Handbook of Differential Geometry • VOLUME II
This page intentionally left blank
Handbook of Differential Geometry ...

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Handbook of Differential Geometry • VOLUME II

This page intentionally left blank

Handbook of Differential Geometry •

VOLUME II Editors

Franki J.E. Dillen Leopold C.A. Verstraelen Katholieke Universiteit Leuven Department of Mathematics Leuven, Belgium

Amsterdam • Boston • Heidelberg • London • New York • Oxford Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo North-Holland is an imprint of Elsevier

ELSEVIER B.V. Radarweg 29 P.O. Box 211, 1000 AE Amsterdam The Netherlands

ELSEVIER Inc. 525 B Street, Suite 1900 San Diego, CA 92101-4495 USA

ELSEVIER Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB UK

ELSEVIER Ltd 84 Theobalds Road London WC1X 8RR UK

© 2006 Elsevier B.V. All rights reserved. This work is protected under copyright by Elsevier B.V., and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier’s Rights Department in Oxford, UK: phone (+44) 1865 843830, fax (+44) 1865 853333, e-mail: [email protected] Requests may also be completed on-line via the Elsevier homepage (http://www.elsevier.com/locate/permissions). In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 20 7631 5555; fax: (+44) 20 7631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of the Publisher is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier’s Rights Department, at the fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.

First edition 2006 Library of Congress Cataloging in Publication Data A catalog record is available from the Library of Congress. British Library Cataloguing in Publication Data A catalogue record is available from the British Library. ISBN-13: 978-0-444-52052-4 ISBN-10: 0-444-52052-X

∞ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.

Dedication In memory of S.S. Chern and T. Willmore

This page intentionally left blank

Preface “Our goal with the volumes which together will constitute the “Handbook of Differential Geometry” is to give a rather complete survey of the field of differential geometry.” Thus reads the opening sentence of the “Handbook of Differential Geometry, Volume I”, and only the presence of the word “rather” saves this goal from being an obvious mission impossible. Let us recall the contents of this Volume I: Differential geometry of webs (M.A. Akivis and V.V. Goldberg), Spaces of metrics and curvature functionals (D.E. Blair), Riemannian submanifolds (B.-Y. Chen), Einstein metrics in dimension four (A. Derdzinski), The Atiyah–Singer index theorem (P.B. Gilkey), Survey of isospectral manifolds (C.S. Gordon), Submanifolds with parallel fundamental form (Ü. Lumiste), Sphere theorems (K. Shiohama), Affine differential geometry (U. Simon), A survey on isoparametric hypersurfaces and their generalizations (G. Thorbergsson), Curves (T. Willmore); with introduction by S.S. Chern. As in Volume I, we allowed the authors in this Volume II as much freedom as possible concerning style and contents. We are confident that the reader will appreciate this pragmatic point of view. Some contributions will emphasize the basics; some will emphasize the classical results; others the recent developments. Needless to say all authors have spent a lot of time and energy in describing their topic, which we appreciate enormously. The contributions to this Volume II are: Some problems on Finsler geometry (J.C. Álvarez Paiva), Foliations (R. Barre and A. El Kacimi), Symplectic geometry (A. Cannas da Silva), Metric Riemannian geometry (K. Fukaya), Contact geometry (H. Geiges), Complex differential geometry (I. Mihai), Compendium on the geometry of Lagrange spaces (R. Miron), Certain actual topics on modern Lorentzian geometry (F.J. Palomo and A. Romero). Obviously the whole field of differential geometry is not yet covered in the two volumes of this “Handbook of Differential Geometry”. Some of the authors explicitly mention topics that should have been covered, but are not for practical reasons; but also other topics are not (yet) treated sufficiently or not treated at all. Recently Professors Chern and Willmore passed away. Both had a great impact on the development of contemporary geometry and were genuine sources of inspiration, guidance and support for many generations of mathematicians through their books and articles, their fantastic lectures and their warm and truly concerned personal contacts. Together with all authors we gratefully dedicate this book to the memories of Professor S.S. Chern and Professor T.J. Willmore.

Franki Dillen and Leopold Verstraelen vii

This page intentionally left blank

List of Contributors Álvarez Paiva, J.C., Polytechnic University, Brooklyn, NY (Ch. 1). Barre, R., Université de Valenciennes, Valenciennes (Ch. 2). Cannas da Silva, A., Instituto Superior Técnico, Lisboa (Ch. 3). El Kacimi Alaoui, A., Université de Valenciennes, Valenciennes (Ch. 2). Fukaya, K., Kyoto University, Kyoto (Ch. 4). Geiges, H., Universität zu Köln, Köln (Ch. 5). Mihai, I., University of Bucharest, Bucharest (Ch. 6). Miron, R., “Al.I. Cuza” University Iasi, Iasi (Ch. 7). Palomo, F.J., Universidad de Málaga, Málaga (Ch. 8). Romero, A., Universidad de Granada, Granada (Ch. 8).

ix

This page intentionally left blank

Contents Preface List of Contributors Contents of Volume I 1. Some problems on Finsler geometry J.C. Álvarez Paiva 2. Foliations R. Barre and A. El Kacimi Alaoui 3. Symplectic geometry A. Cannas da Silva 4. Metric Riemannian geometry K. Fukaya 5. Contact geometry H. Geiges 6. Complex differential geometry I. Mihai 7. Compendium on the geometry of Lagrange spaces R. Miron 8. Certain actual topics on modern Lorentzian geometry F.J. Palomo and A. Romero Author Index Subject Index

vii ix xiii 1 35 79 189 315 383 437 513

547 555

xi

This page intentionally left blank

Contents of Volume I Preface Introduction List of Contributors 1. Differential geometry of webs M.A. Akivis and V.V. Goldberg 2. Spaces of metrics and curvature functionals D.E. Blair 3. Riemannian submanifolds B.-Y. Chen 4. Einstein metrics in dimension four A. Derdzinski 5. The Atiyah–Singer index theorem P.B. Gilkey 6. Survey of isospectral manifolds C.S. Gordon 7. Submanifolds with parallel fundamental form Ü. Lumiste 8. Sphere theorems K. Shiohama 9. Affine differential geometry U. Simon 10. A survey on isoparametric hypersurfaces and their generalizations G. Thorbergsson 11. Curves T. Willmore Author Index Subject Index

v vii ix 1 153 187 419 709 747 779 865 905 963 997

1025 1037

xiii

This page intentionally left blank

CHAPTER 1

Some Problems on Finsler Geometry∗ J.C. Álvarez Paiva∗∗ Department of Mathematics, Polytechnic University, Six MetroTech Center, Brooklyn, NY 11201, USA E-mail: [email protected]

We do like intuitive geometric arguments and uncovering simple geometric reasons underlying seemingly recondite facts. H. Busemann

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . 2. Preliminaries . . . . . . . . . . . . . . . . . . . . 2.1. The Hamiltonian point of view . . . . . . . . 2.2. The Riemannian point of view . . . . . . . . 2.3. Isometries and isometric embeddings . . . . 2.4. Isometric submersions . . . . . . . . . . . . 3. Volume and area in Finsler spaces . . . . . . . . . 4. Unit spheres in Minkowski spaces . . . . . . . . . 5. Symplectic equivalence of Finsler manifolds . . . 5.1. Equivalence of unit co-disc bundles . . . . . 5.2. Equivalence of unit co-sphere bundles . . . . 5.3. Manifolds of geodesics . . . . . . . . . . . . 6. Around Hilbert’s fourth problem . . . . . . . . . . 7. Closed geodesics . . . . . . . . . . . . . . . . . . 8. Differential invariants of Finsler surfaces . . . . . 8.1. Convex geometry and the invariant I . . . . 8.2. The invariant J . . . . . . . . . . . . . . . . 8.3. Curvature of Finsler surfaces . . . . . . . . . 8.4. Cartan’s structure equations . . . . . . . . . 8.5. Applications of Cartan’s structure equations References . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

3 4 5 7 7 8 9 12 14 15 15 17 19 22 23 23 24 24 26 28 31

* This work was partially funded by a crédit aux chercheurs from the FNRS, Belgium. ** I warmly thank G. Berck, D. Burago, C. Durán, E. Fernandes, P. Foulon, I.M. Gelfand, R. Schneider, Z. Shen

and S. Tabachnikov for some very useful conversations and/or criticisms of earlier versions of this work. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 1

2

J.C. Álvarez Paiva

Abstract This chapter is an unorthodox survey of Finsler geometry presenting both results and open problems. It aims to show that recent progress in convex geometry, the calculus of variations, symplectic geometry, and integral geometry can be powerful tools in the study of Finsler manifolds; and that Finsler geometry can prove useful in solving some of the open problems in these fields.

Some problems on Finsler geometry

3

1. Introduction Finsler manifolds, manifolds whose tangent spaces carry a norm that varies smoothly with the base point, were born prematurely in 1854 together with their Riemannian counterparts in Riemann’s ground-breaking Habilitationsvortrag. I say prematurely because in 1854 Minkowski’s work on normed spaces and convex bodies (see [69]) was still forty three years away, and thus not even the infinitesimal geometry on which Finsler manifolds are based was understood or appreciated at the time. Apparently, Riemann did not know what to make of these ‘more general class’ of manifolds whose element of arclength does not originate from a scalar product and, fatefully, put in a bad word for them [44]: Investigation of this more general class would actually require no essential different principles, but it would be rather time-consuming and throw relatively little new light on the study of Space, especially since the results cannot be expressed geometrically.

Given the awe with which we rightfully regard Riemann’s achievements and uncanny geometrical intuition, it is tempting to take the above quotation out of historical context and to dismiss Finsler geometry altogether. But, if we think of the great advances in convex geometry, the calculus of variations, integral geometry, the theory of metric spaces, and symplectic geometry that have taken place since 1854, then we may be moved to reassess Riemann’s statement and to consider applying these new tools to develop the subject in a way that Riemann could not have foreseen. The paper includes eighteen simply-stated open problems, as well as a survey of the more elementary and geometric chapters of Finsler geometry. It presents a detailed discussion of the Holmes–Thompson volume and its role in integral geometry and geometric inequalities, thus complementing the survey by Álvarez and Thompson [16]. The other highlights of the paper are its presentation of Hilbert’s fourth problem and its elementary approach to the differential invariants of Finsler surfaces. These are mostly based on the papers [15,12,13] with I.M. Gelfand, M. Smirnov and E. Fernandes, as well as on the lecture notes [10] written jointly with C. Durán. In view of the often-made criticisms of Finsler geometry—very few concrete and interesting examples, very few non-Riemannian theorems of real geometric content, and too many subindices—I have tried to include as many concrete examples, simply-stated results, and geometric constructions as possible. In this way, many of the jewels, so to speak, of Finsler geometry find their way into the following pages. As anyone writing a survey paper, I have had to make some choices. In matters of taste, I have consistently preferred the concrete to the abstract, the elementary to the advanced, the C ∞ to the C k , and the global to the local. I have stayed clear of Riemann–Finsler geometry and Finsler connections because the book [23] of Bao, Chern and Shen covers the subject in depth as do the lecture notes of Abate and Patrizio [1]. Because of my ignorance of the subject, I have not touched on complex Finsler geometry (see [1] also for this topic) and, despite their undeniable interest and importance, non-reversible Finsler metrics are barely mentioned. Another important topic that is not covered in this survey is Busemann’s G-spaces. This approach, which consists in abstracting the properties of geodesics on Finsler manifolds, is one of the most powerful in Finsler geometry, but it is impossible to outdo Busemann’s own exposition in [37,38].

4

J.C. Álvarez Paiva

The reader wishing to have a broader view of the activity in Finsler geometry should read the short surveys of Chern [44] and Busemann [35]. The book [20] contains survey articles on several topics including non-reversible Finsler metrics, the proceedings [22] contains many open problems and a vista of various approaches to Finsler geometry. The book [84] contains a beautiful exposition of the convex-geometric aspects of the Holmes– Thompson volume as well as most of the convex geometry necessary for the study of Finsler manifolds. The lecture notes [10] are similar in spirit to the present paper. The thesis of Egloff [46] is a good place to learn about the beautiful results of Egloff [47,48] and Foulon [50–52] on the geometry and dynamics of Finsler manifolds with non-positive and negative curvature. Finally, I wholeheartedly recommend looking at the papers [42] and [3] before plunging into other papers where Finsler connections are treated. The reader can find many of the preprints cited in this paper, along with other works on the interaction between convex, integral, metric, and symplectic geometry, in http://www.math.poly.edu/research/finsler. 2. Preliminaries If (V , · ) is a real, finite-dimensional normed space, we define the length of a smooth curve γ : [a, b] → V by the formula length of γ :=

γ˙ (t) dt.

b

a

A smooth submanifold N ⊂ V inherits a metric from the norm: if x and y are two points on N , define their distance as the infimum of the lengths of all smooth curves on N joining x and y. Notice that in order to define the metric on N it suffices to know the restriction of the norm to each tangent space. This motivates the following heuristic definition: A Finsler manifold is a manifold together with the choice of a norm on each tangent space. The precise definition requires us to restrict the class of norms to those where the unit sphere is smooth and quadratically convex (i.e., it has positive principal curvatures for some (and therefore any) Euclidean structure on V ). The intrinsic definition of these norms is as follows: Let V be a vector space and let ϕ : V → [0, ∞) be a norm that is smooth outside the origin. Set L := ϕ 2 /2 and consider the exterior derivative of L, dL, as a map from V minus the origin to V ∗ minus the origin. The norm ϕ is said to be a Minkowski norm if dL is a diffeomorphism. For any non-zero vector v ∈ V , the differential D(dL)(v) : Tv V → TdL(v) V ∗ is an invertible linear map. In fact, using the natural identification of Tv V with V , and TdL(v) V ∗ with V ∗ , we can think of gϕ (v) := D(dL)(v) as a (symmetric) bilinear form on V : gϕ (v)(w1 , w2 ) := D(dL)(v)(w1 ) (w2 ).

Some problems on Finsler geometry

5

The norm ϕ is a Minkowski norm if and only if gϕ is positive definite. When the vector v belongs to the unit sphere, we will denote gϕ (v) as the osculating Euclidean structure at v and the ellipsoid Ev := w ∈ V : gϕ (v)(w, w) = 1 as the osculating ellipsoid at v. D EFINITION 2.1. A Finsler metric on a manifold M is a continuous function defined on its tangent bundle with the property that it is smooth away from the zero section and its restriction to each tangent space is a Minkowski norm. Some examples of Finsler manifolds are submanifolds of Minkowski spaces and flat tori obtained as quotients of Minkowski spaces. If γ : [a, b] → M is a smooth curve on a Finsler manifold (M, ϕ), then the quantity length of γ :=

b

ϕ γ˙ (t) dt

(1)

a

is independent of the parameterization. Using this definition of length we define a metric on M by letting the distance between two points x, y ∈ M to be the infimum of the lengths of all smooth curves joining x and y. Finsler manifolds are length spaces: the length of a curve γ defined by the integral in (1) equals the metric length of the curve given by k−1 sup dist γ (ti ), γ (ti+1 ) : a = t0 < · · · < tk = b is a partition of [a, b] . i=0

The condition that the norms in each tangent space be Minkowski norms is necessary for the study of the geodesics. Namely, we want these to be solutions of a second-order differential equation on M.

2.1. The Hamiltonian point of view If (V , ϕ) is a normed space, then the dual vector space V ∗ inherits a natural norm defined by the equation

ϕ ∗ (ξ ) := sup ξ (v) : ϕ(v) 1 . A related construction on Minkowski spaces is the Legendre transform which assigns to a non-zero vector v ∈ V the covector ϕ(v) dϕ(v) = gϕ (v)(v, ·). It is easy to check that if v belongs to the unit sphere S ⊂ (V , ϕ), then the Legendre transform of v is the unique covector ξ such that ξ = 1 is the hyperplane tangent to S at the point v. This implies that the image of S under the Legendre transform is the unit sphere in (V ∗ , ϕ ∗ ).

6

J.C. Álvarez Paiva

Let (M, ϕ) be a Finsler manifold and for each point m ∈ M let ϕm denote the Minkowski ∗ ) is the dual of the normed space (T M, ϕ ), then the function norm on Tm M. If (Tm∗ M, ϕm m m H : T ∗M → R ∗ (p ) is a Hamiltonian whose energy surfaces are fiber-wise condefined by H (pm ) := ϕm m vex. Applying the Legendre transform on each tangent space of M defines a diffeomorphism L : T M \ 0 → T ∗ M \ 0 with the property that H ◦ L = ϕ. By passing from a Finsler metric to its associated Hamiltonian, we gain access to the techniques of Hamiltonian mechanics and symplectic geometry. Below, we recall some of the basic definitions and constructions. For more information see [18] and [2].

D EFINITION 2.2. Let π : T ∗ M → M be the standard projection and let Dπ : T (T ∗ M) → T M be its differential. The canonical 1-form α on T ∗ M is defined by the equation α(vpm ) = pm (Dπ(vpm )), where pm ∈ Tm∗ M and vpm ∈ Tpm (T ∗ M). The symplectic 2-form is defined as ω := −dα. The form ω is non-degenerate: at each point pm ∈ T ∗ M, the map vpm → ωpm (vpm , ·) is an isomorphism from Tpm (T ∗ M) to Tp∗m (T ∗ M). We can use this isomorphism to pass from 1-forms on T ∗ M to vector fields on T ∗ M. D EFINITION 2.3. Let H : T ∗ M → R be a smooth function. The Hamiltonian vector field of H , XH , is defined by the equality dH = ω(XH , ·). As an easy consequence of the definition, we have that H is constant along the integral curves of the Hamiltonian vector field XH , and that the symplectic form is invariant under the flow of XH . Because of this result, it is usual to disregard the function H in favor of the unit co∗ M := H −1 (1). If α is the canonical 1-form on T ∗ M, then its restriction sphere bundle SH ∗ M, which we denote by α , is a contact form (i.e., the to the unit co-sphere bundle SH H n−1 never vanishes). Using αH , we can define the restriction top-order form αH ∧ (dαH ) of the Hamiltonian vector field XH without any reference to the function H : ∗ M is defined by the equations D EFINITION 2.4. The Reeb vector field XH on SH

αH (XH ) = 1,

dαH (XH , ·) = 0.

The projection to M of the integral curves of this vector field are geodesics parameterized with unit speed. Conversely, if γ is a geodesic on M parameterized with unit speed, then the Legendre transform L maps the velocity curve γ˙ to an orbit of the Reeb vector field. We remark that if γ is any smooth curve on M parameterized with unit speed, then length of γ =

L◦γ

αH .

(2)

Some problems on Finsler geometry

7

Note that the Finsler manifold (M, ϕ) is geodesically complete (or metrically complete, since it is easy to verify that the Hopf–Rinow theorem extends to the Finsler setting) if and only if the Reeb vector field defines a flow. Let us finish this section by remarking that the non-degeneracy of the symplectic form ω on T ∗ M is equivalent to the fact that ωn , n = dim(M), is a volume form. This remark will provide us with a natural way to define the volume of a Finsler manifold. 2.2. The Riemannian point of view Finsler manifolds can also be studied from the point of view of Riemannian manifolds and bundles. Indeed, to every unit vector vm ∈ Tm M we may associate the inner product gϕ (v). In this way, we can define a Riemannian structure on the pullback of the tangent bundle of M to the unit tangent bundle of M. This construction underlies many of the definitions of connections associated to Finsler manifolds. A variation on this theme is to take a nowhere zero vector field X defined on an open subset O ⊂ M and to associate to it the Riemannian metric on O defined by m → gϕ (X(m)). This construction has been used by Shen (see [79] and Section 8) to give a simple description of the Finsler curvature. 2.3. Isometries and isometric embeddings The definitions of isometry and isometric embedding between Finsler manifolds (M, ϕM ) and (N, ϕN ) are the same as for Riemannian manifolds. Namely, an isometry (respectively isometric embedding) is a diffeomorphism (respectively embedding) f : M → N such that f ∗ ϕN = ϕM . Unlike Riemannian manifolds, two Finsler manifolds can fail to be isometric because of what happens at a single tangent space. For example, if at a point m ∈ M the indicatrix Sm M := vm ∈ Tm M: ϕM (vm ) = 1 is an ellipsoid while none of the indicatrices of N are ellipsoids, then M and N are not isometric. This remark points at the important role played by the centro-affine geometry of convex hypersurfaces in Finsler geometry. In [32] Burago and Ivanov showed that any compact Finsler manifold admits an isometric embedding into a finite-dimensional normed space. It is likely, but unproved, that the norm can be chosen to be a Minkowski norm. They also give examples of noncompact Finsler manifolds that cannot be isometrically embedded in any finite-dimensional normed space. Examples of Finsler manifolds that cannot be isometrically embedded in any Minkowski space have been given by Shen (see [79]) and by Álvarez and Durán (see Section 8.5). However, new types of embedding problems arise in Finsler geometry. For example, while it is known that every two-dimensional normed space is isometric to a subspace of L1 ([0, 1]) (see, for example, [29]), it is not clear whether the following problem has an affirmative answer.

8

J.C. Álvarez Paiva

P ROBLEM 1. Does every two-dimensional Finsler manifold admit an isometric embedding into the Banach space L1 ([0, 1])? The relation between intrinsic and extrinsic geometric properties of submanifolds of normed space is not understood. For example, the following problem from [84] is open. P ROBLEM 2 (Thompson). Let X and Y be two normed spaces of dimension n, n > 2, such that its unit spheres are isometric as Finsler manifolds. Does it follow that X and Y are isometric as normed spaces?

2.4. Isometric submersions In Riemannian geometry isometric submersions are used to construct examples of Riemannian manifolds while keeping some control on their geodesics and curvature. The Finslerian generalization of this construction is simple, but perhaps not as well known as it should be. What follows is taken from [11]. D EFINITION 2.5. A surjective linear map π : X → Y between two normed spaces is said to be an isometric submersion if the image of the closed unit ball on X under the map π equals the closed unit ball on Y . Clearly, if x is any vector in X, then π(x)Y xX . The vectors for which the equality holds are called horizontal vectors and form the horizontal cone in X. Notice that if the unit sphere in X is smooth, its intersection with the horizontal cone is the singular set of the restriction of the map π to the unit sphere. This description makes it easy to grasp that, unlike the case where the spaces is Euclidean, the horizontal cone is rarely a subspace. D EFINITION 2.6. A submersion ρ : M → N between Finsler manifolds is said to be isometric if for every point m ∈ M the differential Dm ρ : Tm M → Tρ(m) N is an isometric submersion of normed spaces. More generally, Berestovskii has defined in [24] isometric submersions, or submetries of metric spaces, as maps that send metric balls to metric balls of the same radius. When the metric spaces are Finsler manifolds both notions of isometric submersion agree. However, in the particular case of Finsler manifolds we can also speak of horizontal lifts. D EFINITION 2.7. Let ρ : M → N be an isometric submersion. An immersed curve γ : [a, b] → M is said to be horizontal if for every t ∈ (a, b) the velocity vector γ˙ (t) belongs to the horizontal cone in Tγ (t) M. A curve γ : [a, b] → M is said to be a horizontal lift of an immersed curve σ : [a, b] → N if γ is horizontal and ρ ◦ γ = σ .

Some problems on Finsler geometry

9

The following result is an easy consequence of the definitions and of the basic properties of geodesics in a Finsler manifold. T HEOREM 2.8 (Álvarez and Durán [11]). An immersed curve on N is a geodesic if and only if any of its horizontal lifts is a geodesic on M. In particular, the geodesics of N are precisely the projections of horizontal geodesics on M. As we shall see in Section 6, this theorem is useful in constructing interesting examples of Finsler metrics on complex and quaternionic projective spaces.

3. Volume and area in Finsler spaces The theory of volume and area in normed spaces has long been a major driving force in convex geometry. For example, the Blaschke–Santaló inequality, the Mahler conjecture, the Busemann–Petty problems, the Shephard problem, and the numerous works of Busemann, Ewald and Shephard on the notions of convexity on Grassmannians originated from or have applications to the study of volumes and areas in normed spaces (see [16]). Defining a volume on a finite-dimensional normed space seems easy: a natural volume should be invariant under translations, positive on open sets, and finite on compact sets. By Haar’s theorem, such a volume must be a multiple of the Lebesgue measure. However, the choice of this multiple is crucial. To understand this, suppose we have already decided on how to assign those constants to two-dimensional normed spaces. We can now define the area of a two-dimensional polyhedral surface embedded in a 3-dimensional normed space as the sum of the areas of its faces with their induced norms. Making a different choice of constants leads to a completely different way of measuring the area of polyhedral surfaces. The guiding principle in defining volumes on normed spaces is that the choice of a volume for every k-dimensional normed space leads to the definition of the k-volume integrand in all higher-dimensional normed spaces. Requiring even mild conditions on these area integrands, such as that regions in hyperplanes be area-minimizing, severely restricts our choices for a definition of volume. In fact, in the literature one can only find three reasonable choices of volume on normed spaces: the Busemann definition, the Holmes– Thompson definition, and the Benson definition (also known as Gromov’s mass∗ [56]). Since the Busemann volume of a Finsler manifold coincides with its Hausdorff measure as a metric space, at first sight it seems the most natural and geometric definition. However, the Holmes–Thompson definition, with its ties to Brunn–Minkowski theory, integral, and symplectic geometry, is rapidly becoming the definition of choice. D EFINITION 3.1. The Holmes–Thompson volume of an n-dimensional Finsler manifold (M, ϕ), voln (M, ϕ), is the symplectic volume of its unit co-disc bundle divided by the volume of the Euclidean n-dimensional unit ball. The k-volume of a k-dimensional submanifold is the volume of the submanifold with its induced Finsler metric.

10

J.C. Álvarez Paiva

Note that, using the notation of the previous section, we have that voln (M) =

1 n n!

∗M SH

αH ∧ (dαH )n−1 ,

(3)

where n is the volume of the Euclidean unit ball of dimension n. Using the Blaschke–Santaló inequality, Durán has remarked in [45] that the Hausdorff measure of a Finsler manifold is no less than its Holmes–Thompson volume. Equality holds if and only if the metric is Riemannian. One of the basic problems about a given definition of volume on normed and Finsler spaces is determining the convexity or ellipticity properties of the k-volume integrands. Perhaps the most enticing and difficult question of the kind is the following problem of Busemann: P ROBLEM 3. Let P be a compact polyhedron of dimension k in a normed space. Is the k-volume of any given face less than or equal to the sum of the k-volumes of the remaining faces? In the case of polyhedra of codimension one the answer to this question is affirmative for both the Hausdorff measure (Busemann [34]) and the Holmes–Thompson volume (Holmes and Thompson [62]). For the Hausdorff measure no other results of this kind are known. For the Holmes– Thompson volume the answer to Busemann’s question is known to be affirmative if the normed space is a subspace of L1 ([0, 1]): T HEOREM 3.2 (Busemann et al. [40]). Let P be a compact polyhedron of dimension k in the Banach space L1 ([0, 1]). The (Holmes–Thompson) k-volume of any given face is less than or equal to the sum of the k-volumes of the remaining faces. The latest progress on the question of minimality of flats in normed spaces is the following result of Burago and Ivanov: T HEOREM 3.3 (Burago and Ivanov [33]). Let P be a compact 2-dimensional polyhedron in a normed space. If P is homeomorphic to a sphere, then the (Holmes–Thompson) 2-volume of any given face is less than or equal to the sum of the 2-volumes of the remaining faces. In [64], S. Ivanov shows that this result can be extended to a theorem that is new even in the Riemannian case: T HEOREM 3.4 (Ivanov [64]). Let ϕ be a Finsler metric on the closed two-dimensional disc D such that every two points on D are joined by a unique geodesic. If ψ is another Finsler metric on D such that the distance induced by ψ on the boundary ∂D of D is greater than or equal to the distance induced by ϕ on ∂D, then the Holmes–Thompson volume of (D, ϕ) does not exceed that of (D, ψ).

Some problems on Finsler geometry

11

It follows from Ivanov’s theorem that a two-dimensional totally geodesic submanifold in a Finsler space is minimal—in the usual sense of being an extremal for the 2-volume functional—with respect to the Holmes–Thompson definition of area. This is true in all dimensions: T HEOREM 3.5 (Berck [25]). A totally geodesic submanifold of a Finsler manifold is minimal with the respect to the Holmes–Thompson definition of area. In [9], Álvarez and Berck show that Theorems 3.4 and 3.5 no longer hold if the Holmes– Thompson volume is replaced by the Hausdorff measure. One of the advantages of working with the Holmes–Thompson definition is that there is a remarkably simple formula for the Holmes–Thompson k-volume densities of a Minkowski space in terms of the Fourier transform of its norm. In a different guise, this formula was first obtained by W. Weil [85]. In the present form it was rediscovered by Álvarez and Fernandes in [13]. F OURIER TRANSFORMS OF NORMS . Let φ be a smooth, even homogeneous function of degree one on an n-dimensional vector space V , let e1 , . . . , en be a basis of V , and let ξ 1 , . . . , ξ n be the dual basis in V ∗ . These bases allow us to introduce coordinates (x1 , . . . , xn ) in V and (ξ1 , . . . , ξn ) in V ∗ , which we can use to compute the standard (distributional) Fourier transform ˆ ) := eiξ ·v φ(v) dv. φ(ξ Rn

This transform depends on the choice of basis, or rather on the Lebesgue measure associated to it. However, the form φˆ dξ1 ∧ · · · ∧ dξn does not. Up to a constant factor, the Fourier transform of φ is the contraction of this n-form with the Euler vector field, XE (ξ ) = ξ , in V ∗ : φˇ :=

−1 φˆ dξ1 ∧ · · · ∧ dξn XE . 4(2π)n−1

It is known (see [63, pp. 167–168]) that φˆ is smooth on V ∗ \ {0} and homogeneous of degree −n − 1. It follows that φˇ is a smooth differential form on V ∗ \ {0} which is homogeneous of degree −1. D EFINITION 3.6. Let (V , ϕ) be an n-dimensional Minkowski space. For each integer k, 1 k < n, define the integrand ϕk (v1 ∧ · · · ∧ vk ) := (ξ 1 ,...,ξ k

)∈S ∗k

|ξ 1 ∧ · · · ∧ ξ k · v1 ∧ · · · ∧ vk |ϕˇ k ,

(4)

where S ∗ is any closed hypersurface in V ∗ \ {0} that is star-shaped with respect to the origin.

12

J.C. Álvarez Paiva

T HEOREM 3.7 [13]. Let (V , ϕ) be an n-dimensional Minkowski space. If N ⊂ V is an immersed submanifold of dimension k, 1 k < n, then we have the following formula for the Holmes–Thompson k-area of N : 1 volk (N ) = k

ϕk , N

where k denotes the volume of the Euclidean unit ball of dimension k. One of the main justifications for adopting the Holmes–Thompson volume comes from its role in integral geometry (see [78,12,13,76,77]). In fact, the formula for the Holmes– Thompson k-area density in terms of the Fourier transform of the norm is equivalent to the following Crofton-type formula for Minkowski spaces: T HEOREM 3.8. Let (V , · ) be an n-dimensional Minkowski space and let k, 1 k n − 1, be an integer. There exists a smooth, translation-invariant, and possibly signed measure Φn−k on the manifold Hn,n−k of (n − k)-flats of V such that if N ⊂ V is an immersed k-dimensional submanifold, then volk (N ) =

1 k

#(N ∩ λ)Φn−k ,

(5)

λ∈Hn,n−k

where k is the volume of the Euclidean unit ball of dimension k. This theorem was first proved for finite-dimensional subspaces of L1 ([0, 1]) by Schneider and Wieacker [78]. In the form given above it is due to Álvarez and Fernandes [12]. It has recently been extended by Schneider [77] to generalized hypermetric spaces (i.e., finitedimensional normed spaces where the distributional Fourier transform of the norm is a signed measure).

4. Unit spheres in Minkowski spaces Surprisingly little is known about the Finsler geometry of unit spheres in Minkowski spaces. The classic results in two dimensions are the theorems of Goł¸ab and Schäffer (see [54] and [73]). T HEOREM 4.1 (Goł¸ab). The length of the unit circle of a two-dimensional normed space is greater than or equal to six and less than or equal to eight. Moreover, the lower bound is attained if and only if the unit circle is an affine regular hexagon and the upper bound is attained if and only if the unit circle is a parallelogram. T HEOREM 4.2 (Schäffer). The length of the unit circle of a two-dimensional normed space equals the length of the unit circle of its dual space.

Some problems on Finsler geometry

13

The perimeter of the unit circle is at once its surface area, twice its intrinsic diameter, the length of its shortest closed geodesic, and the length of its shortest closed, symmetric geodesic. Each of these interpretations points to a different, possible, higher-dimensional extension of the theorems of Goł¸ab and Schäffer. In this section, we shall quickly survey what is known about the higher-dimensional analogues of Goł¸ab’s theorem. The generalizations of Schäffer’s theorem have an unexpected relation with symplectic geometry and will be discussed in the next section. We start with the following upper bounds for the Hausdorff measure and the Holmes– Thompson area of unit spheres in finite-dimensional normed spaces. T HEOREM 4.3 (Busemann and Petty [41]). The (intrinsic) Hausdorff measure of the unit sphere of an n-dimensional normed space is at most 2n times the volume of the Euclidean unit ball of dimension n − 1. Equality holds if and only if the unit ball is a parallelotope. Since the Holmes–Thompson area is always less than or equal to the Hausdorff measure, we have the following corollary: C OROLLARY 4.4 (Thompson). The Holmes–Thompson area of the unit sphere in an n-dimensional normed space is at most 2n times the volume of the Euclidean unit ball of dimension n − 1. Equality holds if and only if n = 2 and the unit ball is a parallelogram. The quest of the lower bound is much more challenging and interesting. The only result in dimension greater than two is the following (unpublished) sharp lower bound by Álvarez, Ivanov and Thompson. T HEOREM 4.5. The Holmes–Thompson area of the unit sphere in a normed space of dimension three is at least 36/π . This bound is attained, for example, when the unit sphere of the normed space is the rhombic dodecahedron or the cubo-octahedron. As a corollary, we have that the Hausdorff measure of the unit sphere of a threedimensional normed space is greater than 36/π . This result is, so far, the only contribution to the following problem of Busemann and Petty [41]. P ROBLEM 4 (Busemann–Petty Problem 7). Find the sharp lower bound for the Hausdorff measure of the unit sphere of a normed space of dimension n, n 3. Of course, we have the analogous problem for the Holmes–Thompson definition of volume. P ROBLEM 5 (Thompson [84]). Find the sharp lower bound for the Holmes–Thompson area of the unit sphere of a normed space of dimension n, n > 3. Schäffer has considered the length of the shortest closed geodesic that is symmetric about the origin. This length, which Schäffer calls the girth of the normed space, is twice the length of the shortest non-contractible geodesic, the systole, for the induced Finsler metric on the projective space.

14

J.C. Álvarez Paiva

T HEOREM 4.6 (Schäffer [74]). The girth of an n-dimensional normed space is at most eight and least 4 + 2[n/2]−1 , where [·] denotes the greatest-integer function. Moreover if the girth equals eight, then the space is two-dimensional and its unit ball is a parallelogram. Here is yet another interesting question of Schäffer about the girth of normed spaces. The problem was posed as a conjecture in [74, p. 97]. P ROBLEM 6 (Schäffer [74]). Prove or disprove that girth of a Minkowski space of dimension three is at most 2π and that equality holds if and only the space is Euclidean. So far, we have only considered the shortest geodesic that is symmetric about the origin. Is it possible that on some unit sphere there is a shorter geodesic that is not symmetric? P ROBLEM 7 (Thompson). Is the shortest closed geodesic on the unit sphere of a Minkowski space symmetric with respect to the origin? We now shift our attention to a more classic metric invariant, the inner or intrinsic diameter of the unit sphere. T HEOREM 4.7 (Schäffer [74]). The (intrinsic) diameter of the unit sphere of an n-dimensional normed space is at most four and at least 2+[n/2]−1 , where [·] denotes the greatestinteger function. In particular, the diameter of the unit sphere of a three-dimensional Minkowski space is between three and four. P ROOF. In order to see that the diameter is at most four, let x and y be any two distinct points on the unit sphere and consider a plane passing through these points and the origin. The intersection of the plane with the sphere is a curve whose length, by Goł¸ab’s theorem, is at most eight. It follows that the distance between x and y is at most four. To obtain the lower bound, notice that, by Theorem 4.6, the length of the shortest closed, symmetric curve on the unit sphere is greater than or equal to 4 + 2[n/2]−1 . This implies that there is a pair of antipodal points at a distance greater than or equal to 2 + [n/2]−1 and the inequality follows. Schäffer also characterizes those normed spaces for which the diameter of the unit sphere equals four (see Theorem 9G in [74, p. 58]). P ROBLEM 8 (Schäffer [74]). Is the (inner) diameter of the unit sphere of a finitedimensional normed space attained at a pair of antipodes?

5. Symplectic equivalence of Finsler manifolds In this section we study several notions of symplectic equivalence between Finsler spaces and consider the higher-dimensional generalizations of Schäffer’s Theorem 4.2.

Some problems on Finsler geometry

15

5.1. Equivalence of unit co-disc bundles The unit co-disc bundle of a Finsler manifold is an open subset of the cotangent bundle and, as such, it carries a symplectic structure. A vaguely posed, but possibly fruitful, problem is to relate the symplectic invariants of the unit co-disc bundle to the metric invariants of the Finsler manifold. A large class of examples of Finsler manifolds with symplectomorphic unit co-disc bundles is furnished by the following result: T HEOREM 5.1 (Álvarez [8]). Let (Rn , · 1 ) and (Rn , · 2 ) be two Minkowski spaces and let S1 and S2 denote their unit spheres. The unit co-disc bundle of the Finsler metric on S1 induced by its embedding into (Rn , · 2 ) is symplectomorphic to the unit co-disc bundle of the Finsler metric on S2∗ induced by its embedding into (Rn , · ∗1 ). In particular, the unit sphere of a Minkowski space and its dual have symplectomorphic unit co-disc bundles. The following corollary—the first of our generalizations of Schäffer’s theorem—predated the theorem, and was in effect its motivation. C OROLLARY 5.2 (Holmes and Thompson [62]). The unit sphere of a normed space and that of its dual have the same Holmes–Thompson area. Besides examples of non-isometric Finsler manifolds with symplectomorphic co-disc bundles, it is interesting to look for rigidity results such as the following theorem of Benci and Sikorav (see [80] and [68, p. 365]). T HEOREM 5.3. If the unit co-disc bundles of two flat Finsler tori are symplectomorphic, then the tori are isometric.

5.2. Equivalence of unit co-sphere bundles D EFINITION 5.4. Let M and N be two Finsler manifolds with unit co-sphere bundles S ∗ M and S ∗ N and canonical 1-forms αM and αN . The Finsler manifolds M and N will be said to be exactly contactomorphic if there exists a diffeomorphism F : S ∗ M → S ∗ N and a function f on S ∗ M such that F ∗ αN = αM + df . If F ∗ αN = αM , we shall say that the metrics are α-equivalent. P ROPOSITION 5.5. If M and N are two exactly contactomorphic Finsler manifolds, then their volumes and their length spectra are equal. Moreover, if M and N are α-equivalent, then their geodesic flows are conjugate. P ROOF. To see that the volume of M equals that of N we simply use Eq. (3) expressing the Holmes–Thompson volume of the manifold in terms of the canonical 1-form. The equality of the length spectra follows from Eq. (2), which states that the action of a leaf of the geodesic foliation equals the length of the underlying geodesic. Indeed,

16

J.C. Álvarez Paiva

if F : S ∗ M → S ∗ N is a diffeomorphism satisfying F ∗ αN = αM + df , then F maps the geodesic foliation on S ∗ M to the geodesic foliation on S ∗ N . Moreover, we see that closed leaves are taken to closed leaves with the same action. In the case where F ∗ αN = αM , the expression for the geodesic spray as the Reeb vector field of the canonical 1-form immediately implies that the geodesic flows are conjugate. P ROBLEM 9. Is every (reversible) Finsler metric on the two-sphere exactly contactomorphic or α-equivalent to a Riemannian metric? A large class of examples of exactly contactomorphic Finsler manifolds is provided by the following analogue of Theorem 5.1. T HEOREM 5.6 (Álvarez [8]). Let (Rn , · 1 ) and (Rn , · 2 ) be two Minkowski spaces and let S1 and S2 denote their unit spheres. The Finsler metric on S1 induced by its embedding into (Rn , · 2 ) and the Finsler metric on S2∗ induced by its embedding into (Rn , · ∗1 ) are exactly contactomorphic. Moreover, the diffeomorphism F can be taken such that it takes centrally symmetric closed geodesics to centrally symmetric closed geodesics. As a corollary we obtain our second generalization of Schäffer’s theorem: C OROLLARY 5.7. The length of shortest closed geodesic on the unit sphere of a Minkowski space equals the length of the shortest closed geodesic on the unit sphere its dual. While the length of the shortest closed geodesic on the unit sphere of a Minkowski space is an interesting invariant, it seems very hard to prove that it is continuous with respect to any of the natural topologies in the space of convex bodies. In this respect, the girth— the infimum of the lengths of all centrally symmetric, simple, closed curves—is a much better invariant. Indeed, Schäffer proved in [74, p. 91] that the girth of a normed space is continuous with respect to the topology induced by the Banach–Mazur distance. Schäffer also conjectured that the girth of a normed space equals the girth of its dual and proved that it is enough to consider the case of finite-dimensional normed spaces. Theorem 5.6 together with Schäffer’s results settles the conjecture. T HEOREM 5.8 (Álvarez [8]). The girth of a normed space equals the girth of its dual. The preceding theorem is a third generalization of the fact that the perimeter of the unit circle of a normed plane equals the perimeter of the dual circle. Schäffer showed in [74, p. 110] that a fourth possible generalization—that the (intrinsic) diameter of the unit sphere in a normed space equals the diameter of the dual sphere—is false. In particular, the diameter of a Finsler manifold is not a symplectic invariant of its unit co-disc bundle. The notion of α-equivalence first appeared in Weinstein’s work on the volume of Riemannian manifolds all of whose geodesics are closed. The symplectic and topological nature of his proofs implies that they extend unchanged for Finsler metrics.

Some problems on Finsler geometry

17

T HEOREM 5.9 (Weinstein [86]). Let ϕt , t ∈ [0, 1], be a smooth family of Finsler metrics on a compact manifold M. If for every t the geodesics of the Finsler metric ϕt are all closed and of fixed length L, independent of t, then (M, ϕ0 ) and (M, ϕ1 ) are α-equivalent. In particular, (M, ϕ0 ) and (M, ϕ1 ) have the same volume. Weinstein’s theorem follows from the fact that the manifolds of geodesics of the metrics involved are symplectomorphic. In the next paragraph we will review the natural symplectic structure on spaces of geodesics and some of their applications to integral geometry and the minimality of submanifolds in Finsler spaces. 5.3. Manifolds of geodesics Let M be a Finsler manifold such that its space of oriented geodesics is a manifold G(M). Let S ∗ M denote its unit co-sphere bundle and let π : S ∗ M → G(M) be the canonical projection which sends a given unit covector to the geodesic that has this covector as initial condition. P ROPOSITION –D EFINITION 5.10 ([18] and [27]). Let M be a Finsler manifold with manifold of geodesics G(M) and let S∗M

i

T ∗M

π

G(M) be the canonical projection onto G(M) and the canonical inclusion into T ∗ M. If ωM is the standard symplectic form on T ∗ M, then there is a unique symplectic form ω on G(M) which satisfies the equation π ∗ ω = i ∗ ωM . At first sight there seem to be very few examples of Riemannian or Finsler manifolds whose space of geodesics is smooth. The following examples will convince the reader that this is not so. E XAMPLES . 1. Strictly convex balls and Hadamard manifolds [49]. Around any point x in a Finsler manifold there is an open geodesic ball with the property that the function that assigns to every point in the ball its distance from x is strictly convex. The space of geodesics in such a geodesic ball is a smooth manifold. As a result, the space of geodesics of any complete Riemannian metric on Rn with non-positive sectional curvature (a Hadamard manifold) is a smooth manifold. 2. Projective Finsler metrics. These are Finsler metrics on open, convex subsets of RP n such that projective line segments are geodesics. We will review their construction in the next section.

18

J.C. Álvarez Paiva

3. Zoll manifolds. These are Finsler metrics all of whose geodesics are periodic with the same minimal period. A great number of Riemannian examples has been constructed by Weinstein (see [27]). It is interesting to determine when the manifolds of geodesics of two Finsler manifolds are symplectomorphic. Here are two results in this direction: T HEOREM 5.11 (Ferrand [49]). The manifold of geodesics of a Hadamard manifold of dimension n is symplectomorphic to the cotangent bundle of the (n − 1)-dimensional sphere. T HEOREM 5.12 (Ono [70]). The manifold of geodesics of a Zoll Finsler metric on S 3 is symplectomorphic to a complex hyperquadric in CP 3 . It follows from these theorems that all Hadamard manifolds and all Zoll metrics on S 3 are α-equivalent among themselves. P ROBLEM 10. Is the manifold of geodesics of a Zoll Finsler metric on the n-sphere symplectomorphic to a complex hyperquadric in CP n ? Is the space of all Zoll Finsler metrics on the n-sphere connected? The study of the symplectic geometry of the space of geodesics has interesting applications to the integral geometry of Finsler manifolds. For example, the classical integralgeometric theorem of Cauchy and its extension to finite-dimensional normed spaces is a consequence of the following symplectic equivalence. T HEOREM 5.13 (Álvarez [8]). Let (V , · ) be a Minkowski space, and let M ⊂ V be a smooth quadratically convex hypersurface. The unit co-disc bundle for the induced Finsler metric on M and the set of all oriented lines in V which pass through the interior of M are symplectomorphic. The Crofton formula for hypersurfaces of Finsler spaces, announced by Chakerian in [43], follows easily from the co-area formula and symplectic reduction (see [9] for a proof). T HEOREM 5.14. Let M be an n-dimensional Finsler manifold with manifold of geodesics G(M). If N ⊂ M is an immersed hypersurface and if ωn−1 denotes the Liouville volume form on G(M), then voln−1 (N ) =

1 2n−1 n!

·

γ ∈G(M)

#(N ∩ γ ) ωn−1 ,

where n−1 is the volume of the Euclidean unit ball of dimension n − 1.

Some problems on Finsler geometry

19

6. Around Hilbert’s fourth problem In modern terminology, Hilbert’s fourth problem asks to construct and study all Finsler metrics on open convex subsets of RP n (including RP n itself) such that geodesics lie on projective lines. At the root of this problem are Minkowski’s work on normed spaces and the following generalization of the Cayley–Klein model of hyperbolic geometry given by Hilbert himself: Let D ⊂ Rn be an open domain bounded by a convex hypersurface C. If x and y are two distinct points on D, denote by a and b the points of intersection of C with the line determined by x and y (see Figure 1), and define the distance between these points by the equation d(x, y) :=

1 y − ax − b . ln 2 x − ay − b

(6)

T HEOREM 6.1 (Hilbert [59]). The function d is a distance function on D. Moreover, straight line segments are geodesics. The metric space (D, d) is called a Hilbert geometry. The following elegant description of the Finsler metrics that gives rise to Hilbert geometries, and which I learned from R. Ambartzumian, is apparently well known. Let D ⊂ Rn be an open domain bounded by a smooth, quadratically convex hypersurface C. Define a Finsler metric ϕ on D by setting its value at a non-zero vector vx ∈ Tx D to be ϕ(vx ) := (t1−1 + t2−1 )/2, where t1 and t2 are the two positive real numbers for which x + t1 v and x − t2 v belong to C. P ROPOSITION 6.2. If x and y are two points on D and xy is the line segment joining them, then y − ax − b 1 . ϕ = ln 2 x − ay − b xy Hamel, a student of Hilbert, was the first to study Hilbert’s fourth problem. Among other things he showed that Lagrangians on Rn whose extremals are straight lines are characterized by a system of linear partial differential equations.

Fig. 1.

20

J.C. Álvarez Paiva

T HEOREM 6.3 (Hamel [58]). Let ϕ : T Rn \ 0 → R be a smooth Lagrangian which is homogeneous of order one in the velocities. Straight lines are extremals of the functional

γ → ϕ(γ˙ (t)) dt if and only if ϕ satisfies the system of equations ∂ 2ϕ ∂ 2ϕ = ∂xi ∂vj ∂xj ∂vi

for 1 i, j n.

(7)

It is mainly through the work of Busemann and Pogorelov (see [37,72,82,15]) that the construction of projective Finsler metrics in terms of a class of (possibly signed) measures is now well understood. D EFINITION 6.4 [15,82]. Let D ⊂ RP n be an open convex set and let Hn−1 (D) be the set of all hyperplanes passing through D. A possibly-signed measure on Hn−1 (D) is said to be quasi-positive if for any two line segments xy and yz not on the same line, the measure of the set of hyperplanes intersecting twice the wedge formed by xy and yz is positive. T HEOREM 6.5. A Finsler metric ϕ on an open convex set D ⊂ RP n is projective if and only if there exists a smooth quasi-positive measure Φn−1 on the space of hyperplanes passing through D, Hn−1 (D), such that for any smooth curve γ , 1 ϕ= #(ζ ∩ γ )Φn−1 . (8) 2 ζ ∈Hn−1 (D) γ Notice that, in particular, the length of a line segment equals half the Φn−1 -measure of the set of all hyperplanes intersecting it. In Pogorelov’s approach to Hilbert’s fourth problem, Theorem 6.5 follows from the following integral representation for the solution of Hamel’s equations. T HEOREM 6.6. A Lagrangian ϕ : T Rn \ 0 → R which is homogeneous of order one in the velocities satisfies Hamel’s equations if and only if there exists a smooth even function ν(r, ξ ) on R × S n−1 such that ϕ(x, v) = |ξ · v|ν(ξ · x, ξ )Ω, (9) ξ ∈S n−1

where Ω is the standard area form on the unit sphere in Rn . E XAMPLE [12]. Applying formula (9) to the function ν : R × S 1 → R defined by ν(r, θ ) = 1 + r 2 , we obtain the Finsler metric 1 3 + x12 + x22 v12 + v22 + (x1 v1 + x2 v2 )2 . ϕ(x1 , x2 , v1 , v2 ) = 3 v12 + v22 Theorem 6.5 states that projective Finsler metrics are exactly those Finsler spaces for which there is a Crofton formula for the lengths of curves. Do the Crofton formulas for

Some problems on Finsler geometry

21

the areas of submanifolds also hold? Does the first Crofton formula imply all others? The answer is yes if by area we mean Holmes–Thompson area: T HEOREM 6.7 (Álvarez and Fernandes [12]). Let ϕ be a projective Finsler metric on an open convex domain D ⊂ RP n and let k, 1 k n − 1, be a natural number. There exists a smooth (possibly signed) measure Φn−k on the manifold Hn−k (D) of (n − k)-flats passing through D such that if N ⊂ Rn is an immersed submanifold of dimension k, then 1 volk (N ) = #(N ∩ ζ )Φn−k , (10) k ζ ∈Hn−k (D) where k is the volume of the Euclidean unit ball of dimension k. The construction of the measures Φn−k given in [12] and the formula (4) for the Holmes– Thompson volume imply that all the tangent spaces of a projective Finsler metric are hypermetric if and only if the measures Φn−k , k = 1, . . . , n − 1, are positive. In this case, just like in the case of the standard Riemannian metric on RP n , projective subspaces are area-minimizing. T HEOREM 6.8 (Álvarez and Fernandes). If the geodesics of a Finsler metric on RP n are projective lines and all its tangent spaces are hypermetric, then the projective subspaces are area-minimizing in their homology class. P ROOF. If N ⊂ RP n is a k-dimensional submanifold which is homologous to a projective subspace, then the number of points of intersection of N with a projective subspace of complementary dimension is at least one. Using the Crofton formula (10) and positivity of the measure Φn−k , we have that 1 volk (N ) = #(N ∩ ζ )Φn−k k ζ ∈Hn−k (RP n ) 1 Φn−k = volk RP k . k Hn−k (RP n ) R. Schneider has recently shown in [77] that the two previous theorems remain valid even if the regularity assumptions on both the Finsler metric and the submanifold are significantly weakened. The following results also points to the similarity between projective Finsler metrics and the standard Riemannian metric on RP n . P ROPOSITION 6.9. If (RP n , ϕ) is a projective Finsler space for which the length of the projective lines is equal to π , then the Holmes–Thompson volume of (RP n , ϕ) equals the volume of RP n with its standard Riemannian metric. P ROOF. If ϕ0 denotes the standard Riemannian metric on RP n , then, for each number t, t ∈ [0, 1], the metric ϕt = (1 − t)ϕ0 + tϕ is a projective Finsler metric and the length of

22

J.C. Álvarez Paiva

its closed geodesics is π . Applying Theorem 5.9, we conclude that the Holmes–Thompson volumes of (RP n , ϕ0 ) and (RP n , ϕ) are equal. It seems that the analogues of Hilbert’s fourth problem for rank-one symmetric spaces other than RP n have never been studied. For example, the following problem is open: P ROBLEM 11. Construct all Finsler metrics on CP 2 such that the geodesics coincide as point sets with those of the standard Riemannian metric on CP 2 . In trying to solve this problem, Álvarez and Durán stumbled upon the following construction of Finsler metrics on complex and quaternionic projective spaces for which the projective lines are totally geodesic and the geodesics are circles: Use the Busemann–Pogorelov construction of projective metrics on real projective spaces and spheres to construct a projective metric ϕ on S 2n+1 (respectively S 4n+3 ) that is invariant under the Hopf action. There is a unique Finsler metric ψ on CP n (respectively HP n ) for which the projection map of the Hopf fibration is an isometric submersion from (S 2n+1 , ϕ) to (CP n , ψ) (respectively from (S 4n+3 , ϕ) to (HP n , ψ)). For the metric ψ , projective lines are totally geodesic and geodesics are (geometric) circles. This construction suggests yet another inverse problem: P ROBLEM 12. Construct all Finsler metrics on CP n such that the geodesics are circles. For n = 1 this problem has been solved by Álvarez and Berck (unpublished) in terms of space-type surfaces (also called elliptic congruences) on the Grassmannian of Lagrangian planes in R4 . There have been many attempts to define the Finsler analogue of Kähler metrics, but none seems to have enjoyed any measure of success. Since the metric properties of Kähler manifolds are not so well understood as to proceed in this direction, it makes sense to use some remarkable geometric property of Kähler manifolds in an attempt to define their Finsler analogues. One such remarkable geometric property is that complex submanifolds are minimal. P ROBLEM 13. Construct and study all the Finsler metrics on CP n for which complex submanifolds are minimal. Are (Riemannian) Kähler metrics the only ones with this property? 7. Closed geodesics In Riemannian geometry, the study of closed geodesics has a long and glorious history with contributors like Poincaré, Birkhoff and Morse. However, there do not seem to be many interesting results about closed geodesics in Finsler manifolds. If an existence result in Riemannian geometry depends only on Morse theory and the method of broken geodesics, then it automatically holds in Finsler geometry. A Riemannian result that would be interesting to extend to the Finsler setting is a theorem of Bangert and Franks [19,53] stating that any Riemannian metric on the twodimensional sphere has infinitely many closed geodesics. Here we emphasize that we

Some problems on Finsler geometry

23

are considering symmetric or reversible Finsler metrics. Indeed, Katok has constructed non-reversible Finsler metrics on the 2-sphere which have only two closed geodesics (see [65,87]). P ROBLEM 14 (Bangert). Does every Finsler metric on S 2 have infinitely many closed geodesics? Franks’ work on the periodic points of area preserving maps of the annulus and recent work by Hofer, Wysocki and Zehnder (see [61]) suggest the following question: P ROBLEM 15. Is the number of distinct closed geodesics on a non-reversible Finsler metric on the 2-sphere either two or infinity?

8. Differential invariants of Finsler surfaces In this section we study the differential invariants of Finsler surfaces without the aid of the Cartan connection and then introduce Cartan’s structure equations in order to uncover the relations between these invariants. 8.1. Convex geometry and the invariant I A smooth, centrally symmetric, and quadratically convex curve S on a two-dimensional vector space V parameterizes a family of Euclidean structures on V . Indeed, for each point v ∈ S, there is a unique ellipse Ev which is centered at the origin and osculates S up to second order at this point. We associate to v the Euclidean structure with Ev as unit circle. D EFINITION 8.1. Let (V , ϕ) be a two-dimensional Minkowski space with unit circle S. An orthonormal basis of V is an ordered pair of vectors (v, v⊥ ), where v ∈ S and v⊥ is both of unit length and perpendicular to v with respect to the Euclidean structure associated to v. Using the Euclidean structures associated to the curve we may define a distinguished parameterization of S: orient the vector space V and parameterize the curve S by a map γ in such a way that γ (t), γ˙ (t) is an oriented orthonormal basis. Tabachnikov has remarked (private communication) that this parameterization is, up to shifts in the parameter, the only one satisfying the equation det γ (t), γ˙ (t) = det γ˙ (t), γ¨ (t) . The period of γ is a linear invariant of S which we shall call the total angle of S. T HEOREM 8.2 (Schneider [75]). The total angle of a smooth, centrally-symmetric, and quadratically convex curve on the plane is less than or equal to 2π . Equality holds if and only if the curve is an ellipse.

24

J.C. Álvarez Paiva

D EFINITION 8.3. Let S be a smooth, centrally symmetric, and quadratically convex curve on a two-dimensional vector space V . If γ is a distinguished parameterization of S and v = γ (t) is a point on S, then I (v) is defined by the equation γ¨ (t) := −γ (t) + I (v)γ˙ (t). The quantity I (v) is zero if and only if the osculating ellipse E(v) osculates S up to third order or higher at v. It follows that if I is identically zero, then S is an ellipse. Using a theorem of Ghys of the zeros of the Schwartzian derivative (see [71]), Álvarez has showed in [10] that the invariant I vanishes at least eight times. If (M, ϕ) is a Finsler surface, then every tangent space Tm M, m ∈ M, is a Minkowski space and the indicatrix Sm M := vm ∈ Tm M: ϕ(vm ) = 1 is smooth, centrally symmetric, and quadratically convex. For each unit tangent vector vm we define I (vm ) as the value of the invariant I of Sm M at the point vm . The result is a smooth function, which we again denote by I , defined of the unit circle bundle of M. Clearly, this function is identically zero if and only if the Finsler surface is Riemannian. We can also use the previous geometric constructions to define a vector field on the unit circle bundle of an oriented Finsler surface (M, ϕ): If vm is a unit tangent vector and γ (t) is a distinguished parameterization of Sm M with γ (0) = vm , then X3 (vm ) := γ˙ (0). 8.2. The invariant J We shall now define a smooth function on the unit bundle of M which measures how the invariant I changes along the geodesics. D EFINITION 8.4. Let (M, ϕ) be a Finsler surfaces, let vm ∈ Tm M be a unit tangent vector, and let σ : (−, ) → M be the geodesic with initial condition vm . Define

J (vm ) = d/dtI σ˙ (t) t=0 . Note that J = 0 means that I is an invariant of motion. The Finsler surfaces for which this occurs are called Landsberg surfaces. Originally, they came up in the study of Finslerian analogues of the Gauss–Bonnet theorem (see [66] and Section 8.5). Unfortunately, all the known examples of Landsberg surfaces are either Riemannian or locally isometric to Minkowski planes. For example, the following problem remains unsolved. P ROBLEM 16. Is there any non-Riemannian Landsberg metric on the 2-sphere? 8.3. Curvature of Finsler surfaces The formula of second variation really belongs to variational calculus and not to Riemannian geometry. It should not come then as a surprise that many of the definitions and

Some problems on Finsler geometry

25

theorems in Riemannian geometry, including the notion of curvature, extend to Finsler geometry. Let us start by defining the obvious extensions of Jacobi fields and conjugate points: Given a geodesic γ : [a, b] → M, a geodesic variation of γ is a smooth map Γ : (−, )× [a, b] → M such that • Γ (0, t) = γ (t). • For each fixed s0 , the curve t → Γ (s0 , t) is a geodesic. If Γ (s, t) is a geodesic variation of the geodesic γ , the vector field Y along γ defined by

∂Γ (s, t)

Y (t) = ∂s s=0 is called a Jacobi field. A Jacobi field is said to be proper if, for each t, γ˙ (t) and Y (t) form an orthonormal basis of the Minkowski plane Tm M in the sense of Definition 8.1. D EFINITION 8.5. Let p be a point in a Finsler surface and let γ be a geodesic starting from p. A point γ (s) is said to be conjugate to p along γ if there exists a non-zero Jacobi field Y (t) along γ such that Y (0) = Y (s) = 0. Just as in Riemannian geometry, geodesics minimize length up to their first conjugate point. P ROPOSITION 8.6. Let p be a point in a Finsler surface and let γ be a geodesic starting from p. The curve γ restricted to the interval [0, s] minimizes length if and only if there is no conjugate point between p = γ (0) and γ (s). If we are given a geodesic γ let us orient the tangent spaces Tγ (t) M and let us define γ˙ ⊥ (t) in such a way that γ˙ (t) and γ˙ ⊥ (t) form an oriented orthonormal basis on Tγ (t) M. Note that any proper Jacobi field Y along γ can be uniquely expressed as Y (t) = y(t)γ˙ ⊥ (t), where y is a real-valued function. P ROPOSITION –D EFINITION 8.7. There is a unique smooth function K on the unit circle bundle of M such that for any geodesic γ parameterized with unit speed and for any proper Jacobi field Y (t) = y(t)γ˙ ⊥ (t) the Jacobi equation y (t) + K γ˙ (t) y(t) = 0 holds. The function K is called the curvature of (M, ϕ). In contrast with the Riemannian case, K depends on both the point γ (t) ∈ M and the direction of γ˙ (t).

26

J.C. Álvarez Paiva

The Bonnet–Myers theorem follows, as usual, from the Jacobi equation and Sturm’s comparison theorem. T HEOREM 8.8. If the curvature function of a Finsler surface (M, ϕ) is greater √ than or equal to a positive number δ, then the diameter of M is less than or equal to π/ δ. A very helpful way of describing the Finsler curvature in terms of auxiliary Riemannian metrics has been given by Shen [79]: Let vm ∈ Tm M be a unit vector and let X be a geodesic vector field (i.e., the integral curves of X are geodesics parameterized with unit speed) defined on a neighborhood O of m and such that X(m) = vm . If we define a Riemannian metric on O by x → gϕ (X(x)) as in Section 2, then the Riemannian curvature of this metric at m equals the Finsler curvature of (M, ϕ) at vm . 8.4. Cartan’s structure equations By now we have defined three geometric invariants of Finsler surfaces: I , J and K. The invariant I is a centro-affine invariant which describes the shape of each unit tangent circle, the invariant K belongs to the calculus of variations and measures the focusing of geodesics, and the invariant J , by measuring how I varies along geodesics, partakes of both convex geometry and variational calculus. All three invariants can be defined, as we have done, by elementary geometric and variational considerations, but there is nothing to suggest the deep and interesting relations between the three. D EFINITION 8.9. If (M, ϕ) is a Finsler manifold, the geodesic spray of M is the vector field X1 defined on the unit tangent bundle and whose value at a unit tangent vector vm is defined as follows: take σ (t) to be the geodesic with initial condition vm and set X1 (vm ) := d/dt σ˙ (t)|t=0 . T HEOREM 8.10 (Cartan [42]). Let (M, ϕ) be an oriented Finsler surface, let X1 denote its geodesic spray, and let X3 be the vector field defined at the end of Section 8.1. If we define X2 := [X3 , X1 ], then we have the following equations: [X3 , X1 ] = X2 , [X1 , X2 ] = KX3 , [X3 , X2 ] = −X1 + I X2 + J X3 . Of course, Cartan preferred differential forms to vector fields and he wrote the above equations in terms of the dual forms ω1 , ω2 and ω3 defined by the equations ωi (Xj ) = δij . Cartan’s structure equations are: dω1 = −ω2 ∧ ω3 , dω2 = ω1 ∧ ω3 − I ω2 ∧ ω3 , dω3 = −Kω1 ∧ ω2 − J ω2 ∧ ω3 .

Some problems on Finsler geometry

27

Note that by differentiating the structure equations we obtain the following Bianchi identities: J = I1 , K3 + KI + J1 = 0. In these identities the subindices represent differentiation with respect to the vector fields X1 , X2 and X3 . In general if F is a function on the unit bundle we will write dF = F1 ω1 + F2 ω2 + F3 ω3 . Cartan shows that the forms ω1 , ω2 , and ω3 solve the problem of equivalence. From this it follows, at least in theory, that all microlocal invariant properties of Finsler surfaces can be written in terms of the functions I , J , K, and their derivatives with respect to the vector fields X1 , X2 , and X3 . As an example, we have Berwald’s characterizations of locally Minkowski and projectively flat Finsler surfaces. T HEOREM 8.11 (Berwald). A Finsler surface is locally Minkowski if and only if K, I1 , and I2 are identically zero. Let us recall that a Finsler manifold is said to be projectively flat if around every point we can find a small neighborhood and a diffeomorphism of this neighborhood to an open subset of Euclidean space such that geodesics are mapped onto straight lines. T HEOREM 8.12 (Berwald [26]). The (non-parameterized) geodesics of a Finsler surface are locally the geodesics of some affine connection if and only if the following equation holds: I23 + J33 + 2I (I2 + J3 ) + 6J = 0. The dual system of curves to the geodesics of a Finsler surface are locally the geodesics of some affine connection if and only if the following equation holds: K31 − 3K2 = 0. Moreover, both of the above equations hold if and only if the surface is projectively flat. For dual systems of curves and an elementary exposition of path geometry see Arnold’s book [17, pp. 42–56]. A very clear exposition of Berwald’s theorem and its proof is given in [31]. It is amusing to prove the following classic theorem of Beltrami by using Berwald’s result. C OROLLARY 8.13 (Beltrami). A Riemannian surface is projectively flat if and only if its curvature is constant.

28

J.C. Álvarez Paiva

Berwald was also interested in Finsler metrics whose geodesics coincide as parameterized curves with the geodesics of some affine connection. These Finsler metrics are called Berwald metrics. In two dimensions they are characterized by the equations I1 = 0 and I2 = 0. 8.5. Applications of Cartan’s structure equations Armed with Cartan’s structure equations and their Bianchi identities, we are now in a position to prove some of the deepest results in the theory of Finsler surfaces. T HEOREM 8.14 (Akbar–Zadeh [4]). A compact Finsler surface of constant negative curvature is Riemannian. P ROOF. Using the second Bianchi identity, which tells us that K3 + KI + J1 = 0, we have that if K ≡ c is a constant and σ (t) is a geodesic on M, then the function I (t) := I (σ˙ (t)) satisfies the differential equation d2 I = −cI. dt 2 If c is negative, then I (t) must be a linear combination of exponentials and, therefore, if the initial condition is not I (0) = 0, I (0) = 0, the function I (t) is unbounded. Since I is bounded whenever the Finsler surface is compact, the only possibility that remains is that I be identically zero, and that the surface be Riemannian. The classical non-Riemannian examples of Finsler metrics with constant negative curvature are the Hilbert geometries. Since the existence of an isometric embedding of a Finsler surface on a Minkowski space implies that the invariant I is bounded (this follows immediately from the geometric interpretation of I given in Section 8.1) we have the following remark of Álvarez and Durán [10]: the Hilbert geometry given by a smooth, quadratically convex curve C does not admit an isometric embedding into a Minkowski space unless C is an ellipse. If C is an ellipse, then the Hilbert geometry is the Cayley–Klein model of hyperbolic geometry and, by a theorem of Rosendorn, it admits an explicit isometric embedding into R5 (see [57, p. 276]). T HEOREM 8.15 (Akbar–Zadeh [4]). A compact Finsler surface with zero curvature is locally isometric to a Minkowski plane. P ROOF. By Theorem 8.11, we need to show that a compact Finsler surface of zero curvature also satisfies I1 ≡ 0 and I2 ≡ 0. Reasoning as in the proof of the previous theorem, if σ (t) is a geodesic on M, then the function I (t) := I (σ˙ (t)) satisfies the differential equation d2 I = 0. dt 2

Some problems on Finsler geometry

29

This implies that I is a linear function of t. Since I is bounded, it must be a constant and, therefore, I1 = 0. The proof that I2 = 0 is just slightly more involved. First note from the structure equations that when K ≡ 0 the vector fields X1 and X2 commute. From this we gather that I21 = I12 = 0, and so I2 is constant along geodesics. Using this we have that I2 = X2 I = [X3 , X1 ]I = −I31 . This implies that 0 = I21 = −I311 and, hence, I3 is a linear function of t. Since I3 is bounded, I3 must be constant on geodesics and I2 = −I31 = 0. This proof also shows that a surface with a complete Finsler metric of zero curvature that is isometrically embedded in a Minkowski space must be locally Minkowski. P ROBLEM 17. Is there any complete Finsler metric on R2 with zero curvature that is not locally Minkowski? T HEOREM 8.16. Let (M, ϕ) be a Landsberg surface. If M is connected, then the total angle of any two of its tangent unit circles is the same. The proof is taken from Bryant’s beautiful paper [30]. P ROOF. If x and y be two points on M, the difference of the total angle of the unit circle Sy over y and the total angle of the unit circle Sx over x is given by

ω3 −

Sy

ω3 , Sx

where the orientation over Sy and Sx is taken so that the integrals are positive. Let γ : [0, 1] → M be a smooth curve joining them. Let π : U M → M denote the natural projection and set C be the cylinder π −1 (γ ). Note that the oriented boundary of C is Sy −Sx and that the 2-form ω1 ∧ ω2 vanishes identically on C. Using Stokes theorem and the structure equations we have that

ω3 −

Sy

ω3 =

Sx

C

dω3 =

C

−Kω1 ∧ ω2 − J ω2 ∧ ω3 = 0.

Another application that uses the full power of Cartan’s structure equations is the Finsler version of the Gauss–Bonnet theorem given by Bao and Chern in [21]. Let (M, ϕ) be a compact, oriented Finsler surface and let X be a vector field on M with a finite number of non-degenerate zeros. Cut out small discs, say of radius r, around the zeros of X and denote the resulting manifold with boundary by Mr . Normalizing the vector field X on Mr we obtain a section σr : Mr → SMr over the unit circle bundle of Mr .

30

J.C. Álvarez Paiva

Using the fact that the total angle of the tangent unit circle at a point x is given by the integral of ω3 over Sx M, the equation dω3 = −Kω1 ∧ ω2 − J ω2 ∧ ω3 ,

(11)

and Stokes’ theorem, Chern and Bao arrive at the following result: T HEOREM 8.17 [21]. Let (M, ϕ) be a compact, oriented Finsler surface and let X be a vector field on M with zeros x1 , . . . , xn which are non-degenerate and with indices I(x1 ), . . . , I(xn ). Using the notation above, the limit as r tends to zero of the quantity Mr

σr∗ (−Kω1 ∧ ω2 − J ω2 ∧ ω3 )

(12)

is well defined and equals ni=1 I(xi )A(xi ), where A(x1 ), . . . , A(xn ) are the total angles of the unit tangent circles at the points x1 , . . . , xn . If, as in the case of Landsberg surfaces, the total angle of the unit tangent circles does not vary from point to point, then we have that the Euler characteristic of M can be written as an integral in terms of the differential invariants of the Finsler surface. Unfortunately, the Gauss–Bonnet theorem for Finsler surfaces does not have as many geometric implications as its Riemannian counterpart. For example, it cannot be used to prove that a metric of non-positive, or non-negative, curvature on a two-dimensional Finsler torus must be flat, or that two simple closed geodesics in a positively curved Finsler twosphere must intersect. The reader is invited to prove these seemingly simple results by him/herself and thereby gain some insight into some of the difficulties of extending Riemannian results to the Finsler setting. A greater conceptual challenge is that the standard Riemannian technique of comparing arbitrary metrics to metrics of constant curvature does not generalize to the Finsler setting. As we saw earlier in this section, a compact Finsler surface with constant negative curvature is Riemannian. Likewise, it has been recently remarked by Bryant that the main result of LeBrun and Mason, in [67], implies the following important result: T HEOREM 8.18 (Bryant). A Finsler metric of constant positive curvature on the twosphere is Riemannian. The analogous result in higher dimensions is still open: P ROBLEM 18 (Bryant). Is there any non-Riemannian Finsler metric on S n , n > 2, with constant (positive) curvature? It is possible that we must instead compare arbitrary Finsler metrics to metrics that have a given, simple, dynamical property. For example, there is evidence that the following generalization of the uniformization theorem is true:

Some problems on Finsler geometry

31

C ONJECTURE (Álvarez [6]). If ϕ is a Finsler metric on RP 2 , there exists a smooth function ρ on RP 2 such that all the geodesics of (RP 2 , eρ ϕ) are closed.

References [1] M. Abate and G. Patrizio, Finsler Metrics—A Global Approach, Lecture Notes in Math., vol. 1591, Springer-Verlag (1994). [2] R. Abraham and J.E. Marsden, Foundations of Mechanics, 2nd ed., Benjamin/Cummings, Reading, MA (1978). [3] H. Akbar-Zadeh, Les espaces de Finsler et certaines de leurs généralisations, Ann. Sci. École Norm. Sup. (3) 80 (1963), 1–79. [4] H. Akbar-Zadeh, Sur les espaces de Finsler à courbures sectionelles constantes, Acad. Roy. Belg. Bull. Cl. Sci. (5) 74 (1988), 281–322. [5] J.C. Álvarez Paiva, The symplectic geometry of spaces of geodesics, Ph.D. thesis, Rutgers University (1995). [6] J.C. Álvarez Paiva, Dual mixed volumes and isosystolic inequalities, preprint (2002). [7] J.C. Álvarez Paiva, Hilbert’s fourth problem in two dimensions, Mass Selecta: Teaching and Learning Advanced Undergraduate Mathematics, Amer. Math. Soc., Providence, RI (2003), 165–183. [8] J.C. Álvarez Paiva, Dual spheres have the same girth, Amer. J. Math., to appear. [9] J.C. Álvarez Paiva and G. Berck, What is wrong with the Hausdorff measure in Finsler spaces, Adv. in Math., to appear. [10] J.C. Álvarez Paiva and C. Durán, An Introduction to Finsler Geometry, Notas de la Escuela Venezolana de Matématicas (1998). [11] J.C. Álvarez Paiva and C. Durán, Isometric submersions of Finsler manifolds, Proc. Amer. Math. Soc. 129 (2001), 2409–2417. [12] J.C. Álvarez Paiva and E. Fernandes, Crofton formulas in projective Finsler spaces, Electron. Res. Announc. Amer. Math. Soc. 4 (1998), 91–100. [13] J.C. Álvarez Paiva and E. Fernandes, Fourier transforms and the Holmes–Thompson volume of Finsler manifolds, Int. Math. Res. Notices 19 (1999), 1031–1042. [14] J.C. Álvarez Paiva and E. Fernandes, Crofton formulas and Gelfand transforms, Selecta Math. (NS), to appear. [15] J.C. Álvarez Paiva, I.M. Gelfand and M. Smirnov, Crofton densities, symplectic geometry, and Hilbert’s fourth problem, Arnold–Gelfand Mathematical Seminars, Geometry and Singularity Theory, V.I. Arnold, I.M. Gelfand and V.S. Retakh, eds, Birkhäuser, Boston (1997), 77–92. [16] J.C. Álvarez and A.C. Thompson, Volumes in normed and Finsler spaces, A Sampler in Riemann–Finsler Geometry, D. Bao, R. Bryant, S.S. Chern and Z. Shen, eds, Cambridge Univ. Press (2004), 1–49. [17] V.I. Arnold, Geometrical Methods in the Theory of Ordinary Differential Equations, Grundlehren der Mathematischen Wissenschaften, vol. 250, Springer-Verlag, Berlin (1983). [18] V.I. Arnold and A.B. Givental, Symplectic geometry, Dynamical Systems IV, Encyclopaedia of Mathematical Sciences, vol. 4, V.I. Arnold and S.P. Novikov, eds, Springer-Verlag, Berlin (1990). [19] V. Bangert, On the existence of closed geodesics on the two-spheres, Internat. J. Math. 4 (1993), 403–418. [20] D. Bao, R. Bryant, S.S. Chern and Z. Shen, eds, Some Perspectives in Finsler Geometry, MSRI Series, Cambridge Univ. Press, in press. [21] D. Bao and S.S. Chern, A note on the Gauss–Bonnet theorem for Finsler spaces, Ann. of Math. 143 (1996), 233–252. [22] D. Bao, S.S. Chern and Z. Shen, eds, Finsler Geometry, Contemporary Mathematics, vol. 196, Amer. Math. Soc. (1996). [23] D. Bao, S.S. Chern and Z. Shen, An Introduction to Riemann–Finsler Geometry, Graduate Texts in Mathematics, vol. 200, Springer-Verlag, Berlin (2000). [24] V.N. Berestovskii, Submetries of space forms of nonnegative curvature, Siberian Math. J. 16 (1975), 651– 662. [25] G. Berck, Minimalité des sous-variétés totalement géodésiques, preprint (2004).

32

J.C. Álvarez Paiva

[26] L. Berwald, On Finsler and Cartan geometries. III. Two-dimensional Finsler spaces with rectilinear extremals, Ann. of Math. (2) 42 (1941), 84–112. [27] A. Besse, Manifolds of all Whose Geodesics are Closed, Springer-Verlag, Berlin (1978). [28] G.D. Birkhoff, Dynamical systems with two degrees of freedom, Trans. Amer. Math. Soc. 18 (1917), 199– 300. [29] E.D. Bolker, A class of convex bodies, Trans. Amer. Math. Soc. 145 (1969), 323–345. [30] R.L. Bryant, Finsler structures on the 2-sphere satisfying K = 1, in [22], 27–42. [31] R.L. Bryant, Projectively flat Finsler 2-spheres of constant curvature, Selecta Math. (N.S.) 3 (1997), 161– 203. [32] D. Burago and S. Ivanov, Isometric embeddings of Finsler manifolds, St. Petersburg Math. J. 5 (1994), 159–169. [33] D. Burago and S. Ivanov, On asymptotic volume of Finsler tori, minimal surfaces in normed spaces, and symplectic filling volume, Ann. of Math. (2) 156 (3) (2002), 891–914. [34] H. Busemann, A theorem on convex bodies of the Brunn–Minkowski type, Proc. Nat. Acad. Sci. USA 35 (1949), 27–31. [35] H. Busemann, The geometry of Finsler spaces, Bull. Amer. Math. Soc. 56 (1950), 5–16. [36] H. Busemann, The Geometry of Geodesics, Pure and Applied Mathematics, vol. 6, Academic Press, New York (1955). [37] H. Busemann, Geometries in which the planes minimize area, Ann. Mat. Pura Appl. (4) 55 (1961), 171–190. [38] H. Busemann, Convexity on Grassmann manifolds, Enseign. Math. (2) 7 (1961), 139–152. [39] H. Busemann, Recent Synthetic Differential Geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 54, Springer-Verlag, Berlin (1970). [40] H. Busemann, G. Ewald and G.C. Shephard, Convex bodies and convexity on Grassmann cones: I–IV, Math. Ann. 151 (1963), 1–41. [41] H. Busemann and C.M. Petty, Problems on convex bodies, Math. Scand. 4 (1956), 88–94. [42] E. Cartan, Sur un problème d’équivalence et la théorie des espaces métriques généralisés, Mathematica 4 (1930), 114–136. [43] G.D. Chakerian, Integral geometry in Minkowski spaces, in [22], 43–50. [44] S.S. Chern, Finsler geometry is just Riemannian geometry without the quadratic restriction, Notices Amer. Math. Soc. 43 (9) (1996), 959–963. [45] C.E. Durán, A volume comparison theorem for Finsler manifolds, Proc. Am. Math. Soc. 126 (1998), 3079– 3082. [46] D. Egloff, Some new developments in Finsler geometry, Inaugural-Dissertation, Universität Freiburg (Schweiz) (1995). [47] D. Egloff, Uniform Finsler Hadamard manifolds, Ann. Inst. H. Poincaré Phys. Théor. 66 (1997), 323–357. [48] D. Egloff, On the dynamics of uniform Finsler manifolds of negative flag curvature, Ann. Global Anal. Geom. 15 (1997), 101–116. [49] E. Ferrand, Sur la structure symplectique de l’espace des géodésiques d’un espace de Hadamard, Geom. Dedicata 68 (1997), 79–89. [50] P. Foulon, Géométrie des équations différentielles du second ordre, Ann. Inst. H. Poincaré Phys. Théor. 45 (1986), 1–28. [51] P. Foulon, Locally symmetric Finsler spaces in negative curvature, C. R. Acad. Sci. Paris 324 (1997), 1127– 1132. [52] P. Foulon, Curvature and global rigidity in Finsler manifolds, Houston J. Math. 28 (2002), 263–292. [53] J. Franks, Geodesics on S 2 and periodic points of annulus homeomorphisms, Invent. Math. 108 (1992), 403–418. [54] S. Goł¸ab, Quelques problèmes métriques de la géometrie de Minkowski, Trav. l’Acad. Mines Cracovie 6 (1932), 1–79. [55] P. Goodey and W. Weil, Zonoids and generalisations, Handbook of Convex Geometry, vol. B, P.M. Gruber et al., eds, North-Holland, Amsterdam (1993), 1297–1326. [56] M. Gromov, Filling Riemannian manifolds, J. Differential Geom. 18 (1983), 1–147. [57] M. Gromov, Partial Differential Relations, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 9, Springer-Verlag, Berlin (1986). [58] G. Hamel, Über die Geometrien, in denen die Geraden die kürzesten sind, Math. Ann. 57 (1903), 231–264.

Some problems on Finsler geometry

33

[59] D. Hilbert, Foundations of Geometry, Open Court Classics, Lasalle, IL (1971). [60] D. Hilbert, Mathematical problems, translation in Mathematical developments arising from Hilbert problems, Proceedings of Symposia in Pure Math., vol. XXVII, Part 1, F. Browder, ed., Amer. Math. Soc., Providence, RI (1974). [61] H. Hofer, K. Wysocki and E. Zehnder, The dynamics on three-dimensional strictly convex energy surfaces, Ann. of Math. 148 (1998), 197–289. [62] R.D. Holmes and A.C. Thompson, N -dimensional area and content in Minkowski spaces, Pacific J. Math. 85 (1979), 77–110. [63] L. Hörmander, The Analysis of Linear Partial Differential Operators I, Grundlehren der Mathematischen Wissenschaften, vol. 256, Springer-Verlag, Berlin (1983). [64] S. Ivanov, On two-dimensional minimal fillings, St. Petersburg Math. J. 13 (2002), 17–25. [65] A. Katok, Ergodic perturbations of degenerate integrable Hamiltonian systems (in English); Russian original in Math. USSR Izv. (1974), 535–571. [66] G. Landsberg, Über die Totalkrümmung, Jahresber. Deutsch. Math.-Verein. 16 (1907), 36–46. [67] C. LeBrun and L.J. Mason, Zoll manifolds and complex surfaces, J. Differential Geom. 61 (2002), 453–535. [68] D. McDuff and D. Salamon, Introduction to Symplectic Topology, 2nd. ed., Oxford Mathematical Monographs, Clarendon Press, Oxford (1998). [69] H. Minkowski, Gesammelte Abhandlungen, Chelsea, New York (1967). [70] K. Ono, Space of geodesics on Zoll three-spheres, preprint (1999). [71] V. Ovsienko and S. Tabachnikov, Sturm theory, Ghys theorem on zeros of the Schwartzian derivative, and flattening of Legendrian curves, Selecta Math. (N.S.) 2 (1996), 297–307. [72] A.V. Pogorelov, Hilbert’s Fourth Problem, Scripta Series in Mathematics, Winston and Sons (1979). [73] J.J. Schäffer, The self-circumference of polar convex bodies, Arch. Math. 24 (1973), 87–90. [74] J.J. Schäffer, Geometry of Spheres in Normed Spaces, Lecture Notes in Pure and Applied Mathematics, vol. 20, Dekker, New York (1976). [75] R. Schneider, Über die Finslerräume mit Sij kl = 0, Arch. Math. 19 (1968), 656–658. [76] R. Schneider, Crofton formulas in hypermetric projective Finsler spaces, Festschrift: Erich Lamprecht, Arch. Math. (Basel) 77 (1) (2001), 85–97. [77] R. Schneider, On integral geometry in projective Finsler spaces, Izv. Nats. Akad. Nauk Armenii Mat. 37 (2002) 34–51. [78] R. Schneider and J.A. Wieacker, Integral geometry in Minkowski spaces, Adv. in Math. 129 (1997), 222– 260. [79] Z. Shen, Curvature, distance, and volume in Finsler geometry, IHES preprint (1997). [80] J.-C. Sikorav, Rigidité symplectique dans le cotangent de T n , Duke Math. J. 59 (1989), 759–763. [81] E. Socié-Méthou, Comportements asymptotiques et rigidités en géométries de Hilbert, Thèse, Université de Strasbourg (2000). [82] Z.I. Szabo, Hilbert’s fourth problem, I, Adv. in Math. 59 (1986), 185–301. [83] A.C. Thompson, Applications of various inequalities to Minkowski geometry, Geom. Dedicata 46 (1993), 215–231. [84] A.C. Thompson, Minkowski Geometry, Encyclopedia of Math. and Its Applications, vol. 63, Cambridge Univ. Press, Cambridge (1996). [85] W. Weil, Centrally symmetric convex bodies and distributions II, Israel J. Math. 32 (1979), 173–182. [86] A. Weinstein, Fourier integral operators, quantization, and the spectra of Riemannian manifolds, Géométrie Symplectique et Physique Mathématique, Editions CNRS, Paris (1975). [87] W. Ziller, Geometry of the Katok examples, Ergodic Theory Dynam. Systems 3 (1982), 135–157.

This page intentionally left blank

CHAPTER 2

Foliations∗ Raymond Barre and Aziz El Kacimi Alaoui LAMATH, Université de Valenciennes, Le Mont Houy, 59313 Valenciennes Cedex 9, France E-mail: {raymond.barre, aziz.elkacimi}@univ-valenciennes.fr

Contents 0. Foreword . . . . . . . . . . . . . . . . . . . . . . . 1. Definitions and examples . . . . . . . . . . . . . . . 1.1. Morphisms of foliations . . . . . . . . . . . . 1.2. The concept of holonomy . . . . . . . . . . . . 1.3. Examples of foliations . . . . . . . . . . . . . 1.4. Foliations and differential systems . . . . . . . 1.5. Notations . . . . . . . . . . . . . . . . . . . . . 2. Transverse structures . . . . . . . . . . . . . . . . . 2.1. Measurable foliations . . . . . . . . . . . . . . 2.2. Lie foliations . . . . . . . . . . . . . . . . . . 2.3. Transversely parallelizable foliations . . . . . 2.4. Riemannian foliations . . . . . . . . . . . . . . 2.5. Transversely holomorphic foliations . . . . . . 3. Codimension one foliations . . . . . . . . . . . . . 4. Γ -structures . . . . . . . . . . . . . . . . . . . . . . 5. The leaf space . . . . . . . . . . . . . . . . . . . . . 5.1. V -manifolds . . . . . . . . . . . . . . . . . . . 5.2. QF -manifolds . . . . . . . . . . . . . . . . . . 5.3. Q-manifolds . . . . . . . . . . . . . . . . . . . 6. Characteristic classes . . . . . . . . . . . . . . . . . 6.1. The classifying space and the universal bundle 6.2. Classification of real vector bundles . . . . . . 7. Basic global analysis . . . . . . . . . . . . . . . . . 7.1. Foliated vector bundles and basic sections . . . 7.2. Transversely elliptic operators . . . . . . . . . 7.3. Transversely elliptic complexes . . . . . . . . 8. Deformation theory of foliations . . . . . . . . . . . 9. Some other themes . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

∗ We thank Cédric Rousseau for drawing all the pictures in this paper.

HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 35

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

37 37 39 39 40 45 46 46 46 47 48 48 49 50 51 52 52 53 53 54 54 54 55 55 56 58 61 63

36 9.1. Compact leaves . . . . . . 9.2. When is a manifold a leaf ? 9.3. Minimal leaves . . . . . . References . . . . . . . . . . . . .

R. Barre and A. El Kacimi Alaoui . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

64 64 65 66

Foliations

37

0. Foreword Foliation Theory is the qualitative study of Differential Equations. It was initiated by the works of H. Poincaré, I. Bendixon and developed later by C. Ehresmann, G. Reeb and many other people. Since then the subject has been a wide field in mathematical research. Actually it is almost impossible to describe all the results and the different steps of its development. So the purpose of this chapter is to give definitions, some examples and the fundamental concepts like holonomy, transverse structures, etc. Some themes in the point of view of Differential Geometry are discussed: characteristic classes, basic Hodge theory, deformations, etc. A complete account on Foliation Theory can be found in the book [135] by C. Godbillon. The bibliography is not complete. It is motivated by two reasons: the first one is to indicate references for the reader who wants to learn much more on foliations; the second one is to mention people who make contributions to the subject; for most of them the selected list is nonexhaustive. All foliations considered are regular that is, all leaves have the same dimension. The theory of singular foliations and specially holomorphic singular foliations is well developed with a plentiful literature. It merits to be presented independently. References on the subject can be found on the paper [56] by D. Cerveau. Unless otherwise stated, all the objects (manifolds, maps, functions, etc.) are assumed to be of class C ∞ . Moreover, for simplicity, we will suppose that all the manifolds are orientable. For any manifold M, we denote by A the algebra of functions on M. If E −→ M is a vector bundle, C ∞ (E) will denote the space of its global sections; this is an A-module and, equipped with the C ∞ -topology, it is a Fréchet space. In case E is the tangent bundle T M of M, we denote C ∞ (T M) simply by χ(M) (the space of vector fields tangent to M). For r ∈ N, Ω r (M) is the space of differential forms of degree r on M which is by definition C ∞ (Λr T ∗ M) where Λr T ∗ M −→ M is the vector bundle with fibre at x ∈ M the vector space of skew-symmetric forms of degree r on Tx M; Ω 0 (M) is just A. The other notations will be introduced at need.

1. Definitions and examples Let M be the Euclidean space Rm+n = Rm × Rn with canonical coordinates denoted (x, y) = (x1 , . . . , xm , y1 , . . . , yn ) and consider the family of affine subspaces Fy of M where y ∈ Rn , defined by the differential system: dy1 = · · · = dyn = 0. Then M, considered as a disjoint union of these spaces, is a nonconnected manifold of dimension m. Its topology is the product of the usual topology on Rm and the discrete one on Rn . We say that M, with this structure, is a foliated manifold of dimension m and codimension n. It constitutes the local model of a foliation of codimension n on a manifold of dimension m + n. Let O be an open set of Rm+n ; let us call a plaque of O any intersection of O with a horizontal space Fy . D EFINITION 1. Let M be a manifold of dimension m + n. A codimension n foliation F on M is given by an open cover U = (Ui )i∈I and for each i, a diffeomorphism

38

R. Barre and A. El Kacimi Alaoui

Fig. 1.

ϕi : Rm+n −→ Ui such that, on each nonempty intersection Ui ∩ Uj , the coordinate change ϕj−1 ◦ ϕi : (x, y) ∈ ϕi−1 (Ui ∩ Uj ) −→ (x , y ) ∈ ϕj−1 (Ui ∩ Uj ) has the form: x = ϕij (x, y)

and y = γij (y).

(1)

This means that the diffeomorphism ϕj−1 ◦ ϕi sends a plaque of ϕi−1 (Ui ∩ Uj ) into a

plaque of ϕj−1 (Ui ∩ Uj ). The manifold M is decomposed into connected submanifolds of dimension m. Each of these submanifolds is called a leaf of F . A subset U of M is saturated for F if it is union of leaves that is, if x ∈ U then the leaf passing through x is contained in U . Coordinate patches (Ui , ϕi ) satisfying conditions of Definition 1 are said to be distinguished for the foliation F . Let F be a codimension n foliation on M defined by a maximal atlas (Ui , ϕi )i∈I like in Definition 1. Let π : Rm+n = Rm × Rn −→ Rn be the second projection. Then the map π◦ϕi−1

fi : Ui −→ Rn is a submersion. On Ui ∩Uj = ∅ we have fj = γij ◦fi . The submersions fi and the local diffeomorphisms γij of Rn give a complete characterization of F . D EFINITION 2. A codimension n foliation on M is given by an open cover (Ui )i∈I , submersions fi : Ui −→ T over an n-dimensional transverse manifold T and, for Ui ∩ Uj = ∅, a diffeomorphism γij : fi (Ui ∩ Uj ) ⊂ T −→ fj (Ui ∩ Uj ) ⊂ T satisfying: fj (x) = γij ◦ fi (x)

for x ∈ Ui ∩ Uj .

(2)

We say that {Ui , fi , T , γij } is a foliated cocycle defining F . The proof of the equivalence of Definitions 1 and 2 is not difficult; it is left to the reader. The foliation F is said to be transversely orientable if T can be given an orientation preserved by all the local diffeomorphisms γij .

Foliations

39

1.1. Morphisms of foliations Let M and M be two manifolds endowed respectively with two foliations F and F . A map f : M −→ M will be called foliated or a morphism between F and F if, for every leaf L of F , f (L) is contained in a leaf of F ; we say that f is an isomorphism if, in addition, f is a diffeomorphism; in this case the restriction of f to any leaf L ∈ F is a diffeomorphism on the leaf L = f (L) ∈ F . Suppose now that f is a diffeomorphism of M with a codimension n foliation F . Then for every leaf L ∈ F , f (L) is a leaf of a codimension n foliation F on M; we say that F is the image of F by the diffeomorphism f and we write F = f ∗ (F ). Two foliations F and F on M are said to be C r -conjugated (topologically if r = 0, differentiably if r = ∞ and analytically in the case r = ω) if there exists a C r -homeomorphism f : M −→ M such that f ∗ (F ) = F .

1.2. The concept of holonomy This is a very important notion in Foliation Theory. In many situations it determines completely the structure of the foliation. In this subsection, we will introduce this concept and give the statement of the local and global stability theorems. Let F be a codimension n foliation on M, let L be a leaf of F and x ∈ L. Let T be a small transversal to F passing through x. Let σ : [0, 1] −→ L be a continuous path such that σ (0) = σ (1) = x. Then there exist a finite open cover Ui , i = 0, 1, . . . , k, of M with U0 = Uk and a subdivision 0 = t0 < t1 < · · · < tk = 1 of [0, 1] such that: – σ ([ti−1 , ti ]) ⊂ Ui , – if Ui ∩ Uj = ∅ then Ui ∪ Uj is contained in a distinguished chart of F . We say that Ui is a subordinated chain to σ . For i = 0, 1, . . . , k let Ti be a small transversal to F passing through σi (t) with T0 = Tk = T . For every point z ∈ Ti , sufficiently close to σ (ti ), the plaque of F passing through z intersects Ti+1 in a unique point fi (z). The domain of fi contains a transversal Ti passing through σ (ti ) and homeomorphic to an open ball of Rn . Then, it is clear that the map: fσ = fk−1 ◦ fk−2 ◦ · · · ◦ f0 is well defined on an open neighborhood of x; it is called the holonomy map associated to σ . We can prove (see [39], for instance) that the germ of fσ : – does not depend on the chain Ui , i = 1, . . . , k, and in the choice of σ in its homotopy class in the group π1 (L, x) of the homotopy classes of loops based at x, – satisfies fσ (x) = x. So we get a homomorphism h : [σ ] ∈ π1 (L, x) −→ fσ ∈ G(T , x) where G(T , x) is the group of germs of diffeomorphisms of T fixing the point x. This representation h is called the holonomy of the leaf L at x. It is trivial if L is simply connected. The foliation F is said to be without holonomy if this representation is trivial for every leaf L of F and every point x ∈ L. T HEOREM 1 (Local stability). Suppose that F admits a compact leaf L with finite fundamental group. Then L admits a saturated neighborhood V such that every leaf contained in V is compact with finite fundamental group.

40

R. Barre and A. El Kacimi Alaoui

T HEOREM 2 (Global stability). Suppose that M is compact, the codimension of F is one and that F admits a compact leaf with finite fundamental group. Then all leaves of F are compact with finite fundamental group. The proof can be found in the original paper of G. Reeb [285] or in the book [39] by C. Camacho and A. Lins Neto.

1.3. Examples of foliations (i) Simple foliations. On every manifold M we have a foliation by taking points as leaves. Its codimension is equal to the dimension of M. Also M can be equipped with a codimension zero foliation with only one leaf, namely, M itself. π In general, every submersion M −→ B with connected fibres defines a foliation. The leaves being the fibres π −1 (b), b ∈ B. In particular, every product F × B is a foliation with leaves F × {b}, b ∈ B. These foliations are transversely orientable if, and only if, the manifold B is orientable. These are simple foliations. We shall give more interesting examples in different situations. (ii) One-dimensional foliations. Let us begin by surfaces. Let M˜ = R2 and consider the differential equation dy −α dx = 0 where α is a real number. This equation has y = αx +c, c ∈ R, as general solution. When c varies, we obtain a family of parallel lines which defines ˜ a foliation F˜ in M. The natural action of Z2 on M˜ preserves the foliation F˜ (i.e. the image of any leaf of F˜ by an integer translation is a leaf of F˜ ). Then F˜ induces a foliation F on the torus T2 = R2 /Z2 . The leaves are all diffeomorphic to the circle S1 if α is rational and to the real line if α is not rational (Figure 2). In fact, if α is not rational, every leaf of F is dense; this shows that even if locally a foliation is simple, globally it can be complicated. Let M be a closed orientable surface. The fact that M admits a one-dimensional foliation depends on the topology of M, which is described by the Euler–Poincaré number χ(M);

Fig. 2.

Foliations

41

Fig. 3.

Fig. 4.

this number can be defined as follows: take a triangulation of M, i.e. a decomposition of M into triangles such as shown for the 2-sphere S2 (Figure 3). Let b0 , b1 and b2 be the numbers respectively of vertices, edges and triangles. Then χ(M) = b0 − b1 + b2 is independent of the triangulation; it is called the Euler–Poincaré number of M. (There are many books on Algebraic or Differential Topology where we can find the proof of this fact.) It classifies completely the topology of closed orientable surfaces, i.e. M and M are homeomorphic if, and only if, χ(M) = χ(M ). For the triangulation of S2 in Figure 3 we have b0 = 4, b1 = 6 and b2 = 4. So χ(S2 ) = 2. The Euler–Poincaré number of M is the only obstruction to the existence of dimension one foliation on M: M admits such foliation if, and only if, χ(M) = 0. For example, S2 cannot support a one-dimensional foliation. In fact, T2 is the only one compact orientable surface which admits a foliation of dimension one. The reader can prove, by using an adequate triangulation, that a closed orientable surface Mg of genus g (see in Figure 4 the case g = 2) has χ(Mg ) = 2−2g as Euler–Poincaré number. Then Mg admits a foliation F of dimension one if, and only if, g = 1, i.e. Mg is T2 . Suppose M is compact of dimension n. For each r = 0, 1, . . . , n, let H r (M, R) denote the real rth cohomology space of M which is finite dimensional. Then the number n (−1)r dim H r (M, R) χ(M) = r=0

42

R. Barre and A. El Kacimi Alaoui

is a topological invariant called the Euler–Poincaré number of M. For a surface, it is exactly the number defined above by using a triangulation. The manifold M admits a onedimensional foliation if, and only if, χ(M) = 0. (iii) Reeb foliation on the 3-sphere S3 . Let M be the 3-dimensional sphere S3 = ¯ its closure {(z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 1}. Denote by D the open unit disc in C and D which is the closed unit disc {z ∈ C: |z| 1}. The two subsets: 1 M+ = (z1 , z2 ) ∈ S3 : |z1 |2 2

1 and M− = (z1 , z2 ) ∈ S3 : |z2 |2 2

¯ × S1 . They have T2 as common boundary: are diffeomorphic to D 1 ∂M+ = ∂M− = (z1 , z2 ) ∈ S3 : |z1 |2 = |z2 |2 = 2 and their union is equal to S3 . Then S3 can be obtained by gluing M+ and M− along their boundaries by the diffeomorphism (z1 , z2 ) ∈ ∂M + −→ (z2 , z1 ) ∈ ∂M− , i.e. we identify (z1 , z2 ) with (z2 , z1 ) in the disjoint union M+ M− . Let f : D −→ R be the function defined by: f (z) = exp

1 . 1 − |z|2

Let t denote the second coordinate in D × R. The family of surfaces (St )t∈R obtained by translating the graph S of f along the t-axis defines a foliation on D × R. If we add the ¯ we obtain a codimension one cylinder S1 × R, where S1 is viewed as the boundary of D, ¯ ˜ ˜ foliation F on D × R. By construction, F is invariant by the transformation ¯ × R −→ (z, t + 1) ∈ D ¯ × R; (z, t) ∈ D so it induces a foliation F0 on the quotient: ¯ × R/(z, t) ∼ (z, t + 1) D ¯ × S1 . D It has the boundary T2 = S1 × S1 as a closed leaf. The others are diffeomorphic to R2 (see Figure 5). ¯ × S1 , F0 defines on M+ and M− respecBecause M+ and M− are diffeomorphic to D tively two foliations F+ and F− which give a codimension one foliation F on S3 called the Reeb foliation. (iv) Lie group actions. Let M be a manifold of dimension m + n and G a connected Lie Φ

group of dimension m. An action of G on M is a map G × M −→ M such that: – Φ(e, x) = x for every x ∈ M (where e is the unit element of G), – Φ(g , Φ(g, x)) = Φ(g g, x) for every x ∈ M and every g, g ∈ G.

Foliations

43

Fig. 5.

Suppose that, for every point x ∈ M, the dimension of the isotropy subgroup: Gx = g ∈ G: Φ(g, x) = x is independent of x. Then the action Φ defines a foliation F of dimension = m − dim Gx ; its leaves are the orbits {Φ(g, x): g ∈ G}. In particular, this is the case if Φ is locally free, i.e. if, for every x ∈ M, the isotropy subgroup Gx is discrete. An explicit example is given when M is the quotient H /Γ of a Lie group H by a discrete subgroup Γ and G is a connected Lie subgroup of H ; the action of G on M being induced by the left action of G on H . We say that F is a homogeneous foliation. Let us give an explicit example (for more details see [93]). Let A ∈ SL(m + n − 1, Z), where m + n 3, be a matrix diagonalizable and having all its eigenvalues µ1 , . . . , µm−1 , λ1 , . . . , λn real and positive. We can think of A as a diffeomorphism of the (m + n − 1)-torus Tm+n−1 . Let X1 , . . . , Xm−1 , Y1 , . . . , Yn be linear vector fields on Tm+n−1 such that: A∗ Xj = µj Xj ,

A∗ Yk = λk Yk

for j = 1, . . . , m − 1 and k = 1, . . . , n,

and denote by F0 the foliation on Tm+n−1 defined by the vector fields X1 , . . . , Xm−1 . The product of F0 by R gives a codimension n foliation on Tm+n−1 × R which is invariant by the diffeomorphism φ of Tm+n−1 × R sending (z, t) to (A(z), t + 1). So, it induces a m+n codimension n foliation F on the quotient manifold TA = Tm+n−1 × R/φ. Notice that m+n 1 m+n−1 . In fact Tm+n is the homoTA is a flat bundle over the circle S with fibre T A geneous space H /Γ where H is the semi-direct product of Rm+n−1 by R given by the action: (t, z) ∈ R × Rm+n−1 −→ At z ∈ Rm+n−1

44

R. Barre and A. El Kacimi Alaoui

and Γ is the subgroup:

(m, k) ∈ H | m ∈ Zm+n−1 , k ∈ Z .

If v1 , . . . , vm−1 ∈ Rm+n−1 are eigenvectors of A corresponding respectively to the eigenvalues µ1 , . . . , µm−1 then the subgroup: G=

m−1

ai vi , b ∈ H a1 , . . . , am−1 , b ∈ R

i=1

is isomorphic to the semi-direct product of Rm−1 by R∗+ where R∗+ acts on Rm−1 by homotheties on each factor. The action of G on Tm+n A , induced by this identification, is a locally free action whose orbits define the foliation F . (v) Foliations obtained by suspension. Let B and F be two manifolds, respectively of dimensions m and n. Suppose that the fundamental group π1 (B) of B is finitely generated. Let ρ : π1 (B) −→ Diff(F ) be an injective representation, where Diff(F ) is the diffeomorphism group of F . Denote by B˜ the universal covering of B and F˜ the horizontal foliation on M˜ = B˜ × F , i.e. the foliation whose leaves are the subsets B˜ × {y}, y ∈ F . This foliation is invariant by all the transformations Tγ : M˜ −→ M˜ defined by Tγ (x, ˜ y) = (γ · x, ˜ ρ(γ )(y)) ˜ then F˜ induces a codimension n foliwhere γ · x˜ is the natural action of γ ∈ π1 (B) on B; ation Fρ on the quotient manifold: ˜ x, M = M/( ˜ y) ∼ γ · x, ˜ ρ(γ )(y) . We say that Fρ is the suspension of the diffeomorphism group Γ = ρ(π1 (B)). The leaves of Fρ are transverse to the fibres of the natural fibration induced by the first projection ˜ B˜ × F −→ B.

Fig. 6.

Foliations

45

π

Conversely, suppose that F −→ M −→ B is a fibration with compact fibre F and that F is a codimension n foliation (n = dimension of F ) transverse to the fibres of π . Then there exists a representation ρ : π1 (B) −→ Diff(F ) such that F = Fρ . Concrete example: let B be the circle S1 and F = R+ = [0, +∞[. Let ρ be the representation of Z = π1 (S1 ) in Diff([0, +∞[) defined by ρ(1) = ϕ where ϕ(y) = λy with λ ∈ ]0, 1[. Because ϕ is isotopic to the identity map of F , the manifold M is diffeomorphic to S1 × R+ and the foliation Fρ has one closed leaf diffeomorphic to the circle S1 , corresponding to the fixed point ϕ(0) = 0 (see Figure 6).

1.4. Foliations and differential systems Let M be a manifold of dimension m + n. Denote by T M the tangent bundle of M and let E be a subbundle of rank m. Let U be an open set of M such that on U , T M is equivalent to the product U × Rm+n . At each point x ∈ U , the fibre Ex can be considered as the kernel of n differential 1-forms ω1 , . . . , ωn linearly independent: Ex =

n

ker ωj (x).

(S)

j =1

The subbundle E is called an m-plane field on M. We say that E is involutive, if for every vector field X and Y tangent to E (i.e. sections of E), the bracket [X, Y ] is also tangent to E. We say that E is completely integrable if, through each point x ∈ M, there exists a submanifold Px of dimension m which admits E|Px (the restriction of E to Px ) as tangent bundle. The maximal connected submanifolds satisfying this property are called the integral submanifolds of the differential system (S). They define a partition of M, i.e. a codimension n foliation. We have T HEOREM 3 (Frobenius). Let E be a subbundle of rank m given locally by a differential system like in (S). Then the following assertions are equivalent: – E is involutive, – E is completely integrable, – there n exist differential 1-forms (defined locally) (βij ), i, j = 1, . . . , n, such that dωi = j =1 βij ∧ ωj , i = 1, . . . , n. For example, let ω be a nonsingular 1-form. The corresponding subbundle E has fibre Ex = ker(ωx ). It defines a codimension one foliation if, and only if, there exists a 1-form β such that dω = β ∧ω; this condition is equivalent to dω ∧ω = 0. In particular, if ω is closed it defines a codimension one foliation F . If M is compact, all leaves are diffeomorphic and integration of ω over loops of M gives rise to a morphism h : π1 (M) −→ R. The range Γ = h(π1 (M)) of h is a subgroup of R called the holonomy group of F . Example (ii) is of this type: M = T2 , ω = dy − α dx which is closed. The fundamental group of T2 is Z ⊕ Z and it is easy to see that Γ = {p + qα: p, q ∈ Z}.

46

R. Barre and A. El Kacimi Alaoui

1.5. Notations Let F be a codimension n foliation on M. We denote by T F the tangent bundle to F and νF the quotient T M/T F which is the normal bundle to F ; χ(F) will denote the space of sections of T F (elements of χ(F) are vector fields X ∈ χ(M) tangent to F ). A differential form α ∈ Ω r (M) is said to be basic if it satisfies iX α = 0 and LX α = 0 for every X ∈ χ(F). (Here iX and LX denote respectively the inner product and the Lie derivative with respect to the vector field X.) For a function f : M −→ R, these conditions are equivalent to X · f = 0 for every X ∈ χ(F), i.e. f is constant on the leaves of F ; we denote by Ω r (M/F) the space of basic forms of degree r on the foliated manifold (M, F); this is a module over the algebra Ab of basic functions. A vector field Y ∈ χ(M) is said to be foliated, if for every X ∈ χ(F), the bracket [X, Y ] ∈ χ(F). We can easily see that the set χ(M, F) of foliated vector fields is a Lie algebra and an Ab -module; by definition χ(F) is an ideal of χ(M, F) and the quotient χ(M/F) = χ(M, F)/χ(F) is called the Lie algebra of transverse (or basic) vector fields on the foliated manifold (M, F). Also, it has a module structure over the algebra Ab .

2. Transverse structures Let M be a manifold of dimension m + n endowed with a codimension n foliation F defined by a foliated cocycle {Ui , fi , T , γij } like in Definition 2. D EFINITION 3. A transverse structure to F is a geometric structure on T invariant by the local diffeomorphisms γij . This is a very important notion in Foliation Theory. To make it clear, let us give the main examples.

2.1. Measurable foliations Let BT denote the family of Borel sets on T . A transverse invariant measure to F is a measure µ on BT such that, for any A ∈ BT in the domain of definition of γij , we have µ γij (A) = µ(A). We say that F is a measurable foliation if it admits a transverse measure. The notion of measurable foliation was introduced firstly by J.F. Plante; he obtained many interesting results on the qualitative behavior of codimension one measurable foliations on compact manifolds (cf. [273]).

Foliations

47

2.2. Lie foliations We say that F is a Lie foliation, if T is a Lie group G and γij are restrictions of left translations on G. Such foliation can also be defined by a 1-form ω on M with values in the Lie algebra G such that: (i) ωx : Tx M −→ G is surjective for every x ∈ M, (ii) dω + 12 [ω, ω] = 0. If G is Abelian, ω is given by n linearly independent closed scalar 1-forms ω1 , . . . , ωn . In particular, if n = 1, an important topological property of compact manifolds supporting such foliation is given by the following theorem due to Tischler [350]. T HEOREM 4. If a compact manifold admits a closed nonsingular 1-form, then it is a locally trivial fibration over the circle S1 . The hypothesis G is Abelian is important: D. Lehmann [211] proved that, in general, the result is false even if G is nilpotent. Foliations defined by nonsingular closed 1-forms can be considered as topological prototype of codimension one foliations without holonomy as it is illustrated by Sacksteder’s theorem [311]: T HEOREM 5. Let F be a C r (r 2) codimension one foliation on a connected compact manifold. If F has no holonomy, then it is topologically conjugated to a foliation defined by a nonsingular closed 1-form. In the general case, the structure of a Lie foliation on a compact manifold, is given by the following theorem due to E. Fédida [101]: T HEOREM 6. Let F be a Lie G-foliation on a compact manifold M. Let M˜ be the universal ˜ Then there exist a homomorphism h : π1 (M) −→ G covering of M and F˜ the lift of F to M. ˜ and a locally trivial fibration D : M −→ G whose fibres are the leaves of F˜ and such that, for every γ ∈ π1 (M), the following diagram is commutative: γ −→ M˜ M˜ D↓ ↓D

G

h(γ )

−→

G

˜ where the first line denotes the deck transformation of γ ∈ π1 (M) on M. The subgroup Γ = h(π1 (M)) ⊂ G is called the holonomy group of F although the holonomy of each leaf is trivial. The fibration D : M˜ −→ G is called the developing map of F .

48

R. Barre and A. El Kacimi Alaoui

2.3. Transversely parallelizable foliations We say that F is transversely parallelizable if there exist on M, foliated vector fields Y1 , . . . , Yn , transverse to F and everywhere linearly independent. This means that the manifold T admits a parallelism (Y1 , . . . , Yn ) invariant by all the local diffeomorphisms γij or, equivalently, that the Ab -module χ(M/F) is free of rank n. The structure of a transversely parallelizable foliation on a compact manifold is given by the following theorem due to L. Conlon [60] for n = 2 and in general to P. Molino [243]. T HEOREM 7. Let F be a transversely parallelizable foliation of codimension n on a compact manifold M. Then: (1) the closures of the leaves are submanifolds which are fibres of a locally trivial fibration π : M −→ W where W is a compact manifold, (2) there exists a simply connected Lie group G0 such that the restriction F0 of F to any leaf closure F is a G0 -Lie foliation, (3) the cocycle of the fibration π : M −→ W has values in the group of diffeomorphisms of F preserving F0 . The fibration π : M −→ W and the manifold W are called respectively the basic fibration and the basic manifold associated to F . Theorem 7 says that if, in particular, the leaves of F are closed, then the foliation is just a fibration over W . This is still true even if the leaves are not closed: the manifold M is a fibration over the leaf space M/F which is, in this case, a Q-manifold in the sense of [13]. Theorem 7 is still valid for transversely complete foliations on noncompact manifolds (cf. [242]). It is not difficult to see that any Lie foliation is transversely parallelizable. This is a consequence of the fact that a Lie group is parallelizable and that the parallelism can be chosen invariant by left translations. 2.4. Riemannian foliations The foliation F is said to be Riemannian if there exists on T a Riemannian metric such that the local diffeomorphisms γij are isometries. Using the submersions fi : Ui −→ T one can construct on M a Riemannian metric which can be written in local coordinates: ds 2 =

m i,j =1

θi ⊗ θj +

n

gk (y) dyk ⊗ dy .

k,=1

Equivalently, F is Riemannian, if any geodesic orthogonal to the leaves at a point is orthogonal to the leaves everywhere [289]. Let F be Riemannian. Then there exists a Levi-Civita connection, transverse to the leaves which, by unicity argument, coincides on any distinguished open set, with the pullback of the Levi-Civita connection on the Riemannian manifold T . This connection is said τ to be projectable. Let O(n) −→ M # −→ M be the principal bundle of orthonormal frames transverse to F ; this is an F -bundle, in the sense of Section 7.1. The following theorem is due to P. Molino [243].

Foliations

49

T HEOREM 8. Suppose M is compact. Then, the foliation F can be lifted to a foliation F # on M # of the same dimension and such that: (1) F # is transversely parallelizable, (2) F # is invariant under the action of O(n) on M # and projects, by τ , on F . π#

The basic manifold W # and the basic fibration F # −→ M # −→ W # are called respectively the basic manifold and the basic fibration of F . We have the following properties: – the restriction of τ to a leaf of F # is a covering over a leaf of F . So all leaves of F have the same universal covering, – the closure of any leaf of F is a submanifold of M and the leaf closures define a singular foliation (the leaves have different dimensions) on M. (For more details about this notion see [243].) Another interesting result for Riemannian foliations is the Global Reeb Stability Theorem which is valid even if the codimension is greater than 1. T HEOREM 9. Let F be a Riemannian foliation on a compact manifold M. If there exists a compact leaf with finite fundamental group, then all leaves are compact with finite fundamental group. The property F is Riemannian means that the leaf space Q = M/F is a Riemannian manifold even if Q does not support any differentiable structure!

2.5. Transversely holomorphic foliations The foliation F is said to be transversely holomorphic if T is a complex manifold and the γij are local biholomorphisms. Particular case is a holomorphic foliation: the manifolds M and T are complex, all the fi are holomorphic and all γij are local biholomorphisms. If T is Kählerian and γij biholomorphisms which preserve the Kähler form on T we say that F is transversely Kählerian. For example, any codimension 2 Riemannian foliation which is transversely orientable is transversely Kählerian. Let us give concrete examples of such foliations. Let M be the unit sphere in the Hermitian space Cn+1 : M = S2n+1 = (z1 , . . . , zn+1 ) ∈ Cn+1 :

n+1

|zk |2 = 1 .

k=1

Let Z be the holomorphic vector field on Cn+1 given by the formula: Z=

n+1 k=1

ak zk

∂ , ∂zk

50

R. Barre and A. El Kacimi Alaoui

where ak = αk + iβk ∈ C. There exists a good choice of the numbers ak such that the orbits of Z intersect transversely the sphere M; then Z induces on M a real vector field X which defines a foliation F . It is not difficult to see that F is transversely holomorphic. It is transversely Kählerian if we choose in addition αk = 0 for any k = 1, . . . , n + 1. 3. Codimension one foliations Codimension one foliations constitute a rich theme which was studied extensively by many people. The richness comes from the existence, for such foliations, of nonsingular transverse vector fields which give a way to go from a leaf to an other. Most of the results in Foliation Theory were first obtained in the codimension one case; this section is devoted to summarize some of them. Let F be a codimension one foliation on a manifold M and ν a transverse vector bundle to F . Because ν is of rank one, it is integrable and defines a foliation V transverse to F . So we have clearly χ(M) = 0. It is natural to ask if this condition is sufficient for the existence of a codimension one foliation on M; this was conjectured by E. Thomas [342]. The reader can see the paper [207] by B. Lawson about the history of the different steps for solving this conjecture. The final solution was given by W. Thurston [348] who proved T HEOREM 10 (Thurston). Let M be a compact manifold. Then M admits a codimension one foliation if, and only if, the Euler–Poincaré number χ(M) of M is zero. Recall that two vector bundles E −→ M and E −→ M are said to be homotopic if there exists a continuous family Et −→ M, t ∈ [0, 1], of vector bundles such that E0 = E and E1 = E . So we can formulate the question of existence of codimension one foliations, in general, in the following: Let M be a compact manifold. Then any codimension one plane field on M is homotopic to an integrable one. The first results solving (in some particular cases), this conjecture were obtained by J. Wood (see [376]) and also by P. Schweitzer and W. Thurston in the C 0 -case. As far as we know this conjecture is still open. Notice that the compactness of the manifold is a big constraint. Indeed on open manifolds the answer to this conjecture is positive [266]. The regularity property seems to be very important in the existence of foliations on compact manifolds. In particular, there is a big difference in the treatment between the C ∞ case and the real analytic one. In this direction A. Haefliger proved in [143] the following important theorem. T HEOREM 11 (Haefliger). Let M be a compact manifold with a finite fundamental group. Then M has no real analytic codimension one foliation. Let us end this section with one of the most important results obtained in codimension one foliation theory on 3-manifolds [261].

Foliations

51

T HEOREM 12 (Novikov). Let M be a compact 3-manifold with a finite fundamental group. Then any codimension one foliation on M has a compact leaf diffeomorphic to the torus T2 . 4. Γ -structures This notion was introduced by A. Haefliger and became a key ingredient in studying characteristic classes of foliations. D EFINITION 4. A groupoid is given by a set Γ , a subset Γ (2) of Γ × Γ with a law (γ , σ ) ∈ Γ (2) −→ γ σ ∈ Γ and an inverse map γ ∈ Γ −→ γ −1 ∈ Γ satisfying the following properties: (i) (γ −1 )−1 = γ , (ii) if (γ , σ ), (σ, τ ) ∈ Γ (2) , then (γ σ, τ ), (γ , σ τ ) ∈ Γ (2) and (γ σ )τ = γ (σ τ ), (iii) if (γ −1 , γ ) ∈ Γ (2) and (γ , σ ) ∈ Γ (2) , then γ −1 (γ σ ) = σ , (iv) if (γ , γ −1 ) ∈ Γ (2) and (τ, γ ) ∈ Γ (2) , then (τ γ )γ −1 = τ . For γ ∈ Γ , s(γ ) = γ −1 γ is called the source of γ and r(γ ) = γ γ −1 the range of γ . Then, there are two projections s, r (or α, β) : Γ −→ Γ (0) = Im r. The subset Γ (0) of Γ is called the unit space of Γ . A topological groupoid is a groupoid with a topology compatible with the composition and inverse maps. As a consequence, the two projections s, r on the unit space are also continuous. A differentiable structure on Γ is given by a manifold structure on Γ and Γ (0) compatible with the composition and inverse maps and such that: – s : Γ −→ Γ (0) is a submersion, – the canonical injection Γ (0) −→ Γ is an embedding. The differentiable (or topological) groupoid Γ is étale if s is étale. Let M be a manifold, Γ a topological groupoid and {Ui } an open cover of M; a 1-cocycle on M with values in Γ is given as follows: for each pair (i, j ), let γij : Ui ∩ Uj −→ Γ be a continuous map such that, if x ∈ Ui ∩ Uj ∩ Uk , then (γij (x), γj k (x)) ∈ Γ (2) and γik (x) = γij (x)γj k (x). Two 1-cocycles are said to be cohomologous if they are restrictions of the same cocyle on the union of their coverings. A Γ -structure on M, or an element of H 1 (M, Γ ), is an equivalence class of 1-cocycles. Let Γ be the groupoid of germs of local diffeomorphisms of Rn ; then the unit space Γ (0) may be identified to Rn . A codimension n foliation F on M may be viewed as a particular Γ -structure for which a representative is a 1-cocycle on an open covering {Ui } such that the following maps fi = γii : Ui −→ Γ (0) = Rq are submersions.

52

R. Barre and A. El Kacimi Alaoui

5. The leaf space Let F be a codimension n foliation on M. Let U be a subset of M and denote by Uˆ the union of the leaves intersecting U . Recall that U is saturated if U = Uˆ . It is easy to see that if U is open, so is Uˆ . Then, the equivalence relation on M, x ∼ y if, and only if, x and y are in the same leaf, is open. The set of equivalence classes M/ ∼, endowed with the quotient topology, is called the leaf space of F and usually denoted by M/F . We can think of M/F as follows. The foliation F is the geometric realization of a completely integrable differential system (S) on M. Each integral submanifold is a leaf of F and corresponds to an initial condition of (S). So we can consider Q as a parameter space of the initial conditions of this differential system. In general Q is not a manifold, but we can define on this space many geometrical objects like functions, differential forms, differential operators, etc. (cf., for instance, Section 7). They correspond to their analogues on M invariant along the leaves (in a sense to be determined following the context). There were many attempts to give the leaf space of a foliation a differentiable structure, even if its topology is, generally poor. A first one was from Satake, whose point of view was recovered by W. Thurston. In other domains, let us cite G.W. Mackey [219] who introduced the virtual group notion in Ergodic Theory and M. Artin [12] the algebraic space notion. The former corresponds to the measurable version of the S-atlas of W.T. Van Est, the latter suggested the definition of a Q-manifold. In fact, there is no uniform definition. Each corresponds to a given situation or a particular problem. Nevertheless, the point of view of Noncommutative Geometry, by A. Connes, using the C ∗ -algebra of the groupoid of a foliation, or the crossed-product of the C ∗ -algebra of a manifold by a group acting on it, is attractive and efficient too. For example, there are Longitudinal and Transversal Index Theorems for foliations; one gets also Godbillon–Vey classes, etc.

5.1. V -manifolds Let Ω be an open set in Rn and let Σ be a finite group of diffeomorphisms of Ω. Denote by Ω/Σ the orbit space with quotient topology and p the canonical projection Ω −→ Ω/Σ. If Ω is another open set of Rn , Σ a finite group of diffeomorphisms of Ω and p the canonical projection Ω −→ Ω /Σ , then a morphism from Ω/Σ to Ω /Σ is a continuous map f from Ω/Σ to Ω /Σ , which admits local coverings by smooth local maps from Ω to Ω . An isomorphism is a bijective morphism, the inverse of which is a morphism. If V is a second countable Hausdorff space, a Satake atlas of dimension n is a family A = (Ui , Φi ) where (Ui ) is an open covering of V and Φi : Ui −→ Ωi /Σi is a homeomorphism of Ui on the quotient of an open subset Ωi of Rn by a finite group of diffeomorphisms, with following coherence condition: for all i, j such that Ui ∩ Uj = ∅ the map Φj ◦ Φi−1 : Φi (Ui ∩ Uj ) −→ Φj (Ui ∩ Uj ) is a morphism as previously defined. A V -manifold (or a Satake manifold or an orbifold) of dimension n is a space V with a maximal Satake atlas of dimension n. The following are simple examples illustrating the notion of a V -manifold.

Foliations

53

(i) Let Γ be a finite group of isometries of a Riemannian manifold M of dimension n. Then the quotient space M/Γ is a V -manifold of dimension n. (ii) A waterdrop obtained by gluing two open discs along their boundaries, one of them being implemented with a rotation of 2π 3 around its center. It is proved in [243] that every leaf space of a Riemannian foliation with compact leaves on a compact manifold is a V -manifold. Conversely: every compact V -manifold is the leaf space of a Riemannian foliation with compact leaves on a compact manifold (cf. [132]). 5.2. QF -manifolds Let (X, p, S) be a triple where X is a manifold, S a set and p a surjective map from X to S; this is an étale QF -atlas of S if it satisfies the following conditions: (H) for every pair (x, y) in X 2 such that p(x) = p(y), there are open neighborhoods U and V respectively of x and y and a diffeomorphism h from U to V such that h(x) = y and p ◦ h(t) = p(t) for every t ∈ U , (QF) every morphism from a manifold Z to X such that p ◦ f is constant is locally constant. As usual two étale QF -atlases (X1 , p1 , S) and (X2 , p2 , S) are equivalent if (X, p, S) is an étale QF -atlas where X is the disjoint union of X1 and X2 and p is p1 on X1 and p2 on X2 . A QF -manifold structure on S is an equivalence class of étale QF -atlases on S. All the leaf spaces of foliated second countable manifolds are in this category. 5.3. Q-manifolds Let (X, p, S) be a triple where X is a manifold, S a set and p a surjective map from X to S; this is a Q-atlas of S if it satisfies the following conditions: (H) is as in the definition of an étale QF -atlas, (Q) let f = (f1 , f2 ) be a morphism from a manifold Z to X 2 such that p ◦ f1 = p ◦ f2 ; then the subset T = {z ∈ Z: f1 (z) = f2 (z)} is open in Z. A Q-manifold structure on S is an equivalence class of Q-atlases of S. The following are examples of Q-manifolds: (i) the leaf space of foliated torus with geodesics having irrational slope, (ii) more generally, the leaf space of a transversely parallelizable foliation on a compact manifold. It was first tried to generalize to leaf spaces the classical theorems and tools (Gauss– Bonnet, de Rham cohomology, Poincaré duality, Leray–Serre spectral sequence, fundamental group, etc.) to get results on the transverse structure. The V -manifolds are met in natural way and there exist many examples of them. They appear also with ramified coverings. The Q-manifolds permitted to restore the third Lie theorem for Banach Lie algebras (cf. [268]); they appear also in the structure theorem of P. Molino. Recently, G. Meigniez got a characterization of Godbillon Homotopy Extension Property for foliations, where they play a role (cf. [231]).

54

R. Barre and A. El Kacimi Alaoui

6. Characteristic classes We follow here the lectures of R. Bott as written by L. Conlon in [23]. We will restrict ourself to only one result: Bott vanishing theorem. The reader can find more material in [23] or in [147].

6.1. The classifying space and the universal bundle Let H be a separable real Hilbert space with norm . If u and v are nonzero vectors in H, (u, v) will be the angle defined by u and v; it is immediate to see that for every positive number λ and µ we have (λu, µv) = (u, v). Denote by BGLn the set of n-dimensional subspaces of H. Let τ, σ ∈ BGLn and set δ(τ, σ ) = inf(u, v) where the infimum is taken over all the vector u ∈ τ and v ∈ σ with u = v = 1. It is not difficult to see that δ defines a distance on BGLn . The topological space BGLn is called the classifying space of the group GL(n, R) of linear transformations of the vector space Rn . The cohomology H ∗ (BGLn , R) of BGLn is a polynomial ring R[p1 , . . . , p[n/2] ] where the pi ∈ H 4i (BGLn , R) are the universal Pontryagin classes (cf. [21]). On BGLn we have a canonical real vector bundle S −→ BGLn of rank n whose fibre at each τ is the space τ itself; it is called the universal bundle on BGLn .

6.2. Classification of real vector bundles As M is paracompact, it admits a countable locally finite open cover U = {Ui } which, in addition, can be chosen such that each finite intersection Ui1 ∩ · · · ∩ Ui is contractible. Such an open cover is called a good cover; it always exists: take a Riemannian metric on M π and a countable family of geodesically convex open balls which covers M. If E −→ M is a real vector bundle of rank n, its restriction E|Ui to any Ui is trivial, i.e. there exists a diffeomorphism ϕi : E|Ui −→ Ui × Rn which sends the fibre Ex isomorphically on {x} × Rn . Let (si1 , . . . , sin ) be a basis of the free module C ∞ (E|Ui ) over the algebra A(Ui ) of real valued C ∞ -functions on Ui . Let {ρi } be a partition of the unity subordinated to {Ui } and let Vi be the real vector space spanned by (ρi si1 , . . . , ρi sin ). For each i we set ψi = qi ◦ ϕi where qi : Ui × Rn −→ Rn is the second projection. Now express H as orthogonal direct sum of the Vi Rn and denote by ζi : Vi −→ H the inclusion of the ith summand. Define Φ : E −→ H by Φ(x, ξ ) =

∞

ρi (x) · ζi ψi (x, ξ ) .

i=1

Then Φ is continuous and sends each fibre π −1 (x) of E isomorphically on an n-dimensional subspace of H. Thus f (x) = Φ(π −1 (x)) defines a continuous map f : M −→ BGLn called the classifying map for the vector bundle E namely E is the pullback by f of the universal bundle S −→ BGLn . In fact there is a natural one–one correspondence between

Foliations

55

the set Vectn (M) of isomorphism classes of real vector bundles of rank n on M and the set [M, BGLn ] of homotopy classes of maps M −→ BGLn . The ith-Pontryagin class of the real vector bundle E −→ M is by definition pi (E) = f ∗ (pi ). The graded subring Pont∗ (E) = f ∗ H ∗ (BGLn , R) ⊂ H ∗ (M, R) is called the Pontryagin ring of E. The first important result obtained in the theory of characteristic classes of foliations is the following T HEOREM 13 (Bott vanishing theorem). Let F be a foliation of codimension n with normal bundle νF . Then Ponti (νF) = 0 for i > 2n. As a nontrivial example of characteristic class of a foliation F , we have the Godbillon– Vey invariant GV(F) (discovered by C. Godbillon and J. Vey [136]) which is, in general, nonzero as shown by R. Roussarie. An elementary construction of this invariant in the codimension one case is as follows. Let M be a compact manifold endowed with a codimension one foliation defined by a differential 1-form ω. Then the integrability condition implies the equality ω ∧ dω = 0, i.e. dω = α ∧ ω. It is easy to see that α ∧ dα is closed and that its cohomology class in H 3 (M, R), which is by definition GV(F), is independent of the choice of α. One of the most important results in the study of the Godbillon–Vey invariant for codimension one foliations was obtained by G. Duminy in [73]. Let us describe it briefly; a complete account is given in [119]. Let F be a codimension one foliation on a compact manifold M. A leaf L of F is called resilient if there exist a loop σ : [0, 1] −→ L and a transversal T to F passing through σ (0) such that the following conditions are satisfied: (i) there exists a point x ∈ L ∩ T in the domain of holonomy hσ of σ and different from σ (0); (ii) the sequence hnσ (x) converges to σ (0) as n → +∞. G. Duminy proved that if F has no resilient leaf then the Godbillon–Vey invariant of F is zero. Recently, A. Connes and H. Moscovici have discovered a universal Hopf algebra with cohomology from which one is able to recover the characteristic classes of a foliation without use of Chern–Weil homomorphism or connections (cf. [63]).

7. Basic global analysis Let M be a manifold endowed with a foliation F of codimension n. We suppose for simplicity that F is transversely orientable.

7.1. Foliated vector bundles and basic sections ι

Let P : G → P −→ M be a principal bundle with structural group G ⊂ GL(N, C). The group G acts on P on the right and on its Lie algebra G by the adjoint representation.

56

R. Barre and A. El Kacimi Alaoui

Denote by V the vector bundle whose fibre Vz at a point z ∈ P is the tangent space at z of the fibre of P. A connection on P is a subbundle H of T P such that: – for every z ∈ P, Tz P = Vz ⊕ Hz , – for every g ∈ G and every z ∈ P , Hzg = (Rg )∗ (Hz ) where Rg is the right action of g on P . As is well known the subbundle H is also the kernel of an invariant (under the action of G) 1-form ω on P (called the connection form) with values in G. It is easy to see that the restriction of ι∗ (the derivative of ι) to Hz is an isomorphism onto Tι(z) M. Let τ = ι−1 ∗ (T F). We say that P is foliated if τ is integrable. In this case, τ defines a foliation F˜ on P such that ˜ = dim(F), – dim(F) – F˜ is invariant under the action of G. We say that the connection H is basic, if the ω is basic (cf. Section 1.5). A foliated bundle E is said to be an F -bundle, if it admits a basic connection. Let E −→ M be a complex vector bundle defined by a cocycle {Ui , γij , G} where Ui is an open cover of M and γij : Ui ∩ Uj −→ G ⊂ GL(N, C) are the transition functions. We say that E is an F -bundle, if the associated principal bundle G −→ P −→ M is an F -bundle. Because E = P ×G CN , F˜ induces a foliation FE on E. An F -morphism ϕ : (E, ω) −→ (E , ω ) between two F -bundles is a morphism of vector bundles which sends leaves of FE into leaves of FE . (Notice that the collection of F -bundles and F -morphisms is a category. So we can define the group K(M, F) of foliated K-theory as in the classical case.) Let E −→ M be an F -bundle. Then the dual bundle E ∗ and all its exterior powers Λ∗ E ∗ are F -bundles; also H2 E = {Hermitian forms on E} is an F -bundle.

7.2. Transversely elliptic operators Let E −→ M be a F -foliated vector bundle. Denote by ∇ the covariant derivative ∇

χ(M) × C ∞ (E) −→ C ∞ (E) associated to the connection H. We say that a section α ∈ C ∞ (E) is basic, if it satisfies the condition ∇X α = 0 for every X ∈ χ(F). The space C ∞ (E/F) of basic sections of E is an Ab -module. Let E and E two F -bundles (with the same rank N for simplicity). A basic differential D

operator of order from E to E is a linear map C ∞ (E/F) −→ C ∞ (E /F) such that on local coordinates (x1 , . . . , xm , y1 , . . . , yn ) for which F is defined by the differential equations dy1 = · · · = dyn = 0, D has the expression: D=

|s|

as (y)

∂y1s1

∂ |s| , . . . ∂ynsn

where s = (s1 , . . . , sn ) ∈ Nn , |s| = s1 + · · · + sn and as are (N × N )-matrices whose coefficients are basic functions. The principal symbol of D at the point z = (x, y) and the basic

Foliations

57

covector ξ ∈ νz∗ F is the linear map σ (D)(z, ξ ) : Ez −→ Ez defined by σ (D)(z, ξ )(η) =

ξ1s1 · · · · · ξnsn as (y)(η).

|s|=

We say that D is transversely elliptic if σ (D)(z, ξ ) is an isomorphism for every z ∈ M and every basic covector ξ different from 0. If F is Riemannian, its conormal bundle ν ∗ F is an F -bundle and is equipped with a foliation F ∗ . Then the principal symbol σ (D)(z, ξ ) of a transversely elliptic operator D defines an element [D] in the group K(ν ∗ F, F ∗ ). A Hermitian metric on E is a positive definite section h of H2 E. If h is basic we say that E is a Hermitian F -bundle. Let E −→ M be a Hermitian F -foliated bundle with Hermitian metric h and let D be a basic differential operator of order = 2 on C ∞ (E/F). For every z ∈ M and every basic covector ξ ∈ νz∗ F we define a quadratic form A(D)(z, ξ ) : Ez −→ C by A(D)(z, ξ )(η) = (−1) σ (D)(z, ξ )(η), η . We say that D is strongly transversely elliptic, if A(D)(z, ξ ) is positive definite for every z ∈ M and every nonzero ξ . Obviously every strongly transversely elliptic operator is transversely elliptic. From now on we suppose that M is compact and connected. Assume that the foliation F is Riemannian transversely oriented. Let E # be the pullback of E to the principal bundle p SO(n) −→ M # −→ M of the orthonormal direct frames transverse to F (cf. Theorem 8). Then E # is a SO(n)-bundle and a Hermitian F # -bundle equipped with a Hermitian metric h# . Let W # be the basic manifold associated to the transversely parallelizable foliation F # on M # . The basic sections of E are canonically identified to basic sections of E # which are invariant under the action of SO(n). In particular, if f : M −→ C is a basic function, f ◦ p is a basic function on M # (with respect to F # ); moreover f ◦ p is invariant by the action of SO(n). Because f ◦ p is continuous, it is constant on the leaf closures of F # so it induces an SO(n)-invariant C ∞ function on the basic manifold W # . We can prove, by the converse process, that any SO(n)-invariant C ∞ function on the basic manifold W # defines a C ∞ basic function on M; in other words, the algebra Ab of basic functions on M is canonically isomorphic to the algebra ASO(n) (W # ) of functions on W # invariant by SO(n). The bundle like metric on M # induces a Riemannian metric on W # for which SO(n) acts by isometries. Let µ be the measure on W # associated to this metric. On C ∞ (E/F) we define an inner product as follows. Let α and β be two elements of ∞ C (E/F). The function Θ(α, β) : z ∈ M −→ hz (α(z), β(z)) ∈ C is basic; so it defines an SO(n)-invariant function Θ # (α, β) on W # . We set α, β =

Θ # (α, β)(w) dµ(w). W

For any basic differential operator D from a Hermitian F -bundle E to a Hermitian F -bundle E , denote by N (D) the kernel of D and R(D) its range.

58

R. Barre and A. El Kacimi Alaoui

T HEOREM 14. Let E and E be two Hermitian F -bundles on M and let D be a transversely elliptic operator from C ∞ (E/F) to C ∞ (E /F). Denote by D ∗ the formal adjoint of D which is also a basic transversely elliptic operator from C ∞ (E /F) to C ∞ (E/F). Then N(D) and N (D ∗ ) are finite dimensional and we have an orthogonal decomposition: C ∞ (E/F) = N (D) ⊕ R(D ∗ ). In particular, D has an index: ind(D/F) = dim N (D) − dim N (D ∗ ). All the details of the proof of this theorem can be found in [85].

7.3. Transversely elliptic complexes Let (E r , Dr )r=0,1,...,n be a family of Hermitian F -bundles and basic differential operators of order one Dr : C ∞ (E r /F) −→ C ∞ (E r+1 /F) (by convention Dn = 0) such that the sequence Dr+1 Dr−1 Dr · · · −→ C ∞ (E r /F) −→ C ∞ E r+1 /F −→ · · ·

(∗)

is a differential complex, that is, Dr+1 ◦ Dr = 0 for r = 0, 1, . . . , n − 1. Let z ∈ M and ξ ∈ νz∗ F ; denote by σ (Dr )(z, ξ ) the principal symbol of Dr at (z, ξ ) which is a linear map σ (Dr )(z, ξ ) : Ezr −→ Ezr+1 . Set σr = σ (Dr )(z, ξ ); we say that the complex (∗) is transversely elliptic if its symbol sequence σr−1

σr

σr+1

· · · −→ Ezr −→ Ezr+1 −→ · · ·

(∗ )

is exact for every z and every nonzero ξ . Let Dr∗ : C ∞ (E r+1 /F) −→ C ∞ (E r /F) be the formal adjoint of Dr (with respect to the inner product defined in Section 7.2). Then it is easy to see that the complex (∗) is transversely elliptic if and only if the basic operator of ∗ is strongly order 2: Lr : C ∞ (E r /F) −→ C ∞ (E r /F) defined by Lr = Dr∗ Dr + Dr−1 Dr−1 transversely elliptic. Let (E r , Dr ), r = 0, 1, . . . , n, be a transversely elliptic complex with cohomology r Hb (E ∗ ). Then applying Theorem 14, we have T HEOREM 15. (i) For each r = 0, 1, . . . , n, the kernel Hbr (E ∗ ) of Lr is equal to the space ∗ ). N (Dr ) ∩ N (Dr−1 r (ii) The space Hb (E ∗ ) is finite dimensional and we have an orthogonal decomposition C ∞ (E r /F) = Hbr (E ∗ ) ⊕ R(Dr−1 ) ⊕ R(Dr∗ ). (iii) The orthogonal projection C ∞ (E r /F) −→ Hbr (E ∗ ) induces an isomorphism from Hbr (E ∗ ) to Hbr (E ∗ ).

Foliations

59

We will give two concrete examples to illustrate this result: the basic de Rham complex and the basic Dolbeault complex. Let r ∈ {0, . . . , n} and denote by E r the vector bundle of exterior r-forms on the normal bundle νF . As it was pointed out, E r is a Hermitian F -bundle; its basic sections are exactly basic differential forms Ω r (M/F) of degree r on M. The de Rham exterior differential d restricted to Ω r (M/F) = C ∞ (E r /F) is a basic differential operator d : Ω r (M/F) −→ Ω r+1 (M/F). Thus we obtain a differential complex d

d

d

· · · −→ Ω r (M/F) −→ Ω r+1 (M/F) −→ · · ·

(∗∗)

called the basic de Rham complex of F ; its homology H r (M/F) is called the basic cohomology of F and depends only on the transverse structure of F . Let δb : Ω r+1 (M/F) −→ Ω r (M/F) be the formal adjoint of d; this operator can be described explicitly in terms of coefficients of the transverse metric on νF and the Hermitian metrics on the bundles E r (cf., for instance, [5,352,351,265,289,288,290,291,89, 195–197,85]). Let b = dδb + δb d; this is a basic differential operator of order 2 on Ω r (M/F) called the basic Laplacian. A basic form α ∈ Ω r (M/F) which satisfies the equation b α = 0, or equivalently dα = 0 and δb α = 0, is called a basic harmonic form; denote by Hr (M/F) the space of such forms. Applying Theorem 14 we obtain the following T HEOREM 16. (i) The space Hr (M/F) is finite dimensional and we have an orthogonal decomposition Ω r (M/F) = Hr (M/F) ⊕ R(d) ⊕ R(δb ). (ii) The orthogonal projection Ω r (M/F) −→ Hr (M/F) induces an isomorphism from H r (M/F) to Hr (M/F). (iii) Suppose that the vector space H n (M/F) is nonzero; then there exists a natural nondegenerate pairing Φ : ([α], [β]) ∈ H r (M/F) × H n−r (M/F) −→ Φ([α], [β]) ∈ C. So the basic cohomology satisfies Poincaré duality. During the last decades, many people contributed to the proof of this theorem. It was first proved by B.L. Reinhart in [290]. But Y. Carrière [51] discovered a mistake which makes assertion (iii) false: B.L. Reinhart does not suppose H n (M/F) different from {0} to obtain Poincaré duality; he was probably thinking that this hypothesis is automatically satisfied. Later on F.W. Kamber and P. Tondeur [196] have shown that the Reinhart’s proof is still valid if we suppose the leaves minimal (cf. Section 9.3). Finally the theorem was proved in full generality (without any assumption on the minimality of the leaves) in [89]. Now suppose that F is Hermitian. Let ν be the complexified normal bundle νF ⊗R C of νF . Let J be the automorphism of ν associated to the complex structure; J satisfies the relation J 2 = −id and then has two eigenvalues i and −i with associated eigensubbundles

60

R. Barre and A. El Kacimi Alaoui

respectively denoted ν 10 and ν 01 . We have a splitting ν = ν 10 ⊕ ν 01 which gives rise to a decomposition Λr ν ∗ =

Λp,q ,

p+q=r ∗

∗

where Λp,q = Λp ν 10 ⊗ Λq ν 01 . Basic sections of Λp,q are called basic forms of type (p, q). They form a vector space denoted Ω p,q (M/F). We have Ω r (M/F) =

Ω p,q (M/F).

p+q=r

The exterior differential decomposes into a sum of two operators ∂ : Ω p,q (M/F) −→ Ω p+1,q (M/F)

and

∂¯ : Ω p,q (M/F) −→ Ω p,q+1 (M/F) as in the classical case of a complex manifold. We have ∂¯ 2 = 0; so we obtain, for p fixed, a differential complex ∂¯

∂¯

∂¯

· · · −→ Ω p,q (M/F) −→ Ω p,q+1 (M/F) −→ · · ·

(∗∗∗)

called the basic Dolbeault complex of F ; its homology H p,q (M/F) is the basic Dolbeault cohomology of the foliation F : even though the leaf space is topologically bad, it can be considered as a “complex manifold” whose Dolbeault cohomology is H p,∗ (M/F)! ¯ this is an operator of type (0, −1). The operator Let δb denote the formal adjoint of ∂; ¯ is selfadjoint; a simple computation in local coordinates, like for the basic b = δb ∂¯ + ∂δ b Laplacian, shows that b is strongly transversely elliptic. Therefore the complex (∗∗∗) is transversely elliptic. Let ¯ = 0 and δb α = 0 . Hp,q (M/F) = Ker b = α ∈ Ω p,q (M/F): ∂α Applying Theorem 14, we obtain T HEOREM 17. (i) The space Hp,q (M/F) is finite dimensional and we have an orthogonal decomposition ¯ ⊕ R(δb ). Ω p,q (M/F) = Hp,q (M/F) ⊕ R(b ) = Hp,q (M/F) ⊕ R(∂) (ii) The orthogonal projection Ω p,q (M/F) −→ Hp,q (M/F) induces an isomorphism from H p,q (M/F) to Hp,q (M/F). (iii) Suppose that the vector space H n (M/F) is nonzero; then there exists a natural nondegenerate pairing Ψ : ([α], [β]) ∈ H p,q (M/F) × H n−p,n−q (M/F) −→ Ψ ([α], [β]) ∈ C. So the basic Dolbeault cohomology satisfies Serre duality.

Foliations

61

Suppose now that F is transversely Kählerian with Kähler form ω (it is a basic differential form of degree 2; it is closed and nondegenerate). In this case, we can prove that b = 2b . Because of the decomposition Ω r (M/F) =

Ω p,q (M/F),

p+q=r

every basic differential r-form can be uniquely written as a sum α = p+q=r αpq where αpq ∈ Ω p,q (M/F). Then we have the following assertions. (iv) α is b -harmonic if, and only if, each component αpq is b -harmonic. So we have a direct decomposition H r (M/F) =

H p,q (M/F).

p+q=r

(v) The complex conjugacy induces an isomorphism (of real vector spaces) H p,q (M/F) H q,p (M/F). (vi) For every odd r ∈ {0, . . . , 2n}, the dimension of the space H r (M/F) is even. In particular, if n = 1 we have b1 (M/F) = 2 dim H 01 (M/F). The integer dim H 01 (M/F) will be denoted g(F) and called the genus of the foliation F . It is similar to the genus of a compact Riemann surface; it counts the number of linearly independent basic holomorphic 1-forms. (vii) For every p ∈ {0, . . . , n} the differential form ωp = ω ∧ · · · ∧ ω (wedge product p times) is harmonic. So, the space H p,p (M/F) is nonzero. Notice that this theorem is also a particular case of Theorem 15. It can be used to establish more properties: basic Hodge structures for transversely Kählerian foliations, basic Calabi–Yau theorem [85] and deformation of transversely holomorphic foliations with a fixed differentiable type [90].

8. Deformation theory of foliations We will describe only the real case following Hamilton’s paper [159]. Deformation theory of holomorphic or generally transversely holomorphic foliations is more rich. The reader can find a good account of the subject in [132]. Let M be a manifold of dimension m + n. For each x ∈ M, let G(x, m) be the Grassmanian manifold of m-planes in Tx M. Then: G(m) =

G(x, m)

x∈M

can be given a structure of a differentiable manifold such that the canonical projection (x, τ ) ∈ G(m) −→ x ∈ M is a locally trivial fibration, the fibre being the Grassmanian

62

R. Barre and A. El Kacimi Alaoui

G(m) of m-planes in the space Rm+n . Then a subbundle of rank m of T M is just a section of the bundle G(m) −→ M. Denote by C ∞ (G(m)) the space of sections of this bundle. Let τ ∈ C ∞ (G(m)). By Frobenius theorem, τ is tangent to a foliation if, and only if, for any pair (U, V ) of (global) sections of τ , the Lie bracket [U, V ] is also a section of τ . Let (X1 , . . . , Xm ) be a local basis of τ . Then U=

m

a i Xi

and V =

m

b j Xj .

j =1

i=1

So the bracket [U, V ] can be expressed as [U, V ] =

m i j a b [Xi , Xj ] + a i Xi bj Xj − bj Xj a i Xi . i,j =1

Therefore the value of [U, V ] in ντ = T M/τ at a point x ∈ M depends only on the value of U and V at x. Hence Qτ (U, V ) = π([U, V ]) is a skew-symmetric bilinear map Qτ : τ × τ −→ ντ where π : T M −→ ντ is the canonical projection. In other words, Qτ is a global section of the vector bundle Λ2 (τ, ντ ) of skew-symmetric bilinear forms on the bundle τ . The integrability condition of τ is equivalent to Qτ identically equal to 0. So we get a map Q : C ∞ (G(m)) −→ Σ where Σ is a fibre bundle over G(m) whose fibre over a point σ ∈ G(m) is the infinite-dimensional space Ω 2 (σ, νσ ) of global sections of the bundle Λ2 (τ, ντ ). The space Fol(M, m) of dimension m foliations on M is exactly the set {Q = 0}. It will be equipped with the C ∞ -topology induced by the topology of the Fréchet manifold C ∞ (G(m)) (cf. [160]). Let D be the diffeomorphism group of M; then D acts on C ∞ (G(m)) and the action preserves Fol(M, m). Two foliations F, F ∈ Fol(M, m) are conjugated, if they are in the same orbit of the action of D, that is, there exists ϕ ∈ D such that F = ϕ ∗ (F). Now fix τ in C ∞ (G(m)) and suppose that it is tangent to a foliation F . Then the map Pτ : ϕ ∈ D −→ ϕ ∗ (F) ∈ C ∞ (G(m)) takes its values in Fol(M, m). So we get a sequence of Fréchet manifolds and differentiable maps Q Pτ D −→ C ∞ G(m) −→ Σ. Following R. Hamilton, this sequence is called the nonlinear deformation complex of the foliation F [159]. D EFINITION 5. We say that F is C ∞ -stable if there exist an open neighborhood O of the identity in D and an open neighborhood U of F in Fol(M, m) such that the sequence Pτ

Q

O −→ U −→ Σ is exact, that is, every dimension m foliation F on M, close enough to F in the C ∞ -topology, is conjugated to F by an element of O. An important tool to prove the C ∞ -stability of a foliation is Hamilton’s criterion (cf. [159, p. 47]) that we shall describe. This criterion is based on the implicit function theorem of Nash–Moser which is nicely explained in Hamilton’s paper [160].

Foliations

63

Given a foliation F of dimension m on a compact manifold M, let AkF denote the space of differentiable sections of Λk T F ∗ ⊗ νF . Since νF is a foliated bundle there is a well defined “exterior derivative along the leaves” dF : AkF −→ Ak+1 F given by: dF η(X1 , . . . , Xk+1 ) = (−1)i Xi η(X1 , . . . , Xˆ i , . . . , Xk+1 ) i

+

(−1)i+j η [Xi , Xj ], X1 , . . . , Xˆ i , . . . , Xˆ j , . . . , Xk+1 .

i<j 2 = 0 and thus we obtain a differential complex An easy computation shows that dF dF

dF

dF

dF

0 −→ A0F −→ A1F −→ A2F −→ · · · −→ Am F −→ 0 which is only elliptic along the leaves. Let 0 1 · · · s · · · be an increasing collection of norms (of Sobolev or Hölder type) on the Fréchet space A∗F = AkF . k0

With this notation one has T HEOREM 18 (Hamilton). Assume that there exist continuous linear operators H : A1F −→ A0F and K : A2F −→ A1F fulfilling the following conditions: (i) dF ◦ H + K ◦ dF = id, (ii) there is a fixed number r ∈ N for which we have tame estimates for all s, H (β) Cs βs+r and K(γ ) Cs γ s+r , s s where Cs are positive constants depending only on s. Then the foliation F is C ∞ -stable. Unfortunately Hamilton’s paper is still unpublished. In [93], the authors constructed a class of foliations and, using Hamilton’s criterion, they proved that these foliations are C ∞ -stable. Example 1.3(iv), with some assumptions on the matrix A, is in this class. 9. Some other themes As we have pointed out in the foreword, Foliation Theory is a wide field in Mathematics and so huge to discuss completely here. For this reason we have chosen only some of the themes related to Differential Geometry which is the main topic to which this book is devoted. The nonwarned reader may be inclined to believe that the theory is reduced to this part. Fortunately this is not the case. We devote this section to other themes which are no less important than the above ones.

64

R. Barre and A. El Kacimi Alaoui

9.1. Compact leaves Let M be a connected compact orientable manifold of dimension m + n and F a codimension n foliation on M. A compact leaf of F is a leaf L which is compact as a subset of M. If m = 1 such leaf is a periodic orbit and it describes a stationary state of the dynamical system defined by F . The problem of the existence of compact leaves is highly nontrivial. It was first introduced by H. Poincaré in his studies on limit cycles for ordinary differential equations. One of the famous problems was the Seifert conjecture: Every continuous vector field on the 3-dimensional sphere S3 has a periodic orbit. In 1974, using Denjoy’s example of a vector field with exceptional minimal set on the 2-torus, P. Schweitzer [324] constructed a counterexample in class C 1 . In 1988, J. Harrison [162] gave a C 2 counterexample. Finally in 1993, K. Kuperberg [200] solved completely the problem by constructing in any compact 3-manifold a real analytic vector field without periodic orbit. However, M. Brunella [31] proved that the conjecture is true if the flow is transversely holomorphic; in fact, he established a complete classification of these flows on compact 3-manifolds. The most important result concerning the problem of existence of compact leaves was Novikov’s theorem stated above (Theorem 12). Nothing is known in higher dimensions and the following question is still open: is it true that every codimension one foliation on the odd sphere S2p+1 (where p 2) admits a compact leaf? We say that F is a compact foliation if all leaves are compact. For example, every foliation defined by a locally free action of a connected compact Lie group is a compact foliation. Compact foliations was a theme which interested many people (R. Edwards, K. Millet, D. Sullivan, D. Epstein, E. Vogt, H. Rummler, etc.).

9.2. When is a manifold a leaf ? Let L be a noncompact connected manifold. Does there exist a compact manifold M endowed with a foliation F with a leaf diffeomorphic to L? This question was asked by J. Sondow in [335] where he gave some sufficient conditions on L to be a leaf. J. Cantwell and L. Conlon proved in [46] that every surface is a leaf. Along the same lines, G. Hector and W. Bouma proved in [170] that every noncompact surface can be a leaf of a simple foliation of R3 , i.e. a foliation defined by a submersion R3 −→ R. In [114] E. Ghys observed that the topology of a leaf of a foliation on a compact manifold has to be, in some sense, “recurrent”; then he constructed, for any positive integer d, a noncompact manifold L of dimension d which can not be homeomorphic to any leaf of any foliation on a compact manifold. In [125] he also studies the topology of the generic leaves of a lamination by surfaces on a compact metric space and proved that there exist only six noncompact surfaces which can be realized as leaves: (a) the plane R2 , (b) the cylinder S1 × R, (c) the “Loch-Ness monster”, i.e. the plane with infinitely many handles attached, (d) the “Jacob ladder”,

Foliations

65

(e) the “Cantor tree”, i.e. the sphere S2 with a Cantor set removed, (f) the “flowered Cantor tree”, i.e. the Cantor tree with infinitely many handles attached in all directions.

9.3. Minimal leaves Let M be a Riemannian manifold. Denote by ∇ the covariant derivative associated to the Levi-Civita connection. Let L be a submanifold of M (not necessarily properly embedded). Let x ∈ L and ν a vector field defined on a neighborhood of x and orthogonal to L. For X ∈ Tx L, we set: Wxν (X) = −px (∇X ν), where px : Tx M −→ Tx L is the orthogonal projection. Then Wxν is an endomorphism of the vector space Tx L, symmetric with respect to the induced metric on Tx L; it is called the Weingarten map associated to ν. The trace of Wxν describes the variation at x of the volume element when L moves in the direction of ν. We say that L is minimal, if the trace of Wxν is zero for all vector fields ν orthogonal to L. A foliation F on M is said to be with minimal leaves, if all leaves of F are minimal submanifolds. Given an m-dimensional foliation on a compact manifold M, does there exist a Riemannian metric on M for which the leaves are minimal? This question was discussed by H. Rummler [309] and D. Sullivan [339]. They proved the following criterion: such a metric exists if, and only if, there exists an m-form χ positive on the leaves and relatively closed, namely dχ(X1 , . . . , Xm , Y ) = 0 whenever the vector fields X1 , . . . , Xm are tangent to F . In [149] A. Haefliger proved that the property for F to be with minimal leaves depends only on the transverse structure. He also gave a criterion in terms of transverse invariant currents and used it to give many examples of minimal foliations and nonminimal ones. Suppose now that F is a Riemannian codimension n foliation and denote by v the volume basic form associated to the metric. If F is with minimal leaves then v defines a nonzero class in the basic cohomology H n (M/F). Indeed, let χ be the m-form given by the Rummler–Sullivan criterion. Suppose that v = dβ where β ∈ Ω n−1 (M/F). Then: χ ∧ v = χ ∧ dβ = (−1)m d(χ ∧ β) − dχ ∧ β . But dχ ∧ β = 0 because χ is relatively closed. So χ ∧ v is an exact form. But this is impossible because it is a volume form on the compact orientable manifold M. The converse of this assertion was conjectured by Y. Carrière [51] and proved by X. Masa in [227]. In [112] E. Ghys proved that any Riemannian foliation on a simply connected compact manifold admits a bundle-like metric for which the leaves are minimal. Now let F be a foliation on a compact Riemannian manifold M. We say that F is totally geodesic if every geodesic tangent to a leaf L at a point is tangent to L everywhere. This

66

R. Barre and A. El Kacimi Alaoui

is a special class of foliations with minimal leaves which was studied, for instance, by Y. Carrière, G. Cairns, E. Ghys (see [35,36,33,37,54]). In particular, E. Ghys, in [111], has completely classified all the totally geodesic foliations of codimension one on compact manifolds. References [1] C. Albert and P. Dazord, Théorie des groupoï des symplectiques, II, Pub. Dep. Math. Nouv. Sér. Univ. Claude Bernard, Lyon (1990), 27–99. [2] F. Alcade Cuesta and G. Hector, Feuilletages en surfaces, cycles évaouissants et variétés de Poisson, Monatsh. Math. 124 (3) (1997), 193–213. [3] J. Álvarez-López, Duality in the spectral sequence of a Riemannian foliation, Amer. J. Math. 111 (6) (1989), 905–926. [4] J. Álvarez-López, A finiteness theorem for the spectral sequence of a Riemannian foliation, Illinois J. Math. 33 (1) (1989), 79–92. [5] J. Álvarez-López, The basic component of the mean curvature of Riemannian foliations, Ann. Global Anal. Geom. 10 (1992), 179–194. [6] J. Álvarez-López, Morse inequalities for pseudogroups of local isometries, J. Differential Geom. 37 (3) (1993), 603–638. [7] J. Álvarez-López and P. Tondeur, Hodge decomposition along the leaves of a Riemannian foliation, J. Funct. Anal. 99 (2) (1991), 433–458. [8] C. Anantharman-Delaroche, Purely infinite C ∗ -algebras arising from dynamical systems, Bull. Soc. Math. France 125 (1997), 199–226. [9] J.L. Arraut, A 2-dimensional foliation on R7 , Topology 12 (1973), 243–245. [10] J.L. Arraut and M. Craizer, Foliations of M 3 defined by R2 -actions, Ann. Inst. Fourier 45 (4) (1995), 1018–1118. [11] J.L. Arraut and N.M. Dos Santos, The characteristic mapping of a foliated S1 -bundle, Topology 31 (3) (1992), 545–555. [12] M. Artin, Algebraic spaces, Yale Monograph Study 3 (1971). [13] R. Barre, De quelques aspects de la théorie des Q-variétés différentielles et analytiques, Ann. Inst. Fourier 23 (3) (1973), 227–312. [14] V. Belfi, E. Park and K. Richardson, A Hopf index theorem for foliations, Differential Geom. Appl. 18 (2003), 319–341. [15] M. Belliart and O. Birembaux, Actions localement libres de groupes de Lie résolubles, Ann. Inst. Fourier 44 (5) (1994), 1519–1537. [16] A. Benoist and F. Labourie, Sur les difféomorphismes d’Anosov affines à feuilletages stable et instable, Invent. Math. 111 (2) (1993), 285–308. [17] H. Blas, M. Llabres and A. Reventos, Transverse structures of Lie foliations, J. Math. Soc. Japan 48 (4) (1996), 769–795. [18] R.A. Blumenthal, Transversely homogeneous foliations, Ann. Inst. Fourier 29 (4) (1979), 143–158. [19] C. Bonatti and S. Firmo, Feuilles compactes d’un feuilletage générique en codimension un, Ann. Sci. École Norm. Sup. 27 (4) (1994), 407–461. [20] C. Bonatti and A. Haefliger, Déformations de feuilletages, Topology 29 (2) (1990), 205–229. [21] A. Borel and F. Hirzebruch, Characteristic classes and homogeneous spaces, II, Amer. J. Math. 81 (1959), 315–382. [22] R. Bott, On the topological obstruction to integrability, Sympos. Pure Math., vol. 16, Amer. Math. Soc., Providence, RI (1970), 127–131. [23] R. Bott, Lectures on characteristic classes and foliations, Lecture Notes in Math. 279 (1972). [24] R. Bott, On some formulas for the characteristic classes of group actions, Lecture Notes in Math. 652 (1978), 25–62. [25] R. Bott and A. Haefliger, On characteristic classes of Γ -foliations, Bull. Amer. Math. Soc. 78 (1972), 1039–1044.

Foliations

67

[26] R. Bott and J. Heitsch, A remark on the integral cohomology of BΓq , Topology 11 (1972), 141–146. [27] E. Bouacida, O. Echi and E. Salhi, Feuilletages et topologie spectrale, J. Math. Soc. Japan 522 (2000), 447–464. [28] L.G. Bouma and W.T. Van Est, Manifolds schemes and foliations on the 2-torus and the Klein bottle, I, II, III, Indag. Math. 40 (1978), 313–347. [29] F.G.B. Brito, R. Langevin and H. Rosenberg, Intégrales de courbure sur les variétés feuilletées, J. Differential Geom. 16 (1981), 19–50. [30] M. Brunella, Remarks on structurally stable proper foliations, Math. Proc. Cambridge Philos. Soc. 115 (1994), 111–120. [31] M. Brunella, On transversely holomorphic flows, I, Invent. Math. 126 (2) (1996), 265–279. [32] M. Brunella and E. Ghys, Umbilical foliations and transversely holomorphic flows, J. Differential Geom. 41 (1) (1995), 1–19. [33] G. Cairns, Aspects cohomologiques des feuilletages totalement géodésiques, C. R. Acad. Sci. Paris Sér. I 299 (20) (1984), 1017–1019. [34] G. Cairns, A general description of a totally geodesic foliation, Tohoku Math. J. (2) 38 (1986), 37–55. [35] G. Cairns, Some properties of a cohomology group associated to a totally geodesic foliation, Math. Z. 192 (1986), 391–403. [36] G. Cairns, Feuilletages totalement géodésiques sur les variétés simplement connexes, Sém. Sud Rhodanien Géom. VII Trav. Cours 26 (1988), 1–14. [37] G. Cairns and E. Ghys, Totally geodesic foliations on 4-manifolds, J. Differential Geom. 23 (1986), 241– 254. [38] C. Camacho, Structural stability theorems for integrable differential forms on 3-manifolds, Topology 17 (1978), 143–155. [39] C. Camacho and A. Lins Neto, Geometric Theory of Foliations, Birkhäuser (1985). [40] J. Cantwell and L. Conlon, Growth of leaves, Comment. Math. Helv. 53 (1978), 93–111. [41] J. Cantwell and L. Conlon, Reeb stability for noncompact leaves in foliated 3-manifolds, Proc. Amer. Math. Soc. 83 (1981), 408–410. [42] J. Cantwell and L. Conlon, Poincaré–Bendixon theory for leaves of codimension one, Trans. Amer. Math. Soc. 265 (1981), 181–209. [43] J. Cantwell and L. Conlon, Tischler fibrations of open foliated sets, Ann. Inst. Fourier 31 (2) (1981), 113–135. [44] J. Cantwell and L. Conlon, The endsets of leaves, Topology 21 (1982), 333–352. [45] J. Cantwell and L. Conlon, Analytic foliations and the theory of levels, Math. Ann. 265 (1983), 253–261. [46] J. Cantwell and L. Conlon, Every surface is a leaf, Topology 26 (1987), 265–285. [47] J. Cantwell and L. Conlon, The theory of levels, Contemp. Math. 70 (1988), 1–10. [48] J. Cantwell and L. Conlon, Leafwise hyperbolicity of proper foliations, Comment. Math. Helv. 64 (1989), 329–337. [49] J. Cantwell and L. Conlon, Generic leaves, Comment. Math. Helv. 73 (1998), 306–336. [50] P. Caron and Y. Carrière, Flots transversalement de Lie Rn , flots minimaux, C. R. Acad. Sci. Paris 280 (1980), 477–478. [51] Y. Carrière, Flots riemanniens, Structures transverses des feuilletages, Toulouse 1982, Astérisque 116 (1984), 31–52. [52] Y. Carrière, Feuilletages riemanniens à croissance polynomiale, Comment. Math. Helv. 63 (1984), 1–20. [53] Y. Carrière, Les propriétés topologiques des flots riemanniens retrouvées à l’aide du théorème des variétés presque plates, Math. Z. 186 (1984), 393–400. [54] Y. Carrière and E. Ghys, Feuilletages totalement géodésiques, An. Acad. Brasil. Ciênc. 53 (1981), 427– 432. [55] J. Cerf, 1-formes fermées non singulières sur les variétés compactes de dimension 3, Sém. Bourbaki, Exposé 574, Lecture Notes in Math. 901 (1981), 205–219. [56] D. Cerveau, Feuilletages holomorphes de codimension un. Réduction des singularités en petites dimensions et applications, IRM de Rennes, Prépublication 98-01 (1998). [57] G. Chatelet, Sur les feuilletages induits par l’action de groupes de Lie nilpotents, Ann. Inst. Fourier 27 (2) (1977), 161–189.

68

R. Barre and A. El Kacimi Alaoui

[58] G. Chatelet and H. Rosenberg, Un théorème de conjugaison des feuilletages, Ann. Inst. Fourier 21 (3) (1971), 95–106. [59] J. Chevalley, Theory of Lie Groups, Princeton Mathematical Series, vol. 8 (1946). [60] L. Conlon, Transversally parallelizable foliations of codimension two, Trans. Amer. Math. Soc. 194 (1975), 79–102. [61] A. Connes, Sur la théorie non commutative de l’intégration, Lecture Notes in Math. 725 (1979). [62] A. Connes, Noncommutative Geometry, Academic Press (1994). [63] A. Connes and H. Moscovici, Hopf algebras, cyclic cohomology and the transverse index theorem, Commun. Math. Phys. 198 (1998), 199–246. [64] A. Connes and H. Moscovici, Differentiable cyclic cohomology and Hopf algebraic structures in transverse geometry, Essays on Geometry and Related Topics, vols. I, II, Monogr. Enseign. Math., vol. 38, Geneva (2001), 217–255. [65] M. Craioveanu and M. Puta, Cohomology on a Riemannian foliated manifold with coefficients in the sheaf of germs of foliated currents, Math. Nachr. 99 (1980), 43–53. [66] C. Cumenge, Cohomologie de l’espace des feuilles d’un feuilletage, Thèse 3ème cycle, Toulouse (1981). [67] C. Curras-Bosch and P. Molino, Voisinage d’une feuille compacte dans un voisinage lagrangien, Hokkaido Math. J. 23 (2) (1994), 355–360. [68] A. Dadarlat, C ∗ -Algebras and applications, Contemp. Math. 145 (1993), 393–421. [69] C. Danthony, Feuilletages orientés de surfaces; le problème de la section globale, Ann. Inst. Fourier 38 (2) (1988), 201–227. [70] A. Denjoy, Sur les courbes définies par les équations différentielles à la surface du tore, J. Math. Pures Appl. 11 (1932), 333–375. [71] P.R. Dipolito, Codimension one foliations of closed manifolds, Comment. Math. Helv. 107 (1978), 403– 453. [72] T. Duchamp and M. Kalka, Deformation theory for holomorphic foliations, J. Differential Geom. 14 (1979), 317–337. [73] G. Duminy, L’invariant de Godbillon–Vey d’un feuilletage se localise sur les feuilles ressort, Manuscrit Université de Lille I (1982). [74] G. Duminy, Sur les cycles feuilletées de codimension un, Manuscrit Université de Lille I (1982). [75] G. Duminy and V. Sergiescu, Sur la nullité de l’invariant de Godbillon–Vey, C. R. Acad. Sci. Paris 292 (1981), 821–824. [76] J.L. Dupont, Simplicial de Rham cohomology and characteristic classes of flat bundles, Topology 15 (1976), 233–245. [77] A.H. Durfee, Foliations of odd-dimensional spheres, Ann. of Math. 96 (2) (1972), 407–411. [78] A.H. Durfee and H.B. Lawson, Fibered knots and foliations of highly connected manifolds, Invent. Math. 17 (1972), 203–215. [79] R. Edwards, K. Millet and D. Sullivan, Foliations with all leaves compact, Topology 16 (1977), 13–32. [80] C. Ehresmann, Les connexions infinitésimales dans un espace fibré différentiable, Colloque de Topologie, Bruxelles (1950), 29–53. [81] C. Ehresmann, Sur la théorie des variétés feuilletées, Rend. Mat. Appl. (5) 10 (1951), 64–82. [82] C. Ehresmann, Structures feuilletées, Proc. 5th Canadian Congress, Montréal (1961), 109–172. [83] C. Ehresmann and G. Reeb, Sur les champs d’éléments de contact de dimension p complètement intégrables sur une variété continûment différentiable, C. R. Acad. Sci. Paris 218 (1944), 955–957. [84] C. Ehresmann and S. Weishu, Sur les espaces feuilletées et théorème de stabilité, C. R. Acad. Sci. Paris 243 (1956), 344–346. [85] A. El Kacimi Alaoui, Opérateurs transversalement elliptiques sur les feuilletages riemanniens et applications, Compositio Math. 73 (1990), 57–106. [86] A. El Kacimi Alaoui and E. Gallego, Applications harmoniques feuilletées, Illinois J. Math. 40 (1) (1996), 115–122. [87] A. El Kacimi Alaoui and B. Gmira, Stabilité du caractère kählérien transverse, Israel J. Math. 101 (1997), 323–347. [88] A. El Kacimi Alaoui, G. Guasp and M. Nicolau, On deformations of transversely homogeneous foliations, Topology 40 (2001), 1363–1393.

Foliations

69

[89] A. El Kacimi Alaoui and G. Hector, Décomposition de Hodge basique pour un feuilletage riemannien, Ann. Inst. Fourier 36 (3) (1986), 207–227. [90] A. El Kacimi Alaoui and M. Nicolau, Déformations des feuilletages transversalement holomorphes à type différentiable fixé, Publ. Mat. Barcelona 33 (1989), 485–500. [91] A. El Kacimi Alaoui and M. Nicolau, Structures géométriques invariantes et feuilletages de Lie, Indag. Math. (N.S.) 1 (3) (1990), 323–334. [92] A. El Kacimi Alaoui and M. Nicolau, On the topological invariance of the basic cohomology, Math. Ann. 295 (1993), 627–634. [93] A. El Kacimi Alaoui and M. Nicolau, A class of C ∞ -stable foliations, Ergodic Theory Dynam. Systems 13 (1993), 697–704. [94] A. El Kacimi Alaoui, V. Sergiescu and G. Hector, La cohomologie basique d’un feuilletage riemannien est de dimension finie, Math. Z. 188 (1985), 593–599. [95] D.B.A. Epstein, Periodic flows on 3-manifolds, Ann. of Math. 95 (1972), 68–82. [96] D.B.A. Epstein, Foliations with all leaves compact, Ann. Inst. Fourier 26 (1) (1976), 265–282. [97] D.B.A. Epstein, K. Millet and D. Tischler, Leaves without holonomy, J. London Math. Soc. (2) 16 (1977), 548–552. [98] D.B.A. Epstein and E. Vogt, A counterexemple to the periodic orbit conjecture in codimension 3, Ann. of Math. (2) 108 (1978), 539–552. [99] T. Fack and G. Skandalis, Sur les représentations et idéaux de la C ∗ -algèbre d’un feuilletage, J. Operator Theory 8 (1982), 95–129. [100] F.T. Farrell and L.E. Jones, Foliated control theory, I, II, K-Theory 2 (1988), 357–399 and 401–430. [101] E. Fédida, Sur l’existence des feuilletages de Lie, C. R. Acad. Sci. Paris 278 (1974), 835–837. [102] D. Ferus, Totally geodesic foliations, Math. Ann. 188 (1970), 313–316. [103] H.R. Fisher and F.L. Williams, Complex foliated structures. I. Cohomology of the Dolbeault–Kostant complexes, Trans. Amer. Math. Soc. 252 (1979), 163–195. [104] D.B. Fuks, Characteristic classes of foliations, Russian Math. Surveys 28 (2) (1973), 1–16. [105] E. Gallego and A. Reventos, Lie flows of codimension 3, Trans. Amer. Math. Soc. 326 (2) (1991), 529–541. [106] R.B. Gardner, The Godbillon–Vey invariant and the Bott–Pasternak vanishing theorem, Lecture Notes in Math. 652 (1978), 75–94. [107] L. Garnett, Foliations, the ergodic theorem and Brownian motion, J. Funct. Anal. 51 (1983), 285–311. [108] I.M. Gelfand, B.L. Feigin and D.B. Fuks, Cohomology of the Lie algebra of formal vector fields with coefficients in its adjoint space and variations of characteristic classes of foliations, Funct. Anal. Appl. 8 (1974), 99–112. [109] I.M. Gelfand and D.B. Fuks, PL-foliations, Funct. Anal. Appl. 7 (1973), 278–284 and 8 (1974), 197–200. [110] R. Gerard and A. Sec, Feuilletages de Painlevé, Bull. Soc. Math. France 100 (1972), 47–72. [111] E. Ghys, Classification des feuilletages totalement géodésiques de codimension un, Comment. Math. Helv. 58 (1983), 543–572. [112] E. Ghys, Feuilletages riemanniens sur les variétés simplement connexes, Ann. Inst. Fourier 34 (4) (1984), 203–223. [113] E. Ghys, Actions localement libres du groupe affine, Invent. Math. 82 (1985), 479–526. [114] E. Ghys, Une variété qui n’est pas une feuille, Topology 24 (1985), 67–73. [115] E. Ghys, Classe d’Euler et minimal exceptionnel, Topology 24 (1985), 67–73. [116] E. Ghys, Flots d’Anosov dont les feuilletages stables sont différentiables, Ann. Sci. École Norm. Sup. 20 (1987), 251–270. [117] E. Ghys, Sur l’invariance topologique de la classe de Godbillon–Vey, Ann. Inst. Fourier 37 (4) (1987), 59–76. [118] E. Ghys, Gauss–Bonnet theorem for 2-dimensional foliations, J. Funct. Anal. 77 (1) (1988), 51–59. [119] E. Ghys, L’invariant de Godbillon–Vey, Sém. Bourbaki, Exposé 706, Astérisque 177–178 (1989), 155– 181. [120] E. Ghys, Déformations des flots d’Anosov et groupes fuchsiens, Ann. Inst. Fourier 42 (1–2) (1992), 209– 247. [121] E. Ghys, Rigidité différentiable des groupes fuchsiens, Publ. IHES 78 (1993), 163–185. [122] E. Ghys, Construction de champs de vecteurs sans orbites périodiques, Sém. Bourbaki, Exposé 785, Astérisque 227 (1993–1994), 283–307.

70 [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157]

R. Barre and A. El Kacimi Alaoui E. Ghys, Holomorphic Anosov systems, Invent. Math. 199 (3) (1995), 585–614. E. Ghys, Groupes d’holonomie des feuilletages de Lie, Indag. Math. 47 (1995), 173–182. E. Ghys, Topologie des feuilles génériques, Ann. of Math. (2) 141 (2) (1995), 387–422. E. Ghys, On transversely holomorphic flows, Invent. Math. 126 (2) (1996), 281–286. E. Ghys, R. Langevin and P. Walczak, Entropie géométrique des feuilletages, Acta Math. 16 (1–2) (1988), 105–142. E. Ghys and V. Sergiescu, Stabilité et conjugaison différentiable pour certains feuilletages, Topology 26 (1987), 93–105. E. Ghys and E. Sergiescu, Sur un groupe remarquable de difféomorphismes du cercle, Comment. Math. Helv. 62 (1987), 185–239. E. Ghys and T. Tsuboi, Différentiabilité des conjugaisons entre systèmes dynamiques de dimension un, Ann. Inst. Fourier 38 (1) (1988), 215–244. E. Ghys and A. Verjovsky, Holomorphic actions of the affine group, Proceedings of Geometric Study of Foliations, Tokyo, World Scientific (1994), 201–217. J. Girbau, A. Haefliger and D. Sundararaman, On deformations of transversely holomorphic foliations, J. Reine Angew. Math. 345 (1983), 122–147. J. Girbau and M. Nicolau, On deformations of holomorphic foliations, Ann. Inst. Fourier 39 (2) (1989), 417–449. C. Godbillon, Feuilletages ayant la propriété du prolongement des homotopies, Ann. Inst. Fourier 17 (2) (1967), 219–260. C. Godbillon, Feuilletages. Etudes géométriques, Progress in Mathematics, vol. 98, Birkhäuser, Basel (1991). C. Godbillon and J. Vey, Un invariant des feuilletages de codimension un, C. R. Acad. Sci. Paris 273 (1971), 92–95. X. Gomez Mont, Transversal holomorphic structures, J. Differential Geom. 15 (1980), 161–185. S. Goodman, On the structure of foliated 3-manifolds separated by a compact leaf, Invent. Math. 39 (1977), 213–221. S. Goodman, Vector fields with transverse foliations, Topology 24 (1985), 333–340. S. Goodman and J.F. Plante, Holonomy and averaging in foliated sets, J. Differential Geom. 14 (1979), 401–407. M. Gromov, Stable mappings of foliations into manifolds, Math. USSR Izv. 3 (1969), 671–694. M. Gromov, Transversal mappings of foliations, Soviet Math. Dokl. 9 (1969), 1126–1129. A. Haefliger, Structures feuilletées et cohomologie à valeurs dans un faisceau de groupoï des, Comment. Math. Helv. 32 (1958), 248–329. A. Haefliger, Variétés feuilletées, Ann. Scuola Nom. Sup. Pisa (3) 16 (1962), 367–397. A. Haefliger, Feuilletages sur les variétés ouvertes, Topology 9 (1970), 183–194. A. Haefliger, Homotopy, integrability, Lecture Notes in Math. 197 (1971), 133–163. A. Haefliger, Sur les classes caracteristiques des feuilletages, Sém. Bourbaki 1971/1972, Exposé 412, Lecture Notes in Math. 317 (1973), 239–260. A. Haefliger, Cohomology of Lie algebras and foliations, Lecture Notes in Math. 652 (1978), 1–12. A. Haefliger, Some remarks on foliations with minimal leaves, J. Differential Geom. 15 (1980), 269–284. A. Haefliger, Groupoï des d’holonomie et classifiants, Astérisque 116 (1984), 98–107. A. Haefliger, Pseudo-groups of local isometries, Differential Geometry, Santiago de Compostela, 1984, L. Cordero, ed., Research Notes, vol. 131, Pitman (1985), 174–197. A. Haefliger, Deformations of transversely holomorphic flows on spheres and deformations of Hopf fibrations, Compositio Math. 55 (1985), 241–251. A. Haefliger, Complexification of transversely holomorphic foliations, Math. Ann. 272 (1985), 23–27. A. Haefliger, Feuilletages riemanniens, Sém. Bourbaki 1998/1999, Exposé 707, Astérisque 177–178 (1989), 183–197. A. Haefliger, Travaux de Novikov sur les feuilletages, Sém. Bourbaki 1967/1968, Exposé 339. A. Haefliger and Quach Ngoc Du, Une présentation du groupe fondamental d’une orbifold, Astérisque 116 (1984), 98–107. A. Haefliger and E. Salem, Pseudogroupes d’holonomie des feuilletages riemanniens sur des variétés compactes 1-connexes, Colloq. Geom. Phys. Paris/F (1986), Travaux en cours 33 (1988), 141–160.

Foliations

71

[158] A. Haefliger and E. Salem, Riemannian foliations on simply connected manifolds and actions of tori on orbifolds, Illinois J. Math. 34 (4) (1990), 706–730. [159] R.S. Hamilton, Deformation theory of foliations, Preprint Cornell University, New York (1978). [160] R.S. Hamilton, The inverse function theorem of Nash and Moser, Bull. Amer. Math. Soc. 7 (1982), 65–222. [161] Y. Hantout, Classes caractéristiques de (G, H )-structures et finitude de leur évaluation, Manuscripta Math. 62 (1988), 383–399. [162] J. Harrison, A C 2 counterexample to the Seifert conjecture, Topology 27 (1988), 249–278. [163] G. Hector, Sur les feuilletages presque sans holonomie, C. R. Acad. Sci. Paris Sér. A 274 (1972), 1703– 1706. [164] G. Hector, Ouverts incompressibles et structures des feuilletages de codimension un, C. R. Acad. Sci. Paris Sér. A 274 (1972), 741–744. [165] G. Hector, Quelques exemples de feuilletages. Espèces rares, Ann. Inst. Fourier 26 (1) (1976), 239–264. [166] G. Hector, Feuilletages en cylindres, Lecture Notes in Math. 597 (1976), 252–270. [167] G. Hector, Leaves whose growth is neither exponential nor polynomial, Topology 16 (1977), 449–451. [168] G. Hector, Croissance des feuilletages presque sans holonomie, Lecture Notes in Math. 652 (1978), 141– 182. [169] G. Hector, Architecture des feuilletages de classe C 2 , Astérisque 107–108 (1983), 243–258. [170] G. Hector and W. Bouma, All open surfaces are leaves of simple foliations of R3 , Indag. Math. 45 (1983), 443–452. [171] G. Hector, E. Ghys and Y. Moriyama, On codimension one foliations and a theorem of Malcev, Topology 28 (2) (1989), 197–210. [172] G. Hector and U. Hirsch, Introduction to the Geometry of Foliations, Parts A and B, Vieweg & Sohn (1981). [173] G. Hector and E. Macias, Sur le théorème de de Rham pour les feuilletages de Lie, C. R. Acad. Sci. Paris Sér. I 311 (10) (1990), 633–636. [174] J.L. Heitsch, A cohomology of foliated manifolds, Comment. Math. Helv. 50 (1975), 197–218. [175] J.L. Heitsch, Residues and characteristic classes of foliations, Bull. Amer. Math. Soc. 83 (1977), 397–399. [176] J.L. Heitsch, Independent variation of secondary classes, Ann. of Math. (2) 108 (3) (1978), 421–460. [177] J.L. Heitsch, Flat bundles and residues for foliations, Invent. Math. 73 (1983), 271–285. [178] J.L. Heitsch and S. Hurder, Secondary classes, Weil measures and the geometry of foliations, J. Differential Geom. 20 (1984), 291–309. [179] J.L. Heitsch and H.B. Lawson, Transgressions, Chern–Simons invariants and the classical groups, J. Differential Geom. 9 (1974), 423–434. [180] J.L. Heitsch and C. Lazarov, Lefschetz theorem for foliated manifolds, Topology 29 (2) (1990), 127–162. [181] J.L. Heitsch and C. Lazarov, Homotopy invariance of foliation Betti numbers, Invent. Math. 104 (2) (1991), 321–347. [182] J.L. Heitsch and C. Lazarov, Rigidity theorems by surfaces and spin manifolds, Michigan Math. J. 38 (2) (1991), 285–297. [183] J.L. Heitsch and C. Lazarov, Spectral asymptotics of foliated manifolds, Illinois J. Math. 38 (4) (1994), 653–678. [184] M. Herman, The Godbillon–Vey invariant of foliations by planes of T3 , Lecture Notes in Math. 597 (1977), 294–307. [185] M. Herman, Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations, Publ. Math. IHES 49 (1979), 5–223. [186] R. Hermann, The differential geometry of foliations, I, Ann. of Math. 72 (2) (1960), 445–457. [187] R. Hermann, The differential geometry of foliations, II, J. Math. Mech. 11 (1962), 305–315. [188] M. Hirsch and W. Thurston, Foliated bundles, invariant measures and flat manifolds, Ann. of Math. 101 (1975), 369–390. [189] S. Hurder, Problems on rigidity of group actions and cocycles, Ergodic Theory Dynam. Systems 5 (1985), 473–484. [190] S. Hurder, The Godbillon measure of amenable foliations, J. Differential Geom. 23 (1986), 347–365. [191] S. Hurder, Rigidity for Anosov actions of higher lattices, Ann. of Math. 135 (1992), 361–410. [192] S. Hurder and A. Katok, Ergodic theory and Weil measures for foliations, Ann. of Math. 126 (1987), 221–275.

72

R. Barre and A. El Kacimi Alaoui

[193] S. Hurder and Y. Mitsumatsu, On the intersection products of transverse invariant measures, Indiana Univ. Math. J. 40 (1992), 1169–1183. [194] J.-P. Jouanolou, Une preuve élémentaire d’un théorème de Thurston, Topology 17 (1978), 109–110. [195] F. Kamber and P. Tondeur, Foliated bundles and characteristic classes, Lecture Notes in Math. 493 (1975). [196] F. Kamber and P. Tondeur, Duality for Riemannian foliations, Proc. Sympos. Pure Math. 40 (1983), 609– 618. [197] F. Kamber and P. Tondeur, Hodge de Rham theory for Riemannian foliations, Math. Ann. 277 (1987), 415–431. [198] M. Kellum, Uniformly quasi-isometric foliations, Ergodic Theory Dynam. Systems 13 (1993), 101–122. [199] K. Kodaira and D.C. Spencer, Multifoliate structures, Ann. of Math. 74 (1961), 52–100. [200] K. Kuperberg, A smooth counterexample to the Seifert conjecture, Ann. of Math. 140 (1994), 723–732. [201] C. Lamoureux, Holonomie et feuilles exceptionnelles, Ann. Inst. Fourier 26 (4) (1976), 273–300. [202] R. Langevin and H. Rosenberg, On stability of compact leaves and fibrations, Topology 16 (1977), 107– 111. [203] F. Laudenbach, Formes différentielles de degré 1 fermées non singulières et classes d’homotopie de leurs noyaux, Comment. Math. Helv. 51 (1971), 447–464. [204] F. Laudenbach and R. Roussarie, Un exemple de feuilletage sur S3 , Topology 9 (1970), 63–70. [205] H.B. Lawson, The Quantitative Theory of Foliations, Regional Conference Series in Mathematics, vol. 27, Amer. Math. Soc. (1970). [206] H.B. Lawson, Codimension one foliations of spheres, Ann. of Math. 94 (2) (1971), 494–503. [207] H.B. Lawson, Foliations, Bull. Amer. Math. Soc. 80 (1974), 369–418. [208] C. Lazarov, A vanishing theorem for certain rigid classes, Michigan Math. J. 28 (1981), 89–95. [209] C. Lazarov, An index theorem for foliations, Illinois J. Math. 3 (1) (1986), 101–121. [210] C. Lazarov and J.S. Pasternak, Residues and characteristic classes for Riemannian foliations, J. Differential Geom. 11 (1976), 599–612. [211] D. Lehmann, Sur l’approximation de certains feuilletages nilpotents par des fibrations, C. R. Acad. Sci. Paris 286 (1978), 251–254. [212] J. Lehmann-Lejeune, Cohomologie sur le fibré transverse à un feuilletage, C. R. Acad. Sci. Paris 295 (1982), 495–498. [213] G. Levitt, Foliations and laminations in hyperbolic surfaces, Topology 22 (1983), 119–135. [214] G. Levitt, Groupe fondamental de l’espace des feuilles dans les feuilletages sans holonomie, J. Differential Geom. 31 (3) (1990), 711–761. [215] A. Lichnerowicz, Algèbres de Lie attachées à un feuilletage, Ann. Fac. Sci. Toulouse I (1979), 45–76. [216] W.B.R. Lickorish, A foliation for 3-manifolds, Ann. of Math. 82 (2) (1965), 414–420. [217] J.-J. Lœb and M. Nicolau, Holomorphic flows and complex structures on products of odd dimensional spheres, Math. Ann. 306 (1996), 781–817. [218] D. Mac Duff, On groups of volume preserving diffeomorphisms and foliations with transverse volume form, Proc. London Math. Soc. 43 (1981), 295–320. [219] G.W. Mackey, Ergodic theory and virtual groups, Math. Ann. 166 (1956), 359–363. [220] T. Matsuaka and S. Morita, On characteristic classes of Kähler foliations, Osaka J. Math. 16 (1979), 539–550. [221] S. Matsumoto, Some remarks on foliated S 1 -bundles, Invent. Math. 90 (1987), 343–358. [222] S. Matsumoto, Perturbation of the Hopf flow and transverse foliations, Nagoya Math. J. 122 (1991), 75– 82. [223] S. Matsumoto, Lie affine foliations on 4-manifolds, Invent. Math. 109 (1) (1992), 1–16. [224] S. Matsumoto, Codimension one foliations on solvable manifolds, Comment. Math. Helv. 68 (4) (1993), 633–652. [225] X. Masa, Quelques propriétés des feuilletages de codimension un à connexion transverse projetable, C. R. Acad. Sci. Paris Sér. A 284 (1977), 811–812. [226] X. Masa, Characteristic classes of subfoliations, Ann. Inst. Fourier 31 (2) (1981), 61–86. [227] X. Masa, Duality and minimality in Riemannian foliations, Comment. Math. Helv. 67 (1992), 17–27. [228] G. Meigniez, Bouts d’un groupe opérant sur la droite. Applications à la topologie des feuilletages, Tohoku Math. J. (2) 43 (4) (1991), 473–500.

Foliations

73

[229] G. Meigniez, On completeness of transversely projective foliations, Topology 31 (2) (1992), 421–432. [230] G. Meigniez, Feuilletages de Lie résolubles, Ann. Fac. Sci. Toulouse (6) 4 (1995), 801–817. [231] G. Meigniez, Prolongement des homotopies, Q-variétés et cycles tangents, Ann. Inst. Fourier 47 (3) (1997), 945–965. [232] A. Miernowski and W. Mozgawa, On Molino lifting of Riemannian vector fields, Acta Math. Hungar. 55 (1990), 185–191. [233] J. Milnor, On the existence of a connection with curvature zero, Comment. Math. Helv. 32 (1958), 215– 223. [234] Y. Mitsumatsu, A relation between the topological invariance of the Godbillon–Vey invariant and the differentiability of Anosov foliations, Foliations, Adv. Stud. Pure Math., vol. 5 (1985), 159–167. [235] Y. Mitsumatsu, On the self intersection of foliation cycles, Trans. Amer. Math. Soc. 334 (2) (1992), 1169– 1183. [236] T. Mizutani, S. Morita and T. Tsuboi, The Godbillon–Vey classes of codimension one foliations which are almost without holonomy, Ann. of Math. (2) 113 (1981), 515–527. [237] T. Mizutani, S. Morita and T. Tsuboi, On the cobordism classes of codimension one foliations which are almost without holonomy, Topology 22 (1983), 325–343. [238] I. Moerdijk and D.A. Pronk, Orbifolds, sheaves and groupoids, K-Theory 12 (1) (1997), 3–21. [239] I. Moerdijk and D.A. Pronk, Simplicial cohomology of orbifolds, Indag. Math. (N.S.) 10 (2) (1999), 269– 293. [240] P. Molino, Propriétés cohomologiques et propriétés topologiques des feuilletages à connexion transverse projetable, Topology 12 (1973), 317–325. [241] P. Molino, Sur la géométrie transverse des feuilletages, Ann. Inst. Fourier 25 (1975), 279–284. [242] P. Molino, Etude des feuilletages transversalement complets et applications, Ann. Sci. École Norm. Sup. 10 (1977), 289–307. [243] P. Molino, Géométrie globale des feuilletages riemanniens, Indag. Math. 44 (1982), 45–76. [244] P. Molino, Feuilletages riemanniens, Astérisque 116 (1984). [245] P. Molino, Riemannian Foliations, Birkhäuser (1988). [246] P. Molino, Dualité symplectique, feuilletages et géométrie du moment, Publ. Mat. Barcelona 33 (3) (1989), 533–541. [247] S. Morita, On characteristic classes of Riemannian foliations, Osaka J. Math. 16 (1979), 161–172. [248] R. Moussu, Feuilletages sans holonomie d’une variété fermée, C. R. Acad. Sci. Paris Sér. A 270 (1970), 1308–1311. [249] R. Moussu, Feuilletages presque sans holonomie, C. R. Acad. Sci. Paris Sér. A 272 (1971), 114–117. [250] R. Moussu, Sur les classes exotiques des feuilletages, Lecture Notes in Math. 392 (1974), 37–42. [251] R. Moussu and F. Pelletier, Sur le théorème de Poincaré–Bendixon, Ann. Inst. Fourier 24 (1) (1974), 131–148. [252] R. Moussu and R. Roussarie, Relations de conjugaison et de cobordisme entre certains feuilletages, Publ. Math. IHES 43 (1973), 143–168. [253] W. Mozgawa, Feuilletages de Killing, Collect. Math. 36 (3) (1985), 285–290. [254] M.P. Muller, Sur les composantes de Novikov des feuilletages, Topology 19 (1980), 199–201. [255] Ngo Van Que and R. Roussarie, Sur l’isotopie des formes fermées en dimension 3, Invent. Math. 64 (1981), 69–87. [256] M. Nicolau and A. Reventos, On some geometrical properties of Seifert bundles, Israel J. Math. 47 (1984), 323–334. [257] S. Nishikawa, Residues and secondary characteristic classes for projective foliations, Proc. Japan Acad. 54 (1978), 79–82. [258] S. Nishikawa and H. Sato, On characteristic classes of Riemannian, conformal and projective foliations, J. Math. Soc. Japan 28 (1976), 223–241. [259] S. Nishikawa and M. Takeushi, G-foliations and semisimple flat homogeneous spaces, Tohoku Math. J. 30 (1978), 307–335. [260] T. Nishimori, SRH-decompositions of codimension one foliations and the Godbillon–Vey class, Tohoku Math. J. 32 (1980), 9–34. [261] S.P. Novikov, Foliations of codimension one, Soviet Math. Dokl. 5 (1964), 540–544 and 1023–1025. [262] S.P. Novikov, The topology of foliations, Trans. Moscow Math. Soc. 14 (1965), 268–304.

74

R. Barre and A. El Kacimi Alaoui

[263] J. Palis, Rigidity of the centralizers and structural stability of suspended foliations, Lecture Notes in Math. 652 (1978), 114–121. [264] C.F.B. Palmeira, Open manifolds foliated by planes, Ann. of Math. 107 (1978), 109–131. [265] E. Park and K. Richardson, The basic Laplacian for Riemannian foliations, Amer. J. Math. 118 (1996), 1249–1275. [266] A. Phillips, Foliations on open manifolds, Comment. Math. Helv. 43 (1968), 204–211 and 44 (1969), 367–370. [267] H.P. Pittie, The secondary characteristic classes of parabolic foliations, Comment. Math. Helv. 54 (1979), 601–614. [268] M. Plaisant, Sur l’intégrabilité des algèbres de Lie banachiques dans le cadre des Q-variétés, Thèse 3ème cycle, Valenciennes (1980). [269] J.F. Plante, Anosov flows, Amer. J. Math. 94 (1972), 729–754. [270] J.F. Plante, Asymptotic properties of foliations, Comment. Math. Helv. 47 (1972), 449–456. [271] J.F. Plante, A generalization of the Poincaré–Bendixon theorem for foliations of codimension one, Topology 12 (1973), 177–181. [272] J.F. Plante, Foliations transverse to fibers of a bundle, Proc. Amer. Math. Soc. 42 (1974), 631–635. [273] J.F. Plante, Foliations with measure preserving holonomy, Ann. of Math. 102 (1975), 327–361. [274] J.F. Plante, Foliations of 3-manifolds with solvable fundamental group, Invent. Math. 51 (1979), 219–230. [275] J.F. Plante, Stability of codimension one foliations by compact leaves, Topology 22 (1983), 173–177. [276] J.F. Plante, Polycyclic groups and transversely affine foliations, J. Differential Geom. 36 (3) (1992), 521– 534. [277] J.F. Plante and W.P. Thurston, Polynomial growth in holonomy groups of foliations, Comment. Math. Helv. 51 (1976), 567–584. [278] J. Pradines, Sur une classe remarquable de relations d’équivalence sur les variétés, Rev. An. de Math. Jaca (1973). [279] J. Pradines, Un feuilletage sans holonomie transversale dont le quotient n’est pas une Q-variété, C. R. Acad. Sci. Paris 288 (1979), 245–248. [280] J. Pradines and B. Bigonnet, Graphe d’un feuilletage singulier, C. R. Acad. Sci. Paris 300 (1985), 439– 442. [281] J. Pradines and J. Wouafo-Kamga, La catégorie des QF -variétés, C. R. Acad. Sci. Paris Sér. A 288 (1979), 717–719. [282] M. Puta, Cohomology classes and foliated manifolds, Nonlinear Anal. (1987), 137–159. [283] A. Ramsay, Virtual groups and group actions, Adv. in Math. 6 (1971), 253–322. [284] O.H. Rasmussen, Exotic characteristic classes for holomorphic foliations, Invent. Math. 46 (1978), 153– 171. [285] G. Reeb, Sur certaines propriétés topologiques des variétés feuilletées, Hermann, Paris (1950). [286] G. Reeb, Sur les structures feuilletées de codimension un et sur un théorème de Denjoy, Ann. Inst. Fourier 11 (1961), 185–200. [287] G. Reeb, Feuilletages résultats anciens et nouveaux, Painlevé, Hector and Martinet, eds, Presses Univ. de Montréal (1974). [288] B.L. Reinhart, Harmonic integrals on almost product manifolds, Trans. Amer. Math. Soc. 88 (1958), 243– 276. [289] B.L. Reinhart, Foliated manifolds with bundle-like metrics, Ann. of Math. 69 (1959), 119–132. [290] B.L. Reinhart, Harmonic integrals on foliated manifolds, Amer. J. Math. 81 (1959), 529–586. [291] B.L. Reinhart, Differential Geometry of Foliations, Ergebnisse der Mathematik, vol. 99, Springer (1983). [292] B.L. Reinhart and J.W. Wood, A metric formula for the Godbillon–Vey invariant for foliations, Proc. Amer. Math. Soc. 38 (1973), 427–430. [293] J. Renault, A Groupoid Approach to C ∗ -Algebras, Lecture Notes in Math., vol. 793, Springer (1980). [294] K. Richardson, Asymptotic of heat kernels on Riemannian foliations, Geom. Funct. Anal. 8 (1998), 356– 401. [295] C. Roger, Etude des Γ -structures de codimension un sur la sphère S2 , Ann. Inst. Fourier 23 (4) (1973), 213–227. [296] C. Roger, Méthodes homotopiques et cohomologiques en théorie des feuilletages, Thèse Orsay (1974). [297] H. Rosenberg, Actions of Rn on manifolds, Comment. Math. Helv. 41 (1966), 170–178.

Foliations

75

[298] H. Rosenberg, Foliations by planes, Topology 7 (1968), 143–145. [299] H. Rosenberg and R. Roussarie, Topological equivalence of Reeb foliations, Topology 9 (1970), 231–242. [300] H. Rosenberg and R. Roussarie, Les feuilles exceptionnelles ne sont pas exceptionnelles, Comment. Math. Helv. 45 (1970), 517–523. [301] H. Rosenberg and R. Roussarie, Reeb foliations, Ann. of Math. (2) 91 (1970), 1–24. [302] H. Rosenberg and R. Roussarie, Some remarks on stability of foliations, J. Differential Geom. 10 (1975), 207–219. [303] H. Rosenberg, R. Roussarie and D. Weil, A classification of closed orientable 3-manifolds of rank two, Ann. of Math. 91 (1970), 449–464. [304] H. Rosenberg, R. Roussarie and D. Weil, A classification of the topological types of R2 -actions, Publ. Math. IHES 43 (1973), 261–272. [305] J. Rosenberg, K-theory of group C ∗ -algebras, foliation C ∗ -algebras and crossed products, Index Theory of Elliptic Operators, Foliations and Operator Algebras, New Orleans, Contemp. Math. 70 (1988), 251– 301. [306] R. Roussarie, Sur les feuilletages des variétés de dimension 3, Ann. Inst. Fourier 21 (3) (1971), 13–81. [307] R. Roussarie, Plongement dans les variétés feuilletées et classification des feuilletages sans holonomie, Publ. Math. IHES 43 (1974), 101–141. [308] D. Ruelle and D. Sullivan, Currents, flows and diffeomorphisms, Topology 14 (1975), 319–327. [309] H. Rummler, Quelques notions simples en géométrie riemannienne et leurs applications aux feuilletages compacts, Comment. Math. Helv. 54 (1979), 224–239. [310] S. Sacksteder, Some properties of foliations, Ann. Inst. Fourier 14 (1) (1964), 31–35. [311] S. Sacksteder, Foliations and pseudogroups, Amer. J. Math. 87 (1965), 79–102. [312] S. Sacksteder, A remark on Thurston stability theorem, Ann. Inst. Fourier 25 (2) (1975), 219–220. [313] S. Sacksteder and J. Schwartz, Limit sets for foliations, Ann. Inst. Fourier 15 (2) (1965), 201–213. [314] E. Salem, Une généralisation du théorème de Myers–Steenrod aux pseudo-groupes d’isométries locales, Ann. Inst. Fourier 35 (1985), 137–158. [315] E. Salhi, Sur les ensembles minimaux locaux, C. R. Acad. Sci. Paris 295 (1982), 691–694. [316] E. Salhi, Sur un théorème de structure des feuilletages de codimension 1, C. R. Acad. Sci. Paris 300 (1985), 635–638. [317] E. Salhi, Niveau des feuilles, C. R. Acad. Sci. Paris 301 (1985), 219–222. [318] M. Saralegi, The Euler class for flows of isometries, Differential Geometry, Santiago de Compostela, 1984, Pitman Research Notes in Math., vol. 131 (1985), 220–227. [319] K.S. Sarkaria, The de Rham cohomology of foliated manifolds, Thesis, Stony Brook (1974). [320] K.S. Sarkaria, A finiteness theorem for foliated manifolds, J. Math. Soc. Japan 30 (4) (1978), 687–696. [321] I. Satake, On the generalization of the notion of manifold, Proc. NAS 42 (1956), 359–363. [322] G. Schwarz, On the de Rham cohomology of the leaf space of a foliation, Topology 13 (1974), 185–188. [323] S. Schwartzman, Asymptotic cycles, Ann. of Math. 66 (1957), 270–284. [324] P. Schweitzer, Counterexamples to the Seifert conjecture and opening closed leaves of foliations, Ann. of Math. 100 (1974), 386–400. [325] P. Schweitzer, Stability of compact leaves with trivial linear holonomy, Topology 27 (1) (1988), 37–56. [326] P. Schweitzer, Existence of codimension one foliations with minimal leaves, Ann. Global Anal. Geom. 9 (1) (1991), 77–81. [327] P. Schweitzer, Codimension one foliations without compact leaves, Comment. Math. Helv. 70 (2) (1995), 171–209. [328] P. Scott, The geometry of 3-manifolds, Bull. London Math. Soc. 15 (1983), 401–487. [329] H. Seifert, Closed integral curves in 3-space and isotopic deformations, Proc. Amer. Math. Soc. 1 (1950), 287–302. [330] F. Sergeraert, Feuilletages et difféomorphismes infiniment tangents à l’identité, Invent. Math. 39 (1977), 253–275. [331] V. Sergiescu, Cohomologie basique et dualité des feuilletages riemanniens, Ann. Inst. Fourier de Grenoble 53 (3) (1985), 135–158. [332] C. Series, Foliations of polynomial growth are hyperfinite, Israel J. Math. 34 (1979), 245–258. [333] H. Shulman and J. Stasheff, De Rham theory for BΓ , Lecture Notes in Math. 652 (1978), 62–74.

76

R. Barre and A. El Kacimi Alaoui

[334] J.-C. Sikorav, Formes différentielles fermées sur le n-tore, Comment. Math. Helv. 57 (1982), 79–106. [335] J.D. Sondow, When is a manifold a leaf of some foliation? Bull. Amer. Math. Soc. 81 (1975), 622–624. [336] D. Sullivan, A foliation of geodesics is characterized by having no “tangent homologies”, J. Pure Appl. Algebra 13 (1973), 101–104. [337] D. Sullivan, A counterexample to the periodic orbit conjecture, Publ. Math. IHES 46 (1976), 5–14. [338] D. Sullivan, Cycles for the dynamical study of foliated manifolds and complex manifolds, Invent. Math. 36 (1976), 225–255. [339] D. Sullivan, A homological characterization of foliations consisting of minimal surfaces, Comment. Math. Helv. 54 (1979), 218–223. [340] H. Takai, KK-Theory for the C ∗ -algebras of Anosov foliations, Geometric Methods in Operator Algebras, Pitman Research Notes in Math., vol. 123 (1986), 387–399. [341] I. Tamura, Every odd dimensional homotopy sphere has a codimension one foliation, Comment. Math. Helv. 47 (1972), 73–79. [342] E. Thomas, Vector fields on manifolds, Bull. Amer. Math. Soc. 75 (1969), 643–683. [343] W.P. Thurston, Foliations of 3-manifolds which are circle bundles, Thesis, Berkeley (1972). [344] W.P. Thurston, Noncobordant foliations of S3 , Bull. Amer. Math. Soc. 78 (1972), 511–514. [345] W.P. Thurston, A generalization of the Reeb stability theorem, Topology 13 (1974), 347–352. [346] W.P. Thurston, Foliations and groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 304–307. [347] W.P. Thurston, The theory of foliations of codimension greater than one, Comment. Math. Helv. 49 (1974), 214–231. [348] W.P. Thurston, Existence of codimension one foliations, Ann. of Math. 102 (2) (1976), 249–268. [349] W.P. Thurston, The geometry and topology of 3-manifolds, Mimeographed Notes, Princeton (1977). [350] D. Tischler, On Fibering certain foliated manifolds over S 1 , Topology 9 (1970), 153–154. [351] P. Tondeur, Foliations on Riemannian Manifolds, Springer (1988). [352] P. Tondeur, Geometry of Foliations, Monographs in Mathematics, vol. 90, Birkhäuser (1997). [353] T. Tsuboi, Foliations with trivial F -subgroups, Topology 18 (1979), 223–233. [354] T. Tsuboi, On the cobordism classes of codimension one foliations which are almost without holonomy, Topology 22 (1983), 325–343. [355] T. Tsuboi, Foliated cobordism classes of certain foliated S1 -bundles, Topology 23 (1984), 233–244. [356] T. Tsuboi, On the foliated product of class C 1 , Ann. of Math. 130 (1989), 227–271. [357] T. Tsuboi, The Godbillon–Vey invariant and the foliated cobordism group, Proc. Japan Acad. 68 (1992), 85–90. [358] N. Tsuchiya, The Nishimori decompositions of codimension one foliations and the Godbillon–Vey classes, Tohoku Math. J. 34 (1982), 343–365. [359] H. Uzuki, Characteristic classes of foliated principal GLr -bundles, Hokkaido Math. J. 4 (1975), 159–168. [360] I. Vaisman, Variétés riemanniennes feuilletées, Czechoslovak Math. J. 21 (1971), 46–75. [361] I. Vaisman, Conformal foliations, Kodai Math. J. 2 (1979), 22–37. [362] W.T. Van Est, Rapport sur les S-atlas, Astérisque 116 (1984). [363] A. Verjovsky, A note on Anosov flows, Bol. Soc. Mat. Mexicana (1) 1 (1970), 36–37. [364] A. Verjovsky, Codimension one Anosov flows, Bol. Soc. Mat. Mexicana (2) 19 (1974), 49–77. [365] A. Verjovsky, A uniformisation theorem for holomorphic foliations, The Lefschetz Centenial Conference, Mexico, 1984, Contemp. Math. (3) 58 (1987), 233–253. [366] E. Vogt, Stable foliations of 4-manifolds by closed surfaces. Local structures and free actions of finite cyclic and dihedral groups on surfaces, Invent. Math. 22 (1973), 321–348. [367] E. Vogt, A foliation of R3 and other punctured 3-manifolds by circles, Publ. Math. IHES 69 (1981), 215– 232. [368] E. Vogt, The first cohomology group of leaves and local stability of compact foliations, Manuscripta Math. 37 (1982), 229–267. [369] E. Vogt, Existence of foliations of Euclidean spaces with all leaves compact, Math. Ann. 296 (1) (1993), 159–178. [370] A. Wadsley, Geodesic foliations by circles, J. Differential Geom. 10 (1975), 541–550. [371] P. Walczak, Dynamics of the geodesic flow of a foliation, Ergodic Theory Dynam. Systems 8 (1988), 637–650.

Foliations

77

[372] H.E. Winkelnkemper, The graph of a foliation, Ann. Global Anal. Geom. 1 (1983), 53–75. [373] R. Wolak, Characteristic classes of almost flat structures, Geom. Dedicata 24 (1987), 207–220. [374] R. Wolak, Maximal subalgebras in the algebra of foliated vector fields of a Riemannian foliation, Comment. Math. Helv. 64 (4) (1989), 536–541. [375] R. Wolak, The graph of a totally geodesic foliation, Ann. Polon. Math. 60 (3) (1995), 241–247. [376] J. Wood, Foliations on 3-manifolds, Ann. of Math. 89 (2) (1969), 336–358. [377] J. Wood, Foliations of codimension one, Bull. Amer. Math. Soc. 75 (1970), 1107–1111. [378] J. Wood, Bundles with totally disconnected structure group, Comment. Math. Helv. 46 (1971), 257–273. [379] J. Wouafo-Kamga, Décomposition des G-structures d’ordre supérieur. Structures transverse des feuilletages, Thèse, Toulouse (1979). [380] A. Zeggar, Nombre et classe de Lefschetz pour un feuilletage riemannien, Ann. Fac. Sci. Toulouse (6) 1 (1) (1992), 105–131. [381] A. Zeghib, Laminations et hypersurfaces géodésiques des variétés hyperboliques, Ann. Sci. École Norm. Sup. (4) 24 (2) (1991), 171–188. [382] A. Zeghib, Sur les feuilletages géodésiques continus des variétés hyperboliques, Invent. Math. 114 (1) (1993), 193–206. [383] A. Zeghib, Feuilletages géodésiques appliqués, Math. Ann. 298 (4) (1994), 729–759. [384] A. Zeghib, Feuilletages géodésiques des variétés localement symétriques, Topology 36 (4) (1997), 805– 828. [385] R. Zimmer, On the Mostow rigidity theorem and measurable foliations, Israel J. Math. 43 (1982), 281–290.

This page intentionally left blank

CHAPTER 3

Symplectic Geometry Ana Cannas da Silva1 Departamento de Matemática, Instituto Superior Técnico, 1049-001 Lisboa, Portugal E-mail: [email protected]; [email protected]

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . 1. Symplectic manifolds . . . . . . . . . . . . . . . 1.1. Symplectic linear algebra . . . . . . . . . . 1.2. Symplectic forms . . . . . . . . . . . . . . 1.3. Cotangent bundles . . . . . . . . . . . . . . 1.4. Moser’s trick . . . . . . . . . . . . . . . . . 1.5. Darboux and Moser theorems . . . . . . . 1.6. Symplectic submanifolds . . . . . . . . . . 2. Lagrangian submanifolds . . . . . . . . . . . . . 2.1. First Lagrangian submanifolds . . . . . . . 2.2. Lagrangian neighborhood theorem . . . . . 2.3. Weinstein tubular neighborhood theorem . 2.4. Application to symplectomorphisms . . . . 2.5. Generating functions . . . . . . . . . . . . 2.6. Fixed points . . . . . . . . . . . . . . . . . 2.7. Lagrangians and special Lagrangians in Cn 3. Complex structures . . . . . . . . . . . . . . . . 3.1. Compatible linear structures . . . . . . . . 3.2. Compatible almost complex structures . . 3.3. Integrability . . . . . . . . . . . . . . . . . 3.4. Kähler manifolds . . . . . . . . . . . . . . 3.5. Hodge theory . . . . . . . . . . . . . . . . 3.6. Pseudoholomorphic curves . . . . . . . . . 4. Symplectic geography . . . . . . . . . . . . . . 4.1. Existence of symplectic forms . . . . . . . 4.2. Fibrations and sums . . . . . . . . . . . . . 4.3. Symplectic blow-up . . . . . . . . . . . . . 4.4. Uniqueness of symplectic forms . . . . . . 4.5. Invariants for 4-manifolds . . . . . . . . . 4.6. Lefschetz pencils . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 Current address: Department of Mathematics, Princeton University, Princeton, NJ 08544-1000, USA.

HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 79

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 82 82 84 85 88 90 91 93 93 95 96 98 99 102 104 108 108 110 113 117 120 122 124 124 128 131 134 135 138

80

A. Cannas da Silva

5. Hamiltonian geometry . . . . . . . . . . . . . 5.1. Symplectic and Hamiltonian vector fields 5.2. Arnold conjecture and Floer homology . 5.3. Euler–Lagrange equations . . . . . . . . 5.4. Legendre transform . . . . . . . . . . . . 5.5. Integrable systems . . . . . . . . . . . . . 5.6. Symplectic and Hamiltonian actions . . . 5.7. Convexity . . . . . . . . . . . . . . . . . 6. Symplectic reduction . . . . . . . . . . . . . . 6.1. Marsden–Weinstein–Meyer theorem . . . 6.2. Applications and generalizations . . . . . 6.3. Moment map in gauge theory . . . . . . 6.4. Symplectic toric manifolds . . . . . . . . 6.5. Delzant’s construction . . . . . . . . . . 6.6. Duistermaat–Heckman theorems . . . . . References . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

139 139 142 143 147 150 153 158 160 160 163 166 169 174 178 183

Symplectic geometry

81

Introduction This is an overview of symplectic geometry1 —the geometry of symplectic manifolds. From a language for classical mechanics in the XVIII century, symplectic geometry has matured since the 1960’s to a rich and central branch of differential geometry and topology. A current survey can thus only aspire to give a partial flavor on this exciting field. The following six topics have been chosen for this handbook: 1. Symplectic manifolds are manifolds equipped with symplectic forms. A symplectic form is a closed nondegenerate 2-form. The algebraic condition (nondegeneracy) says that the top exterior power of a symplectic form is a volume form, therefore symplectic manifolds are necessarily even-dimensional and orientable. The analytical condition (closedness) is a natural differential equation that forces all symplectic manifolds to being locally indistinguishable: they all locally look like an even-dimensional Euclidean space equipped with the dxi ∧ dyi symplectic form. All cotangent bundles admit canonical symplectic forms, a fact relevant for analysis of differential operators, dynamical systems, classical mechanics, etc. Basic properties, major classical examples, equivalence notions, local normal forms of symplectic manifolds and symplectic submanifolds are discussed in Section 1. 2. Lagrangian submanifolds2 are submanifolds of symplectic manifolds of half dimension and where the restriction of the symplectic form vanishes identically. By the Lagrangian creed [137], everything is a Lagrangian submanifold, starting with closed 1-forms, real functions modulo constants and symplectomorphisms (diffeomorphisms that respect the symplectic forms). Section 2 also describes normal neighborhoods of Lagrangian submanifolds with applications. 3. Complex structures or almost complex structures abound in symplectic geometry: any symplectic manifold possesses almost complex structures, and even so in a compatible sense. This is the point of departure for the modern technique of studying pseudoholomorphic curves, as first proposed by Gromov [64]. Kähler geometry lies at the intersection of complex, Riemannian and symplectic geometries, and plays a central role in these three fields. Section 3 includes the local normal form for Kähler manifolds and a summary of Hodge theory for Kähler manifolds. 4. Symplectic geography is concerned with existence and uniqueness of symplectic forms on a given manifold. Important results from Kähler geometry remain true in the more general symplectic category, as shown using pseudoholomorphic methods. This viewpoint was more recently continued with work on the existence of certain symplectic 1 The word symplectic in mathematics was coined in the late 1930’s by Weyl [142, p. 165] who substituted the Latin root in complex by the corresponding Greek root in order to label the symplectic group (first studied by Abel). An English dictionary is likely to list symplectic as the name for a bone in a fish’s head. 2 The name Lagrangian manifold was introduced by Maslov [93] in the 1960’s, followed by Lagrangian plane, etc., introduced by Arnold [2].

82

A. Cannas da Silva

submanifolds, in the context of Seiberg–Witten invariants, and with topological descriptions in terms of Lefschetz pencils. Both of these directions are particularly relevant to 4-dimensional topology and to mathematical physics, where symplectic manifolds occur as building blocks or as key examples. Section 4 treats constructions of symplectic manifolds and invariants to distinguish them. 5. Hamiltonian geometry is the geometry of symplectic manifolds equipped with a moment map, that is, with a collection of quantities conserved by symmetries. With roots in Hamiltonian mechanics, moment maps became a consequential tool in geometry and topology. The notion of a moment map arises from the fact that, to any real function on a symplectic manifold, is associated a vector field whose flow preserves the symplectic form and the given function; this is called the Hamiltonian vector field of that (Hamiltonian) function. The Arnold conjecture in the 60’s regarding Hamiltonian dynamics was a major driving force up to the establishment of Floer homology in the 80’s. Section 5 deals mostly with the geometry of moment maps, including the classical Legendre transform, integrable systems and convexity. 6. Symplectic reduction is at the heart of many symplectic arguments. There are infinitedimensional analogues with amazing consequences for differential geometry, as illustrated in a symplectic approach to Yang–Mills theory. Symplectic toric manifolds provide examples of extremely symmetric symplectic manifolds that arise from symplectic reduction using just the data of a polytope. All properties of a symplectic toric manifold may be read from the corresponding polytope. There are interesting interactions with algebraic geometry, representation theory and geometric combinatorics. The variation of reduced spaces is also addressed in Section 6.

1. Symplectic manifolds 1.1. Symplectic linear algebra Let V be a vector space over R, and let Ω : V × V → R be a skew-symmetric bilinear map. By a skew-symmetric version of the Gram–Schmidt process,3 there is a basis u1 , . . . , uk , e1 , . . . , en , f1 , . . . , fn of V for which Ω(ui , v) = Ω(ei , ej ) = Ω(fi , fj ) = 0 and Ω(ei , fj ) = δij for all i, j and all v ∈ V . Although such a basis is not unique, it is commonly referred to as a canonical basis. The dimension k of the subspace U = {u ∈ V | Ω(u, v) = 0 for all v ∈ V } is an invariant of the pair (V , Ω). Since k + 2n = dim V , the even number 2n is also an invariant of (V , Ω), called the rank of Ω. We denote ˜ := Ω(v, u). We say that Ω is symplecby Ω˜ : V → V ∗ the linear map defined by Ω(v)(u) tic (or nondegenerate) if the associated Ω˜ is bijective (i.e., the kernel U of Ω˜ is the trivial space {0}). In that case, the map Ω is called a linear symplectic structure on V , and the 3 Let u , . . . , u be a basis of U := {u ∈ V | Ω(u, v) = 0 for all v ∈ V }, and W a complementary subspace k 1 such that V = U ⊕ W . Take any nonzero e1 ∈ W . There is f1 ∈ W with Ω(e1 , f1 ) = 1. Let W1 be the span of e1 , f1 and W1Ω := {v ∈ V | Ω(v, u) = 0 ∀u ∈ W1 }. Then W = W1 ⊕ W1Ω . Take any nonzero e2 ∈ W1Ω . There is f2 ∈ W1Ω for which Ω(e2 , f2 ) = 1. Let W2 be the span of e2 , f2 , and so on.

Symplectic geometry

83

pair (V , Ω) is called a symplectic vector space. A linear symplectic structure Ω expresses a duality by the bijection Ω˜ : V −→ V ∗ , similar to the (symmetric) case of an inner product. By considering a canonical basis, we see that the dimension of a symplectic vector space (V , Ω) must be even, dim V = 2n, and that V admits a basis e1 , . . . , en , f1 , . . . , fn satisfying Ω(ei , fj ) = δij and Ω(ei , ej ) = 0 = Ω(fi , fj ). Such a basis is then called a symplectic basis of (V , Ω), and, in terms of exterior algebra, Ω = e1∗ ∧ f1∗ + · · · + en∗ ∧ fn∗ , where e1∗ , . . . , en∗ , f1∗ , . . . , fn∗ is the dual basis. With respect to a symplectic basis, the map Ω is represented by the matrix 0 Id . −Id 0 E XAMPLES . 1. The prototype of a symplectic vector space is (R2n , Ω0 ) with Ω0 such that the canonical basis e1 = (1, 0, . . . , 0), . . . , en , f1 , . . . , fn = (0, . . . , 0, 1) is a symplectic basis. Bilinearity then determines Ω0 on other vectors. 2. For any real vector space E, the direct sum V = E ⊕ E ∗ has a canonical symplectic structure determined by the formula Ω0 (u ⊕ α, v ⊕ β) = β(u) − α(v). If e1 , . . . , en is a basis of E, and f1 , . . . , fn is the dual basis, then e1 ⊕ 0, . . . , en ⊕ 0, 0 ⊕ f1 , . . . , 0 ⊕ fn is a symplectic basis for V . Given a linear subspace W of a symplectic vector space (V , Ω), its symplectic orthogonal is the subspace W Ω := {v ∈ V | Ω(v, u) = 0 for all u ∈ W }. By nondegeneracy, we have dim W + dim W Ω = dim V and (W Ω )Ω = W . For subspaces W and Y , we have (W ∩ Y )Ω = W Ω + Y Ω , and if W ⊆ Y then Y Ω ⊆ W Ω . There are special types of linear subspaces of a symplectic vector space (V , Ω). A subspace W is a symplectic subspace if the restriction Ω|W is nondegenerate, that is, W ∩ W Ω = {0}, or equivalently V = W ⊕ W Ω . A subspace W is an isotropic subspace if Ω|W ≡ 0, that is, W ⊆ W Ω . A subspace W is a coisotropic subspace if W Ω ⊆ W . A subspace W is a Lagrangian subspace if it is both isotropic and coisotropic, or equivalently, if it is an isotropic subspace with dim W = 12 dim V . A basis e1 , . . . , en of a Lagrangian subspace can be extended to a symplectic basis: choose f1 in the symplectic orthogonal to the linear span of {e2 , . . . , en }, etc. E XAMPLES . 1. For a symplectic basis as above, the span of e1 , f1 is symplectic, that of e1 , e2 isotropic, that of e1 , . . . , en , f1 coisotropic, and that of e1 , . . . , en Lagrangian. 2. The graph of a linear map A : E → E ∗ is a Lagrangian subspace of E ⊕ E ∗ with the canonical symplectic structure if and only if A is symmetric (i.e., (Au)v = (Av)u). Therefore, the Grassmannian of all Lagrangian subspaces in a 2n-dimensional symplectic vector space has dimension n(n+1) 2 . A symplectomorphism ϕ between symplectic vector spaces (V , Ω) and (V , Ω ) is a linear isomorphism ϕ : V −→ V such that ϕ ∗ Ω = Ω.4 If a symplectomorphism exists, 4 By definition, (ϕ ∗ Ω )(u, v) = Ω (ϕ(u), ϕ(v)).

84

A. Cannas da Silva

(V , Ω) and (V , Ω ) are said to be symplectomorphic. Being symplectomorphic is clearly an equivalence relation in the set of all even-dimensional vector spaces. The existence of canonical bases shows that every 2n-dimensional symplectic vector space (V , Ω) is symplectomorphic to the prototype (R2n , Ω0 ); a choice of a symplectic basis for (V , Ω) yields a symplectomorphism to (R2n , Ω0 ). Hence, nonnegative even integers classify equivalence classes for the relation of being symplectomorphic. Let Ω(V ) be the space of all linear symplectic structures on the vector space V . Take a Ω ∈ Ω(V ), and let Sp(V , Ω) be the group of symplectomorphisms of (V , Ω). The group GL(V ) of all isomorphisms of V acts transitively on Ω(V ) by pullback (i.e., all symplectic structures are related by a linear isomorphism), and Sp(V , Ω) is the stabilizer of the given Ω. Hence, Ω(V ) GL(V )/Sp(V , Ω). 1.2. Symplectic forms Let ω be a de Rham 2-form on a manifold5 M. For each point p ∈ M, the map ωp : Tp M × Tp M → R is skew-symmetric and bilinear on the tangent space to M at p, and ωp varies smoothly in p. D EFINITION 1.1. The 2-form ω is symplectic if ω is closed (i.e., its exterior derivative dω is zero) and ωp is symplectic for all p ∈ M. A symplectic manifold is a pair (M, ω) where M is a manifold and ω is a symplectic form. Symplectic manifolds must be even-dimensional. Moreover, the nth exterior power ωn of a symplectic form ω on a 2n-dimensional manifold is a volume form.6 Hence, any symn plectic manifold (M, ω) is canonically oriented. The form ωn! is called the symplectic volume or Liouville volume of (M, ω). When (M, ω) is a compact 2n-dimensional symplectic manifold, the de Rham cohomology class [ωn ] ∈ H 2n (M; R) must be nonzero by Stokes theorem. Therefore, the class [ω] must be nonzero, as well as its powers [ω]k = [ωk ] = 0. Exact symplectic forms can only exist on noncompact manifolds. Compact manifolds with a trivial even cohomology group H 2k (M; R), k = 0, 1, . . . , n, such as spheres S 2n with n > 1, can thus never be symplectic. On a manifold of dimension greater than 2, a function multiple f ω of a symplectic form ω is symplectic if and only if f is a nonzero locally constant function (this follows from the existence of a symplectic basis). E XAMPLES . 1. Let M = R2n with linear coordinates x1 , . . . , xn , y1 , . . . , yn . The form ω0 =

n

dxi ∧ dyi

i=1 5 Unless otherwise indicated, all vector spaces are real and finite-dimensional, all maps are smooth (i.e., C ∞ ) and all manifolds are smooth, Hausdorff and second countable. 6 A volume form is a nonvanishing form of top degree. If Ω is a symplectic structure on a vector space V of dimension 2n, its nth exterior power Ω n = Ω ∧ · · · ∧ Ω does not vanish. Actually, a skew-symmetric bilinear map Ω is symplectic if and only if Ω n = 0.

Symplectic geometry

2.

3.

4. 5. 6.

85

is symplectic, and the vectors ( ∂x∂ 1 )p , . . . , ( ∂x∂ n t)p , ( ∂y∂ 1 )p , . . . , ( ∂y∂ n )p constitute a symplectic basis of Tp M. Let M = Cn with coordinates z1 , . . . , zn . The form ω0 = 2i dzk ∧ d z¯ k is symplectic. In fact, this form coincides with that of the previous example under the identification Cn R2n , zk = xk + iyk . The 2-sphere S 2 , regarded as the set of unit vectors in R3 , has tangent vectors at p identified with vectors orthogonal to p. The standard symplectic form on S 2 is induced by the standard inner (dot) and exterior (vector) products: ωp (u, v) := p, u × v, for u, v ∈ Tp S 2 = {p}⊥ . This is the standard area form on S 2 with total area 4π . In terms of cylindrical polar coordinates 0 θ < 2π and −1 z 1 away from the poles, it is written ω = dθ ∧ dz. On any Riemann surface, regarded as a 2-dimensional oriented manifold, any area form, that is, any never vanishing 2-form, is a symplectic form. Products of symplectic manifolds are naturally symplectic by taking the sum of the pullbacks of the symplectic forms from the factors. If a (2n + 1)-dimensional manifold X admits a contact form, that is, a 1-form α such that α ∧ (dα)n is never vanishing, then the 2-form d(et α) is symplectic on X × R, and the symplectic manifold (X × R, d(et α)) is called the symplectization of the contact manifold (X, α). For more on contact geometry, see for instance the corresponding contribution in this volume.

D EFINITION 1.2. Let (M1 , ω1 ) and (M2 , ω2 ) be symplectic manifolds. A (smooth) map ψ : M1 → M2 is symplectic if ψ ∗ ω2 = ω1 .7 A symplectic diffeomorphism ϕ : M1 → M2 is a symplectomorphism. (M1 , ω1 ) and (M2 , ω2 ) are said to be symplectomorphic when there exists a symplectomorphism between them. The classification of symplectic manifolds up to symplectomorphism is an open problem in symplectic geometry. However, the local classification is taken care of by the Darboux theorem (Theorem 1.9): the dimension is the only local invariant of symplectic manifolds up to symplectomorphisms. That is, just as any n-dimensional manifold is locally diffeomorphic to Rn , any symplectic manifold (M 2n , ω) is locally symplectomorphic to (R2n , ω0 ). As a consequence, if we prove for (R2n , ω0 ) a local assertion that is invariant under symplectomorphisms, then that assertion holds for any symplectic manifold. We will hence refer to R2n , with linear coordinates (x1 , . . . , xn , y1 , . . . , yn ), and with symplectic form ω0 = ni=1 dxi ∧ dyi , as the prototype of a local piece of a 2n-dimensional symplectic manifold.

1.3. Cotangent bundles Cotangent bundles are major examples of symplectic manifolds. Let (U, x1 , . . . , xn ) be a coordinate chart for a manifold X, with associated cotangent coordinates (T ∗ U, x1 , . . . , xn , 7 By definition of pullback, we have (ψ ∗ ω ) (u, v) = (ω ) 2 p 2 ψ(p) (dψp (u), dψp (v)), at tangent vectors u, v ∈

Tp M1 .

86

A. Cannas da Silva

ξ1 , . . . , ξn ).8 Define a symplectic form on T ∗ U by ω=

n

dxi ∧ dξi .

i=1

One can check that this ω is intrinsically defined by considering the 1-form on T ∗ U , α=

n

ξi dxi ,

i=1

which satisfies ω = −dα and is coordinate-independent: in terms of the natural projection π : M → X, p = (x, ξ ) → x, the form α may be equivalently defined pointwise without coordinates by αp = (dπp )∗ ξ ∈ Tp∗ M, where (dπp )∗ : Tx∗ X → Tp∗ M is the transpose of dπp , that is, αp (v) = ξ((dπp )v) for v ∈ Tp M. Or yet, the form α is uniquely characterized by the property that µ∗ α = µ for every 1-form µ : X → T ∗ X (see Proposition 2.2). The 1-form α is the tautological form (or the Liouville 1-form) and the 2-form ω is the canonical symplectic form on T ∗ X. When referring to a cotangent bundle as a symplectic manifold, the symplectic structure is meant to be given by this canonical ω. Let X1 and X2 be n-dimensional manifolds with cotangent bundles M1 = T ∗ X1 and M2 = T ∗ X2 , and tautological 1-forms α1 and α2 . Suppose that f : X1 → X2 is a diffeomorphism. Then there is a natural diffeomorphism f! : M1 → M2 which lifts f ; namely, for p1 = (x1 , ξ1 ) ∈ M1 we define f! (p1 ) = p2 = (x2 , ξ2 ),

with

x2 = f (x1 ) ∈ X2 and ξ1 = (dfx1 )∗ ξ2 ∈ Tx∗1 X1 ,

where (dfx1 )∗ : Tx∗2 X2 −→ Tx∗1 X1 , so f! |Tx∗ is the inverse map of (dfx1 )∗ . 1

P ROPOSITION 1.3. The lift f! of a diffeomorphism f : X1 → X2 pulls the tautological form on T ∗ X2 back to the tautological form on T ∗ X1 , i.e., (f! )∗ α2 = α1 . 8 If an n-dimensional manifold X is described by coordinate charts (U, x , . . . , x ) with x : U → R, then, at n i 1 any x ∈ U , the differentials (dxi )x form a basis of Tx∗ X, inducing a map

T ∗ U −→ R2n , (x, ξ ) −→ (x1 , . . . , xn , ξ1 , . . . , ξn ), where ξ1 , . . . , ξn ∈ R are the corresponding coordinates of ξ ∈ Tx∗ X: ξ = ni=1 ξi (dxi )x . Then (T ∗ U, x1 , . . . , ∗ xn , ξ1 , . . . , ξn ) is a coordinate chart for the cotangent bundle T X; the coordinates x1 , . . . , xn , ξ1 , . . . , ξn are called the cotangent coordinates associated to the coordinates x1 , . . . , xn on U . One verifies that the transition functions on the overlaps are smooth, so T ∗ X is a 2n-dimensional manifold.

Symplectic geometry

87

P ROOF. At p1 = (x1 , ξ1 ) ∈ M1 , the claimed identity says (df! )∗p1 (α2 )p2 = (α1 )p1 , where p2 = f! (p1 ), that is, p2 = (x2 , ξ2 ) where x2 = f (x1 ) and (dfx1 )∗ ξ2 = ξ1 . This can be proved as follows: (df! )∗p1 (α2 )p2 = (df! )∗p1 (dπ2 )∗p2 ξ2 ∗ = d(π2 ◦ f! ) p ξ2 1 ∗ = d(f ◦ π1 ) p ξ2 1

=

(dπ1 )∗p1 (df )∗x1 ξ2

by definition of α2 by the chain rule because π2 ◦ f! = f ◦ π1 by the chain rule

= (dπ1 )∗p1 ξ1

by definition of f!

= (α1 )p1

by definition of α1 .

As a consequence of this naturality for the tautological form, a diffeomorphism of manifolds induces a canonical symplectomorphism of cotangent bundles: C OROLLARY 1.4. The lift f! : T ∗ X1 → T ∗ X2 of a diffeomorphism f : X1 → X2 is a symplectomorphism for the canonical symplectic forms, i.e., (f! )∗ ω2 = ω1 . In terms of the group (under composition) of diffeomorphisms Diff(X) of a manifold X, and the group of symplectomorphisms Sympl(T ∗ X, ω) of its cotangent bundle, we see that the injection Diff(X) → Sympl(T ∗ X, ω), f → f! is a group homomorphism. Clearly this is not surjective: for instance, consider the symplectomorphism T ∗ X → T ∗ X given by translation along cotangent fibers. E XAMPLE . Let X1 = X2 = S 1 . Then T ∗ S 1 is a cylinder S 1 × R. The canonical form is the area form ω = dθ ∧ dξ . If f : S 1 → S 1 is any diffeomorphism, then f! : S 1 × R → S 1 × R is a symplectomorphism, i.e., is an area-preserving diffeomorphism of the cylinder. Translation along the R direction is area-preserving but is not induced by a diffeomorphism of the base manifold S 1 . There is a criterion for which cotangent symplectomorphisms arise as lifts of diffeomorphisms in terms of the tautological form. First note the following feature of symplectic manifolds with exact symplectic forms. Let α be a 1-form on a manifold M such that ω = −dα is symplectic. There exists a unique vector field v whose interior product with ω is α, i.e., ıv ω = −α. If g : M → M is a symplectomorphism that preserves α (that is, g ∗ α = α), then g commutes with the flow9 of v, i.e., (exp tv) ◦ g = g ◦ (exp tv). When 9 For p ∈ M, (exp tv)(p) is the unique curve in M solving the initial value problem

d dt exp tv(p) = v exp tv(p) , (exp tv)(p)|t=0 = p for t in some neighborhood of 0. The one-parameter group of diffeomorphisms exp tv is called the flow of the vector field v.

88

A. Cannas da Silva

M = T ∗ X is the cotangent bundle of an arbitrary n-dimensional manifold X, and α is the ∂ tautological 1-form on M, the vector field v is just ξi ∂ξi with respect to a cotangent coordinate chart (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ). The flow exp tv, −∞ < t < ∞, satisfies (exp tv)(x, ξ ) = (x, et ξ ), for every (x, ξ ) in M. T HEOREM 1.5. A symplectomorphism g : T ∗ X → T ∗ X is a lift of a diffeomorphism f : X → X if and only if it preserves the tautological form: g ∗ α = α. P ROOF. By Proposition 1.3, a lift f! : T ∗ X → T ∗ X of a diffeomorphism f : X → X preserves the tautological form. Conversely, if g is a symplectomorphism of M that preserves α, then g preserves the cotangent fibration: by the observation above, g(x, ξ ) = (y, η) ⇒ g(x, λξ ) = (y, λη) for all (x, ξ ) ∈ M and λ > 0, and this must hold also for λ 0 by the differentiability of g at (x, 0). Therefore, there exists a diffeomorphism f : X → X such that π ◦g = f ◦π , where π : M → X is the projection map π(x, ξ ) = x, and g = f# . The canonical form is natural also in the following way. Given a smooth function h : X → R, the diffeomorphism τh of M = T ∗ X defined by τh (x, ξ ) = (x, ξ + dhx ) turns out to be always a symplectomorphism. Indeed, if π : M → X, π(x, ξ ) = x, is the projection, we have τh∗ α = α + π ∗ dh, so that τh∗ ω = ω. 1.4. Moser’s trick There are other relevant notions of equivalence for symplectic manifolds10 besides being symplectomorphic. Let M be a manifold with two symplectic forms ω0 , ω1 . D EFINITION 1.6. The symplectic manifolds (M, ω0 ) and (M, ω1 ) are strongly isotopic if there is an isotopy ρt : M → M such that ρ1∗ ω1 = ω0 . (M, ω0 ) and (M, ω1 ) are deformation-equivalent if there is a smooth family ωt of symplectic forms joining ω0 to ω1 . (M, ω0 ) and (M, ω1 ) are isotopic if they are deformation-equivalent and the de Rham cohomology class [ωt ] is independent of t. Hence, being strongly isotopic implies being symplectomorphic, and being isotopic implies being deformation-equivalent. We also have that being strongly isotopic implies being isotopic, because, if ρt : M → M is an isotopy such that ρ1∗ ω1 = ω0 , then ωt := ρt∗ ω1 is a smooth family of symplectic forms joining ω1 to ω0 and [ωt ] = [ω1 ], ∀t, by the homotopy invariance of de Rham cohomology. Moser [105] proved that, on a compact manifold, being isotopic implies being strongly isotopic (Theorem 1.7). McDuff showed that deformation-equivalence is indeed a necessary hypothesis: even if [ω0 ] = [ω1 ] ∈ H 2 (M; R), there are compact examples where (M, ω0 ) and (M, ω1 ) are not strongly isotopic; see Example 7.23 in [99]. In other words, 10 Understanding these notions and the normal forms requires tools, such as isotopies (by isotopy we mean a smooth one-parameter family of diffeomorphisms starting at the identity, like the flow of a vector field), Lie derivative, tubular neighborhoods and the homotopy formula in de Rham theory, covered in differential geometry or differential topology texts.

Symplectic geometry

89

fix c ∈ H 2 (M) and define Sc as the set of symplectic forms ω in M with [ω] = c. On a compact manifold, all symplectic forms in the same path-connected component of Sc are symplectomorphic according to the Moser theorem, though there might be symplectic forms in different components of Sc that are not symplectomorphic. T HEOREM 1.7 (Moser). Let M be a compact manifold with symplectic forms ω0 and ω1 . Suppose that ωt , 0 t 1, is a smooth family of symplectic forms joining ω0 to ω1 with cohomology class [ωt ] independent of t. Then there exists an isotopy ρ : M × R → M such that ρt∗ ωt = ω0 , 0 t 1. Moser applied an extremely useful argument, known as Moser’s trick, starting with the following observation. If there existed an isotopy ρ : M × R → M such that ρt∗ ωt = ω0 , 0 t 1, in terms of the associated time-dependent vector field vt :=

dρt ◦ ρt−1 , dt

t ∈ R,

we would then have for all 0 t 1 that 0=

d ∗ dωt (ρt ωt ) = ρt∗ Lvt ωt + dt dt

⇐⇒

Lvt ω t +

dωt = 0. dt

Conversely, the existence of a smooth time-dependent vector field vt , t ∈ R, satisfying the last equation is enough to produce by integration (since M is compact) the desired isotopy ρ : M × R → M satisfying ρt∗ ωt = ρ0∗ ω0 = ω0 , for all t. So everything boils down t to solving the equation Lvt ωt + dω dt = 0 for vt . d ωt ] = 0, there exists a smooth family of P ROOF. By the cohomology assumption that [ dt 1-forms µt such that

dωt = dµt , dt

0 t 1.

The argument involves the Poincaré lemma for compactly-supported forms, together with the Mayer–Vietoris sequence in order to use induction on the number of charts in a good cover of M; for a sketch, see page 95 in [99]. In the simplest case where ωt = (1 − t)ω0 + t tω1 with [ω0 ] = [ω1 ], we have that dω dt = ω1 − ω0 = dµ is exact. The nondegeneracy assumption on ωt , guarantees that we can pointwise solve the equation, known as Moser’s equation, ıvt ωt + µt = 0 to obtain a unique smooth family of vector fields vt , 0 t 1. Extend vt to all t ∈ R. Thanks to the compactness of M, the vector fields vt generate an isotopy ρ satisfying

90 dρt dt

A. Cannas da Silva

= vt ◦ ρt . Then we indeed have d ∗ dωt ∗ (ρ ωt ) = ρt Lvt ωt + = ρt∗ (dıvt ωt + dµt ) = ρt∗ d(ıvt ωt + µt ) = 0, dt t dt

where we used Cartan’s magic formula in Lvt ωt = dıvt ωt + ıvt dωt .

E XAMPLE . On a compact oriented 2-dimensional manifold M, a symplectic form is just an area form. Let ω0 and ω1 be two area forms on M. If [ω0 ] = [ω1 ], i.e., ω0 and ω1 give the same total area, then any convex combination of them is symplectic (because they induce the same orientation), and there is an isotopy ϕt : M → M, t ∈ [0, 1], such that ϕ1∗ ω0 = ω1 . Therefore, up to strong isotopy, there is a unique symplectic representative in each nonzero 2-cohomology class of M. On a noncompact manifold, given vt , we would need to check the existence for 0 t 1 t of an isotopy ρt solving the differential equation dρ dt = vt ◦ ρt . 1.5. Darboux and Moser theorems By a submanifold of a manifold M we mean either a manifold X with a closed embedding11 i : X → M, or an open submanifold (i.e., an open subset of M). Given a 2n-dimensional manifold M, a k-dimensional submanifold X, neighborhoods U0 , U1 of X, and symplectic forms ω0 , ω1 on U0 , U1 , we would like to know whether there exists a local symplectomorphism preserving X, i.e., a diffeomorphism ϕ : U0 → U1 with ϕ ∗ ω1 = ω0 and ϕ(X) = X. Moser’s Theorem 1.7 addresses the case where X = M. At the other extreme, when X is just one point, there is the classical Darboux theorem (Theorem 1.9). In general, we have: T HEOREM 1.8 (Moser theorem—relative version). Let ω0 and ω1 be symplectic forms on a manifold M, and X a compact submanifold of M. Suppose that the forms coincide, ω0 |p = ω1 |p , at all points p ∈ X. Then there exist neighborhoods U0 and U1 of X in M, and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ restricted to X is the identity map. P ROOF. Pick a tubular neighborhood U0 of X. The 2-form ω1 − ω0 is closed on U0 , and satisfies (ω1 − ω0 )p = 0 at all p ∈ X. By the homotopy formula on the tubular neighborhood, there exists a 1-form µ on U0 such that ω1 − ω0 = dµ and µp = 0 at all p ∈ X. Consider the family ωt = (1 − t)ω0 + tω1 = ω0 + t dµ of closed 2-forms on U0 . Shrinking U0 if necessary, we can assume that ωt is symplectic for t ∈ [0, 1], as nondegeneracy is an open property. Solve Moser’s equation, ıvt ωt = −µ, for vt By integration, shrinking U0 again if necessary, there exists a local isotopy ρ : U0 × [0, 1] → M with ρt∗ ωt = ω0 , for all t ∈ [0, 1]. Since vt |X = 0, we have ρt |X = idX . Set ϕ = ρ1 , U1 = ρ1 (U0 ). 11 A closed embedding is a proper injective immersion. A map is proper when its preimage of a compact set is always compact.

Symplectic geometry

91

T HEOREM 1.9 (Darboux). Let (M, ω) be a symplectic manifold, and let p be any point in M. Then we can find a chart (U, x1 , . . . , xn , y1 , . . . yn ) centered at p where ω=

n

dxi ∧ dyi .

i=1

Such a coordinate chart (U, x1 , . . . , xn , y1 , . . . , yn ) is called a Darboux chart, and the corresponding coordinates are called Darboux coordinates. The classical proof of Darboux’s theorem is by induction on the dimension of the manifold [2], in the spirit of the argument for a symplectic basis (Section 1.1). The proof below, using Moser’s theorem, was first provided by Weinstein [136]. P ROOF. Apply Moser’s relative theorem to X = {p}. More precisely, use any symplectic basis for (Tp M, ωp ) to construct coordinates(x1 , . . . , xn , y1 , . . . yn ) centered at p and dxi ∧ dyi |p . There are two symplectic valid on some neighborhood U , so that ω p = forms on U : the given ω0 = ω and ω1 = dxi ∧ dyi . By Theorem 1.8, there are neighborhoods U0 and U1 of p, and adiffeomorphismϕ : U0 → U1 such that ϕ(p) = p and ϕ ∗ ( dxi ∧ dyi ) = ω. Since ϕ ∗ ( dxi ∧ dyi ) = d(xi ◦ ϕ) ∧ d(yi ◦ ϕ), we simply set new coordinates xi = xi ◦ ϕ, yi = yi ◦ ϕ. Darboux’s theorem is easy in the 2-dimensional case. Being closed ω is locally exact, ω = dα. Every nonvanishing 1-form on a surface can be written locally as α = g dh for suitable functions g, h, where h is a coordinate on the local leaf space of the kernel foliation of α. The form ω = dg ∧ dh is nondegenerate if and only if (g, h) is a local diffeomorphism. By the way, transversality shows that the normal form for a generic12 2-form is x dx ∧ dy near a point where it is degenerate.

1.6. Symplectic submanifolds Moser’s argument permeates many other proofs, including those of the next two results regarding symplectic submanifolds. Let (M, ω) be a symplectic manifold. D EFINITION 1.10. A symplectic submanifold of (M, ω) is a submanifold X of M where, at each p ∈ X, the space Tp X is a symplectic subspace of (Tp M, ωp ). If i : X → M is the inclusion of a symplectic submanifold X, then the restriction of ω to X is a symplectic form, so that (X, i ∗ ω) is itself a symplectic manifold. Let X be a symplectic submanifold of (M, ω). At each p ∈ X, we have Tp M = Tp X ⊕ (Tp X)ωp (Section 1.1), so the map (Tp X)ωp → Tp M/Tp X is an isomorphism. This canonical identification of the normal space of X at p, Np X := Tp M/Tp X, with the symplectic orthogonal (Tp X)ωp , yields a canonical identification of the normal bundle N X 12 Generic here means that the subset of those 2-forms having this behavior is open, dense and invariant under diffeomorphisms of the manifold.

92

A. Cannas da Silva

with the symplectic vector bundle (T X)ω . A symplectic vector bundle is a vector bundle E → X equipped with a smooth13 field Ω of fiberwise nondegenerate skew-symmetric bilinear maps Ωp : Ep × Ep → R. The symplectic normal bundle is the normal bundle of a symplectic submanifold, with the symplectic structure induced by orthogonals. The next theorem, due to Weinstein [136], states that a neighborhood of a symplectic submanifold X is determined by X and (the isomorphism class of) its symplectic normal bundle. T HEOREM 1.11 (Symplectic neighborhood theorem). Let (M0 , ω0 ), (M1 , ω1 ) be symplectic manifolds with diffeomorphic compact symplectic submanifolds X0 , X1 . Let i0 : X0 → M0 , i1 : X1 → M1 be their inclusions. Suppose there is an isomorphism φ˜ : N X0 → NX1 of the corresponding symplectic normal bundles covering a symplectomorphism φ : (X0 , i0∗ ω0 ) → (X1 , i1∗ ω1 ). Then there exist neighborhoods U0 ⊂ M0 , U1 ⊂ M1 of X0 , X1 and a symplectomorphism ϕ : U0 → U1 extending φ such that the restriction of dϕ to ˜ the normal bundle N X0 is φ. As first noted by Thurston [131], the form Ω + π ∗ ωX is symplectic in some neighborhood of the zero section in N X, where π : N X → X is the bundle projection and ωX is the restriction of ω to X. Therefore, a compact symplectic submanifold X always admits a tubular neighborhood in the ambient (M, ω) symplectomorphic to a tubular neighborhood of the zero section in the symplectic normal bundle N X. P ROOF. By the Whitney extension theorem14 there exist neighborhoods U0 ⊂ M0 and U1 ⊂ M1 of X0 and X1 , and a diffeomorphism h : U0 → U1 such that h ◦ i0 = i1 ◦ φ and ˜ Hence ω0 and h∗ ω1 are two the restriction of dh to the normal bundle N X0 is the given φ. symplectic forms on U0 which coincide at all points p ∈ X0 . The result now follows from Moser’s relative theorem (Theorem 1.8). Carefully combining Moser’s argument with the existence of an ambient isotopy that produces a given deformation of a compact submanifold, we can show: T HEOREM 1.12. Let Xt , t ∈ [0, 1], be a (smooth) family of compact symplectic submanifolds of a compact symplectic manifold (M, ω). Then there exists an isotopy ρ : M × R → M such that for all t ∈ [0, 1] we have ρt∗ ω = ω and ρt (X0 ) = Xt . Inspired by complex geometry, Donaldson [32] proved the following theorem on the existence of symplectic submanifolds. A major consequence is the characterization of symplectic manifolds in terms of Lefschetz pencils; see Section 4.6. 13 Smoothness means that, for any pair of (smooth) sections u and v of E, the real-valued function Ω(u, v) : X → R given by evaluation at each point is smooth. 14 Whitney extension theorem. Let M be a manifold and X a submanifold of M. Suppose that at each p ∈ X we are given a linear isomorphism Lp : Tp M −→ Tp M such that Lp |Tp X = IdTp X and Lp depends smoothly on p. Then there exists an embedding h : N → M of some neighborhood N of X in M such that h|X = idX and dhp = Lp for all p ∈ X. A proof relies on a tubular neighborhood model.

Symplectic geometry

93

T HEOREM 1.13 (Donaldson). Let (M, ω) be a compact symplectic manifold. Assume that the cohomology class [ω] is integral, i.e., lies in H 2 (M; Z). Then, for every sufficiently large integer k, there exists a connected codimension-2 symplectic submanifold X representing the Poincaré dual of the integral cohomology class k[ω]. Under the same hypotheses, Auroux extended this result to show that given α ∈ H2m (M; Z) there exist positive k, ∈ Z such that kPD[ωn−m ] + α is realized by a 2mdimensional symplectic submanifold.

2. Lagrangian submanifolds 2.1. First Lagrangian submanifolds Let (M, ω) be a symplectic manifold. D EFINITION 2.1. A submanifold X of (M, ω) is Lagrangian (respectively, isotropic and coisotropic) if, at each p ∈ X, the space Tp X is a Lagrangian (respectively, isotropic and coisotropic) subspace of (Tp M, ωp ). If i : X → M is the inclusion map, then X is a Lagrangian submanifold if and only if i ∗ ω = 0 and dim X = 12 dim M. The problem of embedding15 a compact manifold as a Lagrangian submanifold of a given symplectic manifold is often global. For instance, Gromov [64] proved that there , and more generally can be no Lagrangian spheres in (Cn , ω0 ), except for the circle in C2 no compact exact Lagrangian submanifolds, in the sense that α0 = yj dxj restricts to an exact 1-form. The argument uses pseudoholomorphic curves (Section 3.6). Yet there are immersed Lagrangian spheres (Section 2.7). More recently were found topological and geometrical constraints on manifolds that admit Lagrangian embeddings into compact symplectic manifolds; see, for instance, [16,17,115]. E XAMPLES . 1. Any 1-dimensional submanifold of a symplectic surface is Lagrangian (because a 1-dimensional subspace of a symplectic vector space is always isotropic). Therefore, any product of n embedded curves arises as a Lagrangian submanifold of (a neighborhood of zero in) the prototype (R2n , ω0 ). In particular, a torus Tn = S 1 × · · · × S 1 can be embedded as a Lagrangian submanifold of any 2n-dimensional symplectic manifold, by Darboux’s theorem (Theorem 1.9). 2. Let M = T ∗ X be the cotangent bundle of a manifold X. With respect to a cotangent coordinate chart (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ), the tautological form is α = ξi dxi and the canonical form is ω = −dα = dxi ∧ dξi . The zero section X0 := {(x, ξ ) ∈ T ∗ X | ξ = 0 in Tx∗ X} is an n-dimensional submanifold of T ∗ X whose intersection with T ∗ U is given by the equations ξ1 = · · · = 15 An embedding is an immersion that is a homeomorphism onto its image.

94

A. Cannas da Silva

ξn = 0. Clearly α vanishes on X0 ∩ T ∗ U . Hence, if i0 : X0 → T ∗ X is the inclusion map, we have i0∗ ω = i0∗ dα = 0, and so X0 is Lagrangian. A cotangent fiber Tx∗0 X is an n-dimensional submanifold of T ∗ X given by the equations xi = (x0 )i , i = 1, . . . , n, on T ∗ U . Since the xi ’s are constant, the form α vanishes identically, and Tx∗0 X is a Lagrangian submanifold. Let Xµ be (the image of) an arbitrary section, that is, an n-dimensional submanifold of T ∗ X of the form Xµ = {(x, µx ) | x ∈ X, µx ∈ Tx∗ X}, where the covector µx depends smoothly on x, so µ : X → T ∗ X is a de Rham 1-form. We will investigate when such an Xµ is Lagrangian. Relative to the inclusion i : Xµ → T ∗ X and the cotangent projection π : T ∗ X → X, these Xµ ’s are exactly the submanifolds for which π ◦ i : Xµ → X is a diffeomorphism. P ROPOSITION 2.2. The tautological 1-form α on T ∗ X satisfies µ∗ α = µ, for any 1-form µ : X → T ∗ X. P ROOF. Denote by sµ : X → T ∗ X, x → (x, µx ), the 1-form µ regarded exclusively as a map. From the definition, αp = (dπp )∗ ξ at p = (x, ξ ) ∈ M. For p = sµ (x) = (x, µx ), we have αp = (dπp )∗ µx . Then, since π ◦ sµ = idX , we have ∗ ∗ sµ α x = (dsµ )∗x αp = (dsµ )∗x (dπp )∗ µx = d(π ◦ sµ ) x µx = µx .

The map sµ : X → T ∗ X, sµ (x) = (x, µx ) is an embedding with image the section Xµ . The diffeomorphism τ : X → Xµ , τ (x) := (x, µx ), satisfies i ◦ τ = sµ . P ROPOSITION 2.3. The sections of T ∗ X that are Lagrangian are those corresponding to closed 1-forms on X. P ROOF. Using the previous notation, the condition of Xµ being Lagrangian becomes: i ∗ dα = 0 ⇔ τ ∗ i ∗ dα = 0 ⇔ sµ∗ dα = 0 ⇔ d(sµ∗ α) = 0 ⇔ dµ = 0. When µ = dh for some h ∈ C ∞ (X), such a primitive h is called a generating function for the Lagrangian submanifold Xµ . Two functions generate the same Lagrangian submanifold if and only if they differ by a locally constant function. When X is simply connected, 1 (X) = 0, every Lagrangian Xµ admits a generating function. or at least HdeRham Besides the cotangent fibers, there are lots of Lagrangian submanifolds of T ∗ X not covered by the description in terms of closed 1-forms. Let S be any submanifold of an n-dimensional manifold X. The conormal space of S at x ∈ S is Nx∗ S = ξ ∈ Tx∗ X | ξ(v) = 0 for all v ∈ Tx S . The conormal bundle of S is N ∗ S = {(x, ξ ) ∈ T ∗ X | x ∈ S, ξ ∈ Nx∗ S}. This is an n-dimensional submanifold of T ∗ X. In particular, taking S = {x} to be one point, the conormal bundle is the corresponding cotangent fiber Tx∗ X. Taking S = X, the conormal bundle is the zero section X0 of T ∗ X.

Symplectic geometry

95

P ROPOSITION 2.4. If i : N ∗ S → T ∗ X is the inclusion of the conormal bundle of a submanifold S ⊂ X, and α is the tautological 1-form on T ∗ X, then i ∗ α = 0. P ROOF. Let (U, x1 , . . . , xn ) be a coordinate chart on X adapted to S, so that U ∩ S is described by xk+1 = · · · = xn = 0. Let (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ) be the associated cotangent coordinate chart. The submanifold N ∗ S ∩ T ∗ U is described by xk+1 = · · · = xn = 0 and ξ1 = · · · = ξk = 0. Since α = ξi dxi on T ∗ U , we conclude that, at p ∈ N ∗ S, ∗

(i α)p = αp |Tp (N ∗ S) =

ξi dxi

i>k

span{ ∂x∂ , ik} i

= 0.

C OROLLARY 2.5. For any submanifold S of X, the conormal bundle N ∗ S is a Lagrangian submanifold of T ∗ X.

2.2. Lagrangian neighborhood theorem Weinstein [136] proved that, if a compact submanifold X is Lagrangian with respect to two symplectic forms ω0 and ω1 , then the conclusion of the Moser relative theorem (Theorem 1.8) still holds. We need some algebra for the Weinstein theorem. Suppose that U, W are n-dimensional vector spaces, and Ω : U × W → R is a bilinear ˜ = Ω(u, ·). Then Ω is pairing; the map Ω gives rise to a linear map Ω˜ : U → W ∗ , Ω(u) ˜ nondegenerate if and only if Ω is bijective. P ROPOSITION 2.6. Let (V , Ω) be a symplectic vector space, U a Lagrangian subspace of (V , Ω), and W any vector space complement to U , not necessarily Lagrangian. Then from W we can canonically build a Lagrangian complement to U . P ROOF. From Ω we get a nondegenerate pairing Ω : U × W → R, so Ω˜ : U → W ∗ is bijective. We look for a Lagrangian complement to U of the form W = {w + Aw | w ∈ W } for some linear map A : W → U . For W to be Lagrangian we need that Ω(w1 , w2 ) = Ω˜ (Aw2 )(w1 ) − Ω˜ (Aw1 )(w2 ). Let A = Ω˜ ◦ A, and look for A such that Ω(w1 , w2 ) = A (w2 )(w1 ) − A (w1 )(w2 ) for all w1 , w2 ∈ W . The canonical choice is A (w) = − 12 Ω(w, ·). Set A = (Ω˜ )−1 ◦ A . P ROPOSITION 2.7. Let V be a vector space, let Ω0 and Ω1 be symplectic forms on V , let U be a subspace of V Lagrangian for Ω0 and Ω1 , and let W be any complement to U in V . Then from W we can canonically construct a linear isomorphism L : V −→ V such ∗ that L|U = IdU and L Ω1 = Ω0 . P ROOF. By Proposition 2.6, from W we canonically obtain complements W0 and W1 to U in V such that W0 is Lagrangian for Ω0 and W1 is Lagrangian for Ω1 . The nondegenerate bilinear pairings Ωi : Wi × U → R, i = 0, 1, give isomorphisms Ω˜ i : Wi −→ U ∗ , i = 0, 1, respectively. Let B : W0 → W1 be the linear map satisfying Ω˜ 1 ◦ B = Ω˜ 0 , i.e., Ω0 (w0 , u) =

96

A. Cannas da Silva

Ω1 (Bw0 , u), ∀w0 ∈ W0 , ∀u ∈ U . Let L := IdU ⊕ B : U ⊕ W0 → U ⊕ W1 be the extension of B to the rest of V by setting it to be the identity on U . It satisfies: (L∗ Ω1 )(u ⊕ w0 , u ⊕ w0 ) = Ω1 (u ⊕ Bw0 , u ⊕ Bw0 ) = Ω1 (u, Bw0 ) + Ω1 (Bw0 , u ) = Ω0 (u, w0 ) + Ω0 (w0 , u ) = Ω0 (u ⊕ w0 , u ⊕ w0 ).

T HEOREM 2.8 (Weinstein Lagrangian neighborhood theorem). Let M be a 2n-dimensional manifold, X a compact n-dimensional submanifold, i : X → M the inclusion map, and ω0 and ω1 symplectic forms on M such that i ∗ ω0 = i ∗ ω1 = 0, i.e., X is a Lagrangian submanifold of both (M, ω0 ) and (M, ω1 ). Then there exist neighborhoods U0 and U1 of X in M and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ is the identity on X, i.e., ϕ(p) = p, ∀p ∈ X. P ROOF. Put a Riemannian metric g on M. Fix p ∈ X, and let V = Tp M, U = Tp X and W = U ⊥ , the orthocomplement of U in V relative to the inner product gp (·, ·). Since i ∗ ω0 = i ∗ ω1 = 0, the subspace U is Lagrangian for both (V , ω0 |p ) and (V , ω1 |p ). By Proposition 2.7, we canonically get from U ⊥ a linear isomorphism Lp : Tp M → Tp M depending smoothly on p, such that Lp |Tp X = IdTp X and L∗p ω1 |p = ω0 |p . By the Whitney extension theorem (Section 1.5), there exist a neighborhood N of X and an embedding h : N → M with h|X = idX and dhp = Lp for p ∈ X. Hence, at any p ∈ X, we have (h∗ ω1 )p = (dhp )∗ ω1 |p = L∗p ω1 |p = ω0 |p . Applying the Moser relative theorem (Theorem 1.8) to ω0 and h∗ ω1 , we find a neighborhood U0 of X and an embedding f : U0 → N such that f |X = idX and f ∗ (h∗ ω1 ) = ω0 on Uo . Set ϕ = h ◦ f and U1 = ϕ(U0 ). Theorem 2.8 has the following generalization. For a proof see, for instance, either of [61,70,139]. T HEOREM 2.9 (Coisotropic embedding theorem). Let M be a manifold of dimension 2n, X a submanifold of dimension k n, i : X → M the inclusion, and ω0 and ω1 symplectic forms on M, such that i ∗ ω0 = i ∗ ω1 and X is coisotropic for both (M, ω0 ) and (M, ω1 ). Then there exist neighborhoods U0 and U1 of X in M and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ|X = idX .

2.3. Weinstein tubular neighborhood theorem Let (V , Ω) be a symplectic linear space, and let U be a Lagrangian subspace. Then there is a canonical nondegenerate bilinear pairing Ω : V /U × U → R defined by Ω ([v], u) = Ω(v, u) where [v] is the equivalence class of v in V /U . Consequently, we get a canonical isomorphism Ω˜ : V /U → U ∗ , Ω˜ ([v]) = Ω ([v], ·).

Symplectic geometry

97

In particular, if (M, ω) is a symplectic manifold, and X is a Lagrangian submanifold, then Tp X is a Lagrangian subspace of (Tp M, ωp ) for each p ∈ X and there is a canonical identification of the normal space of X at p, Np X := Tp M/Tp X, with the cotangent fiber Tp∗ X. Consequently the normal bundle N X and the cotangent bundle T ∗ X are canonically identified. T HEOREM 2.10 (Weinstein tubular neighborhood theorem). Let (M, ω) be a symplectic manifold, X a compact Lagrangian submanifold, ω0 the canonical symplectic form on T ∗ X, i0 : X → T ∗ X the Lagrangian embedding as the zero section, and i : X → M the Lagrangian embedding given by inclusion. Then there are neighborhoods U0 of X in T ∗ X, U of X in M, and a diffeomorphism ϕ : U0 → U such that ϕ ∗ ω = ω0 and ϕ ◦ i0 = i. P ROOF. By the standard tubular neighborhood theorem16 and since N X T ∗ X are canonically identified, we can find a neighborhood N0 of X in T ∗ X, a neighborhood N of X in M, and a diffeomorphism ψ : N0 → N such that ψ ◦ i0 = i. Let ω0 be the canonical form on T ∗ X and ω1 = ψ ∗ ω. The submanifold X is Lagrangian for both of these symplectic forms on N0 . By the Weinstein Lagrangian neighborhood theorem (Theorem 2.8), there exist neighborhoods U0 and U1 of X in N0 and a diffeomorphism θ : U0 → U1 such that θ ∗ ω1 = ω0 and θ ◦ i0 = i0 . Take ϕ = ψ ◦ θ and U = ϕ(U0 ). Then ϕ ∗ ω = θ ∗ ψ ∗ ω = θ ∗ ω1 = ω0 . Theorem 2.10 classifies compact Lagrangian embeddings: up to local symplectomorphism, the set of Lagrangian embeddings is the set of embeddings of manifolds into their cotangent bundles as zero sections. The classification of compact isotropic embeddings is also due to Weinstein in [137, 139]. An isotropic embedding of a manifold X into a symplectic manifold (M, ω) is a closed embedding i : X → M such that i ∗ ω = 0. Weinstein showed that neighborhood equivalence of isotropic embeddings is in one-to-one correspondence with isomorphism classes of symplectic vector bundles. The classification of compact coisotropic embeddings is due to Gotay [61]. A coisotropic embedding of a manifold X carrying a closed 2-form α of constant rank into a symplectic manifold (M, ω) is an embedding i : X → M such that i ∗ ω = α and i(X) is coisotropic as a submanifold of M. Let E be the characteristic distribution of a closed form α of constant rank on X, i.e., Ep is the kernel of αp at p ∈ X. Gotay showed that then the total space E ∗ carries a symplectic structure in a neighborhood of the zero section, such that X embeds coisotropically onto this zero section and, moreover, every coisotropic embedding is equivalent to this in some neighborhood of the zero section. 16 Tubular neighborhood theorem. Let M be a manifold, X a submanifold, N X the normal bundle of X in M,

i0 : X → N X the zero section, and i : X → M the inclusion. Then there are neighborhoods U0 of X in N X, U of X in M and a diffeomorphism ψ : U0 → U such that ψ ◦ i0 = i. This theorem can be proved with the exponential map using a Riemannian metric; see, for instance, [120].

98

A. Cannas da Silva

2.4. Application to symplectomorphisms Let (M1 , ω1 ) and (M2 , ω2 ) be two 2n-dimensional symplectic manifolds. Given a diffeo morphism f : M1 −→ M2 , there is a way to express the condition of f being a symplectomorphism in terms of a certain submanifold being Lagrangian. Consider the two projection maps pri : M1 × M2 → Mi , (p1 , p2 ) → pi , i = 1, 2. The twisted product form on M1 × M2 is the symplectic17 form ω˜ = (pr1 )∗ ω1 − (pr2 )∗ ω2 .

P ROPOSITION 2.11. A diffeomorphism f : M1 −→ M2 is a symplectomorphism if and only if the graph of f is a Lagrangian submanifold of (M1 × M2 , ω). ˜ P ROOF. The graph of f is the 2n-dimensional submanifold Graph f = {(p, f (p)) | p ∈ M1 } ⊆ M1 × M2 , which is the image of the embedding γ : M1 → M1 × M2 , p → (p, f (p)). We have γ ∗ ω˜ = γ ∗ pr∗1 ω1 − γ ∗ pr∗2 ω2 = (pr1 ◦ γ )∗ ω1 − (pr2 ◦ γ )∗ ω2 , and pr1 ◦ γ is the identity map on M1 whereas pr2 ◦ γ = f . So Graph f is Lagrangian, i.e., γ ∗ ω˜ = 0, if and only if f ∗ ω2 = ω1 , i.e., f is a symplectomorphism. Lagrangian submanifolds of (M1 × M2 , ω) ˜ are called canonical relations, when viewed as morphisms between (M1 , ω1 ) and (M2 , ω2 ), even if dim M1 = dim M2 . Under a reasonable assumption, there is a notion of composition [137]. Take M1 = M2 = M and suppose that (M, ω) is a compact symplectic manifold and f ∈ Sympl(M, ω). The graphs Graph f and , of f and of the identity map id : M → M, are Lagrangian submanifolds of M × M with ω˜ = pr∗1 ω − pr∗2 ω. By the Weinstein tubular neighborhood theorem, there exist a neighborhood U of in (M × M, ω) ˜ and a neighborhood U0 of M in (T ∗ M, ω0 ) with a symplectomorphism ϕ : U → U0 satisfying ϕ(p, p) = (p, 0), ∀p ∈ M. Suppose that f is sufficiently C 1 -close18 to id, i.e., f is in some sufficiently small neighborhood of the identity id in the C 1 -topology. Hence we can assume that Graph f ⊆ U . Let j : M → U , j (p) = (p, f (p)), be the embedding as Graph f , and i : M → U , i(p) = (p, p), be the embedding as = Graph id. The map j is sufficiently C 1 -close to i. These maps induce embeddings ϕ ◦ j = j0 : M → U0 and ϕ ◦ i = i0 : M → U0 as 0-section, respectively. Since the map j0 is sufficiently C 1 -close to i0 , the image set j0 (M) intersects each fiber Tp∗ M at one point µp depending smoothly on p. Therefore, the image of j0 is the image of a smooth section µ : M → T ∗ M, that is, a 1-form µ = j0 ◦ (π ◦ j0 )−1 . We conclude that Graph f {(p, µp ) | p ∈ M, µp ∈ Tp∗ M}. Conversely, if µ is a 1-form sufficiently C 1 -close to the zero 1-form, then {(p, µp ) | p ∈ M, µp ∈ Tp∗ M} Graph f , for some diffeomorphism f : M → M. 17 More generally, λ (pr )∗ ω + λ (pr )∗ ω is symplectic for all λ , λ ∈ R \ {0}. 1 1 1 2 2 2 1 2 18 Let X and Y be manifolds. A sequence of maps f : X → Y converges in the C 0 -topology (a.k.a. the i

compact-open topology) to f : X → Y if and only if fi converges uniformly on compact sets. A sequence of C 1 maps fi : X → Y converges in the C 1 -topology to f : X → Y if and only if it and the sequence of derivatives dfi : T X → T Y converge uniformly on compact sets.

Symplectic geometry

99

By Proposition 2.3, Graph f is Lagrangian if and only if µ is closed. A small C 1 neighborhood of id in Sympl(M, ω) is thus homeomorphic to a C 1 -neighborhood of zero in the vector space of closed 1-forms on M. So we obtain the model: Tid Sympl(M, ω) µ ∈ Ω 1 (M) | dµ = 0 . In particular, Tid (Sympl(M, ω)) contains the space of exact 1-forms that correspond to generating functions, C ∞ (M)/{locally constant functions}. T HEOREM 2.12. Let (M, ω) be a compact symplectic manifold (and not just one point) 1 (M) = 0. Then any symplectomorphism of M that is sufficiently C 1 -close to with HdeRham the identity has at least two fixed points. P ROOF. If f ∈ Sympl(M, ω) is sufficiently C 1 -close to id, then its graph corresponds to a 1 (M) = 0, we have that µ = dh for some h ∈ C ∞ (M). closed 1-form µ on M. As HdeRham But h must have at least two critical points because M is compact. A point p where µp = dhp = 0 corresponds to a point in the intersection of the graph of f with the diagonal, that is, a fixed point of f . This result has the following analogue in terms of Lagrangian intersections: if X is a 1 (X) = 0, compact Lagrangian submanifold of a symplectic manifold (M, ω) with HdeRham 1 19 then every Lagrangian submanifold of M that is C -close to X intersects X in at least two points.

2.5. Generating functions We focus on symplectomorphisms between the cotangent bundles M1 = T ∗ X1 , M2 = T ∗ X2 of two n-dimensional manifolds X1 , X2 . Let α1 , α2 and ω1 , ω2 be the corresponding tautological and canonical forms. Under the natural identification M1 × M2 = T ∗ X1 × T ∗ X2 T ∗ (X1 × X2 ), the tautological 1-form on T ∗ (X1 × X2 ) is α = pr∗1 α1 + pr∗2 α2 , the canonical 2-form on T ∗ (X1 × X2 ) is ω = −dα = pr∗1 ω1 + pr∗2 ω2 , and the twisted product form is ω˜ = pr∗1 ω1 − pr∗2 ω2 . We define the involution σ2 : M2 → M2 , (x2 , ξ2 ) → (x2 , −ξ2 ), which yields σ2∗ α2 = −α2 . Let σ = idM1 × σ2 : M1 × M2 → M1 × M2 . Then σ ∗ ω˜ = pr∗1 ω1 + pr∗2 ω2 = ω. If L is a Lagrangian submanifold of (M1 × M2 , ω), then its twist Lσ := σ (L) is a Lagrangian ˜ submanifold of (M1 × M2 , ω). For producing a symplectomorphism M1 = T ∗ X1 → M2 = T ∗ X2 we can start with a Lagrangian submanifold L of (M1 × M2 , ω), twist it to obtain a Lagrangian submanifold ˜ and, if Lσ happens to be the graph of some diffeomorphism ϕ : Lσ of (M1 × M2 , ω), M1 → M2 , then ϕ is a symplectomorphism. 19 We say that a submanifold Y of M is C 1 -close to another submanifold X when there is a diffeomorphism X → Y that is, as a map into M, C 1 -close to the inclusion X → M.

100

A. Cannas da Silva

A method to obtain Lagrangian submanifolds of M1 × M2 T ∗ (X1 × X2 ) relies on generating functions. For any f ∈ C ∞ (X1 × X2 ), df is a closed 1-form on X1 × X2 . The Lagrangian submanifold generated by f is Lf := {((x, y), (df )(x,y) ) | (x, y) ∈ X1 × X2 } (cf. Section 2.1). We adopt the loose notation dx f := dx f (x, y) := (df )(x,y) projected to Tx∗ X1 × {0}, dy f := dy f (x, y) := (df )(x,y) projected to {0} × Ty∗ X2 , which enables us to write Lf = {(x, y, dx f, dy f ) | (x, y) ∈ X1 × X2 } and Lσf = (x, y, dx f, −dy f ) | (x, y) ∈ X1 × X2 . When Lσf is in fact the graph of a diffeomorphism ϕ : M1 = T ∗ X1 → M2 = T ∗ X2 , we call ϕ the symplectomorphism generated by f , and call f the generating function of ϕ. The issue now is to determine whether a given Lσf is the graph of a diffeomorphism ϕ : M1 → M2 . Let (U1 , x1 , . . . , xn ), (U2 , y1 , . . . , yn ) be coordinate charts for X1 , X2 , with associated charts (T ∗ U1 , x1 , . . . , xn , ξ1 , . . . , ξn ), (T ∗ U2 , y1 , . . . , yn , η1 , . . . , ηn ) for M1 , M2 . The set Lσf is the graph of ϕ : M1 → M2 exactly when, for any (x, ξ ) ∈ M1 and (y, η) ∈ M2 , we have ϕ(x, ξ ) = (y, η) ⇔ ξ = dx f and η = −dy f . Therefore, given a point (x, ξ ) ∈ M1 , to find its image (y, η) = ϕ(x, ξ ) we must solve the Hamilton look-alike equations ξi = ∂f (x, y), ∂xi ηi = − ∂f (x, y). ∂yi If there is a solution y = ϕ1 (x, ξ ) of the first equation, we may feed it to the second thus obtaining η = ϕ2 (x, ξ ), so that ϕ(x, ξ ) = (ϕ1 (x, ξ ), ϕ2 (x, ξ )). By the implicit function theorem, in order to solve the first equation locally and smoothly for y in terms of x and ξ , we need the condition n ∂f ∂ det = 0. ∂yj ∂xi i,j =1 This is a necessary condition for f to generate a symplectomorphism ϕ. Locally this is also sufficient, but globally there is the usual bijectivity issue. E XAMPLE . Let X1 = X2 = Rn , and f (x, y) = − |x−y| 2 , the square of Euclidean distance up to a constant. In this case, the Hamilton equations are ξi = ∂f = yi − xi , ∂xi y i = x i + ξi , ⇐⇒ ∂f η ηi = − i = ξi . = yi − xi , 2

∂yi

The symplectomorphism generated by f is ϕ(x, ξ ) = (x + ξ, ξ ). If we use the Euclidean inner product to identify T ∗ Rn with T Rn , and hence regard ϕ as ϕ˜ : T Rn → T Rn

Symplectic geometry

101

and interpret ξ as the velocity vector, then the symplectomorphism ϕ corresponds to free translational motion in Euclidean space. The previous example can be generalized to the geodesic flow on a Riemannian manifold.20 Let (X, g) be a geodesically convex Riemannian manifold, where d(x, y) is the Riemannian distance between points x and y. Consider the function f : X × X −→ R,

f (x, y) = −

d(x, y)2 . 2

We want to investigate if f generates a symplectomorphism ϕ : T ∗ X → T ∗ X. Using the identification g˜ x : Tx X −→ Tx∗ X, v → gx (v, ·), induced by the metric, we translate ϕ into a map ϕ˜ : T X → T X. We need to solve

g˜ x (v) = ξ = dx f (x, y), g˜ y (w) = η = −dy f (x, y)

(1)

for (y, η) in terms of (x, ξ ) in order to find ϕ, or, equivalently, for (y, w) in terms (x, v) in order to find ϕ. ˜ Assume that (X, g) is geodesically complete, that is, every geodesic can be extended indefinitely. P ROPOSITION 2.13. Under the identification Tx X Tx∗ X given by the metric, the symplectomorphism generated by f corresponds to the map ϕ˜ : T X −→ T X, dγ (x, v) −→ γ (1), (1) , dt where γ is the geodesic with initial conditions γ (0) = x and

dγ dt

(0) = v.

20 A Riemannian metric on a manifold X is a smooth function g that assigns to each point x ∈ X an inner product gx on Tx X, that is, a symmetric positive-definite bilinear map gx : Tx X × Tx X → R. Smoothness means that for every (smooth) vector field v : X → T X the real-valued function x → gx (vx , vx ) is smooth. A Riemannian manifold is a pair (X, g) where g is a Riemannian metric on the manifold X. The arc-length of a piecewise

dγ dγ smooth curve γ : [a, b] → X on a Riemannian (X, g) is ab dγ dt dt , where dt (t) = dγt (1) ∈ Tγ (t) X and dt = dγ gγ (t) ( dγ dt , dt ) is the velocity of γ . A reparametrization of a curve γ : [a, b] → X is a curve of the form γ ◦ τ : [c, d] → X for some τ : [c, d] → [a, b]. By the change of variable formula for the integral, we see that the arc-length of γ is invariant by reparametrization. The Riemannian distance between two points x and y of a connected Riemannian manifold (X, g) is the infimum d(x, y) of the set of all arc-lengths for piecewise smooth curves joining x to y. A geodesic is a curve that locally minimizes distance and whose velocity is constant. Given any curve γ : [a, b] → X with dγ dt never vanishing, there is a reparametrization γ ◦ τ : [a, b] → X of constant velocity. A minimizing geodesic from x to y is a geodesic joining x to y whose arc-length is the Riemannian distance d(x, y). A Riemannian manifold (X, g) is geodesically convex if every point x is joined to every other point y by a unique (up to reparametrization) minimizing geodesic. For instance, (Rn , ·, ·) is a geodesically convex Riemannian manifold (where gx (v, w) = v, w is the Euclidean inner product on T Rn Rn × Rn ), for which the Riemannian distance is the usual Euclidean distance d(x, y) = |x − y|.

102

A. Cannas da Silva

This map ϕ˜ is called the geodesic flow on (X, g). P ROOF. Given (x, v) ∈ T X, let exp(x, v) : R → X be the unique geodesic with initial con(0) = v. In this notation, we need to show that the ditions exp(x, v)(0) = x and d exp(x,v) dt unique solution of the system of equations (1) is ϕ(x, ˜ v) = (exp(x, v)(1), d exp(x,v) (1)). dt The Gauss lemma in Riemannian geometry (see, for instance, [120]) asserts that geodesics are orthogonal to the level sets of the distance function. To solve the first equation for y = exp(x, u)(1) for some u ∈ Tx X, evaluate both sides at v and at vectors v ∈ Tx X orthogonal to v, |v|2 =

d −d(exp(x, v)(t), y)2 d −d(exp(x, v )(t), y)2 and 0 = dt 2 dt 2 t=0 t=0

to conclude that u = v, and thus y = exp(x, v)(1). We have −dy f (x, y)(w ) = 0 at vectors w ∈ Ty X orthogonal to W := d exp(x,v) (1), dt because f (x, y) is essentially the arc-length of a minimizing geodesic. Hence w = kW must be proportional to W , and k = 1 since d −d(x, exp(x, v)(1 − t))2 = |v|2 . k|v| = gy (kW, W ) = − dt 2 t=0 2

2.6. Fixed points Let X be an n-dimensional manifold, and M = T ∗ X its cotangent bundle equipped with the canonical symplectic form ω. Let f : X × X → R be a smooth function generating a symplectomorphism ϕ : M → M, ϕ(x, dx f ) = (y, −dy f ), with the notation of Section 2.5. To describe the fixed points of ϕ, we introduce the function ψ : X → R, ψ(x) = f (x, x). P ROPOSITION 2.14. There is a one-to-one correspondence between the fixed points of the symplectomorphism ϕ and the critical points of ψ. P ROOF. At x0 ∈ X, dx0 ψ = (dx f + dy f )|(x,y)=(x0 ,x0 ) . Let ξ = dx f |(x,y)=(x0 ,x0 ) . Recalling that Lσf is the graph of ϕ, we have that x0 is a critical point of ψ, i.e., dx0 ψ = 0, if and only if dy f |(x,y)=(x0 ,x0 ) = −ξ , which happens if and only if the point in Lσf corresponding to (x, y) = (x0 , x0 ) is (x0 , x0 , ξ, ξ ), i.e., ϕ(x0 , ξ ) = (x0 , ξ ) is a fixed point. Consider the iterates ϕ N = ϕ ◦ ϕ ◦ · · · ◦ ϕ, N = 1, 2, . . . , given by N successive applications of ϕ. According to the previous proposition, if the symplectomorphism ϕ N : M → M is generated by some function f (N ) , then there is a one-to-one correspondence between the set of fixed points of ϕ N and the set of critical points of ψ (N ) : X → R, ψ (N ) (x) = f (N ) (x, x). It remains to know whether ϕ N admits a generating function. We will see that to a certain extent it does. For each pair x, y ∈ X, define a map X → R, z → f (x, z) + f (z, y). Suppose that this map has a unique critical point z0 and that z0 is nondegenerate. As z0 is determined for

Symplectic geometry

103

each (x, y) implicitly by the equation dy f (x, z0 ) + dx f (z0 , y) = 0, by nondegeneracy, the implicit function theorem assures that z0 = z0 (x, y) is a smooth function. Hence, the function f (2) : X × X −→ R,

f (2) (x, y) := f (x, z0 ) + f (z0 , y)

is smooth. Since ϕ is generated by f , and z0 is critical, we have ϕ 2 x, dx f (2) (x, y) = ϕ ϕ(x, dx f (x, z0 ) = ϕ z0 , −dy f (x, z0 ) = ϕ z0 , dx f (z0 , y) = y, −dy f (z0 , y) = y, −dy f (2) (x, y) . We conclude that the function f (2) is a generating function for ϕ 2 , as long as, for each ξ ∈ Tx∗ X, there is a unique y ∈ X for which dx f (2) (x, y) equals ξ . There are similar partial recipes for generating functions of higher iterates. In the case of ϕ 3 , suppose that the function X × X → R, (z, u) → f (x, z) + f (z, u) + f (u, y), has a unique critical point (z0 , u0 ) and that it is a nondegenerate critical point. A generating function would be f (3) (x, y) = f (x, z0 ) + f (z0 , u0 ) + f (u0 , y). When the generating functions f, f (2) , f (3) , . . . , f (N ) exist given by these formulas, the N -periodic points of ϕ, i.e., the fixed points of ϕ N , are in one-to-one correspondence with the critical points of (x1 , . . . , xN ) −→ f (x1 , x2 ) + f (x2 , x3 ) + · · · + f (xN −1 , xN ) + f (xN , x1 ). E XAMPLE . Let χ : R → R2 be a smooth plane curve that is 1-periodic, i.e., χ(s + 1) = χ(s), and parametrized by arc-length, i.e., | dχ ds | = 1. Assume that the region Y enclosed by the image of χ is convex, i.e., for any s ∈ R, the tangent line {χ(s) + t dχ ds | t ∈ R} intersects the image X := ∂Y of χ only at the point χ(s). Suppose that a ball is thrown into a billiard table of shape Y rolling with constant velocity and bouncing off the boundary subject to the usual law of reflection. The map describing successive points on the orbit of the ball is ϕ : R/Z × (−1, 1) −→ R/Z × (−1, 1), (x, v) −→ (y, w), saying that when the ball bounces off χ(x) with angle θ = arccos v, it will next collide with χ(y) and bounce off with angle ν = arccos w. Then the function f : R/Z × R/Z → R defined by f (x, y) = −|χ(x) − χ(y)| is smooth off the diagonal, and for ϕ(x, v) = (y, w) satisfies

∂f (x, y) = χ(y)−χ(x) · dχ

∂x |χ(x)−χ(y)| ds s=x = cos θ = v,

∂f (x, y) = χ(x)−χ(y) · dχ

∂y |χ(x)−χ(y)| ds s=y = − cos ν = −w.

104

A. Cannas da Silva

We conclude that f is a generating function for ϕ. Similar approaches work for higherdimensional billiard problems. Periodic points are obtained by finding critical points of real functions of N variables in X,

(x1 , . . . , xN ) −→ χ(x1 ) − χ(x2 ) + · · · + χ(xN −1 ) − χ(xN )

+ χ(xN ) − χ(x1 ) , that is, by finding the N -sided (generalized) polygons inscribed in X of critical perimeter. Notice that R/Z × (−1, 1) {(x, v) | x ∈ X, v ∈ Tx X, |v| < 1} is the open unit tangent ball bundle of a circle X, which is an open annulus A, and the map ϕ : A → A is areapreserving, as in the next two theorems. While studying Poincaré return maps in dynamical systems, Poincaré arrived at the following results. T HEOREM 2.15 (Poincaré recurrence theorem). Let ϕ : A → A be a volume-preserving diffeomorphism of a finite-volume manifold A, and U a nonempty open set in A. Then there is q ∈ U and a positive integer N such that ϕ N (q) ∈ U . Hence, under iteration, a mechanical system governed by ϕ will eventually return arbitrarily close to the initial state. P ROOF. Let U0 = U , U1 = ϕ(U), U2 = ϕ 2 (U), . . . . If all of these sets were disjoint, then, since Volume(Ui ) = Volume(U) > 0 for all i, the volume of A would be greater or equal to i Volume(Ui ) = ∞. To avoid this contradiction we must have ϕ k (U) ∩ ϕ (U) = ∅ for some k > , which implies ϕ k− (U) ∩ U = ∅. T HEOREM 2.16 (Poincaré’s last geometric theorem). Suppose that ϕ : A → A is an areapreserving diffeomorphism of the closed annulus A = R/Z × [−1, 1] that preserves the two components of the boundary and twists them in opposite directions. Then ϕ has at least two fixed points. This theorem was proved in 1913 by Birkhoff [18], and hence is also called the Poincaré–Birkhoff theorem. It has important applications to dynamical systems and celestial mechanics. The Arnold conjecture on the existence of fixed points for symplectomorphisms of compact manifolds (see Section 5.2) may be regarded as a generalization of the Poincaré–Birkhoff theorem. This conjecture has motivated a significant amount of research involving a more general notion of generating function; see, for instance, [41,55].

2.7. Lagrangians and special Lagrangians in Cn The standard Hermitian inner product h(·, ·) on Cn has real and imaginary parts given by the Euclidean inner product ·, · and (minus) the symplectic form ω0 , respectively: for

Symplectic geometry

105

v = (x1 + iy1 , . . . , xn + iyn ), u = (a1 + ib1 , . . . , an + ibn ) ∈ Cn , n n n h(v, u) = (xk + iyk )(ak − ibk ) = (xk ak + yk bk ) − i (xk bk − yk ak ) k=1

k=1

k=1

= v, u − iω0 (v, u). L EMMA 2.17. Let W be a subspace of (Cn , ω0 ) and e1 , . . . , en vectors in Cn . Then: (a) W is Lagrangian if and only if W ⊥ = iW ; (b) (e1 , . . . , en ) is an orthonormal basis of a Lagrangian subspace if and only if (e1 , . . . , en ) is a unitary basis of Cn . P ROOF. (a) We always have ω0 (v, u) = − im h(v, u) = re h(iv, u) = iv, u. It follows that, if W is Lagrangian, so that ω0 (v, u) = 0 for all v, u ∈ W , then iW ⊆ W ⊥ . These spaces must be equal because they have the same dimension. Reciprocally, when iv, u = 0 for all v, u ∈ W , the equality above shows that W must be isotropic. Since dim W = dim iW = dim W ⊥ = 2n − dim W , the dimension of W must be n. (b) If (e1 , . . . , en ) is an orthonormal basis of a Lagrangian subspace W , then, by the previous part, (e1 , . . . , en , ie1 , . . . , ien ) is an orthonormal basis of Cn as a real vector space. Hence (e1 , . . . , en ) must be a complex basis of Cn and it is unitary because h(ej , ek ) = ej , ek − iω0 (ej , ek ) = δj k . Conversely, if (e1 , . . . , en ) is a unitary basis of Cn , then the real span of these vectors is Lagrangian (ω0 (ej , ek ) = − im h(ej , ek ) = 0) and they are orthonormal (ej , ek = re h(ej , ek ) = δj k ). The Lagrangian Grassmannian Λn is the set of all Lagrangian subspaces of Cn . It follows from part (b) of Lemma 2.17 that Λn is the set of all subspaces of Cn admitting an orthonormal basis that is a unitary basis of Cn . Therefore, we have Λn U(n)/O(n). Indeed U(n) acts transitively on Λn : given W, W ∈ Λn with orthonormal bases (e1 , . . . , en ), (e1 , . . . , en ), respectively, there is a unitary transformation of Cn that maps (e1 , . . . , en ) to (e1 , . . . , en ) as unitary bases of Cn . And the stabilizer of Rn ∈ Λn is the subgroup of those unitary transformations that preserve this Lagrangian subspace, namely O(n). It follows that Λn is a compact connected manifold of dimension n(n+1) 2 ; cf. the last example of Section 1.1. The Lagrangian Grassmannian comes with a tautological vector bundle τn := (W, v) ∈ Λn × Cn | v ∈ W , whose fiber over W ∈ Λn is the n-dimensional real space W . It is a consequence of part (a) of Lemma 2.17 that the following map gives a well-defined global isomorphism of the complexification τn ⊗R C with the trivial bundle Cn over Λn (i.e., a global trivialization): (W, v ⊗ c) → (W, cv), for W ∈ Λn , v ∈ W , c ∈ C.

106

A. Cannas da Silva

D EFINITION 2.18. A Lagrangian immersion of a manifold X is an immersion f : X → Cn such that dfp (Tp X) is a Lagrangian subspace of (Cn , ω0 ), for every p ∈ X. E XAMPLE . The graph of a map h : Rn → iRn is an embedded n-dimensional submanifold X of Cn . Its tangent space at (p, h(p)) is {v + dhp (v) | v ∈ Rn }. Let e1 , . . . , en be the standard basis of Rn . Since ω0 (ek + dhp (ek ), ej + dhp (ej )) = ek , −i dhp (ej ) + ∂hj n k ej , i dhp (ek ), we see that X is Lagrangian if and only if ∂h ∂xj = ∂xk , ∀j, k, which in R n is if and only if h is the gradient of some H : R → iR. If f : X → Cn is a Lagrangian immersion, we can define a Gauss map λf : X −→ Λn , p −→ dfp (Tp X). Since λ∗f τn = T X and τn ⊗ C Cn , we see that a necessary condition for an immersion X → Cn to exist is that the complexification of T X be trivializable. Using the h-principle (Section 3.2), Gromov [65] showed that this is also sufficient: an n-dimensional manifold X admits a Lagrangian immersion into Cn if and only if the complexification of its tangent bundle is trivializable. E XAMPLE . For the unit sphere S n = {(t, x) ∈ R × Rn : t 2 + |x|2 = 1}, the Whitney sphere immersion is the map f : S n −→ Cn , (t, x) −→ x + itx. The only self-intersection is at the origin where f (−1, 0, . . . , 0) = f (1, 0, . . . , 0). Since T(t,x) S n = (t, x)⊥ , the differential df(t,x) : (u, v) → v + i(tv + ux) is always injective: v + i(tv + ux) = 0 ⇔ v = 0 and ux = 0, but when x = 0 it is t = ±1 and T(±1,0) S n = {0} × Rn , so it must be u = 0. We conclude that f is an immersion. By computing ω0 at two vectors of the form v + i(tv + ux), we find that the image dfp (Tp S n ) is an n-dimensional isotropic subspace of Cn . Therefore, f is a Lagrangian immersion of S n , and the complexification T S n ⊗ C must be always trivializable, though the tangent bundle T S n is only trivializable in dimensions n = 0, 1, 3, 7. The special Lagrangian Grassmannian SΛn is the set of all oriented subspaces of Cn admitting a positive orthonormal basis (e1 , . . . , en ) that is a special unitary basis of Cn . By the characterization of Lagrangian in the part (b) of Lemma 2.17, it follows that the elements of SΛn are indeed Lagrangian submanifolds. Similarly to the case of the Lagrangian Grassmannian, we have that SΛn SU(n)/SO(n) is a compact connected manifold of dimension

n(n+1) 2

− 1.

Symplectic geometry

107

We can single out the special Lagrangian subspaces by expressing the condition on the determinant in terms of the real n-form in Cn , β := im Ω,

where Ω := dz1 ∧ · · · ∧ dzn .

Since for A ∈ SO(n), we have det A = 1 and Ω(e1 , . . . , en ) = Ω(Ae1 , . . . , Aen ), we see that, for an oriented real n-dimensional subspace W ⊂ Cn , the number Ω(e1 , . . . , en ) does not depend on the choice of a positive orthonormal basis (e1 , . . . , en ) of W , thus can be denoted Ω(W ) and its imaginary part β(W ). P ROPOSITION 2.19. A subspace W of (Cn , ω0 ) has an orientation for which it is a special Lagrangian if and only if W is Lagrangian and β(W ) = 0. P ROOF. Any orthonormal basis (e1 , . . . , en ) of a Lagrangian subspace W ⊂ Cn is the image of the canonical basis of Cn by some A ∈ U(n), and Ω(W ) = det A ∈ S 1 . Therefore, W admits an orientation for which such a positive (e1 , . . . , en ) is a special unitary basis of Cn if and only if det A = ±1, i.e., β(W ) = 0. D EFINITION 2.20. A special Lagrangian immersion of an oriented manifold X is a Lagrangian immersion f : X → Cn such that, at each p ∈ X, the space dfp (Tp X) is a special Lagrangian subspace of (Cn , ω0 ). For a special Lagrangian immersion f , the Gauss map λf takes values in SΛn . By Proposition 2.19, the immersion f of an n-dimensional manifold X in (Cn , ω0 ) is special Lagrangian if and only if f ∗ ω0 = 0 and f ∗ β = 0. E XAMPLE . In C2 , writing zk = xk + iyk , we have β = dx1 ∧ dy2 + dy1 ∧ dx2 . We have seen that the graph of the gradient i∇H is Lagrangian, for any function H : R2 → R. ∂H ∂H , i ∂x ) is a Lagrangian immersion. For f to be a special LaSo f (x1 , x2 ) = (x1 , x2 , i ∂x 1 2 grangian immersion, we need the vanish of f ∗ β = dx1 ∧ d

∂H ∂x2

+d

∂H ∂x1

∧ dx2 =

∂ 2H ∂ 2H dx1 ∧ dx2 . + ∂x12 ∂x22

Hence the graph of ∇H is special Lagrangian if and only if H is harmonic. If f : X → Cn is a special Lagrangian immersion, then f ∗ Ω is an exact (real) volume form: f ∗ Ω = d re(z1 dz2 ∧ · · · ∧ dzn ). We conclude, by Stokes theorem, that there can be no special Lagrangian immersion of a compact manifold in Cn . Calabi–Yau manifolds21 are more general manifolds where a definition of special Lagrangian submanifold makes sense and where the space of special Lagrangian embeddings of a compact manifold is interesting. Special Lagrangian geometry was introduced by Harvey and Lawson [71]. For a treatment of Lagrangian and special Lagrangian submanifolds with many examples; see, for instance, [9]. 21 Calabi–Yau manifolds are compact Kähler manifolds (Section 3.4) with vanishing first Chern class.

108

A. Cannas da Silva

3. Complex structures 3.1. Compatible linear structures A complex structure on a vector space V is a linear map J : V → V such that J 2 = −Id. The pair (V , J ) is then called a complex vector space.A complex structure J on V is equivalent to a structure of vector space over C, the map J corresponding to multiplication by i. If (V , Ω) is a symplectic vector space, a complex structure J on V is said to be compatible (with Ω, or Ω-compatible) if the bilinear map GJ : V × V → R defined by GJ (u, v) = Ω(u, J v) is an inner product on V . This condition comprises J being a symplectomorphism (i.e., Ω(J u, J v) = Ω(u, v), ∀u, v) and the so-called taming: Ω(u, J u) > 0, ∀u = 0. E XAMPLE . For the symplectic vector space (R2n , Ω0 ) with symplectic basis e1 = (1, 0, . . . , 0), . . . , en , f1 , . . . , fn = (0, . . . , 0, 1), there is a standard compatible complex structure J0 determined by J0 (ej ) = fj and J0 (fj ) = −ej for all j = 1, . . . , n. This corresponds to a standard identification of R2n with Cn , and Ω0 (u, J0 v) = u, v is the standard Euclidean inner product. With respect to the symplectic basis e1 , . . . , en , f1 , . . . , fn , the map J0 is represented by the matrix !

0 Id

" −Id . 0

The symplectic linear group, Sp(2n) := {A ∈ GL(2n; R) | Ω0 (Au, Av) = Ω0 (u, v) for all u, v ∈ R2n }, is the group of all linear transformations of R2n that preserve the standard symplectic structure. The orthogonal group O(2n) is the group formed by the linear transformations A that preserve the Euclidean inner product, Au, Av = u, v, for all u, v ∈ R2n . The general complex group GL(n; C) is the group of linear transformations A : R2n → R2n commuting with J0 , A(J0 v) = J0 (Av), for all v ∈ R2n .22 The compatibility between the structures Ω0 , ·, · and J0 implies that the intersection of any two of these subgroups of GL(2n; R) is the same group, namely the unitary group U(n). As (R2n , Ω0 ) is the prototype of a 2n-dimensional symplectic vector space, the preceding example shows that compatible complex structures always exist on symplectic vector spaces.23 There is yet a way to produce a canonical compatible complex structure J after the choice of an inner product G on (V , Ω), though the starting G(u, v) is usually different from GJ (u, v) := Ω(u, J v). P ROPOSITION 3.1. Let (V , Ω) be a symplectic vector space, with an inner product G. Then there is a canonical compatible complex structure J on V . 22 Identify the complex n × n matrix X + iY with the real 2n × 2n matrix X −Y . Y X 23 Conversely, given (V , J ), there is a symplectic Ω with which J is compatible: take Ω(u, v) = G(J u, v) for an inner product G such that J t = −J .

Symplectic geometry

109

P ROOF. By nondegeneracy of Ω and G, the maps u → Ω(u, ·) and w → G(w, ·) are both isomorphisms between V and V ∗ . Hence, Ω(u, v) = G(Au, v) for some linear A : V → V . The map A is skew-symmetric, and the product AAt is symmetric24 and positive: G(AAt u, u) = G(At u, At u) > 0, for u = 0. By the spectral theorem, these properties imply that AAt diagonalizes with positive eigenvalues λi , say AAt = B diag(λ1 , . . . , λ2n ) B −1 . We may hence define an arbitrary real power of AAt by rescaling the eigenspaces, in particular, √ # # AAt := B diag( λ1 , . . . , λ2n )B −1 . √ The linear transformation AAt is symmetric, positive-definite and does not depend on the choice of B nor of the ordering of the eigenvalues. It is completely determined by its effect on√each eigenspace of AAt : on the eigenspace √ corresponding to the eigenvalue λk , the map AA√t is defined to be multiplication by λk . √ Let J := ( AAt )−1 A. Since A and AAt commute, J is orthogonal (J J t = Id), as well as skew-symmetric (J t = −J ). It follows that J is a complex structure on V . Compatibility is easily checked: Ω(J u, J v) = G(AJ u, J v) = G(J Au, J v) = G(Au, v) = Ω(u, v) and √ Ω(u, J u) = G(Au, J u) = G(−J Au, u) = G AAt u, u > 0, The factorization A =

for u = 0.

√ AAt J is called the polar decomposition of A.

R EMARK . Being canonical, this construction may be smoothly performed: when (Vt , Ωt ) is a family of symplectic vector spaces with a family Gt of inner products, all depending smoothly on a parameter t, an adaptation of the previous proof shows that there is a smooth family Jt of compatible complex structures on (Vt , Ωt ). Let (V , Ω) be a symplectic vector space of dimension 2n, and let J be a complex structure on V . If J is Ω-compatible and L is a Lagrangian subspace of (V , Ω), then J L is also Lagrangian and J L = L⊥ , where ⊥ indicates orthogonality with respect to the inner product GJ (u, v) = Ω(u, J v). Therefore, a complex structure J is Ω-compatible if and only if there exists a symplectic basis for V of the form e 1 , e 2 , . . . , en ,

f 1 = J e1 ,

f 2 = J e2 ,

...,

fn = J en .

Let J (V , Ω) be the set of all compatible complex structures in a symplectic vector space (V , Ω). 24 A map B : V → V is symmetric, respectively skew-symmetric, when B t = B, respectively B t = −B, where the transpose B t : V → V is determined by G(B t u, v) = G(u, Bv).

110

A. Cannas da Silva

P ROPOSITION 3.2. The set J (V , Ω) is contractible.25 P ROOF. Pick a Lagrangian subspace L0 of (V , Ω). Let L(V , Ω, L0 ) be the space of all Lagrangian subspaces of (V , Ω) that intersect L0 transversally. Let G(L0 ) be the space of all inner products on L0 . The map Ψ : J (V , Ω) −→ L(V , Ω, L0 ) × G(L0 ), J −→ (J L0 , GJ |L0 ) is a homeomorphism, with inverse as follows. Take (L, G) ∈ L(V , Ω, L0 ) × G(L0 ). For v ∈ L0 , v ⊥ = {u ∈ L0 | G(u, v) = 0} is a (n − 1)-dimensional space of L0 ; its symplectic orthogonal (v ⊥ )Ω is (n + 1)-dimensional. Then (v ⊥ )Ω ∩ L is 1-dimensional. Let J v be the unique vector in this line such that Ω(v, J v) = 1. If we take v’s in some G-orthonormal basis of L0 , this defines an element J ∈ J (V , Ω). The set L(V , Ω, L0 ) can be identified with the vector space of all symmetric n × n matrices. In fact, any n-dimensional subspace L of V that is transverse to L0 is the graph of a linear map J L0 → L0 , and the Lagrangian ones correspond to symmetric maps (cf. Section 1.1). Hence, L(V , Ω, L0 ) is contractible. Since G(L0 ) is contractible (it is even convex), we conclude that J (V , Ω) is contractible.

3.2. Compatible almost complex structures An almost complex structure on a manifold M is a smooth26 field of complex structures on the tangent spaces, Jp : Tp M → Tp M, p ∈ M. The pair (M, J ) is then called an almost complex manifold. D EFINITION 3.3. An almost complex structure J on a symplectic manifold (M, ω) is compatible (with ω or ω-compatible) if the map that assigns to each point p ∈ M the bilinear pairing gp : Tp M × Tp M → R, gp (u, v) := ωp (u, Jp v) is a Riemannian metric on M. A triple (ω, g, J ) of a symplectic form, a Riemannian metric and an almost complex structure on a manifold M is a compatible triple when g(·, ·) = ω(·, J ·). If (ω, J, g) is a compatible triple, each of ω, J or g can be written in terms of the other two. E XAMPLES . 1. If we identify R2n with Cn using coordinates zj = xj + iyj , multiplication by i induces a constant linear map J0 on the tangent spaces such that J02 = −Id, known 25 Contractibility of J (V , Ω) means that there exists a homotopy h : J (V , Ω) → J (V , Ω), 0 t 1, starting t at the identity h0 = Id, finishing at a trivial map h1 : J (V , Ω) → {J0 }, and fixing J0 (i.e., ht (J0 ) = J0 , ∀t ) for

some J0 ∈ J (V , Ω). 26 Smoothness means that for any vector field v, the image J v is a (smooth) vector field.

Symplectic geometry

111

as the standard almost complex structure on R2n : J0

∂ ∂xj

∂ = , ∂yj

J0

∂ ∂yj

=−

∂ . ∂xj

For the standard symplectic form ω0 = dxj ∧ dyj and the Euclidean inner product g0 = ·, ·, the compatibility relation holds: ω0 (u, v) = g0 (J0 (u), v). 2. Any oriented hypersurface Σ ⊂ R3 carries a natural symplectic form and a natural compatible almost complex structure induced by the standard inner (or dot) and exterior (or vector) products. They are given by the formulas ωp (u, v) := νp , u × v and Jp (v) = νp × v for v ∈ Tp Σ , where νp is the outward-pointing unit normal vector at p ∈ Σ (in other words, ν : Σ → S 2 is the Gauss map). Cf. Example 3 of Section 1.2. The corresponding Riemannian metric is the restriction to Σ of the standard Euclidean metric ·, ·. 3. The previous example generalizes to the oriented hypersurfaces M ⊂ R7 . Regarding u, v ∈ R7 as imaginary octonions (or Cayley numbers), the natural vector product u × v is the imaginary part of the product of u and v as octonions. This induces a natural almost complex structure on M given by Jp (v) = νp × v, where νp is the outward-pointing unit normal vector at p ∈ M. In the case of S 6 , at least, this J is not compatible with any symplectic form, as S 6 cannot be a symplectic manifold. As a consequence of the remark in Section 3.1, we have: P ROPOSITION 3.4. On any symplectic manifold (M, ω) with a Riemannian metric g, there is a canonical compatible almost complex structure J . Since Riemannian metrics always exist, we conclude that any symplectic manifold has compatible almost complex structures. The metric gJ (·, ·) := ω(·, J ·) tends to be different from the given g(·, ·). P ROPOSITION 3.5. Let (M, J ) be an almost complex manifold where J is compatible with two symplectic forms ω0 , ω1 Then ω0 and ω1 are deformation-equivalent. P ROOF. Simply take the convex combinations ωt = (1 − t)ω0 + tω1 , 0 t 1.

A counterexample to the converse of this proposition is provided by the family ωt = cos πt dx1 ∧ dy1 + sin πt dx1 ∧ dy2 + sin πt dy1 ∧ dx2 + cos πt dx2 ∧ dy2 for 0 t 1. There is no J in R4 compatible with both ω0 and ω1 = −ω0 . A submanifold X of an almost complex manifold (M, J ) is an almost complex submanifold when J (T X) ⊆ T X, i.e., we have Jp v ∈ Tp X, ∀p ∈ X, v ∈ Tp X. P ROPOSITION 3.6. Let (M, ω) be a symplectic manifold equipped with a compatible almost complex structure J . Then any almost complex submanifold X of (M, J ) is a symplectic submanifold of (M, ω).

112

A. Cannas da Silva

P ROOF. Let i : X → M be the inclusion. Then i ∗ ω is a closed 2-form on X. Since ωp (u, v) = gp (Jp u, v), ∀p ∈ X, ∀u, v ∈ Tp X, and since gp |Tp X is nondegenerate, so is ωp |Tp X , and i ∗ ω is nondegenerate. It is easy to see that the set J (M, ω) of all compatible almost complex structures on a symplectic manifold (M, ω) is path-connected. From two almost complex structures J0 , J1 compatible with ω, we get two Riemannian metrics g0 (·, ·) = ω(·, J0 ·), g1 (·, ·) = ω(·, J1 ·). Their convex combinations gt (·, ·) = (1 − t)g0 (·, ·) + tg1 (·, ·),

0 t 1,

form a smooth family of Riemannian metrics. Applying the polar decomposition to the family (ω, gt ), we obtain a smooth path of compatible almost complex structures Jt joining J0 to J1 . The set J (M, ω) is even contractible (this is important for defining invariants). The first ingredient is the contractibility of the set of compatible complex structures on a vector space (Proposition 3.2). Consider the fiber bundle J → M with fiber over p ∈ M being the space Jp := J (Tp M, ωp ) of compatible complex structures on the tangent space at p. A compatible almost complex structure on (M, ω) is a section of J . The space of sections of J is contractible because the fibers are contractible.27 The first Chern class c1 (M, ω) of a symplectic manifold (M, ω) is the first Chern class of (T M, J ) for any compatible J . The class c1 (M, ω) ∈ H 2 (M; Z) is invariant under deformations of ω. We never used the closedness of ω to obtain compatible almost complex structures. The construction holds for an almost symplectic manifold (M, ω), that is, a pair of a manifold M and a nondegenerate 2-form ω, not necessarily closed. We could further work with a symplectic vector bundle, that is, a vector bundle E → M equipped with a smooth field ω of fiberwise nondegenerate skew-symmetric bilinear maps (Section 1.6). The existence of such a field ω is equivalent to being able to reduce the structure group of the bundle from the general linear group to the linear symplectic group. As both Sp(2n) and GL(n; C) retract to their common maximal compact subgroup U(n), a symplectic vector bundle can be always endowed with a structure of complex vector bundle, and vice-versa. Gromov showed in his thesis [63] that any open28 almost complex manifold admits a symplectic form. The books [42, §10.2] and [99, §7.3] contain proofs of this statement using different techniques. T HEOREM 3.7 (Gromov). For an open manifold the existence of an almost complex structure J implies that of a symplectic form ω in any given 2-cohomology class and such that J is homotopic to an almost complex structure compatible with ω. From an almost complex structure J and a metric g, one builds a nondegenerate 2-form ω(u, v) = g(J u, v), which will not be closed in general. Closedness is a differential re27 The base being a (second countable and Hausdorff) manifold, a contraction can be produced using a countable cover by trivializing neighborhoods whose closures are compact subsets of larger trivializing neighborhoods, and such that each p ∈ M belongs to only a finite number of such neighborhoods. 28 A manifold is open if it has no closed connected components, where closed means compact and without boundary.

Symplectic geometry

113

lation, i.e., a condition imposed on the partial derivatives, encoded as a subset of jet space. One says that a differential relation satisfies the h-principle29 if any formal solution (i.e., a solution for the associated algebraic problem, in the present case a nondegenerate 2-form) is homotopic to a holonomic solution (i.e., a genuine solution, in the present case a closed nondegenerate 2-form). Therefore, when the h-principle holds, one may concentrate on a purely topological question (such as the existence of an almost complex structure) in order to prove the existence of a differential solution. Gromov showed that, for an open differential relation on an open manifold, when the relation is invariant under the group of diffeomorphisms of the underlying manifold, the inclusion of the space of holonomic solutions into the space of formal solutions is a weak homotopy equivalence, i.e., induces isomorphisms of all homotopy groups. The previous theorem fits here as an application. For closed manifolds there is no such theorem: as discussed in Section 1.2, the existence of a 2-cohomology class whose top power is nonzero is also necessary for the existence of a symplectic form and there are further restrictions coming from Gromov–Witten theory (see Section 4.5). 3.3. Integrability Any complex manifold30 has a canonical almost complex structure J . It is defined locally over the domain U of a complex chart ϕ : U → V ⊆ Cn , by Jp ( ∂x∂ j |p ) = ∂y∂ j |p and Jp ( ∂y∂ j |p ) = − ∂x∂ j |p , where these are the tangent vectors induced by the real and imaginary parts of the coordinates of ϕ = (z1 , . . . , zn ), zj = xj + iyj . This yields a globally well-defined J , thanks to the Cauchy–Riemann equations satisfied by the components of the transition maps. An almost complex structure J on a manifold M is called integrable when J is induced by some underlying structure of complex manifold on M as above. The question arises whether some compatible almost complex structure J on a symplectic manifold (M, ω) is integrable. To understand what is involved, we review Dolbeault theory and the Newlander–Nirenberg theorem. Let (M, J ) be a 2n-dimensional almost complex manifold. The fibers of the complexified tangent bundle, T M ⊗ C, are 2n-dimensional vector spaces over C. We may extend J linearly to T M ⊗ C by J (v ⊗ c) = J v ⊗ c, v ∈ T M, c ∈ C. Since J 2 = −Id, on the complex vector space (T M ⊗ C)p the linear map Jp has eigenvalues ±i. The (±i)-eigenspaces of J are denoted T1,0 and T0,1 , respectively, and called the spaces of J -holomorphic and of J -anti-holomorphic tangent vectors. We have an isomorphism

(π1,0 , π0,1 ) : T M ⊗ C −→ T1,0 ⊕ T0,1 , 1 v −→ (v − iJ v, v + iJ v), 2 29 There are in fact different h-principles depending on the different possible coincidences of homotopy groups for the spaces of formal solutions and of holonomic solutions. 30 A complex manifold of (complex) dimension n is a set M with a complete complex atlas {(U , V , ϕ ), α ∈ α α α $ index set I } where M = α Uα , the Vα ’s are open subsets of Cn , and the maps ϕα : Uα → Vα are bijections such that the transition maps ψαβ = ϕβ ◦ ϕα−1 : Vαβ → Vβα are biholomorphic (i.e., bijective, holomorphic and with holomorphic inverse) as maps on open subsets of Cn , Vαβ = ϕα (Uα ∩ Uβ ).

114

A. Cannas da Silva

where the maps to each summand satisfy π1,0 ◦ J = iπ1,0 and π0,1 ◦ J = −iπ0,1 . Restricting π1,0 to T M, we see that (T M, J ) T1,0 T0,1 , as complex vector bundles, where the multiplication by i is given by J in (T M, J ) and where T0,1 denotes the complex conjugate bundle of T0,1 . Similarly, J ∗ defined on T ∗ M ⊗ C by J ∗ ξ = ξ ◦ J has (±i)-eigenspaces T 1,0 = (T1,0 )∗ and T 0,1 = (T0,1 )∗ , respectively, called the spaces of complex-linear and of complexantilinear cotangent vectors. Under the two natural projections π 1,0 , π 0,1 , the complexified cotangent bundle splits as 1,0 0,1 ∗ : T M ⊗ C −→ T 1,0 ⊕ T 0,1 , π ,π 1 ξ −→ (ξ − iJ ∗ ξ, ξ + iJ ∗ ξ ). 2 Let Λk (T ∗ M ⊗ C) := Λk T 1,0 ⊕ T 0,1 = Λ,m , +m=k

where Λ,m := (Λ T 1,0 ) ∧ (Λm T 0,1 ), and let Ω k (M; C) be the space of sections of Λk (T ∗ M ⊗ C), called complex-valued k-forms on M. The differential forms of type (, m) on (M, J ) are the sections of Λ,m , and the space of these differential forms is Ω ,m . The decomposition of forms by Dolbeault type is Ω k (M; C) = % denoted ,m . Let π ,m : Λk (T ∗ M ⊗ C) → Λ,m be the projection map, where + m = k. +m=k Ω The usual exterior derivative d (extended linearly to smooth complex-valued forms) composed with two of these projections induces the del and del-bar differential operators, ¯ on forms of type (, m): ∂ and ∂, ∂ := π +1,m ◦ d : Ω ,m −→ Ω +1,m and ∂¯ := π ,m+1 ◦ d : Ω ,m −→ Ω ,m+1 . If β ∈ Ω ,m (M), with k = + m, then dβ ∈ Ω k+1 (M; C): dβ =

¯ + · · · + π 0,k+1 dβ. π r,s dβ = π k+1,0 dβ + · · · + ∂β + ∂β

r+s=k+1

In particular, on complex-valued functions we have df = d(re f ) + i d(im f ) and d = ¯ where ∂ = π 1,0 ◦ d and ∂¯ = π 0,1 ◦ d. A function f : M → C is J -holomorphic ∂ + ∂, at p ∈ M if dfp is complex linear, i.e., dfp ◦ Jp = i dfp (or dfp ∈ Tp1,0 ). A function f is J -holomorphic if it is holomorphic at all p ∈ M. A function f : M → C is J -antiholomorphic at p ∈ M if dfp is complex antilinear, i.e., dfp ◦ Jp = −i dfp (or dfp ∈ Tp0,1 ), ¯ that is, when the conjugate function f¯ is holomorphic at p ∈ M. In terms of ∂ and ∂,

Symplectic geometry

115

¯ = 0, and f is J -anti-holomorphic if and a function f is J -holomorphic if and only if ∂f only if ∂f = 0. When M is a complex % manifold and J is its canonical almost complex structure, the splitting Ω k (M; C) = +m=k Ω ,m is particularly interesting. Let U ⊆ M be the domain of a complex coordinate chart ϕ = (z1 , . . . , zn ), where the corresponding real coordinates x1 , y1 , . . . , xn , yn satisfy zj = xj + iyj . In terms of ∂ 1 ∂ ∂ and := −i ∂zj 2 ∂xj ∂yj

∂ 1 ∂ ∂ , := +i ∂ z¯ j 2 ∂xj ∂yj

the (±i)-eigenspaces of Jp (p ∈ U ) can be written

∂

∂

(T1,0 )p = C-span . : j = 1, . . . , n and (T ) = C-span 0,1 p ∂zj p ∂ z¯ j p Similarly, putting dzj = dxj + i dyj and d z¯ j = dxj − i dyj , we obtain simple formulas ∂b ∂b ¯ for the differentials of a b ∈ C ∞ (U; C), ∂b = ∂zj dzj and ∂b = ∂ z¯ j d z¯ j , and we have T 1,0 = C-span{dzj : j = 1, . . . , n} and T 0,1 = C-span{d z¯ j : j = 1, . . . , n}. If we use multiindex notation J = (j1 , . . . , j ) where 1 j1 < · · · < j n, |J | = and dzJ = dzj1 ∧ dzj2 ∧ · · · ∧ dzj , then the set of (, m)-forms on U is

bJ,K dzJ ∧ d z¯ K | bJ,K ∈ C ∞ (U; C) .

Ω ,m =

|J |=, |K|=m

A form β ∈ Ω k (M; C) may be written over U as β= bJ,K dzJ ∧ d z¯ K . +m=k |J |=, |K|=m

Since d = ∂ + ∂¯ on functions, we get dβ = dbJ,K ∧ dzJ ∧ d z¯ K +m=k |J |=, |K|=m

=

+m=k

&

∂bJ,K ∧ dzJ ∧ d z¯ K

|J |=, |K|=m

'(

)

∈Ω +1,m

+

|J |=, |K|=m

&

¯ J,K ∧ dzJ ∧ d z¯ K ∂b '( ∈Ω ,m+1

¯ = ∂β + ∂β,

)

116

A. Cannas da Silva

and conclude that, on a complex manifold, d = ∂ + ∂¯ on forms of any degree. This cannot be proved for an almost complex manifold, because there are no coordinate functions zj to give a suitable basis of 1-forms. ¯ for any form β ∈ Ω ,m , we have When d = ∂ + ∂, ¯ + ∂∂β ¯ + ∂¯ 2 β 0 = d 2 β = ∂ 2 β + ∂ ∂β &'() & '( ) &'() ∈Ω +2,m

∈Ω +1,m+1 ∂¯

&⇒

∈Ω ,m+2 ∂¯

2 ∂¯ = 0, ¯ = 0, ∂ ∂¯ + ∂∂ 2 ∂ = 0.

∂¯

Since ∂¯ 2 = 0, the chain 0 −→ Ω ,0 −→ Ω ,1 −→ Ω ,2 −→ · · · is a differential complex. Its cohomology groups ,m HDolbeault (M) :=

ker ∂¯ : Ω ,m → Ω ,m+1 im ∂¯ : Ω ,m−1 → Ω ,m

are called the Dolbeault cohomology groups. The Dolbeault theorem states that for com,m (M) H m (M; O(Ω (,0) )), where O(Ω (,0) ) is the sheaf of plex manifolds HDolbeault forms of type (, 0) over M. It is natural to ask whether the identity d = ∂ + ∂¯ could hold for manifolds other than complex manifolds. Newlander and Nirenberg [106] showed that the answer is no: for an almost complex manifold (M, J ), the following are equivalent M is a complex manifold

⇐⇒

N ≡0

⇐⇒

∂¯ 2 = 0,

⇐⇒

d = ∂ + ∂¯

where N is the Nijenhuis tensor: N (X, Y ) := [J X, J Y ] − J [J X, Y ] − J [X, J Y ] − [X, Y ], for vector fields X and Y on M, [·, ·] being the usual bracket.31 The Nijenhuis tensor can be thought of as a measure of the existence of J -holomorphic functions: if there exist n J -holomorphic functions, f1 , . . . , fn , on R2n , that are independent at some point p, i.e., the real and imaginary parts of (df1 )p , . . . , (dfn )p form a basis of Tp∗ R2n , then N vanishes identically at p. More material related to Dolbeault theory or to the Newlander–Nirenberg theorem can be found in [23,37,62,76,141]. E XAMPLE . Out of all spheres, only S 2 and S 6 admit almost complex structures [121, §41.20]. As a complex manifold, S 2 if referred to as the Riemann sphere CP1 . The almost complex structure on S 6 from Example 3 of Section 3.2 is not integrable, but it is not yet known whether S 6 admits a structure of complex manifold. 31 The bracket of vector fields X and Y is the vector field [X, Y ] characterized by the property that L [X,Y ] f := LX (LY f ) − LY (LX f ), for f ∈ C ∞ (M), where LX f = df (X).

Symplectic geometry

117

In the (real) 2-dimensional case N always vanishes simply because N is a tensor, i.e., N (f X, gY ) = f gN (X, Y ) for any f, g ∈ C ∞ (M), and N (X, J X) = 0 for any vector field X. Combining this with the fact that any orientable surface is symplectic, we conclude that any orientable surface is a complex manifold, a result already known to Gauss. However, most almost complex structures on higher-dimensional manifolds are not integrable. In Section 3.5 we see that the existence of a complex structure compatible with a symplectic structure on a compact manifold imposes significant topological constraints. 3.4. Kähler manifolds D EFINITION 3.8. A Kähler manifold is a symplectic manifold (M, ω) equipped with an integrable compatible almost complex structure J . The symplectic form ω is then called a Kähler form. As a complex manifold, a Kähler manifold (M, ω, J ) has Dolbeault cohomology. As it is also a symplectic manifold, it is interesting to understand where the symplectic form ω sits with respect to the Dolbeault type decomposition. ¯ P ROPOSITION 3.9. A Kähler form ω is a ∂- and ∂-closed (1, 1)-form that is given on a local complex chart (U, z1 , . . . , zn ) by ω=

n i hj k dzj ∧ d z¯ k , 2 j,k=1

where, at every point p ∈ U , (hj k (p)) is a positive-definite Hermitian matrix. 1,1 (M). In particular, ω defines a Dolbeault (1, 1)-cohomology class, [ω] ∈ HDolbeault

P ROOF. Being a form in Ω 2 (M; C) = Ω 2,0 ⊕Ω 1,1 ⊕Ω 0,2 , with respect to a local complex chart, ω can be written ω= aj k dzj ∧ dzk + bj k dzj ∧ d z¯ k + cj k d z¯ j ∧ d z¯ k for some aj k , bj k , cj k ∈ C ∞ (U; C). By the compatibility of ω with the complex structure, J is a symplectomorphism, that is, J ∗ ω = ω where (J ∗ ω)(u, v) := ω(J u, J v). Since J ∗ dzj = dzj ◦ J = i dzj and J ∗ d z¯ j = d z¯ j ◦ J = −i d z¯ j , we have J ∗ ω = ω if and only if the coefficients aj k and cj k all vanish identically, that is, if and only if ω ∈ Ω 1,1 . Since ¯ we must have ∂ω = 0 and ∂ω ¯ = 0. Set ω is closed, of type (1, 1) and dω = ∂ω +∂ω, i i i bj k = 2 hj k . As ω is real-valued, i.e., ω = 2 hj k dzj ∧ d z¯ k and ω¯ = − 2 hj k d z¯ j ∧ dzk coincide, we must have hj k = hkj for all j and k. In other words, at every point p ∈ U , the n × n matrix (hj k (p)) is Hermitian. The nondegeneracy amounts to the nonvanishing of n i ω = n! det(hj k ) dz1 ∧ d z¯ 1 ∧ · · · ∧ dzn ∧ d z¯ n . 2 n

118

A. Cannas da Silva

Therefore, at every p ∈ M, the matrix (hj k (p)) must be nonsingular. Finally, the positivity condition ω(v, J v) > 0, ∀v = 0, from compatibility, implies that, at each p ∈ U , the matrix (hj k (p)) is positive-definite. Consequently, if ω0 and ω1 are both Kähler forms on a compact manifold M with 2 [ω0 ] = [ω1 ] ∈ HdeRham (M), then (M, ω0 ) and (M, ω1 ) are strongly isotopic by Moser’s Theorem 1.7. Indeed ωt = (1 − t)ω0 + tω1 is symplectic for t ∈ [0, 1], as convex combinations of positive-definite matrices are still positive-definite. Another consequence is the following recipe for Kähler forms. A smooth real function ρ on a complex manifold M is strictly plurisubharmonic (s.p.s.h.) if, on each local 2 complex chart (U, z1 , . . . , zn ), the matrix ( ∂z∂j ∂ρz¯ k (p)) is positive-definite at all p ∈ U . If ρ ∈ C ∞ (M; R) is s.p.s.h., then the form i ¯ ω = ∂ ∂ρ 2 is Kähler. The function ρ is then called a (global) Kähler potential. E XAMPLE . Let M = Cn R2n , with complex coordinates (z1 , . . . , zn ) and corresponding real coordinates (x1 , y1 , . . . , xn , yn ) via zj = xj + iyj . The function ρ(x1 , y1 , . . . , xn , yn ) =

n 2 xj + yj2 = |zj |2 = zj z¯ j j =1

is s.p.s.h. and is a Kähler potential for the standard Kähler form: i i i ¯ ∂ ∂ρ = δj k dzj ∧ d z¯ k = dzj ∧ d z¯ j = dxj ∧ dyj = ω0 . 2 2 2 j,k

j

j

There is a local converse to the previous construction of Kähler forms. P ROPOSITION 3.10. Let ω be a closed real-valued (1, 1)-form on a complex mani¯ for some fold M and let p ∈ M. Then on a neighborhood U of p we have ω = 2i ∂ ∂ρ ∞ ρ ∈ C (U; R). The proof of this theorem requires holomorphic versions of Poincaré’s lemma, namely, the local triviality of Dolbeault groups (the fact that any point in a complex manifold admits ,m a neighborhood U such that HDolbeault (U) = 0 for all m > 0) and the local triviality of the holomorphic de Rham groups; see [62]. For a Kähler ω, such a local function ρ is called a local Kähler potential. P ROPOSITION 3.11. Let M be a complex manifold, ρ ∈ C ∞ (M; R) s.p.s.h., X a complex submanifold, and i : X → M the inclusion map. Then i ∗ ρ is s.p.s.h.

Symplectic geometry

119

P ROOF. It suffices to verify this locally by considering a complex chart (z1 , . . . , zn ) for M adapted to X so that X is given there by the equations z1 = · · · = zm = 0. Being a 2 principal minor of the positive-definite matrix ( ∂z∂j ∂ z¯ k (0, . . . , 0, zm+1 , . . . , zn )) the matrix 2

∂ ρ ( ∂zm+j ∂ z¯ m+k (0, . . . , 0, zm+1 , . . . , zn )) is also positive-definite.

C OROLLARY 3.12. Any complex submanifold of a Kähler manifold is also Kähler. D EFINITION 3.13. Let (M, ω) be a Kähler manifold, X a complex submanifold, and i : X → M the inclusion. Then (X, i ∗ ω) is called a Kähler submanifold. E XAMPLES . dzj ∧ d z¯ j is Kähler. According to 1. Complex vector space (Cn , ω0 ) where ω0 = 2i Corollary 3.12, every complex submanifold of Cn is Kähler. 2. In particular, Stein manifolds are Kähler. Stein manifolds are the properly embedded complex submanifolds of Cn . They can be alternatively characterized as being the ¯ Kähler manifolds (M, ω) that admit a global proper Kähler potential, i.e., ω = 2i ∂ ∂ρ for some proper function ρ : M → R. 3. The function z → log(|z|2 + 1) on Cn is strictly plurisubharmonic. Therefore the 2-form i ωFS = ∂ ∂¯ log |z|2 + 1 2 is another Kähler form on Cn This is called the Fubini–Study form on Cn . 4. Let {(Uj , Cn , ϕj ), j = 0, . . . , n} be the usual complex atlas for complex projective space.32 The form ωFS is preserved by the transition maps, hence ϕj∗ ωFS and ϕk∗ ωFS agree on the overlap Uj ∩ Uk . The Fubini–Study form on CPn is the Kähler form obtained by gluing together the ϕj∗ ωFS , j = 0, . . . , n. 5. Consequently, all nonsingular projective varieties are Kähler submanifolds. Here by nonsingular we mean smooth, and by projective variety we mean the zero locus of some collection of homogeneous polynomials. 6. All Riemann surfaces are Kähler, since any compatible almost complex structure is integrable for dimension reasons (Section 3.3). 32 The complex projective space CPn is the complex n-dimensional manifold given by the space of complex lines in Cn+1 . It can be obtained from Cn+1 \ {0} by making the identifications (z0 , . . . , zn ) ∼ (λz0 , . . . , λzn ) for all λ ∈ C \ {0}. One denotes by [z0 , . . . , zn ] the equivalence class of (z0 , . . . , zn ), and calls z0 , . . . , zn the homogeneous coordinates of the point p = [z0 , . . . , zn ]. (Homogeneous coordinates are, of course, only determined up to multiplication by a nonzero complex number λ.) Let Uj be the subset of CPn consisting of all points p = [z0 , . . . , zn ] for which zj = 0. Let ϕj : Uj → Cn be the map defined by

z zj −1 zj +1 zn ϕj [z0 , . . . , zn ] = 0 , . . . , , ,..., . zj zj zj zj The collection {(Uj , Cn , ϕj ), j = 0, . . . , n} is the usual complex atlas for CPn . For instance, the transition map z from (U0 , Cn , ϕ0 ) to (U1 , Cn , ϕ1 ) is ϕ0,1 (z1 , . . . , zn ) = ( z1 , z2 , . . . , zzn ) defined from the set {(z1 , . . . , zn ) ∈ 1 1 1 n C | z1 = 0} to itself.

120

A. Cannas da Silva

7. The Fubini–Study form on the chart U0 = {[z0 , z1 ] ∈ CP1 | z0 = 0} of the Riemann sphere CP1 is given by the formula ωFS =

(x 2

dx ∧ dy , + y 2 + 1)2

where zz10 = z = x + iy is the usual coordinate on C. The standard area form ωstd = dθ ∧ dh is induced by regarding CP1 as the unit sphere S 2 in R3 (Example 3 of Section 1.2). Stereographic projection shows that ωFS = 14 ωstd . n n n 8. Complex tori are Kähler. Complex tori look like quotients C /Z where Z is a lattice n in C . The form ω = dzj ∧ d z¯ j induced by the Euclidean structure is Kähler. 9. Just like products of symplectic manifolds are symplectic, also products of Kähler manifolds are Kähler.

3.5. Hodge theory Hodge [73] identified the spaces of cohomology classes of forms with spaces of actual forms, by picking the representative from each class that solves a certain differential equation, namely the harmonic representative. We give a sketch of Hodge’s idea. The first part makes up ordinary Hodge theory, which works for any compact oriented Riemannian manifold (M, g), not necessarily Kähler. At a point p ∈ M, let e1 , . . . , en be a positively oriented orthonormal basis of the cotangent space Tp∗ M, with respect to the induced inner product and orientation. The Hodge star operator is the linear operator on the exterior algebra of Tp∗ M defined by ∗(1) = e1 ∧ · · · ∧ en , ∗(e1 ∧ · · · ∧ en ) = 1, ∗(e1 ∧ · · · ∧ ek ) = ek+1 ∧ · · · ∧ en . We see that ∗ : Λk (Tp∗ M) → Λn−k (Tp∗ M) and satisfies ∗∗ = (−1)k(n−k) . The codifferential and the Laplacian are the operators defined by δ = (−1)n(k+1)+1 ∗ d∗ : Ω k (M) → Ω k−1 (M), = dδ + δd

: Ω k (M) → Ω k (M).

The operator is also called the Laplace–Beltrami operator and satisfies ∗ = ∗. On 2 Ω 0 (Rn ) = C ∞ (Rn ), it is simply the usual Laplacian = − ni=1 ∂ 2 . The inner product ∂xi

on forms of any degree, ·, · : Ω (M) × Ω (M) −→ R, k

k

α, β :=

α ∧ ∗β, M

Symplectic geometry

121

satisfies dα, β = α, δβ, so the codifferential δ is often denoted by d ∗ and called the adjoint33 of d. Also, is self-adjoint (i.e., α, β = α, β), and α, α = |dα|2 + |δα|2 0, where | · | is the norm with respect to this inner product. The harmonic k-forms are the elements of Hk := {α ∈ Ω k | α = 0}. Note that α = 0 if and only if dα = δα = 0. Since a harmonic form is d-closed, it defines a de Rham cohomology class. T HEOREM 3.14 (Hodge). Every de Rham cohomology class on a compact oriented Riemannian manifold (M, g) possesses a unique harmonic representative, i.e., there is an k isomorphism Hk HdeRham (M; R). In particular, the spaces Hk are finite-dimensional. We also have the following orthogonal decomposition with respect to the inner product on forms: Ω k Hk ⊕ (Ω k (M)) Hk ⊕ dΩ k−1 ⊕ δΩ k+1 . This decomposition is called the Hodge decomposition on forms. The proof of this and the next theorem involves functional analysis, elliptic differential operators, pseudodifferential operators and Fourier analysis; see for instance [62,83,141]. Here is where complex Hodge theory begins. When M is Kähler, the Laplacian satisfies ¯ (see, for example, [62]) and preserves the decomposition according to = 2(∂¯ ∂¯ ∗ + ∂¯ ∗ ∂) type, : Ω ,m → Ω ,m . Hence, harmonic forms are also bigraded Hk =

H,m

+m=k

and satisfy a Künneth formula H,m (M × N )

%

p+r=, q+s=m H

p,q (M) ⊗ Hr,s (N ).

T HEOREM 3.15 (Hodge). Every Dolbeault cohomology class on a compact Kähler manifold (M, ω) possesses a unique harmonic representative, i.e., there is an isomorphism ,m H,m HDolbeault (M). Combining the two theorems of Hodge, we find the decomposition of cohomology groups for a compact Kähler manifold k HdeRham (M; C)

,m HDolbeault (M),

+m=k

known as the Hodge decomposition. In particular, the Dolbeault cohomology groups ,m HDolbeault are finite-dimensional and H ,m H m, . k k (M) be the usual Betti numbers of M, and let h,m (M) := Let b (M) := dim HdeRham ,m dim HDolbeault (M) be the Hodge numbers of M. For an arbitrary compact symplectic manifold (M, ω), the even Betti numbers must be k positive, because ωk is closed but not

exact (k = 0, 1, . . . , n). In fact, if it were ω = dα, by Stokes’ theorem we would have M ωn = M d(α ∧ ωn−k ) = 0, which contradicts ωn being a volume form. 33 When M is not compact, we still have a formal adjoint of d with respect to the nondegenerate bilinear pairing ·, · : Ω k (M) × Ωck (M) → R defined by a similar formula, where Ωck (M) is the space of compactly supported k-forms.

122

A. Cannas da Silva

For a compact Kähler manifold (M, ω), there are finer topological consequences coming from the Hodge theorems, as we must have bk = +m=k h,m and h,m = hm, . The odd Betti numbers must be even because b2k+1 = +m=2k+1 h,m = 2 k=0 h,(2k+1−) . The number h1,0 = 12 b1 must be a topological invariant. The numbers h, are positive, , , (M). First of all, [ω ] defines an element of HDolbeault as because 0 = [ω ] ∈ HDolbeault 1,1 , ¯ ω ∈ Ω implies that ω ∈ Ω , and the closedness of ω implies that ∂ω = 0. If it were ¯ for some β ∈ Ω −1, , then ωn = ω ∧ ωn− = ∂(β ¯ ∧ ωn− ) would be ∂-exact. ¯ But ω = ∂β n,n 2n n [ω ] = 0 in HdeRham (M; C) HDolbeault (M) since it is a volume form. A popular diagram to describe relations among Hodge numbers is the Hodge diamond: hn,n hn,n−1 hn,n−2

hn−1,n hn−1,n−1

hn−2,n

.. .

... h2,0

...

h1,1 h1,0

h0,2 h0,1

h0,0 Complex conjugation gives symmetry with respect to the middle vertical, whereas the Hodge star operator induces symmetry about the center of the diamond. The middle vertical axis is all nonzero. ,m for There are further symmetries and ongoing research on how to compute HDolbeault a compact Kähler manifold (M, ω). In particular, the hard Lefschetz theorem states iso n−k n+k morphisms Lk : HdeRham (M) −→ HdeRham (M) given by wedging with ωk at the level of % m forms and the Lefschetz decompositions HdeRham (M) k Lk (ker Ln−m+2k+1 |H m−2k ). The Hodge conjecture claims, for projective manifolds M (i.e., complex submanifolds , (M) ∩ of complex projective space), that the Poincaré duals of elements in HDolbeault H 2 (M; Q) are rational linear combinations of classes of complex codimension subvarieties of M. This has been proved only for the = 1 case (it is the Lefschetz theorem on (1, 1)-classes; see, for instance, [62]).

3.6. Pseudoholomorphic curves Whereas an almost complex manifold (M, J ) tends to have no J -holomorphic functions M → C at all,34 it has plenty of J -holomorphic curves C → M. Gromov first realized that pseudoholomorphic curves provide a powerful tool in symplectic topology in an extremely influential paper [64]. Fix a closed Riemann surface (Σ, j ), that is, a compact complex 1-dimensional manifold Σ without boundary and equipped with the canonical almost complex structure j . 34 However, the study of asymptotically J -holomorphic functions has been recently developed to obtain important results [32,34,13]; see Section 4.6.

Symplectic geometry

123

D EFINITION 3.16. A parametrized pseudoholomorphic curve (or J -holomorphic curve) in (M, J ) is a (smooth) map u : Σ → M whose differential intertwines j and J , that is, dup ◦ jp = Jp ◦ dup , ∀p ∈ Σ. In other words, the Cauchy–Riemann equation du + J ◦ du ◦ j = 0 holds. Pseudoholomorphic curves are related to parametrized 2-dimensional symplectic submanifolds. If a pseudoholomorphic curve u : Σ → M is an embedding, then its image S := u(Σ) is a 2-dimensional almost complex submanifold, hence a symplectic submanifold. Conversely, the inclusion i : S → M of a 2-dimensional symplectic submanifold can be seen as a pseudoholomorphic curve. An appropriate compatible almost complex structure J on (M, ω) can be constructed starting from S, such that T S is J -invariant. The restriction j of J to T S is necessarily integrable because S is 2-dimensional. The group G of complex diffeomorphisms of (Σ, j ) acts on (parametrized) pseudoholomorphic curves by reparametrization: u → u ◦ γ , for γ ∈ G. This normally means that each curve u has a noncompact orbit under G. The orbit space Mg (A, J ) is the set of unparametrized pseudoholomorphic curves in (M, J ) whose domain Σ has genus g and whose image u(Σ) has homology class A ∈ H2 (M; Z). The space Mg (A, J ) is called the moduli space of unparametrized pseudoholomorphic curves of genus g representing the class A. For generic J , Fredholm theory shows that pseudoholomorphic curves occur in finite-dimensional smooth families, so that the moduli spaces Mg (A, J ) can be manifolds, after avoiding singularities given by multiple coverings.35 E XAMPLE . Usually Σ is the Riemann sphere CP1 , whose complex diffeomorphisms are those given by fractional linear transformations (or Möbius transformations). So the 6-dimensional noncompact group of projective linear transformations PSL(2; C) acts on pseudoholomorphic spheres by reparametrization u → u ◦ γA , where A = ac db ∈ PSL(2; C) acts by γA : CP1 → CP1 , γA [z, 1] = [ az+b cz+d , 1]. When J is an almost complex structure compatible with a symplectic form ω, the area of the image of a pseudoholomorphic curve u (with respect to the metric gJ (·, ·) = ω(·, J ·)) is determined by the class A that it represents. The number E(u) := ω(A) = u∗ ω = area of the image of u with respect to gJ Σ

is called the energy of the curve u and is a topological invariant: it only depends on [ω] and on the homotopy class of u. Gromov proved that the constant energy of all the pseudoholomorphic curves representing a homology class A ensured that the space Mg (A, J ), though not necessarily compact, had natural compactifications Mg (A, J ) by including what he called cusp-curves. T HEOREM 3.17 (Gromov’s compactness theorem). If (M, ω) is a compact manifold equipped with a generic compatible almost complex structure J , and if uj is a sequence 35 A curve u : Σ → M is a multiple covering if u factors as u = u ◦ σ where σ : Σ → Σ is a holomorphic map of degree greater than 1.

124

A. Cannas da Silva

of pseudoholomorphic curves in Mg (A, J ), then there is a subsequence that weakly converges to a cusp-curve in Mg (A, J ). Hence the cobordism class of the compactified moduli space Mg (A, J ) might be a nice symplectic invariant of (M, ω), as long as it is not empty or null-cobordant. Actually a nontrivial regularity criterion for J ensures the existence of pseudoholomorphic curves. And even when Mg (A, J ) is null-cobordant, we can define an invariant to be the (signed) number of pseudoholomorphic curves of genus g in class A that intersect a specified set of representatives of homology classes in M [112,128,145]. For more on pseudoholomorphic curves; see, for instance, [100] (for a comprehensive discussion of the genus 0 case) or [11] (for higher genus). Here is a selection of applications of (developments from) pseudoholomorphic curves: • Proof of the nonsqueezing theorem [64]: for R > r there is no symplectic embedding of a ball BR2n of radius R into a cylinder Br2 × R2n−2 of radius r, both in (R2n , ω0 ). • Proof that there are no Lagrangian spheres in (Cn , ω0 ), except for the circle in C2 , and more generally no compact exact Lagrangian submanifolds, in the sense that the tautological 1-form α restricts to an exact form [64]. • Proof that if (M, ω) is a connected symplectic 4-manifold symplectomorphic to (R4 , ω0 ) outside a compact set and containing no symplectic S 2 ’s, then (M, ω) symplectomorphic to (R4 , ω0 ) [64]. • Study questions of symplectic packing [15,98,134] such as: for a given 2n-dimensional symplectic manifold (M, ω), what is the maximal radius R for which there is a symplectic embedding of N disjoint balls BR2n into (M, ω)? • Study groups of symplectomorphisms of 4-manifolds (for a review see [97]). Gromov [64] showed that Sympl(CP2 , ωFS ) and Sympl(S 2 × S 2 , pr∗1 σ ⊕ pr∗2 σ ) deformation retract onto the corresponding groups of standard isometries. • Development of Gromov–Witten invariants allowing to prove, for instance, the nonexistence of symplectic forms on CP2 # CP2 # CP2 or the classification of symplectic structures on ruled surfaces (Section 4.3). • Development of Floer homology to prove the Arnold conjecture on the fixed points of symplectomorphisms of compact symplectic manifolds, or on the intersection of Lagrangian submanifolds (Section 5.2). • Development of symplectic field theory introduced by Eliashberg, Givental and Hofer [40] extending Gromov–Witten theory, exhibiting a rich algebraic structure and also with applications to contact geometry.

4. Symplectic geography 4.1. Existence of symplectic forms The utopian goal of symplectic classification addresses the standard questions: • (Existence) Which manifolds carry symplectic forms? • (Uniqueness) What are the distinct symplectic structures on a given manifold?

Symplectic geometry

125

even-dimensional orientable

almost complex symplectic complex Kähler

Fig. 1.

Existence is tackled through central examples in this subsection and symplectic constructions in the next two sections. Uniqueness is treated in the remainder of this subsection dealing with invariants that allow to distinguish symplectic manifolds. A Kähler structure naturally yields both a symplectic form and a complex structure (compatible ones). Either a symplectic or a complex structure on a manifold implies the existence of an almost complex structure. Figure 1 represents the relations among these structures. In dimension 2, orientability trivially guarantees the existence of all other structures, so the picture collapses. In dimension 4, the first interesting dimension, the picture above is faithful—we will see that there are closed 4-dimensional examples in each region. Closed here means compact and without boundary. Not all 4-dimensional manifolds are almost complex. A result of Wu [146] gives a necessary and sufficient condition in terms of the signature σ and the Euler characteristic χ of a 4-dimensional closed manifold M for the existence of an almost complex structure: 3σ + 2χ = h2 for some h ∈ H 2 (M; Z) congruent with the second Stiefel–Whitney class w2 (M) modulo 2. For example, S 4 and (S 2 × S 2 ) # (S 2 × S 2 ) are not almost complex. When an almost complex structure exists, the first Chern class of the tangent bundle (regarded as a complex vector bundle) satisfies the condition for h. The sufficiency of Wu’s condition is the remarkable part.36 According to Kodaira’s classification of closed complex surfaces [82], such a surface admits a Kähler structure if and only if its first Betti number b1 is even. The necessity of this condition is a Hodge relation on the Betti numbers (Section 3.5). The complex projective plane CP2 with the Fubini–Study form (Section 3.4) might be called the simplest example of a closed Kähler 4-manifold. The Kodaira–Thurston example [131] first demonstrated that a manifold that admits both a symplectic and a complex structure does not have to admit any Kähler structure. 36 Moreover, such solutions h are in one-to-one correspondence with isomorphism classes of almost complex structures.

126

A. Cannas da Silva

even-dimensional and simply connected almost complex (and simply connected) symplectic (and simply connected) complex (and simply connected)

Fig. 2.

Take R4 with dx1 ∧ dy1 + dx2 ∧ dy2 , and Γ the discrete group generated by the four symplectomorphisms: (x1 , x2 , y1 , y2 ) −→ (x1 + 1, x2 , y1 , y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 + 1, y1 , y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 + y2 , y1 + 1, y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 , y1 , y2 + 1). Then M = R4 /Γ is a symplectic manifold that is a 2-torus bundle over a 2-torus. Kodaira’s classification [82] shows that M has a complex structure. However, π1 (M) = Γ , hence H1 (R4 /Γ ; Z) = Γ /[Γ, Γ ] has rank 3, so b1 = 3 is odd. Fernández–Gotay–Gray [44] first exhibited symplectic manifolds that do not admit any complex structure at all. Their examples are circle bundles over circle bundles (i.e., a tower of circle bundles) over a 2-torus. The Hopf surface is the complex surface diffeomorphic to S 1 × S 3 obtained as the quotient C2 \ {0}/Γ where Γ = {2n Id | n ∈ Z} is a group of complex transformations, i.e., we factor C2 \ {0} by the equivalence relation (z1 , z2 ) ∼ (2z1 , 2z2 ). The Hopf surface is not symplectic because H 2 (S 1 × S 3 ) = 0. The manifold CP2 # CP2 # CP2 is almost complex but is neither complex (since it does not fit Kodaira’s classification [82]), nor symplectic as shown by Taubes [126] using Seiberg–Witten invariants (Section 4.5). We could go through the previous discussion restricting to closed 4-dimensional examples with a specific fundamental group. We will do this restricting to simply connected examples, where Figure 2 holds. It is a consequence of Wu’s result [146] that a simply connected manifold admits an almost complex structure if and only if b2+ is odd.37 In particular, the connected sum 37 The intersection form of an oriented topological closed 4-manifold M is the bilinear pairing Q : H 2 (M; Z)× M H 2 (M; Z) → Z, QM (α, β) := α ∪ β, [M], where α ∪ β is the cup product and [M] is the fundamental class.

Symplectic geometry

127

#m CP2 #n CP2 (of m copies of CP2 with n copies of CP2 ) has an almost complex structure if and only if m is odd.38 By Kodaira’s classification [82], a simply connected complex surface always admits a compatible symplectic form (since b1 = 0 is even), i.e., it is always Kähler. Since they are simply connected, S 4 , CP2 # CP2 # CP2 and CP2 live in three of the four regions in the picture for simply connected examples. All of CP2 #m CP2 are also simply connected Kähler manifolds because they are pointwise blow-ups CP2 and the blow-down map is holomorphic; see Section 4.3. There is a family of manifolds obtained from CP2 #9 CP2 =: E(1) by a knot surgery [45] that were shown by Fintushel and Stern to be symplectic and confirmed not to admit a complex structure [109]. The first example of a closed simply connected symplectic manifold that cannot be Kähler, was a 10-dimensional manifold obtained by McDuff [94] as follows. The Kodaira–Thurston example R4 /Γ (not simply connected) embeds symplectically in (CP5 , ωFS ) [65,132]. McDuff’s example is a blow-up of (CP5 , ωFS ) along the image of R4 /Γ . Geography problems are problems on the existence of simply connected closed oriented 4-dimensional manifolds with some additional structure (such as, a symplectic form or a complex structure) for each pair of topological coordinates. As a consequence of the work of Freedman [51] and Donaldson [30] in the 80’s, it became known that the homeomorphism class of a connected simply connected closed oriented smooth 4-manifold is determined by the two integers—the second Betti number and the signature (b2 , σ )— and the parity39 of the intersection form. Forgetting about the parity, the numbers (b2 , σ ) can be treated as topological coordinates. For each pair (b2 , σ ) there could well be infinite different (i.e., nondiffeomorphic) smooth manifolds. Using Riemannian geometry, Cheeger [22] showed that there are at most countably many different smooth types for closed 4-manifolds. There are no known finiteness results for the smooth types of a given topological 4-manifold, in contrast to other dimensions. Traditionally, the numbers used are (c12 , c2 ) := (3σ + 2χ, χ) = (3σ + 4 + 2b2 , 2 + b2 ), and frequently just the slope c12 /c2 is considered. If M admits an almost complex structure J , then (T M, J ) is a complex vector bundle, hence has Chern classes c1 = c1 (M, J ) and c2 = c2 (M, J ). Both c12 := c1 ∪ c1 and c2 may be regarded as numbers since H 4 (M; Z) Z. They satisfy c12 = 3σ +2χ (by Hirzebruch’s signature formula) and c2 = χ (because the top Chern class is always the Euler class), justifying the notation for the topological coordinates in this case. Since QM always vanishes on torsion elements, descending to H 2 (M; Z)/torsion it can be represented by a ma trix. When M is smooth and simply connected, this pairing is QM (α, β) := M α ∧ β since nontorsion elements are representable by 2-forms. As QM is symmetric (in the smooth case, the wedge product of 2-forms is symmetric) and unimodular (the determinant of a matrix representing QM is ±1 by Poincaré duality), it is diagonalizable over R with eigenvalues ±1. We denote by b2+ (respectively b2− ) the number of positive (respectively negative) eigenvalues of QM counted with multiplicities, i.e., the dimension of a maximal subspace where QM is positivedefinite (respectively negative-definite). The signature of M is the difference σ := b2+ − b2− , whereas the second Betti number is the sum b2 = b2+ + b2− , i.e., the rank of QM . The type of an intersection form is definite if it is positive or negative definite (i.e., |σ | = b2 ) and indefinite otherwise. 38 The intersection form of a connected sum M 0 # M1 is (isomorphic to) QM0 ⊕ QM1. 39 We say that the parity of an intersection form Q is even when Q (α, α) is even for all α ∈ H 2 (M; Z), and M M odd otherwise.

128

A. Cannas da Silva

E XAMPLES . The manifold CP2 has (b2 , σ ) = (1, 1), i.e., (c12 , c2 ) = (9, 3). Reversing the orientation CP2 has (b2 , σ ) = (1, −1), i.e., (c12 , c2 ) = (3, 3). Their connected sum CP2 # CP2 has (b2 , σ ) = (2, 0), i.e., (c12 , c2 ) = (8, 0). The product S 2 × S 2 also has (b2 , σ ) = (2, 0), i.e., (c12 , c2 ) = (8, 4). But CP2 # CP2 has an odd intersection form 0 whereas S 2 × S 2 has an even intersection form: 10 −1 vs. 01 01 . Symplectic geography [60,122] addresses the following question: What is the set of pairs of integers (m, n) ∈ Z × Z for which there exists a connected simply connected closed symplectic 4-manifold M having second Betti number b2 (M) = m and signature σ (M) = n? This problem includes the usual geography of simply connected complex surfaces, since all such surfaces are Kähler according to Kodaira’s classification [82]. Often, instead of the numbers (b2 , σ ), the question is equivalently phrased in terms of the Chern numbers (c12 , c2 ) for a compatible almost complex structure, which satisfy c12 = 3σ + 2χ [146] and c2 = χ , where χ = b2 + 2 is the Euler number. Usually only minimal (Section 4.3) or irreducible manifolds are considered to avoid trivial examples. A manifold is irreducible when it is not a connected sum of other manifolds, except when one of the summands is a homotopy sphere. It was speculated that perhaps any simply connected closed smooth 4-manifold other than S 4 is diffeomorphic to a connected sum of symplectic manifolds, where any orientation is allowed on each summand (the so-called minimal conjecture for smooth 4-manifolds). Szabó [124,125] provided counterexamples in a family of irreducible simply connected closed nonsymplectic smooth 4-manifolds. All these problems could be posed for other fundamental groups. Gompf [57] used symplectic sums (Section 4.2) to prove the following theorem. He also proved that his surgery construction can be adapted to produce non-Kähler examples. Since finitely-presented groups are not classifiable, this shows that compact symplectic 4-manifold are not classifiable. T HEOREM 4.1 (Gompf). Every finitely-presented group occurs as the fundamental group π1 (M) of a compact symplectic 4-manifold (M, ω).

4.2. Fibrations and sums Products of symplectic manifolds are naturally symplectic. As we will see, special kinds of twisted products, i.e., fibrations,40 are also symplectic. 40 A fibration (or fiber bundle) is a manifold M (called the total space) with a submersion π : M → X to a manifold X (the base) that is locally trivial in the sense that there is an open covering of X, such that, to each set U in that covering corresponds a diffeomorphism of the form ϕU = (π, sU ) : π −1 (U) → U × F (a local trivialization) where F is a fixed manifold (the model fiber). A collection of local trivializations such that the sets U cover X is called a trivializing cover for π . Given two local trivializations, the second entry of the composition −1 = (id, ψU V ) on (U ∩ V) × F gives the corresponding transition function ψU V (x) : F → F at each ϕV ◦ ϕU x ∈ U ∩ V.

Symplectic geometry

129

D EFINITION 4.2. A symplectic fibration is a fibration π : M → X where the model fiber is a symplectic manifold (F, σ ) and with a trivializing cover for which all the transition functions are symplectomorphisms F → F . In a symplectic fibration each fiber π −1 (x) carries a canonical symplectic form σx defined by the restriction of sU∗ σ , for any domain U of a trivialization covering x (i.e., x ∈ U ). A symplectic form ω on the total space M of a symplectic fibration is called compatible with the fibration if each fiber (π −1 (x), σx ) is a symplectic submanifold of (M, ω), i.e., σx is the restriction of ω to π −1 (x). E XAMPLES . 1. Every compact oriented41 fibration whose model fiber F is an oriented surface admits a structure of symplectic fibration for the following reason. Let σ0 be an area form on F . Each transition function ψUV (x) : F → F pulls σ0 back to a cohomologous area form σ1 (depending on ψUV (x)). Convex combinations σt = (1 − t)σ0 + tσ1 give a path of area forms from σ0 to σ1 with constant class [σt ]. By Moser’s argument (Section 1.4), there exists a diffeomorphism ρ(x) : F → F isotopic to the identity, depending smoothly on x ∈ U ∩ V, such that ψUV (x) ◦ ρ(x) is a symplectomorphism of (F, σ0 ). By successively adjusting local trivializations for a finite covering of the base, we can make all transition functions into symplectomorphisms. 2. Every fibration with connected base and compact fibers having a symplectic form ω for which all fibers are symplectic submanifolds admits a structure of symplectic fibration compatible with ω. Indeed, under trivializations, the restrictions of ω to the fibers give cohomologous symplectic forms in the model fiber F . So by Moser’s Theorem 1.7, all fibers are strongly isotopic to (F, σ ) where σ is the restriction of ω to a chosen fiber. These isotopies can be used to produce a trivializing cover where each sU (x) is a symplectomorphism. In the remainder of this subsection, assume that for a fibration π : M → X the total space is compact and the base is connected. For the existence of a compatible symplectic form on a symplectic fibration, a necessary condition is the existence of a cohomology class in M that restricts to the classes of the fiber symplectic forms. Thurston [131] showed that, when the base admits also a symplectic form, this condition is sufficient. Yet not all symplectic fibrations with a compatible symplectic form have a symplectic base [138]. T HEOREM 4.3 (Thurston). Let π : M → X be a compact symplectic fibration with connected symplectic base (X, α) and model fiber (F, σ ). If there is a class [ν] ∈ H 2 (M) pulling back to [σ ], then, for sufficiently large k > 0, there exists a symplectic form ωk on M that is compatible with the fibration and is in [ν + kπ ∗ α]. P ROOF. We first find a form τ on M in the class [ν] that restricts to the canonical symplectic form on each fiber. Pick a trivializing cover {ϕi = (π, si ) | i ∈ I } with contractible 41 An oriented fibration is a fibration whose model fiber is oriented and there is a trivializing cover for which all

transition functions preserve orientation.

130

A. Cannas da Silva

domains Ui . Let ρi , i ∈ I , be a partition of unity subordinate to this covering and let ρ˜i := ρi ◦ π : M → R. Since [ν] always restricts to the class of the canonical symplectic form [σx ], and the Ui ’s are contractible, on each πi−1 (Ui ) the forms si∗ σ − ν are exact. Choose 1-forms λi such that si∗ σ = ν + dλi , and set τ := ν +

d(ρ˜i λi ).

i∈I

Since τ is nondegenerate on the (vertical) subbundle given by the kernel of dπ , for k > 0 large enough the form τ + kπ ∗ α is nondegenerate on M. C OROLLARY 4.4. Let π : M → X be a compact oriented fibration with connected symplectic base (X, α) and model fiber an oriented surface F of genus g(F ) = 1. Then π admits a compatible symplectic form. P ROOF. By Example 1 above, π : M → X admits a structure of symplectic fibration with model fiber (F, σ ). Since the fiber is not a torus (g(F ) = 1), the Euler class of the tangent bundle T F (which coincides with c1 (F, σ )) is λ[σ ] for some λ = 0. Hence, the first Chern class [c] of the vertical subbundle given by the kernel of dπ (assembling the tangent bundles to the fibers) restricts to λ[σx ] on the fiber over x ∈ X. We can apply Theorem 4.3 using the class [ν] = λ−1 [c]. A pointwise connected sum M0 # M1 of symplectic manifolds (M0 , ω0 ) and (M1 , ω1 ) tends to not admit a symplectic form, even if we only require the eventual symplectic form to be isotopic to ωi on each Mi minus a ball. The reason [7] is that such a symplectic form on M0 # M1 would allow to construct an almost complex structure on the sphere formed by the union of the two removed balls, which is known not to exist except on S 2 and S 6 . Therefore: P ROPOSITION 4.5. Let (M0 , ω0 ) and (M1 , ω1 ) be two compact symplectic manifolds of dimension not 2 nor 6. Then the connected sum M0 # M1 does not admit any symplectic structure isotopic to ωi on Mi minus a ball, i = 1, 2. For connected sums to work in the symplectic category, they should be done along codimension 2 symplectic submanifolds. The following construction, already mentioned in [65], was dramatically explored and popularized by Gompf [57] (he used it to prove Theorem 4.1). Let (M0 , ω0 ) and (M1 , ω1 ) be two 2n-dimensional symplectic manifolds. Suppose that a compact symplectic manifold (X, α) of dimension 2n − 2 admits symplectic embeddings to both i0 : X → M0 , i1 : X → M1 . For simplicity, assume that the corresponding normal bundles are trivial (in general, they need to have symmetric Euler classes). By the symplectic neighborhood theorem (Theorem 1.11), there exist symplectic embeddings j0 : X × Bε → M0 and j1 : X × Bε → M1 (called framings) where Bε is a ball of radius ε and centered at the origin in R2 such that jk∗ ωk = α + dx ∧ dy and jk (p, 0) = ik (p), ∀p ∈ X, k = 0, 1. Chose an area- and orientation-preserving diffeomorphism φ of the annulus Bε \ Bδ for 0 < δ < ε that interchanges the two boundary compo-

Symplectic geometry

131

nents. Let Uk = jk (X × Bδ ) ⊂ Mk , k = 0, 1. A symplectic sum of M0 and M1 along X is defined to be M0 #X M1 := (M0 \ U0 ) ∪φ (M1 \ U1 ), where the symbol ∪φ means that we identify j1 (p, q) with j0 (p, φ(q)) for all p ∈ X and δ < |q| < ε. As ω0 and ω1 agree on the regions under identification, they induce a symplectic form on M0 #X M1 . The result depends on j0 , j1 , δ and φ. Rational blow-down is a surgery on 4-manifolds that replaces a neighborhood of a chain of embedded S 2 ’s with boundary a lens space L(n2 , n − 1) by a manifold with the same rational homology as a ball. This simplifies the homology possibly at the expense of complicating the fundamental group. Symington [123] showed that rational blow-down preserves a symplectic structure if the original spheres are symplectic surfaces in a symplectic 4-manifold.

4.3. Symplectic blow-up Symplectic blow-up is the extension to the symplectic category of the blow-up operation in algebraic geometry. It is due to Gromov according to the first printed exposition of this operation in [94]. Let L be the tautological line bundle over CPn−1 , that is, L = [p], z | p ∈ Cn \ {0}, z = λp for some λ ∈ C with projection to CPn−1 given by π : ([p], z) → [p]. The fiber of L over the point [p] ∈ CPn−1 is the complex line in Cn represented by that point. The blow-up of Cn at the ˜ n . The corresponding bloworigin is the total space of the bundle L, sometimes denoted C n down map is the map β : L → C defined by β([p], z) = z. The total space of L may be decomposed as the disjoint union of two sets: the zero section E := [p], 0 | p ∈ Cn \ {0} and S :=

[p], z | p ∈ Cn \ {0}, z = λp for some λ ∈ C∗ .

The set E is called the exceptional divisor; it is diffeomorphic to CPn−1 and gets mapped to the origin by β. On the other hand, the restriction of β to the complementary set S is a diffeomorphism onto Cn \ {0}. Hence, we may regard L as being obtained from Cn by smoothly replacing the origin by a copy of CPn−1 . Every biholomorphic map f : Cn → Cn with f (0) = 0 lifts uniquely to a biholomorphic map f˜ : L → L with f˜(E) = E. The lift is given by the formula ([f (z)], f (z)) if z = 0, ˜ f [p], z = ([p], 0) if z = 0.

132

A. Cannas da Silva

There are actions of the unitary group U(n) on L, E and S induced by the standard linear action on Cn , and the map β is U(n)-equivariant. For instance, β ∗ ω0 + π ∗ ωFS is a U(n)invariant Kähler form on L. D EFINITION 4.6. A blow-up symplectic form on the tautological line bundle L is a U(n)invariant symplectic form ω such that the difference ω − β ∗ ω0 is compactly supported, i n where ω0 = 2 k=1 dzk ∧ d z¯ k is the standard symplectic form on Cn . Two blow-up symplectic forms are equivalent if one is the pullback of the other by a U(n)-equivariant diffeomorphism of L. Guillemin and Sternberg [69] showed that two blow-up symplectic forms are equivalent if and only if they have equal restrictions to the exceptional divisor E ⊂ L. Let Ω ε (ε > 0) be the set of all blow-up symplectic forms on L whose restriction to the exceptional divisor E CPn−1 is εωFS , where ωFS is the Fubini–Study form (Section 3.4). An ε-blow-up of Cn at the origin is a pair (L, ω) with ω ∈ Ωε. Let (M, ω) be a 2n-dimensional symplectic manifold. It is a consequence of Darboux’s theorem (Theorem 1.9) that, for each point p ∈ M, there exists a complex chart (U, z1 , . . . , zn ) centered at p and with image in Cn where ω|U = 2i nk=1 dzk ∧ d z¯ k . It is shown in [69] that, for ε small enough, we can perform an ε-blow-up of M at p modeled on Cn at the origin, without changing the symplectic structure outside of a small neighborhood of p. The resulting manifold is called an ε-blow-up of M at p. As a manifold, the blow-up of M at a point is diffeomorphic to the connected sum42 M # CPn , where CPn is the manifold CPn equipped with the orientation opposite to the natural complex one. E XAMPLE . Let P(L ⊕ C) be the CP1 -bundle over CPn−1 obtained by projectivizing the direct sum of the tautological line bundle L with a trivial complex line bundle. Consider the map β : CP(L ⊕ C) −→ CPn , [p], [λp : w] −→ [λp : w], where [λp : w] on the right represents a line in Cn+1 , forgetting that, for each [p] ∈ CPn−1 , that line sits in the 2-complex-dimensional subspace L[p] ⊕ C ⊂ Cn ⊕ C. Notice that β maps the exceptional divisor E :=

[p], [0 : . . . : 0 : 1] | [p] ∈ CPn−1 CPn−1

to the point [0 : . . . : 0 : 1] ∈ CPn , and β is a diffeomorphism on the complement S :=

[p], [λp : w] | [p] ∈ CPn−1 , λ ∈ C∗ , w ∈ C CPn \ [0 : . . . : 0 : 1] .

42 The connected sum of two oriented m-dimensional manifolds M and M is the manifold, denoted M 0 1 0 # M1 , obtained from the union of those manifolds each with a small ball removed Mi \ Bi by identifying the boundaries via a (smooth) map φ : ∂B1 → ∂B2 that extends to an orientation-preserving diffeomorphism of neighborhoods of ∂B1 and ∂B2 (interchanging the inner and outer boundaries of the annuli).

Symplectic geometry

133

Therefore, we may regard CP(L ⊕ C) as being obtained from CPn by smoothly replacing the point [0 : . . . : 0 : 1] by a copy of CPn−1 . The space CP(L ⊕ C) is the blow-up of CPn at the point [0 : . . . : 0 : 1], and β is the corresponding blow-down map. The manifold CP(L ⊕ C) for n = 2 is a Hirzebruch surface. When (CPn−1 , ωFS ) is symplectically embedded in a symplectic manifold (M, ω) with image X and normal bundle isomorphic to the tautological bundle L, it can be subject to a blow-down operation. By the symplectic neighborhood theorem (Theorem 1.11), some neighborhood U ⊂ M of the image X is symplectomorphic to a neighborhood U0 ⊂ L of the zero section. It turns out that some neighborhood of ∂U0 in L is symplectomorphic to a spherical shell in (Cn , ω0 ). The blow-down of M along X is a manifold obtained from the union of M \ U with a ball in Cn . For more details, see [99, §7.1]. Following algebraic geometry, we call minimal a 2n-dimensional symplectic manifold (M, ω) without any symplectically embedded (CPn−1 , ωFS ), so that (M, ω) is not the blow-up at a point of another symplectic manifold. In dimension 4, a manifold is minimal if it does not contain any embedded sphere S 2 with self-intersection −1. Indeed, by the work of Taubes [126,129], if such a sphere S exists, then either the homology class [S] or its symmetric −[S] can be represented by a symplectically embedded sphere with self-intersection −1. For a symplectic manifold (M, ω), let i : X → M be the inclusion of a symplectic submanifold. The normal bundle N X to X in M admits a structure of complex vector bundle (as it is a symplectic vector bundle). Let P(N X) → X be the projectivization of the bundle NX → X, let Z be the zero section of N X, let L(N X) be the corresponding tautological line bundle (given by assembling the tautological line bundles over each fiber) and let β : L(N X) → N X be the blow-down map. On the exceptional divisor E :=

[p], 0 ∈ L(NX) | p ∈ N X \ Z P(N X)

the map β is just projection to the zero section Z. The restriction of β to the complement L(NX) \ E is a diffeomorphism to N X \ Z. Hence, L(N X) may be viewed as being obtained from N X by smoothly replacing each point of the zero section by the projectivization of its normal space. We symplectically identify some tubular neighborhood U of X in M with a tubular neighborhood U0 of the zero section Z in N X. A blow-up of the symplectic manifold (M, ω) along the symplectic submanifold X is the manifold obtained from the union of M \ U and β −1 (U0 ) by identifying neighborhoods of ∂U , and equipped with a symplectic form that restricts to ω on M \ U [94]. When X is one point, this construction reduces to the previous symplectic blow-up at a point. Often symplectic geography concentrates on minimal examples. McDuff [95] showed that a minimal symplectic 4-manifold with a symplectically embedded S 2 with nonnegative self-intersection is symplectomorphic either to CP2 or to an S 2 -bundle over a surface. Using Seiberg–Witten theory it was proved: T HEOREM 4.7. Let (M, ω) be a minimal closed symplectic 4-manifold. (a) (Taubes [129]) If b2+ > 1, then c12 0.

134

A. Cannas da Silva

(b) (Liu [89]) If b2+ = 1 and c12 < 0, then M is the total space of an S 2 -fibration over a surface of genus g where ω is nondegenerate on the fibers, and (c12 , c2 ) = (8 − 8g, 4 − 4g), i.e., (M, ω) is a symplectic ruled surface. A symplectic ruled surface43 is a symplectic 4-manifold (M, ω) that is the total space of an S 2 -fibration where ω is nondegenerate on the fibers. A symplectic rational surface is a symplectic 4-manifold (M, ω) that can be obtained from the standard (CP2 , ωFS ) by blowing up and blowing down. With b2+ = 1 and c12 = 0, we have symplectic manifolds CP2 #9 CP2 =: E(1), the Dolgachev surfaces E(1, p, q), the results E(1)K of surgery on a fibered knot K ⊂ S 3 , etc. With b2+ = 1 and c12 > 0, we have symplectic manifolds CP2 , S 2 × S 2 , CP2 #n CP2 for n 8 and the Barlow surface. For b2+ = 1 and c12 0, Park [109] gave a criterion for a symplectic 4-manifold to be rational or ruled in terms of Seiberg–Witten theory.

4.4. Uniqueness of symplectic forms Besides the notions listed in Section 1.4, the following equivalence relation for symplectic manifolds is considered. As it allows the cleanest statements about uniqueness, this relation is simply called equivalence. D EFINITION 4.8. Symplectic manifolds (M, ω0 ) and (M, ω1 ) are equivalent if they are related by a combination of deformation-equivalences and symplectomorphisms. Recall that (M, ω0 ) and (M, ω1 ) are deformation-equivalent when there is a smooth family ωt of symplectic forms joining ω0 to ω1 (Section 1.4), and they are symplectomorphic when there is a diffeomorphism ϕ : M → M such that ϕ ∗ ω1 = ω0 (Section 1.2). Hence, equivalence is the relation generated by deformations and diffeomorphisms. The corresponding equivalence classes can be viewed as the connected components of the moduli space of symplectic forms up to diffeomorphism. This is a useful notion when focusing on topological properties. E XAMPLES . 1. The complex projective plane CP2 has a unique symplectic structure up to symplectomorphism and scaling. This was shown by Taubes [128] relating Seiberg–Witten invariants (Section 4.5) to pseudoholomorphic curves to prove the existence of a pseudoholomorphic sphere. Previous work of Gromov [64] and McDuff [96] showed that the existence of a pseudoholomorphic sphere implies that the symplectic form is standard. Lalonde and McDuff [85] concluded similar classifications for symplectic ruled surfaces and for symplectic rational surfaces (Section 4.3). The symplectic form on 43 A (rational) ruled surface is a complex (Kähler) surface that is the total space of a holomorphic fibration over a Riemann surface with fiber CP1 . When the base is also a sphere, these are the Hirzebruch surfaces P(L ⊕ C) where L is a holomorphic line bundle over CP1 .

Symplectic geometry

135

a symplectic ruled surface is unique up to symplectomorphism in its cohomology class, and is isotopic to a standard Kähler form. In particular, any symplectic form on S 2 × S 2 is symplectomorphic to aπ1∗ σ + bπ2∗ σ for some a, b > 0 where σ is the standard area form on S 2 . Li–Liu [88] showed that the symplectic structure on CP2 #n CP2 for 2 n 9 is unique up to equivalence. 2. McMullen and Taubes [101] first exhibited simply connected closed 4-manifolds admitting inequivalent symplectic structures. Their examples were constructed using 3-dimensional topology, and distinguished by analyzing the structure of Seiberg– Witten invariants to show that the first Chern classes (Section 3.2) of the two symplectic structures lie in disjoint orbits of the diffeomorphism group. In higher dimensions there were previously examples of manifolds with inequivalent symplectic forms; see, for instance, [111]. With symplectic techniques and avoiding gauge theory, Smith [117] showed that, for each n 2, there is a simply connected closed 4-manifold that admits at least n inequivalent symplectic forms, also distinguished via the first Chern classes. It is not yet known whether there exist inequivalent symplectic forms on a 4-manifold with the same first Chern class. 4.5. Invariants for 4-manifolds Very little was known about 4-dimensional manifolds until 1981, when Freedman [51] provided a complete classification of closed simply connected topological 4-manifolds, and shortly thereafter Donaldson [30] showed that the panorama for smooth 4-manifolds was much wilder.44 Freedman showed that, modulo homeomorphism, such topological manifolds are essentially classified by their intersection forms (for an even intersection form there is exactly one class, whereas for an odd intersection form there are exactly two classes distinguished by the Kirby–Siebenmann invariant KS, at most one of which admits smooth representatives—smoothness requires KS = 0). Donaldson showed that, whereas the existence of a smooth structure imposes strong constraints on the topological type of a manifold, for the same topological manifold there can be infinite different smooth structures.45 In other words, by far not all intersection forms can occur for smooth 4-manifolds and the same intersection form may correspond to nondiffeomorphic manifolds. Donaldson’s key tool was a set of gauge-theoretic invariants, defined by counting with signs the equivalence classes (modulo gauge equivalence) of connections on SU(2)- (or SO(3)-) bundles over M whose curvature has vanishing self-dual part. For a dozen years there was hard work on the invariants discovered by Donaldson but limited advancement on the understanding of smooth 4-manifolds. 44 It had been proved by Rokhlin in 1952 that if such a smooth manifold M has even intersection form Q (i.e., M w2 = 0), then the signature of QM must be a multiple of 16. It had been proved by Whitehead and Milnor that two such topological manifolds are homotopy equivalent if and only if they have the same intersection form. 45 It is known that in dimensions 3, each topological manifold has exactly one smooth structure, and in dimensions 5 each topological manifold has at most finitely many smooth structures. For instance, whereas each topological Rn , n = 4, admits a unique smooth structure, the topological R4 admits uncountably many smooth structures.

136

A. Cannas da Silva

E XAMPLES . Finding exotic46 smooth structures on closed simply connected manifolds with small b2 has long been an interesting problem, especially in view of the smooth Poincaré conjecture for 4-manifolds. The first exotic smooth structures on a rational surface CP2 #n CP2 were found in the late 80’s for n = 9 by Donaldson [31] and for n = 8 by Kotschick [84]. There was no progress until the recent work of Park [110] constructing a symplectic exotic CP2 #7 CP2 and using this to exhibit a third distinct smooth structure CP2 #8 CP2 , thus illustrating how the existence of symplectic forms is tied to the existence of different smooth structures. This stimulated research by Fintushel, Ozsváth, Park, Stern, Stipsicz and Szabó, which together shows that there are infinitely many exotic smooth structures on CP2 #n CP2 for n = 5, 6, 7, 8 (the case n = 9 had been shown in the late 80’s by Friedman–Morgan and by Okonek–Van de Ven). In 1994 Witten brought about a revolution in Donaldson theory by introducing a new set of invariants—the Seiberg–Witten invariants—which are much simpler to calculate and to apply. This new viewpoint was inspired by developments due to Seiberg and Witten in the understanding of N = 2 supersymmetric Yang–Mills. Let M be a smooth oriented closed 4-dimensional manifold with b2+ (M) > 1 (there is a version for b2+ (M) = 1). All such 4-manifolds M (with any b2+ (M)) admit a spin-c structure, i.e., a Spinc (4)-bundle over M with an isomorphism of the associated SO(4)-bundle to the bundle of oriented frames on the tangent bundle for some chosen Riemannian metric. Let CM = {a ∈ H 2 (M; Z) | a ≡ w2 (T M)(2)} be the set of characteristic elements, and let Spinc (M) be the set of spin-c structures on M. For simplicity, assume that M is simply connected (or at least that H1 (M; Z) has no 2-torsion), so that Spinc (M) is isomorphic to CM with isomorphism given by the first Chern class of the determinant line bundle (the determinant line bundle is the line bundle associated by a natural group homomorphism Spinc (4) → U(1)). Fix an orientation of a maximal-dimensional positive-definite subspace H+2 (M; R) ⊂ H 2 (M; R). The Seiberg–Witten invariant is the function SWM : CM −→ Z defined as follows. Given a spin-c structure α ∈ Spinc (M) CM , the image SWM (α) = [M] ∈ Hd (B ∗ ; Z) is the homology class of the moduli space M of solutions (called monopoles) of the Seiberg–Witten (SW) equations modulo gauge equivalence. The SW equations are nonlinear differential equations on a pair of a connection A on the determinant line bundle of α and of a section ϕ of an associated U(2)-bundle, called the positive (half) spinor bundle: FA+ = iq(ϕ)

and DA ϕ = 0,

where FA+ is the self-dual part of the (imaginary) curvature of A, q is a squaring operation taking sections of the positive spinor bundle to self-dual 2-forms, and DA is the corresponding Dirac operator. For a generic perturbation of the equations (replacing the first equation by FA+ = iq(ϕ) + iν, where ν is a self-dual 2-form) and of the Riemannian 46 A manifold homeomorphic but not diffeomorphic to a smooth manifold M is called an exotic M.

Symplectic geometry

137

metric, a transversality argument shows that the moduli space M is well-behaved and actually inside the space B ∗ of gauge-equivalence classes of irreducible pairs (those (A, ϕ) for which ϕ = 0), which is homotopy-equivalent to CP∞ and hence has even-degree homology groups Hd (B ∗ ; Z) Z. When the dimension d of M is odd or when M is empty, the invariant SWM (α) is set to be zero. The basic classes are the classes α ∈ CM for which SWM (α) = 0. The set of basic classes is always finite, and if α is a basic class then so is −α. The main results are that the Seiberg–Witten invariants are invariants of the diffeomorphism type of the 4-manifold M and satisfy vanishing and nonvanishing theorems, which allowed to answer an array of questions about specific manifolds. Taubes [128] discovered an equivalence between Seiberg–Witten and Gromov invariants (using pseudoholomorphic curves) for symplectic 4-manifolds, by proving the existence of pseudoholomorphic curves from solutions of the Seiberg–Witten equations and vice-versa. As a consequence, he proved: T HEOREM 4.9 (Taubes). Let (M, ω) be a compact symplectic 4-manifold. If b2+ > 1, then c1 (M, ω) admits a smooth pseudoholomorphic representative. If M = M1 # M2 , then one of the Mi ’s has negative definite intersection form. There are results also for b2+ = 1, and follow-ups describe the set of basic classes of a connected sum M # N in terms of the set of basic classes of M when N is a manifold with negative definite intersection form (starting with CP2 ). In an attempt to understand other 4-manifolds via Seiberg–Witten and Gromov invariants, some analysis of pseudoholomorphic curves has been extended to nonsymplectic 4-manifolds by equipping these with a nearly nondegenerate closed 2-form. In particular, Taubes [130] has related Seiberg–Witten invariants to pseudoholomorphic curves for compact oriented 4-manifolds with b2+ > 0. Any compact oriented 4-manifold M with b2+ > 0 admits a closed 2-form that vanishes along a union of circles and is symplectic elsewhere [54,75]. In fact, for a generic metric on M, there is a self-dual harmonic form ω which is transverse to zero as a section of Λ2 T ∗ M. The vanishing locus of ω is the union of a finite number of embedded circles, and ω is symplectic elsewhere. The generic behavior of closed 2-forms on orientable 4-manifolds is partially understood [3, pp. 23–24]. Here is a summary. Let ω be a generic closed 2-form on a 4-manifold M. At the points of some hypersurface Z, the form ω has rank 2. At a generic point of M, ω is nondegenerate; in particular, has the Darboux normal form dx1 ∧ dy1 + dx2 ∧ dy2 . There is a codimension-1 submanifold Z where ω has rank 2, and there are no points where ω vanishes. At a generic point of Z, the kernel of ω˜ is transverse to Z; the normal form near such a point is x1 dx1 ∧ dy1 + dx2 ∧ dy2 . There is a curve C where the kernel of ω˜ is not transverse to Z, hence sits in T Z. At a generic point of C, the kernel of ω˜ is transverse to C; there are two possible normal forms near such points, 2 3 called elliptic and hyperbolic, d(x − z2 ) ∧ dy + d(xz ± ty − z3 ) ∧ dt. The hyperbolic and elliptic sections of C are separated by parabolic points, where the kernel is tangent to C. It is known that there exists at least one continuous family of inequivalent degeneracies in a parabolic neighborhood [56].

138

A. Cannas da Silva

4.6. Lefschetz pencils Lefschetz pencils in symplectic geometry imitate linear systems in complex geometry. Whereas holomorphic functions on a projective surface must be constant, there are interesting functions on the complement of a finite set, and generic such functions have only quadratic singularities. A Lefschetz pencil can be viewed as a complex Morse function or as a very singular fibration, in the sense that, not only some fibers are singular (have ordinary double points) but all fibers go through some points. D EFINITION 4.10. A Lefschetz pencil on an oriented 4-manifold M is a map f : M \ {b1 , . . . , bn } → CP1 defined on the complement of a finite set in M, called the base locus, that is a submersion away from a finite set {p1 , . . . , pn+1 }, and obeying local models (z1 , z2 ) → z1 /z2 near the bj ’s and (z1 , z2 ) → z1 z2 near the pj ’s, where (z1 , z2 ) are oriented local complex coordinates. Usually it is also required that each fiber contains at most one singular point. By blowing up M at the bj ’s, we obtain a map to CP1 on the whole manifold, called a Lefschetz fibration. Lefschetz pencils and Lefschetz fibrations can be defined on higher-dimensional manifolds where the bj ’s are replaced by codimension 4 submanifolds. By working on the Lefschetz fibration, Gompf [59,58] proved that a structure of Lefschetz pencil (with a nontrivial base locus) gives rise to a symplectic form, canonical up to isotopy, such that the fibers are symplectic. Using asymptotically holomorphic techniques [12,32], Donaldson [34] proved that symplectic 4-manifolds admit Lefschetz pencils. More precisely: T HEOREM 4.11 (Donaldson). Let J be a compatible almost complex structure on a compact symplectic 4-manifold (M, ω) where the class [ω]/2π is integral. Then J can be deformed through almost complex structures to an almost complex structure J such that M admits a Lefschetz pencil with J -holomorphic fibers. The closure of a smooth fiber of the Lefschetz pencil is a symplectic submanifold Poincaré dual to k[ω]/2π ; cf. Theorem 1.13. Other perspectives on Lefschetz pencils have been explored, including in terms of representations of the free group π1 (CP1 \ {p1 , . . . , pn+1 }) in the mapping class group Γg of the generic fiber surface [118]. Similar techniques were used by Auroux [13] to realize symplectic 4-manifolds as branched covers of CP2 , and thus reduce the classification of symplectic 4-manifolds to a (hard) algebraic question about factorization in the braid group. Let M and N be compact oriented 4-manifolds, and let ν be a symplectic form on N . D EFINITION 4.12. A map f : M → N is a symplectic branched cover if for any p ∈ M there are complex charts centered at p and f (p) such that ν is positive on each complex line and where f is given by: a local diffeomorphism (x, y) → (x, y), or a simple branching (x, y) → (x 2 , y), or an ordinary cusp (x, y) → (x 3 − xy, y). T HEOREM 4.13 (Auroux). Let (M, ω) be a compact symplectic 4-manifold where the class [ω] is integral, and let k be a sufficiently large integer. Then there is a symplectic

Symplectic geometry

139

branched cover fk : (M, kω) → CP2 , that is canonical up to isotopy for k large enough. Conversely, given a symplectic branched cover f : M → N , the domain M inherits a symplectic form canonical up to isotopy in the class f ∗ [ν].

5. Hamiltonian geometry 5.1. Symplectic and Hamiltonian vector fields Let (M, ω) be a symplectic manifold and let H : M → R be a smooth function. By nondegeneracy, there is a unique vector field XH on M such that ıXH ω = dH . Supposing that XH is complete (this is always the case when M is compact), let ρt : M → M, t ∈ R, be its flow (cf. Section 1.3). Each diffeomorphism ρt preserves ω, i.e., ρt∗ ω = ω, because d ∗ ∗ ∗ dt ρt ω = ρt LXH ω = ρt (dıXH ω + ıXH dω) = 0. Therefore, every function on (M, ω) produces a family of symplectomorphisms. Notice how this feature involves both the nondegeneracy and the closedness of ω. D EFINITION 5.1. A vector field XH such that ıXH ω = dH for some H ∈ C ∞ (M) is a Hamiltonian vector field with Hamiltonian function H . Hamiltonian vector fields preserve their Hamiltonian functions (LXH H = ıXH dH = field XH ıXH ıXH ω = 0), so each integral curve {ρt (x) | t ∈ R} of a Hamiltonian vector must be contained in a level set of the Hamiltonian function H . In (R2n , ω0 = dxj ∧ ∂H ∂ ∂H ∂ dyj ), the symplectic gradient XH = ( ∂y − ∂x ) and the usual (Euclidean) gradij ∂xj j ∂yj ∂H ∂ ∂H ∂ ent ∇H = j ( ∂xj ∂xj + ∂yj ∂yj ) of a function H are related by J XH = ∇H , where J is the standard almost complex structure. E XAMPLES . 1. For the height function H (θ, h) = h on the sphere (M, ω) = (S 2 , dθ ∧ dh), from ∂ . Thus, ρt (θ, h) = (θ + t, h), which is rotation ıXH (dθ ∧ dh) = dh we get XH = ∂θ about the vertical axis, preserving the height H . 2. Let X be any vector field on a manifold W . There is a unique vector field X! on the cotangent bundle T ∗ W whose flow is the lift of the flow of X. Let α be the tautological form and ω = −dα the canonical symplectic form on T ∗ W . The vector field X! is Hamiltonian with Hamiltonian function H := ıX! α. 3. Consider Euclidean space R2n with coordinates (q1 , . . . , qn , p1 , . . . , pn ) and ω0 = dqj ∧ dpj . The curve ρt = (q(t), p(t)) is an integral curve for a Hamiltonian vector field XH exactly when it satisfies the Hamilton equations: dq

∂H i dt (t) = ∂pi , dpi ∂H dt (t) = − ∂qi .

140

A. Cannas da Silva

4. Newton’s second law states that a particle of mass m moving in configuration space R3 with coordinates q = (q1 , q2 , q3 ) under a potential V (q) moves along a curve q(t) such that m

d 2q = −∇V (q). dt 2

i Introduce the momenta pi = m dq dt for i = 1, 2, 3, and energy function H (q, p) = 1 2 47 R6 = T ∗ R3 with coordinates (q , q , q , p , 1 2 3 1 2m |p| + V (q) on the phase space p2 , p3 ). The energy H is conserved by the motion and Newton’s second law in R3 is then equivalent to the Hamilton equations in R6 :

dqi dt dpi dt

=

1 ∂H m pi = ∂pi , 2 ∂V = m ddtq2i = − ∂q i

= − ∂H ∂qi .

D EFINITION 5.2. A vector field X on M preserving ω (i.e., such that LX ω = 0) is a symplectic vector field. Hence, a vector field X on (M, ω) is called symplectic when ıX ω is closed, and Hamiltonian when ıX ω is exact. In the latter case, a primitive H of ıX ω is called a Hamiltonian function of X. On a contractible open set every symplectic vector field is 1 (M) measures the obstruction for symplectic Hamiltonian. Globally, the group HdeRham vector fields to be Hamiltonian. For instance, the vector field X1 = ∂θ∂ 1 on the 2-torus (M, ω) = (T2 , dθ1 ∧ dθ2 ) is symplectic but not Hamiltonian. A vector field X is a differential operator on functions: X · f := LX f = df (X) for f ∈ C ∞ (M). As such, the bracket W = [X, Y ] is the commutator: LW = [LX , LY ] = LX LY − LY LX (cf. Section 3.3). This endows the set χ(M) of vector fields on a manifold M with a structure of Lie algebra.48 For a symplectic manifold (M, ω), using ı[X,Y ] = [LX , ıY ] and Cartan’s magic formula, we find that ı[X,Y ] ω = dıX ıY ω + ıX dıY ω − ıY dıX ω − ıY ıX dω = d(ω(Y, X)). Therefore: P ROPOSITION 5.3. If X and Y are symplectic vector fields on a symplectic manifold (M, ω), then [X, Y ] is Hamiltonian with Hamiltonian function ω(Y, X). Hence, Hamiltonian vector fields and symplectic vector fields form Lie subalgebras for the Lie bracket [·, ·]. D EFINITION 5.4. The Poisson bracket of two functions f, g ∈ C ∞ (M) is the function {f, g} := ω(Xf , Xg ) = LXg f . 47 The phase space of a system of n particles is the space parametrizing the position and momenta of the particles. The mathematical model for a phase space is a symplectic manifold. 48 A (real) Lie algebra is a (real) vector space g together with a Lie bracket [·, ·], i.e., a bilinear map [·, ·] : g × g → g satisfying antisymmetry, [x, y] = −[y, x], ∀x, y ∈ g, and the Jacobi identity, [x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0, ∀x, y, z ∈ g.

Symplectic geometry

141

By Proposition 5.3 we have X{f,g} = −[Xf , Xg ]. Moreover, the bracket {·, ·} satisfies the Jacobi identity, {f, {g, h}} + {g, {h, f }} + {h, {f, g}} = 0, and the Leibniz rule, {f, gh} = {f, g}h + g{f, h}. D EFINITION 5.5. A Poisson algebra (P, {·, ·}) is a commutative associative algebra P with a Lie bracket {·, ·} satisfying the Leibniz rule. When (M, ω) is a symplectic manifold, (C ∞ (M), {·, ·}) is a Poisson algebra, and the map C ∞ (M) → χ(M), H → XH is a Lie algebra anti-homomorphism. E XAMPLES . 1. For the prototype (R2n , dxi ∧ dyi ), we have Xxi = − ∂y∂ i and Xyi = ∂x∂ i , so that {xi , xj } = {yi , yj } = 0 and {xi , yj } = δij for all i, j . Arbitrary functions f, g ∈ C ∞ (R2n ) have the classical Poisson bracket n ∂f ∂g ∂f ∂g . − {f, g} = ∂xi ∂yi ∂yi ∂xi i=1

2. Let G be a Lie group,49 g its Lie algebra and g∗ the dual vector space of g. The vector field g X # generated by X ∈ g for the adjoint action50 of G on g has value [X, Y ] at Y ∈ g. The vector field X # generated by X ∈ g for the coadjoint action of G on g∗ is Xξ# , Y = ξ, [Y, X], ∀ξ ∈ g∗ , Y ∈ g. The skew-symmetric pairing ω on g defined at ξ ∈ g∗ by ωξ (X, Y ) := ξ, [X, Y ] has kernel at ξ the Lie algebra gξ of the stabilizer of ξ for the coadjoint action. Therefore, ω restricts to a nondegenerate 2-form on the tangent spaces to the orbits of the coadjoint action. As the tangent spaces to an orbit are generated by the vector fields X # , the Jacobi identity in g implies that this form is closed. It is called the canonical symplectic form (or the Lie–Poisson or Kirillov–Kostant–Souriau symplectic structure) on the coadjoint orbits. The corresponding Poisson structure on g∗ is the canonical one induced by the Lie bracket: {f, g}(ξ ) = ξ, [dfξ , dgξ ] 49 A Lie group is a manifold G equipped with a group structure where the group operation G × G → G and inversion G → G are smooth maps. An action of a Lie group G on a manifold M is a group homomorphism G → Diff(M), g → ψg , where the evaluation map M × G → M, (p, g) → ψg (p) is a smooth map. The orbit of G through p ∈ M is {ψg (p) | g ∈ G}. The stabilizer (or isotropy) of p ∈ M is Gp := {g ∈ G | ψg (p) = p}. 50 Any Lie group G acts on itself by conjugation: g ∈ G → ψ ∈ Diff(G), ψ (a) = g · a · g −1 . Let Ad : g → g g g g be the derivative at the identity of ψg : G → G. We identify the Lie algebra g with the tangent space Te G. For matrix groups, Adg X = gXg −1 . Letting g vary, we obtain the adjoint action of G on its Lie algebra Ad : G → GL(g). Let ·, · : g∗ × g → R be the natural pairing ξ, X = ξ(X). Given ξ ∈ g∗ , we define Ad∗g ξ by Ad∗g ξ, X = ξ, Adg −1 X, for any X ∈ g. The collection of maps Ad∗g forms the coadjoint action of G on the dual of its Lie algebra Ad∗ : G → GL(g∗ ). These satisfy Adg ◦ Adh = Adgh and Ad∗g ◦ Ad∗h = Ad∗gh .

142

A. Cannas da Silva

for f, g ∈ C ∞ (g∗ ) and ξ ∈ g∗ . The differential dfξ : Tξ g∗ g∗ → R is identified with an element of g g∗∗ .

5.2. Arnold conjecture and Floer homology There is an important generalization of Poincaré’s last geometric theorem (Theorem 2.16) conjectured by Arnold starting around 1966. Let (M, ω) be a compact symplectic manifold, and ht : M → R a 1-periodic (i.e., ht = ht+1 ) smooth family of functions. Let ρ : M × R → M be the isotopy generated by the time-dependent Hamiltonian vector field vt defined by the equation ω(vt , ·) = dht . The symplectomorphism ϕ = ρ1 is then said to be exactly homotopic to the identity. In other words, a symplectomorphism exactly homotopic to the identity is the time-1 map of the isotopy generated by some time-dependent 1-periodic Hamiltonian function. There is a one-to-one correspondence between the fixed points of ϕ and the period-1 orbits of ρ. When all the fixed points of such ϕ are nondegenerate (generic case), we call ϕ nondegenerate. The Arnold conjecture [2, Appendix 9] predicted that #{fixed points of a nondegenerate ϕ}

2n

dim H i (M; R)

i=0

(or even that the number of fixed points of a nondegenerate ϕ is at least the minimal number of critical points of a Morse function51 ). When the Hamiltonian h : M → R is independent of t, this relation is trivial: a point p is critical for h if and only if dhp = 0, if and only if vp = 0, if and only if ρ(t, p) = p, ∀t ∈ R, which implies that p is a fixed point of ρ1 = ϕ, so the Arnold conjecture reduces to a Morse inequality. Notice that, according to the Lefschetz fixed point theorem, the Euler characteristic of M, i.e., the alternating sum of the Betti numbers, (−1)i dim H i (M; R), is a (weaker) lower bound for the number of fixed points of ϕ. The Arnold conjecture was gradually proved from the late 70’s to the late 90’s by Eliashberg [39], Conley–Zehnder [24], Floer [49], Sikorav [116], Weinstein [140], Hofer– Salamon [74], Ono [108], culminating with independent proofs by Fukaya–Ono [52] and Liu–Tian [90]. There are open conjectures for sharper bounds on the number of fixed points. The breakthrough tool for establishing the Arnold conjecture was Floer homology—an ∞-dimensional analogue of Morse theory. Floer homology was defined by Floer [46–50] and developed through the work of numerous people after Floer’s death. It combines the variational approach of Conley and Zehnder [25], with Witten’s Morse– Smale complex [144], and with Gromov’s compactness theorem for pseudoholomorphic curves [64]. Floer theory starts from a symplectic action functional on the space of loops LM of a symplectic manifold (M, ω) whose zeros of the differential dF : T (LM) → R are the period-1 orbits of the isotopy ρ above. The tangent bundle T (LM) is the space of loops with vector fields over them: pairs (, v), where : S 1 → M and v : S 1 → ∗ (T M) is a 51 A Morse function is a smooth function f : M → R all of whose critical points are nondegenerate, i.e., at any critical point the Hessian matrix is nondegenerate.

Symplectic geometry

143

1 ˙ − Xht ((t), v(t)) dt. The Floer complex52 is the chain section. Then df (, v) = 0 ω((t) complex freely generated by the critical points of F (corresponding to the fixed points of ϕ), with relative grading index(x, y) given by the difference in the number of positive eigenvalues from the spectral flow. The Floer differential is given by counting the number n(x, y) of pseudoholomorphic surfaces (the gradient flow lines joining two fixed points): C∗ =

Zx and ∂x =

x∈Crit(F )

n(x, y)y.

y∈Crit(F ) index(x,y)=1

Pondering transversality, compactness and orientation, Floer’s theorem states that the homology of (C∗ , ∂) is isomorphic to the ordinary homology of M. In particular, the sum of the Betti numbers is a lower bound for the number of fixed points of ϕ. From the above symplectic Floer homology, Floer theory has branched out to tackle other differential geometric problems in symplectic geometry and 3- and 4-dimensional topology. It provides a rigorous definition of invariants viewed as homology groups of infinite-dimensional Morse-type theories, with relations to gauge theory and quantum field theory. There is Lagrangian Floer homology (for the case of Lagrangian intersections, i.e., intersection of a Lagrangian submanifold with a Hamiltonian deformation of itself), instanton Floer homology (for invariants of 3-manifolds), Seiberg–Witten Floer homology, Heegaard Floer homology and knot Floer homology. For more on Floer homology; see, for instance, [35,113].

5.3. Euler–Lagrange equations The equations of motion in classical mechanics arise from variational principles. The physical path of a general mechanical system of n particles is the path that minimizes a quantity called the action. When dealing with systems with constraints, such as the simple 52 The Morse complex for a Morse function on a compact manifold, f : M → R, is the chain complex freely

generated by the critical points of f , graded by the Morse index ı and with differential given by counting the number n(x, y) of flow lines of the negative gradient −∇f (for a metric on X) from the point x to the point y whose indices differ by 1: C∗ = Zx and ∂x = n(x, y)y. x∈Crit(f )

y∈Crit(f ) ı(y)=ı(x)−1

The coefficient n(x, y) is thus the number of solutions (modulo R-reparametrization) u : R → X of the ordinary d u(t) = −∇f (u(t)) with conditions lim differential equation dt t→−∞ u(t) = x, limt→+∞ u(t) = y. The Morse index of a critical point of f is the dimension of its unstable manifold, i.e., the number of negative eigenvalues of the Hessian of f at that point. For a generic metric, the unstable manifold of a critical point W u (x) intersects transversally with the stable manifold of another critical point W s (y). When ı(x) − ı(y) = 1, the intersection W u (x) ∩ W s (y) has dimension 1, so when we quotient out by the R-reparametrization (to count actual image curves) we get a discrete set, which is finite by compactness. That (C∗ , ∂) is indeed a complex, i.e., ∂ 2 = 0, follows from counting broken flow lines between points whose indices differ by 2. Morse’s theorem states that the homology of the Morse complex coincides with the ordinary homology of M. In particular, the sum of all the Betti numbers dim H i (M; R) is a lower bound for the number of critical points of a Morse function.

144

A. Cannas da Silva

pendulum, or two point masses attached by a rigid rod, or a rigid body, the language of variational principles becomes more appropriate than the explicit analogues of Newton’s second laws. Variational principles are due mostly to D’Alembert, Maupertius, Euler and Lagrange. Let M be an n-dimensional manifold, and let F : T M → R be a function on its tangent bundle. If γ : [a, b] → M is a curve on M, the lift of γ to T M is the curve on T M given by γ˜ : [a, b] → T M, t → (γ (t), dγ dt (t)). The action of γ is

b

Aγ :=

∗

(γ˜ F )(t) dt =

a

a

b

dγ F γ (t), (t) dt. dt

For fixed p, q, let P(a, b, p, q) = {γ : [a, b] → M smooth | γ (a) = p, γ (b) = q}. The goal is to find, among all γ ∈ P(a, b, p, q), the curve that locally minimizes Aγ . (Minimizing curves are always locally minimizing.) Assume that p, q and the image of γ lie in a coordinate neighborhood (U, x1 , . . . , xn ). On T U we have coordinates (x1 , . . . , xn , v1 , . . . , vn ) associated with a trivialization of T U by ∂x∂ 1 , . . . , ∂x∂ n . Using this trivialization, a curve γ : [a, b] → U , γ (t) = (γ1 (t), . . . , γn (t)) lifts to γ˜ : [a, b] −→ T U,

dγ1 dγn γ˜ (t) = γ1 (t), . . . , γn (t), (t), . . . , (t) . dt dt

Consider infinitesimal variations of γ . Let c1 , . . . , cn ∈ C ∞ ([a, b]) be such that ck (a) = ck (b) = 0. For ε small, let γε : [a, b] → U be the curve γε (t) = (γ1 (t) + εc1 (t), . . . , γn (t) + εcn (t)). Let Aε := Aγε . A necessary condition for γ = γ0 ∈ P(a, b, p, q) to minimize the action is that ε = 0 be a critical point of Aε . By the Leibniz rule and integration by parts, we have that dAε (0) = dε

a

= a

b ∂F k

dγ0 ∂F dγ0 dck γ0 (t), γ0 , (t) ck (t) + (t) dt ∂xk dt ∂vk dt dt

b ∂F k

d ∂F (. . .) − (. . .) ck (t) dt. ∂xk dt ∂vk

Aε (0) to vanish for all ck ’s satisfying boundary conditions ck (a) = ck (b) = 0, the For ddε path γ0 must satisfy the Euler–Lagrange equations:

d ∂F ∂F dγ0 dγ0 γ0 (t), γ0 (t), (t) = (t) , ∂xk dt dt ∂vk dt

k = 1, . . . , n.

E XAMPLES . 1. Let (M, g) be a Riemannian manifold. Let F : T M → R be the function whose restriction to each tangent space is the quadratic form defined by the Riemannian

Symplectic geometry

145

metric. On a coordinate chart F (x, v) = |v|2 = gij (x)v i v j . Let p, q ∈ M and γ : [a, b] → M a curve joining p to q. The action of γ is b 2

dγ

dt. Aγ =

a dt The Euler–Lagrange equations become the Christoffel equations for a geodesic dγ i dγ j d 2γ k k = 0, + Γ ◦ γ ij dt dt dt 2 where the Christoffel symbols Γijk ’s are defined in terms of the coefficients of the Riemannian metric (g ij is the matrix inverse to gij ) by Γijk

∂gj ∂gij 1 k ∂gi . = g + − 2 ∂xj ∂xi ∂x

2. Consider a point-particle of mass m moving in R3 under a force field G. The work

b of G on a path γ : [a, b] → R3 is Wγ := a G(γ (t)) · dγ dt (t) dt. Suppose that G is conservative, i.e., Wγ depends only on the initial and final points, p = γ (a) and q = γ (b). We can define the potential energy as V : R3 → R, V (q) := Wγ , where γ is a path joining a fixed base point p0 ∈ R3 to q. Let P be the set of all paths going from p to q over time t ∈ [a, b]. By the principle of least action, the physical path is the path γ ∈ P that minimizes a kind of mean value of kinetic minus potential energy, known as the action: Aγ :=

b a

m

dγ

2 (t) − V γ (t) dt. 2 dt

The Euler–Lagrange equations are then equivalent to Newton’s second law: m

d 2x ∂V x(t) = 0 (t) − 2 ∂x dt

⇐⇒

m

d 2x (t) = G x(t) . 2 dt

In the case of the earth moving about the sun, both regarded as point-masses and assuming that the sun to be stationary at the origin, the gravitational potentialV (x) = const |x| yields the inverse square law for the motion. 3. Consider now n point-particles of masses m1 , . . . , mn moving in R3 under a conservative force corresponding to a potential energy V ∈ C ∞ (R3n ). At any instant t, the configuration of this system is described by a vector x = (x1 , . . . , xn ) in configuration space R3n , where xk ∈ R3 is the position of the kth particle. For fixed p, q ∈ R3n , let

146

A. Cannas da Silva

P be the set of all paths γ = (γ1 , . . . , γn ) : [a, b] → R3n from p to q. The action of a path γ ∈ P is

b n mk

dγk

2 (t) − V γ (t) dt. Aγ := 2 dt a

k=1

The Euler–Lagrange equations reduce to Newton’s law for each particle. Suppose that the particles are restricted to move on a submanifold M of R3n called the constraint set. By the principle of least action for a constrained system, the physical path has minimal action among all paths satisfying the rigid constraints. I.e., we single out the actual physical path as the one that minimizes Aγ among all γ : [a, b] → M with γ (a) = p and γ (b) = q. In the case where F (x, v) does not depend on v, the Euler–Lagrange equations are sim∂F 0 (γ0 (t), dγ ply ∂x dt (t)) = 0. These are satisfied if and only if the curve γ0 sits on the critical i set of F . For generic F , the critical points are isolated, hence γ0 (t) mustbe a constant curve. In the case where F (x, v) depends affinely on v, F (x, v) = F0 (x) + nj=1 Fj (x)vj , the Euler–Lagrange equations become n dγj ∂Fj ∂Fi ∂F0 γ (t) (t). γ (t) = − ∂xi ∂xj ∂xi dt j =1

i If the n × n matrix ( ∂F ∂xj −

∂Fj ∂xi

) has an inverse Gij (x), we obtain the system of first order dγ 0 ordinary differential equations dtj (t) = Gj i (γ (t)) ∂F ∂xi (γ (t)). Locally it has a unique solution through each point p. If q is not on this curve, there is no solution at all to the Euler–Lagrange equations belonging to P(a, b, p, q). Therefore, we need nonlinear dependence of F on the v variables in order to have appropriate solutions. From now on, assume the Legendre condition:

∂ 2F det ∂vi ∂vj

= 0. 2

F Letting Gij (x, v) = ( ∂v∂i ∂v (x, v))−1 , the Euler–Lagrange equations become j

d 2 γj dγ dγ dγk ∂F ∂ 2F γ , γ , − . = G G ji ji ∂xi dt ∂vi ∂xk dt dt dt 2 i

i,k

This second order ordinary differential equation has a unique solution given initial con∂2F ditions γ (a) = p and dγ dt (a) = v. Assume that ( ∂vi ∂vj (x, v)) ' 0, ∀(x, v), i.e., with the x variable frozen, the function v → F (x, v) is strictly convex. Then the path γ0 ∈ P(a, b, p, q) satisfying the above Euler–Lagrange equations does indeed locally minimize Aγ (globally it is only critical):

Symplectic geometry

147

P ROPOSITION 5.6. For every sufficiently small subinterval [a1 , b1 ] of [a, b], γ0 |[a1 ,b1 ] is locally minimizing in P(a1 , b1 , p1 , q1 ) where p1 = γ0 (a1 ), q1 = γ0 (b1 ). P ROOF. Take c = (c1 , . . . , cn ) with ci ∈ C ∞ ([a, b]), ci (a) = ci (b) = 0. Let γε = γ0 + εc ∈ P(a, b, p, q), and let Aε = Aγε . Suppose that γ0 : [a, b] → U satisfies the Euler–Lagrange Aε equations, i.e., ddε (0) = 0. Then d 2 Aε (0) = dε 2

b

a

i,j

+2

∂ 2F dγ0 γ0 , ci cj dt ∂xi ∂xj dt

+ a

dcj ∂ 2F dγ0 γ0 , ci dt ∂xi ∂vj dt dt

(B)

∂ 2F dγ0 dci dcj γ0 , dt. ∂vi ∂vj dt dt dt

(C)

b

a

i,j

b i,j

(A)

2

F (x, v)) ' 0 at all x, v, we have Since ( ∂v∂i ∂v j

(A) KA |c|2 2 , L [a,b]

dc

L2 [a,b]

dt L2 [a,b]

(B) KB |c|

and

2

dc

(C) KC

, dt L2 [a,b] where KA , KB , KC are positive constants. By the Wirtinger inequality53 , if b − a is very small, then (C) > |(A)| + |(B)| when c ≡ 0. Hence, γ0 is a local minimum. In Section 5.1 we saw that solving Newton’s second law in configuration space R3 is equivalent to solving in phase space for the integral curve in T ∗ R3 = R6 of the Hamiltonian vector field with Hamiltonian function H . In the next subsection we will see how this correspondence extends to more general Euler–Lagrange equations. 5.4. Legendre transform The Legendre transform gives the relation between the variational (Euler–Lagrange) and the symplectic (Hamilton–Jacobi) formulations of the equations of motion. 53 The Wirtinger inequality states that, for f ∈ C 1 ([a, b]) with f (a) = f (b) = 0, we have

b

df

a dt

2 b 2

dt π |f |2 dt.

(b − a)2 a

This can be proved with Fourier series.

148

A. Cannas da Silva

Let V be an n-dimensional vector space, with e1 , . . . , en a basis of V and v1 , . . . , vn the associated coordinates. Let F : V → R, F = F (v1 , . . . , vn ), be a smooth function. The function F is strictly convex if and only if for every pair of elements p, v ∈ V , v = 0, the restriction of F to the line {p + xv | x ∈ R} is strictly convex.54 It follows from the case of real functions on R that, for a strictly convex function F on V , the following are equivalent:55 (a) F has a critical point, i.e., a point where dFp = 0; (b) F has a local minimum at some point; (c) F has a unique critical point (global minimum); and (d) F is proper, that is, F (p) → +∞ as p → ∞ in V . A strictly convex function F is stable when it satisfies conditions (a)–(d) above. D EFINITION 5.7. The Legendre transform associated to F ∈ C ∞ (V ) is the map LF : V −→ V ∗ , p −→ dFp ∈ Tp∗ V V ∗ , where Tp∗ V V ∗ is the canonical identification for a vector space V . From now on, assume that F is a strictly convex function on V . Then, for every point p ∈ V , LF maps a neighborhood of p diffeomorphically onto a neighborhood of LF (p). Given ∈ V ∗ , let F : V −→ R,

F (v) = F (v) − (v).

Since (d 2 F )p = (d 2 F )p , F is strictly convex if and only if F is strictly convex. The stability set of F is SF = { ∈ V ∗ | F is stable}. The set SF is open and convex, and LF maps V diffeomorphically onto SF . (A way to ensure that SF = V ∗ and hence that LF maps V diffeomorphically onto V ∗ , is to assume that a strictly convex function F has quadratic growth at infinity, i.e., there exists a positive-definite quadratic form Q on V and a constant K such that F (p) Q(p) − K, for all p.) The inverse to LF is the map L−1 F : SF → V described as follows: for ∈ SF , 54 2 nA function F : V → R is strictly convex if at every p ∈ V the Hessian d Fp is positive definite. Let u = u e ∈ V . The Hessian of F at p is the quadratic function on V , i i i=1

∂2F

d2 (p)ui uj = 2 F (p + tu)

. (d 2 F )p (u) := ∂vi ∂vj dt t=0 i,j

55 A smooth function f : R → R is strictly convex if f (x) > 0 for all x ∈ R. Assuming that f is strictly convex, the following four conditions are equivalent: f (x) = 0 at some point, f has a local minimum, f has a unique

(global) minimum, and f (x) → +∞ as x → ±∞. The function f is stable if it satisfies one (and hence all) of these conditions. For instance, ex + ax is strictly convex for any a ∈ R, but it is stable only for a < 0. The function x 2 + ax is strictly convex and stable for any a ∈ R.

Symplectic geometry

149

the value L−1 F () is the unique minimum point p ∈ V of F . Indeed p is the minimum of F (v) − dFp (v). D EFINITION 5.8. The dual function F ∗ to F is F ∗ : SF −→ R,

F ∗ () = − min F (p). p∈V

The dual function F ∗ is smooth and, for all p ∈ V and all ∈ SF , satisfies the Young inequality F (p) + F ∗ () (p). On one hand we have V × V ∗ T ∗ V , and on the other hand, since V = V ∗∗ , we have V × V ∗ V ∗ × V T ∗ V ∗ . Let α1 be the tautological 1-form on T ∗ V and α2 be the tautological 1-form on T ∗ V ∗ . Via the identifications above, we can think of both of these forms as living on V × V ∗ . Since α1 = dβ − α2 , where β : V × V ∗ → R is the function β(p, ) = (p), we conclude that the forms ω1 = −dα1 and ω2 = −dα2 satisfy ω1 = −ω2 . ∗ T HEOREM 5.9. For a strictly convex function F we have that L−1 F = LF .

P ROOF. The graph ΛF of the Legendre transform LF is a Lagrangian submanifold of V × V ∗ with respect to the symplectic form ω1 . Hence, ΛF is also Lagrangian for ω2 . Let pr1 : ΛF → V and pr2 : ΛF → V ∗ be the restrictions of the projection maps V × V ∗ → V and V × V ∗ → V ∗ , and let i : ΛF → V × V ∗ be the inclusion map. Then i ∗ α1 = d(pr1 )∗ F as both sides have value dFp at (p, dFp ) ∈ ΛF . It follows that i ∗ α2 = d(i ∗ β − (pr1 )∗ F ) = d(pr2 )∗ F ∗ , which shows that ΛF is the graph of the inverse of LF ∗ . From this we conclude that the inverse of the Legendre transform associated with F is the Legendre transform associated with F ∗ . Let M be a manifold and F

: T M → R. We return to the Euler–Lagrange equations for minimizing the action Aγ = γ˜ ∗ F . At p ∈ M, let Fp := F |Tp M : Tp M → R. Assume that Fp is strictly convex for all p ∈ M. To simplify notation, assume also that SFp = Tp∗ M. The

Legendre transform on each tangent space LFp : Tp M −→ Tp∗ M is essentially given by the first derivatives of F in the v directions. Collect these and the dual functions Fp∗ : Tp∗ M → R into maps L : T M −→ T ∗ M,

L|Tp M = LFp

and H : T ∗ M −→ R,

H |Tp∗ M = Fp∗ .

The maps H and L are smooth, and L is a diffeomorphism. T HEOREM 5.10. Let γ : [a, b] → M be a curve, and γ˜ : [a, b] → T M its lift. Then γ satisfies the Euler–Lagrange equations on every coordinate chart if and only if L ◦ γ˜ : [a, b] → T ∗ M is an integral curve of the Hamiltonian vector field XH .

150

A. Cannas da Silva

P ROOF. Let (U, x1 , . . . , xn ) be a coordinate chart in M, with associated tangent (T U, x1 , . . . , xn , v1 , . . . , vn ) and cotangent (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ) coordinates. On T U we have F = F (x, v), on T ∗ U we have H = H (x, ξ ), and L : T U −→ T ∗ U,

H : T ∗ U −→ R, (x, ξ ) −→ Fx∗ (ξ ) = ξ · v − F (x, v),

(x, v) −→ (x, ξ ),

where ξ := LFx (v) = ∂F ∂v (x, v) is called the momentum. Integral curves (x(t), ξ(t)) of XH satisfy the Hamilton equations: dx (H)

dt dξ dt

=

∂H ∂ξ (x, ξ ), = − ∂H ∂x (x, ξ ),

whereas the physical path x(t) satisfies the Euler–Lagrange equations: (E–L)

dx d ∂F dx ∂F x, = x, . ∂x dt dt ∂v dt

Let (x(t), ξ(t)) = L(x(t), dx dt (t)). For an arbitrary curve x(t), we want to prove that t → (x(t), ξ(t)) satisfies (H) if and only if t → (x(t), dx dt (t)) satisfies (E–L). The first line of (H) comes automatically from the definition of ξ : ξ = LF x

dx dt

⇐⇒

dx ∂H ∗ = L−1 (x, ξ ). Fx (ξ ) = LFx (ξ ) = dt ∂ξ

If (x, ξ ) = L(x, v), by differentiating both sides of H (x, ξ ) = ξ · v − F (x, v) with respect to x, where ξ = LFx (v) = ξ(x, v) and v = ∂H ∂ξ , we obtain ∂H ∂H ∂ξ ∂ξ ∂F + = ·v− ∂x ∂ξ ∂x ∂x ∂x

⇐⇒

∂F ∂H (x, v) = − (x, ξ ). ∂x ∂x

Using the last equation and the definition of ξ , the second line of (H) becomes (E–L): dξ ∂H =− (x, ξ ) dt ∂x

⇐⇒

d ∂F ∂F (x, v) = (x, v). dt ∂v ∂x

5.5. Integrable systems D EFINITION 5.11. A Hamiltonian system is a triple (M, ω, H ), where (M, ω) is a symplectic manifold and H ∈ C ∞ (M) is the Hamiltonian function. P ROPOSITION 5.12. For a function f on a symplectic manifold (M, ω) we have that {f, H } = 0 if and only if f is constant along integral curves of XH .

Symplectic geometry

151

P ROOF. Let ρt be the flow of XH . Then d (f ◦ ρt ) = ρt∗ LXH f = ρt∗ ıXH df = ρt∗ ıXH ıXf ω = ρt∗ ω(Xf , XH ) dt = ρt∗ {f, H }.

A function f as in Proposition 5.12 is called an integral of motion (or a first integral or a constant of motion). In general, Hamiltonian systems do not admit integrals of motion that are independent of the Hamiltonian function. Functions f1 , . . . , fn are said to be independent if their differentials (df1 )p , . . . , (dfn )p are linearly independent at all points p in some dense subset of M. Loosely speaking, a Hamiltonian system is (completely) integrable if it has as many commuting integrals of motion as possible. Commutativity is with respect to the Poisson bracket. If f1 , . . . , fn are commuting integrals of motion for a Hamiltonian system (M, ω, H ), then ω(Xfi , Xfj ) = {fi , fj } = 0, so at each p ∈ M the Hamiltonian vector fields generate an isotropic subspace of Tp M. When f1 , . . . , fn are independent, by symplectic linear algebra n can be at most half the dimension of M. D EFINITION 5.13. A Hamiltonian system (M, ω, H ) where M is a 2n-dimensional manifold is (completely) integrable if it possesses n independent commuting integrals of motion, f1 = H, f2 , . . . , fn . Any 2-dimensional Hamiltonian system (where the set of nonfixed points is dense) is trivially integrable. Basic examples are the simple pendulum and the harmonic oscillator. A Hamiltonian system (M, ω, H ) where M is 4-dimensional is integrable if there is an integral of motion independent of H (the commutativity condition is automatically satisfied). A basic example is the spherical pendulum. Sophisticated examples of integrable systems can be found in [8,72]. E XAMPLES . 1. The simple pendulum is a mechanical system consisting of a massless rigid rod of length , fixed at one end, whereas the other end has a bob of mass m, which may oscillate in the vertical plane. We assume that the force of gravity is constant pointing vertically downwards and the only external force acting on this system. Let θ be the oriented angle between the rod and the vertical direction. Let ξ be the coordinate along the fibers of T ∗ S 1 induced by the standard angle coordinate on S 1 . ξ2 The energy function H : T ∗ S 1 → R, H (θ, ξ ) = 2m 2 + m(1 − cos θ ), is an appropriate Hamiltonian function to describe the simple pendulum. Gravity is responsible for the potential energy V (θ ) = m(1 − cos θ ), and the kinetic energy is given by 1 2 K(θ, ξ ) = 2m 2ξ . 2. The spherical pendulum consists of a massless rigid rod of length , fixed at one end, whereas the other end has a bob of mass m, which may oscillate freely in all directions. For simplicity let m = = 1. Again assume that gravity is the only external force. Let ϕ, θ (0 < ϕ < π , 0 < θ < 2π ) be spherical coordinates for the bob, inducing coordinates η, ξ along the fibers of T ∗ S 2 . An appropriate Hamiltonian function for this system is the energy function H : T ∗ S 2 → R, H (ϕ, θ, η, ξ ) =

152

A. Cannas da Silva 2

+ (sinξ ϕ)2 ) + cos ϕ. The function J (ϕ, θ, η, ξ ) = ξ is an independent integral of motion corresponding to the group of symmetries given by rotations about the vertical axis (Section 5.6). The points p ∈ T ∗ S 2 where dHp and dJp are linearly dependent are: • the two critical points of H (where both dH and dJ vanish); • if x ∈ S 2 is in the southern hemisphere (x3 < 0), then there exist exactly two points, p+ = (x, η, ξ ) and p− = (x, −η, −ξ ), in the cotangent fiber above x where dHp and dJp are linearly dependent; • since dHp and dJp are linearly dependent along the trajectory of the Hamiltonian vector field of H through p+ , this trajectory is also a trajectory of the Hamiltonian vector field of J and hence its projection onto S 2 is a latitudinal (or horizontal) circle. The projection of the trajectory through p− is the same latitudinal circle traced in the opposite direction. 1 2 2 (η

Let (M, ω, H ) be an integrable system of dimension 2n with integrals of motion f1 = H, f2 , . . . , fn . Let c ∈ Rn be a regular value of f := (f1 , . . . , fn ). The corresponding level set f −1 (c) is a Lagrangian submanifold, as it is n-dimensional and its tangent bundle is isotropic. If the flows are complete on f −1 (c), by following them we obtain global coordinates. Any compact component of f −1 (c) must hence be a torus. These components, when they exist, are called Liouville tori. A way to ensure that compact components exist is to have one of the fi ’s proper. T HEOREM 5.14 (Arnold–Liouville [2]). Let (M, ω, H ) be an integrable system of dimension 2n with integrals of motion f1 = H, f2 , . . . , fn . Let c ∈ Rn be a regular value of f := (f1 , . . . , fn ). The level f −1 (c) is a Lagrangian submanifold of M. (a) If the flows of the Hamiltonian vector fields Xf1 , . . . , Xfn starting at a point p ∈ f −1 (c) are complete, then the connected component of f −1 (c) containing p is a homogeneous space for Rn , i.e., is of the form Rn−k × Tk for some k, 0 k n, where Tk is a k-dimensional torus.. With respect to this affine structure, that component has coordinates ϕ1 , . . . , ϕn , known as angle coordinates, in which the flows of Xf1 , . . . , Xfn are linear. (b) There are coordinates ψ1 , . . . , ψn , known as action coordinates, complementary to the angle coordinates, such that the ψi ’s are integrals of motion and ϕ1 , . . . , ϕn , ψ1 , . . . , ψn form a Darboux chart. Therefore, the dynamics of an integrable system has a simple explicit solution in actionangle coordinates. The proof of part (a)—the easy part of the theorem—is sketched above. For the proof of part (b) see, for instance, [2,36]. Geometrically, regular levels being Lagrangian submanifolds implies that, in a neighborhood of a regular value, the map f : M → Rn collecting the given integrals of motion is a Lagrangian fibration, i.e., it is locally trivial and its fibers are Lagrangian submanifolds. Part (a) states that there are coordinates along the fibers, the angle coordinates,56 in which the flows of Xf1 , . . . , Xfn are linear. Part (b) guarantees the existence of coordinates on Rn , the action coordinates, 56 The name angle coordinates is used even if the fibers are not tori.

Symplectic geometry

153

ψ1 , . . . , ψn , complementary to the angle coordinates, that (Poisson) commute among themselves and satisfy {ϕi , ψj } = δij . The action coordinates are generally not the given integrals of motion because ϕ1 , . . . , ϕn , f1 , . . . , fn do not form a Darboux chart. 5.6. Symplectic and Hamiltonian actions Let (M, ω) be a symplectic manifold, and G a Lie group. D EFINITION 5.15. An action57 ψ : G → Diff(M), g → ψg , is a symplectic action if each ψg is a symplectomorphism, i.e., ψ : G → Sympl(M, ω) ⊂ Diff(M). In particular, symplectic actions of R on (M, ω) are in one-to-one correspondence with complete symplectic vector fields on M:

dψt (p)

, p ∈ M. ψ = exp tX ←→ Xp = dt t=0 We may define a symplectic action ψ of S 1 or R on (M, ω) to be Hamiltonian if the vector field X generated by ψ is Hamiltonian, that is, when there is H : M → R with dH = ıX ω. An action of S 1 may be viewed as a periodic action of R. E XAMPLES . 1. On (R2n , ω0 ), the orbits of the action generated by X = − ∂y∂ 1 are lines parallel to the y1 -axis, {(x1 , y1 − t, x2 , y2 , . . . , xn , yn ) | t ∈ R}. Since X is Hamiltonian with Hamiltonian function x1 , this is a Hamiltonian action of R. 2. On the 2-sphere (S 2 , dθ ∧ dh) in cylindrical coordinates, the one-parameter group of diffeomorphisms given by rotation around the vertical axis, ψt (θ, h) = (θ + t, h) (t ∈ R) is a symplectic action of the group S 1 R/2π, as it preserves the area form dθ ∧ dh. Since the vector field corresponding to ψ is Hamiltonian with Hamiltonian function h, this is a Hamiltonian action of S 1 . When G is a product of S 1 ’s or R’s, an action ψ : G → Sympl(M, ω) is called Hamiltonian when the restriction to each 1-dimensional factor is Hamiltonian in the previous sense with Hamiltonian function preserved by the action of the rest of G. For an arbitrary Lie group G, we use an upgraded Hamiltonian function µ, known as a moment map, determined up to an additive local constant by coordinate functions µi indexed by a basis of the Lie algebra of G. We require that the constant be such that µ is equivariant, i.e., µ intertwines the action of G on M and the coadjoint action of G on the dual of its Lie algebra. (If M is compact, equivariance can be achieved by adjusting the

constant so that M µωn = 0. Similarly when there is a fixed point p (on each component of M) by imposing µ(p) = 0.) Let G be a Lie group, g the Lie algebra of G, and g∗ the dual vector space of g. 57 A (smooth) action of G on M is a group homomorphism G → Diff(M), g → ψ , whose evaluation map g M × G → M, (p, g) → ψg (p), is smooth.

154

A. Cannas da Silva

D EFINITION 5.16. An action ψ : G → Diff(M) on a symplectic manifold (M, ω) is a Hamiltonian action if there exists a map µ : M → g∗ satisfying: • For each X ∈ g, we have dµX = ıX# ω, i.e., µX is a Hamiltonian function for the vector field X # , where – µX : M → R, µX (p) := µ(p), X, is the component of µ along X, – X # is the vector field on M generated by the one-parameter subgroup {exp tX | t ∈ R} ⊆ G. • The map µ is equivariant with respect to the given action ψ on M and the coadjoint action: µ ◦ ψg = Ad∗g ◦µ, for all g ∈ G. Then (M, ω, G, µ) is a Hamiltonian G-space and µ is a moment map. This definition matches the previous one when G is an Abelian group R, S 1 or Tn , for which equivariance becomes invariance since the coadjoint action is trivial. E XAMPLES . 1. Let Tn = {(t1 , . . . , tn ) ∈ Cn : |tj | = 1, for all j } be a torus acting on Cn by (t1 , . . . , tn ) · (z1 , . . . , zn ) = (t1k1 z1 , . . . , tnkn zn ), where k1 , . . . , kn ∈ Z are fixed. This action is Hamiltonian with a moment map µ : Cn → (tn )∗ Rn , µ(z1 , . . . , zn ) = − 12 (k1 |z1 |2 , . . . , kn |zn |2 ). 2. When a Lie group G acts on two symplectic manifolds (Mj , ωj ), j = 1, 2, with moment maps µj : Mj → g∗ , the diagonal action of G on M1 × M2 has moment map µ : M1 × M2 → g∗ , µ(p1 , p2 ) = µ1 (p1 ) + µ2 (p2 ). 3. Equip the coadjoint orbits of a Lie group G with the canonical symplectic form (Section 5.1). Then, for each ξ ∈ g∗ , the coadjoint action on the orbit G · ξ is Hamiltonian with moment map simply the inclusion map µ : G · ξ → g∗ . 4. Identify the Lie algebra of the unitary group U(n) with its dual via the inner product A, B = trace(A∗ B). The natural action of U(n) on (Cn , ω0 ) is Hamiltonian with moment map µ : Cn → u(n) given by µ(z) = 2i zz∗ . Similarly, a moment map for the natural action of U(k) on the space (Ck×n , ω0 ) of complex (k × n)-matrices is given by µ(A) = 2i AA∗ for A ∈ Ck×n . Thus the U(n)-action by conjugation on the space 2

(Cn , ω0 ) of complex (n × n)-matrices is Hamiltonian, with moment map given by µ(A) = 2i [A, A∗ ]. 5. For the spherical pendulum (Section 5.5), the energy-momentum map (H, J ) : T ∗ S 2 → R2 is a moment map for the R × S 1 action given by time flow and rotation about the vertical axis. 6. Suppose that a compact Lie group acts on a symplectic manifold (M, ω) in a Hamiltonian way, and that q ∈ M is a fixed point for the G-action. Then, by an equivariant version of Darboux’s theorem,58 there exists a Darboux chart (U, z1 , . . . , zn ) centered at q that is G-equivariant with respect to a linear action of G on Cn . Consider an ε-blow-up of M relative to this chart, for ε sufficiently small. Then G acts on the blow-up in a Hamiltonian way. 58 Equivariant Darboux theorem [136]. Let (M, ω) be a 2n-dimensional symplectic manifold equipped with

a symplectic action of a compact Lie group G, and let q be a fixed point. Then there exists a G-invariant chart

Symplectic geometry

155

The concept of a moment map was introduced by Souriau [119] under the French name application moment; besides the more standard English translation to moment map, the alternative momentum map is also used, and recently James Stasheff has proposed the short unifying new word momap. The name comes from being the generalization of linear and angular momenta in classical mechanics. Let R3 act on (R6 T ∗ R3 , ω0 = dxi ∧ dyi ) by translations: a ∈ R3 −→ ψa ∈ Sympl R6 , ω0 ,

ψa (x, y) = (x + a, y).

The vector field generated by X = a = (a1 , a2 , a3 ) is X # = a1 ∂x∂ 1 + a2 ∂x∂ 2 + a3 ∂x∂ 3 , and the linear momentum map µ : R6 −→ R3 ,

µ(x, y) = y

is a moment map, with µa (x, y) = µ(x, y), a = y · a. Classically, y is called the momentum vector corresponding to the position vector x. The SO(3)-action on R3 by rotations lifts to a symplectic action ψ on the cotangent bundle R6 . The infinitesimal version of this action is59 a ∈ R3 −→ dψ(a) ∈ χ sympl R6 ,

dψ(a)(x, y) = (a × x, a × y).

Then the angular momentum map µ : R6 −→ R3 ,

µ(x, y) = x × y

is a moment map, with µa (x, y) = µ(x, y), a = (x × y) · a. The notion of a moment map associated to a group action on a symplectic manifold formalizes the Noether principle, which asserts that there is a one-to-one correspondence between symmetries (or one-parameter group actions) and integrals of motion (or conserved quantities) for a mechanical system. (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and G-equivariant with respect to a linear action of G on R2n such that ω|U =

n

dxk ∧ dyk .

k=1

A suitable linear action on R2n is equivalent to the induced action of G on Tq M. The proof relies on an equivariant version of the Moser trick and may be found in [70]. 59 The Lie group SO(3) = {A ∈ GL(3; R) | At A = Id and det A = 1}, has Lie algebra, g = {A ∈ gl(3; R) | A + At = 0}, the space of 3 × 3 skew-symmetric matrices. The standard identification of g with R3 carries the Lie bracket to the exterior product: 0 −a3 a2 A = a3 0 −a1 −→ a = (a1 , a2 , a3 ), −a2 a1 0 [A, B] = AB − BA −→ a × b.

156

A. Cannas da Silva

D EFINITION 5.17. An integral of motion of a Hamiltonian G-space (M, ω, G, µ) is a G-invariant function f : M → R. When µ is constant on the trajectories of a Hamiltonian vector field Xf , the corresponding flow {exp tXf | t ∈ R} (regarded as an R-action) is a symmetry of the Hamiltonian G-space (M, ω, G, µ). T HEOREM 5.18 (Noether). Let (M, ω, G, µ) be a Hamiltonian G-space where G is connected. If f is an integral of motion, the flow of its Hamiltonian vector field Xf is a symmetry. If the flow of some Hamiltonian vector field Xf is a symmetry, then a corresponding Hamiltonian function f is an integral of motion. P ROOF. Let µX = µ, X : M → R for X ∈ g. We have LXf µX = ıXf dµX = ıXf ıX# ω = −ıX# ıXf ω = −ıX# df = −LX# f . So µ is invariant over the flow of Xf if and only if f is invariant under the infinitesimal G-action. We now turn to the questions of existence and uniqueness of moment maps. Let g be a Lie algebra, and let C k := Λk g∗ be the set of k-cochains on g, that is, of alternating k-linear maps g × · · · × g → R. The linear operator δ : C k → C k+1 defined by δc(X0 , . . . , Xk ) = i<j (−1)i+j c([Xi , Xj ], X0 , . . . , Xˆ i , . . . , Xˆ j , . . . , Xk ) satisfies δ 2 = 0. The Lie algebra cohomology groups (or Chevalley cohomology groups) of g are the cohoδ

δ

δ

mology groups of the complex 0 −→ C 0 −→ C 1 −→ · · ·: H k (g; R) :=

ker δ : C k → C k+1 . im δ : C k−1 → C k

It is always H 0 (g; R) = R. If c ∈ C 1 = g∗ , then δc(X, Y ) = −c([X, Y ]). The commutator ideal [g, g] is the subspace of g spanned by {[X, Y ] | X, Y ∈ g}. Since δc = 0 if and only if c vanishes on [g, g], we conclude that H 1 (g; R) = [g, g]0 , where [g, g]0 ⊆ g∗ is the annihilator of [g, g]. An element of C 2 is an alternating bilinear map c : g × g → R, and δc(X, Y, Z) = −c([X, Y ], Z) + c([X, Z], Y ) − c([Y, Z], X). If c = δb for some b ∈ C 1 , then c(X, Y ) = (δb)(X, Y ) = −b([X, Y ]). If g is the Lie algebra of a compact connected Lie group G, then by averaging one can show that the de Rham cohomology may be computed from the subcomplex of G-invariant k (G). forms, and hence H k (g; R) = HdeRham P ROPOSITION 5.19. If H 1 (g; R) = H 2 (g, R) = 0, then any symplectic G-action is Hamiltonian. P ROOF. Let ψ : G → Sympl(M, ω) be a symplectic action of G on a symplectic manifold (M, ω). Since H 1 (g; R) = 0 means that [g, g] = g, and since commutators of symplectic vector fields are Hamiltonian, we have dψ : g = [g, g] → χ ham (M). The action ψ is Hamiltonian if and only if there is a Lie algebra homomorphism µ∗ : g → C ∞ (M) such that the Hamiltonian vector field of µ∗ (ξ ) is dψ(ξ ). We first take an arbitrary vector space lift τ : g → C ∞ (M) with this property, i.e., for each basis vector X ∈ g, we choose τ (X) = τ X ∈ C ∞ (M) such that v(τ X ) = dψ(X). The map X → τ X may not be a Lie algebra homomorphism. By construction, τ [X,Y ] is a Hamiltonian function for [X, Y ]# , and

Symplectic geometry

157

(as computed in Section 5.5) {τ X , τ Y } is a Hamiltonian function for −[X # , Y # ]. Since [X, Y ]# = −[X # , Y # ], the corresponding Hamiltonian functions must differ by a constant: τ [X,Y ] − τ X , τ Y = c(X, Y ) ∈ R. By the Jacobi identity, δc = 0. Since H 2 (g; R) = 0, there is b ∈ g∗ satisfying c = δb, c(X, Y ) = −b([X, Y ]). We define µ∗ : g −→ C ∞ (M), X −→ µ∗ (X) = τ X + b(X) = µX . Now µ∗ is a Lie algebra homomorphism: µ∗ ([X, Y ]) = {τ X , τ Y } = {µX , µY }.

By the Whitehead lemmas (see, for instance, [77, pp. 93–95]) a semisimple Lie group G has H 1 (g; R) = H 2 (g; R) = 0. As a corollary, when G is semisimple, any symplectic G-action is Hamiltonian.60 P ROPOSITION 5.20. For a connected Lie group G, if H 1 (g; R) = 0, then moment maps for Hamiltonian G-actions are unique. P ROOF. Suppose that µ1 and µ2 are two moment maps for an action ψ . For each X ∈ g, X X X # µX 1 and µ2 are both Hamiltonian functions for X , thus µ1 − µ2 = c(X) is locally con∗ stant. This defines c ∈ g , X → c(X). Since the corresponding µ∗i : g → C ∞ (M) are Lie algebra homomorphisms, we have c([X, Y ]) = 0, ∀X, Y ∈ g, i.e., c ∈ [g, g]0 = {0}. Hence, µ1 = µ 2 . In general, if µ : M → g∗ is a moment map, then given any c ∈ [g, g]0 , µ1 = µ + c is another moment map. In other words, moment maps are unique up to elements of the dual of the Lie algebra that annihilate the commutator ideal. The two extreme cases are when • G is semisimple: any symplectic action is Hamiltonian, moment maps are unique; • G is Abelian: symplectic actions may not be Hamiltonian, moment maps are unique up to a constant c ∈ g∗ . 60 A compact Lie group G has H 1 (g; R) = H 2 (g; R) = 0 if and only if it is semisimple. In fact, a compact Lie

group G is semisimple when g = [g, g]. The unitary group U(n) is not semisimple because the multiples of the identity, S 1 · Id, form a nontrivial center; at the level of the Lie algebra, this corresponds to the subspace R · Id of scalar matrices, which are not commutators since they are not traceless. Any Abelian Lie group is not semisimple. Any direct product of the other compact classical groups SU(n), SO(n) and Sp(n) is semisimple. An arbitrary compact Lie group admits a finite cover by a direct product of tori and semisimple Lie groups.

158

A. Cannas da Silva

5.7. Convexity Atiyah, Guillemin and Sternberg [4,68] showed that the image of the moment map for a Hamiltonian torus action on a compact connected symplectic manifold is always a polytope.61 A proof of this theorem can also be found in [99]. T HEOREM 5.21 (Atiyah, Guillemin–Sternberg). Let (M, ω) be a compact connected symplectic manifold. Suppose that ψ : Tm → Sympl(M, ω) is a Hamiltonian action of an m-torus with moment map µ : M → Rm . Then: (a) the levels µ−1 (c) are connected (c ∈ Rm ); (b) the image µ(M) is convex; (c) µ(M) is the convex hull of the images of the fixed points of the action. The image µ(M) of the moment map is called the moment polytope. E XAMPLES . 1. Suppose that Tm acts linearly on (Cn , ω0 ). Let λ(1) , . . . , λ(n) ∈ Zm be the weights appearing in the corresponding weight space decomposition, that is, Cn

n

Vλ(k) ,

k=1 (k)

(k)

where, for λ(k) = (λ1 , . . . , λm ), the torus Tm acts on the complex line Vλ(k) by

(k) j λj tj

(eit1 , . . . , eitm ) · v = ei v. If the action is effective62 , then m n and the (1) (n) weights λ , . . . , λ are part of a Z-basis of Zm . If the action is symplectic (hence Hamiltonian in this case), then the weight spaces Vλ(k) are symplectic subspaces. In this case, a moment map is given by 1 (k) µ(v) = − λ |vλ(k) |2 , 2 n

k=1

where | · | is the standard norm63 and v = vλ(1) + · · · + vλ(n) is the weight space decomposition of v. We conclude that, if Tn acts on Cn in a linear, effective and Hamiltonian way, then any moment map µ is a submersion, i.e., each differential dµv : Cn → Rn (v ∈ Cn ) is surjective. 61 A polytope in Rn is the convex hull of a finite number of points in Rn . A convex polyhedron is a subset of Rn

that is the intersection of a finite number of affine half-spaces. Hence, polytopes coincide with bounded convex polyhedra. 62 An action of a group G on a manifold M is called effective if each group element g = e moves at least one . point p ∈ M, that is, p∈M Gp = {e}, where Gp = {g ∈ G | g · p = p} is the stabilizer of p. 63 The standard inner product satisfies v, w = ω (v, J v) where J ∂ = i ∂ and J ∂ = −i ∂ . In particular, the 0 ∂z ∂z ∂ z¯ ∂ z¯ standard norm is invariant for a symplectic complex-linear action.

Symplectic geometry

159

2. Consider a coadjoint orbit Oλ for the unitary group U(n). Multiplying by i, the orbit Oλ can be viewed as the set of Hermitian matrices with a given eigenvalue spectrum λ = (λ1 · · · λn ). The restriction of the coadjoint action to the maximal torus Tn of diagonal unitary matrices is Hamiltonian with moment map µ : Oλ → Rn taking a matrix to the vector of its diagonal entries. Then the moment polytope µ(Oλ ) is the convex hull C of the points given by all the permutations of (λ1 , . . . , λn ). This is a rephrasing of the classical theorem of Schur (µ(Oλ ) ⊆ C) and Horn (C ⊆ µ(Oλ )). Example 1 is related to the universal local picture for a moment map near a fixed point of a Hamiltonian torus action: T HEOREM 5.22. Let (M 2n , ω, Tm , µ) be a Hamiltonian Tm -space, where q is a fixed point. Then there exists a chart (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and weights λ(1) , . . . , λ(n) ∈ Zm such that ω|U =

n k=1

1 (k) 2 and µ|U = µ(q) − λ xk + yk2 . 2 n

dxk ∧ dyk

k=1

The following two results use the crucial fact that any effective action of an m-torus on a manifold has orbits of dimension m; a proof may be found in [19]. C OROLLARY 5.23. Under the conditions of the convexity theorem, if the Tm -action is effective, then there must be at least m + 1 fixed points. P ROOF. At a point p of an m-dimensional orbit the moment map is a submersion, i.e., (dµ1 )p , . . . , (dµm )p are linearly independent. Hence, µ(p) is an interior point of µ(M), and µ(M) is a nondegenerate polytope. A nondegenerate polytope in Rm has at least m + 1 vertices. The vertices of µ(M) are images of fixed points. P ROPOSITION 5.24. Let (M, ω, Tm , µ) be a Hamiltonian Tm -space. If the Tm -action is effective, then dim M 2m. P ROOF. Since the moment map is constant on an orbit O, for p ∈ O the differential dµp : Tp M → g∗ maps Tp O to 0. Thus Tp O ⊆ ker dµp = (Tp O)ω , where (Tp O)ω is the symplectic orthogonal of Tp O. This shows that orbits O of a Hamiltonian torus action are isotropic submanifolds of M. In particular, by symplectic linear algebra we have that dim O 12 dim M. Now consider an m-dimensional orbit. For a Hamiltonian action of an arbitrary compact Lie group G on a compact symplectic manifold (M, ω), the following non-Abelian convexity theorem was proved by Kirwan [81]: if µ : M → g∗ is a moment map, then the intersection µ(M) ∩ t∗+ of the image of µ with a Weyl chamber for a Cartan subalgebra t ⊆ g is a convex polytope. This had been conjectured by Guillemin and Sternberg and proved by them in particular cases.

160

A. Cannas da Silva

6. Symplectic reduction 6.1. Marsden–Weinstein–Meyer theorem Classical physicists realized that, whenever there is a symmetry group of dimension k acting on a mechanical system, the number of degrees of freedom for the position and momenta of the particles may be reduced by 2k. Symplectic reduction formulates this process mathematically. T HEOREM 6.1 (Marsden–Weinstein, Meyer [92,102]). Let (M, ω, G, µ) be a Hamiltonian G-space (Section 5.6) for a compact Lie group G. Let i : µ−1 (0) → M be the inclusion map. Assume that G acts freely on µ−1 (0). Then (a) the orbit space Mred = µ−1 (0)/G is a manifold, (b) π : µ−1 (0) → Mred is a principal G-bundle, and (c) there is a symplectic form ωred on Mred satisfying i ∗ ω = π ∗ ωred . D EFINITION 6.2. The symplectic manifold (Mred , ωred ) is the reduction (or reduced space, or symplectic quotient) of (M, ω) with respect to G, µ. When M is Kähler and the action of G preserves the complex structure, we can show that the symplectic reduction has a natural Kähler structure. Let (M, ω, G, µ) be a Hamiltonian G-space for a compact Lie group G. To reduce at a level ξ ∈ g∗ of µ, we need µ−1 (ξ ) to be preserved by G, or else take the G-orbit of µ−1 (ξ ), or else take the quotient by the maximal subgroup of G that preserves µ−1 (ξ ). Since µ is equivariant, G preserves µ−1 (ξ ) if and only if Ad∗g ξ = ξ , ∀g ∈ G. Of course, the level 0 is always preserved. Also, when G is a torus, any level is preserved and reduction at ξ for the moment map µ, is equivalent to reduction at 0 for a shifted moment map φ : M → g∗ , φ(p) := µ(p) − ξ . In general, let O be a coadjoint orbit in g∗ equipped with the canonical symplectic formωO (defined in Section 5.1). Let O− be the orbit O equipped with −ωO . The natural product action of G on M × O− is Hamiltonian with moment map µO (p, ξ ) = µ(p) − ξ . If the hypothesis of Theorem 6.1 is satisfied for M × O− , then one obtains a reduced space with respect to the coadjoint orbit O. E XAMPLES . 1. The standard symplectic form on Cn is ω0 = 2i dzi ∧ d z¯ i = dxi ∧ dyi = ri dri ∧ dθi in polar coordinates. The S 1 -action on (Cn , ω0 ) where eit ∈ S 1 acts as multiplication by eit has vector field X # = ∂θ∂ 1 + ∂θ∂ 2 + · · · + ∂θ∂ n . This action is 2 Hamiltonian with moment map µ : Cn → R, µ(z) = − |z|2 , since ıX# ω = ri dri = − 12 dri2 = dµ. The level µ−1 (− 12 ) is the unit sphere S 2n−1 , whose orbit space is the projective space, µ

−1

1 / 1 − S = S 2n−1 /S 1 = CPn−1 . 2

Symplectic geometry

161

The reduced symplectic form at level − 12 is ωred = ωFS the Fubini–Study symplectic form. Indeed, if pr : Cn+1 \ {0} → CPn is the standard projection, the forms pr∗ ωFS = i ¯ 2 2n+1 . 2 ∂ ∂ log(|z| ) and ω0 have the same restriction to S Id k×n 2. Consider the natural action of U(k) on C with moment map µ(A) = 2i AA∗ + 2i k×n −1 k×n ∗ for A ∈ C (Section 5.6). Since µ (0) = {A ∈ C | AA = Id}, the reduced manifold is the Grassmannian of k-planes in Cn : µ−1 (0)/U(k) = G(k, n). For the case where G = S 1 and dim M = 4, here is a glimpse of reduction. Let µ : M → R be the moment map and p ∈ µ−1 (0). Choose local coordinates near p: θ along the orbit through p, µ given by the moment map, and η1 , η2 the pullback of coordinates on Mred = µ−1 (0)/S 1 . Then the symplectic form can be written ω = A dθ ∧ dµ +

Bj dθ ∧ dηj +

Cj dµ ∧ dηj + D dη1 ∧ dη2 .

∂ )ω, we must have A = 1, Bj = 0. Since ω is symplectic, it must be D = 0. As dµ = ı( ∂θ ∗ Hence, i ω = D dη1 ∧ dη2 is the pullback of a symplectic form on Mred . The actual proof of Theorem 6.1 requires some preliminary ingredients. Let µ : M → g∗ be the moment map for an (Hamiltonian) action of a Lie group G on a symplectic manifold (M, ω). Let gp be the Lie algebra of the stabilizer of a point p ∈ M, let g0p = {ξ ∈ g∗ | ξ, X = 0, ∀X ∈ gp } be the annihilator of gp , and let Op be the G-orbit through p. Since ωp (Xp# , v) = dµp (v), X, for all v ∈ Tp M and all X ∈ g, the differential dµp : Tp M → g∗ has

ker dµp = (Tp Op )ωp

and

im dµp = g0p .

Consequently, the action is locally free64 at p if and only if p is a regular point of µ (i.e., dµp is surjective), and we obtain: L EMMA 6.3. If G acts freely on µ−1 (0), then 0 is a regular value of µ, the level µ−1 (0) is a submanifold of M of codimension dim G, and, for p ∈ µ−1 (0), the tangent space Tp µ−1 (0) = ker dµp is the symplectic orthogonal to Tp Op in Tp M. In particular, orbits in µ−1 (0) are isotropic. Since any tangent vector to the orbit is the value of a vector field generated by the group, we can show this directly by computing, for any X, Y ∈ g and p ∈ µ−1 (0), the Hamiltonian function for [Y # , X # ] = [Y, X]# at that point: ωp (Xp# , Yp# ) = µ[Y,X] (p) = 0. L EMMA 6.4. Let (V , Ω) be a symplectic vector space, and I an isotropic subspace. Then Ω induces a canonical symplectic structure Ωred on I Ω /I . 64 The action is locally free at p when g = {0}, i.e., the stabilizer of p is a discrete group. The action is free at p p when the stabilizer of p is trivial, i.e., Gp = {e}.

162

A. Cannas da Silva

P ROOF. Let [u], [v] be the classes in I Ω /I of u, v ∈ I Ω . We have Ω(u + i, v + j ) = Ω(u, v), ∀i, j ∈ I , because Ω(u, j ) = Ω(i, v) = Ω(i, j ) = 0. Hence, we can define Ωred ([u], [v]) := Ω(u, v). This is nondegenerate: if u ∈ I Ω has Ω(u, v) = 0, for all v ∈ I Ω , then u ∈ (I Ω )Ω = I , i.e., [u] = 0. P ROPOSITION 6.5. If a compact Lie group G acts freely on a manifold M, then M/G is a manifold and the map π : M → M/G is a principal G-bundle. P ROOF. We first show that, for any p ∈ M, the G-orbit through p is a compact submanifold of M diffeomorphic to G.65 The G-orbit through p is the image of the smooth injective map evp : G → M, evp (g) = g · p. The map evp is proper because, if A is a compact, hence closed, subset of M, then its inverse image (evp )−1 (A), being a closed subset of the compact Lie group G, is also compact. The differential d(evp )e is injective because d(evp )e (X) = 0 ⇔ Xp# = 0 ⇔ X = 0, ∀X ∈ Te G, as the action is free. At any other point g ∈ G, for X ∈ Tg G we have d(evp )g (X) = 0 ⇔ d(evp ◦ Rg )e ◦ (dRg −1 )g (X) = 0, where Rg : G → G, h → hg, is right multiplication by g. But evp ◦ Rg = evg·p has an injective differential at e, and (dRg −1 )g is an isomorphism. It follows that d(evp )g is always injective, so evp is an immersion. We conclude that evp is a closed embedding. We now apply the slice theorem66 which is an equivariant tubular neighborhood theorem. For p ∈ M, let q = π(p) ∈ M/G. Choose a G-invariant neighborhood U of p as in the slice theorem, so that U G × S where S is an appropriate slice. Then π(U) = U/G =: V is a neighborhood of q in M/G homeomorphic67 to S. Such neighborhoods V are used as charts on M/G. To show that the associated transition maps are smooth, consider two G-invariant open sets U1 , U2 in M and corresponding slices S1 , S2 . Then S12 = S1 ∩ U2 , S21 = S2 ∩ U1 are both slices for the G-action on U1 ∩ U2 . To compute the transition map S12 → S21 , consider the sequence S12 −→ {e} × S12 → G × S12 −→ U1 ∩ U2 and similarly pr for S21 . The composition S12 → U1 ∩ U2 −→ G × S21 −→ S21 is smooth. Finally, we show that π : M → M/G is a principal G-bundle. For p ∈ M, q = π(p), choose a G-invariant neighborhood U of p of the form η : G × S −→ U . Then V = U/G S is the corresponding neighborhood of q in M/G: η

M ⊇U G×S G×V ↓π ↓ M/G ⊇ V = V 65 Even if the action is not free, the orbit through p is a compact submanifold of M. In that case, the orbit of a point p is diffeomorphic to the quotient G/Gp of G by the stabilizer of p. 66 Slice theorem. Let G be a compact Lie group acting on a manifold M such that G acts freely at p ∈ M. Let S be a transverse section to Op at p (this is called a slice). Choose a coordinate chart x1 , . . . , xn centered at p such that Op G is given by x1 = · · · = xk = 0 and S by xk+1 = · · · = xn = 0. Let Sε = S ∩ Bε where Bε is the ball of radius ε centered at 0 with respect to these coordinates. Let η : G × S → M, η(g, s) = g · s. Then, for sufficiently small ε, the map η : G × Sε → M takes G × Sε diffeomorphically onto a G-invariant neighborhood U of the G-orbit through p. In particular, if the action of G is free at p, then the action is free on U , so the set of points where G acts freely is open. 67 We equip the orbit space M/G with the quotient topology, i.e., V ⊆ M/G is open if and only if π −1 (V) is open in M.

Symplectic geometry

163

Since the projection on the right is smooth, π is smooth. By considering the overlap of two trivializations φ1 : U1 → G × V1 and φ2 : U2 → G × V2 , we check that the transition map φ2 ◦ φ1−1 = (σ12 , id) : G × (V1 ∩ V2 ) → G × (V1 ∩ V2 ) is smooth. P ROOF OF T HEOREM 6.1. Since G acts freely on µ−1 (0), by Lemma 6.3 the level µ−1 (0) is a submanifold. Applying Proposition 6.5 to the free action of G on the manifold µ−1 (0), we conclude the assertions (a) and (b). At p ∈ µ−1 (0) the tangent space to the orbit Tp Op is an isotropic subspace of the symplectic vector space (Tp M, ωp ). By Lemma 6.4 there is a canonical symplectic structure on the quotient Tp µ−1 (0)/Tp Op . The point [p] ∈ Mred = µ−1 (0)/G has tangent space T[p] Mred Tp µ−1 (0)/Tp Op . This gives a well-defined nondegenerate 2-form ωred on Mred because ω is G-invariant. By construction i ∗ ω = π ∗ ωred where µ−1 (0) ↓π Mred

i

→ M

The injectivity of π ∗ yields closedness: π ∗ dωred = dπ ∗ ωred = dı ∗ ω = ı ∗ dω = 0.

6.2. Applications and generalizations Let (M, ω, G, µ) be a Hamiltonian G-space for a compact Lie group G. Suppose that another Lie group H acts on (M, ω) in a Hamiltonian way with moment map φ : M → h∗ . Suppose that the H -action commutes with the G-action, that φ is G-invariant and that µ is H -invariant. Assuming that G acts freely on µ−1 (0), let (Mred , ωred ) be the corresponding reduced space. Since the action of H preserves µ−1 (0) and ω and commutes with the G-action, the reduced space (Mred , ωred ) inherits a symplectic action of H . Since φ is preserved by the G-action, the restriction of this moment map to µ−1 (0) descends to a moment map φred : Mred → h∗ satisfying φred ◦ π = φ ◦ i, where π : µ−1 (0) → Mred and i : µ−1 (0) → M. Therefore, (Mred , ωred , H, φred ) is a Hamiltonian H -space. Consider now the action of a product group G = G1 × G2 , where G1 and G2 are compact connected Lie groups. We have g = g1 ⊕ g2 and g∗ = g∗1 ⊕ g∗2 . Suppose that (M, ω, G, ψ) is a Hamiltonian G-space with moment map ψ = (ψ1 , ψ2 ) : M −→ g∗1 ⊕ g∗2 , where ψi : M → g∗i for i = 1, 2. The fact that ψ is equivariant implies that ψ1 is invariant under G2 and ψ2 is invariant under G1 . Assume that G1 acts freely on Z1 := ψ1−1 (0). Let (M1 = Z1 /G1 , ω1 ) be the reduction of (M, ω) with respect to G1 , ψ1 . From the observation above, (M1 , ω1 ) inherits a Hamiltonian G2 -action with moment map µ2 : M1 → g∗2 such that µ2 ◦ π = ψ2 ◦ i, where π : Z1 → M1 and i : Z1 → M. If G acts freely on ψ −1 (0, 0), then G2 acts freely on µ−1 2 (0), and there is a natural symplectomorphism −1 µ−1 2 (0)/G2 ψ (0, 0)/G.

164

A. Cannas da Silva

This technique of performing reduction with respect to one factor of a product group at a time is called reduction in stages. It may be extended to reduction by a normal subgroup H ⊂ G and by the corresponding quotient group G/H . E XAMPLE . Finding symmetries for a mechanical problem may reduce degrees of freedom by two at a time: an integral of motion f for a 2n-dimensional Hamiltonian system (M, ω, H ) may allow to understand the trajectories of this system in terms of the trajectories of a (2n − 2)-dimensional Hamiltonian system (Mred , ωred , Hred ). Locally this process goes as follows. Let (U, x1 , . . . , xn , ξ1 , . . . , ξn ) be a Darboux chart for M such ∂H , the trajectories of that f = ξn .68 Since ξn is an integral of motion, 0 = {ξn , H } = − ∂x n the Hamiltonian vector field XH lie on a constant level ξn = c (Proposition 5.12), and H does not depend on xn . The reduced space is Ured = {(x1 , . . . , xn−1 , ξ1 , . . . , ξn−1 ) | ∃a: (x1 , . . . , xn−1 , a, ξ1 , . . . , ξn−1 , c) ∈ U} and the reduced Hamiltonian is Hred : Ured → R, Hred (x1 , . . . , xn−1 , ξ1 , . . . , ξn−1 ) = H (x1 , . . . , xn−1 , a, ξ1 , . . . , ξn−1 , c) for some a. In order to find the trajectories of the original system on the hypersurface ξn = c, we look for the trajectories (x1 (t), . . . , xn−1 (t), ξ1 (t), . . . , ξn−1 (t)) of the reduced system on Ured , ∂H n and integrate the equation dx dt (t) = ∂ξn to obtain the original trajectories where

xn (t) = xn (0) + ξn (t) = c.

t

∂H 0 ∂ξn (x1 (t), . . . , xn−1 (t), ξ1 (t), . . . , ξn−1 (t), c) dt,

By Sard’s theorem, the singular values of a moment map µ : M → g∗ form a set of measure zero. So, perturbing if necessary, we may assume that a level of µ is regular hence, when G is compact, that any point p of that level has finite stabilizer Gp . Let Op be the orbit of p. By the slice theorem for the case of orbifolds, near Op the orbit space of the level is modeled by S/Gp , where S is a Gp -invariant disk in the level and transverse to Op (a slice). Thus, the orbit space is an orbifold.69 This implies that, when G = Tn is an n-torus, for most levels reduction goes through, however the quotient space is not necessarily a manifold but an orbifold. Roughly speaking, orbifolds are singular manifolds where each singularity is locally modeled on Rm /Γ , for some finite group Γ ⊂ GL(m; R). The differential-geometric notions of vector fields, differential forms, exterior 68 To obtain such a chart, in the proof of Darboux’s Theorem 1.9 start with coordinates (x , . . . , x , y , . . . y ) n 1 n 1 such that yn = f and ∂ = Xf . ∂xn 69 Let |M| be a Hausdorff topological space satisfying the second axiom of countability. An orbifold chart on |M| is a triple (V, Γ, ϕ), where V is a connected open subset of some Euclidean space Rm , Γ is a finite group that acts

linearly on V so that the set of points where the action is not free has codimension at least two, and ϕ : V → |M| is a Γ -invariant map inducing a homeomorphism from V/Γ onto its image U ⊂ |M|. An orbifold atlas A for |M| is a collection of orbifold charts on |M| such that: the collection of images U forms a basis of open sets in |M|, and the charts are compatible in the sense that, whenever two charts (V1 , Γ1 , ϕ1 ) and (V2 , Γ2 , ϕ2 ) satisfy U1 ⊆ U2 , there exists an injective homomorphism λ : Γ1 → Γ2 and a λ-equivariant open embedding ψ : V1 → V2 such that ϕ2 ◦ ψ = ϕ1 . Two orbifold atlases are equivalent if their union is still an atlas. An m-dimensional orbifold M is a Hausdorff topological space |M| satisfying the second axiom of countability, plus an equivalence class of orbifold atlases on |M|. We do not require the action of each group Γ to be effective. Given a point p on an orbifold M, let (V, Γ, ϕ) be an orbifold chart for a neighborhood U of p. The orbifold structure group of p, Γp , is (the isomorphism class of) the stabilizer of a preimage of p under φ. Orbifolds were introduced by Satake in [114].

Symplectic geometry

165

differentiation, group actions, etc., extend naturally to orbifolds by gluing corresponding local Γ -invariant or Γ -equivariant objects. In particular, a symplectic orbifold is a pair (M, ω) where M is an orbifold and ω is a closed 2-form on M that is nondegenerate at every point. E XAMPLES . The S 1 -action on C2 given by eiθ · (z1 , z2 ) = (eikθ z1 , eiθ z2 ), for some integers k and , has moment map µ : C2 → R, (z1 , z2 ) → − 12 (k|z1 |2 + |z2 |2 ). Any ξ < 0 is a regular value and µ−1 (ξ ) is a 3-dimensional ellipsoid. 2πm When = 1 and k 2, the stabilizer of (z1 , z2 ) is {1} if z2 = 0 and is Zk = {ei k | m = 0, 1, . . . , k − 1} if z2 = 0. The reduced space µ−1 (ξ )/S 1 is then called a teardrop orbifold or conehead; it has one cone (or dunce cap) singularity with cone angle 2π k , that is, a point with orbifold structure group Zk . When k, 2 are relatively prime, for z1 , z2 = 0 the stabilizer of (z1 , 0) is Zk , of (0, z2 ) is Z and of (z1 , z2 ) is {1}. The quotient µ−1 (ξ )/S 1 is called a football orbifold: it has two 2π cone singularities, with angles 2π k and . For S 1 acting on Cn by eiθ · (z1 , . . . , zn ) = (eik1 θ z1 , . . . , eikn θ zn ) the reduced spaces are orbifolds called weighted (or twisted) projective spaces. Let (M, ω) be a symplectic manifold where S 1 acts in a Hamiltonian way, ρ : S 1 → Diff(M), with moment map µ : M → R. Suppose that: • M has a unique nondegenerate minimum at q where µ(q) = 0, and • for ε sufficiently small, S 1 acts freely on the level set µ−1 (ε). Let C be equipped with the symplectic form −i dz ∧ d z¯ . Then the action of S 1 on the product ψ : S 1 → Diff(M × C), ψt (p, z) = (ρt (p), t · z), is Hamiltonian with moment map φ : M × C −→ R,

φ(p, z) = µ(p) − |z|2 .

Observe that S 1 acts freely on the ε-level of φ for ε small enough: φ −1 (ε) = (p, z) ∈ M × C | µ(p) − |z|2 = ε = (p, 0) ∈ M × C | µ(p) = ε ∪ (p, z) ∈ M × C | |z|2 = µ(p) − ε > 0 . The reduced space is hence φ −1 (ε)/S 1 µ−1 (ε)/S 1 ∪ p ∈ M | µ(p) > ε . The open submanifold of M given by {p ∈ M | µ(p) > ε} embeds as an open dense submanifold into φ −1 (ε)/S 1 . The reduced space φ −1 (ε)/S 1 is the ε-blow-up of M at q (Section 5.6). This global description of blow-up for Hamiltonian S 1 -spaces is due to Lerman [86], as a particular instance of his cutting technique. Symplectic cutting is the application of symplectic reduction to the product of a Hamiltonian S 1 -space with the standard

166

A. Cannas da Silva

C as above, in a way that the reduced space for the original Hamiltonian S 1 -space embeds symplectically as a codimension 2 submanifold in a symplectic manifold. As it is a local construction, the cutting operation may be more generally performed at a local minimum (or maximum) of the moment map µ. There is a remaining S 1 -action on the cut space ε Mcut := φ −1 (ε)/S 1 induced by τ : S 1 −→ Diff(M × C),

τt (p, z) = ρt (p), z .

In fact, τ is a Hamiltonian S 1 -action on M × C that commutes with ψ, thus descends to ε an action τ˜ : S 1 → Diff(Mcut ), which is also Hamiltonian. Loosely speaking, the cutting technique provides a Hamiltonian way to close the open manifold {p ∈ M | µ(p) > ε}, by using the reduced space at level ε, µ−1 (ε)/S 1 . We may similarly close {p ∈ M | µ(p) < ε}. The resulting Hamiltonian S 1 -spaces are called cut ε ε spaces, and denoted Mcut and Mcut . If another group G acts on M in a Hamiltonian way 1 that commutes with the S -action, then the cut spaces are also Hamiltonian G-spaces.

6.3. Moment map in gauge theory Let G be a Lie group and P a principal G-bundle over B.70 If A is a connection (form)71 1 on P , and if a ∈ Ωhoriz ⊗ g is G-invariant for the product action, then A + a is also a 1 connection on P . Reciprocally, any two connections on P differ by an a ∈ (Ωhoriz ⊗ g)G . 70 Let G be a Lie group and B a manifold. A principal G-bundle over B is a fibration π : P → B (Section 4.2) with a free action of G (the structure group) on the total space P , such that the base B is the orbit space, the map π is the point-orbit projection and the local trivializations are of the form ϕU = (π, sU ) : π −1 (U) → U × G with sU (g · p) = g · sU (p) for all g ∈ G and all p ∈ π −1 (U). A principal G-bundle is represented by a diagram

G →

P ↓π B

For instance, the Hopf fibration is a principal S 1 -bundle over S 2 (= CP1 ) with total space S 3 regarded as unit vectors in C2 where circle elements act by complex multiplication. 71 An action ψ : G → Diff(P ) induces an infinitesimal action dψ : g → χ(P ) mapping X ∈ g to the vector field X # generated by the one-parameter group {exp tX(e) | t ∈ R} ⊆ G. Fix a basis X1 , . . . , Xk of g. Let P be a principal G-bundle over B. Since the G-action is free, the vector fields X1# , . . . , Xk# are linearly independent at each p ∈ P . The vertical bundle V is the rank k subbundle of T P generated by X1# , . . . , Xk# . Alternatively, V is the set of vectors tangent to P that lie in the kernel of the derivative of the bundle projection π , so V is indeed independent of the choice of basis for g. An (Ehresmann) connection on P is a choice of a splitting T P = V ⊕ H , where H (called the horizontal bundle) is a G-invariant subbundle of T P complementary to the vertical bundle V . A connection form on P is a Lie-algebra-valued 1-form A = ki=1 Ai ⊗ Xi ∈ Ω 1 (P ) ⊗ g such that A is G-invariant, with respect to the product action of G on Ω 1 (P ) (induced by the action on P ) and on g (the adjoint action), and A is vertical, in the sense that ıX# A = X for any X ∈ g. A connection T P = V ⊕ H determines a connection (form) A and vice-versa by the formula H = ker A = {v ∈ T P | ıv A = 0}. Given a connection on P , the splitting T P = V ⊕ H induces splittings for bundles T ∗ P = V ∗ ⊕ H ∗ , ∧2 T ∗ P = (∧2 V ∗ ) ⊕ (V ∗ ∧ H ∗ ) ⊕ (∧2 H ∗ ), 1 ⊕ Ω1 2 2 2 2 etc., and for their sections: Ω 1 (P ) = Ωvert horiz , Ω (P ) = Ωvert ⊕ Ωmix ⊕ Ωhoriz , etc. The corresponding 1 connection form A is in Ωvert ⊗ g.

Symplectic geometry

167

We conclude that the set A of all connections on the principal G-bundle P is an affine 1 space modeled on the linear space a = (Ωhoriz ⊗ g)G . Now let P be a principal G-bundle over a compact Riemann surface. Suppose that the group G is compact or semisimple. Atiyah and Bott [6] noticed that the corresponding space A of all connections may be treated as an infinite-dimensional symplectic manifold. This requires choosing a G-invariant inner product ·, · on g, which always exists, either by averaging any inner product when G is compact, or by using the Killing form on semisimple groups. Since A is an affine space, its tangent space at any point A is identified with the model linear space a. With respect to a basis X1 , . . . , Xk for the Lie algebra g, elements a, b ∈ a are written a=

ai ⊗ X i

and b =

bi ⊗ Xi .

If we wedge a and b, and then integrate over B, we obtain a real number: 2 G ω : a × a −→ Ωhoriz (P ) Ω 2 (B) −→ R, ai ∧ bj Xi , Xj −→ ai ∧ bj Xi , Xj . (a, b) −→ i,j

B

i,j

We used that the pullback π ∗ : Ω 2 (B) → Ω 2 (P ) is an isomorphism onto its image 2 (Ωhoriz (P ))G . When ω(a, b) = 0 for all b ∈ a, then a must be zero. The map ω is nondegenerate, skew-symmetric, bilinear and constant in the sense that it does not depend on the base point A. Therefore, it has the right to be called a symplectic form on A, so the pair (A, ω) is an infinite-dimensional symplectic manifold. A diffeomorphism f : P → P commuting with the G-action determines a diffeomorphism fbasic : B → B by projection. Such a diffeomorphism f is called a gauge transformation if the induced fbasic is the identity. The gauge group of P is the group G of all gauge transformations of P . The derivative of an f ∈ G takes an Ehresmann connection T P = V ⊕ H to another connection T P = V ⊕ Hf , and thus induces an action of G in the space A of all connections. Atiyah and Bott [6] noticed that the action of G on (A, ω) is Hamiltonian, where the moment map (appropriately interpreted) is G µ : A −→ Ω 2 (P ) ⊗ g , A −→ curv A, i.e., the moment map is the curvature.72 The reduced space M = µ−1 (0)/G is the space of flat connections modulo gauge equivalence, known as the moduli space of flat connections, which is a finite-dimensional symplectic orbifold. 72 The exterior derivative of a connection A decomposes into three components,

2 2 ⊕ Ω2 ⊕ Ωmix dA = (dA)vert + (dA)mix + (dA)horiz ∈ Ωvert horiz ⊗ g

168

A. Cannas da Silva

E XAMPLE . We describe the Atiyah–Bott construction for the case of a circle bundle S1

→ P ↓π B

Let v be the generator of the S 1 -action on P , corresponding to the basis 1 of g R. A connection form on P is an ordinary 1-form A ∈ Ω 1 (P ) such that Lv A = 0 and ıv A = 1. If we fix one particular connection A0 , then any other connection is of the form A = A0 + a 1 for some a ∈ a = (Ωhoriz (P ))G = Ω 1 (B). The symplectic form on a = Ω 1 (B) is simply ω : a × a −→ Ω 2 (B) −→ R, a ∧ b. (a, b) −→ a ∧ b −→ B

The gauge group is G = Maps(B, S 1 ), because a gauge transformation is multiplication by some element of S 1 over each point in B encoded in a map h : B → S 1 . The action φ : G → Diff(P ) takes h ∈ G to the diffeomorphism φh : p −→ h π(p) · p. The Lie algebra of G is Lie G = Maps(B, R) = C ∞ (B) with dual (Lie G)∗ = Ω 2 (B), where the (smooth) duality is provided by integration C ∞ (B) × Ω 2 (B) → R, (h, β) → B hβ. The gauge group acts on the space of all connections by ψ : G −→ Diff(A), h : x → eiθ(x) −→ (ψh : A → A − π ∗ dθ ). (In the case where P = S 1 × B is a trivial bundle, every connection can be written A = dt + β, with β ∈ Ω 1 (B). A gauge transformation h ∈ G acts on P by φh : (t, x) → (t + θ (x), x) and on A by A → φh∗−1 (A).) The infinitesimal action is dψ : Lie G −→ χ(A), X −→ X # = vector field described by (A → A − dX), so that X # = −dX. It remains to check that µ : A −→ (Lie G)∗ = Ω 2 (B), A −→ curv A

i A ∧ A ⊗ X , where the satisfying (dA)mix = 0 and (dA)vert (X, Y ) = [X, Y ], i.e., (dA)vert = 12 i,,m cm m i i ’s are the structure constants of the Lie algebra with respect to the chosen basis, and defined by [X , X ] = cm m i i,,m cm Xi . So the relevance of dA may come only from its horizontal component, called the curvature form 2 of the connection A, and denoted curv A = (dA)horiz ∈ Ωhoriz ⊗ g. A connection is called flat if its curvature is zero.

Symplectic geometry

169

is indeed a moment map for the action of the gauge group on A. Since in this case curv A = 2 dA ∈ (Ωhoriz (P ))G = Ω 2 (B), the action of G on Ω 2 (B) is trivial and µ is G-invariant, the ∞ X equivariance

condition is satisfied. Take any X ∈ Lie G = C (B). Since the map µ : A → X, dA = B X · dA is linear in A, its differential is dµX : a −→ R, a −→ X da. B

By definition of ω and the Stokes theorem, we have that ω X# , a =

X ·a =−

B

dX · a =

#

B

X · da = dµX (a),

∀a ∈ Ω 1 (B),

B

so we are done in proving that µ is the moment map. The function µ2 : A → R giving the square of the L2 norm of the curvature is the Yang–Mills functional, whose Euler–Lagrange equations are the Yang–Mills equations. Atiyah and Bott [6] studied the topology of A by regarding µ2 as an equivariant Morse function. In general, it is a good idea to apply Morse theory to the norm square of a moment map [80].

6.4. Symplectic toric manifolds Toric manifolds are smooth toric varieties.73 When studying the symplectic features of these spaces, we refer to them as symplectic toric manifolds. Relations between the algebraic and symplectic viewpoints on toric manifolds are discussed in [21]. D EFINITION 6.6. A symplectic toric manifold is a compact connected symplectic manifold (M, ω) equipped with an effective Hamiltonian action of a torus T of dimension equal to half the dimension of the manifold, dim T = 12 dim M, and with a choice of a corresponding moment map µ. Two symplectic toric manifolds, (Mi , ωi , Ti , µi ), i = 1, 2, are equivalent if there exists an isomorphism λ : T1 → T2 and a λ-equivariant symplectomorphism ϕ : M1 → M2 such that µ1 = µ2 ◦ ϕ. E XAMPLES . 1. The circle S 1 acts on the 2-sphere (S 2 , ωstandard = dθ ∧ dh) by rotations, eiν · (θ, h) = (θ + ν, h). with moment map µ = h equal to the height function and moment polytope [−1, 1] (see Figure 3). 73 Toric varieties were introduced by Demazure in [29]. There are many nice surveys of the theory of toric varieties in algebraic geometry; see, for instance, [27,53,79,107]. Toric geometry has recently become an important tool in physics in connection with mirror symmetry [26].

170

A. Cannas da Silva

µ=h

1

−1

Fig. 3.

Analogously, S 1 acts on the Riemann sphere CP1 with the Fubini–Study form ωFS = 14 ωstandard , by eiθ · [z0 , z1 ] = [z0 , eiθ z1 ]. This is Hamiltonian with moment map µ[z0 , z1 ] = − 12 ·

|z1 |2 , |z0 |2 +|z1 |2

and moment polytope [− 12 , 0].

2. For the Tn -action on the product of n Riemann spheres CP1 × · · · × CP1 by iθ e 1 , . . . , eiθn · [z1 , w1 ], . . . , [zn , wn ] = z1 , eiθ1 w1 , . . . , w0 , eiθn w1 , the moment polytope is an n-dimensional cube. 3. Let (CP2 , ωFS ) be 2-(complex-)dimensional complex projective space equipped with the Fubini–Study form defined in Section 3.4. The T2 -action on CP2 by (eiθ1 , eiθ2 ) · [z0 , z1 , z2 ] = [z0 , eiθ1 z1 , eiθ2 z2 ] has moment map µ[z0 , z1 , z2 ] = −

|z1 |2 1 |z2 |2 . , 2 |z0 |2 + |z1 |2 + |z2 |2 |z0 |2 + |z1 |2 + |z2 |2

The image is the isosceles triangle with vertices (0, 0), (− 12 , 0) and (0, − 12 ). 4. For the Tn -action on (CPn , ωFS ) by iθ e 1 , . . . , eiθn · [z0 , z1 , . . . , zn ] = z0 , eiθ1 z1 , . . . , eiθn zn the moment polytope is an n-dimensional simplex. Since the coordinates of the moment map are commuting integrals of motion, a symplectic toric manifold gives rise to a completely integrable system. By Proposition 5.24, symplectic toric manifolds are optimal Hamiltonian torus-spaces. By Theorem 5.21, they have an associated polytope. It turns out that the moment polytope contains enough information to sort all symplectic toric manifolds. We now define the class of polytopes that arise in the classification. For a symplectic toric manifold the weights λ(1) , . . . , λ(n) in Theorem 5.22 form a Z-basis of Zm , hence the moment polytope is a Delzant polytope: D EFINITION 6.7. A Delzant polytope in Rn is a polytope satisfying: • simplicity, i.e., there are n edges meeting at each vertex; • rationality, i.e., the edges meeting at the vertex p are rational in the sense that each edge is of the form p + tui , t 0, where ui ∈ Zn ;

Symplectic geometry

171

Fig. 4.

• smoothness, i.e., for each vertex, the corresponding u1 , . . . , un can be chosen to be a Z-basis of Zn . In R2 the simplicity condition is always satisfied (by nondegenerate polytopes). In R3 , for instance, a square pyramid fails the simplicity condition. E XAMPLES . Figure 4 represents Delzant polytopes in R2 . The following theorem classifies (equivalence classes of) symplectic toric manifolds in terms of the combinatorial data encoded by a Delzant polytope. T HEOREM 6.8 (Delzant [28]). Toric manifolds are classified by Delzant polytopes, and their bijective correspondence is given by the moment map: {toric manifolds} ←→ {Delzant polytopes}, 2n M , ω, Tn , µ −→ µ(M). Delzant’s construction (Section 6.5) shows that for a toric manifold the moment map takes the fixed points bijectively to the vertices of the moment polytope and takes points with a k-dimensional stabilizer to the codimension k faces of the polytope. The moment polytope is exactly the orbit space, i.e., the preimage under µ of each point in the polytope is exactly one orbit. For instance, consider (S 2 , ω = dθ ∧ dh, S 1 , µ = h), where S 1 acts by rotation. The image of µ is the line segment I = [−1, 1]. The product S 1 × I is an open-ended cylinder. We can recover the 2-sphere by collapsing each end of the cylinder to a point. Similarly, we can build CP2 from T2 × where is a rectangular isosceles triangle, and so on. E XAMPLES . 1. By a linear transformation in SL(2; Z), we can make one of the angles in a Delzant triangle into a right angle. Out of the rectangular triangles, only the isosceles one satisfies the smoothness condition. Therefore, up to translation, change of scale and the action of SL(2; Z), there is just one 2-dimensional Delzant polytope with three vertices, namely an isosceles triangle. We conclude that the projective space CP2 is the only 4-dimensional toric manifold with three fixed points, up to choices of a constant in the moment map, of a multiple of ωFS and of a lattice basis in the Lie algebra of T2 . 2. Up to translation, change of scale and the action of SL(n; Z), the standard n-simplex in Rn (spanned by the origin and the standard basis vectors (1, 0, . . . , 0), . . . ,

172

A. Cannas da Silva

p Fig. 5.

(0, . . . , 0, 1)) is the only n-dimensional Delzant polytope with n + 1 vertices. Hence, M = CPn is the only 2n-dimensional toric manifold with n + 1 fixed points, up to choices of a constant in the moment map, of a multiple of ωFS and of a lattice basis in the Lie algebra of TN . 3. A transformation in SL(2; Z) makes one of the angles in a Delzant quadrilateral into a right angle. Automatically an adjacent angle also becomes 90◦ . Smoothness imposes that the slope of the skew side be integral. Thus, up to translation, change of scale and SL(2; Z)-action, the 2-dimensional Delzant polytopes with four vertices are trapezoids with vertices (0, 0), (0, 1), (, 1) and ( + n, 0), for n a nonnegative integer and > 0. Under Delzant’s construction (that is, under symplectic reduction of C4 with respect to an action of (S 1 )2 ), these correspond to the so-called Hirzebruch surfaces—the only 4-dimensional symplectic toric manifolds that have four fixed points up to equivalence as before. Topologically, they are S 2 -bundles over S 2 , either the trivial bundle S 2 × S 2 when n is even or the nontrivial bundle (given by the blowup of CP2 at a point; see Section 4.3) when n is odd. Let be an n-dimensional Delzant polytope, and let (M , ω , Tn , µ ) be the associated symplectic toric manifold. The ε-blow-up of (M , ω ) at a fixed point of the Tn action is a new symplectic toric manifold (Sections 4.3 and 5.6). Let q be a fixed point of the Tn -action on (M , ω ), and let p = µ (q) be the corresponding vertex of . Let u1 , . . . , un be the primitive (inward-pointing) edge vectors at p, so that the rays p + tui , t 0, form the edges of at p. P ROPOSITION 6.9. The ε-blow-up of (M , ω ) at a fixed point q is the symplectic toric manifold associated to the polytope ε obtained from by replacing the vertex p by the n vertices p + εui , i = 1, . . . , n. In other words, the moment polytope for the blow-up of (M , ω ) at q is obtained from by chopping off the corner corresponding to q, thus substituting the original set of vertices by the same set with the vertex corresponding to q replaced by exactly n new vertices. The truncated polytope is Delzant. We may view the ε-blow-up of (M , ω ) as being obtained from M by smoothly replacing q by (CPn−1 , εωFS ) (whose moment polytope is an (n − 1)-dimensional simplex). (See Figure 5.) E XAMPLE . The moment polytope for the standard T2 -action on (CP2 , ωFS ) is a right isosceles triangle . If we blow up CP2 at [0 : 0 : 1] we obtain a symplectic toric manifold associated to the trapezoid below: a Hirzebruch surface (see Figure 6).

Symplectic geometry

β

173

Fig. 6.

M

µ

R

projection

Fig. 7.

Let (M, ω, Tn , µ) be a 2n-dimensional symplectic toric manifold. Choose a suitably generic direction in Rn by picking a vector X whose components are independent over Q. This condition ensures that: • the one-dimensional subgroup TX generated by the vector X is dense in Tn , • X is not parallel to the facets of the moment polytope := µ(M), and • the vertices of have different projections along X. Then the fixed points for the Tn -action are exactly the fixed points of the action restricted to TX , that is, are the zeros of the vector field, X # on M generated by X. The projection of µ along X, µX := µ, X : M → R, is a Hamiltonian function for the vector field X # generated by X. We conclude that the critical points of µX are precisely the fixed points of the Tn -action (see Figure 7). By Theorem 5.22, if q is a fixed point for the Tn -action, then there exists a chart (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and weights λ(1) , . . . , λ(n) ∈ Zn such that n 1 (k) µ |U = µ, X|U = µ (q) − λ , X xk2 + yk2 . 2 X

X

k=1

Since the components of X are independent over Q, all coefficients λ(k) , X are nonzero, so q is a nondegenerate critical point of µX . Moreover, the index74 of q is twice the number of labels k such that −λ(k) , X < 0. But the −λ(k) ’s are precisely the edge vectors ui which satisfy Delzant’s conditions. Therefore, geometrically, the index of q can be read from the moment polytope , by taking twice the number of edges whose inward-pointing 74 A Morse function on an m-dimensional manifold M is a smooth function f : M → R all of whose critical

points (where df vanishes) are nondegenerate (i.e., the Hessian matrix is nonsingular). Let q be a nondegenerate critical point for f : M → R. The index of f at q is the index of the Hessian Hq : Rm × Rm → R regarded as a symmetric bilinear function, that is, the maximal dimension of a subspace of R where H is negative definite.

174

A. Cannas da Silva

edge vectors at µ(q) point up relative to X, that is, whose inner product with X is positive. In particular, µX is a perfect Morse function75 and we have P ROPOSITION 6.10. Let X ∈ Rn have components independent over Q. The degree-2k homology group of the symplectic toric manifold (M, ω, T, µ) has dimension equal to the number of vertices of the moment polytope where there are exactly k (primitive inwardpointing) edge vectors that point up relative to the projection along the X. All odd-degree homology groups of M are zero. By Poincaré duality (or by taking −X instead of X), the words point up may be replaced by point down. The Euler characteristic of a symplectic toric manifold is simply the number of vertices of the corresponding polytope. There is a combinatorial way of understanding the cohomology ring [53]. A symplectic toric orbifold is a compact connected symplectic orbifold (M, ω) equipped with an effective Hamiltonian action of a torus of dimension equal to half the dimension of the orbifold, and with a choice of a corresponding moment map. Symplectic toric orbifolds were classified by Lerman and Tolman [87] in a theorem that generalizes Delzant’s: a symplectic toric orbifold is determined by its moment polytope plus a positive integer label attached to each of the polytope facets. The polytopes that occur are more general than the Delzant polytopes in the sense that only simplicity and rationality are required; the edge vectors u1 , . . . , un need only form a rational basis of Zn . When the integer labels are all equal to 1, the failure of the polytope smoothness accounts for all orbifold singularities.

6.5. Delzant’s construction Following [28,66], we prove the existence part (or surjectivity) in Delzant’s theorem, by using symplectic reduction to associate to an n-dimensional Delzant polytope a symplectic toric manifold (M , ω , Tn , µ ). Let be a Delzant polytope in (Rn )∗76 and with d facets.77 We can algebraically describe as an intersection of d halfspaces. Let vi ∈ Zn , i = 1, . . . , d, be the primitive78 outward-pointing normal vectors to the facets of . Then, for some λi ∈ R, we can write = {x ∈ (Rn )∗ | x, vi λi , i = 1, . . . , d}. 75 A perfect Morse function is a Morse function f for which the Morse inequalities [103,104] are equalities, i.e., bλ (M) = Cλ and bλ (M) − bλ−1 (M) + · · · ± b0 (M) = Cλ − Cλ−1 + · · · ± C0 where bλ (M) = dim Hλ (M) and Cλ be the number of critical points of f with index λ. If all critical points of a Morse function f have even index, then f is a perfect Morse function. 76 Although we identify Rn with its dual via the Euclidean inner product, it may be more clear to see in (Rn )∗ for Delzant’s construction. 77 A face of a polytope is a set of the form F = P ∩ {x ∈ Rn | f (x) = c} where c ∈ R and f ∈ (Rn )∗ satisfies f (x) c, ∀x ∈ P . A facet of an n-dimensional polytope is an (n − 1)-dimensional face. 78 A lattice vector v ∈ Zn is primitive if it cannot be written as v = ku with u ∈ Zn , k ∈ Z and |k| > 1; for instance, (1, 1), (4, 3), (1, 0) are primitive, but (2, 2), (3, 6) are not.

Symplectic geometry

v2

175

(0, 1) v3 (0, 0) (1, 0) v1 Fig. 8.

E XAMPLE . When is the triangle shown in Figure 8, we have ∗ = x ∈ R2 x, (−1, 0) 0, x, (0, −1) 0, x, (1, 1) 1 . For the standard basis e1 = (1, 0, . . . , 0), . . . , ed = (0, . . . , 0, 1) of Rd , consider π : Rd −→ Rn , ei −→ vi . L EMMA 6.11. The map π is onto and maps Zd onto Zn . P ROOF. We need to show that the set {v1 , . . . , vd } spans Zn . At a vertex p, the edge vectors u1 , . . . , un ∈ (Rn )∗ form a basis for (Zn )∗ which, by a change of basis if necessary, we may assume is the standard basis. Then the corresponding primitive normal vectors to the facets meeting at p are −u1 , . . . , −un . π

We still call π the induced surjective map Td = Rd /(2πZd ) → Tn = Rn /(2πZn ). The kernel N of π is a (d − n)-dimensional Lie subgroup of Td with inclusion i : N → Td . Let n be the Lie algebra of N . The exact sequence of tori i

π

1 −→ N −→ Td −→ Tn −→ 1 induces an exact sequence of Lie algebras i

π

0 −→ n −→ Rd −→ Rn −→ 0 with dual exact sequence ∗ i ∗ π∗ 0 −→ (Rn )∗ −→ Rd −→ n∗ −→ 0.

176

A. Cannas da Silva

dzk ∧ d z¯ k , and standard Hamiltonian acConsider Cd with symplectic form ω0 = 2i d it it d 1 tion of T given by (e , . . . , e ) · (z1 , . . . , zd ) = (eit1 z1 , . . . , eitd zd ). A moment map is φ : Cd → (Rd )∗ defined by φ(z1 , . . . , zd ) = −

1 2 |z1 | , . . . , |zd |2 + (λ1 , . . . , λd ), 2

where the constant is chosen for later convenience. The subtorus N acts on Cd in a Hamiltonian way with moment map i ∗ ◦ φ : Cd → n∗ . Let Z = (i ∗ ◦ φ)−1 (0). In order to show that Z (a closed set) is compact it suffices (by the Heine–Borel theorem) to show that Z is bounded. Let be the image of by π ∗ . First we show that φ(Z) = . A value y ∈ (Rd )∗ is in the image of Z by φ if and only if (a)

y is in the image of φ

and (b)

i∗y = 0

if and only if (using the expression for φ and the third exact sequence) (a)

y, ei λi

(b)

y = π ∗ (x)

for i = 1, . . . , d

and

for some x ∈ (Rn )∗ .

Suppose that y = π ∗ (x). Then y, ei λi ,

∀i

⇐⇒ ⇐⇒

x, π(ei ) λi ,

x, vi λi ,

∀i

∀i ⇐⇒

x ∈ .

Thus, y ∈ φ(Z) ⇔ y ∈ π ∗ () = . Since is compact, φ is proper and φ(Z) = , we conclude that Z must be bounded, and hence compact. In order to show that N acts freely on Z, pick a vertex p of , and let I = {i1 , . . . , in } be the set of indices for the n facets meeting at p. Pick z ∈ Z such that φ(z) = π ∗ (p). Then p is characterized by n equations p, vi = λi where i ∈ I : p, vi = λi ⇐⇒ p, π(ei ) = λi ⇐⇒ π ∗ (p), ei = λi ⇐⇒ φ(z), ei = λi ⇐⇒

ith coordinate of φ(z) is equal to λi

⇐⇒

1 − |zi |2 + λi = λi 2 zi = 0.

⇐⇒

Hence, those z’s are points whose coordinates in the set I are zero, and whose other coordinates are nonzero. Without loss of generality, we may assume that I = {1, . . . , n}. The stabilizer of z is d T z = (t1 , . . . , tn , 1, . . . , 1) ∈ Td .

Symplectic geometry

177

As the restriction π : (Rd )z → Rn maps the vectors e1 , . . . , en to a Z-basis v1 , . . . , vn of Zn (respectively), at the level of groups π : (Td )z → Tn must be bijective. Since N = ker(π : Td → Tn ), we conclude that N ∩ (Td )z = {e}, i.e., Nz = {e}. Hence all N -stabilizers at points mapping to vertices are trivial. But this was the worst case, since other stabilizers Nz (z ∈ Z) are contained in stabilizers for points z that map to vertices. We conclude that N acts freely on Z. We now apply reduction. Since i ∗ is surjective, 0 ∈ n∗ is a regular value of i ∗ ◦ φ. Hence, Z is a compact submanifold of Cd of (real) dimension 2d − (d − n) = d + n. The orbit space M = Z/N is a compact manifold of (real) dimension dim Z − dim N = (d + n) − (d − n) = 2n. The point-orbit map p : Z → M is a principal N -bundle over M . Consider the diagram Z p↓ M

j

→

Cd

where j : Z → Cd is inclusion. The Marsden–Weinstein–Meyer theorem (Theorem 6.1) guarantees the existence of a symplectic form ω on M satisfying p ∗ ω = j ∗ ω0 . Since Z is connected, the symplectic manifold (M , ω ) is also connected. It remains to show that (M , ω ) is a Hamiltonian Tn -space with a moment map µ having image µ (M ) = . Let z be such that φ(z) = π ∗ (p) where p is a vertex of . Let σ : Tn → (Td )z be the inverse for the earlier bijection π : (Td )z → Tn . This is a section, i.e., a right inverse for π , in the sequence 1 −→ N

i

−→

Td

π

−→ Tn σ ←−

−→

1,

so it splits, i.e., becomes like a sequence for a product, as we obtain an isomorphism (i, σ ) : N × Tn −→ Td . The action of the Tn factor (or, more rigorously, σ (Tn ) ⊂ Td ) descends to the quotient M = Z/N . Consider the diagram ∗ j φ σ∗ Z → Cd −→ Rd η∗ ⊕ (Rn )∗ −→ (Rn )∗ p↓ M where the last horizontal map is projection onto the second factor. Since the composition of the horizontal maps is constant along N -orbits, it descends to a map µ : M −→ (Rn )∗

178

A. Cannas da Silva

which satisfies µ ◦ p = σ ∗ ◦ φ ◦ j . By reduction for product groups (Section 6.2), this is a moment map for the action of Tn on (M , ω ). The image of µ is µ (M ) = (µ ◦ p)(Z) = (σ ∗ ◦ φ ◦ j )(Z) = (σ ∗ ◦ π ∗ )() = , because φ(Z) = π ∗ () and π ◦ σ = id. We conclude that (M , ω , Tn , µ ) is the required toric manifold corresponding to . This construction via reduction also shows that symplectic toric manifolds are in fact Kähler. E XAMPLE . Here are the details of Delzant’s construction for the case of a segment = [0, a] ⊂ R∗ (n = 1, d = 2). Let v(= 1) be the standard basis vector in R. Then is described by x, −v 0 and x, v a, where v1 = −v, v2 = v, λ1 = 0 and λ2 = a. π The projection R2 −→ R, e1 → −v, e2 → v, has kernel equal to the span of (e1 + e2 ), so that N is the diagonal subgroup of T2 = S 1 × S 1 . The exact sequences become i

π

1 −→ N −→ T2 −→ t −→ (t, t),

S1

−→ 1,

(t1 , t2 ) −→ t1−1 t2 , i

π

R −→ 0, 0 −→ n −→ R2 −→ x −→ (x, x), (x1 , x2 ) −→ x2 − x1 , ∗ i ∗ π∗ 0 −→ R∗ −→ R2 −→ n∗ −→ 0, x −→ (−x, x), (x1 , x2 ) −→ x1 + x2 . The action of the diagonal subgroup N = {(eit , eit ) ∈ S 1 × S 1 } on C2 by it it e , e · (z1 , z2 ) = eit z1 , eit z2 has moment map (i ∗ ◦ φ)(z1 , z2 ) = − 12 (|z1 |2 + |z2 |2 ) + a, with zero-level set (i ∗ ◦ φ)−1 (0) = (z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 2a . Hence, the reduced space is a projective space, (i ∗ ◦ φ)−1 (0)/N = CP1 . 6.6. Duistermaat–Heckman theorems Throughout this subsection, let (M, ω, G, µ) be a Hamiltonian G-space, where G is an n-torus79 and the moment map µ is proper. 79 The discussion in this subsection may be extended to Hamiltonian actions of other compact Lie groups, not necessarily tori; see [66, Exercises 2.1–2.10].

Symplectic geometry

179

If G acts freely on µ−1 (0), it also acts freely on nearby levels µ−1 (t), t ∈ g∗ and t ≈ 0. (Otherwise, assume only that 0 is a regular value of µ and work with orbifolds.) We study the variation of the reduced spaces by relating Mred = µ−1 (0)/G, ωred and Mt = µ−1 (t)/G, ωt . For simplicity, assume G to be the circle S 1 . Let Z = µ−1 (0) and let i : Z → M be the inclusion map. Fix a connection form α ∈ Ω 1 (Z) for the principal bundle S1

→ Z ↓π Mred

that is, LX# α = 0 and ıX# α = 1, where X # is the infinitesimal generator for the S 1 -action. Construct a 2-form on the product manifold Z × (−ε, ε) by the recipe σ = π ∗ ωred − d(xα), where x is a linear coordinate on the interval (−ε, ε) ⊂ R g∗ . (By abuse of notation, we shorten the symbols for forms on Z × (−ε, ε) that arise by pullback via projection onto each factor.) L EMMA 6.12. The 2-form σ is symplectic for ε small enough. ∂ P ROOF. At points where x = 0, the form σ |x=0 = π ∗ ωred + α ∧ dx satisfies σ |x=0 (X # , ∂x ) = 1, so σ is nondegenerate along Z × {0}. Since nondegeneracy is an open condition, we conclude that σ is nondegenerate for x in a sufficiently small neighborhood of 0. Closedness is clear.

Notice that σ is invariant with respect to the S 1 -action on the first factor of Z × (−ε, ε). This action is Hamiltonian with moment map x : Z × (−ε, ε) → (−ε, ε) given by projection onto the second factor (since LX# α = 0 and ıX# α = 1): ıX# σ = −ıX# d(xα) = −LX# (xα) + dıX# (xα) = dx. L EMMA 6.13. There exists an equivariant symplectomorphism between a neighborhood of Z in M and a neighborhood of Z × {0} in Z × (−ε, ε), intertwining the two moment maps, for ε small enough. P ROOF. The inclusion i0 : Z → Z × (−ε, ε) as Z × {0} and the natural inclusion i : Z → M are S 1 -equivariant coisotropic embeddings. Moreover, they satisfy i0∗ σ = i ∗ ω since both sides are equal to π ∗ ωred , and the moment maps coincide on Z because i0∗ x = 0 = i ∗ µ. Replacing ε by a smaller positive number if necessary, the result follows from the equivariant version of the coisotropic embedding theorem (Theorem 2.9).80 80 Equivariant coisotropic embedding theorem: Let (M , ω ), (M , ω ) be symplectic manifolds of dimen0 0 1 1 sion 2n, G a compact Lie group acting on (Mi , ωi ), i = 0, 1, in a Hamiltonian way with moment maps µ0

180

A. Cannas da Silva

Therefore, in order to compare the reduced spaces Mt = µ−1 (t)/S 1 for t ≈ 0, we can work in Z × (−ε, ε) and compare instead the reduced spaces x −1 (t)/S 1 . P ROPOSITION 6.14. The space (Mt , ωt ) is symplectomorphic to (Mred , ωred − tβ) where β is the curvature form of the connection α. P ROOF. By Lemma 6.13, (Mt , ωt ) is symplectomorphic to the reduced space at level t for the Hamiltonian space (Z × (−ε, ε), σ, S 1 , x). Since x −1 (t) = Z × {t}, where S 1 acts on the first factor, all the manifolds x −1 (t)/S 1 are diffeomorphic to Z/S 1 = Mred . As for the symplectic forms, let ιt : Z × {t} → Z × (−ε, ε) be the inclusion map. The restriction of σ to Z × {t} is ι∗t σ = π ∗ ωred − t dα. By definition of curvature, dα = π ∗ β. Hence, the reduced symplectic form on x −1 (t)/S 1 is ωred − tβ. In loose terms, Proposition 6.14 says that the reduced forms ωt vary linearly in t, for t close enough to 0. However, the identification of Mt with Mred as abstract manifolds is not natural. Nonetheless, any two such identifications are isotopic. By the homotopy invariance of de Rham classes, we obtain: T HEOREM 6.15 (Duistermaat–Heckman [38]). Under the hypotheses and notation before, the cohomology class of the reduced symplectic form [ωt ] varies linearly in t. More specif2 (Mred ) is the first Chern class81 of the S 1 -bundle Z → Mred , ically, if c = [−β] ∈ HdeRham we have [ωt ] = [ωred ] + tc. and µ1 , respectively, Z a manifold of dimension k n with a G-action, and ιi : Z → Mi , i = 0, 1, G-equivariant coisotropic embeddings. Suppose that ι∗0 ω0 = ι∗1 ω1 and ι∗0 µ0 = ι∗1 µ1 . Then there exist G-invariant neighborhoods U0 and U1 of ι0 (Z) and ι1 (Z) in M0 and M1 , respectively, and a G-equivariant symplectomorphism ϕ : U0 → U1 such that ϕ ◦ ι0 = ι1 and µ0 = ϕ ∗ µ1 . 81 Often the Lie algebra of S 1 is identified with 2π iR under the exponential map exp : g 2π iR → S 1 , ξ → eξ . Given a principal S 1 -bundle, by this identification the infinitesimal action maps the generator 2π i of 2π iR to the generating vector field X# . A connection form A is then an imaginary-valued 1-form on the total space satisfying LX# A = 0 and ıX# A = 2π i. Its curvature form B is an imaginary-valued 2-form on the base satisfying i B]. π ∗ B = dA. By the Chern–Weil isomorphism, the first Chern class of the principal S 1 -bundle is c = [ 2π 1 Here we identify the Lie algebra of S with R and implicitly use the exponential map exp : g R → S 1 , t → e2π it . Hence, given a principal S 1 -bundle, the infinitesimal action maps the generator 1 of R to X# , and here a connection form α is an ordinary 1-form on the total space satisfying LX# α = 0 and ıX# α = 1. The curvature form β is an ordinary 2-form on the base satisfying π ∗ β = dα. Consequently, we have A = 2π iα, B = 2π iβ and the first Chern class is given by c = [−β].

Symplectic geometry

181

D EFINITION 6.16. The Duistermaat–Heckman measure, mDH , on g∗ is the push-forward of the Liouville measure82 by µ : M → g∗ , that is, for any Borel subset U of g∗ , we have mDH (U) =

ωn . µ−1 (U ) n!

The integral with respect to the Duistermaat–Heckman measure of a compactlysupported function h ∈ C ∞ (g∗ ) is g∗

h dmDH :=

(h ◦ µ) M

ωn . n!

On g∗ regarded as a vector space, say Rn , there is also the Lebesgue (or Euclidean) measure, m0 . The relation between mDH and m0 is governed by the Radon–Nikodym derivative, DH denoted by dm dm0 , which is a generalized function satisfying

g∗

h dmDH =

g∗

h

dmDH dm0 . dm0

T HEOREM 6.17 (Duistermaat–Heckman [38]). Under the hypotheses and notation before, the Duistermaat–Heckman measure is a piecewise polynomial multiple of Lebesgue DH measure on g∗ Rn , that is, the Radon–Nikodym derivative f = dm polydm0 is piecewise

∗ nomial. More specifically, for any Borel subset U of g , we have mDH (U) = U f (x) dx, where dx = dm0 is the Lebesgue volume form on U and f : g∗ Rn → R is polynomial on any region consisting of regular values of µ. This Radon–Nikodym derivative f is called the Duistermaat–Heckman polynomial. In the case of a toric manifold, the Duistermaat–Heckman polynomial is a universal constant equal to (2π)n when is n-dimensional. Thus the symplectic volume of (M , ω ) is (2π)n times the Euclidean volume of . E XAMPLE . For the standard spinning of a sphere (S 2 , ω = dθ ∧ dh, S 1 , µ = h), the image of µ is the interval [−1, 1]. The Lebesgue measure of [a, b] ⊆ [−1, 1] is m0 ([a, b]) = b − a. The Duistermaat–Heckman measure of [a, b] is mDH [a, b] =

{(θ,h)∈S 2 | ahb}

dθ dh = 2π(b − a),

82 On an arbitrary symplectic manifold (M 2n , ω), with symplectic volume ωn , the Liouville measure (or symn!

plectic measure) of a Borel subset U of M is ωn . mω (U) = U n!

The set B of Borel subsets is the σ -ring generated by the set of compact subsets, i.e., if A, B ∈ B, then A \ B ∈ B, $ and if Ai ∈ B, i = 1, 2, . . . , then ∞ i=1 Ai ∈ B.

182

A. Cannas da Silva

i.e., mDH = 2πm0 . Consequently, the area of the spherical region between two parallel planes depends only on the distance between the planes, a result that was known to Archimedes around 230 BC. P ROOF. We sketch the proof of Theorem 6.17 for the case G = S 1 . The proof for the general case, which follows along similar lines, can be found in, for instance, [66], besides the original articles. Let (M, ω, S 1 , µ) be a Hamiltonian S 1 -space of dimension 2n and let (Mx , ωx ) be its reduced space at level x. Proposition 6.14 or Theorem 6.15 imply that, for x in a sufficiently narrow neighborhood of 0, the symplectic volume of Mx , vol(Mx ) = Mx

ωxn−1 = (n − 1)!

Mred

(ωred − xβ)n−1 , (n − 1)!

is a polynomial in x of degree n − 1. This volume can be also expressed as vol(Mx ) = Z

π ∗ (ωred − xβ)n−1 ∧ α, (n − 1)!

where α is a connection form for the S 1 -bundle Z → Mred and β is its curvature form. Now we go back to the computation of the Duistermaat–Heckman measure. For a Borel subset U of (−ε, ε), the Duistermaat–Heckman measure is, by definition, mDH (U) =

ωn . µ−1 (U ) n!

Using the fact that (µ−1 (−ε, ε), ω) is symplectomorphic to (Z × (−ε, ε), σ ) and, moreover, they are isomorphic as Hamiltonian S 1 -spaces, we obtain mDH (U) =

σn . Z×U n!

Since σ = π ∗ ωred − d(xα), its power is σ n = n(π ∗ ωred − x dα)n−1 ∧ α ∧ dx. By the Fubini theorem, we then have mDH (U) =

U

Z

π ∗ (ωred − xβ)n−1 ∧ α ∧ dx. (n − 1)!

Therefore, the Radon–Nikodym derivative of mDH with respect to the Lebesgue measure, dx, is f (x) = Z

π ∗ (ωred − xβ)n−1 ∧ α = vol(Mx ). (n − 1)!

Symplectic geometry

183

The previous discussion proves that, for x ≈ 0, f (x) is a polynomial in x. The same holds for a neighborhood of any other regular value of µ, because we may change the moment map µ by an arbitrary additive constant. Duistermaat and Heckman [38] also applied these results when M is compact to provide

X n a formula for the oscillatory integral M eiµ ωn! for X ∈ g as a sum of contributions of the fixed points of the action of the one-parameter subgroup generated by X. They hence showed that the stationary phase approximation83 is exact in the case of the moment map. When G is a maximal torus of a compact connected simple Lie group acting on a coadjoint orbit, the Duistermaat–Heckman formula reduces to the Harish–Chandra formula. It was observed by Berline and Vergne [14] and by Atiyah and Bott [5] that the Duistermaat– Heckman formula can be derived by localization in equivariant cohomology. This is an instance of Abelian localization, i.e., a formula for an integral (in equivariant cohomology) in terms of data at the fixed points of the action, and typically is used for the case of Abelian groups (or of maximal tori). Later non-Abelian localization formulas were found, where integrals (in equivariant cohomology) are expressed in terms of data at the zeros of the moment map, normally used for the case of non-Abelian groups. Both localizations gave rise to computations of the cohomology ring structure of reduced spaces [80].

References [1] V. Arnold, On a characteristic class entering into conditions of quantization, Funkcional Anal. i Priložen 1 (1967), 1–14. [2] V. Arnold, Mathematical Methods of Classical Mechanics, Graduate Texts in Math., vol. 60, SpringerVerlag, New York (1978). [3] V. Arnold and A. Givental, Symplectic geometry, Dynamical Systems IV, Symplectic Geometry and Its Applications, V. Arnold and S. Novikov, eds, Encyclopaedia of Math. Sciences, vol. 4, Springer-Verlag, Berlin (1990). [4] M. Atiyah, Convexity and commuting Hamiltonians, Bull. London Math. Soc. 14 (1982), 1–15. [5] M. Atiyah and R. Bott, The moment map and equivariant cohomology, Topology 23 (1984), 1–28. [6] M. Atiyah and R. Bott, The Yang–Mills equations over Riemann surfaces, Topology 23 (1984), 1–28; Philos. Trans. Roy. Soc. London 308 (1983), 523–615. [7] M. Audin, Exemples de variétés presque complexes, Enseign. Math. 37 (1991), 175–190. [8] M. Audin, Spinning Tops, A Course on Integrable Systems, Cambridge Studies in Advanced Mathematics, vol. 51, Cambridge Univ. Press, Cambridge (1996). [9] M. Audin, Lagrangian submanifolds, Symplectic Geometry of Integrable Hamiltonian Systems (Barcelona, 2001), Adv. Courses Math. CRM Barcelona, Birkhäuser, Basel (2003), 1–83. [10] M. Audin, Torus Actions on Symplectic Manifolds, Progress in Mathematics, vol. 93, Birkhäuser, Basel (2004). [11] M. Audin and J. Lafontaine, eds, Holomorphic Curves in Symplectic Geometry, Progress in Mathematics, vol. 117, Birkhäuser, Basel (1994). [12] D. Auroux, Asymptotically holomorphic families of symplectic submanifolds, Geom. Funct. Anal. 7 (1997), 971–995. [13] D. Auroux, Symplectic 4-manifolds as branched coverings of CP2 , Invent. Math. 139 (2000), 551–602. [14] N. Berline and M. Vergne, Zéros d’un champ de vecteurs et classes caractéristiques équivariantes, Duke Math. J. 50 (1983), 539–549.

83 The stationary phase lemma gives the asymptotic behavior (for large N ) of integrals ( N )n ig M f e vol, where 2π f and g are real functions and vol is a volume form on a 2n-dimensional manifold M.

184

A. Cannas da Silva

[15] P. Biran, A stability property of symplectic packing, Invent. Math. 136 (1999), 123–155. [16] P. Biran, Geometry of symplectic intersections, Proceedings of the I.C.M., vol. II (Beijing, 2002), Higher Ed. Press, Beijing (2002), 241–255. [17] P. Biran and K. Cieliebak, Symplectic topology on subcritical manifolds, Comment. Math. Helv. 76 (2001), 712–753. [18] G. Birkhoff, Dynamical Systems, Amer. Math. Soc. Colloquium Publications, vol. IX, Amer. Math. Soc., Providence (1966), reprinting of the original 1927 edition, with an addendum by J. Moser. [19] G. Bredon, Introduction to Compact Transformation Groups, Pure and Applied Mathematics, vol. 46, Academic Press, New York (1972). [20] R. Bryant, An introduction to Lie groups and symplectic geometry, Geometry and Quantum Field Theory (Park City, UT, 1991), IAS/Park City Math. Ser., vol. 1, Amer. Math. Soc., Providence (1995), 5–181. [21] A. Cannas da Silva, Symplectic toric manifolds, Symplectic Geometry of Integrable Hamiltonian Systems (Barcelona, 2001), Adv. Courses Math. CRM Barcelona, Birkhäuser, Basel (2003), 1–83. [22] J. Cheeger, Finiteness theorems for Riemannian manifolds, Amer. J. Math. 92 (1970), 61–74. [23] S.S. Chern, Complex Manifolds Without Potential Theory, 2nd ed., Universitext, Springer-Verlag, New York (1979), with an appendix on the geometry of characteristic classes. [24] C. Conley and E. Zehnder, The Birkhoff–Lewis fixed point theorem and a conjecture of V.I. Arnold, Invent. Math. 73 (1983), 33–49. [25] C. Conley and E. Zehnder, Morse-type index theory for flows and periodic solutions for Hamiltonian equations, Comm. Pure Appl. Math. 37 (1984), 207–253. [26] D. Cox, Recent developments in toric geometry, Algebraic Geometry—Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, part 2, Amer. Math. Soc., Providence (1997), 389–436. [27] V. Danilov, The geometry of toric varieties, Uspekhi Mat. Nauk 33 (2) (1978), (200), 85–134, 247; English translation: Russian Math. Surveys 33 (2) (1978), 97–154. [28] T. Delzant, Hamiltoniens périodiques et images convexes de l’application moment, Bull. Soc. Math. France 116 (1988), 315–339. [29] M. Demazure, Sous-groupes algébriques de rang maximum du groupe de Cremona, Ann. Sci. École Norm. Sup. (4) 3 (1970), 507–588. [30] S. Donaldson, An application of gauge theory to four-dimensional topology, J. Differential Geom. 18 (1983), 279–315. [31] S. Donaldson, Irrationality and the h-cobordism conjecture, J. Differential Geom. 26 (1987), 141–168. [32] S. Donaldson, Symplectic submanifolds and almost-complex geometry, J. Differential Geom. 44 (1996), 666–705. [33] S. Donaldson, Lefschetz fibrations in symplectic geometry, Proceedings of the I.C.M., vol. II (Berlin, 1998), Doc. Math., extra vol. II (1998), 309–314. [34] S. Donaldson, Lefschetz pencils on symplectic manifolds, J. Differential Geom. 53 (1999), 205–236. [35] S. Donaldson, Floer Homology Groups in Yang–Mills Theory, Cambridge Tracts in Mathematics, vol. 147, Cambridge Univ. Press, Cambridge (2002), with the assistance of M. Furuta and D. Kotschick. [36] J.J. Duistermaat, On global action-angle coordinates, Comm. Pure Appl. Math. 33 (1980), 687–706. [37] J.J. Duistermaat, The Heat Kernel Lefschetz Fixed Point Formula for the Spin-c Dirac Operator, Progress in Nonlinear Differential Equations and their Applications, vol. 18, Birkhäuser, Boston (1996). [38] J.J. Duistermaat and G. Heckman, On the variation in the cohomology of the symplectic form of the reduced phase space, Invent. Math. 69 (1982), 259–268; Addendum, Invent. Math. 72 (1983), 153–158. [39] Y. Eliashberg, Estimates on the number of fixed points of area-preserving transformations, Syktyvkar University, preprint (1979). [40] Y. Eliashberg, A. Givental and H. Hofer, Introduction to symplectic field theory, GAFA 2000 (Tel Aviv, 1999), Geom. Funct. Anal., special volume, part II (2000), 560–673. [41] Y. Eliashberg and M. Gromov, Lagrangian intersection theory: finite-dimensional approach, Geometry of Differential Equations, Amer. Math. Soc. Transl. Ser. 2, vol. 186, Amer. Math. Soc., Providence (1998), 27–118. [42] Y. Eliashberg and N. Mishachev, Introduction to the h-Principle, Graduate Studies in Mathematics, vol. 48, Amer. Math. Soc., Providence (2002).

Symplectic geometry

185

[43] Y. Eliashberg and L. Traynor, eds, Symplectic Geometry and Topology, Lectures from the Graduate Summer School held in Park City, June 29–July 19, 1997, IAS/Park City Mathematics Series, vol. 7, Amer. Math. Soc., Providence (1999). [44] M. Fernández, M. Gotay and A. Gray, Compact parallelizable four-dimensional symplectic and complex manifolds, Proc. Amer. Math. Soc. 103 (1988), 1209–1212. [45] R. Fintushel and R. Stern, Knots, links, and 4-manifolds, Invent. Math. 134 (1998), 363–400. [46] A. Floer, A relative Morse index for the symplectic action, Comm. Pure Appl. Math. 41 (1988), 393–407. [47] A. Floer, The unregularized gradient flow of the symplectic action, Comm. Pure Appl. Math. 41 (1988), 775–813. [48] A. Floer, Morse theory for Lagrangian intersections, J. Differential Geom. 28 (1988), 513–547. [49] A. Floer, Symplectic fixed points and holomorphic spheres, Comm. Math. Phys. 120 (1989), 575–611. [50] A. Floer, Witten’s complex and infinite-dimensional Morse theory, J. Differential Geom. 30 (1989), 207– 221. [51] M. Freedman, The topology of four-dimensional manifolds, J. Differential Geom. 17 (1982), 357–453. [52] K. Fukaya and K. Ono, Arnold conjecture and Gromov–Witten invariant, Topology 38 (1999), 933–1048. [53] W. Fulton, Introduction to Toric Varieties, Annals of Mathematics Studies, vol. 131, Princeton Univ. Press, Princeton (1993). [54] D. Gay and R. Kirby, Constructing symplectic forms on 4-manifolds which vanish on circles, Geom. Topol. 8 (2004), 743–777. [55] A. Givental, Periodic mappings in symplectic topology, Funktsional. Anal. i Prilozhen 23 (1989), 37–52 (in Russian); translation in Funct. Anal. Appl. 23 (1989), 287–300. [56] M. Golubitsky and D. Tischler, An example of moduli for singular symplectic forms, Invent. Math. 38 (1976/1977), 219–225. [57] R. Gompf, A new construction of symplectic manifolds, Ann. of Math. 142 (1995), 527–595. [58] R. Gompf, Symplectic structures from Lefschetz pencils in high dimensions, Geom. Topol. Monogr. 7 (2004), 267–290. [59] R. Gompf, Toward a topological characterization of symplectic manifolds, J. Symp. Geom., in press. [60] R. Gompf and A. Stipsicz, 4-Manifolds and Kirby Calculus, Graduate Studies in Mathematics, vol. 20, Amer. Math. Soc., Providence (1999). [61] M. Gotay, On coisotropic imbeddings of presymplectic manifolds, Proc. Amer. Math. Soc. 84 (1982), 111–114. [62] P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley Classics Library, Wiley, New York (1994), reprint of the 1978 original. [63] M. Gromov, Stable mappings of foliations into manifolds, Izv. Akad. Nauk SSSR Ser. Mat. 33 (1969), 707–734. [64] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math. 82 (1985), 307–347. [65] M. Gromov, Partial Differential Relations, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 9, Springer-Verlag, Berlin (1986). [66] V. Guillemin, Moment Maps and Combinatorial Invariants of Hamiltonian T n -spaces, Progress in Mathematics, vol. 122, Birkhäuser, Boston (1994). [67] V. Guillemin, V. Ginzburg and Y. Karshon, Moment Maps, Cobordisms, and Hamiltonian Group Actions, Mathematical Surveys and Monographs, vol. 98, Amer. Math. Soc., Providence (2002), with appendix J by M. Braverman. [68] V. Guillemin and S. Sternberg, Convexity properties of the moment mapping, Invent. Math. 67 (1982), 491–513. [69] V. Guillemin and S. Sternberg, Birational equivalence in the symplectic category, Invent. Math. 97 (1989), 485–522. [70] V. Guillemin and S. Sternberg, Symplectic Techniques in Physics, 2nd ed., Cambridge Univ. Press, Cambridge (1990). [71] R. Harvey and H.B. Lawson, Calibrated geometries, Acta Math. 148 (1982), 47–157. [72] N. Hitchin, G. Segal and R. Ward, Integrable Systems. Twistors, Loop groups, and Riemann Surfaces, Oxford Graduate Texts in Mathematics, vol. 4, Clarendon, Oxford Univ. Press, New York (1999). [73] W. Hodge, The Theory and Applications of Harmonic Integrals, 2nd ed., Cambridge Univ. Press, Cambridge (1952).

186

A. Cannas da Silva

[74] H. Hofer and D. Salamon, Floer homology and Novikov rings, The Floer Memorial Volume, Progress in Mathematics, vol. 133, Birkhäuser, Basel (1995), 483–524. [75] K. Honda, Transversality theorems for harmonic forms, Rocky Mountain J. Math. 34 (2004), 629–664. [76] L. Hörmander, An Introduction to Complex Analysis in Several Variables, 3rd ed., North-Holland Mathematical Library, vol. 7, North-Holland, Amsterdam (1990). [77] N. Jacobson, Lie Algebras, Dover, New York (1979), republication of the 1962 original. [78] L. Jeffrey and F. Kirwan, Localization for non-Abelian group actions, Topology 34 (1995), 291–327. [79] G. Kempf, F. Knudsen, D. Mumford and B. Saint-Donat, Toroidal Embeddings, I, Lecture Notes in Mathematics, vol. 339, Springer-Verlag, Berlin (1973). [80] F. Kirwan, Cohomology of Quotients in Symplectic and Algebraic Geometry, Mathematical Notes, vol. 31, Princeton Univ. Press, Princeton (1984). [81] F. Kirwan, Convexity properties of the moment mapping, III, Invent. Math. 77 (1984), 547–552. [82] K. Kodaira, On the structure of compact complex analytic surfaces, I, Amer. J. Math. 86 (1964), 751–798. [83] J. Kohn, Harmonic integrals on strongly pseudo-convex manifolds, I, Ann. of Math. 78 (1963) 112–148. [84] D. Kotschick, On manifolds homeomorphic to CP2 #8 CP2 , Invent. Math. 95 (1989), 591–600. [85] F. Lalonde and D. McDuff, J-curves and the classification of rational and ruled symplectic 4-manifolds, Contact and Symplectic Geometry (Cambridge, 1994), Publ. Newton Inst., vol. 8, Cambridge Univ. Press, Cambridge (1996), 3–42. [86] E. Lerman, Symplectic cuts, Math. Res. Lett. 2 (1995), 247–258. [87] E. Lerman and S. Tolman, Hamiltonian torus actions on symplectic orbifolds and toric varieties, Trans. Amer. Math. Soc. 349 (1997), 4201–4230. [88] T. Li and A. Liu, Symplectic structure on ruled surfaces and a generalized adjunction formula, Math. Res. Lett. 2 (1995), 453–471. [89] A.-K. Liu, Some new applications of general wall crossing formula, Gompf’s conjecture and its applications, Math. Res. Lett. 3 (1996), 569–585. [90] G. Liu and G. Tian, Floer homology and Arnold conjecture, J. Differential Geom. 49 (1998), 1–74. [91] J. Marsden and T. Ratiu, Introduction to Mechanics and Symmetry, A Basic Exposition of Classical Mechanical Systems, Texts in Applied Mathematics, vol. 17, Springer-Verlag, New York (1994). [92] J. Marsden and A. Weinstein, Reduction of symplectic manifolds with symmetry, Rep. Math. Phys. 5 (1974), 121–130. [93] V. Maslov, Perturbation Theory and Asymptotic Methods, Izdat. Moskov Univ., Moscow (1965) (in Russian). [94] D. McDuff, Examples of simply-connected symplectic non-Kählerian manifolds, J. Differential Geom. 20 (1984), 267–277. [95] D. McDuff, Rational and ruled symplectic 4-manifolds, Geometry of low-dimensional manifolds 2 (Durham, 1989), London Math. Soc. Lecture Note Ser., vol. 151, Cambridge Univ. Press, Cambridge (1990), 7–14. [96] D. McDuff, The structure of rational and ruled symplectic 4-manifolds, J. Amer. Math. Soc. 3 (1990), 679–712. [97] D. McDuff, Lectures on groups of symplectomorphisms, Rend. Circ. Mat. Palermo (2) 72 (2004), 43–78. [98] D. McDuff and L. Polterovich, Symplectic packings and algebraic geometry, Invent. Math. 115 (1994), 405–434, with an appendix by Yael Karshon. [99] D. McDuff and D. Salamon, Introduction to Symplectic Topology, Oxford Mathematical Monographs, Oxford Univ. Press, New York (1995). [100] D. McDuff and D. Salamon, J-holomorphic Curves and Symplectic Topology, Amer. Math. Soc. Colloquium Publications, vol. 52, Amer. Math. Soc., Providence (2004). [101] C. McMullen and C. Taubes, 4-manifolds with inequivalent symplectic forms and 3-manifolds with inequivalent fibrations, Math. Res. Lett. 6 (1999), 681–696. [102] K. Meyer, Symmetries and integrals in mechanics, Dynamical Systems (Proc. Sympos., Univ. Bahia, Salvador, 1971), Academic Press, New York (1973), 259–272. [103] J. Milnor, Morse Theory, Annals of Mathematics Studies, vol. 51, Princeton Univ. Press, Princeton (1963), based on lecture notes by M. Spivak and R. Wells. [104] M. Morse, The foundations of a theory in the calculus of variations in the large, Trans. Amer. Math. Soc. 30 (1928), 213–274.

Symplectic geometry

187

[105] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965), 286–294. [106] A. Newlander and L. Nirenberg, Complex analytic coordinates in almost complex manifolds, Ann. of Math. 65 (1957), 391–404. [107] T. Oda, Convex Bodies and Algebraic Geometry—An Introduction to the Theory of Toric Varieties, Ergebnisse der Mathematik und ihrer Grenzgebiete (3), vol. 15, Springer-Verlag, Berlin (1988). [108] K. Ono, On the Arnold conjecture for weakly monotone symplectic manifolds, Invent. Math. 119 (1995), 519–537. [109] J. Park, Non-complex symplectic 4-manifolds with b2+ = 1, Bull. London Math. Soc. 36 (2004), 231–240.

[110] J. Park, Simply connected symplectic 4-manifolds with b2+ = 1 and c12 = 2, Invent. Math., in press. [111] Y. Ruan, Symplectic topology on algebraic 3-folds, J. Differential Geom. 39 (1994), 215–227. [112] Y. Ruan, Topological sigma model and Donaldson-type invariants in Gromov theory, Duke Math. J. 83 (1996), 461–500. [113] D. Salamon, Lectures on Floer homology, Symplectic Geometry and Topology, Y. Eliashberg and L. Traynor, eds, IAS/Park City Math. Ser., vol. 7, Amer. Math. Soc., Providence (1999), 143–229. [114] I. Satake, On a generalization of the notion of manifold, Proc. Nat. Acad. Sci. U.S.A. 42 (1956), 359–363. [115] P. Seidel, Graded Lagrangian submanifolds, Bull. Soc. Math. France 128 (2000), 103–149. [116] J.-C. Sikorav, Points fixes d’un symplectomorphisme homologue à l’identité, C. R. Acad. Sci. Paris Sér. I Math. 299 (1984), 343–346. [117] I. Smith, On moduli spaces of symplectic forms, Math. Res. Lett. 7 (2000), 779–788. [118] I. Smith, Geometric monodromy and the hyperbolic disc, Quart. J. Math. 52 (2001), 217–228. [119] J.-M. Souriau, Structure des Systèmes Dynamiques, Maî trises de Mathématiques, Dunod, Paris (1970). [120] M. Spivak, A Comprehensive Introduction to Differential Geometry, vol. I, 2nd ed., Publish or Perish, Wilmington (1979). [121] N. Steenrod, The Topology of Fibre Bundles, Princeton Mathematical Series, vol. 14, Princeton Univ. Press, Princeton (1951). [122] A. Stipsicz, The geography problem of 4-manifolds with various structures, Acta Math. Hungar. 7 (2000), 267–278. [123] M. Symington, Symplectic rational blowdowns, J. Differential Geom. 50 (1998), 505–518. [124] Z. Szabó, Exotic 4-manifolds with b2+ = 1, Math. Res. Lett. 3 (1996), 731–741. [125] Z. Szabó, Simply-connected irreducible 4-manifolds with no symplectic structures, Invent. Math. 132 (1998), 457–466. [126] C. Taubes, The Seiberg–Witten invariants and symplectic forms, Math. Res. Lett. 1 (1994), 809–822. [127] C. Taubes, More constraints on symplectic forms from Seiberg–Witten invariants, Math. Res. Lett. 2 (1995), 9–13. [128] C. Taubes, The Seiberg–Witten and Gromov invariants, Math. Res. Lett. 2 (1995), 221–238. [129] C. Taubes, SW ⇒ Gr: from the Seiberg–Witten equations to pseudoholomorphic curves, J. Amer. Math. Soc. 9 (1996), 845–918. [130] C. Taubes, Seiberg–Witten invariants and pseudo-holomorphic subvarieties for self-dual, harmonic 2forms, Geom. Topol. 3 (1999), 167–210. [131] W. Thurston, Some simple examples of symplectic manifolds, Proc. Amer. Math. Soc. 55 (1976), 467–468. [132] D. Tischler, Closed 2-forms and an embedding theorem for symplectic manifolds, J. Differential Geom. 12 (1977), 229–235. [133] S. Tolman and J. Weitsman, The cohomology rings of symplectic quotients, Comm. Anal. Geom. 11 (2003), 751–773. [134] L. Traynor, Symplectic packing constructions, J. Differential Geom. 42 (1995), 411–429. [135] C. Viterbo, Symplectic topology as the geometry of generating functions, Math. Ann. 292 (1992), 685–710. [136] A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds, Adv. in Math. 6 (1971), 329–346. [137] A. Weinstein, Lectures on Symplectic Manifolds, Regional Conference Series in Mathematics, vol. 29, Amer. Math. Soc., Providence (1977). [138] A. Weinstein, Fat bundles and symplectic manifolds, Adv. in Math. 37 (1980), 239–250. [139] A. Weinstein, Neighborhood classification of isotropic embeddings, J. Differential Geom. 16 (1981), 125– 128.

188

A. Cannas da Silva

[140] A. Weinstein, On extending the Conley–Zehnder fixed point theorem to other manifolds, Nonlinear Functional Analysis and its Applications, Part 2 (Berkeley, 1983), Proc. Sympos. Pure Math., vol. 45, Part 2, Amer. Math. Soc., Providence (1986), 541–544. [141] R.O. Wells, Differential Analysis on Complex Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 65, Springer-Verlag, New York (1980). [142] H. Weyl, The Classical Groups. Their Invariants and Representations, Princeton Landmarks in Mathematics, Princeton Univ. Press, Princeton (1997). [143] H. Whitney, Analytic extensions of differentiable functions defined in closed sets, Trans. Amer. Math. Soc. 36 (1934), 63–89. [144] E. Witten, Supersymmetry and Morse theory, J. Differential Geom. 17 (1982), 661–692. [145] E. Witten, Topological sigma models, Comm. Math. Phys. 118 (1988), 411–449. [146] W.-T. Wu, Sur les classes caractíristiques des structures fibrées sphériques, Actualités Sci. Ind. 1183 (1952).

CHAPTER 4

Metric Riemannian Geometry Kenji Fukaya∗ Department of Mathematics, Kyoto University, Kitashirakawa Oiwake Cho, Sakyo-ku, Kyoto 606-8502, Japan E-mail: [email protected]

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1. Notations used in this article . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Sphere theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Finiteness theorems and Gromov–Hausdorff distance . . . . . . . . . . . . . . . 4. Geodesic coordinate, injectivity radius, comparison theorems and sphere theorem 5. Packing and precompactness theorem . . . . . . . . . . . . . . . . . . . . . . . . 6. Construction of homeomorphism by isotopy theory . . . . . . . . . . . . . . . . 7. Harmonic coordinate and its application . . . . . . . . . . . . . . . . . . . . . . . 8. Center of mass technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Embedding Riemannian manifolds by distance function . . . . . . . . . . . . . . 10. Almost flat manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Collapsing Riemannian manifolds—I . . . . . . . . . . . . . . . . . . . . . . . . 12. Collapsing Riemannian manifolds—II . . . . . . . . . . . . . . . . . . . . . . . . 13. Collapsing Riemannian manifolds—III . . . . . . . . . . . . . . . . . . . . . . . 14. Morse theory of distance function . . . . . . . . . . . . . . . . . . . . . . . . . . 15. Finiteness theorem by Morse theory . . . . . . . . . . . . . . . . . . . . . . . . . 16. Soul theorem and splitting theorem . . . . . . . . . . . . . . . . . . . . . . . . . 17. Alexandrov space—I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18. Alexandrov space—II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19. First Betti number and fundamental group . . . . . . . . . . . . . . . . . . . . . . 20. Hausdorff convergence of Einstein manifolds . . . . . . . . . . . . . . . . . . . . 21. Sphere theorem and L2 comparison theorem . . . . . . . . . . . . . . . . . . . . 22. Hausdorff convergence and Ricci curvature—I . . . . . . . . . . . . . . . . . . . 23. Hausdorff convergence and Ricci curvature—II . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

191 193 194 195 198 204 208 210 212 215 217 220 224 229 232 237 239 245 254 263 271 274 282 291 308

* The author thanks Professors A. Kasue, Y. Otsu, T. Sakai, T. Shioya and T. Yamaguchi (the authors of [92]) who

provided detailed information on the topics covered in this article (those which are written or not written in [92]). This article grew from the (English translation of) the introduction of [92]. The author also thanks Professor F. Dillen for his patience and encouragement. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen Published by Elsevier B.V. 189

This page intentionally left blank

Metric Riemannian geometry

191

1. Introduction This article is a survey of (a part of) Riemannian geometry. Riemannian geometry is a huge area which occupies, I believe, at least 1/3 of the whole of differential geometry. So obviously we need to restrict our attention to some part of it to write an article in this handbook. (M. Berger’s books [20,19] deals with wider topics.) Let me mention first what is not included in this article but should have been included in a survey of Riemannian geometry. (1) We do not include an elementary or introductory part of Riemannian geometry. For example, topics covered in [103, Sections II, III] or [97] are not in this article. We assume the reader to have some knowledge about it. (2) We focus our attention to global results, and results of local nature are rarely discussed. (3) One powerful tool to study global Riemannian geometry is partial differential equations, especially nonlinear one. We do not discuss it.1 The theory of geodesics (which is a theory of nonlinear ordinary differential equations) is one of the main tools used in this article. Linear partial differential equations, especially the Laplacian, is mentioned only when it is closely related to the other topics included in this article. (4) We do not discuss manifolds of nonpositive curvature. (5) We do not discuss scalar curvature. After removing so many important and interesting topics there are still many things missing in this article. For example, results such as filling volume [73] is not discussed. The study of closed geodesics is not included either. So what is included in this article? We focus the part of Riemannian geometry which describes relations of curvature (sectional or Ricci curvature) to topology of the underlying manifold. Since we do not discuss nonpositively curved manifolds, the main target is manifolds of (almost) nonnegative curvature and more generally the class of manifolds with curvature bounded from below. The study of such Riemannian manifolds started with sphere theorems in the 50’s where comparison theorems are introduced by Rauch as an important tool of study. At the beginning of the 70’s Cheeger (and Weinstein) proved finiteness theorems which provide another kind of statements to be established other than sphere theorems. Soon after that, M. Gromov introduced many new ideas, results and tools, such as Gromov– Hausdorff convergence, almost flat manifold theorem, Betti number estimate, etc., and gave tremendous influence to the area. These present the first turning point of the development of metric Riemannian geometry. In the 1980’s global Riemannian geometry was a very rapidly developing area. Especially the class of Riemannian manifolds with sectional curvature bounded from below and above are studied extensively. An important progress in the 1980’s is the theory of collapsing Riemannian manifolds. Those topics are discussed in Sections 2–13. After a brief review of sphere theorems in Section 2, we describe finiteness theorems in Section 3. In Section 4, while explaining a rough sketch of proofs of sphere theorems, we review several basic facts on global Rie1 So, for example, famous result by Hamilton on the 3 manifold of positive Ricci curvature is not discussed.

192

K. Fukaya

mannian geometry, such as Rauch’s comparison theorem, cut points, conjugate points, injectivity radius, etc. One of the main tools of global Riemannian geometry is the Gromov– Hausdorff distance, which we define in Section 5, and we will prove Gromov’s precompactness theorem. The proofs of finiteness theorems are discussed in Sections 6–9. We try to sketch various (different) techniques used to prove finiteness theorems etc. there, rather than to concentrate on one method and to give its full details. Collapsing Riemannian manifolds (under the bound of the absolute value of sectional curvature) is discussed in Sections 10–13. In Sections 14–18, we discuss the class of Riemannian manifolds with sectional curvature bounded from below (but not above). The basic tool to study it is Morse theory of the distance function, which was initiated by Grove–Shiohama. We discuss it and its application to sphere theorems in Section 14. We explain applications of the same method to finiteness theorems in Section 15. The theme of Section 16 is noncompact manifolds of nonnegative curvature. Besides its own interest, it is used in many places to study compact Riemannian manifold. Our focus in this article is on the compact case, so we restrict our discussion on noncompact manifolds to ones which have a direct application to compact manifolds. New turning points of the development of metric Riemannian geometry came at some point in the 1990’s when several mathematicians belonging to the new generation (such as Perelman and Colding) began to work in this field. In Sections 17 and 18 we discuss Alexandrov spaces. They are metric spaces which have curvature > −∞ in some generalized sense. The notion of curvature on a metric space which is not a manifold was introduced by Alexandrov a long time ago. Recently various applications of it to Riemannian geometry (study of smooth Riemannian manifolds) were discovered. It makes this topic more popular among Riemannian geometers. An important structure theorem of Alexandrov spaces is obtained by Perelman and his collaborators, which we review in Sections 17 and 18. In Sections 19–23 we discuss the class of Riemannian manifolds with Ricci curvature bounded from below. The first Betti number and the fundamental group are topics studied extensively under this curvature assumption. We review some of such studies in Section 19. The theme of Section 20 is (mainly) a special case, that is the case of Einstein manifolds. Our discussion of Einstein manifolds is restricted to those related to the other parts of this article. We discuss Einstein manifolds here since they provide rich examples of a new phenomenon which appears when we replace the assumption sectional curvature const, by the Ricci curvature const. Also it is an area where results we discuss in Sections 21–23 provide (and will provide) a powerful tool. Sections 21–23 are reviews of results obtained recently by Colding and Cheeger–Colding on the class of manifolds whose Ricci curvature is bounded from below. Here we emphasize the geometric part of the story and omit most of the analytic parts of the proofs, though analytic parts are as important as geometric parts. It is of course impossible to write down all details of the proofs in this article. However, rather than stating as many results as possible without proof, the author tried to survey as many ideas, tools, techniques, methods of proofs, etc. as possible. In that sense, the emphasis of this article is on methods of proofs and not on their outcome. (Of course important applications of various techniques are explained.) Since this is a survey article there are no new results in it.

Metric Riemannian geometry

193

1.1. Notations used in this article Tp M = the tangent space, Expp : Tp M → M, the exponential map, Bp (R, X) = x ∈ X | d(x, p) < R , for a metric space (X, d) and p ∈ X, KM = the sectional curvature of M,

Vol(M) = the volume of M,

RicciM = the Ricci curvature of M,

Diam(M) = the diameter of M,

iM (p) = the injectivity radius of M at p

(Definition 4.1),

xy = a minimal geodesic joining x and y, xyz = the angle between xy and yz at y, Sn (D) = M | RicciM −(n − 1), dim = n, Diam(M) D , Sn (D, v) = M ∈ Sn (D) | Vol(M) v , Sn (D, i > ρ) = M ∈ Sn (D) | ∀p, iM (p) ρ , Mn (D) = M | |KM | 1, dim = n, Diam(M) D , Mn (D, v) = M ∈ Mn (D) | Vol(M) v , Mn (D, v) = M | KM −1, Diam(M) D, Vol(M) v ,

dGH (X1 , X2 ) = the Gromov–Hausdorff distance (Definition 3.2), Sn (κ) = simply connected Riemannian manifold with KM ≡ κ, Ap (a, b; M) = x ∈ M | a d(p, x) b , Sp (a; M) = x ∈ M | d(p, x) = a . limGH i→∞ Xi = X. means limi→∞ dGH (Xi , X) = 0. The symbol = means almost equal. The argument using this symbol is not rigorous. We use it only when we sketch the proof. The symbol τ (1 , . . . , k |a1 , . . . , am ) stand for the positive number depending only on 1 , . . . , k , a1 , . . . , am and satisfying lim

1 ,...,k →0

τ (1 , . . . , k |a1 , . . . , am ) = 0,

for each fixed a1 , . . . , am . In other words f (1 , . . . , k |a1 , . . . , am ) < τ (1 , . . . , k |a1 , . . . , am ) is equivalent to the following statement. For each δ, a1 , . . . , am there exists such that if 1 < , . . . , k < then f (1 , . . . , k |a1 , . . . , am ) < δ.

194

K. Fukaya

2. Sphere theorems There are several pioneering works in metric Riemannian geometry (such as Myers’ theorem (Theorem 5.4), Hadamard–Cartan’s theorem (Theorem 4.6), study of convex surfaces in R3 , etc.). But let me set the beginning of metric Riemannian geometry at the time when the following theorem was proved. From now on, we denote by KM the sectional curvature of a Riemannian manifold M. We assume all Riemannian manifolds are complete unless otherwise stated. T HEOREM 2.1 (Rauch’s sphere theorem [129]). There exists a positive constant n depending only on the dimension n such that, if a simply connected Riemannian manifold M satisfies 1 KM 1 − n , then M is homeomorphic to a sphere. This theorem is one of the first theorems which are called “sphere theorems”. In this section, we mention some of the most important sphere theorems.2 T HEOREM 2.2 (Klingenberg [94], Berger [18]). If a simply connected Riemannian manifold M satisfies 1 KM > 1/4, then it is homeomorphic to a sphere. If M satisfies 1 KM 1/4, then M is either homeomorphic to a sphere or is isometric to a symmetric space of compact type.3 Theorem 2.2 is a generalization of Rauch’s theorem, and is an optimal result among those characterizing spheres under an assumption of the sectional curvature bounded from above or below.4 (We remark that the sectional curvature of a complex, or quaternionic projective space, or Cayley plane is between 1 and 1/4.) T HEOREM 2.3 (Bochner [157]). If the curvature tensor R of a simply connected Riemannian manifold M satisfies C −Rij k ξ ij ξ k C 2 ξ for any antisymmetric 2 tensor ξ (where C is a positive constant), then the homology group over R of M is isomorphic to the homology group of the sphere. The assumption of Theorem 2.3 is on the curvature operator and is more restrictive than the one on sectional curvature. Hence Theorem 2.3 follows from Theorem 2.2. (Theorem 2.3 was proved earlier.) We mention Theorem 2.3 since the idea of its proof is quite different from the proof of Theorem 2.2. We mention them later in Section 19. T HEOREM 2.4. If M is simply connected and if 1 KM 1 − , then M is diffeomorphic to a sphere. The difference between Theorems 2.4 and 2.2 is that the conclusion of Theorem 2.4 is one on the diffeomorphism type and is sharper. The constant 1 − in Theorem 2.4 2 In this article we mention only a part of many sphere theorems. The reader may find more in [139]. 3 More precisely, one of the complex or quaternionic projective space or the Cayley plane. 4 Several results which relax the condition of Theorem 2.2 to 1 K 1/4 − are known. See [3]. M

Metric Riemannian geometry

195

was 1 − n where n is a positive number depending only on dimension n and was not explicit, at the time when it was first proved by Gromoll and Shikata in [65,136]. Later it was improved to a constant 1 − which is independent of the dimension. It was further improved and an explicit bound (1 − = 0.87) was found [143]. The explicit bound is improved several times.5 The possibility that “1 KM > 0.25 and π1 (M) = {1} implies that M is diffeomorphic to S n ” was not yet eliminated. The best constant is not yet found. Remark 2.1. Hitchin [85] proved that there are some exotic spheres which do not admit a metric of positive scalar curvature, by using the KO index theorem of the Dirac operator. Gromoll–Myer [66] (and Grove–Ziller [83]) found examples of exotic spheres which have a metric of nonnegative curvature. So far no example of an exotic sphere which has a metric of (strictly) positive sectional curvature is found. T HEOREM 2.5 (Berger [18], Grove–Shiohama [82]). If KM 1/4 and if the diameter of M is greater than π , then M is homeomorphic to a sphere. Berger proved that M is homotopy equivalent to a sphere under the assumption of Theorem 2.5 and Grove–Shiohama proved that M is homeomorphic to a sphere. By the generalized Poincaré conjecture (proved by Smale and Freedman) the latter follows from the former (in case dimension is not 3). But the proof by Grove–Shiohama (which is different from Berger’s) uses Morse theory of functions which are not differentiable. This technique turns out to be very useful to study Riemannian manifold under lower (but not upper) curvature bounds. (See Section 14.) The next theorem is a final form of a series of results due to Shiohama [137], Otsu– Shiohama–Yamaguchi [111], Perelman [114]. We will discuss it in Section 22. T HEOREM 2.6 (Cheeger–Colding [29]). There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n then M is diffeomorphic to a sphere. A sphere theorem is a characterization of a sphere, which is the most basic example of Riemannian manifold. Let us recall the classification of surfaces (two-manifolds). It was first proved that a “simply connected compact 2-dimensional manifold is a sphere”, then the classification in the general case was performed by simplifying a general surface by, say, surgery. In a similar sense, sphere theorems play an important role in metric Riemannian geometry. Especially the techniques used to prove the sphere theorems we mentioned above play an important role to study more general Riemannian manifolds.

3. Finiteness theorems and Gromov–Hausdorff distance Another type of important result in metric Riemannian geometry are finiteness theorems. First examples of that kind are proved by Cheeger and by Weinstein, which appeared at the beginning of the 1970’s. Cheeger’s finiteness theorem is as follows. 5 The best estimate known at the time of writing this article is about 1 − = 0.68 [86,144].

196

K. Fukaya

T HEOREM 3.1 (Cheeger [25]). For all positive numbers D, v, n, the number of diffeomorphism classes of Riemannian manifolds M with Diam(M) D, Vol(M) v, and |KM | 1 is finite. The method of proof of Theorem 3.1 is closely related to the proofs of Rauch’s sphere theorem and of Theorems 2.2, 2.4. We will explain it later. Theorems 2.4 and 3.1 (and their proof) use an idea that if two Riemannian manifolds are “close” to each other then they are diffeomorphic to each other. One way to formulate precisely what we mean by two Riemannian manifolds to be close, is by using the notion Gromov–Hausdorff distance.6 Let us first review the definition of (usual or classical) Hausdorff distance. Let (X, d) be a metric space and Y1 , Y2 be subspaces. We put for any subspace Y of X, N Y = x ∈ X | d(x, Y ) < , where d(x, Y ) = inf{d(x, y) | y ∈ Y }. D EFINITION 3.1. The Hausdorff distance dX (Y1 , Y2 ) between Y1 and Y2 is the infimum of > 0 such that Y2 ⊂ N Y1 , Y1 ⊂ N Y2 . The Hausdorff distance defines a complete metric on the set of all compact subsets of a fixed complete metric space (X, d). The Gromov–Hausdorff distance is an “absolute analogue” of the Hausdorff distance. Namely it defines a distance between two metric spaces (which we do not assume to be embedded somewhere a priori). D EFINITION 3.2. The Gromov–Hausdorff distance dGH ((X1 , d), (X2 , d)) between two metric spaces (X1 , d) and (X2 , d) is an infimum of the Hausdorff distance dZ (X1 , X2 ), where Z is a metric space such that X1 , X2 are embedded to Z by isometries. Hereafter we write limGH i→∞ Xi = X if limi→∞ dGH (Xi , X) = 0. Gromov–Hausdorff distance defines a complete metric on the set of all isometry classes of compact metric spaces. The following version is sometimes convenient. D EFINITION 3.3 [52]. A map ϕ : X1 → X2 is called an -Hausdorff approximation, if |dX1 (ϕ(x), ϕ(y)) − dX2 (x, y)| for all x, y ∈ X1 and if the -neighborhood of the image ϕ(X1 ) is X2 . If dGH (X1 , X2 ) then there exists a 3-Hausdorff approximation X1 → X2 . If there exists an -Hausdorff approximation X1 → X2 then dGH (X1 , X2 ) 3. There are two types of important results on the Gromov–Hausdorff distance which are applied to finiteness theorems. In this section, we explain results which were developed mainly in the 1980’s. 6 See [69,75,57] for more detailed account on it.

Metric Riemannian geometry

197

We first state Gromov’s precompactness theorem on manifolds with Ricci curvature bound. Let n, D be a positive integer and a positive number. We denote by Sn (D) the set of all isometry classes of Riemannian manifolds M such that Ricci −(n − 1) and diameter D. Here and hereafter the diameter Diam(X) of a metric space (X, d) is the supremum of d(x, y) where x, y ∈ X. T HEOREM 3.2 (Gromov [69]). (Sn (D), dGH ) is relatively compact in the space of all isometry classes of compact metric spaces. The method of the proof of Theorem 3.2 is related to the proofs of Rauch’s sphere theorem and of Theorem 2.2. We will explain it in Section 5. We next mention a rigidity theorem. Gromov’s precompactness theorem assumes bounds from below of the Ricci curvature, which is a rather weak assumption. We need the stronger assumption for the rigidity theorem. We first discuss the case that Gromov studied in [69]. For n, D, v, we denote by Mn (D, v) the set of all isometry classes of n-dimensional Riemannian manifolds M such that |KM | 1, Diam(M) D, and Vol(M) v. T HEOREM 3.3 [69,93]. There exists n (D, v) > 0 such that if M1 , M2 ∈ Mn (D, v) and if dGH (M1 , M2 ) n (D, v), then M1 is diffeomorphic to M2 . Attempts to prove a similar conclusion as Theorem 3.3 under an assumption milder than M1 , M2 ∈ Mn (D, v), played a very important role in the development of metric Riemannian geometry. Perelman proved that M1 is homeomorphic to M2 if dGH (M1 , M2 ) n (D, v) under the assumption KM −1, which replaces |KM | 1 in the definition of Mn (D, v). (Theorem 18.2.) Further study is done when we assume Ricci curvature bounds. (See Theorem 22.3.) Theorem 3.1 follows from Theorems 3.2 and 3.3. (We leave its proof as an exercise to the reader.) Theorem 3.2 asserts relative compactness. Namely it implies that, for any sequence Mi of elements of Sn (D), there exists a converging subsequence. Its limit M∞ may be regarded as a “weak solution” of various problems of metric Riemannian geometry (when we regard it as an analogy of functional analysis). Then it is natural and important to study the “regularity” of M∞ . It is closely related to the proof of Theorem 3.3. The next result is related to the “regularity” question. T HEOREM 3.4 [69,64,121]. Each element of Mn (D, v) is a Riemannian manifold of C 1,α class.7 Here α is any positive number with α < 1 and a Riemannian manifold of C 1,α -class is a manifold with metric tensor g whose first derivative is C α -Hölder continuous. The assumption of Theorem 3.4 is rather strong. There are two kinds of study to relax this condition X ∈ Mn (D, v). 7 The proof of this theorem is completed in [64,121] based on the idea of Gromov [69]. There seems to be

various independent research in Russia. (See, for example, [107,108,17].)

198

K. Fukaya

One is to remove the assumption Vol(M) v. It means that we study the limit of a sequence of Riemannian manifolds which will become degenerate. This is called the study of collapsing Riemannian manifolds. We discuss it in Sections 10–13. (See also [57].) The other direction is to relax the assumption |KM | 1. Theorem 3.1 is generalized as follows towards this direction. For n, D, v, we denote by Mn (D, v) the set of all isometry classes of n-dimensional Riemannian manifolds M such that KM −1, Diam(M) D, Vol(M) v. T HEOREM 3.5 (Grove–Petersen–Wu [78,81]). For each n, D, v, the number of homeomorphism classes of elements of Mn (D, v) is finite.8 We explain the proof of Theorem 3.5 in Section 15. The limit of a sequence of manifolds M satisfying KM −1 is an Alexandrov space. We will discuss it in Sections 17 and 18. Remark 3.1. (1) If Mi is a sequence of Riemannian manifolds such that N = limGH i→∞ Mi and N is a Riemannian manifold. Then KMi κ implies KN κ. Moreover we have dim N dim Mi . (2) On the other hand, in the case when Λ KMi κ, Λ KN is, in general, false for limGH i→∞ Mi = N . A counterexample can be constructed as follows. Let Rotθ be the rotation by angle θ of S 2 = {(x, y, z) ∈ R3 | x 2 + y 2 + z2 = 1} around the z axis. We consider the quotient of S 2 × R by the Z action generated by (p, t) → (Rotα (p), t + ). Let M,α be the quotient space with quotient metric. (M,α is diffeomorphic to S 2 × S 1 .) We have 1 KM,α 0 since M,α is locally isometric to S 2 × R. The limit of M,α as → 0 is S 2 with some Riemannian metric gα . 1 (S 2 , gα ) does not hold unless α = 0.

4. Geodesic coordinate, injectivity radius, comparison theorems and sphere theorem The following theorem in differential topology is used in the proof of Theorem 2.2. T HEOREM 4.1. If a compact n-dimensional manifold M is a union of two open sets both of which are diffeomorphic to Rn , then M is homeomorphic to a sphere. In order to apply Theorem 4.1 to the proofs of sphere theorems, we want to cover M by two coordinate neighborhoods. To estimate the size of the coordinate charts plays an important role in the study of other problems. Let us begin with the following P ROPOSITION 4.2. For each compact Riemannian manifold M, there exists a positive number M with the following properties. If the distance between p, q ∈ M is smaller than M , then there exists a unique geodesic of length < M joining p, q. 8 In case the dimension is 3, [78,81] proved only finiteness of homotopy type. Now, Perelman’s stability theorem

(Theorem 18.2) implies the finiteness of homeomorphism classes in general.

Metric Riemannian geometry

199

The proof of Proposition 4.2 is given in many standard text books of Riemannian geometry. (For example, in [97,33].) The uniqueness of such geodesic is essential for our purpose. Let us explain this point. Let M be a complete Riemannian manifold. For each p ∈ M we define the exponential map, Expp : Tp M → M as follows. Let V ∈ Tp (M). There exists a geodesic : R → M, such that d dt (0) = V . We then put (1) = Expp V . Proposition 4.2 implies that Expp : Tp M → M is a diffeomorphism on the ball of radius M . D EFINITION 4.1. The injectivity radius of a Riemannian manifold M is a function iM : M → R which associates to p ∈ M the positive number: iM (p) = sup | Expp : Tp M → M is injective on V ∈ Tp M | V < . Proposition 4.2 implies iM M for a compact Riemannian manifold M. (It is easy to see that iM is continuous. Hence iM M > 0 follows easily from the implicit function theorem. Proposition 4.2 is a bit more involved.) If R < iM (p), then the restriction of the exponential map Expp : Tp M → M to the metric ball of radius R centered at origin, defines a coordinate of a neighborhood of p. We call it the geodesic coordinate. To prove Theorem 2.2, it is important to estimate the injectivity radius iM from below. The next result9 provides such an estimate. T HEOREM 4.3. Suppose that dim M is even. If KM > 0, then iM π and M is simply connected.10 Suppose dim M is odd. If 1 KM 1/4 and if M is simply connected then, iM π . In particular, if M satisfies the assumption of Theorem 2.2, then we have iM π . (There are several results in the nonsimply connected case. We omit it.) Another result we use is the following P ROPOSITION 4.4 (Berger). Let us assume that KM 1/4 and Diam(M) π . We take p, q ∈ M such that d(p, q) = Diam(M). Then we have Int Bp (π, M) ∪ Int Bq (π, M) = M. (Here Int denotes the interior.) The proof is in Section 14. Using Theorem 4.3 and Proposition 4.4, the proof of Theorem 2.2 goes roughly as follows. By Theorem 4.3, the injectivity radius of M is not smaller than π . Especially the diameter of M is not smaller than π . 9 This theorem is due to [18] in even dimension, and to [95,37] in odd dimension. 10 The second assertion is a classical result due to Synge.

200

K. Fukaya

Let us first assume 1 KM > 1/4. We replace the metric gM of M by (1 + δ)gM , where δ is a positive number sufficiently close to 0. The assumption 1 KM > 1/4 is still satisfied. Hence M satisfies the assumption of Proposition 4.4. Hence Int Bp (π, M) ∪ Int Bq (π, M) = M. Moreover Int Bp (π, M) and Int Bq (π, M) are diffeomorphic to the ball by Theorem 4.3. Therefore, by Theorem 4.1, M is homeomorphic to a sphere. We next consider the case when 1 KM 1/4. If the diameter of M is strictly greater than π , then again Proposition 4.4 and Theorems 4.1 and 4.3 imply that M is homeomorphic to a sphere. Finally we consider the case when the diameter of M is π . In this case, we consider the restriction of the exponential map Expp : Tp M → M to the ball D n (π) of radius π . Then it is a diffeomorphism at the interior. So M is obtained from D n (π) by identifying boundary points only. We examine this situation carefully and conclude that M is a symmetric space of compact type. We omit the details. (See, for example, [33, Chapter 7].) We explain the outline of the proof of Theorem 4.3 later in this section. We first explain some basic facts. Let us begin with the following theorem. Let κ be a constant. We put sin t √κ √κ , sκ (t) = t, √ sinh√t −κ , −κ

κ > 0, κ = 0, κ < 0.

(4.1)

T HEOREM 4.5 (Rauch). If KM κ, then the derivative dx Expp of the exponential map Expp satisfies dx Exp (V ) V sκ (r). p √ Here x ∈ Tp (M), x = r, V ∈ Tx Tp (M) ∼ = Tp (M) and we assume r π/ κ in case κ > 0. Let KM κ. In case κ > 0, we assume dtx Expp is invertible for t ∈ [0, 1]. Then we have dx Exp (V ) V sκ (r). p Theorem 4.5 implies that if KM 1, then the restriction of Expp : Tp M → M to the ball of radius π is an immersion. (Namely its Jacobi matrix is invertible.) We remark that the equality in Theorem 4.5 holds in the case when M is of constant curvature κ. Theorem 4.5 is used by Rauch to prove his sphere theorem. We use the Jacobi field in the proof of Theorem 4.5 as follows. Let x, V be as in Theorem 4.5, and define a geodesic s by s (t) = Expp t (x + sV ) .

Metric Riemannian geometry

201

For each s, s is a geodesic. Its derivative

∂s (t)

J (t) = ∈ T0 (t) M ∂s s=0 with respect to s, by definition, is a Jacobi field. Note that dx Expp (V ) = J (1). Therefore, to prove Theorem 4.5, it suffices to estimate the Jacobi field. We use the following equation (which the Jacobi field satisfies) for this purpose. d0 d0 D2 (t), J (t) (t) = 0. J (t) + R dt dt dt 2

(4.2)

D 0 Here dt is a covariant derivative with respect to the tangent vector d dt (t) and R is a curvature tensor. If e1 , e2 is an orthonormal frame of a plane π in the tangent space, then g(R(e1 , e2 )e2 , e1 ) is the sectional curvature of the plane π . (Here g is the metric tensor.) Therefore, the second term of Eq. (4.2) can be written in terms of the sectional curvature. Using it we can compare Eq. (4.2) to the one in case our manifold is of constant curvature. Namely if KM ≡ κ then (4.2) will be

D2 J (t) + κJ (t) = 0. dt 2

(4.3)

Its solution is J (t) = sκ (t)V (t) where ∇(t) ˙ V = 0. Namely J (t) = sκ (t) if KM ≡ κ. This implies Theorem 4.5. Theorem 4.5 implies the following T HEOREM 4.6 (Hadamard–Cartan). If a complete Riemannian manifold M satisfies KM 0, then Expp : Tp M → M is a covering map. In particular the universal covering space of M is diffeomorphic to Rn . In fact, Theorem 4.5 implies that the Jacobi matrix of Expp : Tp M → M is of maximal rank everywhere. To prove that it is a covering map we need a bit more. We use completeness of metric for this last step. We omit it. By integrating the conclusion of Theorem 4.5, we can compare the distance between two points Expp (x), Expp (y) (which are close to p) to the corresponding distance in the space with constant curvature. Actually we can do it more globally and obtain the Toponogov comparison theorem. To state it we need some notation. Let Sn (κ) be the complete simply connected Riemannian manifold with constant curvature κ. Let x , y , z ∈ Sn (κ). We denote by x y , etc. the minimal geodesic joining x and y , etc. Let θ = y x z be the angle between x y and

202

K. Fukaya

Fig. 4.1.

x z at x . We put a = d(x , y ), b = d(x , z ). It is easy to see that d(y , z ) depends only on a, b, θ, κ. We define s(a, b, θ, κ) = d(y , z ).

(4.4)

√ We remark that in case κ > 0, the number s(a, b, θ, κ) is defined only for a, b < π/ κ. Let M be a Riemannian manifold and x, y, z ∈ M. We denote by xy a minimal geodesic joining x and y. (In case there are several minimal geodesics we take any one of them.) Let yxz be the angle between xy and xz at x (see Figure 4.1). T HEOREM 4.7 (Alexandrov–Toponogov). If KM κ then we have d(y, z) s d(x, y), d(x, z), yxz, κ . If KM κ and if d(x, y), d(x, z) iM (x) then d(y, z) s d(x, y), d(x, z), yxz, κ . We remark that in the first inequality we do not need to assume that the triangle x, y, z is small. Actually we only need to assume one of the geodesics √ joining x to y and to z are minimal and the other may be any geodesic of length π/ κ. Theorem 4.7 is proved in many text books (see, for example, [33]). As we already mentioned, Theorem 4.5 implies that, if KM 1, then the exponential map is an immersion on the metric ball of radius π . Especially it is locally an injection there. To prove Theorem 4.1 we need global injectivity. We here introduce several terminology. D EFINITION 4.2. q ∈ M is said to be a conjugate point of p ∈ M if there exists x such that q = Expp (x) and that dx Expp is not of maximal rank. q is said to be a cut point of p ∈ M if there exists x = y ∈ Tp (M) such that Expp x = Expp y = q. E XAMPLE 4.1. We consider the sphere S 2 of constant curvature 1. Every geodesic which starts at the north pole np meets again at the south pole sp. Hence the south pole is a conjugate point of the north pole.

Metric Riemannian geometry

203

Fig. 4.2.

We next divide S 2 by the involution and obtain the real projective space RP 2 . Then np and sp determine the same point x = [np] = [sp] ∈ RP 2 . If c ∈ S 2 is on the equator then there are minimal geodesics 1 , 2 joining c to np, sp, respectively. 1 , 2 induce two minimal geodesics ¯1 , ¯2 in RP 2 joining x to y = [c]. Thus y is a cut point of x. Note that iM (p) > r holds if there exists neither a cut point nor a conjugate point q of p such that d(p, q) r. We can use Theorem 4.5 to estimate the distance to the conjugate point. However the problem to estimate the distance to the cut point is a more global one. We remark the following fact. L EMMA 4.8. If : [a, b] → M is the minimal geodesic, then for t ∈ (a, b), q = (t) is neither a cut point nor a conjugate point of p = (a). In fact if q is a cut point then there is a geodesic joining p to q with | | = |[a,t] | (see Figure 4.2). Then the union ∪ |[t,b] of two geodesics is not smooth and has the same length as the minimal geodesic . This is a contradiction. If q is a conjugate point then by the Morse index theorem (see [103,97,33]), [a,t+] is not minimal. This contradicts the assumption. Here we state the following basic result about cut points. (See, for example, [33, p. 96] for its proof.) T HEOREM 4.9 (Klingenberg). Let M be a Riemannian manifold. We assume that q is not a conjugate point of p, for each p, q ∈ M with d(p, q) < r. If there exists p ∈ M with iM (p) < r then there exists a closed geodesic of length < 2r in M. In view of Theorems 4.5 and 4.9, to prove Theorem 4.3, it suffices to show that the length of a nontrivial closed geodesic of M is greater than 2π . We explain the brief outline of its proof below. (See [33, p. 100] for its details.) We first consider the case dim M is even. Let M be a simply connected Riemannian manifold with 1 KM > 0. Let : S 1 → M be a nontrivial geodesic of minimal length. We regard S 1 ∼ = R/Z. Put p = (0). By the parallel transport along we have a holonomy homomorphism hol : Tp M → Tp M. The tangent vector d dt (0) is an invariant of the holonomy. Since hol is an orthogonal transformation, and dim M is even, it follows that there ex-

204

K. Fukaya

ists a nonzero vector V ∈ Tp M orthogonal to d dt (0) such that hol (V ) = V . The parallel transport of V defines a vector field V (t) ∈ T(t) M, which is a parallel vector field. We put s (t) = Exp(t) sV (t) . Using ∇V (t) = 0 and first variation formula (see, for example, [33, Section 1], [97, Vol. II, s Theorem 5.1], [103, Theorem 12.2]), we find that d ds (0) = 0. Using moreover the second variation formula (see, for example, [33, Section 6], [97, Vol. II, Theorem 5.4], [103, The2 orem 13.1]) and the positivity of curvature, we find dds2s (0) < 0, which contradicts to the minimality of the length of . The proof of the odd-dimensional case is more involved. We remark that the quotient of S 3 by a cyclic group Z/pZ has constant positive curvature one (and is not simply connected). Its injectivity radius converges to 0 as p → ∞. This shows that, to prove Theorem 4.3 in odd-dimensional case, we need to use the assumption that M is simply connected. The proof of the odd-dimensional case is roughly as follows. We assume that there exists a closed geodesic of length < 2π . Since M is simply connected, is null homotopic. Let s be a homotopy such that 0 = , 1 = const. We may assume that the length of is minimal among all nontrivial closed geodesics. By using the assumption that KM > 1/4 we can prove that the length of s is always smaller than 2π . (This is the essential point of the proof. To prove this we use the fact that the Morse index (with respect to the length) of the closed geodesic of length > 2π is not smaller than 2.11 ) Now we consider the exponential map Expp at the tangent space of p = (0). Expp is a submersion on the ball of radius π . Hence it has a similar property to the covering map up to radius π . Especially it has the homotopy lifting property there. Since the length of s is not greater than 2π , its image is of distance π from p. Therefore we can lift s to Tp M. (Note we can lift 1 since it is a constant map.) Hence we obtain a lift ˜0 : S 1 → Tp M. But this is a contradiction since 0 = is a geodesic.12

5. Packing and precompactness theorem A similar argument as in the last section is used in the proof of the finiteness theorem (Theorem 3.1) and of Theorem 3.2. We explain this point here. We first discuss Theorem 3.2. The basic fact we use for its proof is the following 11 Let us consider the round sphere of radius 2 (that is the round sphere of curvature 1/4). The geodesic segment of length 2π , that is the geodesic segment joining north pole with south pole, has Morse index n − 1. (Here we consider the set of all arcs joining north pole with south pole and consider the length as a Morse function on it. n − 1 is the Morse index with respect to this Morse function.) We compare our closed geodesic with this geodesic segment to obtain the conclusion about the Morse index. 12 This argument is not enough to handle the case 1 K 1/4 of Theorem 2.2 (since then we can only show M that π is a submersion at the interior of the ball of radius π ). In that case we need an additional argument. We omit it.

Metric Riemannian geometry

205

P ROPOSITION 5.1. Let D > 0 and N : (0, 1) → N. We denote by Met(D, N ), the set of all isometry classes of complete metric spaces satisfying (1), (2) below. Then Met(D, N ) is compact with respect to the Gromov–Hausdorff distance. (1) The diameter of M D. (2) For each ∈ (0, 1) there exists a finite subset Z of M with the following properties: (2.a) !Z N (). (2.b) For each x ∈ M, there exists x0 ∈ Z satisfying d(x, x0 ) < . The proof of Proposition 5.1 is given, for example, in [57, Section 2]. Here we introduce a notation. D EFINITION 5.1. We call the subset Z an -net if it satisfies (2.b). To deduce Theorem 3.2 from Proposition 5.1, we use the following Theorem 5.2. Let Sn (κ) be the complete simply connected Riemannian manifold with constant curvature κ. Let Bp (R, M) be the metric ball in M of radius R centered at p. T HEOREM 5.2 (Bishop–Gromov). If Ricci (n − 1)κ then the volume Vol(Bp (R, M)) of the metric ball satisfies the following inequality for r < R: Vol(Bp (R, M)) Vol(Bp0 (R, Sn (κ))) . Vol(Bp (r, M)) Vol(Bp0 (r, Sn (κ)))

(5.1)

(5.1) is called the Bishop–Gromov inequality. It plays a key role to study the class of Riemannian manifolds with Ricci curvature bounded from below. The equality holds if M is of constant curvature κ. Let us sketch a proof of Theorem 5.2. We put A(t) =

Vol(Bp (t, M)) . Vol(Bp0 (t, Sn (κ)))

(5.2)

It suffices to show that A is nonincreasing. (In case κ > 0, Theorem 5.4 implies that we need to consider t π only.) Let : [0, a) → M be a minimal geodesic with (0) = p parameterized by arc length. Let v = (d/dt)(0). We take a vector v∗ ∈ Tp0 Sn (κ) with unit length. We put a(v, t) =

det dtv Expp det dtv∗ Expp∗

.

(5.3)

Here det dtv Expp is the determinant of the derivative of the exponential map. We first prove that a(v, t) is a nonincreasing function of t for each fixed v. We can prove it in a way similar as the proof of Theorem 4.5. One difference however is that our assumption in Theorem 5.2 is only on the Ricci curvature while in Theorem 4.5 the assumption is on the sectional curvature. However since we only need to estimate

206

K. Fukaya

the determinant of the Jacobi matrix of the exponential map, the assumption on the Ricci curvature, which is the trace of the curvature tensor, is enough. This is half of the idea of the proof of Proposition 5.2. Let us fix p and move q ∈ M, and consider the set

dp,q (5.4) V= (0) ∈ Tp M

q ∈ M , dt where p,q is the minimal geodesics joining p and q. (If there are several we take all of d them.) (We take a parametrization of p,q so that the length of dtp,q (0) is d(p, q).) We have Vol Bp (R, M) = det dx Expp dx. (5.5) V ∩B0 (R,Rp M)

(Here det dx Exp is the determinant of Jacobi matrix.) (5.5) and the fact that a(v, t) is a nonincreasing function of t implies (5.1) for R, r smaller than the injectivity radius. To prove Theorem 5.2 beyond injectivity radius, we proceed as follows. We remark that V is star shaped (that is if x ∈ V t ∈ [0, 1], then tx ∈ V ). We then modify a to a so that a (t, v) = a(t, v) if tv ∈ V and a (t, v) = 0 if tv ∈ / V . Then a is a nonincreasing function of t. Theorem 5.2 follows. C OROLLARY 5.3. If RicciM κ and p ∈ M then Vol Bp (R, M) Vol Bp0 R, Sn (κ) . This corollary follows from the fact that the function A in (5.2) is nonincreasing and limt→0 A(t) = 1. Theorem 5.2 and Proposition 5.1 imply Theorem 3.2 as follows. Let us assume that M satisfies the assumption of Theorem 3.2. It suffices to show that M satisfies the assumption of Proposition 5.1. Let > 0. We take Z ⊂ M which is maximal (with respect to inclusion) among the subsets of M satisfying “z1 , z2 ∈ Z, z1 = z2 , implies d(z1 , z2 ) > ”. The maximality implies (2.6). On the other hand, since Bz (/2, M), z ∈ Z, are disjoint to each other, it follows that Vol Bz (/2, M) < Vol M. z∈Z

Since Bz (D, M) = M, Proposition 5.1 implies !Z

Vol(Bp0 (D, Sn (κ))) Vol(M) . sup Vol(Bp (/2, M)) Vol(Bp0 (/2, Sn (κ)))

If we let N () be the right-hand side, then the assumption of Proposition 5.1 is satisfied. Theorem 3.2 follows. We remark that the following classical result is actually proved during the proof of Theorem 5.2.

Metric Riemannian geometry

207

T HEOREM 5.4 (Myers). If M is an n-dimensional complete Riemannian manifold √ with Ricci (n − 1)κ > 0, then M is compact and its diameter is not greater than π/ κ. In fact during the proof of Theorem 5.2 we proved the following under the assumption p ∈ M, RicciM κ. “If t → Expp (tv) is a minimal geodesic for t ∈ [0, 1], then det dv Expp is not greater than det dv0 Expp0 , where p0 ∈ Sn (κ), v0 ∈ Tp0 Sn (κ) and |v0 | = |v|.” √ We remark that det dv0 Expp0 = 0 if v0 = π/ κ. Therefore there exists no minimal √ geodesic of length > π/ κ if RicciM κ. Theorem 5.4 follows immediately. In the above argument, Bz (, M), z ∈ Z, covers M. Namely we estimate the number of metric balls (geodesic coordinate) to show Theorem 3.2. If is smaller than the injectivity radius of M, then Bz (, M) is diffeomorphic to D n . The proof of Theorem 3.2 is related to the proof of sphere theorems in this way. Theorem 4.1 deals with the case when two balls cover M and conclude that M is a sphere. If we can replace Theorem 4.1 by a statement such as “if M is covered by the balls whose number is estimated by C, then the number of diffeomorphism classes of such M is estimated by C” then finiteness theorems would follow. Unfortunately the statement in the parenthesis above does not hold. So we need to include information how the balls are glued. Theorem 3.1 can be proved in that way. (See Sections 6–8.) Here we prove a weaker version (Weinstein [150]). P ROPOSITION 5.5. For each D, the number of homotopy equivalence classes of n-dimensional Riemannian manifolds satisfying (1), (2) below is finite. (1) M ∈ Mn (D), (2) The injectivity radius of M is greater than . To prove Proposition 5.5 we use the set Z above. We then obtain an open covering Bz (, M), . z ∈ Z, of M. It is a simple covering. Namely for each z1 , . . . , zk ∈ Z the intersection ki=1 Bzi (, M) is either empty or contractible. It implies that the simplicial complex K(Z) defined below is homotopy equivalent to M. (1) A vertex of K(Z) corresponds to an element of Z. (2) . z0 , . . . , zk ∈ Z is the set of vertices of a k simplex of K(Z) if and only if k i=0 Bzi (, M) = ∅. Since the order of Z is estimated by a number depending only on D and , it follows that there exists only finitely many possibilities for the homotopy type of K(Z). Proposition 5.5 follows. In Theorem 3.1, there is no assumption on the injectivity radius but only a bound of volume from below is assumed. An assumption on the volume is more natural and geometric than one on the injectivity radius. However, in case the absolute value of the sectional curvature is bounded, these two assumptions are equivalent. P ROPOSITION 5.6 (Cheeger [25]). There exists a positive number c(n, D, v) depending only on n, D, v such that if M ∈ Mn (D, v), then iM c(n, D, v).

208

K. Fukaya

The proof of Proposition 5.6 is closely related to the proof of Theorem 3.5. We will explain it in Section 15.

6. Construction of homeomorphism by isotopy theory In Section 5, we discussed an estimate of the number of open sets which cover M and which are diffeomorphic to D n , and we showed how it is used to estimate the number of homotopy types (Proposition 5.5). However as we mentioned there, we need more arguments to estimate the number of diffeomorphism classes (or homeomorphism classes). We will explain some of them in the four sections beginning from this section. We again begin with a sphere theorem, the differentiable sphere theorem (Theorem 2.4) this time. Let M satisfy the assumptions of Theorem 2.4. Namely we assume that M is simply connected and 1 KM 1 − . Then by Proposition 4.4 and Theorem 4.3, M is a union of two balls V1 , V2 such that Vi ∼ = D n . We may assume ∂Vi ∼ = S n−1 . Moreover we may assume V1 ∩ V2 = ∂V1 = ∂V2 . So we obtain a diffeomorphism I : S n−1 ∼ = ∂V1 → ∂V2 ∼ = S n−1 .

(6.1)

It is easy to see that if I is diffeotopic to the identity map (namely if there exists a smooth family It of diffeomorphisms such that I0 = I , I1 = id), then M = V1 ∪ V2 is diffeomorphic to S n . Now we use the following P ROPOSITION 6.1. For each compact Riemannian manifold N there exists N > 0 such that if the C 1 -distance between F : N → N and the identity is smaller than N , then F is diffeotopic to the identity. Here we recall D EFINITION 6.1. Two diffeomorphisms F1 , F2 : N → N are said to be diffeotopic to each other if there exists a smooth map F : [1, 2] × N → N such that F (1, x) = F1 (x), F (2, x) = F2 (x) and that x → F (t, x) is a diffeomorphism for each t. The proof is elementary. To apply Proposition 6.1 to the proof of Theorem 2.5, we use the following lemma. L EMMA 6.2. For each > 0 there exists δn () > 0 with the following properties. Let M be an n-dimensional simply connected Riemannian manifold with 1 > K > 1 − δn (). Then we may choose the gluing map (6.1) so that its C 1 -distance from the identity is smaller than . We omit the proof. See, for example, [33, Chapter 7].

Metric Riemannian geometry

209

Fig. 6.1.

We are going to explain how we use the idea above for the proof of Theorems 3.3 and 3.1. Cheeger’s original proof of Theorem 3.1 [25] is similar to the idea explained in this section. Let M, N be Riemannian manifolds. We assume $ by the same num$ that they are covered ber of metric balls. Namely we assume M = ki=1 Bpi (, M), N = ki=1 Bqi (, N ). We assume also that 10 is smaller than the injectivity radius of M and of N . (We put 10 for a technical reason.) We assume also that intersection pattern of the balls are the same. Namely, for each i, j , Bpi (, M)∩Bpj (, M) = ∅ if and only if Bqi (, N )∩Bqj (, N ) = ∅. We want to find a sufficient condition for M to be diffeomorphic $ to N . For this purpose $ we compare the chart ki=1 Bpi (, M) of M, with the chart N = ki=1 Bqi (, N ) of N . To compare, we want to take the same domain for coordinate transformations. For this purpose we proceed as follows. Let Bpi (, M) ∩ Bpj (, M) = ∅ then Bpi (, M) ⊂ Bpj (10, M) (see Figure 6.1). For each pi , qj , we fix a linear isometry Tpi M ∼ = Rn , Tqj N ∼ = Rn and use n it to identify tangent spaces with R . (There are various choices of identification. We take one and fix it.) We consider the composition n n ϕjMi = Exp−1 pj ◦ Exppi : B () → B (10).

Here B n () is a metric ball of radius in Rn centered at origin, and Exp−1 pj is an inverse of the exponential map Exppj : B n (10) → N . We define ϕjNi in a similar way. In the next proposition we assume that the C 2 norm (or C 1,α norm) of ϕjMi , ϕjNi is smaller than a constant C. P ROPOSITION 6.3. There exists n,k (C) > 0 such that if the C 1 distance between ϕjMi and ϕjNi is smaller than n,k (C), then M is diffeomorphic to N .

210

K. Fukaya

Cheeger proved Proposition 6.3 in the following way. First we use Proposition 6.1 to prove that the coordinate transformation ϕjMi is diffeotopic to ϕjNi . We then use it to $ construct a diffeomorphism K i=1 Ui → N (to its image) by induction in K. For details see [25]. We prove Proposition 6.3 in a slightly different way in Section 7. Proposition 6.3 is used to prove Theorem 3.1. For this purpose, we first observe that there is a constant C such that a Riemannian manifold satisfying the assumption of Theorem 3.1 is covered by metric balls whose number is not greater than C. Since the number of metric balls is bounded, the number of possible intersection patterns among them is also bounded. Let us fix intersection pattern of the metric balls we use. We use Proposition 6.3 and find that, if the coordinate transformations ϕjMi are C 1 close to ϕjNi , then M is diffeomorphic to N . If the coordinate transformations ϕjMi are uniformly bounded in the C 2 norm then Ascoli–Alzera’s theorem implies that they are precompact in C 1 -topology. Theorem 3.1 will follow. We need however to estimate second derivative of the coordinate transformation uniformly. Our assumption in Theorem 3.1 is on curvature, which is a second derivative of the metric tensor. So one may imagine that it implies the estimate of the second derivative of coordinate transformation. However when we use geodesic coordinates, the assumption of (sectional) curvature is not enough to do so. (Cheeger [25] proved it under the additional assumption that a covariant derivative of the curvature tensor is also bounded.) To go around this trouble, Cheeger in [25] proceeds as follows. Instead of using a statement such as “two diffeomorphism is diffeotopic to each other if they are C 1 -close to each other” we can use a statement such as “two homeomorphism are isotopic to each other if they are 13 C 0 -close to each other” $k [48]. And we can use the isotopy extension theorem to construct an homomorphism i=1 Ui → N by induction on k. This argument implies finiteness of homeomorphism classes and is not enough to prove Theorem 3.1 in four dimensions.14 (In higher dimensions, one can use surgery etc. to deduce finiteness of diffeomorphism classes from finiteness of homeomorphism classes by purely topological argument.) We can use harmonic coordinates (which we discuss in the next section) to find a coordinate chart such that the C 2,α -norm of its coordinate transformation can be estimated uniformly.

7. Harmonic coordinate and its application As we mentioned in the last section, in order to obtain an estimate of the Hölder norm of the coordinate transformation, taking geodesic coordinates does not give an optimal result. Harmonic coordinates are the best choice for this purpose.15 There are various other 13 Which is much less elementary than Proposition 6.1 and is based on highly nontrivial results such as Kirby– Siebenmann’s result on the Hauptvermutung. See [48]. 14 [120] added some technical argument and proved Theorem 3.1 in four dimension as well. 15 In mathematical study of gauge theory, we need to take representative of gauge equivalence class in order to kill freedom of gauge transformation. This is an important point to study moduli space of connections. Here we are studying “gravity” and coordinate transformation plays a role of gauge transformation. The process to find a good coordinate is called gauge fixing in Physics. Harmonic coordinates are used in Riemannian geometry around

Metric Riemannian geometry

211

applications of harmonic coordinates.16 It also plays an important role to prove that the limit metric in Theorem 3.4 is of C 1,α -class. Let M be a Riemannian manifold. We assume that the injectivity radius of M is much greater than r. Let p ∈ M and ei (p), i = 1, . . . , n be an orthonormal frame of Tp M. We put vi (p) = Expp (rei (p)), wi (p) = Expp (−rei (p)) and define hp,i (x) =

d(x, wi (p))2 − d(x, vi (p))2 . 4r 2

(7.1)

We call hp,i an almost linear function. (We remark that hp,i is a linear function if M = Rn .) hp = (hp,1 , . . . , hp,n ) defines a coordinate system in a neighborhood of p. However since hp is in principle a distance function, this coordinate does not provide optimal results for the estimate of the Hölder norm of coordinate transformation. We will replace it by a harmonic function. We consider a boundary value problem of the Laplace equation ϕ = 0 as follows. Let us take δ such that r - δ - iM (p), and consider ϕp,i : Bp (δ, M) → R with the following properties: (1) ϕp,i = 0. (2) If q ∈ Sp (δ, M), then ϕp,i (q) = hp,i (q). D EFINITION 7.1. We call ϕp = (ϕp,1 , . . . , ϕp,n ) a harmonic coordinate. p

p

Using the fact that ϕi is C 1 -close to hi we can prove that ϕp defines a coordinate in a neighborhood of p. Now we can prove an estimate of C 2,α norm of the coordinate transformation of the harmonic coordinate as follows. We put D n () = {x ∈ Rn | x < }. We take with 10 < r. Let p, q ∈ M with d(p, q) < . We consider the inverse ϕp−1 of ϕp . Then the image of ϕp−1 : D n () → M is contained in the domain of ϕq : Bq (r, M) → Rn . Therefore we can define M ϕq,p = ϕq ◦ ϕp−1 : D n () → Rn .

(7.2)

T HEOREM 7.1. There exists a positive constant C(r, , α, n) depending only on r, , α and M is not greater than C(r, , α, n). the dimension n, such that the C 2,α -norm of ϕq,p 1,α Also the C -norm of the metric tensor in harmonic coordinates is estimated by C(r, , α, n). The proof is based on a priori estimate of harmonic functions. See [87,88,64], where the second half is proved. The first half follows easily from the second half. Theorem 7.1 is generalized to Theorem 20.7. the same time when Uhlenbeck etc. used Coulomb gauge in the study of moduli space of connections. The proof of Theorem 3.4 we present in this section is very similar to the proof by Uhlenbeck etc. of the compactification of the moduli space of self dual connections on 4 manifolds. 16 We can use it to study Gromov–Hausdorff convergence under weaker assumption also. See Section 20.

212

K. Fukaya

Let us prove Theorem 3.4 as a typical application of Theorem 7.1.17 Let us take a sequence Mk of elements of Mn (D, v). We denote its limit in Gromov–Hausdorff distance by X. By Theorem 4.3, the injectivity radius of Mk is greater than r, a number independent of k. We take such that 10 < r. In the same way as Section 2, we can take a finite subset {pi,k | i = 1, . . . , Ik } ∈ Mk with the following properties: (1) $ Ik is smaller than a number independent of k. (D n ()) = Mk . (2) i ϕp−1 i,k By (1) we may assume that Ik is independent of k by taking a subsequence if neces(D n ()) has only sary. Set I = Ik . Then the intersection pattern of the coordinates ϕp−1 i,k a finite number of possibilities. Hence by taking a subsequence we may assume that the intersection pattern is independent of k. Namely we may assume that for each i, j I , n n D () ∩ ϕp−1 D () ϕp−1 i,k j,k

(7.3)

is empty or not does not depend on k. M Now for any i, j such that (7.3) is not empty, we consider ϕpj,kk ,pi,k defined by (7.2). We fix α < 1, and apply Theorem 7.1 to α with 1 > α > α. We then find that the C 2,α -norm of ϕpMj,kk ,pi,k is estimated by a number independent of k. Hence we may take a subsequence and assume that ϕpMj,kk ,pi,k converges in C 2,α -topology. Let us denotes its limit by ϕpj,∞ ,pi,∞ : D n () → Rn . We use them as a coordinate transformation to obtain a smooth manifold M∞ of C 2,α class. Moreover by the uniform C 1,α -boundedness of metric tensor, we find a Riemannian metric g∞ on M∞ of C 1,α -class which is a limit of metrics on Mk . We can prove that Mk converges to (M∞ , g∞ ) in Gromov–Hausdorff distance. Hence (M∞ , g∞ ) is isometric to X. Theorem 3.4 follows. We next prove Theorem 3.3. We assume that the theorem is false. Then there exist M1,k , M2,k ∈ Mn (D, v) such that dH (M1,k , M2,k ) < 1/k but M1,k is not diffeomorphic to M2,k . We use Theorem 3.3 to show that, after taking a subsequence, M1,k , M2,k converges to X1 , X2 , respectively. By Theorem 3.4, X1 , X2 are Riemannian manifolds of C 1,α -class. By using the center of mass technique we will explain in the next section, we can prove that M1,k is diffeomorphic to X1 and M2,k is diffeomorphic to X2 for large k. On the other hand, since the Gromov–Hausdorff distance between X1 and X2 is zero, it follows that X1 is isometric to X2 . Hence X1 is diffeomorphic to X2 . This is a contradiction. 8. Center of mass technique In Section 6 we explained how the isotopy extension theorem can be used to construct a homeomorphism. In fact the isotopy extension theorem is very difficult to prove. We can use a method called the center of mass technique which simplifies those points. The center 17 The author follows the argument of [90] here.

Metric Riemannian geometry

213

of mass technique can be applied to various other problems, for example, to group actions. In this section we explain it. Let us start the explanation of the center of mass technique by beginning a proof of (a modified version of) Proposition 6.3. In Proposition 6.3, the assumption is about the exponential map Expp or coordinate transformation of geodesic coordinates. We actually use the case of harmonic coordinates. So we consider $ the following situation. $ (a) M = i ϕpi (D n ()), N = i ψqi (D n ()) are open coverings. (b) The intersection pattern of coordinate neighborhoods coincide to each other. Namely ϕpi (D n ()) ∩ ϕpj (D n ()) = ∅ if and only if ψqi (D n ()) ∩ ψqj (D n ()) = ∅. (c) If ϕpi (D n ()) ∩ ϕpj (D n ()) = ∅, then ϕpi (D n ()) ⊆ ϕpj (D n (r)). (d) The C 2,α -norm of the coordinate transformation ◦ ϕpi : D n () → Rn Φij = ϕp−1 i is bounded uniformly above by C. The same holds for ◦ ψqi : D n () → Rn . Ψij = ψq−1 i (e) Φij is close to Ψij in C 1 -norm. Our purpose is to construct a diffeomorphism F : M → N under these assumptions. For each x ∈ ϕpi (D n ()), we put Fi (x) = ψqi ◦ ϕp−1 (x) ∈ N. i

(8.1)

This corresponds to what we defined F on each coordinate chart ϕpi (D n ()). The main point is whether we can glue them to obtain F globally. Namely in case x ∈ ϕpi (D n ()) ∩ ϕpj (D n ()) we need to know whether ψqi ◦ ϕp−1 (x) = ψqj ◦ ϕp−1 (x) i j ?

(8.2)

or not. It is easy to see that (8.2) does not hold. What follows from our assumption (assumption of Proposition 6.3 or the assumption (e) above) is (x), ψqj ◦ ϕp−1 (x) < (8.3) d ψqi ◦ ϕp−1 i j (where is a sufficiently small positive number). (More precisely, (8.3) is in C 0 -norm, but assumption (e) is in C 1 -norm.) The basic idea of the center of mass technique is to take the average of Fi (x) over i with x ∈ ϕpi (D n ()). Before we continue the proof of Proposition 6.3, we explain the center of mass technique in general here. Let m a Borel probability measure on M (namely a measure on M with m(M) = 1). Let us denote the support of m by Supp(m). We define a function dm on M by dm (x) = d(x, p) dm(p). (8.4)

214

K. Fukaya

P ROPOSITION 8.1. We assume √ the injectivity radius of M is larger than 10. We also assume KM κ and 20 < π/ κ.18 If the diameter of Supp(m) is smaller than , then on B3 Supp(m), M = x ∈ M | d x, Supp(m) < 3 , the function dm is convex. Here a function on a Riemannian manifold said to be convex if its restriction to each geodesic is convex. We can √ prove Proposition 8.1 by using the convexity of the distance function dp on Bp (π/ κ, M).19 Now we assume that the diameter of Supp(m) is smaller than . Then outside B3 (Supp(m), M) the value of the function dm is greater than 3, and on Supp(m) the value of the function dm is smaller than . Therefore Supp(m) attains its minimum on the interior of B3 (Supp(m), M). Since dm is convex there, the minimum is attained at unique point. D EFINITION 8.1. The center of mass is the point where dm attains its minimum. We write center of mass by CM(m). We remark that if M = Rn , then CM(m) = x dm(x). Rn

We go back to the $ proof of Proposition 6.3. We take a partition of unity χi associated to the covering M = i Bpi (, M). We define a measure F(x) on N by F(x) =

χi (x)δFi (x) .

i

Here δFi (x) is the delta measure supported at Fi (x) and the summation is taken over all i with x ∈ Bpi (, M). By (8.3) we have Diam(Supp(F(x))) < . Let F (x) be the center of mass of F(x). Namely, χi (x)δFi (x) . F (x) = CM F(x) = CM

(8.5)

i

It is easy to see that F (x) is a continuous function of x. Actually it is smooth. (We can prove it by using implicit function theorem.) We can prove that it is a diffeomorphism by using the following lemma. 18 In case κ 0 the second condition is void. 19 This fact is a consequence of Toponogov’s comparison Theorem 4.7.

Metric Riemannian geometry

215

L EMMA 8.2. If Fi , i = 1, 2, . . . , are C 1 -close to each other, then F , determined by (8.5), is C 1 -close to Fi . The proof is elementary. Then, to prove Proposition 6.3, we only need to show that F is injective. Suppose F (x) = F (y), x = y. By using the fact that the Jacobi matrix of F is invertible, we can show that x cannot be close to y. On the other hand, since F is close of Fi and since Fi is injective, we can prove that x cannot be far from y. This is a contradiction. This is an outline of the proof of Proposition 6.3. There are various other applications of the center of mass technique. Let us mention another application of it, that is an application to group actions. Let M be a Riemannian manifold on which G acts. For simplicity we assume G is a finite group. We assume G has two different actions on M and write them as ψ1 : G → Diff(M) and ψ2 : G → Diff(M). We assume that there exists C such that for each element g ∈ G, the C 2 norm of ψ1 (g), ψ2 (g) are smaller than C. P ROPOSITION 8.3 (Grove–Karcher). There exists a constant depending only on C, the dimension n, the injectivity radius of M, and the maximum of the absolute value of the sectional curvature of M, with the following property. If d(ψ1 (g)(x), ψ2 (g)(x)) < for each g ∈ G, x ∈ M, then there exists a diffeomorphism φ : M → M such that φ(ψ1 (g)(x)) = ψ2 (g)(φ(x)). See [77] for its proof. ([77] is the paper where the center of mass technique first appeared.) Proposition 8.3 is applied to study Riemannian manifold whose sectional curvature is close to 1 but whose fundamental group is not necessary trivial.

9. Embedding Riemannian manifolds by distance function In the last section we explained the center of mass technique which we can use to construct a diffeomorphism. In Section 6 we mentioned another way, that is to use isotopy theory. In this section, we discuss the third method which was introduced and used by Gromov [68, 69]. In [53] the author remarked that this method can be used to construct a smooth map (projection of a fiber bundle) in collapsing situation (Theorem 11.2). It was further generalized by Yamaguchi [153] (Theorem 11.3) to the case when we assume a bound of sectional curvature from below (but not above). We here explain an alternative proof of Theorem 3.3. This proof is completed by Katsuda [93] based on an idea of Gromov [69]. We assume M, N ∈ Mn (D, v), dH (M, N ) < (n, v, D). (We choose = (n, v, D) > 0 later.) Let ψ : M → N be a 3 Hausdorff approximation. We take a 20-net X of M. We can take X such that if x, x ∈ X, x = x ,

then d(x, x ) > 10,

(∗)

216

K. Fukaya

in addition. It is easy to see that ψ(X) is an 30-net of N . It is also easy to see that if x, x ∈ X, x = x ,

then d ψ(x), ψ(x ) > .

(∗∗)

We denote by [0, 1]X the set of all maps X → [0, 1]. It is a finite-dimensional Euclidean space. The idea is to embed M (respectively N ) in [0, 1]X using the distance function from X (respectively ψ(X)). In order to go around the trouble that the distance function is not differentiable, we proceed as follows. We take so that it is much smaller than the injectivity radius of M and N . We next take χ : R>0 → [0, 1] such that χ(t) =

0, t < C, t, t ∈ [C 2 , C 3 ], const, t > C 4 .

Here C is a sufficiently large positive number which will be determined later. We may assume that C 5 is smaller than the injectivity radius of M and N . (Precisely we first choose C and then choose so that this condition is satisfied.) Then we define IM : M → [0, 1]X by IM (p)(x) = χ(d(p, x)) and IN : N → RX by IN (p)(x) = χ(d(p, ψ(x))). Note χ(t) is a constant where t is larger than the injectivity radius. Hence IM , IN are smooth. We can prove the following L EMMA 9.1. (1) IM , IN are smooth embeddings. (2) IM (M) is contained in a tubular neighborhood U (N ) of IN (N ). (3) We identify U (N ) with a normal bundle and let π : U (N ) → N be the projection of the normal bundle. Then the restriction of π to IM (M) is a diffeomorphism. We omit the details of the proof (see [93]), but explain briefly its idea. The reason that (1) holds is that, for each p, there are sufficiently many points q ∈ X with d(q, p) ∈ [C 2 , C 3 ]. Namely using the distance function from such q we can show the Jacobi matrix of IM , IN are invertible in a neighborhood of p. To prove (2) we observe that, if x ∈ X ⊂ M, then the distance between IM (x) and IN (ψ(x)) is small. (Namely it is something like const dH (M, N ) = const .) Moreover, since X, ψ(X) are enough dense in M, N , it follows that IM (M) are sufficiently close to IN (N ). We next need an estimate of the size of the tubular neighborhoods of IM (M), IN (N). This follows from the estimate of the second derivative of IM and IN , which turn out to be a consequence of the assumption on curvature of M, N . To carry out the actual proof we need to estimate the size of the tubular neighborhood and the distance between IM (x) and IN (ψ(x)) more precisely. To prove (3) we need to see that the Jacobi matrix of the restriction of π : U (N ) → N to IM (M) is invertible. This follows from the fact that IM (M) is C 1 -close to IN (N ), namely they are close to each other together with their tangent spaces. Since the derivative of the distance function is written in terms of the angle between edges of geodesic triangles, we can prove this fact by using comparison theorems.

Metric Riemannian geometry

217

Theorem 3.3 follows immediately from Lemma 9.1. Remark 9.1. (A) We used distance functions in the discussion above. We can use eigenfunctions of the Laplace operator (or Green kernel) instead. Then the estimate about the derivatives of the diffeomorphism we get becomes better. (See, for example, [16,56,91].) This approach is closely related to harmonic coordinates. (B) We took a net and embed Riemannian manifolds to a finite-dimensional Euclidean space in the above argument. We can use distance functions from all the points and can embed Riemannian manifolds to a Hilbert or Banach space. This argument is useful for a generalization of Theorem 3.3 to an equivariant version. (Namely in the situation when a Lie group acts on M, N .) If we use the eigenfunction of the Laplace operator as we mentioned in (A), embedding to finite-dimensional Euclidean space is good enough to show the equivariant version also.

10. Almost flat manifold In this section we start discussing the case when the injectivity radius goes to zero. In the earlier sections, we began with sphere theorems and applied the method appeared there to finiteness theorems, etc. In sphere theorems, we study manifolds of positive curvature. We here consider another typical Riemannian manifold that is a flat manifold. We first recall the following famous T HEOREM 10.1 (Bieberbach). If M is a compact Riemannian manifold with KM ≡ 0, then there exists a finite covering M˜ of M such that M˜ is isometric to a flat torus. We want to study a Riemannian manifold (M, g) whose curvature is close to zero. To obtain a nontrivial result, we need some normalization. (In fact, the curvature of (M, kg) tends to 0 as k → ∞ for any (M, g).) To normalize volume is not good enough either. (For example, M × S 1 for any M carries a metric with volume 1 and curvature arbitrary small.) So let us normalize the diameter to 1. In other words, we assume |KM | Diam(M)2 is small. We call such manifold almost flat manifold. However the assumption |KM | Diam(M)2 small does not imply that M is diffeomorphic to a flat manifold. E XAMPLE 10.1. We consider the group N of all 3 × 3 matrix of the form

1 x 0 1 0 0

z y 1

.

We consider a left invariant metric g on N such that g = 2 dx 2 + 2 dy 2 + 4 dz2 at the unit matrix I . Let E1 , E2 , E3 be left invariant vectors such that E1 = ∂/∂x, E2 = ∂/∂y, E3 = ∂/∂z at I . It is well known that the curvature of a Lie group with left invariant metric is calculated as follows. If E, F are left invariant orthonormal vectors then the sectional curvature of the plane spanned by them is not greater 6[E, F ]. (See [24].)

218

K. Fukaya

Hence the sectional curvature K(N,g ) is bounded as → 0. On the other hand, we consider the subgroup NZ consisting of matrix in N such that x, y, z ∈ Z. NZ is a discrete subgroup of N and the quotient space M = NZ \ N is known to be compact. We consider the metric on M induced by g and denote it by g¯ . It is easy to see that the diameter Diam(M, g¯ ) goes to zero. Hence Diam(M, g¯ )2 Kg goes to zero. However no finite cover of M is diffeomorphic to T 3 . This example shows that we need to include not only Abelian but also nilpotent Lie groups to characterize almost flat manifolds. T HEOREM 10.2 (Gromov [68]). There exists n > 0 such that if an n-dimensional compact Riemannian manifold M satisfies |KM | Diam(M)2 < n , then M has a finite cover M˜ which is diffeomorphic to Γ \ N , where N is a nilpotent Lie group and Γ is a discrete subgroup. There is an improvement of Theorem 10.2 due to Ruh [133]. Let N be an nilpotent Lie group. There exists a connection ∇can of T N which is invariant to both left and right actions of N . Let Γ be a discrete subgroup of N . ∇can induces a connection on Γ \ N which we denote by the same symbol. (We remark that ∇can is not equal to the Levi-Civita connection.) Let Λ be a finite subgroup of Aut(Γ \ N, ∇can ). We call Λ \ (Γ \ N ) an infranilmanifold. T HEOREM 10.3 (Ruh). Under the assumption of Theorem 10.2, M is diffeomorphic to an infranilmanifold. Let us sketch some of the essential ideas behind the proof of Theorem 10.2. One important origin is Margulis’ lemma. Margulis’ lemma first appeared in the study of discrete subgroup of Lie group. T HEOREM 10.4 (Zassenhaus, see [69, 8.44]). For each Lie group G there exists a neighborhood U of the unit, such that if Γ ⊂ G is a discrete subgroup then U ∩ Γ generates a nilpotent subgroup. The proof is based on the following fact. Let g1 , g2 ∈ G be in a neighborhood of the unit 1, then d 1, {g1 , g2 } Cd(1, g1 )d(1, g2 ).

(10.1)

Here {g1 , g2 } is the commutator. This formula (10.1) is a consequence of the fact that the derivative of (g1 , g2 ) → {g1 , g2 } at 1 is zero. Once we have (10.1) we can prove Theorem 10.4 as follows. We choose U small enough such that if g ∈ U then d(1, g) < 1/(2C). Then (10.1) implies that if gi ∈ U , then d(1, {g1 , g2 }) is strictly smaller than d(1, gi )/C. We repeat this and find that N hold commutator between elements of U is in the 1/C N neighborhood of 1. Since Γ is discrete, it implies the existence of N such that any N hold

Metric Riemannian geometry

219

commutators between elements of U ∩ Γ are trivial. It follows that U ∩ Γ generates a nilpotent group. There are various Riemannian geometry versions of Theorem 10.4. The following, which is proved by Cheeger–Colding [31] (improving [59]) is one of the strongest versions. T HEOREM 10.5. There exists n with the following properties. Let M be an n-dimensional complete Riemannian manifold with RicciM −(n − 1) and p ∈ M. Then the image of π1 (Bp (n , M)) → π1 (Bp (1, M)) has nilpotent subgroup of finite index. If we apply it to the situation of Theorem 10.2 we find that the fundamental group of M has nilpotent subgroup of finite index. (See Section 19 for more discussions on the fundamental group.) Another idea applied by Gromov to prove Theorem 10.2 is to use local fundamental pseudogroup, which we discuss briefly here. (See [57, Section 7] and [24] for its precise definition.) Let M be a complete Riemannian manifold. We assume KM 1. Let p ∈ M. Then by Theorem 4.5 the exponential map Expp : Tp M → M is an immersion on the ball B0 (π; Tp M). Since B0 (π; Tp M) has a boundary, Expp : B0 (π; Tp M) → M is not a covering map. So we cannot consider its deck transformation group in the usual sense. But we can define a “pseudogroup” in the following way. Let < π/10. We consider the set of all loops : S 1 → Bp (, M) with (0) = p and || < . We say ∼ for such , if there exists a based homotopy t between them such that |t | < for each t. Let us denote the set of equivalence classes by π1 (M, p; ). The loop sum ∗ on π1 (M, p; ) is not necessary defined. But when it is defined, its ∼ equivalence class is well defined. (We need to use the fact that Expp : B0 (π; Tp M) → M is an immersion to show this.) When loop sum is well defined it is associative. (Here the reader may find some flavor of Klingenberg’s argument we mentioned at the end of Section 4.) Thus (π1 (M, p; ), ∗) is something similar to a group. We call it a fundamental pseudogroup. The following pseudogroup version of Margulis’ lemma is used in the proof of Theorem 10.2. L EMMA 10.6. If |KM | 1 and if Diam(M) < n , then there exists a subpseudogroup (π10 (M, p; ), ∗) of (π1 (M, p; ), ∗) such that (π10 (M, p; ), ∗) is embedded ( preserving ∗) into a nilpotent Lie group N , its image generates a discrete subgroup Γ and that the index [π1 (M, p; ) : π10 (M, p; )] is estimated by a number depending only on n. Here n - - 1. Lemma 10.6 is the main part of the proof of Theorem 10.2. (Actually we need a bit more. Namely we have to show that the action of (π1 (M, p; ), ∗) to B0 (π; Tp (M)) is diffeomorphic to an action to U ⊂ N of some subpseudogroup Γ ∩ U , where N is a nilpotent Lie group and Γ is its discrete subgroup.) For the details of the proof, we refer to [24,57].

220

K. Fukaya

11. Collapsing Riemannian manifolds—I Using Theorems 3.3, 3.4, we can describe a sequence of n-dimensional Riemannian manifolds Mi with |KMi | 1 and Vol(Mi ) v > 0 where v is independent of i. Namely the limit X (which exists after taking a subsequence) is a Riemannian manifold of C 1,α -class and X is diffeomorphic to Mi for sufficiently large i. In Section 10, we considered a sequence of Riemannian manifolds Mi with |KMi | 1 and Diam(Mi ) → 0. (The second condition is equivalent to saying that Mi converges to a point.) Theorem 10.2 implies that Mi is an infranilmanifold for large i. These are two extremal cases. We now discuss the intermediate case. Namely we consider the case when a sequence of Riemannian manifolds Mi converges to a metric space X (with respect to the Gromov–Hausdorff distance) such that n > dim X > 0. We say that such sequence collapses to X. Here we discuss results under the assumption |KMi | 1. (The study under weaker assumption is discussed in later sections.) We first explain some examples of collapsing Riemannian manifolds. The first example is due to Berger and is called the Berger sphere. E XAMPLE 11.1. We consider the Hopf fibration π : S 3 → S 2 . (Namely we regard S 3 = {(z1 , z2 ) ∈ C2 | |z1 |2 + |z2 |2 = 1}, and we associate to (z1 , z2 ) the complex onedimensional space spanned by it, which is an element of CP 1 = S 2 .) We put the standard metric on S 3 and regard S 2 as a sphere of radius 1/2. It is easy to see that π is a Riemannian submersion. (Namely if Vh ∈ Tp S 3 and V is perpendicular to the fiber of π containing p, then gS 3 (Vh , Vh ) = gS 2 (π∗ Vh , π∗ Vh ).) We define a metric g on S 3 as follows. Let V , W ∈ Tp S 3 . We write V = Vh + Vv ,

W = Wh + Wv ,

where Vh , Wh are perpendicular to the fiber (with respect to gS 3 ) and Vv , Wv are tangent to the fiber. We set g (V , W ) = gS 3 (Vh , Wh ) + 2 gS 3 (Vv , Wv ). 3 2 It is easy to see that limGH →0 (S , g ) = (S , gS 2 ). We can check that the sectional curvature 3 of (S , g ) is between 0 and 1 if ∈ (0, 1].

We can generalize this construction and prove the following P ROPOSITION 11.1. Let M be a compact manifold on which a torus T m acts. We assume that there is no point p on M such that p is fixed by all the elements of T m . Then there exists a family of metrics g on M such that Kg is bounded from below and above and m that limGH →0 (M, g ) = M/T . To find such a sequence of metrics, we first take a T m invariant Riemannian metric gM on M. We next take X an element of the Lie algebra of T m such that the subgroup ∼ =R

Metric Riemannian geometry

221

generated by X is dense in T m . We regard X as a (Killing) vector field on M. We remark that X never vanishes on M. For V , W ∈ Tp M we put V = Vh + c(V )X(p),

W = Wh + c(W )X(p),

where gM (Vh , Xp ) = gM (Wh , Xp ) = 0. We then define g (V , W ) = gM (Vh , Wh ) + 2 c(V )c(W )gM X(p), X(p) . We can prove that the limit of (M, g ) as → 0 is M/T m with quotient metric and the sectional curvature of (M, g ) is bounded for ∈ (0, 1]. Let us take, for example, M = S 3 . We can find an action of T 2 on S 3 satisfying the assumption of Proposition 11.1. Hence there exists a sequence of metrics on S 3 such that the limit is S 3 /T 2 = [0, 1], the interval. In particular the limit space is not a manifold. This construction is further generalized in [38] (Theorem 12.1). There are two approaches to study collapsing Riemannian manifolds under the assumption |KMi | 1. One is due to Cheeger–Gromov [39,38], the other is due to the author [53, 55,56]. These two approaches are unified in [34]. In this section, we discuss the second approach and in the next section we discuss the first (and the joined) approach. Here we discuss the following two problems. For n, D, we denote by Mn (D) the set of all isometry classes of n-dimensional Riemannian manifolds M such that |KM | 1, and Diam(M) D. P ROBLEM 11.1. Let Mi ∈ Mn (D) and X = limGH i→∞ Mi = X. (1) What kind of singularity can X have? (2) Describe the relations between X and Mi . We remark that if we replace Mn (D) by Mn (D, v), the answers are Theorems 3.3, 3.4. Problem 11.1 will be studied also under milder assumptions on curvature later. We first discuss Problem 3.4(2) in the special case when X is a smooth manifold. T HEOREM 11.2 (Fukaya [53,56]). Let Mi ∈ Mn (D). Suppose B = limGH i→∞ Mi is a smooth Riemannian manifold. Then, for large i, there exists a fiber bundle πi : Mi → B with the following properties: (1) The fiber is diffeomorphic to an infranilmanifold F . (2) The structure group is the group of affine transformations Aff(F, ∇can ), where we define the affine connection ∇can on F as in the last section. (3) πi is an almost Riemannian submersion in the following sense. If V ∈ Tp (Mi ) is perpendicular to the fiber then 1 − i

0 and n ∈ Z+ , there exists ρ = ρ(, n) > 0 with the following properties. Let (M, g) be a complete n-dimensional Riemannian manifold with |Kg | 1. Then there exists a metric g and Up , U˜ p , Γp , Np for each p ∈ M such that: (1) Np is nilpotent. Γp ⊂ Np is a discrete subgroup such that π0 (Np ) is finite and Np is generated by its connected component Np,0 and Γp .

228

K. Fukaya

(2) Up is a neighborhood of p and Up ⊇ Bp (ρ, M). (3) Np acts on (U˜ p , g˜ p ) by isometry. Here U˜ p is a covering space of Up and g˜ p is the metric induced by g . ˜ = p then iU˜ p (p) ˜ > ρ. (4) If p˜ ∈ U˜ p and [p] (5) [Γp : Γp ∩ Np,0 ] < k. (6) For any x ∈ U˜ p Diam(Γp \ Np x) < . Here Np x is an Np -orbit. Moreover we have (7) e− g < g < e g. (8) |∇ g − ∇ g | < . (9) |∇ g Rg | < c(n, i, ), where Rg is the curvature tensor of g and c(n, i, ) depends only on n, i, . Remark 12.2. The existence of g satisfying (7)–(9) is proved by [15,1]. Remark 12.3. The metric satisfying (1)–(6) is called a (ρ, k)-round metric in [34]. We remark that at the point where iM (p) > ρ we may take Np = 1 and U˜ p = Up . Hence the statement above is obvious. On the other hand, condition (4) implies that at the point p where injectivity radius is small, the group Np is nontrivial. Hence, together with (1), we obtain a local action of a torus by restricting the action of Np to the center. Using (6) and the fact that the local action of the torus is compatible with the metric g , we can prove that these actions are compatible in the sense of Theorem 12.2. Moreover, in the case when the diameter of M is smaller than a constant depending only on and n, we can prove that the group Np is independent of the choice of p. Hence its orbits defines a foliation on the frame bundle of M. It implies Theorem 11.5. Thus Theorem 12.7 unifies two approaches for collapsing Riemannian manifolds. The proof of Theorem 12.7 is a combination of the proofs of Theorems 12.2 and 11.5. We use Theorem 11.5 and its proof (together with some improvement) to find Np locally. We then glue them in a way similar to the proof of Theorem 12.2. Finally we take the average and obtain the required metric g . E XAMPLE 12.2. Let Γ be a lattice of a semisimple Lie group G of noncompact type and G/K be a symmetric space. We assume Γ \ G/K is noncompact. Then for each p ∈ G/K the group Γp = {g ∈ Γ | d(p, gp) < } has nilpotent subgroup Γp,0 of finite index [Γp : Γp,0 ] < k by Theorem 10.4. (We remark that Γp may not be contained in a small neighborhood of the unit in G. But its subgroup of finite index is in a small neighborhood of the unit.) The Zariski closure Np ⊂ G of Γp,0 is a nilpotent group. This is our Np . The original metric (the metric of symmetric space) is an invariant of left the Np action. Hattori [84] found the following. Let M = Γ \ G/K be a locally symmetric space of noncompact type. We assume that it is noncompact and of finite volume. Then the limit (M, gM /R) as R goes to infinity is a cone of a simplicial complex T which is called the Tits building. (Here the limit is taken with respect to the pointed Gromov–Hausdorff distance (Definition 16.3).) Now if we take a simplex of T then a “neighborhood” of it in M is diffeomorphic to × [0, ∞) × Γ () \ N (). The dimension of the nilmanifold

Metric Riemannian geometry

229

Γ () \ N () depends on . They are glued appropriately, which gives a structure as in Theorem 12.7. The following addendum to Theorem 12.7 is useful for various applications. P ROPOSITION 12.8 [132]. If a KM b in Theorem 12.7 then we may choose g so that a + KM b − .

13. Collapsing Riemannian manifolds—III In this section, we review some of the applications of collapsing Riemannian manifolds. We recall that Mn (D) is the set of isometry classes of n-dimensional Riemannian manifold M with Diam(M) D, |KM | 1. T HEOREM 13.1 (Fang–Rong [51], Petrunin–Tuschmann [126]). For each n, D the number of diffeomorphism classes of simply connected manifolds M in Mn (D) with finite π2 (M) is finite. T HEOREM 13.2 (Fang–Rong [51], Petrunin–Tuschmann [126]). There exists i(n, δ) > 0 such that if M is simply connected, π2 (M) is finite and if 1 KM δ > 0, then the injectivity radius of M is larger than i(n, δ). We remark that, in case the dimension is even, Theorem 13.2 follows from Theorem 4.3 without assumption on π2 . E XAMPLE 13.1. We first consider the Lens space S 3 /Zp where Zp ⊂ S 1 is a cyclic group of order p. Its curvature is 1 and its limit is S 2 = S 3 /S 1 . This example shows the assumption on π1 (M) is necessary both in Theorems 13.1, 13.2. The three examples below show that the assumption on π2 (M) is also necessary in Theorems 13.1, 13.2. We consider the Lie group SU(3). It has a metric with positive sectional curvature. We consider its maximal torus T 2 ⊆ SU(3). Let pi , qi be coprime integers such that lim pi /qi = α ∈ R \ Q. We identify T 2 = R2 /Z2 and let x, y be coordinates of R2 . We consider Si1 = {[x, y] ∈ T 2 | y = pi x/qi }. We equip Mi = Si1 \ SU(3) with quotient Riemannian metric. Mi is a sequence of 7-dimensional manifolds of positive curvature. Using the fact lim pi /qi is irrational, we can easily find that the limit of Mi with respect to the Gromov–Hausdorff distance is T 2 \ SU(3). We can also prove that the sectional curvature of Mi is uniformly positive. Namely C KMi δ > 0 for some δ, C independent of i. (This is a consequence of the fact that pi /qi converges. We remark that π2 (Mi ) ∼ = π1 (S 1 ) = Z.) In a similar way we can use the T 2 -action to S 3 × S 3 to get a sequence of metrics gi on 2 S × S 3 with C Kgi δ > 0 such that (S 2 × S 3 , gi ) converges to S 2 × S 2 . We next consider an action of T 2 × T 2 on SU(3) where the first factor acts by left multiplication and the second factor acts by right multiplication. Using an appropriate family of

230

K. Fukaya

Si1 ∼ = S 1 ⊆ T 2 × T 2 , Petrunin–Tuschmann [126] (using Eschenburg [50]) found an example of Mi = Si1 \ SU(3) with C KMi > δ > 0 such that Mi converges to T 2 \ SU(3)/T 2 . Remark 13.1. A similar π2 -assumption as in Theorem 13.2 was proposed by the author in [57, Remark 15.10]. However [57, Conjecture 15.7] (by the author) turns out to be false. A counterexample (due to Petrunin–Tuschmann) is the last example in Example 13.1. We now sketch the proof of Theorem 13.1. We start with the following L EMMA 13.3 (Rong [132]). If we assume that π1 (M) is finite in the situation of Theorem 11.5 in addition, then the fiber of π : F Mi → Y in Theorem 11.5 is diffeomorphic to a flat manifold. Using the fact that the fundamental group of Mi is finite (here we assume dim Mi > 2), it follows easily that the fundamental group of the fiber has an index finite Abelian subgroup. Since the fiber is an infranilmanifold the lemma follows immediately. Lemma 13.3 implies that we have an F -structure whose orbits are fibers. (Here our F -structure is one called pure F -structure by Cheeger–Gromov [38]. A pure F -structure is an F -structure such that all the orbits of the local action have the same dimension.) We next apply the averaging process in the proof of Theorem 12.7 to the situation of Theorem 11.5 and of Lemma 13.3. Then we have L EMMA 13.4. In the situation of Lemma 13.3, we can approximate the Riemannian metric on F Mi by g in the same sense as Theorem 12.7(7)–(9) so that g is an invariant of the local T k action and of the O(n) action. Now we start the proof of Theorem 13.1. We assume that Theorem 13.1 is false. Then there exists a sequence Mi ∈ Mn (D) such that Mi is simply connected, π2 (Mi ) is finite, and Mi is not diffeomorphic to Mj for i = j . We take F Mi and may assume that it converges to Y . Since we approximate the metric by one satisfying Theorem 12.7(7)–(9), it follows that Y is a smooth Riemannian manifold. We may replace F Mi by its finite cover F˜ Mi so that it has global a T k × G-action, where G is a compact group20 and a T k -orbits are the fibers of the fibration F˜ Mi → Y . We modify the metric of F˜ Mi so that it is T k × Gequivariant. The next lemma is the place where we use the key assumption that π2 (Mi ) is finite. L EMMA 13.5. If F˜ Mi /T k is G-diffeomorphic to F˜ Mj /T k then F˜ Mi is T k × Gdiffeomorphic to F˜ Mj . In fact the torus bundle T k → E → B is determined by the (T k -analogue of) Euler class ∈ Hom(H2 (B), π1 (T k )) (which is well defined up to Aut(π1 (T k ))). In our case where π2 (F˜ Mj ) is finite and π1 (F˜ Mj ) is trivial, the Euler class is an isomorphism 20 Actually it is finite covering group of O(n). (It may be disconnected.)

Metric Riemannian geometry

231

H2 (B)/Tor → π1 (T k ), hence it is unique up to Aut(π1 (T k )). To obtain the T k -equivariant diffeomorphism F˜ Mi → F˜ Mj which is G-equivariant also, we use the center of mass technique (Proposition 8.3). We remark that F˜ Mj /T k has the same dimension as Y and F˜ Mj /T k converges to Y with respect to the G-Gromov–Hausdorff topology (which was introduced in [52]). Estimate (9) of Theorem 12.7 implies that Y is a smooth manifold. On the other hand, the sectional curvature of F˜ Mj /T k is bounded from below. Hence Theorem 11.3 implies that F˜ Mj /T k is diffeomorphic to Y for large i. We can use the G-equivariant version of Theorem 11.3 (which can be proved in the same way as Theorem 11.3 using an embedding to Hilbert space as in [55]21 ), F˜ Mi /T k is G diffeomorphic to Y for large i. Hence Lemma 13.5 implies that F˜ Mj is G diffeomorphic to F˜ Mj for i, j large. Namely Mi is diffeomorphic to Mj . This is a contradiction. To prove Theorem 13.2 we need another result by Petrunin–Rong–Tuschmann. T HEOREM 13.6 [125]. Let M be a compact manifold. We assume that M admits a sequence of metrics gi . We assume that Λ Kgi λ and that the metric space X = limGH i→∞ (M, gi ) is of dimension strictly smaller than M. We also assume that the distance function di : M × M → R induced by gi converges to a function d which determines a pseudometric22 on M. Then λ 0. Remark 13.2. Klingenberg and Sakai conjectured a similar statement, but their conjecture does not assume the additional assumption that di converges to a pseudometric. To prove Theorem 13.2 using Theorem 13.6 we proceed as follows. We assume that there exists Mi with 1 KKi δ > 0 and that the injectivity radius goes to 0. We can discuss in the same way as in the proof of Theorem 11.3 to show that Mi is diffeomorphic to Mj .23 By looking at the proof carefully we may assume that the diffeomorphism almost preserves distance function. Namely if we identify Mi with Mj then the sequence M = Mi = Mj satisfies the assumption of Theorem 13.6. The conclusion of Theorem 13.6 contradicts to KMi δ > 0. One of the ideas of the proof of Theorem 13.6 is the following observation. If the collapsing occurs in the same way as in the proof of Proposition 11.1, then the sectional curvature of the plain spanned by X and other vector is always converges to zero. To make this simple idea works we need a lot of delicate work which is not described here. We next discuss some other applications of collapsing theory. 21 See Remark 9.1. 22 Namely it satisfies axioms of metric except “d(x, y) implies = y”. 23 We use Proposition 12.8 to show Λ + K Mi λ − .

232

K. Fukaya

T HEOREM 13.7 (Rong [132]). There exists w(n, δ) such that if a compact n-dimensional Riemannian manifold M satisfies 1 KM δ then there exists a cyclic subgroup C of the fundamental group π1 (M) such that [π1 (M) : C] < w(n, δ). Remark 13.3. If we assume 1 KM 0, Diam(M) < D, then there exists an Abelian subgroup C of π1 (M) such that [π1 (M) : C] < w(n, D) [132]. There are results under milder assumption that is the case when M is of almost of nonnegative curvature. See Section 19. The following is another application of collapsing theory. This time we apply to manifolds of almost nonpositive curvature. T HEOREM 13.8 (Fukaya–Yamaguchi [58]). There exists (D, n) such that if a compact n-dimensional Riemannian manifold M satisfies Diam(M) D, (D, n) KM −1 then the universal covering space of M is diffeomorphic to Rn . This is a generalization of Hadamard–Cartan’s theorem (Theorem 4.6), which is the case when KM 0.

14. Morse theory of distance function So far we mainly discussed results assuming the curvature to be bounded from above and below. From this section on, we consider the case when the curvature is bounded from below only. The next theorem is a corollary of Theorem 4.1. T HEOREM 14.1 (Rauch). Let M be a compact manifold without boundary. If there exists a Morse function on M with two critical points, then M is homeomorphic to a sphere. In Section 4, we started with Theorem 4.1 and showed the way to prove sphere theorems, finiteness theorems and compactness theorems by estimating the number of balls we need to cover a manifold. The number of contractible open subsets one needs to cover the space (plus one), is called the Lusternik–Shnirel’man category and is important in Morse theory. In this section we will try to apply Morse theory directly. For a given Riemannian manifold M, a function which is determined automatically from the metric is a distance function dp (x) = d(p, x) from a point. (Note we can use the fact that p = x is the unique critical point of x → dp (x) with d(x, p) < iM (p) to prove that Bp (r, M) is diffeomorphic to a sphere if r < iM (p).) The difficulty to apply Morse theory to the distance function is that x → dp (x) is not differentiable for d(x, p) > iM (p). (dp is not differentiable at p either. But this does not cause serious trouble. We may consider dp2 instead, for example.) During the proof of Theorem 2.5, Grove–Shiohama applied Morse theory away from the ball with radius = iM (p). After that, their method is used in many other places. The main idea of them is the following definition.

Metric Riemannian geometry

233

Fig. 14.1.

D EFINITION 14.1. We say q is a regular point of dp if there exists a nonzero vector V. ∈ Tq M such that for any minimal geodesic : [0, d(p, q)] → M joining p and q, the . angle between d dt (0) and V is not greater than π/2. For example, let p, q be as in Figure 14.1. It is not clear how many minimal geodesics are there joining p with q. But it is easy to see that the direction of any of them is downwards at q. Hence q is a regular point of dp . Remark 14.1. We may consider various situations similar to Definition 14.1. For example, let us consider a continuous function f which is an infimum of finitely many differentiable functions fα locally (namely f = inf fα ).24 In this case we say q is a regular point of f if there exists a vector V. ∈ Tq M such that, for each α with f (q) = fα (q), we have V. (fα ) > 0. We can apply a similar argument to a linear combination of finitely many dp ’s or the infimum of them also. Proposition 14.2 holds for such cases. Based on Definition 14.1, we can prove the following analogue of the Morse lemma for dp . P ROPOSITION 14.2. If q with a dp (q) b is an arbitrary regular point of dp , and if Bp (b, M) is compact, then Bp (b, M) \ Bp (a, M) = {q ∈ M | a d(p, q) b} is homeomorphic to a direct product of ∂Bp (b, M) = {q ∈ M | d(p, q) = b} and [0, 1]. The proof is similar to the proof of the following famous 24 We remark that d may not satisfy this condition in general. p

234

K. Fukaya

T HEOREM 14.3 (Morse lemma). We assume that f : M → R is differentiable, and arbitrary q with f (q) ∈ [a, b] is a regular point of f , and that f −1 ([a, b]) is compact. Then f −1 ([a, b]) is diffeomorphic to f −1 ({a}) × [0, 1]. The proof of Morse lemma uses an integral curve of grad f . (See [103].) Since dp is not differentiable, the vector field grad dp does not make sense. Instead, we will use the vector field V constructed below. For q ∈ Bp (b, M) \ Bp (a, M) let Vq = V be the vector ∈ Tq M as in Definition 14.1. If we can take Vq depending smoothly on q, then we can take the vector field V (q) = Vq in place of − grad f . (The condition in Definition 14.1 implies that dp decreases along the integral curve of V .) To find Vq depending smoothly on q, we proceed as follows. We first take V˜q which may not depend smoothly on q. We extend it to its neighborhood and denote it by the same symbol V˜q . Then if q is in a small neighborhood U (q) of q, then the vector V˜q (q ) ∈ Tq M satisfies the condition of Definition 14.1. We cover Bp (b, M) \ Bp (a, M) by finitely many U (qi )’s. We then take a partition of unity χi and put V (q) =

χi (q)V˜qi (q).

It is easy to see that this V has the required properties. Using this vector field V , the proof of Proposition 14.2 goes in the same way as the proof of Morse lemma. To apply Morse theory of dp to the proof of Theorem 2.5 we need the following lemma. L EMMA 14.4. We assume that M satisfies the assumption of Theorem 2.5. Let p, q ∈ M with d(p, q) = Diam(M), and x ∈ M be a point different from p, q. Let p : [0, d(p, x)] → M, q : [0, d(q, x)] → M be minimal geodesics joining x to p, and x to q, respectively. (In case there are several of them, we assume any of them have the property below.) d d Then the angle between two tangent vectors dtp (0) and dtq (0) ∈ Tx M is greater than π/2 (see Figure 14.2). The proof of Lemma 14.4 uses Toponogov’s comparison theorem (Theorem 4.7). Under the assumption of Lemma 14.4 (that is KM 1/4), Theorem 4.7 implies the following Sublemma 14.5. Let x, y, z ∈ M. We consider the geodesic triangle whose vertices are those three points. We denote the length of its edges by |xy| etc. and angles by xyz etc. We put X = |yz|, Y = |zx|, Z = |xy|. S UBLEMMA 14.5. If zxy π/2, then cos X2 cos Y2 cos Z2 . Note that we have s(Y/2, Z/2, θ, 1) s(Y/2, Z/2, π/2, 1) = cos−1 (cos Y/2 cos Z/2), where s(·, ·, ·, ·) is as in Theorem 4.7.

Metric Riemannian geometry

235

Fig. 14.2.

We start the proof of Lemma 14.4. We put |p | = t, |q | = s, d(p, q) = D. Since dp attains its maximum at q it follows that q is not a regular point of dp . Hence there exists a geodesic joining p and q such that the angle between and q is not greater than π/2. We apply Sublemma 14.5 to the geodesic triangle consisting of , p , q and obtain cos

t s D cos cos . 2 2 2

(14.1)

Since D/2 > π/2 we have cos D2 < 0. Therefore one of cos 2s , cos 2t is positive. We may assume cos 2s > 0. If the angle between p and q is not greater than π/2, then we can again apply Sublemma 14.5 and obtain cos

s t D cos cos . 2 2 2

(14.2)

Since cos 2s > 0, (14.1), (14.2) imply cos

D D s cos2 cos . 2 2 2

We remark that 0 < D/2, s < π .25 This is then a contradiction.

Now Lemma 14.4 implies that if x = p, q then x is a regular point of dp , dq . In fact, let V be the tangent vector of q at x. It follows from Lemma 14.4 that the vector field 25 This is a consequence of Myers’ theorem (Theorem 5.4).

236

K. Fukaya

V satisfies the condition in Definition 14.1. Namely x is a regular point of dp . Now we can use Proposition 14.2 to prove that M is homeomorphic to sphere. Hence we proved Theorem 2.5. We remark that we proved Proposition 4.4 during the proof of Lemma 14.4. In fact, we proved cos t/2 > 0 or cos s/2 > 0 there. It implies t < π or s < π . The method we explained above is very useful to study Riemannian manifolds under the bounds of sectional curvature from below. It is also useful to study Alexandrov spaces (see Sections 17, 18). Theorem 2.5 is a sphere theorem. There are several finiteness theorems corresponding to it. The first one is the following, which is called the Gromov’s Betti number estimate. T HEOREM 14.6 (Gromov [70]). There exists C(n) such that if an n-dimensional compact Riemannian manifold M satisfies KM −κ (κ 0) and if its diameter is D then

rank Hk (M; F ) C(n)1+κD .

k

Here F is an arbitrary field. Note in the case when κ = 0, the right-hand side is independent of D. It follows from Theorem 14.6 that the connected sum of sufficiently many copies of CP 2 does not carry a metric of nonnegative sectional curvature. The proof of Theorem 14.6 is based on Morse theory of a kind of distance function. Namely we use an idea similar to the Morse inequality to estimate the Betti number in terms of the number of critical points. However the proof is more involved since Morse theory of the distance function itself does not work. The actual proof requires more complicated argument, which we omit here. There are many other applications of Morse theory of distance functions to metric Riemannian geometry. For example, Gromov used it to show that a complete manifold M such that 0 > −a 2 KM −b2 and has finite volume is diffeomorphic to the interior of a compact manifold with boundary [67]. Let us add a few more remarks to Theorem 2.5. If we assume 1 KM 1 − in addition in Theorem 2.2, then we can show that M is not only diffeomorphic but is also close to a sphere as a Riemannian manifold. Namely if a sequence of n-dimensional simply connected Riemannian manifolds Mi satisfies 1 KMi 1 − 1/i, then Mi converges to S n with standard metric with respect to the Gromov–Hausdorff distance. On the contrary, the corresponding statement in the situation of Theorem 2.5 does not hold. Namely, let us consider a sequence of Riemannian manifolds Mi such that KMi 1 and that the diameter of Mi converges to π as i goes to infinity. Then Theorem 2.5 implies that Mi is homeomorphic to a sphere. However it is not true that the limit of Mi with respect to the Gromov–Hausdorff distance is isometric to the sphere with standard metric. We remark however a Riemannian manifold with diameter = π and KM 1 (actually weaker assumption Ricci n − 1 is enough) is isometric to the sphere. (Theorem 21.11 [148,41].)

Metric Riemannian geometry

237

Fig. 14.3.

In fact, let us consider the quotient of S 2 by the action of S 2 /Zp generated by the rotation of angle 2π/p around the fixed axis. The quotient is a Riemannian manifold with constant curvature 1, except two points where the axis intersects with S 2 . We approximate the quotient space by a Riemannian manifold with curvature 1 and obtain a sequence of Riemannian manifolds Mi whose diameter converges to π and KMi 1. The limit is S 2 /Zp and is not isometric to the sphere with standard metric. The essential point here is that the Alexandrov space X with diameter = π and KM 1 is not necessary isometric to a sphere with standard metric. (Compare Theorem 23.11.) (See Figure 14.3.) This is related to the fact that the limit of Riemannian manifolds Mi with KMi const is rather different from a Riemannian manifold even in the case when the limit has the same dimension. For example, we consider a boundary S of a convex set in R3 . There is a point of S that has no tangent plane. In the situation when the absolute value of the sectional curvature is bounded, the Gromov–Hausdorff convergence is equivalent to the C 1,α -convergence of the metric tensor (in the situation when the limit has the same dimension), by Theorem 3.4. Therefore the limit space has a tangent space everywhere. By the reason we explained above the following question is yet open. P ROBLEM 14.1. Is there any n > 0 such that if M is an n-dimensional complete Riemannian manifold with KM 1 and Diam(M) π − n , then M is diffeomorphic to a sphere? We remark that in the proof of Theorem 2.5 we consider the distance functions dp , dq simultaneously where p, q lie in the different sides from x. This is similar to the notion strainer used in Alexandrov space. (See Section 17.)

15. Finiteness theorem by Morse theory In this section, we explain the idea of the proof of Theorem 3.5. The first half of it, which was proved in [78], asserts that the number of homotopy classes represented by an element

238

K. Fukaya

of Mn (D, v) is finite. (We recall that M ∈ Mn (D, v) if KM 1, Diam(M) D, and Vol(M) v, dim M = n.) In this section we mainly explain this part. The key of the proof is the following proposition. P ROPOSITION 15.1. There exists = (n, D, v) > 0 such that the following holds for each M ∈ Mn (D, v). Let p, q ∈ M with d(p, q) < , p = q. Then q is a regular point of dp . Moreover we have the following. We put = {(x, x) ∈ M × M | x ∈ M}, () = {(x, y) | d(x, y) < }. Then is a deformation retract of (). The deformation retraction H : () × [0, 1] → () can be chosen so that the length of the curve t → H (p, q, t) is not greater than Cd(p, q). Here C depends only on n, D, v. Using Proposition 15.1, the proof of Theorem 3.5 goes in a way similar to the proof of Proposition 5.5. Namely, from the first half of the Proposition 15.1, we find that the metric balls Bp (, M) of radius are contractible in M. On the other hand, the number of metric balls Bp (, M) we need to cover M is estimated in the same way as in Section 5 by using Proposition 5.2. However since it is not clear whether the intersection of finitely many metric balls Bp (, M) is contractible or not in our case, so we need to modify the proof of Proposition 5.5 a bit. The second half of Proposition 15.1 is used for this purpose. We omit this part of the proof. The proof of Proposition 15.1 is closely related to the proof of Proposition 5.6. So we first sketch the proof of Proposition 5.6. By Theorem 4.9 we only need to estimate the length of closed geodesic of minimal length from below for M ∈ Mn (D, v). Let : S 1 → M be the closed geodesic of length . We take an arbitrary point x ∈ M, and let (t) ∈ (S 1 ) be the point of smallest distance from x. Then is orthogonal to x(t) at (t).26 (Here x(t) is a minimal geodesic joining x and (t).) We put (0) = p. Since d(p, (t)) , it follows that if d(x, p) is sufficiently larger than , then the angle between and xp is close to π/2. We thus have proved the following lemma. L EMMA 15.2. Let δ, ρ > 0. Then there exists depending only on n, D, v, δ, such that if is a closed geodesic with length < and if (0) = p then M is contained in the image of the exponential map of the domain ⊂ Tp M in Figure 15.1. We can choose δ sufficiently small compared to the diameter D, so that the volume of the image of the domain II in Figure 15.1 is smaller than v/2. By choosing ρ small we may assume the volume of the image of the domain I in Figure 15.1 is smaller than v/2 also. Therefore if there exists a closed geodesic of length < , then the volume of M is smaller than v. We turn to the proof of Proposition 15.1. It suffices to show the following lemma. 26 More precisely in case x is a cut point with respect to the geodesic . (The notion of the cut point with respect to a submanifold is defined in a similar way to the notion of cut point from a point. See, for example, [33].) may not be orthogonal to x(t). However this does not cause a trouble for the proof of Lemma 15.2 since the measure of the set of cut points is zero.

Metric Riemannian geometry

239

Fig. 15.1.

Fig. 15.2.

L EMMA 15.3. There exist θ = θ (n, v, D) > 0 and = (n, v, D) > 0 with the following properties. Let M ∈ Mn (D, v), p, q ∈ M, d(p, q) < . Let 1 and 2 are minimal geodesics joining p and q. Then the angle between 1 and 2 at p or q is smaller than π − θ . Lemma 15.3 implies that q in the lemma is a regular point of dp . The first half of Proposition 15.1 follows from Proposition 14.2. The second half can also be proved in the same way by examining the proof of Proposition 14.2 carefully. The proof of Lemma 15.3 is similar to the proof of Proposition 5.6. Namely we replace Figure 15.1 by Figure 15.2. We thus explained an outline of the first half of the proof of Theorem 3.5. The other half is the finiteness of the number of homeomorphism classes and requires another deep argument. The main new technique required is the idea from controlled surgery.

16. Soul theorem and splitting theorem Typical results on noncompact complete Riemannian manifolds of nonnegative curvature are the soul theorem and the splitting theorem. They also are very useful to study the local structure of the Gromov–Hausdorff limit of Riemannian manifolds or its limit. We first explain why the study of noncompact manifolds is useful to study local structure of the limit space. Let us begin with the introducing some notations. Let X be a metric

240

K. Fukaya

space and : [a, b] → X be a continuous map (that is a curve). The length || of is by definition a supremum of the sum d (ti ), (ti+1 ) , where a = t0 < t1 < · · · < tN = b runs over all partitions (N moves also). D EFINITION 16.1. We say that X is a length space if for each p, q ∈ X there exists a curve joining p, q of length d(p, q). A complete Riemannian manifold is a length space. The Gromov–Hausdorff limit of length spaces is also a length space. D EFINITION 16.2. A complete metric space is said to be compactly generated if all of its metric balls are compact. The set of all isometry classes of compact metric spaces is complete with respect to the Gromov–Hausdorff distance. A natural metric to put on the set of all isometry classes of complete compactly generated spaces is pointed the Gromov–Hausdorff distance, which we define below. D EFINITION 16.3. Let X, Y be metric spaces and x ∈ X, y ∈ Y . We say that the pointed Gromov–Hausdorff distance dpGH ((X, x), (Y, y)) between (X, x) and (Y, y) is not greater than , if the Gromov–Hausdorff distance between the metric balls pGH B1/ (x, X) and B1/ (y, Y ) is not greater than . We write limi→∞ (Xi , xi ) = (X, x) if limi→∞ dpGH ((Xi , xi ), (X, x)) = 0. The following can be proved in the same way as Theorem 3.2. T HEOREM 16.1. The set of all isometry classes of a pair (M, p) of an n-dimensional Riemannian manifold M with RicciM −(n − 1) and a point p on it is relatively compact with respect to the pointed Gromov–Hausdorff distance. Now we can define the tangent cone. Let (X, dX ) be a length space and x ∈ X. pGH

D EFINITION 16.4. If the limit limc→∞ ((X, cdX ), x) exists, we call it the tangent cone (at x ∈ X) and write it as Tx X. If X is an n-dimensional Riemannian manifold then the tangent cone of X is isometric to Rn at each point. E XAMPLE 16.1. Let Ω ⊂ Rn be a compact convex set. We put X = ∂Ω and define a length metric on it. (Namely the distance between x, y ∈ X is the infimum of the length of all curves joining x and y in X.)

Metric Riemannian geometry

241

Then tangent cone Tx X is described as follows. We consider every ray (half of the straight line) : [0, ∞) → Rn such that (t) ∈ Ω for small t > 0. The set of such is an open subset of Rn . Its boundary in Rn is the tangent cone Tx X. If the space X is not so wild then we may expect the tangent cone Tx X exists and a neighborhood of x in X is homeomorphic to a neighborhood of the origin (base point) in Tx X. (This holds for Alexandrov spaces, for example. See Theorem 18.1.) Namely we can study the local structure of X by studying the tangent cone Tx X. If X is a Gromov–Hausdorff limit of a sequence of Riemannian manifolds Mi and if the sectional curvature of Mi is bounded from below by a constant independent of i, then we may regard the limit X as the space with “curvature bounded from below”. Then the infimum of the “curvature” of family of length spaces (X, cdX ) as c goes to infinity will become nonnegative. (Note if we multiply the metric by c then the curvature is multiplied by c−2 .) This means that if tangent cone of X exists, then it is of “nonnegative curvature”. (The discussion here is informal and heuristic. So for a moment the curvature may either to the Ricci or the sectional curvature.) This is one of the reasons why the study of noncompact spaces with nonnegative curvature is important in the local theory of spaces which are a limit of Riemannian manifolds. By using Gromov’s precompactness theorem (Theorem 16.1) we have the following P ROPOSITION 16.2. Let Mi be a sequence of Riemannian manifolds with RicciMi > −(n − 1). Let X = limGH i→∞ Mi . Let x ∈ X and ck be a sequence of positive numbers with lim ck = +∞. Then there exists a subsequence of ((X, ck dX ), x) which converges in pointed Gromov–Hausdorff distance. In general ((X, ck dX ), x) itself may not converge. (Namely we need to take a subsequence.) Hence X may not have a tangent cone. This is one of the difficulties to study the family of Riemannian manifolds with Ricci curvature bounded from below. In case X is a limit of Riemannian manifolds with sectional curvature bounded from below (or more generally if X is an Alexandrov space), limc→∞ ((X, cdX ), x) converges without taking a subsequence (Theorem 17.14). Let us now state the soul theorem and the splitting theorem. We first define the notions line and ray. Let X be a length space. A curve : (a, b) → X is called a geodesic if it is length minimizing locally. Namely is a geodesic if, for each t ∈ (a, b), there exists such that d((t − ), (t + )) is equal to the length of the restriction of to (t − , t + ). We use arc length as a parameter in the next definition. D EFINITION 16.5. Let X be a length space. A geodesic : [0, ∞) → X is called a ray if d((t), (s)) = |t − s| for any t, s. A geodesic : (−∞, ∞) → X is called a line if d((t), (s)) = |t − s| for any t, s. (The difference between line and ray is the domain of its definition.) If there exists a tangent cone Tx X = limc→∞ ((X, cdX ), x) then it is a union of its rays such that (0) is the base point. We also have the following

242

K. Fukaya

L EMMA 16.3. Let X be a length space and : (−, ) → X be a minimal geodesic with (0) = x. If the tangent cone Tx X exists, then it contains a line. In fact, since in (X, cdX ) there exists a minimal geodesic of length c containing the origin, its limit in Tx X will be a line. We assume that a complete metric space X is a length space and satisfies one of the following conditions. C ONDITIONS 16.1. (a) X is a Riemannian manifold of nonnegative sectional curvature. pGH (b) X = limi→∞ Mi such that KMi −i , limi→∞ i = 0 and dim X = dim Mi . (c) X is a Riemannian manifold with nonnegative Ricci curvature. pGH (d) X = limi→∞ Mi such that RicciMi −i , limi→∞ i = 0 and Vol(Mi ) v > 0. The next theorem is called the splitting theorem. T HEOREM 16.4. If X satisfies one of the Conditions 16.1 and contains a ray, then X is isometric to a direct product R × X0 . Theorem 16.4 is due to Toponogov [149] in case (a), to Cheeger–Gromoll [35] in case (c), Grove–Petersen [80] and Yamaguchi [153] in case (b) and Cheeger–Colding [28] in case (d). We will explain an idea of the proof of the cases (a), (c) later in this section. (Case (b) is similar to case (a). Case (d) is discussed in Section 23.) We explain more how to apply it to study the local structure of the limit space. Note that we can use Theorem 16.4 repeatedly. Namely if X0 contains a line then we can again apply the theorem and show that it is a direct product. Therefore if we can repeat it dim X times, then we can prove that X = Rn . Lemma 16.3 implies that if x is an interior point of a minimal geodesic, then Tx X contains a line. Therefore if we can find n (= dim X) “independent” geodesic for which x is an interior point, then the tangent cone Tx X is isometric to Rn . This may imply that X is a manifold in a neighborhood of x. This argument appears in Sections 17, 18 and in Sections 20, 22, 23. We next explain an outline of the proof of splitting theorem. The main tool we use is convexity of Busemann function (it is used also in the proof of the soul theorem). Let X be a length space and : [0, ∞) → X be a ray. D EFINITION 16.6. The Busemann function is the limit b (x) = limt→∞ (t − d(x, (t))). P ROPOSITION 16.5. If X satisfies either (a) or (b), then the Busemann function of its ray is convex. If X satisfies (c), then the Busemann function of its ray is subharmonic. In the situation (d) we cannot define subharmonicity in the usual way. So the argument is more involved. See Section 23 and [26,31].

Metric Riemannian geometry

243

The proof of Proposition 16.5 is by a comparison theorem. Namely it follows immediately from the Laplacian and Hessian comparison theorem (Theorem 16.6) for the distance function. We remark that the Hessian Hess f of a function f on a Riemannian manifold is defined by (Hessx f )(V , W ) = V W (f ) − (∇V W )(f )

(16.1)

and is a symmetric bilinear map Tx M ⊗ Tx M → R. A function f is convex if its Hessian Hess f is nonnegative everywhere. The Laplacian f is its trace. Namely f (x) =

n (Hess f )(ei , ei ),

(16.2)

i=1

where ei is an orthonormal basis of Tp M. (We remark that we are using the nonpositive Laplacian. Namely = −(d ∗ d + dd ∗ ).) We say a smooth function is subharmonic if its Laplacian is nonnegative. T HEOREM 16.6. Let M be a Riemannian manifold and p ∈ M. We consider the function dp (x) = d(p, x). (1) If KM κ then Hessx dp

sκ (d(p, x)) (gx − ddp ⊗ ddp ). sκ (d(p, x))

(16.3)

Here ddp : Tx M → R is the exterior derivative of dp . (2) If RicciM κ then f (x) (n − 1)

sκ (d(p, x)) . sκ (d(p, x))

(16.4)

Here sκ is as in (4.1). Remark 16.1. We remark that dp is not differentiable outside the ball Bp (iM (p), M). So we need to be more careful to state Theorem 16.6. Precisely speaking (16.3), (16.4) hold in a barrier sense. See, for example, [26]. We omit the proof of Theorem 16.6. We remark that (16.6) implies Corollary 5.3. In fact d Vol Bp (t, M) = dt

∂Bp (t,M)

grad dp , grad dp Ω∂Bp (t,M)

=

div grad dp ΩM Bp (t,M)

244

K. Fukaya

(n − 1)

d dt

B0 (t,Tp M)

B0 (t,Tp M)

sκ (d(p, x)) ΩRn sκ (d(p, x))

n sκ d(p, x) ΩRn

d Vol Bp0 t, Snκ . dt

Let us explain how we use Proposition 16.5 to prove Theorem 16.4, in cases (a), (b). We assume that X contains a line : R → X. We then have two rays ± : [0, ∞) → X by + (t) = (t),

− (t) = (−t).

We study their Busemann functions b± . The triangle inequality implies b+ (t) + b− (t) 0.

(16.5)

By Proposition 16.5 the right-hand side is convex. Since a bounded convex function is constant, it follows that + (t) + − (t) is constant. (Actually it is 0.) It follows that + (t) = const − − (t) is convex and is concave. Hence its level surface if totally geodesic. (Here we say S ⊂ M is totally geodesic if any minimal geodesic of M joining two points of S are contained in S.) This implies Theorem 16.4. In case (c) we use subharmonicity in place of convexity. We next discuss the soul theorem. T HEOREM 16.7 (Cheeger–Gromoll [36,128]). If a complete Riemannian manifold M has nonnegative sectional curvature then there exists a compact submanifold S ⊆ M without boundary, such that M is diffeomorphic to the normal bundle of S. Moreover S is totally geodesic. We call S the soul of M. The basis of the proof of Theorem 16.7 is Proposition 16.5. It asserts that, for each ray : [0, ∞) → M, the Busemann function b is convex. In particular for any c the closed set H (, c) = x ∈ M | b (x) c is convex. The next lemma is the key of the proof of Theorem 16.7. We fix p ∈ M and let Ray(p) be the set of all rays of M such that (0) = p. L EMMA 16.8. The set Cc (p) =

.

∈Ray(p) H (, c)

is compact.

The proof is by contradiction. Namely we assume that Cc (p) is not compact and let pi ∈ Cc (p) be a divergent sequence. We put d(p, pi ) = ti , and let i : [0, ti ] → M be a minimal geodesic such that i (0) = p, i (ti ) = pi and that it is parameterized by arc length.

Metric Riemannian geometry

245

i Since d dt (0) ∈ Tp M is a unit vector, we may take subsequence so that it converges. Let : [0, ∞) → M be a geodesic such that

d di (0) = (0). i→∞ dt dt lim

Since limi→∞ ti = ∞, it follows that is a ray. On the other hand, we have lim b (pi ) = ∞.

i→∞

This contradicts to pi ∈ Cc (p).

Thus, we obtained a compact convex subset Cc (p) of M. We can find a compact convex submanifold S in it. The argument to do so is rather technical and is omitted. (See [33, Chapter 8].) Perelman [115] proved that if, in the situation of Theorem 16.7, there exists a point where KM > 0, then the soul S is a one point. We refer to [63] for other topics related to the soul theorem. We remark that we already applied Theorem 16.7 in Section 12 to construct on F -structure. 17. Alexandrov space—I In this section and in the next sections, we discuss recent developments [22,119,113] in the theory of Alexandrov space. A good text book on the contents of this section is [138]. (See also [127].) In the following sections, we study compactly generated length spaces of finite Hausdorff dimension only. So we always assume that the length space has this property. The Alexandrov space is a length space with curvature bounded from below. To define the notion of curvature for length space, we use a Toponogov type comparison theorem in the opposite direction. Namely we define the condition KX 1 by using the conclusion of a comparison theorem. However the conclusion of Theorem 4.7 does not (yet) make sense for length space, since it uses angles. So we consider the following slightly different version. We use the notation of Theorem 4.7. Let M be a Riemannian manifold and x, y, z, v, w ∈ M. Let x , y , z , v , w ∈ Sn (κ). We assume v ∈ xy, w ∈ xz, v ∈ x y , w ∈ x z . T HEOREM 17.1 (Alexandrov–Toponogov). We assume KM κ and d(x, y) = d(x , y ), d(x, z) = d(x , z ), d(y, z) = d(y , z ), d(x, v) = d(x , v ), d(x, w) = d(x , w ). u ∈ xy, u ∈ x y , v ∈ xz, v ∈ x y . Then we have d(v, w) d(v , w ). (See Figure 17.1.) D EFINITION 17.1 (Alexandrov). A length space of finite dimension is said to be an Alexandrov space with K κ if the conclusion of Theorem 17.1 holds for X. Remark 17.1. There are several other definitions equivalent to Definition 17.1. We will explain them later (Theorems 17.9 and 17.10).

246

K. Fukaya

Fig. 17.1.

Remark 17.2. There is a notion of Alexandrov space with curvature bounded from above. We do not discuss it in this article. It is proved by Berestovskij that if a length space is an Alexandrov space with curvature bounded from above and below then it is a C 0 Riemannian manifold. This result is related to Theorem 3.4 but was proved earlier than that. See [17]. Hereafter we say Alexandrov space for Alexandrov space with K κ with some κ. The notion of Alexandrov space was introduced by Alexandrov [6] more than 50 years ago. There are several related pioneering works around those old days, like Busemann [23]. In [22], Burago–Gromov–Perelman proved several fundamental theorems on Alexandrov spaces. After that the study of Alexandrov space became very active and important in metric Riemannian geometry. Their main results are T HEOREM 17.2 (Burago–Gromov–Perelman [22]). Let X be an Alexandrov space. Then there exists a dense open subset X0 such that, for each p ∈ X0 , there exists a neighborhood Up and a Lipschitz homeomorphism Up → Vp where Vp ⊂ Rn is an open set. T HEOREM 17.3 (Burago–Gromov–Perelman). The Hausdorff dimension of an Alexandrov space is an integer and is equal to its topological dimension. Remark 17.3. There are several ways to define topological dimension, that is covering dimension (big and small), inductive dimension, etc. Theorem 17.3 also implies that they coincide for Alexandrov spaces. We do not discuss the proof of Theorem 17.3. (It will follow from Corollary 18.3 in the next section.) Before explaining some of the ideas of the proof of Theorems 17.2, we give some examples of Alexandrov spaces. E XAMPLE 17.1. (0) A Riemannian manifold (M, g) is an Alexandrov space with K κ, if and only if the sectional curvature of (M, g) is greater than κ everywhere. (1) Let Ω ⊆ Rn be a compact and convex domain. Let S = ∂Ω. We define the length metric d on S. Namely, the distance between x, y ∈ S is the minimum of the length of the

Metric Riemannian geometry

247

curves in S joining x with y. Then we can prove that (S, d) is an Alexandrov space of curvature 0. (2) Let M be a Riemannian manifold with KM κ and G be a compact group acting on M by isometry. Then the quotient space M/G, equipped with the quotient metric is an Alexandrov space. An important example of an Alexandrov space is a Gromov–Hausdorff limit of an Riemannian manifold. Actually we have P ROPOSITION 17.4. Let Xi be a sequence of compact length spaces and X = limGH i→∞ Xi . If Xi are Alexandrov spaces with K κ, then so is X. (Here κ is independent of i.) The proof is elementary. Remark 17.4. Yamaguchi [153] proved that if M a C ∞ -manifold and G is a compact Lie group acting smoothly on M, then there exists a sequence of metrics gi on M such that Kgi κ for some κ independent and (M, gi ) converges to M/G. Another source of examples is a cone, which we define below. D EFINITION 17.2. Let (Y, d) be a metric space. We consider the product T × [0, ∞) and identify (x, 0) and (y, 0). We thus obtain a space CY . We define a cone metric on it as follows: d (x, t), (y, s) = t 2 + s 2 − 2st cos d(x, y). We denote by o ∈ CY the equivalence class of (x, 0). E XAMPLE 17.2. If Y = S n with KS n ≡ 1, then CS n is isometric to Rn+1 . L EMMA 17.5. If Y is a length space and Diam(Y ) π , then CY is a length space. We can prove an analogue of Myers’ theorem (Theorem 5.4) for Alexandrov spaces. Namely T HEOREM 17.6 [22]. If M is an Alexandrov space with K 1, then Diam(Y ) π . T HEOREM 17.7 [22]. (1) If CY is an Alexandrov space, then Y is an Alexandrov space with K 1. (2) If dim Y > 1 and Y is an Alexandrov space with K 1 then CY is an Alexandrov space with K 0. (3) In case dim Y = 1, the cone CY is an Alexandrov space with K 0 if and only if Diam(Y ) π .

248

K. Fukaya

Fig. 17.2.

We do not discuss the proof. We next discuss an example of a length space which is not an Alexandrov space. E XAMPLE 17.3. Let us consider a simplicial complex X consisting of three arcs which are joined at one point o. (See Figure 17.2.) We can define a metric on it such that the length of each arc is 1. Let x, y, z be interior points of each of the three simplexes, respectively. We can choose v = w on xy ∩ xz = xo. Then d(v, w) = 0. But if we choose x , y , z , v , w as in Theorem 17.1 then d(v , w ) > 0. (For any κ.) So the conclusion of Theorem 17.1 does not hold. Namely X is not an Alexandrov space. The argument of Example 17.3 implies the following. We call a map : (a, b) → X a geodesic if for each c ∈ (a, b) there exists such that the length of the restriction of to (c − , c + ) is d((c − ), (c + )). L EMMA 17.8. If 1 , 2 are geodesics on an Alexandrov space X and if they coincide on an open set, then their union is also a geodesic. In other words, geodesics can never branch. We next explain some other equivalent definitions of Alexandrov spaces. T HEOREM 17.9. Let X be a length space. We assume that for each p ∈ X there exists a neighborhood U such that the conclusion of Theorem 17.1 holds for any x, y, z, u, v ∈ U . Then X is an Alexandrov space with K κ. In other words, the same conclusion holds globally. In fact, usually the assumption of Theorem 17.9 is the definition of Alexandrov space. We discuss another equivalent definition. Let X be a length √ space and x, y, z ∈ X. Let κ ∈ R. In case κ > 0, we assume d(x, y), d(x, z), d(y, z) < π/ κ. We choose x , y , z ∈ Sn (κ) such that d(x, y) = d(x , y ), d(y, z) = d(y , z ), d(x, z) = d(x , z ). We define κ yxz =

y x z .

Metric Riemannian geometry

249

T HEOREM 17.10. Let X be a length space. (1) If, for each p ∈ X, there exists a neighborhood U of p such that κ bac + κ cac + κ cab

2π

for and a, b, c, d ∈ U , then X is an Alexandrov space with K κ. (2) Let X be an Alexandrov space with K κ and let a, b, c, d ∈ X. Then κ bac + κ cac + κ cab

Remark 17.5. By Theorem 17.6

2π.

κ bac

etc. in (2) is well defined.

The idea that if the comparison theorem holds locally, then it holds globally is due to Alexandrov and Toponogov. Theorem 17.10 is proved in [22]. We next discuss the angle between geodesics. Hereafter we assume that geodesics are parameterized by arc length. Let X be an Alexandrov space with K κ and 1 , 2 : [0, c) → X be geodesics such that p = 1 (0) = 2 (0). L EMMA 17.11. If s1 t1 , s2 t2 then κ 1 (s1 )p2 (s2 ) κ 1 (t1 )p2 (t2 ).

This follows easily from the definition. Therefore we can define D EFINITION 17.3. 1 2 = limt1 ,t2 →0 κ 1 (t1 )p2 (t2 ). In case 1 , 2 are minimal geodesics joining p to x, y, respectively, we write xpy = 1 2 . Remark 17.6. (1) The angle xpy is independent of κ. (2) Two geodesics 1 , 2 coincide to each other if 1 2 = 0. T HEOREM 17.12. If X is an Alexandrov space of K κ and x, y, z ∈ X, then we have d(y, z) s(d(x, y), d(x, z), yxz, κ). Here s is defined in (4.4). In other words, Theorem 4.7 holds for an Alexandrov space. The other version of the triangle comparison theorem also holds.

T HEOREM 17.13. If X is an Alexandrov space with K κ and x, y, z ∈ X, then we have yxz κ yxz.

Remark 17.7. The Toponogov type comparison theorem holds in Alexandrov space. Hence we can generalize the argument of the last section to prove the splitting theorem (Theorem 16.4) for an Alexandrov space with K 0. As we mentioned before an Alexandrov space has a tangent cone.

250

K. Fukaya

T HEOREM 17.14 [22]. If (X, d) is an Alexandrov space with K κ and x ∈ X, then limk→∞ ((X, kd), x) converges with respect to the pointed Gromov–Hausdorff distance. The limit in Theorem 17.14 is the tangent cone Tx X. The tangent cone is related to the angle between geodesics as follows. D EFINITION 17.4. Let Σ˜ x0 be the set of all geodesics (parameterized by arc length) : [0, c) → X for some c such that (0) = x. We identify 1 and 2 if they coincide on a neighborhood of 0. We denote by Σx0 the set of this equivalence relation. We can easily show that the angle defines a metric on it. We define the space of directions Σx (X) as the completion of Σx0 . L EMMA 17.15. If X is an Alexandrov space, then Σx (X) is an Alexandrov space with K 1 and Tx X is an Alexandrov space with K 0. T HEOREM 17.16 [22]. The tangent cone Tx X of an Alexandrov space X is isometric to the cone CΣx (X). If dim X = n, then dim Σx (X) = n − 1 and dim Tx X = n. We remark that the second half of Theorem 17.16 is a consequence of Proposition 17.7. Now we start the discussion of the proof of Theorem 17.1. As we mentioned in the last section, if x ∈ X is an interior point of n = dim X “independent” minimal geodesics, then Tx X is isometric to Rn , and this implies that x has neighborhood, homeomorphic to Rn . However the condition about the existence of a geodesic is a bit too strict. So we relax it a bit. Let X be an Alexandrov space with K κ. D EFINITION 17.5. Let x ∈ X and (ai , bi ) ∈ X 2 , i = 1, . . . , n. We say that {(ai , bi )}i=1,2,...,n is an (n, δ)-strainer at x, if κ ai xbi

π − δ,

and κ ai xaj , κ ai xbj , κ bi xbj

δ,

for i = j .

A point x ∈ X is said to be (n, δ)-strained if there exists an (n, δ) strainer at x. (See Figure 17.3.) Remark 17.8. In [22] the strainer is called “explosion” and a strained point is called “burst point”. The name strainer and strained point seems to be more popular now. The main step of the proof of Theorem 17.2 is the following P ROPOSITION 17.17 ([22, Theorem 9.4] or [138, Theorem 7.4]). If p ∈ X is an (n, δ)strained point, then there exists ρ > 0, neighborhoods U ⊂ V of p, and a map ϕ : V → Rn with ϕ(p) = 0 with the following properties:

Metric Riemannian geometry

251

Fig. 17.3.

(1) d(ϕ(x), ϕ(y)) < 2d(x, y). (2) Let x ∈ U and X ∈ Rn , with d(ϕ(x), X) < ρ. Then there exists y ∈ V such that ϕ(y) = X and d(x, y) Cd(ϕ(x), X), where C depends only on n and δ. We remark that (2) implies that ϕ is an open mapping in a neighborhood of x. Hence if ϕ is injective then ϕ gives a chart in a neighborhood of x. We can use the following to show ϕ is injective. L EMMA 17.18. We may choose U small enough so that if ϕ is not injective then there exists an (n + 1, 10δ)-strained point on a small neighborhood of U . We remark that the set of all the (n, δ)-strained points is open. On the other hand Proposition 17.17 implies that if a (n, δ)-strained point exists, then the Hausdorff dimension is not smaller than n. Hence Proposition 17.17 and Lemma 17.18 imply the following. For each open set U , we can find n and a nonempty open subset U0 ⊂ U consisting of (n, δ)-strained points such that there are no (n + 1, 10δ)-strained points on U . Then U0 is an n-dimensional manifold by Proposition 17.17 and Lemma 17.18. The proof of Theorem 17.2 then will be completed by using the next lemma. L EMMA 17.19 [22, Corollary 6.5]. We assume X is connected. If U, V are nonempty open subsets of X, then the Hausdorff dimension of U is equal to the Hausdorff dimension of V .

252

K. Fukaya

We now sketch the proof of Proposition 17.17 and Lemmas 17.18, 17.19. We put µ = inf d(p, a1 ), . . . , d(p, an ), d(p, b1 ), . . . , d(p, bn ) . We first explain the idea of the proof of Proposition 17.17. We put ϕ(x) = − d(x, a1 ), . . . , d(x, an ) + d(p, a1 ), . . . , d(p, an ) . It is easy to see that (1) is satisfied. We show (2). For simplicity, we consider the case x = p, n = 2. For each X = (X1 , X2 ) ∈ B0 (ρ, R2 ), we will find w with ϕ(w) = (X1 , X2 ), d(p, w) Cd(0, X). We assume X1 , X2 > 0. We first take the point q1 ∈ pa1 such that d(p, q1 ) = X1 . We first show |ϕ(q1 ) − (X1 , 0)| τ (ρ, δ|n, κ, µ). d(0, X)

(17.1)

In fact we can prove d(q1 , a2 ) d(p, a2 ) − τ (ρ, δ|n, κ, µ)X1 by applying Theorem 17.1, where we put x = a1 , y = p, z = v = a2 , u = q1 . To prove the opposite inequality we take the point p ∈ b1 q1 such that d(p , q1 ) = X1 . We have d(p, p ) X1 τ (ρ, δ|n, κ, µ). In fact, since κ b1 pa1 > π − δ, it follows that b1 q1 a1 κ b1 q1 a1 > π − δ − τ (ρ|n, κ). Hence pq1 b1 < δ + τ (ρ|n, κ). Theorem 17.12 then implies d(p, p ) X1 τ (ρ, δ|n, κ, µ). We use d(p, p ) X1 τ (ρ, δ|n, κ, µ) to show | b1 q1 a2 − π/2|, | b1 p a2 − π/2| < τ (ρ, δ|n, κ, µ).

(17.2)

We next apply Theorem 17.1 again by putting x = q1 , y = b1 , z = v = a2 , u = p . Then using (17.2) have d(p , a2 ) d(q1 , a2 ) − τ (ρ, δ|n, κ, µ)X1 . Hence d(p, a2 ) d(q1 , a2 ) − τ (ρ, δ|n, κ, µ)X1 . We have proved (17.1). We next take w1 ∈ a2 q1 such that d(w1 , q1 ) = X2 . Then we have |ϕ(w1 ) − (X1 , X2 )| τ (ρ, δ|n, κ, µ). d(0, X) We repeat the process replacing p by w1 and obtain w2 such that d(w1 , w2 ) < C|ϕ(w1 ) − (X1 , X2 )| and |ϕ(w2 ) − (X1 , X2 )| τ (ρ, δ|n, κ, µ). |ϕ(w1 ) − (X1 , X2 )| We can define w3 , . . . in a similar way. wi is a Cauchy sequence whose limit w has the required property.

Metric Riemannian geometry

253

Let us prove Lemma 17.18. Let ϕ(x) = ϕ(y). Let z ∈ xy with d(x, z) = d(x, y). It is easy to see that (ai , bi ), i = 1, . . . , n, and (x, y) is an (n + 1, 2δ)-strainer if d(x, y) is small. Finally we sketch the proof of Lemma 17.19. We may assume X is compact. Take p ∈ V and put D = sup{d(p, x) | x ∈ U }. We take R such that Bp (D/R, X) ⊂ V . We define Φ : U → V as follows. For x ∈ V we take a point Φ(x) ∈ px such that Rd(p, Φ(x)) = d(p, x). (Note the minimal geodesic px may not be unique. So we need some technical argument to find Φ which is measurable.) Definition 17.1 implies that there exists ρ > 0 such that d(Φ(x), Φ(y)) ρd(x, y). It follows that the Hausdorff dimension of Φ(U ) is not smaller than the Hausdorff dimension of U . Therefore the Hausdorff dimension of V is not smaller than the Hausdorff dimension of U . We can prove the opposite inequality in the same way. We thus finished a sketch of the proof of Theorem 17.2.

D EFINITION 17.6. We define the boundary ∂X of an Alexandrov space X by induction on dim X as follows. If dim X = 1, then X is either an arc or a circle. So we can define its boundary in an obvious way. Suppose ∂X is defined for X with dim X < k. Let X be an Alexandrov space of dim X = k. Then we say x ∈ ∂X if ∂Σx (X) = ∅. (We remark that Σx (X) is an Alexandrov space and dim Σx (X) = k − 1.) Theorem 17.2 is improved by Otsu–Shioya [112]. To state their results we define the notion of singular point set in an Alexandrov space more precisely. D EFINITION 17.7. Let X be an n-dimensional Alexandrov space and δ > 0. We put Sδ (X) = x ∈ X | Vol Σx (X) Vol S n−1 − δ , S(X) = Sδ . δ>0

We remark that the Alexandrov space version of the following theorem is a motivation of Definition 17.7. T HEOREM 17.20 (Otsu–Shiohama–Yamaguchi [111]). If an n-dimensional Riemannian manifold M satisfies Vol(M) Vol(S n ) − n , KM 1 then M is diffeomorphic to the sphere. Also M is close to S n with respect to the Hausdorff distance.27 We discuss the idea of the proof of Theorem 17.20 in Section 21. 27 This theorem is improved later to Theorem 21.7 and to Corollary 22.4. Before [111], Shiohama [137] proved that M is homeomorphic to the sphere under a similar but different assumption KM −C, Ricci (n − 1), Vol(M) Vol(S n ) − (n, C).

254

K. Fukaya

T HEOREM 17.21 (Burago–Gromov–Perelman, Otsu–Shioya). Let X be an Alexandrov space of dimension n. Then the Hausdorff dimension of S(X) is not greater than n − 1. The Hausdorff dimension of S(X) \ ∂X is not greater than n − 2. T HEOREM 17.22 (Otsu–Shioya [111]). There exists a C 0 -Riemannian metric on X \ S(X) which induces the metric on X. Moreover there exists X0 ⊂ X \ S(X) such that the (n-dimensional Hausdorff ) measure of X \ X0 is 0 and that there exists manifold structure of C 1.5 -class and a Riemannian structure is of C 0.5 -class on X0 . Remark 17.9. Actually we need to define a C 1.5 -structure etc. in the above theorem. This is because X \ S(X), X0 are not open subsets in general. Hence they are not manifolds. See [112] for the precise statement. Theorem 17.22 is used by Kuwae–Machigashira–Shioya [99] to develop analysis on Alexandrov spaces. We also remark the following T HEOREM 17.23 (Fukaya–Yamaguchi [60]). The isometry group of an Alexandrov space is a Lie group.

18. Alexandrov space—II In [119,113] Perelman proved the following two fundamental results on Alexandrov spaces. T HEOREM 18.1 (Perelman). Let X be an Alexandrov space with K κ. Then, for any x ∈ X, there exists a neighborhood of x homeomorphic to Tx X, the tangent cone. T HEOREM 18.2 (Perelman). Let Xi be a sequence of Alexandrov spaces with K κ where κ is independent of i. We assume X = limGH i→∞ Xi and dim X = dim Xi . Then Xi is homeomorphic to X for large i. Remark 18.1. Both of these theorems are proved in [119]. Later Perelman published another paper [113] where the proof of Theorem 18.1 is given in a simplified way. Perelman says in [113] that a similar method gives a slight simplification of the proof of Theorem 18.2, but the simplification is not so big compared with the one for Theorem 18.1. Unfortunately the paper [119] is not yet published. In fact Theorem 18.1 follows from Theorem 18.2 (and Theorems 17.14, 17.16). However the proof of Theorem 18.2 requires Theorem 18.1. In this section we give a review of the proof of Theorem 18.1. Before that let us mention some of the corollaries of them. We remark that Tx X is homeomorphic to CΣx (X) by Theorem 17.16. Since Σx (X) is again an Alexandrov space, we can apply Theorem 18.1 again. We then find that the

Metric Riemannian geometry

255

singularity of X is obtained locally by taking cones several times. Let us define it more precisely. D EFINITION 18.1. We define a connected metrizable space X to be an MCS-space of dimension n inductively on n as follows. (1) An MCS-space of dimension 2 is a 2-dimensional manifold with or without boundary. (2) X is an MCS-space of dimension n if, for each x ∈ X, there exists a neighborhood U of x and an MCS-space Yx of dimension n − 1, such that there exists a homeomorphism F from the cone of Yx to U such that F sends the cone point to x. The following is immediate from Theorem 18.1. C OROLLARY 18.3. Every Alexandrov space is an MCS-space. The following is also an immediate corollary. C OROLLARY 18.4. For an Alexandrov space X, there $ exists Xk with Xk is a k-dimensional topological manifold and X¯ k = ik Xi .

$

Xk = X such that

C OROLLARY 18.5. An Alexandrov space X is locally contractible. If it is compact then π1 (X) and Hk (X) are finitely generated. Hereafter we assume our Alexandrov space X has no boundary, for simplicity.28 An idea used in [113] to prove these result is to generalize Morse theory of the distance function to an Alexandrov space. Let us give the following definition. Hereafter X is an Alexandrov space with K −1. Let p ∈ X. We put dp (x) = d(x, p). D EFINITION 18.2. x is said to be a regular point of dp if there exists ξ ∈ Σx (X) such that for each minimal geodesic joining x to p we have ξ > π/2. Here ∈ Σx (X) is the equivalence class of in Σx (X). Definition 18.2 is a generalization of Definition 14.1. We can generalize Proposition 14.2 also and we further generalize it to Theorem 18.7. For the proof of Theorem 18.1 we need to use a bit more general function than the distance function and define the “regularity” of a map X → Rk for k > 1 also. To state this generalization we need some notations. D EFINITION 18.3. Let U be an open subset of X. (1) An admissible function f : U → R is a function of the form f (x) =

m

ai φi d(Ai , x) ,

(18.1)

i=1 28 The general case can be handled by taking a double X ∪ ∂X X which is an Alexandrov space by [124].

256

K. Fukaya

where A i is a compact subset of X, φi are smooth functions with 0 φi 1 and ai 0, ai 1. (2) An admissible map F : U → Rk is a composition F = G ◦ f., where G is a biLipschitz homeomorphism and f. = (f 1 , . . . , f k ) with admissible functions fi .

Remark 18.2. In [113] a more general function (map) is called an admissible function (map). But only those in Definition 18.3 are used. For A ⊂ X and x ∈ X we define Σx0 (xA) ⊂ Σx (X) by Σx0 (xA) =

[] | is a minimal geodesic joining x to a point of A .

Let Σx (xA) ⊂ Σx (X) be the closure of Σx0 (xA). For Λ1 , Λ2 ⊂ Σx (X) we put Λ1 Λ2 = inf{ uv | u ∈ Λ1 , v ∈ Λ2 }.

For an admissible function f as in (18.1) we can define its direction derivative Dx f : Σx (X) → R by (Dx f )(u) =

ai φ d(x, Ai ) cos u, Σx (xAi ) .

i

In case X is a manifold Dx f is the direction derivative in the usual sense. If f (1) , f (2) are admissible functions as in (18.1) we put Dx f (1) , Dx f (2) (1) (2) (1) (2) (2) (1) (2) φ d x, Aj cos Ai Aj . ai aj φ (1) d x, Ai =

i,j

This again coincides with the usual inner product between derivatives in case when X is a manifold and f (1) , f (2) are differentiable. D EFINITION 18.4. We say F : U → Rk where F = G ◦ f. is -regular at p ∈ U if the following conditions hold. Let us put f. = (f 1 , . . . , f k ). (1) For each i = j , we have Dp f i , Dp f j < −. (2) There exists ξ ∈ Σp (X) such that (Dp f i )(ξ ) < −, for each i. We say F is regular if it is -regular for some > 0. We say F is -regular on U if it is -regular at every point of U . E XAMPLE 18.1. (1) If f : U → R is defined by f (x) = dp (x). It is an admissible function and hence is an admissible map. It is -regular at x for some > 0 if and only if x is a regular point of f in the sense of Definition 18.2.

Metric Riemannian geometry

257

Fig. 18.1.

(2) Let X be a two-dimensional Alexandrov space and (a1 , b1 ), (a2 , b2 ) be a (2, δ)strainer at x. Let us define ϕ : X → R2 by ϕ(x) = − d(x, a1 ), d(x, a2 ) + d(p, a1 ), (.p, a2 ) , as in the proof of Proposition 17.17. Then ϕ is a homeomorphism in a neighborhood of x. We put √ √ p = ϕ −1 (r, 0), q = ϕ −1 (−r/2, r 2/2), r = ϕ −1 (−r/2, −r 2/2) and define F = f. = (dp , dq ). We also set ξ ∈ Σx (xr). We can prove (1), (2) for sufficiently small . We can generalize this construction to the case of higher dimension and prove that if x is a (k, δ)-strained point, then there exists F : U → Rk from a neighborhood of x which is -regular at x. (See Figure 18.1.) We can prove the following in a way similar to the proof of Proposition 17.17. (See [116, Lemma 2.3 and the argument just after that].) L EMMA 18.6. Let F : Bx (ρ, X) → Rn be an admissible map which is -regular at x. Then there exists a neighborhood U ⊆ Bx (ρ, X) of x and δ > 0, with the following property. If y ∈ U, X ∈ Rk with d(F (v), X) δ, then there exists z ∈ Bx (ρ, X) such that F (z) = X and d(z, y) < Cd(F (v), X). Here C depends only on ρ, δ, . Lemma 18.6 implies that F is an open mapping. In case dim X = k and if there exists an -regular map at x, then Lemma 18.6 shows that a neighborhood of x is a manifold. In the general case, we have to study the situation where k < dim X. The following Proposition 18.7 is the main result in such a case. We need a definition. D EFINITION 18.5. A map F : X → Y between topological spaces is called a topological submersion at x ∈ X if there exists a neighborhood U of x, a neighborhood V of F (x), and a topological space W such that there exists a homeomorphism Φ : U ∼ = V × W satisfying F = P r1 ◦ Φ on U .

258

K. Fukaya

In case X, Y are smooth manifolds and F is a smooth map, F is a topological submersion if its derivative is of maximal rank. P ROPOSITION 18.7 [113, Theorem 1.4]. An admissible map F : X → Rk is a topological submersion at a regular point. The following result is also used in the proof of Theorem 18.1. T HEOREM 18.8 (Siebenmann [142, Corollary 6.14]). Every proper topological submersion between MCS-spaces is a locally trivial fiber bundle. Remark 18.3. (1) We remark that if M, N are smooth manifolds (without boundary) and if F : M → N is a proper smooth submersion then F is a locally trivial fiber bundle. This fact can be proved much more easily than Theorem 18.8. (2) The proof of Theorem 18.8 is based on the isotopy extension theory. We remark that isotopy extension theory for manifolds (see [48]) was used by Cheeger for the proof of his finiteness theorem. (See Section 6.) We next sketch the proof of Proposition 18.7. The difficult case is when X is of dimension greater than k. We try to increase k as much as possible, we then arrive at the following situation. D EFINITION 18.6. Let F : U → Rk be a regular admissible map from an open set U of an Alexandrov space X. We say p ∈ X is imcomplementable if there exists no g such that (f 1 , . . . , f k , g) is regular at p. The case k = 0 is included. Namely in that case p ∈ X is imcomplementable if there exists no admissible function such that p is regular. E XAMPLE 18.2. (1) Let us consider the domain {(r cos θ, r sin θ ) | θ ∈ [−α, +α], r 0}. We glue (r cos α, r sin α) and (r cos −α, r sin −α) to obtain a space Xα . We can show that o = [0, 0] is imcomplementable if and only if α π/2. Actually we put g = d[r,0] . Then g is regular if α > π/2. On the other hand, if α π/2 then the diameter of Σo Xα is not greater than π/2. Hence it is easy to see that (2) in Definition 18.4 can never be satisfied.29 (See Figure 18.2.)

Fig. 18.2. 29 It is easy to see from this argument that in case k = 0 the point p ∈ X is imcomplementable if and only if

Diam Σp (X) π/2.

Metric Riemannian geometry

259

Fig. 18.3.

(2) Let us next take X = Xα × R, where Xα is as above. We define f : X → R by f = d(o,−1) . It is easy to see that (o, 0) is a regular point. Actually we may take ξ = D(o,0) d(o,1) . We can show that f is imcomplementable at (o, 0) if α π/2. (See Figure 18.3.) Now the main technical result in [113] is as follows. L EMMA 18.9 [113, 1.3]. If F : U → Rk is admissible and regular at p ∈ U , and if p is imcomplementable, then there exists an admissible function g : V → R defined on an open neighborhood V of p with the following properties. We write F = G ◦ f., f. = (f 1 , . . . , f k ). (1) g 0 on V and g(p) = 0. (2) F |g −1 (0) : g −1 (0) → Rk defines a homeomorphism onto a neighborhood of F (p). (3) (F, g) : V → Rk+1 is regular on V \ g −1 (0). (4) There exists ρ > 0 such that {x ∈ V | d(F (x), F (p)) ρ, g(x) −ρ} is compact. Let us show how to choose such g in the case of Example 18.2(2). Namely we have U = Xα × R and F = f = d(o,−1) . We write a point of U as ([r cos θ, # r sin gθ ], t) and use r, θ, t as a coordinate. (We take r 0, θ ∈ [−α, α].)#Then f (r, θ, t) = (t + 1)2 + r 2 . We take q = (δ, 0, δ) and put h = dq . Then h(r, θ, t) = (t − δ)2 + r 2 + δ 2 − 2rδ cos |θ |. It is easy to see that (f, h) : U → R2 is regular outside on (B(o,0) (ρ, U ) \ {o}) × R. (We remark {o} × R is the set of singular points.) However if we put g = h then (1), (2) are not satisfied. So we compose it with a homeomorphism of R2 so that (1), (2) will be satisfied. (See Figure 18.4.)

260

K. Fukaya

Fig. 18.4.

We consider the set K(v) = {x ∈ U | f (x) = v, r < ρ} where |ρ| and |v| is small. We can easily check that h(v, 0, 0) = sup h(x) | x ∈ K(v) if α π/2.30 Namely (∗) The restriction of h to K(v) takes its maximum# at a unique point. We remark that f (v, 0, 0) = 1 + v and h(v, 0, 0) = (v − δ)2 + δ 2 . So if we put g(r, v, θ ) = h(r, v, θ ) −

2 f (r, v, θ ) − 1 − δ + δ 2 ,

then (1), (2) are satisfied. We define G : B(1,√2δ) (ρ, R2 ) → R2 by # G(a, b) = a, b − (a − 1 − δ)2 + δ 2 , where ρ - δ. Since (f, g) = G ◦ (f, h), it follows that (f, g) is admissible. It also satisfies (3) since (f, h) satisfies (3). Thus we constructed g in the case of Example 18.2(2). In the general case, we need to choose h more carefully so that it is enough “concave”. (Then (∗) holds.) The proof of Lemma 18.9 is in [113, Section 3]. We can use Lemma 18.9 to complete the proof of Proposition 18.7 as follows. We also prove the following at the same time. 30 This condition is equivalent to the condition that (o, 0) is imcomplementable.

Metric Riemannian geometry

261

P ROPOSITION 18.10. If F : X → Rk is an admissible map and if p ∈ X is a regular point, then F −1 (F (p)) is an MCS-space near p. The proof is by downward induction on k. If k = dim X then both Theorem 18.7 and Proposition 18.10 follow from Lemma 18.6. Let us assume that Theorem 18.7 and Proposition 18.10 are true for k + 1 and prove the case k. We remark that both propositions are local statements on p.31 In case p is not imcomplementable, then we can increase k and use the induction hypothesis. So it suffices to consider the case that p is imcomplementable. We apply Lemma 18.9 to get g. Then (F, g)|V \g −1 (0) : V \ g −1 (0) → Rk+1 is regular. We can use the induction hypothesis to conclude that it is a topological submersion and the fibers are MCS-spaces. Therefore V \ g −1 (0) is an MCS-space. Let (F, g)−1 (B0 (ρ, Rk ) × (−ρ, 0)) = W . Since F : W → B0 (ρ, Rk ) × (−ρ, 0)

(18.2)

is proper, Theorem 18.8 implies that (18.2) is a locally trivial fiber bundle. Since the base space is trivial it follows that (18.2) is a trivial bundle. Hence using Lemma 18.9(2), we can prove Propositions 18.7 and 18.10 for F : U → Rk . Thus the induction works. We remark that Proposition 18.10 implies Theorem 18.1 by putting k = 0. We thus sketched the proof of Theorem 18.1. The proof of Theorem 18.2 uses a similar argument but more involved. See [119]. Let us compare the results we reviewed in the last and this section so far to the one in earlier sections, where we consider the case |KM | is bounded. In Section 11, we asked two questions, Problem 11.1 for a sequence Mi converging to X. (1) was on the singularity of X and (2) was on the relation between topologies of Mi and X. In the case |KMi | 1, an answer to (1) was Theorem 11.4 and an answer to (2) was Theorem 11.5 and 12.7. In our more general case where we assume KMi −1 only, Theorem 18.1 and Corollary 18.3 give a satisfactory answer to (1). However, results on (2) are not satisfactory. In case X = limGH i→∞ Mi satisfies dim X = dim Mi , Theorem 18.2 is a satisfactory answer. This is the noncollapsing case. On the other hand if X is a smooth Riemannian manifold, Theorem 11.3 by Yamaguchi, gives a nice answer. Namely there exists a fiber bundle Mi → X for large i.32 However, the trick (taking frame bundles) we explained in Section 11, does not work in our more general situation to reduce the problem to the case when X is a manifold. So the result is not yet satisfactory. There are however several interesting approaches and partial results about the problem (2) in the case Mi −1, which we review very briefly here. First, Theorem 11.3 is generalized to the case when the limit X has a rather mild singularity. There are two papers about it. In [154], Yamaguchi assumed that for each x ∈ X, there exists a strainer (ai , bi ), i = 1, . . . , n = dim X, with d(x, ai ), d(x, bi ) > µ > 0 where µ is independent of x. Then he concludes that there exists a locally trivial Lipschitz fiber bundle structure Mi → X.33 31 So we prove them by induction without assuming completeness of X. 32 Theorem 11.3 does not say much about the fibers. But there are various results which shows that the fibers are

“of nonnegative curvature” in some sense. 33 In the preprint version of [154] (which the author has), the locally triviality is not asserted. It is proved in [141].

262

K. Fukaya

Perelman in [117] assumed that X has no proper extremal set. Here D EFINITION 18.7. F ⊂ X is extremal if for each p ∈ / F , x ∈ X, and u ∈ Σx (X), we have Dx dp (u) 0. For example, F = {x} consisting of one point is not extremal if and only if there exists an admissible function f which is regular at x. Perelman’s theorem in [117] is that, if there is no extremal set, then there exists fi : Mi → X such that πk (Mi , fi−1 (p)) ∼ = πk (X) for each p ∈ X. The plan proposed by Perelman [116] then is to stratify X using an extremal set and to construct a fiber bundle structure stratawise. This plan is not yet completed. Shioya–Yamaguchi [141] and Yamaguchi [155,156] studied the case when dim Mi = 3, 4 without extra assumption on X and gave a satisfactory description in that case. In this article, we discuss the 3-dimensional case only. Let Mi be 3-dimensional Riemannian manifold with KMi −1, and X = limGH i→∞ Mi . We assume dim X 2. Then X is homeomorphic to a manifold with or without boundary. We assume that X is connected. T HEOREM 18.11 (Shioya–Yamaguchi [141]). We assume dim X = 2. (1) If ∂X = ∅, then there exists a structure of Seifert fibered space Mi → X for large i. (2) If ∂X = ∅, then Mi is homeomorphic to Seii (X) ∪ (∂X × D 2 ) where Seii (X) is a Seifert fibered space over Int X. We glue it with ∂X × D 2 where the fibers of Seii (X) over the boundary point x are glued with {x} × ∂D 2 . In case dim X = 1 there are two possibilities, X ∼ = S 1 or [0, 1]. In the case X = S 1 there 1 exists a fiber bundle Mi → S by Theorem 11.3. T HEOREM 18.12 (Shioya–Yamaguchi [141]). If X ∼ = [0, 1] then Mi is obtained by gluing Bi and Ci along their boundaries where each of Bi , Ci is homeomorphic to one of the following 4-manifolds. (1) D 3 , (2) A nontrivial [0, 1]-bundle over RP 2 , (3) S 1 × D 2 , (4) A nontrivial [0, 1]-bundle over the Klein bottle. The rough idea of the proofs of Theorems 18.11, 18.12 are as follows. In either cases, we can apply a generalization [154] of Theorem 11.3 except in finitely many points (plus ∂X in case (2) of Theorem 18.11). In the neighborhood of those points we scale the metric to obtain a noncompact nonpositively curved Alexandrov space. Then applying the soul theorem (the Alexandrov space analogue of Theorem 16.7). The soul S is an Alexandrov space of dimension 2, so it is a manifold with or without boundary. Actually Shioya– Yamaguchi classified 3-dimensional noncompact complete Alexandrov spaces with K 0. In this way, we can classify neighborhoods ⊂ Mi of a singular point of X locally. Then the last step is to glue those local neighborhoods. In the case when dim X = 0, we can scale the metric of Mi and obtain a limit of nonzero dimension. In this way [141] (improving [153,59]) proved the following

Metric Riemannian geometry

263

T HEOREM 18.13 (Shioya–Yamaguchi). There exists such that if M is a Riemannian 3-manifold with KM Diam(M) −, then a finite cover of M is homeomorphic to S 1 × S 2 , T 3 , a nilmanifold or a simply connected Alexandrov space with K 0.

19. First Betti number and fundamental group So far we discussed results about sectional curvature. In this section we discuss also Ricci curvature. The recent progress mainly due to Cheeger–Colding will be discussed in later sections. In this section, we mainly concern with older results. To study Ricci curvature we need partial differential equations frequently. But we do not mention them so much. We first review Theorem 2.3. The proof of Theorem 2.3 is based on the Bochner trick. The most famous result in metric Riemannian geometry based on the Bochner trick is the following T HEOREM 19.1 (Bochner [157]). If an n-dimensional compact Riemannian manifold M has nonnegative Ricci curvature, then the first Betti number of M is not greater than n. The proof of Theorem 19.1 due to Bochner is as follows. Let u be a one-form on M. Then we have the following equality of Weitzenbeck type. (For proof see [157]. We remark that we use the nonpositive Laplacian (16.2).) 1 −u, u = − u2 + ∇u, ∇u + Ricci(u, u). 2

(19.1)

Let u be a harmonic one-form. We integrate (19.1) over M. The left-hand side is zero (since u is harmonic) and the integral of the first term in the right-hand side vanish. Therefore we have ∇u, ∇uΩM + Ricci(u, u)ΩM = 0. (19.2) M

M

(Here ΩM is the volume element.) The first term of (19.2) is nonnegative. If we assume that the Ricci curvature is nonnegative, then the second term also is nonnegative. Therefore the first and second term both are zero. Namely every harmonic one-form is parallel. Since a parallel one-form is determined by its value at one point (here we are assuming that M is connected), it follows that the dimension of the space of harmonic one forms on M is at most n. Theorem 19.1 follows. When we try to apply a similar argument to the forms of higher degree and try to estimate higher Betti numbers by Ricci curvature, we will meet a trouble. In formula (19.1), the third term involves only Ricci curvature. This is true only for one-forms. A similar formula for forms of higher degree is much more complicated. The assumption that we need to apply a similar argument to forms of higher degree is exactly the assumption in Theorem 2.3, which is much stronger than the one on Ricci curvature.34 34 On the other hand, if we write a formula similar to (19.1) for the spinor and Dirac operator the second term involves only scalar curvature. (See the text book of the Atiyah–Singer index theorem.) A theorem by Lichnerow-

264

K. Fukaya

In Section 16, we discussed the splitting theorem of Riemannian manifolds of nonnegative Ricci curvature (Theorem 16.4). We can prove Theorem 19.1 by using this theorem also. Actually we have the following T HEOREM 19.2 (Cheeger–Gromoll). If M is a compact manifold with nonnegative Ricci curvature, then there exists a finite cover M˜ of M, such that M˜ is isometric to the direct product X × T k , where X is simply connected and T k is a flat torus. ˆ Since we may assume To prove Theorem 19.2, we consider the universal covering M. ˆ we can prove that the fundamental group of M is infinite (otherwise we may take X = M), 35 ˆ ˆ that M contains a line. Now by applying Theorem 16.4, we find M = R × Y . We may split Mˆ = Rk × Y so that Y has no R factor. If Y is not compact, we can show Y contains a line by the same argument. Then, by Theorem 16.4, Y has an R factor, a contradiction. Namely Y is compact. The π1 M action preserves the splitting Mˆ = Rk ×Y . Theorem 19.2 follows easily. In case k = n in Theorem 19.2, or in case when the first Betti number is equal to the dimension in Theorem 19.1, we can show that M is flat. (We can show this fact either by Bochner’s proof using (19.1) or by Cheeger–Gromoll’s proof based on the splitting theorem.) Theorem 19.2 is generalized by Gromov as follows. T HEOREM 19.3 [69, p. 73]. There exists a continuous function b(n, ρ) of ρ ∈ R with b(n, 0) = n, such that the following holds. If M is an n-dimensional Riemannian manifold with diameter 1, Ricci curvature ρ, Then its first Betti number is not greater than b(n, ρ). C OROLLARY 19.4. If M is an n-dimensional Riemannian manifold with diameter 1 and Ricci > −n , then its first Betti number is not greater than n. Here n is a positive number depending only on n. Gromov’s proof is based on the estimate of the growth function by using the Bishop– Gromov inequality (Proposition 5.2) and is closer to the study of the fundamental group we mention later in this section (Theorem 19.9, Theorem 19.10). The analytic proof, using a similar idea to Bochner’s, is given by Gallot [61]. As we mentioned before, the idea of the proof of Theorem 19.1 cannot directly be applied to the study of the second or higher Betti number. In fact, a result similar to Theorem 19.3 does not hold for higher Betti numbers. Namely the statement such as: “If M is an n-dimensional compact Riemannian manifold with diameter 1, Ricci curvature ρ, then its Betti number is smaller than a number depending only on ρ and n”, is icz which asserts “The Aˆ genus of Riemannian manifold of positive scalar curvature is zero” is obtained from this fact. 35 Let p , q ∈ M with d(p , q ) → ∞. Let x be the midpoint of a minimal geodesic joining p and q . Moving i i i i i i i them by an action of π1 (M), we may assume that there exists R independent of i such that d(x, xi ) < R. Then a subsequence of the sequence of geodesics joining pi , xi , qi has a limit. This limit is a line.

Metric Riemannian geometry

265

false. See [135,118] for counter examples. Note that if we replace Ricci ρ by KM ρ in the statement in the parenthesis, then it is Theorem 14.6. Let us consider the case when equality holds in Corollary 19.4, namely the case Ricci > −n and the first Betti number is n. T HEOREM 19.5 (Yamaguchi [153]). There exists a positive number n such that if M is an n-dimensional Riemannian manifold with diameter Diam(M)KM > −n , and its first Betti number is b, then there exists a finite cover M˜ of M and a fiber bundle M˜ → T b over a b-dimensional torus. Moreover, if b = dim M, then M is diffeomorphic to the torus. Remark 19.1. (1) Yamaguchi proved the same conclusion for the fiber of Theorem 11.3. (2) Yamaguchi [151] proved the same conclusion under a different hypothesis KM 1, Diam D, RicciM −(D, n). To prove Theorem 19.5, Yamaguchi used Theorem 16.4 case (b). The second half of Theorem 19.5 is generalized by Colding [46] and Cheeger–Colding [29] as follows. T HEOREM 19.6 [46,29]. If M is an n-dimensional Riemannian manifold with Diam(M) RicciM > −n , and its first Betti number is n, then M is diffeomorphic to a torus. Remark 19.2. The first half of the statement of Theorem 19.5 does not hold under the milder assumption Diam(M) RicciM > −n . Anderson [10] constructed an example of M with Diam(M) RicciM > −n but that has no fibration over T b1 (M) . We here explain some of the ideas used by Yamaguchi in [153] to show Theorem 19.5, which is also used in [46]. (The additional ideas due to [46,29] will be explained in later sections.) For simplicity we consider the case b = n = dim M only. The proof is by contradiction. By scaling we may assume that there exists Mi with Diam(Mi ) = 1, KMi −i but Mi is not diffeomorphic to T n . We consider the covering space Mˆ i → Mi whose covering transformation group is Γi = Zb . We study the limit of the pair (Mˆ i , Γi ). Here we define D EFINITION 19.1 [52]. A sequence of pairs ((Xi , pi ), Γi ) of pointed metric spaces (Xi , pi ) and groups of isometries Γi is said to converge to ((X, p), G) with respect to the equivariant pointed Hausdorff convergence if there exists ϕi : Bpi (1/i , Xi ) → Bp (1/i , X), ϕi : Bp (1/i , X) → Bpi (1/i , Xi ), ψi : Γi → G, ψi : G → Γi with i → 0 such that (1) ϕi , ϕi are i -Hausdorff approximations and d x, ϕi ϕi (x) < i . d x, ϕi ϕi (x) < i , (2) If x, γ (x) ∈ Bpi (1/i , Xi ), γ ∈ Γi , then d ϕi γ (x) , ψi (γ ) ϕi (x) < i .

266

K. Fukaya

(3) If x, γ (x) ∈ Bpi (1/i , X), γ ∈ Γi , then d ϕi γ (x) , ψi (γ ) ϕi (x) < i . We remark that we do not assume that ψi , ψi are homomorphisms. We can prove a similar compactness result as Theorem 16.1. Now let us go back to the proof of Theorem 19.5. Fix pi ∈ Mˆ i . We may consider the limit ((Mˆ i , Γi ), pi ) with respect to the equivariant pointed Hausdorff convergence. However, then the limit may be a continuous group and is a bit hard to handle. So we use the following lemma. L EMMA 19.7 [153]. There exist subgroups Γi ⊂ Γi of finite index and η, η (independent of i) such that (1) For each γ ∈ Γi with γ = 1 we have d(pi , γ (pi )) η. (2) Γi is generated by elements γ1 , . . . , γn such that d(pi , γk (pi )) η . (Here n = dim M.) Lemma 19.7 appeared in the proof by Gromov of Theorem 19.3. The fact that Γi is Abelian plays an important role in the proof. We omit the proof of Lemma 19.7. Now we can consider the limit of the sequence ((Mˆ i , Γi ), pi ). We denote it by ((X, G), p). Using Lemma 19.7 we can easily show that G ∼ = Zn and its action is properly discontinuous. Now we apply the splitting theorem to X and obtain X = Rk × Y where Y is compact.36 Since Zn acts on it properly discontinuously, it follows that k = n. Since dim X dim Mˆ i = n, it follows that X = Rn . We can also prove that Mˆ i /Γi converges to X/G ∼ = T n . We put M˜ i = Mˆ i /Γi . Since M˜ i is n-dimensional and converges to T n , it follows from Theorem 11.3 that M˜ i is diffeomorphic to T n . Using Hn (Mi , Q) = n again, we can show that Mi is homeomorphic to T n . Furthermore we can arrange the covering index M˜ i → Mi so that “Mi is homeomorphic to T N and Mi is diffeomorphic to T n ” imply that Mi is diffeomorphic to T n , if 5. (This point is a standard application of nonsimply connected surgery.) The last step in the low-dimensional case is a bit complicated and is omitted. Remark 19.3. The above argument can be applied to the situation of Theorem 19.6. We only need to replace the splitting theorem to the one by Cheeger–Colding and Theorem 11.3 by Theorem 22.3. Colding’s argument in [46] (though using Lemma 19.7) is slightly different. This is probably because the splitting theorem we need for this purpose was not yet proved at that time. We next remark the following corollary of Theorem 19.2. C OROLLARY 19.8. If a compact Riemannian manifold M has nonnegative Ricci curvature, then its fundamental group π1 (M) contains an Abelian subgroup of finite index. It seems that series of results related to Corollary 19.8 began with the following theorem. 36 I think this was the first place where splitting theorem of the limit (singular) space was applied to study Riemannian manifold.

Metric Riemannian geometry

267

T HEOREM 19.9 (Milnor [104]). If a complete manifold M has nonnegative Ricci curvature and if G is a finitely generated subgroup of π1 (M), then G has polynomial growth. The definition of a group having polynomial growth is as follows. Let G be a finitely generated group and g1 , . . . , gk generate G. Let fG (N ) be the number of elements of G which can be written by a product of at most N of gi or gi−1 . D EFINITION 19.2. We say that G has polynomial growth, if there exists C, K such that fG (N ) < C(N K + 1). It is easy to see that this definition is independent of the choice of generator of G. The proof of Theorem 19.9 is based on Proposition 5.2 and proceeds as follows. Let us assume M is compact for simplicity. Let M˜ be the covering space of M corresponding ˜ By Proposition 5.2 we have to G. Let p ∈ M. ˜ CR n . Vol Bp (R, M) By an elementary argument using fundamental domain, we can show the existence of C with C −1

ceR/C . (We say that G has exponential growth in this case.) On the other hand, G = Zk has polynomial growth. Gromov [71] proved the following T HEOREM 19.10 (Gromov). A finitely generated group G has polynomial growth if and only if G has a nilpotent subgroup of finite index. Let us very briefly sketch its proof here. First we recall the following T HEOREM 19.11 (Tits [147]). Let G be a finitely generated subgroup of GL(n, R). Then either G contains a solvable subgroup of finite index or G contains a noncommutative free group.

268

K. Fukaya

If G contains a noncommutative free group, we can show that G is not of polynomial growth. On the other hand, Milnor proved that a solvable group is of polynomial growth if and only if it contains a nilpotent group of finite index. Hence to prove Theorem 19.10 it suffices to embed G to some Lie group. Gromov’s idea is to do so by using Hilbert’s 5th problem. Let G a group of as in Theorem 19.10. We define a metric (the word metric) on G as follows. Let γ1 , . . . , γn be generators. Let µ1 , µ2 ∈ G. We define d(µ1 , µ2 ) to be the smallest number k such that µ2 = γi11 . . . γikk µ1 . Here ij ∈ {1, . . . , n}, j = ±1. 1 Now we consider the limit limGH N →∞ (G, N d) as N → ∞. The assumption that G is of polynomial growth is used to show that the limit exists. It is easy to see that the limit G has the structure of a group. We then can use the fact that G acts as isometry on itself preserving the metric and a solution of Hilbert’s 5th problem, to show that G is a Lie group. So if we can embed G to G , we are done. But it is not so easy to embed G to G . (Actually in case G is a discrete subgroup of a nilpotent Lie group N , the limit is N but has a strange metric called the Carnot–Carathéodory metric (see [74]).) Therefore we need to discuss it more carefully and some more technical argument is required. We omit it. Theorems 19.9 and 19.10 imply that the finitely generated subgroup of the fundamental group of a complete manifold of nonnegative Ricci curvature has nilpotent subgroup of finite index. This fact is generalized by Fukaya–Yamaguchi [59]37 and further by Cheeger– Colding [46] as follows. T HEOREM 19.12 (Cheeger–Colding). There exists a positive number n such that if an n-dimensional Riemannian manifold satisfies Diam(M)2 RicciM −n , then π1 (M) contains a nilpotent subgroup of finite index. We remark that Theorem 19.12 follows Theorem 10.5. We also remark that Theorem 10.5 implies Theorem 19.9. In Theorem 19.12 we cannot replace the conclusion “nilpotent” by “Abelian”. Namely we cannot replace the assumption Ricci 0 of Corollary 19.8 by −n . The counter example is an almost flat manifold (Example 10.1). Some more results on the fundamental group is proved in [59] and [60] which we review here. A group Γ is said to be polycyclic if there exists 1 = Γ0 ⊂ Γ1 ⊂ · · · ⊂ Γk = Γk ,

(19.3)

such that Γi is a normal subgroup of Γi+1 and Γi+1 /Γi is cyclic. The smallest such number k is called the degree of polycyclicity of Γ . T HEOREM 19.13 ([59, Theorem 0.6, Corollary 7.20] plus [46]). There exists n and wn such that if an n-dimensional Riemannian manifold M satisfies RicciM Diam(M) −n then π1 contains a normal subgroup Γ such that 37 The result of [59] is the same conclusion as Theorem 19.12, but the assumption is on sectional curvature

instead of Ricci curvature.

Metric Riemannian geometry

269

(1) [π1 (M) : Γ ] wn . (2) Γ is polycyclic and its degree of polycyclicity is not greater n. T HEOREM 19.14 (Fukaya–Yamaguchi [60]). For each D, n there exists a finite set of groups G with the following properties. Let M be a manifold with KM −1, Diam(M) D. Then there exists G ∈ G and a surjective homomorphism π1 M → G such that the kernel Γ satisfies (1), (2) of Theorem 19.13. Theorem 19.14 implies the following. For a group G let us put D(G, n) = inf Diam(M) | KM −1, dim M = n, π1 M ⊇ G . Then, for any sequence of noncommutative simple groups Gi with Gi = Gj for i = j , we have limi→∞ D(Gi , n) = ∞. Remark 19.4. Theorem 17.23 plays a key role in the proof of Theorem 19.14. So far the author does not know the proof of the conclusion of Theorem 19.14 under the milder assumption RicciM −(n − 1). The trouble is a generalization of Theorem 17.23 to the limit X of the manifolds Mi with RicciMi −δi , where δi → 0. (Namely the problem whether the isometry group of such X is a Lie group or not.) Cheeger–Colding [30] proved that the group of isometries of X is a Lie group under the additional assumption Vol(Mi ) v > 0. Under this additional assumption there is the following result (Anderson [9]): The number of isomorphism classes of π1 M of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Vol(Mi ) v > 0, Diam(M) D, is finite. We here sketch a part of the proof of Theorem 19.12 given in [59]. Namely we assume the splitting theorem and explain how to deduce Theorem 19.12 from it. Here we consider the case Diam(M)KM −n 38 to simplify the argument. We first need a lemma on the convergence of groups. If Γ acts on a metric space X by isometries and p is a base point of Xwe write Γ (D) = γ | d γ (p), p D . Here A is the subgroup generated by A. L EMMA 19.15 [59, Theorem 3.10]. Let (Xi , Γi , pi ) converge to (X, G, q) in the pointed equivariant Hausdorff distance. We assume that the connected component G0 of G is a Lie group and G/G0 is discrete and finitely presented. We also assume that X/G is compact. Moreover we assume that Xi is simply connected and Γi is properly discontinuous and free. 38 If we use Cheeger–Colding’s splitting theorem, a similar argument works. However we need several modifica-

tions on the technical points of the arguments of [59] or given below. Unfortunately the technical details of such arguments is not written in the literature. The author and T. Yamaguchi are planning to write it and make it public in the near future. But maybe it is too technical to be included in this article.

270

K. Fukaya

Then there exists a sequence of normal subgroups Γi,0 converging to G0 such that Γi /Γi,0 ∼ = G/G0 for large i.39 We omit the proof. Now we prove the following P ROPOSITION 19.16. Let (Mi , pi ) converge to (Rk , 0) with respect to the pointed Hausdorff distance. Assume RicciMi −(n − 1). Then there exists > 0 such that the image of π1 (Bpi (, Mi )) in π1 (Bpi (1, Mi )) has a solvable subgroup of finite index for large i. The solvability in Theorem 19.12 is the case k = 0 of Proposition 19.16. (The proof of a more precise statement as in Theorem 19.13 and nilpotency is omitted.) The proof of Proposition 19.16 is by downward induction on k. The case k = dim Mi follows from Theorem 11.3. We assume Proposition 19.16 is correct for k + m (m > 0) and show it for k by contradiction. Let (Mi , pi ) as in Proposition 19.16. We use Theorem 11.3 to find Vi ⊆ Mi and a fiber bundle fi : Vi → B0 (Ci , Rn ) with Ci → ∞. Here Vi ⊇ Bpi (Ci /2, Mi ). Let δi = √ Diam(fi−1 (0)). We take the metric gi,1 = gi / δi . The limit of (Vi , gi,1 ) with respect to the pointed Hausdorff distance is Rk × Z where Z is an Alexandrov space with K 0. Let Γi = π1 (Fi ) = π1 (Vi ). We take ((V˜i , g˜ i,1 ), Γi , p˜ i ) where (V˜i , g˜ i,1 ) is the covering space of Vi equipped with metric induced from gi,1 . Let us take a subsequence and let (V∞ , G, q) be the limit. We apply the splitting Theorem 16.4 to V∞ and find V∞ = R × Y where Y is compact. Since (R × Y )/G ∼ = Rk × Z we find that V∞ = Rk × R−k × Y such that G −k −k acts only on R × Y and (R × Y )/G = Z. Since G is a Lie group by Theorem 17.23, it follows that we can take its connected component G0 . Since G/G0 is discrete and (R−k × Y )/G is compact we can prove that G/G0 has Abelian subgroup of finite index. (This is easy to see if G acts effectively on R−k . The compact factor Y only contributes a finite group.) To apply Lemma 19.15 we replace V∞ = Rk × R−k × Y by X = B0 (D, Rk ) × R−k × Y for large but fixed D and find a sequence ((Xi , dXi ), Γi , pi ) converging to ((X, dX ), G, q). (We can find such Xi ⊂ Vi easily by using the fiber bundle fi .) We now apply Lemma 19.15 to obtain Γi,0 . Since (X, dX ) is an Alexandrov space, it follows from Theorem 17.2 that we can find q near q and ri → ∞ such that ((X, ri dX ), q ) converges to (Rk+m , 0) with m > 0. (Note that since Diam(Z) = 1, it follows that R−k × Y is not a point.) We may replace ((Xi , dXi ), Γi , pi ) by a subsequence which converges to ((Xi , dXi ), Γi , pi ) very quickly compared to 1/ri . Then we find qi such that ((Xi , ri dXi ), Γi , qi ) converges to (Rk+m , G , 0) for some G . Since we can use the fact that Γi,0 converges to G0 , the connected Lie group and the convergence is quick compared to ri to show that Γi,0 is generated by Γi,0 (δi ) = {γ ∈ Γi,0 | d(γ (qi ), qi ) < δi } where δi → 0. Now we apply induction hypothesis. Then if > δi (where is as in Proposition 19.16), we find that Γi,0 (δi ) has a solvable subgroup with finite index. This is a contradiction since Γi /Γi,0 ∼ = G/G0 has an Abelian subgroup with finite index. 39 This lemma is actually weaker than [59, Theorem 3.10]. But it is enough for the present purpose, since we

now have Theorem 17.23.

Metric Riemannian geometry

271

20. Hausdorff convergence of Einstein manifolds In the last four sections, we discuss Gromov–Hausdorff convergence under the assumption Ricci −(n − 1). We first remark that, when we work under the assumption Ricci −(n − 1), the topology can change when we go to the limit, even in the noncollapsing situation, namely in the situation where we assume Vol v > 0. Such a phenomenon was first observed in the study of 4-dimensional Einstein (or complex 2-dimensional Kähler–Einstein) manifold (at least around 20 years ago as far as I know). Let us start by a review of the case of Einstein manifolds. Let Γ ⊂ SU(2) be a finite subgroup. We consider the quotient C2 /Γ . It is a Kähler orbifold with isolated singularity at the origin. (This singularity is called the Kleinian singularity.) There is a resolution 2 /Γ → C2 /Γ . called minimal resolution of the Kleinian singularity which we denote by C 2 /Γ Eguchi–Hanson [49] and others constructed a Ricci flat Kähler metric g on C C2 /Γ which is asymptotically locally Euclidean (in the sense we define later in Definition 20.1). (Such a metric is called a gravitational instanton.) Asymptotically locally Euclidean met 2 /Γ are classified by Kronheimer [98]. rics on C

Suppose (X, gX ) is a 4-dimensional Riemannian orbifold locally of Kleinian type. (Namely X is locally a quotient of C2 by a finite group Γ ⊂ SU(2). We assume also that the metric on X is a quotient metric with respect to a certain Γ -invariant metric locally.) We assume that X is Ricci flat and Kähler. (Namely its Ricci curvature at regular points is 0 and the metric is Kähler at regular point.) We can locally glue the metric gX on X and the Ricci flat Kähler metric g on X to obtain a metric g on the resolution X˜ of X. g is C2 /Γ almost Ricci flat. We can use the technique of Yau’s proof of the Calabi conjecture [158] to show that there exists a Ricci flat Kähler metric on X˜ near g . (See [96,13].) We remark that ˜ g ) and (X, ˜ g ) converges to (X, g) with respect to the Gromov–Hausdorff distance. (X, A typical example is a Kummer surface where we take X = T 4 /Z2 (and Γ = Z2 ). Thus, we have O BSERVATION 20.1. There exists a family of Riemannian manifolds (X, g ), such that Riccig ≡ 0, Vol(X, g ) v > 0 and the limit of (X, g ) as → 0 converges to a compact metric space X which is not a manifold. The construction here is an analogue of Taubes’ construction [145] of anti-self dual connections on 4 manifolds. Later Joyce (see [89]) generalized this construction and used it to construct (higherdimensional) Riemannian manifolds with exceptional holonomy. (They are in particular Ricci flat.) Namely Joyce started, for example, with a 7-dimensional flat orbifold X = T 7 /Γ , which is obtained by T 7 , divided by a finite group of isometries Γ . In his example, the singular locus of X is a codimension 4 totally geodesic smooth submanifold (actually 2 /Z (equipped with direct product it is a disjoint union of T 3 ). Then Joyce glued T 3 × C 2 metric) along a singularity to obtain a Riemannian manifold and used the implicit function theorem to obtain a manifold with exceptional holonomy. In his construction, we also have

272

K. Fukaya

a family of metrics g which is of exceptional holonomy (and in particular is Ricci flat) and which converges to X. A converse to Observation 20.1 is proved by Nakajima and others as follows. T HEOREM 20.2 (Nakajima [106]). Let gi be a sequence of Einstein metrics with Ricci = ±1 or 0, on a 4 manifolds M, such that Vol(M, gi ) v > 0. (Here v is independent of i.) Let X = limGH i→∞ (M, gi ). Then there exists a finite subset S ⊂ X such that X \ S is an Einstein 4-manifold. Moreover, for every δ > 0 there exists a diffeomorphism Φi : X \ Nδ S → M such that the pullback Riemannian metric Φi∗ gi converges to the Riemannian metric on X in C ∞ topology. Remark 20.1. In case of 4-dimensional Einstein manifolds, the L2 -norm of the curvature

Mi

|RMi |n/2 ΩMi =

|RM |2 ΩM

(20.1)

M

is a topological invariant and is estimated by the Euler number. This fact is essential in the proof of Theorem 20.2.40 In case dim Mi > 4, the same conclusion as Theorem 20.2 holds under the additional hypothesis Mi

|RMi |n/2 ΩMi C.

(20.2)

(In case we assume (20.2) we do not need to fix a topological type of M.) Namely under assumption (20.2) and Vol(Mi ) v > 0, the limit space X of a sequence of Einstein manifolds Mi has only finitely many singular points. We remark however the assumption (20.2) is too restrictive to handle the limit of Einstein manifolds. In the example of Joyce mentioned above the limit of a sequence of 7-dimensional Einstein manifolds is T 7 /Z2 whose singularity is 3-dimensional. In this example, the L2 -norm of the Ricci curvature is bounded but the L3.5 norm is not bounded. To study the structure of Mi or X near a singular point ∈ S, we use the scaling argument as follows. For completeness we include the case when dim M is general. Namely we assume we have a sequence of Einstein manifolds Mi converging to X. We assume (20.2) and Vol(Mi ) v > 0. (Then the singular point set S of X is of finite order.) Let pi ∈ Mi which converges to p∞ ∈ S. We scale the metric gMi to Ri gMi so that |KRi gMi | becomes 1 at pi . We then consider the limit ((Mi , Ri gMi ), pi ) with respect to the pointed Gromov– Hausdorff distance. Theorem 16.1 implies that it has a limit, which we denote by (X, gX ). 40 (20.1) is scale invariant if and only if dim M = 4. (We do not need the Einstein condition for this.) In this sense also the situation is very much similar to the study of the Yang–Mills equation in dimension 4. (Compare also the footnote 15 at the beginning of Section 7.)

Metric Riemannian geometry

273

Using the injectivity radius bound we can show that (X, gX ) is a Ricci flat Riemannian manifold. (It is noncompact but complete.) It also satisfies the following condition:

n/2 Ω X X |RX |

C1 , Vol(Bp (R, X)) C2 R n .

(20.3)

(See [14].) We then can apply the following Theorem 20.3. We define D EFINITION 20.1. A complete pointed Riemannian manifold ((X, g), p) is said to be locally almost Euclidean (abbreviated by ALE hereafter) of order τ > 0, if there exists a finite group Γ ⊂ O(n) and a diffeomorphism Φ : X \ Bp (R, X) → (Rn \ B0 (R, Rn ))/Γ such that

−1 ∗

Φ gX − gcan )(x) C|x|−τ ,

(20.4a)

−1−τ −α |(∇ k Φ −1 )∗ gX (x) − (∇ k Φ −1 )∗ gX (y)| C min |x|, |y| , α |x − y|

(20.4b)

holds for some α and R. Here gcan is the metric on Bp (R, X) C2 R n induced by the Euclidean metric on Rn . T HEOREM 20.3 (Bando–Kasue–Nakajima [14]). If (X, gX ) is an n-dimensional Einstein manifold satisfying (20.3) then it is ALE of order n − 1. If (X, gX ) is Einstein–Kähler and n = 4 then it is ALE of order n. Combining them we have T HEOREM 20.4 ([14], Anderson [11]). The limit space X in Theorem 20.2 is an Einstein orbifold.41 In higher dimensions Theorems 23.16, 23.17 give a natural generalization of the results we explained here. If we remove the assumption Vol(Mi ) v > 0 (namely if we study the collapsing situation), then even in the case of Einstein manifold, not so many things are known. This problem is related to mirror symmetry in string theory and is calling attention of several differential geometers working on it. There is a result by Gross–Wilson [76] which discusses the case of K3-surfaces in the collapsing situation and obtains a singular torus fibration. We now consider more general Riemannian manifolds under the condition of Ricci curvature below. To obtain a result similar to Theorem 3.4 we need to avoid the phenomenon we described in Observation 20.1. There are several results assuming a lower bound of injectivity radius, for this purpose. We denote by Sn (D, i > ρ) the set of all isometry classes of n-dimensional compact Riemannian manifolds (without boundary) such that RicciM −(n − 1), Diam(M) D and iM ρ everywhere. Let α ∈ (0, 1). 41 A similar result holds under an additional assumption (20.2).

274

K. Fukaya

T HEOREM 20.5 (Anderson–Cheeger [12]). Let Mi ∈ Sn (D, i > ρ) and X = limGH i→∞ Mi . Then X is a Riemannian manifold of C α -class and there exist diffeomorphisms ϕi : Mi → X such that (ϕi−1 )∗ gMi converges to gX with respect to the C α -norm. Remark 20.2. Under the stronger assumption | RicciM | (n − 1), Diam(M) D and iM ρ, Anderson [7] proved a stronger result. Namely the limit space X is a C 1,α Riemannian manifold and (ϕi−1 )∗ gMi converges to gX with respect to C 1,α -norm. It was applied (in [7]) to prove a sphere theorem and a pinching theorem for an almost Einstein metric. C OROLLARY 20.6. The number of diffeomorphism classes represented by elements of Sn (D, i > ρ) is finite. The proof of Theorem 20.5 is quite similar to the arguments in Sections 6–8. Namely we construct harmonic coordinate and obtain an appropriate estimate. Then the proof is completed by using the diffeotopy extension theorem (or the center of mass technique which we can apply to a smooth metric near the limit C α metric). So the new result in [12] is the following T HEOREM 20.7 [12, Theorem 0.1]. There exists C(n, ρ), (n, ρ) > 0 with the following property. Let M ∈ Sn (D, i > ρ). We can then cover M by harmonic coordinates Ui such that the C 1,α -norm of the coordinate transformation is smaller than C(n, ρ) and the C α norm of the metric tensor written in this coordinate is smaller than C(n, ρ). Moreover for any p ∈ M, the metric ball Bp ((n, ρ), M) is contained in some Ui . 21. Sphere theorem and L2 comparison theorem In the last three sections, we concern with the class of Riemannian manifolds with Ricci curvature bounded from below. Especially we discuss results obtained by Colding and Cheeger–Colding recently. The surveys [45,47,62] and the book [26] are recommended for their results. The basic tool to study such Riemannian manifolds is Theorem 5.2. So we first draw some of its consequences. We put Ap (a, b; M) = x ∈ M | a d(p, x) b , (21.1) Sp (a; M) = x ∈ M | d(p, x) = a . L EMMA 21.1. If RicciM κ, a < b < c, then Vol(Ap (b, c; M)) Vol(Ap (a, b; M)) Vol(Ap0 (a, b; Sn (κ))) Vol(Ap0 (b, c; Sn (κ)))

(21.2)

Vol(Sp (a; M)) Vol(Ap (a, b; M)) Vol(Sp (b; M)) . n n Vol(Sp0 (a; S (κ))) Vol(Ap0 (a, b; S (κ))) Vol(Sp0 (b; Sn (κ)))

(21.3)

and

Metric Riemannian geometry

275

(21.2) follows from Vol(Ap (0, a; M)) + Vol(Ap (a, b; M)) Vol(Ap0 (0, a; Sn (κ))) + Vol(Ap0 (a, b; Sn (κ)))

Vol(Ap (0, a; M)) + Vol(Ap (a, b; M)) + Vol(Ap (b, c; M)) . Vol(Ap0 (0, a; Sn (κ))) + Vol(Ap0 (a, b; Sn (κ))) + Vol(Ap0 (b, c; Sn (κ)))

By taking the limit b → a and b → c in (21.2) we obtain (21.3).

L EMMA 21.2. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then for each x ∈ M we have d(p, x) + d(q, x) − d(p, q) τ (|n). To show the lemma, let δ = d(p, x) + d(q, x) − d(p, q), r = d(p, x) − δ/2, s = d(p, q) − r = d(q, x) − δ/2. Then (Bp (r, M) ∪ Bq (s, M)) ∩ Bx (δ/2, M) = ∅ and Bp (r, M) ∩ Bq (s, M) = ∅. Therefore, by Theorem 5.2, we have Vol(Bp (r, M) ∪ Bq (s, M)) 1 − τ (|n), Vol(M)

Bx (δ/2, M) Cδ n . Vol(M)

Hence δ < τ (|n) as required.

C OROLLARY 21.3. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then Diam M \ Bp (π − , M) < τ (|n). Corollary 21.3 is an immediate consequence of Lemma 21.2 and Myers’ Theorem 5.4. We remark that Corollary 21.3 is a version of Proposition 4.4. Namely the conclusion of Corollary 21.3 is weaker than that of Proposition 4.4, but it holds under milder assumption. L EMMA 21.4. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then Vol(Sp (π − δ; M)) Vol(Sp (δ; M)) + τ (|δ, n). n Vol(Sp0 (δ; S (1))) Vol(Sp0 (π − δ; Sn (1)))

(21.4)

We remark Vol(Sp (δ; M)) Vol(Sp (π − δ; M)) n Vol(Sp0 (δ; S (1))) Vol(Sp0 (π − δ; Sn (1))) is a consequence of (21.1). Hence (21.4) implies that the ratio of volume Vol(Sp (t; M))/ Vol(Sp0 (t; Sn (1))) is almost constant for t ∈ [δ, π − δ].

276

K. Fukaya

Let us prove Lemma 21.4. Let - ρ - δ. By Corollary 21.3, we have Aq (δ − 2ρ, δ − ρ; M) ⊆ Ap π − δ + ρ − τ (|n), π − δ + 2ρ + τ (|n); M .

(21.5)

We may assume ρ − τ (|n) 0. We remark Vol Ap0 δ − 2ρ, δ − ρ; Sn (1) = Vol Ap0 π − δ + ρ, π − δ + 2ρ; Sn (1) . Therefore (21.5) (together with Lemma 21.1) implies the first inequality of Vol(Aq (δ − 2ρ, δ − ρ; M)) Vol(Ap0 (δ − 2ρ, δ − ρ; Sn (1)))

Vol(Ap (π − δ, π − δ + ρ; M)) + τ (|δ, ρ, n) Vol(Ap0 (π − δ, π − δ + ρ; Sn (1)))

Vol(Ap (δ − 2ρ, δ − ρ; M)) + τ (|δ, ρ, n). Vol(Ap0 (δ − 2ρ, δ − ρ; Sn (1)))

(21.6)

Here the second inequality is a consequence of Lemma 21.1. Changing the role of p and q we have Vol(Ap (δ − ρ, δ; M)) Vol(Aq (δ − ρ, δ; M)) + τ (|δ, ρ, n). (21.7) n Vol(Ap0 (δ − ρ, δ; S (1))) Vol(Ap0 (δ − ρ, δ; Sn (1))) Therefore by (21.6), (21.7) and Lemma 21.1 we have Vol(Sp (π − δ; M)) Vol(Sp (δ; M)) + τ (|δ, ρ, n) Vol(Sp0 (π − δ; Sn (1))) Vol(Sp0 (δ; Sn (1))) as required.

L EMMA 21.5. If RicciM κ and p ∈ M then Vol Bp (R, M) Vol Bp0 R, Sn (κ) . This is an immediate consequence of Theorem 5.2.

We next discuss sphere theorems. The sphere theorem appearing here can be regarded as a generalization of Theorem 17.20. So we first sketch its proof. We first remark: L EMMA 21.6 [79,111]. If KM 1, Vol(M) > Vol(S n ) − , then the Gromov–Hausdorff distance between M and S n is smaller than τ (|n).

Metric Riemannian geometry

277

Let p ∈ M. We identify Tp M with Tp∗ S n for a point p∗ ∈ S n . We then define −1 Φ : M → S n by Φ = Expp∗ ◦ Exp−1 p . Note Expp is discontinuous. Using Corollary 21.3, Lemma 21.5 and the Toponogov comparison Theorem 4.7, we can show that Φ is a τ (|n)Hausdorff approximation. We omit the details since we discuss a sharper result (Theorem 21.8) later. Now under the assumption of Lemma 21.6 we can find points p0 , . . . , pn , q0 , . . . , qn ∈ M such that

d(pi , pj ) − π/2 < τ (|n),

d(pi , qi ) − π < τ (|n).

d(pi , qj ) − π/2 < τ (|n), (21.8) i

In fact, if S n ⊆ Rn+1 , the points pi = (0, . . . , 0, 1, 0, . . . , 0), qi = −pi , satisfy (21.8). Hence we can choose pi = Φ(pi ) where Φ : S n → M is an -Hausdorff approximation. Moreover, in case of S n , the canonical embedding IS n : S n → Rn+1 is obtained by IS n (x) = cos d(p0 , x), . . . , cos d(pn , x) .

(21.9)

Now the idea is to embed M in a neighborhood of S n by using a formula similar to (21.9). Namely we first take a smooth function ϕi which is close to d(x, pi ) up to the first derivative, if x ∈ / Bpi (o(), M) ∪ Bqi (o(), M). We then define IM : M → Rn+1 by IM (x) = ϕ0 (x), . . . , ϕn (x) .

(21.10)

We can then prove that d(IM Φ(x), I (x)) < o() and dist TI (x) S n , TIM Φ(x) IM (M) < τ (|n). Here dist in the above formula is a distance as a codimension one linear subspace in Rn+1 . We can use these two formulas to prove that M is diffeomorphic to S n (in a similar way to Section 9). Theorem 17.20 is generalized by Perelman as follows. T HEOREM 21.7 [114]. There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n , then M is homeomorphic to a sphere. Actually Perelman proved that πk (M) = 1 for k < n under the assumption of Theorem 21.7 and applied the generalized Poincaré conjecture. The idea of the proof is hard to explain for the author in this kind of article. So we refer [114] or [159]. We will discuss the proof of a sharper version, Corollary 22.4, in Section 22. Remark 21.1. We remark that a similar sphere theorem replacing volume by diameter does not holds. Actually Anderson [8] and Otsu [110] found examples of manifolds (M, gi ) such

278

K. Fukaya

that Riccigi (n − 1), Vol(M, gi ) v > 0 and Diam(M, gi ) → π but M = S n . (Otsu’s example is S m × S n−m and Anderson’s example is CP n or CP 2 !CP 2 .) We remark that Vol(Mi ) → Vol(S n ) implies Diam Mi → π (under the assumption Riccigi (n − 1)) by the Bishop–Gromov comparison Theorem 5.2. Now we start the review of the works of Colding, who began with the following theorem closely related to Theorems 17.20 and 21.7. T HEOREM 21.8 (Colding [43,44]). Let M be an n-dimensional Riemannian manifold with RicciM (n − 1). (1) If Vol(M) Vol(S n ) − then dGH (M, S n ) < τ (|n). (2) If dGH (M, S n ) < then Vol(M) Vol(S n ) − τ (|n). The proof is somewhat similar to the proof of Theorem 17.20. However we need several new ideas. Especially we need to develop some method to compare IM (21.10) with IS n (21.9). In the situation of the proof of Theorem 17.20, this was done by Toponogov’s comparison theorem. In our situation, Toponogov’s comparison theorem does not apply since there is no sectional curvature bound. Colding developed the L2 -comparison theorem for this purpose. We describe it below. We consider p∗ ∈ S n and ∗ : [0, α] → S n be a geodesic parameterized by arc length. We put ∗ (0) = q∗ , (d∗ /dt)(0) = v∗ ∈ Tq∗ S n . We then put hp∗ ,α (v∗ , t) = cos d(∗ (t), p∗ ). We can calculate it easily as hp∗ ,α (v∗ , t) =

1 d p∗ , (α) sin(α − t) + d p∗ , (0) sin t . sin α

(21.11)

Now we use (21.11) to define a function on M with which we compare the distance function. Let p ∈ M and : [0, α] → M be a geodesic parameterized by arc length. We put (0) = q, (d/dt)(0) = v ∈ Tq M. ( is determined by v so we write = v .) Let f : M → R be a function. We then define hf,α (v, t) =

1 f v (α) sin(α − t) + f v (0) sin t . sin α

(21.12)

We remark that hf,α may be regarded as a function of (v, t) ∈ SM ×[0, α], where SM is the unit tangent bundle SM = {v ∈ T M | |v| = 1}. In case f (x) = d(p, x), we put hf,α = hp,α . Now the L2 -Toponogov theorem in [44] is as follows. T HEOREM 21.9 [44, Proposition 1.15]. Let a0 ∈ [π/2, π). We assume RicciM (n − 1) and p, q ∈ M with d(p, q) π − . Then, for α α0 , we have 1 α Vol(SM)

v∈SM

< τ (|n, α0 ),

cos d p, v (t) − hp,α (v, t) 2 ΩSM dt

α

0

(21.13)

Metric Riemannian geometry

1 α Vol(SM)

279

2 α

d

cos d p, v (t) − dhp,α (v, t) ΩSM dt

dt dt v∈SM 0

< τ (|n, α0 ).

(21.14)

Here ΩSM is the Liouville measure. (Hereafter we omit the symbol of volume form in case it is clear which volume form we use.) Remark 21.2. We remark that (21.13) means that the length d(p, v (t)) is close to the length of the corresponding triangle in S n in L2 -sense. (21.14) means that the angle pv (t)v (0) is close to the angle in corresponding triangle in S n in L2 -sense. Let us explain a part of the ideas of the proof of Theorem 21.9. We first recall the following. Let λ1 (M) denotes the first nonzero eigenvalue of the Laplacian on (the functions of) M. T HEOREM 21.10 (Lichnerowicz [101], Obata [109]). If an n-dimensional Riemannian manifold satisfies RicciM (n − 1) then λ1 (M) −n. The equality holds if and only if M is isometric to the sphere. The proof can be done by the Bochner formula, in the same way as the argument of Step 1 below. We also remark the following theorem by Cheng which is closely related to Theorem 21.10. T HEOREM 21.11 [41]. Let M be a compact Riemannian manifolds with RicciM (n−1). If Diam(M) π , then M is isometric to S n . What is important for us is that the first eigenfunction of S n is cos d(p, ·) and is exactly the function we want study in Theorem 21.10. So the idea of the proof of Theorem 21.10 goes as follows. Step 1. Let f satisfy f + nf < δ, f = 1. (Here is the L2 -norm.) We prove 1 α Vol(SM)

v∈SM

f v (t) − hf,α (v, t) 2 < τ (δ|n),

α

(21.15)

0

and a similar estimate for the t-derivative of cos f (v (t)) − hf,α (v, t) [44, Lemma 1.4]. This step uses the Bochner–Weitzenbeck formula

2

1 |∇f |2 = Hess(f ) + ∇f, ∇f + Ricci(∇f, ∇f ). 2

(21.16)

Here Hess(f )(X, Y ) = X(Y (f )) − (∇X Y )(f ). (Note we are using the positive Laplacian.) The proof is a kind of an “almost version” of the proofs of Theorems 21.10 and 21.11. To

280

K. Fukaya

clarify the geometric ideas, avoiding analytic details, we consider the case

f = λf , n −λ > 0 and prove f (v (t)) = hf,α (v, t). We integrate (21.16) and using M ∇f1 , ∇f2 = − M f1 , f2 , we find

Hess(f ) 2 − |f |2 + (n − 1)|∇f |2 0. M

Since λ

M

|∇f |2 =

M

M ∇f, ∇f = − M

|f |2 , it follows that

Hess(f ) 2 − λ + n − 1 |f |2 0. λ

By Trace Hess(f ) = f and elementary linear algebra, we find λ = −n and Hess(f ) = −f gM .

(21.17)

Using the fact d 2 f (v (t))/dt 2 = Hess f (˙v (t), ˙v (t)) we have d2 f v (t) = −f v (t) . 2 dt

(21.18)

f (v (t)) = hf,α (v, t) follows. Step 2. Let p, q ∈ M with d(p, q) > π − δ. We consider g(x) = cos d(p, x). We then find f with f + nf < δ and f − gL2 < δ. ( L2 is the Sobolev norm, that is an 1

1

L2 -norm up to the first derivative [44, Lemma 1.10].) The essential part of this step (which is explained below) is to show

2 2

n g − |∇g|

τ (δ) Vol(M), M

(21.19)

M

g < τ (δ) Vol(M).

(21.20)

In fact, (21.20) implies that g is almost perpendicular to the 0th eigenfunction of the Laplacian (the constant). Then we can use (21.19) and λ1 n to get the conclusion. Let a(v, t) be as in the proof of Theorem 5.2. We extend it as 0 outside V . (So precisely speaking a(v, t) is the function which we wrote a (v, t) in the proof of Theorem 5.2.) By Lemma 21.4 we have a(v, δ) a(v, π − δ) + τ (|δ, n). (21.21) v∈S n−1

v∈S n−1

On the other hand, the map t → a(v, t) is nondecreasing by the proof of Theorem 5.2. It follows that

(21.22) a(v, s) − a(v, s ) τ ( | δ, n)

v∈S n−1

v∈S n−1

Metric Riemannian geometry

281

for s, s ∈ [δ, π − δ]. Therefore

π

n−1

g =

a(v, t) cos t sin t

n−1 M v∈S t=0

π/2

n−1

=

a(v, t) − a(v, π − t) cos t sin t

v∈S n−1

t=0

τ (, δ|n) Vol(M).

(21.23)

Moreover using |∇g|2 (x) = sin2 d(p, x) we have

|∇g|2 = M

v∈S n−1

On the other hand 2 |g| = M

v∈S n−1

π

a(v, t) sinn+1 t.

t=0

π

a(v, t) cos2 t sinn−1 t.

t=0

π

π We remark 0 sinn+1 t dt = n 0 cos2 t sinn−1 t dt. Hence using (21.22) we can easily show

2

|∇g|2 − n (21.24) |g| < τ (, δ | n) Vol(M).

M

M

(21.23) and (21.24) complete this step as we mentioned before.

These two steps and some more arguments imply Theorem 21.9. (The integral in Theorem 21.9 is taken with respect to the Liouville measure on the unit sphere bundle. In the argument so far the measure is taken with respect to the measure on M itself (or its products). They are equivalent by Theorem 5.2.) We remark that in Theorem 21.9 we use only a weaker assumption Diam(M) ∼ π and not yet Vol(M) ∼ Vol(S n ). (Compare Remark 21.1, which shows that Diam(M) ∼ π does not imply that dGH (M, S n ) is small.) Now using Theorem 21.9, the proof of Theorem 21.8 goes roughly as follows. We first explain (1). Let us assume Vol(M) Vol(S n ) − δ. It then implies that for each p ∈ M there exists q ∈ M such that d(p, q) > π − τ (δ).42 (This follows from the Bishop– Gromov Theorem 5.2.) Now we claim L EMMA 21.12 [44, Lemma 2.25]. Under the assumption of Theorem 21.8(1) there exist pi , qi (i = 0, . . . , n) such that (21.8) holds. 42 Shiohama–Yamaguchi [140] introduced the notion of radius of M that is inf sup d(p, q). This assertion p q

means that the radius of M is close to π .

282

K. Fukaya

Once we have Lemma 21.12, we can construct a Hausdorff approximation Φ : M → S n by perturbing x → (cos d(x, p0 ), . . . , cos d(x, pn )). In fact, by Theorem 21.8, we can prove that the function x → cos d(x, p0 ) behaves in a similar way (modulo τ (δ)) outside the set of measure τ (δ). This is enough to show that it is a Hausdorff approximation. Remark 21.3. As we mentioned before we can use the L2 -comparison theorem directly to show that a map is a Hausdorff approximation. However we cannot use it directly to find a homeomorphism. This is because the L2 -comparison theorem does not tell what happens on a set of small measure. This point is very different from the Toponogov comparison theorem, which however works only under the assumption of sectional curvature. We can use several ‘indirect’ arguments to obtain various topological information using the L2 comparison theorem. (See the next two sections.) The proof of Lemma 21.12 uses Theorem 21.7 and goes as follows. We construct pi , qi (i = 0, . . . , k) satisfying (21.8) by induction on k. Suppose we have pi , qi (i = 0, . . . , k). We then construct a map Φk : M → Rk by x → (cos d(x, p0 ), . . . , cos d(x, pk )). We coni

struct a set Ak from of p0 , . . . , pk , q0 , . . . , qk . In case M = S n and pi = (0, . . . , 0, 1, 0, . . . , 0) Ak = Rk+1 ∩ S n = S k is obtained by joining pi , qi several times along minimal geodesics. We imitate the construction of Ak from pi , qj in M to obtain Ak ⊂ M. (Actually we need to join only by good geodesics v that is a geodesic such that cos d(p, v (t)) − hp,a (v, t) is small. Theorem 21.9 implies that there are enough such geodesics.) Now the restriction of Φk to Ak is similar to the one for S n . Hence Φk (Ak ) lies in a neighborhood of S k and we may regard Ak ∼ = S k . Since k < n, Theorem 21.7 implies that Ak is homotopic to zero in M. This implies that there exists pk+1 ∈ M such that Φk (pk+1 ) = 0. We take qk+1 with d(pk+1 , qk+1 ) > π − δ. Thus induction works. To prove (2) of Theorem 21.8 we proceed as follows. We take pi , qi (i = 0, . . . , n) such that (21.8) holds. (Since dH (M, S n ) is small we can take such pi , qi .) We use it to ˜ ˜ ˜ = (cos d(p0 , x), . . . , cos d(pn , x)), Φ(x) = Φ(x)/| Φ(x)|. construct Φ : M → S n by Φ(x) Using Theorem 21.9, we find that the determinant of the Jacobi matrix of Φ is almost everywhere close to 1. It follows that | Vol(M) − Vol(Φ(M))| < τ (δ|n). We need another idea to show that Vol(S n \ Φ(M)) < τ (δ|n). Actually for this purpose we need a “local version” of Theorem 21.13 [43, Proposition 4.5]. We omit it. The argument of the proof of Theorem 21.8 is a prototype of the argument which is used by Colding and Cheeger–Colding at several other places. We explain them more in the last two sections where the argument is combined with other arguments which are of more analytic nature.

22. Hausdorff convergence and Ricci curvature—I In Section 21, we compared the distance function of a manifold of positive Ricci curvature to the one of the round sphere, in the sense of L21 -norm. In this section, we compare

Metric Riemannian geometry

283

the distance function of a manifold of almost nonnegative Ricci curvature to the one of Euclidean space. T HEOREM 22.1 (Colding [46, Theorem 0.1]). Let Mi be a sequence of n-dimensional Riemannian manifolds with RicciMi −(n − 1) and let M∞ be another n-dimensional Riemannian manifold. We assume limGH i→∞ Mi = M∞ . Then we have lim Vol(Mi ) = Vol(M∞ ).

i→∞

Remark 22.1. Actually Colding proved the following stronger (local) result in [46]. Let Mi and M∞ be complete Riemannian manifolds. We assume RicciMi −(n − 1). Let pi ∈ Mi , p∞ ∈ M∞ , and r > 0. We assume that limGH i→∞ Bpi (r, Mi ) = Bp∞ (r, M∞ ). Then limi→∞ Vol(Bpi (r, Mi )) = Vol(Bp∞ (r, M∞ )). Together with a result by Perelman and using results of controlled surgery, Theorem 22.1 implies the following T HEOREM 22.2 [46]. In the situation of Theorem 22.1, Mi is homotopy equivalent to M∞ for large i. Moreover Mi is homeomorphic to M∞ for large i if n = 3. Remark 22.2. In case the limit space is singular we cannot prove a result similar to Theorem 22.2 because of Example 21.1 by Anderson and Otsu. The Gromov–Hausdorff limit of the metrics Otsu constructed on S 3 × S 2 is a suspension of S 2 × S 2 and hence is not a topological manifold. Theorem 22.1 follows from Theorem 22.2 roughly in the following way. Choose pj∞ ∈ M∞ , j = 1, . . . , N , and small r > 0 such that N

Bpj∞ (r, M∞ ) = M∞ ,

i=1

and 1−δ

Vol(Bpj∞ (r, M∞ )) Vol0 (B0 (r, Rn ))

1 + δ.

(22.1)

Let Φi : M∞ → Mi be an i Hausdorff approximation with i → 0. We take pji = Φi (pj∞ ) ∈ Mi . Since dH (Bpj (r, Mi ), Bpj (r, M∞ )) is small it follows from Theorem 22.1 ∞ i (more precisely its local version stated in Remark 22.1) together with (22.1) that 1 − 2δ

Vol(Bpi (r, M∞ )) j

Vol0 (B0 (r, Rn ))

1 + 2δ.

(22.2)

284

K. Fukaya

We can then apply the method of Perelman that appeared in the proof of Theorem 21.6. It may43 imply that Bpi ((1−)r, M∞ ) is contractible in Bpi (r, M∞ ). This will imply that Mi j j is homotopy equivalent to M in a way similar to the proof of Theorem 3.5 in Section 15. Using control surgery in a way similar to [122] we can prove that Mi is homeomorphic to M. The proof of Theorem 22.2 is not worked out in so much detail in [46]. However we do not need to worry about it at all now, since Cheeger–Colding [29] improved Theorem 22.2 as follows. T HEOREM 22.3 [29, Theorem A.1.12]. In the situation of Theorem 22.1, Mi is diffeomorphic to M∞ for large i. We discuss its proof later in this section. Theorem 22.3 together with Theorem 21.8 immediately imply the following sharpening of Theorem 21.7. (We stated it as Theorem 2.6 in Section 2.) C OROLLARY 22.4 (Cheeger–Colding [29, Theorem A.1.10]). There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n , then M is diffeomorphic to a sphere. Remark 22.3. We remark Theorem 21.8 is used in the proof of Corollary 22.4. The proof of Theorem 21.8 we sketched in the last section uses Theorem 21.7. However we can avoid it as follows. Let RicciMi (n − 1), Vol(Mi ) Vol(S n ) − i , where i → 0. We may assume that Mi converges to a metric space X. Then, by Theorem 23.11, X is isometric to a metric suspension SY , where SY is defined in Example 23.1(3). Using the assumption on Mi and Theorem 22.5, we can show that the tangent cone Tx X of X at any point x ∈ X is Rn . Therefore, since X = SY , it follows that Y = S n−1 . Hence X = S n (isometric) as required. Remark 22.4. The assumption of Theorem 22.2 plus an additional assumption RicciMi λ implies that the Riemannian metric of Mi converges to the one of M in C 1,α -topology (after identifying the manifolds by an appropriate diffeomorphism). ([46, Theorem 0.6].) We now explain some of the ideas of the proof of Theorem 22.1. The main part of the proof is the proof of (2) of the following theorem. T HEOREM 22.5 [46, Theorem 0.8 and Corollary 2.19]. Let M be an n-dimensional Riemannian manifold with RicciM −λ and p ∈ M. (1) If Vol(Bp (1, M)) Vol(B0 (1, Rn )) − , then we have dGH Bp (1, M), B0 (1, Rn ) < τ (, λ|n). 43 I wrote “may” here since Perelman did not state this result explicitly and only says that “The Main Lemma can obviously be modified. . . ” at [114, p. 300]. Indeed it is very likely so. But I did not check it in detail. By the way, Colding quote [119] in place of [114] at [46, p. 478] just before Theorem 0.4. I believe it is a misprint.

Metric Riemannian geometry

285

(2) If dGH (Bp (1, M), B0 (1, Rn )) < , then we have Vol Bp (1, M) Vol B0 (1, Rn ) − τ (, λ|n). An argument to show Theorem 22.1 by using Theorem 22.5(2) is omitted. Let us sketch how to prove Theorem 22.5(2). We will discuss the proof of Theorem 22.5(1) in the next section. We only show the following version. L EMMA 22.6. If M satisfies dGH (Bp (2R, B), B0 (2R, Rn )) < and RicciM −λ then we have Vol Bp (1, M) 1 − τ (, λ, 1/R|n) Vol B1 (1, Rn ) . The argument to show Theorem 22.5(2) using Lemma 22.6 is tricky but technical. (See [46, p. 494].) (Note the inequality of opposite direction (1 − τ ) Vol Bp (1, M) Vol B1 (1, Rn ) is a consequence of Theorem 5.2.) Theorem 22.5 looks similar to Theorem 21.8. The proof of Lemma 22.6 also is similar. We first need a result corresponding to Theorem 21.13. In the proof of Theorem 21.13 we consider the function x → cos d(p, x) in case there exists q with d(p, q) π − δ. Here i , i = 1, . . . , n, instead. we consider the following function b+ n Let Φ : B0 (2R, R ) → Bp (2R, M) be an -Hausdorff approximation. Let qi = i

Φ(0, . . . , 0, 1, 0, . . . , 0) ∈ M. We put bi (x) = d(x, qi ) − d(p, qi ),

(22.3)

and study it in the ball Bp (1, M). We remark that bi may be regarded as an approximation of the Busemann function (Definition 16.6). In the proof of the Cheeger–Gromoll splitting Theorem 16.4, subharmonicity of the Busemann function is the main point. We choose ρ with 1 - ρ - R. We consider bi : Bp (ρ, M) → R such that bi = 0, bi = bi

(22.4a) on ∂Bp (ρ, M).

(22.4b)

In the case of Euclidean space, the Busemann function is nothing but a linear function. So we compare bi with a linear function. We put gi (v, t) = bi (v (t)). P ROPOSITION 22.7. For r α < 1, we have bi − bi L2 (Bp (1,M)) τ, 1

dgi (v, ·) gi (v, α) − gi (v, 0)

(r) −

< τ,

dt α v∈SBp (1,M)

(22.5a) (22.5b)

286

K. Fukaya

∇bi , ∇bj − δij < τ,

(22.5c)

Hess(bi ) < τ.

(22.5d)

Bp (1,M)

Bp (1,M)

Here τ = τ (λ, ρ/R, 1/ρ|n) and L2 is the L2 -norm up to the first derivative. 1

The proof of (22.5a) is based on Li–Shoen’s Poincaré inequality [100] (estimate of the first eigenvalue of Bp (ρ, M)), and the proof of (22.5d) is based on the Bochner– Weitzenbeck formula (19.1) and Cheng–Yau’s gradient estimate [42]. Then (22.5b) follows in a way similar as the proof of Theorem 21.9. We can use it to prove (22.5c). We put Φ˜ = (b1 , . . . , bn ) : Bp (1, M) → Rn . (22.5a), (22.5b) imply that it induces an τ -Hausdorff approximation to B0 (1, Rn ). (22.5c) implies that Φ˜ almost preserves volume. To complete the proof of Lemma 22.6 we need to show that Vol(B0 (1, Rn ) \ ˜ Φ(Bp (1, M))) is small. We can prove it as follows.44 Using (22.5) we can find a point ˜ 0 )) = {p0 }. (See [26, pp. 53–54] for the proof of p0 ∈ Bp (1/2, M) such that Φ˜ −1 (Φ(p this fact.) On the other hand, since Φ˜ is a τ -Hausdorff approximation, it follows that n n ˜ Φ(∂B p (1, M)) ⊂ B1+τ (0, R ) \ B1−τ (0, R ). Hence Φ˜ ∗ : Hn Bp (1, M), ∂Bp (1, M); Z2 → Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 is well defined. Note Hn Bp (1, M), ∂Bp (1, M); Z2 ∼ = Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 ∼ = Z2 . ˜ 0 )) = {p0 } we can show Using Φ˜ −1 (Φ(p Φ˜ ∗ : Hn Bp (1, M), ∂Bp (1, M); Z2 → Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 ˜ p (1, M)) ⊃ B1−τ (0, Rn ). This completes the proof of is nonzero. This implies Φ(B Lemma 22.6. We next sketch the proof of Theorem 22.3 given in [29, Appendix A]. As is mentioned there this proof is similar to the proof by Cheeger [25] of his finiteness theorem using the diffeotopy extension theorem (which we explained briefly in Section 6). 44 Here we follow [26, pp. 53–54]. Colding’s argument in [46] is a bit different.

Metric Riemannian geometry

287

Let us begin with a definition. Let Z be a complete metric space and , r > 0. (n is a positive integer.) D EFINITION 22.1. We say that Z satisfies the R,r,n condition if for each x ∈ Z there exists s < r such that dGH Bx (s, Z), B0 (s, Rn ) < s.

(22.6)

T HEOREM 22.8 (Cheeger–Colding [29, Theorems A.1.2, A.1.3]). For each n there exists n , independent of r, such that the following holds. If Z satisfies the R,r,n condition with < n , then, for each s < r, we can associate a smooth Riemannian manifold Z(s) with the following properties: (1) There exists a homeomorphism ΦZ,s : Z → Z(s) which is C 1−τ (|n) -Hölder continuous. Namely C −1 d(x, y)1+τ (|n) d ΦZ,s (x), ΦZ,s (y) Cd(x, y)1−τ (|n)

(22.7)

for each x, y ∈ Z. Moreover ΦZ,s is an sτ (|n) Hausdorff approximation. (2) Z(s) is ‘well-defined’ and ‘independent’ of s in the following sense. If u s then there exists a diffeomorphism which is C 1−τ (|n) -Hölder continuous in a way independent of t, u. Namely we have C −1 d(x, y)1+τ (|n) d ΦZ,u,s (x), ΦZ,u,s (y) Cd(x, y)1−τ (|n) ,

(22.8)

where C is independent of u, s, x, y. Moreover ΦZ,u,s is an sτ (|n) approximation and satisfies d ΦZ,u,s ◦ ΦZ,s (x), ΦZ,u (x) < sτ (|n).

(22.9)

(3) If Z is a Riemannian manifold then we may choose ΦZ,s to be a diffeomorphism for sufficiently small s. (4) There exists δ(n, r) > 0 depending n and r such that if Z, Z both satisfy R,r,n condition with < n , and if dGH (Z, Z ) < δ(n, r) then there exists a diffeomorphism Ψ : Z(r/2) → Z (r/2) such that e−τ (,δ|r,n)

d(Ψ (x), Ψ (y)) eτ (,δ|r,n) , d(x, y)

d Ψ ◦ ΦZ,r/2 (x), ΦZ ,r/2 (x) < τ (, δ|r, n).

(22.10)

(22.11)

To apply Theorem 22.8 for the proof of Theorem 22.3 we need the following P ROPOSITION 22.9. Let Mi be a sequence of n-dimensional Riemannian manifolds and let M∞ be another Riemannian manifold of the same dimension. We assume

288

K. Fukaya

limGH i→∞ Mi = M∞ . Then for each there exists r such that Mi for large i and M∞ satisfy the R,r,n condition. Proposition 22.9 and Theorem 22.8 immediately imply Theorem 22.3. Let us prove Proposition 22.9. Under the assumption we have r = r(µ) for each µ such that 1−µ

Vol(Bp (r, Mi )) 1+µ Vol(B0 (r, Rn ))

for large i and i = ∞ and any p ∈ Mi . (See (22.2).) Then we apply Theorem 5.2 to obtain 1−µ

Vol(Bp (s, Mi )) 1+µ Vol(B0 (s, Rn ))

for any s r. We now apply Theorem 22.5(1) after scaling to obtain dGH Bp (s, Mi ), B0 (s, Rn ) < sτ (µ|r, n) as required. (Note that since we scale the metric by a factor 1/s > 1/r, the curvature will be Ricci −(1 − n)r 2 . So the curvature assumption in Theorem 22.5 is satisfied if r is small enough.) We remark that the independence of n of r in Theorem 22.8 played a key role here. We now prove of Theorem 22.8. Let 100s < r. We will construct Z(s) first. We remark that by using assumption (22.6) we can find subsets {xi | i ∈ I } ∈ Z such that

Bxi (s, Z) = Z,

(22.12a)

i∈I

! i ∈ I | Bxi (30s, Z) ∩ Bxj (30s, Z) = ∅ N (n)

(22.12b)

for each j ∈ I . Here N (n) is independent of i, s. Let ϕi : Bxi (100s, Z) → B0 (100s, Rn ) be a τ (|n)s-Hausdorff approximation. We have a τ (|n)s Hausdorff approximation ϕi : B0 (100s, Rn ) → Bxi (100s, Z) such that dist(ϕi ◦ ϕi , id) < τ (|n)s and dist(ϕi ◦ ϕi , id) < τ (|n)s. We consider

ϕj i = ϕj ◦ ϕi B

0 (10s,R

n)

: B0 (10s, Rn ) → B0 (35s, Rn )

(22.13)

for i ∩ j with Bxi (30s, Z) ∩ Bxj (30s, Z) = ∅. It satisfies

d ϕj i (x), ϕj i (x) − d(x, y) < τ (|n)s. We here remark the following simple lemma.

(22.14)

Metric Riemannian geometry

289

L EMMA 22.10. If ϕj i satisfies (22.14) then there exists ψj i : B0 (10s, Rn ) → B0 (35s, Rn ) satisfying (22.14) and e−τ (|n)

d(ψj i (x), ψj i (x))

d(x, y) d ϕj i (x), ψj i (x) < sτ (|n),

eτ (|n) ,

(22.15a) (22.15b)

|ψj i |C k < s −k Ck,n .

(22.15c)

Here Ck,n depends only on k and n. The proof is an elementary smoothing argument. We want to construct a smooth manifold by using ψj i as a coordinate transformation. ◦ ψ = ψ but the following holds if B (20s, Z) ∩ B (20s, Z) ∩ It does not satisfy ψkj xi xj ji ki Bxk (20s, Z) = ∅: ◦ ψj i (x), ψki (x) sτ (|n) d ψkj

(22.16)

for x ∈ B0 (20s, Rn ). We can now use the argument of [25] to approximate ψj i by ψj i which satisfies (22.15) and ψkj ◦ ψj i = ψki .

(22.17)

(Note that the number of steps we need to take to achieve (22.17) is controlled by (22.12b).) We thus constructed a manifold Z(t) whose coordinate transformation is ψj i . We can use a partition of unity to modify the standard metric on Rn so that it is compatible with ψj i . Hence Z(t) is a Riemannian manifold. We will construct ΦZ,s : Z → Z(s) later. At this stage we have ΨZ,s : Z → Z(s) which is an sτ (|n)-Hausdorff approximation. We next show the ‘well-definedness’ property (2). We first consider the case u ∈ [s/2, s]. Let us suppose we have Z(u) for u s. We use the symbol ˜ for points, maps, etc. used to construct Z(u). (Namely we write ϕ˜ i˜ , x˜i˜ , etc.) Let Bx˜i˜ (30u, Z) ∩ Bxj (30s, Z) = ∅. We define Ψj i˜ : B0 (20u, Rn ) → B0 (30s, Rn ) by Ψj i˜ = ϕj ◦ ϕ˜ ˜ . It satisfies i

d Ψ ˜ (x), Ψ ˜ (y) − d(x, y) τ ()s 2τ ()u. ji ji Hence we can approximate it by a smooth map Φ ˜ satisfying (22.15c). It is almost comji

patible with the coordinate transformations ψj i ,ψ˜ j˜i˜ . Hence again by an argument similar to [25] (or by using the center of mass technique) we can approximate it by a diffeomorphism Φj i˜ (x) which is exactly compatible with the coordinate transformation. We thus obtain ΦZ,s,u , if u ∈ [s/2, s]. It is also an sτ (|n)-Hausdorff approximation. Let ΦZ,u,s be

290

K. Fukaya

the inverse of it. We remark that we have an inequality e−τ (|n)

d(ΦZ,u,s (x), ΦZ,u,s (y)) eτ (|n) d(x, y)

(22.18)

which is sharper than (22.10) in case u ∈ [s/2, s]. We remark here that the proof of Theorem 22.8(4) is almost the same as this argument. (So we do not discuss it.) Now we continue the proof of (2) for the general u, s. We may assume u = 2−k s and put ΦZ,u,s = ΦZ,u,2u ◦ · · · ◦ ΦZ,s/2,s .

(22.19)

It is a diffeomorphism. We will check (22.10). Let ρ > 0. We first remark that ΦZ,a,b is a bτ (|n)-Hausdorff approximation for a b. (This is because if b = 2k a then ΦZ,a,b is a k −j j =0 τ (|n)2 b-Hausdorff approximation.) We first take such that e−ρ

−λi with λi → 0 and let (X, p) = limi→∞ (Mi , pi ). Suppose X contains a line. Then X is isometric to the direct product R × X . Remark 23.1. The following slightly more general statement is proved. pGH Let RicciMi −λi with λi → 0, and (X, p) = limi→∞ (Mi , pi ). We assume X ∼ = Rk × Y ∼ and that Y contains a line. Then Y = R × Y . Before explaining the outline of the proof, we mention several of its applications. One important application is Theorems 19.12 and 10.5, which we explained already. To state other applications, we need some definitions. D EFINITION 23.1. A measured metric space (X, µ) is a pair of a metric space X and a Borel measure µ on it. In this article we always assume that µ(X) = 1. For a pointed measured metric space (X, p, µ) we assume µ(Bp (1, X)) = 1. For a Riemannian manifold M we use the renormalized volume form µM = ΩM / Vol(M) and regard it as a measured metric space (unless another measure is specified explicitly). For a pointed Riemannian manifold (M, p), we use the renormalized volume form µM = ΩM / Vol(Bp (1, M)). D EFINITION 23.2 [54]. A sequence of measured metric spaces (Xi , µi ) is said to converge to (X, µ) with respect to the measured Gromov–Hausdorff topology which we write as limmGH i→∞ (Xi , µi ) = (X, µ), if there exists a sequence of i -Hausdorff approximations

292

K. Fukaya

ϕi : Xi → X with i → 0, which are Borel measurable and such that, for any continuous function f on X, we have

lim

i→∞ Xi

(f ◦ ϕi ) dµi =

f dµ. X

The pointed measured Gromov–Hausdorff convergence is defined in the same way. (To be precise we need a net in place of a sequence to define a topology. It is an obvious modification and is omitted.) Remark 23.2. In [75, Chapter 3 12 D], Gromov defined a notion of λ convergence for measured metric spaces. It is similar to, but slightly different from, measured Gromov– Hausdorff topology defined above. Namely there is a situation where the support supp µ of the limit measure is different from X. In that case for λ convergence the limit of (Xi , µi ) is (supp µ, µ), and is different from the limit (X, µ) of the measured Gromov– Hausdorff topology. However if (Xi , µi ) = (Mi , µMi ) is a Riemannian manifold and if RicciMi −(n − 1), then the support of the limit measure is always X itself. (We can prove it using the Bishop–Gromov inequality.) So the two definitions coincide to each other. Measured Gromov–Hausdorff convergence was introduced to study spectra of the Laplace operator. We mention it later. L EMMA 23.2 [54]. If limGH i→∞ Xi = X, and if µi is a probability Borel measure on Xi , then there exists a subsequence ki such that limmGH i→∞ (Xki , µki ) = (X, µ). The proof is elementary. We remark that the limit measure µ depends on the choice of the subsequence in general. f In fact, let us consider T 2 = S 1 × S 1 with metric g = dt 2 + 2 f (t)2 ds 2 , where f : S 1 → R+ is a smooth function. Then (T 2 , g ) converges to S 1 with standard metric and measure f dt with respect to the measured Gromov–Hausdorff topology. On the other hand, the limit in the Gromov–Hausdorff distance is independent of f . We denote by Sn (D) the set of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Diam(M) D. We denote by Sn (∞) the set of n-dimensional pointed Rie¯ n (D), S ¯ n (∞) be the closure mannian manifold (M, p) such that RicciM −(n − 1). Let S of Sn (D), Sn (∞) with respect to the Gromov–Hausdorff distance, the pointed Gromov– Hausdorff distance, respectively. ¯ n (D). We next define the singularity set and the regular set of a length space X ∈ S We recall that the sequence (X, Ri dX , x) with Ri → ∞ always has a subsequence such that (X, Ri dX , x) converges with respect to the pointed Gromov–Hausdorff distance (Proposition 16.2). However the limit is not unique. (Such an example is constructed in [29, Section 8].)

Metric Riemannian geometry

293

D EFINITION 23.3. We say that Tx X is a tangent cone of X at x if there exists a sequence Ri → ∞ such that (X, Ri dX , x) converges to (Tx X, o) with respect to the pointed Gromov–Hausdorff distance.45 ¯ n (∞). We say that a point x ∈ X is in Rk if Rk is a tangent D EFINITION 23.4. Let X ∈ S cone Tx X of x. $ We say x is regular if it is in R = k Rk . Otherwise it is said to be singular and we denote by S the set of all singular points. Remark 23.3. This definition coincides with S(X) in Definition 17.7 by Otsu–Shioya in case when X is an Alexandrov space, because of Theorem 22.5. One of the main results by Cheeger–Colding on the limit space X (in the collapsing situation) is the following T HEOREM 23.3 (Cheeger–Colding [29]). µ(S) = 0 for any limit measure µ. $ Remark 23.4. (1) We remark that Theorem 23.3 implies µ(X \ k Rk ) = 0, but does not imply the existence of k such that µ(X \ Rk ) = 0. (2) In Theorem 23.3 the limit measure µ is used. We do not know how to use the Haus¯ n (D) is an dorff measure since it is not known whether the Hausdorff dimension of X ∈ S integer or not. Here is some very brief idea how a statement like Theorem 23.3 follows from Theo¯ n (D) such that Tx X is a Euclidean rem 23.1. We want to find many points x on X ∈ S space. A naive idea to find such a point may be as follows. First we consider a minimal geodesic xy1 and take an interior point on it and put it x1 . Then any tangent cone Tx1 X contains a line and hence splits as Tx1 X ∼ = R × X1 . We may next take a point x2 near o ∈ Tx1 X which is a midpoint of the minimal geodesic. Then Tx2 X1 contains a line and hence Tx2 X1 splits. This process stops after finitely many stages since we can estimate the dimension of the tangent cone by the Bishop–Gromov inequality. Thus we find near x some kind of ‘point’ for which a tangent cone is Rk . This argument however is too much naive to prove Theorem 23.3. So we need to work more seriously. See [29, Section 2]. We remark that Theorem 23.3 can be applied also to the collapsing situation. Namely it can be applied to the limit X of Mi such that Vol(Mi ) → 0. Several other results are proved by Cheeger–Colding in [29,30]. Nevertheless there are yet many things unclear in the collapsing situation. (In other words, the result in the collapsing situation does not seem to be in the final form.) So we do not discuss it here. (We will discuss one of the main results of [31] later.) 45 I am sorry that this terminology is inconsistent with one in Definition 16.4, where T X is called tangent cone x when any such sequence (X, Ri dX , x) converges to (Tx , 0). In this section we follow Cheeger–Colding and use this terminology. In Definition 16.4 we followed Burago–Gromov–Perelman.

294

K. Fukaya

In the noncollapsing case, Cheeger–Colding obtained more precise results. We discuss some of them here. The following theorem is a generalization of Theorem 22.1. We denote by Sn (D, v) the set of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Diam(M) D, Vol(M) v. We also denote by Sn (∞, v) the set of n-dimensional pointed Riemannian manifold (M, p) such that RicciM −(n − 1), Vol(Bp (1, M)) ¯ n (∞, v) be the closure of Sn (D, v), Sn (∞, v) with respect to the ¯ n (D, v), S v. Let S Gromov–Hausdorff distance, the pointed Gromov–Hausdorff distance, respectively. pGH

T HEOREM 23.4 [29, Theorem 5.9]. Let Mi ∈ Sn (∞, v). We assume that limi→∞ (Mi , pi ) = (X, p). Then for any R we have lim Vol Bpi (R, Mi ) = Hn Bp (R, X) .

i→∞

Here Hn denotes the n-dimensional Hausdorff measure. ¯ n (∞, v) then the Hausdorff dimension of X is n. MoreC OROLLARY 23.5 [29]. If X ∈ S over any limit measure µ is equal to a multiple of the n-dimensional Hausdorff measure. Corollary 23.5 follows from Theorem 23.4 easily. We explain an idea of the proof of Theorem 23.4 later in this section. ¯ n (∞, v) and x ∈ X. Then any tangent cone T HEOREM 23.6 [28, Theorem 5.2]. Let X ∈ S Tx X is isometric to a cone CY of some length space Y of diameter π . Remark 23.5. This result is a kind of generalization of the corresponding result Theorem 17.16 on Alexandrov spaces. However it is not asserted that CY is unique. Actually there is a counter example [29, 8.41]. The conclusion of Theorem 23.6 does not hold in the collapsing situation [29, 8.95]. To prove Theorem 23.6 we need another kind of comparison theorem, which we will explain later. To state the next result we need a definition. ¯ n (∞, v). We say that x ∈ R if every tangent cone Tx X D EFINITION 23.5. Let X ∈ S satisfies dGH B0 (1, Tx X), B0 (1, Rn ) < . We put S = X \ R . Remark 23.6. (1) Using Theorems 22.5, 23.4, we can prove that there exists δ such that Vol(Bx (r, X)) (1 − δ) Vol(B0 (r, Rn )) implies x ∈ R . Thus Definition 23.5 is equivalent to Sδ (X) in Definition 17.7. (2) We can easily see that if < , then R is contained in the interior Int R of R .

Metric Riemannian geometry

295

¯ n (∞, v), we can easily show R = Rn . Using it, we can easily (3) In the$case of X ∈ S . prove S = >0 S , R = >0 R . The following is an analogue of Theorems 17.2 and 17.22. ¯ n (∞, v) and if T HEOREM 23.7 [29, Theorem 5.14]. There exists 0 (n) such that if X ∈ S < 0 (n), then there exists a smooth Riemannian manifold Z() and a homeomorphism Φ : Z() → Int R such that C −1 d(x, y)1+τ (|n) < d Φ (x), Φ (y) < Cd(x, y)1−τ (|n) . Theorem 23.7 actually follows easily from Theorem 22.8. Namely we find, for each > and x ∈ R , a positive number r such that dGH (Bx (r, X), B0 (r, Rn )) < . Therefore any compact subset of Int R satisfies condition R ,r,n for some r. Theorem 23.7 then follows from Theorem 22.8. We next prove Theorem 23.4. For simplicity of notation we assume that X is compact. We first prove Hn (S) = 0. Let µ be a limit measure. We remark that there exists C1 , C2 such that for 0 < r 1 we have C1 r n Vol Bp (r, Mi ) C2 r n , C1 r n µ Bp (r, X) C2 r n .

(23.1a) (23.1b)

In fact (23.1a) is a consequence of the Bishop–Gromov inequality and Vol(Mi ) v > 0. Then (23.1b) follows, since µ is a limit measure. By Theorem 23.3 we have µ(S) = 0. Therefore by (23.1b) and the definition of Hausdorff measure we have Hn (S) = 0. It follows that lim Hn (Int R ) = Hn (X).

(23.2)

→0

We can take the disjoint union of finitely many balls U = Hn (X \ U ) < τ (),

n n

ωn rj − H (X)

< τ (),

$ j

Byj (rj , X) ⊂ R such that (23.3a) (23.3b)

j

dGH Byj (rj , X), B0 (rj , Rn ) < 2rj ,

(23.3c)

where ωn = Vol(B0 (1, Rn )). Here (23.3c) $is a consequence of yj ⊂ R . Then, for large i, we have a disjoint union of balls U,i = j Byj,i (rj , Mi ) ⊂ Mi with dGH Byj,i (rj , Mi ), B0 (rj , X) < 3rj .

(23.4)

296

K. Fukaya

Therefore, by (23.3b), (23.3c), (23.4), and Theorem 22.5 we have

n

H (U ) − Vol(U,i ) < τ (, 1/i|n).

(23.5)

We thus proved Hn (X) Vol(Mi ) + τ (, 1/i|n). To prove the opposite inequality, we take finitely many balls Bza (ta , X) such that X⊆

Byj (rj , X) ∪

j

Bza (ta , X),

(23.6a)

j

tan τ ().

(23.6b)

j

Then for large i we find za,i such that Mi =

Byj,i (rj , Mi ) ∪

j

Bza,i (ta,i , Mi ).

(23.7)

j

Since Vol Bza,i (ta,i , Mi ) < Cn tan , j

it follows that Vol(Mi \ U,i ) τ (|n).

(23.8)

Therefore, Hn (X) Vol(Mi ) − τ (, 1/i|n), as required.

We now sketch the proof of Theorem 23.1. We start with the following situation. (A) M is a Riemannian manifold with RicciM −λ with small λ. (B) We assume dGH (BL (z, M), BL (z , X)) < ρ/10 and there is a line containing z . (Here L is large.) (C) Let p, q ∈ M with d(p, q) = 2L with large L. (D) d(z, pq) ρ/3, |d(z, p) − L| ρ/3, |d(z, q) − L| ρ/3. Here M = Mi , where Mi is as in Theorem 23.1 for large i. Such pair of points p, q exists because of (B). We want to find a length space X such that Bz (R, M) is close to a D ball B(x ,0) × (R, R × X ) in R × X with respect to the Hausdorff distance. (Here 1 - R - L.) We use the following function which is an approximation of the Busemann function: b+ (x) = d(p, x) − d(p, z),

b− (x) = d(q, x) − d(q, z).

(23.9)

Metric Riemannian geometry

297

The argument to control them is similar to the proof of Proposition 22.7. However our problem is a bit different from the situation of Proposition 22.7 where Hausdorff approximation is given by assumption. Our situation is similar to Theorem 22.5(1), where we use another assumption (which was the almost maximality of the volume in case of Theorem 22.5(1)) to find Hausdorff approximation. In our case, we use the following theorem by Abresch–Gromoll to obtain some information on b± and improve it by using a similar argument as in the proof of Proposition 22.7. To state the result by Abresch–Gromoll we need a notation. D EFINITION 23.6. For x, p, q ∈ M, an excess E(x; p, q) is by definition E(x; p, q) = d(x, p) + d(x, q) − d(p, q). T HEOREM 23.8 (Abresch–Gromoll [2]). If RicciM −(n−1)λ, d(z, p) L, d(z, q) L and if E(z; p, q) < ρ, then E(x; p, q) < τ (ρ, λ, 1/L|n, R) for any x ∈ Bz (R, M). Remark 23.7. Abresch–Gromoll stated Theorem 23.8 in the case E(z; p, q) = ρ namely the case z ∈ pq. The above form is a modification by Cheeger–Colding [28, Proposition 6.2]. ([26, Theorem 9.1].) Remark 23.8. Abresch–Gromoll used Theorem 23.8 to show the following Theorem 23.9. It seems that Theorem 23.8 is the first comparison theorem established assuming conditions on Ricci curvature only. For a length space M and B ⊂ A ⊆ M we denote by Diam(B ⊂ A) the following number sup {the length of the shortest curve joining p and q in A}. p,q∈B

T HEOREM 23.9 (Abresch–Gromoll [2]). If M is a complete manifold with RicciM 0, inf KM > −∞, and Diam Sp (3R, M) ⊆ Ap (2R, 4R; M) C/R, then M is homotopy equivalent to an interior of a compact manifold with boundary. Theorem 23.9 is proved by Theorem 23.8 and Morse theory of the distance function in a way similar to Theorem 14.6. Let us go back to the discussion of the proof of Theorem 23.1. Theorem 23.8 and b+ + b− = E(x; p, q) − E(z; p, q) imply −ρ b+ + b− τ L−1 , ρ, λ|R, n .

(23.10)

298

K. Fukaya

Using the fact that b± is “almost subharmonic” we have the following formula (23.12). We define b+ : Bz (R, M) → R by b+ = 0, b+ = b+

(23.11a) on ∂Bz (R, M).

(23.11b)

Then, we can prove b+ − b+ L2 (Bz (R,M)) τ L−1 , ρ, λ|R, n . 1

(23.12)

(Here the right-hand side will become small by taking L large, λ, ρ small.) We now consider the Bochner formula 1 |∇b+ | = | Hess b+ |2 + Ricci(∇b+ , ∇b+ ). 2

(23.13)

We remark |∇b+ | = 1. Hence using (23.12) the integral of the left-hand side of (23.13) is small. Since RicciM −(n − 1)λ it follows from (23.13) Bz (R,M)

Bz (R,M)

| Hess b+ | τ L−1 , ρ, λ|R, n ,

(23.14)

|∇b+ | − 1 τ L−1 , ρ, λ|R, n .

(23.15)

We put X = b−1 + (0). Now we will use (23.14), (23.15) to show that Bx (R, M) is close to a R ball in X × R with respect to the Gromov–Hausdorff distance as follows. Let us take y, z ∈ Bx (R, M). Let y0 , z0 ∈ X such that d(y, y0 ) = d(y, X ),

d(z, z0 ) = d(z, X ).

We will prove

d(y, z)2 − d(y0 , z0 )2 − b+ (y) − b+ (z) 2 τ L−1 , ρ, λ|R, n .

(23.16)

(23.16) obviously implies that y → (y0 , b+ (y)) : Bx (R, M) → B(x,0) (R, X × R) is a Hausdorff approximation and hence dGH Bx (R, M), B(x,0) (R, X × R) τ L−1 , ρ, λ|R, n , which is enough to complete the proof of Theorem 23.1. Let us sketch the proof of (23.16). For simplicity we take b+ (z) = 0 and z = z0 . Let : [0, l] → M be a minimal geodesic joining y0 to y. We put Q(t) = d((t), z). Let γt : [0, Q(t)] → M be a minimal geodesic joining z to (t). (See Figure 23.1.)

Metric Riemannian geometry

299

Fig. 23.1.

We put ht (s) = b+ (γt (s)). We remark that d 2 ht (s) = (Hess b+ ) γ˙t (s), γ˙t (s) - 1. 2 ds

(23.17)

On the other hand ht (Q(t)) = b+ (γ (t)) is almost equal to t.46 Hence dht t ds (s) − Q(t) - 1.

(23.18)

(Here we remark that (23.17), (23.18) do not hold pointwise but only after integrating over some domain. We omit the technical difficulty which arises from this point.) By the first variational formula, we have . dht dQ (s) = γ˙t (s), ∇b+ = cos y0 γ (t)z = (t). ds dt

(23.19)

. (Here and hereafter = means “almost” equal.) Hence Q(t) “almost” satisfies the following differential equation: dQ . t = . dt Q(t)

(23.20)

# The solution of (23.20) with initial value Q(0) = d(y0 , z) is Q(t) = d(y0 , z)2 + t 2 . Hence at t = b+ (y) we have (23.16) with z = z0 . (See [26, Chapter 9] or [28, Section 6] for the details of the proof.) Here we say a few words about the proof of Remark 23.1. In this situation we can take not only p, q but also pi , qi , i = 1, . . . , k. Namely p, q are points close to the line on Y and pi , qi are taken as points close to the point on the coordinate axis of Rk . Using them we 46 We can find (d 2 /dt 2 )(b ◦ γ ) is small in the same way as (23.17). Moreover (d 2 /dt 2 )(b ◦ γ )(0) is close + + to 1 by definition and (23.15).

300

K. Fukaya

obtain b+ together with bi+ , i = 1, . . . , k. They all satisfy (23.14), (23.15). Moreover we . j have ∇bi+ , ∇b+ = δij where b0+ = b+ . We define Φ : X → Rk+1 by Φ = (b0+ , . . . , bk+ ). Using it we can construct a pointed Hausdorff approximation X → Φ −1 (0) × Rk in a way similar to the proof of Theorem 23.1. See [29, pp. 425–426], where similar arguments appears. In Section 20 we reviewed several results obtained by the L2 comparison theorem where we compared a manifold with round sphere. In Section 22 we used L2 comparison theorem where the model space was flat Euclidean space. In the proof Theorem 23.1, we compared a manifold with direct product R × X . In [28], Cheeger–Colding developed a comparison theorem where the model space is a warped product (hereafter we call it the warped product comparison theorem) and gave various applications. We first review some of its applications. One of its applications is Theorem 22.5(1). The following is closely related to it. (Theorem 22.5(1) corresponds to the case when Y = S n−1 .) Cheeger–Colding called this theorem ‘volume cone implies metric cone theorem’. T HEOREM 23.10 (Cheeger–Colding [28]). For each there exists δ = δ(, n) with the following property. Let M be an n-dimensional Riemannian manifold with RicciM −δ(n − 1). We assume Vol(Bp (1, M)) Vol(B0 (1, Rn )) (1 + δ) . Vol(Sp (1, M)) Vol(S0 (1, Rn )) Then there exists a length space Y with Diam(Y ) π such that dGH Bp (1, M), B0 (1, CY ) . We remark that CY = ([0, ∞) × Y )/ ∼ where (0, x) ∼ (0, y) with metric defined in Definition 17.2. Another application of the warped product comparison theorem is Theorem 23.11. To state it we define a warped product. D EFINITION 23.7. Let (X, gX ) be a Riemannian manifold and f : (a, b) → R+ be a smooth function. Then the warped product (a, b) ×f X is by definition a product (a, b) × X equipped with the metric dr 2 ⊕ f (r)2 gX , where r is the coordinate of the interval (a, b). We need to define the warped product for general length space also. Let X be a length space and f : (a, b) → R+ be a smooth function. Let : [α, β] → (a, b) × X be a path which is, say, Lipschitz continuous. We put (t) = (r(t), X (t)). We may change the parameter so that X (t) : [α, β] → X is parameterized by arc length. We then define the length L() of : [α, β] ×f X by

β

L() =

2 2 (dr/dt)(t) + f r(t) dt.

α

We thus defined the length space (a, b) ×f X.

Metric Riemannian geometry

301

E XAMPLE 23.1. (1) The simplest case is f ≡ 1. Then the warped product is the direct product. (2) If (a, b) = (0, ∞) and f (r) = r, then the warped product (0, ∞) ×r X is the cone CX minus 0. If moreover X = S n−1 then it is Rn \ {0}. (3) We take (a, b) = (0, π) and f (r) = sin r. In this case the warped product (0, π) ×sin r X is called the metric suspension SX. In particular the metric suspension SS n−1 of a round sphere S n−1 is the round sphere S n . T HEOREM 23.11 [28, Theorem 5.14]. If RicciM (n − 1), dim M = n and if Diam(M) π − then there exists a length space X such that dGH (M, SX) < τ (|n). Remark 23.9. It is not true in general that M is homeomorphic (or homotopy equivalent to) SX. The counter examples are the ones by Anderson and Otsu we mentioned already. There are several other applications, for example, to the study of the cone at infinity. We omit it. We now explain the idea of the proofs of these theorems. The main idea is to use the warped product comparison theorem. To state it we need some preliminary discussion. We begin with a characterization of a warped product. Let f : (a, b) → R+ be a smooth function, we put F(r) =

r

f (t) dt, a

k(r) =

df (r). dr

(23.21)

L EMMA 23.12. Let X be a Riemannian manifold and M = (a, b) ×f X. Then we have Hess(F) = k(r)gM .

(23.22)

E XAMPLE 23.2. (1) In case M = R ×1 X the direct product. F is linear and k = 0. (2) In case M = Rn = CX \ 0 = (0, ∞) ×r X, we have F = r 2 /2 and k(r) = 1. If X = S n−1 , M = Rn then F(x1 , . . . , xn ) = 12 (x12 + · · · + xn2 ) and (23.22) is obvious. (3) In case M = (0, π) ×sin r S n−1 we have F(r) = −k(r) = cos r. Formula (23.22) is (21.17). Let us prove Lemma 23.12. We put ∂r = ∂/∂r. Hess(F)(∂r , ∂r ) = k is obvious since t → (t, p) is a geodesic. Let V be a vector filed of X, which we regard a vector field on M. We have [V , ∂r ] = 0. Since gM (V , V ) = f 2 gX (V , V ) it follows that −gM (∂r , ∇V V ) = gM (∇V ∂r , V ) = gM (∇∂r V , V ) = f kgX (V , V ). On the other hand, V (F) = 0, ∂r (F) = f . Therefore Hess(F)(V , V ) = −(∇V V )(F) = f 2 kgX (V , V ) = kgM (V , V ), as required.

302

K. Fukaya

The warped product comparison theorem is an ‘almost version’ of the following converse to Lemma 23.12. P ROPOSITION 23.13. If M is a Riemannian manifold F : M → (α, β) is a fiber bundle. Suppose that there exists a function which k : M → R such that Hessx (F) = k(x)gM .

(23.23)

We put X = {x ∈ M | F(p) = F(x)}. Then there exists a function f : (a, b) → R+ such that M∼ = (a, b) ×f X (isometry), r(x) f (t) dt, F(x) =

(23.24a) (23.24b)

r(p)

k(x) =

df r(x) . dr

(23.24c)

Here r : M ∼ = (a, b) ×f X → (a, b) is the projection to the first factor. We now state the warped product comparison theorem. Let M be a complete Riemannian manifold and K be a compact subset. We put r(x) = d(x, K) = inf y ∈ K | d(x, y) , AK (a, b, M) = x ∈ M | a < r(x) < b .

(23.25a) (23.25b)

Let f : (a, b) → R+ be a smooth function and we define F(r) and k(r) as in (23.21). We regard r as a function on AK (a, b, M) then F and k are functions on AK (a, b, M) as well. The following assumptions are a generalization of similar formulae we met several times already. For example, (23.14), (23.14) where k(r) = 0, and (21.17) where k(r) = cos r. A SSUMPTION 23.1. There exists F˜ : AK (a, b, M) → (a, b) such that sup |F˜ − F| , 1 Vol(AK (a, b, M)) 1 Vol(AK (a, b, M))

(23.26a) |∇ F˜ − ∇F| ,

(23.26b)

| Hess F˜ − kgM | .

(23.26c)

AK (a,b,M)

AK (a,b,M)

Theorem 23.14 asserts under Assumption 23.1 plus some more (which will follow), that AK (a, b, M) is Gromov–Hausdorff close to some warped product (a, b) ×f X. A SSUMPTION 23.2. M is an n-dimensional complete Riemannian manifold with KM −Λ. Diam(AK (a, b, M)) D. 0 < α < α, 0 < ξ < α − α .

Metric Riemannian geometry

303

For each x ∈ r −1 (a + α ) there exists y ∈ r −1 (b − α ) such that d (x, y) b − a − 2α + .

(23.27)

T HEOREM 23.14 (Cheeger–Colding [28, Theorem 3.6]). Under Assumptions 23.1 and 23.2, there exists a length space X such that dGH AK (a + α, b − α, M), d , (a + α, b − α) ×f X τ (|α , ξ, n, f, D). Remark 23.10. In Assumption 23.2 and Theorem 23.14 we use the symbol d for the metric of subsets of M. Note that the space AK (a, b, M) is not complete. So when we define the metric function d : AK (a, b, M) × AK (a, b, M) → R using the Riemannian metric, we need to be a bit careful. Namely for p, q ∈ AK (a, b, M) we need to take the infimum of the length of the curves joining them in a slightly larger domain. The metric d stands for such a metric. We do not define it since it is too technical. See [28, pp. 205–206]. Let us explain the idea of the proof of Theorem 23.14. Actually the idea is quite similar to one of the proof of (23.16) we discussed already. We take X = r −1 (a + α). To define a metric on it we consider a broken geodesic on a small neighborhood and take the infimum of the length of them. Now we construct the Hausdorff approximation Φ : AK (a + α, b − α, M) → (a + α, b − α) ×f X. Let y, z ∈ AK (a + α, b − α, M). We take y0 , z0 ∈ X so that d(y, y0 ) = d(y, X), d(z, z0 ) = d(z, X). We remark r(x) = d(x, X) − a − α . We put Φ(y) = r(y), y0 , and will prove that Φ is a Hausdorff approximation. We assume z0 = z for simplicity. Let : [0, l] → M be a minimal geodesic joining y0 to y. We put Q(t) = d((t), z). Let γt : [0, Q(t)] → M be a minimal geodesic joining z to (t). Actually there is a technical trouble here. Namely since AK (a + α, b − α, M) is not complete, we may not be able to take γt . By this reason, we need to take a broken geodesic. (See Figure 23.2.) However since this is a technical point, we forget it and assume that we can take γt . ˜ t (s)). By (23.26), ht ‘almost’ satisfies the differential equation We put ht (s) = F(γ d 2 ht . (s) = H ht (s) , ds 2

(23.28)

where H (c) = k(F˜ −1 (c)). We remark that (23.28) is an ordinary differential equation of second order and hence has unique solution under an appropriate boundary condition. Note . that ht (0) = F(a + α), ht (Q(t)) = F(t + a + α). Thus ht is determined by Q(t). (Precisely we have to say that ht is ‘almost’ determined by Q(t) since (23.28) is only ‘almost’ satisfied.) Moreover we have . dQ dht Q(t) = (t) ds dt

(23.29)

304

K. Fukaya

Fig. 23.2.

by the same reason as (23.19). Thus, (23.29) becomes a differential equation of first order on Q and is determined by f . We remark that Q satisfies an initial value condition Q(0) = d(z, y0 ). Therefore, the value of Q at t = r(y) − a − α is determined by this equation and the initial value d(z, y0 ). (Precisely speaking, we can only say the value of Q is almost determined.) By definition Q(r(y) − a − α) = d(y, z). Since it is almost determined by d(z, y0 ) and r(y) and r(z) (which we assumed to be zero for simplicity), it follows that Φ ‘almost’ preserves the length. The fact that a small neighborhood of the image of Φ contains (a + α, b − α) ×f X follows from (23.27). This is a sketch of the proof of Theorem 23.14. We now discuss applications of Theorem 23.14. We first show how we can use Theorem 23.14 to prove Theorem 22.5(1). Let us assume Vol(Bp (1, M)) Vol(B0 (1, Rn )) − and RicciM −λ. We put f (t) = t. Then k(t) ≡ 1, F(t) = t 2 /2. We need to check Assumptions 23.1, 23.2. Put r(x) = d(x, p). We calculate (2 − n) r 1−n = grad r 2−n · dn = r 2−n . (23.30) Sp (R,M)

Sp (R,M)

Bp (R,M)

Since Vol(Bp (1, M)) Vol(B0 (1, Rn )) − it follows from Lemma 21.1 that Sp (R,M) r 1−n n = cn Vol(S p (R, M))/ Vol(S0 (R, R )) is almost independent of R. Hence (23.30) im 2−n plies Bp (R,M)\Bp (δ,M) r is small. Namely r 2−n is almost a harmonic function on Bp (R, M) \ Bp (δ, M). (We remark that r 2−n is harmonic on Rn .) Then we have r 2−n = (2 − n) div r 1−n grad r = (2 − n)r 1−n r + (2 − n)(1 − n)r −n . Hence . r = (n − 1)r −1 .

(23.31)

Metric Riemannian geometry

305

And hence . r 2 = 2 div(r grad r) = 2 + 2rr = 2n.

(23.32)

We now apply (21.16) to r 2 /2 = F and obtain . . 1 n = r 2 = | Hess F|2 + r 2 Ricci(∇r, ∇r). 2

(23.33)

Since RicciM −λ it follows from (23.33) and (23.32) that . Hess F = gM . Hence if F˜ is an harmonic function which approximates F we can check Assumptions 23.1, 23.2. Therefore Theorem 23.14 implies that there exists X such that dGH Ap (2δ, 1 − 2δ, M), d , (2δ, 1 − 2δ) ×r X τ (|δ, λ, n). To complete the proof it suffices to show that X is close to S n−1 with respect to the Gromov–Hausdorff distance. We can do it by looking at the proof of Theorem 23.14 in our case a bit more carefully. Alternatively we can proceed as follows. Take ρ - 1, with δ - ρ n . By the assumption and the Bishop–Gromov inequality we can find p1 , q1 such that 2d(p, p1 ) = 2d(p, q1 ) = d(p1 , q1 ). We use it in the same way as in the proof of Theorem 23.1 to find V1 ⊃ Bp (ρ, M) such that dGH (V1 , [−ρ, ρ] × X1 ) ρτ (, λ|n). We then . . take points p2 , q2 in a neighborhood of X1 such that 2d(p2 , p) = 2d(q2 , p) = d(p2 , q2 ). Then we use it in the same way as in the proof of Remark 23.1 to find V2 ⊃ Bp (ρ 2 , M) such that dGH (V2 , [−ρ, ρ]2 × X2 ) ρ 2 τ (, λ|n). Repeating this n times, we obtain Vn ⊃ Bp (ρ n , M) such that dGH (Vn , [−ρ, ρ]n ) ρ n τ (, λ|n). Since δ - ρ n it then follows that dGH (X, S n−1 ) τ (, λ|n). This implies Theorem 22.5(1). The proof of Theorem 23.10 is similar to the first half of the proof of Theorem 22.5(1) and is omitted. We next explain the proof of Theorem 23.11. Let RicciM (n − 1) and p, q ∈ M with d(p, q) π − . Put f (r) = sin r, r(x) = d(x, p), F(r) = −k(r) = − cos r. By (the proof of) Theorem 21.9 (see (21.17); we remark that f there is our F ), . Hess(F) = k(r)gM . In this way we can check Assumption 23.1. Assumption 23.2 follows from the Bishop– Gromov inequality in this case. We thus can apply Theorem 23.14 and prove Theorem 23.11. We next explain the idea of the proof of Theorem 23.6. Let ((X, dX ), x) = pGH limi→∞ (Mi , xi ) with (Mi , xi ) ∈ Sn (∞, v). We suppose that a tangent cone Tx X = pGH limi→∞ ((X, ri dX ), x) is not a cone.

306

K. Fukaya

Then there exist δ, R, ρ (and a subsequence of ri which we denote by the same symbols) such that dGH Ax δ, R; (X, ri dX ) , Ao (δ, R; CY ) > ρ for a cone CY . We can take ji → ∞ such that dGH Axji δ, R; (Mji , ri gMji ) , Ao (δ, R; CY ) > ρ/2

(23.34)

for a cone CY . We now claim that Vol(Sxji+1 (δ/ri+1 , (Mji+1 , gMji+1 ))) Vol(S0 (δ/ri+1 , Rn )) > (1 + )

Vol(Sxji (R/ri , (Mji , gMji )))

(23.35)

Vol(S0 (R/ri , Rn ))

for independent of i. In fact, if (23.35) does not hold, then we can apply the argument of the first half of the proof of Theorem 22.5(1) to Axji (δ, R; (Mji , ri gMji )) and using Theorem 23.14, we can show that (23.34) does not hold. Now it is easy to deduce a contradiction from (23.35). By taking a subsequence we may assume that δ/ri > R/ri+1 . Then (23.35) and the Bishop–Gromov inequality implies that Vol(Sxji (R/ri , (Mji , gMji ))) Vol(S0 (R/ri , Rn )) > (1 + )i−1

Vol(Sxj1 (R/r1 , (Mj1 , gMj1 ))) Vol(S0 (R/r1 , Rn ))

.

This is a contradiction since the left-hand side is bounded as i → ∞.

(23.36)

In [31], Cheeger–Colding studied a convergence of the eigenvalue of the Laplace operator using the results explained so far. We state their result (without outline of the proof) here. We start with a simple example to illustrate that the measured Hausdorff convergence is related to the eigenvalue of the Laplace operator. Let us consider T 2 = S 1 × S 2 with

f Riemannian metric g = dt 2 + 2 f (t)2 ds 2 . Here f : S 1 → R+ . We assume f dt = 1. f As we mentioned before the limit of (T 2 , g ) with respect to the measured Hausdorff f 1 topology is S with standard metric and measure f dt. The Dirichlet integral on (T 2 , g ) is D(h, h) =

2 2 dh dh 1 f (t) + dt ds. dt f (t) ds

In case we consider the eigenvalues of the Laplacian which stay bounded as → 0, it suffices to consider h which is constant along the s direction. Hence we are to consider the

Metric Riemannian geometry

307

bilinear form on L2 (S 1 ) defined by D(h, h) =

2 dh f (t) dt. dt

In [54] the author proved that a similar phenomenon occurs in the situation we discussed in Section 11. Cheeger–Colding generalized it much and proved the following Theorem 23.15. T HEOREM 23.15 [31, Theorem 7.9]. Let Mi ∈ Sn (D). We assume that it converges to (X, µ) with respect to the measured Hausdorff topology. Then there exists a (unbounded) symmetric bilinear form D on L2 (X, µ) with discrete spectrum λ0 (D) = 0 < λ1 (D) λ2 (D) · · · such that kth eigenvalue λk (−Mi ) of the Laplace operator (on functions) on Mi converges to λk (D). Remark 23.11. (1) In case the multiplicity of eigenvalue λk (D) is m then we put λk (D) = · · · = λk+m−1 (D). (2) The eigenfunction of −Mi converges to the eigenfunctions of D in an appropriate sense. We finally remark that the study of limits of Einstein manifolds (or manifolds with integral bounds of the curvature tensor) we discussed in Section 20 is improved by [32,27], etc. Here we restrict ourselves to quote the following Theorems 23.16, 23.17. Let Mi be a sequence of n-dimensional Riemannian manifolds. We consider the following integral bounds of the curvature for pi ∈ Mi : |RMi |p ΩMi < C, (23.37) Bpi (1,Mi )

where C is independent of i. Let S, Sk be as in Definitions 23.4. Hm is the m-dimensional Hausdorff measure. We say x ∈ S is (n − 4k)-nonexceptional if there exists a tangent cone Tx X which is not isometric to Rn−4k × C(S 4k−1 /Γ ) where Γ ⊂ O(4k) is a finite group acting freely on S 4k−1 . Otherwise x is said to be (n − 4k)-exceptional. Let Nn−4k ⊂ Sn−4k be the set of all (n − 4k)-nonexceptional points. T HEOREM 23.16 ([32, Theorems 1.15, 1.20], [27, Theorem 6.10]). Let Mi ∈ Sn (∞, v) pGH and limi→∞ (Mi , pi ) = (X, p). We assume (23.37) for 1 p n/2 = dim Mi /2: (1) If p is not an integer then Hn−2p (S) = 0. (2) The Hausdorff dimension of S is not greater than n − 2p. (3) If p = 2, then Hn−4 (Nn−4 ) = 0. (4) If Mi are Kähler and p is an integer, then Hn−2p (S ∩ Bp (R, X)) < ∞ for any R. We remark that in case n = 2 and the manifolds Mi are Einstein, Theorem 23.16(3) is Theorem 20.4.

308

K. Fukaya

T HEOREM 23.17 [27, Theorem 11.1]. In the situation of Theorem 23.16 we have (1) If p = 1, then compact subsets of S are (n − 2) rectifiable. (2) If either p = 2k is an even integer, then Nn−4k ∩ Bp (R, X) are (n − 4k)-rectifiable. (3) Mi are Kähler and p is integer, then S ∩ Bp (R, X) are (n − 2p)-rectifiable. We remark that as in 4-dimensional case, if p = 2 and the Mi are Einstein, the condition (23.37) can be written in terms of characteristic classes and hence is a topological one. These results are parallel to the corresponding results in (higher-dimensional) gauge theory [105,146].

References [1] U. Abresch, Über das glatten Riemannisher Metriken, Habilitationsschrift, Reinishcehn FriedrichWillhelms-Universität, Bonn (1988). [2] U. Abresch and D. Gromoll, On complete manifolds with nonnegative Ricci curvature, J. Amer. Math. Soc. 3 (1990), 355–374. [3] U. Abresch and W. Meyer, Injectivity radius estimates and sphere theorems, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 1–47. [4] U. Abresch and V. Schroeder, Graph manifolds, ends of negatively curved spaces and the hyperbolic 120cell space, J. Differential Geom. 35 (1992), 299–336. [5] A.D. Alexandrov, A theorem on triangles in a metric space and some of its applications, Trudy Mat. Inst. Steklov., vol. 38, Izdat. Akad. Nauk SSSR, Moscow (1951), 5–23 (in Russian). [6] A.D. Alexandrov, Über eine Verallgemeinerung der Riemannschen Geometrie, Schriftenreiche der Institut für Mathematik 1 (1957), 33–84. [7] M.T. Anderson, Convergence and rigidity of manifolds under Ricci curvature bounds, Invent. Math. 102 (1990), 429–445. [8] M.T. Anderson, Metrics of positive Ricci curvature with large diameter, Manuscripta Math. 68 (1990), 405–415. [9] M.T. Anderson, Short geodesics and gravitational instantons, J. Differential Geom. 31 (1990), 265–275. [10] M.T. Anderson, Hausdorff perturbation of Ricci flat manifolds and the splitting theorem, Duke Math. J. 68 (1992), 67–82. [11] M.T. Anderson, The L2 structure of moduli spaces of Einstein metrics on 4-manifolds, Geom. Funct. Anal. 2 (1) (1992), 29–89. [12] M.T. Anderson and J. Cheeger, C α -compactness for manifolds with Ricci curvature and injectivity radius bounded below, J. Differential Geom. 35 (1992), 265–281. [13] S. Bando, Bubbling out of Einstein manifolds, Tohoku Math. J. (2) 42 (1990), 205–216; Correction: Tohoku Math. J. (2) 42 (1990), 587–588. [14] S. Bando, A. Kasue and H. Nakajima, On a construction of coordinates at infinity on manifolds with fast curvature decay and maximal volume growth, Invent. Math. 97 (1989), 313–349. [15] J. Bemelmans, Min-Oo and E. Ruh, Smoothing Riemannian metrics, Math. Z. 188 (1984), 69–74. [16] P. Bérard, G. Besson and S. Gallot, Embedding Riemannian manifolds by their heat kernel, Geom. Funct. Anal. 4 (1994), 373–398. [17] V.N. Berestovskij and I.G. Nikolaev, Multidimensional generalized Riemannian spaces, Geometry, IV, Encyclopaedia Math. Sci., vol. 70, Springer, Berlin (1993), 165–243, 245–250. [18] M. Berger, Les variétés Riemanniennes (1/4)-pincées, Ann. Scuola Norm. Sup. Pisa (3) 14 (1960), 161– 170. [19] M. Berger, Riemannian Geometry During the Second Half of the Twentieth Century, University Lecture Series, vol. 17, American Mathematical Society, Providence, RI (2000). [20] M. Berger, A Panoramic View of Riemannian Geometry, Springer (2003).

Metric Riemannian geometry

309

[21] G. Besson, G. Courtrois and S. Gallot, Entropies et rigidités des espaces localement symétrique de courbure strictment négative, Geom. Funct. Anal. 5 (1995), 731–799. [22] Yu. Burago, M. Gromov and G. Perelman, A.D. Alexandrov’s spaces with curvature bounded from below, Uspekhi Mat. Nauk 47 (1992), 3–51, 222 (in Russian); English transl.: Russian Math. Surveys 47 (1992), 1–58. [23] H. Busemann, The Geometry of Geodesics, Academic Press, New York (1955). [24] P. Buser and H. Karcher, Gromov’s almost flat manifolds, Astérisque, vol. 81, Société Mathématique de France, Paris (1981), 148 pp. [25] J. Cheeger, Finiteness theorems for Riemannian manifolds, Amer. J. Math. 92 (1970), 61–74. [26] J. Cheeger, Degeneration of Riemannian Metrics under Ricci Curvature Bounds, Scuola Normale Superiore, Pisa (2001). [27] J. Cheeger, Integral bounds on curvature elliptic estimates and rectifiability of singular sets, Geom. Funct. Anal. 13 (2003), 20–72. [28] J. Cheeger and T. Colding, Lower bounds on Ricci curvature and the almost rigidity of warped products, Ann. of Math. 144 (1996), 189–237. [29] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. I, J. Differential Geom. 46 (1997), 406–480. [30] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. II, J. Differential Geom. 54 (2000), 13–35. [31] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. III, J. Differential Geom. 54 (2000), 37–74. [32] J. Cheeger, T.H. Colding and G. Tian, On the singularities of spaces with bounded Ricci curvature, Geom. Funct. Anal. 12 (2002), 873–914. [33] J. Cheeger and G. Ebin, Comparison Theorems in Riemannian Geometry, North-Holland Mathematical Library, vol. 9, North-Holland/American Elsevier, Amsterdam/New York (1975). [34] J. Cheeger, K. Fukaya and M. Gromov, Nilpotent structures and invariant metrics on collapsed manifolds, J. Amer. Math. Soc. 5 (1992), 327–372. [35] J. Cheeger and D. Gromoll, The splitting theorem for manifolds of nonnegative Ricci curvature, J. Differential Geom. 6 (1971–1972), 119–128. [36] J. Cheeger and D. Gromoll, On the structure of complete manifolds of nonnegative curvature, Ann. of Math. 96 (1972), 413–443. [37] J. Cheeger and D. Gromoll, On the lower bound for the injectivity radius of 1/4-pinched Riemannian manifolds, J. Differential Geom. 15 (1980), 437–442. [38] J. Cheeger and M. Gromov, Collapsing Riemannian manifolds while keeping their curvature bounded. I, J. Differential Geom. 23 (1986), 309–346. [39] J. Cheeger and M. Gromov, Collapsing Riemannian manifolds while keeping their curvature bounded. II, J. Differential Geom. 32 (1990), 269–298. [40] J. Cheeger and X. Rong, Existence of polarized F -structures on collapsed manifolds with bounded curvature and diameter, Geom. Funct. Anal. 6 (1996), 411–429. [41] S.Y. Cheng, Eigenvalue comparison theorems and its geometric applications, Math. Z. 143 (1975), 289– 297. [42] S.Y. Cheng and S.T. Yau, Differential equations on Riemannian manifolds and their geometric applications, Comm. Pure Appl. Math. 28 (1975), 333–354. [43] T. Colding, Large manifolds with positive Ricci curvature, Invent. Math. 124 (1996), 193–214. [44] T. Colding, Shape of manifolds with positive Ricci curvature, Invent. Math. 124 (1996), 175–191. [45] T. Colding, Aspects of Ricci curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 83–98. [46] T. Colding, Ricci curvature and volume convergence, Ann. of Math. (2) 145 (1997), 477–501. [47] T. Colding, Space with Ricci curvature bound, Documenta Mathematica, Extra Volume, ICM 1998 II, Doc. Math. J. DMV (1999), 299–308. [48] R.D. Edwards and R.C. Kirby, Deformations of spaces of imbeddings, Ann. of Math. 93 (1971), 63–88. [49] T. Eguchi and A. Hanson, Asymptotically flat self-dual solutions to Euclidean gravity, Phys. Lett. B 74 (1978), 249–251.

310

K. Fukaya

[50] J.H. Eschenburg, New examples of manifolds with strictly positive curvature, Invent. Math. 66 (1982), 469–480. [51] F. Fang and X. Rong, Positive pinching, volume and second Betti number, Geom. Funct. Anal. 9 (1999), 641–674. [52] K. Fukaya, Theory of convergence of Riemannian orbifolds, Japan. J. Math. 12 (1984), 121–160. [53] K. Fukaya, Collapsing Riemannian manifolds to ones of lower dimensions, J. Differential Geom. 25 (1987), 139–156. [54] K. Fukaya, Collapsing of Riemannian manifolds and eigenvalues of Laplace operator, Invent. Math. 87 (1987), 517–547. [55] K. Fukaya, A boundary of the set of Riemannian manifolds with bounded curvature and diameter, J. Differential Geom. 28 (1988), 1–21. [56] K. Fukaya, Collapsing Riemannian manifolds to ones with lower dimension. II, J. Math. Soc. Japan 41 (1989), 333–356. [57] K. Fukaya, Hausdorff convergence of Riemannian manifolds and its applications, Recent Topics in Differential and Analytic Geometry, Adv. Stud. Pure Math., vol. 18-I, Academic Press, Boston, MA (1990), 143–238. [58] K. Fukaya and T. Yamaguchi, Almost nonpositively curved manifolds, J. Differential Geom. 33 (1991), 67–90. [59] K. Fukaya and T. Yamaguchi, The fundamental groups of almost non-negatively curved manifolds, Ann. of Math. 136 (1992), 253–333. [60] K. Fukaya and T. Yamaguchi, Isometry groups of singular spaces, Math. Z. 216 (1994), 31–44. [61] S. Gallot, A Sobolev inequality and some geometric applications, Spectra of Riemannian Manifolds, Berger, Murakami and Ochiai, eds, Kaigai, Kyoto (1981). [62] S. Gallot, Volumes, courbure de Ricci et convergence des variétés d’après T.H. Colding et Cheeger– Colding, Séminaire Bourbaki, 50èm année, vol. 835 (1997–1998), 1–33. [63] R.E. Greene, Noncompact manifolds of nonnegative curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 99–131. [64] R.E. Greene and H. Wu, Lipschitz convergence of Riemannian manifolds, Pacific J. Math. 131 (1988), 119–141. [65] D. Gromoll, Differenzierbare Strukturen und Metriken positiver Krümmung auf Sphären, Math. Ann. 164 (1966), 353–371. [66] D. Gromoll and W. Myers, On complete open manifolds of positive curvature, Ann. of Math. 90 (1969), 75–90. [67] M. Gromov, Manifolds of negative curvature, J. Differential Geom. 13 (1978), 223–230. [68] M. Gromov, Almost flat manifolds, J. Differential Geom. 13 (1978), 231–241. [69] M. Gromov, Structures métriques pour les variétés riemannienes, J. Lafontaine and P. Pansu, eds, Textes Mathématiques, vol. 1, CEDIC, Paris (1981). [70] M. Gromov, Curvature, diameter and Betti numbers, Comment. Math. Helv. 56 (1981), 179–195. [71] M. Gromov, Groups of polynomial growth and expanding maps, Comment. Math. Helv. 56 (1981), 179– 195. [72] M. Gromov, Volume and bounded cohomology, Inst. Hautes Etudes Sci. Publ. Math. 56 (1982), 5–99. [73] M. Gromov, Filling Riemannian manifolds, J. Differential Geom. 18 (1983), 1–147. [74] M. Gromov, Carnot–Carathéodory spaces seen from within, Sub-Riemannian Geometry, Progress in Mathematics, vol. 144, Birkhäuser, Basel (1996), 79–323. [75] M. Gromov, Metric structures for Riemannian and non-Riemannian spaces, With appendices by M. Katz, P. Pansu and S. Semmes, Translated from the French by Sean Michael Bates, Progress in Mathematics, vol. 152, Birkhäuser Boston, Boston, MA (1999). [76] M. Gross and P. Wilson, Large complex structure limit of K3 surfaces, J. Differential Geom. 55 (2000), 475–546. [77] K. Grove and H. Karcher, How to conjugate C 1 -close group actions, Math. Z. 132 (1973), 11–20. [78] K. Grove and P. Petersen, Bounding homotopy types by geometry, Ann. of Math. (2) 128 (1988), 195–206. [79] K. Grove and P. Petersen, Homotopy types of positively curved manifolds with large volume, Amer. J. Math. 110 (1988), 1183–1188.

Metric Riemannian geometry

311

[80] K. Grove and P. Petersen, Manifolds near the boundary of existence, J. Differential Geom. 33 (1991), 379–394. [81] K. Grove, P. Petersen and J.Y. Wu, Geometric finiteness theorems via controlled topology, Invent. Math. 99 (1990), 205–213. [82] K. Grove and K. Shiohama, A generalized sphere theorem, Ann. of Math. (2) 106 (1977), 201–211. [83] K. Grove and W. Ziller, Cohomogeneity one manifolds with positive Ricci curvature, Invent. Math. 149 (2002), 619–646. [84] T. Hattori, Asymptotic geometry of arithmetic quotients of symmetric spaces, Math. Z. 222 (1996), 247– 277. [85] N. Hitchin, Harmonic spinor, Adv. Math. 14 (1974), 1–55. [86] H. Im Hof and E. Ruh, An equivariant pinching theorem, Comment. Math. Helv. 50 (1975), 389–401. [87] J. Jost and H. Karcher, Geometrische Methoden zur Gewinnung von a-priori-Schranken für harmonische Abbildungen, Manuscripta Math. 40 (1982), 27–77. [88] J. Jost and H. Karcher, Almost linear functions and a priori estimates for harmonic maps, Global Riemannian Geometry (Durham, 1983), Ellis Horwood Ser. Math. Appl., Horwood, Chichester (1984), 148– 155. [89] D. Joyce, Compact Manifolds with Special Holonomy, Oxford Univ. Press (2000). [90] A. Kasue, A convergence theorem for Riemannian manifolds and some applications, Nagoya Math. J. 114 (1989), 21–51. [91] A. Kasue, Spectral convergence of Riemannian manifolds, Tohoku Math. J. (2) 46 (1994), 147–179. [92] A. Kasue, Y. Otsu, T. Sakai, T. Shioya and T. Yamaguchi, Riemannian manifolds and its limit, Mem. Math. Soc. Japan (in Japanese), in press. [93] A. Katsuda, Gromov’s convergence theorem and its application, Nagoya Math. J. 100 (1985), 11–48. [94] W. Klingenberg, Contributions to Riemannian geometry in the large, Ann. of Math. 69 (1959), 654–666. [95] W. Klingenberg and T. Sakai, Injectivity radius estimate for 1/4-pinched manifolds, Arch. Math. (Basel) 34 (1980), 371–376. [96] R. Kobayashi and A. Todorov, Polarized period map for generalized K3 surface and moduli of Einstein metric, Tohoku Math. J. 39 (1987), 341–363. [97] S. Kobayashi and K. Nomizu, Foundation of Differential Geometry, I, II, Interscience (1963–1969). [98] P. Kronheimer, The construction of ALE spaces as hyper Kähler quotients, J. Differential Geom. 29 (1989), 665–683. [99] K. Kuwae, Y. Machigashira and T. Shioya, Sobolev spaces, Laplacian, and heat kernel on Alexandrov spaces, Math. Z. 238 (2001), 269–316. [100] P. Li and R. Schoen, Lp and mean value properties of subharmonic functions on Riemannian manifolds, Acta Math. 153 (1984), 279–301. [101] A. Lichnerowicz, Geometric des Groupes des transformationes, Dunod, Paris (1958). [102] X. Menguy, Noncollapsing examples with positive Ricci curvature and infinite topological type, Geom. Funct. Anal. 10 (2000), 600–627. [103] J. Milnor, Morse Theory, Annals of Mathematics Studies, vol. 51, Princeton Univ. Press, Princeton, NJ (1963). [104] J. Milnor, A note on curvature and fundamental group, J. Differential Geom. 2 (1968), 1–7. [105] H. Nakajima, Compactness of the moduli space of Yang–Mills connections in higher dimensions, J. Math. Soc. Japan 40 (1988), 383–392. [106] H. Nakajima, Hausdorff convergence of Einstein 4-manifolds, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 35 (1988), 411–424. [107] I.G. Nikolaev, Parallel translation and smoothness of the metric of spaces with bounded curvature, Dokl. Akad. Nauk SSSR 250 (1980), 1056–1058. [108] I.G. Nikolaev, A metric characterization of Riemannian spaces, Siberian Adv. Math. 9 (1999), 1–58. [109] M. Obata, Certain condition for a Riemannian manifold to be isometric to a sphere, J. Math. Soc. Japan 14 (1962), 333–340. [110] Y. Otsu, On manifolds of positive Ricci curvature with large diameter, Math. Z. 206 (1991), 255–264. [111] Y. Otsu, K. Shiohama and T. Yamaguchi, A new version of differentiable sphere theorem, Invent. Math. 98 (1989), 219–228.

312

K. Fukaya

[112] Y. Otsu and T. Shioya, The Riemannian structure of Alexandrov spaces, J. Differential Geom. 39 (1994), 629–658. [113] G. Perelman, Elements of Morse theory on Aleksandrov spaces, Algebra i Analiz 5 (1993), 232–241 (in Russian); English transl.: St. Petersburg Math. J. 5 (1994), 205–213. [114] G. Perelman, Manifolds of positive Ricci curvature with almost maximal volume, J. Amer. Math. Soc. 7 (1994), 299–305. [115] G. Perelman, Proof of the soul conjecture of Cheeger and Gromoll, J. Differential Geom. 40 (1) (1994), 209–212. [116] G. Perelman, Spaces with curvature bounded below, Proceedings of the International Congress of Mathematicians, vols. 1, 2 (Zürich, 1994), Birkhäuser, Basel (1995), 517–525. [117] G. Perelman, Collapsing with no proper extremal subsets, Comparison Geometry (Berkeley, CA, 1993– 1994), Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 149–155. [118] G. Perelman, Construction of manifolds of positive Ricci curvature with big volume and large Betti numbers, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 157–163. [119] G. Perelman, A.D. Alexandrov’s spaces with curvatures bounded from below II, preprint. [120] S. Peters, Cheeger’s finiteness theorem for diffeomorphism classes of Riemannian manifolds, J. Reine Angew. Math. 349 (1984), 77–82. [121] S. Peters, Convergence of Riemannian manifolds, Compositio Math. 62 (1987), 3–16. [122] P. Petersen, A finiteness theorem for metric spaces, J. Differential Geom. 31 (1990), 387–395. [123] P. Petersen, Convergence theorems in Riemannian geometry, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 167–202. [124] A. Petrunin, Applications of quasigeodesics and gradient curves, Comparison Geometry (Berkeley, CA, 1993–1994), Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 203–219. [125] A. Petrunin, X. Rong and W. Tuschmann, Collapsing vs. positive pinching, Geom. Funct. Anal. 9 (1999), 699–735. [126] A. Petrunin and W. Tuschmann, Diffeomorphism finiteness, positive pinching, and second homotopy, Geom. Funct. Anal. 9 (1999), 736–774. [127] C. Plaut, Metric spaces of curvature k, Handbook of Geometric Topology, North-Holland, Amsterdam (2002), 819–898. [128] W.A. Poor, Some results on nonnegatively curved manifolds, J. Differential Geom. 9 (1974), 583–600. [129] H.E. Rauch, A contribution to differential geometry in the large, Ann. of Math. (2) 54 (1951), 38–55. [130] X. Rong, The existence of polarized F -structures on volume collapsed 4-manifolds, Geom. Funct. Anal. 3 (1993), 474–501. [131] X. Rong, The limiting eta invariants of collapsed three-manifolds, J. Differential Geom. 37 (1993), 535– 568. [132] X. Rong, On the fundamental groups of manifolds of positive sectional curvature, Ann. of Math. (2) 143 (1996), 397–411. [133] E. Ruh, Almost flat manifolds, J. Differential Geom. 17 (1982), 1–14. [134] V. Schroeder, Rigidity of nonpositively curved graphmanifolds, Math. Ann. 274 (1986), 19–26. [135] J. Sha and D. Yang, Examples of manifolds of positive Ricci curvature, J. Differential Geom. 29 (1989), 95–103. [136] Y. Shikata, On the differentiable pinching problem, Osaka J. Math. 4 (1967), 279–287. [137] K. Shiohama, A sphere theorem for manifolds of positive Ricci curvature, Trans. Amer. Math. Soc. 275 (1983), 811–819. [138] K. Shiohama, An Introduction to the Geometry of Alexandrov Spaces, Lecture Notes Series, vol. 8, Seoul National University, Seoul, Korea (1993). [139] K. Shiohama, Sphere theorems, Handbook of Differential Geometry, vol. I, North-Holland, Amsterdam (2000), 865–903. [140] K. Shiohama and T. Yamaguchi, Positively curved manifolds with restricted diameters, Geometry of Manifolds (Matsumoto, 1988), Perspect. Math., vol. 8, Academic Press, Boston, MA (1989), 345–350. [141] T. Shioya and T. Yamaguchi, Collapsing three-manifolds under a lower curvature bound, J. Differential Geom. 56 (2000), 1–66.

Metric Riemannian geometry

313

[142] L.C. Siebenmann, Deformation of homeomorphisms on stratified sets. I, II, Comment. Math. Helv. 47 (1972), 123–136; ibid. 47 (1972), 137–163. [143] M. Sugimoto and K. Shiohama, On the differentiable pinching problem, Math. Ann. 195 (1971), 1–16. [144] Y. Suyama, Differentiable sphere theorem by curvature pinching, J. Math. Soc. Japan 43 (3) (1991), 527– 553. [145] C. Taubes, Self-dual Yang–Mills connections on non-self-dual 4-manifolds, J.Differential Geom. 17 (1982), 139–170. [146] G. Tian, Gauge theory and calibrated geometry. I, Ann. of Math. (2) 151 (2000), 193–268. [147] J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972), 250–270. [148] V.A. Toponogov, Riemann spaces with curvature bounded below, Uspekhi Mat. Nauk 14 (1) (1959), 87– 130. [149] V.A. Toponogov, Riemannian spaces containing straight lines, Dokl. Akad. Nauk SSSR 127 (1959), 977– 979. [150] A. Weinstein, On the homotopy type of positively-pinched manifolds, Arch. Math. (Basel) 18 (1967), 523– 524. [151] T. Yamaguchi, Manifolds of almost nonnegative Ricci curvarture, J. Differential Geom. 28 (1988), 157– 167. [152] T. Yamaguchi, Lipschitz convergence of manifolds of positive Ricci curvature with large volume, Math. Ann. 284 (1989), 423–436. [153] T. Yamaguchi, Collapsing and pinching under a lower curvature bound, Ann. of Math. (2) 133 (1991), 317–357. [154] T. Yamaguchi, A convergence theorem in the geometry of Alexandrov spaces, Actes de la Table Ronde de Géométrie Différentielle (Luminy, 1992), Sémin. Congr., vol. 1, Soc. Math. France, Paris (1996), 601–642. [155] T. Yamaguchi, Collapsing Riemannian 4-manifolds, Suugaku 52 (2000), 172–186 (in Japanese). [156] T. Yamaguchi, Collapsing 4-manifolds under a Lower curvature bound, Preprint (2002). [157] K. Yano and S. Bochner, Curvature and Betti Numbers, Annals of Mathematics Studies, vol. 32, Princeton Univ. Press, Princeton, NJ (1953). [158] S.T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge–Ampère equation. I, Comm. Pure Appl. Math. 31 (1978), 339–411. [159] S. Zhu, The comparison geometry of Ricci curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 221–262.

This page intentionally left blank

CHAPTER 5

Contact Geometry Hansjörg Geiges∗ Mathematisches Institut, Universität zu Köln, Weyertal 86-90, 50931 Köln, Germany E-mail: geiges@math.uni-koeln.de

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . 2. Contact manifolds . . . . . . . . . . . . . . . . . . . 2.1. Contact manifolds and their submanifolds . . . . 2.2. Gray stability and the Moser trick . . . . . . . . 2.3. Contact Hamiltonians . . . . . . . . . . . . . . . 2.4. Darboux’s theorem and neighbourhood theorems 2.5. Isotopy extension theorems . . . . . . . . . . . . 2.6. Approximation theorems . . . . . . . . . . . . . 3. Contact structures on 3-manifolds . . . . . . . . . . . 3.1. An invariant of transverse knots . . . . . . . . . 3.2. Martinet’s construction . . . . . . . . . . . . . . 3.3. 2-plane fields on 3-manifolds . . . . . . . . . . . 3.4. Let’s twist again . . . . . . . . . . . . . . . . . . 3.5. Other existence proofs . . . . . . . . . . . . . . 3.6. Tight and overtwisted . . . . . . . . . . . . . . . 3.7. Classification results . . . . . . . . . . . . . . . 4. A guide to the literature . . . . . . . . . . . . . . . . 4.1. Dimension 3 . . . . . . . . . . . . . . . . . . . . 4.2. Higher dimensions . . . . . . . . . . . . . . . . 4.3. Symplectic fillings . . . . . . . . . . . . . . . . 4.4. Dynamics of the Reeb vector field . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

317 317 319 325 328 329 341 345 351 352 354 356 363 367 370 375 376 377 377 378 378 378

* I am grateful to Fan Ding, Jesús Gonzalo and Federica Pasquotto for their attentive reading of the original

manuscript. I also thank John Etnyre and Stephan Schönenberger for allowing me to use a couple of their figures (viz., Figures 2 and 1 of the present text, respectively). HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 315

This page intentionally left blank

Contact geometry

317

1. Introduction Over the past two decades, contact geometry has undergone a veritable metamorphosis: once the ugly duckling known as ‘the odd-dimensional analogue of symplectic geometry’, it has now evolved into a proud field of study in its own right. As is typical for a period of rapid development in an area of mathematics, there are a fair number of folklore results that every mathematician working in the area knows, but no references that make these results accessible to the novice. I therefore take the present article as an opportunity to take stock of some of that folklore. There are many excellent surveys covering specific aspects of contact geometry (e.g., classification questions in dimension 3, dynamics of the Reeb vector field, various notions of symplectic fillability, transverse and Legendrian knots and links). All these topics deserve to be included in a comprehensive survey, but an attempt to do so here would have left this article in the ‘to appear’ limbo for much too long. Thus, instead of adding yet another survey, my plan here is to cover in detail some of the more fundamental differential topological aspects of contact geometry. In doing so, I have not tried to hide my own idiosyncrasies and preoccupations. Owing to a relatively leisurely pace and constraints of the present format, I have not been able to cover quite as much material as I should have wished. Nonetheless, I hope that the reader of the present handbook chapter will be better prepared to study some of the surveys I alluded to—a guide to these surveys will be provided—and from there to move on to the original literature. A book chapter with comparable aims is Chapter 8 in [1]. It seemed opportune to be brief on topics that are covered extensively there, even if it is done at the cost of leaving out some essential issues. I hope to return to the material of the present chapter in a yet to be written more comprehensive monograph.

2. Contact manifolds Let M be a differential manifold and ξ ⊂ T M a field of hyperplanes on M. Locally such a hyperplane field can always be written as the kernel of a non-vanishing 1-form α. One way to see this is to choose an auxiliary Riemannian metric g on M and then to define α = g(X, .), where X is a local non-zero section of the line bundle ξ ⊥ (the orthogonal complement of ξ in T M). We see that the existence of a globally defined 1-form α with ξ = ker α is equivalent to the orientability (hence triviality) of ξ ⊥ , i.e. the coorientability of ξ . Except for an example below, I shall always assume this condition. If α satisfies the Frobenius integrability condition α ∧ dα = 0, then ξ is an integrable hyperplane field (and vice versa), and its integral submanifolds form a codimension 1 foliation of M. Equivalently, this integrability condition can be written as X, Y ∈ ξ

&⇒

[X, Y ] ∈ ξ.

318

H. Geiges

An integrable hyperplane field is locally of the form dz = 0, where z is a coordinate function on M. Much is known, too, about the global topology of foliations, cf. [100]. Contact structures are in a certain sense the exact opposite of integrable hyperplane fields. D EFINITION 2.1. Let M be a manifold of odd dimension 2n + 1. A contact structure is a maximally non-integrable hyperplane field ξ = ker α ⊂ T M, that is, the defining 1-form α is required to satisfy α ∧ (dα)n = 0 (meaning that it vanishes nowhere). Such a 1-form α is called a contact form. The pair (M, ξ ) is called a contact manifold. R EMARK 2.2. Observe that in this case α ∧ (dα)n is a volume form on M; in particular, M needs to be orientable. The condition α ∧ (dα)n = 0 is independent of the specific choice of α and thus is indeed a property of ξ = ker α: Any other 1-form defining the same hyperplane field must be of the form λα for some smooth function λ : M → R \ {0}, and we have n (λα) ∧ d(λα) = λα ∧ (λ dα + dλ ∧ α)n = λn+1 α ∧ (dα)n = 0. We see that if n is odd, the sign of this volume form depends only on ξ , not the choice of α. This makes it possible, given an orientation of M, to speak of positive and negative contact structures. R EMARK 2.3. An equivalent formulation of the contact condition is that we have (dα)n |ξ = 0. In particular, for every point p ∈ M, the 2n-dimensional subspace ξp ⊂ Tp M is a vector space on which dα defines a skew-symmetric form of maximal rank, that is, (ξp , dα|ξp ) is a symplectic vector space. A consequence of this fact is that there exists a complex bundle structure J : ξ → ξ compatible with dα (see [92, Proposition 2.63]), i.e. a bundle endomorphism satisfying • J 2 = −idξ , • dα(J X, J Y ) = dα(X, Y ) for all X, Y ∈ ξ , • dα(X, J X) > 0 for 0 = X ∈ ξ . R EMARK 2.4. The name ‘contact structure’ has its origins in the fact that one of the first historical sources of contact manifolds are the so-called spaces of contact elements (which in fact have to do with ‘contact’ in the differential geometric sense), see [7] and [45]. In the 3-dimensional case the contact condition can also be formulated as X, Y ∈ ξ linearly independent

&⇒

[X, Y ] ∈ / ξ;

Contact geometry

319

this follows immediately from the equation dα(X, Y ) = X α(Y ) − Y α(X) − α [X, Y ] and the fact that the contact condition (in dimension 3) may be written as dα|ξ = 0. In the present article I shall take it for granted that contact structures are worthwhile objects of study. As I hope to illustrate, this is fully justified by the beautiful mathematics to which they have given rise. For an apology of contact structures in terms of their origin (with hindsight) in physics and the multifarious connections with other areas of mathematics I refer the reader to the historical surveys [87] and [45]. Contact structures may also be justified on the grounds that they are generic objects: A generic 1-form α on an odd-dimensional manifold satisfies the contact condition outside a smooth hypersurface, see [89]. Similarly, a generic 1-form α on a 2n-dimensional manifold satisfies the condition α ∧ (dα)n−1 = 0 outside a submanifold of codimension 3; such ‘even-contact manifolds’ have been studied in [51], for instance, but on the whole their theory is not as rich or well motivated as that of contact structures. D EFINITION 2.5. Associated with a contact form α one has the so-called Reeb vector field Rα , defined by the equations (i) dα(Rα , .) ≡ 0, (ii) α(Rα ) ≡ 1. As a skew-symmetric form of maximal rank 2n, the form dα|Tp M has a 1-dimensional kernel for each p ∈ M 2n+1 . Hence equation (i) defines a unique line field Rα on M. The contact condition α ∧ (dα)n = 0 implies that α is non-trivial on that line field, so a global vector field is defined by the additional normalisation condition (ii). 2.1. Contact manifolds and their submanifolds We begin with some examples of contact manifolds; the simple verification that the listed 1-forms are contact forms is left to the reader. E XAMPLE 2.6. On R2n+1 with Cartesian coordinates (x1 , y1 , . . . , xn , yn , z), the 1-form α1 = dz +

n

xj dyj

j =1

is a contact form. E XAMPLE 2.7. On R2n+1 with polar coordinates (rj , ϕj ) for the (xj , yj )-plane, j = 1, . . . , n, the 1-form α2 = dz +

n j =1

is a contact form.

rj2 dϕj = dz +

n (xj dyj − yj dxj ) j =1

320

H. Geiges

Fig. 1. The contact structure ker(dz + x dy).

D EFINITION 2.8. Two contact manifolds (M1 , ξ1 ) and (M2 , ξ2 ) are called contactomorphic if there is a diffeomorphism f : M1 → M2 with Tf (ξ1 ) = ξ2 , where Tf : T M1 → T M2 denotes the differential of f . If ξi = ker αi , i = 1, 2, this is equivalent to the existence of a nowhere zero function λ : M1 → R such that f ∗ α2 = λα1 . E XAMPLE 2.9. The contact manifolds (R2n+1 , ξi = ker αi ), i = 1, 2, from the preceding examples are contactomorphic. An explicit contactomorphism f with f ∗ α2 = α1 is given by f (x, y, z) = (x + y)/2, (y − x)/2, z + xy/2 , where x and y stand for (x1 , . . . , xn ) and (y1 , . . . , yn ), respectively, and xy stands for j xj yj . Similarly, both these contact structures are contactomorphic to ker(dz − y dxj ). Any of these contact structures is called the standard contact structure on j j R2n+1 . E XAMPLE 2.10. The standard contact structure on the unit sphere S 2n+1 in R2n+2 (with Cartesian coordinates (x1 , y1 , . . . , xn+1 , yn+1 )) is defined by the contact form n+1 (xj dyj − yj dxj ). α0 = j =1

With r denoting the radial coordinate on R2n+2 (that is, r 2 = j (xj2 + yj2 )) one checks easily that α0 ∧ (dα0 )n ∧ r dr = 0 for r = 0. Since S 2n+1 is a level surface of r (or r 2 ), this verifies the contact condition. Alternatively, one may regard S 2n+1 as the unit sphere in Cn+1 with complex structure J (corresponding to complex coordinates zj = xj + iyj , j = 1, . . . , n + 1). Then ξ0 = ker α0

Contact geometry

321

defines at each point p ∈ S 2n+1 the complex (i.e. J -invariant) subspace of Tp S 2n+1 , that is, ξ0 = T S 2n+1 ∩ J T S 2n+1 . This follows from the observation that α = −r dr ◦ J . The hermitian form dα(., J.) on ξ0 is called the Levi form of the hypersurface S 2n+1 ⊂ Cn+1 . The contact condition for ξ corresponds to the positive definiteness of that Levi form, or what in complex analysis is called the strict pseudoconvexity of the hypersurface. For more on the question of pseudoconvexity from the contact geometric viewpoint see [1, Section 8.2]. Beware that the ‘complex structure’ in their Proposition 8.14 is not required to be integrable, i.e. constitutes what is more commonly referred to as an ‘almost complex structure’. D EFINITION 2.11. Let (V , ω) be a symplectic manifold of dimension 2n + 2, that is, ω is a closed (dω = 0) and non-degenerate (ωn+1 = 0) 2-form on V . A vector field X is called a Liouville vector field if LX ω = ω, where L denotes the Lie derivative. With the help of Cartan’s formula LX = d ◦ iX + iX ◦ d this may be rewritten as d(iX ω) = ω. Then the 1-form α = iX ω defines a contact form on any hypersurface M in V transverse to X. Indeed, n α ∧ (dα)n = iX ω ∧ d(iX ω) = iX ω ∧ ωn =

1 iX ωn+1 , n+1

which is a volume form on M ⊂ V provided M is transverse to X. E XAMPLE 2.12.With V = R2n+2 , symplectic form ω = j dxj ∧ dyj , and Liouville vector field X = j (xj ∂xj + yj ∂yj )/2 = r ∂r /2, we recover the standard contact structure on S 2n+1 . For finer issues relating to hypersurfaces in symplectic manifolds transverse to a Liouville vector field I refer the reader to [1, Section 8.2]. Here is a further useful example of contactomorphic manifolds. P ROPOSITION 2.13. For any point p ∈ S 2n+1 , the manifold (S 2n+1 \ {p}, ξ0 ) is contactomorphic to (R2n+1 , ξ2 ). P ROOF. The contact manifold (S 2n+1 , ξ0 ) is a homogeneous space under the natural U(n + 1)-action, so we are free to choose p = (0, . . . , 0, −1). Stereographic projection from p does almost, but not quite yield the desired contactomorphism. Instead, we use a map that is well known in the theory of Siegel domains (cf. [3, Chapter 8]) and that looks a bit like a complex analogue of stereographic projection; this was suggested in [92, Exercise 3.64]. Regard S 2n+1 as the unit sphere in Cn+1 = Cn × C with Cartesian coordinates (z1 , . . . , zn , w) = (z, w). We identify R2n+1 with Cn × R ⊂ Cn × C with coordinates

322

H. Geiges

(ζ1 , . . . , ζn , s) = (ζ, s) = (ζ, Re σ ), where ζj = xj + iyj . Then α2 = ds +

n i (xj dyj − yj dxj ) = ds + (ζ d ζ¯ − ζ¯ dζ ) 2 j =1

and i α0 = (z d z¯ − z¯ dz + w d w¯ − w¯ dw). 2 Now define a smooth map f : S 2n+1 \ {(0, −1)} → R2n+1 by i(w − w) ¯ z . ,− (ζ, s) = f (z, w) = 1+w 2|1 + w|2 Then dw d w¯ i dw i d w¯ i(w − w) ¯ i(w − w) ¯ + + + 2(1 + w) |1 + w|2 2(1 + w) ¯ |1 + w|2 2|1 + w|2 2|1 + w|2 w − w¯ w − w¯ i −dw + d w ¯ + dw + d w ¯ = 1+w 1 + w¯ 2|1 + w|2

f ∗ ds = −

and d z¯ z¯ z − d w¯ 1 + w 1 + w¯ (1 + w) ¯ 2 dz z z¯ − dw − 1 + w¯ 1 + w (1 + w)2 dw d w¯ 1 2 z d z ¯ − z ¯ dz + |z| − . = 1 + w 1 + w¯ |1 + w|2

f (ζ d ζ¯ − ζ¯ dζ ) = ∗

Along S 2n+1 we have |z|2 = 1 − |w|2 = (1 − w)(1 + w) ¯ + (w − w) ¯ = (1 − w)(1 ¯ + w) − (w − w), ¯ whence |z|

2

d w¯ w − w¯ w − w¯ dw − = (1 − w) ¯ dw − dw − (1 − w) d w¯ − d w. ¯ 1 + w 1 + w¯ 1+w 1 + w¯

From these calculations we conclude f ∗ α2 = α0 /|1 + w|2 . So it only remains to show that f is actually a diffeomorphism of S 2n+1 \ {(0, −1)} onto R2n+1 . To that end, consider the map f˜ : (Cn × C) \ Cn × {−1} → (Cn × C) \ Cn × {−i/2}

Contact geometry

323

defined by (ζ, σ ) = f˜(z, w) =

z i w−1 ,− . 1+w 2w+1

This is a biholomorphic map with inverse map (ζ, σ ) →

2ζ 1 + 2iσ . , 1 − 2iσ 1 − 2iσ

We compute Im σ = − =

w−1 w¯ − 1 (w − 1)(w¯ + 1) + (w¯ − 1)(w + 1) − =− 4(w + 1) 4(w¯ + 1) 4|1 + w|2

1 − |w|2 . 2|1 + w|2

Hence for (z, w) ∈ S 2n+1 \ {(0, −1)} we have Im σ =

1 |z|2 = |ζ |2 ; 2|1 + w|2 2

conversely, any point (ζ, σ ) with Im σ = |ζ |2 /2 lies in the image of f˜|S 2n+1 \{(0,−1)} , that is, f˜ restricted to S 2n+1 \ {(0, −1)} is a diffeomorphism onto {Im σ = |ζ |2 /2}. Finally, we compute Re σ = − =−

i(w¯ − 1) (w − 1)(w¯ + 1) − (w¯ − 1)(w + 1) i(w − 1) + = −i 4(w + 1) 4(w¯ + 1) 4|1 + w|2 i(w − w) ¯ , 2|1 + w|2

from which we see that for (z, w) ∈ S 2n+1 \ {(0, −1)} and with (ζ, σ ) = f˜(z, w) we have f (z, w) = (ζ, Re σ ). This concludes the proof. At the beginning of this section I mentioned that one may allow contact structures that are not coorientable, and hence not defined by a global contact form. E XAMPLE 2.14. Let M = Rn+1 × RP n with Cartesian coordinates (x0 , . . . , xn ) on the Rn+1 -factor and homogeneous coordinates [y0 : . . . : yn ] on the RP n -factor. Then n yj dxj ξ = ker j =0

324

H. Geiges

is a well-defined hyperplane field on M, because the 1-form on the right-hand side is well defined up to scaling by a non-zero real constant. On the open submanifold Uk = {yk = 0} ∼ = Rn+1 × Rn of M we have ξ = ker αk with yj dxj αk = dxk + yk j =k

an honest 1-form on Uk . This is the standard contact form of Example 2.6, which proves that ξ is a contact structure on M. If n is even, then M is not orientable, so there can be no global contact form defining ξ (cf. Remark 2.2), i.e. ξ is not coorientable. Notice, however, that a contact structure on a manifold of dimension 2n + 1 with n even is always orientable: the sign of (dα)n |ξ does not depend on the choice of local 1-form defining ξ . If n is odd, then M is orientable, so it would be possible that ξ is the kernel of a globally defined 1-form. However, since the sign of α ∧ (dα)n , for n odd, is independent of the choice of local 1-form defining ξ , it is also conceivable that no global contact form exists. (In fact, this consideration shows that any manifold of dimension 2n + 1, with n odd, admitting a contact structure (coorientable or not) needs to be orientable.) This is indeed what happens, as we shall prove now. P ROPOSITION 2.15. Let (M, ξ ) be the contact manifold of the preceding example. Then T M/ξ can be identified with the canonical line bundle on RP n (pulled back to M). In particular, T M/ξ is a non-trivial line bundle, so ξ is not coorientable. P ROOF. For given y = [y0 : . . . : yn ] ∈ RP n , the vector y0 ∂x0 + · · · + yn ∂xn ∈ Tx Rn+1 is well defined up to a non-zero real factor (and independent of x ∈ Rn+1 ), and hence defines a line y in Tx Rn+1 ∼ = Rn+1 . The set E = (t, x, y): x ∈ Rn+1 , y ∈ RP n , t ∈ y ⊂ T Rn+1 × RP n ⊂ T Rn+1 × RP n = T M with projection (t, x, y) → (x, y) defines a line sub-bundle of T M that restricts to the canonical line bundle over {x}×RP n ≡ RP n for each x ∈ Rn+1 . The canonical line bundle over RP n is well known to be non-trivial [95, p. 16], so the same holds for E. Moreover, E is clearly complementary to ξ , i.e. T M/ξ ∼ = E, since n n n yj dxj yk ∂xk = yj2 = 0. j =0

k=0

j =0

This proves that ξ is not coorientable. To sum up, in the example above we have one of the following two situations: • If n is odd, then M is orientable; ξ is neither orientable nor coorientable. • If n is even, then M is not orientable; ξ is not coorientable, but it is orientable.

Contact geometry

325

We close this section with the definition of the most important types of submanifolds. D EFINITION 2.16. Let (M, ξ ) be a contact manifold. (i) A submanifold L of (M, ξ ) is called an isotropic submanifold if Tx L ⊂ ξx for all x ∈ L. (ii) A submanifold M of M with contact structure ξ is called a contact submanifold if T M ∩ ξ |M = ξ . Observe that if ξ = ker α and i : M → M denotes the inclusion map, then the condition for (M , ξ ) to be a contact submanifold of (M, ξ ) is that ξ = ker(i ∗ α). In particular, ξ ⊂ ξ |M is a symplectic sub-bundle with respect to the symplectic bundle structure on ξ given by dα. The following is a manifestation of the maximal non-integrability of contact structures. P ROPOSITION 2.17. Let (M, ξ ) be a contact manifold of dimension 2n + 1 and L an isotropic submanifold. Then dim L n. P ROOF. Write i for the inclusion of L in M and let α be an (at least locally defined) contact form defining ξ . Then the condition for L to be isotropic becomes i ∗ α ≡ 0. It follows that i ∗ dα ≡ 0. In particular, Tp L ⊂ ξp is an isotropic subspace of the symplectic vector space (ξp , dα|ξp ), i.e. a subspace on which the symplectic form restricts to zero. From Linear Algebra we know that this implies dim Tp L (dim ξp )/2 = n. D EFINITION 2.18. An isotropic submanifold L ⊂ (M 2n+1 , ξ ) of maximal possible dimension n is called a Legendrian submanifold. In particular, in a 3-dimensional contact manifold there are two distinguished types of knots: Legendrian knots on the one hand, transverse1 knots on the other, i.e. knots that are everywhere transverse to the contact structure. If ξ is cooriented by a contact form α and γ : S 1 → (M, ξ = ker α) is oriented, one can speak of a positively or negatively transverse knot, depending on whether α(γ˙ ) > 0 or α(γ˙ ) < 0.

2.2. Gray stability and the Moser trick The Gray stability theorem that we are going to prove in this section says that there are no non-trivial deformations of contact structures on closed manifolds. In fancy language, this means that contact structures on closed manifolds have discrete moduli. First a preparatory lemma. L EMMA 2.19. Let ωt , t ∈ [0, 1], be a smooth family of differential k-forms on a manifold M and (ψt )t∈[0,1] an isotopy of M. Define a time-dependent vector field Xt on M by 1 Some people like to call them ‘transversal knots’, but I adhere to J.H.C. Whitehead’s dictum, as quoted in [64]:

“Transversal is a noun; the adjective is transverse.”

326

H. Geiges

Xt ◦ ψt = ψ˙ t , where the dot denotes derivative with respect to t (so that ψt is the flow of Xt ). Then d (ψ ∗ ωt ) = ψt∗ (ω˙ t + LXt ωt ). dt t P ROOF. For a time-independent k-form ω we have d (ψ ∗ ω) = ψt∗ (LXt ω). dt t This follows by observing that (i) the formula holds for functions, (ii) if it holds for differential forms ω and ω , then also for ω ∧ ω , (iii) if it holds for ω, then also for dω, (iv) locally functions and differentials of functions generate the algebra of differential forms. We then compute ψ ∗ ωt+h − ψt∗ ωt d (ψt∗ ωt ) = lim t+h h→0 dt h = lim

∗ ω ∗ ∗ ∗ ψt+h t+h − ψt+h ωt + ψt+h ωt − ψt ωt

h→0

∗ = lim ψt+h h→0

h ψ ∗ ωt − ψt∗ ωt ωt+h − ωt + lim t+h h→0 h h

= ψt∗ (ω˙ t + LXt ωt ). For that last equality observe (regarding the second summand) that ψt+h = ψht ◦ ψt , where ψht denotes, for fixed t and time-variable h, the flow of the time-dependent vector field Xht := Xt+h ; then apply the result for time-independent k-forms. T HEOREM 2.20 (Gray stability). Let ξt , t ∈ [0, 1], be a smooth family of contact structures on a closed manifold M. Then there is an isotopy (ψt )t∈[0,1] of M such that T ψt (ξ0 ) = ξt

for each t ∈ [0, 1].

P ROOF. The simplest proof of this result rests on what is known as the Moser trick, introduced by J. Moser [96] in the context of stability results for (equicohomologous) volume and symplectic forms. J. Gray’s original proof [61] was based on deformation theory à la Kodaira–Spencer. The idea of the Moser trick is to assume that ψt is the flow of a timedependent vector field Xt . The desired equation for ψt then translates into an equation for Xt . If that equation can be solved, the isotopy ψt is found by integrating Xt ; on a closed manifold the flow of Xt will be globally defined.

Contact geometry

327

Let αt be a smooth family of 1-forms with ker αt = ξt . The equation in the theorem then translates into ψt∗ αt = λt α0 , where λt : M → R+ is a suitable smooth family of smooth functions. Differentiation of this equation with respect to t yields, with the help of the preceding lemma, ψt∗ (α˙ t + LXt αt ) = λ˙ t α0 =

λ˙ t ∗ ψ αt , λt t

or, with the help of Cartan’s formula LX = d ◦ iX + iX ◦ d and with µt =

−1 d dt (log λt ) ◦ ψt ,

ψt∗ α˙ t + d αt (Xt ) + iXt dαt = ψt∗ (µt αt ). If we choose Xt ∈ ξt , this equation will be satisfied if α˙ t + iXt dαt = µt αt .

(2.1)

Plugging in the Reeb vector field Rαt gives α˙ t (Rαt ) = µt .

(2.2)

So we can use (2.2) to define µt , and then the non-degeneracy of dαt |ξt and the fact that Rαt ∈ ker(µt αt − α˙ t ) allow us to find a unique solution Xt ∈ ξt of (2.1). R EMARK 2.21. (1) Contact forms do not satisfy stability, that is, in general one cannot find an isotopy ψt such that ψt∗ αt = α0 . For instance, consider the following family of contact forms on S 3 ⊂ R4 : αt = (x1 dy1 − y1 dx1 ) + (1 + t)(x2 dy2 − y2 dx2 ), where t 0 is a real parameter. The Reeb vector field of αt is Rαt = (x1 ∂y1 − y1 ∂x1 ) +

1 (x2 ∂y2 − y2 ∂x2 ). 1+t

The flow of Rα0 defines the Hopf fibration, in particular all orbits of Rα0 are closed. For t ∈ R+ \ Q, on the other hand, Rαt has only two periodic orbits. So there can be no isotopy with ψt∗ αt = α0 , because such a ψt would also map Rα0 to Rαt . (2) Y. Eliashberg [25] has shown that on the open manifold R3 there are likewise no non-trivial deformations of contact structures, but on S 1 × R2 there does exist a continuum of non-equivalent contact structures. (3) For further applications of this theorem it is useful to observe that at points p ∈ M with α˙ t,p identically zero in t we have Xt (p) ≡ 0, so such points remain stationary under the isotopy ψt .

328

H. Geiges

2.3. Contact Hamiltonians A vector field X on the contact manifold (M, ξ = ker α) is called an infinitesimal automorphism of the contact structure if the local flow of X preserves ξ (the study of such automorphisms was initiated by P. Libermann, cf. [80]). By slight abuse of notation, we denote this flow by ψt ; if M is not closed, ψt (for a fixed t = 0) will not in general be defined on all of M. The condition for X to be an infinitesimal automorphism can be written as T ψt (ξ ) = ξ , which is equivalent to LX α = λα for some function λ : M → R (notice that this condition is independent of the choice of 1-form α defining ξ ). The local flow of X preserves α if and only if LX α = 0. T HEOREM 2.22. With a fixed choice of contact form α there is a one-to-one correspondence between infinitesimal automorphisms X of ξ = ker α and smooth functions H : M → R. The correspondence is given by • X → HX = α(X); • H → XH , defined uniquely by α(XH ) = H and iXH dα = dH (Rα )α − dH . The fact that XH is uniquely defined by the equations in the theorem follows as in the preceding section from the fact that dα is non-degenerate on ξ and Rα ∈ ker(dH (Rα )α − dH ). P ROOF. Let X be an infinitesimal automorphism of ξ . Set HX = α(X) and write dHX + iX dα = LX α = λα with λ : M → R. Applying this last equation to Rα yields dHX (Rα ) = λ. So X satisfies the equations α(X) = HX and iX dα = dHX (Rα )α − dHX . This means that XHX = X. Conversely, given H : M → R and with XH as defined in the theorem, we have LXH α = iXH dα + d α(XH ) = dH (Rα )α, so XH is an infinitesimal automorphism of ξ . Moreover, it is immediate from the defini tions that HXH = α(XH ) = H . C OROLLARY 2.23. Let (M, ξ = ker α) be a closed contact manifold and Ht : M → R, t ∈ [0, 1], a smooth family of functions. Let Xt = XHt be the corresponding family of infinitesimal automorphisms of ξ (defined via the correspondence described in the preceding theorem). Then the globally defined flow ψt of the time-dependent vector field Xt is a contact isotopy of (M, ξ ), that is, ψt∗ α = λt α for some smooth family of functions λt : M → R + . P ROOF. With Lemma 2.19 and the preceding proof we have d (ψt∗ α) = ψt∗ (LXt α) = ψt∗ dHt (Rα )α = µt ψt∗ α dt

Contact geometry

329

with µt = dHt (Rα ) ◦ ψt . Since ψ0 = idM (whence ψ0∗ α = α) this implies that, with t λt = exp µs ds , 0

we have ψt∗ α = λt α.

This corollary will be used in Section 2.5 to prove various isotopy extension theorems from isotopies of special submanifolds to isotopies of the ambient contact manifold. In a similar vein, contact Hamiltonians can be used to show that standard general position arguments from differential topology continue to hold in the contact geometric setting. Another application of contact Hamiltonians is a proof of the fact that the contactomorphism group of a connected contact manifold acts transitively on that manifold [12]. (See [8] for more on the general structure of contactomorphism groups.)

2.4. Darboux’s theorem and neighbourhood theorems The flexibility of contact structures inherent in the Gray stability theorem and the possibility to construct contact isotopies via contact Hamiltonians results in a variety of theorems that can be summed up as saying that there are no local invariants in contact geometry. Such theorems form the theme of the present section. In contrast with Riemannian geometry, for instance, where the local structure coming from the curvature gives rise to a rich theory, the interesting questions in contact geometry thus appear only at the global level. However, it is actually that local flexibility that allows us to prove strong global theorems, such as the existence of contact structures on certain closed manifolds. 2.4.1. Darboux’s theorem T HEOREM 2.24 (Darboux’s theorem). Let α be a contact form on the (2n + 1)-dimensional manifold M and p a point on M. Then there are coordinates x1 , . . . , xn , y1 , . . . , yn , z on a neighbourhood U ⊂ M of p such that α|U = dz +

n

xj dyj .

j =1

P ROOF. We may assume without loss of generality that M = R2n+1 and p = 0 is the origin of R2n+1 . Choose linear coordinates x1 , . . . , xn , y1 , . . . , yn , z on R2n+1 such that α(∂z ) = 1, i∂z dα = 0, 2n+1 : on T0 R dα = n dx ∧ dy . ∂ , ∂ ∈ ker α (j = 1, . . . , n), xj

yj

j =1

j

j

This is simply a matter of linear algebra (the normal form theorem for skew-symmetric forms on a vector space).

330

H. Geiges

Now set α0 = dz +

j

xj dyj and consider the family of 1-forms

αt = (1 − t)α0 + tα,

t ∈ [0, 1],

on R2n+1 . Our choice of coordinates ensures that αt = α,

dαt = dα

at the origin.

Hence, on a sufficiently small neighbourhood of the origin, αt is a contact form for all t ∈ [0, 1]. We now want to use the Moser trick to find an isotopy ψt of a neighbourhood of the origin such that ψt∗ αt = α0 . This aim seems to be in conflict with our earlier remark that contact forms are not stable, but as we shall see presently, locally this equation can always be solved. Indeed, differentiating ψt∗ αt = α0 (and assuming that ψt is the flow of some timedependent vector field Xt ) we find ψt∗ (α˙ t + LXt αt ) = 0, so Xt needs to satisfy α˙ t + d αt (Xt ) + iXt dαt = 0.

(2.3)

Write Xt = Ht Rαt + Yt with Yt ∈ ker αt . Inserting Rαt in (2.3) gives α˙ t (Rαt ) + dHt (Rαt ) = 0.

(2.4)

On a neighbourhood of the origin, a smooth family of functions Ht satisfying (2.4) can always be found by integration, provided only that this neighbourhood has been chosen so small that none of the Rαt has any closed orbits there. Since α˙t is zero at the origin, we may require that Ht (0) = 0 and dHt |0 = 0 for all t ∈ [0, 1]. Once Ht has been chosen, Yt is defined uniquely by (2.3), i.e. by α˙ t + dHt + iYt dαt = 0. Notice that with our assumptions on Ht we have Xt (0) = 0 for all t. Now define ψt to be the local flow of Xt . This local flow fixes the origin, so there it is defined for all t ∈ [0, 1]. Since the domain of definition in R × M of a local flow on a manifold M is always open (cf. [15, 8.11]), we can infer2 that ψt is actually defined for all t ∈ [0, 1] on a sufficiently small neighbourhood of the origin in R2n+1 . This concludes the proof of the theorem (strictly speaking, the local coordinates in the statement of the theorem are the coordinates xj ◦ ψ1−1 , etc.). 2 To be absolutely precise, one ought to work with a family α , t ∈ R, where α ≡ α for t ε and α ≡ α for t t t 0 1 t 1 − ε, i.e. a technical homotopy in the sense of [15]. Then Xt will be defined for all t ∈ R, and the reasoning of [15] can be applied.

Contact geometry

331

R EMARK 2.25. The proof of this result given in [1] is incomplete: It is not possible, as is suggested there, to prove the Darboux theorem for contact forms if one requires Xt ∈ ker αt . 2.4.2. Isotropic submanifolds Let L ⊂ (M, ξ = ker α) be an isotropic submanifold in a contact manifold with cooriented contact structure. Write (T L)⊥ ⊂ ξ |L for the sub-bundle of ξ |L that is symplectically orthogonal to T L with respect to the symplectic bundle structure dα|ξ . The conformal class of this symplectic bundle structure depends only on the contact structure ξ , not on the choice of contact form α defining ξ : If α is replaced by λα for some smooth function λ : M → R+ , then d(λα)|ξ = λ dα|ξ . So the bundle (T L)⊥ is determined by ξ . The fact that L is isotropic implies T L ⊂ (T L)⊥ . Following Weinstein [105], we call the quotient bundle (T L)⊥ /T L with the conformal symplectic structure induced by dα the conformal symplectic normal bundle of L in M and write CSN(M, L) = (T L)⊥ /T L. So the normal bundle N L = (T M|L )/T L of L in M can be split as NL ∼ = (T M|L )/(ξ |L ) ⊕ (ξ |L )/(T L)⊥ ⊕ CSN(M, L). Observe that if dim M = 2n + 1 and dim L = k n, then the ranks of the three summands in this splitting are 1, k and 2(n − k), respectively. Our aim in this section is to show that a neighbourhood of L in M is determined, up to contactomorphism, by the isomorphism type (as a conformal symplectic bundle) of CSN(M, L). The bundle (T M|L )/(ξ |L ) is a trivial line bundle because ξ is cooriented. The bundle (ξ |L )/(T L)⊥ can be identified with the cotangent bundle T ∗ L via the well-defined bundle isomorphism Ψ : (ξ |L )/(T L)⊥ → T ∗ L, Y → iY dα|T L . (Ψ is obviously injective and well defined by the definition of (T L)⊥ , and the ranks of the two bundles are equal.) Although Ψ is well defined on the quotient (ξ |L )/(T L)⊥ , to proceed further we need to choose an isotropic complement of (T L)⊥ in ξ |L . Restricted to each fibre ξp , p ∈ L, such an isotropic complement of (Tp L)⊥ exists. There are two ways to obtain a smooth bundle of such isotropic complements. The first would be to carry over Arnold’s corresponding discussion of Lagrangian subbundles of symplectic bundles [6] to the isotropic case in order to show that the space of isotropic complements of U ⊥ ⊂ V , where U is an isotropic subspace in a symplectic vector space V , is convex. (This argument uses generating functions for isotropic subspaces.) Then by a partition of unity argument the desired complement can be constructed on the bundle level. A slightly more pedestrian approach is to define this isotropic complement with the help of a complex bundle structure J on ξ compatible with dα (cf. Remark 2.3). The condition

332

H. Geiges

dα(X, J X) > 0 for 0 = X ∈ ξ implies that (Tp L)⊥ ∩ J (Tp L) = {0} for all p ∈ L, and so a dimension count shows that J (T L) is indeed a complement of (T L)⊥ in ξ |L . (In a similar vein, CSN(M, L) can be identified as a sub-bundle of ξ , viz., the orthogonal complement of T L ⊕ J (T L) ⊂ ξ with respect to the bundle metric dα(., J.) on ξ .) On the Whitney sum T L ⊕ T ∗ L (for any manifold L) there is a canonical symplectic bundle structure ΩL defined by ΩL,p (X + η, X + η ) = η(X ) − η (X) for X, X ∈ Tp L, η, η ∈ Tp∗ L. L EMMA 2.26. The bundle map idT L ⊕ Ψ : T L ⊕ J (T L), dα → (T L ⊕ T ∗ L, ΩL ) is an isomorphism of symplectic vector bundles. P ROOF. We only need to check that idT L ⊕Ψ is a symplectic bundle map. Let X, X ∈ Tp L and Y, Y ∈ Jp (Tp L). Write Y = Jp Z, Y = Jp Z with Z, Z ∈ Tp L. It follows that dα(Y, Y ) = dα(J Z, J Z ) = dα(Z, Z ) = 0, since L is an isotropic submanifold. For the same reason dα(X, X ) = 0. Hence dα(X + Y, X + Y ) = dα(Y, X ) − dα(Y , X) = Ψ (Y )(X ) − Ψ (Y )(X) = ΩL X + Ψ (Y ), X + Ψ (Y ) .

T HEOREM 2.27. Let (Mi , ξi ), i = 0, 1, be contact manifolds with closed isotropic submanifolds Li . Suppose there is an isomorphism of conformal symplectic normal bundles Φ : CSN(M0 , L0 ) → CSN(M1 , L1 ) that covers a diffeomorphism φ : L0 → L1 . Then φ extends to a contactomorphism ψ : N (L0 ) → N (L1 ) of suitable neighbourhoods N (Li ) of Li such that T ψ|CSN(M0 ,L0 ) and Φ are bundle homotopic (as conformal symplectic bundle isomorphisms). C OROLLARY 2.28. Diffeomorphic (closed) Legendrian submanifolds have contactomorphic neighbourhoods. P ROOF. If Li ⊂ Mi is Legendrian, then CSN(Mi , Li ) has rank 0, so the conditions in the theorem, apart from the existence of a diffeomorphism φ : L0 → L1 , are void. E XAMPLE 2.29. Let S 1 ⊂ (M 3 , ξ ) be a Legendrian knot in a contact 3-manifold. Then with a coordinate θ ∈ [0, 2π] along S 1 and coordinates x, y in slices transverse to S 1 , the contact structure cos θ dx − sin θ dy = 0 provides a model for a neighbourhood of S 1 .

Contact geometry

333

P ROOF OF T HEOREM 2.27. Choose contact forms αi for ξi , i = 0, 1, scaled in such a way that Φ is actually an isomorphism of symplectic vector bundles with respect to the symplectic bundle structures on CSN(Mi , Li ) given by dαi . Here we think of CSN(Mi , Li ) as a sub-bundle of T Mi |Li (rather than as a quotient bundle). We identify (T Mi |Li )/(ξi |Li ) with the trivial line bundle spanned by the Reeb vector field Rαi . In total, this identifies N Li = Rαi ⊕ Ji (T Li ) ⊕ CSN(Mi , Li ) as a sub-bundle of T Mi |Li . Let ΦR : Rα0 → Rα1 be the obvious bundle isomorphism defined by requiring that Rα0 (p) map to Rα1 (φ(p)). Let Ψi : Ji (T Li ) → T ∗ Li be the isomorphism defined by taking the interior product with dαi . Notice that T φ ⊕ (φ ∗ )−1 : (T L0 ⊕ T ∗ L0 , ΩL0 ) → (T L1 ⊕ T ∗ L1 , ΩL1 ) is an isomorphism of symplectic vector bundles. With Lemma 2.26 it follows that T φ ⊕ Ψ1−1 ◦ (φ ∗ )−1 ◦ Ψ0 : T L0 ⊕ J0 (T L0 ), dα0 → T L1 ⊕ J1 (T L1 ), dα1 is an isomorphism of symplectic vector bundles. Now let Φ˜ : N L0 → N L1 be the bundle isomorphism (covering φ) defined by Φ˜ = ΦR ⊕ Ψ1−1 ◦ (φ ∗ )−1 ◦ Ψ0 ⊕ Φ. Let τi : N Li → Mi be tubular maps, that is, the τ (I suppress the index i for better readability) are embeddings such that τ |L —where L is identified with the zero section of N L—is the inclusion L ⊂ M, and T τ induces the identity on N L along L (with respect to the splittings T (N L)|L = T L ⊕ N L = T M|L ). Then τ1 ◦ Φ˜ ◦ τ0−1 : N (L0 ) → N (L1 ) is a diffeomorphism of suitable neighbourhoods N (Li ) of Li that induces the bundle map T φ ⊕ Φ˜ : T M0 |L0 → T M1 |L1 . By construction, this bundle map pulls α1 back to α0 and dα1 to dα0 . Hence, α0 and (τ1 ◦ Φ˜ ◦ τ0−1 )∗ α1 are contact forms on N (L0 ) that coincide on T M0 |L0 , and so do their differentials. Now consider the family of 1-forms ∗ βt = (1 − t)α0 + t τ1 ◦ Φ˜ ◦ τ0−1 α1 ,

t ∈ [0, 1].

334

H. Geiges

On T M0 |L0 we have βt ≡ α0 and dβt ≡ dα0 . Since the contact condition α ∧ (dα)n = 0 is an open condition, we may assume—shrinking N (L0 ) if necessary—that βt is a contact form on N (L0 ) for all t ∈ [0, 1]. By the Gray stability theorem (Theorem 2.20) and Remark 2.21(3) following its proof, we find an isotopy ψt of N (L0 ), fixing L0 , such that ψt∗ βt = λt α0 for some smooth family of smooth functions λt : N (L0 ) → R+ . (Since N (L0 ) is not a closed manifold, ψt is a priori only a local flow. But on L0 it is stationary and hence defined for all t. As in the proof of the Darboux theorem (Theorem 2.24) we conclude that ψt is defined for all t ∈ [0, 1] in a sufficiently small neighbourhood of L0 , so shrinking N (L0 ) once again, if necessary, will ensure that ψt is a global flow on N (L0 ).) We conclude that ψ = τ1 ◦ Φ˜ ◦ τ0−1 ◦ ψ1 is the desired contactomorphism. R EMARK 2.30. With a little more care one can actually achieve T ψ1 = id on T M0 |L0 , which implies in particular that T ψ|CSN(M0 ,L0 ) = Φ, cf. [105]. (Remember that there is a certain freedom in constructing an isotopy via the Moser trick if the condition Xt ∈ ξt is dropped.) The key point is the generalised Poincaré lemma, cf. [80, p. 361], which allows us to write a closed differential form γ given in a neighbourhood of the zero section of a bundle and vanishing along that zero section as an exact form γ = dη with η and its partial derivatives with respect to all coordinates (in any chart) vanishing along the zero section. This lemma is applied first to γ = d(β1 − β0 ), in order to find (with the symplectic Moser trick) a diffeomorphism σ of a neighbourhood of L0 ⊂ M0 with T σ = id on T M0 |L0 and such that dβ0 = d(σ ∗ β1 ). It is then applied once again to γ = β0 − σ ∗ β1 . (The proof of the symplectic neighbourhood theorem in [92] appears to be incomplete in this respect.) E XAMPLE 2.31. Let M0 = M1 = R3 with contact forms α0 = dz + x dy and α1 = dz + (x + y) dy and L0 = L1 = 0 the origin in R3 . Thus CSN(M0 , L0 ) = CSN(M1 , L1 ) = span{∂x , ∂y } ⊂ T0 R3 . We take Φ = idCSN . Set αt = dz + (x + ty) dy. The Moser trick with Xt ∈ ker αt yields Xt = −y ∂x , and hence ψt (x, y, z) = (x − ty, y, z). Then 1 −1 0 T ψ1 = 0 1 0 , 0 0 1 which does not restrict to Φ on CSN. However, a different solution for ψt∗ αt = α0 is ψt (x, y, z) = (x, y, z − ty 2 /2), found by integrating Xt = −y 2 ∂z /2 (a multiple of the Reeb vector field of αt ). Here we get 1 0 0 T ψ1 = 0 1 0 , 0 −y 1 hence T ψ1 |T0 R3 = id, so in particular T ψ1 |CSN = Φ.

Contact geometry

335

2.4.3. Contact submanifolds Let (M , ξ = ker α ) ⊂ (M, ξ = ker α) be a contact submanifold, that is, T M ∩ ξ |M = ξ . As before we write (ξ )⊥ ⊂ ξ |M for the symplectically orthogonal complement of ξ in ξ |M . Since M is a contact submanifold (so ξ is a symplectic sub-bundle of (ξ |M , dα)), we have T M ⊕ (ξ )⊥ = T M|M , i.e. we can identify (ξ )⊥ with the normal bundle N M . Moreover, dα induces a conformal symplectic structure on (ξ )⊥ , so we call (ξ )⊥ the conformal symplectic normal bundle of M in M and write CSN(M, M ) = (ξ )⊥ . T HEOREM 2.32. Let (Mi , ξi ), i = 0, 1, be contact manifolds with compact contact submanifolds (Mi , ξi ). Suppose there is an isomorphism of conformal symplectic normal bundles Φ : CSN(M0 , M0 ) → CSN(M1 , M1 ) that covers a contactomorphism φ : (M0 , ξ0 ) → (M1 , ξ1 ). Then φ extends to a contactomorphism ψ of suitable neighbourhoods N (Mi ) of Mi such that T ψ|CSN(M0 ,M0 ) and Φ are bundle homotopic (as conformal symplectic bundle isomorphisms). E XAMPLE 2.33. A particular instance of this theorem is the case of a transverse knot in a contact manifold (M, ξ ), i.e. an embedding S 1 → (M, ξ ) transverse to ξ . Since the sym2n plectic group Sp(2n) n of linear transformations of R preserving the standard symplectic structure ω0 = i=1 dxi ∧ dyi is connected, there is only one conformal symplectic R2n bundle over S 1 up to conformal equivalence. A model for the neighbourhood of a transverse knot is given by

n (xi dyi − yi dxi ) , S × R , ξ = ker dθ +

1

2n

i=1

where θ denotes the S 1 -coordinate; the theorem says that in suitable local coordinates the neighbourhood of any transverse knot looks like this model. P ROOF OF T HEOREM 2.32. As in the proof of Theorem 2.27 it is sufficient to find contact forms αi on Mi and a bundle map T M0 |M0 → T M1 |M1 , covering φ and inducing Φ, that pulls back α1 to α0 and dα1 to dα0 ; the proof then concludes as there with a stability argument. For this we need to make a judicious choice of αi . The essential choice is made separately on each Mi , so I suppress the subscript i for the time being. Choose a contact form α for ξ on M . Write R for the Reeb vector field of α . Given any contact form α for ξ on M we may first scale it such that α(R ) ≡ 1 along M . Then α|T M = α , and hence dα|T M = dα . We now want to scale α further such that its Reeb vector field R coincides with R along M . To this end it is sufficient to find a smooth function f : M → R+ with

336

H. Geiges

f |M ≡ 1 and iR d(f α) ≡ 0 on T M|M . This last equation becomes 0 = iR d(f α) = iR (df ∧ α + f dα) = −df + iR dα

on T M|M .

Since iR dα|T M = iR dα ≡ 0, such an f can be found. The choices of α0 and α1 cannot be made independently of each other; we may first choose α1 , say, and then define α0 = φ ∗ α1 . Then define α0 , α1 as described and scale Φ such that it is a symplectic bundle isomorphism of ⊥ (ξ0 ) , dα0 → (ξ1 )⊥ , dα1 . Then T φ ⊕ Φ : T M0 |M0 → T M1 |M1 is the desired bundle map that pulls back α1 to α0 and dα1 to dα0 .

R EMARK 2.34. The condition that Ri ≡ Ri along M is necessary for ensuring that (T φ ⊕ Φ)(R0 ) = R1 , which guarantees (with the other stated conditions) that (T φ ⊕ Φ)∗ (dα1 ) = dα0 . The condition dαi |T Mi = dαi and the described choice of Φ alone would only give (T φ ⊕ Φ)∗ (dα1 |ξ1 ) = dα0 |ξ0 . 2.4.4. Hypersurfaces Let S be an oriented hypersurface in a contact manifold (M, ξ = ker α) of dimension 2n + 1. In a neighbourhood of S in M, which we can identify with S × R (and S with S × {0}), the contact form α can be written as α = βr + ur dr, where βr , r ∈ R, is a smooth family of 1-forms on S and ur : S → R a smooth family of functions. The contact condition α ∧ (dα)n = 0 then becomes 0 = α ∧ (dα)n = (βr + ur dr) ∧ (dβr − β˙r ∧ dr + dur ∧ dr)n = (−nβr ∧ β˙r + nβr ∧ dur + ur dβr ) ∧ (dβr )n−1 ∧ dr,

(2.5)

where the dot denotes derivative with respect to r. The intersection T S ∩ (ξ |S ) determines a distribution (of non-constant rank) of subspaces of T S. If α is written as above, this distribution is given by the kernel of β0 , and hence, at a given p ∈ S, defines either the full tangent space Tp S (if β0,p = 0) or a 1-codimensional subspace both of Tp S and ξp (if β0,p = 0). In the former case, the symplectically orthogonal complement (Tp S ∩ ξp )⊥ (with respect to the conformal symplectic structure dα on ξp ) is {0}; in the latter case, (Tp S ∩ ξp )⊥ is a 1-dimensional subspace of ξp contained in Tp S ∩ ξp . From that it is intuitively clear what one should mean by a ‘singular 1-dimensional foliation’, and we make the following somewhat provisional definition.

Contact geometry

337

Fig. 2. The characteristic foliation on S 2 ⊂ (R3 , ξ2 ).

D EFINITION 2.35. The characteristic foliation Sξ of a hypersurface S in (M, ξ ) is the singular 1-dimensional foliation of S defined by (T S ∩ ξ |S )⊥ . E XAMPLE 2.36. If dim M = 3 and dim S = 2, then (Tp S ∩ ξp )⊥ = Tp S ∩ ξp at the points p ∈ S where Tp S ∩ ξp is 1-dimensional. Figure 2 shows the characteristic foliation of the unit 2-sphere in (R3 , ξ2 ), where ξ2 denotes the standard contact structure of Example 2.7: The only singular points are (0, 0, ±1); away from these points the characteristic foliation is spanned by (y − xz) ∂x − (x + yz) ∂y + x 2 + y 2 ∂z .

The following lemma helps to clarify the notion of singular 1-dimensional foliation. L EMMA 2.37. Let β0 be the 1-form induced on S by a contact form α defining ξ , and let Ω be a volume form on S. Then Sξ is defined by the vector field X satisfying iX Ω = β0 ∧ (dβ0 )n−1 . P ROOF. First of all, we observe that β0 ∧ (dβ0 )n−1 = 0 outside the zeros of β0 : Arguing by contradiction, assume β0,p = 0 and β0 ∧ (dβ0 )n−1 |p = 0 at some p ∈ S. Then (dβ0 )n |p = 0 by (2.5). On the codimension 1 subspace ker β0,p of Tp S the symplectic form dβ0,p has maximal rank n − 1. It follows that β0 ∧ (dβ0 )n−1 |p = 0 after all, a contradiction. Next we want to show that X ∈ ker β0 . We observe 0 = iX (iX Ω) = β0 (X)(dβ0 )n−1 − (n − 1)β0 ∧ iX dβ0 ∧ (dβ0 )n−2 . Taking the exterior product of this equation with β0 we get β0 (X)β0 ∧ (dβ0 )n−1 = 0. By our previous consideration this implies β0 (X) = 0. It remains to show that for β0,p = 0 we have dβ0 X(p), v = 0 for all v ∈ Tp S ∩ ξp .

(2.6)

338

H. Geiges

For n = 1 this is trivially satisfied, because in that case v is a multiple of X(p). I suppress the point p in the following calculation, where we assume n 2. From (2.6) and with β0 (X) = 0 we have β0 ∧ iX dβ0 ∧ (dβ0 )n−2 = 0.

(2.7)

Taking the interior product with v ∈ T S ∩ ξ yields −dβ0 (X, v)β0 ∧ (dβ0 )n−2 + (n − 2)β0 ∧ iX dβ0 ∧ iv dβ0 ∧ (dβ0 )n−3 = 0. (Thanks to the coefficient n − 2 the term (dβ0 )n−3 is not a problem for n = 2.) Taking the exterior product of that last equation with dβ0 , and using (2.7), we find dβ0 (X, v)β0 ∧ (dβ0 )n−1 = 0, and thus dβ0 (X, v) = 0.

R EMARK 2.38. (1) We can now give a more formal definition of ‘singular 1-dimensional foliation’ as an equivalence class of vector fields [X], where X is allowed to have zeros and [X] = [X ] if there is a nowhere zero function on all of S such that X = f X. Notice that the non-integrability of contact structures and the reasoning at the beginning of the proof of the lemma imply that the zero set of X does not contain any open subsets of S. (2) If the contact structure ξ is cooriented rather than just coorientable, so that α is well defined up to multiplication with a positive function, then this lemma allows to give an orientation to the characteristic foliation: Changing α to λα with λ : M → R+ will change β0 ∧ (dβ0 )n−1 by a factor λn . We now restrict attention to surfaces in contact 3-manifolds, where the notion of characteristic foliation has proved to be particularly useful. The following theorem is due to E. Giroux [52]. T HEOREM 2.39 (Giroux). Let Si be closed surfaces in contact 3-manifolds (Mi , ξi ), i = 0, 1 (with ξi coorientable), and φ : S0 → S1 a diffeomorphism with φ(S0,ξ0 ) = S1,ξ1 as oriented characteristic foliations. Then there is a contactomorphism ψ : N (S0 ) → N (S1 ) of suitable neighbourhoods N (Si ) of Si with ψ(S0 ) = S1 and such that ψ|S0 is isotopic to φ via an isotopy preserving the characteristic foliation. P ROOF. By passing to a double cover, if necessary, we may assume that the Si are orientable hypersurfaces. Let αi be contact forms defining ξi . Extend φ to a diffeomorphism (still denoted φ) of neighbourhoods of Si and consider the contact forms α0 and φ ∗ α1 on a neighbourhood of S0 , which we may identify with S0 × R. By rescaling α1 we may assume that α0 and φ ∗ α1 induce the same form β0 on S0 ≡ S0 × {0}, and hence also the same form dβ0 .

Contact geometry

339

Observe that the expression on the right-hand side of Eq. (2.5) is linear in β˙r and ur . This implies that convex linear combinations of solutions of (2.5) (for n = 1) with the same β0 (and dβ0 ) will again be solutions of (2.5) for sufficiently small r. This reasoning applies to αt := (1 − t)α0 + tφ ∗ α1 ,

t ∈ [0, 1].

(I hope the reader will forgive the slight abuse of notation, with α1 denoting both a form on M1 and its pull-back φ ∗ α1 to M0 .) As is to be expected, we now use the Moser trick to find an isotopy ψt with ψt∗ αt = λt α0 , just as in the proof of Gray stability (Theorem 2.20). In particular, we require as there that the vector field Xt that we want to integrate to the flow ψt lie in the kernel of αt . On T S0 we have α˙ t ≡ 0 (thanks to the assumption that α0 and φ ∗ α1 induce the same form β0 on S0 ). In particular, if v is a vector in S0,ξ0 , then by Eq. (2.1) we have dαt (Xt , v) = 0, which implies that Xt is a multiple of v, hence tangent to S0,ξ0 . This shows that the flow of Xt preserves S0 and its characteristic foliation. More formally, we have LXt αt = d αt (Xt ) + iXt dαt = iXt dαt , so with v as above we have LXt αt (v) = 0, which shows that LXt αt |T S0 is a multiple of α0 |T S0 = β0 . This implies that the (local) flow of Xt changes β0 by a conformal factor. Since S0 is closed, the local flow of Xt restricted to S0 integrates up to t = 1, and so the same is true3 in a neighbourhood of S0 . Then ψ = φ ◦ ψ1 will be the desired diffeomorphism N (S0 ) → N (S1 ). As observed previously in the proof of Darboux’s theorem for contact forms, the Moser trick allows more flexibility if we drop the condition αt (Xt ) = 0. We are now going to exploit this extra freedom to strengthen Giroux’s theorem slightly. This will be important later on when we want to extend isotopies of hypersurfaces. T HEOREM 2.40. Under the assumptions of the preceding theorem we can find ψ : N (S0 ) → N (S1 ) satisfying the stronger condition that ψ|S0 = φ. P ROOF. We want to show that the isotopy ψt of the preceding proof may be assumed to fix S0 pointwise. As there, we may assume α˙ t |T S0 ≡ 0. If the condition that Xt be tangent to ker αt is dropped, the condition Xt needs to satisfy so that its flow will pull back αt to λt α0 is α˙ t + d αt (Xt ) + iXt dαt = µt αt , where µt and λt are related by µt = theorem (Theorem 2.20).

−1 d dt (log λt ) ◦ ψt ,

(2.8) cf. the proof of the Gray stability

3 Cf. the proof (and the footnote therein) of Darboux’s theorem (Theorem 2.24).

340

H. Geiges

Write Xt = Ht Rt + Yt with Rt the Reeb vector field of αt and Yt ∈ ξt = ker αt . Then condition (2.8) translates into α˙ t + dHt + iYt dαt = µt αt .

(2.9)

For given Ht one determines µt from this equation by inserting the Reeb vector field Rt ; the equation then admits a unique solution Yt ∈ ker αt because of the non-degeneracy of dαt |ξt . Our aim now is to ensure that Ht ≡ 0 on S0 and Yt ≡ 0 along S0 . The latter we achieve by imposing the condition α˙ t + dHt = 0

along S0

(2.10)

(which entails with (2.9) that µt |S0 ≡ 0). The conditions Ht ≡ 0 on S0 and (2.10) can be simultaneously satisfied thanks to α˙ t |T S0 ≡ 0. We can therefore find a smooth family of smooth functions Ht satisfying these conditions, and then define Yt by (2.9). The flow of the vector field Xt = Ht Rt + Yt then defines an isotopy ψt that fixes S0 pointwise (and thus is defined for all t ∈ [0, 1] in a neighbourhood of S0 ). Then ψ = φ ◦ ψ1 will be the diffeomorphism we wanted to construct. 2.4.5. Applications Perhaps the most important consequence of the neighbourhood theorems proved above is that they allow us to perform differential topological constructions such as surgery or similar cutting and pasting operations in the presence of a contact structure, that is, these constructions can be carried out on a contact manifold in such a way that the resulting manifold again carries a contact structure. One such construction that I shall explain in detail in Section 3 is the surgery of contact 3-manifolds along transverse knots, which enables us to construct a contact structure on every closed, orientable 3-manifold. The concept of characteristic foliation on surfaces in contact 3-manifolds has proved seminal for the classification of contact structures on 3-manifolds, although it has recently been superseded by the notion of dividing curves. It is clear that Theorem 2.39 can be used to cut and paste contact manifolds along hypersurfaces with the same characteristic foliation. What actually makes this useful in dimension 3 is that there are ways to manipulate the characteristic foliation of a surface by isotoping that surface inside the contact 3-manifold. The most important result in that direction is the Elimination Lemma proved by Giroux [52]; an improved version is due to D. Fuchs, see [26]. This lemma says that under suitable assumptions it is possible to cancel singular points of the characteristic foliation in pairs by a C 0 -small isotopy of the surface (specifically: an elliptic and a hyperbolic point of the same sign—the sign being determined by the matching or non-matching of the orientation of the surface S and the contact structure ξ at the singular point of Sξ ). For instance, Eliashberg [24] has shown that if a contact 3-manifold (M, ξ ) contains an embedded disc D such that Dξ has a limit cycle, then one can actually find a so-called overtwisted disc: an embedded disc D with boundary ∂D tangent to ξ (but D transverse

Contact geometry

341

to ξ along ∂D, i.e. no singular points of Dξ on ∂D) and with Dξ having exactly one singular point (of elliptic type); see Section 3.6. Moreover, in the generic situation it is possible, given surfaces S ⊂ (M, ξ ) and S ⊂ (M , ξ ) with Sξ homeomorphic to Sξ , to perturb one of the surfaces so as to get diffeomorphic characteristic foliations. Chapter 8 of [1] contains a section on surfaces in contact 3-manifolds, and in particular a proof of the Elimination Lemma. Further introductory reading on the matter can be found in the lectures of J. Etnyre [35]; of the sources cited above I recommend [26] as a starting point. In [52] Giroux initiated the study of convex surfaces in contact 3-manifolds. These are surfaces S with an infinitesimal automorphism X of the contact structure ξ with X transverse to S. For such surfaces, it turns out, much less information than the characteristic foliation Sξ is needed to determine ξ in a neighbourhood of S, viz., only the so-called dividing set of Sξ . This notion lies at the centre of most of the recent advances in the classification of contact structures on 3-manifolds [55,71,72]. A brief introduction to convex surface theory can be found in [35].

2.5. Isotopy extension theorems In this section we show that the isotopy extension theorem of differential topology—an isotopy of a closed submanifold extends to an isotopy of the ambient manifold—remains valid for the various distinguished submanifolds of contact manifolds. The neighbourhood theorems proved above provide the key to the corresponding isotopy extension theorems. For simplicity, I assume throughout that the ambient contact manifold M is closed; all isotopy extension theorems remain valid if M has non-empty boundary ∂M, provided the isotopy stays away from the boundary. In that case, the isotopy of M found by extension keeps a neighbourhood of ∂M fixed. A further convention throughout is that our ambient isotopies ψt are understood to start at ψ0 = idM . 2.5.1. Isotropic submanifolds An embedding j : L → (M, ξ = ker α) is called isotropic if j (L) is an isotropic submanifold of (M, ξ ), i.e. everywhere tangent to the contact structure ξ . Equivalently, one needs to require j ∗ α ≡ 0. T HEOREM 2.41. Let jt : L → (M, ξ = ker α), t ∈ [0, 1], be an isotopy of isotropic embeddings of a closed manifold L in a contact manifold (M, ξ ). Then there is a compactly supported contact isotopy ψt : M → M with ψt (j0 (L)) = jt (L). P ROOF. Define a time-dependent vector field Xt along jt (L) by X t ◦ jt =

d jt . dt

To simplify notation later on, we assume that L is a submanifold of M and j0 the inclusion L ⊂ M. Our aim is to find a (smooth) family of compactly supported, smooth functions

342

H. Geiges

H˜ t : M → R whose Hamiltonian vector field X˜ t equals Xt along jt (L). Recall that X˜ t is defined in terms of H˜ t by α(X˜ t ) = H˜ t ,

iX˜ t dα = d H˜ t (Rα )α − d H˜ t ,

where, as usual, Rα denotes the Reeb vector field of α. Hence, we need α(Xt ) = H˜ t ,

iXt dα = d H˜ t (Rα )α − d H˜ t

along jt (L).

(2.11)

Write Xt = Ht Rα + Yt with Ht : jt (L) → R and Yt ∈ ker α. To satisfy (2.11) we need H˜ t = Ht

along jt (L).

(2.12)

This implies d H˜ t (v) = dHt (v)

for v ∈ T jt (L) .

Since jt is an isotopy of isotropic embeddings, we have T (jt (L)) ⊂ ker α. So a prerequisite for (2.11) is that dα(Xt , v) = −dHt (v)

for v ∈ T jt (L) .

(2.13)

We have dα(Xt , v) + dHt (v) = dα(Xt , v) + d α(Xt ) (v) = iv iXt dα + d(iXt α) = iv (LXt α), so Eq. (2.13) is equivalent to iv (LXt α) = 0

for v ∈ T jt (L) .

But this is indeed tautologically satisfied: The fact that jt is an isotopy of isotropic embedd (jt∗ α) = jt∗ (LXt α), as desired. dings can be written as jt∗ α ≡ 0; this in turn implies 0 = dt This means that we can define H˜ t by prescribing the value of H˜ t along jt (L) (with (2.12)) and the differential of H˜ t along jt (L) (with (2.11)), where we are free to impose d H˜ t (Rα ) = 0, for instance. The calculation we just performed shows that these two requirements are consistent with each other. Any function satisfying these requirements along jt (L) can be smoothed out to zero outside a tubular neighbourhood of jt (L), and the Hamiltonian flow of this H˜ t will be the desired contact isotopy extending jt . One small technical point is to ensure that the resulting family of functions H˜ t will be smooth in t. To achieve this, we proceed as follows. Set Mˆ = M × [0, 1] and Lˆ =

q∈L, t∈[0,1]

jt (q), t ,

Contact geometry

343

ˆ Let g be an auxiliary Riemannian metric on M with so that Lˆ is a submanifold of M. respect to which Rα is orthogonal to ker α. Identify the normal bundle N Lˆ of Lˆ in Mˆ with a sub-bundle of T Mˆ by requiring its fibre at a point (p, t) ∈ Lˆ to be the g-orthogonal subspace of Tp (jt (L)) in Tp M. Let τ : N Lˆ → Mˆ be a tubular map. Now define a smooth function Hˆ : N Lˆ → R as follows, where (p, t) always denotes a ˆ point of Lˆ ⊂ N L. ˆ • H (p, t) = α(Xt ), • d Hˆ (p,t) (Rα ) = 0, ˆ • d Hˆ (p,t) (v) = −dα(Xt , v) for v ∈ ker αp ⊂ Tp M ⊂ T(p,t) M, ˆ ˆ ˆ • H is linear on the fibres of N L → L. Let χ : Mˆ → [0, 1] be a smooth function with χ ≡ 0 outside a small neighbourhood N0 ⊂ ˆ of Lˆ and χ ≡ 1 in a smaller neighbourhood N1 ⊂ N0 of L. ˆ For (p, t) ∈ M, ˆ set τ (N L) ˆ −1 H˜ t (p) = χ(p, t)H (τ (p, t)) 0

ˆ for (p, t) ∈ τ (N L), ˆ for (p, t) ∈ / τ (N L).

This is smooth in p and t, and the Hamiltonian flow ψt of H˜ t (defined globally since H˜ t is compactly supported) is the desired contact isotopy. 2.5.2. Contact submanifolds An embedding j : (M , ξ ) → (M, ξ ) is called a contact embedding if j (M ), Tj (ξ ) is a contact submanifold of (M, ξ ), i.e. T j (M) ∩ ξ |j (M) = Tj (ξ ). If ξ = ker α, this can be reformulated as ker j ∗ α = ξ . T HEOREM 2.42. Let jt : (M , ξ ) → (M, ξ ), t ∈ [0, 1], be an isotopy of contact embeddings of the closed contact manifold (M , ξ ) in the contact manifold (M, ξ ). Then there is a compactly supported contact isotopy ψt : M → M with ψt (j0 (M )) = jt (M ). P ROOF. In the proof of this theorem we follow a slightly different strategy from the one in the isotropic case. Instead of directly finding an extension of the Hamiltonian Ht : jt (M ) → R, we first use the neighbourhood theorem for contact submanifolds to extend jt to an isotopy of contact embeddings of tubular neighbourhoods. Again we assume that M is a submanifold of M and j0 the inclusion M ⊂ M. As earlier, NM denotes the normal bundle of M in M. We also identify M with the zero section of N M , and we use the canonical identification T (NM )|M = T M ⊕ N M .

344

H. Geiges

By the usual isotopy extension theorem from differential topology we find an isotopy φt : N M → M with φt |M = jt . Choose contact forms α, α defining ξ and ξ , respectively. Define αt = φt∗ α. Then T M ∩ ker αt = ξ . Let R denote the Reeb vector field of α . Analogous to the proof of Theorem 2.32, we first find a smooth family of smooth functions gt : M → R+ such that gt αt |T M = α , and then a family ft : N M → R+ with ft |M ≡ 1 and dft = iR d(gt αt )

on T (N M )|M .

Then βt = ft gt αt is a family of contact forms on N M representing the contact structure ker(φt∗ α) and with the properties βt |T M = α , dβt |T M = dα , ker(dβt ) = R

along M .

The family (N M , dβt ) of symplectic vector bundles may be thought of as a symplectic vector bundle over M × [0, 1], which is necessarily isomorphic to a bundle pulled back from M × {0} (cf. [74, Corollary 3.4.4]). In other words, there is a smooth family of symplectic bundle isomorphisms Φt : (N M , dβ0 ) → (N M , dβt ). Then idT M ⊕ Φt : T (NM )|M → T (N M )|M is a bundle map that pulls back βt to β0 and dβt to dβ0 . By the now familiar stability argument we find a smooth family of embeddings ϕt : N (M ) → N M for some neighbourhood N (M ) of the zero section M in N M with ϕ0 = inclusion, ϕt |M = idM and ϕt∗ βt = λt β0 , where λt : N (M ) → R+ . This means that φt ◦ ϕt : N (M ) → M is a smooth family of contact embeddings of (N (M ), ker β0 ) in (M, ξ ). Define a time-dependent vector field Xt along φt ◦ ϕt (N (M )) by Xt ◦ φt ◦ ϕt =

d (φt ◦ ϕt ). dt

Contact geometry

345

This Xt is clearly an infinitesimal automorphism of ξ : By differentiating the equation ϕt∗ φt∗ α = µt φ0∗ α (where µt : N (M ) → R+ ) with respect to t we get ϕt∗ φt∗ (LXt α) = µ˙ t φ0∗ α =

µ˙ t ∗ ∗ ϕ φ α, µt t t

so LXt α is a multiple of α (since φt ◦ ϕt is a diffeomorphism onto its image). By the theory of contact Hamiltonians, Xt is the Hamiltonian vector field of a Hamiltonian function Hˆ t defined on φt ◦ ϕt (N (M )). Cut off this function with a bump function so as to obtain Ht : M → R with Ht ≡ Hˆ t near φt ◦ ϕt (M ) and Ht ≡ 0 outside a slightly larger neighbourhood of φt ◦ ϕt (M ). Then the Hamiltonian flow ψt of Ht satisfies our requirements. 2.5.3. Surfaces in 3-manifolds T HEOREM 2.43. Let jt : S → (M, ξ = ker α), t ∈ [0, 1], be an isotopy of embeddings of a closed surface S in a 3-dimensional contact manifold (M, ξ ). If all jt induce the same characteristic foliation on S, then there is a compactly supported isotopy ψt : M → M with ψt (j0 (S)) = jt (S). P ROOF. Extend jt to a smooth family of embeddings φt : S × R → M, and identify S with S × {0}. The assumptions say that all φt∗ α induce the same characteristic foliation on S. By the proof of Theorem 2.40 and in analogy with the proof of Theorem 2.42 we find a smooth family of embeddings ϕt : S × (−ε, ε) → S × R for some ε > 0 with ϕ0 = inclusion, ϕt |S×{0} = idS and ϕt∗ φt∗ α = λt φ0∗ α, where λt : S × (−ε, ε) → R+ . In other words, φt ◦ ϕt is a smooth family of contact embeddings of (S × (−ε, ε), ker φ0∗ α) in (M, ξ ). The proof now concludes exactly as the proof of Theorem 2.42.

2.6. Approximation theorems A further manifestation of the (local) flexibility of contact structures is the fact that a given submanifold can, under fairly weak (and usually obvious) topological conditions, be approximated (typically C 0 -closely) by a contact submanifold or an isotropic submanifold, respectively. The most general results in this direction are best phrased in M. Gromov’s language of h-principles. For a recent text on h-principles that puts particular emphasis on symplectic and contact geometry see [30]; a brief and perhaps more gentle introduction to h-principles can be found in [47]. In the present section I restrict attention to the 3-dimensional situation, where the relevant approximation theorems can be proved by elementary geometric ad hoc techniques.

346

H. Geiges

T HEOREM 2.44. Let γ : S 1 → (M, ξ ) be a knot, i.e. an embedding of S 1 , in a contact 3-manifold. Then γ can be C 0 -approximated by a Legendrian knot isotopic to γ . Alternatively, it can be C 0 -approximated by a positively as well as a negatively transverse knot. In order to prove this theorem, we first consider embeddings γ : (a, b) → (R3 , ξ ) of an open interval in R3 with its standard contact structure ξ = ker α, where α = dz + x dy. Write γ (t) = (x(t), y(t), z(t)). Then α(γ˙ ) = z˙ + x y, ˙ so the condition for a Legendrian curve reads z˙ + x y˙ ≡ 0; for a positively (respectively negatively) transverse curve, z˙ + x y˙ > 0 (respectively < 0). There are two ways to visualise such curves. The first is via its front projection γF (t) = y(t), z(t) , the second via its Lagrangian projection γL (t) = x(t), y(t) . 2.6.1. Legendrian knots If γ (t) = (x(t), y(t), z(t)) is a Legendrian curve in R3 , then y˙ = 0 implies z˙ = 0, so there the front projection has a singular point. Indeed, the curve t → (t, 0, 0) is an example of a Legendrian curve whose front projection is a single point. We call a Legendrian curve generic if y˙ = 0 only holds at isolated points (which we call cusp points), and there y¨ = 0. L EMMA 2.45. Let γ : (a, b) → (R3 , ξ ) be a Legendrian immersion. Then its front projection γF (t) = (y(t), z(t)) does not have any vertical tangencies. Away from the cusp points, γ is recovered from its front projection via x(t) = −

z˙ (t) dz =− , y(t) ˙ dy

i.e. x(t) is the negative slope of the front projection. The curve γ is embedded if and only if γF has only transverse self-intersections. By a C ∞ -small perturbation of γ we can obtain a generic Legendrian curve γ˜ isotopic to γ ; by a C 2 -small perturbation we may achieve that the front projection has only semicubical cusp singularities, i.e. around a cusp point at t = 0 the curve γ˜ looks like γ˜ (t) = t + a, λt 2 + b, −λ 2t 3 /3 + at 2 + c with λ = 0, see Figure 3. Any regular curve in the (y, z)-plane with semi-cubical cusps and no vertical tangencies can be lifted to a unique Legendrian curve in R3 .

Contact geometry

347

Fig. 3. The cusp of a front projection.

P ROOF. The Legendrian condition is z˙ + x y˙ = 0. Hence y˙ = 0 forces z˙ = 0, so γF cannot have any vertical tangencies. Away from the cusp points, the Legendrian condition tells us how to recover x as the dz negative slope of the front projection. (By continuity, the equation x = dy also makes sense at generic cusp points.) In particular, a self-intersecting front projection lifts to a non-intersecting curve if and only if the slopes at the intersection point are different, i.e. if and only if the intersection is transverse. That γ can be approximated in the C ∞ -topology by a generic immersion γ˜ follows from the usual transversality theorem (in its most simple form, viz., applied to the function y(t); the function x(t) may be left unchanged, and the new z(t) is then found by integrating the new −x y). ˙ At a cusp point of γ˜ we have y˙ = z˙ = 0. Since γ˜ is an immersion, this forces x˙ = 0, so γ˜ can be parametrised around a cusp point by the x-coordinate, i.e. we may choose the curve parameter t such that the cusp lies at t = 0 and x(t) = t + a. Since y(0) ¨ = 0 by the genericity condition, we can write y(t) = t 2 g(t) + y(0) with a smooth function g(t) satisfying g(0) = 0 (this is proved like the ‘Morse lemma’ in Appendix 2 of [77]). A C 0 approximation of g(t) by a function h(t) with h(t) ≡ g(0) for t near zero and h(t) ≡ g(t) for |t| greater than some small ε > 0 yields a C 2 -approximation of y(t) with the desired form around the cusp point. E XAMPLE 2.46. Figure 4 shows the front projection of a Legendrian trefoil knot. P ROOF OF T HEOREM 2.44 (Legendrian case). First of all, we consider a curve γ in standard R3 . In order to find a C 0 -close approximation of γ , we simply need to choose a C 0 -close approximation of its front projection γF by a regular curve without vertical tangencies and with isolated cusps (we call such a curve a front) in such a way, that the slope of the front at time t is close to −x(t) (see Figure 5). Then the Legendrian lift of this front is the desired C 0 -approximation of γ . If γ is defined on an interval (a, b) and is already Legendrian near its endpoints, then the approximation of γF may be assumed to coincide with γF near the endpoints, so that the Legendrian lift coincides with γ near the endpoints. Hence, given a knot in an arbitrary contact 3-manifold, we can cut it (by the Lebesgue lemma) into little pieces that lie in Darboux charts. There we can use the preceding recipe to find a Legendrian approximation. Since, as just observed, one can find such approximations on intervals with given boundary condition, this procedure yields a Legendrian approximation of the full knot.

348

H. Geiges

Fig. 4. Front projection of a Legendrian trefoil knot.

Fig. 5. Legendrian C 0 -approximation via front projection.

Locally (i.e. in R3 ) the described procedure does not introduce any self-intersections in the approximating curve, provided we approximate γF by a front with only transverse self-intersections. Since the original knot was embedded, the same will then be true for its Legendrian C 0 -approximation. The same result may be derived using the Lagrangian projection: L EMMA 2.47. Let γ : (a, b) → (R3 , ξ ) be a Legendrian immersion. Then its Lagrangian projection γL (t) = (x(t), y(t)) is also an immersed curve. The curve γ is recovered from γL via z(t1 ) = z(t0 ) −

t1

x dy. t0

Contact geometry

349

Fig. 6. Lagrangian projection of a Legendrian unknot.

A Legendrian immersion γ : S 1 → (R3 , ξ ) has a Lagrangian projection that encloses zero area. Moreover, γ is embedded if and only if every loop in γL (except, in the closed case, the full loop γL ) encloses a non-zero oriented area. Any immersed curve in the (x, y)-plane is the Lagrangian projection of a Legendrian curve in R3 , unique up to translation in the z-direction. P ROOF. The Legendrian condition z˙ + x y˙ implies that if y˙ = 0 then z˙ = 0, and hence, since γ is an immersion, x˙ = 0. So γL is an immersion. The formula for z follows by integrating the Legendrian condition. For a closed curve 5 γL in the (x, y)-plane, the integral γL x dy computes the oriented area enclosed by γL . From that all the other statements follow. E XAMPLE 2.48. Figure 6 shows the Lagrangian projection of a Legendrian unknot. A LTERNATIVE PROOF OF T HEOREM 2.44 (Legendrian case). Again we consider a curve γ in standard R3 defined on an interval. The generalisation to arbitrary contact manifolds and closed curves is achieved as in the proof using front projections. In order to find a C 0 -approximation of γ by a Legendrian curve, one only has to approximate its Lagrangian projection γL by an immersed curve whose ‘area integral’ z(t0 ) −

t

x dy t0

lies as close to the original z(t) as one wishes. This can be achieved by using small loops oriented positively or negatively (see Figure 7). If γL has self-intersections, this approximating curve can be chosen in such a way that along loops properly contained in that curve the area integral is non-zero, so that again we do not introduce any self-intersections in the Legendrian approximation of γ . 2.6.2. Transverse knots The quickest proof of the transverse case of Theorem 2.44 is via the Legendrian case. However, it is perfectly feasible to give a direct proof along the lines of the preceding discussion, i.e. using the front or the Lagrangian projection. Since this picture is useful elsewhere, I include a brief discussion, restricting attention to the front projection. Thus, let γ (t) = (x(t), y(t), z(t)) be a curve in R3 . The condition for γ to be positively transverse to the standard contact structure ξ = ker(dz + x dy) is that z˙ + x y˙ > 0. Hence, if y˙ = 0, if y˙ > 0, if y˙ < 0,

then z˙ > 0, then x > −˙z/y, ˙ then x < −˙z/y. ˙

350

H. Geiges

Fig. 7. Legendrian C 0 -approximation via Lagrangian projection.

Fig. 8. Impossible front projections of positively transverse curve.

The first statement says that there are no vertical tangencies oriented downwards in the front projection. The second statement says in particular that for y˙ > 0 and z˙ < 0 we have x > 0; the third, that for y˙ < 0 and z˙ < 0 we have x < 0. This implies that the situations shown in Figure 8 are not possible in the front projection of a positively transverse curve. I leave it to the reader to check that all other oriented crossings are possible in the front projection of a positively transverse curve, and that any curve in the (y, z)-plane without the forbidden crossing or downward vertical tangencies admits a lift to a positively transverse curve. E XAMPLE 2.49. Figure 9 shows the front projection of a positively transverse trefoil knot. P ROOF OF T HEOREM 2.44 (Transverse case). By the Legendrian case of this theorem, the given knot γ can be C 0 -approximated by a Legendrian knot γ1 . By Example 2.29, a neighbourhood of γ1 in (M, ξ ) looks like a solid torus S 1 × D 2 with contact structure cos θ dx − sin θ dy = 0, where γ1 ≡ S 1 × {0}. Then the curve γ2 (t) = (θ = t, x = δ sin t, y = δ cos t),

t ∈ [0, 2π],

Contact geometry

351

Fig. 9. Front projection of a positively transverse trefoil knot.

is a positively (respectively negatively) transverse knot if δ > 0 (respectively < 0). By choosing |δ| small we obtain as good a C 0 -approximation of γ1 and hence of γ as we wish.

3. Contact structures on 3-manifolds Here is the main theorem proved in this section: T HEOREM 3.1 (Lutz–Martinet). Every closed, orientable 3-manifold admits a contact structure in each homotopy class of tangent 2-plane fields. In Section 3.2 I present what is essentially J. Martinet’s [90] proof of the existence of a contact structure on every 3-manifold. This construction is based on a surgery description of 3-manifolds due to R. Lickorish and A. Wallace. For the key step, showing how to extend over a solid torus certain contact forms defined near the boundary of that torus (which then makes it possible to perform Dehn surgeries), we use an approach due to W. Thurston and H. Winkelnkemper; this allows us to simplify Martinet’s proof slightly. In Section 3.3 we show that every orientable 3-manifold is parallelisable and then build on this to classify (co-)oriented tangent 2-plane fields up to homotopy. In Section 3.4 we study the so-called Lutz twist, a topologically trivial Dehn surgery on a contact manifold (M, ξ ) which yields a contact structure ξ on M that is not homotopic (as 2-plane field) to ξ . We then complete the proof of the main theorem stated above. These results are contained in R. Lutz’s thesis [84]. Of Lutz’s published work, [83] only deals with the 3-sphere (and is only an announcement); [85] also deals with a more restricted problem. I learned the key steps of the construction from an exposition given in V. Ginzburg’s thesis [50]. I have added proofs of many topological details that do not seem to have appeared in a readily accessible source before. In Section 3.5 I indicate two further proofs for the existence of contact structures on every 3-manifold (and provide references to others). The one by Thurston and Winkelnkemper uses a description of 3-manifolds as open books due to J. Alexander; the crucial idea in their proof is the one we also use to simplify Martinet’s argument. Indeed, my discussion of the Lutz twist in the present section owes more to the paper by Thurston– Winkelnkemper than to any other reference. The second proof, by J. Gonzalo, is based

352

H. Geiges

on a branched cover description of 3-manifolds found by H. Hilden, J. Montesinos and T. Thickstun. This branched cover description also yields a very simple geometric proof that every orientable 3-manifold is parallelisable. In Section 3.6 we discuss the fundamental dichotomy between tight and overtwisted contact structures, introduced by Eliashberg, as well as the relation of these types of contact structures with the concept of symplectic fillability. The chapter concludes in Section 3.7 with a survey of classification results for contact structures on 3-manifolds. But first we discuss, in Section 3.1, an invariant of transverse knots in R3 with its standard contact structure. This invariant will be an ingredient in the proof of the Lutz–Martinet theorem, but is also of independent interest. I do not feel embarrassed to use quite a bit of machinery from algebraic and differential topology in this chapter. However, I believe that nothing that cannot be found in such standard texts as [14,77,95] is used without proof or an explicit reference. Throughout this third section, M denotes a closed, orientable 3-manifold.

3.1. An invariant of transverse knots Although the invariant in question can be defined for transverse knots in arbitrary contact manifolds (provided the knot is homologically trivial), for the sake of clarity I restrict attention to transverse knots in R3 with its standard contact structure ξ0 = ker(dz + x dy). This will be sufficient for the proof of the Lutz–Martinet theorem. In Section 3.6 I say a few words about the general situation and related invariants for Legendrian knots. Thus, let γ be a transverse knot in (R3 , ξ0 ). Push γ a little in the direction of ∂x —notice that this is a nowhere zero vector field contained in ξ0 , and in particular transverse to γ —to obtain a knot γ . An orientation of γ induces an orientation of γ . The orientation of R3 is given by dx ∧ dy ∧ dz. D EFINITION 3.2. The self-linking number l(γ ) of the transverse knot γ is the linking number of γ and γ . Notice that this definition is independent of the choice of orientation of γ (but it changes sign if the orientation of R3 is reversed). Furthermore, in place of ∂x we could have chosen any nowhere zero vector field X in ξ0 to define l(γ ): The difference between the selflinking number defined via ∂x and that defined via X is the degree of the map γ → S 1 given by associating to a point on γ the angle between ∂x and X. This degree is computed with the induced map Z ∼ = Z. But the map γ → S 1 factors through = H1 (γ ) → H1 (S 1 ) ∼ 3 R , so the induced homomorphism on homology is the zero homomorphism. Observe that l(γ ) is an invariant under isotopies of γ within the class of transverse knots. We now want to compute l(γ ) from the front projection of γ . Recall that the writhe of an oriented knot diagram is the signed number of self-crossings of the diagram, where the sign of the crossing is given in Figure 10. L EMMA 3.3. The self-linking number l(γ ) of a transverse knot is equal to the writhe w(γ ) of its front projection.

Contact geometry

353

Fig. 10. Signs of crossings in a knot diagram.

Fig. 11. Transverse knots with self-linking number ±3.

P ROOF. Let γ be the push-off of γ as described. Observe that each crossing of the front projection of γ contributes a crossing of γ underneath γ of the corresponding sign. Since the linking number of γ and γ is equal to the signed number of times that γ crosses underneath γ (cf. [98, p. 37]), we find that this linking number is equal to the signed number of self-crossings of γ , that is, l(γ ) = w(γ ). P ROPOSITION 3.4. Every self-linking number is realised by a transverse link in standard R3 . P ROOF. Figure 11 shows front projections of positively transverse knots (cf. Section 2.6.2) with self-linking number ±3. From that the construction principle for realising any odd integer should be clear. With a two component link any even integer can be realised. R EMARK 3.5. It is no accident that I do not give an example of a transverse knot with even self-linking number. By a theorem of Eliashberg [26, Proposition 2.3.1] that relates l(γ ) to the Euler characteristic of a Seifert surface S for γ and the signed number of singular points of the characteristic foliation Sξ , the self-linking number l(γ ) of a knot can only take odd values.

354

H. Geiges

3.2. Martinet’s construction According to Lickorish [81] and Wallace [103] M can be obtained from S 3 by Dehn surgery along a link of 1-spheres. This means that we have to remove a disjoint set of embedded solid tori S 1 × D 2 from S 3 and glue back solid tori with suitable identification by a diffeomorphism along the boundaries S 1 × S 1 . The effect of such a surgery (up to diffeomorphism of the resulting manifold) is completely determined by the induced map in homology H1 S 1 × ∂D 2 → H1 S 1 × ∂D 2 , Z ⊕ Z → Z ⊕ Z, n q which is given by a unimodular matrix m p ∈ GL(2, Z). Hence, denoting coordinates in 1 1 S × S by (θ, ϕ), we may always assume the identification maps to be of the form θ nq θ → . ϕ mp ϕ The curves µ and λ on ∂(S 1 × D 2 ) given respectively by θ = 0 and ϕ = 0 are called meridian and longitude. We keep the same notation µ, λ for the homology classes these curves represent. It turns out that the diffeomorphism type of the surgered manifold is completely determined by the class pµ + qλ, which is the class of the curve that becomes homotopically trivial in the surgered manifold (cf. [98, p. 28]). In fact, the Dehn surgery is completely determined by the surgery coefficient p/q, since the diffeomorphism of ∂(S 1 × D 2 ) given by (λ, µ) → (λ, −µ) extends to a diffeomorphism of the solid torus that we glue back. Similarly, the diffeomorphism given by (λ, µ) → (λ + kµ, µ) extends. By such a change of longitude in S 1 × D 2 ⊂ M, which amounts to choosing a different trivialisation of the n q normal bundle (i.e. framing) of S 1 × {0} ⊂ M, the gluing map is changed to m−kn p−kq . By a change of longitude in the solid torus that we glue back, the gluing map is changed n+kq q to m+kp p . Thus, a Dehn surgery is a so-called handle surgery (or ‘honest surgery’ or simply ‘surgery’) if and only if the surgery coefficient is an integer, that is, q = ±1. For in 0 1 n q exactly this case we may assume m p = 1 0 , and the surgery is given by cutting out S 1 × D 2 and gluing back S 1 × D 2 with the identity map ∂ D2 × S 1 → ∂ S 1 × D2 . The theorem of Lickorish and Wallace remains true if we only allow handle surgeries. However, this assumption does not entail any great simplification of the existence proof for contact structures, so we shall work with general Dehn surgeries. Our aim in this section is to use this topological description of 3-manifolds for a proof of the following theorem, first proved by Martinet [90]. The proof presented here is in spirit the one given by Martinet, but, as indicated in the introduction to this third section, amalgamated with ideas of Thurston and Winkelnkemper [101], whose proof of the same theorem we shall discuss later.

Contact geometry

355

T HEOREM 3.6 (Martinet). Every closed, orientable 3-manifold M admits a contact structure. In view of the theorem of Lickorish and Wallace and the fact that S 3 admits a contact structure, Martinet’s theorem is a direct consequence of the following result. T HEOREM 3.7. Let ξ0 be a contact structure on a 3-manifold M0 . Let M be the manifold obtained from M0 by a Dehn surgery along a knot K. Then M admits a contact structure ξ which coincides with ξ0 outside the neighbourhood of K where we perform surgery. P ROOF. By Theorem 2.44 we may assume that K is positively transverse to ξ0 . Then, by the contact neighbourhood theorem (Example 2.33), we can find a tubular neighbourhood of K diffeomorphic to S 1 ×D 2 (δ0 ), where K is identified with S 1 ×{0} and D 2 (δ0 ) denotes ¯ a disc of radius δ0 , such that the contact structure ξ0 is given as the kernel of d θ¯ + r¯ 2 d ϕ, ¯ polar coordinates on D 2 (δ0 ). with θ¯ denoting the S 1 -coordinate and (¯r , ϕ) n q Now perform a Dehn surgery along K defined by the unimodular matrix m p . This corresponds to cutting out S 1 × D 2 (δ1 ) ⊂ S 1 × D 2 (δ0 ) for some δ1 < δ0 and gluing it back in the way described above. Write (θ ; r, ϕ) for the coordinates on the copy of S 1 × D 2 (δ1 ) that we want to glue back. Then the contact form d θ¯ + r¯ 2 d ϕ¯ given on S 1 × D 2 (δ0 ) pulls back (along S 1 × ∂D 2 (δ1 )) to d(nθ + qϕ) + r 2 d(mθ + pϕ). This form is defined on all of S 1 ×(D 2 (δ1 )−{0}), and to complete the proof it only remains to find a contact form on S 1 × D 2 (δ1 ) that coincides with this form near S 1 × ∂D 2 (δ1 ). It is at this point that we use an argument inspired by the Thurston–Winkelnkemper proof (but which goes back to Lutz). n q 1 2 L EMMA 3.8. Given a unimodular matrix m p , there is a contact form on S × D (δ) that coincides with (n + mr 2 ) dθ + (q + pr 2 ) dϕ near r = δ and with ±dθ + r 2 dϕ near r = 0. P ROOF. We make the ansatz α = h1 (r) dθ + h2 (r) dϕ with smooth functions h1 (r), h2 (r). Then dα = h1 dr ∧ dθ + h2 dr ∧ dϕ and

h α ∧ dα =

1 h1

h2

dθ ∧ dr ∧ dϕ. h2

356

H. Geiges

Fig. 12. Dehn surgery.

So to satisfy the contact condition α ∧ dα = 0, all we have to do is to find a parametrised curve r → h1 (r), h2 (r) ,

0 r δ,

in the plane satisfying the following conditions: 1. h1 (r) = ±1 and h2 (r) = r 2 near r = 0, 2. h1 (r) = n + mr 2 and h2 (r) = q + pr 2 near r = δ, 3. (h1 (r), h2 (r)) is never parallel to (h1 (r), h2 (r)). Since np −mq = ±1, the vector (m, p) is not a multiple of (n, q). Figure 12 shows possible solution curves for the two cases np − mq = ±1. This completes the proof of the lemma and hence that of Theorem 3.7. R EMARK 3.9. On S 3 we have the standard contact forms α± = x dy − y dx ± (z dt − t dz) defining opposite orientations (cf. Remark 2.2). Performing the above surgery construction either on (S 3 , ker α+ ) or on (S 3 , ker α− ) we obtain both positive and negative contact structures on any given M. The same is true for the Lutz construction that we study in the next two sections. Hence: A closed oriented 3-manifold admits both a positive and a negative contact structure in each homotopy class of tangent 2-plane fields.

3.3. 2-plane fields on 3-manifolds First we need the following well-known fact. T HEOREM 3.10. Every closed, orientable 3-manifold M is parallelisable. R EMARK . The most geometric proof of this theorem can be given based on a structure theorem of Hilden, Montesinos and Thickstun. This will be discussed in Section 3.5.2. Another proof can be found in [76]. Here we present the classical algebraic proof.

Contact geometry

357

P ROOF. The main point is to show the vanishing of the second Stiefel–Whitney class w2 (M) = w2 (T M) ∈ H 2 (M; Z2 ). Recall the following facts, which can be found in [14]; for the interpretation of Stiefel–Whitney classes as obstruction classes see also [95]. There are Wu classes vi ∈ H i (M; Z2 ) defined by

Sqi (u), [M] = vi ∪ u, [M]

for all u ∈ H 3−i (M; Z2 ), where Sq denotes the Steenrod squaring operations. Since are v0 = 1 and v1 . The Sqi (u) = 0 for i > 3 − i, the only (potentially) non-zero Wu classes Wu classes and the Stiefel–Whitney classes are related by wq = j Sqq−j (vj ). Hence v1 = Sq0 (v1 ) = w1 , which equals zero since M is orientable. We conclude w2 = 0. Let V2 (R3 ) = SO(3)/ SO(1) = SO(3) be the Stiefel manifold of oriented, orthonormal 2-frames in R3 . This is connected, so there exists a section over the 1-skeleton of M of the 2-frame bundle V2 (T M) associated with T M (with a choice of Riemannian metric on M understood4 ). The obstruction to extending this section over the 2-skeleton is equal to w2 , which vanishes as we have just seen. The obstruction to extending the section over all of M lies in H 3 (M; π2 (V2 (R3 ))), which is the zero group because of π2 (SO(3)) = 0. We conclude that T M has a trivial 2-dimensional sub-bundle ε 2 . The complementary 1-dimensional bundle λ = T M/ε 2 is orientable and hence trivial since 0 = w1 (T M) = w1 (ε 2 ) + w1 (λ) = w1 (λ). Thus T M = ε 2 ⊕ λ is a trivial bundle. Fix an arbitrary Riemannian metric on M and a trivialisation of the unit tangent bundle ST M ∼ = M × S 2 . This sets up a one-to-one correspondence between the following sets, where all maps, homotopies, etc. are understood to be smooth. • Homotopy classes of unit vector fields X on M, • Homotopy classes of (co-)oriented 2-plane distributions ξ in T M, • Homotopy classes of maps f : M → S 2 . (I use the term ‘2-plane distribution’ synonymously with ‘2-dimensional sub-bundle of the tangent bundle’.) Let ξ1 , ξ2 be two arbitrary 2-plane distributions (always understood to be cooriented). By elementary obstruction theory there is an obstruction d 2 (ξ1 , ξ2 ) ∈ H 2 M; π2 S 2 ∼ = H 2 (M; Z) for ξ1 to be homotopic to ξ2 over the 2-skeleton of M and, if d 2 (ξ1 , ξ2 ) = 0 and after homotoping ξ1 to ξ2 over the 2-skeleton, an obstruction (which will depend, in general, on that first homotopy) d 3 (ξ1 , ξ2 ) ∈ H 3 M; π3 S 2 ∼ = H 3 (M; Z) ∼ =Z for ξ1 to be homotopic to ξ2 over all of M. (The identification of H 3 (M; Z) with Z is determined by the orientation of M.) However, rather than relying on general obstruction theory, we shall interpret d 2 and d 3 geometrically, which will later allow us to give a 4 This is not necessary, of course. One may also work with arbitrary 2-frames without reference to a metric. This does not affect the homotopical data.

358

H. Geiges

geometric proof that every homotopy class of 2-plane fields ξ on M contains a contact structure. The only fact that I want to quote here is that, by the Pontrjagin–Thom construction, homotopy classes of maps f : M → S 2 are in one-to-one correspondence with framed cobordism classes of framed (and oriented) links of 1-spheres in M. The general theory can be found in [14] and [77]; a beautiful and elementary account is given in [94]. For given f , the correspondence is defined by choosing a regular value p ∈ S 2 for f and a positively oriented basis b of Tp S 2 , and associating with it the oriented framed link (f −1 (p), f ∗ b), where f ∗ b is the pull-back of b under the fibrewise bijective map Tf : T (f −1 (p))⊥ → Tp S 2 . The orientation of f −1 (p) is the one which together with the frame f ∗ b gives the orientation of M. For a given framed link L the corresponding f is defined by projecting a (trivial) disc bundle neighbourhood L × D 2 of L in M onto the fibre D 2 ∼ = S 2 − p ∗ , where 0 is identified ∗ with p and p denotes the antipode of p, and sending M − (L × D 2 ) to p ∗ . Notice that the orientations of M and the components of L determine that of the fibre D 2 , and hence determine the map f . Before proceeding to define the obstruction classes d 2 and d 3 we make a short digression and discuss some topological background material which is fairly standard but not contained in our basic textbook references [14] and [77]. 3.3.1. Hopf’s Umkehrhomomorphismus If f : M m → N n is a continuous map between smooth manifolds, one can define a homomorphism ϕ : Hn−p (N ) → Hm−p (M) on homology classes represented by submanifolds as follows. Given a homology class [L]N ∈ Hn−p (N ) represented by a codimension p submanifold L, replace f by a smooth approximation transverse to L and define ϕ([L]N ) = [f −1 (L)]M . This is essentially Hopf’s Umkehrhomomorphismus [73], except that he worked with combinatorial manifolds of equal dimension and made no assumptions on the homology class. The following theorem, which in spirit is contained in [41], shows that ϕ is independent of choices (of submanifold L representing a class and smooth transverse approximation to f ) and actually a homomorphism of intersection rings. This statement is not as well known as it should be, and I only know of a proof in the literature for the special case where L is a point [60]. In [14] this map is called transfer map (more general transfer maps are discussed in [60]), but is only defined indirectly via Poincaré duality (though implicitly the statement of the following theorem is contained in [14], see, for instance, p. 377). T HEOREM 3.11. Let f : M m → N n be a smooth map between closed, oriented manifolds and Ln−p ⊂ N n a closed, oriented submanifold of codimension p such that f is transverse to L. Write u ∈ H p (N ) for the Poincaré dual of [L]N , that is, u ∩ [N ] = [L]N . Then [f −1 (L)]M = f ∗ u ∩ [M]. In other words: If u is Poincaré dual to [L]N , then f ∗ u ∈ H p (M) is Poincaré dual to [f −1 (L)]M . P ROOF. Since f is transverse to L, the differential Tf induces a fibrewise isomorphism between the normal bundles of f −1 (L) and L, and we find (closed) tubular neighbourhoods W → L and V = f −1 (W ) → f −1 (L) (considered as disc bundles) such that f : V → W is a fibrewise isomorphism. Write [V ]0 and [W ]0 for the orientation

Contact geometry

359

classes in Hm (V , V − f −1 (L)) and Hn (W, W − L), respectively. We can identify these homology groups with Hm (V , ∂V ) and Hn (W, ∂W ), respectively. Let τW ∈ H p (W, ∂W ) and τV ∈ H p (V , ∂V ) be the Thom classes of these disc bundles defined by τW ∩ [W ]0 = [L]N , τV ∩ [V ]0 = f −1 (L) M . Notice that f ∗ τW = τV since f : W → V is fibrewise isomorphic and the Thom class of an oriented disc bundle is the unique class whose restriction to each fibre is a positive generator of H p (D p , ∂D p ). Writing i : M → (M, M − f −1 (L)) and j : N → (N, N − L) for the inclusion maps we have −1 f (L) M = τV ∩ [V ]0 = f ∗ τW ∩ [V ]0 = f ∗ τW ∩ i∗ [M], where we identify Hm (M, M − f −1 (L)) with Hm (V , V − f −1 (L)) under the excision isomorphism. Then we have further −1 f (L) M = i ∗ f ∗ τW ∩ [M] = f ∗ j ∗ τW ∩ [M]. So it remains to identify j ∗ τW as the Poincaré dual u of [L]N . Indeed, j ∗ τW ∩ [N] = τW ∩ j∗ [N] = τW ∩ [W ]0 = [L]N , where we have used the excision isomorphism between the groups Hn (W, W − L) and Hn (N, N − L). 3.3.2. Representing homology classes by submanifolds We now want to relate elements in H1 (M; Z) to cobordism classes of links in M. T HEOREM 3.12. Let M be a closed, oriented 3-manifold. Any c ∈ H1 (M; Z) is represented by an embedded, oriented link (of 1-spheres) Lc in M. Two links L0 , L1 represent the same class [L0 ] = [L1 ] if and only if they are cobordant in M, that is, there is an embedded, oriented surface S in M × [0, 1] with ∂S = L1 0 (−L0 ) ⊂ M × {1} 0 M × {0}, where 0 denotes disjoint union. P ROOF. Given c ∈ H1 (M; Z), set u = P D(c) ∈ H 2 (M; Z), where P D denotes the Poincaré duality map from homology to cohomology. There is a well-known isomorphism H 2 (M; Z) ∼ = M, K(Z, 2) = [M, CP ∞ ], where brackets denote homotopy classes of maps (cf. [14, VII.12]). So u corresponds to a homotopy class of maps [f ] : M → CP ∞ such that f ∗ u0 = u, where u0 is the positive

360

H. Geiges

generator of H 2 (CP ∞ ) (that is, the one that pulls back to the Poincaré dual of [CP k−1 ]CP k under the natural inclusion CP k ⊂ CP ∞ ). Since dim M = 3, any map f : M → CP ∞ is homotopic to a smooth map f1 : M → CP 1 . Let p be a regular value of f1 . Then P D(c) = u = f1∗ u0 = f1∗ P D[p] = P D f1−1 (p) by our discussion above, and hence c = [f1−1 (p)]. So Lc = f1−1 (p) is the desired link. It is important to note that in spite of what we have just said it is not true that [M, CP ∞ ] = [M, CP 1 ], since a map F : M × [0, 1] → CP ∞ with F (M × {0, 1}) ⊂ CP 1 is not, in general, homotopic rel(M × {0, 1}) to a map into CP 1 . However, we do have [M, CP ∞ ] = [M, CP 2 ]. If two links L0 , L1 are cobordant in M, then clearly [L0 ] = [L1 ] ∈ H1 M × [0, 1]; Z ∼ = H1 (M; Z). For the converse, suppose we are given two links L0 , L1 ⊂ M with [L0 ] = [L1 ]. Choose arbitrary framings for these links and use this, as described at the beginning of this section, to define smooth maps f0 , f1 : M → S 2 with common regular value p ∈ S 2 such that fi−1 (p) = Li , i = 0, 1. Now identify S 2 with the standardly embedded CP 1 ⊂ CP 2 . Let P ⊂ CP 2 be a second copy of CP 1 , embedded in such a way that [P ]CP 2 = [CP 1 ]CP 2 and P intersects CP 1 transversely in p only. This is possible since CP 1 ⊂ CP 2 has selfintersection one. Then the maps f0 , f1 , regarded as maps into CP 2 , are transverse to P and we have fi−1 (P ) = Li , i = 0, 1. Hence fi∗ u0 = fi∗ P D[P ]CP 2 = P D fi−1 (P ) M = P D[Li ]M is the same for i = 0 or 1, and from the identification ∼ = M, CP 2 → H 2 (M, Z), [f ] → f ∗ u0 we conclude that f0 and f1 are homotopic as maps into CP 2 . Let F : M × [0, 1] → CP 2 be a homotopy between f0 and f1 , which we may assume to be constant near 0 and 1. This F can be smoothly approximated by a map F : M ×[0, 1] → CP 2 which is transverse to P and coincides with F near M × 0 and M × 1 (since there the transversality condition was already satisfied). In particular, F is still a homotopy between f0 and f1 , and S = (F )−1 (P ) is a surface with the desired property ∂S = L1 0 (−L0 ). Notice that in the course of this proof we have observed that cobordism classes of links in M (equivalently, classes in H1 (M; Z)) correspond to homotopy classes of maps M → CP 2 , whereas framed cobordism classes of framed links correspond to homotopy classes of maps M → CP 1 . By forming the connected sum of the components of a link representing a certain class in H1 (M; Z), one may actually always represent such a class by a link with only one component, that is, a knot.

Contact geometry

361

3.3.3. Framed cobordisms We have seen that if L1 , L2 ⊂ M are links with [L1 ] = [L2 ] ∈ H1 (M; Z), then L1 and L2 are cobordant in M. In general, however, a given framing on L1 and L2 does not extend over the cobordism. The following observation will be useful later on. Write (S 1 , n) for a contractible loop in M with framing n ∈ Z (by which we mean that S 1 and a second copy of S 1 obtained by pushing it away in the direction of one of the vectors in the frame have linking number n). When writing L = L 0 (S 1 , n) it is understood that (S 1 , n) is not linked with any component of L . Suppose we have two framed links L0 , L1 ⊂ M with [L0 ] = [L1 ]. Let S ⊂ M × [0, 1] be an embedded surface with ∂S = L1 0 (−L0 ) ⊂ M × {1} 0 M × {0}. With D 2 a small disc embedded in S, the framing of L1 and L2 in M extends to a framing of S − D 2 in M × [0, 1] (since S − D 2 deformation retracts to a 1-dimensional complex containing L0 and L1 , and over such a complex an orientable 2-plane bundle is trivial). Now we embed a cylinder S 1 × [0, 1] in M × [0, 1] such that S 1 × [0, 1] ∩ M × {0} = ∅, S 1 × [0, 1] ∩ M × {1} = S 1 × {1}, and S 1 × [0, 1] ∩ S − D 2 = S 1 × {0} = ∂D 2 , see Figure 13. This shows that L0 is framed cobordant in M to L1 0 (S 1 , n) for suitable n ∈ Z.

Fig. 13. The framed cobordism between L0 and L1 0 (S 1 , n).

362

H. Geiges

3.3.4. Definition of the obstruction classes We are now in a position to define the obstruction classes d 2 and d 3 . With a choice of Riemannian metric on M and a trivialisation of ST M understood, a 2-plane distribution ξ on M defines a map fξ : M → S 2 and hence an oriented framed link Lξ as described above. Let [Lξ ] ∈ H1 (M; Z) be the homology class represented by Lξ . This only depends on the homotopy class of ξ , since under homotopies of ξ or choice of different regular values of fξ the cobordism class of Lξ remains invariant. We define d 2 (ξ1 , ξ2 ) = P D[Lξ1 ] − P D[Lξ2 ]. With this definition d 2 is clearly additive, that is, d 2 (ξ1 , ξ2 ) + d 2 (ξ2 , ξ3 ) = d 2 (ξ1 , ξ3 ). The following lemma shows that d 2 is indeed the desired obstruction class. L EMMA 3.13. The 2-plane distributions ξ1 and ξ2 are homotopic over the 2-skeleton M (2) of M if and only if d 2 (ξ1 , ξ2 ) = 0. P ROOF. Suppose d 2 (ξ1 , ξ2 ) = 0, that is, [Lξ1 ] = [Lξ2 ]. By Theorem 3.12 we find a surface S in M × [0, 1] with ∂S = Lξ2 0 (−Lξ1 ) ⊂ M × {1} 0 M × {0}. From the discussion on framed cobordism above we know that for suitable n ∈ Z we find a framed surface S in M × [0, 1] such that ∂S = Lξ2 0 S 1 , n 0 (−Lξ1 ) ⊂ M × {1} 0 M × {0} as framed manifolds. Hence ξ1 is homotopic to a 2-plane distribution ξ1 such that Lξ1 and Lξ2 differ only by one contractible framed loop (not linked with any other component). Then the corresponding maps f1 , f2 differ only in a neighbourhood of this loop, which is contained in a 3-ball, so f1 and f2 (and hence ξ1 and ξ2 ) agree over the 2-skeleton. Conversely, if ξ1 and ξ2 are homotopic over M (2) , we may assume ξ1 = ξ2 on M − D 3 for some embedded 3-disc D 3 ⊂ M without changing [Lξ1 ] and [Lξ2 ]. Now [Lξ1 ] = [Lξ2 ] follows from H1 (D 3 , S 2 ) = 0. R EMARK 3.14. By [99, §37] the obstruction to homotopy between ξ and ξ0 (corresponding to the constant map fξ0 : M → S 2 ) over the 2-skeleton of M is given by fξ∗ u0 , where u0 is the positive generator of H 2 (S 2 ; Z). So u0 = P D[p] for any p ∈ S 2 , and taking p to

Contact geometry

363

be a regular value of fξ we have fξ∗ u0 = fξ∗ P D[p] = P D fξ−1 (p) = P D[Lξ ] = d 2 (ξ, ξ0 ). This gives an alternative way to see that our geometric definition of d 2 does indeed coincide with the class defined by classical obstruction theory. Now suppose d 2 (ξ1 , ξ2 ) = 0. We may then assume that ξ1 = ξ2 on M − int(D 3 ), and we define d 3 (ξ1 , ξ2 ) to be the Hopf invariant H (f ) of the map f : S 3 → S 2 defined as f1 ◦ π+ on the upper hemisphere and f2 ◦ π− on the lower hemisphere, where π+ , π− are the orthogonal projections of the upper respectively lower hemisphere onto the equatorial disc, which we identify with D 3 ⊂ M. Here, given an orientation of M, we orient S 3 in such a way that π+ is orientation-preserving and π− orientation-reversing; the orientation of S 2 is inessential for the computation of H (f ). Recall that H (f ) is defined as the linking number of the preimages of two distinct regular values of a smooth map homotopic to f . Since the Hopf invariant classifies homotopy classes of maps S 3 → S 2 (it is in fact an isomorphism π3 (S 2 ) → Z), this is a suitable definition for the obstruction class d 3 . Moreover, the homomorphism property of H (f ) and the way addition in π3 (S 2 ) is defined entail the additivity of d 3 analogous to that of d 2 . For M = S 3 there is another way to interpret d 3 . Oriented 2-plane distributions on M correspond to sections of the bundle associated to T M with fibre SO(3)/U(1), hence to maps M → SO(3)/U(1) ∼ = S 2 since T M is trivial. Similarly, almost complex structures on 4 D correspond to maps D 4 → SO(4)/U(2) ∼ = SO(3)/U(1) (cf. [61] for this isomorphism). A cooriented 2-plane distribution on M can be interpreted as a triple (X, ξ, J ), where X is a vector field transverse to ξ defining the coorientation, and J a complex structure on ξ defining the orientation. Such a triple is called an almost contact structure. (This notion generalises to higher (odd) dimensions, and by Remark 2.3 every cooriented contact structure induces an almost contact structure, and in fact a unique one up to homotopy as follows from the result cited in that remark.) Given an almost contact structure (X, ξ, J ) on S 3 , one defines an almost complex structure J˜ on T D 4 |S 3 by J˜|ξ = J and J˜N = X, where N denotes the outward normal vector field. So there is a canonical way to identify homotopy classes of almost contact structures on S 3 with elements of π3 (SO(3)/U(1)) ∼ = Z such that the value zero corresponds to the almost contact structure that extends as almost complex structure over D 4 .

3.4. Let’s twist again Consider a 3-manifold M with cooriented contact structure ξ and an oriented 1-sphere K ⊂ M embedded transversely to ξ such that the positive orientation of K coincides with the positive coorientation of ξ . Then in suitable local coordinates we can identify K with S 1 × {0} ⊂ S 1 × D 2 such that ξ = ker(dθ + r 2 dϕ) and ∂θ corresponds to the positive orientation of K (see Example 2.33). Strictly speaking, if, as we shall always assume, S 1 is parametrised by 0 θ 2π , then this formula for ξ holds on S 1 × D 2 (δ) for some,

364

H. Geiges

Fig. 14. Lutz twist.

possibly small, δ > 0. However, to simplify notation we usually work with S 1 × D 2 as local model. We say that ξ is obtained from ξ by a Lutz twist along K and write ξ = ξ K if on 1 S × D 2 the new contact structure ξ is defined by ξ = ker h1 (r) dθ + h2 (r) dϕ with (h1 (r), h2 (r)) as in Figure 14, and ξ coincides with ξ outside S 1 × D 2 . More precisely, (h1 (r), h2 (r)) is required to satisfy the conditions 1. h1 (r) = −1 and h2 (r) = −r 2 near r = 0, 2. h1 (r) = 1 and h2 (r) = r 2 near r = 1, 3. (h1 (r), h2 (r)) is never parallel to (h1 (r), h2 (r)). This is the same as applying the construction of Section 3.2 to the topologically trivial 0 Dehn surgery given by the matrix −1 0 −1 . It will be useful later on to understand more precisely the behaviour of the map fξ : M → S 2 . For the definition of this map we assume—this assumption will be justified below—that on S 1 × D 2 the map fξ was defined in terms of the standard metric dθ 2 + du2 + dv 2 (with u, v Cartesian coordinates on D 2 corresponding to the polar coordinates r, ϕ) and the trivialisation ∂θ , ∂u , ∂v of T (S 1 × D 2 ). Since ξ is spanned by ∂r and h2 (r) ∂θ − h1 (r) ∂ϕ (respectively ∂u , ∂v for r = 0), a vector positively orthogonal to ξ is given by h1 (r) ∂θ + h2 (r) ∂ϕ ,

Contact geometry

365

which makes sense even for r = 0. Observe that the ratio h1 (r)/ h2 (r) (for h2 (r) = 0) is a strictly monotone decreasing function since by the third condition above we have (h1 / h2 ) = (h1 h2 − h1 h2 )/ h22 < 0. This implies that any value on S 2 other than (1, 0, 0) (corresponding to ∂θ ) is regular for the map fξ ; the value (1, 0, 0) is attained along the torus {r = r0 }, with r0 > 0 determined by h2 (r0 ) = 0, and hence not regular. If S 1 × D 2 is endowed with the orientation defined by the volume form dθ ∧ r dr ∧ dϕ = dθ ∧ du ∧ dv (so that ξ and ξ are positive contact structures) and S 2 ⊂ R3 is given its ‘usual’ orientation defined by the volume form x dy ∧ dz + y dz ∧ dx + z dx ∧ dy, then 1 fξ−1 (−1, 0, 0) = S × {0}

with orientation given by −∂θ , since fξ maps the slices {θ } × D 2 (r0 ) orientationreversingly onto S 2 . More generally, for any p ∈ S 2 − {(1, 0, 0)} the preimage fξ−1 (p) (inside the domain 1 2 {(θ, r, ϕ): h2 (r) < 0} = {r < r0 }) is a circle S × {u}, u ∈ D , with orientation given by −∂θ . We are now ready to show how to construct a contact structure on M in any given homotopy class of 2-plane distributions by starting with an arbitrary contact structure and performing suitable Lutz twists. First we deal with homotopy over the 2-skeleton. One way to proceed would be to prove directly, with notation as above, that d 2 (ξ K , ξ ) = −P D[K]. However, it is somewhat easier to compute d 2 (ξ K , ξ ) in the case where ξ is a trivial 2-plane bundle and the trivialisation of ST M is adapted to ξ . Since I would anyway like to present an alternative argument for computing the effect of a Lutz twist on the Euler class of the contact structure, and thus relate d 2 (ξ1 , ξ2 ) with the Euler classes of ξ1 and ξ2 , it seems opportune to do this first and use it to show the existence of a contact structure with Euler class zero. In the next section we shall actually discuss a direct geometric proof, due to Gonzalo, of the existence of a contact structure with Euler class zero. Recall that the Euler class e(ξ ) ∈ H 2 (B; Z) of a 2-plane bundle over a complex B (of arbitrary dimension) is the obstruction to finding a nowhere zero section of ξ over the 2skeleton of B. Since πi (S 1 ) = 0 for i 2, all higher obstruction groups H i+1 (B; πi (S 1 )) are trivial, so a 2-dimensional orientable bundle ξ is trivial if and only if e(ξ ) = 0, no matter what the dimension of B. Now let ξ be an arbitrary cooriented 2-plane distribution on an oriented 3-manifold M. Then T M ∼ = ξ ⊕ ε 1 , where ε 1 denotes a trivial line bundle. Hence w2 (ξ ) = w2 (ξ ⊕ ε 1 ) = w2 (T M) = 0, and since w2 (ξ ) is the mod 2 reduction of e(ξ ) we infer that e(ξ ) has to be even. P ROPOSITION 3.15. For any even element e ∈ H 2 (M; Z) there is a contact structure ξ on M with e(ξ ) = e. P ROOF. Start with an arbitrary contact structure ξ0 on M with e(ξ0 ) = e0 (which we know to be even). Given any even e1 ∈ H 2 (M; Z), represent the Poincaré dual of (e0 − e1 )/2 by

366

H. Geiges

Fig. 15. Effect of Lutz twist on Euler class.

a collection of embedded oriented circles positively transverse to ξ0 . (Here by (e0 − e1 )/2 I mean some class whose double equals e0 − e1 ; in the presence of 2-torsion there is of course a choice of such classes.) Choose a section of ξ0 transverse to the zero section of ξ0 , that is, a vector field in ξ0 with generic zeros. We may assume that there are no zeros on the curves representing P D −1 (e0 − e1 )/2. Now perform a Lutz twist as described above along these curves and call ξ1 the resulting contact structure. It is easy to see that in the local model for the Lutz twist a constant vector field tangent to ξ0 along ∂(S 1 × D 2 (r0 )) extends to a vector field tangent to ξ1 over S 1 × D 2 (r0 ) with zeros of index +2 along S 1 × {0} (Figure 15). So the vector field in ξ0 extends to a vector field in ξ1 with new zeros of index +2 along the curves representing P D −1 (e1 − e0 )/2 (notice that a Lutz twist along a positively transverse knot K turns K into a negatively transverse knot). Since the self-intersection class of M in the total space of a vector bundle is Poincaré dual to the Euler class of that bundle, this proves e(ξ1 ) = e(ξ0 ) + e1 − e0 = e1 . We now fix a contact structure ξ0 on M with e(ξ0 ) = 0 and give M the orientation induced by ξ0 (i.e. the one for which ξ0 is a positive contact structure). Moreover, we fix a Riemannian metric on M and define X0 as the vector field positively orthonormal to ξ0 . Since ξ0 is a trivial plane bundle, X0 extends to an orthonormal frame X0 , X1 , X2 , hence a trivialisation of ST M, with X1 , X2 tangent to ξ0 and defining the orientation of ξ0 . With these choices, ξ0 corresponds to the constant map fξ0 : M → (1, 0, 0) ∈ S 2 . P ROPOSITION 3.16. Let K ⊂ M be an embedded, oriented circle positively transverse to ξ0 . Then d 2 (ξ0K , ξ0 ) = −P D[K]. P ROOF. Identify a tubular neighbourhood of K ⊂ M with S 1 × D 2 with framing defined by X1 , and ξ0 given in this neighbourhood as the kernel of dθ + r 2 dϕ = dθ + u dv − v du. We may then change the trivialisation X0 , X1 , X2 by a homotopy, fixed outside S 1 × D 2 , such that X0 = ∂θ , X1 = ∂u and X2 = ∂v near K; this does not change the homotopical data of 2-plane distributions computed via the Pontrjagin–Thom construction. Then fξ0 is no longer constant, but its image still does not contain the point (−1, 0, 0).

Contact geometry

367

Now perform a Lutz twist along K × {0}. Our discussion at the beginning of this section shows that (−1, 0, 0) is a regular value of the map fξ : M → S 2 associated with ξ = ξ0K and fξ−1 (−1, 0, 0) = −K. Hence, by definition of the obstruction class d 2 we have d 2 (ξ0K , ξ0 ) = −P D[K]. P ROOF OF T HEOREM 3.1. Let η be a 2-plane distribution on M and ξ0 the contact structure on M with e(ξ0 ) = 0 that we fixed earlier on. According to our discussion in Section 3.3.2 and Theorem 2.44, we can find an oriented knot K positively transverse to ξ0 with −P D[K] = d 2 (η, ξ0 ). Then d 2 (η, ξ0 ) = d 2 (ξ0K , ξ0 ) by the preceding proposition, and therefore d 2 (ξ0K , η) = 0. We may then assume that η = ξ0K on M − D 3 , where we choose D 3 so small that ξ0K is in Darboux normal form there (and identical with ξ0 ). By Proposition 3.4 we can find a link K in D 3 transverse to ξ0K with self-linking number l(K ) equal to d 3 (η, ξ0K ). Now perform a Lutz twist of ξ0K along each component of K and let ξ be the resulting contact structure. Since this does not change ξ0K over the 2-skeleton of M, we still have d 2 (ξ, η) = 0. Observe that fξ K |D 3 does not contain the point (−1, 0, 0) ∈ S 2 , and—since fξ K (D 3 ) is 0

0

compact—there is a whole neighbourhood U ⊂ S 2 of (−1, 0, 0) not contained in fξ K (D 3 ). 0

Let f : S 3 → S 2 be the map used to compute d 3 (ξ, ξ0K ), that is, f coincides on the upper hemisphere with fξ |D 3 and on the lower hemisphere with fξ K |D 3 . By the discussion in 0

Section 3.3, the preimage f −1 (u) of any u ∈ U − {(−1, 0, 0)} will be a push-off of −K determined by the trivialisation of ξ0K |D 3 = ξ0 |D 3 . So the linking number of f −1 (u) with f −1 (−1, 0, 0), which is by definition the Hopf invariant H (f ) = d 3 (ξ, ξ0K ), will be equal to l(K ). By our choice of K and the additivity of d 3 this implies d 3 (ξ, η) = 0. So ξ is a contact structure that is homotopic to η as a 2-plane distribution.

3.5. Other existence proofs Here I briefly summarise the other known existence proofs for contact structures on 3-manifolds, mostly by pointing to the relevant literature. In spirit, most of these proofs are similar to the one by Lutz–Martinet: start with a structure theorem for 3-manifolds and show that the topological construction can be performed compatibly with a contact structure. 3.5.1. Open books According to a theorem of Alexander [4], cf. [97], every closed, orientable 3-manifold M admits an open book decomposition. This means that there is a link L ⊂ M, called the binding, and a fibration f : M − L → S 1 , whose fibres are called the pages, see Figure 16. It may be assumed that L has a tubular neighbourhood L × D 2 such that f restricted to L × (D 2 − {0}) is given by f (θ, r, ϕ) = ϕ, where θ is the coordinate along L and (r, ϕ) are polar coordinates on D 2 . At the cost of raising the genus of the pages, one may decrease the number of components of L, and in particular one may always assume L to be a knot. Another way

368

H. Geiges

Fig. 16. An open book near the binding.

to think of such an open book is as follows. Start with a surface Σ with boundary ∂Σ = K ∼ = S 1 and a self-diffeomorphism h of Σ with h = id near K. Form the mapping torus Th = Σ × [0, 2π]/∼, where ‘∼’ denotes the identification (p, 2π) ∼ (h(p), 0). Define a 3-manifold M by M = Th ∪K×S 1 K × D 2 . This M carries by construction the structure of an open book with binding K and pages diffeomorphic to Σ . Here is a slight variation on a simple argument of Thurston and Winkelnkemper [101] for producing a contact structure on such an open book (and hence on any closed, orientable 3-manifold): Start with a 1-form β0 on Σ with β0 = et dθ near ∂Σ = K, where θ denotes the coordinate along K and t is a collar parameter with K = {t = 0} and t < 0 in the interior of Σ. Then dβ0 integrates to 2π over Σ by Stokes’s theorem. Given any area form ω on Σ (with total area equal to 2π ) satisfying ω = et dt ∧ dθ near K, the 2-form ω − dβ0 is, by de Rham’s theorem, an exact 1-form, say dβ1 , where we may assume β1 ≡ 0 near K. Set β = β0 + β1 . Then dβ = ω is an area form (of total area 2π ) on Σ and β = et dθ near K. The set of 1-forms satisfying these two properties is a convex set, so we can find a 1-form (still denoted β) on Th which has these properties when restricted to the fibre over any ϕ ∈ S 1 = [0, 2π]/0∼2π . We may (and shall) require that β = et dθ near ∂Th . Now a contact form α on Th is found by setting α = β + C dϕ for a sufficiently large constant C ∈ R+ , so that in α ∧ dα = (β + C dϕ) ∧ dβ

Contact geometry

369

the non-zero term dϕ ∧ dβ = dϕ ∧ ω dominates. This contact form can be extended to all of M by making the ansatz α = h1 (r) dθ + h2 (r) dϕ on K × D 2 , as described in our discussion of the Lutz twist. The boundary conditions in the present situation are, say, 1. h1 (r) = 2 and h2 (r) = r 2 near r = 0, 2. h1 (r) = e1−r and h2 (r) = C near r = 1. Observe that subject to these boundary conditions a curve (h1 (r), h2 (r)) can be found that does not pass the h2 -axis (i.e. with h1 (r) never being equal to zero). In the 3-dimensional setting this is not essential (and the Thurston–Winkelnkemper ansatz lacked that feature), but it is crucial when one tries to generalise this construction to higher dimensions. This has recently been worked out by Giroux and J.-P. Mohsen [57]. This, however, is only the easy part of their work. Rather strikingly, they have shown that a converse of this result holds: Given a compact contact manifold of arbitrary dimension, it admits an open book decomposition that is adapted to the contact structure in the way described above. Full details have not been published at the time of writing, but see Giroux’s talk [56] at the ICM 2002. 3.5.2. Branched covers A theorem of Hilden, Montesinos and Thickstun [63] states that every closed, orientable 3-manifold M admits a branched covering π : M → S 3 such that the upstairs branch set is a simple closed curve that bounds an embedded disc. (Moreover, the cover can be chosen 3-fold and simple, i.e. the monodromy representation of π1 (S 3 − K), where K is the branching set downstairs (a knot in S 3 ), represents the meridian of K by a transposition in the symmetric group S3 . This, however, is not relevant for our discussion.) It follows immediately, as announced in Section 3.3, that every closed, orientable 3-manifold is parallelisable: First of all, S 3 is parallelisable since it carries a Lie group structure (as the unit quaternions, for instance). Given M and a branched covering π : M → S 3 as above, there is a 3-ball D 3 ⊂ M containing the upstairs branch set. Outside of D 3 , the covering π is unbranched, so the 3-frame on S 3 can be lifted to a frame on M − D 3 . The bundle T M|D 3 is trivial, so the frame defined along ∂D 3 defines an element of SO(3) (cf. the footnote in the proof of Theorem 3.10). Since π2 (SO(3)) = 0, this frame extends over D 3 . In [59], Gonzalo uses this theorem to construct a contact structure on every closed, orientable 3-manifold M, in fact one with zero Euler class: Away from the branching set one can lift the standard contact structure from S 3 (which has Euler class zero: a trivialisation is given by two of the three (quaternionic) Hopf vector fields). A careful analysis of the branched covering map near the branching set then shows how to extend this contact structure over M (while keeping it trivial as 2-plane bundle). A branched covering construction for higher-dimensional contact manifolds is discussed in [43]. 3.5.3. . . . and more The work of Giroux [52], in which he initiated the study of convex surfaces in contact 3-manifolds, also contains a proof of Martinet’s theorem. An entirely different proof, due to S. Altschuler [5], finds contact structures from solutions to a certain parabolic differential equation for 1-forms on 3-manifolds. Some of

370

H. Geiges

these ideas have entered into the more far-reaching work of Eliashberg and Thurston on so-called ‘confoliations’ [32], that is, 1-forms satisfying α ∧ dα 0. 3.6. Tight and overtwisted The title of this section describes the fundamental dichotomy of contact structures in dimension 3 that has proved seminal for the development of the field. In order to motivate the notion of an overtwisted contact structure, as introduced by Eliashberg [21], we consider a ‘full’ Lutz twist as follows. Let (M, ξ ) be a contact 3-manifold and K ⊂ M a knot transverse to ξ . As before, identify K with S 1 × {0} ⊂ S 1 × D 2 ⊂ M such that ξ = ker(dθ + r 2 dϕ) on S 1 × D 2 . Now define a new contact structure ξ as in Section 3.4, with (h1 (r), h2 (r)) now as in Figure 17, that is, the boundary conditions are now h1 (r) = 1

and h2 (r) = r 2

for r ∈ [0, ε] ∪ [1 − ε, 1]

for some small ε > 0. L EMMA 3.17. A full Lutz twist does not change the homotopy class of ξ as a 2-plane field. P ROOF. Let (ht1 (r), ht2 (r)), r, t ∈ [0, 1], be a homotopy of paths such that 1. h01 ≡ 1, h02 (r) = r 2 , 2. h1i ≡ hi , i = 1, 2, 3. hti (r) = hi (r) for r ∈ [0, ε] ∪ [1 − ε, 1].

Fig. 17. A full Lutz twist.

Contact geometry

371

Let χ : [0, 1] → R be a smooth function which is identically zero near r = 0 and r = 1 and χ(r) > 0 for r ∈ [ε, 1 − ε]. Then αt = t (1 − t)χ(r) dr + ht1 (r) dθ + ht2 (r) dϕ is a homotopy from α0 = dθ + r 2 dϕ to α1 = h1 (r) dθ + h2 (r) dϕ through non-zero 1-forms. This homotopy stays fixed near r = 1, and so it defines a homotopy between ξ and ξ as 2-plane fields. Let r0 be the smaller of the two positive radii with h2 (r0 ) = 0 and consider the embedding φ : D 2 (r0 ) → S 1 × D 2 , (r, ϕ) → θ (r), r, ϕ , where θ (r) is a smooth function with θ (r0 ) = 0, θ (r) > 0 for 0 r < r0 , and θ (r) = 0 only for r = 0. We may require in addition that θ (r) = θ (0) − r 2 near r = 0. Then φ ∗ h1 (r) dθ + h2 (r) dϕ = h1 (r)θ (r) dr + h2 (r) dϕ is a differential 1-form on D 2 (r0 ) which vanishes only for r = 0, and along ∂D 2 (r0 ) the vector field ∂ϕ tangent to the boundary lies in the kernel of this 1-form, see Figure 18. In other words, the contact planes ker(h1 (r) dθ + h2 (r) dϕ) intersected with the tangent planes to the embedded disc φ(D 2 (r0 )) induce a singular 1-dimensional foliation on this disc with the boundary of this disc as closed leaf and precisely one singular point in the

Fig. 18. An overtwisted disc.

372

H. Geiges

Fig. 19. Characteristic foliation on an overtwisted disc.

interior of the disc (Figure 19; notice that the leaves of this foliation are the integral curves of the vector field h1 (r)θ (r) ∂ϕ − h2 (r) ∂r ). Such a disc is called an overtwisted disc. A contact structure ξ on a 3-manifold M is called overtwisted if (M, ξ ) contains an embedded overtwisted disc. In order to justify this terminology, observe that in the radially symmetric standard contact structure of Example 2.7, the angle by which the contact planes turn approaches π/2 asymptotically as r goes to infinity. By contrast, any contact manifold which has been constructed using at least one (simple) Lutz twist contains a similar cylindrical region where the contact planes twist by more than π in radial direction (at the smallest positive radius r0 with h2 (r0 ) = 0 the twisting angle has reached π ). We have shown the following: P ROPOSITION 3.18. Let ξ be a contact structure on M. By a full Lutz twist along any transversely embedded circle one obtains an overtwisted contact structure ξ that is homotopic to ξ as a 2-plane distribution. Together with the theorem of Lutz and Martinet we find that M contains an overtwisted contact structure in every homotopy class of 2-plane distributions. In fact, Eliashberg [21] has proved the following much stronger theorem. T HEOREM 3.19 (Eliashberg). On a closed, orientable 3-manifold there is a one-to-one correspondence between homotopy classes of overtwisted contact structures and homotopy classes of 2-plane distributions. This means that two overtwisted contact structures which are homotopic as 2-plane fields are actually homotopic as contact structures and hence isotopic by Gray’s stability theorem. Thus, it ‘only’ remains to classify contact structures that are not overtwisted. In [24] Eliashberg defined tight contact structures on a 3-manifold M as contact structures ξ for which there is no embedded disc D ⊂ M such that Dξ contains a limit cycle. So, by definition, overtwisted contact structures are not tight. In that same paper, as mentioned above in Section 2.4.5, Eliashberg goes on to show the converse with the help of the Elimination Lemma: non-overtwisted contact structures are tight.

Contact geometry

373

There are various ways to detect whether a contact structure is tight. Historically the first proof that a certain contact structure is tight is due to D. Bennequin [9, Corollary 2, p. 150]: T HEOREM 3.20 (Bennequin). The standard contact structure ξ0 on S 3 is tight. The steps of the proof are as follows: (i) First, Bennequin shows that if γ0 is a transverse knot in (S 3 , ξ0 ) with Seifert surface Σ, then the self-linking number of γ satisfies the inequality l(γ0 ) −χ(Σ). (ii) Second, he introduces an invariant for Legendrian knots; nowadays this is called the Thurston–Bennequin invariant: Let γ be a Legendrian knot in (S 3 , ξ0 ). Take a vector field X along γ transverse to ξ0 , and let γ be the push-off of γ in the direction of X. Then the Thurston–Bennequin invariant tb(γ ) is defined to be the linking number of γ and γ . (This invariant has an extension to homologically trivial Legendrian knots in arbitrary contact 3-manifolds.) (iii) By pushing γ in the direction of ±X, one obtains transverse curves γ ± (either of which is a candidate for γ in (ii)). One of these curves is positively transverse, the other negatively transverse to ξ0 . The self-linking number of γ ± is related to the Thurston– Bennequin invariant and a further invariant (the rotation number) of γ . The equation relating these three invariants implies tb(γ ) −χ(Σ), where Σ again denotes a Seifert surface for γ . In particular, a Legendrian unknot γ satisfies tb(γ ) < 0. This inequality would be violated by the vanishing cycle of an overtwisted disc (which has tb = 0), which proves that (S 3 , ξ0 ) is tight. R EMARK 3.21. (1) Eliashberg [25] generalised the Bennequin inequality l(γ0 ) −χ(Σ) for transverse knots (and the corresponding inequality for the Thurston–Bennequin invariant of Legendrian knots) to arbitrary tight contact 3-manifolds. Thus, whereas Bennequin established the tightness (without that name) of the standard contact structure on S 3 by proving the inequality that bears his name, that inequality is now seen, conversely, as a consequence of tightness. (2) In [9] Bennequin denotes the positively (respectively negatively) transverse push-off of the Legendrian knot γ by γ − (respectively γ + ). This has led to some sign errors in the literature. Notably, the ± in Proposition 2.2.1 of [25], relating the described invariants of γ and γ ± , needs to be reversed. C OROLLARY 3.22. The standard contact structure on R3 is tight. P ROOF. This is immediate from Proposition 2.13.

Here are further tests for tightness: 1. A closed contact 3-manifold (M, ξ ) is called symplectically fillable if there exists a compact symplectic manifold (W, ω) bounded by M such that

374

H. Geiges

• the restriction of ω to ξ does not vanish anywhere, • the orientation of M defined by ξ (i.e. the one for which ξ is positive) coincides with the orientation of M as boundary of the symplectic manifold (W, ω) (which is oriented by ω2 ). We then have the following result of Eliashberg [20, Theorem 3.2.1], [22] and Gromov [62, 2.4.D2 (b)], cf. [10]: T HEOREM 3.23 (Eliashberg–Gromov). A symplectically fillable contact structure is tight. E XAMPLE 3.24. The 4-ball D 4 ⊂ R4 with symplectic form ω = dx1 ∧ dy1 + dx2 ∧ dy2 is a symplectic filling of S 3 with its standard contact structure ξ0 . This gives an alternative proof of Bennequin’s theorem. ˜ ξ˜ ) → (M, ξ ) be a covering map and contactomorphism. If (M, ˜ ξ˜ ) is tight, 2. Let (M, then so is (M, ξ ), for any overtwisted disc in (M, ξ ) would lift to an overtwisted disc in ˜ ξ˜ ). (M, E XAMPLE 3.25. The contact structures ξn , n ∈ N, on the 3-torus T 3 defined by αn = cos(nθ1 ) dθ2 + sin(nθ1 ) dθ3 = 0 are tight: Lift the contact structure ξn to the universal cover R3 of T 3 ; there the contact structure is defined by the same equation αn = 0, but now θi ∈ R instead of θi ∈ R/2πZ ∼ = S 1 . Define a diffeomorphism f of R3 by f (x, y, z) = (y/n, z cos y + x sin y, z sin y − x cos y) =: (θ1 , θ2 , θ3 ). Then f ∗ αn = dz + x dy, so the lift of ξn to R3 is contactomorphic to the tight standard contact structure on R3 . Notice that it is possible for a tight contact structure to be finitely covered by an overtwisted contact structure. The first such examples were due to S. Makar-Limanov [88]. Other examples of this kind have been found by V. Colin [18] and R. Gompf [58]. 3. The following theorem of H. Hofer [65] relates the dynamics of the Reeb vector field to overtwistedness. T HEOREM 3.26 (Hofer). Let α be a contact form on a closed 3-manifold such that the contact structure ker α is overtwisted. Then the Reeb vector field of α has at least one contractible periodic orbit. E XAMPLE 3.27. The Reeb vector field Rn of the contact form αn of the preceding example is Rn = cos(nθ1 ) ∂θ2 + sin(nθ1 ) ∂θ3 .

Contact geometry

375

Thus, the orbits of Rn define constant slope foliations of the 2-tori {θ1 = const}; in particular, the periodic orbits of Rn are even homologically non-trivial. It follows, again, that the ξn are tight contact structures on T 3 . (This, admittedly, amounts to attacking starlings with rice puddings fired from catapults.5 )

3.7. Classification results In this section I summarise some of the known classification results for contact structures on 3-manifolds. By Eliashberg’s Theorem 3.19 it suffices to list the tight contact structures, up to isotopy or diffeomorphism, on a given closed 3-manifold. T HEOREM 3.28 (Eliashberg [24]). Any tight contact structure on S 3 is isotopic to the standard contact structure ξ0 . This theorem has a remarkable application in differential topology, viz., it leads to a new proof of Cerf’s theorem [16] that any diffeomorphism of S 3 extends to a diffeomorphism of the 4-ball D 4 . The idea is that the above theorem implies that any diffeomorphism of S 3 is isotopic to a contactomorphism of ξ0 . Eliashberg’s technique [22] of filling by holomorphic discs can then be used to show that such a contactomorphism extends to a diffeomorphism of D 4 . As remarked earlier (Remark 2.21), Eliashberg has also classified contact structures on R3 . Recall that homotopy classes of 2-plane distributions on S 3 are classified by π3 (S 2 ) ∼ = Z. By Theorem 3.19, each of these classes contains a unique (up to isotopy) overtwisted contact structure. When a point of S 3 is removed, each of these contact structures induces one on R3 , and Eliashberg [25] shows that they remain non-diffeomorphic there. Eliashberg shows further that, apart from this integer family of overtwisted contact structures, there is a unique tight contact structure on R3 (the standard one), and a single overtwisted one that is ‘overtwisted at infinity’ and cannot be compactified to a contact structure on S 3 . In general, the classification of contact structures up to diffeomorphism will differ from the classification up to isotopy. For instance, on the 3-torus T 3 we have the following diffeomorphism classification due to Y. Kanda [75]: T HEOREM 3.29 (Kanda). Every (positive) tight contact structure on T 3 is contactomorphic to one of the ξn , n ∈ N, described above. For n = m, the contact structures ξn and ξm are not contactomorphic. Giroux [54] had proved earlier that ξn for n 2 is not contactomorphic to ξ1 . On the other hand, all the ξn are homotopic as 2-plane fields to {dθ1 = 0}. This shows one way how Eliashberg’s classification Theorem 3.19 for overtwisted contact structures can fail for tight contact structures: • There are tight contact structures on T 3 that are homotopic as plane fields but not contactomorphic. 5 This turn of phrase originates from [93].

376

H. Geiges

P. Lisca and G. Mati´c [82] have found examples of the same kind on homology spheres by applying Seiberg–Witten theory to Stein fillings of contact manifolds, cf. also [78]. Eliashberg and L. Polterovich [31] have determined the isotopy classes of diffeomorphisms of T 3 that contain a contactomorphism of ξ1 : they correspond to exactly those elements of SL(3, Z) = π0 (Diff(T 3 )) that stabilise the subspace 0 ⊕ Z2 corresponding to the coordinates (θ2 , θ3 ). In combination with Kanda’s result, this allows one to give an isotopy classification of tight contact structures on T 3 . One particular consequence of the result of Eliashberg and Polterovich is the following: • There are tight contact structures on T 3 that are contactomorphic and homotopic as plane fields, but not isotopic (i.e. not homotopic through contact structures). Again, such examples also exist on homology spheres, as S. Akbulut and R. Matveyev [2] have shown. Another aspect of Eliashberg’s classification of overtwisted contact structures that fails to hold for tight structures is of course the existence of such a structure in every homotopy class of 2-plane fields, as is already demonstrated by the classification of contact structures on S 3 . Etnyre and K. Honda [37] have recently even found an example of a manifold—the connected some of two copies of the Poincaré sphere with opposite orientations—that does not admit any tight contact structure at all. For the classification of tight contact structures on lens spaces and T 2 -bundles over S 1 see [55,71,72]. A partial classification of tight contact structures on lens spaces had been obtained earlier in [34]. As further reading on 3-dimensional contact geometry I can recommend the lucid Bourbaki talk by Giroux [53]. This covers the ground up to Eliashberg’s classification of overtwisted contact structures and the uniqueness of the tight contact structure on S 3 .

4. A guide to the literature In this concluding section I give some recommendations for further reading, concentrating on those aspects of contact geometry that have not (or only briefly) been touched upon in earlier sections. Two general surveys that emphasise historical matters and describe the development of contact geometry from some of its earliest origins are the one by Lutz [87] and one by the present author [45]. One aspect of contact geometry that I have neglected in these notes is the Riemannian geometry of contact manifolds (leading, for instance, to Sasakian geometry). The survey by Lutz has some material on that; D. Blair [11] has recently published a monograph on the topic. There have also been various ideas for defining interesting families of contact structures. Again, the survey by Lutz has something to say on that; one such concept that has exhibited very intriguing ramifications—if this commercial break be permitted—was introduced in [48].

Contact geometry

377

4.1. Dimension 3 As mentioned earlier, Chapter 8 in [1] is in parts complementary to the present notes and has some material on surfaces in contact 3-manifolds. Other surveys and introductory texts on 3-dimensional contact geometry are the introductory lectures by Etnyre [35] and the Bourbaki talk by Giroux [52]. Good places to start further reading are two papers by Eliashberg: [24] for the classification of tight contact structures and [26] for knots in contact 3-manifolds. Concerning the latter, there is also a chapter by Etnyre [36] in a companion Handbook and an article by Etnyre and Honda [38] with an extensive introduction to that subject. The surveys [20] and [27] by Eliashberg are more general in scope, but also contain material about contact 3-manifolds. 3-dimensional contact topology has now become a fairly wide arena; apart from the work of Eliashberg, Giroux, Etnyre–Honda and others described earlier, I should also mention the results of Colin, who has, for instance, shown that surgery of index one (in particular: taking the connected sum) on a tight contact 3-manifold leads again to a tight contact structure [17]. Finally, Etnyre and Ng [40] have compiled a useful list of problems in 3-dimensional contact topology.

4.2. Higher dimensions The article [46] by the present author contains a survey of what was known at the time of writing about the existence of contact structures on higher-dimensional manifolds. One of the most important techniques for constructing contact manifolds in higher dimensions is the so-called contact surgery along isotropic spheres developed by Eliashberg [23] and Weinstein [105]. The latter is a very readable paper. For a recent application of this technique see [49]. Other constructions of contact manifolds (branched covers, gluing along codimension 2 contact submanifolds) are described in my paper [43]. Odd-dimensional tori are of course amongst the manifolds with the simplest global description, but they do not easily lend themselves to the construction of contact structures. In [86] Lutz found a contact structure on T 5 ; since then it has been one of the prize questions in contact geometry to find a contact structure on higher-dimensional tori. That prize, as it were, recently went to F. Bourgeois [13], who showed that indeed all odddimensional tori do admit a contact structure. His construction uses the result of Giroux and Mohsen [56,57] about open book decompositions adapted to contact structures in conjunction with the original proof of Lutz. With the help of the branched cover theorem described in [43] one can conclude further that every manifold of the form M × Σ with M a contact manifold and Σ a surface of genus at least 1 admits a contact structure. Concerning the classification of contact structures in higher dimensions, the first steps have been taken by Eliashberg [28] with the development of contact homology, which has been taken further in [29]. This has been used by Ustilovsky [102] to show that on S 4n+1 there exist infinitely many non-isomorphic contact structures that are homotopically equivalent (in the sense that they induce the same almost contact structure, i.e. reduction of

378

H. Geiges

the structure group of T S 4n+1 to 1 × U(2n)). Earlier results in this direction can be found in [44] in the context of various applications of contact surgery. 4.3. Symplectic fillings A survey on the various types of symplectic fillings of contact manifolds is given by Etnyre [33], cf. also the survey by Bennequin [10]. Etnyre and Honda [39] have recently shown that certain Seifert fibred 3-manifolds M admit tight contact structures ξ that are not symplectically semi-fillable, i.e. there is no symplectic filling W of (M, ξ ) even if W is allowed to have other contact boundary components. That paper also contains a good update on the general question of symplectic fillability. A related question is whether symplectic manifolds can have disconnected boundary of contact type (this corresponds to a stronger notion of symplectic filling defined via a Liouville vector field transverse to the boundary and pointing outwards). For (boundary) dimension 3 this is discussed by McDuff [91]; higher-dimensional symplectic manifolds with disconnected boundary of contact type have been constructed in [42]. Note added in proof: Eliashberg (Geom. Topol. 8 (2004), 277–293) has shown recently that every contact 3-manifold has a concave filling. This implies, in particular, that semifillable contact manifolds are always fillable. 4.4. Dynamics of the Reeb vector field In a seminal paper, Hofer [65] applied the method of pseudo-holomorphic curves, which had been introduced to symplectic geometry by Gromov [62], to solve (for large classes of contact 3-manifolds) the so-called Weinstein conjecture [104] concerning the existence of periodic orbits of the Reeb vector field of a given contact form. (In fact, one of the remarkable aspects of Hofer’s work is that in many instances it shows the existence of a periodic orbit of the Reeb vector field of any contact form defining a given contact structure.) A Bourbaki talk on the state of the art around the time when Weinstein formulated the conjecture that bears his name was given by Desolneux-Moulis [19]; another Bourbaki talk by Laudenbach describes Hofer’s contribution to the problem. The textbook by Hofer and Zehnder [70] addresses these issues, although its main emphasis, as is clear from the title, lies more in the direction of symplectic geometry and Hamiltonian dynamics. Two surveys by Hofer [66,67], and one by Hofer and Kriener [68], are more directly concerned with contact geometry. Of the three, [66] may be the best place to start, since it derives from a course of five lectures. In collaboration with Wysocki and Zehnder, Hofer has expanded his initial ideas into a far-reaching project on the characterisation of contact manifolds via the dynamics of the Reeb vector field, see, e.g., [69]. References [1] B. Aebischer, M. Borer, M. Kälin, Ch. Leuenberger and H.M. Reimann, Symplectic Geometry, Progr. Math., vol. 124, Birkhäuser, Basel (1994). [2] S. Akbulut and R. Matveyev, A note on contact structures, Pacific J. Math. 182 (1998), 201–204.

Contact geometry

379

[3] D.N. Akhiezer, Homogeneous complex manifolds, Several Complex Variables IV, S.G. Gindikin and G.M. Khenkin, eds, Encyclopaedia Math. Sci., vol. 10, Springer, Berlin (1990), 195–244. [4] J.W. Alexander, A lemma on systems of knotted curves, Proc. Nat. Acad. Sci. U.S.A. 9 (1923), 93–95. [5] S. Altschuler, A geometric heat flow for one-forms on three-dimensional manifolds, Illinois J. Math. 39 (1995), 98–118. [6] V.I. Arnold, Characteristic class entering in quantization conditions, Funct. Anal. Appl. 1 (1967), 1–13. [7] V.I. Arnold, Mathematical Methods of Classical Mechanics, Grad. Texts in Math., vol. 60, Springer, Berlin (1978). [8] A. Banyaga, The Structure of Classical Diffeomorphism Groups, Math. Appl., vol. 400, Kluwer, Dordrecht (1997). [9] D. Bennequin, Entrelacements et équations de Pfaff, IIIe Rencontre de Géométrie du Schnepfenried, vol. 1, Astérisque 107–108 (1983), 87–161. [10] D. Bennequin, Topologie symplectique, convexité holomorphe et structures de contact (d’après Y. Eliashberg, D. McDuff et al.), Séminaire Bourbaki, vol. 1989/1990, Astérisque 189–190 (1990), 285–323. [11] D.E. Blair, Riemannian Geometry of Contact and Symplectic Manifolds, Progr. Math., vol. 203, Birkhäuser, Basel (2002). [12] W.M. Boothby, Transitivity of the automorphisms of certain geometric structures, Trans. Amer. Math. Soc. 137 (1969), 93–100. [13] F. Bourgeois, Odd dimensional tori are contact manifolds, Internat. Math. Res. Notices 2002 (2002), 1571– 1574. [14] G.E. Bredon, Topology and Geometry, Grad. Texts in Math., vol. 139, Springer, Berlin (1993). [15] Th. Bröcker and K. Jänich, Einführung in die Differentialtopologie, Springer, Berlin (1973). [16] J. Cerf, Sur les difféomorphismes de la sphère de dimension trois (Γ4 = 0), Lecture Notes in Math., vol. 53, Springer, Berlin (1968). [17] V. Colin, Chirurgies d’indice un et isotopies de sphères dans les variétés de contact tendues, C. R. Acad. Sci. Paris Sér. I Math. 324 (1997), 659–663. [18] V. Colin, Recollement de variétés de contact tendues, Bull. Soc. Math. France 127 (1999), 43–69. [19] N. Desolneux-Moulis, Orbites périodiques des systèmes hamiltoniens autonomes (d’après Clarke, Ekeland-Lasry, Moser, Rabinowitz and Weinstein), Séminaire Bourbaki, vol. 1979/1980, Lecture Notes in Math., vol. 842, Springer, Berlin (1981), 156–173. [20] Y. Eliashberg, Three lectures on symplectic topology in Cala Gonone: Basic notions, problems and some methods, Conference on Differential Geometry and Topology (Sardinia, 1988), Rend. Sem. Fac. Sci. Univ. Cagliari, vol. 58 (1988), 27–49, suppl. [21] Y. Eliashberg, Classification of overtwisted contact structures on 3-manifolds, Invent. Math. 98 (1989), 623–637. [22] Y. Eliashberg, Filling by holomorphic discs and its applications, Geometry of Low-Dimensional Manifolds (Durham, 1989), vol. 2, London Math. Soc. Lecture Note Ser., vol. 151, Cambridge Univ. Press (1990), 45–67. [23] Y. Eliashberg, Topological characterization of Stein manifolds of dimension > 2, Internat. J. Math. 1 (1990), 29–46. [24] Y. Eliashberg, Contact 3-manifolds twenty years since J. Martinet’s work, Ann. Inst. Fourier (Grenoble) 42 (1–2) (1992), 165–192. [25] Y. Eliashberg, Classification of contact structures on R3 , Internat. Math. Res. Notices 1993 (1993), 87–91. [26] Y. Eliashberg, Legendrian and transversal knots in tight contact 3-manifolds, Topological Methods in Modern Mathematics (Stony Brook, 1991), Publish or Perish, Houston (1993), 171–193. [27] Y. Eliashberg, Symplectic topology in the nineties, Differential Geom. Appl. 9 (1998), 59–88. [28] Y. Eliashberg, Invariants in contact topology, Proceedings of the International Congress of Mathematicians (Berlin, 1998), vol. II, Doc. Math., Extra vol. II (1998), 327–338. [29] Y. Eliashberg, A. Givental and H. Hofer, Introduction to symplectic field theory, Geom. Funct. Anal., vol. 2000, Special vol., Part II (2000), 560–673. [30] Y. Eliashberg and N. Mishachev, Introduction to the h-principle, Grad. Stud. Math., vol. 48, American Mathematical Society, Providence (2002). [31] Y. Eliashberg and L. Polterovich, New applications of Luttinger’s surgery, Comment. Math. Helv. 69 (1994), 512–522.

380

H. Geiges

[32] Y. Eliashberg and W.P. Thurston, Confoliations, Univ. Lecture Ser., vol. 13, American Mathematical Society, Providence (1998). [33] J.B. Etnyre, Symplectic convexity in low-dimensional topology, Topology Appl. 88 (1998), 3–25. [34] J.B. Etnyre, Tight contact structures on lens spaces, Commun. Contemp. Math. 2 (2000), 559–577; erratum: ibid. 3 (2001), 649–652. [35] J.B. Etnyre, Introductory lectures on contact geometry, Topology and Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., vol. 71, American Mathematical Society, Providence (2003), 81– 107. [36] J.B. Etnyre, Legendrian and transverse knots, Handbook of Knot Theory, W. Menasco and M. Thistlethwaite, eds, in press. [37] J.B. Etnyre and K. Honda, On the nonexistence of tight contact structures, Ann. of Math. (2) 153 (2001), 749–766. [38] J.B. Etnyre and K. Honda, Knots and contact geometry I: Torus knots and the figure eight knot, J. Symplectic Geom. 1 (2001), 63–120. [39] J.B. Etnyre and K. Honda, Tight contact structures with no symplectic fillings, Invent. Math. 148 (2002), 609–626. [40] J.B. Etnyre and L.L. Ng, Problems in low dimensional contact topology, Topology and Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., vol. 71, American Mathematical Society, Providence (2003), 337–357. [41] H. Freudenthal, Zum Hopfschen Umkehrhomomorphismus, Ann. of Math. (2) 38 (1937), 847–853. [42] H. Geiges, Symplectic manifolds with disconnected boundary of contact type, Internat. Math. Res. Notices 1994 (1994), 23–30. [43] H. Geiges, Constructions of contact manifolds, Math. Proc. Cambridge Philos. Soc. 121 (1997), 455–464. [44] H. Geiges, Applications of contact surgery, Topology 36 (1997), 1193–1220. [45] H. Geiges, A brief history of contact geometry and topology, Exposition. Math. 19 (2001), 25–53. [46] H. Geiges, Contact topology in dimension greater than three, European Congress of Mathematics (Barcelona, 2000), vol. 2, Progress in Math., vol. 202, Birkhäuser, Basel (2001), 535–545. [47] H. Geiges, h-principles and flexibility in geometry, Mem. Amer. Math. Soc. 164 (2003), 779. [48] H. Geiges and J. Gonzalo, Contact geometry and complex surfaces, Invent. Math. 121 (1995), 147–209. [49] H. Geiges and C.B. Thomas, Contact structures, equivariant spin bordism, and periodic fundamental groups, Math. Ann. 320 (2001), 685–708. [50] V.L. Ginzburg, On closed characteristics of 2-forms, Ph.D. thesis, Berkeley (1990). [51] V.L. Ginzburg, Calculation of contact and symplectic cobordism groups, Topology 31 (1992), 767–773. [52] E. Giroux, Convexité en topologie de contact, Comment. Math. Helv. 66 (1991), 637–677. [53] E. Giroux, Topologie de contact en dimension 3 (autour des travaux de Yakov Eliashberg), Séminaire Bourbaki, vol. 1992/1993, Astérisque 216 (1993), 7–33. [54] E. Giroux, Une structure de contact, même tendue, est plus ou moins tordue, Ann. Sci. École Norm. Sup. (4) 27 (1994), 697–705. [55] E. Giroux, Structures de contact en dimension trois et bifurcations des feuilletages de surfaces, Invent. Math. 141 (2000), 615–689. [56] E. Giroux, Géométrie de contact: de la dimension trois vers les dimensions supérieures, Proceedings of the International Congress of Mathematicians (Beijing, 2002), vol. II, Higher Education Press, Beijing (2002), 405–414. [57] E. Giroux and J.-P. Mohsen, Structures de contact et fibrations symplectiques au-dessus du cercle, in preparation. [58] R.E. Gompf, Handlebody construction of Stein surfaces, Ann. of Math. (2) 148 (1998), 619–693. [59] J. Gonzalo, Branched covers and contact structures, Proc. Amer. Math. Soc. 101 (1987), 347–352. [60] D.H. Gottlieb, Partial transfers, Geometric Applications of Homotopy Theory I, Lecture Notes in Math., vol. 657, Springer, Berlin (1978), 255–266. [61] J.W. Gray, Some global properties of contact structures, Ann. of Math. (2) 69 (1959), 421–450. [62] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math. 82 (1985), 307–347. [63] H.M. Hilden, J.M. Montesinos and T. Thickstun, Closed oriented 3-manifolds as 3-fold branched coverings of S 3 of special type, Pacific J. Math. 65 (1976), 65–76. [64] M.W. Hirsch, Differential Topology, Grad. Texts in Math., vol. 33, Springer, Berlin (1976).

Contact geometry

381

[65] H. Hofer, Pseudoholomorphic curves in symplectizations with applications to the Weinstein conjecture in dimension 3, Invent. Math. 114 (1993), 515–563. [66] H. Hofer, Holomorphic curves and dynamics in dimension three, Symplectic Geometry and Topology (Park City, 1997), IAS/Park City Math. Ser., vol. 7, American Mathematical Society, Providence (1999), 35–101. [67] H. Hofer, Holomorphic curves and real three-dimensional dynamics, Geom. Funct. Anal., vol. 2000, Special vol., Part II (2000), 674–704. [68] H. Hofer and M. Kriener, Holomorphic curves in contact dynamics, Differential Equations (La Pietra, 1996), Proc. Sympos. Pure Math., vol. 65, American Mathematical Society, Providence (1999), 77–131. [69] H. Hofer, K. Wysocki and E. Zehnder, A characterisation of the tight three-sphere, Duke Math. J. 81 (1995), 159–226; erratum: ibid. 89 (1997), 603–617. [70] H. Hofer and E. Zehnder, Symplectic Invariants and Hamiltonian Dynamics, Birkhäuser, Basel (1994). [71] K. Honda, On the classification of tight contact structures I, Geom. Topol. 4 (2000), 309–368. [72] K. Honda, On the classification of tight contact structures II, J. Differential Geom. 55 (2000), 83–143. [73] H. Hopf, Zur Algebra der Abbildungen von Mannigfaltigkeiten, J. Reine Angew. Math. 105 (1930), 71–88. [74] D. Husemoller, Fibre Bundles, 3rd ed., Grad. Texts in Math., vol. 20, Springer, Berlin (1994). [75] Y. Kanda, The classification of tight contact structures on the 3-torus, Comm. Anal. Geom. 5 (1997), 413–438. [76] R.C. Kirby, The Topology of 4-Manifolds, Lecture Notes in Math., vol. 1374, Springer, Berlin (1989). [77] A.A. Kosinski, Differential Manifolds, Academic Press, Boston (1993). [78] P.B. Kronheimer and T.S. Mrowka, Monopoles and contact structures, Invent. Math. 130 (1997), 209–255. [79] F. Laudenbach, Orbites périodiques et courbes pseudo-holomorphes, application à la conjecture de Weinstein en dimension 3 (d’après H. Hofer et al.), Séminaire Bourbaki, vol. 1993/1994, Astérisque 227 (1995), 309–333. [80] P. Libermann and C.-M. Marle, Symplectic Geometry and Analytical Mechanics, Math. Appl., vol. 35, Reidel, Dordrecht (1987). [81] W.B.R. Lickorish, A representation of orientable combinatorial 3-manifolds, Ann. of Math. (2) 76 (1962), 531–540. [82] P. Lisca and G. Mati´c, Tight contact structures and Seiberg–Witten invariants, Invent. Math. 129 (1997), 509–525. [83] R. Lutz, Sur l’existence de certaines formes différentielles remarquables sur la sphère S 3 , C. R. Acad. Sci. Paris Sér. A–B 270 (1970), A1597–A1599. [84] R. Lutz, Sur quelques propriétés des formes differentielles en dimension trois, Thèse, Strasbourg (1971). [85] R. Lutz, Structures de contact sur les fibrés principaux en cercles de dimension trois, Ann. Inst. Fourier (Grenoble) 27 (3) (1977), 1–15. [86] R. Lutz, Sur la géométrie des structures de contact invariantes, Ann. Inst. Fourier (Grenoble) 29 (1) (1979), 283–306. [87] R. Lutz, Quelques remarques historiques et prospectives sur la géométrie de contact, Conference on Differential Geometry and Topology (Sardinia, 1988), Rend. Sem. Fac. Sci. Univ. Cagliari, vol. 58 (1988), 361–393, suppl. [88] S. Makar-Limanov, Tight contact structures on solid tori, Trans. Amer. Math. Soc. 350 (1998), 1013–1044. [89] J. Martinet, Sur les singularités des formes différentielles, Ann Inst. Fourier (Grenoble) 20 (1) (1970), 95–178. [90] J. Martinet, Formes de contact sur les variétés de dimension 3, Proc. Liverpool Singularities Sympos. II, Lecture Notes in Math., vol. 209, Springer, Berlin (1971), 142–163. [91] D. McDuff, Symplectic manifolds with contact type boundaries, Invent. Math. 103 (1991), 651–671. [92] D. McDuff and D. Salamon, Introduction to Symplectic Topology, 2nd ed., Oxford Univ. Press (1998). [93] S. Milligan, The Starlings, The Goon Show, Series 4, Special programme, broadcast 31st August 1954. [94] J.W. Milnor, Topology from the Differentiable Viewpoint, The University Press of Virginia, Charlottesville (1965). [95] J.W. Milnor and J.D. Stasheff, Characteristic Classes, Ann. of Math. Studies, vol. 76, Princeton Univ. Press, Princeton (1974). [96] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965), 286–294. [97] D. Rolfsen, Knots and Links, Publish or Perish, Houston (1976).

382

H. Geiges

[98] N. Saveliev, Lectures on the Topology of 3-Manifolds, de Gruyter, Berlin (1999). [99] N. Steenrod, The Topology of Fibre Bundles, Princeton Univ. Press, Princeton (1951). [100] I. Tamura, Topology of Foliations: An Introduction, Transl. Math. Monogr., vol. 97, American Mathematical Society, Providence (1992). [101] W.P. Thurston and H.E. Winkelnkemper, On the existence of contact forms, Proc. Amer. Math. Soc. 52 (1975), 345–347. [102] I. Ustilovsky, Infinitely many contact structures on S 4m+1 , Internat. Math. Res. Notices 1999 (1999), 781–791. [103] A.H. Wallace, Modifications and cobounding manifolds, Canad. J. Math. 12 (1960), 503–528. [104] A. Weinstein, On the hypotheses of Rabinowitz’ periodic orbit theorem, J. Differential Equations 33 (1979), 353–358. [105] A. Weinstein, Contact surgery and symplectic handlebodies, Hokkaido Math. J. 20 (1991), 241–251.

CHAPTER 6

Complex Differential Geometry Ion Mihai∗ Faculty of Mathematics, University of Bucharest, Str. Academiei 14, 010014 Bucharest, Romania E-mail: imihai@fmi.unibuc.ro

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 1. Complex manifolds . . . . . . . . . . . . . . . . . . 2. Almost complex structures . . . . . . . . . . . . . . 3. Dolbeault Lemma . . . . . . . . . . . . . . . . . . . 4. Kaehler manifolds . . . . . . . . . . . . . . . . . . . 5. Complex space forms . . . . . . . . . . . . . . . . . 6. Laplace–Beltrami operator on a Hermitian manifold 7. Harmonic differential forms on Kaehler manifolds . 8. Applications . . . . . . . . . . . . . . . . . . . . . . 9. Chern classes . . . . . . . . . . . . . . . . . . . . . 10. Deformation of complex structures . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

385 385 391 393 395 401 403 407 413 416 426 433

* This chapter was written while the author has visited Michigan State University, Department of Mathematics,

supported by a Fulbright grant. He expresses his hearty thanks to Professor B.Y. Chen for the hospitality he received during this visit. The author is very obliged to Professors D. Blair, B.Y. Chen, F. Defever, F. Dillen, G.D. Ludden, K. Matsumoto, R. Rosca, B. Suceav˘a, Y. Tazawa and L. Verstraelen for valuable discussions on the topics presented therein. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 383

This page intentionally left blank

Complex differential geometry

385

Introduction The geometry of complex manifolds, in particular Kaehler manifolds, is an important research topic in Differential Geometry. In the present chapter, we present the basic notions and certain important results in complex differential geometry. First, we define complex and almost complex manifolds and give standard examples. The most interesting class of complex manifolds are the Kaehler manifolds. Locally, a Kaehlerian metric differs from the Euclidean metric on the complex space Cn starting with the second power of the Taylor series. There are topological obstructions to the existence of Kaehlerian metrics on a compact complex manifold. We give examples and counterexamples. We introduce complex space forms and we state a Schur-like theorem for complex space forms. Next, we study the differential forms on a Kaehler manifold. Then, we prove the complex version of Hodge theorem. As an application, it follows that the Betti numbers of odd order on a compact Kaehler manifold are even. An interesting example of an almost Kaehler manifold which do not admit any Kaehlerian metric is the Thurston–Abbena manifold. The Iwasawa manifold is a complex manifold which does not carry any Kaehlerian metric. The Chern classes are introduced axiomatically and their construction is given. Some applications are derived. Last section deals with the deformation of complex structures in the sense of Kodaira. We state the theorems of existence and completeness. The number of moduli of a compact complex manifold is introduced.

1. Complex manifolds An n-dimensional complex manifold is a pairing (M, A), where M is a non-empty set and A = {(Uα , hα ) | α ∈ A} is a family of mappings satisfying the following properties: (i) For each α ∈ A, Uα is a subset of M and hα : Uα → Cn is one-to-one. (ii) The family {Uα }α∈A is a covering of M, i.e., M=

Uα .

α∈A

(iii) For each α, β ∈ A, the set hα (Uα ∩ Uβ ) is an open subset in Cn and the mapping n hβ ◦ h−1 α : hα (Uα ∩ Uβ ) → C

is holomorphic. (iv) If (M, A ) satisfies the properties (i)–(iii) and A ⊂ A , then A = A (i.e., (M, A) is maximal). R EMARK . If (M, B) = {(Uβ , hβ ) | β ∈ B} is a pairing satisfying the properties (i)–(iii) and we denote by A the set of all the pairs (U, h), with U ⊂ M, h : U → Cn one-to-one

386

I. Mihai

such that for each β ∈ B, h(U ∩ Uβ ), hβ (U ∩ Uβ ) ⊂ Cn are open sets and the mappings n hβ ◦ h−1 : h(U ∩ Uβ ) → Cn and h ◦ h−1 β : hβ (U ∩ Uβ ) → C respectively are holomorphic, then (M, A) is an n-dimensional complex manifold. Thus, in order to define a complex structure on a non-empty set M, it is sufficient to construct a family B satisfying the properties (i)–(iii). E XAMPLES . 1. Each open subset U of Cn admits a canonical complex structure, taking B = {(U, iU )}, where iU : U → Cn is the inclusion map of U into Cn . In particular, GL(n, C) is a complex Lie group. 2. The complex projective space P n (C) is an n-dimensional complex manifold. On − {0}, we define the equivalence relation 1 z , . . . , zn+1 ∼ w 1 , . . . , w n+1

Cn+1

⇔

∃λ ∈ C − {0} such that w i = λzi , i ∈ {1, . . . , n + 1}.

We denote by P n (C) the quotient space P n (C) = z1 , z2 , . . . , zn+1 z1 , z2 , . . . , zn+1 ∈ Cn+1 − {0} . For any α ∈ {1, . . . , n + 1}, we put Uα = {[z1 , z2 , . . . , zn+1 ] | zα = 0} and hα : Uα → Cn , hα z1 , . . . , zn+1 =

z1 zα−1 zα+1 zn+1 . , . . . , , , . . . , zα zα zα zα

Then the family B = {(Uα , hα ), α ∈ {1, . . . , n + 1}} satisfies the properties (i)–(iii). For α = β, the mapping hβ ◦ h−1 α is given by z¯ γ =

zγ , zβ

γ∈ / {α, β};

z¯ α =

1 . zβ

E XERCISE . Determine the holomorphic vector fields on the complex projective spaces P 1 (C) and P 2 (C). Solution. Locally, a holomorphic vector field v on P 1 (C) has the form d , v = az2 + bz + c dz

a, b, c ∈ C,

and a holomorphic vector field v on P 2 (C) is given by 2 ∂ v = c1 + c2 z 1 + c3 z 2 + c7 z 1 + c8 z 1 z 2 ∂z1 2 ∂ + c4 + c5 z 1 + c6 z 2 + c7 z 1 z 2 + c8 z 2 , ∂z2 where c1 , . . . , c8 ∈ R.

Complex differential geometry

387

3. The complex Grassmann manifold. Let Gp (Cn ) be the set of all p-dimensional linear subspaces of Cn . We define on Gp (Cn ) a complex structure of dimension p(n − p). Let {e1 , . . . , en } be the standard basis of Cn . For any sequence 1 i1 < i2 < · · · < ip n, let πi1 ...ip : Cn → Ei1 ...ip = Span{ei1 , . . . , eip } be the orthogonal projection. We denote by Ui1 ...ip = V ∈ Gp (Cn ); πi1 ...ip |V is a linear isomorphism . Let V ∈ Ui1 ...ip . Then, for each k ∈ {1, . . . , p}, there exists a unique fk ∈ V such that πi1 ...ip (fk ) = eik . Obviously,

f k = ei k +

ckl el .

l ∈{i / 1 ,...,ip }

Consider the mapping hi1 ...ip : Ui1 ...ip → Cp(n−p) , hi1 ...ip (V ) = ckl k∈{i

/ 1 ,...,ip } 1 ,...,ip }, l ∈{i

.

If we put B = {hi1 ...ip | 1 i1 < · · · < ip n}, then Gp (Cn ) becomes a complex manifold of dimension p(n − p). R EMARK . G1 (Cn+1 ) = P n (C). 4. The Calabi manifolds S 2m+1 × S 2n+1 [9]. The differentiable manifold M = S 2m+1 × S 2n+1 has real dimension 2m + 2n + 2. We may consider S

2m+1

= ξ ∈C

S

2n+1

= η∈C

m+1

m+1

|ξ | = 1 , α 2

α=1

n+1

β 2

η = 1 ,

n+1

β=1

then M ⊂ Cm+1 × Cn+1 . Let j ∈ {1, . . . , m + 1} and k ∈ {1, . . . , n + 1} and denote by Uj k = (ξ, η) ∈ M | ξ j ηk = 0 . We will construct a family of mappings hj k : Uj k → Cm+n+1 , which defines a complex structure on M. Put

388

I. Mihai

hj k (ξ, η) 1 ξ ξ j −1 ξ j +1 ξ m+1 η1 ηk−1 ηk+1 ηn+1 = ,..., j , j ,..., j , k ,..., k , k ,..., k ,t , ξj ξ ξ ξ η η η η where t=

1 log ξ j + i log ηk 2πi

mod (1, i).

5. Each orientable surface M is a 1-dimensional complex manifold. Let g be a Riemannian metric on M. By a theorem of Lichtenstein, locally, the metric can be written as g = λ2 (dx 2 +dy 2 ), with λ > 0. Putting z = x +iy, M becomes a complex manifold of dimension 1. In particular, the 2-sphere S 2 admits a complex structure. We will give a direct construction of a complex structure on S 2 . It is known that S 2 is a 2-dimensional differentiable manifold. Let S2 =

2 2 2 1 2 3 u , u , u ∈ E3 u1 + u2 + u3 = 1 .

Denote by UN = S 2 − {N } and US = S 2 − {S}, where N = (0, 0, 1) and S = (0, 0, −1) are the north and south poles, respectively. Consider the stereographic projections hN : UN → R2 ,

hS : US → R2

from the north and south poles, respectively. Their equations are u1 u2 , , 1 − u3 1 − u3 1 2 3 u1 u2 , hS u , u , u = , 1 + u3 1 + u3

hN u1 , u2 , u3 =

respectively. We define h˜ N : UN → C and h˜ S : US → C, by h˜ N u1 , u2 , u3 =

u1 u2 +i 3 1−u 1 − u3

h˜ S u1 , u2 , u3 =

u1 u2 − i . 1 + u3 1 + u3

and

Complex differential geometry

389

The mapping hS ◦ h−1 N : C − {0} → C,

1 hS ◦ h−1 N (z) = , z

is holomorphic. Thus the family B = {(UN , U˜ N ), (US , h˜ S )} satisfies the conditions (i)–(iii), and therefore the sphere S 2 is a 1-dimensional complex manifold. Next, we will indicate one procedure for obtaining new complex manifolds. Let M be a complex manifold. The set of all automorphisms of M is a group G endowed with the composition of mappings. Any subgroup G of G is called a group of automorphisms of M. For any p ∈ M, the set Gp = {g(p) | g ∈ G} is the orbit of G at p. Obviously Gp ∩ Gq = ∅ if and only if q ∈ Gp . The set of all orbits, denoted by M/G , is the quotient space of M via G. E XAMPLE . Let M = Cn+1 − {0}. Each g ∈ C∗ defines an automorphism of M by g : z1 , . . . , zn+1 → gz1 , . . . , gzn+1 . Thus C∗ becomes a group of automorphisms of M and M/C∗ = P n (C). D EFINITION . A group G of automorphisms of the complex manifold M is called properly discontinuous if for any compact subsets K1 and K2 in M, the set {g ∈ G | g(K1 )∩K2 = ∅} is finite. In this case, each orbit Gp is a discrete set. We say that G is fixed points free if each g ∈ G − {1M } has no fixed points. R EMARK . In the above example, C∗ is fixed points free, but it is not properly discontinuous. T HEOREM 1.1. Let M be an n-dimensional complex manifold and G a fixed points free and properly discontinuous group of automorphisms. Then M/G carries a natural n-dimensional complex structure induced by the complex structure of M. Using this theorem, we can obtain other examples of complex manifolds. 6. The complex torus is defined as follows. Let {ω1 , . . . , ω2n } be 2n vectors in Cn linearly independent over R. We define a fixed points free and properly discontinuous group of automorphisms of Cn , by G = z → z +

2n

mj ωj mj ∈ Z, j ∈ {1, . . . , 2n} .

j =1

The quotient space T n = Cn /G , which we call the n-dimensional complex torus, becomes a complex manifold.

390

I. Mihai

R EMARK . For n = 1, T 1 is an algebraic curve. If n 2, then T n is not necessarily an algebraic manifold. T n is an algebraic manifold if and only if its period matrix is a Riemann matrix (see [54]). An n-dimensional complex manifold M is said to be complex parallelizable if there exist n holomorphic vector fields which are linearly independent at every point of M. Every complex torus is complex parallelizable. H.G. Wang [61] proved that every compact complex parallelizable manifold can be written as a quotient space G/D of a complex Lie group G by a discrete subgroup D. 7. The Hopf manifolds. Let M = Cn − {0} and α1 , . . . , αn ∈ C such that |αj | > 1 (j ∈ {1, . . . , n}) and let G be the cyclic group spanned by the automorphism g of M, given by g : z = z1 , . . . , zn → g(z) = α1 z1 , . . . , αn zn . Applying the above theorem, M/G has a complex structure; it is called Hopf manifold. It is known (see [37]) that M/G is diffeomorphic with S 1 × S 2n−1 . Thus, Hopf manifolds are a particular case of Calabi manifolds. We want to point out that they are not algebraic manifolds. E XERCISE . Let L = {([z], ξ ) ∈ P n (C) × Cn+1 | ξ ∈ [z]}. Prove that (L, pr1 , P n (C)) is a holomorphic vector bundle of rank 1 (called the tautological vector bundle over the complex projective space). 8. The blowing-up at a point. Consider a small ball B = Br (0) ⊂ Cn , with n 2. Let z = (z1 , . . . , zn ) be the standard ¯ is the complex n-dimensional coordinates of Cn . The blowing-up of B at 0, denoted by B, manifold defined by B¯ = (z, w) ∈ B × P n−1 (C) | zi w j = zj w i , ∀1 i < j n , where w = [w 1 , . . . , w n ] ∈ P n−1 (C). If we cover P n−1 (C) by the open subsets Ui = {w ∈ P n−1 (C) | w i = 0}, i = 1, . . . , n, then B × P

This page intentionally left blank

Handbook of Differential Geometry •

VOLUME II Editors

Franki J.E. Dillen Leopold C.A. Verstraelen Katholieke Universiteit Leuven Department of Mathematics Leuven, Belgium

Amsterdam • Boston • Heidelberg • London • New York • Oxford Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo North-Holland is an imprint of Elsevier

ELSEVIER B.V. Radarweg 29 P.O. Box 211, 1000 AE Amsterdam The Netherlands

ELSEVIER Inc. 525 B Street, Suite 1900 San Diego, CA 92101-4495 USA

ELSEVIER Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB UK

ELSEVIER Ltd 84 Theobalds Road London WC1X 8RR UK

© 2006 Elsevier B.V. All rights reserved. This work is protected under copyright by Elsevier B.V., and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier’s Rights Department in Oxford, UK: phone (+44) 1865 843830, fax (+44) 1865 853333, e-mail: [email protected] Requests may also be completed on-line via the Elsevier homepage (http://www.elsevier.com/locate/permissions). In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+1) (978) 7508400, fax: (+1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 20 7631 5555; fax: (+44) 20 7631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of the Publisher is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier’s Rights Department, at the fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.

First edition 2006 Library of Congress Cataloging in Publication Data A catalog record is available from the Library of Congress. British Library Cataloguing in Publication Data A catalogue record is available from the British Library. ISBN-13: 978-0-444-52052-4 ISBN-10: 0-444-52052-X

∞ The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.

Dedication In memory of S.S. Chern and T. Willmore

This page intentionally left blank

Preface “Our goal with the volumes which together will constitute the “Handbook of Differential Geometry” is to give a rather complete survey of the field of differential geometry.” Thus reads the opening sentence of the “Handbook of Differential Geometry, Volume I”, and only the presence of the word “rather” saves this goal from being an obvious mission impossible. Let us recall the contents of this Volume I: Differential geometry of webs (M.A. Akivis and V.V. Goldberg), Spaces of metrics and curvature functionals (D.E. Blair), Riemannian submanifolds (B.-Y. Chen), Einstein metrics in dimension four (A. Derdzinski), The Atiyah–Singer index theorem (P.B. Gilkey), Survey of isospectral manifolds (C.S. Gordon), Submanifolds with parallel fundamental form (Ü. Lumiste), Sphere theorems (K. Shiohama), Affine differential geometry (U. Simon), A survey on isoparametric hypersurfaces and their generalizations (G. Thorbergsson), Curves (T. Willmore); with introduction by S.S. Chern. As in Volume I, we allowed the authors in this Volume II as much freedom as possible concerning style and contents. We are confident that the reader will appreciate this pragmatic point of view. Some contributions will emphasize the basics; some will emphasize the classical results; others the recent developments. Needless to say all authors have spent a lot of time and energy in describing their topic, which we appreciate enormously. The contributions to this Volume II are: Some problems on Finsler geometry (J.C. Álvarez Paiva), Foliations (R. Barre and A. El Kacimi), Symplectic geometry (A. Cannas da Silva), Metric Riemannian geometry (K. Fukaya), Contact geometry (H. Geiges), Complex differential geometry (I. Mihai), Compendium on the geometry of Lagrange spaces (R. Miron), Certain actual topics on modern Lorentzian geometry (F.J. Palomo and A. Romero). Obviously the whole field of differential geometry is not yet covered in the two volumes of this “Handbook of Differential Geometry”. Some of the authors explicitly mention topics that should have been covered, but are not for practical reasons; but also other topics are not (yet) treated sufficiently or not treated at all. Recently Professors Chern and Willmore passed away. Both had a great impact on the development of contemporary geometry and were genuine sources of inspiration, guidance and support for many generations of mathematicians through their books and articles, their fantastic lectures and their warm and truly concerned personal contacts. Together with all authors we gratefully dedicate this book to the memories of Professor S.S. Chern and Professor T.J. Willmore.

Franki Dillen and Leopold Verstraelen vii

This page intentionally left blank

List of Contributors Álvarez Paiva, J.C., Polytechnic University, Brooklyn, NY (Ch. 1). Barre, R., Université de Valenciennes, Valenciennes (Ch. 2). Cannas da Silva, A., Instituto Superior Técnico, Lisboa (Ch. 3). El Kacimi Alaoui, A., Université de Valenciennes, Valenciennes (Ch. 2). Fukaya, K., Kyoto University, Kyoto (Ch. 4). Geiges, H., Universität zu Köln, Köln (Ch. 5). Mihai, I., University of Bucharest, Bucharest (Ch. 6). Miron, R., “Al.I. Cuza” University Iasi, Iasi (Ch. 7). Palomo, F.J., Universidad de Málaga, Málaga (Ch. 8). Romero, A., Universidad de Granada, Granada (Ch. 8).

ix

This page intentionally left blank

Contents Preface List of Contributors Contents of Volume I 1. Some problems on Finsler geometry J.C. Álvarez Paiva 2. Foliations R. Barre and A. El Kacimi Alaoui 3. Symplectic geometry A. Cannas da Silva 4. Metric Riemannian geometry K. Fukaya 5. Contact geometry H. Geiges 6. Complex differential geometry I. Mihai 7. Compendium on the geometry of Lagrange spaces R. Miron 8. Certain actual topics on modern Lorentzian geometry F.J. Palomo and A. Romero Author Index Subject Index

vii ix xiii 1 35 79 189 315 383 437 513

547 555

xi

This page intentionally left blank

Contents of Volume I Preface Introduction List of Contributors 1. Differential geometry of webs M.A. Akivis and V.V. Goldberg 2. Spaces of metrics and curvature functionals D.E. Blair 3. Riemannian submanifolds B.-Y. Chen 4. Einstein metrics in dimension four A. Derdzinski 5. The Atiyah–Singer index theorem P.B. Gilkey 6. Survey of isospectral manifolds C.S. Gordon 7. Submanifolds with parallel fundamental form Ü. Lumiste 8. Sphere theorems K. Shiohama 9. Affine differential geometry U. Simon 10. A survey on isoparametric hypersurfaces and their generalizations G. Thorbergsson 11. Curves T. Willmore Author Index Subject Index

v vii ix 1 153 187 419 709 747 779 865 905 963 997

1025 1037

xiii

This page intentionally left blank

CHAPTER 1

Some Problems on Finsler Geometry∗ J.C. Álvarez Paiva∗∗ Department of Mathematics, Polytechnic University, Six MetroTech Center, Brooklyn, NY 11201, USA E-mail: [email protected]

We do like intuitive geometric arguments and uncovering simple geometric reasons underlying seemingly recondite facts. H. Busemann

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . 2. Preliminaries . . . . . . . . . . . . . . . . . . . . 2.1. The Hamiltonian point of view . . . . . . . . 2.2. The Riemannian point of view . . . . . . . . 2.3. Isometries and isometric embeddings . . . . 2.4. Isometric submersions . . . . . . . . . . . . 3. Volume and area in Finsler spaces . . . . . . . . . 4. Unit spheres in Minkowski spaces . . . . . . . . . 5. Symplectic equivalence of Finsler manifolds . . . 5.1. Equivalence of unit co-disc bundles . . . . . 5.2. Equivalence of unit co-sphere bundles . . . . 5.3. Manifolds of geodesics . . . . . . . . . . . . 6. Around Hilbert’s fourth problem . . . . . . . . . . 7. Closed geodesics . . . . . . . . . . . . . . . . . . 8. Differential invariants of Finsler surfaces . . . . . 8.1. Convex geometry and the invariant I . . . . 8.2. The invariant J . . . . . . . . . . . . . . . . 8.3. Curvature of Finsler surfaces . . . . . . . . . 8.4. Cartan’s structure equations . . . . . . . . . 8.5. Applications of Cartan’s structure equations References . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . .

3 4 5 7 7 8 9 12 14 15 15 17 19 22 23 23 24 24 26 28 31

* This work was partially funded by a crédit aux chercheurs from the FNRS, Belgium. ** I warmly thank G. Berck, D. Burago, C. Durán, E. Fernandes, P. Foulon, I.M. Gelfand, R. Schneider, Z. Shen

and S. Tabachnikov for some very useful conversations and/or criticisms of earlier versions of this work. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 1

2

J.C. Álvarez Paiva

Abstract This chapter is an unorthodox survey of Finsler geometry presenting both results and open problems. It aims to show that recent progress in convex geometry, the calculus of variations, symplectic geometry, and integral geometry can be powerful tools in the study of Finsler manifolds; and that Finsler geometry can prove useful in solving some of the open problems in these fields.

Some problems on Finsler geometry

3

1. Introduction Finsler manifolds, manifolds whose tangent spaces carry a norm that varies smoothly with the base point, were born prematurely in 1854 together with their Riemannian counterparts in Riemann’s ground-breaking Habilitationsvortrag. I say prematurely because in 1854 Minkowski’s work on normed spaces and convex bodies (see [69]) was still forty three years away, and thus not even the infinitesimal geometry on which Finsler manifolds are based was understood or appreciated at the time. Apparently, Riemann did not know what to make of these ‘more general class’ of manifolds whose element of arclength does not originate from a scalar product and, fatefully, put in a bad word for them [44]: Investigation of this more general class would actually require no essential different principles, but it would be rather time-consuming and throw relatively little new light on the study of Space, especially since the results cannot be expressed geometrically.

Given the awe with which we rightfully regard Riemann’s achievements and uncanny geometrical intuition, it is tempting to take the above quotation out of historical context and to dismiss Finsler geometry altogether. But, if we think of the great advances in convex geometry, the calculus of variations, integral geometry, the theory of metric spaces, and symplectic geometry that have taken place since 1854, then we may be moved to reassess Riemann’s statement and to consider applying these new tools to develop the subject in a way that Riemann could not have foreseen. The paper includes eighteen simply-stated open problems, as well as a survey of the more elementary and geometric chapters of Finsler geometry. It presents a detailed discussion of the Holmes–Thompson volume and its role in integral geometry and geometric inequalities, thus complementing the survey by Álvarez and Thompson [16]. The other highlights of the paper are its presentation of Hilbert’s fourth problem and its elementary approach to the differential invariants of Finsler surfaces. These are mostly based on the papers [15,12,13] with I.M. Gelfand, M. Smirnov and E. Fernandes, as well as on the lecture notes [10] written jointly with C. Durán. In view of the often-made criticisms of Finsler geometry—very few concrete and interesting examples, very few non-Riemannian theorems of real geometric content, and too many subindices—I have tried to include as many concrete examples, simply-stated results, and geometric constructions as possible. In this way, many of the jewels, so to speak, of Finsler geometry find their way into the following pages. As anyone writing a survey paper, I have had to make some choices. In matters of taste, I have consistently preferred the concrete to the abstract, the elementary to the advanced, the C ∞ to the C k , and the global to the local. I have stayed clear of Riemann–Finsler geometry and Finsler connections because the book [23] of Bao, Chern and Shen covers the subject in depth as do the lecture notes of Abate and Patrizio [1]. Because of my ignorance of the subject, I have not touched on complex Finsler geometry (see [1] also for this topic) and, despite their undeniable interest and importance, non-reversible Finsler metrics are barely mentioned. Another important topic that is not covered in this survey is Busemann’s G-spaces. This approach, which consists in abstracting the properties of geodesics on Finsler manifolds, is one of the most powerful in Finsler geometry, but it is impossible to outdo Busemann’s own exposition in [37,38].

4

J.C. Álvarez Paiva

The reader wishing to have a broader view of the activity in Finsler geometry should read the short surveys of Chern [44] and Busemann [35]. The book [20] contains survey articles on several topics including non-reversible Finsler metrics, the proceedings [22] contains many open problems and a vista of various approaches to Finsler geometry. The book [84] contains a beautiful exposition of the convex-geometric aspects of the Holmes– Thompson volume as well as most of the convex geometry necessary for the study of Finsler manifolds. The lecture notes [10] are similar in spirit to the present paper. The thesis of Egloff [46] is a good place to learn about the beautiful results of Egloff [47,48] and Foulon [50–52] on the geometry and dynamics of Finsler manifolds with non-positive and negative curvature. Finally, I wholeheartedly recommend looking at the papers [42] and [3] before plunging into other papers where Finsler connections are treated. The reader can find many of the preprints cited in this paper, along with other works on the interaction between convex, integral, metric, and symplectic geometry, in http://www.math.poly.edu/research/finsler. 2. Preliminaries If (V , · ) is a real, finite-dimensional normed space, we define the length of a smooth curve γ : [a, b] → V by the formula length of γ :=

γ˙ (t) dt.

b

a

A smooth submanifold N ⊂ V inherits a metric from the norm: if x and y are two points on N , define their distance as the infimum of the lengths of all smooth curves on N joining x and y. Notice that in order to define the metric on N it suffices to know the restriction of the norm to each tangent space. This motivates the following heuristic definition: A Finsler manifold is a manifold together with the choice of a norm on each tangent space. The precise definition requires us to restrict the class of norms to those where the unit sphere is smooth and quadratically convex (i.e., it has positive principal curvatures for some (and therefore any) Euclidean structure on V ). The intrinsic definition of these norms is as follows: Let V be a vector space and let ϕ : V → [0, ∞) be a norm that is smooth outside the origin. Set L := ϕ 2 /2 and consider the exterior derivative of L, dL, as a map from V minus the origin to V ∗ minus the origin. The norm ϕ is said to be a Minkowski norm if dL is a diffeomorphism. For any non-zero vector v ∈ V , the differential D(dL)(v) : Tv V → TdL(v) V ∗ is an invertible linear map. In fact, using the natural identification of Tv V with V , and TdL(v) V ∗ with V ∗ , we can think of gϕ (v) := D(dL)(v) as a (symmetric) bilinear form on V : gϕ (v)(w1 , w2 ) := D(dL)(v)(w1 ) (w2 ).

Some problems on Finsler geometry

5

The norm ϕ is a Minkowski norm if and only if gϕ is positive definite. When the vector v belongs to the unit sphere, we will denote gϕ (v) as the osculating Euclidean structure at v and the ellipsoid Ev := w ∈ V : gϕ (v)(w, w) = 1 as the osculating ellipsoid at v. D EFINITION 2.1. A Finsler metric on a manifold M is a continuous function defined on its tangent bundle with the property that it is smooth away from the zero section and its restriction to each tangent space is a Minkowski norm. Some examples of Finsler manifolds are submanifolds of Minkowski spaces and flat tori obtained as quotients of Minkowski spaces. If γ : [a, b] → M is a smooth curve on a Finsler manifold (M, ϕ), then the quantity length of γ :=

b

ϕ γ˙ (t) dt

(1)

a

is independent of the parameterization. Using this definition of length we define a metric on M by letting the distance between two points x, y ∈ M to be the infimum of the lengths of all smooth curves joining x and y. Finsler manifolds are length spaces: the length of a curve γ defined by the integral in (1) equals the metric length of the curve given by k−1 sup dist γ (ti ), γ (ti+1 ) : a = t0 < · · · < tk = b is a partition of [a, b] . i=0

The condition that the norms in each tangent space be Minkowski norms is necessary for the study of the geodesics. Namely, we want these to be solutions of a second-order differential equation on M.

2.1. The Hamiltonian point of view If (V , ϕ) is a normed space, then the dual vector space V ∗ inherits a natural norm defined by the equation

ϕ ∗ (ξ ) := sup ξ (v) : ϕ(v) 1 . A related construction on Minkowski spaces is the Legendre transform which assigns to a non-zero vector v ∈ V the covector ϕ(v) dϕ(v) = gϕ (v)(v, ·). It is easy to check that if v belongs to the unit sphere S ⊂ (V , ϕ), then the Legendre transform of v is the unique covector ξ such that ξ = 1 is the hyperplane tangent to S at the point v. This implies that the image of S under the Legendre transform is the unit sphere in (V ∗ , ϕ ∗ ).

6

J.C. Álvarez Paiva

Let (M, ϕ) be a Finsler manifold and for each point m ∈ M let ϕm denote the Minkowski ∗ ) is the dual of the normed space (T M, ϕ ), then the function norm on Tm M. If (Tm∗ M, ϕm m m H : T ∗M → R ∗ (p ) is a Hamiltonian whose energy surfaces are fiber-wise condefined by H (pm ) := ϕm m vex. Applying the Legendre transform on each tangent space of M defines a diffeomorphism L : T M \ 0 → T ∗ M \ 0 with the property that H ◦ L = ϕ. By passing from a Finsler metric to its associated Hamiltonian, we gain access to the techniques of Hamiltonian mechanics and symplectic geometry. Below, we recall some of the basic definitions and constructions. For more information see [18] and [2].

D EFINITION 2.2. Let π : T ∗ M → M be the standard projection and let Dπ : T (T ∗ M) → T M be its differential. The canonical 1-form α on T ∗ M is defined by the equation α(vpm ) = pm (Dπ(vpm )), where pm ∈ Tm∗ M and vpm ∈ Tpm (T ∗ M). The symplectic 2-form is defined as ω := −dα. The form ω is non-degenerate: at each point pm ∈ T ∗ M, the map vpm → ωpm (vpm , ·) is an isomorphism from Tpm (T ∗ M) to Tp∗m (T ∗ M). We can use this isomorphism to pass from 1-forms on T ∗ M to vector fields on T ∗ M. D EFINITION 2.3. Let H : T ∗ M → R be a smooth function. The Hamiltonian vector field of H , XH , is defined by the equality dH = ω(XH , ·). As an easy consequence of the definition, we have that H is constant along the integral curves of the Hamiltonian vector field XH , and that the symplectic form is invariant under the flow of XH . Because of this result, it is usual to disregard the function H in favor of the unit co∗ M := H −1 (1). If α is the canonical 1-form on T ∗ M, then its restriction sphere bundle SH ∗ M, which we denote by α , is a contact form (i.e., the to the unit co-sphere bundle SH H n−1 never vanishes). Using αH , we can define the restriction top-order form αH ∧ (dαH ) of the Hamiltonian vector field XH without any reference to the function H : ∗ M is defined by the equations D EFINITION 2.4. The Reeb vector field XH on SH

αH (XH ) = 1,

dαH (XH , ·) = 0.

The projection to M of the integral curves of this vector field are geodesics parameterized with unit speed. Conversely, if γ is a geodesic on M parameterized with unit speed, then the Legendre transform L maps the velocity curve γ˙ to an orbit of the Reeb vector field. We remark that if γ is any smooth curve on M parameterized with unit speed, then length of γ =

L◦γ

αH .

(2)

Some problems on Finsler geometry

7

Note that the Finsler manifold (M, ϕ) is geodesically complete (or metrically complete, since it is easy to verify that the Hopf–Rinow theorem extends to the Finsler setting) if and only if the Reeb vector field defines a flow. Let us finish this section by remarking that the non-degeneracy of the symplectic form ω on T ∗ M is equivalent to the fact that ωn , n = dim(M), is a volume form. This remark will provide us with a natural way to define the volume of a Finsler manifold. 2.2. The Riemannian point of view Finsler manifolds can also be studied from the point of view of Riemannian manifolds and bundles. Indeed, to every unit vector vm ∈ Tm M we may associate the inner product gϕ (v). In this way, we can define a Riemannian structure on the pullback of the tangent bundle of M to the unit tangent bundle of M. This construction underlies many of the definitions of connections associated to Finsler manifolds. A variation on this theme is to take a nowhere zero vector field X defined on an open subset O ⊂ M and to associate to it the Riemannian metric on O defined by m → gϕ (X(m)). This construction has been used by Shen (see [79] and Section 8) to give a simple description of the Finsler curvature. 2.3. Isometries and isometric embeddings The definitions of isometry and isometric embedding between Finsler manifolds (M, ϕM ) and (N, ϕN ) are the same as for Riemannian manifolds. Namely, an isometry (respectively isometric embedding) is a diffeomorphism (respectively embedding) f : M → N such that f ∗ ϕN = ϕM . Unlike Riemannian manifolds, two Finsler manifolds can fail to be isometric because of what happens at a single tangent space. For example, if at a point m ∈ M the indicatrix Sm M := vm ∈ Tm M: ϕM (vm ) = 1 is an ellipsoid while none of the indicatrices of N are ellipsoids, then M and N are not isometric. This remark points at the important role played by the centro-affine geometry of convex hypersurfaces in Finsler geometry. In [32] Burago and Ivanov showed that any compact Finsler manifold admits an isometric embedding into a finite-dimensional normed space. It is likely, but unproved, that the norm can be chosen to be a Minkowski norm. They also give examples of noncompact Finsler manifolds that cannot be isometrically embedded in any finite-dimensional normed space. Examples of Finsler manifolds that cannot be isometrically embedded in any Minkowski space have been given by Shen (see [79]) and by Álvarez and Durán (see Section 8.5). However, new types of embedding problems arise in Finsler geometry. For example, while it is known that every two-dimensional normed space is isometric to a subspace of L1 ([0, 1]) (see, for example, [29]), it is not clear whether the following problem has an affirmative answer.

8

J.C. Álvarez Paiva

P ROBLEM 1. Does every two-dimensional Finsler manifold admit an isometric embedding into the Banach space L1 ([0, 1])? The relation between intrinsic and extrinsic geometric properties of submanifolds of normed space is not understood. For example, the following problem from [84] is open. P ROBLEM 2 (Thompson). Let X and Y be two normed spaces of dimension n, n > 2, such that its unit spheres are isometric as Finsler manifolds. Does it follow that X and Y are isometric as normed spaces?

2.4. Isometric submersions In Riemannian geometry isometric submersions are used to construct examples of Riemannian manifolds while keeping some control on their geodesics and curvature. The Finslerian generalization of this construction is simple, but perhaps not as well known as it should be. What follows is taken from [11]. D EFINITION 2.5. A surjective linear map π : X → Y between two normed spaces is said to be an isometric submersion if the image of the closed unit ball on X under the map π equals the closed unit ball on Y . Clearly, if x is any vector in X, then π(x)Y xX . The vectors for which the equality holds are called horizontal vectors and form the horizontal cone in X. Notice that if the unit sphere in X is smooth, its intersection with the horizontal cone is the singular set of the restriction of the map π to the unit sphere. This description makes it easy to grasp that, unlike the case where the spaces is Euclidean, the horizontal cone is rarely a subspace. D EFINITION 2.6. A submersion ρ : M → N between Finsler manifolds is said to be isometric if for every point m ∈ M the differential Dm ρ : Tm M → Tρ(m) N is an isometric submersion of normed spaces. More generally, Berestovskii has defined in [24] isometric submersions, or submetries of metric spaces, as maps that send metric balls to metric balls of the same radius. When the metric spaces are Finsler manifolds both notions of isometric submersion agree. However, in the particular case of Finsler manifolds we can also speak of horizontal lifts. D EFINITION 2.7. Let ρ : M → N be an isometric submersion. An immersed curve γ : [a, b] → M is said to be horizontal if for every t ∈ (a, b) the velocity vector γ˙ (t) belongs to the horizontal cone in Tγ (t) M. A curve γ : [a, b] → M is said to be a horizontal lift of an immersed curve σ : [a, b] → N if γ is horizontal and ρ ◦ γ = σ .

Some problems on Finsler geometry

9

The following result is an easy consequence of the definitions and of the basic properties of geodesics in a Finsler manifold. T HEOREM 2.8 (Álvarez and Durán [11]). An immersed curve on N is a geodesic if and only if any of its horizontal lifts is a geodesic on M. In particular, the geodesics of N are precisely the projections of horizontal geodesics on M. As we shall see in Section 6, this theorem is useful in constructing interesting examples of Finsler metrics on complex and quaternionic projective spaces.

3. Volume and area in Finsler spaces The theory of volume and area in normed spaces has long been a major driving force in convex geometry. For example, the Blaschke–Santaló inequality, the Mahler conjecture, the Busemann–Petty problems, the Shephard problem, and the numerous works of Busemann, Ewald and Shephard on the notions of convexity on Grassmannians originated from or have applications to the study of volumes and areas in normed spaces (see [16]). Defining a volume on a finite-dimensional normed space seems easy: a natural volume should be invariant under translations, positive on open sets, and finite on compact sets. By Haar’s theorem, such a volume must be a multiple of the Lebesgue measure. However, the choice of this multiple is crucial. To understand this, suppose we have already decided on how to assign those constants to two-dimensional normed spaces. We can now define the area of a two-dimensional polyhedral surface embedded in a 3-dimensional normed space as the sum of the areas of its faces with their induced norms. Making a different choice of constants leads to a completely different way of measuring the area of polyhedral surfaces. The guiding principle in defining volumes on normed spaces is that the choice of a volume for every k-dimensional normed space leads to the definition of the k-volume integrand in all higher-dimensional normed spaces. Requiring even mild conditions on these area integrands, such as that regions in hyperplanes be area-minimizing, severely restricts our choices for a definition of volume. In fact, in the literature one can only find three reasonable choices of volume on normed spaces: the Busemann definition, the Holmes– Thompson definition, and the Benson definition (also known as Gromov’s mass∗ [56]). Since the Busemann volume of a Finsler manifold coincides with its Hausdorff measure as a metric space, at first sight it seems the most natural and geometric definition. However, the Holmes–Thompson definition, with its ties to Brunn–Minkowski theory, integral, and symplectic geometry, is rapidly becoming the definition of choice. D EFINITION 3.1. The Holmes–Thompson volume of an n-dimensional Finsler manifold (M, ϕ), voln (M, ϕ), is the symplectic volume of its unit co-disc bundle divided by the volume of the Euclidean n-dimensional unit ball. The k-volume of a k-dimensional submanifold is the volume of the submanifold with its induced Finsler metric.

10

J.C. Álvarez Paiva

Note that, using the notation of the previous section, we have that voln (M) =

1 n n!

∗M SH

αH ∧ (dαH )n−1 ,

(3)

where n is the volume of the Euclidean unit ball of dimension n. Using the Blaschke–Santaló inequality, Durán has remarked in [45] that the Hausdorff measure of a Finsler manifold is no less than its Holmes–Thompson volume. Equality holds if and only if the metric is Riemannian. One of the basic problems about a given definition of volume on normed and Finsler spaces is determining the convexity or ellipticity properties of the k-volume integrands. Perhaps the most enticing and difficult question of the kind is the following problem of Busemann: P ROBLEM 3. Let P be a compact polyhedron of dimension k in a normed space. Is the k-volume of any given face less than or equal to the sum of the k-volumes of the remaining faces? In the case of polyhedra of codimension one the answer to this question is affirmative for both the Hausdorff measure (Busemann [34]) and the Holmes–Thompson volume (Holmes and Thompson [62]). For the Hausdorff measure no other results of this kind are known. For the Holmes– Thompson volume the answer to Busemann’s question is known to be affirmative if the normed space is a subspace of L1 ([0, 1]): T HEOREM 3.2 (Busemann et al. [40]). Let P be a compact polyhedron of dimension k in the Banach space L1 ([0, 1]). The (Holmes–Thompson) k-volume of any given face is less than or equal to the sum of the k-volumes of the remaining faces. The latest progress on the question of minimality of flats in normed spaces is the following result of Burago and Ivanov: T HEOREM 3.3 (Burago and Ivanov [33]). Let P be a compact 2-dimensional polyhedron in a normed space. If P is homeomorphic to a sphere, then the (Holmes–Thompson) 2-volume of any given face is less than or equal to the sum of the 2-volumes of the remaining faces. In [64], S. Ivanov shows that this result can be extended to a theorem that is new even in the Riemannian case: T HEOREM 3.4 (Ivanov [64]). Let ϕ be a Finsler metric on the closed two-dimensional disc D such that every two points on D are joined by a unique geodesic. If ψ is another Finsler metric on D such that the distance induced by ψ on the boundary ∂D of D is greater than or equal to the distance induced by ϕ on ∂D, then the Holmes–Thompson volume of (D, ϕ) does not exceed that of (D, ψ).

Some problems on Finsler geometry

11

It follows from Ivanov’s theorem that a two-dimensional totally geodesic submanifold in a Finsler space is minimal—in the usual sense of being an extremal for the 2-volume functional—with respect to the Holmes–Thompson definition of area. This is true in all dimensions: T HEOREM 3.5 (Berck [25]). A totally geodesic submanifold of a Finsler manifold is minimal with the respect to the Holmes–Thompson definition of area. In [9], Álvarez and Berck show that Theorems 3.4 and 3.5 no longer hold if the Holmes– Thompson volume is replaced by the Hausdorff measure. One of the advantages of working with the Holmes–Thompson definition is that there is a remarkably simple formula for the Holmes–Thompson k-volume densities of a Minkowski space in terms of the Fourier transform of its norm. In a different guise, this formula was first obtained by W. Weil [85]. In the present form it was rediscovered by Álvarez and Fernandes in [13]. F OURIER TRANSFORMS OF NORMS . Let φ be a smooth, even homogeneous function of degree one on an n-dimensional vector space V , let e1 , . . . , en be a basis of V , and let ξ 1 , . . . , ξ n be the dual basis in V ∗ . These bases allow us to introduce coordinates (x1 , . . . , xn ) in V and (ξ1 , . . . , ξn ) in V ∗ , which we can use to compute the standard (distributional) Fourier transform ˆ ) := eiξ ·v φ(v) dv. φ(ξ Rn

This transform depends on the choice of basis, or rather on the Lebesgue measure associated to it. However, the form φˆ dξ1 ∧ · · · ∧ dξn does not. Up to a constant factor, the Fourier transform of φ is the contraction of this n-form with the Euler vector field, XE (ξ ) = ξ , in V ∗ : φˇ :=

−1 φˆ dξ1 ∧ · · · ∧ dξn XE . 4(2π)n−1

It is known (see [63, pp. 167–168]) that φˆ is smooth on V ∗ \ {0} and homogeneous of degree −n − 1. It follows that φˇ is a smooth differential form on V ∗ \ {0} which is homogeneous of degree −1. D EFINITION 3.6. Let (V , ϕ) be an n-dimensional Minkowski space. For each integer k, 1 k < n, define the integrand ϕk (v1 ∧ · · · ∧ vk ) := (ξ 1 ,...,ξ k

)∈S ∗k

|ξ 1 ∧ · · · ∧ ξ k · v1 ∧ · · · ∧ vk |ϕˇ k ,

(4)

where S ∗ is any closed hypersurface in V ∗ \ {0} that is star-shaped with respect to the origin.

12

J.C. Álvarez Paiva

T HEOREM 3.7 [13]. Let (V , ϕ) be an n-dimensional Minkowski space. If N ⊂ V is an immersed submanifold of dimension k, 1 k < n, then we have the following formula for the Holmes–Thompson k-area of N : 1 volk (N ) = k

ϕk , N

where k denotes the volume of the Euclidean unit ball of dimension k. One of the main justifications for adopting the Holmes–Thompson volume comes from its role in integral geometry (see [78,12,13,76,77]). In fact, the formula for the Holmes– Thompson k-area density in terms of the Fourier transform of the norm is equivalent to the following Crofton-type formula for Minkowski spaces: T HEOREM 3.8. Let (V , · ) be an n-dimensional Minkowski space and let k, 1 k n − 1, be an integer. There exists a smooth, translation-invariant, and possibly signed measure Φn−k on the manifold Hn,n−k of (n − k)-flats of V such that if N ⊂ V is an immersed k-dimensional submanifold, then volk (N ) =

1 k

#(N ∩ λ)Φn−k ,

(5)

λ∈Hn,n−k

where k is the volume of the Euclidean unit ball of dimension k. This theorem was first proved for finite-dimensional subspaces of L1 ([0, 1]) by Schneider and Wieacker [78]. In the form given above it is due to Álvarez and Fernandes [12]. It has recently been extended by Schneider [77] to generalized hypermetric spaces (i.e., finitedimensional normed spaces where the distributional Fourier transform of the norm is a signed measure).

4. Unit spheres in Minkowski spaces Surprisingly little is known about the Finsler geometry of unit spheres in Minkowski spaces. The classic results in two dimensions are the theorems of Goł¸ab and Schäffer (see [54] and [73]). T HEOREM 4.1 (Goł¸ab). The length of the unit circle of a two-dimensional normed space is greater than or equal to six and less than or equal to eight. Moreover, the lower bound is attained if and only if the unit circle is an affine regular hexagon and the upper bound is attained if and only if the unit circle is a parallelogram. T HEOREM 4.2 (Schäffer). The length of the unit circle of a two-dimensional normed space equals the length of the unit circle of its dual space.

Some problems on Finsler geometry

13

The perimeter of the unit circle is at once its surface area, twice its intrinsic diameter, the length of its shortest closed geodesic, and the length of its shortest closed, symmetric geodesic. Each of these interpretations points to a different, possible, higher-dimensional extension of the theorems of Goł¸ab and Schäffer. In this section, we shall quickly survey what is known about the higher-dimensional analogues of Goł¸ab’s theorem. The generalizations of Schäffer’s theorem have an unexpected relation with symplectic geometry and will be discussed in the next section. We start with the following upper bounds for the Hausdorff measure and the Holmes– Thompson area of unit spheres in finite-dimensional normed spaces. T HEOREM 4.3 (Busemann and Petty [41]). The (intrinsic) Hausdorff measure of the unit sphere of an n-dimensional normed space is at most 2n times the volume of the Euclidean unit ball of dimension n − 1. Equality holds if and only if the unit ball is a parallelotope. Since the Holmes–Thompson area is always less than or equal to the Hausdorff measure, we have the following corollary: C OROLLARY 4.4 (Thompson). The Holmes–Thompson area of the unit sphere in an n-dimensional normed space is at most 2n times the volume of the Euclidean unit ball of dimension n − 1. Equality holds if and only if n = 2 and the unit ball is a parallelogram. The quest of the lower bound is much more challenging and interesting. The only result in dimension greater than two is the following (unpublished) sharp lower bound by Álvarez, Ivanov and Thompson. T HEOREM 4.5. The Holmes–Thompson area of the unit sphere in a normed space of dimension three is at least 36/π . This bound is attained, for example, when the unit sphere of the normed space is the rhombic dodecahedron or the cubo-octahedron. As a corollary, we have that the Hausdorff measure of the unit sphere of a threedimensional normed space is greater than 36/π . This result is, so far, the only contribution to the following problem of Busemann and Petty [41]. P ROBLEM 4 (Busemann–Petty Problem 7). Find the sharp lower bound for the Hausdorff measure of the unit sphere of a normed space of dimension n, n 3. Of course, we have the analogous problem for the Holmes–Thompson definition of volume. P ROBLEM 5 (Thompson [84]). Find the sharp lower bound for the Holmes–Thompson area of the unit sphere of a normed space of dimension n, n > 3. Schäffer has considered the length of the shortest closed geodesic that is symmetric about the origin. This length, which Schäffer calls the girth of the normed space, is twice the length of the shortest non-contractible geodesic, the systole, for the induced Finsler metric on the projective space.

14

J.C. Álvarez Paiva

T HEOREM 4.6 (Schäffer [74]). The girth of an n-dimensional normed space is at most eight and least 4 + 2[n/2]−1 , where [·] denotes the greatest-integer function. Moreover if the girth equals eight, then the space is two-dimensional and its unit ball is a parallelogram. Here is yet another interesting question of Schäffer about the girth of normed spaces. The problem was posed as a conjecture in [74, p. 97]. P ROBLEM 6 (Schäffer [74]). Prove or disprove that girth of a Minkowski space of dimension three is at most 2π and that equality holds if and only the space is Euclidean. So far, we have only considered the shortest geodesic that is symmetric about the origin. Is it possible that on some unit sphere there is a shorter geodesic that is not symmetric? P ROBLEM 7 (Thompson). Is the shortest closed geodesic on the unit sphere of a Minkowski space symmetric with respect to the origin? We now shift our attention to a more classic metric invariant, the inner or intrinsic diameter of the unit sphere. T HEOREM 4.7 (Schäffer [74]). The (intrinsic) diameter of the unit sphere of an n-dimensional normed space is at most four and at least 2+[n/2]−1 , where [·] denotes the greatestinteger function. In particular, the diameter of the unit sphere of a three-dimensional Minkowski space is between three and four. P ROOF. In order to see that the diameter is at most four, let x and y be any two distinct points on the unit sphere and consider a plane passing through these points and the origin. The intersection of the plane with the sphere is a curve whose length, by Goł¸ab’s theorem, is at most eight. It follows that the distance between x and y is at most four. To obtain the lower bound, notice that, by Theorem 4.6, the length of the shortest closed, symmetric curve on the unit sphere is greater than or equal to 4 + 2[n/2]−1 . This implies that there is a pair of antipodal points at a distance greater than or equal to 2 + [n/2]−1 and the inequality follows. Schäffer also characterizes those normed spaces for which the diameter of the unit sphere equals four (see Theorem 9G in [74, p. 58]). P ROBLEM 8 (Schäffer [74]). Is the (inner) diameter of the unit sphere of a finitedimensional normed space attained at a pair of antipodes?

5. Symplectic equivalence of Finsler manifolds In this section we study several notions of symplectic equivalence between Finsler spaces and consider the higher-dimensional generalizations of Schäffer’s Theorem 4.2.

Some problems on Finsler geometry

15

5.1. Equivalence of unit co-disc bundles The unit co-disc bundle of a Finsler manifold is an open subset of the cotangent bundle and, as such, it carries a symplectic structure. A vaguely posed, but possibly fruitful, problem is to relate the symplectic invariants of the unit co-disc bundle to the metric invariants of the Finsler manifold. A large class of examples of Finsler manifolds with symplectomorphic unit co-disc bundles is furnished by the following result: T HEOREM 5.1 (Álvarez [8]). Let (Rn , · 1 ) and (Rn , · 2 ) be two Minkowski spaces and let S1 and S2 denote their unit spheres. The unit co-disc bundle of the Finsler metric on S1 induced by its embedding into (Rn , · 2 ) is symplectomorphic to the unit co-disc bundle of the Finsler metric on S2∗ induced by its embedding into (Rn , · ∗1 ). In particular, the unit sphere of a Minkowski space and its dual have symplectomorphic unit co-disc bundles. The following corollary—the first of our generalizations of Schäffer’s theorem—predated the theorem, and was in effect its motivation. C OROLLARY 5.2 (Holmes and Thompson [62]). The unit sphere of a normed space and that of its dual have the same Holmes–Thompson area. Besides examples of non-isometric Finsler manifolds with symplectomorphic co-disc bundles, it is interesting to look for rigidity results such as the following theorem of Benci and Sikorav (see [80] and [68, p. 365]). T HEOREM 5.3. If the unit co-disc bundles of two flat Finsler tori are symplectomorphic, then the tori are isometric.

5.2. Equivalence of unit co-sphere bundles D EFINITION 5.4. Let M and N be two Finsler manifolds with unit co-sphere bundles S ∗ M and S ∗ N and canonical 1-forms αM and αN . The Finsler manifolds M and N will be said to be exactly contactomorphic if there exists a diffeomorphism F : S ∗ M → S ∗ N and a function f on S ∗ M such that F ∗ αN = αM + df . If F ∗ αN = αM , we shall say that the metrics are α-equivalent. P ROPOSITION 5.5. If M and N are two exactly contactomorphic Finsler manifolds, then their volumes and their length spectra are equal. Moreover, if M and N are α-equivalent, then their geodesic flows are conjugate. P ROOF. To see that the volume of M equals that of N we simply use Eq. (3) expressing the Holmes–Thompson volume of the manifold in terms of the canonical 1-form. The equality of the length spectra follows from Eq. (2), which states that the action of a leaf of the geodesic foliation equals the length of the underlying geodesic. Indeed,

16

J.C. Álvarez Paiva

if F : S ∗ M → S ∗ N is a diffeomorphism satisfying F ∗ αN = αM + df , then F maps the geodesic foliation on S ∗ M to the geodesic foliation on S ∗ N . Moreover, we see that closed leaves are taken to closed leaves with the same action. In the case where F ∗ αN = αM , the expression for the geodesic spray as the Reeb vector field of the canonical 1-form immediately implies that the geodesic flows are conjugate. P ROBLEM 9. Is every (reversible) Finsler metric on the two-sphere exactly contactomorphic or α-equivalent to a Riemannian metric? A large class of examples of exactly contactomorphic Finsler manifolds is provided by the following analogue of Theorem 5.1. T HEOREM 5.6 (Álvarez [8]). Let (Rn , · 1 ) and (Rn , · 2 ) be two Minkowski spaces and let S1 and S2 denote their unit spheres. The Finsler metric on S1 induced by its embedding into (Rn , · 2 ) and the Finsler metric on S2∗ induced by its embedding into (Rn , · ∗1 ) are exactly contactomorphic. Moreover, the diffeomorphism F can be taken such that it takes centrally symmetric closed geodesics to centrally symmetric closed geodesics. As a corollary we obtain our second generalization of Schäffer’s theorem: C OROLLARY 5.7. The length of shortest closed geodesic on the unit sphere of a Minkowski space equals the length of the shortest closed geodesic on the unit sphere its dual. While the length of the shortest closed geodesic on the unit sphere of a Minkowski space is an interesting invariant, it seems very hard to prove that it is continuous with respect to any of the natural topologies in the space of convex bodies. In this respect, the girth— the infimum of the lengths of all centrally symmetric, simple, closed curves—is a much better invariant. Indeed, Schäffer proved in [74, p. 91] that the girth of a normed space is continuous with respect to the topology induced by the Banach–Mazur distance. Schäffer also conjectured that the girth of a normed space equals the girth of its dual and proved that it is enough to consider the case of finite-dimensional normed spaces. Theorem 5.6 together with Schäffer’s results settles the conjecture. T HEOREM 5.8 (Álvarez [8]). The girth of a normed space equals the girth of its dual. The preceding theorem is a third generalization of the fact that the perimeter of the unit circle of a normed plane equals the perimeter of the dual circle. Schäffer showed in [74, p. 110] that a fourth possible generalization—that the (intrinsic) diameter of the unit sphere in a normed space equals the diameter of the dual sphere—is false. In particular, the diameter of a Finsler manifold is not a symplectic invariant of its unit co-disc bundle. The notion of α-equivalence first appeared in Weinstein’s work on the volume of Riemannian manifolds all of whose geodesics are closed. The symplectic and topological nature of his proofs implies that they extend unchanged for Finsler metrics.

Some problems on Finsler geometry

17

T HEOREM 5.9 (Weinstein [86]). Let ϕt , t ∈ [0, 1], be a smooth family of Finsler metrics on a compact manifold M. If for every t the geodesics of the Finsler metric ϕt are all closed and of fixed length L, independent of t, then (M, ϕ0 ) and (M, ϕ1 ) are α-equivalent. In particular, (M, ϕ0 ) and (M, ϕ1 ) have the same volume. Weinstein’s theorem follows from the fact that the manifolds of geodesics of the metrics involved are symplectomorphic. In the next paragraph we will review the natural symplectic structure on spaces of geodesics and some of their applications to integral geometry and the minimality of submanifolds in Finsler spaces. 5.3. Manifolds of geodesics Let M be a Finsler manifold such that its space of oriented geodesics is a manifold G(M). Let S ∗ M denote its unit co-sphere bundle and let π : S ∗ M → G(M) be the canonical projection which sends a given unit covector to the geodesic that has this covector as initial condition. P ROPOSITION –D EFINITION 5.10 ([18] and [27]). Let M be a Finsler manifold with manifold of geodesics G(M) and let S∗M

i

T ∗M

π

G(M) be the canonical projection onto G(M) and the canonical inclusion into T ∗ M. If ωM is the standard symplectic form on T ∗ M, then there is a unique symplectic form ω on G(M) which satisfies the equation π ∗ ω = i ∗ ωM . At first sight there seem to be very few examples of Riemannian or Finsler manifolds whose space of geodesics is smooth. The following examples will convince the reader that this is not so. E XAMPLES . 1. Strictly convex balls and Hadamard manifolds [49]. Around any point x in a Finsler manifold there is an open geodesic ball with the property that the function that assigns to every point in the ball its distance from x is strictly convex. The space of geodesics in such a geodesic ball is a smooth manifold. As a result, the space of geodesics of any complete Riemannian metric on Rn with non-positive sectional curvature (a Hadamard manifold) is a smooth manifold. 2. Projective Finsler metrics. These are Finsler metrics on open, convex subsets of RP n such that projective line segments are geodesics. We will review their construction in the next section.

18

J.C. Álvarez Paiva

3. Zoll manifolds. These are Finsler metrics all of whose geodesics are periodic with the same minimal period. A great number of Riemannian examples has been constructed by Weinstein (see [27]). It is interesting to determine when the manifolds of geodesics of two Finsler manifolds are symplectomorphic. Here are two results in this direction: T HEOREM 5.11 (Ferrand [49]). The manifold of geodesics of a Hadamard manifold of dimension n is symplectomorphic to the cotangent bundle of the (n − 1)-dimensional sphere. T HEOREM 5.12 (Ono [70]). The manifold of geodesics of a Zoll Finsler metric on S 3 is symplectomorphic to a complex hyperquadric in CP 3 . It follows from these theorems that all Hadamard manifolds and all Zoll metrics on S 3 are α-equivalent among themselves. P ROBLEM 10. Is the manifold of geodesics of a Zoll Finsler metric on the n-sphere symplectomorphic to a complex hyperquadric in CP n ? Is the space of all Zoll Finsler metrics on the n-sphere connected? The study of the symplectic geometry of the space of geodesics has interesting applications to the integral geometry of Finsler manifolds. For example, the classical integralgeometric theorem of Cauchy and its extension to finite-dimensional normed spaces is a consequence of the following symplectic equivalence. T HEOREM 5.13 (Álvarez [8]). Let (V , · ) be a Minkowski space, and let M ⊂ V be a smooth quadratically convex hypersurface. The unit co-disc bundle for the induced Finsler metric on M and the set of all oriented lines in V which pass through the interior of M are symplectomorphic. The Crofton formula for hypersurfaces of Finsler spaces, announced by Chakerian in [43], follows easily from the co-area formula and symplectic reduction (see [9] for a proof). T HEOREM 5.14. Let M be an n-dimensional Finsler manifold with manifold of geodesics G(M). If N ⊂ M is an immersed hypersurface and if ωn−1 denotes the Liouville volume form on G(M), then voln−1 (N ) =

1 2n−1 n!

·

γ ∈G(M)

#(N ∩ γ ) ωn−1 ,

where n−1 is the volume of the Euclidean unit ball of dimension n − 1.

Some problems on Finsler geometry

19

6. Around Hilbert’s fourth problem In modern terminology, Hilbert’s fourth problem asks to construct and study all Finsler metrics on open convex subsets of RP n (including RP n itself) such that geodesics lie on projective lines. At the root of this problem are Minkowski’s work on normed spaces and the following generalization of the Cayley–Klein model of hyperbolic geometry given by Hilbert himself: Let D ⊂ Rn be an open domain bounded by a convex hypersurface C. If x and y are two distinct points on D, denote by a and b the points of intersection of C with the line determined by x and y (see Figure 1), and define the distance between these points by the equation d(x, y) :=

1 y − ax − b . ln 2 x − ay − b

(6)

T HEOREM 6.1 (Hilbert [59]). The function d is a distance function on D. Moreover, straight line segments are geodesics. The metric space (D, d) is called a Hilbert geometry. The following elegant description of the Finsler metrics that gives rise to Hilbert geometries, and which I learned from R. Ambartzumian, is apparently well known. Let D ⊂ Rn be an open domain bounded by a smooth, quadratically convex hypersurface C. Define a Finsler metric ϕ on D by setting its value at a non-zero vector vx ∈ Tx D to be ϕ(vx ) := (t1−1 + t2−1 )/2, where t1 and t2 are the two positive real numbers for which x + t1 v and x − t2 v belong to C. P ROPOSITION 6.2. If x and y are two points on D and xy is the line segment joining them, then y − ax − b 1 . ϕ = ln 2 x − ay − b xy Hamel, a student of Hilbert, was the first to study Hilbert’s fourth problem. Among other things he showed that Lagrangians on Rn whose extremals are straight lines are characterized by a system of linear partial differential equations.

Fig. 1.

20

J.C. Álvarez Paiva

T HEOREM 6.3 (Hamel [58]). Let ϕ : T Rn \ 0 → R be a smooth Lagrangian which is homogeneous of order one in the velocities. Straight lines are extremals of the functional

γ → ϕ(γ˙ (t)) dt if and only if ϕ satisfies the system of equations ∂ 2ϕ ∂ 2ϕ = ∂xi ∂vj ∂xj ∂vi

for 1 i, j n.

(7)

It is mainly through the work of Busemann and Pogorelov (see [37,72,82,15]) that the construction of projective Finsler metrics in terms of a class of (possibly signed) measures is now well understood. D EFINITION 6.4 [15,82]. Let D ⊂ RP n be an open convex set and let Hn−1 (D) be the set of all hyperplanes passing through D. A possibly-signed measure on Hn−1 (D) is said to be quasi-positive if for any two line segments xy and yz not on the same line, the measure of the set of hyperplanes intersecting twice the wedge formed by xy and yz is positive. T HEOREM 6.5. A Finsler metric ϕ on an open convex set D ⊂ RP n is projective if and only if there exists a smooth quasi-positive measure Φn−1 on the space of hyperplanes passing through D, Hn−1 (D), such that for any smooth curve γ , 1 ϕ= #(ζ ∩ γ )Φn−1 . (8) 2 ζ ∈Hn−1 (D) γ Notice that, in particular, the length of a line segment equals half the Φn−1 -measure of the set of all hyperplanes intersecting it. In Pogorelov’s approach to Hilbert’s fourth problem, Theorem 6.5 follows from the following integral representation for the solution of Hamel’s equations. T HEOREM 6.6. A Lagrangian ϕ : T Rn \ 0 → R which is homogeneous of order one in the velocities satisfies Hamel’s equations if and only if there exists a smooth even function ν(r, ξ ) on R × S n−1 such that ϕ(x, v) = |ξ · v|ν(ξ · x, ξ )Ω, (9) ξ ∈S n−1

where Ω is the standard area form on the unit sphere in Rn . E XAMPLE [12]. Applying formula (9) to the function ν : R × S 1 → R defined by ν(r, θ ) = 1 + r 2 , we obtain the Finsler metric 1 3 + x12 + x22 v12 + v22 + (x1 v1 + x2 v2 )2 . ϕ(x1 , x2 , v1 , v2 ) = 3 v12 + v22 Theorem 6.5 states that projective Finsler metrics are exactly those Finsler spaces for which there is a Crofton formula for the lengths of curves. Do the Crofton formulas for

Some problems on Finsler geometry

21

the areas of submanifolds also hold? Does the first Crofton formula imply all others? The answer is yes if by area we mean Holmes–Thompson area: T HEOREM 6.7 (Álvarez and Fernandes [12]). Let ϕ be a projective Finsler metric on an open convex domain D ⊂ RP n and let k, 1 k n − 1, be a natural number. There exists a smooth (possibly signed) measure Φn−k on the manifold Hn−k (D) of (n − k)-flats passing through D such that if N ⊂ Rn is an immersed submanifold of dimension k, then 1 volk (N ) = #(N ∩ ζ )Φn−k , (10) k ζ ∈Hn−k (D) where k is the volume of the Euclidean unit ball of dimension k. The construction of the measures Φn−k given in [12] and the formula (4) for the Holmes– Thompson volume imply that all the tangent spaces of a projective Finsler metric are hypermetric if and only if the measures Φn−k , k = 1, . . . , n − 1, are positive. In this case, just like in the case of the standard Riemannian metric on RP n , projective subspaces are area-minimizing. T HEOREM 6.8 (Álvarez and Fernandes). If the geodesics of a Finsler metric on RP n are projective lines and all its tangent spaces are hypermetric, then the projective subspaces are area-minimizing in their homology class. P ROOF. If N ⊂ RP n is a k-dimensional submanifold which is homologous to a projective subspace, then the number of points of intersection of N with a projective subspace of complementary dimension is at least one. Using the Crofton formula (10) and positivity of the measure Φn−k , we have that 1 volk (N ) = #(N ∩ ζ )Φn−k k ζ ∈Hn−k (RP n ) 1 Φn−k = volk RP k . k Hn−k (RP n ) R. Schneider has recently shown in [77] that the two previous theorems remain valid even if the regularity assumptions on both the Finsler metric and the submanifold are significantly weakened. The following results also points to the similarity between projective Finsler metrics and the standard Riemannian metric on RP n . P ROPOSITION 6.9. If (RP n , ϕ) is a projective Finsler space for which the length of the projective lines is equal to π , then the Holmes–Thompson volume of (RP n , ϕ) equals the volume of RP n with its standard Riemannian metric. P ROOF. If ϕ0 denotes the standard Riemannian metric on RP n , then, for each number t, t ∈ [0, 1], the metric ϕt = (1 − t)ϕ0 + tϕ is a projective Finsler metric and the length of

22

J.C. Álvarez Paiva

its closed geodesics is π . Applying Theorem 5.9, we conclude that the Holmes–Thompson volumes of (RP n , ϕ0 ) and (RP n , ϕ) are equal. It seems that the analogues of Hilbert’s fourth problem for rank-one symmetric spaces other than RP n have never been studied. For example, the following problem is open: P ROBLEM 11. Construct all Finsler metrics on CP 2 such that the geodesics coincide as point sets with those of the standard Riemannian metric on CP 2 . In trying to solve this problem, Álvarez and Durán stumbled upon the following construction of Finsler metrics on complex and quaternionic projective spaces for which the projective lines are totally geodesic and the geodesics are circles: Use the Busemann–Pogorelov construction of projective metrics on real projective spaces and spheres to construct a projective metric ϕ on S 2n+1 (respectively S 4n+3 ) that is invariant under the Hopf action. There is a unique Finsler metric ψ on CP n (respectively HP n ) for which the projection map of the Hopf fibration is an isometric submersion from (S 2n+1 , ϕ) to (CP n , ψ) (respectively from (S 4n+3 , ϕ) to (HP n , ψ)). For the metric ψ , projective lines are totally geodesic and geodesics are (geometric) circles. This construction suggests yet another inverse problem: P ROBLEM 12. Construct all Finsler metrics on CP n such that the geodesics are circles. For n = 1 this problem has been solved by Álvarez and Berck (unpublished) in terms of space-type surfaces (also called elliptic congruences) on the Grassmannian of Lagrangian planes in R4 . There have been many attempts to define the Finsler analogue of Kähler metrics, but none seems to have enjoyed any measure of success. Since the metric properties of Kähler manifolds are not so well understood as to proceed in this direction, it makes sense to use some remarkable geometric property of Kähler manifolds in an attempt to define their Finsler analogues. One such remarkable geometric property is that complex submanifolds are minimal. P ROBLEM 13. Construct and study all the Finsler metrics on CP n for which complex submanifolds are minimal. Are (Riemannian) Kähler metrics the only ones with this property? 7. Closed geodesics In Riemannian geometry, the study of closed geodesics has a long and glorious history with contributors like Poincaré, Birkhoff and Morse. However, there do not seem to be many interesting results about closed geodesics in Finsler manifolds. If an existence result in Riemannian geometry depends only on Morse theory and the method of broken geodesics, then it automatically holds in Finsler geometry. A Riemannian result that would be interesting to extend to the Finsler setting is a theorem of Bangert and Franks [19,53] stating that any Riemannian metric on the twodimensional sphere has infinitely many closed geodesics. Here we emphasize that we

Some problems on Finsler geometry

23

are considering symmetric or reversible Finsler metrics. Indeed, Katok has constructed non-reversible Finsler metrics on the 2-sphere which have only two closed geodesics (see [65,87]). P ROBLEM 14 (Bangert). Does every Finsler metric on S 2 have infinitely many closed geodesics? Franks’ work on the periodic points of area preserving maps of the annulus and recent work by Hofer, Wysocki and Zehnder (see [61]) suggest the following question: P ROBLEM 15. Is the number of distinct closed geodesics on a non-reversible Finsler metric on the 2-sphere either two or infinity?

8. Differential invariants of Finsler surfaces In this section we study the differential invariants of Finsler surfaces without the aid of the Cartan connection and then introduce Cartan’s structure equations in order to uncover the relations between these invariants. 8.1. Convex geometry and the invariant I A smooth, centrally symmetric, and quadratically convex curve S on a two-dimensional vector space V parameterizes a family of Euclidean structures on V . Indeed, for each point v ∈ S, there is a unique ellipse Ev which is centered at the origin and osculates S up to second order at this point. We associate to v the Euclidean structure with Ev as unit circle. D EFINITION 8.1. Let (V , ϕ) be a two-dimensional Minkowski space with unit circle S. An orthonormal basis of V is an ordered pair of vectors (v, v⊥ ), where v ∈ S and v⊥ is both of unit length and perpendicular to v with respect to the Euclidean structure associated to v. Using the Euclidean structures associated to the curve we may define a distinguished parameterization of S: orient the vector space V and parameterize the curve S by a map γ in such a way that γ (t), γ˙ (t) is an oriented orthonormal basis. Tabachnikov has remarked (private communication) that this parameterization is, up to shifts in the parameter, the only one satisfying the equation det γ (t), γ˙ (t) = det γ˙ (t), γ¨ (t) . The period of γ is a linear invariant of S which we shall call the total angle of S. T HEOREM 8.2 (Schneider [75]). The total angle of a smooth, centrally-symmetric, and quadratically convex curve on the plane is less than or equal to 2π . Equality holds if and only if the curve is an ellipse.

24

J.C. Álvarez Paiva

D EFINITION 8.3. Let S be a smooth, centrally symmetric, and quadratically convex curve on a two-dimensional vector space V . If γ is a distinguished parameterization of S and v = γ (t) is a point on S, then I (v) is defined by the equation γ¨ (t) := −γ (t) + I (v)γ˙ (t). The quantity I (v) is zero if and only if the osculating ellipse E(v) osculates S up to third order or higher at v. It follows that if I is identically zero, then S is an ellipse. Using a theorem of Ghys of the zeros of the Schwartzian derivative (see [71]), Álvarez has showed in [10] that the invariant I vanishes at least eight times. If (M, ϕ) is a Finsler surface, then every tangent space Tm M, m ∈ M, is a Minkowski space and the indicatrix Sm M := vm ∈ Tm M: ϕ(vm ) = 1 is smooth, centrally symmetric, and quadratically convex. For each unit tangent vector vm we define I (vm ) as the value of the invariant I of Sm M at the point vm . The result is a smooth function, which we again denote by I , defined of the unit circle bundle of M. Clearly, this function is identically zero if and only if the Finsler surface is Riemannian. We can also use the previous geometric constructions to define a vector field on the unit circle bundle of an oriented Finsler surface (M, ϕ): If vm is a unit tangent vector and γ (t) is a distinguished parameterization of Sm M with γ (0) = vm , then X3 (vm ) := γ˙ (0). 8.2. The invariant J We shall now define a smooth function on the unit bundle of M which measures how the invariant I changes along the geodesics. D EFINITION 8.4. Let (M, ϕ) be a Finsler surfaces, let vm ∈ Tm M be a unit tangent vector, and let σ : (−, ) → M be the geodesic with initial condition vm . Define

J (vm ) = d/dtI σ˙ (t) t=0 . Note that J = 0 means that I is an invariant of motion. The Finsler surfaces for which this occurs are called Landsberg surfaces. Originally, they came up in the study of Finslerian analogues of the Gauss–Bonnet theorem (see [66] and Section 8.5). Unfortunately, all the known examples of Landsberg surfaces are either Riemannian or locally isometric to Minkowski planes. For example, the following problem remains unsolved. P ROBLEM 16. Is there any non-Riemannian Landsberg metric on the 2-sphere? 8.3. Curvature of Finsler surfaces The formula of second variation really belongs to variational calculus and not to Riemannian geometry. It should not come then as a surprise that many of the definitions and

Some problems on Finsler geometry

25

theorems in Riemannian geometry, including the notion of curvature, extend to Finsler geometry. Let us start by defining the obvious extensions of Jacobi fields and conjugate points: Given a geodesic γ : [a, b] → M, a geodesic variation of γ is a smooth map Γ : (−, )× [a, b] → M such that • Γ (0, t) = γ (t). • For each fixed s0 , the curve t → Γ (s0 , t) is a geodesic. If Γ (s, t) is a geodesic variation of the geodesic γ , the vector field Y along γ defined by

∂Γ (s, t)

Y (t) = ∂s s=0 is called a Jacobi field. A Jacobi field is said to be proper if, for each t, γ˙ (t) and Y (t) form an orthonormal basis of the Minkowski plane Tm M in the sense of Definition 8.1. D EFINITION 8.5. Let p be a point in a Finsler surface and let γ be a geodesic starting from p. A point γ (s) is said to be conjugate to p along γ if there exists a non-zero Jacobi field Y (t) along γ such that Y (0) = Y (s) = 0. Just as in Riemannian geometry, geodesics minimize length up to their first conjugate point. P ROPOSITION 8.6. Let p be a point in a Finsler surface and let γ be a geodesic starting from p. The curve γ restricted to the interval [0, s] minimizes length if and only if there is no conjugate point between p = γ (0) and γ (s). If we are given a geodesic γ let us orient the tangent spaces Tγ (t) M and let us define γ˙ ⊥ (t) in such a way that γ˙ (t) and γ˙ ⊥ (t) form an oriented orthonormal basis on Tγ (t) M. Note that any proper Jacobi field Y along γ can be uniquely expressed as Y (t) = y(t)γ˙ ⊥ (t), where y is a real-valued function. P ROPOSITION –D EFINITION 8.7. There is a unique smooth function K on the unit circle bundle of M such that for any geodesic γ parameterized with unit speed and for any proper Jacobi field Y (t) = y(t)γ˙ ⊥ (t) the Jacobi equation y (t) + K γ˙ (t) y(t) = 0 holds. The function K is called the curvature of (M, ϕ). In contrast with the Riemannian case, K depends on both the point γ (t) ∈ M and the direction of γ˙ (t).

26

J.C. Álvarez Paiva

The Bonnet–Myers theorem follows, as usual, from the Jacobi equation and Sturm’s comparison theorem. T HEOREM 8.8. If the curvature function of a Finsler surface (M, ϕ) is greater √ than or equal to a positive number δ, then the diameter of M is less than or equal to π/ δ. A very helpful way of describing the Finsler curvature in terms of auxiliary Riemannian metrics has been given by Shen [79]: Let vm ∈ Tm M be a unit vector and let X be a geodesic vector field (i.e., the integral curves of X are geodesics parameterized with unit speed) defined on a neighborhood O of m and such that X(m) = vm . If we define a Riemannian metric on O by x → gϕ (X(x)) as in Section 2, then the Riemannian curvature of this metric at m equals the Finsler curvature of (M, ϕ) at vm . 8.4. Cartan’s structure equations By now we have defined three geometric invariants of Finsler surfaces: I , J and K. The invariant I is a centro-affine invariant which describes the shape of each unit tangent circle, the invariant K belongs to the calculus of variations and measures the focusing of geodesics, and the invariant J , by measuring how I varies along geodesics, partakes of both convex geometry and variational calculus. All three invariants can be defined, as we have done, by elementary geometric and variational considerations, but there is nothing to suggest the deep and interesting relations between the three. D EFINITION 8.9. If (M, ϕ) is a Finsler manifold, the geodesic spray of M is the vector field X1 defined on the unit tangent bundle and whose value at a unit tangent vector vm is defined as follows: take σ (t) to be the geodesic with initial condition vm and set X1 (vm ) := d/dt σ˙ (t)|t=0 . T HEOREM 8.10 (Cartan [42]). Let (M, ϕ) be an oriented Finsler surface, let X1 denote its geodesic spray, and let X3 be the vector field defined at the end of Section 8.1. If we define X2 := [X3 , X1 ], then we have the following equations: [X3 , X1 ] = X2 , [X1 , X2 ] = KX3 , [X3 , X2 ] = −X1 + I X2 + J X3 . Of course, Cartan preferred differential forms to vector fields and he wrote the above equations in terms of the dual forms ω1 , ω2 and ω3 defined by the equations ωi (Xj ) = δij . Cartan’s structure equations are: dω1 = −ω2 ∧ ω3 , dω2 = ω1 ∧ ω3 − I ω2 ∧ ω3 , dω3 = −Kω1 ∧ ω2 − J ω2 ∧ ω3 .

Some problems on Finsler geometry

27

Note that by differentiating the structure equations we obtain the following Bianchi identities: J = I1 , K3 + KI + J1 = 0. In these identities the subindices represent differentiation with respect to the vector fields X1 , X2 and X3 . In general if F is a function on the unit bundle we will write dF = F1 ω1 + F2 ω2 + F3 ω3 . Cartan shows that the forms ω1 , ω2 , and ω3 solve the problem of equivalence. From this it follows, at least in theory, that all microlocal invariant properties of Finsler surfaces can be written in terms of the functions I , J , K, and their derivatives with respect to the vector fields X1 , X2 , and X3 . As an example, we have Berwald’s characterizations of locally Minkowski and projectively flat Finsler surfaces. T HEOREM 8.11 (Berwald). A Finsler surface is locally Minkowski if and only if K, I1 , and I2 are identically zero. Let us recall that a Finsler manifold is said to be projectively flat if around every point we can find a small neighborhood and a diffeomorphism of this neighborhood to an open subset of Euclidean space such that geodesics are mapped onto straight lines. T HEOREM 8.12 (Berwald [26]). The (non-parameterized) geodesics of a Finsler surface are locally the geodesics of some affine connection if and only if the following equation holds: I23 + J33 + 2I (I2 + J3 ) + 6J = 0. The dual system of curves to the geodesics of a Finsler surface are locally the geodesics of some affine connection if and only if the following equation holds: K31 − 3K2 = 0. Moreover, both of the above equations hold if and only if the surface is projectively flat. For dual systems of curves and an elementary exposition of path geometry see Arnold’s book [17, pp. 42–56]. A very clear exposition of Berwald’s theorem and its proof is given in [31]. It is amusing to prove the following classic theorem of Beltrami by using Berwald’s result. C OROLLARY 8.13 (Beltrami). A Riemannian surface is projectively flat if and only if its curvature is constant.

28

J.C. Álvarez Paiva

Berwald was also interested in Finsler metrics whose geodesics coincide as parameterized curves with the geodesics of some affine connection. These Finsler metrics are called Berwald metrics. In two dimensions they are characterized by the equations I1 = 0 and I2 = 0. 8.5. Applications of Cartan’s structure equations Armed with Cartan’s structure equations and their Bianchi identities, we are now in a position to prove some of the deepest results in the theory of Finsler surfaces. T HEOREM 8.14 (Akbar–Zadeh [4]). A compact Finsler surface of constant negative curvature is Riemannian. P ROOF. Using the second Bianchi identity, which tells us that K3 + KI + J1 = 0, we have that if K ≡ c is a constant and σ (t) is a geodesic on M, then the function I (t) := I (σ˙ (t)) satisfies the differential equation d2 I = −cI. dt 2 If c is negative, then I (t) must be a linear combination of exponentials and, therefore, if the initial condition is not I (0) = 0, I (0) = 0, the function I (t) is unbounded. Since I is bounded whenever the Finsler surface is compact, the only possibility that remains is that I be identically zero, and that the surface be Riemannian. The classical non-Riemannian examples of Finsler metrics with constant negative curvature are the Hilbert geometries. Since the existence of an isometric embedding of a Finsler surface on a Minkowski space implies that the invariant I is bounded (this follows immediately from the geometric interpretation of I given in Section 8.1) we have the following remark of Álvarez and Durán [10]: the Hilbert geometry given by a smooth, quadratically convex curve C does not admit an isometric embedding into a Minkowski space unless C is an ellipse. If C is an ellipse, then the Hilbert geometry is the Cayley–Klein model of hyperbolic geometry and, by a theorem of Rosendorn, it admits an explicit isometric embedding into R5 (see [57, p. 276]). T HEOREM 8.15 (Akbar–Zadeh [4]). A compact Finsler surface with zero curvature is locally isometric to a Minkowski plane. P ROOF. By Theorem 8.11, we need to show that a compact Finsler surface of zero curvature also satisfies I1 ≡ 0 and I2 ≡ 0. Reasoning as in the proof of the previous theorem, if σ (t) is a geodesic on M, then the function I (t) := I (σ˙ (t)) satisfies the differential equation d2 I = 0. dt 2

Some problems on Finsler geometry

29

This implies that I is a linear function of t. Since I is bounded, it must be a constant and, therefore, I1 = 0. The proof that I2 = 0 is just slightly more involved. First note from the structure equations that when K ≡ 0 the vector fields X1 and X2 commute. From this we gather that I21 = I12 = 0, and so I2 is constant along geodesics. Using this we have that I2 = X2 I = [X3 , X1 ]I = −I31 . This implies that 0 = I21 = −I311 and, hence, I3 is a linear function of t. Since I3 is bounded, I3 must be constant on geodesics and I2 = −I31 = 0. This proof also shows that a surface with a complete Finsler metric of zero curvature that is isometrically embedded in a Minkowski space must be locally Minkowski. P ROBLEM 17. Is there any complete Finsler metric on R2 with zero curvature that is not locally Minkowski? T HEOREM 8.16. Let (M, ϕ) be a Landsberg surface. If M is connected, then the total angle of any two of its tangent unit circles is the same. The proof is taken from Bryant’s beautiful paper [30]. P ROOF. If x and y be two points on M, the difference of the total angle of the unit circle Sy over y and the total angle of the unit circle Sx over x is given by

ω3 −

Sy

ω3 , Sx

where the orientation over Sy and Sx is taken so that the integrals are positive. Let γ : [0, 1] → M be a smooth curve joining them. Let π : U M → M denote the natural projection and set C be the cylinder π −1 (γ ). Note that the oriented boundary of C is Sy −Sx and that the 2-form ω1 ∧ ω2 vanishes identically on C. Using Stokes theorem and the structure equations we have that

ω3 −

Sy

ω3 =

Sx

C

dω3 =

C

−Kω1 ∧ ω2 − J ω2 ∧ ω3 = 0.

Another application that uses the full power of Cartan’s structure equations is the Finsler version of the Gauss–Bonnet theorem given by Bao and Chern in [21]. Let (M, ϕ) be a compact, oriented Finsler surface and let X be a vector field on M with a finite number of non-degenerate zeros. Cut out small discs, say of radius r, around the zeros of X and denote the resulting manifold with boundary by Mr . Normalizing the vector field X on Mr we obtain a section σr : Mr → SMr over the unit circle bundle of Mr .

30

J.C. Álvarez Paiva

Using the fact that the total angle of the tangent unit circle at a point x is given by the integral of ω3 over Sx M, the equation dω3 = −Kω1 ∧ ω2 − J ω2 ∧ ω3 ,

(11)

and Stokes’ theorem, Chern and Bao arrive at the following result: T HEOREM 8.17 [21]. Let (M, ϕ) be a compact, oriented Finsler surface and let X be a vector field on M with zeros x1 , . . . , xn which are non-degenerate and with indices I(x1 ), . . . , I(xn ). Using the notation above, the limit as r tends to zero of the quantity Mr

σr∗ (−Kω1 ∧ ω2 − J ω2 ∧ ω3 )

(12)

is well defined and equals ni=1 I(xi )A(xi ), where A(x1 ), . . . , A(xn ) are the total angles of the unit tangent circles at the points x1 , . . . , xn . If, as in the case of Landsberg surfaces, the total angle of the unit tangent circles does not vary from point to point, then we have that the Euler characteristic of M can be written as an integral in terms of the differential invariants of the Finsler surface. Unfortunately, the Gauss–Bonnet theorem for Finsler surfaces does not have as many geometric implications as its Riemannian counterpart. For example, it cannot be used to prove that a metric of non-positive, or non-negative, curvature on a two-dimensional Finsler torus must be flat, or that two simple closed geodesics in a positively curved Finsler twosphere must intersect. The reader is invited to prove these seemingly simple results by him/herself and thereby gain some insight into some of the difficulties of extending Riemannian results to the Finsler setting. A greater conceptual challenge is that the standard Riemannian technique of comparing arbitrary metrics to metrics of constant curvature does not generalize to the Finsler setting. As we saw earlier in this section, a compact Finsler surface with constant negative curvature is Riemannian. Likewise, it has been recently remarked by Bryant that the main result of LeBrun and Mason, in [67], implies the following important result: T HEOREM 8.18 (Bryant). A Finsler metric of constant positive curvature on the twosphere is Riemannian. The analogous result in higher dimensions is still open: P ROBLEM 18 (Bryant). Is there any non-Riemannian Finsler metric on S n , n > 2, with constant (positive) curvature? It is possible that we must instead compare arbitrary Finsler metrics to metrics that have a given, simple, dynamical property. For example, there is evidence that the following generalization of the uniformization theorem is true:

Some problems on Finsler geometry

31

C ONJECTURE (Álvarez [6]). If ϕ is a Finsler metric on RP 2 , there exists a smooth function ρ on RP 2 such that all the geodesics of (RP 2 , eρ ϕ) are closed.

References [1] M. Abate and G. Patrizio, Finsler Metrics—A Global Approach, Lecture Notes in Math., vol. 1591, Springer-Verlag (1994). [2] R. Abraham and J.E. Marsden, Foundations of Mechanics, 2nd ed., Benjamin/Cummings, Reading, MA (1978). [3] H. Akbar-Zadeh, Les espaces de Finsler et certaines de leurs généralisations, Ann. Sci. École Norm. Sup. (3) 80 (1963), 1–79. [4] H. Akbar-Zadeh, Sur les espaces de Finsler à courbures sectionelles constantes, Acad. Roy. Belg. Bull. Cl. Sci. (5) 74 (1988), 281–322. [5] J.C. Álvarez Paiva, The symplectic geometry of spaces of geodesics, Ph.D. thesis, Rutgers University (1995). [6] J.C. Álvarez Paiva, Dual mixed volumes and isosystolic inequalities, preprint (2002). [7] J.C. Álvarez Paiva, Hilbert’s fourth problem in two dimensions, Mass Selecta: Teaching and Learning Advanced Undergraduate Mathematics, Amer. Math. Soc., Providence, RI (2003), 165–183. [8] J.C. Álvarez Paiva, Dual spheres have the same girth, Amer. J. Math., to appear. [9] J.C. Álvarez Paiva and G. Berck, What is wrong with the Hausdorff measure in Finsler spaces, Adv. in Math., to appear. [10] J.C. Álvarez Paiva and C. Durán, An Introduction to Finsler Geometry, Notas de la Escuela Venezolana de Matématicas (1998). [11] J.C. Álvarez Paiva and C. Durán, Isometric submersions of Finsler manifolds, Proc. Amer. Math. Soc. 129 (2001), 2409–2417. [12] J.C. Álvarez Paiva and E. Fernandes, Crofton formulas in projective Finsler spaces, Electron. Res. Announc. Amer. Math. Soc. 4 (1998), 91–100. [13] J.C. Álvarez Paiva and E. Fernandes, Fourier transforms and the Holmes–Thompson volume of Finsler manifolds, Int. Math. Res. Notices 19 (1999), 1031–1042. [14] J.C. Álvarez Paiva and E. Fernandes, Crofton formulas and Gelfand transforms, Selecta Math. (NS), to appear. [15] J.C. Álvarez Paiva, I.M. Gelfand and M. Smirnov, Crofton densities, symplectic geometry, and Hilbert’s fourth problem, Arnold–Gelfand Mathematical Seminars, Geometry and Singularity Theory, V.I. Arnold, I.M. Gelfand and V.S. Retakh, eds, Birkhäuser, Boston (1997), 77–92. [16] J.C. Álvarez and A.C. Thompson, Volumes in normed and Finsler spaces, A Sampler in Riemann–Finsler Geometry, D. Bao, R. Bryant, S.S. Chern and Z. Shen, eds, Cambridge Univ. Press (2004), 1–49. [17] V.I. Arnold, Geometrical Methods in the Theory of Ordinary Differential Equations, Grundlehren der Mathematischen Wissenschaften, vol. 250, Springer-Verlag, Berlin (1983). [18] V.I. Arnold and A.B. Givental, Symplectic geometry, Dynamical Systems IV, Encyclopaedia of Mathematical Sciences, vol. 4, V.I. Arnold and S.P. Novikov, eds, Springer-Verlag, Berlin (1990). [19] V. Bangert, On the existence of closed geodesics on the two-spheres, Internat. J. Math. 4 (1993), 403–418. [20] D. Bao, R. Bryant, S.S. Chern and Z. Shen, eds, Some Perspectives in Finsler Geometry, MSRI Series, Cambridge Univ. Press, in press. [21] D. Bao and S.S. Chern, A note on the Gauss–Bonnet theorem for Finsler spaces, Ann. of Math. 143 (1996), 233–252. [22] D. Bao, S.S. Chern and Z. Shen, eds, Finsler Geometry, Contemporary Mathematics, vol. 196, Amer. Math. Soc. (1996). [23] D. Bao, S.S. Chern and Z. Shen, An Introduction to Riemann–Finsler Geometry, Graduate Texts in Mathematics, vol. 200, Springer-Verlag, Berlin (2000). [24] V.N. Berestovskii, Submetries of space forms of nonnegative curvature, Siberian Math. J. 16 (1975), 651– 662. [25] G. Berck, Minimalité des sous-variétés totalement géodésiques, preprint (2004).

32

J.C. Álvarez Paiva

[26] L. Berwald, On Finsler and Cartan geometries. III. Two-dimensional Finsler spaces with rectilinear extremals, Ann. of Math. (2) 42 (1941), 84–112. [27] A. Besse, Manifolds of all Whose Geodesics are Closed, Springer-Verlag, Berlin (1978). [28] G.D. Birkhoff, Dynamical systems with two degrees of freedom, Trans. Amer. Math. Soc. 18 (1917), 199– 300. [29] E.D. Bolker, A class of convex bodies, Trans. Amer. Math. Soc. 145 (1969), 323–345. [30] R.L. Bryant, Finsler structures on the 2-sphere satisfying K = 1, in [22], 27–42. [31] R.L. Bryant, Projectively flat Finsler 2-spheres of constant curvature, Selecta Math. (N.S.) 3 (1997), 161– 203. [32] D. Burago and S. Ivanov, Isometric embeddings of Finsler manifolds, St. Petersburg Math. J. 5 (1994), 159–169. [33] D. Burago and S. Ivanov, On asymptotic volume of Finsler tori, minimal surfaces in normed spaces, and symplectic filling volume, Ann. of Math. (2) 156 (3) (2002), 891–914. [34] H. Busemann, A theorem on convex bodies of the Brunn–Minkowski type, Proc. Nat. Acad. Sci. USA 35 (1949), 27–31. [35] H. Busemann, The geometry of Finsler spaces, Bull. Amer. Math. Soc. 56 (1950), 5–16. [36] H. Busemann, The Geometry of Geodesics, Pure and Applied Mathematics, vol. 6, Academic Press, New York (1955). [37] H. Busemann, Geometries in which the planes minimize area, Ann. Mat. Pura Appl. (4) 55 (1961), 171–190. [38] H. Busemann, Convexity on Grassmann manifolds, Enseign. Math. (2) 7 (1961), 139–152. [39] H. Busemann, Recent Synthetic Differential Geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 54, Springer-Verlag, Berlin (1970). [40] H. Busemann, G. Ewald and G.C. Shephard, Convex bodies and convexity on Grassmann cones: I–IV, Math. Ann. 151 (1963), 1–41. [41] H. Busemann and C.M. Petty, Problems on convex bodies, Math. Scand. 4 (1956), 88–94. [42] E. Cartan, Sur un problème d’équivalence et la théorie des espaces métriques généralisés, Mathematica 4 (1930), 114–136. [43] G.D. Chakerian, Integral geometry in Minkowski spaces, in [22], 43–50. [44] S.S. Chern, Finsler geometry is just Riemannian geometry without the quadratic restriction, Notices Amer. Math. Soc. 43 (9) (1996), 959–963. [45] C.E. Durán, A volume comparison theorem for Finsler manifolds, Proc. Am. Math. Soc. 126 (1998), 3079– 3082. [46] D. Egloff, Some new developments in Finsler geometry, Inaugural-Dissertation, Universität Freiburg (Schweiz) (1995). [47] D. Egloff, Uniform Finsler Hadamard manifolds, Ann. Inst. H. Poincaré Phys. Théor. 66 (1997), 323–357. [48] D. Egloff, On the dynamics of uniform Finsler manifolds of negative flag curvature, Ann. Global Anal. Geom. 15 (1997), 101–116. [49] E. Ferrand, Sur la structure symplectique de l’espace des géodésiques d’un espace de Hadamard, Geom. Dedicata 68 (1997), 79–89. [50] P. Foulon, Géométrie des équations différentielles du second ordre, Ann. Inst. H. Poincaré Phys. Théor. 45 (1986), 1–28. [51] P. Foulon, Locally symmetric Finsler spaces in negative curvature, C. R. Acad. Sci. Paris 324 (1997), 1127– 1132. [52] P. Foulon, Curvature and global rigidity in Finsler manifolds, Houston J. Math. 28 (2002), 263–292. [53] J. Franks, Geodesics on S 2 and periodic points of annulus homeomorphisms, Invent. Math. 108 (1992), 403–418. [54] S. Goł¸ab, Quelques problèmes métriques de la géometrie de Minkowski, Trav. l’Acad. Mines Cracovie 6 (1932), 1–79. [55] P. Goodey and W. Weil, Zonoids and generalisations, Handbook of Convex Geometry, vol. B, P.M. Gruber et al., eds, North-Holland, Amsterdam (1993), 1297–1326. [56] M. Gromov, Filling Riemannian manifolds, J. Differential Geom. 18 (1983), 1–147. [57] M. Gromov, Partial Differential Relations, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 9, Springer-Verlag, Berlin (1986). [58] G. Hamel, Über die Geometrien, in denen die Geraden die kürzesten sind, Math. Ann. 57 (1903), 231–264.

Some problems on Finsler geometry

33

[59] D. Hilbert, Foundations of Geometry, Open Court Classics, Lasalle, IL (1971). [60] D. Hilbert, Mathematical problems, translation in Mathematical developments arising from Hilbert problems, Proceedings of Symposia in Pure Math., vol. XXVII, Part 1, F. Browder, ed., Amer. Math. Soc., Providence, RI (1974). [61] H. Hofer, K. Wysocki and E. Zehnder, The dynamics on three-dimensional strictly convex energy surfaces, Ann. of Math. 148 (1998), 197–289. [62] R.D. Holmes and A.C. Thompson, N -dimensional area and content in Minkowski spaces, Pacific J. Math. 85 (1979), 77–110. [63] L. Hörmander, The Analysis of Linear Partial Differential Operators I, Grundlehren der Mathematischen Wissenschaften, vol. 256, Springer-Verlag, Berlin (1983). [64] S. Ivanov, On two-dimensional minimal fillings, St. Petersburg Math. J. 13 (2002), 17–25. [65] A. Katok, Ergodic perturbations of degenerate integrable Hamiltonian systems (in English); Russian original in Math. USSR Izv. (1974), 535–571. [66] G. Landsberg, Über die Totalkrümmung, Jahresber. Deutsch. Math.-Verein. 16 (1907), 36–46. [67] C. LeBrun and L.J. Mason, Zoll manifolds and complex surfaces, J. Differential Geom. 61 (2002), 453–535. [68] D. McDuff and D. Salamon, Introduction to Symplectic Topology, 2nd. ed., Oxford Mathematical Monographs, Clarendon Press, Oxford (1998). [69] H. Minkowski, Gesammelte Abhandlungen, Chelsea, New York (1967). [70] K. Ono, Space of geodesics on Zoll three-spheres, preprint (1999). [71] V. Ovsienko and S. Tabachnikov, Sturm theory, Ghys theorem on zeros of the Schwartzian derivative, and flattening of Legendrian curves, Selecta Math. (N.S.) 2 (1996), 297–307. [72] A.V. Pogorelov, Hilbert’s Fourth Problem, Scripta Series in Mathematics, Winston and Sons (1979). [73] J.J. Schäffer, The self-circumference of polar convex bodies, Arch. Math. 24 (1973), 87–90. [74] J.J. Schäffer, Geometry of Spheres in Normed Spaces, Lecture Notes in Pure and Applied Mathematics, vol. 20, Dekker, New York (1976). [75] R. Schneider, Über die Finslerräume mit Sij kl = 0, Arch. Math. 19 (1968), 656–658. [76] R. Schneider, Crofton formulas in hypermetric projective Finsler spaces, Festschrift: Erich Lamprecht, Arch. Math. (Basel) 77 (1) (2001), 85–97. [77] R. Schneider, On integral geometry in projective Finsler spaces, Izv. Nats. Akad. Nauk Armenii Mat. 37 (2002) 34–51. [78] R. Schneider and J.A. Wieacker, Integral geometry in Minkowski spaces, Adv. in Math. 129 (1997), 222– 260. [79] Z. Shen, Curvature, distance, and volume in Finsler geometry, IHES preprint (1997). [80] J.-C. Sikorav, Rigidité symplectique dans le cotangent de T n , Duke Math. J. 59 (1989), 759–763. [81] E. Socié-Méthou, Comportements asymptotiques et rigidités en géométries de Hilbert, Thèse, Université de Strasbourg (2000). [82] Z.I. Szabo, Hilbert’s fourth problem, I, Adv. in Math. 59 (1986), 185–301. [83] A.C. Thompson, Applications of various inequalities to Minkowski geometry, Geom. Dedicata 46 (1993), 215–231. [84] A.C. Thompson, Minkowski Geometry, Encyclopedia of Math. and Its Applications, vol. 63, Cambridge Univ. Press, Cambridge (1996). [85] W. Weil, Centrally symmetric convex bodies and distributions II, Israel J. Math. 32 (1979), 173–182. [86] A. Weinstein, Fourier integral operators, quantization, and the spectra of Riemannian manifolds, Géométrie Symplectique et Physique Mathématique, Editions CNRS, Paris (1975). [87] W. Ziller, Geometry of the Katok examples, Ergodic Theory Dynam. Systems 3 (1982), 135–157.

This page intentionally left blank

CHAPTER 2

Foliations∗ Raymond Barre and Aziz El Kacimi Alaoui LAMATH, Université de Valenciennes, Le Mont Houy, 59313 Valenciennes Cedex 9, France E-mail: {raymond.barre, aziz.elkacimi}@univ-valenciennes.fr

Contents 0. Foreword . . . . . . . . . . . . . . . . . . . . . . . 1. Definitions and examples . . . . . . . . . . . . . . . 1.1. Morphisms of foliations . . . . . . . . . . . . 1.2. The concept of holonomy . . . . . . . . . . . . 1.3. Examples of foliations . . . . . . . . . . . . . 1.4. Foliations and differential systems . . . . . . . 1.5. Notations . . . . . . . . . . . . . . . . . . . . . 2. Transverse structures . . . . . . . . . . . . . . . . . 2.1. Measurable foliations . . . . . . . . . . . . . . 2.2. Lie foliations . . . . . . . . . . . . . . . . . . 2.3. Transversely parallelizable foliations . . . . . 2.4. Riemannian foliations . . . . . . . . . . . . . . 2.5. Transversely holomorphic foliations . . . . . . 3. Codimension one foliations . . . . . . . . . . . . . 4. Γ -structures . . . . . . . . . . . . . . . . . . . . . . 5. The leaf space . . . . . . . . . . . . . . . . . . . . . 5.1. V -manifolds . . . . . . . . . . . . . . . . . . . 5.2. QF -manifolds . . . . . . . . . . . . . . . . . . 5.3. Q-manifolds . . . . . . . . . . . . . . . . . . . 6. Characteristic classes . . . . . . . . . . . . . . . . . 6.1. The classifying space and the universal bundle 6.2. Classification of real vector bundles . . . . . . 7. Basic global analysis . . . . . . . . . . . . . . . . . 7.1. Foliated vector bundles and basic sections . . . 7.2. Transversely elliptic operators . . . . . . . . . 7.3. Transversely elliptic complexes . . . . . . . . 8. Deformation theory of foliations . . . . . . . . . . . 9. Some other themes . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

∗ We thank Cédric Rousseau for drawing all the pictures in this paper.

HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 35

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

37 37 39 39 40 45 46 46 46 47 48 48 49 50 51 52 52 53 53 54 54 54 55 55 56 58 61 63

36 9.1. Compact leaves . . . . . . 9.2. When is a manifold a leaf ? 9.3. Minimal leaves . . . . . . References . . . . . . . . . . . . .

R. Barre and A. El Kacimi Alaoui . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

64 64 65 66

Foliations

37

0. Foreword Foliation Theory is the qualitative study of Differential Equations. It was initiated by the works of H. Poincaré, I. Bendixon and developed later by C. Ehresmann, G. Reeb and many other people. Since then the subject has been a wide field in mathematical research. Actually it is almost impossible to describe all the results and the different steps of its development. So the purpose of this chapter is to give definitions, some examples and the fundamental concepts like holonomy, transverse structures, etc. Some themes in the point of view of Differential Geometry are discussed: characteristic classes, basic Hodge theory, deformations, etc. A complete account on Foliation Theory can be found in the book [135] by C. Godbillon. The bibliography is not complete. It is motivated by two reasons: the first one is to indicate references for the reader who wants to learn much more on foliations; the second one is to mention people who make contributions to the subject; for most of them the selected list is nonexhaustive. All foliations considered are regular that is, all leaves have the same dimension. The theory of singular foliations and specially holomorphic singular foliations is well developed with a plentiful literature. It merits to be presented independently. References on the subject can be found on the paper [56] by D. Cerveau. Unless otherwise stated, all the objects (manifolds, maps, functions, etc.) are assumed to be of class C ∞ . Moreover, for simplicity, we will suppose that all the manifolds are orientable. For any manifold M, we denote by A the algebra of functions on M. If E −→ M is a vector bundle, C ∞ (E) will denote the space of its global sections; this is an A-module and, equipped with the C ∞ -topology, it is a Fréchet space. In case E is the tangent bundle T M of M, we denote C ∞ (T M) simply by χ(M) (the space of vector fields tangent to M). For r ∈ N, Ω r (M) is the space of differential forms of degree r on M which is by definition C ∞ (Λr T ∗ M) where Λr T ∗ M −→ M is the vector bundle with fibre at x ∈ M the vector space of skew-symmetric forms of degree r on Tx M; Ω 0 (M) is just A. The other notations will be introduced at need.

1. Definitions and examples Let M be the Euclidean space Rm+n = Rm × Rn with canonical coordinates denoted (x, y) = (x1 , . . . , xm , y1 , . . . , yn ) and consider the family of affine subspaces Fy of M where y ∈ Rn , defined by the differential system: dy1 = · · · = dyn = 0. Then M, considered as a disjoint union of these spaces, is a nonconnected manifold of dimension m. Its topology is the product of the usual topology on Rm and the discrete one on Rn . We say that M, with this structure, is a foliated manifold of dimension m and codimension n. It constitutes the local model of a foliation of codimension n on a manifold of dimension m + n. Let O be an open set of Rm+n ; let us call a plaque of O any intersection of O with a horizontal space Fy . D EFINITION 1. Let M be a manifold of dimension m + n. A codimension n foliation F on M is given by an open cover U = (Ui )i∈I and for each i, a diffeomorphism

38

R. Barre and A. El Kacimi Alaoui

Fig. 1.

ϕi : Rm+n −→ Ui such that, on each nonempty intersection Ui ∩ Uj , the coordinate change ϕj−1 ◦ ϕi : (x, y) ∈ ϕi−1 (Ui ∩ Uj ) −→ (x , y ) ∈ ϕj−1 (Ui ∩ Uj ) has the form: x = ϕij (x, y)

and y = γij (y).

(1)

This means that the diffeomorphism ϕj−1 ◦ ϕi sends a plaque of ϕi−1 (Ui ∩ Uj ) into a

plaque of ϕj−1 (Ui ∩ Uj ). The manifold M is decomposed into connected submanifolds of dimension m. Each of these submanifolds is called a leaf of F . A subset U of M is saturated for F if it is union of leaves that is, if x ∈ U then the leaf passing through x is contained in U . Coordinate patches (Ui , ϕi ) satisfying conditions of Definition 1 are said to be distinguished for the foliation F . Let F be a codimension n foliation on M defined by a maximal atlas (Ui , ϕi )i∈I like in Definition 1. Let π : Rm+n = Rm × Rn −→ Rn be the second projection. Then the map π◦ϕi−1

fi : Ui −→ Rn is a submersion. On Ui ∩Uj = ∅ we have fj = γij ◦fi . The submersions fi and the local diffeomorphisms γij of Rn give a complete characterization of F . D EFINITION 2. A codimension n foliation on M is given by an open cover (Ui )i∈I , submersions fi : Ui −→ T over an n-dimensional transverse manifold T and, for Ui ∩ Uj = ∅, a diffeomorphism γij : fi (Ui ∩ Uj ) ⊂ T −→ fj (Ui ∩ Uj ) ⊂ T satisfying: fj (x) = γij ◦ fi (x)

for x ∈ Ui ∩ Uj .

(2)

We say that {Ui , fi , T , γij } is a foliated cocycle defining F . The proof of the equivalence of Definitions 1 and 2 is not difficult; it is left to the reader. The foliation F is said to be transversely orientable if T can be given an orientation preserved by all the local diffeomorphisms γij .

Foliations

39

1.1. Morphisms of foliations Let M and M be two manifolds endowed respectively with two foliations F and F . A map f : M −→ M will be called foliated or a morphism between F and F if, for every leaf L of F , f (L) is contained in a leaf of F ; we say that f is an isomorphism if, in addition, f is a diffeomorphism; in this case the restriction of f to any leaf L ∈ F is a diffeomorphism on the leaf L = f (L) ∈ F . Suppose now that f is a diffeomorphism of M with a codimension n foliation F . Then for every leaf L ∈ F , f (L) is a leaf of a codimension n foliation F on M; we say that F is the image of F by the diffeomorphism f and we write F = f ∗ (F ). Two foliations F and F on M are said to be C r -conjugated (topologically if r = 0, differentiably if r = ∞ and analytically in the case r = ω) if there exists a C r -homeomorphism f : M −→ M such that f ∗ (F ) = F .

1.2. The concept of holonomy This is a very important notion in Foliation Theory. In many situations it determines completely the structure of the foliation. In this subsection, we will introduce this concept and give the statement of the local and global stability theorems. Let F be a codimension n foliation on M, let L be a leaf of F and x ∈ L. Let T be a small transversal to F passing through x. Let σ : [0, 1] −→ L be a continuous path such that σ (0) = σ (1) = x. Then there exist a finite open cover Ui , i = 0, 1, . . . , k, of M with U0 = Uk and a subdivision 0 = t0 < t1 < · · · < tk = 1 of [0, 1] such that: – σ ([ti−1 , ti ]) ⊂ Ui , – if Ui ∩ Uj = ∅ then Ui ∪ Uj is contained in a distinguished chart of F . We say that Ui is a subordinated chain to σ . For i = 0, 1, . . . , k let Ti be a small transversal to F passing through σi (t) with T0 = Tk = T . For every point z ∈ Ti , sufficiently close to σ (ti ), the plaque of F passing through z intersects Ti+1 in a unique point fi (z). The domain of fi contains a transversal Ti passing through σ (ti ) and homeomorphic to an open ball of Rn . Then, it is clear that the map: fσ = fk−1 ◦ fk−2 ◦ · · · ◦ f0 is well defined on an open neighborhood of x; it is called the holonomy map associated to σ . We can prove (see [39], for instance) that the germ of fσ : – does not depend on the chain Ui , i = 1, . . . , k, and in the choice of σ in its homotopy class in the group π1 (L, x) of the homotopy classes of loops based at x, – satisfies fσ (x) = x. So we get a homomorphism h : [σ ] ∈ π1 (L, x) −→ fσ ∈ G(T , x) where G(T , x) is the group of germs of diffeomorphisms of T fixing the point x. This representation h is called the holonomy of the leaf L at x. It is trivial if L is simply connected. The foliation F is said to be without holonomy if this representation is trivial for every leaf L of F and every point x ∈ L. T HEOREM 1 (Local stability). Suppose that F admits a compact leaf L with finite fundamental group. Then L admits a saturated neighborhood V such that every leaf contained in V is compact with finite fundamental group.

40

R. Barre and A. El Kacimi Alaoui

T HEOREM 2 (Global stability). Suppose that M is compact, the codimension of F is one and that F admits a compact leaf with finite fundamental group. Then all leaves of F are compact with finite fundamental group. The proof can be found in the original paper of G. Reeb [285] or in the book [39] by C. Camacho and A. Lins Neto.

1.3. Examples of foliations (i) Simple foliations. On every manifold M we have a foliation by taking points as leaves. Its codimension is equal to the dimension of M. Also M can be equipped with a codimension zero foliation with only one leaf, namely, M itself. π In general, every submersion M −→ B with connected fibres defines a foliation. The leaves being the fibres π −1 (b), b ∈ B. In particular, every product F × B is a foliation with leaves F × {b}, b ∈ B. These foliations are transversely orientable if, and only if, the manifold B is orientable. These are simple foliations. We shall give more interesting examples in different situations. (ii) One-dimensional foliations. Let us begin by surfaces. Let M˜ = R2 and consider the differential equation dy −α dx = 0 where α is a real number. This equation has y = αx +c, c ∈ R, as general solution. When c varies, we obtain a family of parallel lines which defines ˜ a foliation F˜ in M. The natural action of Z2 on M˜ preserves the foliation F˜ (i.e. the image of any leaf of F˜ by an integer translation is a leaf of F˜ ). Then F˜ induces a foliation F on the torus T2 = R2 /Z2 . The leaves are all diffeomorphic to the circle S1 if α is rational and to the real line if α is not rational (Figure 2). In fact, if α is not rational, every leaf of F is dense; this shows that even if locally a foliation is simple, globally it can be complicated. Let M be a closed orientable surface. The fact that M admits a one-dimensional foliation depends on the topology of M, which is described by the Euler–Poincaré number χ(M);

Fig. 2.

Foliations

41

Fig. 3.

Fig. 4.

this number can be defined as follows: take a triangulation of M, i.e. a decomposition of M into triangles such as shown for the 2-sphere S2 (Figure 3). Let b0 , b1 and b2 be the numbers respectively of vertices, edges and triangles. Then χ(M) = b0 − b1 + b2 is independent of the triangulation; it is called the Euler–Poincaré number of M. (There are many books on Algebraic or Differential Topology where we can find the proof of this fact.) It classifies completely the topology of closed orientable surfaces, i.e. M and M are homeomorphic if, and only if, χ(M) = χ(M ). For the triangulation of S2 in Figure 3 we have b0 = 4, b1 = 6 and b2 = 4. So χ(S2 ) = 2. The Euler–Poincaré number of M is the only obstruction to the existence of dimension one foliation on M: M admits such foliation if, and only if, χ(M) = 0. For example, S2 cannot support a one-dimensional foliation. In fact, T2 is the only one compact orientable surface which admits a foliation of dimension one. The reader can prove, by using an adequate triangulation, that a closed orientable surface Mg of genus g (see in Figure 4 the case g = 2) has χ(Mg ) = 2−2g as Euler–Poincaré number. Then Mg admits a foliation F of dimension one if, and only if, g = 1, i.e. Mg is T2 . Suppose M is compact of dimension n. For each r = 0, 1, . . . , n, let H r (M, R) denote the real rth cohomology space of M which is finite dimensional. Then the number n (−1)r dim H r (M, R) χ(M) = r=0

42

R. Barre and A. El Kacimi Alaoui

is a topological invariant called the Euler–Poincaré number of M. For a surface, it is exactly the number defined above by using a triangulation. The manifold M admits a onedimensional foliation if, and only if, χ(M) = 0. (iii) Reeb foliation on the 3-sphere S3 . Let M be the 3-dimensional sphere S3 = ¯ its closure {(z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 1}. Denote by D the open unit disc in C and D which is the closed unit disc {z ∈ C: |z| 1}. The two subsets: 1 M+ = (z1 , z2 ) ∈ S3 : |z1 |2 2

1 and M− = (z1 , z2 ) ∈ S3 : |z2 |2 2

¯ × S1 . They have T2 as common boundary: are diffeomorphic to D 1 ∂M+ = ∂M− = (z1 , z2 ) ∈ S3 : |z1 |2 = |z2 |2 = 2 and their union is equal to S3 . Then S3 can be obtained by gluing M+ and M− along their boundaries by the diffeomorphism (z1 , z2 ) ∈ ∂M + −→ (z2 , z1 ) ∈ ∂M− , i.e. we identify (z1 , z2 ) with (z2 , z1 ) in the disjoint union M+ M− . Let f : D −→ R be the function defined by: f (z) = exp

1 . 1 − |z|2

Let t denote the second coordinate in D × R. The family of surfaces (St )t∈R obtained by translating the graph S of f along the t-axis defines a foliation on D × R. If we add the ¯ we obtain a codimension one cylinder S1 × R, where S1 is viewed as the boundary of D, ¯ ˜ ˜ foliation F on D × R. By construction, F is invariant by the transformation ¯ × R −→ (z, t + 1) ∈ D ¯ × R; (z, t) ∈ D so it induces a foliation F0 on the quotient: ¯ × R/(z, t) ∼ (z, t + 1) D ¯ × S1 . D It has the boundary T2 = S1 × S1 as a closed leaf. The others are diffeomorphic to R2 (see Figure 5). ¯ × S1 , F0 defines on M+ and M− respecBecause M+ and M− are diffeomorphic to D tively two foliations F+ and F− which give a codimension one foliation F on S3 called the Reeb foliation. (iv) Lie group actions. Let M be a manifold of dimension m + n and G a connected Lie Φ

group of dimension m. An action of G on M is a map G × M −→ M such that: – Φ(e, x) = x for every x ∈ M (where e is the unit element of G), – Φ(g , Φ(g, x)) = Φ(g g, x) for every x ∈ M and every g, g ∈ G.

Foliations

43

Fig. 5.

Suppose that, for every point x ∈ M, the dimension of the isotropy subgroup: Gx = g ∈ G: Φ(g, x) = x is independent of x. Then the action Φ defines a foliation F of dimension = m − dim Gx ; its leaves are the orbits {Φ(g, x): g ∈ G}. In particular, this is the case if Φ is locally free, i.e. if, for every x ∈ M, the isotropy subgroup Gx is discrete. An explicit example is given when M is the quotient H /Γ of a Lie group H by a discrete subgroup Γ and G is a connected Lie subgroup of H ; the action of G on M being induced by the left action of G on H . We say that F is a homogeneous foliation. Let us give an explicit example (for more details see [93]). Let A ∈ SL(m + n − 1, Z), where m + n 3, be a matrix diagonalizable and having all its eigenvalues µ1 , . . . , µm−1 , λ1 , . . . , λn real and positive. We can think of A as a diffeomorphism of the (m + n − 1)-torus Tm+n−1 . Let X1 , . . . , Xm−1 , Y1 , . . . , Yn be linear vector fields on Tm+n−1 such that: A∗ Xj = µj Xj ,

A∗ Yk = λk Yk

for j = 1, . . . , m − 1 and k = 1, . . . , n,

and denote by F0 the foliation on Tm+n−1 defined by the vector fields X1 , . . . , Xm−1 . The product of F0 by R gives a codimension n foliation on Tm+n−1 × R which is invariant by the diffeomorphism φ of Tm+n−1 × R sending (z, t) to (A(z), t + 1). So, it induces a m+n codimension n foliation F on the quotient manifold TA = Tm+n−1 × R/φ. Notice that m+n 1 m+n−1 . In fact Tm+n is the homoTA is a flat bundle over the circle S with fibre T A geneous space H /Γ where H is the semi-direct product of Rm+n−1 by R given by the action: (t, z) ∈ R × Rm+n−1 −→ At z ∈ Rm+n−1

44

R. Barre and A. El Kacimi Alaoui

and Γ is the subgroup:

(m, k) ∈ H | m ∈ Zm+n−1 , k ∈ Z .

If v1 , . . . , vm−1 ∈ Rm+n−1 are eigenvectors of A corresponding respectively to the eigenvalues µ1 , . . . , µm−1 then the subgroup: G=

m−1

ai vi , b ∈ H a1 , . . . , am−1 , b ∈ R

i=1

is isomorphic to the semi-direct product of Rm−1 by R∗+ where R∗+ acts on Rm−1 by homotheties on each factor. The action of G on Tm+n A , induced by this identification, is a locally free action whose orbits define the foliation F . (v) Foliations obtained by suspension. Let B and F be two manifolds, respectively of dimensions m and n. Suppose that the fundamental group π1 (B) of B is finitely generated. Let ρ : π1 (B) −→ Diff(F ) be an injective representation, where Diff(F ) is the diffeomorphism group of F . Denote by B˜ the universal covering of B and F˜ the horizontal foliation on M˜ = B˜ × F , i.e. the foliation whose leaves are the subsets B˜ × {y}, y ∈ F . This foliation is invariant by all the transformations Tγ : M˜ −→ M˜ defined by Tγ (x, ˜ y) = (γ · x, ˜ ρ(γ )(y)) ˜ then F˜ induces a codimension n foliwhere γ · x˜ is the natural action of γ ∈ π1 (B) on B; ation Fρ on the quotient manifold: ˜ x, M = M/( ˜ y) ∼ γ · x, ˜ ρ(γ )(y) . We say that Fρ is the suspension of the diffeomorphism group Γ = ρ(π1 (B)). The leaves of Fρ are transverse to the fibres of the natural fibration induced by the first projection ˜ B˜ × F −→ B.

Fig. 6.

Foliations

45

π

Conversely, suppose that F −→ M −→ B is a fibration with compact fibre F and that F is a codimension n foliation (n = dimension of F ) transverse to the fibres of π . Then there exists a representation ρ : π1 (B) −→ Diff(F ) such that F = Fρ . Concrete example: let B be the circle S1 and F = R+ = [0, +∞[. Let ρ be the representation of Z = π1 (S1 ) in Diff([0, +∞[) defined by ρ(1) = ϕ where ϕ(y) = λy with λ ∈ ]0, 1[. Because ϕ is isotopic to the identity map of F , the manifold M is diffeomorphic to S1 × R+ and the foliation Fρ has one closed leaf diffeomorphic to the circle S1 , corresponding to the fixed point ϕ(0) = 0 (see Figure 6).

1.4. Foliations and differential systems Let M be a manifold of dimension m + n. Denote by T M the tangent bundle of M and let E be a subbundle of rank m. Let U be an open set of M such that on U , T M is equivalent to the product U × Rm+n . At each point x ∈ U , the fibre Ex can be considered as the kernel of n differential 1-forms ω1 , . . . , ωn linearly independent: Ex =

n

ker ωj (x).

(S)

j =1

The subbundle E is called an m-plane field on M. We say that E is involutive, if for every vector field X and Y tangent to E (i.e. sections of E), the bracket [X, Y ] is also tangent to E. We say that E is completely integrable if, through each point x ∈ M, there exists a submanifold Px of dimension m which admits E|Px (the restriction of E to Px ) as tangent bundle. The maximal connected submanifolds satisfying this property are called the integral submanifolds of the differential system (S). They define a partition of M, i.e. a codimension n foliation. We have T HEOREM 3 (Frobenius). Let E be a subbundle of rank m given locally by a differential system like in (S). Then the following assertions are equivalent: – E is involutive, – E is completely integrable, – there n exist differential 1-forms (defined locally) (βij ), i, j = 1, . . . , n, such that dωi = j =1 βij ∧ ωj , i = 1, . . . , n. For example, let ω be a nonsingular 1-form. The corresponding subbundle E has fibre Ex = ker(ωx ). It defines a codimension one foliation if, and only if, there exists a 1-form β such that dω = β ∧ω; this condition is equivalent to dω ∧ω = 0. In particular, if ω is closed it defines a codimension one foliation F . If M is compact, all leaves are diffeomorphic and integration of ω over loops of M gives rise to a morphism h : π1 (M) −→ R. The range Γ = h(π1 (M)) of h is a subgroup of R called the holonomy group of F . Example (ii) is of this type: M = T2 , ω = dy − α dx which is closed. The fundamental group of T2 is Z ⊕ Z and it is easy to see that Γ = {p + qα: p, q ∈ Z}.

46

R. Barre and A. El Kacimi Alaoui

1.5. Notations Let F be a codimension n foliation on M. We denote by T F the tangent bundle to F and νF the quotient T M/T F which is the normal bundle to F ; χ(F) will denote the space of sections of T F (elements of χ(F) are vector fields X ∈ χ(M) tangent to F ). A differential form α ∈ Ω r (M) is said to be basic if it satisfies iX α = 0 and LX α = 0 for every X ∈ χ(F). (Here iX and LX denote respectively the inner product and the Lie derivative with respect to the vector field X.) For a function f : M −→ R, these conditions are equivalent to X · f = 0 for every X ∈ χ(F), i.e. f is constant on the leaves of F ; we denote by Ω r (M/F) the space of basic forms of degree r on the foliated manifold (M, F); this is a module over the algebra Ab of basic functions. A vector field Y ∈ χ(M) is said to be foliated, if for every X ∈ χ(F), the bracket [X, Y ] ∈ χ(F). We can easily see that the set χ(M, F) of foliated vector fields is a Lie algebra and an Ab -module; by definition χ(F) is an ideal of χ(M, F) and the quotient χ(M/F) = χ(M, F)/χ(F) is called the Lie algebra of transverse (or basic) vector fields on the foliated manifold (M, F). Also, it has a module structure over the algebra Ab .

2. Transverse structures Let M be a manifold of dimension m + n endowed with a codimension n foliation F defined by a foliated cocycle {Ui , fi , T , γij } like in Definition 2. D EFINITION 3. A transverse structure to F is a geometric structure on T invariant by the local diffeomorphisms γij . This is a very important notion in Foliation Theory. To make it clear, let us give the main examples.

2.1. Measurable foliations Let BT denote the family of Borel sets on T . A transverse invariant measure to F is a measure µ on BT such that, for any A ∈ BT in the domain of definition of γij , we have µ γij (A) = µ(A). We say that F is a measurable foliation if it admits a transverse measure. The notion of measurable foliation was introduced firstly by J.F. Plante; he obtained many interesting results on the qualitative behavior of codimension one measurable foliations on compact manifolds (cf. [273]).

Foliations

47

2.2. Lie foliations We say that F is a Lie foliation, if T is a Lie group G and γij are restrictions of left translations on G. Such foliation can also be defined by a 1-form ω on M with values in the Lie algebra G such that: (i) ωx : Tx M −→ G is surjective for every x ∈ M, (ii) dω + 12 [ω, ω] = 0. If G is Abelian, ω is given by n linearly independent closed scalar 1-forms ω1 , . . . , ωn . In particular, if n = 1, an important topological property of compact manifolds supporting such foliation is given by the following theorem due to Tischler [350]. T HEOREM 4. If a compact manifold admits a closed nonsingular 1-form, then it is a locally trivial fibration over the circle S1 . The hypothesis G is Abelian is important: D. Lehmann [211] proved that, in general, the result is false even if G is nilpotent. Foliations defined by nonsingular closed 1-forms can be considered as topological prototype of codimension one foliations without holonomy as it is illustrated by Sacksteder’s theorem [311]: T HEOREM 5. Let F be a C r (r 2) codimension one foliation on a connected compact manifold. If F has no holonomy, then it is topologically conjugated to a foliation defined by a nonsingular closed 1-form. In the general case, the structure of a Lie foliation on a compact manifold, is given by the following theorem due to E. Fédida [101]: T HEOREM 6. Let F be a Lie G-foliation on a compact manifold M. Let M˜ be the universal ˜ Then there exist a homomorphism h : π1 (M) −→ G covering of M and F˜ the lift of F to M. ˜ and a locally trivial fibration D : M −→ G whose fibres are the leaves of F˜ and such that, for every γ ∈ π1 (M), the following diagram is commutative: γ −→ M˜ M˜ D↓ ↓D

G

h(γ )

−→

G

˜ where the first line denotes the deck transformation of γ ∈ π1 (M) on M. The subgroup Γ = h(π1 (M)) ⊂ G is called the holonomy group of F although the holonomy of each leaf is trivial. The fibration D : M˜ −→ G is called the developing map of F .

48

R. Barre and A. El Kacimi Alaoui

2.3. Transversely parallelizable foliations We say that F is transversely parallelizable if there exist on M, foliated vector fields Y1 , . . . , Yn , transverse to F and everywhere linearly independent. This means that the manifold T admits a parallelism (Y1 , . . . , Yn ) invariant by all the local diffeomorphisms γij or, equivalently, that the Ab -module χ(M/F) is free of rank n. The structure of a transversely parallelizable foliation on a compact manifold is given by the following theorem due to L. Conlon [60] for n = 2 and in general to P. Molino [243]. T HEOREM 7. Let F be a transversely parallelizable foliation of codimension n on a compact manifold M. Then: (1) the closures of the leaves are submanifolds which are fibres of a locally trivial fibration π : M −→ W where W is a compact manifold, (2) there exists a simply connected Lie group G0 such that the restriction F0 of F to any leaf closure F is a G0 -Lie foliation, (3) the cocycle of the fibration π : M −→ W has values in the group of diffeomorphisms of F preserving F0 . The fibration π : M −→ W and the manifold W are called respectively the basic fibration and the basic manifold associated to F . Theorem 7 says that if, in particular, the leaves of F are closed, then the foliation is just a fibration over W . This is still true even if the leaves are not closed: the manifold M is a fibration over the leaf space M/F which is, in this case, a Q-manifold in the sense of [13]. Theorem 7 is still valid for transversely complete foliations on noncompact manifolds (cf. [242]). It is not difficult to see that any Lie foliation is transversely parallelizable. This is a consequence of the fact that a Lie group is parallelizable and that the parallelism can be chosen invariant by left translations. 2.4. Riemannian foliations The foliation F is said to be Riemannian if there exists on T a Riemannian metric such that the local diffeomorphisms γij are isometries. Using the submersions fi : Ui −→ T one can construct on M a Riemannian metric which can be written in local coordinates: ds 2 =

m i,j =1

θi ⊗ θj +

n

gk (y) dyk ⊗ dy .

k,=1

Equivalently, F is Riemannian, if any geodesic orthogonal to the leaves at a point is orthogonal to the leaves everywhere [289]. Let F be Riemannian. Then there exists a Levi-Civita connection, transverse to the leaves which, by unicity argument, coincides on any distinguished open set, with the pullback of the Levi-Civita connection on the Riemannian manifold T . This connection is said τ to be projectable. Let O(n) −→ M # −→ M be the principal bundle of orthonormal frames transverse to F ; this is an F -bundle, in the sense of Section 7.1. The following theorem is due to P. Molino [243].

Foliations

49

T HEOREM 8. Suppose M is compact. Then, the foliation F can be lifted to a foliation F # on M # of the same dimension and such that: (1) F # is transversely parallelizable, (2) F # is invariant under the action of O(n) on M # and projects, by τ , on F . π#

The basic manifold W # and the basic fibration F # −→ M # −→ W # are called respectively the basic manifold and the basic fibration of F . We have the following properties: – the restriction of τ to a leaf of F # is a covering over a leaf of F . So all leaves of F have the same universal covering, – the closure of any leaf of F is a submanifold of M and the leaf closures define a singular foliation (the leaves have different dimensions) on M. (For more details about this notion see [243].) Another interesting result for Riemannian foliations is the Global Reeb Stability Theorem which is valid even if the codimension is greater than 1. T HEOREM 9. Let F be a Riemannian foliation on a compact manifold M. If there exists a compact leaf with finite fundamental group, then all leaves are compact with finite fundamental group. The property F is Riemannian means that the leaf space Q = M/F is a Riemannian manifold even if Q does not support any differentiable structure!

2.5. Transversely holomorphic foliations The foliation F is said to be transversely holomorphic if T is a complex manifold and the γij are local biholomorphisms. Particular case is a holomorphic foliation: the manifolds M and T are complex, all the fi are holomorphic and all γij are local biholomorphisms. If T is Kählerian and γij biholomorphisms which preserve the Kähler form on T we say that F is transversely Kählerian. For example, any codimension 2 Riemannian foliation which is transversely orientable is transversely Kählerian. Let us give concrete examples of such foliations. Let M be the unit sphere in the Hermitian space Cn+1 : M = S2n+1 = (z1 , . . . , zn+1 ) ∈ Cn+1 :

n+1

|zk |2 = 1 .

k=1

Let Z be the holomorphic vector field on Cn+1 given by the formula: Z=

n+1 k=1

ak zk

∂ , ∂zk

50

R. Barre and A. El Kacimi Alaoui

where ak = αk + iβk ∈ C. There exists a good choice of the numbers ak such that the orbits of Z intersect transversely the sphere M; then Z induces on M a real vector field X which defines a foliation F . It is not difficult to see that F is transversely holomorphic. It is transversely Kählerian if we choose in addition αk = 0 for any k = 1, . . . , n + 1. 3. Codimension one foliations Codimension one foliations constitute a rich theme which was studied extensively by many people. The richness comes from the existence, for such foliations, of nonsingular transverse vector fields which give a way to go from a leaf to an other. Most of the results in Foliation Theory were first obtained in the codimension one case; this section is devoted to summarize some of them. Let F be a codimension one foliation on a manifold M and ν a transverse vector bundle to F . Because ν is of rank one, it is integrable and defines a foliation V transverse to F . So we have clearly χ(M) = 0. It is natural to ask if this condition is sufficient for the existence of a codimension one foliation on M; this was conjectured by E. Thomas [342]. The reader can see the paper [207] by B. Lawson about the history of the different steps for solving this conjecture. The final solution was given by W. Thurston [348] who proved T HEOREM 10 (Thurston). Let M be a compact manifold. Then M admits a codimension one foliation if, and only if, the Euler–Poincaré number χ(M) of M is zero. Recall that two vector bundles E −→ M and E −→ M are said to be homotopic if there exists a continuous family Et −→ M, t ∈ [0, 1], of vector bundles such that E0 = E and E1 = E . So we can formulate the question of existence of codimension one foliations, in general, in the following: Let M be a compact manifold. Then any codimension one plane field on M is homotopic to an integrable one. The first results solving (in some particular cases), this conjecture were obtained by J. Wood (see [376]) and also by P. Schweitzer and W. Thurston in the C 0 -case. As far as we know this conjecture is still open. Notice that the compactness of the manifold is a big constraint. Indeed on open manifolds the answer to this conjecture is positive [266]. The regularity property seems to be very important in the existence of foliations on compact manifolds. In particular, there is a big difference in the treatment between the C ∞ case and the real analytic one. In this direction A. Haefliger proved in [143] the following important theorem. T HEOREM 11 (Haefliger). Let M be a compact manifold with a finite fundamental group. Then M has no real analytic codimension one foliation. Let us end this section with one of the most important results obtained in codimension one foliation theory on 3-manifolds [261].

Foliations

51

T HEOREM 12 (Novikov). Let M be a compact 3-manifold with a finite fundamental group. Then any codimension one foliation on M has a compact leaf diffeomorphic to the torus T2 . 4. Γ -structures This notion was introduced by A. Haefliger and became a key ingredient in studying characteristic classes of foliations. D EFINITION 4. A groupoid is given by a set Γ , a subset Γ (2) of Γ × Γ with a law (γ , σ ) ∈ Γ (2) −→ γ σ ∈ Γ and an inverse map γ ∈ Γ −→ γ −1 ∈ Γ satisfying the following properties: (i) (γ −1 )−1 = γ , (ii) if (γ , σ ), (σ, τ ) ∈ Γ (2) , then (γ σ, τ ), (γ , σ τ ) ∈ Γ (2) and (γ σ )τ = γ (σ τ ), (iii) if (γ −1 , γ ) ∈ Γ (2) and (γ , σ ) ∈ Γ (2) , then γ −1 (γ σ ) = σ , (iv) if (γ , γ −1 ) ∈ Γ (2) and (τ, γ ) ∈ Γ (2) , then (τ γ )γ −1 = τ . For γ ∈ Γ , s(γ ) = γ −1 γ is called the source of γ and r(γ ) = γ γ −1 the range of γ . Then, there are two projections s, r (or α, β) : Γ −→ Γ (0) = Im r. The subset Γ (0) of Γ is called the unit space of Γ . A topological groupoid is a groupoid with a topology compatible with the composition and inverse maps. As a consequence, the two projections s, r on the unit space are also continuous. A differentiable structure on Γ is given by a manifold structure on Γ and Γ (0) compatible with the composition and inverse maps and such that: – s : Γ −→ Γ (0) is a submersion, – the canonical injection Γ (0) −→ Γ is an embedding. The differentiable (or topological) groupoid Γ is étale if s is étale. Let M be a manifold, Γ a topological groupoid and {Ui } an open cover of M; a 1-cocycle on M with values in Γ is given as follows: for each pair (i, j ), let γij : Ui ∩ Uj −→ Γ be a continuous map such that, if x ∈ Ui ∩ Uj ∩ Uk , then (γij (x), γj k (x)) ∈ Γ (2) and γik (x) = γij (x)γj k (x). Two 1-cocycles are said to be cohomologous if they are restrictions of the same cocyle on the union of their coverings. A Γ -structure on M, or an element of H 1 (M, Γ ), is an equivalence class of 1-cocycles. Let Γ be the groupoid of germs of local diffeomorphisms of Rn ; then the unit space Γ (0) may be identified to Rn . A codimension n foliation F on M may be viewed as a particular Γ -structure for which a representative is a 1-cocycle on an open covering {Ui } such that the following maps fi = γii : Ui −→ Γ (0) = Rq are submersions.

52

R. Barre and A. El Kacimi Alaoui

5. The leaf space Let F be a codimension n foliation on M. Let U be a subset of M and denote by Uˆ the union of the leaves intersecting U . Recall that U is saturated if U = Uˆ . It is easy to see that if U is open, so is Uˆ . Then, the equivalence relation on M, x ∼ y if, and only if, x and y are in the same leaf, is open. The set of equivalence classes M/ ∼, endowed with the quotient topology, is called the leaf space of F and usually denoted by M/F . We can think of M/F as follows. The foliation F is the geometric realization of a completely integrable differential system (S) on M. Each integral submanifold is a leaf of F and corresponds to an initial condition of (S). So we can consider Q as a parameter space of the initial conditions of this differential system. In general Q is not a manifold, but we can define on this space many geometrical objects like functions, differential forms, differential operators, etc. (cf., for instance, Section 7). They correspond to their analogues on M invariant along the leaves (in a sense to be determined following the context). There were many attempts to give the leaf space of a foliation a differentiable structure, even if its topology is, generally poor. A first one was from Satake, whose point of view was recovered by W. Thurston. In other domains, let us cite G.W. Mackey [219] who introduced the virtual group notion in Ergodic Theory and M. Artin [12] the algebraic space notion. The former corresponds to the measurable version of the S-atlas of W.T. Van Est, the latter suggested the definition of a Q-manifold. In fact, there is no uniform definition. Each corresponds to a given situation or a particular problem. Nevertheless, the point of view of Noncommutative Geometry, by A. Connes, using the C ∗ -algebra of the groupoid of a foliation, or the crossed-product of the C ∗ -algebra of a manifold by a group acting on it, is attractive and efficient too. For example, there are Longitudinal and Transversal Index Theorems for foliations; one gets also Godbillon–Vey classes, etc.

5.1. V -manifolds Let Ω be an open set in Rn and let Σ be a finite group of diffeomorphisms of Ω. Denote by Ω/Σ the orbit space with quotient topology and p the canonical projection Ω −→ Ω/Σ. If Ω is another open set of Rn , Σ a finite group of diffeomorphisms of Ω and p the canonical projection Ω −→ Ω /Σ , then a morphism from Ω/Σ to Ω /Σ is a continuous map f from Ω/Σ to Ω /Σ , which admits local coverings by smooth local maps from Ω to Ω . An isomorphism is a bijective morphism, the inverse of which is a morphism. If V is a second countable Hausdorff space, a Satake atlas of dimension n is a family A = (Ui , Φi ) where (Ui ) is an open covering of V and Φi : Ui −→ Ωi /Σi is a homeomorphism of Ui on the quotient of an open subset Ωi of Rn by a finite group of diffeomorphisms, with following coherence condition: for all i, j such that Ui ∩ Uj = ∅ the map Φj ◦ Φi−1 : Φi (Ui ∩ Uj ) −→ Φj (Ui ∩ Uj ) is a morphism as previously defined. A V -manifold (or a Satake manifold or an orbifold) of dimension n is a space V with a maximal Satake atlas of dimension n. The following are simple examples illustrating the notion of a V -manifold.

Foliations

53

(i) Let Γ be a finite group of isometries of a Riemannian manifold M of dimension n. Then the quotient space M/Γ is a V -manifold of dimension n. (ii) A waterdrop obtained by gluing two open discs along their boundaries, one of them being implemented with a rotation of 2π 3 around its center. It is proved in [243] that every leaf space of a Riemannian foliation with compact leaves on a compact manifold is a V -manifold. Conversely: every compact V -manifold is the leaf space of a Riemannian foliation with compact leaves on a compact manifold (cf. [132]). 5.2. QF -manifolds Let (X, p, S) be a triple where X is a manifold, S a set and p a surjective map from X to S; this is an étale QF -atlas of S if it satisfies the following conditions: (H) for every pair (x, y) in X 2 such that p(x) = p(y), there are open neighborhoods U and V respectively of x and y and a diffeomorphism h from U to V such that h(x) = y and p ◦ h(t) = p(t) for every t ∈ U , (QF) every morphism from a manifold Z to X such that p ◦ f is constant is locally constant. As usual two étale QF -atlases (X1 , p1 , S) and (X2 , p2 , S) are equivalent if (X, p, S) is an étale QF -atlas where X is the disjoint union of X1 and X2 and p is p1 on X1 and p2 on X2 . A QF -manifold structure on S is an equivalence class of étale QF -atlases on S. All the leaf spaces of foliated second countable manifolds are in this category. 5.3. Q-manifolds Let (X, p, S) be a triple where X is a manifold, S a set and p a surjective map from X to S; this is a Q-atlas of S if it satisfies the following conditions: (H) is as in the definition of an étale QF -atlas, (Q) let f = (f1 , f2 ) be a morphism from a manifold Z to X 2 such that p ◦ f1 = p ◦ f2 ; then the subset T = {z ∈ Z: f1 (z) = f2 (z)} is open in Z. A Q-manifold structure on S is an equivalence class of Q-atlases of S. The following are examples of Q-manifolds: (i) the leaf space of foliated torus with geodesics having irrational slope, (ii) more generally, the leaf space of a transversely parallelizable foliation on a compact manifold. It was first tried to generalize to leaf spaces the classical theorems and tools (Gauss– Bonnet, de Rham cohomology, Poincaré duality, Leray–Serre spectral sequence, fundamental group, etc.) to get results on the transverse structure. The V -manifolds are met in natural way and there exist many examples of them. They appear also with ramified coverings. The Q-manifolds permitted to restore the third Lie theorem for Banach Lie algebras (cf. [268]); they appear also in the structure theorem of P. Molino. Recently, G. Meigniez got a characterization of Godbillon Homotopy Extension Property for foliations, where they play a role (cf. [231]).

54

R. Barre and A. El Kacimi Alaoui

6. Characteristic classes We follow here the lectures of R. Bott as written by L. Conlon in [23]. We will restrict ourself to only one result: Bott vanishing theorem. The reader can find more material in [23] or in [147].

6.1. The classifying space and the universal bundle Let H be a separable real Hilbert space with norm . If u and v are nonzero vectors in H, (u, v) will be the angle defined by u and v; it is immediate to see that for every positive number λ and µ we have (λu, µv) = (u, v). Denote by BGLn the set of n-dimensional subspaces of H. Let τ, σ ∈ BGLn and set δ(τ, σ ) = inf(u, v) where the infimum is taken over all the vector u ∈ τ and v ∈ σ with u = v = 1. It is not difficult to see that δ defines a distance on BGLn . The topological space BGLn is called the classifying space of the group GL(n, R) of linear transformations of the vector space Rn . The cohomology H ∗ (BGLn , R) of BGLn is a polynomial ring R[p1 , . . . , p[n/2] ] where the pi ∈ H 4i (BGLn , R) are the universal Pontryagin classes (cf. [21]). On BGLn we have a canonical real vector bundle S −→ BGLn of rank n whose fibre at each τ is the space τ itself; it is called the universal bundle on BGLn .

6.2. Classification of real vector bundles As M is paracompact, it admits a countable locally finite open cover U = {Ui } which, in addition, can be chosen such that each finite intersection Ui1 ∩ · · · ∩ Ui is contractible. Such an open cover is called a good cover; it always exists: take a Riemannian metric on M π and a countable family of geodesically convex open balls which covers M. If E −→ M is a real vector bundle of rank n, its restriction E|Ui to any Ui is trivial, i.e. there exists a diffeomorphism ϕi : E|Ui −→ Ui × Rn which sends the fibre Ex isomorphically on {x} × Rn . Let (si1 , . . . , sin ) be a basis of the free module C ∞ (E|Ui ) over the algebra A(Ui ) of real valued C ∞ -functions on Ui . Let {ρi } be a partition of the unity subordinated to {Ui } and let Vi be the real vector space spanned by (ρi si1 , . . . , ρi sin ). For each i we set ψi = qi ◦ ϕi where qi : Ui × Rn −→ Rn is the second projection. Now express H as orthogonal direct sum of the Vi Rn and denote by ζi : Vi −→ H the inclusion of the ith summand. Define Φ : E −→ H by Φ(x, ξ ) =

∞

ρi (x) · ζi ψi (x, ξ ) .

i=1

Then Φ is continuous and sends each fibre π −1 (x) of E isomorphically on an n-dimensional subspace of H. Thus f (x) = Φ(π −1 (x)) defines a continuous map f : M −→ BGLn called the classifying map for the vector bundle E namely E is the pullback by f of the universal bundle S −→ BGLn . In fact there is a natural one–one correspondence between

Foliations

55

the set Vectn (M) of isomorphism classes of real vector bundles of rank n on M and the set [M, BGLn ] of homotopy classes of maps M −→ BGLn . The ith-Pontryagin class of the real vector bundle E −→ M is by definition pi (E) = f ∗ (pi ). The graded subring Pont∗ (E) = f ∗ H ∗ (BGLn , R) ⊂ H ∗ (M, R) is called the Pontryagin ring of E. The first important result obtained in the theory of characteristic classes of foliations is the following T HEOREM 13 (Bott vanishing theorem). Let F be a foliation of codimension n with normal bundle νF . Then Ponti (νF) = 0 for i > 2n. As a nontrivial example of characteristic class of a foliation F , we have the Godbillon– Vey invariant GV(F) (discovered by C. Godbillon and J. Vey [136]) which is, in general, nonzero as shown by R. Roussarie. An elementary construction of this invariant in the codimension one case is as follows. Let M be a compact manifold endowed with a codimension one foliation defined by a differential 1-form ω. Then the integrability condition implies the equality ω ∧ dω = 0, i.e. dω = α ∧ ω. It is easy to see that α ∧ dα is closed and that its cohomology class in H 3 (M, R), which is by definition GV(F), is independent of the choice of α. One of the most important results in the study of the Godbillon–Vey invariant for codimension one foliations was obtained by G. Duminy in [73]. Let us describe it briefly; a complete account is given in [119]. Let F be a codimension one foliation on a compact manifold M. A leaf L of F is called resilient if there exist a loop σ : [0, 1] −→ L and a transversal T to F passing through σ (0) such that the following conditions are satisfied: (i) there exists a point x ∈ L ∩ T in the domain of holonomy hσ of σ and different from σ (0); (ii) the sequence hnσ (x) converges to σ (0) as n → +∞. G. Duminy proved that if F has no resilient leaf then the Godbillon–Vey invariant of F is zero. Recently, A. Connes and H. Moscovici have discovered a universal Hopf algebra with cohomology from which one is able to recover the characteristic classes of a foliation without use of Chern–Weil homomorphism or connections (cf. [63]).

7. Basic global analysis Let M be a manifold endowed with a foliation F of codimension n. We suppose for simplicity that F is transversely orientable.

7.1. Foliated vector bundles and basic sections ι

Let P : G → P −→ M be a principal bundle with structural group G ⊂ GL(N, C). The group G acts on P on the right and on its Lie algebra G by the adjoint representation.

56

R. Barre and A. El Kacimi Alaoui

Denote by V the vector bundle whose fibre Vz at a point z ∈ P is the tangent space at z of the fibre of P. A connection on P is a subbundle H of T P such that: – for every z ∈ P, Tz P = Vz ⊕ Hz , – for every g ∈ G and every z ∈ P , Hzg = (Rg )∗ (Hz ) where Rg is the right action of g on P . As is well known the subbundle H is also the kernel of an invariant (under the action of G) 1-form ω on P (called the connection form) with values in G. It is easy to see that the restriction of ι∗ (the derivative of ι) to Hz is an isomorphism onto Tι(z) M. Let τ = ι−1 ∗ (T F). We say that P is foliated if τ is integrable. In this case, τ defines a foliation F˜ on P such that ˜ = dim(F), – dim(F) – F˜ is invariant under the action of G. We say that the connection H is basic, if the ω is basic (cf. Section 1.5). A foliated bundle E is said to be an F -bundle, if it admits a basic connection. Let E −→ M be a complex vector bundle defined by a cocycle {Ui , γij , G} where Ui is an open cover of M and γij : Ui ∩ Uj −→ G ⊂ GL(N, C) are the transition functions. We say that E is an F -bundle, if the associated principal bundle G −→ P −→ M is an F -bundle. Because E = P ×G CN , F˜ induces a foliation FE on E. An F -morphism ϕ : (E, ω) −→ (E , ω ) between two F -bundles is a morphism of vector bundles which sends leaves of FE into leaves of FE . (Notice that the collection of F -bundles and F -morphisms is a category. So we can define the group K(M, F) of foliated K-theory as in the classical case.) Let E −→ M be an F -bundle. Then the dual bundle E ∗ and all its exterior powers Λ∗ E ∗ are F -bundles; also H2 E = {Hermitian forms on E} is an F -bundle.

7.2. Transversely elliptic operators Let E −→ M be a F -foliated vector bundle. Denote by ∇ the covariant derivative ∇

χ(M) × C ∞ (E) −→ C ∞ (E) associated to the connection H. We say that a section α ∈ C ∞ (E) is basic, if it satisfies the condition ∇X α = 0 for every X ∈ χ(F). The space C ∞ (E/F) of basic sections of E is an Ab -module. Let E and E two F -bundles (with the same rank N for simplicity). A basic differential D

operator of order from E to E is a linear map C ∞ (E/F) −→ C ∞ (E /F) such that on local coordinates (x1 , . . . , xm , y1 , . . . , yn ) for which F is defined by the differential equations dy1 = · · · = dyn = 0, D has the expression: D=

|s|

as (y)

∂y1s1

∂ |s| , . . . ∂ynsn

where s = (s1 , . . . , sn ) ∈ Nn , |s| = s1 + · · · + sn and as are (N × N )-matrices whose coefficients are basic functions. The principal symbol of D at the point z = (x, y) and the basic

Foliations

57

covector ξ ∈ νz∗ F is the linear map σ (D)(z, ξ ) : Ez −→ Ez defined by σ (D)(z, ξ )(η) =

ξ1s1 · · · · · ξnsn as (y)(η).

|s|=

We say that D is transversely elliptic if σ (D)(z, ξ ) is an isomorphism for every z ∈ M and every basic covector ξ different from 0. If F is Riemannian, its conormal bundle ν ∗ F is an F -bundle and is equipped with a foliation F ∗ . Then the principal symbol σ (D)(z, ξ ) of a transversely elliptic operator D defines an element [D] in the group K(ν ∗ F, F ∗ ). A Hermitian metric on E is a positive definite section h of H2 E. If h is basic we say that E is a Hermitian F -bundle. Let E −→ M be a Hermitian F -foliated bundle with Hermitian metric h and let D be a basic differential operator of order = 2 on C ∞ (E/F). For every z ∈ M and every basic covector ξ ∈ νz∗ F we define a quadratic form A(D)(z, ξ ) : Ez −→ C by A(D)(z, ξ )(η) = (−1) σ (D)(z, ξ )(η), η . We say that D is strongly transversely elliptic, if A(D)(z, ξ ) is positive definite for every z ∈ M and every nonzero ξ . Obviously every strongly transversely elliptic operator is transversely elliptic. From now on we suppose that M is compact and connected. Assume that the foliation F is Riemannian transversely oriented. Let E # be the pullback of E to the principal bundle p SO(n) −→ M # −→ M of the orthonormal direct frames transverse to F (cf. Theorem 8). Then E # is a SO(n)-bundle and a Hermitian F # -bundle equipped with a Hermitian metric h# . Let W # be the basic manifold associated to the transversely parallelizable foliation F # on M # . The basic sections of E are canonically identified to basic sections of E # which are invariant under the action of SO(n). In particular, if f : M −→ C is a basic function, f ◦ p is a basic function on M # (with respect to F # ); moreover f ◦ p is invariant by the action of SO(n). Because f ◦ p is continuous, it is constant on the leaf closures of F # so it induces an SO(n)-invariant C ∞ function on the basic manifold W # . We can prove, by the converse process, that any SO(n)-invariant C ∞ function on the basic manifold W # defines a C ∞ basic function on M; in other words, the algebra Ab of basic functions on M is canonically isomorphic to the algebra ASO(n) (W # ) of functions on W # invariant by SO(n). The bundle like metric on M # induces a Riemannian metric on W # for which SO(n) acts by isometries. Let µ be the measure on W # associated to this metric. On C ∞ (E/F) we define an inner product as follows. Let α and β be two elements of ∞ C (E/F). The function Θ(α, β) : z ∈ M −→ hz (α(z), β(z)) ∈ C is basic; so it defines an SO(n)-invariant function Θ # (α, β) on W # . We set α, β =

Θ # (α, β)(w) dµ(w). W

For any basic differential operator D from a Hermitian F -bundle E to a Hermitian F -bundle E , denote by N (D) the kernel of D and R(D) its range.

58

R. Barre and A. El Kacimi Alaoui

T HEOREM 14. Let E and E be two Hermitian F -bundles on M and let D be a transversely elliptic operator from C ∞ (E/F) to C ∞ (E /F). Denote by D ∗ the formal adjoint of D which is also a basic transversely elliptic operator from C ∞ (E /F) to C ∞ (E/F). Then N(D) and N (D ∗ ) are finite dimensional and we have an orthogonal decomposition: C ∞ (E/F) = N (D) ⊕ R(D ∗ ). In particular, D has an index: ind(D/F) = dim N (D) − dim N (D ∗ ). All the details of the proof of this theorem can be found in [85].

7.3. Transversely elliptic complexes Let (E r , Dr )r=0,1,...,n be a family of Hermitian F -bundles and basic differential operators of order one Dr : C ∞ (E r /F) −→ C ∞ (E r+1 /F) (by convention Dn = 0) such that the sequence Dr+1 Dr−1 Dr · · · −→ C ∞ (E r /F) −→ C ∞ E r+1 /F −→ · · ·

(∗)

is a differential complex, that is, Dr+1 ◦ Dr = 0 for r = 0, 1, . . . , n − 1. Let z ∈ M and ξ ∈ νz∗ F ; denote by σ (Dr )(z, ξ ) the principal symbol of Dr at (z, ξ ) which is a linear map σ (Dr )(z, ξ ) : Ezr −→ Ezr+1 . Set σr = σ (Dr )(z, ξ ); we say that the complex (∗) is transversely elliptic if its symbol sequence σr−1

σr

σr+1

· · · −→ Ezr −→ Ezr+1 −→ · · ·

(∗ )

is exact for every z and every nonzero ξ . Let Dr∗ : C ∞ (E r+1 /F) −→ C ∞ (E r /F) be the formal adjoint of Dr (with respect to the inner product defined in Section 7.2). Then it is easy to see that the complex (∗) is transversely elliptic if and only if the basic operator of ∗ is strongly order 2: Lr : C ∞ (E r /F) −→ C ∞ (E r /F) defined by Lr = Dr∗ Dr + Dr−1 Dr−1 transversely elliptic. Let (E r , Dr ), r = 0, 1, . . . , n, be a transversely elliptic complex with cohomology r Hb (E ∗ ). Then applying Theorem 14, we have T HEOREM 15. (i) For each r = 0, 1, . . . , n, the kernel Hbr (E ∗ ) of Lr is equal to the space ∗ ). N (Dr ) ∩ N (Dr−1 r (ii) The space Hb (E ∗ ) is finite dimensional and we have an orthogonal decomposition C ∞ (E r /F) = Hbr (E ∗ ) ⊕ R(Dr−1 ) ⊕ R(Dr∗ ). (iii) The orthogonal projection C ∞ (E r /F) −→ Hbr (E ∗ ) induces an isomorphism from Hbr (E ∗ ) to Hbr (E ∗ ).

Foliations

59

We will give two concrete examples to illustrate this result: the basic de Rham complex and the basic Dolbeault complex. Let r ∈ {0, . . . , n} and denote by E r the vector bundle of exterior r-forms on the normal bundle νF . As it was pointed out, E r is a Hermitian F -bundle; its basic sections are exactly basic differential forms Ω r (M/F) of degree r on M. The de Rham exterior differential d restricted to Ω r (M/F) = C ∞ (E r /F) is a basic differential operator d : Ω r (M/F) −→ Ω r+1 (M/F). Thus we obtain a differential complex d

d

d

· · · −→ Ω r (M/F) −→ Ω r+1 (M/F) −→ · · ·

(∗∗)

called the basic de Rham complex of F ; its homology H r (M/F) is called the basic cohomology of F and depends only on the transverse structure of F . Let δb : Ω r+1 (M/F) −→ Ω r (M/F) be the formal adjoint of d; this operator can be described explicitly in terms of coefficients of the transverse metric on νF and the Hermitian metrics on the bundles E r (cf., for instance, [5,352,351,265,289,288,290,291,89, 195–197,85]). Let b = dδb + δb d; this is a basic differential operator of order 2 on Ω r (M/F) called the basic Laplacian. A basic form α ∈ Ω r (M/F) which satisfies the equation b α = 0, or equivalently dα = 0 and δb α = 0, is called a basic harmonic form; denote by Hr (M/F) the space of such forms. Applying Theorem 14 we obtain the following T HEOREM 16. (i) The space Hr (M/F) is finite dimensional and we have an orthogonal decomposition Ω r (M/F) = Hr (M/F) ⊕ R(d) ⊕ R(δb ). (ii) The orthogonal projection Ω r (M/F) −→ Hr (M/F) induces an isomorphism from H r (M/F) to Hr (M/F). (iii) Suppose that the vector space H n (M/F) is nonzero; then there exists a natural nondegenerate pairing Φ : ([α], [β]) ∈ H r (M/F) × H n−r (M/F) −→ Φ([α], [β]) ∈ C. So the basic cohomology satisfies Poincaré duality. During the last decades, many people contributed to the proof of this theorem. It was first proved by B.L. Reinhart in [290]. But Y. Carrière [51] discovered a mistake which makes assertion (iii) false: B.L. Reinhart does not suppose H n (M/F) different from {0} to obtain Poincaré duality; he was probably thinking that this hypothesis is automatically satisfied. Later on F.W. Kamber and P. Tondeur [196] have shown that the Reinhart’s proof is still valid if we suppose the leaves minimal (cf. Section 9.3). Finally the theorem was proved in full generality (without any assumption on the minimality of the leaves) in [89]. Now suppose that F is Hermitian. Let ν be the complexified normal bundle νF ⊗R C of νF . Let J be the automorphism of ν associated to the complex structure; J satisfies the relation J 2 = −id and then has two eigenvalues i and −i with associated eigensubbundles

60

R. Barre and A. El Kacimi Alaoui

respectively denoted ν 10 and ν 01 . We have a splitting ν = ν 10 ⊕ ν 01 which gives rise to a decomposition Λr ν ∗ =

Λp,q ,

p+q=r ∗

∗

where Λp,q = Λp ν 10 ⊗ Λq ν 01 . Basic sections of Λp,q are called basic forms of type (p, q). They form a vector space denoted Ω p,q (M/F). We have Ω r (M/F) =

Ω p,q (M/F).

p+q=r

The exterior differential decomposes into a sum of two operators ∂ : Ω p,q (M/F) −→ Ω p+1,q (M/F)

and

∂¯ : Ω p,q (M/F) −→ Ω p,q+1 (M/F) as in the classical case of a complex manifold. We have ∂¯ 2 = 0; so we obtain, for p fixed, a differential complex ∂¯

∂¯

∂¯

· · · −→ Ω p,q (M/F) −→ Ω p,q+1 (M/F) −→ · · ·

(∗∗∗)

called the basic Dolbeault complex of F ; its homology H p,q (M/F) is the basic Dolbeault cohomology of the foliation F : even though the leaf space is topologically bad, it can be considered as a “complex manifold” whose Dolbeault cohomology is H p,∗ (M/F)! ¯ this is an operator of type (0, −1). The operator Let δb denote the formal adjoint of ∂; ¯ is selfadjoint; a simple computation in local coordinates, like for the basic b = δb ∂¯ + ∂δ b Laplacian, shows that b is strongly transversely elliptic. Therefore the complex (∗∗∗) is transversely elliptic. Let ¯ = 0 and δb α = 0 . Hp,q (M/F) = Ker b = α ∈ Ω p,q (M/F): ∂α Applying Theorem 14, we obtain T HEOREM 17. (i) The space Hp,q (M/F) is finite dimensional and we have an orthogonal decomposition ¯ ⊕ R(δb ). Ω p,q (M/F) = Hp,q (M/F) ⊕ R(b ) = Hp,q (M/F) ⊕ R(∂) (ii) The orthogonal projection Ω p,q (M/F) −→ Hp,q (M/F) induces an isomorphism from H p,q (M/F) to Hp,q (M/F). (iii) Suppose that the vector space H n (M/F) is nonzero; then there exists a natural nondegenerate pairing Ψ : ([α], [β]) ∈ H p,q (M/F) × H n−p,n−q (M/F) −→ Ψ ([α], [β]) ∈ C. So the basic Dolbeault cohomology satisfies Serre duality.

Foliations

61

Suppose now that F is transversely Kählerian with Kähler form ω (it is a basic differential form of degree 2; it is closed and nondegenerate). In this case, we can prove that b = 2b . Because of the decomposition Ω r (M/F) =

Ω p,q (M/F),

p+q=r

every basic differential r-form can be uniquely written as a sum α = p+q=r αpq where αpq ∈ Ω p,q (M/F). Then we have the following assertions. (iv) α is b -harmonic if, and only if, each component αpq is b -harmonic. So we have a direct decomposition H r (M/F) =

H p,q (M/F).

p+q=r

(v) The complex conjugacy induces an isomorphism (of real vector spaces) H p,q (M/F) H q,p (M/F). (vi) For every odd r ∈ {0, . . . , 2n}, the dimension of the space H r (M/F) is even. In particular, if n = 1 we have b1 (M/F) = 2 dim H 01 (M/F). The integer dim H 01 (M/F) will be denoted g(F) and called the genus of the foliation F . It is similar to the genus of a compact Riemann surface; it counts the number of linearly independent basic holomorphic 1-forms. (vii) For every p ∈ {0, . . . , n} the differential form ωp = ω ∧ · · · ∧ ω (wedge product p times) is harmonic. So, the space H p,p (M/F) is nonzero. Notice that this theorem is also a particular case of Theorem 15. It can be used to establish more properties: basic Hodge structures for transversely Kählerian foliations, basic Calabi–Yau theorem [85] and deformation of transversely holomorphic foliations with a fixed differentiable type [90].

8. Deformation theory of foliations We will describe only the real case following Hamilton’s paper [159]. Deformation theory of holomorphic or generally transversely holomorphic foliations is more rich. The reader can find a good account of the subject in [132]. Let M be a manifold of dimension m + n. For each x ∈ M, let G(x, m) be the Grassmanian manifold of m-planes in Tx M. Then: G(m) =

G(x, m)

x∈M

can be given a structure of a differentiable manifold such that the canonical projection (x, τ ) ∈ G(m) −→ x ∈ M is a locally trivial fibration, the fibre being the Grassmanian

62

R. Barre and A. El Kacimi Alaoui

G(m) of m-planes in the space Rm+n . Then a subbundle of rank m of T M is just a section of the bundle G(m) −→ M. Denote by C ∞ (G(m)) the space of sections of this bundle. Let τ ∈ C ∞ (G(m)). By Frobenius theorem, τ is tangent to a foliation if, and only if, for any pair (U, V ) of (global) sections of τ , the Lie bracket [U, V ] is also a section of τ . Let (X1 , . . . , Xm ) be a local basis of τ . Then U=

m

a i Xi

and V =

m

b j Xj .

j =1

i=1

So the bracket [U, V ] can be expressed as [U, V ] =

m i j a b [Xi , Xj ] + a i Xi bj Xj − bj Xj a i Xi . i,j =1

Therefore the value of [U, V ] in ντ = T M/τ at a point x ∈ M depends only on the value of U and V at x. Hence Qτ (U, V ) = π([U, V ]) is a skew-symmetric bilinear map Qτ : τ × τ −→ ντ where π : T M −→ ντ is the canonical projection. In other words, Qτ is a global section of the vector bundle Λ2 (τ, ντ ) of skew-symmetric bilinear forms on the bundle τ . The integrability condition of τ is equivalent to Qτ identically equal to 0. So we get a map Q : C ∞ (G(m)) −→ Σ where Σ is a fibre bundle over G(m) whose fibre over a point σ ∈ G(m) is the infinite-dimensional space Ω 2 (σ, νσ ) of global sections of the bundle Λ2 (τ, ντ ). The space Fol(M, m) of dimension m foliations on M is exactly the set {Q = 0}. It will be equipped with the C ∞ -topology induced by the topology of the Fréchet manifold C ∞ (G(m)) (cf. [160]). Let D be the diffeomorphism group of M; then D acts on C ∞ (G(m)) and the action preserves Fol(M, m). Two foliations F, F ∈ Fol(M, m) are conjugated, if they are in the same orbit of the action of D, that is, there exists ϕ ∈ D such that F = ϕ ∗ (F). Now fix τ in C ∞ (G(m)) and suppose that it is tangent to a foliation F . Then the map Pτ : ϕ ∈ D −→ ϕ ∗ (F) ∈ C ∞ (G(m)) takes its values in Fol(M, m). So we get a sequence of Fréchet manifolds and differentiable maps Q Pτ D −→ C ∞ G(m) −→ Σ. Following R. Hamilton, this sequence is called the nonlinear deformation complex of the foliation F [159]. D EFINITION 5. We say that F is C ∞ -stable if there exist an open neighborhood O of the identity in D and an open neighborhood U of F in Fol(M, m) such that the sequence Pτ

Q

O −→ U −→ Σ is exact, that is, every dimension m foliation F on M, close enough to F in the C ∞ -topology, is conjugated to F by an element of O. An important tool to prove the C ∞ -stability of a foliation is Hamilton’s criterion (cf. [159, p. 47]) that we shall describe. This criterion is based on the implicit function theorem of Nash–Moser which is nicely explained in Hamilton’s paper [160].

Foliations

63

Given a foliation F of dimension m on a compact manifold M, let AkF denote the space of differentiable sections of Λk T F ∗ ⊗ νF . Since νF is a foliated bundle there is a well defined “exterior derivative along the leaves” dF : AkF −→ Ak+1 F given by: dF η(X1 , . . . , Xk+1 ) = (−1)i Xi η(X1 , . . . , Xˆ i , . . . , Xk+1 ) i

+

(−1)i+j η [Xi , Xj ], X1 , . . . , Xˆ i , . . . , Xˆ j , . . . , Xk+1 .

i<j 2 = 0 and thus we obtain a differential complex An easy computation shows that dF dF

dF

dF

dF

0 −→ A0F −→ A1F −→ A2F −→ · · · −→ Am F −→ 0 which is only elliptic along the leaves. Let 0 1 · · · s · · · be an increasing collection of norms (of Sobolev or Hölder type) on the Fréchet space A∗F = AkF . k0

With this notation one has T HEOREM 18 (Hamilton). Assume that there exist continuous linear operators H : A1F −→ A0F and K : A2F −→ A1F fulfilling the following conditions: (i) dF ◦ H + K ◦ dF = id, (ii) there is a fixed number r ∈ N for which we have tame estimates for all s, H (β) Cs βs+r and K(γ ) Cs γ s+r , s s where Cs are positive constants depending only on s. Then the foliation F is C ∞ -stable. Unfortunately Hamilton’s paper is still unpublished. In [93], the authors constructed a class of foliations and, using Hamilton’s criterion, they proved that these foliations are C ∞ -stable. Example 1.3(iv), with some assumptions on the matrix A, is in this class. 9. Some other themes As we have pointed out in the foreword, Foliation Theory is a wide field in Mathematics and so huge to discuss completely here. For this reason we have chosen only some of the themes related to Differential Geometry which is the main topic to which this book is devoted. The nonwarned reader may be inclined to believe that the theory is reduced to this part. Fortunately this is not the case. We devote this section to other themes which are no less important than the above ones.

64

R. Barre and A. El Kacimi Alaoui

9.1. Compact leaves Let M be a connected compact orientable manifold of dimension m + n and F a codimension n foliation on M. A compact leaf of F is a leaf L which is compact as a subset of M. If m = 1 such leaf is a periodic orbit and it describes a stationary state of the dynamical system defined by F . The problem of the existence of compact leaves is highly nontrivial. It was first introduced by H. Poincaré in his studies on limit cycles for ordinary differential equations. One of the famous problems was the Seifert conjecture: Every continuous vector field on the 3-dimensional sphere S3 has a periodic orbit. In 1974, using Denjoy’s example of a vector field with exceptional minimal set on the 2-torus, P. Schweitzer [324] constructed a counterexample in class C 1 . In 1988, J. Harrison [162] gave a C 2 counterexample. Finally in 1993, K. Kuperberg [200] solved completely the problem by constructing in any compact 3-manifold a real analytic vector field without periodic orbit. However, M. Brunella [31] proved that the conjecture is true if the flow is transversely holomorphic; in fact, he established a complete classification of these flows on compact 3-manifolds. The most important result concerning the problem of existence of compact leaves was Novikov’s theorem stated above (Theorem 12). Nothing is known in higher dimensions and the following question is still open: is it true that every codimension one foliation on the odd sphere S2p+1 (where p 2) admits a compact leaf? We say that F is a compact foliation if all leaves are compact. For example, every foliation defined by a locally free action of a connected compact Lie group is a compact foliation. Compact foliations was a theme which interested many people (R. Edwards, K. Millet, D. Sullivan, D. Epstein, E. Vogt, H. Rummler, etc.).

9.2. When is a manifold a leaf ? Let L be a noncompact connected manifold. Does there exist a compact manifold M endowed with a foliation F with a leaf diffeomorphic to L? This question was asked by J. Sondow in [335] where he gave some sufficient conditions on L to be a leaf. J. Cantwell and L. Conlon proved in [46] that every surface is a leaf. Along the same lines, G. Hector and W. Bouma proved in [170] that every noncompact surface can be a leaf of a simple foliation of R3 , i.e. a foliation defined by a submersion R3 −→ R. In [114] E. Ghys observed that the topology of a leaf of a foliation on a compact manifold has to be, in some sense, “recurrent”; then he constructed, for any positive integer d, a noncompact manifold L of dimension d which can not be homeomorphic to any leaf of any foliation on a compact manifold. In [125] he also studies the topology of the generic leaves of a lamination by surfaces on a compact metric space and proved that there exist only six noncompact surfaces which can be realized as leaves: (a) the plane R2 , (b) the cylinder S1 × R, (c) the “Loch-Ness monster”, i.e. the plane with infinitely many handles attached, (d) the “Jacob ladder”,

Foliations

65

(e) the “Cantor tree”, i.e. the sphere S2 with a Cantor set removed, (f) the “flowered Cantor tree”, i.e. the Cantor tree with infinitely many handles attached in all directions.

9.3. Minimal leaves Let M be a Riemannian manifold. Denote by ∇ the covariant derivative associated to the Levi-Civita connection. Let L be a submanifold of M (not necessarily properly embedded). Let x ∈ L and ν a vector field defined on a neighborhood of x and orthogonal to L. For X ∈ Tx L, we set: Wxν (X) = −px (∇X ν), where px : Tx M −→ Tx L is the orthogonal projection. Then Wxν is an endomorphism of the vector space Tx L, symmetric with respect to the induced metric on Tx L; it is called the Weingarten map associated to ν. The trace of Wxν describes the variation at x of the volume element when L moves in the direction of ν. We say that L is minimal, if the trace of Wxν is zero for all vector fields ν orthogonal to L. A foliation F on M is said to be with minimal leaves, if all leaves of F are minimal submanifolds. Given an m-dimensional foliation on a compact manifold M, does there exist a Riemannian metric on M for which the leaves are minimal? This question was discussed by H. Rummler [309] and D. Sullivan [339]. They proved the following criterion: such a metric exists if, and only if, there exists an m-form χ positive on the leaves and relatively closed, namely dχ(X1 , . . . , Xm , Y ) = 0 whenever the vector fields X1 , . . . , Xm are tangent to F . In [149] A. Haefliger proved that the property for F to be with minimal leaves depends only on the transverse structure. He also gave a criterion in terms of transverse invariant currents and used it to give many examples of minimal foliations and nonminimal ones. Suppose now that F is a Riemannian codimension n foliation and denote by v the volume basic form associated to the metric. If F is with minimal leaves then v defines a nonzero class in the basic cohomology H n (M/F). Indeed, let χ be the m-form given by the Rummler–Sullivan criterion. Suppose that v = dβ where β ∈ Ω n−1 (M/F). Then: χ ∧ v = χ ∧ dβ = (−1)m d(χ ∧ β) − dχ ∧ β . But dχ ∧ β = 0 because χ is relatively closed. So χ ∧ v is an exact form. But this is impossible because it is a volume form on the compact orientable manifold M. The converse of this assertion was conjectured by Y. Carrière [51] and proved by X. Masa in [227]. In [112] E. Ghys proved that any Riemannian foliation on a simply connected compact manifold admits a bundle-like metric for which the leaves are minimal. Now let F be a foliation on a compact Riemannian manifold M. We say that F is totally geodesic if every geodesic tangent to a leaf L at a point is tangent to L everywhere. This

66

R. Barre and A. El Kacimi Alaoui

is a special class of foliations with minimal leaves which was studied, for instance, by Y. Carrière, G. Cairns, E. Ghys (see [35,36,33,37,54]). In particular, E. Ghys, in [111], has completely classified all the totally geodesic foliations of codimension one on compact manifolds. References [1] C. Albert and P. Dazord, Théorie des groupoï des symplectiques, II, Pub. Dep. Math. Nouv. Sér. Univ. Claude Bernard, Lyon (1990), 27–99. [2] F. Alcade Cuesta and G. Hector, Feuilletages en surfaces, cycles évaouissants et variétés de Poisson, Monatsh. Math. 124 (3) (1997), 193–213. [3] J. Álvarez-López, Duality in the spectral sequence of a Riemannian foliation, Amer. J. Math. 111 (6) (1989), 905–926. [4] J. Álvarez-López, A finiteness theorem for the spectral sequence of a Riemannian foliation, Illinois J. Math. 33 (1) (1989), 79–92. [5] J. Álvarez-López, The basic component of the mean curvature of Riemannian foliations, Ann. Global Anal. Geom. 10 (1992), 179–194. [6] J. Álvarez-López, Morse inequalities for pseudogroups of local isometries, J. Differential Geom. 37 (3) (1993), 603–638. [7] J. Álvarez-López and P. Tondeur, Hodge decomposition along the leaves of a Riemannian foliation, J. Funct. Anal. 99 (2) (1991), 433–458. [8] C. Anantharman-Delaroche, Purely infinite C ∗ -algebras arising from dynamical systems, Bull. Soc. Math. France 125 (1997), 199–226. [9] J.L. Arraut, A 2-dimensional foliation on R7 , Topology 12 (1973), 243–245. [10] J.L. Arraut and M. Craizer, Foliations of M 3 defined by R2 -actions, Ann. Inst. Fourier 45 (4) (1995), 1018–1118. [11] J.L. Arraut and N.M. Dos Santos, The characteristic mapping of a foliated S1 -bundle, Topology 31 (3) (1992), 545–555. [12] M. Artin, Algebraic spaces, Yale Monograph Study 3 (1971). [13] R. Barre, De quelques aspects de la théorie des Q-variétés différentielles et analytiques, Ann. Inst. Fourier 23 (3) (1973), 227–312. [14] V. Belfi, E. Park and K. Richardson, A Hopf index theorem for foliations, Differential Geom. Appl. 18 (2003), 319–341. [15] M. Belliart and O. Birembaux, Actions localement libres de groupes de Lie résolubles, Ann. Inst. Fourier 44 (5) (1994), 1519–1537. [16] A. Benoist and F. Labourie, Sur les difféomorphismes d’Anosov affines à feuilletages stable et instable, Invent. Math. 111 (2) (1993), 285–308. [17] H. Blas, M. Llabres and A. Reventos, Transverse structures of Lie foliations, J. Math. Soc. Japan 48 (4) (1996), 769–795. [18] R.A. Blumenthal, Transversely homogeneous foliations, Ann. Inst. Fourier 29 (4) (1979), 143–158. [19] C. Bonatti and S. Firmo, Feuilles compactes d’un feuilletage générique en codimension un, Ann. Sci. École Norm. Sup. 27 (4) (1994), 407–461. [20] C. Bonatti and A. Haefliger, Déformations de feuilletages, Topology 29 (2) (1990), 205–229. [21] A. Borel and F. Hirzebruch, Characteristic classes and homogeneous spaces, II, Amer. J. Math. 81 (1959), 315–382. [22] R. Bott, On the topological obstruction to integrability, Sympos. Pure Math., vol. 16, Amer. Math. Soc., Providence, RI (1970), 127–131. [23] R. Bott, Lectures on characteristic classes and foliations, Lecture Notes in Math. 279 (1972). [24] R. Bott, On some formulas for the characteristic classes of group actions, Lecture Notes in Math. 652 (1978), 25–62. [25] R. Bott and A. Haefliger, On characteristic classes of Γ -foliations, Bull. Amer. Math. Soc. 78 (1972), 1039–1044.

Foliations

67

[26] R. Bott and J. Heitsch, A remark on the integral cohomology of BΓq , Topology 11 (1972), 141–146. [27] E. Bouacida, O. Echi and E. Salhi, Feuilletages et topologie spectrale, J. Math. Soc. Japan 522 (2000), 447–464. [28] L.G. Bouma and W.T. Van Est, Manifolds schemes and foliations on the 2-torus and the Klein bottle, I, II, III, Indag. Math. 40 (1978), 313–347. [29] F.G.B. Brito, R. Langevin and H. Rosenberg, Intégrales de courbure sur les variétés feuilletées, J. Differential Geom. 16 (1981), 19–50. [30] M. Brunella, Remarks on structurally stable proper foliations, Math. Proc. Cambridge Philos. Soc. 115 (1994), 111–120. [31] M. Brunella, On transversely holomorphic flows, I, Invent. Math. 126 (2) (1996), 265–279. [32] M. Brunella and E. Ghys, Umbilical foliations and transversely holomorphic flows, J. Differential Geom. 41 (1) (1995), 1–19. [33] G. Cairns, Aspects cohomologiques des feuilletages totalement géodésiques, C. R. Acad. Sci. Paris Sér. I 299 (20) (1984), 1017–1019. [34] G. Cairns, A general description of a totally geodesic foliation, Tohoku Math. J. (2) 38 (1986), 37–55. [35] G. Cairns, Some properties of a cohomology group associated to a totally geodesic foliation, Math. Z. 192 (1986), 391–403. [36] G. Cairns, Feuilletages totalement géodésiques sur les variétés simplement connexes, Sém. Sud Rhodanien Géom. VII Trav. Cours 26 (1988), 1–14. [37] G. Cairns and E. Ghys, Totally geodesic foliations on 4-manifolds, J. Differential Geom. 23 (1986), 241– 254. [38] C. Camacho, Structural stability theorems for integrable differential forms on 3-manifolds, Topology 17 (1978), 143–155. [39] C. Camacho and A. Lins Neto, Geometric Theory of Foliations, Birkhäuser (1985). [40] J. Cantwell and L. Conlon, Growth of leaves, Comment. Math. Helv. 53 (1978), 93–111. [41] J. Cantwell and L. Conlon, Reeb stability for noncompact leaves in foliated 3-manifolds, Proc. Amer. Math. Soc. 83 (1981), 408–410. [42] J. Cantwell and L. Conlon, Poincaré–Bendixon theory for leaves of codimension one, Trans. Amer. Math. Soc. 265 (1981), 181–209. [43] J. Cantwell and L. Conlon, Tischler fibrations of open foliated sets, Ann. Inst. Fourier 31 (2) (1981), 113–135. [44] J. Cantwell and L. Conlon, The endsets of leaves, Topology 21 (1982), 333–352. [45] J. Cantwell and L. Conlon, Analytic foliations and the theory of levels, Math. Ann. 265 (1983), 253–261. [46] J. Cantwell and L. Conlon, Every surface is a leaf, Topology 26 (1987), 265–285. [47] J. Cantwell and L. Conlon, The theory of levels, Contemp. Math. 70 (1988), 1–10. [48] J. Cantwell and L. Conlon, Leafwise hyperbolicity of proper foliations, Comment. Math. Helv. 64 (1989), 329–337. [49] J. Cantwell and L. Conlon, Generic leaves, Comment. Math. Helv. 73 (1998), 306–336. [50] P. Caron and Y. Carrière, Flots transversalement de Lie Rn , flots minimaux, C. R. Acad. Sci. Paris 280 (1980), 477–478. [51] Y. Carrière, Flots riemanniens, Structures transverses des feuilletages, Toulouse 1982, Astérisque 116 (1984), 31–52. [52] Y. Carrière, Feuilletages riemanniens à croissance polynomiale, Comment. Math. Helv. 63 (1984), 1–20. [53] Y. Carrière, Les propriétés topologiques des flots riemanniens retrouvées à l’aide du théorème des variétés presque plates, Math. Z. 186 (1984), 393–400. [54] Y. Carrière and E. Ghys, Feuilletages totalement géodésiques, An. Acad. Brasil. Ciênc. 53 (1981), 427– 432. [55] J. Cerf, 1-formes fermées non singulières sur les variétés compactes de dimension 3, Sém. Bourbaki, Exposé 574, Lecture Notes in Math. 901 (1981), 205–219. [56] D. Cerveau, Feuilletages holomorphes de codimension un. Réduction des singularités en petites dimensions et applications, IRM de Rennes, Prépublication 98-01 (1998). [57] G. Chatelet, Sur les feuilletages induits par l’action de groupes de Lie nilpotents, Ann. Inst. Fourier 27 (2) (1977), 161–189.

68

R. Barre and A. El Kacimi Alaoui

[58] G. Chatelet and H. Rosenberg, Un théorème de conjugaison des feuilletages, Ann. Inst. Fourier 21 (3) (1971), 95–106. [59] J. Chevalley, Theory of Lie Groups, Princeton Mathematical Series, vol. 8 (1946). [60] L. Conlon, Transversally parallelizable foliations of codimension two, Trans. Amer. Math. Soc. 194 (1975), 79–102. [61] A. Connes, Sur la théorie non commutative de l’intégration, Lecture Notes in Math. 725 (1979). [62] A. Connes, Noncommutative Geometry, Academic Press (1994). [63] A. Connes and H. Moscovici, Hopf algebras, cyclic cohomology and the transverse index theorem, Commun. Math. Phys. 198 (1998), 199–246. [64] A. Connes and H. Moscovici, Differentiable cyclic cohomology and Hopf algebraic structures in transverse geometry, Essays on Geometry and Related Topics, vols. I, II, Monogr. Enseign. Math., vol. 38, Geneva (2001), 217–255. [65] M. Craioveanu and M. Puta, Cohomology on a Riemannian foliated manifold with coefficients in the sheaf of germs of foliated currents, Math. Nachr. 99 (1980), 43–53. [66] C. Cumenge, Cohomologie de l’espace des feuilles d’un feuilletage, Thèse 3ème cycle, Toulouse (1981). [67] C. Curras-Bosch and P. Molino, Voisinage d’une feuille compacte dans un voisinage lagrangien, Hokkaido Math. J. 23 (2) (1994), 355–360. [68] A. Dadarlat, C ∗ -Algebras and applications, Contemp. Math. 145 (1993), 393–421. [69] C. Danthony, Feuilletages orientés de surfaces; le problème de la section globale, Ann. Inst. Fourier 38 (2) (1988), 201–227. [70] A. Denjoy, Sur les courbes définies par les équations différentielles à la surface du tore, J. Math. Pures Appl. 11 (1932), 333–375. [71] P.R. Dipolito, Codimension one foliations of closed manifolds, Comment. Math. Helv. 107 (1978), 403– 453. [72] T. Duchamp and M. Kalka, Deformation theory for holomorphic foliations, J. Differential Geom. 14 (1979), 317–337. [73] G. Duminy, L’invariant de Godbillon–Vey d’un feuilletage se localise sur les feuilles ressort, Manuscrit Université de Lille I (1982). [74] G. Duminy, Sur les cycles feuilletées de codimension un, Manuscrit Université de Lille I (1982). [75] G. Duminy and V. Sergiescu, Sur la nullité de l’invariant de Godbillon–Vey, C. R. Acad. Sci. Paris 292 (1981), 821–824. [76] J.L. Dupont, Simplicial de Rham cohomology and characteristic classes of flat bundles, Topology 15 (1976), 233–245. [77] A.H. Durfee, Foliations of odd-dimensional spheres, Ann. of Math. 96 (2) (1972), 407–411. [78] A.H. Durfee and H.B. Lawson, Fibered knots and foliations of highly connected manifolds, Invent. Math. 17 (1972), 203–215. [79] R. Edwards, K. Millet and D. Sullivan, Foliations with all leaves compact, Topology 16 (1977), 13–32. [80] C. Ehresmann, Les connexions infinitésimales dans un espace fibré différentiable, Colloque de Topologie, Bruxelles (1950), 29–53. [81] C. Ehresmann, Sur la théorie des variétés feuilletées, Rend. Mat. Appl. (5) 10 (1951), 64–82. [82] C. Ehresmann, Structures feuilletées, Proc. 5th Canadian Congress, Montréal (1961), 109–172. [83] C. Ehresmann and G. Reeb, Sur les champs d’éléments de contact de dimension p complètement intégrables sur une variété continûment différentiable, C. R. Acad. Sci. Paris 218 (1944), 955–957. [84] C. Ehresmann and S. Weishu, Sur les espaces feuilletées et théorème de stabilité, C. R. Acad. Sci. Paris 243 (1956), 344–346. [85] A. El Kacimi Alaoui, Opérateurs transversalement elliptiques sur les feuilletages riemanniens et applications, Compositio Math. 73 (1990), 57–106. [86] A. El Kacimi Alaoui and E. Gallego, Applications harmoniques feuilletées, Illinois J. Math. 40 (1) (1996), 115–122. [87] A. El Kacimi Alaoui and B. Gmira, Stabilité du caractère kählérien transverse, Israel J. Math. 101 (1997), 323–347. [88] A. El Kacimi Alaoui, G. Guasp and M. Nicolau, On deformations of transversely homogeneous foliations, Topology 40 (2001), 1363–1393.

Foliations

69

[89] A. El Kacimi Alaoui and G. Hector, Décomposition de Hodge basique pour un feuilletage riemannien, Ann. Inst. Fourier 36 (3) (1986), 207–227. [90] A. El Kacimi Alaoui and M. Nicolau, Déformations des feuilletages transversalement holomorphes à type différentiable fixé, Publ. Mat. Barcelona 33 (1989), 485–500. [91] A. El Kacimi Alaoui and M. Nicolau, Structures géométriques invariantes et feuilletages de Lie, Indag. Math. (N.S.) 1 (3) (1990), 323–334. [92] A. El Kacimi Alaoui and M. Nicolau, On the topological invariance of the basic cohomology, Math. Ann. 295 (1993), 627–634. [93] A. El Kacimi Alaoui and M. Nicolau, A class of C ∞ -stable foliations, Ergodic Theory Dynam. Systems 13 (1993), 697–704. [94] A. El Kacimi Alaoui, V. Sergiescu and G. Hector, La cohomologie basique d’un feuilletage riemannien est de dimension finie, Math. Z. 188 (1985), 593–599. [95] D.B.A. Epstein, Periodic flows on 3-manifolds, Ann. of Math. 95 (1972), 68–82. [96] D.B.A. Epstein, Foliations with all leaves compact, Ann. Inst. Fourier 26 (1) (1976), 265–282. [97] D.B.A. Epstein, K. Millet and D. Tischler, Leaves without holonomy, J. London Math. Soc. (2) 16 (1977), 548–552. [98] D.B.A. Epstein and E. Vogt, A counterexemple to the periodic orbit conjecture in codimension 3, Ann. of Math. (2) 108 (1978), 539–552. [99] T. Fack and G. Skandalis, Sur les représentations et idéaux de la C ∗ -algèbre d’un feuilletage, J. Operator Theory 8 (1982), 95–129. [100] F.T. Farrell and L.E. Jones, Foliated control theory, I, II, K-Theory 2 (1988), 357–399 and 401–430. [101] E. Fédida, Sur l’existence des feuilletages de Lie, C. R. Acad. Sci. Paris 278 (1974), 835–837. [102] D. Ferus, Totally geodesic foliations, Math. Ann. 188 (1970), 313–316. [103] H.R. Fisher and F.L. Williams, Complex foliated structures. I. Cohomology of the Dolbeault–Kostant complexes, Trans. Amer. Math. Soc. 252 (1979), 163–195. [104] D.B. Fuks, Characteristic classes of foliations, Russian Math. Surveys 28 (2) (1973), 1–16. [105] E. Gallego and A. Reventos, Lie flows of codimension 3, Trans. Amer. Math. Soc. 326 (2) (1991), 529–541. [106] R.B. Gardner, The Godbillon–Vey invariant and the Bott–Pasternak vanishing theorem, Lecture Notes in Math. 652 (1978), 75–94. [107] L. Garnett, Foliations, the ergodic theorem and Brownian motion, J. Funct. Anal. 51 (1983), 285–311. [108] I.M. Gelfand, B.L. Feigin and D.B. Fuks, Cohomology of the Lie algebra of formal vector fields with coefficients in its adjoint space and variations of characteristic classes of foliations, Funct. Anal. Appl. 8 (1974), 99–112. [109] I.M. Gelfand and D.B. Fuks, PL-foliations, Funct. Anal. Appl. 7 (1973), 278–284 and 8 (1974), 197–200. [110] R. Gerard and A. Sec, Feuilletages de Painlevé, Bull. Soc. Math. France 100 (1972), 47–72. [111] E. Ghys, Classification des feuilletages totalement géodésiques de codimension un, Comment. Math. Helv. 58 (1983), 543–572. [112] E. Ghys, Feuilletages riemanniens sur les variétés simplement connexes, Ann. Inst. Fourier 34 (4) (1984), 203–223. [113] E. Ghys, Actions localement libres du groupe affine, Invent. Math. 82 (1985), 479–526. [114] E. Ghys, Une variété qui n’est pas une feuille, Topology 24 (1985), 67–73. [115] E. Ghys, Classe d’Euler et minimal exceptionnel, Topology 24 (1985), 67–73. [116] E. Ghys, Flots d’Anosov dont les feuilletages stables sont différentiables, Ann. Sci. École Norm. Sup. 20 (1987), 251–270. [117] E. Ghys, Sur l’invariance topologique de la classe de Godbillon–Vey, Ann. Inst. Fourier 37 (4) (1987), 59–76. [118] E. Ghys, Gauss–Bonnet theorem for 2-dimensional foliations, J. Funct. Anal. 77 (1) (1988), 51–59. [119] E. Ghys, L’invariant de Godbillon–Vey, Sém. Bourbaki, Exposé 706, Astérisque 177–178 (1989), 155– 181. [120] E. Ghys, Déformations des flots d’Anosov et groupes fuchsiens, Ann. Inst. Fourier 42 (1–2) (1992), 209– 247. [121] E. Ghys, Rigidité différentiable des groupes fuchsiens, Publ. IHES 78 (1993), 163–185. [122] E. Ghys, Construction de champs de vecteurs sans orbites périodiques, Sém. Bourbaki, Exposé 785, Astérisque 227 (1993–1994), 283–307.

70 [123] [124] [125] [126] [127] [128] [129] [130] [131] [132] [133] [134] [135] [136] [137] [138] [139] [140] [141] [142] [143] [144] [145] [146] [147] [148] [149] [150] [151] [152] [153] [154] [155] [156] [157]

R. Barre and A. El Kacimi Alaoui E. Ghys, Holomorphic Anosov systems, Invent. Math. 199 (3) (1995), 585–614. E. Ghys, Groupes d’holonomie des feuilletages de Lie, Indag. Math. 47 (1995), 173–182. E. Ghys, Topologie des feuilles génériques, Ann. of Math. (2) 141 (2) (1995), 387–422. E. Ghys, On transversely holomorphic flows, Invent. Math. 126 (2) (1996), 281–286. E. Ghys, R. Langevin and P. Walczak, Entropie géométrique des feuilletages, Acta Math. 16 (1–2) (1988), 105–142. E. Ghys and V. Sergiescu, Stabilité et conjugaison différentiable pour certains feuilletages, Topology 26 (1987), 93–105. E. Ghys and E. Sergiescu, Sur un groupe remarquable de difféomorphismes du cercle, Comment. Math. Helv. 62 (1987), 185–239. E. Ghys and T. Tsuboi, Différentiabilité des conjugaisons entre systèmes dynamiques de dimension un, Ann. Inst. Fourier 38 (1) (1988), 215–244. E. Ghys and A. Verjovsky, Holomorphic actions of the affine group, Proceedings of Geometric Study of Foliations, Tokyo, World Scientific (1994), 201–217. J. Girbau, A. Haefliger and D. Sundararaman, On deformations of transversely holomorphic foliations, J. Reine Angew. Math. 345 (1983), 122–147. J. Girbau and M. Nicolau, On deformations of holomorphic foliations, Ann. Inst. Fourier 39 (2) (1989), 417–449. C. Godbillon, Feuilletages ayant la propriété du prolongement des homotopies, Ann. Inst. Fourier 17 (2) (1967), 219–260. C. Godbillon, Feuilletages. Etudes géométriques, Progress in Mathematics, vol. 98, Birkhäuser, Basel (1991). C. Godbillon and J. Vey, Un invariant des feuilletages de codimension un, C. R. Acad. Sci. Paris 273 (1971), 92–95. X. Gomez Mont, Transversal holomorphic structures, J. Differential Geom. 15 (1980), 161–185. S. Goodman, On the structure of foliated 3-manifolds separated by a compact leaf, Invent. Math. 39 (1977), 213–221. S. Goodman, Vector fields with transverse foliations, Topology 24 (1985), 333–340. S. Goodman and J.F. Plante, Holonomy and averaging in foliated sets, J. Differential Geom. 14 (1979), 401–407. M. Gromov, Stable mappings of foliations into manifolds, Math. USSR Izv. 3 (1969), 671–694. M. Gromov, Transversal mappings of foliations, Soviet Math. Dokl. 9 (1969), 1126–1129. A. Haefliger, Structures feuilletées et cohomologie à valeurs dans un faisceau de groupoï des, Comment. Math. Helv. 32 (1958), 248–329. A. Haefliger, Variétés feuilletées, Ann. Scuola Nom. Sup. Pisa (3) 16 (1962), 367–397. A. Haefliger, Feuilletages sur les variétés ouvertes, Topology 9 (1970), 183–194. A. Haefliger, Homotopy, integrability, Lecture Notes in Math. 197 (1971), 133–163. A. Haefliger, Sur les classes caracteristiques des feuilletages, Sém. Bourbaki 1971/1972, Exposé 412, Lecture Notes in Math. 317 (1973), 239–260. A. Haefliger, Cohomology of Lie algebras and foliations, Lecture Notes in Math. 652 (1978), 1–12. A. Haefliger, Some remarks on foliations with minimal leaves, J. Differential Geom. 15 (1980), 269–284. A. Haefliger, Groupoï des d’holonomie et classifiants, Astérisque 116 (1984), 98–107. A. Haefliger, Pseudo-groups of local isometries, Differential Geometry, Santiago de Compostela, 1984, L. Cordero, ed., Research Notes, vol. 131, Pitman (1985), 174–197. A. Haefliger, Deformations of transversely holomorphic flows on spheres and deformations of Hopf fibrations, Compositio Math. 55 (1985), 241–251. A. Haefliger, Complexification of transversely holomorphic foliations, Math. Ann. 272 (1985), 23–27. A. Haefliger, Feuilletages riemanniens, Sém. Bourbaki 1998/1999, Exposé 707, Astérisque 177–178 (1989), 183–197. A. Haefliger, Travaux de Novikov sur les feuilletages, Sém. Bourbaki 1967/1968, Exposé 339. A. Haefliger and Quach Ngoc Du, Une présentation du groupe fondamental d’une orbifold, Astérisque 116 (1984), 98–107. A. Haefliger and E. Salem, Pseudogroupes d’holonomie des feuilletages riemanniens sur des variétés compactes 1-connexes, Colloq. Geom. Phys. Paris/F (1986), Travaux en cours 33 (1988), 141–160.

Foliations

71

[158] A. Haefliger and E. Salem, Riemannian foliations on simply connected manifolds and actions of tori on orbifolds, Illinois J. Math. 34 (4) (1990), 706–730. [159] R.S. Hamilton, Deformation theory of foliations, Preprint Cornell University, New York (1978). [160] R.S. Hamilton, The inverse function theorem of Nash and Moser, Bull. Amer. Math. Soc. 7 (1982), 65–222. [161] Y. Hantout, Classes caractéristiques de (G, H )-structures et finitude de leur évaluation, Manuscripta Math. 62 (1988), 383–399. [162] J. Harrison, A C 2 counterexample to the Seifert conjecture, Topology 27 (1988), 249–278. [163] G. Hector, Sur les feuilletages presque sans holonomie, C. R. Acad. Sci. Paris Sér. A 274 (1972), 1703– 1706. [164] G. Hector, Ouverts incompressibles et structures des feuilletages de codimension un, C. R. Acad. Sci. Paris Sér. A 274 (1972), 741–744. [165] G. Hector, Quelques exemples de feuilletages. Espèces rares, Ann. Inst. Fourier 26 (1) (1976), 239–264. [166] G. Hector, Feuilletages en cylindres, Lecture Notes in Math. 597 (1976), 252–270. [167] G. Hector, Leaves whose growth is neither exponential nor polynomial, Topology 16 (1977), 449–451. [168] G. Hector, Croissance des feuilletages presque sans holonomie, Lecture Notes in Math. 652 (1978), 141– 182. [169] G. Hector, Architecture des feuilletages de classe C 2 , Astérisque 107–108 (1983), 243–258. [170] G. Hector and W. Bouma, All open surfaces are leaves of simple foliations of R3 , Indag. Math. 45 (1983), 443–452. [171] G. Hector, E. Ghys and Y. Moriyama, On codimension one foliations and a theorem of Malcev, Topology 28 (2) (1989), 197–210. [172] G. Hector and U. Hirsch, Introduction to the Geometry of Foliations, Parts A and B, Vieweg & Sohn (1981). [173] G. Hector and E. Macias, Sur le théorème de de Rham pour les feuilletages de Lie, C. R. Acad. Sci. Paris Sér. I 311 (10) (1990), 633–636. [174] J.L. Heitsch, A cohomology of foliated manifolds, Comment. Math. Helv. 50 (1975), 197–218. [175] J.L. Heitsch, Residues and characteristic classes of foliations, Bull. Amer. Math. Soc. 83 (1977), 397–399. [176] J.L. Heitsch, Independent variation of secondary classes, Ann. of Math. (2) 108 (3) (1978), 421–460. [177] J.L. Heitsch, Flat bundles and residues for foliations, Invent. Math. 73 (1983), 271–285. [178] J.L. Heitsch and S. Hurder, Secondary classes, Weil measures and the geometry of foliations, J. Differential Geom. 20 (1984), 291–309. [179] J.L. Heitsch and H.B. Lawson, Transgressions, Chern–Simons invariants and the classical groups, J. Differential Geom. 9 (1974), 423–434. [180] J.L. Heitsch and C. Lazarov, Lefschetz theorem for foliated manifolds, Topology 29 (2) (1990), 127–162. [181] J.L. Heitsch and C. Lazarov, Homotopy invariance of foliation Betti numbers, Invent. Math. 104 (2) (1991), 321–347. [182] J.L. Heitsch and C. Lazarov, Rigidity theorems by surfaces and spin manifolds, Michigan Math. J. 38 (2) (1991), 285–297. [183] J.L. Heitsch and C. Lazarov, Spectral asymptotics of foliated manifolds, Illinois J. Math. 38 (4) (1994), 653–678. [184] M. Herman, The Godbillon–Vey invariant of foliations by planes of T3 , Lecture Notes in Math. 597 (1977), 294–307. [185] M. Herman, Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations, Publ. Math. IHES 49 (1979), 5–223. [186] R. Hermann, The differential geometry of foliations, I, Ann. of Math. 72 (2) (1960), 445–457. [187] R. Hermann, The differential geometry of foliations, II, J. Math. Mech. 11 (1962), 305–315. [188] M. Hirsch and W. Thurston, Foliated bundles, invariant measures and flat manifolds, Ann. of Math. 101 (1975), 369–390. [189] S. Hurder, Problems on rigidity of group actions and cocycles, Ergodic Theory Dynam. Systems 5 (1985), 473–484. [190] S. Hurder, The Godbillon measure of amenable foliations, J. Differential Geom. 23 (1986), 347–365. [191] S. Hurder, Rigidity for Anosov actions of higher lattices, Ann. of Math. 135 (1992), 361–410. [192] S. Hurder and A. Katok, Ergodic theory and Weil measures for foliations, Ann. of Math. 126 (1987), 221–275.

72

R. Barre and A. El Kacimi Alaoui

[193] S. Hurder and Y. Mitsumatsu, On the intersection products of transverse invariant measures, Indiana Univ. Math. J. 40 (1992), 1169–1183. [194] J.-P. Jouanolou, Une preuve élémentaire d’un théorème de Thurston, Topology 17 (1978), 109–110. [195] F. Kamber and P. Tondeur, Foliated bundles and characteristic classes, Lecture Notes in Math. 493 (1975). [196] F. Kamber and P. Tondeur, Duality for Riemannian foliations, Proc. Sympos. Pure Math. 40 (1983), 609– 618. [197] F. Kamber and P. Tondeur, Hodge de Rham theory for Riemannian foliations, Math. Ann. 277 (1987), 415–431. [198] M. Kellum, Uniformly quasi-isometric foliations, Ergodic Theory Dynam. Systems 13 (1993), 101–122. [199] K. Kodaira and D.C. Spencer, Multifoliate structures, Ann. of Math. 74 (1961), 52–100. [200] K. Kuperberg, A smooth counterexample to the Seifert conjecture, Ann. of Math. 140 (1994), 723–732. [201] C. Lamoureux, Holonomie et feuilles exceptionnelles, Ann. Inst. Fourier 26 (4) (1976), 273–300. [202] R. Langevin and H. Rosenberg, On stability of compact leaves and fibrations, Topology 16 (1977), 107– 111. [203] F. Laudenbach, Formes différentielles de degré 1 fermées non singulières et classes d’homotopie de leurs noyaux, Comment. Math. Helv. 51 (1971), 447–464. [204] F. Laudenbach and R. Roussarie, Un exemple de feuilletage sur S3 , Topology 9 (1970), 63–70. [205] H.B. Lawson, The Quantitative Theory of Foliations, Regional Conference Series in Mathematics, vol. 27, Amer. Math. Soc. (1970). [206] H.B. Lawson, Codimension one foliations of spheres, Ann. of Math. 94 (2) (1971), 494–503. [207] H.B. Lawson, Foliations, Bull. Amer. Math. Soc. 80 (1974), 369–418. [208] C. Lazarov, A vanishing theorem for certain rigid classes, Michigan Math. J. 28 (1981), 89–95. [209] C. Lazarov, An index theorem for foliations, Illinois J. Math. 3 (1) (1986), 101–121. [210] C. Lazarov and J.S. Pasternak, Residues and characteristic classes for Riemannian foliations, J. Differential Geom. 11 (1976), 599–612. [211] D. Lehmann, Sur l’approximation de certains feuilletages nilpotents par des fibrations, C. R. Acad. Sci. Paris 286 (1978), 251–254. [212] J. Lehmann-Lejeune, Cohomologie sur le fibré transverse à un feuilletage, C. R. Acad. Sci. Paris 295 (1982), 495–498. [213] G. Levitt, Foliations and laminations in hyperbolic surfaces, Topology 22 (1983), 119–135. [214] G. Levitt, Groupe fondamental de l’espace des feuilles dans les feuilletages sans holonomie, J. Differential Geom. 31 (3) (1990), 711–761. [215] A. Lichnerowicz, Algèbres de Lie attachées à un feuilletage, Ann. Fac. Sci. Toulouse I (1979), 45–76. [216] W.B.R. Lickorish, A foliation for 3-manifolds, Ann. of Math. 82 (2) (1965), 414–420. [217] J.-J. Lœb and M. Nicolau, Holomorphic flows and complex structures on products of odd dimensional spheres, Math. Ann. 306 (1996), 781–817. [218] D. Mac Duff, On groups of volume preserving diffeomorphisms and foliations with transverse volume form, Proc. London Math. Soc. 43 (1981), 295–320. [219] G.W. Mackey, Ergodic theory and virtual groups, Math. Ann. 166 (1956), 359–363. [220] T. Matsuaka and S. Morita, On characteristic classes of Kähler foliations, Osaka J. Math. 16 (1979), 539–550. [221] S. Matsumoto, Some remarks on foliated S 1 -bundles, Invent. Math. 90 (1987), 343–358. [222] S. Matsumoto, Perturbation of the Hopf flow and transverse foliations, Nagoya Math. J. 122 (1991), 75– 82. [223] S. Matsumoto, Lie affine foliations on 4-manifolds, Invent. Math. 109 (1) (1992), 1–16. [224] S. Matsumoto, Codimension one foliations on solvable manifolds, Comment. Math. Helv. 68 (4) (1993), 633–652. [225] X. Masa, Quelques propriétés des feuilletages de codimension un à connexion transverse projetable, C. R. Acad. Sci. Paris Sér. A 284 (1977), 811–812. [226] X. Masa, Characteristic classes of subfoliations, Ann. Inst. Fourier 31 (2) (1981), 61–86. [227] X. Masa, Duality and minimality in Riemannian foliations, Comment. Math. Helv. 67 (1992), 17–27. [228] G. Meigniez, Bouts d’un groupe opérant sur la droite. Applications à la topologie des feuilletages, Tohoku Math. J. (2) 43 (4) (1991), 473–500.

Foliations

73

[229] G. Meigniez, On completeness of transversely projective foliations, Topology 31 (2) (1992), 421–432. [230] G. Meigniez, Feuilletages de Lie résolubles, Ann. Fac. Sci. Toulouse (6) 4 (1995), 801–817. [231] G. Meigniez, Prolongement des homotopies, Q-variétés et cycles tangents, Ann. Inst. Fourier 47 (3) (1997), 945–965. [232] A. Miernowski and W. Mozgawa, On Molino lifting of Riemannian vector fields, Acta Math. Hungar. 55 (1990), 185–191. [233] J. Milnor, On the existence of a connection with curvature zero, Comment. Math. Helv. 32 (1958), 215– 223. [234] Y. Mitsumatsu, A relation between the topological invariance of the Godbillon–Vey invariant and the differentiability of Anosov foliations, Foliations, Adv. Stud. Pure Math., vol. 5 (1985), 159–167. [235] Y. Mitsumatsu, On the self intersection of foliation cycles, Trans. Amer. Math. Soc. 334 (2) (1992), 1169– 1183. [236] T. Mizutani, S. Morita and T. Tsuboi, The Godbillon–Vey classes of codimension one foliations which are almost without holonomy, Ann. of Math. (2) 113 (1981), 515–527. [237] T. Mizutani, S. Morita and T. Tsuboi, On the cobordism classes of codimension one foliations which are almost without holonomy, Topology 22 (1983), 325–343. [238] I. Moerdijk and D.A. Pronk, Orbifolds, sheaves and groupoids, K-Theory 12 (1) (1997), 3–21. [239] I. Moerdijk and D.A. Pronk, Simplicial cohomology of orbifolds, Indag. Math. (N.S.) 10 (2) (1999), 269– 293. [240] P. Molino, Propriétés cohomologiques et propriétés topologiques des feuilletages à connexion transverse projetable, Topology 12 (1973), 317–325. [241] P. Molino, Sur la géométrie transverse des feuilletages, Ann. Inst. Fourier 25 (1975), 279–284. [242] P. Molino, Etude des feuilletages transversalement complets et applications, Ann. Sci. École Norm. Sup. 10 (1977), 289–307. [243] P. Molino, Géométrie globale des feuilletages riemanniens, Indag. Math. 44 (1982), 45–76. [244] P. Molino, Feuilletages riemanniens, Astérisque 116 (1984). [245] P. Molino, Riemannian Foliations, Birkhäuser (1988). [246] P. Molino, Dualité symplectique, feuilletages et géométrie du moment, Publ. Mat. Barcelona 33 (3) (1989), 533–541. [247] S. Morita, On characteristic classes of Riemannian foliations, Osaka J. Math. 16 (1979), 161–172. [248] R. Moussu, Feuilletages sans holonomie d’une variété fermée, C. R. Acad. Sci. Paris Sér. A 270 (1970), 1308–1311. [249] R. Moussu, Feuilletages presque sans holonomie, C. R. Acad. Sci. Paris Sér. A 272 (1971), 114–117. [250] R. Moussu, Sur les classes exotiques des feuilletages, Lecture Notes in Math. 392 (1974), 37–42. [251] R. Moussu and F. Pelletier, Sur le théorème de Poincaré–Bendixon, Ann. Inst. Fourier 24 (1) (1974), 131–148. [252] R. Moussu and R. Roussarie, Relations de conjugaison et de cobordisme entre certains feuilletages, Publ. Math. IHES 43 (1973), 143–168. [253] W. Mozgawa, Feuilletages de Killing, Collect. Math. 36 (3) (1985), 285–290. [254] M.P. Muller, Sur les composantes de Novikov des feuilletages, Topology 19 (1980), 199–201. [255] Ngo Van Que and R. Roussarie, Sur l’isotopie des formes fermées en dimension 3, Invent. Math. 64 (1981), 69–87. [256] M. Nicolau and A. Reventos, On some geometrical properties of Seifert bundles, Israel J. Math. 47 (1984), 323–334. [257] S. Nishikawa, Residues and secondary characteristic classes for projective foliations, Proc. Japan Acad. 54 (1978), 79–82. [258] S. Nishikawa and H. Sato, On characteristic classes of Riemannian, conformal and projective foliations, J. Math. Soc. Japan 28 (1976), 223–241. [259] S. Nishikawa and M. Takeushi, G-foliations and semisimple flat homogeneous spaces, Tohoku Math. J. 30 (1978), 307–335. [260] T. Nishimori, SRH-decompositions of codimension one foliations and the Godbillon–Vey class, Tohoku Math. J. 32 (1980), 9–34. [261] S.P. Novikov, Foliations of codimension one, Soviet Math. Dokl. 5 (1964), 540–544 and 1023–1025. [262] S.P. Novikov, The topology of foliations, Trans. Moscow Math. Soc. 14 (1965), 268–304.

74

R. Barre and A. El Kacimi Alaoui

[263] J. Palis, Rigidity of the centralizers and structural stability of suspended foliations, Lecture Notes in Math. 652 (1978), 114–121. [264] C.F.B. Palmeira, Open manifolds foliated by planes, Ann. of Math. 107 (1978), 109–131. [265] E. Park and K. Richardson, The basic Laplacian for Riemannian foliations, Amer. J. Math. 118 (1996), 1249–1275. [266] A. Phillips, Foliations on open manifolds, Comment. Math. Helv. 43 (1968), 204–211 and 44 (1969), 367–370. [267] H.P. Pittie, The secondary characteristic classes of parabolic foliations, Comment. Math. Helv. 54 (1979), 601–614. [268] M. Plaisant, Sur l’intégrabilité des algèbres de Lie banachiques dans le cadre des Q-variétés, Thèse 3ème cycle, Valenciennes (1980). [269] J.F. Plante, Anosov flows, Amer. J. Math. 94 (1972), 729–754. [270] J.F. Plante, Asymptotic properties of foliations, Comment. Math. Helv. 47 (1972), 449–456. [271] J.F. Plante, A generalization of the Poincaré–Bendixon theorem for foliations of codimension one, Topology 12 (1973), 177–181. [272] J.F. Plante, Foliations transverse to fibers of a bundle, Proc. Amer. Math. Soc. 42 (1974), 631–635. [273] J.F. Plante, Foliations with measure preserving holonomy, Ann. of Math. 102 (1975), 327–361. [274] J.F. Plante, Foliations of 3-manifolds with solvable fundamental group, Invent. Math. 51 (1979), 219–230. [275] J.F. Plante, Stability of codimension one foliations by compact leaves, Topology 22 (1983), 173–177. [276] J.F. Plante, Polycyclic groups and transversely affine foliations, J. Differential Geom. 36 (3) (1992), 521– 534. [277] J.F. Plante and W.P. Thurston, Polynomial growth in holonomy groups of foliations, Comment. Math. Helv. 51 (1976), 567–584. [278] J. Pradines, Sur une classe remarquable de relations d’équivalence sur les variétés, Rev. An. de Math. Jaca (1973). [279] J. Pradines, Un feuilletage sans holonomie transversale dont le quotient n’est pas une Q-variété, C. R. Acad. Sci. Paris 288 (1979), 245–248. [280] J. Pradines and B. Bigonnet, Graphe d’un feuilletage singulier, C. R. Acad. Sci. Paris 300 (1985), 439– 442. [281] J. Pradines and J. Wouafo-Kamga, La catégorie des QF -variétés, C. R. Acad. Sci. Paris Sér. A 288 (1979), 717–719. [282] M. Puta, Cohomology classes and foliated manifolds, Nonlinear Anal. (1987), 137–159. [283] A. Ramsay, Virtual groups and group actions, Adv. in Math. 6 (1971), 253–322. [284] O.H. Rasmussen, Exotic characteristic classes for holomorphic foliations, Invent. Math. 46 (1978), 153– 171. [285] G. Reeb, Sur certaines propriétés topologiques des variétés feuilletées, Hermann, Paris (1950). [286] G. Reeb, Sur les structures feuilletées de codimension un et sur un théorème de Denjoy, Ann. Inst. Fourier 11 (1961), 185–200. [287] G. Reeb, Feuilletages résultats anciens et nouveaux, Painlevé, Hector and Martinet, eds, Presses Univ. de Montréal (1974). [288] B.L. Reinhart, Harmonic integrals on almost product manifolds, Trans. Amer. Math. Soc. 88 (1958), 243– 276. [289] B.L. Reinhart, Foliated manifolds with bundle-like metrics, Ann. of Math. 69 (1959), 119–132. [290] B.L. Reinhart, Harmonic integrals on foliated manifolds, Amer. J. Math. 81 (1959), 529–586. [291] B.L. Reinhart, Differential Geometry of Foliations, Ergebnisse der Mathematik, vol. 99, Springer (1983). [292] B.L. Reinhart and J.W. Wood, A metric formula for the Godbillon–Vey invariant for foliations, Proc. Amer. Math. Soc. 38 (1973), 427–430. [293] J. Renault, A Groupoid Approach to C ∗ -Algebras, Lecture Notes in Math., vol. 793, Springer (1980). [294] K. Richardson, Asymptotic of heat kernels on Riemannian foliations, Geom. Funct. Anal. 8 (1998), 356– 401. [295] C. Roger, Etude des Γ -structures de codimension un sur la sphère S2 , Ann. Inst. Fourier 23 (4) (1973), 213–227. [296] C. Roger, Méthodes homotopiques et cohomologiques en théorie des feuilletages, Thèse Orsay (1974). [297] H. Rosenberg, Actions of Rn on manifolds, Comment. Math. Helv. 41 (1966), 170–178.

Foliations

75

[298] H. Rosenberg, Foliations by planes, Topology 7 (1968), 143–145. [299] H. Rosenberg and R. Roussarie, Topological equivalence of Reeb foliations, Topology 9 (1970), 231–242. [300] H. Rosenberg and R. Roussarie, Les feuilles exceptionnelles ne sont pas exceptionnelles, Comment. Math. Helv. 45 (1970), 517–523. [301] H. Rosenberg and R. Roussarie, Reeb foliations, Ann. of Math. (2) 91 (1970), 1–24. [302] H. Rosenberg and R. Roussarie, Some remarks on stability of foliations, J. Differential Geom. 10 (1975), 207–219. [303] H. Rosenberg, R. Roussarie and D. Weil, A classification of closed orientable 3-manifolds of rank two, Ann. of Math. 91 (1970), 449–464. [304] H. Rosenberg, R. Roussarie and D. Weil, A classification of the topological types of R2 -actions, Publ. Math. IHES 43 (1973), 261–272. [305] J. Rosenberg, K-theory of group C ∗ -algebras, foliation C ∗ -algebras and crossed products, Index Theory of Elliptic Operators, Foliations and Operator Algebras, New Orleans, Contemp. Math. 70 (1988), 251– 301. [306] R. Roussarie, Sur les feuilletages des variétés de dimension 3, Ann. Inst. Fourier 21 (3) (1971), 13–81. [307] R. Roussarie, Plongement dans les variétés feuilletées et classification des feuilletages sans holonomie, Publ. Math. IHES 43 (1974), 101–141. [308] D. Ruelle and D. Sullivan, Currents, flows and diffeomorphisms, Topology 14 (1975), 319–327. [309] H. Rummler, Quelques notions simples en géométrie riemannienne et leurs applications aux feuilletages compacts, Comment. Math. Helv. 54 (1979), 224–239. [310] S. Sacksteder, Some properties of foliations, Ann. Inst. Fourier 14 (1) (1964), 31–35. [311] S. Sacksteder, Foliations and pseudogroups, Amer. J. Math. 87 (1965), 79–102. [312] S. Sacksteder, A remark on Thurston stability theorem, Ann. Inst. Fourier 25 (2) (1975), 219–220. [313] S. Sacksteder and J. Schwartz, Limit sets for foliations, Ann. Inst. Fourier 15 (2) (1965), 201–213. [314] E. Salem, Une généralisation du théorème de Myers–Steenrod aux pseudo-groupes d’isométries locales, Ann. Inst. Fourier 35 (1985), 137–158. [315] E. Salhi, Sur les ensembles minimaux locaux, C. R. Acad. Sci. Paris 295 (1982), 691–694. [316] E. Salhi, Sur un théorème de structure des feuilletages de codimension 1, C. R. Acad. Sci. Paris 300 (1985), 635–638. [317] E. Salhi, Niveau des feuilles, C. R. Acad. Sci. Paris 301 (1985), 219–222. [318] M. Saralegi, The Euler class for flows of isometries, Differential Geometry, Santiago de Compostela, 1984, Pitman Research Notes in Math., vol. 131 (1985), 220–227. [319] K.S. Sarkaria, The de Rham cohomology of foliated manifolds, Thesis, Stony Brook (1974). [320] K.S. Sarkaria, A finiteness theorem for foliated manifolds, J. Math. Soc. Japan 30 (4) (1978), 687–696. [321] I. Satake, On the generalization of the notion of manifold, Proc. NAS 42 (1956), 359–363. [322] G. Schwarz, On the de Rham cohomology of the leaf space of a foliation, Topology 13 (1974), 185–188. [323] S. Schwartzman, Asymptotic cycles, Ann. of Math. 66 (1957), 270–284. [324] P. Schweitzer, Counterexamples to the Seifert conjecture and opening closed leaves of foliations, Ann. of Math. 100 (1974), 386–400. [325] P. Schweitzer, Stability of compact leaves with trivial linear holonomy, Topology 27 (1) (1988), 37–56. [326] P. Schweitzer, Existence of codimension one foliations with minimal leaves, Ann. Global Anal. Geom. 9 (1) (1991), 77–81. [327] P. Schweitzer, Codimension one foliations without compact leaves, Comment. Math. Helv. 70 (2) (1995), 171–209. [328] P. Scott, The geometry of 3-manifolds, Bull. London Math. Soc. 15 (1983), 401–487. [329] H. Seifert, Closed integral curves in 3-space and isotopic deformations, Proc. Amer. Math. Soc. 1 (1950), 287–302. [330] F. Sergeraert, Feuilletages et difféomorphismes infiniment tangents à l’identité, Invent. Math. 39 (1977), 253–275. [331] V. Sergiescu, Cohomologie basique et dualité des feuilletages riemanniens, Ann. Inst. Fourier de Grenoble 53 (3) (1985), 135–158. [332] C. Series, Foliations of polynomial growth are hyperfinite, Israel J. Math. 34 (1979), 245–258. [333] H. Shulman and J. Stasheff, De Rham theory for BΓ , Lecture Notes in Math. 652 (1978), 62–74.

76

R. Barre and A. El Kacimi Alaoui

[334] J.-C. Sikorav, Formes différentielles fermées sur le n-tore, Comment. Math. Helv. 57 (1982), 79–106. [335] J.D. Sondow, When is a manifold a leaf of some foliation? Bull. Amer. Math. Soc. 81 (1975), 622–624. [336] D. Sullivan, A foliation of geodesics is characterized by having no “tangent homologies”, J. Pure Appl. Algebra 13 (1973), 101–104. [337] D. Sullivan, A counterexample to the periodic orbit conjecture, Publ. Math. IHES 46 (1976), 5–14. [338] D. Sullivan, Cycles for the dynamical study of foliated manifolds and complex manifolds, Invent. Math. 36 (1976), 225–255. [339] D. Sullivan, A homological characterization of foliations consisting of minimal surfaces, Comment. Math. Helv. 54 (1979), 218–223. [340] H. Takai, KK-Theory for the C ∗ -algebras of Anosov foliations, Geometric Methods in Operator Algebras, Pitman Research Notes in Math., vol. 123 (1986), 387–399. [341] I. Tamura, Every odd dimensional homotopy sphere has a codimension one foliation, Comment. Math. Helv. 47 (1972), 73–79. [342] E. Thomas, Vector fields on manifolds, Bull. Amer. Math. Soc. 75 (1969), 643–683. [343] W.P. Thurston, Foliations of 3-manifolds which are circle bundles, Thesis, Berkeley (1972). [344] W.P. Thurston, Noncobordant foliations of S3 , Bull. Amer. Math. Soc. 78 (1972), 511–514. [345] W.P. Thurston, A generalization of the Reeb stability theorem, Topology 13 (1974), 347–352. [346] W.P. Thurston, Foliations and groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 304–307. [347] W.P. Thurston, The theory of foliations of codimension greater than one, Comment. Math. Helv. 49 (1974), 214–231. [348] W.P. Thurston, Existence of codimension one foliations, Ann. of Math. 102 (2) (1976), 249–268. [349] W.P. Thurston, The geometry and topology of 3-manifolds, Mimeographed Notes, Princeton (1977). [350] D. Tischler, On Fibering certain foliated manifolds over S 1 , Topology 9 (1970), 153–154. [351] P. Tondeur, Foliations on Riemannian Manifolds, Springer (1988). [352] P. Tondeur, Geometry of Foliations, Monographs in Mathematics, vol. 90, Birkhäuser (1997). [353] T. Tsuboi, Foliations with trivial F -subgroups, Topology 18 (1979), 223–233. [354] T. Tsuboi, On the cobordism classes of codimension one foliations which are almost without holonomy, Topology 22 (1983), 325–343. [355] T. Tsuboi, Foliated cobordism classes of certain foliated S1 -bundles, Topology 23 (1984), 233–244. [356] T. Tsuboi, On the foliated product of class C 1 , Ann. of Math. 130 (1989), 227–271. [357] T. Tsuboi, The Godbillon–Vey invariant and the foliated cobordism group, Proc. Japan Acad. 68 (1992), 85–90. [358] N. Tsuchiya, The Nishimori decompositions of codimension one foliations and the Godbillon–Vey classes, Tohoku Math. J. 34 (1982), 343–365. [359] H. Uzuki, Characteristic classes of foliated principal GLr -bundles, Hokkaido Math. J. 4 (1975), 159–168. [360] I. Vaisman, Variétés riemanniennes feuilletées, Czechoslovak Math. J. 21 (1971), 46–75. [361] I. Vaisman, Conformal foliations, Kodai Math. J. 2 (1979), 22–37. [362] W.T. Van Est, Rapport sur les S-atlas, Astérisque 116 (1984). [363] A. Verjovsky, A note on Anosov flows, Bol. Soc. Mat. Mexicana (1) 1 (1970), 36–37. [364] A. Verjovsky, Codimension one Anosov flows, Bol. Soc. Mat. Mexicana (2) 19 (1974), 49–77. [365] A. Verjovsky, A uniformisation theorem for holomorphic foliations, The Lefschetz Centenial Conference, Mexico, 1984, Contemp. Math. (3) 58 (1987), 233–253. [366] E. Vogt, Stable foliations of 4-manifolds by closed surfaces. Local structures and free actions of finite cyclic and dihedral groups on surfaces, Invent. Math. 22 (1973), 321–348. [367] E. Vogt, A foliation of R3 and other punctured 3-manifolds by circles, Publ. Math. IHES 69 (1981), 215– 232. [368] E. Vogt, The first cohomology group of leaves and local stability of compact foliations, Manuscripta Math. 37 (1982), 229–267. [369] E. Vogt, Existence of foliations of Euclidean spaces with all leaves compact, Math. Ann. 296 (1) (1993), 159–178. [370] A. Wadsley, Geodesic foliations by circles, J. Differential Geom. 10 (1975), 541–550. [371] P. Walczak, Dynamics of the geodesic flow of a foliation, Ergodic Theory Dynam. Systems 8 (1988), 637–650.

Foliations

77

[372] H.E. Winkelnkemper, The graph of a foliation, Ann. Global Anal. Geom. 1 (1983), 53–75. [373] R. Wolak, Characteristic classes of almost flat structures, Geom. Dedicata 24 (1987), 207–220. [374] R. Wolak, Maximal subalgebras in the algebra of foliated vector fields of a Riemannian foliation, Comment. Math. Helv. 64 (4) (1989), 536–541. [375] R. Wolak, The graph of a totally geodesic foliation, Ann. Polon. Math. 60 (3) (1995), 241–247. [376] J. Wood, Foliations on 3-manifolds, Ann. of Math. 89 (2) (1969), 336–358. [377] J. Wood, Foliations of codimension one, Bull. Amer. Math. Soc. 75 (1970), 1107–1111. [378] J. Wood, Bundles with totally disconnected structure group, Comment. Math. Helv. 46 (1971), 257–273. [379] J. Wouafo-Kamga, Décomposition des G-structures d’ordre supérieur. Structures transverse des feuilletages, Thèse, Toulouse (1979). [380] A. Zeggar, Nombre et classe de Lefschetz pour un feuilletage riemannien, Ann. Fac. Sci. Toulouse (6) 1 (1) (1992), 105–131. [381] A. Zeghib, Laminations et hypersurfaces géodésiques des variétés hyperboliques, Ann. Sci. École Norm. Sup. (4) 24 (2) (1991), 171–188. [382] A. Zeghib, Sur les feuilletages géodésiques continus des variétés hyperboliques, Invent. Math. 114 (1) (1993), 193–206. [383] A. Zeghib, Feuilletages géodésiques appliqués, Math. Ann. 298 (4) (1994), 729–759. [384] A. Zeghib, Feuilletages géodésiques des variétés localement symétriques, Topology 36 (4) (1997), 805– 828. [385] R. Zimmer, On the Mostow rigidity theorem and measurable foliations, Israel J. Math. 43 (1982), 281–290.

This page intentionally left blank

CHAPTER 3

Symplectic Geometry Ana Cannas da Silva1 Departamento de Matemática, Instituto Superior Técnico, 1049-001 Lisboa, Portugal E-mail: [email protected]; [email protected]

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . 1. Symplectic manifolds . . . . . . . . . . . . . . . 1.1. Symplectic linear algebra . . . . . . . . . . 1.2. Symplectic forms . . . . . . . . . . . . . . 1.3. Cotangent bundles . . . . . . . . . . . . . . 1.4. Moser’s trick . . . . . . . . . . . . . . . . . 1.5. Darboux and Moser theorems . . . . . . . 1.6. Symplectic submanifolds . . . . . . . . . . 2. Lagrangian submanifolds . . . . . . . . . . . . . 2.1. First Lagrangian submanifolds . . . . . . . 2.2. Lagrangian neighborhood theorem . . . . . 2.3. Weinstein tubular neighborhood theorem . 2.4. Application to symplectomorphisms . . . . 2.5. Generating functions . . . . . . . . . . . . 2.6. Fixed points . . . . . . . . . . . . . . . . . 2.7. Lagrangians and special Lagrangians in Cn 3. Complex structures . . . . . . . . . . . . . . . . 3.1. Compatible linear structures . . . . . . . . 3.2. Compatible almost complex structures . . 3.3. Integrability . . . . . . . . . . . . . . . . . 3.4. Kähler manifolds . . . . . . . . . . . . . . 3.5. Hodge theory . . . . . . . . . . . . . . . . 3.6. Pseudoholomorphic curves . . . . . . . . . 4. Symplectic geography . . . . . . . . . . . . . . 4.1. Existence of symplectic forms . . . . . . . 4.2. Fibrations and sums . . . . . . . . . . . . . 4.3. Symplectic blow-up . . . . . . . . . . . . . 4.4. Uniqueness of symplectic forms . . . . . . 4.5. Invariants for 4-manifolds . . . . . . . . . 4.6. Lefschetz pencils . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 Current address: Department of Mathematics, Princeton University, Princeton, NJ 08544-1000, USA.

HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 79

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 82 82 84 85 88 90 91 93 93 95 96 98 99 102 104 108 108 110 113 117 120 122 124 124 128 131 134 135 138

80

A. Cannas da Silva

5. Hamiltonian geometry . . . . . . . . . . . . . 5.1. Symplectic and Hamiltonian vector fields 5.2. Arnold conjecture and Floer homology . 5.3. Euler–Lagrange equations . . . . . . . . 5.4. Legendre transform . . . . . . . . . . . . 5.5. Integrable systems . . . . . . . . . . . . . 5.6. Symplectic and Hamiltonian actions . . . 5.7. Convexity . . . . . . . . . . . . . . . . . 6. Symplectic reduction . . . . . . . . . . . . . . 6.1. Marsden–Weinstein–Meyer theorem . . . 6.2. Applications and generalizations . . . . . 6.3. Moment map in gauge theory . . . . . . 6.4. Symplectic toric manifolds . . . . . . . . 6.5. Delzant’s construction . . . . . . . . . . 6.6. Duistermaat–Heckman theorems . . . . . References . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .

139 139 142 143 147 150 153 158 160 160 163 166 169 174 178 183

Symplectic geometry

81

Introduction This is an overview of symplectic geometry1 —the geometry of symplectic manifolds. From a language for classical mechanics in the XVIII century, symplectic geometry has matured since the 1960’s to a rich and central branch of differential geometry and topology. A current survey can thus only aspire to give a partial flavor on this exciting field. The following six topics have been chosen for this handbook: 1. Symplectic manifolds are manifolds equipped with symplectic forms. A symplectic form is a closed nondegenerate 2-form. The algebraic condition (nondegeneracy) says that the top exterior power of a symplectic form is a volume form, therefore symplectic manifolds are necessarily even-dimensional and orientable. The analytical condition (closedness) is a natural differential equation that forces all symplectic manifolds to being locally indistinguishable: they all locally look like an even-dimensional Euclidean space equipped with the dxi ∧ dyi symplectic form. All cotangent bundles admit canonical symplectic forms, a fact relevant for analysis of differential operators, dynamical systems, classical mechanics, etc. Basic properties, major classical examples, equivalence notions, local normal forms of symplectic manifolds and symplectic submanifolds are discussed in Section 1. 2. Lagrangian submanifolds2 are submanifolds of symplectic manifolds of half dimension and where the restriction of the symplectic form vanishes identically. By the Lagrangian creed [137], everything is a Lagrangian submanifold, starting with closed 1-forms, real functions modulo constants and symplectomorphisms (diffeomorphisms that respect the symplectic forms). Section 2 also describes normal neighborhoods of Lagrangian submanifolds with applications. 3. Complex structures or almost complex structures abound in symplectic geometry: any symplectic manifold possesses almost complex structures, and even so in a compatible sense. This is the point of departure for the modern technique of studying pseudoholomorphic curves, as first proposed by Gromov [64]. Kähler geometry lies at the intersection of complex, Riemannian and symplectic geometries, and plays a central role in these three fields. Section 3 includes the local normal form for Kähler manifolds and a summary of Hodge theory for Kähler manifolds. 4. Symplectic geography is concerned with existence and uniqueness of symplectic forms on a given manifold. Important results from Kähler geometry remain true in the more general symplectic category, as shown using pseudoholomorphic methods. This viewpoint was more recently continued with work on the existence of certain symplectic 1 The word symplectic in mathematics was coined in the late 1930’s by Weyl [142, p. 165] who substituted the Latin root in complex by the corresponding Greek root in order to label the symplectic group (first studied by Abel). An English dictionary is likely to list symplectic as the name for a bone in a fish’s head. 2 The name Lagrangian manifold was introduced by Maslov [93] in the 1960’s, followed by Lagrangian plane, etc., introduced by Arnold [2].

82

A. Cannas da Silva

submanifolds, in the context of Seiberg–Witten invariants, and with topological descriptions in terms of Lefschetz pencils. Both of these directions are particularly relevant to 4-dimensional topology and to mathematical physics, where symplectic manifolds occur as building blocks or as key examples. Section 4 treats constructions of symplectic manifolds and invariants to distinguish them. 5. Hamiltonian geometry is the geometry of symplectic manifolds equipped with a moment map, that is, with a collection of quantities conserved by symmetries. With roots in Hamiltonian mechanics, moment maps became a consequential tool in geometry and topology. The notion of a moment map arises from the fact that, to any real function on a symplectic manifold, is associated a vector field whose flow preserves the symplectic form and the given function; this is called the Hamiltonian vector field of that (Hamiltonian) function. The Arnold conjecture in the 60’s regarding Hamiltonian dynamics was a major driving force up to the establishment of Floer homology in the 80’s. Section 5 deals mostly with the geometry of moment maps, including the classical Legendre transform, integrable systems and convexity. 6. Symplectic reduction is at the heart of many symplectic arguments. There are infinitedimensional analogues with amazing consequences for differential geometry, as illustrated in a symplectic approach to Yang–Mills theory. Symplectic toric manifolds provide examples of extremely symmetric symplectic manifolds that arise from symplectic reduction using just the data of a polytope. All properties of a symplectic toric manifold may be read from the corresponding polytope. There are interesting interactions with algebraic geometry, representation theory and geometric combinatorics. The variation of reduced spaces is also addressed in Section 6.

1. Symplectic manifolds 1.1. Symplectic linear algebra Let V be a vector space over R, and let Ω : V × V → R be a skew-symmetric bilinear map. By a skew-symmetric version of the Gram–Schmidt process,3 there is a basis u1 , . . . , uk , e1 , . . . , en , f1 , . . . , fn of V for which Ω(ui , v) = Ω(ei , ej ) = Ω(fi , fj ) = 0 and Ω(ei , fj ) = δij for all i, j and all v ∈ V . Although such a basis is not unique, it is commonly referred to as a canonical basis. The dimension k of the subspace U = {u ∈ V | Ω(u, v) = 0 for all v ∈ V } is an invariant of the pair (V , Ω). Since k + 2n = dim V , the even number 2n is also an invariant of (V , Ω), called the rank of Ω. We denote ˜ := Ω(v, u). We say that Ω is symplecby Ω˜ : V → V ∗ the linear map defined by Ω(v)(u) tic (or nondegenerate) if the associated Ω˜ is bijective (i.e., the kernel U of Ω˜ is the trivial space {0}). In that case, the map Ω is called a linear symplectic structure on V , and the 3 Let u , . . . , u be a basis of U := {u ∈ V | Ω(u, v) = 0 for all v ∈ V }, and W a complementary subspace k 1 such that V = U ⊕ W . Take any nonzero e1 ∈ W . There is f1 ∈ W with Ω(e1 , f1 ) = 1. Let W1 be the span of e1 , f1 and W1Ω := {v ∈ V | Ω(v, u) = 0 ∀u ∈ W1 }. Then W = W1 ⊕ W1Ω . Take any nonzero e2 ∈ W1Ω . There is f2 ∈ W1Ω for which Ω(e2 , f2 ) = 1. Let W2 be the span of e2 , f2 , and so on.

Symplectic geometry

83

pair (V , Ω) is called a symplectic vector space. A linear symplectic structure Ω expresses a duality by the bijection Ω˜ : V −→ V ∗ , similar to the (symmetric) case of an inner product. By considering a canonical basis, we see that the dimension of a symplectic vector space (V , Ω) must be even, dim V = 2n, and that V admits a basis e1 , . . . , en , f1 , . . . , fn satisfying Ω(ei , fj ) = δij and Ω(ei , ej ) = 0 = Ω(fi , fj ). Such a basis is then called a symplectic basis of (V , Ω), and, in terms of exterior algebra, Ω = e1∗ ∧ f1∗ + · · · + en∗ ∧ fn∗ , where e1∗ , . . . , en∗ , f1∗ , . . . , fn∗ is the dual basis. With respect to a symplectic basis, the map Ω is represented by the matrix 0 Id . −Id 0 E XAMPLES . 1. The prototype of a symplectic vector space is (R2n , Ω0 ) with Ω0 such that the canonical basis e1 = (1, 0, . . . , 0), . . . , en , f1 , . . . , fn = (0, . . . , 0, 1) is a symplectic basis. Bilinearity then determines Ω0 on other vectors. 2. For any real vector space E, the direct sum V = E ⊕ E ∗ has a canonical symplectic structure determined by the formula Ω0 (u ⊕ α, v ⊕ β) = β(u) − α(v). If e1 , . . . , en is a basis of E, and f1 , . . . , fn is the dual basis, then e1 ⊕ 0, . . . , en ⊕ 0, 0 ⊕ f1 , . . . , 0 ⊕ fn is a symplectic basis for V . Given a linear subspace W of a symplectic vector space (V , Ω), its symplectic orthogonal is the subspace W Ω := {v ∈ V | Ω(v, u) = 0 for all u ∈ W }. By nondegeneracy, we have dim W + dim W Ω = dim V and (W Ω )Ω = W . For subspaces W and Y , we have (W ∩ Y )Ω = W Ω + Y Ω , and if W ⊆ Y then Y Ω ⊆ W Ω . There are special types of linear subspaces of a symplectic vector space (V , Ω). A subspace W is a symplectic subspace if the restriction Ω|W is nondegenerate, that is, W ∩ W Ω = {0}, or equivalently V = W ⊕ W Ω . A subspace W is an isotropic subspace if Ω|W ≡ 0, that is, W ⊆ W Ω . A subspace W is a coisotropic subspace if W Ω ⊆ W . A subspace W is a Lagrangian subspace if it is both isotropic and coisotropic, or equivalently, if it is an isotropic subspace with dim W = 12 dim V . A basis e1 , . . . , en of a Lagrangian subspace can be extended to a symplectic basis: choose f1 in the symplectic orthogonal to the linear span of {e2 , . . . , en }, etc. E XAMPLES . 1. For a symplectic basis as above, the span of e1 , f1 is symplectic, that of e1 , e2 isotropic, that of e1 , . . . , en , f1 coisotropic, and that of e1 , . . . , en Lagrangian. 2. The graph of a linear map A : E → E ∗ is a Lagrangian subspace of E ⊕ E ∗ with the canonical symplectic structure if and only if A is symmetric (i.e., (Au)v = (Av)u). Therefore, the Grassmannian of all Lagrangian subspaces in a 2n-dimensional symplectic vector space has dimension n(n+1) 2 . A symplectomorphism ϕ between symplectic vector spaces (V , Ω) and (V , Ω ) is a linear isomorphism ϕ : V −→ V such that ϕ ∗ Ω = Ω.4 If a symplectomorphism exists, 4 By definition, (ϕ ∗ Ω )(u, v) = Ω (ϕ(u), ϕ(v)).

84

A. Cannas da Silva

(V , Ω) and (V , Ω ) are said to be symplectomorphic. Being symplectomorphic is clearly an equivalence relation in the set of all even-dimensional vector spaces. The existence of canonical bases shows that every 2n-dimensional symplectic vector space (V , Ω) is symplectomorphic to the prototype (R2n , Ω0 ); a choice of a symplectic basis for (V , Ω) yields a symplectomorphism to (R2n , Ω0 ). Hence, nonnegative even integers classify equivalence classes for the relation of being symplectomorphic. Let Ω(V ) be the space of all linear symplectic structures on the vector space V . Take a Ω ∈ Ω(V ), and let Sp(V , Ω) be the group of symplectomorphisms of (V , Ω). The group GL(V ) of all isomorphisms of V acts transitively on Ω(V ) by pullback (i.e., all symplectic structures are related by a linear isomorphism), and Sp(V , Ω) is the stabilizer of the given Ω. Hence, Ω(V ) GL(V )/Sp(V , Ω). 1.2. Symplectic forms Let ω be a de Rham 2-form on a manifold5 M. For each point p ∈ M, the map ωp : Tp M × Tp M → R is skew-symmetric and bilinear on the tangent space to M at p, and ωp varies smoothly in p. D EFINITION 1.1. The 2-form ω is symplectic if ω is closed (i.e., its exterior derivative dω is zero) and ωp is symplectic for all p ∈ M. A symplectic manifold is a pair (M, ω) where M is a manifold and ω is a symplectic form. Symplectic manifolds must be even-dimensional. Moreover, the nth exterior power ωn of a symplectic form ω on a 2n-dimensional manifold is a volume form.6 Hence, any symn plectic manifold (M, ω) is canonically oriented. The form ωn! is called the symplectic volume or Liouville volume of (M, ω). When (M, ω) is a compact 2n-dimensional symplectic manifold, the de Rham cohomology class [ωn ] ∈ H 2n (M; R) must be nonzero by Stokes theorem. Therefore, the class [ω] must be nonzero, as well as its powers [ω]k = [ωk ] = 0. Exact symplectic forms can only exist on noncompact manifolds. Compact manifolds with a trivial even cohomology group H 2k (M; R), k = 0, 1, . . . , n, such as spheres S 2n with n > 1, can thus never be symplectic. On a manifold of dimension greater than 2, a function multiple f ω of a symplectic form ω is symplectic if and only if f is a nonzero locally constant function (this follows from the existence of a symplectic basis). E XAMPLES . 1. Let M = R2n with linear coordinates x1 , . . . , xn , y1 , . . . , yn . The form ω0 =

n

dxi ∧ dyi

i=1 5 Unless otherwise indicated, all vector spaces are real and finite-dimensional, all maps are smooth (i.e., C ∞ ) and all manifolds are smooth, Hausdorff and second countable. 6 A volume form is a nonvanishing form of top degree. If Ω is a symplectic structure on a vector space V of dimension 2n, its nth exterior power Ω n = Ω ∧ · · · ∧ Ω does not vanish. Actually, a skew-symmetric bilinear map Ω is symplectic if and only if Ω n = 0.

Symplectic geometry

2.

3.

4. 5. 6.

85

is symplectic, and the vectors ( ∂x∂ 1 )p , . . . , ( ∂x∂ n t)p , ( ∂y∂ 1 )p , . . . , ( ∂y∂ n )p constitute a symplectic basis of Tp M. Let M = Cn with coordinates z1 , . . . , zn . The form ω0 = 2i dzk ∧ d z¯ k is symplectic. In fact, this form coincides with that of the previous example under the identification Cn R2n , zk = xk + iyk . The 2-sphere S 2 , regarded as the set of unit vectors in R3 , has tangent vectors at p identified with vectors orthogonal to p. The standard symplectic form on S 2 is induced by the standard inner (dot) and exterior (vector) products: ωp (u, v) := p, u × v, for u, v ∈ Tp S 2 = {p}⊥ . This is the standard area form on S 2 with total area 4π . In terms of cylindrical polar coordinates 0 θ < 2π and −1 z 1 away from the poles, it is written ω = dθ ∧ dz. On any Riemann surface, regarded as a 2-dimensional oriented manifold, any area form, that is, any never vanishing 2-form, is a symplectic form. Products of symplectic manifolds are naturally symplectic by taking the sum of the pullbacks of the symplectic forms from the factors. If a (2n + 1)-dimensional manifold X admits a contact form, that is, a 1-form α such that α ∧ (dα)n is never vanishing, then the 2-form d(et α) is symplectic on X × R, and the symplectic manifold (X × R, d(et α)) is called the symplectization of the contact manifold (X, α). For more on contact geometry, see for instance the corresponding contribution in this volume.

D EFINITION 1.2. Let (M1 , ω1 ) and (M2 , ω2 ) be symplectic manifolds. A (smooth) map ψ : M1 → M2 is symplectic if ψ ∗ ω2 = ω1 .7 A symplectic diffeomorphism ϕ : M1 → M2 is a symplectomorphism. (M1 , ω1 ) and (M2 , ω2 ) are said to be symplectomorphic when there exists a symplectomorphism between them. The classification of symplectic manifolds up to symplectomorphism is an open problem in symplectic geometry. However, the local classification is taken care of by the Darboux theorem (Theorem 1.9): the dimension is the only local invariant of symplectic manifolds up to symplectomorphisms. That is, just as any n-dimensional manifold is locally diffeomorphic to Rn , any symplectic manifold (M 2n , ω) is locally symplectomorphic to (R2n , ω0 ). As a consequence, if we prove for (R2n , ω0 ) a local assertion that is invariant under symplectomorphisms, then that assertion holds for any symplectic manifold. We will hence refer to R2n , with linear coordinates (x1 , . . . , xn , y1 , . . . , yn ), and with symplectic form ω0 = ni=1 dxi ∧ dyi , as the prototype of a local piece of a 2n-dimensional symplectic manifold.

1.3. Cotangent bundles Cotangent bundles are major examples of symplectic manifolds. Let (U, x1 , . . . , xn ) be a coordinate chart for a manifold X, with associated cotangent coordinates (T ∗ U, x1 , . . . , xn , 7 By definition of pullback, we have (ψ ∗ ω ) (u, v) = (ω ) 2 p 2 ψ(p) (dψp (u), dψp (v)), at tangent vectors u, v ∈

Tp M1 .

86

A. Cannas da Silva

ξ1 , . . . , ξn ).8 Define a symplectic form on T ∗ U by ω=

n

dxi ∧ dξi .

i=1

One can check that this ω is intrinsically defined by considering the 1-form on T ∗ U , α=

n

ξi dxi ,

i=1

which satisfies ω = −dα and is coordinate-independent: in terms of the natural projection π : M → X, p = (x, ξ ) → x, the form α may be equivalently defined pointwise without coordinates by αp = (dπp )∗ ξ ∈ Tp∗ M, where (dπp )∗ : Tx∗ X → Tp∗ M is the transpose of dπp , that is, αp (v) = ξ((dπp )v) for v ∈ Tp M. Or yet, the form α is uniquely characterized by the property that µ∗ α = µ for every 1-form µ : X → T ∗ X (see Proposition 2.2). The 1-form α is the tautological form (or the Liouville 1-form) and the 2-form ω is the canonical symplectic form on T ∗ X. When referring to a cotangent bundle as a symplectic manifold, the symplectic structure is meant to be given by this canonical ω. Let X1 and X2 be n-dimensional manifolds with cotangent bundles M1 = T ∗ X1 and M2 = T ∗ X2 , and tautological 1-forms α1 and α2 . Suppose that f : X1 → X2 is a diffeomorphism. Then there is a natural diffeomorphism f! : M1 → M2 which lifts f ; namely, for p1 = (x1 , ξ1 ) ∈ M1 we define f! (p1 ) = p2 = (x2 , ξ2 ),

with

x2 = f (x1 ) ∈ X2 and ξ1 = (dfx1 )∗ ξ2 ∈ Tx∗1 X1 ,

where (dfx1 )∗ : Tx∗2 X2 −→ Tx∗1 X1 , so f! |Tx∗ is the inverse map of (dfx1 )∗ . 1

P ROPOSITION 1.3. The lift f! of a diffeomorphism f : X1 → X2 pulls the tautological form on T ∗ X2 back to the tautological form on T ∗ X1 , i.e., (f! )∗ α2 = α1 . 8 If an n-dimensional manifold X is described by coordinate charts (U, x , . . . , x ) with x : U → R, then, at n i 1 any x ∈ U , the differentials (dxi )x form a basis of Tx∗ X, inducing a map

T ∗ U −→ R2n , (x, ξ ) −→ (x1 , . . . , xn , ξ1 , . . . , ξn ), where ξ1 , . . . , ξn ∈ R are the corresponding coordinates of ξ ∈ Tx∗ X: ξ = ni=1 ξi (dxi )x . Then (T ∗ U, x1 , . . . , ∗ xn , ξ1 , . . . , ξn ) is a coordinate chart for the cotangent bundle T X; the coordinates x1 , . . . , xn , ξ1 , . . . , ξn are called the cotangent coordinates associated to the coordinates x1 , . . . , xn on U . One verifies that the transition functions on the overlaps are smooth, so T ∗ X is a 2n-dimensional manifold.

Symplectic geometry

87

P ROOF. At p1 = (x1 , ξ1 ) ∈ M1 , the claimed identity says (df! )∗p1 (α2 )p2 = (α1 )p1 , where p2 = f! (p1 ), that is, p2 = (x2 , ξ2 ) where x2 = f (x1 ) and (dfx1 )∗ ξ2 = ξ1 . This can be proved as follows: (df! )∗p1 (α2 )p2 = (df! )∗p1 (dπ2 )∗p2 ξ2 ∗ = d(π2 ◦ f! ) p ξ2 1 ∗ = d(f ◦ π1 ) p ξ2 1

=

(dπ1 )∗p1 (df )∗x1 ξ2

by definition of α2 by the chain rule because π2 ◦ f! = f ◦ π1 by the chain rule

= (dπ1 )∗p1 ξ1

by definition of f!

= (α1 )p1

by definition of α1 .

As a consequence of this naturality for the tautological form, a diffeomorphism of manifolds induces a canonical symplectomorphism of cotangent bundles: C OROLLARY 1.4. The lift f! : T ∗ X1 → T ∗ X2 of a diffeomorphism f : X1 → X2 is a symplectomorphism for the canonical symplectic forms, i.e., (f! )∗ ω2 = ω1 . In terms of the group (under composition) of diffeomorphisms Diff(X) of a manifold X, and the group of symplectomorphisms Sympl(T ∗ X, ω) of its cotangent bundle, we see that the injection Diff(X) → Sympl(T ∗ X, ω), f → f! is a group homomorphism. Clearly this is not surjective: for instance, consider the symplectomorphism T ∗ X → T ∗ X given by translation along cotangent fibers. E XAMPLE . Let X1 = X2 = S 1 . Then T ∗ S 1 is a cylinder S 1 × R. The canonical form is the area form ω = dθ ∧ dξ . If f : S 1 → S 1 is any diffeomorphism, then f! : S 1 × R → S 1 × R is a symplectomorphism, i.e., is an area-preserving diffeomorphism of the cylinder. Translation along the R direction is area-preserving but is not induced by a diffeomorphism of the base manifold S 1 . There is a criterion for which cotangent symplectomorphisms arise as lifts of diffeomorphisms in terms of the tautological form. First note the following feature of symplectic manifolds with exact symplectic forms. Let α be a 1-form on a manifold M such that ω = −dα is symplectic. There exists a unique vector field v whose interior product with ω is α, i.e., ıv ω = −α. If g : M → M is a symplectomorphism that preserves α (that is, g ∗ α = α), then g commutes with the flow9 of v, i.e., (exp tv) ◦ g = g ◦ (exp tv). When 9 For p ∈ M, (exp tv)(p) is the unique curve in M solving the initial value problem

d dt exp tv(p) = v exp tv(p) , (exp tv)(p)|t=0 = p for t in some neighborhood of 0. The one-parameter group of diffeomorphisms exp tv is called the flow of the vector field v.

88

A. Cannas da Silva

M = T ∗ X is the cotangent bundle of an arbitrary n-dimensional manifold X, and α is the ∂ tautological 1-form on M, the vector field v is just ξi ∂ξi with respect to a cotangent coordinate chart (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ). The flow exp tv, −∞ < t < ∞, satisfies (exp tv)(x, ξ ) = (x, et ξ ), for every (x, ξ ) in M. T HEOREM 1.5. A symplectomorphism g : T ∗ X → T ∗ X is a lift of a diffeomorphism f : X → X if and only if it preserves the tautological form: g ∗ α = α. P ROOF. By Proposition 1.3, a lift f! : T ∗ X → T ∗ X of a diffeomorphism f : X → X preserves the tautological form. Conversely, if g is a symplectomorphism of M that preserves α, then g preserves the cotangent fibration: by the observation above, g(x, ξ ) = (y, η) ⇒ g(x, λξ ) = (y, λη) for all (x, ξ ) ∈ M and λ > 0, and this must hold also for λ 0 by the differentiability of g at (x, 0). Therefore, there exists a diffeomorphism f : X → X such that π ◦g = f ◦π , where π : M → X is the projection map π(x, ξ ) = x, and g = f# . The canonical form is natural also in the following way. Given a smooth function h : X → R, the diffeomorphism τh of M = T ∗ X defined by τh (x, ξ ) = (x, ξ + dhx ) turns out to be always a symplectomorphism. Indeed, if π : M → X, π(x, ξ ) = x, is the projection, we have τh∗ α = α + π ∗ dh, so that τh∗ ω = ω. 1.4. Moser’s trick There are other relevant notions of equivalence for symplectic manifolds10 besides being symplectomorphic. Let M be a manifold with two symplectic forms ω0 , ω1 . D EFINITION 1.6. The symplectic manifolds (M, ω0 ) and (M, ω1 ) are strongly isotopic if there is an isotopy ρt : M → M such that ρ1∗ ω1 = ω0 . (M, ω0 ) and (M, ω1 ) are deformation-equivalent if there is a smooth family ωt of symplectic forms joining ω0 to ω1 . (M, ω0 ) and (M, ω1 ) are isotopic if they are deformation-equivalent and the de Rham cohomology class [ωt ] is independent of t. Hence, being strongly isotopic implies being symplectomorphic, and being isotopic implies being deformation-equivalent. We also have that being strongly isotopic implies being isotopic, because, if ρt : M → M is an isotopy such that ρ1∗ ω1 = ω0 , then ωt := ρt∗ ω1 is a smooth family of symplectic forms joining ω1 to ω0 and [ωt ] = [ω1 ], ∀t, by the homotopy invariance of de Rham cohomology. Moser [105] proved that, on a compact manifold, being isotopic implies being strongly isotopic (Theorem 1.7). McDuff showed that deformation-equivalence is indeed a necessary hypothesis: even if [ω0 ] = [ω1 ] ∈ H 2 (M; R), there are compact examples where (M, ω0 ) and (M, ω1 ) are not strongly isotopic; see Example 7.23 in [99]. In other words, 10 Understanding these notions and the normal forms requires tools, such as isotopies (by isotopy we mean a smooth one-parameter family of diffeomorphisms starting at the identity, like the flow of a vector field), Lie derivative, tubular neighborhoods and the homotopy formula in de Rham theory, covered in differential geometry or differential topology texts.

Symplectic geometry

89

fix c ∈ H 2 (M) and define Sc as the set of symplectic forms ω in M with [ω] = c. On a compact manifold, all symplectic forms in the same path-connected component of Sc are symplectomorphic according to the Moser theorem, though there might be symplectic forms in different components of Sc that are not symplectomorphic. T HEOREM 1.7 (Moser). Let M be a compact manifold with symplectic forms ω0 and ω1 . Suppose that ωt , 0 t 1, is a smooth family of symplectic forms joining ω0 to ω1 with cohomology class [ωt ] independent of t. Then there exists an isotopy ρ : M × R → M such that ρt∗ ωt = ω0 , 0 t 1. Moser applied an extremely useful argument, known as Moser’s trick, starting with the following observation. If there existed an isotopy ρ : M × R → M such that ρt∗ ωt = ω0 , 0 t 1, in terms of the associated time-dependent vector field vt :=

dρt ◦ ρt−1 , dt

t ∈ R,

we would then have for all 0 t 1 that 0=

d ∗ dωt (ρt ωt ) = ρt∗ Lvt ωt + dt dt

⇐⇒

Lvt ω t +

dωt = 0. dt

Conversely, the existence of a smooth time-dependent vector field vt , t ∈ R, satisfying the last equation is enough to produce by integration (since M is compact) the desired isotopy ρ : M × R → M satisfying ρt∗ ωt = ρ0∗ ω0 = ω0 , for all t. So everything boils down t to solving the equation Lvt ωt + dω dt = 0 for vt . d ωt ] = 0, there exists a smooth family of P ROOF. By the cohomology assumption that [ dt 1-forms µt such that

dωt = dµt , dt

0 t 1.

The argument involves the Poincaré lemma for compactly-supported forms, together with the Mayer–Vietoris sequence in order to use induction on the number of charts in a good cover of M; for a sketch, see page 95 in [99]. In the simplest case where ωt = (1 − t)ω0 + t tω1 with [ω0 ] = [ω1 ], we have that dω dt = ω1 − ω0 = dµ is exact. The nondegeneracy assumption on ωt , guarantees that we can pointwise solve the equation, known as Moser’s equation, ıvt ωt + µt = 0 to obtain a unique smooth family of vector fields vt , 0 t 1. Extend vt to all t ∈ R. Thanks to the compactness of M, the vector fields vt generate an isotopy ρ satisfying

90 dρt dt

A. Cannas da Silva

= vt ◦ ρt . Then we indeed have d ∗ dωt ∗ (ρ ωt ) = ρt Lvt ωt + = ρt∗ (dıvt ωt + dµt ) = ρt∗ d(ıvt ωt + µt ) = 0, dt t dt

where we used Cartan’s magic formula in Lvt ωt = dıvt ωt + ıvt dωt .

E XAMPLE . On a compact oriented 2-dimensional manifold M, a symplectic form is just an area form. Let ω0 and ω1 be two area forms on M. If [ω0 ] = [ω1 ], i.e., ω0 and ω1 give the same total area, then any convex combination of them is symplectic (because they induce the same orientation), and there is an isotopy ϕt : M → M, t ∈ [0, 1], such that ϕ1∗ ω0 = ω1 . Therefore, up to strong isotopy, there is a unique symplectic representative in each nonzero 2-cohomology class of M. On a noncompact manifold, given vt , we would need to check the existence for 0 t 1 t of an isotopy ρt solving the differential equation dρ dt = vt ◦ ρt . 1.5. Darboux and Moser theorems By a submanifold of a manifold M we mean either a manifold X with a closed embedding11 i : X → M, or an open submanifold (i.e., an open subset of M). Given a 2n-dimensional manifold M, a k-dimensional submanifold X, neighborhoods U0 , U1 of X, and symplectic forms ω0 , ω1 on U0 , U1 , we would like to know whether there exists a local symplectomorphism preserving X, i.e., a diffeomorphism ϕ : U0 → U1 with ϕ ∗ ω1 = ω0 and ϕ(X) = X. Moser’s Theorem 1.7 addresses the case where X = M. At the other extreme, when X is just one point, there is the classical Darboux theorem (Theorem 1.9). In general, we have: T HEOREM 1.8 (Moser theorem—relative version). Let ω0 and ω1 be symplectic forms on a manifold M, and X a compact submanifold of M. Suppose that the forms coincide, ω0 |p = ω1 |p , at all points p ∈ X. Then there exist neighborhoods U0 and U1 of X in M, and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ restricted to X is the identity map. P ROOF. Pick a tubular neighborhood U0 of X. The 2-form ω1 − ω0 is closed on U0 , and satisfies (ω1 − ω0 )p = 0 at all p ∈ X. By the homotopy formula on the tubular neighborhood, there exists a 1-form µ on U0 such that ω1 − ω0 = dµ and µp = 0 at all p ∈ X. Consider the family ωt = (1 − t)ω0 + tω1 = ω0 + t dµ of closed 2-forms on U0 . Shrinking U0 if necessary, we can assume that ωt is symplectic for t ∈ [0, 1], as nondegeneracy is an open property. Solve Moser’s equation, ıvt ωt = −µ, for vt By integration, shrinking U0 again if necessary, there exists a local isotopy ρ : U0 × [0, 1] → M with ρt∗ ωt = ω0 , for all t ∈ [0, 1]. Since vt |X = 0, we have ρt |X = idX . Set ϕ = ρ1 , U1 = ρ1 (U0 ). 11 A closed embedding is a proper injective immersion. A map is proper when its preimage of a compact set is always compact.

Symplectic geometry

91

T HEOREM 1.9 (Darboux). Let (M, ω) be a symplectic manifold, and let p be any point in M. Then we can find a chart (U, x1 , . . . , xn , y1 , . . . yn ) centered at p where ω=

n

dxi ∧ dyi .

i=1

Such a coordinate chart (U, x1 , . . . , xn , y1 , . . . , yn ) is called a Darboux chart, and the corresponding coordinates are called Darboux coordinates. The classical proof of Darboux’s theorem is by induction on the dimension of the manifold [2], in the spirit of the argument for a symplectic basis (Section 1.1). The proof below, using Moser’s theorem, was first provided by Weinstein [136]. P ROOF. Apply Moser’s relative theorem to X = {p}. More precisely, use any symplectic basis for (Tp M, ωp ) to construct coordinates(x1 , . . . , xn , y1 , . . . yn ) centered at p and dxi ∧ dyi |p . There are two symplectic valid on some neighborhood U , so that ω p = forms on U : the given ω0 = ω and ω1 = dxi ∧ dyi . By Theorem 1.8, there are neighborhoods U0 and U1 of p, and adiffeomorphismϕ : U0 → U1 such that ϕ(p) = p and ϕ ∗ ( dxi ∧ dyi ) = ω. Since ϕ ∗ ( dxi ∧ dyi ) = d(xi ◦ ϕ) ∧ d(yi ◦ ϕ), we simply set new coordinates xi = xi ◦ ϕ, yi = yi ◦ ϕ. Darboux’s theorem is easy in the 2-dimensional case. Being closed ω is locally exact, ω = dα. Every nonvanishing 1-form on a surface can be written locally as α = g dh for suitable functions g, h, where h is a coordinate on the local leaf space of the kernel foliation of α. The form ω = dg ∧ dh is nondegenerate if and only if (g, h) is a local diffeomorphism. By the way, transversality shows that the normal form for a generic12 2-form is x dx ∧ dy near a point where it is degenerate.

1.6. Symplectic submanifolds Moser’s argument permeates many other proofs, including those of the next two results regarding symplectic submanifolds. Let (M, ω) be a symplectic manifold. D EFINITION 1.10. A symplectic submanifold of (M, ω) is a submanifold X of M where, at each p ∈ X, the space Tp X is a symplectic subspace of (Tp M, ωp ). If i : X → M is the inclusion of a symplectic submanifold X, then the restriction of ω to X is a symplectic form, so that (X, i ∗ ω) is itself a symplectic manifold. Let X be a symplectic submanifold of (M, ω). At each p ∈ X, we have Tp M = Tp X ⊕ (Tp X)ωp (Section 1.1), so the map (Tp X)ωp → Tp M/Tp X is an isomorphism. This canonical identification of the normal space of X at p, Np X := Tp M/Tp X, with the symplectic orthogonal (Tp X)ωp , yields a canonical identification of the normal bundle N X 12 Generic here means that the subset of those 2-forms having this behavior is open, dense and invariant under diffeomorphisms of the manifold.

92

A. Cannas da Silva

with the symplectic vector bundle (T X)ω . A symplectic vector bundle is a vector bundle E → X equipped with a smooth13 field Ω of fiberwise nondegenerate skew-symmetric bilinear maps Ωp : Ep × Ep → R. The symplectic normal bundle is the normal bundle of a symplectic submanifold, with the symplectic structure induced by orthogonals. The next theorem, due to Weinstein [136], states that a neighborhood of a symplectic submanifold X is determined by X and (the isomorphism class of) its symplectic normal bundle. T HEOREM 1.11 (Symplectic neighborhood theorem). Let (M0 , ω0 ), (M1 , ω1 ) be symplectic manifolds with diffeomorphic compact symplectic submanifolds X0 , X1 . Let i0 : X0 → M0 , i1 : X1 → M1 be their inclusions. Suppose there is an isomorphism φ˜ : N X0 → NX1 of the corresponding symplectic normal bundles covering a symplectomorphism φ : (X0 , i0∗ ω0 ) → (X1 , i1∗ ω1 ). Then there exist neighborhoods U0 ⊂ M0 , U1 ⊂ M1 of X0 , X1 and a symplectomorphism ϕ : U0 → U1 extending φ such that the restriction of dϕ to ˜ the normal bundle N X0 is φ. As first noted by Thurston [131], the form Ω + π ∗ ωX is symplectic in some neighborhood of the zero section in N X, where π : N X → X is the bundle projection and ωX is the restriction of ω to X. Therefore, a compact symplectic submanifold X always admits a tubular neighborhood in the ambient (M, ω) symplectomorphic to a tubular neighborhood of the zero section in the symplectic normal bundle N X. P ROOF. By the Whitney extension theorem14 there exist neighborhoods U0 ⊂ M0 and U1 ⊂ M1 of X0 and X1 , and a diffeomorphism h : U0 → U1 such that h ◦ i0 = i1 ◦ φ and ˜ Hence ω0 and h∗ ω1 are two the restriction of dh to the normal bundle N X0 is the given φ. symplectic forms on U0 which coincide at all points p ∈ X0 . The result now follows from Moser’s relative theorem (Theorem 1.8). Carefully combining Moser’s argument with the existence of an ambient isotopy that produces a given deformation of a compact submanifold, we can show: T HEOREM 1.12. Let Xt , t ∈ [0, 1], be a (smooth) family of compact symplectic submanifolds of a compact symplectic manifold (M, ω). Then there exists an isotopy ρ : M × R → M such that for all t ∈ [0, 1] we have ρt∗ ω = ω and ρt (X0 ) = Xt . Inspired by complex geometry, Donaldson [32] proved the following theorem on the existence of symplectic submanifolds. A major consequence is the characterization of symplectic manifolds in terms of Lefschetz pencils; see Section 4.6. 13 Smoothness means that, for any pair of (smooth) sections u and v of E, the real-valued function Ω(u, v) : X → R given by evaluation at each point is smooth. 14 Whitney extension theorem. Let M be a manifold and X a submanifold of M. Suppose that at each p ∈ X we are given a linear isomorphism Lp : Tp M −→ Tp M such that Lp |Tp X = IdTp X and Lp depends smoothly on p. Then there exists an embedding h : N → M of some neighborhood N of X in M such that h|X = idX and dhp = Lp for all p ∈ X. A proof relies on a tubular neighborhood model.

Symplectic geometry

93

T HEOREM 1.13 (Donaldson). Let (M, ω) be a compact symplectic manifold. Assume that the cohomology class [ω] is integral, i.e., lies in H 2 (M; Z). Then, for every sufficiently large integer k, there exists a connected codimension-2 symplectic submanifold X representing the Poincaré dual of the integral cohomology class k[ω]. Under the same hypotheses, Auroux extended this result to show that given α ∈ H2m (M; Z) there exist positive k, ∈ Z such that kPD[ωn−m ] + α is realized by a 2mdimensional symplectic submanifold.

2. Lagrangian submanifolds 2.1. First Lagrangian submanifolds Let (M, ω) be a symplectic manifold. D EFINITION 2.1. A submanifold X of (M, ω) is Lagrangian (respectively, isotropic and coisotropic) if, at each p ∈ X, the space Tp X is a Lagrangian (respectively, isotropic and coisotropic) subspace of (Tp M, ωp ). If i : X → M is the inclusion map, then X is a Lagrangian submanifold if and only if i ∗ ω = 0 and dim X = 12 dim M. The problem of embedding15 a compact manifold as a Lagrangian submanifold of a given symplectic manifold is often global. For instance, Gromov [64] proved that there , and more generally can be no Lagrangian spheres in (Cn , ω0 ), except for the circle in C2 no compact exact Lagrangian submanifolds, in the sense that α0 = yj dxj restricts to an exact 1-form. The argument uses pseudoholomorphic curves (Section 3.6). Yet there are immersed Lagrangian spheres (Section 2.7). More recently were found topological and geometrical constraints on manifolds that admit Lagrangian embeddings into compact symplectic manifolds; see, for instance, [16,17,115]. E XAMPLES . 1. Any 1-dimensional submanifold of a symplectic surface is Lagrangian (because a 1-dimensional subspace of a symplectic vector space is always isotropic). Therefore, any product of n embedded curves arises as a Lagrangian submanifold of (a neighborhood of zero in) the prototype (R2n , ω0 ). In particular, a torus Tn = S 1 × · · · × S 1 can be embedded as a Lagrangian submanifold of any 2n-dimensional symplectic manifold, by Darboux’s theorem (Theorem 1.9). 2. Let M = T ∗ X be the cotangent bundle of a manifold X. With respect to a cotangent coordinate chart (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ), the tautological form is α = ξi dxi and the canonical form is ω = −dα = dxi ∧ dξi . The zero section X0 := {(x, ξ ) ∈ T ∗ X | ξ = 0 in Tx∗ X} is an n-dimensional submanifold of T ∗ X whose intersection with T ∗ U is given by the equations ξ1 = · · · = 15 An embedding is an immersion that is a homeomorphism onto its image.

94

A. Cannas da Silva

ξn = 0. Clearly α vanishes on X0 ∩ T ∗ U . Hence, if i0 : X0 → T ∗ X is the inclusion map, we have i0∗ ω = i0∗ dα = 0, and so X0 is Lagrangian. A cotangent fiber Tx∗0 X is an n-dimensional submanifold of T ∗ X given by the equations xi = (x0 )i , i = 1, . . . , n, on T ∗ U . Since the xi ’s are constant, the form α vanishes identically, and Tx∗0 X is a Lagrangian submanifold. Let Xµ be (the image of) an arbitrary section, that is, an n-dimensional submanifold of T ∗ X of the form Xµ = {(x, µx ) | x ∈ X, µx ∈ Tx∗ X}, where the covector µx depends smoothly on x, so µ : X → T ∗ X is a de Rham 1-form. We will investigate when such an Xµ is Lagrangian. Relative to the inclusion i : Xµ → T ∗ X and the cotangent projection π : T ∗ X → X, these Xµ ’s are exactly the submanifolds for which π ◦ i : Xµ → X is a diffeomorphism. P ROPOSITION 2.2. The tautological 1-form α on T ∗ X satisfies µ∗ α = µ, for any 1-form µ : X → T ∗ X. P ROOF. Denote by sµ : X → T ∗ X, x → (x, µx ), the 1-form µ regarded exclusively as a map. From the definition, αp = (dπp )∗ ξ at p = (x, ξ ) ∈ M. For p = sµ (x) = (x, µx ), we have αp = (dπp )∗ µx . Then, since π ◦ sµ = idX , we have ∗ ∗ sµ α x = (dsµ )∗x αp = (dsµ )∗x (dπp )∗ µx = d(π ◦ sµ ) x µx = µx .

The map sµ : X → T ∗ X, sµ (x) = (x, µx ) is an embedding with image the section Xµ . The diffeomorphism τ : X → Xµ , τ (x) := (x, µx ), satisfies i ◦ τ = sµ . P ROPOSITION 2.3. The sections of T ∗ X that are Lagrangian are those corresponding to closed 1-forms on X. P ROOF. Using the previous notation, the condition of Xµ being Lagrangian becomes: i ∗ dα = 0 ⇔ τ ∗ i ∗ dα = 0 ⇔ sµ∗ dα = 0 ⇔ d(sµ∗ α) = 0 ⇔ dµ = 0. When µ = dh for some h ∈ C ∞ (X), such a primitive h is called a generating function for the Lagrangian submanifold Xµ . Two functions generate the same Lagrangian submanifold if and only if they differ by a locally constant function. When X is simply connected, 1 (X) = 0, every Lagrangian Xµ admits a generating function. or at least HdeRham Besides the cotangent fibers, there are lots of Lagrangian submanifolds of T ∗ X not covered by the description in terms of closed 1-forms. Let S be any submanifold of an n-dimensional manifold X. The conormal space of S at x ∈ S is Nx∗ S = ξ ∈ Tx∗ X | ξ(v) = 0 for all v ∈ Tx S . The conormal bundle of S is N ∗ S = {(x, ξ ) ∈ T ∗ X | x ∈ S, ξ ∈ Nx∗ S}. This is an n-dimensional submanifold of T ∗ X. In particular, taking S = {x} to be one point, the conormal bundle is the corresponding cotangent fiber Tx∗ X. Taking S = X, the conormal bundle is the zero section X0 of T ∗ X.

Symplectic geometry

95

P ROPOSITION 2.4. If i : N ∗ S → T ∗ X is the inclusion of the conormal bundle of a submanifold S ⊂ X, and α is the tautological 1-form on T ∗ X, then i ∗ α = 0. P ROOF. Let (U, x1 , . . . , xn ) be a coordinate chart on X adapted to S, so that U ∩ S is described by xk+1 = · · · = xn = 0. Let (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ) be the associated cotangent coordinate chart. The submanifold N ∗ S ∩ T ∗ U is described by xk+1 = · · · = xn = 0 and ξ1 = · · · = ξk = 0. Since α = ξi dxi on T ∗ U , we conclude that, at p ∈ N ∗ S, ∗

(i α)p = αp |Tp (N ∗ S) =

ξi dxi

i>k

span{ ∂x∂ , ik} i

= 0.

C OROLLARY 2.5. For any submanifold S of X, the conormal bundle N ∗ S is a Lagrangian submanifold of T ∗ X.

2.2. Lagrangian neighborhood theorem Weinstein [136] proved that, if a compact submanifold X is Lagrangian with respect to two symplectic forms ω0 and ω1 , then the conclusion of the Moser relative theorem (Theorem 1.8) still holds. We need some algebra for the Weinstein theorem. Suppose that U, W are n-dimensional vector spaces, and Ω : U × W → R is a bilinear ˜ = Ω(u, ·). Then Ω is pairing; the map Ω gives rise to a linear map Ω˜ : U → W ∗ , Ω(u) ˜ nondegenerate if and only if Ω is bijective. P ROPOSITION 2.6. Let (V , Ω) be a symplectic vector space, U a Lagrangian subspace of (V , Ω), and W any vector space complement to U , not necessarily Lagrangian. Then from W we can canonically build a Lagrangian complement to U . P ROOF. From Ω we get a nondegenerate pairing Ω : U × W → R, so Ω˜ : U → W ∗ is bijective. We look for a Lagrangian complement to U of the form W = {w + Aw | w ∈ W } for some linear map A : W → U . For W to be Lagrangian we need that Ω(w1 , w2 ) = Ω˜ (Aw2 )(w1 ) − Ω˜ (Aw1 )(w2 ). Let A = Ω˜ ◦ A, and look for A such that Ω(w1 , w2 ) = A (w2 )(w1 ) − A (w1 )(w2 ) for all w1 , w2 ∈ W . The canonical choice is A (w) = − 12 Ω(w, ·). Set A = (Ω˜ )−1 ◦ A . P ROPOSITION 2.7. Let V be a vector space, let Ω0 and Ω1 be symplectic forms on V , let U be a subspace of V Lagrangian for Ω0 and Ω1 , and let W be any complement to U in V . Then from W we can canonically construct a linear isomorphism L : V −→ V such ∗ that L|U = IdU and L Ω1 = Ω0 . P ROOF. By Proposition 2.6, from W we canonically obtain complements W0 and W1 to U in V such that W0 is Lagrangian for Ω0 and W1 is Lagrangian for Ω1 . The nondegenerate bilinear pairings Ωi : Wi × U → R, i = 0, 1, give isomorphisms Ω˜ i : Wi −→ U ∗ , i = 0, 1, respectively. Let B : W0 → W1 be the linear map satisfying Ω˜ 1 ◦ B = Ω˜ 0 , i.e., Ω0 (w0 , u) =

96

A. Cannas da Silva

Ω1 (Bw0 , u), ∀w0 ∈ W0 , ∀u ∈ U . Let L := IdU ⊕ B : U ⊕ W0 → U ⊕ W1 be the extension of B to the rest of V by setting it to be the identity on U . It satisfies: (L∗ Ω1 )(u ⊕ w0 , u ⊕ w0 ) = Ω1 (u ⊕ Bw0 , u ⊕ Bw0 ) = Ω1 (u, Bw0 ) + Ω1 (Bw0 , u ) = Ω0 (u, w0 ) + Ω0 (w0 , u ) = Ω0 (u ⊕ w0 , u ⊕ w0 ).

T HEOREM 2.8 (Weinstein Lagrangian neighborhood theorem). Let M be a 2n-dimensional manifold, X a compact n-dimensional submanifold, i : X → M the inclusion map, and ω0 and ω1 symplectic forms on M such that i ∗ ω0 = i ∗ ω1 = 0, i.e., X is a Lagrangian submanifold of both (M, ω0 ) and (M, ω1 ). Then there exist neighborhoods U0 and U1 of X in M and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ is the identity on X, i.e., ϕ(p) = p, ∀p ∈ X. P ROOF. Put a Riemannian metric g on M. Fix p ∈ X, and let V = Tp M, U = Tp X and W = U ⊥ , the orthocomplement of U in V relative to the inner product gp (·, ·). Since i ∗ ω0 = i ∗ ω1 = 0, the subspace U is Lagrangian for both (V , ω0 |p ) and (V , ω1 |p ). By Proposition 2.7, we canonically get from U ⊥ a linear isomorphism Lp : Tp M → Tp M depending smoothly on p, such that Lp |Tp X = IdTp X and L∗p ω1 |p = ω0 |p . By the Whitney extension theorem (Section 1.5), there exist a neighborhood N of X and an embedding h : N → M with h|X = idX and dhp = Lp for p ∈ X. Hence, at any p ∈ X, we have (h∗ ω1 )p = (dhp )∗ ω1 |p = L∗p ω1 |p = ω0 |p . Applying the Moser relative theorem (Theorem 1.8) to ω0 and h∗ ω1 , we find a neighborhood U0 of X and an embedding f : U0 → N such that f |X = idX and f ∗ (h∗ ω1 ) = ω0 on Uo . Set ϕ = h ◦ f and U1 = ϕ(U0 ). Theorem 2.8 has the following generalization. For a proof see, for instance, either of [61,70,139]. T HEOREM 2.9 (Coisotropic embedding theorem). Let M be a manifold of dimension 2n, X a submanifold of dimension k n, i : X → M the inclusion, and ω0 and ω1 symplectic forms on M, such that i ∗ ω0 = i ∗ ω1 and X is coisotropic for both (M, ω0 ) and (M, ω1 ). Then there exist neighborhoods U0 and U1 of X in M and a diffeomorphism ϕ : U0 → U1 such that ϕ ∗ ω1 = ω0 and ϕ|X = idX .

2.3. Weinstein tubular neighborhood theorem Let (V , Ω) be a symplectic linear space, and let U be a Lagrangian subspace. Then there is a canonical nondegenerate bilinear pairing Ω : V /U × U → R defined by Ω ([v], u) = Ω(v, u) where [v] is the equivalence class of v in V /U . Consequently, we get a canonical isomorphism Ω˜ : V /U → U ∗ , Ω˜ ([v]) = Ω ([v], ·).

Symplectic geometry

97

In particular, if (M, ω) is a symplectic manifold, and X is a Lagrangian submanifold, then Tp X is a Lagrangian subspace of (Tp M, ωp ) for each p ∈ X and there is a canonical identification of the normal space of X at p, Np X := Tp M/Tp X, with the cotangent fiber Tp∗ X. Consequently the normal bundle N X and the cotangent bundle T ∗ X are canonically identified. T HEOREM 2.10 (Weinstein tubular neighborhood theorem). Let (M, ω) be a symplectic manifold, X a compact Lagrangian submanifold, ω0 the canonical symplectic form on T ∗ X, i0 : X → T ∗ X the Lagrangian embedding as the zero section, and i : X → M the Lagrangian embedding given by inclusion. Then there are neighborhoods U0 of X in T ∗ X, U of X in M, and a diffeomorphism ϕ : U0 → U such that ϕ ∗ ω = ω0 and ϕ ◦ i0 = i. P ROOF. By the standard tubular neighborhood theorem16 and since N X T ∗ X are canonically identified, we can find a neighborhood N0 of X in T ∗ X, a neighborhood N of X in M, and a diffeomorphism ψ : N0 → N such that ψ ◦ i0 = i. Let ω0 be the canonical form on T ∗ X and ω1 = ψ ∗ ω. The submanifold X is Lagrangian for both of these symplectic forms on N0 . By the Weinstein Lagrangian neighborhood theorem (Theorem 2.8), there exist neighborhoods U0 and U1 of X in N0 and a diffeomorphism θ : U0 → U1 such that θ ∗ ω1 = ω0 and θ ◦ i0 = i0 . Take ϕ = ψ ◦ θ and U = ϕ(U0 ). Then ϕ ∗ ω = θ ∗ ψ ∗ ω = θ ∗ ω1 = ω0 . Theorem 2.10 classifies compact Lagrangian embeddings: up to local symplectomorphism, the set of Lagrangian embeddings is the set of embeddings of manifolds into their cotangent bundles as zero sections. The classification of compact isotropic embeddings is also due to Weinstein in [137, 139]. An isotropic embedding of a manifold X into a symplectic manifold (M, ω) is a closed embedding i : X → M such that i ∗ ω = 0. Weinstein showed that neighborhood equivalence of isotropic embeddings is in one-to-one correspondence with isomorphism classes of symplectic vector bundles. The classification of compact coisotropic embeddings is due to Gotay [61]. A coisotropic embedding of a manifold X carrying a closed 2-form α of constant rank into a symplectic manifold (M, ω) is an embedding i : X → M such that i ∗ ω = α and i(X) is coisotropic as a submanifold of M. Let E be the characteristic distribution of a closed form α of constant rank on X, i.e., Ep is the kernel of αp at p ∈ X. Gotay showed that then the total space E ∗ carries a symplectic structure in a neighborhood of the zero section, such that X embeds coisotropically onto this zero section and, moreover, every coisotropic embedding is equivalent to this in some neighborhood of the zero section. 16 Tubular neighborhood theorem. Let M be a manifold, X a submanifold, N X the normal bundle of X in M,

i0 : X → N X the zero section, and i : X → M the inclusion. Then there are neighborhoods U0 of X in N X, U of X in M and a diffeomorphism ψ : U0 → U such that ψ ◦ i0 = i. This theorem can be proved with the exponential map using a Riemannian metric; see, for instance, [120].

98

A. Cannas da Silva

2.4. Application to symplectomorphisms Let (M1 , ω1 ) and (M2 , ω2 ) be two 2n-dimensional symplectic manifolds. Given a diffeo morphism f : M1 −→ M2 , there is a way to express the condition of f being a symplectomorphism in terms of a certain submanifold being Lagrangian. Consider the two projection maps pri : M1 × M2 → Mi , (p1 , p2 ) → pi , i = 1, 2. The twisted product form on M1 × M2 is the symplectic17 form ω˜ = (pr1 )∗ ω1 − (pr2 )∗ ω2 .

P ROPOSITION 2.11. A diffeomorphism f : M1 −→ M2 is a symplectomorphism if and only if the graph of f is a Lagrangian submanifold of (M1 × M2 , ω). ˜ P ROOF. The graph of f is the 2n-dimensional submanifold Graph f = {(p, f (p)) | p ∈ M1 } ⊆ M1 × M2 , which is the image of the embedding γ : M1 → M1 × M2 , p → (p, f (p)). We have γ ∗ ω˜ = γ ∗ pr∗1 ω1 − γ ∗ pr∗2 ω2 = (pr1 ◦ γ )∗ ω1 − (pr2 ◦ γ )∗ ω2 , and pr1 ◦ γ is the identity map on M1 whereas pr2 ◦ γ = f . So Graph f is Lagrangian, i.e., γ ∗ ω˜ = 0, if and only if f ∗ ω2 = ω1 , i.e., f is a symplectomorphism. Lagrangian submanifolds of (M1 × M2 , ω) ˜ are called canonical relations, when viewed as morphisms between (M1 , ω1 ) and (M2 , ω2 ), even if dim M1 = dim M2 . Under a reasonable assumption, there is a notion of composition [137]. Take M1 = M2 = M and suppose that (M, ω) is a compact symplectic manifold and f ∈ Sympl(M, ω). The graphs Graph f and , of f and of the identity map id : M → M, are Lagrangian submanifolds of M × M with ω˜ = pr∗1 ω − pr∗2 ω. By the Weinstein tubular neighborhood theorem, there exist a neighborhood U of in (M × M, ω) ˜ and a neighborhood U0 of M in (T ∗ M, ω0 ) with a symplectomorphism ϕ : U → U0 satisfying ϕ(p, p) = (p, 0), ∀p ∈ M. Suppose that f is sufficiently C 1 -close18 to id, i.e., f is in some sufficiently small neighborhood of the identity id in the C 1 -topology. Hence we can assume that Graph f ⊆ U . Let j : M → U , j (p) = (p, f (p)), be the embedding as Graph f , and i : M → U , i(p) = (p, p), be the embedding as = Graph id. The map j is sufficiently C 1 -close to i. These maps induce embeddings ϕ ◦ j = j0 : M → U0 and ϕ ◦ i = i0 : M → U0 as 0-section, respectively. Since the map j0 is sufficiently C 1 -close to i0 , the image set j0 (M) intersects each fiber Tp∗ M at one point µp depending smoothly on p. Therefore, the image of j0 is the image of a smooth section µ : M → T ∗ M, that is, a 1-form µ = j0 ◦ (π ◦ j0 )−1 . We conclude that Graph f {(p, µp ) | p ∈ M, µp ∈ Tp∗ M}. Conversely, if µ is a 1-form sufficiently C 1 -close to the zero 1-form, then {(p, µp ) | p ∈ M, µp ∈ Tp∗ M} Graph f , for some diffeomorphism f : M → M. 17 More generally, λ (pr )∗ ω + λ (pr )∗ ω is symplectic for all λ , λ ∈ R \ {0}. 1 1 1 2 2 2 1 2 18 Let X and Y be manifolds. A sequence of maps f : X → Y converges in the C 0 -topology (a.k.a. the i

compact-open topology) to f : X → Y if and only if fi converges uniformly on compact sets. A sequence of C 1 maps fi : X → Y converges in the C 1 -topology to f : X → Y if and only if it and the sequence of derivatives dfi : T X → T Y converge uniformly on compact sets.

Symplectic geometry

99

By Proposition 2.3, Graph f is Lagrangian if and only if µ is closed. A small C 1 neighborhood of id in Sympl(M, ω) is thus homeomorphic to a C 1 -neighborhood of zero in the vector space of closed 1-forms on M. So we obtain the model: Tid Sympl(M, ω) µ ∈ Ω 1 (M) | dµ = 0 . In particular, Tid (Sympl(M, ω)) contains the space of exact 1-forms that correspond to generating functions, C ∞ (M)/{locally constant functions}. T HEOREM 2.12. Let (M, ω) be a compact symplectic manifold (and not just one point) 1 (M) = 0. Then any symplectomorphism of M that is sufficiently C 1 -close to with HdeRham the identity has at least two fixed points. P ROOF. If f ∈ Sympl(M, ω) is sufficiently C 1 -close to id, then its graph corresponds to a 1 (M) = 0, we have that µ = dh for some h ∈ C ∞ (M). closed 1-form µ on M. As HdeRham But h must have at least two critical points because M is compact. A point p where µp = dhp = 0 corresponds to a point in the intersection of the graph of f with the diagonal, that is, a fixed point of f . This result has the following analogue in terms of Lagrangian intersections: if X is a 1 (X) = 0, compact Lagrangian submanifold of a symplectic manifold (M, ω) with HdeRham 1 19 then every Lagrangian submanifold of M that is C -close to X intersects X in at least two points.

2.5. Generating functions We focus on symplectomorphisms between the cotangent bundles M1 = T ∗ X1 , M2 = T ∗ X2 of two n-dimensional manifolds X1 , X2 . Let α1 , α2 and ω1 , ω2 be the corresponding tautological and canonical forms. Under the natural identification M1 × M2 = T ∗ X1 × T ∗ X2 T ∗ (X1 × X2 ), the tautological 1-form on T ∗ (X1 × X2 ) is α = pr∗1 α1 + pr∗2 α2 , the canonical 2-form on T ∗ (X1 × X2 ) is ω = −dα = pr∗1 ω1 + pr∗2 ω2 , and the twisted product form is ω˜ = pr∗1 ω1 − pr∗2 ω2 . We define the involution σ2 : M2 → M2 , (x2 , ξ2 ) → (x2 , −ξ2 ), which yields σ2∗ α2 = −α2 . Let σ = idM1 × σ2 : M1 × M2 → M1 × M2 . Then σ ∗ ω˜ = pr∗1 ω1 + pr∗2 ω2 = ω. If L is a Lagrangian submanifold of (M1 × M2 , ω), then its twist Lσ := σ (L) is a Lagrangian ˜ submanifold of (M1 × M2 , ω). For producing a symplectomorphism M1 = T ∗ X1 → M2 = T ∗ X2 we can start with a Lagrangian submanifold L of (M1 × M2 , ω), twist it to obtain a Lagrangian submanifold ˜ and, if Lσ happens to be the graph of some diffeomorphism ϕ : Lσ of (M1 × M2 , ω), M1 → M2 , then ϕ is a symplectomorphism. 19 We say that a submanifold Y of M is C 1 -close to another submanifold X when there is a diffeomorphism X → Y that is, as a map into M, C 1 -close to the inclusion X → M.

100

A. Cannas da Silva

A method to obtain Lagrangian submanifolds of M1 × M2 T ∗ (X1 × X2 ) relies on generating functions. For any f ∈ C ∞ (X1 × X2 ), df is a closed 1-form on X1 × X2 . The Lagrangian submanifold generated by f is Lf := {((x, y), (df )(x,y) ) | (x, y) ∈ X1 × X2 } (cf. Section 2.1). We adopt the loose notation dx f := dx f (x, y) := (df )(x,y) projected to Tx∗ X1 × {0}, dy f := dy f (x, y) := (df )(x,y) projected to {0} × Ty∗ X2 , which enables us to write Lf = {(x, y, dx f, dy f ) | (x, y) ∈ X1 × X2 } and Lσf = (x, y, dx f, −dy f ) | (x, y) ∈ X1 × X2 . When Lσf is in fact the graph of a diffeomorphism ϕ : M1 = T ∗ X1 → M2 = T ∗ X2 , we call ϕ the symplectomorphism generated by f , and call f the generating function of ϕ. The issue now is to determine whether a given Lσf is the graph of a diffeomorphism ϕ : M1 → M2 . Let (U1 , x1 , . . . , xn ), (U2 , y1 , . . . , yn ) be coordinate charts for X1 , X2 , with associated charts (T ∗ U1 , x1 , . . . , xn , ξ1 , . . . , ξn ), (T ∗ U2 , y1 , . . . , yn , η1 , . . . , ηn ) for M1 , M2 . The set Lσf is the graph of ϕ : M1 → M2 exactly when, for any (x, ξ ) ∈ M1 and (y, η) ∈ M2 , we have ϕ(x, ξ ) = (y, η) ⇔ ξ = dx f and η = −dy f . Therefore, given a point (x, ξ ) ∈ M1 , to find its image (y, η) = ϕ(x, ξ ) we must solve the Hamilton look-alike equations ξi = ∂f (x, y), ∂xi ηi = − ∂f (x, y). ∂yi If there is a solution y = ϕ1 (x, ξ ) of the first equation, we may feed it to the second thus obtaining η = ϕ2 (x, ξ ), so that ϕ(x, ξ ) = (ϕ1 (x, ξ ), ϕ2 (x, ξ )). By the implicit function theorem, in order to solve the first equation locally and smoothly for y in terms of x and ξ , we need the condition n ∂f ∂ det = 0. ∂yj ∂xi i,j =1 This is a necessary condition for f to generate a symplectomorphism ϕ. Locally this is also sufficient, but globally there is the usual bijectivity issue. E XAMPLE . Let X1 = X2 = Rn , and f (x, y) = − |x−y| 2 , the square of Euclidean distance up to a constant. In this case, the Hamilton equations are ξi = ∂f = yi − xi , ∂xi y i = x i + ξi , ⇐⇒ ∂f η ηi = − i = ξi . = yi − xi , 2

∂yi

The symplectomorphism generated by f is ϕ(x, ξ ) = (x + ξ, ξ ). If we use the Euclidean inner product to identify T ∗ Rn with T Rn , and hence regard ϕ as ϕ˜ : T Rn → T Rn

Symplectic geometry

101

and interpret ξ as the velocity vector, then the symplectomorphism ϕ corresponds to free translational motion in Euclidean space. The previous example can be generalized to the geodesic flow on a Riemannian manifold.20 Let (X, g) be a geodesically convex Riemannian manifold, where d(x, y) is the Riemannian distance between points x and y. Consider the function f : X × X −→ R,

f (x, y) = −

d(x, y)2 . 2

We want to investigate if f generates a symplectomorphism ϕ : T ∗ X → T ∗ X. Using the identification g˜ x : Tx X −→ Tx∗ X, v → gx (v, ·), induced by the metric, we translate ϕ into a map ϕ˜ : T X → T X. We need to solve

g˜ x (v) = ξ = dx f (x, y), g˜ y (w) = η = −dy f (x, y)

(1)

for (y, η) in terms of (x, ξ ) in order to find ϕ, or, equivalently, for (y, w) in terms (x, v) in order to find ϕ. ˜ Assume that (X, g) is geodesically complete, that is, every geodesic can be extended indefinitely. P ROPOSITION 2.13. Under the identification Tx X Tx∗ X given by the metric, the symplectomorphism generated by f corresponds to the map ϕ˜ : T X −→ T X, dγ (x, v) −→ γ (1), (1) , dt where γ is the geodesic with initial conditions γ (0) = x and

dγ dt

(0) = v.

20 A Riemannian metric on a manifold X is a smooth function g that assigns to each point x ∈ X an inner product gx on Tx X, that is, a symmetric positive-definite bilinear map gx : Tx X × Tx X → R. Smoothness means that for every (smooth) vector field v : X → T X the real-valued function x → gx (vx , vx ) is smooth. A Riemannian manifold is a pair (X, g) where g is a Riemannian metric on the manifold X. The arc-length of a piecewise

dγ dγ smooth curve γ : [a, b] → X on a Riemannian (X, g) is ab dγ dt dt , where dt (t) = dγt (1) ∈ Tγ (t) X and dt = dγ gγ (t) ( dγ dt , dt ) is the velocity of γ . A reparametrization of a curve γ : [a, b] → X is a curve of the form γ ◦ τ : [c, d] → X for some τ : [c, d] → [a, b]. By the change of variable formula for the integral, we see that the arc-length of γ is invariant by reparametrization. The Riemannian distance between two points x and y of a connected Riemannian manifold (X, g) is the infimum d(x, y) of the set of all arc-lengths for piecewise smooth curves joining x to y. A geodesic is a curve that locally minimizes distance and whose velocity is constant. Given any curve γ : [a, b] → X with dγ dt never vanishing, there is a reparametrization γ ◦ τ : [a, b] → X of constant velocity. A minimizing geodesic from x to y is a geodesic joining x to y whose arc-length is the Riemannian distance d(x, y). A Riemannian manifold (X, g) is geodesically convex if every point x is joined to every other point y by a unique (up to reparametrization) minimizing geodesic. For instance, (Rn , ·, ·) is a geodesically convex Riemannian manifold (where gx (v, w) = v, w is the Euclidean inner product on T Rn Rn × Rn ), for which the Riemannian distance is the usual Euclidean distance d(x, y) = |x − y|.

102

A. Cannas da Silva

This map ϕ˜ is called the geodesic flow on (X, g). P ROOF. Given (x, v) ∈ T X, let exp(x, v) : R → X be the unique geodesic with initial con(0) = v. In this notation, we need to show that the ditions exp(x, v)(0) = x and d exp(x,v) dt unique solution of the system of equations (1) is ϕ(x, ˜ v) = (exp(x, v)(1), d exp(x,v) (1)). dt The Gauss lemma in Riemannian geometry (see, for instance, [120]) asserts that geodesics are orthogonal to the level sets of the distance function. To solve the first equation for y = exp(x, u)(1) for some u ∈ Tx X, evaluate both sides at v and at vectors v ∈ Tx X orthogonal to v, |v|2 =

d −d(exp(x, v)(t), y)2 d −d(exp(x, v )(t), y)2 and 0 = dt 2 dt 2 t=0 t=0

to conclude that u = v, and thus y = exp(x, v)(1). We have −dy f (x, y)(w ) = 0 at vectors w ∈ Ty X orthogonal to W := d exp(x,v) (1), dt because f (x, y) is essentially the arc-length of a minimizing geodesic. Hence w = kW must be proportional to W , and k = 1 since d −d(x, exp(x, v)(1 − t))2 = |v|2 . k|v| = gy (kW, W ) = − dt 2 t=0 2

2.6. Fixed points Let X be an n-dimensional manifold, and M = T ∗ X its cotangent bundle equipped with the canonical symplectic form ω. Let f : X × X → R be a smooth function generating a symplectomorphism ϕ : M → M, ϕ(x, dx f ) = (y, −dy f ), with the notation of Section 2.5. To describe the fixed points of ϕ, we introduce the function ψ : X → R, ψ(x) = f (x, x). P ROPOSITION 2.14. There is a one-to-one correspondence between the fixed points of the symplectomorphism ϕ and the critical points of ψ. P ROOF. At x0 ∈ X, dx0 ψ = (dx f + dy f )|(x,y)=(x0 ,x0 ) . Let ξ = dx f |(x,y)=(x0 ,x0 ) . Recalling that Lσf is the graph of ϕ, we have that x0 is a critical point of ψ, i.e., dx0 ψ = 0, if and only if dy f |(x,y)=(x0 ,x0 ) = −ξ , which happens if and only if the point in Lσf corresponding to (x, y) = (x0 , x0 ) is (x0 , x0 , ξ, ξ ), i.e., ϕ(x0 , ξ ) = (x0 , ξ ) is a fixed point. Consider the iterates ϕ N = ϕ ◦ ϕ ◦ · · · ◦ ϕ, N = 1, 2, . . . , given by N successive applications of ϕ. According to the previous proposition, if the symplectomorphism ϕ N : M → M is generated by some function f (N ) , then there is a one-to-one correspondence between the set of fixed points of ϕ N and the set of critical points of ψ (N ) : X → R, ψ (N ) (x) = f (N ) (x, x). It remains to know whether ϕ N admits a generating function. We will see that to a certain extent it does. For each pair x, y ∈ X, define a map X → R, z → f (x, z) + f (z, y). Suppose that this map has a unique critical point z0 and that z0 is nondegenerate. As z0 is determined for

Symplectic geometry

103

each (x, y) implicitly by the equation dy f (x, z0 ) + dx f (z0 , y) = 0, by nondegeneracy, the implicit function theorem assures that z0 = z0 (x, y) is a smooth function. Hence, the function f (2) : X × X −→ R,

f (2) (x, y) := f (x, z0 ) + f (z0 , y)

is smooth. Since ϕ is generated by f , and z0 is critical, we have ϕ 2 x, dx f (2) (x, y) = ϕ ϕ(x, dx f (x, z0 ) = ϕ z0 , −dy f (x, z0 ) = ϕ z0 , dx f (z0 , y) = y, −dy f (z0 , y) = y, −dy f (2) (x, y) . We conclude that the function f (2) is a generating function for ϕ 2 , as long as, for each ξ ∈ Tx∗ X, there is a unique y ∈ X for which dx f (2) (x, y) equals ξ . There are similar partial recipes for generating functions of higher iterates. In the case of ϕ 3 , suppose that the function X × X → R, (z, u) → f (x, z) + f (z, u) + f (u, y), has a unique critical point (z0 , u0 ) and that it is a nondegenerate critical point. A generating function would be f (3) (x, y) = f (x, z0 ) + f (z0 , u0 ) + f (u0 , y). When the generating functions f, f (2) , f (3) , . . . , f (N ) exist given by these formulas, the N -periodic points of ϕ, i.e., the fixed points of ϕ N , are in one-to-one correspondence with the critical points of (x1 , . . . , xN ) −→ f (x1 , x2 ) + f (x2 , x3 ) + · · · + f (xN −1 , xN ) + f (xN , x1 ). E XAMPLE . Let χ : R → R2 be a smooth plane curve that is 1-periodic, i.e., χ(s + 1) = χ(s), and parametrized by arc-length, i.e., | dχ ds | = 1. Assume that the region Y enclosed by the image of χ is convex, i.e., for any s ∈ R, the tangent line {χ(s) + t dχ ds | t ∈ R} intersects the image X := ∂Y of χ only at the point χ(s). Suppose that a ball is thrown into a billiard table of shape Y rolling with constant velocity and bouncing off the boundary subject to the usual law of reflection. The map describing successive points on the orbit of the ball is ϕ : R/Z × (−1, 1) −→ R/Z × (−1, 1), (x, v) −→ (y, w), saying that when the ball bounces off χ(x) with angle θ = arccos v, it will next collide with χ(y) and bounce off with angle ν = arccos w. Then the function f : R/Z × R/Z → R defined by f (x, y) = −|χ(x) − χ(y)| is smooth off the diagonal, and for ϕ(x, v) = (y, w) satisfies

∂f (x, y) = χ(y)−χ(x) · dχ

∂x |χ(x)−χ(y)| ds s=x = cos θ = v,

∂f (x, y) = χ(x)−χ(y) · dχ

∂y |χ(x)−χ(y)| ds s=y = − cos ν = −w.

104

A. Cannas da Silva

We conclude that f is a generating function for ϕ. Similar approaches work for higherdimensional billiard problems. Periodic points are obtained by finding critical points of real functions of N variables in X,

(x1 , . . . , xN ) −→ χ(x1 ) − χ(x2 ) + · · · + χ(xN −1 ) − χ(xN )

+ χ(xN ) − χ(x1 ) , that is, by finding the N -sided (generalized) polygons inscribed in X of critical perimeter. Notice that R/Z × (−1, 1) {(x, v) | x ∈ X, v ∈ Tx X, |v| < 1} is the open unit tangent ball bundle of a circle X, which is an open annulus A, and the map ϕ : A → A is areapreserving, as in the next two theorems. While studying Poincaré return maps in dynamical systems, Poincaré arrived at the following results. T HEOREM 2.15 (Poincaré recurrence theorem). Let ϕ : A → A be a volume-preserving diffeomorphism of a finite-volume manifold A, and U a nonempty open set in A. Then there is q ∈ U and a positive integer N such that ϕ N (q) ∈ U . Hence, under iteration, a mechanical system governed by ϕ will eventually return arbitrarily close to the initial state. P ROOF. Let U0 = U , U1 = ϕ(U), U2 = ϕ 2 (U), . . . . If all of these sets were disjoint, then, since Volume(Ui ) = Volume(U) > 0 for all i, the volume of A would be greater or equal to i Volume(Ui ) = ∞. To avoid this contradiction we must have ϕ k (U) ∩ ϕ (U) = ∅ for some k > , which implies ϕ k− (U) ∩ U = ∅. T HEOREM 2.16 (Poincaré’s last geometric theorem). Suppose that ϕ : A → A is an areapreserving diffeomorphism of the closed annulus A = R/Z × [−1, 1] that preserves the two components of the boundary and twists them in opposite directions. Then ϕ has at least two fixed points. This theorem was proved in 1913 by Birkhoff [18], and hence is also called the Poincaré–Birkhoff theorem. It has important applications to dynamical systems and celestial mechanics. The Arnold conjecture on the existence of fixed points for symplectomorphisms of compact manifolds (see Section 5.2) may be regarded as a generalization of the Poincaré–Birkhoff theorem. This conjecture has motivated a significant amount of research involving a more general notion of generating function; see, for instance, [41,55].

2.7. Lagrangians and special Lagrangians in Cn The standard Hermitian inner product h(·, ·) on Cn has real and imaginary parts given by the Euclidean inner product ·, · and (minus) the symplectic form ω0 , respectively: for

Symplectic geometry

105

v = (x1 + iy1 , . . . , xn + iyn ), u = (a1 + ib1 , . . . , an + ibn ) ∈ Cn , n n n h(v, u) = (xk + iyk )(ak − ibk ) = (xk ak + yk bk ) − i (xk bk − yk ak ) k=1

k=1

k=1

= v, u − iω0 (v, u). L EMMA 2.17. Let W be a subspace of (Cn , ω0 ) and e1 , . . . , en vectors in Cn . Then: (a) W is Lagrangian if and only if W ⊥ = iW ; (b) (e1 , . . . , en ) is an orthonormal basis of a Lagrangian subspace if and only if (e1 , . . . , en ) is a unitary basis of Cn . P ROOF. (a) We always have ω0 (v, u) = − im h(v, u) = re h(iv, u) = iv, u. It follows that, if W is Lagrangian, so that ω0 (v, u) = 0 for all v, u ∈ W , then iW ⊆ W ⊥ . These spaces must be equal because they have the same dimension. Reciprocally, when iv, u = 0 for all v, u ∈ W , the equality above shows that W must be isotropic. Since dim W = dim iW = dim W ⊥ = 2n − dim W , the dimension of W must be n. (b) If (e1 , . . . , en ) is an orthonormal basis of a Lagrangian subspace W , then, by the previous part, (e1 , . . . , en , ie1 , . . . , ien ) is an orthonormal basis of Cn as a real vector space. Hence (e1 , . . . , en ) must be a complex basis of Cn and it is unitary because h(ej , ek ) = ej , ek − iω0 (ej , ek ) = δj k . Conversely, if (e1 , . . . , en ) is a unitary basis of Cn , then the real span of these vectors is Lagrangian (ω0 (ej , ek ) = − im h(ej , ek ) = 0) and they are orthonormal (ej , ek = re h(ej , ek ) = δj k ). The Lagrangian Grassmannian Λn is the set of all Lagrangian subspaces of Cn . It follows from part (b) of Lemma 2.17 that Λn is the set of all subspaces of Cn admitting an orthonormal basis that is a unitary basis of Cn . Therefore, we have Λn U(n)/O(n). Indeed U(n) acts transitively on Λn : given W, W ∈ Λn with orthonormal bases (e1 , . . . , en ), (e1 , . . . , en ), respectively, there is a unitary transformation of Cn that maps (e1 , . . . , en ) to (e1 , . . . , en ) as unitary bases of Cn . And the stabilizer of Rn ∈ Λn is the subgroup of those unitary transformations that preserve this Lagrangian subspace, namely O(n). It follows that Λn is a compact connected manifold of dimension n(n+1) 2 ; cf. the last example of Section 1.1. The Lagrangian Grassmannian comes with a tautological vector bundle τn := (W, v) ∈ Λn × Cn | v ∈ W , whose fiber over W ∈ Λn is the n-dimensional real space W . It is a consequence of part (a) of Lemma 2.17 that the following map gives a well-defined global isomorphism of the complexification τn ⊗R C with the trivial bundle Cn over Λn (i.e., a global trivialization): (W, v ⊗ c) → (W, cv), for W ∈ Λn , v ∈ W , c ∈ C.

106

A. Cannas da Silva

D EFINITION 2.18. A Lagrangian immersion of a manifold X is an immersion f : X → Cn such that dfp (Tp X) is a Lagrangian subspace of (Cn , ω0 ), for every p ∈ X. E XAMPLE . The graph of a map h : Rn → iRn is an embedded n-dimensional submanifold X of Cn . Its tangent space at (p, h(p)) is {v + dhp (v) | v ∈ Rn }. Let e1 , . . . , en be the standard basis of Rn . Since ω0 (ek + dhp (ek ), ej + dhp (ej )) = ek , −i dhp (ej ) + ∂hj n k ej , i dhp (ek ), we see that X is Lagrangian if and only if ∂h ∂xj = ∂xk , ∀j, k, which in R n is if and only if h is the gradient of some H : R → iR. If f : X → Cn is a Lagrangian immersion, we can define a Gauss map λf : X −→ Λn , p −→ dfp (Tp X). Since λ∗f τn = T X and τn ⊗ C Cn , we see that a necessary condition for an immersion X → Cn to exist is that the complexification of T X be trivializable. Using the h-principle (Section 3.2), Gromov [65] showed that this is also sufficient: an n-dimensional manifold X admits a Lagrangian immersion into Cn if and only if the complexification of its tangent bundle is trivializable. E XAMPLE . For the unit sphere S n = {(t, x) ∈ R × Rn : t 2 + |x|2 = 1}, the Whitney sphere immersion is the map f : S n −→ Cn , (t, x) −→ x + itx. The only self-intersection is at the origin where f (−1, 0, . . . , 0) = f (1, 0, . . . , 0). Since T(t,x) S n = (t, x)⊥ , the differential df(t,x) : (u, v) → v + i(tv + ux) is always injective: v + i(tv + ux) = 0 ⇔ v = 0 and ux = 0, but when x = 0 it is t = ±1 and T(±1,0) S n = {0} × Rn , so it must be u = 0. We conclude that f is an immersion. By computing ω0 at two vectors of the form v + i(tv + ux), we find that the image dfp (Tp S n ) is an n-dimensional isotropic subspace of Cn . Therefore, f is a Lagrangian immersion of S n , and the complexification T S n ⊗ C must be always trivializable, though the tangent bundle T S n is only trivializable in dimensions n = 0, 1, 3, 7. The special Lagrangian Grassmannian SΛn is the set of all oriented subspaces of Cn admitting a positive orthonormal basis (e1 , . . . , en ) that is a special unitary basis of Cn . By the characterization of Lagrangian in the part (b) of Lemma 2.17, it follows that the elements of SΛn are indeed Lagrangian submanifolds. Similarly to the case of the Lagrangian Grassmannian, we have that SΛn SU(n)/SO(n) is a compact connected manifold of dimension

n(n+1) 2

− 1.

Symplectic geometry

107

We can single out the special Lagrangian subspaces by expressing the condition on the determinant in terms of the real n-form in Cn , β := im Ω,

where Ω := dz1 ∧ · · · ∧ dzn .

Since for A ∈ SO(n), we have det A = 1 and Ω(e1 , . . . , en ) = Ω(Ae1 , . . . , Aen ), we see that, for an oriented real n-dimensional subspace W ⊂ Cn , the number Ω(e1 , . . . , en ) does not depend on the choice of a positive orthonormal basis (e1 , . . . , en ) of W , thus can be denoted Ω(W ) and its imaginary part β(W ). P ROPOSITION 2.19. A subspace W of (Cn , ω0 ) has an orientation for which it is a special Lagrangian if and only if W is Lagrangian and β(W ) = 0. P ROOF. Any orthonormal basis (e1 , . . . , en ) of a Lagrangian subspace W ⊂ Cn is the image of the canonical basis of Cn by some A ∈ U(n), and Ω(W ) = det A ∈ S 1 . Therefore, W admits an orientation for which such a positive (e1 , . . . , en ) is a special unitary basis of Cn if and only if det A = ±1, i.e., β(W ) = 0. D EFINITION 2.20. A special Lagrangian immersion of an oriented manifold X is a Lagrangian immersion f : X → Cn such that, at each p ∈ X, the space dfp (Tp X) is a special Lagrangian subspace of (Cn , ω0 ). For a special Lagrangian immersion f , the Gauss map λf takes values in SΛn . By Proposition 2.19, the immersion f of an n-dimensional manifold X in (Cn , ω0 ) is special Lagrangian if and only if f ∗ ω0 = 0 and f ∗ β = 0. E XAMPLE . In C2 , writing zk = xk + iyk , we have β = dx1 ∧ dy2 + dy1 ∧ dx2 . We have seen that the graph of the gradient i∇H is Lagrangian, for any function H : R2 → R. ∂H ∂H , i ∂x ) is a Lagrangian immersion. For f to be a special LaSo f (x1 , x2 ) = (x1 , x2 , i ∂x 1 2 grangian immersion, we need the vanish of f ∗ β = dx1 ∧ d

∂H ∂x2

+d

∂H ∂x1

∧ dx2 =

∂ 2H ∂ 2H dx1 ∧ dx2 . + ∂x12 ∂x22

Hence the graph of ∇H is special Lagrangian if and only if H is harmonic. If f : X → Cn is a special Lagrangian immersion, then f ∗ Ω is an exact (real) volume form: f ∗ Ω = d re(z1 dz2 ∧ · · · ∧ dzn ). We conclude, by Stokes theorem, that there can be no special Lagrangian immersion of a compact manifold in Cn . Calabi–Yau manifolds21 are more general manifolds where a definition of special Lagrangian submanifold makes sense and where the space of special Lagrangian embeddings of a compact manifold is interesting. Special Lagrangian geometry was introduced by Harvey and Lawson [71]. For a treatment of Lagrangian and special Lagrangian submanifolds with many examples; see, for instance, [9]. 21 Calabi–Yau manifolds are compact Kähler manifolds (Section 3.4) with vanishing first Chern class.

108

A. Cannas da Silva

3. Complex structures 3.1. Compatible linear structures A complex structure on a vector space V is a linear map J : V → V such that J 2 = −Id. The pair (V , J ) is then called a complex vector space.A complex structure J on V is equivalent to a structure of vector space over C, the map J corresponding to multiplication by i. If (V , Ω) is a symplectic vector space, a complex structure J on V is said to be compatible (with Ω, or Ω-compatible) if the bilinear map GJ : V × V → R defined by GJ (u, v) = Ω(u, J v) is an inner product on V . This condition comprises J being a symplectomorphism (i.e., Ω(J u, J v) = Ω(u, v), ∀u, v) and the so-called taming: Ω(u, J u) > 0, ∀u = 0. E XAMPLE . For the symplectic vector space (R2n , Ω0 ) with symplectic basis e1 = (1, 0, . . . , 0), . . . , en , f1 , . . . , fn = (0, . . . , 0, 1), there is a standard compatible complex structure J0 determined by J0 (ej ) = fj and J0 (fj ) = −ej for all j = 1, . . . , n. This corresponds to a standard identification of R2n with Cn , and Ω0 (u, J0 v) = u, v is the standard Euclidean inner product. With respect to the symplectic basis e1 , . . . , en , f1 , . . . , fn , the map J0 is represented by the matrix !

0 Id

" −Id . 0

The symplectic linear group, Sp(2n) := {A ∈ GL(2n; R) | Ω0 (Au, Av) = Ω0 (u, v) for all u, v ∈ R2n }, is the group of all linear transformations of R2n that preserve the standard symplectic structure. The orthogonal group O(2n) is the group formed by the linear transformations A that preserve the Euclidean inner product, Au, Av = u, v, for all u, v ∈ R2n . The general complex group GL(n; C) is the group of linear transformations A : R2n → R2n commuting with J0 , A(J0 v) = J0 (Av), for all v ∈ R2n .22 The compatibility between the structures Ω0 , ·, · and J0 implies that the intersection of any two of these subgroups of GL(2n; R) is the same group, namely the unitary group U(n). As (R2n , Ω0 ) is the prototype of a 2n-dimensional symplectic vector space, the preceding example shows that compatible complex structures always exist on symplectic vector spaces.23 There is yet a way to produce a canonical compatible complex structure J after the choice of an inner product G on (V , Ω), though the starting G(u, v) is usually different from GJ (u, v) := Ω(u, J v). P ROPOSITION 3.1. Let (V , Ω) be a symplectic vector space, with an inner product G. Then there is a canonical compatible complex structure J on V . 22 Identify the complex n × n matrix X + iY with the real 2n × 2n matrix X −Y . Y X 23 Conversely, given (V , J ), there is a symplectic Ω with which J is compatible: take Ω(u, v) = G(J u, v) for an inner product G such that J t = −J .

Symplectic geometry

109

P ROOF. By nondegeneracy of Ω and G, the maps u → Ω(u, ·) and w → G(w, ·) are both isomorphisms between V and V ∗ . Hence, Ω(u, v) = G(Au, v) for some linear A : V → V . The map A is skew-symmetric, and the product AAt is symmetric24 and positive: G(AAt u, u) = G(At u, At u) > 0, for u = 0. By the spectral theorem, these properties imply that AAt diagonalizes with positive eigenvalues λi , say AAt = B diag(λ1 , . . . , λ2n ) B −1 . We may hence define an arbitrary real power of AAt by rescaling the eigenspaces, in particular, √ # # AAt := B diag( λ1 , . . . , λ2n )B −1 . √ The linear transformation AAt is symmetric, positive-definite and does not depend on the choice of B nor of the ordering of the eigenvalues. It is completely determined by its effect on√each eigenspace of AAt : on the eigenspace √ corresponding to the eigenvalue λk , the map AA√t is defined to be multiplication by λk . √ Let J := ( AAt )−1 A. Since A and AAt commute, J is orthogonal (J J t = Id), as well as skew-symmetric (J t = −J ). It follows that J is a complex structure on V . Compatibility is easily checked: Ω(J u, J v) = G(AJ u, J v) = G(J Au, J v) = G(Au, v) = Ω(u, v) and √ Ω(u, J u) = G(Au, J u) = G(−J Au, u) = G AAt u, u > 0, The factorization A =

for u = 0.

√ AAt J is called the polar decomposition of A.

R EMARK . Being canonical, this construction may be smoothly performed: when (Vt , Ωt ) is a family of symplectic vector spaces with a family Gt of inner products, all depending smoothly on a parameter t, an adaptation of the previous proof shows that there is a smooth family Jt of compatible complex structures on (Vt , Ωt ). Let (V , Ω) be a symplectic vector space of dimension 2n, and let J be a complex structure on V . If J is Ω-compatible and L is a Lagrangian subspace of (V , Ω), then J L is also Lagrangian and J L = L⊥ , where ⊥ indicates orthogonality with respect to the inner product GJ (u, v) = Ω(u, J v). Therefore, a complex structure J is Ω-compatible if and only if there exists a symplectic basis for V of the form e 1 , e 2 , . . . , en ,

f 1 = J e1 ,

f 2 = J e2 ,

...,

fn = J en .

Let J (V , Ω) be the set of all compatible complex structures in a symplectic vector space (V , Ω). 24 A map B : V → V is symmetric, respectively skew-symmetric, when B t = B, respectively B t = −B, where the transpose B t : V → V is determined by G(B t u, v) = G(u, Bv).

110

A. Cannas da Silva

P ROPOSITION 3.2. The set J (V , Ω) is contractible.25 P ROOF. Pick a Lagrangian subspace L0 of (V , Ω). Let L(V , Ω, L0 ) be the space of all Lagrangian subspaces of (V , Ω) that intersect L0 transversally. Let G(L0 ) be the space of all inner products on L0 . The map Ψ : J (V , Ω) −→ L(V , Ω, L0 ) × G(L0 ), J −→ (J L0 , GJ |L0 ) is a homeomorphism, with inverse as follows. Take (L, G) ∈ L(V , Ω, L0 ) × G(L0 ). For v ∈ L0 , v ⊥ = {u ∈ L0 | G(u, v) = 0} is a (n − 1)-dimensional space of L0 ; its symplectic orthogonal (v ⊥ )Ω is (n + 1)-dimensional. Then (v ⊥ )Ω ∩ L is 1-dimensional. Let J v be the unique vector in this line such that Ω(v, J v) = 1. If we take v’s in some G-orthonormal basis of L0 , this defines an element J ∈ J (V , Ω). The set L(V , Ω, L0 ) can be identified with the vector space of all symmetric n × n matrices. In fact, any n-dimensional subspace L of V that is transverse to L0 is the graph of a linear map J L0 → L0 , and the Lagrangian ones correspond to symmetric maps (cf. Section 1.1). Hence, L(V , Ω, L0 ) is contractible. Since G(L0 ) is contractible (it is even convex), we conclude that J (V , Ω) is contractible.

3.2. Compatible almost complex structures An almost complex structure on a manifold M is a smooth26 field of complex structures on the tangent spaces, Jp : Tp M → Tp M, p ∈ M. The pair (M, J ) is then called an almost complex manifold. D EFINITION 3.3. An almost complex structure J on a symplectic manifold (M, ω) is compatible (with ω or ω-compatible) if the map that assigns to each point p ∈ M the bilinear pairing gp : Tp M × Tp M → R, gp (u, v) := ωp (u, Jp v) is a Riemannian metric on M. A triple (ω, g, J ) of a symplectic form, a Riemannian metric and an almost complex structure on a manifold M is a compatible triple when g(·, ·) = ω(·, J ·). If (ω, J, g) is a compatible triple, each of ω, J or g can be written in terms of the other two. E XAMPLES . 1. If we identify R2n with Cn using coordinates zj = xj + iyj , multiplication by i induces a constant linear map J0 on the tangent spaces such that J02 = −Id, known 25 Contractibility of J (V , Ω) means that there exists a homotopy h : J (V , Ω) → J (V , Ω), 0 t 1, starting t at the identity h0 = Id, finishing at a trivial map h1 : J (V , Ω) → {J0 }, and fixing J0 (i.e., ht (J0 ) = J0 , ∀t ) for

some J0 ∈ J (V , Ω). 26 Smoothness means that for any vector field v, the image J v is a (smooth) vector field.

Symplectic geometry

111

as the standard almost complex structure on R2n : J0

∂ ∂xj

∂ = , ∂yj

J0

∂ ∂yj

=−

∂ . ∂xj

For the standard symplectic form ω0 = dxj ∧ dyj and the Euclidean inner product g0 = ·, ·, the compatibility relation holds: ω0 (u, v) = g0 (J0 (u), v). 2. Any oriented hypersurface Σ ⊂ R3 carries a natural symplectic form and a natural compatible almost complex structure induced by the standard inner (or dot) and exterior (or vector) products. They are given by the formulas ωp (u, v) := νp , u × v and Jp (v) = νp × v for v ∈ Tp Σ , where νp is the outward-pointing unit normal vector at p ∈ Σ (in other words, ν : Σ → S 2 is the Gauss map). Cf. Example 3 of Section 1.2. The corresponding Riemannian metric is the restriction to Σ of the standard Euclidean metric ·, ·. 3. The previous example generalizes to the oriented hypersurfaces M ⊂ R7 . Regarding u, v ∈ R7 as imaginary octonions (or Cayley numbers), the natural vector product u × v is the imaginary part of the product of u and v as octonions. This induces a natural almost complex structure on M given by Jp (v) = νp × v, where νp is the outward-pointing unit normal vector at p ∈ M. In the case of S 6 , at least, this J is not compatible with any symplectic form, as S 6 cannot be a symplectic manifold. As a consequence of the remark in Section 3.1, we have: P ROPOSITION 3.4. On any symplectic manifold (M, ω) with a Riemannian metric g, there is a canonical compatible almost complex structure J . Since Riemannian metrics always exist, we conclude that any symplectic manifold has compatible almost complex structures. The metric gJ (·, ·) := ω(·, J ·) tends to be different from the given g(·, ·). P ROPOSITION 3.5. Let (M, J ) be an almost complex manifold where J is compatible with two symplectic forms ω0 , ω1 Then ω0 and ω1 are deformation-equivalent. P ROOF. Simply take the convex combinations ωt = (1 − t)ω0 + tω1 , 0 t 1.

A counterexample to the converse of this proposition is provided by the family ωt = cos πt dx1 ∧ dy1 + sin πt dx1 ∧ dy2 + sin πt dy1 ∧ dx2 + cos πt dx2 ∧ dy2 for 0 t 1. There is no J in R4 compatible with both ω0 and ω1 = −ω0 . A submanifold X of an almost complex manifold (M, J ) is an almost complex submanifold when J (T X) ⊆ T X, i.e., we have Jp v ∈ Tp X, ∀p ∈ X, v ∈ Tp X. P ROPOSITION 3.6. Let (M, ω) be a symplectic manifold equipped with a compatible almost complex structure J . Then any almost complex submanifold X of (M, J ) is a symplectic submanifold of (M, ω).

112

A. Cannas da Silva

P ROOF. Let i : X → M be the inclusion. Then i ∗ ω is a closed 2-form on X. Since ωp (u, v) = gp (Jp u, v), ∀p ∈ X, ∀u, v ∈ Tp X, and since gp |Tp X is nondegenerate, so is ωp |Tp X , and i ∗ ω is nondegenerate. It is easy to see that the set J (M, ω) of all compatible almost complex structures on a symplectic manifold (M, ω) is path-connected. From two almost complex structures J0 , J1 compatible with ω, we get two Riemannian metrics g0 (·, ·) = ω(·, J0 ·), g1 (·, ·) = ω(·, J1 ·). Their convex combinations gt (·, ·) = (1 − t)g0 (·, ·) + tg1 (·, ·),

0 t 1,

form a smooth family of Riemannian metrics. Applying the polar decomposition to the family (ω, gt ), we obtain a smooth path of compatible almost complex structures Jt joining J0 to J1 . The set J (M, ω) is even contractible (this is important for defining invariants). The first ingredient is the contractibility of the set of compatible complex structures on a vector space (Proposition 3.2). Consider the fiber bundle J → M with fiber over p ∈ M being the space Jp := J (Tp M, ωp ) of compatible complex structures on the tangent space at p. A compatible almost complex structure on (M, ω) is a section of J . The space of sections of J is contractible because the fibers are contractible.27 The first Chern class c1 (M, ω) of a symplectic manifold (M, ω) is the first Chern class of (T M, J ) for any compatible J . The class c1 (M, ω) ∈ H 2 (M; Z) is invariant under deformations of ω. We never used the closedness of ω to obtain compatible almost complex structures. The construction holds for an almost symplectic manifold (M, ω), that is, a pair of a manifold M and a nondegenerate 2-form ω, not necessarily closed. We could further work with a symplectic vector bundle, that is, a vector bundle E → M equipped with a smooth field ω of fiberwise nondegenerate skew-symmetric bilinear maps (Section 1.6). The existence of such a field ω is equivalent to being able to reduce the structure group of the bundle from the general linear group to the linear symplectic group. As both Sp(2n) and GL(n; C) retract to their common maximal compact subgroup U(n), a symplectic vector bundle can be always endowed with a structure of complex vector bundle, and vice-versa. Gromov showed in his thesis [63] that any open28 almost complex manifold admits a symplectic form. The books [42, §10.2] and [99, §7.3] contain proofs of this statement using different techniques. T HEOREM 3.7 (Gromov). For an open manifold the existence of an almost complex structure J implies that of a symplectic form ω in any given 2-cohomology class and such that J is homotopic to an almost complex structure compatible with ω. From an almost complex structure J and a metric g, one builds a nondegenerate 2-form ω(u, v) = g(J u, v), which will not be closed in general. Closedness is a differential re27 The base being a (second countable and Hausdorff) manifold, a contraction can be produced using a countable cover by trivializing neighborhoods whose closures are compact subsets of larger trivializing neighborhoods, and such that each p ∈ M belongs to only a finite number of such neighborhoods. 28 A manifold is open if it has no closed connected components, where closed means compact and without boundary.

Symplectic geometry

113

lation, i.e., a condition imposed on the partial derivatives, encoded as a subset of jet space. One says that a differential relation satisfies the h-principle29 if any formal solution (i.e., a solution for the associated algebraic problem, in the present case a nondegenerate 2-form) is homotopic to a holonomic solution (i.e., a genuine solution, in the present case a closed nondegenerate 2-form). Therefore, when the h-principle holds, one may concentrate on a purely topological question (such as the existence of an almost complex structure) in order to prove the existence of a differential solution. Gromov showed that, for an open differential relation on an open manifold, when the relation is invariant under the group of diffeomorphisms of the underlying manifold, the inclusion of the space of holonomic solutions into the space of formal solutions is a weak homotopy equivalence, i.e., induces isomorphisms of all homotopy groups. The previous theorem fits here as an application. For closed manifolds there is no such theorem: as discussed in Section 1.2, the existence of a 2-cohomology class whose top power is nonzero is also necessary for the existence of a symplectic form and there are further restrictions coming from Gromov–Witten theory (see Section 4.5). 3.3. Integrability Any complex manifold30 has a canonical almost complex structure J . It is defined locally over the domain U of a complex chart ϕ : U → V ⊆ Cn , by Jp ( ∂x∂ j |p ) = ∂y∂ j |p and Jp ( ∂y∂ j |p ) = − ∂x∂ j |p , where these are the tangent vectors induced by the real and imaginary parts of the coordinates of ϕ = (z1 , . . . , zn ), zj = xj + iyj . This yields a globally well-defined J , thanks to the Cauchy–Riemann equations satisfied by the components of the transition maps. An almost complex structure J on a manifold M is called integrable when J is induced by some underlying structure of complex manifold on M as above. The question arises whether some compatible almost complex structure J on a symplectic manifold (M, ω) is integrable. To understand what is involved, we review Dolbeault theory and the Newlander–Nirenberg theorem. Let (M, J ) be a 2n-dimensional almost complex manifold. The fibers of the complexified tangent bundle, T M ⊗ C, are 2n-dimensional vector spaces over C. We may extend J linearly to T M ⊗ C by J (v ⊗ c) = J v ⊗ c, v ∈ T M, c ∈ C. Since J 2 = −Id, on the complex vector space (T M ⊗ C)p the linear map Jp has eigenvalues ±i. The (±i)-eigenspaces of J are denoted T1,0 and T0,1 , respectively, and called the spaces of J -holomorphic and of J -anti-holomorphic tangent vectors. We have an isomorphism

(π1,0 , π0,1 ) : T M ⊗ C −→ T1,0 ⊕ T0,1 , 1 v −→ (v − iJ v, v + iJ v), 2 29 There are in fact different h-principles depending on the different possible coincidences of homotopy groups for the spaces of formal solutions and of holonomic solutions. 30 A complex manifold of (complex) dimension n is a set M with a complete complex atlas {(U , V , ϕ ), α ∈ α α α $ index set I } where M = α Uα , the Vα ’s are open subsets of Cn , and the maps ϕα : Uα → Vα are bijections such that the transition maps ψαβ = ϕβ ◦ ϕα−1 : Vαβ → Vβα are biholomorphic (i.e., bijective, holomorphic and with holomorphic inverse) as maps on open subsets of Cn , Vαβ = ϕα (Uα ∩ Uβ ).

114

A. Cannas da Silva

where the maps to each summand satisfy π1,0 ◦ J = iπ1,0 and π0,1 ◦ J = −iπ0,1 . Restricting π1,0 to T M, we see that (T M, J ) T1,0 T0,1 , as complex vector bundles, where the multiplication by i is given by J in (T M, J ) and where T0,1 denotes the complex conjugate bundle of T0,1 . Similarly, J ∗ defined on T ∗ M ⊗ C by J ∗ ξ = ξ ◦ J has (±i)-eigenspaces T 1,0 = (T1,0 )∗ and T 0,1 = (T0,1 )∗ , respectively, called the spaces of complex-linear and of complexantilinear cotangent vectors. Under the two natural projections π 1,0 , π 0,1 , the complexified cotangent bundle splits as 1,0 0,1 ∗ : T M ⊗ C −→ T 1,0 ⊕ T 0,1 , π ,π 1 ξ −→ (ξ − iJ ∗ ξ, ξ + iJ ∗ ξ ). 2 Let Λk (T ∗ M ⊗ C) := Λk T 1,0 ⊕ T 0,1 = Λ,m , +m=k

where Λ,m := (Λ T 1,0 ) ∧ (Λm T 0,1 ), and let Ω k (M; C) be the space of sections of Λk (T ∗ M ⊗ C), called complex-valued k-forms on M. The differential forms of type (, m) on (M, J ) are the sections of Λ,m , and the space of these differential forms is Ω ,m . The decomposition of forms by Dolbeault type is Ω k (M; C) = % denoted ,m . Let π ,m : Λk (T ∗ M ⊗ C) → Λ,m be the projection map, where + m = k. +m=k Ω The usual exterior derivative d (extended linearly to smooth complex-valued forms) composed with two of these projections induces the del and del-bar differential operators, ¯ on forms of type (, m): ∂ and ∂, ∂ := π +1,m ◦ d : Ω ,m −→ Ω +1,m and ∂¯ := π ,m+1 ◦ d : Ω ,m −→ Ω ,m+1 . If β ∈ Ω ,m (M), with k = + m, then dβ ∈ Ω k+1 (M; C): dβ =

¯ + · · · + π 0,k+1 dβ. π r,s dβ = π k+1,0 dβ + · · · + ∂β + ∂β

r+s=k+1

In particular, on complex-valued functions we have df = d(re f ) + i d(im f ) and d = ¯ where ∂ = π 1,0 ◦ d and ∂¯ = π 0,1 ◦ d. A function f : M → C is J -holomorphic ∂ + ∂, at p ∈ M if dfp is complex linear, i.e., dfp ◦ Jp = i dfp (or dfp ∈ Tp1,0 ). A function f is J -holomorphic if it is holomorphic at all p ∈ M. A function f : M → C is J -antiholomorphic at p ∈ M if dfp is complex antilinear, i.e., dfp ◦ Jp = −i dfp (or dfp ∈ Tp0,1 ), ¯ that is, when the conjugate function f¯ is holomorphic at p ∈ M. In terms of ∂ and ∂,

Symplectic geometry

115

¯ = 0, and f is J -anti-holomorphic if and a function f is J -holomorphic if and only if ∂f only if ∂f = 0. When M is a complex % manifold and J is its canonical almost complex structure, the splitting Ω k (M; C) = +m=k Ω ,m is particularly interesting. Let U ⊆ M be the domain of a complex coordinate chart ϕ = (z1 , . . . , zn ), where the corresponding real coordinates x1 , y1 , . . . , xn , yn satisfy zj = xj + iyj . In terms of ∂ 1 ∂ ∂ and := −i ∂zj 2 ∂xj ∂yj

∂ 1 ∂ ∂ , := +i ∂ z¯ j 2 ∂xj ∂yj

the (±i)-eigenspaces of Jp (p ∈ U ) can be written

∂

∂

(T1,0 )p = C-span . : j = 1, . . . , n and (T ) = C-span 0,1 p ∂zj p ∂ z¯ j p Similarly, putting dzj = dxj + i dyj and d z¯ j = dxj − i dyj , we obtain simple formulas ∂b ∂b ¯ for the differentials of a b ∈ C ∞ (U; C), ∂b = ∂zj dzj and ∂b = ∂ z¯ j d z¯ j , and we have T 1,0 = C-span{dzj : j = 1, . . . , n} and T 0,1 = C-span{d z¯ j : j = 1, . . . , n}. If we use multiindex notation J = (j1 , . . . , j ) where 1 j1 < · · · < j n, |J | = and dzJ = dzj1 ∧ dzj2 ∧ · · · ∧ dzj , then the set of (, m)-forms on U is

bJ,K dzJ ∧ d z¯ K | bJ,K ∈ C ∞ (U; C) .

Ω ,m =

|J |=, |K|=m

A form β ∈ Ω k (M; C) may be written over U as β= bJ,K dzJ ∧ d z¯ K . +m=k |J |=, |K|=m

Since d = ∂ + ∂¯ on functions, we get dβ = dbJ,K ∧ dzJ ∧ d z¯ K +m=k |J |=, |K|=m

=

+m=k

&

∂bJ,K ∧ dzJ ∧ d z¯ K

|J |=, |K|=m

'(

)

∈Ω +1,m

+

|J |=, |K|=m

&

¯ J,K ∧ dzJ ∧ d z¯ K ∂b '( ∈Ω ,m+1

¯ = ∂β + ∂β,

)

116

A. Cannas da Silva

and conclude that, on a complex manifold, d = ∂ + ∂¯ on forms of any degree. This cannot be proved for an almost complex manifold, because there are no coordinate functions zj to give a suitable basis of 1-forms. ¯ for any form β ∈ Ω ,m , we have When d = ∂ + ∂, ¯ + ∂∂β ¯ + ∂¯ 2 β 0 = d 2 β = ∂ 2 β + ∂ ∂β &'() & '( ) &'() ∈Ω +2,m

∈Ω +1,m+1 ∂¯

&⇒

∈Ω ,m+2 ∂¯

2 ∂¯ = 0, ¯ = 0, ∂ ∂¯ + ∂∂ 2 ∂ = 0.

∂¯

Since ∂¯ 2 = 0, the chain 0 −→ Ω ,0 −→ Ω ,1 −→ Ω ,2 −→ · · · is a differential complex. Its cohomology groups ,m HDolbeault (M) :=

ker ∂¯ : Ω ,m → Ω ,m+1 im ∂¯ : Ω ,m−1 → Ω ,m

are called the Dolbeault cohomology groups. The Dolbeault theorem states that for com,m (M) H m (M; O(Ω (,0) )), where O(Ω (,0) ) is the sheaf of plex manifolds HDolbeault forms of type (, 0) over M. It is natural to ask whether the identity d = ∂ + ∂¯ could hold for manifolds other than complex manifolds. Newlander and Nirenberg [106] showed that the answer is no: for an almost complex manifold (M, J ), the following are equivalent M is a complex manifold

⇐⇒

N ≡0

⇐⇒

∂¯ 2 = 0,

⇐⇒

d = ∂ + ∂¯

where N is the Nijenhuis tensor: N (X, Y ) := [J X, J Y ] − J [J X, Y ] − J [X, J Y ] − [X, Y ], for vector fields X and Y on M, [·, ·] being the usual bracket.31 The Nijenhuis tensor can be thought of as a measure of the existence of J -holomorphic functions: if there exist n J -holomorphic functions, f1 , . . . , fn , on R2n , that are independent at some point p, i.e., the real and imaginary parts of (df1 )p , . . . , (dfn )p form a basis of Tp∗ R2n , then N vanishes identically at p. More material related to Dolbeault theory or to the Newlander–Nirenberg theorem can be found in [23,37,62,76,141]. E XAMPLE . Out of all spheres, only S 2 and S 6 admit almost complex structures [121, §41.20]. As a complex manifold, S 2 if referred to as the Riemann sphere CP1 . The almost complex structure on S 6 from Example 3 of Section 3.2 is not integrable, but it is not yet known whether S 6 admits a structure of complex manifold. 31 The bracket of vector fields X and Y is the vector field [X, Y ] characterized by the property that L [X,Y ] f := LX (LY f ) − LY (LX f ), for f ∈ C ∞ (M), where LX f = df (X).

Symplectic geometry

117

In the (real) 2-dimensional case N always vanishes simply because N is a tensor, i.e., N (f X, gY ) = f gN (X, Y ) for any f, g ∈ C ∞ (M), and N (X, J X) = 0 for any vector field X. Combining this with the fact that any orientable surface is symplectic, we conclude that any orientable surface is a complex manifold, a result already known to Gauss. However, most almost complex structures on higher-dimensional manifolds are not integrable. In Section 3.5 we see that the existence of a complex structure compatible with a symplectic structure on a compact manifold imposes significant topological constraints. 3.4. Kähler manifolds D EFINITION 3.8. A Kähler manifold is a symplectic manifold (M, ω) equipped with an integrable compatible almost complex structure J . The symplectic form ω is then called a Kähler form. As a complex manifold, a Kähler manifold (M, ω, J ) has Dolbeault cohomology. As it is also a symplectic manifold, it is interesting to understand where the symplectic form ω sits with respect to the Dolbeault type decomposition. ¯ P ROPOSITION 3.9. A Kähler form ω is a ∂- and ∂-closed (1, 1)-form that is given on a local complex chart (U, z1 , . . . , zn ) by ω=

n i hj k dzj ∧ d z¯ k , 2 j,k=1

where, at every point p ∈ U , (hj k (p)) is a positive-definite Hermitian matrix. 1,1 (M). In particular, ω defines a Dolbeault (1, 1)-cohomology class, [ω] ∈ HDolbeault

P ROOF. Being a form in Ω 2 (M; C) = Ω 2,0 ⊕Ω 1,1 ⊕Ω 0,2 , with respect to a local complex chart, ω can be written ω= aj k dzj ∧ dzk + bj k dzj ∧ d z¯ k + cj k d z¯ j ∧ d z¯ k for some aj k , bj k , cj k ∈ C ∞ (U; C). By the compatibility of ω with the complex structure, J is a symplectomorphism, that is, J ∗ ω = ω where (J ∗ ω)(u, v) := ω(J u, J v). Since J ∗ dzj = dzj ◦ J = i dzj and J ∗ d z¯ j = d z¯ j ◦ J = −i d z¯ j , we have J ∗ ω = ω if and only if the coefficients aj k and cj k all vanish identically, that is, if and only if ω ∈ Ω 1,1 . Since ¯ we must have ∂ω = 0 and ∂ω ¯ = 0. Set ω is closed, of type (1, 1) and dω = ∂ω +∂ω, i i i bj k = 2 hj k . As ω is real-valued, i.e., ω = 2 hj k dzj ∧ d z¯ k and ω¯ = − 2 hj k d z¯ j ∧ dzk coincide, we must have hj k = hkj for all j and k. In other words, at every point p ∈ U , the n × n matrix (hj k (p)) is Hermitian. The nondegeneracy amounts to the nonvanishing of n i ω = n! det(hj k ) dz1 ∧ d z¯ 1 ∧ · · · ∧ dzn ∧ d z¯ n . 2 n

118

A. Cannas da Silva

Therefore, at every p ∈ M, the matrix (hj k (p)) must be nonsingular. Finally, the positivity condition ω(v, J v) > 0, ∀v = 0, from compatibility, implies that, at each p ∈ U , the matrix (hj k (p)) is positive-definite. Consequently, if ω0 and ω1 are both Kähler forms on a compact manifold M with 2 [ω0 ] = [ω1 ] ∈ HdeRham (M), then (M, ω0 ) and (M, ω1 ) are strongly isotopic by Moser’s Theorem 1.7. Indeed ωt = (1 − t)ω0 + tω1 is symplectic for t ∈ [0, 1], as convex combinations of positive-definite matrices are still positive-definite. Another consequence is the following recipe for Kähler forms. A smooth real function ρ on a complex manifold M is strictly plurisubharmonic (s.p.s.h.) if, on each local 2 complex chart (U, z1 , . . . , zn ), the matrix ( ∂z∂j ∂ρz¯ k (p)) is positive-definite at all p ∈ U . If ρ ∈ C ∞ (M; R) is s.p.s.h., then the form i ¯ ω = ∂ ∂ρ 2 is Kähler. The function ρ is then called a (global) Kähler potential. E XAMPLE . Let M = Cn R2n , with complex coordinates (z1 , . . . , zn ) and corresponding real coordinates (x1 , y1 , . . . , xn , yn ) via zj = xj + iyj . The function ρ(x1 , y1 , . . . , xn , yn ) =

n 2 xj + yj2 = |zj |2 = zj z¯ j j =1

is s.p.s.h. and is a Kähler potential for the standard Kähler form: i i i ¯ ∂ ∂ρ = δj k dzj ∧ d z¯ k = dzj ∧ d z¯ j = dxj ∧ dyj = ω0 . 2 2 2 j,k

j

j

There is a local converse to the previous construction of Kähler forms. P ROPOSITION 3.10. Let ω be a closed real-valued (1, 1)-form on a complex mani¯ for some fold M and let p ∈ M. Then on a neighborhood U of p we have ω = 2i ∂ ∂ρ ∞ ρ ∈ C (U; R). The proof of this theorem requires holomorphic versions of Poincaré’s lemma, namely, the local triviality of Dolbeault groups (the fact that any point in a complex manifold admits ,m a neighborhood U such that HDolbeault (U) = 0 for all m > 0) and the local triviality of the holomorphic de Rham groups; see [62]. For a Kähler ω, such a local function ρ is called a local Kähler potential. P ROPOSITION 3.11. Let M be a complex manifold, ρ ∈ C ∞ (M; R) s.p.s.h., X a complex submanifold, and i : X → M the inclusion map. Then i ∗ ρ is s.p.s.h.

Symplectic geometry

119

P ROOF. It suffices to verify this locally by considering a complex chart (z1 , . . . , zn ) for M adapted to X so that X is given there by the equations z1 = · · · = zm = 0. Being a 2 principal minor of the positive-definite matrix ( ∂z∂j ∂ z¯ k (0, . . . , 0, zm+1 , . . . , zn )) the matrix 2

∂ ρ ( ∂zm+j ∂ z¯ m+k (0, . . . , 0, zm+1 , . . . , zn )) is also positive-definite.

C OROLLARY 3.12. Any complex submanifold of a Kähler manifold is also Kähler. D EFINITION 3.13. Let (M, ω) be a Kähler manifold, X a complex submanifold, and i : X → M the inclusion. Then (X, i ∗ ω) is called a Kähler submanifold. E XAMPLES . dzj ∧ d z¯ j is Kähler. According to 1. Complex vector space (Cn , ω0 ) where ω0 = 2i Corollary 3.12, every complex submanifold of Cn is Kähler. 2. In particular, Stein manifolds are Kähler. Stein manifolds are the properly embedded complex submanifolds of Cn . They can be alternatively characterized as being the ¯ Kähler manifolds (M, ω) that admit a global proper Kähler potential, i.e., ω = 2i ∂ ∂ρ for some proper function ρ : M → R. 3. The function z → log(|z|2 + 1) on Cn is strictly plurisubharmonic. Therefore the 2-form i ωFS = ∂ ∂¯ log |z|2 + 1 2 is another Kähler form on Cn This is called the Fubini–Study form on Cn . 4. Let {(Uj , Cn , ϕj ), j = 0, . . . , n} be the usual complex atlas for complex projective space.32 The form ωFS is preserved by the transition maps, hence ϕj∗ ωFS and ϕk∗ ωFS agree on the overlap Uj ∩ Uk . The Fubini–Study form on CPn is the Kähler form obtained by gluing together the ϕj∗ ωFS , j = 0, . . . , n. 5. Consequently, all nonsingular projective varieties are Kähler submanifolds. Here by nonsingular we mean smooth, and by projective variety we mean the zero locus of some collection of homogeneous polynomials. 6. All Riemann surfaces are Kähler, since any compatible almost complex structure is integrable for dimension reasons (Section 3.3). 32 The complex projective space CPn is the complex n-dimensional manifold given by the space of complex lines in Cn+1 . It can be obtained from Cn+1 \ {0} by making the identifications (z0 , . . . , zn ) ∼ (λz0 , . . . , λzn ) for all λ ∈ C \ {0}. One denotes by [z0 , . . . , zn ] the equivalence class of (z0 , . . . , zn ), and calls z0 , . . . , zn the homogeneous coordinates of the point p = [z0 , . . . , zn ]. (Homogeneous coordinates are, of course, only determined up to multiplication by a nonzero complex number λ.) Let Uj be the subset of CPn consisting of all points p = [z0 , . . . , zn ] for which zj = 0. Let ϕj : Uj → Cn be the map defined by

z zj −1 zj +1 zn ϕj [z0 , . . . , zn ] = 0 , . . . , , ,..., . zj zj zj zj The collection {(Uj , Cn , ϕj ), j = 0, . . . , n} is the usual complex atlas for CPn . For instance, the transition map z from (U0 , Cn , ϕ0 ) to (U1 , Cn , ϕ1 ) is ϕ0,1 (z1 , . . . , zn ) = ( z1 , z2 , . . . , zzn ) defined from the set {(z1 , . . . , zn ) ∈ 1 1 1 n C | z1 = 0} to itself.

120

A. Cannas da Silva

7. The Fubini–Study form on the chart U0 = {[z0 , z1 ] ∈ CP1 | z0 = 0} of the Riemann sphere CP1 is given by the formula ωFS =

(x 2

dx ∧ dy , + y 2 + 1)2

where zz10 = z = x + iy is the usual coordinate on C. The standard area form ωstd = dθ ∧ dh is induced by regarding CP1 as the unit sphere S 2 in R3 (Example 3 of Section 1.2). Stereographic projection shows that ωFS = 14 ωstd . n n n 8. Complex tori are Kähler. Complex tori look like quotients C /Z where Z is a lattice n in C . The form ω = dzj ∧ d z¯ j induced by the Euclidean structure is Kähler. 9. Just like products of symplectic manifolds are symplectic, also products of Kähler manifolds are Kähler.

3.5. Hodge theory Hodge [73] identified the spaces of cohomology classes of forms with spaces of actual forms, by picking the representative from each class that solves a certain differential equation, namely the harmonic representative. We give a sketch of Hodge’s idea. The first part makes up ordinary Hodge theory, which works for any compact oriented Riemannian manifold (M, g), not necessarily Kähler. At a point p ∈ M, let e1 , . . . , en be a positively oriented orthonormal basis of the cotangent space Tp∗ M, with respect to the induced inner product and orientation. The Hodge star operator is the linear operator on the exterior algebra of Tp∗ M defined by ∗(1) = e1 ∧ · · · ∧ en , ∗(e1 ∧ · · · ∧ en ) = 1, ∗(e1 ∧ · · · ∧ ek ) = ek+1 ∧ · · · ∧ en . We see that ∗ : Λk (Tp∗ M) → Λn−k (Tp∗ M) and satisfies ∗∗ = (−1)k(n−k) . The codifferential and the Laplacian are the operators defined by δ = (−1)n(k+1)+1 ∗ d∗ : Ω k (M) → Ω k−1 (M), = dδ + δd

: Ω k (M) → Ω k (M).

The operator is also called the Laplace–Beltrami operator and satisfies ∗ = ∗. On 2 Ω 0 (Rn ) = C ∞ (Rn ), it is simply the usual Laplacian = − ni=1 ∂ 2 . The inner product ∂xi

on forms of any degree, ·, · : Ω (M) × Ω (M) −→ R, k

k

α, β :=

α ∧ ∗β, M

Symplectic geometry

121

satisfies dα, β = α, δβ, so the codifferential δ is often denoted by d ∗ and called the adjoint33 of d. Also, is self-adjoint (i.e., α, β = α, β), and α, α = |dα|2 + |δα|2 0, where | · | is the norm with respect to this inner product. The harmonic k-forms are the elements of Hk := {α ∈ Ω k | α = 0}. Note that α = 0 if and only if dα = δα = 0. Since a harmonic form is d-closed, it defines a de Rham cohomology class. T HEOREM 3.14 (Hodge). Every de Rham cohomology class on a compact oriented Riemannian manifold (M, g) possesses a unique harmonic representative, i.e., there is an k isomorphism Hk HdeRham (M; R). In particular, the spaces Hk are finite-dimensional. We also have the following orthogonal decomposition with respect to the inner product on forms: Ω k Hk ⊕ (Ω k (M)) Hk ⊕ dΩ k−1 ⊕ δΩ k+1 . This decomposition is called the Hodge decomposition on forms. The proof of this and the next theorem involves functional analysis, elliptic differential operators, pseudodifferential operators and Fourier analysis; see for instance [62,83,141]. Here is where complex Hodge theory begins. When M is Kähler, the Laplacian satisfies ¯ (see, for example, [62]) and preserves the decomposition according to = 2(∂¯ ∂¯ ∗ + ∂¯ ∗ ∂) type, : Ω ,m → Ω ,m . Hence, harmonic forms are also bigraded Hk =

H,m

+m=k

and satisfy a Künneth formula H,m (M × N )

%

p+r=, q+s=m H

p,q (M) ⊗ Hr,s (N ).

T HEOREM 3.15 (Hodge). Every Dolbeault cohomology class on a compact Kähler manifold (M, ω) possesses a unique harmonic representative, i.e., there is an isomorphism ,m H,m HDolbeault (M). Combining the two theorems of Hodge, we find the decomposition of cohomology groups for a compact Kähler manifold k HdeRham (M; C)

,m HDolbeault (M),

+m=k

known as the Hodge decomposition. In particular, the Dolbeault cohomology groups ,m HDolbeault are finite-dimensional and H ,m H m, . k k (M) be the usual Betti numbers of M, and let h,m (M) := Let b (M) := dim HdeRham ,m dim HDolbeault (M) be the Hodge numbers of M. For an arbitrary compact symplectic manifold (M, ω), the even Betti numbers must be k positive, because ωk is closed but not

exact (k = 0, 1, . . . , n). In fact, if it were ω = dα, by Stokes’ theorem we would have M ωn = M d(α ∧ ωn−k ) = 0, which contradicts ωn being a volume form. 33 When M is not compact, we still have a formal adjoint of d with respect to the nondegenerate bilinear pairing ·, · : Ω k (M) × Ωck (M) → R defined by a similar formula, where Ωck (M) is the space of compactly supported k-forms.

122

A. Cannas da Silva

For a compact Kähler manifold (M, ω), there are finer topological consequences coming from the Hodge theorems, as we must have bk = +m=k h,m and h,m = hm, . The odd Betti numbers must be even because b2k+1 = +m=2k+1 h,m = 2 k=0 h,(2k+1−) . The number h1,0 = 12 b1 must be a topological invariant. The numbers h, are positive, , , (M). First of all, [ω ] defines an element of HDolbeault as because 0 = [ω ] ∈ HDolbeault 1,1 , ¯ ω ∈ Ω implies that ω ∈ Ω , and the closedness of ω implies that ∂ω = 0. If it were ¯ for some β ∈ Ω −1, , then ωn = ω ∧ ωn− = ∂(β ¯ ∧ ωn− ) would be ∂-exact. ¯ But ω = ∂β n,n 2n n [ω ] = 0 in HdeRham (M; C) HDolbeault (M) since it is a volume form. A popular diagram to describe relations among Hodge numbers is the Hodge diamond: hn,n hn,n−1 hn,n−2

hn−1,n hn−1,n−1

hn−2,n

.. .

... h2,0

...

h1,1 h1,0

h0,2 h0,1

h0,0 Complex conjugation gives symmetry with respect to the middle vertical, whereas the Hodge star operator induces symmetry about the center of the diamond. The middle vertical axis is all nonzero. ,m for There are further symmetries and ongoing research on how to compute HDolbeault a compact Kähler manifold (M, ω). In particular, the hard Lefschetz theorem states iso n−k n+k morphisms Lk : HdeRham (M) −→ HdeRham (M) given by wedging with ωk at the level of % m forms and the Lefschetz decompositions HdeRham (M) k Lk (ker Ln−m+2k+1 |H m−2k ). The Hodge conjecture claims, for projective manifolds M (i.e., complex submanifolds , (M) ∩ of complex projective space), that the Poincaré duals of elements in HDolbeault H 2 (M; Q) are rational linear combinations of classes of complex codimension subvarieties of M. This has been proved only for the = 1 case (it is the Lefschetz theorem on (1, 1)-classes; see, for instance, [62]).

3.6. Pseudoholomorphic curves Whereas an almost complex manifold (M, J ) tends to have no J -holomorphic functions M → C at all,34 it has plenty of J -holomorphic curves C → M. Gromov first realized that pseudoholomorphic curves provide a powerful tool in symplectic topology in an extremely influential paper [64]. Fix a closed Riemann surface (Σ, j ), that is, a compact complex 1-dimensional manifold Σ without boundary and equipped with the canonical almost complex structure j . 34 However, the study of asymptotically J -holomorphic functions has been recently developed to obtain important results [32,34,13]; see Section 4.6.

Symplectic geometry

123

D EFINITION 3.16. A parametrized pseudoholomorphic curve (or J -holomorphic curve) in (M, J ) is a (smooth) map u : Σ → M whose differential intertwines j and J , that is, dup ◦ jp = Jp ◦ dup , ∀p ∈ Σ. In other words, the Cauchy–Riemann equation du + J ◦ du ◦ j = 0 holds. Pseudoholomorphic curves are related to parametrized 2-dimensional symplectic submanifolds. If a pseudoholomorphic curve u : Σ → M is an embedding, then its image S := u(Σ) is a 2-dimensional almost complex submanifold, hence a symplectic submanifold. Conversely, the inclusion i : S → M of a 2-dimensional symplectic submanifold can be seen as a pseudoholomorphic curve. An appropriate compatible almost complex structure J on (M, ω) can be constructed starting from S, such that T S is J -invariant. The restriction j of J to T S is necessarily integrable because S is 2-dimensional. The group G of complex diffeomorphisms of (Σ, j ) acts on (parametrized) pseudoholomorphic curves by reparametrization: u → u ◦ γ , for γ ∈ G. This normally means that each curve u has a noncompact orbit under G. The orbit space Mg (A, J ) is the set of unparametrized pseudoholomorphic curves in (M, J ) whose domain Σ has genus g and whose image u(Σ) has homology class A ∈ H2 (M; Z). The space Mg (A, J ) is called the moduli space of unparametrized pseudoholomorphic curves of genus g representing the class A. For generic J , Fredholm theory shows that pseudoholomorphic curves occur in finite-dimensional smooth families, so that the moduli spaces Mg (A, J ) can be manifolds, after avoiding singularities given by multiple coverings.35 E XAMPLE . Usually Σ is the Riemann sphere CP1 , whose complex diffeomorphisms are those given by fractional linear transformations (or Möbius transformations). So the 6-dimensional noncompact group of projective linear transformations PSL(2; C) acts on pseudoholomorphic spheres by reparametrization u → u ◦ γA , where A = ac db ∈ PSL(2; C) acts by γA : CP1 → CP1 , γA [z, 1] = [ az+b cz+d , 1]. When J is an almost complex structure compatible with a symplectic form ω, the area of the image of a pseudoholomorphic curve u (with respect to the metric gJ (·, ·) = ω(·, J ·)) is determined by the class A that it represents. The number E(u) := ω(A) = u∗ ω = area of the image of u with respect to gJ Σ

is called the energy of the curve u and is a topological invariant: it only depends on [ω] and on the homotopy class of u. Gromov proved that the constant energy of all the pseudoholomorphic curves representing a homology class A ensured that the space Mg (A, J ), though not necessarily compact, had natural compactifications Mg (A, J ) by including what he called cusp-curves. T HEOREM 3.17 (Gromov’s compactness theorem). If (M, ω) is a compact manifold equipped with a generic compatible almost complex structure J , and if uj is a sequence 35 A curve u : Σ → M is a multiple covering if u factors as u = u ◦ σ where σ : Σ → Σ is a holomorphic map of degree greater than 1.

124

A. Cannas da Silva

of pseudoholomorphic curves in Mg (A, J ), then there is a subsequence that weakly converges to a cusp-curve in Mg (A, J ). Hence the cobordism class of the compactified moduli space Mg (A, J ) might be a nice symplectic invariant of (M, ω), as long as it is not empty or null-cobordant. Actually a nontrivial regularity criterion for J ensures the existence of pseudoholomorphic curves. And even when Mg (A, J ) is null-cobordant, we can define an invariant to be the (signed) number of pseudoholomorphic curves of genus g in class A that intersect a specified set of representatives of homology classes in M [112,128,145]. For more on pseudoholomorphic curves; see, for instance, [100] (for a comprehensive discussion of the genus 0 case) or [11] (for higher genus). Here is a selection of applications of (developments from) pseudoholomorphic curves: • Proof of the nonsqueezing theorem [64]: for R > r there is no symplectic embedding of a ball BR2n of radius R into a cylinder Br2 × R2n−2 of radius r, both in (R2n , ω0 ). • Proof that there are no Lagrangian spheres in (Cn , ω0 ), except for the circle in C2 , and more generally no compact exact Lagrangian submanifolds, in the sense that the tautological 1-form α restricts to an exact form [64]. • Proof that if (M, ω) is a connected symplectic 4-manifold symplectomorphic to (R4 , ω0 ) outside a compact set and containing no symplectic S 2 ’s, then (M, ω) symplectomorphic to (R4 , ω0 ) [64]. • Study questions of symplectic packing [15,98,134] such as: for a given 2n-dimensional symplectic manifold (M, ω), what is the maximal radius R for which there is a symplectic embedding of N disjoint balls BR2n into (M, ω)? • Study groups of symplectomorphisms of 4-manifolds (for a review see [97]). Gromov [64] showed that Sympl(CP2 , ωFS ) and Sympl(S 2 × S 2 , pr∗1 σ ⊕ pr∗2 σ ) deformation retract onto the corresponding groups of standard isometries. • Development of Gromov–Witten invariants allowing to prove, for instance, the nonexistence of symplectic forms on CP2 # CP2 # CP2 or the classification of symplectic structures on ruled surfaces (Section 4.3). • Development of Floer homology to prove the Arnold conjecture on the fixed points of symplectomorphisms of compact symplectic manifolds, or on the intersection of Lagrangian submanifolds (Section 5.2). • Development of symplectic field theory introduced by Eliashberg, Givental and Hofer [40] extending Gromov–Witten theory, exhibiting a rich algebraic structure and also with applications to contact geometry.

4. Symplectic geography 4.1. Existence of symplectic forms The utopian goal of symplectic classification addresses the standard questions: • (Existence) Which manifolds carry symplectic forms? • (Uniqueness) What are the distinct symplectic structures on a given manifold?

Symplectic geometry

125

even-dimensional orientable

almost complex symplectic complex Kähler

Fig. 1.

Existence is tackled through central examples in this subsection and symplectic constructions in the next two sections. Uniqueness is treated in the remainder of this subsection dealing with invariants that allow to distinguish symplectic manifolds. A Kähler structure naturally yields both a symplectic form and a complex structure (compatible ones). Either a symplectic or a complex structure on a manifold implies the existence of an almost complex structure. Figure 1 represents the relations among these structures. In dimension 2, orientability trivially guarantees the existence of all other structures, so the picture collapses. In dimension 4, the first interesting dimension, the picture above is faithful—we will see that there are closed 4-dimensional examples in each region. Closed here means compact and without boundary. Not all 4-dimensional manifolds are almost complex. A result of Wu [146] gives a necessary and sufficient condition in terms of the signature σ and the Euler characteristic χ of a 4-dimensional closed manifold M for the existence of an almost complex structure: 3σ + 2χ = h2 for some h ∈ H 2 (M; Z) congruent with the second Stiefel–Whitney class w2 (M) modulo 2. For example, S 4 and (S 2 × S 2 ) # (S 2 × S 2 ) are not almost complex. When an almost complex structure exists, the first Chern class of the tangent bundle (regarded as a complex vector bundle) satisfies the condition for h. The sufficiency of Wu’s condition is the remarkable part.36 According to Kodaira’s classification of closed complex surfaces [82], such a surface admits a Kähler structure if and only if its first Betti number b1 is even. The necessity of this condition is a Hodge relation on the Betti numbers (Section 3.5). The complex projective plane CP2 with the Fubini–Study form (Section 3.4) might be called the simplest example of a closed Kähler 4-manifold. The Kodaira–Thurston example [131] first demonstrated that a manifold that admits both a symplectic and a complex structure does not have to admit any Kähler structure. 36 Moreover, such solutions h are in one-to-one correspondence with isomorphism classes of almost complex structures.

126

A. Cannas da Silva

even-dimensional and simply connected almost complex (and simply connected) symplectic (and simply connected) complex (and simply connected)

Fig. 2.

Take R4 with dx1 ∧ dy1 + dx2 ∧ dy2 , and Γ the discrete group generated by the four symplectomorphisms: (x1 , x2 , y1 , y2 ) −→ (x1 + 1, x2 , y1 , y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 + 1, y1 , y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 + y2 , y1 + 1, y2 ), (x1 , x2 , y1 , y2 ) −→ (x1 , x2 , y1 , y2 + 1). Then M = R4 /Γ is a symplectic manifold that is a 2-torus bundle over a 2-torus. Kodaira’s classification [82] shows that M has a complex structure. However, π1 (M) = Γ , hence H1 (R4 /Γ ; Z) = Γ /[Γ, Γ ] has rank 3, so b1 = 3 is odd. Fernández–Gotay–Gray [44] first exhibited symplectic manifolds that do not admit any complex structure at all. Their examples are circle bundles over circle bundles (i.e., a tower of circle bundles) over a 2-torus. The Hopf surface is the complex surface diffeomorphic to S 1 × S 3 obtained as the quotient C2 \ {0}/Γ where Γ = {2n Id | n ∈ Z} is a group of complex transformations, i.e., we factor C2 \ {0} by the equivalence relation (z1 , z2 ) ∼ (2z1 , 2z2 ). The Hopf surface is not symplectic because H 2 (S 1 × S 3 ) = 0. The manifold CP2 # CP2 # CP2 is almost complex but is neither complex (since it does not fit Kodaira’s classification [82]), nor symplectic as shown by Taubes [126] using Seiberg–Witten invariants (Section 4.5). We could go through the previous discussion restricting to closed 4-dimensional examples with a specific fundamental group. We will do this restricting to simply connected examples, where Figure 2 holds. It is a consequence of Wu’s result [146] that a simply connected manifold admits an almost complex structure if and only if b2+ is odd.37 In particular, the connected sum 37 The intersection form of an oriented topological closed 4-manifold M is the bilinear pairing Q : H 2 (M; Z)× M H 2 (M; Z) → Z, QM (α, β) := α ∪ β, [M], where α ∪ β is the cup product and [M] is the fundamental class.

Symplectic geometry

127

#m CP2 #n CP2 (of m copies of CP2 with n copies of CP2 ) has an almost complex structure if and only if m is odd.38 By Kodaira’s classification [82], a simply connected complex surface always admits a compatible symplectic form (since b1 = 0 is even), i.e., it is always Kähler. Since they are simply connected, S 4 , CP2 # CP2 # CP2 and CP2 live in three of the four regions in the picture for simply connected examples. All of CP2 #m CP2 are also simply connected Kähler manifolds because they are pointwise blow-ups CP2 and the blow-down map is holomorphic; see Section 4.3. There is a family of manifolds obtained from CP2 #9 CP2 =: E(1) by a knot surgery [45] that were shown by Fintushel and Stern to be symplectic and confirmed not to admit a complex structure [109]. The first example of a closed simply connected symplectic manifold that cannot be Kähler, was a 10-dimensional manifold obtained by McDuff [94] as follows. The Kodaira–Thurston example R4 /Γ (not simply connected) embeds symplectically in (CP5 , ωFS ) [65,132]. McDuff’s example is a blow-up of (CP5 , ωFS ) along the image of R4 /Γ . Geography problems are problems on the existence of simply connected closed oriented 4-dimensional manifolds with some additional structure (such as, a symplectic form or a complex structure) for each pair of topological coordinates. As a consequence of the work of Freedman [51] and Donaldson [30] in the 80’s, it became known that the homeomorphism class of a connected simply connected closed oriented smooth 4-manifold is determined by the two integers—the second Betti number and the signature (b2 , σ )— and the parity39 of the intersection form. Forgetting about the parity, the numbers (b2 , σ ) can be treated as topological coordinates. For each pair (b2 , σ ) there could well be infinite different (i.e., nondiffeomorphic) smooth manifolds. Using Riemannian geometry, Cheeger [22] showed that there are at most countably many different smooth types for closed 4-manifolds. There are no known finiteness results for the smooth types of a given topological 4-manifold, in contrast to other dimensions. Traditionally, the numbers used are (c12 , c2 ) := (3σ + 2χ, χ) = (3σ + 4 + 2b2 , 2 + b2 ), and frequently just the slope c12 /c2 is considered. If M admits an almost complex structure J , then (T M, J ) is a complex vector bundle, hence has Chern classes c1 = c1 (M, J ) and c2 = c2 (M, J ). Both c12 := c1 ∪ c1 and c2 may be regarded as numbers since H 4 (M; Z) Z. They satisfy c12 = 3σ +2χ (by Hirzebruch’s signature formula) and c2 = χ (because the top Chern class is always the Euler class), justifying the notation for the topological coordinates in this case. Since QM always vanishes on torsion elements, descending to H 2 (M; Z)/torsion it can be represented by a ma trix. When M is smooth and simply connected, this pairing is QM (α, β) := M α ∧ β since nontorsion elements are representable by 2-forms. As QM is symmetric (in the smooth case, the wedge product of 2-forms is symmetric) and unimodular (the determinant of a matrix representing QM is ±1 by Poincaré duality), it is diagonalizable over R with eigenvalues ±1. We denote by b2+ (respectively b2− ) the number of positive (respectively negative) eigenvalues of QM counted with multiplicities, i.e., the dimension of a maximal subspace where QM is positivedefinite (respectively negative-definite). The signature of M is the difference σ := b2+ − b2− , whereas the second Betti number is the sum b2 = b2+ + b2− , i.e., the rank of QM . The type of an intersection form is definite if it is positive or negative definite (i.e., |σ | = b2 ) and indefinite otherwise. 38 The intersection form of a connected sum M 0 # M1 is (isomorphic to) QM0 ⊕ QM1. 39 We say that the parity of an intersection form Q is even when Q (α, α) is even for all α ∈ H 2 (M; Z), and M M odd otherwise.

128

A. Cannas da Silva

E XAMPLES . The manifold CP2 has (b2 , σ ) = (1, 1), i.e., (c12 , c2 ) = (9, 3). Reversing the orientation CP2 has (b2 , σ ) = (1, −1), i.e., (c12 , c2 ) = (3, 3). Their connected sum CP2 # CP2 has (b2 , σ ) = (2, 0), i.e., (c12 , c2 ) = (8, 0). The product S 2 × S 2 also has (b2 , σ ) = (2, 0), i.e., (c12 , c2 ) = (8, 4). But CP2 # CP2 has an odd intersection form 0 whereas S 2 × S 2 has an even intersection form: 10 −1 vs. 01 01 . Symplectic geography [60,122] addresses the following question: What is the set of pairs of integers (m, n) ∈ Z × Z for which there exists a connected simply connected closed symplectic 4-manifold M having second Betti number b2 (M) = m and signature σ (M) = n? This problem includes the usual geography of simply connected complex surfaces, since all such surfaces are Kähler according to Kodaira’s classification [82]. Often, instead of the numbers (b2 , σ ), the question is equivalently phrased in terms of the Chern numbers (c12 , c2 ) for a compatible almost complex structure, which satisfy c12 = 3σ + 2χ [146] and c2 = χ , where χ = b2 + 2 is the Euler number. Usually only minimal (Section 4.3) or irreducible manifolds are considered to avoid trivial examples. A manifold is irreducible when it is not a connected sum of other manifolds, except when one of the summands is a homotopy sphere. It was speculated that perhaps any simply connected closed smooth 4-manifold other than S 4 is diffeomorphic to a connected sum of symplectic manifolds, where any orientation is allowed on each summand (the so-called minimal conjecture for smooth 4-manifolds). Szabó [124,125] provided counterexamples in a family of irreducible simply connected closed nonsymplectic smooth 4-manifolds. All these problems could be posed for other fundamental groups. Gompf [57] used symplectic sums (Section 4.2) to prove the following theorem. He also proved that his surgery construction can be adapted to produce non-Kähler examples. Since finitely-presented groups are not classifiable, this shows that compact symplectic 4-manifold are not classifiable. T HEOREM 4.1 (Gompf). Every finitely-presented group occurs as the fundamental group π1 (M) of a compact symplectic 4-manifold (M, ω).

4.2. Fibrations and sums Products of symplectic manifolds are naturally symplectic. As we will see, special kinds of twisted products, i.e., fibrations,40 are also symplectic. 40 A fibration (or fiber bundle) is a manifold M (called the total space) with a submersion π : M → X to a manifold X (the base) that is locally trivial in the sense that there is an open covering of X, such that, to each set U in that covering corresponds a diffeomorphism of the form ϕU = (π, sU ) : π −1 (U) → U × F (a local trivialization) where F is a fixed manifold (the model fiber). A collection of local trivializations such that the sets U cover X is called a trivializing cover for π . Given two local trivializations, the second entry of the composition −1 = (id, ψU V ) on (U ∩ V) × F gives the corresponding transition function ψU V (x) : F → F at each ϕV ◦ ϕU x ∈ U ∩ V.

Symplectic geometry

129

D EFINITION 4.2. A symplectic fibration is a fibration π : M → X where the model fiber is a symplectic manifold (F, σ ) and with a trivializing cover for which all the transition functions are symplectomorphisms F → F . In a symplectic fibration each fiber π −1 (x) carries a canonical symplectic form σx defined by the restriction of sU∗ σ , for any domain U of a trivialization covering x (i.e., x ∈ U ). A symplectic form ω on the total space M of a symplectic fibration is called compatible with the fibration if each fiber (π −1 (x), σx ) is a symplectic submanifold of (M, ω), i.e., σx is the restriction of ω to π −1 (x). E XAMPLES . 1. Every compact oriented41 fibration whose model fiber F is an oriented surface admits a structure of symplectic fibration for the following reason. Let σ0 be an area form on F . Each transition function ψUV (x) : F → F pulls σ0 back to a cohomologous area form σ1 (depending on ψUV (x)). Convex combinations σt = (1 − t)σ0 + tσ1 give a path of area forms from σ0 to σ1 with constant class [σt ]. By Moser’s argument (Section 1.4), there exists a diffeomorphism ρ(x) : F → F isotopic to the identity, depending smoothly on x ∈ U ∩ V, such that ψUV (x) ◦ ρ(x) is a symplectomorphism of (F, σ0 ). By successively adjusting local trivializations for a finite covering of the base, we can make all transition functions into symplectomorphisms. 2. Every fibration with connected base and compact fibers having a symplectic form ω for which all fibers are symplectic submanifolds admits a structure of symplectic fibration compatible with ω. Indeed, under trivializations, the restrictions of ω to the fibers give cohomologous symplectic forms in the model fiber F . So by Moser’s Theorem 1.7, all fibers are strongly isotopic to (F, σ ) where σ is the restriction of ω to a chosen fiber. These isotopies can be used to produce a trivializing cover where each sU (x) is a symplectomorphism. In the remainder of this subsection, assume that for a fibration π : M → X the total space is compact and the base is connected. For the existence of a compatible symplectic form on a symplectic fibration, a necessary condition is the existence of a cohomology class in M that restricts to the classes of the fiber symplectic forms. Thurston [131] showed that, when the base admits also a symplectic form, this condition is sufficient. Yet not all symplectic fibrations with a compatible symplectic form have a symplectic base [138]. T HEOREM 4.3 (Thurston). Let π : M → X be a compact symplectic fibration with connected symplectic base (X, α) and model fiber (F, σ ). If there is a class [ν] ∈ H 2 (M) pulling back to [σ ], then, for sufficiently large k > 0, there exists a symplectic form ωk on M that is compatible with the fibration and is in [ν + kπ ∗ α]. P ROOF. We first find a form τ on M in the class [ν] that restricts to the canonical symplectic form on each fiber. Pick a trivializing cover {ϕi = (π, si ) | i ∈ I } with contractible 41 An oriented fibration is a fibration whose model fiber is oriented and there is a trivializing cover for which all

transition functions preserve orientation.

130

A. Cannas da Silva

domains Ui . Let ρi , i ∈ I , be a partition of unity subordinate to this covering and let ρ˜i := ρi ◦ π : M → R. Since [ν] always restricts to the class of the canonical symplectic form [σx ], and the Ui ’s are contractible, on each πi−1 (Ui ) the forms si∗ σ − ν are exact. Choose 1-forms λi such that si∗ σ = ν + dλi , and set τ := ν +

d(ρ˜i λi ).

i∈I

Since τ is nondegenerate on the (vertical) subbundle given by the kernel of dπ , for k > 0 large enough the form τ + kπ ∗ α is nondegenerate on M. C OROLLARY 4.4. Let π : M → X be a compact oriented fibration with connected symplectic base (X, α) and model fiber an oriented surface F of genus g(F ) = 1. Then π admits a compatible symplectic form. P ROOF. By Example 1 above, π : M → X admits a structure of symplectic fibration with model fiber (F, σ ). Since the fiber is not a torus (g(F ) = 1), the Euler class of the tangent bundle T F (which coincides with c1 (F, σ )) is λ[σ ] for some λ = 0. Hence, the first Chern class [c] of the vertical subbundle given by the kernel of dπ (assembling the tangent bundles to the fibers) restricts to λ[σx ] on the fiber over x ∈ X. We can apply Theorem 4.3 using the class [ν] = λ−1 [c]. A pointwise connected sum M0 # M1 of symplectic manifolds (M0 , ω0 ) and (M1 , ω1 ) tends to not admit a symplectic form, even if we only require the eventual symplectic form to be isotopic to ωi on each Mi minus a ball. The reason [7] is that such a symplectic form on M0 # M1 would allow to construct an almost complex structure on the sphere formed by the union of the two removed balls, which is known not to exist except on S 2 and S 6 . Therefore: P ROPOSITION 4.5. Let (M0 , ω0 ) and (M1 , ω1 ) be two compact symplectic manifolds of dimension not 2 nor 6. Then the connected sum M0 # M1 does not admit any symplectic structure isotopic to ωi on Mi minus a ball, i = 1, 2. For connected sums to work in the symplectic category, they should be done along codimension 2 symplectic submanifolds. The following construction, already mentioned in [65], was dramatically explored and popularized by Gompf [57] (he used it to prove Theorem 4.1). Let (M0 , ω0 ) and (M1 , ω1 ) be two 2n-dimensional symplectic manifolds. Suppose that a compact symplectic manifold (X, α) of dimension 2n − 2 admits symplectic embeddings to both i0 : X → M0 , i1 : X → M1 . For simplicity, assume that the corresponding normal bundles are trivial (in general, they need to have symmetric Euler classes). By the symplectic neighborhood theorem (Theorem 1.11), there exist symplectic embeddings j0 : X × Bε → M0 and j1 : X × Bε → M1 (called framings) where Bε is a ball of radius ε and centered at the origin in R2 such that jk∗ ωk = α + dx ∧ dy and jk (p, 0) = ik (p), ∀p ∈ X, k = 0, 1. Chose an area- and orientation-preserving diffeomorphism φ of the annulus Bε \ Bδ for 0 < δ < ε that interchanges the two boundary compo-

Symplectic geometry

131

nents. Let Uk = jk (X × Bδ ) ⊂ Mk , k = 0, 1. A symplectic sum of M0 and M1 along X is defined to be M0 #X M1 := (M0 \ U0 ) ∪φ (M1 \ U1 ), where the symbol ∪φ means that we identify j1 (p, q) with j0 (p, φ(q)) for all p ∈ X and δ < |q| < ε. As ω0 and ω1 agree on the regions under identification, they induce a symplectic form on M0 #X M1 . The result depends on j0 , j1 , δ and φ. Rational blow-down is a surgery on 4-manifolds that replaces a neighborhood of a chain of embedded S 2 ’s with boundary a lens space L(n2 , n − 1) by a manifold with the same rational homology as a ball. This simplifies the homology possibly at the expense of complicating the fundamental group. Symington [123] showed that rational blow-down preserves a symplectic structure if the original spheres are symplectic surfaces in a symplectic 4-manifold.

4.3. Symplectic blow-up Symplectic blow-up is the extension to the symplectic category of the blow-up operation in algebraic geometry. It is due to Gromov according to the first printed exposition of this operation in [94]. Let L be the tautological line bundle over CPn−1 , that is, L = [p], z | p ∈ Cn \ {0}, z = λp for some λ ∈ C with projection to CPn−1 given by π : ([p], z) → [p]. The fiber of L over the point [p] ∈ CPn−1 is the complex line in Cn represented by that point. The blow-up of Cn at the ˜ n . The corresponding bloworigin is the total space of the bundle L, sometimes denoted C n down map is the map β : L → C defined by β([p], z) = z. The total space of L may be decomposed as the disjoint union of two sets: the zero section E := [p], 0 | p ∈ Cn \ {0} and S :=

[p], z | p ∈ Cn \ {0}, z = λp for some λ ∈ C∗ .

The set E is called the exceptional divisor; it is diffeomorphic to CPn−1 and gets mapped to the origin by β. On the other hand, the restriction of β to the complementary set S is a diffeomorphism onto Cn \ {0}. Hence, we may regard L as being obtained from Cn by smoothly replacing the origin by a copy of CPn−1 . Every biholomorphic map f : Cn → Cn with f (0) = 0 lifts uniquely to a biholomorphic map f˜ : L → L with f˜(E) = E. The lift is given by the formula ([f (z)], f (z)) if z = 0, ˜ f [p], z = ([p], 0) if z = 0.

132

A. Cannas da Silva

There are actions of the unitary group U(n) on L, E and S induced by the standard linear action on Cn , and the map β is U(n)-equivariant. For instance, β ∗ ω0 + π ∗ ωFS is a U(n)invariant Kähler form on L. D EFINITION 4.6. A blow-up symplectic form on the tautological line bundle L is a U(n)invariant symplectic form ω such that the difference ω − β ∗ ω0 is compactly supported, i n where ω0 = 2 k=1 dzk ∧ d z¯ k is the standard symplectic form on Cn . Two blow-up symplectic forms are equivalent if one is the pullback of the other by a U(n)-equivariant diffeomorphism of L. Guillemin and Sternberg [69] showed that two blow-up symplectic forms are equivalent if and only if they have equal restrictions to the exceptional divisor E ⊂ L. Let Ω ε (ε > 0) be the set of all blow-up symplectic forms on L whose restriction to the exceptional divisor E CPn−1 is εωFS , where ωFS is the Fubini–Study form (Section 3.4). An ε-blow-up of Cn at the origin is a pair (L, ω) with ω ∈ Ωε. Let (M, ω) be a 2n-dimensional symplectic manifold. It is a consequence of Darboux’s theorem (Theorem 1.9) that, for each point p ∈ M, there exists a complex chart (U, z1 , . . . , zn ) centered at p and with image in Cn where ω|U = 2i nk=1 dzk ∧ d z¯ k . It is shown in [69] that, for ε small enough, we can perform an ε-blow-up of M at p modeled on Cn at the origin, without changing the symplectic structure outside of a small neighborhood of p. The resulting manifold is called an ε-blow-up of M at p. As a manifold, the blow-up of M at a point is diffeomorphic to the connected sum42 M # CPn , where CPn is the manifold CPn equipped with the orientation opposite to the natural complex one. E XAMPLE . Let P(L ⊕ C) be the CP1 -bundle over CPn−1 obtained by projectivizing the direct sum of the tautological line bundle L with a trivial complex line bundle. Consider the map β : CP(L ⊕ C) −→ CPn , [p], [λp : w] −→ [λp : w], where [λp : w] on the right represents a line in Cn+1 , forgetting that, for each [p] ∈ CPn−1 , that line sits in the 2-complex-dimensional subspace L[p] ⊕ C ⊂ Cn ⊕ C. Notice that β maps the exceptional divisor E :=

[p], [0 : . . . : 0 : 1] | [p] ∈ CPn−1 CPn−1

to the point [0 : . . . : 0 : 1] ∈ CPn , and β is a diffeomorphism on the complement S :=

[p], [λp : w] | [p] ∈ CPn−1 , λ ∈ C∗ , w ∈ C CPn \ [0 : . . . : 0 : 1] .

42 The connected sum of two oriented m-dimensional manifolds M and M is the manifold, denoted M 0 1 0 # M1 , obtained from the union of those manifolds each with a small ball removed Mi \ Bi by identifying the boundaries via a (smooth) map φ : ∂B1 → ∂B2 that extends to an orientation-preserving diffeomorphism of neighborhoods of ∂B1 and ∂B2 (interchanging the inner and outer boundaries of the annuli).

Symplectic geometry

133

Therefore, we may regard CP(L ⊕ C) as being obtained from CPn by smoothly replacing the point [0 : . . . : 0 : 1] by a copy of CPn−1 . The space CP(L ⊕ C) is the blow-up of CPn at the point [0 : . . . : 0 : 1], and β is the corresponding blow-down map. The manifold CP(L ⊕ C) for n = 2 is a Hirzebruch surface. When (CPn−1 , ωFS ) is symplectically embedded in a symplectic manifold (M, ω) with image X and normal bundle isomorphic to the tautological bundle L, it can be subject to a blow-down operation. By the symplectic neighborhood theorem (Theorem 1.11), some neighborhood U ⊂ M of the image X is symplectomorphic to a neighborhood U0 ⊂ L of the zero section. It turns out that some neighborhood of ∂U0 in L is symplectomorphic to a spherical shell in (Cn , ω0 ). The blow-down of M along X is a manifold obtained from the union of M \ U with a ball in Cn . For more details, see [99, §7.1]. Following algebraic geometry, we call minimal a 2n-dimensional symplectic manifold (M, ω) without any symplectically embedded (CPn−1 , ωFS ), so that (M, ω) is not the blow-up at a point of another symplectic manifold. In dimension 4, a manifold is minimal if it does not contain any embedded sphere S 2 with self-intersection −1. Indeed, by the work of Taubes [126,129], if such a sphere S exists, then either the homology class [S] or its symmetric −[S] can be represented by a symplectically embedded sphere with self-intersection −1. For a symplectic manifold (M, ω), let i : X → M be the inclusion of a symplectic submanifold. The normal bundle N X to X in M admits a structure of complex vector bundle (as it is a symplectic vector bundle). Let P(N X) → X be the projectivization of the bundle NX → X, let Z be the zero section of N X, let L(N X) be the corresponding tautological line bundle (given by assembling the tautological line bundles over each fiber) and let β : L(N X) → N X be the blow-down map. On the exceptional divisor E :=

[p], 0 ∈ L(NX) | p ∈ N X \ Z P(N X)

the map β is just projection to the zero section Z. The restriction of β to the complement L(NX) \ E is a diffeomorphism to N X \ Z. Hence, L(N X) may be viewed as being obtained from N X by smoothly replacing each point of the zero section by the projectivization of its normal space. We symplectically identify some tubular neighborhood U of X in M with a tubular neighborhood U0 of the zero section Z in N X. A blow-up of the symplectic manifold (M, ω) along the symplectic submanifold X is the manifold obtained from the union of M \ U and β −1 (U0 ) by identifying neighborhoods of ∂U , and equipped with a symplectic form that restricts to ω on M \ U [94]. When X is one point, this construction reduces to the previous symplectic blow-up at a point. Often symplectic geography concentrates on minimal examples. McDuff [95] showed that a minimal symplectic 4-manifold with a symplectically embedded S 2 with nonnegative self-intersection is symplectomorphic either to CP2 or to an S 2 -bundle over a surface. Using Seiberg–Witten theory it was proved: T HEOREM 4.7. Let (M, ω) be a minimal closed symplectic 4-manifold. (a) (Taubes [129]) If b2+ > 1, then c12 0.

134

A. Cannas da Silva

(b) (Liu [89]) If b2+ = 1 and c12 < 0, then M is the total space of an S 2 -fibration over a surface of genus g where ω is nondegenerate on the fibers, and (c12 , c2 ) = (8 − 8g, 4 − 4g), i.e., (M, ω) is a symplectic ruled surface. A symplectic ruled surface43 is a symplectic 4-manifold (M, ω) that is the total space of an S 2 -fibration where ω is nondegenerate on the fibers. A symplectic rational surface is a symplectic 4-manifold (M, ω) that can be obtained from the standard (CP2 , ωFS ) by blowing up and blowing down. With b2+ = 1 and c12 = 0, we have symplectic manifolds CP2 #9 CP2 =: E(1), the Dolgachev surfaces E(1, p, q), the results E(1)K of surgery on a fibered knot K ⊂ S 3 , etc. With b2+ = 1 and c12 > 0, we have symplectic manifolds CP2 , S 2 × S 2 , CP2 #n CP2 for n 8 and the Barlow surface. For b2+ = 1 and c12 0, Park [109] gave a criterion for a symplectic 4-manifold to be rational or ruled in terms of Seiberg–Witten theory.

4.4. Uniqueness of symplectic forms Besides the notions listed in Section 1.4, the following equivalence relation for symplectic manifolds is considered. As it allows the cleanest statements about uniqueness, this relation is simply called equivalence. D EFINITION 4.8. Symplectic manifolds (M, ω0 ) and (M, ω1 ) are equivalent if they are related by a combination of deformation-equivalences and symplectomorphisms. Recall that (M, ω0 ) and (M, ω1 ) are deformation-equivalent when there is a smooth family ωt of symplectic forms joining ω0 to ω1 (Section 1.4), and they are symplectomorphic when there is a diffeomorphism ϕ : M → M such that ϕ ∗ ω1 = ω0 (Section 1.2). Hence, equivalence is the relation generated by deformations and diffeomorphisms. The corresponding equivalence classes can be viewed as the connected components of the moduli space of symplectic forms up to diffeomorphism. This is a useful notion when focusing on topological properties. E XAMPLES . 1. The complex projective plane CP2 has a unique symplectic structure up to symplectomorphism and scaling. This was shown by Taubes [128] relating Seiberg–Witten invariants (Section 4.5) to pseudoholomorphic curves to prove the existence of a pseudoholomorphic sphere. Previous work of Gromov [64] and McDuff [96] showed that the existence of a pseudoholomorphic sphere implies that the symplectic form is standard. Lalonde and McDuff [85] concluded similar classifications for symplectic ruled surfaces and for symplectic rational surfaces (Section 4.3). The symplectic form on 43 A (rational) ruled surface is a complex (Kähler) surface that is the total space of a holomorphic fibration over a Riemann surface with fiber CP1 . When the base is also a sphere, these are the Hirzebruch surfaces P(L ⊕ C) where L is a holomorphic line bundle over CP1 .

Symplectic geometry

135

a symplectic ruled surface is unique up to symplectomorphism in its cohomology class, and is isotopic to a standard Kähler form. In particular, any symplectic form on S 2 × S 2 is symplectomorphic to aπ1∗ σ + bπ2∗ σ for some a, b > 0 where σ is the standard area form on S 2 . Li–Liu [88] showed that the symplectic structure on CP2 #n CP2 for 2 n 9 is unique up to equivalence. 2. McMullen and Taubes [101] first exhibited simply connected closed 4-manifolds admitting inequivalent symplectic structures. Their examples were constructed using 3-dimensional topology, and distinguished by analyzing the structure of Seiberg– Witten invariants to show that the first Chern classes (Section 3.2) of the two symplectic structures lie in disjoint orbits of the diffeomorphism group. In higher dimensions there were previously examples of manifolds with inequivalent symplectic forms; see, for instance, [111]. With symplectic techniques and avoiding gauge theory, Smith [117] showed that, for each n 2, there is a simply connected closed 4-manifold that admits at least n inequivalent symplectic forms, also distinguished via the first Chern classes. It is not yet known whether there exist inequivalent symplectic forms on a 4-manifold with the same first Chern class. 4.5. Invariants for 4-manifolds Very little was known about 4-dimensional manifolds until 1981, when Freedman [51] provided a complete classification of closed simply connected topological 4-manifolds, and shortly thereafter Donaldson [30] showed that the panorama for smooth 4-manifolds was much wilder.44 Freedman showed that, modulo homeomorphism, such topological manifolds are essentially classified by their intersection forms (for an even intersection form there is exactly one class, whereas for an odd intersection form there are exactly two classes distinguished by the Kirby–Siebenmann invariant KS, at most one of which admits smooth representatives—smoothness requires KS = 0). Donaldson showed that, whereas the existence of a smooth structure imposes strong constraints on the topological type of a manifold, for the same topological manifold there can be infinite different smooth structures.45 In other words, by far not all intersection forms can occur for smooth 4-manifolds and the same intersection form may correspond to nondiffeomorphic manifolds. Donaldson’s key tool was a set of gauge-theoretic invariants, defined by counting with signs the equivalence classes (modulo gauge equivalence) of connections on SU(2)- (or SO(3)-) bundles over M whose curvature has vanishing self-dual part. For a dozen years there was hard work on the invariants discovered by Donaldson but limited advancement on the understanding of smooth 4-manifolds. 44 It had been proved by Rokhlin in 1952 that if such a smooth manifold M has even intersection form Q (i.e., M w2 = 0), then the signature of QM must be a multiple of 16. It had been proved by Whitehead and Milnor that two such topological manifolds are homotopy equivalent if and only if they have the same intersection form. 45 It is known that in dimensions 3, each topological manifold has exactly one smooth structure, and in dimensions 5 each topological manifold has at most finitely many smooth structures. For instance, whereas each topological Rn , n = 4, admits a unique smooth structure, the topological R4 admits uncountably many smooth structures.

136

A. Cannas da Silva

E XAMPLES . Finding exotic46 smooth structures on closed simply connected manifolds with small b2 has long been an interesting problem, especially in view of the smooth Poincaré conjecture for 4-manifolds. The first exotic smooth structures on a rational surface CP2 #n CP2 were found in the late 80’s for n = 9 by Donaldson [31] and for n = 8 by Kotschick [84]. There was no progress until the recent work of Park [110] constructing a symplectic exotic CP2 #7 CP2 and using this to exhibit a third distinct smooth structure CP2 #8 CP2 , thus illustrating how the existence of symplectic forms is tied to the existence of different smooth structures. This stimulated research by Fintushel, Ozsváth, Park, Stern, Stipsicz and Szabó, which together shows that there are infinitely many exotic smooth structures on CP2 #n CP2 for n = 5, 6, 7, 8 (the case n = 9 had been shown in the late 80’s by Friedman–Morgan and by Okonek–Van de Ven). In 1994 Witten brought about a revolution in Donaldson theory by introducing a new set of invariants—the Seiberg–Witten invariants—which are much simpler to calculate and to apply. This new viewpoint was inspired by developments due to Seiberg and Witten in the understanding of N = 2 supersymmetric Yang–Mills. Let M be a smooth oriented closed 4-dimensional manifold with b2+ (M) > 1 (there is a version for b2+ (M) = 1). All such 4-manifolds M (with any b2+ (M)) admit a spin-c structure, i.e., a Spinc (4)-bundle over M with an isomorphism of the associated SO(4)-bundle to the bundle of oriented frames on the tangent bundle for some chosen Riemannian metric. Let CM = {a ∈ H 2 (M; Z) | a ≡ w2 (T M)(2)} be the set of characteristic elements, and let Spinc (M) be the set of spin-c structures on M. For simplicity, assume that M is simply connected (or at least that H1 (M; Z) has no 2-torsion), so that Spinc (M) is isomorphic to CM with isomorphism given by the first Chern class of the determinant line bundle (the determinant line bundle is the line bundle associated by a natural group homomorphism Spinc (4) → U(1)). Fix an orientation of a maximal-dimensional positive-definite subspace H+2 (M; R) ⊂ H 2 (M; R). The Seiberg–Witten invariant is the function SWM : CM −→ Z defined as follows. Given a spin-c structure α ∈ Spinc (M) CM , the image SWM (α) = [M] ∈ Hd (B ∗ ; Z) is the homology class of the moduli space M of solutions (called monopoles) of the Seiberg–Witten (SW) equations modulo gauge equivalence. The SW equations are nonlinear differential equations on a pair of a connection A on the determinant line bundle of α and of a section ϕ of an associated U(2)-bundle, called the positive (half) spinor bundle: FA+ = iq(ϕ)

and DA ϕ = 0,

where FA+ is the self-dual part of the (imaginary) curvature of A, q is a squaring operation taking sections of the positive spinor bundle to self-dual 2-forms, and DA is the corresponding Dirac operator. For a generic perturbation of the equations (replacing the first equation by FA+ = iq(ϕ) + iν, where ν is a self-dual 2-form) and of the Riemannian 46 A manifold homeomorphic but not diffeomorphic to a smooth manifold M is called an exotic M.

Symplectic geometry

137

metric, a transversality argument shows that the moduli space M is well-behaved and actually inside the space B ∗ of gauge-equivalence classes of irreducible pairs (those (A, ϕ) for which ϕ = 0), which is homotopy-equivalent to CP∞ and hence has even-degree homology groups Hd (B ∗ ; Z) Z. When the dimension d of M is odd or when M is empty, the invariant SWM (α) is set to be zero. The basic classes are the classes α ∈ CM for which SWM (α) = 0. The set of basic classes is always finite, and if α is a basic class then so is −α. The main results are that the Seiberg–Witten invariants are invariants of the diffeomorphism type of the 4-manifold M and satisfy vanishing and nonvanishing theorems, which allowed to answer an array of questions about specific manifolds. Taubes [128] discovered an equivalence between Seiberg–Witten and Gromov invariants (using pseudoholomorphic curves) for symplectic 4-manifolds, by proving the existence of pseudoholomorphic curves from solutions of the Seiberg–Witten equations and vice-versa. As a consequence, he proved: T HEOREM 4.9 (Taubes). Let (M, ω) be a compact symplectic 4-manifold. If b2+ > 1, then c1 (M, ω) admits a smooth pseudoholomorphic representative. If M = M1 # M2 , then one of the Mi ’s has negative definite intersection form. There are results also for b2+ = 1, and follow-ups describe the set of basic classes of a connected sum M # N in terms of the set of basic classes of M when N is a manifold with negative definite intersection form (starting with CP2 ). In an attempt to understand other 4-manifolds via Seiberg–Witten and Gromov invariants, some analysis of pseudoholomorphic curves has been extended to nonsymplectic 4-manifolds by equipping these with a nearly nondegenerate closed 2-form. In particular, Taubes [130] has related Seiberg–Witten invariants to pseudoholomorphic curves for compact oriented 4-manifolds with b2+ > 0. Any compact oriented 4-manifold M with b2+ > 0 admits a closed 2-form that vanishes along a union of circles and is symplectic elsewhere [54,75]. In fact, for a generic metric on M, there is a self-dual harmonic form ω which is transverse to zero as a section of Λ2 T ∗ M. The vanishing locus of ω is the union of a finite number of embedded circles, and ω is symplectic elsewhere. The generic behavior of closed 2-forms on orientable 4-manifolds is partially understood [3, pp. 23–24]. Here is a summary. Let ω be a generic closed 2-form on a 4-manifold M. At the points of some hypersurface Z, the form ω has rank 2. At a generic point of M, ω is nondegenerate; in particular, has the Darboux normal form dx1 ∧ dy1 + dx2 ∧ dy2 . There is a codimension-1 submanifold Z where ω has rank 2, and there are no points where ω vanishes. At a generic point of Z, the kernel of ω˜ is transverse to Z; the normal form near such a point is x1 dx1 ∧ dy1 + dx2 ∧ dy2 . There is a curve C where the kernel of ω˜ is not transverse to Z, hence sits in T Z. At a generic point of C, the kernel of ω˜ is transverse to C; there are two possible normal forms near such points, 2 3 called elliptic and hyperbolic, d(x − z2 ) ∧ dy + d(xz ± ty − z3 ) ∧ dt. The hyperbolic and elliptic sections of C are separated by parabolic points, where the kernel is tangent to C. It is known that there exists at least one continuous family of inequivalent degeneracies in a parabolic neighborhood [56].

138

A. Cannas da Silva

4.6. Lefschetz pencils Lefschetz pencils in symplectic geometry imitate linear systems in complex geometry. Whereas holomorphic functions on a projective surface must be constant, there are interesting functions on the complement of a finite set, and generic such functions have only quadratic singularities. A Lefschetz pencil can be viewed as a complex Morse function or as a very singular fibration, in the sense that, not only some fibers are singular (have ordinary double points) but all fibers go through some points. D EFINITION 4.10. A Lefschetz pencil on an oriented 4-manifold M is a map f : M \ {b1 , . . . , bn } → CP1 defined on the complement of a finite set in M, called the base locus, that is a submersion away from a finite set {p1 , . . . , pn+1 }, and obeying local models (z1 , z2 ) → z1 /z2 near the bj ’s and (z1 , z2 ) → z1 z2 near the pj ’s, where (z1 , z2 ) are oriented local complex coordinates. Usually it is also required that each fiber contains at most one singular point. By blowing up M at the bj ’s, we obtain a map to CP1 on the whole manifold, called a Lefschetz fibration. Lefschetz pencils and Lefschetz fibrations can be defined on higher-dimensional manifolds where the bj ’s are replaced by codimension 4 submanifolds. By working on the Lefschetz fibration, Gompf [59,58] proved that a structure of Lefschetz pencil (with a nontrivial base locus) gives rise to a symplectic form, canonical up to isotopy, such that the fibers are symplectic. Using asymptotically holomorphic techniques [12,32], Donaldson [34] proved that symplectic 4-manifolds admit Lefschetz pencils. More precisely: T HEOREM 4.11 (Donaldson). Let J be a compatible almost complex structure on a compact symplectic 4-manifold (M, ω) where the class [ω]/2π is integral. Then J can be deformed through almost complex structures to an almost complex structure J such that M admits a Lefschetz pencil with J -holomorphic fibers. The closure of a smooth fiber of the Lefschetz pencil is a symplectic submanifold Poincaré dual to k[ω]/2π ; cf. Theorem 1.13. Other perspectives on Lefschetz pencils have been explored, including in terms of representations of the free group π1 (CP1 \ {p1 , . . . , pn+1 }) in the mapping class group Γg of the generic fiber surface [118]. Similar techniques were used by Auroux [13] to realize symplectic 4-manifolds as branched covers of CP2 , and thus reduce the classification of symplectic 4-manifolds to a (hard) algebraic question about factorization in the braid group. Let M and N be compact oriented 4-manifolds, and let ν be a symplectic form on N . D EFINITION 4.12. A map f : M → N is a symplectic branched cover if for any p ∈ M there are complex charts centered at p and f (p) such that ν is positive on each complex line and where f is given by: a local diffeomorphism (x, y) → (x, y), or a simple branching (x, y) → (x 2 , y), or an ordinary cusp (x, y) → (x 3 − xy, y). T HEOREM 4.13 (Auroux). Let (M, ω) be a compact symplectic 4-manifold where the class [ω] is integral, and let k be a sufficiently large integer. Then there is a symplectic

Symplectic geometry

139

branched cover fk : (M, kω) → CP2 , that is canonical up to isotopy for k large enough. Conversely, given a symplectic branched cover f : M → N , the domain M inherits a symplectic form canonical up to isotopy in the class f ∗ [ν].

5. Hamiltonian geometry 5.1. Symplectic and Hamiltonian vector fields Let (M, ω) be a symplectic manifold and let H : M → R be a smooth function. By nondegeneracy, there is a unique vector field XH on M such that ıXH ω = dH . Supposing that XH is complete (this is always the case when M is compact), let ρt : M → M, t ∈ R, be its flow (cf. Section 1.3). Each diffeomorphism ρt preserves ω, i.e., ρt∗ ω = ω, because d ∗ ∗ ∗ dt ρt ω = ρt LXH ω = ρt (dıXH ω + ıXH dω) = 0. Therefore, every function on (M, ω) produces a family of symplectomorphisms. Notice how this feature involves both the nondegeneracy and the closedness of ω. D EFINITION 5.1. A vector field XH such that ıXH ω = dH for some H ∈ C ∞ (M) is a Hamiltonian vector field with Hamiltonian function H . Hamiltonian vector fields preserve their Hamiltonian functions (LXH H = ıXH dH = field XH ıXH ıXH ω = 0), so each integral curve {ρt (x) | t ∈ R} of a Hamiltonian vector must be contained in a level set of the Hamiltonian function H . In (R2n , ω0 = dxj ∧ ∂H ∂ ∂H ∂ dyj ), the symplectic gradient XH = ( ∂y − ∂x ) and the usual (Euclidean) gradij ∂xj j ∂yj ∂H ∂ ∂H ∂ ent ∇H = j ( ∂xj ∂xj + ∂yj ∂yj ) of a function H are related by J XH = ∇H , where J is the standard almost complex structure. E XAMPLES . 1. For the height function H (θ, h) = h on the sphere (M, ω) = (S 2 , dθ ∧ dh), from ∂ . Thus, ρt (θ, h) = (θ + t, h), which is rotation ıXH (dθ ∧ dh) = dh we get XH = ∂θ about the vertical axis, preserving the height H . 2. Let X be any vector field on a manifold W . There is a unique vector field X! on the cotangent bundle T ∗ W whose flow is the lift of the flow of X. Let α be the tautological form and ω = −dα the canonical symplectic form on T ∗ W . The vector field X! is Hamiltonian with Hamiltonian function H := ıX! α. 3. Consider Euclidean space R2n with coordinates (q1 , . . . , qn , p1 , . . . , pn ) and ω0 = dqj ∧ dpj . The curve ρt = (q(t), p(t)) is an integral curve for a Hamiltonian vector field XH exactly when it satisfies the Hamilton equations: dq

∂H i dt (t) = ∂pi , dpi ∂H dt (t) = − ∂qi .

140

A. Cannas da Silva

4. Newton’s second law states that a particle of mass m moving in configuration space R3 with coordinates q = (q1 , q2 , q3 ) under a potential V (q) moves along a curve q(t) such that m

d 2q = −∇V (q). dt 2

i Introduce the momenta pi = m dq dt for i = 1, 2, 3, and energy function H (q, p) = 1 2 47 R6 = T ∗ R3 with coordinates (q , q , q , p , 1 2 3 1 2m |p| + V (q) on the phase space p2 , p3 ). The energy H is conserved by the motion and Newton’s second law in R3 is then equivalent to the Hamilton equations in R6 :

dqi dt dpi dt

=

1 ∂H m pi = ∂pi , 2 ∂V = m ddtq2i = − ∂q i

= − ∂H ∂qi .

D EFINITION 5.2. A vector field X on M preserving ω (i.e., such that LX ω = 0) is a symplectic vector field. Hence, a vector field X on (M, ω) is called symplectic when ıX ω is closed, and Hamiltonian when ıX ω is exact. In the latter case, a primitive H of ıX ω is called a Hamiltonian function of X. On a contractible open set every symplectic vector field is 1 (M) measures the obstruction for symplectic Hamiltonian. Globally, the group HdeRham vector fields to be Hamiltonian. For instance, the vector field X1 = ∂θ∂ 1 on the 2-torus (M, ω) = (T2 , dθ1 ∧ dθ2 ) is symplectic but not Hamiltonian. A vector field X is a differential operator on functions: X · f := LX f = df (X) for f ∈ C ∞ (M). As such, the bracket W = [X, Y ] is the commutator: LW = [LX , LY ] = LX LY − LY LX (cf. Section 3.3). This endows the set χ(M) of vector fields on a manifold M with a structure of Lie algebra.48 For a symplectic manifold (M, ω), using ı[X,Y ] = [LX , ıY ] and Cartan’s magic formula, we find that ı[X,Y ] ω = dıX ıY ω + ıX dıY ω − ıY dıX ω − ıY ıX dω = d(ω(Y, X)). Therefore: P ROPOSITION 5.3. If X and Y are symplectic vector fields on a symplectic manifold (M, ω), then [X, Y ] is Hamiltonian with Hamiltonian function ω(Y, X). Hence, Hamiltonian vector fields and symplectic vector fields form Lie subalgebras for the Lie bracket [·, ·]. D EFINITION 5.4. The Poisson bracket of two functions f, g ∈ C ∞ (M) is the function {f, g} := ω(Xf , Xg ) = LXg f . 47 The phase space of a system of n particles is the space parametrizing the position and momenta of the particles. The mathematical model for a phase space is a symplectic manifold. 48 A (real) Lie algebra is a (real) vector space g together with a Lie bracket [·, ·], i.e., a bilinear map [·, ·] : g × g → g satisfying antisymmetry, [x, y] = −[y, x], ∀x, y ∈ g, and the Jacobi identity, [x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0, ∀x, y, z ∈ g.

Symplectic geometry

141

By Proposition 5.3 we have X{f,g} = −[Xf , Xg ]. Moreover, the bracket {·, ·} satisfies the Jacobi identity, {f, {g, h}} + {g, {h, f }} + {h, {f, g}} = 0, and the Leibniz rule, {f, gh} = {f, g}h + g{f, h}. D EFINITION 5.5. A Poisson algebra (P, {·, ·}) is a commutative associative algebra P with a Lie bracket {·, ·} satisfying the Leibniz rule. When (M, ω) is a symplectic manifold, (C ∞ (M), {·, ·}) is a Poisson algebra, and the map C ∞ (M) → χ(M), H → XH is a Lie algebra anti-homomorphism. E XAMPLES . 1. For the prototype (R2n , dxi ∧ dyi ), we have Xxi = − ∂y∂ i and Xyi = ∂x∂ i , so that {xi , xj } = {yi , yj } = 0 and {xi , yj } = δij for all i, j . Arbitrary functions f, g ∈ C ∞ (R2n ) have the classical Poisson bracket n ∂f ∂g ∂f ∂g . − {f, g} = ∂xi ∂yi ∂yi ∂xi i=1

2. Let G be a Lie group,49 g its Lie algebra and g∗ the dual vector space of g. The vector field g X # generated by X ∈ g for the adjoint action50 of G on g has value [X, Y ] at Y ∈ g. The vector field X # generated by X ∈ g for the coadjoint action of G on g∗ is Xξ# , Y = ξ, [Y, X], ∀ξ ∈ g∗ , Y ∈ g. The skew-symmetric pairing ω on g defined at ξ ∈ g∗ by ωξ (X, Y ) := ξ, [X, Y ] has kernel at ξ the Lie algebra gξ of the stabilizer of ξ for the coadjoint action. Therefore, ω restricts to a nondegenerate 2-form on the tangent spaces to the orbits of the coadjoint action. As the tangent spaces to an orbit are generated by the vector fields X # , the Jacobi identity in g implies that this form is closed. It is called the canonical symplectic form (or the Lie–Poisson or Kirillov–Kostant–Souriau symplectic structure) on the coadjoint orbits. The corresponding Poisson structure on g∗ is the canonical one induced by the Lie bracket: {f, g}(ξ ) = ξ, [dfξ , dgξ ] 49 A Lie group is a manifold G equipped with a group structure where the group operation G × G → G and inversion G → G are smooth maps. An action of a Lie group G on a manifold M is a group homomorphism G → Diff(M), g → ψg , where the evaluation map M × G → M, (p, g) → ψg (p) is a smooth map. The orbit of G through p ∈ M is {ψg (p) | g ∈ G}. The stabilizer (or isotropy) of p ∈ M is Gp := {g ∈ G | ψg (p) = p}. 50 Any Lie group G acts on itself by conjugation: g ∈ G → ψ ∈ Diff(G), ψ (a) = g · a · g −1 . Let Ad : g → g g g g be the derivative at the identity of ψg : G → G. We identify the Lie algebra g with the tangent space Te G. For matrix groups, Adg X = gXg −1 . Letting g vary, we obtain the adjoint action of G on its Lie algebra Ad : G → GL(g). Let ·, · : g∗ × g → R be the natural pairing ξ, X = ξ(X). Given ξ ∈ g∗ , we define Ad∗g ξ by Ad∗g ξ, X = ξ, Adg −1 X, for any X ∈ g. The collection of maps Ad∗g forms the coadjoint action of G on the dual of its Lie algebra Ad∗ : G → GL(g∗ ). These satisfy Adg ◦ Adh = Adgh and Ad∗g ◦ Ad∗h = Ad∗gh .

142

A. Cannas da Silva

for f, g ∈ C ∞ (g∗ ) and ξ ∈ g∗ . The differential dfξ : Tξ g∗ g∗ → R is identified with an element of g g∗∗ .

5.2. Arnold conjecture and Floer homology There is an important generalization of Poincaré’s last geometric theorem (Theorem 2.16) conjectured by Arnold starting around 1966. Let (M, ω) be a compact symplectic manifold, and ht : M → R a 1-periodic (i.e., ht = ht+1 ) smooth family of functions. Let ρ : M × R → M be the isotopy generated by the time-dependent Hamiltonian vector field vt defined by the equation ω(vt , ·) = dht . The symplectomorphism ϕ = ρ1 is then said to be exactly homotopic to the identity. In other words, a symplectomorphism exactly homotopic to the identity is the time-1 map of the isotopy generated by some time-dependent 1-periodic Hamiltonian function. There is a one-to-one correspondence between the fixed points of ϕ and the period-1 orbits of ρ. When all the fixed points of such ϕ are nondegenerate (generic case), we call ϕ nondegenerate. The Arnold conjecture [2, Appendix 9] predicted that #{fixed points of a nondegenerate ϕ}

2n

dim H i (M; R)

i=0

(or even that the number of fixed points of a nondegenerate ϕ is at least the minimal number of critical points of a Morse function51 ). When the Hamiltonian h : M → R is independent of t, this relation is trivial: a point p is critical for h if and only if dhp = 0, if and only if vp = 0, if and only if ρ(t, p) = p, ∀t ∈ R, which implies that p is a fixed point of ρ1 = ϕ, so the Arnold conjecture reduces to a Morse inequality. Notice that, according to the Lefschetz fixed point theorem, the Euler characteristic of M, i.e., the alternating sum of the Betti numbers, (−1)i dim H i (M; R), is a (weaker) lower bound for the number of fixed points of ϕ. The Arnold conjecture was gradually proved from the late 70’s to the late 90’s by Eliashberg [39], Conley–Zehnder [24], Floer [49], Sikorav [116], Weinstein [140], Hofer– Salamon [74], Ono [108], culminating with independent proofs by Fukaya–Ono [52] and Liu–Tian [90]. There are open conjectures for sharper bounds on the number of fixed points. The breakthrough tool for establishing the Arnold conjecture was Floer homology—an ∞-dimensional analogue of Morse theory. Floer homology was defined by Floer [46–50] and developed through the work of numerous people after Floer’s death. It combines the variational approach of Conley and Zehnder [25], with Witten’s Morse– Smale complex [144], and with Gromov’s compactness theorem for pseudoholomorphic curves [64]. Floer theory starts from a symplectic action functional on the space of loops LM of a symplectic manifold (M, ω) whose zeros of the differential dF : T (LM) → R are the period-1 orbits of the isotopy ρ above. The tangent bundle T (LM) is the space of loops with vector fields over them: pairs (, v), where : S 1 → M and v : S 1 → ∗ (T M) is a 51 A Morse function is a smooth function f : M → R all of whose critical points are nondegenerate, i.e., at any critical point the Hessian matrix is nondegenerate.

Symplectic geometry

143

1 ˙ − Xht ((t), v(t)) dt. The Floer complex52 is the chain section. Then df (, v) = 0 ω((t) complex freely generated by the critical points of F (corresponding to the fixed points of ϕ), with relative grading index(x, y) given by the difference in the number of positive eigenvalues from the spectral flow. The Floer differential is given by counting the number n(x, y) of pseudoholomorphic surfaces (the gradient flow lines joining two fixed points): C∗ =

Zx and ∂x =

x∈Crit(F )

n(x, y)y.

y∈Crit(F ) index(x,y)=1

Pondering transversality, compactness and orientation, Floer’s theorem states that the homology of (C∗ , ∂) is isomorphic to the ordinary homology of M. In particular, the sum of the Betti numbers is a lower bound for the number of fixed points of ϕ. From the above symplectic Floer homology, Floer theory has branched out to tackle other differential geometric problems in symplectic geometry and 3- and 4-dimensional topology. It provides a rigorous definition of invariants viewed as homology groups of infinite-dimensional Morse-type theories, with relations to gauge theory and quantum field theory. There is Lagrangian Floer homology (for the case of Lagrangian intersections, i.e., intersection of a Lagrangian submanifold with a Hamiltonian deformation of itself), instanton Floer homology (for invariants of 3-manifolds), Seiberg–Witten Floer homology, Heegaard Floer homology and knot Floer homology. For more on Floer homology; see, for instance, [35,113].

5.3. Euler–Lagrange equations The equations of motion in classical mechanics arise from variational principles. The physical path of a general mechanical system of n particles is the path that minimizes a quantity called the action. When dealing with systems with constraints, such as the simple 52 The Morse complex for a Morse function on a compact manifold, f : M → R, is the chain complex freely

generated by the critical points of f , graded by the Morse index ı and with differential given by counting the number n(x, y) of flow lines of the negative gradient −∇f (for a metric on X) from the point x to the point y whose indices differ by 1: C∗ = Zx and ∂x = n(x, y)y. x∈Crit(f )

y∈Crit(f ) ı(y)=ı(x)−1

The coefficient n(x, y) is thus the number of solutions (modulo R-reparametrization) u : R → X of the ordinary d u(t) = −∇f (u(t)) with conditions lim differential equation dt t→−∞ u(t) = x, limt→+∞ u(t) = y. The Morse index of a critical point of f is the dimension of its unstable manifold, i.e., the number of negative eigenvalues of the Hessian of f at that point. For a generic metric, the unstable manifold of a critical point W u (x) intersects transversally with the stable manifold of another critical point W s (y). When ı(x) − ı(y) = 1, the intersection W u (x) ∩ W s (y) has dimension 1, so when we quotient out by the R-reparametrization (to count actual image curves) we get a discrete set, which is finite by compactness. That (C∗ , ∂) is indeed a complex, i.e., ∂ 2 = 0, follows from counting broken flow lines between points whose indices differ by 2. Morse’s theorem states that the homology of the Morse complex coincides with the ordinary homology of M. In particular, the sum of all the Betti numbers dim H i (M; R) is a lower bound for the number of critical points of a Morse function.

144

A. Cannas da Silva

pendulum, or two point masses attached by a rigid rod, or a rigid body, the language of variational principles becomes more appropriate than the explicit analogues of Newton’s second laws. Variational principles are due mostly to D’Alembert, Maupertius, Euler and Lagrange. Let M be an n-dimensional manifold, and let F : T M → R be a function on its tangent bundle. If γ : [a, b] → M is a curve on M, the lift of γ to T M is the curve on T M given by γ˜ : [a, b] → T M, t → (γ (t), dγ dt (t)). The action of γ is

b

Aγ :=

∗

(γ˜ F )(t) dt =

a

a

b

dγ F γ (t), (t) dt. dt

For fixed p, q, let P(a, b, p, q) = {γ : [a, b] → M smooth | γ (a) = p, γ (b) = q}. The goal is to find, among all γ ∈ P(a, b, p, q), the curve that locally minimizes Aγ . (Minimizing curves are always locally minimizing.) Assume that p, q and the image of γ lie in a coordinate neighborhood (U, x1 , . . . , xn ). On T U we have coordinates (x1 , . . . , xn , v1 , . . . , vn ) associated with a trivialization of T U by ∂x∂ 1 , . . . , ∂x∂ n . Using this trivialization, a curve γ : [a, b] → U , γ (t) = (γ1 (t), . . . , γn (t)) lifts to γ˜ : [a, b] −→ T U,

dγ1 dγn γ˜ (t) = γ1 (t), . . . , γn (t), (t), . . . , (t) . dt dt

Consider infinitesimal variations of γ . Let c1 , . . . , cn ∈ C ∞ ([a, b]) be such that ck (a) = ck (b) = 0. For ε small, let γε : [a, b] → U be the curve γε (t) = (γ1 (t) + εc1 (t), . . . , γn (t) + εcn (t)). Let Aε := Aγε . A necessary condition for γ = γ0 ∈ P(a, b, p, q) to minimize the action is that ε = 0 be a critical point of Aε . By the Leibniz rule and integration by parts, we have that dAε (0) = dε

a

= a

b ∂F k

dγ0 ∂F dγ0 dck γ0 (t), γ0 , (t) ck (t) + (t) dt ∂xk dt ∂vk dt dt

b ∂F k

d ∂F (. . .) − (. . .) ck (t) dt. ∂xk dt ∂vk

Aε (0) to vanish for all ck ’s satisfying boundary conditions ck (a) = ck (b) = 0, the For ddε path γ0 must satisfy the Euler–Lagrange equations:

d ∂F ∂F dγ0 dγ0 γ0 (t), γ0 (t), (t) = (t) , ∂xk dt dt ∂vk dt

k = 1, . . . , n.

E XAMPLES . 1. Let (M, g) be a Riemannian manifold. Let F : T M → R be the function whose restriction to each tangent space is the quadratic form defined by the Riemannian

Symplectic geometry

145

metric. On a coordinate chart F (x, v) = |v|2 = gij (x)v i v j . Let p, q ∈ M and γ : [a, b] → M a curve joining p to q. The action of γ is b 2

dγ

dt. Aγ =

a dt The Euler–Lagrange equations become the Christoffel equations for a geodesic dγ i dγ j d 2γ k k = 0, + Γ ◦ γ ij dt dt dt 2 where the Christoffel symbols Γijk ’s are defined in terms of the coefficients of the Riemannian metric (g ij is the matrix inverse to gij ) by Γijk

∂gj ∂gij 1 k ∂gi . = g + − 2 ∂xj ∂xi ∂x

2. Consider a point-particle of mass m moving in R3 under a force field G. The work

b of G on a path γ : [a, b] → R3 is Wγ := a G(γ (t)) · dγ dt (t) dt. Suppose that G is conservative, i.e., Wγ depends only on the initial and final points, p = γ (a) and q = γ (b). We can define the potential energy as V : R3 → R, V (q) := Wγ , where γ is a path joining a fixed base point p0 ∈ R3 to q. Let P be the set of all paths going from p to q over time t ∈ [a, b]. By the principle of least action, the physical path is the path γ ∈ P that minimizes a kind of mean value of kinetic minus potential energy, known as the action: Aγ :=

b a

m

dγ

2 (t) − V γ (t) dt. 2 dt

The Euler–Lagrange equations are then equivalent to Newton’s second law: m

d 2x ∂V x(t) = 0 (t) − 2 ∂x dt

⇐⇒

m

d 2x (t) = G x(t) . 2 dt

In the case of the earth moving about the sun, both regarded as point-masses and assuming that the sun to be stationary at the origin, the gravitational potentialV (x) = const |x| yields the inverse square law for the motion. 3. Consider now n point-particles of masses m1 , . . . , mn moving in R3 under a conservative force corresponding to a potential energy V ∈ C ∞ (R3n ). At any instant t, the configuration of this system is described by a vector x = (x1 , . . . , xn ) in configuration space R3n , where xk ∈ R3 is the position of the kth particle. For fixed p, q ∈ R3n , let

146

A. Cannas da Silva

P be the set of all paths γ = (γ1 , . . . , γn ) : [a, b] → R3n from p to q. The action of a path γ ∈ P is

b n mk

dγk

2 (t) − V γ (t) dt. Aγ := 2 dt a

k=1

The Euler–Lagrange equations reduce to Newton’s law for each particle. Suppose that the particles are restricted to move on a submanifold M of R3n called the constraint set. By the principle of least action for a constrained system, the physical path has minimal action among all paths satisfying the rigid constraints. I.e., we single out the actual physical path as the one that minimizes Aγ among all γ : [a, b] → M with γ (a) = p and γ (b) = q. In the case where F (x, v) does not depend on v, the Euler–Lagrange equations are sim∂F 0 (γ0 (t), dγ ply ∂x dt (t)) = 0. These are satisfied if and only if the curve γ0 sits on the critical i set of F . For generic F , the critical points are isolated, hence γ0 (t) mustbe a constant curve. In the case where F (x, v) depends affinely on v, F (x, v) = F0 (x) + nj=1 Fj (x)vj , the Euler–Lagrange equations become n dγj ∂Fj ∂Fi ∂F0 γ (t) (t). γ (t) = − ∂xi ∂xj ∂xi dt j =1

i If the n × n matrix ( ∂F ∂xj −

∂Fj ∂xi

) has an inverse Gij (x), we obtain the system of first order dγ 0 ordinary differential equations dtj (t) = Gj i (γ (t)) ∂F ∂xi (γ (t)). Locally it has a unique solution through each point p. If q is not on this curve, there is no solution at all to the Euler–Lagrange equations belonging to P(a, b, p, q). Therefore, we need nonlinear dependence of F on the v variables in order to have appropriate solutions. From now on, assume the Legendre condition:

∂ 2F det ∂vi ∂vj

= 0. 2

F Letting Gij (x, v) = ( ∂v∂i ∂v (x, v))−1 , the Euler–Lagrange equations become j

d 2 γj dγ dγ dγk ∂F ∂ 2F γ , γ , − . = G G ji ji ∂xi dt ∂vi ∂xk dt dt dt 2 i

i,k

This second order ordinary differential equation has a unique solution given initial con∂2F ditions γ (a) = p and dγ dt (a) = v. Assume that ( ∂vi ∂vj (x, v)) ' 0, ∀(x, v), i.e., with the x variable frozen, the function v → F (x, v) is strictly convex. Then the path γ0 ∈ P(a, b, p, q) satisfying the above Euler–Lagrange equations does indeed locally minimize Aγ (globally it is only critical):

Symplectic geometry

147

P ROPOSITION 5.6. For every sufficiently small subinterval [a1 , b1 ] of [a, b], γ0 |[a1 ,b1 ] is locally minimizing in P(a1 , b1 , p1 , q1 ) where p1 = γ0 (a1 ), q1 = γ0 (b1 ). P ROOF. Take c = (c1 , . . . , cn ) with ci ∈ C ∞ ([a, b]), ci (a) = ci (b) = 0. Let γε = γ0 + εc ∈ P(a, b, p, q), and let Aε = Aγε . Suppose that γ0 : [a, b] → U satisfies the Euler–Lagrange Aε equations, i.e., ddε (0) = 0. Then d 2 Aε (0) = dε 2

b

a

i,j

+2

∂ 2F dγ0 γ0 , ci cj dt ∂xi ∂xj dt

+ a

dcj ∂ 2F dγ0 γ0 , ci dt ∂xi ∂vj dt dt

(B)

∂ 2F dγ0 dci dcj γ0 , dt. ∂vi ∂vj dt dt dt

(C)

b

a

i,j

b i,j

(A)

2

F (x, v)) ' 0 at all x, v, we have Since ( ∂v∂i ∂v j

(A) KA |c|2 2 , L [a,b]

dc

L2 [a,b]

dt L2 [a,b]

(B) KB |c|

and

2

dc

(C) KC

, dt L2 [a,b] where KA , KB , KC are positive constants. By the Wirtinger inequality53 , if b − a is very small, then (C) > |(A)| + |(B)| when c ≡ 0. Hence, γ0 is a local minimum. In Section 5.1 we saw that solving Newton’s second law in configuration space R3 is equivalent to solving in phase space for the integral curve in T ∗ R3 = R6 of the Hamiltonian vector field with Hamiltonian function H . In the next subsection we will see how this correspondence extends to more general Euler–Lagrange equations. 5.4. Legendre transform The Legendre transform gives the relation between the variational (Euler–Lagrange) and the symplectic (Hamilton–Jacobi) formulations of the equations of motion. 53 The Wirtinger inequality states that, for f ∈ C 1 ([a, b]) with f (a) = f (b) = 0, we have

b

df

a dt

2 b 2

dt π |f |2 dt.

(b − a)2 a

This can be proved with Fourier series.

148

A. Cannas da Silva

Let V be an n-dimensional vector space, with e1 , . . . , en a basis of V and v1 , . . . , vn the associated coordinates. Let F : V → R, F = F (v1 , . . . , vn ), be a smooth function. The function F is strictly convex if and only if for every pair of elements p, v ∈ V , v = 0, the restriction of F to the line {p + xv | x ∈ R} is strictly convex.54 It follows from the case of real functions on R that, for a strictly convex function F on V , the following are equivalent:55 (a) F has a critical point, i.e., a point where dFp = 0; (b) F has a local minimum at some point; (c) F has a unique critical point (global minimum); and (d) F is proper, that is, F (p) → +∞ as p → ∞ in V . A strictly convex function F is stable when it satisfies conditions (a)–(d) above. D EFINITION 5.7. The Legendre transform associated to F ∈ C ∞ (V ) is the map LF : V −→ V ∗ , p −→ dFp ∈ Tp∗ V V ∗ , where Tp∗ V V ∗ is the canonical identification for a vector space V . From now on, assume that F is a strictly convex function on V . Then, for every point p ∈ V , LF maps a neighborhood of p diffeomorphically onto a neighborhood of LF (p). Given ∈ V ∗ , let F : V −→ R,

F (v) = F (v) − (v).

Since (d 2 F )p = (d 2 F )p , F is strictly convex if and only if F is strictly convex. The stability set of F is SF = { ∈ V ∗ | F is stable}. The set SF is open and convex, and LF maps V diffeomorphically onto SF . (A way to ensure that SF = V ∗ and hence that LF maps V diffeomorphically onto V ∗ , is to assume that a strictly convex function F has quadratic growth at infinity, i.e., there exists a positive-definite quadratic form Q on V and a constant K such that F (p) Q(p) − K, for all p.) The inverse to LF is the map L−1 F : SF → V described as follows: for ∈ SF , 54 2 nA function F : V → R is strictly convex if at every p ∈ V the Hessian d Fp is positive definite. Let u = u e ∈ V . The Hessian of F at p is the quadratic function on V , i i i=1

∂2F

d2 (p)ui uj = 2 F (p + tu)

. (d 2 F )p (u) := ∂vi ∂vj dt t=0 i,j

55 A smooth function f : R → R is strictly convex if f (x) > 0 for all x ∈ R. Assuming that f is strictly convex, the following four conditions are equivalent: f (x) = 0 at some point, f has a local minimum, f has a unique

(global) minimum, and f (x) → +∞ as x → ±∞. The function f is stable if it satisfies one (and hence all) of these conditions. For instance, ex + ax is strictly convex for any a ∈ R, but it is stable only for a < 0. The function x 2 + ax is strictly convex and stable for any a ∈ R.

Symplectic geometry

149

the value L−1 F () is the unique minimum point p ∈ V of F . Indeed p is the minimum of F (v) − dFp (v). D EFINITION 5.8. The dual function F ∗ to F is F ∗ : SF −→ R,

F ∗ () = − min F (p). p∈V

The dual function F ∗ is smooth and, for all p ∈ V and all ∈ SF , satisfies the Young inequality F (p) + F ∗ () (p). On one hand we have V × V ∗ T ∗ V , and on the other hand, since V = V ∗∗ , we have V × V ∗ V ∗ × V T ∗ V ∗ . Let α1 be the tautological 1-form on T ∗ V and α2 be the tautological 1-form on T ∗ V ∗ . Via the identifications above, we can think of both of these forms as living on V × V ∗ . Since α1 = dβ − α2 , where β : V × V ∗ → R is the function β(p, ) = (p), we conclude that the forms ω1 = −dα1 and ω2 = −dα2 satisfy ω1 = −ω2 . ∗ T HEOREM 5.9. For a strictly convex function F we have that L−1 F = LF .

P ROOF. The graph ΛF of the Legendre transform LF is a Lagrangian submanifold of V × V ∗ with respect to the symplectic form ω1 . Hence, ΛF is also Lagrangian for ω2 . Let pr1 : ΛF → V and pr2 : ΛF → V ∗ be the restrictions of the projection maps V × V ∗ → V and V × V ∗ → V ∗ , and let i : ΛF → V × V ∗ be the inclusion map. Then i ∗ α1 = d(pr1 )∗ F as both sides have value dFp at (p, dFp ) ∈ ΛF . It follows that i ∗ α2 = d(i ∗ β − (pr1 )∗ F ) = d(pr2 )∗ F ∗ , which shows that ΛF is the graph of the inverse of LF ∗ . From this we conclude that the inverse of the Legendre transform associated with F is the Legendre transform associated with F ∗ . Let M be a manifold and F

: T M → R. We return to the Euler–Lagrange equations for minimizing the action Aγ = γ˜ ∗ F . At p ∈ M, let Fp := F |Tp M : Tp M → R. Assume that Fp is strictly convex for all p ∈ M. To simplify notation, assume also that SFp = Tp∗ M. The

Legendre transform on each tangent space LFp : Tp M −→ Tp∗ M is essentially given by the first derivatives of F in the v directions. Collect these and the dual functions Fp∗ : Tp∗ M → R into maps L : T M −→ T ∗ M,

L|Tp M = LFp

and H : T ∗ M −→ R,

H |Tp∗ M = Fp∗ .

The maps H and L are smooth, and L is a diffeomorphism. T HEOREM 5.10. Let γ : [a, b] → M be a curve, and γ˜ : [a, b] → T M its lift. Then γ satisfies the Euler–Lagrange equations on every coordinate chart if and only if L ◦ γ˜ : [a, b] → T ∗ M is an integral curve of the Hamiltonian vector field XH .

150

A. Cannas da Silva

P ROOF. Let (U, x1 , . . . , xn ) be a coordinate chart in M, with associated tangent (T U, x1 , . . . , xn , v1 , . . . , vn ) and cotangent (T ∗ U, x1 , . . . , xn , ξ1 , . . . , ξn ) coordinates. On T U we have F = F (x, v), on T ∗ U we have H = H (x, ξ ), and L : T U −→ T ∗ U,

H : T ∗ U −→ R, (x, ξ ) −→ Fx∗ (ξ ) = ξ · v − F (x, v),

(x, v) −→ (x, ξ ),

where ξ := LFx (v) = ∂F ∂v (x, v) is called the momentum. Integral curves (x(t), ξ(t)) of XH satisfy the Hamilton equations: dx (H)

dt dξ dt

=

∂H ∂ξ (x, ξ ), = − ∂H ∂x (x, ξ ),

whereas the physical path x(t) satisfies the Euler–Lagrange equations: (E–L)

dx d ∂F dx ∂F x, = x, . ∂x dt dt ∂v dt

Let (x(t), ξ(t)) = L(x(t), dx dt (t)). For an arbitrary curve x(t), we want to prove that t → (x(t), ξ(t)) satisfies (H) if and only if t → (x(t), dx dt (t)) satisfies (E–L). The first line of (H) comes automatically from the definition of ξ : ξ = LF x

dx dt

⇐⇒

dx ∂H ∗ = L−1 (x, ξ ). Fx (ξ ) = LFx (ξ ) = dt ∂ξ

If (x, ξ ) = L(x, v), by differentiating both sides of H (x, ξ ) = ξ · v − F (x, v) with respect to x, where ξ = LFx (v) = ξ(x, v) and v = ∂H ∂ξ , we obtain ∂H ∂H ∂ξ ∂ξ ∂F + = ·v− ∂x ∂ξ ∂x ∂x ∂x

⇐⇒

∂F ∂H (x, v) = − (x, ξ ). ∂x ∂x

Using the last equation and the definition of ξ , the second line of (H) becomes (E–L): dξ ∂H =− (x, ξ ) dt ∂x

⇐⇒

d ∂F ∂F (x, v) = (x, v). dt ∂v ∂x

5.5. Integrable systems D EFINITION 5.11. A Hamiltonian system is a triple (M, ω, H ), where (M, ω) is a symplectic manifold and H ∈ C ∞ (M) is the Hamiltonian function. P ROPOSITION 5.12. For a function f on a symplectic manifold (M, ω) we have that {f, H } = 0 if and only if f is constant along integral curves of XH .

Symplectic geometry

151

P ROOF. Let ρt be the flow of XH . Then d (f ◦ ρt ) = ρt∗ LXH f = ρt∗ ıXH df = ρt∗ ıXH ıXf ω = ρt∗ ω(Xf , XH ) dt = ρt∗ {f, H }.

A function f as in Proposition 5.12 is called an integral of motion (or a first integral or a constant of motion). In general, Hamiltonian systems do not admit integrals of motion that are independent of the Hamiltonian function. Functions f1 , . . . , fn are said to be independent if their differentials (df1 )p , . . . , (dfn )p are linearly independent at all points p in some dense subset of M. Loosely speaking, a Hamiltonian system is (completely) integrable if it has as many commuting integrals of motion as possible. Commutativity is with respect to the Poisson bracket. If f1 , . . . , fn are commuting integrals of motion for a Hamiltonian system (M, ω, H ), then ω(Xfi , Xfj ) = {fi , fj } = 0, so at each p ∈ M the Hamiltonian vector fields generate an isotropic subspace of Tp M. When f1 , . . . , fn are independent, by symplectic linear algebra n can be at most half the dimension of M. D EFINITION 5.13. A Hamiltonian system (M, ω, H ) where M is a 2n-dimensional manifold is (completely) integrable if it possesses n independent commuting integrals of motion, f1 = H, f2 , . . . , fn . Any 2-dimensional Hamiltonian system (where the set of nonfixed points is dense) is trivially integrable. Basic examples are the simple pendulum and the harmonic oscillator. A Hamiltonian system (M, ω, H ) where M is 4-dimensional is integrable if there is an integral of motion independent of H (the commutativity condition is automatically satisfied). A basic example is the spherical pendulum. Sophisticated examples of integrable systems can be found in [8,72]. E XAMPLES . 1. The simple pendulum is a mechanical system consisting of a massless rigid rod of length , fixed at one end, whereas the other end has a bob of mass m, which may oscillate in the vertical plane. We assume that the force of gravity is constant pointing vertically downwards and the only external force acting on this system. Let θ be the oriented angle between the rod and the vertical direction. Let ξ be the coordinate along the fibers of T ∗ S 1 induced by the standard angle coordinate on S 1 . ξ2 The energy function H : T ∗ S 1 → R, H (θ, ξ ) = 2m 2 + m(1 − cos θ ), is an appropriate Hamiltonian function to describe the simple pendulum. Gravity is responsible for the potential energy V (θ ) = m(1 − cos θ ), and the kinetic energy is given by 1 2 K(θ, ξ ) = 2m 2ξ . 2. The spherical pendulum consists of a massless rigid rod of length , fixed at one end, whereas the other end has a bob of mass m, which may oscillate freely in all directions. For simplicity let m = = 1. Again assume that gravity is the only external force. Let ϕ, θ (0 < ϕ < π , 0 < θ < 2π ) be spherical coordinates for the bob, inducing coordinates η, ξ along the fibers of T ∗ S 2 . An appropriate Hamiltonian function for this system is the energy function H : T ∗ S 2 → R, H (ϕ, θ, η, ξ ) =

152

A. Cannas da Silva 2

+ (sinξ ϕ)2 ) + cos ϕ. The function J (ϕ, θ, η, ξ ) = ξ is an independent integral of motion corresponding to the group of symmetries given by rotations about the vertical axis (Section 5.6). The points p ∈ T ∗ S 2 where dHp and dJp are linearly dependent are: • the two critical points of H (where both dH and dJ vanish); • if x ∈ S 2 is in the southern hemisphere (x3 < 0), then there exist exactly two points, p+ = (x, η, ξ ) and p− = (x, −η, −ξ ), in the cotangent fiber above x where dHp and dJp are linearly dependent; • since dHp and dJp are linearly dependent along the trajectory of the Hamiltonian vector field of H through p+ , this trajectory is also a trajectory of the Hamiltonian vector field of J and hence its projection onto S 2 is a latitudinal (or horizontal) circle. The projection of the trajectory through p− is the same latitudinal circle traced in the opposite direction. 1 2 2 (η

Let (M, ω, H ) be an integrable system of dimension 2n with integrals of motion f1 = H, f2 , . . . , fn . Let c ∈ Rn be a regular value of f := (f1 , . . . , fn ). The corresponding level set f −1 (c) is a Lagrangian submanifold, as it is n-dimensional and its tangent bundle is isotropic. If the flows are complete on f −1 (c), by following them we obtain global coordinates. Any compact component of f −1 (c) must hence be a torus. These components, when they exist, are called Liouville tori. A way to ensure that compact components exist is to have one of the fi ’s proper. T HEOREM 5.14 (Arnold–Liouville [2]). Let (M, ω, H ) be an integrable system of dimension 2n with integrals of motion f1 = H, f2 , . . . , fn . Let c ∈ Rn be a regular value of f := (f1 , . . . , fn ). The level f −1 (c) is a Lagrangian submanifold of M. (a) If the flows of the Hamiltonian vector fields Xf1 , . . . , Xfn starting at a point p ∈ f −1 (c) are complete, then the connected component of f −1 (c) containing p is a homogeneous space for Rn , i.e., is of the form Rn−k × Tk for some k, 0 k n, where Tk is a k-dimensional torus.. With respect to this affine structure, that component has coordinates ϕ1 , . . . , ϕn , known as angle coordinates, in which the flows of Xf1 , . . . , Xfn are linear. (b) There are coordinates ψ1 , . . . , ψn , known as action coordinates, complementary to the angle coordinates, such that the ψi ’s are integrals of motion and ϕ1 , . . . , ϕn , ψ1 , . . . , ψn form a Darboux chart. Therefore, the dynamics of an integrable system has a simple explicit solution in actionangle coordinates. The proof of part (a)—the easy part of the theorem—is sketched above. For the proof of part (b) see, for instance, [2,36]. Geometrically, regular levels being Lagrangian submanifolds implies that, in a neighborhood of a regular value, the map f : M → Rn collecting the given integrals of motion is a Lagrangian fibration, i.e., it is locally trivial and its fibers are Lagrangian submanifolds. Part (a) states that there are coordinates along the fibers, the angle coordinates,56 in which the flows of Xf1 , . . . , Xfn are linear. Part (b) guarantees the existence of coordinates on Rn , the action coordinates, 56 The name angle coordinates is used even if the fibers are not tori.

Symplectic geometry

153

ψ1 , . . . , ψn , complementary to the angle coordinates, that (Poisson) commute among themselves and satisfy {ϕi , ψj } = δij . The action coordinates are generally not the given integrals of motion because ϕ1 , . . . , ϕn , f1 , . . . , fn do not form a Darboux chart. 5.6. Symplectic and Hamiltonian actions Let (M, ω) be a symplectic manifold, and G a Lie group. D EFINITION 5.15. An action57 ψ : G → Diff(M), g → ψg , is a symplectic action if each ψg is a symplectomorphism, i.e., ψ : G → Sympl(M, ω) ⊂ Diff(M). In particular, symplectic actions of R on (M, ω) are in one-to-one correspondence with complete symplectic vector fields on M:

dψt (p)

, p ∈ M. ψ = exp tX ←→ Xp = dt t=0 We may define a symplectic action ψ of S 1 or R on (M, ω) to be Hamiltonian if the vector field X generated by ψ is Hamiltonian, that is, when there is H : M → R with dH = ıX ω. An action of S 1 may be viewed as a periodic action of R. E XAMPLES . 1. On (R2n , ω0 ), the orbits of the action generated by X = − ∂y∂ 1 are lines parallel to the y1 -axis, {(x1 , y1 − t, x2 , y2 , . . . , xn , yn ) | t ∈ R}. Since X is Hamiltonian with Hamiltonian function x1 , this is a Hamiltonian action of R. 2. On the 2-sphere (S 2 , dθ ∧ dh) in cylindrical coordinates, the one-parameter group of diffeomorphisms given by rotation around the vertical axis, ψt (θ, h) = (θ + t, h) (t ∈ R) is a symplectic action of the group S 1 R/2π, as it preserves the area form dθ ∧ dh. Since the vector field corresponding to ψ is Hamiltonian with Hamiltonian function h, this is a Hamiltonian action of S 1 . When G is a product of S 1 ’s or R’s, an action ψ : G → Sympl(M, ω) is called Hamiltonian when the restriction to each 1-dimensional factor is Hamiltonian in the previous sense with Hamiltonian function preserved by the action of the rest of G. For an arbitrary Lie group G, we use an upgraded Hamiltonian function µ, known as a moment map, determined up to an additive local constant by coordinate functions µi indexed by a basis of the Lie algebra of G. We require that the constant be such that µ is equivariant, i.e., µ intertwines the action of G on M and the coadjoint action of G on the dual of its Lie algebra. (If M is compact, equivariance can be achieved by adjusting the

constant so that M µωn = 0. Similarly when there is a fixed point p (on each component of M) by imposing µ(p) = 0.) Let G be a Lie group, g the Lie algebra of G, and g∗ the dual vector space of g. 57 A (smooth) action of G on M is a group homomorphism G → Diff(M), g → ψ , whose evaluation map g M × G → M, (p, g) → ψg (p), is smooth.

154

A. Cannas da Silva

D EFINITION 5.16. An action ψ : G → Diff(M) on a symplectic manifold (M, ω) is a Hamiltonian action if there exists a map µ : M → g∗ satisfying: • For each X ∈ g, we have dµX = ıX# ω, i.e., µX is a Hamiltonian function for the vector field X # , where – µX : M → R, µX (p) := µ(p), X, is the component of µ along X, – X # is the vector field on M generated by the one-parameter subgroup {exp tX | t ∈ R} ⊆ G. • The map µ is equivariant with respect to the given action ψ on M and the coadjoint action: µ ◦ ψg = Ad∗g ◦µ, for all g ∈ G. Then (M, ω, G, µ) is a Hamiltonian G-space and µ is a moment map. This definition matches the previous one when G is an Abelian group R, S 1 or Tn , for which equivariance becomes invariance since the coadjoint action is trivial. E XAMPLES . 1. Let Tn = {(t1 , . . . , tn ) ∈ Cn : |tj | = 1, for all j } be a torus acting on Cn by (t1 , . . . , tn ) · (z1 , . . . , zn ) = (t1k1 z1 , . . . , tnkn zn ), where k1 , . . . , kn ∈ Z are fixed. This action is Hamiltonian with a moment map µ : Cn → (tn )∗ Rn , µ(z1 , . . . , zn ) = − 12 (k1 |z1 |2 , . . . , kn |zn |2 ). 2. When a Lie group G acts on two symplectic manifolds (Mj , ωj ), j = 1, 2, with moment maps µj : Mj → g∗ , the diagonal action of G on M1 × M2 has moment map µ : M1 × M2 → g∗ , µ(p1 , p2 ) = µ1 (p1 ) + µ2 (p2 ). 3. Equip the coadjoint orbits of a Lie group G with the canonical symplectic form (Section 5.1). Then, for each ξ ∈ g∗ , the coadjoint action on the orbit G · ξ is Hamiltonian with moment map simply the inclusion map µ : G · ξ → g∗ . 4. Identify the Lie algebra of the unitary group U(n) with its dual via the inner product A, B = trace(A∗ B). The natural action of U(n) on (Cn , ω0 ) is Hamiltonian with moment map µ : Cn → u(n) given by µ(z) = 2i zz∗ . Similarly, a moment map for the natural action of U(k) on the space (Ck×n , ω0 ) of complex (k × n)-matrices is given by µ(A) = 2i AA∗ for A ∈ Ck×n . Thus the U(n)-action by conjugation on the space 2

(Cn , ω0 ) of complex (n × n)-matrices is Hamiltonian, with moment map given by µ(A) = 2i [A, A∗ ]. 5. For the spherical pendulum (Section 5.5), the energy-momentum map (H, J ) : T ∗ S 2 → R2 is a moment map for the R × S 1 action given by time flow and rotation about the vertical axis. 6. Suppose that a compact Lie group acts on a symplectic manifold (M, ω) in a Hamiltonian way, and that q ∈ M is a fixed point for the G-action. Then, by an equivariant version of Darboux’s theorem,58 there exists a Darboux chart (U, z1 , . . . , zn ) centered at q that is G-equivariant with respect to a linear action of G on Cn . Consider an ε-blow-up of M relative to this chart, for ε sufficiently small. Then G acts on the blow-up in a Hamiltonian way. 58 Equivariant Darboux theorem [136]. Let (M, ω) be a 2n-dimensional symplectic manifold equipped with

a symplectic action of a compact Lie group G, and let q be a fixed point. Then there exists a G-invariant chart

Symplectic geometry

155

The concept of a moment map was introduced by Souriau [119] under the French name application moment; besides the more standard English translation to moment map, the alternative momentum map is also used, and recently James Stasheff has proposed the short unifying new word momap. The name comes from being the generalization of linear and angular momenta in classical mechanics. Let R3 act on (R6 T ∗ R3 , ω0 = dxi ∧ dyi ) by translations: a ∈ R3 −→ ψa ∈ Sympl R6 , ω0 ,

ψa (x, y) = (x + a, y).

The vector field generated by X = a = (a1 , a2 , a3 ) is X # = a1 ∂x∂ 1 + a2 ∂x∂ 2 + a3 ∂x∂ 3 , and the linear momentum map µ : R6 −→ R3 ,

µ(x, y) = y

is a moment map, with µa (x, y) = µ(x, y), a = y · a. Classically, y is called the momentum vector corresponding to the position vector x. The SO(3)-action on R3 by rotations lifts to a symplectic action ψ on the cotangent bundle R6 . The infinitesimal version of this action is59 a ∈ R3 −→ dψ(a) ∈ χ sympl R6 ,

dψ(a)(x, y) = (a × x, a × y).

Then the angular momentum map µ : R6 −→ R3 ,

µ(x, y) = x × y

is a moment map, with µa (x, y) = µ(x, y), a = (x × y) · a. The notion of a moment map associated to a group action on a symplectic manifold formalizes the Noether principle, which asserts that there is a one-to-one correspondence between symmetries (or one-parameter group actions) and integrals of motion (or conserved quantities) for a mechanical system. (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and G-equivariant with respect to a linear action of G on R2n such that ω|U =

n

dxk ∧ dyk .

k=1

A suitable linear action on R2n is equivalent to the induced action of G on Tq M. The proof relies on an equivariant version of the Moser trick and may be found in [70]. 59 The Lie group SO(3) = {A ∈ GL(3; R) | At A = Id and det A = 1}, has Lie algebra, g = {A ∈ gl(3; R) | A + At = 0}, the space of 3 × 3 skew-symmetric matrices. The standard identification of g with R3 carries the Lie bracket to the exterior product: 0 −a3 a2 A = a3 0 −a1 −→ a = (a1 , a2 , a3 ), −a2 a1 0 [A, B] = AB − BA −→ a × b.

156

A. Cannas da Silva

D EFINITION 5.17. An integral of motion of a Hamiltonian G-space (M, ω, G, µ) is a G-invariant function f : M → R. When µ is constant on the trajectories of a Hamiltonian vector field Xf , the corresponding flow {exp tXf | t ∈ R} (regarded as an R-action) is a symmetry of the Hamiltonian G-space (M, ω, G, µ). T HEOREM 5.18 (Noether). Let (M, ω, G, µ) be a Hamiltonian G-space where G is connected. If f is an integral of motion, the flow of its Hamiltonian vector field Xf is a symmetry. If the flow of some Hamiltonian vector field Xf is a symmetry, then a corresponding Hamiltonian function f is an integral of motion. P ROOF. Let µX = µ, X : M → R for X ∈ g. We have LXf µX = ıXf dµX = ıXf ıX# ω = −ıX# ıXf ω = −ıX# df = −LX# f . So µ is invariant over the flow of Xf if and only if f is invariant under the infinitesimal G-action. We now turn to the questions of existence and uniqueness of moment maps. Let g be a Lie algebra, and let C k := Λk g∗ be the set of k-cochains on g, that is, of alternating k-linear maps g × · · · × g → R. The linear operator δ : C k → C k+1 defined by δc(X0 , . . . , Xk ) = i<j (−1)i+j c([Xi , Xj ], X0 , . . . , Xˆ i , . . . , Xˆ j , . . . , Xk ) satisfies δ 2 = 0. The Lie algebra cohomology groups (or Chevalley cohomology groups) of g are the cohoδ

δ

δ

mology groups of the complex 0 −→ C 0 −→ C 1 −→ · · ·: H k (g; R) :=

ker δ : C k → C k+1 . im δ : C k−1 → C k

It is always H 0 (g; R) = R. If c ∈ C 1 = g∗ , then δc(X, Y ) = −c([X, Y ]). The commutator ideal [g, g] is the subspace of g spanned by {[X, Y ] | X, Y ∈ g}. Since δc = 0 if and only if c vanishes on [g, g], we conclude that H 1 (g; R) = [g, g]0 , where [g, g]0 ⊆ g∗ is the annihilator of [g, g]. An element of C 2 is an alternating bilinear map c : g × g → R, and δc(X, Y, Z) = −c([X, Y ], Z) + c([X, Z], Y ) − c([Y, Z], X). If c = δb for some b ∈ C 1 , then c(X, Y ) = (δb)(X, Y ) = −b([X, Y ]). If g is the Lie algebra of a compact connected Lie group G, then by averaging one can show that the de Rham cohomology may be computed from the subcomplex of G-invariant k (G). forms, and hence H k (g; R) = HdeRham P ROPOSITION 5.19. If H 1 (g; R) = H 2 (g, R) = 0, then any symplectic G-action is Hamiltonian. P ROOF. Let ψ : G → Sympl(M, ω) be a symplectic action of G on a symplectic manifold (M, ω). Since H 1 (g; R) = 0 means that [g, g] = g, and since commutators of symplectic vector fields are Hamiltonian, we have dψ : g = [g, g] → χ ham (M). The action ψ is Hamiltonian if and only if there is a Lie algebra homomorphism µ∗ : g → C ∞ (M) such that the Hamiltonian vector field of µ∗ (ξ ) is dψ(ξ ). We first take an arbitrary vector space lift τ : g → C ∞ (M) with this property, i.e., for each basis vector X ∈ g, we choose τ (X) = τ X ∈ C ∞ (M) such that v(τ X ) = dψ(X). The map X → τ X may not be a Lie algebra homomorphism. By construction, τ [X,Y ] is a Hamiltonian function for [X, Y ]# , and

Symplectic geometry

157

(as computed in Section 5.5) {τ X , τ Y } is a Hamiltonian function for −[X # , Y # ]. Since [X, Y ]# = −[X # , Y # ], the corresponding Hamiltonian functions must differ by a constant: τ [X,Y ] − τ X , τ Y = c(X, Y ) ∈ R. By the Jacobi identity, δc = 0. Since H 2 (g; R) = 0, there is b ∈ g∗ satisfying c = δb, c(X, Y ) = −b([X, Y ]). We define µ∗ : g −→ C ∞ (M), X −→ µ∗ (X) = τ X + b(X) = µX . Now µ∗ is a Lie algebra homomorphism: µ∗ ([X, Y ]) = {τ X , τ Y } = {µX , µY }.

By the Whitehead lemmas (see, for instance, [77, pp. 93–95]) a semisimple Lie group G has H 1 (g; R) = H 2 (g; R) = 0. As a corollary, when G is semisimple, any symplectic G-action is Hamiltonian.60 P ROPOSITION 5.20. For a connected Lie group G, if H 1 (g; R) = 0, then moment maps for Hamiltonian G-actions are unique. P ROOF. Suppose that µ1 and µ2 are two moment maps for an action ψ . For each X ∈ g, X X X # µX 1 and µ2 are both Hamiltonian functions for X , thus µ1 − µ2 = c(X) is locally con∗ stant. This defines c ∈ g , X → c(X). Since the corresponding µ∗i : g → C ∞ (M) are Lie algebra homomorphisms, we have c([X, Y ]) = 0, ∀X, Y ∈ g, i.e., c ∈ [g, g]0 = {0}. Hence, µ1 = µ 2 . In general, if µ : M → g∗ is a moment map, then given any c ∈ [g, g]0 , µ1 = µ + c is another moment map. In other words, moment maps are unique up to elements of the dual of the Lie algebra that annihilate the commutator ideal. The two extreme cases are when • G is semisimple: any symplectic action is Hamiltonian, moment maps are unique; • G is Abelian: symplectic actions may not be Hamiltonian, moment maps are unique up to a constant c ∈ g∗ . 60 A compact Lie group G has H 1 (g; R) = H 2 (g; R) = 0 if and only if it is semisimple. In fact, a compact Lie

group G is semisimple when g = [g, g]. The unitary group U(n) is not semisimple because the multiples of the identity, S 1 · Id, form a nontrivial center; at the level of the Lie algebra, this corresponds to the subspace R · Id of scalar matrices, which are not commutators since they are not traceless. Any Abelian Lie group is not semisimple. Any direct product of the other compact classical groups SU(n), SO(n) and Sp(n) is semisimple. An arbitrary compact Lie group admits a finite cover by a direct product of tori and semisimple Lie groups.

158

A. Cannas da Silva

5.7. Convexity Atiyah, Guillemin and Sternberg [4,68] showed that the image of the moment map for a Hamiltonian torus action on a compact connected symplectic manifold is always a polytope.61 A proof of this theorem can also be found in [99]. T HEOREM 5.21 (Atiyah, Guillemin–Sternberg). Let (M, ω) be a compact connected symplectic manifold. Suppose that ψ : Tm → Sympl(M, ω) is a Hamiltonian action of an m-torus with moment map µ : M → Rm . Then: (a) the levels µ−1 (c) are connected (c ∈ Rm ); (b) the image µ(M) is convex; (c) µ(M) is the convex hull of the images of the fixed points of the action. The image µ(M) of the moment map is called the moment polytope. E XAMPLES . 1. Suppose that Tm acts linearly on (Cn , ω0 ). Let λ(1) , . . . , λ(n) ∈ Zm be the weights appearing in the corresponding weight space decomposition, that is, Cn

n

Vλ(k) ,

k=1 (k)

(k)

where, for λ(k) = (λ1 , . . . , λm ), the torus Tm acts on the complex line Vλ(k) by

(k) j λj tj

(eit1 , . . . , eitm ) · v = ei v. If the action is effective62 , then m n and the (1) (n) weights λ , . . . , λ are part of a Z-basis of Zm . If the action is symplectic (hence Hamiltonian in this case), then the weight spaces Vλ(k) are symplectic subspaces. In this case, a moment map is given by 1 (k) µ(v) = − λ |vλ(k) |2 , 2 n

k=1

where | · | is the standard norm63 and v = vλ(1) + · · · + vλ(n) is the weight space decomposition of v. We conclude that, if Tn acts on Cn in a linear, effective and Hamiltonian way, then any moment map µ is a submersion, i.e., each differential dµv : Cn → Rn (v ∈ Cn ) is surjective. 61 A polytope in Rn is the convex hull of a finite number of points in Rn . A convex polyhedron is a subset of Rn

that is the intersection of a finite number of affine half-spaces. Hence, polytopes coincide with bounded convex polyhedra. 62 An action of a group G on a manifold M is called effective if each group element g = e moves at least one . point p ∈ M, that is, p∈M Gp = {e}, where Gp = {g ∈ G | g · p = p} is the stabilizer of p. 63 The standard inner product satisfies v, w = ω (v, J v) where J ∂ = i ∂ and J ∂ = −i ∂ . In particular, the 0 ∂z ∂z ∂ z¯ ∂ z¯ standard norm is invariant for a symplectic complex-linear action.

Symplectic geometry

159

2. Consider a coadjoint orbit Oλ for the unitary group U(n). Multiplying by i, the orbit Oλ can be viewed as the set of Hermitian matrices with a given eigenvalue spectrum λ = (λ1 · · · λn ). The restriction of the coadjoint action to the maximal torus Tn of diagonal unitary matrices is Hamiltonian with moment map µ : Oλ → Rn taking a matrix to the vector of its diagonal entries. Then the moment polytope µ(Oλ ) is the convex hull C of the points given by all the permutations of (λ1 , . . . , λn ). This is a rephrasing of the classical theorem of Schur (µ(Oλ ) ⊆ C) and Horn (C ⊆ µ(Oλ )). Example 1 is related to the universal local picture for a moment map near a fixed point of a Hamiltonian torus action: T HEOREM 5.22. Let (M 2n , ω, Tm , µ) be a Hamiltonian Tm -space, where q is a fixed point. Then there exists a chart (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and weights λ(1) , . . . , λ(n) ∈ Zm such that ω|U =

n k=1

1 (k) 2 and µ|U = µ(q) − λ xk + yk2 . 2 n

dxk ∧ dyk

k=1

The following two results use the crucial fact that any effective action of an m-torus on a manifold has orbits of dimension m; a proof may be found in [19]. C OROLLARY 5.23. Under the conditions of the convexity theorem, if the Tm -action is effective, then there must be at least m + 1 fixed points. P ROOF. At a point p of an m-dimensional orbit the moment map is a submersion, i.e., (dµ1 )p , . . . , (dµm )p are linearly independent. Hence, µ(p) is an interior point of µ(M), and µ(M) is a nondegenerate polytope. A nondegenerate polytope in Rm has at least m + 1 vertices. The vertices of µ(M) are images of fixed points. P ROPOSITION 5.24. Let (M, ω, Tm , µ) be a Hamiltonian Tm -space. If the Tm -action is effective, then dim M 2m. P ROOF. Since the moment map is constant on an orbit O, for p ∈ O the differential dµp : Tp M → g∗ maps Tp O to 0. Thus Tp O ⊆ ker dµp = (Tp O)ω , where (Tp O)ω is the symplectic orthogonal of Tp O. This shows that orbits O of a Hamiltonian torus action are isotropic submanifolds of M. In particular, by symplectic linear algebra we have that dim O 12 dim M. Now consider an m-dimensional orbit. For a Hamiltonian action of an arbitrary compact Lie group G on a compact symplectic manifold (M, ω), the following non-Abelian convexity theorem was proved by Kirwan [81]: if µ : M → g∗ is a moment map, then the intersection µ(M) ∩ t∗+ of the image of µ with a Weyl chamber for a Cartan subalgebra t ⊆ g is a convex polytope. This had been conjectured by Guillemin and Sternberg and proved by them in particular cases.

160

A. Cannas da Silva

6. Symplectic reduction 6.1. Marsden–Weinstein–Meyer theorem Classical physicists realized that, whenever there is a symmetry group of dimension k acting on a mechanical system, the number of degrees of freedom for the position and momenta of the particles may be reduced by 2k. Symplectic reduction formulates this process mathematically. T HEOREM 6.1 (Marsden–Weinstein, Meyer [92,102]). Let (M, ω, G, µ) be a Hamiltonian G-space (Section 5.6) for a compact Lie group G. Let i : µ−1 (0) → M be the inclusion map. Assume that G acts freely on µ−1 (0). Then (a) the orbit space Mred = µ−1 (0)/G is a manifold, (b) π : µ−1 (0) → Mred is a principal G-bundle, and (c) there is a symplectic form ωred on Mred satisfying i ∗ ω = π ∗ ωred . D EFINITION 6.2. The symplectic manifold (Mred , ωred ) is the reduction (or reduced space, or symplectic quotient) of (M, ω) with respect to G, µ. When M is Kähler and the action of G preserves the complex structure, we can show that the symplectic reduction has a natural Kähler structure. Let (M, ω, G, µ) be a Hamiltonian G-space for a compact Lie group G. To reduce at a level ξ ∈ g∗ of µ, we need µ−1 (ξ ) to be preserved by G, or else take the G-orbit of µ−1 (ξ ), or else take the quotient by the maximal subgroup of G that preserves µ−1 (ξ ). Since µ is equivariant, G preserves µ−1 (ξ ) if and only if Ad∗g ξ = ξ , ∀g ∈ G. Of course, the level 0 is always preserved. Also, when G is a torus, any level is preserved and reduction at ξ for the moment map µ, is equivalent to reduction at 0 for a shifted moment map φ : M → g∗ , φ(p) := µ(p) − ξ . In general, let O be a coadjoint orbit in g∗ equipped with the canonical symplectic formωO (defined in Section 5.1). Let O− be the orbit O equipped with −ωO . The natural product action of G on M × O− is Hamiltonian with moment map µO (p, ξ ) = µ(p) − ξ . If the hypothesis of Theorem 6.1 is satisfied for M × O− , then one obtains a reduced space with respect to the coadjoint orbit O. E XAMPLES . 1. The standard symplectic form on Cn is ω0 = 2i dzi ∧ d z¯ i = dxi ∧ dyi = ri dri ∧ dθi in polar coordinates. The S 1 -action on (Cn , ω0 ) where eit ∈ S 1 acts as multiplication by eit has vector field X # = ∂θ∂ 1 + ∂θ∂ 2 + · · · + ∂θ∂ n . This action is 2 Hamiltonian with moment map µ : Cn → R, µ(z) = − |z|2 , since ıX# ω = ri dri = − 12 dri2 = dµ. The level µ−1 (− 12 ) is the unit sphere S 2n−1 , whose orbit space is the projective space, µ

−1

1 / 1 − S = S 2n−1 /S 1 = CPn−1 . 2

Symplectic geometry

161

The reduced symplectic form at level − 12 is ωred = ωFS the Fubini–Study symplectic form. Indeed, if pr : Cn+1 \ {0} → CPn is the standard projection, the forms pr∗ ωFS = i ¯ 2 2n+1 . 2 ∂ ∂ log(|z| ) and ω0 have the same restriction to S Id k×n 2. Consider the natural action of U(k) on C with moment map µ(A) = 2i AA∗ + 2i k×n −1 k×n ∗ for A ∈ C (Section 5.6). Since µ (0) = {A ∈ C | AA = Id}, the reduced manifold is the Grassmannian of k-planes in Cn : µ−1 (0)/U(k) = G(k, n). For the case where G = S 1 and dim M = 4, here is a glimpse of reduction. Let µ : M → R be the moment map and p ∈ µ−1 (0). Choose local coordinates near p: θ along the orbit through p, µ given by the moment map, and η1 , η2 the pullback of coordinates on Mred = µ−1 (0)/S 1 . Then the symplectic form can be written ω = A dθ ∧ dµ +

Bj dθ ∧ dηj +

Cj dµ ∧ dηj + D dη1 ∧ dη2 .

∂ )ω, we must have A = 1, Bj = 0. Since ω is symplectic, it must be D = 0. As dµ = ı( ∂θ ∗ Hence, i ω = D dη1 ∧ dη2 is the pullback of a symplectic form on Mred . The actual proof of Theorem 6.1 requires some preliminary ingredients. Let µ : M → g∗ be the moment map for an (Hamiltonian) action of a Lie group G on a symplectic manifold (M, ω). Let gp be the Lie algebra of the stabilizer of a point p ∈ M, let g0p = {ξ ∈ g∗ | ξ, X = 0, ∀X ∈ gp } be the annihilator of gp , and let Op be the G-orbit through p. Since ωp (Xp# , v) = dµp (v), X, for all v ∈ Tp M and all X ∈ g, the differential dµp : Tp M → g∗ has

ker dµp = (Tp Op )ωp

and

im dµp = g0p .

Consequently, the action is locally free64 at p if and only if p is a regular point of µ (i.e., dµp is surjective), and we obtain: L EMMA 6.3. If G acts freely on µ−1 (0), then 0 is a regular value of µ, the level µ−1 (0) is a submanifold of M of codimension dim G, and, for p ∈ µ−1 (0), the tangent space Tp µ−1 (0) = ker dµp is the symplectic orthogonal to Tp Op in Tp M. In particular, orbits in µ−1 (0) are isotropic. Since any tangent vector to the orbit is the value of a vector field generated by the group, we can show this directly by computing, for any X, Y ∈ g and p ∈ µ−1 (0), the Hamiltonian function for [Y # , X # ] = [Y, X]# at that point: ωp (Xp# , Yp# ) = µ[Y,X] (p) = 0. L EMMA 6.4. Let (V , Ω) be a symplectic vector space, and I an isotropic subspace. Then Ω induces a canonical symplectic structure Ωred on I Ω /I . 64 The action is locally free at p when g = {0}, i.e., the stabilizer of p is a discrete group. The action is free at p p when the stabilizer of p is trivial, i.e., Gp = {e}.

162

A. Cannas da Silva

P ROOF. Let [u], [v] be the classes in I Ω /I of u, v ∈ I Ω . We have Ω(u + i, v + j ) = Ω(u, v), ∀i, j ∈ I , because Ω(u, j ) = Ω(i, v) = Ω(i, j ) = 0. Hence, we can define Ωred ([u], [v]) := Ω(u, v). This is nondegenerate: if u ∈ I Ω has Ω(u, v) = 0, for all v ∈ I Ω , then u ∈ (I Ω )Ω = I , i.e., [u] = 0. P ROPOSITION 6.5. If a compact Lie group G acts freely on a manifold M, then M/G is a manifold and the map π : M → M/G is a principal G-bundle. P ROOF. We first show that, for any p ∈ M, the G-orbit through p is a compact submanifold of M diffeomorphic to G.65 The G-orbit through p is the image of the smooth injective map evp : G → M, evp (g) = g · p. The map evp is proper because, if A is a compact, hence closed, subset of M, then its inverse image (evp )−1 (A), being a closed subset of the compact Lie group G, is also compact. The differential d(evp )e is injective because d(evp )e (X) = 0 ⇔ Xp# = 0 ⇔ X = 0, ∀X ∈ Te G, as the action is free. At any other point g ∈ G, for X ∈ Tg G we have d(evp )g (X) = 0 ⇔ d(evp ◦ Rg )e ◦ (dRg −1 )g (X) = 0, where Rg : G → G, h → hg, is right multiplication by g. But evp ◦ Rg = evg·p has an injective differential at e, and (dRg −1 )g is an isomorphism. It follows that d(evp )g is always injective, so evp is an immersion. We conclude that evp is a closed embedding. We now apply the slice theorem66 which is an equivariant tubular neighborhood theorem. For p ∈ M, let q = π(p) ∈ M/G. Choose a G-invariant neighborhood U of p as in the slice theorem, so that U G × S where S is an appropriate slice. Then π(U) = U/G =: V is a neighborhood of q in M/G homeomorphic67 to S. Such neighborhoods V are used as charts on M/G. To show that the associated transition maps are smooth, consider two G-invariant open sets U1 , U2 in M and corresponding slices S1 , S2 . Then S12 = S1 ∩ U2 , S21 = S2 ∩ U1 are both slices for the G-action on U1 ∩ U2 . To compute the transition map S12 → S21 , consider the sequence S12 −→ {e} × S12 → G × S12 −→ U1 ∩ U2 and similarly pr for S21 . The composition S12 → U1 ∩ U2 −→ G × S21 −→ S21 is smooth. Finally, we show that π : M → M/G is a principal G-bundle. For p ∈ M, q = π(p), choose a G-invariant neighborhood U of p of the form η : G × S −→ U . Then V = U/G S is the corresponding neighborhood of q in M/G: η

M ⊇U G×S G×V ↓π ↓ M/G ⊇ V = V 65 Even if the action is not free, the orbit through p is a compact submanifold of M. In that case, the orbit of a point p is diffeomorphic to the quotient G/Gp of G by the stabilizer of p. 66 Slice theorem. Let G be a compact Lie group acting on a manifold M such that G acts freely at p ∈ M. Let S be a transverse section to Op at p (this is called a slice). Choose a coordinate chart x1 , . . . , xn centered at p such that Op G is given by x1 = · · · = xk = 0 and S by xk+1 = · · · = xn = 0. Let Sε = S ∩ Bε where Bε is the ball of radius ε centered at 0 with respect to these coordinates. Let η : G × S → M, η(g, s) = g · s. Then, for sufficiently small ε, the map η : G × Sε → M takes G × Sε diffeomorphically onto a G-invariant neighborhood U of the G-orbit through p. In particular, if the action of G is free at p, then the action is free on U , so the set of points where G acts freely is open. 67 We equip the orbit space M/G with the quotient topology, i.e., V ⊆ M/G is open if and only if π −1 (V) is open in M.

Symplectic geometry

163

Since the projection on the right is smooth, π is smooth. By considering the overlap of two trivializations φ1 : U1 → G × V1 and φ2 : U2 → G × V2 , we check that the transition map φ2 ◦ φ1−1 = (σ12 , id) : G × (V1 ∩ V2 ) → G × (V1 ∩ V2 ) is smooth. P ROOF OF T HEOREM 6.1. Since G acts freely on µ−1 (0), by Lemma 6.3 the level µ−1 (0) is a submanifold. Applying Proposition 6.5 to the free action of G on the manifold µ−1 (0), we conclude the assertions (a) and (b). At p ∈ µ−1 (0) the tangent space to the orbit Tp Op is an isotropic subspace of the symplectic vector space (Tp M, ωp ). By Lemma 6.4 there is a canonical symplectic structure on the quotient Tp µ−1 (0)/Tp Op . The point [p] ∈ Mred = µ−1 (0)/G has tangent space T[p] Mred Tp µ−1 (0)/Tp Op . This gives a well-defined nondegenerate 2-form ωred on Mred because ω is G-invariant. By construction i ∗ ω = π ∗ ωred where µ−1 (0) ↓π Mred

i

→ M

The injectivity of π ∗ yields closedness: π ∗ dωred = dπ ∗ ωred = dı ∗ ω = ı ∗ dω = 0.

6.2. Applications and generalizations Let (M, ω, G, µ) be a Hamiltonian G-space for a compact Lie group G. Suppose that another Lie group H acts on (M, ω) in a Hamiltonian way with moment map φ : M → h∗ . Suppose that the H -action commutes with the G-action, that φ is G-invariant and that µ is H -invariant. Assuming that G acts freely on µ−1 (0), let (Mred , ωred ) be the corresponding reduced space. Since the action of H preserves µ−1 (0) and ω and commutes with the G-action, the reduced space (Mred , ωred ) inherits a symplectic action of H . Since φ is preserved by the G-action, the restriction of this moment map to µ−1 (0) descends to a moment map φred : Mred → h∗ satisfying φred ◦ π = φ ◦ i, where π : µ−1 (0) → Mred and i : µ−1 (0) → M. Therefore, (Mred , ωred , H, φred ) is a Hamiltonian H -space. Consider now the action of a product group G = G1 × G2 , where G1 and G2 are compact connected Lie groups. We have g = g1 ⊕ g2 and g∗ = g∗1 ⊕ g∗2 . Suppose that (M, ω, G, ψ) is a Hamiltonian G-space with moment map ψ = (ψ1 , ψ2 ) : M −→ g∗1 ⊕ g∗2 , where ψi : M → g∗i for i = 1, 2. The fact that ψ is equivariant implies that ψ1 is invariant under G2 and ψ2 is invariant under G1 . Assume that G1 acts freely on Z1 := ψ1−1 (0). Let (M1 = Z1 /G1 , ω1 ) be the reduction of (M, ω) with respect to G1 , ψ1 . From the observation above, (M1 , ω1 ) inherits a Hamiltonian G2 -action with moment map µ2 : M1 → g∗2 such that µ2 ◦ π = ψ2 ◦ i, where π : Z1 → M1 and i : Z1 → M. If G acts freely on ψ −1 (0, 0), then G2 acts freely on µ−1 2 (0), and there is a natural symplectomorphism −1 µ−1 2 (0)/G2 ψ (0, 0)/G.

164

A. Cannas da Silva

This technique of performing reduction with respect to one factor of a product group at a time is called reduction in stages. It may be extended to reduction by a normal subgroup H ⊂ G and by the corresponding quotient group G/H . E XAMPLE . Finding symmetries for a mechanical problem may reduce degrees of freedom by two at a time: an integral of motion f for a 2n-dimensional Hamiltonian system (M, ω, H ) may allow to understand the trajectories of this system in terms of the trajectories of a (2n − 2)-dimensional Hamiltonian system (Mred , ωred , Hred ). Locally this process goes as follows. Let (U, x1 , . . . , xn , ξ1 , . . . , ξn ) be a Darboux chart for M such ∂H , the trajectories of that f = ξn .68 Since ξn is an integral of motion, 0 = {ξn , H } = − ∂x n the Hamiltonian vector field XH lie on a constant level ξn = c (Proposition 5.12), and H does not depend on xn . The reduced space is Ured = {(x1 , . . . , xn−1 , ξ1 , . . . , ξn−1 ) | ∃a: (x1 , . . . , xn−1 , a, ξ1 , . . . , ξn−1 , c) ∈ U} and the reduced Hamiltonian is Hred : Ured → R, Hred (x1 , . . . , xn−1 , ξ1 , . . . , ξn−1 ) = H (x1 , . . . , xn−1 , a, ξ1 , . . . , ξn−1 , c) for some a. In order to find the trajectories of the original system on the hypersurface ξn = c, we look for the trajectories (x1 (t), . . . , xn−1 (t), ξ1 (t), . . . , ξn−1 (t)) of the reduced system on Ured , ∂H n and integrate the equation dx dt (t) = ∂ξn to obtain the original trajectories where

xn (t) = xn (0) + ξn (t) = c.

t

∂H 0 ∂ξn (x1 (t), . . . , xn−1 (t), ξ1 (t), . . . , ξn−1 (t), c) dt,

By Sard’s theorem, the singular values of a moment map µ : M → g∗ form a set of measure zero. So, perturbing if necessary, we may assume that a level of µ is regular hence, when G is compact, that any point p of that level has finite stabilizer Gp . Let Op be the orbit of p. By the slice theorem for the case of orbifolds, near Op the orbit space of the level is modeled by S/Gp , where S is a Gp -invariant disk in the level and transverse to Op (a slice). Thus, the orbit space is an orbifold.69 This implies that, when G = Tn is an n-torus, for most levels reduction goes through, however the quotient space is not necessarily a manifold but an orbifold. Roughly speaking, orbifolds are singular manifolds where each singularity is locally modeled on Rm /Γ , for some finite group Γ ⊂ GL(m; R). The differential-geometric notions of vector fields, differential forms, exterior 68 To obtain such a chart, in the proof of Darboux’s Theorem 1.9 start with coordinates (x , . . . , x , y , . . . y ) n 1 n 1 such that yn = f and ∂ = Xf . ∂xn 69 Let |M| be a Hausdorff topological space satisfying the second axiom of countability. An orbifold chart on |M| is a triple (V, Γ, ϕ), where V is a connected open subset of some Euclidean space Rm , Γ is a finite group that acts

linearly on V so that the set of points where the action is not free has codimension at least two, and ϕ : V → |M| is a Γ -invariant map inducing a homeomorphism from V/Γ onto its image U ⊂ |M|. An orbifold atlas A for |M| is a collection of orbifold charts on |M| such that: the collection of images U forms a basis of open sets in |M|, and the charts are compatible in the sense that, whenever two charts (V1 , Γ1 , ϕ1 ) and (V2 , Γ2 , ϕ2 ) satisfy U1 ⊆ U2 , there exists an injective homomorphism λ : Γ1 → Γ2 and a λ-equivariant open embedding ψ : V1 → V2 such that ϕ2 ◦ ψ = ϕ1 . Two orbifold atlases are equivalent if their union is still an atlas. An m-dimensional orbifold M is a Hausdorff topological space |M| satisfying the second axiom of countability, plus an equivalence class of orbifold atlases on |M|. We do not require the action of each group Γ to be effective. Given a point p on an orbifold M, let (V, Γ, ϕ) be an orbifold chart for a neighborhood U of p. The orbifold structure group of p, Γp , is (the isomorphism class of) the stabilizer of a preimage of p under φ. Orbifolds were introduced by Satake in [114].

Symplectic geometry

165

differentiation, group actions, etc., extend naturally to orbifolds by gluing corresponding local Γ -invariant or Γ -equivariant objects. In particular, a symplectic orbifold is a pair (M, ω) where M is an orbifold and ω is a closed 2-form on M that is nondegenerate at every point. E XAMPLES . The S 1 -action on C2 given by eiθ · (z1 , z2 ) = (eikθ z1 , eiθ z2 ), for some integers k and , has moment map µ : C2 → R, (z1 , z2 ) → − 12 (k|z1 |2 + |z2 |2 ). Any ξ < 0 is a regular value and µ−1 (ξ ) is a 3-dimensional ellipsoid. 2πm When = 1 and k 2, the stabilizer of (z1 , z2 ) is {1} if z2 = 0 and is Zk = {ei k | m = 0, 1, . . . , k − 1} if z2 = 0. The reduced space µ−1 (ξ )/S 1 is then called a teardrop orbifold or conehead; it has one cone (or dunce cap) singularity with cone angle 2π k , that is, a point with orbifold structure group Zk . When k, 2 are relatively prime, for z1 , z2 = 0 the stabilizer of (z1 , 0) is Zk , of (0, z2 ) is Z and of (z1 , z2 ) is {1}. The quotient µ−1 (ξ )/S 1 is called a football orbifold: it has two 2π cone singularities, with angles 2π k and . For S 1 acting on Cn by eiθ · (z1 , . . . , zn ) = (eik1 θ z1 , . . . , eikn θ zn ) the reduced spaces are orbifolds called weighted (or twisted) projective spaces. Let (M, ω) be a symplectic manifold where S 1 acts in a Hamiltonian way, ρ : S 1 → Diff(M), with moment map µ : M → R. Suppose that: • M has a unique nondegenerate minimum at q where µ(q) = 0, and • for ε sufficiently small, S 1 acts freely on the level set µ−1 (ε). Let C be equipped with the symplectic form −i dz ∧ d z¯ . Then the action of S 1 on the product ψ : S 1 → Diff(M × C), ψt (p, z) = (ρt (p), t · z), is Hamiltonian with moment map φ : M × C −→ R,

φ(p, z) = µ(p) − |z|2 .

Observe that S 1 acts freely on the ε-level of φ for ε small enough: φ −1 (ε) = (p, z) ∈ M × C | µ(p) − |z|2 = ε = (p, 0) ∈ M × C | µ(p) = ε ∪ (p, z) ∈ M × C | |z|2 = µ(p) − ε > 0 . The reduced space is hence φ −1 (ε)/S 1 µ−1 (ε)/S 1 ∪ p ∈ M | µ(p) > ε . The open submanifold of M given by {p ∈ M | µ(p) > ε} embeds as an open dense submanifold into φ −1 (ε)/S 1 . The reduced space φ −1 (ε)/S 1 is the ε-blow-up of M at q (Section 5.6). This global description of blow-up for Hamiltonian S 1 -spaces is due to Lerman [86], as a particular instance of his cutting technique. Symplectic cutting is the application of symplectic reduction to the product of a Hamiltonian S 1 -space with the standard

166

A. Cannas da Silva

C as above, in a way that the reduced space for the original Hamiltonian S 1 -space embeds symplectically as a codimension 2 submanifold in a symplectic manifold. As it is a local construction, the cutting operation may be more generally performed at a local minimum (or maximum) of the moment map µ. There is a remaining S 1 -action on the cut space ε Mcut := φ −1 (ε)/S 1 induced by τ : S 1 −→ Diff(M × C),

τt (p, z) = ρt (p), z .

In fact, τ is a Hamiltonian S 1 -action on M × C that commutes with ψ, thus descends to ε an action τ˜ : S 1 → Diff(Mcut ), which is also Hamiltonian. Loosely speaking, the cutting technique provides a Hamiltonian way to close the open manifold {p ∈ M | µ(p) > ε}, by using the reduced space at level ε, µ−1 (ε)/S 1 . We may similarly close {p ∈ M | µ(p) < ε}. The resulting Hamiltonian S 1 -spaces are called cut ε ε spaces, and denoted Mcut and Mcut . If another group G acts on M in a Hamiltonian way 1 that commutes with the S -action, then the cut spaces are also Hamiltonian G-spaces.

6.3. Moment map in gauge theory Let G be a Lie group and P a principal G-bundle over B.70 If A is a connection (form)71 1 on P , and if a ∈ Ωhoriz ⊗ g is G-invariant for the product action, then A + a is also a 1 connection on P . Reciprocally, any two connections on P differ by an a ∈ (Ωhoriz ⊗ g)G . 70 Let G be a Lie group and B a manifold. A principal G-bundle over B is a fibration π : P → B (Section 4.2) with a free action of G (the structure group) on the total space P , such that the base B is the orbit space, the map π is the point-orbit projection and the local trivializations are of the form ϕU = (π, sU ) : π −1 (U) → U × G with sU (g · p) = g · sU (p) for all g ∈ G and all p ∈ π −1 (U). A principal G-bundle is represented by a diagram

G →

P ↓π B

For instance, the Hopf fibration is a principal S 1 -bundle over S 2 (= CP1 ) with total space S 3 regarded as unit vectors in C2 where circle elements act by complex multiplication. 71 An action ψ : G → Diff(P ) induces an infinitesimal action dψ : g → χ(P ) mapping X ∈ g to the vector field X # generated by the one-parameter group {exp tX(e) | t ∈ R} ⊆ G. Fix a basis X1 , . . . , Xk of g. Let P be a principal G-bundle over B. Since the G-action is free, the vector fields X1# , . . . , Xk# are linearly independent at each p ∈ P . The vertical bundle V is the rank k subbundle of T P generated by X1# , . . . , Xk# . Alternatively, V is the set of vectors tangent to P that lie in the kernel of the derivative of the bundle projection π , so V is indeed independent of the choice of basis for g. An (Ehresmann) connection on P is a choice of a splitting T P = V ⊕ H , where H (called the horizontal bundle) is a G-invariant subbundle of T P complementary to the vertical bundle V . A connection form on P is a Lie-algebra-valued 1-form A = ki=1 Ai ⊗ Xi ∈ Ω 1 (P ) ⊗ g such that A is G-invariant, with respect to the product action of G on Ω 1 (P ) (induced by the action on P ) and on g (the adjoint action), and A is vertical, in the sense that ıX# A = X for any X ∈ g. A connection T P = V ⊕ H determines a connection (form) A and vice-versa by the formula H = ker A = {v ∈ T P | ıv A = 0}. Given a connection on P , the splitting T P = V ⊕ H induces splittings for bundles T ∗ P = V ∗ ⊕ H ∗ , ∧2 T ∗ P = (∧2 V ∗ ) ⊕ (V ∗ ∧ H ∗ ) ⊕ (∧2 H ∗ ), 1 ⊕ Ω1 2 2 2 2 etc., and for their sections: Ω 1 (P ) = Ωvert horiz , Ω (P ) = Ωvert ⊕ Ωmix ⊕ Ωhoriz , etc. The corresponding 1 connection form A is in Ωvert ⊗ g.

Symplectic geometry

167

We conclude that the set A of all connections on the principal G-bundle P is an affine 1 space modeled on the linear space a = (Ωhoriz ⊗ g)G . Now let P be a principal G-bundle over a compact Riemann surface. Suppose that the group G is compact or semisimple. Atiyah and Bott [6] noticed that the corresponding space A of all connections may be treated as an infinite-dimensional symplectic manifold. This requires choosing a G-invariant inner product ·, · on g, which always exists, either by averaging any inner product when G is compact, or by using the Killing form on semisimple groups. Since A is an affine space, its tangent space at any point A is identified with the model linear space a. With respect to a basis X1 , . . . , Xk for the Lie algebra g, elements a, b ∈ a are written a=

ai ⊗ X i

and b =

bi ⊗ Xi .

If we wedge a and b, and then integrate over B, we obtain a real number: 2 G ω : a × a −→ Ωhoriz (P ) Ω 2 (B) −→ R, ai ∧ bj Xi , Xj −→ ai ∧ bj Xi , Xj . (a, b) −→ i,j

B

i,j

We used that the pullback π ∗ : Ω 2 (B) → Ω 2 (P ) is an isomorphism onto its image 2 (Ωhoriz (P ))G . When ω(a, b) = 0 for all b ∈ a, then a must be zero. The map ω is nondegenerate, skew-symmetric, bilinear and constant in the sense that it does not depend on the base point A. Therefore, it has the right to be called a symplectic form on A, so the pair (A, ω) is an infinite-dimensional symplectic manifold. A diffeomorphism f : P → P commuting with the G-action determines a diffeomorphism fbasic : B → B by projection. Such a diffeomorphism f is called a gauge transformation if the induced fbasic is the identity. The gauge group of P is the group G of all gauge transformations of P . The derivative of an f ∈ G takes an Ehresmann connection T P = V ⊕ H to another connection T P = V ⊕ Hf , and thus induces an action of G in the space A of all connections. Atiyah and Bott [6] noticed that the action of G on (A, ω) is Hamiltonian, where the moment map (appropriately interpreted) is G µ : A −→ Ω 2 (P ) ⊗ g , A −→ curv A, i.e., the moment map is the curvature.72 The reduced space M = µ−1 (0)/G is the space of flat connections modulo gauge equivalence, known as the moduli space of flat connections, which is a finite-dimensional symplectic orbifold. 72 The exterior derivative of a connection A decomposes into three components,

2 2 ⊕ Ω2 ⊕ Ωmix dA = (dA)vert + (dA)mix + (dA)horiz ∈ Ωvert horiz ⊗ g

168

A. Cannas da Silva

E XAMPLE . We describe the Atiyah–Bott construction for the case of a circle bundle S1

→ P ↓π B

Let v be the generator of the S 1 -action on P , corresponding to the basis 1 of g R. A connection form on P is an ordinary 1-form A ∈ Ω 1 (P ) such that Lv A = 0 and ıv A = 1. If we fix one particular connection A0 , then any other connection is of the form A = A0 + a 1 for some a ∈ a = (Ωhoriz (P ))G = Ω 1 (B). The symplectic form on a = Ω 1 (B) is simply ω : a × a −→ Ω 2 (B) −→ R, a ∧ b. (a, b) −→ a ∧ b −→ B

The gauge group is G = Maps(B, S 1 ), because a gauge transformation is multiplication by some element of S 1 over each point in B encoded in a map h : B → S 1 . The action φ : G → Diff(P ) takes h ∈ G to the diffeomorphism φh : p −→ h π(p) · p. The Lie algebra of G is Lie G = Maps(B, R) = C ∞ (B) with dual (Lie G)∗ = Ω 2 (B), where the (smooth) duality is provided by integration C ∞ (B) × Ω 2 (B) → R, (h, β) → B hβ. The gauge group acts on the space of all connections by ψ : G −→ Diff(A), h : x → eiθ(x) −→ (ψh : A → A − π ∗ dθ ). (In the case where P = S 1 × B is a trivial bundle, every connection can be written A = dt + β, with β ∈ Ω 1 (B). A gauge transformation h ∈ G acts on P by φh : (t, x) → (t + θ (x), x) and on A by A → φh∗−1 (A).) The infinitesimal action is dψ : Lie G −→ χ(A), X −→ X # = vector field described by (A → A − dX), so that X # = −dX. It remains to check that µ : A −→ (Lie G)∗ = Ω 2 (B), A −→ curv A

i A ∧ A ⊗ X , where the satisfying (dA)mix = 0 and (dA)vert (X, Y ) = [X, Y ], i.e., (dA)vert = 12 i,,m cm m i i ’s are the structure constants of the Lie algebra with respect to the chosen basis, and defined by [X , X ] = cm m i i,,m cm Xi . So the relevance of dA may come only from its horizontal component, called the curvature form 2 of the connection A, and denoted curv A = (dA)horiz ∈ Ωhoriz ⊗ g. A connection is called flat if its curvature is zero.

Symplectic geometry

169

is indeed a moment map for the action of the gauge group on A. Since in this case curv A = 2 dA ∈ (Ωhoriz (P ))G = Ω 2 (B), the action of G on Ω 2 (B) is trivial and µ is G-invariant, the ∞ X equivariance

condition is satisfied. Take any X ∈ Lie G = C (B). Since the map µ : A → X, dA = B X · dA is linear in A, its differential is dµX : a −→ R, a −→ X da. B

By definition of ω and the Stokes theorem, we have that ω X# , a =

X ·a =−

B

dX · a =

#

B

X · da = dµX (a),

∀a ∈ Ω 1 (B),

B

so we are done in proving that µ is the moment map. The function µ2 : A → R giving the square of the L2 norm of the curvature is the Yang–Mills functional, whose Euler–Lagrange equations are the Yang–Mills equations. Atiyah and Bott [6] studied the topology of A by regarding µ2 as an equivariant Morse function. In general, it is a good idea to apply Morse theory to the norm square of a moment map [80].

6.4. Symplectic toric manifolds Toric manifolds are smooth toric varieties.73 When studying the symplectic features of these spaces, we refer to them as symplectic toric manifolds. Relations between the algebraic and symplectic viewpoints on toric manifolds are discussed in [21]. D EFINITION 6.6. A symplectic toric manifold is a compact connected symplectic manifold (M, ω) equipped with an effective Hamiltonian action of a torus T of dimension equal to half the dimension of the manifold, dim T = 12 dim M, and with a choice of a corresponding moment map µ. Two symplectic toric manifolds, (Mi , ωi , Ti , µi ), i = 1, 2, are equivalent if there exists an isomorphism λ : T1 → T2 and a λ-equivariant symplectomorphism ϕ : M1 → M2 such that µ1 = µ2 ◦ ϕ. E XAMPLES . 1. The circle S 1 acts on the 2-sphere (S 2 , ωstandard = dθ ∧ dh) by rotations, eiν · (θ, h) = (θ + ν, h). with moment map µ = h equal to the height function and moment polytope [−1, 1] (see Figure 3). 73 Toric varieties were introduced by Demazure in [29]. There are many nice surveys of the theory of toric varieties in algebraic geometry; see, for instance, [27,53,79,107]. Toric geometry has recently become an important tool in physics in connection with mirror symmetry [26].

170

A. Cannas da Silva

µ=h

1

−1

Fig. 3.

Analogously, S 1 acts on the Riemann sphere CP1 with the Fubini–Study form ωFS = 14 ωstandard , by eiθ · [z0 , z1 ] = [z0 , eiθ z1 ]. This is Hamiltonian with moment map µ[z0 , z1 ] = − 12 ·

|z1 |2 , |z0 |2 +|z1 |2

and moment polytope [− 12 , 0].

2. For the Tn -action on the product of n Riemann spheres CP1 × · · · × CP1 by iθ e 1 , . . . , eiθn · [z1 , w1 ], . . . , [zn , wn ] = z1 , eiθ1 w1 , . . . , w0 , eiθn w1 , the moment polytope is an n-dimensional cube. 3. Let (CP2 , ωFS ) be 2-(complex-)dimensional complex projective space equipped with the Fubini–Study form defined in Section 3.4. The T2 -action on CP2 by (eiθ1 , eiθ2 ) · [z0 , z1 , z2 ] = [z0 , eiθ1 z1 , eiθ2 z2 ] has moment map µ[z0 , z1 , z2 ] = −

|z1 |2 1 |z2 |2 . , 2 |z0 |2 + |z1 |2 + |z2 |2 |z0 |2 + |z1 |2 + |z2 |2

The image is the isosceles triangle with vertices (0, 0), (− 12 , 0) and (0, − 12 ). 4. For the Tn -action on (CPn , ωFS ) by iθ e 1 , . . . , eiθn · [z0 , z1 , . . . , zn ] = z0 , eiθ1 z1 , . . . , eiθn zn the moment polytope is an n-dimensional simplex. Since the coordinates of the moment map are commuting integrals of motion, a symplectic toric manifold gives rise to a completely integrable system. By Proposition 5.24, symplectic toric manifolds are optimal Hamiltonian torus-spaces. By Theorem 5.21, they have an associated polytope. It turns out that the moment polytope contains enough information to sort all symplectic toric manifolds. We now define the class of polytopes that arise in the classification. For a symplectic toric manifold the weights λ(1) , . . . , λ(n) in Theorem 5.22 form a Z-basis of Zm , hence the moment polytope is a Delzant polytope: D EFINITION 6.7. A Delzant polytope in Rn is a polytope satisfying: • simplicity, i.e., there are n edges meeting at each vertex; • rationality, i.e., the edges meeting at the vertex p are rational in the sense that each edge is of the form p + tui , t 0, where ui ∈ Zn ;

Symplectic geometry

171

Fig. 4.

• smoothness, i.e., for each vertex, the corresponding u1 , . . . , un can be chosen to be a Z-basis of Zn . In R2 the simplicity condition is always satisfied (by nondegenerate polytopes). In R3 , for instance, a square pyramid fails the simplicity condition. E XAMPLES . Figure 4 represents Delzant polytopes in R2 . The following theorem classifies (equivalence classes of) symplectic toric manifolds in terms of the combinatorial data encoded by a Delzant polytope. T HEOREM 6.8 (Delzant [28]). Toric manifolds are classified by Delzant polytopes, and their bijective correspondence is given by the moment map: {toric manifolds} ←→ {Delzant polytopes}, 2n M , ω, Tn , µ −→ µ(M). Delzant’s construction (Section 6.5) shows that for a toric manifold the moment map takes the fixed points bijectively to the vertices of the moment polytope and takes points with a k-dimensional stabilizer to the codimension k faces of the polytope. The moment polytope is exactly the orbit space, i.e., the preimage under µ of each point in the polytope is exactly one orbit. For instance, consider (S 2 , ω = dθ ∧ dh, S 1 , µ = h), where S 1 acts by rotation. The image of µ is the line segment I = [−1, 1]. The product S 1 × I is an open-ended cylinder. We can recover the 2-sphere by collapsing each end of the cylinder to a point. Similarly, we can build CP2 from T2 × where is a rectangular isosceles triangle, and so on. E XAMPLES . 1. By a linear transformation in SL(2; Z), we can make one of the angles in a Delzant triangle into a right angle. Out of the rectangular triangles, only the isosceles one satisfies the smoothness condition. Therefore, up to translation, change of scale and the action of SL(2; Z), there is just one 2-dimensional Delzant polytope with three vertices, namely an isosceles triangle. We conclude that the projective space CP2 is the only 4-dimensional toric manifold with three fixed points, up to choices of a constant in the moment map, of a multiple of ωFS and of a lattice basis in the Lie algebra of T2 . 2. Up to translation, change of scale and the action of SL(n; Z), the standard n-simplex in Rn (spanned by the origin and the standard basis vectors (1, 0, . . . , 0), . . . ,

172

A. Cannas da Silva

p Fig. 5.

(0, . . . , 0, 1)) is the only n-dimensional Delzant polytope with n + 1 vertices. Hence, M = CPn is the only 2n-dimensional toric manifold with n + 1 fixed points, up to choices of a constant in the moment map, of a multiple of ωFS and of a lattice basis in the Lie algebra of TN . 3. A transformation in SL(2; Z) makes one of the angles in a Delzant quadrilateral into a right angle. Automatically an adjacent angle also becomes 90◦ . Smoothness imposes that the slope of the skew side be integral. Thus, up to translation, change of scale and SL(2; Z)-action, the 2-dimensional Delzant polytopes with four vertices are trapezoids with vertices (0, 0), (0, 1), (, 1) and ( + n, 0), for n a nonnegative integer and > 0. Under Delzant’s construction (that is, under symplectic reduction of C4 with respect to an action of (S 1 )2 ), these correspond to the so-called Hirzebruch surfaces—the only 4-dimensional symplectic toric manifolds that have four fixed points up to equivalence as before. Topologically, they are S 2 -bundles over S 2 , either the trivial bundle S 2 × S 2 when n is even or the nontrivial bundle (given by the blowup of CP2 at a point; see Section 4.3) when n is odd. Let be an n-dimensional Delzant polytope, and let (M , ω , Tn , µ ) be the associated symplectic toric manifold. The ε-blow-up of (M , ω ) at a fixed point of the Tn action is a new symplectic toric manifold (Sections 4.3 and 5.6). Let q be a fixed point of the Tn -action on (M , ω ), and let p = µ (q) be the corresponding vertex of . Let u1 , . . . , un be the primitive (inward-pointing) edge vectors at p, so that the rays p + tui , t 0, form the edges of at p. P ROPOSITION 6.9. The ε-blow-up of (M , ω ) at a fixed point q is the symplectic toric manifold associated to the polytope ε obtained from by replacing the vertex p by the n vertices p + εui , i = 1, . . . , n. In other words, the moment polytope for the blow-up of (M , ω ) at q is obtained from by chopping off the corner corresponding to q, thus substituting the original set of vertices by the same set with the vertex corresponding to q replaced by exactly n new vertices. The truncated polytope is Delzant. We may view the ε-blow-up of (M , ω ) as being obtained from M by smoothly replacing q by (CPn−1 , εωFS ) (whose moment polytope is an (n − 1)-dimensional simplex). (See Figure 5.) E XAMPLE . The moment polytope for the standard T2 -action on (CP2 , ωFS ) is a right isosceles triangle . If we blow up CP2 at [0 : 0 : 1] we obtain a symplectic toric manifold associated to the trapezoid below: a Hirzebruch surface (see Figure 6).

Symplectic geometry

β

173

Fig. 6.

M

µ

R

projection

Fig. 7.

Let (M, ω, Tn , µ) be a 2n-dimensional symplectic toric manifold. Choose a suitably generic direction in Rn by picking a vector X whose components are independent over Q. This condition ensures that: • the one-dimensional subgroup TX generated by the vector X is dense in Tn , • X is not parallel to the facets of the moment polytope := µ(M), and • the vertices of have different projections along X. Then the fixed points for the Tn -action are exactly the fixed points of the action restricted to TX , that is, are the zeros of the vector field, X # on M generated by X. The projection of µ along X, µX := µ, X : M → R, is a Hamiltonian function for the vector field X # generated by X. We conclude that the critical points of µX are precisely the fixed points of the Tn -action (see Figure 7). By Theorem 5.22, if q is a fixed point for the Tn -action, then there exists a chart (U, x1 , . . . , xn , y1 , . . . , yn ) centered at q and weights λ(1) , . . . , λ(n) ∈ Zn such that n 1 (k) µ |U = µ, X|U = µ (q) − λ , X xk2 + yk2 . 2 X

X

k=1

Since the components of X are independent over Q, all coefficients λ(k) , X are nonzero, so q is a nondegenerate critical point of µX . Moreover, the index74 of q is twice the number of labels k such that −λ(k) , X < 0. But the −λ(k) ’s are precisely the edge vectors ui which satisfy Delzant’s conditions. Therefore, geometrically, the index of q can be read from the moment polytope , by taking twice the number of edges whose inward-pointing 74 A Morse function on an m-dimensional manifold M is a smooth function f : M → R all of whose critical

points (where df vanishes) are nondegenerate (i.e., the Hessian matrix is nonsingular). Let q be a nondegenerate critical point for f : M → R. The index of f at q is the index of the Hessian Hq : Rm × Rm → R regarded as a symmetric bilinear function, that is, the maximal dimension of a subspace of R where H is negative definite.

174

A. Cannas da Silva

edge vectors at µ(q) point up relative to X, that is, whose inner product with X is positive. In particular, µX is a perfect Morse function75 and we have P ROPOSITION 6.10. Let X ∈ Rn have components independent over Q. The degree-2k homology group of the symplectic toric manifold (M, ω, T, µ) has dimension equal to the number of vertices of the moment polytope where there are exactly k (primitive inwardpointing) edge vectors that point up relative to the projection along the X. All odd-degree homology groups of M are zero. By Poincaré duality (or by taking −X instead of X), the words point up may be replaced by point down. The Euler characteristic of a symplectic toric manifold is simply the number of vertices of the corresponding polytope. There is a combinatorial way of understanding the cohomology ring [53]. A symplectic toric orbifold is a compact connected symplectic orbifold (M, ω) equipped with an effective Hamiltonian action of a torus of dimension equal to half the dimension of the orbifold, and with a choice of a corresponding moment map. Symplectic toric orbifolds were classified by Lerman and Tolman [87] in a theorem that generalizes Delzant’s: a symplectic toric orbifold is determined by its moment polytope plus a positive integer label attached to each of the polytope facets. The polytopes that occur are more general than the Delzant polytopes in the sense that only simplicity and rationality are required; the edge vectors u1 , . . . , un need only form a rational basis of Zn . When the integer labels are all equal to 1, the failure of the polytope smoothness accounts for all orbifold singularities.

6.5. Delzant’s construction Following [28,66], we prove the existence part (or surjectivity) in Delzant’s theorem, by using symplectic reduction to associate to an n-dimensional Delzant polytope a symplectic toric manifold (M , ω , Tn , µ ). Let be a Delzant polytope in (Rn )∗76 and with d facets.77 We can algebraically describe as an intersection of d halfspaces. Let vi ∈ Zn , i = 1, . . . , d, be the primitive78 outward-pointing normal vectors to the facets of . Then, for some λi ∈ R, we can write = {x ∈ (Rn )∗ | x, vi λi , i = 1, . . . , d}. 75 A perfect Morse function is a Morse function f for which the Morse inequalities [103,104] are equalities, i.e., bλ (M) = Cλ and bλ (M) − bλ−1 (M) + · · · ± b0 (M) = Cλ − Cλ−1 + · · · ± C0 where bλ (M) = dim Hλ (M) and Cλ be the number of critical points of f with index λ. If all critical points of a Morse function f have even index, then f is a perfect Morse function. 76 Although we identify Rn with its dual via the Euclidean inner product, it may be more clear to see in (Rn )∗ for Delzant’s construction. 77 A face of a polytope is a set of the form F = P ∩ {x ∈ Rn | f (x) = c} where c ∈ R and f ∈ (Rn )∗ satisfies f (x) c, ∀x ∈ P . A facet of an n-dimensional polytope is an (n − 1)-dimensional face. 78 A lattice vector v ∈ Zn is primitive if it cannot be written as v = ku with u ∈ Zn , k ∈ Z and |k| > 1; for instance, (1, 1), (4, 3), (1, 0) are primitive, but (2, 2), (3, 6) are not.

Symplectic geometry

v2

175

(0, 1) v3 (0, 0) (1, 0) v1 Fig. 8.

E XAMPLE . When is the triangle shown in Figure 8, we have ∗ = x ∈ R2 x, (−1, 0) 0, x, (0, −1) 0, x, (1, 1) 1 . For the standard basis e1 = (1, 0, . . . , 0), . . . , ed = (0, . . . , 0, 1) of Rd , consider π : Rd −→ Rn , ei −→ vi . L EMMA 6.11. The map π is onto and maps Zd onto Zn . P ROOF. We need to show that the set {v1 , . . . , vd } spans Zn . At a vertex p, the edge vectors u1 , . . . , un ∈ (Rn )∗ form a basis for (Zn )∗ which, by a change of basis if necessary, we may assume is the standard basis. Then the corresponding primitive normal vectors to the facets meeting at p are −u1 , . . . , −un . π

We still call π the induced surjective map Td = Rd /(2πZd ) → Tn = Rn /(2πZn ). The kernel N of π is a (d − n)-dimensional Lie subgroup of Td with inclusion i : N → Td . Let n be the Lie algebra of N . The exact sequence of tori i

π

1 −→ N −→ Td −→ Tn −→ 1 induces an exact sequence of Lie algebras i

π

0 −→ n −→ Rd −→ Rn −→ 0 with dual exact sequence ∗ i ∗ π∗ 0 −→ (Rn )∗ −→ Rd −→ n∗ −→ 0.

176

A. Cannas da Silva

dzk ∧ d z¯ k , and standard Hamiltonian acConsider Cd with symplectic form ω0 = 2i d it it d 1 tion of T given by (e , . . . , e ) · (z1 , . . . , zd ) = (eit1 z1 , . . . , eitd zd ). A moment map is φ : Cd → (Rd )∗ defined by φ(z1 , . . . , zd ) = −

1 2 |z1 | , . . . , |zd |2 + (λ1 , . . . , λd ), 2

where the constant is chosen for later convenience. The subtorus N acts on Cd in a Hamiltonian way with moment map i ∗ ◦ φ : Cd → n∗ . Let Z = (i ∗ ◦ φ)−1 (0). In order to show that Z (a closed set) is compact it suffices (by the Heine–Borel theorem) to show that Z is bounded. Let be the image of by π ∗ . First we show that φ(Z) = . A value y ∈ (Rd )∗ is in the image of Z by φ if and only if (a)

y is in the image of φ

and (b)

i∗y = 0

if and only if (using the expression for φ and the third exact sequence) (a)

y, ei λi

(b)

y = π ∗ (x)

for i = 1, . . . , d

and

for some x ∈ (Rn )∗ .

Suppose that y = π ∗ (x). Then y, ei λi ,

∀i

⇐⇒ ⇐⇒

x, π(ei ) λi ,

x, vi λi ,

∀i

∀i ⇐⇒

x ∈ .

Thus, y ∈ φ(Z) ⇔ y ∈ π ∗ () = . Since is compact, φ is proper and φ(Z) = , we conclude that Z must be bounded, and hence compact. In order to show that N acts freely on Z, pick a vertex p of , and let I = {i1 , . . . , in } be the set of indices for the n facets meeting at p. Pick z ∈ Z such that φ(z) = π ∗ (p). Then p is characterized by n equations p, vi = λi where i ∈ I : p, vi = λi ⇐⇒ p, π(ei ) = λi ⇐⇒ π ∗ (p), ei = λi ⇐⇒ φ(z), ei = λi ⇐⇒

ith coordinate of φ(z) is equal to λi

⇐⇒

1 − |zi |2 + λi = λi 2 zi = 0.

⇐⇒

Hence, those z’s are points whose coordinates in the set I are zero, and whose other coordinates are nonzero. Without loss of generality, we may assume that I = {1, . . . , n}. The stabilizer of z is d T z = (t1 , . . . , tn , 1, . . . , 1) ∈ Td .

Symplectic geometry

177

As the restriction π : (Rd )z → Rn maps the vectors e1 , . . . , en to a Z-basis v1 , . . . , vn of Zn (respectively), at the level of groups π : (Td )z → Tn must be bijective. Since N = ker(π : Td → Tn ), we conclude that N ∩ (Td )z = {e}, i.e., Nz = {e}. Hence all N -stabilizers at points mapping to vertices are trivial. But this was the worst case, since other stabilizers Nz (z ∈ Z) are contained in stabilizers for points z that map to vertices. We conclude that N acts freely on Z. We now apply reduction. Since i ∗ is surjective, 0 ∈ n∗ is a regular value of i ∗ ◦ φ. Hence, Z is a compact submanifold of Cd of (real) dimension 2d − (d − n) = d + n. The orbit space M = Z/N is a compact manifold of (real) dimension dim Z − dim N = (d + n) − (d − n) = 2n. The point-orbit map p : Z → M is a principal N -bundle over M . Consider the diagram Z p↓ M

j

→

Cd

where j : Z → Cd is inclusion. The Marsden–Weinstein–Meyer theorem (Theorem 6.1) guarantees the existence of a symplectic form ω on M satisfying p ∗ ω = j ∗ ω0 . Since Z is connected, the symplectic manifold (M , ω ) is also connected. It remains to show that (M , ω ) is a Hamiltonian Tn -space with a moment map µ having image µ (M ) = . Let z be such that φ(z) = π ∗ (p) where p is a vertex of . Let σ : Tn → (Td )z be the inverse for the earlier bijection π : (Td )z → Tn . This is a section, i.e., a right inverse for π , in the sequence 1 −→ N

i

−→

Td

π

−→ Tn σ ←−

−→

1,

so it splits, i.e., becomes like a sequence for a product, as we obtain an isomorphism (i, σ ) : N × Tn −→ Td . The action of the Tn factor (or, more rigorously, σ (Tn ) ⊂ Td ) descends to the quotient M = Z/N . Consider the diagram ∗ j φ σ∗ Z → Cd −→ Rd η∗ ⊕ (Rn )∗ −→ (Rn )∗ p↓ M where the last horizontal map is projection onto the second factor. Since the composition of the horizontal maps is constant along N -orbits, it descends to a map µ : M −→ (Rn )∗

178

A. Cannas da Silva

which satisfies µ ◦ p = σ ∗ ◦ φ ◦ j . By reduction for product groups (Section 6.2), this is a moment map for the action of Tn on (M , ω ). The image of µ is µ (M ) = (µ ◦ p)(Z) = (σ ∗ ◦ φ ◦ j )(Z) = (σ ∗ ◦ π ∗ )() = , because φ(Z) = π ∗ () and π ◦ σ = id. We conclude that (M , ω , Tn , µ ) is the required toric manifold corresponding to . This construction via reduction also shows that symplectic toric manifolds are in fact Kähler. E XAMPLE . Here are the details of Delzant’s construction for the case of a segment = [0, a] ⊂ R∗ (n = 1, d = 2). Let v(= 1) be the standard basis vector in R. Then is described by x, −v 0 and x, v a, where v1 = −v, v2 = v, λ1 = 0 and λ2 = a. π The projection R2 −→ R, e1 → −v, e2 → v, has kernel equal to the span of (e1 + e2 ), so that N is the diagonal subgroup of T2 = S 1 × S 1 . The exact sequences become i

π

1 −→ N −→ T2 −→ t −→ (t, t),

S1

−→ 1,

(t1 , t2 ) −→ t1−1 t2 , i

π

R −→ 0, 0 −→ n −→ R2 −→ x −→ (x, x), (x1 , x2 ) −→ x2 − x1 , ∗ i ∗ π∗ 0 −→ R∗ −→ R2 −→ n∗ −→ 0, x −→ (−x, x), (x1 , x2 ) −→ x1 + x2 . The action of the diagonal subgroup N = {(eit , eit ) ∈ S 1 × S 1 } on C2 by it it e , e · (z1 , z2 ) = eit z1 , eit z2 has moment map (i ∗ ◦ φ)(z1 , z2 ) = − 12 (|z1 |2 + |z2 |2 ) + a, with zero-level set (i ∗ ◦ φ)−1 (0) = (z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 2a . Hence, the reduced space is a projective space, (i ∗ ◦ φ)−1 (0)/N = CP1 . 6.6. Duistermaat–Heckman theorems Throughout this subsection, let (M, ω, G, µ) be a Hamiltonian G-space, where G is an n-torus79 and the moment map µ is proper. 79 The discussion in this subsection may be extended to Hamiltonian actions of other compact Lie groups, not necessarily tori; see [66, Exercises 2.1–2.10].

Symplectic geometry

179

If G acts freely on µ−1 (0), it also acts freely on nearby levels µ−1 (t), t ∈ g∗ and t ≈ 0. (Otherwise, assume only that 0 is a regular value of µ and work with orbifolds.) We study the variation of the reduced spaces by relating Mred = µ−1 (0)/G, ωred and Mt = µ−1 (t)/G, ωt . For simplicity, assume G to be the circle S 1 . Let Z = µ−1 (0) and let i : Z → M be the inclusion map. Fix a connection form α ∈ Ω 1 (Z) for the principal bundle S1

→ Z ↓π Mred

that is, LX# α = 0 and ıX# α = 1, where X # is the infinitesimal generator for the S 1 -action. Construct a 2-form on the product manifold Z × (−ε, ε) by the recipe σ = π ∗ ωred − d(xα), where x is a linear coordinate on the interval (−ε, ε) ⊂ R g∗ . (By abuse of notation, we shorten the symbols for forms on Z × (−ε, ε) that arise by pullback via projection onto each factor.) L EMMA 6.12. The 2-form σ is symplectic for ε small enough. ∂ P ROOF. At points where x = 0, the form σ |x=0 = π ∗ ωred + α ∧ dx satisfies σ |x=0 (X # , ∂x ) = 1, so σ is nondegenerate along Z × {0}. Since nondegeneracy is an open condition, we conclude that σ is nondegenerate for x in a sufficiently small neighborhood of 0. Closedness is clear.

Notice that σ is invariant with respect to the S 1 -action on the first factor of Z × (−ε, ε). This action is Hamiltonian with moment map x : Z × (−ε, ε) → (−ε, ε) given by projection onto the second factor (since LX# α = 0 and ıX# α = 1): ıX# σ = −ıX# d(xα) = −LX# (xα) + dıX# (xα) = dx. L EMMA 6.13. There exists an equivariant symplectomorphism between a neighborhood of Z in M and a neighborhood of Z × {0} in Z × (−ε, ε), intertwining the two moment maps, for ε small enough. P ROOF. The inclusion i0 : Z → Z × (−ε, ε) as Z × {0} and the natural inclusion i : Z → M are S 1 -equivariant coisotropic embeddings. Moreover, they satisfy i0∗ σ = i ∗ ω since both sides are equal to π ∗ ωred , and the moment maps coincide on Z because i0∗ x = 0 = i ∗ µ. Replacing ε by a smaller positive number if necessary, the result follows from the equivariant version of the coisotropic embedding theorem (Theorem 2.9).80 80 Equivariant coisotropic embedding theorem: Let (M , ω ), (M , ω ) be symplectic manifolds of dimen0 0 1 1 sion 2n, G a compact Lie group acting on (Mi , ωi ), i = 0, 1, in a Hamiltonian way with moment maps µ0

180

A. Cannas da Silva

Therefore, in order to compare the reduced spaces Mt = µ−1 (t)/S 1 for t ≈ 0, we can work in Z × (−ε, ε) and compare instead the reduced spaces x −1 (t)/S 1 . P ROPOSITION 6.14. The space (Mt , ωt ) is symplectomorphic to (Mred , ωred − tβ) where β is the curvature form of the connection α. P ROOF. By Lemma 6.13, (Mt , ωt ) is symplectomorphic to the reduced space at level t for the Hamiltonian space (Z × (−ε, ε), σ, S 1 , x). Since x −1 (t) = Z × {t}, where S 1 acts on the first factor, all the manifolds x −1 (t)/S 1 are diffeomorphic to Z/S 1 = Mred . As for the symplectic forms, let ιt : Z × {t} → Z × (−ε, ε) be the inclusion map. The restriction of σ to Z × {t} is ι∗t σ = π ∗ ωred − t dα. By definition of curvature, dα = π ∗ β. Hence, the reduced symplectic form on x −1 (t)/S 1 is ωred − tβ. In loose terms, Proposition 6.14 says that the reduced forms ωt vary linearly in t, for t close enough to 0. However, the identification of Mt with Mred as abstract manifolds is not natural. Nonetheless, any two such identifications are isotopic. By the homotopy invariance of de Rham classes, we obtain: T HEOREM 6.15 (Duistermaat–Heckman [38]). Under the hypotheses and notation before, the cohomology class of the reduced symplectic form [ωt ] varies linearly in t. More specif2 (Mred ) is the first Chern class81 of the S 1 -bundle Z → Mred , ically, if c = [−β] ∈ HdeRham we have [ωt ] = [ωred ] + tc. and µ1 , respectively, Z a manifold of dimension k n with a G-action, and ιi : Z → Mi , i = 0, 1, G-equivariant coisotropic embeddings. Suppose that ι∗0 ω0 = ι∗1 ω1 and ι∗0 µ0 = ι∗1 µ1 . Then there exist G-invariant neighborhoods U0 and U1 of ι0 (Z) and ι1 (Z) in M0 and M1 , respectively, and a G-equivariant symplectomorphism ϕ : U0 → U1 such that ϕ ◦ ι0 = ι1 and µ0 = ϕ ∗ µ1 . 81 Often the Lie algebra of S 1 is identified with 2π iR under the exponential map exp : g 2π iR → S 1 , ξ → eξ . Given a principal S 1 -bundle, by this identification the infinitesimal action maps the generator 2π i of 2π iR to the generating vector field X# . A connection form A is then an imaginary-valued 1-form on the total space satisfying LX# A = 0 and ıX# A = 2π i. Its curvature form B is an imaginary-valued 2-form on the base satisfying i B]. π ∗ B = dA. By the Chern–Weil isomorphism, the first Chern class of the principal S 1 -bundle is c = [ 2π 1 Here we identify the Lie algebra of S with R and implicitly use the exponential map exp : g R → S 1 , t → e2π it . Hence, given a principal S 1 -bundle, the infinitesimal action maps the generator 1 of R to X# , and here a connection form α is an ordinary 1-form on the total space satisfying LX# α = 0 and ıX# α = 1. The curvature form β is an ordinary 2-form on the base satisfying π ∗ β = dα. Consequently, we have A = 2π iα, B = 2π iβ and the first Chern class is given by c = [−β].

Symplectic geometry

181

D EFINITION 6.16. The Duistermaat–Heckman measure, mDH , on g∗ is the push-forward of the Liouville measure82 by µ : M → g∗ , that is, for any Borel subset U of g∗ , we have mDH (U) =

ωn . µ−1 (U ) n!

The integral with respect to the Duistermaat–Heckman measure of a compactlysupported function h ∈ C ∞ (g∗ ) is g∗

h dmDH :=

(h ◦ µ) M

ωn . n!

On g∗ regarded as a vector space, say Rn , there is also the Lebesgue (or Euclidean) measure, m0 . The relation between mDH and m0 is governed by the Radon–Nikodym derivative, DH denoted by dm dm0 , which is a generalized function satisfying

g∗

h dmDH =

g∗

h

dmDH dm0 . dm0

T HEOREM 6.17 (Duistermaat–Heckman [38]). Under the hypotheses and notation before, the Duistermaat–Heckman measure is a piecewise polynomial multiple of Lebesgue DH measure on g∗ Rn , that is, the Radon–Nikodym derivative f = dm polydm0 is piecewise

∗ nomial. More specifically, for any Borel subset U of g , we have mDH (U) = U f (x) dx, where dx = dm0 is the Lebesgue volume form on U and f : g∗ Rn → R is polynomial on any region consisting of regular values of µ. This Radon–Nikodym derivative f is called the Duistermaat–Heckman polynomial. In the case of a toric manifold, the Duistermaat–Heckman polynomial is a universal constant equal to (2π)n when is n-dimensional. Thus the symplectic volume of (M , ω ) is (2π)n times the Euclidean volume of . E XAMPLE . For the standard spinning of a sphere (S 2 , ω = dθ ∧ dh, S 1 , µ = h), the image of µ is the interval [−1, 1]. The Lebesgue measure of [a, b] ⊆ [−1, 1] is m0 ([a, b]) = b − a. The Duistermaat–Heckman measure of [a, b] is mDH [a, b] =

{(θ,h)∈S 2 | ahb}

dθ dh = 2π(b − a),

82 On an arbitrary symplectic manifold (M 2n , ω), with symplectic volume ωn , the Liouville measure (or symn!

plectic measure) of a Borel subset U of M is ωn . mω (U) = U n!

The set B of Borel subsets is the σ -ring generated by the set of compact subsets, i.e., if A, B ∈ B, then A \ B ∈ B, $ and if Ai ∈ B, i = 1, 2, . . . , then ∞ i=1 Ai ∈ B.

182

A. Cannas da Silva

i.e., mDH = 2πm0 . Consequently, the area of the spherical region between two parallel planes depends only on the distance between the planes, a result that was known to Archimedes around 230 BC. P ROOF. We sketch the proof of Theorem 6.17 for the case G = S 1 . The proof for the general case, which follows along similar lines, can be found in, for instance, [66], besides the original articles. Let (M, ω, S 1 , µ) be a Hamiltonian S 1 -space of dimension 2n and let (Mx , ωx ) be its reduced space at level x. Proposition 6.14 or Theorem 6.15 imply that, for x in a sufficiently narrow neighborhood of 0, the symplectic volume of Mx , vol(Mx ) = Mx

ωxn−1 = (n − 1)!

Mred

(ωred − xβ)n−1 , (n − 1)!

is a polynomial in x of degree n − 1. This volume can be also expressed as vol(Mx ) = Z

π ∗ (ωred − xβ)n−1 ∧ α, (n − 1)!

where α is a connection form for the S 1 -bundle Z → Mred and β is its curvature form. Now we go back to the computation of the Duistermaat–Heckman measure. For a Borel subset U of (−ε, ε), the Duistermaat–Heckman measure is, by definition, mDH (U) =

ωn . µ−1 (U ) n!

Using the fact that (µ−1 (−ε, ε), ω) is symplectomorphic to (Z × (−ε, ε), σ ) and, moreover, they are isomorphic as Hamiltonian S 1 -spaces, we obtain mDH (U) =

σn . Z×U n!

Since σ = π ∗ ωred − d(xα), its power is σ n = n(π ∗ ωred − x dα)n−1 ∧ α ∧ dx. By the Fubini theorem, we then have mDH (U) =

U

Z

π ∗ (ωred − xβ)n−1 ∧ α ∧ dx. (n − 1)!

Therefore, the Radon–Nikodym derivative of mDH with respect to the Lebesgue measure, dx, is f (x) = Z

π ∗ (ωred − xβ)n−1 ∧ α = vol(Mx ). (n − 1)!

Symplectic geometry

183

The previous discussion proves that, for x ≈ 0, f (x) is a polynomial in x. The same holds for a neighborhood of any other regular value of µ, because we may change the moment map µ by an arbitrary additive constant. Duistermaat and Heckman [38] also applied these results when M is compact to provide

X n a formula for the oscillatory integral M eiµ ωn! for X ∈ g as a sum of contributions of the fixed points of the action of the one-parameter subgroup generated by X. They hence showed that the stationary phase approximation83 is exact in the case of the moment map. When G is a maximal torus of a compact connected simple Lie group acting on a coadjoint orbit, the Duistermaat–Heckman formula reduces to the Harish–Chandra formula. It was observed by Berline and Vergne [14] and by Atiyah and Bott [5] that the Duistermaat– Heckman formula can be derived by localization in equivariant cohomology. This is an instance of Abelian localization, i.e., a formula for an integral (in equivariant cohomology) in terms of data at the fixed points of the action, and typically is used for the case of Abelian groups (or of maximal tori). Later non-Abelian localization formulas were found, where integrals (in equivariant cohomology) are expressed in terms of data at the zeros of the moment map, normally used for the case of non-Abelian groups. Both localizations gave rise to computations of the cohomology ring structure of reduced spaces [80].

References [1] V. Arnold, On a characteristic class entering into conditions of quantization, Funkcional Anal. i Priložen 1 (1967), 1–14. [2] V. Arnold, Mathematical Methods of Classical Mechanics, Graduate Texts in Math., vol. 60, SpringerVerlag, New York (1978). [3] V. Arnold and A. Givental, Symplectic geometry, Dynamical Systems IV, Symplectic Geometry and Its Applications, V. Arnold and S. Novikov, eds, Encyclopaedia of Math. Sciences, vol. 4, Springer-Verlag, Berlin (1990). [4] M. Atiyah, Convexity and commuting Hamiltonians, Bull. London Math. Soc. 14 (1982), 1–15. [5] M. Atiyah and R. Bott, The moment map and equivariant cohomology, Topology 23 (1984), 1–28. [6] M. Atiyah and R. Bott, The Yang–Mills equations over Riemann surfaces, Topology 23 (1984), 1–28; Philos. Trans. Roy. Soc. London 308 (1983), 523–615. [7] M. Audin, Exemples de variétés presque complexes, Enseign. Math. 37 (1991), 175–190. [8] M. Audin, Spinning Tops, A Course on Integrable Systems, Cambridge Studies in Advanced Mathematics, vol. 51, Cambridge Univ. Press, Cambridge (1996). [9] M. Audin, Lagrangian submanifolds, Symplectic Geometry of Integrable Hamiltonian Systems (Barcelona, 2001), Adv. Courses Math. CRM Barcelona, Birkhäuser, Basel (2003), 1–83. [10] M. Audin, Torus Actions on Symplectic Manifolds, Progress in Mathematics, vol. 93, Birkhäuser, Basel (2004). [11] M. Audin and J. Lafontaine, eds, Holomorphic Curves in Symplectic Geometry, Progress in Mathematics, vol. 117, Birkhäuser, Basel (1994). [12] D. Auroux, Asymptotically holomorphic families of symplectic submanifolds, Geom. Funct. Anal. 7 (1997), 971–995. [13] D. Auroux, Symplectic 4-manifolds as branched coverings of CP2 , Invent. Math. 139 (2000), 551–602. [14] N. Berline and M. Vergne, Zéros d’un champ de vecteurs et classes caractéristiques équivariantes, Duke Math. J. 50 (1983), 539–549.

83 The stationary phase lemma gives the asymptotic behavior (for large N ) of integrals ( N )n ig M f e vol, where 2π f and g are real functions and vol is a volume form on a 2n-dimensional manifold M.

184

A. Cannas da Silva

[15] P. Biran, A stability property of symplectic packing, Invent. Math. 136 (1999), 123–155. [16] P. Biran, Geometry of symplectic intersections, Proceedings of the I.C.M., vol. II (Beijing, 2002), Higher Ed. Press, Beijing (2002), 241–255. [17] P. Biran and K. Cieliebak, Symplectic topology on subcritical manifolds, Comment. Math. Helv. 76 (2001), 712–753. [18] G. Birkhoff, Dynamical Systems, Amer. Math. Soc. Colloquium Publications, vol. IX, Amer. Math. Soc., Providence (1966), reprinting of the original 1927 edition, with an addendum by J. Moser. [19] G. Bredon, Introduction to Compact Transformation Groups, Pure and Applied Mathematics, vol. 46, Academic Press, New York (1972). [20] R. Bryant, An introduction to Lie groups and symplectic geometry, Geometry and Quantum Field Theory (Park City, UT, 1991), IAS/Park City Math. Ser., vol. 1, Amer. Math. Soc., Providence (1995), 5–181. [21] A. Cannas da Silva, Symplectic toric manifolds, Symplectic Geometry of Integrable Hamiltonian Systems (Barcelona, 2001), Adv. Courses Math. CRM Barcelona, Birkhäuser, Basel (2003), 1–83. [22] J. Cheeger, Finiteness theorems for Riemannian manifolds, Amer. J. Math. 92 (1970), 61–74. [23] S.S. Chern, Complex Manifolds Without Potential Theory, 2nd ed., Universitext, Springer-Verlag, New York (1979), with an appendix on the geometry of characteristic classes. [24] C. Conley and E. Zehnder, The Birkhoff–Lewis fixed point theorem and a conjecture of V.I. Arnold, Invent. Math. 73 (1983), 33–49. [25] C. Conley and E. Zehnder, Morse-type index theory for flows and periodic solutions for Hamiltonian equations, Comm. Pure Appl. Math. 37 (1984), 207–253. [26] D. Cox, Recent developments in toric geometry, Algebraic Geometry—Santa Cruz 1995, Proc. Sympos. Pure Math., vol. 62, part 2, Amer. Math. Soc., Providence (1997), 389–436. [27] V. Danilov, The geometry of toric varieties, Uspekhi Mat. Nauk 33 (2) (1978), (200), 85–134, 247; English translation: Russian Math. Surveys 33 (2) (1978), 97–154. [28] T. Delzant, Hamiltoniens périodiques et images convexes de l’application moment, Bull. Soc. Math. France 116 (1988), 315–339. [29] M. Demazure, Sous-groupes algébriques de rang maximum du groupe de Cremona, Ann. Sci. École Norm. Sup. (4) 3 (1970), 507–588. [30] S. Donaldson, An application of gauge theory to four-dimensional topology, J. Differential Geom. 18 (1983), 279–315. [31] S. Donaldson, Irrationality and the h-cobordism conjecture, J. Differential Geom. 26 (1987), 141–168. [32] S. Donaldson, Symplectic submanifolds and almost-complex geometry, J. Differential Geom. 44 (1996), 666–705. [33] S. Donaldson, Lefschetz fibrations in symplectic geometry, Proceedings of the I.C.M., vol. II (Berlin, 1998), Doc. Math., extra vol. II (1998), 309–314. [34] S. Donaldson, Lefschetz pencils on symplectic manifolds, J. Differential Geom. 53 (1999), 205–236. [35] S. Donaldson, Floer Homology Groups in Yang–Mills Theory, Cambridge Tracts in Mathematics, vol. 147, Cambridge Univ. Press, Cambridge (2002), with the assistance of M. Furuta and D. Kotschick. [36] J.J. Duistermaat, On global action-angle coordinates, Comm. Pure Appl. Math. 33 (1980), 687–706. [37] J.J. Duistermaat, The Heat Kernel Lefschetz Fixed Point Formula for the Spin-c Dirac Operator, Progress in Nonlinear Differential Equations and their Applications, vol. 18, Birkhäuser, Boston (1996). [38] J.J. Duistermaat and G. Heckman, On the variation in the cohomology of the symplectic form of the reduced phase space, Invent. Math. 69 (1982), 259–268; Addendum, Invent. Math. 72 (1983), 153–158. [39] Y. Eliashberg, Estimates on the number of fixed points of area-preserving transformations, Syktyvkar University, preprint (1979). [40] Y. Eliashberg, A. Givental and H. Hofer, Introduction to symplectic field theory, GAFA 2000 (Tel Aviv, 1999), Geom. Funct. Anal., special volume, part II (2000), 560–673. [41] Y. Eliashberg and M. Gromov, Lagrangian intersection theory: finite-dimensional approach, Geometry of Differential Equations, Amer. Math. Soc. Transl. Ser. 2, vol. 186, Amer. Math. Soc., Providence (1998), 27–118. [42] Y. Eliashberg and N. Mishachev, Introduction to the h-Principle, Graduate Studies in Mathematics, vol. 48, Amer. Math. Soc., Providence (2002).

Symplectic geometry

185

[43] Y. Eliashberg and L. Traynor, eds, Symplectic Geometry and Topology, Lectures from the Graduate Summer School held in Park City, June 29–July 19, 1997, IAS/Park City Mathematics Series, vol. 7, Amer. Math. Soc., Providence (1999). [44] M. Fernández, M. Gotay and A. Gray, Compact parallelizable four-dimensional symplectic and complex manifolds, Proc. Amer. Math. Soc. 103 (1988), 1209–1212. [45] R. Fintushel and R. Stern, Knots, links, and 4-manifolds, Invent. Math. 134 (1998), 363–400. [46] A. Floer, A relative Morse index for the symplectic action, Comm. Pure Appl. Math. 41 (1988), 393–407. [47] A. Floer, The unregularized gradient flow of the symplectic action, Comm. Pure Appl. Math. 41 (1988), 775–813. [48] A. Floer, Morse theory for Lagrangian intersections, J. Differential Geom. 28 (1988), 513–547. [49] A. Floer, Symplectic fixed points and holomorphic spheres, Comm. Math. Phys. 120 (1989), 575–611. [50] A. Floer, Witten’s complex and infinite-dimensional Morse theory, J. Differential Geom. 30 (1989), 207– 221. [51] M. Freedman, The topology of four-dimensional manifolds, J. Differential Geom. 17 (1982), 357–453. [52] K. Fukaya and K. Ono, Arnold conjecture and Gromov–Witten invariant, Topology 38 (1999), 933–1048. [53] W. Fulton, Introduction to Toric Varieties, Annals of Mathematics Studies, vol. 131, Princeton Univ. Press, Princeton (1993). [54] D. Gay and R. Kirby, Constructing symplectic forms on 4-manifolds which vanish on circles, Geom. Topol. 8 (2004), 743–777. [55] A. Givental, Periodic mappings in symplectic topology, Funktsional. Anal. i Prilozhen 23 (1989), 37–52 (in Russian); translation in Funct. Anal. Appl. 23 (1989), 287–300. [56] M. Golubitsky and D. Tischler, An example of moduli for singular symplectic forms, Invent. Math. 38 (1976/1977), 219–225. [57] R. Gompf, A new construction of symplectic manifolds, Ann. of Math. 142 (1995), 527–595. [58] R. Gompf, Symplectic structures from Lefschetz pencils in high dimensions, Geom. Topol. Monogr. 7 (2004), 267–290. [59] R. Gompf, Toward a topological characterization of symplectic manifolds, J. Symp. Geom., in press. [60] R. Gompf and A. Stipsicz, 4-Manifolds and Kirby Calculus, Graduate Studies in Mathematics, vol. 20, Amer. Math. Soc., Providence (1999). [61] M. Gotay, On coisotropic imbeddings of presymplectic manifolds, Proc. Amer. Math. Soc. 84 (1982), 111–114. [62] P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley Classics Library, Wiley, New York (1994), reprint of the 1978 original. [63] M. Gromov, Stable mappings of foliations into manifolds, Izv. Akad. Nauk SSSR Ser. Mat. 33 (1969), 707–734. [64] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math. 82 (1985), 307–347. [65] M. Gromov, Partial Differential Relations, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 9, Springer-Verlag, Berlin (1986). [66] V. Guillemin, Moment Maps and Combinatorial Invariants of Hamiltonian T n -spaces, Progress in Mathematics, vol. 122, Birkhäuser, Boston (1994). [67] V. Guillemin, V. Ginzburg and Y. Karshon, Moment Maps, Cobordisms, and Hamiltonian Group Actions, Mathematical Surveys and Monographs, vol. 98, Amer. Math. Soc., Providence (2002), with appendix J by M. Braverman. [68] V. Guillemin and S. Sternberg, Convexity properties of the moment mapping, Invent. Math. 67 (1982), 491–513. [69] V. Guillemin and S. Sternberg, Birational equivalence in the symplectic category, Invent. Math. 97 (1989), 485–522. [70] V. Guillemin and S. Sternberg, Symplectic Techniques in Physics, 2nd ed., Cambridge Univ. Press, Cambridge (1990). [71] R. Harvey and H.B. Lawson, Calibrated geometries, Acta Math. 148 (1982), 47–157. [72] N. Hitchin, G. Segal and R. Ward, Integrable Systems. Twistors, Loop groups, and Riemann Surfaces, Oxford Graduate Texts in Mathematics, vol. 4, Clarendon, Oxford Univ. Press, New York (1999). [73] W. Hodge, The Theory and Applications of Harmonic Integrals, 2nd ed., Cambridge Univ. Press, Cambridge (1952).

186

A. Cannas da Silva

[74] H. Hofer and D. Salamon, Floer homology and Novikov rings, The Floer Memorial Volume, Progress in Mathematics, vol. 133, Birkhäuser, Basel (1995), 483–524. [75] K. Honda, Transversality theorems for harmonic forms, Rocky Mountain J. Math. 34 (2004), 629–664. [76] L. Hörmander, An Introduction to Complex Analysis in Several Variables, 3rd ed., North-Holland Mathematical Library, vol. 7, North-Holland, Amsterdam (1990). [77] N. Jacobson, Lie Algebras, Dover, New York (1979), republication of the 1962 original. [78] L. Jeffrey and F. Kirwan, Localization for non-Abelian group actions, Topology 34 (1995), 291–327. [79] G. Kempf, F. Knudsen, D. Mumford and B. Saint-Donat, Toroidal Embeddings, I, Lecture Notes in Mathematics, vol. 339, Springer-Verlag, Berlin (1973). [80] F. Kirwan, Cohomology of Quotients in Symplectic and Algebraic Geometry, Mathematical Notes, vol. 31, Princeton Univ. Press, Princeton (1984). [81] F. Kirwan, Convexity properties of the moment mapping, III, Invent. Math. 77 (1984), 547–552. [82] K. Kodaira, On the structure of compact complex analytic surfaces, I, Amer. J. Math. 86 (1964), 751–798. [83] J. Kohn, Harmonic integrals on strongly pseudo-convex manifolds, I, Ann. of Math. 78 (1963) 112–148. [84] D. Kotschick, On manifolds homeomorphic to CP2 #8 CP2 , Invent. Math. 95 (1989), 591–600. [85] F. Lalonde and D. McDuff, J-curves and the classification of rational and ruled symplectic 4-manifolds, Contact and Symplectic Geometry (Cambridge, 1994), Publ. Newton Inst., vol. 8, Cambridge Univ. Press, Cambridge (1996), 3–42. [86] E. Lerman, Symplectic cuts, Math. Res. Lett. 2 (1995), 247–258. [87] E. Lerman and S. Tolman, Hamiltonian torus actions on symplectic orbifolds and toric varieties, Trans. Amer. Math. Soc. 349 (1997), 4201–4230. [88] T. Li and A. Liu, Symplectic structure on ruled surfaces and a generalized adjunction formula, Math. Res. Lett. 2 (1995), 453–471. [89] A.-K. Liu, Some new applications of general wall crossing formula, Gompf’s conjecture and its applications, Math. Res. Lett. 3 (1996), 569–585. [90] G. Liu and G. Tian, Floer homology and Arnold conjecture, J. Differential Geom. 49 (1998), 1–74. [91] J. Marsden and T. Ratiu, Introduction to Mechanics and Symmetry, A Basic Exposition of Classical Mechanical Systems, Texts in Applied Mathematics, vol. 17, Springer-Verlag, New York (1994). [92] J. Marsden and A. Weinstein, Reduction of symplectic manifolds with symmetry, Rep. Math. Phys. 5 (1974), 121–130. [93] V. Maslov, Perturbation Theory and Asymptotic Methods, Izdat. Moskov Univ., Moscow (1965) (in Russian). [94] D. McDuff, Examples of simply-connected symplectic non-Kählerian manifolds, J. Differential Geom. 20 (1984), 267–277. [95] D. McDuff, Rational and ruled symplectic 4-manifolds, Geometry of low-dimensional manifolds 2 (Durham, 1989), London Math. Soc. Lecture Note Ser., vol. 151, Cambridge Univ. Press, Cambridge (1990), 7–14. [96] D. McDuff, The structure of rational and ruled symplectic 4-manifolds, J. Amer. Math. Soc. 3 (1990), 679–712. [97] D. McDuff, Lectures on groups of symplectomorphisms, Rend. Circ. Mat. Palermo (2) 72 (2004), 43–78. [98] D. McDuff and L. Polterovich, Symplectic packings and algebraic geometry, Invent. Math. 115 (1994), 405–434, with an appendix by Yael Karshon. [99] D. McDuff and D. Salamon, Introduction to Symplectic Topology, Oxford Mathematical Monographs, Oxford Univ. Press, New York (1995). [100] D. McDuff and D. Salamon, J-holomorphic Curves and Symplectic Topology, Amer. Math. Soc. Colloquium Publications, vol. 52, Amer. Math. Soc., Providence (2004). [101] C. McMullen and C. Taubes, 4-manifolds with inequivalent symplectic forms and 3-manifolds with inequivalent fibrations, Math. Res. Lett. 6 (1999), 681–696. [102] K. Meyer, Symmetries and integrals in mechanics, Dynamical Systems (Proc. Sympos., Univ. Bahia, Salvador, 1971), Academic Press, New York (1973), 259–272. [103] J. Milnor, Morse Theory, Annals of Mathematics Studies, vol. 51, Princeton Univ. Press, Princeton (1963), based on lecture notes by M. Spivak and R. Wells. [104] M. Morse, The foundations of a theory in the calculus of variations in the large, Trans. Amer. Math. Soc. 30 (1928), 213–274.

Symplectic geometry

187

[105] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965), 286–294. [106] A. Newlander and L. Nirenberg, Complex analytic coordinates in almost complex manifolds, Ann. of Math. 65 (1957), 391–404. [107] T. Oda, Convex Bodies and Algebraic Geometry—An Introduction to the Theory of Toric Varieties, Ergebnisse der Mathematik und ihrer Grenzgebiete (3), vol. 15, Springer-Verlag, Berlin (1988). [108] K. Ono, On the Arnold conjecture for weakly monotone symplectic manifolds, Invent. Math. 119 (1995), 519–537. [109] J. Park, Non-complex symplectic 4-manifolds with b2+ = 1, Bull. London Math. Soc. 36 (2004), 231–240.

[110] J. Park, Simply connected symplectic 4-manifolds with b2+ = 1 and c12 = 2, Invent. Math., in press. [111] Y. Ruan, Symplectic topology on algebraic 3-folds, J. Differential Geom. 39 (1994), 215–227. [112] Y. Ruan, Topological sigma model and Donaldson-type invariants in Gromov theory, Duke Math. J. 83 (1996), 461–500. [113] D. Salamon, Lectures on Floer homology, Symplectic Geometry and Topology, Y. Eliashberg and L. Traynor, eds, IAS/Park City Math. Ser., vol. 7, Amer. Math. Soc., Providence (1999), 143–229. [114] I. Satake, On a generalization of the notion of manifold, Proc. Nat. Acad. Sci. U.S.A. 42 (1956), 359–363. [115] P. Seidel, Graded Lagrangian submanifolds, Bull. Soc. Math. France 128 (2000), 103–149. [116] J.-C. Sikorav, Points fixes d’un symplectomorphisme homologue à l’identité, C. R. Acad. Sci. Paris Sér. I Math. 299 (1984), 343–346. [117] I. Smith, On moduli spaces of symplectic forms, Math. Res. Lett. 7 (2000), 779–788. [118] I. Smith, Geometric monodromy and the hyperbolic disc, Quart. J. Math. 52 (2001), 217–228. [119] J.-M. Souriau, Structure des Systèmes Dynamiques, Maî trises de Mathématiques, Dunod, Paris (1970). [120] M. Spivak, A Comprehensive Introduction to Differential Geometry, vol. I, 2nd ed., Publish or Perish, Wilmington (1979). [121] N. Steenrod, The Topology of Fibre Bundles, Princeton Mathematical Series, vol. 14, Princeton Univ. Press, Princeton (1951). [122] A. Stipsicz, The geography problem of 4-manifolds with various structures, Acta Math. Hungar. 7 (2000), 267–278. [123] M. Symington, Symplectic rational blowdowns, J. Differential Geom. 50 (1998), 505–518. [124] Z. Szabó, Exotic 4-manifolds with b2+ = 1, Math. Res. Lett. 3 (1996), 731–741. [125] Z. Szabó, Simply-connected irreducible 4-manifolds with no symplectic structures, Invent. Math. 132 (1998), 457–466. [126] C. Taubes, The Seiberg–Witten invariants and symplectic forms, Math. Res. Lett. 1 (1994), 809–822. [127] C. Taubes, More constraints on symplectic forms from Seiberg–Witten invariants, Math. Res. Lett. 2 (1995), 9–13. [128] C. Taubes, The Seiberg–Witten and Gromov invariants, Math. Res. Lett. 2 (1995), 221–238. [129] C. Taubes, SW ⇒ Gr: from the Seiberg–Witten equations to pseudoholomorphic curves, J. Amer. Math. Soc. 9 (1996), 845–918. [130] C. Taubes, Seiberg–Witten invariants and pseudo-holomorphic subvarieties for self-dual, harmonic 2forms, Geom. Topol. 3 (1999), 167–210. [131] W. Thurston, Some simple examples of symplectic manifolds, Proc. Amer. Math. Soc. 55 (1976), 467–468. [132] D. Tischler, Closed 2-forms and an embedding theorem for symplectic manifolds, J. Differential Geom. 12 (1977), 229–235. [133] S. Tolman and J. Weitsman, The cohomology rings of symplectic quotients, Comm. Anal. Geom. 11 (2003), 751–773. [134] L. Traynor, Symplectic packing constructions, J. Differential Geom. 42 (1995), 411–429. [135] C. Viterbo, Symplectic topology as the geometry of generating functions, Math. Ann. 292 (1992), 685–710. [136] A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds, Adv. in Math. 6 (1971), 329–346. [137] A. Weinstein, Lectures on Symplectic Manifolds, Regional Conference Series in Mathematics, vol. 29, Amer. Math. Soc., Providence (1977). [138] A. Weinstein, Fat bundles and symplectic manifolds, Adv. in Math. 37 (1980), 239–250. [139] A. Weinstein, Neighborhood classification of isotropic embeddings, J. Differential Geom. 16 (1981), 125– 128.

188

A. Cannas da Silva

[140] A. Weinstein, On extending the Conley–Zehnder fixed point theorem to other manifolds, Nonlinear Functional Analysis and its Applications, Part 2 (Berkeley, 1983), Proc. Sympos. Pure Math., vol. 45, Part 2, Amer. Math. Soc., Providence (1986), 541–544. [141] R.O. Wells, Differential Analysis on Complex Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 65, Springer-Verlag, New York (1980). [142] H. Weyl, The Classical Groups. Their Invariants and Representations, Princeton Landmarks in Mathematics, Princeton Univ. Press, Princeton (1997). [143] H. Whitney, Analytic extensions of differentiable functions defined in closed sets, Trans. Amer. Math. Soc. 36 (1934), 63–89. [144] E. Witten, Supersymmetry and Morse theory, J. Differential Geom. 17 (1982), 661–692. [145] E. Witten, Topological sigma models, Comm. Math. Phys. 118 (1988), 411–449. [146] W.-T. Wu, Sur les classes caractíristiques des structures fibrées sphériques, Actualités Sci. Ind. 1183 (1952).

CHAPTER 4

Metric Riemannian Geometry Kenji Fukaya∗ Department of Mathematics, Kyoto University, Kitashirakawa Oiwake Cho, Sakyo-ku, Kyoto 606-8502, Japan E-mail: [email protected]

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1. Notations used in this article . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Sphere theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. Finiteness theorems and Gromov–Hausdorff distance . . . . . . . . . . . . . . . 4. Geodesic coordinate, injectivity radius, comparison theorems and sphere theorem 5. Packing and precompactness theorem . . . . . . . . . . . . . . . . . . . . . . . . 6. Construction of homeomorphism by isotopy theory . . . . . . . . . . . . . . . . 7. Harmonic coordinate and its application . . . . . . . . . . . . . . . . . . . . . . . 8. Center of mass technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Embedding Riemannian manifolds by distance function . . . . . . . . . . . . . . 10. Almost flat manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11. Collapsing Riemannian manifolds—I . . . . . . . . . . . . . . . . . . . . . . . . 12. Collapsing Riemannian manifolds—II . . . . . . . . . . . . . . . . . . . . . . . . 13. Collapsing Riemannian manifolds—III . . . . . . . . . . . . . . . . . . . . . . . 14. Morse theory of distance function . . . . . . . . . . . . . . . . . . . . . . . . . . 15. Finiteness theorem by Morse theory . . . . . . . . . . . . . . . . . . . . . . . . . 16. Soul theorem and splitting theorem . . . . . . . . . . . . . . . . . . . . . . . . . 17. Alexandrov space—I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18. Alexandrov space—II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19. First Betti number and fundamental group . . . . . . . . . . . . . . . . . . . . . . 20. Hausdorff convergence of Einstein manifolds . . . . . . . . . . . . . . . . . . . . 21. Sphere theorem and L2 comparison theorem . . . . . . . . . . . . . . . . . . . . 22. Hausdorff convergence and Ricci curvature—I . . . . . . . . . . . . . . . . . . . 23. Hausdorff convergence and Ricci curvature—II . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . .

191 193 194 195 198 204 208 210 212 215 217 220 224 229 232 237 239 245 254 263 271 274 282 291 308

* The author thanks Professors A. Kasue, Y. Otsu, T. Sakai, T. Shioya and T. Yamaguchi (the authors of [92]) who

provided detailed information on the topics covered in this article (those which are written or not written in [92]). This article grew from the (English translation of) the introduction of [92]. The author also thanks Professor F. Dillen for his patience and encouragement. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen Published by Elsevier B.V. 189

This page intentionally left blank

Metric Riemannian geometry

191

1. Introduction This article is a survey of (a part of) Riemannian geometry. Riemannian geometry is a huge area which occupies, I believe, at least 1/3 of the whole of differential geometry. So obviously we need to restrict our attention to some part of it to write an article in this handbook. (M. Berger’s books [20,19] deals with wider topics.) Let me mention first what is not included in this article but should have been included in a survey of Riemannian geometry. (1) We do not include an elementary or introductory part of Riemannian geometry. For example, topics covered in [103, Sections II, III] or [97] are not in this article. We assume the reader to have some knowledge about it. (2) We focus our attention to global results, and results of local nature are rarely discussed. (3) One powerful tool to study global Riemannian geometry is partial differential equations, especially nonlinear one. We do not discuss it.1 The theory of geodesics (which is a theory of nonlinear ordinary differential equations) is one of the main tools used in this article. Linear partial differential equations, especially the Laplacian, is mentioned only when it is closely related to the other topics included in this article. (4) We do not discuss manifolds of nonpositive curvature. (5) We do not discuss scalar curvature. After removing so many important and interesting topics there are still many things missing in this article. For example, results such as filling volume [73] is not discussed. The study of closed geodesics is not included either. So what is included in this article? We focus the part of Riemannian geometry which describes relations of curvature (sectional or Ricci curvature) to topology of the underlying manifold. Since we do not discuss nonpositively curved manifolds, the main target is manifolds of (almost) nonnegative curvature and more generally the class of manifolds with curvature bounded from below. The study of such Riemannian manifolds started with sphere theorems in the 50’s where comparison theorems are introduced by Rauch as an important tool of study. At the beginning of the 70’s Cheeger (and Weinstein) proved finiteness theorems which provide another kind of statements to be established other than sphere theorems. Soon after that, M. Gromov introduced many new ideas, results and tools, such as Gromov– Hausdorff convergence, almost flat manifold theorem, Betti number estimate, etc., and gave tremendous influence to the area. These present the first turning point of the development of metric Riemannian geometry. In the 1980’s global Riemannian geometry was a very rapidly developing area. Especially the class of Riemannian manifolds with sectional curvature bounded from below and above are studied extensively. An important progress in the 1980’s is the theory of collapsing Riemannian manifolds. Those topics are discussed in Sections 2–13. After a brief review of sphere theorems in Section 2, we describe finiteness theorems in Section 3. In Section 4, while explaining a rough sketch of proofs of sphere theorems, we review several basic facts on global Rie1 So, for example, famous result by Hamilton on the 3 manifold of positive Ricci curvature is not discussed.

192

K. Fukaya

mannian geometry, such as Rauch’s comparison theorem, cut points, conjugate points, injectivity radius, etc. One of the main tools of global Riemannian geometry is the Gromov– Hausdorff distance, which we define in Section 5, and we will prove Gromov’s precompactness theorem. The proofs of finiteness theorems are discussed in Sections 6–9. We try to sketch various (different) techniques used to prove finiteness theorems etc. there, rather than to concentrate on one method and to give its full details. Collapsing Riemannian manifolds (under the bound of the absolute value of sectional curvature) is discussed in Sections 10–13. In Sections 14–18, we discuss the class of Riemannian manifolds with sectional curvature bounded from below (but not above). The basic tool to study it is Morse theory of the distance function, which was initiated by Grove–Shiohama. We discuss it and its application to sphere theorems in Section 14. We explain applications of the same method to finiteness theorems in Section 15. The theme of Section 16 is noncompact manifolds of nonnegative curvature. Besides its own interest, it is used in many places to study compact Riemannian manifold. Our focus in this article is on the compact case, so we restrict our discussion on noncompact manifolds to ones which have a direct application to compact manifolds. New turning points of the development of metric Riemannian geometry came at some point in the 1990’s when several mathematicians belonging to the new generation (such as Perelman and Colding) began to work in this field. In Sections 17 and 18 we discuss Alexandrov spaces. They are metric spaces which have curvature > −∞ in some generalized sense. The notion of curvature on a metric space which is not a manifold was introduced by Alexandrov a long time ago. Recently various applications of it to Riemannian geometry (study of smooth Riemannian manifolds) were discovered. It makes this topic more popular among Riemannian geometers. An important structure theorem of Alexandrov spaces is obtained by Perelman and his collaborators, which we review in Sections 17 and 18. In Sections 19–23 we discuss the class of Riemannian manifolds with Ricci curvature bounded from below. The first Betti number and the fundamental group are topics studied extensively under this curvature assumption. We review some of such studies in Section 19. The theme of Section 20 is (mainly) a special case, that is the case of Einstein manifolds. Our discussion of Einstein manifolds is restricted to those related to the other parts of this article. We discuss Einstein manifolds here since they provide rich examples of a new phenomenon which appears when we replace the assumption sectional curvature const, by the Ricci curvature const. Also it is an area where results we discuss in Sections 21–23 provide (and will provide) a powerful tool. Sections 21–23 are reviews of results obtained recently by Colding and Cheeger–Colding on the class of manifolds whose Ricci curvature is bounded from below. Here we emphasize the geometric part of the story and omit most of the analytic parts of the proofs, though analytic parts are as important as geometric parts. It is of course impossible to write down all details of the proofs in this article. However, rather than stating as many results as possible without proof, the author tried to survey as many ideas, tools, techniques, methods of proofs, etc. as possible. In that sense, the emphasis of this article is on methods of proofs and not on their outcome. (Of course important applications of various techniques are explained.) Since this is a survey article there are no new results in it.

Metric Riemannian geometry

193

1.1. Notations used in this article Tp M = the tangent space, Expp : Tp M → M, the exponential map, Bp (R, X) = x ∈ X | d(x, p) < R , for a metric space (X, d) and p ∈ X, KM = the sectional curvature of M,

Vol(M) = the volume of M,

RicciM = the Ricci curvature of M,

Diam(M) = the diameter of M,

iM (p) = the injectivity radius of M at p

(Definition 4.1),

xy = a minimal geodesic joining x and y, xyz = the angle between xy and yz at y, Sn (D) = M | RicciM −(n − 1), dim = n, Diam(M) D , Sn (D, v) = M ∈ Sn (D) | Vol(M) v , Sn (D, i > ρ) = M ∈ Sn (D) | ∀p, iM (p) ρ , Mn (D) = M | |KM | 1, dim = n, Diam(M) D , Mn (D, v) = M ∈ Mn (D) | Vol(M) v , Mn (D, v) = M | KM −1, Diam(M) D, Vol(M) v ,

dGH (X1 , X2 ) = the Gromov–Hausdorff distance (Definition 3.2), Sn (κ) = simply connected Riemannian manifold with KM ≡ κ, Ap (a, b; M) = x ∈ M | a d(p, x) b , Sp (a; M) = x ∈ M | d(p, x) = a . limGH i→∞ Xi = X. means limi→∞ dGH (Xi , X) = 0. The symbol = means almost equal. The argument using this symbol is not rigorous. We use it only when we sketch the proof. The symbol τ (1 , . . . , k |a1 , . . . , am ) stand for the positive number depending only on 1 , . . . , k , a1 , . . . , am and satisfying lim

1 ,...,k →0

τ (1 , . . . , k |a1 , . . . , am ) = 0,

for each fixed a1 , . . . , am . In other words f (1 , . . . , k |a1 , . . . , am ) < τ (1 , . . . , k |a1 , . . . , am ) is equivalent to the following statement. For each δ, a1 , . . . , am there exists such that if 1 < , . . . , k < then f (1 , . . . , k |a1 , . . . , am ) < δ.

194

K. Fukaya

2. Sphere theorems There are several pioneering works in metric Riemannian geometry (such as Myers’ theorem (Theorem 5.4), Hadamard–Cartan’s theorem (Theorem 4.6), study of convex surfaces in R3 , etc.). But let me set the beginning of metric Riemannian geometry at the time when the following theorem was proved. From now on, we denote by KM the sectional curvature of a Riemannian manifold M. We assume all Riemannian manifolds are complete unless otherwise stated. T HEOREM 2.1 (Rauch’s sphere theorem [129]). There exists a positive constant n depending only on the dimension n such that, if a simply connected Riemannian manifold M satisfies 1 KM 1 − n , then M is homeomorphic to a sphere. This theorem is one of the first theorems which are called “sphere theorems”. In this section, we mention some of the most important sphere theorems.2 T HEOREM 2.2 (Klingenberg [94], Berger [18]). If a simply connected Riemannian manifold M satisfies 1 KM > 1/4, then it is homeomorphic to a sphere. If M satisfies 1 KM 1/4, then M is either homeomorphic to a sphere or is isometric to a symmetric space of compact type.3 Theorem 2.2 is a generalization of Rauch’s theorem, and is an optimal result among those characterizing spheres under an assumption of the sectional curvature bounded from above or below.4 (We remark that the sectional curvature of a complex, or quaternionic projective space, or Cayley plane is between 1 and 1/4.) T HEOREM 2.3 (Bochner [157]). If the curvature tensor R of a simply connected Riemannian manifold M satisfies C −Rij k ξ ij ξ k C 2 ξ for any antisymmetric 2 tensor ξ (where C is a positive constant), then the homology group over R of M is isomorphic to the homology group of the sphere. The assumption of Theorem 2.3 is on the curvature operator and is more restrictive than the one on sectional curvature. Hence Theorem 2.3 follows from Theorem 2.2. (Theorem 2.3 was proved earlier.) We mention Theorem 2.3 since the idea of its proof is quite different from the proof of Theorem 2.2. We mention them later in Section 19. T HEOREM 2.4. If M is simply connected and if 1 KM 1 − , then M is diffeomorphic to a sphere. The difference between Theorems 2.4 and 2.2 is that the conclusion of Theorem 2.4 is one on the diffeomorphism type and is sharper. The constant 1 − in Theorem 2.4 2 In this article we mention only a part of many sphere theorems. The reader may find more in [139]. 3 More precisely, one of the complex or quaternionic projective space or the Cayley plane. 4 Several results which relax the condition of Theorem 2.2 to 1 K 1/4 − are known. See [3]. M

Metric Riemannian geometry

195

was 1 − n where n is a positive number depending only on dimension n and was not explicit, at the time when it was first proved by Gromoll and Shikata in [65,136]. Later it was improved to a constant 1 − which is independent of the dimension. It was further improved and an explicit bound (1 − = 0.87) was found [143]. The explicit bound is improved several times.5 The possibility that “1 KM > 0.25 and π1 (M) = {1} implies that M is diffeomorphic to S n ” was not yet eliminated. The best constant is not yet found. Remark 2.1. Hitchin [85] proved that there are some exotic spheres which do not admit a metric of positive scalar curvature, by using the KO index theorem of the Dirac operator. Gromoll–Myer [66] (and Grove–Ziller [83]) found examples of exotic spheres which have a metric of nonnegative curvature. So far no example of an exotic sphere which has a metric of (strictly) positive sectional curvature is found. T HEOREM 2.5 (Berger [18], Grove–Shiohama [82]). If KM 1/4 and if the diameter of M is greater than π , then M is homeomorphic to a sphere. Berger proved that M is homotopy equivalent to a sphere under the assumption of Theorem 2.5 and Grove–Shiohama proved that M is homeomorphic to a sphere. By the generalized Poincaré conjecture (proved by Smale and Freedman) the latter follows from the former (in case dimension is not 3). But the proof by Grove–Shiohama (which is different from Berger’s) uses Morse theory of functions which are not differentiable. This technique turns out to be very useful to study Riemannian manifold under lower (but not upper) curvature bounds. (See Section 14.) The next theorem is a final form of a series of results due to Shiohama [137], Otsu– Shiohama–Yamaguchi [111], Perelman [114]. We will discuss it in Section 22. T HEOREM 2.6 (Cheeger–Colding [29]). There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n then M is diffeomorphic to a sphere. A sphere theorem is a characterization of a sphere, which is the most basic example of Riemannian manifold. Let us recall the classification of surfaces (two-manifolds). It was first proved that a “simply connected compact 2-dimensional manifold is a sphere”, then the classification in the general case was performed by simplifying a general surface by, say, surgery. In a similar sense, sphere theorems play an important role in metric Riemannian geometry. Especially the techniques used to prove the sphere theorems we mentioned above play an important role to study more general Riemannian manifolds.

3. Finiteness theorems and Gromov–Hausdorff distance Another type of important result in metric Riemannian geometry are finiteness theorems. First examples of that kind are proved by Cheeger and by Weinstein, which appeared at the beginning of the 1970’s. Cheeger’s finiteness theorem is as follows. 5 The best estimate known at the time of writing this article is about 1 − = 0.68 [86,144].

196

K. Fukaya

T HEOREM 3.1 (Cheeger [25]). For all positive numbers D, v, n, the number of diffeomorphism classes of Riemannian manifolds M with Diam(M) D, Vol(M) v, and |KM | 1 is finite. The method of proof of Theorem 3.1 is closely related to the proofs of Rauch’s sphere theorem and of Theorems 2.2, 2.4. We will explain it later. Theorems 2.4 and 3.1 (and their proof) use an idea that if two Riemannian manifolds are “close” to each other then they are diffeomorphic to each other. One way to formulate precisely what we mean by two Riemannian manifolds to be close, is by using the notion Gromov–Hausdorff distance.6 Let us first review the definition of (usual or classical) Hausdorff distance. Let (X, d) be a metric space and Y1 , Y2 be subspaces. We put for any subspace Y of X, N Y = x ∈ X | d(x, Y ) < , where d(x, Y ) = inf{d(x, y) | y ∈ Y }. D EFINITION 3.1. The Hausdorff distance dX (Y1 , Y2 ) between Y1 and Y2 is the infimum of > 0 such that Y2 ⊂ N Y1 , Y1 ⊂ N Y2 . The Hausdorff distance defines a complete metric on the set of all compact subsets of a fixed complete metric space (X, d). The Gromov–Hausdorff distance is an “absolute analogue” of the Hausdorff distance. Namely it defines a distance between two metric spaces (which we do not assume to be embedded somewhere a priori). D EFINITION 3.2. The Gromov–Hausdorff distance dGH ((X1 , d), (X2 , d)) between two metric spaces (X1 , d) and (X2 , d) is an infimum of the Hausdorff distance dZ (X1 , X2 ), where Z is a metric space such that X1 , X2 are embedded to Z by isometries. Hereafter we write limGH i→∞ Xi = X if limi→∞ dGH (Xi , X) = 0. Gromov–Hausdorff distance defines a complete metric on the set of all isometry classes of compact metric spaces. The following version is sometimes convenient. D EFINITION 3.3 [52]. A map ϕ : X1 → X2 is called an -Hausdorff approximation, if |dX1 (ϕ(x), ϕ(y)) − dX2 (x, y)| for all x, y ∈ X1 and if the -neighborhood of the image ϕ(X1 ) is X2 . If dGH (X1 , X2 ) then there exists a 3-Hausdorff approximation X1 → X2 . If there exists an -Hausdorff approximation X1 → X2 then dGH (X1 , X2 ) 3. There are two types of important results on the Gromov–Hausdorff distance which are applied to finiteness theorems. In this section, we explain results which were developed mainly in the 1980’s. 6 See [69,75,57] for more detailed account on it.

Metric Riemannian geometry

197

We first state Gromov’s precompactness theorem on manifolds with Ricci curvature bound. Let n, D be a positive integer and a positive number. We denote by Sn (D) the set of all isometry classes of Riemannian manifolds M such that Ricci −(n − 1) and diameter D. Here and hereafter the diameter Diam(X) of a metric space (X, d) is the supremum of d(x, y) where x, y ∈ X. T HEOREM 3.2 (Gromov [69]). (Sn (D), dGH ) is relatively compact in the space of all isometry classes of compact metric spaces. The method of the proof of Theorem 3.2 is related to the proofs of Rauch’s sphere theorem and of Theorem 2.2. We will explain it in Section 5. We next mention a rigidity theorem. Gromov’s precompactness theorem assumes bounds from below of the Ricci curvature, which is a rather weak assumption. We need the stronger assumption for the rigidity theorem. We first discuss the case that Gromov studied in [69]. For n, D, v, we denote by Mn (D, v) the set of all isometry classes of n-dimensional Riemannian manifolds M such that |KM | 1, Diam(M) D, and Vol(M) v. T HEOREM 3.3 [69,93]. There exists n (D, v) > 0 such that if M1 , M2 ∈ Mn (D, v) and if dGH (M1 , M2 ) n (D, v), then M1 is diffeomorphic to M2 . Attempts to prove a similar conclusion as Theorem 3.3 under an assumption milder than M1 , M2 ∈ Mn (D, v), played a very important role in the development of metric Riemannian geometry. Perelman proved that M1 is homeomorphic to M2 if dGH (M1 , M2 ) n (D, v) under the assumption KM −1, which replaces |KM | 1 in the definition of Mn (D, v). (Theorem 18.2.) Further study is done when we assume Ricci curvature bounds. (See Theorem 22.3.) Theorem 3.1 follows from Theorems 3.2 and 3.3. (We leave its proof as an exercise to the reader.) Theorem 3.2 asserts relative compactness. Namely it implies that, for any sequence Mi of elements of Sn (D), there exists a converging subsequence. Its limit M∞ may be regarded as a “weak solution” of various problems of metric Riemannian geometry (when we regard it as an analogy of functional analysis). Then it is natural and important to study the “regularity” of M∞ . It is closely related to the proof of Theorem 3.3. The next result is related to the “regularity” question. T HEOREM 3.4 [69,64,121]. Each element of Mn (D, v) is a Riemannian manifold of C 1,α class.7 Here α is any positive number with α < 1 and a Riemannian manifold of C 1,α -class is a manifold with metric tensor g whose first derivative is C α -Hölder continuous. The assumption of Theorem 3.4 is rather strong. There are two kinds of study to relax this condition X ∈ Mn (D, v). 7 The proof of this theorem is completed in [64,121] based on the idea of Gromov [69]. There seems to be

various independent research in Russia. (See, for example, [107,108,17].)

198

K. Fukaya

One is to remove the assumption Vol(M) v. It means that we study the limit of a sequence of Riemannian manifolds which will become degenerate. This is called the study of collapsing Riemannian manifolds. We discuss it in Sections 10–13. (See also [57].) The other direction is to relax the assumption |KM | 1. Theorem 3.1 is generalized as follows towards this direction. For n, D, v, we denote by Mn (D, v) the set of all isometry classes of n-dimensional Riemannian manifolds M such that KM −1, Diam(M) D, Vol(M) v. T HEOREM 3.5 (Grove–Petersen–Wu [78,81]). For each n, D, v, the number of homeomorphism classes of elements of Mn (D, v) is finite.8 We explain the proof of Theorem 3.5 in Section 15. The limit of a sequence of manifolds M satisfying KM −1 is an Alexandrov space. We will discuss it in Sections 17 and 18. Remark 3.1. (1) If Mi is a sequence of Riemannian manifolds such that N = limGH i→∞ Mi and N is a Riemannian manifold. Then KMi κ implies KN κ. Moreover we have dim N dim Mi . (2) On the other hand, in the case when Λ KMi κ, Λ KN is, in general, false for limGH i→∞ Mi = N . A counterexample can be constructed as follows. Let Rotθ be the rotation by angle θ of S 2 = {(x, y, z) ∈ R3 | x 2 + y 2 + z2 = 1} around the z axis. We consider the quotient of S 2 × R by the Z action generated by (p, t) → (Rotα (p), t + ). Let M,α be the quotient space with quotient metric. (M,α is diffeomorphic to S 2 × S 1 .) We have 1 KM,α 0 since M,α is locally isometric to S 2 × R. The limit of M,α as → 0 is S 2 with some Riemannian metric gα . 1 (S 2 , gα ) does not hold unless α = 0.

4. Geodesic coordinate, injectivity radius, comparison theorems and sphere theorem The following theorem in differential topology is used in the proof of Theorem 2.2. T HEOREM 4.1. If a compact n-dimensional manifold M is a union of two open sets both of which are diffeomorphic to Rn , then M is homeomorphic to a sphere. In order to apply Theorem 4.1 to the proofs of sphere theorems, we want to cover M by two coordinate neighborhoods. To estimate the size of the coordinate charts plays an important role in the study of other problems. Let us begin with the following P ROPOSITION 4.2. For each compact Riemannian manifold M, there exists a positive number M with the following properties. If the distance between p, q ∈ M is smaller than M , then there exists a unique geodesic of length < M joining p, q. 8 In case the dimension is 3, [78,81] proved only finiteness of homotopy type. Now, Perelman’s stability theorem

(Theorem 18.2) implies the finiteness of homeomorphism classes in general.

Metric Riemannian geometry

199

The proof of Proposition 4.2 is given in many standard text books of Riemannian geometry. (For example, in [97,33].) The uniqueness of such geodesic is essential for our purpose. Let us explain this point. Let M be a complete Riemannian manifold. For each p ∈ M we define the exponential map, Expp : Tp M → M as follows. Let V ∈ Tp (M). There exists a geodesic : R → M, such that d dt (0) = V . We then put (1) = Expp V . Proposition 4.2 implies that Expp : Tp M → M is a diffeomorphism on the ball of radius M . D EFINITION 4.1. The injectivity radius of a Riemannian manifold M is a function iM : M → R which associates to p ∈ M the positive number: iM (p) = sup | Expp : Tp M → M is injective on V ∈ Tp M | V < . Proposition 4.2 implies iM M for a compact Riemannian manifold M. (It is easy to see that iM is continuous. Hence iM M > 0 follows easily from the implicit function theorem. Proposition 4.2 is a bit more involved.) If R < iM (p), then the restriction of the exponential map Expp : Tp M → M to the metric ball of radius R centered at origin, defines a coordinate of a neighborhood of p. We call it the geodesic coordinate. To prove Theorem 2.2, it is important to estimate the injectivity radius iM from below. The next result9 provides such an estimate. T HEOREM 4.3. Suppose that dim M is even. If KM > 0, then iM π and M is simply connected.10 Suppose dim M is odd. If 1 KM 1/4 and if M is simply connected then, iM π . In particular, if M satisfies the assumption of Theorem 2.2, then we have iM π . (There are several results in the nonsimply connected case. We omit it.) Another result we use is the following P ROPOSITION 4.4 (Berger). Let us assume that KM 1/4 and Diam(M) π . We take p, q ∈ M such that d(p, q) = Diam(M). Then we have Int Bp (π, M) ∪ Int Bq (π, M) = M. (Here Int denotes the interior.) The proof is in Section 14. Using Theorem 4.3 and Proposition 4.4, the proof of Theorem 2.2 goes roughly as follows. By Theorem 4.3, the injectivity radius of M is not smaller than π . Especially the diameter of M is not smaller than π . 9 This theorem is due to [18] in even dimension, and to [95,37] in odd dimension. 10 The second assertion is a classical result due to Synge.

200

K. Fukaya

Let us first assume 1 KM > 1/4. We replace the metric gM of M by (1 + δ)gM , where δ is a positive number sufficiently close to 0. The assumption 1 KM > 1/4 is still satisfied. Hence M satisfies the assumption of Proposition 4.4. Hence Int Bp (π, M) ∪ Int Bq (π, M) = M. Moreover Int Bp (π, M) and Int Bq (π, M) are diffeomorphic to the ball by Theorem 4.3. Therefore, by Theorem 4.1, M is homeomorphic to a sphere. We next consider the case when 1 KM 1/4. If the diameter of M is strictly greater than π , then again Proposition 4.4 and Theorems 4.1 and 4.3 imply that M is homeomorphic to a sphere. Finally we consider the case when the diameter of M is π . In this case, we consider the restriction of the exponential map Expp : Tp M → M to the ball D n (π) of radius π . Then it is a diffeomorphism at the interior. So M is obtained from D n (π) by identifying boundary points only. We examine this situation carefully and conclude that M is a symmetric space of compact type. We omit the details. (See, for example, [33, Chapter 7].) We explain the outline of the proof of Theorem 4.3 later in this section. We first explain some basic facts. Let us begin with the following theorem. Let κ be a constant. We put sin t √κ √κ , sκ (t) = t, √ sinh√t −κ , −κ

κ > 0, κ = 0, κ < 0.

(4.1)

T HEOREM 4.5 (Rauch). If KM κ, then the derivative dx Expp of the exponential map Expp satisfies dx Exp (V ) V sκ (r). p √ Here x ∈ Tp (M), x = r, V ∈ Tx Tp (M) ∼ = Tp (M) and we assume r π/ κ in case κ > 0. Let KM κ. In case κ > 0, we assume dtx Expp is invertible for t ∈ [0, 1]. Then we have dx Exp (V ) V sκ (r). p Theorem 4.5 implies that if KM 1, then the restriction of Expp : Tp M → M to the ball of radius π is an immersion. (Namely its Jacobi matrix is invertible.) We remark that the equality in Theorem 4.5 holds in the case when M is of constant curvature κ. Theorem 4.5 is used by Rauch to prove his sphere theorem. We use the Jacobi field in the proof of Theorem 4.5 as follows. Let x, V be as in Theorem 4.5, and define a geodesic s by s (t) = Expp t (x + sV ) .

Metric Riemannian geometry

201

For each s, s is a geodesic. Its derivative

∂s (t)

J (t) = ∈ T0 (t) M ∂s s=0 with respect to s, by definition, is a Jacobi field. Note that dx Expp (V ) = J (1). Therefore, to prove Theorem 4.5, it suffices to estimate the Jacobi field. We use the following equation (which the Jacobi field satisfies) for this purpose. d0 d0 D2 (t), J (t) (t) = 0. J (t) + R dt dt dt 2

(4.2)

D 0 Here dt is a covariant derivative with respect to the tangent vector d dt (t) and R is a curvature tensor. If e1 , e2 is an orthonormal frame of a plane π in the tangent space, then g(R(e1 , e2 )e2 , e1 ) is the sectional curvature of the plane π . (Here g is the metric tensor.) Therefore, the second term of Eq. (4.2) can be written in terms of the sectional curvature. Using it we can compare Eq. (4.2) to the one in case our manifold is of constant curvature. Namely if KM ≡ κ then (4.2) will be

D2 J (t) + κJ (t) = 0. dt 2

(4.3)

Its solution is J (t) = sκ (t)V (t) where ∇(t) ˙ V = 0. Namely J (t) = sκ (t) if KM ≡ κ. This implies Theorem 4.5. Theorem 4.5 implies the following T HEOREM 4.6 (Hadamard–Cartan). If a complete Riemannian manifold M satisfies KM 0, then Expp : Tp M → M is a covering map. In particular the universal covering space of M is diffeomorphic to Rn . In fact, Theorem 4.5 implies that the Jacobi matrix of Expp : Tp M → M is of maximal rank everywhere. To prove that it is a covering map we need a bit more. We use completeness of metric for this last step. We omit it. By integrating the conclusion of Theorem 4.5, we can compare the distance between two points Expp (x), Expp (y) (which are close to p) to the corresponding distance in the space with constant curvature. Actually we can do it more globally and obtain the Toponogov comparison theorem. To state it we need some notation. Let Sn (κ) be the complete simply connected Riemannian manifold with constant curvature κ. Let x , y , z ∈ Sn (κ). We denote by x y , etc. the minimal geodesic joining x and y , etc. Let θ = y x z be the angle between x y and

202

K. Fukaya

Fig. 4.1.

x z at x . We put a = d(x , y ), b = d(x , z ). It is easy to see that d(y , z ) depends only on a, b, θ, κ. We define s(a, b, θ, κ) = d(y , z ).

(4.4)

√ We remark that in case κ > 0, the number s(a, b, θ, κ) is defined only for a, b < π/ κ. Let M be a Riemannian manifold and x, y, z ∈ M. We denote by xy a minimal geodesic joining x and y. (In case there are several minimal geodesics we take any one of them.) Let yxz be the angle between xy and xz at x (see Figure 4.1). T HEOREM 4.7 (Alexandrov–Toponogov). If KM κ then we have d(y, z) s d(x, y), d(x, z), yxz, κ . If KM κ and if d(x, y), d(x, z) iM (x) then d(y, z) s d(x, y), d(x, z), yxz, κ . We remark that in the first inequality we do not need to assume that the triangle x, y, z is small. Actually we only need to assume one of the geodesics √ joining x to y and to z are minimal and the other may be any geodesic of length π/ κ. Theorem 4.7 is proved in many text books (see, for example, [33]). As we already mentioned, Theorem 4.5 implies that, if KM 1, then the exponential map is an immersion on the metric ball of radius π . Especially it is locally an injection there. To prove Theorem 4.1 we need global injectivity. We here introduce several terminology. D EFINITION 4.2. q ∈ M is said to be a conjugate point of p ∈ M if there exists x such that q = Expp (x) and that dx Expp is not of maximal rank. q is said to be a cut point of p ∈ M if there exists x = y ∈ Tp (M) such that Expp x = Expp y = q. E XAMPLE 4.1. We consider the sphere S 2 of constant curvature 1. Every geodesic which starts at the north pole np meets again at the south pole sp. Hence the south pole is a conjugate point of the north pole.

Metric Riemannian geometry

203

Fig. 4.2.

We next divide S 2 by the involution and obtain the real projective space RP 2 . Then np and sp determine the same point x = [np] = [sp] ∈ RP 2 . If c ∈ S 2 is on the equator then there are minimal geodesics 1 , 2 joining c to np, sp, respectively. 1 , 2 induce two minimal geodesics ¯1 , ¯2 in RP 2 joining x to y = [c]. Thus y is a cut point of x. Note that iM (p) > r holds if there exists neither a cut point nor a conjugate point q of p such that d(p, q) r. We can use Theorem 4.5 to estimate the distance to the conjugate point. However the problem to estimate the distance to the cut point is a more global one. We remark the following fact. L EMMA 4.8. If : [a, b] → M is the minimal geodesic, then for t ∈ (a, b), q = (t) is neither a cut point nor a conjugate point of p = (a). In fact if q is a cut point then there is a geodesic joining p to q with | | = |[a,t] | (see Figure 4.2). Then the union ∪ |[t,b] of two geodesics is not smooth and has the same length as the minimal geodesic . This is a contradiction. If q is a conjugate point then by the Morse index theorem (see [103,97,33]), [a,t+] is not minimal. This contradicts the assumption. Here we state the following basic result about cut points. (See, for example, [33, p. 96] for its proof.) T HEOREM 4.9 (Klingenberg). Let M be a Riemannian manifold. We assume that q is not a conjugate point of p, for each p, q ∈ M with d(p, q) < r. If there exists p ∈ M with iM (p) < r then there exists a closed geodesic of length < 2r in M. In view of Theorems 4.5 and 4.9, to prove Theorem 4.3, it suffices to show that the length of a nontrivial closed geodesic of M is greater than 2π . We explain the brief outline of its proof below. (See [33, p. 100] for its details.) We first consider the case dim M is even. Let M be a simply connected Riemannian manifold with 1 KM > 0. Let : S 1 → M be a nontrivial geodesic of minimal length. We regard S 1 ∼ = R/Z. Put p = (0). By the parallel transport along we have a holonomy homomorphism hol : Tp M → Tp M. The tangent vector d dt (0) is an invariant of the holonomy. Since hol is an orthogonal transformation, and dim M is even, it follows that there ex-

204

K. Fukaya

ists a nonzero vector V ∈ Tp M orthogonal to d dt (0) such that hol (V ) = V . The parallel transport of V defines a vector field V (t) ∈ T(t) M, which is a parallel vector field. We put s (t) = Exp(t) sV (t) . Using ∇V (t) = 0 and first variation formula (see, for example, [33, Section 1], [97, Vol. II, s Theorem 5.1], [103, Theorem 12.2]), we find that d ds (0) = 0. Using moreover the second variation formula (see, for example, [33, Section 6], [97, Vol. II, Theorem 5.4], [103, The2 orem 13.1]) and the positivity of curvature, we find dds2s (0) < 0, which contradicts to the minimality of the length of . The proof of the odd-dimensional case is more involved. We remark that the quotient of S 3 by a cyclic group Z/pZ has constant positive curvature one (and is not simply connected). Its injectivity radius converges to 0 as p → ∞. This shows that, to prove Theorem 4.3 in odd-dimensional case, we need to use the assumption that M is simply connected. The proof of the odd-dimensional case is roughly as follows. We assume that there exists a closed geodesic of length < 2π . Since M is simply connected, is null homotopic. Let s be a homotopy such that 0 = , 1 = const. We may assume that the length of is minimal among all nontrivial closed geodesics. By using the assumption that KM > 1/4 we can prove that the length of s is always smaller than 2π . (This is the essential point of the proof. To prove this we use the fact that the Morse index (with respect to the length) of the closed geodesic of length > 2π is not smaller than 2.11 ) Now we consider the exponential map Expp at the tangent space of p = (0). Expp is a submersion on the ball of radius π . Hence it has a similar property to the covering map up to radius π . Especially it has the homotopy lifting property there. Since the length of s is not greater than 2π , its image is of distance π from p. Therefore we can lift s to Tp M. (Note we can lift 1 since it is a constant map.) Hence we obtain a lift ˜0 : S 1 → Tp M. But this is a contradiction since 0 = is a geodesic.12

5. Packing and precompactness theorem A similar argument as in the last section is used in the proof of the finiteness theorem (Theorem 3.1) and of Theorem 3.2. We explain this point here. We first discuss Theorem 3.2. The basic fact we use for its proof is the following 11 Let us consider the round sphere of radius 2 (that is the round sphere of curvature 1/4). The geodesic segment of length 2π , that is the geodesic segment joining north pole with south pole, has Morse index n − 1. (Here we consider the set of all arcs joining north pole with south pole and consider the length as a Morse function on it. n − 1 is the Morse index with respect to this Morse function.) We compare our closed geodesic with this geodesic segment to obtain the conclusion about the Morse index. 12 This argument is not enough to handle the case 1 K 1/4 of Theorem 2.2 (since then we can only show M that π is a submersion at the interior of the ball of radius π ). In that case we need an additional argument. We omit it.

Metric Riemannian geometry

205

P ROPOSITION 5.1. Let D > 0 and N : (0, 1) → N. We denote by Met(D, N ), the set of all isometry classes of complete metric spaces satisfying (1), (2) below. Then Met(D, N ) is compact with respect to the Gromov–Hausdorff distance. (1) The diameter of M D. (2) For each ∈ (0, 1) there exists a finite subset Z of M with the following properties: (2.a) !Z N (). (2.b) For each x ∈ M, there exists x0 ∈ Z satisfying d(x, x0 ) < . The proof of Proposition 5.1 is given, for example, in [57, Section 2]. Here we introduce a notation. D EFINITION 5.1. We call the subset Z an -net if it satisfies (2.b). To deduce Theorem 3.2 from Proposition 5.1, we use the following Theorem 5.2. Let Sn (κ) be the complete simply connected Riemannian manifold with constant curvature κ. Let Bp (R, M) be the metric ball in M of radius R centered at p. T HEOREM 5.2 (Bishop–Gromov). If Ricci (n − 1)κ then the volume Vol(Bp (R, M)) of the metric ball satisfies the following inequality for r < R: Vol(Bp (R, M)) Vol(Bp0 (R, Sn (κ))) . Vol(Bp (r, M)) Vol(Bp0 (r, Sn (κ)))

(5.1)

(5.1) is called the Bishop–Gromov inequality. It plays a key role to study the class of Riemannian manifolds with Ricci curvature bounded from below. The equality holds if M is of constant curvature κ. Let us sketch a proof of Theorem 5.2. We put A(t) =

Vol(Bp (t, M)) . Vol(Bp0 (t, Sn (κ)))

(5.2)

It suffices to show that A is nonincreasing. (In case κ > 0, Theorem 5.4 implies that we need to consider t π only.) Let : [0, a) → M be a minimal geodesic with (0) = p parameterized by arc length. Let v = (d/dt)(0). We take a vector v∗ ∈ Tp0 Sn (κ) with unit length. We put a(v, t) =

det dtv Expp det dtv∗ Expp∗

.

(5.3)

Here det dtv Expp is the determinant of the derivative of the exponential map. We first prove that a(v, t) is a nonincreasing function of t for each fixed v. We can prove it in a way similar as the proof of Theorem 4.5. One difference however is that our assumption in Theorem 5.2 is only on the Ricci curvature while in Theorem 4.5 the assumption is on the sectional curvature. However since we only need to estimate

206

K. Fukaya

the determinant of the Jacobi matrix of the exponential map, the assumption on the Ricci curvature, which is the trace of the curvature tensor, is enough. This is half of the idea of the proof of Proposition 5.2. Let us fix p and move q ∈ M, and consider the set

dp,q (5.4) V= (0) ∈ Tp M

q ∈ M , dt where p,q is the minimal geodesics joining p and q. (If there are several we take all of d them.) (We take a parametrization of p,q so that the length of dtp,q (0) is d(p, q).) We have Vol Bp (R, M) = det dx Expp dx. (5.5) V ∩B0 (R,Rp M)

(Here det dx Exp is the determinant of Jacobi matrix.) (5.5) and the fact that a(v, t) is a nonincreasing function of t implies (5.1) for R, r smaller than the injectivity radius. To prove Theorem 5.2 beyond injectivity radius, we proceed as follows. We remark that V is star shaped (that is if x ∈ V t ∈ [0, 1], then tx ∈ V ). We then modify a to a so that a (t, v) = a(t, v) if tv ∈ V and a (t, v) = 0 if tv ∈ / V . Then a is a nonincreasing function of t. Theorem 5.2 follows. C OROLLARY 5.3. If RicciM κ and p ∈ M then Vol Bp (R, M) Vol Bp0 R, Sn (κ) . This corollary follows from the fact that the function A in (5.2) is nonincreasing and limt→0 A(t) = 1. Theorem 5.2 and Proposition 5.1 imply Theorem 3.2 as follows. Let us assume that M satisfies the assumption of Theorem 3.2. It suffices to show that M satisfies the assumption of Proposition 5.1. Let > 0. We take Z ⊂ M which is maximal (with respect to inclusion) among the subsets of M satisfying “z1 , z2 ∈ Z, z1 = z2 , implies d(z1 , z2 ) > ”. The maximality implies (2.6). On the other hand, since Bz (/2, M), z ∈ Z, are disjoint to each other, it follows that Vol Bz (/2, M) < Vol M. z∈Z

Since Bz (D, M) = M, Proposition 5.1 implies !Z

Vol(Bp0 (D, Sn (κ))) Vol(M) . sup Vol(Bp (/2, M)) Vol(Bp0 (/2, Sn (κ)))

If we let N () be the right-hand side, then the assumption of Proposition 5.1 is satisfied. Theorem 3.2 follows. We remark that the following classical result is actually proved during the proof of Theorem 5.2.

Metric Riemannian geometry

207

T HEOREM 5.4 (Myers). If M is an n-dimensional complete Riemannian manifold √ with Ricci (n − 1)κ > 0, then M is compact and its diameter is not greater than π/ κ. In fact during the proof of Theorem 5.2 we proved the following under the assumption p ∈ M, RicciM κ. “If t → Expp (tv) is a minimal geodesic for t ∈ [0, 1], then det dv Expp is not greater than det dv0 Expp0 , where p0 ∈ Sn (κ), v0 ∈ Tp0 Sn (κ) and |v0 | = |v|.” √ We remark that det dv0 Expp0 = 0 if v0 = π/ κ. Therefore there exists no minimal √ geodesic of length > π/ κ if RicciM κ. Theorem 5.4 follows immediately. In the above argument, Bz (, M), z ∈ Z, covers M. Namely we estimate the number of metric balls (geodesic coordinate) to show Theorem 3.2. If is smaller than the injectivity radius of M, then Bz (, M) is diffeomorphic to D n . The proof of Theorem 3.2 is related to the proof of sphere theorems in this way. Theorem 4.1 deals with the case when two balls cover M and conclude that M is a sphere. If we can replace Theorem 4.1 by a statement such as “if M is covered by the balls whose number is estimated by C, then the number of diffeomorphism classes of such M is estimated by C” then finiteness theorems would follow. Unfortunately the statement in the parenthesis above does not hold. So we need to include information how the balls are glued. Theorem 3.1 can be proved in that way. (See Sections 6–8.) Here we prove a weaker version (Weinstein [150]). P ROPOSITION 5.5. For each D, the number of homotopy equivalence classes of n-dimensional Riemannian manifolds satisfying (1), (2) below is finite. (1) M ∈ Mn (D), (2) The injectivity radius of M is greater than . To prove Proposition 5.5 we use the set Z above. We then obtain an open covering Bz (, M), . z ∈ Z, of M. It is a simple covering. Namely for each z1 , . . . , zk ∈ Z the intersection ki=1 Bzi (, M) is either empty or contractible. It implies that the simplicial complex K(Z) defined below is homotopy equivalent to M. (1) A vertex of K(Z) corresponds to an element of Z. (2) . z0 , . . . , zk ∈ Z is the set of vertices of a k simplex of K(Z) if and only if k i=0 Bzi (, M) = ∅. Since the order of Z is estimated by a number depending only on D and , it follows that there exists only finitely many possibilities for the homotopy type of K(Z). Proposition 5.5 follows. In Theorem 3.1, there is no assumption on the injectivity radius but only a bound of volume from below is assumed. An assumption on the volume is more natural and geometric than one on the injectivity radius. However, in case the absolute value of the sectional curvature is bounded, these two assumptions are equivalent. P ROPOSITION 5.6 (Cheeger [25]). There exists a positive number c(n, D, v) depending only on n, D, v such that if M ∈ Mn (D, v), then iM c(n, D, v).

208

K. Fukaya

The proof of Proposition 5.6 is closely related to the proof of Theorem 3.5. We will explain it in Section 15.

6. Construction of homeomorphism by isotopy theory In Section 5, we discussed an estimate of the number of open sets which cover M and which are diffeomorphic to D n , and we showed how it is used to estimate the number of homotopy types (Proposition 5.5). However as we mentioned there, we need more arguments to estimate the number of diffeomorphism classes (or homeomorphism classes). We will explain some of them in the four sections beginning from this section. We again begin with a sphere theorem, the differentiable sphere theorem (Theorem 2.4) this time. Let M satisfy the assumptions of Theorem 2.4. Namely we assume that M is simply connected and 1 KM 1 − . Then by Proposition 4.4 and Theorem 4.3, M is a union of two balls V1 , V2 such that Vi ∼ = D n . We may assume ∂Vi ∼ = S n−1 . Moreover we may assume V1 ∩ V2 = ∂V1 = ∂V2 . So we obtain a diffeomorphism I : S n−1 ∼ = ∂V1 → ∂V2 ∼ = S n−1 .

(6.1)

It is easy to see that if I is diffeotopic to the identity map (namely if there exists a smooth family It of diffeomorphisms such that I0 = I , I1 = id), then M = V1 ∪ V2 is diffeomorphic to S n . Now we use the following P ROPOSITION 6.1. For each compact Riemannian manifold N there exists N > 0 such that if the C 1 -distance between F : N → N and the identity is smaller than N , then F is diffeotopic to the identity. Here we recall D EFINITION 6.1. Two diffeomorphisms F1 , F2 : N → N are said to be diffeotopic to each other if there exists a smooth map F : [1, 2] × N → N such that F (1, x) = F1 (x), F (2, x) = F2 (x) and that x → F (t, x) is a diffeomorphism for each t. The proof is elementary. To apply Proposition 6.1 to the proof of Theorem 2.5, we use the following lemma. L EMMA 6.2. For each > 0 there exists δn () > 0 with the following properties. Let M be an n-dimensional simply connected Riemannian manifold with 1 > K > 1 − δn (). Then we may choose the gluing map (6.1) so that its C 1 -distance from the identity is smaller than . We omit the proof. See, for example, [33, Chapter 7].

Metric Riemannian geometry

209

Fig. 6.1.

We are going to explain how we use the idea above for the proof of Theorems 3.3 and 3.1. Cheeger’s original proof of Theorem 3.1 [25] is similar to the idea explained in this section. Let M, N be Riemannian manifolds. We assume $ by the same num$ that they are covered ber of metric balls. Namely we assume M = ki=1 Bpi (, M), N = ki=1 Bqi (, N ). We assume also that 10 is smaller than the injectivity radius of M and of N . (We put 10 for a technical reason.) We assume also that intersection pattern of the balls are the same. Namely, for each i, j , Bpi (, M)∩Bpj (, M) = ∅ if and only if Bqi (, N )∩Bqj (, N ) = ∅. We want to find a sufficient condition for M to be diffeomorphic $ to N . For this purpose $ we compare the chart ki=1 Bpi (, M) of M, with the chart N = ki=1 Bqi (, N ) of N . To compare, we want to take the same domain for coordinate transformations. For this purpose we proceed as follows. Let Bpi (, M) ∩ Bpj (, M) = ∅ then Bpi (, M) ⊂ Bpj (10, M) (see Figure 6.1). For each pi , qj , we fix a linear isometry Tpi M ∼ = Rn , Tqj N ∼ = Rn and use n it to identify tangent spaces with R . (There are various choices of identification. We take one and fix it.) We consider the composition n n ϕjMi = Exp−1 pj ◦ Exppi : B () → B (10).

Here B n () is a metric ball of radius in Rn centered at origin, and Exp−1 pj is an inverse of the exponential map Exppj : B n (10) → N . We define ϕjNi in a similar way. In the next proposition we assume that the C 2 norm (or C 1,α norm) of ϕjMi , ϕjNi is smaller than a constant C. P ROPOSITION 6.3. There exists n,k (C) > 0 such that if the C 1 distance between ϕjMi and ϕjNi is smaller than n,k (C), then M is diffeomorphic to N .

210

K. Fukaya

Cheeger proved Proposition 6.3 in the following way. First we use Proposition 6.1 to prove that the coordinate transformation ϕjMi is diffeotopic to ϕjNi . We then use it to $ construct a diffeomorphism K i=1 Ui → N (to its image) by induction in K. For details see [25]. We prove Proposition 6.3 in a slightly different way in Section 7. Proposition 6.3 is used to prove Theorem 3.1. For this purpose, we first observe that there is a constant C such that a Riemannian manifold satisfying the assumption of Theorem 3.1 is covered by metric balls whose number is not greater than C. Since the number of metric balls is bounded, the number of possible intersection patterns among them is also bounded. Let us fix intersection pattern of the metric balls we use. We use Proposition 6.3 and find that, if the coordinate transformations ϕjMi are C 1 close to ϕjNi , then M is diffeomorphic to N . If the coordinate transformations ϕjMi are uniformly bounded in the C 2 norm then Ascoli–Alzera’s theorem implies that they are precompact in C 1 -topology. Theorem 3.1 will follow. We need however to estimate second derivative of the coordinate transformation uniformly. Our assumption in Theorem 3.1 is on curvature, which is a second derivative of the metric tensor. So one may imagine that it implies the estimate of the second derivative of coordinate transformation. However when we use geodesic coordinates, the assumption of (sectional) curvature is not enough to do so. (Cheeger [25] proved it under the additional assumption that a covariant derivative of the curvature tensor is also bounded.) To go around this trouble, Cheeger in [25] proceeds as follows. Instead of using a statement such as “two diffeomorphism is diffeotopic to each other if they are C 1 -close to each other” we can use a statement such as “two homeomorphism are isotopic to each other if they are 13 C 0 -close to each other” $k [48]. And we can use the isotopy extension theorem to construct an homomorphism i=1 Ui → N by induction on k. This argument implies finiteness of homeomorphism classes and is not enough to prove Theorem 3.1 in four dimensions.14 (In higher dimensions, one can use surgery etc. to deduce finiteness of diffeomorphism classes from finiteness of homeomorphism classes by purely topological argument.) We can use harmonic coordinates (which we discuss in the next section) to find a coordinate chart such that the C 2,α -norm of its coordinate transformation can be estimated uniformly.

7. Harmonic coordinate and its application As we mentioned in the last section, in order to obtain an estimate of the Hölder norm of the coordinate transformation, taking geodesic coordinates does not give an optimal result. Harmonic coordinates are the best choice for this purpose.15 There are various other 13 Which is much less elementary than Proposition 6.1 and is based on highly nontrivial results such as Kirby– Siebenmann’s result on the Hauptvermutung. See [48]. 14 [120] added some technical argument and proved Theorem 3.1 in four dimension as well. 15 In mathematical study of gauge theory, we need to take representative of gauge equivalence class in order to kill freedom of gauge transformation. This is an important point to study moduli space of connections. Here we are studying “gravity” and coordinate transformation plays a role of gauge transformation. The process to find a good coordinate is called gauge fixing in Physics. Harmonic coordinates are used in Riemannian geometry around

Metric Riemannian geometry

211

applications of harmonic coordinates.16 It also plays an important role to prove that the limit metric in Theorem 3.4 is of C 1,α -class. Let M be a Riemannian manifold. We assume that the injectivity radius of M is much greater than r. Let p ∈ M and ei (p), i = 1, . . . , n be an orthonormal frame of Tp M. We put vi (p) = Expp (rei (p)), wi (p) = Expp (−rei (p)) and define hp,i (x) =

d(x, wi (p))2 − d(x, vi (p))2 . 4r 2

(7.1)

We call hp,i an almost linear function. (We remark that hp,i is a linear function if M = Rn .) hp = (hp,1 , . . . , hp,n ) defines a coordinate system in a neighborhood of p. However since hp is in principle a distance function, this coordinate does not provide optimal results for the estimate of the Hölder norm of coordinate transformation. We will replace it by a harmonic function. We consider a boundary value problem of the Laplace equation ϕ = 0 as follows. Let us take δ such that r - δ - iM (p), and consider ϕp,i : Bp (δ, M) → R with the following properties: (1) ϕp,i = 0. (2) If q ∈ Sp (δ, M), then ϕp,i (q) = hp,i (q). D EFINITION 7.1. We call ϕp = (ϕp,1 , . . . , ϕp,n ) a harmonic coordinate. p

p

Using the fact that ϕi is C 1 -close to hi we can prove that ϕp defines a coordinate in a neighborhood of p. Now we can prove an estimate of C 2,α norm of the coordinate transformation of the harmonic coordinate as follows. We put D n () = {x ∈ Rn | x < }. We take with 10 < r. Let p, q ∈ M with d(p, q) < . We consider the inverse ϕp−1 of ϕp . Then the image of ϕp−1 : D n () → M is contained in the domain of ϕq : Bq (r, M) → Rn . Therefore we can define M ϕq,p = ϕq ◦ ϕp−1 : D n () → Rn .

(7.2)

T HEOREM 7.1. There exists a positive constant C(r, , α, n) depending only on r, , α and M is not greater than C(r, , α, n). the dimension n, such that the C 2,α -norm of ϕq,p 1,α Also the C -norm of the metric tensor in harmonic coordinates is estimated by C(r, , α, n). The proof is based on a priori estimate of harmonic functions. See [87,88,64], where the second half is proved. The first half follows easily from the second half. Theorem 7.1 is generalized to Theorem 20.7. the same time when Uhlenbeck etc. used Coulomb gauge in the study of moduli space of connections. The proof of Theorem 3.4 we present in this section is very similar to the proof by Uhlenbeck etc. of the compactification of the moduli space of self dual connections on 4 manifolds. 16 We can use it to study Gromov–Hausdorff convergence under weaker assumption also. See Section 20.

212

K. Fukaya

Let us prove Theorem 3.4 as a typical application of Theorem 7.1.17 Let us take a sequence Mk of elements of Mn (D, v). We denote its limit in Gromov–Hausdorff distance by X. By Theorem 4.3, the injectivity radius of Mk is greater than r, a number independent of k. We take such that 10 < r. In the same way as Section 2, we can take a finite subset {pi,k | i = 1, . . . , Ik } ∈ Mk with the following properties: (1) $ Ik is smaller than a number independent of k. (D n ()) = Mk . (2) i ϕp−1 i,k By (1) we may assume that Ik is independent of k by taking a subsequence if neces(D n ()) has only sary. Set I = Ik . Then the intersection pattern of the coordinates ϕp−1 i,k a finite number of possibilities. Hence by taking a subsequence we may assume that the intersection pattern is independent of k. Namely we may assume that for each i, j I , n n D () ∩ ϕp−1 D () ϕp−1 i,k j,k

(7.3)

is empty or not does not depend on k. M Now for any i, j such that (7.3) is not empty, we consider ϕpj,kk ,pi,k defined by (7.2). We fix α < 1, and apply Theorem 7.1 to α with 1 > α > α. We then find that the C 2,α -norm of ϕpMj,kk ,pi,k is estimated by a number independent of k. Hence we may take a subsequence and assume that ϕpMj,kk ,pi,k converges in C 2,α -topology. Let us denotes its limit by ϕpj,∞ ,pi,∞ : D n () → Rn . We use them as a coordinate transformation to obtain a smooth manifold M∞ of C 2,α class. Moreover by the uniform C 1,α -boundedness of metric tensor, we find a Riemannian metric g∞ on M∞ of C 1,α -class which is a limit of metrics on Mk . We can prove that Mk converges to (M∞ , g∞ ) in Gromov–Hausdorff distance. Hence (M∞ , g∞ ) is isometric to X. Theorem 3.4 follows. We next prove Theorem 3.3. We assume that the theorem is false. Then there exist M1,k , M2,k ∈ Mn (D, v) such that dH (M1,k , M2,k ) < 1/k but M1,k is not diffeomorphic to M2,k . We use Theorem 3.3 to show that, after taking a subsequence, M1,k , M2,k converges to X1 , X2 , respectively. By Theorem 3.4, X1 , X2 are Riemannian manifolds of C 1,α -class. By using the center of mass technique we will explain in the next section, we can prove that M1,k is diffeomorphic to X1 and M2,k is diffeomorphic to X2 for large k. On the other hand, since the Gromov–Hausdorff distance between X1 and X2 is zero, it follows that X1 is isometric to X2 . Hence X1 is diffeomorphic to X2 . This is a contradiction. 8. Center of mass technique In Section 6 we explained how the isotopy extension theorem can be used to construct a homeomorphism. In fact the isotopy extension theorem is very difficult to prove. We can use a method called the center of mass technique which simplifies those points. The center 17 The author follows the argument of [90] here.

Metric Riemannian geometry

213

of mass technique can be applied to various other problems, for example, to group actions. In this section we explain it. Let us start the explanation of the center of mass technique by beginning a proof of (a modified version of) Proposition 6.3. In Proposition 6.3, the assumption is about the exponential map Expp or coordinate transformation of geodesic coordinates. We actually use the case of harmonic coordinates. So we consider $ the following situation. $ (a) M = i ϕpi (D n ()), N = i ψqi (D n ()) are open coverings. (b) The intersection pattern of coordinate neighborhoods coincide to each other. Namely ϕpi (D n ()) ∩ ϕpj (D n ()) = ∅ if and only if ψqi (D n ()) ∩ ψqj (D n ()) = ∅. (c) If ϕpi (D n ()) ∩ ϕpj (D n ()) = ∅, then ϕpi (D n ()) ⊆ ϕpj (D n (r)). (d) The C 2,α -norm of the coordinate transformation ◦ ϕpi : D n () → Rn Φij = ϕp−1 i is bounded uniformly above by C. The same holds for ◦ ψqi : D n () → Rn . Ψij = ψq−1 i (e) Φij is close to Ψij in C 1 -norm. Our purpose is to construct a diffeomorphism F : M → N under these assumptions. For each x ∈ ϕpi (D n ()), we put Fi (x) = ψqi ◦ ϕp−1 (x) ∈ N. i

(8.1)

This corresponds to what we defined F on each coordinate chart ϕpi (D n ()). The main point is whether we can glue them to obtain F globally. Namely in case x ∈ ϕpi (D n ()) ∩ ϕpj (D n ()) we need to know whether ψqi ◦ ϕp−1 (x) = ψqj ◦ ϕp−1 (x) i j ?

(8.2)

or not. It is easy to see that (8.2) does not hold. What follows from our assumption (assumption of Proposition 6.3 or the assumption (e) above) is (x), ψqj ◦ ϕp−1 (x) < (8.3) d ψqi ◦ ϕp−1 i j (where is a sufficiently small positive number). (More precisely, (8.3) is in C 0 -norm, but assumption (e) is in C 1 -norm.) The basic idea of the center of mass technique is to take the average of Fi (x) over i with x ∈ ϕpi (D n ()). Before we continue the proof of Proposition 6.3, we explain the center of mass technique in general here. Let m a Borel probability measure on M (namely a measure on M with m(M) = 1). Let us denote the support of m by Supp(m). We define a function dm on M by dm (x) = d(x, p) dm(p). (8.4)

214

K. Fukaya

P ROPOSITION 8.1. We assume √ the injectivity radius of M is larger than 10. We also assume KM κ and 20 < π/ κ.18 If the diameter of Supp(m) is smaller than , then on B3 Supp(m), M = x ∈ M | d x, Supp(m) < 3 , the function dm is convex. Here a function on a Riemannian manifold said to be convex if its restriction to each geodesic is convex. We can √ prove Proposition 8.1 by using the convexity of the distance function dp on Bp (π/ κ, M).19 Now we assume that the diameter of Supp(m) is smaller than . Then outside B3 (Supp(m), M) the value of the function dm is greater than 3, and on Supp(m) the value of the function dm is smaller than . Therefore Supp(m) attains its minimum on the interior of B3 (Supp(m), M). Since dm is convex there, the minimum is attained at unique point. D EFINITION 8.1. The center of mass is the point where dm attains its minimum. We write center of mass by CM(m). We remark that if M = Rn , then CM(m) = x dm(x). Rn

We go back to the $ proof of Proposition 6.3. We take a partition of unity χi associated to the covering M = i Bpi (, M). We define a measure F(x) on N by F(x) =

χi (x)δFi (x) .

i

Here δFi (x) is the delta measure supported at Fi (x) and the summation is taken over all i with x ∈ Bpi (, M). By (8.3) we have Diam(Supp(F(x))) < . Let F (x) be the center of mass of F(x). Namely, χi (x)δFi (x) . F (x) = CM F(x) = CM

(8.5)

i

It is easy to see that F (x) is a continuous function of x. Actually it is smooth. (We can prove it by using implicit function theorem.) We can prove that it is a diffeomorphism by using the following lemma. 18 In case κ 0 the second condition is void. 19 This fact is a consequence of Toponogov’s comparison Theorem 4.7.

Metric Riemannian geometry

215

L EMMA 8.2. If Fi , i = 1, 2, . . . , are C 1 -close to each other, then F , determined by (8.5), is C 1 -close to Fi . The proof is elementary. Then, to prove Proposition 6.3, we only need to show that F is injective. Suppose F (x) = F (y), x = y. By using the fact that the Jacobi matrix of F is invertible, we can show that x cannot be close to y. On the other hand, since F is close of Fi and since Fi is injective, we can prove that x cannot be far from y. This is a contradiction. This is an outline of the proof of Proposition 6.3. There are various other applications of the center of mass technique. Let us mention another application of it, that is an application to group actions. Let M be a Riemannian manifold on which G acts. For simplicity we assume G is a finite group. We assume G has two different actions on M and write them as ψ1 : G → Diff(M) and ψ2 : G → Diff(M). We assume that there exists C such that for each element g ∈ G, the C 2 norm of ψ1 (g), ψ2 (g) are smaller than C. P ROPOSITION 8.3 (Grove–Karcher). There exists a constant depending only on C, the dimension n, the injectivity radius of M, and the maximum of the absolute value of the sectional curvature of M, with the following property. If d(ψ1 (g)(x), ψ2 (g)(x)) < for each g ∈ G, x ∈ M, then there exists a diffeomorphism φ : M → M such that φ(ψ1 (g)(x)) = ψ2 (g)(φ(x)). See [77] for its proof. ([77] is the paper where the center of mass technique first appeared.) Proposition 8.3 is applied to study Riemannian manifold whose sectional curvature is close to 1 but whose fundamental group is not necessary trivial.

9. Embedding Riemannian manifolds by distance function In the last section we explained the center of mass technique which we can use to construct a diffeomorphism. In Section 6 we mentioned another way, that is to use isotopy theory. In this section, we discuss the third method which was introduced and used by Gromov [68, 69]. In [53] the author remarked that this method can be used to construct a smooth map (projection of a fiber bundle) in collapsing situation (Theorem 11.2). It was further generalized by Yamaguchi [153] (Theorem 11.3) to the case when we assume a bound of sectional curvature from below (but not above). We here explain an alternative proof of Theorem 3.3. This proof is completed by Katsuda [93] based on an idea of Gromov [69]. We assume M, N ∈ Mn (D, v), dH (M, N ) < (n, v, D). (We choose = (n, v, D) > 0 later.) Let ψ : M → N be a 3 Hausdorff approximation. We take a 20-net X of M. We can take X such that if x, x ∈ X, x = x ,

then d(x, x ) > 10,

(∗)

216

K. Fukaya

in addition. It is easy to see that ψ(X) is an 30-net of N . It is also easy to see that if x, x ∈ X, x = x ,

then d ψ(x), ψ(x ) > .

(∗∗)

We denote by [0, 1]X the set of all maps X → [0, 1]. It is a finite-dimensional Euclidean space. The idea is to embed M (respectively N ) in [0, 1]X using the distance function from X (respectively ψ(X)). In order to go around the trouble that the distance function is not differentiable, we proceed as follows. We take so that it is much smaller than the injectivity radius of M and N . We next take χ : R>0 → [0, 1] such that χ(t) =

0, t < C, t, t ∈ [C 2 , C 3 ], const, t > C 4 .

Here C is a sufficiently large positive number which will be determined later. We may assume that C 5 is smaller than the injectivity radius of M and N . (Precisely we first choose C and then choose so that this condition is satisfied.) Then we define IM : M → [0, 1]X by IM (p)(x) = χ(d(p, x)) and IN : N → RX by IN (p)(x) = χ(d(p, ψ(x))). Note χ(t) is a constant where t is larger than the injectivity radius. Hence IM , IN are smooth. We can prove the following L EMMA 9.1. (1) IM , IN are smooth embeddings. (2) IM (M) is contained in a tubular neighborhood U (N ) of IN (N ). (3) We identify U (N ) with a normal bundle and let π : U (N ) → N be the projection of the normal bundle. Then the restriction of π to IM (M) is a diffeomorphism. We omit the details of the proof (see [93]), but explain briefly its idea. The reason that (1) holds is that, for each p, there are sufficiently many points q ∈ X with d(q, p) ∈ [C 2 , C 3 ]. Namely using the distance function from such q we can show the Jacobi matrix of IM , IN are invertible in a neighborhood of p. To prove (2) we observe that, if x ∈ X ⊂ M, then the distance between IM (x) and IN (ψ(x)) is small. (Namely it is something like const dH (M, N ) = const .) Moreover, since X, ψ(X) are enough dense in M, N , it follows that IM (M) are sufficiently close to IN (N ). We next need an estimate of the size of the tubular neighborhoods of IM (M), IN (N). This follows from the estimate of the second derivative of IM and IN , which turn out to be a consequence of the assumption on curvature of M, N . To carry out the actual proof we need to estimate the size of the tubular neighborhood and the distance between IM (x) and IN (ψ(x)) more precisely. To prove (3) we need to see that the Jacobi matrix of the restriction of π : U (N ) → N to IM (M) is invertible. This follows from the fact that IM (M) is C 1 -close to IN (N ), namely they are close to each other together with their tangent spaces. Since the derivative of the distance function is written in terms of the angle between edges of geodesic triangles, we can prove this fact by using comparison theorems.

Metric Riemannian geometry

217

Theorem 3.3 follows immediately from Lemma 9.1. Remark 9.1. (A) We used distance functions in the discussion above. We can use eigenfunctions of the Laplace operator (or Green kernel) instead. Then the estimate about the derivatives of the diffeomorphism we get becomes better. (See, for example, [16,56,91].) This approach is closely related to harmonic coordinates. (B) We took a net and embed Riemannian manifolds to a finite-dimensional Euclidean space in the above argument. We can use distance functions from all the points and can embed Riemannian manifolds to a Hilbert or Banach space. This argument is useful for a generalization of Theorem 3.3 to an equivariant version. (Namely in the situation when a Lie group acts on M, N .) If we use the eigenfunction of the Laplace operator as we mentioned in (A), embedding to finite-dimensional Euclidean space is good enough to show the equivariant version also.

10. Almost flat manifold In this section we start discussing the case when the injectivity radius goes to zero. In the earlier sections, we began with sphere theorems and applied the method appeared there to finiteness theorems, etc. In sphere theorems, we study manifolds of positive curvature. We here consider another typical Riemannian manifold that is a flat manifold. We first recall the following famous T HEOREM 10.1 (Bieberbach). If M is a compact Riemannian manifold with KM ≡ 0, then there exists a finite covering M˜ of M such that M˜ is isometric to a flat torus. We want to study a Riemannian manifold (M, g) whose curvature is close to zero. To obtain a nontrivial result, we need some normalization. (In fact, the curvature of (M, kg) tends to 0 as k → ∞ for any (M, g).) To normalize volume is not good enough either. (For example, M × S 1 for any M carries a metric with volume 1 and curvature arbitrary small.) So let us normalize the diameter to 1. In other words, we assume |KM | Diam(M)2 is small. We call such manifold almost flat manifold. However the assumption |KM | Diam(M)2 small does not imply that M is diffeomorphic to a flat manifold. E XAMPLE 10.1. We consider the group N of all 3 × 3 matrix of the form

1 x 0 1 0 0

z y 1

.

We consider a left invariant metric g on N such that g = 2 dx 2 + 2 dy 2 + 4 dz2 at the unit matrix I . Let E1 , E2 , E3 be left invariant vectors such that E1 = ∂/∂x, E2 = ∂/∂y, E3 = ∂/∂z at I . It is well known that the curvature of a Lie group with left invariant metric is calculated as follows. If E, F are left invariant orthonormal vectors then the sectional curvature of the plane spanned by them is not greater 6[E, F ]. (See [24].)

218

K. Fukaya

Hence the sectional curvature K(N,g ) is bounded as → 0. On the other hand, we consider the subgroup NZ consisting of matrix in N such that x, y, z ∈ Z. NZ is a discrete subgroup of N and the quotient space M = NZ \ N is known to be compact. We consider the metric on M induced by g and denote it by g¯ . It is easy to see that the diameter Diam(M, g¯ ) goes to zero. Hence Diam(M, g¯ )2 Kg goes to zero. However no finite cover of M is diffeomorphic to T 3 . This example shows that we need to include not only Abelian but also nilpotent Lie groups to characterize almost flat manifolds. T HEOREM 10.2 (Gromov [68]). There exists n > 0 such that if an n-dimensional compact Riemannian manifold M satisfies |KM | Diam(M)2 < n , then M has a finite cover M˜ which is diffeomorphic to Γ \ N , where N is a nilpotent Lie group and Γ is a discrete subgroup. There is an improvement of Theorem 10.2 due to Ruh [133]. Let N be an nilpotent Lie group. There exists a connection ∇can of T N which is invariant to both left and right actions of N . Let Γ be a discrete subgroup of N . ∇can induces a connection on Γ \ N which we denote by the same symbol. (We remark that ∇can is not equal to the Levi-Civita connection.) Let Λ be a finite subgroup of Aut(Γ \ N, ∇can ). We call Λ \ (Γ \ N ) an infranilmanifold. T HEOREM 10.3 (Ruh). Under the assumption of Theorem 10.2, M is diffeomorphic to an infranilmanifold. Let us sketch some of the essential ideas behind the proof of Theorem 10.2. One important origin is Margulis’ lemma. Margulis’ lemma first appeared in the study of discrete subgroup of Lie group. T HEOREM 10.4 (Zassenhaus, see [69, 8.44]). For each Lie group G there exists a neighborhood U of the unit, such that if Γ ⊂ G is a discrete subgroup then U ∩ Γ generates a nilpotent subgroup. The proof is based on the following fact. Let g1 , g2 ∈ G be in a neighborhood of the unit 1, then d 1, {g1 , g2 } Cd(1, g1 )d(1, g2 ).

(10.1)

Here {g1 , g2 } is the commutator. This formula (10.1) is a consequence of the fact that the derivative of (g1 , g2 ) → {g1 , g2 } at 1 is zero. Once we have (10.1) we can prove Theorem 10.4 as follows. We choose U small enough such that if g ∈ U then d(1, g) < 1/(2C). Then (10.1) implies that if gi ∈ U , then d(1, {g1 , g2 }) is strictly smaller than d(1, gi )/C. We repeat this and find that N hold commutator between elements of U is in the 1/C N neighborhood of 1. Since Γ is discrete, it implies the existence of N such that any N hold

Metric Riemannian geometry

219

commutators between elements of U ∩ Γ are trivial. It follows that U ∩ Γ generates a nilpotent group. There are various Riemannian geometry versions of Theorem 10.4. The following, which is proved by Cheeger–Colding [31] (improving [59]) is one of the strongest versions. T HEOREM 10.5. There exists n with the following properties. Let M be an n-dimensional complete Riemannian manifold with RicciM −(n − 1) and p ∈ M. Then the image of π1 (Bp (n , M)) → π1 (Bp (1, M)) has nilpotent subgroup of finite index. If we apply it to the situation of Theorem 10.2 we find that the fundamental group of M has nilpotent subgroup of finite index. (See Section 19 for more discussions on the fundamental group.) Another idea applied by Gromov to prove Theorem 10.2 is to use local fundamental pseudogroup, which we discuss briefly here. (See [57, Section 7] and [24] for its precise definition.) Let M be a complete Riemannian manifold. We assume KM 1. Let p ∈ M. Then by Theorem 4.5 the exponential map Expp : Tp M → M is an immersion on the ball B0 (π; Tp M). Since B0 (π; Tp M) has a boundary, Expp : B0 (π; Tp M) → M is not a covering map. So we cannot consider its deck transformation group in the usual sense. But we can define a “pseudogroup” in the following way. Let < π/10. We consider the set of all loops : S 1 → Bp (, M) with (0) = p and || < . We say ∼ for such , if there exists a based homotopy t between them such that |t | < for each t. Let us denote the set of equivalence classes by π1 (M, p; ). The loop sum ∗ on π1 (M, p; ) is not necessary defined. But when it is defined, its ∼ equivalence class is well defined. (We need to use the fact that Expp : B0 (π; Tp M) → M is an immersion to show this.) When loop sum is well defined it is associative. (Here the reader may find some flavor of Klingenberg’s argument we mentioned at the end of Section 4.) Thus (π1 (M, p; ), ∗) is something similar to a group. We call it a fundamental pseudogroup. The following pseudogroup version of Margulis’ lemma is used in the proof of Theorem 10.2. L EMMA 10.6. If |KM | 1 and if Diam(M) < n , then there exists a subpseudogroup (π10 (M, p; ), ∗) of (π1 (M, p; ), ∗) such that (π10 (M, p; ), ∗) is embedded ( preserving ∗) into a nilpotent Lie group N , its image generates a discrete subgroup Γ and that the index [π1 (M, p; ) : π10 (M, p; )] is estimated by a number depending only on n. Here n - - 1. Lemma 10.6 is the main part of the proof of Theorem 10.2. (Actually we need a bit more. Namely we have to show that the action of (π1 (M, p; ), ∗) to B0 (π; Tp (M)) is diffeomorphic to an action to U ⊂ N of some subpseudogroup Γ ∩ U , where N is a nilpotent Lie group and Γ is its discrete subgroup.) For the details of the proof, we refer to [24,57].

220

K. Fukaya

11. Collapsing Riemannian manifolds—I Using Theorems 3.3, 3.4, we can describe a sequence of n-dimensional Riemannian manifolds Mi with |KMi | 1 and Vol(Mi ) v > 0 where v is independent of i. Namely the limit X (which exists after taking a subsequence) is a Riemannian manifold of C 1,α -class and X is diffeomorphic to Mi for sufficiently large i. In Section 10, we considered a sequence of Riemannian manifolds Mi with |KMi | 1 and Diam(Mi ) → 0. (The second condition is equivalent to saying that Mi converges to a point.) Theorem 10.2 implies that Mi is an infranilmanifold for large i. These are two extremal cases. We now discuss the intermediate case. Namely we consider the case when a sequence of Riemannian manifolds Mi converges to a metric space X (with respect to the Gromov–Hausdorff distance) such that n > dim X > 0. We say that such sequence collapses to X. Here we discuss results under the assumption |KMi | 1. (The study under weaker assumption is discussed in later sections.) We first explain some examples of collapsing Riemannian manifolds. The first example is due to Berger and is called the Berger sphere. E XAMPLE 11.1. We consider the Hopf fibration π : S 3 → S 2 . (Namely we regard S 3 = {(z1 , z2 ) ∈ C2 | |z1 |2 + |z2 |2 = 1}, and we associate to (z1 , z2 ) the complex onedimensional space spanned by it, which is an element of CP 1 = S 2 .) We put the standard metric on S 3 and regard S 2 as a sphere of radius 1/2. It is easy to see that π is a Riemannian submersion. (Namely if Vh ∈ Tp S 3 and V is perpendicular to the fiber of π containing p, then gS 3 (Vh , Vh ) = gS 2 (π∗ Vh , π∗ Vh ).) We define a metric g on S 3 as follows. Let V , W ∈ Tp S 3 . We write V = Vh + Vv ,

W = Wh + Wv ,

where Vh , Wh are perpendicular to the fiber (with respect to gS 3 ) and Vv , Wv are tangent to the fiber. We set g (V , W ) = gS 3 (Vh , Wh ) + 2 gS 3 (Vv , Wv ). 3 2 It is easy to see that limGH →0 (S , g ) = (S , gS 2 ). We can check that the sectional curvature 3 of (S , g ) is between 0 and 1 if ∈ (0, 1].

We can generalize this construction and prove the following P ROPOSITION 11.1. Let M be a compact manifold on which a torus T m acts. We assume that there is no point p on M such that p is fixed by all the elements of T m . Then there exists a family of metrics g on M such that Kg is bounded from below and above and m that limGH →0 (M, g ) = M/T . To find such a sequence of metrics, we first take a T m invariant Riemannian metric gM on M. We next take X an element of the Lie algebra of T m such that the subgroup ∼ =R

Metric Riemannian geometry

221

generated by X is dense in T m . We regard X as a (Killing) vector field on M. We remark that X never vanishes on M. For V , W ∈ Tp M we put V = Vh + c(V )X(p),

W = Wh + c(W )X(p),

where gM (Vh , Xp ) = gM (Wh , Xp ) = 0. We then define g (V , W ) = gM (Vh , Wh ) + 2 c(V )c(W )gM X(p), X(p) . We can prove that the limit of (M, g ) as → 0 is M/T m with quotient metric and the sectional curvature of (M, g ) is bounded for ∈ (0, 1]. Let us take, for example, M = S 3 . We can find an action of T 2 on S 3 satisfying the assumption of Proposition 11.1. Hence there exists a sequence of metrics on S 3 such that the limit is S 3 /T 2 = [0, 1], the interval. In particular the limit space is not a manifold. This construction is further generalized in [38] (Theorem 12.1). There are two approaches to study collapsing Riemannian manifolds under the assumption |KMi | 1. One is due to Cheeger–Gromov [39,38], the other is due to the author [53, 55,56]. These two approaches are unified in [34]. In this section, we discuss the second approach and in the next section we discuss the first (and the joined) approach. Here we discuss the following two problems. For n, D, we denote by Mn (D) the set of all isometry classes of n-dimensional Riemannian manifolds M such that |KM | 1, and Diam(M) D. P ROBLEM 11.1. Let Mi ∈ Mn (D) and X = limGH i→∞ Mi = X. (1) What kind of singularity can X have? (2) Describe the relations between X and Mi . We remark that if we replace Mn (D) by Mn (D, v), the answers are Theorems 3.3, 3.4. Problem 11.1 will be studied also under milder assumptions on curvature later. We first discuss Problem 3.4(2) in the special case when X is a smooth manifold. T HEOREM 11.2 (Fukaya [53,56]). Let Mi ∈ Mn (D). Suppose B = limGH i→∞ Mi is a smooth Riemannian manifold. Then, for large i, there exists a fiber bundle πi : Mi → B with the following properties: (1) The fiber is diffeomorphic to an infranilmanifold F . (2) The structure group is the group of affine transformations Aff(F, ∇can ), where we define the affine connection ∇can on F as in the last section. (3) πi is an almost Riemannian submersion in the following sense. If V ∈ Tp (Mi ) is perpendicular to the fiber then 1 − i

0 and n ∈ Z+ , there exists ρ = ρ(, n) > 0 with the following properties. Let (M, g) be a complete n-dimensional Riemannian manifold with |Kg | 1. Then there exists a metric g and Up , U˜ p , Γp , Np for each p ∈ M such that: (1) Np is nilpotent. Γp ⊂ Np is a discrete subgroup such that π0 (Np ) is finite and Np is generated by its connected component Np,0 and Γp .

228

K. Fukaya

(2) Up is a neighborhood of p and Up ⊇ Bp (ρ, M). (3) Np acts on (U˜ p , g˜ p ) by isometry. Here U˜ p is a covering space of Up and g˜ p is the metric induced by g . ˜ = p then iU˜ p (p) ˜ > ρ. (4) If p˜ ∈ U˜ p and [p] (5) [Γp : Γp ∩ Np,0 ] < k. (6) For any x ∈ U˜ p Diam(Γp \ Np x) < . Here Np x is an Np -orbit. Moreover we have (7) e− g < g < e g. (8) |∇ g − ∇ g | < . (9) |∇ g Rg | < c(n, i, ), where Rg is the curvature tensor of g and c(n, i, ) depends only on n, i, . Remark 12.2. The existence of g satisfying (7)–(9) is proved by [15,1]. Remark 12.3. The metric satisfying (1)–(6) is called a (ρ, k)-round metric in [34]. We remark that at the point where iM (p) > ρ we may take Np = 1 and U˜ p = Up . Hence the statement above is obvious. On the other hand, condition (4) implies that at the point p where injectivity radius is small, the group Np is nontrivial. Hence, together with (1), we obtain a local action of a torus by restricting the action of Np to the center. Using (6) and the fact that the local action of the torus is compatible with the metric g , we can prove that these actions are compatible in the sense of Theorem 12.2. Moreover, in the case when the diameter of M is smaller than a constant depending only on and n, we can prove that the group Np is independent of the choice of p. Hence its orbits defines a foliation on the frame bundle of M. It implies Theorem 11.5. Thus Theorem 12.7 unifies two approaches for collapsing Riemannian manifolds. The proof of Theorem 12.7 is a combination of the proofs of Theorems 12.2 and 11.5. We use Theorem 11.5 and its proof (together with some improvement) to find Np locally. We then glue them in a way similar to the proof of Theorem 12.2. Finally we take the average and obtain the required metric g . E XAMPLE 12.2. Let Γ be a lattice of a semisimple Lie group G of noncompact type and G/K be a symmetric space. We assume Γ \ G/K is noncompact. Then for each p ∈ G/K the group Γp = {g ∈ Γ | d(p, gp) < } has nilpotent subgroup Γp,0 of finite index [Γp : Γp,0 ] < k by Theorem 10.4. (We remark that Γp may not be contained in a small neighborhood of the unit in G. But its subgroup of finite index is in a small neighborhood of the unit.) The Zariski closure Np ⊂ G of Γp,0 is a nilpotent group. This is our Np . The original metric (the metric of symmetric space) is an invariant of left the Np action. Hattori [84] found the following. Let M = Γ \ G/K be a locally symmetric space of noncompact type. We assume that it is noncompact and of finite volume. Then the limit (M, gM /R) as R goes to infinity is a cone of a simplicial complex T which is called the Tits building. (Here the limit is taken with respect to the pointed Gromov–Hausdorff distance (Definition 16.3).) Now if we take a simplex of T then a “neighborhood” of it in M is diffeomorphic to × [0, ∞) × Γ () \ N (). The dimension of the nilmanifold

Metric Riemannian geometry

229

Γ () \ N () depends on . They are glued appropriately, which gives a structure as in Theorem 12.7. The following addendum to Theorem 12.7 is useful for various applications. P ROPOSITION 12.8 [132]. If a KM b in Theorem 12.7 then we may choose g so that a + KM b − .

13. Collapsing Riemannian manifolds—III In this section, we review some of the applications of collapsing Riemannian manifolds. We recall that Mn (D) is the set of isometry classes of n-dimensional Riemannian manifold M with Diam(M) D, |KM | 1. T HEOREM 13.1 (Fang–Rong [51], Petrunin–Tuschmann [126]). For each n, D the number of diffeomorphism classes of simply connected manifolds M in Mn (D) with finite π2 (M) is finite. T HEOREM 13.2 (Fang–Rong [51], Petrunin–Tuschmann [126]). There exists i(n, δ) > 0 such that if M is simply connected, π2 (M) is finite and if 1 KM δ > 0, then the injectivity radius of M is larger than i(n, δ). We remark that, in case the dimension is even, Theorem 13.2 follows from Theorem 4.3 without assumption on π2 . E XAMPLE 13.1. We first consider the Lens space S 3 /Zp where Zp ⊂ S 1 is a cyclic group of order p. Its curvature is 1 and its limit is S 2 = S 3 /S 1 . This example shows the assumption on π1 (M) is necessary both in Theorems 13.1, 13.2. The three examples below show that the assumption on π2 (M) is also necessary in Theorems 13.1, 13.2. We consider the Lie group SU(3). It has a metric with positive sectional curvature. We consider its maximal torus T 2 ⊆ SU(3). Let pi , qi be coprime integers such that lim pi /qi = α ∈ R \ Q. We identify T 2 = R2 /Z2 and let x, y be coordinates of R2 . We consider Si1 = {[x, y] ∈ T 2 | y = pi x/qi }. We equip Mi = Si1 \ SU(3) with quotient Riemannian metric. Mi is a sequence of 7-dimensional manifolds of positive curvature. Using the fact lim pi /qi is irrational, we can easily find that the limit of Mi with respect to the Gromov–Hausdorff distance is T 2 \ SU(3). We can also prove that the sectional curvature of Mi is uniformly positive. Namely C KMi δ > 0 for some δ, C independent of i. (This is a consequence of the fact that pi /qi converges. We remark that π2 (Mi ) ∼ = π1 (S 1 ) = Z.) In a similar way we can use the T 2 -action to S 3 × S 3 to get a sequence of metrics gi on 2 S × S 3 with C Kgi δ > 0 such that (S 2 × S 3 , gi ) converges to S 2 × S 2 . We next consider an action of T 2 × T 2 on SU(3) where the first factor acts by left multiplication and the second factor acts by right multiplication. Using an appropriate family of

230

K. Fukaya

Si1 ∼ = S 1 ⊆ T 2 × T 2 , Petrunin–Tuschmann [126] (using Eschenburg [50]) found an example of Mi = Si1 \ SU(3) with C KMi > δ > 0 such that Mi converges to T 2 \ SU(3)/T 2 . Remark 13.1. A similar π2 -assumption as in Theorem 13.2 was proposed by the author in [57, Remark 15.10]. However [57, Conjecture 15.7] (by the author) turns out to be false. A counterexample (due to Petrunin–Tuschmann) is the last example in Example 13.1. We now sketch the proof of Theorem 13.1. We start with the following L EMMA 13.3 (Rong [132]). If we assume that π1 (M) is finite in the situation of Theorem 11.5 in addition, then the fiber of π : F Mi → Y in Theorem 11.5 is diffeomorphic to a flat manifold. Using the fact that the fundamental group of Mi is finite (here we assume dim Mi > 2), it follows easily that the fundamental group of the fiber has an index finite Abelian subgroup. Since the fiber is an infranilmanifold the lemma follows immediately. Lemma 13.3 implies that we have an F -structure whose orbits are fibers. (Here our F -structure is one called pure F -structure by Cheeger–Gromov [38]. A pure F -structure is an F -structure such that all the orbits of the local action have the same dimension.) We next apply the averaging process in the proof of Theorem 12.7 to the situation of Theorem 11.5 and of Lemma 13.3. Then we have L EMMA 13.4. In the situation of Lemma 13.3, we can approximate the Riemannian metric on F Mi by g in the same sense as Theorem 12.7(7)–(9) so that g is an invariant of the local T k action and of the O(n) action. Now we start the proof of Theorem 13.1. We assume that Theorem 13.1 is false. Then there exists a sequence Mi ∈ Mn (D) such that Mi is simply connected, π2 (Mi ) is finite, and Mi is not diffeomorphic to Mj for i = j . We take F Mi and may assume that it converges to Y . Since we approximate the metric by one satisfying Theorem 12.7(7)–(9), it follows that Y is a smooth Riemannian manifold. We may replace F Mi by its finite cover F˜ Mi so that it has global a T k × G-action, where G is a compact group20 and a T k -orbits are the fibers of the fibration F˜ Mi → Y . We modify the metric of F˜ Mi so that it is T k × Gequivariant. The next lemma is the place where we use the key assumption that π2 (Mi ) is finite. L EMMA 13.5. If F˜ Mi /T k is G-diffeomorphic to F˜ Mj /T k then F˜ Mi is T k × Gdiffeomorphic to F˜ Mj . In fact the torus bundle T k → E → B is determined by the (T k -analogue of) Euler class ∈ Hom(H2 (B), π1 (T k )) (which is well defined up to Aut(π1 (T k ))). In our case where π2 (F˜ Mj ) is finite and π1 (F˜ Mj ) is trivial, the Euler class is an isomorphism 20 Actually it is finite covering group of O(n). (It may be disconnected.)

Metric Riemannian geometry

231

H2 (B)/Tor → π1 (T k ), hence it is unique up to Aut(π1 (T k )). To obtain the T k -equivariant diffeomorphism F˜ Mi → F˜ Mj which is G-equivariant also, we use the center of mass technique (Proposition 8.3). We remark that F˜ Mj /T k has the same dimension as Y and F˜ Mj /T k converges to Y with respect to the G-Gromov–Hausdorff topology (which was introduced in [52]). Estimate (9) of Theorem 12.7 implies that Y is a smooth manifold. On the other hand, the sectional curvature of F˜ Mj /T k is bounded from below. Hence Theorem 11.3 implies that F˜ Mj /T k is diffeomorphic to Y for large i. We can use the G-equivariant version of Theorem 11.3 (which can be proved in the same way as Theorem 11.3 using an embedding to Hilbert space as in [55]21 ), F˜ Mi /T k is G diffeomorphic to Y for large i. Hence Lemma 13.5 implies that F˜ Mj is G diffeomorphic to F˜ Mj for i, j large. Namely Mi is diffeomorphic to Mj . This is a contradiction. To prove Theorem 13.2 we need another result by Petrunin–Rong–Tuschmann. T HEOREM 13.6 [125]. Let M be a compact manifold. We assume that M admits a sequence of metrics gi . We assume that Λ Kgi λ and that the metric space X = limGH i→∞ (M, gi ) is of dimension strictly smaller than M. We also assume that the distance function di : M × M → R induced by gi converges to a function d which determines a pseudometric22 on M. Then λ 0. Remark 13.2. Klingenberg and Sakai conjectured a similar statement, but their conjecture does not assume the additional assumption that di converges to a pseudometric. To prove Theorem 13.2 using Theorem 13.6 we proceed as follows. We assume that there exists Mi with 1 KKi δ > 0 and that the injectivity radius goes to 0. We can discuss in the same way as in the proof of Theorem 11.3 to show that Mi is diffeomorphic to Mj .23 By looking at the proof carefully we may assume that the diffeomorphism almost preserves distance function. Namely if we identify Mi with Mj then the sequence M = Mi = Mj satisfies the assumption of Theorem 13.6. The conclusion of Theorem 13.6 contradicts to KMi δ > 0. One of the ideas of the proof of Theorem 13.6 is the following observation. If the collapsing occurs in the same way as in the proof of Proposition 11.1, then the sectional curvature of the plain spanned by X and other vector is always converges to zero. To make this simple idea works we need a lot of delicate work which is not described here. We next discuss some other applications of collapsing theory. 21 See Remark 9.1. 22 Namely it satisfies axioms of metric except “d(x, y) implies = y”. 23 We use Proposition 12.8 to show Λ + K Mi λ − .

232

K. Fukaya

T HEOREM 13.7 (Rong [132]). There exists w(n, δ) such that if a compact n-dimensional Riemannian manifold M satisfies 1 KM δ then there exists a cyclic subgroup C of the fundamental group π1 (M) such that [π1 (M) : C] < w(n, δ). Remark 13.3. If we assume 1 KM 0, Diam(M) < D, then there exists an Abelian subgroup C of π1 (M) such that [π1 (M) : C] < w(n, D) [132]. There are results under milder assumption that is the case when M is of almost of nonnegative curvature. See Section 19. The following is another application of collapsing theory. This time we apply to manifolds of almost nonpositive curvature. T HEOREM 13.8 (Fukaya–Yamaguchi [58]). There exists (D, n) such that if a compact n-dimensional Riemannian manifold M satisfies Diam(M) D, (D, n) KM −1 then the universal covering space of M is diffeomorphic to Rn . This is a generalization of Hadamard–Cartan’s theorem (Theorem 4.6), which is the case when KM 0.

14. Morse theory of distance function So far we mainly discussed results assuming the curvature to be bounded from above and below. From this section on, we consider the case when the curvature is bounded from below only. The next theorem is a corollary of Theorem 4.1. T HEOREM 14.1 (Rauch). Let M be a compact manifold without boundary. If there exists a Morse function on M with two critical points, then M is homeomorphic to a sphere. In Section 4, we started with Theorem 4.1 and showed the way to prove sphere theorems, finiteness theorems and compactness theorems by estimating the number of balls we need to cover a manifold. The number of contractible open subsets one needs to cover the space (plus one), is called the Lusternik–Shnirel’man category and is important in Morse theory. In this section we will try to apply Morse theory directly. For a given Riemannian manifold M, a function which is determined automatically from the metric is a distance function dp (x) = d(p, x) from a point. (Note we can use the fact that p = x is the unique critical point of x → dp (x) with d(x, p) < iM (p) to prove that Bp (r, M) is diffeomorphic to a sphere if r < iM (p).) The difficulty to apply Morse theory to the distance function is that x → dp (x) is not differentiable for d(x, p) > iM (p). (dp is not differentiable at p either. But this does not cause serious trouble. We may consider dp2 instead, for example.) During the proof of Theorem 2.5, Grove–Shiohama applied Morse theory away from the ball with radius = iM (p). After that, their method is used in many other places. The main idea of them is the following definition.

Metric Riemannian geometry

233

Fig. 14.1.

D EFINITION 14.1. We say q is a regular point of dp if there exists a nonzero vector V. ∈ Tq M such that for any minimal geodesic : [0, d(p, q)] → M joining p and q, the . angle between d dt (0) and V is not greater than π/2. For example, let p, q be as in Figure 14.1. It is not clear how many minimal geodesics are there joining p with q. But it is easy to see that the direction of any of them is downwards at q. Hence q is a regular point of dp . Remark 14.1. We may consider various situations similar to Definition 14.1. For example, let us consider a continuous function f which is an infimum of finitely many differentiable functions fα locally (namely f = inf fα ).24 In this case we say q is a regular point of f if there exists a vector V. ∈ Tq M such that, for each α with f (q) = fα (q), we have V. (fα ) > 0. We can apply a similar argument to a linear combination of finitely many dp ’s or the infimum of them also. Proposition 14.2 holds for such cases. Based on Definition 14.1, we can prove the following analogue of the Morse lemma for dp . P ROPOSITION 14.2. If q with a dp (q) b is an arbitrary regular point of dp , and if Bp (b, M) is compact, then Bp (b, M) \ Bp (a, M) = {q ∈ M | a d(p, q) b} is homeomorphic to a direct product of ∂Bp (b, M) = {q ∈ M | d(p, q) = b} and [0, 1]. The proof is similar to the proof of the following famous 24 We remark that d may not satisfy this condition in general. p

234

K. Fukaya

T HEOREM 14.3 (Morse lemma). We assume that f : M → R is differentiable, and arbitrary q with f (q) ∈ [a, b] is a regular point of f , and that f −1 ([a, b]) is compact. Then f −1 ([a, b]) is diffeomorphic to f −1 ({a}) × [0, 1]. The proof of Morse lemma uses an integral curve of grad f . (See [103].) Since dp is not differentiable, the vector field grad dp does not make sense. Instead, we will use the vector field V constructed below. For q ∈ Bp (b, M) \ Bp (a, M) let Vq = V be the vector ∈ Tq M as in Definition 14.1. If we can take Vq depending smoothly on q, then we can take the vector field V (q) = Vq in place of − grad f . (The condition in Definition 14.1 implies that dp decreases along the integral curve of V .) To find Vq depending smoothly on q, we proceed as follows. We first take V˜q which may not depend smoothly on q. We extend it to its neighborhood and denote it by the same symbol V˜q . Then if q is in a small neighborhood U (q) of q, then the vector V˜q (q ) ∈ Tq M satisfies the condition of Definition 14.1. We cover Bp (b, M) \ Bp (a, M) by finitely many U (qi )’s. We then take a partition of unity χi and put V (q) =

χi (q)V˜qi (q).

It is easy to see that this V has the required properties. Using this vector field V , the proof of Proposition 14.2 goes in the same way as the proof of Morse lemma. To apply Morse theory of dp to the proof of Theorem 2.5 we need the following lemma. L EMMA 14.4. We assume that M satisfies the assumption of Theorem 2.5. Let p, q ∈ M with d(p, q) = Diam(M), and x ∈ M be a point different from p, q. Let p : [0, d(p, x)] → M, q : [0, d(q, x)] → M be minimal geodesics joining x to p, and x to q, respectively. (In case there are several of them, we assume any of them have the property below.) d d Then the angle between two tangent vectors dtp (0) and dtq (0) ∈ Tx M is greater than π/2 (see Figure 14.2). The proof of Lemma 14.4 uses Toponogov’s comparison theorem (Theorem 4.7). Under the assumption of Lemma 14.4 (that is KM 1/4), Theorem 4.7 implies the following Sublemma 14.5. Let x, y, z ∈ M. We consider the geodesic triangle whose vertices are those three points. We denote the length of its edges by |xy| etc. and angles by xyz etc. We put X = |yz|, Y = |zx|, Z = |xy|. S UBLEMMA 14.5. If zxy π/2, then cos X2 cos Y2 cos Z2 . Note that we have s(Y/2, Z/2, θ, 1) s(Y/2, Z/2, π/2, 1) = cos−1 (cos Y/2 cos Z/2), where s(·, ·, ·, ·) is as in Theorem 4.7.

Metric Riemannian geometry

235

Fig. 14.2.

We start the proof of Lemma 14.4. We put |p | = t, |q | = s, d(p, q) = D. Since dp attains its maximum at q it follows that q is not a regular point of dp . Hence there exists a geodesic joining p and q such that the angle between and q is not greater than π/2. We apply Sublemma 14.5 to the geodesic triangle consisting of , p , q and obtain cos

t s D cos cos . 2 2 2

(14.1)

Since D/2 > π/2 we have cos D2 < 0. Therefore one of cos 2s , cos 2t is positive. We may assume cos 2s > 0. If the angle between p and q is not greater than π/2, then we can again apply Sublemma 14.5 and obtain cos

s t D cos cos . 2 2 2

(14.2)

Since cos 2s > 0, (14.1), (14.2) imply cos

D D s cos2 cos . 2 2 2

We remark that 0 < D/2, s < π .25 This is then a contradiction.

Now Lemma 14.4 implies that if x = p, q then x is a regular point of dp , dq . In fact, let V be the tangent vector of q at x. It follows from Lemma 14.4 that the vector field 25 This is a consequence of Myers’ theorem (Theorem 5.4).

236

K. Fukaya

V satisfies the condition in Definition 14.1. Namely x is a regular point of dp . Now we can use Proposition 14.2 to prove that M is homeomorphic to sphere. Hence we proved Theorem 2.5. We remark that we proved Proposition 4.4 during the proof of Lemma 14.4. In fact, we proved cos t/2 > 0 or cos s/2 > 0 there. It implies t < π or s < π . The method we explained above is very useful to study Riemannian manifolds under the bounds of sectional curvature from below. It is also useful to study Alexandrov spaces (see Sections 17, 18). Theorem 2.5 is a sphere theorem. There are several finiteness theorems corresponding to it. The first one is the following, which is called the Gromov’s Betti number estimate. T HEOREM 14.6 (Gromov [70]). There exists C(n) such that if an n-dimensional compact Riemannian manifold M satisfies KM −κ (κ 0) and if its diameter is D then

rank Hk (M; F ) C(n)1+κD .

k

Here F is an arbitrary field. Note in the case when κ = 0, the right-hand side is independent of D. It follows from Theorem 14.6 that the connected sum of sufficiently many copies of CP 2 does not carry a metric of nonnegative sectional curvature. The proof of Theorem 14.6 is based on Morse theory of a kind of distance function. Namely we use an idea similar to the Morse inequality to estimate the Betti number in terms of the number of critical points. However the proof is more involved since Morse theory of the distance function itself does not work. The actual proof requires more complicated argument, which we omit here. There are many other applications of Morse theory of distance functions to metric Riemannian geometry. For example, Gromov used it to show that a complete manifold M such that 0 > −a 2 KM −b2 and has finite volume is diffeomorphic to the interior of a compact manifold with boundary [67]. Let us add a few more remarks to Theorem 2.5. If we assume 1 KM 1 − in addition in Theorem 2.2, then we can show that M is not only diffeomorphic but is also close to a sphere as a Riemannian manifold. Namely if a sequence of n-dimensional simply connected Riemannian manifolds Mi satisfies 1 KMi 1 − 1/i, then Mi converges to S n with standard metric with respect to the Gromov–Hausdorff distance. On the contrary, the corresponding statement in the situation of Theorem 2.5 does not hold. Namely, let us consider a sequence of Riemannian manifolds Mi such that KMi 1 and that the diameter of Mi converges to π as i goes to infinity. Then Theorem 2.5 implies that Mi is homeomorphic to a sphere. However it is not true that the limit of Mi with respect to the Gromov–Hausdorff distance is isometric to the sphere with standard metric. We remark however a Riemannian manifold with diameter = π and KM 1 (actually weaker assumption Ricci n − 1 is enough) is isometric to the sphere. (Theorem 21.11 [148,41].)

Metric Riemannian geometry

237

Fig. 14.3.

In fact, let us consider the quotient of S 2 by the action of S 2 /Zp generated by the rotation of angle 2π/p around the fixed axis. The quotient is a Riemannian manifold with constant curvature 1, except two points where the axis intersects with S 2 . We approximate the quotient space by a Riemannian manifold with curvature 1 and obtain a sequence of Riemannian manifolds Mi whose diameter converges to π and KMi 1. The limit is S 2 /Zp and is not isometric to the sphere with standard metric. The essential point here is that the Alexandrov space X with diameter = π and KM 1 is not necessary isometric to a sphere with standard metric. (Compare Theorem 23.11.) (See Figure 14.3.) This is related to the fact that the limit of Riemannian manifolds Mi with KMi const is rather different from a Riemannian manifold even in the case when the limit has the same dimension. For example, we consider a boundary S of a convex set in R3 . There is a point of S that has no tangent plane. In the situation when the absolute value of the sectional curvature is bounded, the Gromov–Hausdorff convergence is equivalent to the C 1,α -convergence of the metric tensor (in the situation when the limit has the same dimension), by Theorem 3.4. Therefore the limit space has a tangent space everywhere. By the reason we explained above the following question is yet open. P ROBLEM 14.1. Is there any n > 0 such that if M is an n-dimensional complete Riemannian manifold with KM 1 and Diam(M) π − n , then M is diffeomorphic to a sphere? We remark that in the proof of Theorem 2.5 we consider the distance functions dp , dq simultaneously where p, q lie in the different sides from x. This is similar to the notion strainer used in Alexandrov space. (See Section 17.)

15. Finiteness theorem by Morse theory In this section, we explain the idea of the proof of Theorem 3.5. The first half of it, which was proved in [78], asserts that the number of homotopy classes represented by an element

238

K. Fukaya

of Mn (D, v) is finite. (We recall that M ∈ Mn (D, v) if KM 1, Diam(M) D, and Vol(M) v, dim M = n.) In this section we mainly explain this part. The key of the proof is the following proposition. P ROPOSITION 15.1. There exists = (n, D, v) > 0 such that the following holds for each M ∈ Mn (D, v). Let p, q ∈ M with d(p, q) < , p = q. Then q is a regular point of dp . Moreover we have the following. We put = {(x, x) ∈ M × M | x ∈ M}, () = {(x, y) | d(x, y) < }. Then is a deformation retract of (). The deformation retraction H : () × [0, 1] → () can be chosen so that the length of the curve t → H (p, q, t) is not greater than Cd(p, q). Here C depends only on n, D, v. Using Proposition 15.1, the proof of Theorem 3.5 goes in a way similar to the proof of Proposition 5.5. Namely, from the first half of the Proposition 15.1, we find that the metric balls Bp (, M) of radius are contractible in M. On the other hand, the number of metric balls Bp (, M) we need to cover M is estimated in the same way as in Section 5 by using Proposition 5.2. However since it is not clear whether the intersection of finitely many metric balls Bp (, M) is contractible or not in our case, so we need to modify the proof of Proposition 5.5 a bit. The second half of Proposition 15.1 is used for this purpose. We omit this part of the proof. The proof of Proposition 15.1 is closely related to the proof of Proposition 5.6. So we first sketch the proof of Proposition 5.6. By Theorem 4.9 we only need to estimate the length of closed geodesic of minimal length from below for M ∈ Mn (D, v). Let : S 1 → M be the closed geodesic of length . We take an arbitrary point x ∈ M, and let (t) ∈ (S 1 ) be the point of smallest distance from x. Then is orthogonal to x(t) at (t).26 (Here x(t) is a minimal geodesic joining x and (t).) We put (0) = p. Since d(p, (t)) , it follows that if d(x, p) is sufficiently larger than , then the angle between and xp is close to π/2. We thus have proved the following lemma. L EMMA 15.2. Let δ, ρ > 0. Then there exists depending only on n, D, v, δ, such that if is a closed geodesic with length < and if (0) = p then M is contained in the image of the exponential map of the domain ⊂ Tp M in Figure 15.1. We can choose δ sufficiently small compared to the diameter D, so that the volume of the image of the domain II in Figure 15.1 is smaller than v/2. By choosing ρ small we may assume the volume of the image of the domain I in Figure 15.1 is smaller than v/2 also. Therefore if there exists a closed geodesic of length < , then the volume of M is smaller than v. We turn to the proof of Proposition 15.1. It suffices to show the following lemma. 26 More precisely in case x is a cut point with respect to the geodesic . (The notion of the cut point with respect to a submanifold is defined in a similar way to the notion of cut point from a point. See, for example, [33].) may not be orthogonal to x(t). However this does not cause a trouble for the proof of Lemma 15.2 since the measure of the set of cut points is zero.

Metric Riemannian geometry

239

Fig. 15.1.

Fig. 15.2.

L EMMA 15.3. There exist θ = θ (n, v, D) > 0 and = (n, v, D) > 0 with the following properties. Let M ∈ Mn (D, v), p, q ∈ M, d(p, q) < . Let 1 and 2 are minimal geodesics joining p and q. Then the angle between 1 and 2 at p or q is smaller than π − θ . Lemma 15.3 implies that q in the lemma is a regular point of dp . The first half of Proposition 15.1 follows from Proposition 14.2. The second half can also be proved in the same way by examining the proof of Proposition 14.2 carefully. The proof of Lemma 15.3 is similar to the proof of Proposition 5.6. Namely we replace Figure 15.1 by Figure 15.2. We thus explained an outline of the first half of the proof of Theorem 3.5. The other half is the finiteness of the number of homeomorphism classes and requires another deep argument. The main new technique required is the idea from controlled surgery.

16. Soul theorem and splitting theorem Typical results on noncompact complete Riemannian manifolds of nonnegative curvature are the soul theorem and the splitting theorem. They also are very useful to study the local structure of the Gromov–Hausdorff limit of Riemannian manifolds or its limit. We first explain why the study of noncompact manifolds is useful to study local structure of the limit space. Let us begin with the introducing some notations. Let X be a metric

240

K. Fukaya

space and : [a, b] → X be a continuous map (that is a curve). The length || of is by definition a supremum of the sum d (ti ), (ti+1 ) , where a = t0 < t1 < · · · < tN = b runs over all partitions (N moves also). D EFINITION 16.1. We say that X is a length space if for each p, q ∈ X there exists a curve joining p, q of length d(p, q). A complete Riemannian manifold is a length space. The Gromov–Hausdorff limit of length spaces is also a length space. D EFINITION 16.2. A complete metric space is said to be compactly generated if all of its metric balls are compact. The set of all isometry classes of compact metric spaces is complete with respect to the Gromov–Hausdorff distance. A natural metric to put on the set of all isometry classes of complete compactly generated spaces is pointed the Gromov–Hausdorff distance, which we define below. D EFINITION 16.3. Let X, Y be metric spaces and x ∈ X, y ∈ Y . We say that the pointed Gromov–Hausdorff distance dpGH ((X, x), (Y, y)) between (X, x) and (Y, y) is not greater than , if the Gromov–Hausdorff distance between the metric balls pGH B1/ (x, X) and B1/ (y, Y ) is not greater than . We write limi→∞ (Xi , xi ) = (X, x) if limi→∞ dpGH ((Xi , xi ), (X, x)) = 0. The following can be proved in the same way as Theorem 3.2. T HEOREM 16.1. The set of all isometry classes of a pair (M, p) of an n-dimensional Riemannian manifold M with RicciM −(n − 1) and a point p on it is relatively compact with respect to the pointed Gromov–Hausdorff distance. Now we can define the tangent cone. Let (X, dX ) be a length space and x ∈ X. pGH

D EFINITION 16.4. If the limit limc→∞ ((X, cdX ), x) exists, we call it the tangent cone (at x ∈ X) and write it as Tx X. If X is an n-dimensional Riemannian manifold then the tangent cone of X is isometric to Rn at each point. E XAMPLE 16.1. Let Ω ⊂ Rn be a compact convex set. We put X = ∂Ω and define a length metric on it. (Namely the distance between x, y ∈ X is the infimum of the length of all curves joining x and y in X.)

Metric Riemannian geometry

241

Then tangent cone Tx X is described as follows. We consider every ray (half of the straight line) : [0, ∞) → Rn such that (t) ∈ Ω for small t > 0. The set of such is an open subset of Rn . Its boundary in Rn is the tangent cone Tx X. If the space X is not so wild then we may expect the tangent cone Tx X exists and a neighborhood of x in X is homeomorphic to a neighborhood of the origin (base point) in Tx X. (This holds for Alexandrov spaces, for example. See Theorem 18.1.) Namely we can study the local structure of X by studying the tangent cone Tx X. If X is a Gromov–Hausdorff limit of a sequence of Riemannian manifolds Mi and if the sectional curvature of Mi is bounded from below by a constant independent of i, then we may regard the limit X as the space with “curvature bounded from below”. Then the infimum of the “curvature” of family of length spaces (X, cdX ) as c goes to infinity will become nonnegative. (Note if we multiply the metric by c then the curvature is multiplied by c−2 .) This means that if tangent cone of X exists, then it is of “nonnegative curvature”. (The discussion here is informal and heuristic. So for a moment the curvature may either to the Ricci or the sectional curvature.) This is one of the reasons why the study of noncompact spaces with nonnegative curvature is important in the local theory of spaces which are a limit of Riemannian manifolds. By using Gromov’s precompactness theorem (Theorem 16.1) we have the following P ROPOSITION 16.2. Let Mi be a sequence of Riemannian manifolds with RicciMi > −(n − 1). Let X = limGH i→∞ Mi . Let x ∈ X and ck be a sequence of positive numbers with lim ck = +∞. Then there exists a subsequence of ((X, ck dX ), x) which converges in pointed Gromov–Hausdorff distance. In general ((X, ck dX ), x) itself may not converge. (Namely we need to take a subsequence.) Hence X may not have a tangent cone. This is one of the difficulties to study the family of Riemannian manifolds with Ricci curvature bounded from below. In case X is a limit of Riemannian manifolds with sectional curvature bounded from below (or more generally if X is an Alexandrov space), limc→∞ ((X, cdX ), x) converges without taking a subsequence (Theorem 17.14). Let us now state the soul theorem and the splitting theorem. We first define the notions line and ray. Let X be a length space. A curve : (a, b) → X is called a geodesic if it is length minimizing locally. Namely is a geodesic if, for each t ∈ (a, b), there exists such that d((t − ), (t + )) is equal to the length of the restriction of to (t − , t + ). We use arc length as a parameter in the next definition. D EFINITION 16.5. Let X be a length space. A geodesic : [0, ∞) → X is called a ray if d((t), (s)) = |t − s| for any t, s. A geodesic : (−∞, ∞) → X is called a line if d((t), (s)) = |t − s| for any t, s. (The difference between line and ray is the domain of its definition.) If there exists a tangent cone Tx X = limc→∞ ((X, cdX ), x) then it is a union of its rays such that (0) is the base point. We also have the following

242

K. Fukaya

L EMMA 16.3. Let X be a length space and : (−, ) → X be a minimal geodesic with (0) = x. If the tangent cone Tx X exists, then it contains a line. In fact, since in (X, cdX ) there exists a minimal geodesic of length c containing the origin, its limit in Tx X will be a line. We assume that a complete metric space X is a length space and satisfies one of the following conditions. C ONDITIONS 16.1. (a) X is a Riemannian manifold of nonnegative sectional curvature. pGH (b) X = limi→∞ Mi such that KMi −i , limi→∞ i = 0 and dim X = dim Mi . (c) X is a Riemannian manifold with nonnegative Ricci curvature. pGH (d) X = limi→∞ Mi such that RicciMi −i , limi→∞ i = 0 and Vol(Mi ) v > 0. The next theorem is called the splitting theorem. T HEOREM 16.4. If X satisfies one of the Conditions 16.1 and contains a ray, then X is isometric to a direct product R × X0 . Theorem 16.4 is due to Toponogov [149] in case (a), to Cheeger–Gromoll [35] in case (c), Grove–Petersen [80] and Yamaguchi [153] in case (b) and Cheeger–Colding [28] in case (d). We will explain an idea of the proof of the cases (a), (c) later in this section. (Case (b) is similar to case (a). Case (d) is discussed in Section 23.) We explain more how to apply it to study the local structure of the limit space. Note that we can use Theorem 16.4 repeatedly. Namely if X0 contains a line then we can again apply the theorem and show that it is a direct product. Therefore if we can repeat it dim X times, then we can prove that X = Rn . Lemma 16.3 implies that if x is an interior point of a minimal geodesic, then Tx X contains a line. Therefore if we can find n (= dim X) “independent” geodesic for which x is an interior point, then the tangent cone Tx X is isometric to Rn . This may imply that X is a manifold in a neighborhood of x. This argument appears in Sections 17, 18 and in Sections 20, 22, 23. We next explain an outline of the proof of splitting theorem. The main tool we use is convexity of Busemann function (it is used also in the proof of the soul theorem). Let X be a length space and : [0, ∞) → X be a ray. D EFINITION 16.6. The Busemann function is the limit b (x) = limt→∞ (t − d(x, (t))). P ROPOSITION 16.5. If X satisfies either (a) or (b), then the Busemann function of its ray is convex. If X satisfies (c), then the Busemann function of its ray is subharmonic. In the situation (d) we cannot define subharmonicity in the usual way. So the argument is more involved. See Section 23 and [26,31].

Metric Riemannian geometry

243

The proof of Proposition 16.5 is by a comparison theorem. Namely it follows immediately from the Laplacian and Hessian comparison theorem (Theorem 16.6) for the distance function. We remark that the Hessian Hess f of a function f on a Riemannian manifold is defined by (Hessx f )(V , W ) = V W (f ) − (∇V W )(f )

(16.1)

and is a symmetric bilinear map Tx M ⊗ Tx M → R. A function f is convex if its Hessian Hess f is nonnegative everywhere. The Laplacian f is its trace. Namely f (x) =

n (Hess f )(ei , ei ),

(16.2)

i=1

where ei is an orthonormal basis of Tp M. (We remark that we are using the nonpositive Laplacian. Namely = −(d ∗ d + dd ∗ ).) We say a smooth function is subharmonic if its Laplacian is nonnegative. T HEOREM 16.6. Let M be a Riemannian manifold and p ∈ M. We consider the function dp (x) = d(p, x). (1) If KM κ then Hessx dp

sκ (d(p, x)) (gx − ddp ⊗ ddp ). sκ (d(p, x))

(16.3)

Here ddp : Tx M → R is the exterior derivative of dp . (2) If RicciM κ then f (x) (n − 1)

sκ (d(p, x)) . sκ (d(p, x))

(16.4)

Here sκ is as in (4.1). Remark 16.1. We remark that dp is not differentiable outside the ball Bp (iM (p), M). So we need to be more careful to state Theorem 16.6. Precisely speaking (16.3), (16.4) hold in a barrier sense. See, for example, [26]. We omit the proof of Theorem 16.6. We remark that (16.6) implies Corollary 5.3. In fact d Vol Bp (t, M) = dt

∂Bp (t,M)

grad dp , grad dp Ω∂Bp (t,M)

=

div grad dp ΩM Bp (t,M)

244

K. Fukaya

(n − 1)

d dt

B0 (t,Tp M)

B0 (t,Tp M)

sκ (d(p, x)) ΩRn sκ (d(p, x))

n sκ d(p, x) ΩRn

d Vol Bp0 t, Snκ . dt

Let us explain how we use Proposition 16.5 to prove Theorem 16.4, in cases (a), (b). We assume that X contains a line : R → X. We then have two rays ± : [0, ∞) → X by + (t) = (t),

− (t) = (−t).

We study their Busemann functions b± . The triangle inequality implies b+ (t) + b− (t) 0.

(16.5)

By Proposition 16.5 the right-hand side is convex. Since a bounded convex function is constant, it follows that + (t) + − (t) is constant. (Actually it is 0.) It follows that + (t) = const − − (t) is convex and is concave. Hence its level surface if totally geodesic. (Here we say S ⊂ M is totally geodesic if any minimal geodesic of M joining two points of S are contained in S.) This implies Theorem 16.4. In case (c) we use subharmonicity in place of convexity. We next discuss the soul theorem. T HEOREM 16.7 (Cheeger–Gromoll [36,128]). If a complete Riemannian manifold M has nonnegative sectional curvature then there exists a compact submanifold S ⊆ M without boundary, such that M is diffeomorphic to the normal bundle of S. Moreover S is totally geodesic. We call S the soul of M. The basis of the proof of Theorem 16.7 is Proposition 16.5. It asserts that, for each ray : [0, ∞) → M, the Busemann function b is convex. In particular for any c the closed set H (, c) = x ∈ M | b (x) c is convex. The next lemma is the key of the proof of Theorem 16.7. We fix p ∈ M and let Ray(p) be the set of all rays of M such that (0) = p. L EMMA 16.8. The set Cc (p) =

.

∈Ray(p) H (, c)

is compact.

The proof is by contradiction. Namely we assume that Cc (p) is not compact and let pi ∈ Cc (p) be a divergent sequence. We put d(p, pi ) = ti , and let i : [0, ti ] → M be a minimal geodesic such that i (0) = p, i (ti ) = pi and that it is parameterized by arc length.

Metric Riemannian geometry

245

i Since d dt (0) ∈ Tp M is a unit vector, we may take subsequence so that it converges. Let : [0, ∞) → M be a geodesic such that

d di (0) = (0). i→∞ dt dt lim

Since limi→∞ ti = ∞, it follows that is a ray. On the other hand, we have lim b (pi ) = ∞.

i→∞

This contradicts to pi ∈ Cc (p).

Thus, we obtained a compact convex subset Cc (p) of M. We can find a compact convex submanifold S in it. The argument to do so is rather technical and is omitted. (See [33, Chapter 8].) Perelman [115] proved that if, in the situation of Theorem 16.7, there exists a point where KM > 0, then the soul S is a one point. We refer to [63] for other topics related to the soul theorem. We remark that we already applied Theorem 16.7 in Section 12 to construct on F -structure. 17. Alexandrov space—I In this section and in the next sections, we discuss recent developments [22,119,113] in the theory of Alexandrov space. A good text book on the contents of this section is [138]. (See also [127].) In the following sections, we study compactly generated length spaces of finite Hausdorff dimension only. So we always assume that the length space has this property. The Alexandrov space is a length space with curvature bounded from below. To define the notion of curvature for length space, we use a Toponogov type comparison theorem in the opposite direction. Namely we define the condition KX 1 by using the conclusion of a comparison theorem. However the conclusion of Theorem 4.7 does not (yet) make sense for length space, since it uses angles. So we consider the following slightly different version. We use the notation of Theorem 4.7. Let M be a Riemannian manifold and x, y, z, v, w ∈ M. Let x , y , z , v , w ∈ Sn (κ). We assume v ∈ xy, w ∈ xz, v ∈ x y , w ∈ x z . T HEOREM 17.1 (Alexandrov–Toponogov). We assume KM κ and d(x, y) = d(x , y ), d(x, z) = d(x , z ), d(y, z) = d(y , z ), d(x, v) = d(x , v ), d(x, w) = d(x , w ). u ∈ xy, u ∈ x y , v ∈ xz, v ∈ x y . Then we have d(v, w) d(v , w ). (See Figure 17.1.) D EFINITION 17.1 (Alexandrov). A length space of finite dimension is said to be an Alexandrov space with K κ if the conclusion of Theorem 17.1 holds for X. Remark 17.1. There are several other definitions equivalent to Definition 17.1. We will explain them later (Theorems 17.9 and 17.10).

246

K. Fukaya

Fig. 17.1.

Remark 17.2. There is a notion of Alexandrov space with curvature bounded from above. We do not discuss it in this article. It is proved by Berestovskij that if a length space is an Alexandrov space with curvature bounded from above and below then it is a C 0 Riemannian manifold. This result is related to Theorem 3.4 but was proved earlier than that. See [17]. Hereafter we say Alexandrov space for Alexandrov space with K κ with some κ. The notion of Alexandrov space was introduced by Alexandrov [6] more than 50 years ago. There are several related pioneering works around those old days, like Busemann [23]. In [22], Burago–Gromov–Perelman proved several fundamental theorems on Alexandrov spaces. After that the study of Alexandrov space became very active and important in metric Riemannian geometry. Their main results are T HEOREM 17.2 (Burago–Gromov–Perelman [22]). Let X be an Alexandrov space. Then there exists a dense open subset X0 such that, for each p ∈ X0 , there exists a neighborhood Up and a Lipschitz homeomorphism Up → Vp where Vp ⊂ Rn is an open set. T HEOREM 17.3 (Burago–Gromov–Perelman). The Hausdorff dimension of an Alexandrov space is an integer and is equal to its topological dimension. Remark 17.3. There are several ways to define topological dimension, that is covering dimension (big and small), inductive dimension, etc. Theorem 17.3 also implies that they coincide for Alexandrov spaces. We do not discuss the proof of Theorem 17.3. (It will follow from Corollary 18.3 in the next section.) Before explaining some of the ideas of the proof of Theorems 17.2, we give some examples of Alexandrov spaces. E XAMPLE 17.1. (0) A Riemannian manifold (M, g) is an Alexandrov space with K κ, if and only if the sectional curvature of (M, g) is greater than κ everywhere. (1) Let Ω ⊆ Rn be a compact and convex domain. Let S = ∂Ω. We define the length metric d on S. Namely, the distance between x, y ∈ S is the minimum of the length of the

Metric Riemannian geometry

247

curves in S joining x with y. Then we can prove that (S, d) is an Alexandrov space of curvature 0. (2) Let M be a Riemannian manifold with KM κ and G be a compact group acting on M by isometry. Then the quotient space M/G, equipped with the quotient metric is an Alexandrov space. An important example of an Alexandrov space is a Gromov–Hausdorff limit of an Riemannian manifold. Actually we have P ROPOSITION 17.4. Let Xi be a sequence of compact length spaces and X = limGH i→∞ Xi . If Xi are Alexandrov spaces with K κ, then so is X. (Here κ is independent of i.) The proof is elementary. Remark 17.4. Yamaguchi [153] proved that if M a C ∞ -manifold and G is a compact Lie group acting smoothly on M, then there exists a sequence of metrics gi on M such that Kgi κ for some κ independent and (M, gi ) converges to M/G. Another source of examples is a cone, which we define below. D EFINITION 17.2. Let (Y, d) be a metric space. We consider the product T × [0, ∞) and identify (x, 0) and (y, 0). We thus obtain a space CY . We define a cone metric on it as follows: d (x, t), (y, s) = t 2 + s 2 − 2st cos d(x, y). We denote by o ∈ CY the equivalence class of (x, 0). E XAMPLE 17.2. If Y = S n with KS n ≡ 1, then CS n is isometric to Rn+1 . L EMMA 17.5. If Y is a length space and Diam(Y ) π , then CY is a length space. We can prove an analogue of Myers’ theorem (Theorem 5.4) for Alexandrov spaces. Namely T HEOREM 17.6 [22]. If M is an Alexandrov space with K 1, then Diam(Y ) π . T HEOREM 17.7 [22]. (1) If CY is an Alexandrov space, then Y is an Alexandrov space with K 1. (2) If dim Y > 1 and Y is an Alexandrov space with K 1 then CY is an Alexandrov space with K 0. (3) In case dim Y = 1, the cone CY is an Alexandrov space with K 0 if and only if Diam(Y ) π .

248

K. Fukaya

Fig. 17.2.

We do not discuss the proof. We next discuss an example of a length space which is not an Alexandrov space. E XAMPLE 17.3. Let us consider a simplicial complex X consisting of three arcs which are joined at one point o. (See Figure 17.2.) We can define a metric on it such that the length of each arc is 1. Let x, y, z be interior points of each of the three simplexes, respectively. We can choose v = w on xy ∩ xz = xo. Then d(v, w) = 0. But if we choose x , y , z , v , w as in Theorem 17.1 then d(v , w ) > 0. (For any κ.) So the conclusion of Theorem 17.1 does not hold. Namely X is not an Alexandrov space. The argument of Example 17.3 implies the following. We call a map : (a, b) → X a geodesic if for each c ∈ (a, b) there exists such that the length of the restriction of to (c − , c + ) is d((c − ), (c + )). L EMMA 17.8. If 1 , 2 are geodesics on an Alexandrov space X and if they coincide on an open set, then their union is also a geodesic. In other words, geodesics can never branch. We next explain some other equivalent definitions of Alexandrov spaces. T HEOREM 17.9. Let X be a length space. We assume that for each p ∈ X there exists a neighborhood U such that the conclusion of Theorem 17.1 holds for any x, y, z, u, v ∈ U . Then X is an Alexandrov space with K κ. In other words, the same conclusion holds globally. In fact, usually the assumption of Theorem 17.9 is the definition of Alexandrov space. We discuss another equivalent definition. Let X be a length √ space and x, y, z ∈ X. Let κ ∈ R. In case κ > 0, we assume d(x, y), d(x, z), d(y, z) < π/ κ. We choose x , y , z ∈ Sn (κ) such that d(x, y) = d(x , y ), d(y, z) = d(y , z ), d(x, z) = d(x , z ). We define κ yxz =

y x z .

Metric Riemannian geometry

249

T HEOREM 17.10. Let X be a length space. (1) If, for each p ∈ X, there exists a neighborhood U of p such that κ bac + κ cac + κ cab

2π

for and a, b, c, d ∈ U , then X is an Alexandrov space with K κ. (2) Let X be an Alexandrov space with K κ and let a, b, c, d ∈ X. Then κ bac + κ cac + κ cab

Remark 17.5. By Theorem 17.6

2π.

κ bac

etc. in (2) is well defined.

The idea that if the comparison theorem holds locally, then it holds globally is due to Alexandrov and Toponogov. Theorem 17.10 is proved in [22]. We next discuss the angle between geodesics. Hereafter we assume that geodesics are parameterized by arc length. Let X be an Alexandrov space with K κ and 1 , 2 : [0, c) → X be geodesics such that p = 1 (0) = 2 (0). L EMMA 17.11. If s1 t1 , s2 t2 then κ 1 (s1 )p2 (s2 ) κ 1 (t1 )p2 (t2 ).

This follows easily from the definition. Therefore we can define D EFINITION 17.3. 1 2 = limt1 ,t2 →0 κ 1 (t1 )p2 (t2 ). In case 1 , 2 are minimal geodesics joining p to x, y, respectively, we write xpy = 1 2 . Remark 17.6. (1) The angle xpy is independent of κ. (2) Two geodesics 1 , 2 coincide to each other if 1 2 = 0. T HEOREM 17.12. If X is an Alexandrov space of K κ and x, y, z ∈ X, then we have d(y, z) s(d(x, y), d(x, z), yxz, κ). Here s is defined in (4.4). In other words, Theorem 4.7 holds for an Alexandrov space. The other version of the triangle comparison theorem also holds.

T HEOREM 17.13. If X is an Alexandrov space with K κ and x, y, z ∈ X, then we have yxz κ yxz.

Remark 17.7. The Toponogov type comparison theorem holds in Alexandrov space. Hence we can generalize the argument of the last section to prove the splitting theorem (Theorem 16.4) for an Alexandrov space with K 0. As we mentioned before an Alexandrov space has a tangent cone.

250

K. Fukaya

T HEOREM 17.14 [22]. If (X, d) is an Alexandrov space with K κ and x ∈ X, then limk→∞ ((X, kd), x) converges with respect to the pointed Gromov–Hausdorff distance. The limit in Theorem 17.14 is the tangent cone Tx X. The tangent cone is related to the angle between geodesics as follows. D EFINITION 17.4. Let Σ˜ x0 be the set of all geodesics (parameterized by arc length) : [0, c) → X for some c such that (0) = x. We identify 1 and 2 if they coincide on a neighborhood of 0. We denote by Σx0 the set of this equivalence relation. We can easily show that the angle defines a metric on it. We define the space of directions Σx (X) as the completion of Σx0 . L EMMA 17.15. If X is an Alexandrov space, then Σx (X) is an Alexandrov space with K 1 and Tx X is an Alexandrov space with K 0. T HEOREM 17.16 [22]. The tangent cone Tx X of an Alexandrov space X is isometric to the cone CΣx (X). If dim X = n, then dim Σx (X) = n − 1 and dim Tx X = n. We remark that the second half of Theorem 17.16 is a consequence of Proposition 17.7. Now we start the discussion of the proof of Theorem 17.1. As we mentioned in the last section, if x ∈ X is an interior point of n = dim X “independent” minimal geodesics, then Tx X is isometric to Rn , and this implies that x has neighborhood, homeomorphic to Rn . However the condition about the existence of a geodesic is a bit too strict. So we relax it a bit. Let X be an Alexandrov space with K κ. D EFINITION 17.5. Let x ∈ X and (ai , bi ) ∈ X 2 , i = 1, . . . , n. We say that {(ai , bi )}i=1,2,...,n is an (n, δ)-strainer at x, if κ ai xbi

π − δ,

and κ ai xaj , κ ai xbj , κ bi xbj

δ,

for i = j .

A point x ∈ X is said to be (n, δ)-strained if there exists an (n, δ) strainer at x. (See Figure 17.3.) Remark 17.8. In [22] the strainer is called “explosion” and a strained point is called “burst point”. The name strainer and strained point seems to be more popular now. The main step of the proof of Theorem 17.2 is the following P ROPOSITION 17.17 ([22, Theorem 9.4] or [138, Theorem 7.4]). If p ∈ X is an (n, δ)strained point, then there exists ρ > 0, neighborhoods U ⊂ V of p, and a map ϕ : V → Rn with ϕ(p) = 0 with the following properties:

Metric Riemannian geometry

251

Fig. 17.3.

(1) d(ϕ(x), ϕ(y)) < 2d(x, y). (2) Let x ∈ U and X ∈ Rn , with d(ϕ(x), X) < ρ. Then there exists y ∈ V such that ϕ(y) = X and d(x, y) Cd(ϕ(x), X), where C depends only on n and δ. We remark that (2) implies that ϕ is an open mapping in a neighborhood of x. Hence if ϕ is injective then ϕ gives a chart in a neighborhood of x. We can use the following to show ϕ is injective. L EMMA 17.18. We may choose U small enough so that if ϕ is not injective then there exists an (n + 1, 10δ)-strained point on a small neighborhood of U . We remark that the set of all the (n, δ)-strained points is open. On the other hand Proposition 17.17 implies that if a (n, δ)-strained point exists, then the Hausdorff dimension is not smaller than n. Hence Proposition 17.17 and Lemma 17.18 imply the following. For each open set U , we can find n and a nonempty open subset U0 ⊂ U consisting of (n, δ)-strained points such that there are no (n + 1, 10δ)-strained points on U . Then U0 is an n-dimensional manifold by Proposition 17.17 and Lemma 17.18. The proof of Theorem 17.2 then will be completed by using the next lemma. L EMMA 17.19 [22, Corollary 6.5]. We assume X is connected. If U, V are nonempty open subsets of X, then the Hausdorff dimension of U is equal to the Hausdorff dimension of V .

252

K. Fukaya

We now sketch the proof of Proposition 17.17 and Lemmas 17.18, 17.19. We put µ = inf d(p, a1 ), . . . , d(p, an ), d(p, b1 ), . . . , d(p, bn ) . We first explain the idea of the proof of Proposition 17.17. We put ϕ(x) = − d(x, a1 ), . . . , d(x, an ) + d(p, a1 ), . . . , d(p, an ) . It is easy to see that (1) is satisfied. We show (2). For simplicity, we consider the case x = p, n = 2. For each X = (X1 , X2 ) ∈ B0 (ρ, R2 ), we will find w with ϕ(w) = (X1 , X2 ), d(p, w) Cd(0, X). We assume X1 , X2 > 0. We first take the point q1 ∈ pa1 such that d(p, q1 ) = X1 . We first show |ϕ(q1 ) − (X1 , 0)| τ (ρ, δ|n, κ, µ). d(0, X)

(17.1)

In fact we can prove d(q1 , a2 ) d(p, a2 ) − τ (ρ, δ|n, κ, µ)X1 by applying Theorem 17.1, where we put x = a1 , y = p, z = v = a2 , u = q1 . To prove the opposite inequality we take the point p ∈ b1 q1 such that d(p , q1 ) = X1 . We have d(p, p ) X1 τ (ρ, δ|n, κ, µ). In fact, since κ b1 pa1 > π − δ, it follows that b1 q1 a1 κ b1 q1 a1 > π − δ − τ (ρ|n, κ). Hence pq1 b1 < δ + τ (ρ|n, κ). Theorem 17.12 then implies d(p, p ) X1 τ (ρ, δ|n, κ, µ). We use d(p, p ) X1 τ (ρ, δ|n, κ, µ) to show | b1 q1 a2 − π/2|, | b1 p a2 − π/2| < τ (ρ, δ|n, κ, µ).

(17.2)

We next apply Theorem 17.1 again by putting x = q1 , y = b1 , z = v = a2 , u = p . Then using (17.2) have d(p , a2 ) d(q1 , a2 ) − τ (ρ, δ|n, κ, µ)X1 . Hence d(p, a2 ) d(q1 , a2 ) − τ (ρ, δ|n, κ, µ)X1 . We have proved (17.1). We next take w1 ∈ a2 q1 such that d(w1 , q1 ) = X2 . Then we have |ϕ(w1 ) − (X1 , X2 )| τ (ρ, δ|n, κ, µ). d(0, X) We repeat the process replacing p by w1 and obtain w2 such that d(w1 , w2 ) < C|ϕ(w1 ) − (X1 , X2 )| and |ϕ(w2 ) − (X1 , X2 )| τ (ρ, δ|n, κ, µ). |ϕ(w1 ) − (X1 , X2 )| We can define w3 , . . . in a similar way. wi is a Cauchy sequence whose limit w has the required property.

Metric Riemannian geometry

253

Let us prove Lemma 17.18. Let ϕ(x) = ϕ(y). Let z ∈ xy with d(x, z) = d(x, y). It is easy to see that (ai , bi ), i = 1, . . . , n, and (x, y) is an (n + 1, 2δ)-strainer if d(x, y) is small. Finally we sketch the proof of Lemma 17.19. We may assume X is compact. Take p ∈ V and put D = sup{d(p, x) | x ∈ U }. We take R such that Bp (D/R, X) ⊂ V . We define Φ : U → V as follows. For x ∈ V we take a point Φ(x) ∈ px such that Rd(p, Φ(x)) = d(p, x). (Note the minimal geodesic px may not be unique. So we need some technical argument to find Φ which is measurable.) Definition 17.1 implies that there exists ρ > 0 such that d(Φ(x), Φ(y)) ρd(x, y). It follows that the Hausdorff dimension of Φ(U ) is not smaller than the Hausdorff dimension of U . Therefore the Hausdorff dimension of V is not smaller than the Hausdorff dimension of U . We can prove the opposite inequality in the same way. We thus finished a sketch of the proof of Theorem 17.2.

D EFINITION 17.6. We define the boundary ∂X of an Alexandrov space X by induction on dim X as follows. If dim X = 1, then X is either an arc or a circle. So we can define its boundary in an obvious way. Suppose ∂X is defined for X with dim X < k. Let X be an Alexandrov space of dim X = k. Then we say x ∈ ∂X if ∂Σx (X) = ∅. (We remark that Σx (X) is an Alexandrov space and dim Σx (X) = k − 1.) Theorem 17.2 is improved by Otsu–Shioya [112]. To state their results we define the notion of singular point set in an Alexandrov space more precisely. D EFINITION 17.7. Let X be an n-dimensional Alexandrov space and δ > 0. We put Sδ (X) = x ∈ X | Vol Σx (X) Vol S n−1 − δ , S(X) = Sδ . δ>0

We remark that the Alexandrov space version of the following theorem is a motivation of Definition 17.7. T HEOREM 17.20 (Otsu–Shiohama–Yamaguchi [111]). If an n-dimensional Riemannian manifold M satisfies Vol(M) Vol(S n ) − n , KM 1 then M is diffeomorphic to the sphere. Also M is close to S n with respect to the Hausdorff distance.27 We discuss the idea of the proof of Theorem 17.20 in Section 21. 27 This theorem is improved later to Theorem 21.7 and to Corollary 22.4. Before [111], Shiohama [137] proved that M is homeomorphic to the sphere under a similar but different assumption KM −C, Ricci (n − 1), Vol(M) Vol(S n ) − (n, C).

254

K. Fukaya

T HEOREM 17.21 (Burago–Gromov–Perelman, Otsu–Shioya). Let X be an Alexandrov space of dimension n. Then the Hausdorff dimension of S(X) is not greater than n − 1. The Hausdorff dimension of S(X) \ ∂X is not greater than n − 2. T HEOREM 17.22 (Otsu–Shioya [111]). There exists a C 0 -Riemannian metric on X \ S(X) which induces the metric on X. Moreover there exists X0 ⊂ X \ S(X) such that the (n-dimensional Hausdorff ) measure of X \ X0 is 0 and that there exists manifold structure of C 1.5 -class and a Riemannian structure is of C 0.5 -class on X0 . Remark 17.9. Actually we need to define a C 1.5 -structure etc. in the above theorem. This is because X \ S(X), X0 are not open subsets in general. Hence they are not manifolds. See [112] for the precise statement. Theorem 17.22 is used by Kuwae–Machigashira–Shioya [99] to develop analysis on Alexandrov spaces. We also remark the following T HEOREM 17.23 (Fukaya–Yamaguchi [60]). The isometry group of an Alexandrov space is a Lie group.

18. Alexandrov space—II In [119,113] Perelman proved the following two fundamental results on Alexandrov spaces. T HEOREM 18.1 (Perelman). Let X be an Alexandrov space with K κ. Then, for any x ∈ X, there exists a neighborhood of x homeomorphic to Tx X, the tangent cone. T HEOREM 18.2 (Perelman). Let Xi be a sequence of Alexandrov spaces with K κ where κ is independent of i. We assume X = limGH i→∞ Xi and dim X = dim Xi . Then Xi is homeomorphic to X for large i. Remark 18.1. Both of these theorems are proved in [119]. Later Perelman published another paper [113] where the proof of Theorem 18.1 is given in a simplified way. Perelman says in [113] that a similar method gives a slight simplification of the proof of Theorem 18.2, but the simplification is not so big compared with the one for Theorem 18.1. Unfortunately the paper [119] is not yet published. In fact Theorem 18.1 follows from Theorem 18.2 (and Theorems 17.14, 17.16). However the proof of Theorem 18.2 requires Theorem 18.1. In this section we give a review of the proof of Theorem 18.1. Before that let us mention some of the corollaries of them. We remark that Tx X is homeomorphic to CΣx (X) by Theorem 17.16. Since Σx (X) is again an Alexandrov space, we can apply Theorem 18.1 again. We then find that the

Metric Riemannian geometry

255

singularity of X is obtained locally by taking cones several times. Let us define it more precisely. D EFINITION 18.1. We define a connected metrizable space X to be an MCS-space of dimension n inductively on n as follows. (1) An MCS-space of dimension 2 is a 2-dimensional manifold with or without boundary. (2) X is an MCS-space of dimension n if, for each x ∈ X, there exists a neighborhood U of x and an MCS-space Yx of dimension n − 1, such that there exists a homeomorphism F from the cone of Yx to U such that F sends the cone point to x. The following is immediate from Theorem 18.1. C OROLLARY 18.3. Every Alexandrov space is an MCS-space. The following is also an immediate corollary. C OROLLARY 18.4. For an Alexandrov space X, there $ exists Xk with Xk is a k-dimensional topological manifold and X¯ k = ik Xi .

$

Xk = X such that

C OROLLARY 18.5. An Alexandrov space X is locally contractible. If it is compact then π1 (X) and Hk (X) are finitely generated. Hereafter we assume our Alexandrov space X has no boundary, for simplicity.28 An idea used in [113] to prove these result is to generalize Morse theory of the distance function to an Alexandrov space. Let us give the following definition. Hereafter X is an Alexandrov space with K −1. Let p ∈ X. We put dp (x) = d(x, p). D EFINITION 18.2. x is said to be a regular point of dp if there exists ξ ∈ Σx (X) such that for each minimal geodesic joining x to p we have ξ > π/2. Here ∈ Σx (X) is the equivalence class of in Σx (X). Definition 18.2 is a generalization of Definition 14.1. We can generalize Proposition 14.2 also and we further generalize it to Theorem 18.7. For the proof of Theorem 18.1 we need to use a bit more general function than the distance function and define the “regularity” of a map X → Rk for k > 1 also. To state this generalization we need some notations. D EFINITION 18.3. Let U be an open subset of X. (1) An admissible function f : U → R is a function of the form f (x) =

m

ai φi d(Ai , x) ,

(18.1)

i=1 28 The general case can be handled by taking a double X ∪ ∂X X which is an Alexandrov space by [124].

256

K. Fukaya

where A i is a compact subset of X, φi are smooth functions with 0 φi 1 and ai 0, ai 1. (2) An admissible map F : U → Rk is a composition F = G ◦ f., where G is a biLipschitz homeomorphism and f. = (f 1 , . . . , f k ) with admissible functions fi .

Remark 18.2. In [113] a more general function (map) is called an admissible function (map). But only those in Definition 18.3 are used. For A ⊂ X and x ∈ X we define Σx0 (xA) ⊂ Σx (X) by Σx0 (xA) =

[] | is a minimal geodesic joining x to a point of A .

Let Σx (xA) ⊂ Σx (X) be the closure of Σx0 (xA). For Λ1 , Λ2 ⊂ Σx (X) we put Λ1 Λ2 = inf{ uv | u ∈ Λ1 , v ∈ Λ2 }.

For an admissible function f as in (18.1) we can define its direction derivative Dx f : Σx (X) → R by (Dx f )(u) =

ai φ d(x, Ai ) cos u, Σx (xAi ) .

i

In case X is a manifold Dx f is the direction derivative in the usual sense. If f (1) , f (2) are admissible functions as in (18.1) we put Dx f (1) , Dx f (2) (1) (2) (1) (2) (2) (1) (2) φ d x, Aj cos Ai Aj . ai aj φ (1) d x, Ai =

i,j

This again coincides with the usual inner product between derivatives in case when X is a manifold and f (1) , f (2) are differentiable. D EFINITION 18.4. We say F : U → Rk where F = G ◦ f. is -regular at p ∈ U if the following conditions hold. Let us put f. = (f 1 , . . . , f k ). (1) For each i = j , we have Dp f i , Dp f j < −. (2) There exists ξ ∈ Σp (X) such that (Dp f i )(ξ ) < −, for each i. We say F is regular if it is -regular for some > 0. We say F is -regular on U if it is -regular at every point of U . E XAMPLE 18.1. (1) If f : U → R is defined by f (x) = dp (x). It is an admissible function and hence is an admissible map. It is -regular at x for some > 0 if and only if x is a regular point of f in the sense of Definition 18.2.

Metric Riemannian geometry

257

Fig. 18.1.

(2) Let X be a two-dimensional Alexandrov space and (a1 , b1 ), (a2 , b2 ) be a (2, δ)strainer at x. Let us define ϕ : X → R2 by ϕ(x) = − d(x, a1 ), d(x, a2 ) + d(p, a1 ), (.p, a2 ) , as in the proof of Proposition 17.17. Then ϕ is a homeomorphism in a neighborhood of x. We put √ √ p = ϕ −1 (r, 0), q = ϕ −1 (−r/2, r 2/2), r = ϕ −1 (−r/2, −r 2/2) and define F = f. = (dp , dq ). We also set ξ ∈ Σx (xr). We can prove (1), (2) for sufficiently small . We can generalize this construction to the case of higher dimension and prove that if x is a (k, δ)-strained point, then there exists F : U → Rk from a neighborhood of x which is -regular at x. (See Figure 18.1.) We can prove the following in a way similar to the proof of Proposition 17.17. (See [116, Lemma 2.3 and the argument just after that].) L EMMA 18.6. Let F : Bx (ρ, X) → Rn be an admissible map which is -regular at x. Then there exists a neighborhood U ⊆ Bx (ρ, X) of x and δ > 0, with the following property. If y ∈ U, X ∈ Rk with d(F (v), X) δ, then there exists z ∈ Bx (ρ, X) such that F (z) = X and d(z, y) < Cd(F (v), X). Here C depends only on ρ, δ, . Lemma 18.6 implies that F is an open mapping. In case dim X = k and if there exists an -regular map at x, then Lemma 18.6 shows that a neighborhood of x is a manifold. In the general case, we have to study the situation where k < dim X. The following Proposition 18.7 is the main result in such a case. We need a definition. D EFINITION 18.5. A map F : X → Y between topological spaces is called a topological submersion at x ∈ X if there exists a neighborhood U of x, a neighborhood V of F (x), and a topological space W such that there exists a homeomorphism Φ : U ∼ = V × W satisfying F = P r1 ◦ Φ on U .

258

K. Fukaya

In case X, Y are smooth manifolds and F is a smooth map, F is a topological submersion if its derivative is of maximal rank. P ROPOSITION 18.7 [113, Theorem 1.4]. An admissible map F : X → Rk is a topological submersion at a regular point. The following result is also used in the proof of Theorem 18.1. T HEOREM 18.8 (Siebenmann [142, Corollary 6.14]). Every proper topological submersion between MCS-spaces is a locally trivial fiber bundle. Remark 18.3. (1) We remark that if M, N are smooth manifolds (without boundary) and if F : M → N is a proper smooth submersion then F is a locally trivial fiber bundle. This fact can be proved much more easily than Theorem 18.8. (2) The proof of Theorem 18.8 is based on the isotopy extension theory. We remark that isotopy extension theory for manifolds (see [48]) was used by Cheeger for the proof of his finiteness theorem. (See Section 6.) We next sketch the proof of Proposition 18.7. The difficult case is when X is of dimension greater than k. We try to increase k as much as possible, we then arrive at the following situation. D EFINITION 18.6. Let F : U → Rk be a regular admissible map from an open set U of an Alexandrov space X. We say p ∈ X is imcomplementable if there exists no g such that (f 1 , . . . , f k , g) is regular at p. The case k = 0 is included. Namely in that case p ∈ X is imcomplementable if there exists no admissible function such that p is regular. E XAMPLE 18.2. (1) Let us consider the domain {(r cos θ, r sin θ ) | θ ∈ [−α, +α], r 0}. We glue (r cos α, r sin α) and (r cos −α, r sin −α) to obtain a space Xα . We can show that o = [0, 0] is imcomplementable if and only if α π/2. Actually we put g = d[r,0] . Then g is regular if α > π/2. On the other hand, if α π/2 then the diameter of Σo Xα is not greater than π/2. Hence it is easy to see that (2) in Definition 18.4 can never be satisfied.29 (See Figure 18.2.)

Fig. 18.2. 29 It is easy to see from this argument that in case k = 0 the point p ∈ X is imcomplementable if and only if

Diam Σp (X) π/2.

Metric Riemannian geometry

259

Fig. 18.3.

(2) Let us next take X = Xα × R, where Xα is as above. We define f : X → R by f = d(o,−1) . It is easy to see that (o, 0) is a regular point. Actually we may take ξ = D(o,0) d(o,1) . We can show that f is imcomplementable at (o, 0) if α π/2. (See Figure 18.3.) Now the main technical result in [113] is as follows. L EMMA 18.9 [113, 1.3]. If F : U → Rk is admissible and regular at p ∈ U , and if p is imcomplementable, then there exists an admissible function g : V → R defined on an open neighborhood V of p with the following properties. We write F = G ◦ f., f. = (f 1 , . . . , f k ). (1) g 0 on V and g(p) = 0. (2) F |g −1 (0) : g −1 (0) → Rk defines a homeomorphism onto a neighborhood of F (p). (3) (F, g) : V → Rk+1 is regular on V \ g −1 (0). (4) There exists ρ > 0 such that {x ∈ V | d(F (x), F (p)) ρ, g(x) −ρ} is compact. Let us show how to choose such g in the case of Example 18.2(2). Namely we have U = Xα × R and F = f = d(o,−1) . We write a point of U as ([r cos θ, # r sin gθ ], t) and use r, θ, t as a coordinate. (We take r 0, θ ∈ [−α, α].)#Then f (r, θ, t) = (t + 1)2 + r 2 . We take q = (δ, 0, δ) and put h = dq . Then h(r, θ, t) = (t − δ)2 + r 2 + δ 2 − 2rδ cos |θ |. It is easy to see that (f, h) : U → R2 is regular outside on (B(o,0) (ρ, U ) \ {o}) × R. (We remark {o} × R is the set of singular points.) However if we put g = h then (1), (2) are not satisfied. So we compose it with a homeomorphism of R2 so that (1), (2) will be satisfied. (See Figure 18.4.)

260

K. Fukaya

Fig. 18.4.

We consider the set K(v) = {x ∈ U | f (x) = v, r < ρ} where |ρ| and |v| is small. We can easily check that h(v, 0, 0) = sup h(x) | x ∈ K(v) if α π/2.30 Namely (∗) The restriction of h to K(v) takes its maximum# at a unique point. We remark that f (v, 0, 0) = 1 + v and h(v, 0, 0) = (v − δ)2 + δ 2 . So if we put g(r, v, θ ) = h(r, v, θ ) −

2 f (r, v, θ ) − 1 − δ + δ 2 ,

then (1), (2) are satisfied. We define G : B(1,√2δ) (ρ, R2 ) → R2 by # G(a, b) = a, b − (a − 1 − δ)2 + δ 2 , where ρ - δ. Since (f, g) = G ◦ (f, h), it follows that (f, g) is admissible. It also satisfies (3) since (f, h) satisfies (3). Thus we constructed g in the case of Example 18.2(2). In the general case, we need to choose h more carefully so that it is enough “concave”. (Then (∗) holds.) The proof of Lemma 18.9 is in [113, Section 3]. We can use Lemma 18.9 to complete the proof of Proposition 18.7 as follows. We also prove the following at the same time. 30 This condition is equivalent to the condition that (o, 0) is imcomplementable.

Metric Riemannian geometry

261

P ROPOSITION 18.10. If F : X → Rk is an admissible map and if p ∈ X is a regular point, then F −1 (F (p)) is an MCS-space near p. The proof is by downward induction on k. If k = dim X then both Theorem 18.7 and Proposition 18.10 follow from Lemma 18.6. Let us assume that Theorem 18.7 and Proposition 18.10 are true for k + 1 and prove the case k. We remark that both propositions are local statements on p.31 In case p is not imcomplementable, then we can increase k and use the induction hypothesis. So it suffices to consider the case that p is imcomplementable. We apply Lemma 18.9 to get g. Then (F, g)|V \g −1 (0) : V \ g −1 (0) → Rk+1 is regular. We can use the induction hypothesis to conclude that it is a topological submersion and the fibers are MCS-spaces. Therefore V \ g −1 (0) is an MCS-space. Let (F, g)−1 (B0 (ρ, Rk ) × (−ρ, 0)) = W . Since F : W → B0 (ρ, Rk ) × (−ρ, 0)

(18.2)

is proper, Theorem 18.8 implies that (18.2) is a locally trivial fiber bundle. Since the base space is trivial it follows that (18.2) is a trivial bundle. Hence using Lemma 18.9(2), we can prove Propositions 18.7 and 18.10 for F : U → Rk . Thus the induction works. We remark that Proposition 18.10 implies Theorem 18.1 by putting k = 0. We thus sketched the proof of Theorem 18.1. The proof of Theorem 18.2 uses a similar argument but more involved. See [119]. Let us compare the results we reviewed in the last and this section so far to the one in earlier sections, where we consider the case |KM | is bounded. In Section 11, we asked two questions, Problem 11.1 for a sequence Mi converging to X. (1) was on the singularity of X and (2) was on the relation between topologies of Mi and X. In the case |KMi | 1, an answer to (1) was Theorem 11.4 and an answer to (2) was Theorem 11.5 and 12.7. In our more general case where we assume KMi −1 only, Theorem 18.1 and Corollary 18.3 give a satisfactory answer to (1). However, results on (2) are not satisfactory. In case X = limGH i→∞ Mi satisfies dim X = dim Mi , Theorem 18.2 is a satisfactory answer. This is the noncollapsing case. On the other hand if X is a smooth Riemannian manifold, Theorem 11.3 by Yamaguchi, gives a nice answer. Namely there exists a fiber bundle Mi → X for large i.32 However, the trick (taking frame bundles) we explained in Section 11, does not work in our more general situation to reduce the problem to the case when X is a manifold. So the result is not yet satisfactory. There are however several interesting approaches and partial results about the problem (2) in the case Mi −1, which we review very briefly here. First, Theorem 11.3 is generalized to the case when the limit X has a rather mild singularity. There are two papers about it. In [154], Yamaguchi assumed that for each x ∈ X, there exists a strainer (ai , bi ), i = 1, . . . , n = dim X, with d(x, ai ), d(x, bi ) > µ > 0 where µ is independent of x. Then he concludes that there exists a locally trivial Lipschitz fiber bundle structure Mi → X.33 31 So we prove them by induction without assuming completeness of X. 32 Theorem 11.3 does not say much about the fibers. But there are various results which shows that the fibers are

“of nonnegative curvature” in some sense. 33 In the preprint version of [154] (which the author has), the locally triviality is not asserted. It is proved in [141].

262

K. Fukaya

Perelman in [117] assumed that X has no proper extremal set. Here D EFINITION 18.7. F ⊂ X is extremal if for each p ∈ / F , x ∈ X, and u ∈ Σx (X), we have Dx dp (u) 0. For example, F = {x} consisting of one point is not extremal if and only if there exists an admissible function f which is regular at x. Perelman’s theorem in [117] is that, if there is no extremal set, then there exists fi : Mi → X such that πk (Mi , fi−1 (p)) ∼ = πk (X) for each p ∈ X. The plan proposed by Perelman [116] then is to stratify X using an extremal set and to construct a fiber bundle structure stratawise. This plan is not yet completed. Shioya–Yamaguchi [141] and Yamaguchi [155,156] studied the case when dim Mi = 3, 4 without extra assumption on X and gave a satisfactory description in that case. In this article, we discuss the 3-dimensional case only. Let Mi be 3-dimensional Riemannian manifold with KMi −1, and X = limGH i→∞ Mi . We assume dim X 2. Then X is homeomorphic to a manifold with or without boundary. We assume that X is connected. T HEOREM 18.11 (Shioya–Yamaguchi [141]). We assume dim X = 2. (1) If ∂X = ∅, then there exists a structure of Seifert fibered space Mi → X for large i. (2) If ∂X = ∅, then Mi is homeomorphic to Seii (X) ∪ (∂X × D 2 ) where Seii (X) is a Seifert fibered space over Int X. We glue it with ∂X × D 2 where the fibers of Seii (X) over the boundary point x are glued with {x} × ∂D 2 . In case dim X = 1 there are two possibilities, X ∼ = S 1 or [0, 1]. In the case X = S 1 there 1 exists a fiber bundle Mi → S by Theorem 11.3. T HEOREM 18.12 (Shioya–Yamaguchi [141]). If X ∼ = [0, 1] then Mi is obtained by gluing Bi and Ci along their boundaries where each of Bi , Ci is homeomorphic to one of the following 4-manifolds. (1) D 3 , (2) A nontrivial [0, 1]-bundle over RP 2 , (3) S 1 × D 2 , (4) A nontrivial [0, 1]-bundle over the Klein bottle. The rough idea of the proofs of Theorems 18.11, 18.12 are as follows. In either cases, we can apply a generalization [154] of Theorem 11.3 except in finitely many points (plus ∂X in case (2) of Theorem 18.11). In the neighborhood of those points we scale the metric to obtain a noncompact nonpositively curved Alexandrov space. Then applying the soul theorem (the Alexandrov space analogue of Theorem 16.7). The soul S is an Alexandrov space of dimension 2, so it is a manifold with or without boundary. Actually Shioya– Yamaguchi classified 3-dimensional noncompact complete Alexandrov spaces with K 0. In this way, we can classify neighborhoods ⊂ Mi of a singular point of X locally. Then the last step is to glue those local neighborhoods. In the case when dim X = 0, we can scale the metric of Mi and obtain a limit of nonzero dimension. In this way [141] (improving [153,59]) proved the following

Metric Riemannian geometry

263

T HEOREM 18.13 (Shioya–Yamaguchi). There exists such that if M is a Riemannian 3-manifold with KM Diam(M) −, then a finite cover of M is homeomorphic to S 1 × S 2 , T 3 , a nilmanifold or a simply connected Alexandrov space with K 0.

19. First Betti number and fundamental group So far we discussed results about sectional curvature. In this section we discuss also Ricci curvature. The recent progress mainly due to Cheeger–Colding will be discussed in later sections. In this section, we mainly concern with older results. To study Ricci curvature we need partial differential equations frequently. But we do not mention them so much. We first review Theorem 2.3. The proof of Theorem 2.3 is based on the Bochner trick. The most famous result in metric Riemannian geometry based on the Bochner trick is the following T HEOREM 19.1 (Bochner [157]). If an n-dimensional compact Riemannian manifold M has nonnegative Ricci curvature, then the first Betti number of M is not greater than n. The proof of Theorem 19.1 due to Bochner is as follows. Let u be a one-form on M. Then we have the following equality of Weitzenbeck type. (For proof see [157]. We remark that we use the nonpositive Laplacian (16.2).) 1 −u, u = − u2 + ∇u, ∇u + Ricci(u, u). 2

(19.1)

Let u be a harmonic one-form. We integrate (19.1) over M. The left-hand side is zero (since u is harmonic) and the integral of the first term in the right-hand side vanish. Therefore we have ∇u, ∇uΩM + Ricci(u, u)ΩM = 0. (19.2) M

M

(Here ΩM is the volume element.) The first term of (19.2) is nonnegative. If we assume that the Ricci curvature is nonnegative, then the second term also is nonnegative. Therefore the first and second term both are zero. Namely every harmonic one-form is parallel. Since a parallel one-form is determined by its value at one point (here we are assuming that M is connected), it follows that the dimension of the space of harmonic one forms on M is at most n. Theorem 19.1 follows. When we try to apply a similar argument to the forms of higher degree and try to estimate higher Betti numbers by Ricci curvature, we will meet a trouble. In formula (19.1), the third term involves only Ricci curvature. This is true only for one-forms. A similar formula for forms of higher degree is much more complicated. The assumption that we need to apply a similar argument to forms of higher degree is exactly the assumption in Theorem 2.3, which is much stronger than the one on Ricci curvature.34 34 On the other hand, if we write a formula similar to (19.1) for the spinor and Dirac operator the second term involves only scalar curvature. (See the text book of the Atiyah–Singer index theorem.) A theorem by Lichnerow-

264

K. Fukaya

In Section 16, we discussed the splitting theorem of Riemannian manifolds of nonnegative Ricci curvature (Theorem 16.4). We can prove Theorem 19.1 by using this theorem also. Actually we have the following T HEOREM 19.2 (Cheeger–Gromoll). If M is a compact manifold with nonnegative Ricci curvature, then there exists a finite cover M˜ of M, such that M˜ is isometric to the direct product X × T k , where X is simply connected and T k is a flat torus. ˆ Since we may assume To prove Theorem 19.2, we consider the universal covering M. ˆ we can prove that the fundamental group of M is infinite (otherwise we may take X = M), 35 ˆ ˆ that M contains a line. Now by applying Theorem 16.4, we find M = R × Y . We may split Mˆ = Rk × Y so that Y has no R factor. If Y is not compact, we can show Y contains a line by the same argument. Then, by Theorem 16.4, Y has an R factor, a contradiction. Namely Y is compact. The π1 M action preserves the splitting Mˆ = Rk ×Y . Theorem 19.2 follows easily. In case k = n in Theorem 19.2, or in case when the first Betti number is equal to the dimension in Theorem 19.1, we can show that M is flat. (We can show this fact either by Bochner’s proof using (19.1) or by Cheeger–Gromoll’s proof based on the splitting theorem.) Theorem 19.2 is generalized by Gromov as follows. T HEOREM 19.3 [69, p. 73]. There exists a continuous function b(n, ρ) of ρ ∈ R with b(n, 0) = n, such that the following holds. If M is an n-dimensional Riemannian manifold with diameter 1, Ricci curvature ρ, Then its first Betti number is not greater than b(n, ρ). C OROLLARY 19.4. If M is an n-dimensional Riemannian manifold with diameter 1 and Ricci > −n , then its first Betti number is not greater than n. Here n is a positive number depending only on n. Gromov’s proof is based on the estimate of the growth function by using the Bishop– Gromov inequality (Proposition 5.2) and is closer to the study of the fundamental group we mention later in this section (Theorem 19.9, Theorem 19.10). The analytic proof, using a similar idea to Bochner’s, is given by Gallot [61]. As we mentioned before, the idea of the proof of Theorem 19.1 cannot directly be applied to the study of the second or higher Betti number. In fact, a result similar to Theorem 19.3 does not hold for higher Betti numbers. Namely the statement such as: “If M is an n-dimensional compact Riemannian manifold with diameter 1, Ricci curvature ρ, then its Betti number is smaller than a number depending only on ρ and n”, is icz which asserts “The Aˆ genus of Riemannian manifold of positive scalar curvature is zero” is obtained from this fact. 35 Let p , q ∈ M with d(p , q ) → ∞. Let x be the midpoint of a minimal geodesic joining p and q . Moving i i i i i i i them by an action of π1 (M), we may assume that there exists R independent of i such that d(x, xi ) < R. Then a subsequence of the sequence of geodesics joining pi , xi , qi has a limit. This limit is a line.

Metric Riemannian geometry

265

false. See [135,118] for counter examples. Note that if we replace Ricci ρ by KM ρ in the statement in the parenthesis, then it is Theorem 14.6. Let us consider the case when equality holds in Corollary 19.4, namely the case Ricci > −n and the first Betti number is n. T HEOREM 19.5 (Yamaguchi [153]). There exists a positive number n such that if M is an n-dimensional Riemannian manifold with diameter Diam(M)KM > −n , and its first Betti number is b, then there exists a finite cover M˜ of M and a fiber bundle M˜ → T b over a b-dimensional torus. Moreover, if b = dim M, then M is diffeomorphic to the torus. Remark 19.1. (1) Yamaguchi proved the same conclusion for the fiber of Theorem 11.3. (2) Yamaguchi [151] proved the same conclusion under a different hypothesis KM 1, Diam D, RicciM −(D, n). To prove Theorem 19.5, Yamaguchi used Theorem 16.4 case (b). The second half of Theorem 19.5 is generalized by Colding [46] and Cheeger–Colding [29] as follows. T HEOREM 19.6 [46,29]. If M is an n-dimensional Riemannian manifold with Diam(M) RicciM > −n , and its first Betti number is n, then M is diffeomorphic to a torus. Remark 19.2. The first half of the statement of Theorem 19.5 does not hold under the milder assumption Diam(M) RicciM > −n . Anderson [10] constructed an example of M with Diam(M) RicciM > −n but that has no fibration over T b1 (M) . We here explain some of the ideas used by Yamaguchi in [153] to show Theorem 19.5, which is also used in [46]. (The additional ideas due to [46,29] will be explained in later sections.) For simplicity we consider the case b = n = dim M only. The proof is by contradiction. By scaling we may assume that there exists Mi with Diam(Mi ) = 1, KMi −i but Mi is not diffeomorphic to T n . We consider the covering space Mˆ i → Mi whose covering transformation group is Γi = Zb . We study the limit of the pair (Mˆ i , Γi ). Here we define D EFINITION 19.1 [52]. A sequence of pairs ((Xi , pi ), Γi ) of pointed metric spaces (Xi , pi ) and groups of isometries Γi is said to converge to ((X, p), G) with respect to the equivariant pointed Hausdorff convergence if there exists ϕi : Bpi (1/i , Xi ) → Bp (1/i , X), ϕi : Bp (1/i , X) → Bpi (1/i , Xi ), ψi : Γi → G, ψi : G → Γi with i → 0 such that (1) ϕi , ϕi are i -Hausdorff approximations and d x, ϕi ϕi (x) < i . d x, ϕi ϕi (x) < i , (2) If x, γ (x) ∈ Bpi (1/i , Xi ), γ ∈ Γi , then d ϕi γ (x) , ψi (γ ) ϕi (x) < i .

266

K. Fukaya

(3) If x, γ (x) ∈ Bpi (1/i , X), γ ∈ Γi , then d ϕi γ (x) , ψi (γ ) ϕi (x) < i . We remark that we do not assume that ψi , ψi are homomorphisms. We can prove a similar compactness result as Theorem 16.1. Now let us go back to the proof of Theorem 19.5. Fix pi ∈ Mˆ i . We may consider the limit ((Mˆ i , Γi ), pi ) with respect to the equivariant pointed Hausdorff convergence. However, then the limit may be a continuous group and is a bit hard to handle. So we use the following lemma. L EMMA 19.7 [153]. There exist subgroups Γi ⊂ Γi of finite index and η, η (independent of i) such that (1) For each γ ∈ Γi with γ = 1 we have d(pi , γ (pi )) η. (2) Γi is generated by elements γ1 , . . . , γn such that d(pi , γk (pi )) η . (Here n = dim M.) Lemma 19.7 appeared in the proof by Gromov of Theorem 19.3. The fact that Γi is Abelian plays an important role in the proof. We omit the proof of Lemma 19.7. Now we can consider the limit of the sequence ((Mˆ i , Γi ), pi ). We denote it by ((X, G), p). Using Lemma 19.7 we can easily show that G ∼ = Zn and its action is properly discontinuous. Now we apply the splitting theorem to X and obtain X = Rk × Y where Y is compact.36 Since Zn acts on it properly discontinuously, it follows that k = n. Since dim X dim Mˆ i = n, it follows that X = Rn . We can also prove that Mˆ i /Γi converges to X/G ∼ = T n . We put M˜ i = Mˆ i /Γi . Since M˜ i is n-dimensional and converges to T n , it follows from Theorem 11.3 that M˜ i is diffeomorphic to T n . Using Hn (Mi , Q) = n again, we can show that Mi is homeomorphic to T n . Furthermore we can arrange the covering index M˜ i → Mi so that “Mi is homeomorphic to T N and Mi is diffeomorphic to T n ” imply that Mi is diffeomorphic to T n , if 5. (This point is a standard application of nonsimply connected surgery.) The last step in the low-dimensional case is a bit complicated and is omitted. Remark 19.3. The above argument can be applied to the situation of Theorem 19.6. We only need to replace the splitting theorem to the one by Cheeger–Colding and Theorem 11.3 by Theorem 22.3. Colding’s argument in [46] (though using Lemma 19.7) is slightly different. This is probably because the splitting theorem we need for this purpose was not yet proved at that time. We next remark the following corollary of Theorem 19.2. C OROLLARY 19.8. If a compact Riemannian manifold M has nonnegative Ricci curvature, then its fundamental group π1 (M) contains an Abelian subgroup of finite index. It seems that series of results related to Corollary 19.8 began with the following theorem. 36 I think this was the first place where splitting theorem of the limit (singular) space was applied to study Riemannian manifold.

Metric Riemannian geometry

267

T HEOREM 19.9 (Milnor [104]). If a complete manifold M has nonnegative Ricci curvature and if G is a finitely generated subgroup of π1 (M), then G has polynomial growth. The definition of a group having polynomial growth is as follows. Let G be a finitely generated group and g1 , . . . , gk generate G. Let fG (N ) be the number of elements of G which can be written by a product of at most N of gi or gi−1 . D EFINITION 19.2. We say that G has polynomial growth, if there exists C, K such that fG (N ) < C(N K + 1). It is easy to see that this definition is independent of the choice of generator of G. The proof of Theorem 19.9 is based on Proposition 5.2 and proceeds as follows. Let us assume M is compact for simplicity. Let M˜ be the covering space of M corresponding ˜ By Proposition 5.2 we have to G. Let p ∈ M. ˜ CR n . Vol Bp (R, M) By an elementary argument using fundamental domain, we can show the existence of C with C −1

ceR/C . (We say that G has exponential growth in this case.) On the other hand, G = Zk has polynomial growth. Gromov [71] proved the following T HEOREM 19.10 (Gromov). A finitely generated group G has polynomial growth if and only if G has a nilpotent subgroup of finite index. Let us very briefly sketch its proof here. First we recall the following T HEOREM 19.11 (Tits [147]). Let G be a finitely generated subgroup of GL(n, R). Then either G contains a solvable subgroup of finite index or G contains a noncommutative free group.

268

K. Fukaya

If G contains a noncommutative free group, we can show that G is not of polynomial growth. On the other hand, Milnor proved that a solvable group is of polynomial growth if and only if it contains a nilpotent group of finite index. Hence to prove Theorem 19.10 it suffices to embed G to some Lie group. Gromov’s idea is to do so by using Hilbert’s 5th problem. Let G a group of as in Theorem 19.10. We define a metric (the word metric) on G as follows. Let γ1 , . . . , γn be generators. Let µ1 , µ2 ∈ G. We define d(µ1 , µ2 ) to be the smallest number k such that µ2 = γi11 . . . γikk µ1 . Here ij ∈ {1, . . . , n}, j = ±1. 1 Now we consider the limit limGH N →∞ (G, N d) as N → ∞. The assumption that G is of polynomial growth is used to show that the limit exists. It is easy to see that the limit G has the structure of a group. We then can use the fact that G acts as isometry on itself preserving the metric and a solution of Hilbert’s 5th problem, to show that G is a Lie group. So if we can embed G to G , we are done. But it is not so easy to embed G to G . (Actually in case G is a discrete subgroup of a nilpotent Lie group N , the limit is N but has a strange metric called the Carnot–Carathéodory metric (see [74]).) Therefore we need to discuss it more carefully and some more technical argument is required. We omit it. Theorems 19.9 and 19.10 imply that the finitely generated subgroup of the fundamental group of a complete manifold of nonnegative Ricci curvature has nilpotent subgroup of finite index. This fact is generalized by Fukaya–Yamaguchi [59]37 and further by Cheeger– Colding [46] as follows. T HEOREM 19.12 (Cheeger–Colding). There exists a positive number n such that if an n-dimensional Riemannian manifold satisfies Diam(M)2 RicciM −n , then π1 (M) contains a nilpotent subgroup of finite index. We remark that Theorem 19.12 follows Theorem 10.5. We also remark that Theorem 10.5 implies Theorem 19.9. In Theorem 19.12 we cannot replace the conclusion “nilpotent” by “Abelian”. Namely we cannot replace the assumption Ricci 0 of Corollary 19.8 by −n . The counter example is an almost flat manifold (Example 10.1). Some more results on the fundamental group is proved in [59] and [60] which we review here. A group Γ is said to be polycyclic if there exists 1 = Γ0 ⊂ Γ1 ⊂ · · · ⊂ Γk = Γk ,

(19.3)

such that Γi is a normal subgroup of Γi+1 and Γi+1 /Γi is cyclic. The smallest such number k is called the degree of polycyclicity of Γ . T HEOREM 19.13 ([59, Theorem 0.6, Corollary 7.20] plus [46]). There exists n and wn such that if an n-dimensional Riemannian manifold M satisfies RicciM Diam(M) −n then π1 contains a normal subgroup Γ such that 37 The result of [59] is the same conclusion as Theorem 19.12, but the assumption is on sectional curvature

instead of Ricci curvature.

Metric Riemannian geometry

269

(1) [π1 (M) : Γ ] wn . (2) Γ is polycyclic and its degree of polycyclicity is not greater n. T HEOREM 19.14 (Fukaya–Yamaguchi [60]). For each D, n there exists a finite set of groups G with the following properties. Let M be a manifold with KM −1, Diam(M) D. Then there exists G ∈ G and a surjective homomorphism π1 M → G such that the kernel Γ satisfies (1), (2) of Theorem 19.13. Theorem 19.14 implies the following. For a group G let us put D(G, n) = inf Diam(M) | KM −1, dim M = n, π1 M ⊇ G . Then, for any sequence of noncommutative simple groups Gi with Gi = Gj for i = j , we have limi→∞ D(Gi , n) = ∞. Remark 19.4. Theorem 17.23 plays a key role in the proof of Theorem 19.14. So far the author does not know the proof of the conclusion of Theorem 19.14 under the milder assumption RicciM −(n − 1). The trouble is a generalization of Theorem 17.23 to the limit X of the manifolds Mi with RicciMi −δi , where δi → 0. (Namely the problem whether the isometry group of such X is a Lie group or not.) Cheeger–Colding [30] proved that the group of isometries of X is a Lie group under the additional assumption Vol(Mi ) v > 0. Under this additional assumption there is the following result (Anderson [9]): The number of isomorphism classes of π1 M of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Vol(Mi ) v > 0, Diam(M) D, is finite. We here sketch a part of the proof of Theorem 19.12 given in [59]. Namely we assume the splitting theorem and explain how to deduce Theorem 19.12 from it. Here we consider the case Diam(M)KM −n 38 to simplify the argument. We first need a lemma on the convergence of groups. If Γ acts on a metric space X by isometries and p is a base point of Xwe write Γ (D) = γ | d γ (p), p D . Here A is the subgroup generated by A. L EMMA 19.15 [59, Theorem 3.10]. Let (Xi , Γi , pi ) converge to (X, G, q) in the pointed equivariant Hausdorff distance. We assume that the connected component G0 of G is a Lie group and G/G0 is discrete and finitely presented. We also assume that X/G is compact. Moreover we assume that Xi is simply connected and Γi is properly discontinuous and free. 38 If we use Cheeger–Colding’s splitting theorem, a similar argument works. However we need several modifica-

tions on the technical points of the arguments of [59] or given below. Unfortunately the technical details of such arguments is not written in the literature. The author and T. Yamaguchi are planning to write it and make it public in the near future. But maybe it is too technical to be included in this article.

270

K. Fukaya

Then there exists a sequence of normal subgroups Γi,0 converging to G0 such that Γi /Γi,0 ∼ = G/G0 for large i.39 We omit the proof. Now we prove the following P ROPOSITION 19.16. Let (Mi , pi ) converge to (Rk , 0) with respect to the pointed Hausdorff distance. Assume RicciMi −(n − 1). Then there exists > 0 such that the image of π1 (Bpi (, Mi )) in π1 (Bpi (1, Mi )) has a solvable subgroup of finite index for large i. The solvability in Theorem 19.12 is the case k = 0 of Proposition 19.16. (The proof of a more precise statement as in Theorem 19.13 and nilpotency is omitted.) The proof of Proposition 19.16 is by downward induction on k. The case k = dim Mi follows from Theorem 11.3. We assume Proposition 19.16 is correct for k + m (m > 0) and show it for k by contradiction. Let (Mi , pi ) as in Proposition 19.16. We use Theorem 11.3 to find Vi ⊆ Mi and a fiber bundle fi : Vi → B0 (Ci , Rn ) with Ci → ∞. Here Vi ⊇ Bpi (Ci /2, Mi ). Let δi = √ Diam(fi−1 (0)). We take the metric gi,1 = gi / δi . The limit of (Vi , gi,1 ) with respect to the pointed Hausdorff distance is Rk × Z where Z is an Alexandrov space with K 0. Let Γi = π1 (Fi ) = π1 (Vi ). We take ((V˜i , g˜ i,1 ), Γi , p˜ i ) where (V˜i , g˜ i,1 ) is the covering space of Vi equipped with metric induced from gi,1 . Let us take a subsequence and let (V∞ , G, q) be the limit. We apply the splitting Theorem 16.4 to V∞ and find V∞ = R × Y where Y is compact. Since (R × Y )/G ∼ = Rk × Z we find that V∞ = Rk × R−k × Y such that G −k −k acts only on R × Y and (R × Y )/G = Z. Since G is a Lie group by Theorem 17.23, it follows that we can take its connected component G0 . Since G/G0 is discrete and (R−k × Y )/G is compact we can prove that G/G0 has Abelian subgroup of finite index. (This is easy to see if G acts effectively on R−k . The compact factor Y only contributes a finite group.) To apply Lemma 19.15 we replace V∞ = Rk × R−k × Y by X = B0 (D, Rk ) × R−k × Y for large but fixed D and find a sequence ((Xi , dXi ), Γi , pi ) converging to ((X, dX ), G, q). (We can find such Xi ⊂ Vi easily by using the fiber bundle fi .) We now apply Lemma 19.15 to obtain Γi,0 . Since (X, dX ) is an Alexandrov space, it follows from Theorem 17.2 that we can find q near q and ri → ∞ such that ((X, ri dX ), q ) converges to (Rk+m , 0) with m > 0. (Note that since Diam(Z) = 1, it follows that R−k × Y is not a point.) We may replace ((Xi , dXi ), Γi , pi ) by a subsequence which converges to ((Xi , dXi ), Γi , pi ) very quickly compared to 1/ri . Then we find qi such that ((Xi , ri dXi ), Γi , qi ) converges to (Rk+m , G , 0) for some G . Since we can use the fact that Γi,0 converges to G0 , the connected Lie group and the convergence is quick compared to ri to show that Γi,0 is generated by Γi,0 (δi ) = {γ ∈ Γi,0 | d(γ (qi ), qi ) < δi } where δi → 0. Now we apply induction hypothesis. Then if > δi (where is as in Proposition 19.16), we find that Γi,0 (δi ) has a solvable subgroup with finite index. This is a contradiction since Γi /Γi,0 ∼ = G/G0 has an Abelian subgroup with finite index. 39 This lemma is actually weaker than [59, Theorem 3.10]. But it is enough for the present purpose, since we

now have Theorem 17.23.

Metric Riemannian geometry

271

20. Hausdorff convergence of Einstein manifolds In the last four sections, we discuss Gromov–Hausdorff convergence under the assumption Ricci −(n − 1). We first remark that, when we work under the assumption Ricci −(n − 1), the topology can change when we go to the limit, even in the noncollapsing situation, namely in the situation where we assume Vol v > 0. Such a phenomenon was first observed in the study of 4-dimensional Einstein (or complex 2-dimensional Kähler–Einstein) manifold (at least around 20 years ago as far as I know). Let us start by a review of the case of Einstein manifolds. Let Γ ⊂ SU(2) be a finite subgroup. We consider the quotient C2 /Γ . It is a Kähler orbifold with isolated singularity at the origin. (This singularity is called the Kleinian singularity.) There is a resolution 2 /Γ → C2 /Γ . called minimal resolution of the Kleinian singularity which we denote by C 2 /Γ Eguchi–Hanson [49] and others constructed a Ricci flat Kähler metric g on C C2 /Γ which is asymptotically locally Euclidean (in the sense we define later in Definition 20.1). (Such a metric is called a gravitational instanton.) Asymptotically locally Euclidean met 2 /Γ are classified by Kronheimer [98]. rics on C

Suppose (X, gX ) is a 4-dimensional Riemannian orbifold locally of Kleinian type. (Namely X is locally a quotient of C2 by a finite group Γ ⊂ SU(2). We assume also that the metric on X is a quotient metric with respect to a certain Γ -invariant metric locally.) We assume that X is Ricci flat and Kähler. (Namely its Ricci curvature at regular points is 0 and the metric is Kähler at regular point.) We can locally glue the metric gX on X and the Ricci flat Kähler metric g on X to obtain a metric g on the resolution X˜ of X. g is C2 /Γ almost Ricci flat. We can use the technique of Yau’s proof of the Calabi conjecture [158] to show that there exists a Ricci flat Kähler metric on X˜ near g . (See [96,13].) We remark that ˜ g ) and (X, ˜ g ) converges to (X, g) with respect to the Gromov–Hausdorff distance. (X, A typical example is a Kummer surface where we take X = T 4 /Z2 (and Γ = Z2 ). Thus, we have O BSERVATION 20.1. There exists a family of Riemannian manifolds (X, g ), such that Riccig ≡ 0, Vol(X, g ) v > 0 and the limit of (X, g ) as → 0 converges to a compact metric space X which is not a manifold. The construction here is an analogue of Taubes’ construction [145] of anti-self dual connections on 4 manifolds. Later Joyce (see [89]) generalized this construction and used it to construct (higherdimensional) Riemannian manifolds with exceptional holonomy. (They are in particular Ricci flat.) Namely Joyce started, for example, with a 7-dimensional flat orbifold X = T 7 /Γ , which is obtained by T 7 , divided by a finite group of isometries Γ . In his example, the singular locus of X is a codimension 4 totally geodesic smooth submanifold (actually 2 /Z (equipped with direct product it is a disjoint union of T 3 ). Then Joyce glued T 3 × C 2 metric) along a singularity to obtain a Riemannian manifold and used the implicit function theorem to obtain a manifold with exceptional holonomy. In his construction, we also have

272

K. Fukaya

a family of metrics g which is of exceptional holonomy (and in particular is Ricci flat) and which converges to X. A converse to Observation 20.1 is proved by Nakajima and others as follows. T HEOREM 20.2 (Nakajima [106]). Let gi be a sequence of Einstein metrics with Ricci = ±1 or 0, on a 4 manifolds M, such that Vol(M, gi ) v > 0. (Here v is independent of i.) Let X = limGH i→∞ (M, gi ). Then there exists a finite subset S ⊂ X such that X \ S is an Einstein 4-manifold. Moreover, for every δ > 0 there exists a diffeomorphism Φi : X \ Nδ S → M such that the pullback Riemannian metric Φi∗ gi converges to the Riemannian metric on X in C ∞ topology. Remark 20.1. In case of 4-dimensional Einstein manifolds, the L2 -norm of the curvature

Mi

|RMi |n/2 ΩMi =

|RM |2 ΩM

(20.1)

M

is a topological invariant and is estimated by the Euler number. This fact is essential in the proof of Theorem 20.2.40 In case dim Mi > 4, the same conclusion as Theorem 20.2 holds under the additional hypothesis Mi

|RMi |n/2 ΩMi C.

(20.2)

(In case we assume (20.2) we do not need to fix a topological type of M.) Namely under assumption (20.2) and Vol(Mi ) v > 0, the limit space X of a sequence of Einstein manifolds Mi has only finitely many singular points. We remark however the assumption (20.2) is too restrictive to handle the limit of Einstein manifolds. In the example of Joyce mentioned above the limit of a sequence of 7-dimensional Einstein manifolds is T 7 /Z2 whose singularity is 3-dimensional. In this example, the L2 -norm of the Ricci curvature is bounded but the L3.5 norm is not bounded. To study the structure of Mi or X near a singular point ∈ S, we use the scaling argument as follows. For completeness we include the case when dim M is general. Namely we assume we have a sequence of Einstein manifolds Mi converging to X. We assume (20.2) and Vol(Mi ) v > 0. (Then the singular point set S of X is of finite order.) Let pi ∈ Mi which converges to p∞ ∈ S. We scale the metric gMi to Ri gMi so that |KRi gMi | becomes 1 at pi . We then consider the limit ((Mi , Ri gMi ), pi ) with respect to the pointed Gromov– Hausdorff distance. Theorem 16.1 implies that it has a limit, which we denote by (X, gX ). 40 (20.1) is scale invariant if and only if dim M = 4. (We do not need the Einstein condition for this.) In this sense also the situation is very much similar to the study of the Yang–Mills equation in dimension 4. (Compare also the footnote 15 at the beginning of Section 7.)

Metric Riemannian geometry

273

Using the injectivity radius bound we can show that (X, gX ) is a Ricci flat Riemannian manifold. (It is noncompact but complete.) It also satisfies the following condition:

n/2 Ω X X |RX |

C1 , Vol(Bp (R, X)) C2 R n .

(20.3)

(See [14].) We then can apply the following Theorem 20.3. We define D EFINITION 20.1. A complete pointed Riemannian manifold ((X, g), p) is said to be locally almost Euclidean (abbreviated by ALE hereafter) of order τ > 0, if there exists a finite group Γ ⊂ O(n) and a diffeomorphism Φ : X \ Bp (R, X) → (Rn \ B0 (R, Rn ))/Γ such that

−1 ∗

Φ gX − gcan )(x) C|x|−τ ,

(20.4a)

−1−τ −α |(∇ k Φ −1 )∗ gX (x) − (∇ k Φ −1 )∗ gX (y)| C min |x|, |y| , α |x − y|

(20.4b)

holds for some α and R. Here gcan is the metric on Bp (R, X) C2 R n induced by the Euclidean metric on Rn . T HEOREM 20.3 (Bando–Kasue–Nakajima [14]). If (X, gX ) is an n-dimensional Einstein manifold satisfying (20.3) then it is ALE of order n − 1. If (X, gX ) is Einstein–Kähler and n = 4 then it is ALE of order n. Combining them we have T HEOREM 20.4 ([14], Anderson [11]). The limit space X in Theorem 20.2 is an Einstein orbifold.41 In higher dimensions Theorems 23.16, 23.17 give a natural generalization of the results we explained here. If we remove the assumption Vol(Mi ) v > 0 (namely if we study the collapsing situation), then even in the case of Einstein manifold, not so many things are known. This problem is related to mirror symmetry in string theory and is calling attention of several differential geometers working on it. There is a result by Gross–Wilson [76] which discusses the case of K3-surfaces in the collapsing situation and obtains a singular torus fibration. We now consider more general Riemannian manifolds under the condition of Ricci curvature below. To obtain a result similar to Theorem 3.4 we need to avoid the phenomenon we described in Observation 20.1. There are several results assuming a lower bound of injectivity radius, for this purpose. We denote by Sn (D, i > ρ) the set of all isometry classes of n-dimensional compact Riemannian manifolds (without boundary) such that RicciM −(n − 1), Diam(M) D and iM ρ everywhere. Let α ∈ (0, 1). 41 A similar result holds under an additional assumption (20.2).

274

K. Fukaya

T HEOREM 20.5 (Anderson–Cheeger [12]). Let Mi ∈ Sn (D, i > ρ) and X = limGH i→∞ Mi . Then X is a Riemannian manifold of C α -class and there exist diffeomorphisms ϕi : Mi → X such that (ϕi−1 )∗ gMi converges to gX with respect to the C α -norm. Remark 20.2. Under the stronger assumption | RicciM | (n − 1), Diam(M) D and iM ρ, Anderson [7] proved a stronger result. Namely the limit space X is a C 1,α Riemannian manifold and (ϕi−1 )∗ gMi converges to gX with respect to C 1,α -norm. It was applied (in [7]) to prove a sphere theorem and a pinching theorem for an almost Einstein metric. C OROLLARY 20.6. The number of diffeomorphism classes represented by elements of Sn (D, i > ρ) is finite. The proof of Theorem 20.5 is quite similar to the arguments in Sections 6–8. Namely we construct harmonic coordinate and obtain an appropriate estimate. Then the proof is completed by using the diffeotopy extension theorem (or the center of mass technique which we can apply to a smooth metric near the limit C α metric). So the new result in [12] is the following T HEOREM 20.7 [12, Theorem 0.1]. There exists C(n, ρ), (n, ρ) > 0 with the following property. Let M ∈ Sn (D, i > ρ). We can then cover M by harmonic coordinates Ui such that the C 1,α -norm of the coordinate transformation is smaller than C(n, ρ) and the C α norm of the metric tensor written in this coordinate is smaller than C(n, ρ). Moreover for any p ∈ M, the metric ball Bp ((n, ρ), M) is contained in some Ui . 21. Sphere theorem and L2 comparison theorem In the last three sections, we concern with the class of Riemannian manifolds with Ricci curvature bounded from below. Especially we discuss results obtained by Colding and Cheeger–Colding recently. The surveys [45,47,62] and the book [26] are recommended for their results. The basic tool to study such Riemannian manifolds is Theorem 5.2. So we first draw some of its consequences. We put Ap (a, b; M) = x ∈ M | a d(p, x) b , (21.1) Sp (a; M) = x ∈ M | d(p, x) = a . L EMMA 21.1. If RicciM κ, a < b < c, then Vol(Ap (b, c; M)) Vol(Ap (a, b; M)) Vol(Ap0 (a, b; Sn (κ))) Vol(Ap0 (b, c; Sn (κ)))

(21.2)

Vol(Sp (a; M)) Vol(Ap (a, b; M)) Vol(Sp (b; M)) . n n Vol(Sp0 (a; S (κ))) Vol(Ap0 (a, b; S (κ))) Vol(Sp0 (b; Sn (κ)))

(21.3)

and

Metric Riemannian geometry

275

(21.2) follows from Vol(Ap (0, a; M)) + Vol(Ap (a, b; M)) Vol(Ap0 (0, a; Sn (κ))) + Vol(Ap0 (a, b; Sn (κ)))

Vol(Ap (0, a; M)) + Vol(Ap (a, b; M)) + Vol(Ap (b, c; M)) . Vol(Ap0 (0, a; Sn (κ))) + Vol(Ap0 (a, b; Sn (κ))) + Vol(Ap0 (b, c; Sn (κ)))

By taking the limit b → a and b → c in (21.2) we obtain (21.3).

L EMMA 21.2. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then for each x ∈ M we have d(p, x) + d(q, x) − d(p, q) τ (|n). To show the lemma, let δ = d(p, x) + d(q, x) − d(p, q), r = d(p, x) − δ/2, s = d(p, q) − r = d(q, x) − δ/2. Then (Bp (r, M) ∪ Bq (s, M)) ∩ Bx (δ/2, M) = ∅ and Bp (r, M) ∩ Bq (s, M) = ∅. Therefore, by Theorem 5.2, we have Vol(Bp (r, M) ∪ Bq (s, M)) 1 − τ (|n), Vol(M)

Bx (δ/2, M) Cδ n . Vol(M)

Hence δ < τ (|n) as required.

C OROLLARY 21.3. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then Diam M \ Bp (π − , M) < τ (|n). Corollary 21.3 is an immediate consequence of Lemma 21.2 and Myers’ Theorem 5.4. We remark that Corollary 21.3 is a version of Proposition 4.4. Namely the conclusion of Corollary 21.3 is weaker than that of Proposition 4.4, but it holds under milder assumption. L EMMA 21.4. If RicciM (n − 1) = dim M − 1 and if p, q ∈ M with d(p, q) > π − , then Vol(Sp (π − δ; M)) Vol(Sp (δ; M)) + τ (|δ, n). n Vol(Sp0 (δ; S (1))) Vol(Sp0 (π − δ; Sn (1)))

(21.4)

We remark Vol(Sp (δ; M)) Vol(Sp (π − δ; M)) n Vol(Sp0 (δ; S (1))) Vol(Sp0 (π − δ; Sn (1))) is a consequence of (21.1). Hence (21.4) implies that the ratio of volume Vol(Sp (t; M))/ Vol(Sp0 (t; Sn (1))) is almost constant for t ∈ [δ, π − δ].

276

K. Fukaya

Let us prove Lemma 21.4. Let - ρ - δ. By Corollary 21.3, we have Aq (δ − 2ρ, δ − ρ; M) ⊆ Ap π − δ + ρ − τ (|n), π − δ + 2ρ + τ (|n); M .

(21.5)

We may assume ρ − τ (|n) 0. We remark Vol Ap0 δ − 2ρ, δ − ρ; Sn (1) = Vol Ap0 π − δ + ρ, π − δ + 2ρ; Sn (1) . Therefore (21.5) (together with Lemma 21.1) implies the first inequality of Vol(Aq (δ − 2ρ, δ − ρ; M)) Vol(Ap0 (δ − 2ρ, δ − ρ; Sn (1)))

Vol(Ap (π − δ, π − δ + ρ; M)) + τ (|δ, ρ, n) Vol(Ap0 (π − δ, π − δ + ρ; Sn (1)))

Vol(Ap (δ − 2ρ, δ − ρ; M)) + τ (|δ, ρ, n). Vol(Ap0 (δ − 2ρ, δ − ρ; Sn (1)))

(21.6)

Here the second inequality is a consequence of Lemma 21.1. Changing the role of p and q we have Vol(Ap (δ − ρ, δ; M)) Vol(Aq (δ − ρ, δ; M)) + τ (|δ, ρ, n). (21.7) n Vol(Ap0 (δ − ρ, δ; S (1))) Vol(Ap0 (δ − ρ, δ; Sn (1))) Therefore by (21.6), (21.7) and Lemma 21.1 we have Vol(Sp (π − δ; M)) Vol(Sp (δ; M)) + τ (|δ, ρ, n) Vol(Sp0 (π − δ; Sn (1))) Vol(Sp0 (δ; Sn (1))) as required.

L EMMA 21.5. If RicciM κ and p ∈ M then Vol Bp (R, M) Vol Bp0 R, Sn (κ) . This is an immediate consequence of Theorem 5.2.

We next discuss sphere theorems. The sphere theorem appearing here can be regarded as a generalization of Theorem 17.20. So we first sketch its proof. We first remark: L EMMA 21.6 [79,111]. If KM 1, Vol(M) > Vol(S n ) − , then the Gromov–Hausdorff distance between M and S n is smaller than τ (|n).

Metric Riemannian geometry

277

Let p ∈ M. We identify Tp M with Tp∗ S n for a point p∗ ∈ S n . We then define −1 Φ : M → S n by Φ = Expp∗ ◦ Exp−1 p . Note Expp is discontinuous. Using Corollary 21.3, Lemma 21.5 and the Toponogov comparison Theorem 4.7, we can show that Φ is a τ (|n)Hausdorff approximation. We omit the details since we discuss a sharper result (Theorem 21.8) later. Now under the assumption of Lemma 21.6 we can find points p0 , . . . , pn , q0 , . . . , qn ∈ M such that

d(pi , pj ) − π/2 < τ (|n),

d(pi , qi ) − π < τ (|n).

d(pi , qj ) − π/2 < τ (|n), (21.8) i

In fact, if S n ⊆ Rn+1 , the points pi = (0, . . . , 0, 1, 0, . . . , 0), qi = −pi , satisfy (21.8). Hence we can choose pi = Φ(pi ) where Φ : S n → M is an -Hausdorff approximation. Moreover, in case of S n , the canonical embedding IS n : S n → Rn+1 is obtained by IS n (x) = cos d(p0 , x), . . . , cos d(pn , x) .

(21.9)

Now the idea is to embed M in a neighborhood of S n by using a formula similar to (21.9). Namely we first take a smooth function ϕi which is close to d(x, pi ) up to the first derivative, if x ∈ / Bpi (o(), M) ∪ Bqi (o(), M). We then define IM : M → Rn+1 by IM (x) = ϕ0 (x), . . . , ϕn (x) .

(21.10)

We can then prove that d(IM Φ(x), I (x)) < o() and dist TI (x) S n , TIM Φ(x) IM (M) < τ (|n). Here dist in the above formula is a distance as a codimension one linear subspace in Rn+1 . We can use these two formulas to prove that M is diffeomorphic to S n (in a similar way to Section 9). Theorem 17.20 is generalized by Perelman as follows. T HEOREM 21.7 [114]. There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n , then M is homeomorphic to a sphere. Actually Perelman proved that πk (M) = 1 for k < n under the assumption of Theorem 21.7 and applied the generalized Poincaré conjecture. The idea of the proof is hard to explain for the author in this kind of article. So we refer [114] or [159]. We will discuss the proof of a sharper version, Corollary 22.4, in Section 22. Remark 21.1. We remark that a similar sphere theorem replacing volume by diameter does not holds. Actually Anderson [8] and Otsu [110] found examples of manifolds (M, gi ) such

278

K. Fukaya

that Riccigi (n − 1), Vol(M, gi ) v > 0 and Diam(M, gi ) → π but M = S n . (Otsu’s example is S m × S n−m and Anderson’s example is CP n or CP 2 !CP 2 .) We remark that Vol(Mi ) → Vol(S n ) implies Diam Mi → π (under the assumption Riccigi (n − 1)) by the Bishop–Gromov comparison Theorem 5.2. Now we start the review of the works of Colding, who began with the following theorem closely related to Theorems 17.20 and 21.7. T HEOREM 21.8 (Colding [43,44]). Let M be an n-dimensional Riemannian manifold with RicciM (n − 1). (1) If Vol(M) Vol(S n ) − then dGH (M, S n ) < τ (|n). (2) If dGH (M, S n ) < then Vol(M) Vol(S n ) − τ (|n). The proof is somewhat similar to the proof of Theorem 17.20. However we need several new ideas. Especially we need to develop some method to compare IM (21.10) with IS n (21.9). In the situation of the proof of Theorem 17.20, this was done by Toponogov’s comparison theorem. In our situation, Toponogov’s comparison theorem does not apply since there is no sectional curvature bound. Colding developed the L2 -comparison theorem for this purpose. We describe it below. We consider p∗ ∈ S n and ∗ : [0, α] → S n be a geodesic parameterized by arc length. We put ∗ (0) = q∗ , (d∗ /dt)(0) = v∗ ∈ Tq∗ S n . We then put hp∗ ,α (v∗ , t) = cos d(∗ (t), p∗ ). We can calculate it easily as hp∗ ,α (v∗ , t) =

1 d p∗ , (α) sin(α − t) + d p∗ , (0) sin t . sin α

(21.11)

Now we use (21.11) to define a function on M with which we compare the distance function. Let p ∈ M and : [0, α] → M be a geodesic parameterized by arc length. We put (0) = q, (d/dt)(0) = v ∈ Tq M. ( is determined by v so we write = v .) Let f : M → R be a function. We then define hf,α (v, t) =

1 f v (α) sin(α − t) + f v (0) sin t . sin α

(21.12)

We remark that hf,α may be regarded as a function of (v, t) ∈ SM ×[0, α], where SM is the unit tangent bundle SM = {v ∈ T M | |v| = 1}. In case f (x) = d(p, x), we put hf,α = hp,α . Now the L2 -Toponogov theorem in [44] is as follows. T HEOREM 21.9 [44, Proposition 1.15]. Let a0 ∈ [π/2, π). We assume RicciM (n − 1) and p, q ∈ M with d(p, q) π − . Then, for α α0 , we have 1 α Vol(SM)

v∈SM

< τ (|n, α0 ),

cos d p, v (t) − hp,α (v, t) 2 ΩSM dt

α

0

(21.13)

Metric Riemannian geometry

1 α Vol(SM)

279

2 α

d

cos d p, v (t) − dhp,α (v, t) ΩSM dt

dt dt v∈SM 0

< τ (|n, α0 ).

(21.14)

Here ΩSM is the Liouville measure. (Hereafter we omit the symbol of volume form in case it is clear which volume form we use.) Remark 21.2. We remark that (21.13) means that the length d(p, v (t)) is close to the length of the corresponding triangle in S n in L2 -sense. (21.14) means that the angle pv (t)v (0) is close to the angle in corresponding triangle in S n in L2 -sense. Let us explain a part of the ideas of the proof of Theorem 21.9. We first recall the following. Let λ1 (M) denotes the first nonzero eigenvalue of the Laplacian on (the functions of) M. T HEOREM 21.10 (Lichnerowicz [101], Obata [109]). If an n-dimensional Riemannian manifold satisfies RicciM (n − 1) then λ1 (M) −n. The equality holds if and only if M is isometric to the sphere. The proof can be done by the Bochner formula, in the same way as the argument of Step 1 below. We also remark the following theorem by Cheng which is closely related to Theorem 21.10. T HEOREM 21.11 [41]. Let M be a compact Riemannian manifolds with RicciM (n−1). If Diam(M) π , then M is isometric to S n . What is important for us is that the first eigenfunction of S n is cos d(p, ·) and is exactly the function we want study in Theorem 21.10. So the idea of the proof of Theorem 21.10 goes as follows. Step 1. Let f satisfy f + nf < δ, f = 1. (Here is the L2 -norm.) We prove 1 α Vol(SM)

v∈SM

f v (t) − hf,α (v, t) 2 < τ (δ|n),

α

(21.15)

0

and a similar estimate for the t-derivative of cos f (v (t)) − hf,α (v, t) [44, Lemma 1.4]. This step uses the Bochner–Weitzenbeck formula

2

1 |∇f |2 = Hess(f ) + ∇f, ∇f + Ricci(∇f, ∇f ). 2

(21.16)

Here Hess(f )(X, Y ) = X(Y (f )) − (∇X Y )(f ). (Note we are using the positive Laplacian.) The proof is a kind of an “almost version” of the proofs of Theorems 21.10 and 21.11. To

280

K. Fukaya

clarify the geometric ideas, avoiding analytic details, we consider the case

f = λf , n −λ > 0 and prove f (v (t)) = hf,α (v, t). We integrate (21.16) and using M ∇f1 , ∇f2 = − M f1 , f2 , we find

Hess(f ) 2 − |f |2 + (n − 1)|∇f |2 0. M

Since λ

M

|∇f |2 =

M

M ∇f, ∇f = − M

|f |2 , it follows that

Hess(f ) 2 − λ + n − 1 |f |2 0. λ

By Trace Hess(f ) = f and elementary linear algebra, we find λ = −n and Hess(f ) = −f gM .

(21.17)

Using the fact d 2 f (v (t))/dt 2 = Hess f (˙v (t), ˙v (t)) we have d2 f v (t) = −f v (t) . 2 dt

(21.18)

f (v (t)) = hf,α (v, t) follows. Step 2. Let p, q ∈ M with d(p, q) > π − δ. We consider g(x) = cos d(p, x). We then find f with f + nf < δ and f − gL2 < δ. ( L2 is the Sobolev norm, that is an 1

1

L2 -norm up to the first derivative [44, Lemma 1.10].) The essential part of this step (which is explained below) is to show

2 2

n g − |∇g|

τ (δ) Vol(M), M

(21.19)

M

g < τ (δ) Vol(M).

(21.20)

In fact, (21.20) implies that g is almost perpendicular to the 0th eigenfunction of the Laplacian (the constant). Then we can use (21.19) and λ1 n to get the conclusion. Let a(v, t) be as in the proof of Theorem 5.2. We extend it as 0 outside V . (So precisely speaking a(v, t) is the function which we wrote a (v, t) in the proof of Theorem 5.2.) By Lemma 21.4 we have a(v, δ) a(v, π − δ) + τ (|δ, n). (21.21) v∈S n−1

v∈S n−1

On the other hand, the map t → a(v, t) is nondecreasing by the proof of Theorem 5.2. It follows that

(21.22) a(v, s) − a(v, s ) τ ( | δ, n)

v∈S n−1

v∈S n−1

Metric Riemannian geometry

281

for s, s ∈ [δ, π − δ]. Therefore

π

n−1

g =

a(v, t) cos t sin t

n−1 M v∈S t=0

π/2

n−1

=

a(v, t) − a(v, π − t) cos t sin t

v∈S n−1

t=0

τ (, δ|n) Vol(M).

(21.23)

Moreover using |∇g|2 (x) = sin2 d(p, x) we have

|∇g|2 = M

v∈S n−1

On the other hand 2 |g| = M

v∈S n−1

π

a(v, t) sinn+1 t.

t=0

π

a(v, t) cos2 t sinn−1 t.

t=0

π

π We remark 0 sinn+1 t dt = n 0 cos2 t sinn−1 t dt. Hence using (21.22) we can easily show

2

|∇g|2 − n (21.24) |g| < τ (, δ | n) Vol(M).

M

M

(21.23) and (21.24) complete this step as we mentioned before.

These two steps and some more arguments imply Theorem 21.9. (The integral in Theorem 21.9 is taken with respect to the Liouville measure on the unit sphere bundle. In the argument so far the measure is taken with respect to the measure on M itself (or its products). They are equivalent by Theorem 5.2.) We remark that in Theorem 21.9 we use only a weaker assumption Diam(M) ∼ π and not yet Vol(M) ∼ Vol(S n ). (Compare Remark 21.1, which shows that Diam(M) ∼ π does not imply that dGH (M, S n ) is small.) Now using Theorem 21.9, the proof of Theorem 21.8 goes roughly as follows. We first explain (1). Let us assume Vol(M) Vol(S n ) − δ. It then implies that for each p ∈ M there exists q ∈ M such that d(p, q) > π − τ (δ).42 (This follows from the Bishop– Gromov Theorem 5.2.) Now we claim L EMMA 21.12 [44, Lemma 2.25]. Under the assumption of Theorem 21.8(1) there exist pi , qi (i = 0, . . . , n) such that (21.8) holds. 42 Shiohama–Yamaguchi [140] introduced the notion of radius of M that is inf sup d(p, q). This assertion p q

means that the radius of M is close to π .

282

K. Fukaya

Once we have Lemma 21.12, we can construct a Hausdorff approximation Φ : M → S n by perturbing x → (cos d(x, p0 ), . . . , cos d(x, pn )). In fact, by Theorem 21.8, we can prove that the function x → cos d(x, p0 ) behaves in a similar way (modulo τ (δ)) outside the set of measure τ (δ). This is enough to show that it is a Hausdorff approximation. Remark 21.3. As we mentioned before we can use the L2 -comparison theorem directly to show that a map is a Hausdorff approximation. However we cannot use it directly to find a homeomorphism. This is because the L2 -comparison theorem does not tell what happens on a set of small measure. This point is very different from the Toponogov comparison theorem, which however works only under the assumption of sectional curvature. We can use several ‘indirect’ arguments to obtain various topological information using the L2 comparison theorem. (See the next two sections.) The proof of Lemma 21.12 uses Theorem 21.7 and goes as follows. We construct pi , qi (i = 0, . . . , k) satisfying (21.8) by induction on k. Suppose we have pi , qi (i = 0, . . . , k). We then construct a map Φk : M → Rk by x → (cos d(x, p0 ), . . . , cos d(x, pk )). We coni

struct a set Ak from of p0 , . . . , pk , q0 , . . . , qk . In case M = S n and pi = (0, . . . , 0, 1, 0, . . . , 0) Ak = Rk+1 ∩ S n = S k is obtained by joining pi , qi several times along minimal geodesics. We imitate the construction of Ak from pi , qj in M to obtain Ak ⊂ M. (Actually we need to join only by good geodesics v that is a geodesic such that cos d(p, v (t)) − hp,a (v, t) is small. Theorem 21.9 implies that there are enough such geodesics.) Now the restriction of Φk to Ak is similar to the one for S n . Hence Φk (Ak ) lies in a neighborhood of S k and we may regard Ak ∼ = S k . Since k < n, Theorem 21.7 implies that Ak is homotopic to zero in M. This implies that there exists pk+1 ∈ M such that Φk (pk+1 ) = 0. We take qk+1 with d(pk+1 , qk+1 ) > π − δ. Thus induction works. To prove (2) of Theorem 21.8 we proceed as follows. We take pi , qi (i = 0, . . . , n) such that (21.8) holds. (Since dH (M, S n ) is small we can take such pi , qi .) We use it to ˜ ˜ ˜ = (cos d(p0 , x), . . . , cos d(pn , x)), Φ(x) = Φ(x)/| Φ(x)|. construct Φ : M → S n by Φ(x) Using Theorem 21.9, we find that the determinant of the Jacobi matrix of Φ is almost everywhere close to 1. It follows that | Vol(M) − Vol(Φ(M))| < τ (δ|n). We need another idea to show that Vol(S n \ Φ(M)) < τ (δ|n). Actually for this purpose we need a “local version” of Theorem 21.13 [43, Proposition 4.5]. We omit it. The argument of the proof of Theorem 21.8 is a prototype of the argument which is used by Colding and Cheeger–Colding at several other places. We explain them more in the last two sections where the argument is combined with other arguments which are of more analytic nature.

22. Hausdorff convergence and Ricci curvature—I In Section 21, we compared the distance function of a manifold of positive Ricci curvature to the one of the round sphere, in the sense of L21 -norm. In this section, we compare

Metric Riemannian geometry

283

the distance function of a manifold of almost nonnegative Ricci curvature to the one of Euclidean space. T HEOREM 22.1 (Colding [46, Theorem 0.1]). Let Mi be a sequence of n-dimensional Riemannian manifolds with RicciMi −(n − 1) and let M∞ be another n-dimensional Riemannian manifold. We assume limGH i→∞ Mi = M∞ . Then we have lim Vol(Mi ) = Vol(M∞ ).

i→∞

Remark 22.1. Actually Colding proved the following stronger (local) result in [46]. Let Mi and M∞ be complete Riemannian manifolds. We assume RicciMi −(n − 1). Let pi ∈ Mi , p∞ ∈ M∞ , and r > 0. We assume that limGH i→∞ Bpi (r, Mi ) = Bp∞ (r, M∞ ). Then limi→∞ Vol(Bpi (r, Mi )) = Vol(Bp∞ (r, M∞ )). Together with a result by Perelman and using results of controlled surgery, Theorem 22.1 implies the following T HEOREM 22.2 [46]. In the situation of Theorem 22.1, Mi is homotopy equivalent to M∞ for large i. Moreover Mi is homeomorphic to M∞ for large i if n = 3. Remark 22.2. In case the limit space is singular we cannot prove a result similar to Theorem 22.2 because of Example 21.1 by Anderson and Otsu. The Gromov–Hausdorff limit of the metrics Otsu constructed on S 3 × S 2 is a suspension of S 2 × S 2 and hence is not a topological manifold. Theorem 22.1 follows from Theorem 22.2 roughly in the following way. Choose pj∞ ∈ M∞ , j = 1, . . . , N , and small r > 0 such that N

Bpj∞ (r, M∞ ) = M∞ ,

i=1

and 1−δ

Vol(Bpj∞ (r, M∞ )) Vol0 (B0 (r, Rn ))

1 + δ.

(22.1)

Let Φi : M∞ → Mi be an i Hausdorff approximation with i → 0. We take pji = Φi (pj∞ ) ∈ Mi . Since dH (Bpj (r, Mi ), Bpj (r, M∞ )) is small it follows from Theorem 22.1 ∞ i (more precisely its local version stated in Remark 22.1) together with (22.1) that 1 − 2δ

Vol(Bpi (r, M∞ )) j

Vol0 (B0 (r, Rn ))

1 + 2δ.

(22.2)

284

K. Fukaya

We can then apply the method of Perelman that appeared in the proof of Theorem 21.6. It may43 imply that Bpi ((1−)r, M∞ ) is contractible in Bpi (r, M∞ ). This will imply that Mi j j is homotopy equivalent to M in a way similar to the proof of Theorem 3.5 in Section 15. Using control surgery in a way similar to [122] we can prove that Mi is homeomorphic to M. The proof of Theorem 22.2 is not worked out in so much detail in [46]. However we do not need to worry about it at all now, since Cheeger–Colding [29] improved Theorem 22.2 as follows. T HEOREM 22.3 [29, Theorem A.1.12]. In the situation of Theorem 22.1, Mi is diffeomorphic to M∞ for large i. We discuss its proof later in this section. Theorem 22.3 together with Theorem 21.8 immediately imply the following sharpening of Theorem 21.7. (We stated it as Theorem 2.6 in Section 2.) C OROLLARY 22.4 (Cheeger–Colding [29, Theorem A.1.10]). There exists n > 0 such that if M satisfies RicciM (n − 1), Vol(M) Vol(S n ) − n , then M is diffeomorphic to a sphere. Remark 22.3. We remark Theorem 21.8 is used in the proof of Corollary 22.4. The proof of Theorem 21.8 we sketched in the last section uses Theorem 21.7. However we can avoid it as follows. Let RicciMi (n − 1), Vol(Mi ) Vol(S n ) − i , where i → 0. We may assume that Mi converges to a metric space X. Then, by Theorem 23.11, X is isometric to a metric suspension SY , where SY is defined in Example 23.1(3). Using the assumption on Mi and Theorem 22.5, we can show that the tangent cone Tx X of X at any point x ∈ X is Rn . Therefore, since X = SY , it follows that Y = S n−1 . Hence X = S n (isometric) as required. Remark 22.4. The assumption of Theorem 22.2 plus an additional assumption RicciMi λ implies that the Riemannian metric of Mi converges to the one of M in C 1,α -topology (after identifying the manifolds by an appropriate diffeomorphism). ([46, Theorem 0.6].) We now explain some of the ideas of the proof of Theorem 22.1. The main part of the proof is the proof of (2) of the following theorem. T HEOREM 22.5 [46, Theorem 0.8 and Corollary 2.19]. Let M be an n-dimensional Riemannian manifold with RicciM −λ and p ∈ M. (1) If Vol(Bp (1, M)) Vol(B0 (1, Rn )) − , then we have dGH Bp (1, M), B0 (1, Rn ) < τ (, λ|n). 43 I wrote “may” here since Perelman did not state this result explicitly and only says that “The Main Lemma can obviously be modified. . . ” at [114, p. 300]. Indeed it is very likely so. But I did not check it in detail. By the way, Colding quote [119] in place of [114] at [46, p. 478] just before Theorem 0.4. I believe it is a misprint.

Metric Riemannian geometry

285

(2) If dGH (Bp (1, M), B0 (1, Rn )) < , then we have Vol Bp (1, M) Vol B0 (1, Rn ) − τ (, λ|n). An argument to show Theorem 22.1 by using Theorem 22.5(2) is omitted. Let us sketch how to prove Theorem 22.5(2). We will discuss the proof of Theorem 22.5(1) in the next section. We only show the following version. L EMMA 22.6. If M satisfies dGH (Bp (2R, B), B0 (2R, Rn )) < and RicciM −λ then we have Vol Bp (1, M) 1 − τ (, λ, 1/R|n) Vol B1 (1, Rn ) . The argument to show Theorem 22.5(2) using Lemma 22.6 is tricky but technical. (See [46, p. 494].) (Note the inequality of opposite direction (1 − τ ) Vol Bp (1, M) Vol B1 (1, Rn ) is a consequence of Theorem 5.2.) Theorem 22.5 looks similar to Theorem 21.8. The proof of Lemma 22.6 also is similar. We first need a result corresponding to Theorem 21.13. In the proof of Theorem 21.13 we consider the function x → cos d(p, x) in case there exists q with d(p, q) π − δ. Here i , i = 1, . . . , n, instead. we consider the following function b+ n Let Φ : B0 (2R, R ) → Bp (2R, M) be an -Hausdorff approximation. Let qi = i

Φ(0, . . . , 0, 1, 0, . . . , 0) ∈ M. We put bi (x) = d(x, qi ) − d(p, qi ),

(22.3)

and study it in the ball Bp (1, M). We remark that bi may be regarded as an approximation of the Busemann function (Definition 16.6). In the proof of the Cheeger–Gromoll splitting Theorem 16.4, subharmonicity of the Busemann function is the main point. We choose ρ with 1 - ρ - R. We consider bi : Bp (ρ, M) → R such that bi = 0, bi = bi

(22.4a) on ∂Bp (ρ, M).

(22.4b)

In the case of Euclidean space, the Busemann function is nothing but a linear function. So we compare bi with a linear function. We put gi (v, t) = bi (v (t)). P ROPOSITION 22.7. For r α < 1, we have bi − bi L2 (Bp (1,M)) τ, 1

dgi (v, ·) gi (v, α) − gi (v, 0)

(r) −

< τ,

dt α v∈SBp (1,M)

(22.5a) (22.5b)

286

K. Fukaya

∇bi , ∇bj − δij < τ,

(22.5c)

Hess(bi ) < τ.

(22.5d)

Bp (1,M)

Bp (1,M)

Here τ = τ (λ, ρ/R, 1/ρ|n) and L2 is the L2 -norm up to the first derivative. 1

The proof of (22.5a) is based on Li–Shoen’s Poincaré inequality [100] (estimate of the first eigenvalue of Bp (ρ, M)), and the proof of (22.5d) is based on the Bochner– Weitzenbeck formula (19.1) and Cheng–Yau’s gradient estimate [42]. Then (22.5b) follows in a way similar as the proof of Theorem 21.9. We can use it to prove (22.5c). We put Φ˜ = (b1 , . . . , bn ) : Bp (1, M) → Rn . (22.5a), (22.5b) imply that it induces an τ -Hausdorff approximation to B0 (1, Rn ). (22.5c) implies that Φ˜ almost preserves volume. To complete the proof of Lemma 22.6 we need to show that Vol(B0 (1, Rn ) \ ˜ Φ(Bp (1, M))) is small. We can prove it as follows.44 Using (22.5) we can find a point ˜ 0 )) = {p0 }. (See [26, pp. 53–54] for the proof of p0 ∈ Bp (1/2, M) such that Φ˜ −1 (Φ(p this fact.) On the other hand, since Φ˜ is a τ -Hausdorff approximation, it follows that n n ˜ Φ(∂B p (1, M)) ⊂ B1+τ (0, R ) \ B1−τ (0, R ). Hence Φ˜ ∗ : Hn Bp (1, M), ∂Bp (1, M); Z2 → Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 is well defined. Note Hn Bp (1, M), ∂Bp (1, M); Z2 ∼ = Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 ∼ = Z2 . ˜ 0 )) = {p0 } we can show Using Φ˜ −1 (Φ(p Φ˜ ∗ : Hn Bp (1, M), ∂Bp (1, M); Z2 → Hn B1+τ (0, Rn ), B1−τ (0, Rn ); Z2 ˜ p (1, M)) ⊃ B1−τ (0, Rn ). This completes the proof of is nonzero. This implies Φ(B Lemma 22.6. We next sketch the proof of Theorem 22.3 given in [29, Appendix A]. As is mentioned there this proof is similar to the proof by Cheeger [25] of his finiteness theorem using the diffeotopy extension theorem (which we explained briefly in Section 6). 44 Here we follow [26, pp. 53–54]. Colding’s argument in [46] is a bit different.

Metric Riemannian geometry

287

Let us begin with a definition. Let Z be a complete metric space and , r > 0. (n is a positive integer.) D EFINITION 22.1. We say that Z satisfies the R,r,n condition if for each x ∈ Z there exists s < r such that dGH Bx (s, Z), B0 (s, Rn ) < s.

(22.6)

T HEOREM 22.8 (Cheeger–Colding [29, Theorems A.1.2, A.1.3]). For each n there exists n , independent of r, such that the following holds. If Z satisfies the R,r,n condition with < n , then, for each s < r, we can associate a smooth Riemannian manifold Z(s) with the following properties: (1) There exists a homeomorphism ΦZ,s : Z → Z(s) which is C 1−τ (|n) -Hölder continuous. Namely C −1 d(x, y)1+τ (|n) d ΦZ,s (x), ΦZ,s (y) Cd(x, y)1−τ (|n)

(22.7)

for each x, y ∈ Z. Moreover ΦZ,s is an sτ (|n) Hausdorff approximation. (2) Z(s) is ‘well-defined’ and ‘independent’ of s in the following sense. If u s then there exists a diffeomorphism which is C 1−τ (|n) -Hölder continuous in a way independent of t, u. Namely we have C −1 d(x, y)1+τ (|n) d ΦZ,u,s (x), ΦZ,u,s (y) Cd(x, y)1−τ (|n) ,

(22.8)

where C is independent of u, s, x, y. Moreover ΦZ,u,s is an sτ (|n) approximation and satisfies d ΦZ,u,s ◦ ΦZ,s (x), ΦZ,u (x) < sτ (|n).

(22.9)

(3) If Z is a Riemannian manifold then we may choose ΦZ,s to be a diffeomorphism for sufficiently small s. (4) There exists δ(n, r) > 0 depending n and r such that if Z, Z both satisfy R,r,n condition with < n , and if dGH (Z, Z ) < δ(n, r) then there exists a diffeomorphism Ψ : Z(r/2) → Z (r/2) such that e−τ (,δ|r,n)

d(Ψ (x), Ψ (y)) eτ (,δ|r,n) , d(x, y)

d Ψ ◦ ΦZ,r/2 (x), ΦZ ,r/2 (x) < τ (, δ|r, n).

(22.10)

(22.11)

To apply Theorem 22.8 for the proof of Theorem 22.3 we need the following P ROPOSITION 22.9. Let Mi be a sequence of n-dimensional Riemannian manifolds and let M∞ be another Riemannian manifold of the same dimension. We assume

288

K. Fukaya

limGH i→∞ Mi = M∞ . Then for each there exists r such that Mi for large i and M∞ satisfy the R,r,n condition. Proposition 22.9 and Theorem 22.8 immediately imply Theorem 22.3. Let us prove Proposition 22.9. Under the assumption we have r = r(µ) for each µ such that 1−µ

Vol(Bp (r, Mi )) 1+µ Vol(B0 (r, Rn ))

for large i and i = ∞ and any p ∈ Mi . (See (22.2).) Then we apply Theorem 5.2 to obtain 1−µ

Vol(Bp (s, Mi )) 1+µ Vol(B0 (s, Rn ))

for any s r. We now apply Theorem 22.5(1) after scaling to obtain dGH Bp (s, Mi ), B0 (s, Rn ) < sτ (µ|r, n) as required. (Note that since we scale the metric by a factor 1/s > 1/r, the curvature will be Ricci −(1 − n)r 2 . So the curvature assumption in Theorem 22.5 is satisfied if r is small enough.) We remark that the independence of n of r in Theorem 22.8 played a key role here. We now prove of Theorem 22.8. Let 100s < r. We will construct Z(s) first. We remark that by using assumption (22.6) we can find subsets {xi | i ∈ I } ∈ Z such that

Bxi (s, Z) = Z,

(22.12a)

i∈I

! i ∈ I | Bxi (30s, Z) ∩ Bxj (30s, Z) = ∅ N (n)

(22.12b)

for each j ∈ I . Here N (n) is independent of i, s. Let ϕi : Bxi (100s, Z) → B0 (100s, Rn ) be a τ (|n)s-Hausdorff approximation. We have a τ (|n)s Hausdorff approximation ϕi : B0 (100s, Rn ) → Bxi (100s, Z) such that dist(ϕi ◦ ϕi , id) < τ (|n)s and dist(ϕi ◦ ϕi , id) < τ (|n)s. We consider

ϕj i = ϕj ◦ ϕi B

0 (10s,R

n)

: B0 (10s, Rn ) → B0 (35s, Rn )

(22.13)

for i ∩ j with Bxi (30s, Z) ∩ Bxj (30s, Z) = ∅. It satisfies

d ϕj i (x), ϕj i (x) − d(x, y) < τ (|n)s. We here remark the following simple lemma.

(22.14)

Metric Riemannian geometry

289

L EMMA 22.10. If ϕj i satisfies (22.14) then there exists ψj i : B0 (10s, Rn ) → B0 (35s, Rn ) satisfying (22.14) and e−τ (|n)

d(ψj i (x), ψj i (x))

d(x, y) d ϕj i (x), ψj i (x) < sτ (|n),

eτ (|n) ,

(22.15a) (22.15b)

|ψj i |C k < s −k Ck,n .

(22.15c)

Here Ck,n depends only on k and n. The proof is an elementary smoothing argument. We want to construct a smooth manifold by using ψj i as a coordinate transformation. ◦ ψ = ψ but the following holds if B (20s, Z) ∩ B (20s, Z) ∩ It does not satisfy ψkj xi xj ji ki Bxk (20s, Z) = ∅: ◦ ψj i (x), ψki (x) sτ (|n) d ψkj

(22.16)

for x ∈ B0 (20s, Rn ). We can now use the argument of [25] to approximate ψj i by ψj i which satisfies (22.15) and ψkj ◦ ψj i = ψki .

(22.17)

(Note that the number of steps we need to take to achieve (22.17) is controlled by (22.12b).) We thus constructed a manifold Z(t) whose coordinate transformation is ψj i . We can use a partition of unity to modify the standard metric on Rn so that it is compatible with ψj i . Hence Z(t) is a Riemannian manifold. We will construct ΦZ,s : Z → Z(s) later. At this stage we have ΨZ,s : Z → Z(s) which is an sτ (|n)-Hausdorff approximation. We next show the ‘well-definedness’ property (2). We first consider the case u ∈ [s/2, s]. Let us suppose we have Z(u) for u s. We use the symbol ˜ for points, maps, etc. used to construct Z(u). (Namely we write ϕ˜ i˜ , x˜i˜ , etc.) Let Bx˜i˜ (30u, Z) ∩ Bxj (30s, Z) = ∅. We define Ψj i˜ : B0 (20u, Rn ) → B0 (30s, Rn ) by Ψj i˜ = ϕj ◦ ϕ˜ ˜ . It satisfies i

d Ψ ˜ (x), Ψ ˜ (y) − d(x, y) τ ()s 2τ ()u. ji ji Hence we can approximate it by a smooth map Φ ˜ satisfying (22.15c). It is almost comji

patible with the coordinate transformations ψj i ,ψ˜ j˜i˜ . Hence again by an argument similar to [25] (or by using the center of mass technique) we can approximate it by a diffeomorphism Φj i˜ (x) which is exactly compatible with the coordinate transformation. We thus obtain ΦZ,s,u , if u ∈ [s/2, s]. It is also an sτ (|n)-Hausdorff approximation. Let ΦZ,u,s be

290

K. Fukaya

the inverse of it. We remark that we have an inequality e−τ (|n)

d(ΦZ,u,s (x), ΦZ,u,s (y)) eτ (|n) d(x, y)

(22.18)

which is sharper than (22.10) in case u ∈ [s/2, s]. We remark here that the proof of Theorem 22.8(4) is almost the same as this argument. (So we do not discuss it.) Now we continue the proof of (2) for the general u, s. We may assume u = 2−k s and put ΦZ,u,s = ΦZ,u,2u ◦ · · · ◦ ΦZ,s/2,s .

(22.19)

It is a diffeomorphism. We will check (22.10). Let ρ > 0. We first remark that ΦZ,a,b is a bτ (|n)-Hausdorff approximation for a b. (This is because if b = 2k a then ΦZ,a,b is a k −j j =0 τ (|n)2 b-Hausdorff approximation.) We first take such that e−ρ

−λi with λi → 0 and let (X, p) = limi→∞ (Mi , pi ). Suppose X contains a line. Then X is isometric to the direct product R × X . Remark 23.1. The following slightly more general statement is proved. pGH Let RicciMi −λi with λi → 0, and (X, p) = limi→∞ (Mi , pi ). We assume X ∼ = Rk × Y ∼ and that Y contains a line. Then Y = R × Y . Before explaining the outline of the proof, we mention several of its applications. One important application is Theorems 19.12 and 10.5, which we explained already. To state other applications, we need some definitions. D EFINITION 23.1. A measured metric space (X, µ) is a pair of a metric space X and a Borel measure µ on it. In this article we always assume that µ(X) = 1. For a pointed measured metric space (X, p, µ) we assume µ(Bp (1, X)) = 1. For a Riemannian manifold M we use the renormalized volume form µM = ΩM / Vol(M) and regard it as a measured metric space (unless another measure is specified explicitly). For a pointed Riemannian manifold (M, p), we use the renormalized volume form µM = ΩM / Vol(Bp (1, M)). D EFINITION 23.2 [54]. A sequence of measured metric spaces (Xi , µi ) is said to converge to (X, µ) with respect to the measured Gromov–Hausdorff topology which we write as limmGH i→∞ (Xi , µi ) = (X, µ), if there exists a sequence of i -Hausdorff approximations

292

K. Fukaya

ϕi : Xi → X with i → 0, which are Borel measurable and such that, for any continuous function f on X, we have

lim

i→∞ Xi

(f ◦ ϕi ) dµi =

f dµ. X

The pointed measured Gromov–Hausdorff convergence is defined in the same way. (To be precise we need a net in place of a sequence to define a topology. It is an obvious modification and is omitted.) Remark 23.2. In [75, Chapter 3 12 D], Gromov defined a notion of λ convergence for measured metric spaces. It is similar to, but slightly different from, measured Gromov– Hausdorff topology defined above. Namely there is a situation where the support supp µ of the limit measure is different from X. In that case for λ convergence the limit of (Xi , µi ) is (supp µ, µ), and is different from the limit (X, µ) of the measured Gromov– Hausdorff topology. However if (Xi , µi ) = (Mi , µMi ) is a Riemannian manifold and if RicciMi −(n − 1), then the support of the limit measure is always X itself. (We can prove it using the Bishop–Gromov inequality.) So the two definitions coincide to each other. Measured Gromov–Hausdorff convergence was introduced to study spectra of the Laplace operator. We mention it later. L EMMA 23.2 [54]. If limGH i→∞ Xi = X, and if µi is a probability Borel measure on Xi , then there exists a subsequence ki such that limmGH i→∞ (Xki , µki ) = (X, µ). The proof is elementary. We remark that the limit measure µ depends on the choice of the subsequence in general. f In fact, let us consider T 2 = S 1 × S 1 with metric g = dt 2 + 2 f (t)2 ds 2 , where f : S 1 → R+ is a smooth function. Then (T 2 , g ) converges to S 1 with standard metric and measure f dt with respect to the measured Gromov–Hausdorff topology. On the other hand, the limit in the Gromov–Hausdorff distance is independent of f . We denote by Sn (D) the set of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Diam(M) D. We denote by Sn (∞) the set of n-dimensional pointed Rie¯ n (D), S ¯ n (∞) be the closure mannian manifold (M, p) such that RicciM −(n − 1). Let S of Sn (D), Sn (∞) with respect to the Gromov–Hausdorff distance, the pointed Gromov– Hausdorff distance, respectively. ¯ n (D). We next define the singularity set and the regular set of a length space X ∈ S We recall that the sequence (X, Ri dX , x) with Ri → ∞ always has a subsequence such that (X, Ri dX , x) converges with respect to the pointed Gromov–Hausdorff distance (Proposition 16.2). However the limit is not unique. (Such an example is constructed in [29, Section 8].)

Metric Riemannian geometry

293

D EFINITION 23.3. We say that Tx X is a tangent cone of X at x if there exists a sequence Ri → ∞ such that (X, Ri dX , x) converges to (Tx X, o) with respect to the pointed Gromov–Hausdorff distance.45 ¯ n (∞). We say that a point x ∈ X is in Rk if Rk is a tangent D EFINITION 23.4. Let X ∈ S cone Tx X of x. $ We say x is regular if it is in R = k Rk . Otherwise it is said to be singular and we denote by S the set of all singular points. Remark 23.3. This definition coincides with S(X) in Definition 17.7 by Otsu–Shioya in case when X is an Alexandrov space, because of Theorem 22.5. One of the main results by Cheeger–Colding on the limit space X (in the collapsing situation) is the following T HEOREM 23.3 (Cheeger–Colding [29]). µ(S) = 0 for any limit measure µ. $ Remark 23.4. (1) We remark that Theorem 23.3 implies µ(X \ k Rk ) = 0, but does not imply the existence of k such that µ(X \ Rk ) = 0. (2) In Theorem 23.3 the limit measure µ is used. We do not know how to use the Haus¯ n (D) is an dorff measure since it is not known whether the Hausdorff dimension of X ∈ S integer or not. Here is some very brief idea how a statement like Theorem 23.3 follows from Theo¯ n (D) such that Tx X is a Euclidean rem 23.1. We want to find many points x on X ∈ S space. A naive idea to find such a point may be as follows. First we consider a minimal geodesic xy1 and take an interior point on it and put it x1 . Then any tangent cone Tx1 X contains a line and hence splits as Tx1 X ∼ = R × X1 . We may next take a point x2 near o ∈ Tx1 X which is a midpoint of the minimal geodesic. Then Tx2 X1 contains a line and hence Tx2 X1 splits. This process stops after finitely many stages since we can estimate the dimension of the tangent cone by the Bishop–Gromov inequality. Thus we find near x some kind of ‘point’ for which a tangent cone is Rk . This argument however is too much naive to prove Theorem 23.3. So we need to work more seriously. See [29, Section 2]. We remark that Theorem 23.3 can be applied also to the collapsing situation. Namely it can be applied to the limit X of Mi such that Vol(Mi ) → 0. Several other results are proved by Cheeger–Colding in [29,30]. Nevertheless there are yet many things unclear in the collapsing situation. (In other words, the result in the collapsing situation does not seem to be in the final form.) So we do not discuss it here. (We will discuss one of the main results of [31] later.) 45 I am sorry that this terminology is inconsistent with one in Definition 16.4, where T X is called tangent cone x when any such sequence (X, Ri dX , x) converges to (Tx , 0). In this section we follow Cheeger–Colding and use this terminology. In Definition 16.4 we followed Burago–Gromov–Perelman.

294

K. Fukaya

In the noncollapsing case, Cheeger–Colding obtained more precise results. We discuss some of them here. The following theorem is a generalization of Theorem 22.1. We denote by Sn (D, v) the set of n-dimensional Riemannian manifolds M with RicciM −(n − 1), Diam(M) D, Vol(M) v. We also denote by Sn (∞, v) the set of n-dimensional pointed Riemannian manifold (M, p) such that RicciM −(n − 1), Vol(Bp (1, M)) ¯ n (∞, v) be the closure of Sn (D, v), Sn (∞, v) with respect to the ¯ n (D, v), S v. Let S Gromov–Hausdorff distance, the pointed Gromov–Hausdorff distance, respectively. pGH

T HEOREM 23.4 [29, Theorem 5.9]. Let Mi ∈ Sn (∞, v). We assume that limi→∞ (Mi , pi ) = (X, p). Then for any R we have lim Vol Bpi (R, Mi ) = Hn Bp (R, X) .

i→∞

Here Hn denotes the n-dimensional Hausdorff measure. ¯ n (∞, v) then the Hausdorff dimension of X is n. MoreC OROLLARY 23.5 [29]. If X ∈ S over any limit measure µ is equal to a multiple of the n-dimensional Hausdorff measure. Corollary 23.5 follows from Theorem 23.4 easily. We explain an idea of the proof of Theorem 23.4 later in this section. ¯ n (∞, v) and x ∈ X. Then any tangent cone T HEOREM 23.6 [28, Theorem 5.2]. Let X ∈ S Tx X is isometric to a cone CY of some length space Y of diameter π . Remark 23.5. This result is a kind of generalization of the corresponding result Theorem 17.16 on Alexandrov spaces. However it is not asserted that CY is unique. Actually there is a counter example [29, 8.41]. The conclusion of Theorem 23.6 does not hold in the collapsing situation [29, 8.95]. To prove Theorem 23.6 we need another kind of comparison theorem, which we will explain later. To state the next result we need a definition. ¯ n (∞, v). We say that x ∈ R if every tangent cone Tx X D EFINITION 23.5. Let X ∈ S satisfies dGH B0 (1, Tx X), B0 (1, Rn ) < . We put S = X \ R . Remark 23.6. (1) Using Theorems 22.5, 23.4, we can prove that there exists δ such that Vol(Bx (r, X)) (1 − δ) Vol(B0 (r, Rn )) implies x ∈ R . Thus Definition 23.5 is equivalent to Sδ (X) in Definition 17.7. (2) We can easily see that if < , then R is contained in the interior Int R of R .

Metric Riemannian geometry

295

¯ n (∞, v), we can easily show R = Rn . Using it, we can easily (3) In the$case of X ∈ S . prove S = >0 S , R = >0 R . The following is an analogue of Theorems 17.2 and 17.22. ¯ n (∞, v) and if T HEOREM 23.7 [29, Theorem 5.14]. There exists 0 (n) such that if X ∈ S < 0 (n), then there exists a smooth Riemannian manifold Z() and a homeomorphism Φ : Z() → Int R such that C −1 d(x, y)1+τ (|n) < d Φ (x), Φ (y) < Cd(x, y)1−τ (|n) . Theorem 23.7 actually follows easily from Theorem 22.8. Namely we find, for each > and x ∈ R , a positive number r such that dGH (Bx (r, X), B0 (r, Rn )) < . Therefore any compact subset of Int R satisfies condition R ,r,n for some r. Theorem 23.7 then follows from Theorem 22.8. We next prove Theorem 23.4. For simplicity of notation we assume that X is compact. We first prove Hn (S) = 0. Let µ be a limit measure. We remark that there exists C1 , C2 such that for 0 < r 1 we have C1 r n Vol Bp (r, Mi ) C2 r n , C1 r n µ Bp (r, X) C2 r n .

(23.1a) (23.1b)

In fact (23.1a) is a consequence of the Bishop–Gromov inequality and Vol(Mi ) v > 0. Then (23.1b) follows, since µ is a limit measure. By Theorem 23.3 we have µ(S) = 0. Therefore by (23.1b) and the definition of Hausdorff measure we have Hn (S) = 0. It follows that lim Hn (Int R ) = Hn (X).

(23.2)

→0

We can take the disjoint union of finitely many balls U = Hn (X \ U ) < τ (),

n n

ωn rj − H (X)

< τ (),

$ j

Byj (rj , X) ⊂ R such that (23.3a) (23.3b)

j

dGH Byj (rj , X), B0 (rj , Rn ) < 2rj ,

(23.3c)

where ωn = Vol(B0 (1, Rn )). Here (23.3c) $is a consequence of yj ⊂ R . Then, for large i, we have a disjoint union of balls U,i = j Byj,i (rj , Mi ) ⊂ Mi with dGH Byj,i (rj , Mi ), B0 (rj , X) < 3rj .

(23.4)

296

K. Fukaya

Therefore, by (23.3b), (23.3c), (23.4), and Theorem 22.5 we have

n

H (U ) − Vol(U,i ) < τ (, 1/i|n).

(23.5)

We thus proved Hn (X) Vol(Mi ) + τ (, 1/i|n). To prove the opposite inequality, we take finitely many balls Bza (ta , X) such that X⊆

Byj (rj , X) ∪

j

Bza (ta , X),

(23.6a)

j

tan τ ().

(23.6b)

j

Then for large i we find za,i such that Mi =

Byj,i (rj , Mi ) ∪

j

Bza,i (ta,i , Mi ).

(23.7)

j

Since Vol Bza,i (ta,i , Mi ) < Cn tan , j

it follows that Vol(Mi \ U,i ) τ (|n).

(23.8)

Therefore, Hn (X) Vol(Mi ) − τ (, 1/i|n), as required.

We now sketch the proof of Theorem 23.1. We start with the following situation. (A) M is a Riemannian manifold with RicciM −λ with small λ. (B) We assume dGH (BL (z, M), BL (z , X)) < ρ/10 and there is a line containing z . (Here L is large.) (C) Let p, q ∈ M with d(p, q) = 2L with large L. (D) d(z, pq) ρ/3, |d(z, p) − L| ρ/3, |d(z, q) − L| ρ/3. Here M = Mi , where Mi is as in Theorem 23.1 for large i. Such pair of points p, q exists because of (B). We want to find a length space X such that Bz (R, M) is close to a D ball B(x ,0) × (R, R × X ) in R × X with respect to the Hausdorff distance. (Here 1 - R - L.) We use the following function which is an approximation of the Busemann function: b+ (x) = d(p, x) − d(p, z),

b− (x) = d(q, x) − d(q, z).

(23.9)

Metric Riemannian geometry

297

The argument to control them is similar to the proof of Proposition 22.7. However our problem is a bit different from the situation of Proposition 22.7 where Hausdorff approximation is given by assumption. Our situation is similar to Theorem 22.5(1), where we use another assumption (which was the almost maximality of the volume in case of Theorem 22.5(1)) to find Hausdorff approximation. In our case, we use the following theorem by Abresch–Gromoll to obtain some information on b± and improve it by using a similar argument as in the proof of Proposition 22.7. To state the result by Abresch–Gromoll we need a notation. D EFINITION 23.6. For x, p, q ∈ M, an excess E(x; p, q) is by definition E(x; p, q) = d(x, p) + d(x, q) − d(p, q). T HEOREM 23.8 (Abresch–Gromoll [2]). If RicciM −(n−1)λ, d(z, p) L, d(z, q) L and if E(z; p, q) < ρ, then E(x; p, q) < τ (ρ, λ, 1/L|n, R) for any x ∈ Bz (R, M). Remark 23.7. Abresch–Gromoll stated Theorem 23.8 in the case E(z; p, q) = ρ namely the case z ∈ pq. The above form is a modification by Cheeger–Colding [28, Proposition 6.2]. ([26, Theorem 9.1].) Remark 23.8. Abresch–Gromoll used Theorem 23.8 to show the following Theorem 23.9. It seems that Theorem 23.8 is the first comparison theorem established assuming conditions on Ricci curvature only. For a length space M and B ⊂ A ⊆ M we denote by Diam(B ⊂ A) the following number sup {the length of the shortest curve joining p and q in A}. p,q∈B

T HEOREM 23.9 (Abresch–Gromoll [2]). If M is a complete manifold with RicciM 0, inf KM > −∞, and Diam Sp (3R, M) ⊆ Ap (2R, 4R; M) C/R, then M is homotopy equivalent to an interior of a compact manifold with boundary. Theorem 23.9 is proved by Theorem 23.8 and Morse theory of the distance function in a way similar to Theorem 14.6. Let us go back to the discussion of the proof of Theorem 23.1. Theorem 23.8 and b+ + b− = E(x; p, q) − E(z; p, q) imply −ρ b+ + b− τ L−1 , ρ, λ|R, n .

(23.10)

298

K. Fukaya

Using the fact that b± is “almost subharmonic” we have the following formula (23.12). We define b+ : Bz (R, M) → R by b+ = 0, b+ = b+

(23.11a) on ∂Bz (R, M).

(23.11b)

Then, we can prove b+ − b+ L2 (Bz (R,M)) τ L−1 , ρ, λ|R, n . 1

(23.12)

(Here the right-hand side will become small by taking L large, λ, ρ small.) We now consider the Bochner formula 1 |∇b+ | = | Hess b+ |2 + Ricci(∇b+ , ∇b+ ). 2

(23.13)

We remark |∇b+ | = 1. Hence using (23.12) the integral of the left-hand side of (23.13) is small. Since RicciM −(n − 1)λ it follows from (23.13) Bz (R,M)

Bz (R,M)

| Hess b+ | τ L−1 , ρ, λ|R, n ,

(23.14)

|∇b+ | − 1 τ L−1 , ρ, λ|R, n .

(23.15)

We put X = b−1 + (0). Now we will use (23.14), (23.15) to show that Bx (R, M) is close to a R ball in X × R with respect to the Gromov–Hausdorff distance as follows. Let us take y, z ∈ Bx (R, M). Let y0 , z0 ∈ X such that d(y, y0 ) = d(y, X ),

d(z, z0 ) = d(z, X ).

We will prove

d(y, z)2 − d(y0 , z0 )2 − b+ (y) − b+ (z) 2 τ L−1 , ρ, λ|R, n .

(23.16)

(23.16) obviously implies that y → (y0 , b+ (y)) : Bx (R, M) → B(x,0) (R, X × R) is a Hausdorff approximation and hence dGH Bx (R, M), B(x,0) (R, X × R) τ L−1 , ρ, λ|R, n , which is enough to complete the proof of Theorem 23.1. Let us sketch the proof of (23.16). For simplicity we take b+ (z) = 0 and z = z0 . Let : [0, l] → M be a minimal geodesic joining y0 to y. We put Q(t) = d((t), z). Let γt : [0, Q(t)] → M be a minimal geodesic joining z to (t). (See Figure 23.1.)

Metric Riemannian geometry

299

Fig. 23.1.

We put ht (s) = b+ (γt (s)). We remark that d 2 ht (s) = (Hess b+ ) γ˙t (s), γ˙t (s) - 1. 2 ds

(23.17)

On the other hand ht (Q(t)) = b+ (γ (t)) is almost equal to t.46 Hence dht t ds (s) − Q(t) - 1.

(23.18)

(Here we remark that (23.17), (23.18) do not hold pointwise but only after integrating over some domain. We omit the technical difficulty which arises from this point.) By the first variational formula, we have . dht dQ (s) = γ˙t (s), ∇b+ = cos y0 γ (t)z = (t). ds dt

(23.19)

. (Here and hereafter = means “almost” equal.) Hence Q(t) “almost” satisfies the following differential equation: dQ . t = . dt Q(t)

(23.20)

# The solution of (23.20) with initial value Q(0) = d(y0 , z) is Q(t) = d(y0 , z)2 + t 2 . Hence at t = b+ (y) we have (23.16) with z = z0 . (See [26, Chapter 9] or [28, Section 6] for the details of the proof.) Here we say a few words about the proof of Remark 23.1. In this situation we can take not only p, q but also pi , qi , i = 1, . . . , k. Namely p, q are points close to the line on Y and pi , qi are taken as points close to the point on the coordinate axis of Rk . Using them we 46 We can find (d 2 /dt 2 )(b ◦ γ ) is small in the same way as (23.17). Moreover (d 2 /dt 2 )(b ◦ γ )(0) is close + + to 1 by definition and (23.15).

300

K. Fukaya

obtain b+ together with bi+ , i = 1, . . . , k. They all satisfy (23.14), (23.15). Moreover we . j have ∇bi+ , ∇b+ = δij where b0+ = b+ . We define Φ : X → Rk+1 by Φ = (b0+ , . . . , bk+ ). Using it we can construct a pointed Hausdorff approximation X → Φ −1 (0) × Rk in a way similar to the proof of Theorem 23.1. See [29, pp. 425–426], where similar arguments appears. In Section 20 we reviewed several results obtained by the L2 comparison theorem where we compared a manifold with round sphere. In Section 22 we used L2 comparison theorem where the model space was flat Euclidean space. In the proof Theorem 23.1, we compared a manifold with direct product R × X . In [28], Cheeger–Colding developed a comparison theorem where the model space is a warped product (hereafter we call it the warped product comparison theorem) and gave various applications. We first review some of its applications. One of its applications is Theorem 22.5(1). The following is closely related to it. (Theorem 22.5(1) corresponds to the case when Y = S n−1 .) Cheeger–Colding called this theorem ‘volume cone implies metric cone theorem’. T HEOREM 23.10 (Cheeger–Colding [28]). For each there exists δ = δ(, n) with the following property. Let M be an n-dimensional Riemannian manifold with RicciM −δ(n − 1). We assume Vol(Bp (1, M)) Vol(B0 (1, Rn )) (1 + δ) . Vol(Sp (1, M)) Vol(S0 (1, Rn )) Then there exists a length space Y with Diam(Y ) π such that dGH Bp (1, M), B0 (1, CY ) . We remark that CY = ([0, ∞) × Y )/ ∼ where (0, x) ∼ (0, y) with metric defined in Definition 17.2. Another application of the warped product comparison theorem is Theorem 23.11. To state it we define a warped product. D EFINITION 23.7. Let (X, gX ) be a Riemannian manifold and f : (a, b) → R+ be a smooth function. Then the warped product (a, b) ×f X is by definition a product (a, b) × X equipped with the metric dr 2 ⊕ f (r)2 gX , where r is the coordinate of the interval (a, b). We need to define the warped product for general length space also. Let X be a length space and f : (a, b) → R+ be a smooth function. Let : [α, β] → (a, b) × X be a path which is, say, Lipschitz continuous. We put (t) = (r(t), X (t)). We may change the parameter so that X (t) : [α, β] → X is parameterized by arc length. We then define the length L() of : [α, β] ×f X by

β

L() =

2 2 (dr/dt)(t) + f r(t) dt.

α

We thus defined the length space (a, b) ×f X.

Metric Riemannian geometry

301

E XAMPLE 23.1. (1) The simplest case is f ≡ 1. Then the warped product is the direct product. (2) If (a, b) = (0, ∞) and f (r) = r, then the warped product (0, ∞) ×r X is the cone CX minus 0. If moreover X = S n−1 then it is Rn \ {0}. (3) We take (a, b) = (0, π) and f (r) = sin r. In this case the warped product (0, π) ×sin r X is called the metric suspension SX. In particular the metric suspension SS n−1 of a round sphere S n−1 is the round sphere S n . T HEOREM 23.11 [28, Theorem 5.14]. If RicciM (n − 1), dim M = n and if Diam(M) π − then there exists a length space X such that dGH (M, SX) < τ (|n). Remark 23.9. It is not true in general that M is homeomorphic (or homotopy equivalent to) SX. The counter examples are the ones by Anderson and Otsu we mentioned already. There are several other applications, for example, to the study of the cone at infinity. We omit it. We now explain the idea of the proofs of these theorems. The main idea is to use the warped product comparison theorem. To state it we need some preliminary discussion. We begin with a characterization of a warped product. Let f : (a, b) → R+ be a smooth function, we put F(r) =

r

f (t) dt, a

k(r) =

df (r). dr

(23.21)

L EMMA 23.12. Let X be a Riemannian manifold and M = (a, b) ×f X. Then we have Hess(F) = k(r)gM .

(23.22)

E XAMPLE 23.2. (1) In case M = R ×1 X the direct product. F is linear and k = 0. (2) In case M = Rn = CX \ 0 = (0, ∞) ×r X, we have F = r 2 /2 and k(r) = 1. If X = S n−1 , M = Rn then F(x1 , . . . , xn ) = 12 (x12 + · · · + xn2 ) and (23.22) is obvious. (3) In case M = (0, π) ×sin r S n−1 we have F(r) = −k(r) = cos r. Formula (23.22) is (21.17). Let us prove Lemma 23.12. We put ∂r = ∂/∂r. Hess(F)(∂r , ∂r ) = k is obvious since t → (t, p) is a geodesic. Let V be a vector filed of X, which we regard a vector field on M. We have [V , ∂r ] = 0. Since gM (V , V ) = f 2 gX (V , V ) it follows that −gM (∂r , ∇V V ) = gM (∇V ∂r , V ) = gM (∇∂r V , V ) = f kgX (V , V ). On the other hand, V (F) = 0, ∂r (F) = f . Therefore Hess(F)(V , V ) = −(∇V V )(F) = f 2 kgX (V , V ) = kgM (V , V ), as required.

302

K. Fukaya

The warped product comparison theorem is an ‘almost version’ of the following converse to Lemma 23.12. P ROPOSITION 23.13. If M is a Riemannian manifold F : M → (α, β) is a fiber bundle. Suppose that there exists a function which k : M → R such that Hessx (F) = k(x)gM .

(23.23)

We put X = {x ∈ M | F(p) = F(x)}. Then there exists a function f : (a, b) → R+ such that M∼ = (a, b) ×f X (isometry), r(x) f (t) dt, F(x) =

(23.24a) (23.24b)

r(p)

k(x) =

df r(x) . dr

(23.24c)

Here r : M ∼ = (a, b) ×f X → (a, b) is the projection to the first factor. We now state the warped product comparison theorem. Let M be a complete Riemannian manifold and K be a compact subset. We put r(x) = d(x, K) = inf y ∈ K | d(x, y) , AK (a, b, M) = x ∈ M | a < r(x) < b .

(23.25a) (23.25b)

Let f : (a, b) → R+ be a smooth function and we define F(r) and k(r) as in (23.21). We regard r as a function on AK (a, b, M) then F and k are functions on AK (a, b, M) as well. The following assumptions are a generalization of similar formulae we met several times already. For example, (23.14), (23.14) where k(r) = 0, and (21.17) where k(r) = cos r. A SSUMPTION 23.1. There exists F˜ : AK (a, b, M) → (a, b) such that sup |F˜ − F| , 1 Vol(AK (a, b, M)) 1 Vol(AK (a, b, M))

(23.26a) |∇ F˜ − ∇F| ,

(23.26b)

| Hess F˜ − kgM | .

(23.26c)

AK (a,b,M)

AK (a,b,M)

Theorem 23.14 asserts under Assumption 23.1 plus some more (which will follow), that AK (a, b, M) is Gromov–Hausdorff close to some warped product (a, b) ×f X. A SSUMPTION 23.2. M is an n-dimensional complete Riemannian manifold with KM −Λ. Diam(AK (a, b, M)) D. 0 < α < α, 0 < ξ < α − α .

Metric Riemannian geometry

303

For each x ∈ r −1 (a + α ) there exists y ∈ r −1 (b − α ) such that d (x, y) b − a − 2α + .

(23.27)

T HEOREM 23.14 (Cheeger–Colding [28, Theorem 3.6]). Under Assumptions 23.1 and 23.2, there exists a length space X such that dGH AK (a + α, b − α, M), d , (a + α, b − α) ×f X τ (|α , ξ, n, f, D). Remark 23.10. In Assumption 23.2 and Theorem 23.14 we use the symbol d for the metric of subsets of M. Note that the space AK (a, b, M) is not complete. So when we define the metric function d : AK (a, b, M) × AK (a, b, M) → R using the Riemannian metric, we need to be a bit careful. Namely for p, q ∈ AK (a, b, M) we need to take the infimum of the length of the curves joining them in a slightly larger domain. The metric d stands for such a metric. We do not define it since it is too technical. See [28, pp. 205–206]. Let us explain the idea of the proof of Theorem 23.14. Actually the idea is quite similar to one of the proof of (23.16) we discussed already. We take X = r −1 (a + α). To define a metric on it we consider a broken geodesic on a small neighborhood and take the infimum of the length of them. Now we construct the Hausdorff approximation Φ : AK (a + α, b − α, M) → (a + α, b − α) ×f X. Let y, z ∈ AK (a + α, b − α, M). We take y0 , z0 ∈ X so that d(y, y0 ) = d(y, X), d(z, z0 ) = d(z, X). We remark r(x) = d(x, X) − a − α . We put Φ(y) = r(y), y0 , and will prove that Φ is a Hausdorff approximation. We assume z0 = z for simplicity. Let : [0, l] → M be a minimal geodesic joining y0 to y. We put Q(t) = d((t), z). Let γt : [0, Q(t)] → M be a minimal geodesic joining z to (t). Actually there is a technical trouble here. Namely since AK (a + α, b − α, M) is not complete, we may not be able to take γt . By this reason, we need to take a broken geodesic. (See Figure 23.2.) However since this is a technical point, we forget it and assume that we can take γt . ˜ t (s)). By (23.26), ht ‘almost’ satisfies the differential equation We put ht (s) = F(γ d 2 ht . (s) = H ht (s) , ds 2

(23.28)

where H (c) = k(F˜ −1 (c)). We remark that (23.28) is an ordinary differential equation of second order and hence has unique solution under an appropriate boundary condition. Note . that ht (0) = F(a + α), ht (Q(t)) = F(t + a + α). Thus ht is determined by Q(t). (Precisely we have to say that ht is ‘almost’ determined by Q(t) since (23.28) is only ‘almost’ satisfied.) Moreover we have . dQ dht Q(t) = (t) ds dt

(23.29)

304

K. Fukaya

Fig. 23.2.

by the same reason as (23.19). Thus, (23.29) becomes a differential equation of first order on Q and is determined by f . We remark that Q satisfies an initial value condition Q(0) = d(z, y0 ). Therefore, the value of Q at t = r(y) − a − α is determined by this equation and the initial value d(z, y0 ). (Precisely speaking, we can only say the value of Q is almost determined.) By definition Q(r(y) − a − α) = d(y, z). Since it is almost determined by d(z, y0 ) and r(y) and r(z) (which we assumed to be zero for simplicity), it follows that Φ ‘almost’ preserves the length. The fact that a small neighborhood of the image of Φ contains (a + α, b − α) ×f X follows from (23.27). This is a sketch of the proof of Theorem 23.14. We now discuss applications of Theorem 23.14. We first show how we can use Theorem 23.14 to prove Theorem 22.5(1). Let us assume Vol(Bp (1, M)) Vol(B0 (1, Rn )) − and RicciM −λ. We put f (t) = t. Then k(t) ≡ 1, F(t) = t 2 /2. We need to check Assumptions 23.1, 23.2. Put r(x) = d(x, p). We calculate (2 − n) r 1−n = grad r 2−n · dn = r 2−n . (23.30) Sp (R,M)

Sp (R,M)

Bp (R,M)

Since Vol(Bp (1, M)) Vol(B0 (1, Rn )) − it follows from Lemma 21.1 that Sp (R,M) r 1−n n = cn Vol(S p (R, M))/ Vol(S0 (R, R )) is almost independent of R. Hence (23.30) im 2−n plies Bp (R,M)\Bp (δ,M) r is small. Namely r 2−n is almost a harmonic function on Bp (R, M) \ Bp (δ, M). (We remark that r 2−n is harmonic on Rn .) Then we have r 2−n = (2 − n) div r 1−n grad r = (2 − n)r 1−n r + (2 − n)(1 − n)r −n . Hence . r = (n − 1)r −1 .

(23.31)

Metric Riemannian geometry

305

And hence . r 2 = 2 div(r grad r) = 2 + 2rr = 2n.

(23.32)

We now apply (21.16) to r 2 /2 = F and obtain . . 1 n = r 2 = | Hess F|2 + r 2 Ricci(∇r, ∇r). 2

(23.33)

Since RicciM −λ it follows from (23.33) and (23.32) that . Hess F = gM . Hence if F˜ is an harmonic function which approximates F we can check Assumptions 23.1, 23.2. Therefore Theorem 23.14 implies that there exists X such that dGH Ap (2δ, 1 − 2δ, M), d , (2δ, 1 − 2δ) ×r X τ (|δ, λ, n). To complete the proof it suffices to show that X is close to S n−1 with respect to the Gromov–Hausdorff distance. We can do it by looking at the proof of Theorem 23.14 in our case a bit more carefully. Alternatively we can proceed as follows. Take ρ - 1, with δ - ρ n . By the assumption and the Bishop–Gromov inequality we can find p1 , q1 such that 2d(p, p1 ) = 2d(p, q1 ) = d(p1 , q1 ). We use it in the same way as in the proof of Theorem 23.1 to find V1 ⊃ Bp (ρ, M) such that dGH (V1 , [−ρ, ρ] × X1 ) ρτ (, λ|n). We then . . take points p2 , q2 in a neighborhood of X1 such that 2d(p2 , p) = 2d(q2 , p) = d(p2 , q2 ). Then we use it in the same way as in the proof of Remark 23.1 to find V2 ⊃ Bp (ρ 2 , M) such that dGH (V2 , [−ρ, ρ]2 × X2 ) ρ 2 τ (, λ|n). Repeating this n times, we obtain Vn ⊃ Bp (ρ n , M) such that dGH (Vn , [−ρ, ρ]n ) ρ n τ (, λ|n). Since δ - ρ n it then follows that dGH (X, S n−1 ) τ (, λ|n). This implies Theorem 22.5(1). The proof of Theorem 23.10 is similar to the first half of the proof of Theorem 22.5(1) and is omitted. We next explain the proof of Theorem 23.11. Let RicciM (n − 1) and p, q ∈ M with d(p, q) π − . Put f (r) = sin r, r(x) = d(x, p), F(r) = −k(r) = − cos r. By (the proof of) Theorem 21.9 (see (21.17); we remark that f there is our F ), . Hess(F) = k(r)gM . In this way we can check Assumption 23.1. Assumption 23.2 follows from the Bishop– Gromov inequality in this case. We thus can apply Theorem 23.14 and prove Theorem 23.11. We next explain the idea of the proof of Theorem 23.6. Let ((X, dX ), x) = pGH limi→∞ (Mi , xi ) with (Mi , xi ) ∈ Sn (∞, v). We suppose that a tangent cone Tx X = pGH limi→∞ ((X, ri dX ), x) is not a cone.

306

K. Fukaya

Then there exist δ, R, ρ (and a subsequence of ri which we denote by the same symbols) such that dGH Ax δ, R; (X, ri dX ) , Ao (δ, R; CY ) > ρ for a cone CY . We can take ji → ∞ such that dGH Axji δ, R; (Mji , ri gMji ) , Ao (δ, R; CY ) > ρ/2

(23.34)

for a cone CY . We now claim that Vol(Sxji+1 (δ/ri+1 , (Mji+1 , gMji+1 ))) Vol(S0 (δ/ri+1 , Rn )) > (1 + )

Vol(Sxji (R/ri , (Mji , gMji )))

(23.35)

Vol(S0 (R/ri , Rn ))

for independent of i. In fact, if (23.35) does not hold, then we can apply the argument of the first half of the proof of Theorem 22.5(1) to Axji (δ, R; (Mji , ri gMji )) and using Theorem 23.14, we can show that (23.34) does not hold. Now it is easy to deduce a contradiction from (23.35). By taking a subsequence we may assume that δ/ri > R/ri+1 . Then (23.35) and the Bishop–Gromov inequality implies that Vol(Sxji (R/ri , (Mji , gMji ))) Vol(S0 (R/ri , Rn )) > (1 + )i−1

Vol(Sxj1 (R/r1 , (Mj1 , gMj1 ))) Vol(S0 (R/r1 , Rn ))

.

This is a contradiction since the left-hand side is bounded as i → ∞.

(23.36)

In [31], Cheeger–Colding studied a convergence of the eigenvalue of the Laplace operator using the results explained so far. We state their result (without outline of the proof) here. We start with a simple example to illustrate that the measured Hausdorff convergence is related to the eigenvalue of the Laplace operator. Let us consider T 2 = S 1 × S 2 with

f Riemannian metric g = dt 2 + 2 f (t)2 ds 2 . Here f : S 1 → R+ . We assume f dt = 1. f As we mentioned before the limit of (T 2 , g ) with respect to the measured Hausdorff f 1 topology is S with standard metric and measure f dt. The Dirichlet integral on (T 2 , g ) is D(h, h) =

2 2 dh dh 1 f (t) + dt ds. dt f (t) ds

In case we consider the eigenvalues of the Laplacian which stay bounded as → 0, it suffices to consider h which is constant along the s direction. Hence we are to consider the

Metric Riemannian geometry

307

bilinear form on L2 (S 1 ) defined by D(h, h) =

2 dh f (t) dt. dt

In [54] the author proved that a similar phenomenon occurs in the situation we discussed in Section 11. Cheeger–Colding generalized it much and proved the following Theorem 23.15. T HEOREM 23.15 [31, Theorem 7.9]. Let Mi ∈ Sn (D). We assume that it converges to (X, µ) with respect to the measured Hausdorff topology. Then there exists a (unbounded) symmetric bilinear form D on L2 (X, µ) with discrete spectrum λ0 (D) = 0 < λ1 (D) λ2 (D) · · · such that kth eigenvalue λk (−Mi ) of the Laplace operator (on functions) on Mi converges to λk (D). Remark 23.11. (1) In case the multiplicity of eigenvalue λk (D) is m then we put λk (D) = · · · = λk+m−1 (D). (2) The eigenfunction of −Mi converges to the eigenfunctions of D in an appropriate sense. We finally remark that the study of limits of Einstein manifolds (or manifolds with integral bounds of the curvature tensor) we discussed in Section 20 is improved by [32,27], etc. Here we restrict ourselves to quote the following Theorems 23.16, 23.17. Let Mi be a sequence of n-dimensional Riemannian manifolds. We consider the following integral bounds of the curvature for pi ∈ Mi : |RMi |p ΩMi < C, (23.37) Bpi (1,Mi )

where C is independent of i. Let S, Sk be as in Definitions 23.4. Hm is the m-dimensional Hausdorff measure. We say x ∈ S is (n − 4k)-nonexceptional if there exists a tangent cone Tx X which is not isometric to Rn−4k × C(S 4k−1 /Γ ) where Γ ⊂ O(4k) is a finite group acting freely on S 4k−1 . Otherwise x is said to be (n − 4k)-exceptional. Let Nn−4k ⊂ Sn−4k be the set of all (n − 4k)-nonexceptional points. T HEOREM 23.16 ([32, Theorems 1.15, 1.20], [27, Theorem 6.10]). Let Mi ∈ Sn (∞, v) pGH and limi→∞ (Mi , pi ) = (X, p). We assume (23.37) for 1 p n/2 = dim Mi /2: (1) If p is not an integer then Hn−2p (S) = 0. (2) The Hausdorff dimension of S is not greater than n − 2p. (3) If p = 2, then Hn−4 (Nn−4 ) = 0. (4) If Mi are Kähler and p is an integer, then Hn−2p (S ∩ Bp (R, X)) < ∞ for any R. We remark that in case n = 2 and the manifolds Mi are Einstein, Theorem 23.16(3) is Theorem 20.4.

308

K. Fukaya

T HEOREM 23.17 [27, Theorem 11.1]. In the situation of Theorem 23.16 we have (1) If p = 1, then compact subsets of S are (n − 2) rectifiable. (2) If either p = 2k is an even integer, then Nn−4k ∩ Bp (R, X) are (n − 4k)-rectifiable. (3) Mi are Kähler and p is integer, then S ∩ Bp (R, X) are (n − 2p)-rectifiable. We remark that as in 4-dimensional case, if p = 2 and the Mi are Einstein, the condition (23.37) can be written in terms of characteristic classes and hence is a topological one. These results are parallel to the corresponding results in (higher-dimensional) gauge theory [105,146].

References [1] U. Abresch, Über das glatten Riemannisher Metriken, Habilitationsschrift, Reinishcehn FriedrichWillhelms-Universität, Bonn (1988). [2] U. Abresch and D. Gromoll, On complete manifolds with nonnegative Ricci curvature, J. Amer. Math. Soc. 3 (1990), 355–374. [3] U. Abresch and W. Meyer, Injectivity radius estimates and sphere theorems, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 1–47. [4] U. Abresch and V. Schroeder, Graph manifolds, ends of negatively curved spaces and the hyperbolic 120cell space, J. Differential Geom. 35 (1992), 299–336. [5] A.D. Alexandrov, A theorem on triangles in a metric space and some of its applications, Trudy Mat. Inst. Steklov., vol. 38, Izdat. Akad. Nauk SSSR, Moscow (1951), 5–23 (in Russian). [6] A.D. Alexandrov, Über eine Verallgemeinerung der Riemannschen Geometrie, Schriftenreiche der Institut für Mathematik 1 (1957), 33–84. [7] M.T. Anderson, Convergence and rigidity of manifolds under Ricci curvature bounds, Invent. Math. 102 (1990), 429–445. [8] M.T. Anderson, Metrics of positive Ricci curvature with large diameter, Manuscripta Math. 68 (1990), 405–415. [9] M.T. Anderson, Short geodesics and gravitational instantons, J. Differential Geom. 31 (1990), 265–275. [10] M.T. Anderson, Hausdorff perturbation of Ricci flat manifolds and the splitting theorem, Duke Math. J. 68 (1992), 67–82. [11] M.T. Anderson, The L2 structure of moduli spaces of Einstein metrics on 4-manifolds, Geom. Funct. Anal. 2 (1) (1992), 29–89. [12] M.T. Anderson and J. Cheeger, C α -compactness for manifolds with Ricci curvature and injectivity radius bounded below, J. Differential Geom. 35 (1992), 265–281. [13] S. Bando, Bubbling out of Einstein manifolds, Tohoku Math. J. (2) 42 (1990), 205–216; Correction: Tohoku Math. J. (2) 42 (1990), 587–588. [14] S. Bando, A. Kasue and H. Nakajima, On a construction of coordinates at infinity on manifolds with fast curvature decay and maximal volume growth, Invent. Math. 97 (1989), 313–349. [15] J. Bemelmans, Min-Oo and E. Ruh, Smoothing Riemannian metrics, Math. Z. 188 (1984), 69–74. [16] P. Bérard, G. Besson and S. Gallot, Embedding Riemannian manifolds by their heat kernel, Geom. Funct. Anal. 4 (1994), 373–398. [17] V.N. Berestovskij and I.G. Nikolaev, Multidimensional generalized Riemannian spaces, Geometry, IV, Encyclopaedia Math. Sci., vol. 70, Springer, Berlin (1993), 165–243, 245–250. [18] M. Berger, Les variétés Riemanniennes (1/4)-pincées, Ann. Scuola Norm. Sup. Pisa (3) 14 (1960), 161– 170. [19] M. Berger, Riemannian Geometry During the Second Half of the Twentieth Century, University Lecture Series, vol. 17, American Mathematical Society, Providence, RI (2000). [20] M. Berger, A Panoramic View of Riemannian Geometry, Springer (2003).

Metric Riemannian geometry

309

[21] G. Besson, G. Courtrois and S. Gallot, Entropies et rigidités des espaces localement symétrique de courbure strictment négative, Geom. Funct. Anal. 5 (1995), 731–799. [22] Yu. Burago, M. Gromov and G. Perelman, A.D. Alexandrov’s spaces with curvature bounded from below, Uspekhi Mat. Nauk 47 (1992), 3–51, 222 (in Russian); English transl.: Russian Math. Surveys 47 (1992), 1–58. [23] H. Busemann, The Geometry of Geodesics, Academic Press, New York (1955). [24] P. Buser and H. Karcher, Gromov’s almost flat manifolds, Astérisque, vol. 81, Société Mathématique de France, Paris (1981), 148 pp. [25] J. Cheeger, Finiteness theorems for Riemannian manifolds, Amer. J. Math. 92 (1970), 61–74. [26] J. Cheeger, Degeneration of Riemannian Metrics under Ricci Curvature Bounds, Scuola Normale Superiore, Pisa (2001). [27] J. Cheeger, Integral bounds on curvature elliptic estimates and rectifiability of singular sets, Geom. Funct. Anal. 13 (2003), 20–72. [28] J. Cheeger and T. Colding, Lower bounds on Ricci curvature and the almost rigidity of warped products, Ann. of Math. 144 (1996), 189–237. [29] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. I, J. Differential Geom. 46 (1997), 406–480. [30] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. II, J. Differential Geom. 54 (2000), 13–35. [31] J. Cheeger and T. Colding, On the structure of spaces with Ricci curvature bounded below. III, J. Differential Geom. 54 (2000), 37–74. [32] J. Cheeger, T.H. Colding and G. Tian, On the singularities of spaces with bounded Ricci curvature, Geom. Funct. Anal. 12 (2002), 873–914. [33] J. Cheeger and G. Ebin, Comparison Theorems in Riemannian Geometry, North-Holland Mathematical Library, vol. 9, North-Holland/American Elsevier, Amsterdam/New York (1975). [34] J. Cheeger, K. Fukaya and M. Gromov, Nilpotent structures and invariant metrics on collapsed manifolds, J. Amer. Math. Soc. 5 (1992), 327–372. [35] J. Cheeger and D. Gromoll, The splitting theorem for manifolds of nonnegative Ricci curvature, J. Differential Geom. 6 (1971–1972), 119–128. [36] J. Cheeger and D. Gromoll, On the structure of complete manifolds of nonnegative curvature, Ann. of Math. 96 (1972), 413–443. [37] J. Cheeger and D. Gromoll, On the lower bound for the injectivity radius of 1/4-pinched Riemannian manifolds, J. Differential Geom. 15 (1980), 437–442. [38] J. Cheeger and M. Gromov, Collapsing Riemannian manifolds while keeping their curvature bounded. I, J. Differential Geom. 23 (1986), 309–346. [39] J. Cheeger and M. Gromov, Collapsing Riemannian manifolds while keeping their curvature bounded. II, J. Differential Geom. 32 (1990), 269–298. [40] J. Cheeger and X. Rong, Existence of polarized F -structures on collapsed manifolds with bounded curvature and diameter, Geom. Funct. Anal. 6 (1996), 411–429. [41] S.Y. Cheng, Eigenvalue comparison theorems and its geometric applications, Math. Z. 143 (1975), 289– 297. [42] S.Y. Cheng and S.T. Yau, Differential equations on Riemannian manifolds and their geometric applications, Comm. Pure Appl. Math. 28 (1975), 333–354. [43] T. Colding, Large manifolds with positive Ricci curvature, Invent. Math. 124 (1996), 193–214. [44] T. Colding, Shape of manifolds with positive Ricci curvature, Invent. Math. 124 (1996), 175–191. [45] T. Colding, Aspects of Ricci curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 83–98. [46] T. Colding, Ricci curvature and volume convergence, Ann. of Math. (2) 145 (1997), 477–501. [47] T. Colding, Space with Ricci curvature bound, Documenta Mathematica, Extra Volume, ICM 1998 II, Doc. Math. J. DMV (1999), 299–308. [48] R.D. Edwards and R.C. Kirby, Deformations of spaces of imbeddings, Ann. of Math. 93 (1971), 63–88. [49] T. Eguchi and A. Hanson, Asymptotically flat self-dual solutions to Euclidean gravity, Phys. Lett. B 74 (1978), 249–251.

310

K. Fukaya

[50] J.H. Eschenburg, New examples of manifolds with strictly positive curvature, Invent. Math. 66 (1982), 469–480. [51] F. Fang and X. Rong, Positive pinching, volume and second Betti number, Geom. Funct. Anal. 9 (1999), 641–674. [52] K. Fukaya, Theory of convergence of Riemannian orbifolds, Japan. J. Math. 12 (1984), 121–160. [53] K. Fukaya, Collapsing Riemannian manifolds to ones of lower dimensions, J. Differential Geom. 25 (1987), 139–156. [54] K. Fukaya, Collapsing of Riemannian manifolds and eigenvalues of Laplace operator, Invent. Math. 87 (1987), 517–547. [55] K. Fukaya, A boundary of the set of Riemannian manifolds with bounded curvature and diameter, J. Differential Geom. 28 (1988), 1–21. [56] K. Fukaya, Collapsing Riemannian manifolds to ones with lower dimension. II, J. Math. Soc. Japan 41 (1989), 333–356. [57] K. Fukaya, Hausdorff convergence of Riemannian manifolds and its applications, Recent Topics in Differential and Analytic Geometry, Adv. Stud. Pure Math., vol. 18-I, Academic Press, Boston, MA (1990), 143–238. [58] K. Fukaya and T. Yamaguchi, Almost nonpositively curved manifolds, J. Differential Geom. 33 (1991), 67–90. [59] K. Fukaya and T. Yamaguchi, The fundamental groups of almost non-negatively curved manifolds, Ann. of Math. 136 (1992), 253–333. [60] K. Fukaya and T. Yamaguchi, Isometry groups of singular spaces, Math. Z. 216 (1994), 31–44. [61] S. Gallot, A Sobolev inequality and some geometric applications, Spectra of Riemannian Manifolds, Berger, Murakami and Ochiai, eds, Kaigai, Kyoto (1981). [62] S. Gallot, Volumes, courbure de Ricci et convergence des variétés d’après T.H. Colding et Cheeger– Colding, Séminaire Bourbaki, 50èm année, vol. 835 (1997–1998), 1–33. [63] R.E. Greene, Noncompact manifolds of nonnegative curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 99–131. [64] R.E. Greene and H. Wu, Lipschitz convergence of Riemannian manifolds, Pacific J. Math. 131 (1988), 119–141. [65] D. Gromoll, Differenzierbare Strukturen und Metriken positiver Krümmung auf Sphären, Math. Ann. 164 (1966), 353–371. [66] D. Gromoll and W. Myers, On complete open manifolds of positive curvature, Ann. of Math. 90 (1969), 75–90. [67] M. Gromov, Manifolds of negative curvature, J. Differential Geom. 13 (1978), 223–230. [68] M. Gromov, Almost flat manifolds, J. Differential Geom. 13 (1978), 231–241. [69] M. Gromov, Structures métriques pour les variétés riemannienes, J. Lafontaine and P. Pansu, eds, Textes Mathématiques, vol. 1, CEDIC, Paris (1981). [70] M. Gromov, Curvature, diameter and Betti numbers, Comment. Math. Helv. 56 (1981), 179–195. [71] M. Gromov, Groups of polynomial growth and expanding maps, Comment. Math. Helv. 56 (1981), 179– 195. [72] M. Gromov, Volume and bounded cohomology, Inst. Hautes Etudes Sci. Publ. Math. 56 (1982), 5–99. [73] M. Gromov, Filling Riemannian manifolds, J. Differential Geom. 18 (1983), 1–147. [74] M. Gromov, Carnot–Carathéodory spaces seen from within, Sub-Riemannian Geometry, Progress in Mathematics, vol. 144, Birkhäuser, Basel (1996), 79–323. [75] M. Gromov, Metric structures for Riemannian and non-Riemannian spaces, With appendices by M. Katz, P. Pansu and S. Semmes, Translated from the French by Sean Michael Bates, Progress in Mathematics, vol. 152, Birkhäuser Boston, Boston, MA (1999). [76] M. Gross and P. Wilson, Large complex structure limit of K3 surfaces, J. Differential Geom. 55 (2000), 475–546. [77] K. Grove and H. Karcher, How to conjugate C 1 -close group actions, Math. Z. 132 (1973), 11–20. [78] K. Grove and P. Petersen, Bounding homotopy types by geometry, Ann. of Math. (2) 128 (1988), 195–206. [79] K. Grove and P. Petersen, Homotopy types of positively curved manifolds with large volume, Amer. J. Math. 110 (1988), 1183–1188.

Metric Riemannian geometry

311

[80] K. Grove and P. Petersen, Manifolds near the boundary of existence, J. Differential Geom. 33 (1991), 379–394. [81] K. Grove, P. Petersen and J.Y. Wu, Geometric finiteness theorems via controlled topology, Invent. Math. 99 (1990), 205–213. [82] K. Grove and K. Shiohama, A generalized sphere theorem, Ann. of Math. (2) 106 (1977), 201–211. [83] K. Grove and W. Ziller, Cohomogeneity one manifolds with positive Ricci curvature, Invent. Math. 149 (2002), 619–646. [84] T. Hattori, Asymptotic geometry of arithmetic quotients of symmetric spaces, Math. Z. 222 (1996), 247– 277. [85] N. Hitchin, Harmonic spinor, Adv. Math. 14 (1974), 1–55. [86] H. Im Hof and E. Ruh, An equivariant pinching theorem, Comment. Math. Helv. 50 (1975), 389–401. [87] J. Jost and H. Karcher, Geometrische Methoden zur Gewinnung von a-priori-Schranken für harmonische Abbildungen, Manuscripta Math. 40 (1982), 27–77. [88] J. Jost and H. Karcher, Almost linear functions and a priori estimates for harmonic maps, Global Riemannian Geometry (Durham, 1983), Ellis Horwood Ser. Math. Appl., Horwood, Chichester (1984), 148– 155. [89] D. Joyce, Compact Manifolds with Special Holonomy, Oxford Univ. Press (2000). [90] A. Kasue, A convergence theorem for Riemannian manifolds and some applications, Nagoya Math. J. 114 (1989), 21–51. [91] A. Kasue, Spectral convergence of Riemannian manifolds, Tohoku Math. J. (2) 46 (1994), 147–179. [92] A. Kasue, Y. Otsu, T. Sakai, T. Shioya and T. Yamaguchi, Riemannian manifolds and its limit, Mem. Math. Soc. Japan (in Japanese), in press. [93] A. Katsuda, Gromov’s convergence theorem and its application, Nagoya Math. J. 100 (1985), 11–48. [94] W. Klingenberg, Contributions to Riemannian geometry in the large, Ann. of Math. 69 (1959), 654–666. [95] W. Klingenberg and T. Sakai, Injectivity radius estimate for 1/4-pinched manifolds, Arch. Math. (Basel) 34 (1980), 371–376. [96] R. Kobayashi and A. Todorov, Polarized period map for generalized K3 surface and moduli of Einstein metric, Tohoku Math. J. 39 (1987), 341–363. [97] S. Kobayashi and K. Nomizu, Foundation of Differential Geometry, I, II, Interscience (1963–1969). [98] P. Kronheimer, The construction of ALE spaces as hyper Kähler quotients, J. Differential Geom. 29 (1989), 665–683. [99] K. Kuwae, Y. Machigashira and T. Shioya, Sobolev spaces, Laplacian, and heat kernel on Alexandrov spaces, Math. Z. 238 (2001), 269–316. [100] P. Li and R. Schoen, Lp and mean value properties of subharmonic functions on Riemannian manifolds, Acta Math. 153 (1984), 279–301. [101] A. Lichnerowicz, Geometric des Groupes des transformationes, Dunod, Paris (1958). [102] X. Menguy, Noncollapsing examples with positive Ricci curvature and infinite topological type, Geom. Funct. Anal. 10 (2000), 600–627. [103] J. Milnor, Morse Theory, Annals of Mathematics Studies, vol. 51, Princeton Univ. Press, Princeton, NJ (1963). [104] J. Milnor, A note on curvature and fundamental group, J. Differential Geom. 2 (1968), 1–7. [105] H. Nakajima, Compactness of the moduli space of Yang–Mills connections in higher dimensions, J. Math. Soc. Japan 40 (1988), 383–392. [106] H. Nakajima, Hausdorff convergence of Einstein 4-manifolds, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 35 (1988), 411–424. [107] I.G. Nikolaev, Parallel translation and smoothness of the metric of spaces with bounded curvature, Dokl. Akad. Nauk SSSR 250 (1980), 1056–1058. [108] I.G. Nikolaev, A metric characterization of Riemannian spaces, Siberian Adv. Math. 9 (1999), 1–58. [109] M. Obata, Certain condition for a Riemannian manifold to be isometric to a sphere, J. Math. Soc. Japan 14 (1962), 333–340. [110] Y. Otsu, On manifolds of positive Ricci curvature with large diameter, Math. Z. 206 (1991), 255–264. [111] Y. Otsu, K. Shiohama and T. Yamaguchi, A new version of differentiable sphere theorem, Invent. Math. 98 (1989), 219–228.

312

K. Fukaya

[112] Y. Otsu and T. Shioya, The Riemannian structure of Alexandrov spaces, J. Differential Geom. 39 (1994), 629–658. [113] G. Perelman, Elements of Morse theory on Aleksandrov spaces, Algebra i Analiz 5 (1993), 232–241 (in Russian); English transl.: St. Petersburg Math. J. 5 (1994), 205–213. [114] G. Perelman, Manifolds of positive Ricci curvature with almost maximal volume, J. Amer. Math. Soc. 7 (1994), 299–305. [115] G. Perelman, Proof of the soul conjecture of Cheeger and Gromoll, J. Differential Geom. 40 (1) (1994), 209–212. [116] G. Perelman, Spaces with curvature bounded below, Proceedings of the International Congress of Mathematicians, vols. 1, 2 (Zürich, 1994), Birkhäuser, Basel (1995), 517–525. [117] G. Perelman, Collapsing with no proper extremal subsets, Comparison Geometry (Berkeley, CA, 1993– 1994), Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 149–155. [118] G. Perelman, Construction of manifolds of positive Ricci curvature with big volume and large Betti numbers, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 157–163. [119] G. Perelman, A.D. Alexandrov’s spaces with curvatures bounded from below II, preprint. [120] S. Peters, Cheeger’s finiteness theorem for diffeomorphism classes of Riemannian manifolds, J. Reine Angew. Math. 349 (1984), 77–82. [121] S. Peters, Convergence of Riemannian manifolds, Compositio Math. 62 (1987), 3–16. [122] P. Petersen, A finiteness theorem for metric spaces, J. Differential Geom. 31 (1990), 387–395. [123] P. Petersen, Convergence theorems in Riemannian geometry, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 167–202. [124] A. Petrunin, Applications of quasigeodesics and gradient curves, Comparison Geometry (Berkeley, CA, 1993–1994), Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 203–219. [125] A. Petrunin, X. Rong and W. Tuschmann, Collapsing vs. positive pinching, Geom. Funct. Anal. 9 (1999), 699–735. [126] A. Petrunin and W. Tuschmann, Diffeomorphism finiteness, positive pinching, and second homotopy, Geom. Funct. Anal. 9 (1999), 736–774. [127] C. Plaut, Metric spaces of curvature k, Handbook of Geometric Topology, North-Holland, Amsterdam (2002), 819–898. [128] W.A. Poor, Some results on nonnegatively curved manifolds, J. Differential Geom. 9 (1974), 583–600. [129] H.E. Rauch, A contribution to differential geometry in the large, Ann. of Math. (2) 54 (1951), 38–55. [130] X. Rong, The existence of polarized F -structures on volume collapsed 4-manifolds, Geom. Funct. Anal. 3 (1993), 474–501. [131] X. Rong, The limiting eta invariants of collapsed three-manifolds, J. Differential Geom. 37 (1993), 535– 568. [132] X. Rong, On the fundamental groups of manifolds of positive sectional curvature, Ann. of Math. (2) 143 (1996), 397–411. [133] E. Ruh, Almost flat manifolds, J. Differential Geom. 17 (1982), 1–14. [134] V. Schroeder, Rigidity of nonpositively curved graphmanifolds, Math. Ann. 274 (1986), 19–26. [135] J. Sha and D. Yang, Examples of manifolds of positive Ricci curvature, J. Differential Geom. 29 (1989), 95–103. [136] Y. Shikata, On the differentiable pinching problem, Osaka J. Math. 4 (1967), 279–287. [137] K. Shiohama, A sphere theorem for manifolds of positive Ricci curvature, Trans. Amer. Math. Soc. 275 (1983), 811–819. [138] K. Shiohama, An Introduction to the Geometry of Alexandrov Spaces, Lecture Notes Series, vol. 8, Seoul National University, Seoul, Korea (1993). [139] K. Shiohama, Sphere theorems, Handbook of Differential Geometry, vol. I, North-Holland, Amsterdam (2000), 865–903. [140] K. Shiohama and T. Yamaguchi, Positively curved manifolds with restricted diameters, Geometry of Manifolds (Matsumoto, 1988), Perspect. Math., vol. 8, Academic Press, Boston, MA (1989), 345–350. [141] T. Shioya and T. Yamaguchi, Collapsing three-manifolds under a lower curvature bound, J. Differential Geom. 56 (2000), 1–66.

Metric Riemannian geometry

313

[142] L.C. Siebenmann, Deformation of homeomorphisms on stratified sets. I, II, Comment. Math. Helv. 47 (1972), 123–136; ibid. 47 (1972), 137–163. [143] M. Sugimoto and K. Shiohama, On the differentiable pinching problem, Math. Ann. 195 (1971), 1–16. [144] Y. Suyama, Differentiable sphere theorem by curvature pinching, J. Math. Soc. Japan 43 (3) (1991), 527– 553. [145] C. Taubes, Self-dual Yang–Mills connections on non-self-dual 4-manifolds, J.Differential Geom. 17 (1982), 139–170. [146] G. Tian, Gauge theory and calibrated geometry. I, Ann. of Math. (2) 151 (2000), 193–268. [147] J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972), 250–270. [148] V.A. Toponogov, Riemann spaces with curvature bounded below, Uspekhi Mat. Nauk 14 (1) (1959), 87– 130. [149] V.A. Toponogov, Riemannian spaces containing straight lines, Dokl. Akad. Nauk SSSR 127 (1959), 977– 979. [150] A. Weinstein, On the homotopy type of positively-pinched manifolds, Arch. Math. (Basel) 18 (1967), 523– 524. [151] T. Yamaguchi, Manifolds of almost nonnegative Ricci curvarture, J. Differential Geom. 28 (1988), 157– 167. [152] T. Yamaguchi, Lipschitz convergence of manifolds of positive Ricci curvature with large volume, Math. Ann. 284 (1989), 423–436. [153] T. Yamaguchi, Collapsing and pinching under a lower curvature bound, Ann. of Math. (2) 133 (1991), 317–357. [154] T. Yamaguchi, A convergence theorem in the geometry of Alexandrov spaces, Actes de la Table Ronde de Géométrie Différentielle (Luminy, 1992), Sémin. Congr., vol. 1, Soc. Math. France, Paris (1996), 601–642. [155] T. Yamaguchi, Collapsing Riemannian 4-manifolds, Suugaku 52 (2000), 172–186 (in Japanese). [156] T. Yamaguchi, Collapsing 4-manifolds under a Lower curvature bound, Preprint (2002). [157] K. Yano and S. Bochner, Curvature and Betti Numbers, Annals of Mathematics Studies, vol. 32, Princeton Univ. Press, Princeton, NJ (1953). [158] S.T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge–Ampère equation. I, Comm. Pure Appl. Math. 31 (1978), 339–411. [159] S. Zhu, The comparison geometry of Ricci curvature, Comparison Geometry, Math. Sci. Res. Inst. Publ., vol. 30, Cambridge Univ. Press, Cambridge (1997), 221–262.

This page intentionally left blank

CHAPTER 5

Contact Geometry Hansjörg Geiges∗ Mathematisches Institut, Universität zu Köln, Weyertal 86-90, 50931 Köln, Germany E-mail: geiges@math.uni-koeln.de

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . 2. Contact manifolds . . . . . . . . . . . . . . . . . . . 2.1. Contact manifolds and their submanifolds . . . . 2.2. Gray stability and the Moser trick . . . . . . . . 2.3. Contact Hamiltonians . . . . . . . . . . . . . . . 2.4. Darboux’s theorem and neighbourhood theorems 2.5. Isotopy extension theorems . . . . . . . . . . . . 2.6. Approximation theorems . . . . . . . . . . . . . 3. Contact structures on 3-manifolds . . . . . . . . . . . 3.1. An invariant of transverse knots . . . . . . . . . 3.2. Martinet’s construction . . . . . . . . . . . . . . 3.3. 2-plane fields on 3-manifolds . . . . . . . . . . . 3.4. Let’s twist again . . . . . . . . . . . . . . . . . . 3.5. Other existence proofs . . . . . . . . . . . . . . 3.6. Tight and overtwisted . . . . . . . . . . . . . . . 3.7. Classification results . . . . . . . . . . . . . . . 4. A guide to the literature . . . . . . . . . . . . . . . . 4.1. Dimension 3 . . . . . . . . . . . . . . . . . . . . 4.2. Higher dimensions . . . . . . . . . . . . . . . . 4.3. Symplectic fillings . . . . . . . . . . . . . . . . 4.4. Dynamics of the Reeb vector field . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

317 317 319 325 328 329 341 345 351 352 354 356 363 367 370 375 376 377 377 378 378 378

* I am grateful to Fan Ding, Jesús Gonzalo and Federica Pasquotto for their attentive reading of the original

manuscript. I also thank John Etnyre and Stephan Schönenberger for allowing me to use a couple of their figures (viz., Figures 2 and 1 of the present text, respectively). HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 315

This page intentionally left blank

Contact geometry

317

1. Introduction Over the past two decades, contact geometry has undergone a veritable metamorphosis: once the ugly duckling known as ‘the odd-dimensional analogue of symplectic geometry’, it has now evolved into a proud field of study in its own right. As is typical for a period of rapid development in an area of mathematics, there are a fair number of folklore results that every mathematician working in the area knows, but no references that make these results accessible to the novice. I therefore take the present article as an opportunity to take stock of some of that folklore. There are many excellent surveys covering specific aspects of contact geometry (e.g., classification questions in dimension 3, dynamics of the Reeb vector field, various notions of symplectic fillability, transverse and Legendrian knots and links). All these topics deserve to be included in a comprehensive survey, but an attempt to do so here would have left this article in the ‘to appear’ limbo for much too long. Thus, instead of adding yet another survey, my plan here is to cover in detail some of the more fundamental differential topological aspects of contact geometry. In doing so, I have not tried to hide my own idiosyncrasies and preoccupations. Owing to a relatively leisurely pace and constraints of the present format, I have not been able to cover quite as much material as I should have wished. Nonetheless, I hope that the reader of the present handbook chapter will be better prepared to study some of the surveys I alluded to—a guide to these surveys will be provided—and from there to move on to the original literature. A book chapter with comparable aims is Chapter 8 in [1]. It seemed opportune to be brief on topics that are covered extensively there, even if it is done at the cost of leaving out some essential issues. I hope to return to the material of the present chapter in a yet to be written more comprehensive monograph.

2. Contact manifolds Let M be a differential manifold and ξ ⊂ T M a field of hyperplanes on M. Locally such a hyperplane field can always be written as the kernel of a non-vanishing 1-form α. One way to see this is to choose an auxiliary Riemannian metric g on M and then to define α = g(X, .), where X is a local non-zero section of the line bundle ξ ⊥ (the orthogonal complement of ξ in T M). We see that the existence of a globally defined 1-form α with ξ = ker α is equivalent to the orientability (hence triviality) of ξ ⊥ , i.e. the coorientability of ξ . Except for an example below, I shall always assume this condition. If α satisfies the Frobenius integrability condition α ∧ dα = 0, then ξ is an integrable hyperplane field (and vice versa), and its integral submanifolds form a codimension 1 foliation of M. Equivalently, this integrability condition can be written as X, Y ∈ ξ

&⇒

[X, Y ] ∈ ξ.

318

H. Geiges

An integrable hyperplane field is locally of the form dz = 0, where z is a coordinate function on M. Much is known, too, about the global topology of foliations, cf. [100]. Contact structures are in a certain sense the exact opposite of integrable hyperplane fields. D EFINITION 2.1. Let M be a manifold of odd dimension 2n + 1. A contact structure is a maximally non-integrable hyperplane field ξ = ker α ⊂ T M, that is, the defining 1-form α is required to satisfy α ∧ (dα)n = 0 (meaning that it vanishes nowhere). Such a 1-form α is called a contact form. The pair (M, ξ ) is called a contact manifold. R EMARK 2.2. Observe that in this case α ∧ (dα)n is a volume form on M; in particular, M needs to be orientable. The condition α ∧ (dα)n = 0 is independent of the specific choice of α and thus is indeed a property of ξ = ker α: Any other 1-form defining the same hyperplane field must be of the form λα for some smooth function λ : M → R \ {0}, and we have n (λα) ∧ d(λα) = λα ∧ (λ dα + dλ ∧ α)n = λn+1 α ∧ (dα)n = 0. We see that if n is odd, the sign of this volume form depends only on ξ , not the choice of α. This makes it possible, given an orientation of M, to speak of positive and negative contact structures. R EMARK 2.3. An equivalent formulation of the contact condition is that we have (dα)n |ξ = 0. In particular, for every point p ∈ M, the 2n-dimensional subspace ξp ⊂ Tp M is a vector space on which dα defines a skew-symmetric form of maximal rank, that is, (ξp , dα|ξp ) is a symplectic vector space. A consequence of this fact is that there exists a complex bundle structure J : ξ → ξ compatible with dα (see [92, Proposition 2.63]), i.e. a bundle endomorphism satisfying • J 2 = −idξ , • dα(J X, J Y ) = dα(X, Y ) for all X, Y ∈ ξ , • dα(X, J X) > 0 for 0 = X ∈ ξ . R EMARK 2.4. The name ‘contact structure’ has its origins in the fact that one of the first historical sources of contact manifolds are the so-called spaces of contact elements (which in fact have to do with ‘contact’ in the differential geometric sense), see [7] and [45]. In the 3-dimensional case the contact condition can also be formulated as X, Y ∈ ξ linearly independent

&⇒

[X, Y ] ∈ / ξ;

Contact geometry

319

this follows immediately from the equation dα(X, Y ) = X α(Y ) − Y α(X) − α [X, Y ] and the fact that the contact condition (in dimension 3) may be written as dα|ξ = 0. In the present article I shall take it for granted that contact structures are worthwhile objects of study. As I hope to illustrate, this is fully justified by the beautiful mathematics to which they have given rise. For an apology of contact structures in terms of their origin (with hindsight) in physics and the multifarious connections with other areas of mathematics I refer the reader to the historical surveys [87] and [45]. Contact structures may also be justified on the grounds that they are generic objects: A generic 1-form α on an odd-dimensional manifold satisfies the contact condition outside a smooth hypersurface, see [89]. Similarly, a generic 1-form α on a 2n-dimensional manifold satisfies the condition α ∧ (dα)n−1 = 0 outside a submanifold of codimension 3; such ‘even-contact manifolds’ have been studied in [51], for instance, but on the whole their theory is not as rich or well motivated as that of contact structures. D EFINITION 2.5. Associated with a contact form α one has the so-called Reeb vector field Rα , defined by the equations (i) dα(Rα , .) ≡ 0, (ii) α(Rα ) ≡ 1. As a skew-symmetric form of maximal rank 2n, the form dα|Tp M has a 1-dimensional kernel for each p ∈ M 2n+1 . Hence equation (i) defines a unique line field Rα on M. The contact condition α ∧ (dα)n = 0 implies that α is non-trivial on that line field, so a global vector field is defined by the additional normalisation condition (ii). 2.1. Contact manifolds and their submanifolds We begin with some examples of contact manifolds; the simple verification that the listed 1-forms are contact forms is left to the reader. E XAMPLE 2.6. On R2n+1 with Cartesian coordinates (x1 , y1 , . . . , xn , yn , z), the 1-form α1 = dz +

n

xj dyj

j =1

is a contact form. E XAMPLE 2.7. On R2n+1 with polar coordinates (rj , ϕj ) for the (xj , yj )-plane, j = 1, . . . , n, the 1-form α2 = dz +

n j =1

is a contact form.

rj2 dϕj = dz +

n (xj dyj − yj dxj ) j =1

320

H. Geiges

Fig. 1. The contact structure ker(dz + x dy).

D EFINITION 2.8. Two contact manifolds (M1 , ξ1 ) and (M2 , ξ2 ) are called contactomorphic if there is a diffeomorphism f : M1 → M2 with Tf (ξ1 ) = ξ2 , where Tf : T M1 → T M2 denotes the differential of f . If ξi = ker αi , i = 1, 2, this is equivalent to the existence of a nowhere zero function λ : M1 → R such that f ∗ α2 = λα1 . E XAMPLE 2.9. The contact manifolds (R2n+1 , ξi = ker αi ), i = 1, 2, from the preceding examples are contactomorphic. An explicit contactomorphism f with f ∗ α2 = α1 is given by f (x, y, z) = (x + y)/2, (y − x)/2, z + xy/2 , where x and y stand for (x1 , . . . , xn ) and (y1 , . . . , yn ), respectively, and xy stands for j xj yj . Similarly, both these contact structures are contactomorphic to ker(dz − y dxj ). Any of these contact structures is called the standard contact structure on j j R2n+1 . E XAMPLE 2.10. The standard contact structure on the unit sphere S 2n+1 in R2n+2 (with Cartesian coordinates (x1 , y1 , . . . , xn+1 , yn+1 )) is defined by the contact form n+1 (xj dyj − yj dxj ). α0 = j =1

With r denoting the radial coordinate on R2n+2 (that is, r 2 = j (xj2 + yj2 )) one checks easily that α0 ∧ (dα0 )n ∧ r dr = 0 for r = 0. Since S 2n+1 is a level surface of r (or r 2 ), this verifies the contact condition. Alternatively, one may regard S 2n+1 as the unit sphere in Cn+1 with complex structure J (corresponding to complex coordinates zj = xj + iyj , j = 1, . . . , n + 1). Then ξ0 = ker α0

Contact geometry

321

defines at each point p ∈ S 2n+1 the complex (i.e. J -invariant) subspace of Tp S 2n+1 , that is, ξ0 = T S 2n+1 ∩ J T S 2n+1 . This follows from the observation that α = −r dr ◦ J . The hermitian form dα(., J.) on ξ0 is called the Levi form of the hypersurface S 2n+1 ⊂ Cn+1 . The contact condition for ξ corresponds to the positive definiteness of that Levi form, or what in complex analysis is called the strict pseudoconvexity of the hypersurface. For more on the question of pseudoconvexity from the contact geometric viewpoint see [1, Section 8.2]. Beware that the ‘complex structure’ in their Proposition 8.14 is not required to be integrable, i.e. constitutes what is more commonly referred to as an ‘almost complex structure’. D EFINITION 2.11. Let (V , ω) be a symplectic manifold of dimension 2n + 2, that is, ω is a closed (dω = 0) and non-degenerate (ωn+1 = 0) 2-form on V . A vector field X is called a Liouville vector field if LX ω = ω, where L denotes the Lie derivative. With the help of Cartan’s formula LX = d ◦ iX + iX ◦ d this may be rewritten as d(iX ω) = ω. Then the 1-form α = iX ω defines a contact form on any hypersurface M in V transverse to X. Indeed, n α ∧ (dα)n = iX ω ∧ d(iX ω) = iX ω ∧ ωn =

1 iX ωn+1 , n+1

which is a volume form on M ⊂ V provided M is transverse to X. E XAMPLE 2.12.With V = R2n+2 , symplectic form ω = j dxj ∧ dyj , and Liouville vector field X = j (xj ∂xj + yj ∂yj )/2 = r ∂r /2, we recover the standard contact structure on S 2n+1 . For finer issues relating to hypersurfaces in symplectic manifolds transverse to a Liouville vector field I refer the reader to [1, Section 8.2]. Here is a further useful example of contactomorphic manifolds. P ROPOSITION 2.13. For any point p ∈ S 2n+1 , the manifold (S 2n+1 \ {p}, ξ0 ) is contactomorphic to (R2n+1 , ξ2 ). P ROOF. The contact manifold (S 2n+1 , ξ0 ) is a homogeneous space under the natural U(n + 1)-action, so we are free to choose p = (0, . . . , 0, −1). Stereographic projection from p does almost, but not quite yield the desired contactomorphism. Instead, we use a map that is well known in the theory of Siegel domains (cf. [3, Chapter 8]) and that looks a bit like a complex analogue of stereographic projection; this was suggested in [92, Exercise 3.64]. Regard S 2n+1 as the unit sphere in Cn+1 = Cn × C with Cartesian coordinates (z1 , . . . , zn , w) = (z, w). We identify R2n+1 with Cn × R ⊂ Cn × C with coordinates

322

H. Geiges

(ζ1 , . . . , ζn , s) = (ζ, s) = (ζ, Re σ ), where ζj = xj + iyj . Then α2 = ds +

n i (xj dyj − yj dxj ) = ds + (ζ d ζ¯ − ζ¯ dζ ) 2 j =1

and i α0 = (z d z¯ − z¯ dz + w d w¯ − w¯ dw). 2 Now define a smooth map f : S 2n+1 \ {(0, −1)} → R2n+1 by i(w − w) ¯ z . ,− (ζ, s) = f (z, w) = 1+w 2|1 + w|2 Then dw d w¯ i dw i d w¯ i(w − w) ¯ i(w − w) ¯ + + + 2(1 + w) |1 + w|2 2(1 + w) ¯ |1 + w|2 2|1 + w|2 2|1 + w|2 w − w¯ w − w¯ i −dw + d w ¯ + dw + d w ¯ = 1+w 1 + w¯ 2|1 + w|2

f ∗ ds = −

and d z¯ z¯ z − d w¯ 1 + w 1 + w¯ (1 + w) ¯ 2 dz z z¯ − dw − 1 + w¯ 1 + w (1 + w)2 dw d w¯ 1 2 z d z ¯ − z ¯ dz + |z| − . = 1 + w 1 + w¯ |1 + w|2

f (ζ d ζ¯ − ζ¯ dζ ) = ∗

Along S 2n+1 we have |z|2 = 1 − |w|2 = (1 − w)(1 + w) ¯ + (w − w) ¯ = (1 − w)(1 ¯ + w) − (w − w), ¯ whence |z|

2

d w¯ w − w¯ w − w¯ dw − = (1 − w) ¯ dw − dw − (1 − w) d w¯ − d w. ¯ 1 + w 1 + w¯ 1+w 1 + w¯

From these calculations we conclude f ∗ α2 = α0 /|1 + w|2 . So it only remains to show that f is actually a diffeomorphism of S 2n+1 \ {(0, −1)} onto R2n+1 . To that end, consider the map f˜ : (Cn × C) \ Cn × {−1} → (Cn × C) \ Cn × {−i/2}

Contact geometry

323

defined by (ζ, σ ) = f˜(z, w) =

z i w−1 ,− . 1+w 2w+1

This is a biholomorphic map with inverse map (ζ, σ ) →

2ζ 1 + 2iσ . , 1 − 2iσ 1 − 2iσ

We compute Im σ = − =

w−1 w¯ − 1 (w − 1)(w¯ + 1) + (w¯ − 1)(w + 1) − =− 4(w + 1) 4(w¯ + 1) 4|1 + w|2

1 − |w|2 . 2|1 + w|2

Hence for (z, w) ∈ S 2n+1 \ {(0, −1)} we have Im σ =

1 |z|2 = |ζ |2 ; 2|1 + w|2 2

conversely, any point (ζ, σ ) with Im σ = |ζ |2 /2 lies in the image of f˜|S 2n+1 \{(0,−1)} , that is, f˜ restricted to S 2n+1 \ {(0, −1)} is a diffeomorphism onto {Im σ = |ζ |2 /2}. Finally, we compute Re σ = − =−

i(w¯ − 1) (w − 1)(w¯ + 1) − (w¯ − 1)(w + 1) i(w − 1) + = −i 4(w + 1) 4(w¯ + 1) 4|1 + w|2 i(w − w) ¯ , 2|1 + w|2

from which we see that for (z, w) ∈ S 2n+1 \ {(0, −1)} and with (ζ, σ ) = f˜(z, w) we have f (z, w) = (ζ, Re σ ). This concludes the proof. At the beginning of this section I mentioned that one may allow contact structures that are not coorientable, and hence not defined by a global contact form. E XAMPLE 2.14. Let M = Rn+1 × RP n with Cartesian coordinates (x0 , . . . , xn ) on the Rn+1 -factor and homogeneous coordinates [y0 : . . . : yn ] on the RP n -factor. Then n yj dxj ξ = ker j =0

324

H. Geiges

is a well-defined hyperplane field on M, because the 1-form on the right-hand side is well defined up to scaling by a non-zero real constant. On the open submanifold Uk = {yk = 0} ∼ = Rn+1 × Rn of M we have ξ = ker αk with yj dxj αk = dxk + yk j =k

an honest 1-form on Uk . This is the standard contact form of Example 2.6, which proves that ξ is a contact structure on M. If n is even, then M is not orientable, so there can be no global contact form defining ξ (cf. Remark 2.2), i.e. ξ is not coorientable. Notice, however, that a contact structure on a manifold of dimension 2n + 1 with n even is always orientable: the sign of (dα)n |ξ does not depend on the choice of local 1-form defining ξ . If n is odd, then M is orientable, so it would be possible that ξ is the kernel of a globally defined 1-form. However, since the sign of α ∧ (dα)n , for n odd, is independent of the choice of local 1-form defining ξ , it is also conceivable that no global contact form exists. (In fact, this consideration shows that any manifold of dimension 2n + 1, with n odd, admitting a contact structure (coorientable or not) needs to be orientable.) This is indeed what happens, as we shall prove now. P ROPOSITION 2.15. Let (M, ξ ) be the contact manifold of the preceding example. Then T M/ξ can be identified with the canonical line bundle on RP n (pulled back to M). In particular, T M/ξ is a non-trivial line bundle, so ξ is not coorientable. P ROOF. For given y = [y0 : . . . : yn ] ∈ RP n , the vector y0 ∂x0 + · · · + yn ∂xn ∈ Tx Rn+1 is well defined up to a non-zero real factor (and independent of x ∈ Rn+1 ), and hence defines a line y in Tx Rn+1 ∼ = Rn+1 . The set E = (t, x, y): x ∈ Rn+1 , y ∈ RP n , t ∈ y ⊂ T Rn+1 × RP n ⊂ T Rn+1 × RP n = T M with projection (t, x, y) → (x, y) defines a line sub-bundle of T M that restricts to the canonical line bundle over {x}×RP n ≡ RP n for each x ∈ Rn+1 . The canonical line bundle over RP n is well known to be non-trivial [95, p. 16], so the same holds for E. Moreover, E is clearly complementary to ξ , i.e. T M/ξ ∼ = E, since n n n yj dxj yk ∂xk = yj2 = 0. j =0

k=0

j =0

This proves that ξ is not coorientable. To sum up, in the example above we have one of the following two situations: • If n is odd, then M is orientable; ξ is neither orientable nor coorientable. • If n is even, then M is not orientable; ξ is not coorientable, but it is orientable.

Contact geometry

325

We close this section with the definition of the most important types of submanifolds. D EFINITION 2.16. Let (M, ξ ) be a contact manifold. (i) A submanifold L of (M, ξ ) is called an isotropic submanifold if Tx L ⊂ ξx for all x ∈ L. (ii) A submanifold M of M with contact structure ξ is called a contact submanifold if T M ∩ ξ |M = ξ . Observe that if ξ = ker α and i : M → M denotes the inclusion map, then the condition for (M , ξ ) to be a contact submanifold of (M, ξ ) is that ξ = ker(i ∗ α). In particular, ξ ⊂ ξ |M is a symplectic sub-bundle with respect to the symplectic bundle structure on ξ given by dα. The following is a manifestation of the maximal non-integrability of contact structures. P ROPOSITION 2.17. Let (M, ξ ) be a contact manifold of dimension 2n + 1 and L an isotropic submanifold. Then dim L n. P ROOF. Write i for the inclusion of L in M and let α be an (at least locally defined) contact form defining ξ . Then the condition for L to be isotropic becomes i ∗ α ≡ 0. It follows that i ∗ dα ≡ 0. In particular, Tp L ⊂ ξp is an isotropic subspace of the symplectic vector space (ξp , dα|ξp ), i.e. a subspace on which the symplectic form restricts to zero. From Linear Algebra we know that this implies dim Tp L (dim ξp )/2 = n. D EFINITION 2.18. An isotropic submanifold L ⊂ (M 2n+1 , ξ ) of maximal possible dimension n is called a Legendrian submanifold. In particular, in a 3-dimensional contact manifold there are two distinguished types of knots: Legendrian knots on the one hand, transverse1 knots on the other, i.e. knots that are everywhere transverse to the contact structure. If ξ is cooriented by a contact form α and γ : S 1 → (M, ξ = ker α) is oriented, one can speak of a positively or negatively transverse knot, depending on whether α(γ˙ ) > 0 or α(γ˙ ) < 0.

2.2. Gray stability and the Moser trick The Gray stability theorem that we are going to prove in this section says that there are no non-trivial deformations of contact structures on closed manifolds. In fancy language, this means that contact structures on closed manifolds have discrete moduli. First a preparatory lemma. L EMMA 2.19. Let ωt , t ∈ [0, 1], be a smooth family of differential k-forms on a manifold M and (ψt )t∈[0,1] an isotopy of M. Define a time-dependent vector field Xt on M by 1 Some people like to call them ‘transversal knots’, but I adhere to J.H.C. Whitehead’s dictum, as quoted in [64]:

“Transversal is a noun; the adjective is transverse.”

326

H. Geiges

Xt ◦ ψt = ψ˙ t , where the dot denotes derivative with respect to t (so that ψt is the flow of Xt ). Then d (ψ ∗ ωt ) = ψt∗ (ω˙ t + LXt ωt ). dt t P ROOF. For a time-independent k-form ω we have d (ψ ∗ ω) = ψt∗ (LXt ω). dt t This follows by observing that (i) the formula holds for functions, (ii) if it holds for differential forms ω and ω , then also for ω ∧ ω , (iii) if it holds for ω, then also for dω, (iv) locally functions and differentials of functions generate the algebra of differential forms. We then compute ψ ∗ ωt+h − ψt∗ ωt d (ψt∗ ωt ) = lim t+h h→0 dt h = lim

∗ ω ∗ ∗ ∗ ψt+h t+h − ψt+h ωt + ψt+h ωt − ψt ωt

h→0

∗ = lim ψt+h h→0

h ψ ∗ ωt − ψt∗ ωt ωt+h − ωt + lim t+h h→0 h h

= ψt∗ (ω˙ t + LXt ωt ). For that last equality observe (regarding the second summand) that ψt+h = ψht ◦ ψt , where ψht denotes, for fixed t and time-variable h, the flow of the time-dependent vector field Xht := Xt+h ; then apply the result for time-independent k-forms. T HEOREM 2.20 (Gray stability). Let ξt , t ∈ [0, 1], be a smooth family of contact structures on a closed manifold M. Then there is an isotopy (ψt )t∈[0,1] of M such that T ψt (ξ0 ) = ξt

for each t ∈ [0, 1].

P ROOF. The simplest proof of this result rests on what is known as the Moser trick, introduced by J. Moser [96] in the context of stability results for (equicohomologous) volume and symplectic forms. J. Gray’s original proof [61] was based on deformation theory à la Kodaira–Spencer. The idea of the Moser trick is to assume that ψt is the flow of a timedependent vector field Xt . The desired equation for ψt then translates into an equation for Xt . If that equation can be solved, the isotopy ψt is found by integrating Xt ; on a closed manifold the flow of Xt will be globally defined.

Contact geometry

327

Let αt be a smooth family of 1-forms with ker αt = ξt . The equation in the theorem then translates into ψt∗ αt = λt α0 , where λt : M → R+ is a suitable smooth family of smooth functions. Differentiation of this equation with respect to t yields, with the help of the preceding lemma, ψt∗ (α˙ t + LXt αt ) = λ˙ t α0 =

λ˙ t ∗ ψ αt , λt t

or, with the help of Cartan’s formula LX = d ◦ iX + iX ◦ d and with µt =

−1 d dt (log λt ) ◦ ψt ,

ψt∗ α˙ t + d αt (Xt ) + iXt dαt = ψt∗ (µt αt ). If we choose Xt ∈ ξt , this equation will be satisfied if α˙ t + iXt dαt = µt αt .

(2.1)

Plugging in the Reeb vector field Rαt gives α˙ t (Rαt ) = µt .

(2.2)

So we can use (2.2) to define µt , and then the non-degeneracy of dαt |ξt and the fact that Rαt ∈ ker(µt αt − α˙ t ) allow us to find a unique solution Xt ∈ ξt of (2.1). R EMARK 2.21. (1) Contact forms do not satisfy stability, that is, in general one cannot find an isotopy ψt such that ψt∗ αt = α0 . For instance, consider the following family of contact forms on S 3 ⊂ R4 : αt = (x1 dy1 − y1 dx1 ) + (1 + t)(x2 dy2 − y2 dx2 ), where t 0 is a real parameter. The Reeb vector field of αt is Rαt = (x1 ∂y1 − y1 ∂x1 ) +

1 (x2 ∂y2 − y2 ∂x2 ). 1+t

The flow of Rα0 defines the Hopf fibration, in particular all orbits of Rα0 are closed. For t ∈ R+ \ Q, on the other hand, Rαt has only two periodic orbits. So there can be no isotopy with ψt∗ αt = α0 , because such a ψt would also map Rα0 to Rαt . (2) Y. Eliashberg [25] has shown that on the open manifold R3 there are likewise no non-trivial deformations of contact structures, but on S 1 × R2 there does exist a continuum of non-equivalent contact structures. (3) For further applications of this theorem it is useful to observe that at points p ∈ M with α˙ t,p identically zero in t we have Xt (p) ≡ 0, so such points remain stationary under the isotopy ψt .

328

H. Geiges

2.3. Contact Hamiltonians A vector field X on the contact manifold (M, ξ = ker α) is called an infinitesimal automorphism of the contact structure if the local flow of X preserves ξ (the study of such automorphisms was initiated by P. Libermann, cf. [80]). By slight abuse of notation, we denote this flow by ψt ; if M is not closed, ψt (for a fixed t = 0) will not in general be defined on all of M. The condition for X to be an infinitesimal automorphism can be written as T ψt (ξ ) = ξ , which is equivalent to LX α = λα for some function λ : M → R (notice that this condition is independent of the choice of 1-form α defining ξ ). The local flow of X preserves α if and only if LX α = 0. T HEOREM 2.22. With a fixed choice of contact form α there is a one-to-one correspondence between infinitesimal automorphisms X of ξ = ker α and smooth functions H : M → R. The correspondence is given by • X → HX = α(X); • H → XH , defined uniquely by α(XH ) = H and iXH dα = dH (Rα )α − dH . The fact that XH is uniquely defined by the equations in the theorem follows as in the preceding section from the fact that dα is non-degenerate on ξ and Rα ∈ ker(dH (Rα )α − dH ). P ROOF. Let X be an infinitesimal automorphism of ξ . Set HX = α(X) and write dHX + iX dα = LX α = λα with λ : M → R. Applying this last equation to Rα yields dHX (Rα ) = λ. So X satisfies the equations α(X) = HX and iX dα = dHX (Rα )α − dHX . This means that XHX = X. Conversely, given H : M → R and with XH as defined in the theorem, we have LXH α = iXH dα + d α(XH ) = dH (Rα )α, so XH is an infinitesimal automorphism of ξ . Moreover, it is immediate from the defini tions that HXH = α(XH ) = H . C OROLLARY 2.23. Let (M, ξ = ker α) be a closed contact manifold and Ht : M → R, t ∈ [0, 1], a smooth family of functions. Let Xt = XHt be the corresponding family of infinitesimal automorphisms of ξ (defined via the correspondence described in the preceding theorem). Then the globally defined flow ψt of the time-dependent vector field Xt is a contact isotopy of (M, ξ ), that is, ψt∗ α = λt α for some smooth family of functions λt : M → R + . P ROOF. With Lemma 2.19 and the preceding proof we have d (ψt∗ α) = ψt∗ (LXt α) = ψt∗ dHt (Rα )α = µt ψt∗ α dt

Contact geometry

329

with µt = dHt (Rα ) ◦ ψt . Since ψ0 = idM (whence ψ0∗ α = α) this implies that, with t λt = exp µs ds , 0

we have ψt∗ α = λt α.

This corollary will be used in Section 2.5 to prove various isotopy extension theorems from isotopies of special submanifolds to isotopies of the ambient contact manifold. In a similar vein, contact Hamiltonians can be used to show that standard general position arguments from differential topology continue to hold in the contact geometric setting. Another application of contact Hamiltonians is a proof of the fact that the contactomorphism group of a connected contact manifold acts transitively on that manifold [12]. (See [8] for more on the general structure of contactomorphism groups.)

2.4. Darboux’s theorem and neighbourhood theorems The flexibility of contact structures inherent in the Gray stability theorem and the possibility to construct contact isotopies via contact Hamiltonians results in a variety of theorems that can be summed up as saying that there are no local invariants in contact geometry. Such theorems form the theme of the present section. In contrast with Riemannian geometry, for instance, where the local structure coming from the curvature gives rise to a rich theory, the interesting questions in contact geometry thus appear only at the global level. However, it is actually that local flexibility that allows us to prove strong global theorems, such as the existence of contact structures on certain closed manifolds. 2.4.1. Darboux’s theorem T HEOREM 2.24 (Darboux’s theorem). Let α be a contact form on the (2n + 1)-dimensional manifold M and p a point on M. Then there are coordinates x1 , . . . , xn , y1 , . . . , yn , z on a neighbourhood U ⊂ M of p such that α|U = dz +

n

xj dyj .

j =1

P ROOF. We may assume without loss of generality that M = R2n+1 and p = 0 is the origin of R2n+1 . Choose linear coordinates x1 , . . . , xn , y1 , . . . , yn , z on R2n+1 such that α(∂z ) = 1, i∂z dα = 0, 2n+1 : on T0 R dα = n dx ∧ dy . ∂ , ∂ ∈ ker α (j = 1, . . . , n), xj

yj

j =1

j

j

This is simply a matter of linear algebra (the normal form theorem for skew-symmetric forms on a vector space).

330

H. Geiges

Now set α0 = dz +

j

xj dyj and consider the family of 1-forms

αt = (1 − t)α0 + tα,

t ∈ [0, 1],

on R2n+1 . Our choice of coordinates ensures that αt = α,

dαt = dα

at the origin.

Hence, on a sufficiently small neighbourhood of the origin, αt is a contact form for all t ∈ [0, 1]. We now want to use the Moser trick to find an isotopy ψt of a neighbourhood of the origin such that ψt∗ αt = α0 . This aim seems to be in conflict with our earlier remark that contact forms are not stable, but as we shall see presently, locally this equation can always be solved. Indeed, differentiating ψt∗ αt = α0 (and assuming that ψt is the flow of some timedependent vector field Xt ) we find ψt∗ (α˙ t + LXt αt ) = 0, so Xt needs to satisfy α˙ t + d αt (Xt ) + iXt dαt = 0.

(2.3)

Write Xt = Ht Rαt + Yt with Yt ∈ ker αt . Inserting Rαt in (2.3) gives α˙ t (Rαt ) + dHt (Rαt ) = 0.

(2.4)

On a neighbourhood of the origin, a smooth family of functions Ht satisfying (2.4) can always be found by integration, provided only that this neighbourhood has been chosen so small that none of the Rαt has any closed orbits there. Since α˙t is zero at the origin, we may require that Ht (0) = 0 and dHt |0 = 0 for all t ∈ [0, 1]. Once Ht has been chosen, Yt is defined uniquely by (2.3), i.e. by α˙ t + dHt + iYt dαt = 0. Notice that with our assumptions on Ht we have Xt (0) = 0 for all t. Now define ψt to be the local flow of Xt . This local flow fixes the origin, so there it is defined for all t ∈ [0, 1]. Since the domain of definition in R × M of a local flow on a manifold M is always open (cf. [15, 8.11]), we can infer2 that ψt is actually defined for all t ∈ [0, 1] on a sufficiently small neighbourhood of the origin in R2n+1 . This concludes the proof of the theorem (strictly speaking, the local coordinates in the statement of the theorem are the coordinates xj ◦ ψ1−1 , etc.). 2 To be absolutely precise, one ought to work with a family α , t ∈ R, where α ≡ α for t ε and α ≡ α for t t t 0 1 t 1 − ε, i.e. a technical homotopy in the sense of [15]. Then Xt will be defined for all t ∈ R, and the reasoning of [15] can be applied.

Contact geometry

331

R EMARK 2.25. The proof of this result given in [1] is incomplete: It is not possible, as is suggested there, to prove the Darboux theorem for contact forms if one requires Xt ∈ ker αt . 2.4.2. Isotropic submanifolds Let L ⊂ (M, ξ = ker α) be an isotropic submanifold in a contact manifold with cooriented contact structure. Write (T L)⊥ ⊂ ξ |L for the sub-bundle of ξ |L that is symplectically orthogonal to T L with respect to the symplectic bundle structure dα|ξ . The conformal class of this symplectic bundle structure depends only on the contact structure ξ , not on the choice of contact form α defining ξ : If α is replaced by λα for some smooth function λ : M → R+ , then d(λα)|ξ = λ dα|ξ . So the bundle (T L)⊥ is determined by ξ . The fact that L is isotropic implies T L ⊂ (T L)⊥ . Following Weinstein [105], we call the quotient bundle (T L)⊥ /T L with the conformal symplectic structure induced by dα the conformal symplectic normal bundle of L in M and write CSN(M, L) = (T L)⊥ /T L. So the normal bundle N L = (T M|L )/T L of L in M can be split as NL ∼ = (T M|L )/(ξ |L ) ⊕ (ξ |L )/(T L)⊥ ⊕ CSN(M, L). Observe that if dim M = 2n + 1 and dim L = k n, then the ranks of the three summands in this splitting are 1, k and 2(n − k), respectively. Our aim in this section is to show that a neighbourhood of L in M is determined, up to contactomorphism, by the isomorphism type (as a conformal symplectic bundle) of CSN(M, L). The bundle (T M|L )/(ξ |L ) is a trivial line bundle because ξ is cooriented. The bundle (ξ |L )/(T L)⊥ can be identified with the cotangent bundle T ∗ L via the well-defined bundle isomorphism Ψ : (ξ |L )/(T L)⊥ → T ∗ L, Y → iY dα|T L . (Ψ is obviously injective and well defined by the definition of (T L)⊥ , and the ranks of the two bundles are equal.) Although Ψ is well defined on the quotient (ξ |L )/(T L)⊥ , to proceed further we need to choose an isotropic complement of (T L)⊥ in ξ |L . Restricted to each fibre ξp , p ∈ L, such an isotropic complement of (Tp L)⊥ exists. There are two ways to obtain a smooth bundle of such isotropic complements. The first would be to carry over Arnold’s corresponding discussion of Lagrangian subbundles of symplectic bundles [6] to the isotropic case in order to show that the space of isotropic complements of U ⊥ ⊂ V , where U is an isotropic subspace in a symplectic vector space V , is convex. (This argument uses generating functions for isotropic subspaces.) Then by a partition of unity argument the desired complement can be constructed on the bundle level. A slightly more pedestrian approach is to define this isotropic complement with the help of a complex bundle structure J on ξ compatible with dα (cf. Remark 2.3). The condition

332

H. Geiges

dα(X, J X) > 0 for 0 = X ∈ ξ implies that (Tp L)⊥ ∩ J (Tp L) = {0} for all p ∈ L, and so a dimension count shows that J (T L) is indeed a complement of (T L)⊥ in ξ |L . (In a similar vein, CSN(M, L) can be identified as a sub-bundle of ξ , viz., the orthogonal complement of T L ⊕ J (T L) ⊂ ξ with respect to the bundle metric dα(., J.) on ξ .) On the Whitney sum T L ⊕ T ∗ L (for any manifold L) there is a canonical symplectic bundle structure ΩL defined by ΩL,p (X + η, X + η ) = η(X ) − η (X) for X, X ∈ Tp L, η, η ∈ Tp∗ L. L EMMA 2.26. The bundle map idT L ⊕ Ψ : T L ⊕ J (T L), dα → (T L ⊕ T ∗ L, ΩL ) is an isomorphism of symplectic vector bundles. P ROOF. We only need to check that idT L ⊕Ψ is a symplectic bundle map. Let X, X ∈ Tp L and Y, Y ∈ Jp (Tp L). Write Y = Jp Z, Y = Jp Z with Z, Z ∈ Tp L. It follows that dα(Y, Y ) = dα(J Z, J Z ) = dα(Z, Z ) = 0, since L is an isotropic submanifold. For the same reason dα(X, X ) = 0. Hence dα(X + Y, X + Y ) = dα(Y, X ) − dα(Y , X) = Ψ (Y )(X ) − Ψ (Y )(X) = ΩL X + Ψ (Y ), X + Ψ (Y ) .

T HEOREM 2.27. Let (Mi , ξi ), i = 0, 1, be contact manifolds with closed isotropic submanifolds Li . Suppose there is an isomorphism of conformal symplectic normal bundles Φ : CSN(M0 , L0 ) → CSN(M1 , L1 ) that covers a diffeomorphism φ : L0 → L1 . Then φ extends to a contactomorphism ψ : N (L0 ) → N (L1 ) of suitable neighbourhoods N (Li ) of Li such that T ψ|CSN(M0 ,L0 ) and Φ are bundle homotopic (as conformal symplectic bundle isomorphisms). C OROLLARY 2.28. Diffeomorphic (closed) Legendrian submanifolds have contactomorphic neighbourhoods. P ROOF. If Li ⊂ Mi is Legendrian, then CSN(Mi , Li ) has rank 0, so the conditions in the theorem, apart from the existence of a diffeomorphism φ : L0 → L1 , are void. E XAMPLE 2.29. Let S 1 ⊂ (M 3 , ξ ) be a Legendrian knot in a contact 3-manifold. Then with a coordinate θ ∈ [0, 2π] along S 1 and coordinates x, y in slices transverse to S 1 , the contact structure cos θ dx − sin θ dy = 0 provides a model for a neighbourhood of S 1 .

Contact geometry

333

P ROOF OF T HEOREM 2.27. Choose contact forms αi for ξi , i = 0, 1, scaled in such a way that Φ is actually an isomorphism of symplectic vector bundles with respect to the symplectic bundle structures on CSN(Mi , Li ) given by dαi . Here we think of CSN(Mi , Li ) as a sub-bundle of T Mi |Li (rather than as a quotient bundle). We identify (T Mi |Li )/(ξi |Li ) with the trivial line bundle spanned by the Reeb vector field Rαi . In total, this identifies N Li = Rαi ⊕ Ji (T Li ) ⊕ CSN(Mi , Li ) as a sub-bundle of T Mi |Li . Let ΦR : Rα0 → Rα1 be the obvious bundle isomorphism defined by requiring that Rα0 (p) map to Rα1 (φ(p)). Let Ψi : Ji (T Li ) → T ∗ Li be the isomorphism defined by taking the interior product with dαi . Notice that T φ ⊕ (φ ∗ )−1 : (T L0 ⊕ T ∗ L0 , ΩL0 ) → (T L1 ⊕ T ∗ L1 , ΩL1 ) is an isomorphism of symplectic vector bundles. With Lemma 2.26 it follows that T φ ⊕ Ψ1−1 ◦ (φ ∗ )−1 ◦ Ψ0 : T L0 ⊕ J0 (T L0 ), dα0 → T L1 ⊕ J1 (T L1 ), dα1 is an isomorphism of symplectic vector bundles. Now let Φ˜ : N L0 → N L1 be the bundle isomorphism (covering φ) defined by Φ˜ = ΦR ⊕ Ψ1−1 ◦ (φ ∗ )−1 ◦ Ψ0 ⊕ Φ. Let τi : N Li → Mi be tubular maps, that is, the τ (I suppress the index i for better readability) are embeddings such that τ |L —where L is identified with the zero section of N L—is the inclusion L ⊂ M, and T τ induces the identity on N L along L (with respect to the splittings T (N L)|L = T L ⊕ N L = T M|L ). Then τ1 ◦ Φ˜ ◦ τ0−1 : N (L0 ) → N (L1 ) is a diffeomorphism of suitable neighbourhoods N (Li ) of Li that induces the bundle map T φ ⊕ Φ˜ : T M0 |L0 → T M1 |L1 . By construction, this bundle map pulls α1 back to α0 and dα1 to dα0 . Hence, α0 and (τ1 ◦ Φ˜ ◦ τ0−1 )∗ α1 are contact forms on N (L0 ) that coincide on T M0 |L0 , and so do their differentials. Now consider the family of 1-forms ∗ βt = (1 − t)α0 + t τ1 ◦ Φ˜ ◦ τ0−1 α1 ,

t ∈ [0, 1].

334

H. Geiges

On T M0 |L0 we have βt ≡ α0 and dβt ≡ dα0 . Since the contact condition α ∧ (dα)n = 0 is an open condition, we may assume—shrinking N (L0 ) if necessary—that βt is a contact form on N (L0 ) for all t ∈ [0, 1]. By the Gray stability theorem (Theorem 2.20) and Remark 2.21(3) following its proof, we find an isotopy ψt of N (L0 ), fixing L0 , such that ψt∗ βt = λt α0 for some smooth family of smooth functions λt : N (L0 ) → R+ . (Since N (L0 ) is not a closed manifold, ψt is a priori only a local flow. But on L0 it is stationary and hence defined for all t. As in the proof of the Darboux theorem (Theorem 2.24) we conclude that ψt is defined for all t ∈ [0, 1] in a sufficiently small neighbourhood of L0 , so shrinking N (L0 ) once again, if necessary, will ensure that ψt is a global flow on N (L0 ).) We conclude that ψ = τ1 ◦ Φ˜ ◦ τ0−1 ◦ ψ1 is the desired contactomorphism. R EMARK 2.30. With a little more care one can actually achieve T ψ1 = id on T M0 |L0 , which implies in particular that T ψ|CSN(M0 ,L0 ) = Φ, cf. [105]. (Remember that there is a certain freedom in constructing an isotopy via the Moser trick if the condition Xt ∈ ξt is dropped.) The key point is the generalised Poincaré lemma, cf. [80, p. 361], which allows us to write a closed differential form γ given in a neighbourhood of the zero section of a bundle and vanishing along that zero section as an exact form γ = dη with η and its partial derivatives with respect to all coordinates (in any chart) vanishing along the zero section. This lemma is applied first to γ = d(β1 − β0 ), in order to find (with the symplectic Moser trick) a diffeomorphism σ of a neighbourhood of L0 ⊂ M0 with T σ = id on T M0 |L0 and such that dβ0 = d(σ ∗ β1 ). It is then applied once again to γ = β0 − σ ∗ β1 . (The proof of the symplectic neighbourhood theorem in [92] appears to be incomplete in this respect.) E XAMPLE 2.31. Let M0 = M1 = R3 with contact forms α0 = dz + x dy and α1 = dz + (x + y) dy and L0 = L1 = 0 the origin in R3 . Thus CSN(M0 , L0 ) = CSN(M1 , L1 ) = span{∂x , ∂y } ⊂ T0 R3 . We take Φ = idCSN . Set αt = dz + (x + ty) dy. The Moser trick with Xt ∈ ker αt yields Xt = −y ∂x , and hence ψt (x, y, z) = (x − ty, y, z). Then 1 −1 0 T ψ1 = 0 1 0 , 0 0 1 which does not restrict to Φ on CSN. However, a different solution for ψt∗ αt = α0 is ψt (x, y, z) = (x, y, z − ty 2 /2), found by integrating Xt = −y 2 ∂z /2 (a multiple of the Reeb vector field of αt ). Here we get 1 0 0 T ψ1 = 0 1 0 , 0 −y 1 hence T ψ1 |T0 R3 = id, so in particular T ψ1 |CSN = Φ.

Contact geometry

335

2.4.3. Contact submanifolds Let (M , ξ = ker α ) ⊂ (M, ξ = ker α) be a contact submanifold, that is, T M ∩ ξ |M = ξ . As before we write (ξ )⊥ ⊂ ξ |M for the symplectically orthogonal complement of ξ in ξ |M . Since M is a contact submanifold (so ξ is a symplectic sub-bundle of (ξ |M , dα)), we have T M ⊕ (ξ )⊥ = T M|M , i.e. we can identify (ξ )⊥ with the normal bundle N M . Moreover, dα induces a conformal symplectic structure on (ξ )⊥ , so we call (ξ )⊥ the conformal symplectic normal bundle of M in M and write CSN(M, M ) = (ξ )⊥ . T HEOREM 2.32. Let (Mi , ξi ), i = 0, 1, be contact manifolds with compact contact submanifolds (Mi , ξi ). Suppose there is an isomorphism of conformal symplectic normal bundles Φ : CSN(M0 , M0 ) → CSN(M1 , M1 ) that covers a contactomorphism φ : (M0 , ξ0 ) → (M1 , ξ1 ). Then φ extends to a contactomorphism ψ of suitable neighbourhoods N (Mi ) of Mi such that T ψ|CSN(M0 ,M0 ) and Φ are bundle homotopic (as conformal symplectic bundle isomorphisms). E XAMPLE 2.33. A particular instance of this theorem is the case of a transverse knot in a contact manifold (M, ξ ), i.e. an embedding S 1 → (M, ξ ) transverse to ξ . Since the sym2n plectic group Sp(2n) n of linear transformations of R preserving the standard symplectic structure ω0 = i=1 dxi ∧ dyi is connected, there is only one conformal symplectic R2n bundle over S 1 up to conformal equivalence. A model for the neighbourhood of a transverse knot is given by

n (xi dyi − yi dxi ) , S × R , ξ = ker dθ +

1

2n

i=1

where θ denotes the S 1 -coordinate; the theorem says that in suitable local coordinates the neighbourhood of any transverse knot looks like this model. P ROOF OF T HEOREM 2.32. As in the proof of Theorem 2.27 it is sufficient to find contact forms αi on Mi and a bundle map T M0 |M0 → T M1 |M1 , covering φ and inducing Φ, that pulls back α1 to α0 and dα1 to dα0 ; the proof then concludes as there with a stability argument. For this we need to make a judicious choice of αi . The essential choice is made separately on each Mi , so I suppress the subscript i for the time being. Choose a contact form α for ξ on M . Write R for the Reeb vector field of α . Given any contact form α for ξ on M we may first scale it such that α(R ) ≡ 1 along M . Then α|T M = α , and hence dα|T M = dα . We now want to scale α further such that its Reeb vector field R coincides with R along M . To this end it is sufficient to find a smooth function f : M → R+ with

336

H. Geiges

f |M ≡ 1 and iR d(f α) ≡ 0 on T M|M . This last equation becomes 0 = iR d(f α) = iR (df ∧ α + f dα) = −df + iR dα

on T M|M .

Since iR dα|T M = iR dα ≡ 0, such an f can be found. The choices of α0 and α1 cannot be made independently of each other; we may first choose α1 , say, and then define α0 = φ ∗ α1 . Then define α0 , α1 as described and scale Φ such that it is a symplectic bundle isomorphism of ⊥ (ξ0 ) , dα0 → (ξ1 )⊥ , dα1 . Then T φ ⊕ Φ : T M0 |M0 → T M1 |M1 is the desired bundle map that pulls back α1 to α0 and dα1 to dα0 .

R EMARK 2.34. The condition that Ri ≡ Ri along M is necessary for ensuring that (T φ ⊕ Φ)(R0 ) = R1 , which guarantees (with the other stated conditions) that (T φ ⊕ Φ)∗ (dα1 ) = dα0 . The condition dαi |T Mi = dαi and the described choice of Φ alone would only give (T φ ⊕ Φ)∗ (dα1 |ξ1 ) = dα0 |ξ0 . 2.4.4. Hypersurfaces Let S be an oriented hypersurface in a contact manifold (M, ξ = ker α) of dimension 2n + 1. In a neighbourhood of S in M, which we can identify with S × R (and S with S × {0}), the contact form α can be written as α = βr + ur dr, where βr , r ∈ R, is a smooth family of 1-forms on S and ur : S → R a smooth family of functions. The contact condition α ∧ (dα)n = 0 then becomes 0 = α ∧ (dα)n = (βr + ur dr) ∧ (dβr − β˙r ∧ dr + dur ∧ dr)n = (−nβr ∧ β˙r + nβr ∧ dur + ur dβr ) ∧ (dβr )n−1 ∧ dr,

(2.5)

where the dot denotes derivative with respect to r. The intersection T S ∩ (ξ |S ) determines a distribution (of non-constant rank) of subspaces of T S. If α is written as above, this distribution is given by the kernel of β0 , and hence, at a given p ∈ S, defines either the full tangent space Tp S (if β0,p = 0) or a 1-codimensional subspace both of Tp S and ξp (if β0,p = 0). In the former case, the symplectically orthogonal complement (Tp S ∩ ξp )⊥ (with respect to the conformal symplectic structure dα on ξp ) is {0}; in the latter case, (Tp S ∩ ξp )⊥ is a 1-dimensional subspace of ξp contained in Tp S ∩ ξp . From that it is intuitively clear what one should mean by a ‘singular 1-dimensional foliation’, and we make the following somewhat provisional definition.

Contact geometry

337

Fig. 2. The characteristic foliation on S 2 ⊂ (R3 , ξ2 ).

D EFINITION 2.35. The characteristic foliation Sξ of a hypersurface S in (M, ξ ) is the singular 1-dimensional foliation of S defined by (T S ∩ ξ |S )⊥ . E XAMPLE 2.36. If dim M = 3 and dim S = 2, then (Tp S ∩ ξp )⊥ = Tp S ∩ ξp at the points p ∈ S where Tp S ∩ ξp is 1-dimensional. Figure 2 shows the characteristic foliation of the unit 2-sphere in (R3 , ξ2 ), where ξ2 denotes the standard contact structure of Example 2.7: The only singular points are (0, 0, ±1); away from these points the characteristic foliation is spanned by (y − xz) ∂x − (x + yz) ∂y + x 2 + y 2 ∂z .

The following lemma helps to clarify the notion of singular 1-dimensional foliation. L EMMA 2.37. Let β0 be the 1-form induced on S by a contact form α defining ξ , and let Ω be a volume form on S. Then Sξ is defined by the vector field X satisfying iX Ω = β0 ∧ (dβ0 )n−1 . P ROOF. First of all, we observe that β0 ∧ (dβ0 )n−1 = 0 outside the zeros of β0 : Arguing by contradiction, assume β0,p = 0 and β0 ∧ (dβ0 )n−1 |p = 0 at some p ∈ S. Then (dβ0 )n |p = 0 by (2.5). On the codimension 1 subspace ker β0,p of Tp S the symplectic form dβ0,p has maximal rank n − 1. It follows that β0 ∧ (dβ0 )n−1 |p = 0 after all, a contradiction. Next we want to show that X ∈ ker β0 . We observe 0 = iX (iX Ω) = β0 (X)(dβ0 )n−1 − (n − 1)β0 ∧ iX dβ0 ∧ (dβ0 )n−2 . Taking the exterior product of this equation with β0 we get β0 (X)β0 ∧ (dβ0 )n−1 = 0. By our previous consideration this implies β0 (X) = 0. It remains to show that for β0,p = 0 we have dβ0 X(p), v = 0 for all v ∈ Tp S ∩ ξp .

(2.6)

338

H. Geiges

For n = 1 this is trivially satisfied, because in that case v is a multiple of X(p). I suppress the point p in the following calculation, where we assume n 2. From (2.6) and with β0 (X) = 0 we have β0 ∧ iX dβ0 ∧ (dβ0 )n−2 = 0.

(2.7)

Taking the interior product with v ∈ T S ∩ ξ yields −dβ0 (X, v)β0 ∧ (dβ0 )n−2 + (n − 2)β0 ∧ iX dβ0 ∧ iv dβ0 ∧ (dβ0 )n−3 = 0. (Thanks to the coefficient n − 2 the term (dβ0 )n−3 is not a problem for n = 2.) Taking the exterior product of that last equation with dβ0 , and using (2.7), we find dβ0 (X, v)β0 ∧ (dβ0 )n−1 = 0, and thus dβ0 (X, v) = 0.

R EMARK 2.38. (1) We can now give a more formal definition of ‘singular 1-dimensional foliation’ as an equivalence class of vector fields [X], where X is allowed to have zeros and [X] = [X ] if there is a nowhere zero function on all of S such that X = f X. Notice that the non-integrability of contact structures and the reasoning at the beginning of the proof of the lemma imply that the zero set of X does not contain any open subsets of S. (2) If the contact structure ξ is cooriented rather than just coorientable, so that α is well defined up to multiplication with a positive function, then this lemma allows to give an orientation to the characteristic foliation: Changing α to λα with λ : M → R+ will change β0 ∧ (dβ0 )n−1 by a factor λn . We now restrict attention to surfaces in contact 3-manifolds, where the notion of characteristic foliation has proved to be particularly useful. The following theorem is due to E. Giroux [52]. T HEOREM 2.39 (Giroux). Let Si be closed surfaces in contact 3-manifolds (Mi , ξi ), i = 0, 1 (with ξi coorientable), and φ : S0 → S1 a diffeomorphism with φ(S0,ξ0 ) = S1,ξ1 as oriented characteristic foliations. Then there is a contactomorphism ψ : N (S0 ) → N (S1 ) of suitable neighbourhoods N (Si ) of Si with ψ(S0 ) = S1 and such that ψ|S0 is isotopic to φ via an isotopy preserving the characteristic foliation. P ROOF. By passing to a double cover, if necessary, we may assume that the Si are orientable hypersurfaces. Let αi be contact forms defining ξi . Extend φ to a diffeomorphism (still denoted φ) of neighbourhoods of Si and consider the contact forms α0 and φ ∗ α1 on a neighbourhood of S0 , which we may identify with S0 × R. By rescaling α1 we may assume that α0 and φ ∗ α1 induce the same form β0 on S0 ≡ S0 × {0}, and hence also the same form dβ0 .

Contact geometry

339

Observe that the expression on the right-hand side of Eq. (2.5) is linear in β˙r and ur . This implies that convex linear combinations of solutions of (2.5) (for n = 1) with the same β0 (and dβ0 ) will again be solutions of (2.5) for sufficiently small r. This reasoning applies to αt := (1 − t)α0 + tφ ∗ α1 ,

t ∈ [0, 1].

(I hope the reader will forgive the slight abuse of notation, with α1 denoting both a form on M1 and its pull-back φ ∗ α1 to M0 .) As is to be expected, we now use the Moser trick to find an isotopy ψt with ψt∗ αt = λt α0 , just as in the proof of Gray stability (Theorem 2.20). In particular, we require as there that the vector field Xt that we want to integrate to the flow ψt lie in the kernel of αt . On T S0 we have α˙ t ≡ 0 (thanks to the assumption that α0 and φ ∗ α1 induce the same form β0 on S0 ). In particular, if v is a vector in S0,ξ0 , then by Eq. (2.1) we have dαt (Xt , v) = 0, which implies that Xt is a multiple of v, hence tangent to S0,ξ0 . This shows that the flow of Xt preserves S0 and its characteristic foliation. More formally, we have LXt αt = d αt (Xt ) + iXt dαt = iXt dαt , so with v as above we have LXt αt (v) = 0, which shows that LXt αt |T S0 is a multiple of α0 |T S0 = β0 . This implies that the (local) flow of Xt changes β0 by a conformal factor. Since S0 is closed, the local flow of Xt restricted to S0 integrates up to t = 1, and so the same is true3 in a neighbourhood of S0 . Then ψ = φ ◦ ψ1 will be the desired diffeomorphism N (S0 ) → N (S1 ). As observed previously in the proof of Darboux’s theorem for contact forms, the Moser trick allows more flexibility if we drop the condition αt (Xt ) = 0. We are now going to exploit this extra freedom to strengthen Giroux’s theorem slightly. This will be important later on when we want to extend isotopies of hypersurfaces. T HEOREM 2.40. Under the assumptions of the preceding theorem we can find ψ : N (S0 ) → N (S1 ) satisfying the stronger condition that ψ|S0 = φ. P ROOF. We want to show that the isotopy ψt of the preceding proof may be assumed to fix S0 pointwise. As there, we may assume α˙ t |T S0 ≡ 0. If the condition that Xt be tangent to ker αt is dropped, the condition Xt needs to satisfy so that its flow will pull back αt to λt α0 is α˙ t + d αt (Xt ) + iXt dαt = µt αt , where µt and λt are related by µt = theorem (Theorem 2.20).

−1 d dt (log λt ) ◦ ψt ,

(2.8) cf. the proof of the Gray stability

3 Cf. the proof (and the footnote therein) of Darboux’s theorem (Theorem 2.24).

340

H. Geiges

Write Xt = Ht Rt + Yt with Rt the Reeb vector field of αt and Yt ∈ ξt = ker αt . Then condition (2.8) translates into α˙ t + dHt + iYt dαt = µt αt .

(2.9)

For given Ht one determines µt from this equation by inserting the Reeb vector field Rt ; the equation then admits a unique solution Yt ∈ ker αt because of the non-degeneracy of dαt |ξt . Our aim now is to ensure that Ht ≡ 0 on S0 and Yt ≡ 0 along S0 . The latter we achieve by imposing the condition α˙ t + dHt = 0

along S0

(2.10)

(which entails with (2.9) that µt |S0 ≡ 0). The conditions Ht ≡ 0 on S0 and (2.10) can be simultaneously satisfied thanks to α˙ t |T S0 ≡ 0. We can therefore find a smooth family of smooth functions Ht satisfying these conditions, and then define Yt by (2.9). The flow of the vector field Xt = Ht Rt + Yt then defines an isotopy ψt that fixes S0 pointwise (and thus is defined for all t ∈ [0, 1] in a neighbourhood of S0 ). Then ψ = φ ◦ ψ1 will be the diffeomorphism we wanted to construct. 2.4.5. Applications Perhaps the most important consequence of the neighbourhood theorems proved above is that they allow us to perform differential topological constructions such as surgery or similar cutting and pasting operations in the presence of a contact structure, that is, these constructions can be carried out on a contact manifold in such a way that the resulting manifold again carries a contact structure. One such construction that I shall explain in detail in Section 3 is the surgery of contact 3-manifolds along transverse knots, which enables us to construct a contact structure on every closed, orientable 3-manifold. The concept of characteristic foliation on surfaces in contact 3-manifolds has proved seminal for the classification of contact structures on 3-manifolds, although it has recently been superseded by the notion of dividing curves. It is clear that Theorem 2.39 can be used to cut and paste contact manifolds along hypersurfaces with the same characteristic foliation. What actually makes this useful in dimension 3 is that there are ways to manipulate the characteristic foliation of a surface by isotoping that surface inside the contact 3-manifold. The most important result in that direction is the Elimination Lemma proved by Giroux [52]; an improved version is due to D. Fuchs, see [26]. This lemma says that under suitable assumptions it is possible to cancel singular points of the characteristic foliation in pairs by a C 0 -small isotopy of the surface (specifically: an elliptic and a hyperbolic point of the same sign—the sign being determined by the matching or non-matching of the orientation of the surface S and the contact structure ξ at the singular point of Sξ ). For instance, Eliashberg [24] has shown that if a contact 3-manifold (M, ξ ) contains an embedded disc D such that Dξ has a limit cycle, then one can actually find a so-called overtwisted disc: an embedded disc D with boundary ∂D tangent to ξ (but D transverse

Contact geometry

341

to ξ along ∂D, i.e. no singular points of Dξ on ∂D) and with Dξ having exactly one singular point (of elliptic type); see Section 3.6. Moreover, in the generic situation it is possible, given surfaces S ⊂ (M, ξ ) and S ⊂ (M , ξ ) with Sξ homeomorphic to Sξ , to perturb one of the surfaces so as to get diffeomorphic characteristic foliations. Chapter 8 of [1] contains a section on surfaces in contact 3-manifolds, and in particular a proof of the Elimination Lemma. Further introductory reading on the matter can be found in the lectures of J. Etnyre [35]; of the sources cited above I recommend [26] as a starting point. In [52] Giroux initiated the study of convex surfaces in contact 3-manifolds. These are surfaces S with an infinitesimal automorphism X of the contact structure ξ with X transverse to S. For such surfaces, it turns out, much less information than the characteristic foliation Sξ is needed to determine ξ in a neighbourhood of S, viz., only the so-called dividing set of Sξ . This notion lies at the centre of most of the recent advances in the classification of contact structures on 3-manifolds [55,71,72]. A brief introduction to convex surface theory can be found in [35].

2.5. Isotopy extension theorems In this section we show that the isotopy extension theorem of differential topology—an isotopy of a closed submanifold extends to an isotopy of the ambient manifold—remains valid for the various distinguished submanifolds of contact manifolds. The neighbourhood theorems proved above provide the key to the corresponding isotopy extension theorems. For simplicity, I assume throughout that the ambient contact manifold M is closed; all isotopy extension theorems remain valid if M has non-empty boundary ∂M, provided the isotopy stays away from the boundary. In that case, the isotopy of M found by extension keeps a neighbourhood of ∂M fixed. A further convention throughout is that our ambient isotopies ψt are understood to start at ψ0 = idM . 2.5.1. Isotropic submanifolds An embedding j : L → (M, ξ = ker α) is called isotropic if j (L) is an isotropic submanifold of (M, ξ ), i.e. everywhere tangent to the contact structure ξ . Equivalently, one needs to require j ∗ α ≡ 0. T HEOREM 2.41. Let jt : L → (M, ξ = ker α), t ∈ [0, 1], be an isotopy of isotropic embeddings of a closed manifold L in a contact manifold (M, ξ ). Then there is a compactly supported contact isotopy ψt : M → M with ψt (j0 (L)) = jt (L). P ROOF. Define a time-dependent vector field Xt along jt (L) by X t ◦ jt =

d jt . dt

To simplify notation later on, we assume that L is a submanifold of M and j0 the inclusion L ⊂ M. Our aim is to find a (smooth) family of compactly supported, smooth functions

342

H. Geiges

H˜ t : M → R whose Hamiltonian vector field X˜ t equals Xt along jt (L). Recall that X˜ t is defined in terms of H˜ t by α(X˜ t ) = H˜ t ,

iX˜ t dα = d H˜ t (Rα )α − d H˜ t ,

where, as usual, Rα denotes the Reeb vector field of α. Hence, we need α(Xt ) = H˜ t ,

iXt dα = d H˜ t (Rα )α − d H˜ t

along jt (L).

(2.11)

Write Xt = Ht Rα + Yt with Ht : jt (L) → R and Yt ∈ ker α. To satisfy (2.11) we need H˜ t = Ht

along jt (L).

(2.12)

This implies d H˜ t (v) = dHt (v)

for v ∈ T jt (L) .

Since jt is an isotopy of isotropic embeddings, we have T (jt (L)) ⊂ ker α. So a prerequisite for (2.11) is that dα(Xt , v) = −dHt (v)

for v ∈ T jt (L) .

(2.13)

We have dα(Xt , v) + dHt (v) = dα(Xt , v) + d α(Xt ) (v) = iv iXt dα + d(iXt α) = iv (LXt α), so Eq. (2.13) is equivalent to iv (LXt α) = 0

for v ∈ T jt (L) .

But this is indeed tautologically satisfied: The fact that jt is an isotopy of isotropic embedd (jt∗ α) = jt∗ (LXt α), as desired. dings can be written as jt∗ α ≡ 0; this in turn implies 0 = dt This means that we can define H˜ t by prescribing the value of H˜ t along jt (L) (with (2.12)) and the differential of H˜ t along jt (L) (with (2.11)), where we are free to impose d H˜ t (Rα ) = 0, for instance. The calculation we just performed shows that these two requirements are consistent with each other. Any function satisfying these requirements along jt (L) can be smoothed out to zero outside a tubular neighbourhood of jt (L), and the Hamiltonian flow of this H˜ t will be the desired contact isotopy extending jt . One small technical point is to ensure that the resulting family of functions H˜ t will be smooth in t. To achieve this, we proceed as follows. Set Mˆ = M × [0, 1] and Lˆ =

q∈L, t∈[0,1]

jt (q), t ,

Contact geometry

343

ˆ Let g be an auxiliary Riemannian metric on M with so that Lˆ is a submanifold of M. respect to which Rα is orthogonal to ker α. Identify the normal bundle N Lˆ of Lˆ in Mˆ with a sub-bundle of T Mˆ by requiring its fibre at a point (p, t) ∈ Lˆ to be the g-orthogonal subspace of Tp (jt (L)) in Tp M. Let τ : N Lˆ → Mˆ be a tubular map. Now define a smooth function Hˆ : N Lˆ → R as follows, where (p, t) always denotes a ˆ point of Lˆ ⊂ N L. ˆ • H (p, t) = α(Xt ), • d Hˆ (p,t) (Rα ) = 0, ˆ • d Hˆ (p,t) (v) = −dα(Xt , v) for v ∈ ker αp ⊂ Tp M ⊂ T(p,t) M, ˆ ˆ ˆ • H is linear on the fibres of N L → L. Let χ : Mˆ → [0, 1] be a smooth function with χ ≡ 0 outside a small neighbourhood N0 ⊂ ˆ of Lˆ and χ ≡ 1 in a smaller neighbourhood N1 ⊂ N0 of L. ˆ For (p, t) ∈ M, ˆ set τ (N L) ˆ −1 H˜ t (p) = χ(p, t)H (τ (p, t)) 0

ˆ for (p, t) ∈ τ (N L), ˆ for (p, t) ∈ / τ (N L).

This is smooth in p and t, and the Hamiltonian flow ψt of H˜ t (defined globally since H˜ t is compactly supported) is the desired contact isotopy. 2.5.2. Contact submanifolds An embedding j : (M , ξ ) → (M, ξ ) is called a contact embedding if j (M ), Tj (ξ ) is a contact submanifold of (M, ξ ), i.e. T j (M) ∩ ξ |j (M) = Tj (ξ ). If ξ = ker α, this can be reformulated as ker j ∗ α = ξ . T HEOREM 2.42. Let jt : (M , ξ ) → (M, ξ ), t ∈ [0, 1], be an isotopy of contact embeddings of the closed contact manifold (M , ξ ) in the contact manifold (M, ξ ). Then there is a compactly supported contact isotopy ψt : M → M with ψt (j0 (M )) = jt (M ). P ROOF. In the proof of this theorem we follow a slightly different strategy from the one in the isotropic case. Instead of directly finding an extension of the Hamiltonian Ht : jt (M ) → R, we first use the neighbourhood theorem for contact submanifolds to extend jt to an isotopy of contact embeddings of tubular neighbourhoods. Again we assume that M is a submanifold of M and j0 the inclusion M ⊂ M. As earlier, NM denotes the normal bundle of M in M. We also identify M with the zero section of N M , and we use the canonical identification T (NM )|M = T M ⊕ N M .

344

H. Geiges

By the usual isotopy extension theorem from differential topology we find an isotopy φt : N M → M with φt |M = jt . Choose contact forms α, α defining ξ and ξ , respectively. Define αt = φt∗ α. Then T M ∩ ker αt = ξ . Let R denote the Reeb vector field of α . Analogous to the proof of Theorem 2.32, we first find a smooth family of smooth functions gt : M → R+ such that gt αt |T M = α , and then a family ft : N M → R+ with ft |M ≡ 1 and dft = iR d(gt αt )

on T (N M )|M .

Then βt = ft gt αt is a family of contact forms on N M representing the contact structure ker(φt∗ α) and with the properties βt |T M = α , dβt |T M = dα , ker(dβt ) = R

along M .

The family (N M , dβt ) of symplectic vector bundles may be thought of as a symplectic vector bundle over M × [0, 1], which is necessarily isomorphic to a bundle pulled back from M × {0} (cf. [74, Corollary 3.4.4]). In other words, there is a smooth family of symplectic bundle isomorphisms Φt : (N M , dβ0 ) → (N M , dβt ). Then idT M ⊕ Φt : T (NM )|M → T (N M )|M is a bundle map that pulls back βt to β0 and dβt to dβ0 . By the now familiar stability argument we find a smooth family of embeddings ϕt : N (M ) → N M for some neighbourhood N (M ) of the zero section M in N M with ϕ0 = inclusion, ϕt |M = idM and ϕt∗ βt = λt β0 , where λt : N (M ) → R+ . This means that φt ◦ ϕt : N (M ) → M is a smooth family of contact embeddings of (N (M ), ker β0 ) in (M, ξ ). Define a time-dependent vector field Xt along φt ◦ ϕt (N (M )) by Xt ◦ φt ◦ ϕt =

d (φt ◦ ϕt ). dt

Contact geometry

345

This Xt is clearly an infinitesimal automorphism of ξ : By differentiating the equation ϕt∗ φt∗ α = µt φ0∗ α (where µt : N (M ) → R+ ) with respect to t we get ϕt∗ φt∗ (LXt α) = µ˙ t φ0∗ α =

µ˙ t ∗ ∗ ϕ φ α, µt t t

so LXt α is a multiple of α (since φt ◦ ϕt is a diffeomorphism onto its image). By the theory of contact Hamiltonians, Xt is the Hamiltonian vector field of a Hamiltonian function Hˆ t defined on φt ◦ ϕt (N (M )). Cut off this function with a bump function so as to obtain Ht : M → R with Ht ≡ Hˆ t near φt ◦ ϕt (M ) and Ht ≡ 0 outside a slightly larger neighbourhood of φt ◦ ϕt (M ). Then the Hamiltonian flow ψt of Ht satisfies our requirements. 2.5.3. Surfaces in 3-manifolds T HEOREM 2.43. Let jt : S → (M, ξ = ker α), t ∈ [0, 1], be an isotopy of embeddings of a closed surface S in a 3-dimensional contact manifold (M, ξ ). If all jt induce the same characteristic foliation on S, then there is a compactly supported isotopy ψt : M → M with ψt (j0 (S)) = jt (S). P ROOF. Extend jt to a smooth family of embeddings φt : S × R → M, and identify S with S × {0}. The assumptions say that all φt∗ α induce the same characteristic foliation on S. By the proof of Theorem 2.40 and in analogy with the proof of Theorem 2.42 we find a smooth family of embeddings ϕt : S × (−ε, ε) → S × R for some ε > 0 with ϕ0 = inclusion, ϕt |S×{0} = idS and ϕt∗ φt∗ α = λt φ0∗ α, where λt : S × (−ε, ε) → R+ . In other words, φt ◦ ϕt is a smooth family of contact embeddings of (S × (−ε, ε), ker φ0∗ α) in (M, ξ ). The proof now concludes exactly as the proof of Theorem 2.42.

2.6. Approximation theorems A further manifestation of the (local) flexibility of contact structures is the fact that a given submanifold can, under fairly weak (and usually obvious) topological conditions, be approximated (typically C 0 -closely) by a contact submanifold or an isotropic submanifold, respectively. The most general results in this direction are best phrased in M. Gromov’s language of h-principles. For a recent text on h-principles that puts particular emphasis on symplectic and contact geometry see [30]; a brief and perhaps more gentle introduction to h-principles can be found in [47]. In the present section I restrict attention to the 3-dimensional situation, where the relevant approximation theorems can be proved by elementary geometric ad hoc techniques.

346

H. Geiges

T HEOREM 2.44. Let γ : S 1 → (M, ξ ) be a knot, i.e. an embedding of S 1 , in a contact 3-manifold. Then γ can be C 0 -approximated by a Legendrian knot isotopic to γ . Alternatively, it can be C 0 -approximated by a positively as well as a negatively transverse knot. In order to prove this theorem, we first consider embeddings γ : (a, b) → (R3 , ξ ) of an open interval in R3 with its standard contact structure ξ = ker α, where α = dz + x dy. Write γ (t) = (x(t), y(t), z(t)). Then α(γ˙ ) = z˙ + x y, ˙ so the condition for a Legendrian curve reads z˙ + x y˙ ≡ 0; for a positively (respectively negatively) transverse curve, z˙ + x y˙ > 0 (respectively < 0). There are two ways to visualise such curves. The first is via its front projection γF (t) = y(t), z(t) , the second via its Lagrangian projection γL (t) = x(t), y(t) . 2.6.1. Legendrian knots If γ (t) = (x(t), y(t), z(t)) is a Legendrian curve in R3 , then y˙ = 0 implies z˙ = 0, so there the front projection has a singular point. Indeed, the curve t → (t, 0, 0) is an example of a Legendrian curve whose front projection is a single point. We call a Legendrian curve generic if y˙ = 0 only holds at isolated points (which we call cusp points), and there y¨ = 0. L EMMA 2.45. Let γ : (a, b) → (R3 , ξ ) be a Legendrian immersion. Then its front projection γF (t) = (y(t), z(t)) does not have any vertical tangencies. Away from the cusp points, γ is recovered from its front projection via x(t) = −

z˙ (t) dz =− , y(t) ˙ dy

i.e. x(t) is the negative slope of the front projection. The curve γ is embedded if and only if γF has only transverse self-intersections. By a C ∞ -small perturbation of γ we can obtain a generic Legendrian curve γ˜ isotopic to γ ; by a C 2 -small perturbation we may achieve that the front projection has only semicubical cusp singularities, i.e. around a cusp point at t = 0 the curve γ˜ looks like γ˜ (t) = t + a, λt 2 + b, −λ 2t 3 /3 + at 2 + c with λ = 0, see Figure 3. Any regular curve in the (y, z)-plane with semi-cubical cusps and no vertical tangencies can be lifted to a unique Legendrian curve in R3 .

Contact geometry

347

Fig. 3. The cusp of a front projection.

P ROOF. The Legendrian condition is z˙ + x y˙ = 0. Hence y˙ = 0 forces z˙ = 0, so γF cannot have any vertical tangencies. Away from the cusp points, the Legendrian condition tells us how to recover x as the dz negative slope of the front projection. (By continuity, the equation x = dy also makes sense at generic cusp points.) In particular, a self-intersecting front projection lifts to a non-intersecting curve if and only if the slopes at the intersection point are different, i.e. if and only if the intersection is transverse. That γ can be approximated in the C ∞ -topology by a generic immersion γ˜ follows from the usual transversality theorem (in its most simple form, viz., applied to the function y(t); the function x(t) may be left unchanged, and the new z(t) is then found by integrating the new −x y). ˙ At a cusp point of γ˜ we have y˙ = z˙ = 0. Since γ˜ is an immersion, this forces x˙ = 0, so γ˜ can be parametrised around a cusp point by the x-coordinate, i.e. we may choose the curve parameter t such that the cusp lies at t = 0 and x(t) = t + a. Since y(0) ¨ = 0 by the genericity condition, we can write y(t) = t 2 g(t) + y(0) with a smooth function g(t) satisfying g(0) = 0 (this is proved like the ‘Morse lemma’ in Appendix 2 of [77]). A C 0 approximation of g(t) by a function h(t) with h(t) ≡ g(0) for t near zero and h(t) ≡ g(t) for |t| greater than some small ε > 0 yields a C 2 -approximation of y(t) with the desired form around the cusp point. E XAMPLE 2.46. Figure 4 shows the front projection of a Legendrian trefoil knot. P ROOF OF T HEOREM 2.44 (Legendrian case). First of all, we consider a curve γ in standard R3 . In order to find a C 0 -close approximation of γ , we simply need to choose a C 0 -close approximation of its front projection γF by a regular curve without vertical tangencies and with isolated cusps (we call such a curve a front) in such a way, that the slope of the front at time t is close to −x(t) (see Figure 5). Then the Legendrian lift of this front is the desired C 0 -approximation of γ . If γ is defined on an interval (a, b) and is already Legendrian near its endpoints, then the approximation of γF may be assumed to coincide with γF near the endpoints, so that the Legendrian lift coincides with γ near the endpoints. Hence, given a knot in an arbitrary contact 3-manifold, we can cut it (by the Lebesgue lemma) into little pieces that lie in Darboux charts. There we can use the preceding recipe to find a Legendrian approximation. Since, as just observed, one can find such approximations on intervals with given boundary condition, this procedure yields a Legendrian approximation of the full knot.

348

H. Geiges

Fig. 4. Front projection of a Legendrian trefoil knot.

Fig. 5. Legendrian C 0 -approximation via front projection.

Locally (i.e. in R3 ) the described procedure does not introduce any self-intersections in the approximating curve, provided we approximate γF by a front with only transverse self-intersections. Since the original knot was embedded, the same will then be true for its Legendrian C 0 -approximation. The same result may be derived using the Lagrangian projection: L EMMA 2.47. Let γ : (a, b) → (R3 , ξ ) be a Legendrian immersion. Then its Lagrangian projection γL (t) = (x(t), y(t)) is also an immersed curve. The curve γ is recovered from γL via z(t1 ) = z(t0 ) −

t1

x dy. t0

Contact geometry

349

Fig. 6. Lagrangian projection of a Legendrian unknot.

A Legendrian immersion γ : S 1 → (R3 , ξ ) has a Lagrangian projection that encloses zero area. Moreover, γ is embedded if and only if every loop in γL (except, in the closed case, the full loop γL ) encloses a non-zero oriented area. Any immersed curve in the (x, y)-plane is the Lagrangian projection of a Legendrian curve in R3 , unique up to translation in the z-direction. P ROOF. The Legendrian condition z˙ + x y˙ implies that if y˙ = 0 then z˙ = 0, and hence, since γ is an immersion, x˙ = 0. So γL is an immersion. The formula for z follows by integrating the Legendrian condition. For a closed curve 5 γL in the (x, y)-plane, the integral γL x dy computes the oriented area enclosed by γL . From that all the other statements follow. E XAMPLE 2.48. Figure 6 shows the Lagrangian projection of a Legendrian unknot. A LTERNATIVE PROOF OF T HEOREM 2.44 (Legendrian case). Again we consider a curve γ in standard R3 defined on an interval. The generalisation to arbitrary contact manifolds and closed curves is achieved as in the proof using front projections. In order to find a C 0 -approximation of γ by a Legendrian curve, one only has to approximate its Lagrangian projection γL by an immersed curve whose ‘area integral’ z(t0 ) −

t

x dy t0

lies as close to the original z(t) as one wishes. This can be achieved by using small loops oriented positively or negatively (see Figure 7). If γL has self-intersections, this approximating curve can be chosen in such a way that along loops properly contained in that curve the area integral is non-zero, so that again we do not introduce any self-intersections in the Legendrian approximation of γ . 2.6.2. Transverse knots The quickest proof of the transverse case of Theorem 2.44 is via the Legendrian case. However, it is perfectly feasible to give a direct proof along the lines of the preceding discussion, i.e. using the front or the Lagrangian projection. Since this picture is useful elsewhere, I include a brief discussion, restricting attention to the front projection. Thus, let γ (t) = (x(t), y(t), z(t)) be a curve in R3 . The condition for γ to be positively transverse to the standard contact structure ξ = ker(dz + x dy) is that z˙ + x y˙ > 0. Hence, if y˙ = 0, if y˙ > 0, if y˙ < 0,

then z˙ > 0, then x > −˙z/y, ˙ then x < −˙z/y. ˙

350

H. Geiges

Fig. 7. Legendrian C 0 -approximation via Lagrangian projection.

Fig. 8. Impossible front projections of positively transverse curve.

The first statement says that there are no vertical tangencies oriented downwards in the front projection. The second statement says in particular that for y˙ > 0 and z˙ < 0 we have x > 0; the third, that for y˙ < 0 and z˙ < 0 we have x < 0. This implies that the situations shown in Figure 8 are not possible in the front projection of a positively transverse curve. I leave it to the reader to check that all other oriented crossings are possible in the front projection of a positively transverse curve, and that any curve in the (y, z)-plane without the forbidden crossing or downward vertical tangencies admits a lift to a positively transverse curve. E XAMPLE 2.49. Figure 9 shows the front projection of a positively transverse trefoil knot. P ROOF OF T HEOREM 2.44 (Transverse case). By the Legendrian case of this theorem, the given knot γ can be C 0 -approximated by a Legendrian knot γ1 . By Example 2.29, a neighbourhood of γ1 in (M, ξ ) looks like a solid torus S 1 × D 2 with contact structure cos θ dx − sin θ dy = 0, where γ1 ≡ S 1 × {0}. Then the curve γ2 (t) = (θ = t, x = δ sin t, y = δ cos t),

t ∈ [0, 2π],

Contact geometry

351

Fig. 9. Front projection of a positively transverse trefoil knot.

is a positively (respectively negatively) transverse knot if δ > 0 (respectively < 0). By choosing |δ| small we obtain as good a C 0 -approximation of γ1 and hence of γ as we wish.

3. Contact structures on 3-manifolds Here is the main theorem proved in this section: T HEOREM 3.1 (Lutz–Martinet). Every closed, orientable 3-manifold admits a contact structure in each homotopy class of tangent 2-plane fields. In Section 3.2 I present what is essentially J. Martinet’s [90] proof of the existence of a contact structure on every 3-manifold. This construction is based on a surgery description of 3-manifolds due to R. Lickorish and A. Wallace. For the key step, showing how to extend over a solid torus certain contact forms defined near the boundary of that torus (which then makes it possible to perform Dehn surgeries), we use an approach due to W. Thurston and H. Winkelnkemper; this allows us to simplify Martinet’s proof slightly. In Section 3.3 we show that every orientable 3-manifold is parallelisable and then build on this to classify (co-)oriented tangent 2-plane fields up to homotopy. In Section 3.4 we study the so-called Lutz twist, a topologically trivial Dehn surgery on a contact manifold (M, ξ ) which yields a contact structure ξ on M that is not homotopic (as 2-plane field) to ξ . We then complete the proof of the main theorem stated above. These results are contained in R. Lutz’s thesis [84]. Of Lutz’s published work, [83] only deals with the 3-sphere (and is only an announcement); [85] also deals with a more restricted problem. I learned the key steps of the construction from an exposition given in V. Ginzburg’s thesis [50]. I have added proofs of many topological details that do not seem to have appeared in a readily accessible source before. In Section 3.5 I indicate two further proofs for the existence of contact structures on every 3-manifold (and provide references to others). The one by Thurston and Winkelnkemper uses a description of 3-manifolds as open books due to J. Alexander; the crucial idea in their proof is the one we also use to simplify Martinet’s argument. Indeed, my discussion of the Lutz twist in the present section owes more to the paper by Thurston– Winkelnkemper than to any other reference. The second proof, by J. Gonzalo, is based

352

H. Geiges

on a branched cover description of 3-manifolds found by H. Hilden, J. Montesinos and T. Thickstun. This branched cover description also yields a very simple geometric proof that every orientable 3-manifold is parallelisable. In Section 3.6 we discuss the fundamental dichotomy between tight and overtwisted contact structures, introduced by Eliashberg, as well as the relation of these types of contact structures with the concept of symplectic fillability. The chapter concludes in Section 3.7 with a survey of classification results for contact structures on 3-manifolds. But first we discuss, in Section 3.1, an invariant of transverse knots in R3 with its standard contact structure. This invariant will be an ingredient in the proof of the Lutz–Martinet theorem, but is also of independent interest. I do not feel embarrassed to use quite a bit of machinery from algebraic and differential topology in this chapter. However, I believe that nothing that cannot be found in such standard texts as [14,77,95] is used without proof or an explicit reference. Throughout this third section, M denotes a closed, orientable 3-manifold.

3.1. An invariant of transverse knots Although the invariant in question can be defined for transverse knots in arbitrary contact manifolds (provided the knot is homologically trivial), for the sake of clarity I restrict attention to transverse knots in R3 with its standard contact structure ξ0 = ker(dz + x dy). This will be sufficient for the proof of the Lutz–Martinet theorem. In Section 3.6 I say a few words about the general situation and related invariants for Legendrian knots. Thus, let γ be a transverse knot in (R3 , ξ0 ). Push γ a little in the direction of ∂x —notice that this is a nowhere zero vector field contained in ξ0 , and in particular transverse to γ —to obtain a knot γ . An orientation of γ induces an orientation of γ . The orientation of R3 is given by dx ∧ dy ∧ dz. D EFINITION 3.2. The self-linking number l(γ ) of the transverse knot γ is the linking number of γ and γ . Notice that this definition is independent of the choice of orientation of γ (but it changes sign if the orientation of R3 is reversed). Furthermore, in place of ∂x we could have chosen any nowhere zero vector field X in ξ0 to define l(γ ): The difference between the selflinking number defined via ∂x and that defined via X is the degree of the map γ → S 1 given by associating to a point on γ the angle between ∂x and X. This degree is computed with the induced map Z ∼ = Z. But the map γ → S 1 factors through = H1 (γ ) → H1 (S 1 ) ∼ 3 R , so the induced homomorphism on homology is the zero homomorphism. Observe that l(γ ) is an invariant under isotopies of γ within the class of transverse knots. We now want to compute l(γ ) from the front projection of γ . Recall that the writhe of an oriented knot diagram is the signed number of self-crossings of the diagram, where the sign of the crossing is given in Figure 10. L EMMA 3.3. The self-linking number l(γ ) of a transverse knot is equal to the writhe w(γ ) of its front projection.

Contact geometry

353

Fig. 10. Signs of crossings in a knot diagram.

Fig. 11. Transverse knots with self-linking number ±3.

P ROOF. Let γ be the push-off of γ as described. Observe that each crossing of the front projection of γ contributes a crossing of γ underneath γ of the corresponding sign. Since the linking number of γ and γ is equal to the signed number of times that γ crosses underneath γ (cf. [98, p. 37]), we find that this linking number is equal to the signed number of self-crossings of γ , that is, l(γ ) = w(γ ). P ROPOSITION 3.4. Every self-linking number is realised by a transverse link in standard R3 . P ROOF. Figure 11 shows front projections of positively transverse knots (cf. Section 2.6.2) with self-linking number ±3. From that the construction principle for realising any odd integer should be clear. With a two component link any even integer can be realised. R EMARK 3.5. It is no accident that I do not give an example of a transverse knot with even self-linking number. By a theorem of Eliashberg [26, Proposition 2.3.1] that relates l(γ ) to the Euler characteristic of a Seifert surface S for γ and the signed number of singular points of the characteristic foliation Sξ , the self-linking number l(γ ) of a knot can only take odd values.

354

H. Geiges

3.2. Martinet’s construction According to Lickorish [81] and Wallace [103] M can be obtained from S 3 by Dehn surgery along a link of 1-spheres. This means that we have to remove a disjoint set of embedded solid tori S 1 × D 2 from S 3 and glue back solid tori with suitable identification by a diffeomorphism along the boundaries S 1 × S 1 . The effect of such a surgery (up to diffeomorphism of the resulting manifold) is completely determined by the induced map in homology H1 S 1 × ∂D 2 → H1 S 1 × ∂D 2 , Z ⊕ Z → Z ⊕ Z, n q which is given by a unimodular matrix m p ∈ GL(2, Z). Hence, denoting coordinates in 1 1 S × S by (θ, ϕ), we may always assume the identification maps to be of the form θ nq θ → . ϕ mp ϕ The curves µ and λ on ∂(S 1 × D 2 ) given respectively by θ = 0 and ϕ = 0 are called meridian and longitude. We keep the same notation µ, λ for the homology classes these curves represent. It turns out that the diffeomorphism type of the surgered manifold is completely determined by the class pµ + qλ, which is the class of the curve that becomes homotopically trivial in the surgered manifold (cf. [98, p. 28]). In fact, the Dehn surgery is completely determined by the surgery coefficient p/q, since the diffeomorphism of ∂(S 1 × D 2 ) given by (λ, µ) → (λ, −µ) extends to a diffeomorphism of the solid torus that we glue back. Similarly, the diffeomorphism given by (λ, µ) → (λ + kµ, µ) extends. By such a change of longitude in S 1 × D 2 ⊂ M, which amounts to choosing a different trivialisation of the n q normal bundle (i.e. framing) of S 1 × {0} ⊂ M, the gluing map is changed to m−kn p−kq . By a change of longitude in the solid torus that we glue back, the gluing map is changed n+kq q to m+kp p . Thus, a Dehn surgery is a so-called handle surgery (or ‘honest surgery’ or simply ‘surgery’) if and only if the surgery coefficient is an integer, that is, q = ±1. For in 0 1 n q exactly this case we may assume m p = 1 0 , and the surgery is given by cutting out S 1 × D 2 and gluing back S 1 × D 2 with the identity map ∂ D2 × S 1 → ∂ S 1 × D2 . The theorem of Lickorish and Wallace remains true if we only allow handle surgeries. However, this assumption does not entail any great simplification of the existence proof for contact structures, so we shall work with general Dehn surgeries. Our aim in this section is to use this topological description of 3-manifolds for a proof of the following theorem, first proved by Martinet [90]. The proof presented here is in spirit the one given by Martinet, but, as indicated in the introduction to this third section, amalgamated with ideas of Thurston and Winkelnkemper [101], whose proof of the same theorem we shall discuss later.

Contact geometry

355

T HEOREM 3.6 (Martinet). Every closed, orientable 3-manifold M admits a contact structure. In view of the theorem of Lickorish and Wallace and the fact that S 3 admits a contact structure, Martinet’s theorem is a direct consequence of the following result. T HEOREM 3.7. Let ξ0 be a contact structure on a 3-manifold M0 . Let M be the manifold obtained from M0 by a Dehn surgery along a knot K. Then M admits a contact structure ξ which coincides with ξ0 outside the neighbourhood of K where we perform surgery. P ROOF. By Theorem 2.44 we may assume that K is positively transverse to ξ0 . Then, by the contact neighbourhood theorem (Example 2.33), we can find a tubular neighbourhood of K diffeomorphic to S 1 ×D 2 (δ0 ), where K is identified with S 1 ×{0} and D 2 (δ0 ) denotes ¯ a disc of radius δ0 , such that the contact structure ξ0 is given as the kernel of d θ¯ + r¯ 2 d ϕ, ¯ polar coordinates on D 2 (δ0 ). with θ¯ denoting the S 1 -coordinate and (¯r , ϕ) n q Now perform a Dehn surgery along K defined by the unimodular matrix m p . This corresponds to cutting out S 1 × D 2 (δ1 ) ⊂ S 1 × D 2 (δ0 ) for some δ1 < δ0 and gluing it back in the way described above. Write (θ ; r, ϕ) for the coordinates on the copy of S 1 × D 2 (δ1 ) that we want to glue back. Then the contact form d θ¯ + r¯ 2 d ϕ¯ given on S 1 × D 2 (δ0 ) pulls back (along S 1 × ∂D 2 (δ1 )) to d(nθ + qϕ) + r 2 d(mθ + pϕ). This form is defined on all of S 1 ×(D 2 (δ1 )−{0}), and to complete the proof it only remains to find a contact form on S 1 × D 2 (δ1 ) that coincides with this form near S 1 × ∂D 2 (δ1 ). It is at this point that we use an argument inspired by the Thurston–Winkelnkemper proof (but which goes back to Lutz). n q 1 2 L EMMA 3.8. Given a unimodular matrix m p , there is a contact form on S × D (δ) that coincides with (n + mr 2 ) dθ + (q + pr 2 ) dϕ near r = δ and with ±dθ + r 2 dϕ near r = 0. P ROOF. We make the ansatz α = h1 (r) dθ + h2 (r) dϕ with smooth functions h1 (r), h2 (r). Then dα = h1 dr ∧ dθ + h2 dr ∧ dϕ and

h α ∧ dα =

1 h1

h2

dθ ∧ dr ∧ dϕ. h2

356

H. Geiges

Fig. 12. Dehn surgery.

So to satisfy the contact condition α ∧ dα = 0, all we have to do is to find a parametrised curve r → h1 (r), h2 (r) ,

0 r δ,

in the plane satisfying the following conditions: 1. h1 (r) = ±1 and h2 (r) = r 2 near r = 0, 2. h1 (r) = n + mr 2 and h2 (r) = q + pr 2 near r = δ, 3. (h1 (r), h2 (r)) is never parallel to (h1 (r), h2 (r)). Since np −mq = ±1, the vector (m, p) is not a multiple of (n, q). Figure 12 shows possible solution curves for the two cases np − mq = ±1. This completes the proof of the lemma and hence that of Theorem 3.7. R EMARK 3.9. On S 3 we have the standard contact forms α± = x dy − y dx ± (z dt − t dz) defining opposite orientations (cf. Remark 2.2). Performing the above surgery construction either on (S 3 , ker α+ ) or on (S 3 , ker α− ) we obtain both positive and negative contact structures on any given M. The same is true for the Lutz construction that we study in the next two sections. Hence: A closed oriented 3-manifold admits both a positive and a negative contact structure in each homotopy class of tangent 2-plane fields.

3.3. 2-plane fields on 3-manifolds First we need the following well-known fact. T HEOREM 3.10. Every closed, orientable 3-manifold M is parallelisable. R EMARK . The most geometric proof of this theorem can be given based on a structure theorem of Hilden, Montesinos and Thickstun. This will be discussed in Section 3.5.2. Another proof can be found in [76]. Here we present the classical algebraic proof.

Contact geometry

357

P ROOF. The main point is to show the vanishing of the second Stiefel–Whitney class w2 (M) = w2 (T M) ∈ H 2 (M; Z2 ). Recall the following facts, which can be found in [14]; for the interpretation of Stiefel–Whitney classes as obstruction classes see also [95]. There are Wu classes vi ∈ H i (M; Z2 ) defined by

Sqi (u), [M] = vi ∪ u, [M]

for all u ∈ H 3−i (M; Z2 ), where Sq denotes the Steenrod squaring operations. Since are v0 = 1 and v1 . The Sqi (u) = 0 for i > 3 − i, the only (potentially) non-zero Wu classes Wu classes and the Stiefel–Whitney classes are related by wq = j Sqq−j (vj ). Hence v1 = Sq0 (v1 ) = w1 , which equals zero since M is orientable. We conclude w2 = 0. Let V2 (R3 ) = SO(3)/ SO(1) = SO(3) be the Stiefel manifold of oriented, orthonormal 2-frames in R3 . This is connected, so there exists a section over the 1-skeleton of M of the 2-frame bundle V2 (T M) associated with T M (with a choice of Riemannian metric on M understood4 ). The obstruction to extending this section over the 2-skeleton is equal to w2 , which vanishes as we have just seen. The obstruction to extending the section over all of M lies in H 3 (M; π2 (V2 (R3 ))), which is the zero group because of π2 (SO(3)) = 0. We conclude that T M has a trivial 2-dimensional sub-bundle ε 2 . The complementary 1-dimensional bundle λ = T M/ε 2 is orientable and hence trivial since 0 = w1 (T M) = w1 (ε 2 ) + w1 (λ) = w1 (λ). Thus T M = ε 2 ⊕ λ is a trivial bundle. Fix an arbitrary Riemannian metric on M and a trivialisation of the unit tangent bundle ST M ∼ = M × S 2 . This sets up a one-to-one correspondence between the following sets, where all maps, homotopies, etc. are understood to be smooth. • Homotopy classes of unit vector fields X on M, • Homotopy classes of (co-)oriented 2-plane distributions ξ in T M, • Homotopy classes of maps f : M → S 2 . (I use the term ‘2-plane distribution’ synonymously with ‘2-dimensional sub-bundle of the tangent bundle’.) Let ξ1 , ξ2 be two arbitrary 2-plane distributions (always understood to be cooriented). By elementary obstruction theory there is an obstruction d 2 (ξ1 , ξ2 ) ∈ H 2 M; π2 S 2 ∼ = H 2 (M; Z) for ξ1 to be homotopic to ξ2 over the 2-skeleton of M and, if d 2 (ξ1 , ξ2 ) = 0 and after homotoping ξ1 to ξ2 over the 2-skeleton, an obstruction (which will depend, in general, on that first homotopy) d 3 (ξ1 , ξ2 ) ∈ H 3 M; π3 S 2 ∼ = H 3 (M; Z) ∼ =Z for ξ1 to be homotopic to ξ2 over all of M. (The identification of H 3 (M; Z) with Z is determined by the orientation of M.) However, rather than relying on general obstruction theory, we shall interpret d 2 and d 3 geometrically, which will later allow us to give a 4 This is not necessary, of course. One may also work with arbitrary 2-frames without reference to a metric. This does not affect the homotopical data.

358

H. Geiges

geometric proof that every homotopy class of 2-plane fields ξ on M contains a contact structure. The only fact that I want to quote here is that, by the Pontrjagin–Thom construction, homotopy classes of maps f : M → S 2 are in one-to-one correspondence with framed cobordism classes of framed (and oriented) links of 1-spheres in M. The general theory can be found in [14] and [77]; a beautiful and elementary account is given in [94]. For given f , the correspondence is defined by choosing a regular value p ∈ S 2 for f and a positively oriented basis b of Tp S 2 , and associating with it the oriented framed link (f −1 (p), f ∗ b), where f ∗ b is the pull-back of b under the fibrewise bijective map Tf : T (f −1 (p))⊥ → Tp S 2 . The orientation of f −1 (p) is the one which together with the frame f ∗ b gives the orientation of M. For a given framed link L the corresponding f is defined by projecting a (trivial) disc bundle neighbourhood L × D 2 of L in M onto the fibre D 2 ∼ = S 2 − p ∗ , where 0 is identified ∗ with p and p denotes the antipode of p, and sending M − (L × D 2 ) to p ∗ . Notice that the orientations of M and the components of L determine that of the fibre D 2 , and hence determine the map f . Before proceeding to define the obstruction classes d 2 and d 3 we make a short digression and discuss some topological background material which is fairly standard but not contained in our basic textbook references [14] and [77]. 3.3.1. Hopf’s Umkehrhomomorphismus If f : M m → N n is a continuous map between smooth manifolds, one can define a homomorphism ϕ : Hn−p (N ) → Hm−p (M) on homology classes represented by submanifolds as follows. Given a homology class [L]N ∈ Hn−p (N ) represented by a codimension p submanifold L, replace f by a smooth approximation transverse to L and define ϕ([L]N ) = [f −1 (L)]M . This is essentially Hopf’s Umkehrhomomorphismus [73], except that he worked with combinatorial manifolds of equal dimension and made no assumptions on the homology class. The following theorem, which in spirit is contained in [41], shows that ϕ is independent of choices (of submanifold L representing a class and smooth transverse approximation to f ) and actually a homomorphism of intersection rings. This statement is not as well known as it should be, and I only know of a proof in the literature for the special case where L is a point [60]. In [14] this map is called transfer map (more general transfer maps are discussed in [60]), but is only defined indirectly via Poincaré duality (though implicitly the statement of the following theorem is contained in [14], see, for instance, p. 377). T HEOREM 3.11. Let f : M m → N n be a smooth map between closed, oriented manifolds and Ln−p ⊂ N n a closed, oriented submanifold of codimension p such that f is transverse to L. Write u ∈ H p (N ) for the Poincaré dual of [L]N , that is, u ∩ [N ] = [L]N . Then [f −1 (L)]M = f ∗ u ∩ [M]. In other words: If u is Poincaré dual to [L]N , then f ∗ u ∈ H p (M) is Poincaré dual to [f −1 (L)]M . P ROOF. Since f is transverse to L, the differential Tf induces a fibrewise isomorphism between the normal bundles of f −1 (L) and L, and we find (closed) tubular neighbourhoods W → L and V = f −1 (W ) → f −1 (L) (considered as disc bundles) such that f : V → W is a fibrewise isomorphism. Write [V ]0 and [W ]0 for the orientation

Contact geometry

359

classes in Hm (V , V − f −1 (L)) and Hn (W, W − L), respectively. We can identify these homology groups with Hm (V , ∂V ) and Hn (W, ∂W ), respectively. Let τW ∈ H p (W, ∂W ) and τV ∈ H p (V , ∂V ) be the Thom classes of these disc bundles defined by τW ∩ [W ]0 = [L]N , τV ∩ [V ]0 = f −1 (L) M . Notice that f ∗ τW = τV since f : W → V is fibrewise isomorphic and the Thom class of an oriented disc bundle is the unique class whose restriction to each fibre is a positive generator of H p (D p , ∂D p ). Writing i : M → (M, M − f −1 (L)) and j : N → (N, N − L) for the inclusion maps we have −1 f (L) M = τV ∩ [V ]0 = f ∗ τW ∩ [V ]0 = f ∗ τW ∩ i∗ [M], where we identify Hm (M, M − f −1 (L)) with Hm (V , V − f −1 (L)) under the excision isomorphism. Then we have further −1 f (L) M = i ∗ f ∗ τW ∩ [M] = f ∗ j ∗ τW ∩ [M]. So it remains to identify j ∗ τW as the Poincaré dual u of [L]N . Indeed, j ∗ τW ∩ [N] = τW ∩ j∗ [N] = τW ∩ [W ]0 = [L]N , where we have used the excision isomorphism between the groups Hn (W, W − L) and Hn (N, N − L). 3.3.2. Representing homology classes by submanifolds We now want to relate elements in H1 (M; Z) to cobordism classes of links in M. T HEOREM 3.12. Let M be a closed, oriented 3-manifold. Any c ∈ H1 (M; Z) is represented by an embedded, oriented link (of 1-spheres) Lc in M. Two links L0 , L1 represent the same class [L0 ] = [L1 ] if and only if they are cobordant in M, that is, there is an embedded, oriented surface S in M × [0, 1] with ∂S = L1 0 (−L0 ) ⊂ M × {1} 0 M × {0}, where 0 denotes disjoint union. P ROOF. Given c ∈ H1 (M; Z), set u = P D(c) ∈ H 2 (M; Z), where P D denotes the Poincaré duality map from homology to cohomology. There is a well-known isomorphism H 2 (M; Z) ∼ = M, K(Z, 2) = [M, CP ∞ ], where brackets denote homotopy classes of maps (cf. [14, VII.12]). So u corresponds to a homotopy class of maps [f ] : M → CP ∞ such that f ∗ u0 = u, where u0 is the positive

360

H. Geiges

generator of H 2 (CP ∞ ) (that is, the one that pulls back to the Poincaré dual of [CP k−1 ]CP k under the natural inclusion CP k ⊂ CP ∞ ). Since dim M = 3, any map f : M → CP ∞ is homotopic to a smooth map f1 : M → CP 1 . Let p be a regular value of f1 . Then P D(c) = u = f1∗ u0 = f1∗ P D[p] = P D f1−1 (p) by our discussion above, and hence c = [f1−1 (p)]. So Lc = f1−1 (p) is the desired link. It is important to note that in spite of what we have just said it is not true that [M, CP ∞ ] = [M, CP 1 ], since a map F : M × [0, 1] → CP ∞ with F (M × {0, 1}) ⊂ CP 1 is not, in general, homotopic rel(M × {0, 1}) to a map into CP 1 . However, we do have [M, CP ∞ ] = [M, CP 2 ]. If two links L0 , L1 are cobordant in M, then clearly [L0 ] = [L1 ] ∈ H1 M × [0, 1]; Z ∼ = H1 (M; Z). For the converse, suppose we are given two links L0 , L1 ⊂ M with [L0 ] = [L1 ]. Choose arbitrary framings for these links and use this, as described at the beginning of this section, to define smooth maps f0 , f1 : M → S 2 with common regular value p ∈ S 2 such that fi−1 (p) = Li , i = 0, 1. Now identify S 2 with the standardly embedded CP 1 ⊂ CP 2 . Let P ⊂ CP 2 be a second copy of CP 1 , embedded in such a way that [P ]CP 2 = [CP 1 ]CP 2 and P intersects CP 1 transversely in p only. This is possible since CP 1 ⊂ CP 2 has selfintersection one. Then the maps f0 , f1 , regarded as maps into CP 2 , are transverse to P and we have fi−1 (P ) = Li , i = 0, 1. Hence fi∗ u0 = fi∗ P D[P ]CP 2 = P D fi−1 (P ) M = P D[Li ]M is the same for i = 0 or 1, and from the identification ∼ = M, CP 2 → H 2 (M, Z), [f ] → f ∗ u0 we conclude that f0 and f1 are homotopic as maps into CP 2 . Let F : M × [0, 1] → CP 2 be a homotopy between f0 and f1 , which we may assume to be constant near 0 and 1. This F can be smoothly approximated by a map F : M ×[0, 1] → CP 2 which is transverse to P and coincides with F near M × 0 and M × 1 (since there the transversality condition was already satisfied). In particular, F is still a homotopy between f0 and f1 , and S = (F )−1 (P ) is a surface with the desired property ∂S = L1 0 (−L0 ). Notice that in the course of this proof we have observed that cobordism classes of links in M (equivalently, classes in H1 (M; Z)) correspond to homotopy classes of maps M → CP 2 , whereas framed cobordism classes of framed links correspond to homotopy classes of maps M → CP 1 . By forming the connected sum of the components of a link representing a certain class in H1 (M; Z), one may actually always represent such a class by a link with only one component, that is, a knot.

Contact geometry

361

3.3.3. Framed cobordisms We have seen that if L1 , L2 ⊂ M are links with [L1 ] = [L2 ] ∈ H1 (M; Z), then L1 and L2 are cobordant in M. In general, however, a given framing on L1 and L2 does not extend over the cobordism. The following observation will be useful later on. Write (S 1 , n) for a contractible loop in M with framing n ∈ Z (by which we mean that S 1 and a second copy of S 1 obtained by pushing it away in the direction of one of the vectors in the frame have linking number n). When writing L = L 0 (S 1 , n) it is understood that (S 1 , n) is not linked with any component of L . Suppose we have two framed links L0 , L1 ⊂ M with [L0 ] = [L1 ]. Let S ⊂ M × [0, 1] be an embedded surface with ∂S = L1 0 (−L0 ) ⊂ M × {1} 0 M × {0}. With D 2 a small disc embedded in S, the framing of L1 and L2 in M extends to a framing of S − D 2 in M × [0, 1] (since S − D 2 deformation retracts to a 1-dimensional complex containing L0 and L1 , and over such a complex an orientable 2-plane bundle is trivial). Now we embed a cylinder S 1 × [0, 1] in M × [0, 1] such that S 1 × [0, 1] ∩ M × {0} = ∅, S 1 × [0, 1] ∩ M × {1} = S 1 × {1}, and S 1 × [0, 1] ∩ S − D 2 = S 1 × {0} = ∂D 2 , see Figure 13. This shows that L0 is framed cobordant in M to L1 0 (S 1 , n) for suitable n ∈ Z.

Fig. 13. The framed cobordism between L0 and L1 0 (S 1 , n).

362

H. Geiges

3.3.4. Definition of the obstruction classes We are now in a position to define the obstruction classes d 2 and d 3 . With a choice of Riemannian metric on M and a trivialisation of ST M understood, a 2-plane distribution ξ on M defines a map fξ : M → S 2 and hence an oriented framed link Lξ as described above. Let [Lξ ] ∈ H1 (M; Z) be the homology class represented by Lξ . This only depends on the homotopy class of ξ , since under homotopies of ξ or choice of different regular values of fξ the cobordism class of Lξ remains invariant. We define d 2 (ξ1 , ξ2 ) = P D[Lξ1 ] − P D[Lξ2 ]. With this definition d 2 is clearly additive, that is, d 2 (ξ1 , ξ2 ) + d 2 (ξ2 , ξ3 ) = d 2 (ξ1 , ξ3 ). The following lemma shows that d 2 is indeed the desired obstruction class. L EMMA 3.13. The 2-plane distributions ξ1 and ξ2 are homotopic over the 2-skeleton M (2) of M if and only if d 2 (ξ1 , ξ2 ) = 0. P ROOF. Suppose d 2 (ξ1 , ξ2 ) = 0, that is, [Lξ1 ] = [Lξ2 ]. By Theorem 3.12 we find a surface S in M × [0, 1] with ∂S = Lξ2 0 (−Lξ1 ) ⊂ M × {1} 0 M × {0}. From the discussion on framed cobordism above we know that for suitable n ∈ Z we find a framed surface S in M × [0, 1] such that ∂S = Lξ2 0 S 1 , n 0 (−Lξ1 ) ⊂ M × {1} 0 M × {0} as framed manifolds. Hence ξ1 is homotopic to a 2-plane distribution ξ1 such that Lξ1 and Lξ2 differ only by one contractible framed loop (not linked with any other component). Then the corresponding maps f1 , f2 differ only in a neighbourhood of this loop, which is contained in a 3-ball, so f1 and f2 (and hence ξ1 and ξ2 ) agree over the 2-skeleton. Conversely, if ξ1 and ξ2 are homotopic over M (2) , we may assume ξ1 = ξ2 on M − D 3 for some embedded 3-disc D 3 ⊂ M without changing [Lξ1 ] and [Lξ2 ]. Now [Lξ1 ] = [Lξ2 ] follows from H1 (D 3 , S 2 ) = 0. R EMARK 3.14. By [99, §37] the obstruction to homotopy between ξ and ξ0 (corresponding to the constant map fξ0 : M → S 2 ) over the 2-skeleton of M is given by fξ∗ u0 , where u0 is the positive generator of H 2 (S 2 ; Z). So u0 = P D[p] for any p ∈ S 2 , and taking p to

Contact geometry

363

be a regular value of fξ we have fξ∗ u0 = fξ∗ P D[p] = P D fξ−1 (p) = P D[Lξ ] = d 2 (ξ, ξ0 ). This gives an alternative way to see that our geometric definition of d 2 does indeed coincide with the class defined by classical obstruction theory. Now suppose d 2 (ξ1 , ξ2 ) = 0. We may then assume that ξ1 = ξ2 on M − int(D 3 ), and we define d 3 (ξ1 , ξ2 ) to be the Hopf invariant H (f ) of the map f : S 3 → S 2 defined as f1 ◦ π+ on the upper hemisphere and f2 ◦ π− on the lower hemisphere, where π+ , π− are the orthogonal projections of the upper respectively lower hemisphere onto the equatorial disc, which we identify with D 3 ⊂ M. Here, given an orientation of M, we orient S 3 in such a way that π+ is orientation-preserving and π− orientation-reversing; the orientation of S 2 is inessential for the computation of H (f ). Recall that H (f ) is defined as the linking number of the preimages of two distinct regular values of a smooth map homotopic to f . Since the Hopf invariant classifies homotopy classes of maps S 3 → S 2 (it is in fact an isomorphism π3 (S 2 ) → Z), this is a suitable definition for the obstruction class d 3 . Moreover, the homomorphism property of H (f ) and the way addition in π3 (S 2 ) is defined entail the additivity of d 3 analogous to that of d 2 . For M = S 3 there is another way to interpret d 3 . Oriented 2-plane distributions on M correspond to sections of the bundle associated to T M with fibre SO(3)/U(1), hence to maps M → SO(3)/U(1) ∼ = S 2 since T M is trivial. Similarly, almost complex structures on 4 D correspond to maps D 4 → SO(4)/U(2) ∼ = SO(3)/U(1) (cf. [61] for this isomorphism). A cooriented 2-plane distribution on M can be interpreted as a triple (X, ξ, J ), where X is a vector field transverse to ξ defining the coorientation, and J a complex structure on ξ defining the orientation. Such a triple is called an almost contact structure. (This notion generalises to higher (odd) dimensions, and by Remark 2.3 every cooriented contact structure induces an almost contact structure, and in fact a unique one up to homotopy as follows from the result cited in that remark.) Given an almost contact structure (X, ξ, J ) on S 3 , one defines an almost complex structure J˜ on T D 4 |S 3 by J˜|ξ = J and J˜N = X, where N denotes the outward normal vector field. So there is a canonical way to identify homotopy classes of almost contact structures on S 3 with elements of π3 (SO(3)/U(1)) ∼ = Z such that the value zero corresponds to the almost contact structure that extends as almost complex structure over D 4 .

3.4. Let’s twist again Consider a 3-manifold M with cooriented contact structure ξ and an oriented 1-sphere K ⊂ M embedded transversely to ξ such that the positive orientation of K coincides with the positive coorientation of ξ . Then in suitable local coordinates we can identify K with S 1 × {0} ⊂ S 1 × D 2 such that ξ = ker(dθ + r 2 dϕ) and ∂θ corresponds to the positive orientation of K (see Example 2.33). Strictly speaking, if, as we shall always assume, S 1 is parametrised by 0 θ 2π , then this formula for ξ holds on S 1 × D 2 (δ) for some,

364

H. Geiges

Fig. 14. Lutz twist.

possibly small, δ > 0. However, to simplify notation we usually work with S 1 × D 2 as local model. We say that ξ is obtained from ξ by a Lutz twist along K and write ξ = ξ K if on 1 S × D 2 the new contact structure ξ is defined by ξ = ker h1 (r) dθ + h2 (r) dϕ with (h1 (r), h2 (r)) as in Figure 14, and ξ coincides with ξ outside S 1 × D 2 . More precisely, (h1 (r), h2 (r)) is required to satisfy the conditions 1. h1 (r) = −1 and h2 (r) = −r 2 near r = 0, 2. h1 (r) = 1 and h2 (r) = r 2 near r = 1, 3. (h1 (r), h2 (r)) is never parallel to (h1 (r), h2 (r)). This is the same as applying the construction of Section 3.2 to the topologically trivial 0 Dehn surgery given by the matrix −1 0 −1 . It will be useful later on to understand more precisely the behaviour of the map fξ : M → S 2 . For the definition of this map we assume—this assumption will be justified below—that on S 1 × D 2 the map fξ was defined in terms of the standard metric dθ 2 + du2 + dv 2 (with u, v Cartesian coordinates on D 2 corresponding to the polar coordinates r, ϕ) and the trivialisation ∂θ , ∂u , ∂v of T (S 1 × D 2 ). Since ξ is spanned by ∂r and h2 (r) ∂θ − h1 (r) ∂ϕ (respectively ∂u , ∂v for r = 0), a vector positively orthogonal to ξ is given by h1 (r) ∂θ + h2 (r) ∂ϕ ,

Contact geometry

365

which makes sense even for r = 0. Observe that the ratio h1 (r)/ h2 (r) (for h2 (r) = 0) is a strictly monotone decreasing function since by the third condition above we have (h1 / h2 ) = (h1 h2 − h1 h2 )/ h22 < 0. This implies that any value on S 2 other than (1, 0, 0) (corresponding to ∂θ ) is regular for the map fξ ; the value (1, 0, 0) is attained along the torus {r = r0 }, with r0 > 0 determined by h2 (r0 ) = 0, and hence not regular. If S 1 × D 2 is endowed with the orientation defined by the volume form dθ ∧ r dr ∧ dϕ = dθ ∧ du ∧ dv (so that ξ and ξ are positive contact structures) and S 2 ⊂ R3 is given its ‘usual’ orientation defined by the volume form x dy ∧ dz + y dz ∧ dx + z dx ∧ dy, then 1 fξ−1 (−1, 0, 0) = S × {0}

with orientation given by −∂θ , since fξ maps the slices {θ } × D 2 (r0 ) orientationreversingly onto S 2 . More generally, for any p ∈ S 2 − {(1, 0, 0)} the preimage fξ−1 (p) (inside the domain 1 2 {(θ, r, ϕ): h2 (r) < 0} = {r < r0 }) is a circle S × {u}, u ∈ D , with orientation given by −∂θ . We are now ready to show how to construct a contact structure on M in any given homotopy class of 2-plane distributions by starting with an arbitrary contact structure and performing suitable Lutz twists. First we deal with homotopy over the 2-skeleton. One way to proceed would be to prove directly, with notation as above, that d 2 (ξ K , ξ ) = −P D[K]. However, it is somewhat easier to compute d 2 (ξ K , ξ ) in the case where ξ is a trivial 2-plane bundle and the trivialisation of ST M is adapted to ξ . Since I would anyway like to present an alternative argument for computing the effect of a Lutz twist on the Euler class of the contact structure, and thus relate d 2 (ξ1 , ξ2 ) with the Euler classes of ξ1 and ξ2 , it seems opportune to do this first and use it to show the existence of a contact structure with Euler class zero. In the next section we shall actually discuss a direct geometric proof, due to Gonzalo, of the existence of a contact structure with Euler class zero. Recall that the Euler class e(ξ ) ∈ H 2 (B; Z) of a 2-plane bundle over a complex B (of arbitrary dimension) is the obstruction to finding a nowhere zero section of ξ over the 2skeleton of B. Since πi (S 1 ) = 0 for i 2, all higher obstruction groups H i+1 (B; πi (S 1 )) are trivial, so a 2-dimensional orientable bundle ξ is trivial if and only if e(ξ ) = 0, no matter what the dimension of B. Now let ξ be an arbitrary cooriented 2-plane distribution on an oriented 3-manifold M. Then T M ∼ = ξ ⊕ ε 1 , where ε 1 denotes a trivial line bundle. Hence w2 (ξ ) = w2 (ξ ⊕ ε 1 ) = w2 (T M) = 0, and since w2 (ξ ) is the mod 2 reduction of e(ξ ) we infer that e(ξ ) has to be even. P ROPOSITION 3.15. For any even element e ∈ H 2 (M; Z) there is a contact structure ξ on M with e(ξ ) = e. P ROOF. Start with an arbitrary contact structure ξ0 on M with e(ξ0 ) = e0 (which we know to be even). Given any even e1 ∈ H 2 (M; Z), represent the Poincaré dual of (e0 − e1 )/2 by

366

H. Geiges

Fig. 15. Effect of Lutz twist on Euler class.

a collection of embedded oriented circles positively transverse to ξ0 . (Here by (e0 − e1 )/2 I mean some class whose double equals e0 − e1 ; in the presence of 2-torsion there is of course a choice of such classes.) Choose a section of ξ0 transverse to the zero section of ξ0 , that is, a vector field in ξ0 with generic zeros. We may assume that there are no zeros on the curves representing P D −1 (e0 − e1 )/2. Now perform a Lutz twist as described above along these curves and call ξ1 the resulting contact structure. It is easy to see that in the local model for the Lutz twist a constant vector field tangent to ξ0 along ∂(S 1 × D 2 (r0 )) extends to a vector field tangent to ξ1 over S 1 × D 2 (r0 ) with zeros of index +2 along S 1 × {0} (Figure 15). So the vector field in ξ0 extends to a vector field in ξ1 with new zeros of index +2 along the curves representing P D −1 (e1 − e0 )/2 (notice that a Lutz twist along a positively transverse knot K turns K into a negatively transverse knot). Since the self-intersection class of M in the total space of a vector bundle is Poincaré dual to the Euler class of that bundle, this proves e(ξ1 ) = e(ξ0 ) + e1 − e0 = e1 . We now fix a contact structure ξ0 on M with e(ξ0 ) = 0 and give M the orientation induced by ξ0 (i.e. the one for which ξ0 is a positive contact structure). Moreover, we fix a Riemannian metric on M and define X0 as the vector field positively orthonormal to ξ0 . Since ξ0 is a trivial plane bundle, X0 extends to an orthonormal frame X0 , X1 , X2 , hence a trivialisation of ST M, with X1 , X2 tangent to ξ0 and defining the orientation of ξ0 . With these choices, ξ0 corresponds to the constant map fξ0 : M → (1, 0, 0) ∈ S 2 . P ROPOSITION 3.16. Let K ⊂ M be an embedded, oriented circle positively transverse to ξ0 . Then d 2 (ξ0K , ξ0 ) = −P D[K]. P ROOF. Identify a tubular neighbourhood of K ⊂ M with S 1 × D 2 with framing defined by X1 , and ξ0 given in this neighbourhood as the kernel of dθ + r 2 dϕ = dθ + u dv − v du. We may then change the trivialisation X0 , X1 , X2 by a homotopy, fixed outside S 1 × D 2 , such that X0 = ∂θ , X1 = ∂u and X2 = ∂v near K; this does not change the homotopical data of 2-plane distributions computed via the Pontrjagin–Thom construction. Then fξ0 is no longer constant, but its image still does not contain the point (−1, 0, 0).

Contact geometry

367

Now perform a Lutz twist along K × {0}. Our discussion at the beginning of this section shows that (−1, 0, 0) is a regular value of the map fξ : M → S 2 associated with ξ = ξ0K and fξ−1 (−1, 0, 0) = −K. Hence, by definition of the obstruction class d 2 we have d 2 (ξ0K , ξ0 ) = −P D[K]. P ROOF OF T HEOREM 3.1. Let η be a 2-plane distribution on M and ξ0 the contact structure on M with e(ξ0 ) = 0 that we fixed earlier on. According to our discussion in Section 3.3.2 and Theorem 2.44, we can find an oriented knot K positively transverse to ξ0 with −P D[K] = d 2 (η, ξ0 ). Then d 2 (η, ξ0 ) = d 2 (ξ0K , ξ0 ) by the preceding proposition, and therefore d 2 (ξ0K , η) = 0. We may then assume that η = ξ0K on M − D 3 , where we choose D 3 so small that ξ0K is in Darboux normal form there (and identical with ξ0 ). By Proposition 3.4 we can find a link K in D 3 transverse to ξ0K with self-linking number l(K ) equal to d 3 (η, ξ0K ). Now perform a Lutz twist of ξ0K along each component of K and let ξ be the resulting contact structure. Since this does not change ξ0K over the 2-skeleton of M, we still have d 2 (ξ, η) = 0. Observe that fξ K |D 3 does not contain the point (−1, 0, 0) ∈ S 2 , and—since fξ K (D 3 ) is 0

0

compact—there is a whole neighbourhood U ⊂ S 2 of (−1, 0, 0) not contained in fξ K (D 3 ). 0

Let f : S 3 → S 2 be the map used to compute d 3 (ξ, ξ0K ), that is, f coincides on the upper hemisphere with fξ |D 3 and on the lower hemisphere with fξ K |D 3 . By the discussion in 0

Section 3.3, the preimage f −1 (u) of any u ∈ U − {(−1, 0, 0)} will be a push-off of −K determined by the trivialisation of ξ0K |D 3 = ξ0 |D 3 . So the linking number of f −1 (u) with f −1 (−1, 0, 0), which is by definition the Hopf invariant H (f ) = d 3 (ξ, ξ0K ), will be equal to l(K ). By our choice of K and the additivity of d 3 this implies d 3 (ξ, η) = 0. So ξ is a contact structure that is homotopic to η as a 2-plane distribution.

3.5. Other existence proofs Here I briefly summarise the other known existence proofs for contact structures on 3-manifolds, mostly by pointing to the relevant literature. In spirit, most of these proofs are similar to the one by Lutz–Martinet: start with a structure theorem for 3-manifolds and show that the topological construction can be performed compatibly with a contact structure. 3.5.1. Open books According to a theorem of Alexander [4], cf. [97], every closed, orientable 3-manifold M admits an open book decomposition. This means that there is a link L ⊂ M, called the binding, and a fibration f : M − L → S 1 , whose fibres are called the pages, see Figure 16. It may be assumed that L has a tubular neighbourhood L × D 2 such that f restricted to L × (D 2 − {0}) is given by f (θ, r, ϕ) = ϕ, where θ is the coordinate along L and (r, ϕ) are polar coordinates on D 2 . At the cost of raising the genus of the pages, one may decrease the number of components of L, and in particular one may always assume L to be a knot. Another way

368

H. Geiges

Fig. 16. An open book near the binding.

to think of such an open book is as follows. Start with a surface Σ with boundary ∂Σ = K ∼ = S 1 and a self-diffeomorphism h of Σ with h = id near K. Form the mapping torus Th = Σ × [0, 2π]/∼, where ‘∼’ denotes the identification (p, 2π) ∼ (h(p), 0). Define a 3-manifold M by M = Th ∪K×S 1 K × D 2 . This M carries by construction the structure of an open book with binding K and pages diffeomorphic to Σ . Here is a slight variation on a simple argument of Thurston and Winkelnkemper [101] for producing a contact structure on such an open book (and hence on any closed, orientable 3-manifold): Start with a 1-form β0 on Σ with β0 = et dθ near ∂Σ = K, where θ denotes the coordinate along K and t is a collar parameter with K = {t = 0} and t < 0 in the interior of Σ. Then dβ0 integrates to 2π over Σ by Stokes’s theorem. Given any area form ω on Σ (with total area equal to 2π ) satisfying ω = et dt ∧ dθ near K, the 2-form ω − dβ0 is, by de Rham’s theorem, an exact 1-form, say dβ1 , where we may assume β1 ≡ 0 near K. Set β = β0 + β1 . Then dβ = ω is an area form (of total area 2π ) on Σ and β = et dθ near K. The set of 1-forms satisfying these two properties is a convex set, so we can find a 1-form (still denoted β) on Th which has these properties when restricted to the fibre over any ϕ ∈ S 1 = [0, 2π]/0∼2π . We may (and shall) require that β = et dθ near ∂Th . Now a contact form α on Th is found by setting α = β + C dϕ for a sufficiently large constant C ∈ R+ , so that in α ∧ dα = (β + C dϕ) ∧ dβ

Contact geometry

369

the non-zero term dϕ ∧ dβ = dϕ ∧ ω dominates. This contact form can be extended to all of M by making the ansatz α = h1 (r) dθ + h2 (r) dϕ on K × D 2 , as described in our discussion of the Lutz twist. The boundary conditions in the present situation are, say, 1. h1 (r) = 2 and h2 (r) = r 2 near r = 0, 2. h1 (r) = e1−r and h2 (r) = C near r = 1. Observe that subject to these boundary conditions a curve (h1 (r), h2 (r)) can be found that does not pass the h2 -axis (i.e. with h1 (r) never being equal to zero). In the 3-dimensional setting this is not essential (and the Thurston–Winkelnkemper ansatz lacked that feature), but it is crucial when one tries to generalise this construction to higher dimensions. This has recently been worked out by Giroux and J.-P. Mohsen [57]. This, however, is only the easy part of their work. Rather strikingly, they have shown that a converse of this result holds: Given a compact contact manifold of arbitrary dimension, it admits an open book decomposition that is adapted to the contact structure in the way described above. Full details have not been published at the time of writing, but see Giroux’s talk [56] at the ICM 2002. 3.5.2. Branched covers A theorem of Hilden, Montesinos and Thickstun [63] states that every closed, orientable 3-manifold M admits a branched covering π : M → S 3 such that the upstairs branch set is a simple closed curve that bounds an embedded disc. (Moreover, the cover can be chosen 3-fold and simple, i.e. the monodromy representation of π1 (S 3 − K), where K is the branching set downstairs (a knot in S 3 ), represents the meridian of K by a transposition in the symmetric group S3 . This, however, is not relevant for our discussion.) It follows immediately, as announced in Section 3.3, that every closed, orientable 3-manifold is parallelisable: First of all, S 3 is parallelisable since it carries a Lie group structure (as the unit quaternions, for instance). Given M and a branched covering π : M → S 3 as above, there is a 3-ball D 3 ⊂ M containing the upstairs branch set. Outside of D 3 , the covering π is unbranched, so the 3-frame on S 3 can be lifted to a frame on M − D 3 . The bundle T M|D 3 is trivial, so the frame defined along ∂D 3 defines an element of SO(3) (cf. the footnote in the proof of Theorem 3.10). Since π2 (SO(3)) = 0, this frame extends over D 3 . In [59], Gonzalo uses this theorem to construct a contact structure on every closed, orientable 3-manifold M, in fact one with zero Euler class: Away from the branching set one can lift the standard contact structure from S 3 (which has Euler class zero: a trivialisation is given by two of the three (quaternionic) Hopf vector fields). A careful analysis of the branched covering map near the branching set then shows how to extend this contact structure over M (while keeping it trivial as 2-plane bundle). A branched covering construction for higher-dimensional contact manifolds is discussed in [43]. 3.5.3. . . . and more The work of Giroux [52], in which he initiated the study of convex surfaces in contact 3-manifolds, also contains a proof of Martinet’s theorem. An entirely different proof, due to S. Altschuler [5], finds contact structures from solutions to a certain parabolic differential equation for 1-forms on 3-manifolds. Some of

370

H. Geiges

these ideas have entered into the more far-reaching work of Eliashberg and Thurston on so-called ‘confoliations’ [32], that is, 1-forms satisfying α ∧ dα 0. 3.6. Tight and overtwisted The title of this section describes the fundamental dichotomy of contact structures in dimension 3 that has proved seminal for the development of the field. In order to motivate the notion of an overtwisted contact structure, as introduced by Eliashberg [21], we consider a ‘full’ Lutz twist as follows. Let (M, ξ ) be a contact 3-manifold and K ⊂ M a knot transverse to ξ . As before, identify K with S 1 × {0} ⊂ S 1 × D 2 ⊂ M such that ξ = ker(dθ + r 2 dϕ) on S 1 × D 2 . Now define a new contact structure ξ as in Section 3.4, with (h1 (r), h2 (r)) now as in Figure 17, that is, the boundary conditions are now h1 (r) = 1

and h2 (r) = r 2

for r ∈ [0, ε] ∪ [1 − ε, 1]

for some small ε > 0. L EMMA 3.17. A full Lutz twist does not change the homotopy class of ξ as a 2-plane field. P ROOF. Let (ht1 (r), ht2 (r)), r, t ∈ [0, 1], be a homotopy of paths such that 1. h01 ≡ 1, h02 (r) = r 2 , 2. h1i ≡ hi , i = 1, 2, 3. hti (r) = hi (r) for r ∈ [0, ε] ∪ [1 − ε, 1].

Fig. 17. A full Lutz twist.

Contact geometry

371

Let χ : [0, 1] → R be a smooth function which is identically zero near r = 0 and r = 1 and χ(r) > 0 for r ∈ [ε, 1 − ε]. Then αt = t (1 − t)χ(r) dr + ht1 (r) dθ + ht2 (r) dϕ is a homotopy from α0 = dθ + r 2 dϕ to α1 = h1 (r) dθ + h2 (r) dϕ through non-zero 1-forms. This homotopy stays fixed near r = 1, and so it defines a homotopy between ξ and ξ as 2-plane fields. Let r0 be the smaller of the two positive radii with h2 (r0 ) = 0 and consider the embedding φ : D 2 (r0 ) → S 1 × D 2 , (r, ϕ) → θ (r), r, ϕ , where θ (r) is a smooth function with θ (r0 ) = 0, θ (r) > 0 for 0 r < r0 , and θ (r) = 0 only for r = 0. We may require in addition that θ (r) = θ (0) − r 2 near r = 0. Then φ ∗ h1 (r) dθ + h2 (r) dϕ = h1 (r)θ (r) dr + h2 (r) dϕ is a differential 1-form on D 2 (r0 ) which vanishes only for r = 0, and along ∂D 2 (r0 ) the vector field ∂ϕ tangent to the boundary lies in the kernel of this 1-form, see Figure 18. In other words, the contact planes ker(h1 (r) dθ + h2 (r) dϕ) intersected with the tangent planes to the embedded disc φ(D 2 (r0 )) induce a singular 1-dimensional foliation on this disc with the boundary of this disc as closed leaf and precisely one singular point in the

Fig. 18. An overtwisted disc.

372

H. Geiges

Fig. 19. Characteristic foliation on an overtwisted disc.

interior of the disc (Figure 19; notice that the leaves of this foliation are the integral curves of the vector field h1 (r)θ (r) ∂ϕ − h2 (r) ∂r ). Such a disc is called an overtwisted disc. A contact structure ξ on a 3-manifold M is called overtwisted if (M, ξ ) contains an embedded overtwisted disc. In order to justify this terminology, observe that in the radially symmetric standard contact structure of Example 2.7, the angle by which the contact planes turn approaches π/2 asymptotically as r goes to infinity. By contrast, any contact manifold which has been constructed using at least one (simple) Lutz twist contains a similar cylindrical region where the contact planes twist by more than π in radial direction (at the smallest positive radius r0 with h2 (r0 ) = 0 the twisting angle has reached π ). We have shown the following: P ROPOSITION 3.18. Let ξ be a contact structure on M. By a full Lutz twist along any transversely embedded circle one obtains an overtwisted contact structure ξ that is homotopic to ξ as a 2-plane distribution. Together with the theorem of Lutz and Martinet we find that M contains an overtwisted contact structure in every homotopy class of 2-plane distributions. In fact, Eliashberg [21] has proved the following much stronger theorem. T HEOREM 3.19 (Eliashberg). On a closed, orientable 3-manifold there is a one-to-one correspondence between homotopy classes of overtwisted contact structures and homotopy classes of 2-plane distributions. This means that two overtwisted contact structures which are homotopic as 2-plane fields are actually homotopic as contact structures and hence isotopic by Gray’s stability theorem. Thus, it ‘only’ remains to classify contact structures that are not overtwisted. In [24] Eliashberg defined tight contact structures on a 3-manifold M as contact structures ξ for which there is no embedded disc D ⊂ M such that Dξ contains a limit cycle. So, by definition, overtwisted contact structures are not tight. In that same paper, as mentioned above in Section 2.4.5, Eliashberg goes on to show the converse with the help of the Elimination Lemma: non-overtwisted contact structures are tight.

Contact geometry

373

There are various ways to detect whether a contact structure is tight. Historically the first proof that a certain contact structure is tight is due to D. Bennequin [9, Corollary 2, p. 150]: T HEOREM 3.20 (Bennequin). The standard contact structure ξ0 on S 3 is tight. The steps of the proof are as follows: (i) First, Bennequin shows that if γ0 is a transverse knot in (S 3 , ξ0 ) with Seifert surface Σ, then the self-linking number of γ satisfies the inequality l(γ0 ) −χ(Σ). (ii) Second, he introduces an invariant for Legendrian knots; nowadays this is called the Thurston–Bennequin invariant: Let γ be a Legendrian knot in (S 3 , ξ0 ). Take a vector field X along γ transverse to ξ0 , and let γ be the push-off of γ in the direction of X. Then the Thurston–Bennequin invariant tb(γ ) is defined to be the linking number of γ and γ . (This invariant has an extension to homologically trivial Legendrian knots in arbitrary contact 3-manifolds.) (iii) By pushing γ in the direction of ±X, one obtains transverse curves γ ± (either of which is a candidate for γ in (ii)). One of these curves is positively transverse, the other negatively transverse to ξ0 . The self-linking number of γ ± is related to the Thurston– Bennequin invariant and a further invariant (the rotation number) of γ . The equation relating these three invariants implies tb(γ ) −χ(Σ), where Σ again denotes a Seifert surface for γ . In particular, a Legendrian unknot γ satisfies tb(γ ) < 0. This inequality would be violated by the vanishing cycle of an overtwisted disc (which has tb = 0), which proves that (S 3 , ξ0 ) is tight. R EMARK 3.21. (1) Eliashberg [25] generalised the Bennequin inequality l(γ0 ) −χ(Σ) for transverse knots (and the corresponding inequality for the Thurston–Bennequin invariant of Legendrian knots) to arbitrary tight contact 3-manifolds. Thus, whereas Bennequin established the tightness (without that name) of the standard contact structure on S 3 by proving the inequality that bears his name, that inequality is now seen, conversely, as a consequence of tightness. (2) In [9] Bennequin denotes the positively (respectively negatively) transverse push-off of the Legendrian knot γ by γ − (respectively γ + ). This has led to some sign errors in the literature. Notably, the ± in Proposition 2.2.1 of [25], relating the described invariants of γ and γ ± , needs to be reversed. C OROLLARY 3.22. The standard contact structure on R3 is tight. P ROOF. This is immediate from Proposition 2.13.

Here are further tests for tightness: 1. A closed contact 3-manifold (M, ξ ) is called symplectically fillable if there exists a compact symplectic manifold (W, ω) bounded by M such that

374

H. Geiges

• the restriction of ω to ξ does not vanish anywhere, • the orientation of M defined by ξ (i.e. the one for which ξ is positive) coincides with the orientation of M as boundary of the symplectic manifold (W, ω) (which is oriented by ω2 ). We then have the following result of Eliashberg [20, Theorem 3.2.1], [22] and Gromov [62, 2.4.D2 (b)], cf. [10]: T HEOREM 3.23 (Eliashberg–Gromov). A symplectically fillable contact structure is tight. E XAMPLE 3.24. The 4-ball D 4 ⊂ R4 with symplectic form ω = dx1 ∧ dy1 + dx2 ∧ dy2 is a symplectic filling of S 3 with its standard contact structure ξ0 . This gives an alternative proof of Bennequin’s theorem. ˜ ξ˜ ) → (M, ξ ) be a covering map and contactomorphism. If (M, ˜ ξ˜ ) is tight, 2. Let (M, then so is (M, ξ ), for any overtwisted disc in (M, ξ ) would lift to an overtwisted disc in ˜ ξ˜ ). (M, E XAMPLE 3.25. The contact structures ξn , n ∈ N, on the 3-torus T 3 defined by αn = cos(nθ1 ) dθ2 + sin(nθ1 ) dθ3 = 0 are tight: Lift the contact structure ξn to the universal cover R3 of T 3 ; there the contact structure is defined by the same equation αn = 0, but now θi ∈ R instead of θi ∈ R/2πZ ∼ = S 1 . Define a diffeomorphism f of R3 by f (x, y, z) = (y/n, z cos y + x sin y, z sin y − x cos y) =: (θ1 , θ2 , θ3 ). Then f ∗ αn = dz + x dy, so the lift of ξn to R3 is contactomorphic to the tight standard contact structure on R3 . Notice that it is possible for a tight contact structure to be finitely covered by an overtwisted contact structure. The first such examples were due to S. Makar-Limanov [88]. Other examples of this kind have been found by V. Colin [18] and R. Gompf [58]. 3. The following theorem of H. Hofer [65] relates the dynamics of the Reeb vector field to overtwistedness. T HEOREM 3.26 (Hofer). Let α be a contact form on a closed 3-manifold such that the contact structure ker α is overtwisted. Then the Reeb vector field of α has at least one contractible periodic orbit. E XAMPLE 3.27. The Reeb vector field Rn of the contact form αn of the preceding example is Rn = cos(nθ1 ) ∂θ2 + sin(nθ1 ) ∂θ3 .

Contact geometry

375

Thus, the orbits of Rn define constant slope foliations of the 2-tori {θ1 = const}; in particular, the periodic orbits of Rn are even homologically non-trivial. It follows, again, that the ξn are tight contact structures on T 3 . (This, admittedly, amounts to attacking starlings with rice puddings fired from catapults.5 )

3.7. Classification results In this section I summarise some of the known classification results for contact structures on 3-manifolds. By Eliashberg’s Theorem 3.19 it suffices to list the tight contact structures, up to isotopy or diffeomorphism, on a given closed 3-manifold. T HEOREM 3.28 (Eliashberg [24]). Any tight contact structure on S 3 is isotopic to the standard contact structure ξ0 . This theorem has a remarkable application in differential topology, viz., it leads to a new proof of Cerf’s theorem [16] that any diffeomorphism of S 3 extends to a diffeomorphism of the 4-ball D 4 . The idea is that the above theorem implies that any diffeomorphism of S 3 is isotopic to a contactomorphism of ξ0 . Eliashberg’s technique [22] of filling by holomorphic discs can then be used to show that such a contactomorphism extends to a diffeomorphism of D 4 . As remarked earlier (Remark 2.21), Eliashberg has also classified contact structures on R3 . Recall that homotopy classes of 2-plane distributions on S 3 are classified by π3 (S 2 ) ∼ = Z. By Theorem 3.19, each of these classes contains a unique (up to isotopy) overtwisted contact structure. When a point of S 3 is removed, each of these contact structures induces one on R3 , and Eliashberg [25] shows that they remain non-diffeomorphic there. Eliashberg shows further that, apart from this integer family of overtwisted contact structures, there is a unique tight contact structure on R3 (the standard one), and a single overtwisted one that is ‘overtwisted at infinity’ and cannot be compactified to a contact structure on S 3 . In general, the classification of contact structures up to diffeomorphism will differ from the classification up to isotopy. For instance, on the 3-torus T 3 we have the following diffeomorphism classification due to Y. Kanda [75]: T HEOREM 3.29 (Kanda). Every (positive) tight contact structure on T 3 is contactomorphic to one of the ξn , n ∈ N, described above. For n = m, the contact structures ξn and ξm are not contactomorphic. Giroux [54] had proved earlier that ξn for n 2 is not contactomorphic to ξ1 . On the other hand, all the ξn are homotopic as 2-plane fields to {dθ1 = 0}. This shows one way how Eliashberg’s classification Theorem 3.19 for overtwisted contact structures can fail for tight contact structures: • There are tight contact structures on T 3 that are homotopic as plane fields but not contactomorphic. 5 This turn of phrase originates from [93].

376

H. Geiges

P. Lisca and G. Mati´c [82] have found examples of the same kind on homology spheres by applying Seiberg–Witten theory to Stein fillings of contact manifolds, cf. also [78]. Eliashberg and L. Polterovich [31] have determined the isotopy classes of diffeomorphisms of T 3 that contain a contactomorphism of ξ1 : they correspond to exactly those elements of SL(3, Z) = π0 (Diff(T 3 )) that stabilise the subspace 0 ⊕ Z2 corresponding to the coordinates (θ2 , θ3 ). In combination with Kanda’s result, this allows one to give an isotopy classification of tight contact structures on T 3 . One particular consequence of the result of Eliashberg and Polterovich is the following: • There are tight contact structures on T 3 that are contactomorphic and homotopic as plane fields, but not isotopic (i.e. not homotopic through contact structures). Again, such examples also exist on homology spheres, as S. Akbulut and R. Matveyev [2] have shown. Another aspect of Eliashberg’s classification of overtwisted contact structures that fails to hold for tight structures is of course the existence of such a structure in every homotopy class of 2-plane fields, as is already demonstrated by the classification of contact structures on S 3 . Etnyre and K. Honda [37] have recently even found an example of a manifold—the connected some of two copies of the Poincaré sphere with opposite orientations—that does not admit any tight contact structure at all. For the classification of tight contact structures on lens spaces and T 2 -bundles over S 1 see [55,71,72]. A partial classification of tight contact structures on lens spaces had been obtained earlier in [34]. As further reading on 3-dimensional contact geometry I can recommend the lucid Bourbaki talk by Giroux [53]. This covers the ground up to Eliashberg’s classification of overtwisted contact structures and the uniqueness of the tight contact structure on S 3 .

4. A guide to the literature In this concluding section I give some recommendations for further reading, concentrating on those aspects of contact geometry that have not (or only briefly) been touched upon in earlier sections. Two general surveys that emphasise historical matters and describe the development of contact geometry from some of its earliest origins are the one by Lutz [87] and one by the present author [45]. One aspect of contact geometry that I have neglected in these notes is the Riemannian geometry of contact manifolds (leading, for instance, to Sasakian geometry). The survey by Lutz has some material on that; D. Blair [11] has recently published a monograph on the topic. There have also been various ideas for defining interesting families of contact structures. Again, the survey by Lutz has something to say on that; one such concept that has exhibited very intriguing ramifications—if this commercial break be permitted—was introduced in [48].

Contact geometry

377

4.1. Dimension 3 As mentioned earlier, Chapter 8 in [1] is in parts complementary to the present notes and has some material on surfaces in contact 3-manifolds. Other surveys and introductory texts on 3-dimensional contact geometry are the introductory lectures by Etnyre [35] and the Bourbaki talk by Giroux [52]. Good places to start further reading are two papers by Eliashberg: [24] for the classification of tight contact structures and [26] for knots in contact 3-manifolds. Concerning the latter, there is also a chapter by Etnyre [36] in a companion Handbook and an article by Etnyre and Honda [38] with an extensive introduction to that subject. The surveys [20] and [27] by Eliashberg are more general in scope, but also contain material about contact 3-manifolds. 3-dimensional contact topology has now become a fairly wide arena; apart from the work of Eliashberg, Giroux, Etnyre–Honda and others described earlier, I should also mention the results of Colin, who has, for instance, shown that surgery of index one (in particular: taking the connected sum) on a tight contact 3-manifold leads again to a tight contact structure [17]. Finally, Etnyre and Ng [40] have compiled a useful list of problems in 3-dimensional contact topology.

4.2. Higher dimensions The article [46] by the present author contains a survey of what was known at the time of writing about the existence of contact structures on higher-dimensional manifolds. One of the most important techniques for constructing contact manifolds in higher dimensions is the so-called contact surgery along isotropic spheres developed by Eliashberg [23] and Weinstein [105]. The latter is a very readable paper. For a recent application of this technique see [49]. Other constructions of contact manifolds (branched covers, gluing along codimension 2 contact submanifolds) are described in my paper [43]. Odd-dimensional tori are of course amongst the manifolds with the simplest global description, but they do not easily lend themselves to the construction of contact structures. In [86] Lutz found a contact structure on T 5 ; since then it has been one of the prize questions in contact geometry to find a contact structure on higher-dimensional tori. That prize, as it were, recently went to F. Bourgeois [13], who showed that indeed all odddimensional tori do admit a contact structure. His construction uses the result of Giroux and Mohsen [56,57] about open book decompositions adapted to contact structures in conjunction with the original proof of Lutz. With the help of the branched cover theorem described in [43] one can conclude further that every manifold of the form M × Σ with M a contact manifold and Σ a surface of genus at least 1 admits a contact structure. Concerning the classification of contact structures in higher dimensions, the first steps have been taken by Eliashberg [28] with the development of contact homology, which has been taken further in [29]. This has been used by Ustilovsky [102] to show that on S 4n+1 there exist infinitely many non-isomorphic contact structures that are homotopically equivalent (in the sense that they induce the same almost contact structure, i.e. reduction of

378

H. Geiges

the structure group of T S 4n+1 to 1 × U(2n)). Earlier results in this direction can be found in [44] in the context of various applications of contact surgery. 4.3. Symplectic fillings A survey on the various types of symplectic fillings of contact manifolds is given by Etnyre [33], cf. also the survey by Bennequin [10]. Etnyre and Honda [39] have recently shown that certain Seifert fibred 3-manifolds M admit tight contact structures ξ that are not symplectically semi-fillable, i.e. there is no symplectic filling W of (M, ξ ) even if W is allowed to have other contact boundary components. That paper also contains a good update on the general question of symplectic fillability. A related question is whether symplectic manifolds can have disconnected boundary of contact type (this corresponds to a stronger notion of symplectic filling defined via a Liouville vector field transverse to the boundary and pointing outwards). For (boundary) dimension 3 this is discussed by McDuff [91]; higher-dimensional symplectic manifolds with disconnected boundary of contact type have been constructed in [42]. Note added in proof: Eliashberg (Geom. Topol. 8 (2004), 277–293) has shown recently that every contact 3-manifold has a concave filling. This implies, in particular, that semifillable contact manifolds are always fillable. 4.4. Dynamics of the Reeb vector field In a seminal paper, Hofer [65] applied the method of pseudo-holomorphic curves, which had been introduced to symplectic geometry by Gromov [62], to solve (for large classes of contact 3-manifolds) the so-called Weinstein conjecture [104] concerning the existence of periodic orbits of the Reeb vector field of a given contact form. (In fact, one of the remarkable aspects of Hofer’s work is that in many instances it shows the existence of a periodic orbit of the Reeb vector field of any contact form defining a given contact structure.) A Bourbaki talk on the state of the art around the time when Weinstein formulated the conjecture that bears his name was given by Desolneux-Moulis [19]; another Bourbaki talk by Laudenbach describes Hofer’s contribution to the problem. The textbook by Hofer and Zehnder [70] addresses these issues, although its main emphasis, as is clear from the title, lies more in the direction of symplectic geometry and Hamiltonian dynamics. Two surveys by Hofer [66,67], and one by Hofer and Kriener [68], are more directly concerned with contact geometry. Of the three, [66] may be the best place to start, since it derives from a course of five lectures. In collaboration with Wysocki and Zehnder, Hofer has expanded his initial ideas into a far-reaching project on the characterisation of contact manifolds via the dynamics of the Reeb vector field, see, e.g., [69]. References [1] B. Aebischer, M. Borer, M. Kälin, Ch. Leuenberger and H.M. Reimann, Symplectic Geometry, Progr. Math., vol. 124, Birkhäuser, Basel (1994). [2] S. Akbulut and R. Matveyev, A note on contact structures, Pacific J. Math. 182 (1998), 201–204.

Contact geometry

379

[3] D.N. Akhiezer, Homogeneous complex manifolds, Several Complex Variables IV, S.G. Gindikin and G.M. Khenkin, eds, Encyclopaedia Math. Sci., vol. 10, Springer, Berlin (1990), 195–244. [4] J.W. Alexander, A lemma on systems of knotted curves, Proc. Nat. Acad. Sci. U.S.A. 9 (1923), 93–95. [5] S. Altschuler, A geometric heat flow for one-forms on three-dimensional manifolds, Illinois J. Math. 39 (1995), 98–118. [6] V.I. Arnold, Characteristic class entering in quantization conditions, Funct. Anal. Appl. 1 (1967), 1–13. [7] V.I. Arnold, Mathematical Methods of Classical Mechanics, Grad. Texts in Math., vol. 60, Springer, Berlin (1978). [8] A. Banyaga, The Structure of Classical Diffeomorphism Groups, Math. Appl., vol. 400, Kluwer, Dordrecht (1997). [9] D. Bennequin, Entrelacements et équations de Pfaff, IIIe Rencontre de Géométrie du Schnepfenried, vol. 1, Astérisque 107–108 (1983), 87–161. [10] D. Bennequin, Topologie symplectique, convexité holomorphe et structures de contact (d’après Y. Eliashberg, D. McDuff et al.), Séminaire Bourbaki, vol. 1989/1990, Astérisque 189–190 (1990), 285–323. [11] D.E. Blair, Riemannian Geometry of Contact and Symplectic Manifolds, Progr. Math., vol. 203, Birkhäuser, Basel (2002). [12] W.M. Boothby, Transitivity of the automorphisms of certain geometric structures, Trans. Amer. Math. Soc. 137 (1969), 93–100. [13] F. Bourgeois, Odd dimensional tori are contact manifolds, Internat. Math. Res. Notices 2002 (2002), 1571– 1574. [14] G.E. Bredon, Topology and Geometry, Grad. Texts in Math., vol. 139, Springer, Berlin (1993). [15] Th. Bröcker and K. Jänich, Einführung in die Differentialtopologie, Springer, Berlin (1973). [16] J. Cerf, Sur les difféomorphismes de la sphère de dimension trois (Γ4 = 0), Lecture Notes in Math., vol. 53, Springer, Berlin (1968). [17] V. Colin, Chirurgies d’indice un et isotopies de sphères dans les variétés de contact tendues, C. R. Acad. Sci. Paris Sér. I Math. 324 (1997), 659–663. [18] V. Colin, Recollement de variétés de contact tendues, Bull. Soc. Math. France 127 (1999), 43–69. [19] N. Desolneux-Moulis, Orbites périodiques des systèmes hamiltoniens autonomes (d’après Clarke, Ekeland-Lasry, Moser, Rabinowitz and Weinstein), Séminaire Bourbaki, vol. 1979/1980, Lecture Notes in Math., vol. 842, Springer, Berlin (1981), 156–173. [20] Y. Eliashberg, Three lectures on symplectic topology in Cala Gonone: Basic notions, problems and some methods, Conference on Differential Geometry and Topology (Sardinia, 1988), Rend. Sem. Fac. Sci. Univ. Cagliari, vol. 58 (1988), 27–49, suppl. [21] Y. Eliashberg, Classification of overtwisted contact structures on 3-manifolds, Invent. Math. 98 (1989), 623–637. [22] Y. Eliashberg, Filling by holomorphic discs and its applications, Geometry of Low-Dimensional Manifolds (Durham, 1989), vol. 2, London Math. Soc. Lecture Note Ser., vol. 151, Cambridge Univ. Press (1990), 45–67. [23] Y. Eliashberg, Topological characterization of Stein manifolds of dimension > 2, Internat. J. Math. 1 (1990), 29–46. [24] Y. Eliashberg, Contact 3-manifolds twenty years since J. Martinet’s work, Ann. Inst. Fourier (Grenoble) 42 (1–2) (1992), 165–192. [25] Y. Eliashberg, Classification of contact structures on R3 , Internat. Math. Res. Notices 1993 (1993), 87–91. [26] Y. Eliashberg, Legendrian and transversal knots in tight contact 3-manifolds, Topological Methods in Modern Mathematics (Stony Brook, 1991), Publish or Perish, Houston (1993), 171–193. [27] Y. Eliashberg, Symplectic topology in the nineties, Differential Geom. Appl. 9 (1998), 59–88. [28] Y. Eliashberg, Invariants in contact topology, Proceedings of the International Congress of Mathematicians (Berlin, 1998), vol. II, Doc. Math., Extra vol. II (1998), 327–338. [29] Y. Eliashberg, A. Givental and H. Hofer, Introduction to symplectic field theory, Geom. Funct. Anal., vol. 2000, Special vol., Part II (2000), 560–673. [30] Y. Eliashberg and N. Mishachev, Introduction to the h-principle, Grad. Stud. Math., vol. 48, American Mathematical Society, Providence (2002). [31] Y. Eliashberg and L. Polterovich, New applications of Luttinger’s surgery, Comment. Math. Helv. 69 (1994), 512–522.

380

H. Geiges

[32] Y. Eliashberg and W.P. Thurston, Confoliations, Univ. Lecture Ser., vol. 13, American Mathematical Society, Providence (1998). [33] J.B. Etnyre, Symplectic convexity in low-dimensional topology, Topology Appl. 88 (1998), 3–25. [34] J.B. Etnyre, Tight contact structures on lens spaces, Commun. Contemp. Math. 2 (2000), 559–577; erratum: ibid. 3 (2001), 649–652. [35] J.B. Etnyre, Introductory lectures on contact geometry, Topology and Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., vol. 71, American Mathematical Society, Providence (2003), 81– 107. [36] J.B. Etnyre, Legendrian and transverse knots, Handbook of Knot Theory, W. Menasco and M. Thistlethwaite, eds, in press. [37] J.B. Etnyre and K. Honda, On the nonexistence of tight contact structures, Ann. of Math. (2) 153 (2001), 749–766. [38] J.B. Etnyre and K. Honda, Knots and contact geometry I: Torus knots and the figure eight knot, J. Symplectic Geom. 1 (2001), 63–120. [39] J.B. Etnyre and K. Honda, Tight contact structures with no symplectic fillings, Invent. Math. 148 (2002), 609–626. [40] J.B. Etnyre and L.L. Ng, Problems in low dimensional contact topology, Topology and Geometry of Manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., vol. 71, American Mathematical Society, Providence (2003), 337–357. [41] H. Freudenthal, Zum Hopfschen Umkehrhomomorphismus, Ann. of Math. (2) 38 (1937), 847–853. [42] H. Geiges, Symplectic manifolds with disconnected boundary of contact type, Internat. Math. Res. Notices 1994 (1994), 23–30. [43] H. Geiges, Constructions of contact manifolds, Math. Proc. Cambridge Philos. Soc. 121 (1997), 455–464. [44] H. Geiges, Applications of contact surgery, Topology 36 (1997), 1193–1220. [45] H. Geiges, A brief history of contact geometry and topology, Exposition. Math. 19 (2001), 25–53. [46] H. Geiges, Contact topology in dimension greater than three, European Congress of Mathematics (Barcelona, 2000), vol. 2, Progress in Math., vol. 202, Birkhäuser, Basel (2001), 535–545. [47] H. Geiges, h-principles and flexibility in geometry, Mem. Amer. Math. Soc. 164 (2003), 779. [48] H. Geiges and J. Gonzalo, Contact geometry and complex surfaces, Invent. Math. 121 (1995), 147–209. [49] H. Geiges and C.B. Thomas, Contact structures, equivariant spin bordism, and periodic fundamental groups, Math. Ann. 320 (2001), 685–708. [50] V.L. Ginzburg, On closed characteristics of 2-forms, Ph.D. thesis, Berkeley (1990). [51] V.L. Ginzburg, Calculation of contact and symplectic cobordism groups, Topology 31 (1992), 767–773. [52] E. Giroux, Convexité en topologie de contact, Comment. Math. Helv. 66 (1991), 637–677. [53] E. Giroux, Topologie de contact en dimension 3 (autour des travaux de Yakov Eliashberg), Séminaire Bourbaki, vol. 1992/1993, Astérisque 216 (1993), 7–33. [54] E. Giroux, Une structure de contact, même tendue, est plus ou moins tordue, Ann. Sci. École Norm. Sup. (4) 27 (1994), 697–705. [55] E. Giroux, Structures de contact en dimension trois et bifurcations des feuilletages de surfaces, Invent. Math. 141 (2000), 615–689. [56] E. Giroux, Géométrie de contact: de la dimension trois vers les dimensions supérieures, Proceedings of the International Congress of Mathematicians (Beijing, 2002), vol. II, Higher Education Press, Beijing (2002), 405–414. [57] E. Giroux and J.-P. Mohsen, Structures de contact et fibrations symplectiques au-dessus du cercle, in preparation. [58] R.E. Gompf, Handlebody construction of Stein surfaces, Ann. of Math. (2) 148 (1998), 619–693. [59] J. Gonzalo, Branched covers and contact structures, Proc. Amer. Math. Soc. 101 (1987), 347–352. [60] D.H. Gottlieb, Partial transfers, Geometric Applications of Homotopy Theory I, Lecture Notes in Math., vol. 657, Springer, Berlin (1978), 255–266. [61] J.W. Gray, Some global properties of contact structures, Ann. of Math. (2) 69 (1959), 421–450. [62] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math. 82 (1985), 307–347. [63] H.M. Hilden, J.M. Montesinos and T. Thickstun, Closed oriented 3-manifolds as 3-fold branched coverings of S 3 of special type, Pacific J. Math. 65 (1976), 65–76. [64] M.W. Hirsch, Differential Topology, Grad. Texts in Math., vol. 33, Springer, Berlin (1976).

Contact geometry

381

[65] H. Hofer, Pseudoholomorphic curves in symplectizations with applications to the Weinstein conjecture in dimension 3, Invent. Math. 114 (1993), 515–563. [66] H. Hofer, Holomorphic curves and dynamics in dimension three, Symplectic Geometry and Topology (Park City, 1997), IAS/Park City Math. Ser., vol. 7, American Mathematical Society, Providence (1999), 35–101. [67] H. Hofer, Holomorphic curves and real three-dimensional dynamics, Geom. Funct. Anal., vol. 2000, Special vol., Part II (2000), 674–704. [68] H. Hofer and M. Kriener, Holomorphic curves in contact dynamics, Differential Equations (La Pietra, 1996), Proc. Sympos. Pure Math., vol. 65, American Mathematical Society, Providence (1999), 77–131. [69] H. Hofer, K. Wysocki and E. Zehnder, A characterisation of the tight three-sphere, Duke Math. J. 81 (1995), 159–226; erratum: ibid. 89 (1997), 603–617. [70] H. Hofer and E. Zehnder, Symplectic Invariants and Hamiltonian Dynamics, Birkhäuser, Basel (1994). [71] K. Honda, On the classification of tight contact structures I, Geom. Topol. 4 (2000), 309–368. [72] K. Honda, On the classification of tight contact structures II, J. Differential Geom. 55 (2000), 83–143. [73] H. Hopf, Zur Algebra der Abbildungen von Mannigfaltigkeiten, J. Reine Angew. Math. 105 (1930), 71–88. [74] D. Husemoller, Fibre Bundles, 3rd ed., Grad. Texts in Math., vol. 20, Springer, Berlin (1994). [75] Y. Kanda, The classification of tight contact structures on the 3-torus, Comm. Anal. Geom. 5 (1997), 413–438. [76] R.C. Kirby, The Topology of 4-Manifolds, Lecture Notes in Math., vol. 1374, Springer, Berlin (1989). [77] A.A. Kosinski, Differential Manifolds, Academic Press, Boston (1993). [78] P.B. Kronheimer and T.S. Mrowka, Monopoles and contact structures, Invent. Math. 130 (1997), 209–255. [79] F. Laudenbach, Orbites périodiques et courbes pseudo-holomorphes, application à la conjecture de Weinstein en dimension 3 (d’après H. Hofer et al.), Séminaire Bourbaki, vol. 1993/1994, Astérisque 227 (1995), 309–333. [80] P. Libermann and C.-M. Marle, Symplectic Geometry and Analytical Mechanics, Math. Appl., vol. 35, Reidel, Dordrecht (1987). [81] W.B.R. Lickorish, A representation of orientable combinatorial 3-manifolds, Ann. of Math. (2) 76 (1962), 531–540. [82] P. Lisca and G. Mati´c, Tight contact structures and Seiberg–Witten invariants, Invent. Math. 129 (1997), 509–525. [83] R. Lutz, Sur l’existence de certaines formes différentielles remarquables sur la sphère S 3 , C. R. Acad. Sci. Paris Sér. A–B 270 (1970), A1597–A1599. [84] R. Lutz, Sur quelques propriétés des formes differentielles en dimension trois, Thèse, Strasbourg (1971). [85] R. Lutz, Structures de contact sur les fibrés principaux en cercles de dimension trois, Ann. Inst. Fourier (Grenoble) 27 (3) (1977), 1–15. [86] R. Lutz, Sur la géométrie des structures de contact invariantes, Ann. Inst. Fourier (Grenoble) 29 (1) (1979), 283–306. [87] R. Lutz, Quelques remarques historiques et prospectives sur la géométrie de contact, Conference on Differential Geometry and Topology (Sardinia, 1988), Rend. Sem. Fac. Sci. Univ. Cagliari, vol. 58 (1988), 361–393, suppl. [88] S. Makar-Limanov, Tight contact structures on solid tori, Trans. Amer. Math. Soc. 350 (1998), 1013–1044. [89] J. Martinet, Sur les singularités des formes différentielles, Ann Inst. Fourier (Grenoble) 20 (1) (1970), 95–178. [90] J. Martinet, Formes de contact sur les variétés de dimension 3, Proc. Liverpool Singularities Sympos. II, Lecture Notes in Math., vol. 209, Springer, Berlin (1971), 142–163. [91] D. McDuff, Symplectic manifolds with contact type boundaries, Invent. Math. 103 (1991), 651–671. [92] D. McDuff and D. Salamon, Introduction to Symplectic Topology, 2nd ed., Oxford Univ. Press (1998). [93] S. Milligan, The Starlings, The Goon Show, Series 4, Special programme, broadcast 31st August 1954. [94] J.W. Milnor, Topology from the Differentiable Viewpoint, The University Press of Virginia, Charlottesville (1965). [95] J.W. Milnor and J.D. Stasheff, Characteristic Classes, Ann. of Math. Studies, vol. 76, Princeton Univ. Press, Princeton (1974). [96] J. Moser, On the volume elements on a manifold, Trans. Amer. Math. Soc. 120 (1965), 286–294. [97] D. Rolfsen, Knots and Links, Publish or Perish, Houston (1976).

382

H. Geiges

[98] N. Saveliev, Lectures on the Topology of 3-Manifolds, de Gruyter, Berlin (1999). [99] N. Steenrod, The Topology of Fibre Bundles, Princeton Univ. Press, Princeton (1951). [100] I. Tamura, Topology of Foliations: An Introduction, Transl. Math. Monogr., vol. 97, American Mathematical Society, Providence (1992). [101] W.P. Thurston and H.E. Winkelnkemper, On the existence of contact forms, Proc. Amer. Math. Soc. 52 (1975), 345–347. [102] I. Ustilovsky, Infinitely many contact structures on S 4m+1 , Internat. Math. Res. Notices 1999 (1999), 781–791. [103] A.H. Wallace, Modifications and cobounding manifolds, Canad. J. Math. 12 (1960), 503–528. [104] A. Weinstein, On the hypotheses of Rabinowitz’ periodic orbit theorem, J. Differential Equations 33 (1979), 353–358. [105] A. Weinstein, Contact surgery and symplectic handlebodies, Hokkaido Math. J. 20 (1991), 241–251.

CHAPTER 6

Complex Differential Geometry Ion Mihai∗ Faculty of Mathematics, University of Bucharest, Str. Academiei 14, 010014 Bucharest, Romania E-mail: imihai@fmi.unibuc.ro

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 1. Complex manifolds . . . . . . . . . . . . . . . . . . 2. Almost complex structures . . . . . . . . . . . . . . 3. Dolbeault Lemma . . . . . . . . . . . . . . . . . . . 4. Kaehler manifolds . . . . . . . . . . . . . . . . . . . 5. Complex space forms . . . . . . . . . . . . . . . . . 6. Laplace–Beltrami operator on a Hermitian manifold 7. Harmonic differential forms on Kaehler manifolds . 8. Applications . . . . . . . . . . . . . . . . . . . . . . 9. Chern classes . . . . . . . . . . . . . . . . . . . . . 10. Deformation of complex structures . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

385 385 391 393 395 401 403 407 413 416 426 433

* This chapter was written while the author has visited Michigan State University, Department of Mathematics,

supported by a Fulbright grant. He expresses his hearty thanks to Professor B.Y. Chen for the hospitality he received during this visit. The author is very obliged to Professors D. Blair, B.Y. Chen, F. Defever, F. Dillen, G.D. Ludden, K. Matsumoto, R. Rosca, B. Suceav˘a, Y. Tazawa and L. Verstraelen for valuable discussions on the topics presented therein. HANDBOOK OF DIFFERENTIAL GEOMETRY, VOL. II Edited by F.J.E. Dillen and L.C.A. Verstraelen © 2006 Elsevier B.V. All rights reserved 383

This page intentionally left blank

Complex differential geometry

385

Introduction The geometry of complex manifolds, in particular Kaehler manifolds, is an important research topic in Differential Geometry. In the present chapter, we present the basic notions and certain important results in complex differential geometry. First, we define complex and almost complex manifolds and give standard examples. The most interesting class of complex manifolds are the Kaehler manifolds. Locally, a Kaehlerian metric differs from the Euclidean metric on the complex space Cn starting with the second power of the Taylor series. There are topological obstructions to the existence of Kaehlerian metrics on a compact complex manifold. We give examples and counterexamples. We introduce complex space forms and we state a Schur-like theorem for complex space forms. Next, we study the differential forms on a Kaehler manifold. Then, we prove the complex version of Hodge theorem. As an application, it follows that the Betti numbers of odd order on a compact Kaehler manifold are even. An interesting example of an almost Kaehler manifold which do not admit any Kaehlerian metric is the Thurston–Abbena manifold. The Iwasawa manifold is a complex manifold which does not carry any Kaehlerian metric. The Chern classes are introduced axiomatically and their construction is given. Some applications are derived. Last section deals with the deformation of complex structures in the sense of Kodaira. We state the theorems of existence and completeness. The number of moduli of a compact complex manifold is introduced.

1. Complex manifolds An n-dimensional complex manifold is a pairing (M, A), where M is a non-empty set and A = {(Uα , hα ) | α ∈ A} is a family of mappings satisfying the following properties: (i) For each α ∈ A, Uα is a subset of M and hα : Uα → Cn is one-to-one. (ii) The family {Uα }α∈A is a covering of M, i.e., M=

Uα .

α∈A

(iii) For each α, β ∈ A, the set hα (Uα ∩ Uβ ) is an open subset in Cn and the mapping n hβ ◦ h−1 α : hα (Uα ∩ Uβ ) → C

is holomorphic. (iv) If (M, A ) satisfies the properties (i)–(iii) and A ⊂ A , then A = A (i.e., (M, A) is maximal). R EMARK . If (M, B) = {(Uβ , hβ ) | β ∈ B} is a pairing satisfying the properties (i)–(iii) and we denote by A the set of all the pairs (U, h), with U ⊂ M, h : U → Cn one-to-one

386

I. Mihai

such that for each β ∈ B, h(U ∩ Uβ ), hβ (U ∩ Uβ ) ⊂ Cn are open sets and the mappings n hβ ◦ h−1 : h(U ∩ Uβ ) → Cn and h ◦ h−1 β : hβ (U ∩ Uβ ) → C respectively are holomorphic, then (M, A) is an n-dimensional complex manifold. Thus, in order to define a complex structure on a non-empty set M, it is sufficient to construct a family B satisfying the properties (i)–(iii). E XAMPLES . 1. Each open subset U of Cn admits a canonical complex structure, taking B = {(U, iU )}, where iU : U → Cn is the inclusion map of U into Cn . In particular, GL(n, C) is a complex Lie group. 2. The complex projective space P n (C) is an n-dimensional complex manifold. On − {0}, we define the equivalence relation 1 z , . . . , zn+1 ∼ w 1 , . . . , w n+1

Cn+1

⇔

∃λ ∈ C − {0} such that w i = λzi , i ∈ {1, . . . , n + 1}.

We denote by P n (C) the quotient space P n (C) = z1 , z2 , . . . , zn+1 z1 , z2 , . . . , zn+1 ∈ Cn+1 − {0} . For any α ∈ {1, . . . , n + 1}, we put Uα = {[z1 , z2 , . . . , zn+1 ] | zα = 0} and hα : Uα → Cn , hα z1 , . . . , zn+1 =

z1 zα−1 zα+1 zn+1 . , . . . , , , . . . , zα zα zα zα

Then the family B = {(Uα , hα ), α ∈ {1, . . . , n + 1}} satisfies the properties (i)–(iii). For α = β, the mapping hβ ◦ h−1 α is given by z¯ γ =

zγ , zβ

γ∈ / {α, β};

z¯ α =

1 . zβ

E XERCISE . Determine the holomorphic vector fields on the complex projective spaces P 1 (C) and P 2 (C). Solution. Locally, a holomorphic vector field v on P 1 (C) has the form d , v = az2 + bz + c dz

a, b, c ∈ C,

and a holomorphic vector field v on P 2 (C) is given by 2 ∂ v = c1 + c2 z 1 + c3 z 2 + c7 z 1 + c8 z 1 z 2 ∂z1 2 ∂ + c4 + c5 z 1 + c6 z 2 + c7 z 1 z 2 + c8 z 2 , ∂z2 where c1 , . . . , c8 ∈ R.

Complex differential geometry

387

3. The complex Grassmann manifold. Let Gp (Cn ) be the set of all p-dimensional linear subspaces of Cn . We define on Gp (Cn ) a complex structure of dimension p(n − p). Let {e1 , . . . , en } be the standard basis of Cn . For any sequence 1 i1 < i2 < · · · < ip n, let πi1 ...ip : Cn → Ei1 ...ip = Span{ei1 , . . . , eip } be the orthogonal projection. We denote by Ui1 ...ip = V ∈ Gp (Cn ); πi1 ...ip |V is a linear isomorphism . Let V ∈ Ui1 ...ip . Then, for each k ∈ {1, . . . , p}, there exists a unique fk ∈ V such that πi1 ...ip (fk ) = eik . Obviously,

f k = ei k +

ckl el .

l ∈{i / 1 ,...,ip }

Consider the mapping hi1 ...ip : Ui1 ...ip → Cp(n−p) , hi1 ...ip (V ) = ckl k∈{i

/ 1 ,...,ip } 1 ,...,ip }, l ∈{i

.

If we put B = {hi1 ...ip | 1 i1 < · · · < ip n}, then Gp (Cn ) becomes a complex manifold of dimension p(n − p). R EMARK . G1 (Cn+1 ) = P n (C). 4. The Calabi manifolds S 2m+1 × S 2n+1 [9]. The differentiable manifold M = S 2m+1 × S 2n+1 has real dimension 2m + 2n + 2. We may consider S

2m+1

= ξ ∈C

S

2n+1

= η∈C

m+1

m+1

|ξ | = 1 , α 2

α=1

n+1

β 2

η = 1 ,

n+1

β=1

then M ⊂ Cm+1 × Cn+1 . Let j ∈ {1, . . . , m + 1} and k ∈ {1, . . . , n + 1} and denote by Uj k = (ξ, η) ∈ M | ξ j ηk = 0 . We will construct a family of mappings hj k : Uj k → Cm+n+1 , which defines a complex structure on M. Put

388

I. Mihai

hj k (ξ, η) 1 ξ ξ j −1 ξ j +1 ξ m+1 η1 ηk−1 ηk+1 ηn+1 = ,..., j , j ,..., j , k ,..., k , k ,..., k ,t , ξj ξ ξ ξ η η η η where t=

1 log ξ j + i log ηk 2πi

mod (1, i).

5. Each orientable surface M is a 1-dimensional complex manifold. Let g be a Riemannian metric on M. By a theorem of Lichtenstein, locally, the metric can be written as g = λ2 (dx 2 +dy 2 ), with λ > 0. Putting z = x +iy, M becomes a complex manifold of dimension 1. In particular, the 2-sphere S 2 admits a complex structure. We will give a direct construction of a complex structure on S 2 . It is known that S 2 is a 2-dimensional differentiable manifold. Let S2 =

2 2 2 1 2 3 u , u , u ∈ E3 u1 + u2 + u3 = 1 .

Denote by UN = S 2 − {N } and US = S 2 − {S}, where N = (0, 0, 1) and S = (0, 0, −1) are the north and south poles, respectively. Consider the stereographic projections hN : UN → R2 ,

hS : US → R2

from the north and south poles, respectively. Their equations are u1 u2 , , 1 − u3 1 − u3 1 2 3 u1 u2 , hS u , u , u = , 1 + u3 1 + u3

hN u1 , u2 , u3 =

respectively. We define h˜ N : UN → C and h˜ S : US → C, by h˜ N u1 , u2 , u3 =

u1 u2 +i 3 1−u 1 − u3

h˜ S u1 , u2 , u3 =

u1 u2 − i . 1 + u3 1 + u3

and

Complex differential geometry

389

The mapping hS ◦ h−1 N : C − {0} → C,

1 hS ◦ h−1 N (z) = , z

is holomorphic. Thus the family B = {(UN , U˜ N ), (US , h˜ S )} satisfies the conditions (i)–(iii), and therefore the sphere S 2 is a 1-dimensional complex manifold. Next, we will indicate one procedure for obtaining new complex manifolds. Let M be a complex manifold. The set of all automorphisms of M is a group G endowed with the composition of mappings. Any subgroup G of G is called a group of automorphisms of M. For any p ∈ M, the set Gp = {g(p) | g ∈ G} is the orbit of G at p. Obviously Gp ∩ Gq = ∅ if and only if q ∈ Gp . The set of all orbits, denoted by M/G , is the quotient space of M via G. E XAMPLE . Let M = Cn+1 − {0}. Each g ∈ C∗ defines an automorphism of M by g : z1 , . . . , zn+1 → gz1 , . . . , gzn+1 . Thus C∗ becomes a group of automorphisms of M and M/C∗ = P n (C). D EFINITION . A group G of automorphisms of the complex manifold M is called properly discontinuous if for any compact subsets K1 and K2 in M, the set {g ∈ G | g(K1 )∩K2 = ∅} is finite. In this case, each orbit Gp is a discrete set. We say that G is fixed points free if each g ∈ G − {1M } has no fixed points. R EMARK . In the above example, C∗ is fixed points free, but it is not properly discontinuous. T HEOREM 1.1. Let M be an n-dimensional complex manifold and G a fixed points free and properly discontinuous group of automorphisms. Then M/G carries a natural n-dimensional complex structure induced by the complex structure of M. Using this theorem, we can obtain other examples of complex manifolds. 6. The complex torus is defined as follows. Let {ω1 , . . . , ω2n } be 2n vectors in Cn linearly independent over R. We define a fixed points free and properly discontinuous group of automorphisms of Cn , by G = z → z +

2n

mj ωj mj ∈ Z, j ∈ {1, . . . , 2n} .

j =1

The quotient space T n = Cn /G , which we call the n-dimensional complex torus, becomes a complex manifold.

390

I. Mihai

R EMARK . For n = 1, T 1 is an algebraic curve. If n 2, then T n is not necessarily an algebraic manifold. T n is an algebraic manifold if and only if its period matrix is a Riemann matrix (see [54]). An n-dimensional complex manifold M is said to be complex parallelizable if there exist n holomorphic vector fields which are linearly independent at every point of M. Every complex torus is complex parallelizable. H.G. Wang [61] proved that every compact complex parallelizable manifold can be written as a quotient space G/D of a complex Lie group G by a discrete subgroup D. 7. The Hopf manifolds. Let M = Cn − {0} and α1 , . . . , αn ∈ C such that |αj | > 1 (j ∈ {1, . . . , n}) and let G be the cyclic group spanned by the automorphism g of M, given by g : z = z1 , . . . , zn → g(z) = α1 z1 , . . . , αn zn . Applying the above theorem, M/G has a complex structure; it is called Hopf manifold. It is known (see [37]) that M/G is diffeomorphic with S 1 × S 2n−1 . Thus, Hopf manifolds are a particular case of Calabi manifolds. We want to point out that they are not algebraic manifolds. E XERCISE . Let L = {([z], ξ ) ∈ P n (C) × Cn+1 | ξ ∈ [z]}. Prove that (L, pr1 , P n (C)) is a holomorphic vector bundle of rank 1 (called the tautological vector bundle over the complex projective space). 8. The blowing-up at a point. Consider a small ball B = Br (0) ⊂ Cn , with n 2. Let z = (z1 , . . . , zn ) be the standard ¯ is the complex n-dimensional coordinates of Cn . The blowing-up of B at 0, denoted by B, manifold defined by B¯ = (z, w) ∈ B × P n−1 (C) | zi w j = zj w i , ∀1 i < j n , where w = [w 1 , . . . , w n ] ∈ P n−1 (C). If we cover P n−1 (C) by the open subsets Ui = {w ∈ P n−1 (C) | w i = 0}, i = 1, . . . , n, then B × P