Finsler geometry generalizes Riemannian geometry in the same sense that Banach spaces generalize Hilbert spaces. This book presents an expository account of seven important topics in Riemann–Finsler geometry, which have recently undergone significant development but have not had a detailed pedagogical treatment elsewhere. Each article will open the door to an active area of research and is suitable for a special topics course in graduatelevel differential geometry. ´ Alvarez and Thompson discuss the theory of volumes for normed spaces and Finsler spaces and show how it unifies a wide range of geometric inequalities. Bellettini studies the evolution of crystals, where the driving agent is the mean curvature of the facets. Aikou reviews the essential role played by Finsler metrics in complex differential geometry. Chandler and Wong explain why parametrized jet bundles admit only Finsler metrics and develop machinery which they use to prove the Kobayashi conjecture (1960) and a special case of the Green–Griffiths (1979) conjecture. Bao and Robles focus on the flag and Ricci curvatures of Finsler manifolds, with an emphasis on Einstein metrics of Randers type. Rademacher gives a detailed and new account of his Sphere Theorem for nonreversible Finsler metrics. Shen’s article explains why Finsler manifolds are colorful objects and examines the interplay among the flag, S, and Landsberg curvatures in Finsler geometry.
Mathematical Sciences Research Institute Publications
50
A Sampler of Riemann–Finsler Geometry
Mathematical Sciences Research Institute Publications 1 2 3 4 5 6 7 8 9 10–11 12–13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
Freed/Uhlenbeck: Instantons and FourManifolds, second edition Chern (ed.): Seminar on Nonlinear Partial Differential Equations Lepowsky/Mandelstam/Singer (eds.): Vertex Operators in Mathematics and Physics Kac (ed.): Infinite Dimensional Groups with Applications Blackadar: KTheory for Operator Algebras, second edition Moore (ed.): Group Representations, Ergodic Theory, Operator Algebras, and Mathematical Physics Chorin/Majda (eds.): Wave Motion: Theory, Modelling, and Computation Gersten (ed.): Essays in Group Theory Moore/Schochet: Global Analysis on Foliated Spaces Drasin/Earle/Gehring/Kra/Marden (eds.): Holomorphic Functions and Moduli Ni/Peletier/Serrin (eds.): Nonlinear Diffusion Equations and Their Equilibrium States Goodman/de la Harpe/Jones: Coxeter Graphs and Towers of Algebras Hochster/Huneke/Sally (eds.): Commutative Algebra Ihara/Ribet/Serre (eds.): Galois Groups over Q Concus/Finn/Hoffman (eds.): Geometric Analysis and Computer Graphics Bryant/Chern/Gardner/Goldschmidt/Griffiths: Exterior Differential Systems Alperin (ed.): Arboreal Group Theory Dazord/Weinstein (eds.): Symplectic Geometry, Groupoids, and Integrable Systems Moschovakis (ed.): Logic from Computer Science Ratiu (ed.): The Geometry of Hamiltonian Systems Baumslag/Miller (eds.): Algorithms and Classification in Combinatorial Group Theory Montgomery/Small (eds.): Noncommutative Rings Akbulut/King: Topology of Real Algebraic Sets Judah/Just/Woodin (eds.): Set Theory of the Continuum Carlsson/Cohen/Hsiang/Jones (eds.): Algebraic Topology and Its Applications Clemens/Koll´ ar (eds.): Current Topics in Complex Algebraic Geometry Nowakowski (ed.): Games of No Chance Grove/Petersen (eds.): Comparison Geometry Levy (ed.): Flavors of Geometry Cecil/Chern (eds.): Tight and Taut Submanifolds Axler/McCarthy/Sarason (eds.): Holomorphic Spaces Ball/Milman (eds.): Convex Geometric Analysis Levy (ed.): The Eightfold Way Gavosto/Krantz/McCallum (eds.): Contemporary Issues in Mathematics Education Schneider/Siu (eds.): Several Complex Variables Billera/Bj¨ orner/Green/Simion/Stanley (eds.): New Perspectives in Geometric Combinatorics Haskell/Pillay/Steinhorn (eds.): Model Theory, Algebra, and Geometry Bleher/Its (eds.): Random Matrix Models and Their Applications Schneps (ed.): Galois Groups and Fundamental Groups Nowakowski (ed.): More Games of No Chance Montgomery/Schneider (eds.): New Directions in Hopf Algebras Buhler/Stevenhagen (eds.): Algorithmic Number Theory Jensen/Ledet/Yui: Generic Polynomials: Constructive Aspects of the Inverse Galois Problem Rockmore/Healy (eds.): Modern Signal Processing Uhlmann (ed.): Inside Out: Inverse Problems and Applications Gross/Kotiuga: Electromagnetic Theory and Computation: A Topological Approach Darmon/Zhang (ed.): Heegner Points and Rankin LSeries Bao/Bryant/Chern/Shen (ed.): A Sampler of Riemann–Finsler Geometry Volumes 1–4 and 6–27 are published by SpringerVerlag
A Sampler of Riemann–Finsler Geometry
Edited by
David Bao University of Houston
Robert L. Bryant Duke University
ShiingShen Chern UC Berkeley and Nankai Institute of Mathematics
Zhongmin Shen Indiana University – Purdue University Indianapolis
David Bao Department of Mathematics University of Houston Houston, TX 772043008 United States
[email protected] Robert L. Bryant Duke University ShiingShen Chern Nankai Institute of Mathematics Zhongmin Shen IUPUI
Series Editor Silvio Levy Mathematical Sciences Research Institute 17 Gauss Way Berkeley, CA 94720 United States MSRI Editorial Committee Hugo Rossi (chair) Alexandre Chorin Silvio Levy Jill Mesirov Robert Osserman Peter Sarnak
The Mathematical Sciences Research Institute wishes to acknowledge support by the National Science Foundation. This material is based upon work supported by NSF Grant 9810361.
published by the press syndicate of the university of cambridge The Pitt Building, Trumpington Street, Cambridge, United Kingdom cambridge university press The Edinburgh Building, Cambridge CB2 2RU, UK 40 West 20th Street, New York, NY 100114211, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia Ruiz de Alarc´ on 13, 28014 Madrid, Spain Dock House, The Waterfront, Cape Town 8001, South Africa http://www.cambridge.org c Mathematical Sciences Research Institute 2004
First published 2004 Printed in the United States of America A catalogue record for this book is available from the British Library. Library of Congress Cataloging in Publication data ISBN 0 521 83181 4 hardback
Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Contents
Preface
ix
Synopses
xi
Volumes on Normed and Finsler Spaces ´ J. C. Alvarez Paiva and A. C. Thompson
1
Anisotropic and Crystalline Mean Curvature Flow Giovanni Bellettini
49
Finsler Geometry on Complex Vector Bundles Tadashi Aikou
83
Finsler Geometry of Holomorphic Jet Bundles Karen Chandler and PitMann Wong
107
Ricci and Flag Curvatures in Finsler Geometry David Bao and Colleen Robles
197
Nonreversible Finsler Metrics of Positive Flag Curvature HansBert Rademacher
261
Landsberg Curvature, SCurvature and Riemann Curvature Zhongmin Shen
303
Index
357
vii
Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Preface
This volume contains seven expository articles and concerns three facets of Riemann–Finsler geometry that have undergone important recent developments: 1. The concept of volumes on normed spaces and Finsler manifolds, and crystalline motion by mean curvature in phase transitions. 2. The essential role played by Finsler metrics in complex manifold theory, together with the resolution of the Kobayashi conjecture and a special case of the Green–Griffiths conjecture. 3. The significance of the flag, Ricci, and Scurvatures of Finsler metrics, as well as the Sphere Theorem for nonreversible Finsler structures. Conspicuously absent from the above are two highly geometrical areas: Bryant’s use of exterior differential systems to understand Finsler metrics of constant flag curvature, and Foulon’s dynamical systems approach to Finsler geometry. They are not included here because reasonable expositions already exist in a special Chern issue of the Houston Journal of Mathematics 28 (2002), 221–262 (Bryant) and 263–292 (Foulon). Our goal is to render the aforementioned developments accessible to the differential geometry community at large. It is not our intention to present an encyclopedic picture of the field. What we do covet are concrete examples, instructive graphics, meaningful computations, and care in organizing technical arguments. The resulting articles appear to have met these criteria at an aboveaverage level. All the articles have been refereed. In fact, a total of 26 referee reports were obtained, some addressing the mathematics, others critiquing expository matters. After a few rounds of revision, each article was lineedited by at least one mathematician who is not familiar with the topic in question, in the hope that this would uncover most typographical mistakes. It is certain that, in spite of these attempts at quality control, layers of blemishes are still awaiting their turn to surface. We urge readers to bring those mistakes to the attention of one of us (Bao). ix
x
PREFACE
Thanks are due all reviewers who generously donated their time. We are also grateful to Dr. Silvio Levy, Editor of the MSRI Series, for his prompt responses to all our queries.
David Bao, Houston, USA Robert L. Bryant, Durham, USA ShiingShen Chern, Nankai, China Zhongmin Shen, Indianapolis, USA
Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Synopses
1. Part One ´ • Alvarez–Thompson: This paper is a concise introduction to the theory of volumes on normed and Finsler spaces. The definitions of Holmes–Thompson, Busemann, and Benson–Gromov are studied and their convexity (ellipticity) properties are discussed in detail. The authors show how the theory of volumes provides a unified context for a diverse range of geometric inequalities. The article is intended for students and researchers in differential, integral, and convex geometry. • Bellettini: Crystalline motion driven by the mean curvature is an evolution process arising in material science and phase transitions. It is an anisotropic flow in an ambient space endowed with a piecewise linear norm. For threedimensional crystals, the crystalline mean curvature of a facet is defined and identified with the initial velocity in the evolution process. Facets with constant crystalline mean curvature are important because they are expected not to break or curve under the evolution. The problem of characterizing such facets is discussed.
2. Part Two • Aikou: This article highlights the essential role played by Finsler metrics in complex differential geometry. It describes a few situations for which techniques based solely on Hermitian metrics are hopelessly inadequate. These include Kobayashi’s characterization of negative holomorphic vector bundles over compact complex manifolds, in terms of the existence of negatively curved pseudoconvex Finsler metrics. • Chandler–Wong: The authors present the proof of the Kobayashi conjecture (1960) on the hyperbolicity of generic algebraic surfaces of degree d > 5 in P3 . They also address the Green–Griffiths conjecture (1979) that every holomorphic map f : C → X to a surface X of general type is algebraically degenerate. Their paper establishes the latter for the special case where X is xi
xii
SYNOPSES
minimal, Pic(X) ∼ = Z, and pg (X) > 0. The main tool used is their generalization of the classical Schwarz lemma for complex curves, to varieties of every dimension. In this crucial step, algebraic geometric arguments are used to construct a Finsler metric of logarithmic type, thereby reducing the problem to one in which a certain estimate (the lemma of logarithmic derivatives) is applicable.
3. Part Three • Bao–Robles: Many recent developments have advanced our understanding of the flag and Ricci curvatures of Finsler metrics. This paper is a uniform presentation of these results and their underlying techniques. Included is a geometric definition of Einstein–Finsler metrics. Einstein metrics of Randers type are studied via their representation as solutions to Zermelo navigation on Riemannian manifolds. This viewpoint leads to the classification of all constant flag curvature Randers metrics. It also yields a Schur lemma, and settles a question of rigidity in three dimensions, for Einstein–Randers metrics. The theory is illustrated with a diverse array of explicit examples. • Rademacher: The author shows in detail how the classical Sphere Theorem in Riemannian geometry is extended to the case of nonreversible Finsler metrics. The proof hinges on a fruitful definition of the notion of reversibility, and how that can be used to effect some crucial estimates, such as the injectivity radius, the length of nonminimal geodesics between two fixed points, and the length of nonconstant geodesic loops. The proof also capitalizes on an idea of Klingenberg: that Morse theory of the energy functional allows us to circumvent Toponogov’s comparison theorem for geodesic triangles. This idea renders irrelevant the “handicap” that, in Finsler geometry proper, there is no Toponogov’s theorem. • Shen: This paper is about the interaction between the generalized Riemann curvature and other nonRiemannian quantities in Finsler geometry. The latter include the Scurvature, whose vanishing is equivalent to having constant “distortion” along each geodesic. The Scurvature quantifies some aspect of the change in the Minkowski model (“infinitesimal color pattern”) as one moves from one tangent space to another, along geodesics on a Finslerian landscape. The information it provides is complementary to that supplied by a certain contracted version of the Berwald curvature. These theoretical constructs are exemplified by Finsler metrics with a broad array of special curvature properties. The author also proves a number of local and global theorems using certain curvature equations along geodesics.
Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Volumes on Normed and Finsler Spaces ´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Contents 1. Introduction 2. A Short Review of the Geometry of Normed Spaces 2.1. Maps between normed spaces 2.2. Dual spaces and polar bodies 2.3. Sociology of normed spaces 3. Volumes on Normed Spaces 3.1. Examples of definitions of volume in normed spaces 3.2. The volume of the unit ball 3.3. Relationship between the definitions of volume 3.4. Extension to Finsler manifolds 4. kVolume Densities 4.1. Examples of kvolume densities 4.2. Convexity of (n−1)volume densities 4.3. Convexity properties of kvolume densities 5. Length and Area in TwoDimensional Normed Spaces 6. Area on FiniteDimensional Normed Spaces 6.1. Injectivity and range of the area definition 6.2. Area of the unit sphere 6.3. Mixed volumes and the isoperimetrix 6.4. Geometry of the isoperimetrix Acknowledgments References
1 3 5 6 7 9 10 13 14 15 17 18 20 26 30 35 36 39 40 43 45 45
1. Introduction The study of volumes and areas on normed and Finsler spaces is a relatively new field that comprises and unifies large domains of convexity, geometric tomography, and integral geometry. It opens many classical unsolved problems in these fields to powerful techniques in global differential geometry, and suggests new challenging problems that are delightfully geometric and simple to state. Keywords: Minkowski geometry, Hausdorff measure, Holmes–Thompson volume, Finsler manifold, isoperimetric inequality.
1
2
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
The theory starts with a simple question: How does one measure volume on a finitedimensional normed space? At first sight, this question may seem either unmotivated or trivial: normed spaces are metric spaces and we can measure volume using the Hausdorff measure, period. However, if one starts asking simple, naive questions one discovers the depth of the problem. Even if one is unwilling to consider that definitions of volume other than the Hausdorff measure are not only possible but may even be better, one is faced with questions such as these: What is the (n−1)dimensional Hausdorff measure of the unit sphere of an ndimensional normed space? Do flat regions minimize area? For what normed spaces are metric balls also the solutions of the isoperimetric problem? These questions, first posed in convexgeometric language by Busemann and Petty [1956], are still open, at least in their full generality. However, one should not assume too quickly that the subject is impossible. Some beautiful results and striking connections have been found. For example, the fact that the (n−1)Hausdorff measure in a normed space determines the norm is equivalent to the fact that the areas of the central sections determine a convex body that is symmetric with respect to the origin. This, in turn, follows from the study of the spherical Radon transform. The fact that regions in hyperplanes are areaminimizing is equivalent to the fact that the intersection body of a convex body that is symmetric with respect to the origin is also convex. But the true interest of the theory can only be appreciated if one is willing to challenge Busemann’s dictum that the natural volume in a normed or Finsler space is the Hausdorff measure. Indeed, thinking of a normed or Finsler space as an anisotropic medium where the speed of a light ray depends on its direction, we are led to consider a completely different notion of volume, which has become known as the Holmes–Thompson volume. This notion of volume, introduced in [Holmes and Thompson 1979], uncovers striking connections between integral geometry, convexity, and Hamiltonian systems. For example, in a recent series of papers, [Schneider and Wieacker 1997], [Alvarez and Fernandes 1998], [Alvarez and Fernandes 1999], [Schneider 2001], and [Schneider 2002], it was shown that the classical integral geometric formulas in Euclidean spaces can be extended to normed and even to projective Finsler spaces (the solutions of Hilbert’s fourth problem) if the areas of submanifolds are measured with the Holmes–Thompson definition. That extensions of this kind are not possible with the Busemann definition was shown by Schneider [Schneider 2001]. Using Finsler techniques, Burago and Ivanov [2001] have proved that a flat twodimensional disc in a finitedimensional normed space minimizes area among all other immersed discs with the same boundary. Ivanov [2001] has shown, among other things, that Pu’s isosystolic inequality for Riemannian metrics in the projective plane extends to Finsler metrics, and the Finslerian extension of Berger’s infinitesimal isosystolic inequality for Riemannian metrics on real ´ projective spaces of arbitrary dimension has been proved by Alvarez [2002].
VOLUMES ON NORMED AND FINSLER SPACES
3
Despite these and other recent interdisciplinary results, we believe that the most surprising feature of the Holmes–Thompson definition is the way in which it organizes a large portion of convexity into a coherent theory. For example, the sharp upper bound for the volume of the unit ball of a normed space is given by the Blaschke–Santal´ o inequality; the conjectured sharp lower bound is Mahler’s conjecture; and the reconstruction of the norm from the area functional is equivalent to the famous Minkowski’s problem of reconstructing a convex body from the knowledge of its curvature as a function of its unit normals. In this paper, we have attempted to provide students and researchers in Finsler and global differential geometry with a clear and concise introduction to the theory of volumes on normed and Finsler spaces. To do this, we have avoided the temptation to use auxiliary Euclidean structures to describe the various concepts and constructions. While these auxiliary structures may render some of the proofs simpler, they hinder the understanding of the subject and make the application of the ideas and techniques to Finsler spaces much more cumbersome. We also believe that by presenting the results and techniques in intrinsic terms we can make some of the beautiful results of convexity more accessible and enticing to differential geometers. In the course of our writing we had to make some tough choices as to what material should be left out as either too advanced or too specialized. At the end we decided that we would concentrate on the basic questions and techniques of the theory, while doing our best to exhibit the general abstract framework that makes the theory of volumes on normed spaces into a sort of Grand Unified Theory for many problems in convexity and Finsler geometry. As a result there is little Finsler geometry per se in the pages that follow. However, just as tensors, forms, spinors, and Clifford algebras developed in invariant form have immediate use in Riemannian geometry, the more geometric constructions with norms, convex bodies, and kvolume densities that make up the heart of this paper have immediate applications to Finsler geometry.
2. A Short Review of the Geometry of Normed Spaces This section is a quick review of the geometry of finitedimensional normed spaces adapted to the needs and language of Finsler geometry. Unless stated otherwise, all vector spaces in this article are real and finitedimensional. We suggest that the reader merely browse through this section and come back to it if and when it becomes necessary. Definition 2.1. A norm on a vector space X is a function k · k : X → [0, ∞) satisfying the following properties of positivity, homogeneity, and convexity: (1) If kxk = 0, then x = 0;
4
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
(2) If t is a real number, then ktxk = tkxk; (3) For any two vectors x and y in X, kx + yk ≤ kxk + kyk. If (X, k · k) is a finitedimensional normed space, the set BX := {x ∈ X : kxk ≤ 1} is the unit ball of X and the boundary of BX , SX , is its unit sphere. Notice that BX is a compact, convex set with nonempty interior. In short, it is a convex body in X. Moreover, it is symmetric with respect to the origin. Conversely, if B ⊂ X is a centered convex body (i.e., a convex body symmetric with respect to the origin), it is the unit ball of the norm kxk := inf {t ≥ 0 : ty = x for some y ∈ B}. We shall now describe various classes of normed spaces that will appear repeatedly throughout the paper. Euclidean spaces. A Euclidean structure on a finitedimensional vector space X is a choice of a symmetric, positivedefinite quadratic form Q : X → R. The normed space (X, Q1/2 ) will be called a Euclidean space. Note that a normed space is Euclidean if and only if its unit sphere is an ellipsoid, which is if and only if the norm satisfies the parallelogram identity: kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 . Exercise 2.2. Let B ⊂ R n be a convex body symmetric with respect to the origin. Show that if the intersection of B with every 2dimensional plane passing through the origin is an ellipse, then B is an ellipsoid. The `p spaces. If p ≥ 1 is a real number, the function 1/p kxkp := x1 p + · · · + xn p
is a norm on R n . When p tends to infinity, kxkp converges to kxk∞ := max{x1 , . . . , xn }. The normed space (R n , k · kp ), 1 ≤ p ≤ ∞, is denoted by `np . The unit ball of `n∞ is the ndimensional cube of side length two, while the unit ball of `n1 is the ndimensional crosspolytope. In general, norms whose unit balls are polytopes are called crystalline norms. Subspaces of L1 ([0, 1], dx). The space of measurable functions f : [0, 1] → R satisfying Z 1 kf k := f (x) dx < ∞ 0
is a normed space denoted by L1 ([0, 1], dx). The geometry of finitedimensional subspaces of L1 ([0, 1], dx) is closely related to problems of volume, area, and integral geometry on normed and Finsler spaces. In the next few paragraphs,
VOLUMES ON NORMED AND FINSLER SPACES
5
we will summarize the properties of these subspaces that will be used in the rest of the paper. For proofs, references, and to learn more about hypermetric spaces, we recommend the landmark paper [Bolker 1969], as well as the surveys [Schneider and Weil 1983] and [Goodey and Weil 1993]. First we begin with a beautiful metric characterization of the subspaces of L1 ([0, 1], dx). Definition 2.3. A metric space (M, d) is said to be hypermetric if it satisfies the following stronger version of the triangle inequality: If m 1 , . . . , mk are elements P of M and b1 , . . . , bk are integers with i bi = 1, then k X
d(mi , mj )bi bj ≤ 0.
i,j=1
Theorem 2.4. A finitedimensional normed space is hypermetric if and only if it is isometric to a subspace of L1 ([0, 1], dx). An important analytic characterization of a hypermetric normed space can be given through the Fourier transform of its norm: Theorem 2.5. A norm on R n is hypermetric if and only if its distributional Fourier transform is a nonnegative measure. The characterizations above, important as they are, are hard to grasp at first sight. A much more visual approach will be given after we review the duality of normed spaces. Minkowski spaces. Minkowski spaces are normed spaces with strict smoothness and convexity properties. In precise terms, a norm k · k on a vector space X is said to be a Minkowski norm if it is smooth outside the origin and the Hessian 2 of the function k · k at every nonzero point is a positivedefinite quadratic form. The unit sphere of a Minkowski space X is a smooth convex hypersurface S X such that for any Euclidean structure on X the principal curvatures of S X are positive. 2.1. Maps between normed spaces. An important feature of the geometry of normed spaces is that the space of linear maps between two normed spaces carries a natural norm. Definition 2.6. If T : X → Y is a linear map between normed spaces (X, k · kX ) and (Y, k · kY ), we define the norm of T as the supremum of kT xkY taken over all vectors x ∈ X with kxkX ≤ 1. A linear map T : X → Y is said to be short if its norm is less than or equal to one. In other words, a short linear map does not increase distances. Two important types of short linear maps between normed spaces are isometric embeddings and isometric submersions:
6
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Definition 2.7. An injective linear map T : X → Y between normed spaces (X, k · kX ) and (Y, k · kY ) is said to be an isometric embedding if kT xkY = kxkX for all vectors x ∈ X. Definition 2.8. A surjective linear map T : X → Y between normed spaces (X, k · kX ) and (Y, k · kY ) is said to be an isometric submersion if kT xkY = inf {kvkX : v ∈ X and T v = T x} for all vectors x ∈ X. In terms of the unit balls, T : X → Y is an isometric embedding if and only if T (BX ) = T (X) ∩ BY , and T is an isometric submersion if and only if T (BX ) = BY . 2.2. Dual spaces and polar bodies. From the previous paragraph, we see that if (X, k · k) is a normed space, then the set of all linear maps onto the onedimensional normed space (R,  · ) carries a natural norm. The resulting normed ∗ space is called the dual of (X, k · k) and is denoted by (X ∗ , k · k ). It is easy to see that the double dual (i.e., the dual of the dual) of a finitedimensional normed space can be naturally identified with the space itself. The unit ball of ∗ (X ∗ , k · k ) is said to be the polar of the unit ball of (X, k · k). Example. H¨ older’s inequality implies that if p > 1, the dual of ` np is `nq , where 1/p + 1/q = 1. Likewise, it is easy to see that the dual of ` n1 is `n∞ . If T : X 7→ Y is a linear map then the dual map T ∗ : Y ∗ 7→ X ∗ is defined by (T ∗ ξ)(x) = ξ(T x). Note that kT ∗ k = kT k. Exercise 2.9. Show that if T : X → Y is an isometric embedding between normed spaces X and Y , the dual map T ∗ : Y ∗ → X ∗ is an isometric submersion. Many of the geometric constructions in convex geometry and the geometry of normed spaces are functorial. More precisely, if we denote by N the category whose objects are finitedimensional normed spaces and whose morphisms are short linear maps, many classical constructions define functors from N to itself. ∗
Exercise 2.10. Show that the assignment (X, k · k) 7→ (X ∗ , k · k ) is a contravariant functor from N to N. Duals of hypermetric normed spaces. As advertised earlier in this section, the notion of duality allows us to give a more geometric characterization of hypermetric spaces. Definition 2.11. A polytope in a vector space X is said to be a zonotope if all of its faces are symmetric. A convex body is said to be a zonoid if it is the limit (in the Hausdorff topology) of zonotopes.
VOLUMES ON NORMED AND FINSLER SPACES
7
Notice that an ndimensional cube, as well as all its linear projections, are zonotopes. In fact, it can be shown that any zonotope is the linear projection of a cube (see, for example, Theorem 3.3 in [Bolker 1969]). Theorem 2.12. Let X be a finitedimensional normed space with unit ball BX . The dual of X is hypermetric if and only if BX is a zonoid . Notice that this immediately implies that the space ` n1 , n ≥ 1, is hypermetric. Duality in Minkowski spaces. If (X, k · k) is a Minkowski space, the differential 2 of the function L := k · k /2, 1 d kx + tyk2t=0 , 2 dt is a continuous linear map from X to X ∗ that is smooth outside the origin and homogeneous of degree one. This map is usually called the Legendre transform, although that term is also used to describe some related concepts (see, for example, § 2.2 in [H¨ ormander 1994]). The following exercise describes the most important properties of the Legendre transform. dL(x)(y) :=
Exercise 2.13. Let (X, k · k) be a Minkowski space and let L : X \ 0 → X∗ \ 0 be its Legendre transform. (1) Show that if x ∈ X is a unit vector, then L(x) is the unique covector ξ ∈ X ∗ such that the equation ξ · y = 1 describes the tangent plane to the unit sphere SX at the point x. (2) Show that the Legendre transform defines a diffeomorphism between the unit sphere and its polar. (3) Show that the inverse of the Legendre transform from X \ 0 to X ∗ \ 0 is just the Legendre transform from X ∗ \ 0 to X \ 0. (4) Show that the Legendre transform is linear if and only if X is a Euclidean space. Exercise 2.14. Show that a normed space is a Minkowski space if its unit sphere and the unit sphere of its dual are smooth. 2.3. Sociology of normed spaces. If k · k1 and k · k2 are two norms on a finitedimensional vector space X, it is easy to see that there are positive numbers m and M such that mk · k2 ≤ k · k1 ≤ M k · k2 . If we take the numbers m and M such that the inequalities are sharp, then log(M/m) is a good measure of how far away one norm is from the other. For example, the following wellknown result states that we can always approximate a norm by one whose unit sphere is a polytope or by one such that its unit sphere and the unit sphere of its dual are smooth.
8
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Proposition 2.15. Let k · k be a norm on the finitedimensional vector space X. Given ε > 0, there exist a crystalline norm k · k1 and a Minkowski norm k · k2 on X such that k · k1 ≤ k · k ≤ (1 + ε)k · k1 , k · k2 ≤ k · k ≤ (1 + ε)k · k2 . For a short proof see Lemma 2.3.2 in [H¨ ormander 1994] . In many circumstances, one wants to measure how far is one normed space from being isometric to another. The straightforward adaptation of the previous idea leads us to the following notion: Definition 2.16. The Banach–Mazur distance between ndimensional normed spaces X and Y , is the infimum of the numbers log(kT kkT −1 k), where T ranges over all invertible linear maps from X to Y . Notice that the Banach–Mazur distance is a distance on the set of isometry classes of ndimensional normed spaces: two such spaces are at distance zero if and only if they are isometric. An important question is to determine how far a general ndimensional normed space is from being Euclidean. The answer rests on two results of independent interest: Theorem 2.17 (Loewner). If B is a convex body in an ndimensional vector space X, there exists a unique ndimensional ellipsoid E ⊂ B such that for any Lebesgue measure on X, the ratio vol(B)/vol(E) is minimal . Theorem 2.18 [John 1948]. Let X be an ndimensional normed space with unit ball B. If E ⊂ B is the Loewner ellipsoid of B, then √ E ⊂ B ⊂ nE. Exercise 2.19. Show that the Banach–Mazur distance from an ndimensional normed space to a Euclidean space is at most log(n)/2. The structure of the set of isometry classes of ndimensional normed spaces is given by the following theorem (see [Thompson 1996, page 73] for references and some of the history on the subject): Theorem 2.20. The set of isometry classes of ndimensional normed spaces, Mn , provided with the Banach–Mazur distance is a compact, connected metric space. The Banach–Mazur compactum, Mn , enters naturally into Finsler geometry by the following construction: Let π : ζ → M be a vector bundle with ndimensional fibers such that every fiber ζm = π −1 (m) carries a norm that varies continuously with the base point (a Finsler bundle). The (continuous) map I : M −→ Mn
VOLUMES ON NORMED AND FINSLER SPACES
9
that assigns to each point m ∈ M the isometry class of ζ m measures how the norms vary from point to point. Currently, there are not many results that describe the map I under different geometric and/or topological hypotheses on the bundle. However the following exercise (and its extension in [Gromov 1967]) shows that such results are possible. Exercise 2.21. Let π : ζ → S 2 be a Finsler bundle whose fibers are 2dimensional. Show that if the bundle is nontrivial and the map I is constant, then the image of S 2 under I is the isometry class of 2dimensional Euclidean spaces. A corollary of this exercise is that if X is a threedimensional normed space such that all its twodimensional subspaces are isometric, then X is Euclidean. Another interesting corollary is that a Berwald (Finsler) metric on S 2 must be Riemannian.
3. Volumes on Normed Spaces In defining the notion of volume on normed spaces, it is best to adopt an axiomatic approach. We shall impose some minimal set of conditions that are reasonable and then try to find out to what extent they can be satisfied, and to what point they determine our choices. In a normed space, all translations are isometries. This suggests that we require the volume of a set to be invariant under translations. Since any finitedimensional normed space is a locally compact, commutative group, we can apply the following theorem of Haar: Theorem 3.1. If µ is a translationinvariant measure on R n for which all compact sets have finite measure and all open sets have positive measure, then µ is a constant multiple of the Lebesgue measure. Proofs of this theorem can be found in many places. A full account is given in [Cohn 1980] and an abbreviated version in [Thompson 1996]. In the light of Haar’s theorem, in order to give a definition of volume in every normed space, we must assign to every normed space X a multiple of the Lebesgue measure. Since the Lebesgue measure is not intrinsically defined (it depends on a choice of basis for X), it is best to describe this assignment as a choice of a norm µ in the 1dimensional vector space Λ n X, where n is the dimension of X; if x1 , . . . , xn ∈ X, we define µ(x1 ∧ x2 ∧ · · · ∧ xn ) as the volume of the parallelotope formed by these vectors. Another natural requirement is monotonicity: if X and Y are ndimensional normed spaces and T : X → Y is a short linear map (i.e., a linear map that does not increase distances), we require that T does not increase volumes. Notice that this implies that isometries between normed spaces are volumepreserving.
10
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
The monotonicity requirement makes a definition of volume on normed spaces into a functor from N to itself that takes the ndimensional normed space (X, k · k) to the 1dimensional normed space (Λn X, µ). While we shall often abandon this viewpoint, it is a guiding principle throughout the paper with which we would like to acquaint the reader early on. Definition 3.2. A definition of volume on normed spaces assigns to every ndimensional, n ≥ 1, normed space X a normed space (Λ n X, µX ) with the following properties: (1) If X and Y are ndimensional normed spaces and T : X → Y is a short linear map , then the induced linear map T∗ : Λn X → Λn Y is also short. (2) The map X 7→ (Λn X, µX ) is continuous with respect to the topology induced by the Banach–Mazur distance. (3) If X is Euclidean, then µX is the standard Euclidean volume on X. Before presenting the principal definitions of volume in normed spaces, let us make the first link between these concepts and the affine geometry of convex bodies. Exercise 3.3. Assume we have a definition of volume in normed spaces and use it to assign a number to any centrally symmetric convex body B ⊂ R n by the following procedure: Consider R n as the normed space X whose unit ball is B and compute Z V(B) := µX (B) =
µX .
B
Show that if T : R n → R n is an invertible linear map, then V(B) = V(T (B)), and write the monotonicity condition in terms of the affine invariant V. Notice that we can turn the tables and start by considering a suitable affine invariant V of centered convex bodies and give a definition of volume in normed spaces by prescribing that the volume of the unit ball B of a normed space X be given by V(B). Exercise 3.4. Let µ be a definition of volume for 2dimensional normed spaces. Use John’s theorem to show that if B is the unit disc of a twodimensional normed space X, then π/2 ≤ µX (B) ≤ 2π. 3.1. Examples of definitions of volume in normed spaces. The study of the four definitions of volume we shall describe below makes up the most important part of the theory of volumes on normed and Finsler spaces. The Busemann definition. The Busemann volume of an ndimensional normed space is that multiple of the Lebesgue measure for which the volume of the unit ball equals the volume of the Euclidean unit ball in dimension n, ε n , . In other words, we have chosen as our affine invariant the constant ε n , where n is the dimension of the space.
VOLUMES ON NORMED AND FINSLER SPACES
11
Another way to define the Busemann volume of a normed space X is by setting εn , µb (x1 ∧ x2 ∧ · · · ∧ xn ) = vol(B; x1 ∧ x2 ∧ · · · ∧ xn ) where the notation vol(B; x1 ∧ x2 ∧ · · · ∧ xn ) indicates the volume of B in the Lebesgue measure determined by the basis x1 , . . . , xn . Using Brunn–Minkowski theory, Busemann showed in [1947] that the Busemann volume of an ndimensional normed space equals its ndimensional Hausdorff measure. Hence, from the viewpoint of metric geometry, this is a very natural definition. Exercise 3.5. Show that the Busemann definition of volume satisfies the axioms in Definition 3.2. The Holmes–Thompson definition. Let X be an ndimensional normed space and let B ∗ ⊂ X ∗ be the dual unit ball. If x1 , . . . , xn is a basis of X and ξ 1 , . . . , ξ n is its dual basis, define ∗ µht (x1 ∧ x2 ∧ · · · ∧ xn ) := ε−1 n vol(B ; ξ 1 ∧ ξ 2 ∧ · · · ∧ ξ n ).
Another way of defining the Holmes–Thompson volume is by considering the set B × B ∗ in the product space X × X ∗ . Since X × X ∗ has a natural symplectic structure defined by ω((x1 , ξ 1 ), (x2 , ξ 2 )) := ξ 2 (x1 ) − ξ 1 (x2 ), it has a canonical volume (the symplectic or Liouville volume) defined by the nth exterior power ω n of ω, divided by n!. The Holmes–Thompson volume of the ndimensional normed space X is the multiple of the Lebesgue measure for which the volume of the unit ball equals the Liouville volume of B × B ∗ divided by the volume of the Euclidean unit ball of dimension n. We mention in passing that in convex geometry it is usual to denote the Liouville volume of B × B ∗ as the volume product of B, vp(B). The Holmes–Thompson definition — introduced in [Holmes and Thompson 1979] — was originally motivated by purely geometric considerations. However, from the physical point of view it is the natural definition of volume if we think of normed spaces as homogeneous, anisotropic media: media in which the speed of light varies with the direction of the light ray, but not with the point at which the propagation of light originates. It is interesting to remark that the Busemann definition and the Holmes– Thompson definition are dual functors: to obtain the Holmes–Thompson volume of an ndimensional normed space X we pass to the dual normed space X ∗ , we apply the “Busemann functor” to obtain (Λn X ∗ , µbX ∗ ) and then pass to the dual of the normed space (Λn X ∗ , µbX ∗ ). Exercise 3.6. Consider a definition of volume (X, k · k) 7→ (Λ n X, µX ), where n is the dimension of X, and define its dual definition by the map (X, k · k) 7→
12
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
(Λn X, µ∗X ) := (Λn X ∗ , µX ∗ )∗ . Show that µ∗ also satisfies the axioms in Definition 3.2. The notion of duality is somewhat mysterious and is closely related to the duality between intersections and projections proposed in [Lutwak 1988], and which led to the development of the dual Brunn–Minkowski theory. We shall have a little more to say about this duality after presenting a second dual pair of volume definitions. Gromov’s mass. If X is an ndimensional normed space, define µ m : Λn X → [0, ∞) by the formula µ (a) := inf m
n Y
kxi k : x1 ∧ x2 ∧ · · · ∧ xn = a .
i=1
Another way to define the mass of an ndimensional normed space X is as the multiple of the Lebesgue measure for which the volume of the maximal crosspolytope inscribed to the unit ball is 2n /n!. Exercise 3.7. Consider the 2dimensional normed space whose unit disc D is a regular hexagon. What is µm (D)? The Benson definition or Gromov’s mass∗. One way to make the Benson definition is as the dual of mass: given an ndimensional normed space X together with a basis x1 , . . . , xn , we take the dual basis ξ 1 , . . . , ξ n in X ∗ and define µm∗ X (x1 ∧ x2 ∧ · · · ∧ xn ) :=
1 µm X ∗ (ξ 1
∧ ξ2 ∧ · · · ∧ ξn )
.
This is Gromov’s definition [1983]. Benson [1962] originally defined the mass∗ of an ndimensional normed space as the multiple of the Lebesgue measure for which the volume of a minimal parallelotope circumscribed to the unit ball is 2n . Exercise 3.8. Consider the 2dimensional normed space whose unit disc D is a regular hexagon. What is µm∗ (D)? The following exercise gives a third characterization of mass∗. Exercise 3.9. Let X be an ndimensional normed space and let B be its unit ball. A basis ξ 1 , . . . , ξ n of X ∗ is said to be short if ξ i (x) ≤ 1 for all x ∈ B and all i, 1 ≤ i ≤ n (i.e., if all the vectors in the basis are in the dual unit ball). Show that for any nvector a ∈ Λn X µm∗ (a) = sup{ξ 1 ∧ ξ2 ∧ · · · ∧ ξ n (a) : ξ 1 , . . . , ξ n is a short basis of X ∗ } It is not hard to come up with other definitions of volume. For example, instead of considering inscribed crosspolytopes and circumscribed parallelotopes one might consider maximal inscribed or minimal circumscribed ellipsoids (as in Loewner’s theorem cited above) and then specify the volume of either to be
VOLUMES ON NORMED AND FINSLER SPACES
13
εn . However, as we shall see in the next two sections, a good definition of volume must satisfy some additional conditions that are very hard to verify. The examples given above are important mainly because their study provides a common context to many problems in convex, integral, and differential geometry. 3.2. The volume of the unit ball. If we are given a definition of volume and a normed space, we would like to compute the volume of the unit ball. This is, of course, trivial if we work with the Busemann definition, but for the other definitions it is a challenging problem. Let us start with some simple experiments. Example 3.10. In the table below we use 7 different norms in R 3 whose unit balls are, in order, the Euclidean unit ball; the cube with vertices at (±1, ±1, ±1); the octahedron with vertices at ±(1, 0, 0), ±(0, 1, 0), ±(0, 0, 1); the right cylinder over the unit circle in the xyplane and with −1 ≤ z ≤ 1; its dual, the double cone which is the convex hull of the unit circle in the xyplane and the points ±(0, 0, 1); the affine image of the cuboctahedron that has vertices at ±(1, 0, 0), ±(0, 1, 0), ±(0, 0, 1), ±(1, −1, 0), ±(1, 0, −1), ±(0, 1, −1); and its dual, the affine image of the rhombic dodecahedron, that has vertices at ±(1, 1, 1), ±(0, 1, 1), ±(1, 0, 1), ±(1, 1, 0), ±(0, 0, 1), ±(0, 1, 0), ±(1, 0, 0). These are listed in the first column. In the subsequent columns are the volumes of each unit ball using the different definitions of volume. The ball B
µb (B)
µht (B)
µm∗ (B)
µm (B)
ball cube octahedron cylinder double cone cuboctahedron rhombic dodecahedron
4π/3 4π/3 4π/3 4π/3 4π/3 4π/3 4π/3
4π/3 8/π 8/π π π 10/π 10/π
4π/3 8 16/3 2π 4π/3 20/3 4
4π/3 2 4/3 π 2π/3 10/3 2
Exercise 3.11. Verify these numbers. Given a definition of volume, an interesting problem is to determine sharp upper and lower bounds for the volume of the unit balls of ndimensional normed spaces. In the case of the Holmes–Thompson definition, this question has a classical reformulation: give sharp upper and lower bounds for the volume product of an ndimensional centrallysymmetric convex body. ´ inequality). The Holmes–Thompson Theorem 3.12 (Blaschke–Santalo volume of the unit ball of an ndimensional normed space is less than or equal to the volume of the Euclidean unit ball of dimension n. Moreover , equality holds if and only if the space is Euclidean. The sharp lower bound for the Holmes–Thompson volume of unit balls is a reformulation of a longstanding conjecture of Mahler [1939]:
14
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Conjecture. The Holmes–Thompson volume of the unit ball of an ndimensional normed space is greater than or equal to 4n /εn n!. Moreover , equality holds if and only if the unit ball is a parallelotope or a crosspolytope. This conjecture has been verified by Mahler [1939] in the twodimensional case and by Reisner [1985, 1986] in the case when either the normed space or its dual is hypermetric. For µm∗ (B) the upper bound of 2n is attained for a parallelotope and for m µ (B) the equivalent lower bound of 2n /n! is attained by a crosspolytope. One also has µm∗ (B) ≥ 2n /n! and µm (B) ≤ 2n but these are far from optimal; better bounds will be obtained after studying the relationship between the different definitions of volume. 3.3. Relationship between the definitions of volume. There are several relationships between the various measures we are considering. For example, the Blaschke–Santal´ o inequality is clearly equivalent to the following theorem: b Theorem 3.13. If X is an ndimensional normed space, then µht X ≤ µX with equality if and only if X is Euclidean.
For mass and mass∗ we have the following inequality: m∗ Proposition 3.14. If X is an ndimensional normed space, then µm X ≤ µX .
Proof. Let P be a minimal circumscribed parallelotope to the unit ball B. Then (see for example [Thompson 1999], but there are many other possible references) the midpoint of each facet of P is a point of contact with B. The convex hull of these midpoints is a crosspolytope C inscribed to B. Also, if P is given the volume 2n , then C has volume 2n /n!. Hence, in this situation, a maximal inscribed crosspolytope will have volume greater than or equal to 2n /n!. b Theorem 3.15. If X is an ndimensional normed space, then µm X ≤ µX with equality if and only if X is Euclidean.
The proof depends on the following theorem of McKinney [1974]: Theorem 3.16. Let K ⊂ X be a convex set symmetric about the origin and let S be a maximal simplex contained in K with one vertex at the origin, then for any Lebesgue measure λ λ(S)/λ(K) ≥ 1/n!εn with equality if and only if K is an ellipsoid . Proof of Theorem 3.15. If B is the unit ball of X then µb (B) = εn and µm (B) = 2n λ(B)/n!λ(C) where C is a maximal crosspolytope inscribed to B. Moreover, C is the convex hull of S ∪ −S, where S is a maximal simplex inscribed to B with one vertex at the origin. It follows from the theorem that λ(C)/λ(B) ≥ 2n /n!εn which, upon rearrangement, gives µb (B) ≥ µm (B).
VOLUMES ON NORMED AND FINSLER SPACES
15
The relationship between mass∗ and the Holmes–Thompson volume follows from Theorem 3.15 and the following simple exercise: Exercise 3.17. Let µ and ν be two definitions of volume, and let µ ∗ and ν ∗ be their dual definitions. Show that if for every normed space X µX ≤ ν X ,
∗ νX ≤ µ∗X .
then
m∗ Corollary 3.18. If X is an ndimensional normed space, then µht X ≤ µX with equality if and only if X is Euclidean.
The previous inequalities are summarized by the diagram µm∗
µb @ @
? µht
@
@ @ R @? µm
Notice that as a consequence of the Mahler–Reisner inequality we have the following lower bounds for the mass and mass∗ of unit balls in normed spaces and their duals. Corollary 3.19. For any unit ball B, we have µm (B) ≤ εn and , if B is either a zonoid or the dual of a zonoid , µm∗ (B) ≥ 4n /n!εn . Problem. Is the mass∗ of the unit ball of an ndimensional normed space at least 4n /n!εn ? This is a weaker version of Mahler’s conjecture. 3.4. Extension to Finsler manifolds Definition 3.20. A volume density on anndimensional manifold M is a continuous function Φ : Λn T M −→ R such that for every point m ∈ M the restriction of Φ to Λ n Tm M is a norm. A volume density is said to be smooth if the function Φ is smooth outside the zero section. If M is an oriented manifold and Φ is a volume density on M , then we can define a volume form Ω on M whose value at a basis x1 , . . . , xn of Tm M is Φ(m; x1 ∧ x2 ∧ · · · ∧ xn ) if the basis is positively oriented and −Φ(m; x1 ∧ x2 ∧ · · · ∧ xn ) if it is negatively oriented. For any positively oriented ndimensional submanifold U ⊂ M we have that Z Z Φ= Ω. U
U
However, the integral of a volume density does not depend on the orientation and volume densities can be defined in nonorientable manifolds like the projective plane where no volume form exists.
16
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Definition 3.21. A continuous Finsler metric F on a manifold M assigns a norm, F (m, ·), to each tangent space Tm M in such a way that the norm varies continuously with the base point. A continuous Finsler manifold is a pair (M, F ), where M is a manifold and F is a continuous Finsler metric on M . An important class of examples of Finsler manifolds are finitedimensional submanifolds of normed spaces. If M is a submanifold of a finitedimensional normed space X, at each point m ∈ M the tangent space Tm M can be thought of as a subspace of X and, as such, it inherits a norm. If γ : [a, b] → M is a differentiable curve on a continuous Finsler manifold (M, F ), we define Z b F (γ(t), γ(t)) ˙ dt. length(γ) := a
This definition can be extended in the obvious way to piecewisedifferentiable curves. If x and y are two points in M , we define the distance between x and y as the infimum of the lengths of all piecewisedifferentiable curves that join them. Thus, continuous Finsler manifolds are metric spaces and metric techniques can be used to study them. Each definition of volume on normed spaces gives a definition of volume on continuous Finsler manifolds: if we are given a volume definition µ and an ndimensional continuous Finsler manifold (M, F ), then the map that assigns to every point m the norm µTm M on Λn Tm M is a volume density on M . Notice that, in particular, a definition of volume on normed spaces immediately yields a way to measure the volumes of submanifolds of a normed space X because, as remarked above, the tangent space Tm M of such a submanifold can be regarded as a subspace of X and so inherits both a norm and a volume. This will be studied from an extrinsic viewpoint and in much more detail in the next section. Exercise 3.22. Show that if a definition of volume on normed spaces is used to define a volume on Finsler manifolds, it satisfies the following two properties: (1) If the Finsler manifold is Riemannian, its volume is the standard Riemannian volume; (2) If ϕ : M → N is a short map (i.e., does not increase distances) between two Finsler manifolds of the same dimension, then ϕ does not increase volumes. Extending our four volume definitions from normed spaces to continuous Finsler manifolds, we may speak of the Busemann, Holmes–Thompson, mass, and mass∗ definition of volumes on continuous Finsler manifolds. To end the section we relate the Busemann and Holmes–Thompson definitions with wellknown geometric constructions. As was previously remarked, Finsler manifolds are metric spaces and, as such, we can define their volume as their Hausdorff measure: if (M, F ) is an ndimensional Finsler manifold and r > 0, we cover M by a family of metric balls of radius at most r, B(m1 , r1 ), B(m2 , r2 ), . . ., and consider the quantity
VOLUMES ON NORMED AND FINSLER SPACES
17
εn (r1n + r2n + · · ·). We now take the infimum of this quantity over all possible covering families and take the limit as r tends to zero. The resulting number is the ndimensional Hausdorff measure of (M, F ). Theorem 3.23 [Busemann 1947]. The Busemann volume of a continuous Finsler manifold is equal to its Hausdorff measure. To explain the second construction, we need to recall some standard facts about the geometry of cotangent bundles. If π : T ∗ M → M is the canonical projection, we define the canonical 1form α on T ∗ M by the formula α(v p ) := p(π∗ (v p )). Pn In standard coordinates (x1 , . . . , xn , p1 , . . . , pn ), α := i=1 pi dqi . The canonical symplectic form on T ∗ M is defined as ω := −dα and the Liouville volume is defined by Ω := ω n /n!. If (M, F ) is a continuous Finsler manifold, each tangent space T m M carries ∗ the norm F (m, ·) and, hence, each cotangent space T m M carries the dual norm ∗ ∗ ∗ F (m, ·). Let us denote the unit ball in Tm M by Bm and define the unit codisc bundle of M as the set [ ∗ B ∗ (M ) := Bm ⊂ T ∗ M. m∈M
Proposition 3.24. The Holmes–Thompson volume of an ndimensional , continuous Finsler manifold is equal to the Liouville volume of its unit codisc bundle divided by the volume of the ndimensional Euclidean unit ball . Proof. It suffices to verify the result on normed spaces where it easily follows from the definitions. Theorem 3.25 [Duran 1998]. If M is a Finsler manifold , then the Holmes– Thompson volume of M is less than or equal to its Hausdorff measure with equality if and only if M is Riemannian. Proof. By the Blaschke–Santal´ o inequality, at each point m ∈ M we have that b µht Tm M ≤ µTm M with equality if and only if Tm M is Euclidean. The result now follows immediately from Theorems 3.23 and 3.24 .
4. kVolume Densities The theory of volumes and areas on Euclidean and Riemannian spaces is based on the fact that a Euclidean structure on a vector space induces natural Euclidean structures on its exterior powers: if x 1 , . . . , xn is an orthonormal basis of a Euclidean space X, then the vectors xi1 ∧ xi2 ∧ · · · ∧ xik ,
1 ≤ i1 < i2 < · · · < ik ≤ n,
18
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
form an orthonormal basis of Λk X. If we want to know the area of the parallelogram formed by the vectors x and y, we need to compute the norm of x ∧ y in Λ2 X. In normed and Finsler spaces these simple algebraic constructions, which should be seen as functors from the category of Euclidean spaces onto itself, cannot be reproduced and we need to understand their geometry to see how they may be naturally extended to these spaces. The first important remark is that in order to compute kdimensional volumes (kvolumes from now on), we do not need to define a norm on all of Λ k X. It suffices to define the magnitude of vectors of the form v 1 ∧ v 2 ∧ · · · ∧ v k , where v 1 , . . . , v k are vectors in X. In this paper we shall refer to these kvectors as simple and denote the set of all simple kvectors in X by Λ ks X. Note that for k = 1, n − 1 every kvector is simple, which makes the study of (n−1)volume in ndimensional normed spaces a richer and more approachable subject than the study of volumes in higher codimension. Indeed, when k 6= 1, n − 1, Λ ks X is not a vector subspace of Λk X, but just an algebraic cone. Exercise 4.1. Let Q : Λ2 R 4 → Λ4 R 4 be the quadratic form defined by Q(a) = a ∧ a. Show that Λ2s R 4 is the quadric Q = 0 and use this to prove that the intersection of Λ2s R 4 with the (Euclidean) unit sphere in Λ2 R 4 is a product of two 2dimensional spheres. In general, the intersection of Λks R n with the Euclidean unit sphere in Λk R n is n the Pl¨ ucker embedding of the Grassmannian of oriented kplanes in R n , G+ k (R ). Let us recall that this Grassmannian is a smooth manifold of dimension k(n−k). Following our first remark, we see that in order to compute kvolumes in a normed space X, we need to define a “norm” on the cone of simple kvectors of X. The fact that Λks X is not a vector space complicates matters since it is not clear how to write the triangle inequality, and, even if an apparent analogue could be found, it would have to be justified in terms of its geometric significance. Nevertheless, the homogeneity and positivity of a norm are easy to generalize: Definition 4.2. A kdensity on a vector space X is a continuous function φ : Λks X −→ R that is homogeneous of degree one (i.e., φ(λa) = λφ(a)). A kdensity φ is said to be a kvolume density if φ(a) ≥ 0 with equality if and only if a = 0. 4.1. Examples of kvolume densities. In the previous chapter we studied four natural volume definitions on normed spaces. Each one of these definitions yields natural constructions of kvolume densities on the spaces of simple kvectors of a normed space X. Given a volume definition µ and a kvector a in a kdimensional normed space Y , we can compute µ(a). To be perfectly rigorous, we should include the normed space as a variable in µ, for example, by writing µY (a). If a is a simple kvector in an ndimensional normed space X, then we may consider it as a kvector on
VOLUMES ON NORMED AND FINSLER SPACES
19
the kdimensional normed space “spanned by a”, hai := {x ∈ X : a ∧ x = 0} ⊂ X, (provided with the induced norm), and compute µhai (a). Thus, once we have chosen a way to define volume in all finitedimensional normed spaces, we have a way to associate to each norm on an ndimensional vector space X a family of kvolume densities, with 1 ≤ k ≤ n. The Busemann kvolume densities. Let X be a normed space of dimension n with unit ball B, and let k be a positive integer less than n. The Busemann kvolume density on X is defined by the formula µb (a) :=
εk . vol(B ∩ hai; a)
The Holmes–Thompson kvolume densities. Let X be a normed space of dimension n, and let k be a positive integer less than n. If a is a simple kvector spanning the kdimensional subspace hai, we consider the inclusion of hai into ∗ X and the dual projection π : X ∗ → hai . Regarding a as a volume form on ∗ hai we define Z −1 ht µ (a) := εk a. ∗ ) π(BX
The mass kvolume densities. Let (X, k · k) be a normed space of dimension n, and let k be a positive integer less than n. The mass kvolume density on X is defined by the formula µm (a) := inf
k Y
i=1
kxi k : x1 ∧ x2 ∧ · · · ∧ xk = a .
The mass∗ kvolume densities. According to the characterization of mass∗ given in Exercise 3.9, we may describe the mass∗ kvolume densities as follows: Let X be a normed space and let W ⊂ X be a kdimensional subspace. If a is a simple kvector on X, we define µm∗ (a) as the supremum of the numbers ∗ ξ1 ∧ ξ 2 ∧ · · · ∧ ξ k (a), where ξ 1 , . . . , ξ k ranges over all short bases of hai . However, there is a simpler description: Exercise 4.3. Using the Hahn–Banach theorem and the notation above, show that µm∗ (a) is the supremum of the numbers ξ 1 ∧ ξ 2 ∧ · · · ∧ ξ k (a), where ∗ ξ 1 , . . . , ξ k ∈ BX . In the study of volumes and areas on Finsler manifolds, we shall also need to work with kdensities and smooth kdensities on manifolds. For this purpose we introduce the bundle of simple tangent kvectors on a manifold M , Λ ks T M . This is a subbundle of algebraic cones of the vector bundle Λ k T M , and if we omit the zero section it is a smooth manifold.
20
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Definition 4.4. A kdensity φ (resp. kvolume density) on a manifold M is a continuous function φ : Λks T M → R such that at each point m, the restriction of φ to Λks Tm M is a kdensity (resp. kvolume density). If the function φ is smooth outside the zero section, we shall say that the density is smooth. Since every tangent space of a Finsler manifold (M, F ) is a normed space, we may define the Busemann, Holmes–Thompson, mass, and mass∗ kvolume densities on (M, F ) by assigning to each tangent space its respective kdensity. It is easy to show that if F is a smooth Finsler metric, then the Busemann and Holmes– Thompson kvolume densities are smooth. This is probably not the case with mass and mass∗, but we have no explicit examples to illustrate this. Just like differential kforms, kdensities can be pulled back: if f : N → M is a smooth map and φ is a kdensity on M , then f ∗ φ(v 1 ∧ v 2 ∧ · · · ∧ v k ) = φ Df (v 1 ) ∧ Df (v 2 ) ∧ · · · ∧ Df (v k ) . Remark that if f : N → M is an immersion and φ is a kvolume density on M , then f ∗ φ is a kvolume density on N . Also like differential forms, kdensities can be integrated over kdimensional submanifolds: if N ⊂ M is a kdimensional submanifold of M and i : N → M is the inclusion map, then i∗ φ is a volume density on N , and its integral over N was defined in Section 3. This integral is independent of the parameterization and orientation of N . In the same way, we may define the integral of a kdensity over a kchain. For the rest of the chapter, we associate to a given kvolume density φ on a vector space X the functional N 7−→
Z
φ, N
and investigate the relationship between the behavior of the functional and certain convexity properties of φ. The easiest case is when φ is an (n−1)volume density in an ndimensional vector space. 4.2. Convexity of (n−1)volume densities. This case is special because every (n−1)vector in an ndimensional vector space, X, is simple and we may impose the condition that an (n−1)volume density be a norm in Λ n−1 X. This is, for example, satisfied by (n−1)volume densities for the Busemann, Holmes– Thompson, and mass∗ definitions of volume. Nevertheless, it remains to see why such a condition is desirable. The next result is the first of four characterizations of norms on Λ n−1 X. Theorem 4.5. Let φ be an (n−1)volume density on an ndimensional vector space X. The following conditions on φ are equivalent:
VOLUMES ON NORMED AND FINSLER SPACES
21
• φ is a norm; • If P ⊂ X is a closed (n−1)dimensional polyhedron in X, then the area of any one of its facets is less than or equal to the sum of the areas of the remaining facets. Before proving this theorem, we need to introduce a classical construction that associates to any kdimensional polyhedral surface on X a set of simple kvectors. This set will be called the Gaussian image of the polyhedron (see also [Burago and Ivanov 2002], where the almost identical notion of Gaussian measure is used). If P ⊂ X is a polyhedron with facets F1 , . . . , Fm we associate to each facet Fi the simple kvector ai such that hai i is parallel to Fi and such that vol(Fi ; ai ) = 1. The Gaussian image of P is the set {a1 , . . . , am } ⊂ Λks X. If φ is a kvolume density in X, the kvolume of P (with respect to φ) is just φ(a 1 ) + · · · + φ(am ). Exercise 4.6. Show that if {a1 , . . . , am } ⊂ Λks X is the Gaussian image of a closed polyhedron in X, then a1 + · · · + am = 0. In general, the condition that the sum of a set of simple kvectors be zero, does not imply that it is the Gaussian image of a closed kdimensional polyhedron in X. However, in codimension one we have the following celebrated theorem of Minkowski. Theorem 4.7 (Minkowski). A set of (n−1)vectors of an ndimensional vector space X is the Gaussian image of a closed , convex polyhedron if and only if the (n−1)vectors span Λn−1 X and their sum equals zero. To prove Theorem 4.5, we shall need an easy particular case of Minkowski’s result: Exercise 4.8. Let X be an ndimensional vector space and let a 1 , . . . , an be a basis of Λn−1 X. Show that there exists a simplex in X whose Gaussian image is the set {a1 , . . . , an , −(a1 + · · · + an )}. Proof of Theorem 4.5. Assume that φ is a norm, and let P ⊂ X be an (n−1)dimensional closed polyhedron with Gaussian image {a 0 , a1 , . . . , am }. Since the sum of the ai ’s is zero, we may use the triangle inequality to write φ(a0 ) = φ(a1 + · · · + am ) ≤ φ(a1 ) + · · · + φ(am ). In other words, the area of the facet corresponding to a 0 is less than or equal to the sum of the areas of the remaining facets. To prove the converse, we take any two (n−1)vectors a 1 and a2 , which we assume to be linearly independent, and use them as part of a basis a 1 , . . . , an of Λn−1 X. By Exercise 4.8, the set {a1 , . . . , an , −(a1 + · · · + an )} is the Gaussian image of a simplex in X. Then, by assumption, φ(a1 + · · · + an ) ≤ φ(a1 ) + · · · + φ(an ).
22
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
By letting a3 , . . . , an tend to zero in the above inequality we obtain the triangle inequality φ(a1 + a2 ) ≤ φ(a1 ) + φ(a2 ), and, therefore, φ is a norm. Exercise 4.9. Consider the tetrahedron in the normed space ` 3∞ with vertices (0, 0, 0), (−1, 1, 1), (1, −1, 1), (1, 1, −1), and show that the mass of the facet opposite the origin is greater than the sum of the masses of the three other facets. Hint. Use the definition of the mass 2volume density in terms of minimal circumscribed parallelograms. By Theorem 4.5, the previous exercise shows that the mass (n−1)volume density of a normed space X is not necessarily a norm in Λn−1 X. As we shall see in the rest of this chapter, this is a good reason to disqualify mass as a satisfactory definition of volume on normed spaces. Our second characterization of norms in Λn−1 X is another variation on the theme of flats minimize. Theorem 4.10. Let φ be an (n−1)volume density on an ndimensional vector space X. The following conditions on φ are equivalent: • φ is a norm; • Whenever C and C 0 are (n−1)chains with real coefficients such that ∂C = ∂C 0 and the image of C is contained in a hyperplane, then the area of C is less than or equal to the area of C 0 . In order to prove this theorem we need to introduce the concept of calibration formalized by Harvey and Lawson [1982]. Definition 4.11. A closed kform ω is said to calibrate a kdensity φ if for all simple kvectors a in T M we have that ω(a) ≤ φ(a) and equality is attained on a nonempty subset of Λks T M . The homogeneity of ω and φ allows us to consider the set where they coincide as a subset E of the bundle of oriented kdimensional subspaces of T M , G + k (T M ). Proposition 4.12 [Harvey and Lawson 1982]. Let φ be a kvolume density on a manifold M , let ω be a closed kform on M that calibrates φ and let E ⊂ G+ k (T M ) be the set where φ and ω coincide. If N ⊂ M is a kdimensional oriented submanifold all of whose tangent planes belong to E, and N 0 is another submanifold of M homologous to N , then Z Z φ≤ φ. N
N0
Proof. Using that φ = ω on the tangent spaces of N and Stokes’ formula, we have Z Z Z Z φ= ω= ω≤ φ. N
N
N0
N0
VOLUMES ON NORMED AND FINSLER SPACES
23
Proof of Theorem 4.10. Assume that φ is a norm, let C and C 0 be as in the statement of the theorem, and let a be an (n−1)vector on X such that φ(a) = 1 and the subspace hai is parallel to the hyperplane containing the image of C. Since the unit sphere in (Λn−1 X, φ) is a convex hypersurface, it has a supporting hyperplane that touches it at a. This hyperplane can be given as the set ω = 1, where ω is a constantcoefficient (n−1)form on X. Since the unit sphere lies in the halfspace ω ≤ 1, we have ω ≤ φ and, thus, ω calibrates φ. Using that ω = φ on C, that dω = 0, and that C + (−C 0 ) is a closed chain homologous to zero, we have Z Z Z Z φ= ω= ω≤ φ. C
C
C0
C0
To prove the converse, note that the second condition in the theorem immediately implies that the (n−1)volume of the facet of a closed polyhedron is less than or equal to the sum of the (n−1)volumes of the remaining facets. By Theorem 4.5, this implies that φ is a norm. In Euclidean geometry, the orthogonal projection onto a kdimensional subspace is areadecreasing. This can be generalized as follows: Theorem 4.13. Let φ be an (n−1)volume density on an ndimensional vector space X. The following conditions on φ are equivalent: • φ is a norm; • For every (n−1)dimensional subspace W ⊂ X there is a φdecreasing linear projection PW : X → W . The proof of this theorem rests on a simple lemma in multilinear algebra. Lemma 4.14. Let X be an ndimensional vector space and let W ⊂ X be a kdimensional subspace. If w 1 , . . . , w k is a basis of W and ω ∈ Λk X ∗ is such that ω(w 1 ∧ w 2 ∧ · · · ∧ w k ) = 1, then the linear map P x :=
k X
(−1)i ω(x ∧ w 1 ∧ · · · ∧ wˆi ∧ · · · ∧ w k )w i
i=1
is a projector with range W . Moreover , ω is simple if and only if for any vectors x1 , . . . , xk ∈ X, P x1 ∧ P x2 ∧ · · · ∧ P xk = ω(x1 ∧ x2 ∧ · · · ∧ xk ) w 1 ∧ w 2 ∧ · · · ∧ w k . The proof of the lemma is left as an exercise to the reader. Proof of Theorem 4.13. Assume that φ is a norm in Λn−1 X and let W ⊂ X be an (n−1)dimensional subspace. Choose a basis of W , w 1 , . . . , w n−1 , such that φ(w 1 ∧ w 2 ∧ · · · ∧ w n−1 ) = 1 and consider the support hyperplane of the unit sphere of (Λn−1 X, φ) at the point w 1 ∧ w 2 ∧ · · · ∧ w n−1 . This hyperplane is
24
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
given by an equation of the form ω = 1, where ω is an (n−1)form with constant coefficients. In other words, ω ∈ Λn−1 X ∗ . We claim that the linear projection P x :=
n−1 X
(−1)i ω(x ∧ w 1 ∧ · · · ∧ wˆi ∧ · · · ∧ w n−1 )w i
i=1
is φdecreasing. Indeed, since ω is an (n−1)form on an ndimensional space, it is simple. Using the second part of Lemma 4.14, we have, for any (n−1)vector a := x1 ∧ x2 ∧ · · · ∧ xn−1 , φ(P x1 ∧ P x2 ∧ · · · ∧ P xn−1 ) = ω(a)φ(w 1 ∧ w 2 ∧ · · · ∧ w n−1 ) = ω(a) ≤ φ(a). To prove the converse, we note that the existence of a φdecreasing linear projection onto any given hyperplane implies that the (n−1)volume of the facet of any closed (n−1)dimensional polyhedron is less than or equal to the sum of the areas of the remaining facets. The argument is quite simple: if the closed polyhedron has facets F0 , F1 , . . . , Fm , we use the φdecreasing projection P to project the whole polyhedron onto the hyperplane containing F 0 . Note that P (F1 ) ∪ · · · ∪ P (Fm ) contains P (F0 ) = F0 and, therefore, the sum of the (n−1)volumes of the P (Fi ), 1 ≤ i ≤ m, is greater than or equal to the (n−1)volume of F0 . Since P is φdecreasing, this gives us that the sum of the (n−1)volumes of the Fi , 1 ≤ i ≤ m, is greater than or equal to the (n−1)volume of F 0 . We now state the fourth and last of our characterizations of norms on the space of (n−1)vectors in an ndimensional normed space. Theorem 4.15. Let φ be an (n−1)volume density on an ndimensional vector space X. The following conditions on φ are equivalent: • φ is a norm; • If K ⊂ K 0 are two nested convex bodies in X, then the area of ∂K is less than or equal to the area of ∂K 0 with equality if and only if K equals K 0 . The proof of this theorem is a simple consequence of the relation between norms in the space of (n−1)vectors and the theory of mixed volumes that will be developed in Section 6. At the beginning of this section we mentioned that for any ndimensional normed space X the Busemann, Holmes–Thompson, and mass∗ (n−1)volume densities on X are norms in Λn−1 X. For the Busemann definition this is a celebrated theorem of Busemann [1949a]. For the mass∗ definition this result is due to Benson [1962]. We shall follow [Gromov 1983] and give a proof of a much stronger result later in this section. For the Holmes–Thompson definition, the result — under a different formulation — goes back to Minkowski. Theorem 4.16 (Minkowski). The Holmes–Thompson (n−1)volume density of an ndimensional normed space X is itself a norm in Λn−1 X.
VOLUMES ON NORMED AND FINSLER SPACES
25
In order to prove the convexity of the Holmes–Thompson (n−1)volume density, we shall first give an integral representation for it. This representation depends, in turn, on two classical constructions: the Gauss map and the surfacearea measure. Our approach follows [Fernandes 2002]. Let X be an ndimensional vector space and let φ be an (n−1)volume density on X. If N ⊂ X is an oriented hypersurface and n ∈ N , we define G φ (n) as the unique (n−1)vector in Λn−1 Tn N ⊂ Λn−1 X that is positively oriented and satisfies φ(Gφ (n)) = 1. Notice that when N is a strictly convex hypersurface, the Gauss map Gφ : N −→ Σ := {a ∈ Λn−1 X : φ(a) = 1} is a diffeomorphism. In this case, we define the surfacearea measure of N as the (n−1)volume density dSN := G−1∗ φ φ on Σ. Lemma 4.17. Let π : X → Y be a surjective linear map between an ndimensional vector space X and an (n−1)dimensional vector space Y , and let φ be an (n−1)volume density on X with unit sphere Σ ⊂ Λn−1 X. If N ⊂ X is a smooth, strictly convex hypersurface and ω is a volume form on Y , then Z Z 1 ω = π ∗ ω(a) dSN . 2 a∈Σ π(N ) Proof. By the definition of the Gauss map, if n ∈ N and x1 ∧ x2 ∧ · · · ∧ xn−1 ∈ Λn−1 Tn N , x1 ∧ x2 ∧ · · · ∧ xn−1 = φ(x1 ∧ x2 ∧ · · · ∧ xn−1 )Gφ (n). Therefore, π ∗ ω(x1 ∧ x2 ∧ · · · ∧ xn−1 ) = π ∗ ω(Gφ (n))φ(x1 ∧ x2 ∧ · · · ∧ xn−1 ). Then Z Z Z 1 1 π ∗ ω = π ∗ ω(Gφ (n))φ ω = 2 N 2 n∈N π(N ) Z Z 1 1 ∗ G−1∗ π ω(G (n))φ = π ∗ ω(a)G−1∗ = φ φ φ 2 Σ φ 2 a∈Σ Z 1 = π ∗ ω(a) dSN . 2 a∈Σ Proof of Theorem 4.16. By a standard approximation argument, it suffices to consider the case where the dual unit sphere ∂B ∗ ⊂ X ∗ is smooth and strictly convex. Applying the previous lemma to the surface N = ∂B ∗ and to an arbitrary (n−1)volume density on X ∗ , we have µht (a) = ε−1 n−1
Z
a = ε−1 n−1 π(B ∗ )
Z
ξ∈Σ
a(ξ) dS∂B ∗ .
26
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Since the surfacearea measure dS∂B ∗ is nonnegative, Z µht (a + b) = ε−1 a(ξ) + b(ξ) dS∂B ∗ n−1 ξ∈Σ Z Z −1 ∗ + ε ≤ ε−1 a(ξ) dS ∂B n−1 n−1 ξ∈Σ
= µht (a) + µht (b).
b(ξ) dS∂B ∗ ξ∈Σ
Exercise 4.18. Let X be an ndimensional vector space and let φ be an (n−1)volume density on X. Show that if φ is a norm, then compact hypersurfaces cannot by minimal. 4.3. Convexity properties of kvolume densities. We now pass to the more delicate subject that Busemann, Ewald, and Shephard studied extensively under the heading of convexity on Grassmannians. Most of what follows can be found in their papers, “Convex bodies and convexity on Grassmannian cones” I–XI, but we have tried to make the language and proofs more accessible. We shall see that there are several notions and degrees of convexity for kvolume densities. These are closely related to the concept of ellipticity in geometric measure theory and, historically, to the generalization of the Legendre condition for variational problems. Weakly convex kdensities. Let X be an ndimensional vector space and let Λks X, 1 ≤ k ≤ n − 1, be the cone of simple kvectors on X. If Y is a (k + 1)dimensional subspace of X, then the subspace Λk Y ⊂ Λk X lies inside Λks X. This motivates a definition: Definition 4.19. A kvolume density φ on an ndimensional vector space X, n > k, is said to be weakly convex if for any linear subspace Y of dimension k + 1, the restriction of φ to the linear space Λk Y is a norm. From the previous section, we know that the kvolume densities of any normed space for the Busemann, Holmes–Thompson, or mass∗ definitions of volume are weakly convex. Exercise 4.20. Show that a kvolume density in a vector space X is weakly convex if for every (k + 1)dimensional simplex in X the area of any one facet is less than or equal to the sum of the areas of the remaining facets. Extendibly convex kvolume densities Definition 4.21. A kvolume density φ on an ndimensional vector space X, n > k, is said to be extendibly convex if it is the restriction of a norm on Λ k X to the cone of simple kvectors in X. Equivalently, φ is extendibly convex if and only if there is a support hyperplane for the unit sphere S := {a ∈ Λks X : φ(a) = 1}
VOLUMES ON NORMED AND FINSLER SPACES
27
passing through any of its points. Theorem 4.22. If φ is an extendibly convex kvolume density on a vector space X, then any kchain with real coefficients whose image is contained in a kdimensional flat is φminimizing. The proof — by the method of calibrations — is nearly identical to the proof of Theorem 4.10 and is left as an exercise for the reader. Notice that a corollary to Theorem 4.22 is that if P ⊂ X is a closed kdimensional polyhedron, the area of any of its facets is less than or equal to the sum of the areas of the remaining facets. The problem of determining whether the Busemann kvolume densities are extendibly convex was posed by Busemann in several of his papers as a major problem in convexity. So far, there are no results in this direction. Problem. Is the Busemann 2volume density of a 4dimensional normed space extendibly convex ? In the case of the Holmes–Thompson definition, Busemann, Ewald, and Shephard have given explicit examples of norms for which the kvolume densities, 1 < k < n − 1, are not extendibly convex (see [Busemann et al. 1963]). A simpler example has been given recently by Burago and Ivanov: Theorem 4.23 [Burago and Ivanov 2002]. Consider the norm k · k on R 4 whose dual unit ball in R 4∗ is the convex hull of the curve γ(t) := (sin t, cos t, sin 3t, cos 3t),
0 ≤ t ≤ 2π.
4
The Holmes–Thompson 2volume density for (R , k · k) is not extendibly convex . Despite these examples, in many important cases the Holmes–Thompson kvolume densities are extendibly convex. Theorem 4.24. The Holmes–Thompson kvolume densities of a hypermetric normed space are extendibly convex . In order to prove this result, we shall derive a formula for the Holmes–Thompson kvolume densities of a Minkowski space in terms of the Fourier transform of its norm. In a somewhat different guise, this formula was first obtained by W. Weil ´ [1979]. In the present form it was rediscovered by Alvarez and Fernandes in [1999], where it was shown to follow from the Crofton formula for Minkowski spaces. Let φ be a smooth, even, homogeneous function of degree one on an ndimensional vector space X, let e1 , . . . , en be a basis of X, and let ξ 1 , . . . , ξ n be the dual basis in X ∗ . Using the basis e1 , . . . , en and its dual to identify both X and X ∗ with R n , we can compute the (distributional) Fourier transform of φ, Z ˆ φ(ξ) := eiξ·x φ(x) dx. Rn
28
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
The form φˆ dξ1 ∧ · · · ∧ dξn does not depend on the choice of basis in X. Up to a constant factor, we define the form φˇ as the contraction of this nform with the Euler vector field, XE (ξ) = ξ, in X ∗ : φˇ :=
−1 φˆ dξ1 ∧ · · · ∧ dξn cXE . 4(2π)n−1
It is known (see [H¨ ormander 1983, pages 167–168]) that φˆ is smooth on X ∗ \ 0 and homogeneous of degree −n − 1; therefore φˇ is a smooth differential form on X ∗ \ 0 that is homogeneous of degree −1. Denoting by φˇk the product form in the product space (X ∗ \ 0)k , we have the following result: Theorem 4.25. Let (X, φ) be an ndimensional Minkowski space. For any simple kvector a on X, 1 ≤ k < n, we have Z 1 µht (a) = ξ ∧ · · · ∧ ξ k · a φˇk , εk (ξ1 ,...ξk )∈S ∗k 1 where S ∗ is any closed hypersurface in X ∗ \ 0 that is starshaped with respect to the origin. Notice that this formula allows us to extend the definition of the Holmes– Thompson kvolume density of any Minkowski space to all of Λ k X. It remains to see when this extension is a norm. Proof of Theorem 4.24. It is enough to prove convexity in the case the hypermetric space (X, φ) is also a Minkowski space. This allows us to use the integral representation given above. Since X is hypermetric, Theorem 2.5 tells us that the form φˆ dξ1 ∧ · · · ∧ dξn is a volume form, and, therefore, the restriction of φˇk to the manifold S ∗k defines a nonnegative measure. Then for any two kvectors a and b we have Z ht µ (a + b) = ξ 1 ∧ · · · ∧ ξ k · (a + b) φˇk S ∗k Z Z ≤ ξ ∧ · · · ∧ ξ · a φˇk + ξ ∧ · · · ∧ ξ · b φˇk 1
S ∗k ht
= µ (a) + µht (b).
1
k
k
S ∗k
Totally convex kdensities Definition 4.26. A kdensity φ on an ndimensional vector space X, n > k, is said to be totally convex if through every point of the unit sphere of Λ k X there passes a supporting hyperplane of the form ξ = 1 with ξ a simple kvector in Λk X ∗ . Total convexity implies extendible convexity and, in turn, weak convexity. The following result, stated in [Busemann 1961] gives an important characterization of totally convex kdensities in terms of what Gromov [1983] calls the compressing property.
VOLUMES ON NORMED AND FINSLER SPACES
29
Theorem 4.27. A kdensity φ on an ndimensional vector space X is totally convex if and only if for every kdimensional linear subspace there exists a φdecreasing linear projection onto that subspace. The proof, using Lemma 4.14, is nearly identical to the proof of Theorem 4.13. Of all the four volume definitions we have studied, mass∗ has by far the strongest convexity property: Theorem 4.28. The mass∗ kvolume densities of an ndimensional normed space X, 1 ≤ k ≤ n − 1, are totally convex . Proof. By Theorem 4.27, it is enough to show that given any kdimensional subspace W , there exists a linear projection P : X → W that is mass∗decreasing. Choose a basis ξ 1 , . . . , ξ k of W ∗ which satisfies two properties: (1) It is short (i.e., ξ i (x) ≤ 1 for all x ∈ B ∩ W ); (2) The integral of the volume density ξ 1 ∧ ξ2 ∧ · · · ∧ ξ k  over B ∩ W is maximal among all short bases. Notice that for any basis w 1 , . . . , w k of W , we have that µm∗ (w 1 ∧ w 2 ∧ · · · ∧ w k ) = ξ 1 ∧ ξ 2 ∧ · · · ∧ ξ k (w 1 ∧ w 2 ∧ · · · ∧ w k ), and that if w 1 , . . . , w k is dual to ξ 1 , . . . , ξ k , then µm∗ (w 1 ∧ w 2 ∧ · · · ∧ w k ) = 1. By the Hahn–Banach theorem, there exist covectors ξˆ1 , . . . , ξˆk ∈ X ∗ such that (1) ξˆi (x) ≤ 1 for all x ∈ B and for all i, 1 ≤ i ≤ k; (2) the restriction of ξˆi to W equals ξ i for all i, 1 ≤ i ≤ k. We may now define the projection P : X → W by the formula P (x) :=
k X
ξˆi (x)w i ,
i=1
and show that it is µm∗ decreasing. Indeed, if a = v 1 ∧ v 2 ∧ · · · ∧ v k is a simple kvector in X, P (v 1 ) ∧ P (v 2 ) ∧ · · · ∧ P (v k ) = ˆ ξ1 ∧ ˆ ξ2 ∧ · · · ∧ ˆ ξ k (a)w 1 ∧ w 2 ∧ · · · ∧ w k , and therefore µm∗ (P (v 1 ) ∧ P (v 2 ) ∧ · · · ∧ P (v k )) = ˆ ξ1 ∧ ˆ ξ2 ∧ · · · ∧ ˆ ξ k (a). ∗ Since the restriction of ξˆi , 1 ≤ i ≤ k, to hai form a short basis of hai , we have
µm∗ (P (v 1 ) ∧ P (v 2 ) ∧ · · · ∧ P (v k )) = ˆ ξ1 ∧ ˆ ξ2 ∧ · · · ∧ ˆ ξ k (a) ≤ µm∗ (a).
We end the section with an exercise and an open problem: Exercise 4.29. Show that the sum of two totally convex 2volume densities in R 4 is not necessarily totally convex. On the other had, show that the maximum of two totally convex kvolume densities is a totally convex kvolume density.
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
30
Problem. For what (hypermetric) normed spaces are the Holmes–Thompson kvolume densities totally convex ?
5. Length and Area in TwoDimensional Normed Spaces Before going into the rich and beautiful theory of (hypersurface) area on finitedimensional normed spaces, we shall sharpen our intuition by carefully considering the case of twodimensional normed spaces. This case is fundamentally simpler because the notion of hypersurface area coincides with that of length and is thus independent of our volume definition. Nevertheless, we shall see that the theory of length and area on twodimensional normed spaces is far from trivial and provides a platform from which to jump to higher dimensions. We start with two theorems that involve solely the notion of length: Theorem 5.1 [Gol¸ab 1932]. The perimeter of the unit circle of a twodimensional normed space is between six and eight. Moreover , the length is equal to six if and only if the unit ball is an affine regular hexagon and is equal to eight if and only if it is a parallelogram. Full proofs can be found in [Sch¨ affer 1967] and [Thompson 1996]. We stress that the length of the unit circle is measured with the definition of length given by the norm: If γ : [a, b] → X is a continuous curve on the normed space (X, k · k), the length of γ is defined as the supremum of the quantities n−1 X
kγ(ti+1 ) − γ(ti )k
i=0
taken over all partitions a = t0 < t1 < · · · < tn = b of the interval [a, b]. Notice that if γ is differentiable, we can also compute its length by the integral Z b `(γ) = kγ(t)k ˙ dt. a
It is convenient to denote the length of a curve γ on the normed space X with unit disc B by `B (γ). Note that `B (∂B) is an affine invariant of the convex body B. Theorem 5.2 [Sch¨ affer 1973]. If B and D are unit balls of two norms in a twodimensional space X and if B ∗ and D∗ are the dual balls in X ∗ then `D (∂B) = `B ∗ (∂D∗ ). In particular , `B (∂B) = `B ∗ (∂B ∗ ). A complete proof is available in [Thompson 1996]. For those who like simply stated open problems, we pass on the following question of Sch¨ affer (private communication):
VOLUMES ON NORMED AND FINSLER SPACES
31
Problem. Given an arbitrary convex body B ⊂ R 3 that is symmetric with respect to the origin, does there always exist a plane Π passing through the origin and for which `Π∩B (∂(Π ∩ B)) is less than or equal to 2π? Next we discuss the relation between length and area on twodimensional normed spaces. The first important question that arises is the isoperimetric problem: Of all convex bodies in a twodimensional normed space X with a given perimeter find those that enclose the largest area. The solution of this problem passes through the representation of the length as a mixed volume (in this case a mixed area). This permits the use of Brunn– Minkowski theory to solve the isoperimetric problem and to also give further properties of the length functional. The reader is referred to [Schneider 1993] for a complete discussion of the theory, but our needs can be met in just a few paragraphs. Let X be an ndimensional vector space and let λ be a Lebesgue measure on X. If K and L are two subsets of X, the Minkowski sum of K and L is the set K + L := {x + y ∈ X : x ∈ K, y ∈ L}. If L is the unit ball of a norm in X, we may think of K + L as the set of all points in X whose distance from K is less than or equal to one. In other words, the tube of radius one about the set K. The mixed volume V (K[n − 1], L) of two closed, bounded convex sets K and L in X is defined as a “directional derivative” of the Lebesgue measure: V (K[n − 1], L) =
λ(K + tL) − λ(K) 1 lim . n t→+0 t
In the twodimensional, case V (K, L) := V (K[1], L) is linear and monotonic in each variable. The key result in the solution of the isoperimetric problem in normed spaces is the Minkowski mixed volume inequality: V (K[n − 1], L) ≥ λ(K)n−1 λ(L). Moreover, if K and L are convex bodies, then equality holds if and only if K is obtained from L by translation and dilation. Back to the twodimensional case, if we’re given a centered convex body B, we may define the magnitude of a vector x in two different ways: (1) Take B to be the unit ball of a norm k · k on X and set the magnitude of x to be kxk. (2) Let [x] ⊂ X denote the line segment from the origin to x and define the magnitude of x as V ([x], B). Exercise 5.3. Show that for any convex body B that is symmetric with respect to the origin, the map x 7→ V ([x], B) is a norm, but that in general its unit disc is different from B.
32
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
The first step in solving the isoperimetric problem in a normed space X is to find a centrally symmetric convex body I such that kxk = V ([x], I), for all x ∈ X. Of course, I will also depend on the choice of Lebesgue measure λ used to define the mixed volume. However, given a volume definition the body I will be defined intrinsically in terms of the norm. The construction of I is extremely simple: Let B be the unit ball of X and let Ω be the volume form on X that satisfies Ω(x ∧ y) = λ(x ∧ y) for all positive bases x, y of X (we are forced to take an orientation of X at this point, but the result will not depend on the choice). If iΩ : X −→ X ∗ is defined by iΩ (v)(w) := Ω(v ∧ w), the set I is given by (iΩ B)∗ . Summarizing: Proposition 5.4. Let X be a twodimensional normed space with unit ball B and volume form Ω. If I denotes the body (iΩ B)∗ , then kxk = V ([x], I) for all vectors x ∈ X. The proof will be postponed to the next section where we will treat the ndimensional version of the proposition. Notice that if the orientation of X is changed, the form Ω changes sign, but the symmetry of the unit disc B implies that the body I stays the same. Exercise 5.5. Show that if K is a convex body in X, its perimeter equals 2V (K, I) and that, in particular, the perimeter of I is twice its area. Hint: Try first with bodies whose boundaries are polygons and use the previous proposition. The representation of length as a mixed volume gives an easy proof of the following monotonicity property of length in twodimensional normed spaces. Proposition 5.6. If K1 ⊂ K2 are nested convex bodies in a twodimensional normed space, then `(∂K1 ) ≤ `(∂K2 ). The proof is left as a simple exercise to the reader. The following related exercise is, perhaps, somewhat harder. Exercise 5.7. Show that a Finsler metric on the plane satisfies the monotonicity property in the previous proposition if and only if its geodesics are straight lines. Theorem 5.8. Let X be a twodimensional normed space with unit disc B and area form Ω ∈ Λ2 X ∗ . Of all convex bodies in X with a given perimeter the one that encloses the largest area is, up to translations, a dilate of I := (iΩ B)∗ .
VOLUMES ON NORMED AND FINSLER SPACES
33
Proof. Let K ⊂ X be a convex body and let `B (∂K) = 2V (K, I) be its perimeter. By Minkowski’s mixed volume inequality, we have `B (∂K)2 = V (K, I)2 ≥ λ(K)λ(I) 4 with equality if an only if K and I are homothetic. Thus, the area enclosed by K is maximal for a given perimeter if and only if K is a dilate of I. Definition 5.9. Let Y 7→ (Λ2 Y, µY ) be a volume definition for twodimensional normed spaces. If X is a twodimensional normed space with unit ball B, the isoperimetrix of X corresponding to the volume definition µ is the body I X := ∗ iΩX (B) , where ΩX is a 2form on X satisfying ΩX  = µX .
We shall denote the isoperimetrices of a twodimensional normed space X with respect to the Busemann, Holmes–Thompson, mass, and mass∗ volume definim m∗ tions by I bX , I ht X , I X , and I X . If T : X → X is an invertible linear transformation, the isoperimetrix, with respect to any volume definition, of the norm with unit ball T (B X ) is T (I X ). Exercise 5.10. If X is a twodimensional normed space with unit ball B and if µ is a particular choice of volume definition, then `B (∂I X ) = 2µX (I X ) and µX (I X ) = µ∗X (B ∗ ).
Using this exercise, we can give sharp estimates on the area and perimeter of the isoperimetrix of a twodimensional normed space for Busemann, Holmes– Thompson, and mass∗ volume definitions. Indeed, it follows trivially from the exercise that µ bX (I bX ) = vp(BX )/π and ht that µht o inequalities, we have X (I X ) = π. Using the Mahler and Blaschke–Santal´ 8 ≤ µbX (I bX ) ≤ π. π m∗ The fact that µm∗ X (I X ) ≤ π with equality if and only if X is Euclidean is equivalent to the inequality µm∗ ≥ µht for twodimensional normed spaces. m∗ Exercise 5.11. Find the sharp lower bound for µm∗ X (I X ).
It is interesting to note that the Blaschke–Santal´ o inequality implies that µbX (BX ) ≥ µbX (I bX )
ht ht and µht X (BX ) ≤ µX (I X ),
with equality in both cases if and only if B is an ellipse. Of course this implies that for both the Busemann and Holmes–Thompson definitions B X = I X if and only if X is Euclidean. Notice that whether a unit disc is equal to its isoperimetrix depends on the volume definition we are using. However, whether the unit disc is a dilate of its isoperimetrix does not depend on such a choice.
34
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Definition 5.12. Let X be a twodimensional normed space. If B X is a dilate of I X for one (and, therefore, any) volume definition, the unit circle, ∂B X , is said to be a Radon curve. For comparison with the higherdimensional case we summarize the properties of the map I that sends a unit disc BX to I X . This maps sends convex bodies to convex bodies; it is a bijection; it commutes with linear maps in the sense that T (BX ) is sent to T (I X ) for all invertible linear maps T ; it maps polygons to polygons, smooth bodies to strictly convex bodies and strictly convex bodies to smooth bodies; and the only fixed points for the µ b and µht normalizations are ellipses. A good, very elementary account of the construction of the isoperimetrix from first principles and its relationship to physics and symplectic geometry (the ball is used for measuring position and the isoperimetrix for measuring velocity) is given by Wallen [1995]. Finally, we explore the relationship between the perimeter and area of the unit ball. The motivation is that `(∂I) = 2µ(I B ) and that in the Euclidean case this holds for the ball. Theorem 5.13. If X is a twodimensional normed space with unit ball B then 2µm (B) ≤ `(∂B) ≤ 2µm∗ (B) with equality on the left if and only if ∂B is a Radon curve and on the right if and only if ∂B is an equiframed curve. For the definition of equiframed curves and a proof of the theorem we refer the reader to [Martini et al. 2001] where the history of this result is also discussed. Exercise 5.14. Use this result and properties of I X to show that BX = I m X if and only if ∂BX is a Radon curve; and that BX = I m∗ if and only if ∂B is X X equiframed. There is a further recent result in this direction. Theorem 5.15 [Moustafaev]. If X is a twodimensional normed space, then 2µht X (BX ) ≤ `(∂BX ), with equality if and only if X is Euclidean. Proof. By definition of the isoperimetrix and Minkowski’s mixed volume inequality, we have ht ht ht `(BX )2 = 4V (BX , I ht X ) ≥ 4µX (BX )µX (I X ). ht ht Using that µht X (I X ) = π and that µX (BX )/π ≤ 1, we have ht `(BX )2 ≥ 4πµht X (BX ) ≥ 4πµX (BX )
µht 2 X (BX ) = 4µht X (BX ) . π
VOLUMES ON NORMED AND FINSLER SPACES
35
Exercise 5.16. If X is a twodimensional normed space, show that 2 ≤ µm X (BX ) ≤ π, 3 ≤ µm∗ X (BX ) ≤ 4, and (using inequalities from Section 3) 8/π ≤ µht X (BX ) ≤ π. Give the equality cases.
6. Area on FiniteDimensional Normed Spaces In Section 4, we saw that the Busemann, Holmes–Thompson, and mass∗ volume definitions induce kvolume densities that are weakly convex. In the special case where the dimension of the normed space X is n = k + 1, then the (n−1)volume densities are norms on the space Λn−1 X. It follows from the properties of the volume definitions that, in all three cases, the map that assigns to the normed space X the normed space Λ n−1 X has the following properties: (1) If T : X → Y is a short linear map between normed spaces X and Y , then the induced map T∗ : Λn−1 X → Λn−1 Y is also short. (2) The map X 7→ Λn−1 X is continuous with respect to the topology induced by the Banach–Mazur distance. (3) If X is a Euclidean space, then the (n−1)volume density is the standard Euclidean area on X. (4) If the dimension of X is two, the map X 7→ Λ1 X is the identity. Notice that property (1) states that for the Busemann, Holmes–Thompson, and mass∗ definitions, the map that takes the normed space X to the normed space Λn−1 X is a covariant functor in the category N of finitedimensional normed spaces. Definition 6.1. A definition of area on normed spaces assigns to every ndimensional, n ≥ 2, normed space X a normed space (Λ n−1 X, σX ) in such a way that properties (1)–(4) above are satisfied. For simplicity, we shall speak of the Busemann, Holmes–Thompson, and mass∗ definitions of area to refer to the definitions of area induced, respectively, by the Busemann, Holmes–Thompson, and mass∗ volume definitions. Definitions of area in normed spaces are related to important constructions in convex geometry such as intersection bodies, projection bodies, and Wulff shapes. However, let us start by posing a few natural questions that arise whenever we have a definition of area. The answer to some of these questions, once specialized to the Busemann and Holmes–Thompson definitions, are deep results in the
36
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
theory of convex bodies. Other questions are longstanding open problems, and yet others seem to be new. Given a definition of area X 7→ (Λn−1 X, σX ) on normed spaces, we may ask: Is the map X 7→ (Λn−1 X, σX ) injective? What is its range? Does it send crystalline norms to crystalline norms? Does it send Minkowski spaces to Minkowski spaces? In what numeric range is the area of the unit sphere of an ndimensional normed space? Other problems arise when we consider the relationship between length, area, and volume, but, for now, let us concentrate on the questions we have just posed. 6.1. Injectivity and range of the area definition Let us start the study of the injectivity and range of the Busemann definition of area by describing the unit ball of the (n−1)volume density in terms of a wellknown construction in convex geometry. Busemann area and intersection bodies. Consider R n with its Euclidean structure and its unit sphere S n−1 . If K ⊂ R n is a starshaped body containing the origin, the intersection body of K, IK, is defined by the following simple construction: if x ∈ R n is a unit vector, let A(K ∩ x⊥ ) denote the area of the intersection of K with the hyperplane perpendicular to x, and let IK be the starshaped body enclosed by the surface {x/A(K ∩ x⊥ ) ∈ R n : x ∈ S n−1 }. A celebrated theorem of Busemann, which is equivalent to the weak convexity of the Busemann volume definition, states that if K is a centered convex body, then IK is also a centered convex body. Let X be an ndimensional normed space. Choose a basis of X and use it to identify X with R n . Take the Euclidean structure in R n for which the basis is orthonormal and use the resulting Euclidean structure to identify the spaces X ∗ and Λn−1 X, as well as to define the unit sphere S n−1 in R n . Exercise 6.2. Show that with all these identifications, the convex body {x ∈ b b R n : σX (x) ≤ 1} is εn−1 times the intersection body of BX ; (here σX is the ∗ n−1 norm induced on X by the norm on Λ X). Notice that we can now write the question of whether the Busemann definition of area is injective in the following classical form: Is a centered convex body determined uniquely by the area of its intersections with hyperplanes passing through the origin? The answer is affirmative (see [Lutwak 1988] and [Gardner 1995]), and so we have the following result: Theorem 6.3. The Busemann area definition is injective. Determining the range of the Busemann area definition is somewhat trickier. Thanks to the efforts of R. Gardner, G. Zhang, and others in the solution of the first of the Busemann–Petty problems, it is known (see [Gardner 1994], [Gardner et al. 1999], and [Zhang 1999] and the references therein) that in dimensions
VOLUMES ON NORMED AND FINSLER SPACES
37
two, three, and four every convex body symmetric with respect to the origin is the intersection body of some starshaped body. It is not clear at this point whether those bodies that are intersection bodies of centered convex bodies can be characterized effectively. For dimensions greater than four, not every centered convex body is an intersection body ([Gardner et al. 1999]). For further information about intersection bodies see, for example, [Gardner 1995] and [Lutwak 1988]. Examples in [Thompson 1996] show that the Busemann area definition does not take crystalline norms to crystalline norms. We don’t know whether it takes Minkowski norms to Minkowski norms. Let us now pass to the Holmes–Thompson definition. Holmes–Thompson area and projection bodies. Consider R n with its Euclidean structure and its unit sphere S n−1 . If K ⊂ R n is a convex body, the projection body of K, ΠK, is given by the following simple construction: if x ∈ R n is a unit vector, let A(Kx⊥ ) denote the area of the orthogonal projection of K onto the hyperplane perpendicular to x, and let the polar of ΠK be the body enclosed by the surface {A(Kx⊥ )x ∈ R n : x ∈ S n−1 }. As in the case of the Busemann definition of area, identifying a normed space X with R n allows us to write the unit ball for the (n−1)volume density in terms of this nonintrinsic construction. Exercise 6.4. Show that by identifying a normed space X with R n as in the ht previous exercise, the convex body {x ∈ R n : σX (x) ≤ 1} is 1/εn−1 times the ∗ polar of the projection body of BX . The question of the injectivity of the Holmes–Thompson definition of area can now be formulated in classical terms: Is a centered convex body determined uniquely by the area of its orthogonal projections onto hyperplanes? The answer, in the affirmative, follows from a celebrated result of Alexandrov [1933] (see also [Gardner 1995]). We then have the following result: Theorem 6.5. The Holmes–Thompson area definition is injective. It is known, basically from the time of Minkowski, that a centered convex body B is the projection body of another if and only if it is a zonoid (see [Gardner 1995]). By Theorem 2.12, this means that for any ndimensional normed space ht X the normed space (Λn−1 X, σX ) is hypermetric. Moreover, because of the integral formula for the Holmes–Thompson (n−1)volume density in terms of the surface area measure of the dual sphere given in the proof of Theorem 4.16, the problem of reconstructing the norm from the Holmes–Thompson (n−1)volume density is precisely the famous Minkowski problem: Reconstruct a convex body from the knowledge of its Gauss curvature as a function of its unit normals. The next two theorems follow directly from
38
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
the work of Minkowski, Pogorelov, and Nirenberg (see [Pogorelov 1978] for a detailed presentation). Theorem 6.6. The range of the Holmes–Thompson area definition is the set of hypermetric normed spaces. Theorem 6.7. Let X be an ndimensional vector space and let σ : Λn−1 X → [0, ∞) be a Minkowski norm. If (Λn−1 X, σ) is hypermetric, then there exists a unique Minkowski norm k · k on X such that σ is the Holmes–Thompson (n−1)volume density of the normed space (X, k · k). Another important feature of the Holmes–Thompson area is the following (for a proof see [Thompson 1996]): Theorem 6.8. The Holmes–Thompson area definition takes Minkowski spaces to Minkowski spaces and crystalline norms to crystalline norms. Mass* area and wedge bodies. Let B be a centered convex body in an ndimensional vector space X, and let B n−1 be the (n−1)fold product of B in the n(n − 1)dimensional space X n−1 . If Alt : X n−1 → Λn−1 X denotes the (nonlinear) map (x1 , . . . , xn−1 ) 7−→ x1 ∧ x2 ∧ · · · ∧ xn−1 , we define the wedge body of B, denoted by WB, as the convex hull of Alt(B n−1 ) in Λn−1 X. We remark that even if B n−1 ⊂ X n−1 is a centered convex body, Alt(B n−1 ) is not necessarily convex. Theorem 6.9. The unit ball in Λn−1 X for the mass∗ (n−1)volume density of ∗ ∗ a normed space X is the body (WBX ) . Proof. By Exercise 4.3, we have m∗ ∗ σX (a) = sup{ξ 1 ∧ ξ 2 ∧ · · · ∧ ξ n−1 · a : ξ 1 , . . . , ξ n−1 ∈ BX }. ∗ n−1 m∗ But this is just the supremum of η · a, where η ∈ Alt (BX ) . Therefore σX ∗ is the dual to the norm in Λn−1 X ∗ whose unit ball is WBX .
It is quite easy to do calculations for WB ∗ in the case when the centered convex body B is a simple object. The following statements are based on such calculations, the details of which are left as exercises (see also [Thompson 1999]). Proposition 6.10. The mass∗ area definition is not injective. Sketch of the proof.. All we must do is find two centered convex bodies B and K such that WB ∗ = WK ∗ , but B 6= K. Let B be the cube with vertices at (±1, ±1, ±1). In this case, B ∗ is the octahedron with vertices (±1, 0, 0), (0, ±1, 0), (0, 0, ±1) and WB ∗ = B ∗ .
VOLUMES ON NORMED AND FINSLER SPACES
39
Let K be the cuboctahedron with vertices (±1, ±1, 0), (±1, 0, ±1), (0, ±1, ±1). The dual ball K ∗ is the rhombic dodecahedron with vertices ±(1, 0, 0), ±(0, 1, 0), ±(0, 0, 1) and (± 12 , ± 12 , ± 12 ). A simple calculation shows that WB ∗ = W K ∗ . In fact, if L is any centered convex body that lies between the cube and the cubeoctahedron then WL∗ = WB ∗ . While it seems unlikely that the wedge body of the unit ball in a Minkowski space is the ball of a Minkowski space, it is not hard to show that the wedge body of a polytope is a polytope. Then: Proposition 6.11 [Thompson 1999]. The mass∗ area definition takes crystalline norms to crystalline norms. The question of determining the range for the mass∗ area definition is completely open. Is it possible that any centered convex body is a wedge body? 6.2. Area of the unit sphere. In this section we give the higherdimensional analogues (as far as we know them) of the theorems of Sch¨ affer and Gol¸ab discussed in Section 5. The Holmes–Thompson definition was designed originally to yield a generalization of Sch¨ affer’s result and we have the following theorem. Theorem 6.12 [Holmes and Thompson 1979]. If B and K are the unit balls of two norms k · kB and k · kK in the vector space X, the Holmes–Thompson area of ∂K in the normed space (X, k · kB ) equals the Holmes–Thompson area of ∂B ∗ ∗ in the normed space (X ∗ , k · kK ). Notice that in particular, the Holmes–Thompson area of the unit sphere of a normed space equals the Holmes–Thompson area of the unit sphere of its dual. Simple calculations show that neither the Busemann, the mass∗, nor the mass definition have this property. In fact, Daniel Hug (private communication) has shown that Theorem 6.12 characterizes the Holmes–Thompson definition. However, the following question remains open. Problem [Thompson 1996]. Is the Holmes–Thompson definition of volume characterized by the fact that the area of the unit sphere of a normed space equals the area of the unit sphere of its dual ? The first result extending Gol¸ab’s theorem to higher dimension is the following sharp upper bound for the Busemann area of a unit sphere. Theorem 6.13 [Busemann and Petty 1956]. The Busemann area of the unit sphere of an ndimensional normed space is at most 2nεn−1 with equality if and only if B is a parallelotope. For n ≥ 3 no sharp lower bound for the Busemann area of the unit sphere of an ndimensional normed space has been proved. It is conjectured that the minimum is nεn attained by the Euclidean ball. However, when n = 3 it is also attained by the rhombic dodecahedron.
40
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
Since µb ≥ µht , an upper bound for the Busemann area is also an upper bound for the Holmes–Thompson area. Corollary 6.14. The Holmes–Thompson area of the unit sphere of an ndimensional normed space is less than 2nεn−1 . While the sharp upper bound for the Holmes–Thompson area of the unit sphere in any dimension greater than two is not known, the sharp lower bound in ´ dimension three is given by the following unpublished result of Alvarez, Ivanov, and Thompson: Theorem 6.15. The Holmes–Thompson area of the unit sphere of a threedimensional normed space is at least 36/π. Moreover , equality holds if the unit ball is a cuboctahedron or a rhombic dodecahedron. Since µb ≥ µh t and µm∗ ≥ µht , we have the following lower bound for the Busemann and mass∗ areas of the unit sphere of a threedimensional normed space. Corollary 6.16. The Busemann and mass∗ areas of unit sphere of a threedimensional normed space is greater than 36/π. Although these bounds are not sharp, they are the best bounds known so far. It is possible to use a variety of inequalities including the Petty projection inequality (in the case of σ ht ) and the Busemann intersection inequality (in the case of σ b ) to give nonsharp lower bounds. The reader is referred to [Thompson 1996] for examples of what one can get. 6.3. Mixed volumes and the isoperimetrix. We now pass to questions concerning the relationship between areas and volumes, and, in particular, to the solution of the isoperimetric problem in finitedimensional normed spaces. The subject is classical and has been studied from different viewpoints by convex geometers, geometric measure theorists, and crystallographers (see, for example, [Busemann 1949b], [Taylor 1978], and [Ambrosio and Kirchheim 2000]). Nevertheless, being interested in a particular intrinsic viewpoint and relations to area on normed and Finsler spaces that are not treated elsewhere, we shall give a short account of the subject. Let X be an ndimensional vector space and let λ be a Lebesgue measure on X. If I ⊂ X is a centered convex body, we can define an (n−1)volume density on X by the following construction: given n − 1 linearly independent vectors x1 , . . . , xn−1 ∈ X, we denote the parallelotope they define by [x1 , . . . , xn−1 ] and set σI (x1 ∧ x2 ∧ · · · ∧ xn−1 ) :=
λ([x1 , . . . , xn−1 ] + tI) − λ([x1 , . . . , xn−1 ]) 1 lim . n t→+0 t
It is easy to see that σI is well defined and that by changing λ for another Lebesgue measure on X we simply multiply σI by a constant. Note also that
VOLUMES ON NORMED AND FINSLER SPACES
41
although the measure of [x1 , . . . , xn−1 ] is zero, we have included it in the formula to stress its relationship with the ndimensional mixed volume of two bodies, V (K[n − 1], L) :=
λ(K + tL) − λ(K) 1 lim . n t→+0 t
With this definition, if K is a convex body in X, Z σI = nV (K[n − 1], I). ∂K
Exercise 6.17. Show that the (n−1)volume density σ I constructed above is a norm on Λn−1 (X), and that Z σI = nλ(I). ∂I
We would also like to reverse this construction: Starting from a norm σ : Λn−1 X → [0, ∞) and a Lebesgue measure λ on X construct a convex body I ⊂ X such that σ = σI . The construction is quite simple: Let Ω be a volume form on X such that Ω = λ and consider the linear isomorphism iΩ : Λn−1 X −→ X ∗ defined by iΩ (x1 ∧ x2 ∧ · · · ∧ xn−1 )(x) = Ω(x1 ∧ x2 ∧ · · · ∧ xn−1 ∧ x). The body I is given by (iΩ B)∗ , where B ⊂ Λn−1 X is the unit ball of σ. In terms of mixed volumes, we have the following result: Proposition 6.18. Let X be an ndimensional vector space, let σ be a norm on Λn−1 X with unit ball B and let λ be a Lebesgue measure on X. Using the notation above, if I := (iΩ B)∗ , we have Z σ = nV (K[n − 1], I) ∂K
for all convex bodies K ⊂ X. To prove the proposition, let us give a simpler, more visual relationship between σ and I := (iΩ B)∗ that is of independent interest. Given a nonzero (n−1)vector a ∈ Λn−1 X, we shall say that a vector v ∈ X is normal to a with respect to I if v ∈ ∂I, the hyperplane parallel to hai and passing through v supports I, and Ω(a ∧ v) > 0. When I is smooth and strictly convex the normal is unique, but this is of no importance to what follows. Notice, and this is important, that v is constructed in such a way that Ω(a ∧ v) = sup{Ω(a ∧ x) : x ∈ I}. Lemma 6.19. Let X be an ndimensional vector space, let σ be a norm on Λn−1 X with unit ball B and let Ω ∈ Λn X ∗ be a volume form on X. If a is a nonzero (n−1)vector on X and v ∈ X is normal to a with respect to I := (iΩ B)∗ , then σ(a) = Ω(a ∧ v).
42
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON ∗
Proof. Let k · k denote the norm in X ∗ whose unit ball is I ∗ = iΩ B. Trivially, we have σ(a) = kiΩ (a)k∗ for any a ∈ Λn−1 X. Therefore, σ(a) = sup{Ω(a ∧ x) : x ∈ I} = Ω(a ∧ v).
In other terms, if x1 , . . . , xn−1 are linearly independent vectors in X and v is normal to x1 ∧ x2 ∧ · · · ∧ xn−1 with respect to I, then the volume of the ndimensional parallelotope [x1 , . . . , xn−1 , v] is the area of the (n−1)dimensional parallelotope [x1 , . . . , xn−1 ]. Proof of Proposition 6.18. Let x1 , . . . , xn−1 be linearly independent vectors in X and let [x1 , . . . , xn−1 ] denote the parallelotope spanned by them. Notice that if v is normal to x1 ∧ x2 ∧ · · · ∧ xn−1 with respect to I, then for any t > 0, the union of the ndimensional parallelotopes [x1 , . . . , xn−1 , tv] and [x1 , . . . , xn−1 , −tv], which we denote by P (t), is contained in the set [x 1 , . . . , xn−1 ] + tI. Moreover, since up to terms of order 2 and higher in t the volumes of P (t) and [x1 , . . . , xn−1 ] + tI are the same, we have λ([x1 , . . . , xn−1 ] + tI) λ(P (t)) 1 1 lim = lim = σ(x1 ∧ x2 ∧ · · · ∧ xn−1 ), n t→+0 t n t→+0 t and this concludes the proof.
We are now ready to solve the isoperimetric problem for convex bodies: Theorem 6.20. Let X be an ndimensional vector space, let σ be a norm on Λn−1 X with unit ball B, and let Ω ∈ Λn X ∗ be a volume form on X. Of all convex bodies in X with a given surface area the one that encloses the largest volume is, up to translations, a dilate of I := (iΩ B)∗ . Proof. Let K ⊂ X be a convex body and let Z σ = nV (K[n − 1], I) ∂K
be its surface area. By Minkowski’s mixed volume inequality, we have Z n σ = nn V (K[n − 1], I)n ≥ nn λ(K)n−1 λ(I) ∂K
with equality if an only if K and I are homothetic. Thus, the volume enclosed by K is maximal for a given surface area if and only if K is a dilate of I. We shall denote the isoperimetrices of a normed space X with respect to the m∗ Busemann, Holmes–Thompson, and mass∗ definitions by I bX , I ht X , and I X , respectively. In the case of the Busemann and Holmes–Thompson definitions, the isoperimetrices can be given, nonintrinsically, in terms of intersection and projection bodies.
VOLUMES ON NORMED AND FINSLER SPACES
43
Exercise 6.21. Using Exercises 6.2 and 6.4, and the construction of the isoperimetrix, show that εn−1 εn ∗ ∗ −1 ∗ I bX = λ(BX )(IBX )∗ and I ht λ (BX ) ΠBX , X = εn εn−1 where λ and λ∗ are, respectively, the Euclidean volumes on X and X ∗ given by their identification with R n . Exercise 6.22. Describe the isoperimetrix I m∗ X B in terms of wedge bodies. 6.4. Geometry of the isoperimetrix. We now turn our attention to problems relating the unit ball of a normed space and its isoperimetrix with respect to some volume definition. Let us start with the deceptively simple problem of estimating the volume of the isoperimetrix. Identifying the normed space X with R n as in Exercise 6.21, we see that the Holmes–Thompson volume of I ht X is ht −1 µht X (I X ) = εn
ε n n εn−1
∗ −n+1 ∗ λ∗ (BX ) λ(ΠBX ).
The statement that this quantity is greater than or equal to ε n with equality if and only if X is Euclidean is known as Petty’s conjectured projection inequality, and is one of the major open problems in the theory of affine geometric inequalities. m∗ Sharp lower bounds for µbX (I bX ) and µm∗ X (I X ) are also unknown, although as observed in [Thompson 1996] the inequality µbX (I bX ) ≥ εn for n ≥ 3 would easily yield (exercise!) that the Busemann area of unit sphere of a normed space of dimension n is at least nεn . Another interesting affine invariant involving the isoperimetrix is the symplectic volume of BX × I ∗X in X × X ∗ . In the twodimensional case this simply yields the square of the area of the unit disc, but in higher dimension it is a much more interesting invariant: Exercise 6.23. Pick up either [Gardner 1995] or [Thompson 1996] and, using Exercise 6.21, prove that the inequality svol(BX × I ∗X ) ≤ εn µX (BX ) is true for the Busemann (resp. Holmes–Thompson) definition by showing that it is equivalent to Busemann’s intersection inequality (resp. Petty’s projection inequality). It would be interesting to complete the picture by having a sharp upper bound ∗ m∗ for svol(BX × (I m∗ X (BX ). X ) ) in terms of µ We finish the paper by considering some questions relating length, area, and volume. In terms of the isoperimetrix they have very simple statements: Given a volume definition, when is the isoperimetrix equal to the unit ball, when is it a multiple of the ball, and when is it inside the ball?
44
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
These simple questions are really about the existence of a coarea formula or inequality for the different definitions of volume on normed and Finsler spaces. Many Riemannian and Euclidean results depend, or seem to depend, on the simple fact that volume = base × height. To what extent is this true in normed and Finsler spaces? In order to relate the coarea formula and inequality with the geometry of the isoperimetrix, let us first define the height of a parallelotope [x 1 , . . . , xn−1 , xn ] in a vector space X with respect to a centered convex body B ⊂ X by the following construction: let ξ ∈ X ∗ be a covector in ∂B ∗ such that ξ(xi ) = 0 for all i between 1 and n − 1. The quantity ξ(xn ), which is independent of the choice of ξ, will be called the height of [x1 , . . . , xn−1 , xn ] with respect to B. By the construction of the isoperimetrix, we know that the volume of the parallelotope [x1 , . . . , xn−1 , xn ] equals the area of its base, [x1 , . . . , xn−1 ], times its height with respect to the isoperimetrix. Therefore, if the volume of every parallelotope in a normed space equals the area of its base times its height with respect to the unit ball, the ball equals the isoperimetrix. If the volume is greater than the area of the base times the height with respect to the unit ball, then the isoperimetrix is contained in the ball, and so on. The first clear sign that the relationship between length, area and volume may not go smoothly on normed and Finsler spaces is the following result of Thompson: Proposition 6.24 [Thompson 1996]. The isoperimetric of a normed space X for the Holmes–Thompson definition is contained in the unit ball if and only if the space is Euclidean. In which case, the ball and the isoperimetric are equal . In other words, the coarea equality or inequality “volume ≥ base × height” for the Holmes–Thompson definition is true only for Euclidean spaces. ∗ ht ∗ Proof. If I ht X ⊂ BX , then BX ⊂ (I X ) and, therefore, ∗ ∗ ht svol(BX × (I ht X ) ) ≥ svol(BX × BX ) = εn µX (BX ).
By Exercise 6.23, the only way this can happen is if X is Euclidean.
However, for the mass∗ definition the coarea inequality is always true: Theorem 6.25 [Gromov 1983]. If X is a finitedimensional normed space, then I m∗ X ⊂ BX . Proof. We must show that if [v 1 , . . . , v n ] is a parallelotope, µm∗ (v 1 ∧ v 2 ∧ · · · ∧ v n ) ≥ σ m∗ (v 1 ∧ v 2 ∧ · · · ∧ v n−1 )ξ(v n ), ∗ where ξ ∈ ∂BX and ξ(v i ) = 0, 1 ≤ i ≤ n − 1. Without loss of generality we may suppose that v 1 , v 2 , . . . v n−1 is an extremal basis in the subspace V ⊂ X they span, i.e. each vector v i is a point of contact between BX ∩ V and a minimal circumscribing parallelotope for B ∩ V . Let u
VOLUMES ON NORMED AND FINSLER SPACES
45
be such that kuk = ξ(u) = 1 and set v n = αu + x where x ∈ V . The right hand side of the above inequality is α. Let ξ 1 , ξ 2 , . . . , ξ n−1 be the dual basis to the v i ’s in V and extend these to the whole of X by setting ξ i (u) = 0. Then ξ 1 , ξ 2 , . . . , ξ n−1 , ξ are all of norm 1 and form the dual basis to v 1 , v 2 , . . . , v n−1 , u. Now m∗ µm∗ X (v 1 ∧ v 2 ∧ · · · ∧ v n ) = αµX (v 1 ∧ v 2 ∧ · · · ∧ v n−1 ∧ u) −1 = α(µm X ∗ (ξ 1 ∧ ξ 2 ∧ · · · ∧ ξ n−1 ∧ ξ)) Y ≥ α(kξk kξi k)−1 = α.
The inequality comes from the definition of mass.
As we have said, the problem of determining for what normed spaces metric balls are solutions to the isoperimetric problem, i.e. when is the isoperimetrix a multiple of the unit ball, is completely open for all three definitions of volume in dimensions greater than two.
Acknowledgments It is a pleasure to acknowledge the hospitality and wonderful working conditions of the Research in Pairs program at the Mathematisches Forschungsinstitut Oberwolfach, where the authors did a substantial part of the research going into ´ this paper. Alvarez also thanks the MAPA Institute at the Universit´e Catholique de Louvain for its hospitality and technical support in the final stages of this work. Thompson is grateful to NSERC for its support through grant #A4066. Many of the ideas in this paper concerning the abstract framework of the subject were first developed in the Geometry Seminar at the Universit´e Catholique de Louvain and were greatly influenced by discussions with Emmanuel Fernandes and Gautier Berck. The thesis [Fernandes 2002] was for us a steppingstone for many of the subjects developed in this work.
References [Alexandrov 1933] A. D. Aleksandrov, “A theorem on convex polyhedra” (Russian), Trudy Fiz.Mat. Inst. Steklov. 4 (1933), 87. ´ [Alvarez 2002] J. C. Alvarez Paiva, “Dual mixed volumes and isosystolic inequalities”, preprint, 2002. ´ [Alvarez and Fernandes 1998] J. C. Alvarez Paiva and E. Fernandes, “Crofton formulas in projective Finsler spaces”, Electronic Research Announcements of the Amer. Math. Soc. 4 (1998), 91–100. ´ [Alvarez and Fernandes 1999] J. C. Alvarez Paiva and E. Fernandes, “Fourier transforms and the Holmes–Thompson volume of Finsler manifolds”, Int. Math. Res. Notices 19 (1999), 1032–104. [Ambrosio and Kirchheim 2000] L. Ambrosio and B. Kirchheim, “Currents in metric spaces”, Acta Math. 185 (2000), 1–80.
46
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
[Benson 1962] R. V. Benson, “The geometry of affine areas”, Ph.D. Thesis, University of Southern California, Los Angeles, (1962). (University Microfilms Inc., Ann Arbor, Michigan 626037). [Bolker 1969] E. D. Bolker, “A class of convex bodies”, Trans. Amer. Math. Soc. 145 (1969), 323–345. [Bourgain and Lindenstrauss 1988] J. Bourgain and J. Lindenstrauss, “Projection bodies”, pp. 250–270 in Geometric aspects of functional analysis (Tel Aviv, 1986/87), edited by J. Lindenstrauss and V. D. Milman, Lecture Notes in Math. 1317, Springer, Berlin, 1988. [Burago and Ivanov 2002] 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. [Busemann 1947] H. Busemann, “Intrinsic area”, Ann. of Math. (2) 48 (1947), 234–267. [Busemann 1949a] H. Busemann, “A theorem on convex bodies of the Brunn–Minkowski type”, Proc. Nat. Acad. Sci. USA 35 (1949), 27–31. [Busemann 1949b] H. Busemann, “The isoperimetric problem for Minkowski area”, Amer. J. Math. 71 (1949), 743–762. [Busemann 1961] H. Busemann, “Convexity on Grassmann manifolds”, Enseign. Math. (2) 7 (1961), 139–152. [Busemann 1969] H. Busemann, “Convex bodies and convexity on Grassmann cones: XI”, Math. Scand. 24 (1969), 93–101. [Busemann and Petty 1956] H. Busemann and C. M. Petty, “Problems on convex bodies”, Math. Scand. 4 (1956), 88–94. [Busemann et al. 1963] H. Busemann, G. Ewald, and G. C. Shephard, “Convex bodies and convexity on Grassmann cones: I–IV”, Math. Ann. 151 (1963), 1–41. [Busemann et al. 1962] H. Busemann, G. Ewald, and G. C. Shephard, “Convex bodies and convexity on Grassmann cones: V”, Arch. Math. 13 (1962), 512–526. [Busemann and Sphephard 1965] H. Busemann and G. C. Shephard, “Convex bodies and convexity on Grassmann cones: X”, Ann. Mat. Pura Appl. 70 (1965), 271–294. [Cohn 1980] D. L. Cohn, Measure theory, Birkh¨ auser, Basel, 1980. [Duran 1998] C. E. Dur´ an, “A volume comparison theorem for Finsler manifolds”, Proc. Amer. Math. Soc. 126 (1998), 3079–3082. [Ewald 1964a] G. Ewald, “Convex bodies and convexity on Grassmann cones: VII”, Abh. Math. Sem. Univ. Hamburg 27 (1964), 167–170. [Ewald 1964b] G. Ewald, “Convex bodies and convexity on Grassmann cones: IX”, Math. Ann. 157 (1964), 219–230. [Fernandes 2002] E. Fernandes, “Double fibrations: a modern approach to integral geometry and Crofton formulas in projective Finsler spaces”, Ph.D. Thesis, Universit´e Catholique de Louvain, 2002. [Gardner 1995] R. J. Gardner, Geometric Tomography, Encyclopedia of Mathematics and its Applications 58, Cambridge University Press, Cambridge, 1995. [Gardner 1994] R. J. Gardner, “A positive answer to the Busemann–Petty problem in three dimensions”, Ann. of Math. (2) 140:2 (1994), 435–447. [Gardner et al. 1999] R. J. Gardner, A. Koldobsky, and T. Schlumprecht, “An analytic solution to the Busemann–Petty problem on sections of convex bodies”, Ann. of Math. 149:2 (1999), 691–703.
VOLUMES ON NORMED AND FINSLER SPACES
47
[Gol¸ab 1932] S. Gol¸ab, “Quelques probl`emes m´etriques de la g´eometrie de Minkowski”, Trav. l’Acad. Mines Cracovie 6 (1932), 1–79. [Goodey and Weil 1993] P. Goodey and W. Weil, “Zonoids and generalisations”, pp. 1297–1326 in Handbook of convex geometry, Vol. B, edited by P. M. Gruber et al., NorthHolland, Amsterdam, 1993. [Gromov 1967] M. Gromov, “On a geometric hypothesis of Banach” (Russian), Izv. Akad. Nauk SSSR Ser. Mat. 31 (1967), 1105–1114. [Gromov 1983] M. Gromov, “Filling Riemannian manifolds”, J. Diff. Geom. 18 (1983), 1–147. [Harvey and Lawson 1982] R. Harvey and H. B. Lawson, “Calibrated geometries”, Acta Math. 148 (1982), 47–157. [Holmes and Thompson 1979] R. D. Holmes and A. C. Thompson, “N dimensional area and content in Minkowski spaces”, Pacific J. Math. 85 (1979), 77–110. [H¨ ormander 1994] L. H¨ ormander, Notions of convexity, Progress in Mathematics 127, Birkh¨ auser, Boston, 1994. [H¨ ormander 1983] L. H¨ ormander, The Analysis of Linear Partial Differential Operators I, Grundlehren der mathematischen Wissenschaften 256, Springer, Berlin, 1983. [Ivanov 2002] S. Ivanov, “On twodimensional minimal fillings”, St. Petersburg Math. J. 13 (2002), 17–25. [John 1948] F. John, “Extremum problems with inequalities as subsidiary conditions”, pp. 187–204 in Studies and essays presented to R. Courant on his 60th birthday, edited by K. O. Friedrichs, O. Neuegebauer, and J. J. Stoker, Interscience, New York, 1948. [Lutwak 1988] E. Lutwak, “Intersection bodies and dual mixed volumes”, Adv. Math. 71 (1988), 232–261. [Mahler 1939] K. Mahler, “Ein Minimalproblem f¨ ur konvexe Polygone”, Mathematica (Zutphen) B7 (1939), 118–127. [Martini et al. 2001] H. Martini, K. J. Swanepoel, and Weiß, G., “The geometry of Minkowski spaces: a survey, part I”, Exposition. Math. 19 (2001), 97–142. [McKinney 1974] J. R. McKinney, “On maximal simplices inscribed to a central convex set”, Mathematika 21 (1974), 38–44. [Moustafaev] Z. Moustafaev, “The ratio of the length of the unit circle to the area of the unit disc in Minkowski planes”, to appear in Proc. Amer. Math. Soc. [Nirenberg 1953] L. Nirenberg, “The Weyl and Minkowski problems in differential geometry in the large”, Comm. Pure Appl. Math. 6 (1953), 337–394. [Pogorelov 1978] A. V. Pogorelov, The Minkowski multidimensional problem, Scripta Series in Mathematics, V. H. Winston, Washington (DC) and Halsted Press, New York, 1978. [Reisner 1985] S. Reisner, “Random polytopes and the volumeproduct of symmetric convex bodies”, Math. Scand. 57 (1985), 386–392. [Reisner 1986] S. Reisner, “Zonoids with minimal volume product”, Math. Z. 192 (1986), 339–346. [Sch¨ affer 1967] J. J. Sch¨ affer, “Inner diameter, perimeter, and girth of spheres”, Math. Ann. 173 (1967), 59–79. [Sch¨ affer 1973] J. J. Sch¨ affer, “The selfcircumference of polar convex bodies”, Arch. Math. 24 (1973), 87–90.
48
´ J. C. ALVAREZ PAIVA AND A. C. THOMPSON
[Schneider 1993] R. Schneider, Convex bodies: the Brunn–Minkowski Theory, Encyclopedia of Math. and Its Appl. 44, Cambridge University Press, New York, 1993. [Schneider 2001] R. Schneider, “On the Busemann area in Minkowski spaces”, Beitr. Algebra Geom. 42 (2001), 263–273. [Schneider 2001] R. Schneider, “Crofton formulas in hypermetric projective Finsler spaces”, Festschrift: Erich Lamprecht. Arch. Math. (Basel) 77:1 (2001), 85–97. [Schneider 2002] R. Schneider, “On integral geometry in projective Finsler spaces”, Izv. Nats. Akad. Nauk Armenii Mat. 37 (2002) 34–51. [Schneider and Weil 1983] R. Schneider and W. Weil, “Zonoids and related topics”, pp. 296–317 in Convexity and its applications, edited by P. M. Gruber and J. M. Wills, Birkh¨ auser, Basel, 1983. [Schneider and Wieacker 1997] R. Schneider and J. A. Wieacker, “Integral geometry in Minkowski spaces”, Adv.in Math. 129 (1997), 222–260. [Shephard 1964a] G. C. Shephard, “Convex bodies and convexity on Grassmann cones: VI”, J. London Math. Soc. 39 (1964), 307–319. [Shephard 1964b] G. C. Shephard, “Convex bodies and convexity on Grassmann cones: VIII”, J. London Math. Soc. 39 (1964), 417–423. [Taylor 1978] J. Taylor, “Crystalline variational problems”, Bull. Amer. Math. Soc. 84:4 (1978), 568–588. [Thompson 1996] A. C. Thompson, Minkowski Geometry, Encyclopedia of Mathematics and its Applications 63, Cambridge University Press, New York, 1996. [Thompson 1999] A. C. Thompson, “On Benson’s definition of area in Minkowski space”, Canad. Math. Bull. 42:2 (1999) 237–247. [Wallen 1995] L. J. Wallen, “Kepler, the taxicab metric and beyond: an isoperimetric primer”, College J. Math. 26 (1995), 178–190. [Weil 1979] W. Weil, “Centrally symmetric convex bodies and distributions II”, Israel J. Math. 32 (1979), 173–182. [Zhang 1999] G. Zhang, “A positive solution to the Busemann–Petty problem in R4 ”. Ann. of Math. (2) 149:2 (1999), 535–543. ´ J. C. Alvarez Paiva Department of Mathematics Polytechnic University Six MetroTech Center Brooklyn, New York, 11201 United States jalvarez duke.poly.edu A. C. Thompson Department of Mathematics and Statistics Dalhousie University Halifax, Nova Scotia Canada B3H 3J5
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Anisotropic and Crystalline Mean Curvature Flow GIOVANNI BELLETTINI
Contents 1. Introduction 2. Notation 2.1. The norm φ and the distance dφ 2.2. The dual norm φo . The duality maps 3. Area Measures in (R n , φ) 3.1. The (n−1)dimensional Hausdorff measure 3.2. The (n−1)dimensional Minkowski content 3.3. From the Hausdorff measure to the Minkowski content 4. First Variation of Area and Mean Curvature Flow. Regular Case 4.1. Preliminaries on the Euclidean case 4.2. The anisotropic regular case 4.3. On space dependent norms 5. The Crystalline Case: Lipschitz φRegular Sets 6. On Facets of Polyhedral Lipschitz φRegular Sets 6.1. Some notation 7. φMean Curvature on a Facet 7.1. Characterization of φcalibrable facets in special cases 8. Concluding Remarks References
49 53 53 55 57 57 59 60 60 60 62 64 65 69 70 72 75 76 77
1. Introduction Motion by mean curvature of an embedded smooth hypersurface without boundary has been the subject of several recent papers, because of its geometric interest and of its application to different areas, see for instance the pioneering book [Brakke 1978], or the papers [Allen and Cahn 1979], [Huisken 1984], [Osher and Sethian 1988], [Evans and Spruck 1991], [Almgren et al. 1993]. A smooth boundary ∂E of an open set E = E(0) ⊂ R n flows by mean curvature if there 49
50
GIOVANNI BELLETTINI
exists a timedependent family (∂E(t))t∈[0,T ] of smooth boundaries satisfying the following property: the normal velocity of any point x ∈ ∂E(t) is equal to the sum of the principal curvatures of ∂E(t) at x. One can show that, at each time t, such an evolution process reduces the area of ∂E(t) as fast as possible. Mean curvature flow has therefore a variational character, since it can be interpreted as the gradient flow associated with the area functional ∂E → H n−1 (∂E), where Hn−1 indicates the (n−1)dimensional Hausdorff measure in R n . In several physical processes (for instance in certain models of dendritic growth and crystal growth, see [Cahn et al. 1992], or in statistical physics (see for example [Spohn 1993]) it turns out, however, that the evolution of the surface is not simply by mean curvature, but is an anisotropic evolution. From the energy point of view, this means that the functional of which we are taking the gradient flow is not the area of ∂E anymore, but is a weighted area, which can be derived by looking at R n as a normed space. Let φ : R n → [0, +∞[ be a norm on R n . One of the most common area measures of ∂E in the normed space (R n , φ) is the socalled Minkowski content Mn−1 dφ (∂E) of ∂E induced by φ. Denoting by dφ (x, y) := φ(y − x) the distance on R n induced by φ and by Hdnφ the ndimensional Hausdorff measure with respect to d φ , Mdn−1 is defined φ as 1 n Mn−1 (1–1) Hdφ {z ∈ R n : inf dφ (z, x) < ρ} . dφ (∂E) := lim+ x∈∂E ρ→0 2ρ Since it is possible to prove that Hdnφ coincides with the Lebesgue measure  ·  multiplied by the factor ωn , cn,φ := {ξ ∈ R n : φ(ξ) ≤ 1} ωn being a normalizing constant, we have 1 {z ∈ R n : inf dφ (z, x) < ρ} . Mn−1 dφ (∂E) = cn,φ lim+ x∈∂E ρ→0 2ρ
(1–2)
Therefore, Mn−1 (∂E) measures (for small ρ > 0) the ratio between the volume φ of a ρtubular neighborhood of ∂E and ρ. Definition (1–1) can be made more explicit, since it turns out that Z Mn−1 (∂E) = c φo (ν E ) dHn−1 . (1–3) n,φ dφ ∂E
Here ν is the Euclidean unit normal to ∂E pointing outside of E, and the function φo : R n → [0, +∞[ is the dual norm of φ. Physically, φo (ν) plays the rˆ ole of a surface tension of a flat surface whose normal is ν, and can be considered as the anisotropy. The functional (1–3) is the above mentioned weighted area, whose gradient flow gives raise to the socalled anisotropic motion by mean curvature. In the regular case, that is, when φ2 is smooth and strictly convex, the relevant quantity is the socalled Cahn–Hoffman vector field n E φ on ∂E, which is the im1 o 2 E E o E age of νφ := ν /φ (ν ) through the map 2 ∇((φ ) ), and whose divergence is E
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
51
the anisotropic mean curvature of ∂E (denoted by κ E φ ). The anisotropic mean curvature is derived (again by a variational principle) in the computation of the first variation of (1–3). Therefore, for any time t, anisotropic mean curvature flow is defined in such a way to decrease Mn−1 dφ (∂E(t)) as fast as possible. Besides the regular case, other anisotropies can be considered; we are particularly interested in the crystalline case, when the function φ is piecewise linear (or equivalently when the unit ball Bφ := {φ ≤ 1} is a polytope). From the mathematical point of view, this field of research was initiated by the work of J. Taylor [Taylor 1978], [Taylor 1986], [Taylor 1991], [Cahn et al. 1992], [Taylor 1992], [Taylor 1993], [Almgren and Taylor 1995]. See also the papers [Hoffman and Cahn 1972], [Cahn and Hoffman 1974], [Cahn et al. 1993]. Recently, several authors contributed to the subject: see for instance • [Girao and Kohn 1994], [Girao 1995], [Rybka 1997], [Ishii and Soner 1999] and [Giga and Giga 2000] for general properties of the crystalline flow in two dimensions and for the convergence of a crystalline algorithm; • [Stancu 1996] for selfsimilar solutions of the crystalline flow in two dimensions; • [Fukui and Giga 1996], [Giga and Giga 1997], [Giga and Giga 1998b], [Giga and Giga 1999] for the crystalline evolution of graphs in two dimensions; • [Giga and Gurtin 1996] for a comparison theorem for crystalline evolutions in two dimensions; • [Roosen and Taylor 1994] for the crystalline evolution in a diffusion field, and [Giga and Giga 1998a] for the crystalline flow with a driving force in two dimensions; • [Ambrosio et al. 2002] for some regularity properties of solutions to crystalline variational problems in two dimensions; • [Yunger 1998] and [Paolini and Pasquarelli 2000] for some properties of the crystalline flow in three dimensions. In two dimensions (that is, for crystalline curvature evolution of curves) the situation is essentially understood, since the notion of crystalline curvature is clear, as well as the corresponding geometric evolution law. For instance, for polygonal initial curves (whose geometry is compatible with the geometry of ∂Bφ ) a comparison principle is available and the flow admits local existence and uniqueness. It turns out that each edge of the curve translates in normal direction during the flow, and the evolution can be described by a system of ordinary differential equations. However, in three space dimensions the situation is not so clear. As in the twodimensional case, it is necessary to redefine what is a smooth boundary ∂E, in order to assign to our interface some notion of φmean curvature. To this purpose we recall a (rather strong) notion of smoothness, the Lipschitz φregularity. Even with this notion at our disposal, the definition of the crystalline mean curvature is quite involved. In addition, once the crystalline mean curvature κ E φ is defined
52
GIOVANNI BELLETTINI
on ∂E, one realizes that, in general, it is not constant on twodimensional facets F of ∂E. This fact is a source of difficulties, since (being κ E φ identified with the normal velocity of ∂E at the initial time) facets can split in several pieces, or can even bend (forming curved regions) during the subsequent evolutionary process. These phenomena (which probably should not be considered as singularities of the flow) partially explain why when n = 3, a short time existence theorem for crystalline mean curvature flow is still missing (even for convex initial data E). Concerning this kind of behavior, we refer also to the work [Yunger 1998]. Before illustrating the plan of the paper, we observe that other choices of area measures are possible in (R n , φ), which are as natural as the φMinkowski content. For example, one could consider the (n−1)dimensional Hausdorff measure (∂E) of ∂E with respect to dφ . Also for this notion of area, an integral repHdn−1 φ resentation theorem is available, which shows in particular that M n−1 dφ (∂E) and n−1 n−1 Hdφ (∂E) may differ. Therefore, taking the first variation of H dφ would give a notion of mean curvature (see [Shen 1998]) which is different from κ E φ . This, in turn, implies that the gradient flow of the dφ Hausdorff measure functional is a different geometric evolution process. Our viewpoint will be to work with the φMinkowski content of ∂E. We remark that all the theory that we develop can be similarly constructed for Hdn−1 : indeed, what is really relevant is that φ n−1 Hdφ (∂E) can also be represented as an integral on ∂E, by means of an integrand (weighting ν E ) which is convex. Finally, we recall that the mean curvature obtained from the first variation of the volume form for the Minkowski content on regular hypersurfaces in a Finsler manifold is considered in [Shen 2001]. The content of the paper is the following. In Section 2 we give some notation. In particular, in Subsection 2.1 we introduce the norm φ, its unit ball B φ and the induced distance dφ . In Subsection 2.2 we recall the main properties of the dual φo of φ and of the duality maps Tφ and Tφo . In particular, we discuss the geometric properties of such maps, also in the crystalline case. In Subsection 3.1 we discuss the integral representation of Hdn−1 (Theorem 3.3). In Subsection 3.2 φ n−1 we discuss the integral representation of Mdφ (Theorem 3.7). The relations between Hdn−1 and Mn−1 are considered in Subsection 3.3. Section 3 is based dφ φ on the results proved in [Bellettini et al. 1996]. In Section 4 we recall the first variation of area in the Euclidean case, and the main definitions and properties of Euclidean mean curvature flow. We rely heavily on the notion of oriented distance function from ∂E. In Subsection 4.2 we focus attention on the regular case and on the first variation of the weighted area. The definition of φmean curvature is given in (4–9). Also here the oriented φdistance function plays a crucial rˆ ole. In Subsection 4.3 we state some generalizations of the previous results when the norm is spacedependent. Subsections 4.2 and 4.3 are based on the results proved in [Bellettini and Paolini 1996]. The crystalline case is deepened in Section 5. Lipschitz φregularity is introduced in Definition 5.2, and illustrated with examples. The geometry of a facet F ⊂ ∂E is studied in Section
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
53
6, which is preliminary to the definition of crystalline mean curvature on F (Definition 7.2). We will restrict for simplicity to polyhedral Lipschitz φregular sets. After some examples, in Subsection 7.1 we illustrate some properties of those facets having constant crystalline mean curvature. Sections 5, 6 and 7 are based on the results originally proved in [Bellettini et al. 1999], [Bellettini et al. 2001a], [Bellettini et al. 2001b], [Bellettini et al. 2001c]. We conclude the paper with Section 8, where we briefly summarize the main ideas and motivations behind our approach.
2. Notation Given two vectors v, w ∈ R n , n p ≥ 2, we denote by hv, wi the scalar product between v and w. We also set v := hv, vi and S n−1 := {v ∈ R n : v = 1}. Given an integer k ∈ [0, n], we denote by H k the kdimensional Hausdorff measure in R n (see [Federer 1969] and [Ambrosio et al. 2000]). If B ⊂ R n is a Borel set, we let B to be the Lebesgue measure of B (which equals H n (B)) and dist(x, B) to be the distance of the point x ∈ R n from B, defined as inf{y −x : y ∈ B}. Even if B is a smooth hypersurface, it is not difficult to realize thatthe distance funcR +∞ tion dist( · , B) is not differentiable on B. We let ω m := π m/2 0 sm/2 e−s ds, which turns out to be the Lebesgue measure of {x ∈ R m : x ≤ 1}, for an integer m ∈ [0, n]. We say that the set M ⊂ R n is an (n−1)dimensional Lipschitz manifold if M can be written, locally, as the graph of a Lipschitz function (with respect to a suitable orthogonal system of coordinates) defined on an open subset of R n−1 . We say that the open set E ⊂ R n is Lipschitz (or that its topological boundary ∂E is Lipschitz) if ∂E can be written, locally, as the graph of a Lipschitz function of (n−1) variables (with respect to a suitable orthogonal system of coordinates) and E is locally the subgraph. We recall (see [Federer 1969]) that, if E has Lipschitz boundary, then at Hn−1 almost every x ∈ ∂E, the unit (Euclidean) normal vector to ∂E pointing toward R n \ E is well defined and, in the sequel, will be denoted by ν E (x). If f is a smooth function defined on an open subset of R n , we denote by ∇f = (∂f /∂x1 , . . . , ∂f /∂xn ) the gradient of f and by ∆f the Laplacian of f . 2.1. The norm φ and the distance dφ . In what follows we indicate by φ : R n → [0, +∞[ a convex function which satisfies the properties φ(ξ) ≥ λξ,
ξ ∈ Rn,
(2–1)
for a suitable constant λ ∈ ]0, +∞[, and φ(aξ) = a φ(ξ),
ξ ∈ R n , a ∈ R.
(2–2)
Notice that our assumptions ensure that there exists a constant Λ ≥ λ such that φ(ξ) ≤ Λξ for any ξ ∈ R n . The function φ is a norm on R n , called a Minkowski
54
GIOVANNI BELLETTINI
norm (or Minkowski metric). The vector space R n endowed with φ is a normed space called Minkowski space and is probably the simplest example of a Finsler manifold: in this case, the manifold is R n , and φ is a norm (independent of the position) on its tangent space which obviously is a copy of R n . We set Bφ := {ξ ∈ R n : φ(ξ) ≤ 1} (the unit ball of φ), a bounded convex set containing the origin in its interior and centrally symmetric (a socalled symmetric convex body). B φ is usually called the indicatrix (sometimes also Wulff shape). The function φ can be identified with Bφ , since given a symmetric convex body K, the function ξ → inf{α > 0 : ξ ∈ αK} is a convex function satisfying (2–1), (2–2) and having K as unit ball. Clearly φ is uniquely determined by its values on the unit sphere S n−1 . Remark 2.1. If we weaken assumption (2–2) into φ(aξ) = aφ(ξ) for any ξ ∈ R n and any a ≥ 0, we have that Bφ is not centrally symmetric anymore. Some of the results in the next sections can be generalized to nonsymmetric functions φ; in the sequel, for simplicity we will restrict to the symmetric case. It is always useful to keep in mind the Euclidean case, which corresponds to the choice φ(ξ) = ξ. The p Riemannian case corresponds to a norm φ whose B φ is an ellipsoid: φ(ξ) := hAξ, ξi for a real positive definite symmetric matrix A.
Definition 2.2. We say that φ is regular if Bφ has boundary of class C ∞ and each principal curvature of ∂Bφ is strictly positive at each point of ∂Bφ . Pn Example 2.1. Let p ∈ [1, +∞[, p 6= 2 and set φ(ξ) := ( i=1 ξi p )1/p . Then φ is not regular: indeed, if p > 2 then Bφ is of class C 2 but the requirement on principal curvatures in Definition 2.2 is not fulfilled. On the other hand, if p < 2 then ∂Bφ is not of class C 2 . Definition 2.3. We say that φ is crystalline if Bφ is a polytope. Pn Example 2.2. The norms φ(ξ) := i=1 ξn  and φ(ξ) := max{ξ1 , . . . , ξn } are crystalline. It is well known that, given the norm φ, we can measure distances in R n by “integrating” φ as follows: the φdistance dφ (x, y) between two points x, y ∈ R n is given by nZ 1 o n dφ (x, y) = inf φ(γ) ˙ dt : γ ∈ AC([0, 1]; R ), γ(0) = x, γ(1) = y 0
= φ(y − x),
(2–3)
where AC([0, 1]; R n ) is the class of all absolutely continuous curves from [0, 1] to R n . The last equality in (2–3) is, for instance, a consequence of Jensen’s inequality. The function dφ is nonnegative, symmetric, vanishes only if x = y, and satisfies the triangular property. To be consistent with the beginning of this
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
55
section, when φ(ξ) = ξ (Euclidean case) we omit the subscript φ in the notation of the distance function. Remark 2.4. We recall the following interesting fact. Given a distance d : R n × R n → [0, +∞[ on R n , by differentiation we can construct a new function ψd : R n × R n → [0, +∞] which, under suitable assumptions, turns out to be a Finsler metric (which, in this case, depends on the position x): ψd (x, ξ) := lim sup t→0
d(x, x + tξ) , t
x ∈ Rn, ξ ∈ Rn.
Some of the properties of the functions ψd , dψd have been investigated for instance in the papers [De Giorgi 1989], [De Giorgi 1990], [Venturini 1992], [Bellettini et al. 1996], [Amar et al. 1998] (see also [De Cecco and Palmieri 1993], [De Cecco and Palmieri 1995] for related results on Lipschitz manifolds). 2.2. The dual norm φo . The duality maps. Given the norm φ acting on vectors, we define the dual norm φo : R n → [0, +∞[ of φ (acting on covectors) as φo (ξ o ) := sup{hξ, ξ o i : ξ ∈ Bφ },
(2–4)
ξo ∈ Rn. n
It is not difficult to verify that φo is a norm on (the dual of) R , and that (φo )o = φ. It is possible to prove that if φ is regular then also φ o is regular, and that if φ is crystalline then φo is crystalline. The function φo is strictly related to the Legendre–Fenchel transform φ∗ of φ, defined as φ∗ (ξ o ) := sup{hξ, ξ o i − φ(ξ) : ξ ∈ R n }. Indeed φ∗ (ξ o ) = +∞ if ξ ∈ / Bφo , and φ∗ (ξ o ) = 0 if ξ ∈ Bφo . Example 2.3. Figure 1 describes how to construct B φo starting from φ. Assume we have been given a smooth symmetric convex body {φ ≤ 1} as (the ellipse) in Figure 1. Let ν ∈ S n−1 be a unit vector. By definition and using the homogeneity, computing φo (ν) is equivalent to solve the maximum problem max{hν, zi : z ∈ ∂Bφ }. The vector ξ = ξ(ν) in Figure 1 is the solution, hence φ o (ν) = hν, ξi. Observe that the strict convexity of Bφ ensures uniqueness of the solution of the maximum problem (2–4); it is clear that if ∂Bφ contains some flat region, problem (2–4) has in general more than one solution. Remark 2.5. Notice that φo (ν) =
1 1 H (prν (Bφ )), 2
(2–5)
where prν (Bφ ) denotes the orthogonal projection of Bφ onto the line Rν; see Figure 1. The map described in Figure 1 associating to the vector ξ ∈ ∂B φ the vector ν/φo (ν) ∈ ∂Bφo (extended in a onehomogeneous way on the whole of R n ) is called the duality map, and can also be defined in the crystalline case.
56
GIOVANNI BELLETTINI
ξ Bφ
ν
S1
H 1 (prν (Bφ )) −ξ
Figure 1. ν is a unit vector. The vector ξ ∈ ∂Bφ (dotted line) is such that φo (ν) = hν, ξi. Observe that the line tangent to ∂Bφ at ξ is orthogonal to ν. The distance between this line and its parallel line tangent to ∂Bφ at −ξ is the length of the orthogonal projection of Bφ onto the onedimensional subspace Rν, and equals 2φo (ν).
Definition 2.6. By Tφ and Tφo we denote the (possibly multivalued) duality maps defined as Tφ (ξ) := {ξ o ∈ R n : hξ o , ξi = φ(ξ)2 = (φo (ξ o ))2 }, Tφo (ξ o ) := {ξ ∈ R n : hξ, ξ o i = (φo (ξ o ))2 = φ(ξ)2 },
ξ ∈ R n, ξo ∈ Rn.
(2–6)
Possibly adopting the conventions on multivalued mappings (see for instance [Brezis 1973]) one can check that Tφ (aξ) = aTφ (ξ) for any ξ ∈ R n and any a ∈ R, and similarly for Tφo . Moreover Tφ takes ∂Bφ onto ∂Bφo , Tφo takes ∂Bφo onto ∂Bφ , and Tφ Tφo = Tφo Tφ = Id. Remark 2.7. Let φ be regular. Then Tφ and Tφo are single valued. Moreover Tφ =
1 ∇((φ)2 ) = φ∇φ, 2
T φo =
1 ∇((φo )2 ) = φo ∇φo . 2
(2–7)
Example 2.4. When φ(ξ) = ξ (Euclidean case), then T φ = Id. When φ(ξ) = p hAξ, ξi is Riemannian, then Tφ (ξ) = Aξ.
Figure 2 illustrates how to construct Tφ in a regular case. First of all, since Tφ is onehomogeneous, it is enough to evaluate Tφ on ∂Bφ . The point ξ belongs to ∂Bφ ; since Tφ (ξ) = ∇φ(ξ) and ∂Bφ is a level set of φ, it is clear that Tφ (ξ) is orthogonal to ξ. In addition, φo (∇φ(ξ)) = 1 = hξ, ∇φ(ξ)i, which implies Tφ (ξ) = hξ, ∇φ(ξ)/∇φ(ξ)i−1 . In what follows, it is important to keep in mind that the duality maps in (2–6) are still well defined under our assumptions on a Minkowski norm, in particular in the crystalline case. If φ is not regular, the equalities in (2–7) become Tφ = 21 ∇− ((φ)2 ) = φ∇− φ,
Tφo = 12 ∇− ((φo )2 ) = φo ∇− φo ,
(2–8)
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
57
ξo ξ ξo
Tφ
Bφ
B φo
Figure 2. The point ξ belongs to ∂Bφ . The point Tφ (ξ) := ξ o ∈ ∂Bφo is the unit normal ν Bφ (ξ) to ∂Bφ at ξ, multiplied by the factor hξ, ν Bφ (ξ)i−1 .
where ∇− denotes the usual subdifferential in convex analysis (see for instance [Rockafellar 1972]); the main feature of the maps T φ and Tφo is that they are no longer onetoone. Geometrically, if ξ ∈ ∂Bφ , then Tφ (ξ) is the intersection of the closed outward normal cone to ∂Bφ with ∂Bφo . In Figure 3 we show an example of Bφ and of its dual body Bφo : Bφ is the Cartesian product of a planar regular hexagon with the interval [−1, 1]. If ξ ∈ ∂Bφ is a point in the relative interior of a facet, then the normal cone Tφ (ξ) to ∂Bφ at ξ is a singleton (a vertex in ∂Bφo ); if ξ ∈ ∂Bφ is a point in the relative interior of an edge, then T φ (ξ) is a onedimensional closed segment (a closed edge in ∂Bφo ); if ξ ∈ ∂Bφ is a vertex, then Tφ (ξ) is a closed triangle (a closed facet in ∂Bφo ).
Bφ
B φo
Figure 3. Dual polytopes. Duality maps take vertices into closed facets, points in the relative interior of a facet into vertices, and points in the relative interior of an edge into closed edges. The Bφo depicted here is supposed to look like an umbrella viewed from above.
3. Area Measures in (R n , φ) 3.1. The (n−1)dimensional Hausdorff measure. We recall the definition of Hausdorff measure in the metric space (R n , dφ ); see [Federer 1969].
58
GIOVANNI BELLETTINI
Definition 3.1. If A ⊆ R n is a Borel subset of R n and m ∈ {n − 1, n} we set Hdmφ (A)
ωm := m lim+ inf 2 δ→0
+∞
X
(diamdφ (Si ))
m
i=1
:A⊆
S+∞ i=1
Si , diamdφ (Si ) < δ ,
(3–1) where diamdφ (Si ) := sup{dφ (s, σ) : (s, σ) ∈ Si × Si } is the diameter of the set Si with respect to dφ . It is not difficult to prove that the limit in (3–1) exists. Consistent with the notation in Section 2, when φ is the Euclidean norm we omit the subscript d φ in the notation of the Hausdorff measure. The following integral representation result provides an explicit formula for computing H dnφ and Hdn−1 . φ Theorem 3.2. Let φ be a norm on R n . Then Z ωn ωn dx = A, Hdnφ (A) = B  B φ φ A for any Borel set A ⊆ R n . Given ν ∈ S n−1 , denote by Sν (Bφ ) := {ξ ∈ Bφ : hν, ξi = 0} the section of Bφ with the hyperplane orthogonal to ν passing through the origin; moreover, set 1 Iφ(ν) := n−1 . (3–2) H (Sν (Bφ )) Theorem 3.3. Let φ be a norm on R n . Let M be a (n−1)dimensional Lipschitz manifold in R n . Then Z n−1 Hdφ (A ∩ M ) = ωn−1 Iφ(ν M ) dHn−1 , (3–3) A∩M
where ν (x) is a unit vector normal to M at x and A ⊆ R n is a Borel set. M
The measure Hdn−1 is also called Busemann surface measure, see the books φ [Thompson 1996], [Schneider 1993] for detailed information on this topic and for complete references. We will still denote by Iφ the onehomogeneous extension of the function in (3–2) on the whole of R n . We then have that the unit ball of Iφ can be written as {ξ ∈ R n : Iφ(ξ) ≤ 1} = ξ ∈ R n : ξ ≤ Hn−1 (Sξ/ξ (Bφ )) ,
that is, {Iφ ≤ 1} is the socalled intersection body I(B φ ) of Bφ . An interesting result of Busemann ensures that I(Bφ ) is convex (see [Busemann 1947], [Busemann 1949], [Thompson 1996], [Schneider 1993] and references therein). This, in turn, is essentially equivalent to the following semicontinuity property of H dn−1 : φ
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
59
if {Ek }k is a sequence of finite perimeter sets whose characteristic functions converge in L1 (R n ) to the characteristic function of a finite perimeter set E, then Hdn−1 (∂E) ≤ lim inf k→+∞ Hdn−1 (∂Ek ). φ φ 3.2. The (n−1)dimensional Minkowski content. Definition 3.4. Let M be a (n−1)dimensional Lipschitz manifold. We define the (n−1)dimensional Minkowski content Mn−1 dφ (M ) of M with respect to dφ as Hdnφ ({z ∈ R n : distφ (z, M ) < ρ}) . (3–4) Mn−1 (M ) := lim dφ 2ρ ρ→0+ Under our regularity assumption on ∂E it is possible to prove that the limit in (3–4) exists (see [Federer 1969], [Ambrosio et al. 2000]). In the Euclidean case φ(ξ) = ξ, the (n−1)dimensional Minkowski content coincides with the (n−1)dimensional Hausdorff measure. Observe that Hdnφ and Mn−1 are invariant under isometries between the φ normed ambient spaces, while cn,φ is not invariant. Remark 3.5. The Minkowski content Mn−1 provides a notion of surface meadφ sure which is constructed by means of the orthogonal projections of B φ onto the (onedimensional) normal spaces to the manifold M ; on the other hand the is constructed by means of the intersections of B φ with Hausdorff measure Hdn−1 φ the ((n−1)dimensional) tangent spaces to M . Other notions of surface measure, different in general from these two notions, can be considered, such as the Holmes–Thompson measure, see [Thompson 1996], or the definitions introduced in [De Giorgi 1995] (see [Ambrosio and Kirchheim 2000]). Remark 3.6. It can be proved that, for a Lipschitz set E, M n−1 dφ (∂E) coincides with the perimeter of the set E with respect to φ, whose definition is given in a distributional way. The following representation result provides an explicit integral formula for computing the Minkowski content of a sufficiently smooth set. Theorem 3.7. Let M be a (n−1)dimensional Lipschitz manifold . Then Z n−1 Mdφ (M ) = cn,φ φo (ν M ) dHn−1 . (3–5) M
The validity of Theorem 3.7 can be explained as follows: the measure H dnφ is cn,φ times the Lebesgue measure. Moreover the Lebesgue measure of the ρtubular neighborhood (in the distance dφ ) in (3–4) is approximately Hn−1 (M ) multiplied by the 1dimensional length of the orthogonal projection of {φ ≤ ρ} in the direction ν, see Figure 4. This length equals 2ρφ o (ν), hence (3–5) follows. Since φo is convex, the following semicontinuity property of M dn−1 holds: φ if {Ek }k is a sequence of finite perimeter sets whose characteristic functions converge in L1 (R n ) to the characteristic function of a finite perimeter set E, n−1 then Mn−1 dφ (∂E) ≤ lim inf k→+∞ Mdφ (∂Ek ).
60
GIOVANNI BELLETTINI
ν
M
2φo (ν)
{φ ≤ ρ}
Figure 4. We assume for simplicity that M is flat, and that ν ∈ S n−1 is orthogonal to M . The ρtubular neighborhood (in the distance dφ ) is obtained by centering rescaled sets {φ ≤ ρ} of {φ ≤ 1} at points of M .
3.3. From the Hausdorff measure to the Minkowski content. As already remarked in the Introduction, in what follows we will work with M n−1 dφ and this will affect the value of the φmean curvature. However, one could consider H dn−1 φ as well, and change all subsequent definitions by replacing φ o with Iφ. We refer to [Shen 1998] for a definition of mean curvature obtained by considering the intrinsic Hausdorff measure. The following result follows from Theorems 3.3 and 3.7. Proposition 3.8. Let φ be a norm on R n . Then 1/(n−1) ωn−1 B(Iφ)o  n−1 n−1 (Iφ)o . Hdφ = Mdψ , ψ := ωn
(3–6)
4. First Variation of Area and Mean Curvature Flow. Regular Case In this section we define the anisotropic φmean curvature. We first recall some facts concerning the Euclidean case. 4.1. Preliminaries on the Euclidean case. Let E ⊂ R n be an open set with smooth compact boundary. It is known (see for instance [Giusti 1984], [Ambrosio 1999]) that, under these assumptions, there exists a tubular neighborhood U of ∂E such that the oriented distance function dE from ∂E negative inside E, defined as dE (z) := dist(z, E) − dist(z, R n \ E), z ∈ R n , is smooth on U , and ∇dE (z) = 1 for any z ∈ U (eikonal equation). Hence, given any x ∈ ∂E, ∇dE (x) is the outer unit normal ν E (x) to ∂E at x. In addition ∆dE (x) is the sum of the principal curvatures (the mean curvature) of ∂E at x. Therefore −∆dE (x)∇dE (x) is the mean curvature vector of ∂E at x.
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
61
In order to compute the first variation of area, we need to introduce a class of admissible variations. We define a family Ψλ of compactly supported diffeomorphisms of the ambient space R n as follows. Denote by Ψ : R n+1 → R n a smooth vector field. Given any λ ∈ R, define Ψλ : R n → R n as Ψλ (x) := Ψ(x, λ). Assume that Ψ0 = Id and that Ψλ − Id has compact support for any λ ∈ R \ {0}. The following theorem is the classical result on the first variation of area, see for instance [Giusti 1984]. Theorem 4.1. For any λ ∈ R define Eλ := Ψλ (E). Then Z d n−1 ∆dE hX, ∇dE i dHn−1 , H (∂Eλ )λ=0 = dλ ∂E
(4–1)
where X := (∂Ψλ /∂λ)λ=0 . Observe that only the normal component of X enters in formula (4–1). We are now in a position to define the smooth mean curvature flow starting from a given ∂E. Definition 4.2. Let E ⊂ R n be an open set with smooth compact boundary. Let T > 0 and, for any t ∈ [0, T ], let E(t) be a set with compact boundary. We say that (E(t))t∈[0,T ] is a smooth mean curvature flow in [0, T ] starting from E = E(0) if the following conditions hold: (i) there exists an open set A ⊂ R n containing ∂E(t) for any t ∈ [0, T ] such that, if we set ¯ t) := dist(z, E(t)) − dist(z, R n \ E(t)), d(z, we have d¯ ∈ C ∞ A × [0, T ] ; (ii)
∂ ¯ ¯ t), d(x, t) = ∆d(x, ∂t
z ∈ R n , t ∈ [0, T ],
x ∈ ∂E(t), t ∈ [0, T ].
(4–2)
Condition (i) implies that each ∂E(t) is a smooth boundary, smoothly evolving ¯ ¯ t) is the projection of the velocity of the in time. The vector −(∂ d/∂t)∇ d(x, point x on the normal space to ∂E(t) at x (see for instance [Ambrosio 1999]). Example 4.1. The main explicit example of mean curvature flow is the one of the sphere E = {z < R0 }, which shrinks selfsimilarly. Indeed in this case ¯ t) = z − R(t), and the equation (4–2) becomes R˙ = −(n − 1)/R. Its d(z, p solution represents the evolving sphere of radius R(t) = R02 − 2(n − 1)t for t ∈ [0, 12 R02 /(n − 1)], which disappears for times larger than 21 R02 /(n − 1). It is customary to say that the evolution law (4–2) is the gradient flow of the area functional Hn−1 (∂E); the idea is that, at each time, the set E(t) evolves in such a way to make Hn−1 (∂E(t)) as small as possible. This assertion has been made rigorous in the paper [Almgren et al. 1993], where it is shown that the
62
GIOVANNI BELLETTINI
correct (nonsymmetric) distance between sets to use in order to obtain the mean curvature flow is Z (E, F ) ⊂ R n × R n → dist(x, ∂F ) dx. (4–3) (E\F )∪(F \E)
Mean curvature flow has been the subject of several recent papers, of which we list some. We refer to: • [Gage and Hamilton 1986] and [Evans and Spruck 1992a] for a local in time existence and uniqueness theorem of a smooth solution; • [Ecker and Huisken 1989] and [Ecker and Huisken 1991] for the evolution of graphs; • [Huisken 1984] for the evolution of convex sets; • [Grayson 1987] and [Angenent 1991a] for local and global properties of the flow of curves; • [Barles et al. 1993], [Evans and Spruck 1992b], [Evans and Spruck 1995], [White 1995], [White 2000], [White 2003] for some qualitative properties of the flow. • Concerning global in time solutions, defined after the onset of singularities, we refer to: • [Brakke 1978], where a geometric measure theory approach is introduced; • [Evans and Spruck 1991] and [Chen et al. 1991] for the level set method and viscosity solutions; • [Ilmanen 1992], [Ilmanen 1993a], [De Giorgi 1993], [De Giorgi 1994] and [Bellettini and Paolini 1995] for the barriers method (see also [Soner 1993]); • [Almgren et al. 1993] for a variational approach based on time discretization; • [Ilmanen 1993b] for the approximation of mean curvature flow by means of a sequence of reactiondiffusion equations; • [Ilmanen 1994] for the elliptic regularization method; • [Angenent 1991a], [Soner and Souganidis 1993], [Altschuler et al. 1995] and [Bellettini and Paolini 1994] for the analysis of some kind of singularity of the flow; • [Fierro and Paolini 1996], [Paolini and Verdi 1992], [Angenent et al. 1995] for numerical simulations of certain singularities. 4.2. The anisotropic regular case. We assume in this subsection that φ is a regular norm. Let E be a set with smooth compact boundary. Also in this case it is possible to prove that there exists a tubular neighborhood U of ∂E such that the oriented φdistance function dE φ from ∂E negative inside E, defined as n dE φ (z) := distφ (z, E) − distφ (z, R \ E),
z ∈ Rn,
is smooth on U , and φo (∇dE φ (z)) = 1 for any z ∈ U (anisotropic eikonal equation). Therefore o E o E o E Tφo (∇dE φ (z)) = φ (∇dφ (z))∇φ (∇dφ (z)) = ∇φ (∇dφ (z)),
z ∈ U.
(4–4)
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
63
In particular, given any x ∈ ∂E, we have ∇dE φ (x) =
ν E (x) =: νφE (x). φo (ν E (x))
(4–5)
The following result is a generalization of Theorem 4.1, and shows also the rˆ ole of the duality map Tφo . Let Ψλ and Eλ be as in Subsection 4.1. Theorem 4.3. For any λ ∈ R define Eλ := Ψλ (E). Then Z d E o E n−1 Mn−1 (∂E ) = c div nE , λ n,φ λ=0 φ hX, ∇dφ i φ (ν )dH dλ dφ ∂E where X :=
∂ ∂λ Ψλ λ=0 ,
(4–6)
and E nE φ (z) := Tφo (∇dφ (z)),
z ∈ U.
(4–7)
The vector field nφ is sometimes called the Cahn–Hoffman vector field, and satisfies E E φ(nE (4–8) φ ) = 1 = h∇dφ , nφ i on U. E E In the Euclidean case φ(ξ) = ξ we have nE on ∂E. φ = νφ = ν
Remark 4.4. The left hand side of (4–6) depends on the values of φ o only on S n−1 (recall (3–5)). Hence also the right hand side of (4–6), written in terms of the onehomogeneous extension of φo on the whole of R n , must depend only on the values of φo on S n−1 . Observe that d Mn−1 (∂Eλ )λ=0 = dλ dφ
Z
n−1 E div nE , φ hX, ∇dφ i dPφ ∂E
where dPφn−1 is the measure on ∂E having cn,φ φo (ν E ) as density with respect to Hn−1 . We are in a position to give the following definition. Definition 4.5. Let E be an open set with smooth compact boundary. We define the φmean curvature κE φ of ∂E as E κE φ (x) := div nφ (x),
x ∈ ∂E.
(4–9)
E It is possible to prove that κE φ (x) is also the tangential divergence of n φ evaluated E E at x ∈ ∂E. Indeed, define f (z) := hν (x), nφ (z)i for any z ∈ U . Thanks to (4–8), f has a maximum at x (with value φo (ν E (x))). Therefore ∇f vanishes at E x, that is, ν E (x)∇nE φ (x) = 0. This implies that the tangential divergence of n φ E at x equals div nφ (x).
Example 4.2. The φmean curvature of ∂Bφ is constantly equal to n−1. Indeed B B ∇dφ φ (z) = ∇φ(z) and Tφo (∇dφ φ (z)) = z/φ(z) for any z in R n \ {0}. Then a computation gives div(z/φ(z)) = n − 1 on ∂Bφ .
64
GIOVANNI BELLETTINI
Example 4.3. Let n = 2, and write φo (ξ o ) = ξ o φo (ξ o /ξ o ) =: ρψ(θ), where E 00 E ξo = (ξ1o , ξ2o ) = (ρ cos θ, ρ sin θ). Then κE is the φ = κ (ψ + ψ ), where κ Euclidean curvature of ∂E. 2 o E 2 E Example 4.4. Observe that from (4–4) we derive κE φ = tr(∇ φ (∇dφ )∇ dφ ) 2 on ∂E, where ∇ denotes the Hessian matrix and tr is the trace operator.
Remark 4.6. In the paper [Bellettini and Fragal` a 2002] the second variation of Mn−1 is computed. What replaces the squared length of the second fundamental dφ E E form of ∂E is the term tr(∇nφ ∇nφ ). In the same paper, a sort of Laplace– Beltrami operator is introduced, see also [Bao et al. 2000] (for references therein), [Shen 2001], [Mugnai 2003]. We are now in a position to define what is a smooth anisotropic mean curvature flow, for a regular anisotropy φ. Definition 4.7. Let E ⊂ R n be an open set with smooth compact boundary. Let T > 0 and, for any t ∈ [0, T ], let E(t) be a set with compact boundary. We say that (E(t))t∈[0,T ] is a smooth φmean curvature flow in [0, T ] starting from E = E(0) if the following conditions hold: (i) there exists an open set A ⊂ R n containing ∂E(t) for any t ∈ [0, T ] such that, if we set d¯φ (z, t) := distφ (z, E(t)) − distφ (z, R n \ E(t)), we have dφ ∈ C ∞ A × [0, T ] ;
z ∈ R n , t ∈ [0, T ],
∂ E(t) dφ (x, t) = divnφ (x), x ∈ ∂E(t), t ∈ [0, T ]. (4–10) ∂t Example 4.5. We show that {ξ ∈ R n : φ(ξ) < R0 } shrinks selfsimilarly under the flow (4–10). We have in this case d¯φ (z, t) = φ(z)−R(t), and (4–10) (thanks to Example 4.2) becomes R˙ = −(n − 1)/R. p Its solution represents the1 evolving set n {ξ ∈ R : φ(ξ) < R(t)}, where R(t) = R02 − 2(n − 1)t for t ∈ [0, 2 R02 /(n − 1)], which disappears for times larger than 12 R02 /(n − 1).
(ii)
The evolution law (4–10) is the gradient flow of M n−1 dφ (∂E). This can be seen, for instance, by using the (nonsymmetric) distance between sets in (4–3), where dist(x, ∂F ) is replaced by distφ (x, ∂F ). For what concerns anisotropic mean curvature flows, we refer to the following (largely incomplete) list of papers: [Hoffman and Cahn 1972], [Angenent 1991b], [Spohn 1993], [Cahn et al. 1993], [Gage 1994], [Giga and Goto 1992] (see also [Giga et al. 1991] for weak solutions to a large class of anisotropic equations). 4.3. On space dependent norms. In this subsection we list some generalization of the previous results. Assume that φ = φ(x, ξ) : R n × R n → [0, +∞[ depends on the position x, φ2 is smooth, φ2 (x, · ) is strictly convex (in the sense of Definition 2.2) and φ satisfies (2–1), (2–2) for any x ∈ R n , with λ independent of
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
65
x. Define φo (x, ξ o ) := sup{hξ, ξ o i : ξ ∈ Bφ (x)}, Bφ (x) := {ξ ∈ R n : φ(x, ξ) ≤ 1}, and dφ as in the first equality of (2–3), with φ(γ, γ) ˙ in place of φ(γ). ˙ T φ and Tφo are defined as in (2–8) taking x fixed. Then (3–3) holds with Iφ(x, ν M (x)) =
1 Hn−1 (Sν (Bφ (x))
in place of Iφ(ν M (x)). The ndimensional dφ Hausdorff measure Hdnφ in R n has the integral representation Z 1 dx, Hdnφ (A) = ωn A Bφ (x) for any Borel set A ⊆ R n . Then (3–5) becomes Z 1 n−1 φo (x, ν M (x)) Mdφ (M ) = ωn dHn−1 (x). B φ (x) M In addition, the function ψ in (3–6) becomes ψ(x, ξ) = f (x)(Iφ) o (x, ξ), where f depends only on x and has the expression 1/(n−1) ωn−1 B(Iφ)o (x) f (x) = . ωn Concerning the first variation of area and φmean curvature, (4–6) of Theorem 4.3 reads as Z 1 d n−1 E M φo (x, ν E )dHn−1 , (∂Eλ )λ=0 = ωn divφ nE φ hX, ∇dφ i dλ dφ B φ (x) ∂E where the operator divφ acts on a smooth vector field v as follows: divφ v := div v + and
1 v, ∇ log Bφ (x)
,
o E nE φ (z) := ∇φ (z, ∇dφ (z)),
∇φo being the gradient of φo with respect to the ξ variable.
5. The Crystalline Case: Lipschitz φRegular Sets In this section we assume that φ is a crystalline norm. The main difficulties when trying to generalize the notion of φcurvature given in (4.5) to the crystalline case are due to the loss of regularity, both of ∂E and of the norm φ o . Observe that the explicit computation of κE φ in (4.5) requires the computation of the hessian of φo which, in the crystalline case, is just a (nonnegative) measure. Recall also that, in the crystalline case, the duality maps are not single valued anymore. This will force us to consider inclusions in place of equalities, and suitable selection principles will be required. Finally, we have to keep in
66
GIOVANNI BELLETTINI
mind that, whichever definition of smoothness we choose, the set ∂B φ must be smooth. Unlike the regular case, in the crystalline case we have to redefine what is a smooth boundary. The idea is to define smoothness of ∂E by requiring the existence of at least one Lipschitz selection of a normal (in a suitable sense) vector field. Before giving formal definitions, we recall some notation. At points x ∈ ∂E where ν E (x) exists we set νφE (x) := ν E (x)/φo (ν E (x)). We indicate by Lip(∂E; R n ) the class of all Lipschitz vector fields defined on the Lipschitz boundary ∂E. Definition 5.1. If E ⊂ R n is Lipschitz we define Norφ (∂E; R n ) := {N ∈ L∞ (∂E; R n ) : N (x)∈ T o (νφE (x)) for Hn−1 a.e. x ∈ ∂E}, Lipν,φ (∂E; R n ) := Lip(∂E; R n ) ∩ Norφ (∂E; R n ). As we shall see, in general smooth sets E in the usual sense do not admit even one element in the class Lipν,φ (∂E; R n ). The best of smoothness that we can hope is described by the following definition [Bellettini and Novaga 1998]: Definition 5.2. Let E ⊂ R n be an open set with compact boundary. We say that E is Lipschitz φregular if ∂E is Lipschitz continuous and there exists a vector field η ∈ Lipν,φ (∂E; R n ). With a small abuse of notation, the pair (E, η) will also denote a Lipschitz φregular set. Observe that, unlike νφE , the vector field η is defined everywhere on ∂E. To better understand the meaning of Definition 5.2, we consider examples. B φo a
a
Bφ
b
b T φo p x
c b
E
q c c
ω Figure 5. The open ball E is not Lipschitz φregular when we endow R 2 with the norm φ(ξ) = max{ξ1 , ξ2 }.
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
67
Example 5.1. Let n = 2 and φ(ξ) := max{ξ1 , ξ2 }, in such a way that Bφ is the square of side 2, see Figure 5. Let E := {z ∈ R 2 : z < 1} be the open unit disk. Then E is not Lipschitz φregular. Indeed, to see that E is Lipschitz φregular, we have to compute T o (νφE (x)), for x ∈ ∂E, and to show that we can produce a vector field η on ∂E which is a Lipschitz selection for the multivalued map x ∈ ∂E → Tφo (νφE (x)). Observe now that T o (νφE (p)) is the upper horizontal segment [a, b] of ∂Bφ ; we depict therefore a corresponding dotted triangle at p. Similarly, Tφo (νφE (q)) is the right vertical segment [b, c] of ∂Bφ , and again we depict the corresponding dotted triangle at q. On the other hand, any point x on ∂E lying in the (relatively) open arc A between p and q is such that Tφo (νφE (x)) = b. We deduce that η ≡ b on A, and η ≡ c on the open arc on ∂E between q and ω. Hence, any vector we choose inside the dotted triangles (for instance, the triangle at q) will produce a discontinuity in the vector field η (at q). We can conclude that the circle, considered in (R 2 , φ), is not Lipschitz φregular, and that it takes the rˆ ole of the square in the usual Euclidean plane. B φo
Bφ Tφo
E
Figure 6. Example of a Lipschitz φregular set E when φ(ξ) = max{ξ1 , ξ2 }. The values of η are uniquely determined at the vertices and on the curved arc of ∂E. Any Lipschitz extension of these values on the interior of the edges, which lies in the dotted triangles, produces a Lipschitz vector field satisfying the required properties (that is, making E Lipschitz φregular). Examples are depicted in Figures 10 and 13.
Example 5.2. Let φ(ξ) := max{ξ1 , ξ2 }. In Figure 6 we show an example of a Lipschitz φregular set E. At the vertices of ∂E the vector νφE is not defined. Let v be a vertex of ∂E, and let F1 and F2 be the two arcs of ∂E having v as a vertex (arcs can also be flat, i.e., segments). For any x in the relative interior of F i , the closed convex set Tφo (νφE (x)) is either a segment or a singleton; in both cases is independent
68
GIOVANNI BELLETTINI
of x and depends only on Fi . Denote it by Ki . The crucial property that makes E Lipschitz φregular is that the intersection ∩ 2i=1 Ki is a singleton, see Figure 6. This produces a unique vector at each vertex of ∂E; at this point, it is easy to realize that we can construct infinitely many vector fields η ∈ Lip ν,φ (∂E; R 2 ) lying inside the dotted triangles with the assigned values at the vertices (see for instance Figures 10 and 13). Tφo
F2
B φo
Bφ
x q
p
F1 F3
E Figure 7. An example of Lipschitz φregular set E for the norm φ whose unit ball is the product of an hexagon with [−1, 1].
Example 5.3. In Figure 7 it is shown an example of Lipschitz φregular set in R 3 when the unit ball of φ is prism with hexagonal basis. Observe that the vector νφE is not defined on the vertices and on the edges of ∂E. Let p ∈ ∂E be a vertex of ∂E and let F1 , F2 , F3 be the three (relatively) closed facets of ∂E having p as a common vertex. For any x in the relative interior of F i , the convex set Tφo (νφE (x)) is a closed facet of ∂Bφ which is independent of x and depends only on Fi . Denote it by Ki . The intersection ∩3i=1 Ki is a singleton (and is the corresponding vertex of ∂Bφ ). In Figure 7 we have depicted such an intersection as a vector at the point p. On the other hand, if F1 and F2 have in common the segment S, and q is a point in the relative interior of S, then ∩ 2i=1 Ki is the corresponding edge of ∂Bφ . We have depicted this set as a triangle. Finally, if x is a point in the relative interior of a facet (for instance, the top facet F ), then νφE (x) coincides with the top vertex of the ∂Bφo , and therefore Tφo (νφ (F )) is the top facet of ∂Bφ , and we have depicted this set on the interior of F as a pyramid. Showing that E is Lipschitz φregular means to exhibit a Lipschitz vector field η : ∂E → R 3 which on the vertices of ∂E is fixed (to be the corresponding vertices of ∂Bφ ), on the relative interior of an edge of ∂E is constrained to lie in the corresponding segment of ∂Bφ , and in the relative interior of a facet of ∂E is constrained to lie in the corresponding facet of ∂B φ . It is at this point easy
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
69
to realize that such a choice can be made (in infinitely many different ways) for the set E in Figure 7. a
a
b
B φo
b
Bφ
C
E Figure 8. A non polyhedral Lipschitz φregular set E.
Example 5.4. In Figure 8 we show an example of Lipschitz φregular set E in R 3 which is not polyhedral. The curved region C is ruled ; if S is any horizontal segment in C (see the dotted lines), any vector field η ∈ Lip ν,φ (∂E; R 3 ) must lie, on S, in the corresponding segment [a, b] of ∂Bφ . In two dimensions the structure of Lipschitz φregular sets E having a finite number of arcs (arcs can be also segments) can be described as follows. The arcs are located in a precise order consistent with φregularity, and are divided into two classes. In the first class there are edges which are parallel to some facet of Bφ and have the same exterior Euclidean normal vector (and we say that the edge corresponds to a facet of Bφ ). The second class consists of the arcs (some of which can be flat) not belonging to the first class, where there is only one possible choice of the vector field η consistent with the φregularity. The arcs of the second class have therefore zero φcurvature (see [Taylor 1993], [Giga and Gurtin 1996]).
6. On Facets of Polyhedral Lipschitz φRegular Sets We have seen in Example 5.2 that, even in two dimensions, if (E, η) is a Lipschitz φregular set, there are in general infinitely many vector fields η ∈ Lipν,φ (∂E; R 2 ) such that (E, η) is Lipschitz φregular. Therefore the divergence of each of these η could be considered as a φmean curvature of ∂E, and in this way the Lipschitz boundary ∂E would have infinitely many different φmean curvatures. This approach could be pursued to some extent; however, we shall
70
GIOVANNI BELLETTINI
see that, among all vector fields satisfying the required constraints, there are some which are distinguished, have the same divergence, and such a uniquely defined divergence is what we can call the φmean curvature of ∂E, at least from the evolutionary point of view. Indeed, this notion of φmean curvature should be identified with the initial velocity of the interface under crystalline mean curvature flow. To simplify notation, in this section we shall consider, when n ≥ 3, only polyhedral Lipschitz φregular sets E with a finite number of facets, that will be understood as (relatively) closed connected (n−1) dimensional flats with Lipschitz (polyhedral) boundary. The symbol F will always denote such a facet. This is surely a restriction, since in general facets can produce curved regions during the flow. 6.1. Some notation. If F ⊂ ∂E is a facet, we denote by ∂F and int(F ) the relative boundary and relative interior of F . An edge of ∂E with vertices p, q will be denoted by [p, q], and its relative interior by ]p, q[. We denote by Π F the affine hyperplane spanned by the facet F . ν(F )
ν˜F
ΠF F ν˜F
ν˜F Figure 9. A facet F . The dotted lines delimit the solid set E having F as a facet.
We define ν(F ) to be the unit normal to int(F ) which points outside of E and we set νφ (F ) := ν(F )/φo (νφ (F )). We indicate by ν˜F the (Hn−2 almost everywhere defined) unit normal to ∂F pointing outside of F ; see Figure 9. Only facets F such that Tφo (νφ (F )) is a facet of Bφ (that is., facets of ∂E corresponding to some facet of ∂Bφ ) will be considered. Definition 6.1. Let (E, η) be a Lipschitz φregular set. We define the trace function cF ∈ L∞ (∂F ) as cF := hη, ν˜F i.
(6–1)
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
71
E
Figure 10. The normal trace of η on the boundary of each onedimensional facet of ∂E is independent of η itself (among all vector fields making E Lipschitz φregular).
Example 6.1. In Figure 10 we depict a vector field η which makes ∂E Lipschitz φregular (Bφ is the square as in Figure 6). Since the values of η are uniquely determined at the vertices of ∂E, the constants c F do not depend on the particular choice of η. The dotted vectors at the vertices indicate the unit normals (in the line containing the facet F ) pointing outward F (that is, ν˜ F ). Bφ F
q E
p
Figure 11. A Lipschitz φregular set when Bφ is the cube.
Example 6.2. Consider the Lipschitz φregular set (E, η) of Figure 11 (B φ is the unit cube). In Figure 12 the bold vectors at the vertices of ∂E are the unique possible values for η. The vector field ν˜F points outside F , and on ]p, q[ points inside E. The pyramids with vertex on the relative interior of the two facets having [p, q] in common represent the corresponding facets of ∂B φ (for instance, Tφo (νφ (F )) for the facet F ), that is, the range of admissibility of η. It follows that cF is negative on ]p, q[, while cF is positive on the remaining relatively open edges of ∂F . Given a Lipschitz φregular set (E, η), in general it is possible to prove that cF does not depend on the choice of η in Lipν,φ (∂E; R n ). More precisely, for
72
GIOVANNI BELLETTINI
Bφ
ν˜F ν˜F
q ν˜F
p ν˜F
F
E
Figure 12. On the relative interior of [p, q] the function cF is negative (and constant).
Hn−1 almost every x ∈ ∂F , we have ( max{hξ, ν˜F (x)i : ξ ∈ Tφo (νφ (F ))} cF (x) = min{hξ, ν˜F (x)i : ξ ∈ Tφo (νφ (F ))}
if ν˜F (x) points outside E, if ν˜F (x) points inside E.
(6–2)
7. φMean Curvature on a Facet In this section we want to define (pointwise almost everywhere) the φmean curvature on a facet F of a polyhedral Lipschitz φregular set (E, η). We need some preliminaries. We let Norφ (F ; ΠF ) := N ∈ L∞ (int(F ); ΠF ) : N (x) ∈ Tφo (νφ (F )) for Hn−1 a.e. x ∈ int(F ) .
It is possible to prove (see [Giga et al. 1998]) that any N ∈ Nor φ (F ; ΠF ) with divN ∈ L2 (int(F )) admits a normal trace hN, ν˜F i on ∂F , for which the Gauss– Green Theorem holds on F (see [Anzellotti 1983]). We set
H(F ; ΠF ) := N ∈ Norφ (F ; ΠF ) : divN ∈ L2 (int(F )), hN, ν˜F i = cF Hn−2 a.e. on ∂F .
Remark 7.1. Thanks to (6–2), the class H(F ; ΠF ) does not depend on the choice of the vector field η making E Lipschitz φregular. We define the functional F( · , F ) : H(F ; ΠF ) → [0, +∞[ as Z F(N, F ) := cn,φ (divN )2 φo (ν E )dHn−1 .
(7–1)
int(F )
The right hand side of (7–1) equals cn,φ φo (ν(F )) flat.
R
F
(divN )2 dHn−1 , since F is
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
73
Our definition of φmean curvature is based on the following result: the minimum problem inf {F(N, F ) : N ∈ H(F ; ΠF )} (7–2) admits a solution, and any two minimizers have the same divergence. F F Denote by Nmin a solution of problem (7–2); since divNmin is independent of F the choice of Nmin among all minimizers of (7–2), we can give the definition of crystalline mean curvature. Definition 7.2. We define the φmean curvature κF φ on the relative interior of F as F κF Hn−1 a.e. x ∈ int(F ). φ (x) := divNmin (x), 2 E F Observe that κF φ is only a function in L (int(F )). We then set κφ := κφ on each facet F of ∂E: it turns out that the orthogonal projection of minimizing vector fields on the orthogonal to ∂F is continuous on ∂F .
Remark 7.3. The minimum problem (7–2), which is at the basis of Definition 7.2, arises when looking at the best way to decrease the M n−1 dφ (∂E) through deformations of the ambient space, precisely in the computation of the first variation n−1 Mn−1 dφ (∂Eλ ) − Mdφ (∂E) lim inf λ λ→0+ n−1 of Mdφ at ∂E. Here, using the notation of Theorem 4.3, we have E λ = Ψλ (E) and Ψλ (x) = x + λX(x), where X is a suitable Lipschitz vector field. The φmean curvature of ∂Bφ is constantly equal to n − 1. Indeed, the vector field x/φ(x) has constant divergence on ∂Bφ , hence it solves the Euler–Lagrange inequality derived from (7–2). We now use the (strict) convexity in the divergence to show that x/φ(x) is actually a minimizer of F( · , F ) on any facet F ⊂ ∂B φ . The following example concerning crystalline curvature of curves is enlightening.
F
E
2 Figure 13. The vector field N E min : ∂E → R is, on each facet F , the linear combination of the values of η at the vertices.
74
GIOVANNI BELLETTINI
Example 7.1. Let n = 2. We compute explicitly the φcurvature of a twodimensional Lipschitz φregular set (E, η). Given a facet F ⊂ ∂E (in this case F equals a segment [z, w]), the minimum problem (7–2) reads as Z (N 0 (s))2 dH1 (s) : N ∈ L2 (]z, w[; Π[z,w] ), N 0 ∈ L2 (]z, w[), inf ]z,w[
N (x) ∈ T
φo
[z,w] (νφ (x))
for a.e. x ∈ ]z, w[, N (z) = cz , N (w) = cw ,
where cz and cw are the orthogonal projections of η(z) and η(w) on the line Π[z,w] , with the correct sign. We now observe that the above minimum problem has a unique solution N F min , which is simply the linear function connecting c z at z with cw at w. Hence, when n = 2, not only the divergence of a minimizer is unique, but also the minimizer itself. If we now repeat this procedure for any facet, and on each facet we add to NF min the proper (constant) normal component to F , we end up with the vector 2 field N E min : ∂E → R whose divergence is the φcurvature of ∂E. An example of this vector field is depicted in Figure 13. Curved regions in ∂E have zero φcurvature. On the other hand, if F is a facet of ∂E ⊂ R 2 and BF ⊂ ∂Bφ is the corresponding facet in ∂Bφ , κF φ is constant on F and κF φ = δF
BF  F 
on int(F ),
(7–3)
where δF ∈ {0, ±1} is a convexity factor: δF = 1, −1 or 0 depending on whether E is locally convex at F , locally concave at F , or neither. In two dimensions (7–3) is used to define the curvature flow of a Lipschitz φregular set (see the references quoted in the Introduction). If ∂E has a finite number of arcs, crystalline curvature flow can be described with a system of ordinary differential equations, since each arc (with nonzero φcurvature) moves in normal direction in the evolution process: it cannot split or curve since, as dictated by (7–3), its normal velocity is constant. On the other hand, arcs or segments with zero φcurvature stay still, and are progressively eaten by the other evolving arcs. When the space dimension n is larger than or equal to 3, the computation of the φmean curvature on a facet is not, in general, an easy problem. As already mentioned in the Introduction, a short time existence theorem of a crystalline mean curvature flow is still missing. Concerning the comparison principle, only an indirect proof is available, in a certain class of crystalline evolutions, see [Bellettini et al. 2000]. Definition 7.4. We say that E is convex at F if E, locally around F , lies on one side of ΠF . The following results show that the φcurvature enjoys some additional regularity properties.
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
75
∞ F Theorem 7.5. κF φ ∈ L (int(F )). Moreover , κφ has bounded variation on int(F ). Finally, if F is convex and E is convex at F , then κF φ is convex on int(F ).
Since the jump set of a function with bounded variation is well defined (see for instance [Ambrosio et al. 2000]), this theorem makes it possible to speak of the jump set of κF φ on int(F ), which should describe, at time zero, where the facet splits under crystalline mean curvature flow. For small times in the evolution problem, F is expected to translate parallel to itself if κ F φ is constant on int(F ) or to bend if κF is continuous but not constant on int(F ). Facets φ with constant φmean curvature have been isolated and studied in [Bellettini et al. 1999], [Bellettini et al. 2001c], where the following notation was introduced. Definition 7.6. We say that F is φcalibrable if κF φ is constant on int(F ). More explicitly, F is φcalibrable provided there exists a vector field N : int(F ) → ΠF which solves the following problem: ∞ N ∈ L (int(F ); ΠF ), N (x) ∈ Tφo (νφ (F )) for Hn−1 a.e. x ∈ int(F ), (7–4) hN, ν˜F i = cFZ Hn−2 a.e. on ∂F, 1 divN = cF dHn−1 . F  ∂F Observe that the constant on the right hand side of the differential equation in (7–4) is determined by using the Gauss–Green theorem on F . The complete characterization of φcalibrable facets F is not yet available. We conclude the paper by pointing out some known results in this direction.
7.1. Characterization of φcalibrable facets in special cases. Assume that n = 3. Let BF ⊂ ∂Bφ be the facet corresponding to F . If necessary, we identify BF with its orthogonal projection on the plane parallel to Π F passing through the origin of R 3 . We also assume that BF contains the origin in its interior and that it is symmetric (this latter assumption can be weakened). Therefore ˜ BF can be considered as the unit ball of a norm (in R 2 ), which we denote by φ. F ˜ ˜ We assume that F is Lipschitz φregular. Denote by κφ˜ the φcurvature of ∂F o ˜ The following result holds. and by φ˜ the dual of φ. Theorem 7.7. Let n = 3. Assume that F is convex and that E is convex at F . Then F is φcalibrable if and only if Z 1 φ˜o (˜ ν F ) dH1 . (7–5) sup κF ≤ ˜ φ F  ∂F ∂F ∞ The sup in (7–5) is an essential supremum, since κ F ˜ is a function in L (∂F ). φ Hence, under the assumptions of Theorem 7.7, problem (7–4) is solvable if and ˜ only if the φcurvature of ∂F is bounded above by the constant on the right
76
GIOVANNI BELLETTINI
hand side of (7–5); this means, roughly speaking, that the edges of ∂F cannot be too “short”. Finally, let us mention that examples of facets which are not φcalibrable are given in [Bellettini et al. 1999], and that the problem of calibrability when B φ is a portion of a cylinder (hence not in a crystalline setting) has been recently considered, under rather mild assumptions, in the papers [Bellettini et al. 2002] and [Bellettini et al. 2003].
8. Concluding Remarks Fix a norm φ on R n and denote by dφ the distance induced by φ. As a starting point of our approach let us consider the (n−1)dimensional measure Mn−1 dφ , defined as in (1–1) on compact and sufficiently smooth boundaries ∂E of solid sets E. Such a notion is called the φMinkowski content of the manifold ∂E, and is a geometric invariant under isometries of the ambient space (because the ndimensional Hausdorff measure Hdnφ with respect to dφ is invariant and the tubular neighborhoods are computed with respect to the distance d φ ). On the other hand cn,φ is not invariant (recall that ωn is a normalizing constant). We recall that, in the generic (finite dimensional) normed space (R n , φ), there are other meaningful notions of surface measure, such as for instance the (n−1)dimensional Hausdorff measure Hdn−1 with respect to dφ , the Holmes–Thompson φ measure and the measure considered in [De Giorgi 1995]. Even in n = 2 dimensions, there are examples of norms φ on R 2 for which M1dφ and Hd1φ are different. Roughly speaking, this can be explained as follows. M n−1 dφ (∂E) is constructed by taking the projections of Bφ := {φ ≤ 1} onto the (onedimensional) normal spaces to ∂E. On the other hand, Hdn−1 (∂E) is constructed by taking the interφ sections of Bφ with the ((n−1)dimensional) tangent spaces to ∂E. We notice that, in any case, Hdn−1 can be seen as the Minkowski content with respect to φ another norm ψ. Beside its geometric interest, our choice of working with M n−1 is motivated dφ also by the physics of phase transitions, where it happens that some relevant phenomena are concentrated in a very thin tubular neighborhood of the interface (sometimes called diffuse interface), and lead in the limit to the Minkowski content. The next step in our approach consists in the definition of the φmean curvature κE φ of ∂E. This concept, which as usual depends also on the immersion of the manifold, is obtained by computing the first variation of M n−1 on ∂E, and dφ is identified with the normal velocity of the initial datum ∂E = ∂E(0) under the anisotropic mean curvature flow. The computation of the first variation of Mn−1 is rather direct when ∂Bφ is dφ smooth and all its principal curvatures are strictly positive (regular case), but becomes more involved in the crystalline case (that is, when B φ a polytope). In
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
77
this latter situation, assuming that ∂E is polyhedral and has a geometry locally resembling the geometry of ∂Bφ , we compute κE φ on its facets, by applying a suitable minimization principle. We then discuss the problem of characterizing φcalibrable facets of ∂E in n = 3 dimensions, that is, those facets F ⊂ ∂E for which κE φ is constant on the relative interior int(F ) of F . We conclude by recalling that (a scalar multiple of) B φ is a solution of the socalled isoperimetric problem, that is, the problem of minimizing M n−1 dφ (∂E) among all finite perimeter sets E with Hdnφ (E) fixed. This is in agreement with the fact that ∂Bφ has constant φmean curvature, precisely equal to n − 1.
References [Allen and Cahn 1979] S. M. Allen and J. W. Cahn, “A macroscopic theory for antiphase boundary motion and its application to antiphase domain coarsening”, Acta Metall. Mater. 27 (1979), 1085–1095. [Almgren and Taylor 1995] F. J. Almgren and J. E. Taylor, “Flat flow is motion by crystalline curvature for curves with crystalline energies”, J. Differential Geom. 42 (1995), 1–22. [Almgren et al. 1993] F. J. Almgren, J. E. Taylor and L. Wang, “Curvaturedriven flows: a variational approach”, SIAM J. Control Optim. 31 (1993), 387–437. [Altschuler et al. 1995] S. Altschuler, S. Angenent and Y. Giga, “Mean curvature flow through singularities for surfaces of rotation”, J. Geom. Anal. 5 (1995), 293–358. [Amar et al. 1998] M. Amar, G. Bellettini and S. Venturini, “Integral representation of functionals defined on curves of W 1,p ”, Proc. Roy. Soc. Edinb. 128A (1998), 193–217. [Ambrosio 1999] L. Ambrosio, “Lecture notes on geometric evolution problems, distance function and viscosity solutions”, pp. 5–94 in Calculus of variations and partial differential equations: topics on geometrical evolution problems and degree theory, edited by G. Buttazzo et al., Springer, Berlin, 1999. [Ambrosio et al. 2000] L. Ambrosio, N. Fusco and D. Pallara, Functions of bounded variation and free discontinuity problems, Clarendon Press, Oxford, 2000. [Ambrosio and Kirchheim 2000] L. Ambrosio and B. Kirchheim, “Currents in metric spaces”, Acta Math. 185 (2000), 1–80. [Ambrosio et al. 2002] L. Ambrosio, M. Novaga and E. Paolini, “Some regularity results for minimal crystals”, ESAIM Control Optim. Calc. Var. 8 (2002), 69–103. [Angenent 1991a] S. Angenent, “On the formation of singularities in the curve shortening flow”, J. Differential Geom. 33 (1991) 601–633. [Angenent 1991b] S. Angenent, “Parabolic equations for curves and surfaces Part II. Intersections, blowup and generalized solutions”, Ann. of Math. 133 (1991), 171– 215. [Angenent et al. 1995] S. Angenent, D. L. Chopp and T. Ilmanen, “A computed example of nonuniqueness of mean curvature flow in R 3 ”, Comm. Partial Differential Equations 20 (1995), 1937–1958.
78
GIOVANNI BELLETTINI
[Anzellotti 1983] G. Anzellotti, “Pairings between measures and bounded functions and compensated compactness”, Ann. Mat. Pura Appl. 135 (1983), 293–318. [Bao et al. 2000] D. Bao, S.S. Chern, and Z. Shen, An introduction to Riemann–Finsler geometry, Graduate Texts in Mathematics 200, Springer, New York, 2000. [Barles et al. 1993] G. Barles, H.M. Soner and P. E. Souganidis, “Front propagation and phase field theory”, SIAM J. Control Optim. 31 (1993), 439–469. [Bellettini and Fragal` a 2002] G. Bellettini and I. Fragala, “Elliptic approximations of prescribed mean curvature surfaces in Finsler geometry”, Asymptot. Anal. 22 (2002), 87–111. [Bellettini and Novaga 1998] G. Bellettini and M. Novaga, “Approximation and comparison for nonsmooth anisotropic motion by mean curvature in R N ”, Math. Mod. Meth. Appl. Sc. 10 (2000), 1–10. [Bellettini and Paolini 1994] G. Bellettini and M. Paolini, “Two examples of fattening for the curvature flow with a driving force”, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. 5 (1994), 229–236. [Bellettini and Paolini 1995] G. Bellettini and M. Paolini, “Some results on minimal barriers in the sense of De Giorgi applied to driven motion by mean curvature”, Rend. Atti Acc. Naz. Sci. XL Mem. Mat. XIX (1995), 43–67. Errata in XXVI (2002), 161–165. [Bellettini and Paolini 1996] G. Bellettini and M. Paolini, “Anisotropic motion by mean curvature in the context of Finsler geometry”, Hokkaido Math. J. 25 (1996), 537– 566. [Bellettini et al. 1996] G. Bellettini and M. Paolini and S. Venturini, “Some results on surface measures in Calculus of Variations”, Ann. Mat. Pura Appl. (4) IV (1996), 329–359. [Bellettini et al. 1999] G. Bellettini, M. Novaga and M. Paolini, “Facetbreaking for threedimensional crystals evolving by mean curvature”, Interfaces Free Bound. 1 (1999), 39–55. [Bellettini et al. 2000] G. Bellettini, R. Goglione and M. Novaga, “Approximation to driven motion by crystalline curvature in two dimensions”, Adv. Math. Sci. Appl. 10 (2000), 467–493. [Bellettini et al. 2001a] G. Bellettini, M. Novaga and M. Paolini, “On a crystalline variational problem, part I: first variation and global L∞ regularity”, Arch. Rational Mech. Anal. 157 (2001), 165–191. [Bellettini et al. 2001b] G. Bellettini, M. Novaga and M. Paolini, “On a crystalline variational problem, part II: BV regularity and structure of minimizers on facets”, Arch. Rational Mech. Anal. 157 (2001), 193–217. [Bellettini et al. 2001c] G. Bellettini, M. Novaga and M. Paolini, “Characterization of facets–breaking for nonsmooth mean curvature flow in 3D in the convex case”, Interfaces Free Bound. 3 (2001), 415–446. [Bellettini et al. 2002] G. Bellettini, V. Caselles and M. Novaga, “The total variation flow in R n ”, J. Diff. Equations 184 (2002), 475–525. [Bellettini et al. 2003] G. Bellettini, V. Caselles and M. Novaga, “Explicit solutions of the eigenvalue problem −div(Du/Du) = u”, preprint, Univ. Pisa, 2003.
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
79
[Brakke 1978] K. A. Brakke, The motion of a surface by its mean curvature, Princeton Univ. Press, Princeton, 1978. [Brezis 1973] H. Brezis, Op´erateurs maximaux monotones et semigroupes de contractions dans les espaces de Hilbert, NorthHolland, Amsterdam, 1973. [Busemann 1947] H. Busemann, “Intrinsic area”, Ann. Math. 48 (1947), 234–267. [Busemann 1949] H. Busemann, “A theorem on convex bodies of the Brunn–Minkowski type”, Proc. Nat. Acad. Sci. USA 35 (1949), 27–31. [Cahn and Hoffman 1974] J. W. Cahn and D. W. Hoffman, “A vector thermodynamics for anisotropic interfaces, 2: curved and faceted surfaces”, Acta Metall. Mater. 22 (1974), 1205–1214. [Cahn et al. 1992] J. W. Cahn, C. A. Handwerker and J. E. Taylor, “Geometric models of crystal growth”, Acta Metall. Mater. 40 (1992), 1443–1474. [Cahn et al. 1993] J. W. Cahn and C. A. Handwerker, “Equilibrium geometries of anisotropic surfaces and interfaces”, Mat. Sci. Eng. A162 (1993), 83–95. [Chen et al. 1991] Y.G. Chen, Y. Giga and S. Goto, “Uniqueness and existence of viscosity solutions of generalized mean curvature flow equations”, J. Differential Geom. 3 (1991), 749–786. [De Cecco and Palmieri 1993] G. De Cecco and G. Palmieri, “Distanza intrinseca su una variet` a finsleriana di Lipschitz”, Rend. Atti Acc. Naz. Sci. XL Mem. Mat. (5) 17 (1993), 129–151. [De Cecco and Palmieri 1995] G. De Cecco and G. Palmieri, “Lip manifolds: from metric to Finslerian structure”, Math. Z. 218 (1995), 223–237. [De Giorgi 1989] E. De Giorgi, “Su alcuni problemi comuni all’Analisi e alla Geometria”, Note di Matematica , IX Suppl., (1989), 59–71. [De Giorgi 1990] E. De Giorgi, “Conversazioni di matematica”, Quaderni Dip. Mat. Univ. Lecce 2 (1990). [De Giorgi 1993] E. De Giorgi, “Congetture riguardanti barriere, superfici minime, movimenti secondo la curvatura media”, manuscript, Lecce, November 4, 1993. [De Giorgi 1994] E. De Giorgi, “Barriers, boundaries, motion of manifolds”, Conference held at Department of Mathematics of Pavia, March 18, 1994. [De Giorgi 1995] E. De Giorgi, Problema di Plateau generale e funzionali geodetici, Atti Sem. Mat. Fis. Univ. Modena 43 (1995), 285–292. [Ecker and Huisken 1989] K. Ecker and G. Huisken, “Mean curvature flow of entire graphs”, Ann. of Math. (2) 130 (1989), 453–471. [Ecker and Huisken 1991] K. Ecker and G. Huisken, “Interior estimates for hypersurfaces moving by mean curvature”, Invent. Math. 103 (1991), 547–569. [Evans and Spruck 1991] L. C. Evans and J. Spruck, “Motion of level sets by mean curvature I”, J. Differential Geom. 33 (1991), 635–681. [Evans and Spruck 1992a] L. C. Evans and J. Spruck, “Motion of level sets by mean curvature II”, Trans. Amer. Math. Soc. 330 (1992), 321–332. [Evans and Spruck 1992b] L. C. Evans and J. Spruck, “Motion of level sets by mean curvature III”, J. Geom. Anal. 2 (1992), 121–160.
80
GIOVANNI BELLETTINI
[Evans and Spruck 1995] L. C. Evans and J. Spruck, “Motion of level sets by mean curvature IV”, J. Geom. Anal. 5 (1995), 77–114. [Federer 1969] H. Federer, Geometric measure theory, Springer, Berlin, 1969. [Fierro and Paolini 1996] F. Fierro and M. Paolini, “Numerical evidence of fattening for the mean curvature flow”, Math. Models Methods Appl. Sci. 6 (1996), 793–813. [Fukui and Giga 1996] T. Fukui and Y. Giga, “Motion of a graph by nonsmooth weighted curvature”, pp. 47–56 in World Congress of Nonlinear Analysis ’92, edited by V. Lakshmikantham, de Gruyter, Berlin, 1996. [Gage 1994] M. E. Gage, “Evolving plane curves by curvature in relative geometries, II”, Duke Math. J. 75 (1994), 79–98. [Gage and Hamilton 1986] M. E. Gage and R. S. Hamilton, “The heat equation shrinking convex plane curves”, J. Differential Geom. 23 (1986), 69–96. [Giga and Giga 1997] M.H. Giga and Y. Giga, “Remarks on convergence of evolving graphs by nonlocal curvature”, pp. 99–116 in Progress in partial differential equations (Pont´ aMousson, 1997), vol. 1, edited by H. Amann et al., Pitman research notes in mathematics 383, Longman, Harlow (Essex), 1998. [Giga and Giga 1998a] M.H. Giga and Y. Giga, “A subdifferential interpretation of crystalline motion under nonuniform driving force”, pp. 276–287 in Dynamical systems and differential equations (Springfield, MO, 1996), vol. 1, edited by Wenxiong Chen and Shouchuan Hu, Discrete Contin. Dynam. Systems, added volume I (1998). [Giga and Giga 1998b] M.H. Giga and Y. Giga, “Evolving graphs by singular weighted curvature”, Arch. Rational Mech. Anal. 141 (1998), 117–198. [Giga and Giga 1999] M.H. Giga and Y. Giga, “Stability of evolving graphs by nonlocal weighted curvature”, Comm. Partial Differential Equations 24 (1999), 109–184. [Giga and Giga 2000] M.H. Giga and Y. Giga, “Crystalline and level set flowconvergence of a crystalline algorithm for a general anisotropic curvature flow in the plane”, pp. 64–79 in Free boundary problems: theory and applications (Chiba, 1999), vol. I, edited by N. Kenmochi, GAKUTO Intern. Ser. Math. Appl. 13, Gakk¯ otosho, Tokyo, 2000. [Giga and Goto 1992] Y. Giga and s. Goto, “Geometric evolutions of phaseboundaries”, pp. 51–66 in On the evolution of phaseboundaries, edited by M. Gurtin and Mc Fadden, IMA Vols. Math. Appl. 43, Springer, Berlin, 1992. [Giga and Gurtin 1996] Y. Giga and M. E. Gurtin, “A comparison theorem for crystalline evolutions in the plane”, Quart. Appl. Math. 54 (1996), 727–737. [Giga et al. 1991] Y. Giga, S. Goto, H. Ishii and M.H. Sato, “Comparison principle and convexity preserving properties for singular degenerate parabolic equations on unbounded domains”, Indiana Univ. Math. J. 40 (1991), 443–470. [Giga et al. 1998] Y. Giga, M. E. Gurtin and J. Matias, “On the dynamics of crystalline motion”, Japan J. Indust. Appl. Math. 15 (1998), 7–50. [Girao 1995] P. M. Girao, “Convergence of a crystalline algorithm for the motion of a simple closed convex curve by weighted curvature”, SIAM J. Numer. Anal. 32 (1995), 886–899. [Girao and Kohn 1994] P. M. Girao and R. V. Kohn, “Convergence of a crystalline algorithm for the heat equation in one dimension and for the motion of a graph”, Numer. Math. 67 (1994), 41–70.
ANISOTROPIC AND CRYSTALLINE MEAN CURVATURE FLOW
81
[Giusti 1984] E. Giusti, Minimal surfaces and functions of bounded variation, Birkh¨ auser, Boston, 1984. [Grayson 1987] M. A. Grayson, “The heat equation shrinks embedded plane curves to round points”, J. Differential Geom. 26 (1987), 285–314. [Hoffman and Cahn 1972] D. W. Hoffman and J. W. Cahn, “A vector thermodynamics for anisotropic surfaces, I: fundamentals and application to plane surface junctions”, Surf. Sci. 31 (1972), 368–388. [Huisken 1984] G. Huisken, “Flow by mean curvature of convex surfaces into spheres”, J. Differential Geom. 20 (1984), 237–266. [Ilmanen 1992] T. Ilmanen, “Generalized flow of sets by mean curvature on a manifold”, Indiana Univ. Math. J. 41 (1992), 671–705. [Ilmanen 1993a] T. Ilmanen, “The level set flow on a manifold”, pp. 193–204 in Differential geometry: partial differential equations on manifolds (Los Angeles, 1990), edited by Robert Greene and S.T. Yau, Proc. Sympos. Pure Math. 54, Part 1, Amer. Math. Soc., Providence, RI, 1993. [Ilmanen 1993b] T. Ilmanen, “Convergence of the Allen–Cahn equation to Brakke’s motion by mean curvature”, J. Differential Geom. 38 (1993), 417–461. [Ilmanen 1994] T. Ilmanen, “Elliptic regularization and partial regularity for motion by mean Curvature”, Mem. Amer. Math. Soc. 108 (1994). [Ishii and Soner 1999] K. Ishii and H. M. Soner, “Regularity and convergence of crystalline motion”, SIAM J. Math. Anal. 30 (1999), 19–37. [Mugnai 2003] L. Mugnai, “Relaxation and variational approximation for curvaturedepending functionals in 2D”, Ph.D. Thesis, Univ. of Pisa, 2003. [Osher and Sethian 1988] S. Osher and J. A. Sethian, “Fronts propagating with curvature dependent speed: algorithms based on Hamilton–Jacobi formulations”, J. Comput. Phys. 79 (1988), 12–49. [Paolini and Pasquarelli 2000] M. Paolini and F. Pasquarelli, “Numerical simulations of crystalline curvature flow in 3D by interface diffusion”, pp. 376–389 in: Free boundary problems: theory and applications (Chiba, 1999), vol. II, edited by N. Kenmochi, GAKUTO Intern. Ser. Math. Sci. Appl. 14 Gakk¯ otosho, Tokyo, 2000. [Paolini and Verdi 1992] M. Paolini and C. Verdi, “Asymptotic and numerical analyses of the mean curvature flow with a spacedependent relaxation parameter”, Asymptot. Anal. 5 (1992), 553–574. [Rockafellar 1972] R. T. Rockafellar, Convex analysis, Princeton Univ. Press, Princeton, 1972. [Roosen and Taylor 1994] A. R. Roosen and J. Taylor, “Modeling crystal growth in a diffusion field using fully faceted interfaces”, J. Comput. Phys. 114 (1994), 444451. [Rybka 1997] P. Rybka, “A crystalline motion: uniqueness and geometric properties”, SIAM J. Appl. Math. 57 (1997), 53–72. [Schneider 1993] R. Schneider, Convex Bodies: the Brunn–Minkowski theory, Encyclopedia of Math. and its Appl. 4, Cambridge Univ. Press, New York, 1993. [Shen 1998] Z. Shen, “On Finsler geometry of submanifolds”, Math. Ann. 311 (1998), 549–576. [Shen 2001] Z. Shen, Lectures on Finsler geometry, World Scientific, Singapore, 2001.
82
GIOVANNI BELLETTINI
[Soner 1993] H.M. Soner, “Motion of a set by the curvature of its boundary”, J. Differential Equations 101 (1993), 313–372. [Soner and Souganidis 1993] H.M. Soner and P. E. Souganidis, “Singularities and uniqueness of cylindrically symmetric surfaces moving by mean curvature”, Comm. Partial Differential Equations 18 (1993), 859–894. [Spohn 1993] H. Spohn, “Interface motion in models with stochastic dynamics”, J. Stat. Phys. 71 (1993), 1081–1132. [Stancu 1996] A. Stancu, “Uniqueness of selfsimilar solutions for a crystalline flow”, Indiana Univ. Math. J. 45 (1996), 1157–1174. [Taylor 1978] J. E. Taylor, “Crystalline variational problems”, Bull. Amer. Math. Soc. (N. S.) 84 (1978), 568–588. [Taylor 1986] J. E. Taylor, “Complete catalog of minimizing embedded crystalline cones”, pp. 379–403 in Geometric measure theory and the calculus of variations (Arcata, CA, 1984), edited by William K. Allard and Frederick J. Almgren, Jr., Proc. Symposia Math. 44, Amer. Math. Soc., Providence, 1986. [Taylor 1991] J. E. Taylor, “Constructions and conjectures in crystalline nondifferential geometry”, pp. 321–336 in Differential geometry: a symposium in honour of Manfredo do Carmo, edited by Blaine Lawson and Keti Tenenblat, Pitman monographs and surveys in pure and applied mathematics 52, Longman, Harlow (Essex), 1991. [Taylor 1992] J. E. Taylor, “Mean curvature and weighted mean curvature”, Acta Metall. Mater. 40 (1992), 1475–1485. [Taylor 1993] J. E. Taylor, “Motion of curves by crystalline curvature, including triple junctions and boundary points”, pp. 417–438 in Differential Geometry: partial differential equations on manifolds (Los Angeles, 1990), edited by Robert Greene and S.T. Yau, Proc. Sympos. Pure Math. 54, Part 1, Amer. Math. Soc., Providence, RI, 1993. [Thompson 1996] A. C. Thompson, Minkowski geometry, Encyclopedia of Math. and its Appl. 63, Cambridge Univ. Press, New York, 1996. [Venturini 1992] S. Venturini, “Derivation of distance functions on R n ”, preprint, 1992. [White 1995] B. White, “The topology of hypersurfaces moving by mean curvature”, Comm. Anal. Geom. 3 (1995), 317–333. [White 2000] B. White, “The size of the singular set in mean curvature flow of meanconvex sets”, J. Amer. Math. Soc. 13 (2000), 665–695. [White 2003] B. White, “The nature of singularities in mean curvature flow”, J. Amer. Math. Soc. 16 (2003), 123–138. [Yunger 1998] J. Yunger, “Facet stepping and motion by crystalline curvature”, Ph.D. thesis, Rutgers University, 1998. Giovanni Bellettini Dipartimento di Matematica ` di Roma “Tor Vergata” Universita Via della Ricerca Scientifica 00133 Roma Italy
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Finsler Geometry on Complex Vector Bundles TADASHI AIKOU
Contents 1. Introduction 2. Ampleness 2.1. Ample line bundles 2.2. Ample vector bundles 3. Finsler Connections 3.1. Finsler connection 3.2. Curvature 3.3. Holomorphic sectional curvature 4. Ruled Manifolds 4.1. Projective bundle 4.2. Ruled manifolds References
83 85 85 88 91 92 94 97 99 99 102 103
1. Introduction A Finsler metric of a manifold or vector bundle is defined as a smooth assignment for each base point a norm on each fibre space, and thus the class of Finsler metrics contains Riemannian metrics as a special subclass. For this reason, Finsler geometry is usually treated as a generalization of Riemannian geometry. In fact, there are many contributions to Finsler geometry which contain Riemannian geometry as a special case (see e.g., [Bao et al. 2000], [Matsumoto 1986], and references therein). On the other hand, we can treat Finsler geometry as a special case of Riemannian geometry in the sense that Finsler geometry may be developed as differential geometry of fibred manifolds (e.g., [Aikou 2002]). In fact, if a Finsler metric in the usual sense is given on a vector bundle, then it induces a Riemannian inner product on the vertical subbundle of the total space, and thus, Finsler geometry is translated to the geometry of this Riemannian vector bundle. It is natural to question why we need Finsler geometry at all. To answer this question, we shall describe a few applications of complex Finsler geometry to some subjects which are impossible to study via Hermitian geometry. 83
84
TADASHI AIKOU
The notion of complex Finsler metric is old and goes back at least to Carath´eodory who introduced the socalled Carath´eodory metric. The geometry of complex Finsler manifold, via tensor analysis, was started by [Rizza 1963], and afterwards, the connection theory on complex Finsler manifolds has been developed by [Rund 1972], [Icijy¯ o 1994], [Fukui 1989], and [Cao and Wong 2003], etc.. Recently, from the viewpoint of the geometric theory of several complex variables, complex Finsler metric has become an interesting subject. In particular, an intrinsic metric on a complex manifold, namely the Kobayashi metric, is a holomorphic invariant metric on a complex manifold. The Kobayashi metric is, by its definition, a pseudo Finsler metric. However, by the fundamental work of [Lempert 1981], the Kobayashi metric on a smoothly bounded strictly convex domain in Cn is a smooth pseudoconvex Finsler metric. The interest in complex Finsler geometry also arises from the study of holomorphic vector bundles. The characterization of ample (or negative) vector bundles due to Kobayashi [Kobayashi 1975] shows the importance of Finsler geometry. In fact, he has proved that E is ample if and only if its dual E ∗ admits a “negatively curved” pseudoconvex Finsler metric (Theorem 3.2). The meaning of the term “negatively curved” is defined by using the curvature tensor of the Finsler connection on a Finsler bundle (E, F ). Another example of interest in complex Finsler geometry arises from the geometry of geometrically ruled surfaces X . A geometrically ruled surface X is, by definition (see [Yang 1991]), an algebraic surface with a holomorphic projection φ : X → M , M a compact Riemann surface, such that each fibre is isomorphic to the complex projective line P1 . Every geometrically ruled surface is isomorphic to P(E) for some holomorphic vector bundle π : E → M of rank(E) = 2. Then, every geometrically ruled surface X = P(E) is also a compact K¨ ahler manifold by Lemma 6.37 in [Shiffmann and Sommese 1985], and any K¨ ahler metric g X on X induces a Finsler metric F , which is not a Hermitian metric in general, on the bundle E. Thus the geometry of (X , gX ) is translated to the geometry of the Finsler bundle (E, F ). In general, an algebraic curve (or polarized manifold) ϕ : X → P N has a K¨ ahler metric ωX = ϕ∗ ωF S induced from the Fubini–Study metric ωF S on PN . An interesting subject in complex geometry is to investigate how metrics of this kind are related to constant curvature metrics, and moreover, it is interesting to investigate how constant scalar curvature metrics should be related to algebrogeometric stability. LeBrun [LeBrun 1995] has investigated minimal ruled surfaces X = P(E) over a compact Riemann surface M of genus g(M ) ≥ 2 with constant scalar curvature. He showed that, roughly speaking, X admits such a K¨ ahler metric gX if and only if the bundle E is semistable in the sense of Mumford–Takemoto. By the statement above, an arbitrary K¨ ahler metric on a minimal ruled surface X determines a Finsler metric F on E by the identity √ ahler form ωX . The geometry of such a minimal ωX = −1 ∂ ∂¯ log F for the K¨
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
85
ruled surface can also be investigated by the study of the Finsler bundle (E, F ) (see [Aikou 2003b]). In this article, we shall report on the geometry of complex vector bundles with Finsler metrics, i.e., Finsler bundles. Let F be a Finsler metric on a holomorphic vector bundle π : E → M over a complex manifold M . The geometry of a Finsler bundle (E, F ) is the study of the vertical bundle VE = ker π∗ with a Hermitian metric GVE induced from the given Finsler metric. The main tool of the investigation in Finsler geometry is the Finsler connection. The connection is a unique one on the Hermitian bundle (V E , gVE ), satisfying some geometric condition (see definition below). Although it is natural to investigate (VE , gVE ) by using the Hermitian connection of (VE , gVE ), it is convenient to use the Finsler connection for investigating some special Finsler metrics. For example, the flatness of the Hermitian connection of (V E , gVE ) implies that the Finsler metric F is reduced to a flat Hermitian metric. However, if the Finsler connection is flat, then the metric F belongs to an important class, the socalled locally Minkowski metrics (we simply call these special metrics flat Finsler metrics). If the Finsler connection is induced from a connection on E, then the metric F belongs to another important class, the socalled Berwald metrics (sometimes a Berwald metric is said to be modeled on a Minkowski space). In this sense, the big difference between Hermitian geometry and Finsler geometry is the connection used for the investigation of the bundle (V E , gVE ).
2. Ampleness 2.1. Ample line bundles. Let L be a holomorphic line bundle over a compact complex manifold M . We denote by O(L) the sheaf of germs of holomorphic sections of L. Since M is compact, dimC H 0 (M, O(L)) is finite. Let {f 0 , . . . , f N } be a set of linearly independent sections of L, from the complex vector space of global sections. The vector space spanned by these sections is called a linear system on M . If the vector space consists of all global sections of L, it is called a complete linear system on M . Using these sections {f 0 , . . . , f N }, a rational map ϕL : M → PN is defined by ϕL (z) = [f 0 (z) : · · · : f N (z)].
(2.1)
This rational map is defined on the open set in M which is complementary to the common zeroset of the sections f i (0 ≤ i ≤ N ). It is verified that the rational map ϕ˜L obtained from another basis {f˜0 , . . . , f˜N } is transformed by an automorphism of PN . Definition 2.1. A line bundle L over M is said to be very ample if the rational map ϕL : M → PN determined by its complete linear system L is an embedding. L is said to be ample if there exists some integer m > 0 such that L ⊗m is very ample.
86
TADASHI AIKOU
Let L be a very ample line bundle over a compact complex manifold M , and {f 0 , . . . , f N } a basis of H 0 (M, O(L)) which defines an embedding ϕL : M → PN . Embedding M into PN , we may consider the line bundle as the hyperplane bundle over M ⊂ PN . We define an open covering {U(j) } of M by U(j) = z ∈ M : f j (z) 6= 0 . With respect to this covering, the local trivialization ϕ j α i ). The transition cocycle {ljk : over U(j) × C is given by ϕj (f i ) = (z(j) , f(j) ∗ U(j) ∩ U(k) → C } is given by ljk (z) =
i f(k) i f(j)
.
(2.2)
Let {hjk } be the transition cocycle of the hyperplane bundle H with respect to the standard covering {U(j) } of PN . Then, {hjk } is given by hjk = ξ k /ξ j in terms of the homogeneous coordinate system [ξ 0 : · · · : ξ N ] of PN . Since (2.2) implies i f(k) hjk ◦ ϕL = i = ljk , f(j) we obtain L = ϕ∗L H. Lemma 2.1. Let L be a very ample line bundle over a complex manifold M . Then L is isomorphic to the pullback bundle ϕ∗L H of the hyperplane bundle H over the target space PN of ϕL . Example 2.1. (1) The hyperplane bundle H over PN is very ample. (2) Let E be a holomorphic line bundle and L an ample line bundle over a compact complex manifold M . For some sufficiently large integer k, the line bundle E ⊕ L⊗k is very ample (see [Griffith and Harris 1978, p. 192]). As we can see from the above, it is an algebrogeometric issue to determine whether a holomorphic line bundle is ample or not. However, the Kodaira embedding theorem provides a differential geometric way to check ampleness; see Theorem 2.1 and Proposition 2.1 below. The key idea is to relate ampleness to the notion of positivity, defined as follows. Definition 2.2. A holomorphic line bundle L is said to be positive if its Chern class c1 (L) ∈ H 2 (M, R) is represented by a positive real (1, 1)form. A holomorphic line bundle L is said to be negative if its dual L ∗ is positive. Since c1 (L∗ ) = −c1 (L), the holomorphic line bundle L is negative if c 1 (L) is represented by a negative real (1, 1)form. A Hermitian metric g on L is given by the family g(j) of local positive functions g(j) : U(j) → R, satisfying g(k) = ljk 2 g(j) on U(j) ∩U(k) for the transition cocycle ¯ log g(j) = ∂∂ ¯ log g(k) , and thus {ljk } of L. Since ljk are holomorphic, we have ∂∂ ¯ log g(j) } defines a global (1, 1)form on M , which will be denoted by ∂∂ ¯ log g, {∂∂ and is just the curvature form of (L, g). The Chern form c 1 (L, g) defined by √ ¯ log g is a representative of the Chern class c1 (L). c1 (L, g) = −1 ∂∂
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
87
By the definition above, a holomorphic line bundle L is positive if and only if L admits a Hermitian metric g whose Chern form c1 (L, g) is positivedefinite. A compact complex manifold M is called a Hodge manifold if there exists a positive line bundle L over M . If M is a Hodge manifold, then there exists a Hermitian line bundle (L, g) whose Chern form c1 (L, g) is positivedefinite, and thus c1 (L, g) defines a K¨ ahler metric on M . Consequently, every Hodge manifold is K¨ ahler. N The hyperplane In fact, if we define a function 0 bundleNH overN P jis positive. g(j) on V(j) = [ξ : · · · : ξ ] ∈ P : ξ 6= 0 by ξ j 2 g(j) = P k 2 , ξ 
(2.3)
the family {g(j) }j=0,...,N satisfies g(k) = hjk 2 g(j) on V(j) ∩ V(k) , and thus it determines a Hermitian metric gH on H. Then we have √ −1 ¯ c1 (H) = ∂∂ log gH > 0. (2.4) 2π The closed real (1, 1)form representing c1 (H) induces a K¨ ahler metric on PN , which is called the Fubini–Study metric gF S with the K¨ ahler form √ 2 ωF S = −1 ∂ ∂¯ log kξk , P 2 where we put kξk = ξ i 2 . The following wellknown theorem shows that every Hodge manifold M is algebraic, i.e., M is holomorphically embedded in a projective space P N . Theorem 2.1 (Kodaira’s embedding theorem). Let L be a holomorphic line bundle over a compact complex manifold M . If L is positive, then it is ample, i .e., there exists some integer n0 > 0 such that for all m ≥ n0 the map ϕL⊗m  : M → PN is a holomorphic embedding. Conversely, we suppose that L is ample. Then, by definition, there exists a basis {f 0 , . . . , f N } of H 0 (M, O(L⊗m )) such that ϕL⊗m  : M → PN defined by (2.1) is an embedding. By Lemma 2.1, the line bundle L⊗m is identified with ϕ∗L⊗m  H. Thus L⊗m admits a Hermitian metric g = ϕ∗L⊗m  gH , and c1 (L⊗m ) is given by √ −1 ¯ ∂∂ log g . c1 (L⊗m ) = mc1 (L) = 2π √ ¯ log g is positive, and thus Since H is positive, the (1, 1)form −1 ∂∂ √ −1 ¯ 1 ∂∂ log g c1 (L) = m 2π is positive. Consequently: Proposition 2.1. A holomorphic line bundle L over a compact complex manifold M is ample if and only if L is positive.
88
TADASHI AIKOU
Let M be a compact Riemann surface. The integer g(M ) defined by g(M ) := dimC H 1 (M, OM ) = dimC H 0 (M, O(KM )) ∗ is called the genus of M , where KM = TM is the canonical line bundle over M , and OM is the sheaf of germs of holomorphic functions on M . The degree of a line bundle L is defined by Z deg L = c1 (L) ∈ Z. M
Applying the Riemann–Roch theorem dimC H 0 (M, O(L)) − dimC H 1 (M, O(L)) = deg L + 1 − g(M ) to the case of L = KM , we have ∗ dimC H 1 (M, O(KM )) = dimC H 0 (M, Ω 1 (KM )) = dimC H 0 (M, OM ) = 1,
since M is compact. Consequently we have deg KM = 2g(M ) − 2, and the Euler characteristic χ(M ) is given by Z χ(M ) = c1 (TM ) = − deg KM = 2 − 2g(M ). M
By the uniformisation theorem (e.g., Theorem 4.41 in [Jost 1997]), any compact Riemann surface M is determined completely by its genus g(M ). If g(M ) = 0, then M is isomorphic to the Riemannian sphere S 2 ∼ = P1 and its holomorphic tangent bundle TM is ample. In the case of g(M ) = 1, then M is isomorphic to a torus T = C/Λ, where Λ is a module over Z of rank two, and T M is trivial. In the last case of g(M ) ≥ 2, it is wellknown that M is hyperbolic, i.e., M admits a K¨ ahler metric of negative constant curvature, and T M is negative since c1 (TM ) < 0. In the case of dimC M ≥ 2, Hartshone’s conjecture (“If the tangent bundle TM is ample, then M is biholomorphic to the projective space P n ”) was solved affirmatively by an algebrogeometric method ([Mori 1979]). Then, it is natural to investigate complex manifolds with negative tangent bundles. We next discuss the negativity and ampleness of holomorphic vector bundles. 2.2. Ample vector bundles. Let π : E → M be a holomorphic vector bundle of rank(E) = r + 1 (≥ 2) over a compact complex manifold M , and φ : P(E) = E × /C× → M the projective bundle associated with E. Here and in the sequel, we put E × = E − {0} and C× = C − {0}. We also denote by L(E) the tautological line bundle over P(E), i.e., L(E) = {(V, v) ∈ P(E) × E  v ∈ V } . The dual line bundle H(E) = L(E)∗ is called the hyperplane bundle over P(E).
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
89
Since L(E) is obtained from E by blowing up the zero section of E to P(E), the manifold L(E)× is biholomorphic to E × . This biholomorphism is given by the holomorphic map τ : E × 3 v → ([v], v) ∈ P(E) × E × .
(2.5)
Then, for a arbitrary Hermitian metric gL(E) on L(E), we define the norm kvkE of v ∈ E × by q (2.6) kvkE = gL(E) (τ (v)). Extending this definition to the whole of E continuously, we obtain a function F : E → R by 2
F (v) = kvkE
(2.7)
for every v ∈ E. This function satisfies the following conditions. (F.1) F (v) ≥ 0, and F (v) = 0 if and only if v = 0, (F.2) F (λv) = λ2 F (v) for any λ ∈ C× = C\{0}, (F.3) F (v) is smooth outside of the zerosection. Definition 2.3. Let π : E → M be a holomorphic vector bundle over a complex manifold M . A real valued function F : E → R satisfying the conditions (F 1) ∼ (F 3) is called a Finsler metric on E, and the pair (E, F ) is called a Finsler bundle. If a Finsler metric F satisfies, in addition, √ ¯ is positivedefinite on each fibre Ez , (F.4) the real (1, 1)form −1 ∂ ∂F √ √ ¯ , not −1 ∂ ∂¯ log F .) then F is said to be pseudoconvex . (Note: it’s −1 ∂ ∂F This discussion shows that any Hermitian metric on L(E) defines a Finsler metric on E. Conversely, an arbitrary Finsler metric F on E determines a Hermitian metric gL(E) on L(E), i.e., we obtain Proposition 2.2 [Kobayashi 1975]. There exists a onetoone correspondence between the set of Hermitian metrics on L(E) and the set of Finsler metrics on E. Definition 2.4 [Kobayashi 1975]. A holomorphic vector bundle π : E → M over a compact complex manifold M is said to be negative if its tautological line bundle L(E) → P(E) is negative, and E is said to be ample if its dual E ∗ is negative. The Chern class c1 (L(E)) is represented by the closed real (1, 1)form √ −1 ¯ ∂∂ log F c1 (L(E), F ) = 2π for a Finsler metric F on E. Thus, E is negative if and only if E admits a Finsler metric F satisfying c1 (L(E), F ) < 0. Consequently:
90
TADASHI AIKOU
Proposition 2.3. Let E be a holomorphic vector bundle over a compact complex manifold M . Then E is negative if and only if E admits a Finsler metric F √ ¯ log F < 0. satisfying −1 ∂∂ Given any negative holomorphic vector bundle E over M , we shall construct a pseudoconvex Finsler metric F on E, with c1 (L(E), F ) < 0 (see [Aikou 1999] and [Wong 1984]). By definition the line bundle L(E) is negative, and so L(E)∗ is ample. Hence there exists a sufficiently large positive m ∈ Z such that L := L(E)∗⊗m is very ample. By the definition of very ampleness, we can take f 0 , . . . , f N ∈ H 0 (P(E), L) such that ϕL : P(E) 3 [v] → f 0 ([v]) : · · · : f N ([v]) ∈ PN defines a holomorphic embedding ϕL : P(E) → PN . Then, by Lemma 2.1, we have L ∼ = ϕ∗L H for the hyperplane bundle H → PN . Since c1 (H) is given by (2.4), we have c1 (L) = c1 (ϕ∗L H) = ϕ∗L c1 (H) > 0, and the induced metric gL is given by gL = ϕ∗L gH for the metric gH on H defined by (2.3). Since L = (L(E)∗ ) −m we have gL = gL(E) , and thus the induced metric gL(E) on L(E) is given by
gL(E) =
h
ϕ∗L gH
i−1/m
=
"
1 ϕ∗L gH
#1/m
⊗m
,
.
Because of (2.3), the metric gL is locally given by f j 2 g(j) = P k 2 , f 
and the Finsler metric F on E corresponding to gL(E) is given by F (v) =
h
(ϕ∗L gH ) (τ (v))
i−1/m
=
f k ([v])2 f j ([v])2
P
1/m
v j 2
(2.8)
for v = (v 1 , . . . , v r+1 ) ∈ Ez . The Finsler metric F obtained as above satisfies the condition c1 (L(E), F ) < 0. The pseudoconvexity of F will be shown by more local computations (see Theorem 3.2). Remark 2.1. Every pseudoconvex Finsler metric on a holomorphic vector bundle E is obtained from a pseudoK¨ ahler metric on P(E) (Propositions 4.1 and 4.2). In a later section, we shall show a theorem of Kobayashi’s (Theorem 3.2) which characterizes negative vector bundles in terms of the curvature of Finsler metrics. For this purpose, in the next section, we shall discuss the theory of Finsler connections on (E, F ).
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
91
3. Finsler Connections Let π : E → M be a holomorphic vector bundle of rank(E) = r + 1 over a complex manifold M . We denote by TM the holomorphic tangent bundle of M . We also denote by TE the holomorphic tangent bundle of the total space E. Then we have an exact sequence of holomorphic vector bundles ∗ 0 −→ VE −→ TE −→ π ∗ TM −→ 0,
i
π
(3.1)
where VE = ker π∗ is the vertical subbundle of TE . A connection of the bundle π : E → M is a smooth splitting of this sequence. Definition 3.1. A connection of a fibre bundle π : E → M is a smooth VE valued (1, 0)form θE ∈ Ω1,0 (VE ) satisfying θE (Z) = Z
(3.2)
for all Z ∈ VE . A connection θE defines a smooth splitting TE = V E ⊕ H E
(3.3)
of the sequence (3.1), where HE ⊂ TE is a the horizontal subbundle defined by HE = ker θE . The complex general linear group GL(r + 1, C) acts on E in a natural way. A connection θE is called a linear connection if the horizontal subspace at each point is GL(r + 1, C)invariant. A Hermitian metric on E defines a unique linear connection θE . On the other hand, the multiplier group C× = C\{0} ∼ = {c · I  c ∈ C× } ⊂ GL(r + 1, C) also acts on the total space E by multiplication L λ : E 3 v → λv ∈ E on the fibres for all v ∈ E and λ ∈ C× . In this paper, we assume that a connection θE is C× invariant. We denote by E ∈ OE (VE ) the tautological section of VE generated by the action of C× , i.e., E is defined by E(v) = (v, v) for all v ∈ E. The invariance of θE under the action of C× is equivalent to LE H E ⊂ H E ,
(3.4)
where LE denotes the Lie derivative by E. In this sense, a C× invariant connection θE is called a nonlinear connection. Example 3.1. Let E be a holomorphic vector bundle with a Hermitian metric ¯ be the Hermitian connection of h, and ∇ = π ∗ ∇ ¯ the induced connection h. Let ∇ 1,0 on VE . If we define θ ∈ Ω (VE ) by ¯ θE = ∇E, then θE defines a linear connection on the bundle π : E → M .
(3.5)
92
TADASHI AIKOU
If a pseudoconvex Finsler metric F is given on E, then it defines a canonical nonlinear connection θE (see below). If a connection θE is given on E, we put X V = θE (X), X H = X − X V , and (dV f )(X) = df X V , (dH f )(X) = df X H
for every X ∈ TE and f ∈ C ∞ (E). These operators can be decomposed as dV = ∂ V + ∂¯V and dH = ∂ H + ∂¯H . Furthermore the partial derivatives are also decomposed as ∂ = ∂ H + ∂ V and ∂¯ = ∂¯H + ∂¯V . 3.1. Finsler connection. We define a partial connection ∇ H : VE → Ω1 (VE ) on VE by H ∇H (3.6) Y Z = θE [Y , Z]
for all Y ∈ TE and Z ∈ VE , where [ · , · ] denotes the Lie bracket on TE . This operator ∇H is linear in X and satisfies the Leibnitz rule ∇H (f Z) = (dH f ) ⊗ Z + f ∇H Z for all f ∈ C ∞ (E). On the other hand, since E is a fibre bundle over M , the projection map π : E → M can be used to pullback the said fibre bundle, generating a π ∗ E which sits over E. Note that VE ∼ = π ∗ E. Thus VE admits a canonical relatively flat connection ∇V : VE → Ω1 (VE ) defined by ∇VX (π ∗ s) = 0 for every local holomorphic section s of E, i.e., ∇V = d V .
(3.7)
Then a connection ∇ : VE → Ω1 (VE ) is defined by ∇Z = ∇H Z + dV Z
(3.8)
for every Y ∈ TE and Z ∈ VE . We note here that the connection ∇ is determined uniquely from the connection θE on the bundle π : E → M . Proposition 3.1. Let ∇ : VE → Ω1 (VE ) be the connection on VE defined by (3.8), from a connection θE on the bundle π : E → M . Then ∇ satisfies ∇E = θE .
(3.9)
Proof. Since θE is invariant by the action of C× , we have ∇H E = 0. In fact, H H ∇H X E = θE [X , E] = −θE LE X ] = 0. Furthermore, since dV E (X) = X V , we obtain V V ∇X E = ∇ H = θE (X) X E + d E (X) = X for every X ∈ TE . Hence we have (3.9).
˜
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
93
For the rest of this paper, we shall use the following local coordinate system on M and E. Let U be an open set in M with local coordinates (z 1 , . . . , z n ), and s = P i (s0 , . . . , sr ) a holomorphic local frame field on U . By setting v= ξ si (z) on for each v ∈ π −1 (U ), we take (z, ξ) = z 1 , . . . , z n , ξ 0 , . . . , ξ r as a local coordinate system on π −1 (U ) ⊂ E. We use the notation ∂α =
∂ ∂z α
and ∂j =
∂ . ∂ξ j
We also denote by ∂α¯ and ∂¯j their conjugates. We suppose that a pseudoconvex Finsler metric F is given on E. Then, by √ ¯ is a closed real (1, 1)definition in the previous section, the form ωE = −1 ∂ ∂F form on the total space E such that the restriction ω z on each fibre Ez = π −1 (z) defines a K¨ ahler metric Gz on Ez , and ωE defines a Hermitian metric GVE on the bundle VE . Thus we shall investigate the geometry of the Hermitian bundle {VE , GVE }. AFinsler metric F on E is pseudoconvex if and only if the Hermitian matrix Gi¯j defined by Gi¯j (z, ξ) = ∂i ∂¯j F
(3.10)
is positivedefinite. Each fibre Ez may be considered as a K¨ ahler manifold with √ P i j ¯ K¨ ahler form ωz = −1 Gi¯j dξ ∧ dξ . The family {Ez , ωz }z∈M is considered as a family of K¨ ahler manifolds and the bundle is considered as the associated fibred manifold. The Hermitian metric GVE on VE is defined by GVE (si , sj ) = Gi¯j ,
(3.11)
where we consider s = (s0 , . . . , sr ) as a local holomorphic frame field for VE ∼ = π ∗ E. We denote by k · kE the norm defined by the Hermitian metric GVE . Then, because of the homogeneity (F 2), we have 2
kEkE = GVE (E, E) = F (z, ξ)
(3.12)
and (3.13) L E GV E = G V E P i for the tautological section E(z, ξ) = ξ si (z). Let θ = θ1 , . . . , θ r be the dual frame field for the dual bundle VE∗ , i.e., θi (sj ) = δji . A connection θE for the bundle π : E → M is written as θE = s ⊗ θ =
X
si ⊗ θ i .
Proposition 3.2. Let F be a pseudoconvex Finsler metric on a holomorphic vector bundle E. Then there exists a unique connection θE such that (3.3) is the orthogonal splitting with respect to ωE .
94
TADASHI AIKOU
Proof. We can easily show that the required (1, 0)form θ i is defined by θ i = P dξ i + Nαi dz α , where the local functions Nαi (z, ξ) are given by X ¯ Nαi = Gim ∂α ∂m ¯F ¯ for the inverse matrix Gim of (Gim ¯ ). By the homogeneity (F 2), these functions i i satisfy Nα (z, λξ) = λNα (z, ξ) for every λ ∈ C× , which implies that θE is C× invariant. ˜ For the rest of this paper, we shall adopt the connection θ E obtained in Proposition 3.2 on a pseudoconvex Finsler bundle (E, F ). Proposition 3.3 [Aikou 1998]. The connection θE satisfies ∂ H ◦ ∂ H ≡ 0.
(3.14)
Such a connection θE determines a unique connection ∇ on VE . Definition 3.2. The connection ∇ : VE → Ω1 (VE ) on (VE , GVE ) defined by (3.8) is called the Finsler connection of (E, F ). The connection ∇ defined by (3.8) is canonical in the following sense. Proposition 3.4 [Aikou 1998]. Let ∇ be the Finsler connection on a pseudoconvex Finsler bundle (E, F ). Then ∇ = ∇H +dV satisfies the following metrical condition. dH GVE (Y, Z) = GVE (∇H Y, Z) + GVE (Y, ∇H Z) (3.15) for all Y, Z ∈ VE . The connection form ω = ωji of ∇ with respect to a local holomorphic frame P field s = (s0 , . . . , sr ) is defined by ∇sj = si ⊗ ωji . By the identity (3.15), ω is given by ω = G−1 ∂ H G. (3.16) 3.2. Curvature. Let ∇ be the Finsler connection on (E, F ). We also denote by ∇ : Ωk (VE ) → Ωk+1 (VE ) the covariant exterior derivative defined by ∇. Definition 3.3. The section R = ∇ ◦ ∇ ∈ Ω2 (End(VE )) is called the curvature of ∇. With respect to the local frame field s = (s0 , . . . , sr ), we put X R(sj ) = si ⊗ Ωji . In matrix notation, the curvature form Ω = Ωji of ∇ is given by Ω = dω + ω ∧ ω.
(3.17)
The curvature form Ω is decomposed as Ω = dH ω + ω ∧ ω + dV ω, which can be simplified to Ω = ∂¯H ω + dV ω. (3.18) This is made possible by
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
95
Proposition 3.5. The horizontal (2, 0)part of Ω vanishes, i .e., ∂ H ω + ω ∧ ω ≡ 0. Proof. Since ω = G
(3.19)
−1 H
∂ G, (3.14) implies ∂ H ω + ω ∧ ω = ∂ H G−1 ∂ H G + ω ∧ ω
= −G−1 ∂ H G ∧ G−1 ∂ H G + G−1 ∂ H ◦ ∂ H G + ω ∧ ω = G−1 ∂ H ◦ ∂ H G = 0.
˜
We give the definition of flat Finsler metrics. Definition 3.4. A Finsler metric F is said to be flat if there exists a holomorphic local frame field s = (s0 , . . . , sr ) around every point of M such that F = F (ξ), i.e., F is independent of the base point z ∈ M . Theorem 3.1 [Aikou 1999]. A Finsler metric F is flat if and only if the curvature R vanishing identically. Let RH be the curvature of the partial connection ∇H , i.e., RH = ∇H ◦ ∇H . From (3.18), the curvature form Ω H of RH is given by Ω H = ∂¯H ω. If we put X i α ∂¯H ωji = Rjα zβ , β¯ dz ∧ d¯
the curvature RH is given by
RH (sj ) =
X
i α si ⊗ Rjα zβ . β¯ dz ∧ d¯
For the curvature form Ω H of ∇H , we define a horizontal (1, 1)form Ψ by H (E), E GVE RXY 1 X = Ri¯jαβ¯ (z, ξ)ξ i ξ¯j X α Y¯ β Ψ (X, Y ) = 2 F kEkE for any horizontal vector fields X, Y at (z, ξ) ∈ E, where we put R i¯jαβ¯ = P m Gm¯j Riα . We set β¯ 1 X Ri¯jαβ¯ ξ i ξ¯j . (3.20) Ψαβ¯ (z, ξ) = F P In [Kobayashi 1975], this (1, 1)form Ψ = Ψαβ¯ dz α ∧ d¯ z β is also called the curvature of F . Definition 3.5. If the curvature form Ψ satisfies the negativity condition, i.e., Ψ (Y, Z) < 0 for all Y, Z ∈ HE , then we say that (E, F ) is negatively curved . Direct computation gives: Proposition 3.6 [Aikou 1998]. For a pseudoconvex Finsler metric F on a √ holomorphic vector bundle E, the real (1, 1) form −1 ∂ ∂¯ log F = −c1 (L(E), F ) is given by √ √ −Ψαβ¯ O −1 ∂ ∂¯ log F = −1 O ∂i ∂¯j log F
96
TADASHI AIKOU
with respect to the orthogonal decomposition (3.3), i .e., √
−1 ∂ ∂¯ log F =
√
−1 −
X
Ψαβ¯ dz ∧ d¯ z + α
β
X
∂i ∂¯j (log F )θ ∧ θ¯j . (3.21) i
Analyzing the negativity of the form c1 (L(E), F ), we have the following theorem of Kobayashi. Theorem 3.2 [Kobayashi 1975]. A holomorphic vector bundle π : E → M over a compact complex manifold M is negative if and only if E admits a negatively curved pseudoconvex Finsler metric. Proof. By Proposition 2.3, E is negative if and only if there exists a Finsler √ ¯ log F < 0. Since ∂ and ∂¯ anticommute, this charmetric F such that −1 ∂∂ √ acterization is equivalent to −1 ∂ ∂¯ log F > 0. Thus, E is negative if and only if the right hand side of (3.21) is positive. Denote by Fz the restriction of F to each fibre Ez = π −1 (z). Then, we have √ √ ¯ z = −1 Fz ∂ ∂¯ log Fz + ∂ log Fz ∧ ∂¯ log Fz . −1 ∂ ∂F √ If (3.21) has positive right hand side, then −1 Ψ must be negative, and √ −1 ∂ ∂¯ log Fz must be positive. The latter, in conjunction with the displayed √ ¯ z . Thus F is pseudoconvex and negformula, implies the positivity of −1 ∂ ∂F atively curved. Conversely, suppose F is pseudoconvex and negatively curved. That is, we √ √ ¯ z > 0 and −1 Ψ < 0. Now, the pseudoconvexity of F implies have −1 ∂ ∂F that the second term on the right hand side of (3.21) is positivedefinite (see section 4.1 for details). Thus the entire right hand side of (3.21) is positive. ˜ Remark 3.1. In this section, the horizontal (1, 1)part R H of R plays an important role. In Finsler geometry, there are other important tensors. The V E valued 2form TE defined by TE = ∇θE (3.22) is called the torsion form of ∇, which is expressed by X X TE = s ⊗ (dθ + ω ∧ θ) = si ⊗ dθi + ωji ∧ θj .
Because of (3.9), the torsion form TE is also given by TE = R(E).
The torsion form TE vanishes if and only if the horizontal subbundle HE defined by θE is holomorphic and integrable (see [Aikou 2003b]). On the other hand, the mixed part RHV of R defined by X RHV (sj ) = si ⊗ dV ωji is also an important curvature form. The vanishing of R HV shows that (E, F ) is modeled on a complex Minkowski space, i.e., there exists a Hermitian metric h F
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
97
¯ for on E such that the Finsler connection ∇ on (E, F ) is obtained by ∇ = π ∗ ∇ ¯ the Hermitian metric ∇ of (E, hF ) (see [Aikou 1995]). Hence a Finsler metric F is flat if and only if (E, F ) is modeled on a complex Minkowski space and the associated Hermitian metric hF is flat. 3.3. Holomorphic sectional curvature. We now study the holomorphic tangent bundle TM of a complex manifold with a pseudoconvex Finsler metric F : TM → R. The pair (M, F ) is called a complex Finsler manifold . This is the special case of E = TM in Definition 2.3, and we have the natural identifications VE ∼ = HE ∼ = π ∗ TM . Let ∆(r) = {ζ ∈ C : ζ < r} be the disk of radius r in C with the Poincar´e metric 4r2 ¯ dζ ⊗ dζ. gr = 2 (r − ζ2 )2 For every point (z, ξ) ∈ TM , there exists a holomorphic map ϕ : ∆(r) → M satisfying ϕ(0) = z and ∂ ϕ∗ (0) := ϕ∗ = ξ. (3.23) ∂ζ ζ=0 Then, the pullback ϕ∗ F defines a Hermitian metric on ∆(r) by ¯ ϕ∗ F = E(ζ) dζ ⊗ dζ, where we put E(ζ) = F (ϕ(ζ), ϕ∗ (ζ)). The Gauss curvature Kϕ∗ F (z, ξ) is defined by 1 ∂ 2 log E . Kϕ∗ F (z, ξ) = − E ∂ζ ∂ ζ¯ ζ=0
Definition 3.6 [Royden 1986]. The holomorphic sectional curvature K F of (M, F ) at (z, ξ) ∈ TM is defined by KF (z, ξ) = sup {Kϕ∗ F (z, ξ) : ϕ(0) = z, ϕ∗ (0) = ξ} , ϕ
where ϕ ranges over all holomorphic maps from a small disk into M satisfying ϕ(0) = z and (3.23). Then KF has a computable expression in terms of the curvature tensor of the Finsler connection ∇. Proposition 3.7 [Aikou 1991]. The holomorphic sectional curvature KF of a complex Finsler manifold (M, F ) is given by 1 X Ψ (E, E) = 2 KF (z, ξ) = Ri¯jk¯l (z, ξ)ξ i ξ¯j ξ k ξ¯l , (3.24) 2 F kEkE P m where Ri¯jk¯l = Gm¯j Rik is the curvature tensor of the Finsler connection ∇ ¯ l on (TM , F ). Then we have a Schwarztype lemma:
98
TADASHI AIKOU
Proposition 3.8 [Aikou 1991]. Let F be a pseudoconvex Finsler metric on the holomorphic tangent bundle of a complex manifold M . Suppose that its holomorphic sectional curvature KF (z, ξ) at every point (z, ξ) ∈ TM is bounded above by a negative constant −k. Then, for every holomorphic map ϕ : ∆(r) → M satisfying ϕ(0) = z and (3.23), we have gr ≥ kϕ∗ F.
(3.25)
The Kobayashi metric FM on a complex manifold M is a positive semidefinite pseudo metric defined by (3.26) FM (z, ξ) = inf 1r : ϕ(0) = z, ϕ∗ (0) = ξ . ϕ
In general, FM is not smooth. FM is only upper semicontinuous, i.e., for every X ∈ TM and every ε > 0 there exists a neighborhood U of X such that F M (Y ) < FM (X)+ε for all Y ∈ U (see [Kobayashi 1998], [Lang 1987]). Even though F M is not a Finsler metric in our sense, the decreasing principle shows the importance of the Kobayashi metric, i.e., for every holomorphic map ϕ : N → M , we have the inequality FN (X) ≥ FM (ϕ∗ (X)). (3.27) This principle shows that FM is holomorphically invariant, i.e., if ϕ : N → M is biholomorphic, then we have FN = ϕ∗ FM . In this sense, FM is an intrinsic metric on complex manifolds. A typical example of Kobayashi metrics is the one on a domain in C n . It is wellknown that, if M is a strongly convex domain with smooth boundary in C n , then FM is a pseudoconvex Finsler metric in our sense (see [Lempert 1981]). A complex manifold M is said to be Kobayashi hyperbolic if its Kobayashi metric FM is a metric in the usual sense. If M admits a pseudoconvex Finsler metric F whose holomorphic sectional curvature K F is bounded above by a negative constant −k, then (3.25) implies the inequality 2 FM ≥ kF,
(3.28)
and thus M is Kobayashi hyperbolic. Theorem 3.3 [Kobayashi 1975]. Let M be a compact complex manifold . If its holomorphic tangent bundle TM is negative, then M is Kobayashi hyperbolic. Proof. We suppose that TM is negative. Then, Theorem 3.2 implies that there exists a pseudoconvex Finsler metric F on TM with negativedefinite Ψ . By the definition (3.20), the negativity of Ψ and (3.24) imply KF (z, ξ) =
Ψ (E, E) 2
< 0.
kEkE Since M is compact, P(E) is also compact. Moreover, since K F is a function on P(E), the negativity of KF shows that KF is bounded by a negative constant −k. Hence we obtain (3.28), and M is Kobayashi hyperbolic. ˜
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
99
Remark 3.2. Recently Cao and Wong [Cao and Wong 2003] have introduced the notion of “mixed holomorphic bisectional curvature” for Finsler bundles (E, F ), which equals the usual holomorphic bisectional curvature in the case of E = TM . They also succeeded in showing that a holomorphic vector bundle E of rank(E) ≥ 2 over a compact complex manifold M is ample if and only if E admits a Finsler metric with positive mixed holomorphic bisectional curvature.
4. Ruled Manifolds 4.1. Projective bundle. Let φ : P(E) → M be the projective bundle associated with a holomorphic vector bundle E over M . ¯ Definition 4.1. A locally ∂ ∂exact real (1, 1)form ωP(E) on the total space P(E) is called a pseudoK¨ ahler metric on P(E) if its restriction to each fibre defines a K¨ ahler metric on P(Ez ) ∼ = Pr . If a pseudoK¨ ahler metric ωP(E) is given on P(E), then its restriction to each fibre φ−1 (z) = P(Ez ) is a K¨ ahler form on P(Ez ). We shall show that a pseudoconvex Finsler metric on E defines a pseudoK¨ ahler metric on P(E). For this purpose, we use the socalled Euler sequence (e.g., [Zheng 2000]). We denote by ρ : E × → P(E) the natural projection. We also denote by VP(E) := ker φ∗ the vertical subbundle of TP(E) . Let s = (s0 , . . . , sr ) be a holomorphic local frame field of E on an open set U ⊂ M , which is naturally considered as a holomorphic local frame field of VE on π −1 (U ). Then, the vertical subbundle VP(E) ⊂ TP(E) is locally spanned by {ρ∗ s0 , . . . , ρ∗ sr } with the relation ρ∗ E = 0.
(4.1)
Then the Euler sequence 0 −→ L(E) −→ φ∗ E −→ L(E) ⊗ VP(E) −→ 0 i
implies the following exact sequence of vector bundles: P
0 −→ 1P(E) −→ H(E) ⊗ φ∗ E −→ VP(E) −→ 0, i
(4.2)
where H(E) = L(E)∗ is the hyperplane bundle over P(E) and the bundle morphism P : H(E) ⊗ φ∗ E → VP(E) is defined as follows. Since any section Z of P j H(E) ⊗ φ∗ E can be naturally identified with a section Z = Z sj of VE satisfying the homogeneity Z j (λv) = λZ j (v), the definition ρ∗ Z(v) = (ρ∗ Z) ([v]) makes sense. Then P is defined by X P(Z) = ρ∗ (4.3) Z j sj .
Moreover, since ρ∗ E = 0, 1P(E) (= ker P) is the trivial line bundle spanned by E. Then, since ker P = 1P(E) is spanned by E, the morphism P is expressed as P(Z) = Z −
GVE (Z, E) 2
kEkE
E
100
TADASHI AIKOU
for a Hermitian metric GVE on VE . Since P is surjective, for any sections Z˜ ˜ of VP(E) , there exist sections Z and W of VE satisfying P(Z) = Z˜ and and W ˜ . Then, a Hermitian metric GV on VP(E) is defined by P(W ) = W P(E) 2
˜ W ˜ ) = kEkE GVE (Z, W ) − GVE (Z, E)GVE (E, W ) GVP(E) (Z, 4 kEkE
(4.4)
for the Hermitian metric GVE on VE defined by (3.11), which induces the orthogonal decomposition H(E) ⊗ φ∗ E = 1P(E) ⊕ VP(E) . Because of (4.1) and (4.4), the components of the metric G VP(E) with respect to the local frame {ρ∗ (sj )} is given by GVP(E) (ρ∗ si , ρ∗ sj ) = ∂i ∂¯j (log F ).
(4.5)
Consequently we have Proposition 4.1. If F is a pseudoconvex Finsler metric on a holomorphic √ vector bundle E, then the real (1, 1)form −1 ∂ ∂¯ log F defines a pseudoK¨ ahler metric on P(E). Conversely: Proposition 4.2. If ωP(E) is a pseudoK¨ ahler metric on P(E), then ωP(E) √ defines a pseudoconvex Finsler metric F on E such that ρ∗ ωP(E) = −1 ∂ ∂¯ log F . Such a pseudoconvex Finsler metric F is unique up to a local positive function σU on U ⊂ M . Proof. On each open set U(j) = {[v] = (z, [ξ]) ∈ φ−1 (U ) : ξ j 6= 0} of P(E), we express the pseudoK¨ ahler metric ωP(E) by √ ¯ (j) , ωP(E) U(j) = −1 ∂ ∂g where {g(j) } is a family of local smooth functions g(j) : U(j) → R. Since the restriction of this form to each fibre Pz ⊂ U(j) is a K¨ ahler form ωz on Pz , we may put √ ¯ z,(j) , ωz = −1 ∂ ∂g where the local functions gz,(j) = g(j) Pz depend on z ∈ U smoothly. Then we define a function Fz : Ez× → R by Fz (ξ) = ξ j 2 exp gz,(j) . Since Fz also depends on z ∈ U smoothly, we extend this function to a smooth function F : E × → R by F (z, ξ) = Fz (ξ). It is easily verified that F defines a pseudoconvex Finsler metric on E.
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
101
We note that another K¨ ahler potential {˜ g(j) } for ωP(E) which induces the K¨ ahler metric ωz on each Pz is given by g˜(j) (z, [ξ]) = σU (z) + g(j) (z, [ξ])
(4.6)
for some functions σU (z) defined on U . Hence the Finsler metric F˜ determined from the potential {˜ gj } is connected to the function F by the relation F˜ = σU (z) e F on each U . ˜ Similar to Definition 3.4, we say a pseudoK¨ ahler metric ω P(E) on P(E) is flat if there exists an open cover {U, s} of E so that we can choose K¨ ahler potentials g(j) for ωP(E) which are independent of the base point z ∈ M . Now we define the projectiveflatness of Finsler metrics. Definition 4.2. A Finsler metric on E is said to be projectively flat if it is obtained from a flat pseudoK¨ ahler metric on P(E). In a previous paper, we proved this result: Theorem 4.1 [Aikou 2003a]. A pseudoconvex Finsler metric is projectively flat if and only if the tracefree part of the curvature form Ω vanishes identically, i .e., Ω = A(z) ⊗ Id
(4.7)
for some (1, 1)form A on M . Remark 4.1. A Finsler metric F is projectively flat if and only if there exists a local function σU (z) on U such that F is of the form F (z, ξ)U = exp σU (z) · ξ j 2 exp g(j) ([ξ])
(4.8)
on each U . In other words, a Finsler metric F is projectively flat if and only if there exists a local function σU (z) on U such that the local metric e−σU (z) F is a flat Finsler metric on U . In the previous paper [Aikou 1997], such a Finsler metric F is said to be conformally flat. We suppose that a pseudoconvex Finsler metric F is projectively flat. Then, since the curvature form Ω is given by (3.18), we have R HV ≡ 0, and thus (E, F ) is modeled on a complex Minkowski space. We can easily show that the associated Hermitian metric hF is also projectively flat. Hence: Theorem 4.2. A holomorphic vector bundle E of rank(E) = r + 1 admits a projectively flat Finsler metric if and only if P(E) is flat, i .e., ˜ ×ρ Pr , P(E) = M
(4.9)
˜ is the universal cover of M , and ρ : π1 (M ) → P U (r) is a representation where M of the fundamental group π1 (M ) in the projective unitary group P U (r).
102
TADASHI AIKOU
4.2. Ruled manifolds. An algebraic surface X is said to be ruled if it is birational to M × P1 , where M is a compact Riemann surface. An algebraic surface X is said to be geometrically ruled if there exists a holomorphic projection φ : X → M such that every fibre φ−1 (z) = Xz is holomorphically isomorphic to P1 . As is well known, a geometrically ruled surface is ruled (see [Beauville 1983]), and every geometrically ruled surface X is holomorphically isomorphic to P(E) for some holomorphic vector bundle π : E → M of rank(E) = 2 (see [Yang 1991], for example). An algebraic manifold X is said to be a ruled manifold if X is a holomorphic Pr bundle with structure group P GL(r + 1, C) = GL(r + 1, C)/C × . Any holomorphic Pr bundle over M is classified by H 1 (M, P GL(r + 1, OM )), and any rank r + 1 holomorphic vector bundle over M is classified by the elements of H 1 (M, GL(r + 1, OM )). The exact sequence ∗ 0 → OM → GL(r + 1, OM ) → P GL(r + 1, OM ) → 0
implies the sequence of cohomology groups: ∗ · · · → H 1 (M, GL(r + 1, OM )) → H 1 (M, P GL(r + 1, OM )) → H 2 (M, OM ) → ··· ∗ ) = 0, the following is obtained. Since H 2 (M, OM
Proposition 4.3. Every ruled manifold X over a compact Riemann surface M is holomorphically isomorphic to P(E) for some holomorphic vector bundle π : E → M of rank(E) = r + 1. Such a bundle E is uniquely determined up to tensor product with a holomorphic line bundle. If E is a holomorphic vector bundle over a compact K¨ ahler manifold M , then P(E) is also a compact K¨ ahler manifold. In fact, we can construct a K¨ ahler metric ωP(E) on P(E) of the form ωP(E) = φ∗ ωM + εη. Here, ωM is a K¨ ahler metric on M , ε is a small positive constant, and η is a closed (1, 1)form on P(E) such that η is positivedefinite on the fibres of φ (see Lemma (6.37) in [Shiffmann and Sommese 1985]). Thus every ruled manifold X over a compact Riemann surface M is a compact K¨ ahler manifold, and φ : X = P(E) → M is a holomorphic submersion from a compact K¨ ahler manifold X to M . Then we have Theorem 4.3. Let X be a ruled manifold over a compact Riemann surface M with a K¨ ahler metric ωX . Then there exists a negative vector bundle π : E → M such that X = P(E), √ and a negatively curved pseudoconvex Finsler metric F on E satisfying ρ∗ ωX = 1∂ ∂¯ log F . Proof. Let ωX be a K¨ ahler metric on X . Propositions 4.3 and 4.2 imply that there exists a holomorphic vector bundle E satisfying X = P(E) with a pseudoconvex Finsler metric F . Then, since √ −1 ∂ ∂¯ log F = ωX > 0,
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
F is negatively curved, and hence Theorem 2.3 implies that E is negative.
103
˜
LeBrun [LeBrun 1995] has investigated minimal ruled surfaces (i.e., geometrically ruled surface) over a compact Riemann surface of genus g(M ) ≥ 2 with constant negative scalar curvature. Roughly speaking, he proved that such a minimal ruled surface X is obtained by a semistable vector bundle over M so that X = P(E). Since the semistability of vector bundles over a compact Riemann surface is equivalent to the existence of a projectively flat Hermitian metric on E, such a surface is written in the form (4.9). On the other hand, by Theorem 4.3, the geometry of a minimal ruled surface X is naturally translated to the geometry of a negative vector bundle E with a negatively curved pseudoconvex Finsler metric F . From this viewpoint, we have also investigated minimal ruled surfaces, and we have concluded that each minimal ruled surface φ : X → M over a compact Riemann surface of genus g(M ) ≥ 2 with constant negative scalar curvature is a K¨ ahler submersion with isometric fibres (see [Aikou 2003b]).
References [Abate–Patrizio 1994] M. Abate and G. Patrizio, Finsler metrics: a global approach, Lecture Notes in Math. 1591, Springer, Berlin, 1994. [Aikou 1991] T. Aikou, “On complex Finsler manifolds”, Rep. Fac. Sci. Kagoshima Univ. 24 (1991), 9–25. [Aikou 1995] T. Aikou, “Complex manifolds modeled on a complex Minkowski space”, J. Math. Kyoto Univ. 35 (1995), 83–111. [Aikou 1996] T. Aikou, “Some remarks on locally conformal complex Berwald spaces”, pp. 109–120 in Finsler geometry (Seattle, 1995), edited by David Bao et al., Contemporary Math. 196, Amer. Math. Soc., Providence, 1996. [Aikou 1997] T. Aikou, “Einstein–Finsler vector bundles”, Publ. Math. Debrecen 51 (1997), 363–384. [Aikou 1998] T. Aikou, “A partial connection on complex Finsler bundles and its applications”, Illinois J. Math. 42 (1998), 481–492. [Aikou 1999] T. Aikou, “Conformal flatness of complex Finsler structures”, Publ. Math. Debrecen 54 (1999), 165–179. [Aikou 2000] T. Aikou, “Some remarks on the conformal equivalence of complex Finsler structures”, pp. 35–52 in Finslerian geometry: a meeting of minds, edited by P. Antonelli, Kluwer, Amsterdam, 2000. [Aikou 2002] T. Aikou, “Applications of Bott connection to Finsler geometry”, pp. 1– 13 in Steps in differential geometry, University of Debrecen Institute of Mathematics and Informatics, Debrecen, 2001. [Aikou 2003a] T. Aikou, “Projective flatness of complex Finsler metrics”, Publ. Math. Debrecen 63 (2003), 343–362. [Aikou 2003b] T. Aikou, “A note on projectively flat complex Finsler metrics”, preprint, 2003.
104
TADASHI AIKOU
[Bao et al. 1996] D. Bao, S.S. Chern and Z. Shen (editors), Finsler geometry, Contemporary Math. 196, Amer. Math. Soc., Providence, 1996. [Bao et al. 2000] D. Bao, S.S. Chern and Z. Shen, An introduction to Riemann–Finsler geometry, Graduate Texts in Math. 200, Springer, New York, 2000. [Beauville 1983] A. Beauville, Complex algebraic surfaces, Cambridge Univ. Press, Cambridge, 1983. [Cao and Wong 2003] J.K. Cao and P.M. Wong, “Finsler geometry of projectivized vector bundles”, to appear in J. Math. Kyoto Univ. [Chern et al. 1999] S.S. Chern, W. H. Chen and K. S. Lam, Lectures on differential geometry, Series on Univ. Math. 1, World Scientific, Singapore, 1999. [Dineen 1989] S. Dineen, The Schwarz Lemma, Clarendon Press, Oxford, 1989. [Duchamp and Kalka 1988] T. Duchamp and M. Kalka, “Invariants of complex foliations and the Monge–Amp`ere equation”, Michigan Math. J. 35 (1988), 91–115. [Faran 1990] J. J. Faran, “Hermitian Finsler metrics and the Kobayashi metric”, J. Differential Geometry 31 (1990), 601–625. [Fukui 1989] M. Fukui, “Complex Finsler manifolds”, J. Math. Kyoto Univ. 29 (1989), 609–624. [Griffith and Harris 1978] G. Griffith and J. Harris, Principles of algebraic geometry, Wiley, New York, 1978. [Icijy¯ o 1988] Y. Icijy¯ o, “Finsler metrics on almost complex manifolds”, Riv. Mat. Univ. Parma 14 (1988), 1–28. [Icijy¯ o 1994] Y. Icijy¯ o, “Kaehlerian Finsler manifolds”, J. Math. Tokushima Univ. 28 (1994), 19–27. [Icijy¯ o 1998] Y. Icijy¯ o, “Kaehlerian Finsler manifolds of Chern type”, Research Bull. Tokushima Bunri Univ. 57 (1999), 9–16. [Jost 1997] J. Jost, Compact Riemann surfaces, Universitext, Springer, Berlin, 1997. [Kobayashi 1975] S. Kobayashi, “Negative vector bundles and complex Finsler structures”, Nagoya Math. J. 57 (1975), 153–166. [Kobayashi 1987] S. Kobayashi, Differential geometry of complex vector bundles, Iwanami, Tokyo, and Princeton Univ. Press, Priceton, 1987. [Kobayashi 1996] S. Kobayashi, “Complex Finsler vector bundles”, pp. 145–153 in Finsler geometry (Seattle, 1995), edited by David Bao et al., Contemporary Math. 196, Amer. Math. Soc., Providence, 1996. [Kobayashi 1998] S. Kobayashi, Hyperbolic complex spaces, Springer, Berlin, 1998. [Kobayashi and Wu 1970] S. Kobayashi and H.H. Wu, “On holomorphic sections on certain Hermitian vector bundles”, Math. Ann. 189 (1970), 1–4. [Kobayashi and Nomizu 1963] Foundations of differential geometry, vol. II, Wiley, New York, 1963. [Lang 1987] S. Lang, Introduction to complex hyperbolic spaces, Springer, 1987. [LeBrun 1995] C. LeBrun, “Polarized 4manifolds, extremal K¨ ahler metrics, and Seiberg–Witten theory”, Math. Res. Lett. 2 (1995), 653–662.
FINSLER GEOMETRY ON COMPLEX VECTOR BUNDLES
105
[Lempert 1981] L. Lempert, “La m´etrique de Kobayashi et la repr´esentation des domaines sur la boule”, Bull. Soc. Math. France 109 (1981), 427–474. [Matsumoto 1986] M. Matsumoto, Foundations of Finsler geometry and special Finsler spaces, Kaiseisha Press, Otsu, Japan, 1986. [Mori 1979] S. Mori, “Projective manifolds with ample tangent bundles”, Ann. of Math. (2) 110 (1979), 593–606. [Pang 1992] M.Y. Pang, “Finsler metrics with properties of the Kobayashi metric of convex domains”, Publicacions Matem` atiques (Barcelona) 36 (1992), 131–155. [Patrizio and Wong 1983] G. Patrizio and P.M. Wong, “Stability of the Monge–Amp`ere Foliation”, Math. Ann. 263 (1983), 1329. [Rizza 1963] G. B. Rizza, “Strutture di Finsler di tipo quasi Hermitiano”, Riv. Mat. Univ. Parma 4 (1963), 83–106. [Royden 1986] H. L. Royden, “Complex Finsler metrics”, pp. 261–271 in Complex differential geometry and nonlinear differential equations, edited by YumTong Siu, Contemporary Math. 49, Amer. Math. Soc., Providence, 1986. [Rund 1959] H. Rund, The differential geometry of Finsler spaces, Springer, Berlin, 1959. [Rund 1972] H. Rund, “The curvature theory of directiondependent connections on complex manifolds”, Tensor (N.S.) 24 (1972), 189–205. [Shiffmann and Sommese 1985] B. Shiffmann and A. J. Sommese, Vanishing theorems on complex manifolds, Birkh¨ auser, Boston, 1985. [Yang 1991] K. Yang, Complex algebraic geometry: an introduction to curves and surfaces, Dekker, New York, 1991. [Watt 1996] C. Watt, “Complex sprays, Finsler metrics and horizontal complex curves”, Ph.D. Thesis, Trinity College Dublin, 1996. [Wong 1977] B. Wong, “On the holomorphic curvature of some intrinsic metrics”, Proc. Amer. Math. Soc. 65 (1977), 57–61. [Wong 1984] B. Wong, “A class of compact complex manifolds with negative tangent bundles”, Math. Z. 185 (1984), 217–223. [Wu 1973] B. Wu, “A remark on holomorphic sectional curvature”, Indiana Univ. Math. J. 22 (1973), 1103–1108. [Zheng 2000] F. Zheng, Complex differential geometry, AMS/IP studies in advanced mathematics 18, Amer. Math. Soc., Providence, 2000. Tadashi Aikou Department of Mathematics and Computer Science Faculty of Science Kagoshima University Kagoshima 8900065 Japan
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Finsler Geometry of Holomorphic Jet Bundles KAREN CHANDLER AND PITMANN WONG
Contents Introduction 1. Holomorphic Jet Bundles 2. Chern Classes and Cohomology Groups: The Case of Curves 3. Computation of Chern Classes: The Case of Surfaces 4. Finsler Geometry of Projectivized Vector Bundles 5. Weighted Projective Spaces and Projectivized Jet Bundles 6. The Lemma of Logarithmic Derivatives and the Schwarz Lemma 7. Surfaces of General Type Acknowledgements References
107 113 120 139 148 153 164 173 192 192
Introduction A complex manifold X is Brody hyperbolic if every holomorphic map f : C → X is constant. For compact complex manifolds this is equivalent to the condition that the Kobayashi pseudometric κ1 (see (1.12)) is a positive definite Finsler metric. One may verify the hyperbolicity of a manifold by exhibiting a Finsler metric with negative holomorphic sectional curvature. The construction of such a metric motivates the use of parametrized jet bundles, as defined by Green–Griffiths. (The theory of these bundles goes back to [Ehresmann 1952].) We examine the algebraicgeometric properties (ample, big, nef, spanned and the dimension of the base locus) of these bundles that are relevant toward the metric’s existence. To do this, we start by determining (and computing) basic invariants of jet bundles. Then we apply Nevanlinna theory, via the construction of an appropriate singular Finsler metric of logarithmic type, to obtain precise extensions of the classical Schwarz Lemma on differential forms toward jets. Particularly, this allows direct control over the analysis of the jets j k f of a holomorphic map f : C → X; namely, the image of j k f must be contained in the base locus of the jet differentials. For an algebraically nondegenerate holomorphic map we show by means of reparametrization that the algebraic 107
108
KAREN CHANDLER AND PITMANN WONG
closure of j k f is quite large while, under appropriate conditions, the base locus is relatively small. This contradiction shows that the map f must be algebraically degenerate. We apply this method to verify that a generic smooth hypersurface of P 3 , of degree d ≥ 5, is hyperbolic (see Corollary 7.21). Using this we show also the existence of a smooth curve C of degree d = 5 in P 2 such that P 2 \ C is Kobayashi hyperbolic. In the classical theory of curves (Riemann surfaces) the most important invariant is the genus. The genus g of a curve is the number of independent global regular 1forms: g = h0 (KX ) = dim H 0 (KX ), where KX = T ∗ X is the canonical bundle (which in the case of curves is also the cotangent bundle). A curve is hyperbolic if and only if g ≥ 2. One way to see this is to take a basis ω 1 , . . . , ωg of regular 1forms and define a metric ρ on the tangent bundle by setting ρ(v) =
g X
ωi (v)
i=1
2
1/2
, v ∈ T X.
(∗)
For g = 1 the metric is flat, that is, the Gaussian or holomorphic sectional curvature (hsc) is zero. Hence X is an elliptic curve. For g ≥ 2 the curvature of this metric is strictly negative which, by the classical Poincar´e–Schwarz Lemma, implies that X is hyperbolic. Algebraic geometers take the dual approach by interpreting ρ as defining a metric along the fibers of the dual T ∗ X = KX and, for g ≥ 2, the Chern form c1 (KX , ρ) is positive, that is, the canonical bundle is ample. Indeed the following four conditions are equivalent: (i) g ≥ 2; (ii) X is hyperbolic; (iii) T ∗ X is ample; and (iv) There exists a negatively curved metric. For a complex compact manifold of higher dimension the number of independent 1forms g = h0 (T ∗ X) is known as the irregularity of the manifold. If g ≥ 1, we may define ρ as in (∗). More generally, for each m, we may choose a basis Jm ∗ Jm ∗ ω1 , . . . , ωgm of H 0 ( T X), where T X is the mfold symmetric product, and define, if gm ≥ 1, ρ(v) =
gm X
ωi (v)1/m .
(∗∗)
i=1
In dimension 2 or higher, ρ cannot, in general, be positive definite and it is only a Finsler rather than a hermitian metric. However the holomorphic sectional curvature may be defined for a Finsler metric and the condition that the curvature is negative implies that X is hyperbolic. It is known [Cao and Wong 2003] that the ampleness of T ∗ X is equivalent to the existence of a Finsler metric with negative holomorphic bisectional curvature (hbsc):
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
109
Theorem [Aikou 1995; 1998; Cao and Wong 2003]. T ∗ X is ample ⇐⇒ Finsler metric has negative hbsc =⇒ Finsler metric has negative hsc =⇒ X is hyperbolic. In our view the fundamental problems in hyperbolic geometry are the following. Problem 1. Find an algebraic geometric characterization of the concept of negative hsc. Problem 2. Find an algebraic geometric and a differential geometric characterization of hyperbolicity. It is known that there are hyperbolic hypersurfaces in P n (for each n). On the other hand, there are no global regular 1forms on hypersurfaces in P n for n ≥ 3; Jm ∗ T X) = 0 for all m. These hyperbolic hypersurfaces are discovindeed h0 ( ered using, in one form or another, the Second Main Theorem of Nevanlinna Theory, which involves higherorder information; see for example [Wong 1989; Stoll and Wong 1994; Fujimoto 2001]. This leads us to the concept of the (parametrized) jet bundles [Ehresmann 1952], formalized (for complex manifolds) and studied by Green and Griffiths [1980]. Observe that a complex tangent v at a point x of a manifold may be represented by the first order derivative f 0 (0) of a local holomorphic map f : ∆r → X, f (0) = x for some disc ∆r of radius r in the complex plane C (more precisely, v is the equivalence class of such maps, as different maps may have the same derivative at the origin). A kjet is defined as the equivalence class of the first kth order derivatives of local holomorphic maps and the kjet bundle, denoted J k X, is just the collection of all (equivalence classes of) kjets. Note that J 1 X = T X. For k ≥ 2 these bundles are C ∗ bundles but not vector bundles. The nonlinear structure is reflected in reparametrization. Namely, given a kjet j k f (0) = (f (0), f 0 (0), . . . , f (k) (0)) we obtain another kjet by composing f with another local holomorphic self map φ in C that preserves the origin, then taking j k (f ◦ φ)(0). In particular, if φ is given by multiplication by a complex number λ we see that j k (f ◦ φ)(0) = (f (0), λf 0 (0), λ2 f 00 (0), . . . , λk f (k) (0)). Equivalence under this action is denoted by λ · j f (0) and this is the C ∗ action on J k X; in general there is no vector bundle structure on J k X. We write: λ · (v1 , . . . , vk ) = (λv1 , λ2 v2 , . . . , λk vk ), (v1 , . . . , vk ) ∈ J k X
(∗∗∗)
and assign the weight i to the variable vi . A 1form ω may be regarded as a holomorphic function on the tangent bundle ω : T X → C satisfying the condition ω(λ · v) = λω(v), that is, linearity along the fibers. More generally, an element Jm ∗ T X) is a holomorphic function on the tangent bundle ω : T X → C ω ∈ H 0( that is a homogeneous polynomial of degree m along the fibers. Analogously we define a kjet differential ω of weight m to be a holomorphic function on the kjet bundle ω : J k X → C which is a weighted homogeneous polynomial of degree m along the fibers. The sheaf of kjet differentials of weight m will be denoted by
110
KAREN CHANDLER AND PITMANN WONG
Jkm X. Taking a basis ω1 , . . . , ωN of H 0 (Jkm X) we define ρ(v1 , . . . , vk ) =
N X ωi (v1 , . . . , vk ) 1/m , i=1
and from (∗∗∗) we see (since each ωi is a weighted homogeneous polynomial of degree m) that ρ is a Finsler pseudometric, that is, ρ(λ · (v1 , . . . , vk )) =
N X
λm ωi (v1 , . . . , vk )
i=1
= λ
N X
ωi (v1 , . . . , vk )
i=1
1/m
1/m
= λ ρ(v1 , . . . , vk ).
The positive definiteness is a separate issue that one must deal with in higher dimension. The algebraic geometric concept that is equivalent to the positive definiteness of ρ is that the sheaf Jkm X is spanned (meaning that global sections span the fiber at each point). Other relevant concepts here are whether such a sheaf is ample, nef (numerically effective), or big. These concepts are intimately related to the Chern numbers of the sheaf Jkm X and the dimensions of the cohomology groups hi (Jkm X), 0 ≤ i ≤ n = dim X. The starting point here is the computation of the Euler characteristic in the case of a manifold of general type by the Riemann–Roch Formula. An asymptotic expansion of χ(J km X) was given in [Green and Griffiths 1980] with a sketch of the proof. Often in articles making reference to this result readers questioned the validity of the statement. A detailed proof, in the case of general type surfaces (complex dimension 2), of this formula was given in [Stoll and Wong 2002] using a different approach to that given in by Green and Griffiths. Indeed explicit formulas, not merely asymptotic expansions, were given for Jkm X, k = 2 and 3. The method of computation also shows that Jkm X is not semistable (see Section 3, Remark 3.5) in the sense of Mumford–Takemoto despite the fact (see [Maruyama 1981; Tsuji Jm ∗ 1987; 1988]) that all tensor products ⊗T ∗ X and symmetric products T X are semistable if X is of general type. In this article we also introduce an analogue of semistability in the sense of Gieseker–Maruyama (see [Okonek et al. 1980]) and show that Jkm X is not semistable (see Section 7) in this sense either. We have (see Section 5 for the reason in choosing the weight k! below) T ∗ X is ample =⇒ Jkk! X is ample for all k =⇒ Jkk! X is ample for some k ⇐⇒ there exists Finsler metric on J k X with negative hbsc =⇒ there exists Finsler metric on J k X with negative hsc =⇒ X is hyperbolic.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
111
The condition that Jkk! X is ample is much stronger than hyperbolicity of X. A weaker condition is that Jkk! X is big; this says that h0 (Jkk!m X) = dim(X, Jkk!m X) = O(mn(k+1)−1 ), where n = dim X. From the differential geometric point of view, this means that there is a pseudoFinsler metric on J k X that is generically positive definite and has generically negative hbsc (as defined wherever the metric is positive definite). The condition that Jkk! X is big implies that, for any ample divisor D on X, there exists m0 = m0 (D) such that Jkk!m0 X ⊗ [−D] (the sheaf of kjet differentials of weight k!m0 vanishing along D) is big. This, however, is not quite enough to guarantee hyperbolicity; the problem is that the base locus of J kk! X ⊗ [−D] may be “too big”. As a natural correction, we verify, using the Schwarz Lemma for jet differentials (see Theorem 6.1, Corollaries 6.2 and 6.3), that the assumption Jkk!m0 X ⊗ [−D] is big and spanned (that is, the base locus is empty) does imply hyperbolicity. However, the condition that a sheaf is big and spanned may be difficult to verify (unless, perhaps, it is already ample and we know of no hypersurfaces in P 3 satisfying this condition). To alleviate this, we refine the form of Schwarz’s Lemma (see Theorem 6.4 and Corollary 6.5) to establish the result that every holomorphic map f : C → X is algebraically degenerate if the dimension of the base locus of Jkk!m0 X ⊗ [−D] in the projectivized jet bundle P(J k X) is no more than n + k − 1. From this we show in Section 7, using the explicit computation of the invariants of the jet bundles in the first 3 sections (see Theorem 3.9 and Corollary 3.10), that the dimension of the base locus of a generic hypersurface of degree ≥ 5 in P 3 is, indeed, at most n + k − 1 = k + 1 (n = 2 in this case) and consequently, hyperbolic. The key ingredient is the extension of the inductive cutting procedure of the base locus, of [Lu and Yau 1990] and [Lu 1991] (see also [Dethloff et al. 1995a; 1995b]) in the case of 1jets, to kjets. There is a delicate point in the cutting procedure, namely that intersections of irreducible varieties may not be irreducible. We show, again using the Schwarz Lemma, that under the algebraically nondegenerate assumption on f , there is no loss of generality in assuming that the intersection is irreducible (see the proof of Theorems 7.18 and 7.20). The crucial analytic tools here are the Schwarz Lemma for jet differentials and its refined form. These are established using Nevanlinna Theory. We remark that jet differentials are used routinely in Nevanlinna Theory without a priori knowledge of whether regular jet differentials exist at all. The main idea of the proof of the Schwarz Lemma is to use jet differentials with logarithmic poles; to determine conditions under which the sheaf of such jet differentials is spanned and provides a singular Finsler metric that is positive definite in the extended sense. The classical Nevanlinna Theory is seen to work well with nonhermitian Finsler metrics with logarithmic poles on account of the fundamental principle
112
KAREN CHANDLER AND PITMANN WONG
(the Lemma of logarithmic derivatives of Nevanlinna) that logarithmic poles are relatively harmless (see the proof of Theorem 6.1 in Section 6 for details). The article is organized as follows. We describe the parametrized jet bundles of Green–Griffiths, which differ from the usual jet bundles; for example, they are C ∗ bundles but in general not vector bundles. The definitions are recalled in Section 1. For the usual jet bundles there is the question of interpolation: Find all varieties with prescribed jets, say, at a finite number of points. This problem, for 1jets, is equivalent to the Waring problem concerning when a general homogeneous polynomial is the sum of powers of linear forms. The Waring problem is related to the explicit construction (not merely existence) of hyperbolic hypersurfaces in P n for any n. Limitation of space does not allow us to discuss this problem in this article. Solutions of the interpolation problem for a collection of points can be found in [Alexander and Hirschowitz 1992a; 1992b; 1995; Chandler 1995; 1998a; 2002]. The analogous problem concerning the Green–Griffiths jet bundles is still open. In Section 2 we give a fairly detailed account of the jet bundles of curves. We calculate the Chern number c1 (Jkm X) and the invariants h0 (Jkm X) and h1 (Jkm X). We show, by examples, how to construct jet differentials explicitly, in terms of the defining polynomial, in the case of curves of degree d ≥ 4 in P 2 . Jet bundles may also be defined for varieties defined over fairly general fields (even in positive characteristic). The explicit construction of sections of powers m of the canonical bundle, KX , was useful in the solution of the “strong uniqueness polynomial problem” (see Section 2 and the articles [An et al. 2004] in the complex case and [An et al. 2003a; 2003b] in the case of positive characteristic). The formulas for invariants of the jet differentials for surfaces (the Chern numbers, the index, the Euler characteristic, the dimensions of cohomology groups) are given in Sections 3, 4 and 7. The calculations are similar to those over curves, though combinatorially much more complicated. We provide computations in special cases; the details are given in [Stoll and Wong 2002]. For example the explicit computation in Section 7 (see Theorem 7.7) shows that, for a smooth hypersurface in P 3 the Euler characteristic χ(J2m X) is big if and only if the degree is ≥ 16. In Section 6 we prove a Schwarz Lemma for jet differentials. This is the generalization of the classical result for differential forms on curves: if ω ∈ H 0 (X, KX ⊗ [−D]), that is, if ω is a regular 1form vanishing on an effective ample divisor D in the curve X, then f ∗ (ω) ≡ 0 for any holomorphic map f : C → X. This says that f 0 vanishes identically, that is, f is a constant. The proof given in Section 6 of the Schwarz Lemma for jets j k f has been in circulation since 1994 but was never formally published; it was used, for example, in the thesis of Jung [1995] and by Cherry–Ru in the context of padic jet differentials.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
113
For surfaces of general type the fact that Jkk! X is big for k 0 is equivalent k m to the “hyperplane” line bundle Lk! k being big on P(J X). (Note that Lk is locally free only if m is divisible by k!; see Section 5 for more details.) Schwarz’s Lemma then implies that the lifting [j k f ] : C → P(J k X) of a holomorphic map f : C → X must be contained in some divisor Y ⊂ P(J k X). The idea is to show k that Lk! k Y is again big so that the image of [j f ] is contained in a divisor Z of Y . k! Then we show that Lk Z is big and continue until we reach the critical dimension n + k − 1 = k + 1. For surfaces of general type the sheaf of 1jet differentials L 11 is big if c21 > c2 . In order for the restriction of L11 to subvarieties to be big, the condition that the index c21 − 2c2 is positive is required. The proof is based on the intersection theory of the projectivized tangent bundle P(T X) and the fact that the cotangent bundle T ∗ X of a surface of general type is semistable (in the sense of Mumford–Takemoto) relative to the canonical class. As remarked earlier the kjet differentials are not semistable for any k ≥ 2. However, by our explicit computation, for minimal surfaces of general type the index of J kk! X is positive for k 0. Indeed, we may write the index as ι(Jkk! X) = c(αk c21 − βk c2 ) where c, αk and βk are positive and we show that limk→∞ αk /βk = ∞ (see Corollary 3.10). This is crucial in showing that Lk! k Z is big in the cutting procedure. For example, for a smooth hypersurface of degree 5 in P 3 , ι(Jkk! X) is positive and the ratio αk /βk must be greater than 11 in order to establish the degeneracy of a map from C to X. Using the explicit expressions for α k and βk we show, with the aid of computer, that this occurs precisely for k ≥ 199 (see the table at the end of Section 3 and Example 7.6 in Section 7). However, in order for the index of the restriction of the sheaf to subvarieties (in the cutting procedure) to be positive (verifying the hyperbolicity of X), k must be even larger. Using our formulas in the proof of Theorem 7.20, Professor B. Hu, using the computer, checked that k ≥ 2283 is sufficient. Note. Experts who are familiar with parametrized jet bundles and are interested mainly in the proof of the Kobayashi conjecture may skip the first five sections (with the exception of Theorem 3.9 and Corollary 3.10) and proceed directly to Sections 6 and 7.
1. Holomorphic Jet Bundles Summary. Two notions of jet bundles, the full and the parametrized bundles, are introduced . The parametrized jet bundle is only a C ∗ bundle, not a vector bundle in general . For the resolution of the Kobayashi conjecture, as dictated by analysis, it is necessary to work with the parametrized jet bundle. (See Section 6 on the Schwarz Lemma.) Some basic facts are recalled here, all of which may be found in [Green and Griffiths 1980; Stoll and Wong 2002]. There are, in the literature, two different concepts of jet bundles of a complex manifold. The first is used by analysts (PDE), algebraic geometers [Chandler
114
KAREN CHANDLER AND PITMANN WONG
1995; 1998a; 2002] and also by number theorists (see Faltings’s work on rational points of an ample subvariety of an abelian variety and integral points of the complement of an ample divisor of an abelian variety [Faltings 1991]); it was used implicitly in [Ru and Wong 1991] (see also [Wong 1993b]) for the proof that there are only finitely many integral points in the complement of 2n + 1 hyperplanes in general position in P n . The second is the jet bundles introduced by Green and Griffiths [1980] (see also [Stoll and Wong 2002]). The first notion shall henceforth be referred to as the full jet bundle and these bundles are holomorphic vector bundles (locally free). The second notion of jet bundle shall be referred to as the parametrized jet bundle. These bundles are coherent sheaves that are holomorphic C ∗ bundles which, in general, are not locally free. For a complex manifold X the (locally free) sheaf of germs of holomorphic tangent vector fields (differential operators of order 1) of X shall be denoted by T 1 X or simply T X. An element of T 1 X acts on the sheaf of germs of holomorphic functions by differentiation: (D, f ) ∈ T 1 X × OX 7→ Df ∈ OX and the action is linear over C; in symbols, D ∈ HomC (OX , OX ). This concept may be extended as follows: Definition 1.1. Let X be a complex manifold of dimension n. The sheaf of germs of holomorphic kjets (differential operators of order k), denoted T k X, is the subsheaf of the sheaf of germs of homomorphisms Hom C (OX , OX ) consisting of elements (differential operators) of the form k X X
Di 1 ◦ · · · ◦ D i j ,
j=1 ij ∈N
where Dij ∈ T 1 X. In terms of holomorphic coordinates z1 , . . . , zn an element of T k X is expressed as k X
X
j=1 1≤i1 ,...,ij ≤n
ai1 ,...,ij
∂j , ∂zi1 . . . ∂zij
where the coefficients ai1 ,...,ij are symmetric in the indices i1 , . . . , ij . The bundle T k X is locally free. One may see this by observing that T k−1 X injects into T k X and there is an exact sequence of sheaves: 0 → T k−1 X → T k X → T k X/T k−1 X → 0, (1.1) Jk 1 T X is the sheaf of germs of kfold symmetric prodwhere T k X/T k−1 X ∼ = ucts of T 1 X. These exact sequences imply, by induction, that T k X is locally Jk 1 T X, a symmetric product of the tangent sheaf, is locally free as each sheaf free. A proof of (1.1) can be found in [Stoll and Wong 2002].
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
115
The parametrized kjet bundles for complex manifolds are introduced by Green– Griffiths. (These are special cases of the general theory of jets due to Ehresmann [1952] for differentiable manifolds.) These bundles are defined as follows. Let Hx , x ∈ X, be the sheaf of germs of holomorphic curves: {f : ∆ r → X, f (0) = x} where ∆r is the disc of radius r in C. For k ∈ N, define an equivalence relation (p) (p) ∼k by designating two elements f, g ∈ Hx as kequivalent if fj (0) = gj (0) for all 1 ≤ p ≤ k, where fi = zi ◦ f and z1 , . . . , zn are local holomorphic coordinates near x. The sheaf of parametrized kjets is defined by [ Hx / ∼ k . (1.2) J kX = x∈X
Elements of J X will be denoted by j k f (0) = (f (0), f 0 (0), . . . , f (k) (0)). The fact that J k X, k ≥ 2, is in general not locally free may be seen from the nonlinearity of change of coordinates: k
(wj ◦ f )0 =
n X ∂wj i=1
(wj ◦ f )00 =
n X ∂wj i=1
and for each k, (wj ◦ f )
(k)
=
∂zi
n X ∂wj i=1
∂zi
∂zi
(f )(zi ◦ f )00 +
(f )(zi ◦ f )
(f )(zi ◦ f )0 ,
n X ∂ 2 wj (f )(zi ◦ f )0 (zk ◦ f )0 ∂zi ∂zk
i,k=1
(k)
+P
∂ l wj (f ), (wj ◦ f )(l) , ∂zi1 . . . ∂zil
where P is an integercoefficient polynomial in ∂ l wj /∂zi1 . . . ∂zil and (wj ◦ f )(l) for j = 1, . . . , n and l = 1, . . . , k. There is, however, a natural C ∗ action on J k X defined via parameterization. Namely, for λ ∈ C ∗ and f ∈ Hx a map fλ ∈ (k) Hx is defined by fλ (t) = f (λt). Then j k fλ (0) = (fλ (0), fλ0 (0), . . . , fλ (0)) = (f (0), λf 0 (0), . . . , λk f (k) (0)). So the C ∗ action is given by λ · j k f (0) = (f (0), λf 0 (0), . . . , λk f (k) (0)).
(1.3)
Definition 1.2. The parametrized kjet bundle is defined to be J k X together with the C ∗ action defined by (1.3) and shall simply be denoted by J k X. It is clear that, for a complex manifold of (complex) dimension n, J k X is a holomorphic C ∗ bundle of rank r = kn and T k X is a holomorphic vector bundle Pk of rank r = i=1 Cin+i−1 where Cij are the usual binomial coefficients. Although J 1 X = T 1 X = T X these bundles differ for k ≥ 2. The nonlinearity of the change of coordinates formulas above shows that there is in general no natural way of injecting J k−1 X into J k X as opposed to the case of T k X (see (1.1)). There is however a natural projection map (the forgetting map) p kl : J k X → J l X for any l ≤ k defined simply by pkl (j k f (0)) = j l f (0), (1.4)
116
KAREN CHANDLER AND PITMANN WONG
which then respects the C ∗ action defined by (1.3) and so is a C ∗ bundle morphism. If Φ : X → Y is a holomorphic map between the complex manifolds X and Y then the usual differential Φ∗ : T 1 X → T 1 Y is defined. More generally, the kth order differential Φk∗ : T k X → T k Y is given by def
Φk∗ = (D1 ◦ · · · ◦ Dk )(g) = D1 ◦ · · · ◦ Dk (g ◦ Φ)
(1.5)
for any g ∈ OY . The kth order induced map for the parametrized jet bundle, denoted J k Φ : J k X → J k Y , can also be defined: def
J k Φ(j k f (0)) = (Φ ◦ f )(k) (0)
(1.6)
for any j k f (0) ∈ J k X. For the parametrized jet bundle J k X there is another notion closely related to the differential: the natural lifting of a holomorphic curve. Namely, given any holomorphic map f : ∆r → X(0 < r ≤ ∞), the lifting j k f : ∆r/2 → J k X is defined by j k f (ζ) = j k g(0), ζ ∈ ∆r/2
(1.7)
where g(ξ) = f (ζ + ξ) is holomorphic for ξ ∈ ∆r/2 . Definition 1.3. The dual of the full jet bundles T k X shall be called the sheaf of germs of kjet forms and shall be denoted by Tk∗ X. For m ∈ N the mfold Jm ∗ symmetric product shall be denoted by Tk X and its global sections shall be called kjet forms of weight m. In this article we shall focus on the dual of the parametrized jet bundles defined as follows. Definition 1.4. The dual of J k X (i.e., the sheaf associated to the presheaf consisting of holomorphic maps ω : j k XU → C such that ω(λ·j k f ) = λm ω(j k f ) for all λ ∈ C ∗ and positive integer m) shall be referred to as the sheaf of germs of kjet differentials of weight m and shall be denoted by Jkm X. It follows from the definition of the C ∗ action on J k X that a kjet differential ω of weight m is of the form: X aI1 ,...,Ik (z)(f 0 )I1 . . . (f (k) )Ik , (1.8) ω(j k f ) = I1 +2I2 +···+kIk =m
where aI1 ,...Ik are holomorphic functions, Ij = (i1j , . . . , inj ), n = dim X are the multiindices with each ilj being a nonnegative integer and Ij  = i1j + · · · + inj . In terms of local coordinates (z1 , . . . , zn ), (k)
(f 0 )I1 . . . (f (k) )Ik = (f10 )i11 . . . (fn0 )in1 . . . (f1 )i1k . . . (fn(k) )ink . Further, the coefficients aI1 ,...Ik (z) are symmetric with respect to the indices in each Ij . More precisely, a(iσ1 (1)1 ,...,iσ1 (n)1 ),...,(iσk (1)k ,...,iσk (n)k ) = a(i11 ,...,in1 ),...,(i1k ,...,ink ) ,
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
117
where each σj , j = 1, . . . , n, is a permutation on n elements. For example, (f10 )2 (f200 )2 + f1000 f20 f200 + f10000 f200 + f100000 f20 is a 5jet differential of weight 6. There are several important naturally defined operators on jet differentials; m+1 the first is a derivation δ : Jkm X → Jk+1 X defined by def
δω(j k+1 f ) = (ω(j k f ))0 .
(1.9)
Note that in contrast to the exterior differentiation of differential forms δ ◦ δ 6= 0 on jet differentials. In particular, given a holomorphic function φ defined on some open neighborhood U in X, the kth iteration δ (k) of δ, δ (k) φ(j k f ) = (φ ◦ f )(k) ,
(1.10)
is a kjet differential of weight k. Another difference between jet differentials and exterior differential forms is that a lower order jet differential can be naturally associated to a jet differential of higher order. The natural projection pkl : J k X → J l X defined by pkl (j k f ) = j l f , for k ≥ l, induces an injection p∗kl : Jlm X → Jkm X defined by “forgetting” those derivatives higher than l: def
p∗kl ω(j k f ) = ω(pkl (j k f )) = ω(j l f ).
(1.11)
We shall simply write ω(j k f ) = ω(j l f ) if no confusion arises. The wedge (exterior) product of differential forms is replaced by taking symmetric product; the symmetric product of a kjet differential of weight m and a k 0 jet differential of weight m0 is a max{k, k 0 }jet differential of weight m + m0 . Pn Example 1.5. A 1jet differential is a differential 1form ω = i=1 ai (z)dzi . Let f = (f1 , . . . , fn ) : ∆r → X be a holomorphic map. Then ω(j 1 f ) =
n X
ai (f )dzi (f 0 ) =
i=1
n X
ai (f )fi0
i=1
and δω is a 2jet differential of weight 2, given by 2
1
0
δω(j f ) = (ω(j f )) =
n X i=1
ai (f )fi0
0
=
n n X X ∂ai (f )fi0 fj0 + ai (f )fi00 . ∂z j i=1 i,j=1
Analogously, δ 2 ω is a 3jet differential of weight 3, given by δ 2 ω(j 3 f ) =
n X
n n X X ∂ 2 ai ∂ai (f )fi0 fj0 fk0 + 3 (f )fi00 fj0 + ai (f )fi000 . ∂z ∂z ∂z j k j i,j=1 i=1 ij=1
The concept of jet bundles extends also to singular spaces. Let us remark on how this may be defined. One may locally embed an open set U of X as a subvariety in a smooth variety U ⊂ Y . At a point x ∈ U the stalk jet (J k Y )x is then defined, as Y is smooth. The stalk (J k X)x is defined as the subset k j f (0) ∈ (J k Y )x  f : ∆r → Y is holomorphic, f (0) = x and f (∆r ) ⊂ U .
118
KAREN CHANDLER AND PITMANN WONG
From the differential geometric point of view, properties of the full jet bundle Tk∗ X, as a vector bundle, are reflected by the curvatures of hermitian metrics along its fibers. The parametrized jet bundles, however, are only C ∗ bundles hence can only be equipped with Finsler metrics. A Finsler pseudometric (or a kjet pseudometric) on X is a map ρ = ρk : J k X → R ≥0 satisfying the condition ρ(λ · j k ) = λρ(j k ) for all λ ∈ C and j k ∈ J k X. It is said to be a Finsler metric if it is positive outside of the zero section. A (k − 1)jet (pseudo)metric (k ≥ 2) ρ k−1 can be considered as a kjet (pseudo)metric by the forgetting map: ρk−1 (j k ) := ρk−1 (j k−1 ). where j k = j k f (0) and j k−1 = j k−1 f (0). Define, for j k ∈ J k M, k ≥ 1, κk (j k ) = inf {1/r},
(1.12)
where the infimum is taken over all r such that Hrk (ζ) = {f : ∆r → X  f is holomorphic and j k f (0) = j k } is nonempty. For k = 1 this is the usual Kobayashi–Royden pseudometric on J 1 X = T X. Henceforth we shall refer to κk as the kth infinitesimal Kobayashi–Royden pseudometric. We shall also say that X is kjet hyperbolic if κk is indeed a Finsler metric; that is, κk (j k ) > 0 for each nonzero kjet j k . Thus 1jet hyperbolicity is the same as Kobayashi hyperbolicity. Since a holomorphic map f : ∆r → X such that j k f (0) = (z, ζ1 , . . . , ζk ) also satisfies j k−1 f (0) = (z, ζ1 , . . . , ζk−1 ), we obtain: κk (z, ζ1 , . . . , ζk ) ≥ κk−1 (z, ζ1 , . . . , ζk−1 ).
(1.13)
From this we see that (k − 1)jet hyperbolicity implies kjet hyperbolicity. Remark 1.6. The notion introduced above is not to be confused with the kdimensional (1 ≤ k ≤ n = dim X) Kobayashi pseudometric in the literature (see [Lang 1987], for example); ndimensional Kobayashi hyperbolicity is more commonly known as measure hyperbolicity. In general the kth Kobayashi–Royden metric does not have a good regularity property. It is wellknown that κ1 is uppersemicontinuous (see [Royden 1971] or [Kobayashi 1970]); a similar proof shows that the same is true for κ k for any k. It is also known that κ1 , in general, is not continuous; however it is continuous if X is complete hyperbolic (that is, the distance function associated to the metric κ 1 is complete). In particular, κ1 is continuous on a compact hyperbolic manifold X. On the other hand, using a partition of unity one may construct kjet metrics
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
119
that are continuous everywhere and smooth outside of the zero section. Consider first the space C nk = C n × . . . × C n {z }  k
∗
with C action λ · (z 1 , z 2 , . . . , z k ) 7→ (λz 1 , λ2 z 2 , . . . , λk z k ). Define, for Z = (z 1 , . . . , z k ) ∈ C nk : rk (Z) = (z 1 2k! + z 2 2k!/2 + · · · + z k 2k!/k )1/2k!
(1.14)
where z i  is the usual Euclidean norm on C n . Observe that rk (λ·Z) = λ rk (Z) and that rk is continuous on C nk , smooth outside of the origin. Indeed rk2k! is smooth on all of C nk . Alternatively we can take rk (Z) = z 1  + z 2 1/2 + · · · + z k 1/k ,
(1.15) C nk ,
which also satisfies rk (λ·Z) = λ rk (Z) and is continuous on smooth outside of the set [z 1 · z k = 0]. On a local trivialization J k XU ∼ = U × C nk we define simply ρk (z, Z) = rk (Z) on J k XU so that a global kjet metric is defined via a partition of unity subordinate to a locally finite trivialization cover. This general construction is of limited use as it does not take into account the geometry of the manifold. In the case of a compact manifold a more useful construction can be carried out by taking a basis ω1 , . . . , ωN of global holomorphic kjet differentials (provided that these exist), and defining ρk (j f ) = k
N X i=1
ωi (j f ) k
2
1/2
.
(1.16)
Then, since a jet differential is a linear functional on the kjet bundle (that is, ω(λ·j k f ) = λω(j k f )), we see readily that ρk (λ·j k f ) = λρk (j k f ). It is clear from the definition that ρk is continuous on J k X, real analytic on J k X\{zero section}; indeed, ρ2k is real analytic on J k X. For k = 1 use a basis of global holomorphic 1forms. The number N = h0 (T ∗ X) is the irregularity of X (for a Riemann surface this is just the genus of X). Thus the invariants h 0 (Jkm X) play an important role in the determination of hyperbolicity. The jet bundles may be defined, in an analogous way, over fairly general fields. We conclude this section by introducing a very interesting problem: Interpolation Problem. Find all subvarieties in P n = P n (K) (where K is an infinite field ) of a given degree d with prescribed jet spaces at a finite number of points. More precisely, given subspaces V1 ⊂ Txk1 P n , . . . , VN ⊂ TxkN P n (or V1 ⊂ Jxk1 P n , . . . , VN ⊂ JxkN P n ), find all varieties X of degree d such that V1 = Txk1 X, . . . , VN ⊂ TxkN X. At this time little is known about the problem for the bundle J k X however much is known in the case of T k X. For example, the following is known (see [Chandler
120
KAREN CHANDLER AND PITMANN WONG
1998a; 2002] or else [Alexander 1988; Alexander and Hirschowitz 1992a; 1992b; 1995]). Theorem 1.7. Let Ψ be a general collection of d points in P n . The codimension in H 0(OP n (3)) of the space of sections singular on Ψ is min{(n+1)d, (n+3)!/3!n!} unless n = 4 and d = 7. More generally, the codimension in H 0 (OP n (m)) of the space of sections singular on Ψ is equal to min{(n + 1)d, (n + m)!/m! n!} unless (n, m, d) = (2, 4, 5), (3, 4, 9), (4, 3, 7) or (4, 4, 14). The problem is related also to the Waring problem for linear forms: when can a general degree m form in n+1 variables be expressed as a sum of mth powers of linear forms? Let P S(n, m, d) be the space of homogeneous polynomials in n + 1 m variables expressible as Lm 1 + · · · + Ld , where L1 , . . . , Ld are linear forms. Then: Theorem 1.8. With the notation above, we have dim P S(n, m, d) = min{(n + 1)d, (n + m)!/m! n!} unless (n, m, d) = (2, 4, 5), (3, 4, 9), (4, 3, 7) and (4, 4, 14). For details, see the articles by Chandler and by Alexander and Hirschowitz in the references, as well as [Iarrobino and Kanev 1999].
2. Chern Classes and Cohomology Groups The Case of Curves Summary. The theory of parametrized jet bundles is complicated by their not being vector bundles. This section discusses the case of curves to acquaint readers with the theory in the simplest situation. The theory is based on the fundamental result of Green and Griffiths on the filtration of the parametrized jet bundles (see Theorem 2.3 and Corollary 2.4). The explicit computations of this section have numerous applications (see for example [An et al. 2003a; 2004; 2003b]). In this section we compute the Chern numbers and the invariants h i (Jkm X), i = 0, 1, of the jet bundles for curves. In the case of curves in P 2 we are interested in finding an explicit expression of a basis for h 0 (Jkm X). The procedure introduced here for the construction works as well for singular curves and in varieties defined over general differential fields. For applications in this direction to the strong uniqueness polynomial problem and the unique range set problem; see [An et al. 2004] in the complex case and [An et al. 2003a; 2003b] in the case of fields of positive characteristic. For the full jet bundles the computation of Chern classes and cohomology groups is straightforward. Dualizing the defining sequence (1.1) we get an exact sequence Jk ∗ ∗ T1 X → Tk∗ X → Tk−1 X → 0. (2.1) 0→
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
121
For example, for k = 3 the exact sequences J3 ∗ T1 X → T3∗ X → T2∗ X → 0, 0→ J2 ∗ 0→ T1 X → T2∗ X → T1∗ X → 0
and Whitney’s Formula yields J3 ∗ J2 ∗ J3 ∗ T1 X). c1 (T3∗ X) = c1 (T2∗ X) + c1 ( T1 X) + c1 ( T1 X) = c1 (T1∗ X) + c1 (
In general, we have, by induction:
Theorem 2.1. The first Chern number of the bundle Tk∗ X is given by c1 (Tk∗ X) =
k X
c1 (
j=1
Jj
T1∗ X).
In particular , if X is a Riemann surface, c1 (Tk∗ X) =
k X j=1
where KX =
T1∗ X
jc1 (T1∗ X) =
k(k + 1) c1 (KX ) = k(k + 1)(g − 1) 2
is the canonical bundle of X and g is the genus.
For a line bundle L and nonnegative integer i the ifold tensor product is denoted by Li and L−i is the dual of Li . (Recall that tensor product and symmetric product on line bundles are equivalent.) Theorem 2.2. Let X be a smooth curve of genus g ≥ 2. Then h0 (Tk∗ X) = k 2 (g − 1) + 1 and h1 (Tk∗ X) = 1. Proof. By Riemann–Roch for curves, i i i i h0 (KX ) − h1 (KX ) = χ(KX ) = χ(OX ) + c1 (KX )
= h0 (OX ) − h1 (OX ) + 2(g − 1)i = 1 − g + 2(g − 1)i i for any nonnegative integer i. Thus h0 (KX ) = 1−i 1−i 0 0 h (KX ) + (2i − 1)(g − 1). As h (KX ) = 1 for i i ≥ 2 we get 0, 1, i h0 (KX )= g, (2i − 1)(g − 1), 1−i i By duality, h1 (KX ) = h0 (KX ); hence 0, 1, i h1 (KX )= g, (1 − 2i)(g − 1),
i h1 (KX ) + (2i − 1)(g − 1) = 1−i ) = 0 for = 1 and h0 (KX
i < 0, i = 0, i = 1, i ≥ 2.
(2.2a)
i ≥ 2, i = 1, i = 0, i < 0.
(2.2b)
122
KAREN CHANDLER AND PITMANN WONG
From the short exact sequence (2.1) we get the exact sequence k ∗ k 0 → H 0 (KX ) → H 0 (Tk∗ X) → H 0 (Tk−1 X) → H 1 (KX )→ ∗ → H 1 (Tk∗ X) → H 1 (Tk−1 X) → 0. ∗ From (2.2a,b) we deduce that, for k ≥ 2, H 1 (Tk∗ X) = H 1 (Tk−1 X) and that ∗ k h0 (Tk∗ X) = h0 (Tk−1 X) + h0 (KX ).
These imply that h1 (Tk∗ X) = h1 (T ∗ X) = h1 (KX ) = h0 (OX ) = 1 for all k ≥ 1 and that h0 (Tk∗ X) =
k X i=1
i h0 (KX ) = g+
k X
(2i − 1)(g − 1) = k 2 (g − 1) + 1.
˜
i=2
The computation of Chern classes and cohomology groups for the parametrized jet bundles is somewhat more complicated. This depends on the fundamental filtration for these bundles due to Green–Griffiths. Let 0 → S 0 → S → S 00 → 0 be an exact sequence of sheaves. Then for any m there is a filtration 0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F m ⊂ F m+1 =
Jm
S
Ji 0 Jm−i 00 Jm S . AnalS ⊗ S, such that F i /F i−1 ∼ of the symmetric product = Vm ogously, for the exterior product S we have a filtration 0 = F 0 ⊂ F 1 ⊂ · · · ⊂ F m ⊂ F m+1 =
Vm
S
Vi 0 Vm−i 00 such that F i /F i−1 ∼ S ⊗ S . These filtrations connect the cohomology = groups of higher symmetric (resp. exterior) products to the cohomology groups of lower symmetric (resp. exterior) products. The analogue of these is the following theorem of Green and Griffiths (the proof can be found in [Stoll and Wong 2002]): Theorem 2.3. There exists a filtration of Jkm X: [m/k]
m Jk−1 X = Fk0 ⊂ Fk1 ⊂ · · · ⊂ Fk
= Jkm X
(where [m/k] is the greatest integer smaller than or equal to m/k) such that Ji ∗ m−ki T X). Fki /Fki−1 ∼ X ⊗( = Jk−1
As an immediate consequence [Green and Griffiths 1980], we have:
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
123
Corollary 2.4. Let X be a smooth projective variety. Then Jkm X admits a composition series whose factors consist precisely of all bundles of the form: Ji Ji ( 1 T ∗ X) ⊗ · · · ⊗ ( k T ∗ X) where ij ranges over all nonnegative integers satisfying i1 + 2i2 + · · · + kik = m. The first Chern number of Jkm X is given by X Ji Ji c1 (Jkm X) = c1 ( 1 T ∗ X) ⊗ · · · ⊗ ( k T ∗ X) . i1 +2i2 +···+kik =m ij ∈Z ≥0
In particular , if X is a curve then X c1 (Jkm X) =
(i1 + i2 + · · · + ik )c1 (T ∗ X).
i1 +2i2 +···+kik =m ij ∈Z ≥0
Example 2.5 [Stoll and Wong 2002]. It is clear that for m < k the filtration m m degenerates and we have Jkm X = Jk−1 X = . . . = Jm X. In particular, J21 X = 1 ∗ J1 X = T X. For m = k = 2, the filtration is given by J2 ∗ T X = J12 X = S20 ⊂ S21 = J22 X, S21 /S20 ∼ = T ∗ X,
so we have the exact sequence J2 ∗ T X → J22 X → T ∗ X → 0. 0→
Thus the first Chern numbers are related by the formula J2 ∗ c1 (J22 X) = c1 ( T X) + c1 (T ∗ X).
Analogously, J31 X = J21 X = J11 X = T ∗ X and J32 X = J22 X. The filtration of J33 X is as follows: J33 X = S31 ⊃ S30 = J23 X, J33 X/J23 X = S31 /S30 ∼ = T ∗ X. Hence we have an exact sequence 0 → J23 X → J33 X → T ∗ X → 0. Now the filtration of J23 X is J23 X = S21 ⊃ S20 = J13 X, J23 X/J13 X ∼ = T ∗X ⊗ T ∗X J3 ∗ and, since J13 X = T X, we have an exact sequence J3 ∗ 0→ T X → J23 X → T ∗ X ⊗ T ∗ X → 0. From these two exact sequences we get
c1 (J33 X) = c1 (T ∗ X) + c1 (T ∗ X ⊗ T ∗ X) + c1 (
J3
T ∗ X).
From basic representation theory (or linear algebra in this special case) we have V2 ∗ J2 ∗ T X hence T X⊕ T ∗X ⊗ T ∗X = V2 J3 ∗ J2 ∗ T X) + c1 ( T ∗ X). T X) + c1 ( c1 (J33 X) = c1 (T ∗ X) + c1 (
124
KAREN CHANDLER AND PITMANN WONG
V2 ∗ ∗ T X is the Weyl module W1,1 X associated to the In representation theory partition {1, 1} (see [Fulton and Harris 1991]). Thus we have c1 (J33 X) =
3 X
c1 (
j=1
Jj
In the special case of a Riemann surface curve we have
∗ T ∗ X) + c1 (W1,1 X).
V2
T ∗ X is the zerosheaf. Thus for a
c1 (J33 X) = (1 + 2 + 3)c1 (T ∗ X) = 6c1 (T ∗ X). For m = k = 4, we have the filtrations J44 X = S41 ⊃ S40 = J34 X, J44 X/J34 X = S41 /S40 ∼ = T ∗ X, J34 X = S31 ⊃ S30 = J24 X, J34 X/J24 X = S31 /S30 ∼ = T ∗ X ⊗T ∗ X, and 4 2 1 0 4 J2 X = S2 ⊃ S2 ⊃ S2 = J1 X, with J2 ∗ J2 ∗ J24 X/S21 = T X). T X, S21 /S20 ∼ = T ∗X ⊗ (
Thus the Chern number is given by
c1 (J44 X) = c1 (T ∗ X) + c1 (T ∗ X ⊗ T ∗ X) + c1 (
J2
T ∗ X) J4 ∗ J2 ∗ T X). T X)) + c1 ( + c1 (T ∗ X ⊗ (
From elementary representation theory we obtain Jk ∗ Jk+1 ∗ ∗ T ∗X ⊗ ( T X) = Wk,1 X ⊕( T X)
∗ where Wk,1 is the Weyl module associated to the partition {k, 1} so that
c1 (J44 X) = c1 (
J2
T ∗ X) +
In particular, if X is a curve,
P4
i=1 c1 (
Ji
T ∗ X) +
P2
∗ i=1 c1 (Wj,1 X).
c1 (J44 X) = (1 + 2 + 2 + 3 + 4)c1 (T ∗ X) = 12c1 (T ∗ X). The procedure can be carried out further; for instance, Jj ∗ P3 c1 (J55 X) = j=2 c1 ( T X) P5 Jj ∗ P2 P3 ∗ ∗ + j=1 c1 ( T X) + j=1 c1 (Wj,1 X) + j=1 c1 (Wj,1 X), J 3 ∗ 6 ∗ ∗ ∗ ∗ T X) c1 (J6 X) = c1 (T X) + 3c1 (T X ⊗ T X) + 2c1 T X ⊗ ( J2 ∗ J3 ∗ J2 ∗ J2 ∗ + c1 ( T X) + c1 ( T X) + c1 ( T X) ⊗ ( T X) J6 ∗ J4 ∗ ∗ T X). T X) + c1 ( + c1 T X ⊗ (
So if X is a curve we have
c1 (J55 X) = (1 + 2 + 3 + 2 + 3 + 4 + 5)c1 (T ∗ X) = 20c1 (T ∗ X), c1 (J66 X) = (1 + 6 + 8 + 2 + 3 + 4 + 5 + 6)c1 (T ∗ X) = 35c1 (T ∗ X).
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
125
The calculation of the sum X
i1 + · · · + i k
(2.3)
i1 +2i2 +···+kik =m
can be carried out using standard combinatorial results which we now describe. Definition 2.6. (i) A maximal set of mutually conjugate elements of S m (the symmetric group on m elements) is said to be a class of S m . (ii) A partition of a natural number m is a set of positive integers i 1 , . . . , iq such that m = i1 + · · · + iq . The following asymptotic result concerning the number of partitions of a positive integer m is wellknown in representation theory and in combinatorics [Hardy and Wright 1970]: Theorem 2.7. The number of partitions of m, the number of classes of Sm and the number of (inequivalent) irreducible representations of Sm are equal . This common number p(m) is asymptotically approximated by the formula of Hardy–Ramanujan: √ eπ 2m/3 √ . p(m) ∼ 4m 3 The first few partition numbers are p(1) = 1, p(2) = 2, p(3) = 3, p(4) = 5, p(5) = 7, p(6) = 11, p(7) = 15, p(8) = 22, p(9) = 30, p(10) = 42, p(11) = 56, p(12) = 77, p(13) = 101. Consider first the case of partitioning a number by partitions of a fixed length k. Denote by pk (m) the number of positive integral solutions of the equation x1 + · · · + x k = m with the condition that 1 ≤ xk ≤ xk−1 ≤ . . . ≤ x1 . This number is equal to the number of integer solutions of the equation y1 + · · · + y k = m − k with the condition that the solutions be nonnegative and 0 ≤ y k ≤ yk−1 ≤ . . . ≤ y1 . If exactly i of the integers {y1 , . . . , yk } are positive then these are the solutions of x1 + · · · + xi = m − k and so there are pi (m − k) of such solutions. Consequently we have (see [Stoll and Wong 2002] for more details): Pm Lemma 2.8. With the notation above we have: p(m) = k=1 pk (m), where pk (m) =
k X
pi (m − k),
i=0
for 1 ≤ k ≤ m and with the convention that p0 (0) = 1, p0 (m) = 0 if m > 0 and pk (m) = 0 if k > m. Moreover , the number pk (m) satisfies the following recursive relation: pk (m) = pk−1 (m − 1) + pk (m − k).
126
KAREN CHANDLER AND PITMANN WONG
Example 2.9. We shall compute p(6) and p(7) using the preceding lemma. We have p1 (m) = pm (m) = 1 and p2 (m) = m/2 or (m − 1)/2 according to m being even or odd; thus p1 (6) = 1, p2 (6) = 3, p6 (6) = 1. Analogously, we have: p3 (m) = p2 (m − 1) + p3 (m − 3), p4 (m) = p3 (m − 1) + p4 (m − 4), p5 (m) = p4 (m − 1) + p5 (m − 5), so that, for example: p3 (6) = p2 (5) + p3 (3) = 2 + 1 = 3, p4 (6) = p3 (5) + p4 (2) = p2 (4) = 2, p5 (6) = p4 (5) = p3 (4) = p2 (3) = 1. Since p(m) =
Pm
k=1
pk (m) we have
p(6) =
6 X
pk (6) = 1 + 3 + 3 + 2 + 1 + 1 = 11.
k=1
For m = 7 we have p1 (7) = 1, p2 (7) = 3, p7 (7) = 1, p3 (7) = p2 (6) + p3 (4) = p2 (6) + p2 (3) = 4, p4 (7) = p3 (6) = 3, p5 (7) = p4 (6) = 2, p6 (7) = p5 (6) = 1; hence p(7) =
7 X
pk (7) = 1 + 3 + 4 + 3 + 2 + 1 + 1 = 15.
k=1
For k ≤ m denote by Lk (m) the sum of the lengths of all partitions λ of m whose length lλ is at most k: X lλ . Lk (m) = λ, lλ ≤k
The next lemma follows from the definitions [Wong 1999; Stoll and Wong 2002]:
Lemma 2.10. With notation as above we have Lk (m) =
X
λ,lλ ≤k
lλ =
k X
jpj (m) =
k X X
ij ,
λ, lλ ≤k j=1
j=1
where the sum on the right is taken over all partitions λ = (λ1 , . . . , λρλ ) of m, 1 ≤ λlλ ≤ . . . ≤ λ2 ≤ λ1 , lλ ≤ k and ij is the number of j’s in {λ1 , . . . , λlλ }. For k = m, L(m) = Lm (m) is the total length of all possible partitions of m. For example if m = 6 then L(6) = 1 + 6 + 9 + 8 + 5 + 6 = 35 and for m = 7, L(7) = 1 + 6 + 12 + 12 + 10 + 6 + 7 = 54. Indeed we have: Theorem 2.11. If X is a nonsingular projective curve, the Chern number of m Jm X is m X m c1 (Jm X) = Lm (m)c1 (KX ) = jpj (m)c1 (KX ), j=1
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
127
where KX is the canonical bundle of X. There is a formula for the asymptotic behavior of pk (m): Theorem 2.12. For k fixed and m → ∞ the number pk (m) is asymptotically pk (m) ∼
mk−1 . (k−1)! k!
We give below the explicit calculation of the above in the first few cases. For m = k = 3, we have p(3) = 3 and the possible indices are Pk λ l λ d λ i1 i2 i3 j=1 ij 1 2 3
(1, 1, 1) (2, 1) (3)
3 2 1
1 2 1
3 1 0
0 1 0
0 0 1
3 2 1
The cases cases correspond to the possible partitions of 3: 1+1+1 = 3, 2+1 = 3 and 3 = 3, of respective lengths 3, 2, and 1. The Chern number c 1 (J33 X) of a curve X is obtained by summing the last column: c 1 (J33 X) = (1 + 2 + 3) × c1 (T ∗ X) = 6c1 (T ∗ X). For m = k = 4 the number of partitions is p(4) = 5 and we have Pk λ l λ d λ i1 i2 i3 i4 j=1 ij 1 2 3 4 5
(1, 1, 1, 1) (2, 1, 1) (3, 1) (2, 2) (4)
4 3 2 2 1
1 3 3 2 1
4 2 1 0 0
0 1 0 2 0
0 0 1 0 0
0 0 0 0 1
4 3 2 2 1
λ
ρλ
dλ
i1
i2
i3
i4
i5
(1, 1, 1, 1, 1) (2, 1, 1, 1) (3, 1, 1) (2, 2, 1) (4, 1) (3, 2) (5)
5 4 3 3 2 2 1
1 4 6 5 4 15 1
5 3 2 1 1 0 0
0 1 0 2 0 1 0
0 0 1 0 0 1 0
0 0 0 0 1 0 0
0 0 0 0 0 0 1
and c1 (J44 X) = 12c1 (T ∗ X). For m = k = 5, p(5) = 7,
1 2 3 4 5 6 7
and c1 (J55 X) = 20c1 (T ∗ X). For m = k = 6, p(6) = 11,
Pk
j=1 ij
5 4 3 3 2 2 1
128
KAREN CHANDLER AND PITMANN WONG
λ 1 (1, 1, 1, 1, 1, 1) 2 (2, 1, 1, 1, 1) 3 (3, 1, 1, 1) 4 (2, 2, 1, 1) 5 (4, 1, 1) 6 (3, 2, 1) (2, 2, 2) 7 8 (5, 1) 9 (4, 2) 10 (3, 3) 11 (6)
lλ
dλ
i1
i2
i3
i4
i5
i6
6 5 4 4 3 3 3 2 2 2 1
1 5 10 9 10 36 5 30 9 5 1
6 4 3 2 2 1 0 1 0 0 0
0 1 0 2 0 1 3 0 1 0 0
0 0 1 0 0 1 0 0 0 2 0
0 0 0 0 1 0 0 0 1 0 0
0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0 1
Pk
j=1 ij
6 5 4 4 3 3 3 2 2 2 1
and c1 (J66 X) = 35c1 (T ∗ X). The next few values of Lk (k) are L7 (7) = 54, L8 (8) = 86, L9 (9) = 128, L10 (10) = 192, L11 (11) = 275, L12 (12) = 399, L13 (13) = 556, L14 (14) = 780, L15 (15) = 1068, L16 (16) = 1463. Next we deal with the problem of computing the invariants: h i (Jkm X) = dim H i (Jkm X) for a curve X of genus g ≥ 2. We have h0 (J11 X) = h0 (KX ) = g, h1 (J11 X) = h0 (OX ) = 1. For curves the filtration of Green–Griffiths takes the form [m/k]
Jkm X = Sk
m ⊃ · · · ⊃ Sk0 = Jk−1 X,
m−ki i (X). Ski /Ski−1 = KX ⊗ Jk−1
m−2i m−i i i Hence, for k = 2, S2i /S2i+1 = KX ⊗ J1m−2i = KX ⊗ KX = KX . It is clear 1 1 ∼ from the filtration that J1 X = J2 X (the isomorphism is given by the forgetting map (1.11)). For J22 X the filtration yields the short exact sequence J2 ∗ 2 0 → KX = T X → J22 X → T ∗ X = KX → 0,
from which we get the exact sequence
2 2 0 → H 0 (KX ) → H 0 (J22 X) → H 0 (KX ) → H 1 (KX ) → H 1 (J22 X) → H 1 (KX ) → 0. 2 By (2.2b) we have h1 (KX ) = 1 and h1 (KX ) = 0 if d ≥ 2. Hence, as KX = OX (3 − d), 2 h0 (J22 X) = h0 (KX ) + h0 (KX ) = 4g − 3,
h1 (J22 X) = h1 (T ∗ X) = h0 (OX ) = 1. For J23 X we obtain the short exact sequence from the filtration, 3 2 0 → KX → J23 X → KX → 0,
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
129
and the exact cohomology sequence 3 2 3 2 0 → H 0 (KX ) → H 0 (J23 X) → H 0 (KX ) → H 1 (KX ) → H 1 (J23 X) → H 1 (KX ) → 0. 2 3 Since h1 (KX ) = h1 (KX ) = 0 for g ≥ 2 we find
h1 (J23 X) = 0, 2 3 h0 (J23 X) = h0 (KX ) + h0 (KX ) = 8(g − 1).
For J24 X the filtration is given by 4 J24 X = S22 ⊃ S21 ⊃ S20 = J14 X = KX 2 3 with S22 /S21 = KX , S21 /S20 = KX . From the filtration we have two short exact sequences, 2 0 → S21 → J24 X → KX →0
4 3 and 0 → KX → S21 → KX → 0.
3 For g ≥ 2 we get from the second exact sequence and the fact that h 1 (KX )= 1 4 1 1 h (KX ) = 0 that h (S2 ) = 0. This and the first exact sequence imply that 2 h1 (J24 X) = h1 (KX ) = 0 and 2 3 4 h0 (J24 X) = h0 (KX ) + h0 (KX ) + h0 (KX ) = 15(g − 1).
We get, inductively: Theorem 2.13. For a smooth curve with genus g ≥ 2 the following equalities hold : (i) J21 X = KX ; hence h1 (J21 X) = 1, h0 (J21 X) = genus of X; 2 (ii) h1 (J22 X) = 1, h0 (J22 X) = h0 (KX ) + h0 (KX ) = 4g − 3; 1 m (iii) h (J2 X) = 0, and for m ≥ 3, h0 (J2m X) = (iv) for i ≥ 1,
P[m/2] j=0
m−j m h0 (KX ) = (2m − [ m 2 ] − 1)([ 2 ] + 1)(g − 1);
i+1 i h0 (J21 X ⊗ KX ) = h0 (KX ) = (2i + 1)(g − 1), i+1 i+2 i h0 (J22 X ⊗ KX ) = h0 (KX ) + h0 (KX ) = 4(i + 1)(g − 1),
and for m ≥ 3, i h0 (J2m X ⊗ KX )=
P[m/2] j=0
m+i−j m h0 (KX ) = (2m + 2i − [ m 2 ] − 1)([ 2 ] + 1)(g − 1).
Proof. Parts (i), (ii) and (iii) are clear. For part (iv), tensoring the exact 2 i sequence 0 → KX → J22 X → KX → 0 by KX yields the exact sequence 0 → i+2 i+1 2 i KX → J2 X ⊗ KX → KX → 0. From the associated long exact cohomology
130
KAREN CHANDLER AND PITMANN WONG
i+1 i+2 i sequence one sees that h0 (J22 X ⊗ KX ) = h0 (KX ) + h0 (KX ) as claimed. From the exact sequences
0→
[m/2]−1
Sm,2
m−2[m/2]
K[m/2] ⊗J1
→ J2m X →
[m/2]−2 Sm,2
[m/2]−1 Sm,2
m−[m/2]
X = KX
[m/2]−1 m−2([m/2]−1) KX ⊗J1 X
→ 0,
m−[m/2]+1 KX
0→ → → = → 0, ............................................................................. m 0 → J1m X = KX →
1 Sm,2
m−1 KX ⊗J1m−2 X = KX
→
→ 0,
i we obtain, by tensoring with KX , i ≥ 0, the exact sequences [m/2]−1
0 → Sm,2
m+i−[m/2]
i i ⊗KX → J2m X⊗KX → KX
[m/2]−2 i Sm,2 ⊗KX
[m/2]−1 i Sm,2 ⊗KX
→ 0,
m+i−[m/2]+1 KX
0→ → → → 0, ......................................................... 0→
m+i KX
→
1 i Sm,2 ⊗KX
→
m+i−1 KX [m/2]−j
i from which we deduce that h1 (J2m X ⊗ KX ) = h1 (Sm,2 j ≤ [m/2] and that
→ 0,
i ⊗ KX ) = 0 for 0 ≤
[m/2] i h0 (J2m X ⊗ KX )=
X
m+i−j ), h0 (KX
j=0
as claimed.
˜
The coefficient of (g − 1) in part (iv) of the preceding lemma may be expressed as ( 1 (3m2 +4m(i+1)+8i−4) = 41 (m+2)(3m+4i−2), m even, α(m, i, 2) = 14 1 2 4 (3m +2m(2i+1)+4i−1) = 4 (m+1)(3m+4i−1)4, m odd. The coefficient of (g − 1) in part (iii) is α(m, 2) = α(m, 0, 2). Corollary 2.14. For a smooth curve of genus g ≥ 2 we have, for m ≥ 3, ( 1 (3m2 + 4m − 4)(g − 1), m even, 0 m h (J2 X) = 41 2 4 (3m + 2m − 1)(g − 1), m odd , and c1 (J2m X)
=
(
1 2 4 (3m 1 2 4 (3m
+ 6m)(g − 1),
m even,
+ 4m + 1)(g − 1),
m odd .
Proof. The first formula is given by part (iii) of Theorem 2.13. The second formula is a consequence of the Riemann–Roch for curves: h0 (J2m X) − h1 (J2m X) = c1 (J2m X) − (rk J2m X)(g − 1), using the fact that rk J2m X = [m/2] + 1 and that h1 (J2m X) = 0 if m ≥ 3.
˜
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
131
We now deal with the case of general k. We shall be content with asymptotic formulas as the general formulas become complicated by the fact that the general formula for sums of powers is only given recursively. However the highest order term is quite simple: m X
id =
i=1
md+1 + O(md ). d+1
(2.4)
The filtration theorem of Green–Griffiths implies that X
rk Jkm X =
rk SI .
I∈Ik,m
For a curve, SI = Hence
J i1
T ∗X ⊗ · · · ⊗
Jik−1
T ∗X ⊗
J ik
I
i1 +···+ik T ∗ X = K X = KX .
Pk rk Jkm X = #Ik,m , Ik,m = I = (i1 , . . . , ik )  j=1 jij = m .
Alternatively, since SI = SI 0 ⊗ we have
J ik
T ∗ X, where I 0 = (i1 , . . . , ik−1 ) ∈ Ik−1,m−kik ,
[m/k]
rk
Jkm X
=
X
[m/k]
rk
m−kik (Jk−1 X
ik =0
ik ⊗ KX )
=
X
m−kik rk Jk−1 X;
ik =0
equivalently, [m/k]
#Ik,m =
X
#Ik−1,m−kik .
ik =0
Pk Theorem 2.15. Let Ik,m = I = (i1 , . . . , ik )  j=1 jij = m . Then, for a curve X, rk Jkm X = #Ik,m =
mk−1 + O(mk−2 ). k!(k − 1)!
Proof. It is clear that rk J1m X = 1 and we have seen that rk J2m X = [m/2]+1, thus writing rk Jkm X = ak mk−1 + O(mk−2 ) we get, via (2.4), [m/k]
ak mk−1 + O(mk−2 ) = ak−1
X
(m − kik )k−2 + O(mk−2 )
ik =0 [m/k]
= ak−1
X
ik =0
(m − kik )k−2 + O(mk−2 )
132
KAREN CHANDLER AND PITMANN WONG
= ak−1
k−2 X
(−1)j
j=0
= ak−1
k−2 X
k
(−1)j
j=0
=
[m/k] X j (k − 2)! mk−2−j k j ik + O(mk−2 ) j!(k − 2 − j)! i =0
(k − 2)! mj+1 + O(mk−2 ) mk−2−j k j j!(k − 2 − j)! (j + 1)k j+1
k−2 ak−1 X (−1)j (k − 2)! mk−1 + O(mk−2 ). k j=0 j + 1 j!(k − 2 − j)!
The following formula is easily verified by double induction: Lemma 2.16. For any positive integers 1 ≤ l ≤ k, we have k X (−1)j k j=0
j +l
j
=
(l − 1)! k! . (k + l)!
Using this lemma we obtain a recursive formula for k ≥ 2: ak−1 ak = , a1 = 1. k(k − 1) The first few values of ak are a1 = 1, a2 = 1/2, a3 = 1/22 3, a4 = 1/(24 32 ), a5 = 1/(26 32 5), a6 = 1/(27 33 52 ). The recursive formula also yields the general formula for ak : 1 1 = . ˜ ak = Q k (k − 1)! k! l=2 (l − 1)l
The filtration also yields a formula for a curve of genus g: X X X I(g − 1), Ic1 (KX ) = 2 c1 (SI ) = c1 (Jkm X) = I∈Ik,m
I∈Ik,m
I∈Ik,m
where I = i1 + · · · + ik . On the other hand we have [m/k]
c1 (Jkm X) =
X
X
(c1 (SI 0 ) + ik c1 (KX ))
ik =0 I 0 ∈Ik−1,m−kik [m/k]
=2
X
ik =0
It is clear that
I 0 ∈I
X
(I 0  + ik )(g − 1).
k−1,m−kik
c1 (J1m X)
= 2m(g − 1) and we have seen that ( 1 (3m2 + 6m)(g − 1), m even, m c1 (J2 X) = 14 2 4 (3m + 4m + 1)(g − 1), m odd.
Theorem 2.17. For a curve of genus g ≥ 2 we have, for each k ≥ 2, c1 (Jkm X) =
k
2(g − 1) X 1 mk + O(mk−1 ). (k!)2 i=1 i
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
133
Proof. It is clear that asymptotically c1 (Jkm X) = O(mk ). Hhence, writing c1 (Jkm X) = 2bk mk (g − 1) + O(mk−1 ) , we get via Theorem 2.15 that, for k ≥ 3, 2bk mk (g − 1) + O(mk−1 ) m/k
=2
X
X
(bk−1 (m − kik )k−1 + ik )(g − 1) + O(mk−1 )
ik =0 I 0 ∈Ik−1,m−kik m/k
= 2(g − 1)
X
ik =0
ak−1 ik (m − kik )k−2 + bk−1 (m − kik )k−1 + O(mk−1 ).
By Lemma 2.16 we have m/k
X
ik (m − kik )k−2 =
ik =0
k−2 1 1 X (−1)j (k − 2)! mk = 3 2 k j=0 j + 2 j!(k − 2 − j)! k (k − 1)
and m/k
X
(m − kik )k−1 =
ik =0
k−1 1 X (−1)j (k − 1)! 1 (k − 1)! 1 mk = = 2. k j=0 j + 1 j!(k − 1 − j)! k k! k
From these we obtain 2bk mk (g − 1) + O(mk−1 ) = 2(g − 1)
a k−1
k 3 (k − 1)
+
bk−1 + O(mk−1 ) k2
and hence the recursive relation: bk =
ak−1 3 k (k − 1)
+
bk−1 1 bk−1 1 1 = 2 + 2 = 2 (bk−1 + ) 2 k k k!(k − 1)! k k k!(k − 1)!
with a1 = 1 and b1 = 1. An explicit formula is obtained by repeatedly using the recursion. More precisely, we first apply the recursive formula to b k−1 : bk−1 =
1 1 bk−2 + 2 (k − 1) (k − 1)!(k − 2)!
and substitution yields 1 bk = 2 k
1 1 1 bk−2 + + . 2 (k − 1) (k − 1)!(k − 2)! k!(k − 1)!
The procedure above may be repeated until we reach b 1 = 1. Induction shows that k 1 X1 . ˜ bk = (k!)2 i=1 i
134
KAREN CHANDLER AND PITMANN WONG
For general m > k ≥ 2 we get from the filtrations the following [m/k] exact sequences: [m/k]−1
0 → Sm,k
[m/k]
→ Jkm X →
[m/k]−2
KX
[m/k]−1
m−k[m/k]
⊗Jk−1
[m/k]−1
X
→0
m−k([m/k]−1)
0 → Sm,k → Sm,k → KX ⊗Jk−1 X→0 ............................................................ m 0 → Jk−1 X →
1 Sm,k
→ 0.
m−k KX ⊗Jk−1 X
→
(2.5)
Observe that m − k[m/k] = 0 or 1 depending on whether m is divisible by k. m−ki i i By induction, we see that h1 (KX ⊗ Jk−1 X) = 0 implies that h1 (Sm,k ) = 0, for 1 m 0 ≤ i ≤ [m/k]. Hence h (Jk X) = 0 for any k ≥ 2, and as a result we also have [m/k] m h0 (Jkm X) = h0 (Jk−1 X) +
X
m−ki i h0 (KX X) if m > k. ⊗ Jk−1
(2.6)
i=1
Corollary 2.18. Let X be a curve of genus g ≥ 2. Then, h1 (Jkm X) = 0 if m ≥ k, and for all k ≥ 2 we have
h0 (Jkm X) =
k
2(g − 1) X 1 mk + O(mk−1 ). (k!)2 i=1 i
Proof. By Riemann–Roch for curves, we have h0 (Jkm X) − h1 (Jkm X) = c1 (Jkm X) − (rk Jkm X)(g − 1). As observed, h1 (Jkm X) vanishes. By Theorem 2.15, (rk Jkm X)(g−1) is of lower order, so h0 (Jkm X) = c1 (Jkm X) and the result follows from Theorem 2.17. ˜ By induction we get from (2.6) h0 (Jkm X) [m/3]
= h0 (J2m X) +
X
[m/k] i h0 (KX ⊗ J2m−3i X) + · · · +
i=1
[m/2]
=
X
X i=1
[m/3]
h
0
m−i (KX )+
i=0
m−ki i h0 (KX ⊗ Jk−1 X)
X
[m/k]
h
0
i (KX
⊗ J2m−3i X) + · · · +
i=1
X
m−ki i h0 (KX ⊗ Jk−1 X).
i=1
i Tensoring (2.5) with KX yields exact sequences [m/k]−1
0 → Sm,k
i i ⊗KX → Jkm X⊗KX →
[m/k]−2 i ⊗KX Sm,k
[m/k]−1 i ⊗KX Sm,k
[m/k]+i
KX
m−k[m/k]
⊗Jk−1
X
→ 0,
m−k([m/k]−1) [m/k]+i−1 X ⊗Jk−1 KX
→ 0, 0→ → → ........................................................................ m i 0 → Jk−1 X⊗KX →
1 Sm,k
→
i+1 m−k X KX ⊗Jk−1
→ 0.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
135
These imply that [m/k]
X
i h0 (Jkm X ⊗ KX )=
m−kj i+j X). ⊗ Jk−1 h0 (KX
j=0
Thus, for k = 3 we get by Theorem 2.13: [m/3]
[m/2]
h
0
(J3m X)
=
X
h
0
m−i (KX )+
i h0 (KX ⊗ J2m−3i X)
i=1
i=0
[m/3] [(m−3i)/2]
[m/2]
=
X
X
m−i h0 (KX )+
X i=1
i=0
X
m−2i−j h0 (KX ).
j=0
With this it is possible to write down the explicit formulas. In the case of J 2m X there are only two cases depending on the parity of m. For 3jets there are the following cases: (1a) m = 3q, q even; (1b) m = 3q, q odd; (2a) m = 3q + 1, q even; (2b) m = 3q + 1, q odd; (3a) m = 3q + 1, q even; and (3b) m = 3q + 2, q odd. For simplicity we shall only do this for case (1a). First we observe that the rank of J3m X is given by the number rk J3m X =
hmi 2
[m/3]h
X
+1+
i=1
m − 3i + 1 = O(m2 ). 2
i
If m is divisible by 3! then m/6
rk J3m X = =
m/6
m X m − 3(2l − 1) X m − 3(2l) m +1+ + + 2 3 2 2
l=1 1 12 (m + 3)(m + 4).
l=1
(2.7)
For k = 3 we get, by Theorem 2.13, h0 (J3m X) [m/2]
=
X
[m/3] m−i h0 (KX )+
i=0
X
i h0 (KX ⊗ J2m−3i X)
i=1
[m/3] [(m−3i)/2]
[m/2]
=
X
m−i h0 (KX )+
i=0
X
X
m−2i−j h0 (KX )
j=0
i=1
= (g − 1) ×
2m−1−
h m ih m i 2
2
+1 +
[m/3]
X i=1
2m−4i−1−
h m−3i ih m−3i i 2
2
+1
.
136
KAREN CHANDLER AND PITMANN WONG
If m is divisible by 3!, both m and [m/3] are even. Then, denoting the second sum above by S, S=
m/6
X
2m − 4(2l − 1) − 1 −
l=1
m − 1 − 3(2l − 1) 2 +
m/6
X
m − 1 − 3(2l − 1)
2m − 8l − 1 −
l=1
2
m − 6l 2
+1
m − 6l 2
+1
m/6 1 1X (3m2 + 10m + 6 + 60l2 − (28m + 36)l) = 2 3 m(11m2 − 18m − 18). = 2 2 3 l=1
Thus for m divisible by 3! we have 1 1 0 m 2 h (J3 X) = m(11m − 18m − 18) + 2 (m + 2)(3m − 2) (g − 1), 22 33 2 and, by Riemann–Roch and (2.7), c1 (J3m X) = (g − 1) × 1 1 1 2 m(11m − 18m − 18) + (m + 2)(3m − 2) + (m + 3)(m + 4) . 22 33 22 22 3 Example 2.19. The filtration of J36 X is given by J36 X = S 2 ⊃ S 1 ⊃ S 0 = 2 J26 X, and the associated exact sequences are 0 → S 1 → J36 X → KX → 0 and 0 → J26 X → S 1 → KX ⊗ J23 X → 0. 2 )+h0 (KX ⊗J23 X). From Hence h0 (J36 X) = 0 and h0 (J36 X) = h0 (J26 X)+h0 (KX 3 3 2 the exact sequence 0 → KX → J2 X → KX → 0 we obtain the exact sequence 4 3 0 → KX → KX ⊗ J23 X → KX →0
from which we conclude that 2 3 4 h0 (J36 X) = h0 (J26 X) + h0 (KX ) + h0 (KX ) + h0 (KX ) 2 3 4 5 6 = h0 (KX ) + 2(h0 (KX ) + h0 (KX )) + h0 (KX ) + h0 (KX )
= (3 + 2(5 + 7) + 9 + 11)(g − 1) = 47(g − 1). Next we consider the problem of constructing an explicit basis for H 0 (Jkm X). First we recall the construction of a basis for H 0 (J11 X) = H 0 (KX ). The procedure of this construction works in any algebraically closed field and has been used toward resolving the uniqueness problem for rational and meromorphic functions. (The reader is referred to [An et al. 2003a; 2004; 2003b] for details.) Let z 0 , z1 , z2 be the homogeneous coordinates on P 2 . Then z1 z2 z z dz − z dz dz1 dz2 i j i i j d = (2.8) = zj zj2 zj2
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
137
is a welldefined rational 1form on P n . Let P (z0 , z1 , z2 ) be a homogeneous polynomial of degree d and X = {[z0 , z1 , z2 ] ∈ P 2 (C)  P (z0 , z1 , z2 ) = 0}. Then, by Euler’s Theorem, for [z0 , z1 , z2 ] ∈ X, we have z0
∂P ∂P ∂P (z0 , z1 , z2 ) + z1 (z0 , z1 , z2 ) + z2 (z0 , z1 , z2 ) = 0. ∂z0 ∂z1 ∂z2
The tangent space of X is defined by the equation P (z 0 , z1 , z2 ) = 0 and dz0
∂P ∂P ∂P (z0 , z1 , z2 ) + dz1 (z0 , z1 , z2 ) + dz2 (z0 , z1 , z2 ) = 0. ∂z0 ∂z1 ∂z2
These may be expressed as ∂P (z0 , z1 , z2 ) + z1 ∂z0 ∂P dz0 (z0 , z1 , z2 ) + dz1 ∂z0 z0
∂P ∂P (z0 , z1 , z2 ) = −z2 (z0 , z1 , z2 ), ∂z1 ∂z2 ∂P ∂P (z0 , z1 , z2 ) = −dz2 (z0 , z1 , z2 ). ∂z1 ∂z2
Then by Cramer’s rule, we have on X W (z1 , z2 ) ∂P ∂P W (z2 , z0 ) ∂P ∂P = , = ∂z0 W (z0 , z1 ) ∂z2 ∂z1 W (z0 , z1 ) ∂z2 provided that the Wronskian W (z0 , z1 ) = z0 dz1 − z1 dz0 6≡ 0 on any component of X; that is, the defining homogeneous polynomial of X has no linear factor of the form az0 + bz1 . Thus W (z2 , z0 ) W (z0 , z1 ) W (z1 , z2 ) = = ∂P (z , z , z ) ∂P (z , z , z ) ∂P (z , z , z ) 0 1 2 0 1 2 0 1 2 ∂z0 ∂z1 ∂z2
(2.9)
is a globally welldefined rational 1form on any component of π −1 (X) ⊂ C 3 \{0}, where (π : C 3 \ {0} → P 2 (C) is the Hopf fibration), provided that the expressions make sense (that is, the denominators are not identically zero when restricted to a component of ψ −1 (X)). For our purpose, we also require that the form given by (2.9) is not identically trivial when restricted to a component of π −1 (X). This is equivalent to the condition that the Wronskians in the formula above are not identically zero; in other words, the defining homogeneous polynomial of X has no linear factor of the form azi + bzj where a, b ∈ C, 0 ≤ i, j ≤ 2 and i 6= j. If P , ∂P/∂z0 , ∂P/∂z1 , ∂P/∂z2 never vanish all at once (that is, X is smooth) then, at each point, one of the expressions in (2.9) is regular at the point. Hence so are the other expressions. This means that z1 z2 dz1 dz2 η= (2.10) ∂P/∂z0
138
KAREN CHANDLER AND PITMANN WONG
is regular on π −1 (X). (Note that the form η is not welldefined on X unless n = 3; see (2.8)). The form z1 z1 z2 z2 dz1 dz2 z n−1 dz1 dz2 0 z0n−3 = z0n−3 η, ω= ∂P/∂z0 = 2 z0 ∂P/∂z0 with n = deg P , is a welldefined (again by (2.8)) rational 1form on X. Moreover, as η is regular on X, the 1form ω is also regular if n ≥ 3. If n = 3 then ω = η and if n ≥ 4 then ω is regular and vanishes along the ample divisor [z0n−3 =0]∩X. Thus for any homogeneous polynomial Q = Q(z 0 , z1 , z2 ) of degree n − 3, the 1form Q n−3 ω = Qη z0 is regular on C and vanishes along [Q=0]. Note that the dimension of the vector space of homogeneous polynomials of degree n−3 (a basis is given by all possible monomials) is 1 2 (n − 1)(n − 2)
= genus of X.
We summarize these observations: Proposition 2.20. Let X = [z0 , z1 , z2 ] ∈ P 2 (C)  P (z0 , z1 , z2 ) = 0 be a nonsingular curve of degree d ≥ 3. If d = 3 then the space of regular 1forms on X is {cη  c ∈ C}, where η is defined by (2.2). If d ≥ 4 take the set Qi  Qi is a monomial of degree d − 3 for 1 ≤ i ≤ 12 (d − 1)(d − 2) as an ordered basis of homogeneous polynomials of degree d − 3. Then ωi = Qi η  1 ≤ i ≤ 21 (d − 1)(d − 2)
is a basis of the space of regular 1forms on X.
Using the preceding we may write down explicitly a basis for H 0 (Jkm X). We 2 demonstrate via examples. For d = 4, h0 (J22 X) = h0 (KX ) + h0 (KX ) = 6 + 3 = 9 and, since the genus is 3, there are 3 linearly independent 1forms ω 1 , ω2 , ω3 which, as shown above, may be taken as ω1 =
z0 (z0 dz1 − z1 dz0 ) , ∂P/∂z2
ω2 =
z1 (z0 dz1 − z1 dz0 ) , ∂P/∂z2
ω3 =
z2 (z0 dz1 − z1 dz0 ) . ∂P/∂z3
A basis for H 0 (J22 X) is given by ω1⊗2 , ω2⊗2 , ω3⊗2 , ω1 ⊗ ω2 , ω1 ⊗ ω3 , ω2 ⊗ ω3 , δω1 , δω2 , δω3 , where δ is the derivation defined in (1.9). The first six of these provide a basis 2 of H 0 (KX ) and the last three may be identified with a basis of H 0 (KX ). For
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
139
J23 X we have 2 3 h0 (J23 X) = h0 (KX ) + h0 (KX ) = h0 OX (2(d − 3)) + h0 OX (3(d − 3))
= C22d−4 − C2d−4 + C23d−7 − C22d−7 .
2 3 In particular, for d = 4, h0 (J23 X) = h0 (KX ) + h0 (KX ) = 6 + 10 = 16. A basis 0 3 2 for H (J2 X) is given by the six elements (identified with a basis of H 0 (KX ))
δω1⊗2 , δω2⊗2 , δω3⊗2 , δ(ω1 ⊗ ω2 ), δ(ω1 ⊗ ω3 ), δ(ω2 ⊗ ω3 ) 3 and the 10 elements (a basis of H 0 (KX )):
ω1⊗3 , ω2⊗3 , ω3⊗3 , ω1 ⊗ ω2 ⊗ ω3 , ω1⊗2 ⊗ ω2 , ω1⊗2 ⊗ ω3 , ω2⊗2 ⊗ ω1 , ω2⊗2 ⊗ ω3 , ω3⊗2 ⊗ ω1 , ω3⊗2 ⊗ ω2 .
3. Computation of Chern Classes The Case of Surfaces Summary. We exhibit here the explicit formulas due to [Stoll and Wong 2002] (see also [Green and Griffiths 1980]) for the Chern numbers of the projectivized parametrized jet bundles of a compact complex surface. The most important is the index formula given in Theorem 3.9: ι(Jkm X) = (αk c21 − βk c2 )m2k+1 + O(m2k ) where ci = ci (X), αk = βk + γk and k
βk =
X 1 2 , 2 (k!) (2k + 1)! i=1 i2
k
γk =
i−1
X1X1 2 . 2 (k!) (2k + 1)! i=1 i j=1 j
This implies that αk /βk → ∞ hence αk /βk > c2 /c21 for k sufficiently large provided that c21 > 0. For example, c2 /c21 = 11 for a smooth hypersurface of degree d = 5 and the explicit formula shows that αk /βk > 11 for all k ≥ 199. (See the table at the end of this section). The explicit formulas for αk and βk are crucial in the proof of the Kobayashi conjecture in Section 7. We now treat the case of a complex surface (complex dimension 2). The computations here are more complicated than those of Section 2 as we must deal with the second Chern number. The computation of the first Chern class is relatively easy since the Whitney formula is linear in this case; that is, if 0 → S 0 → S → S 00 → 0 is exact, then c1 (S) = c1 (S 0 ) + c1 (S 00 ). The Whitney formula for the second Chern classes on the other hand is nonlinear: c 2 (S) = c2 (S 0 ) + c2 (S 00 ) + c1 (S 0 )c1 (S 00 ). The (minor) nonlinearity may seem harmless at first but for filtrations the nonlinearity carries over at each step and the complexity increases rapidly. Thus the correct way to deal with the problem is not to calculate the second Chern class directly but to calculate the
140
KAREN CHANDLER AND PITMANN WONG
index ι(Jkm X) = c21 (Jkm X) − 2c2 (Jkm X) which does behave linearly, that is, ι(S) = ι(S 0 ) + ι(S 00 ). We then recover the second Chern class from the formula c2 (Jkm X) = (c21 (Jkm X) − ι(Jkm X))/2. In order to compute the jet differentials we must first calculate the Chern classes and indices of the sheaves J i1 ∗ J ik ∗ SI = T X ⊗···⊗ T X where I = (i1 , . . . , ik ). For details of the computations see [Stoll and Wong 2002]. By results from [Tsuji 1987; 1988; Maruyama 1981], the exterior, symmetric and tensor products of the cotangent sheaf of a manifold of general type are semistable in the sense of Mumford–Takemoto. For a coherent sheaf S on a variety of dimension n the index of stability relative to the canonical class is defined to be cn−1 (S) c1 (T ∗ X) µ(S) = 1 . (rk S)cn1 (T ∗ X) A sheaf S is said to be semistable in the sense of Mumford–Takemoto (relative to the canonical class) if µ(S 0 ) ≤ µ(S) for all coherent subsheaves S 0 of S. For a nonsemistable sheaf a subsheaf S 0 satisfying µ(S 0 ) > µ(S) is said to be a destabilizing subsheaf . In view of Tsuji’s result it would seem reasonable to expect that the sheaves of jet differentials are also semistable. However using the explicit formulas for the Chern classes computed below we shall see that this is not the case. Tsuji’s result is used in [Lu and Yau 1990] (see also [Lu 1991]) to show that a projective surface X satisfying the conditions that K X is nef and c21 (T ∗ X)−2c2 (T ∗ X) > 0 contains no rational nor elliptic curves. The instability of the jet differentials implies that the analogous result of Lu–Yau requires a different argument. We list below some basic but very useful formulas (see [Wong 1999; Stoll and Wong 2002]): Lemma 3.1. Let X be a nonsingular complex surface and E be a vector bundle Jm of rank 2 over X. Then rk ( E) = m + 1 and Jm 1 c1 ( E) = 2 m(m + 1)c1 (E), Jm 1 c2 ( E) = 24 m(m2 − 1)(3m + 2)c21 (E) + 61 m(m + 1)(m + 2)c2 (E). Consequently the index is given by the formula: Jm E) = 16 m(m + 1)(2m + 1)c21 (E) − 31 m(m + 1)(m + 2)c2 (E). ι( Moreover , if c21 (E) 6= 0 then
def
Jm 1 E) c2 ( J = . m 2 m→∞ c ( 2 E) 1
δ∞ (E) = lim
Note that δ∞ (E) is independent of c2 (E)/c21 (E). The next formula gives the Chern numbers for tensor products of different bundles. Lemma 3.2. Let Ei , i = 1, . . . , k, be holomorphic vector bundles, of respective Qk rank ri , over a nonsingular complex surface X. Let R = l=1 rl . Then:
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
(i) c1 (ii) c2
k O
i=1 k O
Ei Ei
i=1
=
k X
(r1 . . . ri−1 ri+1 . . . rk )c1 (Ei ) = Rri
i=1
=R
k X c1 (Ei ) i=1
k X c2 (Ei ) i=1
ri
+ (R − 1)
In particular :
X
1≤i<j≤k
ri
141
.
c1 (Ei )c1 (Ej ) ri rj k Qk X l=1,l6=i rj c21 (Ei ). + 2 i=1
r2 2 r1 2 (iii) c2 (E1 ⊗ E2 ) = c1 (E1 ) + (r1 r2 − 1)c1 (E1 )c1 (E2 ) + c (E2 ) 2 2 1 +r2 c2 (E1 ) + r1 c2 (E2 ); 3 3 X c1 (Ei )c1 (Ej ) O X c2 (Ei ) + (r1 r2 r3 − 1) (iv) c2 E i = r 1 r2 r3 ri ri rj i=1 i=1 1≤i<j≤3 3 X r1 r2 r3 /ri 2 + c1 (Ei ); 2 i=1 and the index ι(E1 ⊗ E2 ) = c21 (E1 ⊗ E2 ) − 2c2 (E1 ⊗ E2 ) is given by
(v) ι(E1 ⊗ E2 ) = r2 c21 (E1 ) + r1 c21 (E2 ) + 2c1 (E1 )c1 (E2 ) − 2r2 c2 (E1 ) − 2r1 c2 (E2 ). With the preceding formulas the computation of the Chern numbers for J km X can now be carried out by using the filtration given in Theorem 2.3, reducing the calculation to the Chern numbers of bundles of the form J i1 ∗ J ik ∗ SI = T X ⊗···⊗ T X,
where the indices I = (i1 , . . . , ik ) satisfy the condition i1 + 2i2 + · · · + kik = m. More precisely, take Ikm = I = (i1 , . . . , ik )  ij ∈ N ∪ {0}, i1 + 2i2 + · · · + kik = m together with a fixed ordering of Ikm (say, the lexicographical ordering). Then a brute force computation, applying Lemma 3.2 and Lemma 3.3 repeatedly yields the following formulas:
Theorem 3.3. Let X be a nonsingular complex surface and let Sm−2i,i = Jm−2i ∗ Ji ∗ T X⊗ T X. Denote by c1 = c1 (T ∗ X), c2 = c2 (T ∗ X). Then rk (Sm−2i,i ) = (m−2i+1)(i+1) = (m+1) + (m−1)i − 2i2 , c1 (Sm−2i,i ) = 12 (m−i)(m−2i+1)(i+1)c1 = 21 m(m+1) + (m2 −2m−1)i − (3m−1)i2 +2i3 c1 , c2 (Sm−2i,i ) =
2 4 3 2 1 24 {m(m −1)(3m+2) + 2(3m −5m −3m +4m+1)i 4 3 2 2 3 2
+(3m −30m +12m +6m−7)i − 2(9m −27m +5m+1)i3
+ (39m2 −42m+7)i4 − 4(3m−1)i5 +4i6 }c21 + 16 bm−2i,i c2 ,
142
KAREN CHANDLER AND PITMANN WONG
where ci = ci (T ∗ X). The index is given by ι(Sm−2i,i ) = 61 (am−2i,i c21 − 2bm−2i,i c2 ), where am−2i,i and bm−2i,i are polynomials given by am−2i,i = m(m+1)(2m+1) + (2m3 −6m2 −7m−1)i − (9m2 −6m−5)i2 + (14m−2)i3 −8i4 , bm−2i,i = m(m+1)(m+2) + (m3 −3m2 −8m−2)i − (6m2 −3m−7)i2 + (13m−1)i3 − 10i4 . The rank rk J2m X of the sheaf of J2m X is given by 1 1 (m + 1)(m + 3)(m + 5) = 3 (m3 + 9m2 + 23m + 15), 24 2 3 1 1 (m + 2)(m + 3)(m + 4) = 3 (m3 + 9m2 + 26m + 24), 24 2 3
if m is odd, if m is even;
and the first Chern class of same sheaf , c1 (J2m X), is 3m4 +28m3 +78m2 +68m+15 (m+1)(m+3)(m+5)(3m+1) c1 = c1 , 192 26 3
m odd,
m(m+2)(m+4)(3m+10) 3m4 +28m3 +84m2 +80m c1 = c1 , 192 26 3
m even.
The index of J2m X is given by ι(J2m X) = c21 (J2m X) − 2c2 (J2m X) = am c21 − bm c2 , where the coefficients am and bm are polynomials in m given by ( 1 (7m5 + 75m4 + 270m3 + 390m2 + 203m + 15) if m is odd , 6 am = 2 115 5 4 3 2 if m is even; 26 15 (7m + 75m + 280m + 420m + 208m) bm =
(
1 5 26 15 (5m 1 5 26 15 (5m
+ 75m4 + 390m3 + 810m2 + 565m + 75)
if m is odd ,
+ 75m4 + 400m3 + 900m2 + 720m)
if m is even.
The formula for the index also yields the formula for c 2 (J2m X): c2 (J2m X) =
1 2 m 1 {c (J X) − (am c21 − bm c2 )} = {λm c21 + bm c2 } 2 1 2 2
(3.1)
where the coefficients am and bm are given by Theorem 3.3, and the coefficient λm is given by ( 2 1 2 − am , m odd, 192 (m + 1)(m + 3)(3m + 16m + 5) λm = 2 1 − am , m even. 192 m(m + 2)(m + 4)(3m + 10) In particular:
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
143
Corollary 3.4. Let X be a nonsingular complex surface and assume that c21 (T ∗ X) > 0. Then 1 c2 (J2m X) = . m→∞ c2 (J m X) 2 1 2
δ(J2m X) = lim
For simplicity, set c1 = c1 (T ∗ X), c2 = c2 (T ∗ X). For any sheaf S, define ι(S) = c21 (S) − 2c2 (S),
µ(S) =
c1 (S) c1 , (rk S)c21
δ(S) =
c2 (S) , c21 (S)
(3.2)
provided that the denominators are not zero. Denote for simplicity δ = δ(T X) = c2 /c21 . It is well known that c21 ≤ 3c2 and c2 ≤ 5c21 + 36 for a surface of general type with c21 > 0 [Barth et al. 1984, p. 228]. Thus, for such surfaces, δ satisfies the estimate 36 1 (3.3) ≤ δ ≤ 5 + 2 ≤ 41. 3 c1 We give the precise numbers for a few special cases: • J22 X. In this case k = 2, m = 2 and there are two weighted partitions (i1 , i2 ) corresponding to the two solutions of i1 + 2i2 = 2 (Example 2.9), namely J2 ∗ T X, I1 = (2, 0) and I2 = (0, 1). The corresponding sheaves are SI1 = SI2 = T ∗ X. The various invariants of these sheaves are as follows: I (2, 0) (0, 1)
rank c1 (S)
S J2
∗
T X T X J22 X ∗
3 2 5
3c1 c1 4c1
c2 (S) 2c21
+ 4c2 c2 5c21 + 5c2
ι(S) 5c21 − 8c2 c21 − 2c2 6c21 − 10c2
µ(S)
δ(S)
1
1 9 (2 + 4δ)
1 2 4 5
5 16 (1 + δ)
δ
The Chern numbers are calculated using Lemma 3.1 and Lemma 3.2. Note that J2 ∗ J2 ∗ T X is a subsheaf of J22 X (by Example 2.5, 0 → T X → J22 X → J 2 ∗ ∗ 2 T X → 0 is an exact sequence) with µ( T X) > µ(J2 X). A subsheaf with such a property is said to be a destabilizing subsheaf. On the other hand T ∗ X is a quotient sheaf of J22 X with µ(T ∗ X) < µ(J22 X). A quotient sheaf with such a property is said to be a destabilizing quotient sheaf. • J23 X. In this case k = 2, m = 3 and there are two weighted partitions I1 = (3, 0) and I2 = (1, 1) corresponding to the two solutions of i1 + 2i2 = 3. S rk c1 (S) c2 (S) ι(S) µ(S) δ(S) J3 ∗ 1 3 (3, 0) T X 4 6c1 11c21 + 10c2 14c21 − 20c2 2 36 (11 + 10δ) N2 ∗ 1 2 2 (1, 1) T X 4 4c1 5c1 + 4c2 6c1 − 8c2 1 16 (5 + δ) 1 5 J23 X 8 10c1 40c21 + 14c2 20c21 − 28c2 4 50 (20 + 7δ) J3 ∗ N2 ∗ The sheaves T X and T X are respectively a destabilizing subsheaf and J3 ∗ a destabilizing quotient sheaf of J23 X. The sequence 0 → T X → J23 X → N2 ∗ T X → 0 is exact, by Example 2.5. I
144
KAREN CHANDLER AND PITMANN WONG
• J24 X. In this case k = 2, m = 4 and there are 3 weighted partitions I1 = (4, 0), I2 = (2, 1) and I3 = (0, 2) corresponding to the 3 solutions of i1 + 2i2 = 4. I (4, 0) (2, 1) (0, 2)
rk c1 (S)
S J2
J4
∗
T X
∗
∗
T X ⊗T X J2 ∗ T X J24 X
J4
c2 (S)
ι(S) 30c21 − 40c2 19c21 − 22c2 5c21 − 8c2
1
54c21 − 70c2
11 7
5
10c1
6
9c1
35c21 + 20c2 31c21 + 11c2
3
3c1
2c21 + 4c2
14 J2
22c1 215c21 + 35c2
µ(S)
δ(S)
2
7 1 20 + 5 δ 31 11 81 + 81 δ 4 2 9 + 9δ 215 35 484 + 484 δ
3 2
T ∗ X are respectively a destabilizing subsheaf and T ∗ X and The sheaves J2 ∗ T X ⊗ T ∗ X is neither a a destabilizing quotient sheaf of J24 X. Note that 4 subsheaf nor a quotient sheaf of J2 X. We have two exact sequences: J2 ∗ 0 → F21 → J24 X → T X → 0, J2 ∗ J4 ∗ 1 T X ⊗ T ∗ X → 0. T X → F2 → 0→ • J25 X. In this case k = 2, m = 5 and there are 3 weighted partitions I1 = (5, 0), I2 = (3, 1) and I3 = (1, 2) corresponding to the 3 solutions of i1 + 2i2 5. I (5, 0) (3, 1) (1, 2)
rk c1 (S)
S J3
J5
∗
T X
T ∗X ⊗ T ∗X J2 ∗ T X T ∗X ⊗
6 8 6
c2 (S)
ι(S)
µ(S)
15c1 85c21 + 35c2 16c1 106c21 + 24c2
55c21 − 70c2 44c21 − 48c2
5 2
2
31c21 + 11c2
19c21 − 22c2
3 2
9c1
δ(S) 17 45 53 128 31 81 741 1600
7 + 45 δ 3 + 32 δ 11 + 81 δ
7 J25 X 20 40c1 741c21 + 70c2 118c21 − 140c2 2 + 160 δ J2 ∗ J5 ∗ T X ⊗ T ∗ X are respectively a destabilizing subT X and The sheaves J3 ∗ sheaf and a destabilizing quotient sheaf of J25 X. Note that T X ⊗ T ∗ X is neither a subsheaf nor a quotient sheaf of J25 X. We have two exact sequences: J2 ∗ T X → 0, 0 → F21 → J25 X → T ∗ X ⊗ J3 ∗ J5 ∗ 1 T X ⊗ T ∗ X → 0. T X → F2 → 0→
• J26 X. In this case k = 2, m = 6 and there are 4 weighted partitions I1 = (6, 0), I2 = (4, 1), I3 = (2, 1) and I4 = (0, 3) corresponding to the 3 solutions of i1 + 2i2 = 6. S
I (6, 0) (4, 1) (2, 2) (0, 3)
J6
rk c1 (S)
T ∗X 7 J4 ∗ ∗ T X ⊗T X 10 J2 ∗ J T X ⊗ 2 T ∗X 9 J3 ∗ T X 4 6 J2 X 30
c2 (S)
ι(S)
µ(S)
21c1 175c21 +56c2 91c21 −112c2 3 2 25c1 270c1 +45c2 85c21 −90c2 52 2 18c1 138c1 +24c2 48c21 −48c2 2 2 6c1 11c1 +10c2 14c21 −20c2 32 70c1 2331c21 +135c2 238c21 −270c2 73
δ(S) 25 8 + 63 δ 63 54 9 + 125 δ 125 23 2 + δ 54 27 11 5 + δ 36 18 333 27 + 980 δ 700
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
145
We have three exact sequences: J3 ∗ 0 → F22 → J26 X → T X → 0, J2 ∗ J2 ∗ 1 2 0 → F 2 → F2 → T X → 0, T X⊗ J2 ∗ J4 ∗ J6 ∗ 1 T X → 0. T X⊗ T X → F2 → 0→
J3 ∗ J6 ∗ T X are respectively a destabilizing subsheaf and T X and The sheaves a destabilizing quotient sheaf of J26 X. Remark 3.5. For each partition I = (i1 , i2 ) satisfying i1 +2i2 = m we associate the (nonweighted) sum I = i1 +i2 . Let Imax = maxI {I} and Imin = maxI {I}. Then the sheaf SImax is a destabilizing subsheaf and the sheaf SImin is a destabilizing quotient sheaf. We now deal with the case of general k. We shall be content with asymptotic formulas as the general formulas become complicated since the general formula for sums of powers can only be given recursively. However the highest order term is quite simple; indeed, we have m X
id =
i=1
md+1 + O(md ). d+1
(3.4)
Before dealing with the jet bundles Jkm X we must first find the formulas for the J i1 ∗ J ik ∗ sheaves SI = T X ⊗···⊗ T X. This is easier due to the symmetry of the sheaves and we know, a priori, that the formulas can be expressed in terms of the symmetric functions in the exponents i1 , . . . , ik . For general k we introduce some notation for the jth symmetric functions on k indices: s0;k = 1, s1;k =
k X
ip , s2;k =
k X
ip iq , . . . , sk;k =
1≤p 0 then ι(Jkm X) = cm2k+1 c21 + O(m2k ) for some positive constant c. The asymptotic expansion for c2 (Jkm X) now follows readily from Corollary 3.10 along with Theorems 3.8 and 3.9: Theorem 3.11. Let X be a nonsingular complex surface. For any positive integer k, c2 (Jkm X) = 21 (c21 (Jkm X) − ι(Jkm X)) = 21 c21 (Jkm X) = 21 A2k c21 m4k + O(m4k−1 ).
148
KAREN CHANDLER AND PITMANN WONG
We tabulate the ratios αk /βk on the next page (they can be checked readily using Mathematica or Maple): k 2 3 4 5 6 7 8 9 10 11 12 197 198 199 200
αk
βk
7 26 15 17 27 36 7 83 216 38 7 1717 217 38 56 11 1927 221 311 56 11 13 726301 222 312 57 76 11 13 3144919 234 312 57 76 11 13 17 2754581 235 320 57 76 13 17 19 2923673 239 321 510 77 13 17 19 315566191 240 321 510 77 116 17 19 23 330851461 247 324 512 77 116 17 19 23
5 26 15 7 27 36 5 41 216 38 7 479 217 38 56 7 59 221 311 56 11 266681 222 312 57 76 11 13 63397 234 312 57 76 11 13 514639 235 320 57 76 11 13 17 178939 239 321 59 77 13 17 19 10410343 240 321 59 77 116 13 17 19 18500393 247 324 511 77 116 17 19 23
αk /βk 1.40000 1.73469 2.02439 2.28108 2.51239 2.72348 2.91804 3.09879 3.26779 3.42666 3.57670 10.9808 10.9987 11.0165 11.0345
4. Finsler Geometry of Projectivized Vector Bundles Summary. Our use of projectivized jet bundles is initiated by the recognition that, for projectivized vector bundles, the algebraic geometric concept of ampleness is equivalent to the existence of a Finsler (not hermitian in general ) metric with negative mixed holomorphic bisectional curvature. It is known, at least in the case of the tangent bundle that, even for Finsler metrics, negative holomorphic bisectional curvature implies hyperbolicity. We provide in this section some of the basic notions from Finsler geometry. For more details see [Cao and Wong 2003; Chandler and Wong 2004] and the references there. Many questions concerning a complex vector bundle E of rank greater than 1 may be reduced to problems about the tautological line bundle (or its dual) over the projectivization P(E). For example the algebraic geometric concept of ampleness (and the numerical effectiveness) of a holomorphic vector bundle
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
149
E may be interpreted in terms of Finsler geometry (see [Cao and Wong 2003], and also [Aikou 1995; 1998]; for general theory on Finsler geometry we refer to [Bao and Chern 1991; Bao et al. 1996; Abate and Patrizio 1994]). For the relationship with the Monge–Amp`ere equation see [Wong 1982]. We also provide some implications of this reformulation. For applications of the formulation using projectivized bundles to complex analysis see [Dethloff et al 1995a, b]. The dual of a vector bundle E will be denoted by E ∗ . For any positive integer k, denote Jk E the kfold symmetric product. The dual vector bundle E ∗ is said to by be ample if and only if the line bundle LP(E) is ample. By a Finsler metric along the fibers of E we mean a function h : E → R ≥0 with the following properties: (FM1) h is of class C 0 on E and is of class C ∞ on E \ {zero section}. (FM2) h(z, λv) = λh(z, v) for all λ ∈ C. (FM3) h(z, v) > 0 on E \ {zero section}. (FM4) For z and v fixed, the function ηz,v (λ) = h2 (z, λv) is smooth even at λ = 0. (FM5) hEz is a strictly pseudoconvex function on Ez \ {0} for all z ∈ M . Denote by π : T E → E the projection and V = ker π ⊂ T E the vertical subbundle. A Finsler metric F defines naturally a hermitian inner product on the vertical bundle V ⊂ T E by hV, W iV =
r X
j
gi¯j (z, v)V i W ,
∂ 2 F 2 (z, v) ∂v i ∂¯ vj
gi¯j (z, v) =
(4.1)
i,j=1
P P for horizontal vector fields V = i V i ∂/∂v i , W = i W i ∂/∂v i ∈ V on E where v1 , . . . , vr are the fiber coordinates. (The difference between a Finsler metric and a hermitian metric is that, for a hermitian metric, the components (g i¯j ) of the hermitian inner product on the vertical bundle are independent of the fiber coordinates). The hermitian inner product defines uniquely a hermitian connection (known as the Chern connection) θ = (θik ) and the associate hermitian curvature Θ = (Θki ). If (gi¯j ) comes from a hermitian metric then the curvature forms depend only on the base coordinates; however if it comes from a general Finsler metric then the curvature forms will have horizontal, vertical and mixed components: Θki =
n X
α,β=1
k α Kiα zβ + β¯ dz ∧ d¯
r X
j,l=1
κkij ¯l dv j ∧ d¯ vl +
n X r X
µkiα¯l dz α ∧ d¯ vl
α=1 l=1
+
n r X X
k j νij zβ . β¯ dv ∧ d¯
j=1 β=1
Pr Denote by P = i=1 v i ∂/∂v i the position vector field on E. The mixed holomorphic bisectional curvature of the Finsler metric is defined, for any nonzero
150
KAREN CHANDLER AND PITMANN WONG
vector field X ∈ Γ(M, T M ), to be
Θ(X, X)P, P
V
=
r X
n X
β
(4.2)
k i j α gk¯j Kiα ¯ X X , β¯ v v
i,j,k=1 α,β=1
where the inner product is defined by (4.1). The following result can be found in [Cao and Wong 2003]. Theorem 4.1. Let E be a rank r ≥ 2 holomorphic vector bundle over a compact complex manifold X. The following statements are equivalent: (1) E ∗ is ample (resp. nef ). Jk ∗ E is ample (resp. nef ) for some positive integer k. (2) (3) The dual LP(Jk E) of the tautological line bundle over the projectivized bundle Jk E) is ample (resp. nef ) for some positive integer k. P( (4) There exists a Finsler metric along the fibers of E with negative (resp. nonpositive) mixed holomorphic bisectional curvature. (5) For some positive integer k there exists a Finsler metric along the fibers of Jk E with negative (resp. nonpositive) mixed holomorphic bisectional curvature. From the algebraic geometric point of view the key relationship between a vector bundle and its projectivization is the Fundamental Theorem of Grothendieck [Grothendieck 1958]: Theorem 4.2. Let p : E → X be a holomorphic vector bundle of rank r ≥ 2 over a complex manifold X of dimension n. Then for any analytic sheaf S on X and any m ≥ 1, Jm ∗ E , if i = 0, ∼ pi∗ Lm = P(E) 0, if i > 0, m where pi∗ Lm P(E) is the ith direct image of LP(E) . Consequently,
H i (X,
Jm
∗ E ∗ ⊗ S) ∼ = H i (P(E), Lm P(E) ⊗ p S)
for all i ≥ 0.
The theorem implies that the cohomology groups vanish beyond the dimension n Jm ∗ of X although the dimension of P(E) is n+r−1 > n; moreover, χ( E ⊗S) = ∗ χ(Lm ⊗ p S). For a vector bundle F over a smooth surface X, the Chern P(E) character and the Todd class are defined by ch(F ) = rk(F ) + c1 (F ) + 12 (c21 (F ) − 2c2 (F )),
(4.3)
1 td(F ) = 1 + 12 c1 (F ) + 12 (c21 (F ) + c2 (F )).
The Riemann–Roch formula is χ(F ) = ch(F ) · td(T X)[X] =
2 1 1 2 (ι(F ) − c1 (F ) c1 ) + 12 rk(F )(c1 + c2 )
[X], (4.4)
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
151
where ci = ci (T ∗ X) = −ci (T X). The notation ω[X] indicates the evaluation of a form of top degree on the fundamental cycle [X], that is, Z ω[X] = ω. X
Assume rank F = 2 over a nonsingular complex surface X. Then, by Lemma 3.1, Jm ι( F ) = 16 m(m + 1)(2m + 1)c21 (F ) − 31 m(m + 1)(m + 2)c2 (F ), Jm 1 F ) = m + 1 + 21 m(m + 1)c1 (F ) + 12 ch( m(m + 1) (2m + 1)c21 (F ) − 2(m + 2)c2 (F ) , Jm 1 F ) = 12 χ( m(m + 1) (2m + 1)c21 (F ) − 2(m + 2)c2 (F ) 1 (m + 1)(c21 + c2 ). − 41 m(m + 1)c1 (F ) c1 + 12
For example, taking F = T ∗ X, Jm ∗ 1 T X) = 12 χ( (m + 1) (2m2 − 2m + 1)c21 − (2m2 + 4m − 1)c2 ) ; in particular,
χ(T ∗ X) = 16 (c21 − 5c2 ), In any case we have:
χ(
J2
T ∗ X) = 41 (5c21 − 15c2 ).
Theorem 4.3. Let p : E → X be a holomorphic vector bundle of rank r = 2 over a complex surface X. Then dim P(E) = 3 and for any positive integer m, m3 3 m m3 2 ∗ (c1 (E )−c2 (E ∗ ))+O(m2 ) = c (L )+O(m2 ). 3! 3! 1 P(E) Jm 2 Suppose that h2 (Lm E)) = O(m2 ) and that c31 (Lm P(E) ) (= h ( P(E) ) > 0 (equiva2 lently, c1 (E)−c2 (E) > 0). The preceding theorem implies that E (or equivalently LP(E) ) is big, that is, 0 Jm h0 (Lm E) ≥ Cm3 P(E) ) = h (
χ(
Jm
E ∗ ) = χ(Lm P(E) ) =
for some constant C > 0. Recall the following fact (from [Cao and Wong 2003] or [Kobayashi and Ochiai 1970], for example):
Theorem 4.4. Let E be a holomorphic vector bundle of rank r ≥ 2 over a complex manifold X. Then the canonical bundles of X and P(E) are related by the formula KP(E) ∼ = [pE ]∗ (KX ⊗ det E ∗ ) ⊗ L−r P(E) where L−r P(E) is the dual of the rfold tensor product of LP(E) . In particular , we have 2 ∼ −n ∗ ⊗ L−n KP(T X) ∼ = [pT X ]∗ KX P(T X) and KP(T X) = LP(T ∗ X) where n = dim X.
152
KAREN CHANDLER AND PITMANN WONG
Corollary 4.5. Let X be a complex manifold of dimension n. −1 (i) T X is ample (resp. nef) if and only if KP(T ∗ X) is ample (resp. nef). n (ii) If KX is nef then KP(T X) ⊗ LP(T X) is nef .
(iii) If T ∗ X is ample then KP(T X) ⊗LnP(T X) is nef and KP(T X) ⊗Ln+1 P(T X) is ample. We have the following vanishing theorem [Cao and Wong 2003] (for variants see [Chandler and Wong 2004]): Corollary 4.6. Let E be a nef holomorphic vector bundle of rank r ≥ 2 over a compact complex manifold M of dimension n. Then Jm E ⊗ det E ⊗ KX = 0, H i X, Nk Jm Nk E) ⊗ KX = 0, E) ⊗ det( ( H i X, Jm T X) = 0 for all for all i, m, k ≥ 1. Consequently, if E = T X then H i (X, i, m ≥ 1. For a holomorphic line bundle L over a compact complex manifold Y with h0 (Lm ) > 0, m a positive integer, define a meromorphic map Φm = [σ0 , . . . , σN ] : Y → P N where σ0 , . . . , σN is a basis of H 0 (Lm ). The Kodaira–Iitaka dimension of L is defined to be −∞, if h0 (Lm ) = 0 for all m, κ(L) = max{dim Φm (X)  h0 (Lm ) > 0}, otherwise. The line bundle L is said to be big if k(L) = dim Y . This is equivalent to saying that, for m 0 h0 (Lm ) ≥ Cmdim Y for some positive constant C; in other words, the dimension of the space of sections h0 (Lm ) has maximum possible growth rate. See [Chandler and Wong 2004] for a discussion of the differential geometric meaning of big bundles. Riemann– Y Roch asserts that if cdim (L) > 0 the Euler characteristic is big: 1 χ(Lm ) =
Y cdim (L) dim Y 1 m + O(mdim Y −1 ). (dim Y )!
This, in general, is not enough to conclude that L is big. However, Corollary 4.6 implies that if T ∗ X is nef then the cohomology groups H i (X, T ∗ X) = 0 for all i ≥ 1. Hence T ∗ X is big if the Euler characteristic is big. In fact, for surfaces the weaker condition that KX is nef suffices: Corollary 4.7. Suppose that the canonical bundle KX of a nonsingular surface dim P(T X) is nef and that c1 (LP(T X) ) > 0. Then LP(T X) is big.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
153
A vector bundle E of rank > 1 is said to be big if the line bundle L P(E) is big. By dim P(T X)
Theorem 4.3, for a surface X the condition c1 (LP(T X) ) > 0 is equivalent 2 2 ∗ to the condition that c1 (T X) − c2 (T X) = c1 (T X) − c2 (T ∗ X) is positive. Thus we may restate Corollary 4.7 as follows: Corollary 4.8. Let X be a nonsingular compact surface such that c21 (T ∗ X) − c2 (T ∗ X) > 0 and KX is nef . Then T ∗ X is big. The preceding corollary implies the following theorem which may be viewed as an analogue, for surfaces, of the classical theorem that a curve of positive genus is hyperbolic [Lu and Yau 1990; Lu 1991; Dethloff et al. 1995a; 1995b]: Theorem 4.9. Let X be a nonsingular surface such that c21 (T ∗ X)−2c2 (T ∗ X) > 0 and KX is nef . Then X is hyperbolic. We refer the readers to [Dethloff et al. 1995a; 1995b] for further information and refinements of the preceding theorem. The condition c 21 (T ∗ X) − c2 (T ∗ X) > 0 is not satisfied by hypersurfaces in P 3 which is the main reason that jet differentials are introduced. The computations in the previous section will provide conditions (on the Chern numbers c21 (T ∗ X) and c2 (T ∗ X)) under which the sheaves of jet differentials Jkm X must be big.
5. Weighted Projective Spaces and Projectivized Jet Bundles Summary. The fibers of the kjet bundles P(J k X) are special types of weighted projective spaces. We collect some of the known facts of these spaces in this section. The main point is that these spaces are, in general , not smooth but with very mild singularities and we show that the usual theory of fiber integration for smooth manifolds extends to P(J k X). This will be used in later sections. We follow the approach of the previous section by reducing questions concerning kjet differentials to questions about the line bundle over the projectivization P(J k X). Since J k X is only a C ∗ bundle rather than a vector bundle the fibers of the projectivized bundle P(J k X) is not the usual projective space but a special type of weighted projective space. We give below a brief account concerning these spaces; see [Beltrametti and Robbiano 1986; Dolgachev 1982; Dimca 1992] for more detailed discussions and further references. The general theory of the projectivization of coherent sheaves can be found in [Banica and Stanasila 1976]. Consider C r+1 together with a vector Q = (q0 , . . . , qr ) of positive integers. The space C r+1 is then denoted (C r+1 , Q) and we say that each coordinate zi , 0 ≤ i ≤ r, has weight (or degree) qi . A C ∗ action is defined on (C r+1 , Q) by λ.(z0 , . . . , zr ) = (λq0 z0 , . . . , λqr zr )
for λ ∈ C ∗ .
The quotient space P(Q) = (C r+1, Q)/C ∗ is the weighted projective space of type Q. The equivalence class of an element (z0 , . . . , zr ) is denoted by [z0 , . . . , zr ]Q .
154
KAREN CHANDLER AND PITMANN WONG
For Q = (1, . . . , 1) = 1, P(Q) = P r is the usual complex projective space of dimension r and an element of P r is denoted simply by [z0 , . . . , zr ]. The case r = 1 is special as it can be shown that P(q0 , q1 ) ∼ = P 1 for any tuple (q0 , q1 ). This is not so if r ≥ 2. For a tuple Q define a map ψQ : (C r+1 , 1) → (C r+1 , Q) by ψQ (z0 , . . . , zr ) = (z0q0 , . . . , zrqr ). It is easily seen that ρQ is compatible with the respective C ∗ actions and hence descends to a welldefined morphism: [ψQ ] : P r → P(Q),
[ψQ ]([z0 , . . . , zr ]) = [z0q0 , . . . , zrqr ]Q .
(5.1)
The weighted projective space can also be described as follows. Denote by Θ qi Lr the group consisting of all qi th roots of unity. The group ΘQ = i=0 Θqi acts on P r by coordinatewise multiplication: (θ0 , . . . , θr ).[z0 , . . . , zr ] = [θ0 z0 , . . . , θr zr ],
θ i ∈ Θ qi ,
and the quotient space is denoted by P r /ΘQ . The next result is easily verified [Dimca 1992]: Theorem 5.1. The weighted projective space P(Q) is isomorphic to the quotient P r /ΘQ . In particular , P(Q) is irreducible and normal (the singularities are cyclic quotients and hence rational ). Given a tuple Q we assign the degree (or weight) q i to the variable zi (i = 1, . . . , q) and denote by SQ (m) the space of homogeneous polynomials of degree m. In other words, a polynomial P is in S(Q)(m) if and only if P (λ · (z 0 , . . . , zr )) = λm P (z0 , . . . , zr ). We may express such a polynomial explicitly as X P = ai0 ...ir z0i0 . . . zrir , (i0 ,...,ir )∈IQ,m
where the index set IQ,m is defined by Pr IQ,m = (i0 , . . . , ir ) j=0 qj ij = m, il ∈ N ∪ {0} .
The sheaf OP(Q) (m), m ∈ N, is by definition the sheaf over P(Q) whose global regular sections are precisely the elements of S Q (m), i.e., H 0 (P(Q), OP(Q) (m)) = SQ (m). For a negative integer −m the sheaf OP(Q) (−m) is defined to be the dual of OP(Q) (m) and OP(Q) (0) is the structure sheaf OP(Q) of P(Q). Here are some basic properties of these sheaves (see [Beltrametti and Robbiano 1986]): Theorem 5.2. Let Q = (q0 , . . . , qr ) be an r + 1tuple of positive integers.
(i) For any for any m ∈ Z, the line sheaf OP(Q) (m) is a reflexive coherent sheaf . (ii) OP(Q) (m) is locally free if m is divisible by each qi (hence by the least common multiple). (iii) Let mQ be the least common multiple of {q0 , . . . , qr }. Then OP(Q) (mQ ) is ample.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
155
(iv) There exists an integer n0 depending only on Q such that OP(Q) (nmQ ) is very ample for all n ≥ n0 . (v) OP(Q) (αmQ ) ⊗ OP(Q) (β) ∼ = OP(Q) (αmQ + β) for any α, β ∈ Z. For Q = 1 the assertions of the preceding theorem reduce to wellknown properties of the usual twisted structure sheaves of the projective space. For any subset J ⊂ {0, 1, . . . , r} denote by mJ the least common multiple of {qj , j ∈ J} and define q r+1 P X 1 X #J=i mJ , m(Q) = − qi + r−1 r i=2 i−2 i=0
where #J is the number of elements in the set J. It is known that we may take n0 = m(Q) + 1 in assertion (iv) above. In general the line sheaf O P(Q) (m) is not invertible if m is not an integral multiple of m Q . It can be shown that for Q = (1, 1, 2) the sheaf OP(Q) (1) is not invertible and hence, neither is OP(Q) (1) ⊗ OP(Q) (1). On the other hand, by part (ii) of the preceding theorem we know that OP(Q) (2) is invertible, thus OP(Q) (1) ⊗ OP(Q) (1) 6∼ = OP(Q) (2). The following theorem on the cohomologies of the sheaf O P (Q)(p) is similar to the case of standard projective space (see [Beltrametti and Robbiano 1986] or [Dolgachev 1982]): Theorem 5.3. If Q = (q0 , . . . , qr ) is an (r + 1)tuple of for p ∈ Z, {0}, i H (P(Q), OP (Q)(p)) = SQ (p), S(Q)(−p − Q),
positive integers then i 6= 0, r i = 0, i = r,
where Q = q0 + · · · + qr .
The cohomology group H i (P(Q), OP(Q) (p)) vanishes provided that i 6= 0, r. Let Q = (q0 , . . . , qr ) be a (r+1)tuple of positive integers and define, for k = 1, . . . , r, lQ,k = lcm
n
q i0 . . . q ik 0 ≤ i0 < · · · < i k ≤ r . gcd (q0 , . . . , qik )
o
For integral cohomology we have:
Theorem 5.4. Let Q be an (r + 1)tuple of positive integers. Then Z, if i is even, H i (P(Q); Z) ∼ = 0, if i is odd . Further , take [ψQ ] : P r → P(Q) as the quotient map defined by (5.1). Then the diagram [ψQ]∗ 2k r H (P ; Z) H 2k (P(Q); Z) ∼ = ? Z
lQ,k
∼ = ?  Z
156
KAREN CHANDLER AND PITMANN WONG
commutes, where the lower map is given by multiplication by the number lQ,k . Note that the number lQ,r is precisely the number of preimages of a point in P(Q) under the quotient map [ψQ ] (see (5.1)). The proof of the preceding Theorem for k = r is easy; the reader is referred to [Kawasaki 1973] for the general case. Let Q = (q0 , q1 , . . . , qr ), r ≥ 1, be an (r +1)tuple of positive integers. The tuple Q is said to be reduced if the greatest common divisor (gcd) of (q 0 , q1 , . . . , qr ) is 1. In general, if the gcd is d, the tuple Qred = Q/d = (q0 /d, . . . , qr /d) is called the reduction of Q. Let d0 = gcd(q1 , . . . , qr ), di = gcd(q0 , q1 , . . . , qi−1 , qi+1 , . . . , qr ),
1 ≤ i ≤ r − 1,
dr = gcd(q0 , . . . , qr−1 ) and define a0 = lcm(d1 , . . . , dr ), ai = lcm(d0 , d1 , . . . , di−1 , di+1 , . . . , dr ),
1 ≤ i ≤ r − 1,
ar = lcm(d0 , . . . , dr−1 ), where “lcm” is short for “least common multiple”. Define the normalization of Q by Qnorm = (q0 /a0 , . . . , qr /ar ). A tuple Q is said to be normalized if Q = Qnorm . Theorem 5.5. Let Q be a normalized (r+1)tuple of positive integers. Then the Picard group Pic(P(Q)) and the divisor class group Cl(P(Q)) are both isomorphic to Z, and are generated , respectively, by mQ LP(Q) = OP(Q) (mQ ) and LP(Q) = OP(Q) (1) .
Note that the generators of the two groups are different in general. For the standard projective space we have mQ = 1 and so the generators are the same. For the kjet bundles the fibers of their projectivization are weighted projective spaces with mQ = k!, so we shall only be concerned with the case where n, k ≥ 1 are positive integers and Q = (1, . . . , 1), (2, . . . , 2), . . . , (k, . . . , k ) ,  {z }  {z }  {z } n
n
n
which is normalized. In this case we shall write P n,k for P(Q). Note that r = dim P n,k = nk − 1; the least common multiple of Q is mQ = k!
and lQ,r = (k!)n .
(5.2)
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
157
Define a positive function ρQ (z0 , . . . , zr ) =
r X
zi 2/qi
(5.3)
i=0
on (C r+1 , Q) \ {0}. Then r X
ρQ (λq0 z0 , . . . , λqr zr ) = λ2
zi 2/qi = λ2 ρQ (z0 , . . . , zr )
i=0
and ψ ∗ (ρQ )(z0 , . . . , zr ) =
r X
ziqi 2/qi =
i=0
r X
zi 2 = ρ1 (z0 , . . . , zr )
i=0
is the standard Euclidean norm function on (C r+1 , 1). The function ρQ is not S differentiable along Z = {[zqi = 0], qi 6= 1}. However, on C r+1 \ Z, we deduce from the above that ∂ ∂¯ log ρQ (λq0 z0 , . . . , λqr zr ) = ∂ ∂¯ log ρQ (z0 , . . . , zr ) and that ∗ ψQ (∂ ∂¯ log ρQ ) = ∂ ∂¯ log ρ1 .
The first identity shows that ∂ ∂¯ log ρQ is invariant under the C ∗ action hence descends to a welldefined (1, 1)form ωQ on P(Q) \ πQ (Z). The second identity ∗ says that ψQ (ωQ ) is the Fubini–Study metric ωF S on the standard projective r space P \ π(Z) (hence actually extends smoothly across π(Z)). The Fubini– Study metric [ωF S ] is the first Chern form of OP r (1) which is the (positive) generator of Pic P r = Cl P r . Hence [ωF S ] is the positive generator of H 2 (P r , Z). Theorem 5.4 implies that [lQ,1 ωQ ] is the generator of H 2 (P(Q), Z). Pr 2κ 1/κ , for κ a positive integer, defined on Consider the function i=0 zi  C r+1 . It clearly satisfies Pr
i=0
λzi 2κ
1/κ
= λ2 (
Pr
i=0
zi 2κ )1/κ ;
hence is a metric along the fibers of the tautological line bundle over P r . Moreover, the form Pr 2κ 1/κ ∂ ∂¯ log i=0 zi 
descends to a welldefined form on the standard projective space P r , indeed a Chern form, denoted by ηκ , for the hyperplane bundle of P r ; moreover it is cohomologous to the Fubini–Study form. With this we may define an alternative to ρQ , τQ (z0 , . . . , zr ) =
r X i=0
zi 2κ/qi
1/κ
,
κ=
r Y
i=0
qi .
(5.4)
158
KAREN CHANDLER AND PITMANN WONG
It is of class C ∞ on C r+1 \ {0}. Just like ρQ , the function τQ satisfies τQ (λ z0 , . . . , λ zr ) = q0
qr
r X
2κ
λ zi 
2κ/qi
i=0
1/κ
= λ
2
r X
zi 
2κ/qi
i=0
1/κ
and ∗
(ψ τQ )(z0 , . . . , zr ) =
τQ (z0q0 , . . . , zrqr )
=
r X i=0
zi 
2κ
1/κ
.
These equalities imply that ∂ ∂¯ log τQ descends to a welldefined form γQ on P(Q) ∗ with the property that ψQ γQ = ηk , and consequently is cohomologous to ωQ . k Let π : J X → X be the parametrized kjet bundle of a complex manifold X and denote by p : P(J k X) → X and pr : p∗ J k X → P(J k X) the corresponding projection maps. The following diagram is commutative: p∗ J k X
p∗ 
pr
π ? p X
? P(J k X)
J kX
and the tautological subsheaf of p∗ J k X is the line sheaf defined by {([ξ], η) ∈ p∗ J k X  [ξ] ∈ P(J k X), p([ξ]) = π(η) = x, [η] = [ξ]} where, for ξ (resp. η) in J k X, its equivalence class in P(J k X) is denoted by [ξ] (resp. [η]). The “hyperplane sheaf ”, denoted L = L k , is defined to be the dual of the tautological line sheaf. The fiber P(Jxk X) over a point x ∈ X is the weighted projective space of type Q = ((1, . . . , 1); . . . ; (k, . . . , k)) and the restriction of L k to P(Jxk X) is the line sheaf OP(Q) (1) as defined in Theorem 5.2. The next result follows readily from Theorem 5.2: Theorem 5.6. Let X be a complex manifold . (i) For any m ∈ Z, Lm P(J k X) is a reflexive coherent sheaf . k! (ii) LP(J k X) is the generator of Pic(P(J k X)), that is, Lm P(J k X) is locally free if m is divisible by k!. β ∼ k!α+β (iii) For any α, β ∈ Z, Lk!α P(J k X) ⊗ LP(J k X) = LP(J k X) . The Chern class of the bundle Lk! P(J k X) is k! ωQ = lQ,1 ωQ , where ωQ is constructed after Theorem 5.5. By (5.3) the function ρ k! Q is a Finsler metric along k! the fibers of LP(J k X) . The same is of course also true if we use γ Q and τQ instead. Just as in the case of projectivized vector bundles we still have the identification \ {0} with J k X \ {0}, which is compatible with the respecof the spaces L−1 P(J k X) ∗ tive C action. Thus, as in the case of vector bundles, we conclude that a metric along the fibers of L−1 is identified with a Finsler metric along the fibers of P(J k X)
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
159
J k X. As J k X, in general, is only a C ∗ bundle and not a vector bundle, we see that Finsler geometry is indispensable. Note that, for k ≥ 2, the sheaf LP(J k X) is not locally free and, in general, LaP(J k X) ⊗ LbP(J k X) (a) 6∼ . = La+b P(J k X) Hence some of the proofs of the results that are valid for projectivized vector bundles require modifications. Basically, things work well if we use integer multiples of k! (that is, Jkmk! X); for example, Grothendieck’s Theorem (Theorem 4.2) remains valid: Theorem 5.7. Let X be a complex manifold and p : P(J k X) → X the kth parametrized jet bundle and let S be an analytic sheaf on X. For any m ≥ 1, mk! Jk X, if i = 0, i mk! ∼ p∗ LP(J k X) = 0, if i > 0, mk! where pi∗ Lmk! P(J k X) is the ith direct image of LP(J k X) . Consequently, we have ∗ H i (X, Jkmk! X ⊗ S) ∼ = H i (P(J k X), Lmk! P(J k X) ⊗ p S)
for all i. In the case of vector bundles, Theorem 4.3 provides a relation between the Chern numbers of the bundle and that of the line bundle over the projectivization. Theorem 4.3 may be proved directly via fiber integration. Although the projectivized kjet bundles are not smooth for k ≥ 2 this correspondence is still valid. These technicalities are needed when we deal with problem of degeneration; as we shall see in Sections 6 and 7, under the condition that L k! k is big, kjets of holomorphic maps into X are algebraically degenerate, that is, the images are contained in some (special type of) subvarieties of P(J k X) which may be very singular. In order to calculate the Euler characteristic of L k! k X of these subvarieties it is necessary to compute the intersection numbers, as usual, via Chern classes and this is best handled by going down, via fiber integration, to the base variety X which is nonsingular. We take this opportunity to formulate a criterion for certain type of singular spaces on which fiber integration works well. The purpose here is not to exhibit the most general results but results general enough for our purpose. First we recall some basic facts concerning fiber integration. Let P and X be complex manifolds and p : P → X be a holomorphic surjection. The map p is said to be regular at a point y ∈ P if the Jacobian of p at y is of maximal rank. The set of regular points is an open subset of P and p is said to be regular if every point of P is a regular point. The following statements concerning fiber integration are wellknown (see [Stoll 1965], for example): Theorem 5.8. Let P and X be connected complex manifolds of dimension N and n respectively. Let p : P → X be a regular holomorphic surjection. Let r, s be integers with r, s ≥ N − n = q. Then for any (r, s)form ω of class C k on
160
KAREN CHANDLER AND PITMANN WONG
P that is integrable along the fibers of p, the fiber integral p∗ ω is a well defined (r − q, s − q) form of class C k on X. Moreover : (i) For any (N − r, N − s)form on X such that ω ∧ p∗ η is integrable on P , we have Z Z ∗ ω∧p η = p∗ ω ∧ η. P
X
(ii) If ω is of class C 1 and if ω and dω are integrable along the fibers p then dp∗ ω = p∗ dω, ∂p∗ ω = p∗ ∂ω and ∂p∗ ω = p∗ ∂ω. (iii) If ω is nonnegative and integrable along the fibers of p then p∗ ω is also nonnegative. (iv) Suppose that Y is another connected complex manifold of dimension n0 with a regular holomorphic surjection π : X → Y . Assume that ω is a (r, s)form such that r, s ≥ q +q 0 where q = N −n and q 0 = n−n0 . If ω is integrable along the fibers of p and p∗ ω is integrable along the fibers of π then π∗ p∗ ω = (π◦p)∗ ω. If ω is a form of bidegree (r, s) so that either r < q or s < q, where q is the fiber dimension, then we set p∗ ω = 0. If p : P → X is a holomorphic fiber bundle with smooth fiber S, then p is a regular surjection and the preceding Theorem is applicable. Consider now P , an irreducible complex space of complex dimension N , with a holomorphic surjection p : P → X where X is nonsingular and of complex dimension n. The map p is said to be regular if there exists a connected complex manifold P˜ of the same dimension as P and a surjective morphism τ : P˜ → P such that the composite map p˜ = p ◦ τ : P˜ → X is regular. Let 0 U ⊂ P be an open set and ι : U → V ⊂ C N a local embedding, where V is 0 an open set of C N for some N 0 . If η is a differential form on V then ι∗ η is a differential form on U . Conversely, a differential form ω on U is of the form ι ∗ η for some embedding ι : U → V and some differential form η on V . Suppose that ω is a differential form on P of bidegree (r, s); hence τ ∗ ω is a differential form on P˜ of bidegree (r, s). If either r or s is less than the fiber dimension q = N − n then p∗ ω is defined to be zero. For the case r, s ≥ q and assuming that τ ∗ ω is integrable along the fibers of p˜ (for example, this is the case if ω is integrable along the fibers of p), we are in the nonsingular situation; hence p˜∗ τ ∗ ω is defined. The pushforward p∗ ω is naturally defined by def
p∗ ω = p˜∗ τ ∗ ω.
(5.5)
From this definition it is clear (since p˜ = p ◦ τ ) that the basic properties of fiber integrals remain valid in the more general situation: Theorem 5.9. Let τ˜ : P˜ → P, p : P → X and p˜ : P˜ → X be as above and let ω be a form of bidegree (r, s) on P with r, s ≥ N − n where n = dim X, N = dim P = dim P˜ . Then: (i) If τ ∗ ω is integrable along the fibers P˜y = p˜−1 (x) for almost all x ∈ X then for any (N − r, N − s)form on X such that ω ∧ p∗ η is integrable on P and
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
161
τ ∗ ω ∧ p˜∗ η is integrable on P˜ , Z Z Z ω ∧ p∗ η = p∗ ω ∧ η = p˜∗ τ ∗ ω ∧ τ ∗ η. P
X
P˜
(ii) If ω is of class C 1 and if τ ∗ ω and τ ∗ dω are integrable along the fibers P˜x for all x ∈ X then dp∗ ω = p∗ dω, ∂p∗ ω = p∗ ∂ω and ∂p∗ ω = p∗ ∂ω. ˜ x for all x ∈ X then p∗ ω is a form of type (iii) If τ ∗ ω is integrable along M (p − N + n, q − N + n) on X. (iv) If ω is a continuous nonnegative form and τ ∗ ω is integrable along P˜x for all x ∈ X then p∗ ω is also nonnegative. The converse of part (iv) is not true in general. The next theorem shows that the preceding theorem is applicable to the projectivized kjet bundles (we refer the readers to [Stoll and Wong 2002] for details). Theorem 5.10. Let X be a complex manifold of complex dimension n and let p : P = P(J k X) → X be the projectivized kjet bundle of X. Then there exists a complex manifold P˜ of the same dimension as P and a surjective finite morphism τ˜ : P˜ → P such that p˜ = p ◦ τ : P˜ → X is a regular holomorphic surjection. Moreover , P˜ can be chosen so that each of the fibers of p˜ is the complex projective space P q where q = nk − 1. A similar argument (see [Stoll and Wong 2002]) shows that in general we have: Theorem 5.11. Let X be a connected complex manifold of complex dimension n. Suppose that P is an irreducible complex space for which there exists a holomorphic surjective morphism p : P → X that is locally trivial ; that is, for any x ∈ X there exists an open neighborhood V of X, a complex space Y and a biholomorphic map αV : p−1 V → V × Y such that the diagram p−1 (V )
αVV ×Y ∼ =
pV ? ? V ========= V
p
commutes, where pV is the projection onto the first factor . Then there exists a complex manifold P˜ of the same dimension as P and a surjective morphism τ : P˜ → P such that p˜ = p ◦ τ : P˜ → X is a regular holomorphic surjection. Next we extend the definition of pushforward of forms to subvarieties of a complex space P with a projection map p : P → X satisfying the local triviality condition of the preceding theorem. In general the pushforwards exist only as currents. Suppose that Y ⊂ P is an irreducible subvariety of dimension ν of P and assume that pY : Y → X is surjective. Let Σ ⊂ Y be the set of singular points of Y ; so the set S1 = {z ∈ X  (pY )−1 (z) ⊂ Σ} is a subvariety of
162
KAREN CHANDLER AND PITMANN WONG
codimension at least one in X. Note that pY \Σ : Y \ Σ → X \ S1 is surjective, hence generically regular; that is, there exists a subvariety S 2 ⊂ X of codimension at least 1 such that pY \(Σ∪(pY )−1 (S2 )) : Y1 = Y \ (Σ ∪ (pY )−1 (S2 )) → X1 = X \ (S1 ∪ S2 ) is a regular surjection. Let ω be a smooth (r, s)form on Y , r, s ≥ N − n = generic fiber dimension of pY1 , which is integrable along the fibers of pY1 . Then (pY1 )∗ ω is a (p − N + n, q − N + n)form on X1 . Meanwhile, the pushforward (pY )∗ ω exists as a current on X, that is, def
(pY )∗ ω((pY )∗ φ) = ω(φ) =
Z
(pY )∗ φ ∧ ω
(5.6)
Y
for any (N −r, N −s)form φ with compact support on X. Clearly, (p Y )∗ ωX1 = (pY )∗ ω. Note that as a current the pushforward commutes with exterior differentiation, that is, d(pY )∗ ω = p∗ dω, ∂(pY )∗ ω = (pY )∗ ∂ω and ∂(pY )∗ ω = (pY )∗ ∂ω. Also, by definition, the pushforward preserves nonnegativity. The Riemann–Roch formulas for jet differentials follow from those of the J i1 ∗ J ik ∗ bundles T X ⊗···⊗ T X, given below (see [Stoll and Wong 2002] for details): Theorem 5.12. Let X be a smooth compact complex surface. Set I = (i1 , . . . , ik ) J i1 ∗ Ji and SI = T X⊗· · ·⊗ k T ∗ X, where each ij is a nonnegative integer . Then
χ(X; SI ) =
2 1 12 µk (2s1;k
− 2s1;k − s2;k + 1)c21 (T ∗ X)
1 − 12 µk (2s21;k + 4s1;k − 4s2;k − 1)c2 (T ∗ X),
where sj;k , for 1 ≤ j ≤ k, is the degreej symmetric function in i1 , . . . , ik and Pk µk = j=0 sj;k is as in (3.6). Given an exact sequence of coherent sheaves 0 → E1 → E2 → E3 → 0 the ranks, the first Chern classes, the Chern characters, the indices and the Euler characteristics are additive in the sense that rk E 2 = rk E1 + rk E3 , c1 (E2 ) = c1 (E1 ) + c1 (E3 ), ι(E2 ) = ι(E1 ) + ι(E3 ), ch(E2 ) = ch(E1 ) + ch(E3 ) and χ(X; E2 ) = χ(X; E1 ) + χ(X; E3 ). The Euler characteristic of Jkk! X is given thus: Theorem 5.13. Let X be a nonsingular surface. We have, for m k, χ(Jkk!m X) = 12 ι(Jkk!m X) + O(m2k ) = 12 (k!)2k+1 (αk c21 − βk c2 ) m2k+1 + O(m2k ), where αk and βk are constants given in Theorem 3.9.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
163
Example 5.14. We record below explicit formulas for the sheaves that occur in the preceding computations: ch(S)
S ∗
T X
∗
( (
J3
c1 + 3c1 +
1 (5c21 2
6c1 +
7c21
J2
T X
J3
∗
T X
4+
J4
T ∗X
5 + 10c1 +
J5
∗
T X
J6
∗
J7
∗
3+
− 2c2 ) − 8c2 )
1 (5c21 4
− 15c2 )
− 10c2
1 (13c21 3
− 29c2 )
15c21 − 20c2
1 (125c21 12
6 + 15c1 +
T X
7 + 21c1 +
1 (91c21 2
T ∗X ∗
∗
∗
∗
∗
T X) ⊗ T X
T X) ⊗ T X J ( T ∗ X) ⊗ ( 2 T ∗ X) J ( 4 T ∗ X) ⊗ T ∗ X J J ( 3 T ∗ X) ⊗ ( 2 T ∗ X) J ( 5 T ∗ X) ⊗ T ∗ X J2
J22 X
χ(S) 1 2 (c 6 1
1 (55c21 2
T X ⊗T X J2
2+
1 2 (c 2 1
− 70c2 )
1 (41c21 2
− 5c2 )
− 235c2 ) − 84c2 )
− 112c2 )
1 (427c21 12
− 665c2 )
7 + 21c1 + 21 (91c21 − 112c2 )
1 (170c21 3
− 250c2 )
4c1 +
3c21
9c1 +
1 (19c21 2
8 + 16c1 +
22c21
− 24c2
1 (44c21 3
9 + 18c1 +
24c21 − 24c2
1 (63c21 4
4+ 6+
10 + 25c1 +
1 (85c21 2
12 + 30c1 +
49c21
12 + 36c1 +
73c21
1 (4c21 3
− 11c2 )
− 22c2 )
1 (11c21 2
− 21c2 ) − 70c2 ) − 93c2 )
− 4c2
− 90c2 )
1 (185c21 6
− 265c2 )
− 46c2
35c21
− 45c2
− 76c2
56c21
− 75c2
1 (17c21 12
− 55c2 )
− 14c2
1 (23c21 4
− 53c2 )
4c1 +
3c21
J23 X
8 + 10c1 +
10c21
J24 X
14 + 22c1 +
27c21 − 35c2
1 (103c21 6
− 207c2 )
20 + 40c1 +
59c21
− 70c2
1 (122c21 3
− 205c2 )
30 + 70c1 +
119c21
− 135c2
1 (173c21 2
− 265c2 )
40 + 110c1 +
214c21
− 200c2
1 (487c21 3
− 590c2 )
J25 X J26 X J27 X
5+
− 5c2
Although the space P(J k X) is not smooth, the following Riemann–Roch Theorem is still valid, by Theorems 5.7 and 5.13: Theorem 5.15. Let X be a nonsingular surface and p : P(J k X) → X the kjet bundle. Then m k k χ(Lm P(J k X) ) = ch(L ) . td(P(J X))[P(J X)]
= ch(Lm ) . td(Tp ) . p∗ td(X)[P(J k X)] = p∗ ch(Lm ) . td(Tp ) .td(X)[X],
where Tp is the relative tangent sheaf of the projection p : P(J k X) → X, that is, the restriction of Tp to each fiber of p is the tangent sheaf of the weighted projective space P(Q).
164
KAREN CHANDLER AND PITMANN WONG
On P(J k X) we have ch(Lm ) =
2k+1 X i=0
2k+1 X ci (L) ci1 (Lm ) 1 = mi i! i! i=0
which implies that χ(Lm ) =
c2k+1 (L) 2k+1 1 m + O(m2k ). (2k + 1)!
Theorems 5.13 and 5.15 imply: Corollary 5.16. Let X be a nonsingular surface and p : P(J k X) → X the kjet bundle. Then p∗
c2k+1 (L) 1 (2k + 1)!
=
1 1 ι(Jk X) = (αk c21 − βk c2 ) 2 2
where αk and βk are constants given in Theorem 3.9.
6. The Lemma of Logarithmic Derivatives and the Schwarz Lemma Summary. In this section we use Nevanlinna Theory to show (Corollary 6.2) that if ω is a holomorphic kjet differential of weight m vanishing on an effective divisor of a projective manifold X then f ∗ ω ≡ 0 for any holomorphic map f : C → X. (For our application in Section 7 it is enough to assume that the divisor is a hyperplane section.) This implies (see Theorem 6.4) that if f : C → X is an algebraically nondegenerate holomorphic map then the irreducible component of the base locus containing [j k f ] is of codimension at most (n−1)k; equivalently the dimension is at least n + k − 1, n = dim X. This result is crucial in the proof of our main result in Section 7. We must point out that the method of this section works only for the parametrized jet bundles but not the full jet bundles. (Otherwise we could have avoided the complicated computations of the Chern numbers of the parametrized jet bundles; computing the Chern numbers of the full jet bundles, as honest vector bundles, is much simpler !) The idea of the proof is relatively standard from the point of view of Nevanlinna Theory. The main step is to construct, using a standard algebraic geometric argument, a Finsler metric of logarithmic type, reducing the problem to a situation in which the Lemma of Logarithmic Derivatives is applicable. If f ∗ ω 6≡ 0 this lemma implies that the integral of log f ∗ ω is small . On the other hand , the first Crofton formula in Nevanlinna Theory asserts that the integral of log f ∗ ω (as the counting function of the zeros of f ∗ ω by the Poincar´e–Lelong formula ) is not small . This contradiction establishes Theorem 6.1 and Corollary 6.2. Theorem 6.4 and Corollary 6.5 then follow from the Schwarz Lemma via a reparametrization argument often used in Nevanlinna Theory. The main point is that a reparametrization does
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
165
not change, as a set, the (algebraic closure of the) image of the map f but does change the image of f ∗ ω if ω is a kjet differential and k ≥ 2. The classical Schwarz Lemma in one complex variable asserts that a holomorphic map f : C → X is constant for a compact Riemann surface X of genus ≥ 2; that is, there are at least 2 independent regular 1forms on X. Further, there is a noncompact version; namely, let X be a compact Riemann surface and let D be a finite number of points in X. Then a holomorphic map f : C → X \D is constant if the logarithmic genus is at least 2, that is, there are at least 2 independent 1forms regular on X \ D and with no worse than logarithmic singularity at each of the points in D. There are of course many proofs of this classical result, one of which is to find a nontrivial holomorphic 1form (or, logarithmic 1form) on X such that f ∗ ω ≡ 0 for any entire holomorphic map f : C → X. This is not so difficult to do because g ≥ 2 implies that T ∗ X is ample (and a priori, spanned). The main difficulty of proving the preceding comes from the fact that a big bundle is not necessary spanned. A coherent sheaf S is said to be spanned (by global regular sections) if, for every v ∈ Sx there is a global regular section σ ∈ H 0 (S) such that σ(x) = v. However, it is easily seen that a coherent sheaf S is spanned by global rational sections. For example, the complex projective space has no global regular 1form. Hence it cannot span any of the fibers of T ∗ P n . However take any point x ∈ P n , assuming without loss of generality that x = [x0 , . . . , xn ] with x0 6= 0, then Tx∗ P n is spanned by dti , i = 1, . . . , n, where ti = xi /x0 . Now dti is a global rational oneform since ti is a global rational function; in fact x0 dxi − xi dx0 dti = x20 has a pole of order 2 along the “hyperplane at infinity”, [x 0 =0]. This shows that T ∗ P n is spanned by global rational oneforms. In fact we can do better, namely, we may replace dti by d log ti d log ti =
dti dxi dx0 = − . ti xi x0
A simple argument shows that there is a finite set of logarithmic oneforms {dLi /Li } where each Li is a rational function which span T ∗ P n at every point. The mild singularity can be dealt with using the classical Lemma of Logarithmic Derivatives in Nevanlinna Theory and a weak form of the analytic B´ezout Theorem known as Crofton’s Formula. It is not hard to see that the preceding procedure can be extended to deal with jet differentials. The details are given in the next theorem. The most convenient way to get to the Schwarz Lemma is via Nevanlinna Theory. First we recall some standard terminology. The characteristic function of a map f : C → X is Z r Z 1 f ∗ c1 (H), Tf (r) = 0 t ∆t
166
KAREN CHANDLER AND PITMANN WONG
where H is a hyperplane section in X and the characteristic function of a nontrivial holomorphic function F : C → C is Z 2π √ dθ log+ F (re −1θ ) . TF (r) = 2π 0 Note that ω(j k f ) is a holomorphic function if f is a holomorphic map and ω is a kjet differential of weight m. Theorem 6.1 (Lemma of Logarithmic Derivatives). Let X be a projective variety and let (i) D be an effective divisor with simple normal crossings, or (ii) D be the trivial divisor in X (that is, the support of D is empty or equivalently, the line bundle associated to D is trivial). Let f : C → X be an algebraically nondegenerate holomorphic map and ω ∈ H 0 (X, Jkm X(log D)) (resp. H 0 (X, Jkm X) in case (ii)) a jet differential such that ω ◦ j k f is not identically zero. Then Z 2π √ dθ ≤ O(log Tf (r)) + O(log r). log+ ω(j k f (re −1θ )) Tω◦j k f (r) = 2π 0
Proof. We claim that there exist a finite number of rational functions t 1 , . . . , tq on X such that (†) the logarithmic jet differentials {(d(j) ti /ti )m/j  1 ≤ i ≤ q, 1 ≤ j ≤ k} span the fibers of Jkm X(log D) (resp.Jkm X) over every point of X.
Note that rational jet differentials span the fibers of J km X(log D) (resp.Jkm X); the claim here is that this can be achieved by those of logarithmic type. Without loss of generality we may assume that D is ample; otherwise we may replace D by D + D 0 so that D + D 0 is ample. (This is so because a section of Jkm X(log D) is a priori a section of Jkm X(log(D + D 0 )).) Observe that if s is a function that is holomorphic on a neighborhood U such that [s = 0] = D ∩ U then [s τ = 0] = τ D ∩ U for any rational number τ . Thus δ (j) (log sτ ) = τ δ (j) (log s) is still a jet differential with logarithmic singularity along D ∩ U so the multiplicity causes no problem. This implies that we may assume without loss of generality that D is very ample (after perhaps replacing D with τ D for some τ for which τ D is very ample). Let u ∈ H 0 (X, [D]) be a section such that D = [u = 0]. At a point x ∈ D choose a section v1 ∈ H 0 (X, [D]) so that E1 = [v1 = 0] is smooth, D + E1 is of simple normal crossings and v1 is nonvanishing at x. (This is possible because the line bundle [D] is very ample.) The rational function t 1 = u1 /v1 is regular on the affine open neighborhood X \E1 of x and (X \E1 )∩[t1 =0] = (X \E1 )∩D. Choose rational functions t2 = u2 /v2 , . . . , tn = un /vn where ui and vi are sections of a very ample bundle L so that t2 , . . . , tn are regular at x, the divisors Di = [ui =0], Ei = [vi =0] are smooth and the divisor D +D2 +· · ·+Dn +E1 +· · ·+En is of simple normal crossings. Further, since the bundles involved are very ample the sections can be chosen so that dt1 ∧· · ·∧dtn is nonvanishing at x; the complete
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
167
system of sections provides an embedding. Hence at each point there are n + 1 sections with the property that n of the quotients of these n + 1 sections form a local coordinate system on some open neighborhood U x of x. This implies that (†) is satisfied over Ux . Since D is compact it is covered by a finite number of such open neighborhoods, say U1 , . . . , Up , and a finite number of rational functions S (constructed as above for each Ui ) on X so that (†) is satisfied on 1≤i≤p Ui . Moreover, there exist relatively compact open subsets U i0 of Ui (1 ≤ i ≤ p) such S that 1≤i≤p Ui0 still covers D. T Next we consider a point x in the compact set X \ 1≤i≤p Ui0 . Repeating the procedure as above we may find rational functions s 1 = a1 /b1 , . . . , sn = an /bn where ai and bi are sections of some very ample line bundle so that s 1 , . . . , sn form a holomorphic local coordinate system on some open neighborhood V x of x. Thus (†) is satisfied on Vx by the rational functions s1 , . . . , sn . Note that we must also choose these sections so that the divisor H = [s1 . . . sn = 0] together with those divisors (finite in number), which have been already constructed above, is still a divisor with simple normal crossings (this is possible by the very ampleness of the T line bundle L.) Since X \ 1≤i≤p Ui0 is compact, it is covered by a finite number of such coordinate neighborhoods. The coordinates are rational functions and finite in number and by construction it is clear that the condition (†) is satisfied T S T on X \ 1≤i≤p Ui0 . Since 1≤i≤p Ui together with X \ 1≤i≤p Ui0 covers X, the condition (†) is satisfied on X. If D is the trivial divisor, then it is enough to use only the second part of the construction above and again (†) is verified with Jkm X(log D) = Jkm X. To obtain the estimate of the theorem observe that the function ρ : J k X(− log D) → [0, ∞] defined by ρ(ξ) =
q X k X (j) (d ti /ti )m/j (ξ) 2 ,
ξ ∈ J k X(− log D),
(6.1)
i=1 j=1
{ti } being the family of rational functions satisfying condition (†), is continuous in the extended sense; it is continuous in the usual sense outside the fibers over the divisor E (the sum of the divisors associated to the rational functions {t }; note that E contains D). Over the fiber of each point x ∈ X − E, (j) i m/j 2 (d ti /ti ) (ξ) is finite for ξ ∈ J k X(− log D)x , thus ρ is not identically infinite. Moreover, since (j) (d ti /ti )m/j  1 ≤ i ≤ q, 1 ≤ j ≤ k
span the fiber of Jkm X(log D) over every point of X, ρ is strictly positive (possibly +∞) outside the zero section of J k X(− log D). The quotient ω2 /ρ : J k X(− log D) → [0, ∞]
does not take on the extended value ∞ when restricted to J k X(− log D)\ {zero section} because, as we have just observed, ρ is nonvanishing (although it does blow up along the fibers over E so that the reciprocal 1/ρ is zero there) and the
168
KAREN CHANDLER AND PITMANN WONG
singularity of ω is no worse than that of ρ since the singularity of ω occurs only along D (which is contained in E) and is of log type. Thus the restriction to Jk X(− log D) \ {zero section}, ω2 /ρ : J k X(− log D) \ {zero section} → [0, ∞), is a continuous nonnegative function. Moreover, ω and ρ have the same homogeneity, ω(λ.ξ)2 = λ2m ω(λ.ξ)2 and ρ(λ.ξ) = λ2m ρ(ξ), for all λ ∈ C ∗ and ξ ∈ J k X(− log D); therefore ω2 /ρ descends to a welldefined function on P(Ek,D ) = (J k X(− log D) \ {zero section})/C ∗ , that is, ω2 /ρ : P(Ek,D ) → [0, ∞) is a welldefined continuous function and so, by compactness, there exists a constant c with the property that ω2 ≤ cρ. This implies that Z 2π √ dθ Tω◦j k f (r) = log+ ω(j k f (re −1θ )) 2π 0 Z 2π √ dθ ≤ log+ ρ(j k f (re −1θ )) + O(1). 2π 0 Since ti is a rational function on X, the function
(d(j) ti /ti )m/j (j k f ) = (ti ◦f )(j)/ti ◦f
m/j
(m is divisible by k!) is meromorphic on C and so, by the definition of ρ, log+ ρ(j k f ) ≤ O max1≤i≤q, 1≤j≤k log+ (ti ◦f )(j)/ti ◦f  + O(1).
Now by the classical lemma of logarithmic derivatives for meromorphic functions, Z 2π dθ ·≤· O(log r) + O(log Tti ◦f (r)), log+ (ti ◦f )(j)/ti ◦f 2π 0
where ·≤· indicates that the estimate holds outside a set of finite Lebesgue measure in R + . Since ti is a rational function, log Tti ◦f (r) ≤ O(log Tf (r)) + O(log r) and we arrive at the estimate Z 2π Z 2π √ k dθ + + −1θ θ (j) )) log ρ(j f (re ≤ O log (ti ◦f ) /ti ◦f  + O(1) 2π 2π 0 0 ·≤· O(log Tf (r)) + O(log r).
This implies that Tω ◦j k f (r) ·≤· O(log Tf (r)) + O(log r), as claimed.
˜
We obtain as a consequence the following Schwarz type lemma for logarithmic jet differentials.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
169
Corollary 6.2. Let X be a projective variety and D be an effective divisor (possibly the trivial divisor) with simple normal crossings. Let f : C → X \ D be a holomorphic map. Then ω(j k f ) ≡ 0
for all ω ∈ H 0 (X, Jkm X(log D) ⊗ [−H]),
where H is a generic hyperplane section (and hence any hyperplane section). Proof. If f is constant the corollary holds trivially, so we may assume that f is nonconstant. Now suppose that ω ◦ j k f 6≡ 0. Since f is nonconstant, we may assume without loss of generality that log r = o(T f (r)) (after perhaps replacing f with f ◦ φ, where φ is a transcendental function on C). By Theorem 6.1, we have Z 2π dθ = Tω◦j k f (r) ·≤· O log(rTf (r)) . log+ ω ◦ j k f  2π 0
On the other hand, since ω vanishes on H and H is generic, we obtain via Jensen’s Formula , Tf (r) ≤ Nf (H; r) + O log(rTf (r)) Z 2π dθ + O(log rTf (r)) , log ω ◦ j k f  = 2π 0 which, together with the preceding estimate, implies that Tf (r) ≤ O log(rTf (r)) .
This is impossible; hence we must have ω ◦ j k f ≡ 0. If H1 = [s1 = 0] is any hyperplane section then it is linearly equivalent to a generic hyperplane section H = [s=0]. If ω vanishes along H 0 then (s/s1 )ω vanishes along H. The preceding discussion implies that (s/s1 )ω(j k f ) ≡ 0. Further, this implies that ω(j k f ) ≡ 0 as we may choose a generic section H so that the image of f is not entirely contained in H. ˜ Interpreting this corollary via Grothendieck’s isomorphism we may restate the ∗ result in terms of sections of Lm P(J k X) Y ⊗ pY [−D] on the projectivized bundle: Corollary 6.3. Let Y ⊂ P(J k X) be a subvariety and suppose that there exists a nontrivial section ∗ σ ∈ H 0 Y, Lm P(J k X) Y ⊗ pY [−D] ,
where D is an ample divisor in X and p : P(J k X) → X is the projection map. If the image of the lifting [j k f ] : C → P(J k X) of a holomorphic curve f : C → X is contained in Y , then σ([j k f ]) ≡ 0. Theorem 6.1 and Corollaries 6.2 and 6.3 tell us about the base locus B km (D) of ∗ m m the line sheaves Lm k ⊗ p [−D], where we write for simplicity L k = LP(J k X) and
170
KAREN CHANDLER AND PITMANN WONG
D is an ample divisor in X; by that we mean the (geometric) intersection of all possible sections of powers of Lk : \ Bkm (D) = [σ = 0]. (6.2) σ∈H 0 (P(J k X),Lm k )
Indeed, Corollary 6.3 implies that the image of the (projectivized kjet) [j k f ] : C → P(J k X) of a nonconstant holomorphic map f : C → X must be contained in Bkm (D) for all m ∈ N and D ∈ A = the cone of all ample divisors; that is, the image [j k f ](C) is contained in \ \ Bkm (D), (6.3) Bk (Lk ) = m∈N D∈A
which is a subvariety of P(J k X). Moreover, the image [j k f ](C), being a connected set, must be contained in an irreducible component of B k (L). If f is algebraically nondegenerate then dim f (C) = dim X = n. Since p ∗ [j k f (C)] = f (C) and [j k f (C)] ⊂ Bk (L) (where p : P(J k X) → X is the projection) we conclude that the dimension of the base locus is at least n = dim X if f is algebraically nondegenerate. We shall show that the dimension is actually higher, for k ≥ 2, by considering a reparametrization of the curve f . Define A = {φ  φ : C → C is a nonconstant holomorphic map}, Aζ0 = {φ ∈ A  φ(ζ0 ) = ζ0 , φ0 (ζ0 ) 6= 0}, Aζ0 ,ζ1 = {φ ∈ A  φ(ζ0 ) = ζ1 , φ0 (ζ0 ) 6= 0}. By a reparametrization of f we mean the composite map f ◦ φ : C → X, where φ ∈ A. It is clear that, as a set, the algebraic closure of the image of f is invariant by reparametrization. Moreover, since a reparametrization is again a curve in X, the Schwarz Lemma implies that its kjet is contained in the base locus B k (L). As remarked earlier, if f is algebraically nondegenerate the dimension of the base locus is at least n. The first order jet of a reparametrization is given by j 1 (f ◦ φ) = (f (ζ), f 0 (φ)φ0 ). Thus, if φ ∈ A0 (that is, φ(0) = 0), then j 1 (f ◦ φ)(0) = f (φ(0)), f 0 (φ(0))φ0 (0) = f (0), f 0 (0)φ0 (0) ,
which implies that the projectivization satisfies
[j 1 (f ◦ φ)(0)] = [f 0 (0)φ0 (0)] = [f 0 (0)] = [j 1 f (0)]; that is, the fiber P f (ζ0 ) (J 1 X) is invariant by φ ∈ Aζ0 .
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
171
Assume from here on that the map f is algebraically nondegenerate. For k ≥ 2 we have j 2 (f (φ)) = f (φ), f 0 (φ)φ0 , f 0 (φ)φ00 + f 00 (φ)(φ0 )2 . Moreover, φ ∈ A0 implies
j 2 (f ◦ φ)(0) = (f ◦ φ)(0), (f ◦ φ)0 (0), (f ◦ φ)00 (0) = f (0), sφ f 0 (0), tφ f 0 (0) + s2φ f 00 (0) ,
where sφ = φ0 (0) and tφ = φ00 (0). We are free to prescribe the complex numbers sφ and tφ . The bundle P(J 2 X) is algebraic and locally trivial, hence locally algebraically trivial (as a C ∗ bundle). In particular, we have a C ∗ isomorphism Jf2(0) X ∼ = C n ⊕ C n , where λ(z, w) = (λz, λ2 w) for (z, w) ∈ C n ⊕ C n . The Jacobian matrix of the map (sφ , tφ ) 7→ (sφ f 0 (0), tφ f 0 (0) + s2φ f 00 (0)) is given by ∂(f ◦ φ)0 (0) ∂sφ ∂(f ◦ φ)0 (0) ∂tφ
∂(f ◦ φ)00 (0) 0 ∂sφ = f (0) 0 ∂(f ◦ φ)00 (0) ∂tφ
2sφ f 00 (0) f 0 (0)
.
It is clear that the rank is 2 if f 0 (0) 6= 0 (which we may assume without loss of generality because f 0 6≡ 0 so f 0 (ζ) 6= 0 for generic ζ). Thus, as φ varies through the space A0 , j 2 (f ◦ φ)(0) sweeps out a complex 2dimensional set in the fiber Jf2(0) X over the point f (0) ∈ X, and the projectivization is a set of dimension at least 1 in P(Jf2(0) X). If f is algebraically nondegenerate, the algebraic closure of [j 2 f (C)] is of dimension n = dim X, as remarked earlier. The preceding argument shows that [ [j 2 (f ◦ φ)(C)] φ∈A
S is of dimension at least n+1. By Schwarz’s Lemma the set φ∈A [j 2 (f ◦ φ)(C)] is contained in the base locus B2 (L) thus dim B2 (L) ≥ n + 1. Since dim P(J 2 X) = n(2 + 1) − 1, the codimension of B2 (L) in P(J 2 X) is at most 3n − 1 − (n + 1) = 2(n − 1). For k = 3 we get j 3 (f (φ)) = f (φ), f 0 (φ)φ0 , f 0 (φ)φ00 + f 00 (φ)(φ0 )2 , f 0 (φ)φ000 + 3f 00 (φ)φ0 φ00 + f 000 (φ)(φ0 )3 .
Hence, for φ ∈ A0 ,
j 3 (f ◦φ)(0) = f (0), sφ f 0 (0), tφ f 0 (0)+s2φ f 00 (0), uφ f 0 (0)+3sφ tφ f 00 (0)+s3φ f 000 (0) , where sφ = φ0 (0), tφ = φ00 (0), and uφ = φ000 (0). The Jacobian matrix of the map (sφ , tφ , uφ ) 7→ sφ f 0 (0), tφ f 0 (0) + s2φ f 00 (0), uφ f 0 (0) + 3sφ tφ f 00 (0) + s3φ f 000 (0)
172
KAREN CHANDLER AND PITMANN WONG
is ∂(f ◦ φ)0 (0)
∂sφ ∂(f ◦ φ)0 (0) ∂tφ ∂(f ◦ φ)0 (0) ∂uφ
∂(f ◦ φ)00 (0) ∂sφ ∂(f ◦ φ)00 (0) ∂tφ ∂(f ◦ φ)00 (0) ∂uφ
∂(f ◦ φ)000 (0) ∂sφ ∂(f ◦ φ)000 (0) ∂tφ 000 ∂(f ◦ φ) (0) ∂uφ 0 f (0) 2sφ f 00 (0) = 0 f 0 (0) 0
3tφ f 00 (0) + 3s2φ f 000 (0) 3tφ f 00 (0) f 0 (0)
0
.
It is clear that the rank is 3 if f 0 (0) 6= 0 (which we may assume without loss of generality). Thus, as φ varies through the space A 0 , j 3 (f ◦ φ)(0) sweeps out a complex 3dimensional set in the fiber Jf3(0) X over the point f (0) ∈ X and the projectivization is a set of dimension at least 2 in P(J f3(0) X). If f is algebraically S nondegenerate then the set φ∈A [j 3 (f ◦ φ)(C)] is of dimension at least n + 2 in P(J 3 X). By Schwarz’s Lemma this same set is contained in the base locus B 3 (L) thus dim B3 (L) ≥ n + 2. Since dim P(J 3 X) = n(3 + 1) − 1, the codimension of B3 (L) in P(J 3 X) is at most 4n − 1 − (n + 2) = 3(n − 1). The case for general k is argued in a similar fashion. Define polynomials P ij , 1 ≤ j ≤ i, by setting P1,1 = φ0 , P2,1 = φ00 , P2,2 = (φ0 )2 and, for i ≥ 3, Pi,1 = φ(i) , 0 Pi,2 = Pi−1,1 + Pi−1,2 ,... 0 Pi−1,i−1 = Pi−1,i−2 + Pi−1,i−1 ,
Pi,i = (φ0 )i . In particular, Pi,1 is the only polynomial involving φ(i) ; each Pi,j , for j ≥ 2, involves only derivatives of φ of order less than i. We get, by induction: (f ◦ φ)(i) =
i X
f (j) (φ)Pi,j = f 0 (φ)φ(i) +
j=1
i X
f (j) (φ)Pi,j .
j=2
Thus the kth jet j k (f ◦ φ) is given by (f 0 (φ)φ0 , · · · , f 0 (φ)φ(i) +
i X j=2
f (j) (φ)Pi,j , · · · , f (k) (φ)φ(k) +
k X
f (j) (φ)Pk,j ),
j=2
and we have k parameters sφ,i = φ(i) (0), i = 1, . . . , k. The Jacobian matrix of the map (with φ(0) = 0) (sφ,1 , . . . , sφ,k ) 7→ (sφ,1 f 0 (0), sφ,2 f 0 (0)+s2φ,1 f 00 (0), . . . , sφ,k f 0 (0)+
k X j=2
f (j) (0)Pk,j )
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
is given by the k × nk matrix 0 f (0) 2sφ f 00 (0) 3tφ f 00 (0) + 3s2φ f 000 (0) 0 f 0 (0) 3tφ f 00 (0) 0 0 f 0 (0) . . 0 . . . 0 0 .
... ... ... ... ... ...
173
. . . . . . . . . f 0 (0) . 0 f 0 (0)
It is clear that the rank is k if f 0 (0) 6= 0 (which we may assume without loss of generality). Thus, as φ varies through the space A 0 , j k (f ◦ φ)(0) sweeps out a complex kdimensional set in the fiber Jfk(0) X over the point f (0) ∈ X, and the projectivization is a set of dimension at least k − 1 in P(J fk(0) X). If f is S algebraically nondegenerate then the set φ∈A [j k (f ◦ φ)(C)] is of dimension at least n+k −1 in P(J k X). By Schwarz’s Lemma this same set is contained in the base locus Bk (L) thus dim Bk (L) ≥ n + k − 1. Since dim P(J k X) = n(k + 1) − 1, the codimension of Bk (L) in P(J k X) is at most (k+1)n−1−(n+k−1) = k(n−1). This completes the proof of the following Theorem: Theorem 6.4. Let X be a connected compact manifold of dimension n and let Lk be the dual of the tautological line bundle over P(J k X), k ≥ 2. Suppose that f : C → X is an algebraically nondegenerate holomorphic map. Then the irreducible component of the base locus containing [j k f ] is of codimension at most (n − 1)k; equivalently the dimension is at least n + k − 1. Corollary 6.5. Let X be a connected projective manifold of complex dimension n and let Lk be the dual of the tautological line bundle over P(J k X). If the dimension of the base locus Bk (Lk ) ≤ n + k − 2 then every holomorphic map f : C → X is algebraically degenerate.
7. Surfaces of General Type Summary. In this section we shall show that every holomorphic map f : C → X is algebraically degenerate, where X is a minimal surface of general type such that pg (X) > 0 and Pic X ∼ = Z. These conditions, together with the explicit calculations in Section 3, imply that Jk X is big (equivalently, the line bundle Lk over P(J k X) is big) for k 0. The Schwarz Lemma of the preceding section implies that the image of the lifting [j k f ] : C → P(J k X) is contained in the base locus Bk (Lk ) (see (6.3)). (Note that the dimension of P(J k X) is 2k + 1.) Moreover , if f is algebraically nondegenerate, dim Bk (Lk ) ≥ k + 1. On the other hand , we show (Theorem 7.20) that the base locus is at most of dimension k. This contradiction establishes the theorem. The result in Theorem 7.20 is obtained by a cutting procedure (each cut lowers the dimension of the base locus by one) pioneered by Lu and Yau and extended by Dethloff– Schumacher–Wong (in which the condition Pic X ∼ = Z was first introduced ).
174
KAREN CHANDLER AND PITMANN WONG
The starting point in the process is the explicit formulas obtained by Stoll and Wong in Theorem 3.9 and Corollary 3.10, namely, the index ι(Jkm X) = 2k 2 2k+1 χ(Lm + O(m2k ) (here ci = ci (X)) is very big; k ) + O(m ) = (αk c1 − βk c2 )m indeed we have, limk→∞ αk /βk = ∞. Consequently if c21 > 0, which is the case 2k+1 2 if X is minimal , then χ(Lm c1 + O(m2k ) for some positive constant k ) = cm 2 c (as, eventually, αk /βk > c2 /c1 ). If the base locus Y1 were of codimension one (which we show that there is no loss of generality in assuming that it is irreducible) then for k 0, χ(Lk Y1 ) is still big and Schwarz Lemma implies that the base locus must be of codimension 2. The computation is based on the intersection formulas obtained in Lemma 7.15 (requiring the assumption Pic X ∼ = Z) and Theorem 7.16. The cutting procedure can be repeated and , as to be expected , each time with a loss which can be explicitly estimated using the intersection formulas. These losses are compensated by taking a larger k. In the proof of Theorem 7.20 we show that, after k cuts, the Euler characteristic is bounded below by k X 1 2 c2 , µk δ k c 1 − i2 i=1 where µk is a positive integer and δk =
k k i−1 X 1 X1X1 + i2 i=2 i j=1 j i=1
It remains to show that
1 − 4
δk Pk
1 i=1 i2
k X 1
i
i=1
>
2
(k + 1) + 4(k!)2
k X 1
i
i=1
2
.
c2 c21
for k sufficiently large. A little bit of combinatorics shows that δk lim Pk
k→∞
1 i=1 i2
=∞
(compare the proof of Corollary 3.10). This completes the proof of our main result. Indeed , for a hypersurface of degree d ≥ 5 in P 3 , our colleague B . Hu checked , using Maple, that k ≥ 2283 is sufficient. This, together with a result of Xu implies that a generic hypersurface of degree d ≥ 5 in P 3 is hyperbolic. We recall first some wellknown results on manifolds of general type. The following result can be found in [Barth et al. 1984]: Theorem 7.1. Let X be a minimal surface of general type. The following Chernnumber inequalities hold : (i) (ii) (iii) (iv) (v)
c21 (T ∗ X)[X] > 0. c2 (T ∗ X)[X] > 0. c21 (T ∗ X)[X] ≤ 3c2 (T ∗ X)[X]. 5c21 (T ∗ X)[X] − c2 (T ∗ X[X]) + 36 ≥ 0 if c21 (T ∗ X)[X] is even. 5c21 (T ∗ X)[X] − c2 (T ∗ X)[X] + 30 ≥ 0 if c21 (T ∗ X)[X] is odd.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
175
Let L0 be a nef line bundle on a variety X of complex dimension n. A coherent sheaf E over X is said to be semistable (or semistable in the sense of Mumford– Takemoto) with respect to L0 if c1 (E) . cn−1 (L0 ) ≥ 0 and if, for any coherent 1 subsheaf S of E with 1 ≤ rk S < rk E, we have µS,L0 ≤ µE,L0 , where def
µS,L0 =
n−1 (L0 ) c1 (S) . cn−1 (L0 ) def c1 (E) . c1 1 [X] and µE,L0 = [X]. rk S rk E
(7.1)
It is said to be stable if the inequality is strict, that is, µ S,L0 < µE,L0 . The number µS,L0 shall be referred to as the normalized degree relative to L0 . We shall write µS for µS,L0 if L0 is the canonical bundle. If X is of general type then (see [Maruyama 1981] in the case of surfaces and [Tsuji 1987, 1988] for general dimensions): Theorem 7.2. Let X be a smooth variety of general type. Then the bundles Jm ∗ Nm ∗ T X are semistable with respect to the canonical bundle KX . T X, Recall from Section 2 that for a vector bundle E of rank r, rk
Jm
E=
(m + r − 1)! , (r − 1)! m!
c1 (
Jm
E) =
(m + r − 1)! c1 (E). r! (m − 1)!
Thus, for surfaces of general type, we have
µ m T ∗ X = 21 mc21 (T ∗ X)[X] with respect to the canonical bundle. More generally: Theorem 7.3. Let X be a surface of general type. If D is a divisor in X Ji J ik ∗ such that H 0 (X, SI ⊗ [−D]) 6= 0 where SI = ( 1 T ∗ X ⊗ · · · ⊗ T X) and I = (i1 , . . . , ik ) is a ktuple of positive integers satisfying m = i1 +2i2 +· · · +kik , then Pk j=1 ij 2 c1 (T ∗ X)[X] ≤ 21 mc21 (T ∗ X)[X], µ[D] ≤ µSI = 2 where [D] is the line bundle associated to the divisor D. The examples at the end of Section 2 show that the sheaves of kjet differentials are not semistable unless k = 1. However we do have (by Theorems 3.7 and 3.8): Theorem 7.4. Let X be a surface of general type. Then P c1 (SI ) c1 (T ∗ X) X rkSI m P = µSI ≤ c21 (T ∗ X), µJkm X = I P rkS rkS 2 I I I I I
and equality holds if and only if k = 1; moreover , asymptotically, Pk 1 i=1 i m + O(1) c21 (T ∗ X). µJkm X = 2k
176
KAREN CHANDLER AND PITMANN WONG
A coherent sheaf E is said to be Euler semistable if for any coherent subsheaf S of E with 1 ≤ rk S < rk E, we have χ(S) χ(E) ≤ (7.2) rk S rk E It is said to be Euler stable if the inequality is strict. There is a concept of semistability due to Gieseker–Maruyama (see [Okonek et al. 1980]) for coherent sheaves on P n in terms of the Euler characteristic that differs from the concept introduced here. Example 7.5. From the exact sequence J2 ∗ T X → J22 X → T ∗ X → 0, 0→
we get, via the table on page 163, J2 ∗ χ(J22 X) = χ(T ∗ X)+χ( T X) = 16 (c21 −5c2 )+ 14 (5c21 −15c2 ) =
1 2 12 (17c1 −55c2 ).
Theorem 6.1 yields c21 − 3c2 ≤ 0, which implies that χ(T ∗ X) =
Thus χ(J22 X) < χ( (7.2).
J2
c21 − 5c2 < 0. 6
T ∗ X), that is, J22 X is not semistable in the sense of
Recall that the index of each of the sheaves SI and Jkm X of a surface X is of the form ac21 (T ∗ X)+bc2 (T ∗ X). Thus the ratio γ(X) = γ(T ∗ X) = c2 (T ∗ X)/c21 (T ∗ X) is an important invariant. More generally, we define γ(S) =
c2 (S) , c21 (S)
(7.3)
provided that c21 (S) 6= 0. Let X be a smooth hypersurface in P 3 . Then c1 = c1 (T X) = −c1 (T ∗ X) = d − 4, c2 = c2 (T X) = c2 (T ∗ X) = d2 − 4d + 6. Hence the ratio of c21 (T ∗ X) and c2 (T ∗ X) is given by γd (J1 X) = γd (T ∗ X) =
4d − 10 d2 − 4d + 6 c2 (T ∗ X) = 1+ , = 2 ∗ c1 (T X) (d − 4)2 (d − 4)2
(7.4)
provided that d 6= 4. Note that γ∞ (T ∗ X) = limd→∞ γd (T ∗ X) = 1. Table A on the next page shows the first few values of γg = γd (J1 X). Recall from Theorem 5.12 that Jm ∗ 1 T X) = 12 χ(X; (m+1) (2m2 −2m+1)c21 − (2m2 +4m−1)c2 1 (m+1) (2m2 −2m+1)(d−4)2 − (2m2 +4m−1)(d2 −4d+6) = 12 = 13 (5−2d)3m3 + O(m2 ).
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
d
γd
d
5
11
12
51 32
13
14 27
14
73 50
6
9 2
7
= 4.5 3
γd
d
γd
19
97 75
∼ 1.2934
∼ 1.5175
20
163 128
∼ 1.2735
= 1.46
21
363 289
∼ 1.2561
∼ 1.4132
22
67 54
∼ 1.2408
= 1.375
23
443 361
∼ 1.2244 = 1.215
∼ 1.5937
177
8
19 8
= 2.375
15
171 121
9
51 25
= 2.04
16
11 8
10
11 6
= 1.8¯ 3
17
227 169
∼ 1.3432
24
243 200
11
83 49
∼ 1.6939
18
129 98
∼ 1.3164
25
59 49
∼ 1.2041
Table A. Values of γd (J1 X) as a function of d.
Jm ∗ T X) < 0 for all m ≥ 1 if d ≥ 3. If d ≥ 5 it is wellIt is clear that χ(X; J m ∗ known that H 0 (X, T X) = 0, whence the following nonvanishing theorem: Theorem 7.6. Let X be a smooth hypersurface of degree d ≥ 5 in P 3 . Then Jm ∗ Jm ∗ Jm ∗ T X) T X)−dim H 2 (X, T X) ≥ dim H 1 (X, dim H 1 (X,
= 61 (m+1) 2(2d−5)m2 −(3d2 −16d+28)m−(d2 −6d+11) = 31 (2d−5)3m3 +O(m2 )
for all m 0. Next we consider the case of 2jets. We have, by Riemann–Roch: χ(J2m X) =
1 1 ι(J2m X) − c1 (J2m X) · c1 + (rk J2m X)(c21 + c2 ). 2 12
(Here c1 = c1 (T ∗ X), c2 = c2 (T ∗ X) and, using the formulas for c1 (J2m X), rk J2m X and ι(J2m X) in Theorem 3.3 we get: χ(J2m X) =
1 27 32 5
(pm c21 − qm c2 )
with pm = qm =
21m5 + 180m4 + 410m3 + 180m2 + 49m + 120, if m is odd, 21m5 + 180m4 + 420m3 + 180m2 − 56m + 480, if m is even; 15m5 + 225m4 + 1150m3 + 2250m2 + 1235m − 75, if m is odd, 15m5 + 225m4 + 1180m3 + 2520m2 + 1640m − 480, if m is even.
The index χ(J2m X) is positive if and only if pm /qm > c2 /c21 , and taking the limit as m → ∞ yields the inequality c2 /c21 ≤ 75 . For a smooth hypersurface of degree d in P 3 the ratio c2 /c21 = 1 + (4d − 10)/(d − 4)2 and we arrive at the inequality 4d − 10 2 ≤ , (d − 4)2 5
178
KAREN CHANDLER AND PITMANN WONG
which is equivalent to the inequality 0 ≤ d2 −18d+41 = (d−9)2 −40. We deduce: Theorem 7.7. Let X be a smooth hypersurface in P 3 . Then χ(J2m X) is big if and only if d = deg X ≥ 16. We use the terminology that the Euler characteristic is big if and only if there is a constant c > 0 such that χ(J2m X) ≥ cm5 + O(m4 ) for all m 0. In order to lower the degree in the preceding theorem we must use jet differentials of higher order. We see from Table A on page 177 that the ratio c2 /c21 of a hypersurface of degree d ≥ 5 in P 3 is bounded above by 11. By Theorem 3.7, ι(Jkm X) = (αk c21 − βk c2 )m2k+1 + O(m2k ) thus the index is positive if and only if αk c2 > 2. βk c1 In the table on page 148 we see that the ratio αk /βk crosses the threshold 11 as k increases from 198 to 199. Putting this together with Theorem 5.13, we get: Theorem 7.8. Let X be a generic smooth hypersurface of degree d ≥ 5 in P 3 . For each k ≥ 199, χ(Jkm X) ≥ cm5 + O(m4 ) for all m k. For a minimal surface of general type, Theorem 7.1 implies that ( 5 + 36c−2 if c21 is even, c2 (X) 1 1 ≤ 41 ≤ γ(X) = 2 ≤ 3 c1 (X) if c21 is odd. 5 + 30c−2 1 ≤ 34 The ratio αk /βk was shown to tend to ∞ as k → ∞. Thus Theorem 7.8 extends to any minimal surface of general type: Theorem 7.9. Let X be a smooth minimal surface of general type. Then χ(Jkm X) ≥ cm5 + O(m4 )
for all m k 0.
In [Green and Griffiths 1980] we find the following result: Theorem 7.10. Let X be a smooth surface of general type. If i1 +· · ·+ik is even J i1 J ik then a nontrivial section of the bundle TX ⊗···⊗ T X ⊗ K(i1 +···+ik )/2 is nonvanishing. Using this, Green and Griffiths deduced the following vanishing Theorem. We include their argument here, with minor modifications. Theorem 7.11. Let X be a smooth surface of general type. Assume that the canonical bundle KX admits a nontrivial section. Then H 2 (X, Jkm X) = 0 for all k ≥ 1 and m > 2k.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
179
Proof. Let σ be a nontrivial section of KX , so that we have an exact sequence: ⊗σ
0 → SI ⊗ K(/i1 +···+ik )2−1 → SI ⊗ K(/i1 +···+ik )2 → SI ⊗ K(/i1 +···+ik )2 D → 0 J i1 Ji where D = [σ = 0], SI = T X ⊗· · ·⊗ k T X and i1 +· · ·+ik is even. Hence, ⊗σ
0 → H 0 (X, SI ⊗ K(/i1 +···+ik )2−1 ) → H 0 (X, SI ⊗ K(/i1 +···+ik )2 )
is exact. By Theorem 7.10 the image of the map ⊗σ is 0; hence H 0 (X, SI ⊗ K(/i1 +···+ik )2−1 ) = 0. The argument applies also to the exact sequence: ⊗σ
0 → SI ⊗ K(/i1 +···+ik )2−l → SI ⊗ K(/i1 +···+ik )2=l+1 → SI ⊗ K(/i1 +···+ik )2 D → 0 for any l ≥ 1 and we conclude via induction that H 0 (X, SI ⊗ Kq ) = 0 for all q < (i1 + · · · + ik )/2. If i1 + · · · + ik is odd then taking ik+1 = 1 we have H 0 (X, Si1 ,...,ik ,ik+1 ⊗ Kq+1 ) = 0 provided that q + 1 < (i1 + · · · + ik + 1)/2 (equivalently q < (i1 + · · · + ik − 1)/2). Suppose that H 0 (X, SI ⊗ Kq ) 6= 0. Then there exists a nontrivial section ρ of H 0 (X, SI ⊗ Kq ) and we obtain a nontrivial section ρ ⊗ σ of Si1 ,...,ik ,ik+1 ⊗ Kq+1 . This shows that: 0, for all q < 21 (i1 + · · · + ik − 1) if i1 + · · · + ik is odd, H 0 (X, SI ⊗ Kq ) = 0 for all q < 12 (i1 + · · · + ik ) if i1 + · · · + ik is even. By Serre duality,
0, for all q < 21 (i1 + · · · + ik − 1) if i1 + · · · + ik is odd, 0 for all q < 12 (i1 + · · · + ik ) if i1 + · · · + ik is even, J i1 ∗ J ik ∗ where SI = T X ⊗···⊗ T X. If I = i1 + · · · + ik ≥ 3 then we may take q = 1 in the formulas above. Thus we have: H 2 (X, SI ) = 0, if I ≥ 3. Note Pk Jkm X admits a composition series by SI satisfying the condition j=1 jij = m. Thus H 2 (X, Jkm X) = 0 if each of these SI satisfies the condition I ≥ 3. If k = 2 we have: H 2 (X, SI ⊗ K1−q ) =
i1 + 2i2 = m ⇐⇒ i2 = (m − i1 )/2 ⇐⇒ i1 + i2 = (m + i1 )/2. Thus i1 + i2 ≥ 3 if and only if m ≥ 6 − i1 . Since i1 ≥ 0 we conclude that m ≥ 6 implies i1 + i2 ≥ 3. If k = 3 then i1 +2i2 +3i3 = m ⇐⇒ (i1 +i3 ) + 2(i2 +i3 ) = m ⇐⇒ i2 +i3 = 12 (m−i1 −i3 ) ⇐⇒ i1 +i2 +i3 = 12 (m+i1 −i3 ).
180
KAREN CHANDLER AND PITMANN WONG
Thus i1 + i2 + i3 ≥ 3 if and only if m ≥ 6 − i1 + i3 ≥ 6 + i3 . Since i3 is at most [m/3] we conclude that i1 + i2 + i3 ≥ 3 if m ≥ 9. The case of general k can be established by an induction argument. ˜ For our purpose only the following weaker result is needed: Theorem 7.12. Let E be a holomorphic vector bundle of rank r ≥ 2 over a nonsingular projective surface X. Assume that (i) KX is nef and not the trivial bundle; (ii) Pic X ∼ = Z; ∗ (iii) det E is nef ; (iv) there exists a positive integer s with the property that there is a nontrivial global regular section ρ of (KX ⊗ det E ∗ )s such that the zero divisor [ρ = 0] is smooth. Jm ∗ E ) = 0 for all i ≥ 2 and for m sufficiently large. Then H i (X,
The canonical bundle KX of a minimal surface X of general type is nef. If J i1 ∗ J ik ∗ Pic X ∼ T X ⊗···⊗ T X is ample = Z then KX is ample, so KX ⊗ det for any nonnegative integers i1 , . . . , ik . Hence: Corollary 7.13. Let X be a nonsingular minimal surface of general type. Assume that Pic X ∼ = Z and pg (X) > 0. Let I = (i1 , . . . , ik ) be a ktuple of J i1 ∗ Ji T X ⊗· · ·⊗ k T ∗ X = 0 if i1 +· · ·+ik nonnegative integers. Then H 2 X, is sufficiently large; consequently, H 2 (X, Jkm X) = 0 if m k. Corollary 7.14. Let X be a nonsingular minimal surface of general type with Pic X ∼ = Z. Then h0 (X, Jkk!m X) ≥ cm2k+1 + O(m2k ) for some positive constant c; that is, Jkk! X is big. A good source for the general theory of vanishing theorems is [Esnault and Viehweg 1992]. Next we deal with the question of algebraic degeneration of holomorphic maps and hyperbolicity of surfaces of general type. The condition that J kk! X is big implies that Jkk! X ⊗ [−D] is big for any ample divisor D on X. We may write D = a0 D0 for a0 > 0, with D0 as the positive generator of Pic X. The Schwarz Lemma for jet differentials implies that the image of [j k f ] is contained in the zero set of all kjet differentials vanishing along an ample divisor. Thus we may assume that [j k f ](C) is contained in an effective irreducible divisor in P(J k X) and is the zero set of a section (7.5) σ ∈ H 0 P(J k X), Lkk!mk ⊗ p∗ [−νk D0 ] , ∼ ZhLk! i ⊕ Pic X). where we abbreviate Lk = LP(J k X) (note that Pic P(J k X) = k Our aim is to show that the restriction Lk! k [σ=0] is big. First we need a lemma:
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
181
Lemma 7.15. Let X be a nonsingular minimal surface of general type with Pic X ∼ = Z. Suppose that H 0 (Jkk!m X ⊗ [−D]) 6= 0 for some divisor D in X. Then for all m 0, c1 ([D]) ≤
Bk k! mc1 (T ∗ X) + O(1) Ak
where Ak and Bk are the constants defined in Theorems 3.7 and 3.8. Proof. The assumption that Pic X ∼ = Z implies that c1 (Jkk! X ⊗ [−D]) = qc1 , ∗ where c1 = c1 (T X) and q ∈ Q. Let σ be a nontrivial section of Jkk! X ⊗ [−D]. The Poincar´e–Lelong formula implies that Z Z Z 0= ddc log σ2 ∧ c1 ≥ c1 − c1 (Jkk!m X ⊗ [−D]) ∧ c1 X [σ=0] X Z Z = c1 − q c21 , [σ=0]
X
implying q > 0. On the other hand, the usual formula for Chern classes yields 0 < c1 (Jkk!m X ⊗ [−D]) = c1 (Jkk!m X) − (rkJkk!m X)c1 ([D]). By the asymptotic formula in Section 3, we have c1 (Jkk!m X) = Bk (k!m)2k c1 + O((k!m)2k−1 ); hence the preceding inequality may be written as Ak c1 ([D])(k!m)2k−1 = (rkJkk!m X)c1 ([D]) < Bk (k!m)2k c1 + O((k!m)2k−1 ), where Ak and Bk are the constants defined in Theorems 3.7 and 3.8. Thus we get the estimate Bk k! mc1 + O(1). ˜ c1 ([D]) ≤ Ak Theorem 7.16. Let X be a smooth surface and Lk be the “hyperplane line sheaf ” over P(J k X). Then k! 2k+1 1 (Lk! p∗ c2k+1 (αk c21 − βk c2 ) k ) = (2k + 1)!χ(Lk ) = 2 (2k + 1)!(k!) 1
= (k!) k! ∗ p∗ c2k 1 (Lk )p c1
=
2k−1
i−1 k k X 1 X1X1 2 + c − i2 i=2 i j=1 j 1 i=1
2k 2 1 2 (k!) (2k!)Bk c1
=
k X 1 c2 , 2 i i=1
k (k!)2k−2 X 1 2 c , 2 i 1 i=1
2k−1 ∗ 2 (2k − 1)!Ak c21 = (k!)2k−3 c21 . p∗ c12k−1 (Lk! k )p c1 = (k!)
Proof. Let E be a coherent sheaf of rank r and L be a line bundle. Then ck (E ⊗ L) =
k X i=0
(r − i)! ci (E)c1 (L)k−i . (k − i)!(r − k)!
182
KAREN CHANDLER AND PITMANN WONG
For a surface we have only two Chern classes, c1 (E ⊗ L) = rc1 (L) + c1 (E) and c2 (E ⊗ L) = 12 r(r − 1)c21 (L) + (r − 1)c1 (E)c1 (L) + c2 (E). From this we get ι(E ⊗ L) = r 2 c21 (L) + 2rc1 (L)c1 (E) + c21 (E) − r(r − 1)c21 (L) − 2(r − 1)c1 (L)c1 (E) − 2c2 (E) = =
c21 (E) − 2c2 (E) + rc21 (L) + 2c1 (L)c1 (E) ι(E) + rc21 (L) + 2c1 (L)c1 (E)
and the Euler characteristic (with ci = ci (T ∗ X)): 1 χ(E ⊗ L) = 21 ι(E ⊗ L) − c1 (E ⊗ L)c1 + rk(E ⊗ L)(c21 + c2 ) 12 1 = 12 ι(E) + rc21 (L) + 2c1 (L)c1 (E) − (rc1 (L) + c1 (E))c1 + 12 r(c21 + c2 ) = χ(E) + 21 rc21 (L) + 2c1 (L)c1 (E) − rc1 (L) c1 .
For the sheaf of jet differentials we have the asymptotic expansions c1 (Jkm X) = Bk m2k c1 + O(m2k−1 ), rk Jkm X = Ak m2k−1 + O(m2k−2 )
χ(Jkm X) = χ(Lm ) = 12 (αk c21 − βk c2 )m2k+1 + O(m2k )0 ; hence c1 (Jkk!m X) = (k!)2k Bk m2k c1 + O(m2k−1 ), rk Jkk!m X = (k!)2k−1 Ak m2k−1 + O(m2k−2 ) χ(Jkk!m X) = χ(Lk!m ) = 21 (k!)2k+1 (αk c21 − βk c2 )m2k+1 + O(m2k ). We get from these the asymptotic expansion for χ(J kk!m X ⊗ Lm ): χ(Jkk!m X ⊗ Lm ) = χ(Jkk!m X) + 12 m m(rk Jkk!m X)c21 (L) + c1 (L)c1 (Jkk!m X) − (rk Jkk!m X)c1 (L)c1 =
1 2
(k!)2k+1 (αk c21 − βk c2 )
If c1 (L) = λc1 then
+ (k!)2k−1 Ak c21 (L) + (k!)2k Bk c1 (L)c1 m2k+1 + O(m2k ).
∗ m χ((Lk! k ⊗ p L) )
= χ((Jkk!m X ⊗ Lm ) =
1 2
(k!)2k+1 (αk c21 − βk c2 ) + (λ2 (k!)2k−1 Ak + λ(k!)2k Bk )c21 m2k+1 + O(m2k )
= χ(Jkk!m X) + 21 λ2 (k!)2k−1 Ak c21 m2k+1 + 21 λ(k!)2k Bk c21 m2k+1 + O(m2k ). Since ci1 (p∗ L) = 0 for all i ≥ 3, we have ∗ (Lk! c2k+1 k ⊗ p L) 1 ∗ 2k+1 = (c1 (Lk! k ) + c1 (p L))
2k−1 2k k! ∗ 2 ∗ = c2k+1 (Lk! (Lk! k ) + (2k + 1)c1 (Lk )c1 (p L) + k(2k + 1)c1 k )c1 (p L), 1
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
183
and we get, up to O(m2k ), ∗ m χ((Lk! k ⊗ p L) )
=
∗ c2k+1 (Lk! 1 k ⊗ p L) 2k+1 m (2k + 1)!
=
2k−1 2k k! ∗ 2 ∗ c2k+1 (Lk! (Lk! 1 k ) + (2k + 1)c1 (Lk )c1 (p L) + k(2k + 1)c1 k )c1 (p L) 2k+1 m (2k + 1)!
=
∗ 2 ∗ c2k (Lk! 1 c2k−1 (Lk! c2k+1 (Lk! 1 k )c1 (p L) 2k+1 k )c1 (p L) 2k+1 k ) 2k+1 m + 1 m + 1 m (2k + 1)! (2k)! 2 (2k − 1)!
=
∗ ∗ 2 c2k+1 (Lk! c2k (Lk! 1 c2k−1 (Lk! 1 k ) 2k+1 k )p c1 2k+1 k )p c1 2k+1 m +λ 1 m + λ2 1 m . (2k + 1)! (2k)! 2 (2k − 1)!
∗ m Comparing the two expressions for χ((Lk! k ⊗ p L) ) we deduce that k! p∗ c2k+1 (Lk! k ) = (2k + 1)!χ(Lk ), 1 k! ∗ 2k 2 1 p∗ c2k 1 (Lk )p c1 = 2 (k!) (2k!)Bk c1 , ∗ 2 2k−1 p∗ c12k−1 (Lk! (2k − 1)!Ak c21 . k )p c1 = (k!)
The theorem follows from these by substituting the asymptotic expansions for χ(Lk! ˜ k ), Ak and Bk into the expressions above. As a means toward understanding the general case we treat the special case of 2jets and 3jets (for the case of P(T X), that is, 1jets, see [Miyaoka 1977; Lu and Yau 1990; Lu 1991; Dethloff et al. 1995b]). For 2jets the intersection formulas in Lemma 7.15 and Theorem 7.16 read as: c1 ([D]) ≤ 34 mc1 , p∗ c51 (L22 ) = 14c21 − 10c2 , p∗ c41 (L22 )p∗ c1 = 3c21 ,
(7.6)
p∗ c31 (L22 )p∗ c21 = 2c21 . We shall use these formulas to deal with holomorphic maps from the complex plane into a minimal surface X of general type satisfying the conditions that Pic X ∼ = Z and KX is effective and nontrivial (for example X is a hypersurface 3 in P of degree d ≥ 5). The condition was first introduced in [Dethloff et al. 1995b] and is crucial in the rest of this article. We shall use the following terminology. An irreducible subvariety Y in P(J k X) is said to be horizontal if p(Y ) = X, where p : P(J k X) → X is the projection; otherwise it is said to be vertical. A variety is said to be horizontal (resp. vertical) if every irreducible component is horizontal (resp. vertical). A subvariety Y may be decomposed as Y = Y hor + Y ver , where Y hor and Y ver consist respectively of the horizontal and vertical components. Note that Y ver = (p−1 C) ∩ Y , where C is a subvariety of X; indeed C = p(Y ver ). We shall need a lemma:
184
KAREN CHANDLER AND PITMANN WONG
Lemma 7.17. Let X be a surface such that pg (X) > 0 and Pic X ∼ = Z with ample generator [D0 ]. There exist positive integers m and a and a nontrivial section σ ∈ ∗ hor H 0 P(J k X), Lk!m ⊗ p [−aD ] such that [σ = 0] is reduced and irreducible, 0 k that is, there exists exactly one horizontal component with multiplicity 1. For the proof of the case k = 1, see [Dethloff et al. 1995b, Lemmas 3.5 and 3.6]. The proof depends only on the assumption Pic X ∼ = Z, which implies that 1 ∼ Pic P(J X) = Z ⊕ Z. This is of course also valid for Pic P(J k X) for any k. Indeed the proof (with J 1 X replaced by J k X) is word for word the same. Theorem 7.18. Let X be a minimal surface of general type with effective ample canonical bundle such that Pic X ∼ = Z, pg (X) > 0, and 17c21 (T ∗ X) − 16c2 (T ∗ X) > 0. (This is satisfied if X is a hypersurface of degree d ≥ 70.) Then every holomorphic map f : C → X is algebraically degenerate. Proof. We start with the weaker assumption 7c21 (T ∗ X) − 5c2 (T ∗ X) > 0. (By Theorem 7.7, this is satisfied for smooth hypersurfaces in P 3 if and only if deg X ≥ 16.) Under this assumption the sheaf J22 X is big. This implies that, ∗ for any ample divisor D in X there is a section 0 6≡ σ1 ∈ H 0 (L2m 2 ⊗ p [−aK]) provided that m 0 where a > 0 and K is the canonical divisor. By Schwarz Lemma (Corollary 6.3) the image of [j 2 f ] (as f is algebraically nondegenerate) is contained in the horizontal component of [σ1 =0]. By Lemma 7.17 we may assume that the horizontal component of [σ1 = 0] is irreducible. The vertical component of [σ1 = 0] must be of the form p∗ (bK) for some b ≥ 0 which admits a section sb . 1 ⊗ p∗ [−(a − b)K]), Y1 = [σ1 ⊗ s−b = 0] is Replacing σ1 with σ1 ⊗ s−b ∈ H 0 (L2m 2 horizontal, irreducible and contains the image of [j 2 f ]. Since dim P(J 2 X) = 5 the dimension of Y1 is 4. As remarked earlier we may assume that a1 = a−b ≥ 0. We get from the first and third intersection formulas of (7.6): 1 c41 (L22 Y1 ) = c41 (L22 ) . (c1 (L2m ) − a1 p∗ c1 ) ≥ m1 c51 (L22 ) − a1 c41 (L22 ) . p∗ c1 2 m1 ≥ m1 (14c21 − 10c2 ) − 49 c21 = 2 (47c21 − 40c2 ) > 0. 2
(For a hypersurface of degree d in P n we have c21 = (d − 4)2 , c2 = d2 − 4d + 6. Thus, for d = 16, 47c21 = 6768 and 40c2 = 7920, so χ(L22 Y ) < 0; however, 47c21 − 40c2 = 47(d2 − 8d + 16) − 40(d2 − 4d + 6) = 7d(d − 30) − 2(3d − 256) is positive if and only if d ≥ 40.) We claim that L22 Y1 is big. It suffices to show 1 ⊗ [−Y1 ]) = 0 for m 0. To see this consider the exact sequence that H 2 (L2m 2 ⊗σ
1 1 1 0 → L2m ⊗ [−Y1 ] → L2m → L2m  Y1 → 0 2 2 2
and the induced exact sequence ⊗σ
1 1 1 Y1 ) → 0. ) → H 2 (L2m ⊗ [−Y1 ]) → H 2 (L2m · · · → H 2 (L2m 2 2 2
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
185
1 The vanishing of H 2 (L2m Y1 ) for m1 0 follows from the vanishing of H 2 (L2m 2 ). 2 By Schwarz’s Lemma, the image of [j k f ] is contained in the zero set of any 1 nontrivial section σ2 ∈ H 0 (Y1 , L2m Y1 ⊗ p∗ [−a2 K]), a2 > 0 and m2 0. Since 2 Y1 is irreducible Y2 = [σ2 =0] ∩ Y1 is of codimension 2 (so dim Y2 = 3) in P(J 2 X) 2 where σ2 ∈ H 0 (L2m Y1 ⊗ [−a2 D]). By Schwarz’s Lemma the reparametrized 2 k kjets {[j (f ◦ φ)]} is contained in Y2 . Pn Y , where We may assume that Y2 is irreducible. Otherwise Y2 = i=1 Nn2,i n ≥ 2 and each Y2,i , is irreducible and hence effective. We have i=1 [Y2,i ] = k!m2 ∗ [Y2 ] = Lk ⊗ p [−a2 K]Y1 (we use the notation [Z] to denote the line bundle associated to a divisor Z). The image [j k f ](P n ) is contained in Y2,i0 for some 1 ≤ i0 ≤ n. Let si be the (regular) section such that [si = 0] = Y2,i (an effective divisor in Y1 ); then we have an exact sequence
ρi
0 → [Y2,i0 ] →0 Lkk!m2 ⊗ p∗ [−a2 K]Y1 → Lkk!m2 ⊗ p∗ [−a2 K]Y2,i0 → 0. In particular, we have an injection ρi
0 → [Y2,i0 ] →0 Lkk!m2 ⊗ p∗ [−a2 K]Y1 , Nn where the map ρi0 is defined by multiplication with the section i=1,i6=i0 si . In other words we may consider each [Y2,i0 ] as a subsheaf of Lkk!m2 ⊗ p∗ [−a2 K]Y1 hence a section of [Y2,i0 ] is identified also as a section of Lkk!m2 ⊗ p∗ [−a2 K]Y1 . The Schwarz Lemma applies and we conclude that s i0 ([j k f ]) ≡ 0 for each i. Thus we may assume that Y2 is irreducible by replacing Y2 with Y2,i0 . We now repeat the previous calculation for Y1 to Y2 using again the intersection formulas listed above; we get 1 2 c31 (L22 Y2 ) = c31 (L22 ) . (c1 (L2m ) − a1 p∗ c1 ) . (c1 (L2m ) − a 2 p ∗ c1 ) 2 2
≥ m1 m2 c51 (L22 ) − (a1 m2 + m1 a2 )c41 (L22 ) . p∗ c1 + a1 a2 c31 (L22 ) . p∗ c21 = m1 m2 c51 (L22 ) − (l1 + l2 )c41 (L22 ) . p∗ c1 + l1 l2 c31 (L22 ) . p∗ c21 = m1 m2 (14c21 − 10c2 ) − 3(l1 + l2 )c21 + 2l1 l2 c21 ,
where 0 ≤ li = ai /mi ≤ 43 , for i = 1, 2. Elementary calculus shows that the function 14−3(l1 +l2 )+2l1 l2 achieves its minimum value 14−3( 43 + 43 )+2( 34 )2 = 85 8 at l1 = l2 = 43 ; thus we get 2 c31 (L22 Y2 ) ≥ m1 m2 ( 85 8 c1 − 10c2 ) =
5m1 m2 (17c21 − 16c2 ) > 0. 23
This shows that L22 Y2 is big and the image of [j 2 f ] is contained in Y3 = Y2 ∩ [σ3 = 0] (where the intersection is taken over all global sections σ 3 of L2m 2 Y2 vanishing on an ample divisor), which is of dimension 2. By Corollary 6.5 the dimension of the base locus is at least 3 if f is algebraically nondegenerate. Thus f must
186
KAREN CHANDLER AND PITMANN WONG
be algebraically degenerate (and if X contains no rational or elliptic curve then X is hyperbolic). ˜ Note that the intersection procedure was applied twice. For a smooth hypersurface X in P 3 , condition (iii) is satisfied if and only degree of X ≥ 70 (this is easily checked from the formulas c21 = (d − 4)2 , c2 = d2 − 4d + 6). This can be improved if we use 3jets. For 3jets the intersection formulas of Lemma 7.15 and Theorem 7.16 are given explicitly as follows: 11m c1 , 3! 17 7 7!(3!)7 2 3 2 c c − p∗ c71 (L3! ) = 2 = (3!) (85c1 − 49c2 ), 3 2 27 36 7 1 27 36 5 (3!)3 11 2 ∗ p∗ c61 (L3! c1 , 3 )p c1 = 2 3 2 ∗ 2 p∗ c51 (L3! 3 )p c1 = (3!) c1 . c1 ([D]) ≤
Theorem 7.19. Let X be a minimal surface with Pic X ∼ = Z, pg (X) > 0, and 389c21 (T ∗ X) − 294c2 (T ∗ X) > 0.
(∗)
Then every holomorphic map f : C → X is algebraically degenerate. Proof. The sheaf J3 X is big if and only if degree d ≥ 11. As in the case of 2jets we know that the image of [j 3 f ] is contained in Y1 = [σ1 = 0] for some 1 ⊗ p∗ [−a1 K]). Since dim P(J 3 X) = 7 the dimension of Y1 is 6. σ1 ∈ H 0 (L2m 2 From the intersection formulas listed above we get 2m1 ∗ 3! 6 ) − a1 p∗ c1 ) ≥ m1 c71 (L3! c61 (L22 Y1 ) = c61 (L3! 2 ) − a1 c1 (L2 ) . p c1 2 ) . (c1 (L2 1 1 ≥ (3!)3 m1 (85c21 − 49c2 ) − 12 112 c21 = (3!)3 12 m1 (899c21 − 588c2 ) > 0. For a smooth hypersurface in P 3 , 899c21 − 588c2 > 0 if and only if d ≥ 13. Continuing as in the case of 2jets, we see that the image of [j 3 f ] is contained (3!)m1 Y1 ⊗ p∗ [−a2 K] , in the zero set of any nontrivial section σ2 ∈ H 0 Y1 , L2 a2 > 0 and m2 0. The dimension of Y2 = [σ2 = 0] ∩ Y1 is 5. By Schwarz’s Lemma the reparametrized 3jet {[j 3 (f ◦ φ)]} is contained in Y2 . As in the case of 2jets we may assume that Y2 is irreducible. We now repeat the previous calculation using the intersection formulas above: (3!)m1
5 3! c51 (L3! 3 Y2 ) = c1 (L3 ) . (c1 (L3
(3!)m2
) − a1 p∗ c1 ) . (c1 (L3
) − a 2 p ∗ c1 )
6 3! ∗ 5 3! ∗ 2 ≥ m1 m2 c71 (L3! 3 ) − (a1 m2 + m1 a2 )c1 (L3 ) . p c1 + a1 a2 c1 (L3 ) . p c1 ∗ 2 3! ∗ 5 3! 6 = m1 m2 c71 (L3! 3 ) − (l1 + l2 )c1 (L3 ) . p c1 + l1 l2 c1 (L3 ) . p c1 2 2 = (3!)3 m1 m2 (85c21 − 49c2 ) − 11 2 (l1 + l2 )c1 + l1 l2 c1
where 0 ≤ li = ai /mi ≤ 11 6 for i = 1, 2. Elementary calculus shows that the 11 function 85 − 2 (l1 + l2 ) + l1 l2 achieves its minimum value at l1 = l2 = 11 6 ; thus
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
187
we have 3 c51 (L3! 3 Y2 ) ≥ (3!) m1 m2
2455 2 36 c1
− 49c2 > 0.
For a smooth hypersurface in P 3 this occurs if and only if d ≥ 18. Now the image of [j 3 f ] is contained in Y3 = [σ3 = 0] ∩ Y2 and has dimension 4. Moreover, an argument identical to the case of Y2 shows that we may assume Y3 irreducible. Continuing with the procedure we get 4 3! c41 (L3! 3 Y3 ) = c1 (L3 )
3 Y
i=1
(3!)mi
c1 (L3
) − a i p ∗ c1 .
Expanding the righthand side above yields (note that p ∗ c31 ≡ 0 because the dimension of the base space is 2 hence c31 = c31 (X) ≡ 0) 6 3! ∗ m1 m2 m3 c71 (L3! 3 ) − (a1 m2 m3 + m1 a2 m3 + m1 m2 a3 )c1 (L3 ) . p c1 ∗ 2 + (a1 a2 m3 + a1 m2 a3 + m1 a2 a3 )c51 (L3! 3 ) . p c1 ,
so we have 7 3! 6 3! ∗ c41 (L3! 3 Y3 ) ≥ m1 m2 m3 c1 (L3 ) − (a1 m2 m3 + m1 a2 m3 + m1 m2 a3 )c1 (L3 ) . p c1 ∗ 2 + (a1 a2 m3 + a1 m2 a3 + m1 a2 a3 )c51 (L3! 3 ) . p c1 6 3! ∗ = m1 m2 m3 c71 (L3! 3 ) − (l1 + l2 + l3 )c1 (L3 ) . p c1 ∗ 2 + (l1 l2 + l2 l3 + l3 l1 )c51 (L3! 3 ) . p c1 2 = (3!)3 m1 m2 m3 (85c21 − 49c2 ) − 11 2 (l1 + l2 + l3 )c1
+ (l1 l2 + l2 l3 + l3 l1 )c21 ,
where 0 ≤ li = ai /mi ≤ 11 6 for i = 1, 2, 3. Elementary calculus shows that the 11 function 85 − 2 (l1 + l2 + l3 ) + (l1 l2 + l2 l3 + l3 l1 ) achieves its minimum value at l1 = l2 = l3 = 11 6 ; thus we get c41 (L3! 3  Y3 ) ≥
(3!)3 m1 m2 m3 (389c21 − 294c2 ) > 0. 6
For hypersurfaces in P 3 this happens if and only if d ≥ 20. Thus the image of [j 3 f ] is contained in a subvariety Y4 = Y3 ∩ [σ4 = 0] which is of dimension 3. By Corollary 6.5 the map f must be algebraically degenerate. ˜ Note that the intersection procedure was applied three times. In order to remove condition (∗) in Theorem 7.19 we must use very high order jets, and if we use kjets then it is necessary to carry out the intersection procedure k times. The preceding proof underscores the importance of the explicit formulas obtained in Section 3. Theorem 7.20. Let X be a smooth minimal surface of general type with pg (X) > 0 and Pic X ∼ = Z. Then every holomorphic map f : C → X is algebraically degenerate. If , in addition, the surface X contains no rational nor elliptic curve then X is hyperbolic.
188
KAREN CHANDLER AND PITMANN WONG
Proof. As remarked earlier, we have to work with the kjet bundles for k sufficiently large. In the case of 2jets the cutting procedure was applied twice and for 3jets, 3 times. Now we have to do this ktimes, each time making sure (by using the explicit formulas of section 3) that the bundle is still big. The assumption implies that Lk! k is big for k 0 hence there exists m1 k and a1 > 0 such that h0 (P(J k X), Lkk!m1 ⊗ p∗ [−a1 K]) > 0 where K is the canonical divisor. As in the proof of Theorem 7.18 (and Theorem 7.19) we may, by Lemma 7.17, assume that there exists σ1 ∈ H 0 (P(J k X), Lk!m ⊗ p∗ [−a1 K]) k such that Y1 = [σ1 = 0] is horizontal and irreducible. This implies that codim Y1 = 1 (equivalently, dim Yi = dim P(J k X) − 1 = 2k + 1 − 1 = 2k). By the Schwarz Lemma of the preceding section, we conclude that the image of [j k f ] is contained in Y1 . The proof of Theorem 7.19 shows that LP(J k X) Y1 is still big and so there exists σ2 ∈ H 0 (Y1 , Lkk!m2 ⊗ p∗ [−a2 K]), m2 , a2 > 0 and (because Y1 is irreducible) that Y2 = [σ2 = 0] is of codimension 2 in P(J k X). Schwarz’s Lemma implies that the image of [j k f ] is contained in Y2 . As was shown earlier, we may assume that Y2 is irreducible. A calculation similar to that of Theorem 7.19 shows that Lk! k Y2 is still big (see the calculation below). The process can be continued k times, resulting in a sequence of reduced and irreducible horizontal subvarieties, P(J k X) = Y0 ⊃ Y1 ⊃ Y2 ⊃ · · · ⊃ Yk ⊃ [j k f ](C), where codim Yi = i (equivalently, dim Yi = 2k + 1 − i as dim P(J k X) = 2k + 1), and each of the subvarieties is the zero set of a section σ i : Yi = [σi = 0], σi ∈ H 0 Yi−1 , Lkk!mi ⊗ p∗ [−ai K] for 1 ≤ i ≤ k.
We claim that Lk! k Yi is big for 1 ≤ i ≤ k, by a calculation (to be carried out below ) analogous to that in Theorems 7.19 and 7.20.
Assuming this for the moment, we see that there exists a nontrivial section σk+1 ∈ H 0 Yk , Lkk!mk ⊗ p∗ [−ak+1 K] and [j k f ](C) is contained in an irreducible component of [σk+1 = 0] ∩ Yk . Since Yk is irreducible this component, denoted Yk+1 , is of codimension k + 1 (equivalently, dim Yk+1 = 2k + 1 − (k + 1) = k). This however contradicts Corollary 6.5 that the component containing all the reparametrization [j k (f ◦φ)](C) must be of codimension at most k (equivalently, dimension at least k + 1) if f is algebraically nondegenerate. Thus the map f must be algebraically degenerate. Since the image of an algebraically degenerate map must be contained in a rational or an elliptic curve in X, we conclude readily that X is hyperbolic if it contains no rational nor elliptic curve. It remains to verify the claim by carrying out the computations for kjets — more precisely, computations for the Chern numbers c 12k+1−λ (Lk! k Yλ ), for 1≤λ≤k — using Theorem 7.16 the intersection formulas obtained in Lemma 7.15: c12k+1−λ (Lkk! Yλ )
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
= c12k+1−λ (Lk! k )
λ Y
mi c1 (Lk! k ) − a i c1
i=1
=
λ Y
mi c2k+1 (Lk! k )− 1
i=1
λ X
Y
ai
= (k!)2k−3
λ Y
i=1
mi
(k!)2
k! mj c2k 1 (Lk X)c1
1≤j6=i≤λ
i=1
+
k X i=1
−
X
1≤i<j≤l k X
λ X
li
i=1
Y
ai aj
mq c2k−1 (Lk! k ) 1
1≤q6=i,j≤λ
i−1 X
1 1 + i2 i=2 i
189
1 2 c − (k!)2 j 1
j=1
k
k! X 1 2 c + 2 i=1 i 1
c21
k X 1 c2 2 i i=1
X
!
li lj c21
1≤i<j≤λ
for 1 ≤ lj = aj /mj ≤ (Bk /Ak )k! and 1 ≤ λ ≤ k. The coefficient of c21 is Dk,λ = (k!)
2
k i−1 k X 1 X1X1 + i2 i=2 i j=1 j i=1
−
λ X
li
i=1
k
k! X 1 2 i=1 i
+
X
li lj .
1≤i<j≤λ
The minimum occurs at k Bk (k −1)! X 1 lj = k! = Ak 2 i i=1
for all 1 ≤ j ≤ λ ≤ k. By the intersection formulas in Lemma 7.15 and Theorem 7.16, we have: Dk,λ ≥ (k!)
2
i−1 k k X (k − 1)! k! 1 X1X1 + −λ 2 i i=2 i j=1 j 4 i=1
k X 1
i
i=1
2
λ(λ + 1) + 4k
k X 1
i
i=1
2
.
It is clear that the worst case occurs for λ = k, namely, D k,λ ≥ Dk,k , and that Dk,k ≥ (k!)
2
k i−1 k X 1 X1X1 + i2 i=2 i j=1 j i=1
(k!)2 − 4
k X 1
i
i=1
2
(k!)2 (k + 1) + 4(k!)2
k X 1
i
i=1
2
.
In other words, denoting the expression on the righthand side above by (k!) 2 δk , we have 2k+1−λ c12k+1−λ (Lk! (Lk! k  Yλ ) = c 1 k )
λ Y
mi c1 (Lk! k X) − ai c1
i=1
≥ (k!)
2k−3
= (k!)2k−1
λ Y
mi
(k!)
mi
δk c21 −
i=1 λ Y
i=1
2
δk c21
− (k!)
2
k X 1 c2 2 i i=1
k X 1 c2 2 i i=1
190
KAREN CHANDLER AND PITMANN WONG
for 1 ≤ λ ≤ k. It remains to show that δk c21
−
or, equivalently,
k X 1 c2 > 0 2 i i=1
δk
>
Pk
2 i=1 1/i for k sufficiently large. We claim that
δk
lim Pk
k→∞
i=1
where δk =
k i−1 k X X 1 1X1 + i2 i j=1 j i=1 i=2
1/i2
1 4
−
c2 c21
(7.7)
= ∞,
(7.8)
k X 1
i
i=1
2
+
(k + 1) 4(k!)2
k X 1
i
i=1
2
.
Observe that i−1 k X 1X1 = i j=1 j i=2
1 ij
X
and
1≤i<j≤k
k X 1
i
i=1
2
=
k X X 1 + 2 i2 i=1
1 ; ij
1≤i<j≤k
hence δk ≥
1 2
1≤i<j≤k
and the ratio satisfies
1 ≥ Pk 2 2 i=1 1/i δk
Since
X
1≤i<j≤k
we must show that lim
k
X
Pk
P
3X 1 1 − , ij 4 i=1 i2
1≤i<j≤k 1/(ij) Pk 2 i=1 1/i
3 − . 4
i−1 k X 1X1 1 = , ij i j=1 j i=2
1/i Pk
i=2
Pi−1
j=1
1/j
= ∞, 1/i2 just as in the limit in Corollary 3.10. But this is clear, because k→∞
i=1
i−1 k k X X 1 1X1 ≥ lim (i − 1) 2 = ∞, k→∞ k→∞ i j i j=1 i=2 i=2
lim
Pk whereas limk→∞ i=1 1/i2 < ∞. Thus (7.7) is verified. We remark that c2 /c21 = 11 for a smooth hypersurface of degree d = 5 in P 3 . Thus, by (7.7), it is enough to choose k so that δk Pk
i=1
1/i2
> 11.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
191
With the aid of a computer, we found that this occurs at k = 2283 (for k = 2282 the ratio on the left above is approximately 10.9998). By Theorem 7.1, 2 5c1 − c2 + 36 ≥ 0, if c21 is even, 5c21 − c2 + 30 ≥ 0, if c21 is odd,
which implies that
23 ≥ 5 + (36/c21 ) ≥ c2 /c21 ,
if c21 is even,
35 ≥ 5 + (30/c21 ) ≥ c2 /c21 ,
if c21 is odd. Pk Thus, by (7.7), we need k so that the ratio δk / i=1 1/i2 is > 23 if c21 is even and > 35 if it is odd. We did not find the explicit k satisfying these conditions; this would take a lot of time, even for the computer. However we do know from (7.8) that k exists. This shows that c12k+1−λ (Lk! k Yλ ) > 0 hence 2 X k! (−1)i H i (Yλ , Lk! k Yλ ) = χ(Lk Yλ ) > 0. i=0
To show that is big it is sufficient to show that H 2 (Yλ , Lk! k Yλ ) = 0 for 0 ≤ λ ≤ k. This is done as in Theorems 7.18 and 7.19 by considering the exact sequences Lk! k  Yλ
⊗σ
0 → Lkk!mλ Yλ−1 ⊗ [−Yλ ] −→λ Lkk!mλ Yλ−1 → Lkk!mλ Yλ → 0 and the induced exact sequence λ · · · → H 2 Lkk!mλ Yλ−1 ⊗ [−Yλ ] → H 2 (Lkk!mλ Yλ−1 ) → H 2 (Lk!m Yλ ) → 0. k
By induction H 2 (Lkk!mλ Yλ−1 ) = 0 for mλ 0 and the exact sequence above implies the vanishing of H 2 (Lkk!mλ Yλ ). This completes the proof of the theorem. ˜ Corollary 7.21. A generic hypersurface surface of degree d ≥ 5 in P 3 is hyperbolic. Proof. The assumptions of Theorem 7.19 are satisfied by a generic hypersurface of degree d ≥ 5 in P 3 . Thus the image of a holomorphic map f : C → X is contained in a curve, necessarily rational or elliptic curve. By a theorem of Xu [1994] a generic hypersurface surface of degree d ≥ 5 in P 3 contains no rational nor elliptic curve. Hence f must be a constant. ˜ The generic condition in Xu means that the statement holds for all curves outside a countable union of Zariski closed sets. A variety X satisfying the condition that every holomorphic curve f : C → X is constant is usually referred to as Brody hyperbolic. In general Kobayashi hyperbolic implies Brody hyperbolic. For compact varieties Brody hyperbolic is equivalent to Kobayashi hyperbolic but for open varieties this is not the case. As a consequence of Corollary 7.21 we have:
192
KAREN CHANDLER AND PITMANN WONG
Corollary 7.22. There exists a curve C of degree d = 5 in P 2 such that P 2 \C is Kobayashi hyperbolic. Proof. It is wellknown that the complement of 5 lines, in general position, in P 2 is Kobayashi hyperbolic. By a Theorem of Zaidenberg [1989] any sufficiently small (in the sense of the classical topology, rather than the Zariski topology) deformation of a Brody hyperbolic manifold is Brody hyperbolic. Thus, for any curve C of degree 5 in a sufficiently small open (in the classical topology) neighborhood U , of 5 lines in general position, the complement P 2 \ C is Brody S hyperbolic. Let Zi be a countable union of Zariski closed sets in the space of surfaces of degree 5 in P 3 such that any surface S 6∈ ∪Zi is hyperbolic. Embed P 2 in P 3 as a linear subspace. Any curve C ∈ C = {S ∩ P 2  S 6∈ ∪Zi } is a curve of degree 5 and is hyperbolic. It is clear that C ∩ U is nonempty. Thus there exists a hyperbolic curve C of degree 5 in P 2 such that P 2 \ C is Brody hyperbolic. It is wellknown that this implies that P 2 \ C is Kobayashi hyperbolic. ˜
Acknowledgements The authors would like to express their gratitude to Professors B. Hu, J.G. Cao, the editor and the referees for many helpful suggestions. We would also like to thank one of the referees for pointing out the work of Ehresmann [1952] on jet bundles.
References [Abate and Patrizio 1994] M. Abate and G. Patrizio, Finsler metrics: a global approach, Lecture Notes in Math. 1591, Springer, Berlin, 1994. [Aikou 1995] T. Aikou, “Complex manifolds modelled on a complex Minkowski space”, J. Math. Kyoto Uni. 35 (1995), 83–101. [Aikou 1998] T. Aikou, “A partial connection on complex Finsler bundles and its applications”, Illinois J. Math. 42 (1998), 481–492. [Alexander 1988] J. Alexander, “Singularit´es imposables en position g´enerale aux hypersurfaces de Pn ”, Compositio Math. 68 (1988), 305–354. [Alexander and Hirschowitz 1992a] J. Alexander and A. Hirschowitz, “Un lemme d’Horace diff´erentiel: application aux singularit´es hyperquartiques de P5 ”, J. Algebraic Geom. 1 (1992), 411–426. [Alexander and Hirschowitz 1992b] J. Alexander and A. Hirschowitz, “La m´ethode d’Horace eclat´ee: application a l’interpolation en degr´e quatre.” Invent. Math. 107 (1992), 585–602. [Alexander and Hirschowitz 1995] J. Alexander and A. Hirschowitz, “Polynomial interpolation in several variables.” J. Algebraic Geom. 4 (1995), 201–222. [An et al. 2003a] T.T.H. An, J. T.Y. Wang and P.M. Wong, “Unique range sets and uniqueness polynomials in positive characteristic”, Acta Arithm. 109 (2003), 259–280.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
193
[An et al. 2003b] T.T.H. An, J. T.Y. Wang and P.M. Wong, “Unique range sets and Uniqueness polynomials in positive characteristic II”, preprint, 2003. [An et al. 2004] T.T.H. An, J. T.Y. Wang and P.M. Wong, “Strong uniqueness polynomials: the complex case”, Complex Variables 9 (2004), 25–54. [Banica and Stanasila 1976] C. Banica and O. Stanasila, Algebraic methods in the global theory of complex spaces, Wiley, New York, 1976. Translation of the original Romanian book of the same title published in 1974. [Besse 1978] A.L. Besse, Manifolds all of whose geodesics are closed, Ergebnisse der Math. (n.F.) 93, Springer, Berlin, 1978. [Bloch 1926] A. Bloch, “Sur les systemes de fonctions uniformes satisfaisant a l’´equation d’une variet´e algebraique dont l’irregularit´e d´epasse la dimension”, J. de Math. 5 (1926), 19–66. [Bogomolov 1977] F. Bogomolov, “Families of curves on a surface of general type”, Sov. Math. Dokl. 18 (1977), 1294–1297. [Bogomolov 1979] F. Bogomolov, “Holomorphic tensors and vector bundles on projective varieties”, Math. USSR Izv. 13 (1979), 499–555. [Bao and Chern 1991] D. Bao and S.S. Chern, On a notable connection in Finsler geometry, Houston J. Math. 19 (1995), 135–180. [Bao et al. 1996] D. Bao, S.S. Chern and Z. Shen (editors), Finsler geometry, Contemporary Math. 196, Amer. Math. Soc., Providence, 1996. [Barth et al. 1984] W. Barth, C. Peters and A. Van de Ven, Compact complex surfaces Ergebnisse der Math. (3) 4, Springer, Berlin, (1984). [Beltrametti and Robbiano 1986] M. Beltrametti and L. Robbiano, “Introduction to the theory of weighted projective spaces”, Exposition. Math. 4 (1986), 111–162. [Cao and Wong 2003] J.K. Cao and P.M. Wong, “Finsler geometry of projectivized vector bundles”, J. Math. Kyoto Univ. 43 (2003), 369–410. [Chandler 1995] K. Chandler, “Geometry of dots and ropes”, Trans. Amer. Math. Soc. 347 (1995), 767–784. [Chandler 1998a] K. Chandler, “A brief proof of a maximal rank theorem for general points in projective space”, Trans. Amer. Math. Soc. 353 (1998), 1907–1920. [Chandler 1998b] K. Chandler, “Higher infinitesimal neighborhoods”, J. of Algebra 205 (1998), 460–479. [Chandler 2002] K. Chandler, “Linear systems of cubics singular at general points in projective space”, Compositio Math. 134 (2002), 269–282. [Chandler 2003] K. Chandler, “The geometric interpretation of Froberg–Iarrobino conjectures on the Hilbert functions of fat points”, to appear in J. Algebra. [Chandler and Wong 2004] K. Chandler and P.M. Wong, “On the holomorphic sectional and bisectional curvatures in complex Finsler geometry”, Periodica Mathematica Hungarica 48 (2004), 1–31. [Dimca 1992] A. Dimca, Singularities and topology of hypersurfaces, Universitext, Springer, Berlin, 1992.
194
KAREN CHANDLER AND PITMANN WONG
[Dolgachev 1982] I. Dolgachev, “Weighted projective varieties”, pp. 34–71 in “Group actions and vector fields” (Vancouver, 1981), edited by J. B. Carrell, Lecture Notes in Math. 956, Springer, New York, 1982. [Dethloff et al. 1995a] G. Dethloff, G. Schumacher and P.M. Wong, “Hyperbolicity of the complements of plane algebraic curves”, Amer. J. Math. 117 (1995), 573–599. [Dethloff et al. 1995b] G. Dethloff, G. Schumacher and P.M. Wong, “Hyperbolicity of the complements of plane algebraic curves: the three components case”, Duke Math. J. 78 (1995), 193–212. [Ehresmann 1952] C. Ehresmann, “Introduction a ` la th´eorie des structures infinit´esimales et des pseudogroupes de Lie”, pp. 97–110 in G´eom´etrie diff´erentielle (Strasbourg, 1953), Colloques internationaux du Centre National de la Recherche Scientifique 52, Paris, CNRS, 1953. [Esnault and Viehweg 1992] H. Esnault and E. Viehweg, Lectures on Vanishing Theorems, DMV Seminar 20, Birkh¨ auser, Basel, 1992. [Faltings 1991] G. Faltings, “Diophantine approximation on abelian varieties”, Ann. Math. 133 (1991), 549–576. [Fujimoto 2001] H. Fujimoto, “A family of hyperbolic hpersurfaces in the complex projective space”, Complex Variables 43 (2001), 273–283. [Fulton and Harris 1991] W. Fulton and J. Harris, Representation Theory, Grad. Texts in Math. 129, Springer, New York, 1991. [Grant 1986] C. G. Grant, “Entire holomorphic curves in surfaces”, Duke Math. J. 53 (1986), 345–358. [Grant 1989] C. G. Grant, “Hyperbolicity of surfaces modular rational and elliptic Curves”, Pacific J. Math. 139 (1989), 241–249. [Grauert 1989] H. Grauert, “Jetmetriken und hyperbolische Geometrie”, Math. Z. 200 (1989), 149–168. [Griffiths 1969] P. Griffiths, “Hermitian differential geometry, Chern classes, and positive vector bundles”, pp. 185–251 in Global Analysis, edited by D. C. Spencer and S. Iyanaga, University of Tokyo Press and Princeton University Press (1969). [Green and Griffiths 1980] M. Green and P. Griffiths, “Two applications of algebraic geometry to entire holomorphic mappings”, pp. 41–74 in The Chern Symposium (Berkeley, 1979), Springer, New York, 1980. [Grothendieck 1958] A. Grothendieck, “La theorie des classes de Chern”, Bull. Soc. math. France 86, 137–154 (1958). [Hardy and Wright 1970] G. H. Hardy and E. M. Wright, “An introduction to the theory of numbers”, 3rd ed., Clarendon Press, Oxford, 1970. [Hartshorne 1968] R. Hartshorne, “Ample vector bundles”, Publ. IHES 29 (1968), 319– 350. [Hirzebruch 1966] F. Hirzebruch, “Topological methods in algebraic geometry”, Grundlehren der Math. Wiss. 131, Springer, Berlin, 1966. [Iarrobino and Kanev 1999] A. Iarrobino and V. Kanev, Power sums, Gorenstein algebras and determinantal loci, Lectures Notes in Math. 1721 Springer, Berlin, 1999.
FINSLER GEOMETRY OF HOLOMORPHIC JET BUNDLES
195
[Jung 1995] E.K. Jung, “Holomorphic Curves in Projective Varieties”, Ph.D. thesis, Univ. Notre Dame, Notre Dame (IN), 1995. [Kawasaki 1973] T. Kawasaki, “Cohomology of twisted projective spaces and lens complexes”, Math. Ann. 206, 243–248 (1973). [Kobayashi 1970] S. Kobayashi, “Hyperbolic manifolds and holomorphic mappings”, Dekker, New York, 1970. [Kobayashi and Ochiai 1970] S. Kobayashi and T. Ochiai, “On complex manifolds with positive tangent bundles”, J. Math. Soc. Japan 22 (1970), 499–525. [Lang 1987] S. Lang, “Introduction to Complex Hyperbolic Spaces”, Springer, New York, 1987. [Lu 1991] S. Lu, “On meromorphic maps into varieties of loggeneral type”, pp. 305– 333 in Several complex variables and complex geometry, edited by Eric Bedford et al., Proc. Symp. Amer. Math. Soc. 52, Amer. Math. Soc., Providence (RI), 1991. [Lu and Yau 1990] S. Lu and S.T. Yau, “Holomorphic curves in surfaces of general type”, Proc. Natl. Acad. Sci. USA 87 (1990), 80–82. [Maruyama 1981] M. Maruyama, “The theorem of Grauert–Mulich–Splindler”, Math. Ann. 255 (1981), 317–333. [Miyaoka 1977] Y. Miyaoka, “On the Chern numbers of surfaces of general type”, Invent. Math. 42 (1977), 225–237. [Nadel 1989] A. Nadel, “Hyperbolic surfaces in P 3 ”, Duke Math. J. 58 (1989), 749–771. [Okonek et al. 1980] C. Okonek, M. Schneider and H. Spindler, Vector bundles on complex projective spaces, Progress in Math. 3, Birkh¨ auser, Boston, 1980. [Royden 1971] H. Royden, “Integral points of Pn \ {2n+1 hyperplanes in general position}”, Invent. Math. 106 (1991), 195–216. [Ru and Wong 1991] M. Ru and P.M. Wong, Integral points of Pn \{2n+1 hyperplanes in general position}, Invent. Math. 106 (1991), 195–216. [Sakai 1980] F. Sakai, “Semistable curves on algebraic surfaces”, Math. Ann. 254 (1980), 89–120. [Stoll 1965] W. Stoll, Value distribution of holomorphic maps into compact complex manifolds, Lectures Notes in Math. 135, Springer, Berlin, 1965. [Stoll and Wong 1994] W. Stoll and P.M. Wong, “Second Main Theorem of Nevanlinna Theory for nonequidimensional meromorphic maps”, Amer. J. Math. 116 (1994), 1031–1071. [Stoll and Wong 2002] W. Stoll and P.M. Wong, “On holomorphic jet bundles”, preprint, 2002. [Tsuji 1987] H. Tsuji, “An inequality of Chern numbers for open algebraic varieties”, Math. Ann. 277 (1987), 483–487. [Tsuji 1988] H. Tsuji, “Stability of tangent bundles of minimal algebraic varieties”, Topology 22 (1988), 429–441. [Wong 1982] P.M. Wong, “Geometry of the complex homogeneous Monge–Amp`ere equation”, Invent. Math. 67 (1982), 261–274. [Wong 1989] P.M. Wong, “On the Second Main Theorem of Nevanlinna Theory”, Amer. J. of Math. 111 (1989), 549–583.
196
KAREN CHANDLER AND PITMANN WONG
[Wong 1993a] P.M. Wong, “Holomorphic curves in spaces of constant curvature”, pp. 201–223 in Complex geometry, edited by G. Komatsu and Y. Sakane, Lecture Notes in Pure and Applied Math. 143, Dekker, New York, 1993. [Wong 1993b] P.M. Wong, “Recent results in hyperbolic geometry and diophantine geometry”, pp. 120–135 in Proc. International Symposium on Holomorphic Mappings, Diophantine Geometry and Related Topics, edited by J. Noguchi, RIMS, Kyoto, 1993. [Wong 1999] P.M. Wong, Topics on higher dimensional Nevanlinna theory and intersection theory, Six lectures at the Center of Theoretical Science, National TsingHua University, 1999. [Xu 1994] G. Xu, “Subvarieties of general hypersurfaces in projective space”, J. Diff. Geom. 39 (1994), 139–172. [Zaidenberg 1989] M.G. Zaidenberg, “Stability of hyperbolic embeddedness and construction of examples”, Math. USSR Sbornik 63 (1989), 351–361. Karen Chandler Department of Mathematics University of Notre Dame Notre Dame, IN 46556 United States
[email protected] PitMann Wong Department of Mathematics University of Notre Dame Notre Dame IN 46556 United States
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Ricci and Flag Curvatures in Finsler Geometry DAVID BAO AND COLLEEN ROBLES
Contents Introduction 1. Flag and Ricci Curvatures 1.1. Finsler metrics 1.2. Flag curvature 1.3. Ricci curvature 2. Randers Metrics in Their Defining Form 2.1. Basics 2.2. Characterising Einstein–Randers metrics 2.3. Characterising constant flag curvature Randers metrics 3. Randers Metrics Through Zermelo Navigation 3.1. Zermelo navigation 3.2. Navigation description of curvature conditions 3.3. Issues resolved by the navigation description 4. Einstein Metrics of Nonconstant Flag Curvature 4.1. Examples with Riemannian–Einstein navigation data 4.2. Examples with K¨ ahler–Einstein navigation data 4.3. Rigidity and a Ricciflat example 5. Open Problems 5.1. Randers and beyond 5.2. Chern’s question 5.3. Geometric flows Acknowledgments References
198 199 199 207 214 218 218 220 228 230 230 236 241 244 244 247 251 254 254 255 256 256 256
Mathematics Subject Classification: 53B40, 53C60. Keywords: Finsler metric, flag curvature, Ricci curvature, Einstein metric, Zermelo navigation, Randers space, Riemannian metric, sectional curvature. Bao’s research was supported in part by R. Uomini and the S.S. Chern Foundation for Mathematical Research. Robles’ research was supported in part by the UBC Graduate Fellowship Program.
197
198
DAVID BAO AND COLLEEN ROBLES
Introduction It is our goal in this article to present a current and uniform treatment of flag and Ricci curvatures in Finsler geometry, highlighting recent developments. (The flag curvature is a natural extension of the Riemannian sectional curvature to Finsler manifolds.) Of particular interest are the Einstein metrics, constant Ricci curvature metrics and, as a special case, constant flag curvature metrics. Our understanding of Einstein spaces is inchoate. Much insight may be gained by considering the examples that have recently proliferated in the literature. This motivates us to discuss many of these metrics. Happily, the theory is developing as well. The Einstein and constant flag curvature metrics of spaces of Randers type, a fecund class of Finsler spaces, are now properly understood. Enlightenment comes from being able to identify the class as solutions to Zermelo’s problem of navigation, a perspective that allows a very apt characterisation of the Einstein spaces. When specialised to flag curvature, the navigation description yields a complete classification of the constant flag curvature Randers metrics. We hope to bring out the rich variety of behaviour displayed by these metrics. For example, Finsler metrics of constant flag curvature exhibit qualities not found in their constant sectional curvature Riemannian counterparts. • Beltrami’s theorem guarantees that a Riemannian metric is projectively flat if and only if it has constant sectional curvature. On the other hand, there are many Finsler metrics of constant flag curvature which are not projectively flat. See Section 3.2.3 and [Shen 2004]. • Every Riemannian metric of constant sectional curvature K is locally isometric to a round sphere, Euclidean space, or a hyperbolic space, depending on K. Hence, for each K, there is only one Riemannian standard model, up to isometry. By contrast, on S n , Rn , and the unit ball B n , there are numerous nonisometric Randers metrics of constant flag curvature K. In fact, isometry classes of Randers type standard models make up a moduli space M K whose dimension is linear in n. See the table on page 242 and [Bryant 2002]. A roadmap. This article is written with a variety of readers in mind, ranging from the geometric neophyte to the Finsler aficionado. We anticipate that these users will approach the manuscript with distinct aims. The outline below is intended to help readers navigate the article efficiently. Section 1 introduces Finsler metrics and their curvatures, as well as tools and constructions that are endemic to nonRiemannian Finsler geometry. The reader conversant with Finsler metrics might only skim this section to glean our notation and conventions. In Section 2 we develop a useful characterisation of the Einstein spaces among a ubiquitous class of Finsler metrics. This description generalises a characterisation of constant flag curvature Randers metrics ([Bao–Robles 2003] and
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
199
[Matsumoto–Shimada 2002]) to the Einstein realm. The resulting conditions form a tensorial, coupled system of nonlinear second order partial differential equations, whose unknowns consist of Riemannian metrics a and 1forms b. These equations provide a substantial step forward in computational efficiency over the defining Einstein criterion (which stipulates that the ‘average flag curvature’ is to be a function of position only). Indeed, study of these equations has led to the construction of the Finslerian Poincar´e metric ([Okada 1983] and [Bao et al. 2000]), as well as the S 3 metric [Bao–Shen 2002]. However, while the characterisation improves the computational accessibility of Einstein metrics, it does little to advance our understanding of their geometry. It is in the following section that we pursue this geometric insight. The sine qua non here is Shen’s observation that Randers metrics may be identified with solutions to Zermelo’s problem of navigation on Riemannian manifolds. This navigation structure establishes a bijection between Randers spaces (M, F = α+β) and pairs (h, W ) of Riemannian metrics h and vector fields W on the manifold M . From this perspective, the characterisation of Section 2 is parlayed into a breviloquent geometric description of Einstein metrics. Explicitly, the Randers metric F with navigation data (h, W ) is Einstein if and only if h is Einstein and W is an infinitesimal homothety of h. (In particular, these h and W solve the system of partial differential equations in Section 2.) The transparent nature of the navigation description immediately yields a Schur lemma for the Ricci scalar, together with a certain rigidity in three dimensions. The variety of examples in the article may be categorised as follows. • • • • •
Metrics in their defining form: Sections 1.1.1, 1.2.1, 2.1.1, 2.3.2, 3.1.2 Solutions to Zermelo navigation: Sections 1.1.1, 3.1.1, 3.1.2, 3.2.3 Of constant flag curvature: Sections 1.2.1, 2.3.2, 3.1.2, 3.2.3, 4.1.1 Einstein but not of const. flag curvature: Sections 4.1.1, 4.1.2, 4.2.3, 4.3.3 Ricciflat Berwald: Sections 3.1.1 (locally Minkowski), 4.3.3 (not loc. Mink.)
In Section 4, the emphasis is on Einstein metrics of nonconstant flag curvature, especially those on compact boundaryless manifolds. The spaces studied include Finsler surfaces with Ricci scalar a function on M alone (the scalar is a priori a function on the tangent bundle), as well as nonRiemannian Ricciconstant solutions of Zermelo navigation on Cartesian products and K¨ ahler–Einstein manifolds. Section 5 discusses open problems.
1. Flag and Ricci Curvatures 1.1. Finsler metrics 1.1.1. Definition and examples. A Finsler metric is a continuous function F : T M → [0, ∞) with the following properties:
200
DAVID BAO AND COLLEEN ROBLES
(i) Regularity: F is smooth on uniformized in favor of : T M r 0 := {(x, y) ∈ T M : y 6= 0}. (ii) Positive homogeneity: F (x, cy) = cF (x, y) for all c > 0. (iii) Strong convexity: the fundamental tensor gij (x, y) := 12 F 2 yi yj is positive definite for all (x, y) ∈ T M r 0. Here the subscript partial differentiation by y i .
yi
denotes
Strong convexity implies that {y ∈ Tx M : F (x, y) 6 1} is a strictly convex set, but not vice versa; see [Bao et al. 2000]. p The function F for a Riemannian metric a is F (x, y) := aij (x)y i y j . In this case, one finds that gij := ( 21 F 2 )yi yj is simply aij . Thus the fundamental tensor for general Finsler metrics may be thought of as a directiondependent Riemannian metric. This viewpoint is treated more carefully in Section 1.1.2. Many calculations in Finsler geometry are simplified, or magically facilitated, by Euler’s theorem for homogeneous functions: Let φ be a real valued function on Rn , differentiable at all y 6= 0. The following two statements are equivalent. • φ(cy) = cr φ(y) for all c > 0 (positive homogeneity of degree r). • y i φyi = rφ; that is, the radial derivative of φ is r times φ. (See, for example, [Bao et al. 2000] for a proof.) This theorem, for instance, lets us invert the defining relation of the fundamental tensor given above to get F 2 (x, y) = gij (x, y)y i y j . Consequently, strong convexity implies that F must be positive at all y 6= 0. The converse, however, is false; positivity does not in general imply strong convexity. This is because while gij (x, y)y i y j = F 2 (x, y) may be positive for y 6= 0, the quadratic gij (x, y) y˜i y˜j could still be 6 0 for some nonzero y˜. Here are some 2dimensional examples. Being in two dimensions, we revert to the common notation of denoting position coordinates by x, y rather than x 1 , x2 , and components of tangent vectors by u, v rather than y 1 , y 2 . Example (Quartic metric). Let F (x, y; u, v) := (u4 + v 4 )1/4 . Positivity is manifest. However, det(gij ) = 3u2 v 2 /(u4 + v 4 ) along the tangent vector u∂x + v∂y based at the point (x, y). Thus (gij ) fails to be a positive definite matrix whenu or v vanishes. For instance, if u = 0 but v 6= 0, we have gij (x, y; 0, v) = 10 00 , which is not positive definite. It is shown in [Bao et al. 2000] that F can be regularised to restore strong convexity. ♦
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
201
We shall see in Section 2.1.2 that, surprisingly, positivity of F does imply strong convexity for Randers metrics. Next, consider a surface S given by the graph of a smooth function f (x, y). Parametrise S via (x, y) 7→ (x, y, f ). By a slight abuse of notation, set ∂ x := (1, 0, fx ) and ∂y := (0, 1, fy ), and denote the natural dual of this basis by dx, dy. The Euclidean metric of R3 induces a Riemannian metric on S: h := 1 + fx2 dx ⊗ dx + fx fy (dx ⊗ dy + dy ⊗ dx) + 1 + fy2 dy ⊗ dy.
If Y := u∂x + v∂y is an arbitrary tangent vector on S, we have Y 2 := h(Y, Y ) = u2 + v 2 + (ufx + vfy )2 .
We note for later use that the contravariant description of df = f x dx + fy dy is the vector field (df )] =
1
(fx ∂x + fy ∂y ), 2
1 + fx2 + fy
with
(df )] 2 =
fx2 + fy2 . 1 + fx2 + fy2
Example (Metric from Zermelo navigation). If we assume that gravity acts perpendicular to S, a person’s weight does not affect his motion along the surface. Now introduce a wind W = W x ∂x + W y ∂y blowing tangentially to S. The norm function F that measures travel time on S can be derived using a procedure (Section 3.1) due to Zermelo and generalised by Shen. With W 2 = (W x )2 + (W y )2 + (W x fx + W y fy )2 , h(W, Y ) = uW x + v W y + (ufx + vfy )(W x fx + W y fy ), and λ := 1 − W 2 , the formula for F reads p h(W, Y )2 + Y 2 λ h(W, Y ) F (x, y; u, v) = − . λ λ This Zermelo navigation metric F is strongly convex if and only if W  < 1. The unit circle of h in each tangent plane represents the destinations reachable in one unit of time when there is no wind. It will be explained (Section 3.1) that the effect of the wind is to take this unit circle and translate it rigidly by the amount W . The resulting figure is offcentered and represents the locus of unit time destinations under windy conditions, namely the indicatrix of F . Since the latter lacks central symmetry, F could not possibly be Riemannian; the above formula makes explicit this fact. ♦ Example (Matsumoto’s slopeofamountain metric). Take the same surface S, but without the wind. View S as the slope of a mountain resting on level ground, with gravity pointing down instead of perpendicular to S. A person who can walk with speed c on level ground navigates this hillside S along a path that makes an angle θ with the steepest downhill direction. The acceleration of gravity (of magnitude g), being perpendicular to level ground, has a
202
DAVID BAO AND COLLEEN ROBLES
p component of magnitude gk = g (fx2 + fy2 )/(1 + fx2 + fy2 ) along the steepest downhill direction. The hiker then experiences an acceleration g k cos θ along her path, and compensates against the gk sin θ which tries to drag her offcourse. Under suitable assumptions about frictional forces, the acceleration g k cos θ rapidly effects a terminal addition 21 gk cos θ to the pace c generated by her leg muscles. In other words, her speed is effectively of the form c + a cos θ, where a is independent of θ. Thus the locus of unit time destinations is a lima¸con. The unit circle of h, instead of undergoing a rigid translation as in Zermelo navigation, has now experienced a directiondependent deformation. The norm function F with this lima¸con as indicatrix measures travel time on S. It was worked out by Matsumoto, after being inspired by a letter from P. Finsler, and reads Y 2 ; cY  − (g/2)(ufx + vfy ) see [Matsumoto 1989] and [Antonelli et al. 1993]. For simplicity, specialise to the case c = g/2. Multiplication by c then converts this norm function to F (x, y; u, v) :=
(df )(Y ) Y 2 = Y  ϕ , Y  − (ufx + vfy ) Y 
with ϕ(s) := 1/(1 − s). We see from [Shen 2004] in this volume that metrics of the type αϕ(β/α) are strongly convex whenever the function ϕ(s) satisfies ϕ(s) > 0, ϕ(s)−sϕ0 (s) > 0, and ϕ00 (s) > 0. For the ϕ at hand, this is equivalent to (df )(Y ) < 12 Y , which is in turn equivalent to (df )]  < 21 . (In one direction, set Y = (df )] ; the converse follows from a Cauchy–Schwarz inequality.) Using the formula for (df )] 2 presented earlier, this criterion produces fx2 + fy2 < 13 . Whenever this holds, F defines a Finsler metric. Such is the case for f (x, y) := 12 x but not for f (x, y) := x, even though the surface S is an inclined plane in both instances. As for the elliptic paraboloid given by the graph of f (x, y) := 100−x2 −y 2 , we have strong convexity only in a circular vicinity of the hilltop. ♦ The functions F in these two examples are not absolutely homogeneous (and therefore nonRiemannian) because at any given juncture, the speed with which one could move forward typically depends on the direction of travel. Our discussion also raises a tantalising question: if the wind were blowing on the slope of a mountain, would the indicatrix of the resulting F be a rigid translate of the lima¸con? 1.1.2. The pulledback bundle and the fundamental tensor. The matrix g ij involved in the definition of strong convexity is known as the fundamental tensor, and has a geometric meaning, which is most transparent through the use of the pulledback bundle introduced by Chern.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
(π ∗ T M )(x,y)
Tx M
x
π
(x,y)
203
TM r0
M
Figure 1. The pulledback tangent bundle π T M is a vector bundle over the “parameter space” T M r 0. The fiber over any point (x, y) is a copy of Tx M . The dotted part is the deleted zero section. ∗
The gij depend on both x and y ∈ Tx M . Over each fixed point (x, y) ∈ T M r 0, the bundle π ∗ T M provides a copy of Tx M . Endow this Tx M with the symmetric bilinear form g := gij (x, y) dxi ⊗ dxj . Since the Finsler metric is strongly convex, this bilinear form is positive definite, which renders it an inner product. Thus, every Finsler metric endows the fibres of the pulledback bundle with a Riemannian metric. Two facts stand out: (1) The fundamental tensor gij (x, y) is invariant under y 7→ λy for all λ > 0. This invariance directly follows from the hypothesis that F is positively homogeneous of degree 1 in y. Thus, over the points {(x, λy) : λ > 0} in the parameter space T M r 0, not only the fibres are identical, but the inner products are too. (2) That gij arises as the yHessian of 21 F 2 imposes a stringent symmetry condition. Namely, (gij )yk must be totally symmetric in its three indices i, j, k. For this reason, not every Riemannian metric on the fibres of the pulledback bundle comes from a Finsler metric. Property (1) is a redundancy that will be given a geometrical interpretation below. We will also explain why the symmetry criterion (2) is an integrability condition in disguise. Let’s switch our perspective from the pulledback bundle π ∗ T M to the tangent bundle T M itself. Recall that a Riemannian metric on M is a smooth assignment of inner products, one for each tangent space Tx M . By contrast, a Finsler metric F gives rise to a family of inner products gij (x, y) dxi ⊗ dxj on each tangent space Tx M . Item (1) above means that there is exactly one inner product for each direction. This sphere’s worth of inner products on each tangent space has to satisfy the symmetry condition described in (2). The converse is also true. Suppose we are given a family of inner products gij (x, y) on each tangent space Tx M , smoothly dependent on x and nonzero
204
DAVID BAO AND COLLEEN ROBLES
y ∈ Tx M , invariant under positive rescaling of y (that is, maps y 7→ λy, for λ > 0) and such that (gij )yk is totally symmetric in i, j, and k. We construct a Finsler function as follows: q F (x, y) := gpq (x, y)y p y q . This F is smooth on the entire slit tangent bundle T M r 0. It is positively homogeneous of degree 1 in y because gpq (x, y) is invariant under positive rescaling in y. Also, with the total symmetry of (gij )yk , we have (F 2 )yi yj = (gpq y p y q )yi yj = (gpq )yi y p y q + giq y q + gpi y p yj = (giq )yj y q + gij + (gpi )yj y p + gji = 2gij .
Thus F is strongly convex because, for each nonzero y, the matrix g ij (x, y) is positive definite. In the calculation above, the quantity (gpq )yi y p on the first line and the terms (giq )yj y q and (gpi )yj y p on the second line are all zero because of the hypothesis that (grs )yt is totally symmetric in r, s, t. For instance, (gpq )yi y p = (giq )yp y p = 0, where the last equality follows from Euler’s theorem, and the assumption that giq (x, y) is positive homogeneous of degree 0 in y. The symmetry hypothesis therefore plays the role of an integrability condition. We conclude that the concept of a Finsler metric on M is equivalent to an assignment of a sphere’s worth of inner products gij (x, y) on each tangent space Tx M , such that (gij )yk is totally symmetric in i, j, k and appropriate smoothness holds. Similarly, a Finsler metric on M is also equivalent to a smooth Riemannian metric on the fibres of the pulledback bundle π ∗ T M , satisfying the above redundancy (1) and integrability condition (2). The pulledback bundle π ∗ T M and its natural dual π ∗ T ∗ M each contains an important global section. They are ` :=
yi ∂xi , F (x, y)
the distinguished section, and ω := Fyi dxi , the Hilbert form. As another application of the integrability condition, we differentiate the statement F 2 = gij y i y j to find g
Fy i =
yi = gij `j =: `i , F
with
g
yi := gij y j .
This readily gives `i `i = 1 and `i;j = gij − `i `j , where the semicolon abbreviates the differential operator F ∂ yi .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
205
1.1.3. Geodesic spray coefficients and the Chern connection. We have seen in Section 1.1.1 that if F is the Finsler function of a Riemannian metric, its fundamental tensor has no ydependence. The converse follows from the fact that gij y i y j reconstructs F 2 for us, thanks to Euler’s theorem. Thus, the y derivative of gij measures the extent to which F fails to be Riemannian. More formally, we define the Cartan tensor Aijk := 12 F (gij )yk = 14 F (F 2 )yi yj yk , which is totally symmetric in all its indices. As an illustration of Euler’s theorem, note that since g is homogeneous of degree zero in y, we have y i Aijk = 0. Besides the Cartan tensor, we can also associate to g its formal Christoffel symbols of the second kind , γ i jk := 12 g is (gsj,xk − gjk,xs + gks,xj ), and the geodesic spray coefficients Gi := 21 γ i jk y j y k . The latter are so named because the (constant speed) geodesics of F are the solutions of the differential equation x ¨ i + x˙ j x˙ k γ i jk (x, x) ˙ = 0, which may be i i abbreviated as x ¨ + 2G (x, x) ˙ = 0. Caution: the Gi defined here is equal to half the Gi in [Bao et al. 2000]. Covariant differentiation of local sections of the pulledback bundle π ∗ T M requires a connection, for which there are many namebrand ones. All of them have their genesis in the nonlinear connection N i j := (Gi )yj . This N i j is a connection in the Ehresmann sense, because it specifies a distribution of horizontal vectors on the manifold T M r 0, with basis δ := ∂xj − N i j ∂yi . δxj As another application of Euler’s theorem, note that N i j y j = 2Gi , since Gi is homogeneous of degree 2 in y. We also digress to observe that the Finsler function F is constant along such horizontal vector fields: Fj :=
δ F = 0. δxj
The key lies in the following sketch of a computation: N i j `i = (Gi `i )yj −
1 i G `i;j = Fxj , F
in which establishing (`i Gi )yj = 12 (Fxj + y k Fxk yj ) and its companion statement −(1/F )Gi `i;j = 12 (Fxj − y k Fxk yj ) takes up the bulk of the work.
206
DAVID BAO AND COLLEEN ROBLES
Using A, γ, and N , we can now state the formula of the Chern connection in natural coordinates [Bao et al. 2000]: Γi jk = γ i jk −
1 is g (Asjt N t k − Ajkt N t s + Akst N t j ), F
with associated connection 1forms ωj i := Γi jk dxk . These ωj i represent the unique, torsionfree (Γi kj = Γi jk ) connection which is almost gcompatible: dgij − gkj ωi k − gik ωj k =
2 Aijk (dy k + N k l dxl ). F
As yet another application of Euler’s Theorem, we shall show that the nonlinear connection N can be recovered from the Chern connection Γ as follows: Γi jk y k = N i j . A crucial ingredient in the derivation is (γ i st )yj y s y t = 2(g ip )yj Gp . Indeed, (γ i st )yj y s y t = (g ip γpst )yj y s y t = (g )yj γpst y y + g ip
s t
= 2(g ip )yj Gp + 0.
ip
1
F
Apsj
−
xt
1
F
Astj
+
xp
1
F
Atpj
ys yt
xs
Whence, with the help of Aijk y k = 0 and N t k y k = 2Gt , we have 2 is g Astj Gt = γ i jk y k − g is (gst )yj Gt F = γ i jk y k + (g is )yj gst Gt
Γi jk y k = γ i jk y k −
= γ i jk y k + 12 (γ i st )yj y s y t = ( 21 γ i st y s y t )yj = (Gi )yj = N i j , as claimed. It is now possible to covariantly differentiate sections of π ∗ T M (and its tensor products) along the horizontal vector fields δ/δx k of the manifold T M r 0. For example, δ T i jk := k T i j + T s j Γi sk − T i s Γs jk . δx If F arises from a Riemannian metric a, then A = 21 F (aij )yk = 0 because a has no ydependence. In that case, Γ is given by the Christoffel symbols of a. If the tensor T also has no ydependence, T i jk reduces to the familiar covariant derivative in Riemannian geometry. Let’s return to the Finsler setting. Using the symmetry Γ i sk = Γi ks and the recovery property y s Γi ks = N i k , we have y i k =
δ i y + y s Γi ks = 0, δxk
hence `i k = 0.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
207
Also, the covariant derivative of the Cartan tensor A along the special horizontal vector field `s δ/δxs gives the Landsberg tensor, which makes frequent appearances in Finsler geometry: A˙ ijk := Aijks `s . Note that A˙ is totally symmetric, and its contraction with y vanishes. The Chern connection gives rise to two curvature tensors: δ δ i Γ jl − l Γi jk + Γi sk Γs jl − Γi sl Γs jk , k δx δx ∂ i = −F l Γ jk (note the symmetry: Pj i kl = Pk i jl ), ∂y
Rj i kl = Pj i kl
both invariant under positive rescaling in y. In the case of Riemannian metrics, Γ reduces to the standard LeviCivita (Christoffel) connection, which is indepen∂ δ Γ becomes ∂x Γ. The curvature R is then the usual Riemann dent of y; hence δx tensor, and P is zero. In Finsler geometry, there are many Bianchi identities. A leisurely account of their derivation can be found in [Bao et al. 2000]. 1.2. Flag curvature. This is a generalisation of the sectional curvature of Riemannian geometry. Alternatively, flag curvatures can be treated as Jacobi endomorphisms [Foulon 2002]. The flag curvature has also led to a pinching (sphere) theorem for Finsler metrics; see [Rademacher 2004] in this volume. 1.2.1. The flag curvature versus the sectional curvature. Installing a flag on a Finsler manifold (M, F ) implies choosing ◦ a basepoint x ∈ M at which the flag will be planted, ◦ a flagpole given by a nonzero y ∈ Tx M , and ◦ an edge V ∈ Tx M transverse to the flagpole. See Figure 2. Note that the flagpole y 6= 0 singles out an inner product gy := gij (x, y) dxi ⊗ dxj from among the sphere’s worth of inner products described in Section 1.1.2. This gy allows us to measure the angle between V and y. It also enables us to calculate the area of the parallelogram formed by V and ` := y/F (x, y). The flag curvature is defined as K(x, y, V ) :=
V i (y j Rjikl y l )V k , gy (y, y)gy (V, V ) − gy (y, V )2
where the index i on Rj i kl has been lowered by gy . When the Finsler function F comes from a Riemannian metric, gy is simply the Riemannian metric, Rjikl is the usual Riemann tensor, and K(x, y, V ) reduces to the familiar sectional curvature of the 2plane spanned by {y, V }.
208
DAVID BAO AND COLLEEN ROBLES
Tx M transverse edge V
flag `
flagpole y
basepoint x Finsler manifold M
Figure 2. A typical flag, based at the point x on a Finsler manifold M . The flagpole is y, and the “cloth” part of the flag is ` ∧ V . The entire flag lies in the tangent space Tx M .
Since gy (y, y) = F 2 (x, y), the flag curvature can be reexpressed as K(x, y, V ) =
V i (`j Rjikl `l )V k . gy (V, V ) − gy (`, V )2
The denominator here is the areasquared of the parallelogram formed by V and the gy unit vector ` = y/F . The tensor Rik := `j Rjikl `l is called the predecessor of the flag curvature. It is proved in [Bao et al. 2000] that Rki = Rik . We also note that `i Rik = 0 = Rik `k . The second equality is immediate because Rj i kl is manifestly skewsymmetric (Section 1.1.3) in k and l. The first equality follows from the symmetry of Rik . One can use y i Rik = 0 = Rik y k to show that if {y, V1 } and {y, V2 } have the same span, then K(x, y, V1 ) = K(x, y, V2 ). In other words, K depends on x, y, and span{y, V }. A Finsler metric is of scalar curvature if K(x, y, V ) does not depend on V , that is, if Rik V i V k = K(x, y)(gik − `i `k )V i V k . This says that two symmetric bilinear forms generate the same quadratic form. A polarisation identity then tells us that the bilinear forms in question must be equal. So, Finsler metrics of scalar curvature are described by Rik = K(x, y)(gik − `i `k ), where `i = Fyi = gyi /F (see Section 1.1.2).
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
209
Example (Numata metrics). Numata [1978] has shown that the Finsler metrics q F (x, y) = qij (y)y i y j + bi (x)y i ,
where q is positive definite and b is closed, are of scalar curvature. Note that the first term of F is a locally Minkowski norm, Riemannian only when the q ij are constant. Consider the case that qij = δij and b = df , where f is a smooth p function on Rn . If necessary, scale f so that the open set M := x ∈ Rn : δ ij fxi fxj < 1 is nonempty. Then a straightforward calculation reveals that F is of scalar curvature on M with 3 1 1 1 K(x, y) = (fxi xj y i y j )2 − (f i j k y i y j y k ). 4 4F 2 F3 x x x The computation in [Bao et al. 2000] utilises the spray curvature and Berwald’s formula, to be discussed in Section 1.2.3. The Numata metrics are projectively flat. For Finsler metrics of (nonconstant) scalar curvature but not projectively flat, see [Shen 2004] in this volume. ♦ The flag curvature is an important geometric invariant because its sign governs the growth of Jacobi fields, which in turn gives qualitative information about short geodesic rays with close initial data. See [Bao et al. 2000]. To bring out the essential difference between the Finslerian and Riemannian settings, we consider the case of surfaces. There, once the basepoint x and the flagpole y are chosen, span{y, V } is equal to the tangent plane Tx M for all transverse edges V . Thus every Finsler surface is of scalar curvature K(x, y). Remark. It is evident from the Numata metrics that the sign of K(x, y) can depend on the direction of the flagpole y ∈ Tx M . By contrast, when the surface is Riemannian, this K(x, y) reduces to the usual Gaussian curvature K(x), which does not depend on y. The implication of this difference is profound when we survey the immediate vicinity of any fixed x. If the landscape is Riemannian, the sign of K(x) creates only one type of geometry near x: hyperbolic, flat, or spherical. If the landscape is Finslerian, the sign of K(x, y) can depend on the direction y of our line of sight, making it possible to encounter all three geometries during the survey! If K is a constant (namely, it depends neither on V , nor y, nor x), the Finsler metric F is said to be of constant flag curvature. The tensorial criterion is Rik = K (gik − `i `k ), with K constant. For later purposes, we rewrite it as F 2 Ri k = K (F 2 δ i k − y i gyk ), where gyk := gks y s (see Section 1.1.2). Example (Bryant’s metrics). Bryant discovered an interesting 2parameter family of projectively flat Finsler metrics on the sphere S 2 , with constant flag
210
DAVID BAO AND COLLEEN ROBLES
curvature K = 1. Here we single out one metric from this family for presentation; see [Bryant 1997] for the geometry behind the construction. p Parametrise the hemispheres of S 2 via the map (x, y) 7→ x, y, s 1−x2 −y 2 , with s = ±1. Denote tangent vectors by (u, v) = u∂x + v∂y ∈ T(x,y) S 2 , and introduce the following abbreviations: r2 := x2 + y 2 ,
P 2 := 1 − r 2 ,
R2 := u2 + v 2 ,
C := xu + yv;
B := 2r 2 − 1;
2
a := (1 + B ) (P 2 R2 + C 2 ) + B(P 2 R2 − C 2 ) + 8(1 + B)C 2 P 4 ; b := (1 + B 2 ) (P 2 R2 − C 2 ) − B (P 2 R2 + C 2 ) − 8(0 + B)C 2 P 2 .
We emphasise that in b, the very last term contains C 2 P 2 and not C 2 P 4 . The formula for the Finsler function is then s √ 1 a + a2 + b 2 1 + 2C . F (x, y; u, v) = 1 + B2 P 2 Note that a is a quadratic and a2 + b2 is a quartic. All the geodesics of F are arcs of great circles with Finslerian length 2π. As a comparison, the corre2 sponding √ Finsler function for the standard Riemannian metric on S is simply 2 2 2 (1/P ) P R + C . For additional discussions, see [Bao et al. 2000; Sabau 2003; Shen 2004]. ♦ In dimension greater than two, a Schur lemma says that if K does not depend on V and y, it must be constant. This was proved in [del Riego 1973], then in [Matsumoto 1986]; see also [Berwald 1947]. In two dimensions, K can be a function of position x only without being constant. All Riemannian surfaces with nonconstant Gaussian curvature belong to this category; for nonRiemannian examples, see Section 4.1.1. 1.2.2. Rapsc´ak’s identity. We now prepare to relate the flag curvature to one of Berwald’s spray curvatures. Suppose F and F are two arbitrary Finsler metrics, with geodesic spray coefficients Gi and G i . We think of F as a “background” metric, and use a colon to denote horizontal covariant differentiation with respect to the Chern connection of F. (Note that F:j vanishes by Section 1.1.3, but not F:j .) Finally, let g ij denote the inverse of the fundamental tensor of F , and let the subscript 0 abbreviate contraction with y: for instance, F :0 = F:j y j . Then Rapcs´ ak’s identity [Rapcs´ ak 1961] reads Gi = G i + y i
1
2F
F:0 + 21 F g ij (F:0 )yj − 2F:j .
Since Finsler metrics generally involve square roots, another form of Rapcs´ ak’s identity is more userfriendly: Gi = G i + 14 g ij (F 2 :0 )yj − 2F 2 :j .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
211
The derivation of the identity involves three key steps, in which the basic fact F Fyi = gij y j will be used repeatedly without mention. We first verify that 2Gi = (F Fxk )yi y k − F Fxi . 2Gi = γijk y j y k = 12 (gij,xk − gjk,xi + gki,xj )y j y k = (gij y j )xk y k − 21 (gjk y j y k )xi = ( 12 F 2 )yi xk y k − ( 12 F 2 )xi ,
(†)
where grs := ( 12 F 2 )yr ys , and the last displayed equality follows from two applications of Euler’s theorem. Next, observe that F Fxr = F F:r + N j r gyj , where gyj := gjs y s and N is the nonlinear connection of F. Indeed, horizontal differentiation with respect to F is given by ( · · · ):r = ( · · · )xr − N j r ( · · · )yj . Thus F Fxr = F (F:r + N j r Fyj ) = F F:r + N j r (gjs y s ).
(‡)
Here we have chosen to reexpress Fxr using the nonlinear connection of F rather than that of F , thereby opening the door for G to enter the picture. Finally, we substitute equality (‡) into the purpose of (†), getting 2Gi = Fyi F:k y k +F (F:k )yi y k −F F:i +(N j k )yi gyj y k +N j k (gyj )yi y k −N j i gyj . Note that (F:k )yi y k = (F:0 )yi − F:i . Also, since N j k = (G j )yk , Euler’s theorem gives (N j k )yi gyj y k = (G j )yi gyj = N j i gyj , N j k (gyj )yi y k = (2G j ) ( 21 F 2 )yj ys yi y s + gjs δ s i = (2G j )(0 + gji ).
These two statements constitute the heart of the entire derivation. After using them to simplify the above expression for 2Gi , and relabelling i as r, we have 2Gr = 2G j gjr +
1g yr F:0 + F (F:0 )yr − 2F:r . F
Contracting with 21 g ir yields Rapcs´ ak’s identity in its original form. The userfriendly version follows without much trouble. 1.2.3. Spray curvatures and Berwald’s formulae. The definition of the flag curvature through a connection (for example, Chern’s) has theoretical appeal, but is not practical if one wants to compute it. Even for relatively simple Finsler metrics, the machine computation of any namebrand connection is already a daunting and often insurmountable task, let alone the curvature. This is where Berwald’s spray curvatures come to the rescue. They originate from his study of systems of coupled second order differential equations, and are defined entirely in terms of the geodesic spray coefficients [Berwald 1929]. K i k := 2(Gi )xk − (Gi )yj (Gj )yk − y j (Gi )xj yk + 2Gj (Gi )yj yk , Gj i kl := (Gi )yj yk yl .
212
DAVID BAO AND COLLEEN ROBLES
These spray curvatures are related to the predecessor of the flag curvature, and to the P curvature of Chern’s, in the following manner: F 2 Ri k = y j Rj i kl y l = K i k , 1 i Pj kl = −Gj i kl + (A˙ i jk )yl . F The statement about P follows from the fact that the Berwald connection (G) yy ˙ This, together with can be obtained from the Chern connection Γ by adding A. a companion formula, is discussed in the reference [Bao et al. 2000] (whose G i is twice the Gi in the present article). Explicitly, A˙ i jk = (Gi )yj yk − Γi jk
and
A˙ ijk = − 21 gys (Gs )yi yj yk ,
where gys := gst y t . The key that helps establish the first claim is the realisation that Gi = 12 Γi pq y p y q , which holds because contracting A with y gives zero. When calculating the yHessian of Gi , we need the latter part of Section 1.1.3 and the Bianchi identity `j Pj i kl = −A˙ i kl . As for the second claim, here is a sketch of the derivation: ◦ Start with A˙ s jk = (Gs )yj yk − Γs jk , apply ∂yi , contract with gys , and use P = −F ∂y Γ (Section 1.1.3). We get gys (A˙ s jk )yi = gys (Gs )yi yj yk + `s Pjski . ◦ With the two Bianchi identities (3.4.8) and (3.4.9) of [Bao et al. 2000], it can be shown that `s Pjski = −`s Psjki = A˙ jki . ◦ The term gys (A˙ s jk )yi on the left is equal to `s (A˙ s jk );i , which in turn = (`s A˙ s jk );i − `s;i A˙ s jk = 0 − (gsi − `i `s )A˙ s jk = −A˙ ijk . ◦ These manoeuvres produce A˙ ijk = − 12 gys (Gs )yi yj yk as stated. We are now in a position to express the Chern curvatures in terms of the Berwald spray curvatures. Note that applying ∂yl to A˙ i jk = (Gi )yj yk − Γi jk immediately yields the statement involving P . The derivation of the formula F 2 Ri k = K i k makes frequent use of the relationship y k Γi jk = N i j ,
(∗)
from Section 1.1.3. We first demonstrate that δ δT y r s = s (y r T ) + N r s T. δx δx
(∗∗)
This arises from 0 = y r s = δxδ s y r + y k Γr ks ; see Section 1.1.3. Property (∗) gives y k Γr ks = y k Γr sk = N r s . Thus δxδ s y r = −N r s , and (∗∗) follows from the product rule. Using (∗∗) and (∗), we ascertain that F 2 Ri k = y j
δ
δx
N ij − k
δ N ik . δxj
(∗∗∗)
Indeed, F 2 Ri k is obtained by contracting with y j y l the explicit formula for Rj i kl (Section 1.1.3). After several uses of (∗), we get the intermediate expression
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
213
y j (y l δxk Γi jl ) − y l (y j δxl Γi jk ) + y j Γi hk N h j − N i h N h k . Now apply (∗∗) to the first two sets of parentheses, and use (∗) again. The result simplifies to (∗∗∗). Finally, recall from Section 1.1.3 that N i j := (Gi )yj , so Euler’s theorem gives j i y N j = 2Gi . Consequently, yj
−y j
δ δ N i j = k (2Gi ) + N i j N j k by (∗∗) δxk δx = (2Gi )xk − N j k (2Gi )yj + N i j N j k = (2Gi )xk − (Gi )yj (Gj )yk , δ N i k = −y j (N i k )xj + (y j N h j )(N i k )yh = −y j (Gi )yk xj + 2Gj (Gi )yk yj . δxj
Summing these two conclusions gives Berwald’s formula for K i k . (For ease of exposition, we shall refer to K i k simply as the spray curvature.) We now return to the setting of two arbitrary Finsler metrics F and F, as discussed in the previous subsection. Denote their respective spray curvatures by K i k and Ki k . According to Rapcs´ ak’s identity, the geodesic coefficients of F i i i and F are related by G = G + ζ . Inspired by Shibata–Kitayama, we now show that substituting this decomposition into Berwald’s formula for K i k allows us to rewrite the latter in a split and covariantised form K i k = Ki k + 2(ζ i ):k − (ζ i )yj (ζ j )yk − y j (ζ i :j )yk + 2ζ j (ζ i )yj yk + 3ζ j A˙ i jk , where the colon refers to horizontal covariant differentiation with respect to the Chern connection of F, and A˙ is associated to F as well. Remarks. 1. Had we used the Berwald connection (which equals the Chern ˙ that 3ζ A˙ term in the above formula would have been connection plus A), absorbed away. 2. If the background metric were Riemannian, F 2 = aij (x)y i y j , then Aijk = 1 ˙ 2 F(aij )y k would be zero because a is independent of y. Hence A = 0 as well. Since A = 0, the Chern connection is given by the usual Christoffel symbols of a. Also, the Chern and Berwald connections coincide for Riemannian ˙ metrics because they differ merely by A. Here is a sketch of the derivation of the split and covariantised formula. First we replace G by G + ζ in Berwald’s original formula, obtaining K i k = Ki k + 2(ζ i )xk − (ζ i )yj (ζ j )yk − y j (ζ i )xj yk + 2ζ j (ζ i )yj yk − N i j (ζ j )yk − (ζ i )yj N j k + 2G j (ζ i )yj yk + 2ζ j (G i )yj yk . (†) ˜ denote the Chern connection of F. Then horizontal covariant Next, let Γ differentiation of ζ is ˜ i lj , ζ i :j = (ζ i )xj − N l j (ζ i )yl + ζ l Γ
214
DAVID BAO AND COLLEEN ROBLES
from which we solve for (ζ i )xj in terms of ζ i :j . The result is used to replace (“covariantise”) all the xderivatives in (†). The outcome reads K i k = Ki k + 2(ζ i ):k − (ζ i )yj (ζ j )yk − y j (ζ i :j )yk + 2ζ j (ζ i )yj yk ˜ i lk + 2(ζ i )yl N l k + y j (ζ l )yk Γ ˜ i lj + y j ζ l (Γ ˜ i lj )yk − 2ζ l Γ − y j (N l j )yk (ζ i )yl − y j N l j (ζ i )yl yk − N i j (ζ j )yk − (ζ i )yj N j k + 2G j (ζ i )yj yk + 2ζ j (G i )yj yk . It remains to check that the last ten terms on the above righthand side actually coalesce into the single term 3ζ j A˙ i jk . To that end, homogeneity and Euler’s theorem enable us to make the substitutions y j (N l j )yk = (N l j y j )yk − N l k = 2(G l )yk − N l k = N l k , ˜ i lj )yk = (Γ ˜ i lj y j )yk − Γ ˜ i lk = (N i l )yk − Γ ˜ i lk , y j (Γ ˜ i lj = y j Γ ˜ i jl = N i j , yj Γ y j N l j = 2G l . After some cancellations, those final ten terms consolidate into ˜ i jl )yk = 3ζ j (G i )yj yk − Γ ˜ i jk = 3ζ j A˙ i jk . 3ζ j y l (Γ
1.3. Ricci curvature. The importance of Ricci curvatures (defined below, in Section 1.3.1) can be seen from the following Bonnet–Myers theorem: Let (M, F ) be a forwardcomplete connected Finsler manifold of dimension n. Suppose its Ricci curvature has the uniform positive lower bound Ric > (n − 1)λ > 0; equivalently, y i y j Ricij (x, y) > (n − 1)λF 2 (x, y), with λ > 0. Then: √ (i) Every geodesic of length at least π/√ λ contains conjugate points. (ii) The diameter of M is at most π/ λ. (iii) M is in fact compact. (iv) The fundamental group π(M, x) is finite. The Riemannian version of this result is one of the most useful comparison theorems in differential geometry; see [Cheeger–Ebin 1975]. It was first extended to Finsler manifolds in [Auslander 1955]. See [Bao et al. 2000] for a leisurely exposition and references. 1.3.1. Ricci scalar and Ricci tensor. Our geometric definition of the Ricci curvature begins with K(x, y, V ) = V i Rik V k / gy (V, V ) − gy (`, V )2 , a formula for the flag curvature (Section 1.2.1). If, with respect to gy , the transverse edge V has unit length and is orthogonal to the flagpole y, that formula simplifies to K(x, y, V ) = V i Rik V k .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
215
Using gy to measure angles and length, we take any collection of n−1 orthonormal transverse edges {eν : ν = 1, . . . , n − 1} perpendicular to the flagpole. They give rise to n − 1 flags whose flag curvatures are K(x, y, e ν ) = (eν )i Rik (eν )k . The inclusion of en := ` = y/F completes our collection into a gy orthonormal basis B for Tx M . Note that K(x, y, eν ) is simply Rνν (no sum), the (ν, ν) component of the tensor Rik with respect to B. Also, as mentioned in Section 1.2.1, `i Rik `k = 0. Thus Rnn = 0 with respect to the orthonormal basis B. Define, geometrically, the Ricci scalar Ric(x, y) as the sum of those n − 1 flag curvatures K(x, y, eν ). Then Ric(x, y) :=
n−1 X
Rνν =
ν=1
n X
Raa = Ra a = Ri i =
1 1 j i l (y Rj il y ) = 2 K i i , F2 F
a=1
where the last equality follows from Section 1.2.3. Remarks. 1. The indices on R are to be manipulated by the fundamental tensor, and the latter is the Kronecker delta in the gy orthonormal basis B. Thus each component Raa has the same numerical value as Ra a (no sum). 2. The fact that Rik is a tensor ensures that its trace is independent of the basis used to carry that out. Hence Ra a = Ri i . Consequently, the definition of the Ricci scalar is independent of the choice of those n−1 orthonormal transverse edges. 3. The invariance of Rj i kl under positive rescaling in y makes clear that Ric(x, y) has the same property. It is therefore meant to be a function on the projectivised sphere bundle of (M, F ), but could just as well live on the slit tangent bundle T M r 0. In any case, being a function justifies the name scalar. We obtain the Ricci tensor from the Ricci scalar as follows: Ricij := ( 21 F 2 Ric)yi yj = 21 (y k Rk s sl y l )yi yj . This definition, due to AkbarZadeh, is motivated by the fact that, when F arises from any Riemannian metric a, the curvature tensor depends on x alone and the yHessian in question reduces to the familiar expression aRi s sj , which is a Ricij . The Ricci tensor has the same geometrical content as the Ricci scalar. It can be shown that Ric = `i `k Ricik , Ricik = gik Ric + 34 (`i Ric ;k + `k Ric ;i ) + 41 (Ric ;i;k + Ric ;k;i ), where the semicolon means F ∂y . See [Bao et al. 2000]. 1.3.2. Einstein metrics. We defined the Ricci scalar Ric as the sum of n − 1 appropriately chosen flag curvatures. We showed in the last section that this sum depends only on the position x and the flagpole y, not on the specific n − 1 flags with transverse edges orthogonal to y. Thus it is legitimate to think of Ric as n−1 times the average flag curvature at x in the direction y. In Riemannian
216
DAVID BAO AND COLLEEN ROBLES
geometry this is the average sectional curvature among sections spanned by y and a vector orthogonal to y; so generically the result again depends on both x and y. Using the above perspective, it would seem quite remarkable if the said average does not depend on the flagpole y. Finsler metrics F with such a property, namely Ric = (n − 1)K(x) for some function K on M , are called Einstein metrics. This nomenclature is due to AkbarZadeh. The reciprocal relationship (Section 1.3.1) between the Ricci scalar and the Ricci tensor tells us that Ric = (n − 1)K(x)
⇐⇒
Ricij = (n − 1)K(x)gij .
Going one step further, if that average does not depend on the location x either, F is said to be Ricciconstant; in this case, the function K is constant. Remarks. 1. Every Riemannian surface is Einstein (because Ric equals the familiar Gaussian curvature K(x)), but not necessarily Ricciconstant. On the other hand, Finsler surfaces are typically not Einstein, with counterexamples provided by the Numata metrics in Section 1.2.1. 2. In dimension at least 3, a Schur type lemma ensures that every Riemannian Einstein metric is necessarily Ricciconstant. The proof uses the second Bianchi identity for the Riemann curvature tensor. 3. It is not known at the moment whether such a Schur lemma holds for Finsler Einstein metrics in general. However, if we restrict our Finsler metrics to those of Randers type, then there is indeed a Schur lemma for dimM > 3; see [Robles 2003] and Section 3.3.1. A good number of nonRiemannian Einstein metrics and Ricciconstant metrics are presented in Section 4. It follows immediately from our geometric definition of Ric that every Finsler metric of constant flag curvature K must be Einstein with constant Ricci scalar (n − 1)K. As a consistency check, we derive the same fact from the constant flag curvature criterion in Section 1.2.1, R i k = K(δ i k − y i gyk /F 2 ). Indeed, tracing on i and k, and noting that y i gyi = y i y j gij = F 2 , we get Ric = Ri i = (n − 1)K. 1.3.3. Known rigidity results and topological obstructions. We summarise a few basic results on Riemannian Einstein manifolds. References for this material include [Besse 1987; LeBrun–Wang 1999]. In two dimensions: Every 2dimensional manifold M admits a complete Riemannian metric of constant (Gaussian) curvature, which is therefore Einstein. The construction involves gluing together manifolds with boundary, and is due to Thurston; see [Besse 1987] for a summary. If M is compact, there is a variational approach. In [Berger 1971], existence was established by starting with any Riemannian metric on M and deforming it conformally to one with constant curvature. For an exposition of the hard analysis behind this socalled (Melvyn) Berger problem, see [Aubin 1998].
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
217
In three dimensions: For 3dimensional Riemannian manifolds (M, h), the Weyl conformal curvature tensor is automatically zero. This has the immediate consequence that h is Einstein if and only if it is of constant sectional curvature. Though such rigidity extends to Finsler metrics of Randers type (Section 3.3.2), it is not known whether the same holds for arbitrary Finsler metrics. The said rigidity in the Riemannian setting precludes some topological manifolds from admitting Einstein metrics. Take, for example, M = S 2 × S 1 . Were M to admit an Einstein metric h, the latter would perforce be of constant sectional curvature. Now M is compact, so h is complete and Hopf’s classification of Riemannian space forms implies that the universal cover of M is either compact or contractible. This is a contradiction because the universal cover of M is S 2 ×R. In four dimensions: The basic result is the Hitchin–Thorpe Inequality [Hitchin 1974; Thorpe 1969; LeBrun 1999]: If a smooth compact oriented 4dimensional manifold M admits an Einstein metric, then χ(M ) > 23 τ (M ). Here χ(M ) is the Euler characteristic, τ (M ) = 31 p1 (M ) is the signature, and p1 (M ) is the first Pontryagin number. The key lies in the following formulae peculiar to four dimensions [Besse 1987]: Z √ 1 τ (M ) = (W + 2 − W − 2 ) h dx, 2 12π M Z √ 1 2 + 2 − 2 h dx, (S + W  + W  ) χ(M ) = 8π 2 M where S and W are respectively the scalar curvature part and the Weyl part of the Riemann curvature tensor of h. In the second formula, the fact that h is Einstein has already been used to zero out an otherwise negative term from the integrand. The complex projective space CP 2 has the Fubini–Study metric, which is Einstein; there, τ = 1 and χ = 3. On the other hand, the connected sum of 4 or more copies of CP 2 fails the inequality and hence cannot admit any Einstein metric. Finally, the said inequality is not sufficient. See [LeBrun 1999] for compact simply connected 4manifolds which satisfy χ > 32 τ , but which do not admit Einstein metrics. In three dimensions, rigidity equips us with all the tools available to space forms. In particular, there are wellunderstood universal models. Without this structure in dimensions at least 4, the analysis of Einstein metrics becomes considerably more difficult. The saving grace in four dimensions may be attributed to the fact [Singer–Thorpe 1969] that a 4manifold is Einstein if and only if its curvature operator (as a selfadjoint linear operator on 2forms) commutes with the Hodge star operator. This is the property which leads us to the Hitchin– Thorpe Inequality. The tools above do not apply in dimensions greater than four, where there are no known topological obstructions.
218
DAVID BAO AND COLLEEN ROBLES
2. Randers Metrics in Their Defining Form 2.1. Basics 2.1.1. Definition and examples. Randers metrics were introduced by Randers [1941] in the context of general relativity, and later named by Ingarden [1957]. In the positive definite category, they are Finsler spaces built from • a Riemannian metric a := aij dxi ⊗ dxj , and • a 1form b := bi dxi , with equivalent description b] := bi ∂xi , both living globally on the smooth ndimensional manifold M . The Finsler function of a Randers metric has the simple form F = α + β, where q α(x, y) := aij (x)y i y j , β(x, y) := bi (x)y i . Generic Randers metrics are only positively homogeneous. No Randers metric can satisfy absolute homogeneity F (x, cy) = cF (x, y) unless b = 0, in which case it is Riemannian. Examples. The Zermelo navigation metric (page 201) is a Randers metric with defining data λhij + Wi Wj −Wi and bi = , where λ := 1 − h(W, W ). λ2 λ Matsumoto’s slopeofamountain metric (page 201) is not of Randers type. This is because it has the form F = α2 /(α − β), where α comes from the Riemannian metric on the graph of a certain function f , and β = df . Of the examples in Section 1.2.1, a subclass of the Numata metrics — those with constant qij (which serves as our aij ) and closed 1forms b — is Randers, while Bryant’s metrics are manifestly not Randers. ♦ aij =
2.1.2. Criterion for strong convexity. In order that F = α + β : T M → R be a Finsler function, it must be nonnegative, regular, positively homogeneous, and strongly convex (Section 1.1.1). Regularity and positive homogeneity can be established by inspection. Strong convexity concerns the positive definiteness of the fundamental tensor, which for Randers metrics is
a a a yj yi yi ayj F + bi + bj , where ayi := aij y j . aij − + α α α α α It turns out that the following three criteria are equivalent:
gij =
(1) The anorm kbk of b is strictly less than 1 on M . (2) F (x, y) is positive for all y 6= 0. (3) The fundamental tensor gij (x, y) is positive definite at all y 6= 0. Proof: (1) =⇒ (2). Suppose kbk < 1. A Cauchy–Schwarz type argument gives q ±β 6 β = bi y i  6 kbkkyk < 1 · aij y i y j = α.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
219
In particular, F = α + β is positive. (2) =⇒ (3). (As stressed in Section 1.1.1, this is false for general Finsler metrics.) Suppose F = α + β is positive. Then so is Ft := α + tβ, where t 6 1. Let gt denote the fundamental tensor of Ft . Starting with the cited formula for the fundamental tensor of Randers metrics, a standard matrix identity gives det(gt ) =
F n+1 t
α
det(a).
For a leisurely treatment, see [Bao et al. 2000]. This tells us that g t has positive determinant, hence none of its eigenvalues can vanish. These eigenvalues depend continuously on t. At t = 0, they are all positive because g 0 = a is Riemannian. If any eigenvalue were to become nonpositive, it would have to go through zero at some t, in which case det(gt ) could not possibly remain positive. Thus all eigenvalues stay positive; in particular, g = g1 is positive definite. (3) =⇒ (2). As in Section 1.1.1, this follows from F 2 (x, y) = gij (x, y)y i y j . (2) =⇒ (1). Suppose F is positive. Then F (x, −b(x)) = kbk(1 − kbk) forces kbk < 1 wherever b(x) 6= 0. At points where b(x) vanishes, the said inequality certainly holds. 2.1.3. Explicit formula of the spray curvature. Let aK i k denote the spray curvature tensor of the Riemannian metric a. p Then the spray curvature tensor K ik of the Randers metric F (x, y) := α + β = aij (x)y i y j + bi (x)y i can be expressed in terms of aK i j and the quantities lieij := bij + bji , a
yi := aij y j ,
curlij := bij − bji , ξ :=
1 2
lie00 − αθ0 ,
θj := bi curlij , ξ0 :=
1 2
lie000 − αθ00 ,
through the use of Berwald’s formula (Section 1.2.3) in a split and covariantised form. When applying Section 1.2.3, set the background metric F to be the Riemannian a, with Christoffel symbols aγ i jk , and let  instead of : denote the corresponding covariant differentiation. We also need the fact, derived in [Bao et al. 2000], that Gi = aGi + ζ i with 2ζ i = (y i /F )ξ + α curli 0 . The resulting formula for the spray curvature (also independently obtained by Shen) reads K i k = aK i k + yyCoeff y i ayk + ybCoeff y i bk + δCoeff δ i k + 14 curli j curlj 0 ayk − 41 α2 curli j curlj k + 34 curli 0 curlk0 + 14 (α2 /F )y i θj curlj k − 34 (1/F )y i θ0 curlk0 + 21 (α/F )y i curlj 0 liejk − 41 (α/F )y i liej0 curlj k + α curli 0k − 12 α curli k0 − 21 (1/α) curli 00 ayk + 12 (α/F )y i θk0 − (α/F )y i θ0k + 21 (1/F )y i lie00k − 12 (1/F )y i liek00 .
220
DAVID BAO AND COLLEEN ROBLES
The three suppressed coefficients are yyCoeff := α/(2F 2 ) − 1/(4F ) curlj 0 θj − 1/(2F 2 ) + 1/(4F α) curlj 0 liej0
+ 21 θ00 /(F α) − 43 ξ 2 /(F 3 α) + 12 ξ0 /(F 2 α),
ybCoeff := 21 (α2/F 2 ) curlj 0 θj − 21 (α/F 2 ) curlj 0 liej0 − 34 (1/F 3 )ξ 2 + 21 (1/F 2 )ξ0 , δCoeff := − 12 (α2/F ) curlj 0 θj + 21 (α/F ) curlj 0 liej0 + 34 (1/F 2 )ξ 2 − 12 (1/F )ξ0 . Remark. Covariant differentiation with respect to the Riemannian metric a, indicated by our vertical slash, can be lifted horizontally to T M r 0, using the nonlinear connection and the Christoffel symbols of a. The section y of π ∗ T M then satisfies y i k = 0; see Section 1.1.3. So, in the above expressions, we can interpret the subscript 0 as contraction with y either before or after the vertical slash has been carried out, with no difference in the outcome. 2.2. Characterising Einstein–Randers metrics. In this section we derive necessary and sufficient conditions on a and b for the Randers metric to be Einstein. Recall that F is Einstein with Ricci scalar Ric(x) if and only if K i i = Ric(x)F 2 (Section 1.3.2). We begin by assuming that this equality holds, and deduce the necessary conditions for the metric to be Einstein. Then we show that these necessary conditions are also sufficient. Compute K i i by tracing the expression for K i k in Section 2.1.3 to arrive at 0 = K i i − F 2 Ric(x) α 1 θ00 − 41 (n − 1) lie000 F F α α2 i 1 1 + 2 (n − 1) curl 0 liei0 − 2 (n − 1) θi curli0 + 21 curli 0 curli0 F F 1 α ij 3 1 2 + 4 α curl curlij + 16 (n − 1) 2 (lie00 )2 − 34 (n − 1) 2 lie00 θ0 F F α2 + 34 (n − 1) 2 (θ0 )2 − F 2 Ric(x). F
= a Ric00 +α curli 0i + 21 (n − 1)
Here, we have used the fact that aK i k , the spray curvature of the Riemannian metric a, is related to the latter’s Riemann tensor via aK i k = y j aRj i kl y l , as shown in Section 1.2.3. Hence aK i i = y j aRj i il y l = y j a Ricjl y l = a Ric00 . Multiplying this displayed equation by F 2 removes y from the denominators. The criterion for a Randers metric to be Einstein then takes the form q Rat + αIrrat = 0, where α := aij (x)y i y j . Here Rat and Irrat are homogeneous polynomials in y, of degree 4 and 3 respectively, whose coefficients are functions of x. Their formulae are given below. As observed by Crampin, the displayed equation becomes Rat − αIrrat = 0 if we replace y by −y. The two equations then effect Rat = 0 and αIrrat = 0.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
221
Being homogeneous in y, Irrat certainly vanishes at y = 0. At nonzero y, we have α > 0 because aij is positivedefinite. Hence Irrat = 0. p Lemma 1. Let F (x, y) := aij (x)y i y j + bi (x)y i be a Randers metric with positive definite (i .e., Riemannian) aij . Then F is Einstein if and only if Rat = 0 and Irrat = 0. The formulae for Rat and Irrat are Rat = (α2 + β 2 ) a Ric00 +2α2 β curli 0i + 21 (α2 + β 2 ) curli 0 curli0 + 41 α2 (α2 + β 2 ) curlij curlij − (α4 + 6α2 β 2 + β 4 ) Ric(x) + 12 (n − 1) α2 θ00 − 12 β lie000 + α2 curli 0 liei0 − α2 β θi curli0 + 83 (lie00 )2 + 32 α2 (θ0 )2 ,
Irrat = 2β a Ric00 +(α2 + β 2 ) curli 0i + β curli 0 curli0 + 21 α2 β curlij curlij − 4β(α2 + β 2 ) Ric(x)
+ 21 (n − 1) β θ00 − 12 lie000 + β curli 0 liei0 − α2 θi curli0 − 32 lie00 θ0 .
From these two expressions we will derive the preliminary form of three necessary and sufficient conditions for a Randers metric to be Einstein. 2.2.1. Preliminary form of the characterisation. Assume F is Einstein, so that Rat = 0 and Irrat = 0. For convenience abbreviate Ric(x) by Ric. Then the weaker statement Rat − β Irrat = 0 certainly holds, and reads 0 = (α2 −β 2 ) a Ric00 +β curli 0i + 21 curli 0 curli0 + 14 α2 curlij curlij −(α2 +3β 2 ) Ric + 21 (n − 1)(curli 0 liei0 + 32 (θ0 )2 + θ00 ) 2 3 (n − 1) lie00 + 2β θ0 . + 16
Fix x. Considering the righthand side as a polynomial in y, we see that α 2 − β 2 divides (lie00 +2βθ0 )2 . The polynomial α2 −β 2 is irreducible, because if it were to factor — necessarily into two linear terms — its zero set would contain a hyperplane, contradicting the strong convexity condition (kbk < 1), which requires that it be positive at all y 6= 0 (Section 2.1.2). Being irreducible, α2 − β 2 must divide not just the square but lie00 + 2βθ0 itself. Thus there exists a scalar function σ(x) on M such that lie00 + 2β θ0 = σ(x)(α2 − β 2 ).
This is our Basic Equation, the first necessary condition for a Randers metric to be Einstein. Differentiating with respect to y i and y k gives an equivalent version: lieik + bi θk + bk θi = σ(x)(aik − bi bk ). To recover the original version, just contract this with y i y k . The Basic Equation is equivalent to the statement that the Scurvature of the Randers metric F is given by S = 41 (n + 1)σ(x)F ; see [Chen–Shen 2003].
222
DAVID BAO AND COLLEEN ROBLES
Now return to the expression for 0 = Rat−β Irrat. We use the Basic Equation to replace lie00 +2βθ0 with σ(x)(α2 −β 2 ), and then divide off by a uniform factor of α2 − β 2 . The result reads a
Ric00 = (α2 + 3β 2 ) Ric −β curlj 0j − 41 α2 curlhj curlhj − 12 curlj 0 curlj0 2 3 2 8 σ (x)(α
− 12 (n − 1)
− β 2 ) + curlj 0 liej0 + 32 (θ0 )2 + θ00 .
This is the Ricci Curvature Equation, so named because it describes the Ricci tensor of a. We obtain the indexed version by differentiating with respect to y i and y k , and making use of the symmetry a Ricik = a Ricki . a
Ricik = (aik + 3bi bk ) Ric − 21 (bi curlj kj + bk curlj ij ) − 14 aik curlhj curlhj − 12 curlj i curljk − 21 (n − 1)
3 2 8 σ (x)(aik
− bi bk ) + 21 (curlj i liejk + curlj k lieji ) + 32 θi θk + 12 (θik + θki ) .
From the Basic and Ricci Curvature Equations we derive the final characterising condition, which we call the E23 Equation (the number 23 being of some chronological significance in our research notes). Two pieces of information from the Basic Equation are required. To reduce clutter, abbreviate σ(x) as σ. First, differentiate to obtain lie000 = σ0 (α2 − β 2 ) − lie00 (σβ + θ0 ) − 2β θ00 . Next, contract the indexed form of the Basic Equation with y i curlk 0 to get curlj 0 liej0 = −β θ j curlj0 − (θ0 )2 − σβ θ0 .
Return to the equation 0 = Irrat. Replace the term a Ric00 by the righthand side of the Ricci Curvature Equation. Then, wherever possible, insert the expressions for lie00 , lie000 and curlj 0 liej0 given by the Basic Equation. After dividing off a factor of α2 − β 2 , we obtain the E23 Equation: (E23 ) curlj 0j = 2 Ric β + (n − 1) 18 σ 2 β + 12 σθ0 + 21 θj curlj0 + 41 σ0 . Again, differentiating by y i produces the indexed version curlj ij = 2 Ric bi + (n − 1)
1 2 8 σ bi
+ 21 σθi + 12 θj curlji + 14 σi .
The Basic, Ricci Curvature and E23 Equations are all necessary conditions for the Randers metric F to be Einstein. Together, they are also sufficient. In view of Lemma 1 (page 221), we can demonstrate this by showing that they imply Rat = 0 = Irrat. Recall that we deduced the E23 Equation from Irrat = 0 by ◦ expressing a Ric00 via the Ricci Curvature Equation, ◦ computing lie00 , lie000 and curlj 0 liej0 with the Basic Equation, and ◦ dividing by a uniform factor of α2 − β 2 .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
223
Reversing these three algebraic steps allows us to recover Irrat = 0 from the E 23 Equation. Likewise, the Ricci Curvature Equation came from Rat − β Irrat = 0 by ◦ using the Basic Equation to replace lie00 + 2βθ0 with σ(α2 − β 2 ), and ◦ dividing by α2 − β 2 . Again, reversing the two steps above will give us Rat − β Irrat = 0, whence Rat = 0 because Irrat = 0. To summarise, the Basic Equation, the Ricci Curvature Equation and the E23 Equation characterise strongly convex Einstein Randers metrics. In the next section we will refine the three characterising equations by showing that σ must be constant. 2.2.2. Constancy of the Scurvature. In the previous section we commented that the Scurvature of any Randers metric F satisfying the Basic Equation is given by S = 14 (n + 1)σ(x)F [Chen–Shen 2003]. The Scurvature is positively homogeneous of degree 1 in y. In Section 1 we demonstrated a strong preference for working with objects that are positively homogeneous of degree zero in y. That is because such objects naturally live on the projectivised sphere bundle SM as well as the larger slit tangent bundle T M r 0. The compact parameter space provided by the sphere bundle is generally better suited for global and analytic considerations. So the object we are really interested in is not S, but S/F , which is homogeneous of degree zero in y. In this context, when we say that the Scurvature of any Randers metric satisfying the Basic Equation is isotropic, we mean that the quotient FS is a function of x alone. Similarly, when σ(x) is constant, F is said to be a metric of constant Scurvature. The following lemma plays a crucial role in establishing the constancy of the Scurvature for Einstein Randers metrics. Lemma 2. The covariant derivative of the tensor curl associated to any Randers metric is given by curlijk = −2bs aRksij + lieikj − liekji . Proof. Using Ricci identities and the definition of lie ij we have bijk − bikj =
bs aRisjk ,
bikj + bkij =
lieikj ,
−bkij + bkji = −bs aRksij , −bkji − bjki = −liekji , bjki − bjik =
bs aRjski .
Summing these five equalities and applying the first Bianchi identity produces the desired formula.
224
DAVID BAO AND COLLEEN ROBLES
Proposition 3. Let F be a strongly convex Randers metric on a connected manifold , satisfying the Basic Equation (with σ a function of x) and the Ricci Curvature Equation. Then F is of constant Scurvature (i .e., σ is constant) if and only if the E23 Equation holds. In practice, the E23 Equation has proved to be remarkably useful. Proposition 3 shows us that such efficacy is attributable to the constancy of the Scurvature. Also, since strongly convex Einstein Randers metrics satisfy the Basic, Ricci Curvature and E23 Equations, the following corollary is immediate. Corollary 4. Any strongly convex Einstein Randers metric on a connected manifold is necessarily of constant Scurvature. Proof of Proposition. The key is to compute a formula for the tensor curl i 0i . Lemma 2 plays a pivotal role. We first contract that lemma with a ik y j to obtain curli 0i = 2bi a Rici0 + liei i0 − liei 0i ,
(∗)
a preliminary formula around which all further analysis is centered. Another contracted version of Lemma 2 will be needed when we calculate a certain term in 2bi a Rici0 and − liei 0i . Before plunging into details, here is an outline: If σ is constant, we use the Basic and Ricci Curvature Equations to finish calculating the righthand side of (∗). A third contracted version of Lemma 2 will come into play. The outcome is none other than the E 23 Equation. Conversely, if the E23 Equation is presumed to hold, we immediately get one formula for curli 0i . We use the Basic, Ricci Curvature, and E23 Equations to finish calculating the righthand side of (∗), thereby deducing a second formula for curli 0i . A comparison of the two then tells us that σ is constant. Now for the calculations. We first reexpress the terms in the righthand side of (∗). The last two terms are handled using the Basic Equation: liei i0 = (n − kbk2 )σ0 − σ (1 − kbk2 )(σβ + θ0 ), liei 0i = σ0 − β bi σi − 21 σ 2 (n − 2kbk2 + 1)β + 21 σ(2kbk2 − n)θ0 + 12 βθi θi + 21 θi curli0 − βθ i i − bi θ0i . The remaining term, 2bi a Rici0 , is handled by the Ricci Curvature Equation: 2bi a Rici0 = θi curli0 − (n−1) 14 kbk2 σ0 + 21 bi (θi0 + θ0i ) +β (2(1 + kbk2 ) Ric − 12 curlij curlij − (n−1)
2 1 i 1 2 8 σ (3 − kbk ) + 4 b σi
).
Next we compute the quantities bi θi0 , bi θ0i , and θ i i that occur in these formulae. We will use without explicit mention the equalities bij = 12 (lieij + curlij )
and
bij curlij =
1 2
curlij curlij .
For bi θi0 , notice that bi θi = bi bj curlij = 0, because curlij is skewsymmetric. Differentiating bi θi = 0 and using the Basic Equation gives bi θi0 = 21 θi θi β − 21 σθ0 − 12 θi curli0 .
(∗∗)
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
225
To compute bi θ0i , expand it as 12 bi (lieik − curlik ) curlk 0 + bi bk curli0k . By Lemma 2, bi bk curli0k = bi bk (−2bj aRijk0 + lieik0 + liei0k ). However, bi bj aRijk0 vanishes because aR is skewsymmetric in the first two indices. Hence bi θ0i = 21 bi (lieik − curlik ) curlk 0 + bi bk (lieik0 + liei0k ). Our calculation of bi θ0i can now be completed in three steps as follows. ◦ Use the Basic Equation to remove all occurrences of the tensor lie and its covariant derivatives. ◦ Replace the bi θi0 term, which resurfaces twice, by the righthand side of (∗∗). ◦ Simplification leads to an expression of the form (1 − kbk 2 )( · · · ) for the quantity (1 − kbk2 )bi θ0i . Strong convexity (Section 2.1.2) allows us to divide both sides by 1 − kbk2 to obtain bi θ0i = 12 θi θi β + 21 σθ0 − 21 θi curli0 + kbk2 σ0 − bi σi β. The last term of interest, θ i i , is computed separately for each direction of the proof. First, if we assume that F is of constant Scurvature (σ is constant), the Ricci Curvature Equation simplifies. Carrying out an appropriate trace on Lemma 2, followed by contracting with b and rearranging, we find that θi i =
1 2
curlij curlij − bi (2bj a Ricij + liej ji − liej ij ).
Now replace the Ricci tensor by the righthand side of the Ricci Curvature Equation, and use the Basic Equation (with σ constant) to remove all occurrences of lie and its covariant derivatives. After simplifying we find that each term contains a factor of 1 + kbk2 (not the 1 − kbk2 occurring previously). Dividing out by that factor gives θi i = 21 curlij curlij − 2 Ric + 18 (n − 1)σ 2 kbk2 + 21 (n − 1)θi θi . (†)
Conversely, assume that the E23 Equation holds. Without the constancy of σ it is not possible to compute θ i i via Lemma 2. Happily, the hypothesised E23 Equation saves the day: 1 2
curlij curlij − bi curlj ij = 12 curlij curlij − 2 Ric + 81 (n−1)σ 2 kbk2 + 21 (n−1)θi θi − 41 (n−1)bi σi . (‡)
θi i = (bi curlij )j =
Remark. This is the only place where the E23 Equation gets used in the proof. We are now ready to complete the proof. Consider the expressions for 2b i a Rici0 , liei i0 , and − liei 0i found on the previous page. By (∗), the sum of the three is curli 0i . Now substitute into this sum the formulae for bi θi0 , bi θ0i , and θ i i just found, and simplify. Under the hypothesis that F is of constant Scurvature, using (†) for the value of θ i i , we get curli 0i = 2 Ric β + (n − 1) 18 σ 2 β + 21 σθ0 + 21 θi curli0 .
226
DAVID BAO AND COLLEEN ROBLES
This is the E23 Equation (with σ constant, i.e. σi = 0). If instead we assume that F satisfies the E23 Equation, and use (‡) as the value of θ i i , we obtain curli 0i = 2 Ric β + (n − 1) 18 σ 2 β + 21 σθ0 + 12 θi curli0 + σ0 − 43 kbk2 σ0 .
Comparing this formula for curli 0i with the one given by the E23 Equation indicates that 34 (1 − kbk2 )σ0 = 0. Since kbk < 1, we must have σ0 = 0; equivalently, all covariant derivatives σi vanish. But σ is a function of x, so all its partial derivatives are zero. Therefore σ is constant on the connected M . 2.2.3. Final characterisation of Einstein–Randers metrics. In Section 2.2.1 we showed that strongly convex Einstein Randers metrics are characterised by the preliminary form of the Basic, Ricci Curvature and E 23 Equations. The constancy of σ, established in Corollary 4, can now be used to refine these conditions to their final form. Let’s begin with the final form of the Basic Equation. Since the equation involves no derivatives of σ, it undergoes little cosmetic alteration. We simply write σ instead of σ(x) to emphasise the constancy of the function: lie00 + 2β θ0 = σ (α2 − β 2 ). Equivalently, the indexed form reads lieik + bi θk + bk θi = σ (aik − bi bk ). The final form of the Ricci Curvature Equation is derived in two steps. First, use the Basic Equation above to remove the tensor lie and its covariant derivatives from the preliminary expression in Section 2.2.1. Then replace the curl j 0j term with the formula given by the E23 Equation in Section 2.2.1. Keep in mind that the covariant derivatives σi vanish because σ is constant. After simplifying, the result is a
Ric00 = (α2 + β 2 ) Ric(x) − 41 α2 curlhj curlhj − 12 curlj 0 curlj0 −(n − 1)
2 1 2 16 σ (3α
− β 2 ) + 41 (θ0 )2 + 12 θ00 .
Differentiating by y i and y k and applying the symmetry of a Ricik produces the indexed version a
Ricik = (aik + bi bk ) Ric(x) − 41 aik curlhj curlhj − 21 curlj i curljk −(n − 1)
1 2 16 σ (3aik
− bi bk ) + 14 θi θk + 41 (θik + θki ) .
The constancy of σ updates the E23 Equation to curlj0j = 2 Ric(x)β + (n − 1)
1 2 8σ β
+ 21 σθ0 + 12 θj curlj0 ,
curljij = 2 Ric(x)bi + (n − 1)
1 2 8 σ bi
+ 21 σθi + 12 θj curlji .
or
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
227
Remark. The final forms of the Basic, E23 and Ricci Curvature Equations are equivalent to the preliminary forms. In the case of the Basic and E 23 Equations, this follows immediately from the constancy of σ. As for the Ricci Curvature Equation, its final form was deduced from the preliminary form by replacing the terms lie00 , lie000 , curlj 0 liej0 and curlj 0j with the expressions given by the Basic and E23 Equations. Reversing this algebraic substitution resurrects the preliminary form of the Ricci Curvature Equation. We saw in Section 2.2.1 that the preliminary forms characterise strongly convex Einstein metrics. Therefore the final forms of the Basic, Ricci Curvature and E 23 Equations are necessary and sufficient conditions for the metric to be Einstein. Moreover, Proposition 3 assures us that, with σ constant, the Basic and Ricci Curvature Equations alone do the trick. Theorem 5 (Einstein Characterisation). Let F = α + β be a strongly convex Randers metric on a smooth manifold M of dimension n > 2, with α2 = aij (x)y i y j and β = bi (x)y i . Then (M, F ) is Einstein with Ricci scalar Ric(x) if and only if the Basic Equation lieik + bi θk + bk θi = σ(aik − bi bk ) and the Ricci Curvature Equation a
Ricik = (aik + bi bk ) Ric(x) − 41 aik curlhj curlhj − 21 curlj i curljk −(n − 1)
1 2 16 σ (3aik
− bi bk ) + 41 θi θk + 14 (θik + θki )
are satisfied for some constant σ. Tracing the Basic Equation tells us that σ, besides being related to the Scurvature (Section 2.2.2), also has the geometrically significant value σ=
2divb] , n − kbk2
where divb] := bi i is the divergence of the vector field b] := bi ∂xi . Remark. The Basic Equation, Ricci Curvature Equation, and E 23 Equation are tensorial equations, and highly nonlinear due to the presence of a Ricik . They constitute a coupled system of second order partial differential equations. Their redeeming feature is being polynomial in the tangent space coordinates y i , whereas the original Einstein criterion is not (unless b = 0). This greatly reduces computational complexity. While testing Randers metrics to see whether they satisfy the Einstein criterion K i i = F 2 Ric(x), we encountered cases in which the software Maple was unable to complete computations of K i i . So, for those examples we could not directly verify the Einstein criterion. Maple was able, however, to work efficiently with the three indirect characterising equations.
228
DAVID BAO AND COLLEEN ROBLES
2.3. Characterising constant flag curvature Randers metrics 2.3.1. The result. Recall from Section 1.3.1 that the Ricci scalar Ric is the sum of n − 1 appropriately chosen flag curvatures. Thus, Finsler metrics of constant flag curvature K necessarily have constant Ricci scalar (n − 1)K, and are therefore Einstein. By Corollary 4, they must have constant Scurvature. Computationally, the equation K i k = KF 2 (δ i k −`i `k ) characterizing constant flag curvature is even more challenging than the Einstein equation, which already gives Maple trouble (see end of previous section). The need for machinefriendly characterisation equations is acute. Partly motivated by this, Randers metrics of constant flag curvature have been characterised in [Bao–Robles 2003]. The same conclusion was simultaneously obtained in [Matsumoto–Shimada 2002], albeit by a different method. The result is similar to that described in Theorem 5. Theorem 6 (Constant flag curvature characterisation). Let F = α+β be a strongly convex Randers metric on a smooth manifold M of dimension n > 2, with α2 = aij (x)y i y j and β = bi (x)y i . Then (M, F ) is of constant flag curvature K if and only if there exists a constant σ such that the Basic Equation lieik + bi θk + bk θi = σ (aik − bi bk ) holds and the Riemann tensor of a satisfies the Curvature Equation Rhijk = ξ (aij ahk − aik ahj ) − 14 aij curlt h curltk + 14 aik curlt h curltj
a
+ 41 ahj curlt i curltk − 14 ahk curlt i curltj −
1 4
curlij curlhk + 14 curlik curlhj + 12 curlhi curljk ,
1 2 3 2 σ ) + (K + 16 σ )kbk2 − 41 θi θi . with ξ := (K − 16
2.3.2. Utility. Theorem 6 provides an indirect but efficient way for checking whether a given Randers metric is of constant flag curvature. As we mentioned in Section 2.2.3, tracing the Basic Equation reveals that the constant σ, whenever it exists, is equal to 2 div b] /(n−kbk2 ). Hence, this quotient is one of the first items we must compute. If the answer is not a constant, it is pointless to proceed any further. If the computed value for σ is constant, surviving the Basic Equation constitutes the next checkpoint. After that, we solve the Curvature Equation for K, and see whether it is constant. ´ disc). This metric is implicit in [Okada Example (Finslerian Poincare 1983], and is extensively discussed in [Bao et al. 2000]. Let r and θ denote polar coordinates on the open disc of radius 2 in R2 . The Randers metric in question is defined by the following Riemannian metric a and 1form b: a=
dr ⊗ dr + r 2 dθ ⊗ dθ , (1 − 14 r2 )2
b=
r dr (1 +
1 2 1 2 4 r )(1 − 4 r )
.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
229
Note that a is the Riemannian Poincar´ e model of constant sectional curvature −1, and b = d log (4+r 2 )/(4−r 2 ) is exact. This metric is interesting because its geodesic trajectories agree with those of the Riemannian Poincar´e model a. However, as shown in [Bao et al. 2000], the travel time from the boundary to the center is finite (log 2 seconds), while the return trip takes infinite time! We summarise below three key steps in ascertaining that our Randers metric has constant flag curvature K = − 41 . ◦ The value of the Scurvature σ is computed to be 2. ◦ Since b is exact, it is closed and curl = 0. In particular, θ = 0 and lie ik = 2bik . The Basic Equation is shown to hold with σ = 2. ◦ Since curl = 0 and a has constant curvature −1, the Curvature Equation ♦ reduces to (K + 14 )(1 + kbk2 ) = 0, which gives K = − 41 . Another byproduct of Theorem 6 is the corrected Yasuda–Shimada theorem, proved in [Bao–Robles 2003; Matsumoto–Shimada 2002], and discussed near the end of [Bao et al. 2003]. That theorem characterises, within the family of Randers metrics satisfying θ = 0, those that have constant flag curvature K. (Those with K > 0 and θ = 0 were classified in [Bejancu–Farran 2002; 2003].) Lastly, Theorem 6 provides an important link in the complete classification of constant flag curvature Randers metrics; see [Bao et al. 2003]. 2.3.3. Comparing with the Einstein case. In Sections 1.3.2 and 2.3.1, we pointed out that Finsler metrics of constant flag curvature are necessarily Einstein. In particular, Randers metrics characterised by Theorem 6 should satisfy the criteria stipulated in Theorem 5. This is indeed the case. The two Basic Equations are identical; and tracing the Curvature Equation of Theorem 6 produces the Ricci Curvature Equation of Theorem 5. In [Bao–Robles 2003], the characterisation result includes a third condition, called the CC(23) Equation, which gives a formula for the covariant derivative curlijk of curl. We excluded this equation from our statement of Theorem 6 because, like the E23 Equation, it is automatically satisfied whenever the Basic and Curvature Equations hold with constant σ. Proposition 7. Suppose F is a strongly convex Randers metric on a connected manifold , satisfying the preliminary form of the Basic and Curvature Equations described in [Bao–Robles 2003]. Then F is of constant Scurvature (σ is constant) if and only if the CC (23) Equation of [Bao–Robles 2003] holds. We omit the proof because it is structurally similar to the one we gave for Proposition 3. For constant σ, the CC(23) Equation reads: curlijk = aik (2K + 18 σ 2 )bj + 21 curlh j θh + 12 σ θj −ajk (2K + 81 σ 2 )bi + 12 curlh i θh + 12 σ θi .
Tracing the CC(23) Equation on its first and third indices gives the E 23 Equation (with σ constant) in Section 2.2.1.
230
DAVID BAO AND COLLEEN ROBLES
3. Randers Metrics Through Zermelo Navigation 3.1. Zermelo navigation. Zermelo [1931] posed and answered the following question (see also [Carath´eodory 1999]): Consider a ship sailing on the open sea in calm waters. Suppose a mild breeze comes up. How must the ship be steered in order to reach a given destination in the shortest time? Zermelo assumed that the open sea was R2 with the flat/Euclidean metric. Recently, Shen generalised the problem to the setting where the sea is an arbitrary Riemannian manifold (M, h). Shen [2002] finds that, when the wind is timeindependent, the paths of shortest time are the geodesics of a Randers metric. This will be established in Section 3.1.1. For the remainder of this section, we develop some intuition by considering the problem on the infinitesimal scale. Given any Riemannian metric h on a differentiable manifold M , denote the corresponding normsquared of tangent vectors y ∈ T x M by y2 := hij y i y j = h(y, y). Think of y as measuring the time it takes, using an engine with a fixed power output, to travel from the basepoint of the vector y to its tip. Note the symmetry property −y = y. The unit tangent sphere in each Tx M consists of all those tangent vectors u such that u = 1. Now introduce a vector field W such that W  < 1, the spatial velocity vector of our mild wind on the Riemannian landscape (M, h). Before W sets in, a journey from the base to the tip of any u would take 1 unit of time, say, 1 second. The effect of the wind is to cause the journey to veer off course (or merely off target if u is collinear with W ). Within the same 1 second, we traverse not u but the resultant v = u + W instead. As an example, suppose W  = 12 . If u points along W (that is, u = 2W ), then v = 32 u. Alternatively, if u points opposite to W (u = −2W ), then v = 12 u. In these two scenarios, v equals 32 and 21 instead of 1. So, with the wind present, our Riemannian metric h no longer gives the travel time along vectors. This prompts the introduction of a function F on the tangent bundle T M , to keep track of the travel time needed to traverse tangent vectors y under windy conditions. For all those resultants v = u + W mentioned above, we have F (v) = 1. In other words, within each tangent space Tx M , the unit sphere of F is simply the W translate of the unit sphere of h. Since this W translate is no longer centrally symmetric about the origin 0 of Tx M , the Finsler function F cannot be Riemannian. Given any Finsler manifold (M, F ), the indicatrix in Tx M each tangent space is Sx (F ) := {y ∈ Tx M : F (x, y) = 1}. F =1 The indicatrices of h and the Randers metric F with navigation data (h, W ) are related by a rigid translaWx 0 p tion: Sx (F ) = Sx (h) + Wx . In particular, the Randers h=1 indicatrix is simply an ellipse centered at the tip of W x .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
231
In the following section we will algebraically derive an expression for F , showing that it is a Randers metric. Then we demonstrate that the paths of shortest time are indeed the geodesics of this F . 3.1.1. Algebraic and calculusofvariations aspects. We return to the earlier discussion and consider those u ∈ Tx M with u = 1; equivalently, h(u, u) = 1. Into this, we substitute u = v − W and then h(v, W ) = vW  cos θ. Introducing the abbreviation λ := 1 − W 2 , we have v2 − (2W  cos θ) v − λ = 0. Since W  < 1, p the resultant v is never zero, hence v > 0. This leads to v = W  cos θ+ W 2 cos2 θ + λ, which we abbreviate as p+q. Since F (v) = 1, we see that p [h(W, v)]2 + v2 λ h(W, v) q−p 1 = = v 2 − . F (v) = 1 = v 2 q+p q −p λ λ It remains to deduce F (y) for an arbitrary y ∈ T M . Note that every nonzero y is expressible as a positive multiple c of some v with F (v) = 1. For c > 0, traversing y = cv under the windy conditions should take c seconds. Consequently, F is positively homogeneous: F (y) = cF (v). Using this homogeneity and the formula derived for F (v), we find that p [h(W, y)]2 + y2 λ h(W, y) F (y) = − . λ λ Here, F (y) abbreviates F (x, y); the basepoint x has been suppressed temporarily. p As promised, F isi a Randers metric. Namely, it has the form F (x, y) = aij (x)y i y j + bi (x)y , where a is a Riemannian metric and b a differential 1form. Explicitly, hij Wi Wj −Wi aij = + , bi = . λ λ λ λ Here Wi := hij W j and λ = 1 − W i Wi . In particular, there is a canonical Randers metric associated to each Zermelo navigation problem with data (h, W ). Incidentally, the inverse of a is given by aij = λ(hij − W i W j ),
and bi := aij bj = −λW i .
Under the influence of W , the most efficient navigational paths are no longer the geodesics of the Riemannian metric h; instead, they are the geodesics of the Finsler metric F . To see this, let x(t), for t ∈ [0, τ ], be a curve in M from point p to point q. Return to our imaginary ship sailing about M , with velocity vector u but not necessarily with constant speed. If the ship is to travel along the curve x(t) while the wind blows, the captain must continually adjust the ship’s direction u/u so that the resultant u + W is tangent to x(t). The travel time along any infinitesimal segment x˙ dt of the curve is F (x, x˙ dt) = F (x, x) ˙ dt, because as explained above it is the positively homogeneous F (not h) that keeps
232
DAVID BAO AND COLLEEN ROBLES
track of travel times. The captain’s task is to select a path x(t) from p to q that minimises the total travel time Z τ F (x, x) ˙ dt. 0
This quantity is independent of orientationpreserving parametrisations due to the positive homogeneity of F and the changeofvariables theorem. Such an efficient path is precisely a geodesic of the Finsler metric F , which is said to have solved Zermelo’s problem of navigation under the external influence W . Let’s look at some 2dimensional examples. Being in two dimensions, we revert to the common notation of denoting position coordinates by x, y rather than x1 , x2 , and components of tangent vectors by u, v rather than y 1 , y 2 . Example (Minkowski space). Consider R2 equipped with the Euclidean metric hij = δij . Let W = (p, q) be a constant vector field. The resulting Randers metric is of Minkowski type, and is given by q (pu+qv)2 + (u2 +v 2 ) 1 − (p2 +q 2 ) −(pu + qv) F (x, y; u, v) = + . 1 − (p2 +q 2 ) 1 − (p2 +q 2 ) The condition W 2 = p2 + q 2 < 1 ensures that F is strongly convex. The geodesics of both h and F are straight lines. Thus, when navigating on flatland under the influence of a constant wind, the correct strategy is to steer the ship so that it travels along a straight line. This means the captain should aim the ship, not straight toward the desired destination, but slightly off course with a velocity V , selected so that V + W points at the destination. ♦ Example (Shen’s fish pond). Fish are kept in a pond with a rotational current. The pond occupies the unit disc in R2 , and hij is again the Euclidean metric δij . The current’s velocity field is W := −y∂x + x∂y , which describes a counterclockwise circulation of angular speed 1. These navigation data give rise to the Randers metric [Shen 2002] p (−yu + xv)2 + (u2 + v 2 )(1 − x2 − y 2 ) −yu + xv + , F (x, y; u, v) = 2 2 1−x −y 1 − x2 − y 2 with W 2 = x2 +y 2 < 1. A feeding station is placed at a fixed location along the perimeter of the pond. The fish, eager to get to the food, swim along geodesics of F . As observed by a stationary viewer at the pond’s edge, these geodesics are spirals because it is the fish’s instinct to approach the perimeter quickly, where it can obtain the most help from the current’s linear velocity (which equals angular speed times radial distance). An experienced fish will aim itself with a velocity V , selected so that V + W is tangent to the spiral. ♦ For the pond described above, Shen has found numerically that the geodesics of F are spirals as expected. Now suppose that instead of fish we have an excited octopus which secretes ink as it swims toward the feeding station. The resulting
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
233
ink trail comoves with the swirling water. Its exact shape can be deduced by reexpressing the above spiral with respect to a frame which is comoving with the water. Shen agrees with this trend of thought and finds, somewhat surprisingly, that the trail in question is a straight ray. fish pond
ink trail
geodesic in space food
food
food
food
counterclockwise current
no current
3.1.2. Inverse problem. A question naturally arises: Can every strongly convex Randers metric be realised through the perturbation of some Riemannian metric h by some vector field W satisfying W  < 1? The answer is yes. Indeed, let us be given an arbitrary Randers metric F with Riemannian metric a and differential 1form b, satisfying kbk 2 := aij bi bj < 1. Set bi := aij bj and ε := 1 − kbk2 . Construct h and W as follows: hij := ε(aij − bi bj ),
W i := −bi /ε.
Note that F is Riemannian if and only if W = 0, in which case h = a. Also, Wi := hij W j = −εbi . Using this, it can be directly checked that perturbing h by the stipulated W gives back the Randers metric we started with. Furthermore, W 2 := hij W i W j = aij bi bj =: kbk2 < 1. Incidentally, the inverse of hij is hij = ε−1 aij + ε−2 bi bj . This hij , together with W i , defines a Cartan metric F ∗ of Randers type on the cotangent bundle T ∗ M . A comparison with [Hrimiuc–Shimada 1996] shows that F ∗ is the Legendre dual of the Finsler–Randers metric F on T M . It is remarkable that the Zermelo navigation data of any strongly convex Randers metric F is so simply related to its Legendre dual. See also [Ziller 1982; Shen 2002; 2004]. We summarise:
234
DAVID BAO AND COLLEEN ROBLES
Proposition 8. A strongly convex Finsler metric F is of Randers type if and only if it solves the Zermelo navigation problem on some Riemannian manifold (M, h), under the influence of a wind W with h(W, W ) < 1. Also, F is Riemannian if and only if W = 0. Example (Finslerian Poincar´ e disc). As an illustration of the inverse procedure, we apply it to the Randers metric F that describes the Finslerian Poincar´e disc (Section 2.3.2). With r and θ denoting polar coordinates on the open disc of radius 2 in R2 , the Randers metric in question is defined by a=
dr ⊗ dr + r 2 dθ ⊗ dθ , (1 − 14 r2 )2
b=
r dr . (1 + 14 r2 )(1 − 41 r2 )
The underlying Zermelo navigation data is h=
(1 − 14 r2 )2 dr ⊗ dr + r 2 (1 + 41 r2 )2 dθ ⊗ dθ , (1 + 14 r2 )4
W =
−r(1 + 41 r2 ) ∂r . 1 − 14 r2
It turns out that this Riemannian landscape h on which the navigation takes place is flat (!), and the associated 1form W [ = −r
1 − 41 r2 r2 1 2 dr = −8 d (4 + r 2 )2 1 + 4r
is exact. The wind here is blowing radially toward the center of the disc, and its strength decreases to zero there. The geodesics of h are straight lines. Those of F have been analysed in detail in [Bao et al. 2000]. Their trajectories coincide with those of the Riemannian Poincar´e model: straight rays to and from the origin, and circular arcs intersecting the rim of the disc at Euclidean right angles. ♦ 3.1.3. General relation between two covariant derivatives. Our goal in Section 3.2 will be to reexpress the Einstein characterisation of Theorem 5 in terms of the navigation data (h, W ). To that end, it is helpful to first relate the covariant derivative bij of b (with respect to a) to the covariant derivative W i:j of W (with respect to h). Let aγ i ij and hγ i jk denote the respective Christoffel symbols of the Riemannian metrics a and h. We have bij = bi,xj − bs aγ s ij
and Wi:j = Wi,xj − Ws hγ s ij .
Since a i
γ
ij
= (aGi )yj yk ,
h i
γ
ij
= (hGi )yj yk ,
it suffices to compare the geodesic spray coefficients a
Gi :=
j k 1a i 2 γ jk y y
=:
1a i 2 γ 00
and
h
Gi :=
j k 1h i 2 γ jk y y
=:
1h i 2 γ 00 .
The tool that effects this comparison is Rapcs´ ak’s identity (Section 1.2.2).
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
235
The symbol α denotes the Finsler norm associated to the Riemannian metric a: α2 (x, y) := aij (x)y i y j = a(y, y). According to Section 1.2.2, the userfriendly form of Rapcs´ ak’s identity gives a i G = hGi + 41 aij (α2 :0 )yj − 2α2 :j .
Using the formula aij =
1 λ hij
+ λ12 Wi Wj derived in Section 1.3.1, we find that
aij = λ(hij − W i W j )
and α2 =
1 1 h00 + 2 W0 2 . λ λ
Here, Wi := hij W j , λ := 1 − W i Wi , and the formula for the inverse aij is ascertained by inspection. A straightforward but tedious computation of the righthand side of Rapcs´ ak’s identity yields a
Gi = hGi + ζ i ,
where (using W s:i to abbreviate W s :r hri ) ζ i :=
1 1 1 i s y W Ws:0 + 21 W i W0:0 + h00 + 2 W0 2 (W i W s W t Ws:t − Ws W s:i ) λ 2λ λ 1 + W0 W i W s (Ws:0 + W0:s ) + (W i :0 − W0 :i ) . 2λ
Differentiating with respect to y j , y k gives aγ i jk = hγ i jk + (ζ i )yj yk , whence bjk = bj:k − bi (ζ i )yj yk . Into the righthand side we substitute bs = −(1/λ)Ws (Section 1.3.1), resulting in a formula for bjk in terms of the covariant derivatives of W with respect to h. For later purposes, it is best to split the answer into its symmetric and antisymmetric parts, bjk = 21 (bjk + bkj ) + 12 (bjk − bkj ) =:
1 2
liejk + 12 curljk ,
and to introduce the abbreviations Ljk := Wj:k + Wk:j , Then the said computation gives bjk =
1 2
liejk + 12 curljk , with
2 Wj Wk W s W t Lst λ λ2 1 1 + W 2 i W (Wi:j Wk + Wi:k Wj ) − (Wj Wk:i + Wk Wj:i )W i , + 2 λ λ 2 1 = − Cjk + 2 W i (Wi:j Wk − Wi:k Wj ). λ λ
liejk = −Ljk −
curljk
1
Cjk := Wj:k − Wk:j .
hjk +
Observe that since curljk = ∂xk bj −∂xj bk and liejk = bi ∂xi ajk +aik ∂xj bi +aji ∂xk bi (where bi = −λW i ), the last two conclusions could have been obtained without relying on the explicit formula of aγ. In any case, the relation above between bjk and the covariant derivatives of W is valid without any assumption on b.
236
DAVID BAO AND COLLEEN ROBLES
3.2. Navigation description of curvature conditions. Theorem 5 (p. 227) characterises Einstein Randers metrics via the defining Riemannian metric a and 1form b. It says that a Randers metric F is Einstein if and only if both the Basic and Ricci Curvature Equations hold with constant σ (or Scurvature). Though this is a substantial improvement over the Einstein criterion K i i = Ric(x)F 2 , most notably in the realm of computation, the characterisation does little to describe the geometry of these metrics. Surprisingly, the breakthrough lies in a change of dependent variables. We find that replacing the defining data (a, b) by the navigation data (h, W ) (discussed in Section 3.1.2) yields a breviloquent geometric description of Einstein Randers metrics. Explicitly, this change of variables reveals that the Riemannian metric h must be an Einstein metric itself, and the vector field W an infinitesimal homothety of h. The next two subsections are devoted to developing this “navigation description”. 3.2.1. Consequences of the Basic Equation. Our first step is to derive the navigation version of the Basic Equation lieik + bi θk + bk θi = σ(aik − bi bk ) of page 226. For that we replace aik by (1/λ)hik + (1/λ2 )Wi Wk , bi by −(1/λ)Wi , and lieik by the expression derived on the previous page. We also use the formula for curljk there to compute θk := bj curljk . Since bj := ajs bs = −λW j , we get θk = T k −
2 1 (W i W j Lij )Wk + W 2 W i Wi:k . λ λ
Here, an abbreviation has been introduced for the ubiquitous quantity Tk := W j (Wj:k − Wk:j ) = W j Cjk . These manoeuvres, followed by some rearranging, convert the Basic Equation to λLik + L(W, W )hik = −σhik , where L(W, W ) stands for Lst W s W t . Contracting with W i , W k and using λ := 1 − W 2 shows that L(W, W ) = −σW 2 . Consequently, the navigation version of the Basic Equation is Lik := Wi:k + Wk:i = −σhik ,
that is, LW h = −σh.
We name this the LW Equation. It says that W is an infinitesimal homothety of h; see [Kobayashi–Nomizu 1996, p. 309]. In this equation, σ must be zero whenever h is not flat. (In particular, σ must vanish whenever h is not Ricciflat.) Indeed, let ϕ t denote the time t flow of the vector field W . The LW Equation tells us that ϕ∗t h = e−σt h. Since ϕt is a diffeomorphism, e−σt h and h must be isometric; therefore they have the same sectional curvatures. If h is not flat, this condition on sectional curvatures mandates that e−σt = 1, hence σ = 0. The argument we presented was pointed out to us by Bryant. Incidentally, the use of h = ε(a − b ⊗ b) and W = −b] /ε (Section 3.1.2) allows us to recover the Basic Equation from the LW Equation. Thus the two equations
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
237
are equivalent. This remains so even if σ were to be a function of x, because neither equation contains any derivative of σ. We now turn to the derivation of the navigation versions of the Ricci Curvature Equation (Theorem 5, page 227) and the Curvature Equation (Theorem 6, page 228). The LW Equation affords simplified expressions for many quantities that enter into the curvature equations of Theorems 5 and 6. The key in all such simplifications can invariably be traced back to the statement Wi:j = 12 (Lij + Cij ) = − 21 σhij + 21 Cij . We first address all but one of the terms on the righthand sides of the curvature equations. Keep in mind that indices on curl, θ, and ayi := aij y j are manipulated by the Riemannian metric a, while those on C, T , and hyi := hij y j are manipulated by the Riemannian metric h. The relevant formulae are 1 1 curlij = − Cij + 2 (Ti Wj − Tj Wi ), λ λ 1 curli j = −C i j + T i Wj , λ curlij = −λC ij ;
1 1h yi + 2 W i W 0 , λ λ 1 θ i = Ti , λ θi = T i .
a
yi =
This does not address the term θ00 (equivalently θik +θki ), which appears on the righthand side of the Ricci Curvature Equation (Section 2.2.3). To tackle this, as well as the lefthand sides of those curvature equations, we shall need the relation between the geodesic spray coefficients aGi and hGi . That relation, first derived in Section 3.1.3, undergoes a dramatic simplification in the presence of the LW Equation. The result is a
Gi = hGi + ζ i ,
where ζ i = yi
1 1 1 i 1 (T0 − σ W0 ) − T i h00 + 2 W0 2 + C 0 W0 . 2λ 4λ 2λ 2λ
Hence θ00 = θ0:0 − 2θi ζ i =
1 1 1 1 T0:0 − 2 W0 T i Ci0 + h00 + 3 W0 2 T i Ti . λ λ 2λ2 λ
Differentiating these two statements with respect to y j , y k gives an explicit relation between the Christoffel symbols aγ i jk and hγ i jk , as well as a formula for θjk +θkj in terms of the navigation data (h, W ). However, we refrain from doing so. In the following two subsections, we shall determine the navigation version of the Ricci Curvature Equation in Theorem 5 and of the Curvature Equation in Theorem 6. It is found, in retrospect, that the computational tedium is significantly lessened by working with a Ric00 , aK i k rather than a Ricij , aRhijk . Consequently, the relation aGi = hGi + ζ i and the formula for θ00 should suffice.
238
DAVID BAO AND COLLEEN ROBLES
3.2.2. Einstein–Randers metrics. The contracted form of the Ricci Curvature Equation (Section 2.2.3) reads a
Ric00 = (α2 + β 2 ) Ric(x) − 41 α2 curlhj curlhj − 12 curlj 0 curlj0 −(n − 1)
2 1 2 16 σ (3α
− β 2 ) + 41 (θ0 )2 + 12 θ00 .
With α2 = a00 = (1/λ)h00 + (1/λ2 )W0 2 and β = b0 = −(1/λ)W0 , the simplified formulae in Section 3.2.1, and Ric(x) =: (n − 1)K(x), all the terms on the righthand side are accounted for. For the lefthand side, note first that a Ric00 = aK i i . Specialising the split and covariantised form of Berwald’s formula (Section 1.2.3) to F = a and F = h, and taking the natural trace, we have a
Ric00 = h Ric00 + (2ζ i ):i − (ζ i )yj (ζ j )yi − y j (ζ i :j )yi + 2ζ j (ζ i )yj yi .
Though ζ i has a simplified formula (Section 3.2.1), computing the four terms dependent on ζ is still tedious. That task is helped by the L W Equation and the navigation description [Robles 2003] of the E23 Equation: 1 2 W i :0:i = (n − 1) K(x) + 16 σ W0 . The result is unexpectedly elegant: h
1 2 σ h00 . Ric00 = (n − 1) K(x) + 16
Differentiating away the contracted y i , y k gives h Ricik . Conversely, it has been checked that, via h = ε(a − b ⊗ b) and W = −b ] /ε (Section 3.1.2), the above navigation description reproduces the characterisation in Theorem 5. Thus the characterisation (in terms of a, b) is equivalent to the navigation description (in terms of h, W ). In particular, Theorem 5 implies: Theorem 9 (Einstein navigation description). Suppose the Randers metric F solves Zermelo’s problem of navigation on the Riemannian manifold (M, h) under the external influence W , W  < 1. Then (M, F ) is Einstein with Ricci scalar Ric(x) =: (n − 1)K(x) if and only if there exists a constant σ such that 1 2 (i) h is Einstein with Ricci scalar (n − 1) K(x) + 16 σ , that is, h 1 2 σ hik , Ricik = (n − 1) K(x) + 16 and (ii) W is an infinitesimal homothety of h, namely,
(LW h)ik = Wi:k + Wk:i = −σhik . Furthermore, σ must vanish whenever h is not Ricciflat. We call this a ‘description’ rather than a ‘characterisation’ because, in contrast with Theorem 5, it makes explicit the underlying geometry of Einstein–Randers metrics. Section 4 will illustrate Theorem 9 with a plethora of examples.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
239
3.2.3. Constant flag curvature Randers metrics. For Randers metrics of constant flag curvature, the equation destined to be recast into navigational form is given in Theorem 6. Contracting it with y h , y k , raising the index i with a, and relabelling j as k, we obtain the following expression for the spray curvature a i K k (= y j aRj i kl y l ; Section 1.2.3): a i 1 2 s 3 2 σ )+(K + 16 σ )b bs − 14 θs θs (δ i k α2 −y i ayk ) K k = (K − 16
+ 41 curls 0 (curls i ayk +y i curlsk −curls0 δ i k )− 41 α2 curlsi curlsk − 34 curli 0 curlk0 .
Here ayk := aik y i . This is equivalent to the Curvature Equation because a Rhijk = 31 (aKij )yk yh − (aKik )yj yh .
(∗)
We now recast the equality just given for aK i k . All the terms on the righthand side are routinely computed, using α 2 = a00 = (1/λ)h00 + (1/λ2 )W0 2 , bs = −(1/λ)Ws , bs = −λW s , and the simplified formulae in Section 3.2.1. For the lefthand side, we first specialise the split and covariantised form of Berwald’s formula (Section 1.2.3) to F = a and F = h, getting a
K i k = hK i k + (2ζ i ):k − (ζ i )ys (ζ s )yk − y s (ζ i :s )yk + 2ζ s (ζ i )ys yk .
Into this we substitute the simplified formula for ζ (Section 3.2.1). The ensuing computation is assisted by the prodigious use of the L W Equation and the navigation description of the CC(23) Equation: 1 2 σ (hik Wj − hjk Wi ). Wi:j:k = K + 16 The result is as elegant as the Einstein case: h i 1 2 K k = K + 16 σ (δ i k h00 − y i hyk ),
where hyk := hik y i . Lowering the index i with the Riemannian metric h and differentiating in the same fashion as formula (∗) gives hRhijk . We have verified that the use of h = ε(a − b ⊗ b) and W = −b] /ε (Section 3.1.2) converts the above navigation description in terms of h, W back to the characterisation in terms of a, b presented by Theorem 6. So the two pictures are indeed equivalent, and Theorem 6 implies: Theorem 10 (Constant flag curvature navigation description). Suppose the Randers metric F solves Zermelo’s problem of navigation on the Riemannian manifold (M, h) under the external influence W , W  < 1. Then (M, F ) is of constant flag curvature K if and only if there exists a constant σ such that 1 2 σ ), that is, (i) h is of constant sectional curvature (K + 16 h 1 2 σ (hij hhk − hik hhj ) , Rhijk = K + 16
and
240
DAVID BAO AND COLLEEN ROBLES
(ii) W is an infinitesimal homothety of h, namely, (LW h)ik = Wi:k + Wk:i = −σhik . Furthermore, σ must vanish whenever h is not flat. Examples. We present three favorite examples to illustrate the use of Theorem 10. In the examples below: position coordinates are denoted x, y, z rather than x1 , x2 , x3 ; components of tangent vectors are u, v, w instead of y 1 , y 2 , y 3 . Funk disc. The Finslerian Poincar´e example was discussed in Section 2.3.2 and Section 3.1.2. Here we revisit it a third time, for the sake of those who prefer to work with simple navigation data. Fix the angle θ and contract the radius via r 7→ r/ 1 + 41 r2 . This map is an isometry of the Finslerian Poincar´e model onto the Funk metric of the unit disc [Funk 1929; Okada 1983; Shen 2001]. The navigation data of the Funk metric is simple: h is the Euclidean metric and the radial W = −r∂ r is an infinitesimal homothety with σ = 2. Writing tangent vectors at (r, θ) as u∂ r + v∂θ , we have p u2 + r2 (1 − r 2 )v 2 ru F = + , with r 2 = x2 + y 2 . 2 1−r 1 − r2 1 2 σ = 0. Hence the Funk metric on the unit disc has By Theorem 10, K + 16 constant flag curvature K = − 41 . The isometry above is a global change of coordinates which transforms the navigation data in the Section 3.1.2 example into a more computationally friendly format. ♦
A 3sphere that is not projectively flat. We start with the unit sphere S 3 in R4 , parametrised by its tangent spaces at the poles, as in [Bao–Shen 2002]. For each constant K > 1, let h be 1/K times the standard Riemannian metric induced on S 3 . The rescaled metric has sectional curvature K. Perturb h by the Killing vector field √ W = K − 1 − s(1 + x2 ), z − sxy, −y − sxz , p with s = ±1 depending on the hemisphere. Then W  = (K − 1)/K and W is tangent to the S 1 fibers in the Hopf fibration of S 3 . By Theorem 10, the resulting Randers metric F has constant flag curvature K. Explicitly, F = α + β, where p K(su − zv + yw)2 + (zu + sv − xw)2 + (−yu + xv + sw)2 α= , 1 + x2 + y 2 + z 2 √ K − 1(su − zv + yw) . β= 1 + x2 + y 2 + z 2 This Randers metric is not projectively flat [Bao–Shen 2002]. This is in stark contrast with the Riemannian case because, according to Beltrami’s theorem, a Riemannian metric is locally projectively flat if and only if it is of constant sectional curvature. ♦
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
241
Shen’s fish tank. This example, first presented in [Shen 2002], is a threedimensional variant of Shen’s fish pond (Section 3.1.1). Consider a cylindrical fish tank x2 + y 2 < 1 in R3 , equipped with the standard Euclidean metric h. Suppose the tank has a rotational current with velocity vector W = y∂ x − x∂y + 0∂z , and a big inquisitive mosquito hovers just above the water. Wishing to reach the bug as soon as possible, the hungry fish swim along a path of shortest time — that is, along a geodesic of the Randers metric F = α + β with Zermelo navigation data h and the infinitesimal rotation W . Explicitly, p (−yu + xv)2 + (u2 + v 2 + w2 )(1 − x2 − y 2 ) , α= 1 − x2 − y 2 −yu + xv β= , with W 2 = x2 + y 2 . 1 − x2 − y 2 Since W is a Killing field of Euclidean space we have σ = 0, and Theorem 10 tells us that F is of constant flag curvature K = 0. The same conclusion holds for the fish pond. ♦ Theorem 10 implies that every constant flag curvature Randers metric is locally isometric to a “standard model” with navigation data (h, W ), where h is a standard Riemannian space form (sphere, Euclidean space, or hyperbolic space), and W is one of its infinitesimal homotheties. It remains to sort these standard models into Finslerian isometry classes. Let (M1 , F1 ), (M2 , F2 ) be any two Randers spaces, with navigation data (h1 , W1 ) and (h2 , W2 ). It is a fact that F1 , F2 are isometric as Finsler metrics if and only if there exists a Riemannian isometry ϕ : (M 1 , h1 ) → (M2 , h2 ) such that ϕ∗ W1 = W2 . For each standard Riemannian space form (M, h), the isometry group G of h leaves h invariant, but acts on its infinitesimal homotheties W via pushforward. By the cited fact, all (h, W ) which lie on the same Gorbit generate mutually isometric standard models. This redundancy can be suppressed by collapsing each Gorbit to a point. For any fixed K, the resulting collection of such “points” constitutes the moduli space M K for strongly convex Randers metrics of constant flag curvature K. Lie theory effects (a parametrisation and hence) a dimension count of MK ; see [Bao et al. 2003] for details. The table on the next page includes, for comparison, similar information about the Riemannian setting and the case θ := curl(b] , ·) = 0 (Section 2.3.2). 3.3. Issues resolved by the navigation description 3.3.1. Schur lemma for the Ricci scalar. In essence, this lemma constrains the geometry of Einstein metrics in dimension > 3 by forcing the Ricci scalar to be constant. Historically, this is the second Schur lemma in (nonRiemannian) Finsler geometry. The first Finslerian Schur lemma concerns the flag curvature; see [del Riego 1973; Matsumoto 1986; Berwald 1947]. An exposition can be found in [Bao et al. 2000].
242
DAVID BAO AND COLLEEN ROBLES
CFC metrics
dim M
Dimension of moduli space K0 K=0 σ=0 σ 6= 0
Riemannian n>2 0 empty b equiv. W = 0 Yasuda–Shimada even n 0∗ 1 0∗ 0† θ=0 odd n 1 Unrestricted even n n/2 Randers odd n (n + 1)/2 (n − 1)/2 The moduli spaces of dimension 0 consist of a single point. ∗ The single isometry class is Riemannian. † The single isometry class is nonRiemannian, of Funk type. Table 1. The dimension of the moduli space for several families of constant flag curvature (CFC) Randers metrics.
In two dimensions, the Ricci scalar of a Riemannian metric is the Gaussian curvature of the surface. Hence all Riemannian surfaces are Einstein in the sense of Section 1.3.2. Since the Gaussian curvature is not constant in general, the Schur lemma fails for Riemannian (and therefore Randers) metrics in two dimensions. It is natural to ask whether the Schur lemma also fails for nonRiemannian (W 6= 0) Randers surfaces. The answer is yes. Section 4.1 develops a class of nonRiemannian Randers surfaces whose Ricci scalars are nonconstant functions of x alone. In particular, these nonRiemannian surfaces are Einstein, but fail the Schur lemma. In dimension n > 3, every Riemannian Einstein metric h is Ricciconstant. This follows readily from tracing the second Bianchi identity and realising that, for such metrics, h Ricij = (S/n)hij , where S denotes the scalar curvature of h. Lemma 11 (Schur lemma). The Ricci scalar of any Einstein Randers metric in dimension greater than two is necessarily constant. Proof. Suppose F is an Einstein metric of Randers type, with navigation data (h, W ) and Ricci scalar Ric(x) = (n − 1)K(x). Theorem 9 says that h must be 1 2 Einstein with Ricci scalar (n − 1) K + 16 σ , for some constant σ. Since n > 2 1 2 here, the Riemannian Schur lemma forces K + 16 σ to be constant. The same must then hold for K and Ric = (n − 1)K. Another proof of the Schur lemma, based on the Einstein characterisation of Section 2.2.3 (Theorem 5 on page 227), is given in [Robles 2003]. 3.3.2. Three dimensional Einstein–Randers metrics. For Riemannian metrics in three dimensions, being Einstein and having constant sectional curvature are equivalent conditions because the conformal Weyl curvature tensor automatically vanishes. It is not known whether this rigidity holds for Einstein–Finsler metrics
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
243
in general. However, the said rigidity does hold for Randers metrics. The proof rests on a comparison between the navigation descriptions for Einstein (Section 3.2.2) and constant flag curvature (Section 3.2.3) Randers metrics. Proposition 12 (Threedimensional rigidity). Let F be a Randers metric in three dimensions. Then F is Einstein if and only if it has constant flag curvature. Proof. Metrics of constant flag curvature are always Einstein. As for the converse, let F be an Einstein Randers metric with navigation data (h, W ). The Ricci scalar of F is (n − 1)K = 2K; in view of the Schur lemma above, K has to be constant. According to Theorem 9, h is Einstein with Ricci scalar 1 2 2 K + 16 σ , for some constant σ. By Riemannian threedimensional rigidity, h 1 2 σ . The navigation description in must have constant sectional curvature K + 16 Theorem 10 then forces F to be of constant flag curvature K. Interestingly, the two navigation descriptions also tell us that any Einstein Randers metric that arises as a solution to Zermelo’s problem of navigation on a Riemannian space form must be of constant flag curvature. 3.3.3. The Matsumoto identity. This identity first came to light in a letter from Matsumoto to the first author. It says that any Randers metric of constant flag curvature K satisfies 1 2 σ = 0. σ K + 16 Since metrics of constant flag curvature are Einstein, it is natural to wonder whether this identity can be extended to Einstein Randers metrics. The answer is yes, by the following result:
Proposition 13. Let F be an Einstein Randers metric whose Ricci scalar Ric(x) we reexpress as (n − 1)K(x). Then σ K+
1 2 16 σ
=
W i K:i 0
when n = 2, when n > 2.
Here, σ is the constant supplied by the navigation data (h, W ) of F . According 1 2 to Theorem 9, h must be Einstein with Ricci scalar (n − 1)(K + 16 σ ), and W satisfies the LW Equation Wi:j + Wj:i = −σ hij . Proof. We begin with the Ricci identity for the tensor C ij := Wi:j − Wj:i , namely Cij:k:h − Cij:h:k = Csj hRi s kh + Cis hRj s kh , where hRhi jk is the curvature tensor of h. Trace this identity on (i, k) and (h, j) to obtain W i:j :i − W j:i :i
:j
= W i:j − W j:i
h
Ricij = 0,
where the second equality follows because h Ricij is symmetric.
244
DAVID BAO AND COLLEEN ROBLES
Next, we compute W i:j :i − W j:i :i . To that end, differentiating the LW Equation gives Wp:q:r + Wq:p:r = 0. This and the Ricci identity for W imply that Wi:j:k − Wj:i:k = (Wi:j:k − Wi:k:j ) − (Wk:i:j − Wk:j:i ) + (Wj:k:i − Wj:i:k ) = W s hRisjk − W s hRksij + W s hRjski . Using h to trace on (i, k) and raise j, we get W i:j :i − W j:i :i = 2Ws h Ricsj . Since 1 2 h Ric = (n − 1)(K + 16 σ )h, we are led to 1 2 σ W j. (∗) W i:j :i − W j:i :i = 2(n − 1) K + 16 Finally, tracing the LW Equation gives 2W j :j = −nσ, whence
1 2 0 = (W i:j :i − W j:i :i ):j = 2(n − 1) W j K:j − 21 nσ(K + 16 σ ) .
The identity now follows from the Schur lemma (Section 3.3.1).
Remark. If we assume that the LW Equation (or the Basic Equation, which amounts to the same) holds, the E23 Equation (Section 2.2.3) can be reexpressed as (∗). Thus (∗) is the navigation version of the E23 Equation. It can be further refined, using Wi:j:k + Wj:i:k = 0, to read 1 2 σ Wj . W i :j:i = (n − 1) K + 16 A second derivation of the Matsumoto Identity, based on Theorem 5 (page 227), may be found in [Robles 2003].
4. Einstein Metrics of Nonconstant Flag Curvature We now present a variety of nonRiemannian Randers metrics that are either Einstein or Ricciconstant. Apart from the 2sphere, which is included merely because of its simplicity, all examples have nonconstant flag curvatures. Section 3.1.1 will be used without mention. 4.1. Examples with Riemannian–Einstein navigation data 4.1.1. Surfaces of revolution. Our first class of examples comprises surfaces of rotation in R3 . We shall see that solutions to Zermelo’s problem of navigation under infinitesimal rotations are Einstein, with Ricci scalar Ric(x) equal to the Gaussian curvature of the original Riemannian surface. Among the examples below, two (the elliptic paraboloid and the torus) have nonconstant Ricci scalar. These solutions of Zermelo navigation are nonRiemannian counterexamples to Schur’s lemma in dimension 2. To begin, take any surface of revolution M , obtained by revolving a profile curve ϕ 7→ 0, f (ϕ), g(ϕ)
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
245
in the right half of the yzplane around the z axis. The ambient Euclidean space induces a Riemannian metric h on M . Parametrise M by (θ, ϕ) 7→ f (ϕ) cos θ, f (ϕ) sin θ, g(ϕ) , 0 6 θ 6 2π.
Now consider the infinitesimal isometry W := ε∂θ , where ε is a constant. By limiting the size of our profile curve if necessary, there is no loss of generality in assuming that f is bounded. Choose ε so that εf  < 1 for all ϕ. Expressing h in the given parametrisation, we find that the solution to Zermelo’s problem is the Randers metric F = α + β on M , with q u2 f 2 + v 2 (1 − ε2 f 2 )(f˙2 + g˙ 2 ) −εuf 2 α= , β= , 2 2 1−ε f 1 − ε2 f 2 and kbk2 = ε2 f 2 = W 2 . Here, u∂θ +v ∂ϕ represents an arbitrary tangent vector on M , and f˙, g˙ are the derivatives of f , g with respect to ϕ. Note that the hypothesis εf  < 1 ensures strong convexity. Because W is a Killing vector field, σ vanishes. The Einstein navigation description (Theorem 9 on page 238) then says that the Ricci scalar Ric of F is identical to that of the Riemannian metric h. The latter is none other than the Gaussian curvature K of h. Hence g˙ (f˙ g¨ − f¨g) ˙ , Ric(x) = K(x) = 2 2 ˙ f (f + g˙ )2 where the dots indicate derivatives with respect to ϕ. We examine three special cases of surfaces of revolution: Sphere. The unit sphere is given as a surface of revolution by f (ϕ) = cos ϕ and g(ϕ) = sin ϕ. We will consider the infinitesimal rotation W = ε∂ θ , with ε < 1 to effect the necessary εf  < 1. The Randers metric solving Zermelo’s problem of navigation on the sphere under the influence of W is of constant flag curvature K = 1, with p u2 cos2 ϕ + v 2 (1 − ε2 cos2 ϕ) −εu cos2 ϕ , β= . ♦ α= 2 2 1 − ε cos ϕ 1 − ε2 cos2 ϕ Elliptic paraboloid. This is the surface z = x2 + y 2 in R 3 . Set the multiple ε in W to be 1. The resulting Randers metric lives on the x 2 + y 2 < 1 portion of the elliptic paraboloid, and has Ricci scalar 4/(1 + 4x 2 + 4y 2 )2 . It reads q (−yu + xv)2 + (1 + 4x2 )u2 + 8xy uv + (1 + 4y 2 )v 2 (1 − x2 − y 2 ) α= , 1 − x2 − y 2 yu − xv β= , with kbk2 = x2 + y 2 . ♦ 1 − x2 − y 2 Torus. Specialize to a torus of revolution with parametrisation (θ, ϕ) 7→ (2 + cos ϕ) cos θ, (2 + cos ϕ) sin θ, sin ϕ .
246
DAVID BAO AND COLLEEN ROBLES
Set the multiple ε in W to be 14 . The resulting Randers metric on the torus has Ricci scalar cos ϕ/(2 + cos ϕ). It is given by q 4 16(2 + cos ϕ)2 u2 + 16 − (2 + cos ϕ)2 v 2 , α= 16 − (2 + cos ϕ)2 −4(2 + cos ϕ)2 u 1 (2 + cos ϕ)2 . ♦ β= , with kbk2 = 16 16 − (2 + cos ϕ)2 4.1.2. Certain Cartesian products. Recall from Section 1.3.2 the geometrical definition of the Ricci scalar and of Einstein metrics. When specialised to Riemannian nmanifolds, it says that the Ricci scalar Ric is obtained by summing the sectional curvatures of n − 1 appropriately chosen sections that share a common flagpole. A Ricciconstant metric is remarkable because this sum is a constant. A moment’s thought convinces us of the following: The Cartesian product of two Riemannian Einstein metrics with the same constant Ricci scalar ρ is again Ricciconstant, and has Ric = ρ. As we will illustrate, this allows us to construct a wealth of Ricciconstant Randers metrics with nonconstant flag curvature. Fix ρ. For i = 1, 2, let Mi be an ni dimensional Riemannian manifold with constant sectional curvature ρ/(ni − 1). Therefore Mi is Einstein with Ricci scalar ρ. Let Wi be a Killing field on Mi . Let h denote the product Riemannian metric on the Cartesian product M = M1 × M2 . Then h has constant Ricci scalar ρ, and admits W = (W1 , W2 ) as a Killing field. By Theorem 9 (page 238), the Randers metric F generated by the navigation data (h, W ) on M is Einstein, with constant Ricci scalar ρ. When ρ is nonzero, the Einstein metric h on M is not of constant sectional curvature. Hence Theorem 10 (page 239) assures us that F will not be of constant flag curvature. Proposition 8 (page 234) then says that the Randers metric F is nonRiemannian if and only if the wind W is nonzero. So it suffices to select a nonzero W1 . e 1 be the n1 dimensional, complete, simply connected stanTo that end, let M dard model of constant sectional curvature ρ/(n 1 − 1). The space of globally e 1 is a Lie algebra g of dimension 1 n1 (n1 + 1). Select a defined Killing fields on M 2 ˜ 1 from g. The isometry group G of M e acts on g via pushforwards. Let nonzero W ˜ 1 . Then we have a natural H be any finite subgroup of the isotropy group of W e e e 1 /H is of constant projection π : M 1 → M 1 /H. The quotient space M1 := M ˜ 1. sectional curvature ρ/(n1 − 1), and has a nonzero Killing field W1 := π∗ W As a concrete illustration, we specialise the discussion to spheres. For simplicity, we specify the finite subgroup H to be trivial. √ Example. Let Mi (i = 1, 2) be the ni sphere of radius ni − 1, ni > 2. Then Mi has constant sectional curvature 1/(ni − 1), and is therefore Einstein with Ricci scalar 1. The Cartesian product M = M1 ×M2 , equipped with the product
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
247
metric h, is an (n1 + n2 )dimensional Riemannian Einstein manifold with Ricci scalar 1, and it is not of constant sectional curvature. The Lie algebra of Killing fields on the nsphere S n (r) with radius r is isomorphic to so(n + 1), regardless of the size of r. The following description accounts for all such vector fields. View points p ∈ S n (r) as row vectors in Rn+1 . For each Ω ∈ so(n + 1), the assignment p 7→ pΩ ∈ Tp (S n (r)) is a globally defined Killing vector field on S n (r). Now, for i = 1, 2, take Ωi ∈ so(ni + 1). Denote points of M1 , M2 by p and q, respectively. The map (p, q) 7→ (pΩ1 , q Ω2 ) ∈ T(p,q) M defines a Killing field W of h. This W is nonzero as long as Ω1 , Ω2 are not both zero. Zermelo navigation on (M, h) under the influence of W generates a nonRiemannian Randers metric with constant Ricci scalar 1, and which is not of constant flag curvature. ♦ 4.2. Examples with K¨ ahler–Einstein navigation data. In this section, we construct Einstein metrics of Randers type, with navigation data h, which is a K¨ ahler–Einstein metric, and W , which is a Killing field of h. We choose h from among K¨ ahler metrics of constant holomorphic sectional curvature, because the formula can be explicitly written down. There exist much more general K¨ ahler– Einstein metrics, for instance those with positive sectional curvature; see [Tian 1997]. 4.2.1. K¨ahler manifolds of constant holomorphic sectional curvature Suppose (M, h) is a K¨ ahler manifold of complex dimension m (real dimension n = 2m) with complex structure J. Let (z 1 , . . . , z m ), where z α := xα + ixα¯ , denote local ¯ ¯ are the 2m real coordinates. complex coordinates. Here, x1 , . . . , xm ; x1 , . . . , xm In our notation, lowercase Greek indices run from 1 to m. The complex coordinate vector fields are Zα := ∂zα = 12 (∂xα − i∂xα¯ ), Zα¯ := ∂z¯α = 21 (∂xα + i∂xα¯ ), respectively eigenvectors of J with eigenvalues +i and −i. In what follows, uppercase Latin indices run through 1, . . . , m, ¯1, . . . , m. ¯ Also, z¯α := xα − ixα¯ abbreviates the complex conjugate of z α . The K¨ ahler metric h is a Riemannian (real) metric with the Jinvariance property h(JX, JY ) = h(X, Y ), and such that the 2form (X, Y ) 7→ h(X, JY ), known as the K¨ ahler form, is closed. Let hAB := h(ZA , ZB ). The Jinvariance of h implies hαβ = 0 = hα¯ β¯ . Expanding in terms of the complex basis gives h = hαβ¯ (dz α ⊗ d¯ z β + d¯ z β ⊗ dz α ), where (hαβ¯ ) is an m × m complex Hermitian matrix. By contrast, using the real basis and setting HAB := h(∂xA , ∂xB ), we have h = HAB dxA ⊗ dxB , with Hα¯ β¯ = Hαβ and Hαβ ¯ = −Hαβ¯ : the two diagonal blocks are symmetric and identical, while the offdiagonal blocks are skewsymmetric and negatives of each other.
248
DAVID BAO AND COLLEEN ROBLES
The K¨ ahler form being closed is equivalent to either Z γ (hαβ¯ ) = Zα (hγ β¯ ) or Zγ¯ (hαβ¯ ) = Zβ¯ (hα¯γ ), which puts severe restrictions on the usual Riemannian connection of h. Consequently, if we expand the curvature operator hR(ZC , ZD )ZB as hRB A CD ZA in the complex basis, it is not surprising to find considerable economy among the coefficients: Rβ¯ α CD = hRβ α¯ CD = 0 = hRB A γδ = hRB A γ¯ δ¯.
h
For each Y in the real tangent space Tx M , the 2plane spanned by {Y, JY } is known as a holomorphic section because it is invariant under J, and the corresponding sectional curvature (in the usual Riemannian sense) is called a holomorphic sectional curvature. If all such curvatures are equal to the same constant c, the K¨ ahler metric is said to be of constant holomorphic sectional curvature c. Such metrics are characterised by their curvature tensor having the following form in the complex basis {ZA }: c h (hβ α¯ hγ δ¯ + hγ α¯ hβ δ¯). Rβ αγ ¯ δ¯ = 2 Equivalently, c h Rβ α γ δ¯ = (δβ α hγ δ¯ + δ α γ hβ δ¯). 2 Discussion of all this may be found in [Kobayashi–Nomizu 1996]; but one must adjust for the fact that their KABCD is our RBACD , and that their definition of the curvature operator agrees with ours. Remarks. 1. Suppose h is a K¨ ahler metric of constant holomorphic sectional curvature c. Return to the expression for hRβ α γ δ¯ above. Tracing on the indices α and γ, we see that h Ric = (m + 1)(c/2)h. That is, h must be an Einstein metric with constant Ricci scalar (m + 1)c/2. 2. If a K¨ ahler metric h were to satisfy the stronger condition of constant sectional curvature c, then it would necessarily be Einstein with Ricci scalar (n − 1)c = (2m − 1)c; see Section 1.3.1. At the same time, (1) implies that the Ricci scalar is (m + 1)c/2. Hence we would have to have either c = 0 or m = 1. Thus a K¨ ahler manifold (M, h) can have constant sectional curvature in only two ways: either h is flat, or the real dimension of M is 2. This rigidity indicates that in the K¨ ahler category, the weaker concept of constant holomorphic sectional curvature is more appropriate. In analogy with the constant sectional curvature case we have a classification theorem for K¨ ahler metrics of constant holomorphic sectional curvature [Kobayashi–Nomizu 1996]: Any simply connected complete K¨ ahler manifold of constant holomorphic sectional curvature c is holomorphically isometric to one of three standard models: c > 0: the Fubini–Study metric on CP n (see below), c = 0: the standard Euclidean metric on Cn , c < 0: the Bergmann metric on the unit ball in Cn .
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
249
4.2.2. Killing fields of the Fubini–Study metric. The Fubini–Study metric is a K¨ ahler metric on CP m of constant holomorphic sectional curvature c > 0. Complex projective space is obtained from Cm+1 r 0 by quotienting out the equivalence relation ζ ∼ λζ, where 0 6= λ ∈ C. Denote class the equivalence m j m j of ζ by [ζ]. CP is covered by the charts U := [ζ] ∈ CP : ζ 6= 0 , j = 0, 1, . . . , m, with holomorphic coordinate mapping [ζ] 7→
1 0 (ζ , . . . , ζbj , . . . , ζ m ) =: (z 1 , . . . , z m ) = z ∈ Cm . ζj
In these coordinates, the Fubini–Study metric of constant holomorphic sectional curvature c > 0 has components hαβ¯ := h(Zα , Zβ¯ ) =
1 2 1 δαβ − 2 z¯α zβ , c ρ ρ
where z¯α := δαβ z¯β , zβ := δβτ z τ , and ρ := 1 + z α z¯α . Conventional wisdom in complex manifold theory prompts us to construct some explicit Killing vector field of h by considering ξ := π ∗ (P i j ζ j ∂ζ i ). Here, π : Cm+1 r0 → CP m is the natural projection and P ∈ u(m+1), the Lie algebra of the unitary group U (m + 1). Note that U (m + 1) is the group of holomorphic isometries of Euclidean Cm+1 r 0. For concreteness, we compute ξ in the local coordinates of the chart U 0 . We have π(ζ 0 , ζ 1 , . . . , ζ m ) := (1/ζ 0 )(ζ 1 , . . . , ζ m ) =: (z 1 , . . . , z m ), from which it follows that 1 1 π∗ (∂ζ 0 ) = − 0 z α ∂zα and π∗ (∂ζ α ) = 0 ∂zα ζ ζ because the differential of π is simply the matrix −z 1 1 0 1 .. .. π∗ = 0 . . . ζ 0 m −z 1 Thus
π∗ (P i j ζ j ∂ζ i ) = P 0 j ζ j −
1 α 1 z ∂z α + P α j ζ j 0 ∂z α 0 ζ ζ
= (−P 0 0 − P 0 β z β )z α ∂zα + (P α 0 + P α β z β )∂zα . With the skewHermitian property P t = −P¯ , and introducing the decomposition 0 E := P 0 is pure imaginary, t ¯ E C P = , where C = (C α ) := (P α 0 ) ∈ Cm , C Q Q = (Qα β ) := (P α β ) ∈ u(m), we see that
¯ α ∂z α . ξ := π∗ (P i j ζ j ∂ζ i ) = Qα β z β + C α + (C¯ · z + E)z
250
DAVID BAO AND COLLEEN ROBLES
The real and imaginary parts of ξ give two real vector fields. A straightforward calculation shows that only Re ξ is Killing. The failure of Im ξ to be Killing persists even for CP 1 . In that case, the skewHermitian Q is simply a pure imaginary number, C is a single complex constant, and LIm ξ h =
2 2 1− 2 ρ ρ
¯ (dx1 ⊗ dx1 + dx¯1 ⊗ dx¯1 ), ¯ + 8 Im(E C) Im(E + Q) ρ3 ¯
where ρ := 1 + z2 . (We hasten to add that for CP m with m > 1, the dxα dxβ and dxα¯ dxβ components of LIm ξ h do not vanish.) Thus, our construction of ξ gives rise to the real Killing fields W := Re ξ = 12 (ξ α ∂zα + ξ α ∂z¯α ) = 21 (Re ξ α )∂xα + (Im ξ α )∂xα¯ ,
where
¯ α. ξ α := Qα β z β + C α + (C¯ · z + E)z 4.2.3. NonRiemannian Einstein metrics on CP m . Theorem 9 (page 238) assures us that the Randers metric F = α + β generated by the navigation data h, W is a globally defined Einstein metric on CP m , provided that the Killing field W satisfies W  < 1. Since, with ρ := 1 + z γ z¯γ , the Fubini–Study metric h :=
2 1 1 δαβ − 2 z¯α zβ (dz α ⊗ d¯ z β + d¯ z β ⊗ dz α ) c ρ ρ
does not have constant sectional curvature for m > 1, Theorem 10 (page 239) ensures that F will not be of constant flag curvature. Moreover, the Ricci scalar of F equals that of h, which has the constant value (m + 1)c/2 (Section 4.2.1). Example. To make explicit the Riemannian metric a underlying α and the 1form b that gives β, it is necessary to have available the covariant description W [ of the Killing field W . If we write W = W B ∂xB and HAB := h(∂xA , ∂xB ), then W [ := (HAB W B ) dxA . Using Hαβ = 2 Re hαβ¯ = Hα¯ β¯ and Hαβ¯ = 2 Im hαβ¯ = −Hαβ ¯ , a computation tells us that z α ). W [ = (Re ξα¯ ) dxα + (Im ξα¯ ) dxα¯ = 21 (ξα¯ dz α + ξα¯ d¯ B β Here, ξα¯ := hαB ¯ ξ = hαβ ¯ ξ has the formula
ξα¯ =
2 ¯ − Qβγ z¯β z γ )zα , ρ(Qαβ z β + Cα ) + (C¯ · z − C · z¯ + E cρ2
where indices on Q, C, z are raised and lowered by the Kronecker delta. As long as W 2 := h(W, W ) = 21 Re(ξ α ξα¯ ) < 1, the Randers metric F with defining data a :=
1 1 h + 2 W [ ⊗ W [, λ λ
1 b := − W [ , λ
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
251
where λ := 1 − W 2 , will be strongly convex. Note that W [ ⊗ W [ does have dz α dz β and d¯ z α d¯ z β components, whereas h doesn’t; thus the Riemannian metric a is not Hermitian unless W = 0. For any choice of the constant quantities Q, C, E, the resulting function W 2 is continuous on the compact CP m , and is therefore bounded. Normalising these quantities by a common positive number if necessary, the strong convexity criterion W 2 < 1 can always be met. ♦ 4.3. Rigidity and a Ricciflat example. In this final set of examples, we consider Einstein–Randers metrics on compact boundaryless manifolds M , with an eye toward those with nonpositive constant Ricci scalar. The information obtained will complement the Riccipositive example presented in Section 4.2.3. We begin by observing that any infinitesimal homothety of (M, h) is necessarily Killing; that is, σ = 0. The proof follows from a divergence lemma (for compact boundaryless manifolds) and the trace of the L W Equation: Z 0= W i :i dVh by the divergence lemma [Bao et al. 2000] M Z = − 21 nσ dVh by tracing Wi:j + Wj:i = −σ hij , M
where dVh :=
√
h dx. In conjunction with Theorem 9 (page 238), we have:
Lemma 14. Let (M, h) be a compact boundaryless Riemannian manifold . Every infinitesimal homothety W must be a Killing field ; equivalently, σ = 0. In particular , if h is Einstein with Ricci scalar Ric, then the navigation data (h, W ) generates an Einstein Randers metric F with Ricci scalar Ric. 4.3.1. Killing fields versus eigenforms. Let W be any Killing vector field on a Riemannian Einstein manifold (M, h) with constant Ricci scalar Ric. Let W [ := Wi dxi denote the 1form dual to W . The action of the Laplace–Beltrami operator ∆ := dδ + δd on W [ is given by the Weitzenb¨ ock formula ∆W [ = (−Wi :j :j + h Rici j Wj ) dxi .
(†)
See, for example, [Bao et al. 2000]. Given that h Ricij = Ric hij , we have h
Rici j Wj = Ric Wi .
By a Ricci identity, W j :j:i −W j :i:j = −h Rici s Ws . Since W is Killing, Wi:j +Wj:i vanishes. Thus W j :j = 0 and −W j :i = Wi :j , which reduce that Ricci identity to −Wi :j :j = h Rici s Ws = Ric Wi . (Note for later use that if W is parallel but 6≡ 0, then Ric = 0.) The Weitzenb¨ ock formula (†) now becomes ∆W [ = 2 Ric W [ .
(∗)
252
DAVID BAO AND COLLEEN ROBLES
Therefore, whenever h is Einstein with constant Ricci scalar Ric, all its Killing 1forms must be eigenforms of ∆, with eigenvalue 2 Ric. Now suppose M is compact and boundaryless, so that integration by parts can be carried out freely without generating any boundary terms. Let h , i denote the L2 inner product on kforms; namely, Z √ 1 hω, ηi := hi1 j1 · · · hik jk ωi1 ···ik ηj1 ···jk h dx. k! M Since the codifferential δ is the L2 adjoint of d, we have 2 Ric hW [ , W [ i = h∆W [ , W [ i = hδW [ , δW [ i + hdW [ , dW [ i > 0.
(‡)
In particular, if Ric is negative, then W must be zero. On the other hand, using (†) and h Ricij = Ric hij , we get Z √ h∆W [ , W [ i = (Wi:j W i:j + Ric W 2 ) h dx. M
This enables us to make a Bochner type argument: if Ric = 0, so that W [ is harmonic by (∗), then Wi:j must vanish identically. 4.3.2. Digression on Berwald spaces. In anticipation of our discussion of nonpositive Ricci curvature, we review Berwald spaces. (See [Szab´ o 1981; 2003] for a complete classification of such spaces.) Let (M, F ) be an arbitrary Finsler space. M need not be compact boundaryless, and F need not be Einstein or of Randers type. Let G i denote the geodesic spray coefficients of F . Then (M, F ) is a Berwald space if the Berwald connection coefficients (Gi )yj yk do not depend on y. In particular, all Riemannian and locally Minkowski spaces are Berwald; for explicit examples belonging to neither of these two camps, see [Bao et al. 2000]. Now suppose F = α + β is Randers but not necessarily Einstein, and has navigation data (h, W ). It is known that F is Berwald if and only if the defining 1form b is parallel. This elegant theorem is due to the efforts of the Japanese school. See [Bao et al. 2000] for an account of the history and references therein; see also the errata for a proof by Mike Crampin. Decompose bij = 21 lieij + 12 curlij into its symmetric and skewsymmetric parts. Look back at the expression for lie and curl at the end of Section 3.1.3. (We reiterate here that they were derived under no assumptions on F .) Observe that Wi:j = 0 implies bij = 0. We prove the converse: suppose bij = 0. ◦ We have 0 = 2bjk W j W k = liejk W j W k = −(1/λ)Wj:k W j W k by referring to Section 3.1.3. Hence Wj:k W j W k = 0. ◦ Using this, a similar calculation gives 0 = 2bjk W j = −(2/λ)W j Wj:k and 0 = 2bjk W k = −(2/λ)Wj:k W k . That is, W j Wj:k = 0 = Wj:k W k . ◦ The formulae for liejk and curljk now simplify to 0 = liejk = −Ljk and 0 = curljk = −(1/λ)Cjk . Hence Wj:k = 12 (Ljk + Cjk ) = 0.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
253
Lemma 15. Let F be any Randers metric, with defining data (a, b) and navigation data (h, W ). The following three conditions are equivalent: • F is Berwald . • b is parallel with respect to a. • W is parallel with respect to h. 4.3.3. A rigidity theorem. We now focus on Einstein Randers metrics of nonpositive constant Ricci scalar, and show that there is considerable rigidity. We begin by addressing the Ricciflat case. We saw at the end of 4.3.1 that Ric = 0 implies that W is parallel. Conversely, if W is parallel and not identically zero, then Ric = 0 (see parenthetic remark just before (∗) on page 251). Whence, in conjunction with Lemma 15 and Proposition 8 (page 234), we obtain: Proposition 16. Let F be an Einstein Randers metric on a compact boundaryless manifold M . • If Ric = 0, then F must be Berwald . • If F is nonRiemannian and Berwald , then Ric = 0. The second conclusion is false if we remove the stipulation that F be nonRiemannian. A Riemannian metric is always Berwald and Randers, and being Einstein certainly does not mandate it to be Ricciflat. Next, we turn our attention to compact boundaryless Einstein Randers spaces of constant negative Ricci scalar. In this case, by (‡) on page 252, W must be identically zero. Equivalently, F = h is Riemannian. Proposition 17. Let F be an Einstein Randers metric with constant negative Ricci scalar on a compact boundaryless manifold M . Then F is Riemannian. Together, these two propositions imply the following rigidity theorem. Theorem 18 (Ricci rigidity). Suppose (M, F ) is a connected compact boundaryless Einstein Randers manifold with constant Ricci scalar Ric. • If Ric < 0, then (M, F ) is Riemannian. • If Ric = 0, then (M, F ) is Berwald . Note that locally Minkowskian spaces, being Berwald and of zero flag curvature, are obvious examples of the second camp. The following arguments show that there exist Ricciflat nonRiemannian Berwald–Randers metrics which are not locally Minkowskian. Example. Take any K3 surface† , namely a complex surface with zero first Chern class and no nontrivial global holomorphic 1forms. All K3 surfaces admit K¨ ahler † According to [Weil 1979, v. 2, p. 546], K3 surfaces are named after Kummer, Kodaira, K¨ ahler, and “the beautiful mountain K2 in Kashmir” — the second tallest peak in the world. One may conjecture that with this last reference Weil was implying that such surfaces are as hard to conquer as the K2. . .
254
DAVID BAO AND COLLEEN ROBLES
metrics (a result due to Todorov and to Siu), and hence Ricciflat K¨ ahler metrics, by Yau’s proof of the Calabi conjecture. Since χ(K3) = 24 by Riemann–Roch, these metrics are not flat by virtue of the Gauss–Bonnet–Chern theorem. See [Besse 1987] for details and references therein. It is futile to consider the Killing fields of such Ricciflat metrics because, by an argument involving Serre duality, the isometry groups in question are all discrete. To circumvent this difficulty, set M := K3 × S 1 (a compact boundaryless real 5manifold) and consider the product metric h on M ; it can be checked that h is also Ricciflat but not flat. The vector field W := 0⊕∂/∂t on M is parallel, hence Killing, with respect to h. Theorem 9 (page 238) tells us that the Randers metric F on M with navigation data (h, W ) is Ricciflat, while Proposition 8 (page 234) guarantees that it is not Riemannian. Theorem 10 (page 239) ensures that F is not of constant (zero) flag curvature; hence it could not be locally Minkowskian. ♦ Theorem 18 generalises a result of AkbarZadeh’s for Finsler metrics of constant flag curvature [AkbarZadeh 1988]: Suppose (M, F ) is a connected compact boundaryless Finsler manifold of constant flag curvature λ. • If λ < 0, then (M, F ) is Riemannian. • If λ = 0, then (M, F ) is locally Minkowski. The Ricci rigidity theorem is a straightforward extension of AkbarZadeh’s result when Ric < 0. To appreciate the generalisation when Ric = 0, it is helpful to note that locally Minkowski spaces are precisely Berwald spaces of constant flag curvature K = 0; see, for instance, [Bao et al. 2000]. AkbarZadeh’s theorem holds for all compact boundaryless Finsler spaces of constant flag curvature, while the Ricci rigidity theorem above is restricted to the Randers setting. So, towards a complete generalisation of AkbarZadeh’s result: What should the conclusions be if we replace ‘Randers’ by ‘Finsler’ in the Ricci rigidity theorem?
5. Open Problems 5.1. Randers and beyond. Table 2 summarises some key information about Randers metrics, which are the simplest members in the much larger family of strongly convex (α, β) metrics. Matusmoto’s slopeofamountain metric (Section 1.1.1) is a prime example from this family. Formal discussions of (α, β) metrics are given in [Matsumoto 1986; Antonelli et al. 1993]; see also [Shen 2004] in this volume. For strongly convex (α, β) metrics, how should the entries of the table be modified? Also, Randers metrics exhibit threedimensional rigidity (Section 3.3.2), and their Ricci scalars obey a Schur lemma (Section 3.3.1).
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
255
Property
Characterisation with (a, b)
Description with (h, W )
Strong convexity
kbk < 1
W  < 1
a
Berwald Constant flag curvature Einstein
a
∇b = 0
h
∇W = 0
Lb] a = σ(a − bb) − (bθ + θb) a R = poly(K, σ, a, b, curl) (Theorem 5)
h is a space form, LW h = −σh
Lb] a = σ(a − bb) − (bθ + θb) Ric = poly(Ric, σ, a, b, curl, a∇θ) (Theorem 6)
h is Einstein, LW h = −σh
Table 2. Summary of information about Randers metrics
• Does the passage from Randers metrics to (α, β) metrics allow us to construct, in three dimensions, a Ricciconstant metric which is not of constant flag curvature? • Does every Einstein (α, β) metric (Ric a function of x only) in dimension > 3 have to be Ricciconstant? Finally, fans of Randers metrics can aim to append an extra row to the above table, characterising Randers metrics of scalar curvature (Section 1.2.1). 5.2. Chern’s question. Professor S.S. Chern has openly asked the following question on many occasions: Does every smooth manifold admit a Finsler Einstein metric? Topological obstructions prevent some manifolds, such as S 2 ×S 1 , from admitting Riemannian Einstein metrics; see Section 1.3.3 and references therein. By the navigation description of Theorem 9 (page 238), any manifold that admits an Einstein Randers metric must also admit a Riemannian Einstein metric. Thus the same topological obstructions confront Einstein metrics of Randers type. As a prelude to answering Chern’s question, it would be prudent to first settle the issue for concrete examples such as S 2 ×S 1 . The discussion above shows that in searching for a Finsler Einstein metric on this 3manifold, we must look beyond those of Randers type.
256
DAVID BAO AND COLLEEN ROBLES
5.3. Geometric flows. On the slit tangent bundle T M r 0, there are two interesting curvature invariants: the Ricci scalar and the Scurvature. They open the door to evolution equations which may be used to deform Finsler metrics. In this frame of mind, we wonder: • Would a flow driven by the Ricci scalar, such as† ∂t log F = −Ric, enable us to prove the existence of Finsler metrics with coveted curvature properties? • Can deformations tailored to the Scurvature be used to ascertain the existence of Landsberg metrics (A˙ = 0) which are not of Berwald type (P = 0)?
Acknowledgments We thank G. Beil and Z. Shen for discussions about the physics and the geodesics, respectively, in Zermelo’s navigation problem. We thank A. Todorov for information about K3 surfaces, and P.M. Wong for guidance on the Killing fields of CP m . Sections 3.2.1, 3.2.2, 3.3, 4.1.2, and 4.3 are based on Robles’ doctoral dissertation.
References [AkbarZadeh 1988] H. AkbarZadeh, “Sur les espaces de Finsler a ´ courbures sectionnelles constantes”, Acad. Roy. Belg. Bull. Cl. Sci. 74 (1988), 281–322. [Antonelli et al. 1993] P. L. Antonelli, R. S. Ingarden and M. Matsumoto, The theory of sprays and Finsler spaces with applications in physics and biology, Fundamental theories of physics 58, Kluwer, Dordrecht, 1993. [Aubin 1998] T. Aubin, Some nonlinear problems in Riemannian geometry, Springer monographs in mathematics, Springer, Berlin, 1998. [Auslander 1955] L. Auslander, “On curvature in Finsler geometry”, Trans. Amer. Math. Soc. 79 (1955), 378–388. [Bao et al. 2000] D. Bao, S.S. Chern and Z. Shen, An introduction to Riemann–Finsler geometry, Graduate Texts in Mathematics 200, Springer, New York, 2000. [Bao et al. 2003] D. Bao, C. Robles and Z. Shen, “Zermelo navigation on Riemannian manifolds”, to appear in J. Diff. Geom. [Bao–Robles 2003] D. Bao and C. Robles, “On Randers spaces of constant flag curvature”, Rep. on Math. Phys. 51 (2003), 9–42. [Bao–Shen 2002] D. Bao and Z. Shen, “Finsler metrics of constant positive curvature on the Lie group S 3 ”, J. London Math. Soc. 66 (2002), 453–467. [Bejancu–Farran 2002] A. Bejancu and H. Farran, “Finsler metrics of positive constant flag curvature on Sasakian space forms”, Hokkaido Math. J. 31:2 (2002), 459–468. † The equation for ∂ log F has its genesis in ∂ g t t ij = −2 Ricij , which is formally identical to the unnormalised Ricci flow of Richard Hamilton for Riemannian metrics (see [Cao–Chow 1999] and references therein). In the Finsler setting, this equation makes sense on T M r 0. Contracting with y i , y j (and using Sections 1.1.1 and 1.3.1) gives ∂t F 2 = −2F 2 Ric. Since F is strictly positive on the slit tangent bundle, the stated equation follows readily.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
257
[Bejancu–Farran 2003] A. Bejancu and H. Farran, “Randers manifolds of positive constant curvature”, Int. J. Math. Mathematical Sci. 18 (2003), 1155–1165. [Berger 1971] M. Berger, “On Riemannian structures of prescribed Gaussian curvature for compact 2manifolds”, J. Diff. Geom. 5 (1971), 325–332. ¨ [Berwald 1929] L. Berwald, “Uber die ndimensionalen Geometrien konstanter Kr¨ ummung, in denen die Geraden die k¨ urzesten sind”, Math. Z. 30 (1929), 449–469. [Berwald 1947] L. Berwald, “Ueber Finslersche und Cartansche Geometrie IV. Projektivkr¨ ummung allgemeiner affiner R¨ aume und Finslersche R¨ aume skalarer Kr¨ ummung”, Ann. Math. (2) 48 (1947), 755–781. [Besse 1987] A. L. Besse, Einstein manifolds, Ergebnisse der Math. (3) 10, Springer, Berlin, 1987. [Bryant 1997] R. Bryant, “Projectively flat Finsler 2spheres of constant curvature”, Selecta Math. (N.S.) 3 (1997), 161–203. [Bryant 2002] R. Bryant, “Some remarks on Finsler manifolds with constant flag curvature”, Houston J. Math. 28:2 (2002), 221–262. [Cao–Chow 1999] H.D. Cao and Bennett Chow, “Recent developments on the Ricci flow”, Bull. Amer. Math. Soc. 36 (1999), 59–74. [Carath´eodory 1999] C. Carath´eodory, Calculus of variations and partial differential equations of the first order , 2 v., HoldenDay, San Francisco, 1965–1967; reprinted by Chelsea, 1999. [Cheeger–Ebin 1975] J. Cheeger and D. Ebin, Comparison theorems in Riemannian geometry, North Holland, Amsterdam, 1975. [Chen–Shen 2003] X. Chen and Z. Shen, “Randers metrics with special curvature properties”, Osaka J. Math. 40 (2003), 87101. [Foulon 2002] P. Foulon, “Curvature and global rigidity in Finsler manifolds”, Houston J. Math. 28 (2002), 263–292. ¨ [Funk 1929] P. Funk, “Uber Geometrien, bei denen die Geraden die K¨ urzesten sind”, Math. Ann. 101 (1929), 226–237. [Hitchin 1974] N. Hitchin, “On compact fourdimensional Einstein manifolds”, J. Diff. Geom. 9 (1974), 435–442. [Hrimiuc–Shimada 1996] D. Hrimiuc and H. Shimada, “On the Lduality between Lagrange and Hamilton manifolds”, Nonlinear World 3 (1996), 613–641. [Ingarden 1957] R. S. Ingarden, “On the geometrically absolute optical representation in the electron microscope”, Trav. Soc. Sci. Lettr. Wroclaw B45 (1957), 3–60. [Kobayashi–Nomizu 1996] S. Kobayashi and K. Nomizu, Foundations of differential geometry, v. 2, Interscience, New York, 1969; reprinted by Wiley, New York, 1996. [LeBrun 1999] C. LeBrun, “Fourdimensional Einstein manifolds, and beyond”, pp. 247– 285 in [LeBrun–Wang 1999]. [LeBrun–Wang 1999] C. LeBrun and M. Wang (editors), Surveys in differential geometry: essays on Einstein maniolds, v. 6, International Press, Boston, 1999. [Matsumoto 1986] M. Matsumoto, Foundations of Finsler geometry and special Finsler spaces, Kaiseisha Press, Japan, 1986.
258
DAVID BAO AND COLLEEN ROBLES
[Matsumoto 1989] M. Matsumoto, “A slope of a mountain is a Finsler surface with respect to a time measure”, J. Math. Kyoto Univ. 29:1 (1989), 17–25. [Matsumoto–Shimada 2002] M. Matsumoto and H. Shimada, “The corrected fundamental theorem on Randers spaces of constant curvature”, Tensor (N.S.) 63 (2002), 43–47. [Numata 1978] S. Numata, “On the torsion tensors Rjhk and Phjk of Finsler spaces with a metric ds = (gij (dx)dxi dxj )1/2 + bi (x)dxi ”, Tensor (N.S.) 32 (1978), 27–32. [Okada 1983] T. Okada, “On models of projectively flat Finsler spaces of constant negative curvature”, Tensor (N.S.) 40 (1983), 117–124. [Rademacher 2004] H.B. Rademacher, “Nonreversible Finsler metrics of positive flag curvature”, pp. 263–304 A sampler of Riemann–Finsler geometry, edited by D. Bao et al., Math. Sci. Res. Inst. Publ. 50, Cambridge Univ. Press, Cambridge, 2004. [Randers 1941] G. Randers, “On an asymmetrical metric in the fourspace of general relativity”, Phys. Rev. 59 (1941), 195–199. ¨ [Rapcs´ ak 1961] A. Rapcs´ ak, “Uber die bahntreuen Abbidungen metrischer R¨ aume”, Publ. Math., Debrecen 8 (1961), 285–290. [del Riego 1973] L. del Riego, Tenseurs de Weyl d’un spray de directions, doctoral dissertation, Universit´e Scientifique et M´edicale de Grenoble, 1973. [Robles 2003] C. Robles, Einstein metrics of Randers type, doctoral dissertation, University of British Columbia, 2003. [Rund 1959] H. Rund, The differential geometry of Finsler spaces, Grundlehren der math. Wissenschaften 101, Springer, Berlin, 1959. [Sabau 2003] S. V. Sabau, “On projectively flat Finsler spheres (remarks on a theorem of R. L. Bryant)”, pp. 181–191 in Finsler and Lagrange Geometries (Ia¸si, 2001), edited by M. Anastasiei and P. L. Antonelli, Kluwer, Dordrecht, 2003. [Shen 2001] Z. Shen, Differential geometry of sprays and Finsler spaces, Kluwer, Dordrecht 2001. [Shen 2002] Z. Shen, “Finsler metrics with K = 0 and S = 0”, Canadian J. Math. 55 (2003), 112–132. [Shen 2003] Z. Shen, “Projectively flat Finsler metrics of constant flag curvature”, Trans. Amer. Math. Soc. 355 (2003), 1713–1728. [Shen 2004] Z. Shen, “Landsberg curvature, Scurvature and Riemann curvature”, pp. 305–356 in A sampler of Riemann–Finsler geometry, edited by D. Bao et al., Math. Sci. Res. Inst. Publ. 50, Cambridge Univ. Press, Cambridge, 2004. [Singer–Thorpe 1969] I. M. Singer and J. A. Thorpe, “The curvature of 4dimensional Einstein spaces”, pp. 355–365 in Global analysis: papers in honour of K. Kodaira, edited by D. C. Spencer and S. Iyanaga, Princeton mathematical series 29, Princeton Univ. Press, Princeton, 1969. [Szab´ o 1981] Z. I. Szab´ o, “Positive definite Berwald spaces (structure theorems on Berwald spaces)”, Tensor (N.S.) 35 (1981), 25–39. [Szab´ o 2003] Z. I. Szab´ o, “Classification of Berwald and symmetric Finsler manifolds via explicit constructions”, preprint, 2003.
RICCI AND FLAG CURVATURES IN FINSLER GEOMETRY
259
[Thorpe 1969] J. Thorpe, “Some remarks on the Gauss–Bonnet formula”, M. Math. Mech. 18 (1969), 779–786. [Tian 1997] G. Tian, “K¨ ahler–Einstein metrics with positive scalar curvature”, Invent. Math. 130 (1997), 1–39. [Weil 1979] Andr´e Weil, Œuvres scientifiques, Springer, Berlin, 1979. ¨ [Zermelo 1931] E. Zermelo, “Uber das Navigationsproblem bei ruhender oder ver¨ anderlicher Windverteilung”, Z. Angew. Math. Mech. 11 (1931), 114–124. [Ziller 1982] W. Ziller, “Geometry of the Katok examples, Ergodic Theory Dynam. Systems 3 (1982), 135–157. David Bao Department of Mathematics University of Houston Houston, TX 772043008 United States
[email protected] Colleen Robles Department of Mathematics University of Rochester Rochester, NY 146270001 United States
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Nonreversible Finsler Metrics of Positive Flag Curvature HANSBERT RADEMACHER
Contents 1. Introduction 2. Conventions 3. Finsler Metrics 4. First Variation of the Energy Functional 5. Flag curvature, Jacobi Fields and Conjugate Points 6. Second Variation of the Energy Functional 7. Results from Topology 8. Morse Theory of the Energy Functional 9. Shortest Nonminimal Geodesics and the Sphere Theorem 10. Length of Closed Geodesics in Even Dimensions 11. An Example References
261 263 265 271 273 276 278 283 289 297 299 300
1. Introduction Finsler geometry is an essential extension of Riemannian geometry. Instead of an inner product on every tangent space one considers Minkowski norms on every tangent space. For a Finsler metric the unit sphere in each tangent space is a strictly convex hypersurface. One obtains for every nonzero tangent vector an inner product, arising from Minkowski norm; in the Riemannian case these inner products all coincide on a fixed tangent space. The length of a smooth curve is welldefined. Geodesics — locally lengthminimizing curves parametrized with constant speed — are uniquely defined for a given initial direction. From the viewpoint of the calculus of variations Finsler metrics are a suitable generalization of Riemannian metrics such that the variational problem for the length of curves between two fixed points is positive and positive regular. In terms of physics a Finsler metric describes a Lagrangian system without a potential; a Riemannian metric can be viewed as the special case of quadratic kinetic energy. 261
262
HANSBERT RADEMACHER
But in contrast to the Riemannian case there is no canonical connection, so several connections have been used in Finsler geometry. We use here the one introduced by S.S. Chern [Bao et al. 2000, Chapter 2], transposed to vector fields on the manifold for a fixed direction field: Given a nowhere vanishing vector field V in an open nonempty subset U , there is a uniquely determined torsionfree connection ∇V that is almost metric. Using this connection one can define the flag curvature, which generalizes the sectional curvature in Riemannian geometry and controls the infinitesimal behavior of geodesics. Given a geodesic c and a nowhere vanishing geodesic vector field V in an open neighborhood of c extending the velocity field c0 of the geodesic, there is a Riemannian metric gV on U such that c is also a geodesic of the Riemannian manifold (U, g V ) and the flag curvature K(c0 ; σ) for any plane σ containing c0 coincides with the sectional curvature K(σ) of gV . In particular the Jacobi fields of the Finsler metric and of gV coincide along the geodesic c. Thus the flag curvature can be introduced without selecting a connection. The flag curvature does not completely determine the metric. For example, in contrast to the Riemannian case, there are Finsler metrics of constant positive flag curvature on spheres that are not isometric to the standard Riemannian metric; see [Bao et al. 2003; Bryant 2002; Shen 2002] for their characterization. We call a Finsler metric F reversible (or symmetric) if opposite vectors have the same length: F (X) = F (−X) for all tangent vectors X. In this case the unit sphere Tp1 M = {X ∈ Tp M  F (X) = 1} is symmetric under reflection through the origin. But this assumption excludes many interesting examples, for example Randers metrics, which are Finsler metrics defined pby adding a oneform to the norm induced by a Riemannian metric: F (X) = g(X, X) + α(X), where g is a Riemannian metric and α is a oneform. In Riemannian geometry a metric with constant positive sectional curvature on a compact simply connected manifold is isometric to the standard sphere of the same curvature. The now classical Sphere Theorem states that a compact, simply connected manifold of dimension n with sectional curvature K such that 1 4 < K ≤ 1 everywhere is homeomorphic to the nsphere [Klingenberg 1995, § 2.8; Abresch and Meyer 1997]. In this form the result is contained in [Klingenberg 1961]; earlier contributions are due to M. Berger, H. Rauch and V. A. Toponogov [Berger 1998, I A 2]. The proof uses an estimate for the injectivity radius and the Toponogov comparison theorem for geodesic triangles. In [Klingenberg 1963] it is shown that one can prove the Sphere Theorem without making use of Toponogov’s comparison result for geodesic triangles. Instead one uses Morse theory of the energy functional on the space of curves between two fixed points and on the space of loops. The injectivity radius is bounded from below by π, so geodesic loops have length at least 2π and their Morse index is bounded below by n−1. This implies that the loop space is (n−2)connected and therefore the manifold is homotopy equivalent to the nsphere.
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
263
Though in Finsler geometry the condition of constant positive flag curvature no longer determines the metric up to isometry, one can show using the exponential map that a simply connected Finsler manifold of constant positive flag curvature is homeomorphic to the nsphere. P. Dazord [1968a; 1968b] remarked that in the case of a reversible Finsler metric one can carry over the Morsetheoretic proof of the Sphere Theorem found in [Klingenberg 1963]. The original proof of the Sphere Theorem does not carry over, since the triangle comparison result cannot be extended to the Finsler case. The main topic of this article is to show in detail how the estimates for the injectivity radius, the length of a nonminimal geodesic between two fixed points, the length of a nonconstant geodesic loop and the length of a nonconstant closed geodesic can be extended to the case of a nonreversible Finsler metric by introducing the notion of reversibility λ := sup{F (−X)  F (X) = 1} ≥ 1. We will derive from a length estimate (Proposition 9.9) and from Theorem 9.10 the following Sphere Theorem for nonreversible Finsler metrics: Theorem 9.11. A simply connected and compact Finsler manifold of dimension 2 1 0, we have 1 d2 1 d2 2 hV, V iV = F (V + tV ) = (1 + t)2 F 2 (V ) = F 2 (V ) 2 dt2 t=0 2 dt2 t=0
In coordinates,
1 i j ∂2 yy F 2 (x, y) = F 2 (x, y). 2 ∂y i ∂y j (c) Euler’s Theorem states that for a positively homogeneous function f : V → R of order k on a vector space V (meaning that f (µX) = µk f (X) for X ∈ V and µ > 0), the radial derivative coincides up to the factor k with f itself: n X
yi
∂ f (y) = kf (y). ∂y i
i=1
See [Bao et al. 2000, Theorem 1.2.1]. Definition 3.3. The Legendre transformation on a Finsler manifold (M, F ) is the map LF : T M → T ∗ M defined by LF (V )(W ) = gV (V, W ). One can view LF (V ) as the 1form dual to V with respect to the metric g V .
266
HANSBERT RADEMACHER
Lemma 3.4. If F is a Finsler metric, F (X + Y ) ≤ F (X) + F (Y ) for all X, Y ∈ Tx M . Equality holds only if Y = µX for some µ ≥ 0. This implies that hX, Y iY ≤ F (X)F (Y ) for all X, Y ∈ Tx M , with equality if and only if Y = µX for some µ ≥ 0 [Shen 2001a, § 1.2; Bao et al. 2000, p. 10]. Lemma 3.5 [Bao et al. 2000, Proposition 14.8.1; Shen 2001a, Lemma 1.2.4]. Let V, W ∈ Tx M be nonzero. If hX, V iV = hX, W iW for all X ∈ Tx M , then V = W . Consequently, the Legendre transformation is an isomorphism. A Riemannian metric g on a manifold M is symmetric (2, 0)tensor field g : VM × VM → FM such that for every x ∈ M the bilinear map gx : Tx M × Tx M → p R is positive definite. The associated Finsler metric is defined by F (X) = g(X, X); forming the metric gV for any nonzero V we obtain g for every nonzero V . Example 3.6 (Randers metrics). Suppose given a Riemannian metric α and a differential 1form β. There is a vector field ζ satisfying β(X) =p α(X, ζ) for all X; we say ζ is dual to β with respect to α. Define kβk := kζk = α(ζ, ζ). If kβk < 1 everywhere, p p F (X) := α(X, X) + β(X) = α(X, X) + α(X, ζ) defines a Finsler metric. This type p of Finsler metric is called a Randers metric. Since X 6= 0 implies kXk = α(X, X) > 0, we have
F (X) = kXk 1 + α
X
kXk
,ζ
≥ kXk (1 − kζk) > 0,
showing that F satisfies condition (2) of Definition 3.1. Condition (1) is obvious. For the proof of (3) we refer to [Bao et al. 2000, Chapter 11]. p A Randers metric F (X) = α(X, X) + α(X, ζ) is only positively homogeneous. If α(X, ζ) 6= 0 then F (−X) 6= F (X). This motivates the following notion: Definition 3.7. On a Finsler manifold (M, F ) the reversibility function λ : M → R + is defined by λ(x) := sup F (−X)  X ∈ Tx M, F (X) = 1 . The number λ = λ(M, F ) = sup λ(x)  x ∈ M is called the reversibility of the Finsler manifold (M, F ), if it exists — for example, if M is compact. One has to show that the function λ is continuous, which is done using a standard argument: The subspace Tx1 M = {X ∈ Tx M  F (X) = 1} is called the unit sphere or indicatrix at the point x. It is a compact space diffeomorphic to the sphere S n−1 . The subspaces Tx1 M , x ∈ M , form a sphere bundle over M . The function X ∈ Tx1 M 7→ F (−X) ∈ R is continuous, therefore the supremum is actually the maximum of this function. The sphere bundle T 1 M → M is locally trivial, that is, for small open sets U ⊂ M the restriction T 1 U → U can
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
267
be identified via a fiberpreserving diffeomorphism with the canonical projection U ×S n−1 → U on the first factor. The function T 1 U ∼ = U ×S n−1 7→ F (−X) ∈ R is continuous. A proof by contradiction, using the sequential compactness of S n−1 , then shows that the reversibility function λ : M → R is continuous. If M is compact, the function is bounded and the supremum is actually a maximum. Since F is positively homogeneous we could also write λ(x) = max
o n F (−X) X ∈ Tx M, X 6= 0 . F (X)
We call a Finsler metric reversible if F (−X) = F (X) for all X ∈ T M . Then obviously λ = 1. If there is a tangent vector X such that F (−X) 6= F (X) then F (−X) F (X) > 1. , λ ≥ max F (X) F (−X) Hence a Finsler metric is reversible if and only if λ = 1. In this case the indicatrix Tp1 M is symmetric with respect to reflection X 7→ −X. Sometimes a reversible metric is also called symmetric, but this terminology conflicts with other notions such as symmetric quadratic forms and symmetric spaces. For an arbitrary Finsler metric on a compact manifold M we obtain λ−1 F (X) ≤ F (−X) ≤ λF (X). If γ : [0, 1] → M is a smooth curve on M , we define the length of γ as L(γ) = R1 F (γ 0 (t)) dt. We also introduce γ −1 : [0, 1] → M , the curve γ run in reverse: 0 −1 γ (t) = γ(1 − t). The lengths of γ and γ −1 satisfy 1 L(γ) ≤ L(γ −1 ) ≤ λL(γ). (3–1) λ p Example 3.8. Let F (X) = α(X, X) + α(X, ζ) be a Randers metric, with vector field ζ. For fixed x ∈ M , we will find λ(x) by looking at the quotient F (−X)/F (X) on the unit ball {X ∈ Tx M  α(X, X) = 1} of α: X 7→
1 − α(X, ζ(x)) F (−X) = . F (X) 1 + α(X, ζ(x))
This quotient attains its maximum for X = −ζ(x)/kζ(x)k, and we obtain λ(x) =
1 + kζ(x)k . 1 − kζ(x)k
For a nonzero tangent vector V ∈ Tx M , we define on Tx M the trilinear form hX1 , X2 , X3 iV := 41 X 1 X 2 X 3 F 2 (V )
3
X ∂3 1 si X i . F2 V + = 4 ∂s1 ∂s2 ∂s3 (s1 ,s2 ,s3 )=(0,0,0) i=1
For a given everywhere nonzero vector field V defined on an open subset U ⊂ M , we obtain a symmetric (0, 3)tensor, called the Cartan tensor ; its coefficients are
268
HANSBERT RADEMACHER
usually denoted Cijk (that is, the functions Cijk at V ∈ Tx M express the trilinear form h , , iV in a given system of canonical coordinates). The Cartan tensor vanishes if and only if the Finsler metric comes from a Riemannian metric g (meaning that F 2 (X) = g(X, X)). Euler’s theorem implies that hV, X, Y iV = hX, V, Y iV = hX, Y, V iV = 0 (3–2) for all vector fields X, Y . A distance function on a differentiable manifold M is a smooth function θ : M × M → [0, ∞) such that θ(p, q) = 0 if and only if p = q, and such that the triangle inequality is satisfied: θ(p, q) ≤ θ(p, r) + θ(r, q)
for all p, q, r ∈ M.
Given a connected Finsler manifold (M, F ), the induced distance θ : M × M → R is defined by θ(p, q) := inf {L(c)  c : [0, 1] → M piecewise smooth, c(0) = p, c(1) = q}. (Piecewise smooth means that c is continuous and there is a finite partition of the interval [0, 1] such the restriction of c to each closed subinterval is smooth.) It is easy to check that θ is a distance function. If the Finsler metric is nonreversible, the induced metric is not symmetric: there are points p, q with θ(p, q) 6= θ(q, p). Lemma 3.9. The induced distance of a Finsler manifold (M, F ) with reversibility λ ≥ 1 satisfies 1 θ(p, q) ≤ θ(q, p) ≤ λθ(p, q) (3–3) λ Proof. For every k ∈ N, let γk : [0, 1] → M be a piecewise smooth curve with p = γk (0), q = γk (1) and L(γk ) ≤ θ(p, q) + 1/k. Then (3–1) gives L(γk−1 ) ≤ λθ(p, q) + λ/k for all k. Thus θ(q, p) ≤ L(γk ) ≤ λ θ(p, q) + 1/k for all k. ˜
Given a Finsler manifold (M, F ), the symmetrized distance d : M × M → R is defined by d(p, q) = 21 θ(p, q) + θ(q, p) . The distance functions θ and d of a Finsler manifold coincide if and only if the Finsler metric is reversible. For U an open subset of a manifold M , recall that VU is the space of smooth vector fields on U , and let VU + ⊂ VU be the subset of nowhere vanishing vector fields. For the next theorem we recall the definition of an affine connection: a map ∇V : (X, Y ) ∈ VU × VU 7→ ∇VX Y ∈ VU , linear in Y and satisfying
∇VX (f Y ) = f ∇VX Y + X(f )Y and ∇VfX Y = f ∇X Y for all f ∈ FU, X, Y ∈ VU . Theorem 3.10. Let (M, F ) be a Finsler manifold and U ⊂ X an open subset. There is a map ∇ : (V, X, Y ) ∈ VU + × VU × VU 7→ ∇VX Y ∈ VU with the following properties:
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
269
(a) for every V ∈ VU + , the map ∇V : (X, Y ) ∈ VU × VU 7→ ∇VX Y ∈ VU is an affine connection. (b) ∇V is torsionfree, that is, ∇VX Y − ∇VY X = [X, Y ]
for all X, Y ∈ VU .
(c) ∇V is almost metric, that is, X hY, ZiV = h∇VX Y, ZiV + hY, ∇VX ZiV + 2h∇VX V, Y, ZiV .
(3–4)
(3–5)
Moreover we have 2h∇VX Y, ZiV = X hY, ZiV + Y hZ, XiV − Z hX, Y iV
+ [X, Y ], Z V − [Y, Z], X V + [Z, X], Y V
− 2h∇VX V, Y, ZiV − 2h∇VY V, Z, XiV + 2h∇VZ V, X, Y iV
(3–6)
for all vector fields X, Y, Z ∈ VU , and this equation, called the generalized Koszul formula, uniquely determines ∇. Sketch of proof. From the requirements (3–4) and (3–5) we obtain, through straightforward calculations, the generalized Koszul formula (3–6). Equations (3–5), (3–6) and (3–2) then imply that
h∇VV V, ZiV = 2V hV, ZiV − Z hV, V iV + 2 [Z, V ], V V and
2h∇VX V, ZiV = X hV, ZiV + V hZ, XiV − Z hX, V iV
+ [X, V ], Z V − [V, Z], X V + [Z, X], V V − 2h∇VV V, Z, XiV .
(3–7)
Thus the righthand side of (3–6) can be expanded into an expression devoid of any reference to ∇, showing that ∇VX Y is uniquely determined. Then one has to check that the ∇V thus defined is in fact an affine connection, torsionfree and almost metric. ˜ Remark 3.11. (a) In the Riemannian case the connection ∇ V = ∇ is indepen dent of V ∈ VU , it is metric, meaning that X hY, Zi = h∇X Y, Zi + hY, ∇X Zi, and it is determined by the Koszul formula: 2h∇X Y, Zi = X hY, Zi + Y hZ, Xi − Z hX, Y i
+ [X, Y ], Z − [Y, Z], X + [Z, X], Y .
(3–8)
∇ is called the LeviCivita connection or canonical connection.
(b) We point out the correspondence between the development adopted above (which one can find in [Matthias 1980, Chapter 2]) and the description given in [Bao et al. 2000, Chapter 2] and [Shen 2001a, § 5.2]. In canonical coordinates
270
HANSBERT RADEMACHER
(x, y) = (x1 , . . . , xn , y 1 , . . . , y n ) of the tangent bundle T U of an open set U ⊂ M , we obtain the functions gij : (x, y) ∈ T U 7→ gij (x, y) = g(x, y)
∂
∂x
, i
∂ ∂xj
=
1 ∂2 F 2 (x, y). 2 ∂y i ∂y j
The coefficients Cijk = Cijk (x, y) of the Cartan tensor are Cijk (x, y) =
1 ∂2 F 2 (x, y) = 4 ∂y i ∂y j
D ∂
∂x
, i
E
∂ ∂ , . ∂xj ∂xk
We also define g ij (x, y) as the coefficients of the inverse matrix of gij (x, y). Then i one can define formal Christoffel symbols γjk : T U → R: i γjk (x, y) = 12 g il (x, y)
∂g
lj ∂xk
(x, y) −
∂gjk ∂gkl (x, y) + (x, y) . ∂xl ∂xj
We can raise and lower indices by contracting with the coefficients g ij and gij ; i for example, Cjk = g il Cljk . Here we use the Einstein summation convention. Then we define the quantities i i k r s Nji = Nji (x, y) = γjk (x, y)y k − Cjk (x, y)γrs y y .
It turns out that the coefficients of the Chern connection are i Γijk (x, y) = γjk − g li (Cijr Nkr − Cjkr Nir + Ckir Njr ),
(see [Bao et al. 2000, (2.4.9)]), the Chern connection is given by ∇∂/∂xi
∂ ∂ (x, y) = Γkij (x, y) k , ∂xj ∂x
and one shows that Nij (x, y) = Γjik (x, y)y k . The Chern connection is torsionfree, which implies that Γ kij = Γijk . It is also almost metric, which implies (see [Shen 2001a, (5.22), (5.29)]) ∂gjl k (x, y) = gkl Γijm + gkj Γklm + 2Cjkl Nm ∂xm = gkl Γijm + gkj Γklm + 2Cjkl Γkmr y r . In particular: Lemma 3.12. For two vector fields V, W ∈ VU + and a point p ∈ U with V (p) = W (p) and for arbitrary vector fields X, Y ∈ VU we have: ∇VX Y (p) = ∇W X Y (p).
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
271
Using the connection ∇V we introduce the covariant derivative ∇V/dt along a curve c : [a, b] → M . For a vector field X along the curve c with tangent vector field c0 define (∇V/dt)X(t) = ∇Vc0 X(t), where on the righthand side one has to take extensions of the vector fields V, X, c0 onto an open subset containing the curve. This expression is independent of the chosen extensions. If the vector fields V, c0 coincide, we also write simply (∇V/dt)X = (∇/dt)X. For a differentiable map H : [0, 1] × [0, 1] → M and a vector field X(s, t) along F (meaning that X(s, t) ∈ TH(s,t) M ), we define (∇V/∂t)X(t) as a vector field along the curve t 7→ H(s1 , t) for a fixed s1 and (∇V/∂s)X(t) as a vector field along the curve s 7→ H(s, t1 ) for a fixed t1 . Then we obtain the following rule for exchanging the order of differentiation: ∇V ∂H ∇V ∂H = ∂t ∂s ∂s ∂t This rule follows since the connection ∇V is torsionfree.
4. First Variation of the Energy Functional In the Morsetheoretic proof of the Sphere Theorem we use the energy functional E on a suitable space of curves as the Morse function. For a smooth curve c : [0, 1] → M , the energy is defined as Z 1 1 2 0 F c (t) dt. E(c) = 2 0 For a variation cs in the first variation formula one studies the first derivative (d/ds)s=0 E(cs ): Lemma 4.1 (First variational formula). If cs : [a, b] → M , for s ∈ (−ε, ε), is a smooth variation of the curve c = c0 with variation vector field V (t) = (∂/∂s)s=0 cs (t), then Z bD E
0
0 d ∇ 0 dt. (4–1) c ,V E(cs ) = c (b), V (b) c0 (b) − c (a), V (a) c0 (a) − ds s=0 dt c0 a Proof.
1 ∂ 0 0 hc , c ic0 = 2 ∂s s s s
D ∇c0s ∂s
D
c0s , c0s
E
∂ ∂cs 0 ,c = ∂t ∂s s
+
c0s
E
c0s
−
D ∇c0s
E
D
∂cs 0 c0s , c0s , c0s ,c = (∇/∂t) ∂s ∂s s c0s  {z }
D ∂c
s
∂s
=0
, (∇/∂t)c0s
E
c0s
−
D ∇c0s
E
c0s
E
∂cs 0 c0s , ,c ∂s ∂s s c0s  {z } =0
Hence we conclude that Z b Z b
b
d 0 0 0 0 − hc , c i dt = V (t), c (t) V (t), (∇/dt)c0 c0 dt. 0 s s cs c ds s=0 a a a
˜
272
HANSBERT RADEMACHER
Corollary 4.2. If c : [0, 1] → M is a piecewise smooth curve such that no other piecewise smooth curve joining p = c(0) and q = c(1) is shorter , then c is a geodesic, that is, c is smooth and (∇/dt)c0 = 0. Proof. Let c be smooth when restricted to each subinterval [t j , tj+1 ] of a partition 0 = t0 < t1 < · · · < tk = 1 of [0, 1]. We first want to prove that these restrictions are geodesics, so that c is a broken geodesic (also known as a geodesic polygon). If (∇/dt)c0 (s) 6= 0 for some s ∈ (tj , tj+1 ), we choose a vector field V (t) along the image of c as follows: V (t) = φ(t)
∇ 0 c (t), dt
where φ : [0, 1] → [0, 1] is a smooth function with φ(s) = 1 and φ(t) = 0 for t∈ / (tj , tj+1 ). Then we take a smooth variation cs : [0, 1] → M of the piecewise smooth curve c with variation vector field V (t) = (∂/∂s) s=0 cs (t). (Saying that the family cs is a smooth variation of c is saying that, for each j = 0, . . . , k − 1, the restrictions cs [tj , tj+1 ] form a smooth variation of c[tj , tj+1 ].) Then the first variation formula (Lemma 4.1) gives 0=
d E(cs ) = ds s=0
Z 1D 0
∇ 0 c ,V dt
E
dt =
c˙
Z
1 0
∇ 0 2 c dt.
φ(t)
dt
Since φ(t)≥0 for all t and φ(s)=1, the righthand side is positive, a contradiction. Hence no such s exists, and c is a broken geodesic. Now fix l ∈ {1, 2, . . . , k−1} and choose any tangent vector V 0 ∈ Tc(tl ) M and a variation vector field V = V (t) along the broken geodesic c such that V (t l ) = V0 and V (tj ) = 0 for j 6= l. Again, Lemma 4.1 shows that 0=
d − c0 (t− E(cs ) = c0 (t+ l ), V0 c0 (t+ l ), V0 c0 (t− l ) l ) ds s=0
for all V0 ∈ Tc(tl ) M . (Here, as usual, c0 (t± ) = limε→0, ε>0 c0 (t ± ε).) We con0 − clude that c0 (t+ l ) = c (tl ), since the Legendre transformation is an isomorphism (Lemma 3.5). Hence c is a smooth curve. ˜ A vector field V ∈ VU is called a geodesic vector field if ∇VV V = 0, which says that the flow lines of V are geodesics of the Finsler metric. These lines can also be seen as geodesics of an associated Riemannian metric: Lemma 4.3. Let V be a nowhere vanishing geodesic field defined on an open subset U ⊂ M . Denote by ∇ the LeviCivita connection of the Riemannian manifold (U, gV ). Then ∇VX V = ∇X V for all vector fields X; in particular , the vector field V is also geodesic for the Riemannian manifold (U, gV ).
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
273
Proof. ∇ is uniquely determined by the Koszul formula (see (3–8), with ∇ playing the role of ∇). Since ∇VV V = 0 we conclude from Equation 3–7 that h∇X V, ZiV = h∇VX V, ZiV
(4–2)
for all vector fields X, Y .
˜
We obtain a similar statement if we restrict to vector fields along a given geodesic: Lemma 4.4. Let c : [0, 1] → M be a nonselfintersecting geodesic of the Finsler manifold (M, F ), and V ∈ VU + an extension of the velocity vector field c0 onto an open neighborhood U of c([0, 1]). We call the Riemannian manifold (U, gV ) an osculating Riemannian manifold , denote its LeviCivita connection by ∇ and the covariant derivative along c by (∇/dt). Then c is also a geodesic of the osculating Riemannian metric gV and ∇ ∇ X(t) = X(t) dt dt for any vector field X along c. Proof. As in the proof of Lemma 4.3 we show that (4–2) holds along the given geodesic: h∇X V, ZiV (c(t)) = h∇VX V, ZiV (c(t)). Then
(∇V/dt)c0 X(t) = ∇Vc0 X(t) = ∇VX c0 (t) + [c0 , X] = ∇X c0 + [c0 , X] = ∇Vc0 X(t) = (∇/dt)X(t).
For X = c0 it follows that c is also a geodesic of the osculating Riemannian metric. ˜
5. Flag curvature, Jacobi Fields and Conjugate Points For a Riemannian manifold (M, g) with LeviCivita connection ∇, the Riemann curvature tensor is a (1, 3)tensor defined by R(X, Y )Z = ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z,
forX, Y, Z ∈ VM .
It is determined by the Jacobi operators or directional curvature operators R X , X ∈ Tx M , given by Y ∈ Tx M 7→ RX (Y ) := R(Y, X)X. The sectional curvature K(σ) = K(X, Y ) of a plane σ ⊂ Tx M spanned by the tangent vectors X, Y is defined by
X
R (Y ), Y R(X, Y )Y, X = . K(X, Y ) = X2 Y 2 − hX, Y i2 X2 Y 2 − hX, Y i2 The Finsler geometry counterparts of these entities can be introduced by considering the osculating Riemannian metric. In the next statement we make
274
HANSBERT RADEMACHER
use of the notion of a geodesic variation of a geodesic c; this is simply a variation cs of c such that each curve cs : [0, 1] → M is geodesic. Proposition 5.1. (a) [Shen 2001b, Lemma 8.1.1] For every nonzero tangent vector X ∈ Tx M on a Finsler manifold M with geodesic c = cX : [0, 1] → M and c0X (0) = X, the map RX : Tx M → Tx M given by RX (Y (0)) = −
∇2 Y (0), dt2
where Y (t) = (∂/∂s)s=0 cs (t) is the variation vector field of a geodesic variation of c, is welldefined and linear . It is called the Riemann curvature (operator ) of the Finsler manifold . (b) [Shen 2001b, Proposition 8.4.1] For every nonzero tangent vector X ∈ T x M with a nonzero geodesic vector field V extending X ∈ Tx M in an open neighborhood U of x, the Riemann curvature operator RX of the Finsler manifold coincides with the Jacobi operator RX of the osculating Riemannian metric g¯ = g V defined on U . We can now introduce the Finsler counterpart of the sectional curvature. In contrast with the Riemannian case, the notion depends not only on the choice of a twodimensional tangent plane but also on a direction in this plane. Definition 5.2. For a Finsler manifold (M, F ) and a flag (X, σ) consisting of a nonzero tangent vector X ∈ Tx M and a plane σ ⊂ Tx M spanned by the tangent vectors X, Y , the flag curvature is defined as
X R (Y ), Y X . K(X; σ) = K(X; Y ) = X2X Y 2X − hX, Y i2X The notation δ < K ≤ 1, where δ ∈ (0, 1), will be used often; it is a shorthand for the condition δ < K(X; σ) ≤ 1 for all flags (X; σ) in the tangent bundle. Given a nonzero vector field V on a Finsler manifold (M, F ) with connection ∇ V (see Theorem 3.10), one can consider the curvature tensor R V defined by RV (X, Y )Z = ∇VX ∇VY Z − ∇VY ∇VX Z − ∇V[X,Y ] Z. If the vector field V is geodesic, it follows from the definition of the Riemannian curvature that RV (Y ) = RV (Y, V )V = −∇VV ∇VY V − ∇V[Y,V ] V. As in the Riemannian case, the flag curvature geometrically controls the infinitesimal behavior of geodesics, as described by the Jacobi fields along a geodesic: Definition 5.3. On a Finsler manifold (M, F ) we call a vector field Y = Y (t) along a geodesic c : [0, 1] → M a Jacobi field if it satisfies the differential equation 0 ∇2 Y (t) + Rc (Y ) = 0. 2 dt
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
275
It follows from Proposition 5.1 that the Jacobi fields of an osculating Riemannian metric (U, gV ) along the geodesic coincide with the Jacobi fields of the Finsler metric along c. Therefore the following wellknown facts of Riemannian geometry (see [Klingenberg 1995, 1.12], for example) carry over to the Finsler case immediately: Lemma 5.4. Let (M, F ) be a Finsler manifold and c : [0, 1] → M a geodesic. (a) For any Y0 , Y1 ∈ Tc(0) M , there is a uniquely determined Jacobi field Y along c with initial conditions Y (0) = Y0 and (∇/dt)Y (0) = Y1 . (b) If , in addition, hY0 , c0 (0)ic0 (0) = 0 = hY1 , c0 (0)ic0 (0) , the Jacobi field Y thus defined satisfies Y (t), c0 (t) c0 (t) = 0 for all t ∈ [0, 1]. The Jacobi equation of Definition 5.3 is the linearization of the geodesic equation: Lemma 5.5. Let (M, F ) be a Finsler manifold . (a) The variation vector field V (t) = (∂/∂s)s=0 cs (t) of a geodesic variation cs : [0, 1] → M on M is a Jacobi field . (b) For every Jacobi field Y = Y (t) along the geodesic c : [0, 1] → M there is a geodesic variation cs : [0, 1] → M whose variation vector field coincides with Y . The proof from the Riemannian case carries over; see [Klingenberg 1995, 1.12.4]. Definition 5.6. Let (M, F ) be a Finsler manifold and X a unit tangent vector in T M . Let the geodesic parametrized by arc length with initial direction X be defined (at least) in the closed interval [0, a], and denote it by c : [0, a] → M , so that c0 (0) = X. Suppose that for some s ∈ (0, a) there is a nontrivial Jacobi field Y = Y (t) along c that vanishes for t = 0 and t = s. (Nontrivial means that (∇/dt)Y (0) 6= 0.) Then the point c(s) is called conjugate to p = c(0) along c. Moreover, Y can be chosen so that hY, c0 ic0 = 0, and the set of all Y satisfying all these conditions is a vector space whose dimension is called the multiplicity of the conjugate point c(s). We define conjX ∈ (0, ∞] as the smallest positive number r such that c(r) is conjugate to p along c. The point c(r) is called the first conjugate point to p along c. The conjugate locus is the set of all first conjugate points to p. The conjugate radius conjp of a point p ∈ M is the infimum of the set {conjX  X ∈ Tp M, F (X) = 1}. The conjugate locus of p consists of critical points of the exponential map (see Section 8). The function X ∈ Tp1 M 7→ conjX ∈ R + ∪ {∞} is continuous. We denote by conj := inf {conjp  p ∈ M } the conjugate radius of M . If M is compact this is a positive real number or ∞. Remark 5.7. In the case of constant flag curvature one can describe Jacobi fields explicitly. Let c : R → M be a geodesic parametrized by arc length on a Finsler manifold (M, F ) and assume that the flag curvature along c is a constant δ, meaning that K(c0 (t); V ) = δ for every t and every V ∈ Tc(t) M forming a flag with c0 (t). One can choose an orthonormal basis (e1 , e2 , . . . , en ) of the tangent
276
HANSBERT RADEMACHER
space Tc(0) M with respect to the metric h · , · ic0 with en = c0 (0). Using parallel transport defined by the covariant derivative ∇/dt along c we obtain a frame (e1 (t), e2 (t), . . . , en (t)) along c = c(t) orthonormal with respect to h · , · i c0 and Pn satisfying e1 (t) = c0 (t) for all t ∈ R. Then with Y (t) = i=2 yi (t)ei (t) the Jacobi equation for a Jacobi field orthogonal to c0 (with respect to h · , · ic0 ) decouples because of the identities n
X ∇2 Y (t) = yi00 (t)ei (t) 2 dt i=2 and 0
Rc (Y, c0 )c0 (t) =
n X
i,j=2
n n X X
0 yi Rc (ei ), ej c0 ej (t) = yi K(e1 ; ei )ei (t) = δ yi ei (t) i=2
i=2
into n−1 ordinary differential equations yi00 (t) + δyi (t) = 0,
i = 2, . . . , n.
The solutions y 00 + δy = 0, y(0) = 0, y 0 (0) = 1 are √ √ if δ > 0, 1/ δ sin( δ t) yδ (t) = t if δ = 0, √ √ 1/ −δ sinh( −δ t) if δ < 0.
Hence in the case of constant flag √ curvature K(c(t)) = δ along a geodesic c, we obtain for conjc0 (0) the value π/ δ if δ > 0, and ∞ if δ ≤ 0.
6. Second Variation of the Energy Functional The first variational formula shows that geodesics can be seen as critical points of the energy functional. Therefore it is natural to study the secondorder behavior of the energy functional at a geodesic. This leads to the second variational formula. For a piecewise smooth curve c : [0, 1] → M denote by W c the set of piecewise smooth vector fields along c. The index form of a geodesic c : [0, 1] → M is the symmetric bilinear form Ic : Wc × Wc → R defined by Z 1 D E 0 ∇ ∇ Ic (X, Y ) := X, Y (t) − hRc (X), c0 ic0 (t) dt. dt dt c0 0
Lemma 6.1 [Shen 2001a, § 10.2; Shen 2001b, § 8.5]. Let c s : [0, 1] → M , for s ∈ (−ε, ε), be a variation of the geodesic c = c0 with fixed end points cs (0) = c(0), cs (1) = c(1); or let cs : S 1 → M be a variation of the closed geodesic c = c0 by closed curves. Let the variation vector field be V (t) = (∂cs /∂s)s=0 cs (t). Then
d2 E(cs ) = Ic (V, V ). ds2 s=0
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
277
For a geodesic c : [0, 1] → M on a Finsler manifold (M, F ) with an osculating Riemannian metric g¯ = gV defined in a neighborhood of c([0, 1]), the index forms Ic , I c with respect to the Finsler metric and with respect to the osculating Riemannian metric coincide. We introduce some subspaces of the space Wc of vector fields along c defined at the beginning of this section. Wc0 denotes the subspace consisting of vector fields that vanish at the end points. If c is a closed geodesic, also known as a periodic geodesic (that is, c(0) = c(1) and c0 (0) = c0 (1)), we denote by Wc1 the subspace of Wc consisting of periodic vector fields (X(0) = X(1)). Note that a closed geodesic can be though of as having domain S 1 = [0, 1]/{0, 1}. Using Lemma 6.1 we obtain: Corollary 6.2. (a) For points p, q ∈ M let c : [0, 1] → M be a geodesic joining p = c(0) and q = c(1). The restriction of the index form of c to Wc0 is denoted by Ic0 . If cs : [0, 1] → M , s ∈ (−ε, ε), is a piecewise smooth variation of c with variation vector field Y ∈ Wc0 and with fixed end points p = cs (0), q = cs (1), we have d2 E(cs ) = Ic0 (Y, Y ). ds2 s=0 (b) Let c : S 1 → M be a closed geodesic and denote the restriction of the index form Ic to Wc1 by Ic1 . If cs : S 1 → M , s ∈ (−ε, ε) is a variation of c by closed curves with variation vector field Y , we have
d2 E(cs ) = Ic1 (Y, Y ). ds2 s=0
We now define important invariants of geodesics and closed geodesics. Definition 6.3. (a) The index ind c of a geodesic c : [0, 1] → M joining points p and q of a Finsler manifold is by definition the same as the index ind W c0 of the quadratic form Ic0 : Wc0 × Wc0 → R, that is, the maximal dimension of a subspace on which I c0 is negative definite. The nullity nul c is the maximal dimension of a subspace W 0 ⊂ Wc0 such that Ic0 (X, Y ) = 0 for all X ∈ W 0 and Y ∈ Wc0 . (b) The Λindex indΛ c of a closed geodesic c : S 1 → M on a Finsler manifold is the maximal dimension of a subspace on which the index form Ic1 : Wc1 × Wc1 → R is negative definite. The Λnullity nul c is the maximal dimension of a subspace W 00 ⊂ Wc1 such that Ic1 (X, Y ) = 0 for all X ∈ W 00 and Y ∈ Wc1 . As in the Riemannian case, one can show that these numbers are finite: Lemma 6.4. The index ind c and the nullity nul c of a geodesic c are finite. So are the Λindex indΛ c and the Λnullity nul c of a closed geodesic c.
278
HANSBERT RADEMACHER
Proof. Let c : [0, 1] → M be a geodesic. Choose a partition 0 = t 0 < t1 < · · · < tk = 1 of the unit interval such that there is no pair of conjugate points in [ti , ti+1 ]. Define the following subspaces of the vector space W c0 of piecewise smooth vector fields along c vanishing at the endpoints: J := {X ∈ Wc0  X[ti , ti+1 ] is a Jacobi field for i = 0, . . . , k−1}, H := {X ∈ Wc0  X(ti ) = 0 for i = 1, . . . , k−1}, Wc⊥ = {X ∈ Wc0  hX, c0 ic0 = 0 and X[ti , ti+1 ] is smooth for i = 0, . . . , k−1}. Then, for X, Y ∈ Wc⊥ , Z 1 D E
0 ∇ ∇ Ic (X, Y ) = X, Y − Rc (X, c0 )c0 , Y c0 dt dt dt c0 0 Z 1D 2 E ∇ c0 0 0 X + R (X, c )c , Y dt =− dt2 c0 0 +
k−1 XD
∇ ∇ X(t− X(t+ i )− i ), Y (ti ) dt dt
i=1
E
.
(6–1)
c0
It follows that J and H ∩ Wc⊥ are orthogonal with respect to Ic , that is, Ic (X, Y ) = 0 for all X ∈ J and Y ∈ H ∩ Wc⊥ . Therefore we can conclude that Wc0 is the direct orthogonal sum J ⊕ (H ∩ Wc⊥ ) with the following argument: For every Xi ∈ Tc(ti ) M and Xi+1 ∈ Tc(ti+1 ) M there is a unique Jacobi field Y along c[ti , ti+1 ] with Y (ti ) = Xi and Y (ti+1 ) = Xi+1 , since c[ti , ti+1 ] is by assumption free of pairs of conjugate points. On the other hand, the index form Ic0 is positive definite on H since there is no conjugate point c(t ∗ ), t∗ ∈ (ti , ti+1 ), to the point c(ti ) along c[ti , t∗ ]. This shows that the indices and nullities match: ind c = ind Ic0 = ind(Ic0 J),
nul c = nul Ic0 = nul(Ic0 J).
(6–2)
Since J is finitedimensional these invariants are finite. An analogous proof shows that the Λindex and Λnullity of a closed geodesic are also finite. ˜ We call a geodesic c nondegenerate if nul c = 0. This implies that the point q = c(1) is not conjugate to p = c(0) along c. We call a closed geodesic c : S 1 → M nondegenerate if nulΛ c = 1. Since in this case Ic1 (c0 , c0 ) = 0, the nullity is at least 1. (That’s why some other authors define the nullity of a closed geodesic as nulΛ c−1.) A Finsler metric all of whose closed geodesics are nondegenerate is called bumpy.
7. Results from Topology Using the energy functional on the space Ωpq M of curves on a Finsler manifold M joining two fixed points p, q ∈ M , we obtain a CWdecomposition of the space Ωpq M . The Morse indices of the geodesics in Ωpq are related to the dimensions
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
279
of the cells of the CWdecomposition. In this chapter we review results from the topology of CWcomplexes and the relation between the topology of Ω pq M and that of M . As general references we cite [Milnor 1969; Bredon 1995; Spanier 1966]. Definition 7.1. Let A, X be topological spaces with A ⊂ X. We say that X is obtained from A by adjoining kcells ekj , j ∈ Jk , if the following conditions hold: (a) For every j ∈ Jk the set ekj is a closed subset of X. (b) Let e˙ kj := ekj ∩ A. Then for all i, j ∈ Jk with i 6= j the subsets eki − e˙ ki and ekj − e˙ kj are disjoint. S (c) The topology of X = A ∪ j∈Jk ekj is the weak topology with respect to the component subsets A and ekj , for j ∈ Jk . (This means that U ⊂ X is open if and only if U ∩ A is open in A and U ∩ ekj is open in ekj for each j.) (d) For every j ∈ Jk there is a continuous map φj : (D k , S k−1 ) → (ekj , e˙ kj ) with φj (Dk ) = ekj such that the restriction φj : Dk − S k−1 → ekj − e˙ kj is a homeomorphism, and such that a subset U ⊂ ekj is open if and only if k U ∩ e˙ kj ⊂ e˙ kj and φ−1 j (U ) ⊂ D are open subsets. For k = 0 the space X is the disjoint union of the space A and a discrete space. Definition 7.2. A (relative) CWcomplex (X, A) consists of a topological space X, a closed subspace A ⊂ X, and a sequence of closed subspaces (X, A) k ⊂ X, S k ≥ 0 (where (X, A)k is called the kskeleton) with X = A ∪ k≥0 (X, A)k such that the following conditions hold: (a) (X, A)0 is obtained form A by adjoining 0cells and for every k ≥ 1 the space (X, A)k is obtained from (X, A)k−1 by adjoining kcells. S (b) The topology of X is the weak topology of the union A ∪ k≥0 (X, A)k .
Hence it is possible to build up a CWcomplex recursively. Start with the topological space (X, A)−1 := A, and recursively assume that (X, A)k−1 has been defined. Given continuous maps φ˜j : S k−1 → (X, A)k−1 , j ∈ Jk (called attaching maps), we form the subset (X, A)k as follows: Let e˙ kj = φj (S k−1 ) and ekj := D k ∪φj e˙ kj , meaning that ekj is the quotient space of the disjoint union of Dk and e˙ kj with the equivalence relation identifying each x ∈ S k−1 with φj (x) ∈ e˙ kj . The sets D k \ S k−1 and ekj \ e˙ kj are homeomorphic and can be identified. Next define the characteristic map φj : (D k , S k−1 ) → (ekj , e˙ kj ) by φj (x) = x, for x ∈ D k − S k−1 , and φj (y) = φ˜j (y), for y ∈ S k−1 . Then set S (X, A)k = (X, A)k−1 ∪ j∈Jk ekj . A subset U ⊂ (X, A)k is open if and only if the intersection U ∩ (X, A)k−1 is open in (X, A)k−1 and for all j ∈ Jk the preimages k k k φ−1 j (U ∩ ej ) are open subsets of D . The subsets ej , j ∈ Jk , are called the closed kcells of the CWcomplex. Because of possible identifications on the boundary,
280
HANSBERT RADEMACHER
these cells are in general not homeomorphic to D k . The subsets ekj − e˙ kj are called open cells; they are homeomorphic to D k − S k−1 , but in general these sets are not open subsets of the kskeleton (X, A)k . A subcomplex of a CWcomplex is the union of closed cells of the CWcomplex with the same attaching resp. characteristic maps which is itself a CWcomplex. For example, the kskeleton (X, A)k is a subcomplex. For A = ? we simply write X k = (X, ?)k . Remark 7.3. The advantage of using CWcomplexes is that one generally needs fewer cells to write a space as a CWcomplexes than to triangulate it. A simple but important example: for n ≥ 1 the ndimensional sphere S n = {x ∈ R n+1  kxk2 = 1} has the structure of a CWcomplex with one 0cell e 0 and one ncell; the attaching map is the constant map φ : S n−1 → e0 . CWcomplexes are very useful if one considers homotopy properties of topological spaces. For example, Morse theory shows that any manifold is homotopy equivalent to a CWcomplexes; see the proof of Corollary 7.7. Two continuous maps f, g : X → Y are called homotopic if there is a continuous map F : X × [0, 1] → Y , (x, t) 7→ Ft (x) = F (x, t), with F0 = f and F1 = g. There is also a relative version of this concept: Let (X, A) and (Y, B) be pairs of topological spaces, meaning that A ⊂ X and B ⊂ Y . A continuous map f : (X, A) → (Y, B) between these pairs is a continuous map f : X → Y with the property f (A) ⊂ B. Two continuous maps f, g : (X, A) → (Y, B) agreeing on A are homotopic relative to A, or homotopic rel A, if there is a continuous map F : (X, A) × [0, 1] 7→ (Y, B), (x, t) 7→ Ft (x) = F (x, t), with F0 = f , F1 = g and Ft A = f A = gB for all t ∈ [0, 1]. Two topological spaces X, Y are called homotopy equivalent if there are continuous maps f : X → Y and g : Y → X such that the compositions f ◦ g and g ◦ f are homotopic to the identity maps idY and idX , respectively. Then f is called a homotopy equivalence. We call a pathwise connected topological space X nconnected for some n ≥ 1 if for every j ∈ {1, . . . , n} every continuous map f : S j → X is homotopic to a constant map. (This implies that the homotopy groups πj (X, p) = 0 vanish for all j = 1, . . . , n, but we do not need this concept here.) A 1connected space is also called simply connected. A topological pair (X, A) is called nconnected for some n ≥ 1 if for every j ∈ {1, . . . , n} every continuous map f : S j → X is homotopic to a continuous map whose image lies in A. A continuous map f : X → Y between CWcomplexes is called cellular if it respects the CWstructure, that is, if the image f (X k ) of the kskeleton lies in the kskeleton of Y : f (X k ) ⊂ Y k . Proposition 7.4 (Cellular Approximation Theorem). Every continuous map f : X → Y between CWcomplexes is homotopic to a cellular map. (See [Spanier 1966, Theorem 7.6.17] for a proof.)
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
281
Thus every continuous map f : S j → S n between spheres is homotopic to a cellular map F : S j → S n , where the spheres have the CWstructure described in Remark 7.3. Now suppose that j < n; since the jskeleton of S n equals the 0skeleton (a single point), S n is (n−1)connected. The same argument shows: Proposition 7.5. If a pathwise connected CWcomplex X has no jdimensional cells for any j ∈ {1, 2, . . . , k}, the space X is kconnected . We note the following consequences of Whitehead’s and Hurewicz’s theorems [Bredon 1995, Theorems VII11.2 and VII10.7]: Proposition 7.6 [Bredon 1995, Corollary VII11.14]. (a) Let f : X → Y be a continuous map between two simply connected CWcomplexes, such that the induced homomorphism f∗ : Hi (X; Z) → Hi (Y ; Z) of the singular homology groups with integer coefficients is an isomorphism for all i. Then f is a homotopy equivalence. (b) If a topological space X is (n−1)connected for some n ≥ 2, then for every homology class h ∈ Hn (X; Z) there is a continuous map fh : S n → X and a generator g ∈ Hn (S n ; Z) such that (fh )∗ (g) = h. (Hence every ndimensional homology class can be represented by a spherical cycle.) Corollary 7.7. A compact and (n−1)connected differentiable manifold M is homotopy equivalent to the nsphere. Proof. We use the fact that an ndimensional manifold has the homotopy type of a finite CWcomplex X (meaning there are only finitely many cells) of dimension n (meaning the maximal cell dimension is n). This can be proved using a Morse function on M ; one obtains a CWstructure where every critical point of index k corresponds to a kdimensional cell. Since the critical points are nondegenerate and the manifold is compact, there are only finitely many of them. The index of a critical point on a differentiable manifold is bounded above by the dimension of the manifold. The manifold is simply connected, therefore orientable. Hence H n (M ; Z) ∼ =Z [Bredon 1995, Corollary VI7.2]. It follows from Proposition 7.6(b) that there is a continuous map f : S n → M inducing an isomorphism f∗ : Hn (S n ; Z) → Hn (M ; Z). Since S n and M n are (up to homotopy equivalence) CWcomplexes, we conclude from Proposition 7.6(a) that f is a homotopy equivalence. ˜ Remark 7.8. The Poincar´e conjecture states that for n ≥ 3, an ndimensional simply connected and compact manifold homotopy equivalent to the nsphere is homeomorphic to the nsphere. For n ≥ 5 the conjecture was proven by S. Smale, for n = 4 by M. Freedman. There is an announcement of a proof for n = 3, by G. Perelman [Milnor 2003]. Given a topological space X and a point p ∈ X, we define the loop space Ωp (X) = {γ : [0, 1] → X  γ continuous, γ(0) = γ(1) = p}.
282
HANSBERT RADEMACHER
Proposition 7.9. If X is a pathwise connected topological space and Ωp X is the loop space of X with p ∈ X, the homotopy groups satisfy πk−1 (Ωp X) ∼ = πk X
for all k ≥ 1.
Proof. Consider the cube I k = [0, 1]k , with boundary I˙k = {(x1 , . . . , xk ) ∈ I k  there is j = 1, . . . , k with xj ∈ {0, 1}} Given a continuous map f : (I k+1 , I˙k+1 ) → (X, p) for some point p ∈ X, we define F (f ) : (I k , I˙k ) → (Ωp X, p) by F (f )(x1 , x2 , . . . , xk )(t) = f (x1 , x2 , . . . , xk+1 ). Also, for a continuous map g : (I k , I˙k ) → (ΩX, p) we define a continuous map G(g) : (I k+1 , I˙k+1 ) → (X, p) by G(g)(x1 , . . . , xk+1 ) = g(x1 , . . . , xk )(xk+1 ). This lets us define the isomorphism between the homotopy groups π k (ΩX, p) and πk+1 (X, p). ˜ Corollary 7.10. Let M be a simply connected , compact, ndimensional manifold and p, q ∈ M arbitrary points. Consider the space Ωpq M := {γ : [0, 1] → M  γ piecewise smooth, γ(0) = p, γ(1) = q}. If this space is homotopy equivalent to a CWcomplex with no cells of dimension j ∈ {1, . . . , k}, then M is kconnected . In particular , if k = n − 1 the manifold is homotopy equivalent to the nsphere. Proof. The space Ωpq M of piecewise continuous curves joining p and q is homotopy equivalent to the space Ω∗pq M of continuous curves joining p and q with the compactopen topology. This can be shown by using the finitedimensional approximation Ω(k, a) for Ωa M = {γ ∈ Ωpq M  E(γ) ≤ a} with a sufficiently large k [Milnor 1969, Theorem 17.1]. The next step is that the homotopy type of the spaces Ωpq M does not depend on the chosen points. For two curves γ1 , γ2 : [0, 1] → M with γ2 (1) = γ1 (0), denote by γ1 ∗ γ2 : [0, 1] → M their composition, defined by ( γ2 (2t) for 0 ≤ t ≤ 21 , γ1 ∗ γ2 (t) = γ1 (2t − 1) for 21 ≤ t ≤ 1. Now fix a curve γ1 ∈ Ωqr M ; the map γ ∈ Ωpq M 7→ γ1 ∗ γ ∈ Ωr M defines a homotopy equivalence. Hence we can conclude from Proposition 7.5 that the loop space is kconnected. This finally implies by Proposition 7.9 that the manifold is (k−1)connected. If k = n − 1 we conclude from Corollary 7.7 that the manifold is homotopy equivalent to the nsphere. ˜
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
283
8. Morse Theory of the Energy Functional For a Finsler manifold (M, F ) and two points p, q ∈ M we consider the energy functional on the space Ωpq of curves joining the points p and q. A critical value κ equals the energy E(c) of a geodesic c joining these points. Morse theory provides a connection between invariants of critical points of a function on a manifold and global topological invariants, in our case a connection between homology or homotopy invariants of the loop space and indices of geodesics. We introduce the space Ωpq M of absolutely continuous curves γ : [0, 1] → M satisfying γ(0) = p, γ(1) = q, and Z 1 F 2 (γ 0 (t))dt < ∞. 0
The energy functional E : Ωpq M −→ R, given by Z 1 1 2 0 E(γ) = F (γ (t)) dt, 2 0 is C 1,1 differentiable, that is, it is C 1 differentiable and its derivative is locally Lipschitz continuous [Mercuri 1971]. If cs : [0, 1] → M , s ∈ (−ε, ε), is a variation of c = c0 with fixed end points p = c(0), q = c(1), we conclude from the first variation formula (Lemma 4.1) that the variation vector field Y (t) = (d/ds)s=0 cs (t) satisfies Z 1D E ∇ 0 ∂ (t) dt. E(c ) = c , Y dE(c)Y := s ∂s s=0 dt c0 0 The curve c is a critical point of the energy functional if dE(c)Y = 0 for every vector field Y ∈ Wc0 . Then it follows from Corollary 4.2 that the critical points are the geodesics c : [0, 1] → M , starting at p = c(0) and ending at q = c(1). For nonnegative κ ≥ 0 we define the sublevel sets Ωκpq = Ωκ := {σ ∈ Ωpq M  E(σ) ≤ κ}; then Ωlpq M with 2l = θ(p, q)2 contains the minimal geodesics joining p and q. For l small enough there is a unique minimal geodesic, that is, Ω lpq M contains exactly one element. If the case p = q the subset Ω 0pp consists of the point curve p. We choose an arbitrary Riemannian metric g on the manifold, which induces a Hilbert space structure on Ωpq M . If c : [0, 1] → M is a smooth curve with c(0) = c(1) and X, Y are smooth vector fields along c, a Riemannian metric on Ωpq M is defined by Z 1 Z 1 ∇ ∇ X(t), Y (t) dt, g1 (X, Y ) = g X(t), Y (t) dt + g dt dt 0 0 where ∇/dt is the covariant derivative along c induced by the LeviCivita connection of the Riemannian manifold. The energy functional induces a Lipschitz
284
HANSBERT RADEMACHER
continuous gradient vector field grad E through the equation g1 grad E(c), X = dE(c)X
for all X. The energy functional satisfies the Palais–Smale condition and the negative gradient flow, that is, the flow of the vector field −grad E on λ, is defined for every t ≥ 0 [Mercuri 1971]. For the Morse theory of the energy functional we have to consider also the second derivatives of the energy functional at the critical points. For a nonRiemannian Finsler metric the square F 2 of the Finsler metric F is not C 2 differentiable at the zero section. Hence E is C 2 differentiable only at the regular curves, those curves c with c0 (t) 6= 0 for all t. Geodesics of positive length are regular, so we can use the statement in Corollary 6.2. As remarked just before that corollary, the index form Ic0 of the geodesic defined with the Finsler metric coincides with the index form I c of an osculating Riemannian metric. We will not go into details in this construction since instead of the space Ω pq one can use a finitedimensional approximation. This allows us to use Morse theory for a finitedimensional compact manifold instead of the infinitedimensional Hilbert manifold Ωpq M . The finitedimensional approximation was introduced by M. Morse and is explained in [Milnor 1969, Chapter 16]. We start with a compact Finsler manifold (M, F ) with injectivity radius inj > 0 (see Definition 9.1 below). For every pair of points p, q ∈ M with distance θ(p, q) < inj, there is a unique minimal geodesic from p to q. Choose a > 0 and k ∈ N such that 1/k < (inj)2 /(2a), and set ti := i/k for i ∈ N. Define Ωpq (k, a) := c ∈ Ωapq M  c[ti , ti+1 ] is a geodesic . Since
2a 2 < inj2 , θ2 c(ti ), c(ti+1 ) ≤ L2 c[ti , ti+1 ] ≤ E c[ti , ti+1 ] ≤ k k a curve c ∈ Ωpq (k, a) is uniquely determined by the points c(t1 ), . . . , c(tk−1 ) ∈ M × · · · × M . Therefore we can identify Ωpq (k, a) with the submanifold with boundary (x1 , . . . , xk−1 ) ∈ M × · · · × M  θ(xi , xi+1 ) ≤ 21 inj for i = 0, . . . , k − 1 of the product manifold M × · · · × M , where we set p = x0 , q = xk . We conclude that Ωpq (k, a) has the structure of a compact manifold with boundary of dimension (k−1) dim M . Then there is a strong deformation retraction ru : Ωapq → Ωapq ,
u ∈ [0, 1],
with r0 (c) = c for all c ∈ Ωapq , ru (c) = c for all c ∈ Ωpq (k, a), u ∈ [0, 1], and r1 (c) ∈ Ωpq (k, a) for all Ωapq . It is defined for u ∈ [ti , ti+1 ], as follows: For t ≤ ti , ru (c)(t) is the broken geodesic with corners c(0), c(t 1 ), c(t2 ), . . . , c(ti ); ru (t) for
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
285
t ∈ [ti , u] is the minimal geodesic between c(ti ) and c(u) and ru (c)(t) = c(t) for t ≥ ti+1 . Then the restriction E 0 : Ωpq (k, a) → R of E given by Pk E 0 (x1 , . . . , xk ) = 12 i=1 θ2 (xi , xi+1 )
is a C 1 smooth function. The first variational formula implies that the critical points are the geodesics from p to q with E(c) ≤ a, and the function is C ∞ differentiable in the neighborhood of critical points of positive length. For a broken geodesic c = (x1 , . . . , xk ) (smooth except at t = t1 , . . . , tk−1 , where ti no longer bears the meaning i/k), a tangent vector y(t)=(∂/∂s) s=0 cs (t) is given by a variation cs = (xs1 , . . . , xsk ) ∈ Ωpq (k, a), s ∈ (−ε, ε), that is, a curve in Ωpq (k, a). Since the variation vector field of a geodesic variation is a Jacobi field (Lemma 5.5), the tangent space Tx Ωpq consists of broken Jacobi fields, Tx Ωpq (k, a) = X ∈ Wc  X[ti , ti+1 ] is a Jacobi field .
In particular, if c is a geodesic (smooth throughout), the tangent space T c Ωpq coincides with the space J introduced in the Proof of Lemma 6.4. Therefore the Proof of Lemma 6.4 also shows that restricting the index form to the tangent space Tc Ωpq (k, a) changes neither the index nor the nullity. The energy functional E 0 : Ωpq M (k, a) → R is a differentiable function on the compact manifold Ωpq M (k, a). It is a Morse function if all critical points are nondegenerate, that is, if all geodesics c joining the points p and q with energy ≤ a are nondegenerate — in symbols, nul c = 0. Assume c ∈ Ω pq (k, a) is degenerate, so there is X ∈ Wc0 with X 6= 0 and Ic0 (X, Y ) = 0 for all Y ∈ Wc0 . Since X is a piecewise smooth vector field, choose 0 = t 0 < t1 < · · · < tk = 1 such that X is smooth when restricted to each subinterval [t i , ti+1 ], and also such that no subinterval contains a pair of conjugate points. Then we obtain from Equation 6–1: Z 1D 2 E ∇ c0 0 0 dt X + R (X, c )c , Y 0 = Ic (X, Y ) = − dt2 c0 0 +
k−1 XD
∇ ∇ X(t− X(t+ i )− i ), Y (ti ) dt dt
i=1
E
.
(8–1)
c0
+ Let Y ∈ Wc0 be a broken Jacobi field with Y (ti ) := (∇/dt)X(t− i ) − (∇/dt)X(ti ), so the restrictions Y [ti , ti+1 ] are Jacobi fields along c[ti , ti+1 ]. Then Equation 8–1 implies
Ic (X, Y ) =
2 ∇
∇ + ) − X(t )
X(t− i i 0 = 0,
k−1 X i=1
dt
dt
c
hence X is a smooth vector field. We have shown:
Lemma 8.1. The energy functional E 0 : Ωpq (k, a) → R is a Morse function if and only if the point q is not conjugate to p along any geodesic c with E(c) ≤ a joining p and q.
286
HANSBERT RADEMACHER
The next proposition relates the conjugate points of p ∈ M to the critical points of the exponential map at p. (Recall that the exponential map exp p : Tp M → M is defined by expp (X) = cX (1), where for X ∈ Tp M we denote by cX : R → M the geodesic with c0X (0) = X. This assumes that the metric is complete — more precisely, forward geodesically complete — an assumption that is satisfied in particular if the manifold is compact, thanks to the Finsler version of the Hopf–Rinow Theorem [Bao et al. 2000, § 6.6].) Proposition 8.2. Let (M, F ) be a complete Finsler manifold and let p ∈ M . A point q = expp (X) is a critical point of the exponential map expp : Tp M → M if and only if q is a conjugate point of p along the geodesic t ∈ [0, 1] 7→ expp (tX) ∈ M from p to q. The proof of the Riemannian case [Milnor 1969, Theorem 18.1] carries over. As an application of Sard’s Theorem one obtains: Corollary 8.3 [Milnor 1969, Corollary 18.2]. Let (M, F ) be a compact Finsler manifold and let p ∈ M . For almost all points q ∈ M (that is, up to a set of measure zero) the point q is not a conjugate point to p along any geodesic from p to q. For almost all q ∈ M the energy functional E : Ωpq M → R is a Morse function. It is the chief observation of Morse theory that the topology of the sublevel sets Ωκpq M := {σ ∈ Ωpq  E(σ) ≤ κ} and Ωκpq (k, b) can only change if κ is a critical value. The change in topology can be described by the indices of the corresponding critical points. Applied to the energy functional, this line of argumentation yields (compare [Milnor 1969, Theorem 17.3]): Theorem 8.4 (Fundamental Theorem of Morse Theory). Let (M, F ) be a compact Finsler manifold and p ∈ M an arbitrary point. For almost all q ∈ M and for all a > 0 the function E 0 : Ωpq (k, a) → R is a Morse function and there are only finitely many geodesics c joining p and q with E(c) ≤ a. The spaces Ωκpq M and Ωκpq (k, a) have the homotopy type of a CWcomplex having as many mcells as there are geodesics c joining p and q with E(c) ≤ a and ind c = m. Sketch of proof. As remarked in Proposition 8.2, for almost all q ∈ M and all a > 0 the energy functional E 0 : Ωpq (k, a) → R is a Morse function. If there is no critical value in [α, β], one can use the flow of the negative gradient field −grad E 0 on Ωpq (k, a) and retract Ωβpq (k, a) onto Ωα pq (k, a). The behavior of a Morse function near a critical point is described by the Morse Lemma [Milnor 1969, Lemma 2.2]. Applied to E 0 it states that near a geodesic c one can introduce local coordinates y = (y 1 , . . . , yr ), with c corresponding to 0 = (0, . . . , 0), such that E 0 (y1 , . . . , yr ) = E(c) −
ind Xc j=1
yj2 +
r X
j=ind c+1
yj2 .
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
287
Here r = (k−1) dim M = dim Ωpq (k, a). Hence the index describes the dimension of a subspace on which the energy of nearby curves decreases quadratically, whereas on a complementary subspace the energy grows quadratically. This implies that the geodesics are isolated. Since Ω pq (k, a) is compact, there are only finitely many geodesics joining p and q with energy ≤ a. Assume for simplicity that there is only one geodesic of energy a joining p and q. Let ind c = m. Then one can show that for sufficiently small ε > 0 a−ε the set Ωa+ε pq (k, a) has the homotopy type of Ωpq (k, a) with an mdimensional cell attached [Milnor 1969, Theorem 3.2]. This mcell corresponds to the set 2 (y1 , . . . , yk , 0, . . . , 0)  y12 + . . . + ym < ε in the coordinates used in the Morse Lemma. ˜ Remark 8.5. In Remark 5.7 we discussed Jacobi fields along a geodesic where the flag curvature is constant and positive. Now we consider the index form Ic (δ, l) of a geodesic c = cl (δ) : [0, 1] → M of length l with constant flag curvature K(c0 (t); σ). We can use bounds for the flag curvature to estimate the index and the conjugate radius, as in the Riemannian case. We choose e1 , e2 , . . . , en in Tc(0) M , orthonormal with respect to h · , · ic0 and such that c0 (0) = F (c0 )e1 . We extend this frame by parallel transport with respect to (∇/dt) along c. We can write vector fields X = X(t) along c as Pn X(t) = i=1 xi (t)ei (t), for smooth functions xi : [0, 1] → R. Then Ic (δ, l)(X, X) =
Z
1
0
x0i (t)2 − δx2i (t) dt
and one shows that ind cl (δ) = ind Ic (δ, l) = k (n − 1) (8–2) √ for l ∈ kπ/ δ, (k+1)π/ δ . See [Klingenberg 1995, Example 2.5.7]. Now let γ : [0, 1] → M be a geodesic of a Finsler metric with a lower bound for the flag curvature: K ≥ δ. We again choose along γ an orthonormal frame (e1 , e2 , . . . , en )(t) parallel with respect to (∇/dt). We can estimate the indexes ind γ and ind Iγ by comparing them with the index ind cl (δ) of a geodesic cl (δ) of the same length on a space form with constant sectional curvature: Z 1 x0i (t)2 − K(e1 (t); e2 (t))x2i (t) dt Ic (X, X) = √
0
≤
Z
1
0
x0i (t)2 − δx2i (t) dt = Ic (δ, l)(X, X).
This computation and a similar one in the case of an upper bound for the flag curvature lead to the following estimates for the distance of conjugate points and indices of geodesics. Here we use the fact that the index form I γ of γ = γX is positive definite for L(γ) < conjX and degenerate for L(γ) = conjX .
288
HANSBERT RADEMACHER
Lemma 8.6. Let γ = γX : [0, a] → M be a geodesic parametrized by arc length 0 on a Finsler manifold (M, F ), with γX (0) = X. (a) If the flag√curvature K = K(γ 0 ;√σ) satisfies K ≤ ∆ (resp. K < ∆) then conjX ≥ π/ ∆ (resp. conjX > π/ ∆). (b) If the flag√curvature K = K(γ 0√ ; σ) satisfies δ ≤ K (resp. δ < K) then conjX ≤ π/ ∆ (resp. conjX < π/ ∆). 2 (c) If the Ricci curvature Ric √ = Ric(γ 0 )F 2 satisfies Ric √ ≥ δ(n−1)F (resp. Ric > δ(n − 1)) then conjX ≤ π/ δ (resp. conjX < π/ δ). Sketch of proof. The argument for cases (a) and (b) is given in Remark 8.5. The argument in case (c) is the same as in the Riemannian case. As in the preceding remark √ we choose an orthonormal frame (e 1 , . . . , en )(t) along γ with L(γ) = a = π/ δ, parallel with respect to ∇/dt, and such that γ 0 (0) = X = F (X)e1 (0). Then we define the vector fields √ Xi (t) = sin( δ t)ei (t), i = 2, . . . , n. We compute for the index form Iγ : n X i=2
Iγ (Xi , Xi ) =
n Z X i=2
a
0
= (n − 1)
√ √ cos2 ( δ t) − K(e1 (t); ei (t)) sin2 ( δ t) dt a X 0
≤ (n − 1)δ
Z
a 0
Pn
√ K(e1 ; ei ) 2 δ cos ( δ t ) − sin ( δ t ) dt n−1 √ √ cos2 ( δ t) − sin2 ( δ t) dt = 0.
√ We conclude that conjX ≤ π/ δ.
2
√
i=2
˜
The diameter of a complete Finsler manifold M is the maximal distance of two points. By the Hopf–Rinow theorem [Bao et al. 2000, § 6.6] there is a minimal geodesic between two points of maximal distance. Since a geodesic is not minimal after the first conjugate point, the diameter is at most max {conj X  X ∈ T 1 M }. Therefore we obtain as a consequence of Lemma 8.6 the following generalization of the Bonnet–Myers theorem of Riemannian geometry: Corollary 8.7 [Auslander 1955]. Let (M, F ) be a complete Finsler manifold of dimension n with Ricci curvature Ric(V ) ≥ δ (n−1)F 2 (V ) for some positive δ and for all √ nonzero tangent vectors V . Then M is compact and its diameter is at most π/ δ. Since this estimate also holds for the universal covering space, we conclude that the universal covering space is also compact, so the fundamental group of the manifold is finite. In the proof of the Sphere Theorem the following statement is of importance:
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
289
Proposition 8.8. Let (M, F ) be a compact and simply connected Finsler manifold of dimension n, and let p, q ∈ M be such that q is not conjugate to p along any geodesic joining p and q. Assume there is a number m ≥ 2 such that every nonminimal geodesic c from p to q has index at least m. Then: (a) The manifold is mconnected (see page 280 for definition). (b) If m = n − 1, the manifold is homotopy equivalent to the nsphere. Proof. It follows from the Fundamental Theorem of Morse Theory (Theorem 8.4) that the space Ωpq has the homotopy type of a CWcomplex with no cells of dimension k ∈ {1, 2, . . . , m − 1}. By Proposition 7.5 this implies that the space Ωpq is (m−1)connected; Proposition 7.9 then implies that M itself is mconnected. Part (b) follows from Corollary 7.7. ˜
9. Shortest Nonminimal Geodesics and the Sphere Theorem Now we come to the crucial geometric argument in the proof of the Sphere Theorem. We obtain a lower bound for the length of a nonminimal geodesic c joining two points p, q or a nonconstant geodesic loop. In contrast to a minimal geodesic, this geodesic will meet the cut locus, after which the geodesic is not minimal anymore. The exponential map expp : Tp M → M is C ∞ smooth on Tp M \ {0} and 1 C smooth on Tp M [Shen 2001a, § 11.1]. The differential at 0 ∈ Tp M is an isomorphism; hence there is an ε > 0 such that the restriction expp : Bε (Tp M ) = {X ∈ Tp M F (X) < ε} → M is a local diffeomorphism onto its image Bε (p) ⊂ M . If a piecewise smooth curve c : [0, a] → M is minimal, that is, L(c) = θ c(0), c(a) , it follows from Corollary 4.2 that c is a smooth geodesic. Definition 9.1. For a unit tangent vector X ∈ Tp M , set t(X) = sup s > 0  θ(expp (sX), p) = s .
Then q = expp (t(X)X) is called a cut point. The cut locus Cut(p) := expp (t(X)X)  F (X) = 1, t(X) < ∞
is the union of all cut points on geodesics starting from p. The injectivity radius at p is inj p := inf θ(p, q)  q ∈ Cut(p) . If the manifold is compact we define the injectivity radius of M as inj = inj(M ; F ) = inf inj p  p ∈ M . The symmetrized injectivity radius at p is d(p) := inf d(p, q)  q ∈ Cut(p) . If the manifold is compact, we define the symmetrized injectivity radius d = d(M ; F ) = inf d(p)  p ∈ M . Finally, given two points p, q we define ϑ(p, q) := inf θ(p, r) + θ(r, q)  r ∈ Cut p }.
290
HANSBERT RADEMACHER
Hence d(p) is the symmetrized distance between p and its cut locus, whereas inj p is the distance θ(p, Cut(p)) with respect to the distance function θ. For a reversible Finsler metric these functions coincide. In general we have the bounds 1 1 2 1 + 1/λ inj p ≤ d(p) ≤ 2 (1 + λ) inj p,
2 2λ d(p) ≤ inj p ≤ d(p), 1+λ 1+λ which imply the corresponding estimates for global injectivity radii inj and d in case of a compact manifold. Obviously ϑ(p, p) = 2d(p) and the triangle inequality for θ implies ϑ(p, q) + θ(q, p) ≥ 2d(p). (9–1) If the manifold is compact, the cut locus of a point is also compact, so the infima in the above definitions of the injectivity radius and the symmetrized injectivity radius are actually minima. Definition 9.2. A broken geodesic with one r corner joining p and q is a continuous curve c1 c : [0, b] → M such that p = c(0), q = c(b), and for some a ∈ (0, b) the restrictions c1 = c2 c[0, a] → M and c2 = c[a, b] → M are miniq mal geodesics. The point r = c(a) is the corner p 0 0 of c. We call c smooth at r if c1 (a) = c2 (a). The length of c is given by L(c) = L(c1 ) + L(c2 ) = θ(p, r) + θ(r, q). If p = q, we have a closed broken geodesic, and its length is twice the symmetrized distance between p and r: L(c) = 2d(p, r). Lemma 9.3. Let (M, F ) be a compact Finsler manifold with reversibility λ and flag curvature K ≤ 1, and let p ∈ M be a point on M . If there is a cut point r ∈ Cut p with θ(p, r) < π, there is a local hypersurface H ⊂ M with r ∈ H such that for every smooth curve τ : (−1, 1) → H with τ (0) = r there are two geodesic variations c1,s , c2,s : [0, θ(p, q)] → M with c1,s (θ(p, r)) = c2,s (θ(p, r)) = τ (s), L(c1,s ) = L(c2,s ) and such that c1 = c1,0 , c2 = c2,0 are two distinct minimal geodesics joining p and r. H c1
c1,s
τ (s)
r
c2,s
c2 p
Proof. We conclude from Lemma 8.6 that, since r is not conjugate to p along a minimal geodesic, there are distinct minimal geodesics c 1 , c2 : [0, θ(p, r)] → M
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
291
parametrized by arc length with c1 (0) = c2 (0) = p and c1 (θ(p, r)) = c2 (θ(p, r)) = r. Since r is not conjugate to p along c1 or c2 , we can choose an open neighborhood U ⊂ M of r and open disjoint neighborhoods Uj ⊂ Tp M of c0j (0), for j = 1, 2, such that the restrictions of the exponential map exp p : Tp M → M to U1 , U2 are diffeomorphisms; see Proposition 8.2. We define functions f 1 , f2 : U → R by setting fj (v) = F (expp Uj )−1 (v) , j = 1, 2. These functions are differentiable and of maximal rank, with f j (r) = θ(p, r),
grad fj (r) = c0j (θ(p, r)) and dfj (r)(X) = c0j (θ(p, r)), X c0 (θ(p,r)) j
for j = 1, 2 and for all X, as follows from the Gauss lemma (see [Bao et al. 2000, § 6.1] and [Shen 2001a, 11.2.1]). One can view f1 , f2 as distance functions on the Finsler manifolds (U10 , F ) and (U20 , F ), where Uj0 is a small open neigborhood of the image of the curve c [0, θ(p, r)] . Since grad f1 (r) = c01 (θ(p, r)) 6= c02 (θ(p, r)) = grad f2 (r),
the function f1 − f2 has maximal rank in an open neighborhood of r, which we again denote by U ; thus H = f −1 (0) = {x ∈ U  f1 (x) = f2 (x)} is a smooth hypersurface with r ∈ H. We finish by setting c 01,s (0) = f1−1 (τ (s)) and c02,s (0) = f2−1 (τ (s)). ˜ Lemma 9.4. Let (M, F ) be a compact Finsler manifold with reversibility λ and with flag curvature K ≤ 1. Let p, q ∈ M be two points with ϑ(p, q) + θ(q, p) < π(1 + λ−1 ). Then there is a cut point r ∈ Cut p and a geodesic c of length ϑ(p, q) parametrized by arc length from p to q going through r.
c3 H q
c3,s c1
τ (s)
r
Proof. Choose r ∈ Cut p such that ϑ(p, q) = c2,s θ(p, r) + θ(r, q). If θ(p, r) ≥ π, the definition c1,s of the reversibility (Lemma 3.9) implies that θ(r, p) ≥ π/λ, hence ϑ(p, q) + θ(q, p) = θ(p, r) + c2 θ(r, q) + θ(q, p) ≥ θ(p, r) + θ(r, p) = 2d(p, r) ≥ p π(1+λ−1 ), contradicting the assumption. Thus we have proved that θ(p, r) < π. On the other hand, if θ(r, q) ≥ π, we have θ(q, r) ≥ π/λ and ϑ(p, q) + θ(q, p) = θ(p, r) + θ(r, q) + θ(q, p) ≥ θ(r, q) + θ(q, r) = d(r, q) ≥ π(1 + λ −1 ). Therefore it follows also that θ(r, q) < π. Since θ(p, r) < π, the point r is not conjugate to p along a minimal geodesic joining p. Therefore Lemma 9.3 gives an open hyersurface H ⊂ M with r ∈ H such that for every smooth curve τ : (−1, 1) → H with τ (0) = r there are variations c1,s , c2,s : [0, θ(p, r)] → M such that L (˜ c1,s ) = L (˜ c1,s ), p = c1,s (0) =
292
HANSBERT RADEMACHER
c2,s (0), τ (s) = c1,s (1) = c2,s (1) and c1 = c1,0 , c2 = c2,0 : [0, θ(p, r)] → M are distinct minimal geodesics joining p and r. Now let c3 : [θ(p, r), ϑ(p, q)] → M be a minimal geodesic parametrized by arc length joining c3 (θ(p, r)) = r and c3 (2d(p, q)) = p. Since θ(r, q) < π, the point q is not conjugate to r along a minimal geodesic from r to q. Therefore we can choose a geodesic variation c3,s : [θ(p, r), ϑ(p, q)] → M with c3,0 = c3 , c3,s (θ(p, r)) = τ (s) and c3,s (ϑ(p, q)) = q for all s ∈ (−1, 1). Now we can combine the smooth geodesic variations c 1,s , c2,s and c3,s to obtain two piecewise smooth variations c˜1,s , c˜2,s : [0, ϑ(p, q)] → M with ( cj,s (t) if t ∈ 0, θ(p, r) , c˜j,s (t) = c3,s (t) if t ∈ θ(p, r), ϑ(p, q) .
These are variations of the broken geodesics c˜1 = (c1 , c3 ) and c˜2 = (c2 , c3 ) by broken geodesics with fixed end points p = c˜j (0) and q = c˜j (ϑ(p, q)) and with τ 0 (0) = (∂/∂s)s=0 c˜j,s (θ(p, r)). We assume that c01 (θ(p, r)) 6= c03 (θ(p, r)) and c02 (θ(p, r)) 6= c03 (θ(p, r)). Since c1 and c2 are distinct, c01 (θ(p, r)), c02 (θ(p, r)), c03 (θ(p, r)) are pairwise disjoint. Recall from Definition 3.3 the Legendre transformation L(X)(Y ) = hY, Xi X . Given three pairwise distinct nonzero vectors X 1 , X2 , X3 ∈ Tr M , we have dim Y ∈ Tr M  L(X1 )(Y ) = L(X2 )(Y ) = L(X3 )(Y ) ≤ n − 2. Applying this c01 (θ(p, r)), c02 (θ(p, r)), c03 (θ(p, r)), we see that there is a tangent vector V ∈ Tr H ⊂ Tr M such that L c3 (θ(p, r)) (V ) is not equal simultaneously to L c1 (θ(p, r)) (V ) and L c2 (θ(p, r)) (V ). We assume without loss of generality that L c1 (θ(p, r)) (V ) − L c3 (θ(p, r)) (V ) 6= 0. The first variational formula for the energy functional (Lemma 4.1), applied to the variation c˜1,s of the broken geodesic c˜1 , yields
d E(˜c1 ) = L c1 (θ(p, r)) − L c3 (θ(p, r)) 6= 0. ds s=0
By using s 7→ τ (−s) instead of s 7→ τ (s), if necessary, we can assume that
d E(˜c1,s ) < 0. ds s=0 It follows that θ p, c˜1,s (θ(p, r)) + θ c˜1,s (θ(p, r)), q < θ(p, r) + θ(r, q) for small s > 0. Since for sufficiently small s > 0 the geodesics c˜1,s , c˜2,s : [0, θ(p, r)] → M intersect at t = θ(p, r), the cut point c˜1,s (t1,s ) of c˜1,s occurs no later than θ(p, r), that is, t1,s ≤ θ(p, r). Since θ p, c˜1,s (t1,s ) + θ c˜1,s (t1,s ), q ≤ θ p, c˜1,s (θ(p, r)) + θ c˜1,s (θ(p, r)), c˜1,s (ϑ(p, q)) = L c˜1,s < ϑ(p, q),
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
293
we have found for sufficiently small s > 0 a cut point r1,s = c˜1,s (t1,s ) ∈ Cut p satisfying θ(p, r1,s ) + θ(r1,s , q) < θ(p, r) + θ(r, q) = ϑ(p, q), which contradicts the definition of ϑ(p, q). Hence c01 (θ(p, r)) = c03 (θ(p, r)), that is, the broken geodesic (c1 , c3 ) with break point r is actually smooth. ˜ Lemma 9.5. Let (M, F ) be a compact Finsler manifold with reversibility λ and flag curvature K ≤ 1. (a) Let p, q ∈ M with q 6∈ Cut p and assume that ϑ(p, q) + θ(q, p) < π(1 + λ −1 ). Then ϑ(p, q) is the length of the shortest nonminimal geodesic from p to q. (b) If the symmetrized injectivity radius d of M satisfies d < π(1 + λ−1 )/2, there is a shortest geodesic loop c with initial point p and a point q ∈ Cut(p) on this loop with L(c) = 2d = 2d(p, q). Proof. (a) The cut locus Cut p is a closed subset; hence there exists r ∈ Cut p with θ(p, r)+θ(r, q) = ϑ(p, q). It follows from Lemma 9.4 that there is a geodesic c from p to q through r with L(c) = ϑ(p, q). Since r is a cut point and r 6= q, this geodesic is not minimal. (b) Let q ∈ Cut(p) be a point with d = d(p, q). We know from Lemma 9.4 that there is a geodesic loop c with c(0) = p and L(c) = 2d. We only have to show that this curve is a shortest geodesic loop. If c1 is a shortest geodesic loop with c1 (0) = p and with cut point q = c1 (t0 ), the restriction c1 [0, t0 ] is minimal and L(c1 ) ≥ 2d(p, q). But we showed in Lemma 9.4 that there is a geodesic loop c ∈ Ωp with L(c) = 2d = 2d(p). This finally implies that L(c 1 ) = 2d. ˜ Remark 9.6. If the Finsler metric is reversible, the proof of Lemma 9.4 simplifies considerably. The argument for this case was introduced by Klingenberg [1995, 2.1.11] in the Riemannian setting. The minimal geodesic c 3 coincides with one of the minimal geodesics c1 , c2 (say c1 ) up to orientation, that is, θ(p, r) = d(p, r) and c3 (t) = c1 (2d(p, r) − t). By using the same argument exchanging the roles of p, r one can prove then that there is a closed geodesic c of length 2d. If c is parametrized by arc length, c(d) is the cut point, and there is no shorter geodesic loop. If we use the same argument as in the proof of Lemma 9.4, we obtain: Lemma 9.7. Let (M, F ) be a compact Finsler manifold with reversibility λ and flag curvature K ≤ 1. If the symmetrized injectivity radius satisfies d < π 1 + λ1 , there is a point p ∈ M and a cut point r ∈ Cut p such that either
(a) there is a closed geodesic c : [0, 2d] → M parametrized by arclength with L(c) = 2d and c(0) = p, c(θ(p, r)) = r, or (b) there are two distinct geodesic loops c1 , c2 : [0, 2d] → M parametrized by arc length (that is, both have the same length 2d) with c1 (0) = p = c2 (θ(r, p)) and c1 (θ(p, r)) = r = c2 (0).
294
HANSBERT RADEMACHER
r
r
c1 c1 p
c2
c2
c3 p
c3
Proof. If we use the statement of Lemma 9.4 and the argument in the proof of the same lemma, exchanging the roles of p and q, we reach the following statement: There are three minimal geodesics c 1 , c2 : [0, θ(p, q)] → M and c3 : [θ(q, p), 2d] → M with d = d(p, q), p = c1 (0) = c2 (0) = c3 (2d), and q = c1 (θ(p, q)) = c2 (θ(p, q)) = c3 (θ(p, q)). Without loss of generality we can assume that the broken geodesic (c1 , c3 ) with corner q is smooth at q: c01 (θ(p, q)) = c03 (θ(p, q)). In addition there are two cases: Either the broken geodesic formed by c3 , c1 is smooth at p, so that c03 (2d) = c01 (0), which means that the geodesics c1 , c3 form a closed geodesic, or the broken geodesic formed by c3 , c2 is smooth at p. ˜ In the reversible case we have case (a) of the lemma. It is not clear whether case (b) occurs in the nonreversible case. Lemma 9.8. Let (M, F ) be a compact Finsler manifold with reversibility λ and flag curvature K ≤ 1 and let p, q ∈ M be two points with distance θ(p, q) < inj p. Let c1 : [0, 1] → M be a nonminimal geodesic with c1 (0) = p and c1 (1) = q, and : [0, 1] → M , let c0 : [0, 1] → M be a curve such that the reversed curve c−1 0 c−1 (t) = c (1 − t) is a minimal geodesic with q = c (1), p = c (0) and length 0 0 0 0 L(c0 ) = θ(q, p). If cs : [0, 1] → M , s ∈ [0, 1], is a homotopy of piecewise smooth curves with fixed endpoints p = cs (0), q = cs (1) for all s ∈ [0, 1] between the curves c0 , c1 , then 1 θ(q, p) + max L(cs ) ≥ π 1 + . λ s∈[0,1] Proof. We are assuming that θ(q, p) + L(cs ) < π (1 + λ−1 ) for all s ∈ (0, 1]; hence there is a ρ > 0 such that θ(q, p)+L(cs ) ≤ (π −ρ)(1+λ−1 ) for all s ∈ [0, 1]. We show by contradiction that θ(p, cs (t)) ≤ π − ρ for all s, t ∈ [0, 1]. If there are s, t ∈ [0, 1] such that θ(p, cs (t)) > π − ρ, then θ(q, p) + L(cs ) ≥ θ(p, cs (t)) + θ(cs (t), q) + θ(q, p) ≥ θ(p, cs (t)) + θ(cs (t), p) ≥ θ(p, cs (t))(1 + λ−1 ) > (π − ρ)(1 + λ−1 ),
(9–2)
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
295
which contradicts our assumption. Define the closed ball B a (Tp M ) := {v ∈ Tp M  F(v) ≤ a} of radius a > 0 in the tangent space Tp M at p. The subset B a (p) := x ∈ M  θ(p, x) ≤ a ⊂ M equals the image expp (B a (Tp M )) of the exponential map expp : Tp M → M for an arbitrary a > 0. It follows from 9–2 that cs (t) ∈ B π−ρ (p) for all s, t ∈ [0, 1]. The restriction F := expp : B π−ρ (Tp M ) → B π−ρ (p) has everywhere maximal rank since the flag curvature satisfies K ≤ 1; thus it is a local diffeomorphism. The restriction expp : Binj p (Tp M ) → Binj p (p) is a diffeomorphism, since θ(q, p) < inj p for sufficiently small η > 0 we have cs ([0, 1]) ⊂ Binj p (p) for all s ∈ [0, η). Hence there is a uniquely defined lift c˜s : t ∈ [0, 1) 7→ c˜s (t) ∈ B inj p (Tp M ) for s ∈ [0, η) with cs (t) = expp (˜ cs (t)), 0p = c˜s (0), and X = c˜0 (1) = c˜s (1) for all s ∈ [0, η). Since the restriction F = expp B π−δ (Tp M ). is a local diffeomorphism, there is a uniquely determined extension c˜s : t ∈ [0, 1) 7→ c˜s (t) ∈ B π−ρ (Tp M ) with cs (t) = expp (˜ cs (t)) of the lift for all s ∈ [0, 1]. It remains to show that this lift is a homotopy with fixed cs (0) = 0p , end points. Define J0 := s ∈ [0, 1]˜ J1 := s ∈ [0, 1]˜ cs (1) = X ; these subsets contain the nonempty interval [0, η) and are closed in [0, 1]. Since the restriction F of the exponential map is a local diffeomorphism the subsets J0 , J1 ⊂ [0, 1] are also open, hence J0 = J1 = [0, 1]. By assumption c1 : [0, 1] → M is a geodesic from p to q; hence c˜1 (t) = tc01 (0) for all t ∈ [0, 1], contradicting c˜1 (1) = 0p . Therefore we arrive at a contradiction 1 starting from the assumption θ(q, p) + L(cs ) < π 1 + λ for all s ∈ [0, 1], which finally proves the claim. ˜ With the long homotopy lemma we are able to gain a lower bound for the length of nonminimal geodesics: Proposition 9.9. Let (M, F ) be a simply connected , compact Finsler manifold 2 1 < K ≤ 1. of dimension n ≥ 3, with reversibility λ and flag curvature 1 − 1+λ If p ∈ M , there exists for every ε > 0 a point q that is a regular point of expp and that satisfies θ(q, p) < ε and ϑ(p, q) + θ(q, p) ≥ π 1 + λ1 . (Recall that ϑ(p, q) is the length of a shortest nonminimal geodesic from p to q.) 2 1 Proof. Since M is compact we can choose δ > 1 − 1+λ such that the flag curvature K of satisfies δ < K ≤ 1. For a given ε > 0 we choose a regular point q of expp with θ(q, p) < ε and
θ(q, p) < π 1 +
1 1 −√ . λ δ
296
HANSBERT RADEMACHER
Now we assume that there is a shortest nonminimal geodesic c 1 : [0, 1] → M from p = c1 (0) to q = c1 (1) with length L(c1 ) and satisfying L(c1 ) + θ(q, p) < π 1 + λ1 . M is simply connected, so there is a path s ∈ [0, 1] 7→ c s ∈ Ωpq M going from c0 (the reverse of the minimal geodesic c−1 0 from q to p of length θ(q, p)) to the geodesic c1 . This path describes a map H : ([0, 1], {0, 1}) → Ωpq M, Ωκ− pq M ,
with H(s) = cs , Ωκ− pq M := {γ ∈ Ωpq M  E(γ) < κ}, and κ defined by √ −1 2 κ = π(1 + λ ) − θ(q, p). Let c∗ : [0, 1] → √ M be a geodesic from p to q with length L(c∗ ) ≥ π 1 + λ1 − θ(q, p) ≥ π/ δ. Lemma 8.6 gives the bound ind c ≥ n − 1, so ind c ≥ 2 in view of the assumption n ≥ 3. We can conclude from the Fundamental Theorem of Morse Theory (Theorem 8.4) that the pair (Ωpq M, Ωκ− pq M ) has the homotopy type of a CWcomplex with no 1dimensional cells; hence the relative homotopy group π1 Ωp M, Ωκ− p M = 0, is 1connected, as is the pair Ωpq M, Ωκ− pq M ; see Proposition 7.5. Therefore there is a map Φ : (u, s) ∈ [0, 1] × ([0, 1], {0, 1}) 7→ Φ u (s) ∈ (Ωpq M, Ωκ− pq M ) with Φu (0)(t) = c0 (t), Φu (1)(t) = c(t) for all t, u ∈ [0, 1] and Φ0 (s) = cs and c¯s = Φ1 (s) ∈ Ωκ− cs ) < π 1 + λ1 − θ(q, p) pq for all s ∈ [0, 1]. This implies that L(¯ for all s ∈ [0, 1], that and c¯0 = c0 is up to orientation the minimal geodesic, and that c¯1 = c is a shortest nonminimal geodesic joining p and q. But we ∗ conclude from the Long Homotopy Lemma 9.8 that there is a s ∈ (0, 1) with 1 ˜ L(¯ cs∗ ) ≥ 1 + λ − θ(q, p), which is a contradiction. The assumption that q is a regular value of the exponential map exp p ensures that the energy functional is a Morse function, so all geodesics joining p and q are nondegenerate. If one aims at estimating the length of geodesic loops or closed geodesics it won’t be the case in general that p itself is a regular value of the exponential map expp . For example, on the standard sphere every point p is conjugate to itself along a great circle; in particular every point p is a critical point of expp . But the statement of Proposition 9.9 is also correct if we remove this assumption. In that case either one has to use a version of Morse theory including degenerate critical points [Rademacher 2004, Theorem 3] or one can argue as follows:
Theorem 9.10. Let (M, F ) be a simply connected , compact Finsler manifold of 2 1 dimension n ≥ 3, with reversibility λ and flag curvature 1 − 1+λ < K ≤ 1. 1 Then every nonconstant geodesic loop c has length at least π 1 + λ and the injectivity radius satisfies inj ≥ π/λ. Proof. For a point p ∈ M the function (q, r) ∈ M × Cut p 7→ θ(p, r) + θ(r, q) ∈ R
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
297
is continuous, hence also the map q ∈ M 7→ ϑ(p, q) = inf θ(p, r) + θ(r, q)  r ∈ Cut p .
Choose a sequence (qi )i∈N of regular points of the exponential map expp with limi→∞ θ(qi , p) = 0. We conclude from Proposition 9.9 that
1 . i→∞ λ The estimate for the injectivity radius follows from Equation 9–1. 2d(p) = lim ϑ(p, qi ) ≥ π 1 +
˜
Now we can prove the Sphere Theorem: Theorem 9.11. A simply connected and compact Finsler manifold of dimension 2 1 n ≥ 3 with reversibility λ and with flag curvature K satisfying 1− 1+λ 0 is diffeomorphic to the 2sphere. Proof of the Sphere Theorem. Since M is compact we can choose δ > π 1 + λ−1 such that the flag curvature K satisfies δ < K ≤ 1. We choose ε > 0 with 1 1 ε 0, contradicting the assumption that c is a shortest closed curve in the given homotopy class. Hence π1 (M ) = 0. If M is nonorientable one passes to the orientable double cover, which by (a) is simply connected, so π1 (M ) = Z 2 . ˜ Theorem 10.3. Let (M, F ) be a simply connected compact Finsler manifold of even dimension n ≥ 2 with reversibility λ and with flag curvature 0< K ≤ 1. Then every nonconstant closed geodesic c has length L(c) ≥ π 1 + λ1 . Proof. Let c : S 1 → M be a shortest closed geodesic with 0 < L(c) < π 1 + λ1 . By Lemma 10.1, there exists a parallel unit vector field W along c; it follows that the index form Ic on Vc⊥ satisfies Ic (W, W ) < 0. Let cs , s ∈ (−ε, ε), be a variation of c = c0 with variation vector field W . It follows from the second variation formula that E(cs ) < E(c0 ) for all s ∈ (−ε, 0) ∪ (0, ε). Since there are no critical values of E in the interval 0, E(c0 ) , there is a map hs : S 1 → M , s ∈ [−1, 1], with c = h0 , L(h1 ) = L(h−1 ) = 0 and L(hs ) < L(c) = L(h0 ) for all nonzero s ∈ (−1, 1). One can generalize the Long Homotopy Lemma 9.8 to the case of homotopies cs : S 1 → M of freely homotopic closed curves. This generalization yields a contradiction. ˜
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
299
11. An Example Shen [2002] constructed Finsler metrics of constant flag curvature and Randers type. It turns out that in the Hamiltonian description these are the metrics introduced in [Katok 1973] and investigated in [Ziller 1982], as observed in [Rademacher 2004, Chapter 4]. These examples show that the estimates in Proposition 9.9 and Theorem 9.10, for the lengths of nonminimal geodesics between fixed points and of nonconstant geodesic loops, are sharp. One can describe a Finsler metric using the Legendre transformation with a Hamiltonian function [Ziller 1982, Chapter 1]. The Katok examples on S 2 = {(x, y, z) ∈ R 3  x2 + y 2 + z 2 = 1} can be introduced as follows. We start with the standard Riemannian metric g on the 2sphere S 2 , letting g ∗ be the dual metric on the cotangent bundle T ∗ M . In the Hamiltonian description the standard metric is determined by the quadratic Hamiltonian function y ∈ T ∗ S 2 7→ g ∗ (y, y) ∈ R, p or by the 1homogeneous Hamiltonian H0 : T ∗ S 2 → R, H0 (y) = g ∗ (y, y). Let ψ0t : T ∗ S 2 → T ∗ S 2 be the corresponding Hamiltonian flow. Then t ∈ R 7→ τ ∗ (ψ0t (y)) is a geodesic of the standard metric; here τ ∗ : T ∗ S 2 → S 2 is the projection of the cotangent bundle. Let V (x, y, z) = (y, −x, 0) be the Killing field belonging to the 1parameter subgroup φt : S 2 → S 2 generated by the rotations around the zaxis. A. Katok introduced the following perturbation of Randers type: p Hε : T ∗ S 2 → R; Hε (y) = g ∗ (y, y) + εy(V ). In [Bao et al. 2003] these perturbations are connected to Zermelo navigation. For ε ∈ [0, 1) this defines a quadratic Hamiltonian 12 Hε2 and using the Legendre transformation of this Hamiltonian we obtain a Finsler metric F ε . The description of the geodesics appears to be easier in the Hamiltonian picture: Since φt is a group of isometries leaving H0 invariant, the Hamiltonian flow ψεt of the quadratic Hamiltonian 12 Hε2 is generated by two commuting flows, ψεt = ψ0t ◦ (φεt )∗ . Here (φt )∗ : T ∗ S 2 → T ∗ S 2 is the flow on the cotangent bundle induced by differentiating φt . The projection of the Hamiltonian flow onto the 2sphere yields the geodesics of the Finsler metric. As described in [Ziller 1982, Chapter 1] one can visualize the geodesic flow of these Finsler metrics by identifying the cotangent bundle T ∗ S 2 with the tangent bundle T∗ S 2 via the standard Riemannian metric g. Then the geodesic flow can be seen as the geodesic flow of the standard metric observed from a coordinate system rotating around the zaxis with constant speed 2πε, as shown in the figure on the next page. For irrational ε the only closed geodesics are c± (t) = (cos 2πt, ± sin 2πt, 0), t ∈ [0, 1], i.e., the equator with both orientations. (We consider c + and c− geometrically distinct; for example their lengths L(c± ) = 2π/(1 ± ε) differ.)
300
HANSBERT RADEMACHER
β
Using the results [Hrimiuc and Shimada 1996, Theorem 5.8], [Shen 2001a, Example 3.1.1] one obtains in geodesic polar coordinates (r, φ) ∈ (0, π) × [0, 2π] of the standard metric the following formula for Fε : q (1 − ε2 sin2 r) dr 2 + sin2 (r) dφ2 − ε sin2 r dφ . (11–1) Fε = 1 − ε2 sin2 r It is shown in [Shen 2002, Remark 3.1] that this metric has constant flag curvature 1. The reversibility of the Finsler metrics F ε can be computed by max Hε (−y)  y ∈ T ∗ S 2 , Hε (y) = 1 . Collecting the results from [Ziller 1982] and [Rademacher 2004, Chapter 5] one obtains:
Theorem 11.1 [Rademacher 2004, Theorem 5]. There is a oneparameter family Fε , ε ∈ [0, 1), of Finsler metrics on the 2sphere S 2 of constant flag curvature 1. These Finsler metrics are nonreversible for ε ∈ (0, 1) and F0 is the standard metric. The reversibility is λ = (1 + ε)/(1 − ε). If ε is irrational there are exactly two geometrically distinct closed geodesics c± of length L(c± ) = 2π(1 ± ε)−1 . In particular the shortest closed geodesic c+ satisfies L(c+ ) = 2π(1 + λ−1 ) = π/(1 + ε). The injectivity radius and the diameter equal π.
References [Abresch and Meyer 1997] U. Abresch and W. Meyer, Injectivity radius estimates and sphere theorems, pp. 1–47 in: Comparison geometry, MSRI Publications 30 (1997). [Auslander 1955] L. Auslander, “On curvature in Finsler geometry”, Trans. Amer. Math. Soc. 79 (1955), 378–388. [Bao et al. 2000] D. Bao, S.S. Chern, and Z. Shen, An introduction to Riemann–Finsler geometry, Grad. Texts Math. 200, Springer, New York, 2000. [Bao et al. 2003] D. Bao, C. Robles, and Z. Shen, “Zermelo navigation on Riemannian manifolds”, preprint, 2003.
NONREVERSIBLE FINSLER METRICS OF POSITIVE FLAG CURVATURE
301
[Berger 1998] M. Berger, “Riemannian geometry during the second half of the twentieth century”, Jber. d. Dt. Math.Verein. 100 (1998), 45–208. [Bredon 1995] G. E. Bredon, Topology and geometry, Grad. Texts Math. 139, Springer, New York, corr. 2nd printing, 1995. [Bryant 2002] R. Bryant, “Some remarks on Finsler manifolds with constant flag curvature”, Houston J. Math. 28 (2002), 221–262. [Dazord 1968a] P. Dazord, “Vari´et´es finsl´eriennes de dimension paire δpinc´ees”, C. R. Acad. Sc. Paris 266 (1968), 496–498. [Dazord 1968b] P. Dazord, “Vari´et´es finsl´eriennes en forme des sph`eres”, C. R. Acad. Sc. Paris 267 (1968), 353–355. [Hrimiuc and Shimada 1996] D. Hrimiuc and H. Shimada, “On the Lduality between Lagrange and Hamilton manifolds”, Nonlin. World 3 (1996), 613–641. [Katok 1973] A. Katok, “Ergodic properties of degenerate integrable Hamiltonian systems” (Russian), Izs. Akad. Nauk SSSR Ser. Mat. 37 (1973), 539–576; translated in Math. USSR Izv. 7 (1973), 535–572. ¨ [Klingenberg 1961] W. Klingenberg, “Uber Riemannsche Mannigfaltigkeiten mit positiver Kr¨ ummung”, Comment. Math. Helv. 35 (1961), 47–54. [Klingenberg 1963] W. Klingenberg, “Manifolds with restricted conjugate locus”, Ann. Math. 78 (1963), 527–547. [Klingenberg 1995] W. Klingenberg, Riemannian geometry, de Gruyter Studies Math. 1, 2nd rev. ed., de Gruyter, Berlin, New York, 1995. [Matthias 1980] H. H. Matthias, Zwei Verallgemeinerungen eines Satzes von Gromoll und Meyer, Bonner Math. Schr. 126 (1980). [Mercuri 1971] F. Mercuri, “The critical point theory for the closed geodesic problem”, Math. Zeitschr. 156 (1971), 231–245. [Milnor 1969] J. Milnor, Morse theory, Annals of Math. Studies 51 Princeton Univ. Press, Princeton 1969. [Milnor 2003] J. Milnor, “Towards the Poincar´e conjecture and the classification of 3manifolds”, Notices Amer. Math. Soc. 50:10 (2003), 1226–1233. [Rademacher 1992] H.B. Rademacher, MorseTheorie und Geschlossene Geod¨ atische, Bonner Math. Schr. 229 (1992). [Rademacher 2004] H.B. Rademacher, “A sphere theorem for nonreversible Finsler metrics”, Math. Ann. 328 (2004), 373–387. [Shen 2001a] Z. Shen, “Lectures on Finsler geometry”, World Scientific, Singapore, 2001. [Shen 2001b] Z. Shen, Differential geometry of spray and Finsler spaces, Kluwer, Dordrecht, 2001. [Shen 2002] Z. Shen, “Twodimensional Finsler metrics with constant curvature”, Manuscr. Math. 109 (2002), 349–366. [Spanier 1966] E. H. Spanier, “Algebraic Topology”, McGrawHill, New York, 1966; reprinted by Springer, New York, 1989. [Ziller 1982] W. Ziller, “Geometry of the Katok examples”, Ergod. Th. Dyn. Syst. 3 (1982), 135–157.
302
HANSBERT RADEMACHER
HansBert Rademacher ¨ t Leipzig Universita Mathematisches Institut Augustusplatz 10/11 04109 Leipzig Germany
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Landsberg Curvature, SCurvature and Riemann Curvature ZHONGMIN SHEN
Contents 1. Introduction 2. Finsler Metrics 3. Cartan Torsion and Matsumoto Torsion 4. Geodesics and Sprays 5. Berwald Metrics 6. Gradient, Divergence and Laplacian 7. SCurvature 8. Landsberg Curvature 9. Randers Metrics with Isotropic SCurvature 10. Randers Metrics with Relatively Isotropic LCurvature 11. Riemann Curvature 12. Projectively Flat Metrics 13. The Chern Connection and Some Identities 14. Nonpositively Curved Finsler Manifolds 15. Flag Curvature and Isotropic SCurvature 16. Projectively Flat Metrics with Isotropic SCurvature 17. Flag Curvature and Relatively Isotropic LCurvature References
303 304 311 313 315 317 319 321 323 325 327 330 333 337 341 342 348 351
1. Introduction Roughly speaking, Finsler metrics on a manifold are regular, but not necessarily reversible, distance functions. In 1854, B. Riemann attempted to study a special class of Finsler metrics — Riemannian metrics — and introduced what is now called the Riemann curvature. This infinitesimal quantity faithfully reveals the local geometry of a Riemannian manifold and becomes the central concept of Riemannian geometry. It is a natural problem to understand general regular distance functions by introducing suitable infinitesimal quantities. For more than half a century, there had been no essential progress until P. Finsler studied the variational problem in a Finsler manifold. However, it was L. Berwald who 303
304
ZHONGMIN SHEN
first successfully extended the notion of Riemann curvature to Finsler metrics by introducing what is now called the Berwald connection. He also introduced some nonRiemannian quantities via his connection [Berwald 1926; 1928]. Since then, Finsler geometry has been developed gradually. The Riemann curvature is defined using the induced spray, which is independent of any wellknown connection in Finsler geometry. It measures the shape of the space. The Cartan torsion and the distortion are two primary geometric quantities describing the geometric properties of the Minkowski norm in each tangent space. Differentiating them along geodesics gives rise to the Landsberg curvature and the Scurvature. These quantities describe the rates of change of the “color pattern” on the space. In this article, I am going to discuss the geometric meaning of the Landsberg curvature, the Scurvature, the Riemann curvature, and their relationship. I will give detailed proofs for several important local and global results.
2. Finsler Metrics By definition, a Finsler metric on a manifold is a family of Minkowski norms on the tangent spaces. A Minkowski norm on a vector space V is a nonnegative function F : V → [0, ∞) with the following properties: (i) F is positively yhomogeneous of degree one, i.e., F (λy) = λF (y) for any y ∈ V and any λ > 0; (ii) F is C ∞ on V \ {0} and for any tangent vector y ∈ V \ {0}, the following bilinear symmetric form g y : V × V → R is positive definite: 1 ∂2 F 2 (y + su + tv) s=t=0 . 2 ∂s∂t A Minkowski norm is said to be reversible if F (−y) = F (y) for y ∈ V. In this article, Minkowski norms are not assumed to be reversible. From (i) and (ii), one can show that F (y) > 0 for y 6= 0 and F (u + v) ≤ F (u) + F (v) for u, v ∈ V. See [Bao et al. 2000] for a proof. Let h , i denote thepstandard inner product on Rn , defined by hu, vi := Pn i i hy, yi is called the standard Euclidean norm on R n . i=1 u v . Then y := Let b ∈ Rn with b < 1, then F = y + hb, yi is a Minkowski norm on R n , which is called a Randers norm. S Let M be a connected, ndimensional, C ∞ manifold. Let T M = x∈M Tx M be the tangent bundle of M , where Tx M is the tangent space at x ∈ M . We denote a typical point in T M by (x, y), where y ∈ Tx M , and set T M0 := T M \{0} where {0} stands for {(x, 0)  x ∈ X, 0 ∈ Tx M }. A Finsler metric on M is a function F : T M → [0, ∞) with the following properties: g y (u, v) :=
(a) F is C ∞ on T M0 ; (b) At each point x ∈ M , the restriction Fx := F Tx M is a Minkowski norm on Tx M .
LANDSBERG, S AND RIEMANN CURVATURES
305
The pair (M, F ) is called a Finsler manifold. Let (M, F ) be a Finsler manifold. Let (xi , y i ) be a standard local coordinate system in T M , i.e., y i ’s are determined by y = y i (∂/∂xi )x . For a vector y = y i (∂/∂xi )x 6= 0, let gij (x, y) := 21 [F 2 ]yi yj (x, y). The induced inner product g y is given by g y (u, v) = gij (x, y)ui v j , where u = ui (∂/∂xi )x and v = v i (∂/∂xi )x . By the homogeneity of F , q q F (x, y) = g y (y, y) = gij (x, y)y i y j . A Finsler metric F = F (x, y) is called a Riemannian metric if the g ij = gij (x) are functions of x ∈ M only. There are three special Riemannian metrics. Example 2.1 (Euclidean metric). The simplest metric is the Euclidean metric α0 = α0 (x, y) on Rn , which is defined by α0 (x, y) := y,
y = (y i ) ∈ Tx Rn ∼ = Rn .
(2–1)
We will simply denote (R , α0 ) by R , which is called Euclidean space. n
n
Example 2.2 (Spherical metric). Let Sn := {x ∈ Rn+1  x = 1} denote the standard unit sphere in Rn+1 . Every tangent vector y ∈ Tx Sn can be identified with a vector in Rn+1 in a natural way. The induced metric α+1 on Sn is defined by α+1 = kykx , for y ∈ Tx Sn ⊂ Rn+1 , where k·kx denotes the induced Euclidean norm on Tx Sn . Let ϕ : Rn → Sn ⊂ Rn+1 be defined by ϕ(x) :=
x
1
p , p . 1 + x2 1 + x2
(2–2)
Then ϕ pulls back α+1 on the upper hemisphere to a Riemannian metric on R n , which is given by p y2 + (x2 y2 − hx, yi2 ) (2–3) , y ∈ T x Rn ∼ α+1 = = Rn . 1 + x2 Example 2.3 (Hyperbolic metric). Let Bn denote the unit ball in Rn . Define p y2 − (x2 y2 − hx, yi2 ) α−1 := , y ∈ T x Bn ∼ (2–4) = Rn . 1 − x2 We call α−1 the Klein metric and denote (Bn , α−1 ) by Hn . The metrics (2–1), (2–3) and (2–4) can be combined into one formula: p (1 + µx2 )y2 − µhx, yi2 . αµ = 1 + µx2
(2–5)
Of course, there are many nonRiemannian Finsler metrics on R n with special geometric properties. We just list some of them below and discuss their geometric properties later.
306
ZHONGMIN SHEN
Example 2.4 (Funk metric). Let p y2 − (x2 y2 − hx, yi2 ) + hx, yi , Θ := 1 − x2
y ∈ T x Bn ∼ = Rn .
Θ = Θ(x, y) is a Finsler metric on Bn , called the Funk metric on Bn . For an arbitrary constant vector a ∈ Rn with a < 1, let p y2 − (x2 y2 − hx, yi2 ) + hx, yi ha, yi + . Θa := 2 1 − x 1 + ha, xi
(2–6)
(2–7)
where y ∈ Tx Bn ∼ = Rn . Θa = Θa (x, y) is a Finsler metric on Bn . Note that Θ0 = Θ is the Funk metric on Bn . We call Θa the generalized Funk metric on Bn [Shen 2003a]. Example 2.5 [Shen 2003b]. Let δ be an arbitrary number with δ < 1. Let p p y2 −(x2 y2 −hx, yi2 )+hx, yi y2 −δ 2 (x2 y2 −hx, yi2 )+δhx, yi Fδ := −δ , 2 2(1−x ) 2(1−δ 2 x2 ) where y ∈ Tx Bn ∼ = Rn . Fδ is a Finsler metric on Bn . Note that F−1 = α−1 is the Klein metric on Bn . Let Θ be the Funk metric on Bn defined in (2–6). We can express Fδ by Fδ = 21 Θ(x, y) − δΘ(δx, y) . Example 2.6 [Berwald 1929b]. Let
p 2 y2 − (x2 y2 − hx, yi2 ) + hx, yi p B := , (1 − x2 )2 y2 − (x2 y2 − hx, yi2 )
(2–8)
where y ∈ Tx Bn ∼ = Rn . Then B = B(x, y) is a Finsler metric on Bn . Example 2.7. Let ε be an arbitrary number with ε < 1. Let q p √ Ψ 12 ( Φ2 +(1−ε2 )y4 +Φ) + (1−ε2 )hx, yi2 + 1−ε2 hx, yi Fε := , Ψ
(2–9)
where Φ := εy2 + (x2 y2 − hx, yi2 ),
Ψ := 1 + 2εx2 + x4 .
Fε = Fε (x, y) is a Finsler metric on Rn . Note that if ε = 1, then F1 = α+1 is the spherical metric on Rn . R. Bryant [1996; 1997] classified Finsler metrics on the standard unit sphere S2 with constant flag curvature equal to +1 and geodesics being great circles. The Finsler metrics Fε in (2–9) is a special family of Bryant’s metrics expressed in a local coordinate system. See Example 12.7 for further discussion.
LANDSBERG, S AND RIEMANN CURVATURES
307
The examples of Finsler metrics above all have special geometric properties. They are locally projectively flat with constant flag curvature. Some belong to the class of (α, β)metrics, that is, those of the form F = αφ
β α
,
(2–10)
p where α = αx (y) = aij (x)y i y j is a Riemannian metric, β = βx (y) = bi (x)y i is a 1form, and φ is a C ∞ positive function on some interval I = [−r, r] big enough that r ≥ β/α for all x and y ∈ Tx M . It is easy to see that any such F is positively homogeneous of degree one. Let kβk x := supy∈Tx M βx (y)/αx (y). Using a Maple program, we find that the Hessian gij := 12 [F 2 ]yi yj is gij = ρaij + ρ0 bi bj + ρ1 (bi αj + bj αi ) + ρ2 αi αj , where αi = αyi and ρ = φ2 − sφφ0 ,
ρ0 = φφ00 + φ0 φ0 ,
ρ1 = −s(φφ00 + φ0 φ0 ) + φφ0 ,
ρ2 = s2 (φφ00 + φ0 φ0 ) − sφφ0 ,
where s := β/α with s ≤ kβkx ≤ r. Then det (gij ) = φn+1 (φ − sφ0 )
n−2
If φ = φ(s) satisfies φ(s) > 0,
φ(s) − sφ0 (s) > 0,
(φ − sφ0 ) + (kβk2x − s2 )φ00 det (aij ) . (φ − sφ0 ) + (b2 − s2 )φ00 (s) ≥ 0
(2–11)
for all s with s ≤ b ≤ r, then (gij ) is positive definite; hence F is a Finsler metric. Sabau and Shimada [2001] have classified (α, β)metrics into several classes and computed the Hessian gij for each class. Below are some special (α, β)metrics. (a) φ(s) = 1 + s. The defined function F = α + β is a Finsler metric if and only if the norm of β with respect to α is less than 1 at any point: q kβkx := aij (x)bi (x)bj (x) < 1, x ∈ M. A Finsler metric in this form is called a Randers metric. The Finsler metrics in Example 2.4 are Randers metrics. The Finsler metrics in Example 2.5 is the sum of two Randers metrics. (b) φ(s) = (1 + s)2 . The defined function F = (α + β)2 /α is a Finsler metric if and only if kβkx < 1 at any point x ∈ M . The Finsler metric in Example 2.6 is in this form. By a Finsler structure on a manifold M we usually mean a Finsler metric. Sometimes, we also define a Finsler structure as a scalar function F ∗ on T ∗ M such that F ∗ is C ∞ on T ∗ M \ {0} and Fx∗ := F ∗ Tx∗ M is a Minkowski norm on
308
ZHONGMIN SHEN
Tx∗ M for x ∈ M . Such a function is called a coFinsler metric. Given a coFinsler metric, one can define a Finsler metric via the standard duality defined below. Let F ∗ = F ∗ (x, ξ) be a coFinsler metric on a manifold M . Define a nonnegative scalar function F = F (x, y) on T M by ξ(y) . F ∗ (x, ξ)
F (x, y) := sup ξ∈Tx∗ M
Then F = F (x, y) is a Finsler metric on M , said to be dual to F ∗ . In the same sense, F ∗ = F ∗ (x, ξ) is also dual to F : F ∗ (x, ξ) = sup y∈Tx M
ξ(y) . F (x, y)
Every vector y ∈ Tx M \ {0} uniquely determines a covector ξ ∈ Tx∗ M \ {0} by ξ(w) :=
1 d F 2 (x, y + tw) t=0 , 2 dt
w ∈ Tx M.
1 d F ∗2 (x, ξ + tη) t=0 , 2 dt
η ∈ Tx∗ M.
The resulting map `x : y ∈ Tx M → ξ ∈ Tx∗ M is called the Legendre transformation at x. Similarly, every covector ξ ∈ Tx∗ M \ {0} uniquely determines a vector y ∈ Tx M \ {0} by η(y) :=
The resulting map `∗x : ξ ∈ Tx∗ M → y ∈ Tx M is called the inverse Legendre transformation at x. Indeed, `x and `∗x are inverses of each other. Moreover, they preserve the Minkowski norms: F (x, y) = F ∗ (x, `x (y)),
F ∗ (x, ξ) = F (x, `∗x (ξ)).
(2–12)
Let Φ = Φ(x, y) be a Finsler metric on a manifold M and let Φ ∗ = Φ∗ (x, ξ) be the coFinsler metric dual to Φ. By the above formulas, one can easily show that if y ∈ Tx M \ {0} and ξ ∈ Tx∗ M \ {0} satisfy
Then
d ∗ Φ (x, ξ + tη) t=0 = η(y), dt
η ∈ Tx∗ M.
Φ(x, y) = 1.
(2–13)
Let V be a vector field on M with Φ(x, −Vx ) < 1 and let V denote the 1form dual to V , defined by Vx∗ (ξ) = ξ(Vx ),
∗
∗
: T M → [0, ∞)
ξ ∈ Tx∗ M.
We have Φ∗ (x, −Vx∗ ) = Φ(x, −Vx ) < 1. Thus F ∗ := Φ∗ + V ∗ is a coFinsler metric on M . Define F = F (x, y) by F (x, y) := sup ξ∈Tx∗ M
ξ(y) , F ∗ (x, ξ)
y ∈ Tx M.
(2–14)
LANDSBERG, S AND RIEMANN CURVATURES
309
F is a Finsler metric on M , called the Finsler metric generated from the pair (Φ, V ). One can also define F in a different way without using the duality: Lemma 2.8. Let Φ = Φ(x, y) be a Finsler metric on M and let V be a vector field on M with Φ(x, −Vx ) < 1 for all x ∈ M . Then F = F (x, y) defined in (2–14) satisfies y (2–15) − Vx = 1, y ∈ Tx M. Φ x, F (x, y)
Conversely, if F = F (x, y) is defined by (2–15), it is dual to the coFinsler metric F ∗ := Φ∗ + V ∗ as defined in (2–14).
Proof. For the coFinsler metric F ∗ = Φ∗ + V ∗ , let F = F (x, y) be defined in (2–14). Fix an arbitrary nonzero vector y ∈ Tx M . There is a covector ξ ∈ Tx∗ M such that ξ(y) F (x, y) = ∗ . (2–16) F (x, ξ) Let η ∈ Tx∗ M be an arbitrary covector. Consider the function h(t) :=
ξ(y) + tη(y) . Φ∗ (x, ξ + tη) + ξ(Vx ) + tη(Vx )
Then h(t) ≤ h(0) = F (x, y). Thus h0 (0) = 0: η(y)F ∗ (x, ξ) − ξ(y)
d
dt
Φ∗ (x, ξ + tη) t=0 + η(Vx ) = 0.
Dividing by F ∗ (x, ξ) and using (2–16), one obtains η(y) − F (x, y)
d
dt
From this identity it follows that
Φ∗ (x, ξ + tη) t=0 + η(Vx ) = 0.
y d ∗ − Vx , Φ (x, ξ + tη) t=0 = η dt F (x, y)
η ∈ Tx∗ M.
Thus F (x, y) satisfies (2–15) as we have explained in (2–13). Conversely, let F = F (x, y) be defined by (2–15). Then for any ξ ∈ T x∗ M , Φ∗ (x, ξ) = sup η y∈Tx M
y − Vx . F (x, y)
One obtains sup y∈Tx M
y ξ(y) = sup ξ − Vx + ξ(Vx ) = Φ∗ (x, ξ) + Vx∗ (ξ) = F ∗ (x, ξ). F (x, y) y∈Tx M F (x, y)
Thus F ∗ is dual to F and so F is dual to F ∗ , that is, F is given by (2–14).
310
ZHONGMIN SHEN
p Let Φ = φij (x)y i y j be a Riemannian metric and let V = V i (x)(∂/∂xi ) be a vector field on a manifold M with q Φ(x, −Vx ) = kV kx := φij (x)V i (x)V j (x) < 1, x ∈ M. Solving (2–15) for F = F (x, y), one obtains p (1 − φij V i V j )φij y i y j + (φij y i V j )2 − φij y i V j F = . 1 − φij V i V j
(2–17)
Clearly, F is a Randers that any Randers metric p metric. It is easy to verify i i j F = α+β, where α = aij (x)y y and β = bi (x)y , can be expressed in the form (2–17). According to Lemma 2.8, any Randers metric F = α + β expressed in p the form (2–17) can be constructed in the following way. Let Φ ∗ := φij (x)ξi ξj p be the Riemannian metric dual to Φ = φij (x)y i y j and V ∗ := ξ(Vx ) = V i (x)ξi p be the 1form dual to V . Then F ∗ := Φ∗ (x, ξ) + V ∗ (ξ) = φij (x)ξi ξj + V i (x)ξi is a coFinsler metric on M . Moreover, the dual Finsler metric F of F ∗ is given by (2–17). This is proved in [Hrimiuc and Shimada 1996]. It was discovered in [Shen 2003c; 2002] that if Φ is a Riemannian metric of constant curvature and V is a special vector field, the generated metric F is of constant flag curvature. This discovery opens the door to a classification of Randers metrics of constant flag curvature [Bao et al. 2003]. But Maple programs played an important role in the computations that led to it. Example 2.9. Let φ = φ(y) be a Minkowski norm on Rn and let Uφ := y ∈ Rn  φ(y) < 1 .
Define
Φ(x, y) := φ(y),
y ∈ T x Rn ∼ = Rn .
Φ = Φ(x, y) is called a Minkowski metric on Rn . Let Vx := −x, for x ∈ Rn . V is a radial vector field pointing toward the origin. For any x ∈ U φ , Φ x, −Vx = φ(x) < 1.
The pair (Φ, V ) generates a Finsler metric Θ = Θ(x, y) on U φ by (2–15): Θ(x, y) = φ y + Θ(x, y)x . (2–18)
Differentiating with respect to xk and y k separately, one obtains 1 − φwl (w)xl Θxk (x, y) = φwk (w)Θ(x, y), 1 − φwl (w)xl Θyk (x, y) = φwk (w), where w := y + Θ(x, y)x. It follows that
Θxk (x, y) = Θ(x, y)Θyk (x, y). This argument is from [Okada 1983].
(2–19)
LANDSBERG, S AND RIEMANN CURVATURES
311
A domain Uφ in Rn defined by a Minkowski norm φ is called a strongly convex domain. A Finsler metric Θ = Θ(x, y) defined in (2–18) is called the Funk metric on a strongly convex domain in Rn . When φ = y is the standard Euclidean metric on Rn , Uφ = Bn is the standard unit ball and Θ = Θ(x, y) is given by (2–6). Equation (2–19) is the key property of Θ, from which one can derive other geometric properties of Θ. Definition 2.10. A Finsler function Θ = Θ(x, y) on an open subset in R n is called a Funk metric if it satisfies (2–19). Example 2.11. Let Φ(x, y) := y be the standard Euclidean metric on R n and let V = V (x) be a vector field on Rn defined by Vx := x2 a − 2ha, xix, where a ∈ Rn is a constant vector. Note that p Φ(x, −Vx ) = φij V i V j = Vx  = ax2 < 1,
x ∈ Bn (1/
and that
p
a),
φij y i V j = x2 ha, yi − 2ha, xihx, yi. Given the pair (Φ, V ) above, one obtains, by solving (2–15) for F , q 2 x2 ha, yi−2ha, xihx, yi + y2 1−a2 x4 − x2 ha, yi−2ha, xihx, yi F = . 1 − a2 x4 (2–20) This Randers metric F has very important properties. It is of scalar curvature and isotropic Scurvature. But the flag curvature and the Scurvature are not constant. See Example 11.2 below for further discussion.
3. Cartan Torsion and Matsumoto Torsion To characterize Euclidean norms, E. Cartan [1934] introduced what is now called the Cartan torsion. Let F = F (y) be a Minkowski norm on a vector space V. Fix a basis {bi } for V. Then F = F (y i bi ) is a function of (y i ). Let gij := 12 [F 2 ]yi yj ,
Cijk := 41 [F 2 ]yi yj yk (y),
Ii := g jk (y)Cijk (y),
where (g ij ) := (gij )−1 . It is easy to see that Ii =
p ∂ ln det(gjk ) . ∂y i
(3–1)
For y ∈ V \ {0}, set C y (u, v, w) := Cijk (y)ui v j wk ,
I y (u) := Ii (y)ui ,
where u := ui bi , v := v j bj and w := w k bk . The family C := {C y  y ∈ V \ {0}} is called the Cartan torsion and the family I := {I y  y ∈ V \ {0}} is called the
312
ZHONGMIN SHEN
mean Cartan torsion. They are not tensors in a usual sense. In later sections, we will convert them to tensors on T M0 and call them the (mean) Cartan tensor. We view a Minkowski norm F on a vector space V as a color pattern. When F is Euclidean, the color pattern is trivial or Euclidean. The Cartan torsion C y describes the nonEuclidean features of the color pattern in the direction y ∈ V \ {0}. And the mean Cartan torsion I y is the average value of C y . A trivial fact is that a Minkowski norm F on a vector space V is Euclidean if and only if C y = 0 for all y ∈ V \ {0}. This can be improved: Proposition 3.1 [Deicke 1953]. A Minkowski norm is Euclidean if and only if I = 0. To characterize Randers norms, M. Matsumoto introduces the quantity Mijk := Cijk −
1 Ii hjk + Ij hik + Ik hij , n+1
(3–2)
where hij := F Fyi yj = gij − gip y p gjq y q/F 2 . For y ∈ V \ {0}, set M y (u, v, w) := Mijk (y)ui v j wk , where u = ui bi , v = v j bj and w = w k bk . The family M := {M y  y ∈ V\{0}} is called the Matsumoto torsion. A Minkowski norm is called Creducible if M = 0. Lemma 3.2 [Matsumoto 1972b]. Every Randers metric satisfies M = 0. Proof. Let p F = α + β be an iarbitrary Randers norm on a vector space V, where α = aij y i y j and β = bi y with kβkα < 1. By a direct computation, the gij := 21 [F 2 ]yi yj are given by F gij = α
α yi yj yi + aij − bi + α α F α
yj bj + α
(3–3)
,
where yi := aij y j . The hij = F Fyi yj = gij − gip y p gjq y q/F 2 are given by hij =
yi yj α+β aij − 2 . α α
The inverse matrix (g ij ) = (gij )−1 is given by g ij =
α F
aij − (1 − kβk2 )
α yi yj + F F F
The determinant det(gij ) is det (gij ) = From this and (3–1), one obtains Ii =
bi −
α + β n+1 α
yi α
bj −
(3–4)
yj F
− b i bj
.
(3–5)
det (aij ) .
n+1 yi β bi − . 2(α + β) αα
(3–6)
LANDSBERG, S AND RIEMANN CURVATURES
313
Differentiating (3–3) yields Cijk = implying that Mijk = 0.
1 Ii hjk + Ij hik + Ik hij , n+1
(3–7)
Matsumoto and H¯ oj¯ o later proved that the converse is true as well, if dim V ≥ 3. Proposition 3.3 [Matsumoto 1972b; Matsumoto and H¯ oj¯ o 1978]. If F is a Minkowski norm on a vector space V of dimension at least 3, the Matsumoto torsion of F vanishes if and only if F is a Randers norm. Their proof is long and I could not find a shorter proof which fits into this article.
4. Geodesics and Sprays Every Finsler metric F on a connected manifold M defines a length structure LF on oriented curves in M . Let c : [a, b] → M be a piecewise C ∞ curve. The length of c is defined by Z b LF (c) := F c(t), c(t) ˙ dt. a
For any two points p, q ∈ M , define dF (p, q) := inf c LF (c), where the infimum is taken over all piecewise C ∞ curves c from p to q. The quantity dF = dF (p, q) is a nonnegative function on M × M . It satisfies (a) dF (p, q) ≥ 0, with equality if and only if p = q; and (b) dF (p, q) ≤ dF (p, r) + dF (r, q) for any p, q, r ∈ M . We call dF the distance function induced by F . This is a weaker notion than the distance function of metric spaces, since dF need not satisfy dF (p, q) = dF (q, p) for p, q ∈ M . But if the Finsler metric F is reversible, that is, if F (x, −y) = F (x, y) for all y ∈ Tx M , then dF is symmetric. A piecewise C ∞ curve σ : [a, b] → M is minimizing if it has least length: LF (σ) = dF σ(a), σ(b) .
It is locally minimizing if, for any t0 ∈ I := [a, b], there is a small number ε > 0 such that σ is minimizing when restricted to [t0 − ε, t0 + ε] ∩ I. Definition 4.1. A C ∞ curve σ(t), t ∈ I, is called a geodesic if it is locally minimizing and has constant speed (meaning that F (σ(t), σ(t)) ˙ is constant).
Lemma 4.2. A C ∞ curve σ(t) in a Finsler manifold (M, F ) is a geodesic if and only if it satisfies the system of second order ordinary differential equations σ ¨ i (t) + 2Gi σ(t), σ(t) ˙ = 0, (4–1)
314
ZHONGMIN SHEN
where the Gi = Gi (x, y) are local functions on T M defined by Gi := 41 g il (x, y) [F 2 ]xk yl (x, y)y k − [F 2 ]xl (x, y) .
(4–2)
This is shown by the calculus of variations; see, for example, [Shen 2001a; 2001b]. Let {∂/∂xi , ∂/∂y i } denote the natural local frame on T M in a standard local coordinate system, and set G := y i
∂ ∂ − 2Gi i , ∂xi ∂y
(4–3)
where the Gi = Gi (x, y), which are given in (4–2), satisfy the homogeneity property Gi (x, λy) = λ2 Gi (x, y), λ > 0. (4–4) G is a welldefined vector field on T M . Any vector field G on T M having the form (4–3) and satisfying the homogeneity condition (4–4) is called a spray on M , and the Gi are its spray coefficients. Let δ ∂ ∂ ∂Gi , := − Nji j . Nji = ∂y j δxi ∂xi ∂y Then HT M := span{δ/δxi } and V T M := span{∂/∂y i } are welldefined and T (T M0 ) = HT M ⊕ V T M . That is, every spray naturally determines a decomposition of T (T M0 ). For a Finsler metric on a manifold M and its spray G, a C ∞ curve σ(t) in M is a geodesic of F if and only if the canonical lift γ(t) := σ(t) ˙ in T M is an integral curve of G. One can use this to define the notion of geodesics for sprays. It is usually difficult to compute the spray coefficients of a Finsler metric. However, for an (α, β)metric F , given by equation (2–10), the computation is relatively simple using a Maple program. Let Gi be the spray coefficients of the Riemannian metric α, with coefficients Γ ijk , so that Gi = 12 Γijk (x)y j y k . These coefficients are called the Christoffel symbols of α. By (4–2), they are given by Γijk =
ail ∂ajl ∂akl ∂ajk + − . 2 ∂xk ∂xj ∂xl
To find a formula for the spray coefficients Gi = Gi (x, y) of F in terms of α and β, we introduce the covariant derivatives of β with respect to α. Let θ i := dxi and θj i := Γijk dxk . We have dθi = θj ∧ θj i ,
daij = akj θi k + aik θj k .
Define bi;j by bi;j θj := dbi − bj θi j . Let rij :=
1 2
bi;j + bj;i ,
si j := aih shj ,
sj := bi si j ,
sij :=
1 2
bi;j − bj;i ,
eij := rij + bi sj + bj si .
By (4–2) and using a Maple program, one obtains the following relationship:
LANDSBERG, S AND RIEMANN CURVATURES
315
Lemma 4.3. The geodesic coefficients Gi are related to Gi by Gi = G i +
φφ0 − s(φφ00 + φ0 φ0 ) αφ0 i s + φ − sφ0 0 2φ (φ − sφ0 ) + (b2 − s2 )φ00 ×
−2αφ0 φ − sφ
s + r00 0 0
y i α
+
φφ00 bi , 0 φφ − s(φφ00 + φ0 φ0 )
where s = β/α, si 0 = si j y j , s0 = si y i , r00 = rij y i y j and b2 = aij bi bj .
(4–5)
Consider the metric F =
(α + β)2 , α
p where α = aij (x)y i y j is a Riemannian metric and β = bi (x)y i is a 1form with kβkx < 1 for every x ∈ M . By (4–5), we obtain a formula for the spray coefficients of F : Gi = G i +
2α α(α − 2β) −4α αsi 0 + 2 2 αs0 + r00 α−β 2α b + α2 − 3β 2 α − β
y i α
+
α bi , α − 2β
where b = kβkx . Given a spray G, we define the covariant derivatives of a vector field X = i X (t)(∂/∂xi )c(t) along a curve c by ∂ , ∂xi c(t) ∂ ∇c˙ X(t) := X˙ i (t) + X j (t)Nji (c(t), X(t)) . ∂xi c(t)
Dc˙ X(t) := X˙ i (t) + X j (t)Nji (c(t), c(t)) ˙
(4–6)
Dc˙ X(t) is the linear covariant derivative and ∇c˙ X(t) the covariant derivative of X(t) along c. The field X is linearly parallel along c if Dc˙ X(t) = 0, and parallel along c if ∇c˙ X(t) = 0. For linearly parallel vector fields X = X(t) and Y = Y (t) X(t), Y (t) is constant, and for a parallel along a geodesic c, the expression g c(t) ˙ vector field X = X(t) along a curve c, F c(t), X(t) is constant.
5. Berwald Metrics Consider a Riemannian metric F = we obtain Gi = 12 Γijk (x)y j y k , where Γijk (x) := 14 g il (x)
∂g
p
gij (x)y i y j on a manifold M . By (4–2),
lk (x) ∂xj
+
∂gjl ∂gjk (x) − (x) . ∂xk ∂xl
(5–1)
In this case the Gi = Gi (x, y) are quadratic in y ∈ Tx M at any point x ∈ M . A Finsler metric F = F (x, y) is called a Berwald metric if in any standard local coordinate system, the spray coefficients G i = Gi (x, y) are quadratic in y ∈ Tx M . There are many nonRiemannian Berwald metrics.
316
ZHONGMIN SHEN
¯ and (M , α) be Riemannian manifolds and let M = Example 5.1. Let (M , α) M × M . Let f : [0, ∞) × [0, ∞) → [0, ∞) be a C ∞ function satisfying f (λs, λt) = λf (s, t) for λ > 0 and f (s, t) 6= 0 if (s, t) 6= 0. Define q F (x, y) := f [¯ α(¯ x, y¯)]2 , [α(x, y)]2 , (5–2) where x = (¯ x, x) ∈ M and y = y¯ ⊕ y ∈ Tx M ∼ = Tx¯ M ⊕ Tx M . We now find additional conditions on f (s, t) under which the matrix g ij := 1 2 i j x a¯ , y¯a¯ ) 2 [F ]y y is positive definite. Take standard local coordinate systems (¯ a ¯ a ¯ a a a a ¯ and y = y ∂/∂x . Then in T M and (x , y ) in T M , so that y¯ = y¯ ∂/∂ x i j a ¯ ¯ a a a (x , y ) := (¯ x , x , y¯ , y ) is a standard local coordinate system in T M . Express α ¯ and α by q q α ¯ (¯ x, y¯) = g¯a¯¯b (¯ x)¯ y a¯ y¯¯b , α(x, y) = g ab (x)y a y b , We obtain gab = 2fss y¯a¯ y¯¯b + fs g¯a¯¯b ,
gab = 2fst y¯a¯ y β ,
gab = 2ftt y a y b + ft g ab ,
(5–3)
¯
where y¯a¯ := g¯a¯¯b y¯b and y a := g ab y b . By an elementary argument, one can show that (gij ) is positive definite if and only if f (s, t) satisfies the conditions fs > 0,
ft > 0,
fs + 2sfss > 0,
ft + 2tftt > 0,
fs ft − 2f fst > 0.
In this case,
where
det (gij ) = h [¯ α]2 , [α]2 det (¯ ga¯¯b ) det g ab ,
(5–4)
h := (fs )n¯ −1 (ft )n−1 (fs ft − 2f fst ),
(5–5)
where n ¯ := dim M and n := dim M . By a direct computation, one can show that the spray coefficients of F split as the direct sum of the spray coefficients of α ¯ and α: x, y¯), Ga¯ (x, y) = Ga¯ (¯
Ga (x, y) = Ga (x, y),
(5–6)
where Ga¯ and Ga are the spray coefficients of α ¯ and α. From (5–6), one can see that the spray of F is independent of the choice of a particular function f (s, t). In particular, the Gi (x, y) are quadratic in y ∈ Tx M . Thus F is a Berwald metric. A typical example of f is p k f = s + t + ε sk + tk , where ε is a nonnegative real number and k is a positive integer. The Berwald metric obtained with this choice of f is discussed in [Szab´ o 1981].
LANDSBERG, S AND RIEMANN CURVATURES
317
Let (M, F ) be a Berwald manifold and p, q ∈ M be an arbitrary pair of points in M . Let c : [0, 1] → M be a geodesic going from p = c(0) to q = c(1). Define a linear isomorphism T : Tp M → Tq M by T (X(0)) := X(1), where X(t) is a linearly parallel vector field along c, so Dc˙ X(t) = 0. Since F is a Berwald metric, the linear covariant derivative ∇c˙ coincides with the covariant derivative Dc˙ along c, by (4–6). Thus X(t) is also parallel along c, that is, ∇c˙ X(t) = 0. Therefore, F (c(t), X(t)) is constant. This implies that T : (T p M, Fp ) → (Tq M, Fq ) preserves the Minkowski norms. We have proved the following wellknown result: Proposition 5.2 [Ichijy¯ o 1976]. On a Berwald manifold (M, F ), all tangent spaces (Tx M, Fx ) are linearly isometric to each other . On a Finsler manifold (M, F ), we view the Minkowski norm F x on Tx M as an infinitesimal color pattern at x. As we mentioned early in Section 3, the Cartan torsion C y describes the nonEuclidean features of the pattern in the direction y ∈ Tx M \ {0}. In the case when F is a Berwald metric on a manifold M , by Proposition 5.2, all tangent spaces (Tx M, Fx ) are linearly isometric, and (M, F ) is modeled on a single Minkowski space. More precisely, for any pair points x, x0 ∈ M and a geodesic from x to x0 , (linearly) parallel translation defines a linear isometry T : (Tx M, Fx ) → (Tx0 M, Fx0 ). This linear isometry T maps the infinitesimal color pattern at x to that at x 0 . Thus the infinitesimal color patterns do not change over the manifold. If one looks at a Berwald manifold on a large scale, with the infinitesimal color pattern at each point shrunken to a single spot of color, one can only see a space with uniform color. The color depends on the Minkowski model. A Finsler metric F on a manifold M is said to be affinely equivalent to another Finsler metric F¯ on M if F and F¯ induce the same sprays. By (5–6), one can see that the family of Berwald metrics in (5–2) are affinely equivalent. Proposition 5.3 [Szab´ o 1981]. Every Berwald metric on a manifold is affinely equivalent to a Riemannian metric. Based on this observation, Z. I. Szab´ o [1981] determined the local structure of Berwald metrics.
6. Gradient, Divergence and Laplacian Let F = F (x, y) be a Finsler metric on a manifold M and let F ∗ = F ∗ (x, ξ) be dual to F . Let f be a C 1 function on M . At a point x ∈ M , the differential dfx ∈ Tx∗ M is a 1form. Define the dual vector ∇fx ∈ Tx M by ∇fx := `∗x (dfx ),
(6–1)
where `∗x : Tx∗ M → Tx M is the inverse Legendre transformation. By definition, ∇fx is uniquely determined by 1 d F ∗2 (x, dfx + tη) t=0 , η ∈ Tx∗ M. η(∇fx ) := 2 dt
318
ZHONGMIN SHEN
∇fx is called the gradient of f at x. We have F (x, ∇fx ) = F ∗ (x, dfx ). If f is C k (k ≥ 1), then ∇f is C k−1 on {dfx 6= 0} and C 0 at any point x ∈ M with dfx = 0. Given a closed subset A ⊂ M and a point x ∈ M , let d(A, x) := supz∈A d(z, x) and d(x, A) := supz∈A d(x, z). The function ρ defined by ρ(x) := d(x, A) is locally Lipschitz, hence differentiable almost everywhere. It is easy to verify [Shen 2001b] that F (x, ∇ρx ) = F ∗ (x, dρx ) = 1 almost everywhere.
(6–2)
Any function ρ satisfying satisfies (6–2) is called a distance function of the Finsler metric F ; another example is the function ρ defined by ρ(x) := −d(x, A). In general, a distance function on a Finsler manifold (M, F ) is C ∞ on an open dense subset of M . For example, when A = {p} is a point, the distance function ρ defined by ρ(x) := d(x, p) or rho(x) := d(p, x) is C ∞ on an open dense subset of M . Let ρ be a distance function of a Finsler metric F on a manifold M . Assume that itpis C ∞ on an open subset U ⊂ M . Then ∇ρ induces a Riemannian metric ˆ is Fˆ := g ∇ρ (v, v) on U . Moreover, ρ is a distance function of Fˆ and ∇ρ = ∇ρ ˆ the gradient of ρ with respect to F . See [Shen 2001b]. Every Finsler metric F defines a volume form dVF := σF (x) dx1 · · · dxn , where σF :=
Vol
(y i )
∈
Rn
Vol Bn .  F (x, y i (∂/∂xi )x ) < 1
(6–3)
Here Vol denotes Euclidean volume in Rn . It was proved by H. Busemann [1947] that if F is reversible, the Hausdorff measure of the induced distance function p dF is represented by dVF . When F = gij (x)y i y j is Riemannian, p p σF = det(gij (x)) and dVF = det(gij (x)) dx1 · · · dxn . For a vector field X = X i (x)(∂/∂xi )x on M , the divergence of X is div X :=
∂ 1 σF (x)X i (x) . i σF (x) ∂x
(6–4)
The Laplacian of a C 2 function f with df 6= 0 is ∆f := div ∇f. ∆ is a nonlinear elliptic operator. If there are points x at which df x = 0, then ∇f is only C 0 at these points. In this case, ∆f is only defined weakly in the sense of distributions.
LANDSBERG, S AND RIEMANN CURVATURES
319
For a C ∞ distance function ρ on an open subset U ⊂ M , dρx 6= 0 at every point x ∈ U and the level set Nr := ρ−1 (r) ⊂ U is a C ∞ hypersurface in U . Thus ∆ρ can be defined by the above formula and its restriction to N r , H := ∆ρNr , is called the mean curvature of Nr with respect to the normal vector n = ∇ρNr .
7. SCurvature Consider an ndimensional Finsler manifold (M, F ). As mentioned in Section 5, we view the Minkowski norm Fx on Tx M as an infinitesimal color pattern at x. The Cartan torsion C y describes the nonEuclidean features of the pattern in the direction y ∈ Tx M \ {0}. The mean Cartan torsion I y is the average value of C y . Besides the (mean) Cartan torsion, there is another geometric quantity associated with Fx . Take a standard local coordinate system (xi , y i ) and let q det gij (x, y) , (7–1) τ := ln σF (x) where σF is defined in (6–3). τ is called the distortion [Shen 1997; 2001b]. Intuitively, the distortion τ (x, y) is the directional twisting number of the infinitesimal color pattern at x. Observe that q 1 ∂ ∂gjk det gjk (x, y) = g jk τy i = ln = g jk Cijk =: Ii . (7–2) ∂y i 2 ∂y i Here σF does not occur in the first equality, because it is independent of y at each point x. If the distortion is isotropic at x, that is, if τ (x) is independent of the direction y ∈ Tx M , then τ (x) = 0 and Fx is Euclidean (Proposition 3.1). In this case, the infinitesimal color pattern is in the simplest form at every point. It is natural to study the rate of change of the distortion along geodesics. For y ∈ Tx M \ {0}, let σ(t) be the geodesic with σ(0) = x and σ(0) ˙ = y. Let S :=
d τ σ(t), σ(t) ˙ . dt t=0
(7–3)
S is called the Scurvature. It is positively homogeneous of degree one in y: S(x, λy) = λS(x, y),
λ > 0.
In a standard local coordinate system (xi , y i ), let Gi = Gi (x, y) denote the spray coefficients of F . Contracting (8–2) with g ij yields ∂gml i ∂Gm = 21 g ml y − 2Ii Gi , ∂y m ∂xi so S = yi
∂gml i ∂τ ∂τ ∂ − 2 i Gi = 21 g ml y − 2Ii Gi − y m m ln σF ∂xi ∂y ∂xi ∂x
320
ZHONGMIN SHEN
and finally S=
∂ ∂Gm − y m m ln σF . ∂y m ∂x
(7–4)
Proposition 7.1 [Shen 1997]. For any Berwald metric, S = 0. However, many metrics of zero Scurvature are nonBerwaldian. Some comparison theorems in Riemannian geometry are still valid for Finsler metrics of zero Scurvature; see [Shen 1997; 2001b]. By definition, the Scurvature is the covariant derivative of the distortion along geodesics. Let σ(t) be a geodesic and set τ (t) := τ σ(t), σ(t) ˙ , S(t) := S σ(t), σ(t) ˙ .
By (7–3), S(t) = τ 0 (t), so if S vanishes, τ (t) is a constant. Intuitively, the distortion (twisting number) of the infinitesimal color pattern in the direction σ(t) ˙ does not change along any geodesic. However, the distortion might take different values along different geodesics. A Finsler metric F is said to have isotropic Scurvature if S = (n + 1)cF. More generally, F is said to have almost isotropic Scurvature if S = (n + 1)(cF + η), where c = c(x) is a scalar function and η = ηi (x)y i is a closed 1form. Differentiating the Scurvature gives rise to another quantity. Let Eij := 12 S yi yj (x, y).
(7–5)
For y ∈ Tx M \ {0}, the form E y = Eij (x, y) dxi ⊗ dxj is a symmetric bilinear form on Tx M . We call the family E := {E y  y ∈ T M \ {0}} the mean Berwald curvature, or simply the Ecurvature [Shen 2001a]. Let h y := hij (x, y) dxi ⊗ dxj , where hij := F Fyi yj . We say that F has isotropic Ecurvature if E = 12 (n + 1)cF −1 h, where c = c(x) is a scalar function on M . Clearly, if the Scurvature is almost isotropic, the Ecurvature is isotropic. Conversely, if the Ecurvature is isotropic, there is a 1form η = ηi (x) dxi such that S = (n + 1)(cF + η). However, this η is not closed in general. Finally, we mention another geometric meaning of the Scurvature. Let ρ = ρ(x) be a C ∞ distance function on an open subset U ⊂ M (see (6–2)). The gradient ∇ρ induces a Riemannian metric Fˆ = Fˆ (z, v) on U by p Fˆ (z, v) := g ∇ρ (v, v), v ∈ Tz U.
LANDSBERG, S AND RIEMANN CURVATURES
321
ˆ denote the Laplacians on functions with respect to F and Fˆ . Then Let ∆ and ∆ ˆ = ∆ρ ˆ N are the mean curvature of Nr := ρ−1 (r) with respect H = ∆ρNr and H r ˆ to F and F , respectively. The Scurvature can be expressed by ˆ − ∆ρ = H ˆ − H. S(∇ρ) = ∆ρ ˆ and obtain an estimate on ∆ρ under From these identities, one can estimate ∆ a Ricci curvature bound and an Scurvature bound. Then one can establish a volume comparison on the metric balls. See [Shen 2001b] for more details.
8. Landsberg Curvature The (mean) Cartan torsion is a geometric quantity that characterizes the Euclidean norms among Minkowski norms on a vector space. On a Finsler manifold (M, F ), one can view the Minkowski norm Fx on Tx M as an infinitesimal color pattern at x. The Cartan torsion C y describes the nonEuclidean features of the pattern in the direction y ∈ Tx M \ {0}. The mean Cartan torsion I y is the average value of C y . They reveal the nonEuclidean features which are different from that revealed by the distortion. Therefore, it is natural to study the rate of change of the (mean) Cartan torsion along geodesics. Let (M, F ) be a Finsler manifold. To differentiate the (mean) Cartan torsion along geodesics, we need linearly parallel vector fields along a geodesic. Recall that a vector field U (t) := U i (T )(∂/∂xi )σ(t) along a geodesic σ(t) is linearly parallel along σ if Dσ˙ U (t) = 0: U˙ i (t) + U j (t)Nji (σ(t), σ(t)) ˙ = 0.
(8–1)
On the other hand, by a direct computation using (4–2), one can verify that g ij satisfy the following identity: ym
∂gij ∂gij − 2Gm m = gim Njm + gmj Nim ∂xm ∂y
(8–2)
Using (8–1) and (8–2), one can verify that for two linearly parallel vector fields U (t), V (t) along σ, d U (t), V (t) = 0. g σ(t) ˙ dt In this sense, the family of inner products g y does not change along geodesics. However, for linearly parallel vector fields U (t), V (t) and W (t) along σ, the U (t) do change, in general. Set U (t), V (t), W (t) and I σ(t) functions C σ(t) ˙ ˙ Ly (u, v, w) := and
d C σ(t) (U (t), V (t), W (t)) t=0 ˙ dt
(8–3)
d I σ(t) (U (t)) t=0 , ˙ dt where u = U (0), v = V (0), w = W (0) and y = σ(0) ˙ ∈ Tx M . The family L = {Ly  y ∈ T M \ {0}} is called the Landsberg curvature, or Lcurvature, and J y (u) :=
322
ZHONGMIN SHEN
the family J = {J y  y ∈ T M \ {0}} is called the mean Landsberg curvature, or Jcurvature. A Finsler metric is called a Landsberg metric if L = 0, and a weakly Landsberg metric if J = 0. Let (xi , y i ) be a standard local coordinate system in T M and set C ijk := 1 2 i j k 4 [F ]y i y j y k . From the definition, Ly = Lijk dx ⊗ dx ⊗ dx is given by Lijk = y m
∂Cijk ∂Cijk − 2Gm − Cmjk Nim − Cimk Njm − Cijm Nkm , m ∂x ∂y m
(8–4)
and J = Ji dxi is given by Ji = y m
∂Ii ∂Ii − 2Gm m − Im Nim . m ∂x ∂y
(8–5)
We have Ji = g jk Lijk . It follows from (4–2) that gsm Gm =
∂gkm k m 1 ∂gsk y y . 2 − 4 ∂xm ∂xs
Differentiating with respect to y i , y j , y k and contracting the resulting identity by 21 y s , one obtains ∂ 3 Gm Lijk = − 12 y s gsm i j k . (8–6) ∂y ∂y ∂y Thus if Gm = Gm (x, y) are quadratic in y ∈ Tx M , then Lijk = 0. This proves the following wellknown result. Proposition 8.1. Every Berwald metric is a Landsberg metric. By definition, the (mean) Landsberg curvature is the covariant derivative of the (mean) Cartan torsion along a geodesic. Let σ = σ(t) be a geodesic and U = U (t), V = V (t), W = W (t) be parallel vector fields along σ. Let L(t) := Lσ(t) U (t), V (t), W (t) , C(t) := C σ(t) U (t), V (t), W (t) . ˙ ˙
By (8–3),
L(t) = C 0 (t). If F is Landsbergian, the Cartan torsion C σ˙ in the direction σ(t) ˙ is constant along σ. Intuitively, the infinitesimal color pattern in the direction σ(t) ˙ does not change along σ. But the patterns might look different at neighboring points. It is easy to see that in dimension two, a Finsler metric is Berwaldian if and only if E = 0 (or S = 0) and J = 0. It seems that E and L are complementary to each other. So we may ask: Is a Finsler metric Berwaldian if E = 0 and L = 0? A more difficult problem is: Is a Finsler metric Berwaldian if L = 0? So far, we do not know. Finsler metrics with L = 0 can be generalized as follows. Let F be a Finsler metric on an ndimensional manifold M . We say that F has relatively isotropic
LANDSBERG, S AND RIEMANN CURVATURES
323
Lcurvature L + cF C = 0, where c = c(x) is a scalar function on M . We say that F has relatively isotropic Jcurvature if J + cF I = 0. There are many interesting Finsler metrics having isotropic Lcurvature or (almost) isotropic Scurvature. We will discuss them in the following two sections.
9. Randers Metrics with Isotropic SCurvature We now discuss Randers metrics of isotropic Scurvature. Let p F = α + β be a Randers metric on an ndimensional manifold M , with α = aij (x)y i y j and β = bi (x)y i . Recall from page 307 that this is a special case of an (α, β)metric, with φ(s) = 1 + s. By (4–5), the spray coefficients Gi of F and Gi of α are related via Gi = G i + P y i + Q i , (9–1) where
e00 − s0 , Qi = αsi 0 , (9–2) 2F and e00 := eij y i y j , s0 := si y i , si 0 := si j y j . The formula above can be found in [Antonelli et al. 1993]. Let p ρ := ln 1 − kβk2x . P :=
The volume forms dVF and dVα are related by
dVF = e(n+1)ρ(x) dVα . Since sij = sji , s00 := sij y i y j = 0 and si i = aij sij = 0. Observe that ∂P m ∂(P y m ) = y + nP = (n+1)P, m ∂y ∂y m
∂Qm = α−1 s00 + αsmm = 0. ∂y m
Since α is Riemannian, we have ∂Gm ∂ = y m m ln σα . m ∂y ∂x Thus one obtains ∂Gm ∂ S= − y m m ln σF ∂y m ∂x ∂(P y m ) ∂Qm ∂ρ ∂ ∂Gm + + − (n+1)y m m − y m m ln σα = m ∂y ∂y m ∂y m ∂x ∂x e 00 = (n+1)(P − ρ0 ) = (n+1) − (s0 + ρ0 ) , 2F
(9–3)
where ρ0 := ρxi (x)y i .
Lemma 9.1 [Chen and Shen 2003a]. For a Randers metric F = α + β on an ndimensional manifold M , the following conditions are equivalent:
324
ZHONGMIN SHEN
(a) The Scurvature is isotropic, i .e., S =(n+1)cF for a scalar function c on M. (b) The Scurvature is almost isotropic, i .e., S = (n+1)(cF + η) for a scalar function c and a closed 1form η on M . (c) The Ecurvature is isotropic, i .e., E = 21 (n+1)cF −1 h for a scalar function c on M . (d) e00 = 2c(α2 − β 2 ) for a scalar function c on M . Proof. (a) =⇒ (b) and (b) =⇒ (c) are obvious. (c) =⇒ (d). First, S = (n + 1)(cF + η), where η is a 1form on M . By (9–3), (c) is equivalent to the following e00 = 2cF 2 + 2θF , where θ := s0 + ρ0 + η. This implies that e00 = 2c(α2 + β 2 ) + 2θβ, 0 = 4cβ + 2θ. Solving for θ in the second of these equations and plugging the result into the first, one obtains (d). (d) =⇒ (a). Plugging e00 = 2c(α2 − β 2 ) into (9–3) yields S = (n + 1) c(α − β) − (s0 + ρ0 ) . (9–4)
On the other hand, contracting eij = 2c(aij −bi bj ) with bj gives si +ρi +2cbi = 0. Thus s0 + ρ0 = −2cβ. Plugging this into (9–4) yields (a). Example 9.2. Let V = (A, B, C) be a vector field on a domain U ⊂ R 3 , where A = A(r, s, t), B = B(r, s, t) and C = C(r, s, t) are C ∞ functions on U with p V (x) = A(x)2 + B(x)2 + C(x)2 < 1, ∀x = (r, s, t) ∈ U. Let Φ := y be the standard Euclidean metric on R 3 . Define F = α + β by (2–15) for the pair (Φ, V ). α and β are given by p hV (x), yi2 + y2 (1 − V (x)2 ) hV (x), yi α= , β=− , 1 − V (x)2 1 − V (x)2 where y = (u, v, w) ∈ Tx U ∼ = R3 . One can easily verify that kβkx < 1 for x ∈ U . By a direct computation, one obtains B 2 (Ar −Bs ) + C 2 (Ar −Ct ) − Ar + H , 1−A2 −B 2 −C 2 A2 (Bs −Ar ) + C 2 (Bs −Ct ) − Bs + H , = 1−A2 −B 2 −C 2 A2 (Ct −Ar ) + B 2 (Ct −Bs ) −Ct + H = , 1−A2 −B 2 −C 2 = − 21 (As +Br ), e13 = − 12 (At +Cr ), e23 = ,− 12 (Bt +Cs ),
e11 = e22 e33 e12
where H := 2ABe12 + 2ACe13 + 2BCe23 . Here as usual we write Ar = ∂A/∂r, etc. On the other hand, aij − bi bj =
δij . 1 − A2 − B 2 − C 2
LANDSBERG, S AND RIEMANN CURVATURES
325
It is easy to verify that eij = 2c(aij − bi bj ) if and only if A, B, and C satisfy Ar = B s = C t ,
At + Cr = 0,
As + Br = 0,
Bt + Cs = 0.
In this case, c = − 21 Ar = − 21 Bs = − 21 Ct . By Lemma 9.1, we know that S = 4cF . If F = α + βp on an ndimensional manifold M is generated from the pair (Φ, V ), where Φ = φij y i y j is a Riemannian metric and V = V i (∂/∂xi ) is a vector field on M with φij (x)V i (x)v j (x) < 1 for any x ∈ M , then F has isotropic Scurvature, S = (n + 1)c(x)F , if and only if Vi;j + Vj;i = −4cφij , where Vi = φij V j and Vi;j are the covariant derivatives of V with respect to Φ. This observation is made by Xing [2003]. It also follows from [Bao and Robles 2003b], although it is not proved there directly.
10. Randers Metrics with Relatively Isotropic LCurvature We now study Randers metrics with relatively isotropic (mean) Landsberg curvature. From its definition, the mean Landsberg curvature is the mean value of the Landsberg curvature. Thus if a Finsler metric has isotropic Landsberg curvature, it must have isotropic mean Landsberg curvature. I don’t know whether the converse is true as well; no counterexample has been found. Nevertheless, for Randers metrics, “having isotropic mean Landsberg curvature” implies “having isotropic Landsberg curvature”. According to Lemma 3.2, the Cartan torsion is given by (3–7). Differentiating (3–7) along a geodesic and using (8–4) and (8–5), we obtain 1 (Ji hjk + Jj hik + Jk hij ). (10–1) Lijk = n+1 Here we have used the fact that the angular form hy is constant along geodesics. By (3–7) and (10–1), one can easily show that Ji + cF Ii = 0 if and only if Lijk + cF Cijk = 0. This proves the claim. Lemma 10.1 [Chen and Shen 2003a]. For a nonRiemannian Randers metric F = α + β on an ndimensional manifold M , these conditions are equivalent: (a) (b) (c) (d)
J + cF I = 0 (or L + cF C = 0). S = (n+1)cF and β is closed . E = 21 cF −1 h and β is closed . e00 = 2c(α2 − β 2 ) and β is closed .
Here c = c(x) is a scalar function on M .
326
ZHONGMIN SHEN
Proof. By (10–1), to compute Lijk , it suffices to compute Ji . First, the mean Cartan torsion is 1 (10–2) Ii = (n + 1)F −1 α−2 (α2 bi − βyi ), 2 where yi := aij y j . By a direct computation using (8–5), one obtains Ji = 41 (n + 1)F −2 α−2 (2α (ei0 α2 − yi e00 ) − 2β(si α2 − yi s0 ) + si0 (α2 + β 2 )
+α2 (ei0 β − bi e00 ) + β(ei0 α2 − yi e00 ) − 2(si α2 − yi s0 )(α2 + β 2 ) + 4si0 α2 β ). Using this and (10–2), one can easily prove the lemma.
Thus, for any Randers metric F = α + β, the Jcurvature vanishes if and only if e00 = 0 and dβ = 0. This is equivalent to bi;j = 0, in which case, the spray coefficients of F coincide with that of α. This observation leads to the following result, first established by the collective efforts found in [Matsumoto 1974; Hashiguchi and Ichijy¯ o 1975; Kikuchi 1979; Shibata et al. 1977]. Proposition 10.2. For a Randers metric F = α + β, the following conditions are equivalent: (a) (b) (c) (d)
F F F β
is a weakly Landsberg metric, J = 0. is a Landsberg metric, L = 0. is a Berwald metric. is parallel with respect to α.
Example 10.3. Consider the Randers metric F = α + β on Rn defined by α :=
p
(1 − ε2 )hx, yi2 + εy2 (1 + εx2 ) , 1 + εx2
√ β :=
1 − ε2 hx, yi , 1 + εx2
where ε is an arbitrary constant with 0 < ε ≤ 1. Since β is closed, s ij = 0 and si = 0. After computing bi;j , one obtains eij =
√ ε 1 − ε2 δij . (1 + εx2 )(ε + x2 )
On the other hand, aij − bi bj =
ε δij . Thus eij = 2c(aij − bi bj ) with 1 + εx2
√ 1 − ε2 c := . 2(ε + x2 ) By Lemma 10.1, F satisfies L + cF C = 0, S = (n + 1)cF , and E = 21 cF −1 h. See [Mo and Yang 2003] for a family of more general Randers metrics with nonconstant isotropic Scurvature.
LANDSBERG, S AND RIEMANN CURVATURES
327
11. Riemann Curvature The Riemann curvature is an important quantity in Finsler geometry. It was first introduced by Riemann for Riemannian metrics in 1854. Berwald [1926; 1928] extended it to Finsler metrics using the Berwald connection. His extension of the Riemann curvature is a milestone in Finsler geometry. Let (M, F ) be a Finsler manifold and let G = y i (∂/∂xi ) − 2Gi (∂/∂y i ) be the induced spray. For a vector y ∈ Tx M \ {0}, set Ri k := 2
2 i 2 i ∂Gi ∂Gj ∂Gi j ∂ G j ∂ G − y + 2G − . ∂xk ∂xj ∂y k ∂y j ∂y k ∂y j ∂y k
(11–1)
The local curvature functions Ri k and Rjk := gij Ri k satisfy Ri k y k = 0,
Rjk = Rkj .
(11–2)
Ry = Ri k (∂/∂xi ) ⊗ dxk : Tx M → Tx M is a welldefined linear map. We call the family R = {Ry  y ∈ T M \ {0}} the Riemann curvature. The Riemann curvature is actually defined for sprays, as shown in [Kosambi 1933; 1935]. When the Finsler metric is Riemannian, then Ri k (x, y) = Rji kl (x)y j y l , where R(u, v)w = Rji kl (x)w j ui v j (∂/∂xi )i denotes the Riemannian curvature tensor. Namely, Ry (u) = R(u, y)y. The geometric meaning of the Riemann curvature lies in the second variation of geodesics. Let σ(t), for a ≤ t ≤ b, be a geodesic in M . Take a geodesic variation H(t, s) of σ(t), that is, a family of curves σ s (t) := H(t, s), a ≤ t ≤ b, each of which is a geodesic, with σ0 = σ. Let J(t) :=
∂H (t, 0). ∂s
Then J(t) satisfies the Jacobi equation Dσ˙ Dσ˙ J(t) + Rσ(t) (J(t)) = 0, ˙
(11–3)
where Dσ˙ is defined in (4–6). See [Kosambi 1933; 1935]. There is another way to define the Riemann curvature. Any vector y ∈ T x M can be extended to a nonzero C ∞ geodesic field Y in an open neighborhood U of x; a geodesic field is one for which every integral curve is a geodesic. Define p Fˆ (z, v) := g Yz (v, v), v ∈ Tz U, z ∈ U.
ˆ = g Y be the inner product Then Fˆ = Fˆ (z, v) is a Riemannian metric on U . Let g ˆ be the Riemann curvature of Fˆ . It is wellknown in induced by Fˆ and let R Riemannian geometry that ˆ y (u) = 0, g ˆ y (u), v = g ˆ y (v) , ˆ R ˆ u, R R (11–4)
328
ZHONGMIN SHEN
where u, v ∈ Tx U . An important fact is ˆ y (u), Ry (u) = R
u ∈ Tx M.
(11–5)
ˆ x = g y . It See [Shen 2001b, Proposition 6.2.2] for a proof of (11–5). Note that g follows from (11–4) and (11–5) that Ry (y) = 0,
g y (Ry (u), v) = g y (u, Ry (v)),
(11–6)
where u, v ∈ Tx M . In local coordinates, this equation is just (11–2). See [Shen 2001b] for the application of (11–5) in comparison theorems in conjunction with the Scurvature. For a twodimensional subspace Π ⊂ Tx M and a nonzero vector y ∈ Π, define K(Π, y) :=
g y (Ry (u), u) , g y (y, y)g y (u, u) − g y (y, u)2
(11–7)
where u ∈ Π such that Π = span {y, u}. One can use (11–6) to show that K(Π, y) is independent of the choice of a vector u, but it is usually dependent on y. We p call K(Π, y) the flag curvature of the flag (Π, y). When F = gij (x)y i y j is a Riemannian metric, K(Π, y) = K(Π) is independent of y ∈ Π, in which case K(Π) is usually called the sectional curvature of the section Π ⊂ T x M . A Finsler metric F on a manifold M is said to be of scalar curvature K = K(x, y) if for any y ∈ Tx M \ {0} the flag curvature K(Π, y) = K(x, y) is independent of the tangent planes Π containing y. From the definition, the flag curvature is a scalar function K = K(x, y) if and only if in a standard local coordinate system, Ri k = KF 2 hik , (11–8) where hik := g ij hjk = g ij F Fyj yk . F is of constant flag curvature if this K is a constant. For a Riemannian metric, if the flag curvature K(Π, y) = K(x, y) is a scalar function on T M , then K(x, y) = K(x) is independent of y ∈ T x M and it is a constant when n ≥ 3 by the Schur Lemma. In the next section we show that any locally projectively flat Finsler metric is of scalar curvature. Such metrics are for us a rich source of Finsler metrics of scalar curvature. Classifying Finsler metrics of scalar curvature, in particular those of constant flag curvature, is one of the important problems in Finsler geometry. The local structures of projectively flat Finsler metrics of constant flag curvature were characterized in [Shen 2003b]. R. Bryant [1996; 1997; 2002] had earlier classified the global structures of projectively flat Finsler metrics of K = 1 on S n , and given some ideas for constructing non–projectively flat metrics of K = 1 on S n . Very recently, some non–projectively flat metrics of constant flag curvature have been explicitly constructed; see [Bao–Shen 2002; Bejancu–Farran 2002; Shen 2002; 2003a; 2003b; 2003c; Bao–Robles 2003], for example. These are all Randers metrics. Therefore the classification of Randers metrics of constant flag curvature is a natural problem, first tackled in [Yasuda and Shimada 1977; Matsumoto 1989]. These authors obtained conditions they believed were necessary
LANDSBERG, S AND RIEMANN CURVATURES
329
and sufficient for a Randers metric to be of constant flag curvature. Using their result strictly as inspiration, Bao and Shen [2002] constructed a family of Randers metrics on S3 with K = 1; these metrics do satisfy Yasuda and Shimada’s conditions. Later, however, examples were found of Randers metrics of constant flag curvature not satisfying those conditions [Shen 2002; 2003c], showing that the earlier characterization was incorrect. Shortly thereafter, Randers metrics of constant flag curvature were characterized in [Bao–Robles 2003] using a system of PDEs, a result also obtained in [Matsumoto and Shimada 2002] by a different method. This subsequently led to a corrected version of the Yasuda–Shimada theorem. Finally, using the characterization in [Bao–Robles 2003], and motivated by some constructions in [Shen 2002; 2003c], Bao, Robles and Shen have classified Randers metrics of constant flag curvature with the help of formula (2–17): p Theorem 11.1 [Bao et al. 2003]. Let Φ = φij y i y j be a Riemannian metric and let V = V i (∂/∂xi ) be a vector field on a manifold M with Φ(x, Vx ) < 1 for all x ∈ M . Let F be the Randers metric defined by (2–17). F is of constant flag curvature K = λ if and only if (a) there is a constant c such that V satisfies Vij + Vji = −4cφij , where Vi := φij V j , and ˜ = λ + c2 , (b) Φ has constant sectional curvature K where

denotes the covariant derivative with respect to Φ and c is a constant.
The equation Vij + Vji = −4cφij of part (a) is by itself always equivalent to S = (n + 1)cF , for c a scalar function on M [Xing 2003]. An analogue of Theorem 11.1 still holds for Randers metrics of isotropic Ricci curvature, i.e., Ric = (n − 1)λF 2 , where λ = λ(x) is a scalar function on M . See [Bao and Robles 2003b] in this volume. We have not extended the result above to Randers metrics of scalar curvature. Usually, the isotropic Scurvature condition simplifies the classification problem. It seems possible to classify Randers metrics of scalar curvature and isotropic Scurvature. The following example is our first attempt to understand Randers metrics of scalar curvature and isotropic Scurvature. Example 11.2. Let F = α + β be p the Randers metric defined in (2–20). Set ∆ := 1 − a2 x4 . We can write α = aij (x)y i y j and β = bi (x)y i , where x2 ai − 2ha, xixi x2 aj − 2ha, xixj δij aij = , + ∆ ∆2
bi = −
x2 ai − 2ha, xixi . ∆
330
ZHONGMIN SHEN
Using Maple for the computation, we obtain, with the notations of Section 9, 2ha, xiy2 = 2ha, xi(α2 − β 2 ), ∆ ha, yixj − hx, yiaj sj0 = 2 , ∆2 a2 x2 hx, yi + ha, xiha, vi . s0 = bi si0 = 2 ∆ By Lemma 9.1, we see that F has isotropic Scurvature: e00 =
S = (n+1)ha, xiF. By (9–1), the spray coefficients Gi = Gi (x, y) of F are Gi = Gi + P y i + αaij sj0 , where P = e00 /(2F ) − s0 = ha, xi(α − β) − s0 . Using the formulas for Gi and Ri k in (11–1), we can show that F is also of scalar curvature with flag curvature K=3
ha, yi + 3ha, xi2 − 2a2 x2 . F
12. Projectively Flat Metrics A Finsler metric F = F (x, y) on an open subset U ⊂ Rn is projectively flat if every geodesic σ(t) is straight in U , that is, if σ i (t) = xi + f (t)y i , where f (t) is a C ∞ function with f (0) = 0, f 0 (0) = 1 and x = (xi ), y = (y i ) are constant vectors. This is equivalent to Gi = P y i , where P = P (x, y) is positively yhomogeneous of degree one. P is called the projective factor. It is generally difficult to compute the Riemann curvature, but for locally projectively flat Finsler metrics, the formula is relatively simple. Consider a projectively flat Finsler metric F = F (x, y) on an open subset U ⊂ Rn . By definition, its spray coefficients are in the form G i = P y i . Plugging them into (11–1), one obtains Ri k = Ξδki + τk y i ,
(12–1)
where Ξ = P 2 − P xk y k ,
τk = 3(Pxk − P Pyk ) + Ξyk .
Using (11–6), one can show that τk = −ΞF −1 Fyk and Fy k i y . F Thus F is of scalar curvature with flag curvature
Ri k = Ξ δki −
K=
P 2 − P xk y k Ξ = . F2 F2
(12–2)
(12–3)
LANDSBERG, S AND RIEMANN CURVATURES
331
Using (7–4), one obtains S = (n + 1)P (x, y) − y m
∂ ln σF (x). ∂xm
(12–4)
By (12–2), one immediately obtains the following result. (See also [Szab´ o 1977] and [Matsumoto 1980] for related discussions.) Proposition 12.1 [Berwald 1929a; 1929b]. Every locally projectively flat Finsler metric is of scalar curvature. There is another way to characterize projectively flat Finsler metrics. Theorem 12.2 [Hamel 1903; Rapcs´ ak 1961]. Let F = F (x, y) be a Finsler metric on an open subset U ⊂ Rn . F is projectively flat if and only if F satisfies Fxk yl y k − Fxl = 0,
(12–5)
in which case, the spray coefficients are given by Gi = P y i with P = 12 Fxk y k /F . Proof. Let Gi = Gi (x, y) denote the spray coefficients of F in the standard coordinate system in T U ∼ = U × Rn . One can rewrite (4–2) as Gi = P y i + Q i ,
(12–6)
where Fx k y k , Qi = 12 F g il (Fxk yl y k − Fxl ). 2F Thus F is projectively flat if and only if there is a scalar function P˜ = P˜ (x, y) such that Gi = P˜ y i , i.e., P y i + Qi = P˜ y i . (12–7) P =
Observe that gij y j Qi = 21 F y l (Fxk yl y k − Fxl ) = 0. Assume that (12–7) holds. Contracting with yi := gij y j yields P = P˜ . Then Qi = 0 by (12–7). This implies (12–5).
Since equation (12–5) is linear, if F1 and F2 are projectively flat on an open subset U ⊂ Rn , so is their sum. If F = F (x, y) is projectively flat on U ⊂ R n , so is its reverse F¯ := F (x, −y). Thus the symmetrization F˜ := 12 F (x, y) + F (x, −y) is projectively flat. The Finsler metric F =αµ (x, y) in (2–5) satisfies (12–5), so it’s projectively flat.
Theorem 12.3. (Beltrami) A Riemannian metric F = F (x, y) on a manifold M is locally projectively flat if and only if it is locally isometric to the metric αµ in (2–5).
332
ZHONGMIN SHEN
Using the formula (9–1), one can easily prove the following: Theorem 12.4. A Randers metric F = α+β on a manifold is locally projectively flat if and only if α is locally projectively flat and β is closed . Besides projectively flat Randers metrics, we have the following examples. Example 12.5. Let φ = φ(y) be a Minkowski norm on Rn and let U be the strongly convex domain enclosed by the indicatrix of φ. Let Θ = Θ(x, y) be the Funk metric on U (Example 2.9). By (2–19), Θxk yl y k = (ΘΘyk )yl y k = 12 (Θ2 )yk yl y k = 21 [Θ2 ]yl = Θxl . Thus Θ is projectively flat with projective factor P =
ΘΘyk y k Θ xk y k = = 12 Θ. 2Θ 2Θ
By (12–3), the flag curvature is K=
Θ2 − 2Θxk y k Θ2 − 2Θ2 1 = =− . 2 2 4Θ 4Θ 4
Example 12.6 [Shen 2003b]. Let φ = φ(y) be a Minkowski norm on R n and let U be the strongly convex domain enclosed by the indicatrix of φ. Let Θ = Θ(x, y) be the Funk metric on U and define F := Θ(x, y) 1 + Θyk (x, y)xk . Since F (0, y) = Θ(0, y) = φ(y) is a Minkowski norm, by continuity, F is a Finsler metric for x nearby the origin. By (2–19), one can verify that Fx k y l y k = F x l ,
Fxk y k = 2ΘF.
Thus F is projectively flat with projective factor P = Θ(x, y). By (2–19) and (12–3), we obtain K=
Θ2 − ΘΘyk y k Θ 2 − Θ xk y k = = 0. F2 F2
Now we take a look at the Finsler metric F = Fε (x, y) defined in (2–9). Example 12.7. Let q p √ Ψ 12 ( Φ2 + (1 − ε2 )y4 + Φ) + (1 − ε2 )hx, yi2 + 1 − ε2 hx, yi F := , Ψ (12–8) where Φ := εy2 + x2 y2 − hx, yi2 , Ψ := 1 + 2εx2 + x4 .
LANDSBERG, S AND RIEMANN CURVATURES
333
First, one can verify that F = Fε (x, y) satisfies (12–5). Thus F is projectively flat with spray coefficients Gi = P y i , where P = 21 Fxk (x, y)y k /F (x, y). A Maple computation gives q p Ψ 21 ( Φ2 + (1 − ε2 )y4 + Φ) − (1 − ε2 )hx, yi2 − (ε + x2 )hx, yi P = . Ψ (12–9) Further, one can verify that P satisfies Px k y k = P 2 − F 2 . Thus K=
P 2 − (P 2 − F 2 ) P 2 − P xk y k = = 1. 2 F F2
That is, F has constant flag curvature K = 1. The projectively flat Finsler metrics constructed above are incomplete. They can be pulled back to Sn by (2–2) to form complete irreversible projectively flat Finsler metrics of constant flag curvature K = 1. See [Bryant 1996; 1997].
13. The Chern Connection and Some Identities The previous sections introduced several geometric quantities, such as the Cartan torsion, the Landsberg curvature, the Scurvature and the Riemann curvature. These quantities are not all independent. To reveal their relationships, we use the Chern connection to describe them as tensors on the slit tangent bundle, and use the exterior differentiation method to derive some important identities. Let M be an ndimensional manifold and T M its tangent bundle. As usual, a typical element of T M will be denoted by (x, y), with y ∈ T x M . The natural projection π : T M → M pulls back the tangent bundle T M over M to a vector bundle π ∗ T M over the slit tangent bundle T M0 . The fiber of π ∗ T M at each point (x, y) ∈ T M0 is a copy of Tx M . Thus we write a typical element of π ∗ T M as (x, y, v), where y ∈ Tx M \ {0} and v ∈ Tx M . Let ∂i(x,y) := x, y, (∂/∂xi )x . Then {∂i } is a local frame for π ∗ T M . Let (xi , y i ) be a standard local coordinate system in T M0 . Then HT ∗ M := span{dxi } is a welldefined subbundle of T ∗ (T M0 ). Let δy i := dy i − Nji dxj , where Nji := ∂Gi /∂y j . Then V T ∗ M := span {δy i } is a welldefined subbundle of T ∗ (T M0 ), so that T ∗ (T M0 ) = HT ∗ M ⊕ V T ∗ M . The Chern connection is a linear connection on π ∗ T M , locally expressed by DX = (dX i + X j ωj i ) ⊗ ∂i ,
X = X i ∂i ,
334
ZHONGMIN SHEN
where the set of 1forms {ωj i } is uniquely determined by dω i = ω j ∧ ωj i , dgij = gik ωj k + gkj ωi k + 2Cijk ω n+k ,
(13–1)
where gij := 21 [F 2 ]yi yj , Cijk := 14 [F 2 ]yi yj yk , ω i := dxi , and ω n+i := δy i . See [Bao and Chern 1993; Bao et al. 2000; Chern 1943; 1948; 1992]. Each 1form ωj i is horizontal: ωj i = Γijk dxk . The coefficients Γijk = Γijk (x, y) are called the Christoffel symbols. We have Nji = y k Γijk . Thus ω n+i = dy i + y j ωj i .
(13–2)
Ωi := dω n+i − ω n+j ∧ ωj i .
(13–3)
Put One can express Ω as i
Ωi =
1 i k R ω ∧ ω l − Li kl ω k ∧ ω n+l , 2 kl
where Ri kl =
i i ∂Nki ∂Nli s ∂Nk s ∂Nl − + N − N , l k ∂xk ∂xl ∂y s ∂y s
and Li kl := y j
∂Γijk ∂Nki = − Γikl . ∂y l ∂y l
Let Ri k be defined in (11–1) and Lijk be defined in (8–4). Then Ri k = Ri kl y l ,
Ri kl =
Put
1 ∂Ri k ∂Ri l − , 3 ∂y l ∂y k
Li kl = g ij Ljkl .
(13–4)
Ωji := dωj i − ωj k ∧ ωki . One can express Ωji as Ω ji =
1 i k R ω ∧ ω l + Pj ikl ω k ∧ ω n+l . 2 j kl
Differentiating (13–2) yields Ωi = y j Ωji . Thus Ri kl = y j Rji kl ,
Li kl = −y j Pj ikl .
There is a canonical way to define the covariant derivatives of a tensor on T M0 using the Chern connection. For the distortion τ on T M \ {0}, define τ m and τ·m by dτ = τi ω i + τ·i ω n+i . (13–5) It follows from (7–2) that τ·i =
∂τ = Ii . ∂y i
(13–6)
LANDSBERG, S AND RIEMANN CURVATURES
335
For the induced Riemannian tensor, g = gij ω i ⊗ ω j , define gijk and gij·k by dgij − gkj ωi k − gik ωj k = gijk ω k + gij·k ω n+k . It follows from (13–1) that gijk = 0,
gij·k = 2Cijk .
Similarly, one can define Cijkl at Iil . Equations (8–4) and (8–5) become Lijk = Cijkm y m ,
Ji = Iim y m .
(13–7)
Differentiating (13–3) yields the Bianchi identity dΩi = −Ωj ∧ ωj i + ω n+j ∧ Ωji .
(13–8)
It follows from (13–8) that Rji kl = Ri kl·j + Li kjl − Li ljk + Li lm Lmkj − Li km Lmlj .
(13–9)
We are going to find other relationships among curvature tensors. Differentiating (13–1) yields 0 = gik Ωjk + gkj Ωi k + 2(Cijkl ω l + Cijk·l ω n+l ) ∧ ω n+k + 2Cijk Ωk . It follows that Rjikl + Rijkl + 2Cijm Rmkl = 0, where Rjikl :=
gim Rjmkl ,
(13–10)
and
Pjikl + Pijkl + 2Cijlk − 2Cijm Lmkl = 0, where Pjikl := gim Pj mkl . Then (13–9) can be expressed by Rjikl = gim Rmkl·j + Likjl − Liljk + Lilm Lmkj − Likm Lmlj . Plugging the formulas for Rjikl and Rijkl into (13–10) yields Lijkl − Lijlk = − 21 (gim Rmkl·j + gjm Rmkl·i ) − Cijm Rmkl , Ikl − Ilk = −Rmkl·m − Im Rmkl .
(13–11)
The expression for Ri kl in (13–4) can be written as Ri kl = 13 (Ri k·l − Ri l·k ).
(13–12)
Lemma 13.1 [Mo 1999]. Lijk and Ri k are related by Cijkpq y p y q + Cijm Rmk = − 31 gim Rmk·j − 31 gjm Rmk·i − 16 gim Rmj·k − 16 gjm Rmi·k .
(13–13)
In particular , Ikpq y p y q + Im Rmk = − 13 (2Rmk·m + Rmm·k ).
(13–14)
336
ZHONGMIN SHEN
Proof. By (13–7), we have Lijkm y m = Cijkpq y p y q ,
Jkm y m = Ikpq y p y q .
Contracting the first line of (13–11) with y l yields (13–13), and contracting (13–13) with g ij yields (13–14). Here we have made use of (13–12). The equations above are crucial in the study of Finsler metrics of scalar curvature. Let F = F (x, y) be a Finsler metric of scalar curvature with flag curvature K = K(x, y). Then (11–8) holds. Plugging (11–8) into (13–13) and (13–14) yields Cijkpq y p y q + KF 2 Cijk = − 31 F 2 (K ·i hjk + K ·j hik + K ·k hij ), Ikpq y p y q + KF 2 Ik = − 31 (n + 1)F 2 K ·k .
(13–15)
Using the first of these equations, one shows that any compact Finsler manifold of negative constant flag curvature must be Riemannian [AkbarZadeh 1988]. It follows from (13–15) that for any Finsler metric F of scalar curvature with flag curvature K, the Matsumoto torsion satisfies Mijkpq y p y q + KF 2 Mijk = 0.
(13–16)
One can use (13–16) to show that any Landsberg metric of scalar curvature with K 6= 0 it is Riemannian, in dimension n ≥ 3 [Numata 1975]. See also Corollary 17.4. Using (13–16), one can easily prove this: Theorem 13.2 [Mo and Shen 2003]. Let (M, F ) be a compact Finsler manifold of dimension n ≥ 3. If F is of scalar curvature with negative flag curvature, F must be a Randers metric. Now we derive some important identities for the Scurvature. Differentiating (13–5) and using (13–3) and (13–6), one obtains 0 = d2 τ = (τkl ω l + τk·l ω n+l ) ∧ ω k + (Ikl ω l + Ik·l ω n+l ) ∧ ω k + Im Ωn+m . This yields the Ricci identities τkl =τlk + Ip Rpkl , τk·l =Ilk −
Ip Lpkl .
(13–17) (13–18)
From the definition (7–3), the Scurvature can be regarded as S = τm y m .
(13–19)
Contracting (13–17) with y k yields S ·k = (τm y m )·k = τm·k y m + τk = Ikm y m − Ip Lpmk y m + τk = Jk + τk , where we have made use of (13–17) and (13–19). We restate this equation as S ·k = τk + Jk .
(13–20)
LANDSBERG, S AND RIEMANN CURVATURES
337
Lemma 13.3 [Mo 2002; Mo and Shen 2003]. The Scurvature satisfies S ·km y m − S k = − 31 (2Rmk·m + Rmm·k ).
(13–21)
Proof. It follows from (13–20) that S ·kl = τkl + Jkl .
(13–22)
By (13–17) and (13–22), one obtains S ·km y m − S k = (S ·km − S ·mk )y m = (τkm − τmk )y m + (Jkm − Jmk )y m = Ip Rpkm y m + Jkm y m = Ip Rpk − Ip Rpk − 31 Im (Rmk·l − Rml·k ) = − 31 Im (Rmk·l − Rml·k ).
14. Nonpositively Curved Finsler Manifolds We now use some of the identities derived in the previous section to establish global rigidity theorems. First, consider the mean Cartan torsion. Let (M, F ) be an ndimensional Finsler manifold. The norm of the mean Cartan torsion I at a point x ∈ M is defined by q kIkx :=
sup
Ii (x, y)g ij (x, y)Ij (x, y).
06=y∈Tx M
It is known that if F = α + β is a Randers metric, then q p n+1 n+1 1 − 1 − kβk2x < √ . kIkx ≤ √ 2 2 The bound in dimension two was suggested by B. Lackey. See [Shen 2001b, Proposition 7.1.2] or [Ji and Shen 2002] for a proof. Below is our first global rigidity theorem. Theorem 14.1 [Shen 2003d]. Let (M, F ) be an ndimensional complete Finsler manifold with nonpositive flag curvature. Suppose that F has almost constant Scurvature S = (n + 1)(cF + η) (with c a constant and η a closed 1form) and bounded mean Cartan torsion supx∈M kIkx < ∞. Then J = 0 and R ◦ I = 0. Moreover F is Riemannian at points where the flag curvature is negative. Proof. It follows from (13–14) and (13–21) that Ikpq y p y q + Im Rmk = S ·km y m − S k .
(14–1)
Assume that the Scurvature is almost isotropic: S = (n + 1)(cF + η), where c = c(x) is a scalar function on M and η = ηi dxi is a closed 1form on M . Observe that ∂η ∂ηm m k − y = 0. η·km y m − ηk = (ηkm − ηmk )y m = ∂xm ∂xk
338
ZHONGMIN SHEN
Thus S ·km y m − S k = (n + 1)(cxm y m F·k − ck F + η·km y m − ηk ) = (n + 1)(cxm y m F·k − ck F ). In this case, (13–21) becomes 2Rmk·m + Rmm·k = −3(n + 1)(cxm y m F·k − ck F )
(14–2)
and (14–1) becomes Ikpq y p y q + Im Rmk = (n + 1)(cxm y m F·k − ck F ). By assumption, c is constant, so this last equation reduces Ikpq y p y q + Im Rmk = 0.
(14–3)
Let y ∈ Tx M be an arbitrary vector and let σ(t) be the geodesic with σ(0) = x and σ(0) ˙ = y. Since the Finsler metric is complete, one may assume that σ(t) is defined on (−∞, ∞). The mean Cartan torsion I and the mean Landsberg curvature J restricted to σ(t) are vector fields along σ(t): I(t) := I i (σ(t), σ(t)) ˙
∂ , ∂xi σ(t)
J(t) := J i (σ(t), σ(t)) ˙
It follows from (8–5) or (13–7) that
Dσ˙ I(t) = I im (σ(t), σ(t)) ˙ σ˙ m (t) It follows from (14–3) that
∂ . ∂xi σ(t)
∂ = J(t). ∂xi σ(t)
Dσ˙ Dσ˙ I(t) + Rσ(t) (I σ(t) ) = 0. ˙ ˙ Setting ϕ(t) := g σ(t) I(t), I(t) , ˙
we obtain
ϕ00 (t) = 2g σ(t) Dσ˙ Dσ˙ I(t), I(t) + 2g σ(t) Dσ˙ I(t), Dσ˙ I(t) ˙ ˙ = −2g σ(t) Rσ(t) (I(t)), I(t) + 2g σ(t) J (t), J (t) . ˙ ˙ ˙
(14–4)
By assumption, K ≤ 0. Thus
g σ(t) Rσ(t) (I(t)), I(t) ≤ 0. ˙ ˙
It follows from (14–4) that
ϕ00 (t) ≥ 0. Thus ϕ(t) is convex and nonnegative. Suppose that ϕ 0 (t0 ) 6= 0 for some t0 . By an elementary argument, limt→+∞ ϕ(t) = ∞ or limt→−∞ ϕ(t) = ∞. This
LANDSBERG, S AND RIEMANN CURVATURES
339
implies that the mean Cartan torsion is unbounded, which contradicts the assumption. Therefore, ϕ0 (t) = 0 and hence ϕ00 (t) = 0. Since each term in (14–4) is nonnegative, one concludes that Rσ(t) (I(t)), I(t) = 0, J (t) = 0. g σ(t) ˙ ˙ Setting t = 0 yields
g y Ry (I y ), I y = 0
(14–5)
and J y = 0. By (11–6), Ry (y) = 0 and Ry is selfadjoint with respect to g y , i.e., g y Ry (u), v = g y (u, Ry (v)), for u, v ∈ Tx M . Thus there is an orthonormal basis {ei }ni=1 with en = y such that Ry (ei ) = λi ei ,
i = 1, . . . , n,
with λn = 0. By assumption, the flag curvature is nonpositive. Then g y (Ry (ei ), ei ) = λi ≤ 0,
i = 1, . . . , n − 1.
Since I y is perpendicular to y with respect to g y , one can express it as I y = µ1 e1 + · · · + µn−1 en−1 . By (14–5), one obtains X n−1 0 = g y Ry (I y ), I y = µ2i λi . i=1
Since each term µ2i λi is nonpositive, one concludes that µi λi = 0, or yet Ry (I y ) =
n−1 X
µi λi = 0.
(14–6)
i=1
Now suppose that F has negative flag curvature at a point x ∈ M . Then λi < 0 for i = 1, . . . , n−1. By (14–6), one concludes that µ i = 0, i = 1, . . . , n−1, namely, I y = 0. By Deicke’s theorem [Deicke 1953], F is Riemannian. Corollary 14.2. Every complete Berwald manifold with negative flag curvature is Riemannian. Proof. For a Berwald metric F on a manifold M , the Minkowski spaces (Tx M, Fx ) are all linearly isometric (Proposition 5.2). Thus the Cartan torsion is bounded from above. Meanwhile, the Scurvature vanishes (Proposition 7.1). Thus F must be Riemannian. Example 14.3. Let (M , α ¯ ) and (M , α) be Riemannian manifolds and let F = F (x, y) be the product metric on M = M × M , defined in Example 5.1. We computed the spray coefficients of F in Example 5.1. Using (11–1), one obtains the Riemann tensor of F : Ra¯¯b = Ra¯¯b ,
Ra¯b = 0 = R ¯b , a
R
a b
a
= R b,
340
ZHONGMIN SHEN
¯ and a. Let where Ra¯¯b and R b are the coefficients of the Riemann tensors of a Rij := gik Rkj as usual, and define Ra¯¯b and Rab similarly. Using (5–3), one obtains Ra¯¯b = fs Ra¯¯b , Ra¯b = 0 = Ra¯b , Rab = ft Rab . a
For any vector v = v i (∂/∂xi )x ∈ Tx M , ¯ g y Ry (v), v = fs Ra¯¯b v a¯ v b + ft Rab v a v b .
Thus if α1 and α2 both have nonpositive sectional curvature, F has nonpositive flag curvature. Using (5–4), one can compute the mean Cartan torsion. First, observe that q p ∂ ∂ ln det(gjk ) = ln h [α1 ]2 , [α2 ]2 , Ii = i i ∂y ∂y where h = h(s, t) is defined in (5–5). One obtains Ia¯ =
hs y¯a¯ , h
Ia =
ht y¯a , h
¯
where y¯a¯ := g¯a¯¯b y b and y¯a := g¯ab y b . Since y¯a¯ Ra¯¯b = 0 and y¯a R
a b
= 0, one obtains
hs ht ¯ a g y Ry (I y ), I y = Ii Ri j I j = y¯a¯ Ra¯¯b I b + y¯a R b I b = 0. h h
Since Ry is selfadjoint and nonpositive definite with respect to g y , Ry (I y ) = 0. Therefore F satisfies the conditions and conclusions in Theorem 14.1. The next example shows that completeness in Theorem 14.1 cannot be replaced by positive completeness. Example 14.4. Let φ(y) be a Minkowski norm on Rn . Let Θ = Θ(x, y) be the Funk metric on U := {y ∈ Rn  φ(y) < 1} defined in (2–18). Let a ∈ Rn be an arbitrary constant vector. Let F := Θ(x, y) +
ha, yi , 1 + ha, xi
y ∈ TU ∼ = U × Rn .
Clearly, F is a Finsler metric near the origin. By (2–19), one sees that the spray coefficients of F are given by Gi = P y i , where P :=
1 ha, yi Θ(x, y) − . 2 1 + ha, xi
Using this and (12–3), one obtains
K=
2
1 ha, yi Θ2 + 2 1 + ha, xi 2 ha, yi Θ(x, y) + 1 + ha, xi
1 ha, yi Θ− 4 1 + ha, xi
−
2
1 =− . 4
Thus F has constant flag curvature K = − 14 . See also [Shen 2003b, Example 5.3].
LANDSBERG, S AND RIEMANN CURVATURES
341
Now we compute the Scurvature of F . A direct computation gives ∂Gm = (n + 1)P. ∂y m Let dV = σF (x) dx1 · · · dxn be the Finsler volume form on M . From (12–4), we obtain ha, yi ∂ n+1 F (x, y) − (n + 1) − y m m ln σF (x) S= 2 1 + ha, xi ∂x = (n + 1) 21 F (x, y) + dϕx (y) , where ϕ(x) := − ln (1 + ha, xi)σF (x)1/(n+1) . Thus E = 14 (n + 1)F −1 h,
where hy = hij (x, y) dxi ⊗ dxj is given by hij = F (x, y)Fyi yj (x, y). When φ(y) = y is the standard Euclidean norm, U = B n is the standard unit ball in Rn and q y2 − x2 y2 − hx, yi2 Θ= . 1 − x2 Thus
q
y2 − x2 y2 − hx, yi2
ha, yi + . 1 − x2 1 + ha, xi Assume that a < 1. It is easy to verify that F is a Randers metric defined on the whole Bn , with constant Scurvature S = 12 (n + 1)F (x, y). One can show that F is positively complete on Bn , so that every geodesic defined on an interval (λ, µ) can be extended to a geodesic defined on (λ, +∞). F =
15. Flag Curvature and Isotropic SCurvature It is a difficult task to classify Finsler metrics of scalar curvature. All known Randers metrics of scalar curvature have isotropic Scurvature. Thus it is a natural idea to investigate Finsler metrics of scalar curvature which also have isotropic Scurvature. Proposition 15.1 [Chen et al. 2003]. Let (M, F ) be an ndimensional Finsler manifold of scalar curvature with flag curvature K = K(x, y). Suppose that the Scurvature is almost isotropic, S = (n + 1)(cF + η), where c = c(x) is a scalar function on M and η = ηi (x)y i is a closed 1form. Then there is a scalar function σ = σ(x) on M such that the flag curvature equals c xm y m K=3 + σ. (15–1) F
342
ZHONGMIN SHEN
Proof. By assumption, the flag curvature K = K(x, y) is a scalar function on T M0 . Thus (11–8) holds. Plugging (11–8) into (14–2) yields cxm y m F·k − cxk F = − 31 K yk F 2 . Rewriting (15–2) as
1
K−
3 one concludes that the quantity
c xm y m F
yk
(15–2)
= 0,
3cxm y m F is a scalar function on M . This proves the proposition. σ := K −
Corollary 15.2 [Mo 2002]. Let F be an ndimensional Finsler metric of scalar curvature. If F has almost constant Scurvature, the flag curvature is a scalar function on M . From the definition of flag curvature, one can see that every twodimensional Finsler metric is of scalar curvature. One immediately obtains the following: Corollary 15.3. Let F be a twodimensional Finsler metric with almost isotropic Scurvature. Then the flag curvature is in the form (15–1). Let F = F (x, y) be a twodimensional Berwald metric on a surface M . It follows from Corollaries 15.3 and 15.2 that the Gauss curvature K = K(x) is a scalar function of x ∈ M . Since F is a Berwald metric, the Gi = 21 Γijk (x)y j y k are quadratic in y = y i (∂/∂xi )x ∈ Tx M . By (11–1), the Riemann curvature, Ri k = Ri k (x, y), are quadratic in y. This implies that the Ricci scalar Ric = R mm (x, y) is quadratic in y. Suppose that K(x0 ) 6= 0 at some point x0 ∈ M . Then F (x0 , y)2 =
Ric(x0 , y) K(x0 )
is quadratic in y ∈ Tx0 M . Namely, Fx0 = F Tx0 M is Euclidean at x0 . By Proposition 5.2, all tangent spaces (Tx M, Fx ) are linearly isometric to each other. One concludes that Fx is Euclidean for any x ∈ M and F is Riemannian. Now we suppose that K ≡ 0. Since F is Berwaldian, F must be locally Minkowskian. See [Szab´ o 1981] for a different argument.
16. Projectively Flat Metrics with Isotropic SCurvature Recall that a Finsler metric F on a manifold M is locally projectively flat if at any point x ∈ M , there is a local coordinate system (x i ) in M such that every geodesic σ(t) is straight, i.e., σ i (t) = f (t)ai + bi . This is equivalent to saying that in the standard local coordinate system (x i , y i ), the spray coefficients Gi are in the form Gi = P y i with P = Fxk y k /(2F ). It is wellknown that any locally projectively flat Finsler metric F is of scalar curvature, and its flag curvature
LANDSBERG, S AND RIEMANN CURVATURES
343
equals K = (P 2 − Pxk y k )/F 2 (see Proposition 12.1). Our goal is to characterize those with almost isotropic Scurvature. First, by Beltrami’s theorem and the Cartan classification theorem, a Riemannian metric is locally projectively flat if and only if it is of constant sectional curvature. Every Riemannian metric of constant sectional curvature µ is locally isometric to the metric αµ on a ball in Rn , defined in (2–5). A Randers metric F = α + β is locally projectively flat if and only if α is locally projectively flat (hence of constant sectional curvature) and β is closed. This follows directly from a result in [B´ acs´ o and Matsumoto 1997] and the Beltrami theorem on projectively flat Riemannian metrics. If in addition, the Scurvature is almost isotropic, then β can be determined explicitly. Proposition 16.1 [Chen et al. 2003]. Let F = α + β be a locally projectively flat Randers metric on an ndimensional manifold M . Suppose that F has almost isotropic Scurvature, S = (n + 1)(cF + η), where c is a scalar function on M and η is a closed 1form on M . Then: (a) α is locally isometric to αµ and β is a closed 1form satisfying (µ + 4c2 )β = −cxk y k . (b) The flag curvature is given by K=
α−β µ 3cxk y k + 3c2 + µ = 34 (µ + 4c2 ) + . α+β α+β 4
(16–1)
(c) If µ + 4c2 ≡ 0, then c is a constant and the flag curvature equals −c2 . In this case, F = α + β is either locally Minkowskian (c = 0) or , up to scaling (c = ± 12 ), locally isometric to the generalized Funk metric Θa = Θa (x, y) of ¯ a = Θa (x, −y). (2–7) or its reverse Θ 2 (d) If µ + 4c 6= 0, then F = α + β must be locally given by α = αµ (x, y),
β=−
2cxk (x)y k µ + 4c2
where c := cµ (x) is given by r µ , (λ + ha, xi) ±(1 + µx2 ) − (λ + ha, xi)2 cµ = ±1 , p 2 λ + 2ha, xi + x2
(16–2)
µ 6= 0, µ = 0,
for a ∈ Rn a constant vector and λ ∈ R a constant number . p Proof. Let αµ = aij (x)y i y j and β = bi (x)y i . We may assume that α = αµ in a local coordinate system aij =
δij µxi xj − . 2 1 + µx (1 + µx2 )2
344
ZHONGMIN SHEN
The Christoffel symbols of α are given by Γijk = −µ
xj δki + xk δji . 1 + µx2
Thus Gi = −
µhx, yi i y. 1 + µx2
The spray coefficients of F are given by Gi = Gi +P y i +Qi , where P = e00 /(2F ) and Qi = αsi 0 are given by (9–1) and (9–2). Since β is closed, sij := 21 (bi;j + bj;i ) = 0 and si := bj sji = 0. Thus Qi = 0. By assumption, S = (n + 1)(cF + η) and Lemma 9.1, e00 = βk y k = 2c(α2 − β 2 ).
(16–3)
Thus P = e00 /(2F ) − s0 = c(α − β) and βxk y k = βk y k + 2Gk βyk = 2c(α2 − β 2 ) −
2µhx, yiβ . 1 + µx2
µhx, yi + c(α − β). By (12–3), we obtain Then Gi = P˜ y i , where P˜ = − 1 + µx2 KF 2 = P˜ 2 − P˜xk y k = µα2 + c2 (3α + β)(α − β) − cxk y k (α − β). On the other hand, by Theorem 15.1, the flag curvature is in the following form K=
3cxk y k + σ, α+β
where σ = σ(x) is a scalar function on M . It follows from the last two displayed equations that 2 2cxk y k + (σ + c2 )β α + 2cxk y k + (σ + c2 )β β + (σ − 3c2 − µ)α2 = 0. This gives
2cxk y k + (σ + c2 )β = 0,
σ − 3c2 − µ = 0.
Solving the second of these equations for σ and substituting into the first we get (µ + 4c2 )β = −2cxk y k .
(16–4)
To prove part (c) of the Proposition, suppose that µ + 4c 2 ≡ 0. Then c is constant. It follows from (16–1) that K = 3c2 + µ = −c2 . The local structure of F can be easily determined [Shen 2003a]. Now suppose instead that µ + 4c2 6= 0 on an open subset U ⊂ M . By (16–4), β=−
2cxk y k . µ + 4c2
(16–5)
LANDSBERG, S AND RIEMANN CURVATURES
345
Note that β is exact. It follows from (16–3) and (16–5) that c xi xj +
µ(xi cxj + xj cxi ) 1 + µx2 = −c(µ + 4c2 )
δij µxi xj − 1 + µx2 (1 + µx2 )2
We are going to solve for c. Let p 2c 1 + µx2 , µ 6= 0, p 2) ±(µ + 4c f := 1, µ = 0, c2
+
12ccxi cxj . µ + 4c2
(16–6)
where the sign is chosen so that the radicand ±(µ + 4c 2 ) > 0. Then (16–6) reduces to 0, µ 6= 0, f xi xj = 8δij , µ = 0. We obtain f=
λ + ha, xi, µ 6= 0, 4 λ + 2ha, xi + x2 , µ = 0,
where a ∈ Rn is a constant vector and λ is a constant. This gives part (d).
By Proposition 16.1, one immediately obtains: Corollary 16.2. Let F = α + β be a locally projectively flat Randers metric on an ndimensional manifold M . Suppose that F has almost constant Scurvature S = (n + 1)(cF + η), where c is a constant. Then F is locally Minkowskian, or Riemannian with constant curvature, or up to a scaling, locally isometric to the generalized Funk metric in (2–7). Proof. Let µ be the constant sectional curvature of α. If µ + 4c 2 = 0, by Proposition 16.1(c), F = α + β is either locally Minkowskian or, up to a scaling, locally isometric to the generalized Funk metric in (2–7). If µ + 4c 2 6= 0 instead, F = α + β is given by (16–2). Since cxk = 0, we get β = 0 and F = α is a Riemannian metric. Proposition 16.1 completely classifies projectively flat Randers metrics of almost isotropic Scurvature. If a Randers metric has almost isotropic Scurvature, its the Ecurvature is isotropic. By Lemma 9.1, the Scurvature is isotropic. Thus a Randers metric is of almost isotropic Scurvature if and only if it is of isotropic Scurvature. This is not true for general Finsler metrics: if Θ(x, y) is the Funk metric on a strongly convex domain U ⊂ Rn , the Finsler metric F = Θ(x, y) +
ha, yi , 1 + ha, xi
y ∈ Tx U ∼ = Rn ,
346
ZHONGMIN SHEN
is projectively flat with almost isotropic Scurvature, according to Example 14.4. Thus it has isotropic Ecurvature. However, this F is of isotropic Scurvature only for certain U ’s such as the standard unit ball. A natural problem is whether there are other types of projectively flat Finsler metrics of almost isotropic Scurvature. Here is the answer: Proposition 16.3 [Chen and Shen 2003b]. Let F = F (x, y) be a projectively flat Finsler metric on a simply connected open subset U ⊂ Rn . Suppose that F has almost isotropic Scurvature, S = (n + 1)c(F + η),
(16–7)
where c is a scalar function on M and η is a closed 1form on U . (a) If K is not of the form −c2 +cxm y m /F at every point x ∈ U , then F = α+β ¯ =µ is a Randers metric on U . Further , α is of constant sectional curvature K 2 with µ + 4c 6= 0 and α and β are as in Proposition 16.1(c). (b) If K ≡ −c2 + cxm y m /F on U , then c is a constant, and either F is locally Minkowskian (c = 0) or there exist a Funk metric Θ and a constant vector a ∈ Rn such that F has the form F = where Ψ = Θ(x, y) if c =
1 2
n
ha, yi 1 , Ψ+ 2c 1 + ha, xi
and Ψ = −Θ(x, −y) if c = − 21 .
Proof. Since F is projectively flat, the spray coefficients are given by G i = P y i , where F k yk . P := x 2F Thus the Scurvature is given by (12–4) and the flag curvature of F is given by (12–3). By assumption, S is of the form (16–7). Since η is closed on U , it can be written as η(x, y) = dhx (y), where h = h(x) is a scalar function on U . Thus P = cF + dϕx ,
(16–8)
where ϕ(x) := h(x) + (ln σF (x))/(n + 1). It follows from the last two displayed equations that Fxi y i = 2F P = 2F (cF + ϕxi y i ). Using this together with (16–8) and (12–3), one obtains (cF + ϕxi y i )2 − (cxi y i F + cFxi y i + ϕxi xj y i y j ) F2 2 2 m −c F − cxm y F + (ϕxi ϕxj − ϕxi xj )y i y j . = F2
K=
(16–9)
LANDSBERG, S AND RIEMANN CURVATURES
347
On the other hand, since F is of scalar curvature, by Proposition 15.1, the flag curvature of F is given by (15–1). Comparing (16–9) with (15–1) yields (σ + c2 )F 2 + 4cxm y m F + (ϕxi xj − ϕxi ϕxj )y i y j = 0.
(16–10)
Assume that K 6= −c2 + cxm y m /F at every point x ∈ U . Then, by (15–1), for any x ∈ U , there is a nonzero vector y ∈ Tx U such that σ + c2 +
2cxm y m 6= 0. F
We claim that σ + c2 6= 0 on U . If not, there is a point x0 ∈ U such that σ(x0 ) + c(x0 )2 = 0. The inequality above implies that dc 6= 0 at x0 . Then (16–10) at x0 reduces (16–11) 4cxm (x0 )y m F (x0 , y) + ϕxi xj (x0 ) − ϕxi (x0 )ϕxj (x0 ) y i y j = 0. Differentiating with respect to y i , then restricting to the hyperplane V := {y  cxm (x0 )y m = 0}, one obtains 4cxi (x0 )F (x0 , y) + ϕxi xj (x0 ) − ϕxi (x0 )ϕxj (x0 ) y j = 0.
In other words, F (x0 , y) is a homogeneous linear function of y ∈ V . This is impossible, because F (x0 , y) is always positive for y ∈ V \ {0}. Now we may assume that σ + c2 6= 0 on U . One can solve the quadratic equation (16–10) for F , p (σ + c2 )(ϕxi xj − ϕxi ϕxj )y i y j + 4(cxm y m )2 − 2cxm y m . F = σ + c2 p That is, F is expressed in the form F = α + β, where α = aij y i y j and β = bi y i are given by aij =
(σ + c2 )(ϕxi xj − ϕxi ϕxj ) + 4cxi cxj , (σ + c2 )2
bi = −
2cxi . σ + c2
Since F is a Randers metric, by Lemma 9.1, one concludes that S is isotropic, i.e., η = 0 and S = (n + 1)cF. ¯ = µ and β Since F is projectively flat, α is of constant sectional curvature K is closed. Moreover, by Proposition 16.1, the flag curvature is given by (16–1). Note that σ + c2 6= 0 is equivalent to the inequality µ + 4c2 6= 0. By Proposition 16.1(d), F is given by (16–2). We now assume that K ≡ −c2 + cxi y i /F . It follows from (15–1) that σ + c2 +
2cxm y m ≡ 0. F
348
ZHONGMIN SHEN
Suppose that cxm (x0 )y m 6= 0 at some point x0 . From the preceding identity, one sees that σ(x0 ) + c(x0 )2 6= 0. Thus F (x0 , y) = −
2cxm (x0 )y m σ(x0 ) + c(x0 )2
is a linear function. This is impossible. One concludes that c xm y m = 0 on U , and hence c is a constant and σ(x) = −c2 is a constant too. In this case, the flag curvature is given by K = −c2 . Equation (16–10) reduces to ϕxi xj − ϕxi ϕxj = 0, which is easily solved to yield ϕ = − ln 1 + ha, xi + C,
where a ∈ Rn is a constant vector and C is a constant. Assume that c = 0. Then K = −c2 = 0. It follows from (16–8) that the projective factor P = dϕx is a 1form, hence the spray coefficients Gi = P y i are quadratic in y ∈ Tx U . By definition, F is a Berwald metric, and every Berwald metric with K = 0 is locally Minkowskian (see [Bao et al. 2000] for a proof). Assume that c 6= 0. By (16–8), P = cF +dϕ. With Ψ := P +cF = 2cF +dϕ x , we have 1 1 ha, yi F = (Ψ(x, y) − dϕx ) = Ψ(x, y) + . 2c 2c 1 + ha, xi Since F is projectively flat and P is the projective factor, Fxk = (P F )yk ,
P x k = P P y k + c 2 F Fy k .
These equations imply that Ψxi = ΨΨyi . Let Θ := Ψ(x, y) if c > 0 and Θ := −Ψ(x, −y) if c < 0. Then Θ is a Funk metric and F is of the form stated in part (b) of the theorem.
17. Flag Curvature and Relatively Isotropic LCurvature Although the relatively isotropic Jcurvature condition is stronger than the isotropic Scurvature condition for Randers metrics (Lemma 10.1), it seems that there is no direct relationship between these two conditions. Nevertheless, for Finsler metrics of scalar curvature, the relatively isotropic Jcurvature condition also implies that the flag curvature takes a special form in certain cases. Proposition 17.1 [Chen et al. 2003]. Let F be an ndimensional Finsler manifold of scalar curvature and of relatively constant Jcurvature, so that J + cF I = 0, for some constant c. Then K = −c2 + σe−3τ /(n+1) ,
(17–1)
LANDSBERG, S AND RIEMANN CURVATURES
349
where τ = τ (x, y) is the distortion and σ = σ(x) is a scalar function on M . Proof. By assumption, Jk = −cF Ik . Using (13–6) and (13–7), one obtains Ikpq y p y q = Jkm y m = −cF Ikm y m = c2 F 2 τ·k . Plugging this into the second line of (13–15) yields 1 3 (n
+ 1)K ·k + (K + c2 )τ·k = 0.
This implies that (K + c2 )(n+1)/3 eτ
·k
= (K + c2 )(n−2)/3 eτ
1 3 (n
+ 1)K ·k + Kτ·k = 0.
Thus the function (K + c2 )(n+1)/3 eτ is independent of y ∈ Tx M .
Proposition 17.1 in the particular case c = 0 was essentially achieved in the proof of [Matsumoto 1972a, Proposition 26.2], where it is assumed that F is a Landsberg metric, but what is needed is merely that J = 0. Since the notion of distortion is not introduced in [Matsumoto 1972a], the result is stated in a local coordinate system. Corollary 17.2. Let F be a Finsler metric on a manifold M . Suppose that F has isotropic flag curvature not equal to −c2 and that F has relatively constant Jcurvature. Then F is Riemannian. Proof. By Proposition 17.1, K(x) = −c2 + σ(x)e−3τ /(n+1) . Since K(x) 6= −c2 , one concludes that σ(x) 6= 0, so τ = τ (x) is independent of y ∈ Tx M . It follows from (7–2) that Ii = τyi = 0. Thus F is Riemannian by Deicke’s theorem [Deicke 1953]. Proposition 17.3. Let F be a Finsler metric of scalar curvature on an ndimensional manifold . Suppose that F has relatively isotropic Lcurvature, so L + cF C = 0,
(17–2)
where c is a scalar function on M . (a) If c is constant, then K = −c2 + σe−3τ /(n+1) , where σ is a scalar function on M . (b) If n ≥ 3 and K 6= −c2 + cxm y m /F for almost all y ∈ Tx M \ {0} at any point x in an open domain U of M , then F = α + β is a Randers metric in U .
350
ZHONGMIN SHEN
Proof. If F has relatively isotropic Lcurvature, (17–1) holds by taking the average of (17–2) on both sides. Statement (a) then follows from Proposition 17.1. Now we assume that K 6= −c2 + cxm (x)y m /F for almost all y ∈ Tx M \ {0} at any point x in an open domain U ⊂ M . By assumption, Lijk = −cF Cijk , one obtains c xm y m F 2 Cijk . Cijkpq y p y q = −cxm y m F Cijk − cF Lijk = c2 − F Since Jk = −cF Ik by (17–1), we have
Ikpq y p y q = −cxm y m F Ik − cF Jk = c2 − By the formula for Mijk in (3–2), one obtains
Mijkpq y p y q = c2 −
c xm y m F 2 Ik . F
c xm y m F 2 Mijk . F
Since F is of scalar curvature, equation (13–16) holds. One obtains
K + c2 −
c xm y m F 2 Mijk = 0. F
It follows that Mijk = 0, so the Matsumoto torsion vanishes. By Proposition 3.3, F = α + β is a Randers metric on U . Proposition 17.3 was proved by H. Izumi [1976; 1977; 1982], The particular case c = 0 is proved by S. Numata [1975]. Corollary 17.4 [Numata 1975]. Let F be a Finsler metric of scalar curvature on an ndimensional manifold , with n ≥ 3. Suppose that L = 0 and K 6= 0. Then F is Riemannian. Proof. By Proposition 17.3, F = α + β is a Randers metric with L = 0. By Lemma 10.1, S = 0 and β is closed. By Proposition 15.1, one concludes that K = σ(x) is a scalar function on M . It follows from (13–14) that 0 = −F 2 σ(x)Ik . By assumption, K = σ(x) 6= 0. Thus Ik = 0 and F is Riemannian by Deicke’s theorem. We may ask again: is there a nonBerwaldian Finsler metric satisfying K = 0 and L = 0 (or J = 0)? If such a metric exists, it cannot be locally projectively flat and it cannot be a Randers metric. (Why?) Example 17.5. Let F = α + β be the Randers metric on Rn defined by
Note that
hx, yi F := y + p , 1 + x2 kβk2 =
y ∈ T x Rn ∼ = Rn .
x2 < 1. 1 + x2
LANDSBERG, S AND RIEMANN CURVATURES
351
F is indeed a Randers metric on the whole of Rn . One can verify that F satisfies (12–5). Thus it is a projectively flat Randers metric on R n . Further, the spray coefficients Gi = P y i are given by
hx, yi , P = c y − p 1 + x2 p p p where c = 1/ 2 1 + x2 . Let ρ := ln 1 − kβk2 = − ln 1 + x2 . By (9–3), one obtains S = (n+1)(P −ρ0 ) = (n+1)cF . Since β is closed, this is equivalent, by Proposition 10.1, to the identity L + cF C = 0. Since F is projectively flat, it is of scalar curvature. Further computation yields the flag curvature: p y 1 + x2 − hx, yi 3 P 2 − P xk y k = · p . K= F2 4(1 + x2 ) y 1 + x2 + hx, yi Note that K 6= −c2 + cxk (x)y k /F (x, y) and that F is a Randers metric. This matches the conclusion in Proposition 17.3(b).
The Randers metric in Example 17.5 is locally projectively flat. There are non– projectively flat Randers metrics of scalar curvature and isotropic Scurvature; see Example 11.2. This example is a Randers metric generated by a special vector field on the Euclidean space by (2–15). In fact, we can determine all vector fields V on a Riemannian space form (M, αµ ) of constant curvature µ such that the generated Randers metric F = α + β by (αµ , V ) is of scalar curvature and isotropic Scurvature. This work will appear elsewhere.
References [AkbarZadeh 1988] H. AkbarZadeh, “Sur les espaces de Finsler a ´ courbures sectionnelles constantes”, Acad. Roy. Belg. Bull. Cl. Sci. 74 (1988), 281–322. [Antonelli et al. 1993] P. Antonelli, R. Ingarden and M. Matsumoto, The theory of sprays and Finsler spaces with applications in physics and biology, Kluwer, Dordrecht, 1993. [Auslander 1955] L. Auslander, “On curvature in Finsler geometry”, Trans. Amer. Math. Soc. 79 (1955), 378–388. [B´ acs´ o and Matsumoto 1997] S. B´ acs´ o and M. Matsumoto, “On Finsler spaces of Douglas type: a generalization of the notion of Berwald space”, Publ. Math. Debrecen, 51 (1997), 385–406. [Bao and Chern 1993] D. Bao and S. S. Chern, “On a notable connection in Finsler geometry”, Houston J. Math. 19:1 (1993), 135–180. [Bao et al. 2000] D. Bao, S. S. Chern and Z. Shen, An Introduction to Riemann–Finsler geometry, Springer, New York, 2000. [Bao et al. 2003] D. Bao, C. Robles and Z. Shen, Zermelo navigation on Riemannian manifolds, preprint, 2003.
352
ZHONGMIN SHEN
[Bao–Robles 2003] D. Bao and C. Robles, “On Randers spaces of constant flag curvature”, Rep. on Math. Phys. 51 (2003), 9–42. [Bao and Robles 2003b] D. Bao and C. Robles, Ricci and flag curvatures in Finsler geometry, pp. 199–261 in A sampler of Riemann–Finsler Geometry, edited by D. Bao et al., Math. Sci. Res. Inst. Publ. 50, Cambridge Univ. Press, Cambridge, 2004. [Bao–Shen 2002] D. Bao and Z. Shen, “Finsler metrics of constant positive curvature on the Lie group S 3 ”, J. London Math. Soc. 66 (2002), 453–467. [Bejancu–Farran 2002] A. Bejancu and H. Farran, “Finsler metrics of positive constant flag curvature on Sasakian space forms”, Hokkaido Math. J. 31:2 (2002), 459–468. [Bejancu and Farran 2003] A. Bejancu and H. R. Farran, “Randers manifolds of positive constant flag curvature”, Int. J. Math. Sci. 18 (2003), 1155–1165. [Berwald 1926] L. Berwald, “Untersuchung der Kr¨ ummung allgemeiner metrischer R¨ aume auf Grund des in ihnen herrschenden Parallelismus”, Math. Z. 25 (1926), 40–73. [Berwald 1928] L. Berwald, “Parallel¨ ubertragung in allgemeinen R¨ aumen”, Atti Congr. Intern. Mat. Bologna 4 (1928), 263–270. ¨ [Berwald 1929a] L. Berwald, “Uber eine characteristic Eigenschaft der allgemeinen R¨ aume konstanter Kr¨ ummung mit gradlinigen Extremalen”, Monatsh. Math. Phys. 36 (1929), 315–330. ¨ [Berwald 1929b] L. Berwald, “Uber die ndimensionalen Geometrien konstanter Kr¨ ummung, in denen die Geraden die k¨ urzesten sind”, Math. Z. 30 (1929), 449–469. [Bryant 1996] R. Bryant, “Finsler structures on the 2sphere satisfying K = 1”, pp. 27– 42 in Finsler Geometry (Seattle, 1995), edited by D. Bao et al., Contemporary Mathematics 196, Amer. Math. Soc., Providence, 1996. [Bryant 1997] R. Bryant, “Projectively flat Finsler 2spheres of constant curvature”, Selecta Math. (N.S.) 3 (1997), 161–203. [Bryant 2002] R. Bryant, “Some remarks on Finsler manifolds with constant flag curvature”, Houston J. Math. 28:2 (2002), 221–262. [Busemann 1947] H. Busemann, “Intrinsic area”, Ann. of Math. 48 (1947), 234–267. [Cartan 1934] E. Cartan, Les espaces de Finsler, Actualit´es scientifiques et industrielles 79, Hermann, Paris, 1934. [Chen et al. 2003] X. Chen, X. Mo and Z. Shen, “On the flag curvature of Finsler metrics of scalar curvature”, J. London Math. Soc. (2) 68:3 (2003), 762–780. [Chen and Shen 2003a] X. Chen and Z. Shen, “Randers metrics with special curvature properties”, Osaka J. Math. 40 (2003), 87–101. [Chen and Shen 2003b] X. Chen and Z. Shen, “Projectively flat Finsler metrics with almost isotropic Scurvature”, preprint, 2003. [Chern 1943] S. S. Chern, “On the Euclidean connections in a Finsler space”, Proc. National Acad. Soc., 29 (1943), 33–37; reprinted as pp. 107–111 in Selected Papers, v. 2, Springer, New York, 1989. [Chern 1948] S. S. Chern, “Local equivalence and Euclidean connections in Finsler spaces”, Science Reports Nat. Tsing Hua Univ. 5 (1948), 95–121.
LANDSBERG, S AND RIEMANN CURVATURES
353
[Chern 1992] S. S. Chern, “On Finsler geometry”, C. R. Acad. Sc. Paris 314 (1992), 757–761. ¨ [Deicke 1953] A. Deicke, “Uber die FinslerR¨ aume mit Ai = 0”, Arch. Math. 4 (1953), 45–51. [Foulon 2002] P. Foulon, “Curvature and global rigidity in Finsler geometry”, Houston J. Math. 28 (2002), 263–292. ¨ [Funk 1929] P. Funk, “Uber Geometrien bei denen die Geraden die K¨ urzesten sind”, Math. Ann. 101 (1929), 226–237. ¨ [Funk 1936] P. Funk, “Uber zweidimensionale Finslersche R¨ aume, insbesondere u ¨ber solche mit geradlinigen Extremalen und positiver konstanter Kr¨ ummung”, Math. Z. 40 (1936), 86–93. ¨ [Hamel 1903] G. Hamel, “Uber die Geometrien in denen die Geraden die K¨ urzesten sind”, Math. Ann. 57 (1903), 231–264. [Hashiguchi and Ichijy¯ o 1975] M. Hashiguchi and Y. Ichijy¯ o, “On some special (α, β)metrics”, Rep. Fac. Sci. Kagoshima Univ. 8 (1975), 39–46. [Hrimiuc and Shimada 1996] H. Hrimiuc and H. Shimada, “On the Lduality between Finsler and Hamilton manifolds”, Nonlinear World 3 (1996), 613–641. [Ichijy¯ o 1976] Y. Ichijy¯ o, “Finsler spaces modeled on a Minkowski space”, J. Math. Kyoto Univ. 16 (1976), 639–652. [Izumi 1976] H. Izumi, “On ∗ PFinsler spaces, I”, Memoirs of the Defense Academy 16 (1976), 133–138. [Izumi 1977] H. Izumi, “On ∗ PFinsler spaces, II”, Memoirs of the Defense Academy 17 (1977), 1–9. [Izumi 1982] H. Izumi, “On ∗ PFinsler spaces of scalar curvature”, Tensor (N.S.) 38 (1982), 220–222. [Ji and Shen 2002] M. Ji and Z. Shen, “On strongly convex indicatrices in Minkowski geometry”, Canad. Math. Bull. 45:2 (2002), 232–246. [Kikuchi 1979] S. Kikuchi, “On the condition that a space with (α, β)metric be locally Minkowskian”, Tensor (N.S.) 33 (1979), 242–246. [Kim and Yim 2003] C.W. Kim and J.W. Yim, “Finsler manifolds with positive constant flag curvature”, Geom. Dedicata 98 (2003), 47–56. [Kosambi 1933] D. Kosambi, “Parallelism and pathspaces”, Math. Z. 37 (1933), 608– 618. [Kosambi 1935] D. Kosambi, “Systems of differential equations of second order”, Quart. J. Math. 6 (1935), 1–12. [Matsumoto 1972a] M. Matsumoto, Foundations of Finsler geometry and special Finsler spaces, Kaiseisha Press, Japan, 1986. [Matsumoto 1972b] M. Matsumoto, “On Creducible Finsler spaces”, Tensor (N.S.) 24 (1972), 29–37. [Matsumoto 1974] M. Matsumoto, “On Finsler spaces with Randers metric and special forms of important tensors”, J. Math. Kyoto Univ. 14 (1974), 477–498. [Matsumoto 1980] M. Matsumoto, “Projective changes of Finsler metrics and projectively flat Finsler spaces”, Tensor (N.S.) 34 (1980), 303–315.
354
ZHONGMIN SHEN
[Matsumoto 1989] M. Matsumoto, “Randers spaces of constant curvature”, Rep. Math. Phys. 28 (1989), 249–261. [Matsumoto and H¯ oj¯ o 1978] M. Matsumoto and S. H¯ oj¯ o, “A conclusive theorem on Creducible Finsler spaces”, Tensor (N.S.) 32 (1978), 225–230. [Matsumoto and Shimada 2002] M. Matsumoto and H. Shimada, “The corrected fundamental theorem on Randers spaces of constant curvature”, Tensor (N.S.) 63 (2002), 43–47. [Mo 1999] X. Mo, “The flag curvature tensor on a closed Finsler space”, Results Math. 36 (1999), 149–159. [Mo 2002] X. Mo, On the flag curvature of a Finsler space with constant Scurvature, to appear in Houston J. Math. [Mo and Shen 2003] X. Mo and Z. Shen, On negatively curved Finsler manifolds of scalar curvature, to appear in Canadian Math. Bull.. [Mo and Yang 2003] X. Mo and C. Yang, “Nonreversible Finsler metrics with nonzero isotropic Scurvature”, preprint, 2003. [Numata 1975] S. Numata, “On Landsberg spaces of scalar curvature”, J. Korea Math. Soc. 12 (1975), 97–100. [Okada 1983] T. Okada, “On models of projectively flat Finsler spaces of constant negative curvature”, Tensor (N.S.) 40 (1983), 117–123. ¨ [Rapcs´ ak 1961] A. Rapcs´ ak, “Uber die bahntreuen Abbildungen metrischer R¨ aume”, Publ. Math. Debrecen 8 (1961), 285–290. [Sabau and Shimada 2001] V. S. Sabau and H. Shimada, “Classes of Finsler spaces with (α, β)metrics”, Rep. Math. Phy. 47 (2001), 31–48. [Shen 1996] Z. Shen, “Finsler spaces of constant positive curvature”, pp. 83–92 in Finsler Geometry (Seattle, 1995), edited by D. Bao et al., Contemporary Mathematics 196, Amer. Math. Soc., Providence, 1996. [Shen 1997] Z. Shen, “Volume comparison and its applications in Riemann–Finsler geometry”, Advances in Math. 128 (1997), 306–328. [Shen 2001a] Z. Shen, Differential geometry of sprays and Finsler spaces, Kluwer, Dordrecht 2001. [Shen 2001b] Z. Shen, Lectures on Finsler geometry, World Scientific, Singapore, 2001. [Shen 2002] Z. Shen, “Twodimensional Finsler metrics of constant flag curvature”, Manuscripta Mathematica 109:3 (2002), 349–366. [Shen 2003a] Z. Shen, “Projectively flat Randers metrics of constant flag curvature”, Math. Ann. 325 (2003), 19–30. [Shen 2003b] Z. Shen, “Projectively flat Finsler metrics of constant flag curvature”, Trans. Amer. Math. Soc. 355(4) (2003), 1713–1728. [Shen 2003c] Z. Shen, “Finsler metrics with K = 0 and S = 0”, Canadian J. Math. 55:1 (2003), 112–132. [Shen 2003d] Z. Shen, “Nonpositively curved Finsler manifolds with constant Scurvature”, Math. Z. (to appear). [Shibata et al. 1977] C. Shibata, H. Shimada, M. Azuma and H. Yasuda, “On Finsler spaces with Randers’ metric”, Tensor (N.S.) 31 (1977), 219–226.
LANDSBERG, S AND RIEMANN CURVATURES
355
[Szab´ o 1977] Z. I. Szab´ o, “Ein Finslerscher Raum ist gerade dann von skalarer Kr¨ ummung, wenn seine Weylsche Projektivkr¨ ummung verschwindet”, Acta Sci. Math. (Szeged) 39 (1977), 163–168. [Szab´ o 1981] Z. I. Szab´ o, “Positive definite Berwald spaces (structure theorems on Berwald spaces)”, Tensor (N.S.) 35 (1981), 25–39. [Xing 2003] H. Xing, “The geometric meaning of Randers metrics with isotropic Scurvature”, preprint. [Yasuda and Shimada 1977] H. Yasuda and H. Shimada, “On Randers spaces of scalar curvature”, Rep. Math. Phys. 11 (1977), 347–360. Zhongmin Shen Department of Mathematical Sciences Indiana University – Purdue University Indianapolis 402 N. Blackford Street Indianapolis, IN 462023216 United States
[email protected] Riemann–Finsler Geometry MSRI Publications Volume 50, 2004
Index Note. This is a partial index and the level of coverage varies from article to article. If you don’t find your entry here, it doesn’t mean it’s not in the book! Searchable PDF files for the articles are available at www.msri.org. affine invariant, 10, 30 affinely equivalent metrics, 317 AkbarZadeh’s rigidity theorem, 254 Alexandrov, A. D., 37 algebraically degenerate, 159 almost isotropic Scurvature, 320 almost metric, 269 (α, β)metric, 307 ´ Alvarez Paiva, J. C., 2, 27, 40 Ambrosio, L., 40 ample, 85, 89, 110, 149 very, 85 anisotropic mean curvature, 51 anisotropic mean curvature flow, 51, 64 anisotropy, 50, 64 Banach–Mazur compactum, 8 Banach–Mazur distance, 8, 10, 35 base locus, 169 Basic Equation, 221, 224, 226, 228, 229, 236 Benson volume, 12 Benson, R. V., 12, 24 Berck, G., 45 Berwald metric, 9, 315 Berwald spaces, 252, 255 Berwald’s formula, 211 split and covariantised, 213 big, 110, 111 Blaschke–Santal´ o inequality, 3, 13, 14, 17, 33 Bolker, E. D., 5, 7 Bonnet–Myers theorem, 214 Brody hyperbolic, 107, 191
broken geodesic, 272 Brunn–Minkowski theory, 31 bumpy metric, 278 bundle ample, 89 cotangent, 263 Finsler, 89 horizontal sub, 91 hyperplane, 88 negative, 86 parametrized kjet, 115 positive, 86 projective, 88 projectivized parametrized jet, 139 pulledback, 202–204 tangent, 263 tautological line, 88 vertical sub, 91 Burago, D., 2, 21, 27 Busemann area, 39 Busemann area definition, 36 Busemann intersection inequality, 40, 43 Busemann volume, 10, 16, 17 Busemann, H., 11, 17, 26–28, 36, 40, 318 Busemann–Petty problem, 36 Cahn–Hoffman vector field, 50, 63 calibration, 22, 27 canonical connection, 269 Cartan tensor, 205, 267, 312 Cartan torsion, 312 CC(23) Equation, 229, 239 Chern character, 150
357
358
PARTIAL INDEX
numbers of Jm k X, 141 of tensor products of bundles, 140 Christoffel symbols, 314, 334 Christoffel symbols, formal, 270 classes of a permutation group, 125 closed geodesic, 277 codisc bundle, 17 coFinsler metric, 308 coarea equality, 44 coarea formula, 44 coarea inequality, 44 coarea inequality for mass∗, 44 Cohn, D. L., 9 color pattern, 312 complete linear system, 85 complex Finsler manifold, 97 compressing property, 28 condition, Legendre, 265 conjX , 275 conjugate locus, 275 conjugate point, 275 first, 275 conjugate radius, 275 connection, 91 canonical, 269 Chern, 149, 205–207 Finsler, 94 LeviCivita, 269 linear, 91 nonlinear, 91, 205 contangent bundle, 263 convex body, 4 convex hypersurface, 5 cotangent bundles, 17 covariant derivative, 271, 314, 315 Crofton formula, 27, 164 crosspolytope, 4, 14 maximal inscribed, 12 crystalline curvature, 51, 73 curvature flow, 74 mean curvature, 51, 53, 73 mean curvature flow, 52, 70, 74, 75 norm, 4, 8, 36–39, 65 cube, 13, 38 cuboctahedron, 13, 39, 40 curvature, 94 constant flag, 209, 216, 228, 239, 255 characterisation, 228
nagivation description, 239 flag, 207–214, 274 holomorphic bisectional, 108 mixed, 149 holomorphic sectional, 97, 107 Ricci, 214–217 Riemann, 274, 303, 327 S, 221, 223, 224, 229, 304, 319 scalar, 208, 209 sectional, 207, 328 spray, 211, 219 tensors, 207 Curvature Equation, 228, 229, 237, 239 cut locus, 289 cut point, 289 CWcomplex, 279 cylinder, right, 13 ∇V , 269 definition of area, 35 degree, 88 Deicke, 312 derivation, 117 derivative covariant, 271 Lie, 264 destabilizing subsheaf, 143 diameter, 288 distance, 268, 313 symmetrized, 268 distinguished section, 204 distortion, 319 divisor class group, 156 divisor, effective, 164 double cone, 13 double dual space, 6 double tangent bundle, 264 dual Brunn–Minkowski theory, 12 dual definition of volume, 11 dual functor, 11 dual linear map, 6 dual norm, 50, 55 dual space, 6 duality map, 55, 63 Duran, C. E., 17 Ecurvature, 320 E23 Equation, 222, 224, 226, 238 effective divisor, 164 Einstein characterisation, 227 Einstein metric, 215–216, 224, 255
PARTIAL INDEX
Einstein navigation description, 238 Einstein rigidity, 217 ellipsoid maximal inscribed, 12 minimal circumscribed, 12 εn Euclidean volume of Euclidean ball, 10 equiframed curve, 34 Euclidean norm, 304 Euclidean space, 305 Euclidean structure, 4 Euler stable, semistable, 176 Euler’s theorem, 200, 265 Ewald, G., 26, 27 example 3sphere, 240 Bryant’s metrics, 209 Cartesian products, 246–247 Finslerian Poincar´e disc, 228, 234, 240 Fubini–Study, 249 Funk disc, 240 K¨ ahler–Einstein, 247–251 Matsumoto’s slopeofamountain metric, 201 Minkowski space, 232 Numata metrics, 208 paraboloid, 245 Ricciflat, 253 Shen’s fish pond, 232 Shen’s fish tank, 241 sphere, 245 surface of revolution, 244–246 torus, 245 Zermelo navigation, 201 exponential map, 286 extendibly convex, 26 facet, 52, 57, 68–72 Fernandes, E., 2, 25, 27, 45 Finsler bundle, 8, 89 Finsler manifold, 305 area on, 19 continuous, 16, 17 volume on, 19 Finsler metric, 9, 32, 89, 118, 149, 199, 265 continuous, 16 smooth, 20 Finsler pseudometric, 118 Finsler space, 40, 44 Finsler structure, 307
359
first conjugate point, 275 first variational formula, 271 flag, 207 flag curvature, 274, 328 flagpole, 207 flat, 95, 101 conformally, 101 projectively, 101 forgetting map, 115, 118 form connection, 94 curvature, 94 torsion, 96 formal Christoffel symbols, 270 Fourier transform, 5, 27 function, reversibility, 266 fundamental cycle, 151 fundamental tensor, 200, 202–204 Fundamental Theorem of Grothendieck, 150 Funk metric, 306, 311, 346 Gardner, R. J., 36, 37, 43 Gauss map, 25 Gaussian image, 21 Gaussian measure, 21 generalized Funk metric, 306 generalized Koszul formula, 269 genus, 88 geodesic, 272, 313 broken, 272, 290 closed, 277 periodic, 277 geodesic field, 327 geodesic polygon, 272 geodesic spray coefficients, 205–207 geodesic variation, 327 geodesic vector field, 272 geometrically ruled, 102 Gol¸ab theorem, 39 Gol¸ab, S., 30, 39 Goodey, P., 5 Grassmannian, 18 convexity on, 26 Gromov mass, 12 Gromov’s mass∗, 12 Gromov, M., 9, 12, 24, 28 H¨ older’s inequality, 6 H¨ ormander, L., 7, 8, 28 Haar’s theorem, 9
360
PARTIAL INDEX
Hahn–Banach theorem, 19, 29 Hamiltonian systems, 2 Hardy–Ramanujan formula, 125 Harvey, R., 22 Hausdorff measure, 2, 11, 16, 17, 50, 52, 53, 57, 60 hexagon, regular, 12 Hilbert form, 204 Hitchin–Thorpe inequality, 217 Hodge manifold, 87 Holmes, R. D., 2, 11, 39 Holmes–Thompson area definition, 37 Holmes–Thompson volume, 2, 11, 16, 17 homogeneous positively, 200, 265 Hug, D., 39 hyperbolic Kobayashi, 98, 108, 191 manifold, 88 hypermetric normed space, 6 hypermetric space, 5, 14, 27, 28, 37, 38 Ic , 276 index, 277 ind c, 277 index form, 276 index of Jm k X, 147 index of stability, 140 indicatrix, 266 infinitesimal color pattern, 317 infinitesimal homothety, 236 injectivity radius, 289 symmetrized, 289 integrability condition, 203 intersection, 12 intersection body, 2, 35–37, 42, 58 irregularity, 108 isometric embedding, 6 isometric spaces, 8 isometric submersion, 6 isometry classes, 8 isoperimetric problem, 31, 40 isoperimetrix, 33, 42, 43 geometry of, 43 area of, 33 perimeter of, 33 isosystolic inequality Berger, 2 Pu, 2 isotropic Ecurvature, 321
isotropic Scurvature, 320 Ivanov, S., 2, 21, 27, 40 Jcurvature, 322 Jacobi field, 274 Jensen’s Formula, 169 John’s theorem, 10 John, F., 8 kdensity, 18, 20 kjet differential ω of weight m, 109 kvolume, 18 kvolume density, 3, 18, 20, 26 mass∗, 19, 29 Busemann, 19 convexity of, 26 Holmes–Thompson, 19 mass, 19 Kirchheim, B., 40 Klein metric, 305 Kobayashi conjecture, 139 Kobayashi hyperbolic, 108, 191 Kobayashi metric, 98 Kobayashi pseudometric, 107 Kodaira–Iitaka dimension, 152 Koszul formula, 269 generalized, 269 Lcurvature, 322 LW Equation, 236 `p space, 4 λ, 266 Landsberg curvature, 304, 322 Landsberg metric, 322 Landsberg tensor, 207, 212 Lawson, H. B., 22 Lebesgue measure, 9 Legendre condition, 265 Legendre transform, 7 Legendre transformation, 265, 308 length, 313 LeviCivita connection, 269 Lie derivative, 264 linear covariant derivative, 315 linear system, 85 linearly parallel vector field, 315, 321 Liouville volume, 11, 17 Lipschitz φregular, 65, 69 locally Minkowski, 209 locus, conjugate, 275 Loewner ellipsoid, 8
PARTIAL INDEX
Loewner theorem, 12 Loewner, C., 8 Logarithmic Derivatives Lemma, 164 logarithmic jet differential, 166 loop space, 281 Lutwak, E., 12, 36 Mahler conjecture, 13, 15 Mahler inequality, 33 Mahler’s conjecture, 3 Mahler, K., 14 Mahler–Reisner inequality, 15 map, exponential, 286 Martini, H., 34 mass, 16 mass∗, 16 mass∗ area definition, 38 Matsumoto identity, 243 Matsumoto torsion, 312 maximal set, 125 McKinney, J. R., 14 mean Berwald curvature, 321 mean Cartan tensor, 312 mean Cartan torsion, 312 mean curvature, 319 mean curvature flow, 50, 52, 60, 62 mean Landsberg curvature, 322 metric, 269 almost, 269 bumpy, 278 Finsler, 89, 118, 149, 199, 265 Fubini–Study, 87 Kobayashi, 98 Minkowski, 310 osculating Riemannian, 273 Poincar´e, 97 pseudoK¨ ahler, 99 Randers, 266 Riemannian, 266, 305 minimizing curve, 313 Minkowski content, 50, 52, 59, 60 Minkowski metric, 310 Minkowski mixed volume inequality, 31, 33, 42 Minkowski norm, 8, 304 Minkowski problem, 3, 37 Minkowski space, 5, 7, 36, 38 Minkowski sum, 31 Minkowski theorem, 21 Minkowski, H., 21, 24, 38
mixed area, 31 mixed volume, 31, 32, 41 Monge–Amp`ere equation, 149 monotonicity, 9 Moustafaev, Z., 34 µht , Holmes–Thompson volume, 11 µb , Busemann volume, 11 µm∗ , Gromov’s mass∗, 12 µm , Gromov mass, 12 multiplicity, 275 nef, 110 negatively curved, 95 Nirenberg, L., 38 nondegenerate, 278 norm, 3 Minkowski, 5, 37, 304 norm of linear map, 5 normalized degree, 175 normalized tuple, 156 nullity, 277 octahedron, 13, 38 of constant flag curvature, 328 of scalar curvature, 328 Okada, T., 310 oriented distance function, 52, 60 osculating Riemannian metric, 273 parallel vector field, 315 parallelogram minimal circumscribed, 22 parallelogram identity, 4 parallelotope minimal circumscribed, 12 partition, 125 perimeter of unit circle, 30 periodic geodesic, 277 φcalibrable facets, 75 Picard group, 156 piecewise smooth, 268 Pl¨ ucker embedding, 18 Pogorelov, A. V., 38 Poincar´e–Lelong formula, 164 point conjugate, 275 first conjugate, 275 polar, 6, 37 polygon, geodesic, 272 polytope, 7 positively homogeneous, 200, 265
361
362
PARTIAL INDEX
principal curvature, 5 projection, 12 projection body, 35, 37, 42 projection inequality Petty, 40, 43 Petty conjectured, 43 projective factor, 330 projective Finsler spaces, 2 projectively flat, 101, 209, 330 pseudoconvex, 89 strictly, 149 pushforward, 161 quartic metric, 200 radius conjugate, 275 injectivity, 289 symmetrized injectivity, 289 Radon curve, 34 Randers metric, 266, 307 definition of, 218 Randers norm, 304 Rapsc´ ak’s identity, 210–211 reduced, 156 reduction, 156 regular, 159 regularity, 200 Reisner, S., 14 relatively isotropic Jcurvature, 322 relatively isotropic Lcurvature, 322 reparametrization, 164, 170 reversibility, 266 reversibility function, 266 reversible, 267 reversible metric, 304 rhombic dodecahedron, 13, 39, 40 Ricci Curvature Equation, 222, 224, 226, 237, 238 Ricci identity, 336 Ricci rigidity theorem, 253 Ricci scalar, 214–215 Ricci tensor, 214–215 Riemann curvature, 274, 303, 327 Riemann–Roch for curves, 130 formula, 150 for jet differentials, 162 Riemann–Roch theorem, 88 Riemannian metric, 266, 305
osculating, 273 ruled algebraic surface, 102 ruled manifold, 102 ruled, geometrically, 102 Scurvature, 304, 319 Sch¨ affer, J. J., 30, 39 Schneider, R., 2, 5, 31 Schur lemma, 210, 241 Schwarz Lemma, 168 classical, 165 sectional curvature, 328 sheaf destabilizing sub, 140 semistable, 140 sheaf of germs of kjet differentials of weight m, 116 of kjet forms, 116 of holomorphic kjets, 114 Shephard, G. C., 26, 27 short basis, 12 short linear map, 5, 35 σ, 221, 223, 227, 229 simple kvector, 18 simple normal crossings, 166 simple tangent kvector, 19 smooth, piecewise, 268 spanned, 110, 165 spray, 314 spray coefficients, 314 stable, 175 Euler semi, 176 semi, 175 starshaped body, 37 strong convexity, 200, 255 Randers, 218 strongly convex domain, 310 subbundle, vertical, 91, 264 subspace of L1 , 4 supporting hyperplane, 28 surfacearea measure, 25 symmetrized distance, 268 symmetrized injectivity radius, 289 symplectic form, 17 symplectic structure, 11 symplectic volume, 11, 43 tangent bundle, 263 double, 264 tangent space, vertical, 264 tautological section, 91
PARTIAL INDEX
Taylor, J., 40 tensor, Cartan, 267 theorem, Euler’s, 265 ϑ(p, q), 289 θ, 268 Thompson, A. C., 2, 11, 38–40, 43, 44 3d rigidity, 242 Todd class, 150 totally convex, 28, 29 transformation, Legendre, 265 uniqueness problem for rational and meromorphic functions, 136 unit ball, 4 volume of, 13 vanishing theorem, 152 variational formula, first, 271 vector field, geodesic, 272 vertical subbundle, 91, 264 vertical tangent space, 264 volume density, 15, 48; see also kvolume density (n−1)volume density mass∗, 24 Busemann, 24 Holmes–Thompson, 24
363
volume form, 32 volume of submanifold, 16 volume on continuous Finsler manifold, 16 volume product, 11, 13 volume, definition of, 10 Wallen, L. J., 34 Waring problem, 120 weakly convex, 26, 35 weakly Landsberg metric, 322 wedge body, 38, 39, 43 weight, 153 weighted projective space, 153 Weil, W., 5, 27 Weyl module, 124 Whitney formula, 139 Wieacker, J., 2 Wronskian, 137 Wulff shape, 35 Zermelo navigation, 230–234, 238 example, 201 Zhang, G., 36 zonoid, 6, 15, 37 zonotope, 6