Metric Structures for Riemannian and Non

Riemannian Spaces Mikhail Gromov pdf download

https://ebookgate.com/product/metric-structures-for-riemannian-
and-non-riemannian-spaces-mikhail-gromov/

Get the full ebook with Bonus Features for a Better Reading Experience on ebookgate.com
Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Riemannian geometry a modern introduction 2nd Edition

Isaac Chavel

https://ebookgate.com/product/riemannian-geometry-a-modern-
introduction-2nd-edition-isaac-chavel/

ebookgate.com

Minimal Submanifolds in Pseudo riemannian Geometry 1st

Edition Henri Anciaux

https://ebookgate.com/product/minimal-submanifolds-in-pseudo-
riemannian-geometry-1st-edition-henri-anciaux/

ebookgate.com

The geometry of curvature homogeneous pseudo Riemannian

manifolds First Edition Peter B. Gilkey

https://ebookgate.com/product/the-geometry-of-curvature-homogeneous-
pseudo-riemannian-manifolds-first-edition-peter-b-gilkey/

ebookgate.com

Introduction to Metric and Topological Spaces 2nd Edition

Wilson A Sutherland

https://ebookgate.com/product/introduction-to-metric-and-topological-
spaces-2nd-edition-wilson-a-sutherland/

ebookgate.com
An Introduction to Metric Spaces and Fixed Point Theory
1st Edition Mohamed A. Khamsi

https://ebookgate.com/product/an-introduction-to-metric-spaces-and-
fixed-point-theory-1st-edition-mohamed-a-khamsi/

ebookgate.com

Classifying Spaces of Degenerating Polarized Hodge

Structures AM 169 Kazuya Kato

https://ebookgate.com/product/classifying-spaces-of-degenerating-
polarized-hodge-structures-am-169-kazuya-kato/

ebookgate.com

Metric Survey Specifications for Cultural Heritage 2nd,

Revised Edition Paul Bryan

https://ebookgate.com/product/metric-survey-specifications-for-
cultural-heritage-2nd-revised-edition-paul-bryan/

ebookgate.com

Geometry of Time Spaces Non Commutative Algebraic Geometry

Applied to Quantum Theory 1st Edition Audal

https://ebookgate.com/product/geometry-of-time-spaces-non-commutative-
algebraic-geometry-applied-to-quantum-theory-1st-edition-audal/

ebookgate.com

Peptidomics methods and applications 1st Edition Mikhail

Soloviev

https://ebookgate.com/product/peptidomics-methods-and-
applications-1st-edition-mikhail-soloviev/

ebookgate.com
Modern Birkhauser Classics

Many of the original research and survey monographs in pure and

applied mathematics published by Birkhauser in recent decades
have been groundbreaking and have come to be regarded as foun-
dational to the subject. Through the MBC Series, a select number of
these modern classics, entirely uncorrected, are being re-released in
paperback (and as eBooks) to ensure that these treasures remain ac-
cessible to new generations of students, scholars, and researchers.
 Metric Structures
for Riemannian and
Non-Riemannian Spaces

Misha Gromov
With Appendices by
M. Katz, P. Pansu, and S. Semmes

English translation
by Sean Michael Bates

Reprint of the 2001 Edition

Birkhauser
Boston • Basel • Berlin
 Mikhail Gromov
Institut des Hautes Etudes Scientifiques
Departement de Mathematiques
F-91440 Bures-sur-Yvette
France

Originally published as volume 152 in the series Progress in Mathematics

Cover design by Alex Gerasev.

Mathematics Subject Classification (2000): 01A60, 01A75, 51K99, 53-02, 53C20, 53C23,
53C70, 57N65

Library of Congress Control Number: 2006937425

ISBN-10: 0-8176-4582-9 e-ISBN-10: 0-8176-4583-7

ISBN-13: 978-0-8176-4582-3 e-ISBN-13: 978-0-8176-4583-0

Printed on acid-free paper

©2007 Birkhauser Boston BirUhaUSCr

Based on Structures Metriques des Varietes Riemanniennes
Edited by J. LaFontaine and P. Pansu
All rights reserved. This work may not be translated or copied in whole or in part without the writ-
ten permission of the publisher (Birkhauser Boston, c/o Springer Science+Business Media LLC, 233
Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or
scholarly analysis. Use in connection with any form of information storage and retrieval, electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereafter de-
veloped is forbidden.
The use in this publication of trade names, trademarks, service marks and similar terms, even if they
are not identified as such, is not to be taken as an expression of opinion as to whether or not they are
subject to proprietary rights.

9 8 7 6 5 4 3 2 1

www.birkhauser.com (IBT)
Misha Gromov
with Appendices by
M. Katz, P. Pansu, and S. Semmes"

Metric Structures
for Riemannian and
Non-Riemannian Spaces
Based on Structures Metriques des Varietes Riemanniennes
Edited by J. LaFontaine and P. Pansu
English Translation by Sean Michael Bates

Birkhauser
Boston • Basel • Berlin
Author Translator (English edition)
Mikhail Gromov Sean Michael Bates
Departement de Mathematiques Department of Mathematics
Institut des Hautes Etudes Scientifiques Columbia University
Bures-sur-Yvette, France New York, NY 10027, USA

Editors (French Edition)

Jacques LaFontaine Pierre Pansu
Departement des Sciences Mathematiques Departement de Mathematiques
Universite de Montpellier Universite de Paris-Sud
2, Place E. Bataillon, 91405 Orsay Cedex, France
34095 Montpellier Cedex 5, France

Library of Congress Cataloging-in-Publication Data

Gromov, Mikhail
[Structures metriques des varietes riemanniennes. English]
Metric structures for Riemannian and non-Riemannian spaces /
Mikhail Gromov ; translated by Sean Michael Bates,
p. cm. - (Progress in mathematics ; v. 152)
Includes bibliographical references and index.
ISBN 0-8176-3898-9 (acid-free paper)
1. Riemannian manifolds. I. Title. II. Series: Progress in
mathematics (Boston, Mass.) ; vol. 152.
QA629.G8313 1998 97-24633
516.373-DC21 CIP

AMS Subject Classifications: 53B21, 53C20, 53C21, 53C23, 58D17, 54E35, 51H20, 51H25,
54E35, 28A78

Printed on acid-free paper.

© 1999 Birkhauser Boston, first printing . .. 5^'^
© 2001, second printing with corrections BirUndUSer j ^
English edition based on
Structures Metriques des Varietes Riemanniennes
© 1999, S. Semmes, Appendix B: Metric Spaces and
Mappings Seen at Many Scales

All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Birkhauser Boston, c/o Springer-Verlag New York, Inc., 175 Fifth Avenue,
New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use
in connection with any form of information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the former
are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks
and Merchandise Marks Act, may accordingly be used freely by anyone.
ISBN 0-8176-3898-9 SPIN 10835928
ISBN 3-7643-3898-9

Typeset in lATfeX by Sean Michael Bates, New York, NY

Printed and bound by Hamilton Printing, Rensselaer, NY
Printed in the United States in America

9 8 7 6 5 4 3 2
 ...Meme ceux qui furent favorables a ma perception des verites que je
voulais ensuite graver dans le temple, me feliciterent de les avoir decouvertes
au microscope, quand je m^etais au contraire servi d'un telescope pour
apercevoir des choses, tres petites en effet, mais parce qu ^elles etaient situees
a une grande distance, et qui etaient chacune un monde.

Marcel Proust, Le temps retrouve

(Pleiade, Paris, 1954, p. 1041)
Contents

Preface to the French Edition xi

Preface to the English Edition xiii

Introduction: Metrics Everywhere xv

Length Structures: Path Metric Spaces 1

A. Length structures 1
B. Path metric spaces 6
C. Examples of path metric spaces 10
D. Arc-wise isometrics 22

2 Degree and Dilatation 27

A. Topological review 27
B. Elementary properties of dilatations for spheres 30
C. Homotopy counting Lipschitz maps 35
D. Dilatation of sphere-valued mappings 41
E4- Degrees of short maps between compact and
noncompact manifolds 55

3 Metric Structures on Families of Metric Spaces 71

A. Lipschitz and Hausdorff distance 71
B. The noncompact case 85
C. The Hausdorff-Lipschitz metric,
quasi-isometries, and word metrics 89
D-(. First-order metric invariants
and ultralimits 94
E_^ Convergence with control 98

3^ Convergence and Concentration of Metrics and Measures 113

A. A review of measures and mm spaces 113
B. DA-convergence of mm spaces 116
C. Geometry of measures in metric spaces 124
viii Contents

D. Basic geometry of the space A! 129

E. Concentration phenomenon 140
F. Geometric invariants of measures
related to concentration 181
G. Concentration, spectrum, and
the spectral diameter 190
H. Observable distance H;^ on the space X and concentration
X"" -^X 200
I. The Lipschitz order on A', pyramids, and asymptotic con-
centration 212
J. Concentration versus dissipation 221

4 Loewner Rediscovered 239

A. First, some history (in dimension 2) 239
B. Next, some questions in dimensions > 3 244
C. Norms on homology and Jacobi varieties 245
D. An application of geometric integration theory 261
E+ Unstable systolic inequalities and filling 264
F4. Finer inequalities and systoles
of universal spaces 269

5 Manifolds with Bounded Ricci Curvature 273

A. Precompactness 273
B. Growth of fundamental groups 279
C. The first Betti number 284
D. Small loops 288
E+ Applications of the packing inequalities 294
F+ On the nilpotency of TTI 295
G+ Simplicial volume and entropy 302
H_|_ Generalized simplicial norms and
the metrization of homotopy theory 307
1+ Ricci curvature beyond coverings 316

6 Isoperimetric Inequalities and Amenability 321

A. Quasiregular mappings 321
B. Isoperimetric dimension of a manifold 322
C. Computations of isoperimetric dimension 327
D. Generalized quasiconformality 336
E+ The Varopoulos isoperimetric inequality 346

7 Morse Theory and Minimal Models 351

A. Application of Morse theory
to loop spaces 351
B. Dilatation of mappings between
simply connected manifolds 357
Contents ix

8+ Pinching and Collapse 365

A. Invariant classes of metrics
and the stabiUty problem 365
B. Sign and the meaning of curvature 369
C. Elementary geometry of collapse 375
D. Convergence without collapse 384
E. Basic features of collapse 390

A "Quasiconvex" Domains
in M^ 393

B Metric Spaces and Mappings Seen

at Many Scales 401
I. Basic concepts and examples 402
1. Euclidean spaces, hyperbolic spaces, and ideas from analysis 402
2. Quasimetrics, the doubling condition, and
examples of metric spaces 404
3. Doubling measures and regular metric spaces, deformations
of geometry, Riesz products and Riemann surfaces 411
4. Quasisymmetric mappings and deformations of geometry from
doubling measures 417
5. Rest and recapitulation 422
II. Analysis on general spaces 423
6. Holder continuous functions on metric spaces 423
7. Metric spaces which are doubling 430
8. Spaces of homogeneous type 435
9. Holder continuity and mean oscillation 437
10. Vanishing mean oscillation 439
11. Bounded mean oscillation 443
III. Rigidity and structure 445
12. Differentiability almost everywhere 445
13. Pause for reflection 448
14. Almost flat curves 448
15. Mappings that almost preserve distances 452
16. Almost flat hypersurfaces 455
17. The ^oo condition for doubling measures 458
18. Quasisymmetric mappings and doubling measures 462
19. Metric doubling measures 464
20. Bi-Lipschitz embeddings 468
21. Ai weights 470
22. Interlude: bi-Lipschitz mappings between Cantor sets . . . 471
23. Another moment of reflection 471
24. Rectifiabihty 471
25. Uniform rectifiabihty 475
X Contents

26. Stories from the past 477

27. Regular mappings 479
28. Big pieces of bi-Lipschitz mappings 480
29. Quantitative smoothness for Lipschitz functions 482
30. Smoothness of uniformly rectifiable sets 488
31. Comments about geometric complexity 490
IV. An introduction to real-variable methods 491
32. The Maximal function 491
33. Covering lemmas 493
34. Lebesgue points 495
35. Differentiability almost everywhere 497
36. Finding Lipschitz pieces inside functions 502
37. Maximal functions and snapshots 505
38. Dyadic cubes 505
39. The Calderon-Zygmund approximation 507
40. The John-Nirenberg theorem 508
41. Reverse Holder inequalities 511
42. Two useful lemmas 513
43. Better methods for small oscillations 515
44. Real-variable methods and geometry 517

C Paul Levy's Isoperimetric Inequality 519

D Systolically Free Manifolds 531

Bibliography 545

Glossary of Notation 575

Index 577
Preface to the
French Edition
This book arose from a course given at the University of Paris VII during the
third trimester of 1979. My purpose was to describe some of the connections
between the curvature of a Riemannian manifold V and some of its global
properties. Here, the word global refers not only to the topology of V but
also to a family of metric invariants, defined for Riemannian manifolds
and mappings between these spaces. The simplest metric invariants of
such V are, for example, its volume and diameter; an important invariant
of a mapping from Vi into V2 is its dilatation. In fact, such invariants
also appear in a purely topological context and provide an important link
between the given infinitesimal information about V (usually expressed as
some restriction on the curvature) and the topology of V.

For example, the classical Gauss-Bonnett theorem gives an upper bound

for the diameter of a positively curved manifold V, from which one can de-
duce the finitude of the fundamental group of V. For a deeper topological
study of Riemannian manifolds, we need more subtle invariants than di-
ameter or volume; I have attempted to present a systematic treatment of
these invariants, but this treatment is far from exhaustive.
Messrs. J. Lafontaine and R Pansu have successfully completed the
almost insurmountable task of transforming into a rigorous mathematical
text the chaos of my course, which was scattered with imprecise statements
and incomplete proofs. I thank them as well as M. Berger, without whose
assistance and encouragement this book would never have come into being.
I also thank the Editions Cedic for the liberty it afforded the authors at
the time that the book was edited.

M. Gromov
Paris, June 1980
Preface to the
English Edition
The metric theory described in this book covers a domain stretching some-
where between the fields of topology and global Riemannian geometry. The
boundary of this domain has dramatically exploded since 1979 and accord-
ingly, in the course of its translation from the 1979 French version into En-
glish, the book has approximately quadrupled in size, even though I tried
not to maim the original text with unnecessary incisions, insertions, and
corrections, but rather to add several new sections indicated by the + sub-
script. The most voluminous additions are Chapter 3^ , which attempts to
link geometry and probability theory, and Appendix B.^, where an analyst
lays down his view on metric spaces. Here, the reader can painlessly learn
several key ideas of real analysis made accessible to us geometers by the
masterful exposition of Stephen Semmes, who has adapted his material to
our non-analytic minds.
Additionally, Appendix C_^ reproduces my 1980-rendition of Paul Levy's
inequality, while Misha Katz gives an overview of systolic freedom in Ap-
pendix D+.
Acknowledgements: The initiative to publish an English translation of the
book with Birkhauser is due to Alan Weinstein. It was my pleasure to
cooperate with Sean Bates, who translated the original version of the book
and helped me edit the new sections. I am also grateful to Marcel Berger,
Keith Burns, and Richard Montgomery for calling my attention to errors
in the first version of the book.
M. Gromov
Bures-sur-Yvette, May 1997
Introduction:
Metrics Everywhere

The conception of "distance" is already present in everyday language where

it refers to two physical objects or two abstract ideas being mutually close or
far apart. The most common (but by no means most general) mathematical
incarnation of this vague idea is the notion of metric space^ that is, an
abstract set X where the distance between its elements, called points x e X,
is measured by positive real numbers. Thus a metric space is a set X with a
given function in two variables, say d : X x X —^ R_|_ satisfying the famous
triangle inequality
d{x, x'') < d{x, x') + d{x', x'')

for all triples of points x, x' and x'^ in X.

Besides, one insists that the distance function be symmetric, that is,
d{x,x') = d{x',x). (This unpleasantly limits many applications: the effort
of climbing up to the top of a mountain, in real life as well as in mathe-
matics, is not at all the same as descending back to the starting point).
Finally, one assumes d{x, x) = 0 for all x G X and add the following
separation axiom. \i x ^ x', then c/(x,x') = 0. This seems to be an
innocuous restriction, as one can always pass to the quotient space by
identifying x and x' whenever (i(x, x') = 0. But sometimes the separation
becomes a central issue, e.g., for Kohayashi and Hofer metrics, where such
identification may reduce X to a single point, for instance).
The archtypical example of a metric space is the ordinary Euclidean
space R^ with the pythagorean distance between the points x = {xi,... ,Xn)
and x' = {x[,... ,x'^) defined by

d{x,x') = ^(xi-x;)2 + ... + (xn-x;,)2.

Next come subsets in M^ with this metric providing many appealing
examples, such as the sphere S^~^ = {x G IR^IXir=i^i ~ ^} ^^^ ^^^ ^^^
xvi Metric Structures for Riemannian and Non-Riemannian Spaces

of vertices of the unit cube, i.e., {0,1}^ C R^ with the induced Euclidean
(Pythagorean) distance. If X is a smooth connected submanifold in R^
(as the above sphere) then, besides the induced EucHdean distance distjRn
on X, one has the induced Riemannian distance^ distx(^, x') defined as the
infimum of lengths of curves contained in X and joining x and x\ (One may
be tempted to use the above as a quick definition of a Riemannian metric.
Indeed, every Riemannian manifold admits a smooth embedding into some
R^ preserving the length of the curves according to the Nash theorem. But
Euclidean embeddings hide rather than reveal the true metric structure of
Riemannian manifolds due to uncontrolled distortion distx | dist^n. Besides
the full beauty and power of Riemannian geometry depend not only on the
metric but also on the associated elliptic Riemannian equations, such as
Laplace-Hodge, Dirac, Yang-Mills, and so on. These naturally come along
with the Riemannian tensor but are nearly invisible on X embedded to R'^.
Our approach to general metric spaces bears the undeniable imprint
of early exposure to Euclidean geometry. We just love spaces sharing a
common feature with R^. Thus there is a long tradition of the study
of homogeneous spaces X where the isometry group acts transitively on
X. (In the Riemannian case the metric on such X is fully determined by
prescribing a positive definite quadratic form go on a single tangent space
Tx^{X). But the simplicity of this description is illusory; it is quite hard to
evaluate metric invariants of X in terms of po- For example, one has a very
limited idea of how systoles (see below) behave as one varies a left invariant
metric on a Lie group SO{n) or U{n) for instance.) Besides isometrics,
R^ possesses many nontrivial self-similarities^ i.e., transformations / , such
that /*(dist) = A dist for some constant A ^ 0,1. There are no self-
similar spaces besides R'^ in the Riemannian category — this is obvious —
but there are many such non-Riemannian examples such as p-adic vector
spaces (these are totally disconnected) and some connected nilpotent Lie
groups (e.g., the Heisenherg group) with Carnot-Caratheodory metrics (see
1.4, 1.18 and 2.6 in Appendix B).
Switching the mental wavelength, one introduces spaces with curvature
X < 0 and K >^hy requiring their small geodesic triangles to be "thinner"
(correspondingly, "thicker") than the Euclidean ones (see 1.19+). Here
one is guided by the geometry of symmetric spaces that are distinguished
homogeneous spaces, such as S^ and CP^ where K" > 0 and SLn^/SO{n)
with K <().
Apart from direct Euclidean descendants there are many instances of
metrics associated to various structures, sometimes in a rather unexpected
and subtle way. Here are a few examples.
Introduction: Metrics Everywhere xvii

Complex manifolds. The complex space C^,n > 1, carries no metric

invariant under holomorphic automorphisms. There are just too many of
them! Yet, many complex (and almost complex) manifolds, e.g., bounded
domains in C^, do possess such natural metrics, for example, the Kohayashi
metric (see 1.8 bis^.).

Symplectic manifolds. No such manifold X of positive dimension carries

an invariant metric, again because the group of symplectomorphisms is too
large. Yet, the infinite dimensional space of closed lagrangian submanifolds
in X (or rather each "hamiltonian component" of this space) does admit
such a metric. (The construction of the metric is easy but the proof of
the separation property, due to Hofer, is quite profound. Alas, we have no
room for Hofer's metric in our book).

Homotopy category. Once can functorially associate an infinite dimen-

sional metric polyhedron to the homotopy class of each topological space X,
such that continuous maps between spaces transform to distance decreasing
maps between these polyhedra. Amazingly, the metric invariants of such
polyhedra (e.g., its systoles, the volumes of minimal subvarieties realizing
prescribed homology classes) lead to new homotopy invariants which are
most useful for (aspherical) spaces X with large fundamental groups (see
Ch. 5H+).

Discrete groups. A group with a distinguished finite set of generators

carries a natural discrete metric which only moderately, i.e., bi-Lipschitzly,
changes with a change of generators. Then, by adopting ideas from the
geometry of noncompact Riemannian manifolds, one defines a variety of
asymptotic invariants of infinite groups that shed new light on the whole
body of group theory (see 3C, 5B, and 6B, C).
Lipschitz and bi-Lipschitz. What are essential maps between metric
spaces? The answer "isometric" leads to a poor and rather boring category.
The most generous response "continuous" takes us out of geometry to the
realm of pure topology. We mediate between the two extremes by empha-
sizing distance decreasing maps / : X —> F as well as general X-Lipschitz
maps / which enlarge distances at most by a factor A for some A > 0.
Isomorphisms in this categeory are X-bi-Lipschitz homeomorphisms and
most metric invariants defined in our book transform in a A-controUed way
under A-Lipschitz maps, as does for example, the diameter of a space,
DiamX = sup^ ^.z dist(a:,x'). We study several classes of such invariants
with a special treatment of systoles measuring the minimal volumes of ho-
mology classes in X (see Ch. 4 and App. D) and of isoperimetric profiles
of complete Riemannian manifolds and infinite groups which are linked in
xviii Metric Structures for Riemannian and Non-Riemannian Spaces

Ch. 6 to quasiconformal and quasiregular mappings.

Asymptotic viewpoint. Since every diffeomorphism between compact

Riemannian manifolds is A-bi-Lipschitz for some A < CXD, our invariants tell
us preciously little if we look at a fixed compact manifold X. What is
truly interesting in the A-Lipschitz environment is the metric behavior of
sequences of compact spaces. This ideology applies, for example, to an in-
dividual noncompact space X, where the asymptotic geometry may be seen
as X is exhausted by a growing sequence of compact subspaces. We also
study sequences of maps and homotopy classes of maps between fixed com-
pact spaces, say fi'.X—^Y (see Ch. 2, 7) and relate this asymptotic metric
behavior to the homotopy structure of X and Y (with many fundamental
questions remaining open).

Metric sociology. As our perspective shifts from individual spaces X

to families (e.g., sequences) of these, we start looking at all metric spaces
simultaneously and observe that there are several satisfactory notions of
distance between metric spaces (see Ch. 3). Thus we may speak of various
kinds of metric convergence of a sequence Xi to a space X and study the
asymptotics of particular sequences. For example, if Xj, i = 1,2,... , are
Riemannian manifolds of dimension n with a fixed bound on their sectional
curvatures, then there is a subsequence that converges (or collapses) to a
mildly singular space of dimension m <n (see Ch. 8).
Metric, Measure and Probability. Suppose our metric spaces are
additionally given some measures, e.g., the standard Riemannian measures
if we deal with compact Riemannian manifolds. Here one has several no-
tions of metric convergence of spaces modulo subsets of measure —> 0 (see
Ch. 3^ ). Then there is a weaker convergence most suitable for sequences
Xi with dimXi -^ oo. According to this, unit spheres S* C W'^^ with
normalized Riemannian measures converge (or concentrate, see 3^ ) to a
single atom of unit mass! This can be regarded as a geometric version of the
law of large numbers that is derived in the present case from the spherical
isoperimetric inequality (see 3^ E and Appendix C).

From local to global. This is a guiding principle in geometry as well

as in much of analysis, and the reader will find it in all corners of our book.
It appears most clearly in Ch. 5 where we explain how the lower bound on
the Ricci curvature of a manifold X implies the measure doubling property^
saying that the volume of each not very large 2r-ball B{2r) C X does not
exceed const Vol B{r) for the concentric ball B{r) C 5 ( 2 r ) . This leads to
several topological consequences concerning the fundamental group of X
(see Ch. 5).
Introduction: Metrics Everywhere xix

Besides the volumes of balls, the Ricci curvature controls the isoperi-
metric profile of X. For example, the spherical isoperimetric inequality
generalizes to the manifolds with Ricci > — const (see Appendix C).
Analysis on metric spaces. Smooth manifolds and maps, being in-
finitesimally linear, appear plain and uneventful when looked upon through
a microscope. But singular fractal spaces and maps display a kaleidoscopic
variety of patterns on all local scales. Some of these spaces and maps are
suHiciently regular, e.g., they may possess the doubhng property, and pro-
vide a fertile soil for developing rich geometric analysis. This is exposed by
Stephen Semmes in Appendix B.
I have completed describing what is in the book. It would take another
volume just to indicate the full range of applications of the metric idea in
various domains of mathematics.

Misha Gromov
April 1999
Chapter 1

Length Structures:
P a t h Metric Spaces

Introduction
In classical Riemannian geometry, one begins with a C ^ manifold X
and then studies smooth, positive-definite sections g of the bundle S^T*X.
In order to introduce the fundamental notions of covariant derivative and
curvature (cf. [Grl-Kl-Mey] or [MilnorjMT, Ch. 2), use is made only of the
differentiability of g and not of its positivity, as illustrated by Lorentzian
geometry in general relativity. By contrast, the concepts of the length of
curves in X and of the geodesic distance associated with the metric g rely
only on the fact that g gives rise to a family of continuous norms on the
tangent spaces T^X of X. We will study the associated notions of length
and distance for their own sake.

A. Length structures
1.1. Definition: The dilatation of a mapping / between metric spaces
X, Y is the (possibly infinite) number

where "oJ" stands for the metrics (distances) distx in X and disty in Y.
The local dilatation of / at 2: is the number

diU(/) = l i m d i l ( / | B ( . , . ) ) .
2 Metric Structures for Riemannmn and Non-Riemannian Spaces

A map / is called Lipschitz if dil(/) < OD; it is called X-Lipschitz if dil(/) <
A, in which case, the infimal such A is called the Lipschitz constant of / .

If / is a Lipschitz mapping of an interval [a, b] into X, then the function

11—^ d\\t{f) is measurable.

1.2. Definition: The length of a Lipschitz map / : [a, 6] —^X is the

number

J a

If / is merely continuous, we can define £{f) as the supremum of all sums

of the form Yl'i^od{f{U), fiU+i)) where a = to < ti < -- - < tn^i = 6 is a
finite partition of [a,b].
If (/? is a homeomorphism of a closed interval / ' onto I = [a, 6], then i
satisfies i{foif) = i{f), as follows from the fact that (f is strictly monotone
(invariance under change of parameter).

The two definitions of i{f) stated above are equivalent when / is ab-
solutely continuous (cf. [Rinow], p. 106). This fact permits us to define
£{f) as the integral of the local dilatation when / is Lipschitz and to set
i{f o if) = £{f) for each homeomorphism (^ of 7 onto / ' . More generally,

1.3. Definition: A length structure on a set X consists of a family C{I)

of mappings / : / —^ X for each interval I and a map i of C = [jC{I) into
R having the following properties:

(a) Positivity: We have £(/) > 0 for each f e C, and £{f) = 0 if and
only if / is constant (we assume of course that the constant functions
belong to C).
(6) Restriction, juxtaposition: If / C J, then the restriction to /
of any member of C{J) is contained in C{I). If / G C([a,6]) and
g E C{[b,c]), then the function h obtained by juxtaposition of / and
g lies in C{[a, c]) and i{h) = £{f) + ({g).

(c) Invariance under change of parameter: If (/? is a homeomor-

phism from / onto J and if / G C(J), then f o ip e C{I) and
eifoip) = e{f).
(d) Continuity: For each / = [a, 6], the map 11-> £{f\[a,t]) is continuous.

Using conditions (a), (6), and (c), we can define a pseudo-metric di on X

called the length metric by setting

d,{x,y) = mi{i{f) :feC,x,ye im(/)}.

1. Length Structures: Path Metric Spaces 3

As usual, this pseudo-metric induces a topology on X.

It is common to define i{f) = oc when the map f : I ^^ X is not
contained in C(/).

1.4. Examples:

(a) A metric space (X, d) is equipped with a canonical length structure:

The set C consists of all continuous mappings from intervals into X, and
the function £ is defined as in 1.2 above. The resulting structure is called
the metric length structure of (X, d); in general, however, the length metric
di differs from d, and their corresponding topologies may also be distinct.

(64.) Tits-like metrics and snowflakes: Consider R^ equipped with polar

coordinates (r, 5), where r G [0,oo) and s G 5^"-^, the unit sphere in R^.
Define
d{xi,X2) = In - r 2 | -f r||5i - S 2 p / ^ ,
where Xi = (r^, Sj) G R'^, i = 1,2, ||si — 52II denotes the Euclidean distance
on S^~^ C R", and r = min{ri, r2}. This d gives rise to the usual topology
on R^, but
de{x,,x,) = l l^^-^^l f'^^Z'^
and so (R'^, di) becomes the disjoint union of the Euclidean rays [0, 00) x 5
for all s € 5"~^, all glued together at the origin only. In particular, the
unit sphere S^~^ C R^ is discrete with respect to de. Metrics of this type
naturally appear on (the ideal boundaries of) manifolds with nonpositive
sectional curvatures and are collectively referred to as Tits metrics (cf.
[Ba-Gr-Sch]).
An analog of the metric de can be constructed on the subset of Euclidean
3-space consisting of the straight cone X C R^ over the Koch snowflake
5 C R^. (Here, the snowflake is the base of the cone and plays the role of the
sphere S^~^ in the Tits-like example above). The only curves in X having
finite Euclidean length are those contained in the (straight) generating lines
of the cone, and so these lines are disjoint with respect to di away from the
vertex (compare [Rinow], p. 117, and Appendix B_^ of this book).
In general, the metrics d, di always satisfy the inequality d < di, so that
their corresponding topologies coincide if and only if for each x e X and
£ > 0, there exists a ^-neighborhood of x in which each point is connected
to X by a curve of length at most e.
 Metric Structures for Riemannian and Non-Riemannian Spaces

(c) If X is a manifold, then any Riemannian or Finslerian structure on

X naturally gives rise to a length structure: One proceeds as in 1.2, noting
that when / is differentiable, its local dilatation at a point x equals the
norm of its derivative at x.

(d) Induced length structures: If X is equipped with a length structure

and (^ is a map from a set Y into X, then we obtain a length structure on
Y by setting

for each f:I-^Y.

(e) First exposure to Carnot-Caratheodory spaces. We can associate a

length structure on a Riemannian manifold (V, g) with any tangent subbun-
dle E C TV by defining the length of a curve c to be its usual Riemannian
length if c is absolutely continuous and its tangent vector lies within E at
a.e. point, and by setting ^(c) = oo otherwise. If E is integrable, then the
topology defined by d(_ is none other than the leaf topology. The case of
nonintegrable E is of great interest.
A basic example of the latter structure is provided by the 3-dimensional
Heisenberg group M^ of matrices of the form
1. Length Structures: Path Metric Spaces 5

equipped with a left-invariant metric. The quotient of H^ by its center C

(isomorphic to R) defines a Riemannian fibration (see [Ber-Gau-Maz], Ch.
1) of H^ over the Euchdean plane M^/C c^ R^ rpj^^ subbundle E then
consists of the horizontal subbundle of this fibration, which coincides with
the kernel of the 1-form dz — xdy.

1.5. Suppose X is equipped with a length structure ^, and let i be the

length structure defined by the metric d£. The following criterion, which is
nothing more than an axiomatic version of the classical properties of the
lengths of curves in metric spaces, describes when these two structures are
identical.

1.6. Proposition: //, for each interval I, the function i is lower semi-
continuous on C{I) with respect to the compact-open topology, then £ = i.

Proof. By 1.3(d), the function t \-^ ^(/|[a,t]) is uniformly continuous on

/ = [a, b]. For each £ > 0, there exists 77 > 0 such that if |t — t'| < r/, then
de{m,f{t'))<s.
Let a = to < ti < ' " < tn-f 1 = 6 be a partition of / having increments
no larger than rj. For each integer i between 0 and n, there exists a map gi
in C{[ti,ti^i]) having the same values as / at ti,ti^i such that
i{gi)<de{f{ti)J{ti^i))-hs/n.
By juxtaposing the gi, we obtain a curve h^ satisfying
n n
i{h,) = Y,i{9i) < YldeifiU), fiti+i)) + e < i{f) + e

and such that for each t G / , we have d(,{h^{t), f{t)) < 3e.
From the hypothesis that i is lower semicontinuous, it follows that
i{f)<\imini£{h,)<i{f),
whereas the opposite inequality is an immediate consequence of the defini-
tion of L

Remark: If i is the length structure associated with a metric d, then the

same argument as above shows that £ = i, using the semicontinuity of
length with respect to d (cf. 1.2 and [Choq], p. 137). In other words, by
following the sequence of constructions
(X, d), a metric metric length (X, d^), a new metric
space —y structure £ —> associated with the
on X length structure.
6 Metric Structures for Riemannian and Non-Riemannian Spaces

we obtain the same length structure. Nevertheless, we again emphasize

that £ 7^ £ in general.

1.6^ Locality of the length structure. If two length structures agree

on some open subsets covering X, then they are obviously equal. Con-
versely, if we are given a covering of X by open subsets Xi for i G / ,
together with length structures £{ on the Xi which are compatible on the
intersections Xi r\Xj for all i,j £ I, then there obviously exists a (unique)
length structure on X that restricts to £i on each Xi. (In other words,
the length structures comprise a sheaf over X.) On the other hand, metrics
on X are not local (they form only a presheaf over X), but they can be
localized as follows: Given a metric d on X, we consider all metrics d' that
are locally majorized by d. This means that for each x E X, there exists a
neighborhood Y^ C X of x such that d|y^ > d'ly^. Now take the supremum
of all these d' and call it dm- (Note that the supremum of a bounded family
of metrics is again a metric. In general, this supremum may be infinite at
some pairs of points in X, but otherwise it looks like a metric.) Clearly
dm ^ di in any metric space (X, d); if (X, d) is complete^ then dm =" d£, as
a trivial argument shows (see Section 1.8 below).

B. P a t h metric spaces

1.7. Definition: A metric space (X^d) is a path metric space if the

distance between each pair of points equals the infimum of the lengths of
curves joining the points (i.e., ii d = di).

Examples: Note that, according to this definition the Euclidean plane is

a path metric space, but the plane with a segment removed is not.

The n-sphere S^ is not a path metric space when equipped with the
metric induced by that of R"^"^^, but it is a path metric space for the
geodesic metric by Proposition 1.6.

Path metric spaces admit the following simple characterization.

1.8. Theorem: The following properties of a metric space (X, d) are equiv-
1. Length Structures: Path Metric Spaces 7

alent:

1. For arbitrary points x,y ^ X and e > 0, there is a z such that

sup(d(x, z), d{z, y)) < - d{x, y) + e,

2. For arbitrary x^y E X and ri,r2 > 0 with ri + r2 < d{x^y), we have

d{B{x, ri), B{y, r2)) < d{x, y) - ri - r2,

for
d{B^,B2)= inf dix',y').
y' e B2

Every path metric space has these properties, and conversely, if (X, d) is
complete and satisfies (1) or (2), then it is a path metric space.

Proof. Let (X, d) be a complete metric space satisfying condition (1), and
set 5 = d{x,y). Given a sequence (sk) of positive numbers, there is a point
2:1/2 such that max(d(x, 2:1/2), d{zi/2, y)) £ 6/2-\-£iS/2, and points 2:1/4,2:3/4
for which each of the distances d(x, 2:1/4), (^(2:1/4, 2:1/2), d{zi^2i ^3/4)7 <^(^3/4) v)
are less than

l/2{S/2 + EiS/2) + S2{d/2 + siS/2), etc.

By choosing the sequence (sk) so that ^i^Sk < 00, we can define a map /
from the dyadic rationals in [0,1] into X satisfying

If (X, d) is complete, then this map extends to the entire interval [0,1].
Since we can choose the Sk so that the product 11(1 •^" ^A;) is arbitrarily
close to 1, we obtain curves whose lengths tend to 6 = d(x, ?/), which proves
the last assertion.
The implication (1) => (2) is proven in the same way, whereas (2) =^ (1)
and the assertion that a path metric space satisfies (1),(2) are trivial.

Path metric spaces enjoy some of the same geometric properties as Rie-
mannian manifolds.

l.Sbis. Property: If (X, d) is a path metric space, and if / is a map of

X into a metric space Y, then the dilatation of / obviously equals the
supremum of its local dilatation, i.e., dil(/) = sup^^;^ dila;(/). Note that
8 Metric Structures for Riemannian and Non-Riemannian Spaces

if X and Y are Riemannian manifolds, and if / is differentiable, then the

differential Dfx'. T^X -^ Tf(^j.)Y satisfies dila:(/) = ||Z)/a;||.

1.8bis-j_ Kobayashi metrics. Let A be a path metric space and let X

be an arbitrary (say, topological) space with a distinguished set of maps
f:A—^X. Consider all metrics d' on X for which these / have dil(/) < 1,
i.e., for which the mappings are (nonstrictly) distance decreasing, and define
dx as the supremum of the metrics d' on X. Here, it is convenient to
admit degenerate metrics d' (in the sense that d'[x^y) — 0 for perhaps
some X 7^ ?/), so that dK may itself be degenerate. In fact, this dx is a
(possibly degenerate) path metric by the property above.
In the classical example, due to Kobayashi, A is the unit open disk
equipped with the Poincare metric (i.e., the hyperbolic plane), X is a com-
plex analytic space, and the collection of distinguished maps consists of all
holomorphic mappings A —> X. The usefulness of this metric is based on
the Schwarz lemma (and its various generalizations), which implies that
dK is nondegenerate for many X. Such X are said to be (Kobayashi) hy-
perbolic. For example, the disk A is itself hyperbolic since dx in this case
equals the Poincare metric (following from the fact that every holomorphic
map A —)• A is distance decreasing with respect to the Poincare metric, a
consequence of the classical Schwarz-Ahlfors lemma). The basic features of
the Kobayashi metric and hyperbolicity do not depend on the integrability
of the implied (almost) complex structure of X and therefore extend to all
almost complex manifolds X (via the theory of pseudo-holomorphic curves
in X, cf. [McD-Sal]). For example, hyperbolicity is stable under small
(possibly singular) perturbations of almost complex structures on compact
manifolds and (suitably defined) singular almost complex spaces (compare
[Kobay], [Brody], [Krug-0ver]).
There is also a real analog of Kobayashi hyperbolicity, in which X is
a Riemannian manifold, A is as above, and the set of distinguished maps
consists of all conformal, globally area minimizing mappings A —^ X. In
this case, hyperbolicity ofX is equivalent to ^-hyperbolicity (see (e) in 1.19^_
below) under mild restrictions on X, which are satisfied, for example, if X
is the universal cover of a compact manifold. (In fact, X does not have to
be a manifold here — it can be a rather singular space, e.g., a simplicial
polyhedron as in 1.1 S^, see [GroJnG, [GroJHMGA-)

1.9. Definition: A minimizing geodesic in a path metric space {X,d) is

any curve f: I -^ X such that d{f{t)J{t')) = \t - f\ for each t,t' e L A
geodesic in X is any curve f: I -^ X whose restriction to any sufficiently
small subinterval in / is a minimizing geodesic.
1. Length Structures: Path Metric Spaces 9

In this connection, we have the following :

Hopf—Rinow theorem. If (X, d) is a complete, locally compact path met-

ric space, then

1. Closed balls are com,pact, or, equivalently, each hounded, closed do-
main is compact.

2. Each pair of points can he joined hy a minimizing geodesic.

Before turning to the proof of the theorem, we observe that if (X, d) is

a complete, locally compact metric space, then there are many noncompact
balls for the metric d' = inf(l,d).
1.11 Compactness of closed balls. Note that if a is a point in X, then
the ball B{a^r) is by hypothesis compact if r is sufficiently small. We will
first show that if B{a^r) is compact for all r in an interval [0,/o), then
B{a^p) is compact as well.
Let {xn) be a sequence of points in B{a^p). We may suppose that the
distances o?(a,Xn) tend to p; otherwise, there is a ball B{a^r) with r < p
containing infinitely many of the Xn and thus a limit point of the sequence.
Let (sp) be a sequence of positive real numbers tending to zero. By applying
property (2) of Theorem 1.8, we find that for each p, there exists an integer
n{p) such that for each n > n{p), there is a point y^ satisfying

y^ e B{a, p - 2£p) and d{xn, y^) < Sp.

For each p, the sequence (t/^) lies within a compact set; by a diagonal
argument (or since the product of compact sets is compact), it follows
that there is a sequence of integers (uk) such that the subsequence (^/n^)
converges for all p. The sequence (xn^), which is the uniform limit of the
{Vrik)^ is a Cauchy sequence and therefore converges by completeness of X.
By the preceding remarks, the supremum of the r for which S(a, r) is
compact is infinite: if instead it equalled /? < oc, then we could find p' > p
such that B{a,p) would be compact, by using a finite covering of the sphere
5(a, p) by compact balls.

1.12. Existence of a minimizing geodesic joining two arbitrary

points

We first consider the case when X is compact.

Lemma: / / {X, d) is a compact path metric space and a, 6 € X, then there

exists a curve of length d{a^ b) joining a and b.
10 Metric Structures for Riemannian and Non-Riemannian Spaces

Proof. It suffices to consider curves / : [0,1] -^ X which are parametrized

by arc length. From the definition of path metric spaces, if follows that for
each positive integer n, there exists such a curve fn joining a to 6 and having
length less than (i(a, 6) -h 1/n. The set of fn is therefore equicontinuous,
and by Ascoli's theorem, there exists a subsequence fn,, that converges
uniformly to a curve / : [0,1] -^ X. Since the length function £ is lower
semicontinuous, we have

i{f)<\imM£{fn,) = d{a,b).

In the case of a complete, locally compact, but noncompact path metric

space, it suffices to note that the images of the curves fn chosen in the
preceding paragraph lie within the compact ball B{a,2d{a,b)).

1.13. R e m a r k s : (a) In the case of Riemannian manifolds, this proof

fulfills the promise made in the introduction that use would be made only
of the associated length structure.

(6) The equicontinuity argument of Lemma 1.12 also shows that in a com-
pact path metric space, every free homotopy class is represented by a length-
minimizing curve^ and that the minimizing curves are geodesies. Moreover,
if X is a manifold, then for each real r, there is only a finite number of
homotopy classes represented by curves of length less than r (again, it suf-
fices to use Ascoli's theorem and the fact that the homotopy classes are
open subsets of C^{S^,X); cf. [Dieu], p. 188). These results also hold for
homotopy classes of curves based at a point x e X and geodesies based at
X (but not necessarily smooth at x) and will play a key role, particularly
in Chapter 5.

Examples of path metric spaces

1.14. Riemannian manifolds with boundary and subsets of R^ with

smooth boundary. Let X be a domain in R^ with smooth boundary,
equipped with the metric and length structure induced by that of R^, and
let / be the identity map

(X, induced metric) -^ (X, induced length metric).

It is easy to see that if the boundary of X is smooth, then dilx(/) = 1 for

each X G X, and that dil(/) = 1 if and only if X is convex.
1. Length Structures: Path Metric Spaces 11

Distortion.^: More generally, let X be a subset of a path metric space

A and let distort(X) denote the dilatation of the identity map f: X —^ X
with respect to the two induced metrics, i.e.,

d..i s^t o^r/tv(\X ) = s u (length

p - ^ ^ ^ ^dist)IX
- .

Our first observation is the following:

(a) Let X be a compact subset of W^. //distort (X) < | (which means
that every two points in X that lie within a Euclidean distance of d from
one another can be joined by a curve in X of length < dn/2), then X is
simply connected.

Proof. To prove this assertion, we argue by contradiction. Suppose TTI (X) 7^

0 and let a be a nontrivial homotopy class in which there exists a curve
of minimal length among all homotopically nontrivial loops (the existence
of such a is guaranteed by Remark 1.13(b)). Let Y be the image of c
and g : Y -^ Y the identity map of the space Y with its induced length
structure. We claim that dil(^) = dil(/|y).

To prove the claim, let 1/1, yo be two points of Y and fix a parametrization
of Y by arc length, i.e., a map c: [0,£] -^ Y such that c(0) = y^ = c{£) and
2/1 = c{d) for some d E [0, £] such that d < i — d. Then c|[o,d] is the shortest
path joining yo to yi in X. Indeed, if there were a strictly shorter path
c' from yo to t/i, then the two loops obtained by adjoining c' and the two
parts of c defined by the parameters 0 and d would be strictly shorter than
c. Since their product is homotopic to c, however, we could conclude that
one of the two is not homotopic to 0 in X, which contradicts the minimality
of c. Since the path c|[o,d] lies within y , it follows that d is the distance
from yo to yi for the length metrics of X and Y, and that

dil(2/o,i/i)(^) = d i l ( ^ o , 2 / i ) ( / ) -
12 Metric Structures for Riemannian and Non-Riemannian Spaces

Thus, we have dil(^) < 7r/2. Extend c to a periodic function on R and

set r{s) = d{c{s),c{s-\-£/2)), so that the inequaHty r{s) > £/2dil{g) holds.
Set u{s) = {c{s + £/2) — c{s))/r{s). The curve u is differentiable almost
everywhere, and its image lies within the unit sphere of R^. Moreover,
u{s -h ^/2) = —u{s), so that the length of u is at least 27r. Thus,

du 4dil(g)y
ds e ) '
and so £{u) < 4dil(^) < 27r, which is the desired contradiction.

Remark: If di[{d) = 7r/2, and if X is not simply connected, then X

contains a round circle.

(6) //distort(X) < 7T/2y/2, then X is contractible.

Proof^_ . The idea is to homotopy-retract R" to X by following the flow of

a suitable vector field d which plays the role of — grad d{y) for the distance
function d\ y \^ d\st{y^X) = inf^^^x \\y — ^||- In general, the function d is
nonsmooth, even on the complement R^ \ X. Nonsmoothness at a point
y ^ W^ \X \s due to the fact that the sphere Sy~^ at y of radius d{y)
can meet X at several points. These points essentially realize the above
infimum, since the open ball bounded by Sy~^ does not intersect X, while
the set XnSy'^ is nonempty and d{y) = \\x-y\\ for all x G XnSy~'^. Now
we observe that the normal projection X —^ ^y~^ is distance-decreasing and
thus (length dist)j^(a:i,X2) > (length dist)^n-i (2:1,3:2) for all pairs of points
in the intersection X D Sy~^. It follows that the latter distance is bounded
by 7rd{y)/2^ and since we assume the strict inequality distort(X) < 7r/2v^,
the distance above is bounded by S7rd{y)/2 for some S < 1 independent
of y. Consequently, the intersection X n Sy~^ is strictly contained in a
hemisphere, or, in other words, there exists a unit vector dy at every y G
R^ \ X such that

dy\\x-y\\<-e<0 for all x G X 0 S^-\ (*)

where dy\\x — y\\ denotes the dyderivative of the (distance) function y t-^
\\x — y\\. In fact, one can take dy to be the vector which points towards
center of the minimal spherical cap in Sy~^ containing X fi Sy~^, thus
obtaining a (Borel) measurable vector field y ^-^ dy satisfying (*).
Finally, we can easily smooth this vector field, so that the resulting (now
smooth!) unit vector field, say y ^-^ dy on W^ \ X, satisfies (*) (possibly
with a smaller £ > 0) as well. Clearly, every forward orbit of such a field
1. Length Structures: Path Metric Spaces 13

converges to a point in X, and so the flow generated by d eventually retracts

M^ to X.

Remark^: A sharper result is proved in Appendix A, where one allows

distort(X) < l-\-an for some specific a^ > 1 —n/2y/2. Furthermore, there
are many examples of (necessarily contractible) subsets X with arbitrarily
large (even infinite) distortion for which X nSy~^ is still strictly contained
in an open hemisphere for each y e W^ \ X and to which our argument
applies. On the other hand, we do not know the precise value of an for
which distort(J^) < 1 -f a^^ necessarily implies that X is contractible.

Exercise^: Construct a closed, convex surface X in R^ with distort(X) <

7r/2. (Compare Appendix A.)

Problem^: Given a topological space X, evaluate the infimum of all dis-

tortions induced by embeddings X -^ M"" or of those distortions induced
by embeddings which lie in a fixed isotopy class. The first interesting case
arises when X is the circle and we minimize the distortion for X knotted
in M^ in a prescribed way (compare [Gro]HED» [O'HarajEK)-
Remark^: The geometry of subsets X C R^ satisfying distort (X) < l-f-a^
can be rather complicated, even for small a^^ > 0. For example, there are
simple smooth arcs in R^ with arbitrarily small (i.e. close to 1) distortion
which have an arbitrarily large turn of the tangent direction. To see this,
consider diffeomorphisms T^: R^ —» R^ with the following properties: Each
Ti fixes the complement of the disk of radius 2""^ around the origin and
isometrically maps the disk of radius 2~^~-^ into itself by rotating it by a
small angle a > 0.
Clearly, one can choose the Ti so that they and their inverses are (1+6:)-
14 Metric Structures for Riemannian and Non-Riemannian Spaces

Lipschitz for £ < 10a. Since the distortion of such a Ti is located in the
annulus between 2~* and 2~^"^^, the diffeomorphism given by the composi-
tion Ui = Ti 0T2 o' • • oTi are also (1 -h^) Lipschitz along with their inverses
Ui, since the "distortion supports" of the Ti are mutually disjoint. Note,
however, that for large z, the map Ui is the identity outside of the unit disk
and rotates the 2~*~-^ disk by the angle za, which can be quite large for
large i. Also, the J/j-image of the unit segment in R^ passing through the
origin may have arbitrarily large rotation of the tangent direction, while its
distortion does not exceed 1 + £.

The full richness of this picture becomes apparent in Appendix B_^.

Asymptotic distortion.^. Define
distort(X; D) = supdist(xi,X2),
where the supremum ranges over all xi,X2 £ X C A with dist^(xi,X2) >
D with respect to the induced length metric, and study the asymptotic
behavior of this distortion as Z? —^ 00. For example, if X is a properly
embedded, noncompact submanifold in hyperbolic space H^, then either
distort(X) < 00 or distort(X; D) grows exponentially as D -^ 00. On the
other hand, the distortion of every connected subgroup X of a Lie group A
is at most exponential. This is easy to see for A = GLn and for algebraic
groups in general (see [GrojAi); the (more difficult) nonalgebraic case was
recently settled by Varopoulos (see [Var]).

1.15+ Polyhedral metrics. Assign a length structure to the standard

n-simplex

n+l
A"" = { x o , x i , . . . .Xn > 0 : ^ x ^^i —= -1}
L; C
^ ii^

i=0
1. Length Structures: Path Metric Spaces 15

that is invariant under the automorphisms of A^ corresponding to permu-

tations of the vertices of A^ (labelled by the indices 0 , 1 , . . . ,n). Suppose
this is done coherently for all A^, for n = 1,2,... , i.e., so that the chosen
length structure on A^~-^ coincides with the length structure induced by the
standard (face) embedding A^~-^ C A " from the chosen length structure
on A^. For example, we can take the Euclidean structure on A'^ C M^"^^,
or the spherical structure induced by the radial projection A'^ —> S^. Fur-
thermore, we can modify the ambient metric in R^"^-^, e.g., by using the
^i-norm ||a:||^^ = J^^^Q 1^*1' ^-"^ we can identify A " with the regular simplex
of diameter d in hyperbolic n-space for a given d G [0, oo].
Once the length structure on A^ has been chosen, every simplicial poly-
hedron K acquires a unique length structure which restricts to this struc-
ture on all of its simplices (compare 1.6^ ). The resulting path metric
spaces exhibit an unexpected richness of geometry. Even in the simplest
case where i^ is a tree (i.e., a contractible 1-complex), the picture is not
quite trivial (just try to figure out when a 4 point metric space admits an
isometric embedding into such a tree). And if we allow infinite-dimensional
polyhedra K^ then we will be able to apply metric treatment to arbitrary
homotopy types (of spaces) via their (semi-) simplicial models (see 5.42).

1.15^ Semialgebraic sets. A subset X C R"^ is called (real) algebraic if

it is defined via some polynomials P^, 2 = 1 , . . . , / , on R^ as follows:

X = {xeW : Pi{x) = 0 for i = 1 , . . . , / } .

We call X special semialgebraic if, in addition to the polynomial equations

above, we allow finitely many strict and nonstrict polynomial inequalities:

X = {xeW : Pi{x) = 0]Qj{x) > 0;Rk{x) > 0}.

The basic measure of the geometric complexity of such X is the sum of the
degrees of these polynomials:

D{Pi, Qj.Rk) = ^ deg(P,) -f J2 deg(Qi) + ^ deg{Rk),

i j k

or, more precisely, the infimum algdeg(X) of the numbers D{Pi,Qj^Rk),

where the P^, Qj.Rk range over all polynomials which represent a given X as
above (the representation of X by P^, Qj^Rk is highly nonunique. Note, for
instance, that X remains the same when we replace the given polynomials
by their cubes P^.Q^^Rl). Finally, semialgebraic X are defined as a finite
union of special semialgebraic subsets.
It is not hard to see that the induced length structure on X C R^ is
rather regular (albeit far from fully understood). In particular, the length
16 Metric Structures for Riemannian and Non-Riemannian Spaces

metric induces the same topology on (each connected component of) X

as the induced metric. In fact, if X is an algebraic curve of degree d,
then it meets each hyperplane in at most d points, and so the length of
the intersection of X with a ball of radius R does not exceed 2Rd by the
Crofton formula. This gives a rough idea of the length metric on such X
when we recall (the well known fact) that for each x G X, there exists
R > 0 such that the intersection of X with the Euclidean r-ball around
X is connected for all r < R. (In fact, X H B{x,r) is contractihle for all
semialgebraic X and r < R = R{x) > 0.) This implies that

lengthdist(x,2/) < 2ddist{x,y)

for every x e X and all y e X H B{x,r). Moreover, by projecting a

semialgebraic curve X onto lines in R^ and then isolating "branches" of X,
one can improve the above estimate to

lengthdist(x, y) < (1 -f s) dist(x, y), {<)s

where s = e{x^r) -^ 0 as r —> 0. (Probably {<)s holds for all semialgebraic

sets X.) Note that the convergence s —^ 0 might be nonuniform in x,
as is seen in the case when X consists of two smooth curves which are
tangent at a single point. But, one can show that every bounded, connected
semialgebraic set X satisfies

lengthdist(x, y) < Cdist^{x^y)^ (^)a

for some C = C{X) and a = a(algdeg(X)) > 0. This is weaker than (<)£
for individual x, but it has the advantage of being satisfied by all x and
y, with C independent of these points. (See [Grojvom? [Grojscss for more
information and references about the geometry of semialgebraic sets)

1.16+ Quotient spaces X/T. If (X, dx) is a metric space and f: X -^ Y

is a surjective map, then Y can be equipped with the quotient metric d^
equal to the supremum of those d' on Y for which the map / is (nonstrictly)
distance decreasing (i.e., with dil(/) < 1). Notice that the metric d may
degenerate (as did the Kobayashi metrics), but if y = X/T for a group F
that acts isometrically on X and that has closed orbits, then d is nondegen-
erate. (Here, the distance between the orbits yi = T{xi) and i/2 = r(x2)
clearly satisfies d{yi,y2) = inf^^r d{xi,^{x2)).) Hdx is a path metric, then
so is d^ since its construction is local. In fact, the length structure in Y can
sometimes be recaptured (but not always!) by lifting curves c from F to X
and then defining lengthy (c) as the infimum of the X-lengths of all such
lifts. Ideally, we would like to have horizontal lifts c of curves c in F to X
1, Length Structures: Path Metric Spaces 17

so that lengthj5^(c) = lengthy (c), and this is possible, for example, when
X is a Riemannian manifold and Y = X/T. Here, we declare a smooth
curve c in X to be horizontal if it is normal to the orbits of F (which is no
restriction at all if F is discrete) and introduce the length structure in Y
using the projection c of these horizontal ctoX, where we define
lengthy (c) = length;^ (c).
Then the metric in Y associated with this length structure equals the quo-
tient metric c? in F , as easily follows from the slice theorem for isometry
groups. Note that this Y is singular unless the action of F is free. This
remark will play an important role in Chapters 3 and 8.

1.17_j_ Covering metrics. If X -^ X is a, covering map, then length

structures, being local, lift from X to X, as do path metrics. In fact, every
path metric d on X lifts to a unique path metric d for which the covering
map is a local isometry.
Although the definition of d is obvious, the actual determination of its
essential properties can become insurmountably difficult. For example, if
X is a compact Riemannian manifold or a finite simplicial polyhedron, then
all we may care to ask about d stands clearly before our eyes. But if we pose
some seemingly inocuous question about d on the universal cover X ^> X,
e.g., whether X has finite or infinite diameter, then we encounter an (al-
gorithmically) unsolvable decision problem of finiteness of the fundamental
group of X . Here, the logical complexity of the problem is ingrained in the
very definition of the path metric, the infimum of length over all curves be-
tween given points. This ''all" makes the passage length—^distance highly
non-effective! Thus, complete knowledge about "length" does not immedi-
ately translate into comparable information about "distance," even in such
simple cases as a Lie group X equipped with a left invariant Riemannian
metric d. Here d = dg is uniquely determined by the prescription of a
positive definite quadratic form ^ on a single tangent space, TgX, but eval-
uating dg{xi^X2) for given Xi and X2 may be diflScult. (To see the point,
take X = GL^R C R^ and measure the c?o-distance between the matrices
id and — id in terms of g on T[^X = R^ , or try to estimate the diameter
of X = (50(n), dg) in terms of g.)

1.18. Carnot—Caratheodory metrics associated with a tangent sub-

bundle of a Riemannian manifold. One can show that the length
structure associated with a completely nonintegrable subbundle (cf. 1.4(e))
— as in the case of the Heisenberg group — gives rise to a distance on V
that induces the same topology but radically alters the metric properties
(e.g., Hausdorff dimension, etc.) of V.
18 Metric Structures for Riemannian and Non-Riemannlan Spaces

1.19+ Curvature. It is hardly possible to find a convincing definition of the

curvature (tensor) for an arbitrary metric space X, but one can distinguish
certain classes of metric spaces corresponding to Riemannian manifolds
with curvatures of a given type. This can be done, for example, by imposing
inequalities between mutual distances of finite configurations of points in X.
More precisely, let Mr denote the space of positive symmetric rxr matrices,
and let Kr{X) C Mr be the subset realizable by the distances among r-
tuples of points in X. Thus, (m^j) G Kr{X) if and only if there exist points
x i , . . . ,Xr in X, such that dist{xi^Xj) = rriij^ for i^j = 1 , . . . ,r. Then
every subset /C C Mr defines the (global) JC-curvature class, which consists
of the spaces X with Kr{X) C /C, and the local JC-curvature class, where
each point x G X is required to admit a neighborhood U with Kr[U) C /C.
In this setting, we have the basic

Curvature problem: Given K C Mr^ describe the spaces X in the (local

or global) AT-curvature class.

In general, a /C-condition on X does not seem to lead to an interesting

theory unless one exercises good judgement in choosing the set /C. Geome-
ters usually start with some standard (model) space Y (or with a family of
such spaces) and take K = Kr{Y) (or the union of these for a given family
of spaces Y). If one is willing to sacrifice the combinatorial aspect of the
problem, then the phenomena associated with discrete spaces can be ruled
out by imposing various connectivity restrictions on X, such as local path
connectedness, local contract ability, controlled contract ability, etc. In par-
ticular, one can require X to be a path metric space, in which case one can
go amazingly far with a judiciously chosen /C. But it is difficult to decide
by looking at a given /C whether the corresponding /C-class is "interesting."
For instance, when does there exist a path metric space X with Kr{X) = /C
for a given JC C Mr^

Examples: (a) If X is compact and connected, then (obviously) the only

information contained in K2{X) is the diameter of X, i.e., supdist(x 1,^2),
where the supremum ranges over all xi,X2 G X.

(6) If r = 3, then for any metric space X, the triangle inequality requires
Ks{X) to lie within the subset Tri C M3 consisting of matrices {rriij) that
satisfy rriij < rriik + rrikj for all i,j,k running through 1,2,3. It seems
that most path metric spaces X with infinite diameter satisfy Ks{X) = Tri
(with some notable exceptions such as X = R or R x compact), but I do
not know a good criterion that guarantees this property. For example, all
uniformly contractible manifolds X of dimension > 2 have large Ks{X),
probably always equal to Tri.
i. Length Structures: Path Metric Spaces 19

(c) The curvature story starts (and essentially ends, in the present state
of the art) with r = 4. For example, K4 characterizes Euclidean spaces as
follows:

If a complete path metric space X has K4{X) C K4(R'^), then it is isometric

to a convex subset in M^. If K4{X) C K4(R^), then X is isometric to a
convex subset in a Hilbert space.

The proof is straightforward and easily extends to complete simply con-

nected spaces Y with constant curvature in place of R"^ (see [Rinow] §41).
However, a similar characterization has not been worked out for other Rie-
mannian symmetric spaces F , where one expects every complete path met-
ric space X with Kr{X) C KriX) for some (sufficiently large) r = r{Y) to
be (isometric to) a convex subset in another (possibly infinite dimensional)
symmetric space Y' that naturally extends the geometry of Y (as, for ex-
ample, the Grassmannian manifold Grfc(M^) extends Grfc(R")). On the
other hand, it is easy to see that (convex subsets in) normed vector spaces
are characterized by X5, since the homogeneity of a norm is a five point
property.

{d) Alexandrov spaces with K > 0: One could define this class by taking
the union of K4{Y) C M4 for all convex complete surfaces F C M^ equipped
with their induced path metrics. Customarily, however, one uses a some-
what larger /C(> 0) C M4, where the /C-positivity condition is felt only by
those quadruples of points {xi} C X, i = 1 , . . . ,4, with the property that
X4 lies between xi and X3, i.e., X4 satisfies

dist(x4, X2) + dist(x4, X3) = dist(x2,0:3). (A)

In other words, one defines if > 0 by requiring that every 1-Lipschitz map
/ : {xi,X2,X3} —> R^ extend to a 1-Lipschitz map {xi,X2,X3,X4} —• R^
for all quadruples {xi\ in X satisfying (A), where "1-Lipschitz" means
dil(/) < 1 (see [Alex], [Rinow]). The resulting relation /C(> 0) C M4,
called the Alexandrov-Topogonov comparison inequality for K > 0^ is quite
resilient, e.g., locally K > 0 implies globally K >0 (the proof is nontrivial,
see [Top], [Bu-Gr-Per]), isometric quotients X/T inherit K > 0 from X
(this is straightforward), etc., and thus leads to a full-fledged geometric
theory without any regularity assumptions on X (see [Per]scBB and the
references therein).
The Alexandrov-Topogonov inequality can be thought of as a kind of con-
cavity condition on the distance function x 1-^ dist(xo,x) that rules out,
for example, all non-Euchdean normed spaces (see [Rinow], p. 33). These,
20 Metric Structures for Riemannian and Non-Riemannian Spaces

however, reside within a larger class of path metric spaces X having a less
stringent concavity condition on dist. This condition can be described, for
example, by requiring that inward equidistant deformations preserve con-
vexity of hypersurfaces in X. However, the corresponding theory (starting
from the K^ or KQ characterization of this property) has not yet been
developed.
The K > 0 theory generalizes to K > K for given /c G M, but other differ-
ential geometric curvature positivity conditions have not been extended to
singular spaces so far. One might expect that the positivity of the curva-
ture operator and that of the complexified sectional curvature (especially
on the isotropic planes) admit Kr renditions with some not very large r.
On the other hand, the curvature classes Ricci > 0 and Sc > 0 lie beyond
the reach of the Kr language. (See [Grojsign, [Gro]pcMD for a survey of
various positive curvatures; also see Chapters 5 and 8 of this book.)

(e) Alexandrov spaces with K < 0: These spaces are defined via a suitable
^ ( ^ 0) C M4 that expresses the Cartan-Alexandrov-Topogonov compar-
ison inequality for K <^. (Here, the test spaces Y are convex space-like
surfaces in the light cone of the Lorentz (-h H—)-space.) The singular path
spaces with K < 0 (also called CAT(O)-spaces) are by far more numerous
than those with K > 0. For example, simplicial (as well as nonsimplicial)
trees equipped with length metrics and cartesian products of these spaces
have K <Q. In fact, there are plenty of polyhedra with K <^ locally at
each point, and in practice, the inequality K <^ can be confirmed by an
elementary combinatorial argument. On the other hand, the local condition
K < 0 implies strong global properties of X, such as its asphericity (a nice
property that is hard to prove using traditional topology). In fact, those
polyhedra with K <0 are the core of the geometro-combinatorial approach
to infinite groups initiated by Dehn at the turn of the century.
Following Busemann [Busemann], one can extend the notion K < 0 to
encompass normed linear spaces. The enlarged class is distinguished by the
(local) convexity of the distance function, x 1—> dist(xo, x) on every geodesic
segment in X, or by the property that the convexity of hypersurfaces is
preserved under outward equidistant deformations (compare [GroJHMGA,
[E-H-S]).
Another notion (generalizing K < -K < 0, rather than iC < 0) is as follows:
Consider all simplicial trees Y and let Tre C M4 be the union of if4(F)
over all trees Y. Then take the (5-neighborhood Ties C M4 of Tre for some
(5 > 0 (defined with respect to the obvious Euchdean metric in M4) and
introduce S-hyperbolic spaces X as those with K4{X) C Tre<5. Here, the
1. Length Structures: Path Metric Spaces 21

essential geometry appears at large scales » 6, where the behavior of (5-

hyperbolic spaces is indistinguishable from those with K <0 (see [GrojHc)-
The traditional definition of /C(< 0) C M4 is dual to that for given /C(> 0)
since it involves the extendability of 1-Lipschitz maps from R^ into X.
Namely, for each quadruple {xi} C M^ satisfying equation (A) in {d) above,
every 1-Lipschitz map {xi, X2,0:3} -^ X should extend to a 1-Lipschitz map
{xi} -^ X, z = 1 , . . . ,4.

Exercise: Let X^ satisfy the Alexandrov-Topogonov condition K < 0,

and let Z C X^ be the union of the minimizing geodesic segments whose
ends comprise a three point subset {xi,X2,a:3} C X+. Show that every
1-Lipschitz map {xi,X2,xs} -^ X admits a 1-Lipschitz extension to Z.

Kirszbraun—Lang—Schroeder^ Theorem: Let X+ and X- be path met-

ric spaces having K > 0 and K < 0 respectively in the Alexandrov sense.
Then, every l-Lipschitz map X'_^ —^ X- for X'^ C X^ admits a 1-Lipschitz
extension to all of X^.

If X+ = X- = R^, this is a classical result (see [Kirsz], [FedJcMT? [GrojMv),

while the general case is quite recent (see [Lan-Sch], where the authors treat
the case when K{X^) > K> K{X-) for all K G R). Notice that the Eu-
clidean (and the spherical) case of this theorem follows from monotonicity
of the volume of intersections of balls in R'^, i.e., VO1(JB(XO, ro)nB{xi,ri) D
• • • nB{xn,rn)) is monotone decreasing in the distances \\xi — Xj\\ between
the centers of the balls (which was probably known in antiquity). Also
observe the following equivalent version of the Kirszbraun theorem.

(•) Let f: A —^ S^ be a X-Lipschitz map from an arbitrary subset A C S^

where A < 1. Then f{A) is strictly contained in a hemisphere of diameter
< 7T — £ < 7T. (If X = 1 and f{A) is not contained in a hemisphere, then f
is isometric.)

Exercises: Prove the monotonicity claim above and derive Kirszbraun's

theorem from it. Derive (•) from Kirszbraun's theorem, and conversely
show that (•) implies Kirszbraun's theorem by studying optimal A-Lipschitz
extensions of 1-Lipschitz (partially defined) maps with minimal possible A.
(A reader versed in K >, < 0 geometry may notice that the Lang-Schroeder

^When I met Petrunin in Spring 96, I asked him whether the Kirszbraun theorem
extended to Alexandrov spaces. He responded with "yes, for dim < 00," hardly hiding
his surprise at my inability to furnish a proof by myself. The next day, I found the
preprint by Lang and Schroeder with a complete proof of the most general case, and
similar ideas were communicated to me somewhat earlier by Branka Pavlovic.
22 Metric Structures for Riemannian and Non-Riemannian Spaces

theorem for (smooth!) Riemannian manifolds also follows from (•) via the
study of optimal A-Lipschitz maps).

Remark: It is unclear which spaces Y (say with K{Y) < 0) admit Lipschitz
extensions of maps from ^4 —^ y to X for arbitrary metric spaces X and
subsets A C X, where the implied Lipschitz constant of the extension
X —> Y is allowed to be bigger than that for the original map A —^ Y.
It is rather obvious that F = R has this property (with Lip |R = Lip | ^ ) ,
and consequently this is true for R^ (with Lip \^n < Sn Lip |A). The same
applies to trees (in place of R) and their products. The question is open,
however, in the case of the hyperbolic plane,^ for example (see [Ball] for
some results in this direction).

(/) Finally, we mention the possibility of having /C^ C Mr depend on x €

X and then consider the (variable) local /C^;-curvature class at each x e X.
This notion captures, for example, the idea of local pinching, which requires
that the sectional curvatures of X at each point x be pinched between K{x)
and KK{X), for some positive (or negative) functions K{x) on X and a local
pinching constant K > 0. (This also seems to be the appropriate framework
for the Kr-rendition of the positivity of the curvature operator and the
complexified sectional curvatures mentioned earlier).

D. Arc-wise isometries

1.20. The requirements for a map to be an isometry or even local isometry

are too stringent to provide a sufficiently rich class of morphisms for path
metric spaces. For example, any Riemannian n-manifold that is locally
isometric to R^ must be flat. A more flexible notion is the following.

1.21. Definition: An arc-wise isometry of path metric spaces X,Y is a

map f:X-^Y such that i{f o c) = £{c) for each Lipschitz curve c: I -^ X
oiX.

Examples:

(1) Every closed, piecewise C^ curve admits an arc-wise isometric mapping

into R.

^This was recently settled by Branka Pavlovic.

1. Length Structures: Path Metric Spaces 23

(2) Every flat manifold of dimension n < 5 admits an arc-wise isometry

into R^ that is additionally piecewise linear (see [Zalg]). The question
is open for n > 5.

(3) An intuitively plausible property, such as the nonexistence of arc-

wise isometrics from X to y when dim(X) > dim(F), is trivial to
prove for C^ maps but less transparent in general. The proof re-
quires Rademacher's theorem (cf. [FedJcMT 3.1.6), which states that
Lipschitz mappings are differentiable almost everywhere.

We end this chapter with a few results obtained using the methods of
Nash and Kuiper.

1.22. IfX^Y are Riemannian manifolds with dim(y) > dim(X), then there
exists an arc-wise isometry from X into Y.
In particular, incredible as it seems, every n-dimensional Riemannian man-
ifold X admits an arc-wise isometry X -^ W^. Of course, such mappings
will not be C^ in general!

1.23 A n approximation problem. Given a Lipschitz mapping / Q : X —>

Y and £ > 0, does there exist an arc-wise isometry f^: X —^ Y such that

d{fo, fe) = s u p d{fo{x), fe[x)) < sl

xex
Evidently, we must have dil(/o) < 1.
1.24. Definition: A mapping / between path metric spaces is called
short if dil(/) < 1 and strictly short if dil(/) < 1.
1.25. T h e o r e m (cf. [GroJpDR): If X^Y are Riemannian manifolds with
dim(y) < dim(X)^ and if f is a strictly short mapping from X into Y,
then the approximation problem has a positive solution.

This brings us back to the basic problem of the existence of a homotopy

of a given continuous map to a short map which is addressed in the following
chapter.

1.25^ . If one feels disgusted by the spineless flexibility of arc-wise iso-

metric maps, then some rigidifying conditions can be added so that the
remaining mappings / in question have no folds. (The construction under-
lying the proof of 1.25 uses multiply folded maps X -^ Y.) For example,
one may insist that / : X ^ y be open (i.e., that / map open subsets of X
onto open subsets of y ) , which is quite a strong assumption for Lipschitz
24 Metric Structures for Riemannian and Non-Riemannian Spaces

(and especially arc-wise isometric) maps. Secondly, if X and Y are oriented

manifolds of the same dimension, then one may require the local degree of
/ to be positive. In the present setting, this means (usually one uses a
stronger condition locdeg^ > 0, x G X) that whenever / properly maps
some open subset U C X onto a subset V C Y, the degree of f: U -^ V
as defined in 2.A is > 0. Thirdly, one may express the same idea by re-
quiring that the inequality Jacobian(/)a: > 0 holds for almost all points
X €: X (which makes sense for Lipschitz maps). Finally, one can introduce
co-Lipschitz maps (compare [B-J-L-P-S] as follows: Let

COdlK/) = sup ; r ,
yiL dist(2/i,2/2)
where (2/1,2/2) runs over all pairs of distinct points in Y and dist// denotes
the Hausdorff distance (see Section 3.B). If codil(/) < A, then / is called
A-CO Lipschitz. The quotient map X —^ Y = X/T for a proper isometry
group r acting on X is 1-co-Lipschitz (as well as 1-Lipschitz), for example.
Next, consider the i?-balls Bx{R) in X around some point x e X and
let r be the radius of the maximal ball in Y centered at f{x) and contained
in the image /(B^(ii)) C Y. We define

codila;(/) = limsup —

and observe that if y is a path-metric space and X is locally compact (which

can often be relaxed), then

codil(/) = supcodila:(/).
xex

In fact, if codila:(/) < c < 00, then every path p: [0,1] ^^Y can be covered
by a path p: [0,1] -^ Y issuing from a given point y £ Y over p{0) e X.
Furthermore, if p is 1-Lipschitz, then p can be chosen to be c-Lipschitz.

Exercises:

(a) Show that if X is complete and f: X -^ Y is co-Lipschitz, then Y is

also complete.

(6) Study the following implications between the preceding four condi-
tions for Lipschitz maps between suitably oriented equidimensional
Riemannian manifolds:

co-Lipschitz => Jacob > 0 <=> locdeg > 0 =^ open.

1. Length Structures: Path Metric Spaces 25

(c) Decide which of these four conditions ensure that arc-wise isometric
maps are local isometrics. (I did not solve (b) and (c) myself, but I
have prepared a good excuse for this.)

(d) Compare "Lipschitz + co-Lipschitz" with quasiregular maps (see Sec-

tion 6. A).

1.26-|_ Historical and terminological remarks. The idea of length

structures and path metric spaces goes back at least as far as Gauss, who
studied how the bending of surfaces in R^ distorts the induced metric but
preserves the length structure. In classical geometry, one distinguishes the
induced path metric with the name inner metric, and the word "inner"
is often used for general length spaces as well. Traditionally, geometers
were most interested in those "inner metric spaces" that are similar to Rie-
mannian manifolds, and much effort was spent trying to identify and study
such spaces with inner (i.e., path) metrics (see [Busemann], [Rinow], [Alex],
[Alex-Zalg]).
In this book, we emphasize a different point of view, in which the
Riemannian idea is stretched to its limits in order to capture such non-
Riemannian beasts as Carnot-Caratheodory spaces (also called "nonholo-
nomic spaces" and "sub-Riemannian" or "subelliptic" manifolds) and trees
with infinite branching at every (!) point. In fact, one should not even
restrict attention to the length structure, especially if one is attracted by
fractal spaces (one can also play around with area instead of length, as in-
spired by symplectic geometry, string theory, and/or the area isoperimetric
inequalities in geometric group theory).
Finally, we call the reader's attention to the paper [Bing], where it is
shown that every compact, connected and locally connected metrizable space
admits a path metric.
Chapter 2

Degree and Dilatation

A. Topological review
Throughout this chapter, M and N will denote connected, oriented,
C^ manifolds having the same dimension n. Additionally, M is assumed
to be compact and without boundary.

2.1. Proposition: Let f: M -^ N be a C^ map, so that there exist regular

values of f, i.e., y E N such that for each x E f~^{y) the differential Dfx
has rank n. For each such y E N, the set f~^{y) is finite, and if we set
o{x) = 1 if Dfx preserves orientation and o{x) = — 1 otherwise, then the
number
deg(/, y)= ^ o{x)
xef-Hy)

does not depend on the regular value y.

(See [MilnorJTDV or [Berger]cours, Ch. 7).

2.2. Definition: The degree deg(/) of a smooth map f: M -^ N is the
number deg(/, y), for any regular value y oi f.

2.3. Proposition (see [Milnor]TDv)« If f,g: M -^ N are homotopic,

thendeg{f) = deg{g).

2.4. Definition: If / : M -^ iV is continuous, then the degree of / is

defined as the degree of any smooth map homotopic to / .

2.5. Remark: If some y E N does not lie in the image of a continuous

map f:M-^N, then deg(/) = 0. Note that if ^ is smooth and suflaciently
28 Metric Structures for Riemannian and Non-Riemannian Spaces

approximates / , then g is homotopic to / , and y does not lie in the image

of g. Thus, 2/ is a regular value of g, and deg(/) = deg{g) = deg(^, y) = 0.

2.6. Proposition: The integration of n-forms defines an isomorphism

H^{M]W) -^ R. Moreover, the condition that Jj^,j<^ = 1 determines a
unique class UJM ^ J?^(M;R); similarly, if N is compact, there exists a
corresponding class tuj^ ^ H^{N;M). If f : M -^ N is continuous, then
f*(jON is proportional to UJM, ^^C? the proportionality constant equals the
degree of f defined above. (See [Massey] and [Godb], p. 212).

2.7. Remark: If M is a compact, oriented Riemannian manifold, then

its volume form is defined as the unique differential n-form VM , which for
every x e M has value 1 on each oriented, orthonormal frame in T^M. The
volume of M equals the integral vol(M) = J^^ UM- If N is another compact,
oriented Riemannian manifold with volume form z/^v, and / : M -^ iV is
smooth, then deg(/) = {YO\{N))~^ J^j f^u^. At each point m G M, the
skew-symmetric 77,-form f^i^N is proportional to I/M by a factor known as
the Jacobian of / at the point m, denoted J ( / , m). If / is a diffeomorphism
from a subset ^ of M onto a subset B of N, then by the change-of-variables
formula, we have

/ /*i^iv = / J{f,rn)uM= / z/iv=vol(B),

JA JA JB
so that we are justified in setting

voi(/u) ^= I r^N.
JA
even if / is not a diffeomorphism. In this notation, we have

deg(/)=vol(/|M)/vol(iV),

where
VO1(/)1I'VO1(/|M).

In general, the degree of a map from M to N does not completely

determine its homotopy class. For example, the mapping S^ xS^ —» 5^ x S^
given by (x^y) 1—> (1,2/) is not surjective and therefore has zero degree,
although the map is not homotopic to 0. Homotopy classes of mappings
into spheres, however, are determined by their degree (see [MilnorJTDV? P-
51).

2.8. Theorem (H. Hopf): If M is a compact, oriented n-manifold, then

two mappings M —^ S^ are homotopic if and only if they have the same
degree.
2. Degree and Dilatation 29

2.8^ Remarks about degree.

(a) The pointwise degree deg(/, y) can also be defined for open and/or dis-
connected manifolds M and AT; all one needs is the finiteness of the pullback
f~^{y) C M. This finiteness condition is (obviously) satisfied for regular
values y of proper maps / , i.e., those / for which f~^{Y) C M is compact
for all compact subsets Y C N (when M and A^ are allowed to have bound-
aries 9M, dN^ respectively, the definition of properness usually includes the
requirement that f{dM) C dN). If / is proper^, then deg(/, y) is locally
constant on each connected component of N. In particular, deg(/, y) is in-
dependent of y if A^ is connected, in which case it is regarded as deg(/). This
number does not change under homotopies of proper maps and therefore
extends to all continuous proper maps f:M—^N (compare [MilnorJxDv)-
If M is an open connected manifold without boundary, then the degree of
proper maps f:M—^W^, denoted f H^ deg(/) G Z, establishes a bisection
between the homotopy classes of such maps and Z, provided that n > 2,
by a (trivial) modification of Hopf's theorem. (If n = 1, then every proper
map M -^ M has degree ± 1 for connected M, which must be homeomorphic
toM.)

(6) The notion of degree extends to certain nonproper maps f: M —^ N

such as:

(i) Maps / that are constant at infinity and on dM, i.e., constant outside
a compact subset K C Int(M).
(ii) If dim(M) = dim(A/") > 2, one can also allow maps M -^ N which
are locally constant outside some compact subset K C Int(M).
(Hi) Maps / which send lnt{M)\K onto a subset of topological codimen-
sion > 2 in A^, i.e., those having rank(/|aM) < d i m ^ M for compact
M.

In all of these cases, deg(/) is homotopy invariant and is defined for the
continuous maps satisfying (z), (zz), or {Hi). Such maps also satisfy the
conclusion of Hopf's theorem. For example, if M is compact and connected,
then the homotopy classes of maps / : (M^dM) —> (S^.SQ) bijectively
correspond with Z via / ^^ deg(/).

(c) Let A^ be a Riemannian manifold with finite total volume, vol(A^) <
cxD, and let / : M ^ A^ be a smooth map with

< oo,
JM
M
 Another Random Document on
Scribd Without Any Related Topics
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com