Introduction To Riemannian Manifolds

Graduate Texts in Mathematics
John M. Lee
Introduction to
Riemannian
Manifolds
Second Edition
Graduate Texts in Mathematics 176
Series Editors:
Sheldon Axler
San Francisco State University, San Francisco, CA, USA
Kenneth Ribet
University of California, Berkeley, CA, USA
Advisory Board:
Alejandro Adem, University of British Columbia

David Eisenbud, University of California, Berkeley & MSRI
Brian C. Hall, University of Notre Dame
J.F. Jardine, University of Western Ontario
Jeffrey C. Lagarias, University of Michigan
Ken Ono, Emory University
Jeremy Quastel, University of Toronto
Fadil Santosa, University of Minnesota
Barry Simon, California Institute of Technology
Ravi Vakil, Stanford University
Steven H. Weintraub, Lehigh University
Graduate Texts in Mathematics bridge the gap between passive study and
creative understanding, offering graduate-level introductions to advanced topics in
mathematics. The volumes are carefully written as teaching aids and highlight
characteristic features of the theory. Although these books are frequently used as
textbooks in graduate courses, they are also suitable for individual study.
More information about this series at http://www.springer.com/series/136

John M. Lee
Introduction to
Riemannian Manifolds
Second Edition
123
John M. Lee
Department of Mathematics
University of Washington
Seattle, WA, USA
ISSN 0072-5285 ISSN 2197-5612 (electronic)

ISBN 978-3-319-91754-2 ISBN 978-3-319-91755-9 (eBook)
https://doi.org/10.1007/978-3-319-91755-9
Library of Congress Control Number: 2018943719
Mathematics Subject Classification (2010): 53-01, 53C20, 53B20
Originally published with title “Riemannian Manifolds: An Introduction to Curvature”

1st edition: © Springer-Verlag New York, Inc. 1997
2nd edition: © Springer International Publishing AG 2018
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
Riemannian geometry is the study of manifolds endowed with Riemannian metrics,

which are, roughly speaking, rules for measuring lengths of tangent vectors and
angles between them. It is the most “geometric” branch of differential geometry.
Riemannian metrics are named for the great German mathematician Bernhard
Riemann (1826–1866).
This book is designed as a textbook for a graduate course on Riemannian
geometry for students who are familiar with the basic theory of smooth manifolds.
It focuses on developing an intimate acquaintance with the geometric meaning of
curvature, and in particular introducing many of the fundamental results that relate
the local geometry of a Riemannian manifold to its global topology (the kind of
results I like to call “local-to-global theorems,” as explained in Chapter 1). In so
doing, it introduces and demonstrates the uses of most of the main technical tools
needed for a careful study of Riemannian manifolds.
The book is meant to be introductory, not encyclopedic. Its coverage is reason-
ably broad, but not exhaustive. It begins with a careful treatment of the machinery
of metrics, connections, and geodesics, which are the indispensable tools in the sub-
ject. Next comes a discussion of Riemannian manifolds as metric spaces, and the
interactions between geodesics and metric properties such as completeness. It then
introduces the Riemann curvature tensor, and quickly moves on to submanifold
theory in order to give the curvature tensor a concrete quantitative interpretation.
The first local-to-global theorem I discuss is the Gauss–Bonnet theorem for com-
pact surfaces. Many students will have seen a treatment of this in undergraduate
courses on curves and surfaces, but because I do not want to assume such a course
as a prerequisite, I include a complete proof.
From then on, all efforts are bent toward proving a number of fundamental
local-to-global theorems for higher-dimensional manifolds, most notably the Killing–
Hopf theorem about constant-curvature manifolds, the Cartan–Hadamard theorem
about nonpositively curved manifolds, and Myers’s theorem about positively curved
ones. The last chapter also contains a selection of other important local-to-global
theorems.
v
vi Preface
Many other results and techniques might reasonably claim a place in an introduc-
tory Riemannian geometry book, but they would not fit in this book without dras-
tically broadening its scope. In particular, I do not treat the Morse index theorem,
Toponogov’s theorem, or their important applications such as the sphere theorem;
Hodge theory, gauge theory, minimal surface theory, or other applications of elliptic
partial differential equations to Riemannian geometry; or evolution equations such as
the Ricci flow or the mean curvature flow. These important topics are for other, more
advanced, books.
When I wrote the first edition of this book twenty years ago, a number of superb
reference books on Riemannian geometry were already available; in the intervening
years, many more have appeared. I invite the interested reader, after reading this
book, to consult some of those for a deeper treatment of some of the topics introduced
here, or to explore the more esoteric aspects of the subject. Some of my favorites are
Peter Petersen’s admirably comprehensive introductory text [Pet16]; the elegant
introduction to comparison theory by Jeff Cheeger and David Ebin [CE08] (which
was out of print for a number of years, but happily has been reprinted by the American
Mathematical Society); Manfredo do Carmo’s much more leisurely treatment of the
same material and more [dC92]; Barrett O’Neill’s beautifully integrated introduction
to pseudo-Riemannian and Riemannian geometry [O’N83]; Michael Spivak’s classic
multivolume tome [Spi79], which can be used as a textbook if plenty of time is
available, or can provide enjoyable bedtime reading; the breathtaking survey by
Marcel Berger [Ber03], which richly earns the word “panoramic” in its title; and
the “Encyclopaedia Britannica” of differential geometry books, Foundations of
Differential Geometry by Shoshichi Kobayashi and Katsumi Nomizu [KN96]. At the
other end of the spectrum, Frank Morgan’s delightful little book [Mor98] touches on
most of the important ideas in an intuitive and informal way with lots of pictures—I
enthusiastically recommend it as a prelude to this book. And there are many more to
recommend: for example, the books by Chavel [Cha06], Gallot/Hulin/Lafontaine
[GHL04], Jost [Jos17], Klingenberg [Kli95], and Jeffrey Lee [LeeJeff09] are all
excellent in different ways.
It is not my purpose to replace any of these. Instead, I hope this book fills a niche
in the literature by presenting a selective introduction to the main ideas of the
subject in an easily accessible way. The selection is small enough to fit (with some
judicious cutting) into a single quarter or semester course, but broad enough, I hope,
to provide any novice with a firm foundation from which to pursue research or
develop applications in Riemannian geometry and other fields that use its tools.
This book is written under the assumption that the student already knows the
fundamentals of the theory of topological and smooth manifolds, as treated, for
example, in my two other graduate texts [LeeTM, LeeSM]. In particular, the
student should be conversant with general topology, the fundamental group, covering
spaces, the classification of compact surfaces, topological and smooth manifolds,
immersions and submersions, submanifolds, vector fields and flows, Lie brackets and
Lie derivatives, tensors, differential forms, Stokes’s theorem, and the basic theory of
Lie groups. On the other hand, I do not assume any previous acquaintance with
Riemannian metrics, or even with the classical theory of curves and surfaces in R3 .
(In this subject, anything proved before 1950 can be considered “classical”!)
Preface vii
Although at one time it might have been reasonable to expect most mathematics
students to have studied surface theory as undergraduates, many current North
American undergraduate math majors never see any differential geometry. Thus the
fundamentals of the geometry of surfaces, including a proof of the Gauss–Bonnet
theorem, are worked out from scratch here.
The book begins with a nonrigorous overview of the subject in Chapter 1,
designed to introduce some of the intuitions underlying the notion of curvature and
to link them with elementary geometric ideas the student has seen before. Chapter 2
begins the course proper, with definitions of Riemannian metrics and some of their
attendant flora and fauna. Here I also introduce pseudo-Riemannian metrics, which
play a central role in Einstein’s general theory of relativity. Although I do not
attempt to provide a comprehensive introduction to pseudo-Riemannian geometry,
throughout the book I do point out which of the constructions and theorems of
Riemannian geometry carry over easily to the pseudo-Riemannian case and which
do not.
Chapter 3 describes some of the most important “model spaces” of Riemannian
and pseudo-Riemannian geometry—those with lots of symmetry—with a great deal
of detailed computation. These models form a sort of leitmotif throughout the text,
serving as illustrations and testbeds for the abstract theory as it is developed.
Chapter 4 introduces connections, together with some fundamental constructions
associated with them such as geodesics and parallel transport. In order to isolate the
important properties of connections that are independent of the metric, as well as to
lay the groundwork for their further study in arenas that are beyond the scope of this
book, such as the Chern–Weil theory of characteristic classes and the Donaldson and
Seiberg–Witten theories of gauge fields, connections are defined first on arbitrary
vector bundles. This has the further advantage of making it easy to define the induced
connections on tensor bundles. Chapter 5 investigates connections in the context of
Riemannian (and pseudo-Riemannian) manifolds, developing the Levi-Civita con-
nection, its geodesics, the exponential map, and normal coordinates. Chapter 6
continues the study of geodesics, focusing on their distance-minimizing properties.
First, some elementary ideas from the calculus of variations are introduced to prove
that every distance-minimizing curve is a geodesic. Then the Gauss lemma is used to
prove the (partial) converse—that every geodesic is locally minimizing.
Chapter 7 unveils the first fully general definition of curvature. The curvature
tensor is motivated initially by the question whether all Riemannian metrics are
“flat” (that is, locally isometric to the Euclidean metric). It turns out that the failure
of parallel transport to be path-independent is the primary obstruction to the
existence of a local isometry. This leads naturally to a qualitative interpretation of
curvature as the obstruction to flatness. Chapter 8 is an investigation of submanifold
theory, leading to the definition of sectional curvatures, which give curvature a
more quantitative geometric interpretation.
The last four chapters are devoted to the development of some of the most
important global theorems relating geometry to topology. Chapter 9 gives a simple
moving-frames proof of the Gauss–Bonnet theorem, based on a careful treatment of
Hopf’s rotation index theorem (often known by its German name, the Umlaufsatz).
Chapter 10 has a largely technical nature, covering Jacobi fields, conjugate points,
viii Preface
the second variation formula, and the index form for later use in comparison
theorems. Chapter 11 introduces comparison theory, using a simple comparison
theorem for matrix Riccati equations to prove the fundamental fact that bounds on
curvature lead to bounds (in the opposite direction) on the size of Jacobi fields,
which in turn lead to bounds on many fundamental geometric quantities, such as
distances, diameters, and volumes. Finally, in Chapter 12 comes the denouement:
proofs of some of the most important local-to-global theorems illustrating the ways
in which curvature and topology affect each other.
Exercises and Problems
This book contains many questions for the reader that deserve special mention.
They fall into two categories: “exercises,” which are integrated into the text, and
“problems,” grouped at the end of each chapter. Both are essential to a full
understanding of the material, but they are of somewhat different characters and
serve different purposes.
The exercises include some background material that the student should have
seen already in an earlier course, some proofs that fill in the gaps from the text,
some simple but illuminating examples, and some intermediate results that are used
in the text or the problems. They are, in general, elementary, but they are not
optional—indeed, they are integral to the continuity of the text. They are chosen
and timed so as to give the reader opportunities to pause and think over the material
that has just been introduced, to practice working with the definitions, and to
develop skills that are used later in the book. I recommend that students stop and do
each exercise as it occurs in the text, or at least convince themselves that they know
what is involved in the solution of each one, before going any further.
The problems that conclude the chapters are generally more difficult than the
exercises, some of them considerably so, and should be considered a central part
of the book by any student who is serious about learning the subject. They not only
introduce new material not covered in the body of the text, but they also provide the
student with indispensable practice in using the techniques explained in the text,
both for doing computations and for proving theorems. If the result of a problem is
used in an essential way in the text, or in a later problem, the page where it is used is
noted at the end of the problem statement. Instructors might want to present some
of these problems in class if more than a semester is available.
At the end of the book there are three appendices that contain brief reviews of
background material on smooth manifolds, tensors, and Lie groups. I have omitted
most of the proofs, but included references to other books where they may be
found. The results are collected here in order to clarify what results from topology
and smooth manifold theory this book will draw on, and also to establish definitions
and conventions that are used throughout the book. I recommend that most readers
at least glance through the appendices before reading the rest of the book, and
consider consulting the indicated references for any topics that are unfamiliar.
Preface ix
About the Second Edition
This second edition, titled Introduction to Riemannian Manifolds, has been adapted
from my earlier book Riemannian Manifolds: An Introduction to Curvature,
Graduate Texts in Mathematics 176, Springer 1997.
For those familiar with the first edition, the first difference you will notice about
this edition is that it is considerably longer than the first. To some extent, this is due
to the addition of more thorough explanations of some of the concepts. But a much
more significant reason for the increased length is the addition of many topics that
were not covered in the first edition. Here are some of the most important ones: a
somewhat expanded treatment of pseudo-Riemannian metrics, together with more
consistent explanations of which parts of the theory apply to them; a more detailed
treatment of which homogeneous spaces admit invariant metrics; a new treatment of
general distance functions and semigeodesic coordinates; introduction of the Weyl
tensor and the transformation laws for various curvatures under conformal changes
of metric; derivation of the variational equations for hypersurfaces that minimize
area with fixed boundary or fixed enclosed volume; an introduction to symmetric
spaces; and a treatment of the basic properties of the cut locus. Most importantly,
the entire treatment of comparison theory has been revamped and expanded based
on Riccati equations, and a handful of local-to-global theorems have been added
that were not present in the first edition: Cartan’s torsion theorem, Preissman’s
theorem, Cheng’s maximal diameter theorem, Milnor’s theorem on polynomial
growth of the fundamental group, and Synge’s theorem. I hope these will make the
book much more useful.
I am aware, though, that one of the attractions of the first edition for some
readers was its brevity. For those who would prefer a more streamlined path toward
the main local-to-global theorems in Chapter 12, here are topics that can be omitted
on a first pass through the book without essential loss of continuity.
• Chapter 2: Other generalizations of Riemannian metrics
• Chapter 3: Other homogeneous Riemannian manifolds and model pseudo-
Riemannian manifolds
• Chapter 5: Tubular neighborhoods, Fermi coordinates, and Euclidean and
non-Euclidean geometries
• Chapter 6: Distance functions and semigeodesic coordinates
• Chapter 7: The Weyl tensor and curvatures of conformally related metrics
• Chapter 8: Computations in semigeodesic coordinates, minimal hypersur-
faces, and constant-mean-curvature hypersurfaces
• Chapter 9: The entire chapter
• Chapter 10: Locally symmetric spaces and cut points
• Chapter 11: Günther’s volume comparison theorem and the Bishop–Gromov
volume comparison theorem
• Chapter 12: All but the theorems of Killing–Hopf, Cartan–Hadamard, and
Myers
x Preface
In addition to the major changes listed above, there are thousands of minor ones
throughout the book. Of course, I have attempted to correct all of the mistakes that I
became aware of in the first edition. Unfortunately, I surely have not been able to
avoid introducing new ones, so if you find anything that seems amiss, please let me
know by contacting me through the website listed below. I will keep an updated list
of corrections on that website.
I have also adjusted my notation and terminology to be consistent with my two
other graduate texts [LeeSM, LeeTM] and hopefully to be more consistent with
commonly accepted usage. Like those books, this one now has a notation index just
before the subject index, and it uses the same typographical conventions: mathema-
tical terms are typeset in bold italics when they are officially defined; exercises in
the text are indented, numbered consecutively with the theorems, and marked with
the special symbol I to make them easier to find; the ends of numbered examples
are marked with the symbol //; and the entire book is now set in Times Roman,
supplemented by the MathTime Professional II mathematics fonts created by
Personal TEX, Inc.
Acknowledgements
I owe an unpayable debt to the authors of the many Riemannian geometry books I
have used and cherished over the years, especially the ones mentioned above—I
have done little more than rearrange their ideas into a form that seems handy for
teaching. Beyond that, I would like to thank my Ph.D. advisor, Richard Melrose,
who many years ago introduced me to differential geometry in his eccentric but
thoroughly enlightening way; my colleagues Judith Arms, Yu Yuan, and Jim
Isenberg, who have provided great help in sorting out what topics should be
included; and all of the graduate students at the University of Washington who have
suffered with amazing grace through the many flawed drafts of both editions of this
book and have provided invaluable feedback, especially Jed Mihalisin, David
Sprehn, Collin Litterell, and Maddie Burkhart. And my deepest gratitude goes to
Ina Mette of Springer-Verlag (now at the AMS), who first convinced me to turn my
lecture notes into a book; without her encouragement, I would never have become a
textbook author.
Finally, I would like to dedicate this book to the memory of my late colleague
Steve Mitchell, who by his sparkling and joyful example taught me more about
teaching and writing than anyone.
Seattle, Washington, USA John M. Lee

June 2018 www.math.washington.edu/*lee/
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
1 What Is Curvature? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
The Euclidean Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Surfaces in Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Curvature in Higher Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Riemannian Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Methods for Constructing Riemannian Metrics . . . . . . . . . . . . . . . . . . 15
Basic Constructions on Riemannian Manifolds . . . . . . . . . . . . . . . . . . 25
Lengths and Distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Pseudo-Riemannian Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Other Generalizations of Riemannian Metrics . . . . . . . . . . . . . . . . . . . 46
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3 Model Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Symmetries of Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 55
Euclidean Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Hyperbolic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Invariant Metrics on Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Other Homogeneous Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . 72
Model Pseudo-Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 79
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
The Problem of Differentiating Vector Fields . . . . . . . . . . . . . . . . . . . 85
Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Covariant Derivatives of Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . 95
Vector and Tensor Fields Along Curves . . . . . . . . . . . . . . . . . . . . . . . 100
xi
xii Contents
Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Parallel Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Pullback Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5 The Levi-Civita Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

The Tangential Connection Revisited . . . . . . . . . . . . . . . . . . . . . . . . . 115
Connections on Abstract Riemannian Manifolds . . . . . . . . . . . . . . . . . 117
The Exponential Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Normal Neighborhoods and Normal Coordinates . . . . . . . . . . . . . . . . 131
Tubular Neighborhoods and Fermi Coordinates . . . . . . . . . . . . . . . . . 133
Geodesics of the Model Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Euclidean and Non-Euclidean Geometries . . . . . . . . . . . . . . . . . . . . . 142
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6 Geodesics and Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

Geodesics and Minimizing Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Uniformly Normal Neighborhoods . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Distance Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Semigeodesic Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Local Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
The Curvature Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Flat Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Symmetries of the Curvature Tensor . . . . . . . . . . . . . . . . . . . . . . . . . 202
The Ricci Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Ricci and Scalar Curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
The Weyl Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Curvatures of Conformally Related Metrics . . . . . . . . . . . . . . . . . . . . 216
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8 Riemannian Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

The Second Fundamental Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
Hypersurfaces in Euclidean Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Sectional Curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
9 The Gauss–Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

Some Plane Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
The Gauss–Bonnet Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Contents xiii
The Gauss–Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
10 Jacobi Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

The Jacobi Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
Basic Computations with Jacobi Fields . . . . . . . . . . . . . . . . . . . . . . . 287
Conjugate Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
The Second Variation Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
Cut Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
11 Comparison Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

Jacobi Fields, Hessians, and Riccati Equations . . . . . . . . . . . . . . . . . . 320
Comparisons Based on Sectional Curvature . . . . . . . . . . . . . . . . . . . . 327
Comparisons Based on Ricci Curvature . . . . . . . . . . . . . . . . . . . . . . . 336
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
12 Curvature and Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Manifolds of Constant Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Manifolds of Nonpositive Curvature . . . . . . . . . . . . . . . . . . . . . . . . . 352
Manifolds of Positive Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
Appendix A: Review of Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 371

Topological Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
Smooth Manifolds and Smooth Maps . . . . . . . . . . . . . . . . . . . . . . . . . . 374
Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
Submanifolds. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
Vector Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
The Tangent Bundle and Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . 384
Smooth Covering Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
Appendix B: Review of Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391

Tensors on a Vector Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
Tensor Bundles and Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
Differential Forms and Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Appendix C: Review of Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407

Definitions and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
The Lie Algebra of a Lie Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
Group Actions on Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Notation Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Subject Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Chapter 1
What Is Curvature?
If you have spent some time studying modern differential geometry, with its intricate
web of manifolds, submanifolds, vector fields, Lie derivatives, tensor fields, differ-
ential forms, orientations, and foliations, you might be forgiven for wondering what
it all has to do with geometry. In most people’s experience, geometry is concerned
with properties such as distances, lengths, angles, areas, volumes, and curvature.
These concepts, however, are often barely mentioned in typical beginning graduate
courses in smooth manifold theory.
The purpose of this book is to introduce the theory of Riemannian manifolds:
these are smooth manifolds equipped with Riemannian metrics (smoothly varying
choices of inner products on tangent spaces), which allow one to measure geometric
quantities such as distances and angles. This is the branch of differential geometry in
which “geometric” ideas, in the familiar sense of the word, come to the fore. It is the
direct descendant of Euclid’s plane and solid geometry, by way of Gauss’s theory of
curved surfaces in space, and it is a dynamic subject of contemporary research.
The central unifying theme in current Riemannian geometry research is the
notion of curvature and its relation to topology. This book is designed to help you
develop both the tools and the intuition you will need for an in-depth exploration
of curvature in the Riemannian setting. Unfortunately, as you will soon discover, an
adequate development of curvature in an arbitrary number of dimensions requires
a great deal of technical machinery, making it easy to lose sight of the underlying
geometric content. To put the subject in perspective, therefore, let us begin by asking
some very basic questions: What is curvature? What are some important theorems
about it? In this chapter, we explore these and related questions in an informal way,
without proofs. The “official” treatment of the subject begins in Chapter 2.
The Euclidean Plane

To get a sense of the kinds of questions Riemannian geometers address and where
these questions came from, let us look back at the very roots of our subject. The
© Springer International Publishing AG 2018 1
J. M. Lee, Introduction to Riemannian Manifolds, Graduate Texts
in Mathematics 176, https://doi.org/10.1007/978-3-319-91755-9 1
2 1 What Is Curvature?
treatment of geometry as a mathematical subject began with Euclidean plane geom-

etry, which you probably studied in secondary school. Its elements are points, lines,
distances, angles, and areas; and its most fundamental relationship is congruence—
two plane figures are congruent if one can be transformed into the other by a rigid
motion of the plane, which is a bijective transformation from the plane to itself that
preserves distances. Here are a couple of typical theorems.
Theorem 1.1 (Side-Side-Side). Two Euclidean triangles are congruent if and only
if the lengths of their corresponding sides are equal.
Theorem 1.2 (Angle-Sum Theorem). The sum of the interior angles of a Euclidean
triangle is .
As trivial as they may seem, these theorems serve to illustrate two major types of
results that permeate the study of geometry; in this book, we call them “classification
theorems” and “local-to-global theorems.”
The side-side-side (SSS) theorem is a classification theorem. Such a theorem
tells us how to determine whether two mathematical objects are equivalent (under
some appropriate equivalence relation). An ideal classification theorem lists a small
number of computable invariants (whatever “small” may mean in a given context),
and says that two objects are equivalent if and only if all of these invariants match.
In this case the equivalence relation is congruence, and the invariants are the three
side lengths.
The angle-sum theorem is of a different sort. It relates a local geometric property
(angle measure) to a global property (that of being a three-sided polygon or triangle).
Most of the theorems we study in this book are of this type, which, for lack of a
better name, we call local-to-global theorems.
After proving the basic facts about points and lines and the figures constructed
directly from them, one can go on to study other figures derived from the basic
elements, such as circles. Two typical results about circles are given below; the first
is a classification theorem, while the second is a local-to-global theorem. (It may not
be obvious at this point why we consider the second to be a local-to-global theorem,
but it will become clearer soon.)
Theorem 1.3 (Circle Classification Theorem). Two circles in the Euclidean plane
are congruent if and only if they have the same radius.
Theorem 1.4 (Circumference Theorem). The circumference of a Euclidean circle

of radius R is 2R.
If we want to continue our study of plane geometry beyond figures constructed

from lines and circles, sooner or later we have to come to terms with other curves in
the plane. An arbitrary curve cannot be completely described by one or two numbers
such as length or radius; instead, the basic invariant is curvature, which is defined
using calculus and is a function of position on the curve.
Formally, the curvature of a plane curve is defined to be .t / D j 00 .t /j, the
length of the acceleration vector, when is given a unit-speed parametrization.
The Euclidean Plane 3
Fig. 1.1: Osculating circle
(Here and throughout this book, the word “curve” refers to a parametrized curve,
not a set of points. Typically, a curve will be defined as a smooth function of a real
variable t , with a prime representing an ordinary derivative with respect to t .)
Geometrically, the curvature has the following interpretation. Given a point
p D .t /, there are many circles tangent to at p—namely, those circles whose
velocity vector at p is the same as that of when both are given unit-speed
parametrizations; these are the circles whose centers lie on the line that passes
through p and is orthogonal to 0 .p/. Among these circles, there is exactly one
unit-speed parametrized circle whose acceleration vector at p is the same as that of
; it is called the osculating circle (Fig. 1.1). (If the acceleration of is zero, replace
the osculating circle by a straight line, thought of as a “circle with infinite radius.”)
The curvature is then .t / D 1=R, where R is the radius of the osculating circle.
The larger the curvature, the greater the acceleration and the smaller the osculating
circle, and therefore the faster the curve is turning. A circle of radius R has constant
curvature 1=R, while a straight line has curvature zero.
It is often convenient for some purposes to extend the definition of the curvature
of a plane curve, allowing it to take on both positive and negative values. This is done
by choosing a continuous unit normal vector field N along the curve, and assigning
the curvature a positive sign if the curve is turning toward the chosen normal or a
negative sign if it is turning away from it. The resulting function N along the curve
is then called the signed curvature.
Here are two typical theorems about plane curves.
Theorem 1.5 (Plane Curve Classification Theorem). Suppose and z W Œa; b !

R2 are smooth, unit-speed plane curves with unit normal vector fields N and Nz ,
and N .t /, Nz .t / represent the signed curvatures at .t / and z.t/, respectively.
Then and z are congruent by a direction-preserving congruence if and only if
N .t / D Nz .t / for all t 2 Œa; b.
Theorem 1.6 (Total Curvature Theorem). If W Œa; b ! R2 is a unit-speed simple

closed curve such that 0 .a/ D 0 .b/, and N is the inward-pointing normal, then
Z b
N .t / dt D 2:
a
The first of these is a classification theorem, as its name suggests. The second
is a local-to-global theorem, since it relates the local property of curvature to the
global (topological) property of being a simple closed curve. We will prove both of
these theorems later in the book: the second will be derived as a consequence of a
more general result in Chapter 9 (see Corollary 9.6); the proof of the first is left to
Problem 9-12.
It is interesting to note that when we specialize to circles, these theorems reduce
to the two theorems about circles above: Theorem 1.5 says that two circles are con-
gruent if and only if they have the same curvature, while Theorem 1.6 says that if
a circle has curvature and circumference C , then C D 2. It is easy to see that
these two results are equivalent to Theorems 1.3 and 1.4. This is why it makes sense
to regard the circumference theorem as a local-to-global theorem.
Surfaces in Space
The next step in generalizing Euclidean geometry is to start working in three dimen-
sions. After investigating the basic elements of “solid geometry”—points, lines,
planes, polyhedra, spheres, distances, angles, surface areas, volumes—one is led
to study more general curved surfaces in space (2-dimensional embedded subman-
ifolds of R3 , in the language of differential geometry). The basic invariant in this
setting is again curvature, but it is a bit more complicated than for plane curves,
because a surface can curve differently in different directions.
The curvature of a surface in space is described by two numbers at each point,
called the principal curvatures. We will define them formally in Chapter 8, but here
is an informal recipe for computing them. Suppose S is a surface in R3 , p is a point
in S , and N is a unit normal vector to S at p.
1. Choose a plane ˘ passing through p and parallel to N . The intersection of ˘
with a neighborhood of p in S is a plane curve ˘ containing p (Fig. 1.2).
2. Compute the signed curvature N of at p with respect to the chosen unit
normal N .
3. Repeat this for all normal planes ˘ . The principal curvatures of S at p,
denoted by 1 and 2 , are the minimum and maximum signed curvatures so
obtained.
Although the principal curvatures give us a lot of information about the geome-
try of S , they do not directly address a question that turns out to be of paramount
importance in Riemannian geometry: Which properties of a surface are intrinsic?
Roughly speaking, intrinsic properties are those that could in principle be measured
or computed by a 2-dimensional being living entirely within the surface. More pre-
cisely, a property of surfaces in R3 is called intrinsic if it is preserved by isometries
(maps from one surface to another that preserve lengths of curves).
To see that the principal curvatures are not intrinsic, consider the following two
embedded surfaces S1 and S2 in R3 (Figs. 1.3 and 1.4): S1 is the square in the xy-
Surfaces in Space 5
Fig. 1.2: Computing principal curvatures
Fig. 1.3: S1 Fig. 1.4: S2
plane where 0 < x < and 0 < y < , and S2 is the half-cylinder f.x; y; z/ W z D
p
1 y 2 ; 0 < x < ; jyj < 1g. If we follow the recipe above for computing principal
curvatures (using, say, the downward-pointing unit normal), we find that, since all
planes intersect S1 in straight lines, the principal curvatures of S1 are 1 D 2 D 0.
On the other hand, it is not hard to see that the principal curvatures of S2 are 1 D 0
and 2 D 1. However, the map taking .x; y; 0/ to .x; cos y; sin y/ is a diffeomorphism
from S1 to S2 that preserves lengths of curves, and is thus an isometry.
Even though the principal curvatures are not intrinsic, the great German mathe-
matician Carl Friedrich Gauss made the surprising discovery in 1827 [Gau65] that
a particular combination of them is intrinsic. (See also [Spi79, Vol. 2] for an excel-
lent discussion of the details of Gauss’s paper.) He found a proof that the product
K D 1 2 , now called the Gaussian curvature, is intrinsic. He thought this result
was so amazing that he named it Theorema Egregium. (This does not mean “totally
awful theorem” as its English cognate egregious might suggest; a better translation
into modern colloquial English might be “totally awesome theorem.”)
Fig. 1.5: K > 0 Fig. 1.6: K < 0
To get a feeling for what Gaussian curvature tells us about surfaces, let us look at
a few examples. Simplest of all is any surface that is an open subset of a plane: as we
have seen, such a surface has both principal curvatures equal to zero and therefore
has constant Gaussian curvature equal to zero. The half-cylinder described above
also has K D 1 2 D 0 1 D 0, as the Theorema Egregium tells us it must, being iso-
metric to a square. Another simple example is a sphere of radius R. Every normal
plane intersects the sphere in a great circle, which has radius R and therefore curva-
ture ˙1=R (with the sign depending on whether we choose the outward-pointing or
inward-pointing normal). Thus the principal curvatures are both equal to ˙1=R, and
the Gaussian curvature is 1 2 D 1=R2 . Note that while the signs of the principal
curvatures depend on the choice of unit normal, the Gaussian curvature does not: it
is always positive on the sphere.
Similarly, any surface that is “bowl-shaped” or “dome-shaped” has positive
Gaussian curvature (Fig. 1.5), because the two principal curvatures always have the
same sign, regardless of which normal is chosen. On the other hand, the Gaussian
curvature of any surface that is “saddle-shaped” (Fig. 1.6) is negative, because the
principal curvatures are of opposite signs.
The model spaces of surface theory are the surfaces with constant Gaussian cur-
vature. We have already seen two of them: the Euclidean plane R2 (K D 0), and the
sphere of radius R (K D 1=R2 ). The most important model surface with constant
negative Gaussian curvature is called the hyperbolic plane, and will be defined in
Chapter 3. It is not so easy to visualize because it cannot be realized globally as a
smoothly embedded surface in R3 (see [Spi79, Vol. 3, pp. 373–385] for a proof).
Surface theory is a highly developed branch of geometry. Of all its results, two—
a classification theorem and a local-to-global theorem—are generally acknowledged
as the most important.
Theorem 1.7 (Uniformization Theorem). Every connected 2-manifold is diffeo-
morphic to a quotient of one of the constant-curvature model surfaces described
above by a discrete group of isometries without fixed points. Thus every connected
2-manifold has a complete Riemannian metric with constant Gaussian curvature.
Theorem 1.8 (Gauss–Bonnet Theorem). Suppose S is a compact Riemannian 2-
manifold. Then
Z
K dA D 2.S /;
S
where .S / is the Euler characteristic of S .

Curvature in Higher Dimensions 7
The uniformization theorem is a classification theorem, because it replaces the

problem of classifying surfaces with that of classifying certain discrete groups of
isometries of the models. The latter problem is not easy by any means, but it sheds
a great deal of new light on the topology of surfaces nonetheless. In Chapter 3, we
sketch a proof of the uniformization theorem for the case of compact surfaces.
Although stated here as a geometric-topological result, the uniformization the-
orem is usually stated somewhat differently and proved using complex analysis. If
you are familiar with complex analysis and the complex version of the uniformiza-
tion theorem, it will be an enlightening exercise after you have finished this book to
prove that the complex version of the theorem is equivalent to the one stated here.
The Gauss–Bonnet theorem, on the other hand, is purely a theorem of differential
geometry, arguably the most fundamental and important one of all. It relates a local
geometric property (the curvature) with a global topological invariant (the Euler
characteristic). We give a detailed proof in Chapter 9.
Taken together, these theorems place strong restrictions on the types of metrics
that can occur on a given surface. For example, one consequence of the Gauss–
Bonnet theorem is that the only compact, connected, orientable surface that admits
a metric of strictly positive Gaussian curvature is the sphere. On the other hand, if
a compact, connected, orientable surface has nonpositive Gaussian curvature, the
Gauss–Bonnet theorem rules out the sphere, and then the uniformization theorem
tells us that its universal covering space is topologically equivalent to the plane.
Curvature in Higher Dimensions
We end our survey of the basic ideas of Riemannian geometry by mentioning briefly
how curvature appears in higher dimensions. Suppose M is an n-dimensional Rie-
mannian manifold. As with surfaces, the basic geometric invariant is curvature, but
curvature becomes a much more complicated quantity in higher dimensions because
a manifold may curve in so many different directions.
The first problem we must contend with is that, in general, Riemannian mani-
folds are not presented to us as embedded submanifolds of Euclidean space. There-
fore, we must abandon the idea of cutting out curves by intersecting our manifold
with planes, as we did when defining the principal curvatures of a surface in R3 .
Instead, we need a more intrinsic way of sweeping out submanifolds. Fortunately,
geodesics—curves that are the shortest paths between nearby points—are ready-
made tools for this and many other purposes in Riemannian geometry. Examples
are straight lines in Euclidean space and great circles on a sphere.
The most fundamental fact about geodesics, which we prove in Chapter 4, is
that given any point p 2 M and any vector v tangent to M at p, there is a unique
geodesic starting at p with initial velocity v.
Here is a brief recipe for computing some curvatures at a point p 2 M .
1. Choose a 2-dimensional subspace ˘ of the tangent space to M at p.
2. Look at all the geodesics through p whose initial velocities lie in the selected
plane ˘ . It turns out that near p these sweep out a certain 2-dimensional sub-
manifold S˘ of M , which inherits a Riemannian metric from M .
3. Compute the Gaussian curvature of S˘ at p, which the Theorema Egregium
tells us can be computed from the Riemannian metric that S˘ inherits from M .
This gives a number, denoted by sec.˘ /, called the sectional curvature of M
at p associated with the plane ˘ .
Thus the “curvature” of M at p has to be interpreted as a map
˚
sec W 2-planes in Tp M ! R:
As we will see in Chapter 3, we again have three classes of constant (sectional)

curvature model spaces: Rn with its Euclidean metric (for which sec 0); the n-
sphere of radius R, with the Riemannian metric inherited from RnC1 (sec 1=R2 );
and hyperbolic space of radius R (with sec 1=R2 ). Unfortunately, however,
there is as yet no satisfactory uniformization theorem for Riemannian manifolds in
higher dimensions. In particular, it is definitely not true that every manifold pos-
sesses a metric of constant sectional curvature. In fact, the constant-curvature met-
rics can all be described rather explicitly by the following classification theorem.
Theorem 1.9 (Characterization of Constant-Curvature Metrics). The complete,

connected, n-dimensional Riemannian manifolds of constant sectional curvature
are, up to isometry, exactly the Riemannian quotients of the form M = , where M
is a Euclidean space, sphere, or hyperbolic space with constant sectional curvature,
that acts freely on M
and is a discrete group of isometries of M .
On the other hand, there are a number of powerful local-to-global theorems,

which can be thought of as generalizations of the Gauss–Bonnet theorem in var-
ious directions. They are consequences of the fact that positive curvature makes
geodesics converge, while negative curvature forces them to spread out. Here (in
somewhat simplified form) are two of the most important such theorems.
Theorem 1.10 (Cartan–Hadamard). Suppose M is a complete, connected Rie-

mannian n-manifold with all sectional curvatures less than or equal to zero. Then
the universal covering space of M is diffeomorphic to Rn .
Theorem 1.11 (Myers). Suppose M is a complete, connected Riemannian mani-

fold with all sectional curvatures bounded below by a positive constant. Then M is
compact and has a finite fundamental group.
Looking back at the remarks concluding the section on surfaces above, you can
see that these last three theorems generalize some of the consequences of the uni-
formization and Gauss–Bonnet theorems, although not their full strength. It is the
primary goal of this book to prove Theorems 1.9, 1.10, and 1.11, among others; it is
a primary goal of current research in Riemannian geometry to improve upon them
and further generalize the results of surface theory to higher dimensions.
Chapter 2
Riemannian Metrics
In this chapter we officially define Riemannian metrics, and discuss some of the
basic computational techniques associated with them. After the definitions, we
describe a few standard methods for constructing Riemannian manifolds as subman-
ifolds, products, and quotients of other Riemannian manifolds. Then we introduce
some of the elementary geometric constructions provided by Riemannian metrics,
the most important of which is the Riemannian distance function, which turns every
connected Riemannian manifold into a metric space.
At the end of the chapter, we discuss some important generalizations of Rie-
mannian metrics—most importantly, the pseudo-Riemannian metrics, followed by
brief mentions of sub-Riemannian and Finsler metrics.
Before you read this chapter, it would be a good idea to skim through the three
appendices after Chapter 12 to get an idea of the prerequisite material that will be
assumed throughout this book.
Definitions
Everything we know about the Euclidean geometry of Rn can be derived from its
dot product, which is defined for v D .v 1 ; : : : ; v n / and w D .w 1 ; : : : ; w n / by
X
n
vw D vi wi :
iD1
The dot product has a natural generalization to arbitrary vector spaces. Given a
vector space V (which we always assume to be real), an inner product on V is
a map V V ! R, typically written .v; w/ 7! hv; wi, that satisfies the following
properties for all v; w; x 2 V and a; b 2 R:
(i) S YMMETRY: hv; wi D hw; vi.
(ii) B ILINEARITY: hav C bw; xi D ahv; xi C bhw; xi D hx; av C bwi.

10 2 Riemannian Metrics
(iii) P OSITIVE D EFINITENESS : hv; vi 0, with equality if and only if v D 0.

A vector space endowed with a specific inner product is called an inner product
space.
An inner product on V allows us to make sense of geometric quantities such as
lengths of vectors and angles between vectors. First, we define the length or norm
of a vector v 2 V as
jvj D hv; vi1=2 : (2.1)
The following identity shows that an inner product is completely determined by
knowledge of the lengths of all vectors.
Lemma 2.1 (Polarization Identity). Suppose h; i is an inner product on a vector
space V . Then for all v; w 2 V ,

hv; wi D 14 hv C w; v C wi hv w; v wi : (2.2)
I Exercise 2.2. Prove the preceding lemma.
The angle between two nonzero vectors v; w 2 V is defined as the unique 2

Œ0; satisfying
hv; wi
cos D : (2.3)
jvj jwj
Two vectors v; w 2 V are said to be orthogonal if hv; wi D 0, which means that
either their angle is =2 or one of the vectors is zero. If S V is a linear subspace,
the set S ? V , consisting of all vectors in V that are orthogonal to every vector in
S , is also a linear subspace, called the orthogonal complement of S .
Vectors v1 ; : : : ; vk are called orthonormal if they are of length 1 and pairwise
orthogonal, or equivalently if hvi ; vj i D ıij (where ıij is the Kronecker delta symbol
defined in Appendix B; see (B.1)). The following well-known proposition shows
that every finite-dimensional inner product space has an orthonormal basis.
Proposition 2.3 (Gram–Schmidt Algorithm). Let V be an n-dimensional inner
product space, and suppose .v1 ; : : : ; vn / is any ordered basis for V: Then there is an
orthonormal ordered basis .b1 ; : : : ; bn / satisfying the following conditions:
span.b1 ; : : : ; bk / D span.v1 ; : : : ; vk / for each k D 1; : : : ; n: (2.4)
Proof. The basis vectors b1 ; : : : ; bn are defined recursively by

v1
b1 D ; (2.5)
jv1 j
P 1
vj jiD1 hvj ; bi ibi
bj D ˇ Pj 1 ˇ; 2 j n: (2.6)
ˇ ˇ
ˇvj iD1 hvj ; bi ibi ˇ
Because v1 ¤ 0 and vj … span.b1 ; : : : ; bj 1 / for each j 2, the denominators are all

nonzero. These vectors satisfy (2.4) by construction, and are orthonormal by direct
computation.
Definitions 11
If two vector spaces V and W are both equipped with inner products, de-
noted by h; iV and h; iW , respectively, then a map F W V ! W is called a lin-
ear isometry if it is a vector space isomorphism that preserves inner products:
hF .v/; F .v 0 /iW D hv; v 0 iV . If V and W are inner product spaces of dimension n,
then given any choices of orthonormal bases .v1 ; : : : ; vn / for V and .w1 ; : : : ; wn / for
W , the linear map F W V ! W determined by F .vi / D wi is easily seen to be a lin-
ear isometry. Thus all inner product spaces of the same finite dimension are linearly
isometric to each other.
Riemannian Metrics
To extend these geometric ideas to abstract smooth manifolds, we define a structure

that amounts to a smoothly varying choice of inner product on each tangent space.
Let M be a smooth manifold. A Riemannian metricon M is a smooth covariant
2-tensor field g 2 T 2 .M / whose value gp at each p 2 M is an inner product on
Tp M ; thus g is a symmetric 2-tensor field that is positive definite in the sense that
gp .v; v/ 0 for each p 2 M and each v 2 Tp M , with equality if and only if v D 0.
A Riemannian manifold is a pair .M; g/, where M is a smooth manifold and g is a
specific choice of Riemannian metric on M . If M is understood to be endowed with
a specific Riemannian metric, we sometimes say “M is a Riemannian manifold.”
The next proposition shows that Riemannian metrics exist in great abundance.
Proposition 2.4. Every smooth manifold admits a Riemannian metric.
I Exercise 2.5. Use a partition of unity to prove the preceding proposition.
We will give a number of examples of Riemannian metrics, along with several

systematic methods for constructing them, later in this chapter and in the next.
If M is a smooth manifold with boundary, a Riemannian metric on M is defined
in exactly the same way: a smooth symmetric 2-tensor field g that is positive definite
everywhere. A Riemannian manifold with boundary is a pair .M; g/, where M is
a smooth manifold with boundary and g is a Riemannian metric on M . Many of the
results we will discuss in this book work equally well for manifolds with or without
boundary, with the same proofs, and in such cases we will state them in that gen-
erality. But when the treatment of a boundary would involve additional difficulties,
we will generally restrict attention to the case of manifolds without boundary, since
that is our primary interest. Many problems involving Riemannian manifolds with
boundary can be addressed by embedding into a larger manifold without boundary
and extending the Riemannian metric arbitrarily to the larger manifold; see Propo-
sition A.31 in Appendix A.
A Riemannian metric is not the same as a metric in the sense of metric spaces
(though, as we will see later in this chapter, the two concepts are related). In this
book, when we use the word “metric” without further qualification, it always refers
to a Riemannian metric.
Let g be a Riemannian metric on a smooth manifold M with or without boundary.

Because gp is an inner product on Tp M for each p 2 M , we often use the following
angle-bracket notation for v; w 2 Tp M :
hv; wig D gp .v; w/:
Using this inner product, we can define lengths of tangent vectors, angles between
nonzero tangent vectors, and orthogonality of tangent vectors as described above.
The length of a vector v 2 Tp M is denoted by jvjg D hv; vi1=2 g . If the metric is
understood, we sometimes omit it from the notation, and write hv; wi and jvj in
place of hv; wig and jvjg , respectively.
The starting point for Riemannian geometry is the following fundamental exam-
ple.
Example 2.6 (The Euclidean Metric). The Euclidean metric is the Riemannian
metric gx on Rn whose value at each x 2 Rn is just the usual dot product on Tx Rn
1Tx R Š R . This P w 2 Tx Rn writ-
n n
under the natural identification means that for v;P
ten in standard coordinates x ; : : : ; x as v D i v @i jx , w D j w j @j jx , we
n i
have
Xn
hv; wigx D vi wi :
iD1
When working with R as a Riemannian manifold, we always assume we are using

n
the Euclidean metric unless otherwise specified. //
Isometries

Suppose .M; g/ and M ; gz are Riemannian manifolds with or without boundary.

An isometry from .M ;g/ to M f ,e
g is a diffeomorphism ' W M ! M such that

' gz D g. Unwinding the definitions shows that this is equivalent to the requirement
that ' be a smooth bijection and each d'p W Tp M ! T'.p/ M be a lin-
differential

ear isometry. We say .M; g/ and M ; gz are isometric if there exists an isometry
between them.
A composition of isometries and the inverse of an isometry are again isometries,
so being isometric is an equivalence relation on the class of Riemannian manifolds
with or without boundary. Our subject, Riemannian geometry, is concerned primar-
ily with properties of Riemannian
manifolds that are preserved by isometries.
If .M; g/ and M ; gz are Riemannian manifolds, a map ' W M ! M is a local
isometry if each point p 2 M has a neighborhood U such that 'jU is an isometry
onto an open subset of M .

;g
I Exercise 2.7. Prove that if .M; g/ and M z are Riemannian manifolds of the same
is a local isometry if and only if ' g
dimension, a smooth map ' W M ! M z D g.
A Riemannian n-manifold is said to be flat if it is locally isometric to a Euclidean

space, that is, if every point has a neighborhood that is isometric to an open set in
Definitions 13
Rn with its Euclidean metric. Problem 2-1 shows that all Riemannian 1-manifolds
are flat; but we will see later that this is far from the case in higher dimensions.
An isometry from .M; g/ to itself is called an isometry of .M ,g/. The set of
all isometries of .M; g/ is a group under composition, called the isometry group
of .M ,g/; it is denoted by Iso.M; g/, or sometimes just Iso.M / if the metric is
understood.
A deep theorem of Sumner B. Myers and Norman E. Steenrod [MS39] shows
that if M has finitely many components, then Iso.M; g/ has a topology and smooth
structure making it into a finite-dimensional Lie group acting smoothly on M . We
will neither prove nor use the Myers–Steenrod theorem, but if you are interested, a
good source for the proof is [Kob72].
Local Representations for Metrics

Suppose .M; g/ is a Riemannian manifold with or without boundary. If x 1 ; : : : ; x n
are any smooth local coordinates on an open subset U M , then g can be written
locally in U as
g D gij dx i ˝ dx j (2.7)
for some collection of n2 smooth functions gij for i; j D 1; : : : ; n. (Here and
throughout the book, we use the Einstein summation convention; see p. 375.)
The component functions of this ˝ tensor field ˛ constitute a matrix-valued function
.gij /, characterized by gij .p/ D @i jp ; @j jp , where @i D @=@x i is the i th coordi-
nate vector field; this matrix is symmetric in i and j and depends smoothly on
p 2 U . If v D v i @i jp is a vector in Tp M such that gij .p/v j D 0, it follows that
hv; vi D gij .p/v i v j D 0, which implies v D 0; thus the matrix .gij .p// is always
nonsingular. The notation for g can be shortened by expressing it in terms of the
symmetric product (see Appendix B): using the symmetry of gij , we compute
g D gij dx i ˝ dx j

D 12 gij dx i ˝ dx j C gj i dx i ˝ dx j

D 12 gij dx i ˝ dx j C gij dx j ˝ dx i
D gij dx i dx j :
For example, the Euclidean metric on Rn (Example 2.6) can be expressed in

standard coordinates in several ways:
X X 2
gx D dx i dx i D dx i D ıij dx i dx j : (2.8)
i i
The matrix of gx in these coordinates is thus gxij D ıij .

More generally, if .E1 ; : : : ; En / is any smooth local frame for TM on an open
subset U M and ."1 ; : : : ; "n / is its dual coframe, we can write g locally in U as
g D gij "i "j ; (2.9)

˝ ˛
where gij .p/ D Ei jp ; Ej jp , and the matrix-valued function .gij / is symmetric and
smooth as before.
A Riemannian metric g acts on smooth vector fields X; Y 2 X.M / to yield a
real-valued function hX; Y i. In terms of any smooth local frame, this function is
expressed locally by hX; Y i D gij X i Y j and therefore is smooth. Similarly, we
obtain a nonnegative real-valued function jX j D hX; X i1=2 , which is continuous
everywhere and smooth on the open subset where X ¤ 0.
A local frame .Ei / for M on an open set U is said to be an orthonormal frame
if the vectors E1 jp ; : : : ; En jp are an orthonormal basis for Tp M at each p 2 U .
Equivalently, .Ei / is an orthonormal frame if and only if
hEi ; Ej i D ıij ;
in which case g has the local expression
g D ."1 /2 C C ."n /2 ;
where ."i /2 denotes the symmetric product "i "i D "i ˝ "i .
Proposition 2.8 (Existence of Orthonormal Frames). Let .M; g/ be a Riemannian
n-manifold with or without boundary. If .Xj / is any smooth local frame for TM
over an open subset U M , thenthere is a smooth orthonormal
frame .Ej / over
U such that span E1 jp ; : : : ; Ek jp D span X1 jp ; : : : ; Xk jp for each k D 1; : : : ; n
and each p 2 U . In particular, for every p 2 M , there is a smooth orthonormal
frame .Ej / defined on some neighborhood of p.

Proof. Applying the Gram–Schmidt algorithm to the vectors X1 jp ; : : : ; Xn jp at
each p 2 U , we obtain an ordered n-tuple of rough orthonormal vector fields
.E1 ; : : : ; En / over U satisfying the span conditions. Because the vectors whose
norms appear in the denominators of (2.5)–(2.6) are nowhere vanishing, those for-
mulas show that each vector field Ej is smooth. The last statement of the proposition
follows by applying this construction to any smooth local frame in a neighborhood
of p.
Warning: A common mistake made by beginners is to assume that one can find
coordinates near p such that the coordinate frame .@i / is orthonormal. Proposition
2.8 does not show this. In fact, as we will see in Chapter 7, this is possible only
when the metric is flat, that is, locally isometric to the Euclidean metric.
For a Riemannian manifold .M; g/ with or without boundary, we define the unit
tangent bundle to be the subset U TM TM consisting of unit vectors:
˚
U TM D .p; v/ 2 TM W jvjg D 1 : (2.10)
Proposition 2.9 (Properties of the Unit Tangent Bundle). If .M; g/ is a Rie-

mannian manifold with or without boundary, its unit tangent bundle U TM is a
Methods for Constructing Riemannian Metrics 15
smooth, properly embedded codimension-1 submanifold with boundary in TM , with

@.U TM / D 1 .@M / (where W U TM ! M is the canonical projection). The unit
tangent bundle is connected if and only if M is connected, and compact if and only
if M is compact.
I Exercise 2.10. Use local orthonormal frames to prove the preceding proposition.
Methods for Constructing Riemannian Metrics
Many examples of Riemannian manifolds arise naturally as submanifolds, products,

and quotients of other Riemannian manifolds. In this section, we introduce some of
the tools for constructing such metrics.
Riemannian Submanifolds
Every submanifold of a Riemannian manifold automatically inherits a Riemannian

metric, and many interesting Riemannian metrics are defined in this way. The key
fact is the following lemma.

Lemma 2.11. Suppose M ; gz is a Riemannian manifold with or without boundary,
M is a smooth manifold with or without boundary, and F W M ! M is a smooth

map. The smooth 2-tensor field g D F gz is a Riemannian metric on M if and only
if F is an immersion.
I Exercise 2.12. Prove Lemma 2.11.

Suppose M ; gz is a Riemannian manifold with or without boundary. Given a
smooth immersion F W M ! M , the metric g D F gz is called the metric induced
by F . On the other hand, if M is already endowed with a given Riemannian met-
ric g, an immersion or embedding F W M ! M satisfying F gz D g is called an
isometric immersion or isometric embedding, respectively. Which terminology is
used depends on whether the metric on M is considered to be given independently
of the immersion or not.
The most important examples of induced metrics occur on submanifolds. Sup-
pose M M is an (immersed or embedded) submanifold, with or without bound-
ary. The induced metric on M is the metric g D gz induced by the inclusion map
W M ,! M . With this metric, M is called a Riemannian submanifold (or Rie-
mannian submanifold with boundary) of M . We always consider submanifolds
(with or without boundary) of Riemannian manifolds to be endowed with the
induced metrics unless otherwise specified.
If .M; g/ is a Riemannian submanifold of M ; gz , then for every p 2 M and v; w 2
Tp M , the definition of the induced metric reads

gp .v; w/ D gzp d p .v/; d p .w/ :
Because we usually identify Tp M with its image in Tp M under d p , and think of

d p as an inclusion map, what this really amounts to is gp .v; w/ D gzp .v; w/ for
v; w 2 Tp M . In other words, the induced metric g is just the restriction of gz to
vectors tangent to M . Many of the examples of Riemannian metrics that we will
encounter are obtained in this way, starting with the following.
Example 2.13 (Spheres). For each positive integer n, the unit n-sphere Sn RnC1
is an embedded n-dimensional submanifold. The Riemannian metric induced on Sn
by the Euclidean metric is denoted by gV and known as the round metric or standard
metric on Sn . //
The next lemma describes

one of the most important tools for studying Riemann-
ian submanifolds. If M ; gz is an m-dimensional smooth Riemannian manifold and
M M is an n-dimensional submanifold (both with or without boundary), a local
frame .E1 ; : : : ; Em / for M on an open subset Uz M is said to be adapted to M
if the first n vector fields .E1 ; : : : ; En / are tangent to M . In case M has empty
boundary (so that slice coordinates are available), adapted local orthonormal frames
are easy to find.

Proposition 2.14 (Existence of Adapted Orthonormal Frames). Let M ; gz be a
Riemannian manifold (without boundary), and let M M be an embedded smooth
submanifold with or without boundary. Given p 2 M , there exist a neighborhood Uz
of p in M and a smooth orthonormal frame for M on Uz that is adapted to M .
I Exercise 2.15. Prove the preceding proposition. [Hint: Apply the Gram–Schmidt algo-
rithm to a coordinate frame in slice coordinates (see Prop. A.22).]

Suppose M ; gz is a Riemannian manifold and M M is a smooth submanifold

with or without boundary in M . Given p 2 M , a vector 2 Tp M is said to be
normal to M if h; wi D 0 for every w 2 Tp M . The space of all vectors normal
to M at p is a subspace of Tp M , called the normal space at p and denoted by
?
Np M D .Tp M / . At each p 2 M , the ambient tangent space Tp M splits as an

orthogonal direct sum Tp M D Tp M ˚ Np M . A section N of the ambient tangent
jM is called a normal vector field along M if Np 2 Np M for each
bundle T M
p 2 M . The set a
NM D Np M
p2M
is called the normal bundle of M .
Proposition 2.16 (The Normal Bundle). If M is a Riemannian m-manifold and

M M is an immersed or embedded n-dimensional submanifold with or without
boundary, then NM is a smooth rank-.m n/ vector subbundle of the ambient tan-
jM . There are smooth bundle homomorphisms
gent bundle T M
jM ! TM;
> W T M jM ! NM ;
? W T M
called the tangential and normal projections, that for each p 2 M restrict to
to Tp M and Np M , respectively.
orthogonal projections from Tp M
Proof. Given any point p 2 M , Theorem A.16 shows that there is a neighborhood
U of p in M that is embedded in M , and then Proposition 2.14 shows that there is a
smooth orthonormal frame .E1 ; : : : ; Em / that is adapted to U on some neighborhood
Uz of p in M
. This means that the restrictions of .E1 ; : : : ; En / to Uz \ U form a local
orthonormal frame for M . Given such an adapted frame, the restrictions of the last
m n vector fields .EnC1 ; : : : ; Em / to M form a smooth local frame for NM , so it
follows from Lemma A.34 that NM is a smooth subbundle.
The bundle homomorphisms > and ? are defined pointwise as orthogonal
projections onto the tangent and normal spaces, respectively, which shows that they
are uniquely defined. In terms of an adapted orthonormal frame, they can be written

> X 1 E1 C C X m Em D X 1 E1 C C X n En ;

? X 1 E1 C C X m Em D X nC1 EnC1 C C X m Em ;
which shows that they are smooth.

In case M is a manifold with boundary, the preceding constructions do not
always work, because there is not a fully general construction of slice coordinates
in that case. However, there is a satisfactory result in case the submanifold is the
boundary itself, using boundary coordinates in place of slice coordinates.
Suppose .M; g/ is a Riemannian manifold with boundary. We will always con-
sider @M to be a Riemannian submanifold with the induced metric.
Proposition 2.17 (Existence of Outward-Pointing Normal). If .M; g/ is a smooth

Riemannian manifold with boundary, the normal bundle to @M is a smooth rank-1
vector bundle over @M , and there is a unique smooth outward-pointing unit normal
vector field along all of @M .
I Exercise 2.18. Prove this proposition. [Hint: Use the paragraph preceding Prop. B.17
as a starting point.]
Computations on a submanifold M M are usually carried out most conve-

niently in terms of a smooth local parametrization: this is a smooth map X W U !
M , where U is an open subset of Rn (or Rn in case M has a boundary), such that
C
X.U / is an open subset of M , and such that X , regarded as a map from U into M ,
is a diffeomorphism onto its image. Note that we can think of X either as a map into
M or as a map into M ; both maps are typically denoted by the same symbol X . If
we put V D X.U / M and ' D X 1 W V ! U , then .V; '/ is a smooth coordinate
chart on M .
Suppose .M; g/ is a Riemannian submanifold of M ; gz and X W U ! M
is a
smooth local parametrization of M . The coordinate representation of g in these
coordinates is given by the following 2-tensor field on U :
1
' g D X g D X gz D . ı X / gz:
Since ıX is just the map X itself, regarded as a map into M , this is really just X gz.
The simplicity of the formula for the pullback of a tensor field makes this expression
exceedingly easy to compute, once a coordinate expression for gz is known. For
example, if M is an immersed n-dimensional Riemannian submanifold of Rm and
X W U ! Rm is a smooth local parametrization of M , the induced metric on U is
just
0 12
Xm Xm Xn
@X i Xm Xn
@X i @X i j k
g D X gx D .dX i /2 D @ dujA
D du du :
@uj @uj @uk
iD1 iD1 j D1 iD1 j;kD1
Example 2.19 (Metrics in Graph Coordinates). If U Rn is an open set and

f W U ! R is a smooth function, then the graph of f is the subset .f / D
f.x; f .x// W x 2 U g RnC1 , which is an embedded submanifold of dimension n. It
has a global parametrization X W U ! RnC1 called a graph parametrization, given
by X.u/ D .u; f .u//; the corresponding coordinates .u1 ; : : : ; un / on M are called
graph coordinates. In graph coordinates, the induced metric of .f / is
2 2 1 2
X gx D X dx 1 C C dx nC1 D du C C .dun /2 C df 2 :
Applying this to the upper hemisphere of S2 with the parametrization X W B2 ! R3

given by
p
X.u; v/ D u; v; 1 u2 v 2 ;
we see that the round metric on S2 can be written locally as

u du C v dv 2
gV D X gx D du2 C dv 2 C p
1 u2 v 2
2 2
1 v du C 1 u dv C 2uv du dv
2 2
D : //
1 u2 v 2
Example 2.20 (Surfaces of Revolution). Let H be the half-plane f.r; z/ W r > 0g,
and suppose C H is an embedded 1-dimensional submanifold. The surface of
revolution determined by C is the subset SC R3 given by
n p o
SC D .x; y; z/ W x2 C y2; z 2 C :
The set C is called its generating curve (see Fig. 2.1). Every smooth local parametri-
zation .t / D .a.t /; b.t // for C yields a smooth local parametrization for SC of the
form
X.t; / D .a.t / cos ; a.t / sin ; b.t //; (2.11)
provided that .t; / is restricted to a sufficiently small open set in the plane. The
t -coordinate curves t 7! X.t; 0 / are called meridians, and the -coordinate curves
7! X.t0 ; / are called latitude circles. The induced metric on SC is
Fig. 2.1: A surface of revolution
X gx D d.a.t / cos /2 C d.a.t / sin /2 C d.b.t //2

D .a0 .t / cos dt a.t / sin d /2
C .a0 .t / sin dt C a.t / cos d /2 C .b 0 .t /dt /2

D a0 .t /2 C b 0 .t /2 dt 2 C a.t /2 d 2 :
In particular, if is a unit-speed curve (meaning that j 0 .t /j2 D a0 .t /2 Cb 0 .t /2 1),

this reduces to dt 2 C a.t /2 d 2 .
Here are some examples of surfaces of revolution and their induced metrics.
If C is the semicircle r 2 C z 2 D 1, parametrized by .'/D.sin '; cos '/ for 0<
' < , then SC is the unit sphere (minus the north and south poles). The map
X.'; / D .sin ' cos ; sin ' sin ; cos '/ constructed above is called the spher-
ical coordinate parametrization, and the induced metric is d' 2 C sin2 ' d 2 .
(This example is the source of the terminology for meridians and latitude cir-
cles.)
If C is the circle .r 2/2 Cz 2 D 1, parametrized by .t / D .2Ccos t; sin t /, we
obtain a torus of revolution, whose induced metric is dt 2 C .2 C cos t /2 d 2 .
If C is a vertical line parametrized by .t / D .1; t /, then SC is the unit cylinder
x 2 C y 2 D 1, and the induced metric is dt 2 C d 2 . Note that this means that
the parametrization X W R2 ! R3 is an isometric immersion. //
Example 2.21 (The n-Torus as a Riemannian Submanifold). The smooth cover-

ing map X W Rn ! T n described in Example A.52 restricts to a smooth local pa-
rametrization on any sufficiently small open subset of Rn , and the induced metric is
equal to the Euclidean metric in .ui / coordinates, and therefore the induced metric
on T n is flat. //
I Exercise 2.22. Verify the claims in Examples 2.19–2.21.
Riemannian Products
Next we consider products. If .M1 ; g1 / and .M2 ; g2 / are Riemannian manifolds, the
product manifold M1 M2 has a natural Riemannian metric g D g1 ˚ g2 , called the
product metric, defined by
ˇ ˇ
g.p1 ;p2 / .v1 ; v2 /; .w1 ; w2 / D g1 ˇp .v1 ; w1 / C g2 ˇp .v2 ; w2 /; (2.12)
1 2
where .v1 ; v2 / and .w1 ; w2 / are elements of Tp1 M1 ˚ Tp2 M2 , which is natu-
rally identified with T.p1 ;p2 / .M1 M2 /. Smooth local coordinates .x 1 ; : : : ; x n / for
M1 and .x nC1 ; : : : ; x nCm / for M2 give coordinates .x 1 ; : : : ; x nCm / for M1 M2 .
In terms of these coordinates, the product metric has the local expression g D
gij dx i dx j , where .gij / is the block diagonal matrix

.g1 /ab 0
.gij / D I
0 .g2 /cd
here the indices a; b run from 1 to n, and c; d run from n C 1 to n C m. Product

metrics on products of three or more Riemannian manifolds are defined similarly.
I Exercise 2.23. Show that the induced metric on T n described in Exercise 2.21 is equal
to the product metric obtained from the usual induced metric on S1 R2 .
Here is an important generalization of product metrics. Suppose .M1 ; g1 / and

.M2 ; g2 / are two Riemannian manifolds, and f W M1 ! RC is a strictly positive
smooth function. The warped product M1 f M2 is the product manifold M1 M2
endowed with the Riemannian metric g D g1 ˚ f 2 g2 , defined by
ˇ ˇ
g.p1 ;p2 / .v1 ; v2 /; .w1 ; w2 / D g1 ˇp .v1 ; w1 / C f .p1 /2 g2 ˇp .v2 ; w2 /;
1 2
where .v1 ; v2 /; .w1 ; w2 / 2 Tp1 M1 ˚ Tp2 M2 as before. (Despite the similarity with
the notation for product metrics, g1 ˚ f 2 g2 is generally not a product metric unless
f is constant.) A wide variety of metrics can be constructed in this way; here are
just a few examples.
Example 2.24 (Warped Products).

(a) With f 1, the warped product M1 f M2 is just the space M1 M2 with
the product metric.
(b) Every surface of revolution can be expressed as a warped product, as follows.
Let H be the half-plane f.r; z/ W r > 0g, let C H be an embedded smooth
1-dimensionalsubmanifold,andletSC R3 denotethecorrespondingsurface of
revolution as in Example 2.20. Endow C with the Riemannian metric induced
from the Euclidean metric on H , and let S1 be endowed with its standard
metric. Let f W C ! R be the distance to the z-axis: f .r; z/ D r. Then Problem
2-3 shows that SC is isometric to the warped product C f S1 .
(c) If we let denote the standard coordinate function on RC R, then the map
˚. ; !/ D ! gives an isometry from the warped product RC Sn1 to
Rn X f0g with its Euclidean metric (see Problem 2-4). //
Riemannian Submersions
Unlike submanifolds and products of Riemannian manifolds, which automatically

inherit Riemannian metrics of their own, quotients of Riemannian manifolds inherit
Riemannian metrics only under very special circumstances. In this section, we see
what those circumstances are.
Suppose M and M are smooth manifolds, W M ! M is a smooth submersion,
and gz is a Riemannian metric on M . By the submersion level set theorem (Corollary
y D 1 .y/ is a properly embedded smooth submanifold of M
A.25), each fiber M .

At each point x 2 M , we define two subspaces of the tangent space Tx M as follows:
the vertical tangent space at x is

Vx D Ker d x D Tx M .x/
(that is, the tangent space to the fiber containing x), and the horizontal tangent
space at x is its orthogonal complement:
Hx D .Vx /? :
Then the tangent space Tx M decomposes as an orthogonal direct sum Tx M D

Hx ˚ Vx . Note that the vertical space is well defined for every submersion, because
it does not refer to the metric; but the horizontal space depends on the metric.
A vector field on M is said to be a horizontal vector field if its value at each point
lies in the horizontal space at that point; a vertical vector field is defined similarly.
Given a vector field X on M , a vector field Xz on M is called a horizontal lift of X

if X is horizontal and -related to X . (The latter property means that d x Xz x D
z
X.x/ for each x 2 M .)
The next proposition is the principal tool for doing computations on Riemannian
submersions.
and M be
Proposition 2.25 (Properties of Horizontal Vector Fields). Let M

smooth manifolds, let W M ! M be a smooth submersion, and let gz be a Rie-
mannian metric on M.
(a) Every smooth vector field W on M can be expressed uniquely in the form
W D W H C W V , where W H is horizontal, W V is vertical, and both W H
and W V are smooth.
.
(b) Every smooth vector field on M has a unique smooth horizontal lift to M
(c) For every x 2 M and v 2 Hx , there is a vector field X 2 X.M / whose hori-
zontal lift Xz satisfies Xz x D v.
be arbitrary. Because is a smooth submersion, the rank theorem

Proof. Let p 2 M
(Theorem A.15) shows that there exist smooth coordinate charts Uz ; x i centered
j
at p and U; u centered at .p/ in which has the coordinate representation

x 1 ; : : : ; x n ; x nC1 ; : : : ; x m D x 1 ; : : : ; x n ;
where m D dim M and n D dim M . It follows that at each point q 2 Uz , the

vertical space Vq is spanned by the vectors @nC1 jq ; : : : ; @m jq . (It probably will
not be the case, however, that the horizontal space is spanned by the other n
basis
vectors.) If we apply the Gram–Schmidt algorithm to the ordered frame
@nC1 ; : : : ; @m ; @1 ; : : : ; @n , we obtain a smooth orthonormal frame .E1 ; : : : ; Em / on
Uz such that Vq is spanned by .E1 jq ; : : : ; Emn jq / at each q 2 Uz . It follows that Hq
is spanned by .EmnC1 jq ; : : : ; Em jq /.

Now let W 2 X M be arbitrary. At each point q 2 M , Wq can be written
uniquely as a sum of a vertical vector plus a horizontal vector, thus defining a
decomposition W D W V C W H into rough vertical and horizontal vector fields.
To see that they are smooth, just note that in a neighborhood of each point we
can express W in terms of a frame .E1 ; : : : ; Em / of the type
constructed above as
W D W 1 E1 C C W m Em with smooth coefficients W i , and then it follows that
W V D W 1 E1 C C W mn Emn and W H D W mnC1 EmnC1 C C W m Em ,
both of which are smooth.
The proofs of (b) and (c) are left to Problem 2-5.
The fact that every horizontal vector at a point of M can be extended to a hor-

izontal lift on all of M (part (c) of the preceding proposition) is highly useful for
computations. It is important to be aware, though, that not every horizontal vector
field on M is a horizontal lift, as the next exercise shows.
I Exercise 2.26. Let W R2 ! R be the projection map .x; y/ D x, and let W be the
smooth vector field y@x on R2 . Show that W is horizontal, but there is no vector field on
R whose horizontal lift is equal to W .
Now we can identify some quotients of Riemannian manifolds that inherit met-
rics of their own.
Let
us begin by describing what such a metric should look like.
Suppose M ; gz and .M; g/ are Riemannian manifolds, and W M ! M is a
smooth submersion. Then is said to be a Riemannian submersion if for each
x2M , the differential d x restricts to a linear isometry from Hx onto T.x/ M . In
other words, gzx .v; w/ D g.x/ .d x .v/; d x .w// whenever v; w 2 Hx .
Example 2.27 (Riemannian Submersions).
(a) The projection W RnCk ! Rn onto the first n coordinates is a Riemannian

submersion if RnCk and Rn are both endowed with their Euclidean metrics.
(b) If M and N are Riemannian manifolds and M N is endowed with the product
metric, then both projections M W M N ! M and N W M N ! N are
Riemannian submersions.
(c) If M f N is a warped product manifold, then the projection M W M f N !

M is a Riemannian submersion, but N typically is not. //

Given a Riemannian manifold M ; gz and a surjective submersion W M ! M,
it is almost never the case that there is a metric on M that makes into a Riemannian

submersion. It is not hard to see why: for this to be the case, whenever p1 ; p2 2 M
are two points in the same fiber 1 .y/, the linear maps d pi jHpi 1 W Ty M !
Hpi both have to pull gz back to the same inner product on Ty M .
There is, however, an important special case in which there is such a metric.
Suppose W M ! M is a smooth surjective submersion, and G is a group acting on
M . (See Appendix C for a review of the basic definitions and terminology regarding
group actions on manifolds.) We say that the action is vertical if every element
' 2 G takes each fiber to itself, meaning that .' p/ D .p/ for all p 2 M . The

action is transitive on fibers if for each p; q 2 M such that .p/ D .q/, there exists
' 2 G such that ' p D q.
If in addition M is endowed with a Riemannian metric, the action is said to be
an isometric action or an action by isometries, and the metric is said to be invariant
under G , if the map x 7! ' x is an isometry for each ' 2 G. In that case, provided
the action is effective (so that different elements of G define different isometries of
M ), we can identify G with a subgroup of Iso M ; g . Since an isometry is, in par-
ticular, a diffeomorphism, every isometric action is an action by diffeomorphisms.

Theorem 2.28. Let M ; gz be a Riemannian manifold, let W M ! M be a surjec-
tive smooth submersion, and let G be a group acting on M . If the action is isometric,
vertical, and transitive on fibers, then there is a unique Riemannian metric on M
such that is a Riemannian submersion.
Proof. Problem 2-6.

The next corollary describes one important situation to which the preceding the-
orem applies.

Corollary 2.29. Suppose M ; gz is a Riemannian manifold, and G is a Lie group
. Then the orbit space M D
acting smoothly, freely, properly, and isometrically on M

M =G has a unique smooth manifold structure and Riemannian metric such that
is a Riemannian submersion.
Proof. Under the given hypotheses, the quotient manifold theorem (Thm. C.17)
shows that M has a unique smooth manifold structure such that the quotient map
W M ! M is a smooth submersion. It follows easily from the definitions in that
case that the given action of G on M is vertical and transitive on fibers. Since the
action is also isometric, Theorem 2.28 shows that M inherits a unique Riemannian
metric making into a Riemannian submersion.
Here is an important example of a Riemannian metric defined in this way. A
larger class of such metrics is described in Problem 2-7.
Example 2.30 (The Fubini–Study Metric). Let n be a positive integer, and con-
sider the complex projective space CP n defined in Example C.19. That example
shows that the map W C nC1 X f0g ! CP n sending each point in C nC1 X f0g to
its span is a surjective smooth submersion. Identifying C nC1 with R2nC2 endowed
with its Euclidean metric, we can view the unit sphere S2nC1 with its round metric gV
as an embedded Riemannian submanifold of C nC1 X f0g. Let p W S2nC1 ! CP n de-
note the restriction of the map . Then p is smooth, and it is surjective, because every
1-dimensional complex subspace contains elements of unit norm. We need to show
that it is a submersion. Let z0 2 S2nC1 and set
0 D p.z0 / 2 CP n . Since is a smooth
submersion, it has a smooth local section W U ! C nC1 defined on a neighborhood
U of
0 and satisfying .
0 / D z0 (Thm. A.17). Let W C nC1 X f0g ! S2nC1 be the
radial projection onto the sphere:
z
.z/ D :
jzj
Since dividing an element of C nC1 by a nonzero scalar does not change its span, it
follows that p ı D . Therefore, if we set z D ı , we have p ı z D p ı ı D
ı D IdU , so z is a local section of p. By Theorem A.17, this shows that p is a
submersion.
Define an action of S1 on S2nC1 by complex multiplication:

z 1 ; : : : ; z nC1 D z 1 ; : : : ; z nC1 ;

for 2 S1 (viewed as a complex number of norm 1) and z D z 1 ; : : : ; z nC1 2
S2nC1 . This is easily seen to be isometric, vertical, and transitive on fibers of p.
By Theorem 2.28, therefore, there is a unique metric on CP n such that the map
p W S2nC1 ! CP n is a Riemannian submersion. This metric is called the Fubini–
Study metric; you will have a chance to study its geometric properties in Problems
3-19 and 8-13. //
Riemannian Coverings
Another important special case of
Riemannian submersions occurs in the context of
covering maps. Suppose M ; gz and .M; g/ are Riemannian manifolds. A smooth
! M is called a Riemannian covering if it is a local isometry.
covering map W M
Proposition 2.31. Suppose W M ! M is a smooth normal covering map, and gz is

any metric on M that is invariant under all covering automorphisms. Then there is
a unique metric g on M such that is a Riemannian covering.
Proof. Proposition A.49 shows that is a surjective smooth submersion. The au-
tomorphism group acts vertically by definition, and Proposition C.21 shows that it
acts transitively on fibers when the covering is normal. It then follows from Theorem
2.28 that there is a unique metric g on M such that is a Riemannian submersion.
Basic Constructions on Riemannian Manifolds 25
Since a Riemannian submersion between manifolds of the same dimension is a local

isometry, it follows that is a Riemannian covering.

Proposition 2.32. Suppose M ; gz is a Riemannian manifold, and is a discrete
Lie group acting smoothly, freely, properly, and isometrically on M . Then M
=

has a unique Riemannian metric such that the quotient map W M ! M = is a
normal Riemannian covering.
Proof. Proposition C.23 shows that is a smooth normal covering map, and Propo-
= has a unique Riemannian metric such that is a
sition 2.31 shows that M D M
Riemannian covering.

Corollary 2.33. Suppose .M; g/ and M ; gz are connected Riemannian manifolds,

W M ! M is a normal Riemannian covering map, and D Aut M . Then M

is isometric to M = .
Proof. Proposition C.20 shows that with the discrete topology, is a discrete Lie
group acting smoothly, freely, and properly on M , and then Proposition C.23 shows

that M = is a smooth manifold and the quotient map q W M !M = is a smooth
normal covering map. The fact that both and q are normal coverings implies that
acts transitively on the fibers of both maps, so the two maps are constant on
each other’s fibers. Proposition A.19 then implies that there is a diffeomorphism
F W M !M = that satisfies q ı F D . Because both q and are local isometries,
F is too, and because it is bijective it is a global isometry.
Example 2.34. The two-element group D f˙1g acts smoothly, freely, properly,
and isometrically on Sn by multiplication. Example C.24 shows that the quotient
space is diffeomorphic to the real projective space RP n and the quotient map
q W Sn ! RP n is a smooth normal covering map. Because the action is isomet-
ric, Proposition 2.32 shows that there is a unique metric on RP n such that q is a
Riemannian covering. //
Example 2.35 (The Open Möbius Band). The open Möbius band is the quotient
space M D R2 =Z, where Z acts on R2 by n .x; y/ D .x C n; .1/n y/. This action
is smooth, free, proper, and isometric, and therefore M inherits a flat Riemannian
metric such that the quotient map is a Riemannian covering. (See Problem 2-8.) //
I Exercise 2.36. Let T n R2n be the n-torus with its induced metric. Show that the
map X W Rn ! T n of Example 2.21 is a Riemannian covering.
Basic Constructions on Riemannian Manifolds
Every Riemannian metric yields an abundance of useful constructions on manifolds,

besides the obvious ones of lengths of vectors and angles between them. In this
section we describe the most basic ones. Throughout this section M is a smooth
manifold with or without boundary.
Raising and Lowering Indices
One elementary but important property of Riemannian metrics is that they allow us
to convert vectors to covectors and vice versa. Given a Riemannian metric g on M ,
we define a bundle homomorphism gy W TM ! T M by setting
gy.v/.w/ D gp .v; w/
for all p 2 M and v; w 2 Tp M . If X and Y are smooth vector fields on M , this

yields
gy.X /.Y / D g.X; Y /;
which implies, first, that gy.X /.Y / is linear over C 1 .M / in Y and thus gy.X / is
a smooth covector field by the tensor characterization lemma (Lemma B.6); and
second, that the covector field gy.X / is linear over C 1 .M / as a function of X , and
thus gy is a smooth bundle homomorphism.
Given a smooth local frame .Ei / and its dual coframe "i , let g D gij "i "j be
the local expression for g. If X D X i Ei is a smooth vector field, the covector field
gy.X / has the coordinate expression

gy.X / D gij X i "j :
Thus the matrix of gy in any local frame is the same as the matrix of g itself.
Given a vector field X , it is standard practice to denote the components of the
covector field gy.X / by
Xj D gij X i ;
so that
gy.X / D Xj "j ;
and we say that gy.X / is obtained from X by lowering an index. With this in mind,
the covector field gy.X / is denoted by X [ and called X flat, borrowing from the
musical notation for lowering a tone.
Because the matrix .gij / is nonsingular at each point, the map gy is invertible, and
1
of gy is just the inverse matrix of .gij /. We denote this inverse matrix
the matrix
by g ij , so that g ij gj k D gkj g j i D ıki . The symmetry of gij easily implies that
ij
g is also symmetric in i and j . In terms of a local frame, the inverse map gy1 is
given by
gy1 .!/ D ! i Ei ;
where
! i D g ij !j : (2.13)
If ! is a covector field, the vector field gy1 .!/ is called (what else?) ! sharp and
denoted by ! ] , and we say that it is obtained from ! by raising an index. The two
inverse isomorphisms [ and ] are known as the musical isomorphisms.
Probably the most important application of the sharp operator is to extend the
classical gradient operator to Riemannian manifolds. If g is a Riemannian metric
on M and f W M ! R is a smooth function, the gradient of f is the vector field

grad f D .df /] obtained from df by raising an index. Unwinding the definitions,
we see that grad f is characterized by the fact that
˝ ˛
dfp .w/ D grad f jp ; w for all p 2 M; w 2 Tp M; (2.14)
and has the local basis expression

grad f D g ij Ei f Ej :
Thus if .Ei / is an orthonormal frame, then grad f is the vector field whose
components are the same as the components of df ; but in other frames, this will
not be the case.
The next proposition shows that the gradient has the same geometric interpre-
tation on a Riemannian manifold as it does in Euclidean space. If f is a smooth
real-valued function on a smooth manifold M , recall that a point p 2 M is called a
regular point of f if dfp ¤ 0, and a critical point of f otherwise; and a level set
f 1 .c/ is called a regular level set if every point of f 1 .c/ is a regular point of f
(see Appendix A). Corollary A.26 shows that each regular level set is an embedded
smooth hypersurface in M .
Proposition 2.37. Suppose .M; g/ is a Riemannian manifold, f 2 C 1 .M /, and

R M is the set of regular points of f . For each c 2 R, the set Mc D f 1 .c/ \ R,
if nonempty, is an embedded smooth hypersurface in M , and grad f is everywhere
normal to Mc .
Proof. Problem 2-9.

The flat and sharp operators can be applied to tensors of any rank, in any index
position, to convert tensors from covariant to contravariant or vice versa. Formally,
this operation is defined as follows: if F is any .k; l/-tensor and i 2 f1; : : : ; k C lg is
any covariant index position for F (meaning that the i th argument is a vector, not a
covector), we can form a new tensor F ] of type .k C 1; l 1/ by setting

F ] ˛1 ; : : : ; ˛kCl D F ˛1 ; : : : ; ˛i1 ; ˛i] ; ˛iC1 ; : : : ; ˛kCl
whenever ˛1 ; : : : ; ˛kC1 are vectors or covectors as appropriate. In any local frame,

the components of F ] are obtained by multiplying the components of F by g kl and
contracting one of the indices of g kl with the i th index of F . Similarly, if i is a
contravariant index position, we can define a .k 1; l C 1/-tensor F [ by

F [ ˛1 ; : : : ; ˛kCl D F ˛1 ; : : : ; ˛i1 ; ˛i[ ; ˛iC1 ; : : : ; ˛kCl :
In components, it is computed by multiplying by gkl and contracting.

For example, if A is a mixed 3-tensor given in terms of a local frame by
A D Ai j k "i ˝ Ej ˝ "k
(see (B.6)), we can lower its middle index to obtain a covariant 3-tensor A[ with
components
Aij k D gj l Ai l k :
To avoid overly cumbersome notation, we use the symbols F ] and F [ without
explicitly specifying which index position the sharp or flat operator is to be applied
to; when there is more than one choice, we will always stipulate in words what is
meant.
Another important application of the flat and sharp operators is to extend the
trace operator introduced in Appendix B to covariant tensors. If h is any covariant
k-tensor field on a Riemannian manifold with k 2, we can raise one of its indices
(say the last one for definiteness) and obtain a .1; k 1/-tensor h] . The trace of h] is
thus a well-defined covariant .k 2/-tensor field (see Exercise B.3). We define the
trace of h with respect tog as

trg h D tr h] :
Sometimes we may wish to raise an index other than the last, or to take the trace on
a pair of indices other than the last covariant and contravariant ones. In each such
case, we will say in words what is meant.
The most important case is that of a covariant 2-tensor field. In this case, h] is
a .1; 1/-tensor field, which can equivalently be regarded as an endomorphism field,
and trg h is just the ordinary trace of this endomorphism field. In terms of a basis,
this is
trg h D hi i D g ij hij :
In particular, in an orthonormal frame this is the ordinary trace of the matrix .hij /
(the sum of its diagonal entries); but if the frame is not orthonormal, then this trace
is different from the ordinary trace.
I Exercise 2.38. If g is a Riemannian metric on M and .E i / is a local frame on M ,

there is a potential ambiguity about what the expression g ij represents: we have defined
it to mean the inverse matrix of .gij /, but one could also interpret it as the components of
the contravariant 2-tensor field g ]] obtained by raising both of the indices of g. Show that
these two interpretations lead to the same result.
Inner Products of Tensors
A Riemannian metric yields, by definition, an inner product on tangent vectors at

each point. Because of the musical isomorphisms between vectors and covectors, it
is easy to carry the inner product over to covectors as well.
Suppose g is a Riemannian metric on M , and x 2 M . We can define an inner
product on the cotangent space Tx M by
h!; ig D h! ] ; ] ig :
(Just as with inner products of vectors, we might sometimes omit g from the notation
when the metric is understood.) To see how to compute this, we just use the basis
formula (2.13) for the sharp operator, together with the relation gkl g ki D glk g ki D
ıli , to obtain

h!; i D gkl g ki !i g lj j
D ıli g lj !i j
D g ij !i j :
In other
words, the inner product on covectors is represented by the inverse ma-
trix g ij . Using our conventions for raising and lowering indices, this can also be
written
h!; i D !i i D ! j j :
I Exercise 2.39. Let .M; g/ be a Riemannian manifold with or without boundary, let
.Ei / be a local frame for M , and let ."i / be its dual coframe. Show that the following are
equivalent:
(a) .Ei / is orthonormal.
(b) ."i / is orthonormal.
(c) ."i /] D Ei for each i.
This construction can be extended to tensor bundles of any rank, as the following
proposition shows. First a bit of terminology: if E ! M is a smooth vector bundle, a
smooth fiber metric on E is an inner product on each fiber Ep that varies smoothly,
in the sense that for any (local) smooth sections ; of E, the inner product h; i
is a smooth function.
Proposition 2.40 (Inner Products of Tensors). Let .M; g/ be an n-dimensional

Riemannian manifold with or without boundary. There is a unique smooth fiber
metric on each tensor bundle T .k;l/ TM with the property that if ˛1 ; : : : ; ˛kCl ,
ˇ1 ; : : : ; ˇkCl are vector or covector fields as appropriate, then
h˛1 ˝ ˝ ˛kCl ; ˇ1 ˝ ˝ ˇkCl i D h˛1 ; ˇ1 i : : : h˛kCl ; ˇkCl i: (2.15)
With this inner product, if .E1 ; : : : ; En / is a local orthonormal frame for TM and
."1 ; : : : ; "n / is the corresponding dual coframe, then the collection of tensor fields
Ei1 ˝ ˝ Eik ˝ "j1 ˝ ˝ "jl as all the indices range from 1 to n forms a local
orthonormal frame for T .k;l/ .Tp M /. In terms of any (not necessarily orthonormal)
frame, this fiber metric satisfies
i :::i
hF; Gi D gi1 r1 gik rk g j1 s1 g jl sl Fj11:::jkl Gsr11:::s
:::rk
l
: (2.16)
If F and G are both covariant, this can be written
hF; Gi D Fj1 :::jl G j1 :::jl ;

where the last factor on the right represents the components of G with all of its
indices raised:
G j1 :::jl D g j1 s1 g jl sl Gs1 :::sl :
Proof. Problem 2-11.
The Volume Form and Integration

Another important construction provided by a metric on an oriented manifold is a
canonical volume form.
Proposition 2.41 (The Riemannian Volume Form). Let .M; g/ be an oriented Rie-
mannian n-manifold with or without boundary. There is a unique n-form d Vg on
M , called the Riemannian volume form, characterized by any one of the following
three equivalent properties:

(a) If "1 ; : : : ; "n is any local oriented orthonormal coframe for T M , then
d Vg D "1 ^ ^ "n :
(b) If .E1 ; : : : ; En / is any local oriented orthonormal frame for TM , then
d Vg .E1 ; : : : ; En / D 1:
1
(c) If x ; : : : ; x n are any oriented local coordinates, then
q
d Vg D det.gij / dx 1 ^ ^ dx n :

The significance of the Riemannian volume form is that it allows us to integrate
functions on an oriented Riemannian manifold, not just differential forms. If f is
a continuous, compactly supported real-valued function on an oriented Riemannian
n-manifold .M; g/ with or without
R boundary, then f d Vg is a compactly supported
n-form. Therefore, the integral M f d Vg makes sense, and we define it to be the
integral of f over M . Similarly, if M is compact, the volume of M is defined to
be Z Z
Vol.M / D d Vg D 1 d Vg :
M M
In particular, if D M is a regular domain (a closed, embedded codimension-0
submanifold with boundary), we can apply these definitions to D with its induced
metric and thereby make sense of the integral of f over D and, in case D is com-
pact, the volume of D. R
The notation d Vg is chosen to emphasize the similarity of the integral M f d Vg
with the standard integral of a function over an open subset of Rn . It is not meant
to imply that d Vg is an exact form; in fact, if M is a compact oriented manifold
without boundary, then d Vg is never exact, because its integral over M is positive,
and exact forms integrate to zero by Stokes’s theorem.
Because there are two conventions in common use for the wedge product (see
p. 401), it should be noted that properties (a) and (c) of Proposition 2.41 are the
same regardless of which convention is used; but property (b) holds only for the
determinant convention that we use. If the Alt convention is used, the number 1
should be replaced by 1=nŠ in that formula.

I Exercise 2.42. Suppose .M; g/ and M;g
z are oriented Riemannian manifolds, and
is an orientation-preserving isometry. Prove that ' d Vgz D d Vg .
'W M ! M
For Riemannian hypersurfaces, we have the following important characterization

of the volume form on the hypersurface in terms of that of the ambient manifold.
If X is a vector field and is a differential form, recall that X ³ denotes interior
multiplication of by X (see p. 401).
Proposition
2.43. Suppose M is a hypersurface in an oriented Riemannian mani-
fold M ; gz and g is the induced metric on M . Then M is orientable if and only if
there exists a global unit normal vector field N for M , and in that case the volume
form of .M; g/ is given by
ˇ
d Vg D N ³ d Vgz ˇM : (2.17)

When M is not orientable, we can still define integrals of functions, but now we
have to use densities instead of differential forms (see pp. 405–406).
Proposition 2.44 (The Riemannian Density). If .M; g/ is any Riemannian mani-
fold, then there is a unique smooth positive density on M , called the Riemannian
density, with the property that
.E1 ; : : : ; En / D 1 (2.18)
for every local orthonormal frame .Ei /.

I Exercise 2.45. Prove this proposition by showing that can be defined in terms of any
local orthonormal frame by
D j"1 ^ ^ "n j:
Let .M; g/ be a Riemannian manifold (with or without boundary). If M is ori-

ented and d Vg is its Riemannian volume form, then its Riemannian density is easily
seen to be equal to jd Vg j. On the other hand, the Riemannian density is defined
whether M is oriented or not. It is customary to denote the Riemannian density by
the same notation d Vg that we use for the Riemannian volume form, and to specify
when necessary whether the notation refers to a density or a form. In either case,
R can define the integral of a compactly supported smooth function f W M ! R as
we
M f d Vg . This is to be interpreted as the integral of a density when M is nonori-
entable; when M is orientable, it can be interpreted either as the integral of a density
or as the integral of an n-form (with respect to some choice of orientation), because

both give the same result.
The Divergence and the Laplacian
In advanced calculus, you have undoubtedly been introduced to three important

differential operators involving vector fields on R3 : the gradient (which takes real-
valued functions to vector fields), divergence (vector fields to functions), and curl
(vector fields to vector fields). We have already described how the gradient operator
can be generalized to Riemannian manifolds (see equation (2.14)); now we can show
that the divergence operator also generalizes easily to that setting. Problem 2-27
describes a similar, but more limited, generalization of the curl.
Suppose .M; g/ is an oriented Riemannian n-manifold with or without boundary,
and d Vg is its volume form. If X is a smooth vector field on M , then X ³ d Vg is
an .n 1/-form. The exterior derivative of this .n 1/-form is a smooth n-form, so
it can be expressed as a smooth function multiplied by d Vg . That function is called
the divergence of X , and denoted by div X ; thus it is characterized by the following
formula:
d.X ³ d Vg / D .div X /d Vg : (2.19)
Even if M is nonorientable, in a neighborhood of each point we can choose an
orientation and define the divergence by (2.19), and then note that reversing the ori-
entation changes the sign of d Vg on both sides of the equation, so div X is well
defined, independently of the choice of orientation. In this way, we can define the
divergence operator on any Riemannian manifold with or without boundary, by
requiring that it satisfy (2.19) for any choice of orientation in a neighborhood of
each point.
The most important application of the divergence operator is the divergence the-
orem, which you will be asked to prove in Problem 2-22.
Using the divergence operator, we can define another important operator, this one
acting on real-valued functions. The Laplacian (or Laplace–Beltrami operator) is
the linear operator W C 1 .M / ! C 1 .M / defined by
u D div.grad u/: (2.20)
(Note that many books, including [LeeSM] and the first edition of this book, define
the Laplacian as div.grad u/. The main reason for choosing the negative sign is
so that the operator will have nonnegative eigenvalues; see Problem 2-24. But the
definition we give here is much more common in Riemannian geometry.)
The next proposition gives alternative formulas for these operators.
2.46. Let .M; g/ be a Riemannian manifold with or without boundary,

Proposition
and let x i be any smooth local coordinates on an open set U M . The coordinate
representations of the divergence and Laplacian are as follows:
Lengths and Distances 33

@ 1 @ ip
div X i i D p X det g ;
@x det g @x i

1 @ ij
p @u
u D p g det g ;
det g @x i @x j
where det g D det.gkl / is the determinant of the component matrix of g in these co-
ordinates. On Rn with the Euclidean metric and standard coordinates, these reduce
to
X n
@ @X i
div X i i D ;
@x @x i
iD1
X
n
@2 u
u D :
.@x i /2
iD1
Lengths and Distances
Perhaps the most important tool that a Riemannian metric gives us is the ability to
measure lengths of curves and distances between points. Throughout this section,
.M; g/ denotes a Riemannian manifold with or without boundary.
Without further qualification, a curve in M always means a parametrized curve,
that is, a continuous map W I ! M , where I R is some interval. Unless
otherwise specified, we will not worry about whether the interval is bounded or
unbounded, or whether it includes endpoints or not. To say that is a smooth curve
is to say that it is smooth as a map from the manifold (with boundary) I to M . If
I has one or two endpoints and M has empty boundary, then is smooth if and
only if it extends to a smooth curve defined on some open interval containing I . (If
@M ¤ ¿, then smoothness of has to be interpreted as meaning that each coor-
dinate representation of has a smooth extension to an open interval.) A curve
segment is a curve whose domain is a compact interval.
A smooth curve W I ! M has a well-defined velocity 0 .t / 2 T.t/ M for each
t 2 I . We say that is a regular curve if 0 .t / ¤ 0 for t 2 I . This implies that is
an immersion, so its image has no “corners” or “kinks.”
We wish to use curve segments as “measuring tapes” to define distances between
points in a connected Riemannian manifold. Many aspects of the theory become
technically much simpler if we work with a slightly larger class of curve segments
instead of just the regular ones. We now describe the appropriate class of curves.
If Œa; b R is a closed bounded interval, a partition of Œa; b is a finite sequence
.a0 ; : : : ; ak / of real numbers such that a D a0 < a1 < < ak D b. Each interval
Œai1 ; ai is called a subinterval of the partition. If M is a smooth manifold with
or without boundary, a (continuous) curve segment W Œa; b ! M is said to be
piecewise regular if there exists a partition .a0 ; : : : ; ak / of Œa; b such that jŒai 1 ;ai
is a regular curve segment (meaning it is smooth with nonvanishing velocity) for
i D 1; : : : ; k. For brevity, we refer to a piecewise regular curve segment as an admis-
sible curve, and any partition .a0 ; : : : ; ak / such that jŒai 1 ;ai is smooth for each
i an admissible partition for . (There are many admissible partitions for a given
admissible curve, because we can always add more points to the partition.) It is also
convenient to consider any map W fag ! M whose domain is a single real number
to be an admissible curve.
Suppose is an admissible curve and .a0 ; : : : ; ak / is an admissible partition for
it. At each of the intermediate partition points a1 ; : : : ; ak1 , there are two one-sided
velocity vectors, which we denote by

0 ai D lim 0 .t /;
t%ai

0 C
ai D lim 0 .t /:
t&ai
They are both nonzero, but they need not be equal.

If W I ! M is a smooth curve, we define a reparametrization of to be a curve
of the form z D ı ' W I 0 ! M , where I 0 R is another interval and ' W I 0 ! I
is a diffeomorphism. Because intervals are connected, ' is either strictly increasing
or strictly decreasing on I 0 . We say that z is a forward reparametrization if ' is
increasing, and a backward reparametrization if it is decreasing.
For an admissible curve W Œa; b ! M , we define a reparametrization of a
little more broadly, as a curve of the form z D ı ', where ' W Œc; d ! Œa; b is
a homeomorphism for which there is a partition .c0 ; : : : ; ck / of Œc; d such that the
restriction of ' to each subinterval Œci1 ; ci is a diffeomorphism onto its image.
If W Œa; b ! M is an admissible curve, we define the length of to be
Z b
Lg . / D j 0 .t /jg dt:
a
The integrand is bounded and continuous everywhere on Œa; b except possibly at

the finitely many points where is not smooth, so this integral is well defined.
Proposition 2.47 (Properties of Lengths). Suppose .M; g/ is a Riemannian mani-

fold with or without boundary, and W Œa; b ! M is an admissible curve.
(a) A DDITIVITY OF L ENGTH : If a < c < b, then
ˇ ˇ
Lg . / D Lg ˇŒa;c C Lg ˇŒc;b :
(b) PARAMETER
I NDEPENDENCE : If z is a reparametrization of , then Lg . / D
Lg z .
(c) I SOMETRY I NVARIANCE : If .M; g/ and M ; gz are Riemannian manifolds
(with or without boundary) and ' W M ! M is a local isometry, then Lg . / D
Lgz .' ı /.
I Exercise 2.48. Prove Proposition 2.47.
Suppose W Œa; b ! M is an admissible curve. The arc-length function of is

the function s W Œa; b ! R defined by
Z t

s.t / D Lg jŒa;t D j 0 .u/jg du:
a
It is continuous everywhere, and it follows from the fundamental theorem of calcu-

lus that it is smooth wherever is, with derivative s 0 .t / D j 0 .t /j.
For this reason, if W I ! M is any smooth curve (not necessarily a curve seg-
ment), we define the speed of at any time t 2 I to be the scalar j 0 .t /j. We say
that is a unit-speed curve if j 0 .t /j D 1 for all t , and a constant-speed curve if
j 0 .t /j is constant. If is a piecewise smooth curve, we say that has unit speed
if j 0 .t /j D 1 wherever is smooth.
If W Œa; b ! M is a unit-speed admissible curve, then its arc-length function
has the simple form s.t / D t a. If, in addition, its parameter interval is of the form
Œ0; b for some b > 0, then the arc-length function is s.t / D t . For this reason, a
unit-speed admissible curve whose parameter interval is of the form Œ0; b is said to
be parametrized by arc length.
Proposition 2.49. Suppose .M; g/ is a Riemannian manifold with or without bound-

ary.
(a) Every regular curve in M has a unit-speed forward reparametrization.
(b) Every admissible curve in M has a unique forward reparametrization by arc
length.
Proof. Suppose W I ! M is a regular curve. Choose an arbitrary t0 2 I , and define

s W I ! R by Z t
s.t / D j 0 .u/jg du:
t0
Since s 0 .t / D j 0 .t /jg > 0, it follows that s is a strictly increasing local diffeo-

morphism, and thus maps I diffeomorphically onto an interval I 0 R. If we let
' D s 1 W I 0 ! I , then z D ı ' is a forward reparametrization of , and the chain
rule gives
ˇ 0 ˇ 1
ˇz .t /ˇ D j' 0 .s/ 0 .'.s//jg D j 0 .'.s//jg D 1:
g s 0 .'.s//
Thus z is a unit-speed reparametrization of .

Now let W Œa; b ! M be an admissible curve. We prove the existence statement
in part (b) by induction on the number of smooth segments in an admissible parti-
tion. If has only one smooth segment, then it is a regular curve segment, and (b)
follows by applying (a) in the special case I D Œa; b and choosing t0 D a. Assuming
that the result is true for admissible curves with k smooth segments, suppose
has an admissible partition .a0 ; : : : ; akC1 /. The inductive hypothesis gives piece-
wise regular homeomorphisms ' W Œ0; c ! Œa; ak and W Œ0; d ! Œak ; b such
that ı ' is an arc-length reparametrization of jŒa;ak and ı is an arc-length
reparametrization of jŒak ;b . If we define 'z W Œ0; c C d ! Œa; b by
(
'.s/; s 2 Œ0; c;
z D
'.s/
.s c/; s 2 Œc; c C d ;
then ı 'z is a reparametrization of by arc length.

To prove uniqueness, suppose that z D ı ' and y D ı are both for-
ward reparametrizations of by arc length. Since both have the same length,
it follows that ' and both have the same parameter domain Œ0; c, and thus
both are piecewise regular homeomorphisms from Œ0; c to Œa; b. If we define
D ' 1 ı W Œ0; c ! Œ0; c, then is a piecewise regular increasing homeomor-
phism satisfying y D ı ' ı D z ı . The fact that both z and y are of unit speed
implies the following equality for all s 2 Œ0; c except the finitely many values at
which z or is not smooth:
ˇ ˇ ˇ ˇ ˇ ˇ
1 D ˇy0 .s/ˇg D ˇz0 ..s//0 .s/ˇg D ˇz0 ..s//ˇg 0 .s/ D 0 .s/:
Since is continuous and .0/ D 0, it follows that .s/ D s for all s 2 Œ0; c, and
thus z D y.
The Riemannian Distance Function
We are now in a position to introduce one of the most important concepts from
classical geometry into the Riemannian setting: distances between points.
Suppose .M; g/ is a connected Riemannian manifold with or without boundary.
For each pair of points p; q 2 M , we define the Riemannian distance from p to
q, denoted by dg .p; q/, to be the infimum of the lengths of all admissible curves
from p to q. The following proposition guarantees that dg .p; q/ is a well-defined
nonnegative real number for each p; q 2 M .
Proposition 2.50. If M is a connected smooth manifold (with or without boundary),
then any two points of M can be joined by an admissible curve.
Proof. Let p; q 2 M be arbitrary. Since a connected manifold is path-connected,
p and q can be connected by a continuous path c W Œa; b ! M . By compactness,
there is a partition of Œa; b such that c Œai1 ; ai is contained in a single smooth
coordinate ball (or half-ball in case @M ¤ ¿) for each i . Then we may replace each
such curve segment by a straight-line path in coordinates, yielding an admissible
curve between the same points (Fig. 2.2).
For convenience, if .M; g/ is a disconnected Riemannian manifold, we also let
dg .p; q/ denote the Riemannian distance from p to q, provided that p and q lie in
the same connected component of M . (See also Problem 2-30.)
Fig. 2.2: Any two points can be connected by an admissible curve
Proposition 2.51 (Isometry

Invariance of the Riemannian Distance Function).
Suppose .M; g/ and M ; gz are connected Riemannian manifolds with or without
boundary, and ' W M ! M is an isometry. Then dgz .'.x/; '.y// D dg .x; y/ for all
x; y 2 M .
I Exercise 2.52. Prove the preceding proposition.
(Note that unlike lengths of curves, Riemannian distances are not necessarily pre-
served by local isometries; see Problem 2-31.)
We wish to show that the Riemannian distance function turns M into a metric
space, whose metric topology is the same as its original manifold topology. To do
so, we need the following lemmas.
Lemma 2.53. Let g be a Riemannian metric on an open subset W Rn or RnC , and

let gx denote the Euclidean metric on W . For every compact subset K W , there
are positive constants c; C such that for all x 2 K and all v 2 Tx Rn ,
cjvjgx jvjg C jvjgx : (2.21)
Proof. Define a continuous function F W T W ! R by F .x; v/ D jvjg D gx .v; v/1=2

for x 2 W and v 2 Tx Rn . Let K W be any compact subset, and define L T Rn
by ˚
L D .x; v/ 2 T Rn W x 2 K; jvjgx D 1 :
Under the canonical identification of T Rn with Rn Rn , L is equal to the prod-
uct set K Sn1 , which is compact. Since F is positive on L, there are positive
constants c; C such that
c jvjg C for all .x; v/ 2 L:
For each x 2 K and v 2 Tx Rn with v ¤ 0, we can write v D y

v , where D jvjgx and
vy D v=. Then x; vy 2 L, and it follows from the homogeneity of j jg that
ˇ ˇ
jvjg D ˇvyˇg C D C jvjgx :
The same inequality holds trivially when v D 0. Arguing similarly in the other direc-
tion, we conclude that (2.21) holds for all x 2 K and v 2 Tx Rn .
Fig. 2.3: dg .p; q/ D Fig. 2.4: The triangle inequality
The next lemma shows how to transfer this result to Riemannian manifolds.
Lemma 2.54. Let .M; g/ be a Riemannian manifold with or without boundary and
let dg be its Riemannian distance function. Suppose U is an open subset of M and
p 2 U . Then p has a coordinate neighborhood V U with the property that there
are positive constants C; D satisfying the following inequalities:
(a) If q 2 V , then dg .p; q/ Cdgx .p; q/, where gx is the Euclidean metric in the
given coordinates on V .
(b) If q 2 M X V , then dg .p; q/ D.
Proof. Let W be any neighborhood of p contained in U on which there exist smooth
coordinates x i . Using these coordinates, we can identify W with an open subset
of Rn or RnC . Let K be a compact subset of W containing a neighborhood V of
p, chosen as follows: If p is an interior point of M , let K be the closed Euclidean
ball Bx" .0/, with " > 0 chosen small enough that K W , and let V D B" .0/. If
p 2 @M , let K be a closedP half-ball Bx" .0/\RnC with " chosen similarly, and let V D
B" .0/ \ RC . Let gx D i .dx / denote the Euclidean metric in these coordinates,
n i 2
and let c; C be constants satisfying (2.21). If is any admissible curve whose image
lies entirely in V , it follows that
cLgx . / Lg . / CLgx . /: (2.22)
To prove (a), suppose q 2 V . Letting W Œa; b ! V be a parametrization of the

line segment from p to q, we conclude from (2.22) that
dg .p; q/ Lg . / CLgx . / D Cdgx .p; q/:
To prove (b), suppose q 2 M X V . If W Œa; b ! M is any admissible curve from

let t0 denote the infimum of times t 2 Œa; b such that .t / … V . It follows
p to q,
that Œa; t0 K and dgx .p; .t0 // D " by continuity (Fig. 2.3), so (2.22) implies

Lg . / Lg jŒa;t0 cLgx jŒa;t0 c dgx .p; .t0 // D c":
Taking the infimum over all such curves , we obtain (b) with D D c".
Theorem 2.55 (Riemannian Manifolds as Metric Spaces). Let .M; g/ be a con-

nected Riemannian manifold with or without boundary. With the distance function
dg , M is a metric space whose metric topology is the same as the given manifold
topology.
Proof. It is immediate from the definition of dg that dg .p; q/ D dg .q; p/ 0 and

dg .p; p/ D 0. On the other hand, suppose p; q 2 M are distinct. Let U M be an
open set that contains p but not q, and choose a coordinate neighborhood V of p
contained in U and satisfying the conclusion of Lemma 2.54. Then Lemma 2.54(b)
shows that dg .p; q/ D > 0.
The triangle inequality follows from the fact that an admissible curve from p
to q can be combined with one from q to r (possibly changing the starting time
of the parametrization of the second) to yield one from p to r whose length is the
sum of the lengths of the two given curves (Fig. 2.4). (This is one reason for defining
distance using piecewise regular curves instead of just regular ones.) This completes
the proof that dg turns M into a metric space.
It remains to show that the metric topology is the same as the manifold topology.
Suppose first that U M is open in the manifold topology. For each p 2 U , we can
choose a coordinate neighborhood V of p contained in U with positive constants
C; D satisfying the conclusions of Lemma 2.54. The contrapositive of part (b) of
that lemma says dg .p; q/ < D ) q 2 V U , which means that the metric ball of
radius D is contained in U . Thus U is open in the metric topology induced by dg .
On the other hand, suppose U 0 is open in the metric topology. Given p 2 U 0 ,
choose ı > 0 such that the dg -metric ball of radius ı around p is contained in U 0 .
Let V be any neighborhood of p that is open in the manifold topology and satisfies
the conclusions of Lemma 2.54, with corresponding constants C; D. (We are not
claiming that V U 0 .) Choose " small enough that C " < ı. Lemma 2.54(a) shows
that if q is a point of V such that dgx .p; q/ < ", then dg .p; q/ C " < ı, and thus
q lies in the metric ball of radius ı about p, and hence in U 0 . Since the set fq 2
V W dgx .p; q/ < "g is open in the given manifold topology, this shows that U 0 is also
open in the manifold topology.
Thanks to the preceding theorem, it makes sense to apply all the concepts of the
theory of metric spaces to a connected Riemannian manifold .M; g/. For example,
we say that M is (metrically) complete if every Cauchy sequence in M converges. A
subset A M is bounded if there is a constant C such that dg .p; q/ C for all
p; q 2 A; if this is the case, the diameter of A is the smallest such constant:
diam.A/ D supfdg .p; q/ W p; q 2 Ag:
Since every compact metric space is bounded, every compact connected Riemannian
manifold with or without boundary has finite diameter. (Note that the unit sphere
with the Riemannian distance determined by the round metric has diameter , not
2, since the Riemannian distance between antipodal points is . See Problem 6-2.)
Pseudo-Riemannian Metrics
From the point of view of geometry, Riemannian metrics are by far the most impor-
tant structures that manifolds carry. However, there is a generalization of Rieman-
nian metrics that has become especially important because of its application to
physics.
Before defining this generalization, we begin with some linear-algebraic pre-
liminaries. Suppose V is a finite-dimensional vector space, and q is a symmetric
covariant 2-tensor on V (also called a symmetric bilinear form). Just as for an inner
product, there is a linear map qy W V ! V defined by
qy.v/.w/ D q.v; w/ for all v; w 2 V:
We say that q is nondegenerate if qy is an isomorphism.

Lemma 2.56. Suppose q is a symmetric covariant 2-tensor on a finite-dimensional
vector space V . The following are equivalent:
(a) q is nondegenerate.
(b) For every nonzero v 2 V , there is some
w 2 V such that q.v; w/ ¤ 0.
(c) Ifq D qij "i "j intermsofsomebasis "j forV ,thenthematrix.qij /isinvertible.
Every inner product is a nondegenerate symmetric bilinear form, as is every sym-

metric bilinear form that is negative definite (which is defined by obvious analogy
with positive definite). But there are others that are neither positive definite nor negative
definite, as we will see below.
We use the term scalar product to denote any nondegenerate symmetric bilinear
form on a finite-dimensional vector space V , and reserve the term inner product for
the special case of a positive definite scalar product. A scalar product space is a
finite-dimensional vector space endowed with a scalar product. When convenient,
we will often use a notation like h; i to denote a scalar product. We say that vectors
v; w 2 V are orthogonal if hv; wi D 0, just as in the case of an inner product. Given
a vector v 2 V , we define the norm of v to be jvj D jhv; vij1=2 . Note that in the
indefinite case, it is possible for a nonzero vector to be orthogonal to itself, and thus
to have norm zero. Thus jvj is not technically a norm in the sense defined on page 47
below, but it is customary to call it “the norm of v” anyway.
I Exercise 2.58. Prove that the polarization identity (2.2) holds for every scalar product.
If S V is any linear subspace, the set of vectors in V that are orthogonal to

every vector in S is a linear subspace denoted by S ? .
Lemma 2.59. Suppose .V; q/ is a finite-dimensional scalar product space, and S
V is a linear subspace.
(a) dim S C dim S ? D dim V .
(b) .S ? /? D S .
Pseudo-Riemannian Metrics 41
Proof. Define a linear map ˚ W V ! S by ˚.v/ D qy.v/jS . Note that v 2 Ker ˚ if

and only if q.v; x/ D qy.v/.x/ D 0 for all x 2 S , so Ker ˚ D S ? . If ' 2 S is arbi-
trary, there is a covector 'z 2 V whose restriction to S is equal to '. (For example,
such a covector is easily constructed after choosing a basis for S and extending it to
a basis for V .) Since qy is an isomorphism, there exists v 2 V such that qy.v/ D '.
z It
follows that ˚.v/ D ', and therefore ˚ is surjective. By the rank–nullity theorem,
the dimension of S ? D Ker ˚ is equal to dim V dim S D dim V dim S . This
proves (a).
To prove (b), note that every v 2 S is orthogonal to every element of S ? by
definition, so S .S ? /? . Because these finite-dimensional vector spaces have the
same dimension by part (a), they are equal.
An ordered k-tuple .v1 ; : : : ; vk / of elements of V is said to be orthonormal if
jvi j D 1 for each i and hvi ; vj i D 0 for i ¤ j , or equivalently, if hvi ; vj i D ˙ıij for
each i and j . We wish to prove that every scalar product space has an orthonormal
basis. Note that the usual Gram–Schmidt algorithm does not always work in this
situation, because the vectors that appear in the denominators in (2.5)–(2.6) might
have vanishing norms. In order to get around this problem, we introduce the fol-
lowing definitions. If .V; q/ is a finite-dimensional scalar product space, a subspace
S V is said to be nondegenerate if the restriction of q to S S is nondegener-
ate. An ordered k-tuple of vectors .v1 ; : : : ; vk / in V is said to be nondegenerate if
for each j D 1; : : : ; k, the vectors .v1 ; : : : ; vj / span a nondegenerate j -dimensional
subspace of V . For example, every orthonormal basis is nondegenerate.
Lemma 2.60. Suppose .V; q/ is a finite-dimensional scalar product space, and S

V is a linear subspace. The following are equivalent:
(a) S is nondegenerate.
(b) S ? is nondegenerate.
(c) S \ S ? D f0g.
(d) V D S ˚ S ? .
Lemma 2.62 (Completion of Nondegenerate Bases). Suppose .V; q/ is an n-

dimensional scalar product space, and .v1 ; : : : ; vk / is a nondegenerate k-tuple in
V with 0 k < n. Then there exist vectors vkC1 ; : : : ; vn 2 V such that .v1 ; : : : ; vn /
is a nondegenerate basis for V .
Proof. Let S D span.v1 ; : : : ; vk / V . Because k < n, S ? is a nontrivial subspace

of V , and Lemma 2.60 shows that S ? is nondegenerate and V D S ˚ S ? . By the
nondegeneracy of S ? , there must be a vector in S ? with nonzero length, because
otherwise the polarization identity would imply that all inner products of pairs of
elements of S would be zero. If vkC1 2 S ? is any vector with nonzero length, then
.v1 ; : : : ; vkC1 / is easily seen to be a nondegenerate .k C 1/-tuple. Repeating this
argument for vkC2 ; : : : ; vn completes the proof.
Proposition 2.63 (Gram–Schmidt Algorithm for Scalar Products). Suppose

.V; q/ is an n-dimensional scalar product space. If .vi / is a nondegenerate basis for
V , then there is an orthonormal basis .bi / with the property that span.b1 ; : : : ; bk / D
span.v1 ; : : : ; vk / for each k D 1; : : : ; n.
Proof. As in the positive definite case, the basis .bi / is constructed recursively,
starting with b1 D v1 =jv1 j and noting that the assumption that v1 spans a nonde-
generate subspace ensures that jv1 j ¤ 0. At the inductive step, assuming we have
constructed .b1 ; : : : ; bk /, we first set
X
k
hvkC1 ; bi i
z D vkC1 bi :
hbi ; bi i
iD1
Each denominator hbi ; bi i is equal to ˙1, so this defines z as a nonzero element

of V orthogonal to b1 ; : : : ; bk , and with the property that span.b1 ; : : : ; bk ; z/ D
span.v1 ; : : : ; vkC1 /. If hz; zi D 0, then z is orthogonal to span.v1 ; : : : ; vkC1 /, con-
tradicting the nondegeneracy assumption. Thus we can complete the inductive step
by putting bkC1 D z=jzj.
Corollary
2.64.
Suppose .V; q/ is an n-dimensional scalar product space. There is
a basis ˇ i for V with respect to which q has the expression
2 2 2 2
q D ˇ 1 C C ˇ r ˇ rC1 ˇ rCs ; (2.23)
for some nonnegative integers r; s with r C s D n.

Proof. Let .bi / be an orthonormal basis for V , and let ˇ i be the dual basis for V .
A computation shows that q has a basis expression of the form (2.23), but perhaps
with the positive and negative terms in a different order. Reordering the basis so that
the positive terms come first, we obtain (2.23).
It turns out that the numbers r and s in (2.23) are independent of the choice of
basis. The key to proving this is the following classical result from linear algebra.
Proposition 2.65 (Sylvester’s Law of Inertia). Suppose .V; q/ is a finite-dimen-

sional scalar product space. If q has the representation (2.23) in some basis, then the
number r is the maximum dimension among all subspaces on which the restriction of
q is positive definite, and thus r and s are independent of the choice of basis.

The integer s in the expression (2.23) (the number of negative terms) is called
the index of q, and the ordered pair .r; s/ is called the signature of q.
Now suppose M is a smooth manifold. A pseudo-Riemannian metric on M
(called by some authors a semi-Riemannian metric) is a smooth symmetric 2-tensor
field g that is nondegenerate at each point of M , and with the same signature every-
where. Every Riemannian metric is also a pseudo-Riemannian metric.
Proposition 2.66 (Orthonormal Frames for Pseudo-Riemannian Manifolds). Let

.M; g/ be a pseudo-Riemannian manifold. For each p 2 M , there exists a smooth
orthonormal frame on a neighborhood of p in M .
I Exercise 2.68. Suppose .M1 ; g1 / and .M2 ; g2 / are pseudo-Riemannian manifolds of

signatures .r1 ; s1 / and .r2 ; s2 /, respectively. Show that .M1 M2 ; g1 ˚ g2 / is a pseudo-
Riemannian manifold of signature .r1 C r2 ; s1 C s2 /.
For nonnegative integers r and s, we define the pseudo-Euclidean space of sig-

nature .r;s/, denoted by Rr;s , to be the manifold RrCs , with standard coordinates
denoted by 1 ; : : : ; r ; 1 ; : : : ; s , and with the pseudo-Riemannian metric qx.r;s/
defined by
2 2 2 2
qx.r;s/ D d 1 C C d r d 1 d s : (2.24)
By far the most important pseudo-Riemannian metrics (other than the Riemann-
ian ones) are the Lorentz metrics, which are pseudo-Riemannian metrics of index 1,
and thus signature .r; 1/. (Some authors, especially in the physics literature, prefer
to use signature .1; r/; either one can be converted to the other by multiplying the
metric by 1, so there is no significant difference.)
The pseudo-Euclidean metric qx.r;1/ is a Lorentz metric called the Minkowski
metric, and the Lorentz manifold Rr;1 is called .r C 1/-dimensional Minkowski
space. If we denote standard coordinates on Rr;1 by 1 ; : : : ; r ; , then the Min-
kowski metric is given by
2 2
qx.r;1/ D d 1 C C d r .d /2 : (2.25)
In the special case of R3;1 , the Minkowski metric is the fundamental invariant of
Albert Einstein’s special theory of relativity, which can be expressed succinctly by
saying that if gravity is ignored, then spacetime is accurately modeled by .3 C 1/-
dimensional Minkowski space, and the laws of physics have the same form in every
coordinate system in which the Minkowski metric has the expression (2.25). The
differing physical characteristics of “space” (the i directions) and “time” (the
direction) arise from the fact that they are subspaces on which the metric is positive
definite and negative definite, respectively. The general theory of relativity includes
gravitational effects by allowing the Lorentz metric to vary from point to point.
Many, but not all, results from the theory of Riemannian metrics apply equally
well to pseudo-Riemannian metrics. Throughout this book, we will attempt to point
out which results carry over directly to pseudo-Riemannian metrics, which ones
can be adapted with minor modifications, and which ones do not carry over at all.
As a rule of thumb, proofs that depend only on the nondegeneracy of the metric
tensor, such as properties of the musical isomorphisms and existence and uniqueness
of geodesics, work fine in the pseudo-Riemannian setting, while proofs that use
positivity in an essential way, such as those involving lengths of curves, do not.
One notable result that does not carry over to the pseudo-Riemannian case is
Proposition 2.4, about the existence of metrics. For example, the following result
characterizes those manifolds that admit Lorentz metrics.
Theorem 2.69. A smooth manifold M admits a Lorentz metric if and only if it

admits a rank-1 tangent distribution (i.e., a rank-1 subbundle of TM ).

With some more sophisticated tools from algebraic topology, it can be shown
that every noncompact connected smooth manifold admits a Lorentz metric, and a
compact connected smooth manifold admits a Lorentz metric if and only if its Euler
characteristic is zero (see [O’N83, p. 149]). It follows that no even-dimensional
sphere admits a Lorentz metric, because S2n has Euler characteristic equal to 2.
For a thorough treatment of pseudo-Riemannian metrics from a mathematical
point of view, see the excellent book [O’N83]; a more physical treatment can be
found in [HE73].
Pseudo-Riemannian Submanifolds
The theory of submanifolds
is only slightly more complicated in the pseudo-
Riemannian case. If M ; gz is a pseudo-Riemannian manifold, a smooth submani-
fold W M ,! M is called a pseudo-Riemannian submanifold of M if gz is non-
degenerate with constant signature. If this is the case, we always consider M to be
endowed with the induced pseudo-Riemannian metric gz. In the special case in
which gz is positive definite, M is called a Riemannian submanifold.
The nondegeneracy hypothesis is not automatically satisfied: for example, if
M R1;1 is the submanifold f.; / W D g and W M ! R1;1 is inclusion, then
the pullback tensor qx.1;1/ is identically zero on M .
For hypersurfaces (submanifolds of codimension 1), the nondegeneracy condi-
tion is easy to check. If M M is a smooth submanifold and p 2 M , then a vector

2 Tp M is said to be normal to M if h; xi D 0 for all x 2 Tp M , just as in the
Riemannian case. The space of all normal vectors at p is a subspace of Tp M de-
noted by Np M .

Proposition 2.70. Suppose M ; gz is a pseudo-Riemannian manifold of signature
.r; s/. Let M be a smooth hypersurface in M , and let W M ,! M
be the inclusion

map. Then the pullback tensor field gz is nondegenerate if and only if gz.; / ¤ 0
for every p 2 M and every nonzero normal vector 2 Np M . If gz.; / > 0 for
every nonzero normal vector to M , then M is a pseudo-Riemannian submanifold of
signature .r 1; s/; and if gz.; / < 0 for every such vector, then M is a pseudo-
Riemannian submanifold of signature .r; s 1/.
Proof. Given p 2 M , Lemma 2.60 shows that Tp M is a nondegenerate subspace of

Tp M if and only if the one-dimensional subspace .Tp M /? D Np M is nondegen-
erate, which is the case if and only if every nonzero 2 Np M satisfies gz.; / ¤ 0.
Now suppose gz.; / > 0 for every nonzero normal vector . Let p 2 M be
arbitrary, and let be a nonzero element of Np M . Writing n D dim M , we can

complete to a nondegenerate basis .; w2 ; : : : ; wn / for Tp M by Lemma 2.62, and
then use the Gram–Schmidt algorithm to find an orthonormal basis .b1 ; : : : ; bn/ for
Tp M such that span.b1 / D Np M . It follows that span.b2 ; : : : ; bn / D Tp M . If ˇ j
is the dual basis to .bi /, then gzp has a basis representation of the form .ˇ 1 /2 ˙
.ˇ 2 /2 ˙ ˙ .ˇ n /2 , with a total of r positive terms and s negative ones, and with
a positive sign on the first term .ˇ 1 /2 . Therefore, gzp D ˙.ˇ 2 /2 ˙ ˙ .ˇ n /2 has
signature .r 1; s/. The argument for the case gz.; / < 0 is similar.

If M ; gz is a pseudo-Riemannian manifold and f 2 C 1 .M /, then the gradient of
f is defined as the smooth vector field grad f D .df /] just as in the Riemannian
case.

Corollary 2.71. Suppose M ; gz is a pseudo-Riemannian manifold of signature

.r; s/, f 2 C 1 M , and M D f 1 .c/ for some c 2 R. If gz.grad f; grad f / > 0
everywhere on M , then M is an embedded pseudo-Riemannian submanifold of M
of signature .r 1; s/; and if gz.grad f; grad f / < 0 everywhere on M , then M is an
embedded pseudo-Riemannian submanifold of M of signature .r; s 1/. In either
case, grad f is everywhere normal to M .
Proposition
2.72 (Pseudo-Riemannian Adapted Orthonormal Frames). Suppose
M ; gz is a pseudo-Riemannian manifold, and M M is an embedded pseudo-
Riemannian or Riemannian submanifold. For each p 2 M , there exists a smooth
orthonormal frame on a neighborhood of p in M that is adapted to M .
Proof. Write m D dim M and n D dim M , and let p 2 M be arbitrary. Propo-

sition 2.66 shows that there is a smooth orthonormal frame .E1 ; : : : ; En / for M
on some neighborhood of p in M . Then by Lemma 2.62, we can find vectors
vnC1 ; : : : ; vm 2 Tp M such that E1 jp ; : : : ; En jp ; vnC1 ; : : : ; vm is a nondegener-
ate basis for Tp M . Now extend vnC1 ; : : : ; vm arbitrarily to smooth vector fields
VnC1 ; : : : ; Vm on a neighborhood of p in M . By continuity, the ordered m-tuple
.E1 ; : : : ; En ; VnC1 ; : : : ; Vm / will be a nondegenerate frame for M in some (possibly
smaller) neighborhood of p. Applying the Gram–Schmidt algorithm (Prop. 2.63) to
this local frame yields a smooth local orthonormal frame .E1 ; : : : ; Em / for M with
the property that .E1 ; : : : ; En / restricts to a local orthonormal frame for M .
The next corollary is proved in the same way as Proposition 2.16.
Corollary
2.73
(Normal Bundle to a Pseudo-Riemannian Submanifold). Sup-
pose M ; gz is a pseudo-Riemannian manifold, and M M is an embedded
pseudo-Riemannian or Riemannian submanifold. The set of vectors normal to M
jM , called the normal bundle to M .
is a smooth vector subbundle of T M
Other Generalizations of Riemannian Metrics
Pseudo-Riemannian metrics are obtained by relaxing the positivity requirement for

Riemannian metrics. In addition, there are other useful generalizations that result
when we relax other requirements. In this section we touch briefly on two of those
generalizations. We will not treat these anywhere else in the book, but it is useful to
know the definitions.
Sub-Riemannian Metrics
The first generalization arises when we relax the requirement that the metric be
defined on the whole tangent space.
A sub-Riemannian metric (also sometimes known as a singular Riemannian
metric or Carnot–Carathéodory metric) on a smooth manifold M is a smooth fiber
metric on a smooth tangent distribution S TM (i.e., a vector subbundle of TM ).
Since lengths make sense only for vectors in S , the only curves whose lengths can
be measured are those whose velocity vectors lie everywhere in S . Therefore, one
usually imposes some condition on S that guarantees that any two nearby points
can be connected by such a curve. This is, in a sense, the opposite of the Frobenius
integrability condition, which would restrict every such curve to lie in a single leaf
of a foliation.
Sub-Riemannian metrics arise naturally in the study of the abstract models of
real submanifolds of complex space C n , called CR manifolds (short for Cauchy–
Riemann manifolds). CR manifolds are real manifolds endowed with a tangent dis-
tribution S TM whose fibers carry the structure of complex vector spaces (with
an additional integrability condition that need not concern us here). In the model
case of a submanifold M C n , Spis the set of vectors tangent to M that remain
tangent after multiplication by i D 1 in the ambient complex coordinates. If S is
sufficiently far from being integrable, choosing a fiber metric on S results in a sub-
Riemannian metric whose geometric properties closely reflect the complex-analytic
properties of M as a subset of C n .
Another motivation for studying sub-Riemannian metrics arises from control the-
ory. In this subject, one is given a manifold with a vector field depending on parame-
ters called controls, with the goal being to vary the controls so as to obtain a solution
curve with desired properties, often one that minimizes some function such as arc
length. If the vector field is constrained to be everywhere tangent to a distribution S
on the manifold (for example, in the case of a robot arm whose motion is restricted
by the orientations of its hinges), then the function can often be modeled as a sub-
Riemannian metric and optimal solutions modeled as sub-Riemannian geodesics.
A useful introduction to the geometry of sub-Riemannian metrics is provided in
the article [Str86].
Problems 47
Finsler Metrics
Another important generalization arises from relaxing the requirement that norms
of vectors be defined in terms of an inner product on each tangent space.
In general, a norm on a vector space V is a real-valued function on V , usually
written v 7! jvj, that satisfies
(i) H OMOGENEITY: jcvj D jcj jvj for v 2 V and c 2 R;
(ii) P OSITIVITY: jvj 0 for v 2 V , with equality if and only if v D 0;
(iii) T RIANGLE I NEQUALITY: jv C wj jvj C jwj for v; w 2 V .
For example, the length function associated with an inner product is a norm.
Now suppose M is a smooth manifold. A Finsler metric on M is a continuous
function F W TM ! R, smooth on the set of nonzero vectors, whose restriction to
each tangent space Tp M is a norm. Again, the norm function associated with a
Riemannian metric is a special case.
The inventor of Riemannian geometry himself, Bernhard Riemann, clearly envis-
aged an important role in n-dimensional geometry for what we now call Finsler met-
rics; he restricted his investigations to the “Riemannian” case purely for simplicity
(see [Spi79, Vol. 2]). However, it was not until the late twentieth century that Finsler
metrics began to be studied seriously from a geometric point of view.
The recent upsurge of interest in Finsler metrics has been motivated in part by
the fact that two different Finsler metrics appear very naturally in the theory of
several complex variables. For certain bounded open sets in C n (the ones with
smooth, strictly convex boundaries, for example), the Kobayashi metric and the
Carathéodory metric are intrinsically defined, biholomorphically invariant Finsler
metrics. Combining differential-geometric and complex-analytic methods has led
to striking new insights into both the function theory and the geometry of such
domains. We do not treat Finsler metrics further in this book, but you can consult
one of the recent books on the subject [AP94, BCS00, JP13].
Problems
2-1. Show that every Riemannian 1-manifold is flat. (Used on pp. 13, 193.)
2-2. Suppose V and W are finite-dimensional real inner product spaces of the
same dimension, and F W V ! W is any map (not assumed to be linear or
even continuous) that preserves the origin and all distances: F .0/ D 0 and
jF .x/ F .y/j D jx yj for all x; y 2 V . Prove that F is a linear isometry.
[Hint: First show that F preserves inner products, and then show that it is linear.]
(Used on p. 187.)
2-3. Given a smooth embedded 1-dimensional submanifold C H as in Exam-
ple 2.24(b), show that the surface of revolution SC R3 with its induced
metric is isometric to the warped product C a S1 , where a W C ! R is the
distance to the z-axis.
2-4. Let W RC ! R be the restriction of the standard coordinate function, and

let RC Sn1 denote the resulting warped product (see Example 2.24(c)).
Define ˚ W RC Sn1 ! Rn X f0g by ˚. ; !/ D !. Show that ˚ is an
isometry between the warped product metric and the Euclidean metric on
Rn X f0g. (Used on p. 293.)
2-5. Prove parts (b) and (c) of Proposition 2.25 (properties of horizontal vector
fields). (Used on p. 146.)
2-6. Prove Theorem 2.28 (if W M ! M is a surjective smooth submersion,
and a group acts on M isometrically, vertically, and transitively on fibers,
then M inherits a unique Riemannian metric such that is a Riemannian
submersion).
2-7. For 0 < k < n, the set Gk .Rn / of k-dimensional linear subspaces of Rn
is called a Grassmann manifold or Grassmannian. The group GL.n; R/
acts transitively on Gk .Rn / in an obvious way, and Gk .Rn / has a unique
smooth manifold structure making this action smooth (see [LeeSM, Exam-
ple 21.21]).
(a) Let Vk .Rn / denote the set of orthonormal ordered k-tuples of vectors
in Rn . By arranging the vectors in k columns, we can view Vk .Rn /
as a subset of the vector space M.n k; R/ of all n k real matrices.
Prove that Vk .Rn / is a smooth submanifold of M.n k; R/ of dimen-
sion k.2n k 1/=2, called a Stiefel manifold. [Hint: Consider the
map ˚ W M.n k; R/ ! M.k k; R/ given by ˚.A/ D AT A.]
(b) Show that the map W Vk .Rn / ! Gk .Rn / that sends a k-tuple to its
span is a surjective smooth submersion.
(c) Give Vk .Rn / the Riemannian metric induced from the Euclidean met-
ric on M.n k; R/. Show that the right action of O.k/ on Vk .Rn / by
matrix multiplication on the right is isometric, vertical, and transitive
on fibers of , and thus there is a unique metric on Gk .Rn / such that
is a Riemannian submersion. [Hint: It might help to note that the
Euclidean inner
product
on M.n k; R/ can be written in the form
hA; Bi D tr AT B .]
(Used on p. 82.)
2-8. Prove that the action of Z on R2 defined in Example 2.35 is smooth, free,
proper, and isometric, and therefore the open Möbius band inherits a flat
Riemannian metric such that the quotient map is a Riemannian covering.
2-9. Prove Proposition 2.37 (the gradient is orthogonal to regular level sets).
2-10. Suppose .M; g/ is a Riemannian manifold, f 2 C 1 .M /, and X 2 X.M /
is a nowhere-vanishing vector field. Prove that X D grad f if and only if
Xf jX j2g and X is orthogonal to the level sets of f at all regular points
of f . (Used on pp. 161, 180.)
2-11. Prove Proposition 2.40 (inner products on tensor bundles).
Problems 49
2-12. Prove Proposition 2.41 (existence and uniqueness of the Riemannian volume
form).
2-13. Prove Proposition 2.43 (characterizing the volume form of a Riemannian
hypersurface). [Hint: To prove (2.17), use an adapted orthonormal frame.]

2-14. Suppose M ; gz and .M; g/ are compact connected Riemannian manifolds,

and W M ! M is a k-sheeted Riemannian covering. Prove that Vol M D
k Vol.M /. (Used on p. 363.)
2-15. Suppose .M1 ; g1 / and .M2 ; g2 / are oriented Riemannian manifolds of di-
mensions k1 and k2 , respectively. Let f W M1 ! RC be a smooth func-
tion, and let g D g1 ˚ f 2 g2 be the corresponding warped product metric
on M1 f M2 . Prove that the Riemannian volume form of g is given by
d Vg D f k2 d Vg1 ^ d Vg2 ;
where f , d Vg1 , and d Vg2 are understood to be pulled back to M1 M2 by

the projection maps. (Used on p. 295.)
2-16. Let .M; g/ be a Riemannian n-manifold. Show that for each k D 1; : : : ; n,
there is a unique fiber metric h; ig on the bundle ƒk T M that satisfies
˝ 1 ˛ ˝ ˛
! ^ ^ ! k ; 1 ^ ^ k g D det ! i ; j g (2.26)
whenever ! 1 ; : : : ; ! k ; 1 ; : : : ; k are covectors at a point p 2 M . [Hint: De-

fine the inner product locally by declaring the set of k-covectors
n ˇ o
"i1 ^ ^ "ik ˇp W i1 < < ik

to be an orthonormal basis for ƒk Tp M whenever "i is a local orthonor-
mal coframe for T M , and then prove that the resulting inner product satis-
fies (2.26) and is independent of the choice of frame.]
2-17. Because we regard the bundle ƒk T M of k-forms as a subbundle of
the bundle T k T M of covariant k-tensors, we have two inner products to
choose from on k-forms: the one defined in Problem 2-16, and the restriction
of the tensor inner product defined in Proposition 2.40. For this problem, we
use the notation h; i to denote the inner product of Problem 2-16, and hh; ii
to denote the restriction of the tensor inner product.
(a) Using the convention for the wedge product that we use in this book (see
p. 400), prove that
1
h; ig D hh; iig :
kŠ
(b) Show that if the Alt convention is used for the wedge product (p. 401),
the formula becomes
h; ig D kŠhh; iig :

2-18. Let .M; g/ be an oriented Riemannian n-manifold.
(a) For each k D 0; : : : ; n; show that there is a unique smooth bundle
homomorphism
W ƒk T M ! ƒnk T M , called the Hodge star op-
erator, satisfying
! ^
D h!; ig d Vg
for all smooth k-forms !, , where h; ig is the inner product on k-
forms defined in Problem 2-16. (For k D 0; interpret the inner product
as ordinary multiplication.) [Hint: First prove uniqueness, and then de-
fine
locally by setting

"i1 ^ ^ "ik D ˙"j1 ^ ^ "jnk

in terms of an orthonormal coframe "i , where the indices j1 ; : : : ; jnk
are chosen such that .i1 ; : : : ; ik ; j1 ; : : : ; jnk / is some permutation of
.1; : : : ; n/.]
(b) Show that
W ƒ0 T M ! ƒn T M is given by
f D f d Vg .
(c) Show that

! D .1/k.nk/ ! if ! is a k-form.
2-19. Regard Rn as a Riemannian manifold with the Euclidean metric and the
standard orientation, and let
denote the Hodge star operator defined in
Problem 2-18.
(a) Calculate
dx i for i D1; : : : ; n.
(b) Calculate
dx i ^ dx j in the case n D 4.
2-20. Let M be an oriented Riemannian 4-manifold. A 2-form ! on M is said to
be self-dual if
! D !; and anti-self-dual if
! D !.
(a) Show that every 2-form ! on M can be written uniquely as a sum of a
self-dual form and an anti-self-dual form.
(b) On M D R4 with the Euclidean metric, determine the self-dual and anti-
self-dual forms in standard coordinates.
2-21. Prove Proposition 2.46 (the coordinate formulas for the divergence and the
Laplacian).
2-22. Suppose .M; g/ is a compact Riemannian manifold with boundary.
(a) Prove the following divergence theorem for X 2 X.M /:
Z Z
.div X /d Vg D hX; N ig d Vgy ;
M @M
where N is the outward unit normal to @M and gy is the induced metric

on @M . [Hint: Prove it first in the case that M is orientable, and then
apply that case to the orientation covering of M (Prop. B.18).]
Problems 51
(b) Show that the divergence operator satisfies the following product rule for
u 2 C 1 .M / and X 2 X.M /:
div.uX / D u div X C hgrad u; X ig ;
and deduce the following “integration by parts” formula:

Z Z Z
hgrad u; X ig d Vg D uhX; N ig d Vgy u div X d Vg :
M @M M
What does this say when M is a compact interval in R?

(Used on p. 149.)
2-23. Let .M; g/ be a compact Riemannian manifold with or without boundary.
A function u 2 C 1 .M / is said to be harmonic if u D 0, where is the
Laplacian defined on page 32.
(a) Prove Green’s identities:
Z Z Z
uv d Vg D u N v d Vgy hgrad u; grad vig d Vg ;
Z M
Z@M M
.uv vu/ d Vg D .u N v v N u/d Vgy ;

M @M
where N is the outward unit normal vector field on @M and gy is the

induced metric on @M .
(b) Show that if M is connected, @M ¤ ¿, and u, v are harmonic functions
on M whose restrictions to @M agree, then u v.
(c) Show that if M is connected and @M D ¿, then the only harmonic
functions
R on M are the constants, and every smooth function u satisfies
M u d V g D 0.
(Used on pp. 149, 223.)

2-24. Let .M; g/ be a compact Riemannian manifold (without boundary). A real
number is called an eigenvalue of M if there exists a smooth function u
on M , not identically zero, such that u D u. In this case, u is called an
eigenfunction corresponding to .
(a) Prove that 0 is an eigenvalue of M , and that all other eigenvalues are
strictly positive.
(b) Show that if Ru and v are eigenfunctions corresponding to distinct eigen-
values, then M uv d Vg D 0.
(Used on p. 149.)
2-25. Let .M; g/ be a compact connected Riemannian n-manifold with nonempty
boundary. A number 2 R is called a Dirichlet eigenvalue for M if there
exists a smooth real-valued function u on M , not identically zero, such that
u D u and uj@M D 0. Similarly, is called a Neumann eigenvalue if
there exists such a u satisfying u D u and N uj@M D 0, where N is the

outward unit normal.
(a) Show that every Dirichlet eigenvalue is strictly positive.
(b) Show that 0 is a Neumann eigenvalue, and all other Neumann eigenval-
ues are strictly positive.
2-26. D IRICHLET ’ S P RINCIPLE : Suppose .M; g/ is a compact connected Rie-
mannian n-manifold with nonempty boundary. ProveR that a function u 2
C 1 .M / is harmonic if and only if it minimizes M j grad uj2 d Vg among
all smooth functions with the same boundary values.
R [Hint: For any func-
tion f 2 C 1 .M / that vanishes on @M; expand M j grad.u C "f /j2 d Vg and
use Problem 2-22.]
2-27. Suppose .M; g/ is an oriented Riemannian 3-manifold.
(a) Define ˇ W TM ! ƒ2 T M by ˇ.X / D X ³ d Vg . Show that ˇ is a
smooth bundle isomorphism, and thus we can define the curl of a vector
field X 2 X.M / by

curl X D ˇ 1 d X [ :
(b) Show that the following diagram commutes:
grad
- X.M / curl
- X.M / div
- C 1 .M /
C 1 .M /
Id [ ˇ
(2.27)
? ? ? ?
0 .M / - 1 .M / - 2 .M / - 3 .M /;
d d d
where
.f / D f d Vg , and use this to prove that curl.grad f / D 0 for
every f 2 C 1 .M /, and div.curl X / D 0 for every X 2 X.M /.
(c) Compute the formula for the curl in standard coordinates on R3 with
the Euclidean metric.
2-28. Let .M; g/ be an oriented Riemannian manifold and let
denote its Hodge
star operator (Problem 2-18). Show that for every X 2 X.M /,

X ³ d Vg D
X [ ;

div X D
d
X [ ;
and, when dim M D 3,

]
curl X D
d X [ ;
where the curl of a vector field is defined as in Problem 2-27.

2-29. Let .M; g/ be a compact oriented Riemannian n-manifold. For 1 k n,
define a map d W k .M / ! k1 .M / by d ! D .1/n.kC1/C1
d
!,
Problems 53
where
is the Hodge star operator defined in Problem 2-18. Extend this
definition to 0-forms by defining d ! D 0 for ! 2 0 .M /.
(a) Show that d ı d D 0.
(b) Show that the formula
Z
.!; / D h!; ig d Vg
M
defines an inner product on k .M / for each k, where h; ig is the inner
product on forms defined in Problem 2-16.
(c) Show that .d !; / D .!; d/ for all ! 2 k .M / and 2 k1 .M /.
2-30. Suppose .M; g/ is a (not necessarily connected) Riemannian manifold.
Show that there is a distance function d on M that induces the given topol-
ogy and restricts to the Riemannian distance on each component of M .
(Used on p. 187.)

2-31. Suppose .M; g/ and M ; gz are connected Riemannian manifolds, and
'W M ! M is a local isometry. Show that dgz .'.x/; '.y// dg .x; y/ for
all x; y 2 M . Give an example to show that equality need not hold. (Used
on p. 37.)
2-32. Let .M; g/ be a Riemannian manifold and W Œa; b ! M a smooth curve
segment. For each continuous function f W Œa;
R b ! R, we define the integral
of f with respect to arc length, denoted by f ds, by
Z Z b
f ds D f .t / j 0 .t /jg dt:
a
R
(a) Show that f ds is independent of parametrization in the following
sense: if ' W Œc; d ! Œa; b is a diffeomorphism, then
Z Z
b d ˇ ˇ
0
f .t / j .t /jg dt D fz.u/ ˇz0 .u/ˇg du;
a c
where fz D f ı ' and z D ı '.

(b) Suppose now that is a smooth embedding, so that C D .Œa; b/ is an
embedded submanifold with boundary in M . Show that
Z Z
f ds D .f ı 1 / d Vz ;
C
where d Vz is the Riemannian volume element on C associated with the

induced metric and the orientation determined by .
(Used on p. 273.)
2-33. Prove Proposition 2.65 (Sylvester’s law of inertia).
2-34. Prove Theorem 2.69 (existence of Lorentz metrics), as follows.

(a) For sufficiency, assume that D TM is a 1-dimensional distribution, and
choose any Riemannian metric g on M . Show that locally it is possible
to choose a g-orthonormal frame .Ei / and dual coframe "i such that
E1 spans D; and then show that the Lorentz metric ."1 /2 C ."2 /2 C
C ."n /2 is independent of the choice of frame.
(b) To prove necessity, suppose that g is a Lorentz metric on M , and let
g0 be any Riemannian metric. Show that for each p 2 M , there are
exactly two g0 -unit vectors v0 , v0 on which the function v 7! g.v; v/
takes its minimum among all unit vectors in Tp M , and use Lagrange
multipliers to conclude that there exists a number .p/ < 0 such that
g.v0 ; w/ D .p/g0 .v0 ; w/ for all w 2 Tp M . You may use the following
standard result from perturbation theory: if U is an open subset of Rn
and A W U ! GL.n; R/ is a smooth matrix-valued function such that
A.x/ is symmetric and has exactly one negative eigenvalue for each
x 2 U , then there exist smooth functions W U ! .1; 0/ and X W U !
Rn X f0g such that A.x/X.x/ D .x/X.x/ for all x 2 U .
2-35. Prove Corollary 2.71 (about level sets in pseudo-Riemannian manifolds).
(Used on p. 63.)
Chapter 3
Model Riemannian Manifolds
Before we delve into the general theory of Riemannian manifolds, we pause to give
it some substance by introducing a variety of “model Riemannian manifolds” that
should help to motivate the general theory. These manifolds are distinguished by
having a high degree of symmetry.
We begin by describing the most symmetric model spaces of all—Euclidean
spaces, spheres, and hyperbolic spaces. We analyze these in detail, and prove that
each one has a very large isometry group: not only is there an isometry taking any
point to any other point, but in fact one can find an isometry taking any
orthonormal basis at one point to any orthonormal basis at any other point. As we
will see in Chapter 8, this has strong consequences for the curvatures of these man-
ifolds.
After introducing these very special models, we explore some more general
classes of Riemannian manifolds with symmetry—the invariant metrics on Lie
groups, homogeneous spaces, and symmetric spaces.
At the end of the chapter, we give a brief introduction to some analogous models
in the pseudo-Riemannian case. For the particular case of Lorentz manifolds, these
are the Minkowski spaces, de Sitter spaces, and anti-de Sitter spaces, which are
important model spaces in general relativity.
Symmetries of Riemannian Manifolds

The main feature of the Riemannian manifolds we are going to introduce in this
chapter is that they are all highly symmetric, meaning that they have large groups of
isometries.
Let .M; g/ be a Riemannian manifold. Recall that Iso.M; g/ denotes the set of all
isometries from M to itself, which is a group under composition. We say that .M; g/
is a homogeneous Riemannian manifold if Iso.M; g/ acts transitively on M , which
is to say that for each pair of points p; q 2 M , there is an isometry ' W M ! M such
that '.p/ D q.
56 3 Model Riemannian Manifolds
The isometry group does more than just act on M itself. For every ' 2 Iso.M; g/,
the global differential d' maps TM to itself and restricts to a linear isometry
d'p W Tp M ! T'.p/ M for each p 2 M .
Given a point p 2 M , let Isop .M; g/ denote the isotropy subgroup at p, that
is, the subgroup of Iso.M; g/ consisting of isometries that fix p. For each ' 2
Isop .M; g/, the linear map d'p takes Tp M to itself, and the map Ip W Isop .M; g/ !
GL.Tp M / given by Ip .'/ D d'p is a representation of Isop .M; g/, called the
isotropy representation. We say that M is isotropic at p if the isotropy represen-
tation of Isop .M; g/ acts transitively on the set of unit vectors in Tp M . If M is
isotropic at every point, we say simply that M is isotropic.
There is an even stronger kind of symmetry than isotropy. Let O.M / denote the
set of all orthonormal bases for all tangent spaces of M :
a˚
O.M / D orthonormal bases for Tp M :
p2M
There is an induced action of Iso.M; g/ on O.M /, defined by using the differential

of an isometry ' to push an orthonormal basis at p forward to an orthonormal basis
at '.p/:
' .b1 ; : : : ; bn / D d'p .b1 /; : : : ; d'p .bn / : (3.1)
We say that .M; g/ is frame-homogeneous if this induced action is transitive on
O.M /, or in other words, if for all p; q 2 M and choices of orthonormal bases at
p and q, there is an isometry taking p to q and the chosen basis at p to the one at q.
(Warning: Some authors, such as [Boo86, dC92, Spi79], use the term isotropic to
refer to the property we have called frame-homogeneous.)
Proposition 3.1. Let .M; g/ be a Riemannian manifold.

(a) If M is isotropic at one point and it is homogeneous, then it is isotropic.
(b) If M is frame-homogeneous, then it is homogeneous and isotropic.
Proof. Problem 3-3. t

u
A homogeneous Riemannian manifold looks geometrically the same at every
point, while an isotropic one looks the same in every direction. It turns out that
an isotropic Riemannian manifold is automatically homogeneous; however, a Rie-
mannian manifold can be isotropic at one point without being isotropic (for example,
the paraboloid z D x 2 C y 2 in R3 with the induced metric); homogeneous without
being isotropic anywhere (for example, the Berger metrics on S3 discussed in Prob-
lem 3-10 below); or homogeneous and isotropic without being frame-homogeneous
(for example, the Fubini–Study metrics on complex projective spaces discussed in
Example 2.30). The proofs of these claims will have to wait until we have developed
the theories of geodesics and curvature (see Problems 6-18, 8-5, 8-16, and 8-13).
As mentioned in Chapter 1, the Myers–Steenrod theorem shows that Iso.M; g/
is always a Lie group acting smoothly on M . Although we will not use that result,
in many cases we can identify a smooth Lie group action that accounts for at least
Euclidean Spaces 57
some of the isometry group, and in certain cases we will be able to prove that it is
the entire isometry group.
Euclidean Spaces
The simplest and most important model Riemannian manifold is of course
n-dimensional Euclidean space, which is just Rn with the Euclidean metric gx given
by (2.8).
Somewhat more generally, if V is any n-dimensional real vector space
endowed with an inner product, we can set g.v; w/ D hv; wi for any p 2 V and
any v; w 2 Tp V Š V . Choosing an orthonormal basis .b1 ; : : : ; bn / for V defines a
basis isomorphism from Rn to V that sends .x 1 ; : : : ; x n / to x i bi ; this is easily seen
to be an isometry of .V; g/ with R ; gx , so all n-dimensional inner product spaces
n
are isometric to each other as Riemannian manifolds.

It is easy to construct isometries of the Riemannian manifold Rn ; gx : for exam-
ple, every orthogonal linear transformation A W Rn ! Rn preserves the Euclidean
metric, as does every translation x 7! b C x. It follows that every map of the form
x 7! b C Ax, formed by first applying the orthogonal map A and then translating by
b, is an isometry.
It turns out that the set of all such isometries can be realized as a Lie group
acting smoothly on Rn . Regard Rn as a Lie group under addition, and let
W O.n/ Rn ! Rn be the natural action of O.n/ on Rn . Define the Euclidean
group E.n/ to be the semidirect product Rn Ì O.n/ determined by this action: this
is the Lie group whose underlying manifold is the product space Rn O.n/, with
multiplication given by .b; A/.b 0 ; A0 / D .b C Ab 0 ; AA0 / (see Example C.12). It has a
faithful representation given by the map W E.n/ ! GL.n C 1; R/ defined in block
form by
Ab
.b; A/ D ;
0 1
where b is considered an n 1 column matrix.
The Euclidean group acts on Rn via
.b; A/ x D b C Ax: (3.2)
Problem 3-1 shows that when Rn is endowed n with the Euclidean metric, this action
is isometric and the induced action on O R is transitive. (Later, we will see that
this is in fact the full isometry group of R ; gx —see Problem 5-11—but we do not
n
need that fact now.) Thus each Euclidean space is frame-homogeneous.

Fig. 3.1: Transitivity of O.n C 1/ on O.Sn .R//
Spheres
Our second class of model Riemannian manifolds comes in a family, with one for
each positive real number. Given R > 0, let Sn .R/ denote the sphere of radius R
centered at the origin in RnC1 , endowed with the metric gV R (called the round metric
of radius R) induced from the Euclidean metric on RnC1 . When R D 1, it is the
round metric on Sn introduced in Example 2.13, and we use the notation gV D gV 1 .
One of the first things one notices about the spheres is that like Euclidean spaces,
they are highly symmetric. We can immediately write down a large group of isome-
tries of Sn .R/ by observing that the linear action of the orthogonal group O.n C 1/
on RnC1 preserves Sn .R/ and the Euclidean metric, so its restriction to Sn .R/ acts
isometrically on the sphere. (Problem 5-11 will show that this is the full isometry
group.)
Proposition 3.2. The group O.n C 1/ acts transitively on O .Sn .R//, and thus each
round sphere is frame-homogeneous.
Proof. It suffices to show that given any p 2 Sn .R/ and any orthonormal ba-
sis .bi / for Tp Sn .R/, there is an orthogonal map that takes the “north pole”
N D .0; : : : ; 0; R/ to p and the basis .@1 ; : : : ; @n / for TN Sn .R/ to .bi /.
To do so, think of p as a vector of length R in RnC1 , and let py D p=R denote
the unit vector in the same direction (Fig. 3.1). Since the basis vectors .bi / are
tangent to the sphere, they are orthogonal to p, y is an orthonormal
y so .b1 ; : : : ; bn ; p/
Spheres 59
basis for RnC1 . Let ˛ be the matrix whose columns are these basis vectors. Then
˛ 2 O.n C 1/, and by elementary linear algebra, ˛ takes the standard basis vectors
y It follows that ˛.N / D p. Moreover, since ˛ acts
.@1 ; : : : ; @nC1 / to .b1 ; : : : ; bn ; p/.
linearly on RnC1 , its differential d˛N W TN RnC1 ! Tp RnC1 is represented in stan-
dard coordinates by the same matrix as ˛ itself, so d˛N .@i / D bi for i D 1; : : : ; n,
and ˛ is the desired orthogonal map. t
u
Another important feature of the round metrics—one that is much less evident
than their symmetry—is that they bear a certain close relationship to the Euclidean
metrics, which we now describe. Two metrics g1 and g2 on a manifold M are said
to be conformally related (or pointwise conformal or just conformal) to each other
if there is a positive function f 2 C 1 .M / such that g2 D fg1 . Given two Rie-
mannian manifolds .M; g/ and M ; gz , a diffeomorphism ' W M ! M is called a
conformal diffeomorphism (or a conformal transformation) if it pulls gz back to a
metric that is conformal to g:
' gz D fg for some positive f 2 C 1 .M /:
Problem 3-6 shows that conformal diffeomorphisms are the same as angle-pre-
serving diffeomorphisms. Two Riemannian manifolds are said to be conformally
equivalent if there is a conformal diffeomorphism between them.
A Riemannian manifold .M; g/ is said to be locally conformally flat if every
point of M has a neighborhood that is conformally equivalent to an open set in
.Rn ; gx/.
I Exercise 3.3. (a) Show that for every smooth manifold M , conformality is an equiva-
lence relation on the set of all Riemannian metrics on M .
(b) Show that conformal equivalence is an equivalence relation on the class of all Rie-
mannian manifolds.
I Exercise 3.4. Suppose g1 and g2 D fg1 are conformally related metrics on an ori-
ented n-manifold. Show that their volume forms are related by d Vg2 D f n=2 d Vg1 .
A conformal equivalence between Rn and Sn .R/ minus a point is provided by

1 W S n.R/X fN g !
n
stereographic projection from the north pole. This is the map
R that sends a point P 2 S .R/ X fN g, written P D ; : : : ; ; , to u D
n n
.u1 ; : : : ; un / 2 Rn , where U D .u1 ; : : : ; un ; 0/ is the point where the line through

N and P intersects the hyperplane f.; / W D 0g in RnC1 (Fig. 3.2). Thus U is
characterized by the fact that .U N / D .P N / for some nonzero scalar . Writ-
ing N D .0; R/, U D .u; 0/, and P D .; / 2 RnC1 D Rn R, we obtain the system
of equations
ui D i ;
(3.3)
R D . R/:
Solving the second equation for and plugging it into the first equation, we get
the following formula for stereographic projection from the north pole of the sphere
of radius R:
Fig. 3.2: Stereographic projection
R
.; / D u D : (3.4)
R
It follows from this formula that is defined and smooth on all of Sn .R/ X fN g.
The easiest way to see that it is a diffeomorphism is to compute its inverse. Solving
the two equations of (3.3) for and i gives
ui 1
i D ; DR : (3.5)

The point P D 1 .u/ is characterized by these equations and the fact that P is on
the sphere. Thus, substituting (3.5) into jj2 C 2 D R2 gives
juj2 2 . 1/
2
C R D R2 ;
2 2
from which we conclude
juj2 C R2
D :
2R2
Inserting this back into (3.5) gives the formula

2R2 u juj2 R2
1 .u/ D .; / D ; R ; (3.6)
juj2 C R2 juj2 C R2
which by construction maps Rn back to Sn .R/ X fN g and shows that is a diffeo-

morphism.
Spheres 61
Proposition 3.5. Stereographic projection is a conformal diffeomorphism between

Sn .R/ X fN g and Rn .
Proof. The inverse map 1 is a smooth parametrization of Sn .R/ X fN g, so we

can use it to compute the pullback metric. Using the usual technique of substitution
to compute pullbacks, we obtain the following coordinate representation of gV R in
stereographic coordinates:
X 2R2 uj 2 juj2 R2 2

1 1
. / gV R D . / gx D d C d R 2 :
juj2 C R2 juj C R2
j
If we expand each of these terms individually, we get

P
2R2 uj 2R2 duj 4R2 uj i ui dui
d D 2 I
juj2 C R2 juj C R2 .juj2 C R2 /2
P P
juj2 R2 2R i ui dui 2R.juj2 R2 / i ui dui
d R 2 D
juj C R2 juj2 C R2 .juj2 C R2 /2
P
4R3 i ui dui
D :
.juj2 C R2 /2
Therefore,
P j 2 P i i 2 P i i 2
1
4R4 j du / 16R4 i u du 16R4 juj2 i u du
. / gV R D C
.juj C R /
2 2 2 .juj C R /
2 2 3 .juj C R2 /4
2
P 2
16R6 i
i u du
i
C
.juj2 C R2 /4
4
P j 2
4R j du /
D :
.juj C R2 /2
2
In other words,
4R4
. 1 / gV R D gx; (3.7)
.juj2 C R2 /2
where gx now represents the Euclidean metric on Rn , and so is a conformal
diffeomorphism. t
u
Corollary 3.6. Each sphere with a round metric is locally conformally flat.
Proof. Stereographic projection gives a conformal equivalence between a neighbor-

hood of any point except the north pole and Euclidean space; applying a suitable ro-
tation and then stereographic projection (or stereographic projection from the south
pole), we get such an equivalence for a neighborhood of the north pole as well. u t
Hyperbolic Spaces
Our third class of model Riemannian manifolds is perhaps less familiar than the
other two. For each n 1 and each R > 0 we will define a frame-homogeneous
Riemannian manifold Hn .R/, called hyperbolic space of radius R. There are four
equivalent models of the hyperbolic spaces, each of which is useful in certain con-
texts. In the next theorem, we introduce all of them and show that they are isometric.
Theorem 3.7. Let n be an integer greater than 1. For each fixed R > 0, the following
Riemannian manifolds are all mutually isometric.
(a) (H YPERBOLOID MODEL ) Hn .R/ 1is the submanifold
of Minkowski space Rn;1
defined in standard coordinates ; : : : ; ; as the “upper sheet” f > 0g of
n
2 2
the two-sheeted hyperboloid 1 C C n 2 D R2 , with the induced
metric
1
gM R D qx;
where W Hn .R/ ! Rn;1 is inclusion, and qx D qx.n;1/ is the Minkowski metric:
2 2
qx D d 1 C C d n .d /2 : (3.8)
(b) (B ELTRAMI –K LEIN MODEL ) Kn .R/ is the ball of radius R centered at the
origin in Rn , with the metric given in coordinates .w 1 ; : : : ; w n / by
1
n 2
2 w dw C C w dw
1 n
2 .dw / C C .dw n /2
1 2
2
gM R DR CR : (3.9)
R2 jwj2 R2 jwj2 2
(c) (P OINCAR É BALL MODEL ) Bn .R/ is the ball of radius R centered at the
origin in Rn , with the metric given in coordinates .u1 ; : : : ; un / by
3 .du1 /2 C C .dun /2
gM R D 4R4 :
.R2 juj2 /2
(d) (P OINCAR É HALF - SPACE MODEL ) U n .R/ is the upper half-space in Rn

defined in coordinates .x 1 ; : : : ; x n1 ; y/ by U n .R/ D f.x; y/ W y > 0g, endowed
with the metric
4 .dx 1 /2 C C .dx n1 /2 C dy 2

gM R D R2 :
y2
Proof. Let R > 0 be given. We need to verify that Hn .R/ is actually a Riemannian
submanifold of Rn;1 , or in other words that gM R 1
is positive definite. One way to do
this is to show, as we will below, that it is the pullback of gM R
2
or gM R
3
(both of which
are manifestly positive definite) by a diffeomorphism. Alternatively, here is a direct
proof using some of the theory of submanifolds of pseudo-Riemannian manifolds
developed in Chapter 1.
Hyperbolic Spaces 63
Fig. 3.3: Isometries among the hyperbolic models
Note that Hn .R/ is an open subset of a level set of the smooth function
2 2
f W Rn;1 ! R given by f .; / D 1 C C n 2 . We have
df D 2 1 d 1 C C 2 n d n 2 d ;
and therefore the gradient of f with respect to qx is given by
@ @ @
grad f D 2 1 C C 2 n n C 2 : (3.10)
@ 1 @ @
Direct computation shows that
X
i 2
qx.grad f; grad f / D 4 2 ;
i
which is equal to 4R2 at points of Hn .R/. Thus it follows from Corollary 2.71 that
Hn .R/ is a pseudo-Riemannian submanifold of signature .n; 0/, which is to say it
is Riemannian.
We will show that all four Riemannian manifolds are mutually isometric by defin-
ing isometries c W Hn .R/ ! Kn .R/, W Hn .R/ ! Bn .R/, and
W Bn .R/ ! U n .R/
(shown schematically in Fig. 3.3).
We begin with a geometric construction of a diffeomorphism called central pro-
jection from the hyperboloid to the ball,
c W Hn .R/ ! Kn .R/;
which turns out to be an isometry

between the two metrics given in (a) and (b).
For any P D 1 ; : : : ; n ; 2 Hn .R/ Rn;1 , set c.P / D w 2 Kn .R/, where
W D .w; R/ 2 Rn;1 is the point where the line from the origin to P intersects the
hyperplane f.; / W D Rg (Fig. 3.4). Because W is characterized as the unique
scalar multiple of P whose last coordinate is R, we have W D RP = , and therefore
Fig. 3.4: Central projection from the hyperboloid to the Beltrami–Klein model
c is given by the formula
R
c.; / D : (3.11)

The relation jj2 2 D R2 guarantees that jc.; /j2 D R2 .1 R2 = 2 / < R2 , so
c maps Hn .R/ into Kn .R/. To show that c is a diffeomorphism, we determine its
inverse map. Let w 2 Kn .R/ be arbitrary. The unique positive scalar such that the
point .; / D .w; R/ lies on Hn .R/ is characterized by 2 jwj2 2 R2 D R2 ,
and therefore
R
D p :
R2 jwj2
It follows that the following smooth map is an inverse for c:
!
1 Rw R2
c .w/ D .; / D p ;p : (3.12)
R2 jwj2 R2 jwj2
Thus c is a diffeomorphism. To show that it is an isometry between gM R 1

and gM R
2
, we
use the fact that gM R is the metric induced from qx, analogously to the computation
1
we did for stereographic projection above. With .; / defined by (3.12), we have
P
i R dw i Rw i j w j dw j
d D p C 3=2 ;
R2 jwj2 R2 jwj2
P
R2 j w j dw j
d D 3=2 :
R2 jwj2
1 P
It is then straightforward to compute that c 1 gM R D i .d i /2 .d /2 D gM R
2
.
Hyperbolic Spaces 65
Fig. 3.5: Hyperbolic stereographic projection
Next we describe a diffeomorphism
W Hn .R/ ! Bn .R/
from the hyperboloid to the ball, called hyperbolic stereographic projection, which
is an isometry between the metrics 1of (a) and (c). Let S 2 Rn;1 denote the point
S D .0; : : : ; 0; R/. For any P D ; : : : ; ; 2 H .R/ Rn;1 , set .P / D u 2
n n
Bn .R/, where U D .u; 0/ 2 Rn;1 is the point where the line through S and P inter-
sects the hyperplane f.; / W D 0g (Fig. 3.5). The point U is characterized by
.U S / D .P S / for some nonzero scalar , or
ui D i ;
(3.13)
R D . C R/:
These equations can be solved in the same manner as in the spherical case to yield
R
.; / D u D ;
RC
which takes its values in Bn .R/ because j .; /j2 D R2 . 2 R2 /. 2 C R2 / < R2 .
A computation similar to the ones before shows that the inverse map is

1 2R2 u R2 C juj2
.u/ D .; / D ;R 2 :
R2 juj2 R juj2
We will show that . 1 / gM R

1
D gM R
3
. The computation proceeds just as in the
spherical case, so we skip over most of the details:
X 2R2 uj 2 R2 C juj2 2

. 1 / gM R
1
D d d R 2
R2 juj2 R juj2
j
P
4R4 j duj /2
D
.R2 juj2 /2
3
D gM R :
Next we consider the Poincaré half-space model, by constructing an explicit dif-

feomorphism

W U n .R/ ! Bn .R/:
In this case it is more convenient to write the coordinates on the ball as .u; v/ D
.u1 ; : : : ; un1 ; v/. In the 2-dimensional case,
is easy to write down in complex
notation w D uCiv and z D x Ciy. It is a variant of the classical Cayley transform:
z iR

.z/ D w D iR : (3.14)
z C iR
Elementary complex analysis shows that this is a complex-analytic diffeomorphism
taking U 2 .R/ onto B2 .R/. Separating z into real and imaginary parts, we can also
write this in real terms as

2R2 x jxj2 C jyj2 R2

.x; y/ D .u; v/ D ;R 2 : (3.15)
jxj2 C .y C R/2 jxj C .y C R/2
This same formula makes sense in any dimension n if we interpret x to mean

.x 1 ; : : : ; x n1 /, and it is easy to check that it maps the upper half-space fy > 0g
into the ball of radius R. A direct computation shows that its inverse is

2R2 u R2 juj2 v 2

1 .u; v/ D .x; y/ D ; R ;
juj2 C .v R/2 juj2 C .v R/2
so
is a diffeomorphism, called the generalized Cayley transform. The verification
that
gM R
3
D gM R
4
is basically a long calculation, and is left to Problem 3-4. t
u
We often use the generic notation Hn .R/ to refer to any one of the Riemannian
manifolds of Theorem 3.7, and gM R to refer to the corresponding metric; the special
case R D 1 is denoted by .Hn ; g/M and is called simply hyperbolic space, or in the
2-dimensional case, the hyperbolic plane.
Because all of the models for a given value of R are isometric to each other, when
analyzing them geometrically we can use whichever model is most convenient for
the application we have in mind. The next corollary is an example in which the
Poincaré ball and half-space models serve best.
Corollary 3.8. Each hyperbolic space is locally conformally flat.

Invariant Metrics on Lie Groups 67
Proof. In either the Poincaré ball model or the half-space model, the identity map
gives a global conformal equivalence with an open subset of Euclidean space. t
u
The examples presented so far might give the impression that most Riemannian
manifolds are locally conformally flat. This is far from the truth, but we do not yet
have the tools to prove it. See Problem 8-25 for some explicit examples of Riemann-
ian manifolds that are not locally conformally flat.
The symmetries of Hn .R/ are most easily seen in the hyperboloid model. Let
O.n; 1/ denote the group of linear maps from Rn;1 to itself that preserve the
Minkowski metric, called the .n C 1/-dimensional Lorentz group. Note that each
element of O.n; 1/ preserves the hyperboloid f 2 jj2 D R2 g, which has two com-
ponents determined by > 0 and < 0. We let OC .n; 1/ denote the subgroup of
O.n; 1/ consisting of maps that take the > 0 component of the hyperboloid to
itself. (This is called the orthochronous Lorentz group, because physically it repre-
sents coordinate changes that preserve the forward time direction.) Then OC .n; 1/
preserves Hn .R/, and because it preserves qx it acts isometrically on Hn .R/.
(Problem 5-11 will show that this is the full isometry group.) Recall that O Hn .R/
denotes the set of all orthonormal bases for all tangent spaces of Hn .R/.

Proposition 3.9. The group OC .n; 1/ acts transitively on O Hn .R/ , and therefore
Hn .R/ is frame-homogeneous.
Proof. The argument is entirely analogous to the proof of Proposition 3.2, so we

give only a sketch. Suppose p 2 Hn .R/ and .bi / is an orthonormal basis for
Tp Hn .R/. Identifying p 2 Rn;1 with an element of Tp Rn;1 in the usual way, we
can regard py D p=R as a qx-unit vector in Tp Rn;1 , and (3.10) shows that it is a
scalar multiple of the qx-gradient of the
defining function f and thus is orthogonal
to Tp Hn .R/ with respect to qx. Thus b1 ; : : : ; bn ; bnC1 D py is a qx-orthonormal
basis
for Rn;1 , and qx has the following expression in terms of the dual basis ˇ j :
qx D .ˇ 1 /2 C C .ˇ n /2 .ˇ nC1 /2 :
Thus the matrix whose columns are .b1 ; : : : ; bnC1 / is an element of OC .n; 1/ send-
ing N D .0; : : : ; 0; R/ to p and @i to bi (Fig. 3.6). t
u
Invariant Metrics on Lie Groups

Lie groups provide us with another large class of homogeneous Riemannian mani-
folds. (See Appendix C for a review of the basic facts about Lie groups that we will
use.)
Let G be a Lie group. A Riemannian metric g on G is said to be left-invariant
if it is invariant under all left translations: L' g D g for all ' 2 G. Similarly, g is
right-invariant if it is invariant under all right translations, and bi-invariant if it is
Fig. 3.6: Frame homogeneity of Hn .R/
both left- and right-invariant. The next lemma shows that left-invariant metrics are
easy to come by.
Lemma 3.10. Let G be a Lie group and let g be its Lie algebra of left-invariant
vector fields.
(a) A Riemannian metric g on G is left-invariant if and only if for all X; Y 2 g, the
function g.X; Y / is constant on G.
(b) The restriction map g 7! ge 2 †2 Te G together with the natural identifica-
tion Te G Š g gives a bijection between left-invariant Riemannian metrics on
G and inner products on g.
Thus all we need to do to construct a left-invariant metric is choose any inner

product on g, and define a metric on G by applying that inner product to left-
invariant vector fields. Right-invariant metrics can be constructed in a similar way
using right-invariant vector fields. Since a Lie group acts transitively on itself by
either left or right translation, every left-invariant or right-invariant metric is homo-
geneous.
Much more interesting are the bi-invariant metrics, because, as you will be able
to prove later (Problems 7-13 and 8-17), their curvatures are intimately related to
the structure of the Lie algebra of the group. But bi-invariant metrics are generally
much rarer than left-invariant or right-invariant ones; in fact, some Lie groups have
no bi-invariant metrics at all (see Problems 3-12 and 3-13). Fortunately, there is a
complete answer to the question of which Lie groups admit bi-invariant metrics,
which we present in this section.
We begin with a proposition that shows how to determine whether a given left-
invariant metric is bi-invariant, based on properties of the adjoint representation of
the group. Recall that this is the representation Ad W G ! GL.g/ given by Ad.'/ D
.C' / W g ! g, where C' W G ! G is the automorphism defined by conjugation:
C' . / D ' ' 1 . See Appendix C for more details.
Proposition 3.12. Let G be a Lie group and g its Lie algebra. Suppose g is a
left-invariant Riemannian metric on G, and let h; i denote the corresponding inner
product on g as in Lemma 3.10. Then g is bi-invariant if and only if h; i is invariant
under the action of Ad.G/ GL.g/, in the sense that hAd.'/X; Ad.'/Y i D hX; Y i
for all X; Y 2 g and ' 2 G.
Proof. We begin the proof with some preliminary computations. Suppose g is left-
invariant and h; i is the associated inner product on g. Let ' 2 G be arbitrary,
and note that C' is the composition of left multiplication by ' followed by right
multiplication by ' 1 . Thus for every X 2 g, left-invariance implies .R' 1 / X D
.R' 1 / .L' / X D .C' / X D Ad.'/X . Therefore, for all 2 G and X; Y 2 g, we
have

R' 1 g .X ; Y / D g ' 1 R' 1 X ' 1 ; R' 1 Y ' 1

D g ' 1 .Ad.'/X / ' 1 ; .Ad.'/Y / ' 1
D hAd.'/X; Ad.'/Y i:
Now assume that h; i is invariant under Ad.G/. Then the expression
on the last
line above is equal to hX; Y i D g .X ; Y /, which shows that R' 1 g D g. Since
this is true for all ' 2 G, it follows that g is bi-invariant.
Conversely, assuming that g is bi-invariant, we have R' 1 g D g for each
' 2 G, so the above computation yields

hX; Y i D g .X ; Y / D R' 1 g .X ; Y / D hAd.'/X; Ad.'/Y i;
which shows that h; i is Ad.G/-invariant. t

u
In order to apply the preceding proposition, we need a lemma about finding
invariant inner products on vector spaces. Recall from Appendix C that for every
finite-dimensional real vector space V , GL.V / denotes the Lie group of all invertible
linear maps from V to itself. If H is a subgroup of GL.V /, an inner product h; i on V
is said to be H -invariant if hhx; hyi D hx; yi for all x; y 2 V and h 2 H .
Lemma 3.13. Suppose V is a finite-dimensional real vector space and H is a sub-
group of GL.V /. There exists an H -invariant inner product on V if and only if H
has compact closure in GL.V /.
Proof. Assume first that there exists an H -invariant inner product h; i on V . This
implies that H is contained in the subgroup O.V / GL.V / consisting of linear
isomorphisms of V that are orthogonal with respect to this inner product. Choos-
ing an orthonormal basis of V yields a Lie group isomorphism between O.V / and
O.n/ GL.n; R/ (where n D dim V ), so O.V / is compact; and the closure of H is
a closed subset of this compact group, and thus is itself compact.
Conversely, suppose H has compact closure in GL.V /, and let K denote the
closure. A simple limiting argument shows that K is itself a subgroup, and thus it is
a Lie group by the closed subgroup theorem (Thm. C.8). Let h; i0 be an arbitrary
inner product on V , and let be a right-invariant volume form on K (for example,
the volume form of some right-invariant metric on K). For fixed x; y 2 V , define
a smooth function fx;y W K ! R by fx;y .k/ D hkx; kyi0 . Then define a new inner
product h; i on V by Z
hx; yi D fx;y :
K
It follows directly from the definition that h; i is symmetric and bilinear over R.
For each nonzero x 2 V , we have fx;x > 0 everywhere on K, so hx; xi > 0, showing
that h; i is indeed an inner product.
To see that it is invariant under K, let k0 2 K be arbitrary. Then for all x; y 2 V
and k 2 K, we have
˝ ˛
fk0 x;k0 y .k/ D kk0 x; kk0 y 0
D fx;y ı Rk0 .k/;
where Rk0 W K ! K is right translation by k0 . Because is right-invariant, it fol-

lows from diffeomorphism invariance of the integral that
Z
hk0 x; k0 yi D fk0 x;k0 y
ZK

D fx;y ı Rk0
ZK

D Rk0 fx;y
ZK
D fx;y D hx; yi:
K
Thus h; i is K-invariant, and it is also H -invariant because H K. t

u
Theorem 3.14 (Existence of Bi-invariant Metrics). Let G be a Lie group and g its
Lie algebra. Then G admits a bi-invariant metric if and only if Ad.G/ has compact
closure in GL.g/.
Proof. Proposition 3.12 shows that there is a bi-invariant metric on G if and only if
there is an Ad.G/-invariant inner product on g, and Lemma 3.13 in turn shows that
the latter is true if and only if Ad.G/ has compact closure in GL.g/. t
u
The most important application of the preceding theorem is to compact groups.
Corollary 3.15. Every compact Lie group admits a bi-invariant Riemannian metric.
Proof. If G is compact, then Ad.G/ is a compact subgroup of GL.g/ because

Ad W G ! GL.g/ is continuous. t
u
Another important application is to prove that certain Lie groups do not admit
bi-invariant metrics. One way to do this is to note that if Ad.G/ has compact closure
in GL.g/, then every orbit of Ad.G/ must be a bounded subset of g with respect to
any choice of norm, because it is contained in the image of the compact set Ad.G/
under a continuous map of the form ' 7! '.X0 / from GL.g/ to g. Thus if one can
find an element X0 2 g and a subset S G such that the elements of the form
Ad.'/X are unbounded in g for ' 2 S , then there is no bi-invariant metric.
Here are some examples.
Example 3.16 (Invariant Metrics on Lie Groups).
(a) Every left-invariant metric on an abelian Lie group is bi-invariant, because the
adjoint representation is trivial. Thus the Euclidean metric on Rn and the flat
metric on T n of Example 2.21 are both bi-invariant.
(b) If a metric g on a Lie group G is left-invariant, then the induced metric on
every Lie subgroup H G is easily seen to be left-invariant. Similarly, if g is
bi-invariant, then the induced metric on H is bi-invariant.
(c) The Lie group SL.2; R/ (the group of 2 2 real matrices of determinant 1)
admits many left-invariant metrics (as does every positive-dimensional Lie
group), but no bi-invariant ones. To see this, recall that the Lie algebra of
SL.2; R/ is isomorphic to the algebra sl.2; R/ of trace-free 2 2 matrices,
1
and the adjoint representation
0 1 is given by Ad.A/Xc D0
AXA (see Example
C.10). If we let X0 D 0 0 2 sl.2; R/ and Ac D 0 1=c 2 SL.2; R/ for c > 0,

then Ad.Ac /X0 D 00 0c 2 , which is unbounded as c ! 1. Thus the orbit of
X0 is not contained in any compact subset, which implies that there is no bi-
invariant metric on SL.2; R/. A similar argument shows that SL.n; R/ admits
no bi-invariant metric for any n 2. In view of (b) above, this shows also
that GL.n; R/ admits no bi-invariant metric for n 2. (Of course, GL.1; R/
does admit bi-invariant metrics because it is abelian.)
(d) With S3 regarded as a submanifold of C 2 , the map

w z
.w; z/ 7! (3.16)
x
z wx
gives a diffeomorphism from S3 to SU.2/. Under the inverse of this map, the
round metric on S3 pulls back to a bi-invariant metric on SU.2/, as Problem
3-10 shows.
(e) Let o.n/ denote the Lie algebra of O.n/, identified with the algebra of skew-
symmetric n n matrices, and define a bilinear form on o.n/ by

hA; Bi D tr AT B :
This is an Ad-invariant inner product, and thus determines a bi-invariant Rie-

mannian metric on O.n/ (see Problem 3-11).
(f) Let U n be the upper half-space as defined in Theorem 3.7. We canregard
U n as a Lie group by identifying each point .x; y/ D x 1 ; : : : ; x n1 ; y 2 U n
with an invertible n n matrix as follows:

In1 0
.x; y/ ! ;
xT y
where In1 is the .n 1/ .n 1/ identity matrix. Then the hyperbolic metric

gM R
4
is left-invariant on U n but not right-invariant (see Problem 3-12).
(g) For n 1, the .2n C 1/-dimensional Heisenberg group is the Lie subgroup
Hn GL.n C 2; R/ defined by
˚1 x z T

Hn D 0 1 y W x; y 2 Rn ; z 2 R ;
0 0 1
where x and y are treated as column matrices. These are the simplest examples
of nilpotent Lie groups, meaning that the series of subgroups G ŒG; G
ŒG; ŒG; G eventually reaches the trivial subgroup (where for any sub-
groups G1 ; G2 G, the notation ŒG1 ; G2 means the subgroup of G generated
by all elements of the form x1 x2 x11 x21 for xi 2 Gi ). There are many left-
invariant metrics on Hn , but no bi-invariant ones, as Problem 3-13 shows.
(h) Our last example is a group that plays an important role in the classification
of 3-manifolds. Let Sol denote the following 3-dimensional Lie subgroup of
GL.3; R/: ˚ ez 0 x

Sol D 0 e z y W x; y; z 2 R :
0 0 1
This group is the simplest nonnilpotent example of a solvable Lie group,
meaning that the series of subgroups G ŒG; G ŒŒG; G; ŒG; G even-
tually reaches the trivial subgroup. Like the Heisenberg groups, Sol admits
left-invariant metrics but not bi-invariant ones (Problem 3-14). //
In fact, John Milnor showed in 1976 [Mil76] that the only Lie groups that admit
bi-invariant metrics are those that are isomorphic to direct products of compact
groups and abelian groups.
Other Homogeneous Riemannian Manifolds
There are many homogeneous Riemannian manifolds besides the frame-homoge-

neous ones and the Lie groups with invariant metrics. To identify other examples, it
is natural to ask the following question: If M is a smooth manifold endowed with a
smooth, transitive action by a Lie group G (called a homogeneous G-space or just
a homogeneous space), is there a Riemannian metric on M that is invariant under
the group action?
The next theorem gives a necessary and sufficient condition for existence of an
invariant Riemannian metric that is usually easy to check.
Other Homogeneous Riemannian Manifolds 73
Theorem 3.17 (Existence of Invariant Metrics on Homogeneous Spaces). Sup-

pose G is a Lie group and M is a homogeneous G-space. Let p0 be a point in M ,
and let Ip0 W Gp0 ! GL Tp0 M denote the isotropy representation at p0 . There
exists a G-invariant
Riemannian
metric on M if and only if Ip0 Gp0 has compact
closure in GL Tp0 M .
Proof. Assume first that g is a G-invariant metric on M . Then the inner product gp0
on Tp0 M is invariant under the isotropy representation,
so it follows from Lemma
3.13 that Ip0 .Gp0 / has compact closure
in GL T p 0
M .
Conversely, assume that Ip0 Gp0 has compact closure in GL Tp0 M . Lemma
3.13 shows that there is an inner product gp0 on Tp0 .M / that is invariant under
the isotropy representation. For arbitrary p 2 M , we define an inner product gp on
Tp M by choosing an element ' 2 G such that '.p/ D p0 and setting

gp D d'p gp0 :
If '1 ; '2 are any two such elements of G, then '1 D h'2 with h D '1 '21 2 Gp0 , so

d'1 jp gp0 D d.h'2 /p gp0 D d'2 jp dhp0 gp0 D d'2 jp gp0 ;
showing that g is well defined as a rough tensor field on M . An easy computation

shows that g is G-invariant, so it remains only to show that it is smooth.
The map W G ! M given by . / D p0 is a smooth surjection because the
action is smooth and transitive. Given ' 2 G, if we let ' W M ! M denote the map
p 7! ' p and L' W G ! G the left translation by ', then the map satisfies
ı L' . / D .' / p0 D ' . p0 / D ' ı . /; (3.17)
so it is equivariant with respect to these two actions. Thus it is a submersion by the

equivariant rank theorem (Thm. C.14).
Define a rough 2-tensor field on G by D g. (It will typically not be positive
definite, because e .v; w/ D 0 if either v or w is tangent to the isotropy group Gp0
and thus in the kernel of d e .) For all ' 2 G, (3.17) implies

L' D L' g D ı L' g D ' ı g D ' g D g D ;
where the next-to-last equality follows from the G-invariance of g. Thus is a

left-invariant tensor field on G. Every basis .X1 ; : : : ; Xn / for the Lie algebra of G
forms a smooth global left-invariant frame for G, and with respect to such a frame
the components .Xi ; Xj / are constant; thus is a smooth tensor field on G.
For each p 2 M , the fact that is a surjective smooth submersion implies that
there exist a neighborhood U of p and a smooth local section W U ! G (Thm.
A.17). Then ˇ
g ˇU D . ı / g D g D ;
showing that g is smooth on U . Since this holds in a neighborhood of each point, g
is smooth. t
u
The next corollary, which follows immediately from Theorem 3.17, addresses
the most commonly encountered case. (Other necessary and sufficient conditions
for the existence of invariant metrics are given in [Poo81, 6.58–6.59].)
Corollary 3.18. If a Lie group G acts smoothly and transitively on a smooth man-
ifold M with compact isotropy groups, then there exists a G-invariant Riemannian
metric on M . t
u
I Exercise 3.19. Suppose G is a Lie group and M is a homogeneous G-space that

admits at least one g-invariant metric. Show that for each p 2 M , the map g 7! gp gives
a bijection between G-invariant metrics on M and Ip .Gp /-invariant inner products on
Tp M .
Locally Homogeneous Riemannian Manifolds
A Riemannian manifold .M; g/ is said to be locally homogeneous if for every pair

of points p; q 2 M there is a Riemannian isometry from a neighborhood of p to a
neighborhood of q that takes p to q. Similarly, we say that .M; g/ is locally frame-
homogeneous if for every p; q 2 M and every pair of orthonormal bases .vi / for
Tp M and .wi / for Tq M , there is an isometry from a neighborhood of p to a neigh-
borhood of q that takes p to q, and whose differential takes vi to wi for each i .
Every homogeneous Riemannian manifold is locally homogeneous, and every
frame-homogeneous one is locally frame-homogeneous. Every proper open subset of
a homogeneous or frame-homogeneous Riemannian manifold is locally homoge-
neous or locally frame-homogeneous, respectively. More interesting examples arise
in the following way.

Proposition 3.20. Suppose M ; gz is a homogeneous Riemannian manifold, .M; g/

is a Riemannian manifold, and W M ! M is a Riemannian covering. Then .M; g/

is locally homogeneous. If M ; gz is frame-homogeneous, then .M; g/ is locally
frame-homogeneous.
I Exercise 3.21. Prove this proposition.
Locally homogeneous Riemannian metrics play an important role in classifica-

tion theorems for manifolds, especially in low dimensions. The most fundamental
case is that of compact 2-manifolds, for which we have the following important
theorem.
Theorem 3.22 (Uniformization of Compact Surfaces). Every compact, connected,

smooth 2-manifold admits a locally frame-homogeneous Riemannian metric, and
a Riemannian covering by the Euclidean plane, hyperbolic plane, or round unit
sphere.
Sketch of proof. The proof relies on the topological classification of compact sur-
faces (see, for example, [LeeTM, Thms. 6.15 and 10.22]), which says that every
Fig. 3.7: A connected sum of tori Fig. 3.8: Constructing a Riemannian covering
connected compact surface is homeomorphic to a sphere, a connected sum of one

or more tori, or a connected sum of one or more projective planes. The crux of
the proof is showing that each of the model surfaces on this list has a metric that
admits a Riemannian covering by one of the model frame-homogeneous manifolds,
and therefore is locally frame-homogeneous by Proposition 3.20. We consider each
model surface in turn.
The 2-sphere: S2 , of course, has its round metric, and the identity map is a Rie-
mannian covering.
metric on T described in Example
2
The 2-torus: Exercise 2.36 shows that the flat
2.21 admits a Riemannian covering by R ; gx . 2
A connected sum of n 2 copies of T 2 : It is shown in [LeeTM, Example 6.13]

that such a surface is homeomorphic to a quotient of a regular 4n-sided polygo-
nal region by side identifications indicated schematically by the sequence of labels
a1 b1 a11 b11 : : : an bn an1 bn1 (Fig. 3.7 illustrates the case n D 2). Let G GL.2; C/
be the following subgroup:

˛ˇ 2 2
GD x̌ ˛ W ˛; ˇ 2 C; j˛j jˇj > 0 : (3.18)
x
Problem 3-8 shows that G acts transitively and isometrically on the Poincaré disk
with its hyperbolic metric. It is shown in [LeeTM] that for each n 2, there is a
discrete subgroup n G such that the quotient map B2 ! B2 = n is a covering
map, and B2 = n is homeomorphic to a connected sum of n tori. The group is found
by first identifying a compact region P B2 bounded by a “regular geodesic poly-
gon,” which is a union of 4n congruent circular arcs making interior angles that all
measure exactly =2n, so that 4n of them fit together locally to fill out a neighbor-
hood of a point (see Fig. 3.8). (The name “geodesic polygon” reflects the fact that
these circular arcs are segments of geodesics with respect to the hyperbolic metric,
as we will see in Chapter 5.) Then n is the group generated by certain elements
of G that take an edge with a label an or bn to the other edge with the same label.
Because n acts isometrically, it follows that such a connected sum admits a locally
Fig. 3.9: The Klein bottle Fig. 3.10: Connected sum of three projective planes
frame-homogeneous metric and a Riemannian covering by the Poincaré disk. For

details, see the proofs of Theorems 12.29 and 12.30 of [LeeTM].
The projective plane: Example 2.34 2 shows
that RP 2 has a metric that admits a
2-sheeted Riemannian covering by S ; gV .
A connected sum of two copies of RP 2 : It is shown in [LeeTM, Lemma 6.16]
that RP 2 # RP 2 is homeomorphic to the Klein bottle, which is the quotient space of
the unit square Œ0; 1 Œ0; 1 by the equivalence relation generated by the following
relations (see Fig. 3.9):
.x; 0/ .x; 1/ for 0 x 1;

(3.19)
.0; y/ .1; 1 y/ for 0 y 1:
Let E.2/ be the Euclidean group in two dimensions as defined earlier in this chapter,
and let E.2/ be the subgroup defined by

1 0
D .b; A/ 2 E.2/ W b D .k; l/ with k; l 2 Z; and A D : (3.20)
0 .1/k
It turns out that acts freely and properly on R2 , and every -orbit has a rep-
resentative in the unit square such that two points in the square are in the same
orbit if and only if they satisfy the equivalence relation generated by the relations in
(3.19). Thus the orbit space R2 = is homeomorphic to the Klein bottle, and since
the group action is isometric, it follows that the Klein bottle inherits a flat, locally
frame-homogeneous metric and the quotient map is a Riemannian covering. Prob-
lem 3-18 asks you to work out the details.
A connected sum of n 3 copies of RP 2 : Such a surface is homeomorphic to a
quotient of a regular 2n-sided polygonal region by side identifications according to
a1 a1 a2 a2 : : : an an [LeeTM, Example 6.13]. As in the case of a connected sum of
tori, there is a compact region P B2 bounded by a 2n-sided regular geodesic poly-
gon whose interior angles are all =n (see Fig. 3.10), and there is a discrete group
of isometries that realizes the appropriate side identifications and yields a quotient
homeomorphic to the connected sum. The new ingredient here is that because such
a connected sum is not orientable, we must work with the full group of isometries
of B2 , not just the (orientation-preserving) ones determined by elements of G; but
otherwise the argument is essentially the same as the one for connected sums of tori.
The details can be found in [Ive92, Section VII.1].
There is one remaining step. The arguments above show that each compact topo-
logical 2-manifold possesses a smooth structure and a locally frame-homogeneous
Riemannian metric, which admits a Riemannian covering by one of the three
frame-homogeneous model spaces. However, we started with a smooth compact
2-manifold, and we are looking for a Riemannian metric that is smooth with respect
to the given smooth structure. To complete the proof, we appeal to a result by James
Munkres [Mun56], which shows that any two smooth structures on a 2-manifold
are related by a diffeomorphism; thus after pulling back the metric by this diffeo-
morphism, we obtain a locally frame-homogeneous metric on M with its originally
given smooth structure. t
u
Locally homogeneous metrics also play a key role in the classification of compact
3-manifolds. In 1982, William Thurston made a conjecture about the classification
of such manifolds, now known as the Thurston geometrization conjecture. The
conjecture says that every compact, orientable 3-manifold can be expressed as a
connected sum of compact manifolds, each of which either admits a Riemannian
covering by a homogeneous Riemannian manifold or can be cut along embedded
tori so that each piece admits a finite-volume locally homogeneous Riemannian
metric. An important ingredient in the analysis leading up to the conjecture was his
classification of all simply connected homogeneous Riemannian 3-manifolds that
admit finite-volume Riemannian quotients. Thurston showed that there are exactly
eight such manifolds (see [Thu97] or [Sco83] for a proof):

R3 with the Euclidean metric

S3 with a round metric

H3 with a hyperbolic metric

S2 R with a product of a round metric and the Euclidean metric

H2 R with a product of a hyperbolic metric and the Euclidean metric

The Heisenberg group H1 of Example 3.16(g) with a left-invariant metric

The group Sol of Example 3.16(h) with a left-invariant metric

The universal covering group of SL.2; R/ with a left-invariant metric
The Thurston geometrization conjecture was proved in 2003 by Grigori Perel-
man. The proof is described in several books [BBBMP, KL08, MF10, MT14].
Symmetric Spaces
We end this section with a brief introduction to another class of Riemannian mani-
folds with abundant symmetry, called symmetric spaces. They turn out to be inter-
mediate between frame-homogeneous and homogeneous Riemannian manifolds.
Here is the definition. If .M; g/ is a Riemannian manifold and p 2 M , a
point reflection at p is an isometry ' W M ! M that fixes p and satisfies d'p D
Id W Tp M ! Tp M . A Riemannian manifold .M; g/ is called a (Riemannian) sym-
metric space if it is connected and for each p 2 M there exists a point reflection at
p. (The modifier “Riemannian” is included to distinguish such spaces from other

kinds of symmetric spaces that can be defined, such as pseudo-Riemannian sym-
metric spaces and affine symmetric spaces; since we will be concerned only with
Riemannian symmetric spaces, we will sometimes refer to them simply as “sym-
metric spaces” for brevity.)
Although we do not yet have the tools to prove it, we will see later that every
Riemannian symmetric space is homogeneous (see Problem 6-19). More generally,
.M; g/ is called a (Riemannian) locally symmetric space if each p 2 M has a neigh-
borhood U on which there exists an isometry ' W U ! U that is a point reflection at
p. Clearly every Riemannian symmetric space is locally symmetric.
The next lemma can be used to facilitate the verification that a given Riemannian
manifold is symmetric.
Lemma 3.23. If .M; g/ is a connected homogeneous Riemannian manifold that pos-

sesses a point reflection at one point, then it is symmetric.
Proof. Suppose .M; g/ satisfies the hypothesis, and let ' W M ! M be a point re-
flection at p 2 M . Given any other point q 2 M , by homogeneity there is an isometry
W M ! M satisfying .p/ D q. Then 'z D ı ' ı 1 is an isometry that fixes
q. Because d p is linear, it commutes with multiplication by 1, so

d 'zq D d p ı IdTpM ı d 1 q D IdTqM ı d p ı d 1 q
D IdTqM :
Thus 'z is a point reflection at q. t

u
Example 3.24 (Riemannian Symmetric Spaces).

(a) Suppose .M; g/ is any connected frame-homogeneous Riemannian manifold.
Then for each p 2 M , we can choose an orthonormal basis .bi / for Tp M ,
and frame homogeneity guarantees that there is an isometry ' W M ! M that
fixes p and sends .bi / to .bi /, which implies that d'p D Id. Thus every
frame-homogeneous Riemannian manifold is a symmetric space. In particular,
all Euclidean spaces, spheres, and hyperbolic spaces are symmetric.
(b) Suppose G is a connected Lie group with a bi-invariant Riemannian metric g.
If we define ˚ W G ! G by ˚.x/ D x 1 , then it is straightforward to check
that d ˚e .v/ D v for every v 2 Te G, from which it follows that d ˚e .ge / D
ge . To see that ˚ is an isometry, let p 2 G be arbitrary. The identity q 1 D
.p 1 q/1 p 1 for all q 2 G implies that ˚ D Rp1 ı ˚ ı Lp1 , and therefore
it follows from bi-invariance of g that
.˚ g/p D d ˚p gp1 D d.Lp1 /p ı d ˚e ı d.Rp1 /e gp1 D gp :
Therefore ˚ is an isometry of g and hence a point reflection at e. Lemma 3.23

then implies that .G; g/ is a symmetric space.
(c) The complex projective spaces introduced in Example 2.30 and the Grassmann
manifolds introduced in Problem 2-7 are all Riemannian symmetric spaces (see
Problems 3-19 and 3-20).
Model Pseudo-Riemannian Manifolds 79
(d) Every product of Riemannian symmetric spaces is easily seen to be a sym-

metric space when endowed with the product metric. A symmetric space is
said to be irreducible if it is not isometric to a product of positive-dimensional
symmetric spaces. //
Model Pseudo-Riemannian Manifolds
The definitions of the Euclidean, spherical, and hyperbolic metrics can easily be
adapted to give analogous classes of frame-homogeneous pseudo-Riemannian mani-
folds.
The first example is one we have already seen: the pseudo-Euclidean
space of
signature .r; s/ is the pseudo-Riemannian manifold Rr;s ; qx.r;s/ , where qx.r;s/ is the
pseudo-Riemannian metric defined by (2.24).
There are also pseudo-Riemannian analogues of the spherical and hyperbolic
metrics. For nonnegative integers r and s and a positive real number R, we define the
.r;s/ .r;s/
pseudosphere Sr;s .R/; qV R and the pseudohyperbolic space Hr;s .R/; qM R as
follows. As manifolds, Sr;s .R/ RrC1;s and Hr;s .R/ Rr;sC1 are defined by
n 2 2 2 2 o
Sr;s .R/ D .; / W 1 C C rC1 1 s D R2 ;
n 2 2 2 2 o
Hr;s .R/ D .; / W 1 C C r 1 sC1 D R2 :
The metrics are the ones induced from the respective pseudo-Euclidean metrics:
.r;s/ .r;s/
qV R D qx.rC1;s/ on Sr;s .R/, and qM R D qx.r;sC1/ on Hr;s .R/.
Theorem 3.25. For all r, s, and R as above, Sr;s .R/ and Hr;s .R/ are pseudo-
Riemannian manifolds of signature .r; s/.

u
It turns out that these pseudo-Riemannian manifolds all have the same degree
of symmetry as the three classes of model Riemannian manifolds introduced ear-
lier. For pseudo-Riemannian manifolds, though, it is necessary to modify the defi-
nition of frame homogeneity slightly. If .M; g/ is a pseudo-Riemannian manifold
of signature .r; s/, let us say that an orthonormal basis for some tangent space
Tp M is in standard order if the expression for gp in terms of the dual basis ."i / is
."1 /2 C C ."r /2 ."rC1 /2 ."rCs /2 , with all positive terms coming before
the negative terms. With this understanding, we define O.M / to be the set of all
standard-ordered orthonormal bases for all tangent spaces to M , and we say that
.M; g/ is frame-homogeneous if the isometry group acts transitively on O.M /.
Theorem 3.26. All pseudo-Euclidean spaces, pseudospheres, and pseudohyperbolic

spaces are frame-homogeneous.

u
.n;1/
In the particular case of signature .n; 1/, the Lorentz manifolds Sn;1 .R/; qV R
.n;1/
and Hn;1 .R/; qM R are called de Sitter space of radius R and anti-de Sitter space
of radius R, respectively.
Problems
3-1. Show that (3.2) defines a smooth isometric action of E.n/ on .Rn ; gx/, and
the induced action on O .Rn / is transitive. (Used on p. 57.)
3-2. Prove that the metric on RP n described in Example 2.34 is frame-homo-
geneous. (Used on p. 145)
3-3. Prove Proposition 3.1 (about homogeneous and isotropic Riemannian mani-
folds).
3-4. Complete the proof of Theorem 3.7 by showing that
gM R
3
D gM R
4
.
n n 2
3-5. (a) Prove that S .R/; gV R is isometric to S ; R gV for each R > 0.

(b) Prove that Hn .R/; gM R is isometric to Hn ; R2 gM for each R > 0.
(c) We could also have defined a family of metrics on Rn by gxR D R2 gx.
Why did we not bother?
(Used on p. 185.)
3-6. Show that two Riemannian metrics g1 and g2 are conformal if and only if
they define the same angles but not necessarily the same lengths, and that a
diffeomorphism is a conformal equivalence if and only if it preserves angles.
[Hint: Let .Ei / be a local orthonormal frame for g1 , and consider the g2 -angle
between Ei and .cos /Ei C .sin /Ej .] (Used on p. 59.)
3-7. Let U 2 denote the upper half-plane model of the hyperbolic plane (of radius
1), with the metric gM D .dx 2 C dy 2 /=y 2 . Let SL.2; R/ denote the group of
2 2 real matrices of determinant 1. Regard U 2 as a subset of the complex
plane with coordinate z D x C iy, and let

az C b ab
Az D ; AD 2 SL.2; R/:
cz C d cd
Show that this defines a smooth,

transitive,
orientation-preserving, and iso-
metric
action of SL.2; R/ on U 2
; M
g . Is the induced action transitive on
O U2 ?
3-8. Let B2 denote the Poincaré disk model of the hyperbolic plane (of radius 1),
with the metric gM D .du2 Cdv 2 /=.1u2 v 2 /2 , and let G GL.2; C/ be the
subgroup defined by (3.18). Regarding B2 as a subset of the complex plane
with coordinate w D u C iv, let G act on B2 by

˛ˇ ˛z C ˇ
x̌ ˛
x wD
x̌z C ˛
x
:
Problems 81
Show that this defines a smooth,

transitive, orientation-preserving, and iso-
metric action of G on B2 ; gM . [Hint: One way to proceed is to define an
action of G on the upper half-plane by A z D
1 ı A ı
.z/, where
is the
Cayley transform defined by (3.14) in the case R D 1, and use the result of
Problem 3-7.] (Used on pp. 73, 185.)
3-9. Suppose G is a compact Lie group with a left-invariant metric g and a left-
invariant orientation. Show that the Riemannian volume form d Vg is bi-
invariant. [Hint: Show that d Vg is equal to the Riemannian volume form for
a bi-invariant metric.]
3-10. Consider the basis

01 0i i 0
XD ; Y D ; ZD
1 0 i 0 0 i
for the Lie algebra su.2/. For each positive real number a, define a left-
invariant metric ga on the group SU.2/ by declaring X; Y; aZ to be an
orthonormal frame.
(a) Show that ga is bi-invariant if and only if a D 1.
(b) Show that the map defined by (3.16) is an isometry between S3 ; gV
and .SU.2/; g1 /. [Remark: SU.2/ with any of these metrics is called a
Berger sphere, named after Marcel Berger.]
(Used on pp. 56, 71, 259.)

3-11. Prove that the formula hA; Bi D tr AT B defines a bi-invariant Riemannian
metric on O.n/. (See Example 3.16(e).)
3-12. Regard the upper half-space U n as a Lie group as described in Example
3.16(f).
(a) Show that for each R > 0, the hyperbolic metric gM R 4
on U n is left-
invariant.
(b) Show that U n does not admit any bi-invariant metrics.
(Used on pp. 68, 72.)
3-13. Write down an explicit formula for an arbitrary left-invariant metric on the
Heisenberg group Hn of Example 3.16(g) in terms of global coordinates
.x 1 ; : : : ; x n ; y 1 ; : : : ; y n ; z/, and show that the group has no bi-invariant met-
rics. (Used on pp. 68, 72.)
3-14. Repeat Problem 3-13 for the group Sol of Example 3.16(h). (Used on p. 72.)
3-15. Let Rn;1 be the .n C 1/-dimensional Minkowski space with coordinates
P 2
.; / D 1 ; : : : ; n ; and with the Minkowski metric qx.n;1/ D i d i
d 2 . Let S Rn;1 be the set
˚
S D .; / W . 1 /2 C C . n /2 D D 1 :
(a) Prove that S is a smooth submanifold diffeomorphic to Sn1 , and with

the induced metric g D S qx.n;1/ it is isometric to the round unit .n 1/-
sphere.
(b) Define an action of the orthochronous Lorentz group OC .n; 1/ on S as
follows: For every p 2 S , let hpi denote the 1-dimensional subspace of
Rn;1 spanned by p. Given A 2 OC .n; 1/, show that the image set A.hpi/
is a 1-dimensional subspace that intersects S in exactly one point, so we
can define A p to be the unique point in S \ A.hpi/. Prove that this is a
smooth transitive action on S .
(c) Prove that OC .n; 1/ acts by conformal diffeomorphisms of .S; g/.
3-16. Prove that there is no Riemannian metric on the sphere that is invariant under
the group action described in Problem 3-15.
3-17. Given a Lie group G, define an action of the product group G G on G
by .'1 ; '2 / D '1 '21 . Show that this action is transitive, and that the
isotropy group of the identity is the diagonal subgroup D f.'; '/ W ' 2 Gg.
Then show that the following diagram commutes:
Š G
Ie Ad
? ?
GL.Te G/ Š GL.g/;
where Ie is the isotropy representation of and g is the Lie algebra of G,
and use this to give an alternative proof of Theorem 3.14.
3-18. Let E.2/ be the subgroup defined by (3.20). Prove that acts freely and
properly on R2 and the orbit space is homeomorphic to the Klein bottle, and
conclude that the Klein bottle has a flat metric and a Riemannian covering by
the Euclidean plane.
3-19. Show that the Fubini–Study metric on CP n (Example 2.30) is homogeneous,
isotropic, and symmetric. (Used on p. 78.)
3-20. Show that the metric on the Grassmannian Gk .Rn / defined in Problem 2-7 is
homogeneous, isotropic, and symmetric. (Used on p. 78.)

3-21. Let M ; gz be a simply connected Riemannian manifold, and suppose 1

and 2 are countable subgroups of Iso M ; gz acting smoothly, freely, and
properly on M (when endowed with the discrete topology). For i D 1; 2, let
Mi D M = i , and let gi be the Riemannian metric on Mi that makes the
quotient map i W M ! Mi a Riemannian covering (see Prop. 2.32). Prove
that the Riemannian manifolds .M1 ; g1 / and .M 2 ; g2 / are isometric if and
only if 1 and 2 are conjugate subgroups of Iso M ; gz .
3-22. Prove Theorem 3.25 (showing that pseudospheres and pseudohyperbolic
spaces are pseudo-Riemannian manifolds). [Hint: Mimic the argument in the
proof of Theorem 3.7 that Hn .R/ is Riemannian.]
Problems 83
3-23. Prove Theorem 3.26 (pseudo-Euclidean spaces, pseudospheres, and pseudo-

hyperbolic spaces are frame-homogeneous).
3-24. Prove that for all positive integers r and s and every real number R >0, both the
pseudohyperbolic space Hr;s .R/ and the pseudosphere Ss;r .R/ are diffeo-
morphic to Rr Ss . [Hint: Consider the maps '; W Rr Ss ! RrCsC1 given
by
p p
'.x; y/ D Rx; 1 C jxj2 Ry ; .x; y/ D 1 C jxj2 Ry; Rx :

3-25. Let Kr .R/; gM R
2
be the r-dimensional ball of radius R with the Beltrami–
z r;1 .R/ be the product manifold Kr .R/ R with
Klein metric (3.9), and let H
the pseudo-Riemannian warped product metric q D gM R 2
˚ .f 2 dt 2 /, where
C
f W K .R/ ! R is given by
r
R2
f .w/ D p :
R2 jwj2
z r;1 .R/ ! Rr;2 by

Define F W H
.Rw; R2 cos t; R2 sin t /

F .w; t / D p :
R2 jwj2
Prove that the image of F is the anti-de Sitter space Hr;1 .R/, and F de-
.r;1/ r;1
fines a pseudo-Riemannian covering of Hr;1 .R/; qR by H z .R/; q .
[Remark: We are tacitly extending the notions of warped product metric and
Riemannian coverings to the pseudo-Riemannian case in the obvious ways.
It follows from the result of Problem 3-24 that Hr;s .R/ is simply
connected
when s 2 but Hr;1 .R/ is not. This shows that H z r;1 .R/; q , called uni-
versal anti-de Sitter space of radius R, is the universal pseudo-Riemannian
.r;1/
covering manifold of Hr;1 .R/; qR .]
Chapter 4
Connections
Our ultimate goal is to define a notion of curvature that makes sense on arbitrary Rie-
mannian manifolds, and to relate it to other geometric and topological properties.
Before we can do so, however, we need to study geodesics, the generalizations to
Riemannian manifolds of straight lines in Euclidean space. There are two key prop-
erties satisfied by straight lines in Rn , either of which serves to characterize them
uniquely: first, every segment of a straight line is the unique shortest path between its
endpoints; and second, straight lines are the only curves that have parametrizations
with zero acceleration.
The first of these characterizations—as shortest paths—is probably the most “ge-
ometric,” so it is tempting to try to use it as a definition of geodesics in Riemannian
manifolds. However, this property turns out to be technically difficult to work with
as a definition, so instead we will use “zero acceleration” as the defining property
and generalize that.
To make sense of acceleration on a manifold, we have to introduce a new object
called a connection—essentially a coordinate-independent set of rules for taking
directional derivatives of vector fields.
We begin this chapter by examining more closely the problem of finding an in-
variant interpretation for the acceleration of a curve, as a way to motivate the defini-
tions that follow. We then give a rather general definition of a connection, in terms
of directional derivatives of sections of vector bundles. After deriving some basic
properties of connections, we show how to use them to differentiate vector fields
along curves, to define geodesics, and to define “parallel transport” of vectors along
curves.
The Problem of Differentiating Vector Fields

To see why we need a new kind of differentiation operator, let us begin by thinking
informally about curves in Rn . Let I R be an interval
1 and nW I ! R a smooth
n
curve, written in standard coordinates as .t / D .t /; : : : ; .t / . Such a curve

86 4 Connections
has a well-defined velocity 0 .t / and acceleration 00 .t / at each t 2 I , computed by

differentiating the components:
ˇ ˇ
0 1 @ ˇˇ n @ ˇˇ
.t / D P .t / 1 ˇ C C P .t / n ˇ ; (4.1)
@x .t/ @x .t/
ˇ ˇ
@ ˇ @ ˇ
00 .t / D R 1 .t / 1 ˇˇ C C R n .t / n ˇˇ : (4.2)
@x
.t/ @x
.t/
(Here and throughout the book, we use dots to denote ordinary derivatives with
respect to t when there are superscripts that would make primes hard to read.) A
curve in Rn is a straight line if and only if it has a parametrization for which
00 .t / 0.
We can also make sense of directional derivatives of vector fields on Rn , just by
computing ordinary directional derivatives
of
the component functions in standard
coordinates: given a vector field Y 2 X Rn and a vector v 2 Tp Rn , we define the
Euclidean directional derivative of Y in the direction v by the formula
ˇ ˇ
ˇ ˇ
rxv Y D v Y 1 @ ˇ C C v Y n @ ˇ ;
1
@x p ˇ @x ˇp
n

where for each i , v Y i is the result of applying the vector v to the function Y i :
@Y i .p/ i
n @Y .p/
v Y i D v1 C C v :
@x 1 @x n
x X Y by evaluating
If X is another vector field on Rn , we obtain a new vector field r
x Xp Y at each point:
r

xX Y D X Y 1 @ C C X Y n @ :
r (4.3)
@x 1 @x n
More generally, we can play the same game with curves and vector fields on a
submanifold of Rn . Suppose M Rn is an embedded submanifold, and consider
a smooth curve W I ! M . We want to think of a geodesic in M as a curve in M
that is “as straight as possible.” Of course, if M itself is curved, then 0 .t / (thought
of as a vector in Rn ) will probably have to vary, or else the curve will leave M .
But we can try to insist that the velocity not change any more than necessary for
the curve to stay in M . One way to do this is to compute the Euclidean acceleration
00 .t / as above, and then apply the tangential projection > W T.t/ Rn ! T.t/ M
(see Prop. 2.16). This yields a vector 00 .t /> D > . 00 .t // tangent to M , which we
call the tangential acceleration of . It is reasonable to say that is as straight as it
is possible for a curve in M to be if its tangential acceleration is zero.
Similarly, suppose Y is a smooth vector field on (an open subset of) M , and we
wish to ask how much Y is varying in M in the direction of a vector v 2 Tp M . Just
as in the case of velocity vectors, if we look at it from the point of view of Rn , the
vector field Y might be forced to vary just so that it can remain tangent to M . But
The Problem of Differentiating Vector Fields 87
Fig. 4.1: Cartesian coordinates Fig. 4.2: Polar coordinates
one plausible way to answer the question is to extend Y to a smooth vector field Yz
on an open subset of Rn , compute the Euclidean directional derivative of Yz in the
direction v, and then project orthogonally onto Tp M . Let us define the tangential
directional derivative of Y in the direction v to be

rv> Y D > r x v Yz : (4.4)
Problem 4-1 shows that the tangential directional derivative is well defined and pre-
served by rigid motions of Rn . However, at this point there is little reason to believe
that the tangential directional derivative is an intrinsic invariant of M (one that de-
pends only on the Riemannian geometry of M with its induced metric).
On an abstract Riemannian manifold, for which there is no “ambient Euclidean
space” in which to differentiate, this technique is not available. Thus we have to
find some way to make sense of the acceleration of a smooth curve in an abstract
manifold. Let W I ! M be such a curve. As you know from your study of smooth
manifold theory, at each time t 2 I , the velocity of is a well-defined vector 0 .t / 2
T.t/ M (see Appendix A), whose representation in any coordinates is given by (4.1),
just as in Euclidean space.
However, unlike velocity, acceleration has no such coordinate-independent in-
terpretation. For example, consider the parametrized circle in the plane given in
Cartesian coordinates by .t / D .x.t /; y.t // D .cos t; sin t / (Fig. 4.1). As a smooth
curve in R2 , it has an acceleration vector at time t given by
ˇ ˇ
00 00 @ ˇˇ 00 @ ˇˇ
.t / D x .t / C y .t /
@x ˇ.t/ @x ˇ.t/
ˇ ˇ
@ ˇˇ @ ˇˇ
D cos t sin t :
@x ˇ .t/ @x ˇ.t/
88 4 Connections
Fig. 4.3: 0 .t/ and 0 .t C h/ lie in different vector spaces
But in polar coordinates, the same curve is described by .r.t /; .t // D .1; t / (Fig.
4.2). In these coordinates, if we try to compute the acceleration vector by the anal-
ogous formula, we get
ˇ ˇ
@ˇ @ ˇˇ
00 .t / D r 00 .t / ˇˇ C 00 .t / D 0:
@r .t/ @ ˇ.t/
The problem is this: to define 00 .t / by differentiating 0 .t / with respect to t ,

we have to take a limit of a difference quotient involving the vectors 0 .t C h/ and
0 .t /; but these live in different vector spaces (T.tCh/ M and T.t/ M respectively),
so it does not make sense to subtract them (Fig. 4.3). The definition of acceleration
works in the special case of smooth curves in Rn expressed in standard coordinates
(or more generally, curves in any finite-dimensional vector space expressed in linear
coordinates) because each tangent space can be naturally identified with the vector
space itself. On a general smooth manifold, there is no such natural identification.
The velocity vector 0 .t / is an example of a vector field along a curve, a concept
for which we will give a rigorous definition presently. To interpret the acceleration
of a curve in a manifold, what we need is some coordinate-independent way to
differentiate vector fields along curves. To do so, we need a way to compare values
of the vector field at different points, or intuitively, to “connect” nearby tangent
spaces. This is where a connection comes in: it will be an additional piece of data
on a manifold, a rule for computing directional derivatives of vector fields.
Connections
It turns out to be easiest to define a connection first as a way of differentiating

sections of vector bundles. The definition is meant to capture the essential properties
of the Euclidean and tangential directional derivative operators (rx and r > ) that we
defined above. (We will verify later that those operators actually are connections;
Connections 89
see Examples 4.8 and 4.9.) After defining connections in this general setting, we
will adapt the definition to the case of vector fields along curves.
Let W E ! M be a smooth vector bundle over a smooth manifold M with
or without boundary, and let .E/ denote the space of smooth sections of E. A
connection in E is a map
r W X.M / .E/ ! .E/;
written .X; Y / 7! rX Y , satisfying the following properties:

(i) rX Y is linear over C 1 .M / in X : for f1 ; f2 2 C 1 .M / and X1 ; X2 2 X.M /,
rf1 X1 Cf2 X2 Y D f1 rX1 Y C f2 rX2 Y:
(ii) rX Y is linear over R in Y : for a1 ; a2 2 R and Y1 ; Y2 2 .E/,
rX .a1 Y1 C a2 Y2 / D a1 rX Y1 C a2 rX Y2 :
(iii) r satisfies the following product rule: for f 2 C 1 .M /,
rX .f Y / D f rX Y C .Xf /Y:
The symbol r is read “del” or “nabla,” and rX Y is called the covariant derivative
of Y in the direction X . (This use of the word “covariant” has nothing to do with
covariant functors in category theory. It is related, albeit indirectly, to the use of
the word in the context of covariant and contravariant tensors, in that it reflects the
fact that the components of the covariant derivative have a transformation law that
“varies correctly” to give a well-defined meaning independent of coordinates. From
the modern coordinate-free point of view, “invariant derivative” would probably be
a better term.)
There is a variety of types of connections that are useful in different circum-
stances. The type of connection we have defined here is sometimes called a Koszul
connection to distinguish it from other types. Since we have no need to consider
other types of connections in this book, we refer to Koszul connections simply as
connections.
Although a connection is defined by its action on global sections, it follows from
the definitions that it is actually a local operator, as the next lemma shows.
Lemma 4.1 (Locality). Suppose r is a connection in a smooth vector bundle E !
M . For every X 2 X.M /, Y 2 .E/, and p 2 M , the covariant derivative rX Y jp
depends only on the values of X and Y in an arbitrarily small neighborhood of
p. More precisely, if X D Xz and Y D Yz on a neighborhood of p, then rX Y jp D
rXz Yz jp .
Proof. First consider Y . Replacing Y by Y Yz shows that it suffices to prove

rX Y jp D 0 if Y vanishes on a neighborhood of p.
Thus suppose Y is a smooth section of E that is identically zero on a neighbor-
hood U of p. Choose a bump function ' 2 C 1 .M / with support in U such that
90 4 Connections
'.p/ D 1. The hypothesis that Y vanishes on U implies that 'Y 0 on all of M ,

so for every X 2 X.M /, we have rX .'Y / D rX .0 'Y / D 0rX .'Y / D 0. Thus the
product rule gives
0 D rX .'Y / D .X'/Y C '.rX Y /: (4.5)
Now Y 0 on the support of ', so the first term on the right is identically zero.
Evaluating (4.5) at p shows that rX Y jp D 0. The argument for X is similar but
easier.
I Exercise 4.2. Complete the proof of Lemma 4.1 by showing that rX Y and rXz Y
z on a neighborhood of p.
agree at p if X D X
Proposition 4.3 (Restriction of a Connection). Suppose r is a connection in a

smooth vector bundle E ! M . For every open subset U M , there is a unique
connection r U on the restricted bundle EjU that satisfies the following relation for
every X 2 X.M / and Y 2 .E/:
ˇ
U
r.Xj U/
.Y jU / D .rX Y / ˇU : (4.6)
Proof. First we prove uniqueness. Suppose r U is any such connection and X 2

X.U / and Y 2 .EjU / are arbitrary. Given p 2 U , we can use a bump function to
construct a smooth vector field Xz 2 X.M / and a smooth section Yz 2 .E/ such that
z U agrees with X and Yz jU with Y on some neighborhood of p, and then Lemma
Xj
4.1 together with (4.6) implies
U ˇ
ˇ ˇ ˇ
rX Y p D rUz Yz jU ˇp D rXz Yz ˇp : (4.7)
XjU
Since the right-hand side is completely determined by r, this shows that r U is

uniquely defined if it exists.
To prove existence, given X 2 X.U / and Y 2 .EjU /, for every p 2 U we just
construct Xz and Yz as above, and define rX
U
Y jp by (4.7). This is independent of the
z z
choices of X and Y by Lemma 4.1, and it is smooth because the same formula holds
on some neighborhood of p. The fact that it satisfies the properties of a connection
is an easy exercise.
I Exercise 4.4. Complete the proof of the preceding proposition by showing that r U is
a connection.
In the situation of this proposition, we typically just refer to the restricted con-
nection as r instead of r U ; the proposition guarantees that there is no ambiguity in
doing so.
Lemma 4.1 tells us that we can compute the value of rX Y at p knowing only the
values of X and Y in a neighborhood of p. In fact, as the next proposition shows,
we need only know the value of X at p itself.
Proposition 4.5. Under the hypotheses of Lemma 4.1, rX Y jp depends only on the
values of Y in a neighborhood of p and the value of X at p.
Connections 91
Proof. The claim about Y was proved in Lemma 4.1. To prove the claim about X ,
it suffices by linearity to assume that Xp D 0 and show that rX Y jp D 0. Choose
a coordinate neighborhood U of p, and write X D X i @i in coordinates on U , with
X i .p/ D 0. Thanks to Proposition 4.3, it suffices to work with the restricted con-
nection on U , which we also denote by r. For every Y 2 .EjU /, we have
rX Y jp D rX i @i Y jp D X i .p/r@i Y jp D 0: t
u
Thanks to Propositions 4.3 and 4.5, we can make sense of the expression rv Y
when v is some element of Tp M and Y is a smooth local section of E defined only
on some neighborhood of p. To evaluate it, let X be a vector field on a neighbor-
hood of p whose value at p is v, and set rv Y D rX Y jp . Proposition 4.5 shows
that the result does not depend on the extension chosen. Henceforth, we will inter-
pret covariant derivatives of local sections of bundles in this way without further
comment.
Connections in the Tangent Bundle
For Riemannian or pseudo-Riemannian geometry, our primary concern is with con-

nections in the tangent bundle, so for the rest of the chapter we focus primarily on
that case. A connection in the tangent bundle is often called simply a connection on
M . (The terms affine connection and linear connection are also sometimes used
in this context, but there is little agreement on the precise definitions of these terms,
so we avoid them.)
Suppose M is a smooth manifold with or without boundary. By the definition we
just gave, a connection in TM is a map
r W X.M / X.M / ! X.M /
satisfying properties (i)–(iii) above. Although the definition of a connection resem-

bles the characterization of .1; 2/-tensor fields given by the tensor characterization
lemma (Lemma B.6), a connection in TM is not a tensor field because it is not linear
over C 1 .M / in its second argument, but instead satisfies the product rule.
For computations, we need to examine how a connection appears in terms of a
local frame. Let .Ei / be a smooth local frame for TM on an open subset U M .
For every choice of the indices i and j , we can expand the vector field rEi Ej in
terms of this same frame:
rEi Ej D ijk Ek : (4.8)
As i , j , and k range from 1 to n D dim M , this defines n3 smooth functions
ijk W U ! R, called the connection coefficients of r with respect to the given
frame. The following proposition shows that the connection is completely deter-
mined in U by its connection coefficients.
Proposition 4.6. Let M be a smooth manifold with or without boundary, and let
r be a connection in TM . Suppose .Ei / is a smooth local frame over an open sub-
92 4 Connections
˚
set U M , and let ijk be the connection coefficients of r with respect to this
frame. For smooth vector fields X; Y 2 X.U /, written in terms of the frame as
X D X i Ei , Y D Y j Ej , one has

rX Y D X Y k C X i Y j ijk Ek : (4.9)
Proof. Just use the defining properties of a connection and compute:

rX Y D rX Y j Ej

D X Y j Ej C Y j rX i Ei Ej

D X Y j Ej C X i Y j rEi Ej

D X Y j Ej C X i Y j ijk Ek :
Renaming the dummy index in the first term yields (4.9).

Once the connection coefficients (and thus the connection) have been determined
in some local frame, they can be determined in any other local frame on the same
open set by the result of the following proposition.
Proposition 4.7 (Transformation Law for Connection Coefficients). Let M be

a smooth manifold with or without boundary, and let r be a connection in TM .
Suppose we are given two smooth local frames .Ei / and Ezj for TM on an open

subset U M , related by Ezi D Aji Ej for some matrix of functions Aji . Let ijk
z k denote the connection coefficients of r with respect to these two frames.
and ij
Then k
zijk D A1 k Aq Arj qr
p
C A1 p Aqi Eq Apj : (4.10)
p i
Proof. Problem 4-3.

Observe that the first term above is exactly what the transformation law would
be if ijk were the components of a .1; 2/-tensor field; but the second term is of a

different character, because it involves derivatives of the transition matrix Apj .
Existence of Connections
So far, we have studied properties of connections but have not produced any, so you
might be wondering whether they are plentiful or rare. In fact, they are quite plenti-
ful, as we will show shortly. Let us begin with the simplest example.
Example 4.8 (The Euclidean Connection). In T Rn , define the Euclidean connec-

tion rx by formula (4.3). It is easy to check that this satisfies the required proper-
ties for a connection, and that its connection coefficients in the standard coordinate
frame are all zero. //
Here is a way to construct a large class of examples.

Connections 93
Example 4.9 (The Tangential Connection on a Submanifold of Rn ). Let M Rn

be an embedded submanifold. Define a connection r > on TM , called the tangential
connection, by setting ˇ
>
rX Y D > r x z Yz ˇ ;
X M
where > is the orthogonal projection onto TM (Prop. 2.16), r x is the Euclidean
z z
connection on R (Example 4.8), and X and Y are smooth extensions of X and
n
Y to an open set in Rn . (Such extensions exist by the result of Exercise A.23.)

Since the value ofr x z Yz at a point p 2 M depends only on Xzp D Xp , this just boils
X
> >
down to defining rX Y p to be equal to the tangential directional derivative rX p
Y
that we defined in (4.4) above. Problem 4-1 shows that this value is independent of
the choice of extension Yz , so r > is well defined. Smoothness is easily verified by
expressing rx z Yz in terms of an adapted orthonormal frame (see Prop. 2.14).
X
>
It is immediate from the definition that rX Y is linear over C 1 .M / in X and
>
over R in Y , so to show that r is a connection, only the product rule needs to be
checked. Let f 2 C 1 .M /, and let fz be an extension of f to a neighborhood of M
in Rn . Then fzYz is a smooth extension of f Y to a neighborhood of M , so
ˇ
>
rX x z fzYz ˇ
.f Y / D > r X M
ˇ ˇ
D > Xz fz Yz ˇM C > fzr
x z Yz ˇ
X M
>
D .Xf /Y C f rX Y:
Thus r > is a connection. //
In fact, there are many connections on Rn , or indeed on every smooth manifold

that admits a global frame (for example, every manifold covered by a single smooth
coordinate chart). The following lemma shows how to construct all of them explic-
itly.
Lemma 4.10. Suppose M is a smooth n-manifold with or without boundary, and

M admits a global frame .Ei /. Formula (4.9) gives a one-to-one correspondence ˚
between connections in TM and choices of n3 smooth real-valued functions ijk
on M .
˚
Proof. Every connection determines functions ijk by (4.8), and Proposition 4.6
˚
shows that those functions satisfy (4.9). On the other hand, given ijk , we can
define rX Y by (4.9); it is easy to see that the resulting expression is smooth if X and
Y are smooth, linear over R in Y , and linear over C 1 .M / in X . To prove that it is a
connection, only the product rule requires checking; this is a straightforward
computation left as an exercise.
I Exercise 4.11. Complete the proof of Lemma 4.10.
Proposition 4.12. The tangent bundle of every smooth manifold with or without
boundary admits a connection.
94 4 Connections
Proof. Let M be a smooth manifold with or without boundary, and cover M with
coordinate charts fU˛ g; the preceding lemma guarantees the existence of a connec-
tion r ˛ on each U˛ . Choose a partition of unity f'˛ g subordinate to fU˛ g. We would
like to patch the various r ˛ ’s together by the formula
X
˛
rX Y D '˛ rX Y: (4.11)
˛
Because the set of supports of the '˛ ’s is locally finite, the sum on the right-hand
side has only finitely many nonzero terms in a neighborhood of each point, so it
defines a smooth vector field on M . It is immediate from this definition that rX Y is
linear over R in Y and linear over C 1 .M / in X . We have to be a bit careful with the
product rule, though, since a linear combination of connections is not necessarily a
connection. (You can check, for example, that if r 0 and r 1 are connections, then
neither 2r 0 nor r 0 C r 1 satisfies the product rule.) By direct computation,
X
˛
rX .f Y / D '˛ rX .f Y /
˛
X
˛
D '˛ .Xf /Y C f rX Y
˛
X X
˛
D .Xf /Y '˛ C f '˛ rX Y
˛ ˛
D .Xf /Y C f rX Y: t
u
Although a connection is not a tensor field, the next proposition shows that the
difference between two connections is.
Proposition 4.13 (The Difference Tensor). Let M be a smooth manifold with or

without boundary. For any two connections r 0 and r 1 in TM , define a map
D W X.M / X.M / ! X.M / by
1 0
D.X; Y / D rX Y rX Y:
Then D is bilinear over C 1 .M /, and thus defines a .1; 2/-tensor field called the
difference tensor between r 0 and r 1 .
Proof. It is immediate from the definition that D is linear over C 1 .M / in its first
argument, because both r 0 and r 1 are. To show that it is linear over C 1 .M / in the
second argument, expand D.X; f Y / using the product rule, and note that the two
terms in which f is differentiated cancel each other.
Now that we know there is always one connection in TM , we can use the result
of the preceding proposition to say exactly how many there are.
Theorem 4.14. Let M be a smooth manifold with or without boundary, and let r 0
be any connection in TM . Then the set A.TM / of all connections in TM is equal
to the following affine space:
Covariant Derivatives of Tensor Fields 95
˚
A.TM / D r 0 C D W D 2 T .1;2/ TM ;

where D 2 T .1;2/ TM is interpreted as a map from X.M / X.M / to X.M / as
in Proposition B.1, and r 0 C D W X.M / X.M / ! X.M / is defined by
0 0
r C D X Y D rX Y C D.X; Y /:
Proof. Problem 4-4.
Covariant Derivatives of Tensor Fields
By definition, a connection in TM is a rule for computing covariant derivatives of

vector fields. We show in this section that every connection in TM automatically in-
duces connections in all tensor bundles over M , and thus gives us a way to compute
covariant derivatives of tensor fields of any type.
r be a connection in TM . Then r uniquely determines a connection in each tensor
bundle T .k;l/ TM , also denoted by r, such that the following four conditions are
satisfied.
(i) In T .1;0/ TM D TM , r agrees with the given connection.
(ii) In T .0;0/ TM D M R, r is given by ordinary differentiation of functions:
rX f D Xf:
(iii) r obeys the following product rule with respect to tensor products:
rX .F ˝ G/ D .rX F / ˝ G C F ˝ .rX G/ :
(iv) r commutes with all contractions: if “tr” denotes a trace on any pair of in-
dices, one covariant and one contravariant, then
rX .tr F / D tr .rX F / :
This connection also satisfies the following additional properties:

(a) r obeys the following product rule with respect to the natural pairing between
a covector field ! and a vector field Y :
rX h!; Y i D hrX !; Y i C h!; rX Y i :

.k;l/
(b) For all F 2 T TM , smooth 1-forms ! 1 ; : : : ; ! k , and smooth vector
fields Y1 ; : : : ; Yl ,
96 4 Connections

.rX F / ! 1 ; : : : ; ! k ; Y1 ; : : : ; Yl D X F ! 1 ; : : : ; ! k ; Y1 ; : : : ; Yl
X
k

F ! 1 ; : : : ; rX ! i ; : : : ; ! k ; Y1 ; : : : ; Yl
iD1 (4.12)
X
l

F ! 1 ; : : : ; ! k ; Y1 ; : : : ; rX Yj ; : : : ; Yl :
j D1
Proof. First we show that every family of connections on all tensor bundles satisfying
(i)–(iv) also satisfies (a) and (b). Suppose we are given such a family of connections,
all denoted by r. To prove (a), note that h!; Y i D tr.! ˝ Y /, as can be seen by
evaluating both sides in coordinates, where they both reduce to !i Y i . Therefore,
(i)–(iv) imply

rX h!; Y i D rX tr.! ˝ Y /

D tr rX .! ˝ Y /

D tr rX ! ˝ Y C ! ˝ rX Y
D hrX !; Y i C h!; rX Y i :
Then (b) is proved by induction using a similar computation applied to

F ! 1 ; : : : ; ! k ; Y1 ; : : : ; Yl D tr
„ ıƒ‚ tr F ˝ ! 1 ˝ ˝ ! k ˝ Y1 ˝ ˝ Yl ;
ı…
kCl
where each trace operator acts on an upper index of F and the lower index of the
corresponding 1-form, or a lower index of F and the upper index of the correspond-
ing vector field.
Next we address uniqueness. Assume again that r represents a family of connec-
tions satisfying (i)–(iv), and hence also (a) and (b). Observe that (ii) and (a) imply
that the covariant derivative of every 1-form ! can be computed by
.rX !/.Y / D X.!.Y // !.rX Y /: (4.13)
It follows that the connection on 1-forms is uniquely determined by the original

connection in TM . Similarly, (b) gives a formula that determines the covariant
derivative of every tensor field F in terms of covariant derivatives of vector fields
and 1-forms, so the connection in every tensor bundle is uniquely determined.
Now to prove existence, we first define covariant derivatives of 1-forms by (4.13),
and then we use (4.12) to define r on all other tensor bundles. The first thing that
needs to be checked is that the resulting expression is multilinear over C 1 .M / in
each ! i and Yj , and therefore defines a smooth tensor field. This is done by inserting
f ! i in place of ! i , or f Yj in place of Yj , and expanding the right-hand side, noting
that the two terms in which f is differentiated cancel each other out. Once we know
that rX F is a smooth tensor field, we need to check that it satisfies the defining
properties of a connection. Linearity over C 1 .M / in X and linearity over R in F
are both evident from (4.12) and (4.13), and the product rule in F follows easily
from the fact that differentiation of functions by X satisfies the product rule.
While (4.12) and (4.13) are useful for proving the existence and uniqueness of the
connections in tensor bundles, they are not very practical for computation, because
computing the value of rX F at a point requires extending all of its arguments to
vector fields and covector fields in an open set, and computing a great number of
derivatives. For computing the components of a covariant derivative in terms of a
local frame, the formulas in the following proposition are far more useful.
r be a connection in TM . Suppose .Ei / is a local frame for M , "j is its dual
˚
coframe, and ijk are the connection coefficients of r with respect to this frame.
Let X be a smooth vector field, and let X i Ei be its local expression in terms of this
frame.
(a) The covariant derivative of a 1-form ! D !i "i is given locally by

rX .!/ D X.!k / X j !i ji k "k :

(b) If F 2 T .k;l/ TM is a smooth mixed tensor field of any rank, expressed
locally as
i :::i
F D Fj11:::jkl Ei1 ˝ ˝ Eik ˝ "j1 ˝ ˝ "jl ;
then the covariant derivative of F is given locally by
Xk X
l
i :::i i :::p:::i i :::ik p
rX F D X Fj11:::jkl C X m Fj11:::jl k mp
is
X m Fj11:::p:::j
l mjs

sD1 sD1
Ei1 ˝ ˝ Eik ˝ "j1 ˝ ˝ "jl :
Proof. Problem 4-5.

Because the covariant derivative rX F of a tensor field (or, as a special case,
a vector field) is linear over C 1 .M / in X , the covariant derivatives of F in all
directions can be handily encoded in a single tensor field whose rank is one more
than the rank of F , as follows.
Proposition 4.17 (The Total Covariant Derivative). Let M be a smooth mani-

fold
with or without
boundary and let r be a connection in TM . For every F 2
T .k;l/ TM , the map
rF W 1 .M / 1 .M / X.M / X.M / ! C 1 .M /
„ ƒ‚ … „ ƒ‚ …
k copies lC1 copies
given by

.rF / ! 1 ; : : : ; ! k ; Y1 ; : : : ; Yl ; X D .rX F / ! 1 ; : : : ; ! k ; Y1 ; : : : ; Yl (4.14)
98 4 Connections
defines a smooth .k; l C 1/-tensor field on M called the total covariant derivative
of F.
Proof. This follows immediately from the tensor characterization lemma (Lemma
B.6): rX F is a tensor field, so it is multilinear over C 1 .M / in its k C l arguments;
and it is linear over C 1 .M / in X by definition of a connection.
When we write the components of a total covariant derivative in terms of a local
frame, it is standard practice to use a semicolon to separate indices resulting from
differentiation from the preceding indices. Thus, for example, if Y is a vector field
written in coordinates as Y D Y i Ei , the components of the .1; 1/-tensor field rY
are written Y i Ij , so that
rY D Y i Ij Ei ˝ "j ;
with
Y i Ij D Ej Y i C Y k ji k :
For a 1-form !, the formulas read
r! D !iIj "i ˝ "j ; with !iIj D Ej !i !k jki :
More generally, the next lemma gives a formula for the components of total co-
variant derivatives of arbitrary tensor fields.
Proposition 4.18. Let M be a smooth manifold with or without boundary and let r
be a connection in TM ; and let .Ei / be a smooth local frame for TM and fijk g
the corresponding connection coefficients. The components of the total covariant
derivative of a .k; l/-tensor field F with respect to this frame are given by
X
k X
l
i :::i i1 :::ik i1 :::p:::ik is i :::ik
Fj11:::jkl Im D Em Fj1 :::jl C Fj1 :::jl mp Fj11:::p:::j p :
l mjs
sD1 sD1
I Exercise 4.20. Suppose F is a smooth .k; l/-tensor field and G is a smooth .r; s/-
tensor field. Show that the components of the total covariant derivative of F ˝ G are given
by
i1 :::i p1 :::pr i :::i i :::i p :::p
r.F ˝ G/ j1 :::jkl q1 :::qs Im D Fj11:::jkl Im Gqp11:::q
:::pr
s
C Fj11:::jkl Gq11:::qsrIm :
[Remark: This formula is often written in the following way, more suggestive of the product
rule for ordinary derivatives:

i :::i i :::i i :::i p :::p
Fj11:::jkl Gqp11:::q
:::pr
s
D Fj11:::jkl Im Gqp11:::q
:::pr
s
C Fj11:::jkl Gq11:::qsrIm :
Im
Notice that this does not say that r.F ˝ G/ D .rF / ˝ G C F ˝ .rG/, because in the
first term on the right-hand side of this latter formula, the index resulting from differentia-
tion is not the last lower index.]
Second Covariant Derivatives
Having defined the tensor field rF for a .k; l/-tensor field F , we can in turn take its
total covariant derivative and obtain a .k; l C 2/-tensor field r 2 F D r.rF /. Given
vector fields X; Y 2 X.M /, let us introduce the notation rX;Y2
F for the .k; l/-tensor
field obtained by inserting X; Y in the last two slots of r 2 F :
2
rX;Y F .: : : / D r 2 F .: : : ; Y; X /:
Note the reversal of order of X and Y : this is necessitated by our convention that
the last index position in rF is the one resulting from differentiation, while it is
conventional to let rX;Y
2
stand for differentiating first in the Y direction, then in the
X direction. (For this reason, some authors adopt the convention that the new index
position introduced by differentiation is the first instead of the last. As usual, be sure
to check each author’s conventions when you read.)
It is important to be aware that rX;Y
2
F is not the same as rX .rY F /. The main
reason is that the former is linear over C 1 .M / in Y , while the latter is not. The
relationship between the two expressions is given in the following proposition.
Proposition 4.21. Let M be a smooth manifold with or without boundary and let r
be a connection in TM . For every smooth vector field or tensor field F ,
2
rX;Y F D rX .rY F / r.rX Y / F:
Proof. A covariant derivative rY F can be expressed as the trace of rF ˝ Y on its

last two indices:
rY F D tr.rF ˝ Y /;
as you can verify by noting that both expressions have the same component formula,
i :::i
Fj11:::jkl Im Y m . Similarly, rX;Y
2
F can be expressed as an iterated trace:
2

rX;Y F D tr tr.r 2 F ˝ X / ˝ Y :
(First trace the last index of r 2 F with that of X , and then trace the last remaining
free index—originally the second-to-last in r 2 F —with that of Y .)
Therefore, since rX commutes with contraction and satisfies the product rule
with respect to tensor products (Prop. 4.15), we have

rX .rY F / D rX tr.rF ˝ Y /

D tr rX .rF ˝ Y /

D tr rX .rF / ˝ Y C rF ˝ rX Y

D tr tr.r 2 F ˝ X / ˝ Y C tr rF ˝ rX Y
2
D rX;Y F C r.rX Y / F: t
u
100 4 Connections
Example 4.22 (The Covariant Hessian). Let u be a smooth function on M . Then

ru 2 T .0;1/ TM D 1 .M / is just the 1-form du, because both tensors have
the same action on vectors: ru.X / D rX u D Xu D du.X /. The 2-tensor r 2 u D
r.du/ is called the covariant Hessian of u. Proposition 4.21 shows that its action
on smooth vector fields X; Y can be computed by the following formula:
r 2 u.Y; X / D rX;Y
2
u D rX .rY u/ r.rX Y / u D Y.Xu/ .rY X /u:
In any local coordinates, it is
r 2 u D uIij dx i ˝ dx j ; with uIij D @j @i u jki @k u: //
Vector and Tensor Fields Along Curves
Now we can address the question that originally motivated the definition of connec-
tions: How can we make sense of the derivative of a vector field along a curve?
Let M be a smooth manifold with or without boundary. Given a smooth curve
W I ! M , a vector field along is a continuous map V W I ! TM such that
V .t / 2 T.t/ M for every t 2 I ; it is a smooth vector field along if it is smooth as
a map from I to TM . We let X. / denote the set of all smooth vector fields along
. It is a real vector space under pointwise vector addition and multiplication by
constants, and it is a module over C 1 .I / with multiplication defined pointwise:
.f X /.t / D f .t /X.t /:
The most obvious example of a vector field along a smooth curve is the curve’s
velocity: 0 .t / 2 T.t/ M for each t , and its coordinate expression (4.1) shows that
it is smooth. Here is another example: if is a curve in R2 , let N.t / D R 0 .t /,
where R is counterclockwise rotationby =2, so N.t / is normal to 0 .t /. In standard
coordinates, N.t / D P .t /; P 1 .t / , so N is a smooth vector field along .
2
A large supply of examples is provided by the following construction: suppose

W I ! M is a smooth curve and Vz is a smooth vector field on an open subset of
M containing the image of . Define V W I ! TM by setting V .t / D Vz.t/ for each
t 2 I . Since V is equal to the composition Vz ı , it is smooth. A smooth vector
field along is said to be extendible if there exists a smooth vector field Vz on
a neighborhood of the image of that is related to V in this way (Fig. 4.4). Not
every vector field along a curve need be extendible; for example, if .t1 / D .t2 /
but 0 .t1 / ¤ 0 .t2 / (Fig. 4.5), then 0 is not extendible. Even if is injective, its
velocity need not be extendible, as the next example shows.
Example 4.23. Consider the figure eight curve W .; / ! R2 defined by
.t / D .sin 2t; sin t /:

Vector and Tensor Fields Along Curves 101
Fig. 4.4: Extendible vector field Fig. 4.5: Nonextendible vector field
Fig. 4.6: The image of the figure eight curve of Example 4.23
Its image is a set that looks like a figure eight in the plane (Fig. 4.6). Problem 4-7
asks you to show that is an injective smooth immersion, but its velocity vector
field is not extendible. //
More generally, a tensor field along is a continuous

map from I to some
tensor bundle T .k;l/ TM such that .t / 2 T .k;l/ T.t/ M for each t 2 I . It is a
smooth tensor field along if it is smooth as a map from I to T .k;l/ TM , and it is
extendible if there is a smooth tensor field z on a neighborhood of .I / such that
D z ı .
Covariant Derivatives Along Curves
Here is the promised interpretation of a connection as a way to take derivatives of

vector fields along curves.
Theorem 4.24 (Covariant Derivative Along a Curve). Let M be a smooth mani-

fold with or without boundary and let r be a connection in TM . For each smooth
curve W I ! M , the connection determines a unique operator
Dt W X. / ! X. /;
102 4 Connections
called the covariant derivative along , satisfying the following properties:

(i) L INEARITY OVER R:
Dt .aV C bW / D aDt V C bDt W for a; b 2 R:
(ii) P RODUCT RULE :
Dt .f V / D f 0 V C f Dt V for f 2 C 1 .I /:
(iii) If V 2 X. / is extendible, then for every extension Vz of V ,
Dt V .t / D r 0 .t/ Vz :
There is an analogous operator on the space of smooth tensor fields of any type
along .
Proof. For simplicity, we prove the theorem for the case of vector fields along ;
the proof for arbitrary tensor fields is essentially identical except for notation.
First we show uniqueness. Suppose Dt is such an operator, and let t0 2 I be
arbitrary. An argument similar to that of Lemma 4.1 shows that the value of Dt V at
t0 depends only on the values of V in any interval .t0 "; t0 C "/ containing t0 . (If
t0 is an endpoint of I , extend a coordinate representation of to a slightly bigger
and then restrict back to I .)
open interval, prove the lemma there,
Choose smooth coordinates x i for M in a neighborhood of .t0 /, and write
ˇ
V .t / D V j .t /@j ˇ.t/
for t near t0 , where V 1 ; : : : ; V n are smooth real-valued functions defined on some

neighborhood of t0 in I . By the properties of Dt , since each @j is extendible,
ˇ ˇ
Dt V .t / D VP j .t /@j ˇ.t/ C V j .t /r 0 .t/ @j ˇ.t/
ˇ (4.15)
D VP k .t / C P i .t /V j .t /ijk ..t // @k ˇ.t/ :
This shows that such an operator is unique if it exists.

For existence, if .I / is contained in a single chart, we can define Dt V by (4.15);
the easy verification that it satisfies the requisite properties is left as an exercise. In
the general case, we can cover .I / with coordinate charts and define Dt V by this
formula in each chart, and uniqueness implies that the various definitions agree when-
ever two or more charts overlap.
(It is worth noting that in the physics literature, the covariant derivative along a
curve is sometimes called the absolute derivative.)
I Exercise 4.25. Complete the proof of Theorem 4.24 by showing that the operator Dt
defined in coordinates by (4.15) satisfies properties (i)–(iii).
Geodesics 103
Apart from its use in proving existence of the covariant derivative along a curve,
(4.15) also gives a practical formula for computing such covariant derivatives in
coordinates.
Now we can improve Proposition 4.5 by showing that rv Y actually depends only
on the values of Y along any curve through p whose velocity is v.
Proposition 4.26. Let M be a smooth manifold with or without boundary, let r be a

connection in TM , and let p 2 M and v 2 Tp M . Suppose Y and Yz are two smooth
vector fields that agree at points in the image of some smooth curve W I ! M such
that .t0 / D p and 0 .t0 / D v. Then rv Y D rv Yz .
Proof. We can define a smooth vector field Z along by Z.t / D Y.t/ D Yz.t/ .
Since both Y and Yz are extensions of Z, it follows from condition (iii) in Theorem
4.24 that both rv Y and rv Yz are equal to Dt Z.t0 /.
Geodesics
Armed with the notion of covariant differentiation along curves, we can now define
acceleration and geodesics.
Let M be a smooth manifold with or without boundary and let r be a connection
in TM . For every smooth curve W I ! M , we define the acceleration of to be the
vector field Dt 0 along . A smooth curve is called a geodesic (with respect i to
r) if its acceleration is zero: Dt 0 0. In terms
of smooth coordinates
x , if
we write the component functions of as .t / D x 1 .t /; : : : ; x n .t / , then it follows
from (4.15) that is a geodesic if and only if its component functions satisfy the
following geodesic equation:
xR k .t / C xP i .t /xP j .t /ijk .x.t // D 0; (4.16)
where we use x.t / as an abbreviation for the n-tuple of component functions

x 1 .t /; : : : ; x n .t / . This is a system of second-order ordinary differential equations
(ODEs) for the real-valued functions x 1 ; : : : ; x n . The next theorem uses ODE theory
to prove existence and uniqueness of geodesics with suitable initial conditions. (Be-
cause difficulties can arise when a geodesic starts on the boundary or later hits the
boundary, we state and prove this theorem only for manifolds without boundary.)
Theorem 4.27 (Existence and Uniqueness of Geodesics). Let M be a smooth

manifold and r a connection in TM . For every p 2 M , w 2 Tp M , and t0 2 R,
there exist an open interval I R containing t0 and a geodesic W I !M satisfy-
ing .t0 / D p and 0 .t0 / D w. Any two such geodesics agree on their common
domain.

Proof. Let x i be smooth coordinates
on someneighborhood U of p. A smooth
curve in U , written as .t / D x 1 .t /; : : : ; x n .t / , is a geodesic if and only if its
104 4 Connections
Fig. 4.7: Uniqueness of geodesics
component functions satisfy (4.16). The standard trick for proving existence and
uniqueness for such a second-order system is to introduce auxiliary variables v i D
xP i to convert it to the following equivalent first-order system in twice the number of
variables:
xP k .t / D v k .t /;
(4.17)
vP k .t / D v i .t /v j .t /ijk .x.t //:

Treating x 1 ; : : : ; x n ; v 1 ; : : : ; v n / as coordinates on U Rn , we can recognize (4.17)
as the equations for the flow of the vector field G 2 X.U Rn / given by
ˇ ˇ
k @ ˇ
ˇ i j k @ ˇˇ
G.x;v/ D v v v ij .x/ k ˇ : (4.18)
@x k ˇ.x;v/ @v .x;v/
By the fundamental theorem on flows (Thm. A.42), for each .p; w/ 2 U Rn and
t0 2 R, there exist an open interval I0 containing t0 and a unique smooth solution
W I0 ! U Rn to this system satisfying the initial i condition .t0 / D .p; w/. If
we write the component functions 1 of as .t/ D x .t /; v i
.t / , then we can easily
check that the curve .t / D x .t /; : : : ; x .t / in U satisfies the existence claim of
n
the theorem.
To prove the uniqueness claim, suppose ; z W I ! M are both geodesics defined
on some open interval with .t0 / D z.t0 / and 0 .t0 / D z0 .t0 /. In any local coordi-
nates around .t0 /, we can define smooth curves ; z W .t0 "; t0 C "/ ! U Rn
as above. These curves both satisfy the same initial value problem for the sys-
tem (4.17), so by the uniqueness of ODE solutions, they agree on .t0 "; t0 C "/
for some " > 0. Suppose for the sake of contradiction that .b/ ¤ z.b/ for some
b 2 I . First suppose b > t0 , and let ˇ be the infimum of numbers b 2 I such that
b > t0 and .b/ ¤ z.b/ (Fig. 4.7). Then ˇ 2 I , and by continuity, .ˇ/ D z.ˇ/ and
0 .ˇ/ D z0 .ˇ/. Applying local uniqueness in a neighborhood of ˇ, we conclude that
and z agree on a neighborhood of ˇ, which contradicts our choice of ˇ. Arguing
similarly to the left of t0 , we conclude that z on all of I .
A geodesic W I ! M is said to be maximal if it cannot be extended to a
geodesic on a larger interval, that is, if there does not exist a geodesic z W Iz ! M
defined on an interval Iz properly containing I and satisfying zjI D . A geodesic
segment is a geodesic whose domain is a compact interval.
Parallel Transport 105
Fig. 4.8: A parallel vector field along a curve
Corollary 4.28. Let M be a smooth manifold and let r be a connection in TM .

For each p 2 M and v 2 Tp M , there is a unique maximal geodesic W I ! M
with .0/ D p and 0 .0/ D v, defined on some open interval I containing 0.
Proof. Given p 2 M and v 2 Tp M , let I be the union of all open intervals con-
taining 0 on which there is a geodesic with the given initial conditions. By Theo-
rem 4.27, all such geodesics agree where they overlap, so they define a geodesic
W I ! M , which is obviously the unique maximal geodesic with the given initial
conditions.
I Exercise 4.29. Show that the maximal geodesics on Rn with respect to the Euclidean
connection (4.3) are exactly the constant curves and the straight lines with constant-speed
parametrizations.
The unique maximal geodesic with .0/ D p and 0 .0/ D v is often called
simply the geodesic with initial point p and initial velocity v, and is denoted by v .
(For simplicity, we do not specify the initial point p in the notation; it can implicitly
be recovered from v by p D .v/, where W TM ! M is the natural projection.)
Parallel Transport
Another construction involving covariant differentiation along curves that will be
useful later is called parallel transport. As we did with geodesics, we restrict atten-
tion here to manifolds without boundary.
Let M be a smooth manifold and let r be a connection in TM . A smooth vector
or tensor field V along a smooth curve is said to be parallel along (with respect
to r) if Dt V 0 (Fig. 4.8). Thus a geodesic can be characterized as a curve whose
velocity vector field is parallel along the curve.
106 4 Connections
I Exercise 4.30. Let W I ! Rn be a smooth curve, and let V be a smooth vector field
along . Show that V is parallel along with respect to the Euclidean connection if and
only if its component functions (with respect to the standard basis) are constants.
The fundamental fact about parallel vector and tensor fields along curves is that
every tangent vector or tensor at any point on a curve can be uniquely extended to a
parallel field along the entire curve. Before we prove this claim, let us examine what
the equation of parallelism looks like in coordinates. Given a smooth curve with a
local coordinate representation .t / D 1 .t /; : : : ; n .t / , formula (4.15) shows that
a vector field V is parallel along if and only if
VP k .t / D V j .t /P i .t /ijk ..t //; k D 1; : : : ; n; (4.19)
with analogous expressions based on Proposition 4.18 for tensor fields of other
types. In each case, this is a system of first-order linear ordinary differential equa-
tions for the unknown
coefficients of the vector or tensor field—in the vector case,
the functions V 1 .t /; : : : ; V n .t / . The usual ODE theorem guarantees the existence
and uniqueness of a solution for a short time, given any initial values at t D t0 ; but
since the equation is linear, we can actually show much more: there exists a unique
solution on the entire parameter interval.
Theorem 4.31 (Existence, Uniqueness, and Smoothness for Linear ODEs). Let
I R be an open interval, and for 1 j; k n, let Akj W I ! R be smooth func-

tions. For all t0 2 I and every initial vector c 1 ; : : : ; c n 2 Rn , the linear initial value
problem
VP k .t / D Akj .t /V j .t /;
(4.20)
V k .t0 / D c k ;
has a unique smooth solution on all of I , and the solution depends smoothly on
.t; c/ 2 I Rn .

Proof. First assume t0 D 0. Let x 0 ; x 1 ; : : : ; x n denote standard
coordinates
on the
manifold I Rn RnC1 , and consider the vector field Y 2 X I Rn defined by
@ @ @
Y D 0
C A1j x 0 x j 1 C C Anj x 0 x j n :
@x @x @x
1
If V .t / D V .t /; : : : ; V n .t / is a solution
to (4.20) with t0 D 0 defined on some
interval I0 I , then the curve .t / D t; V 1 .t /; : : : ; V n .t / is an integral curve of
Y defined on I0 satisfying the initial condition

.0/ D 0; c 1 ; : : : ; c n : (4.21)
1
Conversely, for each c ; : : : ; c n 2 Rn , there is an integral curve of Y defined
on some open interval I0 I containing 0 and satisfying (4.21).
If we write the
component functions of as .t / D 0 .t /; 1 .t /; : : : ; n .t/ , then P 0 .t / 1 and
0 .0/ D 0, so .t / D t for all t . It then follows that V .t / D 1 .t /; : : : ; n .t / solves
(4.20) with t0 D 0. Thus there is a one-to-one correspondence between solutions to

(4.20) and integral curves of Y satisfying (4.21).
The fundamental
theorem on flows of vector fields (Thm. A.42) guarantees that
for each c 1 ; : : : ; c n 2 Rn , there exists a maximal integral curve of Y defined
on some open interval containing 0 and satisfying the initial condition (4.21), and
the solutions depend smoothly on .t; c/. Therefore, there is a solution to (4.20) for
t in some maximal interval I0 I , and we need only show that I0 D I . Write
I D .a; b/ and I0 D .a0 ; b0 / (where a; b; a0 ; b0 can be finite or infinite), and assume
for the sake of contradiction that b0 < b.
Let us use the differential equation to estimate the derivative of jV .t /j2 :
d X
jV .t /j2 D 2 VP k .t /V k .t /
dt
k
X
D2 Akj .t /V j .t /V k .t /
j;k
D 2V .t /T A.t /V .t / 2jA.t /j jV .t /j2 :
Here j j denotes the Frobenius norm of a vector or matrix, obtained by summing

the squares of all the components and taking the square root. (It is just the Euclidean
norm of the components.) On the compact interval Œ0; b0
I , the functions Akj are
all bounded, so there is a constant M such that jA.t /j M there. It then follows
that
d 2M t d
e jV .t /j2 D e 2M t jV .t /j2 2M jV .t /j2 0;
dt dt
so the expression e 2M t jV .t /j2 is a nonincreasing function of t , and thus is bounded
for all t 2 Œ0; b0 / by its initial value jV .0/j2 . This implies jV .t /j2 e 2M t jV .0/j2 for
t 2 Œ0; b0 /, which in turn implies that the corresponding integral curve of Y stays in
the compact set Œ0; b0
BxR .0/ I Rn , where R D e M b0 jV .0/j. This contradicts
the escape lemma (Lemma A.43), and shows that b0 D b. The possibility that a < a0
can be ruled out by applying the same reasoning to the vector field Y .
Finally, the case of general t0 2 I can be reduced to the previous case by making
the substitutions V k .t / D Vz k .t t0 / and Akj .t / D Azkj .t t0 /.
Theorem 4.32 (Existence and Uniqueness of Parallel Transport). Suppose M
is a smooth manifold with or without boundary, and r is a connection in TM .
Given a smooth curve W I ! M , t0 2 I , and a vector v 2 T.t0 / M or tensor v 2
T .k;l/ .T.t/ M /, there exists a unique parallel vector or tensor field V along such
that V .t0 / D v.
Proof. As in the proof of Theorem 4.24, we carry out the proof for vector fields.
The case of tensor fields differs only in notation.
First suppose .I / is contained in a single coordinate chart. Then V is parallel
along if and only if its components satisfy the linear system of ODEs (4.19).
Theorem 4.31 guarantees the existence and uniqueness of a solution on all of I with
any initial condition V .t0 / D v.
108 4 Connections
Fig. 4.9: Existence and uniqueness of parallel transports
Now suppose .I / is not covered by a single chart. Let ˇ denote the supremum
of all b > t0 for which a unique parallel transport exists on Œt0 ; b
. (The argument
for t < t0 is similar.) We know that ˇ > t0 , since for b close enough to t0 , .Œt0 ; b
/
is contained in a single chart and the above argument applies. Then a unique par-
allel transport V exists on Œt0 ; ˇ/ (Fig. 4.9). If ˇ is equal to sup I , we are done. If
not, choose smooth coordinates on an open set containing .ˇ ı; ˇ C ı/ for some
positive ı. Then there exists a unique parallel vector field Vz on .ˇ ı; ˇ C ı/ sat-
isfying the initial condition Vz .ˇ ı=2/ D V .ˇ ı=2/. By uniqueness, V D Vz on
their common domain, and therefore Vz is a parallel extension of V past ˇ, which is
a contradiction.
The vector or tensor field whose existence and uniqueness are proved in Theorem
4.32 is called the parallel transport of v along . For each t0 ; t1 2 I , we define a map
Pt0 t1 W T.t0 / M ! T.t1 / M; (4.22)

called the parallel transport map, by setting Pt0 t1 .v/ D V .t1 /
for each v 2 T.t0 / M ,
where V is the parallel transport of v along . This map is linear, because the equa-
tion of parallelism is linear. It is in fact an isomorphism, because Pt1 t0 is an inverse
for it.
It is also useful to extend the parallel transport operation to curves that are merely
piecewise smooth. Given an admissible curve W Œa; b
! M , a map V W Œa; b
!
TM such that V .t / 2 T.t/ M for each t is called a piecewise smooth vector field
along if V is continuous and there is an admissible partition .a0 ; : : : ; ak / for
such that V is smooth on each subinterval Œai1 ; ai
. We will call any such partition
an admissible partition for V . A piecewise smooth vector field V along is said to
be parallel along if Dt V D 0 wherever V is smooth.
Corollary 4.33 (Parallel Transport Along Piecewise Smooth Curves). Suppose

M is a smooth manifold with or without boundary, and r is a connection in TM .
Given an admissible curve W Œa; b
! M and a vector v 2 T.t0 / M or tensor v 2
T .k;l/ .T.t/ M /, there exists a unique piecewise smooth parallel vector or tensor
field V along such that V .a/ D v, and V is smooth wherever is.
Proof. Let .a0 ; : : : ; ak / be an admissible partition for . First define V jŒa0 ;a1 to
be the parallel transport of v along the first smooth segment jŒa0 ;a1 ; then de-
fine V jŒa1 ;a2 to be the parallel transport of V .a1 / along the next smooth segment
jŒa1 ;a2 ; and continue by induction.
Here is an extremely useful tool for working with parallel transport. Given any
basis .b1 ; : : : ; bn / for T.t0 / M , we can parallel transport the vectors bi along ,
thus obtaining an n-tuple of parallel vector fields .E1 ; : : : ; En / along . Because
each parallel transport map is an isomorphism, the vectors .Ei .t // form a basis
for T.t/ M at each point .t /. Such an n-tuple of vector fields along is called a
parallel frame along . Every smooth (or piecewise smooth) vector field along can
be expressed in terms of such a frame as V .t / D V i .t /Ei .t /, and then the properties
of covariant derivatives along curves, together with the fact that the Ei ’s are parallel,
imply
Dt V .t / D VP i .t /Ei .t / (4.23)
wherever V and are smooth. This means that a vector field is parallel along if
and only if its component functions with respect to the frame .Ei / are constants.
The parallel transport map is the means by which a connection “connects” nearby
tangent spaces. The next theorem and its corollary show that parallel transport de-
termines covariant differentiation along curves, and thereby the connection itself.
Theorem 4.34 (Parallel Transport Determines Covariant Differentiation). Let
M be a smooth manifold with or without boundary, and let r be a connection in
TM . Suppose W I ! M is a smooth curve and V is a smooth vector field along .
For each t0 2 I ,
Pt1 t0 V .t1 / V .t0 /
Dt V .t0 / D lim : (4.24)
t1 !t0 t1 t0
Proof. Let .Ei / be a parallel frame along , and write V .t / D V i .t /Ei .t / for t 2 I .
On the one hand, (4.23) shows that Dt V .t0 / D VP i .t0 /Ei .t0 /.
On the other hand, for every fixed t1 2 I , the parallel transport of the vector V .t1 /
along is the constant-coefficient vector field W .t / D V i .t1 /Ei .t / along , so
Pt1 t0 V .t1 / D V i .t1 /Ei .t0 /. Inserting these formulas into (4.24) and taking the limit
as t1 ! t0 , we conclude that the right-hand side is also equal to VP i .t0 /Ei .t0 /.
Corollary 4.35 (Parallel Transport Determines the Connection). Let M be a
smooth manifold with or without boundary, and let r be a connection in TM . Sup-
pose X and Y are smooth vector fields on M . For every p 2 M ,
ˇ P Y.h/ Yp
rX Y ˇp D lim h0 ; (4.25)
h!0 h
110 4 Connections
where W I ! M is any smooth curve such that .0/ D p and 0 .0/ D Xp .

Proof. Given p 2 M and a smooth curve such that .0/ D p and 0 .0/ D Xp , let
V .t / denote the vector field along determined by Y , so V .t / D Y.t/ . By property
(iii) of Theorem 4.24, rX Y jp is equal to Dt V .0/, so the result follows from Theo-
rem 4.34.
A smooth vector or tensor field on M is said to be parallel (with respect to r) if
it is parallel along every smooth curve in M . For example, Exercise 4.30 shows that
every constant-coefficient vector field on Rn is parallel.
Proposition 4.36. Suppose M is a smooth manifold with or without boundary, r
is a connection in TM , and A is a smooth vector or tensor field on M . Then A is
parallel on M if and only if rA 0.
Although Theorem 4.32 showed that it is always possible to extend a vector at
a point to a parallel vector field along any given curve, it may not be possible in
general to extend it to a parallel vector field on an open subset of the manifold.
The impossibility of finding such extensions is intimately connected with the phe-
nomenon of curvature, which will occupy a major portion of our attention in the
second half of the book.
Pullback Connections
Like vector fields, connections in the tangent bundle cannot be either pushed for-
ward or pulled back by arbitrary smooth maps. However, there is a natural way to
pull back such connections by means of a diffeomorphism. In this section we define
this operation and enumerate some of its most important properties.
Suppose M and M are smooth manifolds and ' W M ! M is a diffeomorphism.
For a smooth vector field X 2 X.M /, recall that the pushforward of X is the
that satisfies d'p .Xp / D .' X /'.p/ for all p 2 M .
unique vector field ' X 2 X M
(See Lemma A.36.)
Lemma 4.37 (Pullback Connections). Suppose M and M are smooth manifolds
z is a connection in T M
with or without boundary. If r and ' W M ! M is a diffeo-
morphism, then the map ' rz W X.M / X.M / ! X.M / defined by

' r z Y D ' 1 r z ' X .' Y / (4.26)
X
is a connection in TM , called the pullback of r z by '.

Proof. It is immediate from the definition that ' rz Y is linear over R in Y . To
X
see that it is linear over C 1 .M / in X , let f 2 C 1 .M /, and let fz D f ı ' 1 , so
' .f X / D fz' X . Then
Problems 111

z
' r Y D ' 1 rzz .' Y /
fX f ' X

D ' 1 fzrz ' X .' Y /

D f 'rz Y:
X
Finally, to prove the product rule in Y , let f and fz be as above, and note that (A.7)

implies .' X / fz D .Xf / ı ' 1 . Thus

z .f Y / D ' 1 r
' r z ' X fz' Y
X

D ' 1 fzr z ' X .' Y / C .' X / fz ' Y

D f 'rz Y C .Xf /Y: t
u
X
The next proposition shows that various important concepts defined in terms of
connections—covariant derivatives along curves, parallel transport, and geodesics—
all behave as expected with respect to pullback connections.
Proposition 4.38 (Properties of Pullback Connections). Suppose M and M are

smooth manifolds with or without boundary, and ' W M ! M is a diffeomorphism.
z be a connection in T M
Let r z be the pullback connection in TM .
and let r D ' r
Suppose W I ! M is a smooth curve.
(a) ' takes covariant derivatives along curves to covariant derivatives along
curves: if V is a smooth vector field along , then
z t .d' ı V /;
d' ı Dt V D D
where Dt is covariant differentiation along with respect to r, and D z t is

z
covariant differentiation along ' ı with respect to r.
(b) ' takes geodesics to geodesics: if is a r-geodesic in M , then ' ı is a
z
r-geodesic in M .
(c) ' takes parallel transport to parallel transport: for every t0 ; t1 2 I ,
d'.t1 / ı Pt0 t1 D Pt'ı

0 t1
ı d'.t0 / :
Problems
4-1. Let M Rn be an embedded submanifold and Y 2 X.M /. For every point
p 2 M and vector v 2 Tp M , define rv> Y by (4.4).
(a) Show that rv> Y does not depend on the choice of extension Yz of Y .
[Hint: Use Prop. A.28.]
112 4 Connections
(b) Show that rv> Y is invariant under rigid motions of Rn , in the following

sense: if F 2 E.n/ and Mz D F .M /, then dFp rv> Y D r > .F Y /.
dFp .v/
(Used on pp. 87, 93.)

4-2. In your study of smooth manifolds, you have already seen another way of
taking “directional derivatives of vector fields,” the Lie derivative LX Y
(which is equal to the Lie bracket ŒX; Y
; see Prop. A.46). Suppose M is a
smooth manifold of positive dimension.
(a) Show that the map L W X.M / X.M / ! X.M / is not a connection.
(b) Show that there are smooth vector fields X and Y on R2 such that
X D Y D @1 along the x 1 -axis, but the Lie derivatives LX .@2 / and
LY .@2 / are not equal on the x 1 -axis.
4-3. Prove Proposition 4.7 (the transformation law for the connection coeffi-
cients).
4-4. Prove Theorem 4.14 (characterizing the space of connections).
4-5. Prove Proposition 4.16 (local formulas for covariant derivatives of tensor
fields).
4-6. Let M be a smooth manifold and let r be a connection in TM . Define a
map W X.M / X.M / ! X.M / by
.X; Y / D rX Y rY X ŒX; Y
:
(a) Show that is a .1; 2/-tensor field, called the torsion tensor of r .
(b) We say that r is symmetric if its torsion vanishes identically. Show that
r is symmetric if and only if its connection coefficients with respect
to every coordinate frame are symmetric: ijk D jki . [Warning: They
might not be symmetric with respect to other frames.]
(c) Show that r is symmetric if and only if the covariant Hessian r 2 u of
every smooth function u 2 C 1 .M / is a symmetric 2-tensor field. (See
Example 4.22.)
(d) Show that the Euclidean connection r x on Rn is symmetric.
(Used on pp. 113, 121, 123.)

4-7. Let W .; / ! R2 be the figure eight curve defined in Example 4.23.
Prove that is an injective smooth immersion, but its velocity vector field is
not extendible.
4-8. Suppose M is a smooth manifold (without boundary), I R is an interval
(bounded or not, with or without endpoints), and W I ! M is a smooth
curve.
(a) Show that for every t0 2 I such that 0 .t0 / ¤ 0, there is a connected
neighborhood J of t0 in I such that every smooth vector field along
jJ is extendible.
Problems 113
(b) Show that if I is an open interval or a compact interval and is

a smooth embedding, then every smooth vector field along is
extendible.
4-9. Let M be a smooth manifold, and let r 0 and r 1 be two connections on
TM .
(a) Show that r 0 and r 1 have the same torsion (Problem 4-6) if and only
if their difference tensor is symmetric, i.e., D.X; Y / D D.Y; X / for all X
and Y .
(b) Show that r 0 and r 1 determine the same geodesics if and only if their
difference tensor is antisymmetric, i.e., D.X; Y / D D.Y; X / for all X
and Y .
(Used on p. 145.)
4-10. Suppose M is a smooth manifold endowed with a connection, W I ! M is
a smooth curve, and Y 2 X. /. Prove that if Y is parallel along , then it is
parallel along every reparametrization of .
4-11. Suppose G is a Lie group.
(a) Show that there is a unique connection r in T G with the property that
every left-invariant vector field is parallel.
(b) Show that the torsion tensor of r (Problem 4-6) is zero if and only if
G is abelian.
4-12. Prove Proposition 4.36 (a vector or tensor field A is parallel if and only if
rA 0).
4-13. Prove Proposition 4.38 (properties of pullback connections).
4-14. Let M be a smooth n-manifold and r a connection in TM , let .Ei / be a
local frame on some open subset U M , and let "i be the dual coframe.
(a) Show that
there is a uniquely determined n n matrix of smooth 1-
forms !i j on U , called the connection 1-forms for this frame, such
that
rX Ei D !i j .X /Ej
for all X 2 X.U /.
(b) C ARTAN ’ S FIRST STRUCTURE EQUATION : Prove that these forms sat-
isfy the following equation, due to Élie Cartan:
d "j D "i ^ !i j C j ;
where 1 ; : : : ; n 2 2 .M / are the torsion 2-forms, defined in terms of

the torsion tensor (Problem 4-6) and the frame .Ei / by
.X; Y / D j .X; Y /Ej :
(Used on pp. 145, 222.)

Chapter 5
The Levi-Civita Connection
If we are to use geodesics and covariant derivatives as tools for studying Riemann-
ian geometry, it is evident that we need a way to single out a particular connection
on a Riemannian manifold that reflects the properties of the metric. In this chap-
ter, guided by the example of the tangential connection on a submanifold of Rn ,
we describe two properties that determine a unique connection on every Riemann-
ian manifold. The first property, compatibility with the metric, is easy to motivate
and understand. The second, symmetry, is a bit more mysterious; but it is motivated
by the fact that it is invariantly defined, and is always satisfied by the tangential
connection. It turns out that these two conditions are enough to determine a unique
connection associated with any Riemannian or pseudo-Riemannian metric, called the
Levi-Civita connection after the early twentieth-century Italian differential geometer
Tullio Levi-Civita.
After defining the Levi-Civita connection, we investigate the exponential map,
which conveniently encodes the collective behavior of geodesics and allows us to
study how they change as the initial point and initial velocity vary. Having estab-
lished the properties of this map, we introduce normal neighborhoods and normal
coordinates, which are essential computational and theoretical tools for studying
local geometric properties near a point. Then we introduce the analogous notion
for studying properties near a submanifold: tubular neighborhoods and Fermi
coordinates. Finally, we return to our three main model Riemannian manifolds and
determine their geodesics.
Except where noted otherwise, the results and proofs of this chapter do not use
positivity of the metric, so they apply equally well to Riemannian and pseudo-
Riemannian manifolds.
The Tangential Connection Revisited

We are eventually going to show that on each Riemannian manifold there is a natural
connection that is particularly well suited to computations in Riemannian geome-
116 5 The Levi-Civita Connection
try. Since we get most of our intuition about Riemannian manifolds from studying
submanifolds of Rn with the induced metric, let us start by examining that case.
Let M Rn be an embedded submanifold. As a guiding principle, consider the
idea mentioned at the beginning of Chapter 4: a geodesic in M should be “as straight
as possible.” A reasonable way to make this rigorous is to require that the geodesic
have no acceleration in directions tangent to the manifold, or in other words that its
acceleration vector have zero orthogonal projection onto TM .
The tangential connection defined in Example 4.9 is perfectly suited to this task,
because it computes covariant derivatives on M by taking ordinary derivatives in
Rn and projecting them orthogonally to TM .
It is easy to compute covariant derivatives along curves in M with respect to
the tangential connection. Suppose W I ! M is a smooth curve. Then can be
regarded as either a smooth curve in M or a smooth curve in Rn , and a smooth
vector field V along that takes its values in TM can be regarded as either a
vector field along in M or a vector field along in Rn . Let D x t V denote the
n
covariant derivative of V along (as a curve in R ) with respect to the Euclidean
connection r, x and let D > V denote its covariant derivative along (as a curve in
t
M ) with respect to the tangential connection r > . The next proposition shows that
the two covariant derivatives along have a simple relationship to each other.
Proposition 5.1. Let M Rn be an embedded submanifold, W I ! M a smooth

curve in M , and V a smooth vector field along that takes its values in TM . Then
for each t 2 I ,
Dt> V .t / D > D x t V .t / :
Proof. Let t0 2 I be arbitrary. By Proposition 2.14, on some neighborhood U of

.t0 / in Rn there is an adapted orthonormal frame for TM , that is, a local orthonor-
mal frame .E1 ; : : : ; En / for T Rn such that .E1 ; : : : ; Ek / restricts to an orthonormal
frame
for TM atpoints of M \ U (where k D dim M ). If " > 0 is small enough that
.t0 "; t0 C "/ U , then for t 2 .t0 "; t0 C "/ we can write
ˇ ˇ
V .t / D V 1 .t /E1 ˇ.t/ C C V k .t /Ek ˇ.t/ ;
for some smooth functions V 1 ; : : : ; V k W .t0 "; t0 C "/ ! R. Formula (4.15) yields
X
k
ˇ ˇ
> Dx t V .t / D > VP i .t /Ei ˇ.t/ C V i .t /r
x 0 .t/ Ei ˇ
.t/
iD1
k
X ˇ ˇ
D VP i .t /Ei ˇ.t/ C V i .t / > rx 0 .t/ Ei ˇ
.t/
iD1
k
X ˇ ˇ
D VP i .t /Ei ˇ.t/ C V i .t /r>0 .t/ Ei ˇ.t/
iD1
D Dt> V .t /: t
u
Connections on Abstract Riemannian Manifolds 117
Corollary 5.2. Suppose M Rn is an embedded submanifold. A smooth curve

W I ! M is a geodesic with respect to the tangential connection on M if and
only if its ordinary acceleration 00 .t / is orthogonal to T.t/ M for all t 2 I .
Proof. As noted in Example 4.8, the connection coefficients of the Euclidean con-
nection on Rn are all zero. Thus it follows from (4.15) that the Euclidean covariant
x t 0 .t / D 00 .t /. The corol-
derivative of 0 along is just its ordinary acceleration: D
lary then follows from Proposition 5.1. t
u
These
considerations
can be extended to pseudo-Riemannian manifolds as well.
Let Rr;s ; qx.r;s/ be the pseudo-Euclidean space of signature .r; s/. If M Rr;s
is an embedded Riemannian or pseudo-Riemannian submanifold, then for each
p 2 M , the tangent space Tp Rr;s decomposes as a direct sum Tp M ˚ Np M , where
Np M D .Tp M /? is the orthogonal complement of Tp M with respect to qx.r;s/ . We
let > W Tp Rr;s ! Tp M be the qx.r;s/ -orthogonal projection, and define the tangen-
tial connection r > on M by

rX>
Y D > r x z Yz ;
X
where Xz and Yz are smooth extensions of X and Y to a neighborhood of M , and

x is the ordinary Euclidean connection on Rr;s . This is a well-defined connection
r
on M by the same argument as in the Euclidean case, and the next proposition is
proved in exactly the same way as Corollary 5.2.
Proposition 5.3. Suppose M is an embedded Riemannian or pseudo-Riemannian

submanifold of the pseudo-Euclidean space Rr;s . A smooth curve W I ! M is a
geodesic with respect to r > if and only if 00 .t / is qx.r;s/ -orthogonal to T.t/ M for
all t 2 I .
Connections on Abstract Riemannian Manifolds

There is a celebrated (and hard) theorem of John Nash [Nas56] that says that every
Riemannian metric on a smooth manifold can be realized as the induced metric
of some embedding in a Euclidean space. That theorem was later generalized
independently by Robert Greene [Gre70] and Chris J. S. Clarke [Cla70] to pseudo-
Riemannian metrics. Thus, in a certain sense, we would lose no generality by study-
ing only submanifolds of Euclidean and pseudo-Euclidean spaces with their induced
metrics, for which the tangential connection would suffice. However, when we are
trying to understand intrinsic properties of a Riemannian manifold, an embedding
introduces a great deal of extraneous information, and in some cases actually makes
it harder to discern which geometric properties depend only on the metric. Our
task in this chapter is to distinguish some important properties of the tangential
connection that make sense for connections on an abstract Riemannian or pseudo-

Riemannian manifold, and to use them to single out a unique connection in the
abstract case.
Metric Connections
The Euclidean connection on Rn has one very nice property with respect to the
Euclidean metric: it satisfies the product rule
˝ ˛ ˝ ˛
x X hY; Zi D r
r xX Z ;
x X Y; Z C Y; r
as you can verify easily by computing in terms of the standard basis. (In this for-
mula, the left-hand side represents the covariant derivative of the real-valued func-
tion hY; Zi regarded as a .0; 0/-tensor field, which is really just X hY; Zi by virtue
of property (ii) of Prop. 4.15.) The Euclidean connection has the same property with
respect to the pseudo-Euclidean metric on Rr;s . It is almost immediate that the tan-
gential connection on a Riemannian or pseudo-Riemannian submanifold satisfies
the same product rule, if we now interpret all the vector fields as being tangent to M
and interpret the inner products as being taken with respect to the induced metric on
M (see Prop. 5.8 below).
This property makes sense on an abstract Riemannian or pseudo-Riemannian
manifold. Let g be a Riemannian or pseudo-Riemannian metric on a smooth mani-
fold M (with or without boundary). A connection r on TM is said to be compatible
with g, or to be a metric connection, if it satisfies the following product rule for all
X; Y; Z 2 X.M /:
rX hY; Zi D hrX Y; Zi C hY; rX Zi : (5.1)
The next proposition gives several alternative characterizations of compatibility
with a metric, any one of which could be used as the definition.
Proposition 5.5 (Characterizations of Metric Connections). Let .M; g/ be a Rie-

mannian or pseudo-Riemannian manifold (with or without boundary), and let r be
a connection on TM . The following conditions are equivalent:
(a) r is compatible with g: rX hY; Zi D hrX Y; Zi C hY; rX Zi.
(b) g is parallel with respect to r: rg 0.
(c) In terms of any smooth local frame .Ei /, the connection coefficients of r
satisfy
l l
ki glj C kj gil D Ek .gij /: (5.2)
(d) If V; W are smooth vector fields along any smooth curve , then
d
hV; W i D hDt V ; W i C hV; Dt W i : (5.3)
dt
(e) If V; W are parallel vector fields along a smooth curve in M , then hV; W i is
constant along .
Fig. 5.1: A parallel orthonormal frame
(f) Given any smooth curve in M , every parallel transport map along is a
linear isometry.
(g) Given any smooth curve in M , every orthonormal basis at a point of can
be extended to a parallel orthonormal frame along (Fig. 5.1).
Proof. First we prove (a) , (b). By (4.14) and (4.12), the total covariant derivative
of the symmetric 2-tensor g is given by
.rg/.Y; Z; X / D .rX g/.Y; Z/ D X.g.Y; Z// g.rX Y; Z/ g.Y; rX Z/:
This is zero for all X; Y; Z if and only if (5.1) is satisfied for all X; Y; Z.
To prove (b) , (c), note that Proposition 4.18 shows that the components of rg
in terms of a smooth local frame .Ei / are
l l
gij Ik D Ek .gij / ki glj kj gil :
These are all zero if and only if (5.2) is satisfied.

Next we prove (a) , (d). Assume (a), and let V; W be smooth vector fields along
a smooth curve W I ! M . Given t0 2 I , in a neighborhood of .t0 / we may choose
coordinates x i and write V D V i @i and W D W j @j for some smooth functions
V i ; W j W I ! R. Applying (5.1) to the extendible vector fields @i ; @j , we obtain
d d i j
hV; W i D V W h@i ; @j i
dt dt
D VP i W j C V i WP j h@i ; @j i C V i W j hr 0 .t/ @i ; @j i C h@i ; r 0 .t/ @j i
D hDt V; W i C hV; Dt W i;
which proves (d). Conversely, if (d) holds, then in particular it holds for extendible
vector fields along , and then (a) follows from part (iii) of Theorem 4.24.
Now we will prove (d) ) (e) ) (f) ) (g) ) (d). Assume first that (d) holds. If
V and W are parallel along , then (5.3) shows that hV; W i has zero derivative with
respect to t , so it is constant along .
Now assume (e). Let v0 ; w0 be arbitrary vectors in T.t0 / M , and let V; W

be their parallel transports along , so that V .t0 / D v0 , W .t0 / D w0 , Pt0 t1 v0 D
V .t1 /, and Pt0 t1 w0 D W .t1 /. Because hV; W i is constant along , it follows that
hPt0 t1 v0 ; Pt0 t1 w0 i D hV .t1 /; W .t1 /i D hV .t0 /; W .t0 /i D hv0 ; w0 i, so Pt0 t1 is a lin-
ear isometry.
Next, assuming (f), we suppose W I ! M is a smooth curve and .bi / is an
orthonormal basis for T.t0 / M , for some t0 2 I . We can extend each bi by parallel
transport to obtain a smooth parallel vector field Ei along , and the assumption
that parallel transport is a linear isometry guarantees that the resulting n-tuple .Ei /
is an orthonormal frame at all points of .
Finally, assume that (g) holds, and let .Ei / be a parallel orthonormal frame along
. Given smooth vector fields V and W along , we can express them in terms of
this frame as V D V i Ei and W D W j Ej . The fact that the frame is orthonormal
means that the metric coefficients gij D hEi ; Ej i are constants along (˙1 or 0),
and the fact that it is parallel means that Dt V D VP i Ei and Dt W D WP i Ei . Thus
both sides of (5.3) reduce to the following expression:

gij VP i W j C V i WP j : (5.4)
This proves (d). t

u
Corollary 5.6. Suppose .M; g/ is a Riemannian or pseudo-Riemannian manifold
with or without boundary, r is a metric connection on M , and W I ! M is a
smooth curve.
(a) j 0 .t /j is constant if and only if Dt 0 .t / is orthogonal to 0 .t / for all t 2 I .
(b) If is a geodesic, then j 0 .t /j is constant.
I Exercise 5.7. Prove the preceding corollary.
Proposition 5.8. If M is an embedded Riemannian or pseudo-Riemannian subman-

ifold of Rn or Rr;s , the tangential connection on M is compatible with the induced
Riemannian or pseudo-Riemannian metric.
Proof. We will show that r > satisfies (5.1). Suppose X; Y; Z 2 X.M /, and let
z Yz ; Z
X; z be smooth extensions of them to an open subset of Rn or Rr;s . At points of
M , we have
˝ ˛
rX>
hY; Zi D X hY; Zi D Xz Yz ; Z z
˝ ˛
Drx z Yz ; Z z
X
˝ ˛ ˝ ˛
D r x z Yz ; Z x zZ
z C Yz ; r z
X X
˝ ˛ ˝ ˛
D > r x z Yz ; Z
z C Yz ; > r x zZ
z
X X
˝ > ˛ ˝ >
˛
D rX Y; Z C Y; rX Z;
z and Yz are tangent to

where the next-to-last equality follows from the fact that Z
M. u
t
Symmetric Connections
It turns out that every abstract Riemannian or pseudo-Riemannian manifold admits

many different metric connections (see Problem 5-1), so requiring compatibility
with the metric is not sufficient to pin down a unique connection on such a manifold.
To do so, we turn to another key property of the tangential connection. Recall the
definition (4.3) of the Euclidean connection. The expression on the right-hand side
of that definition is reminiscent of part of the coordinate expression for the Lie
bracket:
@ @
ŒX; Y D X Y i i
Y Xi :
@x @x i
In fact, the two terms in the Lie bracket formula are exactly the coordinate expressions
for rx X Y and r x Y X . Therefore, the Euclidean connection satisfies the following
identity for all smooth vector fields X; Y :
xX Y r
r x Y X D ŒX; Y :
This expression has the virtue that it is coordinate-independent and makes sense
for every connection on the tangent bundle. We say that a connection r on the
tangent bundle of a smooth manifold M is symmetric if
rX Y rY X ŒX; Y for all X; Y 2 X.M /:
The symmetry condition can also be expressed in terms of the torsion tensor
of the connection, which was introduced in Problem 4-6; this is the smooth .1; 2/-
tensor field W X.M / X.M / ! X.M / defined by
.X; Y / D rX Y rY X ŒX; Y :
Thus a connection r is symmetric if and only if its torsion vanishes identically. It

follows from the result of Problem 4-6 that a connection is symmetric if and only
if its connection coefficients in every coordinate frame satisfy ijk D jki ; this is the
origin of the term “symmetric.”
Proposition 5.9. If M is an embedded (pseudo-)Riemannian submanifold of a

(pseudo-)Euclidean space, then the tangential connection on M is symmetric.
Proof. Let M be an embedded Riemannian or pseudo-Riemannian submanifold

of Rn , where Rn is endowed either with the Euclidean metric or with a pseudo-
Euclidean metric qx.r;s/ , r C s D n. Let X; Y 2 X.M /, and let X; z Yz be smooth ex-
tensions of them to an open subset of the ambient space. If W M ,! Rn represents the
inclusion map, it follows that X and Y are -related to Xz and Yz , respectively,
and
thus by the naturality of the Lie bracket (Prop. A.39), ŒX; Y is -related to Xz ; Yz .

In particular, X;z Yz is tangent to M , and its restriction to M is equal to ŒX; Y .
Therefore,
ˇ ˇ
>
rX Y rY> X D > r x z Yz ˇ r x z Xz ˇ
X M Y M
ˇ
D > X; z Yz ˇ
M
ˇ
z
D X; Y Mz ˇ
D ŒX; Y : t
u
The last two propositions show that if we wish to single out a connection on each
Riemannian or pseudo-Riemannian manifold in such a way that it matches the tan-
gential connection when the manifold is presented as an embedded submanifold of
Rn or Rr;s with the induced metric, then we must require at least that the connec-
tion be compatible with the metric and symmetric. It is a pleasant fact that these two
conditions are enough to determine a unique connection.
Theorem 5.10 (Fundamental Theorem of Riemannian Geometry). Let .M; g/

be a Riemannian or pseudo-Riemannian manifold (with or without boundary).
There exists a unique connection r on TM that is compatible with g and symmetric.
It is called the Levi-Civita connection of g (or also, when g is positive definite,
the Riemannian connection).
Proof. We prove uniqueness first, by deriving a formula for r. Suppose, therefore,

that r is such a connection, and let X; Y; Z 2 X.M /. Writing the compatibility
equation three times with X; Y; Z cyclically permuted, we obtain
X hY; Zi D hrX Y; Zi C hY; rX Zi;

Y hZ; X i D hrY Z; X i C hZ; rY X i;
ZhX; Y i D hrZ X; Y i C hX; rZ Y i:
Using the symmetry condition on the last term in each line, this can be rewritten as
X hY; Zi D hrX Y; Zi C hY; rZ X i C hY; ŒX; Zi;

Y hZ; X i D hrY Z; X i C hZ; rX Y i C hZ; ŒY; X i;
ZhX; Y i D hrZ X; Y i C hX; rY Zi C hX; ŒZ; Y i:
Adding the first two of these equations and subtracting the third, we obtain
X hY; Zi C Y hZ; X iZhX; Y i D

2hrX Y; Zi C hY; ŒX; Zi C hZ; ŒY; X i hX; ŒZ; Y i:
Finally, solving for hrX Y; Zi, we get

hrX Y; Zi D 12 X hY; Zi C Y hZ; X i ZhX; Y i

hY; ŒX; Zi hZ; ŒY; X i C hX; ŒZ; Y i : (5.5)
Now suppose r 1 and r 2 are two connections on TM that are symmetric and
compatible with g. Since the right-hand side of (5.5) does not depend on the con-
1 2
nection, it follows that hrX Y rX Y; Zi D 0 for all X; Y; Z. This can happen only
1 2
if rX Y D rX Y for all X and Y , so r 1 D r 2 .
To prove existence, we use (5.5), or rather a coordinate version of it. It suffices
to prove that such a connection exists in each coordinate chart, for then uniqueness
ensures that the
connections in different charts agree where they overlap.
Let U; x i be any smooth local coordinate chart. Applying (5.5) to the coordi-
nate vector fields, whose Lie brackets are zero, we obtain
˝ ˛ ˝ ˛ ˝ ˛
r@i @j ; @l D 12 @i @j ; @l C @j h@l ; @i i @l @i ; @j : (5.6)
Recall the definitions of the metric coefficients and the connection coefficients:
˝ ˛
gij D @i ; @j ; r@i @j D ijm @m :
Inserting these into (5.6) yields

ijm gml D 1
2
@i gj l C @j gil @l gij : (5.7)
Finally, multiplying both sides by the inverse matrix g kl and noting that gml g kl D
k
ım , we get

ijk D 12 g kl @i gj l C @j gil @l gij : (5.8)
This formula certainly defines a connection in each chart, and it is evident from
the formula that ijk D jki , so the connection is symmetric by Problem 4-6(b). Thus
only compatibility with the metric needs to be checked. Using (5.7) twice, we get
l l

ki glj C kj gil D 12 @k gij C @i gkj @j gki C 12 @k gj i C @j gki @i gkj
D @k gij :
By Proposition 5.5(c), this shows that r is compatible with g. t

u
A bonus of this proof is that it gives us explicit formulas that can be used for
computing the Levi-Civita connection in various circumstances.
Corollary 5.11 (Formulas for the Levi-Civita Connection). Let .M; g/ be a Rie-
mannian or pseudo-Riemannian manifold (with or without boundary), and let r be
its Levi-Civita connection.
(a) I N TERMS OF VECTOR FIELDS : If X; Y; Z are smooth vector fields on M , then

hrX Y; Zi D 12 X hY; Zi C Y hZ; X i ZhX; Y i

hY; ŒX; Zi hZ; ŒY; X i C hX; ŒZ; Y i : (5.9)
(This is known as Koszul’s formula.)

(b) I N COORDINATES : In any smooth coordinate chart for M , the coefficients of
the Levi-Civita connection are given by

ijk D 12 g kl @i gj l C @j gil @l gij : (5.10)
(c) I N A LOCAL FRAME : Let .Ei / be a smooth local frame on an open subset
U M , and let cijk W U ! R be the n3 smooth functions defined by
ŒEi ; Ej D cijk Ek : (5.11)
Then the coefficients of the Levi-Civita connection in this frame are

ijk D 12 g kl Ei gj l C Ej gil El gij gj m cil
m
glm cjmi C gim clj
m
: (5.12)
(d) I N A LOCAL ORTHONORMAL FRAME : If g is Riemannian, .Ei / is a smooth

local orthonormal frame, and the functions cijk are defined by (5.11), then
j
ijk D 1
2
cijk cik cji k : (5.13)
Proof. We derived (5.9) and (5.10) in the proof of Theorem 5.10. To prove (5.12),
apply formula (5.9) with X D Ei , Y D Ej , and Z D El , to obtain
ijq gql D hrEi Ej ; El i

m

D 12 Ei gj l C Ej gil El gij gj m cil glm cjmi C gim clj
m
:
Multiplying both sides by g kl and simplifying yields (5.12). Finally, under the
hypotheses of (d), we have gij D ıij , so (5.12) reduces to (5.13) after rearranging
and using the fact that cijk is antisymmetric in i; j . t
u
On every Riemannian or pseudo-Riemannian manifold, we will always use the
Levi-Civita connection from now on without further comment. Geodesics with respect
to this connection are called Riemannian (or pseudo-Riemannian) geodesics, or
simply “geodesics” as long as there is no risk of confusion. The connection coeffi-
cients ijk of the Levi-Civita connection in coordinates, given by (5.10), are called
the Christoffel symbols of g.
The next proposition shows that these connections are familiar ones in the case
of embedded submanifolds of Euclidean or pseudo-Euclidean spaces.
Proposition 5.12.
(a) The Levi-Civita connection on a (pseudo-)Euclidean space is equal to the
Euclidean connection.
(b) Suppose M is an embedded (pseudo-)Riemannian submanifold of a (pseudo-)
Euclidean space. Then the Levi-Civita connection on M is equal to the tangen-
tial connection r > .
Proof. We observed earlier in this chapter that the Euclidean connection is sym-
metric and compatible with both the Euclidean metric gx and the pseudo-Euclidean
metrics qx.r;s/ , which implies (a). Part (b) then follows from Propositions 5.8 and
5.9. t
u
An important consequence of the definition is that because Levi-Civita connec-

tions are defined in coordinate-independent terms, they behave well with respect to
isometries. Recall the definition of the pullback of a connection (see Lemma 4.37).
Proposition
5.13 (Naturality of the Levi-Civita Connection). Suppose .M; g/
and M
; gz are Riemannian or pseudo-Riemannian manifolds with or without
boundary, and let r denote the Levi-Civita connection of g and r z that of gz. If

z
' W M ! M is an isometry, then ' r D r.
Proof. By uniqueness of the Levi-Civita connection, it suffices to show that the

z is symmetric and compatible with g. The fact that ' is an
pullback connection ' r
isometry means that for any X; Y 2 X.M / and p 2 M ,
hYp ; Zp i D hd'p .Yp /; d'p .Zp /i D h.' Y /'.p/ ; .' Z/'.p/ i;
or in other words, hY; Zi D h' Y; ' Zi ı '. Therefore,
X hY; Zi D X.h' Y; ' Zi ı '/

D .' X /h' Y; ' Zi ı '
˝ ˛ ˝ ˛
D r z ' X .' Y /; ' Z C ' Y; rz ' X .' Z/ ı '
˝ ˛ ˝ ˛
D ' 1 r z ' X .' Y /; Z C Y; ' 1 r z .' Z/
' X
˝ ˛ ˝ ˛
z X Y; Z C Y; ' r
D .' r/ z Z;
X
which shows that the pullback connection is compatible with g. Symmetry is proved
as follows:

z Y 'r
' r z X D ' 1 r z ' X .' Y / r
z ' Y .' X /
X Y

D ' 1 ' X; ' Y
D ŒX; Y : t
u

Corollary 5.14 (Naturality of Geodesics). Suppose .M; g/ and M
; gz are Rie-
mannian or pseudo-Riemannian manifolds with or without boundary, and ' W M !
is a local isometry. If is a geodesic in M , then ' ı is a geodesic in M

M
.
Proof. This is an immediate consequence of Proposition 4.38, together with the fact
that being a geodesic is a local property. t
u
Like every connection on the tangent bundle, the Levi-Civita connection induces
connections on all tensor bundles.
Proposition 5.15. Suppose .M; g/ is a Riemannian or pseudo-Riemannian mani-

fold. The connection induced on each tensor bundle by the Levi-Civita connec-
tion is compatible with the induced inner product on tensors, in the sense that
X hF; Gi D hrX F; GiChF;
rX Gi
for every vector field X and every pair of smooth
tensor fields F; G 2 T .k;l/ TM .
Proof. Since every tensor field can be written as a sum of tensor products of vector
and/or covector fields, it suffices to consider the case in which F D ˛1 ˝ ˝ ˛kCl
and G D ˇ1 ˝ ˝ ˇkCl , where ˛i and ˇi are covariant or contravariant 1-tensor
fields, as appropriate. In this case, the formula follows from (2.15) by a routine
computation. t
u
Proposition 5.16. Let .M; g/ be an oriented Riemannian manifold. The Riemannian

volume form of g is parallel with respect to the Levi-Civita connection.
Proof. Let p 2 M and v 2 Tp M be arbitrary, and let W ."; "/ ! M be a smooth

curve satisfying .0/ D p and 0 .0/ D v. Let .E1 ; : : : ; En / be a parallel oriented
orthonormal frame along . Since d Vg .E1 ; : : : ; En / 1 and Dt Ei 0 along ,
formula (4.12) shows that rv .d Vg / D Dt .d Vg /jtD0 D 0. t
u
Proposition 5.17. The musical isomorphisms commute with the total covariant
derivative operator: if F is any smooth tensor field with a contravariant i th index
position, and [ represents the operation of lowering the i th index, then

r F [ D .rF /[ : (5.14)
Similarly, if G has a covariant i th position and ] denotes raising the i th index, then

r G ] D .rG/] : (5.15)
Proof. The discussion on page 27 shows that F [ D tr.F ˝ g/, where the trace is
taken on the i th and last indices of F ˝ g. Because g is parallel, for every vector
field X we have rX .F ˝ g/ D .rX F / ˝ g. Because rX commutes with traces,
therefore,

rX F [ D rX tr.F ˝ g/ D tr .rX F / ˝ g D .rX F /[ :
This shows that when X is inserted into the last index position on both sides of
(5.14), the results are equal. Since X is arbitrary, this proves (5.14).
Because the sharp and flat operators are inverses of each other when applied
to the same index position, (5.15) follows by substituting F D G ] into (5.14) and
applying ] to both sides. t
u
The Exponential Map
Throughout this section, we let .M; g/ be a Riemannian or pseudo-Riemannian

n-manifold, endowed with its Levi-Civita connection. Corollary 4.28 showed that
each initial point p 2 M and each initial velocity vector v 2 Tp M determine a
unique maximal geodesic v . To deepen our understanding of geodesics, we need to
study their collective behavior, and in particular, to address the following question:
How do geodesics change if we vary the initial point or the initial velocity? The
The Exponential Map 127
dependence of geodesics on the initial data is encoded in a map from the tangent
bundle into the manifold, called the exponential map, whose properties are funda-
mental to the further study of Riemannian geometry.
(It is worth noting that the existence of the exponential map and the basic prop-
erties expressed in Proposition 5.19 below hold for every connection in TM , not just
for the Levi-Civita connection. For simplicity, we restrict attention here to the latter
case, because that is all we need. We also restrict to manifolds without boundary, in
order to avoid complications with geodesics running into a boundary.)
The next lemma shows that geodesics with proportional initial velocities are
related in a simple way.
Lemma 5.18 (Rescaling Lemma). For every p 2 M , v 2 Tp M , and c; t 2 R,
cv .t / D v .ct /; (5.16)
whenever either side is defined.
Proof. If c D 0, then both sides of (5.16) are equal to p for all t 2 R, so we may
assume that c ¤ 0. It suffices to show that cv .t / exists and (5.16) holds whenever
the right-hand side is defined. (The same argument with the substitutions v D c 0 v 0 ,
t D c 0 t 0 , and c D 1=c 0 then implies that the conclusion holds when only the left-hand
side is known to be defined.)
Suppose the maximal domain of v is the open interval I R. For simplicity,
write D v , and define a new curve z W c 1 I ! M by z.t/ D .ct /, where c 1 I D
fc 1 t W t 2 I g. We will show that z is a geodesic with initial point p and initial
velocity cv; it then follows by uniqueness and maximality that it must be equal to
cv .
It is immediate from the definition that z.0/ D .0/ D p. Choose any smooth
local
1 coordinates on M and write the coordinate representation of as .t / D
.t /; : : : ; n .t / ; then the chain rule gives
d
zP i .t / D i .ct /
dt
D c P i .ct /:
In particular, it follows that z0 .0/ D c 0 .0/ D cv.

Now let Dt and D z t denote the covariant differentiation operators along and z,
respectively. Using the chain rule again in coordinates yields

Dz t z0 .t / D d zP k .t / C ijk z.t / zP i .t /zP j .t / @k
dt

D c 2 R k .ct / C c 2 ijk ..ct //P i .ct /P j .ct / @k
D c 2 Dt 0 .ct / D 0:
Thus z is a geodesic, so z D cv , as claimed. t

u
The assignment v 7! v defines a map from TM to the set of geodesics in M .

More importantly, by virtue of the rescaling lemma, it allows us to define a map
from (a subset of) the tangent bundle to M itself, which sends each line through the
origin in Tp M to a geodesic.
Define a subset E TM , the domain of the exponential map, by
E D fv 2 TM W v is defined on an interval containing Œ0; 1g;
and then define the exponential map exp W E ! M by
exp.v/ D v .1/:
For each p 2 M , the restricted exponential map at p, denoted by expp , is the

restriction of exp to the set Ep D E \ Tp M .
The exponential map of a Riemannian manifold should not be confused with
the exponential map of a Lie group. The two are closely related for bi-invariant
metrics (see Problem 5-8), but in general they need not be. To avoid confusion, we
always designate the exponential map of a Lie group G by expG , and reserve the
undecorated notation exp for the Riemannian exponential map.
The next proposition describes some essential features of the exponential map.
Recall that a subset of a vector space V is said to be star-shaped with respect to a
point x 2 S if for every y 2 S , the line segment from x to y is contained in S .
Proposition 5.19 (Properties of the Exponential Map). Let .M; g/ be a Riemann-
ian or pseudo-Riemannian manifold, and let exp W E ! M be its exponential map.
(a) E is an open subset of TM containing the image of the zero section, and each
set Ep Tp M is star-shaped with respect to 0.
(b) For each v 2 TM , the geodesic v is given by
v .t / D exp.t v/ (5.17)
for all t such that either side is defined.

(c) The exponential map is smooth.
(d) For each point p 2 M , the differential d expp 0 W T0 .Tp M / Š Tp M ! Tp M
is the identity map of Tp M , under the usual identification of T0 .Tp M / with
Tp M .
Proof. Write n D dim M . The rescaling lemma with t D 1 says precisely that
exp.cv/ D cv .1/ D v .c/ whenever either side is defined; this is (b). Moreover,
if v 2 Ep , then by definition v is defined at least on Œ0; 1. Thus for 0 t 1, the
rescaling lemma says that
expp .t v/ D tv .1/ D v .t /
is defined. This shows that Ep is star-shaped with respect to 0.

Next we will show that E is open and exp is smooth. To do so, we revisit the
proof of the existence and uniqueness theorem for geodesics (Theorem 4.27) and
reformulate it in a more invariant way. Let x i be any smooth local coordinates
The Exponential Map 129

on an open set U M , let W TM ! M be the projection, and let x i ; v i denote
the associated natural coordinates for 1 .U / TM (see p. 384). In terms of these
coordinates, formula (4.18) defines a smooth vector field G on 1 .U /. The integral
curves of G are the curves .t / D x 1 .t /; : : : ; x n .t /; v 1 .t /; : : : ; v n .t / that satisfy the
system of ODEs given by (4.17), which is equivalent to the geodesic equation under
the substitution v k D xP k , as we observed in the proof of Theorem 4.27. Stated some-
what more invariantly, every integral curve of G on 1 .U / projects to a geodesic
under W TM ! M (which in these coordinates
is just .x; v/ D x); conversely,
every geodesic .t / D x 1 .t /; : : : ; x n .t / in U lifts to an integral curve of G in
1 .U / by setting v i .t / D xP i .t /.
The importance of G stems from the fact that it actually defines a global vector
field on the total space of TM , called the geodesic vector field. We could verify this
by computing the transformation law for the components of G under a change of
coordinates and showing that they take the same form in every coordinate chart; but
fortunately there is a way to avoid that messy computation. The key observation, to
be proved below, is that G acts on a function f 2 C 1 .TM / by
ˇ
d ˇˇ
Gf .p; v/ D ˇ f v .t /; v0 .t / : (5.18)
dt tD0
(Here and whenever convenient, we use the notations .p; v/ and v interchangeably
for an element v 2 Tp M , depending on whether we wish to emphasize the point at
which v is tangent.) Since this formula is independent of coordinates, it shows that
the various definitions of G given by (4.18) in different coordinate systems agree.
To prove that G satisfies (5.18), we write the components of the geodesic v .t /
as x i .t / and those of its velocity as v i .t / D xP i .t /. Using the chain rule and the
geodesic equation in the form (4.17), we can write the right-hand side of (5.18) as
ˇ
@f @f ˇ
k
k
.x.t /; v.t //xP .t / C k
.x.t /; v.t //vP .t / ˇˇ
k
@x @v tD0
@f @f
D k .p; v/v k k .p; v/v i v j ijk .p/
@x @v
D Gf .p; v/:
The fundamental theorem on flows (Thm. A.42) shows that there exist an open
set D R TM containing f0g TM and a smooth map W D ! TM , such that
each curve .p;v/ .t / D .t; .p; v// is the unique maximal integral curve of G starting
at .p; v/, defined on an open interval containing 0.
Now suppose .p; v/ 2 E. This means that the geodesic v is defined at least on
the interval Œ0; 1, and therefore so is the integral curve of G starting at .p; v/ 2 TM .
Since .1; .p; v// 2 D, there is a neighborhood of .1; .p; v// in R TM on which the
flow of G is defined (Fig. 5.2). In particular, this means that there is a neighborhood of
.p; v/ on which the flow exists for t 2 Œ0; 1, and therefore on which the exponential
map is defined. This shows that E is open.
Fig. 5.2: E is open
Since geodesics are projections of integral curves of G, it follows that the expo-
nential map can be expressed as
expp .v/ D v .1/ D ı .1; .p; v//
wherever it is defined, and therefore expp .v/ is a smooth function of .p; v/.

To compute d expp 0 .v/ for an arbitrary vector v 2 Tp M , we just need to
choose a curve in Tp M starting at 0 whose initial velocity is v, and compute
the initial velocity of expp ı . A convenient curve is .t / D t v, which yields
ˇ ˇ ˇ
d ˇˇ d ˇˇ d ˇˇ
d expp 0 .v/ D .expp ı /.t / D expp .t v/ D v .t / D v:
dt ˇtD0 dt ˇtD0 dt ˇtD0

Thus d expp 0 is the identity map. t
u
Corollary 5.14 on the naturality of geodesics translates into the following impor-
tant property of the exponential map.
Proposition
5.20 (Naturality of the Exponential Map). Suppose .M; g/ and
M
; gz are Riemannian or pseudo-Riemannian manifolds and ' W M ! M
is a
local isometry. Then for every p 2 M , the following diagram commutes:
d'p- z
Ep E'.p/
expp exp'.p/
? ?
' -
M M;
Normal Neighborhoods and Normal Coordinates 131
where Ep Tp M and Ez'.p/ T'.p/ M

are the domains of the restricted exponen-
tial maps expp (with respect to g) and exp'.p/ (with respect to gz), respectively.
An important consequence of the naturality of the exponential map is the follow-

ing proposition, which says that local isometries of connected manifolds are com-
pletely determined by their values and differentials at a single point.

Proposition 5.22. Let .M; g/ and M
; gz be Riemannian or pseudo-Riemannian
manifolds, with M connected. Suppose '; W M ! M
are local isometries such
that for some point p 2 M , we have '.p/ D .p/ and d'p D d p . Then ' .
u
A Riemannian or pseudo-Riemannian manifold .M; g/ is said to be geodesically
complete if every maximal geodesic is defined for all t 2 R, or equivalently if the
domain of the exponential map is all of TM . It is easy to construct examples of
manifolds that are not geodesically complete; for example, in every proper open
subset of Rn with its Euclidean metric or with a pseudo-Euclidean metric, there are
geodesics that reach the boundary in finite time. Similarly, on Rn with the metric
.
1 / gV obtained from the sphere by stereographic projection, there are geodesics
that escape to infinity in finite time. Geodesically complete manifolds are the nat-
ural setting for global questions in Riemannian or pseudo-Riemannian geometry;
beginning with Chapter 6, most of our attention will be focused on them.
Normal Neighborhoods and Normal Coordinates
We continue to let .M; g/ be a Riemannian or pseudo-Riemannian manifold

of dimension n (without boundary). Recall that for every p 2 M , the restricted
exponential map expp maps the open subset Ep Tp M smoothly into M . Because

d expp 0 is invertible, the inverse function theorem guarantees that there exist a
neighborhood V of the origin in Tp M and a neighborhood U of p in M such
that expp W V ! U is a diffeomorphism. A neighborhood U of p 2 M that is the
diffeomorphic image under expp of a star-shaped neighborhood of 0 2 Tp M is
called a normal neighborhood of p.
Every orthonormal basis .bi / for Tp M determines a basis isomorphism B W Rn !
Tp M by B.x 1 ; : : : ; x n / D x i bi . If U D expp .V / is a normal neighborhood of p, we
can combine this isomorphism with the exponential map to get a smooth coordinate
map ' D B 1 ı .expp jV /1 W U ! Rn :
B 1
- Rn
Tp M
-
6 '
.expp jV /1
U:
Such coordinates are called (Riemannian or pseudo-Riemannian) normal co-

ordinates centered at p.
Proposition 5.23 (Uniqueness of Normal Coordinates). Let .M; g/ be a Riemann-

ian or pseudo-Riemannian n-manifold, p a point of M , and U a normal neigh-
borhood of p. For every normal coordinate chart on U centered at p, the coordinate
orthonormal basis .bi / for Tp M , there is
basis is orthonormal at p; and for every
a unique normal coordinate chart x i on U such that @i jp D bi for i D
1; : : : ; n. In
the Riemannian case, any two normal coordinate charts x i and xzj are related
by
xzj D Aji x i (5.19)
j
for some (constant) matrix Ai 2 O.n/.
Proof. Let ' be a normal coordinate chart on U centered at p, with coordinate func-
tions x i . By definition, this means that ' D B 1 ıexp1
p , where B W R n
! T p M is
the basis isomorphism
determined by some orthonormal
basis .b i / for T p M . Note
that d'p1 D d expp 0 ıdB0 D B because d expp 0 is the identity and B is linear.
Thus @i jp D d'p1 .@i j0 / D B.@i j0 / D bi , which shows that the coordinate basis is
orthonormal at p. Conversely, every orthonormal basis .bi / for Tp M yields a basis
isomorphism B and thus a normal coordinate chart ' D B 1 ıexp1 p , which satisfies
@i jp D bi by the computation above.
If 'z D Bz 1 ı exp1
p is another such chart, then
'z ı ' 1 D Bz 1 ı exp1 z 1 ı B;

p ı expp ıB D B
which is a linear isometry of Rn and therefore

has the form (5.19) in terms of stan-
dard coordinates on Rn . Since xzj and x i are the same coordinates if and only

if Aji is the identity matrix, this shows that the normal coordinate chart associated
with a given orthonormal basis is unique. t
u
Let .M; g/ be a Riemann-

Proposition 5.24 (Properties of Normal Coordinates).
ian or pseudo-Riemannian n-manifold, and let U; x i be any normal coordinate
chart centered at p 2 M .
(a) The coordinates of p are .0; : : : ; 0/.
(b) The components of the metric at p are gij D ıij if g is Riemannian, and gij D
˙ıij otherwise.
(c) For every v D v i @i jp 2 Tp M , the geodesic v starting at p with initial velocity
v is represented in normal coordinates by the line

v .t / D t v 1 ; : : : ; t v n ; (5.20)
as long as t is in some interval I containing 0 such that v .I / U .

(d) The Christoffel symbols in these coordinates vanish at p.
(e) All of the first partial derivatives of gij in these coordinates vanish at p.
Tubular Neighborhoods and Fermi Coordinates 133
Proof. Part (a) follows directly from the definition of normal coordinates, and parts
(b) and (c) follow from Propositions 5.23 and 5.19(b), respectively.
To prove (d), let v D v i @i jp 2 Tp M be arbitrary. The geodesic equation (4.16)
for v .t / D t v 1 ; : : : ; t v n simplifies to
ijk .t v/v i v j D 0:
Evaluating this expression at t D 0 shows that ijk .0/v i v j D 0 for every index k and
k
every vector v. In particular, with v D @a for some fixed a, this shows that aa D0
for each a and k (no summation). Substituting v D @a C @b and v D @b @a for any
k
fixed pair of indices a and b and subtracting, we conclude also that ab D 0 at p for
all a; b; k. Finally, (e) follows from (d) together with (5.2) in the case Ek D @k . u t
Because they are given by the simple formula (5.20), the geodesics starting at p
and lying in a normal neighborhood of p are called radial geodesics. (But be warned
that geodesics that do not pass through p do not in general have a simple form in
normal coordinates.)
Tubular Neighborhoods and Fermi Coordinates

The exponential map and normal coordinates give us a good understanding of the
behavior of geodesics starting a point. In this section, we generalize those construc-
tions to geodesics starting on any embedded submanifold. We restrict attention to
the Riemannian case, because we will be using the Riemannian distance function.
Suppose .M; g/ is a Riemannian manifold, P M is an embedded submanifold,
and W NP ! P is the normal bundle of P in M . Let E TM denote the domain
of the exponential map of M , and let EP D E \ NP . Let E W EP ! M denote
the restriction of exp (the exponential map of M ) to EP . We call E the normal
exponential map of P in M.
A normal neighborhood of P in M is an open subset U M that is the dif-
feomorphic image under E of an open subset V EP whose intersection with each
fiber Nx P is star-shaped with respect to 0. We will be primarily interested in normal
neighborhoods of the following type: a normal neighborhood of P in M is called a
tubular neighborhood if it is the diffeomorphic image under E of a subset V EP
of the form ˚
V D .x; v/ 2 NP W jvjg < ı.x/ ; (5.21)
for some positive continuous function ı W P ! R (Fig. 5.3). If U is the diffeomor-
phic image of such a set V for a constant function ı.x/ ", then it is called a
uniform tubular neighborhood of radius ", or an "-tubular neighborhood.
Theorem 5.25 (Tubular Neighborhood Theorem). Let .M; g/ be a Riemannian

manifold. Every embedded submanifold of M has a tubular neighborhood in M ,
and every compact submanifold has a uniform tubular neighborhood.
Fig. 5.3: A tubular neighborhood Fig. 5.4: Injectivity of E
Proof. Let P M be an embedded submanifold, and let P0 NP be the sub-

set f.x; 0/ W x 2 P g (the image of the zero section of NP ). We begin by showing
that the normal exponential map E is a local diffeomorphism on a neighborhood of
P0 . By the inverse function theorem, it suffices to show that the differential dE.x;0/
is bijective at each point .x; 0/ 2 P0 . The restriction of E to P0 is just the dif-
feomorphism P0 ! P followed by the embedding P ,! M , so dE.x;0/ maps the
subspace T.x;0/ P0 T.x;0/ NP isomorphically onto Tx P . On the other hand, on the
fiber Nx P , E agrees with the restricted exponential map expx , which is a diffeo-
morphism near 0, so dE.x;0/ maps T.x;0/ .Nx P / T.x;0/ NP isomorphically onto
Nx P . Since Tx M D Tx P ˚ Nx P , this shows that dE.x;0/ is surjective, and hence it
is bijective for dimensional reasons. Thus E is a diffeomorphism on a neighborhood
of .x; 0/ in NP , which we can take to be of the form
˚
Vı .x/ D .x 0 ; v 0 / 2 NP W dg .x; x 0 / < ı; jv 0 jg < ı (5.22)
for some ı > 0. (Here we are using the fact that P is embedded in M , so it has the
subspace topology.)
To complete the proof, we need to show that there is a set V EP of the form
(5.21) on which E is a diffeomorphism onto its image. For each point x 2 P , define
˚
.x/ D sup ı 1 W E is a diffeomorphism from Vı .x/ to its image : (5.23)
The argument in the preceding paragraph implies that .x/ is positive for each
x. Note that E is injective on the entire set V.x/ .x/, because any two points
.x1 ; v1 /; .x2 ; v2 / in this set are in Vı .x/ for some ı < .x/. Because it is an injective
local diffeomorphism, E is actually a diffeomorphism from V.x/ .x/ onto its image.
Next we show that the function W P ! R is continuous. For any x; x 0 2 P ,
if dg .x; x 0 / < .x/, then the triangle inequality shows that Vı .x 0 / is contained in
V.x/ .x/ for ı D .x/ dg .x; x 0 /, which implies that .x 0 / .x/ dg .x; x 0 /,
or
.x/ .x 0 / dg .x; x 0 /: (5.24)
Tubular Neighborhoods and Fermi Coordinates 135
If dg .x; x 0 / .x/, then (5.24) holds for trivial reasons. Reversing the roles of x
and x 0 yields an analogous inequality, which shows that j.x/ .x 0 /j dg .x; x 0 /,
so is continuous.
˚
Let V D .x; v/ 2 NP W jvjg < 12 .x/ , which is an open subset of NP contain-
ing P0 . We show that E is injective on V . Suppose .x; v/ and .x 0 ; v 0 / are points in
V such that E.x; v/ D E.x 0 ; v 0 / (Fig. 5.4). Assume without loss of generality that
.x 0 / .x/. Because expx .v/ D expx 0 .v 0 /, there is an admissible curve from x
to x 0 of length jvjg C jv 0 jg , and thus
dg .x; x 0 / jvjg C jv 0 jg < 12 .x/ C 12 .x 0 / .x/:
Therefore, both .x; v/ and .x 0 ; v 0 / are in V.x/ .x/. Since E is injective on this set,
this implies .x; v/ D .x 0 ; v 0 /.
The set U D E.V / is open in M because EjV is a local diffeomorphism and
thus an open map, and E W V ! U is a diffeomorphism. Therefore, U is a tubular
neighborhood of P .
Finally, if P is compact, then the continuous function 12 achieves a minimum
value " > 0 on P , so U contains a uniform tubular neighborhood of radius ". t
u
Fermi Coordinates
Now we will construct coordinates on a tubular neighborhood that are analogous

to Riemannian normal coordinates around a point. Let P be an embedded
p-dimensional submanifold of a Riemannian n-manifold .M; g/, and let U M
be a normal neighborhood of P , with U D E.V / for some appropriate open subset
V NP .
Let .W0 ; / be a smooth coordinate chart for P , and let .E1 ; : : : ; Enp / be a local
orthonormal frame for the normal bundle NP ; by shrinking W0 if necessary, we can
assume that the coordinates and the local frame are defined on the same open subset
W0 P . Let W 0 D .W0 / Rp , and let NP jW0 be the portion of the normal
bundle over W0 . The coordinate map and frame .Ej / yield a diffeomorphism
BW W 0 Rnp ! NP jW0 defined by

B x 1 ; : : : ; x p ; v 1 ; : : : ; v np D q; v 1 E1 jq C C v np Enp jq ;
where q D 1 .x 1 ; : : : ; x p /. Let V0 D V \ NP jW0 NP and U0 D E.V0 / M ,

1
and define a smooth coordinate map ' W U0 ! Rn by ' D B 1 ı EjV0 :
B 1
- W 0 Rnp Rn
NP jW0
-
1 6 '
EjV0
U0 :
The coordinate map can also be written

' W E q; v 1 E1 jq C C v np Enp jq 7! x 1 .q/; : : : ; x p .q/; v 1 ; : : : ; v np : (5.25)
Coordinates of this form are called Fermi coordinates, after the Italian physicist
Enrico Fermi (1901–1954), who first introduced them in the special case in which
P is the image of a geodesic in M . The generalization to arbitrary submanifolds
was first introduced by Alfred Gray [Gra82]. (See also [Gra04] for a detailed study
of the geometry of tubular neighborhoods.)
Here is the analogue of Proposition 5.24 for Fermi coordinates.
Proposition 5.26 (Properties of Fermi Coordinates). Let P be an embedded

p-dimensional submanifold of a Riemannian
n-manifold .M; g/, let U be a nor-
mal neighborhood of P in M , and let x 1 ; : : : ; x p ; v 1 ; : : : ; v np be Fermi coordi-
nates on an open subset U0 U . For convenience, we also write x pCj D v j for
j D 1; : : : ; n p.
(a) P \ U0 is the set of points where x pC1 D D x n D 0.
(b) At each point q 2 P \ U0 , the metric components satisfy the following:
(
0; 1 i p and p C 1 j n;
gij D gj i D
ıij ; p C 1 i; j n:
(c) For every q 2 P \ U0 and v D v 1 E1 jq C C v np Enp jq 2 Nq P , the

geodesic v starting at q with initial velocity v is the curve with coordinate
expression v .t / D x 1 .q/; : : : ; x p .q/; t v 1 ; : : : ; t v np .
(d) At each q 2 P \U0 , the Christoffel symbols in these coordinates satisfy ijk D0,
provided p C 1 i; j n.
(e) At each q 2P \U0 , the partial derivatives @i gj k .q/ vanish for p C1 i; j; k
n.

u
Geodesics of the Model Spaces
In this section we determine the geodesics of the three types of frame-homogeneous

Riemannian manifolds defined in Chapter 3. We could, of course, compute the
Christoffel symbols of these metrics in suitable coordinates, and try to find the
geodesics by solving the appropriate differential equations; but for these spaces,
much easier methods are available based on symmetry and other geometric consid-
erations.
Geodesics of the Model Spaces 137
Euclidean Space
On Rn with the Euclidean metric, Proposition 5.12 shows that the Levi-Civita
connection is the Euclidean connection. Therefore, as one would expect, constant-
coefficient vector fields are parallel, and the Euclidean geodesics are straight lines
with constant-speed parametrizations (Exercises 4.29 and 4.30). Every Euclidean
space is geodesically complete.
Spheres
Because the round metric on the sphere Sn .R/ is induced by the Euclidean metric on
RnC1 , it is easy to determine the geodesics on a sphere using Corollary 5.2. Define
a great circle on Sn .R/ to be any subset of the form Sn .R/ \ ˘ , where ˘ RnC1
is a 2-dimensional linear subspace.
Proposition 5.27. A nonconstant curve on Sn .R/ is a maximal geodesic if and only

if it is a periodic constant-speed curve whose image is a great circle. Thus every
sphere is geodesically complete.
Proof. Let p 2 Sn .R/ be arbitrary. Because f .x/ D jxj2 is a defining function for
Sn .R/, a vector v 2 Tp RnC1 is tangent to Sn .R/ if and only if dfp .v/ D 2hv; pi D
0, where we think of p as a vector by means of the usual identification of RnC1 with
Tp RnC1 . Thus Tp Sn .R/ is exactly the set of vectors orthogonal to p.
Suppose v is an arbitrary nonzero vector in Tp Sn .R/. Let a D jvj=R and vy D v=a
v j D R), and consider the smooth curve W R ! RnC1 given by
(so jy
.t / D .cos at /p C .sin at /y

v:
By direct computation, j.t /j2 D R2 , so .t / 2 Sn .R/ for all t . Moreover,
0 .t / D a.sin at /p C a.cos at /y
v;
00 .t / D a2 .cos at /p a2 .sin at /y
v:
Because 00 .t / is proportional to .t / (thinking of both as vectors in RnC1 ), it fol-

lows that 00 .t / is gx-orthogonal to T.t/ Sn .R/, so is a geodesic in Sn .R/ by Corol-
lary 5.2. Since .0/ D p and 0 .0/ D ay v D v, it follows that D v .
Each v is periodic of period 2=a, and has constant speed by Corollary 5.6 (or
by direct computation). The image of v is the great circle formed by the intersection
of Sn .R/ with the linear subspace spanned by fp; vyg, as you can check.
Conversely, suppose C is a great circle formed by intersecting Sn .R/ with a
2-dimensional subspace ˘ , and let fv; wg be an orthonormal basis for ˘ . Then C
is the image of the geodesic with initial point p D Rw and initial velocity v. t
u
Fig. 5.5: A great hyperbola
Fig. 5.6: Geodesics of Kn .R/ Fig. 5.7: Geodesics of Bn .R/
Hyperbolic Spaces
The geodesics of hyperbolic spaces can be determined by an analogous procedure

using the hyperboloid model.
Proposition 5.28. A nonconstant curve in a hyperbolic space is a maximal geodesic

if and only if it is a constant-speed embedding of R whose image is one of the
following:
(a) H YPERBOLOID MODEL : The intersection of Hn .R/ with a 2-dimensional lin-
ear subspace of Rn;1 , called a great hyperbola (Fig. 5.5).
(b) B ELTRAMI –K LEIN MODEL : The interior of a line segment whose endpoints
both lie on @Kn .R/ (Fig. 5.6).
Fig. 5.8: Geodesics of U n .R/
(c) BALL MODEL : The interior of a diameter of Bn .R/, or the intersection of

Bn .R/ with a Euclidean circle that intersects @Bn .R/ orthogonally (Fig. 5.7).
(d) H ALF - SPACE MODEL : The intersection of U n .R/ with one of the following: a
line parallel to the y-axis or a Euclidean circle with center on @U n .R/ (Fig.
5.8).
Every hyperbolic space is geodesically complete.
Proof. We begin with the hyperboloid model, for which the proof is formally quite
similar to what we just did for the sphere. Since the Riemannian connection on
Hn .R/ is equal to the tangential connection by Proposition 5.12, it follows from
Corollary 5.2 that a smooth curve W I ! Hn .R/ is a geodesic if and only if its
acceleration 00 .t / is everywhere qx-orthogonal to T.t/ Hn .R/ (where qx D qx.n;1/ is
the Minkowski metric).
Let p 2 Hn .R/ be arbitrary. Note that f .x/ D qx.x; x/ is a defining function
for Hn .R/, and (3.10) shows that the gradient of f at p is equal to 2p (where
we regard p as a vector in Tp Rn;1 as before). It follows that a vector v 2 Tp Rn;1
is tangent to Hn .R/ if and only if qx.p; v/ D 0. Let v 2 Tp Hn .R/ be an arbitrary
nonzero vector. Put a D jvjqx =R D qx.v; v/1=2 =R and vy D v=a, and define W R !
Rn;1 by
.t / D .cosh at /p C .sinh at /y
v:
Direct computation shows that takes its values in Hn .R/ and that its accel-
eration vector is everywhere proportional to .t /. Thus 00 .t / is qx-orthogonal to
T.t/ Hn .R/, so is a geodesic in Hn .R/ and therefore has constant speed. Because
it satisfies the initial conditions .0/ D p and 0 .0/ D v, it is equal to v . Note that
v is a smooth embedding of R into Hn .R/ whose image is the great hyperbola
formed by the intersection between Hn .R/ and the plane spanned by fp; vyg.
Conversely, suppose ˘ is any 2-dimensional linear subspace of Rn;1 that has
nontrivial intersection with Hn .R/. Choose p 2 ˘ \ Hn .R/, and let v be another
nonzero vector in ˘ that is qx-orthogonal to p, which implies v 2 Tp Hn .R/. Using the
computation above, we see that the image of the geodesic v is the great hyperbola
formed by the intersection of ˘ with Hn .R/.
Before considering the other three models, note that since maximal geodesics in
Hn .R/ are constant-speed embeddings of R, it follows from naturality that maximal
geodesics in each of the other models are also constant-speed embeddings of R.
Thus each model is geodesically complete, and to determine the geodesics in the
other models we need only determine their images.
Consider the Beltrami–Klein model. Recall the isometry c W Hn .R/ ! Kn .R/
given by c.; / D R= (see (3.11)). The image of a maximal geodesic in Hn .R/
is a great hyperbola, which is the set of points .; / 2 Hn .R/ that solve a system
of n 1 independent linear equations. Simple algebra shows that .; / satisfies a
linear equation ˛i i C ˇ D 0 if and only if w D c.; / D R= satisfies the affine
equation ˛i w i D ˇR. Thus c maps each great hyperbola onto the intersection of
Kn .R/ with an affine subspace of Rn , and since it is the image of a smooth curve,
it must be the intersection of Kn .R/ with a straight line.
Next consider the Poincaré ball model. First consider the 2-dimensional case,
and recall the inverse hyperbolic stereographic projection 1 W B2 .R/ ! H2 .R/
constructed in Chapter 3:

2R2 u R2 C juj2
1 .u/ D .; / D ; R :
R2 juj2 R2 juj2
In this case, a great hyperbola is the set of points on H2 .R/ that satisfy a single
linear equation ˛i i C ˇ D 0. In the special case ˇ D 0, this hyperbola is mapped
by to a straight line segment through the origin, as can easily be seen from the
geometric definition of . If ˇ ¤ 0, we can assume (after multiplying through by a
constant if necessary) that ˇ D 1, and write the linear equation as D ˛i i D ˛
(where the dot represents the Euclidean dot product between elements of R2 ). Under
1 , this pulls back to the equation
R2 C juj2 2R2 ˛ u
R 2 2
D 2
R juj R juj2
on the disk, which simplifies to
juj2 2R˛ u C R2 D 0:
Completing the square, we can write this as
ju R˛j2 D R2 .j˛j2 1/: (5.26)
If j˛j2 1, this locus is either empty or a point on @B2 .R/, so it contains no points
in B2 .R/. Since we are assuming that it is the image of a maximal geodesic, we must
therefore havepj˛j2 > 1. In that case, (5.26) is the equation of a circle with center R˛
and radius R j˛j2 1. At a point u0 where the circle intersects @B2 .R/, the three
points 0, u0 , and R˛ form a triangle with sides ju0 j D R, jR˛j, and ju0 R˛j (Fig.
5.9), which satisfy the Pythagorean identity by (5.26); therefore the circle meets
@B2 .R/ in a right angle.
Fig. 5.9: Geodesics are arcs of circles orthogonal to the boundary of H2 .R/
In the higher-dimensional case, a geodesic on Hn .R/ is determined by a 2-plane.

If the 2-plane contains the point .0; : : : ; 0; R/, then the corresponding geodesic on
Bn .R/ is a line through the origin as before. Otherwise, we can use an ortho-
gonal transformation in the . 1 ; : : : ; n / variables (which preserves gM R ) to move
this 2-plane so that it lies in the . 1 ; 2 ; / subspace, and then we are in the same
situation as in the 2-dimensional case.
Finally, consider the upper half-space model. The 2-dimensional case is easi-
est to analyze using complex notation. Recall the complex formula for the Cayley
transform W U 2 .R/ ! B2 .R/ given in Chapter 3:
z iR
.z/ D w D iR :
z C iR
Substituting this into equation (5.26) and writing w D u C iv and ˛ D a C i b in
place of u D .u1 ; u2 /, ˛ D .˛ 1 ; ˛ 2 /, we get
jz iRj2 z iR zN C iR
R2 2
iR2 ˛N C iR2 ˛ C R2 j˛j2 D R2 .j˛j2 1/:
jz C iRj z C iR zN iR
Multiplying through by .z C iR/.zN iR/=2R2 and simplifying yields
.1 b/jzj2 2aRx C .b C 1/R2 D 0:
This is the equation of a circle with center on the x-axis, unless b D 1, in which
case the condition j˛j2 > 1 forces a ¤ 0, and then it is a straight line x D constant.
The other class of geodesics on the ball, line segments through the origin, can be
handled similarly.
In the higher-dimensional case, suppose first that W R ! U n .R/ ˚ is a maximal
geodesic such that .0/ lies on the y-axis and 0 .0/ is in the span of @=@x 1 ; @=@y .
From the explicit formula (3.15) for , it˚follows that ı .0/ lies on the v-axis in
the ball, and . ı /0 .0/ is in the span of @=@u1 ; @=@v . The image of the geodesic
ı is either part of a line through the origin or an arc of a circle perpendicu-
lar to @Bn .R/, both of which are contained in the u1 ; v -plane. By the argument
in 1the preceding paragraph, it then follows that the image of is contained in the
x ; y -plane and is either a vertical half-line or a semicircle centered on the y D 0
hyperplane. For the general case, note that translations and orthogonal transforma-
tions in the x-variables preserve vertical half-lines and circles centered on the y D 0
3
hyperplane in U n .R/, and they also preserve the metric gM R . Given an arbitrary max-
n
imal geodesic W R ! U .R/, after applying an x-translation we may assume that
.0/ lies on the y-axis, and after an orthogonal
˚ transformation
in the x variables,
we may assume that 0 .0/ is in the span of @=@x 1 ; @=@y ; then the argument above
shows that the image of is either a vertical half-line or a semicircle centered on
the y D 0 hyperplane. t
u
Euclidean and Non-Euclidean Geometries
In two dimensions, our model spaces can be interpreted as models of classical

Euclidean and non-Euclidean plane geometries.
Euclidean Plane Geometry

Euclid’s axioms for plane and spatial geometry, written around 300 BCE, became a
model for axiomatic treatments of geometry, and indeed for all of mathematics. As
standards of rigor evolved, mathematicians revised and added to Euclid’s axioms
in various ways. One axiom system that meets modern standards of rigor was cre-
ated by David Hilbert [Hil71]. Here (in somewhat simplified form) are his axioms
for plane geometry. (See [Hil71, Gan73, Gre93] for more complete treatments of
Hilbert’s axioms, and see [LeeAG] for a different axiomatic approach based on the
real number system.)
Hilbert’s Axioms For Euclidean Plane Geometry. The terms point, line, lies on,
between, and congruent are primitive terms, and are thus left undefined. We make
the following definitions:
Given a line l and a point P , we say that l contains P if P lies on l.
A set of points is said to be collinear if there is a line that contains them all.
Given two distinct points A; B, the segment AB is the set consisting of A, B,
and all points C such that C is between A and B.
The notation AB Š A 0 B 0 means that AB is congruent to A0 B 0 .
!
Given two distinct points A; B, the ray AB is the set consisting of A, B, and
all points C such that either C is between A and B or B is between A and C .
! !
An interior point of the ray AB is a point that lies on AB and is not equal to A.
Given three noncollinear points A; O; B, the angle †AOB is the union of the
! !
rays OA and OB.
Euclidean and Non-Euclidean Geometries 143
The notation †ABC Š †A 0 B 0 C 0 means that †ABC is congruent to †A0 B 0 C 0 .

Given a line l and two points A; B that do not lie on l, we say that A and B are
on the same side of l if no point of AB lies on l.
Two lines are said to be parallel if there is no point that lies on both of them.
These terms are assumed to satisfy the following postulates, among others:
Incidence Postulates:
(a) For any two distinct points A; B, there exists a unique line that contains
both of them.
(b) There exist at least two points on each line, and there exist at least three
noncollinear points.
Order Postulates:
(a) If a point B lies between a point A and a point C , then A; B; C are three
distinct points of a line, and B also lies between C and A.
(b) Given two distinct points A and C , there always exists at least one point B
such that C lies between A and B.
(c) Given three distinct points on a line, no more than one of them lies between
the other two.
(d) Let A; B; C be three noncollinear points, and let l be a line that does not
contain any of them. If l contains a point of AB, then it also contains a
point of AC or BC .
Congruence Postulates:
(a) If A; B are two points on a line l, and A0 is a point on a line l 0 , then it is
always possible to find a point B 0 on a given ray of l 0 starting at A0 such
that AB Š A0 B 0 .
(b) If segments A0 B 0 and A00 B 00 are congruent to the same segment AB, then
A0 B 0 and A00 B 00 are congruent to each other.
(c) On a line l, let AB and BC be two segments that, except for B, have no
points in common. Furthermore, on the same or another line l 0 , let A0 B 0
and B 0 C 0 be two segments that, except for B 0 , have no points in common.
In that case, if AB Š A0 B 0 and BC Š B 0 C 0 , then AC Š A0 C 0 .
(d) Let †rs be an angle and l 0 a line, and let a definite side of l 0 be given. Let

!
r 0 be a ray on l 0 starting at a point O 0 . Then there exists one and only one

!
ray s 0 such that †r 0 s 0 Š †rs and at the same time all the interior points

!
of s 0 lie on the given side of l 0 .
(e) If for two triangles 4ABC and 4A0 B 0 C 0 the congruences AB Š A0 B 0 ,
AC Š A0 C 0 , and †BAC Š †B 0 A0 C 0 hold, then †ABC Š †A0 B 0 C 0 and
†ACB Š †A0 C 0 B 0 as well.
Euclidean Parallel Postulate: Given a line l and a point A that does not lie on
l, there exists a unique line that contains A and is parallel to l.
Given an axiomatic system such as this one, an interpretation of the system is
simply an assignment of a definition for each of the primitive terms. An interpre-
tation is called a model of the axiomatic system provided that each of the axioms
becomes a theorem when the primitive terms are given the assigned meanings.
We are all familiar with the Cartesian plane as an interpretation of Euclidean

plane geometry. Formally, in this interpretation, we make the following definitions:
A point is an element of R2 .
A line is the image of a maximal geodesic with respect to the Euclidean metric.
Given a point A and a line l, we say that A lies on l if A 2 l.
Given three distinct points A; B; C , we say that B is between A and C if B is
on the geodesic segment joining A to C .
Given two sets of points S and S 0 , we say that S is congruent to S 0 if there is
a Euclidean isometry ' W R2 ! R2 such that '.S / D S 0 .
With this interpretation, it will come as no surprise that Hilbert’s postulates are
all theorems; proving them is just a standard exercise in plane analytic geometry.
I Exercise 5.29. Verify that all of Hilbert’s axioms are theorems when the primitive
terms are given the interpretations listed above.
More interesting is the application of Riemannian geometry to non-Euclidean

geometry. Hilbert’s axioms can be easily modified to yield axioms for plane hyper-
bolic geometry, simply by replacing the Euclidean parallel postulate by the follow-
ing:
Hyperbolic Parallel Postulate: Given a line l and a point A that does not lie
on l, there exist at least two distinct lines that contain A and are parallel to l.
We obtain an interpretation of this new axiomatic system by giving definitions
to the primitive terms just as we did above, but now with R2 replaced by H2 and
gx replaced by any hyperbolic metric gM R . (The axioms we have listed here do not
distinguish among hyperbolic metrics of different radii.) In Problem 5-19, you will be
asked to prove that some of Hilbert’s axioms are theorems under this interpretation.
In addition to hyperbolic geometry, it is possible to construct another version of
non-Euclidean geometry, in which the Euclidean parallel postulate is replaced by
the following assertion:
Elliptic Parallel Postulate: No two lines are parallel.
Unfortunately, we cannot simply replace the Euclidean parallel postulate with
this one and leave the other axioms alone, because it already follows from Hilbert’s
other axioms that for every line l and every point A … l there exists at least one
line through A that is parallel to l (for a proof, see [Gre93], for example). Nonethe-
less, we already know of an interesting geometry that satisfies the elliptic parallel
postulate—the sphere S2 with the round metric g. V To construct a consistent
axiomatic system including the elliptic parallel postulate, some of the other axioms
need to be modified.
If we take the sphere as a guide, with images of maximal geodesics as lines,
then we can see already that the first incidence postulate needs to be abandoned,
because if A; B 2 S2 are antipodal points (meaning that B D A), then there are
infinitely many lines containing A and B. Any axiomatic system for which S2 ; gV
is a model is called double elliptic geometry, because every pair of distinct lines
intersects in exactly two points.
Problems 145
It is also possible to construct an elliptic geometry in which the incidence postu-

lates hold, as in the following example.
Example 5.30. The real projective plane RP 2 has a frame-homogeneous Riemann-

ian metric g that is locally isometric to a round metric on Sn (see Example 2.34 and
Problem 3-2). As Problem 5-20 shows, single elliptic geometry satisfies Hilbert’s
incidence postulates as well as the elliptic parallel postulate. This interpretation is
called single elliptic geometry. //
Problems
5-1. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold, and let r be

its Levi-Civita connection. Suppose r z is another connection on TM , and
D is the difference tensor between r and r z (Prop. 4.13). Let D [ denote the
[
covariant 3-tensor field defined by D .X; Y; Z/ D hD.X; Y /; Zi. Show that
z is compatible with g if and only if D [ is antisymmetric in its last two
r
arguments: D [ .X; Y; Z/ D D [ .X; Z; Y / for all X; Y; Z 2 X.M /. Conclude
that on every Riemannian or pseudo-Riemannian manifold of dimension
at least 2, the space of metric connections is an infinite-dimensional affine
space. (Used on p. 121.)
5-2. Let r be a connection on the tangent bundle of a Riemannian manifold
.M; g/. Show that r is compatible with g if and only if the connection
1-forms !i j (Problem 4-14) with respect to each local frame .Ei / satisfy
gj k !i k C gik !j k D dgij :
Show that
this
implies that with respect to every local orthonormal frame, the
matrix !i j is skew-symmetric.
5-3. Define a connection on R3 by setting (in standard coordinates)
3 1 2
12 D 23 D 31 D 1;
3 1 2
21 D 32 D 13 D 1;
with all other connection coefficients equal to zero. Show that this connec-
tion is compatible with the Euclidean metric and has the same geodesics as
the Euclidean connection, but is not symmetric. (See Problem 4-9.)
5-4. Let C be an embedded smooth curve in the half-plane H D f.r; z/ W r > 0g,
and let SC R3 be the surface of revolution determined by C as in Example
2.20. Let .t / D .a.t /; b.t // be a unit-speed local parametrization of C , and
let X be the parametrization of SC given by (2.11).
(a) Compute the Christoffel symbols of the induced metric on SC in .t; /
coordinates.
(b) Show that each meridian is the image of a geodesic on SC .

(c) Determine necessary and sufficient conditions for a latitude circle to be
the image of a geodesic.
5-5. Recall that a vector field Y defined on (an open subset of) a Riemannian
manifold is said to be parallel if rY 0.
(a) Let p 2 Rn and v 2 Tp Rn . Show that there is a unique parallel vector
field Y on Rn such that Yp D v.
(b) Let X.'; / D .sin ' cos ; sin ' sin ; cos '/ be the spherical coordinate
parametrization of an open subset U of the unit sphere S2 (see Example
2.20 and Problem 5-4), and let X D X .@ /, X' D X .@' / denote the
coordinate vector fields associated with this parametrization. Compute
rX .X' / and rX' .X' /, and conclude that X' is parallel along the
equator and along each meridian D 0 .
(c) Let p D .1; 0; 0/ 2 S2 . Show that there is no parallel vector field W on
any neighborhood of p in S2 such that Wp D X' jp .
2
2noneighborhood of p in S ; gV is isomet-
(d) Use (a) and (c) to show that
ric to an open subset of R ; gx .
(Used on p. 194.)

5-6. Suppose W M
; gz ! .M; g/ is a Riemannian submersion. If Z is any vec-
z denote its horizontal lift to M
tor field on M , we let Z
(see Prop. 2.25).
(a) Show that for every pair of vector fields X; Y 2 X.M /, we have
˝ ˛
z Yz D hX; Y i ı I
X;

A
z Yz H D ŒX; Y I
X;

z W is vertical if W 2 X M
X;
is vertical.
(b) Let rz and r denote the Levi-Civita connections of gz and g, respec-

tively. Show that for every pair of vector fields X; Y 2 X.M /, we have
A
z Xz Yz D rX Y C 1 Xz ; Yz V :
r 2
(5.27)
z be a horizontal lift and W a vertical vector field on M
[Hint: Let Z
,
˝ ˛ ˝ ˛
and compute r z Xz Yz ; Z
z and rz Xz Yz ; W using (5.9).]
(Used on p. 224.)
5-7. Suppose .M1 ; g1 / and .M2 ; g2 / are Riemannian manifolds.
(a) Prove that if M1 M2 is endowed with the product metric, then a curve
W I ! M1 M2 of the form .t / D .1 .t /; 2 .t // is a geodesic if and
only if i is a geodesic in .Mi ; gi / for i D 1; 2.
(b) Now suppose f W M1 ! RC is a strictly positive smooth function, and
M1 f M2 is the resulting warped product (see Example 2.24). Let
1 W I ! M1 be a smooth curve and q0 a point in M2 , and define
Problems 147
W I ! M1 f M2 by .t / D .1 .t /; q0 /. Prove that is a geodesic

with respect to the warped product metric if and only if 1 is a geodesic
with respect to g1 .
(Used on p. 316.)
5-8. Let G be a Lie group and g its Lie algebra. Suppose g is a bi-invariant
Riemannian metric on G, and h; i is the corresponding inner product on g
(see Prop. 3.12). Let ad W g ! gl.g/ denote the adjoint representation of g
(see Appendix C).
(a) Show that ad.X / is a skew-adjoint endomorphism of g for every X 2 g:
had.X /Y; Zi D hY; ad.X /Zi:

˝ ˛
[Hint: Take the derivative of Ad expG tX Y; Ad expG tX Z with
respect to t at t D 0, where expG is the Lie group exponential map of
G, and use the fact that Ad D ad.]
(b) Show that rX Y D 12 ŒX; Y whenever X and Y are left-invariant vector
fields on G.
(c) Show that the geodesics of g starting at the identity are exactly the one-
parameter subgroups. Conclude that under the canonical isomorphism
of g Š Te G described in Proposition C.3, the restricted Riemannian
exponential map at the identity coincides with the Lie group exponential
map expG W g ! G. (See Prop. C.7.)
(d) Let RC be the set of positive real numbers, regarded as a Lie group
under multiplication. Show that g D t 2 dt 2 is a bi-invariant metric on
RC , and the restricted Riemannian exponential map at 1 is given by
c@=@t 7! e c .
(Used on pp. 128, 224.)

5-9. Suppose .M; g/ is a Riemannian manifold and U; ' is a smooth coordinate
chart on a neighborhood of p 2 M such that '.p/ D 0 and '.U / is star-
shaped with respect to 0. Prove that this chart is a normal coordinate chart
for g if and only if gij .p/ D ıij and the following identity is satisfied on U :
x i x j ijk .x/ 0:
5-10. Prove Proposition 5.22 (a local isometry is determined by its value and dif-
ferential at one point).
5-11. Recall the groups E.n/, O.n C 1/, and OC .n; 1/ defined
in Chapter
n 3, which
n
act isometrically
n on the model Riemannian manifolds R ; x
g , S .R/; gV R ,
and H .R/; gM R , respectively.
(a) Show that

Iso Rn ; gx D E.n/;

Iso Sn .R/; gV R D O.n C 1/;

Iso Hn .R/; gM R D OC .n; 1/:
(b) Show that in each case, for each point p in Rn , Sn .R/, or Hn .R/, the
isotropy group at p is a subgroup isomorphic to O.n/.
(c) Strengthen the result above
by showing
that if .M;
g/ is one of the
Riemannian manifolds Rn ; gx , Sn .R/; gV R , or Hn .R/; gM R , U is
a connected open subset of M , and ' W U ! M is a local isometry,
then ' is the restriction to U of an element of Iso.M; g/.
(Used on pp. 57, 58, 67, 348, 349.)
5-12. Suppose M is a connected n-dimensional Riemannian manifold, and G is
a Lie group acting isometrically and effectively on M . Show that dim G
n.n C 1/=2. (Used on p. 261.)
5-13. Let .M; g/ be a Riemannian manifold.
(a) Show that the following formula holds for every smooth 1-form 2
1 .M /:
d.X; Y / D .rX /.Y / .rY /.X /:
(b) Generalize this to an arbitrary k-form 2 k .M / as follows:
d D .1/k .k C 1/ Alt.r/;
where Alt denotes the alternation operator defined in (B.9). [Hint: For
each p 2 M , do the computation in normal coordinates centered at p,
and note that both sides of the equation are well defined, independently
of the choice of coordinates.]
(Used on p. 209.)
5-14. Suppose .M; g/ is a Riemannian manifold, and let div and be the diver-
gence and Laplace operators defined on pages 32–33.
(a) Show that for every vector field X 2 X.M /, div X can be written in
terms of the total covariant derivative as div X D tr.rX /, and that if
X D X i Ei in terms of some local frame, then div X D X i Ii . [Hint:
Show that it suffices to prove the formulas at the origin in normal
coordinates.]
(b) Show that the Laplace operator acting on a smooth function u can be
expressed as
u D trg r 2 u ; (5.28)
and in terms of any local frame,
u D g ij uIij D uIi i : (5.29)
(Used on pp. 218, 256, 333.)

Problems 149
5-15. Suppose .M; g/ is a compact Riemannian n-manifold (without boundary)

and u 2 C 1 .M / is an eigenfunction of M , meaning that u D u for
some constant (see Prob. 2-24). Prove that
Z Z
ˇ 2 ˇ2
2
j grad uj d Vg n ˇr uˇ d Vg :
M M
[Hint: Consider the 2-tensor field r 2 u n1 .u/g, and use one of Green’s
identities (Prob. 2-23).] (Used on p. 223.)
5-16. By analogy with the formula div X D tr.rX / developed in Problem 5-14,
we can define a divergence operator on tensor fields of any rank on a Rie-
mannian manifold. If F is any smooth k-tensor field (covariant, contravari-
ant, or mixed), we define the divergence of F by
div F D trg .rF /;
where the trace is taken on the last two indices of the .k C 1/-tensor field
rF . (If F is purely contravariant, then trg can be replaced with tr, because
the next-to-last index of rF is already an upper index.) Extend the integra-
tion by parts formula of Problem 2-22 as follows: if F is a smooth covariant
k-tensor field and G is a smooth covariant .k C 1/-tensor field on a compact
smooth Riemannian manifold .M; g/ with boundary, then
Z Z Z
˝ ˛
hrF; Gi d Vg D F ˝ N [ ; G d Vgy hF; div Gi d Vg ;
M @M M
where gy is the induced metric on @M . This is often written more sugges-

tively as
Z
Fi1 :::ik Ij G i1 :::ik j d Vg
M
Z Z
D Fi1 :::ik G i1 :::ik j Nj d Vgy Fi1 :::ik G i1 :::ik j Ij d Vg :
@M M
5-17. Suppose .M; g/ is a Riemannian manifold and P M is an embedded sub-

manifold. Show that P has a tubular neighborhood that is diffeomorphic to
the total space of the normal bundle NP , by a diffeomorphism that sends
the zero section of NP to P . [Hint: First show that the function ı in (5.21)
can be chosen to be smooth.]
5-18. Prove Proposition 5.26 (properties of Fermi coordinates).
5-19. Use H2 with the metric gM to construct an interpretation of Hilbert’s axioms
with the hyperbolic parallel postulate substituted for the Euclidean one, and
prove that the incidence postulates, congruence postulate (e), and the hyper-
bolic parallel postulate are theorems in this geometry. (Used on p. 144.)
5-20. Show that single elliptic geometry (Example 5.30) satisfies Hilbert’s inci-
dence postulates and the elliptic parallel postulate if points are defined as
elements of RP 2 and lines are defined as images of maximal geodesics.
5-21. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold and p 2 M .
Show that for every orthonormal basis .b1 ; : : : ; bn / for Tp M; there is a
smooth orthonormal frame .Ei / on a neighborhood of p such that Ei jp D bi
and .rEi /p D 0 for each i .
5-22. A smooth vector field X on a Riemannian manifold is called a Killing vector
field if the Lie derivative of the metric with respect to X vanishes. By Propo-
sition B.10, this is equivalent to the requirement that the metric be invariant
under the flow of X . Prove that X is a Killing vector field if and only if
the covariant 2-tensor field .rX /[ is antisymmetric. [Hint: Use Prop. B.9.]
(Used on pp. 190, 315.)
5-23. Let .M; g/ be a connected Riemannian manifold and p 2 M . An admissible
loop based at p is an admissible curve W Œa; b ! M such that .a/ D
.b/ D p. For each such loop , let P denote the parallel
transport
operator

Pab W Tp M ! Tp M along , and let Hol.p/ GL Tp M denote the set of
all automorphisms of Tp M obtained in this way:
˚
Hol.p/ D P W is an admissible loop based at p :

(a) Show that Hol.p/ is a subgroup of O Tp M (the set of all linear
isometries of Tp M ), called the holonomy group at p.
(b) Let Hol0 .p/ Hol.p/ denote the subset obtained by restricting to
loops that are path-homotopic to the constant loop. Show that Hol0 .p/
is a normal subgroup of Hol.p/, called the restricted holonomy group
at p.
(c) Given p; q2 M , show that there is an isomorphism of GL Tp M with
GL Tq M that takes Hol.p/ to Hol.q/.
(d) Show that M is orientable if and only if Hol.p/ SO Tp M (the set
of linear isometries with determinant C1) for some p 2 M .
(e) Show that g is flat if and only if Hol0 .p/ is the trivial group for some
p 2 M.
Chapter 6
Geodesics and Distance
In this chapter, we explore the relationships among geodesics, lengths, and distances
on a Riemannian manifold. A primary goal is to show that all length-minimizing
curves are geodesics, and that all geodesics are locally length-minimizing. Later, we
prove the Hopf–Rinow theorem, which states that a connected Riemannian manifold
is geodesically complete if and only if it is complete as a metric space. At the end
of the chapter, we study distance functions (which express the distance to a point
or other subset) and show that they can be used to construct coordinates, called
semigeodesic coordinates, that put a metric in a particularly simple form.
Most of the results of this chapter do not apply to general pseudo-Riemannian
metrics, at least not without substantial modification. For this reason, we restrict
our focus here to the Riemannian case. (For a treatment of lengths of curves in the
pseudo-Riemannian setting, see [O’N83].) Also, the theory of minimizing curves
becomes considerably more complicated in the presence of a nonempty boundary;
thus, unless otherwise stated, throughout this chapter we assume that .M; g/ is a Rie-
mannian manifold without boundary. And because we will be using the Riemannian
distance function, we assume for most results that M is connected.
Geodesics and Minimizing Curves
Let .M; g/ be a Riemannian manifold. An admissible curve in M is said to be

a minimizing curve if Lg . / Lg z for every admissible curve z with the same
endpoints. When M is connected, it follows from the definition of the Riemannian
distance that is minimizing if and only if Lg . / is equal to the distance between
its endpoints.
Our first goal in this section is to show that all minimizing curves are geodesics.
To do so, we will think of the length function Lg as a functional on the set of all ad-
missible curves in M with fixed starting and ending points. (Real-valued functions
whose domains are themselves sets of functions are typically called functionals.)
Our project is to search for minima of this functional.
152 6 Geodesics and Distance
From calculus, we might expect that a necessary condition for a curve to be

minimizing would be that the “derivative” of Lg vanish at , in some sense. This
brings us to the brink of the subject known as the calculus of variations: the use of
calculus to identify and analyze extrema of functionals defined on spaces of func-
tions or maps. In its fully developed state, the calculus of variations allows one to
apply all the usual tools of multivariable calculus in the infinite-dimensional setting
of function spaces—tools such as directional derivatives, gradients, critical points,
local extrema, saddle points, and Hessians. For our purposes, however, we do not
need to formalize the theory of calculus in the infinite-dimensional setting. It suf-
fices to note that if is a minimizing curve, and fs W s 2 ."; "/g is a one-parameter
family of admissible curves with the same endpoints such that Lg .s / is a differ-
entiable function of s and 0 D , then by elementary calculus, the s-derivative of
Lg .s / must vanish at s D 0 because Lg .s / attains a minimum there.
Families of Curves
To make this rigorous, we introduce some more definitions. Let .M; g/ be a Rie-
mannian manifold.
Given intervals I; J R, a continuous map W J I ! M is called a one-
parameter family of curves. Such a family defines two collections of curves in M :
the main curves s .t / D .s; t / defined for t 2 I by holding s constant, and the
transverse curves .t/ .s/ D .s; t / defined for s 2 J by holding t constant.
If such a family is smooth (or at least continuously differentiable), we denote
the velocity vectors of the main and transverse curves by
@t .s; t / D .s /0 .t / 2 T .s;t/ M I @s .s; t / D .t/0 .s/ 2 T .s;t/ M:
Each of these is an example of a vector field along ; which is a continuous map

V W J I ! TM such that V .s; t / 2 T .s;t/ M for each .s; t /.
The families of curves that will interest us most in this chapter are of the fol-
lowing type. A one-parameter family is called an admissible family of curves
if (i) its domain is of the form J Œa; b for some open interval J ; (ii) there is a
partition .a0 ; : : : ; ak / of Œa; b such that is smooth on each rectangle of the form
J Œai1 ; ai ; and (iii) s .t / D .s; t / is an admissible curve for each s 2 J (Fig.
6.1). Every such partition is called an admissible partition for the family.
If W Œa; b ! M is a given admissible curve, a variation of is an admissible
family of curves W J Œa; b ! M such that J is an open interval containing 0
and 0 D . It is called a proper variation if in addition, all of the main curves have
the same starting and ending points: s .a/ D .a/ and s .b/ D .b/ for all s 2 J .
In the case of an admissible family, the transverse curves are smooth on J for
each t , but the main curves are in general only piecewise regular. Thus the velocity
vector fields @s and @t are smooth on each rectangle J Œai1 ; ai , but not
generally on the whole domain.
Geodesics and Minimizing Curves 153
Fig. 6.1: An admissible family
We can say a bit more about @s , though. If is an admissible family, a piece-

wise smooth vector field along is a (continuous) vector field along whose
restriction to each rectangle J Œai1 ; ai is smooth for some admissible parti-
tion .a0 ; : : : ; ak / for . In fact, @s is always such a vector field. To see that it
is continuous on the whole domain J Œa; b, note on the one hand that for each
i D 1; : : : ; k 1, the values of @s along the set J fai g depend only on the val-
ues of on that set, since the derivative is taken only with respect to the s vari-
able; on the other hand, @s is continuous (in fact smooth) on each subrectangle
J Œai1 ; ai and J Œai ; aiC1 , so the right-hand and left-hand limits at t D ai
must be equal. Therefore @s is always a piecewise smooth vector field along .
(However, @t is typically not continuous at t D ai .)
If is a variation of , the variation field of is the piecewise smooth vector
field V .t / D @s .0; t / along . We say that a vector field V along is proper if
V .a/ D 0 and V .b/ D 0; it follows easily from the definitions that the variation field
of every proper variation is itself proper.
Lemma 6.1. If is an admissible curve and V is a piecewise smooth vector field

along , then V is the variation field of some variation of . If V is proper, the
variation can be taken to be proper as well.
Proof. Suppose and V satisfy the hypotheses, and set .s; t / D exp.t/ .sV .t //
(Fig. 6.2). By compactness of Œa; b, there is some positive " such that is de-
fined on ."; "/ Œa; b. By composition, is smooth on ."; "/ Œai1 ; ai for
each subinterval Œai1 ; ai on which V is smooth, and it is continuous on its whole
domain. By the properties of the exponential map, the variation field of is V .
Moreover, if V .a/ D 0 and V .b/ D 0, the definition gives .s; a/ .a/ and
.s; b/ .b/, so is proper. t
u
If V is a piecewise smooth vector field along ; we can compute the covariant
derivative of V either along the main curves (at points where V is smooth) or along
the transverse curves; the resulting vector fields along are denoted by Dt V and
Ds V respectively.
A key ingredient in the proof that minimizing curves are geodesics is the sym-
metry of the Levi-Civita connection. It enters into our proofs in the form of the
Fig. 6.2: Every vector field along is a variation field
following lemma. (Although we state and use this lemma only for the Levi-Civita
connection, the proof shows that it is actually true for every symmetric connection in
TM .)
Lemma 6.2 (Symmetry Lemma). Let W J Œa; b ! M be an admissible family

of curves in a Riemannian manifold. On every rectangle J Œai1 ; ai where is
smooth,
Ds @t D Dt @s :
i
Proof. This is a local question, so we may compute in local coordinates
1 x around
a point .s0 ; t0 /. Writing the components of as .s; t / D x .s; t /; : : : ; x n .s; t / ,
we have
@x k @x k
@t D @k I @s D @k :
@t @s
Then, using the coordinate formula (4.15) for covariant derivatives along curves, we
obtain !
@2 x k @x i @x j k
Ds @t D C @k I
@s@t @t @s j i
!
@2 x k @x i @x j k
Dt @s D C @k :
@t @s @s @t j i
Reversing the roles of i and j in the second line above and using the symmetry
condition jki D ijk , we conclude that these two expressions are equal. t
u
Minimizing Curves Are Geodesics
We can now compute an expression for the derivative of the length functional along
a variation of a curve. Traditionally, the derivative of a functional on a space of maps
is called its first variation.
Theorem 6.3 (First Variation Formula). Let .M; g/ be a Riemannian manifold.

Suppose W Œa; b ! M is a unit-speed admissible curve, W J Œa; b ! M is a
Fig. 6.3: i 0 is the “jump” in 0 at ai
variation of , and V is its variation field. Then Lg .s / is a smooth function of s,

and
ˇ Z b X˝
k1
d ˇˇ ˝ 0
˛ ˛
ˇ L g . s / D V; D t dt V .ai /; i 0
ds sD0 a iD1
˝ ˛ ˝ ˛
C V .b/; 0 .b/ V .a/; 0 .a/ ; (6.1)
where .a0 ; : : : ; ak / is an admissible partition for V , and for each i D 1; : : : ; k 1,

i 0 D 0 .aiC / 0 .ai / is the “jump” in the velocity vector field 0 at ai (Fig. 6.3).
In particular, if is a proper variation, then
ˇ Z b X˝
k1
d ˇˇ ˝ 0
˛ ˛
ˇ L g . s / D V; D t dt V .ai /; i 0 : (6.2)
ds sD0 a iD1
Proof. On every rectangle J Œai1 ; ai where is smooth, since the integrand in

Lg .s / is smooth and the domain of integration is compact, we can differentiate
under the integral sign as many times as we wish. Because Lg .s / is a finite sum of
such integrals, it follows that it is a smooth function of s.
For brevity, let us introduce the notations
T .s; t / D @t .s; t /; S.s; t / D @s .s; t /:
Differentiating on the interval Œai1 ; ai yields

Z ai
d ˇ @
Lg s ˇŒa ;a D hT; T i1=2 dt
ds i 1 i
ai 1 @s
Z ai
1
D hT; T i1=2 2 hDs T ; T i dt (6.3)
ai 1 2
Z ai
1
D hDt S; T i dt;
ai 1 jT j
where we have used the symmetry lemma in the last line. Setting s D 0 and noting
that S.0; t / D V .t / and T .0; t / D 0 .t / (which has length 1), we get
ˇ Z
d ˇˇ ˇ ˝
ai ˛
Lg s ˇŒa D Dt V ; 0 dt
ds ˇ sD0
i 1 ;ai
a
Z ia1
i d ˝ ˛
D hV; 0 i V; Dt 0 dt
a dt
˝ i 1 ˛ ˝ C
˛
D V .ai /; 0 .ai / V .ai1 /; 0 .ai1 /
Z ai
˝ ˛
V; Dt 0 dt:
ai 1
(The second equality follows from (5.3), and the third from the fundamental theorem
of calculus.) Finally, summing over i , we obtain (6.1). t
u
Because every admissible curve has a unit-speed parametrization and length is
independent of parametrization, the requirement in the above proposition that be
of unit speed is not a real restriction, but rather just a computational convenience.
Theorem 6.4. In a Riemannian manifold, every minimizing curve is a geodesic

when it is given a unit-speed parametrization.
Proof. Suppose W Œa; b ! M is minimizing and of unit speed, and .a0 ; : : : ; ak / is an

admissible partition for . If is any proper variation of , then Lg .s / is a smooth
function of s that
achieves its minimum at s D 0, so it follows from elementary
calculus that d Lg .s / =ds D 0 when s D 0. Since every proper vector field along
is the variation field of some proper variation, the right-hand side of (6.2) must
vanish for every such V .
First we show that Dt 0 D 0 on each subinterval Œai1 ; ai , so is a “broken
geodesic.” Choose one such interval, and let ' 2 C 1 .R/ be a bump function such
that ' > 0 on .ai1 ; ai / and ' D 0 elsewhere. Then (6.2) with V D 'Dt 0 becomes
Z ai
ˇ ˇ2
0D ' ˇDt 0 ˇ dt:
ai 1
Since the integrand is nonnegative and ' > 0 on .ai1 ; ai /, this shows that Dt 0 D 0
on each such subinterval.
Next we need to show that i 0 D 0 for each i between 0 and k, which is to say
that has no corners. For each such i , we can use a bump function in a coordinate
chart to construct a piecewise smooth vector field V along such that V .ai / D i 0
and V .aj / D 0 for j ¤ i . Then (6.2) reduces to ji 0 j2 D 0, so i 0 D 0 for each
i.
Finally, since the two one-sided velocity vectors of match up at each ai , it
follows from uniqueness of geodesics that jŒai ;ai C1 is the continuation of the
geodesic jŒai 1 ;ai , and therefore is smooth. t
u
The preceding proof has an enlightening geometric interpretation. Under the as-
sumption that Dt 0 ¤ 0, the first variation with V D 'Dt 0 is negative, which shows
that deforming in the direction of its acceleration vector field decreases its length
(Fig. 6.4). Similarly, the length of a broken geodesic is decreased by deforming it
Fig. 6.4: Deforming in the direction of the

acceleration Fig. 6.5: Rounding the corner
in the direction of a vector field V such that V .ai / D i 0 (Fig. 6.5). Geometrically,
this corresponds to “rounding the corner.”
The first variation formula actually tells us a bit more than is claimed in Theorem
6.4. In proving that is a geodesic, we did not use the full strength of the assumption
that the length of s takes a minimum when s D 0; we only used the fact that its
derivative is zero. We say that an admissible curve is a critical point of Lg if for
every proper variation s of , the derivative of Lg .s / with respect to s is zero at
s D 0. Therefore we can strengthen Theorem 6.4 in the following way.
Corollary 6.5. A unit-speed admissible curve is a critical point for Lg if and only
if it is a geodesic.
Proof. If is a critical point, the proof of Theorem 6.4 goes through without mod-
ification to show that is a geodesic. Conversely, if is a geodesic, then the first
term on the right-hand side of (6.2) vanishes by the geodesic equation, and the sec-
ond term vanishes because 0 has no jumps. t
u
The geodesic equation Dt 0 D 0 thus characterizes the critical points of the
length functional. In general, the equation that characterizes critical points of a func-
tional on a space of maps is called the variational equation or the Euler–Lagrange
equation of the functional. Many interesting equations in differential geometry arise
as variational equations. We touch briefly on three others in this book: the Einstein
equation (7.34), the Yamabe equation (7.59), and the minimal hypersurface equation
H 0 (Thm. 8.18).
Geodesics Are Locally Minimizing
Next we turn to the converse of Theorem 6.4. It is easy to see that the literal con-
verse is not true, because not every geodesic segment is minimizing. For example,
every geodesic segment on Sn that goes more than halfway around the sphere is
not minimizing, because the other portion of the same great circle is a shorter curve
segment between the same two points. For that reason, we concentrate initially on
local minimization properties of geodesics.
As usual, let .M; g/ be a Riemannian manifold. A regular (or piecewise regular)
curve W I ! M is said to be locally minimizing if every t0 2 I has a neighborhood
I0 I such that whenever a; b 2 I0 with a < b, the restriction of to Œa; b is
minimizing.
Fig. 6.6: The radial vector field in a normal neighborhood
Lemma 6.6. Every minimizing admissible curve segment is locally minimizing.
Our goal in this section is to show that geodesics are locally minimizing. The
proof will be based on a careful analysis of the geodesic equation in Riemannian
normal coordinates.
If " is a positive number such that expp is a diffeomorphism from the ball
B" .0/ Tp M to its image (where the radius of the ball is
measured with respect to
the norm defined by gp ), then the image set expp B" .0/ is a normal neighborhood
of p, called a geodesic ball in M , or sometimes an open geodesic ball for clarity.
Also, if the closed ball Bx" .0/ is contained in an open
set V Tp M on which
expp is a diffeomorphism onto its image, then expp Bx" .0/ is called a closed

geodesic ball, and expp @B" .0/ is called a geodesic sphere. Given such a V , by
compactness there exists "0 > " such that B"0 .0/ V , so every closed geodesic ball
is contained in an open geodesic ball of larger radius. In Riemannian normal co-
ordinates centered at p, the open and closed geodesic balls and geodesic spheres
centered at p are just the coordinate balls and spheres.
iSuppose
U is a normal neighborhood of p 2 M . Given any normal coordinates
x on U centered at p, define the radial distance function r W U ! R by
p
r.x/ D .x 1 /2 C C .x n /2 ; (6.4)
and the radial vector field on U X fpg (see Fig. 6.6), denoted by @r , by
xi @
@r D : (6.5)
r.x/ @x i
In Euclidean space, r.x/ is the distance to the origin, and @r is the unit vector field
pointing radially outward from the origin. (The notation is suggested by the fact that
@r is a coordinate derivative in polar or spherical coordinates.)
Lemma 6.8. In every normal neighborhood U of p 2 M , the radial distance function

and the radial vector field are well defined, independently of the choice of normal
coordinates. Both r and @r are smooth on U X fpg, and r 2 is smooth on all of U .
Proof. Proposition 5.23 shows that any two normal coordinate

charts on U are re-
lated by xzi D Aij x j for some orthogonal matrix Aij , and a straightforward com-
putation shows that both r and @r are invariant under such coordinate changes. The
smoothness statements follow directly from the coordinate formulas. t
u
The crux of the proof that geodesics are locally minimizing is the following de-
ceptively simple geometric lemma.
Theorem 6.9 (The Gauss Lemma). Let .M; g/ be a Riemannian manifold, let U
be a geodesic ball centered at p 2 M , and let @r denote the radial vector field on
U Xfpg. Then @r is a unit vector field orthogonal to the geodesic spheres in U Xfpg.

Proof. We will work entirely in normal coordinates x i on U centered at p, using
the properties of normal coordinates described in Proposition 5.24.
Let q 2 U X fpg be arbitrary, with coordinate representation q 1 ; : : : ; q n , and let
p
b D r.q/ D .q 1 /2 C C .q n /2 , where r is the radial distance function defined by
(6.4). It follows that @r jq has the coordinate representation
ˇ
ˇ q i @ ˇˇ
ˇ
@r q D :
b @x i ˇq
Let v D v i @i jp 2 Tp M be the tangent vector at p with components v i D q i =b.

By Proposition 5.24(c), the radial geodesic with initial velocity v is given in these
coordinates by
v .t / D t v 1 ; : : : ; t v n :
It satisfies v .0/ D p, v .b/ D q, and v0 .b/ D v i @i jq D @r jq . Because gp is equal
to the Euclidean metric in these coordinates, we have
ˇ 0 ˇ p p
ˇ .0/ˇ D jvjg D .v 1 /2 C C .v n /2 D 1 .q 1 /2 C C .q n /2 D 1;
v g b
so v is a unit vector, and thus v is a unit-speed geodesic. It follows that @r jq D v0 .b/
is also a unit vector.
To prove that @r is orthogonal
to the geodesic spheres let q, b, and v be as above,
and let ˙b D expp @Bb .0/ be the geodesic sphere containing q. In these coordi-
nates, ˙b is the set of points satisfying the equation .x 1 /2 C C˝ .x n /2 D 2
˛ b . Let
w 2 Tq M be any vector tangent to ˙b at q. We need to show that w; @r jq g D 0.
0
Choose a smooth curve W ."; "/! ˙b satisfying .0/ D q and .0/ D w, and
write its coordinate representation in x i -coordinates as .s/ D 1 .s/; : : : ; n .s/ .
The fact that .s/ lies in ˙b for all s means that
1 2 2
.s/ C C n .s/ D b 2 : (6.6)
Fig. 6.7: Proof of the Gauss lemma
Define a smooth map W ."; "/ Œ0; b ! U (Fig. 6.7) by

t 1 t
.s; t / D .s/; : : : ; n .s/ :
b b
For each s 2 ."; "/, s is a geodesic by Proposition 5.24(c). Its initial velocity is
s0 .0/ D .1=b/ i .s/@i jp, which is a unit vector by (6.6) and the fact that gp is the
Euclidean metric in coordinates; thus each s is a unit-speed geodesic.
As before, let S D @s and T D @t . It follows from the definitions that
ˇ
d ˇˇ
S.0; 0/ D s .0/ D 0I
ds ˇsD0
ˇ
d ˇˇ
T .0; 0/ D v .t / D vI
dt ˇtD0
ˇ
d ˇˇ
S.0; b/ D .s/ D wI
ds ˇsD0
ˇ
d ˇˇ
T .0; b/ D v .t / D v0 .b/ D @r jq :
dt ˇtDb
˝ ˛
Therefore hS; T i is zero when .s; t / D .0; 0/ and equal to w; @r jq when .s; t / D
.0; b/, so to prove the theorem it suffices to show that hS; T i is independent of t .
We compute
@
hS; T i D hDt S; T i C hS; Dt T i (compatibility with the metric)
@t
D hDs T ; T i C hS; Dt T i (symmetry lemma)
(6.7)
D hDs T ; T i C 0 (each s is a geodesic)
1 @
D jT j2 D 0 (jT j D js0 j 1 for all .s; t /):
2 @s
Fig. 6.8: Radial geodesics are minimizing
This proves the theorem. t

u
We will use the Gauss lemma primarily in the form of the next corollary.
Corollary 6.10. Let U be a geodesic ball centered at p 2 M , and let r and @r be

the radial distance and radial vector field as defined by (6.4) and (6.5). Then
grad r D @r on U X fpg.
Proof. By the result of Problem 2-10, it suffices to show that @r is orthogonal to the
level sets of r and @r .r/ j@r j2g . The first claim follows directly from the Gauss
lemma, and the second from the fact that @r .r/ 1 by direct computation in normal
coordinates, which in turn is equal to j@r j2g by the Gauss lemma. t
u
Here is the payoff: our first step toward proving that geodesics are locally mini-
mizing. Note that this is not yet the full strength of the theorem we are aiming for,
because it shows only that for each point on a geodesic, sufficiently small segments
of the geodesic starting at that point are minimizing. We will remove this restriction
after a little more work below.
Proposition 6.11. Let .M; g/ be a Riemannian manifold. Suppose p 2 M and q is

contained in a geodesic ball around p. Then (up to reparametrization) the radial
geodesic from p to q is the unique minimizing curve in M from p to q.

Proof. Choose " > 0 such that expp B" .0/ is a geodesic ball containing q. Let
W Œ0; c ! M be the radial geodesic from p to q parametrized by arc length, and
write .t / D expp .t v/ for some unit vector v 2 Tp M . Then Lg . / D c, since has
unit speed.
To show that is minimizing, we need to show that every other admissible curve
from p to q has length at least c. Let W Œ0; b ! M be an arbitrary admissible curve
from p to q, which we may assume to be parametrized by arc length as well. Let
a0 2 Œ0; b denote the last time that .t / D p, and b0 2 Œ0; b the first time after a0
that .t / meets the geodesic sphere ˙c of radius c around p (Fig. 6.8). Then the
composite function r ı is continuous on Œa0 ; b0 and piecewise smooth in .a0 ; b0 /,
so we can apply the fundamental theorem of calculus to conclude that
Z b0
d
r. .b0 // r. .a0 // D r. .t // dt
a0 dt
Z b0
D dr. 0 .t // dt
a0
Z b0 ˝ ˛
D grad rj.t/ ; 0 .t / dt
a0 (6.8)
Z b0 ˇ ˇˇ ˇ
ˇgrad rj.t/ ˇ ˇ 0 .t /ˇ dt
a0
Z b0
D j 0 .t /j dt
a0

D Lg jŒa0 ;b0 Lg . /:
Thus Lg . / r. .b0 // r. .a0 // D c, so is minimizing.
Now suppose Lg . / D c. Then b D c, and both inequalities in (6.8) are equal-
ities. Because we assume that is a unit-speed curve, the second of these equali-
ties implies that a0 D 0 and b0 D b D c, since otherwise the segments of before
t D a0 and after t D b0 would contribute ˇ positive lengths.
ˇˇ ˇ The
˝ first equality then ˛
implies that the nonnegative expression ˇgrad rj.t/ ˇ ˇ 0 .t /ˇ grad rj.t/ ; 0 .t / is
identically zero on Œ0; b, which is possible only if 0 .t / is a positive multiple
of grad rj.t/ for each t . Since we assume that has unit speed, we must have
0 .t / D grad rj.t/ D @r j.t/ . Thus and are both integral curves of @r passing
through q at time t D c, so D . t
u
The next two corollaries show how radial distance functions, balls, and spheres
in normal coordinates are related to their global metric counterparts.
Corollary 6.12. Let .M; g/ be a connected Riemannian manifold and p 2 M . Within

every open or closed geodesic ball around p, the radial distance function r.x/
defined by (6.4) is equal to the Riemannian distance from p to x in M .
Proof. Since every closed geodesic ball is contained in an open geodesic ball of
larger radius, we need only consider the open case. If x is in the open geodesic ball
expp Bc .0/ , the radial geodesic from p to x is minimizing by Proposition 6.11.
Since its velocity is equal to @r , which is a unit vector in both the g-norm and the
Euclidean norm in normal coordinates, the g-length of is equal to its Euclidean
length, which is r.x/. t
u
Corollary 6.13. In a connected Riemannian manifold, every open or closed geodesic

ball is also an open or closed metric ball of the same radius, and every geodesic
sphere is a metric sphere of the same radius.
g/ be a Riemannian manifold, and let p 2 M be arbitrary. First,

Proof. Let .M;
let V D expp Bxc .0/ M be a closed geodesic ball of radius c > 0 around p.
Suppose q is an arbitrary point of M . If q 2 V , then Corollary 6.12 shows that q
Uniformly Normal Neighborhoods 163
is also in the closed metric ball of radius c. Conversely, suppose q … V . Let S be
the geodesic sphere expp @Bc .0/ . The complement of S is the disjoint union of

the open sets expp Bc .0/ and M X expp Bxc .0/ , and hence disconnected. Thus if
W Œa; b ! M is any admissible curve from p to q, there must be a time t0 2 .a; b/
when .t0 / 2 S , and then Corollary 6.12 shows that the length of jŒa;t0 must be at
least c. Since jŒt0 ;b must have positive length, it follows that dg .p; q/ > c, so q is
not in the closed metric ball of radius c around p.
Next, let W D expp Bc .0/ be an open geodesic ball of radius c. Since W is
the union of all closed geodesic balls around p of radius less than c, and the open
metric ball of radius c is similarly the union of all closed metric metric balls of
smaller radii, the result of the preceding paragraph shows that W is equal to the
open metric ball of radius c.
Finally, if S D expp @Bc .0/ is a geodesic sphere of radius c, the arguments
above show that S is equal to the closed metric ball of radius c minus the open
metric ball of radius c, which is exactly the metric sphere of radius c. t
u
The last corollary suggests a simplified notation for geodesic balls and spheres
in M . From now on, we will use the notations Bc .p/ D expp Bc .0/ , Bxc .p/ D

expp Bxc .0/ , and Sc .p/ D expp @Bc .0/ for open and closed geodesic balls and
geodesic spheres, which we now know are also open and closed metric balls and
spheres. (To avoid confusion, we refrain from using this notation for other metric
balls and spheres unless explicitly stated.)
Uniformly Normal Neighborhoods
We continue to let .M; g/ be a Riemannian manifold. In order to prove that

geodesics in M are locally minimizing, we need the following refinement of the
concept of normal neighborhoods. A subset W M is called uniformly normal
if there exists some ı > 0 such that W is contained in a geodesic ball of radius ı
around each of its points (Fig. 6.9). If ı is any such constant, we will also say that
W is uniformly ı-normal. Clearly every subset of a uniformly ı-normal set is itself
uniformly ı-normal.
Lemma 6.14 (Uniformly Normal Neighborhood Lemma). Given p 2 M and any

neighborhood U of p, there exists a uniformly normal neighborhood of p contained
in U .

Proof. Choose
i a inormal
coordinate chart U0 ; x i centered at p and contained in
U , and let x ; v be the corresponding natural coordinates for 1 .U0 / TM .
Because this is a local question, we might as well identify U0 with an open sub-
set of Rn , and identify TM with U0 Rn . The exponential map for the Riemann-
ian manifold .U0 ; g/ is defined on an open subset E U0 Rn . Consider the map
E W E ! U0 U0 defined by
Fig. 6.9: Uniformly normal neighborhood
E.x; v/ D .x; expx v/:

The differential of E at .p; 0/ is represented by the matrix
˙ @x i
@x i

@x j @v j Id 0
dE.p;0/ D D ;
Id
@ expi @ expi
@x j @v j
which is invertible. By the inverse function theorem, therefore, there are neighbor-
hoods U U0 Rn of .p; 0/ and V U0 U0 of .p; p/ such that E restricts to a
diffeomorphism from U to V . Shrinking both neighborhoods if necessary, we may
assume that U is a product set of the form W B" .0/, where W is a neighborhood
of p and B" .0/ is a Euclidean ball in v-coordinates. It follows that for each x 2 W ,
expx maps B" .0/ smoothly onto the open set Vx D fy W .x; y/ 2 V g, and it is a dif-
feomorphism because its inverse is given explicitly by exp1 x .y/ D 2 ı E
1
.x; y/,
where 2 W U0 R ! R is the projection. Shrinking W still further if necessary,
n n
we may assume that the metric g satisfies an estimate of the form (2.21) for all
x 2 W . This means that for each x 2 W , the coordinate ball B" .0/ Tx M contains
a gx -ball of radius at least "=c. Put ı D "=c; we have shown that for each x 2 W ,
there is a g-geodesic ball of radius ı in M centered at x.
Now, shrinking W once more, we may assume that its diameter (with respect
to the metric dg ) is less than ı. It follows that for each x 2 W , the entire set W
is contained in the metric ball of radius ı around x, and Corollary 6.13 shows that
this metric ball is also a geodesic ball of radius ı. Thus W is the required uniformly
normal neighborhood of p. t
u
We are now ready to prove the main result of this section.
Uniformly Normal Neighborhoods 165
Fig. 6.10: Geodesics are locally minimizing
Theorem 6.15. Every Riemannian geodesic is locally minimizing.
Proof. Let .M; g/ be a Riemannian manifold. Suppose W I ! M is a geodesic,

which we may assume to be defined on an open interval, and let t0 2 I . Let W be a
uniformly normal neighborhood of .t0 /, and let I0 I be the connected compo-
nent of 1 .W / containing t0 . If a; b 2 I0 with a < b, then the definition of uniformly
normal neighborhood implies that .b/ is contained in a geodesic ball centered at
.a/ (Fig. 6.10). Therefore, by Proposition 6.11, the radial geodesic segment from
.a/ to .b/ is the unique minimizing curve segment between these points. How-
ever, the restriction of to Œa; b is also a geodesic segment from .a/ to .b/
lying in the same geodesic ball, and thus jŒa;b must coincide with this minimizing
geodesic. t
u
It is interesting to note that the Gauss lemma and the uniformly normal neighbor-
hood lemma also yield another proof that minimizing curves are geodesics, without
using the first variation formula. On the principle that knowing more than one proof
of an important fact always deepens our understanding of it, we present this proof
for good measure.
Another proof of Theorem 6.4. Suppose W Œa; b ! M is a minimizing admissible
curve. Just as in the preceding proof, for every t0 2 Œa; b we can find a connected
neighborhood I0 of t0 such that .I0 / is contained in a uniformly normal neigh-
borhood W . Then for every a0 ; b0 2 I0 , the same argument as above shows that the
unique minimizing curve segment from .a0 / to .b0 / is a radial geodesic. Since
the restriction of to Œa0 ; b0 is such a minimizing curve segment, it must coincide
with this radial geodesic. Therefore solves the geodesic equation in a neighbor-
hood of t0 . Since t0 was arbitrary, is a geodesic. t
u
Given a Riemannian manifold .M; g/ (without boundary), for each point p 2 M
we define the injectivity radius of M at p, denoted by inj.p/, to be the supremum
of all a > 0 such that expp is a diffeomorphism from Ba .0/ Tp M onto its image.
If there is no upper bound to the radii of such balls (as is the case, for example, on
Rn ), then we set inj.p/ D 1. Then we define the injectivity radius of M , denoted
by inj.M /, to be the infimum of inj.p/ as p ranges over points of M . It can be zero,
positive, or infinite. (The terminology is explained by Problem 10-24.)
Lemma 6.16. If .M; g/ is a compact Riemannian manifold, then inj.M / is positive.
Proof. For each x 2 M , there is a positive number ı.x/ such that x is contained in a
uniformly ı.x/-normal neighborhood Wx , and inj.x 0 / ı.x/ for each x 0 2 W . Since
M is compact, it is covered by finitely many such neighborhoods Wx1 ; : : : ; Wxk .
Therefore, inj.M / is at least equal to the minimum of ı.x1 /; : : : ; ı.xk /. It cannot be
infinite, because a compact metric space is bounded, and a geodesic ball of radius c
contains points whose distances from the center are arbitrarily close to c. t
u
In addition to uniformly normal neighborhoods, there is another, more special-
ized, kind of normal neighborhood that is frequently useful. Let .M; g/ be a Rie-
mannian manifold. A subset U M is said to be geodesically convex if for each
p; q 2 U , there is a unique minimizing geodesic segment from p to q in M , and the
image of this geodesic segment lies entirely in U .
The next theorem says that every sufficiently small geodesic ball is geodesically
convex.
Theorem 6.17. Let .M; g/ be a Riemannian manifold. For each p 2 M , there exists
"0 > 0 such that every geodesic ball centered at p of radius less than or equal to "0
is geodesically convex.
u
Completeness
Suppose .M; g/ is a connected Riemannian manifold. Now that we can view M as
a metric space, it is time to address one of the most important questions one can ask
about a metric space: Is it complete? In general, the answer is no: for example, if M
is an open ball in Rn with its Euclidean metric, then every sequence in M that converges
in Rn to a point in @M is Cauchy, but not convergent in M .
In Chapter 5, we introduced another notion of completeness for Riemannian and
pseudo-Riemannian manifolds: recall that such a manifold is said to be geodesically
complete if every maximal geodesic is defined for all t 2 R. For clarity, we will
use the phrase metrically complete for a connected Riemannian manifold that is
complete as a metric space with the Riemannian distance function, in the sense that
every Cauchy sequence converges.
The Hopf–Rinow theorem, which we will state and prove below, shows that these
two notions of completeness are equivalent for connected Riemannian manifolds.
Before we prove it, let us establish a preliminary result, which will have other im-
portant consequences besides the Hopf–Rinow theorem itself.
Completeness 167
Fig. 6.11: Proof that jŒ0;" aims at q
Lemma 6.18. Suppose .M; g/ is a connected Riemannian manifold, and there is a

point p 2 M such that expp is defined on the whole tangent space Tp M .
(a) Given any other q 2 M , there is a minimizing geodesic segment from p to q.
(b) M is metrically complete.
Proof. Let q 2 M be arbitrary. If W Œa; b ! M is a geodesic segment starting at

p, let us say that aims at q if is minimizing and
dg .p; q/ D dg .p; .b// C dg ..b/; q/: (6.9)
(This would be the case, for example, if were an initial segment of a minimizing
geodesic from p to q; but we are not assuming that.) To prove (a), it suffices to show
that there is a geodesic segment W Œa; b ! M that begins at p, aims at q, and has
length equal to dg .p; q/, for then the fact that is minimizing means that
dg .p; .b// D Lg . / D dg .p; q/, and (6.9) becomes
dg .p; q/ D dg .p; q/ C dg ..b/; q/;
which implies .b/ D q. Since is a segment from p to q of length dg .p; q/, it is

the desired minimizing geodesic segment.
Choose " > 0 such that there is a closed geodesic ball Bx" .p/ around p that does
not contain q. Since the distance function on a metric space is continuous, there
is a point x in the geodesic sphere S" .p/ where dg .x; q/ attains its minimum on
the compact set S" .p/. Let be the maximal unit-speed geodesic whose restriction
to Œ0; " is the radial geodesic segment from p to x (Fig. 6.11); by assumption, is
defined for all t 2 R.
We begin by showing that jŒ0;" aims at q. Since it is minimizing by Proposition
6.11 (noting that every closed geodesic ball is contained in a larger open one), we
need only show that (6.9) holds with b D ", or
dg .p; q/ D dg .p; x/ C dg .x; q/: (6.10)
To this end, let W Œa0 ; b0 ! M be any admissible curve from p to q. Let t0 be the
first time hits S" .p/, and let 1 and 2 denote the restrictions of 1 to Œa0 ; t0 and
Œt0 ; b0 , respectively (Fig. 6.11). Since every point in S" .p/ is at a distance " from
Fig. 6.12: Proof that A D T
p, we have Lg .1 / dg .p; .t0 // D dg .p; x/; and by our choice of x we have
Lg .2 / dg . .t0 /; q/ dg .x; q/. Putting these two inequalities together yields
Lg . / D Lg .1 / C Lg .2 / dg .p; x/ C dg .x; q/:
Taking the infimum over all such , we find that dg .p; q/ dg .p; x/ C dg .x; q/.
The opposite inequality is just the triangle inequality, so (6.10) holds.
Now let T D dg .p; q/ and
˚ ˇ
A D b 2 Œ0; T W ˇŒ0;b aims at q :
We have just shown that " 2 A. Let A D sup A ". By continuity of the distance
function, it is easy to see that A is closed in Œ0; T , and therefore A 2 A. If A D T ,
then jŒ0;T is a geodesic of length T D dg .p; q/ that aims at q, and by the remark
above we are done. So we assume A < T and derive a contradiction.
Let y D .A/, and choose ı > 0 such that there is a closed geodesic ball Bxı .y/
around y, small enough that it does not contain q (Fig. 6.12). The fact that A 2 A
means that
dg .y; q/ D dg .p; q/ dg .p; y/ D T A:
Let z 2 Sı .y/ be a point where dg .z; q/ attains its minimum, and let W Œ0; ı ! M
be the unit-speed radial geodesic from y to z. By exactly the same argument as
before, aims at q, so
dg .z; q/ D dg .y; q/ dg .y; z/ D .T A/ ı: (6.11)
By the triangle inequality and (6.11),
dg .p; z/ dg .p; q/ dg .z; q/

D T .T A ı/ D A C ı:
Therefore, the admissible curve consisting of jŒ0;A (of length A) followed by (of
length ı) is a minimizing curve from p to z. This means that it has no corners, so z must
lie on , and in fact, z D .A C ı/. But then (6.11) says that
Completeness 169
Fig. 6.13: Cauchy sequences converge
dg .p; q/ D T D .A C ı/ C dg .z; q/ D dg .p; z/ C dg .z; q/;
so jŒ0;ACı aims at q and A C ı 2 A, which is a contradiction. This completes the

proof of (a).
To prove (b), we need to show that every Cauchy sequence in M converges. Let
.qi / be a Cauchy sequence in M . For each i , let i .t / D expp .t vi / be a unit-speed
minimizing geodesic from p to qi , and let di D dg .p; qi /, so that qi D expp .di vi /
(Fig. 6.13). The sequence .di / is bounded in R (because Cauchy sequences in a
metric space are bounded), and the sequence .vi / consists of unit vectors in Tp M ,
so the sequence of vectors .di vi / in Tp M is bounded. Therefore a subsequence
.dik vik / converges to some v 2 Tp M . By continuity of the exponential map, qik D
expp .dik vik / ! expp v, and since the original sequence .qi / is Cauchy, it converges
to the same limit. t
u
The next theorem is the main result of this section.
Theorem 6.19 (Hopf–Rinow). A connected Riemannian manifold is metrically

complete if and only if it is geodesically complete.
Proof. Let .M; g/ be a connected Riemannian manifold. Suppose first that M is

geodesically complete. Then in particular, it satisfies the hypothesis of Lemma 6.18,
so it is metrically complete.
Conversely, suppose M is metrically complete, and assume for the sake of
contradiction that it is not geodesically complete. Then there is some unit-speed
geodesic W Œ0; b/ ! M that has no extension to a geodesic on any interval Œ0; b 0 /
for b 0 > b. Let .ti / be any increasing sequence in Œ0; b/ that approaches b, and set
qi D .ti /. Since is parametrized by arc length, the length of jŒti ;tj is exactly
jtj ti j, so dg .qi ; qj / jtj ti j and .qi / is a Cauchy sequence in M . By complete-
ness, .qi / converges to some point q 2 M .
Fig. 6.14: extends past q
Let W be a uniformly ı-normal neighborhood of q for some ı > 0. Choose j

large enough that tj > b ı and qj 2 W (Fig. 6.14). The fact that Bı .qj / is a
geodesic ball means that every unit-speed geodesic starting at qj exists at least for
t 2 Œ0; ı/. In particular, this is true of the geodesic with .0/ D qj and 0 .0/ D
0 .tj /. Define z W Œ0; tj C ı/ ! M by
(
.t /; t 2 Œ0; b/;
z.t / D
.t tj /; t 2 .tj ı; tj C ı/:
Note that both expressions on the right-hand side are geodesics, and they have the
same position and velocity when t D tj . Therefore, by uniqueness of geodesics, the
two definitions agree where they overlap. Since tj C ı > b, z is an extension of
past b, which is a contradiction. t
u
A connected Riemannian manifold is simply said to be complete if it is either
geodesically complete or metrically complete; the Hopf–Rinow theorem then im-
plies that it is both. For disconnected manifolds, we interpret “complete” to mean
geodesically complete, which is equivalent to the requirement that each component
be a complete metric space. As mentioned in the previous chapter, complete mani-
folds are the natural setting for global questions in Riemannian geometry.
We conclude this section by stating three important corollaries, whose proofs are
easy applications of Lemma 6.18 and the Hopf–Rinow theorem.
Corollary 6.20. If M is a connected Riemannian manifold and there exists a point

p 2 M such that the restricted exponential map expp is defined on all of Tp M , then
M is complete. t
u
Corollary 6.21. If M is a complete, connected Riemannian manifold, then any

two points in M can be joined by a minimizing geodesic segment. t
u
Corollary 6.22. If M is a compact Riemannian manifold, then every maximal

geodesic in M is defined for all time. t
u
Completeness 171
Fig. 6.15: Lifting geodesics
The Hopf–Rinow theorem and Corollary 6.20 are key ingredients in the follow-
ing theorem about Riemannian covering maps. This theorem will play a key role in
the proofs of some of the local-to-global theorems in Chapter 12.

Theorem 6.23. Suppose M ; gz and .M; g/ are connected Riemannian manifolds
with M complete, and W M ! M is a local isometry. Then M is complete and
is a Riemannian covering map.
Proof. A fundamental property of covering maps is the path-lifting property (Prop.

A.54(b)): if is a covering map, then every continuous path W I !M lifts to a path
z in M such that ı z D . We begin by proving that possesses the path-lifting
property for geodesics (Fig. 6.15): if p 2 M is a point in the image of , W I ! M
is any geodesic starting at p, and pz is any point in 1 .p/, then has a unique
z The lifted curve is necessarily also a geodesic because is a local
lift starting at p.
isometry.
To prove the path-lifting property for geodesics, suppose p 2 .M / and pz 2
1 .p/, and let W I ! M be any geodesic with p D .0/. Let v D 0 .0/ and
(which is well defined because d pz is an isomorphism),
vz D .d pz /1 .v/ 2 Tpz M
and let z be the geodesic in M with initial point pz and initial velocity vz. Because M

is complete, z is defined on all of R. Since is a local isometry, ittakes geodesics
to geodesics; and since by construction z.0/ D .0/ and d pz z0 .0/ D 0 .0/,
we must have ı z D on I , so zjI is a lift of starting at p.
z
S
z˛ \ U
Fig. 6.16: Proof that U zˇ D ¿ Fig. 6.17: Proof that 1 .U / z
˛ U˛
To show that M is complete, let p be any point in the image of . If W I ! M

is any geodesic starting at p, then has a lift z W I ! M . Because M is complete,
ı z is a geodesic defined for all t that coincides with on I , so extends to all of
R. Thus M is complete by Corollary 6.20.
Next we show that is surjective. Choose some point pz 2 M , write p D .p/,z
and let q 2 M be arbitrary. Because M is connected and complete, there is a mini-
mizing unit-speed geodesic segment p to q. Letting z be the lift of starting
from
at pz and r D dg .p; q/, we have z.r/ D .r/ D q, so q is in the image of .
To show that is a smooth covering map, we need to show that every point of M
has a neighborhood U that is evenly covered, which means that 1 .U / is a disjoint
union of connected open sets Uz˛ such that jUz ˛ W Uz ˛ ! U is a diffeomorphism. We
will show, in fact, that every geodesic ball is evenly covered.
Let p 2 M , and let U D B" .p/ be a geodesic ball centered at p. Write 1 .p/ D
fpz˛ g˛2A , and for each ˛ let Uz˛ denote the metric ball of radius " around pz˛ (we are
not claiming that Uz˛ is a geodesic ball). The first step is to show that the various sets
Uz˛ are disjoint. For every ˛ ¤ ˇ, there is a minimizing geodesic segment z from pz˛
to pzˇ because M is complete. The projected curve D ı z is a geodesic segment
that starts and ends at p (Fig. 6.16), whose length is the same as that of z. Such
a geodesic must leave U and reenter it (since all geodesics passing through p and
lying in U are radial line segments), and thus must have length at least 2". This
means that dgz .pz˛ ; pzˇ / 2", and thus by the triangle inequality, Uz˛ \ Uzˇ D ¿.
S
The next step is to show that 1 .U / D ˛ Uz˛ . If qz is any point in Uz˛ , then there
is a geodesic z of length less than " from pz˛ to qz, and then ı z is a geodesic
S of the
same length from p to .z q /, showing that .zq / 2 U . It follows that ˛ Uz˛
1 .U /.
Completeness 173
Conversely, suppose qz 2 1 .U /, and set q D .z q /. This means that q 2 U , so

there is a minimizing radial geodesic in U from q to p, and r Ddg .q; p/ < ".
Let z be the lift of starting at qz (Fig. 6.17). It follows that z.r/ D .r/ D p.
Therefore z.r/ D pz˛ for some ˛, and dgz .z / D r < ", so qz 2 Uz ˛ .
q ; pz˛ / Lg .z
It remains only to show that W Uz˛ ! U is a diffeomorphism for each ˛. It is
certainly a local diffeomorphism (because is). It is bijective because its inverse
can be constructed explicitly: it is the map sending each radial geodesic starting at
p to its lift starting at pz˛ . This completes the proof. t
u
and M are connected Riemannian manifolds, and
Corollary 6.24. Suppose M
! M is a Riemannian covering map. Then M is complete if and only if
W M
is complete.
M
Proof. A Riemannian covering map is, in particular, a local isometry. Thus if M is
complete, satisfies the hypotheses of Theorem 6.23, which implies that M is also
complete.

suppose M iscomplete. Let pz 2 M and vz 2 Tp M be arbitrary, and
Conversely,
let p D pz and v D d pz vz . Completeness of M implies that the geodesic with
starting
.0/ D p and 0 .0/ D v is defined for all t 2 R, and then its lift z W R ! M

at pz is a geodesic in M with initial velocity vz, also defined for all t . t
u
Corollary 6.21 to the Hopf–Rinow theorem shows that any two points in a com-
plete, connected Riemannian manifold can be joined by a minimizing geodesic seg-
ment. The next proposition gives a refinement of that statement.
Proposition 6.25. Suppose .M; g/ is a complete, connected Riemannian manifold,
and p; q 2 M . Every path-homotopy class of paths from p to q contains a geodesic
segment that minimizes length among all admissible curves in the same path-
homotopy class.
Proof. Let W M ! M be the universal covering manifold of M , endowed with
the pullback metric gz D g. Given p; q 2 M and a path W Œ0; 1 ! M from p to
be the lift of starting at p,
q, choose a point pz 2 1 .p/, and let z W Œ0; 1 ! M z
and set qz D z .1/. By Corollary 6.21, there is a minimizing gz-geodesic segment z
from pz to qz, and because is a local isometry, D ı z is a geodesic in M from
p to q. If 1 is any other admissible curve from p to q in the same path-homotopy
class, then by the monodromy theorem (Prop. A.54(c)), its lift z1 starting at pz also
ends at qz. Thus z1 is no longer than z, which implies 1 is no longer than . t
u
Closed Geodesics
Suppose .M; g/ is a connected Riemannian manifold. A closed geodesic in M is a

nonconstant geodesic segment W Œa; b ! M such that .a/ D .b/ and 0 .a/ D
0 .b/.
I Exercise 6.26. Show that a geodesic segment is closed if and only if it extends to a
periodic geodesic defined on all of R.
Round spheres have the remarkable property that all of their geodesics are closed
when restricted to appropriate intervals. Of course, this is not typically the case,
even for compact Riemannian manifolds; but it is natural to wonder whether closed
geodesics exist in more general manifolds. Much work has been done in Riemannian
geometry to determine how many closed geodesics exist in various situations. Here
we can only touch on the simplest case; these results will be useful in some of the
proofs of local-to-global theorems in Chapter 12.
A continuous path W Œ0; 1 ! M is called a loop if .0/ D .1/. Two loops
0 ; 1 W Œ0; 1 ! M are said to be freely homotopic if they are homotopic through
closed paths (but not necessarily preserving the base point), that is, if there exists a
homotopy H W Œ0; 1 Œ0; 1 ! M satisfying
H.s; 0/ D 0 .s/ and H.s; 1/ D 1 .s/ for all s 2 Œ0; 1;

(6.12)
H.0; t / D H.1; t / for all t 2 Œ0; 1:
This is an equivalence relation on the set of all loops in M , and an equivalence class
is called a free homotopy class. The trivial free homotopy class is the equivalence
class of any constant path.
I Exercise 6.27. Given a connected manifold M and a point x 2 M , show that a loop
based at x represents the trivial free homotopy class if and only if it represents the identity
element of 1 .M; x/.
The next proposition shows that closed geodesics are easy to find on compact
Riemannian manifolds that are not simply connected.
Proposition 6.28 (Existence of Closed Geodesics). Suppose .M; g/ is a compact,
connected Riemannian manifold. Every nontrivial free homotopy class in M is rep-
resented by a closed geodesic that has minimum length among all admissible loops
in the given free homotopy class.
u
The previous proposition guarantees the existence of at least one closed geodesic
on every non-simply-connected compact Riemannian manifold. In fact, it was
proved in 1951 by the Russian mathematicians Lazar Lyusternik and Abram Fet
that closed geodesics exist on every compact Riemannian manifold, but the proof
in the simply connected case is considerably harder. Proofs can be found in [Jos17]
and [Kli95].
Distance Functions
Suppose .M; g/ is a connected Riemannian manifold and S M is any subset. For
each point x 2 M , we define the distance from x to S to be
˚
dg .x; S / D inf dg .x; p/ W p 2 S :
Distance Functions 175
Lemma 6.29. Suppose .M; g/ is a connected Riemannian manifold and S M is

any subset.
(a) dg .x; S / dg .x; y/ C dg .y; S / for all x; y 2 M .
(b) x 7! dg .x; S / is a continuous function on M .
The simplest example of a distance function occurs when the set S is just a
singleton, S D fpg. Inside a geodesic ball around p, Corollary 6.12 shows that
dg .x; S / D r.x/, the radial distance function, and Corollary 6.10 shows that it has
unit gradient where it it smooth (everywhere inside the geodesic ball except at p
itself). The next theorem is a far-reaching generalization of that result.
Theorem 6.31. Suppose .M; g/ is a connected Riemannian manifold, S M is

arbitrary, and f W M ! Œ0; 1/ is the distance to S , that is, f .x/ D dg .x; S / for all
x 2 M . If f is continuously differentiable on some open subset U M X S , then
j grad f j 1 on U .
Proof. Suppose U M X S is an open subset onˇ which f ˇ is continuously dif-

ˇ ˇ
ˇferentiable,ˇ and x 2 U . We will show first that grad f jx 1, and then that
ˇ grad f jx ˇ 1.
To prove the first inequality, we may assume grad f jx ¤ 0, for otherwise the
inequality is trivial. Let v 2 Tx M be any unit vector, and let be the unit-
speed geodesic with .0/ D x and 0 .0/ D v. Then for every positive t sufficiently
small that jŒ0;t is minimizing, Lemma 6.29 gives dg ..t /; S / dg ..t /; .0// C
dg ..0/; S /, or equivalently f ..t // t C f ..0//. Therefore, since f is differen-
tiable at x, ˇ
d ˇˇ f ..t // f ..0//
ˇ f ..t // D lim 1:
dt tD0 t&0 t
ˇ ˇ
In particular, taking v D .grad f jx /=ˇ grad f jx ˇ (the unit vector in the direction of
grad f jx ), we obtain
ˇ
d ˇˇ ˝ ˛ ˇ ˇ
ˇ f ..t // D dfx 0 .0/ D grad f jx ; v D ˇ grad f jx ˇ:
dt tD0
ˇ ˇ
This proves that ˇ grad f jx ˇ 1.
ˇ To prove ˇ the reverse inequality, assume for the sake of contradiction that
ˇ grad f jx ˇ<1. Since we are assuming that grad f is continuous on U , there exist
ı; " > 0suchthat Bx" .x/ is a closed geodesic ball contained in U and j grad f j 1ı
on Bx" .x/. Let c be a positive constant less than "ı. By definition of dg .x; S /, there
is an admissible curve ˛ W Œ0; b ! M (which we may assume to be parametrized by
arc length) such that ˛.0/ D x, ˛.b/ 2 S , and b D Lg .˛/ < dg .x; S / C c.
Since we are assuming Bx" .x/ U M X S , we have that b > ", so ˛jŒ";b is an
admissible curve from ˛."/ to S . On the one hand,

dg .˛."/; S / Lg ˛jŒ";b D b " < dg .x; S / C c ": (6.13)
On the other hand, for 0 t ", the fact that ˛.t / 2 Bx" .x/ implies
ˇ ˇ
ˇd ˇ ˇ˝ ˛ˇ ˇ ˇ
ˇ f .˛.t //ˇ D ˇ grad f j˛.t/ ; ˛ 0 .t / ˇ ˇgrad f j˛.t/ ˇ j˛ 0 .t /j 1 ı:
ˇ dt ˇ
Thus f .˛.t // f .x/ .1 ı/t for all such t . In particular, for t D ", this means
that
dg .˛."/; S / dg .x; S / .1 ı/": (6.14)
Combining (6.13) and (6.14) yields c > "ı, contradicting our choice of c. t
u
Motivated by the previous theorem, if .M; g/ is a Riemannian manifold and U
M is an open subset, we define a local distance function on U to be a continuously
differentiable function f W U ! R such that j grad f jg 1 in U . Theorem 6.34
and Corollary 6.35 below will justify this terminology. But first, we develop some
important general properties of local distance functions.
Theorem 6.32. Suppose .M; g/ is a Riemannian manifold and f is a smooth local

distance function on an open subset U M . Then rgrad f .grad f / 0, and each
integral curve of grad f is a unit-speed geodesic.
Proof. Let F 2 X.U / denote the unit vector field grad f . The definition of the
gradient shows that for every vector field W , we have
Wf D df .W / D hF; W i; (6.15)
and therefore
Ff D hF; F i D j grad f j2 D 1: (6.16)
For every smooth vector field W on U , we have
hW; rF F i D F hW; F i hrF W; F i .compatibility with g/

D F Wf hŒF; W ; F i hrW F; F i .(6.15), symmetry of r/
D F Wf ŒF; W f 12 W jF j2 .(6.15), compatibility with g/
2
D W Ff 1
2
W jF j .definition of ŒF; W /
D0 .since Ff jF j2 1/:
Since W is arbitrary, this proves that rF F 0.

If W I ! U is an integral curve of F , then the fact that 0 is extendible implies
Dt 0 .t / D rF F j.t/ D 0;
so is a geodesic. t
u
Lemma 6.33. Suppose .M; g/ is a Riemannian manifold, K M , and f W K ! R

is a continuous function whose restriction to some open set W K is a smooth local
distance
function.
For every admissible curve W Œa0 ; b0 ! K such that
.a0 ; b0 / W , we have
Fig. 6.18: Proving that f .x/ D dg .x; S/ in a neighborhood of S
Lg . / jf . .b0 // f . .a0 //j:
Proof. This is proved exactly as in (6.8), noting that the only properties of r we used
in that computation were that it is continuous on the image of and continuously
differentiable on .a0 ; b0 / , and its gradient has unit length there. t
u
The next theorem and its corollary explain why the name “local distance func-
tion” is justified. Its proof is an adaptation of the proof of Proposition 6.11.
Theorem 6.34. Suppose .M; g/ is a Riemannian manifold, U M is an open sub-

set, S U , and f W U ! Œ0; 1/ is a continuous function such that f 1 .0/ D S
and f is a smooth local distance function on U X S . Then there is a neighborhood
U0 U of S in which f .x/ is equal to the distance in M from x to S .
Proof. For each p 2 S , there are positive numbers "p ; ıp such that B"p .p/ is a
uniformly ıp -normal geodesic ball and B2"p .p/ U . This means that B"p .p/ is
contained in the open geodesic ball of radius ıp around each of its points. In partic-
ular, B"p .p/ Bıp .p/, which means that ıp "p , and thus every geodesic starting
at a point of B"p .p/ is defined at least for t 2 ."p ; "p /. Let U0 be the union of all
of the geodesic balls B"p .p/ for p 2 S , which is a neighborhood of S contained in
U (Fig. 6.18).
Let x 2 U0 be arbitrary, and let c D f .x/. We will show that dg .x; S / D c. If
x 2 S , then dg .x; S / D 0 D c, so we may as well assume x … S .
There is some p 2 S such that x 2 B"p .p/, which means that dg .x; S / < "p
and geodesics starting at x are defined at least on ."p ; "p /. Also, if ˛ W Œ0; b !
B" .p/ is the radial geodesic segment from p to x, it follows from Lemma 6.33 that
Lg .˛/ jf .x/ f .p/j D c, and we conclude that c Lg .˛/ < "p as well.
Let W ."p ; "p / ! U be the unit-speed geodesic starting at x with initial veloc-
ity equal to grad f jx . By Theorem 6.32 and uniqueness of geodesics, coincides
with an integral curve of grad f as long as .t / 2 U X S , which is to say as long
as f ..t // ¤ 0. For all such t we have
d ˝ ˛ ˇ ˇ2
f ..t // D grad f j.t/ ; 0 .t / D ˇgrad f j.t/ ˇ D 1;
dt
so f ..t // D c t as long as t < c, and by continuity, f ..c// D 0. This means

that .c/ 2 S , and jŒ0;c is a curve segment of length c connecting x with S , so
dg .x; S / c.
To prove the reverse inequality, suppose ˛ W Œa; b ! M is any admissible curve
starting at x and ending at a point of S . Assume first that ˛.t / 2 U for all t 2 Œa; b,
and let b0 2 Œa;
b bethe first time that ˛.b0 / 2 S . Then Lemma 6.33 shows that
Lg .˛/ Lg ˛jŒa;b0 jf .˛.b0 // f .˛.a//j D c. On the other hand, suppose
˛.t / 2 M X U for some t . The triangle inequality impliesB"p .x/ B2"p .p/
U , so thereis a first time b0 2 Œa; b such that dg x; ˛.b0 / "p . Then Lg .˛/
Lg ˛jŒa;b0 "p > c. Taken together, these two inequalities show that Lg .˛/ c
for every such ˛, which implies dg .x; S / c. t
u
See Problem 6-27 for a global version of the preceding theorem.
Corollary 6.35. Let .M; g/ be a Riemannian manifold, and let f be a smooth local
distance function on an open subset U M . If c is a real number such that S D
f 1 .c/ is nonempty, then there is a neighborhood U0 of S in U on which jf .x/ cj
is equal to the distance in M from x to S .
I Exercise 6.36. Prove the preceding corollary.
Distance Functions for Embedded Submanifolds
The most important local distance functions are those associated with embedded
submanifolds. As we will see in this section, such distance functions are always
smooth near the manifold.
Suppose .M; g/ is a Riemannian n-manifold (without boundary) and P M
is an embedded k-dimensional submanifold. Let NP denote the normal bundle of
P in M , and let U M be a normal neighborhood of P in M , which is the dif-
feomorphic image of a certain open subset V NP under the normal exponential
map. (Such a neighborhood always exists by Thm. 5.25.) We begin by constructing
generalizations of the radial distance function and radial vector field (see (6.4) and
(6.5)). Recall the definition of Fermi coordinates from Chapter 5 (see Prop. 5.26).
Proposition 6.37. Let P be an embedded submanifold of a Riemannian manifold
.M; g/ and let U be any normal neighborhood of P in M . There exist a unique
continuous function r W U ! Œ0; 1/ and smooth vector field @r on U X P that
have
1 the following coordinate
representations in terms of any Fermi coordinates
x ; : : : ; x k ; v 1 ; : : : ; v nk for P on a subset U0 U :
1 k 1 nk
q
r x ;:::;x ;v ;:::;v D .v 1 /2 C C .v nk /2 ; (6.17)
1 nk
v @ v @
@r D 1
CC : (6.18)
r.x; v/ @v r.x; v/ @v nk
The function r is smooth on U X P , and r 2 is smooth on all of U .

Proof. The uniqueness, continuity, and smoothness claims follow immediately from
the coordinate expressions (6.17) and (6.18), so we need only prove that r and @r
can be globally defined so as to have the indicated coordinate expressions in any
Fermi coordinates.
Let V NS be the subset that is mapped diffeomorphically onto U by the
normal exponential map E. Define a function
W V ! Œ0; 1/ by
.p; v/ D jvjg ,
and define r W U ! Œ0; 1/ by r D
ı E 1 . Any Fermi coordinates for P are
defined by choosing local coordinates x 1 ; : : : ; x k for P and a local orthonormal
1 k 1 nk

frame .E˛ / for NP , and assigning
the coordinates x .p/; : : : ; x .p/; v ; : : : ; v
to the point E p; v ˛ E˛ jp . (Here we are using the summation convention with
Greek indices
running from 1 to n k.) Because the frame is orthonormal, for each
.p; v/ D p; v ˛ E˛ jp 2 V we have r.E.p; v//2 D
.p; v/2 D .v 1 /2 C C.v nk /2 ,
which shows that r has the coordinate representation (6.17).
To define @r , let q be an arbitrary point in U X P . Then q D expp .v/ for a unique
.p; v/ 2 V , and the curve W Œ0; 1 ! U given by .t / D expp .t v/ is a geodesic
from p to q. Define
1 0
@r jq D .1/; (6.19)
r.q/
which is independent of the choice of coordinates. Proposition 5.26 shows that in
coordinates, if we writev D v E˛ jp , then has the coordinate formula
˛
any Fermi
.t / D x 1 ; : : : ; x k ; t v 1 ; : : : ; t v nk , and therefore 0 .t / D v ˛ @=@v ˛ j.t/ . It follows
that @r has the coordinate formula (6.18). t
u
By analogy with the special case in which P is a point, we call r the radial
distance function for P and @r the radial vector field for P.
Theorem 6.38 (Gauss Lemma for Submanifolds). Let P be an embedded sub-
manifold of a Riemannian manifold .M; g/, let U be a normal neighborhood of P
in M , and let r and @r be defined as in Proposition 6.37. On U X P , @r is a unit
vector field orthogonal to the level sets of r.
Proof. The proof is a dressed-up version of the proof of the ordinary Gauss lemma.
Let q 2 U X P be arbitrary, and let x 1 ; : : : ; x k ; v 1 ; : : : ; v nk be the coordinate
representation of q in some choice of Fermi coordinates associated with a local
orthonormal frame .E˛ / for NP . As in the proof of Proposition 6.37, q D .1/,
where is the geodesic expp .t v/ for some p 2 P and v D v ˛ E˛ jp 2 Np M . Since
the frame .E˛ / is orthonormal, we have
q
j 0 .0/jg D jvjg D .v 1 /2 C C .v nk /2 D r.q/:
Since geodesics have constant speed, it follows that j 0 .1/jg D r.q/ as well, and
then (6.19) shows that @r jq is a unit vector.
Next we show that @r is orthogonal to the level sets of r. Suppose q 2 U X P , and
write q D expp0 .v0 / for some p0 2 P and v0 2 Np P with v0 ¤ 0. Let b D r.q/ D
jv0 jg , so q lies in the level set r 1 .b/. The coordinate representation (6.17) shows
that this is a regular level set, and hence an embedded submanifold of U .
Let w 2 Tq M be an arbitrary vector tangent to this level set, and let W ."; "/ !
U be a smooth curve lying in the same level set, with .0/ D q and 0 .0/ D w. We
can write .s/ D expx.s/ .v.s//, where x.s/ 2 P and v.s/ 2 Nx.s/ P with jv.s/jg D
b. The initial condition .0/ D q translates to x.0/ D p0 and v.0/ D v0 . Define a
smooth one-parameter family of curves W ."; "/ Œ0; b ! M by

t 1 t n
.s; t / D expx.s/ v .s/; : : : ; v .s/ :
b b
Since jv.s/jg =b 1, each s is a unit-speed geodesic.

Write T .s; t / D @t .s; t / and S.s; t / D @s .s; t / as in the proof of Theorem 6.9.
We have the following endpoint conditions:
ˇ
d ˇˇ
S.0; 0/ D x.s/ D x 0 .0/I
ds ˇsD0
ˇ
d ˇˇ t 1
T .0; 0/ D expp0 v0 D v0 I
dt ˇtD0 b b
ˇ
d ˇˇ
S.0; b/ D .s/ D wI
ds ˇsD0
ˇ
d ˇˇ t ˇ
T .0; b/ D ˇ expp0 v0 D @r ˇq :
dt tDb b
Then the same computation as in (6.7) shows that .@=@t /hS; T ig 0, and therefore
hw; @r jq ig D hS.0; b/; T .0; b/ig D hS.0; 0/; T .0; 0/ig D .1=b/hx 0 .0/; v0 ig , which
is zero because x 0 .0/ is tangent to P and v0 is normal to it. This proves that @r is
orthogonal to the level sets of r. t
u
Corollary 6.39. Assume the hypotheses of Theorem 6.38.
(a) @r is equal to the gradient of r on U X P .
(b) r is a local distance function.
(c) Each unit-speed geodesic W Œa; b/ ! U with 0 .a/ normal to P coincides with
an integral curve of @r on .a; b/.
(d) P has a tubular neighborhood in which the distance in M to P is equal to r.
Proof. By direct computation in Fermi coordinates using formulas (6.17) and
(6.18), @r .r/ 1, which is equal to j@r j2g by the previous theorem. Thus @r D grad r
on U X P by Problem 2-10. Because grad r is a unit vector field, r is a local dis-
tance function. By Proposition 5.26, the geodesics in U that start normal to P are
represented in any Fermi coordinates by t 7! .x 1 ; : : : ; x k ; t v 1 ; : : : ; t v nk /, and such
a geodesic has unit speed if and only if .v 1 /2 C C .v nk /2 D 1. Another direct
computation shows that each such curve is an integral curve of @r wherever r ¤ 0.
Finally, to prove (d), note that Theorem 6.34 shows that there is some neighborhood
Uz0 of P in M on which r.x/ D dg .x; P /; if we take U0 to be a tubular neighborhood
of P in Uz0 , then U0 satisfies the conclusion. t
u
When P is compact, we can say more.
Semigeodesic Coordinates 181
Theorem 6.40. Suppose .M; g/ is a connected Riemannian manifold, P M is a

compact submanifold, and U" is an "-tubular neighborhood of P . Then U" is also
an "-neighborhood in the metric space sense, and inside U" , the distance in M to
P is equal to the function r defined in Proposition 6.37.
Proof. First we show that r can be extended continuously to Ux" by setting r.q/ D "
for q 2 @U" . Indeed, suppose q 2 @U" and qi is any sequence of points in U" converg-
ing to q. Then lim supi r.qi / " because r.qi / < " for each i . Let c D lim infi r.qi /;
we will prove the result by showing that c D ". Suppose for the sake of contra-
diction that c < ". By passing to a subsequence, we may assume that r.qi / ! c.
We can write qi D exppi .vi / for pi 2 P and vi 2 Npi P , and because P is com-
pact and limi jvi jg D limi r.qi / D c, we can pass to a further subsequence and
assume that .pi ; vi / ! .p; v/ 2 NP with jvjg D c < ". Then we have q D limi qi D
limi exppi .vi / D expp v, which lies in the open set U" , contradicting our assumption
that q 2 @U" . Henceforth, we regard r as a continuous function on Ux" .
Now to prove the theorem, let W" denote the "-neighborhood of P in the metric
space sense. Suppose first that q 2 M X U" , and suppose ˛ W Œa; b ! M is any
admissible curve from a point of P to q. There is a first time b0 2 Œa; b that ˛.b0 / 2
@U" , and then Lemma 6.33 shows that

Lg .˛/ Lg ˛jŒa;b0 jr.˛.b0 // r.˛.a//j D ":
Thus q … U" ) q … W" , or equivalently W" U" .

Conversely, suppose q 2 U" . Then q is connected to P by a geodesic seg-
ment of length r.q/, so dg .q; P / r.q/. To prove the reverse inequality, suppose
˛ W Œa; b ! M is any admissible curve starting at a point of P and ending at q. If
˛.t / remains in U for all t 2 Œa; b, then Lemma 6.33 shows that

Lg .˛/ Lg ˛jŒa0 ;b jr..b// r..a0 //j D r.q/;
where a0 is the last time that ˛.a0 / 2 P . On the other hand, if ˛.t / does not remain
in U , then there is a first time b0 such that ˛.b0 / 2 @U" , and the argument in the
preceding paragraph shows that Lg .˛/ " > r.q/. Thus dg .q; P / D r.q/ for all
q 2 U" . Since r.q/ < " for all such q, it follows also that U" W" . t
u
Semigeodesic Coordinates
Local distance functions can be used to build coordinate charts near submanifolds
in which the metric has a particularly simple form. We begin by describing the kind
of coordinates we are looking for.
1Let .M;n g/ be an n-dimensional Riemannian manifold. Smooth local coordinates
x ; : : : ; x on an open subset
U M are called semigeodesic coordinates if each
x n -coordinate curve t 7! x 1 ; : : : ; x n1 ; t is a unit-speed geodesic that meets each
level set of x n orthogonally.
Because of the distinguished role played by the last coordinate function, through-
out the rest of this section we will use the summation convention with Latin indices
running from 1 to n and Greek indices running from 1 to n 1.
We will see below that semigeodesic coordinates are easy to construct. But first,
let us develop some alternative characterizations of them.
Proposition 6.41 (Characterizations of Semigeodesic

Coordinates). Let .M; g/
be a Riemannian n-manifold, and let x 1 ; : : : ; x n be smooth coordinates on an open
subset of M . The following are equivalent:

(a) x i are semigeodesic coordinates.
(b) j@n jg 1 and h@˛ ; @n i g 0 for ˛ D 1; : : : ; n 1.
(c) jdx n jg 1 and hdx ˛ ; dx n i g 0 for ˛ D 1; : : : ; n 1.
(d) j grad x n jg 1 and hgrad x ˛ ; grad x n i g 0 for ˛ D 1; : : : ; n 1.
(e) x n is a local distance function and x 1 ; : : : ; x n1 are constant along the integral
curves of grad x n .
(f) grad x n @n .
Proof. We begin by showing that (b) , (c) , (d) , (e) and (c) ,(f). Note that (b)
is equivalent to the coordinate matrix of g having the block form 0 10 , where the
asterisk represents an arbitrary .n 1/ .n 1/ positive definite symmetric matrix,
while (c) is equivalent to the inverse matrix having the same form. It follows from
Cramer’s rule that the matrix of g has this form if and only if its inverse does, and
thus (b) is equivalent to (c).
The equivalence of (c) and (d) follows from the definitions of the gradient and of
the inner product on 1-forms: for all i; j D 1; : : : ; n,
˝ i ˛ ˝ ˛ ˝ ˛
dx ; dx j g D .dx i /# ; .dx j /# g D grad x i ; grad x j g :
The equivalence of (d) and (e) also follows from the definition of the gradient:
hgrad x ˛ ; grad x n i g D dx ˛ grad x n D .grad x n /.x ˛ / for each ˛, which means that
x ˛ is constant along the grad x n integral curves if and only if hgrad x ˛ ; grad x n i g D
0. Finally, by examining the individual components of the coordinate formula
grad x n D g nj @j , we see that (c) is also equivalent to (f).
To complete the proof, we show that (a) , (b). Assume first that (a) holds.
Because the x n -coordinate curves have unit speed, it follows that j@n jg 1. The
tangent space to any x n -level set is spanned at each point by @1 ; : : : ; @n1 , and (a)
guarantees that @n is orthogonal to each of these, showing that (b) holds. Con-
versely, if we assume (b), the first part of the proof shows that (f) holds as well,
so j grad x n jg j@n jg 1, showing that x n is a local distance function. Thus the
x n -coordinate curves are also integral curves of grad x n and hence are unit-speed
geodesics by Theorem 6.32. The fact that h@˛ ; @n i D 0 for ˛ D 2; : : : ; n implies that
these geodesics are orthogonal to the level sets of x n , thus proving (a). t
u
Part (b) of this proposition leads to the following simplified coordinate represen-
tations for the metric and Christoffel symbols in semigeodesic coordinates. Recall
that implied summations with Greek indices run from 1 to n 1.
Semigeodesic Coordinates 183

Corollary 6.42. Let x i be semigeodesic coordinates on an open subset of a Rie-
mannian n-manifold .M; g/.
(a) The metric has the following coordinate expression:
2
g D dx n C g˛ˇ x 1 ; : : : ; x n dx ˛ dx ˇ :
(b) The Christoffel symbols of g have the following coordinate expressions:

n ˛ n n
nn D nn D ˛n D n˛ D 0;
n
˛ˇ D 12 @n g˛ˇ ;
ˇ ˇ
(6.20)
n˛ D ˛n D 12 g ˇ @n g˛ ;
D y ;
˛ˇ ˛ˇ
y are the Christoffel symbols

where for each fixed value of x n , the quantities ˛ˇ
in .x / coordinates for the induced metric gy on the level set x n D constant.
˛
Proof. Part (a) follows immediately from part (b) of Proposition 6.41, and (b) is
proved by inserting gnn D 1 and g˛n D gn˛ D 0 into formula (5.8) for the Christoffel
symbols. t
u
Proposition 6.41(e) gives us an effective way to construct semigeodesic coordi-
nates: if r is any smooth local distance function (for example, the distance from
a point or a smooth submanifold), just set x n D r, choose any local coordinates
x 1 ; : : : ; x n1 for a level set of r, and then extend them to be constant along the
integral curves of grad r. Here are some explicit examples.
Example 6.43 (Fermi Coordinates for a Hypersurface). Suppose P is anembed-

ded hypersurface in a Riemannian manifold .M; g/, and let x 1 ; : : : ; x n1 ; v be any
Fermi coordinates for P on an open subset
U M (see (5.25)). In this case, the
function r defined by (6.17) is just r x 1 ; : : : ; x n1 ; v D .v 2 /1=2 D jvj, so v is a
local distance function on U X P . It follows from Corollary 6.39 that there is a
neighborhood U0 of P on which jvj is equal to the distance from P . Moreover,
(6.18) reduces to @r D ˙@=@v , which is equal to grad jvj by Corollary 6.39, so it
follows from Proposition 6.41(f) that Fermi coordinates for a hypersurface are auto-
matically semigeodesic coordinates. //
Example 6.44 (Boundary Normal Coordinates). Suppose .M; g/ is a smooth Rie-

mannian manifold with nonempty boundary. The results in this chapter do not
apply directly to manifolds with boundary, but we can embed M in its double M
(Prop. A.31), extend the metric smoothly to M , and construct Fermi coordinates
1
x ; : : : ; x n1 ; v for @M in M . By replacing v with v if necessary, we can arrange
that v>0 in Int M , and then these Fermi coordinates restrict to smooth boundary
coordinates for M that are also semigeodesic coordinates. Such coordinates
are called boundary normal coordinates for M . //
Fig. 6.19: Polar normal coordinates
Example 6.45 (Polar Normal Coordinates). Polar coordinates for Rn are con-
structed by choosing a smooth local parametrization y W Uy ! U Sn1 for an
open subset U of Sn1 , and defining y W Uy RC ! Rn by y 1 ; : : : ; n1 ; r D

r y 1 ; : : : ; n1 . It is straightforward to show that the differential of y van-
y y 1 is a smooth coordinate map on the open subset
ishes nowhere,
nD
so
y y C
U D U R R X f0g. Familiar examples are ordinary polar coordinates in
the plane and spherical coordinates in R3 . Such coordinates have the property
that the last coordinate function is r.x/ D jxj.
Now let .M; g/ be a Riemannian n-manifold, p a point in M , and ' any normal
coordinate chart defined
on a normal neighborhood V of p. For every choice of polar
coordinates U; y for Rn X f0g, we obtain a smooth coordinate map D y ı ' on
an open subset of V X fpg (see Fig. 6.19). Such coordinates are called polar normal
coordinates. They have the property that the last coordinate function r is the radial
distance function on V , and the other coordinates are constant along the integral
curves of grad r, so they are semigeodesic coordinates. //
Example 6.46 (Polar Fermi Coordinates). Now let P be an embedded subman-

ifold of .M; g/, and let ' D x 1 ; : : : ; x k ; v 1 ; : : : ; v nk be Fermi coordinates on a
neighborhood U0 of a point p 2 P . Then anypolar coordinate map y for Rnk
can be applied to the variables v 1 ; : : : ; v nk to yield a coordinate chart D

IdRk y ı ' on an open subset of U0 X P , taking values in Rk Rnk1 RC .

If we write the coordinate functions as x 1 ; : : : ; x k ; 1 ; : : : ; nk ; r , it follows from

Proposition 5.26 that each coordinate curve t 7! x 1 ; : : : ; x k ; 1 ; : : : ; nk ; t is a
unit-speed geodesic. Thus these are semigeodesic coordinates, called polar Fermi
coordinates. The polar normal coordinates described above are just the special case
P D fpg. //
Problems 185
Problems
6-1. Suppose M is a nonempty connected Riemannian 1-manifold. Show that if

M is noncompact, then it is isometric to an open interval in R with the Euclidean
metric, while if it is compact, it is isometric to a circle S1 .R/ D fx 2 R2 W
jxj D Rg with its induced metric for some R > 0, using the following steps.
(a) Let W I ! M be any maximal unit-speed geodesic. Show that its im-
age is open and closed, and therefore is surjective.
(b) Show that if is injective, then it is an isometry between I with its
Euclidean metric and M .
(c) Now suppose .t1 / D .t2 / for some t1 ¤ t2 . In case 0 .t1 / D 0 .t2 /,
show that is periodic, and descends to a global isometry from an
appropriate circle to M .
(d) It remains only to rule out the case .t1 / D .t2 / and 0 .t1 / D 0 .t2 /.
If this occurs, let t0 D .t1 C t2 /=2, and define geodesics ˛ and ˇ by
˛.t / D .t0 C t /; ˇ.t / D .t0 t /:
Use uniqueness of geodesics to conclude that ˛ ˇ on their common

domain, and show that this contradicts the fact that is injective on some
neighborhood of t0 .
6-2. Let n be a positive integer and R a positive real number.
(a) Prove that the Riemannian distance between any two points p; q in
Sn .R/ with the round metric is given by
hp; qi
dgV R .p; q/ D R arccos ;
R2
where h; i is the Euclideaninner product
on R
nC1
.
(b) Prove that the metric space S .R/; dgV R has diameter R.
n
(Used on pp. 39, 359.)

6-3. Let n be a positive integer and R a positive real number. Prove that the
Riemannian
n distance between any two points in the Poincaré ball model
B .R/; gM R of hyperbolic space of radius R is given by
!
2R2 jp qj2
dgM R .p; q/ D R arccosh 1 C 2 ;
R jpj2 R2 jqj2
where j j represents the Euclidean norm in Rn . [Hint: First use the result of
Problem 3-5 to show that it suffices to consider the case R D 1. Then use a
rotation to reduce to the case n D 2, and use the group action of Problem 3-8
to show that it suffices to consider the case in which p is the origin.]
6-4. In Chapter 2, we started with a Riemannian metric and used it to define the
Riemannian distance function. This problem shows how to go back the other
way: the distance function determines the Riemannian metric. Let .M; g/ be
a connected Riemannian manifold.
(a) Show that if W ."; "/ ! M is any smooth curve, then
dg ..0/; .t //
j 0 .0/jg D lim :
t&0 t
(b) Show that if g and gz are two Riemannian metrics on M such that
dg .p; q/ D dgz .p; q/ for all p; q 2 M , then g D gz.
6-5. Prove Theorem 6.17 (sufficiently small geodesic balls are geodesically con-
vex) as follows.
(a) Let .M; g/ be a Riemannian manifold, let p 2 M be fixed, and let W
be a uniformly normal neighborhood of p. For " > 0 small enough that
B3" .p/ W , define a subset W" TM R by
W" D f.q; v; t / 2 TM R W q 2 B" .p/; v 2 Tq M; jvj D 1; jt j < 2"g:
Define f W W" ! R by
f .q; v; t / D dg .p; expq .t v//2 :
Show that f is smooth. [Hint: Use normal coordinates centered at p.]

(b) Show that if " is chosen small enough, then @2 f =@t 2 > 0 on W" . [Hint:
Compute f .p; v; t / explicitly and use continuity. Be careful to verify that
" can be chosen independently of v.]
(c) Suppose " is chosen as in (b). Show that if q1 ; q2 2 B" .p/ and is a
minimizing geodesic segment from q1 to q2 , then dg .p; .t // attains its
maximum at one of the endpoints of .
(d) Show that B" .p/ is geodesically convex.
6-6. Suppose .M; g/ is a Riemannian manifold. For each x 2 M , define the con-
vexity radius of M at x, denoted by conv.x/, to be the supremum of all " > 0
such that there is a geodesically convex geodesic ball of radius " centered at
x. Show that conv.x/ is a continuous function of x.
6-7. We now have two kinds of “metrics” on a connected Riemannian manifold:
the Riemannian metric and the distance function. Correspondingly, there are
two definitions of “isometry” between connected Riemannian manifolds: a
Riemannian isometry is a diffeomorphism that pulls one Riemannian metric
back to the other, and a metric isometry is a homeomorphism that preserves
distances. Proposition 2.51 shows that every Riemannian isometry is a metric
This problem outlines a proof of the converse. Suppose .M; g/ and
isometry.

M ; gz are connected Riemannian manifolds, and ' W M ! M is a metric
isometry.
Problems 187
(a) Show that for every p 2 M and v; w 2 Tp M , we have
dg .expp t v; expp t w/
lim D jv wjg :
t!0 t
[Hint: Use the Taylor series for g in Riemannian normal coordinates on
a convex geodesic ball centered at p.]
(b) Show that ' takes geodesics to geodesics.
(c) For each p 2 M , show that there exist an open ball B" .0/ Tp M
and a continuous map W B" .0/ ! T'.p/ M satisfying .0/ D 0 and
exp'.p/ .v/ D '.expp v/ for all v 2 B" .0/.
(d) With as above, show that j .v/ .w/jgz D jv wjg for all v; w 2
B" .0/, and conclude from Problem 2-2 that is the restriction of a
linear isometry.
(e) With p and as above, show that ' is smooth on a neighborhood of p
and d'p D .
(f) Conclude that ' is a Riemannian isometry.

6-8. Suppose .M; g/ and M ; gz are connected Riemannian manifolds (not nec-
essarily complete), and for each i 2 ZC , 'i W M ! M is a Riemannian isom-
etry such that 'i converges pointwise to a map ' W M ! M . Show that '
is a Riemannian isometry. [Hint: Once you have shown that ' is a local
isometry, to show that ' is surjective, suppose y is a limit point of '.M /.
Choose x 2 M such that '.x/ lies in a uniformly normal neighborhood of
sequence of points .xi ; vi / 2 TM
y, and show that there exists a convergent
such that 'i .xi / D '.x/ and 'i expxi vi D y.]
6-9. Suppose .M; g/ is a (not necessarily connected) Riemannian manifold. In
Problem 2-30, you were asked to show that there is a distance function
d W M M ! R that induces the given topology and restricts to the Rie-
mannian distance on each connected component of M . Show that if each
component of M is geodesically complete, then d can be chosen so that
.M; d / is a complete metric space. Show also that if M has infinitely many
components, then d can be chosen so that .M; d / is not complete.
6-10. A curve W Œ0; b/ ! M (with 0 < b 1) is said to diverge to infinity if
for every compact set K M , there is a time T 2 Œ0; b/ such that .t / … K
for t > T . (For those who are familiar with one-point compactifications, this
means that .t / converges to the “point at infinity” in the one-point com-
pactification of M as t % b.) Prove that a connected Riemannian manifold is
complete if and only if every regular curve that diverges to infinity has
infinite length. (The length of a curve whose domain is not compact is just
the supremum of the lengths of its restrictions to compact subintervals.)
6-11. Suppose .M; g/ is a connected Riemannian manifold, P M is a connected
embedded submanifold, and gy is the induced Riemannian metric on P .
(a) Show that dgy .p; q/ dg .p; q/ for p; q 2 P .

(b) Prove that if .M; g/ is complete and P is closed in M , then P; gy is
complete.
(c) Give an example of a complete Riemannian manifold .M; g/ and a con-
nected embedded submanifold P M that is complete but not closed
in M .
6-12. Let .M; g/ be a connected Riemannian manifold.
(a) Suppose there exists ı > 0 such that for each p 2 M , every maximal
unit-speed geodesic starting at p is defined at least on an interval of the
form .ı; ı/. Prove that M is complete.
(b) Prove that if M has positive or infinite injectivity radius, then it is com-
plete.
(c) Prove that if M is homogeneous, then it is complete.
(d) Give an example of a complete, connected Riemannian manifold that
has zero injectivity radius.
6-13. Let G be a connected compact Lie group. Show that the Lie group expo-
nential map of G is surjective. [Hint: Use Problem 5-8.]
6-14. Let .M; g/ be a connected Riemannian manifold.
(a) Show that M is complete if and only if the compact subsets of M are
exactly the closed and bounded ones.
(b) Show that M is compact if and only if it is complete and bounded.
6-15. Let S be the unit 2-sphere minus its north and south poles, and let M D
.0; / R. Define q W M ! S by q.'; / D .sin ' cos ; sin ' sin ; cos '/,
and let g be the metric on M given by pulling back the round metric: g D
q g.
V (Think of M as an infinitely long onion skin wrapping infinitely many
times around an onion.) Prove that the Riemannian manifold .M; g/ has diam-
eter , but contains infinitely long properly embedded geodesics.
6-16. Suppose .M; g/ is a complete, connected Riemannian manifold with positive
or infinite injectivity radius.
(a) Let
2 .0; 1 denote the injectivity radius of M , and define T M to be
the subset of TM consisting of vectors of length less than
, and D to
be the subset f.p; q/ W dg .p; q/ <
g M M . Define E W T M ! D
by E.x; v/ D .x; expx v/. Prove that E is a diffeomorphism.
(b) Use part (a) to prove that if B is a topological space and F; G W B ! M
are continuous maps such that dg .F .x/; G.x// < inj.M / for all x 2 B,
then F and G are homotopic.
6-17. Prove Proposition 6.28 (existence of a closed geodesic in a free homotopy
class). [Hint: Use Prop. 6.25 to show that the given free homotopy class is
represented by a geodesic loop, i.e., a geodesic whose starting and ending
points are the same. Show that the lengths of such loops have a positive great-
est lower bound; then choose a sequence of geodesic loops whose lengths
approach that lower bound, and show that a subsequence converges uniformly
Problems 189
to a geodesic loop whose length is equal to the lower bound. Use Problem
6-16 to show that the limiting curve is in the given free homotopy class, and
apply the first variation formula to show that the limiting curve is in fact a
closed geodesic.]
6-18. A connected Riemannian manifold .M; g/ is said to be k-point homoge-
neous if for any two ordered k-tuples .p1 ; : : : ; pk / and .q1 ; : : : ; qk / of points
in M such that dg .pi ; pj / D dg .qi ; qj / for all i; j , there is an isometry
' W M ! M such that '.pi / D qi for i D 1; : : : ; k. Show that .M; g/ is 2-point
homogeneous if and only if it is isotropic. [Hint: Assuming that M is isotropic,
first show that it is homogeneous by considering the midpoint of a geodesic
segment joining sufficiently nearby points p; q 2 M , and then use the result
of Problem 6-12(c) to show that it is complete.] (Used on pp. 56, 261.)
6-19. Prove that every Riemannian symmetric space is homogeneous. [Hint: Pro-
ceed as in Problem 6-18.] (Used on p. 78.)
6-20. A connected Riemannian manifold is said to be extendible if it is isometric
to a proper open subset of a larger connected Riemannian manifold.
(a) Show that every complete, connected Riemannian manifold is nonex-
tendible.
(b) Show that the converse is false by giving an example of a nonextendible
connected Riemannian manifold that is not complete.
6-21. Let .M; g/ be a complete, connected, noncompact Riemannian manifold.
Define a ray in M to be a geodesic whose domain is Œ0; 1/, and such that
the restriction of to Œ0; b is minimizing for every b > 0. Prove that for each
p 2 M there is a ray in M starting at p.
6-22. Let .M; g/ be a connected Riemannian manifold with boundary. Prove that
.M; dg / is a complete metric space if and only if the following condition
holds: for every geodesic W Œ0; b/ ! M that cannot be extended to a geodesic
on any interval Œ0; b 0 / with b 0 > b, .t / converges to a point of @M as t % b.
6-23. In some treatments of Riemannian geometry, instead of minimizing the
length functional, one considers the following energy functional for an ad-
missible curve W Œa; b ! M :
Z b
E. / D 1
2
j 0 .t /j2 dt:
a
(Note that E. / is not independent of parametrization.)

(a) Prove that an admissible curve is a critical point for E (with respect
to proper variations) if and only if it is a geodesic (which means, in
particular, that it has constant speed).
(b) Prove that if is an admissible curve that minimizes energy among
admissible curves with the same endpoints, then it also minimizes length.
(c) Prove that if is an admissible curve that minimizes length among

admissible curves with the same endpoints, then it minimizes energy if
and only if it has constant speed.
[Remark: For our limited purposes, it is easier and more straightforward to

use the length functional. But because the energy functional does not involve
the square root function, and its critical points automatically have constant-
speed parametrizations, it is sometimes more useful for proving the existence
of geodesics with certain properties.]
6-24. Let .M; g/ be a Riemannian manifold. Recall that a vector field X 2 X.M /
is called a Killing vector field if LX g D 0 (see Problem 5-22).
(a) Prove that a Killing vector field that is normal to a geodesic at one point
is normal everywhere along the geodesic.
(b) Prove that if a Killing vector field vanishes at a point p, then it is tangent
to geodesic spheres centered at p.
(c) Prove that a Killing vector field on an odd-dimensional manifold cannot
have an isolated zero.
(Used on p. 315.)
6-25. Suppose .M; g/ is a Riemannian manifold and f W M ! R is a smooth local
distance function. Prove that f is a Riemannian submersion.
6-26. Suppose .M; g/ is a complete, connected Riemannian manifold and S M
is a closed subset.
(a) Show that for every p 2 M X S , there is a geodesic segment from p to a
point of S whose length is equal to dg .p; S /.
(b) In case S is a properly embedded submanifold of M , show that every
geodesic segment satisfying the conclusion of (a) intersects S orthogo-
nally.
(c) Give a counterexample to (a) if S is not closed.
6-27. Suppose .M; g/ is a complete, connected Riemannian manifold, S M is a
closed subset, and f W M ! R is a continuous function such that f 1 .0/ D
S and f is smooth with unit gradient on M X S . Prove that f .x/ D dg .x; S /
for all x 2 M .
6-28. Suppose .M; g/ is a complete, connected Riemannian n-manifold whose
isometry group is a Lie group acting smoothly on M . (The Myers–Steenrod
theorem shows that this is always the case when M is connected, but we have
not proved this.) Prove that if G is a closed subgroup of Iso.M /, then G acts
properly on M , using the following outline. Suppose .'m / is a sequence in G
and .pm / is a sequence in M such that pm ! p0 and 'm .pm / ! q0 .
(a) Prove that 'm .p0 / ! q0 .
(b) Let Br .p0 / be a geodesic ball centered at p0 ; let 0 < " < r; let
.b1 ; : : : ; bn / be an orthonormal basis for Tp0 M ; and let pi D expp0 ."bi /
for i D 1; : : : ; n. Prove that there exist a linear isometry A W Tp0 M !
Problems 191
Tq0 M and a subsequence .'mj / such that 'mj .pi / ! expq0 ."Abi / for
i D 1; : : : ; n. [Hint: Use Prop. 5.20.]
(c) Prove that there is an isometry ' 2 G such that 'mj ! ' pointwise,
meaning that 'mj .x/ ! '.x/ for every x 2 M .
(d) Prove that 'mj ! ' in the topology of G. [Hint: Define a map F W G !
M nC1 by F . / D . .p0 /; .p1 /; : : : ; .pn //, where p1 ; : : : ; pn are
the points introduced in part (b). Show that F is continuous, injective,
and closed, and therefore is a homeomorphism onto its image.]
(Used on p. 349.)
6-29. Suppose .M; g/ is a Riemannian n-manifold that admits a nonzero parallel
vector field X . Show that for each p 2 M, there exist a neighborhood U of
p, a Riemannian .n 1/-manifold N, and an open interval I R with its
Euclidean metric such that U is isometric to the Riemannian product N I .
[Hint: It might be helpful to prove that the 1-form X [ is closed and to compute
the Lie derivative LX g.]
6-30. Suppose .M; g/ is a Riemannian n-manifold, p 2 M , and B" .p/ is a
geodesic ball centered at p. Prove that for every ı such that 0 < ı < ",
Z
ı
Vol Bxı .p/ D Area.@Br .p// dr;
0
where Area.@Br .p// represents the .n 1/-dimensional volume of @Br .p/

with its induced Riemannian metric.
6-31. Suppose .M; g/ is a Riemannian n-manifold, P M is a compact embedded
submanifold, and U is an "-tubular neighborhood of P for some " > 0. For
0 < ı < ", define Uı and Pı by
Uı D fx 2 U W dg .x; P / ıg;
Pı D fx 2 U W dg .x; P / D ıg:
(a) Prove that Uı is a regular domain and Pı is a compact, embedded sub-

manifold of M for each ı 2 .0; "/.
(b) Generalize the result of Problem 6-30 by proving that
Z ı
Vol.Uı / D Area.Pr / dr;
0
where Area.Pr / denotes the .n 1/-dimensional volume of Pr with the

induced Riemannian metric.
Chapter 7
Curvature
In this chapter, we begin our study of the local invariants of Riemannian metrics.
Starting with the question whether all Riemannian metrics are locally isometric,
we are led to a definition of the Riemannian curvature tensor as a measure of the
failure of second covariant derivatives to commute. Then we prove the main result
of this chapter: a Riemannian manifold has zero curvature if and only if it is flat, or
locally isometric to Euclidean space. Next, we derive the basic symmetries of the
curvature tensor, and introduce the Ricci, scalar, and Weyl curvature tensors. At the
end of the chapter, we explore how the curvature can be used to detect conformal
flatness. As you will see, the results of this chapter apply essentially unchanged to
pseudo-Riemannian metrics.
Local Invariants
For any geometric structure defined on smooth manifolds, it is of great interest to

address the local equivalence question: Are all examples of the structure locally
equivalent to each other (under an appropriate notion of local equivalence)?
There are some interesting and useful structures in differential geometry that
have the property that all such structures on manifolds of the same dimension
are locally equivalent to each other. For example:
N ONVANISHING VECTOR FIELDS : Every nonvanishing vector field can be writ-
ten as X D @=@x 1 in suitable local coordinates, so they are all locally equivalent.
R IEMANNIAN METRICS ON A 1- MANIFOLD : Problem 2-1 shows that every
Riemannian 1-manifold is locally isometric to R with its Euclidean metric.
S YMPLECTIC FORMS : A symplectic form on a smooth manifold M is a closed
2-form ! that is nondegenerate at each p 2 M , meaning that !p .v; w/ D 0 for
all w 2 Tp M only if v D 0. By the theorem of Darboux [LeeSM,PThm. 22.13],
every symplectic form can be written in suitable coordinates as i dx i ^ dy i .
Thus all symplectic forms on 2n-manifolds are locally equivalent.

194 7 Curvature
Fig. 7.1: Result of parallel transport along the x 1 -axis and the x 2 -coordinate lines
On the other hand, Problem 5-5 showed that the round 2-sphere and the Euclidean
plane are not locally isometric.
The most important technique for proving that two geometric structures are not
locally equivalent is to find local invariants, which are quantities that must be pre-
served by local equivalences. In order to address the general problem of local equiv-
alence of Riemannian or pseudo-Riemannian metrics, we will define a local invari-
ant for all such metrics called curvature. Initially, its definition will have nothing to
do with the curvature of curves described in Chapter 1, but later we will see that the
two concepts are intimately related.
To motivate the definition of curvature, let us look back at the argument outlined
in Problem 5-5 for showing that the sphere and the plane are not locally isometric.
The key idea is that every tangent vector in the plane can be extended to a parallel
vector field, so every Riemannian manifold that is locally isometric to R2 must have
the same property locally.
Given a Riemannian 2-manifold M , here is one way to attempt to construct a
parallel extension of a vector z 2 Tp M : working in any smooth local coordinates
.x 1 ; x 2 / centered at p, first parallel transport z along the x 1 -axis, and then parallel
transport the resulting vectors along the coordinate lines parallel to the x 2 -axis (Fig.
7.1). The result is a vector field Z that, by construction, is parallel along every x 2 -
coordinate line and along the x 1 -axis. The question is whether this vector field is
parallel along x 1 -coordinate lines other than the x 1 -axis, or in other words, whether
r@1 Z 0. Observe that r@1 Z vanishes when x 2 D 0. If we could show that
r@2 r@1 Z D 0; (7.1)
then it would follow that r@1 Z 0, because the zero vector field is the unique
parallel transport of zero along the x 2 -curves. If we knew that
Local Invariants 195
r@2 r@1 Z D r@1 r@2 Z; (7.2)
then (7.1) would follow immediately, because r@2 Z D 0 everywhere by construc-

tion. Indeed, on R2 with the Euclidean metric, direct computation shows that

x@ r
r x ZDr x @ @1 Z k @k
2 @1 2

D @2 @1 Z k @k ;
and r x@ r x Z is equal to the same thing, because ordinary second partial deriva-
1 @2
tives commute. However, (7.2) might not hold for an arbitrary Riemannian metric;
indeed, it is precisely the noncommutativity of such second covariant derivatives
that forces this construction to fail on the sphere. Lurking behind this noncommuta-
tivity is the fact that the sphere is “curved.”
To express this noncommutativity in a coordinate-independent way, let us look
more closely at the quantity r xX r xY r
xY Z r x X Z when X , Y , and Z are smooth
vector fields. On R2 with the Euclidean connection, we just showed that this always
vanishes if X D @1 and Y D @2 ; however, for arbitrary vector fields this may no
longer be true. In fact, in Rn with the Euclidean connection we have

xX r
r xY Z D r x X Y Z k @k D X Y Z k @k ;

and similarly rx r x Z D YX Z k @k . The difference between these two expressions
k Y X k
is X Y Z YX Z @k D r x ŒX;Y Z. Therefore, the following relation holds for
all vector fields X; Y; Z defined on an open subset of Rn :
xX r
r xY r
xY Z r x ŒX;Y Z:
xX Z D r
Recall that a Riemannian manifold is said to be flat if it is locally isometric

to a Euclidean space, that is, if every point has a neighborhood that is isometric
to an open set in Rn with its Euclidean metric. Similarly, a pseudo-Riemannian
manifold is flat if it is locally isometric to a pseudo-Euclidean space. The compu-
tation above leads to the following simple necessary condition for a Riemannian or
pseudo-Riemannian manifold to be flat. We say that a connection r on a smooth
manifold M satisfies the flatness criterion if whenever X; Y; Z are smooth vector
fields defined on an open subset of M , the following identity holds:
rX rY Z rY rX Z D rŒX;Y Z: (7.3)
Example 7.1. The metric on the n-torus induced by the embedding in R2n given in
Example 2.21 is flat, because each point has a coordinate neighborhood in which
the metric is Euclidean. //
Proposition 7.2. If .M; g/ is a flat Riemannian or pseudo-Riemannian manifold,

then its Levi-Civita connection satisfies the flatness criterion.
196 7 Curvature
Proof. We just showed that the Euclidean connection on Rn satisfies (7.3). By nat-
urality, the Levi-Civita connection on every manifold that is locally isometric to a
Euclidean or pseudo-Euclidean space must also satisfy the same identity. t
u
The Curvature Tensor

Motivated by the computation in the preceding section, we make the following def-
inition. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold, and define a
map R W X.M / X.M / X.M / ! X.M / by
R.X; Y /Z D rX rY Z rY rX Z rŒX;Y Z:
Proposition 7.3. The map R defined above is multilinear over C 1 .M /, and thus
defines a .1; 3/-tensor field on M .
Proof. The map R is obviously multilinear over R. For f 2 C 1 .M /,
R.X; f Y /Z D rX rf Y Z rf Y rX Z rŒX;f Y Z
D rX .f rY Z/ f rY rX Z rf ŒX;Y C.Xf /Y Z
D .Xf /rY Z C f rX rY Z f rY rX Z
f rŒX;Y Z .Xf /rY Z
D fR.X; Y /Z:
The same proof shows that R is linear over C 1 .M / in X , because R.X; Y /Z D

R.Y; X /Z from the definition. The remaining case to be checked is linearity over
C 1 .M / in Z; this is left to Problem 7-1.
By the tensor characterization lemma (Lemma B.6), the fact that R is multilinear
over C 1 .M / implies that it is a .1; 3/-tensor field. t
u
Thanks to this proposition, for each pair of vector fields X; Y 2 X.M /, the map
R.X; Y / W X.M / ! X.M / given by Z 7! R.X; Y /Z is a smooth bundle endomor-
phism of TM , called the curvature endomorphism determined by X and Y.
The tensor field R itself is called the (Riemann) curvature endomorphism or the
(1, 3)-curvature tensor. (Some authors call it simply the curvature tensor, but we
reserve that name instead for another closely related tensor field, defined below.)
As a .1; 3/-tensor field, the curvature endomorphism can be written in terms of
any local frame with one upper and three lower indices. We adopt the convention
that the last index is the contravariant (upper) one. (This is contrary to our default
assumption that covector arguments come first.) Thus, for example,
the curvature
endomorphism can be written in terms of local coordinates x i as
R D Rij k l dx i ˝ dx j ˝ dx k ˝ @l ;
where the coefficients Rij k l are defined by

The Curvature Tensor 197

R @i ; @j @k D Rij k l @l :
The next proposition shows how to compute the components of R in coordinates.
Proposition 7.4. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold. In

terms of any smooth local coordinates, the components of the .1; 3/-curvature tensor
are given by
Rij k l D @i jl k @j ik
l
C jmk im
l m l
ik j m : (7.4)
u
Importantly for our purposes, the curvature endomorphism also measures the
failure of second covariant derivatives along families of curves to commute. Given a
smooth one-parameter family of curves W J I ! M , recall from Chapter 6 that
the velocity fields @t .s; t / D .s /0 .t / and @s .s; t / D .t/0 .s/ are smooth vector
fields along .
Proposition 7.5. Suppose .M; g/ is a smooth Riemannian or pseudo-Riemannian
manifold and W J I ! M is a smooth one-parameter family of curves in M .
Then for every smooth vector field V along ,
Ds Dt V Dt Ds V D R.@s ; @t /V: (7.5)
Proof. This is a local question, so for each .s; t / 2 J I , we can choose smooth
coordinates x i defined on a neighborhood of .s; t / and write
ˇ
.s; t / D 1 .s; t /; : : : ; n .s; t / ; V .s; t / D V j .s; t /@j ˇ .s;t/ :
Formula (4.15) yields

@V i
@i C V i Dt @i :
Dt V D
@t
Therefore, applying (4.15) again, we get
@2 V i @V i @V i
Ds Dt V D @i C Ds @i C Dt @i C V i Ds Dt @i :
@s@t @t @s
Interchanging s and t and subtracting, we see that all the terms except the last cancel:
Ds Dt V Dt Ds V D V i .Ds Dt @i Dt Ds @i / : (7.6)
Now we need to compute the commutator in parentheses. For brevity, let us write
@ k @ j
S D @s D @k I T D @t D @j :
@s @t
Because @i is extendible,
@ j
Dt @i D rT @i D r@j @i ;
@t
198 7 Curvature
and therefore, because r@j @i is also extendible,

@ j
Ds Dt @i D Ds r@j @i
@t
@2 j @ j
D r@j @i C rS r@j @i
@s@t @t
@2 j @ j @ k
D r@ @i C r@k r@j @i :
@s@t j @t @s
Interchanging s $ t and j $ k and subtracting, we find that the first terms cancel,
and we get
@ j @ k
Ds Dt @i Dt Ds @i D r@k r@j @i r@j r@k @i
@t @s
@ j @ k
D R.@k ; @j /@i D R.S; T /@i :
@t @s
Finally, inserting this into (7.6) yields the result. t
u
For many purposes, the information contained in the curvature endomorphism is
much more conveniently encoded in the form of a covariant 4-tensor. We define the
(Riemann) curvature tensor to be the .0; 4/-tensor field Rm D R[ (also denoted by
Riem by some authors) obtained from the .1; 3/-curvature tensor R by lowering its
last index. Its action on vector fields is given by
Rm.X; Y; Z; W / D hR.X; Y /Z; W ig : (7.7)
In terms of any smooth local coordinates it is written
Rm D Rij kl dx i ˝ dx j ˝ dx k ˝ dx l ;
where Rij kl D glm Rij k m . Thus (7.4) yields

p m
Rij kl D glm @i jmk @j ikm
C jpk ip
m
ik jp : (7.8)
It is appropriate to note here that there is much variation in the literature with
respect to index positions in the definitions of the curvature endomorphism and
curvature tensor. While almost all authors define the .1; 3/-curvature tensor as we
have, there are a few (notably [dC92, GHL04]) whose definition is the negative of
ours. There is much less agreement on the definition of the .0; 4/-curvature tensor:
whichever definition is chosen for the curvature endomorphism, you will see the
curvature tensor defined as in (7.7) but with various permutations of .X; Y; Z; W /
on the right-hand side. After applying the symmetries of the curvature tensor that we
will prove later in this chapter, however, all of the definitions agree up to sign. There
are various arguments to support one choice or another; we have made a choice that
makes equation (7.7) easy to remember. You just have to be careful when you begin
reading any book or article to determine the author’s sign convention.
Flat Manifolds 199
The next proposition gives one reason for our interest in the curvature tensor.
Proposition
7.6. The curvature tensor is a local isometry invariant: if .M; g/ and
M ; gz are Riemannian or pseudo-Riemannian manifolds and ' W M ! M is a
f D Rm.
local isometry, then ' Rm
Flat Manifolds
To give a qualitative geometric interpretation to the curvature tensor, we will show
that it is precisely the obstruction to being locally isometric to Euclidean (or pseudo-
Euclidean) space. (In Chapter 8, after we have developed more machinery, we will
be able to give a far more detailed quantitative interpretation.) The crux of the proof
is the following lemma.
Lemma 7.8. Suppose M is a smooth manifold, and r is any connection on M sat-

isfying the flatness criterion. Given p 2 M and any vector v 2 Tp M , there exists a
parallel vector field V on a neighborhood of p such that Vp D v.

Proof. Let p 2 M and v 2 Tp M be arbitrary, and let x 1 ; : : : ; x n be any smooth
coordinates for M centered at p. By shrinking the coordinate neighborhood if
necessary, we may assume that the image of the coordinate map is an open cube
C" D fx W jx i j < "; i D 1; : : : ; ng. We use the coordinate map to identify the coordi-
nate domain with C" .
Begin by parallel transporting v along the x 1 -axis; then from each point on the
x -axis, parallel transport along the coordinate line parallel to the x 2 -axis; then
1
successively parallel transport along coordinate lines parallel to the x 3 through x n -

axes (Fig. 7.2). The result is a vector field V defined in C" . The fact that V is
smooth follows from an inductive application of the theorem concerning smooth
dependence of solutions to linear ODEs on initial conditions (Thm. 4.31); the details
are left as an exercise.
Since rX V is linear over C 1 .M / in X , to show that V is parallel, it suffices
to show that r@i V D 0 for each i D 1; : : : ; n. By construction, r@1 V D 0 on the
x 1 -axis, r@2 V D 0 on the .x 1 ; x 2 /-plane, and in general r@k V D 0 on the slice
Mk C" defined by x kC1 D D x n D 0. We will prove the following fact by
induction on k:
r@1 V D D r@k V D 0 on Mk : (7.9)
For k D 1, this is true by construction, and for k D n, it means that V is parallel
on the whole cube C" . So assume that (7.9) holds for some k. By construction,
r@kC1 V D 0 on all of MkC1 , and for i k, the inductive hypothesis shows that
r@i V D 0 on the hyperplane Mk MkC1 .
Since Œ@kC1 ; @i D 0, the flatness criterion gives
200 7 Curvature
Fig. 7.2: Proof of Lemma 7.8
r@kC1 .r@i V / D r@i .r@kC1 V / D 0 on MkC1 :
This shows that r@i V is parallel along the x kC1 -curves starting on Mk . Since r@i V
vanishes on Mk and the zero vector field is the unique parallel transport of zero, we
conclude that r@i V is zero on each x kC1 -curve. Since every point of MkC1 is on
one of these curves, it follows that r@i V D 0 on all of MkC1 . This completes the
inductive step to show that V is parallel. t
u
I Exercise 7.9. Prove that the vector field V constructed in the preceding proof is
smooth.
Theorem 7.10. A Riemannian or pseudo-Riemannian manifold is flat if and only if

its curvature tensor vanishes identically.
Proof. One direction is immediate: Proposition 7.2 showed that the Levi-Civita con-
nection of a flat metric satisfies the flatness criterion, so its curvature endomorphism
is identically zero, which implies that the curvature tensor is also zero.
Now suppose .M; g/ has vanishing curvature tensor. This means that the cur-
vature endomorphism vanishes as well, so the Levi-Civita connection satisfies the
flatness criterion. We begin by showing that g shares one important property with
Euclidean and pseudo-Euclidean metrics: it admits a parallel orthonormal frame in
a neighborhood of each point.
Let p 2 M , and choose an orthonormal basis .b1 ; : : : ; bn / for Tp M . In the
pseudo-Riemannian case, we may assume that the basis is in standard order (with posi-
tive entries before negative ones in the matrix gij D gp .bi ; bj /). Lemma 7.8 shows
that there exist parallel vector fields E1 ; : : : ; En on a neighborhood U of p such that
Ei jp D bi for each i D 1; : : : ; n. Because parallel transport preserves inner products,
the vector fields .Ej / are orthonormal (and hence linearly independent) in all of U .
Because the Levi-Civita connection is symmetric, we have
Flat Manifolds 201
Fig. 7.3: The curvature endomorphism and parallel transport around a closed loop
ŒEi ; Ej D rEi Ej rEj Ei D 0:
Thus the vector fields .E1 ; : : : ; En / form a commuting orthonormal frame on U . The
canonical form theorem
forcommuting vector fields (Prop. A.48) shows that there
are coordinates y 1 ; : : : ; y n on a (possibly smaller) neighborhood of p such that
Ei D @=@y i for i D 1; : : : ; n. In any such coordinates, gij D g.@i ; @j / D g.Ei ; Ej / D
˙ıij , so the map y D .y 1 ; : : : ; y n / is an isometry from a neighborhood of p to an
open subset of the appropriate Euclidean or pseudo-Euclidean space. t
u
Using similar ideas, we can give a more precise interpretation of the meaning of
the curvature tensor: it is a measure of the extent to which parallel transport around
a small rectangle fails to be the identity map.
Theorem 7.11. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold; let I

be an open interval containing 0; let W I I ! M be a smooth one-parameter
family of curves; and let p D .0; 0/, x D @s .0; 0/, and y D @t .0; 0/. For any
s1 ; s2 ; t1 ; t2 2 I , let Pss11;t;t12 W T .s1 ;t1 / M ! T .s1 ;t2 / M denote parallel transport
along the curve t 7! .s1 ; t / from time t1 to time t2 , and let Pss12;t;t11 W T .s1 ;t1 / M !
T .s2 ;t1 / M denote parallel transport along the curve s 7! .s; t1 / from time s1 to
time s2 . (See Fig. 7.3.) Then for every z 2 Tp M ,
0;0 ı;0 ı;" 0;"
Pı;0 ı Pı;" ı P0;" ı P0;0 .z/ z
R.x; y/z D lim : (7.10)
ı;"!0 ı"
Proof. Define a vector field Z along by first parallel transporting z along the
curve t 7! .0; t /, and then for each t , parallel transporting Z.0; t / along the curve
s 7! .s; t /. The resulting vector field along is smooth by another application
of Theorem 4.31; and by construction, it satisfies Dt Z.0; t / D 0 for all t 2 I , and
Ds Z.s; t / D 0 for all .s; t / 2 I I . Proposition 7.5 shows that
R.x; y/z D Ds Dt Z.0; 0/ Dt Ds Z.0; 0/ D Ds Dt Z.0; 0/:

202 7 Curvature
Thus we need only show that Ds Dt Z.0; 0/ is equal to the limit on the right-hand
side of (7.10).
From Theorem 4.34, we have
s;0
Ps;" .Z.s; "// Z.s; 0/
Dt Z.s; 0/ D lim ; (7.11)
"!0 "
0;0
Pı;0 .Dt Z.ı; 0// Dt Z.0; 0/
Ds Dt Z.0; 0/ D lim : (7.12)
ı!0 ı
Evaluating (7.11) first at s D ı and then at s D 0, and inserting the resulting expres-
sions into (7.12), we obtain
Ds Dt Z.0; 0/
0;0 ı;0 0;0 0;0
Pı;0 ı Pı;" .Z.ı; "// Pı;0 .Z.ı; 0// P0;" .Z.0; "// C Z.0; 0/
D lim : (7.13)
ı;"!0 ı"
Here we have used the fact that parallel transport is linear, so the "-limit can be
0;0
pulled past Pı;0 .
Now, the fact that Z is parallel along t 7! .0; t / and along all of the curves
s 7! .s; t / implies
0;0 0;0
Pı;0 .Z.ı; 0// D P0;" .Z.0; "// D Z.0; 0/ D z;
ı;" ı;" 0;"
Z.ı; "/ D P0;" .Z.0; "// D P0;" ı P0;0 .z/:
Inserting these relations into (7.13) yields (7.10). t

u
Symmetries of the Curvature Tensor

The curvature tensor on a Riemannian or pseudo-Riemannian manifold has a num-
ber of symmetries besides the obvious skew-symmetry in its first two arguments.
Proposition 7.12 (Symmetries of the Curvature Tensor). Let .M; g/ be a Rie-

mannian or pseudo-Riemannian manifold. The .0; 4/-curvature tensor of g has the
following symmetries for all vector fields W , X , Y , Z:
(a) Rm.W; X; Y; Z/ D Rm.X; W; Y; Z/.
(b) Rm.W; X; Y; Z/ D Rm.W; X; Z; Y /.
(c) Rm.W; X; Y; Z/ D Rm.Y; Z; W; X /.
(d) Rm.W; X; Y; Z/ C Rm.X; Y; W; Z/ C Rm.Y; W; X; Z/ D 0.
Before we begin the proof, a few remarks are in order. First, as the proof will
show, (a) is a trivial consequence of the definition of the curvature endomorphism;
(b) follows from the compatibility of the Levi-Civita connection with the metric; (d)
Symmetries of the Curvature Tensor 203
follows from the symmetry of the connection; and (c) follows from (a), (b), and (d).
The identity in (d) is called the algebraic Bianchi identity (or more traditionally
but less informatively, the first Bianchi identity). It is easy to show using (a)–(d)
that a three-term sum obtained by cyclically permuting any three arguments of Rm
is also zero. Finally, it is useful to record the form of these symmetries in terms of
components with respect to any basis:
(a0 ) Rij kl D Rj ikl .
(b0 ) Rij kl D Rij lk .
(c0 ) Rij kl D Rklij .
(d0 ) Rij kl C Rj kil C Rkij l D 0.
Proof of Proposition 7.12. Identity (a) is immediate from the definition of the
curvature tensor, because R.W; X /Y D R.X; W /Y . To prove (b), it suffices to
show that Rm.W; X; Y; Y / D 0 for all Y , for then (b) follows from the expansion of
Rm.W; X; Y C Z; Y C Z/ D 0. Using compatibility with the metric, we have
W X jY j2 D W .2hrX Y; Y i/ D 2hrW rX Y; Y i C 2hrX Y; rW Y iI

X W jY j2 D X.2hrW Y; Y i/ D 2hrX rW Y; Y i C 2hrW Y; rX Y iI
ŒW; X jY j2 D 2hrŒW;X Y; Y i:
When we subtract the second and third equations from the first, the left-hand side
is zero. The terms 2hrX Y; rW Y i and 2hrW Y; rX Y i cancel on the right-hand side,
giving
0 D 2hrW rX Y; Y i 2hrX rW Y; Y i 2hrŒW;X Y; Y i

D 2hR.W; X /Y; Y i
D 2Rm.W; X; Y; Y /:
Next we prove (d). From the definition of Rm, this will follow immediately from
R.W; X /Y C R.X; Y /W C R.Y; W /X D 0:
Using the definition of R and the symmetry of the connection, the left-hand side
expands to
.rW rX Y rX rW Y rŒW;X Y /
C .rX rY W rY rX W rŒX;Y W /
C .rY rW X rW rY X rŒY;W X /
D rW .rX Y rY X / C rX .rY W rW Y / C rY .rW X rX W /
rŒW;X Y rŒX;Y W rŒY;W X
D rW ŒX; Y C rX ŒY; W C rY ŒW; X
rŒW;X Y rŒX;Y W rŒY;W X
D ŒW; ŒX; Y C ŒX; ŒY; W C ŒY; ŒW; X :
204 7 Curvature
This is zero by the Jacobi identity.

Finally, we show that identity (c) follows from the other three. Writing the alge-
braic Bianchi identity four times with indices cyclically permuted gives
Rm.W; X; Y; Z/ C Rm.X; Y; W; Z/ C Rm.Y; W; X; Z/ D 0;

Rm.X; Y; Z; W / C Rm.Y; Z; X; W / C Rm.Z; X; Y; W / D 0;
Rm.Y; Z; W; X / C Rm.Z; W; Y; X / C Rm.W; Y; Z; X / D 0;
Rm.Z; W; X; Y / C Rm.W; X; Z; Y / C Rm.X; Z; W; Y / D 0:
Now add up all four equations. Applying (b) four times makes all the terms in
the first two columns cancel. Then applying (a) and (b) in the last column yields
2Rm.Y; W; X; Z/ 2Rm.X; Z; Y; W / D 0, which is equivalent to (c). t
u
There is one more identity that is satisfied by the covariant derivatives of the
curvature tensor on every Riemannian manifold. Classically, it was called the sec-
ond Bianchi identity, but modern authors tend to use the more informative name
differential Bianchi identity.
Proposition 7.13 (Differential Bianchi Identity). The total covariant derivative of

the curvature tensor satisfies the following identity:
rRm.X; Y; Z; V; W / C rRm.X; Y; V; W; Z/ C rRm.X; Y; W; Z; V / D 0: (7.14)
In components, this is
Rij klIm C Rij lmIk C Rij mkIl D 0: (7.15)
Proof. First of all, by the symmetries of Rm, (7.14) is equivalent to
rRm.Z; V; X; Y; W / C rRm.V; W; X; Y; Z/ C rRm.W; Z; X; Y; V / D 0: (7.16)
This can be proved by a long and tedious computation, but there is a standard short-
cut for such calculations in Riemannian geometry that makes our task immeasurably
easier. To prove that (7.16) holds at a particular point p, it suffices by multilinear-
ity to prove the formula when X; Y; Z; V; W are basis vectors with respect to some
frame. The shortcut consists in choosing a special frame for each point p to simplify
the computations there.
Let p be an arbitrary point, let x i be normal coordinates centered at p, and let
X; Y; Z; V; W be arbitrary coordinate basis vector fields. These vectors satisfy two
properties that simplify our computations enormously: (1) their commutators vanish
identically, since Œ@i ; @j 0; and (2) their covariant derivatives vanish at p, since
ijk .p/ D 0 (Prop. 5.24(d)).
Using these facts and the compatibility of the connection with the metric, the first
term in (7.16) evaluated at p becomes
The Ricci Identities 205

.rW Rm/.Z; V; X; Y / D rW Rm.Z; V; X; Y /
D rW hR.Z; V /X; Y i
D rW hrZ rV X rV rZ X rŒZ;V X; Y i
D hrW rZ rV X rW rV rZ X; Y i:
Write this equation three times, with the vector fields W; Z; V cyclically permuted.
Summing all three gives
rRm.Z; V; X; Y; W / C rRm.V; W; X; Y; Z/ C rRm.W; Z; X; Y; V /

D hrW rZ rV X rW rV rZ X
C rZ rV rW X rZ rW rV X
C rV rW rZ X rV rZ rW X; Y i
D hR.W; Z/.rV X / C R.Z; V /.rW X / C R.V; W /.rZ X /; Y i
D 0;
where the last line follows because rV X D rW X D rZ X D 0 at p. t

u
The Ricci Identities
The curvature endomorphism also appears as the obstruction to commutation of

total covariant derivatives. Recall from Chapter 4 that if F is any smooth tensor
field of type .k; l/, then its second covariant derivative r 2 F D r.rF / is a smooth
2
.k; l C 2/-tensor field, and for vector fields X and Y , the notation rX;Y F denotes
r F .: : : ; Y; X /. Given vector fields X and Y , let R.X; Y / W T M ! T M denote
2
the dual map to R.X; Y /, defined by

R.X; Y / .Z/ D R.X; Y /Z :
Theorem 7.14 (Ricci Identities). On a Riemannian or pseudo-Riemannian mani-

fold M , the second total covariant derivatives of vector and tensor fields satisfy the
following identities. If Z is a smooth vector field,
2 2
rX;Y Z rY;X Z D R.X; Y /Z: (7.17)
If ˇ is a smooth 1-form,
2 2
rX;Y ˇ rY;X ˇ D R.X; Y / ˇ: (7.18)
And if B is a smooth .k; l/-tensor field,

206 7 Curvature
2 2

rX;Y B rY;X B .! 1 ; : : : ; ! k ; V1 ; : : : ; Vl /

D B R.X; Y / ! 1 ; ! 2 ; : : : ; ! k ; V1 ; : : : ; Vl C

C B ! 1 ; : : : ; ! k1 ; R.X; Y / ! k ; V1 ; : : : ; Vl (7.19)

B ! 1 ; : : : ; ! k ; R.X; Y /V1 ; V2 ; : : : ; Vl

B ! 1 ; : : : ; ! k ; V1 ; : : : ; Vl1 ; R.X; Y /Vl ;
for all covector fields ! i and vector fields Vj . In terms of any local frame, the
component versions of these formulas read
Z i Ipq Z i Iqp D Rpq m i Z m ; (7.20)

m
ˇj Ipq ˇj Iqp D Rpqj ˇm ; (7.21)
i :::i i :::i mi2 :::ik
i1 iki :::i m
Bj11 :::jkl Ipq Bj11 :::jkl Iqp D Rpq m Bj1 :::j l
Rpq m Bj11 :::jk1l
m i1 :::ik i :::i
C Rpqj1 Bmj2 :::jl C C Rpqjl m Bj11 :::jkl1 m : (7.22)
Proof. For any tensor field B and vector fields X; Y , Proposition 4.21 implies
2 2
rX;Y B rY;X B D rX rY B r.rX Y / B rY rX B C r.rY X/ B
(7.23)
D rX rY B rY rX B rŒX;Y B;
where the last equality follows from the symmetry of the connection. In particular,
this holds when B D Z is a vector field, so (7.17) follows directly from the definition
of the curvature endomorphism.
Next we prove (7.18). Using (4.13) repeatedly, we compute

.rX rY ˇ/.Z/ D X .rY ˇ/.Z/ .rY ˇ/.rX Z/

D X Y.ˇ.Z// ˇ.rY Z/ .rY ˇ/.rX Z/
D X Y.ˇ.Z// .rX ˇ/.rY Z/ ˇ.rX rY Z/ .rY ˇ/.rX Z/:
(7.24)
Reversing the roles of X and Y , we get
.rY rX ˇ/.Z/ D YX.ˇ.Z// .rY ˇ/.rX Z/ ˇ.rY rX Z/ .rX ˇ/.rY Z/;

(7.25)
and applying (4.13) one more time yields

rŒX;Y ˇ .Z/ D ŒX; Y .ˇ.Z// ˇ rŒX;Y Z : (7.26)
Now subtract (7.25) and (7.26) from (7.24): all but three of the terms cancel, yielding

rX rY ˇ rY rX ˇ rŒX;Y ˇ .Z/ D ˇ rX rY Z rY rX Z rŒX;Y Z

D ˇ R.X; Y /Z ;
which is equivalent to (7.18).

2 2
Next consider the action of rX;Y rY;X on an arbitrary tensor product:
Ricci and Scalar Curvatures 207
2 2

rX;Y rY;X .F ˝ G/

D rX rY rY rX rŒX;Y .F ˝ G/
D rX rY F ˝ G C rY F ˝ rX G C rX F ˝ rY G C F ˝ rX rY G
rY rX F ˝ G rX F ˝ rY G rY F ˝ rX G F ˝ rY rX G
rŒX;Y F ˝ G F ˝ rŒX;Y G
2 2
2 2

D rX;Y F rY;X F ˝ G C F ˝ rX;Y G rY;X G :
A simple induction using this relation together with (7.17) and (7.18) shows that for
all smooth vector fields W1 ; : : : ; Wk and 1-forms 1 ; : : : ; l ,
2 2

rX;Y rY;X W1 ˝ ˝ Wk ˝ 1 ˝ ˝ l

D R.X; Y /W1 ˝ W2 ˝ ˝ Wk ˝ 1 ˝ ˝ l C

C W1 ˝ ˝ Wk1 ˝ R.X; Y /Wk ˝ 1 ˝ ˝ l

C W1 ˝ ˝ Wk ˝ R.X; Y / 1 ˝ 2 ˝ ˝ l C

C W1 ˝ ˝ Wk ˝ 1 ˝ ˝ l1 ˝ R.X; Y / l :
Since every tensor field can be written as a sum of tensor products of vector fields
and 1-forms, this implies (7.19).
Finally, the component formula (7.22) follows by applying (7.19) to
2 2

rEq ;Ep B rE p ;Eq
B "i1 ; : : : ; "ik ; Ej1 ; : : : ; Ejl ;
where .Ej / and ."i / represent a local frame and its dual coframe, respectively, and
using
R.Eq ; Ep /Ej D Rqpj m Em D Rpqj m Em ;

R.Eq ; Ep / "i D Rqpm i "m D Rpq m i "m :
The other two component formulas are special cases of (7.22). t

u
Ricci and Scalar Curvatures

Suppose .M; g/ is an n-dimensional Riemannian or pseudo-Riemannian manifold.
Because 4-tensors are so complicated, it is often useful to construct simpler tensors
that summarize some of the information contained in the curvature tensor. The most
important such tensor is the Ricci curvature or Ricci tensor, denoted by Rc (or often
Ric in the literature), which is the covariant 2-tensor field defined as the trace of the
curvature endomorphism on its first and last indices. Thus for vector fields X; Y ,

Rc.X; Y / D tr Z 7! R.Z; X /Y :
208 7 Curvature
The components of Rc are usually denoted by Rij , so that
Rij D Rkij k D g km Rkij m :
The scalar curvature is the function S defined as the trace of the Ricci tensor:
S D trg Rc D Ri i D g ij Rij :
It is probably not clear at this point why the Ricci tensor or scalar curvature might be
interesting, and we do not yet have the tools to give them geometric interpretations.
But be assured that there is such an interpretation; see Proposition 8.32.
Lemma 7.15. The Ricci curvature is a symmetric 2-tensor field. It can be expressed
in any of the following ways:
Rij D Rkij k D Rik k j D Rki k j D Rikj k :
I Exercise 7.16. Prove Lemma 7.15, using the symmetries of the curvature tensor.
It is sometimes useful to decompose the Ricci tensor into a multiple of the metric
and a complementary piece with zero trace. Define the traceless Ricci tensor of g
as the following symmetric 2-tensor:
1
V D Rc Sg:
Rc
n
Proposition 7.17. Let .M; g/ be a Riemannian or pseudo-Riemannian n-manifold.
V 0, and the Ricci tensor decomposes orthogonally as
Then trg Rc
1
V C Sg:
Rc D Rc (7.27)
n
Therefore, in the Riemannian case,
ˇ ˇ2 1 2
Vˇ C S :
jRcj2g D ˇRc g
(7.28)
n
Remark. The statement about norms, and others like it that we will prove below,
works only in the Riemannian case because of the additional absolute value signs
required to compute norms in the˝ pseudo-Riemannian
˛ case. The pseudo-Riemannian
analogue would be hRc; Rcig D Rc; V Rc
V C 1 S 2 , but this is not as useful.
g n
Proof. Note that in every local frame, we have
trg g D gij g j i D ıii D n:
V that trg Rc
It then follows directly from the definition of Rc V 0 and (7.27) holds.
The fact that the decomposition is orthogonal follows easily from the fact that for
every symmetric 2-tensor h, we have
hh; gi D g ik g j l hij gkl D g ij hij D trg h;

˝ ˛
and therefore Rc;V g D trg Rc
V D 0. Finally, (7.28) follows from (7.27) and the fact
that hg; gi D trg g D n. t
u
The next proposition, which follows directly from the differential Bianchi iden-
tity, expresses some important relationships among the covariant derivatives of the
various curvature tensors. To express it concisely, it is useful to introduce another
operator on tensor fields. If T is a smooth 2-tensor field on a Riemannian or pseudo-
Riemannian manifold, we define the exterior covariant derivative of T to be the
3-tensor field DT defined by
.DT /.X; Y; Z/ D .rT /.X; Y; Z/ C .rT /.X; Z; Y /: (7.29)
In terms of components, this is
.DT /ij k D Tij Ik C TikIj :
(This operator is a generalization of the ordinary exterior derivative of a 1-form,

which can be expressed in terms of the total covariant derivative by .d/.Y; Z/ D
.r/.Y; Z/ C .r/.Z; Y / by the result of Problem 5-13. The exterior covariant
derivative can be generalized to other types of tensors as well, but this is the only
case we need.)
Proposition 7.18 (Contracted Bianchi Identities). Let .M; g/ be a Riemannian or

pseudo-Riemannian manifold. The covariant derivatives of the Riemann, Ricci, and
scalar curvatures of g satisfy the following identities:
trg .rRm/ D D.Rc/; (7.30)

1
trg .rRc/ D 2
dS; (7.31)
where the trace in each case is on the first and last indices. In components, this is
Rij klI i D Rj kIl Rj lIk ; (7.32)

i 1
RilI D S :
2 Il
(7.33)
Proof. Start with the component form (7.15) of the differential Bianchi identity,
raise the index m, and then contract on the indices i; m to obtain (7.32). (Note that
covariant differentiation commutes with contraction by Proposition 4.15 and with
the musical isomorphisms by Proposition 5.17, so it does not matter whether the
indices that are raised and contracted come before or after the semicolon.) Then do
the same with the indices j; k and simplify to obtain (7.33). The coordinate-free
formulas (7.30) and (7.31) follow by expanding everything out in components. u t
It is important to note that if the sign convention chosen for the curvature tensor
is the opposite of ours, then the Ricci tensor must be defined as the trace of Rm on
the first and third (or second and fourth) indices. (The trace on the first two or last
210 7 Curvature
two indices is always zero by antisymmetry.) The definition is chosen so that the
Ricci and scalar curvatures have the same meaning for everyone, regardless of the
conventions chosen for the full curvature tensor. So, for example, if a manifold is
said to have positive scalar curvature, there is no ambiguity as to what is meant.
A Riemannian or pseudo-Riemannian metric is said to be an Einstein metric if
its Ricci tensor is a constant multiple of the metric—that is,
Rc D g for some constant : (7.34)
This equation is known as the Einstein equation. As the next proposition shows,
for connected manifolds of dimension greater than 2, it is not necessary to assume
that is constant; just assuming that the Ricci tensor is a function times the metric
is sufficient.
Proposition 7.19 (Schur’s Lemma). Suppose .M; g/ is a connected Riemannian

or pseudo-Riemannian manifold of dimension n 3 whose Ricci tensor satisfies
Rc D fg for some smooth real-valued function f . Then f is constant and g is an
Einstein metric.
Proof. Taking traces of both sides of Rc D fg shows that f D n1 S , so the trace-

V 0. Because the covariant
less Ricci tensor is identically zero. It follows that r Rc
derivative of the metric is zero, this implies the following equation in any coordinate
chart:
1
0 D Rij Ik SIk gij :
n
Tracing this equation on i and k, and comparing with the contracted Bianchi identity
(7.33), we conclude that
1 1
0 D SIj SIj :
2 n
Because n 3, this implies SIj D 0. But SIj is the component of rS D dS , so
connectedness of M implies that S is constant and thus so is f . t
u
Corollary 7.20. If .M; g/ is a connected Riemannian or pseudo-Riemannian mani-

V D 0.
fold of dimension n 3, then g is Einstein if and only if Rc
Proof. Suppose first that g is an Einstein metric with Rc D g. Taking traces of both
V D Rc g D 0. Conversely, if Rc
sides, we find that D n1 S , and therefore Rc V D 0,
Schur’s lemma implies that g is Einstein. t
u
By an argument analogous to those of Chapter 6, Hilbert showed (see [Bes87,
Thm. 4.21]) that Einstein
R metrics are critical points for the total scalar curvature
functional .g/ D M S d Vg on the space of all metrics on M with fixed volume.
Thus Einstein metrics can be viewed as “optimal” metrics in a certain sense, and as
such they form an appealing higher-dimensional analogue of locally homogeneous
metrics on 2-manifolds, with which one might hope to prove some sort of general-
ization of the uniformization theorem (Thm. 3.22). Although the statement of such
a theorem cannot be as elegant as that of its 2-dimensional ancestor because there
are known examples of smooth, compact manifolds that admit no Einstein metrics
[Bes87, Chap. 6], there is still a reasonable hope that “most” higher-dimensional
manifolds (in some sense) admit Einstein metrics. This is an active field of current
research; see [Bes87] for a sweeping survey of Einstein metrics.
The term “Einstein metric” originated, as you might guess, in physics: the central
assertion of Einstein’s general theory of relativity is that physical spacetime is mod-
eled by a 4-manifold that carries a Lorentz metric whose Ricci curvature satisfies
the following Einstein field equation:
1
Rc Sg D T; (7.35)
2
where T is a certain symmetric 2-tensor field (the stress–energy tensor) that
describes the density, momentum, and stress of the matter and energy present at
each point in spacetime. It is shown in physics books (e.g., [CB09, pp. 51–53]) that
(7.35) is the variational equation for a certain functional, called the Einstein–Hilbert
action, on the space of all Lorentz metrics on a given 4-manifold. Einstein’s theory
can then be interpreted as the assertion that a physically realistic spacetime must be
a critical point for this functional.
In the special case T 0, (7.35) reduces to the vacuum Einstein field
equation Rc D 12 Sg. Taking traces of both sides and recalling that trg g D dim M D
4, we obtain S D 2S , which implies S D 0. Therefore, the vacuum Einstein equation
is equivalent to Rc D 0, which means that g is a (pseudo-Riemannian) Einstein met-
ric in the mathematical sense of the word. (At one point in the development of the
theory, Einstein considered adding a term g to the left-hand side of (7.35), where
is a constant that he called the cosmological constant. With this modification the
vacuum Einstein field equation would be exactly the same as the mathematicians’
Einstein equation (7.34). Einstein soon decided that the cosmological constant
was a mistake on physical grounds; however, researchers in general relativity have
recently begun to believe that a theory with a nonzero cosmological constant might in
fact have physical relevance.)
Other than these special cases and the obvious formal similarity between (7.35)
and (7.34), there is no direct connection between the physicists’ version of the
Einstein equation and the mathematicians’ version. The mathematical interest in
Riemannian Einstein metrics stems more from their potential applications to uni-
formization in higher dimensions than from their relation to physics.
Another approach to generalizing the uniformization theorem to higher dimen-
sions is to search for metrics of constant scalar curvature. These are also critical
points of the total scalar curvature functional, but only with respect to variations of
the metric with fixed volume within a given conformal equivalence class. Thus it
makes sense to ask whether, given a metric g on a manifold M , there exists a metric
gz conformal to g that has constant scalar curvature. This is called the Yamabe prob-
lem, because it was first posed in 1960 by Hidehiko Yamabe, who claimed to have
proved that the answer is always yes when M is compact. Yamabe’s proof was
later found to be in error, and it was two dozen years before the proof was finally
completed by Richard Schoen; see [LP87] for an expository account of Schoen’s
212 7 Curvature
solution. When M is noncompact, the issues are much subtler, and much current
research is focused on determining exactly which conformal classes contain metrics
of constant scalar curvature.
The Weyl Tensor

As noted above, the Ricci and scalar curvatures contain only part of the information
encoded into the curvature tensor. In this section, we introduce a tensor field called
the Weyl tensor, which encodes all the rest.
We begin by considering some linear-algebraic aspects of tensors that have the
symmetries of the curvature tensor. Suppose V is an n-dimensional real vector
space. Let R.V / T 4 .V / denote the vector space of all covariant 4-tensors T
on V that have the symmetries of the .0; 4/ Riemann curvature tensor:
(a) T .w; x; y; z/ D T .x; w; y; z/.
(b) T .w; x; y; z/ D T .w; x; z; y/.
(c) T .w; x; y; z/ D T .y; z; w; x/.
(d) T .w; x; y; z/ C T .x; y; w; z/ C T .y; w; x; z/ D 0.
(As the proof of Prop. 7.12 showed, (c) follows from the other three symmetries, so
it would suffice to assume only (a), (b), and (d); but it is more convenient to include
all four symmetries in the definition.) An element of R.V / is called an algebraic
curvature tensor on V .
Proposition 7.21. If the vector space V has dimension n, then

n2 n2 1
dim R.V / D : (7.36)
12
Proof. Let B.V / denote the linear subspace 4
of T .V / consisting of tensors satis-
2 2
fying properties (a)–(c), and let † ƒ .V / denote the space of symmetric bilinear
2
forms on the vector
2
2 space

ƒ .V /of alternating contravariant 2-tensors on V . Define
a map ˚ W † ƒ .V / ! B.V / as follows:
˚.B/.w; x; y; z/ D B.w ^ x; y ^ z/:
It is easy to check that ˚.B/ satisfies (a)–(c), so ˚.B/ 2 B.V /, and that ˚ is a
linear map. In fact, it is an isomorphism, which we prove by constructing an inverse
for it. Choose a basis .b1 ; : : : ; bn / for V , so the fbi ^ bj W i < j g is a basis
collection
for ƒ2 .V /. Define a map W B.V / ! †2 ƒ2 .V / by setting
.T /.bi ^ bj ; bk ^ bl / D T .bi ; bj ; bk ; bl /
when i < j and k < l, and extending by bilinearity. A straightforward computation

shows that is an inverse for ˚.
The Weyl Tensor 213
The upshot of the preceding construction is that

n n

2 2
2 2
C1 n.n 1/.n2 n C 2/
dim B.V / D dim † ƒ .V / D D ;
2 8

where we have used the facts that dim ƒ2 .V / D n2 D n.n 1/=2 and the dimension
of the space of symmetric bilinear forms on a vector space of dimension m is
m.m C 1/=2.
Now consider the linear map
W B.V / ! T 4 .V / defined by

.T /.w; x; y; z/ D 13 T .w; x; y; z/ C T .x; y; w; z/ C T .y; w; x; z/ :
By definition, R.V / is equal to the kernel of

. In fact,
is equal to the restriction
to B.V / of the operator Alt W T 4 .V / ! ƒ4 .V / defined by (B.9): thanks to the
symmetries (a)–(c), the 24 terms in the definition of Alt T can be arranged in three
groups of eight in such a way that all the terms in each group reduce to one of the
terms in the definition of
. Thus the image of
is contained in ƒ4 .V /. In fact, the
image is all of ƒ4 .V /: every T 2 ƒ4 .V / satisfies (a)–(c) and thus lies in B.V /,
and
.T / D Alt T D T for each such tensor.
Therefore, using the rank–nullity theorem of linear algebra, we conclude that
!
4 n.n 1/.n2 n C 2/ n
dim R.V / D dim B.V / dim ƒ .V / D ;
8 4
and simplification yields (7.36). t

u
Let us now assume that our vector space V is endowed with a (not necessarily
positive definite) scalar product g 2 †2 .V /. Let trg W R.V / ! †2 .V / denote the
trace operation (with respect to g) on the first and last indices (so that, for example,
Rc D trg .Rm/). It is natural to wonder whether this operator is surjective and what
its kernel is, as a way of asking how much of the information contained in the
Riemann curvature tensor is captured by the Ricci tensor. One way to try to answer
the question is to attempt to construct a right inverse for the trace operator—a linear
map G W †2 .V / ! R.V / such that trg .G.S // D S for all S 2 †2 .V /.
Such an operator must start with a symmetric 2-tensor and construct a 4-tensor,
using only the given 2-tensor and the metric. It turns out that there is a natural way
to construct an algebraic curvature tensor out of two symmetric 2-tensors, which we
now describe. Given h; k 2 †2 .V /, we define a covariant 4-tensor h k, called the
Kulkarni–Nomizu product of h and k, by the following formula:
h k.w; x; y; z/ D h.w; z/k.x; y/Ch.x; y/k.w; z/

h.w; y/k.x; z/ h.x; z/k.w; y/: (7.37)
In terms of any basis, the components of h k are
.h k/ij lm D him kj l C hj l kim hil kj m hj m kil :

214 7 Curvature
(It should be noted that the Kulkarni–Nomizu product must be defined as the nega-
tive of this expression when the Riemann curvature tensor is defined as the negative
of ours.)
Lemma 7.22 (Properties of the Kulkarni–Nomizu Product). Let V be an n-
dimensional vector space endowed with a scalar product g, let h and k be sym-
metric 2-tensors on V , let T be an algebraic curvature tensor on V , and let trg
denote the trace on the first and last indices.
(a) h k is an algebraic curvature tensor.
(b) h k D k h.
(c) trg .h g/ D .n 2/h C .trg h/g.
(d) trg .g g/ D 2.n 1/g.
(e) hT; h gig D 4htrg T; hig .
(f) In case g is positive definite, jg hj2g D 4.n 2/jhj2g C 4.trg h/2 .
Proof. It is evident from the definition that h k has three of the four symmetries of
an algebraic curvature tensor: it is antisymmetric in its first two arguments and also
in its last two, and its value is unchanged when the first two and last two arguments
are interchanged. Thus to prove (a), only the algebraic Bianchi identity needs to
be checked. This is a straightforward computation: when h k.w; x; y; z/ is written
three times with the arguments w; x; y cyclically permuted and the three expressions
are added together, all the terms cancel due to the symmetry of h and k.
Part (b) is immediate from the definition. To prove (c), choose a basis and use the
definition to compute

.trg .h g//j k D g im him gj l C hj l gim hil gj m hj m gil
D hi i gj l C nhj l hj l hj l ;
which is equivalent to (c). Then (d) follows from (c) and the fact that trg g D n.
The proofs of (e) and (f) are left to Problem 7-9. t
u
Here is the primary application of the Kulkarni–Nomizu product.
Proposition 7.23. Let .V; g/ be an n-dimensional scalar product space with n 3,
and define a linear map G W †2 .V / ! R.V / by

1 trg h
G.h/ D h g g: (7.38)
n2 2.n 1/
Then G is a right inverse for trg , and its image is the orthogonal complement of the
kernel of trg in R.V /.
Proof. The fact that G is a right inverse is a straightforward computation based
on the definition and Lemma 7.22(c,d). This implies that G is injective and trg
is surjective, so the dimension of Im G is equal to the codimension of Ker.trg /,
which in turn is equal to the dimension of Ker.trg /? . If T 2 R.V / is an algebraic
curvature tensor such that trg T D 0, then Lemma 7.22(e) shows that hT; G.h/i D 0,
so it follows by dimensionality that Im G D Ker.trg /? . t
u
The Weyl Tensor 215
Now suppose g is a Riemannian or pseudo-Riemannian metric. Define the Schouten

tensor of g, denoted by P , to be the following symmetric 2-tensor field:

1 S
PD Rc g I
n2 2.n 1/
and define the Weyl tensor of g to be the following algebraic curvature tensor field:
W D Rm P g
1 S
D Rm Rc g C g g:
n2 2.n 1/.n 2/
Proposition 7.24. For every Riemannian or pseudo-Riemannian manifold .M; g/ of

dimension n 3, the trace of the Weyl tensor is zero, and Rm D W C P g is the
orthogonal decomposition of Rm corresponding to R.V / D Ker.trg / ˚ Ker.trg /? .
Proof. This follows immediately from Proposition 7.23 and the fact that P g D
G.Rc/ D G.trg Rm/. t
u
These results lead to some important simplifications in low dimensions.
Corollary 7.25 Let V be an n-dimensional real vector space.
(a) If n D 0 or n D 1, then R.V / D f0g.
(b) If n D 2, then R.V / is 1-dimensional, spanned by g g.
(c) If n D 3, then R.V / is 6-dimensional, and G W †2 .V / ! R.V / is an iso-
morphism.
Proof. The dimensional results follow immediately from Proposition 7.21. In the
case n D 2, Lemma 7.22(d) shows that trg .g g/ D 2g ¤ 0, which implies that
g g is nonzero and therefore spans the 1-dimensional space R.V /.
Now consider n D 3. Proposition 7.23 shows that trg ıG is the identity, which
means that G W †2 .V / ! R.V / is injective. On the other hand, Proposition 7.21
shows that dim R.V / D 6 D dim †2 .V /, so G is also surjective. t
u
The next corollary shows that the entire curvature tensor is determined by the
Ricci tensor in dimension 3.
Corollary 7.26 (The Curvature Tensor in Dimension 3). On every Riemannian
or pseudo-Riemannian manifold .M; g/ of dimension 3, the Weyl tensor is zero, and
the Riemann curvature tensor is determined by the Ricci tensor via the formula
Rm D P g D Rc g 14 Sg g: (7.39)
Proof. Corollary 7.25 shows that G W †2 .V / ! R.V / is an isomorphism in

dimension 3. Since trg ıG is the identity, it follows that trg is also an isomorphism.
Because trg W is always zero by Proposition 7.24, it follows that W is always zero.
Formula (7.39) then follows from the definition of the Weyl tensor. t
u
In dimension 2, the definitions of the Weyl and Schouten tensors do not make
sense; but we have the following analogous result instead.
216 7 Curvature
Corollary 7.27 (The Curvature Tensor in Dimension 2). On every Riemannian or

pseudo-Riemannian manifold .M; g/ of dimension 2, the Riemann and Ricci tensors
are determined by the scalar curvature as follows:
Rm D 14 Sg g; Rc D 12 Sg:
Proof. In dimension 2, it follows from Corollary 7.25(b) that there is some scalar
function f such that Rm D fg g. Taking traces, we find from Lemma 7.22(d) that
Rc D trg .Rm/ D 2fg, and then S D trg .Rc/ D 2f trg .g/ D 4f . The results follow
by substituting f D 14 S back into these equations. t
u
Although the traceless Ricci tensor is always zero on a 2-manifold, this does not
imply that S is constant, since the proof of Schur’s lemma fails in dimension 2.
Einstein metrics in dimension 2 are simply those with constant scalar curvature.
Returning now to dimensions greater than 2, we can use (7.27) to further decom-
pose the Schouten tensor into a part determined by the traceless Ricci tensor and a
purely scalar part. The next proposition is the analogue of Proposition 7.17 for the
full curvature tensor.
Proposition 7.28 (The Ricci Decomposition of the Curvature Tensor). Let .M; g/
be a Riemannian or pseudo-Riemannian manifold of dimension n 3. Then the
.0; 4/-curvature tensor of g has the following orthogonal decomposition:
1 1
Rm D W C V gC
Rc S g g: (7.40)
n2 2n.n 1/
Therefore, in the Riemannian case,
1 ˇ ˇ 1
jRmj2g D jW j2g C V g ˇ2 C
ˇRc jS g gj2g
.n 2/2 g 4n2 .n 1/2
(7.41)
4 ˇˇ ˇˇ2 2
D jW j2g C V C
Rc S 2:
n2 g n.n 1/
Proof. The decomposition (7.40) follows immediately by substituting (7.27) into
the definition of the Weyl tensor and simplifying. The decomposition is orthogonal
thanks to Lemma 7.22(e), and (7.41) follows from Lemma 7.22(f). t
u
Curvatures of Conformally Related Metrics
Recall that two Riemannian or pseudo-Riemannian metrics on the same manifold

are said to be conformal to each other if one is a positive function times the other.
For example, we have seen that the round metrics and the hyperbolic metrics are all
conformal to Euclidean metrics, at least locally.
If g and gz are conformal metrics on a smooth manifold M, there is no reason to
expect that the curvature tensors of g and gz should be closely related to each other.
Curvatures of Conformally Related Metrics 217
But it is a remarkable fact that the Weyl tensor has a very simple transformation law
under conformal changes of metric. In this section, we derive that law.
First we need to determine how the Levi-Civita connection changes when a met-
ric is changed conformally. Given conformal metrics g and gz, we can always write
gz D e 2f g for some smooth real-valued function f .
Proposition 7.29 (Conformal Transformation of the Levi-Civita Connection).

Let .M; g/ be a Riemannian or pseudo-Riemannian n-manifold (with or without
z denote the
boundary), and let gz D e 2f g be any metric conformal to g. If r and r
Levi-Civita connections of g and gz, respectively, then
z X Y D rX Y C .Xf /Y C .Yf /X hX; Y ig grad f;
r (7.42)
where the gradient on the right-hand side is that of g. In any local coordinates, the
Christoffel symbols of the two connections are related by
zijk D ijk C fIi ıjk C fIj ıik g kl fIl gij ;
(7.43)
where fIi D @i f is the i th component of rf D df D fIi dx i .
Proof. Formula (7.43) is a straightforward computation using formula (5.10) for the
Christoffel symbols in coordinates, and then (7.42) follows by expanding everything
in coordinates and using (7.43). t
u
This result leads to transformation laws for the various curvature tensors.
Theorem 7.30 (Conformal Transformation of the Curvature). Let g be a Rie-

mannian or pseudo-Riemannian metric on an n-manifold M with or without
boundary, f 2 C 1 .M /, and gz D e 2f g. In the Riemannian case, the curvature
tensors of gz (represented with tildes) are related to those of g by the following for-
mulas:

f D e 2f Rm .r 2 f / g C .df ˝ df / g 1 jdf j2g .g g/ ;
Rm (7.44)
2

e D Rc .n 2/.r 2 f / C .n 2/.df ˝ df / f C .n 2/jdf j2g g; (7.45)
Rc

Sz D e 2f S 2.n 1/f .n 1/.n 2/jdf j2g ; (7.46)
where the curvatures and covariant derivatives on the right are those of g, and
f D div.grad f / is the Laplacian of f defined in Chapter 2. If in addition n 3,
then
Pz D P r 2 f C .df ˝ df / 12 jdf j2g g; (7.47)

e D e 2f W:
W (7.48)
In the pseudo-Riemannian case, the same formulas hold with each occurrence of
jdf j2g replaced by hdf; df i2g .
218 7 Curvature
Proof. We begin with (7.44). The plan is to choose local coordinates and insert for-
mula (7.43) for the Christoffel symbols into the coordinate formula (7.8) for the co-
efficients of the curvature tensor. As in the proof of the differential Bianchi identity,
we can make the computations much more tractable by computing the components
of these tensors at a point p 2 M in normal coordinates for g centered at p, so that
the equations gij D ıij , @k gij D 0, and ijk D 0 hold at p. This has the following
consequences at p:
fIij D @j @i f;
zijk D fIi ıjk C fIj ıik g kl fIl gij ;

zijk D @m ijk C fIim ıjk C fIj m ıik g kl fIlm gij ;
@m
Rij k l D @i jl k @j ik
l
:
Now start with

zij kl D e 2f glm @i
R z m @j
zm C
zp zm zm :
zp
jk ik j k ip ik jp
Inserting the relations above and simplifying, we eventually obtain

Rzij kl D e 2f Rij kl .fIil gj k C fIj k gil fIil gj k fIj k gil /
C .fIi fIl gj k C fIj fIk gil fIi fIk gj l fIj fIl gik /

g mp fIm fIp .gil gj k gik gj l / ;
which is the coordinate version of (7.44). (See Problem 5-14 for the formula for the
Laplacian in terms of covariant derivatives.) The rest of the formulas follow easily
from this, using the identities of Lemma 7.22 and the fact that g ij D e 2f gzij . u
t
The next corollary begins to explain the geometric significance of the Weyl ten-
sor. Recall that a Riemannian manifold is said to be locally conformally flat if
every point has a neighborhood that is conformally equivalent to an open subset of
Euclidean space. Similarly, a pseudo-Riemannian manifold is locally conformally
flat if every point has a neighborhood conformally equivalent to an open set in a
pseudo-Euclidean space.

of dimension n 3. If g is locally conformally flat, then its Weyl tensor vanishes
identically.
Proof. Suppose .M; g/ is locally conformally flat. Then for each p 2 M there exist
a neighborhood U and an embedding ' W U ! Rn such that ' pulls back a flat
(Riemannian or pseudo-Riemannian) metric on Rn to a metric of the form gz D
e 2f g. This implies that gz has zero Weyl tensor, because its entire curvature tensor
is zero. By virtue of (7.48), the Weyl tensor of g is also zero. t
u
Thus a necessary condition for a smooth manifold of dimension at least 3 to

be locally conformally flat is that its Weyl tensor vanish identically. As Theorem
7.37 below will show, in dimensions 4 and higher, this condition is also sufficient.
But in 3 dimensions, Corollary 7.26 shows that the Weyl tensor vanishes identically
on every manifold, so to understand that case, we must introduce one more tensor
field. On a Riemannian or pseudo-Riemannian manifold, the Cotton tensor is
defined as the negative of the exterior covariant derivative of the Schouten tensor:
C D DP . This is the 3-tensor field whose expression in terms of any local frame is
Cij k D Pij Ik PikIj : (7.49)

fold of dimension n 3, and let W and C denote its Weyl and Cotton tensors,
respectively. Then
trg .rW / D .n 3/C; (7.50)
where the trace is on the first and last indices of the 5-tensor rW .
Proof. Writing W D Rm P g and using the component form of the first con-
tracted Bianchi identity (7.32), we obtain
Wij klI i D Rj kIl Rj lIk PilI i gj k Pj kI i gil C PikI i gj l C Pj lI i gik : (7.51)
Note that Pj kI i gil D Pj kIl and Pj lI i gik D Pj lIk . To simplify the other two terms,
we use the definition of the Schouten tensor and the second contracted Bianchi
identity (7.33) to obtain that

i 1 i S Ii 1
PilI D RilI gil D SIl :
n2 2.n 1/ 2.n 1/
The analogous formula holds for PikI i . When we insert these expressions into (7.51)
and simplify, we find that the terms involving derivatives of the Ricci and scalar
curvatures combine to yield
Wij klI i D .n 2/.Pj kIl Pj lIk / Pj kIl C Pj lIk D .n 3/Cj kl ;
which is the component version of (7.50). t

u
Corollary 7.33. Suppose .M; g/ is a Riemannian or pseudo-Riemannian manifold.

If dim M 4 and the Weyl tensor vanishes identically, then so does the Cotton
tensor. t
u
The next proposition expresses another important feature of the Cotton tensor.
Proposition 7.34 (Conformal Invariance of the Cotton Tensor in Dimension 3).

Suppose .M; g/ is a Riemannian or pseudo-Riemannian 3-manifold, and gz D e 2f g
for some f 2 C 1 .M /. If C and Cz denote the Cotton tensors of g and gz, respec-
tively, then Cz D C .
220 7 Curvature

u
Corollary 7.35. If .M; g/ is a locally conformally flat 3-manifold, then the Cotton
tensor of g vanishes identically.
I Exercise 7.36. Prove this corollary.
The real significance of the Weyl and Cotton tensors is explained by the following
important theorem.
Theorem 7.37 (Weyl–Schouten). Suppose .M; g/ is a Riemannian or pseudo-

Riemannian manifold of dimension n 3.
(a) If n 4, then .M; g/ is locally conformally flat if and only if its Weyl tensor is
identically zero.
(b) If n D 3, then .M; g/ is locally conformally flat if and only if its Cotton tensor
is identically zero.
Proof. The necessity of each condition was proved in Corollaries 7.31 and 7.35. To
prove sufficiency, suppose .M; g/ satisfies the hypothesis appropriate to its dimen-
sion; then it follows from Corollaries 7.26 and 7.33 that the Weyl and Cotton tensors
of g are both identically zero. Every metric gz D e 2f g conformal to g also has zero
f D Pz gz. We will show that in
Weyl tensor, and therefore its curvature tensor is Rm
a neighborhood of each point, the function f can be chosen to make Pz 0, which
implies that Rmf 0 and therefore gz is flat.
From (7.47), it follows that Pz 0 if and only if
P r 2 f C .df ˝ df / 12 hdf; df i2g g D 0: (7.52)
This equation can be written in the form r.df / D A.df /, where A is the map from
1-tensors to symmetric 2-tensors given by
A./ D . ˝ / 12 h; i2g g C P;
or in components,
A./ij D i j 12 m m gij C Pij : (7.53)
We will solve this equation by first looking for a smooth 1-form that satisfies
r D A./. In any local coordinates, if we write D j dx j , this becomes an
overdetermined system of first-order partial differential equations for the n unknown
functions 1 ; : : : ; n . (A system of differential equations is said to be overdetermined
if there are more equations than unknown functions.) If is a solution to r D A./
on an open subset of M , then the Ricci identity (7.21) shows that A./ satisfies
A./ij Ik A./ikIj D iIj k iIkj D Rj ki l l : (7.54)
Lemma 7.38 below shows that this condition is actually sufficient: more precisely,
r D A./ has a smooth solution in a neighborhood of each point, provided that for
every smooth covector field , the covariant derivatives of A./ satisfy A./ij Ik

A./ikIj D Rj ki l l when A./ij is substituted for iIj wherever it appears.
To see that this condition is satisfied, differentiate (7.53) to obtain
Aij Ik AikIj
D iIk j C i j Ik mIk m gij iIj k i kIj C mIj m gik C Cij k :
Now substitute the right-hand side of (7.53) (with appropriate index substitutions)
for iIj wherever it appears, subtract Rj ki l l , and use the relation Rm D W CP g.
After extensive cancellations, we obtain
A./ij Ik A./ikIj Rj ki l l D Wj ki l l C Cij k : (7.55)
Our hypotheses guarantee that the right-hand side is identically zero, so there is a
solution to r D A./ in a neighborhood of each point.
Because A./ij is symmetric in i and j , it follows that the ordinary deriva-
tives @j i D iIj C ijk k are also symmetric, and thus is a closed 1-form. By
the Poincaré lemma, in some (possibly smaller) neighborhood of each point, there
is a smooth function f such that D df D rf ; this f is the function we seek. u t
Here is the lemma that was used in the proof of the preceding theorem.
Lemma 7.38. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold, and

consider the overdetermined system of equations
r D A./; (7.56)
where A W T M ! T 2 T M is a smooth map satisfying the following compatibility

condition: for any smooth covector field , the 3-tensor field r.A.// satisfies the
following identity when A./ is substituted for r wherever it appears:
A./ij Ik A./ikIj D Rj ki l l : (7.57)
Then for every p 2 M and every covector 0 2 Tp M , there is a smooth solution to
(7.56) on a neighborhood of p satisfying p D 0 .

Proof. Let p 2 M be given. In smooth local coordinates x i on a neighborhood of
p, (7.56) is equivalent to the overdetermined system
@i .x/
D aij .x; .x//;
@x j
where
aij .x; / D ijk .x/k C A./ij :
An application of the Frobenius theorem [LeeSM, Prop. 19.29] shows that there is
a smooth solution to this overdetermined system in a neighborhood of p with p
arbitrary, provided that
222 7 Curvature
@aij @aij @aik @aik

k
C akl D j
C aj l : (7.58)
@x @l @x @l
By
virtue of the
chain rule, this identity means exactly that the first derivatives
@ aij .x; .x// =@x k , after substituting akl .x; .x// in place of @k .x/=@x l , are
m
symmetric in the indices j; k. Because kIl D @l k kl m , this substitution is
equivalent to substituting A./kl for kIl wherever it occurs. After expanding the
hypothesis (7.57) in terms of Christoffel symbols, we find after some manipulation
that it reduces to (7.58). t
u
Because all Riemannian 1-manifolds are flat, the only nontrivial case that is not
addressed by the Weyl–Schouten theorem is that of dimension 2. Smooth coordi-
nates that provide a conformal equivalence between an open subset of a Riemannian
or pseudo-Riemannian 2-manifold and an open subset of R2 are called isothermal
coordinates, and it is a fact that such coordinates always exist locally. For the Rie-
mannian case, there are various proofs available, all of which involve more machin-
ery from partial differential equations and complex analysis than we have at our dis-
posal; see [Che55] for a reasonably elementary proof. For the pseudo-Riemannian
case, see Problem 7-15. Thus every Riemannian or pseudo-Riemannian 2-manifold
is locally conformally flat.
Problems
7-1. Complete the proof of Proposition 7.3 by showing that R.X; Y /.f Z/ D
fR.X; Y /Z for all smooth vector fields X; Y; Z and smooth real-valued
functions f .
7-2. Prove Proposition 7.4 (the formula for the curvature tensor in coordinates).
7-3. Show that the curvature tensor of a Riemannian locally symmetric space is
parallel: rRm 0. (Used on pp. 297, 351.)

7-4. Let M be a Riemannian or pseudo-Riemannian manifold, and let x i be
normal coordinates centered at p 2 M . Show that the following holds at p:
1
Rij kl D @j @l gik C @i @k gj l @i @l gj k @j @k gil :
2
7-5. Let r be the Levi-Civita connection on a Riemannian or pseudo-Riemannian
manifold .M; g/, and let !i j be its connection 1-forms with respect to a local
frame .Ei / (Problem 4-14). Define a matrix of 2-forms i j , called the cur-
vature 2-forms, by
1
i j D Rkli j "k ^ "l ;
2
i
where " is the coframe dual to .Ei /. Show that the curvature 2-forms
satisfy Cartan’s second structure equation:
Problems 223
i j D d!i j !i k ^ !k j :
[Hint: Expand R.Ek ; El /Ei in terms of r and !i j .]

7-6. Suppose .M1 ; g1 / and .M2 ; g2 / are Riemannian manifolds, and M1 M2 is
endowed with the product metric g D g1 ˚ g2 as in (2.12). Show that the
Riemann curvature, Ricci curvature, and scalar curvature of g are given by
the following formulas:
Rm D
1 Rm1 C
2 Rm2 ;
Rc D
1 Rc1 C
2 Rc2 ;
S D
1 S1 C
1 S2 ;
where Rmi , Rci , and Si are the Riemann, Ricci, and scalar curvatures of
.Mi ; gi /, and
i W M1 M2 ! Mi is the projection. (Used on pp. 257, 261.)
7-7. Suppose .M; g/ is a Riemannian manifold and u 2 C 1 .M /. Use (5.29) and
(7.21) to prove Bochner’s formula:
1 ˇ ˇ2 ˝ ˛
2
j grad uj2 D ˇr 2 uˇ C grad.u/; grad u C Rc .grad u; grad u/ :
7-8. L ICHNEROWICZ ’ S T HEOREM : Suppose .M; g/ is a compact Riemannian

n-manifold, and there is a positive constant such that the Ricci tensor of
g satisfies Rc.v; v/ jvj2 for all tangent vectors v. If is any positive
eigenvalue of M , then n=.n 1/. [Hint: Use Probs. 2-23(c), 5-15, and
7-7.]
7-9. Prove parts (e) and (f) of Lemma 7.22 (properties of the Kulkarni–Nomizu
product).
7-10. Prove Proposition 7.34 (conformal invariance of the Cotton tensor in dimen-
sion 3).
7-11. Let .M; g/ be a Riemannian manifold of dimension n > 2. Define the con-
formal Laplacian L W C 1 .M / ! C 1 .M / by the formula
4.n 1/
Lu D u C S u;
.n 2/
where is the Laplace–Beltrami operator of g and S is its scalar curvature.
Prove that if gz D e 2f g for some f 2 C 1 .M /, and Lz denotes the conformal
Laplacian with respect to gz, then for every u 2 C 1 .M /,
nC2

z D L e n2
e 2 f Lu 2 fu :
Conclude that a metric gz conformal to g has constant scalar curvature

if and only if it can be expressed in the form gz D ' 4=.n2/ g, where ' is a
smooth positive solution to the Yamabe equation:
nC2
L' D ' n2 : (7.59)
224 7 Curvature
7-12. Let M be a smooth manifold and let r be any connection on TM . We can

define the curvature endomorphism of r by the same formula as in the
Riemannian case: R.X; Y /Z D rX rY Z rY rX Z rŒX;Y Z. Then r is
said to be a flat connection if R.X; Y /Z 0. Prove that the following are
equivalent:
(a) r is flat.
(b) For every point p 2 M , there exists a parallel local frame defined on a
neighborhood of p.
(c) For all p; q 2 M , parallel transport along an admissible curve segment
from p to q depends only on the path-homotopy class of .
(d) Parallel transport around any sufficiently small closed curve is the iden-
tity; that is, for every p 2 M , there exists a neighborhood U of p such that
if W Œa; b ! U is an admissible curve in U starting and ending at p, then
Pab W Tp M ! Tp M is the identity map.
7-13. Let G be a Lie group with a bi-invariant metric g. Show that the following
formula holds whenever X; Y; Z are left-invariant vector fields on G:
1
R.X; Y /Z D ŒZ; ŒX; Y :
4
(See Problem 5-8.)

7-14. Suppose
W M ; gz ! .M; g/ is a Riemannian submersion. Using the nota-
tion and results of Problem 5-6, prove O’Neill’s formula:
Rm.W; X; Y; Z/ı
D Rm W e

; X;

z 1 W
z Yz ; Z
2
D

; Xz V ; Yz ; Z

E
z V
1 D z
V z z
V E 1 D z
V z z
V E
W ; Y ; X; Z C W ; Z ; X; Y :
4 4
(Used on p. 258.)
7-15. Suppose .M; g/ is a 2-dimensional pseudo-Riemannian manifold of signature
.1; 1/, and p 2 M .
(a) Show that there is a smooth local frame .E1 ; E2 / in a neighborhood of
p such that g.E1 ; E1 / D g.E2 ; E2 / D 0.
(b) Show that there are smooth coordinates .x; y/ in a neighborhood of
p such that .dx/2 .dy/2 D fg for some smooth, positive real-valued
function f . [Hint: Use Prop. A.45 to show that there exist coordinates
.t; u/ in which E1 D @=@t , and coordinates .v; w/ in which E2 D @=@v,
and set x D u C w, y D u w.]
(c) Show that .M; g/ is locally conformally flat.
Chapter 8
Riemannian Submanifolds
This chapter has a dual purpose: first to apply the theory of curvature to Riemann-
ian submanifolds, and then to use these concepts to derive a precise quantitative
interpretation of the curvature tensor.
After introducing some basic definitions and terminology concerning Riemann-
ian submanifolds, we define a vector-valued bilinear form called the second fun-
damental form, which measures the way a submanifold curves within the ambient
manifold. We then prove the fundamental relationships between the intrinsic and
extrinsic geometries of a submanifold, including the Gauss formula, which relates
the Riemannian connection on the submanifold to that of the ambient manifold, and
the Gauss equation, which relates their curvatures. We then show how the second
fundamental form measures the extrinsic curvature of submanifold geodesics.
Using these tools, we focus on the special case of hypersurfaces, and use the
second fundamental form to define some real-valued curvature quantities called the
principal curvatures, mean curvature, and Gaussian curvature. Specializing even
more to hypersurfaces in Euclidean space, we describe various concrete geomet-
ric interpretations of these quantities. Then we prove Gauss’s Theorema Egregium,
which shows that the Gaussian curvature of a surface in R3 can be computed intrin-
sically from the curvature tensor of the induced metric.
In the last section, we introduce the promised quantitative geometric interpreta-
tion of the curvature tensor. It allows us to compute sectional curvatures, which are
just the Gaussian curvatures of 2-dimensional submanifolds swept out by geodesics
tangent to 2-planes in the tangent space. Finally, we compute the sectional curva-
tures of our frame-homogeneous model Riemannian manifolds—Euclidean spaces,
spheres, and hyperbolic spaces.
The Second Fundamental Form

Suppose .M; g/ is a Riemannian submanifold of a Riemannian manifold M ; gz .
endowed with the induced metric
Recall that this means that M is a submanifold of M

226 8 Riemannian Submanifolds
g D M gz (where M W M ,! M is the inclusion map). Our goal in this chapter is

to study the relationship between the geometry of M and that of M .
Although we focus our attention in this chapter on embedded submanifolds for
simplicity, the results we present are all local, so they apply in much greater gener-
ality. In particular, if M M is an immersed submanifold, then every point of M
has a neighborhood in M that is embedded in M , so the results of this chapter can
be applied by restricting to such a neighborhood. Even more generally, if .M; g/
is any Riemannian manifold and F W M ! M is an isometric immersion (meaning

that F gz D g), then again every point p 2 M has a neighborhood U M such that
F jU is an embedding, so the results apply to F .U / M . We leave it to the reader
to sort out the minor modifications in notation and terminology needed to handle
these more general situations.
The results in the first section of this chapter apply virtually without modification
to Riemannian submanifolds of pseudo-Riemannian manifolds (ones on which the
induced metric is positive definite), so we state most of our theorems in that case.
Recall that when the ambient metric gz is indefinite, this includes the assumption
that the induced metric g D M gz is positive definite. Some of the results can also
be extended to pseudo-Riemannian submanifolds of mixed signature, but there are
various pitfalls to watch out for in that case; so for simplicity we restrict to the
case of Riemannian submanifolds. See [dC92] for a thorough treatment of pseudo-
Riemannian submanifolds.
Also, most of these results can be adapted to manifolds and submanifolds with
boundary, simply by embedding everything in slightly larger manifolds without
boundary, but one might need to be careful about the statements of some of the
results when the submanifold intersects the boundary. Since the interaction of sub-
manifolds with boundaries is not our primary concern here, for simplicity we state
all of these results in the case of empty boundary only.
Throughout this chapter, therefore, we assume that M ; gz is a Riemannian
or pseudo-Riemannian manifold of dimension m, and .M; g/ is an embedded n-
dimensional Riemannian submanifold of M . We call M the ambient manifold. We
will denote covariant derivatives and curvatures associated .M; g/ in the usual
with
way (r, R, Rm, etc.), and write those associated with M z R,
; gz with tildes (r, z Rm,
f
etc.). We can unambiguously use the inner product notation hv; wi to refer either
to g or to gz, since g is just the restriction of gz to pairs of vectors in TM .
Our first main task is to compare the Levi-Civita connection of M with that of
M . The starting point for doing so is the orthogonal decomposition of sections of
the ambient tangent bundle T M jM into tangential and orthogonal components. Just
as we did for submanifolds of Rn in Chapter 4, we define orthogonal projection
maps called tangential and normal projections:
jM ! TM;
> W T M
?
W TM jM ! NM :
In terms of an adapted orthonormal frame .E1 ; : : : ; Em / for M in M , these are just

the usual projections onto span.E1 ; : : : ; En / and span.EnC1 ; : : : ; Em / respectively,
The Second Fundamental Form 227
Fig. 8.1: The second fundamental form
so both projections are smooth bundle homomorphisms (i.e., they are linear on fibers
and map smooth sections to smooth sections). If X is a section of T M jM , we often
use the shorthand notations X > D > X and X ? D ? X for its tangential and
normal projections.
If X; Y are vector fields in X.M /, we can extend them to vector fields on an
open subset of M (still denoted by X and Y ), apply the ambient covariant derivative
z and then decompose at points of M to get
operator r,

zX Y D r
r zX Y > C rzX Y ?: (8.1)
We wish to interpret the two terms on the right-hand side of this decomposition.
Let us focus first on the normal component. We define the second fundamental
form of M to be the map II W X.M / X.M / ! .NM / (read “two”) given by

II.X; Y / D rzX Y ?;
(Fig. 8.1). Since ?

where X and Y are extended arbitrarily to an open subset of M
maps smooth sections to smooth sections, II.X; Y / is a smooth section of NM .
The term first fundamental form, by the way, was originally used to refer to the
induced metric g on M . Although that usage has mostly been replaced by more
descriptive terminology, we seem unfortunately to be stuck with the name “second
fundamental form.” The word “form” in both cases refers to bilinear form, not dif-
ferential form.
Proposition 8.1 (Properties of the Second Fundamental Form). Suppose .M; g/

is an embedded
Riemannian submanifold of a Riemannian or pseudo-Riemannian
manifold M ; gz , and let X; Y 2 X.M /.
.
(a) II.X; Y / is independent of the extensions of X and Y to an open subset of M
1
(b) II.X; Y / is bilinear over C .M / in X and Y .
(c) II.X; Y / is symmetric in X and Y .
(d) The value of II.X; Y / at a point p 2 M depends only on Xp and Yp .
Proof. Choose particular extensions of X and Y to a neighborhood of M in M ,

and for simplicity denote the extended vector fields also by X and Y . We begin
by proving that II.X; Y / is symmetric in X and Y when defined in terms of these
extensions. The symmetry of the connection rz implies

II.X; Y / II.Y; X / D r z Y X ? D ŒX; Y ? :
zXY r
Since X and Y are tangent to M at all points of M , so is their Lie bracket (Cor.
A.40). Therefore ŒX; Y ? D 0, so II is symmetric.
Because rz X Y jp depends only on Xp , it follows that the value of II.X; Y / at p
depends only on Xp , and in particular
is independent of the extension chosen for X .
Because rz X Y is linear over C 1 M
in X , and every f 2 C 1 .M / can be extended
to a smooth function on a neighborhood of M in M , it follows that II.X; Y / is linear
over C 1 .M / in X . By symmetry, the same claims hold for Y . t
u
As a consequence of the preceding proposition, for every p 2 M and all vectors
v; w 2 Tp M , it makes sense to interpret II.v; w/ as the value of II.V; W / at p,
where V and W are any vector fields on M such that Vp D v and Wp D W , and we
will do so from now on without further comment.
We have not yet identified the tangential term in the decomposition of r zXY .
Proposition 5.12(b) showed that in the special case of a submanifold of a Euclidean
or pseudo-Euclidean space, it is none other than the covariant derivative with respect
to the Levi-Civita connection of the induced metric on M . The following theorem
shows that the same is true in the general case. Therefore, we can interpret the
second fundamental form as a measure of the difference between the intrinsic Levi-
Civita connection on M and the ambient Levi-Civita connection on M .
Theorem 8.2 (The Gauss Formula). Suppose .M; g/ is an embedded Riemannian

submanifold of a Riemannian or pseudo-Riemannian manifold M ; gz . If X; Y 2
X.M / are extended arbitrarily to smooth vector fields on a neighborhood of M in
M , the following formula holds along M :
z X Y D rX Y C II.X; Y /:
r
Proof. Because of the decomposition (8.1)and the definition of the second funda-
z X Y > D rX Y at all points of M .
mental form, it suffices to show that r
>
Define a map r W X.M / X.M / ! X.M / by

>
rX Y D rz X Y >;
where X and Y are extended arbitrarily to an open subset of M . We examined

a special case of this construction, in which gz is a Euclidean or pseudo-Euclidean
metric, in Example 4.9. It follows exactly as in that example that r > is a connection
on M , and exactly as in the proofs of Propositions 5.8 and 5.9 that it is symmetric
and compatible with g. The uniqueness of the Riemannian connection on M then
shows that r > D r. t
u
The Gauss formula can also be used to compare intrinsic and extrinsic covariant
derivatives along curves. If W I ! M is a smooth curve and X is a vector field
along that is everywhere tangent to M , then we can regard X as either a vector
field along in M or a vector field along in M . We let Dz t X and Dt X denote

its covariant derivatives along as a curve in M and as a curve in M , respectively.
The next corollary shows how the two covariant derivatives are related.
Corollary 8.3 (The Gauss Formula Along a Curve). Suppose .M; g/ is an em-
bedded
Riemannian submanifold of a Riemannian or pseudo-Riemannian manifold
M ; gz , and W I ! M is a smooth curve. If X is a smooth vector field along that
is everywhere tangent to M , then

Dz t X D Dt X C II 0 ; X : (8.2)
Proof. For each t0 2 I , we can find an adapted orthonormal frame .E1 ; : : : ; Em / in a

neighborhood of .t0 /. (Recall that our default assumption is that D m and
Pndim M
dim M D n.) In terms of this frame, X can be written X.t / D iD1 X i .t /Ei j.t/ .
Applying the product rule and the Gauss formula, and using the fact that each vector
field Ei is extendible, we get
n
X i
zt X D
D z 0 Ei
XP Ei C X i r
iD1
n
X i
D XP Ei C X i r 0 Ei C X i II. 0 ; Ei /
iD1

D Dt X C II 0 ; X : t
u
Although the second fundamental form is defined in terms of covariant deriva-

tives of vector fields tangent to M , it can also be used to evaluate extrinsic covari-
ant derivatives of normal vector fields, as the following proposition shows. To ex-
press it concisely, we introduce one more notation. For each normal vector field
N 2 .NM /, we obtain a scalar-valued symmetric bilinear form IIN W X.M /
X.M / ! C 1 .M / by
IIN .X; Y / D hN; II.X; Y /i: (8.3)
Let WN W X.M / ! X.M / denote the self-adjoint linear map associated with this
bilinear form, characterized by
hWN .X /; Y i D IIN .X; Y / D hN; II.X; Y /i: (8.4)
The map WN is called the Weingarten map in the direction of N. Because the
second fundamental form is bilinear over C 1 .M /, it follows that WN is linear over
C 1 .M / and thus defines a smooth bundle homomorphism from TM to itself.
Proposition 8.4 (The Weingarten Equation). Suppose .M; g/ is an embedded
Rie-
mannian submanifold of a Riemannian or pseudo-Riemannian manifold M ; gz . For
every X 2 X.M / and N 2 .NM /, the following equation holds:

rz X N > D WN .X /; (8.5)
.
when N is extended arbitrarily to an open subset of M
z X N is independent of the
Proof. Note that at points of M , the covariant derivative r
choice of extensions of X and N by Proposition 4.26. Let Y 2 X.M / be arbitrary,
extended to a vector field on an open subset of M . Since hN; Y i vanishes identically
along M and X is tangent to M , the following holds at points of M :
0 D X hN; Y i
˝ ˛ ˝ ˛
D r zXY
z X N; Y C N; r
˝ ˛ ˝ ˛
D r z X N; Y C N; rX Y C II.X; Y /
˝ ˛ ˝ ˛
D r z X N; Y C N; II.X; Y /
˝ ˛ ˝ ˛
D r z X N; Y C WN .X /; Y :
Since Y was an arbitrary vector field tangent to M , this implies

0D rz X N C WN .X / > D rz X N > C WN .X /;
which is equivalent to (8.5). t

u
In addition to describing the difference between the intrinsic and extrinsic con-
nections, the second fundamental form plays an even more important role in de-
scribing the difference between the curvature tensors of M and M . The explicit
formula, also due to Gauss, is given in the following theorem.
Theorem 8.5 (The Gauss Equation). Suppose .M; g/ is an embedded

Riemann-

ian submanifold of a Riemannian or pseudo-Riemannian manifold M ; gz . For
all W; X; Y; Z 2 X.M /, the following equation holds:
f
Rm.W; X; Y; Z/ D Rm.W; X; Y; Z/ hII.W; Z/; II.X; Y /i C hII.W; Y /; II.X; Z/i:
Proof. Let W; X; Y; Z be extended arbitrarily to an open subset of M . At points of

M , using the definition of the curvature and the Gauss formula, we get
f
Rm.W; zW r
X; Y; Z/ D hr zXY r zXrzW Y r z ŒW;X Y; Zi
z W .rX Y C II.X; Y // r
D hr z X .rW Y C II.W; Y //
rz ŒW;X Y; Zi:
Apply the Weingarten equation to each of the terms involving II (with II.X; Y / or
II.W; Y / playing the role of N ) to get
˝ ˛ ˝ ˛
f
Rm.W; X; Y; Z/ D r z W rX Y; Z II.X; Y /; II.W; Z/
˝ ˛ ˝ ˛ ˝ ˛
r z X rW Y; Z C II.W; Y /; II.X; Z/ rz ŒW;X Y; Z :
Decomposing each term involving r z into its tangential and normal components, we
see that only the tangential component survives, because Z is tangent to M . The
Gauss formula allows each such term to be rewritten in terms of r, giving
f
Rm.W; X; Y; Z/ D hrW rX Y; Zi hrX rW Y; Zi hrŒW;X Y; Zi
hII.X; Y /; II.W; Z/i C hII.W; Y /; II.X; Z/i
D hR.W; X /Y; Zi
hII.X; Y /; II.W; Z/i C hII.W; Y /; II.X; Z/i: t
u
There is one other fundamental submanifold equation, which relates the normal
part of the ambient curvature endomorphism to derivatives of the second fundamen-
tal form. We will not have need for it, but we include it here for completeness. To
state it, we need to introduce a connection on the normal bundle of a Riemannian
submanifold.
If .M; g/
is aRiemannian submanifold of a?Riemannian or pseudo-Riemannian
manifold M ; gz , the normal connection r W X.M / .NM / ! .NM / is
defined by

rX?
ND r zX N ?;
.
where N is extended to a smooth vector field on a neighborhood of M in M
Proposition 8.6. If .M; g/ is an embedded
Riemannian
submanifold of a Riemann-
ian or pseudo-Riemannian manifold M ; gz , then r ? is a well-defined connection
in NM , which is compatible with gz in the sense that for any two sections N1 ; N2
of NM and every X 2 X.M /, we have
˝ ? ˛ ˝ ?
˛
X hN1 ; N2 i D rX N1 ; N2 C N1 ; rX N2 :
We need the normal connection primarily to make sense of tangential covariant

derivatives of the second fundamental form. To do so, we make the following def-
initions. Let F ! M denote the bundle whose fiber at each point p 2 M is the
set of bilinear maps Tp M Tp M ! Np M . It is easy to check that F is a smooth
vector bundle over M , and that smooth sections of F correspond to smooth maps
X.M / X.M / ! .NM / that are bilinear over C 1 .M /, such as the second fun-
damental form. Define a connection r F in F as follows: if B is any smooth section
F
of F , let rX B be the smooth section of F defined by
F ?

rX B .Y; Z/ D rX .B.Y; Z// B rX Y; Z B Y; rX Z :
I Exercise 8.8. Prove that r F is a connection in F .
Now we are ready to state the last of the fundamental equations for submanifolds.
This equation was independently discovered (in the special case of surfaces in R3 )
by Karl M. Peterson (1853), Gaspare Mainardi (1856), and Delfino Codazzi (1868–
1869), and is sometimes designated by various combinations of these three names.
For the sake of simplicity we use the traditional but historically inaccurate name
Codazzi equation.
Theorem 8.9 (The Codazzi Equation). Suppose .M; g/ is an embedded

Riemann-

ian submanifold of a Riemannian or pseudo-Riemannian manifold M ; gz . For all
W; X; Y 2 X.M /, the following equation holds:
? F F
z
R.W; X /Y D rW II .X; Y / rX II .W; Y /: (8.6)
Proof. It suffices to show that both sides of (8.6) give the same result when we take
their inner products with an arbitrary smooth normal vector field N along M :
˝ ˛ ˝ F ˛ ˝ F ˛
z
R.W; X /Y; N D rW II .X; Y /; N rX II .W; Y /; N : (8.7)
We begin as in the proof of the Gauss equation: after extending the vector fields
to a neighborhood of M and applying the Gauss formula, we obtain
˝
f
Rm.W; X; Y; N / D r z W rX Y C II.X; Y / r z X rW Y C II.W; Y /
˛
r z ŒW;X Y; N :
Now when we expand the covariant derivatives, we need only pay attention to the
normal components. This yields
f .W; X; Y; N /
Rm
˝ F ˛
D II .W; rX Y / C rW II .X; Y / C II .rW X; Y / C II .X; rW Y / ; N
˝ F ˛
II .X; rW Y / C rX II .W; Y / C II .rX W; Y / C II .W; rX Y / ; N
˝ ˛
II .ŒW; X ; Y / ; N :
The terms involving rX Y and rW Y cancel each other in pairs, and three other
terms sum to zero because rW X rX W ŒW; X D 0. What remains is (8.7). u
t
Curvature of a Curve
By studying the curvatures of curves, we can give a more geometric interpreta-

tion of the second fundamental form. Suppose .M; g/ is a Riemannian or pseudo-
Riemannian manifold, and W I ! M is a smooth unit-speed curve in M . We define
the (geodesic) curvature of as the length of the acceleration vector field, which is
the function W I ! R given by
ˇ ˇ
.t / D ˇDt 0 .t /ˇ :
If is an arbitrary regular curve in a Riemannian manifold (not necessarily of unit

speed), we first find a unit-speed reparametrization z D ı ', and then define the
curvature of at t to be the curvature of z at ' 1 .t /. In a pseudo-Riemannian
manifold, the same approach works, but we have to restrict the definition to curves
such that j 0 .t /j is everywhere nonzero. Problem 8-6 gives a formula that can
be used in the Riemannian case to compute the geodesic curvature directly without
explicitly finding a unit-speed reparametrization.
From the definition, it follows that a smooth unit-speed curve has vanishing
geodesic curvature if and only if it is a geodesic, so the geodesic curvature of a
curve can be regarded as a quantitative measure of how far it deviates from being a
geodesic. If M D Rn with the Euclidean metric, the geodesic curvature agrees with
the notion of curvature
introduced
in advanced calculus courses.
Now suppose M ; gz is a Riemannian or pseudo-Riemannian manifold and
.M; g/ is a Riemannian submanifold. Every regular curve W I ! M has two dis-
tinct geodesic curvatures: its intrinsic curvature as a curve in M , and its extrinsic
curvature z as a curve in M . The second fundamental form can be used to compute
the relationship between the two.
Proposition 8.10 (Geometric Interpretation of II). Suppose .M; g/ is an embed-

ded
Riemannian
submanifold of a Riemannian or pseudo-Riemannian manifold
; gz , p 2 M , and v 2 Tp M .
M
(a) II.v; v/ is the gz-acceleration at p of the g-geodesic v .
(b) If v is a unit vector, then jII.v; v/j is the gz-curvature of v at p.
Proof. Suppose W ."; "/ ! M is any regular curve with .0/ D p and 0 .0/ D v.
Applying the Gauss formula (Corollary 8.3) to the vector field 0 along , we obtain
z t 0 D Dt 0 C II. 0 ; 0 /:
D
If is a g-geodesic in M , this formula simplifies to

z t 0 D II. 0 ; 0 /:
D
Both conclusions of the proposition follow from this. t

u
Note that the second fundamental form is completely determined by its values of
the form II.v; v/ for unit vectors v, by the following lemma.
Lemma 8.11. Suppose V is an inner product space, W is a vector space, and

B; B 0 W V V ! W are symmetric and bilinear. If B.v; v/ D B 0 .v; v/ for every
unit vector v 2 V , then B D B 0 .
Proof. Every vector v 2 V can be written v D y v for some unit vector vy, so the
bilinearity of B and B 0 implies B.v; v/ D B 0 .v; v/ for every v, not just unit vectors.
The result then follows from the following polarization identity, which is proved in
exactly the same way as its counterpart (2.2) for inner products:

B.v; w/ D 14 B.v C w; v C w/ B.v w; v w/ : t
u
Because the intrinsic and extrinsic accelerations of a curve are usually different,
it is generally not the case that a gz-geodesic that starts tangent to M stays in M ;
just think of a sphere
in Euclidean space, for example. A Riemannian submanifold
.M; g/ of M ; gz is said to be totally geodesic if every gz-geodesic that is tangent to
M at some time t0 stays in M for all t in some interval .t0 "; t0 C "/.
Proposition 8.12. Suppose .M; g/ is an embedded
Riemannian submanifold of a
Riemannian or pseudo-Riemannian manifold M ; gz , The following are equivalent:
.
(a) M is totally geodesic in M
(b) Every g-geodesic in M is also a gz-geodesic in M .
(c) The second fundamental form of M vanishes identically.
Proof. We will prove (a) ) (b) ) (c) ) (a). First assume that M is totally
geodesic. Let W I ! M be a g-geodesic. For each t0 2 I , let z W Iz ! M be the gz-
0 0
geodesic with z.t0 / D .t0 / and z .t0 / D .t0 /. The hypothesis implies that there
is some open interval I0 containing t0 such that z.t / 2 M for t 2 I0 . On I0 , the
Gauss formula (8.2) for z0 reads

0DD z t z0 D Dt z0 C II z0 ; z0 :
Because the first term on the right is tangent to M and the second is normal, the two
terms must vanish individually. In particular, Dt z0 0 on I0 , which means that z
is also a g-geodesic there. By uniqueness of geodesics, therefore, D z on I0 , so it
follows in turn that is a gz-geodesic there. Since the same is true in a neighborhood
of every t0 2 I , it follows that is a gz-geodesic on its whole domain.
Next assume that every g-geodesic is a gz-geodesic. Let p 2 M and v 2 Tp M be
arbitrary, and let D v W I ! M be the g-geodesic with .0/ D p and 0 .0/ D v.
The hypothesis implies that is also a gz-geodesic. Thus D z t 0 D Dt 0 D 0, so the
0 0
Gauss formula yields II. ; / D 0 along . In particular, II.v; v/ D 0. By Lemma
8.11, this implies that II is identically zero.
Finally, assume that II 0, and let z W I ! M be a gz-geodesic such that z.t0 / 2
0
M and z .t0 / 2 TM for some t0 2 I . Let W I ! M be the g-geodesic with the same
initial conditions: .t0 / D z.t0 / and 0 .t0 / D z0 .t0 /. The Gauss formula together
with the hypothesis II 0 implies that D z t D Dt D 0, so is also a gz-geodesic.
By uniqueness of geodesics, therefore, z D on the intersection of their domains,
which implies that z.t/ lies in M for t in some open interval around t0 . t
u
Hypersurfaces
Now we specialize the preceding considerations to the case in which M is a hyper-
surface (i.e., a submanifold of codimension 1) in M . Throughout this section, our
default assumption is that .M; g/ is an embedded n-dimensional
Riemannian sub-
manifold of an .n C 1/-dimensional Riemannian manifold M ; gz . (The analogous
formulas in the pseudo-Riemannian case are a little different; see Problem 8-19.)
Hypersurfaces 235
In this situation, at each point of M there are exactly two unit normal vectors.
In terms of any local adapted orthonormal frame .E1 ; : : : ; EnC1 /, the two choices
are ˙EnC1 . In a small enough neighborhood of each point of M , therefore, we can
always choose a smooth unit normal vector field along M .
If both M and M are orientable, we can use an orientation to pick out a global
smooth unit normal vector field along all of M . In general, though, this might or
might not be possible. Since all of our computations in this chapter are local, we
will always assume that we are working in a small enough neighborhood that a
smooth unit normal field exists. We will address as we go along the question of how
various quantities depend on the choice of normal vector field.
The Scalar Second Fundamental Form and the Shape Operator
Having chosen a distinguished smooth unit normal vector field N on the hypersur-
face M M , we can replace the vector-valued second fundamental form II by a
simpler scalar-valued form. The scalar second fundamental
form of M is the sym-
metric covariant 2-tensor field h 2 †2 T M defined by h D IIN (see (8.3)); in
other words,
h.X; Y / D hN; II.X; Y /i: (8.8)
Using the Gauss formula r z X Y D rX Y C II.X; Y / and noting that rX Y is orthog-
onal to N , we can rewrite the definition as
˝ ˛
zXY :
h.X; Y / D N; r (8.9)
Also, since N is a unit vector spanning NM at each point, the definition of h is

equivalent to
II.X; Y / D h.X; Y /N: (8.10)
Note that replacing N by N multiplies h by 1, so the sign of h depends on which
unit normal is chosen; but h is otherwise independent of the choices.
The choice of unit normal field also determines a Weingarten map WN W X.M / !
X.M / by (8.4); in the case of a hypersurface, we use the notation s D WN and call
it the shape operator of M . Alternatively, we can think of s as the .1; 1/-tensor field
on M obtained from h by raising an index. It is characterized by
hsX; Y i D h.X; Y / for all X; Y 2 X.M /:
Because h is symmetric, s is a self-adjoint endomorphism of TM , that is,
hsX; Y i D hX; sY i for all X; Y 2 X.M /:
As with h, the sign of s depends on the choice of N .

In terms of the tensor fields h and s, the formulas of the last section can be
rewritten somewhat more simply. For this purpose, we will use two tensor operations
defined in Chapter 7: the Kulkarni–Nomizu product of symmetric 2-tensors h; k is
h k.w; x; y; z/ D h.w; z/k.x; y/ C h.x; y/k.w; z/

h.w; y/k.x; z/ h.x; z/k.w; y/;
and the exterior covariant derivative of a smooth symmetric 2-tensor field T is
.DT /.x; y; z/ D .rT /.x; y; z/ C .rT /.x; z; y/:
Theorem 8.13 (Fundamental Equations for a Hypersurface).

Suppose .M; g/ is
a Riemannian hypersurface in a Riemannian manifold M ; gz , and N is a smooth
unit normal vector field along M .
(a) T HE G AUSS F ORMULA FOR A H YPERSURFACE : If X; Y 2 X.M / are extended
, then
to an open subset of M
z X Y D rX Y C h.X; Y /N:
r
(b) T HE G AUSS F ORMULA FOR A C URVE IN A H YPERSURFACE : If W I ! M

is a smooth curve and X W I ! TM is a smooth vector field along , then
z t X D Dt X C h. 0 ; X /N:
D
(c) T HE W EINGARTEN E QUATION FOR A H YPERSURFACE : For every X 2

X.M /,
rz X N D sX: (8.11)
(d) T HE G AUSS E QUATION FOR A H YPERSURFACE : For all W; X; Y; Z 2 X.M /,
e
Rm.W; X; Y; Z/ D Rm.W; X; Y; Z/ 12 .h h/.W; X; Y; Z/:
(e) T HE C ODAZZI E QUATION FOR A H YPERSURFACE : For all W; X; Y 2 X.M /,
e
Rm.W; X; Y; N / D .Dh/.Y; W; X /: (8.12)
Proof. Parts (a), (b), and (d) follow immediately from substituting (8.10) into the
general versions of the Gauss formula and Gauss equation. To prove (c), note first

that the general version of the Weingarten equation can be written rz X N > D sX .
˝ ˛
Since r z X N; N D 1 X jN j2 D 0, it follows that r z X N is tangent to M , so (c)
2
follows.
To prove the hypersurface Codazzi equation, note that the fact that N is a unit
vector field implies ˝ ? ˛
0 D X jN j2gz D 2 rX N; N gz :
Since N spans the normal bundle, this implies that N is parallel with respect to the
normal connection. Moreover,
Hypersurfaces 237
F ?
rW II .X; Y / D rW .II.X; Y // II.rW X; Y / II.X; rW Y /
?
D rW .h.X; Y /N / II.rW X; Y / II.X; rW Y /

D W .h.X; Y // h.rW X; Y / h.X; rW Y / N
D rW .h/.X; Y /N:
Inserting this into the general form (8.6) of the Codazzi equation and using the fact
that rh is symmetric in its first two indices yields
f
Rm.W; X; Y; N / D rW .h/.X; Y / rX .h/.W; Y /
D .rh/.X; Y; W / .rh/.W; Y; X /
D .rh/.Y; X; W / .rh/.Y; W; X /;

u
Principal Curvatures
At every point p 2 M , we have seen that the shape operator s is a self-adjoint linear
endomorphism of the tangent space Tp M . To analyze such an operator, we recall
some linear-algebraic facts about self-adjoint endomorphisms.
Lemma 8.14. Suppose V is a finite-dimensional inner product space and s W V ! V

is a self-adjoint linear endomorphism. Let C denote the set of unit vectors in V .
There is a vector v0 2 C where the function v 7! hsv; vi achieves its maximum
among elements of C , and every such vector is an eigenvector of s with eigenvalue
0 D hsv0 ; v0 i.
I Exercise 8.15. Use the Lagrange multiplier rule (Prop. A.29) to prove this lemma.
Proposition 8.16 (Finite-Dimensional Spectral Theorem). Suppose V is a finite-

dimensional inner product space and s W V ! V is a self-adjoint linear endomor-
phism. Then V has an orthonormal basis of s-eigenvectors, and all of the eigenval-
ues are real.
Proof. The proof is by induction on n D dim V . The n D 1 result is easy, so assume

that the theorem holds for some n 1 and suppose dim V D n C 1. Lemma 8.14
shows that s has a unit eigenvector b0 with a real eigenvalue 0 . Let B V be
the span of b0 . Since s.B/ B, self-adjointness of s implies s B ? B ? . The
inductive hypothesis applied to sjB ? implies that B ? has an orthonormal basis
.b1 ; : : : ; bn / of s-eigenvectors with real eigenvalues, and then .b0 ; b1 ; : : : ; bn / is the
desired basis of V . t
u
Applying this proposition to the shape operator s W Tp M ! Tp M , we see that
s has real eigenvalues 1 ; : : : ; n , and there is an orthonormal basis .b1 ; : : : ; bn / for
Tp M consisting of s-eigenvectors, with sbi D i bi for each i (no summation). In
this basis, both h and s are represented by diagonal matrices, and h has the expres-
sion
h.v; w/ D 1 v 1 w 1 C C n v n w n :
The eigenvalues of s at a point p 2 M are called the principal curvatures
of M at p, and the corresponding eigenspaces are called the principal directions.
The principal curvatures all change sign if we reverse the normal vector, but the
principal directions and principal curvatures are otherwise independent of the choice
of coordinates or bases.
There are two combinations of the principal curvatures that play particularly
important roles for hypersurfaces. The Gaussian curvature is defined as K D det.s/,
and the mean curvature as H D .1=n/ tr.s/ D .1=n/ trg .h/. Since the determinant
and trace of a linear endomorphism are basis-independent, these are well defined
once a unit normal is chosen. In terms of the principal curvatures, they are
1
K D 1 2 n ; HD .1 C C n /;
n
as can be seen by expressing s in terms of an orthonormal basis of eigenvectors. If
N is replaced by N , then H changes sign, while K is multiplied by .1/n .
Computations in Semigeodesic Coordinates
Semigeodesic coordinates (Prop. 6.41) provide an extremely convenient tool for

computing
the invariants of hypersurfaces.
Let M ; gz be an .nC1/-dimensional Riemannian manifold, and let x 1 ; : : : ; x n ; v
be semigeodesic coordinates on an open subset U M . (For example, they might
1
be Fermi coordinates for the hypersurface M0 D v .0/; see Example 6.43.) For
each real number a such that v 1 .a/ ¤ ¿, the level set Ma D v 1 .a/ is a properly
embedded hypersurface in U . Let ga denote the induced metric on Ma . Corollary
6.42 shows that gz is given by
n
X
gz D dv 2 C g˛ˇ x 1 ; : : : ; x n ; v dx ˛ dx ˇ : (8.13)
˛;ˇ D1

The restrictions of x 1 ; : : : ; x n / give smooth coordinates for each hypersurface Ma ,
and in those coordinates the induced metric ga is given by ga D g˛ˇ dx ˛ dx ˇ jvDa .
(Here we use the summation convention with Greek indices running from 1 to n.)
The vector field @v D @nC1 restricts to a unit normal vector field along each hyper-
surface Ma .
As the next proposition shows, semigeodesic coordinates give us a simple for-
mula for the second fundamental forms of all of the submanifolds Ma at once.

Proposition 8.17. With notation as above, the components in x 1 ; : : : ; x n -coor-
dinates of the scalar second fundamental form, the shape operator, and the mean
Hypersurfaces 239
curvature of .Ma ; ga / .denoted by ha ; sa ; and Ha ; respectively/ with respect to the

normal N D @v are given by
ˇ
1 ˇ
.ha /˛ˇ D @v g˛ˇ ˇ ;
2 vDa
ˇ
1 ˇ
.sa /˛ˇ D g ˛ @v gˇ ˇ ;
2 vDa
ˇ
1 ˇ
Ha D g ˛ˇ @v g˛ˇ ˇ :
2n vDa
Proof. The normal component of r z @ @ˇ is

z nC1 @v , which Corollary 6.42 shows
˛ ˛ˇ
is equal to 12 @v g˛ˇ @r (noting that the roles of g and gy in that corollary are being
played here by gz and ga , respectively). Equation (8.9) evaluated at points of Ma
gives ˝ ˛ ˝ ˛
.ha /˛ˇ D r z @ @ˇ ; N D 1 @v g˛ˇ @v ; @v D 1 @v g˛ˇ :
˛ gz 2 gz 2
The formulas for sa and Ha follow by using .g ˛ / (the inverse matrix of .g˛ /) to
raise an index and then taking the trace. t
u
Minimal Hypersurfaces
A natural question that has received a great deal of attention over the past century is
this: Given a simple closed curve C in R3 , is there an embedded or immersed sur-
face M with @M D C that has least area among all surfaces with the same bound-
ary? If so, what is it? Such surfaces are models of the soap films that are produced
when a closed loop of wire is dipped in soapy water.
More generally, we can consider the analogous question for hypersurfaces in
Riemannian manifolds. Suppose M is a compact codimension-1 submanifold with
nonempty boundary in an .n C 1/-dimensional Riemannian manifold M ; gz . By
analogy with the case of surfaces in R3 , it is traditional to use the term area to
refer to the n-dimensional volume of M with its induced Riemannian metric, and
to say that M is area-minimizing if it has the smallest area among all compact
embedded hypersurfaces in M with the same boundary. One key observation is the
following theorem, which is an analogue for hypersurfaces of Theorem 6.4 about
length-minimizing curves.
Theorem 8.18. Let M be a compact codimension-1 submanifold

with nonempty
boundary in an .n C 1/-dimensional Riemannian manifold M ; gz . If M is area-
minimizing, then its mean curvature is identically zero.
Proof. Let g denote the induced metric on M . The fact that M minimizes area
among hypersurfaces with the same boundary means, in particular, that it minimizes
area among small perturbations of M in a neighborhood of a single point. We will
exploit this idea to prove that M must
have zero mean curvature everywhere.
Let p 2 Int M be arbitrary, let x 1 ; : : : ; x n ; v be Fermi coordinates for M on an
open set Uz M containing p (see Example 6.43), and let U D Uz \ M . By taking
Fig. 8.2: The hypersurface Mt
Uz sufficiently small, we can arrange

that U is a regular coordinate ball in M (see p.
374) and Uz \ @M D ¿. We use x 1 ; : : : ; x n as coordinates on M , and observe the
summation convention with Greek indices running from 1 to n.
Let ' be an arbitrary smooth real-valued function on M with compact support in
by
U . For sufficiently small t , define a set Mt M
˚
Mt D .M X U / [ z 2 Uz W v.z/ D t ' x 1 .z/; : : : ; x n .z/ :
(See Fig. 8.2.) Then Mt is an embedded smooth hypersurface in M , which agrees

with M outside of U and which coincides with the graph of v D t ' in Uz . Let
ft W U ! Uz be the graph parametrization of Mt \ Uz , given in Fermi coordinates by

ft x 1 ; : : : ; x n D x 1 ; : : : ; x n ; t '.x/ : (8.14)
Using this map, for each t we can define a diffeomorphism Ft W M ! Mt by

z; z 2 M X supp ';
Ft .z/ D
ft .z/; z 2 U:
For each t , let gyt D Mt gz denote the induced Riemannian metric on Mt , and
let gt D Ft gyt D Ft gz denote the pulled-back metric on M . When t D 0, we have
M0 D M , and both g0 and gy0 are equal to the induced metric g on M . Since gz
is given by (8.13) in Fermi coordinates, a simple computation shows that in U ,
gt D Ft gz has the coordinate expression gt D .gt /˛ˇ dx ˛ dx ˇ , where
@' @'
.gt /˛ˇ D t 2 ˛
.x/ ˇ .x/ C g˛ˇ x; t '.x/ ; (8.15)
@x @x
while on M X U , gt is equal to g and thus is independent of t .
Since each Mt is a smooth hypersurface with the same boundary as M , our
hypothesis guarantees that Area .Mt ; gyt / achieves a minimum at t D 0. Because Ft
is an isometry from .M; gt / to .Mt ; gyt /, we can express this area as follows:
Area .Mt ; gyt / D Area .M; gt / D Area .M X U; g/ C Area .U; gt /:

Hypersurfaces 241
The first term on the right is independent of t , and we can compute the second term
explicitly in coordinates x 1 ; : : : ; x n on U :
Z p
Area .U; gt / D det gt dx 1 dx n ;
U

where det gt denotes the determinant of the matrix 1 .gt /˛ˇ defined by (8.15). The
integrand above is a smooth function of t and x ; : : : ; x n , so the area is a smooth
function of t . We have
ˇ Z ˇ
d ˇˇ @ ˇˇ p
ˇ Area .Mt ; gyt / D ˇ det gt dx 1 dx n
dt tD0 U @t tD0
Z ˇ (8.16)
1 1=2 @ ˇ
ˇ 1 n
D .det g/ .det gt / dx dx :
U 2 @t ˇtD0
(The differentiation under the integral sign is justified because the integrand is
smooth and has compact support in U .)
To compute the derivative of the determinant, note that the expansion by minors
along, say, row ˛ shows that the partial derivative of det with respect to the matrix
entry in position .˛; ˇ/ is equal to the cofactor cof ˛ˇ , and thus by the chain rule,
ˇ ˇ
@ ˇˇ ˛ˇ @ ˇ
ˇ
ˇ .det gt / D cof .gt /˛ˇ : (8.17)
@t tD0 @t ˇtD0
On the other hand, Cramer’s rule shows that the .˛; ˇ/ component of the inverse
matrix is given by g ˛ˇ D .det g/1 cof ˛ˇ . Thus from (8.17) and (8.15) we obtain
ˇ ˇ
@ ˇˇ ˛ˇ @ ˇ
ˇ @g˛ˇ
ˇ .det g / D .det g/g .gt /˛ˇ D .det g/g ˛ˇ ':
@t ˇtD0
t
@t tD0 @v
Inserting this into (8.16) and using the result of Proposition 8.17, we conclude that
ˇ Z
d ˇˇ 1 @g˛ˇ
ˇ Area .M t ; y
g t / D .det g/1=2 g ˛ˇ ' dx 1 dx n
dt tD0 U 2 @v
Z (8.18)
D n H ' d Vg ;
U
where H is the mean curvature

R of .M; g/. Since Area .Mt ; gyt / attains a minimum at
t D 0, we conclude that U H ' d Vg D 0 for every such '.
Now suppose for the sake of contradiction that H.p/ ¤ 0. If H.p/>0, we can let '
be a smooth nonnegative bump function that is positive at p and supportedR in a small
neighborhood of p on which H > 0. The argument above shows that U H ' d Vg D
0, which is impossible because the integrand is nonnegative on U and positive on an
open set. A similar argument rules out H.p/ < 0. Since p was an arbitrary point in
Int M , we conclude that H 0 on Int M , and then by continuity on all of M . t
u
Because of the result of Theorem 8.18, a hypersurface (immersed or embed-

ded, with or without boundary) that has mean curvature identically equal to zero is
called a minimal hypersurface (or a minimal surface when it has dimension 2). It is
an unfortunate historical accident that the term “minimal hypersurface” is defined in
this way, because in fact, a minimal hypersurface is just a critical point for the area,
not necessarily area-minimizing. It can be shown that as in the case of geodesics, a
small enough piece of every minimal hypersurface is area-minimizing.
As a complement to the above theorem about hypersurfaces that minimize area
with fixed boundary, we have the following result about hypersurfaces that minimize
area while enclosing a fixed volume.

Theorem 8.19. Suppose M ; g is a Riemannian .n C 1/-manifold, D M is a
compact regular domain, and M D @D. If M has the smallest surface area among
boundaries of compact regular domains with the same volume as D, then M has
constant mean curvature (computed with respect to the outward unit normal).
Proof. Let g denote the induced metric on M . Assume for the sake of contradiction
that the mean curvature H of M is not constant, and let p; q 2 M be points such
that H.p/ < H.q/.
Since M is compact, it has an "-tubular
neighborhood for some " > 0 by Theorem
5.25. As in the previous proof, let x 1 ; : : : ; x n ; v be Fermi coordinates for M on
an open set Uz M containing p, and let U D Uz \ M . We may assume that U
is a regular coordinate ball in M and the image of the chart is a set of the form
Uy ."; "/ for some open subset Uy Rn . Similarly, let y 1 ; : : : ; y n ; w be Fermi
coordinates for M on an open set W M containing q and satisfying the analogous

conditions, and let W D W \ M . By replacing v with its negative if necessary,
we can arrange that D \ Uz is the set where v 0, and similarly for w. Also, by
shrinking both domains, we can assume that the mean curvature of M satisfies H
H1 on U and H H2 on W , where H1 ; H2 are constants such that H.p/ < H1 <
H2 < H.q/.
Let ' and be smooth real-valued functions on M , Rwith ' compactly R supported
in U and compactly supported in W , and satisfying U ' d Vg D W d Vg D 1.
For sufficiently small s; t 2 R, define a subset Ds;t M as follows:
˚
Ds;t D z 2 Uz W v.z/ s' x 1 .z/; : : : ; x n .z/
˚
[ z2W W w.z/ t y 1 .z/; : : : ; y n .z/

[ D X Uz [ W ;
and let Ms;t D @Ds;t , so D0;0 D D and M0;0 D M (see Fig. 8.3). For sufficiently
small s and t , the set Ds;t is a regular domain and Ms;t is a compact smooth hy-
persurface, and Vol.Ds;t / and Area .Ms;t / are both smooth functions of .s; t /. For
convenience, write V .s; t / D Vol.Ds;t / and A.s; t / D Area .Ms;t /.
The same argument that led to (8.18) shows that
Z Z
@A @A
.0; 0/ D n H ' d Vg ; .0; 0/ D n H d Vg :
@s U @t W
Hypersurfaces 243
Fig. 8.3: The domain Ds;t
To compute the partial derivatives of the volume, we just note that if we hold t D 0
fixed and let s vary, the only change in volume occurs in the part of Ds;t contained
in Uz , so the fundamental theorem of calculus gives
ˇ
@V d ˇˇ
.0; 0/ D Vol Ds;0 \ Uz
@s ds ˇsD0
ˇ Z Z s'.x/ p !
d ˇˇ
D det gz.x; v/ dv dx 1 dx n
ds ˇsD0 U "
Z ˇ Z s'.x/ p !
d ˇˇ
D det gz.x; v/ dv dx 1 dx n
U ds ˇsD0 "
Z p Z
D '.x/ det gz.x; 0/ dx 1 dx n D ' d Vg D 1;
U U
where the differentiation under the integral sign in the third line is justified just
like (8.16), and in the last line we used the fact that g˛ˇ .x/ D gz˛ˇ .x; 0/ in these
coordinates. Similarly, @V =@t .0; 0/ D 1.
Because V .0; 0/ D Vol.D/ and @V =@t .0; 0/ ¤ 0, the implicit function theorem
guarantees that there is a smooth function W .ı; ı/ ! R for some ı > 0 such that
V .s; .s// Vol.D/. The chain rule then implies
ˇ
d ˇˇ @V @V
0D ˇ V .s; .s// D .0; 0/ C 0 .0/ .0; 0/ D 1 C 0 .0/:
ds sD0 @s @t
Thus 0 .0/ D 1.

Our hypothesis that M minimizes area implies that
ˇ
d ˇˇ @A @A
0D ˇ A.s; .s// D .0; 0/ C 0 .0/ .0; 0/
ds sD0 @s @t
Z Z
D n H ' d Vg C n H d Vg ;
U W
R R
and thus U RH ' d Vg D RW H d Vg . But our choice of R U and V together with
the
R fact that U ' d V g D W d V g D 1 guarantees that U H ' d Vg H 1 < H 2
W H d V g , which is a contradiction. t
u
We do not pursue minimal or constant-mean-curvature hypersurfaces any further
in this book, but you can find a good introductory treatment in [CM11].
Hypersurfaces in Euclidean Space

Now we specialize even further, to hypersurfaces in Euclidean space. In this section,
we assume that M RnC1 is an embedded n-dimensional submanifold with the
induced Riemannian metric. The Euclidean metric will be denoted as usual by gx,
and covariant derivatives and curvatures associated with gx will be indicated by a
bar. The induced metric on M will be denoted by g.
In this setting, because Rm 0, the Gauss and Codazzi equations take even sim-
pler forms:
1
2
hhD Rm; (8.19)
Dh D 0; (8.20)
or in terms of a local frame for M ,
hil hj k hik hj l D Rij kl ; (8.21)

hij Ik hikIj D 0: (8.22)
In particular, this means that the Riemann curvature tensor of a hypersurface

in RnC1 is completely determined by the second fundamental form. A symmetric
2-tensor field that satisfies Dh D 0 is called a Codazzi tensor, so (8.20) can be
expressed succinctly by saying that h is a Codazzi tensor.
I Exercise 8.20. Show that a smooth 2-tensor field h on a Riemannian manifold is a

Codazzi tensor if and only if both h and rh are symmetric.
The equations (8.19)–(8.20) can be viewed as compatibility conditions for the

existence of an embedding or immersion into Euclidean space with prescribed first
and second fundamental forms. If .M; g/ is a Riemannian n-manifold and h is a
given smooth symmetric 2-tensor field on M , then Theorem 8.13 shows that these
two equations are necessary conditions for the existence of an isometric immersion
M ! RnC1 for which h is the scalar second fundamental form. (Note that an im-
mersion is locally an embedding, so the theorem applies in a neighborhood of each
Hypersurfaces in Euclidean Space 245
Fig. 8.4: Geometric interpretation of h.v; v/
point.) It is a remarkable fact that the Gauss and Codazzi equations are actually
sufficient, at least locally. A sketch of a proof of this fact, called the fundamental
theorem of hypersurface theory, can be found in [Pet16, pp. 108–109].
In the setting of a hypersurface M RnC1 , we can give some very concrete
geometric interpretations of the quantities we have defined so far. We begin with
curves. For every unit vector v 2 Tp M , let D v W I ! M be the g-geodesic in M
with initial velocity v. Then the Gauss formula shows that the ordinary Euclidean
x t 0 .0/ D h.v; v/Np (Fig. 8.4). Thus jh.v; v/j is
acceleration of at 0 is 00 .0/ D D
the Euclidean curvature of at 0, and h.v; v/ D h 00 .0/; Np i > 0 if and only if 00 .0/
points in the same direction as Np . In other words, h.v; v/ is positive if is curving
in the direction of Np , and negative if it is curving away from Np .
The next proposition shows that this Euclidean curvature can be interpreted in
terms of the radius of the “best circular approximation,” as mentioned in Chapter 1.
Proposition 8.21. Suppose W I ! Rm is a unit-speed curve, t0 2 I , and .t0 / ¤ 0.

(a) There is a unique unit-speed parametrized circle c W R ! Rm , called the oscu-
lating circle at .t 0/ , with the property that c and have the same position,
velocity, and acceleration at t D t0 .
(b) The Euclidean curvature of at t0 is .t0 / D 1=R, where R is the radius of
the osculating circle.
Proof. An easy geometric argument shows that every circle in Rm with center q and
radius R has a unit-speed parametrization of the form
t t t t
0 0
c.t / D q C R cos v C R sin w;
R R
where .v; w/ is a pair of orthonormal vectors in Rm . By direct computation, such a
parametrization satisfies
1
c.t0 / D q C Rv; c 0 .t0 / D w; c 00 .t0 / D v:
R
Thus if we put
1 1
RD D ; v D R 00 .t0 /; w D 0 .t0 /; q D .t0 / Rv;
j 00 .t0 /j .t0 /
we obtain a circle satisfying the required conditions, and its radius is equal to
1=.t0 / by construction. Uniqueness is left as an exercise. u
t
I Exercise 8.22. Complete the proof of the preceding proposition by proving uniqueness
of the osculating circle.
Computations in Euclidean Space

When we wish to compute the invariants of a Euclidean hypersurface M RnC1 ,
it is usually unnecessary to go to all the trouble of computing Christoffel symbols.
Instead, it is usually more effective to use either a defining function or a parametriza-
tion to compute the scalar second fundamental form, and then use (8.21) to compute
the curvature. Here we describe several contexts in which this computation is not too
hard.
Usually the computations are simplest if the hypersurface is presented in terms of
a local parametrization. Suppose M RnC1 is a smooth embedded hypersurface,
nC1
1 let X nW U ! R nbe a smooth local parametrization of M . The coordinates
and
u ; : : : ; u on U R thus give local coordinates for M . The coordinate vector
fields @i D @=@ui push forward to vector fields dX .@i / on M , which we can view as
nC1
1 T R nC1
sections of the restricted tangent bundle jM , or equivalently
as RnC1 -valued
functions. If we think of X.u/ D X .u/; : : : ; X .u/ as a vector-valued function
of u, these vectors can be written as

dXu .@i / D @i X.u/ D @i X 1 .u/; : : : ; @i X nC1 .u/ :
For simplicity, write Xi D @i X .

Once these vector fields are computed, a unit normal field can be computed
as follows: Choose any coordinate vector field @=@x j0 that is not contained in
span .X1 ; : : : ; Xn / (there will always be one, at least in a neighborhood
of each point).
Then apply the Gram–Schmidt algorithm to the local frame X1 ; : : : ; Xn ; @=@x j0
along M to obtain an adapted orthonormal frame .E1 ; : : : ; EnC1 /. The two choices
of unit normal are N D ˙EnC1 .
The next proposition gives a formula for the second fundamental form that is
often easy to use for computation.
Proposition 8.23. Suppose M RnC1 is an embedded hypersurface, X W U ! M

is a smooth local parametrization of M , .X1 ; : : : ; Xn / is the local frame for TM
determined by X , and N is a unit normal field on M . Then the scalar second fun-
damental form is given by
2
@ X
h Xi ; Xj D ;N : (8.23)
@ui @uj

Proof. Let u0 D u10 ; : : : ; un0 be an arbitrary point of U and let p D X.u0 / 2 M . For

each i 2 f1; : : : ; ng, the curve .t / D X u10 ; : : : ; ui0 C t; : : : ; un0 is a smooth curve in
M whose initial velocity is Xi . Regarding the normal field N as a smooth map
from M to RnC1 , we have
@ x X N.X.u0 //:
N.X.u0 // D .N ı /0 .0/ D r
@ui i
Because Xj D @X=@uj is tangent to M and N is normal, the following expression

is zero for all u 2 U :
@X
.u/; N.X.u// :
@uj
Differentiating with respect to ui and using the product rule for ordinary inner prod-
ucts in RnC1 yields
2

@ X @X x
0D .u/; N.X.u// C .u/; rXi .u/ N.X.u// :
@ui @uj @uj
By the Weingarten equation (8.11), the last term on the right becomes
˝ ˛
Xj .u/; s.Xi .u// D h.Xj .u/; Xi .u//:
Inserting this above yields (8.23). t

u
Here is an application of this formula: it shows how the principal curvatures give
a concise description of the local shape of an embedded hypersurface by approxi-
mating the surface with the graph of a quadratic function.
Proposition 8.24. Suppose M RnC1 is a Riemannian hypersurface. Let p 2 M ,

and let 1 ; : : : ; n denote the principal curvatures of M at p with respect to some
choice of unit normal. Then there is an isometry ' W RnC1 ! RnC1 that takes p
to the origin and takes a neighborhood of p in M to a graph of the form x nC1 D
f .x 1 ; : : : ; x n /, where
1
f .x/ D 1 .x 1 /2 C C n .x n /2 C O jxj3 : (8.24)
2
Proof. Replacing M by its image under a translation and a rotation (which are
Euclidean isometries), we may assume that p is the origin and Tp M is equal to the
span of .@1 ; : : : ; @n /. Then after reflecting in the x 1 ; : : : ; x n -hyperplane if neces-
sary, we may assume that the chosen unit normal is .0; : : : ; 0; 1/. By an orthogonal
transformation in the first n variables, we can also arrange that the scalar second fun-
damental form at 0 is diagonal with respect to the basis .@1 ; : : : ; @n /, with diagonal
entries .1 ; : : : ; n /.
It follows from the implicit function theorem that there is some neighborhood
U of 0 such that M \ U is the graph of a smooth function of the form x nC1 D
f x 1 ; : : : ; x n with f .0/ D 0. A smooth local parametrization of M is then given by

X.u/ D u1 ; : : : ; un ; f .u/ , and the fact that T0 M is spanned by @=@x 1 ; : : : ; @=@x n
guarantees that @1 f .0/ D D @n f .0/ D 0. Because Xi D @=@x i at 0, Proposition
8.23 then yields

@ @ @2 f @2 f
h i
; j
D 0; : : : ; 0; i j
.0/ ; .0; : : : ; 0; 1/ D i j .0/:
@x @x @x @x @x @x
It follows from our normalization that the matrix of second derivatives of f at 0

is diagonal, and its diagonal entries are the principal curvatures of M at that point.
Then (8.24) follows from Taylor’s theorem. t
u
Here is another approach. When it is practical to write down a smooth vector
field N D N i @i on an open subset of RnC1 that restricts to a unit normal vector
field along M , then the shape operator can be computed straightforwardly using the
Weingarten equation and observing that the Euclidean covariant derivatives of N
are just ordinary directional derivatives in Euclidean space. Thus for every vector
X D X j @j tangent to M , we have
nC1
X
xX N D
sX D r X j .@j N i /@i : (8.25)
i;j D1
One common way to produce such a smooth vector field is to work with a local
defining function for M : Recall that this is a smooth real-valued function defined
on some open subset U RnC1 such that U \ M is a regular level set of F (see
Prop. A.27). The definition ensures that grad F (the gradient of F with respect to
gx) is nonzero on some neighborhood of M \ U , so a convenient choice for a unit
normal vector field along M is
grad F
ND :
j grad F j
Here is an application.
Example 8.25 (Shape Operators of Spheres). The function F W RnC1 ! R

defined by F .x/ D jxj2 is a smooth defining n
P i function for each sphere S .R/.n The
gradient of this function is grad F D 2 i x @i , which has length 2R along S .R/.
The smooth vector field
nC1
1 X i
ND x @i
R
iD1
n
thus restricts to a unit normal along S .R/. (It is the outward pointing normal.) The
shape operator is now easy to compute:
nC1
1 X j 1
sX D X .@j x i /@i D X:
R R
i;j D1
Therefore s D .1=R/Id. The principal curvatures, therefore, are all equal to 1=R,
and it follows that the mean curvature is H D 1=R and the Gaussian curvature is
.1=R/n . //
For surfaces in R3 , either of the above methods can be used. When a parametriza-
tion X is given, the normal vector field is particularly easy to compute: because X1
and X2 span the tangent space to M at each point, their cross product is a nonzero
normal vector, so one choice of unit normal is
X1 X2
ND : (8.26)
jX1 X2 j
Problems 8-1, 8-2, and 8-3 will give you practice in carrying out these computations
for surfaces presented in various ways.
The Gaussian Curvature of a Surface Is Intrinsic
Because the Gaussian and mean curvatures are defined in terms of a particular
embedding of M into RnC1 , there is little reason to suspect that they have much to
do with the intrinsic Riemannian geometry of .M; g/. The next exercise illustrates
the fact that the mean curvature has no intrinsic meaning.
I Exercise 8.26. Let M1 R3 be the plane fz D 0g, and let M2 R3 be the cylinder
fx 2 C y 2 D 1g. Show that M1 and M2 are locally isometric, but the former has mean
curvature zero, while the latter has mean curvature ˙ 12 , depending on which normal is
chosen.
The amazing discovery made by Gauss was that the Gaussian curvature of a
surface in R3 is actually an intrinsic invariant of the Riemannian manifold .M; g/.
He was so impressed with this discovery that he called it Theorema Egregium, Latin
for “excellent theorem.”
Theorem 8.27 (Gauss’s Theorema Egregium). Suppose .M; g/ is an embedded

2-dimensional Riemannian submanifold of R3 . For every p 2 M , the Gaussian
curvature of M at p is equal to one-half the scalar curvature of g at p, and thus
the Gaussian curvature is a local isometry invariant of .M; g/.
Proof. Let p 2 M be arbitrary, and choose an orthonormal basis .b1 ; b2 / for Tp M .

In this basis g is represented by the identity matrix, and the shape operator has
the same matrix as the scalar second fundamental form. Thus K.p/ D det sji D

det hij , and the Gauss equation (8.21) reads

Rmp .b1 ; b2 ; b2 ; b1 / D h11 h22 h12 h21 D det hij D K.p/:
On the other hand, Corollary 7.27 shows that Rm D 14 Sg g, and thus by the defi-
nition of the Kulkarni–Nomizu product we have
Fig. 8.5: A plane section

Rmp .b1 ; b2 ; b2 ; b1 / D 14 S.p/ 2g11 g22 2g12 g21 D 12 S.p/: t
u
Motivated by the Theorema Egregium, for an abstract Riemannian 2-manifold

.M; g/, not necessarily embedded in R3 , we define the Gaussian curvature to be
K D 12 S , where S is the scalar curvature. If M is a Riemannian submanifold of R3 ,
then the Theorema Egregium shows that this new definition agrees with the original
definition of K as the determinant of the shape operator. The following result is a
restatement of Corollary 7.27 using this new definition.
Corollary 8.28. If .M; g/ is a Riemannian 2-manifold, the following relationships

hold:
Rm D 12 Kg g; Rc D Kg; S D 2K: t
u
Sectional Curvatures
Now, finally, we can give a quantitative geometric interpretation to the curvature

tensor in dimensions higher than 2. Suppose M is a Riemannian n-manifold (with
n 2), p is a point of M , and V Tp M is a star-shaped neighborhood of zero
on which expp is a diffeomorphism onto an open set U M . Let ˘ be any 2-
dimensional linear subspace of Tp M . Since ˘ \ V is an embedded 2-dimensional
submanifold of V , it follows that S˘ D expp .˘ \V / is an embedded 2-dimensional
submanifold of U M containing p (Fig. 8.5), called the plane section determined
by ˘ . Note that S˘ is just the set swept out by geodesics whose initial velocities lie
in ˘ , and Tp S˘ is exactly ˘ .
We define the sectional curvature of ˘ , denoted by sec.˘ /, to be the intrinsic
Gaussian curvature at p of the surface S˘ with the metric induced from the embed-
ding S˘ M . If .v; w/ is any basis for ˘ , we also use the notation sec.v; w/ for
sec.˘ /.
The next theorem shows how to compute the sectional curvatures in terms of the
curvature of .M; g/. To make the formula more concise, we introduce the following
notation. Given vectors v; w in an inner product space V , we set
Sectional Curvatures 251
p
jv ^ wj D jvj2 jwj2 hv; wi2 : (8.27)
It follows from the Cauchy–Schwarz inequality that jv ^ wj 0, with equality if
and only if v and w are linearly dependent, and jv ^ wj D 1 when v and w are
orthonormal. (One can define an inner product on the space ƒ2 .V / of contravariant
alternating 2-tensors, analogous to the inner product on forms defined in Problem
2-16, and this is the associated norm; see Problem 8-33(a).)
Proposition 8.29 (Formula for the Sectional Curvature). Let .M; g/ be a Rie-
mannian manifold and p 2 M . If v; w are linearly independent vectors in Tp M ,
then the sectional curvature of the plane spanned by v and w is given by
Rmp .v; w; w; v/
sec.v; w/ D : (8.28)
jv ^ wj2
Proof. Let ˘ Tp M be the subspace spanned by .v; w/. For this proof, we de-
note the induced metric on S˘ by gy, and its associated curvature tensor by Rm. c By
y
definition, sec.v; w/ is equal to K.p/, the Gaussian curvature of gy at p.
We show first that the second fundamental form of S˘ in M vanishes at p. To
see why, let z 2 ˘ be arbitrary, and let D z be the g-geodesic with initial velocity
z, whose image lies in S˘ for t sufficiently near 0. By the Gauss formula for vector
fields along curves,
0 D Dt 0 D D y t 0 C II. 0 ; 0 /:
Since the two terms on the right-hand side are orthogonal, each must vanish iden-
tically. Evaluating at t D 0 gives II.z; z/ D 0. Since z was an arbitrary element of
˘ D Tp .S˘ / and II is symmetric, polarization shows that II D 0 at p. (We cannot in
general expect II to vanish at other points of S˘ —it is only at p that all g-geodesics
starting tangent to S˘ remain in S˘ .) The Gauss equation then tells us that the
curvature tensors of M and S˘ are related at p by
c p .u; v; w; x/
Rmp .u; v; w; x/ D Rm
whenever u; v; w; x 2 ˘ .
Now choose an orthonormal basis .b1 ; b2 / for ˘ . Based on the observations
above, we see that the sectional curvature of ˘ is
y
K.p/ y
D 12 S.p/
2
X
D 1 c p b i ; bj ; bj ; bi
Rm
2
i;j D1
1c c p .b2 ; b1 ; b1 ; b2 /
D 2
Rmp .b1 ; b2 ; b2 ; b1 / C 12 Rm
c p .b1 ; b2 ; b2 ; b1 / D Rmp .b1 ; b2 ; b2 ; b1 / :
D Rm
To see how to compute this in terms of an arbitrary basis, let .v; w/ be any basis for
˘ . The Gram–Schmidt algorithm yields an orthonormal basis as follows:
v
b1 D I
jvj
w hw; b1 ib1
b2 D :
jw hw; b1 ib1 j
Then by the preceding computation,
y
K.p/ D Rmp .b1 ; b2 ; b2 ; b1 /

v w hw; b1 ib1 w hw; b1 ib1 v
D Rmp ; ; ;
jvj jw hw; b1 ib1 j jw hw; b1 ib1 j jvj (8.29)
Rmp .v; w; w; v/
D 2 ;
jvj jw hw; b1 ib1 j2
where we have used the fact that Rm.v; b1 ; ; / D Rm.; ; b1 ; v/ D 0 because b1 is

a multiple of v. To simplify the denominator of this last expression, we substitute
b1 D v=jvj to obtain

2 2 2 2 hw; vi2 hw; vi2
jvj jw hw; b1 ib1 j D jvj jwj 2 C
jvj2 jvj2
D jvj2 jwj2 hv; wi2 D jv ^ wj2 :
Inserting this into (8.29) proves the theorem. t

u
I Exercise 8.30. Suppose .M; g/ is a Riemannian manifold and gz D g for some pos-
itive constant . Use Theorem 7.30 to prove that for every p 2 M and plane ˘ Tp M ,
z and g are related by sf
the sectional curvatures of ˘ with respect to g ec.˘ / D 1 sec.˘ /.
Proposition 8.29 shows that one important piece of quantitative information pro-
vided by the curvature tensor is that it encodes the sectional curvatures of all plane
sections. It turns out, in fact, that this is all of the information contained in the curva-
ture tensor: as the following proposition shows, the sectional curvatures completely
determine the curvature tensor.
Proposition 8.31. Suppose R1 and R2 are algebraic curvature tensors on a finite-

dimensional inner product space V . If for every pair of linearly independent vectors
v; w 2 V ,
R1 .v; w; w; v/ R2 .v; w; w; v/
D ;
jv ^ wj2 jv ^ wj2
then R1 D R2 .
Proof. Let R1 and R2 be tensors satisfying the hypotheses, and set D D R1 R2 .

Then D is an algebraic curvature tensor, and D.v; w; w; v/ D 0 for all v; w 2 V .
(This is true by hypothesis when v and w are linearly independent, and it is true by
the symmetries of D when they are not.) We need to show that D D 0.
For all vectors v; w; x, the symmetries of D give
Sectional Curvatures 253
0 D D.v C w; x; x; v C w/
D D.v; x; x; v/ C D.v; x; x; w/ C D.w; x; x; v/ C D.w; x; x; w/
D 2D.v; x; x; w/:
From this it follows that

0 D D.v; x C u; x C u; w/
D D.v; x; x; w/ C D.v; x; u; w/ C D.v; u; x; w/ C D.v; u; u; w/
D D.v; x; u; w/ C D.v; u; x; w/:
Therefore D is antisymmetric in every adjacent pair of arguments. Now the alge-

braic Bianchi identity yields
0 D D.v; w; x; u/ C D.w; x; v; u/ C D.x; v; w; u/

D D.v; w; x; u/ D.w; v; x; u/ D.v; x; w; u/
D 3D.v; w; x; u/: t
u
We can also give a geometric interpretation of the Ricci and scalar curvatures on
a Riemannian manifold. Since the Ricci tensor is symmetric and bilinear, Lemma
8.11 shows that it is completely determined by its values of the form Rc.v; v/ for
unit vectors v.
Proposition 8.32 (Geometric Interpretation of Ricci and Scalar Curvatures).

Let .M; g/ be a Riemannian n-manifold and p 2 M .
(a) For every unit vector v 2 Tp M , Rcp .v; v/ is the sum of the sectional curva-
tures of the 2-planes spanned by .v; b2 /; : : : ; .v; bn /, where .b1 ; : : : ; bn / is any
orthonormal basis for Tp M with b1 D v.
(b) The scalar curvature at p is the sum of all sectional curvatures of the 2-planes
spanned by ordered pairs of distinct basis vectors in any orthonormal basis.
Proof. Given any unit vector v 2 Tp M , let .b1 ; : : : ; bn / be as in the hypothesis. Then
Rcp .v; v/ is given by
n
X n
X
Rcp .v; v/ D R11 .p/ D Rk11 k .p/ D Rmp .bk ; b1 ; b1 ; bk / D sec.b1 ; bk /:
kD1 kD2
For the scalar curvature, we let .b1 ; : : : ; bn / be any orthonormal basis for Tp M , and
compute
n
X n
X

S.p/ D Rj j .p/ D Rcp bj ; bj D Rmp bk ; bj ; bj ; bk
j D1 j;kD1
X
D sec bj ; bk : t
u
j ¤k
One consequence of this proposition is that if .M; g/ is a Riemannian manifold

in which all sectional curvatures are positive, then the Ricci and scalar curvatures
are both positive as well. The analogous statement holds if “positive” is replaced by
“negative,” “nonpositive,” or “nonnegative.”
If the opposite sign convention is chosen for the curvature tensor, then the right-
hand side of formula (8.28) has to be adjusted accordingly, with Rmp .v; w; v; w/
taking the place of Rmp .v; w; w; v/. This is so that whatever sign convention is
chosen for the curvature tensor, the notion of positive or negative sectional, Ricci,
or scalar curvature has the same meaning for everyone.
Sectional Curvatures of the Model Spaces
We can now compute the sectional curvatures of our three families of frame-
homogeneous model spaces. A Riemannian metric or Riemannian manifold is said
to have constant sectional curvature if the sectional curvatures are the same for all
planes at all points.
Lemma 8.33. If a Riemannian manifold .M; g/ is frame-homogeneous, then it has

constant sectional curvature.
Proof. Frame homogeneity implies, in particular, that given two 2-planes at the
same or different points, there is an isometry taking one to the other. The result
follows from the isometry invariance of the curvature tensor. t
u
Thus to compute the sectional curvature of one of our model spaces, it suffices
to compute the sectional curvature for one plane at one point in each space.
Theorem 8.34 (Sectional Curvatures of the Model Spaces). The following Rie-
mannian manifolds have the indicated constant sectional curvatures:

(a) Rn ; gx has constant sectional curvature 0.
n
(b) S .R/; gV R has constant sectional curvature 1=R2 .
(c) Hn .R/; gM R has constant sectional curvature 1=R2 .
Proof. First we consider the simplest case: Euclidean space. Since the curvature
tensor of Rn is identically zero, clearly all sectional curvatures are zero. This is also
easy to see geometrically, since each plane section is actually a plane, which has
zero Gaussian curvature.
Next consider the sphere Sn .R/. We need only compute the sectional curvature
of the plane ˘ spanned by .@1 ; @2 / at the point .0; : : : ; 0; 1/. The geodesics with
initial velocities in ˘ are great circles in the .x 1 ; x 2 ; x nC1 / subspace. Therefore
S˘ is isometric to the round 2-sphere of radius R embedded in R3 . As Example
8.25 showed, S2 .R/ has Gaussian curvature 1=R2 . Therefore Sn .R/ has constant
sectional curvature equal to 1=R2 .
Finally, the proof for hyperbolic spaces is left to Problem 8-28. t
u
Problems 255
Note that for every real number c, exactly one of the model spaces listed above
has constant sectional curvature c. These spaces will play vital roles in our compar-
ison and local-to-global theorems in the last two chapters of the book.
I Exercise 8.35. Show that the metric on real projective space RP n defined in Example
2.34 has constant positive sectional curvature.
Since the sectional curvatures determine the curvature tensor, one would expect
to have an explicit formula for the full curvature tensor when the sectional curvature
is constant. Such a formula is provided in the following proposition.
Proposition 8.36. A Riemannian metric g has constant sectional curvature c if and
only if its curvature tensor satisfies
Rm D 12 cg g:
In this case, the Ricci tensor and scalar curvature of g are given by the formulas
Rc D .n 1/cgI S D n.n 1/c;
and the Riemann curvature endomorphism is

R.v; w/x D c hw; xiv hv; xiw :
In terms of any basis,
Rij kl D c.gil gj k gik gj l /I Rij D .n 1/cgij :

u
The basic concepts of Riemannian metrics and sectional curvatures in arbitrary
dimensions were introduced by Bernhard Riemann in a famous 1854 lecture at
Göttingen University [Spi97, Vol. II, Chap. 4]. These were just a few of the seminal
accomplishments in his tragically short life.
Problems
8-1. Suppose U is an open set in Rn and f W U ! R is a smooth function. Let
M D f.x; f .x// W x 2 U g RnC1 be the graph of f , endowed with the
induced Riemannian metric and upward unit normal (see Example 2.19).
(a) Compute the components of the shape operator in graph coordinates, in
terms of f and its partial derivatives.
(b) Let M RnC1 be the n-dimensional paraboloid defined as the graph
of f .x/ D jxj2 . Compute the principal curvatures of M .
(Used on p. 361.)
8-2. Let .M; g/ be an embedded Riemannian hypersurface in a Riemannian
manifold M ; gz , let F be a local defining function for M , and let N D
grad F=j grad F j.
(a) Show that the scalar second fundamental form of M with respect to the
unit normal N is given by
z 2 F .X; Y /
r
h.X; Y / D
j grad F j
for all X; Y 2 X.M /.

(b) Show that the mean curvature of M is given by

1 grad F
H D divgz ;
n j grad F j
where n D dim M and divgz is the divergence operator of gz (see Prob-

lem 5-14). [Hint: First prove the following linear-algebra lemma: If V is
a finite-dimensional inner product space, w 2 V is a unit vector,
and
A W V ! V is a linear map that takes w ? to w ? , then tr Ajw ? D trB,
where B W V ! V is defined by Bx D Ax hx; wiAw.]
8-3. Let C be an embedded smooth curve in the half-plane H D f.r; z/ W r > 0g,
and let SC R3 be the surface of revolution determined by C as in Example
2.20. Let .t / D .a.t /; b.t // be a unit-speed local parametrization of C , and
let X be the local parametrization of SC given by (2.11).
(a) Compute the shape operator and principal curvatures of SC in terms of
a and b, and show that the principal directions at each point are tangent
to the meridians and latitude circles.
(b) Show that the Gaussian curvature of SC at a point X.t; / is equal to
a00 .t /=a.t /.
8-4. Show that there is a surface of revolution in R3 that has constant Gaussian
curvature equal to 1 but does not have constant principal curvatures. [Remark:
We will see in Chapter 10 that this surface is locally isometric to S2 (see Cor.
10.15), so this gives an example of two nonflat hypersurfaces in R3 that are
locally isometric but have different principal curvatures.]
8-5. Let S R3 be the paraboloid given by z D x 2 C y 2 , with the induced metric.
Prove that S is isotropic at only one point. (Used on p. 56.)
8-6. Suppose .M; g/ is a Riemannian manifold, and W I ! M is a regular (but
not necessarily unit-speed) curve in M . Show that the geodesic curvature of
at t 2 I is
j 0 .t / ^ Dt 0 .t /j
.t / D ;
j 0 .t /j3
where the norm in the numerator is the one defined by (8.27). Show also that
in R3 with the Euclidean metric, the formula can be written
j 0 .t / 00 .t /j
.t / D :
j 0 .t /j3
Problems 257
8-7. For w > 0, let Mw R3 be the surface of revolution obtained by revolving

the curve .t / D .w cosh.t =w/; t / around the z-axis, called a catenoid. Show
that Mw is a minimal surface for each w.
8-8. For h > 0, let Ch R3 be the one-dimensional submanifold f.x; y; z/ W x 2 C
y 2 D 1; z D ˙hg; it is the union of two unit circles lying in the z D ˙h planes.
Let h0 D 1= sinh c, where c is the unique positive solution to the equation
c tanh c D 1. Prove that if h < h0 , then there are two distinct positive values
w1 < w2 such that the portions of the catenoids Mw1 and Mw2 lying in the
region jzj h are minimal surfaces with boundary Ch (using the notation of
Problem 8-7), while if h D h0 , there is exactly one such value, and if h > h0 ,
there are none. [Remark: This phenomenon can be observed experimentally.
If you dip two parallel wire circles into soapy water, as long as they are close
together they will form an area-minimizing soap film in the shape of a portion
of a catenoid (the one with the larger “waist,” Mw2 in the notation above). As
you pull the circles apart, the “waist” of the catenoid gets smaller, until the
ratio of the distance between the circles to their diameter reaches h0 0:6627,
at which point the film will burst and the only soap film that can be formed is
two disjoint flat disks.]
8-9. Let M RnC1 be a Riemannian hypersurface, and let N be a smooth unit
normal vector field along M . At each point p 2 M , Np 2 Tp RnC1 can be
thought of as a unit vector in RnC1 and therefore as a point in Sn . Thus each
choice of unit normal vector field defines a smooth map
W M ! Sn , called
the Gauss map of M . Show that
d VgV D .1/n K d Vg , where K is the
Gaussian curvature of M .
8-10. Suppose g D g1 ˚ g2 is a product metric on M1 M2 as in (2.12). (See also
Problem 7-6.)
(a) Show that for each point pi 2 Mi , the submanifolds M1 fp2 g and
fp1 g M2 are totally geodesic.
(b) Show that sec.˘ / D 0 for every 2-plane ˘ T.p1 ;p2 / .M1 M2 / spanned
by v1 2 Tp1 M1 and v2 2 Tp2 M2 .
(c) Show that if M1 and M2 both have nonnegative sectional curvature, then
M1 M2 does too; and if M1 and M2 both have nonpositive sectional
curvature, then M1 M2 does too.
(d) Now suppose .Mi ; gi / has constant sectional curvature ci for i D 1; 2;
let ni D dim Mi . Show that .M; g/ is Einstein if and only if c1 .n1 1/ D
c2 .n2 1/; and .M; g/ has constant sectional curvature if and only if
n1 D n2 D 1 or c1 D c2 D 0.
(Used on p. 366.)
8-11. Suppose M R3 is a smooth surface, p 2 M , and Np is a normal vector to
M at p. Show that if ˘ R3 is any affine plane containing p and parallel
to Np , then there is a neighborhood U of p in R3 such that M \ ˘ \ U
is a smooth embedded curve in ˘ , and the principal curvatures of M at
p are equal to the minimum and maximum signed Euclidean curvatures of
these curves as ˘ varies. [Remark: This justifies the informal recipe for
computing principal
curvatures given in Chapter 1.]
8-12. Suppose W M ; gz ! .M; g/ is a Riemannian submersion, and M ; gz has
all sectional curvatures bounded below by a constant c. Use O’Neill’s for-
mula (Problem 7-14) to show that the sectional curvatures of .M; g/ are
bounded below by the same constant.
8-13. Let p W S2nC1 ! CP n be the Riemannian submersion described in Exam-
ple 2.30. In this problem, we identify C nC1 with R2nC2 by means of coor-
dinates .x 1 ; y 1 ; : : : ; x nC1 ; y nC1 / defined by z j D x j C iy j .
(a) Show that the vector field
@ @
S D xj j
yj j
@y @x
on C nC1 is tangent to S2nC1 and spans the vertical space Vz at each

point z 2 S2nC1 . (The implicit summation here is from 1 to n C 1.)
(b) Show that for all horizontal vector fields W; Z on S2nC1 ,
ŒW; ZV D d!.W; Z/S D 2hW; J ZiS;
where ! is the 1-form on C nC1 given by

X
! D S[ D x j dy j y j dx j ;
j
and J W T C nC1 ! T C nC1 is the real-linear orthogonal map given by

@ @ @ @
J aj j C b j j D aj j b j j :
@x @y @y @x
p
(This is just multiplication by i D 1 in complex coordinates. Notice
that J ı J D Id.)
(c) Using O’Neill’s formula (Problem 7-14), show that the curvature tensor
of CP n satisfies
Rm.w; x; y; z/ D hw;
z zzihz z hw;
x ; yi z yihz
z x ; zzi
2hw;
z J xzihy;
z J zzi hw;
z J yihz
z x ; J zzi
C hw;
z J zzihz z
x ; J yi;
for every q 2 CP n and w; x; y; z 2 Tq CP n , where w; z xz; y;

z zz are hor-
izontal lifts of w; x; y; z to an arbitrary point qz 2 p 1 .q/ S2nC1 .
(d) Using the notation of part (c), show that for orthonormal vectors w; x 2
Tq CP n , the sectional curvature of the plane spanned by fw; xg is
˝ ˛2
sec.w; x/ D 1 C 3 w;
z J xz :
Problems 259
(e) Show that for n 2, the sectional curvatures at each point of CP n take
on all values between 1 and 4, inclusive, and conclude that CP n is not
frame-homogeneous.
(f) Compute the Gaussian curvature of CP 1 .
(Used on pp. 56, 262, 367.)
8-14. Show that a Riemannian 3-manifold is Einstein if and only if it has constant
sectional curvature.
8-15. Suppose .M; g/ is a 3-dimensional Riemannian manifold that is homoge-
neous and isotropic. Show that g has constant sectional curvature. Show that
the analogous result in dimension 4 is not true. [Hint: See Problem 8-13.]
8-16. For each a > 0, let ga be the Berger metric on SU.2/ defined in Problem 3-10.
Compute the sectional curvatures with respect to ga of the planes spanned by
.X; Y /, .Y; Z/, and .Z; X /. Prove that if a ¤ 1, then .SU.2/; ga / is homoge-
neous but not isotropic anywhere. (Used on p. 56.)
8-17. Let G be a Lie group with a bi-invariant metric g (see Problem 7-13).
(a) Suppose X and Y are orthonormal elements of Lie.G/. Show that
ˇ ˇ2
sec.Xp ; Yp / D 14 ˇŒX; Y ˇ for each p 2 G, and conclude that the sec-
tional curvatures of .G; g/ are all nonnegative.
(b) Show that every Lie subgroup of G is totally geodesic in G.
(c) Now suppose G is connected. Show that G is flat if and only if it is
abelian.
(Used on p. 282.)
8-18. Suppose
.M; g/ is a Riemannian hypersurface in a Riemannian manifold
M ; gz , and N is a unit normal vector field along M . We say that M is
convex (with respect to N ) if its scalar second fundamental form satisfies
h.v; v/ 0 for all v 2 TM . Show that if M is convex and M has sectional
curvatures bounded below by a constant c, then all sectional curvatures of M
are bounded below by c.
8-19. Suppose .M; g/ is a Riemannian
hypersurface in an .n C 1/-dimensional
Lorentz manifold M ; gz , and N is a smooth unit normal vector field along
M . Define the scalar second fundamental form h and the shape operator s
by requiring that II.X; Y / D h.X; Y /N and hsX; Y i D h.X; Y / for all X; Y 2
X.M /. Prove the following Lorentz analogues of the formulas of Theorem
8.13 (with notation as in that theorem):
(a) G AUSS F ORMULA : r z X Y D rX Y C h.X; Y /N .
(b) G AUSS F ORMULA FOR A C URVE : D z t X D Dt X C h. 0 ; X /N .
(c) W EINGARTEN E QUATION : r z X N D sX .

(d) f
G AUSS E QUATION : Rm.W; X; Y; Z/ D Rm C 12 h h .W; X; Y; Z/.
(e) f
C ODAZZI E QUATION : Rm.W; X; Y; N / D .Dh/.Y; W; X /.

8-20. Suppose M ; gz is an .n C 1/-dimensional Lorentz manifold, and assume
that gz satisfies the Einstein field equation with a cosmological constant:
e 1 Szgz C z
Rc g D T , where is a constant and T is a smooth symmetric
2
,
2-tensor field (see p. 211). Let .M; g/ be a Riemannian hypersurface in M
and let h be its scalar second fundamental form as defined in Problem 8-19.
Use the results of Problem 8-19 to show that g and h satisfy the following
Einstein constraint equations on M :
2
S 2 jhj2g C trg h D 2;

div h d trg h D J;
where S is the scalar curvature of g, div h the divergence of h with respect

to g as defined in Problem 5-16, is the function D T .N; N /, and J is
the 1-form J.X / D T .N; X /. [Remark: When J and are both zero, these
equations, known as the vacuum Einstein constraint equations, are neces-
sary conditions for g and h to be the metric and second fundamental form
of a Riemannian hypersurface in an .n C 1/-dimensional Lorentz manifold
satisfying the vacuum Einstein field equation with cosmological constant .
It was proved in 1950 by Yvonne Choquet-Bruhat that these conditions are
also sufficient; for a proof, see [CB09, pp. 166–168].]
8-21. For every linear endomorphism A W Rn ! Rn , the associated quadratic form
is the function Q W Rn ! R defined by Q.x/ D hAx; xi. Prove that
Z
1
Q d VgV D .tr A/Vol Sn1 :
Sn1 n
R
[Hint: Show that Sn1 x i x j d VgV D 0 when i ¤ j by examining the effect
of the isometry
.x 1 ; : : : ; x i ; : : : ; x j ; : : : ; x n / 7! .x 1 ; : : : ; x j ; : : : ; x i ; : : : ; x n /;
R 2 P i 2
and compute Sn1 x i d VgV using the fact that i x D 1 on the
sphere.] [Remark: It is a standard fact of linear algebra that the trace of A
is independent of the choice of basis. This gives a geometric interpretation to
the trace as n times the average value of the associated quadratic form Q.]
(Used on p. 313.)
8-22. Let .M; g/ be a Riemannian n-manifold and p 2 M . Proposition 8.32 gave
a geometric interpretation of the Ricci curvature at p based on a choice of
orthonormal basis. This problem describes an interpretation that does not
refer to a basis. For each unit vector v 2 Tp M , prove that
Z
n1
Rcp .v; v/ D sec.v; w/ d Vgy ;
Vol Sn2 w2Sv?
where Sv? denotes the set of unit vectors in Tp M that are orthogonal to v and
gy denotes the Riemannian metric on Sv? induced from the flat metric gp on
Tp M . [Hint: Use Problem 8-21.]
Problems 261
8-23. Suppose .M; g/ is a connected n-dimensional Riemannian manifold, and a

Lie group G acts isometrically and effectively on M . Problem 5-12 showed
that dim G n.n C 1/=2. Prove that if equality holds, then g has constant
sectional curvature.
8-24. Let .M; g/ be a connected Riemannian manifold. Recall the definition of
k-point homogeneous from Problem 6-18. Prove the following:
(a) If .M; g/ is homogeneous, then it has constant scalar curvature.
(b) If .M; g/ is 2-point homogeneous, then it is Einstein.
(c) If .M; g/ is 3-point homogeneous, then it has constant sectional curva-
ture.
8-25. For i D 1; 2, suppose .Mi ; gi / is a Riemannian manifold of dimension ni
2 with constant sectional curvature ci ; and let g D g1 ˚ g2 be the product
metric on M D M1 M2 . Show that the Weyl tensor of g is given by
c1 C c2
W D n2 .n2 1/h1 h1
2.n 1/.n 2/

2.n1 1/.n2 1/h1 h2 C n1 .n1 1/h2 h2 ;
where n D n1 C n2 and hi D i gi for i D 1; 2. Conclude that .M; g/ is

locally conformally flat if and only if c2 D c1 . (See also Problem 7-6.)
(Used on p. 67.)
8-26. Let .M; g/ be a 4-dimensional Riemannian manifold. Given p 2 M and an
orthonormal basis .bi / for Tp M , prove that the Weyl tensor at p satisfies
1 1
W1221 D .k12 C k34 / .k13 C k14 C k23 C k24 /;
3 6

where kij is the sectional curvature of the plane spanned by bi ; bj .
8-27. Prove that the Fubini–Study metric on CP 2 is not locally conformally flat.
[Hint: Use Problems 8-13 and 8-26.]
8-28. Complete the proof of Theorem 8.34 by showing in two ways that the hy-
perbolic space of radius R has constant sectional curvature equal to 1=R2 .
(a) In the hyperboloid model, compute the second fundamental form of
Hn .R/ Rn;1 at the point .0; : : : ; 0; R/, and use either the general form
of the Gauss equation (Thm. 8.5) or the formulas of Problem 8-19.
(b) In the Poincaré ball model, use formula (7.44) for the conformal trans-
formation of the curvature to compute the Riemann curvature tensor at
the origin.
8-29. Prove Proposition 8.36 (curvature tensors on constant-curvature spaces).
8-30. Suppose M RnC1 is a hypersurface with the induced Riemannian metric.
Show that the Ricci tensor of M satisfies
˝ ˛
Rc.v; w/ D nH sv s 2 v; w ;
where H and s are the mean curvature and shape operator of M , respec-
tively, and s 2 v D s.sv/.
8-31. For 1 < k n, show that any k points in the hyperbolic space Hn .R/ lie
in a totally geodesic .k 1/-dimensional submanifold, which is isometric to
Hk1 .R/.
8-32. Suppose .M; g/ is a Riemannian manifold and G is a Lie group acting
smoothly and isometrically on M . Let S M be the fixed point set of G,
that is, the set of points p 2 M such that '.p/ D p for all ' 2 G. Show
that each connected component of S is a smoothly embedded totally geodesic
submanifold of M . (The reason for restricting to a single connected comp-
onent is that different components may have different dimensions.)
8-33. Suppose .M; g/ is a Riemannian manifold. Let ƒ2 .TM / denote the bundle
of 2-vectors (alternating contravariant 2-tensors) on M .
(a) Show that there is a unique fiber metric on ƒ2 .TM / whose associated
norm satisfies
jw ^ xj2 D jwj2 jxj2 hw; xi2
for all tangent vectors w; x at every point q 2 M .
(b) Show that there is a unique bundle endomorphism R W ƒ2 .TM / !
ƒ2 .TM /, called the curvature operator of g, that satisfies
˝ ˛
R.w ^ x/; y ^ z D Rm.w; x; y; z/ (8.30)
for all tangent vectors w; x; y; z at a point of M , where the inner prod-

uct on the left-hand side is the one described in part (a).
(c) A Riemannian metric is said to have positive curvature operator if R is
positive definite, and negative curvature operator if R is negative def-
inite. Show that positive curvature operator implies positive sectional
curvature, and negative curvature operator implies negative sectional
curvature. [Remark: This is the reason for the negative sign in (8.30).
If the Riemann curvature tensor is defined as the negative of ours, the
negative sign should be omitted.]
(d) Show that the converse need not be true, by using the results of Problem
8-13 to compute the following expression on CP 2 with the Fubini–
Study metric: ˝ ˛
R.w ^ x C y ^ z/; w ^ x C y ^ z ;
where w; x; y; z are orthonormal vectors at an arbitrary point of CP 2 ,
chosen so that their horizontal lifts satisfy yz D J w
z and zz D J xz.
Chapter 9
The Gauss–Bonnet Theorem
All the work we have done so far has been focused on purely local properties of
Riemannian manifolds and their submanifolds. We are finally in a position to prove
our first major local-to-global theorem in Riemannian geometry: the Gauss–Bonnet
theorem. The grandfather of all such theorems in Riemannian geometry, it is a local-
to-global theorem par excellence, because it asserts the equality of two very differ-
ently defined quantities on a compact Riemannian 2-manifold: the integral of the
Gaussian curvature, which is determined by the local geometry, and 2 times the
Euler characteristic, which is a global topological invariant. Although in this form
it applies only in two dimensions, it has provided a model and an inspiration for
innumerable local-to-global results in higher-dimensional geometry, some of which
we will prove in Chapter 12.
This chapter begins with some not-so-elementary notions from plane geome-
try, leading up to a proof of Hopf’s rotation index theorem, which expresses the
intuitive idea that the velocity vector of a simple closed plane curve, or more gen-
erally of a “curved polygon,” makes a net rotation through an angle of exactly 2
as one traverses the curve counterclockwise. Then we investigate curved polygons
on Riemannian 2-manifolds, leading to a far-reaching generalization of the rota-
tion index theorem called the Gauss–Bonnet formula, which relates the exterior an-
gles, geodesic curvature of the boundary, and Gaussian curvature in the interior of
a curved polygon. Finally, we use the Gauss–Bonnet formula to prove the global
statement of the Gauss–Bonnet theorem.
Some Plane Geometry

Look back for a moment at the three local-to-global theorems about plane geometry
stated in Chapter 1: the angle-sum theorem, the circumference theorem, and the total
curvature theorem. When looked at correctly, these three theorems all turn out to be
manifestations of the same phenomenon: as one traverses a simple closed plane

264 9 The Gauss–Bonnet Theorem
Fig. 9.1: A closed curve with 0 .a/ D 0 .b/
curve, the velocity vector makes a net rotation through an angle of exactly 2. Our
task in the first part of this chapter is to make these notions precise.
Throughout this section, W Œa; b ! R2 is an admissible curve in the plane. We
say that is a simple closed curve if .a/ D .b/ but is injective on Œa; b/. We do not
assume that has unit speed; instead, we define the unit tangent vector field of as
the vector field T along each smooth segment of given by
0 .t /
T .t / D :
j 0 .t /j
Since each tangent space to R2 is naturally identified with R2 itself, we can think
of T as a map into R2 , and since T has unit length, it takes its values in S1 .
If is smooth (or at least continuously differentiable), we define a tangent an-
gle function for to be a continuous function W Œa; b ! R such that T .t / D
.cos .t /; sin .t // for all t 2 Œa; b. It follows from the theory of covering spaces
that such a function exists: the map q W R ! S1 given by q.s/ D .cos s; sin s/ is
a smooth covering map, and the path-lifting property of covering maps (Prop.
A.54(b)) ensures that there is a continuous function W Œa; b ! R that satisfies
q..t // D T .t /. By the unique lifting property (Prop. A.54(a)), a lift is uniquely
determined once its value at any single point is determined, and thus any two lifts
differ by a constant integral multiple of 2.
If is a continuously differentiable simple closed curve such that 0 .a/ D 0 .b/
(Fig. 9.1), then .cos .a/; sin .a// D .cos .b/; sin .b//, so .b/ .a/ must be an
integral multiple of 2. For such a curve, we define the rotation index of to be
the following integer:
1
. / D .b/ .a/ ;
2
where is any tangent angle function for . For any other choice of tangent angle
function, .a/ and .b/ would change by addition of the same constant, so the
rotation index is independent of the choice of .
We would also like to extend the definition of the rotation index to certain piece-
wise regular closed curves. For this purpose, we have to take into account the
“jumps” in the tangent angle function at corners. To do so, suppose W Œa; b ! R2
Some Plane Geometry 265
Fig. 9.2: An exterior angle Fig. 9.3: A cusp vertex
is an admissible simple closed curve, and let .a0 ; : : : ; ak / be an admissible parti-

tion of Œa; b. The points .ai / are called the vertices of , and the curve segments
jŒai 1 ;ai are called its edges or sides.
At each vertex .ai /, recall that has left-hand and right-hand velocity vec-
tors denoted by 0 .ai / and 0 .aiC /, respectively; let T .ai / and T .aiC / denote the
corresponding unit vectors. We classify each vertex into one of three categories:
If T .ai / ¤ ˙T .aiC /, then .ai / is an ordinary vertex.
If T .ai / D T .aiC /, then .ai / is a flat vertex.
If T .ai / D T .aiC /, then .ai / is a cusp vertex.
At each ordinary vertex, define the exterior angle at .ai / to be the oriented mea-
sure "i of the angle from T .ai / to T .aiC /, chosen to be in the interval .; /, with
a positive sign if T .ai /; T .aiC / is an oriented basis for R2 , and a negative sign
otherwise (Fig. 9.2). At a flat vertex, the exterior angle is defined to be zero. At a
cusp vertex, there is no simple way to choose unambiguously between and
(Fig. 9.3), so we leave the exterior angle undefined. The vertex .a/ D .b/ is han-
dled in the same way, with T .b/ and T .a/ playing the roles of T .ai / and T .aiC /,
respectively. If .ai / is an ordinary or a flat vertex, the interior angle at .ai / is
defined to be i D "i ; our conventions ensure that 0 < i < 2.
The curves we wish to consider are of the following type: a curved polygon in
the plane is an admissible simple closed curve without cusp vertices, whose image
is the boundary of a precompact open set ˝ R2 . The set ˝ is called the interior
of (not to be confused with the topological interior of its image as a subset of R2 ,
which is the empty set).
Suppose W Œa; b ! R2 is a curved polygon. If is parametrized so that at
smooth points, 0 is positively oriented with respect to the induced orientation on
@˝ in the sense of Stokes’s theorem, we say that is positively oriented (Fig. 9.4).
Intuitively, this means that is parametrized in the counterclockwise direction, or
that ˝ is always to the left of .
We define a tangent angle function for a curved polygon to be a piecewise
continuous function W Œa; b ! R that satisfies T .t / D .cos .t /; sin .t // at each
point t where is smooth, that is continuous from the right at each ai with
Fig. 9.4: A positively oriented curved polygon
Fig. 9.5: Tangent angle at a vertex Fig. 9.6: Tangent angle function
.ai / D lim .t / C "i ; (9.1)

t%ai
and that satisfies

.b/ D lim .t / C "k ; (9.2)
t%b
where "k is the exterior angle at .b/. (See Figs. 9.5 and 9.6.) Such a function always
exists: start by defining .t / for t 2 Œa; a1 / to be any lift of T on that interval; then on
Œa1 ; a2 / define .t / to be the unique lift that satisfies (9.1), and continue by induc-
tion, ending with .b/ defined by (9.2). Once again, the difference between any two
such functions is aconstant integral
multiple of 2. We define the rotation index of
1
to be . / D 2 .b/ .a/ just as in the smooth case. As before, . / is an in-
teger, because the definition ensures that .cos .b/; sin .b// D .cos .a/; sin .a//.
The following theorem was first proved by Heinz Hopf in 1935. (For a readable
version of Hopf’s proof, see [Hop89, p. 42].) It is frequently referred to by the
German name given to it by Hopf, the Umlaufsatz.
Theorem 9.1 (Rotation Index Theorem). The rotation index of a positively ori-
ented curved polygon in the plane is C1.
Fig. 9.7: The curve after changing the parameter interval and translating .a/ to the origin
Proof. Let W Œa; b ! R2 be such a curved polygon. Assume first that all the ver-
tices of are flat. This means, in particular, that 0 is continuous and 0 .a/ D 0 .b/.
Since .a/ D .b/, we can extend to a continuous map from R to R2 by requiring
it to be periodic of period b a, and our hypothesis 0 .a/ D 0 .b/ guarantees that
the extended map still has continuous first derivatives. Define T .t / D 0 .t /=j 0 .t /j
as before.
Let W R ! R be any lift of T W R ! S1 . Then jŒa;b is a tangent angle function
for , and thus .b/ D .a/ C 2. /. If we set z.t / D .t C b a/ 2. /, then

cos z.t /; sin z.t / D cos .t C b a/; sin .t C b a/ D T .t C b a/ D T .t /;
so z is also a lift of T . Because .a/

z D .a/, it follows that z , or in other words
the following equation holds for all t 2 R:
.t C b a/ D .t / C 2. /: (9.3)
If a1 is an arbitrary point in Œa; b and b1 D a1 C b a, then jŒa1 ;b1 is also a

positively oriented curved polygon with only flat vertices, and jŒa1 ;b1 is a tangent
angle function for it. Note that (9.3) implies
.b1 / .a1 / D .a1 C b a/ .a1 / D ..a1 / C 2. // .a1 / D 2. /;
so jŒa1 ;b1 has the same rotation index as jŒa;b . Thus we obtain the same result by
restricting to any closed interval of length b a.
Using this freedom, we can assume that the parameter interval Œa; b has been
chosen so that the y-coordinate of achieves its minimum at t D a. Moreover, by a
translation in the xy-plane (which does not change 0 or ), we may as well assume
that .a/ is the origin. With these adjustments, the image of remains in the closed
upper half-plane, and T .a/ D T .b/ D .1; 0/ (Fig. 9.7). By adding a constant integral
multiple of 2 to if necessary, we can also assume that .a/ D 0.
Next, we define a continuous secant angle function, denoted by '.t1 ; t2 /, repre-
senting the angle between the positive x-direction and the vector from .t1 / to .t2 /.
To be precise, let R2 be the triangular region D f.t1 ; t2 / W a t1 t2 bg
Fig. 9.8: The domain of ' Fig. 9.9: The secant angle function
(Fig. 9.8), and define a map V W ! S1 by

„ .t / .t /
2 1
; t1 < t2 and .t1 ; t2 / ¤ .a; b/I
j.t2 / .t1 /j
V .t1 ; t2 / D
T .t1 /; t1 D t2 I
T .b/; .t1 ; t2 / D .a; b/:
The function V is continuous at points where t1 < t2 and .t1 ; t2 / ¤ .a; b/, because
is continuous and injective there. To see that it is continuous elsewhere, note that
for t1 < t2 , the fundamental theorem of calculus gives
Z 1 Z 1
d
.t2 / .t1 / D t1 C s.t2 t1 / ds D 0 t1 C s.t2 t1 / .t2 t1 / ds;
0 ds 0
and thus
ˇ ˇ Z 1
ˇ .t2 / .t1 / ˇ ˇ 0 ˇ
ˇ .t /ˇˇ
0 ˇ t1 C s.t2 t1 / 0 .t /ˇ ds:
ˇ t t
2 1 0
Because 0 is uniformly continuous on the compact set Œa; b, this last expression
can be made as small as desired by taking .t1 ; t2 / close to .t; t /. It follows that
.t2 / .t1 /
lim D 0 .t /;
.t1 ;t2 /!.t;t/ t2 t1
t1 <t2
and therefore
.t2 / .t1 /
lim V .t1 ; t2 / D lim
.t1 ;t2 /!.t;t/ .t1 ;t2 /!.t;t/ j.t2 / .t1 /j
t1 <t2 t1 <t2
,ˇ ˇ
.t2 / .t1 / ˇ .t2 / .t1 / ˇ
D lim ˇ ˇ
.t1 ;t2 /!.t;t/ t2 t1 ˇ t t ˇ
2 1
t1 <t2
0
.t /
D D T .t / D V .t; t /:
j 0 .t /j
Similarly, because T is continuous,
lim V .t1 ; t2 / D lim T .t1 / D T .t / D V .t; t /:

.t1 ;t2 /!.t;t/ t1 !t
t1 Dt2
It follows that V is continuous at .t; t /.

To prove that V is continuous at .a; b/, recall that we have extended to be
periodic of period b a. The argument above gives
.t2 / .t1 C b a/
lim V .t1 ; t2 / D lim
.t1 ;t2 /!.a;b/ .t1 ;t2 /!.a;b/ j.t2 / .t1 C b a/j
t1 <t2 t1 <t2
.s2 / .s1 /
D lim D T .b/ D V .a; b/:
.s1 ;s2 /!.b;b/ j.s2 / .s1 /j
s1 >s2
Thus V is continuous.
Since is simply connected, Corollary A.57 guarantees that V W ! S1 has a
continuous lift ' W ! R, which is unique if we require '.a; a/ D 0 (Fig. 9.9). This
is our secant angle function.
We can express the rotation index in terms of the secant angle function as follows:
1 1 1
. / D .b/ .a/ D '.b; b/ '.a; a/ D '.b; b/:
2 2 2
Observe that along the side of where t1 D a and t2 2 Œa; b, the vector V .a; t2 /
has its tail at the origin and its head in the upper half-plane. Since we stipulate that
'.a; a/ D 0, we must have '.a; t2 / 2 Œ0; on this segment. By continuity, there-
fore, '.a; b/ D (since '.a; b/ represents the tangent angle of T .b/ D .1; 0/).
Similarly, on the side where t2 D b, the vector V .t1 ; b/ has its head at the origin and
its tail in the upper half-plane, so '.t1 ; b/ 2 Œ; 2. Therefore, since '.b; b/ repre-
sents the tangent angle of T .b/ D .1; 0/, we must have '.b; b/ D 2 and therefore
. / D 1. This completes the proof for the case in which 0 is continuous.
Now suppose has one or more ordinary vertices. It suffices to show there is a
curve with a continuous velocity vector field that has the same rotation index as .
We will construct such a curve by “rounding the corners” of . It will simplify the
proof somewhat if we choose the parameter interval Œa; b so that .a/ D .b/ is not
one of the ordinary vertices.
Fig. 9.10: Isolating the change in the tangent angle at a vertex
Fig. 9.11: Rounding a corner
Let .ai / be any ordinary vertex, let "i be its exterior angle, and let ˛ be a
positive number less than 12 . j"i j/. Recall that is continuous from the right
at ai and limt%ai .t / D .ai / "i . Therefore, we can choose ı small enough that
j.t / ..ai / "i /j < ˛ when t 2 .ai ı; ai /, and j.t / .ai /j < ˛ when t 2
.ai ; ai C ı/.
The image under of Œa; bX.ai ı; ai Cı/ is a compact set that does not contain
.ai /, so we can choose r small enough that does not enter Bxr ..ai // except when
t 2 .ai ı; ai C ı/. Let t1 2 .ai ı; ai / denote a time when enters Bxr ..ai //, and
t2 2 .ai ; ai C ı/ a time when it leaves (Fig. 9.10). By our choice of ı, the total
change in .t / is not more than ˛ when t 2 Œt1 ; ai /, and again not more than ˛ when
t 2 .ai ; t2 . Therefore, the total change in .t / during the time interval Œt1 ; t2 is
between "i 2˛ and "i C 2˛, which implies < < .
Now we simply replace jŒt1 ;t2 with a smooth curve segment that has the same
velocity as at .t1 / and .t2 /, and whose tangent angle increases or decreases
monotonically from .t1 / to .t2 /; an arc of a hyperbola will do (Fig. 9.11). Since
The Gauss–Bonnet Formula 271
Fig. 9.12: A curved polygon on a surface
the change in tangent angle of is between and and represents the angle
between T .t1 / and T .t2 /, it must be exactly . Repeating this process for each
vertex, we obtain a new curved polygon with a continuous velocity vector field
whose rotation index is the same as that of , thus proving the theorem. t
u
From the rotation index theorem, it is not hard to deduce the three local-to-global
theorems mentioned at the beginning of the chapter as corollaries. (The angle-sum
theorem is immediate; for the total curvature theorem, the trick is to show that 0 .t /
is equal to the signed curvature of ; the circumference theorem follows from the to-
tal curvature theorem as mentioned in Chapter 1.) However, instead of proving them
directly, we will prove a general formula, called the Gauss–Bonnet formula, from
which these results and more follow easily. You will easily see how the statement
and proof of Theorem 9.3 below can be simplified in case the metric is Euclidean.
The Gauss–Bonnet Formula
We now direct our attention to the case of an oriented Riemannian 2-manifold

.M; g/. In this setting, an admissible simple closed curve W Œa; b ! M is called a
curved polygon in M if the image of is the boundary of a precompact open set
˝ M , and there is an oriented smooth coordinate disk containing x̋ under whose
image is a curved polygon in the plane (Fig. 9.12). As in the planar case, we call
˝ the interior of . A curved polygon whose edges are all geodesic segments is
called a geodesic polygon.
For a curved polygon in M , our previous definitions go through almost
unchanged. We say that is positively oriented if it is parametrized in the direc-
tion of its Stokes orientation as the boundary of ˝. On each smooth segment of ,
we define the unit tangent vector field T .t / D 0 .t /=j 0 .t /jg . If .ai / is an ordinary
or flat vertex, we define the exterior angle of at .ai / as the oriented measure "i
of the angle from T .ai / to T .aiC / with respect to the g-inner product and the given
orientation of M ; explicitly, this is

d Vg T .ai /; T .aiC / ˝ ˛
"i D ˇ ˇ arccos T .ai /; T .aiC / g : (9.4)
ˇd Vg T .a /; T .aC / ˇ
i i
The corresponding interior angle of at .ai / is i D "i . Exterior and interior

angles at .a/ D .b/ are defined similarly.
We need a version of the rotation index theorem for curved polygons in M . Sup-
pose W Œa; b ! M is a curved polygon and ˝ is its interior, and let .U; '/ be an
oriented smooth chart such that U contains x̋ . Using the coordinate map ' to trans-
fer , ˝, and g to the plane, we may as well assume that g is a metric on some
open subset Uy R2 , and is a curved polygon in Uy . Let .E1 ; E2 / be the ori-
ented orthonormal frame for g obtained by applying the Gram–Schmidt algorithm
to .@x ; @y /, so that E1 is a positive scalar multiple of @x everywhere in Uy .
We define a tangent angle function for to be a piecewise continuous function
W Œa; b ! R that satisfies
T .t / D cos .t /E1 j.t/ C sin .t /E2 j.t/
at each t where 0 is continuous, and that is continuous from the right and satisfies
(9.1) and (9.2) at vertices. The existence of such a function follows as in the planar
case, using the fact that
T .t / D u1 .t /E1 j.t/ C u2 .t /E2 j.t/ (9.5)
for a pair of piecewise continuous functions u1 ; u2 W Œa; b ! R that can be regarded

as the coordinate functions of a map .u1 ; u2 / W Œa; b ! S1 because T has unit
length.
1
The rotation index of is . / D 2 .b/ .a/ . Because of the role played
by the specific frame .E1 ; E2 / in the definition, it is not obvious that the rotation
index has any coordinate-independent meaning; however, the following easy conse-
quence of the rotation index theorem shows that it does not depend on the choice of
coordinates.
Lemma 9.2. If M is an oriented Riemannian 2-manifold, the rotation index of every

positively oriented curved polygon in M is C1.
Proof. If we use the given oriented coordinate chart to regard as a curved polygon
in the plane, we can compute its tangent angle function either with respect to g or
with respect to the Euclidean metric gx. In either case, . / is an integer because
.a/ and .b/ both represent the angle between the same two vectors, calculated
with respect to some inner product. Now for 0 s 1, let gs D sg C .1 s/x g . By
the same reasoning, the rotation index gs . / with respect to gs is also an integer
for each s, so the function f .s/ D gs . / is integer-valued.
In fact, the function f is continuous in s, as can be deduced easily from the

following observations: (1) Our preferred gs -orthonormal frame E1.s/ ; E2.s/
depends continuously on s, as can be seen from the formulas (2.5)–(2.6) used to
implement the Gram–Schmidt algorithm. (2) On every interval Œai1 ; ai where is
smooth, the functions u1 and u2 satisfying the gs -analogue of (9.5) can be expressed
˝ ˛
as uj .t; s/ D Ts .t /; Ej.s/ j.t/ gs , where Ts .t / D 0 .t /=j 0 .t /jgs . Thus u1 and u2 de-
pend continuously on .t; s/ 2 Œai1 ; ai Œ0; 1, so the function .u1 ; u2 / W Œai1 ; ai
Fig. 9.13: N.t/ is the inward-pointing normal
Œ0; 1 ! S1 has a continuous lift W Œai1 ; ai Œ0; 1 ! R, uniquely determined by

its value at one point. (3) At each vertex, it follows from formula (9.4) that the
exterior angle depends continuously on gs .
Because f is continuous and integer-valued, it follows that g . / D f .1/ D
f .0/ D gx . / D 1, which was to be proved. t
u
From this point onward, we assume for convenience that our curved polygon
is given a unit-speed parametrization, so the unit tangent vector field T .t / is equal
to 0 .t /. There is a unique unit normal vector field N along the smooth portions
of such that . 0 .t /; N.t // is an oriented orthonormal basis for T.t/ M for each t .
If is positively oriented as the boundary of ˝, this is equivalent to N being the
inward-pointing normal to @˝ (Fig. 9.13). We define the signed curvature of at
smooth points of by

N .t / D hDt 0 .t /; N.t /ig :
By differentiating j 0 .t /j2g 1, we see that Dt 0 .t / is orthogonal to 0 .t /, and there-
fore we can write Dt 0 .t / D
N .t /N.t /, and the (unsigned) geodesic curvature of
is
.t / D j
N .t /j. The sign of
N .t / is positive if is curving toward ˝, and
negative if it is curving away.
Theorem 9.3 (The Gauss–Bonnet Formula). Let .M; g/ be an oriented Riemann-

ian 2-manifold. Suppose is a positively oriented curved polygon in M , and ˝ is
its interior. Then
Z Z Xk
K dA C
N ds C "i D 2; (9.6)
˝ iD1
where K is the Gaussian curvature of g, dA is its Riemannian volume form,

"1 ; : : : ; "k are the exterior angles of , and the second integral is taken with respect
to arc length (Problem 2-32).
Proof. Let .a0 ; : : : ; ak / be an admissible partition of Œa; b, and let .x; y/ be oriented
smooth coordinates on an open set U containing x̋ . Let W Œa; b ! R be a tangent
angle function for . Using the rotation index theorem and the fundamental theorem
of calculus, we can write
k
X k Z
X ai
2 D .b/ .a/ D "i C 0 .t / dt: (9.7)
iD1 iD1 ai 1
To prove (9.6), we need to derive a relationship among 0 ,

N , and K.
Let .E1 ; E2 / be the oriented g-orthonormal frame obtained by applying the
Gram–Schmidt algorithm to .@=@x; @=@y/ as before. Then by definition of and
N , the following formulas hold at smooth points of :
0 .t / D cos .t /E1 j.t/ C sin .t /E2 j.t/ I

N.t / D sin .t /E1 j.t/ C cos .t /E2 j.t/ :
Differentiating 0 (and omitting the t dependence from the notation for simplicity),
we get
Dt 0 D .sin / 0 E1 C .cos /r 0 E1 C .cos / 0 E2 C .sin /r 0 E2

(9.8)
D 0 N C .cos /r 0 E1 C .sin /r 0 E2 :
Next we analyze the covariant derivatives of E1 and E2 . Because .E1 ; E2 / is an

orthonormal frame, for every vector v we have
0 D rv jE1 j2 D 2hrv E1 ; E1 i;
0 D rv jE2 j2 D 2hrv E2 ; E2 i;
0 D rv hE1 ; E2 i D hrv E1 ; E2 i C hE1 ; rv E2 i:
The first two equations show that rv E1 is a multiple of E2 and rv E2 is a multiple

of E1 . Define a 1-form ! by
!.v/ D hE1 ; rv E2 i D hrv E1 ; E2 i:
It follows that the covariant derivatives of the basis vectors are given by
rv E1 D !.v/E2 I
(9.9)
rv E2 D !.v/E1 :
Thus the 1-form ! completely determines the connection in U . (In fact, when the
connection is expressed in terms of the local frame .E1 ; E2 / as in Problem 4-14, this
computation shows that the connection 1-forms are just !2 1 D !1 2 D !, !1 1 D
!2 2 D 0; but it is simpler in this case just to derive the result directly as we have
done.)
Using (9.8) and (9.9), we can compute

N D hDt 0 ; N i
D h 0 N; N i C cos hr 0 E1 ; N i C sin hr 0 E2 ; N i
D 0 cos h!. 0 /E2 ; N i C sin h!. 0 /E1 ; N i
D 0 cos2 !. 0 / sin2 !. 0 /
D 0 !. 0 /:
Therefore, (9.7) becomes

k
X k Z
X ai k Z
X ai
2 D "i C
N .t / dt C !. 0 .t // dt
iD1 iD1 ai 1 iD1 ai 1
k
X Z Z
D "i C
N ds C !:
iD1
The theorem will therefore be proved if we can show that

Z Z
!D K dA: (9.10)
˝
Because ˝ is a smooth manifold with corners (see [LeeSM, pp. 415–419]), we

canR apply Stokes’s theorem and conclude that the left-hand side of (9.10) is equal
to ˝ d!. The last step of the proof is to show that d! D K dA. This follows from
the general formula relating the curvature tensor and the connection 1-forms given
in Problem 7-5; but in the case of two dimensions we can give an easy direct proof.
Since .E1 ; E2 / is an oriented orthonormal frame, it follows from Proposition 2.41
that dA.E1 ; E2 / D 1. Using (9.9), we compute
K dA.E1 ; E2 / D K D Rm.E1 ; E2 ; E2 ; E1 /
˝ ˛
D rE1 rE2 E2 rE2 rE1 E2 rŒE1 ;E2 E2 ; E1
˝ ˛
D rE1 .!.E2 /E1 / rE2 .!.E1 /E1 / !.ŒE1 ; E2 /E1 ; E1
˝
D E1 .!.E2 //E1 C !.E2 /rE1 E1 E2 .!.E1 //E1
˛
!.E1 /rE2 E1 !.ŒE1 ; E2 /E1 ; E1
D E1 .!.E2 // E2 .!.E1 // !.ŒE1 ; E2 /
D d!.E1 ; E2 /:
This completes the proof. t

u
The three local-to-global theorems of plane geometry stated in Chapter 1 follow
from the Gauss–Bonnet formula as easy corollaries.
Corollary 9.4 (Angle-Sum Theorem). The sum of the interior angles of a Euclidean
triangle is . t
u
Corollary 9.5 (Circumference Theorem). The circumference of a Euclidean circle
of radius R is 2R. t
u
(a) (b)
Fig. 9.14: Valid and invalid intersections of triangles in a triangulation
Corollary 9.6 (Total Curvature Theorem). If W Œa; b ! R2 is a smooth, unit-

speed, simple closed curve such that 0 .a/ D 0 .b/, and N is the inward-pointing
normal, then
Z b

N .t / dt D 2: t
u
a
The Gauss–Bonnet Theorem
It is now a relatively easy matter to “globalize” the Gauss–Bonnet formula to obtain

the Gauss–Bonnet theorem. The link between the local and global results is pro-
vided by triangulations, so we begin by discussing this construction borrowed from
algebraic topology.
Let M be a smooth, compact 2-manifold. A curved triangle in M is a curved
polygon with exactly three edges and three vertices. A smooth triangulation of M
is a finite collection of curved triangles with disjoint interiors such that the union of
the triangles with their interiors is M , and the intersection of any pair of triangles (if
not empty) is either a single vertex of each or a single edge of each (Fig. 9.14). Every
smooth, compact surface possesses a smooth triangulation. In fact, it was proved by
Tibor Radó [Rad25] in 1925 that every compact topological 2-manifold possesses
a triangulation (without the assumption of smoothness of the edges, of course), in
which every edge belongs to exactly two triangles. There is a proof for the smooth
case that is not terribly hard, based on choosing geodesic triangles contained in
convex geodesic balls (see Problem 9-5).
If M is a triangulated 2-manifold, the Euler characteristic of M (with respect
to the given triangulation) is defined to be
.M / D V E C F;
where V is the number of vertices in the triangulation, E is the number of edges,

and F is the number of faces (the interiors of the triangles). It is an important result
The Gauss–Bonnet Theorem 277
of algebraic topology that the Euler characteristic is in fact a topological invariant,

and is independent of the choice of triangulation (see [LeeTM, Cor. 10.25]), but we
do not need that result here.
Theorem 9.7 (The Gauss–Bonnet Theorem). If .M; g/ is a smoothly triangulated

compact Riemannian 2-manifold, then
Z
K dA D 2.M /;
M
where K is the Gaussian curvature of g and dA is its Riemannian density.
Proof. We may as well assume that M is connected, because if not we can prove
the theorem for each connected component and add up the results.
First consider the case in which
R M is orientable. In this case, we can choose an
orientation for M , and then M K dA gives the same result whether we interpret
dA as the Riemannian density or as the Riemannian volume form, so we will use
the latter interpretation for the proof. Let f˝i W i D 1; : : : ; F g denote the faces of the
triangulation, and for each i , let fij W j D 1; 2; 3g be the edges of ˝i and fij W j D
1; 2; 3g its interior angles. Since each exterior angle is minus the corresponding
interior angle, applying the Gauss–Bonnet formula to each triangle and summing
over i gives
F Z
X X 3 Z
F X F X
X 3 F
X
K dA C
N ds C . ij / D 2: (9.11)
iD1 ˝i iD1 j D1 ij iD1 j D1 iD1
Note that each edge integral appears exactly twice in the above sum, with oppo-
site orientations, so the integrals of
N all cancel out. Thus (9.11) becomes
Z F X
X 3
K dA C 3F ij D 2F: (9.12)
M iD1 j D1
Note also that each interior angle ij appears exactly once. At each vertex, the angles
that touch that vertex must have interior measures that add up to 2 (Fig. 9.15); thus
the angle sum can be rearranged to give exactly 2V . Equation (9.12) thus can be
written Z
K dA D 2V F: (9.13)
M
Finally, since each edge appears in exactly two triangles, and each triangle has
exactly three edges, the total number of edges counted with multiplicity is 2E D 3F ,
where we count each edge once for each triangle in which it appears. This means
that F D 2E 2F , so (9.13) finally becomes
Z
K dA D 2V 2E C 2F D 2.M /:
M
Fig. 9.15: Interior angles at a vertex add up to 2
Now suppose M is nonorientable. Then Proposition B.18 shows that there is an ori-
entable connected smooth manifold M that admits a 2-sheeted smooth covering
yW M
! M , and Exercise A.62 shows that M is compact. If we endow M with the
metric gy D c
y g, then the Riemannian density of gy is given by dA D y dA, and
its Gaussian curvature is Ky D
y K, so c
y .K dA/ D Ky dA. The result of Problem
R R
2-14 shows that M Ky c
dA D 2 K dA.
M
To compare Euler characteristics, we will show that the given triangulation of M
“lifts” to a smooth triangulation of M . To see this, let be any curved triangle in
M and let ˝ be its interior. By definition, this means that there exists a smooth chart
x̋ 2
.U;
'/ whose domain contains and whose image is a disk D R , and such that
' x̋ D x̋0 , where ˝0 is the interior of a curved triangle 0 in R2 . Then ' 1 is an
embedding of D into M , which restricts to a diffeomorphism F W x̋0 ! x̋ . Because
D is simply connected, it follows from Corollary A.57 that ' 1 (and therefore also
F ) has a lift to M , which is smooth because y is a local diffeomorphism; and
because the covering is two-sheeted, there are exactly two such lifts F1 ; F2 . Each
lift is injective because y ı Fi D F , which is injective, and their images are disjoint
because if two lifts agree at a point, they must be identical. From this it is straightfor-

of M with twice
ward to verify that the lifted curved triangles form a triangulation

as many vertices, edges, and faces as that of M , and thus M D 2.M /. Substi-
tuting these relations into the Gauss–Bonnet theorem for M and dividing through
by 2, we obtain the analogous relation for M . t
u
The significance of this theorem cannot be overstated. Together with the classi-
fication theorem for compact surfaces, it gives us very detailed information about
the possible Gaussian curvatures for metrics on compact surfaces. The classification
theorem [LeeTM, Thms. 6.15 and 10.22] says that every compact, connected, ori-
entable 2-manifold M is homeomorphic to a sphere or a connected sum of n tori,
and every nonorientable one is homeomorphic to a connected sum of n copies of the
real projective plane RP 2 ; the number n is called the genus of M . (The sphere is
said to have genus zero.) By constructing simple triangulations, one can show that
The Gauss–Bonnet Theorem 279
the Euler characteristic of an orientable surface of genus n is 2 2n, and that of a

nonorientable one is 2 n. The following corollary follows immediately from the
Gauss–Bonnet theorem.
Corollary 9.8 Let .M; g/ be a compact Riemannian 2-manifold and let K be its
Gaussian curvature.
(a) If M is homeomorphic to the sphere or the projective plane, then K > 0 some-
where.
(b) If M is homeomorphic to the torus or the Klein bottle, then either K 0 or K
takes on both positive and negative values.
(c) If M is any other compact surface, then K < 0 somewhere. t
u
This corollary has a remarkable converse, proved in the mid-1970s by Jerry Kaz-
dan and Frank Warner: If K is any smooth function on a compact 2-manifold M
satisfying the necessary sign condition of Corollary 9.8, then there exists a Rie-
mannian metric on M for which K is the Gaussian curvature. The proof is a deep
application of the theory of nonlinear partial differential equations. (See [Kaz85] for
a nice expository account.)
In Corollary 9.8 we assumed we knew the topology of M and drew conclusions
about the possible curvatures it could support. In the following corollary we reverse
our point of view, and use assumptions about the curvature to draw conclusions
about the topology of the manifold.
Corollary 9.9 Let .M; g/ be a compact Riemannian 2-manifold and K its Gaussian
curvature.
(a) If K > 0 everywhere on M , then the universal covering manifold of M is
homeomorphic to S2 , and 1 .M / is either trivial or isomorphic to the two-
element group Z=2.
(b) If K 0 everywhere on M , then the universal covering manifold of M is
homeomorphic to R2 , and 1 .M / is infinite.
Proof. Suppose first that M has positive Gaussian curvature. From the Gauss–
Bonnet theorem, M has positive Euler characteristic. The classification theorem
for compact surfaces shows that the only such surfaces are the sphere (with trivial
fundamental group) and the projective plane (with fundamental group isomorphic
to Z=2), both of which are covered by the sphere.
On the other hand, suppose M has nonpositive Gaussian curvature. Then its Euler
characteristic is nonpositive, so it is either an orientable surface of genus n 1 or
a nonorientable one of genus n 2. Thus the universal covering space of M is
R2 if M is the torus or the Klein bottle, and B2 in all other cases (see [LeeTM,
Thm. 12.29]), both of which are homeomorphic to R2 . The fact that the universal
covering space is noncompact implies that the universal covering map has infinitely
many sheets by the result of Exercise A.62, and then Proposition A.61 shows that
1 .M / is infinite. t
u
Much of the effort in contemporary Riemannian geometry is aimed at generaliz-

ing the Gauss–Bonnet theorem and its topological consequences to higher dimen-
sions. As we will see in the next few chapters, most of the interesting results have
required the development of different methods.
However, there is one rather direct generalization of the Gauss–Bonnet theorem
that deserves mention. For a compact manifold M of any dimension, the Euler
characteristic of M , denoted by .M /, can be defined analogously to that of a
surface and is a topological invariant (see [LeeTM, Thm. 13.36]). It turns out that
it is always zero for an odd-dimensional compact manifold, but it is a nontrivial
invariant in each even-dimensional case.
The Chern–Gauss–Bonnet theorem equates the Euler characteristic of an even-
dimensional compact oriented Riemannian manifold to a certain curvature integral.
Versions of this theorem were proved by Heinz Hopf in 1925 for an embedded Rie-
mannian hypersurface in Euclidean space, and independently by Carl Allendoerfer
and Werner Fenchel in 1940 for an embedded Riemannian submanifold of any
Euclidean space (well before Nash’s 1956 proof that every Riemannian manifold
has such an embedding). Finally, an intrinsic proof for the general case was discov-
ered by Shiing-Shen Chern in 1944 (see [Spi79, Vol. 5] for a complete discussion
with references). The theorem asserts that for each 2n-dimensional oriented inner
product space V , there exists a basis-independent function
Pf W R.V / ! ƒ2n .V /;
called the Pfaffian, with the property that for every oriented compact Riemannian
2n-manifold M , Z
Pf .Rm/ D .2/n .M /:
M
(Depending on how the Pfaffian is defined, you will see different choices of nor-
malization constants on the right-hand side of this equation.) For example, in four
dimensions, the theorem can be written in terms of familiar curvature quantities as
follows: Z

jRmj2 4jRcj2 C S 2 d Vg D 32 2 .M /: (9.14)
M
In a certain sense, this might be considered a very satisfactory generalization of
Gauss–Bonnet. The only problem with this result is that the relationship between the
Pfaffian and sectional curvatures is obscure in higher dimensions, so it is very hard
to interpret the theorem geometrically. For example, after he proved the version
of the theorem for Euclidean hypersurfaces, Hopf conjectured in the 1930s that
every compact even-dimensional manifold that admits a metric with strictly positive
sectional curvature must have positive Euler characteristic; to date, the conjecture is
known to be true in dimensions 2 and 4, but it is still open in higher dimensions (see
[Pet16, p. 320]).
Problems 281
Problems
9-1. Let .M; g/ be an oriented Riemannian 2-manifold with nonpositive Gaus-

sian curvature everywhere. Prove that there are no geodesic polygons with
exactly 0, 1, or 2 ordinary vertices. Give examples of all three if the curvature
hypothesis is not satisfied.
9-2. Let .M; g/ be a Riemannian 2-manifold. If is a geodesic polygon in M
with n vertices, the angle excess of is defined as
X
n
E. / D i .n 2/;
iD1
where 1 ; : : : ; n are the interior angles of . Show that if M has constant

Gaussian curvature K, then every geodesic polygon has angle excess equal to
K times the area of the region bounded by the polygon.
9-3. Given h 2 .R; R/, let Ch be the circle in S2 .R/ R3 where z D h (where
we label the standard coordinates in R3 as .x; y; z/), and let ˝ be the subset
of S2 .R/ where z > h. Compute the signed curvature of Ch and verify the
Gauss–Bonnet formula in this case.
9-4. Let T R3 be the torus of revolution obtained by revolving the circle
.r 2/2 C z 2 D 1 around the z-axis (see p. 19). Compute the Gaussian curva-
ture of T and verify the Gauss–Bonnet theorem in this case.
9-5. This problem outlines a proof that every compact smooth 2-manifold has a
smooth triangulation.
(a) Show that it suffices to prove that there exist finitely many convex geodesic
polygons whose interiors cover M , and each of which lies in a uni-
formly normal convex geodesic ball. (A curved polygon is called con-
vex if the union of the polygon and its interior is a geodesically convex
subset of M .)
(b) Using Theorem 6.17, show that there exist finitely many points v1 , . . . ,
vk and a positive number " such that the geodesic balls B3" .vi / are
geodesically convex and uniformly normal, and the balls B" .vi / cover M .
(c) For each i , show that there is a convex geodesic polygon in B3" .vi /
whose interior contains B" .vi /. [Hint: Let the vertices be sufficiently
nearby points on the circle of radius 2" around vi .]
(d) Prove the result.
(Used on p. 276.)
9-6. Let M R3 be a compact, embedded, 2-dimensional Riemannian subman-
ifold. Show that M cannot have K 0 everywhere. [Hint: Look at a point
where the distance from the origin takes a maximum.]
9-7. Suppose M is either the 2-sphere of radius R or the hyperbolic plane of

radius R for some R > 0. Show that similar triangles in M are congruent.
More precisely, if 1 and 2 are geodesic triangles in M such that corre-
sponding side lengths are proportional and corresponding interior angles are
equal, then there exists an isometry of M taking 1 to 2 .
9-8. Use the Gauss–Bonnet theorem to prove that every compact connected Lie
group of dimension 2 is isomorphic to the direct product group S1 S1 .
[Hint: See Problem 8-17.]
9-9. (a) Show that there is an upper bound for the areas of geodesic triangles in
the hyperbolic plane H2 .R/, and compute the least upper bound.
(b) Two distinct maximal geodesics in the hyperbolic plane H2 are said to be
asymptotically parallel if they have unit-speed parametrizations 1 ; 2 W
R ! H2 such that dgM .1 .t /; 2 .2// remains bounded as t ! C1 or as
t ! 1. An ideal triangle in H2 is a region whose boundary consists of
three distinct maximal geodesics, any two of which are asymptotically
parallel to each other. Show that all ideal triangles have the same finite
area, and compute it. Be careful to justify any limits.
9-10. T HE G AUSS –B ONNET THEOREM FOR SURFACES WITH BOUNDARY: Sup-

pose .M; g/ is a compact Riemannian 2-manifold with boundary, endowed
with a smooth triangulation such that the intersection of each curved triangle
with @M , if not empty, is either a single vertex or a single edge. Then
Z Z
K dA C
N ds D 2.M /;
M @M
where
N is the signed geodesic curvature of @M with respect to the inward-
pointing normal N .
9-11. Suppose g is a Riemannian metric on the cylinder S1 Œ0; 1 such that both
boundary curves are totally geodesic. Prove that the Gaussian curvature of
g either is identically zero or attains both positive and negative values. Give
examples of both possibilities.
9-12. Prove the plane curve classification theorem (Theorem 1.5). [Hint: Show that
every smooth unit-speed plane curve .t / D .x.t /; y.t // satisfies the second-
order ODE 00 .t / D
N .t /N.t /, where N is the unit normal vector field given
by N.t / D .y 0 .t /; x 0 .t //.] (Used on p. 4.)
9-13. Use the four-dimensional Chern–Gauss–Bonnet formula (9.14) to prove that
a compact 4-dimensional Einstein manifold must have positive Euler char-
acteristic unless it is flat.
Chapter 10
Jacobi Fields
Our goal for the remainder of this book is to generalize to higher dimensions some
of the geometric and topological consequences of the Gauss–Bonnet theorem. We
need to develop a new approach: instead of using Stokes’s theorem and differential
forms to relate the curvature to global topology as in the proof of the Gauss–Bonnet
theorem, we study the way that curvature affects the behavior of nearby geodesics.
Roughly speaking, positive curvature causes nearby geodesics to converge, while
negative curvature causes them to spread out (Fig. 10.1). In order to draw topological
consequences from this fact, we need a quantitative way to measure the effect of
curvature on a one-parameter family of geodesics.
We begin by deriving the Jacobi equation, which is an ordinary differential equa-
tion satisfied by the variation field of any one-parameter family of geodesics. A
vector field satisfying this equation along a geodesic is called a Jacobi field. We
then introduce conjugate points, which are pairs of points along a geodesic where
some nontrivial Jacobi field vanishes. Intuitively, if p and q are conjugate along a
geodesic, one expects to find a one-parameter family of geodesic segments that start
at p and end (almost) at q.
After defining conjugate points, we prove a simple but essential fact: the points
conjugate to p are exactly the points where expp fails to be a local diffeomorphism.
We then derive an expression for the second derivative of the length functional
with respect to proper variations of a geodesic, called the second variation formula.
Using this formula, we prove another essential fact about conjugate points: once a
geodesic passes its first conjugate point, it is no longer minimizing. The converse of
this statement, however, is untrue: a geodesic can cease to be minimizing before it
reaches its first conjugate point. In the last section of the chapter, we study the set
of points where geodesics starting at a given point p cease to minimize, called the
cut locus of p.
In the next two chapters, we will derive geometric and topological consequences
of these facts.

284 10 Jacobi Fields
Fig. 10.1: Geodesics converge in positive curvature, and spread out in negative curvature
The Jacobi Equation

Let .M; g/ be an n-dimensional Riemannian or pseudo-Riemannian manifold. In
order to study the effect of curvature on nearby geodesics, we focus on variations
through geodesics. Suppose, therefore, that I; K R are intervals, W I ! M is a
geodesic, and W K I ! M is a variation of (as defined in Chapter 6). We say
that is a variation through geodesics if each of the main curves s .t / D .s; t /
is also a geodesic. (In particular, this requires that be smooth.) Our first goal is to
derive an equation that must be satisfied by the variation field of a variation through
geodesics.
Theorem 10.1 (The Jacobi Equation). Let .M; g/ be a Riemannian or pseudo-

Riemannian manifold, let be a geodesic in M , and let J be a vector field along
. If J is the variation field of a variation through geodesics, then J satisfies the
following equation, called the Jacobi equation:

Dt2 J C R J; 0 0 D 0: (10.1)
Proof. Write T .s; t / D @t .s; t / and S.s; t / D @s .s; t / as in Chapter 6. The

geodesic equation tells us that
Dt T 0
for all .s; t /. We can take the covariant derivative of this equation with respect to s,
yielding
Ds Dt T 0:
Using Proposition 7.5 to commute the covariant derivatives along , we compute
0 D Ds D t T
D Dt Ds T C R.S; T /T
D Dt Dt S C R.S; T /T;
where the last step follows from the symmetry lemma. Evaluating at s D 0, where
S.0; t / D J.t / and T .0; t / D 0 .t /, we get (10.1). t
u
A smooth vector field along a geodesic that satisfies the Jacobi equation is called
a Jacobi field. As the following proposition shows, the Jacobi equation can be writ-
The Jacobi Equation 285
ten as a system of second-order linear ordinary differential equations, so it has a

unique solution given initial values for J and Dt J at one point.
Proposition 10.2 (Existence and Uniqueness of Jacobi Fields). Let .M; g/ be a

Riemannian or pseudo-Riemannian manifold. Suppose I R is an interval, W I !
M is a geodesic, a 2 I , and p D .a/. For every pair of vectors v; w 2 Tp M , there
is a unique Jacobi field J along satisfying the initial conditions
J.a/ D v; Dt J .a/ D w:
Proof. Choose a parallel orthonormal frame .Ei / along , and write v D v i Ei .a/,
w D w i Ei .a/, and 0 .t / D y i .t /Ei .t / in terms of this frame. Writing an unknown
vector field J 2 X. / as J.t / D J i .t /Ei .t /, we can express the Jacobi equation as
JR i .t / C Rj kl i ..t //J j .t /y k .t /y l .t / D 0:
This is a system of n linear second-order ODEs for the n functions J i W I ! R.

Making the substitution W i D JP i converts it to the following
equivalent first-order
linear system for the 2n unknown functions J i ; W i :
JP i .t / D W i .t /;
WP i .t / D Rj kl i ..t //J j .t /y k .t /y l .t /:
Then Theorem 4.31 guarantees the existence and uniqueness of a smooth solution
on the whole interval I with arbitrary initial conditions J i .a/ D v i , W i .a/ D w i .
Since Dt J.a/ D JP i .a/Ei .a/ D W i .a/Ei .a/ D w, it follows that J.t / D J i .t /Ei .t /
is the desired Jacobi field. t
u
Given a geodesic , let J. / X. / denote the set of Jacobi fields along .

of dimension n, and is any geodesic in M . Then J. / is a 2n-dimensional linear
subspace of X. /.
Proof. Because the Jacobi equation is linear, J. / is a linear subspace of X. /. Let
p D .a/ be any point on , and consider the linear map from J. / to Tp M ˚ Tp M
by sending J to .J.a/; Dt J.a//. The preceding proposition says precisely that this
map is bijective. t
u
The following proposition is a converse to Theorem 10.1; it shows that each
Jacobi field along a geodesic segment tells us how some family of geodesics
behaves, at least to first order along .
Proposition 10.4. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold,

and let W I ! M be a geodesic. If M is complete or I is a compact interval,
then every Jacobi field along is the variation field of a variation of through
geodesics.
Proof. Let J be a Jacobi field along . After applying a translation in t (which does
not affect either the fact that is a geodesic or the fact that J is a Jacobi field), we
can assume that the interval I contains 0, and write p D .0/ and v D 0 .0/. Note
that this implies .t / D expp .t v/ for all t 2 I .
Choose a smooth curve W ."; "/ ! M and a smooth vector field V along
satisfying
.0/ D p; V .0/ D v;
0
.0/ D J.0/; Ds V .0/ D Dt J.0/;
where Ds and Dt denote covariant differentiation along and , respectively. (They

are easily constructed in local coordinates around p.) We wish to define a variation
of by setting
.s; t / D exp.s/ tV .s/ : (10.2)
If M is geodesically complete, this is defined for all .s; t / 2 ."; "/ I . On the other
hand, if I is compact, the fact that the domain of the exponential map is an open
subset of TM that contains the compact set f.p; t v/ W t 2 I g guarantees that there is
some ı > 0 such that .s; t / is defined for all .s; t / 2 .ı; ı/ I .
Note that

.0; t / D exp.0/ tV .0/ D expp .t v/ D .t /; (10.3)
.s; 0/ D exp.s/ .0/ D .s/: (10.4)
In particular, (10.3) shows that is a variation of . The properties of the exponen-

tial map guarantee that is a variation through geodesics, and therefore its variation
field W .t / D @s .0; t / is a Jacobi field along .
Now, (10.4) implies
ˇ
@ ˇˇ
W .0/ D .s; 0/ D 0 .0/ D J.0/:
@s ˇsD0
If we can show that Dt W .0/ D Dt J.0/ as well, it then follows from the uniqueness
of Jacobi fields that W J , and the proposition is proved.
Formula (10.2) shows that each main curve s .t / is a geodesic whose initial
velocity is V .s/, so
ˇ
@ ˇˇ
@t .s; 0/ D ˇ s .t / D V .s/:
@t tD0
It follows from the symmetry lemma that Dt @s D Ds @t , and our choice of V

gives
Dt W .0/ D Dt @s .0; 0/ D Ds @t .0; 0/ D Ds V .0/ D Dt J.0/:
It follows that W J , as claimed. t

u
Basic Computations with Jacobi Fields 287
Fig. 10.2: Tangential Jacobi fields
Proposition
10.5 (Local Isometry Invariance of Jacobi Fields). Suppose .M; g/
and M ; gz are Riemannian or pseudo-Riemannian manifolds and ' W M ! M is
a local isometry. Let W I !
M and z W I ! M be geodesics related by z D ' ı ,
and let J 2 X. /, Jz 2 X z be related by d'.t/ .J.t // D Jz.t / for all t 2 I . Then
J is a Jacobi field if and only if Jz is.
Basic Computations with Jacobi Fields

There are various situations in which Jacobi fields can be computed explicitly. We
begin by describing the most important of these.
Tangential and Normal Jacobi Fields
Along every geodesic W I ! M , there are always two Jacobi fields that
we can
write down immediately (see Fig. 10.2). Because Dt 0 0 and R 0 ; 0 0 0 by
antisymmetry of R, the vector fields J0 .t / D 0 .t / and J1 .t / D t 0 .t / both satisfy
the Jacobi equation by direct computation. If I is compact or M is complete, the
vector field J0 is the variation field of the variation .s; t / D .s C t /, while J1 is
the variation field of .s; t / D ..1 C s/t /. Therefore, these two Jacobi fields just
reflect the possible reparametrizations of , and do not tell us anything about the
behavior of geodesics other than itself.
To distinguish these trivial cases from more informative ones, we make the
following definitions. Given a regular curve W I ! M , for each t 2 I we let
>
T.t/ M T.t/ M denote the one-dimensional subspace spanned by 0 .t /, and
?
T.t/ M its orthogonal complement. A tangential vector field along is a vector
>
field V 2 X. / such that V .t / 2 T.t/ M for all t , and a normal vector field along
?
is one such that V .t / 2 T.t/ M for all t . Thus a normal Jacobi field along is
a Jacobi field J satisfying J.t / ? 0 .t / for all t . Let X? . / and X> . / denote the
spaces of smooth normal and tangential vector fields along , respectively. When
is a geodesic, J ? . / and J > . / denote the spaces of normal and tangential Jacobi
fields along , respectively.
Proposition 10.7. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold.

Suppose W I ! M is a geodesic and J is a Jacobi field along . Then the fol-
lowing are equivalent:
(a) J is a normal Jacobi field.
(b) J is orthogonal to 0 at two distinct points.
(c) Both J and Dt J are orthogonal to 0 at one point.
(d) Both J and Dt J are orthogonal to 0 everywhere along .
Proof. Define a function f W I ! R by f .t / D hJ.t /; 0 .t /i, so that f .t / D 0 if and

only if J.t / ? 0 .t /. Using compatibility with the metric and the fact that Dt 0 0,
we compute ˝ ˛
f 00 D Dt2 J; 0
˝ ˛
D R J; 0 0 ; 0

D Rm J; 0 ; 0 ; 0 D 0
by the symmetries of the curvature tensor. Thus, by elementary calculus, f is an
affine function of t .
Note that f 0 .t / D hDt J.t /; 0 .t /i, which vanishes at t if and only if Dt J.t / ?
.t /. It follows that J.a/ and Dt J.a/ are orthogonal to 0 .a/ for some a 2 I if and
0
only if f and its first derivative vanish at a, which happens if and only if f 0.
Similarly, J is orthogonal to 0 at two points if and only if f vanishes at two points,
which happens if and only if f is identically zero. If this is the case, then f 0 0 as
well, which implies that both J and Dt J are orthogonal to 0 for all t . t
u
Corollary 10.8. Suppose .M; g/ is a Riemannian or pseudo-Riemannian n-manifold

and W I ! M is any nonconstant geodesic. Then J ? . / is a .2n 2/-dimensional
subspace of J. /, and J > . / is a 2-dimensional subspace. Every Jacobi field can
be uniquely decomposed as a sum of a tangential Jacobi field plus a normal Jacobi
field.
Proof. As we noted in the proof of Corollary 10.3, for every a 2 I , the map from
J. / to T.a/ M ˚ T.a/ M given by J 7! .J.a/; Dt J.a// is an isomorphism, and
Proposition 10.7 shows that J ? . / is exactly the preimage of the subspace consist-
ing of all pairs .v; w/ 2 T.a/ M ˚ T.a/ M such that hv; 0 .a/i D hw; 0 .a/i D 0.
Because this subspace has dimension 2n 2, it follows that J ? . / has the same
dimension.
On the other hand, J > . / contains J0 .t / D 0 .t / and J1 .t / D t 0 .t /, which are
linearly independent over R because 0 .t / never vanishes, so it is a subspace of
dimension at least 2. Because J ? . / \ J > . / D f0g, the dimension of J > . / must
be exactly 2, and we have a direct sum decomposition J. / D J ? . / ˚ J > . /.
This implies that every J 2 J. / has a unique decomposition J D J ? C J > , with
J ? 2 J ? . / and J > 2 J > . /. t
u
Fig. 10.3: The variation of Lemma 10.9
Jacobi Fields Vanishing at a Point

For many purposes, we will be primarily interested in Jacobi fields that vanish at a
particular point. For these, there are some useful explicit formulas.
Lemma 10.9. Let .M; g/ be a Riemannian or pseudo-Riemannian manifold, I R

an interval containing 0, and W I ! M a geodesic. Suppose J W I ! M is
a Jacobi field such that J.0/ D 0. If M is geodesically complete or I is com-
pact, then J is the variation field of the following variation of through geodesics
(Fig. 10.3):
.s; t / D expp t .v C sw/ ; (10.5)
where p D .0/, v D 0 .0/, and w D Dt J.0/.
Proof. The proof of Proposition 10.4 showed that J is the variation field of a vari-
ation of the form (10.2), with any smooth curve satisfying .0/ D p and
0 .0/ D 0, and V a smooth vector field along with V .0/ D v and Ds V .0/ D w. In
this case, we can take .s/ p and V .s/ D v C sw 2 Tp M , yielding (10.5). t
u
This result leads to some explicit formulas for all of the Jacobi fields vanishing
at a point.
Proposition 10.10 (Jacobi Fields Vanishing at a Point). Let .M; g/ be a Rie-

mannian or pseudo-Riemannian n-manifold and p 2 M . Suppose W I ! M is
a geodesic such that 0 2 I and .0/ D p. For every w 2 Tp M , the Jacobi field J
along such that J.0/ D 0 and Dt J.0/ D w is given by

J.t / D d expp tv .t w/; (10.6)
Fig. 10.4: A Jacobi field in normal coordinates
where v D 0 .0/, and we regard t w as an element of Ttv .Tp M / by means of the
canonical identification Ttv .Tp M / Š Tp M . If x i are normal coordinates on a
normal neighborhood of p containing the image of , then J is given by the formula
ˇ
J.t / D t w i @i ˇ.t/ ; (10.7)
ˇ
where w i @i ˇ0 is the coordinate representation of w.
Proof. Under the given hypotheses, Lemma 10.9 showed that the restriction of J
to any compact interval containing 0 is the variation field of a variation through
geodesics of the form (10.5). Using the chain rule to compute J.t / D @s .0; t /,
we arrive at (10.6). Because every t in the domain of is contained in some such
compact interval, the formula holds for all such t .
In normal coordinates, the coordinate representation of the exponential map is
the identity, so can be written explicitly in coordinates as

.s; t / D t .v 1 C sw 1 /; : : : ; t .v n C sw n / :
(See Fig. 10.4.) Differentiating .s; t / with respect to s and setting s D 0 shows that
its variation field J is given by (10.7). t
u

and U is a normal neighborhood of p 2 M . For each q 2 U X fpg, every vector in
Tq M is the value of a Jacobi field J along a radial geodesic such that J vanishes
at p.
Fig. 10.5: The graph of sc
1
Proof. Let x i be normal coordinates on 1U . Given q
D q ; : : : ; q n
2 U X fpg and
i n
w D w @i jq 2 Tq M , the curve .t / D t q ; : : : ; t q is a radial geodesic satisfying
.0/ D p and .1/ D q. The previous proposition showed that J.t / D t w i @i j.t/ is
a Jacobi field along . Because J.0/ D 0 and J.1/ D w, the result follows. t
u
Jacobi Fields in Constant-Curvature Spaces

For metrics with constant sectional curvature, we have a different kind of explicit
formula for Jacobi fields—this one expresses a Jacobi field as a scalar multiple of
a parallel vector field. To handle the various cases concisely, for each c 2 R, let us
define a function sc W R ! R (Fig. 10.5) by
† t; if c D 0I
t 1
sc .t / D R sin ; if c D > 0I (10.8)
R R2
t 1
R sinh ; if c D 2 < 0:
R R
Proposition 10.12 (Jacobi Fields in Constant Curvature). Suppose .M; g/ is a

Riemannian manifold with constant sectional curvature c, and is a unit-speed
geodesic in M . The normal Jacobi fields along vanishing at t D 0 are the vector
fields of the form
J.t / D ksc .t /E.t /; (10.9)
where E is any parallel unit normal vector field along , k is an arbitrary constant,
and sc is defined by (10.8). The initial derivative of such a Jacobi field is
Dt J.0/ D kE.0/; (10.10)
and its norm is

jJ.t /j D jsc .t /j jDt J.0/j: (10.11)
Proof. Since g has constant curvature, its curvature endomorphism is given by the
formula of Proposition 8.36:

R.v; w/x D c hw; xiv hv; xiw :
Substituting this into the Jacobi equation, we find that a normal Jacobi field J satis-
fies
0 D Dt2 J C c h 0 ; 0 iJ hJ; 0 i 0
(10.12)
D Dt2 J C cJ;
where we have used the facts that j 0 j2 D 1 and hJ; 0 i D 0.
Since (10.12) says that the second covariant derivative of J is a multiple of J
itself, it is reasonable to try to construct a solution by choosing an arbitrary parallel
unit normal vector field E along and setting J.t / D u.t /E.t / for some function u
to be determined. Plugging this into (10.12), we find that J is a Jacobi field if and
only if u is a solution to the differential equation
u00 .t / C cu.t / D 0:
It is an easy matter to solve this ODE explicitly. In particular, the solutions satisfying
u.0/ D 0 are constant multiples of sc . This construction yields all the normal Jacobi
fields vanishing at 0, since there is an .n 1/-dimensional space of them, and the
space of parallel normal vector fields has the same dimension.
To prove the last two statements, suppose J is given by (10.9), and compute
Dt J.0/ D ksc0 .0/E.0/ D kE.0/;
since sc0 .0/ D 1 in every case. Because E is a unit vector field, jDt J.0/j D jkj, and
(10.11) follows. t
u
Here is our first significant application of Jacobi fields. Because every tangent
vector in a normal neighborhood is the value of a Jacobi field vanishing at the origin
by Corollary 10.11, Proposition 10.12 yields explicit formulas for constant-
curvature metrics in normal coordinates. To set the stage, we will rewrite the
Euclidean metric on Rn in a form that is somewhat more convenient for these comp-
utations.
Let W Rn X f0g ! Sn1 be the radial projection
x
.x/ D ; (10.13)
jxj
and define a symmetric 2-tensor field on Rn X f0g by
gy D g;
V (10.14)
where gV is the round metric of radius 1 on Sn1 .
Lemma 10.13. On Rn X f0g, the metric gy defined by (10.14) and the Euclidean
metric gx are related by
gx D dr 2 C r 2 gy; (10.15)
where r.x/ D jxj is the Euclidean distance from the origin.
Proof. Example 2.24 observed that the map ˚ W RC Sn1 ! Rn X f0g given by
˚.; !/ D ! (10.16)
is an isometry when RC Sn1 has the warped product metric d2 ˚ 2 gV and
Rn X f0g has the Euclidean metric (see
1 also2 Problem
2-4).2 Because ˚ 1 .x/ D
2 2
.r.x/; .x//, this means that gx D ˚ d ˚ gV D dr C r gy. t
u
Theorem 10.14 (Constant-Curvature Metrics in Normal Coordinates). Suppose

.M; g/is a Riemannian manifold with constant sectional curvature c. Given p 2 M ,
let x i be normal coordinates on a normal neighborhood U of p; let r be the radial
distance function on U defined by (6.4); and let gy be the symmetric 2-tensor defined
in x-coordinates by (10.14). On U X fpg, the metric g can be written
g D dr 2 C sc .r/2 gy; (10.17)
where sc is defined by (10.8).
Proof. Let gx denote the Euclidean metric in x-coordinates, and let gc denote the
metric defined by the formula on the right-hand side of (10.17). By the properties
of normal coordinates, at points of U X fpg, all three metrics g, gx, and gc make the
radial vector field @r a unit vector orthogonal to the level sets of r. Thus we need
only show that g.w1 ; w2 / D gc .w1 ; w2 / when w1 ; w2 are tangent to a level set of r,
and by polarization it suffices to show that g.w; w/ D gc .w; w/ for every such
vector w. Note that if w is tangent to a level set r D b, then formulas (10.17) and
(10.15) imply
sc .b/2
gc .w; w/ D sc .b/2 gy.w; w/ D gx.w; w/:
b2
Let q 2 U X fpg and w 2 Tq M , and assume that w is tangent to the r-level
set containing q. Let b D dg .p; q/, and let W Œ0; b ! U be the unit-speed radial
geodesic from p to q, so the coordinate representation of is

t 1 t n
.t / D q ;:::; q ;
b b
1
where q ; : : : ; q n is the coordinate representation of q. Let J 2 X. / be the vector
field along given by
Fig. 10.6: Local isometry constructed from normal coordinate charts
t i ˇˇ
J.t / D w @i .t/ ; (10.18)
b
where w i @i jq is the coordinate representation for w. By Proposition 10.10, J is a
Jacobi field satisfying Dt J.0/ D .1=b/w i @i jp , and it follows from the definition
that J.b/ D w. Because J is orthogonal to 0 at p and q, it is normal by Proposition
10.7. Thus by Proposition 10.12,
jwj2g D jJ.b/j2g D sc .b/2 jDt J.0/j2g

1 ˇˇ ˇ ˇˇ2 1
D sc .b/2 2 ˇ w i @i ˇp ˇ D sc .b/2 2 jwj2gx D jwj2gc : t
u
b g b
Corollary
10.15 (Local Uniqueness of Constant-Curvature Metrics). Let .M; g/
and M ; gz be Riemannian manifolds of the same dimension with constant sectional
, there exist neighborhoods U of p and Uz
curvature c. For all points p 2 M , pz 2 M
z
of pz and an isometry ' W U ! U .
Proof. Choose p 2 M and pz 2 M , and let U and Uz be geodesic balls of small

radius " around p and p, z respectively. Riemannian normal coordinates give maps
W U ! B" .0/ Rn and z W Uz ! B" .0/ Rn , under which both metrics are
given by formula (10.17) on the complement of the origin (Fig. 10.6). At the origin,
gij D gzij D ıij . Therefore z 1 ı is the required local isometry. t
u
Corollary 10.16 (Constant-Curvature Metrics as Warped Products). Suppose

.M; g/ is a Riemannian manifold with constant sectional curvature c, and U is a
geodesic ball of radius b centered at p 2 M . Then U X fpg is isometric to a warped
product of the form .0; b/ sc Sn1 , where .0; b/ R has the Euclidean metric, and
Sn1 is the unit sphere with the round metric g.V
Proof. By virtue of Theorem 10.14, we may consider g to be a metric on the ball
of radius b in Rn given by formula (10.17). Let ˚ W .0; b/ Sn1 ! U X fpg and
W Rn X f0g ! Sn1 be the maps defined by (10.16) and (10.13). Because ı
˚ restricts to the identity on fg Sn1 for each fixed , it follows that ˚ gy D
˚ gV D g,
V and thus
˚ g D d2 ˚ sc ./2 g:
V t
u
The next corollary describes a formula for integration in polar coordinates with
respect to a constant-curvature metric. It will be useful in our proofs of volume
comparison theorems in the next chapter.
Corollary 10.17 (Polar Decomposition of Integrals). Suppose .M; g/ is a Rie-
mannian manifold with constant sectional curvature c, and U is an open or closed
geodesic ball of radius b around a point p 2 M . If f W U ! R is any bounded
integrable function, then the integral of f over U can be expressed as
Z Z Z b
f d Vg D f ı ˚.; !/sc ./n1 d d VgV ;
U Sn1 0
where d Vg is the Riemannian density of g, and ˚ W .0; b/ Sn1 ! U X fpg is

defined in normal coordinates by (10.16).
Proof. Because every geodesic ball is orientable, we might as well choose an ori-
entation on U and interpret d Vg as a differential form. Since the boundary of a
geodesic ball has measure zero, it does not matter whether U is open or closed.
Similarly, integrating over U X fpg instead of U does not change the value of the
integral. The claim therefore follows from Corollary 10.16 together with the result
of Problem 2-15, which shows that the volume form of the warped product metric
d2 ˚ sc ./2 gV can be written sc ./n1 d ^ d VgV . t
u
Locally Symmetric Spaces
Here is another application of the theory of Jacobi fields. Recall that a Riemannian
manifold is a locally symmetric space if every p 2 M has a neighborhood that
admits a point reflection at p. Problem 7-3 showed that every locally symmetric
space has parallel curvature tensor. Now we can prove the converse. The key is the
following lemma due to Élie Cartan. We will use the lemma again in Chapter 12 to
prove a more global result (see Thm. 12.6).
Lemma 10.18. Suppose .M; g/ is a Riemannian manifold with parallel curvature
tensor, and for some points p; py 2 M we are given a linear map A W Tp M ! Tpy M
satisfying A .gpy / D gp and A .Rmpy / D Rmp . Then there exist a neighborhood U
of p and a local isometry ' W U ! M such that '.p/ D py and d'p D A. If M is
complete, then U can be taken to be any normal neighborhood of p.
Proof. The hypothesis means that rRm 0, and because covariant differentia-
tion commutes with raising and lowering indices, the curvature endomorphism is
also parallel. If M is complete, let U be any normal neighborhood of p; other-
wise choose U D B" .p/, where " > 0 is chosen small enough that both B" .p/
and B" .p/ y are geodesic balls. Our choice guarantees that ' D exppy ıA ı exp1 p is
a smooth map from U into M , and it satisfies '.p/ D py and d'p D A. We will
show that jd'q .x/jg D jxjg for every tangent vector x at every point q 2 U . It then
follows by polarization that hd'q .x/; d'q .y/ig D hx; yig for all x; y 2 Tq M , and
thus ' g D g.
Because d'p D A is a linear isometry, we have jd'p .x/jg D jxjg for x 2 Tp M ,
so we need only consider points q ¤ p. Let q 2 U X fpg and x 2 Tq M be arbitrary.
Let W Œ0; 1 ! U be the radial geodesic from p to q, given explicitly by .t / D
expp .t v/ for some v 2 Tp M . It follows from Corollary 10.11 that there is a Jacobi
field J along such that J.0/D 0 and J.1/ D x; and Proposition 10.10 shows
that it is of the form J.t / D d expp tv .t w/ for some w 2 Tp M . Let vy D A.v/
and w y D A.w/ 2 Tpy M , and define y.t / D exppy .t vy/ and Jy.t / D d.exppy /t vy .t w/
y for
y
t 2 Œ0; 1. Then y is a geodesic from py D '.p/ to '.q/, and J is a Jacobi field along
' and the chain rule (using the fact that dAv D A
y. It follows from the definition of
because A is linear) that d'q ı d expp v D d.exppy /vy ı A, and thus
Jy.1/ D d.exppy /vy .w/

y D d.exppy /vy ı A.w/

D d'q ı d expp v .w/ D d'q .J.1// D d'q .x/;
ˇ ˇ
so to prove the theorem it suffices to show that ˇJy.1/ˇg D jJ.1/jg .

Let .E1 ; : : : ; En / be a parallel orthonormal frame along , and let Ey1 ; : : : ; Eyn
be the parallel orthonormal frame along y such that Eyi .0/ D A.Ei .0//. At points of
, we can express the curvature endomorphism in terms of the frame .Ei / as

R Ei .t /; Ej .t / Ek .t / D Rij k l .t / El .t /;
for some smooth functions Rij k l W Œ0; 1 ! R. The parallel curvature hypothesis and
the fact that each Ei is parallel imply
0
0 D .Dt R/ Ei .t /; Ej .t / Ek .t / D Dt R.Ei .t /; Ej .t //Ek .t / D Rij k l .t / El .t /;
so in fact the coefficients Rij k l .t / are constant in t . The same argument shows
that the curvature
endomorphism has constant coefficients along y in terms of the
frame Eyi . Because our hypotheses guarantee that the coefficients of the two cur-
vature endomorphisms agree at t D 0, they are in fact the same constants along both
geodesics; we write those constants henceforth as Rij k l .
Also, we can write J.t / D J i .t /Ei .t / and Jy.t / D Jyi .t /Eyi .t / for some smooth
functions J i ; Jyi W Œ0; 1 ! R. The Jacobi equations for J and Jy, written in terms of
our parallel frames, read
Conjugate Points 297
l 00
J .t / C Rij k l J i .t /v j v k D 0;
l 00
Jy .t / C Rij k l Jyi .t /v j v k D 0:
Proposition 10.10 shows that Dt J.0/ D w, which we can write as w D w l El .0/.

It also follows that D y t Jy.0/ D w y D A.w/ D w l Eyl .0/, so the functions J l and Jyl
satisfy the same system of differential
0 equations
0 with the same initial conditions
J l .0/ D Jyl .0/ D 0 and J l .0/ D Jyl .0/ D w l . Uniqueness of ODE solutions
implies J l .t /D Jyl .t / for all t , and in particular J l .1/ D Jyl .1/. Because the frames
.Ei / and Eyi are orthonormal, we have
n
X n
i 2 X i 2 ˇ ˇ2
jJ.1/j2g D J .1/ D Jy .1/ D ˇJy.1/ˇg ;
iD1 iD1
thus completing the proof. t

u
Theorem 10.19 (Characterization of Locally Symmetric Spaces). A Riemannian

manifold is a locally symmetric space if and only if its curvature tensor is parallel.
Proof. One direction is taken care of by Problem 7-3. To prove the converse, sup-
pose .M; g/ is a Riemannian manifold with rRm 0. Let p 2 M be arbitrary, and
let U be a geodesic ball centered at p. The linear map A D Id W Tp M ! Tp M
satisfies A gp D gp , and for all w; x; y; z 2 Tp M , we have
.A Rmp /.w; x; y; z/ D Rmp .w; x; y; z/ D Rmp .w; x; y; z/:
It follows that A D Id satisfies the hypotheses of Lemma 10.18 with p D p, y and

thus there is a local isometry ' W U ! M such that '.p/ D p and d'p D Id. Since
a local isometry takes geodesics to geodesics, '.U / is also a geodesic ball centered
at p of the same radius as U , so ' actually maps U to U . If we take the radius of U
small enough, then ' is an isometry from U to itself. t
u
Conjugate Points
Our next application of Jacobi fields is to study the question of when the exponential
map is a local diffeomorphism.
Suppose .M; g/ is a Riemannian or pseudo-Riemannian manifold and p 2 M .
The restricted exponential map expp is defined on an open subset Ep Tp M , and
because it is a smooth map between manifolds of the same dimension, the inverse
function theorem guarantees that it is a local diffeomorphism
in a neighborhood of
each of its regular points (points v 2 Tp M where d expp v is surjective and thus
invertible). To see where this fails, we need to identify the critical points of expp
(the points where its differential is singular). Proposition 5.19(d) guarantees that 0
is a regular point, but it may well happen that it has critical points elsewhere in Ep .
Fig. 10.7: The exponential map of the sphere
Fig. 10.8: Conjugate points
An enlightening example is provided by the sphere Sn .R/ (Fig. 10.7). All

geodesics starting at a given point p meet at the antipodal point, which is at a dis-
tance of R along each geodesic. The exponential map is a diffeomorphism on the
ball BR .0/ Tp Sn .R/, but every point on the boundary of that ball is a critical
point. Moreover, Proposition 10.12 shows that each Jacobi field on Sn .R/ vanishing
at p has its first zero precisely at distance R.
On the other hand, formula (10.7) shows that if U is a normal neighborhood of
p (the image of a star-shaped open set on which expp is a diffeomorphism), then no
Jacobi field that vanishes at p can vanish at any other point in U . We might thus be
led to expect a relationship between zeros of Jacobi fields and critical points of the
exponential map.
Let .M; g/ be a Riemannian or pseudo-Riemannian manifold, W I ! M a
geodesic, and p D .a/, q D .b/ for some a; b 2 I . We say that p and q are
conjugate along if there is a Jacobi field along vanishing at t D a and t D b
but not identically zero (Fig. 10.8). The order (or multiplicity) of conjugacy is the
dimension of the space of Jacobi fields vanishing at a and b. From the existence
and uniqueness theorem for Jacobi fields, there is an n-dimensional space of Jacobi
fields that vanish at a; since tangential Jacobi fields vanish at most at one point, the
Conjugate Points 299
order of conjugacy of two points along can be at most n 1. This bound is sharp:
Proposition 10.12 shows that if is a geodesic joining antipodal points p and q on
Sn .R/, then there is a Jacobi field vanishing at p and q for each parallel normal
vector field along ; thus in that case p and q are conjugate to order exactly n 1.
The most important fact about conjugate points is that they are the images of
critical points of the exponential map, as the following proposition shows.

fold, p 2 M , and v 2 Ep Tp M . Let D v W Œ0; 1 ! M be the geodesic segment
.t / D expp .t v/, and let q D .1/ D expp .v/. Then v is a critical point of expp if
and only if q is conjugate to p along .
Proof. Suppose first that v is a critical point of expp . Then there is a nonzero vector

w 2 Tv .Tp M / such that d expp v .w/ D 0. Because Tp M is a vector space, we
can identify Tv .Tp M / with Tp M as usual and regard w as a vector in Tp M . Let
be the variation of defined by (10.5), and let J.t / D @s .0; t / be its variation
field. We can compute J.1/ as follows:
ˇ
@ ˇˇ
J.1/ D @s .0; 1/ D ˇ expp .v C sw/ D d expp v .w/ D 0:
@s sD0
Thus J is a nontrivial Jacobi field vanishing at t D 0 and t D 1, so q is conjugate to

p along .
Conversely, if q is conjugate to p along , then there is some nontrivial Jacobi
field J along such that J.0/ D 0 and J.1/ D 0. Lemma 10.9 shows that J is the
D Dt J.0/ 2 Tp M , and
variation field of a variation of of the form (10.5) with w
the computation in the preceding paragraph shows that d expp v .w/ D J.1/ D 0.
Thus v is a critical point for expp . t
u
As Proposition 10.2 shows, the “natural” way to specify a unique Jacobi field
is by giving its initial value and initial derivative. However, in Corollary 10.11 and
Proposition 10.20, we had to construct Jacobi fields along a geodesic satisfying
J.0/ D 0 and J.1/ D w for some specific vector w. More generally, one can pose the
two-point boundary problem for Jacobi fields: given v 2 T.a/ M and w 2 T.b/ M ,
find a Jacobi field J along such that J.a/ D v and J.b/ D w. Another interesting
property of conjugate points is that they are the obstructions to solving the two-point
boundary problem, as the next proposition shows.
Proposition 10.21 (The Two-Point Boundary Problem for Jacobi Fields). Sup-
pose .M; g/ is a Riemannian or pseudo-Riemannian manifold, and W Œa; b ! M
is a geodesic segment. The two-point boundary problem for Jacobi fields along is
uniquely solvable for every pair of vectors v 2 T.a/ M and w 2 T.b/ M if and only
if .a/ and .b/ are not conjugate along .

u
The Second Variation Formula
Our next task is to study the question of which geodesic segments are minimizing.
In the remainder of the chapter, because of the complications involved in studying
lengths on pseudo-Riemannian manifolds, we restrict our attention to the Riemann-
ian case.
In our proof that every minimizing curve is a geodesic, we imitated the first-
derivative test of elementary calculus: if a geodesic is minimizing, then the first
derivative of the length functional must vanish for every proper variation of . Now
we imitate the second-derivative test: if is minimizing, the second derivative must
be nonnegative. First, we must compute this second derivative. In keeping with clas-
sical terminology, we call it the second variation of the length functional.
Theorem 10.22 (Second Variation Formula). Suppose .M; g/ is a Riemannian

manifold. Let W Œa; b ! M be a unit-speed geodesic segment, W J Œa; b ! M
a proper variation of , and V its variation field. The second variation of Lg .s / is
given by the following formula:
ˇ Z b
d 2 ˇˇ ˇ ˇ
L . / D ˇDt V ? ˇ2 Rm V ? ; 0 ; 0 ; V ? dt; (10.19)
2 ˇ
ds sD0
g s
a
where V ? is the normal component of V .
Proof. As usual, write T D @t and S D @s , and let .a0 ; : : : ; ak / be an admissible

partition for . We begin, as we did when computing the first variation formula,
by restricting to a rectangle J Œai1 ; ai where is smooth. From (6.3) we have,
for every s, Z ai
d hDt S; T i
Lg s jŒai 1 ;ai D 1=2
dt:
ds ai 1 hT; T i
Differentiating again with respect to s, and using the symmetry lemma and Propo-
sition 7.5, we obtain
d2
2
Lg s jŒai 1 ;ai
ds Z
ai
hDs Dt S; T i hDt S; Ds T i 1 hDt S; T i2hDs T; T i
D C dt
ai 1 hT; T i1=2 hT; T i1=2 2 hT; T i3=2
Z ai
hDt Ds S C R.S; T /S; T i hDt S; Dt S i hDt S; T i2
D C dt:
ai 1 jT j jT j jT j3
Now restrict to s D 0, where jT j D 1:

ˇ Z ai
d 2 ˇˇ
2 ˇ L g . s j Œa ;a
i 1 i / D hDt Ds S; T i Rm.S; T; T; S /
ds sD0 ai 1
(10.20)
ˇˇ
C jDt S j2 hDt S; T i2 dt ˇ :
sD0
The Second Variation Formula 301
Because Dt T D Dt 0 D 0 when s D 0, the first term in (10.20) can be integrated

as follows:
Z ai Z ai ˇtDai
d ˇ
hDt Ds S; T idt D hDs S; T idt D hDs S; T iˇˇ : (10.21)
ai 1 ai 1 dt tDai 1
Notice that S.s; t / D 0 for all s at the endpoints t D a0 D a and t D ak D b because

is a proper variation, so Ds S D 0 there. Moreover, along the boundaries ft D ai g
of the smooth regions, Ds S D Ds .@s / depends only on the values of when
t D ai , and it is smooth up to the line ft D ai g from both sides; therefore Ds S is
continuous for all .s; t /. Thus when we insert (10.21) into (10.20) and sum over i ,
the boundary contributions from the first term all cancel, and we get
ˇ Z b ˇ
d 2 ˇˇ ˇ
2 2 ˇ
L . / D jD S j hD S; T i Rm.S; T; T; S / dt ˇ
ds 2 ˇsD0
g s t t
a ˇ
sD0 (10.22)
Z b

D jDt V j2 hDt V; 0 i2 Rm.V; 0 ; 0 ; V / dt:
a
Every vector field V along can be written uniquely as V D V > C V ? , where

V is tangential and V ? is normal. Explicitly,
>
V > D hV; 0 i 0 I V ? D V V >:
Because Dt 0 D 0, it follows that

Dt V > D hDt V; 0 i 0 D .Dt V /> I Dt V ? D .Dt V /? :
Therefore,
jDt V j2 D j.Dt V /> j2 C j.Dt V /? j2 D hDt V; 0 i2 C jDt V ? j2 :

Also, the fact that Rm 0 ; 0 ; ; D Rm ; ; 0 ; 0 D 0 implies
Rm.V; 0 ; 0 ; V / D Rm.V ? ; 0 ; 0 ; V ? /:
Substituting these relations into (10.22) gives (10.19). t

u
It should come as no surprise that the second variation depends only on the nor-
mal component of V , because the tangential component of V contributes only to a
reparametrization of , and length is independent of parametrization. For this rea-
son, we will generally restrict our attention to variations of the following type: if
is an admissible curve, a variation of is called a normal variation if its variation
field is a normal vector field along .
Given a geodesic segment W Œa; b ! M , we define a symmetric bilinear form
I , called the index form of , on the space of normal vector fields along by
Z b
I.V; W / D hDt V ; Dt W i Rm.V; 0 ; 0 ; W / dt: (10.23)
a
You should think of I.V; W / as a sort of “Hessian” or second derivative of the length
functional. Because every proper normal vector field along is the variation field
of some proper normal variation, the preceding theorem can be rephrased in terms
of the index form in the following way.
Corollary 10.23. Suppose .M; g/ is a Riemannian manifold. Let W Œa; b ! M be

a unit-speed geodesic, a proper normal variation of , and V its variation field.
The second variation of Lg .s / is I.V; V /. If is minimizing, then I.V; V / 0 for
every proper normal vector field along . t
u
The next proposition gives another expression for I , which makes the role of the
Jacobi equation more evident.
Proposition 10.24. Let .M; g/ be a Riemannian manifold and let W Œa; b ! M be

a geodesic segment. For every pair of piecewise smooth normal vector fields V; W
along ,
Z b˝ ˛
I.V; W / D Dt2 V C R.V; 0 / 0 ; W dt
a
ˇtDb k1
X
ˇ
C hDt V; W iˇ hi Dt V ; W .ai /i ; (10.24)
tDa
iD1
where .a0 ; : : : ; ak / is an admissible partition for V and W , and i Dt V is the jump

in Dt V at t D ai .
Proof. On every subinterval Œai1 ; ai where V and W are smooth,
d ˝ ˛
hDt V ; W i D Dt2 V ; W C hDt V ; Dt W i :
dt
Thus, by the fundamental theorem of calculus,
Z ai Z ai ˇai
˝ 2 ˛ ˇ
hDt V ; Dt W i dt D Dt V ; W dt C hDt V ; W iˇˇ :
ai 1 ai 1 ai 1
Summing over i , and noting that W is continuous at t D ai for i D 1; : : : ; k 1, we

get (10.24). t
u
Corollary 10.25. If is a geodesic segment and V is a proper normal piecewise

smooth vector field along , then I.V; W / D 0 for every proper normal piecewise
smooth vector field W along if and only if V is a Jacobi field.

u
Geodesics Do Not Minimize Past Conjugate Points
We can use the second variation formula to prove another extremely important fact
about conjugate points: no geodesic is minimizing past its first conjugate point. The
geometric intuition is as follows. Suppose W Œa; c ! M is a minimizing geodesic
segment, and .b/ is conjugate to .a/ along for some a < b < c. If J is a Jacobi
field along that vanishes at t D a and t D b, then there is a variation of through
geodesics, all of which start at .a/. Since J.b/ D 0, we can expect them to end
“almost” at .b/. If they really did all end at .b/, we could construct a broken
geodesic by following some s from .a/ to .b/ and then following from .b/
to .c/, which would have the same length and thus would also be a minimizing
curve. But this is impossible: as the proof of Theorem 6.4 shows, a broken geodesic
can always be shortened by rounding the corner.
The problem with this heuristic argument is that there is no guarantee that we
can construct a variation through geodesics that actually end at .b/. The proof of
the following theorem is based on an “infinitesimal” version of rounding the corner
to obtain a shorter curve.
Given a geodesic segment W Œa; c ! M , we say that has a conjugate point
if there is some b 2 .a; c such that .b/ is conjugate to .a/ along , and has an
interior conjugate point if there is such a b 2 .a; c/.
Theorem 10.26. Let .M; g/ be a Riemannian manifold and p; q 2 M . If is a unit-

speed geodesic segment from p to q that has an interior conjugate point, then there
exists a proper normal vector field X along such that I.X; X / < 0. Therefore,
is not minimizing.
Proof. Suppose W Œa; c ! M is a unit-speed geodesic segment, and .b/ is con-

jugate to .a/ along for some a < b < c. This means that there is a nontrivial
normal Jacobi field J along that vanishes at t D a and t D b. Define a vector field
V along all of by (
J.t /; t 2 Œa; bI
V .t / D
0; t 2 Œb; c:
This is a proper, normal, piecewise smooth vector field along .
Let W be a smooth proper normal vector field along such that W .b/ is equal to
the jump Dt V at t D b (Fig. 10.9). Such a vector field is easily constructed with
the help of an orthonormal frame along and a bump function. Note that Dt V D
Dt J.b/ is not zero, because otherwise J would be a Jacobi field satisfying J.b/ D
Dt J.b/ D 0, and thus would be identically zero.
For small positive ", let X" D V C "W . Then
I.X" ; X" / D I.V C "W; V C "W /

D I.V; V / C 2"I.V; W / C "2 I.W; W /:
Since V satisfies the Jacobi equation on each subinterval Œa; b and Œb; c, and
V .b/ D 0, (10.24) gives
Fig. 10.9: Constructing a vector field X with I.X; X / < 0
I.V; V / D hDt V ; V .b/i D 0:
Similarly,
I.V; W / D hDt V ; W .b/i D jW .b/j2 :
Thus
I.X" ; X" / D 2" jW .b/j2 C "2 I.W; W /:
If we choose " small enough, this is strictly negative. t
u
There is a partial converse to the preceding theorem, which says that a geodesic
without conjugate points has the shortest length among all nearby curves in any
proper variation. Before we prove it, we need the following technical lemma.
Lemma 10.27. Let W Œa; b ! M be a geodesic segment, and suppose J1 and
J2 are Jacobi fields along . Then hDt J1 .t /; J2 .t /i hJ1 .t /; Dt J2 .t /i is constant
along .
Proof. Let f .t / D hDt J1 .t /; J2 .t /i hJ1 .t /; Dt J2 .t /i. Using the Jacobi equation,
we compute
f 0 .t / D hDt2 J1 .t /; J2 .t /i C hDt J1 .t /; Dt J2 .t /i
hDt J1 .t /; Dt J2 .t /i hJ1 .t /; Dt2 J2 .t /i

D Rm J1 .t /; 0 .t /; 0 .t /; J2 .t / C Rm J2 .t /; 0 .t /; 0 .t /; J1 .t / D 0;
where the last equality follows from the symmetries of the curvature tensor. t
u
Theorem 10.28. Let .M; g/ be a Riemannian manifold. Suppose W Œa; b ! M is
a unit-speed geodesic segment without interior conjugate points. If V is any proper
normal piecewise smooth vector field along , then I.V; V / 0, with equality if and
only if V is a Jacobi field. In particular, if .b/ is not conjugate to .a/ along ,
then I.V; V / > 0.
Proof. To simplify the notation, we can assume (after replacing t by t a) that
a D 0. Let p D .0/, and let .w1 ; : : : ; wn / be an orthonormal basis for Tp M , chosen
so that w1 D 0 .0/. For each ˛ D 2; : : : ; n, let J˛ be the unique normal Jacobi field
along satisfying J˛ .0/ D 0 and Dt J˛ .0/ D w˛ .
Our assumption that has no interior conjugate points guarantees that no linear
combination of the J˛ .t /’s can vanish for any t 2 .0; b/, and thus .J˛ .t // forms a
basis for the orthogonal complement of 0 .t / in T.t/ M for each such t . Thus, given
V as in the statement of the theorem, for t 2 .0; b/ we can write
V .t / D v ˛ .t /J˛ .t / (10.25)
for some piecewise smooth functions v ˛ W .0; b/ ! M . (Here and in the remainder
of this proof, the summation convention is in effect, with Greek indices running
from 2 to n.)
˛
In each function v has a piecewise smooth extension to Œ0; b. To see why,
ifact,
let x be the normal coordinates centered at p determined by the basis .wi /. For
sufficiently small t > 0, we can express J˛ .t / and V .t / in normal coordinates as
ˇ
@ ˇˇ
J˛ .t / D t ; ˛ D 2; : : : ; n;
@x˛ ˇ.t/
ˇ
@ ˇˇ
V .t / D v ˛ .t /J˛ .t / D t v ˛ .t / :
@x ˇ
˛ .t/
(The formula for J˛ .t / follows from Prop. 10.10.) Because V is smooth on Œ0; ı/
for some ı > 0 and V .0/ D 0, it follows from Taylor’s theorem that the components
of V .t /=t extend smoothly to Œ0; ı/, which shows that v ˛ is smooth there. Because
V .b/ D 0, it follows similarly that v ˛ extends smoothly to t D b as well. (If J˛ .b/ D
0, the argument is the same as for t D 0, while if not, it is even easier.)
Let .a0 ; : : : ; ak / be an admissible partition for V . On each subinterval .ai1 ; ai /
where V is smooth, define vector fields X and Y along by
X D v ˛ Dt J˛ ; Y D vP ˛ J˛ :
Then Dt V D X C Y on each such interval, and the fact that V is piecewise smooth
implies that Dt V , X , and Y extend smoothly to Œai1 ; ai for each i , with one-sided
derivatives at the endpoints.
To compute I.V; V /, we will use the following identity, which holds on each
subinterval Œai1 ; ai :
d
jDt V j2 Rm.V; 0 ; 0 ; V / D hV; X i C jY j2 : (10.26)
dt
Granting this for now, we use the fundamental theorem of calculus to compute
k Z
X ai
I.V; V / D jDt V j2 Rm.V; 0 ; 0 ; V / dt
iD1 ai 1
k
X ˇtDai Z b
ˇ
D hV; X iˇ C jY j2 dt;
tDai 1 0
iD1
where the boundary terms are to be interpreted as limits from above and below.
Because X and V are both continuous on Œ0; b, the boundary terms at t Da1 ; : : : ; ak1
all cancel, and because V .0/ D V .b/ D 0, the boundary terms at t D 0 and t D b
Rb
are both zero. It follows that I.V; V / D 0 jY j2 dt 0. If I.V; V / D 0, then Y is
identically zero on .0; b/. Since the J˛ ’s are linearly independent there, this implies
that vP ˛ 0 for each ˛, so each v ˛ is constant. Thus V is a linear combination of
Jacobi fields with constant coefficients, so it is a Jacobi field.
It remains only to prove (10.26). Note that
d
hV; X i D hDt V; X i C hV; Dt X i D hX C Y; X i C hV; Dt X i: (10.27)
dt
The Jacobi equation gives
Dt X D vP ˛ Dt J˛ C v ˛ Dt2 J˛ D vP ˛ Dt J˛ v ˛ R.J˛ ; 0 / 0 D vP ˛ Dt J˛ R.V; 0 / 0 :
Therefore,
hDt X; V i D hvP ˛ Dt J˛ ; v ˇ Jˇ i Rm.V; 0 ; 0 ; V /: (10.28)
Because hDt J˛ ; Jˇ i hJ˛ ; Dt Jˇ i D 0 at t D 0, it follows from Lemma 10.27 that
hDt J˛ ; Jˇ i D hJ˛ ; Dt Jˇ i all along . Thus we can simplify the first term in (10.28)
as follows:
hvP ˛ Dt J˛ ; v ˇ Jˇ i D vP ˛ v ˇ hDt J˛ ; Jˇ i D vP ˛ v ˇ hJ˛ ; Dt Jˇ i

D hvP ˛ J˛ ; v ˇ Dt Jˇ i D hY; X i:
Inserting this into (10.28), and then inserting the latter into (10.27) yields
d
hV; X i D hX C Y; X i C hY; X i Rm.V; 0 ; 0 ; V /
dt
D jX C Y j2 jY j2 Rm.V; 0 ; 0 ; V /;

u
The next corollary summarizes the results of Theorems 10.26 and 10.28.
Corollary 10.29. Let .M; g/ be a Riemannian manifold, and let W Œa; b ! M be

a unit-speed geodesic segment.
(a) If has an interior conjugate point, then it is not minimizing.
Cut Points 307
Fig. 10.10: Geodesics on the cylinder
(b) If .a/ and .b/ are conjugate but has no interior conjugate points, then
for every proper normal variation of , the curve s is strictly longer
than for all sufficiently small nonzero s unless the variation field of is
a Jacobi field.
(c) If has no conjugate points, then for every proper normal variation of ,
the curve s is strictly longer than for all sufficiently small nonzero s. u
t
There is a far-reaching quantitative generalization of Theorems 10.26 and 10.28,

called the Morse index theorem, which we do not treat here. The index of a geodesic
segment is defined to be the maximum dimension of a linear space of proper normal
vector fields along the segment on which I is negative definite. Roughly speaking,
the index is the number of independent directions in which can be deformed to
decrease its length. (Analogously, the index of a critical point of a function on Rn
is defined as the number of negative eigenvalues of its Hessian.) The Morse index
theorem says that the index of every geodesic segment is finite, and is equal to the
number of its interior conjugate points counted with multiplicity. (Proofs can be
found in [Mil63, CE08, dC92].)
Cut Points
Theorem 10.26 showed that once a geodesic passes its first conjugate point, it ceases
to be minimizing. The converse, however, is not true: a geodesic can cease to be
minimizing without reaching a conjugate point. For example, on the cylinder S1 R
with the product metric, there are no conjugate points along any geodesic; but no
geodesic segment that wraps more than halfway around the cylinder is minimizing
(Fig. 10.10).
Therefore it is useful to make the following definitions. Suppose .M; g/ is a
complete, connected Riemannian manifold, p is a point of M , and v 2 Tp M . Define
the cut time of .p; v/ by
˚
tcut .p; v/ D sup b > 0 W the restriction of v to Œ0; b is minimizing ;
where v is the maximal geodesic starting at p with initial velocity v. Because v is

minimizing as long as its image stays inside a geodesic ball (Prop. 6.11), tcut .p; v/
is always positive; but it might be C1.
If tcut .p; v/ < 1, the cut point of p along v is the point v tcut .p; v/ 2 M .
The cut locus of p, denoted by Cut.p/, is the set of all q 2 M such that q is the
cut point of p along some geodesic. Because the question whether a geodesic
is minimizing is independent of parametrization, the cut point of p along v is the
same as the cut point along v for every positive constant , so we may as well
restrict attention to unit vectors v. Theorem 10.26 says that the cut point (if it exists)
occurs at or before the first conjugate point along every geodesic.
The determination of the cut locus of a point is typically very difficult; but the
next example gives some special cases in which it is relatively simple.
Example 10.30 (Cut Loci).

(a) If Sn .R/; gV R is a sphere with a round metric, the cut locus of every point
p 2 Sn .R/ is the singleton set containing only the antipodal point p.
(b) On a product space Sn .R/ Rm with the product metric, the cut locus of every
point .p; x/ is the set fpg Rm . The case n D m D 1 is illustrated in Figure
10.10. //
I Exercise 10.31. Verify the claims in the preceding example.
Proposition 10.32 (Properties of Cut Times). Suppose .M; g/ is a complete, con-

nected Riemannian manifold, p 2 M , and v is a unit vector in Tp M . Let c D
tcut .p; v/ 2 .0; 1.
(a) If 0 < b < c, then v jŒ0;b has no conjugate points and is the unique unit-speed
minimizing curve between its endpoints.
(b) If c < 1, then v jŒ0;c is minimizing, and one or both of the following condi-
tions are true:
v .c/ is conjugate to p along v .
There are two or more unit-speed minimizing geodesics from p to v .c/.
Proof. Suppose first that 0 < b < c. By definition of tcut .p; v/, there is a time b 0
such that b < b 0 < c and v jŒ0;b 0 is minimizing. Then v .t / cannot be conjugate
to p along v for any 0 < t b (Thm. 10.26), and v jŒ0;b is minimizing because
a shorter admissible curve from p to v .b/ could be combined with v jŒb;b 0 to
produce a shorter admissible curve from p to v .b 0 /, contradicting the fact that
v jŒ0;b 0 is minimizing.
To see that v jŒ0;b is the unique unit-speed minimizing curve between its end-
points, suppose for the sake of contradiction that W Œ0; b ! M is another. Note
that 0 .b/ ¤ v0 .b/, since otherwise and v would agree on Œ0; b by uniqueness of
geodesics. Define a new unit-speed admissible curve z W Œ0; b 0 ! M that is equal to
.t / for t 2 Œ0; b and equal to v .t / for t 2 Œb; b 0 . Then z has length b 0 , so it is also
a minimizing curve from p to v .b 0 /; but it is not smooth at t D b, contradicting the
fact that minimizing curves are smooth geodesics. This completes the proof of (a).
Cut Points 309
Now suppose c < 1. By definition of tcut .p; v/, there is a sequence of times
bi % c such that the restriction of v to Œ0; bi is minimizing. By continuity of the
distance function, therefore,
dg .p; v .c// D lim dg .p; v .bi // D lim bi D c;

i!1 i!1
which shows that v is minimizing on Œ0; c. To prove that one of the options in
(b) must hold, assume that v .c/ is not conjugate to p along v . We will prove the
existence of a second unit-speed minimizing geodesic from p to v .c/.
Let .bi / be a sequence of real numbers such that bi & c. By definition of cut time,
v jŒ0;bi is not minimizing, so for each i there is a unit-speed minimizing geodesic
i W Œ0; ai ! M such that i .0/ D p, i .ai / D v .bi /, and ai < bi . Set wi D i0 .0/ 2
Tp M , so each wi is a unit vector. By compactness of the unit sphere, after passing
to a subsequence we may assume that wi converges to some unit vector w. Since the
ai ’s are all positive and bounded above by b1 , by passing to a further subsequence,
we may also assume that ai converges to some number a. Then by continuity of the
exponential map, i .ai / D expp .ai wi / converges to expp .aw/. But we also know
that i .ai / D v .bi /, which converges to v .c/, so expp .aw/ D v .c/. Moreover,
by continuity of the distance function,
c D dg .p; v .c// D lim dg .p; i .ai // D lim ai D a:

i!1 i!1
Thus W Œ0; c ! M given by .t / D expp .t w/ is also a unit-speed minimizing

geodesic from p to v .c/. We need to show that it is not equal to v .
The assumption that v .c/ is not conjugate to p along v implies that cv is a
regular point of expp (Prop. 10.20), so expp is injective in some neighborhood V
of cv. Note that expp .ai wi / D expp .bi v/ for each i , while ai wi ¤ bi v, since wi
and v are unit vectors and ai < bi . Since bi v converges to cv, we conclude that
bi v 2 V for sufficiently large i , and thus by injectivity ai wi … V for these values of
i . Therefore cw D limi!1 ai wi ¤ cv, which implies w ¤ v and thus ¤ v , as
claimed. t
u
Next we examine how the cut time varies as the initial point and initial velocity
of the geodesic vary. Recall that the unit tangent bundle of a Riemannian manifold
.M; g/ is the subset U TM D f.p; v/ 2 TM W jvjg D 1g TM . In the next theorem,
we interpret continuity of a function into .0; 1 using the usual definition of infinite
limits as in ordinary calculus.
Theorem 10.33. Suppose .M; g/ is a complete, connected Riemannian manifold.

The function tcut W U TM ! .0; 1 is continuous.
Proof. Let .p; v/ 2 U TM be arbitrary, and let .pi ; vi / be any sequence in U TM

converging to .p; v/. Put ci D tcut .pi ; vi /, and
b D lim inf ci ; c D lim sup ci :

i!1 i!1
We will show that c tcut .p; v/ b, which implies ci ! tcut .p; v/.
To show that c tcut .p; v/, suppose first that c < 1. By passing to a subse-
quence, we may assume that ci is finite for each i and ci ! c. Proposition 10.32
shows that vi is minimizing on Œ0; ci . By continuity of the exponential map,
exp.pi ; ci vi / ! exp.p; cv/ as i ! 1, and therefore by continuity of the distance
function we have
dg .p; exp.p; cv// D lim dg .pi ; exp.pi ; ci vi // D lim ci D c:

i!1 i!1
This shows that v is minimizing on Œ0; c, and therefore tcut .p; v/ c, as claimed.
Now suppose c D 1. Again, by passing to a subsequence, we may assume
ci ! 1. It follows that for every positive number c0 , the geodesic vi is mini-
mizing on Œ0; c0 for i sufficiently large, and it follows by continuity as above that
v is minimizing on Œ0; c0 . Since c0 was arbitrary, this means that tcut .p; v/ D 1.
Next we show that tcut .p; v/ b. If b D 1, there is nothing to prove, so assume
b < 1. Again by passing to a subsequence, we may assume that ci is finite for each
i and ci ! b. By virtue of Proposition 10.32, either there are infinitely many indices
i for which vi .ci / is conjugate to pi along vi , or there are infinitely many i for
which there exists a second minimizing unit-speed geodesic i from pi to vi .ci /.
In the first case, because conjugate points are critical values of the restricted
exponential map, which can be detected in coordinates by the vanishing of a deter-
minant of a matrix of first derivatives, it follows by continuity that v .b/ is also a
critical value, and thus v .b/ is conjugate to p along v . Then Theorem 10.26 shows
that tcut .p; v/ b.
In the second case, let wi be the unit vector in Tpi M such that i D wi . Because
the components of wi with respect to a local orthonormal frame lie in Sn1 , by
passing to a subsequence we may assume .pi :wi / ! .p; w/. If v .b/ is conjugate
to p along v , then tcut .p; v/ b as above, so we may assume that v .b/ is not a
conjugate point. This means that bv is a regular point of the restricted exponential
map expp . Since the set of such regular points is an open subset of TM , there is
some " > 0 such that exppi is injective on the "-neighborhood of ci wi for all i
sufficiently large. This implies that jci wi ci vi jg " for all such i , and therefore
the limits bw and bv are distinct. Thus w jŒ0;b is another minimizing geodesic from
p to v .b/, which by Proposition 10.32 implies that tcut .p; v/ b. t
u
Given p 2 M , we define two subsets of Tp M as follows: the tangent cut locus
of p is the set ˚
TCL.p/ D v 2 Tp M W jvj D tcut .p; v=jvj/ ;
and the injectivity domain of p is
˚
ID.p/ D v 2 Tp M W jvj < tcut .p; v=jvj/ :

It is immediate that TCL.p/ D @ ID.p/, and Cut.p/ D expp TCL.p/ . Further
properties of Cut.p/ and ID.p/ are described in the following theorem.
Cut Points 311
Theorem 10.34. Let .M; g/ be a complete, connected Riemannian manifold and

p 2 M.
(a) The cut locus of p is a closed subset of M of measure zero.
(b) The restriction of expp to ID.p/ is surjective.
(c) The restriction of expp to ID.p/ is a diffeomorphism onto M X Cut.p/.
Proof. To prove that the cut locus is closed, suppose .qi / is a sequence of points
in Cut.p/ converging to some q 2 M . Write qi D expp .tcut .p; vi /vi / for unit
vectors vi . By compactness of the unit sphere, we may assume after passing to
a subsequence that vi converges to some unit vector v, and by Theorem 10.33,
tcut .p; v/ D limi!1 tcut .p; vi /. Because convergent sequences in a metric space are
bounded, the sequence .tcut .p; vi // is bounded, and therefore tcut .p; v/ < 1. By
continuity of the exponential map, therefore, q must be equal to expp .tcut .p; v//,
which shows that q 2 Cut.p/, and thus Cut.p/ is closed.
1To see n1that Cut.p/
has measure zero, note first that in any polar coordinates

;:::;
; r on Tp M , the set TCL.p/ can be expressed locally as the graph of
the continuous function r D tcut p; .
1 ; : : : ;
n1 / , using the fact that .
1 ; : : : ;
n1 /
form smooth local coordinates for the unit sphere in Tp M . Since graphs of continu-
ous functions have measure zero (see, for example, [LeeSM, Prop. 6.3]), it follows
that TCL.p/ is a union of finitely many sets of measure zero and thus has measure
zero in Tp M ; and because smooth maps take sets of measure zero to sets of measure
zero (see [LeeSM, Prop. 6.4]), Cut.p/ D expp TCL.p/ has measure zero in M .
This proves (a).
Part (b) follows from the fact that every point of M can be connected to p by a
minimizing geodesic.
To prove (c), note that it follows easily from the definitions
that expp ID.p/ D M X Cut.p/. Also, the definition of ID.p/ guarantees that no

point in expp ID.p/ can be a cut point of p, and thus no such point can be a
conjugate point either. The absence of cut points implies that expp is injective on
ID.p/, and the absence of conjugate points implies that it is a local diffeomorphism
there. Together these two facts imply that it is a diffeomorphism onto its image. u t
The preceding theorem leads to the following intriguing topological result about
compact manifolds.
Corollary 10.35. Every compact, connected, smooth n-manifold is homeomorphic
x n by an equivalence relation that identifies only points on
to a quotient space of B
the boundary.
Proof. Let M be a compact, connected, smooth n-manifold, let p be any point of
M , and let g be any Riemannian metric on M . Because a compact metric space has
finite diameter, every unit vector in Tp M has a finite cut time, no greater than the
diameter of M . Let Bx1 .0/ Tp M denote the closed unit ball in Tp M , and define a
map f W Bx1 .0/ ! ID.p/ by
˚

v

tcut p; v; v ¤ 0;
f .v/ D jvjg
0 v D 0:
It follows from Theorem 10.33 that f is continuous, and it is easily seen to be

bijective, so it is a homeomorphism by the closed map lemma (Lemma A.4).
Since every orthonormal basis for Tp M yields a homeomorphism of Bx1 .0/ with
x n , it follows that ID.p/ is homeomorphic to B
B x n and the homeomorphism takes
n1
TCL.p/ D @ ID.p/ to S .
By Theorem 10.34, expp restricts to a surjective map from ID.p/ to M , and it
is a quotient map by the closed map lemma. It follows that M is homeomorphic to
the quotient of ID.p/ by the equivalence relation v
w if and only if expp .v/ D
expp .w/. Since expp is injective on ID.p/ and the images of ID.p/ and @ ID.p/ D
TCL.p/ are disjoint, the equivalence relation identifies only points on the boundary
of ID.p/. t
u
Recall from Chapter 6 that the injectivity radius of M at p, denoted by inj.p/,
is the supremum of all positive numbers a such that expp is a diffeomorphism from
Ba .0/ Tp M to its image. The injectivity radius is closely related to the cut locus,
as the next proposition shows.
Proposition 10.36. Let .M; g/ be a complete, connected Riemannian manifold. For
each p 2 M , the injectivity radius at p is the distance from p to its cut locus if the
cut locus is nonempty, and infinite otherwise.
Proof. Given p 2 M , let d denote the distance from p to its cut locus, with the
convention that d D 1 if the cut locus is empty. Let a 2 .0; 1 be arbitrary, and
let Ba Tp M denote the set of vectors v 2 Tp M with jvjg < a (so Ba D Tp M if
a D 1). We will show that the restriction of expp to Ba is a diffeomorphism onto
its image if and only if a d , from which the result follows.
First suppose a d . By definition of d , no point of the form expp .v/ with v 2 Ba
can be a cut point of p, so Ba ID.p/. It follows from Theorem 10.34(c) that expp
is a diffeomorphism from Ba onto its image.
On the other hand, if a > d , then p has a cut point q whose distance from p
is less than a. It follows from the definition of cut points that the radial geodesic
from p to q is not minimizing past q, so Proposition 6.11 shows that there is no
geodesic ball of radius greater than dg .p; q/. In particular, the restriction of expp to
Ba cannot be a diffeomorphism onto its image. t
u
Proposition 10.37. Let .M; g/ be a complete, connected Riemannian manifold. The
function inj W M ! .0; 1 is continuous.
Proof. Let p 2 M be arbitrary. Proposition 10.32(b) shows that for each point q 2
Cut.p/, there is a minimizing unit-speed geodesic v from p to q whose length is
tcut .p; v/, and therefore the distance from p to Cut.p/ is the infimum of the cut
times of unit-speed geodesics starting at p. By the previous proposition, therefore,
˚
inj.p/ D inf tcut .p; v/ W v 2 Tp M with jvjg D 1 :
Suppose .pi / is a sequence in M converging to a point p 2 M . As in the proof of

Theorem 10.33, we will prove that inj.pi / ! inj.p/ by showing that c inj.p/ b,
where
Problems 313
b D lim inf inj.pi /; c D lim sup inj.pi /:

i!1 i!1
First we show that inj.p/ b. By passing to a subsequence, we may assume

inj.pi / ! b. By compactness of the unit sphere, for each i there is a unit vector vi 2
Tpi M such that inj.pi / D tcut .pi ; vi /, and after passing to a further subsequence,
we may assume .pi ; vi / ! .p; v/ for some v 2 Tp M . By continuity of tcut , we have
tcut .p; v/ D limi tcut .pi ; vi / D b, so inj.p/ b.
Next we show that inj.p/ c. Once again, by passing to a subsequence of
the original sequence .pi /, we may assume inj.pi / ! c. Suppose for the sake of
contradiction that inj.p/ < c, and choose c0 such that inj.p/ < c0 < c. Let w be a
unit vector in Tp M such that tcut .p; w/ D inj.p/. We can choose some sequence of
unit vectors wi 2 Tpi M such that .pi ; wi / ! .p; w/, so tcut .pi ; wi / ! tcut .p; w/ D
inj.p/. For i sufficiently large, this implies tcut .pi ; wi / < c0 < c, contradicting the
facts that tcut .pi ; wi / inj.pi / and inj.pi / ! c. t
u
Problems
10-1. Suppose .M; g/ is a Riemannian manifold and p 2 M . Show that the

second-order Taylor series of g in normal coordinates centered at p is
1X
gij .x/ D ıij Riklj .p/x k x l C O jxj3 :
3
k;l
[Hint: Let .t / D .t v 1 ; : : : ; t v n / be a radial geodesic starting at p, let J.t / D

t w i @i j.t/ be a Jacobi field along , and compute the first four t -derivatives
of jJ.t /j2 at t D 0 in two ways.]
10-2. Suppose .M; g/ is a Riemannian manifold and p 2 M . Let S˘ M be the
2-dimensional submanifold obtained by applying the exponential map to a
plane ˘ Tp M , as on page 250. For sufficiently small r, let A.r/ be the
area of the geodesic disk of radius r about p in S˘ with its induced metric.
Using the results of Problems 10-1 and 8-21, find the Taylor series of A.r/
to fourth order in r. Then use this result to express sec.˘ / in terms of a
limit involving the difference r 2 A.r/.
10-3. Extend the result of Proposition 10.12 by finding a basis for the space of
all Jacobi fields along a geodesic in the constant-curvature case, not just the
normal ones that vanish at 0.
10-4. Prove
nthat the volume of the round sphere Sn .R/ is given for n 0 by
n
Vol S .R/ D ˛n R , where
‚22kC1
k kŠ
; n D 2k; k 2 Z;
.2k/Š
˛n D
2 kC1
; n D 2k C 1; k 2 Z:
kŠ
(The volume of a compact 0-manifold is just the number of points.)
(a) First show that it suffices to prove the volume formula for R D 1.
(b) Use Corollary 10.17 to prove the recurrence relation

Vol SnC1 D n Vol Sn ;
where Z
n D .sin r/n dr:
0
(c) By differentiating the function .sin r/n cos r, prove that

n
nC1 D n1 ;
nC1
and use this to prove that
2
n n1 D :
n
(d) Prove the result by induction on k.
(Used on p. 342.)

10-5. For r > 0, let Br .0/ denote the ball of radius r in Euclidean space Rn ; gx ,

n 1. Prove that Vol Br .0/ D n1 ˛n1 r n , with ˛n1 as in Problem 10-4.
10-6. Let p be a point in a Riemannian n-manifold .M; g/. Use the results of
Problems 10-1 and 8-21 to show that as r & 0,

1 S.p/r 2
Vol.Br .p// D ˛n1 r n 1 C O r3 ;
n 6.n C 2/
where S.p/ is the scalar curvature of g at p and ˛n1 is as in Problem 10-4.

10-7. Suppose .M; g/ is a Riemannian manifold with nonpositive sectional cur-
vature. Prove that no point of M has conjugate points along any geodesic.
[Hint: Consider derivatives of jJ.t /j2 when J is a Jacobi field.] (Used on
p. 333.)
10-8. Prove Proposition 10.21 (solvability of the two-point boundary problem).
10-9. Suppose .M; g/ is a Riemannian manifold and M1 ; M2 M are embedded
submanifolds. Let W Œa; b ! M be a unit-speed geodesic segment that
meets M1 orthogonally at t D a and meets M2 orthogonally at t D b, and
let W K Œa; b ! M be a normal variation of such that .s; a/ 2 M1
Problems 315
and .s; b/ 2 M2 for all s. Prove the following generalization of the second
variation formula:
ˇ Z b
d 2 ˇˇ 2
0 0

L . / D jD V j Rm V; ; ; V dt
ds 2 ˇsD0
g s t
a
˝ ˛ ˝ ˛
C II2 .V .b/; V .b//; 0 .b/ II1 .V .a/; V .a//; 0 .a/ ;
where V is the variation field of , and IIi is the second fundamental form
of Mi for i D 1; 2. (Used on p. 365.)
10-10. Prove the following theorem of Theodore Frankel [Fra61], generalizing the
well-known fact that any two great circles on S2 must intersect: Suppose
.M; g/ is a complete, connected Riemannian manifold with positive sec-
tional curvature. If M1 ; M2 M are compact, totally geodesic subman-
ifolds such that dim M1 C dim M2 dim M , then M1 \ M2 ¤ ¿. [Hint:
Assuming that the intersection is empty, show that there exist a shortest
geodesic segment connecting M1 and M2 and a parallel vector field along
that is tangent to M1 and M2 at the endpoints; then apply the second
variation formula of Problem 10-9 to derive a contradiction.]
10-11. Prove Corollary 10.25 (I.V; W / D 0 for all W if and only if V is a Jacobi
field). [Hint: Adapt the proof of Theorem 6.4.]
10-12. Let .M; g/ be a Riemannian manifold. Suppose W Œa; b ! M is a unit-
speed geodesic segment with no interior conjugate points, J is a normal
Jacobi field along , and V is any other piecewise smooth normal vector
field along such that V .a/ D J.a/ and V .b/ D J.b/.
(a) Show that I.V; V / I.J; J /.
(b) Now assume in addition that .b/ is not conjugate to .a/ along .
Show that I.V; V / D I.J; J / if and only if V D J .
10-13. Suppose .M; g/ is a Riemannian manifold and X 2 X.M / is a Killing vec-
tor field (see Problems 5-22 and 6-24). Show that if W Œa; b ! M is any
geodesic segment, then X restricts to a Jacobi field along .
10-14. Suppose P is an embedded submanifold of a Riemannian manifold M , and
W I ! M is a geodesic that meets P orthogonally at t D a for some a 2 I .
A Jacobi field J along is said to be transverse to P if its restriction to each
compact subinterval of I containing a is the variation field of a variation of
through geodesics that all meet P orthogonally at t D a.
(a) Prove that the tangential Jacobi field J.t / D .t a/ 0 .t / along is
transverse to P .
(b) Prove that a normal Jacobi field J along is transverse to P if and
only if
Dt J.a/ D W 0 .a/ .J.a//;
where W 0 .a/ is the Weingarten map of P in the direction of 0 .a/.
(c) When M has dimension n, prove that the set of transverse Jacobi fields
along is an n-dimensional linear subspace of J. /, and the set of
transverse normal Jacobi fields is an .n 1/-dimensional subspace of
that.
(Used on p. 342.)

10-15. Let x 1 ; : : : ; x n / be any semigeodesic coordinates on an open subset U in a
Riemannian n-manifold .M; g/ (see Prop. 6.41 and Examples 6.43–6.46),
and let .t / D x 1 ; : : : ; x n1 ; t be an xn -coordinate curve defined on some
interval I . Prove that for all constants a1 ; : : : ; an1 , the following vector
field along is a normal Jacobi field along that is transverse to each of
the level sets of x n (in the sense defined in Problem 10-14):
n1
X ˇ
i @ ˇˇ
J.t / D a :
@x i ˇ.t/
iD1
10-16. Suppose P is an embedded k-dimensional submanifold of a Riemannian n-

manifold .M; g/, and x 1 ; : : : ; x k ; v 1 ; : : : ; v nk
are
Fermi coordinates for
P on some open subset U0 M . For fixed x i ; v j , let W I ! M be the

curve with coordinate representation .t / D x 1 ; : : : ; x k ; t v 1 ; : : : ; t v nk ;
Proposition 5.26 shows that is a geodesic that meets P orthogonally at
t D 0. Prove that the Jacobi fields along that are transverse to P are
exactly the vector fields of the form
k
X ˇ nk
X ˇ
@ ˇˇ j @ ˇ
ˇ
J.t / D ai ˇ C t b ;
i
@x .t/ @v ˇ.t/
j
iD1 j D1
for arbitrary constants a1 ; : : : ; ak ; b 1 ; : : : ; b nk .

10-17. Suppose P is an embedded submanifold of a Riemannian manifold .M; g/,
and U is a normal neighborhood of P in M . Prove that every tangent vector
to U X P is the value of a transverse Jacobi field: more precisely, for each
q 2 U X P and each w 2 Tq M , there is a g-geodesic segment W Œ0; b ! U
such that .0/ 2 P , 0 .0/ ? T.0/ P , and .b/ D q, and a Jacobi field J
along that is transverse to P and satisfies J.b/ D w. [Hint: Use Problem
10-16.]
10-18. Suppose .M1 ; g1 / and .M2 ; g2 / are Riemannian manifolds, f W M1 ! RC
is a smooth positive function, and M1 f M2 is the resulting warped product
manifold. Let q0 2 M2 be arbitrary, let W I ! M1 be a g1 -geodesic, and let
z W I ! M1 f M2 be the curve z.t/ D ..t /; q0 /. It follows from the result
of Problem 5-7 that z is a geodesic meeting each submanifold f.t /g M2
orthogonally. Given any fixed vector w 2 Tq0 M2 , define a vector field J
along z by J.t / D .0; w/ 2 T.t/ M1 ˚ Tq0 M2 . Prove that J is a Jacobi field
that is transverse to each submanifold f.t /g M2 (in the sense defined in
Problem 10-14).
Problems 317
10-19. Suppose P is an embedded submanifold in a Riemannian manifold .M; g/.

A point q 2 M is said to be a focal point of P if it is a critical value of the
normal exponential map E W EP ! M (see p. 133). Show that q is a focal
point of P if and only if there exist a geodesic segment W Œ0; b ! M that
starts normal to P and ends at q and a nontrivial Jacobi field J 2 J. /
that is transverse to P in the sense defined in Problem 10-14 and satisfies
J.b/ D 0. (If this is the case, we say that q is a focal point of P along .)
10-20. Suppose .M; g/ is a Riemannian manifold with nonpositive sectional curv-
ature, and P M is a totally geodesic embedded submanifold. Prove that
P has no focal points. [Hint: See Problem 10-7.] (Used on p. 370.)
10-21. Determine the cut locus of an arbitrary point in the n-torus T n with the flat
metric of Examples 2.21 and 7.1.
10-22. Suppose .M; g/ is a connected, compact Riemannian manifold, p 2 M ,
and C M is the cut locus of p. Prove that C is homotopy equivalent to
M X fpg.
10-23. Let .M; g/ be a complete Riemannian manifold, and suppose p; q 2 M are
points such that dg .p; q/ is equal to the distance from p to its cut locus.
(a) Prove that either q is conjugate to p along some minimizing geodesic
segment, or there are exactly two minimizing geodesic segments from
p to q, say 1 ; 2 W Œ0; b ! M , such that 10 .b/ D 20 .b/.
(b) Now suppose in addition that inj.p/ D inj.M /. Prove that if q is not
conjugate to p along any minimizing geodesic, then there is a closed
geodesic W Œ0; 2b ! M such that .0/ D .2b/ D p and .b/ D q,
where b D dg .p; q/.
(Used on p. 343.)
10-24. Let .M; g/ be a complete Riemannian manifold and p 2 M . Show that

inj.p/ is equal to the radius of the largest open ball in Tp M on which expp
is injective.
(Used on p. 166.)
Chapter 11
Comparison Theory
The purpose of this chapter is to show how upper or lower bounds on curvature can
be used to derive bounds on other geometric quantities such as lengths of tangent
vectors, distances, and volumes. The intuition behind all the comparison theorems
is that negative curvature forces geodesics to spread apart faster as you move away
from a point, and positive curvature forces them to spread slower and eventually to
begin converging.
One of the most useful comparison theorems is the Jacobi field comparison
theorem (see Thm. 11.9 below), which gives bounds on the sizes of Jacobi fields
based on curvature bounds. Its importance is based on four observations: first, in a
normal neighborhood of a point p, every tangent vector can be represented as the
value of a Jacobi field that vanishes at p (by Cor. 10.11); second, zeros of Jacobi
fields correspond to conjugate points, beyond which geodesics cannot be minimizing;
third, Jacobi fields represent the first-order behavior of families of geodesics; and
fourth, each Jacobi field satisfies a differential equation that directly involves the
curvature.
In the first section of the chapter, we set the stage for the comparison theorems
by showing how the growth of Jacobi fields in a normal neighborhood is controlled
by the Hessian of the radial distance function, which satisfies a first-order differ-
ential equation called a Riccati equation. We then state and prove a fundamental
comparison theorem for Riccati equations.
Next we proceed to derive some of the most important geometric comparison the-
orems that follow from the Riccati comparison theorem. The first few comparison
theorems are all based on upper or lower bounds on sectional curvatures. Then we
explain how some comparison theorems can also be proved based on lower bounds
for the Ricci curvature. In the next chapter, we will see how these comparison the-
orems can be used to prove significant local-to-global theorems in Riemannian ge-
ometry.
Since all of the results in this chapter are deeply intertwined with lengths and
distances, we restrict attention throughout the chapter to the Riemannian case.

320 11 Comparison Theory
Jacobi Fields, Hessians, and Riccati Equations
Our main aim in this chapter is to use curvature inequalities to derive consequences
about how fast the metric grows or shrinks, based primarily on size estimates for
Jacobi fields. But first, we need to make one last stop along the way.
The Jacobi equation is a second-order differential equation, but comparison the-
ory for differential equations generally works much more smoothly for first-order
equations. In order to get the sharpest results about Jacobi fields and other geomet-
ric quantities, we will derive a first-order equation, called a Riccati equation, that is
closely related to the Jacobi equation.
Let .M; g/ be an n-dimensional Riemannian manifold, let U be a normal neigh-
borhood of a point p 2 M , and let r W U ! R be the radial distance function as
defined by (6.4). The Gauss lemma shows that the gradient of r on U X fpg is the
radial vector field @r .
On U X fpg, we can form the symmetric covariant 2-tensor field r 2 r (the covari-
ant Hessian of r) and the .1; 1/-tensor field Hr D r.@r /. Because @r D grad r D
.rr/] and r commutes with the musical isomorphisms (Prop. 5.17), we have
]
Hr D r.@r / D r .rr/] D r 2 r :
In other words, Hr is obtained from r 2 r by raising one of its indices.

B.1, we can also interpret the .1; 1/-tensor field Hr as an
Using Proposition

element of End TM jU Xfpg (that is, a field of endomorphisms of TM over
U X fpg), defined by
Hr .w/ D rw @r (11.1)
for all w 2 TM jU Xfpg . The endomorphism field Hr is called the Hessian operator
of r. It is related to the .0; 2/-Hessian by

hHr .v/; wi D r 2 r .v; w/; for all q 2 U X fpg and v; w 2 Tq M: (11.2)
The next lemma summarizes some of its basic algebraic properties.

Lemma 11.1. Let r, @r , and Hr be defined as above.
(a) Hr is self-adjoint.
(b) Hr .@r / 0.
(c) The restriction of Hr to vectors tangent to a level set of r is equal to the shape
operator of the level set associated with the normal vector field N D @r .
Proof. Since the covariant Hessian r 2 r is symmetric, equation (11.2) shows that
the Hessian operator is self-adjoint. Part (b) follows immediately from the fact that
Hr .@r / D r@r @r D 0 because the integral curves of @r are geodesics.
Next, note that @r is a unit vector field normal to each level set of r by the Gauss
lemma, so (c) follows from the Weingarten equation 8.11. t
u
Problem 11-1 gives another geometric interpretation of r 2 r, as the radial deriva-
tive of the nonconstant components of the metric in polar normal coordinates.
Jacobi Fields, Hessians, and Riccati Equations 321
The Hessian operator also has a close relationship with Jacobi fields.
Proposition 11.2. Suppose .M; g/ is a Riemannian manifold, U M is a normal

neighborhood of p 2 M , and r is the radial distance function on U . If W Œ0; b ! U
is a unit-speed radial geodesic segment starting at p, and J 2 J ? . / is a normal
Jacobi field along that vanishes at t D 0, then the following equation holds for all
t 2 .0; b:
Dt J.t / D Hr .J.t //: (11.3)
Proof. Let v D 0 .0/, so jvjg D 1 and .t / D expp .t v/. Proposition 10.10 shows
that
J.t / D d expp tv .t w/;
where Dt J.0/ D w D w i @i jp . Because we are assuming that J is normal, it follows
from Proposition 10.7 that w ? v.
Because w ? v ensures that w is tangent to the unit sphere in Tp M at v, we
can choose a smooth curve W ."; "/ ! Tp M that satisfies j .s/jg D 1 for all
s 2 ."; "/, with initial conditions .0/ D v and 0 .0/ D w. (As always, we are
using the canonical identification between Tv .Tp M / and Tp M .) Define a smooth
family of curves W ."; "/ Œ0; b ! M by .s; t / D expp .t .s//. Then .0; t / D
expp .t v/ D .t /, so is a variation of . The stipulation that j .s/jg 1 ensures
that each main curve s .t / D .s; t / is a unit-speed radial geodesic, so its velocity
satisfies the following identity for all .s; t /:
ˇ
@t .s; t / D .s /0 .t / D @r ˇ .s;t/ : (11.4)
The chain rule yields

@s .0; t / D d expp t.0/ .t 0 .0// D d expp tv .t w/ D J.t /;
so J is the variation field of . By the symmetry lemma,
Dt J.t / D Dt @s .0; t / D Ds @t .0; t /: (11.5)
This last expression is the covariant derivative of @t .s; t / along the curve .t/ .s/ D
.s; t / evaluated at s D 0. Since the velocity of this curve at s D 0 is @s .0; t / D
J.t / and @t .s; t / D @r is an extendible vector field by (11.4), we obtain
Ds @t .0; t / D r .t/0 .0/ .@r / D rJ.t/ .@r / D Hr .J.t //:
Combining this with (11.5) yields the result. t

u
In order to compare the Hessian operator of an arbitrary metric with those of the
constant-curvature models, we need the following explicit formula for the constant-
curvature case.
Proposition 11.3. Suppose .M; g/ is a Riemannian manifold, U M is a normal

neighborhood of p 2 M , and r is the radial distance function on U . Then g has
constant sectional curvature c on U if and only if the following formula holds at all
points of U X fpg:
s 0 .r/
Hr D c r ; (11.6)
sc .r/
where sc is defined by (10.8), and for each q 2 U X fpg, r W Tq M ! Tq M is the
orthogonal projection onto the tangent space of the level set of r (equivalently, onto
the orthogonal complement of @r jq ).
Proof. First suppose g has constant sectional curvature c on U . Let q 2 U X fpg,

and let W Œ0; b ! U be the unit-speed radial geodesic from p to q, so b D r.q/.
Let .E1 ; : : : ; En / be a parallel orthonormal frame along , chosen so that En .t / D
0 .t / D @r j.t/ . It follows from Proposition 10.12 that for i D 1; : : : ; n 1, the vector
fields Ji .t / D sc .t /Ei .t / are normal Jacobi fields along that vanish at t D 0. The
assumption that U is a normal neighborhood of p means that U D expp .V / for
some star-shaped neighborhood V of 0 2 Tp M , and every point of V is a regular
point for expp . Thus Proposition 10.20 shows that p has no conjugate points along
, which implies that sc .t / ¤ 0 for t 2 .0; b. (For c 0, this is automatic, because
sc vanishes only at 0; but in the case c D 1=R2 > 0, it means that b < R.)
For 1 i n 1, we use Proposition 11.2 to compute
Dt Ji .t / D Hr .Ji .t // D sc .t /Hr .Ei .t //:
On the other hand, because each Ei is parallel along ,
Dt Ji .t / D sc0 .t /Ei .t /:
Comparing these two equations at t D b and dividing by sc .b/, we obtain
sc0 .b/ s 0 .b/

Hr .Ei .b// D Ei .b/ D c r .Ei .b//:
sc .b/ sc .b/
On the other hand, Lemma 11.1(b) shows that
s 0 .b/
Hr .En .b// D Hr @r jq D 0 D c r .En .b//;
sc .b/
because r .En .b// D r .@r jq / D 0. Since .Ei .b// is a basis for Tq M , this proves
(11.6).
Conversely, suppose Hr is given by (11.6). Let be a radial geodesic starting at
p, and let J be a normal Jacobi field along that vanishes at t D 0. By Proposi-
tion 11.2, Dt J.t / D Hr J.t / D sc0 .t /J.t /=sc .t /. A straightforward computation then
shows that sc .t /1 J.t / is parallel along . Thus we can write every such Jacobi field
in the form J.t / D ksc .t /E.t / for some constant k and some parallel unit normal
vector field E along . Proceeding exactly as in the proof of Theorem 10.14, we
conclude that g is given by formula (10.17) in these coordinates, and therefore has
constant sectional curvature c. t
u
Fig. 11.1: The graph of sc0 =sc
˚
For convenience, we record the exact formulas for the quotient sc0 =sc that ap-
peared in the previous proposition (see Fig. 11.1):
1
; if c D 0I
t
sc0 .t / 1 t 1
D cot ; if c D > 0I
sc .t / R R R2
1 t 1
coth ; if c D 2 < 0:
R R R
Now we are in a position to derive the first-order equation mentioned at the begin-
ning of this section. (Problem 11-3 asks you to show, with a different argument, that
the conclusion of the next theorem holds for the Hessian operator of every smooth
local distance function, not just the radial distance function in a normal neighbor-
hood.) This theorem concerns the covariant derivative of the endomorphism field
Hr along a curve . We can compute the action of Dt Hr on every V 2 X. / by
noting that Hr .V .t // is a contraction of Hr ˝ V .t /, so the product rule implies
Dt .Hr .V // D .Dt Hr /V C Hr .Dt V /.
Theorem 11.4 (The Riccati Equation). Let .M; g/ be a Riemannian manifold; let
U be a normal neighborhood of a point p 2 M ; let r W U ! R be the radial distance
function; and let W Œ0; b ! U be a unit-speed radial geodesic. The Hessian oper-
ator Hr satisfies the following equation along j.0;b , called a Riccati equation:
Dt Hr C Hr2 C R 0 D 0; (11.7)
where Hr2 and R 0 are the endomorphism

0 fields along defined by Hr2 .w/ D
0
Hr .Hr .w// and R 0 .w/ D R w; , with R the curvature endomorphism of g.
Proof. Let t0 2 .0; b and w 2 T.t0 / M be arbitrary. We can decompose w as w D

y C z, where y is a multiple of @r and z is tangent to a level set of r. Since (11.7)
is an equation between linear operators, we can prove the equation by evaluating it
separately on y and z.
Because is a unit-speed radial geodesic, its velocity is equal to @r , and thus
Dt @r D 0 along . It follows that .Dt Hr /.@r / D Dt .Hr .@r // Hr .Dt @r / D 0.
Since all three terms on the left-hand side of (11.7) annihilate @r , the equation holds
when applied to any multiple of @r .
Next we consider a vector z 2 T.t0 / M that is tangent to a level set of r, and thus
by the Gauss lemma orthogonal to 0 .t0 /. By Corollary 10.11, z can be expressed as
the value at t D t0 of a Jacobi field J along vanishing at t D 0. Because J.0/ and
J.t0 / are orthogonal to 0 , it follows that J is a normal Jacobi field, so Proposition
11.2 shows that Dt J.t / D Hr .J.t // for all t 2 Œ0; b. Differentiation yields
Dt2 J D Dt .Hr .J // D .Dt Hr /J C Hr .Dt J / D .Dt Hr /J C Hr .Hr .J //:
On the other hand, the Jacobi equation gives Dt2 J D R 0 .J /, so

Dt Hr C Hr2 C R 0 .J / 0:
Evaluating this at t D t0 proves the result. t

u
The Riccati equation is named after Jacopo Riccati, an eighteenth-century Italian
mathematician who studied scalar differential equations of the form v 0 C pv 2 C
qv C r D 0, where p; q; r are known functions and v is an unknown function of one
real variable. As is shown in some ODE texts, a linear second-order equation in one
variable of the form au00 C bu0 C cu D 0 can be transformed to a Riccati equation
wherever u ¤ 0 by making the substitution v D u0 =u. The relation (11.3) generalizes
this, and allows us to replace the analysis of the second-order linear Jacobi equation
by an analysis of the first-order nonlinear Riccati equation.
The primary tool underlying all of our geometric comparison theorems is a fun-
damental comparison theorem for solutions to Riccati equations. It says, roughly,
that a larger curvature term results in a smaller solution, and vice versa. When we
apply this to (11.3), it will yield an analogous comparison for Jacobi fields.
In the statement and proof of this theorem, we will compare self-adjoint en-
domorphisms by comparing the quadratic forms they determine. Given a finite-
dimensional inner product space V and self-adjoint endomorphisms A; B W V ! V ,
the notation A B means that hAv; vi hBv; vi for all v 2 V , or equivalently
that B A is positive semidefinite. In particular, B 0 means that B is positive
semidefinite. Note that˝ the square
˛ of every self-adjoint endomorphism is positive
semidefinite, because B 2 v; v D hBv; Bvi 0 for all v 2 V .
Theorem 11.5 (Riccati Comparison Theorem). Suppose .M; g/ is a Riemann-

ian manifold and W Œa; b ! M is a unit-speed geodesic segment. Suppose ; z
are self-adjoint endomorphism fields along j.a;b that satisfy the following Riccati
equations:
Dt C 2 C D 0; Dt z C z2 C z D 0; (11.8)
where and z are continuous self-adjoint endomorphism fields along satisfying
z .t / .t / for all t 2 Œa; b: (11.9)

Suppose further that limt&a z.t / .t / exists and satisfies

lim z.t / .t / 0:
t&a
Then
z.t / .t / for all t 2 .a; b:
To prove this theorem, we will express the endomorphism fields , z, , and z in
terms of a parallel orthonormal frame along . In this frame, they become symmetric
matrix-valued functions, and then the Riccati equations for and z become ordinary
differential equations for these matrix-valued functions. The crux of the matter is the
following comparison theorem for solutions to such matrix-valued equations.
Let M.n; R/ be the space of all n n real matrices, viewed as linear endomor-
phisms of Rn , and let S.n; R/ M.n; R/ be the subspace of symmetric matrices,
corresponding to self-adjoint endomorphisms of Rn with respect to the standard
inner product.
z W .a; b !
Theorem 11.6 (Matrix Riccati Comparison Theorem). Suppose H; H
S.n; R/ satisfy the following matrix Riccati equations:
H 0 C H 2 C S D 0; z0 CH
H z 2 C Sz D 0; (11.10)
where S; Sz W Œa; b ! S.n; R/ are continuous and satisfy

z S.t / for all t 2 Œa; b:
S.t/ (11.11)

Suppose further that limt&a H z .t / H.t / exists and satisfies

lim Hz .t / H.t / 0: (11.12)
t&a
Then
z .t / H.t / for all t 2 .a; b:
H (11.13)
Proof. Define functions A; B W .a; b ! S.n; R/ by
z .t / H.t /; 1 z
A.t / D H B.t / D H .t / C H.t / :
2
The hypothesis implies that A extends to a continuous matrix-valued function (still
denoted by A) on Œa; b satisfying A.a/ 0. We need to show that A.t / 0 for all
t 2 .a; b.
Simple computations show that the following equalities hold on .a; b:
A0 D BA C AB Sz C S;
1 z2 z
B0 D H CS CH2 CS :
2
Our hypotheses applied to these formulas imply
A0 BA C AB; (11.14)
B 0 k Id; (11.15)
where the last inequality holds for some (possibly negative) real number k because
z 2 are positive semidefinite and S and Sz are continuous on all of Œa; b and
H 2 and H
thus bounded below. Therefore, for every t 2 .a; b,
Z b
B.t / D B.b/ B 0 .u/ du
t (11.16)
B.b/ .b t /k Id B.b/ C jb aj jkj Id;
which shows that B.t / is bounded above on .a; b. Let K be a constant large enough
that B.t / K Id is negative definite for all such t .
Define a continuous function f W Œa; b Sn1 ! R by
f .t; x/ D e 2Kt hA.t /x; xi :
To prove the theorem, we need to show that f .t; x/ 0 for all .t; x/ 2 Œa; b Sn1 .
Suppose this is not the case; then by compactness of Œa; b Sn1 , f takes on a
positive maximum at some .t0 ; x0 / 2 Œa; b Sn1 . Since f .a; x/ D 0 for all x,
we must have a < t0 b. Because hA.t0 /x; xi hA.t0 /x0 ; x0 i for all x 2 Sn1 ,
it follows from Lemma 8.14 that x0 is an eigenvector of A.t0 / with eigenvalue
0 D hA.t0 /x0 ; x0 i > 0.
Since f is differentiable at .t0 ; x0 / and f .t; x0 / f .t0 ; x0 / for a < t < t0 , we
have
@f f .t; x0 / f .t0 ; x0 /
.t0 ; x0 / D lim 0:
@t t%t0 t t0
(We have to take a one-sided limit here to accommodate the fact that t0 might equal
b.) On the other hand, from (11.14) and the fact that A.t0 / and B.t0 / are self-adjoint,
we have
@f ˝ ˛
.t0 ; x0 / D e 2Kt0 A0 .t0 /x0 ; x0 2K hA.t0 /x0 ; x0 i
@t
e 2Kt0 .hB.t0 /A.t0 /x0 ; x0 i C hA.t0 /B.t0 /x0 ; x0 i 2K hA.t0 /x0 ; x0 i/
D e 2Kt0 .2 hA.t0 /x0 ; B.t0 /x0 i 2K hA.t0 /x0 ; x0 i/
D e 2Kt0 .2 0 hx0 ; .B.t0 / K Id/x0 i/ < 0:
These two inequalities contradict each other, thus completing the proof. t
u
Comparisons Based on Sectional Curvature 327
Proof of Theorem 11.5. Suppose , z, , and z are self-adjoint endomorphism fields
along satisfying the hypotheses of the theorem. Let .E1 ; : : : ; En / be a parallel
orthonormal frame along , and let H; H z W .a; b ! S.n; R/ and S; Sz W Œa; b !
S.n; R/ be the symmetric matrix-valued functions defined by
.t /.Ei .t // D Hij .t /Ej .t /; z j .t /Ej .t /;

z.t /.Ei .t // D H i
.t /.Ei .t // D Sij .t /Ej .t /; z .t /.Ei .t // D Szij .t /Ej .t /:
Because Dt Ej 0, the Riccati equations (11.8) reduce to the ordinary differen-

tial equations (11.10) for these matrix-valued functions. Theorem 11.6 shows that
Hz .t / H.t / for all t 2 .a; b, which in turn implies that z.t / .t /. t
u
Comparisons Based on Sectional Curvature

Now we are ready to establish some comparison theorems for metric quantities
based on comparing sizes of Hessian operators and Jacobi fields for an arbitrary
metric with those of the constant-curvature models.
The most fundamental comparison theorem is the following result, which com-
pares the Hessian of the radial distance function with its counterpart for a constant-
curvature metric.
Theorem 11.7 (Hessian Comparison). Suppose .M; g/ is a Riemannian n-mani-

fold, p 2 M , U is a normal neighborhood of p, and r is the radial distance function
on U .
(a) If all sectional curvatures of M are bounded above by a constant c, then the
following inequality holds in U0 X fpg:
sc0 .r/
Hr r ; (11.17)
sc .r/
where sc and r are defined as in Proposition 11.3, and U0 D U if c 0,
while U0 D fq 2 U W r.q/ < Rg if c D 1=R2 > 0.
(b) If all sectional curvatures of M are bounded below by a constant c, then the
following inequality holds in all of U X fpg:
sc0 .r/
Hr r : (11.18)
sc .r/

Proof. Let x 1 ; : : : ; x n be Riemannian normal coordinates on U centered at p, let
r be the radial distance function on U , and let sc be the function defined by (10.8).
Let U0 U be the subset on which sc .r/ > 0; when c 0, this is all of U , but when
c D 1=R2 > 0, it is the subset where r < R. Let Hrc be the endomorphism field on
U0 X fpg given by
sc0 .r/
Hrc D r :
sc .r/
Let q 2 U0 X fpg be arbitrary, and let W Œ0; b ! U0 be the unit-speed radial
geodesic from p to q. Note that at every point .t / for 0 < t b, the endomorphism
field r can be expressed as r .w/ D w hw; @r i@r D w hw; 0 i 0 , and in this
form it extends smoothly to an endomorphism field along all of . Moreover, since
Dt 0 D 0 along , it follows that Dt r D 0 along as well. Therefore, direct com-
putation using the facts that sc00 D csc and r2 D r shows that Hrc satisfies the
following Riccati equation along j.0;b :
Dt Hrc C .Hrc /2 C cr D 0:
On the other hand, Theorem 11.4 shows that Hr satisfies
Dt Hr C Hr2 C R 0 D 0:
The sectional curvature hypothesis implies that R 0 cr in ˛ R 0 cr

˝ case (a) and
in case (b), using the facts that R 0 . 0 / D 0 D r . 0 /, and R 0 .w/; w D sec. 0 ; w/
if w is a unit vector orthogonal to 0 .
In order to apply the Riccati comparison theorem to Hr and Hrc , we need to show
that Hr Hrc has a finite limit along as t & 0. A straightforward series expansion
shows that no matter what c is,
sc0 .r/=sc .r/ D 1=r C O.r/ (11.19)
as r & 0. We will show that Hr satisfies the analogous estimate:

1
Hr D r C O.r/: (11.20)
r
The easiest way to verify (11.20) is to note that on U X fpg, Hr has the following
coordinate expression in normal coordinates:

Hr D g ij @j @k r jmk @m r @i ˝ dx k :
Now @m r D x m =r, which is bounded on U X fpg, and @j @k r D O.r 1 /. Moreover,

g ij D ı ij C O.r 2 / and jmk D O.r/, so Hr is equal to ı ij .@j @k r/@i ˝ dx k plus
terms that are O.r/ in these coordinates. But this last expression is exactly the coor-
dinate expression for Hr in the case of the Euclidean metric in normal coordinates,
which Proposition 11.3 shows is equal to .1=r/r . This proves (11.20), from which
we conclude that Hr Hrc approaches zero along as t & 0.
If the sectional curvatures of g are bounded above by c, then the arguments above
show that the hypotheses of the Riccati comparison theorem are satisfied along
jŒ0;b with .t / D Hr j.t/ , .t / D R 0 j.t/ , z.t / D Hrc j.t/ , and z .t / D cr j.t/ . It
follows that Hr Hrc at q D .b/, thus proving (a).
On the other hand, if the sectional curvatures are bounded below by c, the same
argument with the roles of Hr and Hrc reversed shows that Hr Hrc on U0 . It
remains only to show that U0 D U in this case. If c 0, this is automatic. If c D
1=R2 > 0, then sc0 .r/=sc .r/ ! 1 as r % R; since Hr is defined and smooth in
all of U X fpg and bounded above by Hrc , it must be the case that r < R in U ,
which implies that U0 D U . t
u
Corollary 11.8 (Principal Curvature Comparison). Suppose .M; g/ is a Rie-

mannian n-manifold, p 2 M , U is a normal neighborhood of p, r is the radial
distance function on U , and sc and r are defined as in Proposition 11.3.
(a) If all sectional curvatures of M are bounded above by a constant c, then the
principal curvatures of the r-level sets in U0 X fpg (with respect to the inward
unit normal) satisfy
s 0 .r/

c ;
sc .r/
where U0 D U if c 0, while U0 D fq 2 U W r.q/ < Rg if c D 1=R2 > 0.
(b) If all sectional curvatures of M are bounded below by a constant c, then the
principal curvatures of the r-level sets in U X fpg (with respect to the inward
unit normal) satisfy
s 0 .r/

c :
sc .r/
Proof. This follows immediately from the fact that the shape operator of each r-
level set is the restriction of Hr by Lemma 11.1(c). u
t
Because Jacobi fields describe the behavior of families of geodesics, the next
theorem gives some substance to the intuitive notion that negative curvature tends
to make nearby geodesics spread out, while positive curvature tends to make them
converge. More precisely, an upper bound on curvature forces Jacobi fields to be at
least as large as their constant-curvature counterparts, and a lower curvature bound
constrains them to be no larger.
Theorem 11.9 (Jacobi Field Comparison). Suppose .M; g/ is a Riemannian man-

ifold, W Œ0; b ! M is a unit-speed geodesic segment, and J is any normal Jacobi
field along such that J.0/ D 0. For each c 2 R, let sc be the function defined by
(10.8).
(a) If all sectional curvatures of M are bounded above by a constant c, then
jJ.t /j sc .t /jDt J.0/j (11.21)
for all t 2 Œ0; b1 , where b1 D b if c 0, and b1 D min.b; R/ if c D 1=R2 > 0.

(b) If all sectional curvatures of M are bounded below by a constant c, then
jJ.t /j sc .t /jDt J.0/j (11.22)
for all t 2 Œ0; b2 , where b2 is chosen so that .b2 / is the first conjugate point
to .0/ along if there is one, and otherwise b2 D b.
Proof. If Dt J.0/ D 0, then J vanishes identically, so we may as well assume that

Dt J.0/ ¤ 0. Let b0 be the largest time in .0; b such that has no conjugate points
in .0; b0 / and sc .t / > 0 for t 2 .0; b0 /. Let p D .0/, and assume temporarily
that Œ0; b0 / is contained in a normal neighborhood U of p. Define a function
f W .0; b0 / ! R by
f .t / D log sc .t /1 jJ.t /j :
Differentiating with respect to t and using Dt J D Hr .J /, we get
˝ ˛ ˝ ˛
0 d d Dt J; J sc0 Hr .J /; J sc0
f D log jJ j log sc D D :
dt dt jJ j2 sc jJ j2 sc
Under hypothesis (a), it follows from the Hessian comparison theorem that f 0 .t /
0 for all t 2 .0; b0 /, so f .t / is nondecreasing, and thus so is sc .t /1 jJ.t /j. Simi-
larly, under hypothesis (b), we get f 0 .t / 0, which implies that sc .t /1 jJ.t /j is
nonincreasing.
Next we consider the limit of sc .t /1 jJ.t /j as t & 0. Two applications of
l’Hôpital’s rule yield
jJ j2 2hDt J; J i 2hDt2 J; J i C 2jDt J j2

lim D lim D lim
t&0 sc2 t&0 2sc0 sc t&0 2sc00 sc C 2sc02
2hR.J; 0 / 0 ; J i C 2jDt J j2
D lim ;
t&0 2sc00 sc C 2sc02
provided the last limit exists. Since J ! 0, sc ! 0, and sc0 ! 1 as t & 0, this last
limit does exist and is equal to jDt J.0/j2 . Combined with the derivative estimates
above, this shows that the appropriate conclusion
(11.21) or (11.22) holds on .0; b0 /,
and thus by continuity on Œ0; b0 , when Œ0; b0 / is contained in a normal neighbor-
hood of p.
Now suppose is an arbitrary geodesic segment, not assumed to be contained in
a normal neighborhood of p. Let v D 0 .0/, so that .t / D expp .t v/ for t 2 Œ0; b,
and define b0 as above. The definition ensures that .t / is not conjugate to .0/ for
t 2 .0; b0 /, and therefore expp is a local diffeomorphism on some neighborhood of
the set L D ft v W 0 t < b0 g. Let W Tp M be a convex open set containing L on
which expp is a local diffeomorphism, and let gz D expp g, which is a Riemannian
metric on W that satisfies the same curvature estimates as g (Fig. 11.2). By construc-
tion, W is a normal neighborhood of 0, and expp is a local isometry from W; gz
to .M; g/. The curve z.t / D t v for t 2 Œ0; b0 / is a radial geodesic in W , and Propo-
1
sition 10.5 shows that the vector field Jz.t / D d expp tv .J.t // is a Jacobi field
along z thatˇ vanishes
ˇ at t Dˇ 0. Therefore,
ˇ for t 2 .0; ˇb0 /, the
ˇ preceding
ˇ argumentˇ
z t Jz.0/ˇ in case (a) and ˇJz.t /ˇ sc .t /ˇD
implies that ˇJz.t /ˇgz sc .t /ˇD g
z g
z
z t Jz.0/ˇ in
g
z
case (b). This implies that the conclusions of the theorem hold for J on the interval
.0; b0 / and thus by continuity on Œ0; b0 .
To complete the proof, we need to show that b0 b1 in case (a) and b0 b2 in
case (b). Assuming the hypothesis of (a), suppose for contradiction that b0 <b1 . The
Fig. 11.2: Pulling the metric back to Tp M
only way this can occur is if .b0 / is conjugate to .0/ along , while sc .b0 />0.
This means that there is a nontrivial normal Jacobi field J 2 J. / satisfying J.0/ D
0 D J.b0 /. But the argument above showed that every such Jacobi field satisfies
jJ.t /j sc .t /jDt J.0/j for t 2 Œ0; b0 and thus jJ.b0 /j sc .b0 /jDt J.0/j > 0, which
is a contradiction. Similarly, in case (b), suppose b0 < b2 . Then sc .b0 / D 0, but .b0 /
is not conjugate to .0/ along . If J is any nontrivial normal Jacobi field along
that vanishes at t D 0, the argument above shows that jJ.t /j sc .t /jDt J.0/j for
t 2 Œ0; b0 , so jJ.b0 /j sc .b0 /jDt J.0/j D 0; but this is impossible because .b0 / is
not conjugate to .0/. t
u
There is a generalization of the preceding theorem, called the Rauch comparison
theorem, that allows for comparison of Jacobi fields in two different Riemannian
manifolds when neither is assumed to have constant curvature. The statement and
proof can be found in [CE08, Kli95].
Because all tangent vectors in a normal neighborhood are values of Jacobi fields
along radial geodesics, the Jacobi field comparison theorem leads directly to the
following comparison theorem for metrics.
Theorem
11.10
(Metric Comparison). Let .M; g/ be a Riemannian manifold, and
let U; x i be any normal coordinate chart for g centered at p 2 M . For each
c 2 R, let gc denote the constant-curvature metric on U X fpg given in the same
coordinates by formula (10.17).
(a) Suppose all sectional curvatures of g are bounded above by a constant c. If
c 0, then for all q 2 U X fpg and all w 2 Tq M , we have g.w; w/ gc .w; w/.
If c D 1=R2 > 0, then the same holds, provided that dg .p; q/ < R.
(b) If all sectional curvatures of g are bounded below by a constant c, then for all
q 2 U X fpg and all w 2 Tq M , we have g.w; w/ gc .w; w/.
Proof. Let q 2 U Xfpg, satisfying the restriction dg .p; q/ < R if we are in case (a)
and c D 1=R2 > 0, but otherwise arbitrary; and let b D dg .p; q/. Given w 2 Tq M ,
we can decompose w as a sum w D y C z, where y is a multiple of the radial
vector field @r and z is tangent to the level set r D b. Then gc .y; z/ D 0 by direct
computation, and the Gauss lemma shows that g.y; z/ D 0, so
g.w; w/ D g.y; y/ C g.z; z/;

gc .w; w/ D gc .y; y/ C gc .z; z/:
Because @r is a unit vector with respect to both g and gc , it follows that g.y; y/ D
gc .y; y/. So it suffices to prove the comparison for z.
There is a radial geodesic W Œ0; b ! U satisfying .0/ D p and .b/ D q, and
Corollary 10.11 shows that z D J.b/ for some Jacobi field J along vanishing at
t D 0, which is normal because it is orthogonal to 0 at t D 0 and t D b. Proposition
i
10.10
1 shows n
iformula J.t / D t a @i j.t/ for some constants
that J has the coordinate
a ; : : : ; a . Since the coordinates x / are normal coordinates for both g and gc ,
it follows that is also a radial geodesic for gc , and the same vector field J is also
a normal Jacobi field for gc along . Therefore, g.z; z/ D jJ.b/j2g and gc .z; z/ D
jJ.b/j2gc . In case (a), our hypothesis guarantees that sc .b/ > 0, so
g.z; z/ D jJ.b/j2g jsc .b/j2 jDt J.0/j2g (Jacobi field comparison theorem)
2
D jsc .b/j jDt J.0/j2gc (since g and gc agree at p)
D jJ.b/j2gc D gc .z; z/ (by Prop. 10:12):
In case (b), the same argument with the inequalities reversed shows that g.z; z/
gc .z; z/. t
u
The next three comparison theorems (Laplacian, conjugate point, and volume
comparisons) can be proved equally easily under the assumption of either an upper
bound or a lower bound for the sectional curvature, just like the preceding theorems.
However, we state these only for the case of an upper bound, because we will prove
stronger theorems later in the chapter based on lower bounds for the Ricci curvature
(see Thms. 11.15, 11.16, and 11.19).
The first of the three is a comparison of the Laplacian of the radial distance func-
tion with its constant-curvature counterpart. Our primary interest in the Laplacian of
the distance function stems from its role in volume and conjugate point comparisons
(see Thms. 11.14, 11.16, and 11.19 below); but it also plays an important role in the
study of various partial differential equations on Riemannian manifolds.
Theorem 11.11 (Laplacian Comparison I). Suppose .M; g/ is a Riemannian n-

manifold whose sectional curvatures are all bounded above by a constant c. Suppose
p 2 M , U is a normal neighborhood of p, r is the radial distance function on U ,
and sc is defined as in Proposition 11.3. Then on U0 X fpg, we have
sc0 .r/
r .n 1/ ;
sc .r/
where U0 D U if c 0, while U0 D fq 2 U W r.q/ < Rg if c D 1=R2 > 0.

Proof. By the result of Problem 5-14, r D trg r 2 r D tr Hr . The result then
follows from the Hessian comparison theorem, using the fact that tr.r / D n 1,
which can be verified easily by expressing r locally in an adapted orthonormal
frame for the r-level sets. t
u
The next theorem shows how an upper curvature bound prevents the formation
of conjugate points. It will play a decisive role in the proof of the Cartan–Hadamard
theorem in the next chapter.
Theorem 11.12 (Conjugate Point Comparison I). Suppose .M; g/ is a Riemann-

ian n-manifold whose sectional curvatures are all bounded above by a constant c. If
c 0, then no point of M has conjugate points along any geodesic. If c D 1=R2 > 0,
then there is no conjugate point along any geodesic segment shorter than R.
Proof. The case c 0 is covered by Problem 10-7, so assume c D 1=R2 > 0. Let
W Œ0; b ! M be a unit-speed geodesic segment, and suppose J is a nontrivial
normal Jacobi field along that vanishes at t D 0. The Jacobi field comparison
theorem implies that jJ.t /j .constant/ sin.t =R/ > 0 as long as 0 < t < R. t
u
The last of our sectional curvature comparison theorems is a comparison of vol-
ume growth of geodesic balls. Before proving it, we need the following lemma,
which shows how the Riemannian volume form is related to the Laplacian of the
radial distance function.

Lemma 11.13. Suppose .M; g/ is a Riemannian manifold and x i are Riemannian
normal coordinates on a normal neighborhood U of p 2 M . Let det g denote the
determinant of the matrix .gij / in these coordinates, let r be the radial distance
function, and let @r be the unit radial vector field. The following identity holds on
U X fpg: p
r D @r log r n1 det g : (11.23)
Proof. Corollary 6.10 to the Gauss lemma shows that the vector fields grad r and
@r are equal on U X fpg. Comparing the components of these two vector fields in
normal coordinates, we conclude (using the summation convention as usual) that
xi
g ij @j r D :
r
Based on the formula for r from Proposition 2.46, we compute
1 p
r D p @i g ij det g @j r
det g
i
1 x p
Dp @i det g
det g r
i
x 1 xi p
D @i Cp @i det g
r det g r
n1 1 p
D Cp @r det g;
r det g
where the first term in the last line follows by direct computation using r D
P i 2 1=2
i x . This is equivalent to (11.23). u
t
The following result was proved by Paul Günther in 1960 [Gün60]. (Günther also
proved an analogous result in the case of a lower sectional curvature bound, but that
result has been superseded by the Bishop–Gromov theorem, Thm. 11.19 below.)
Theorem 11.14 (Günther’s Volume Comparison). Suppose .M; g/ is a connected

Riemannian n-manifold whose sectional curvatures are all bounded above by a
constant c. Given p 2 M , let ı0 D inj.p/ if c 0, and ı0 D min.R; inj.p// if
c D 1=R2 > 0. For every positive number ı ı0 , let Vg .ı/ denote the volume of the
geodesic ball Bı .p/ in .M; g/, and let Vc .ı/ denote the volume of a geodesic ball
of radius ı in the n-dimensional Euclidean space, hyperbolic space, or sphere with
constant sectional curvature c. Then for every 0 < ı ı0 , we have
Vg .ı/ Vc .ı/; (11.24)
and the quotient Vg .ı/=Vc .ı/ is a nondecreasing function of ı that approaches 1 as

ı & 0. If equality holds in (11.24) for some ı 2 .0; ı0 , then g has constant sectional
curvature c on the entire geodesic ball Bı .p/.
Proof. The volume estimate (11.24) follows easily from the metric comparison the-
orem, which implies
p the determinants of the metrics g and gc in normal coor-
that p
dinates satisfy det g det gc . If that were all we needed, we could stop here;
but to prove the other statements, we need a more involved argument, which inci-
dentally provides another proof of (11.24) that does not rely directly on the metric
comparison theorem, and therefore can be adapted more easily to the case in which
we have only
an estimate of the Ricci curvature (see Thm. 11.19 below).
Let x i be normal coordinates on Bı0 .p/ (interpreted as all of M if ı0 D 1).
Using these coordinates, we might as well consider g to be a Riemannian metric
on an open subset of Rn and p to be the origin. Let gx denote the Euclidean metric
in these coordinates, and let gc denote the constant-curvature metric in the same
coordinates, given on the complement of the origin by (10.17).
The Laplacian comparison theorem together with Lemma 11.13 shows that
p s 0 .r/
@r log r n1 det g D r .n 1/ c D @r log sc .r/n1 : (11.25)
sc .r/
p
Thus log r n1 det g=sc .r/n1 p is a nondecreasing function of r along each radial
geodesic, and so is the ratio r n1 p det g=sc .r/n1 . To compute the limit as r & 0,
note that gij D ıij at the origin, so det g convergesp uniformly to 1 as r & 0. Also,
n1
for every c, we have sc .r/=r
p ! 1 as r & 0, so r det g=sc .r/n1 ! 1.
We can write d Vg D det g d Vgx . Corollary 10.17 in the case c D 0 shows that
Z Z ı p
Vg .ı/ D det g ı ˚.; !/n1 d d VgV ;
Sn1 0
where ˚.; !/ D ! for 2 .0; ı0 / and ! 2 Sn1 . The same corollary shows that
Z Z Z ! Z
ı ı
Vc .ı/ D sc ./n1 d d VgV D sc ./n1 d d VgV :
Sn1 0 0 Sn1
Therefore,
Z Z ı p
det g ı ˚.; !/n1 d d VgV
Vg .ı/ Sn1 0
D Z ! Z
Vc .ı/ ı
sc ./n1 d d VgV
0
˙Z ı
Sn1

n1
Z .; !/sc ./ d
1 0
D Z ı d VgV ; (11.26)
Vol.Sn1 / Sn1
sc ./n1 d
0
where we have written

p
det g ı ˚.; !/n1
.; !/ D :
sc ./n1
The argument above (together with the fact that r ı ˚ D ) shows that .; !/ is a
nondecreasing function of for each !, which approaches 1 uniformly as & 0.
We need to show that the quotient in parentheses in the last line of (11.26) is also
nondecreasing as ı increases. Suppose 0 < ı1 < ı2 . Because is nondecreasing, we
have
Z ı1 Z ı2
n1
.; !/sc ./ d sc ./n1 d
0 ı1
Z ! Z !
ı1 ı2
sc ./n1 d .ı1 ; !/ sc ./n1 d
0 ı1
Z ı1 Z ı2
sc ./n1 d .; !/sc ./n1 d;
0 ı1
and therefore (suppressing the dependence of the integrands on and ! for brevity),
Z ı1 Z ı2
scn1 d scn1 d
0 0
Z Z Z !
ı1 ı1 ı2
D scn1 d scn1 d C scn1 d
0 0 ı1
Z ı1 Z ı1 Z ı1 Z ı2
scn1 d scn1 d C scn1 d scn1 d
0 0 0 ı1
Z ı1 Z ı2
D scn1 d scn1 d:
0 0
Evaluating (11.26) at ı1 and ı2 and inserting the inequality above shows that the
ratio Volg .ı/= Volc .ı/ is nondecreasing as a function of ı, and it approaches 1 as
ı & 0 because .; !/ ! 1 as & 0. It follows that Vg .ı/ Vc .ı/ for all ı 2 .0; ı0 .
It remains only to consider the case in which the volume ratio is equal to 1 for
some ı. If .; !/ is not identically 1 on the set where 0 < < ı, then it is strictly
greater than 1 on a nonempty open subset, which implies that the volume ratio
in (11.26) is strictly greater than 1; so Volg .ı/ D Volc .ı/ implies
p .; !/ 1 on
.0; ı/ Sn1 , and pulling back to U via ˚ 1 shows that r n1 det g sc .r/n1 on
Bı .p/. By virtue of (11.25), we have r .n 1/sc0 .r/=sc .r/, or in other words,
tr.Hr / D tr..sc0 .r/=sc .r//r /.It follows from
the Hessian comparison theorem that
the endomorphism field Hr sc0 .r/=sc .r/ r is positive semidefinite, so its eigen-
values are all nonnegative. Since its trace is zero, the eigenvalues must all be zero.
In other words, Hr .sc0 .r/=sc .r//r on the geodesic ball Bı .p/. It then follows
from Proposition 11.3 that g has constant sectional curvature c on that ball. t
u
Comparisons Based on Ricci Curvature
All of our comparison theorems so far have been based on assuming an upper or
lower bound for the sectional curvature. It is natural to wonder whether anything can be
said if we weaken the hypotheses and assume only bounds on other curvature quan-
tities such as Ricci or scalar curvature.
It should be noted that except in very low dimensions, assuming a bound on Ricci
or scalar curvature is a strictly weaker hypothesis than assuming one on sectional
curvature. Recall Proposition 8.32, which says that on an n-dimensional Rieman-
nian manifold, the Ricci curvature evaluated on a unit vector is a sum of n 1 sec-
tional curvatures, and the scalar curvature is a sum of n.n 1/ sectional curvatures.
Thus if .M; g/ has sectional curvatures bounded below by c, then its Ricci curvature
satisfies Rc.v; v/ .n 1/c for all unit vectors v, and its scalar curvature satisfies
S n.n 1/c, with analogous inequalities if the sectional curvature is bounded
above. However, the converse is not true: an upper or lower bound on the Ricci cur-
vature implies nothing about individual sectional curvatures, except in dimensions
2 and 3, where the entire curvature tensor is determined by the Ricci curvature (see
Comparisons Based on Ricci Curvature 337
Cors. 7.26 and 7.27). For example, in every even dimension greater than or equal to
4, there are compact Riemannian manifolds called Calabi–Yau manifolds that have
zero Ricci curvature but nonzero sectional curvatures (see, for example, [Bes87,
Chap. 11]).
In this section we investigate the extent to which bounds on the Ricci curva-
ture lead to useful comparison theorems. The strongest theorems of the preceding
section, such as the Hessian, Jacobi field, and metric comparison theorems, do not
generalize to the case in which we merely have bounds on Ricci curvature. How-
ever, it is a remarkable fact that Laplacian, conjugate point, and volume comparison
theorems can still be proved assuming only a lower (but not upper) bound on the
Ricci curvature. (The problem of drawing global conclusions from scalar curvature
bounds is far more subtle, and we do not pursue it here. A good starting point for
learning about that problem is [Bes87].)
The next theorem is the analogue of Theorem 11.11.
Theorem 11.15 (Laplacian Comparison II). Let .M; g/ be a Riemannian n-mani-

fold, and suppose there is a constant c such that the Ricci curvature of M satisfies
Rc.v; v/ .n 1/c for all unit vectors v. Given any point p 2 M , let U be a normal
neighborhood of p and let r be the radial distance function on U . Then the following
inequality holds on U0 X fpg:
sc0 .r/
r .n 1/ ; (11.27)
sc .r/
where sc is defined by (10.8), and U0 D U if c 0, while U0 D fq 2 U W r.q/ < Rg

if c D 1=R2 > 0.
Proof. Let q 2 U0 X fpg be arbitrary, and let W Œ0; b ! U0 be the unit-speed radial
geodesic from p to q. We will show that (11.27) holds at .t/ for 0 < t b.
Because covariant differentiation commutes with the trace operator (since it is
just a particular kind of contraction), the Riccati
equation
(11.7) for Hr implies the
following scalar equation along for r D tr Hr :
d
r C tr Hr2 C tr R 0 D 0: (11.28)
dt
We need to analyze the last two terms on the left-hand side. We begin with the
last term. In terms of any local orthonormal frame .Ei /, we have
n
X n n
˝ ˛ X ˝ ˛ X
tr R 0 D R 0 .Ei /; Ei D R.Ei ; 0 / 0 ; Ei D Rm.Ei ; 0 ; 0 ; Ei /
iD1 iD1 iD1
D Rc. 0 ; 0 /:
To analyze the Hr2 term, let us set
r
HV r D Hr r : (11.29)
n1
We compute
r r .r/2 2
tr HV 2r D tr Hr2 tr.Hr ı r / tr.r ı Hr / C tr r :
n1 n1 .n 1/2

To simplify this, note that r2 D r because r is a projection, and thus tr r2 D
tr.r / D n 1. Also, Hr .@r / 0 implies that Hr ı r D Hr ; and since Hr is self-
adjoint, hHr v; @r ig D hv; Hr @r ig D 0 for all v, so the image of Hr is contained in
the orthogonal complement of @r , and it follows
that r ı Hr D Hr as well. Thus
the last three terms in the formula for tr HV 2r combine to yield
.r/2
tr HV 2r D tr Hr2 :
n1

Solving this for tr Hr2 , substituting into (11.28), and dividing by n 1, we obtain

d r r 2 tr HV 2r C Rc. 0 ; 0 /
C C D 0: (11.30)
dt n1 n1 n1
Let H.t / D sc0 .t /=sc .t /, so that
H 0 .t / C H.t /2 C c 0:
We wish to apply the 1 1 case of the matrix Riccati comparison theorem (Thm.
11.6) with H.t / as above, S.t / c,
ˇ ˇ
.r/ˇ.t/ tr HV 2r ˇ.t/ C Rc. 0 .t /; 0 .t //
Hz .t / D ; and S.t/z D :
n1 n1
Note that HV 2r is positive semidefinite, which means that all of its eigenvalues are
nonnegative, so its trace (which is the sum of the eigenvalues) is also nonnegative.
(This is the step that does not work in the case of an upper bound on Ricci curvature.)
Thus our hypothesis on the Ricci curvature guarantees that S.t/ z c for all t 2 .0; b.
To apply Theorem 11.6, we need to verify that Sz has a continuous exten-
sion to Œ0; b and that H z .t / H.t / has a nonnegative limit as t & 0. Recall
that we showed in (11.19) and (11.20) that sc0 .r/=sc .r/ D 1=r C O.r/ and Hr D
.1=r/r C O.r/ as r & 0. This implies that r D tr.Hr / D .n 1/=r C O.r/,
and therefore both HV r j.t/ and H z .t / H.t / approach 0 and Sz.t / approaches
Rc. 0 .0/; 0 .0//=.n 1/ c as t & 0. Therefore, we can apply Theorem 11.6 to
conclude that H z .t / H.t / for t 2 .0; b. Since .b/ D q was arbitrary, this com-
pletes the proof. t
u
The next theorem and its two corollaries will be crucial ingredients in the proofs
of our theorems in the next chapter about manifolds with positive Ricci curvature
(see Thms. 12.28 and 12.24).
Theorem 11.16 (Conjugate Point Comparison II). Let .M; g/ be a Riemannian n-

manifold, and suppose there is a positive constant c D 1=R2 such that the Ricci cur-
vature of M satisfies Rc.v; v/ .n 1/c for all unit vectors v. Then every geodesic
segment of length at least R has a conjugate point.
Proof. Let U be a normal neighborhood of an arbitrary point p 2 M . The second

Laplacian comparison theorem (Thm. 11.15) combined with Lemma 11.13 shows
that
p s 0 .r/
@r log r n1 det g D r .n 1/ c D @r log sc .r/n1
sc .r/
n1
p
on the subset U0 Upwhere r < R. Since r det g=sc .r/n1 ! 1 as r & 0,
n1 n1
this implies that r det g=sc .r/ 1 everywhere in U0 , or equivalently,
p
det g sc .r/n1 =r n1 : (11.31)
Suppose U contains a point q where r R, and let W Œ0; b ! U be the unit-speed
radial geodesic from p to q. Because sc .R/ D 0, (11.31) shows that det g..t // !
0 as t % R, and therefore by continuity det g D 0 at .b/, which contradicts the
fact that det g > 0 in every coordinate neighborhood. The upshot is that no normal
neighborhood can include points where r R.
Now suppose W Œ0; b ! M is a unit-speed geodesic with b R, and assume
for the sake of contradiction that has no conjugate points. Let pD.0/ and vD 0 .0/,
so .t / D expp .t v/ for t 2 Œ0; b. As in the proof of Theorem 11.9, because has no
conjugate points, we can choose a star-shaped open subset W Tp M containing
the set L D ft v W 0 t < b0 g Tp M on which expp is a local diffeomorphism,
and let gz be the pulled-back metric expp g on W , which satisfies the same curvature
estimates as g. Then z.t / D t v is a radial gz-geodesic in W of length greater than or
equal to R, which contradicts the argument in the preceding paragraph. t
u
Corollary 11.17 (Injectivity Radius Comparison). Let .M; g/ be a Riemannian

n-manifold, and suppose there is a positive constant c D 1=R2 such that the Ricci
curvature of M satisfies Rc.v; v/ .n 1/c for all unit vectors v. Then for every
point p 2 M , we have inj.p/ R.
Proof. Every radial geodesic segment in a geodesic ball is minimizing, but the pre-
ceding theorem shows that no geodesic segment of length R or greater is minimiz-
ing. Thus no geodesic ball has radius greater than R. t
u
Corollary 11.18 (Diameter Comparison). Let .M; g/ be a complete, connected

Riemannian n-manifold, and suppose there is a positive constant c D 1=R2 such
that the Ricci curvature of M satisfies Rc.v; v/ .n 1/c for all unit vectors v.
Then the diameter of M is less than or equal to R.
Proof. This follows from the fact that any two points of M can be connected by a
minimizing geodesic segment, and the conjugate point comparison theorem implies
that no such segment can have length greater than R. t
u
Our final comparison theorem is a powerful volume estimate under the assump-
tion of a lower bound on the Ricci curvature. We will use it in the proof of Theorem
12.28 in the next chapter, and it plays a central role in many of the more advanced
results of Riemannian geometry.
A weaker version of this result was proved by Paul Günther in 1960 [Gün60]
for balls within the injectivity radius under the assumption of a lower bound on
sectional curvature; it was improved by Richard L. Bishop in 1963 (announced in
[Bis63], with a proof in [BC64]) to require only a lower Ricci curvature bound; and
then it was extended by Misha Gromov in 1981 [Gro07] to cover all metric balls in
the complete case, not just those inside the injectivity radius.
Theorem 11.19 (Bishop–Gromov Volume Comparison). Let .M; g/ be a con-

nected Riemannian n-manifold, and suppose there is a constant c such that the
Ricci curvature of M satisfies Rc.v; v/ .n 1/c for all unit vectors v. Let p 2 M
be given, and for every positive number ı, let Vg .ı/ denote the volume of the metric
ball of radius ı about p in .M; g/, and let Vc .ı/ denote the volume of a metric ball
of radius ı in the n-dimensional Euclidean space, hyperbolic space, or sphere with
constant sectional curvature c. Then for every 0 < ı inj.p/, we have
Vg .ı/ Vc .ı/; (11.32)
and the quotient Vg .ı/=Vc .ı/ is a nonincreasing function of ı that approaches 1 as

ı & 0. If .M; g/ is complete, the same is true for all positive ı, not just ı inj.p/.
In either case, if equality holds in (11.32) for some ı, then g has constant sectional
curvature on the entire metric ball of radius ı about p.
Proof. First consider ı inj.p/, in which case a metric ball of radius ı in M is

actually a geodesic ball. With the exception of the first and last paragraphs, the proof
of Theorem 11.14 goes through with all of the inequalities reversed, and with the
first Laplacian comparison theorem replaced by its counterpart Theorem 11.15, to
show that Vg .ı/=Vc .ı/ is a nonincreasing function of ı that approaches 1 as ı & 0,
and (11.32) follows.
In case M is complete, expp is defined and smooth on all of Tp M . Theorem
10.34 shows that Cut.p/ has measure zero in M and expp maps the open subset
ID.p/ Tp M diffeomorphically onto the complement of Cut.p/ in M . Therefore,

for every ı > 0, the metric ball of radius ı is equal to expp Bı .0/ \ ID.p/ up to
a set of measure zero, where Bı .0/ denotes the ı-ball about 0 in Tp M . Using an
orthonormal basis to identify .Tp M; gp / with .Rn ; gx/, we can compute the volume
of a metric ı-ball as
Z Z p
Vg .ı/ D expp d Vg D det g d Vgx ;
ID.p/\Bı .0/ ID.p/\Bı .0/
where det g denotes the determinant of the matrix of g in the normal coordinates
determined by the choice of basis.
Let ˚ W RC Sn1 ! Tp M X f0g Š Rn X f0g be the map ˚.; !/ D ! as in
Corollary 10.17, and define zsc W RC ! Œ0; 1/ by zsc ./ D sc ./ if c 0, while in
the case c D 1=R2 > 0,

(
sc ./; < R;
zsc ./ D
0; R:
Corollary 11.18 shows that the cut time of every unit vector in Tp M is less than or
equal to R. Thus zsc ./ > 0 whenever ˚.; !/ 2 ID.p/, and we can define z W RC
n1
S ! R by
p
det g ı ˚.; !/n1
z !/ D ; ˚.; !/ 2 ID.p/;
.; zsc ./n1
0; ˚.; !/ … ID.p/;
and just as in the proof of Theorem 11.14, we can write

Z Z ı
Vg .ı/ D z !/zsc ./n1 d d V ;
.; gV
Sn1 0
Z Z ı
Vc .ı/ D sc ./n1 d d VgV :
z
Sn1 0
The arguments of Theorem 11.14 (with inequalities reversed) show that for each
z !/ is nonincreasing in for all positive , and it follows
! 2 Sn1 , the function .;
just as in that proof that Vg .ı/=Vc .ı/ (now interpreted as a ratio of volumes of
metric balls) is nonincreasing for all ı > 0 and approaches 1 as ı & 0, and (11.32)
follows.
Finally, suppose Vg .ı/ D Vc .ı/ for some ı > 0, and assume first that ı inj.p/.
An argument exactly analogous to the one at the end of the proof of Theorem 11.14
shows that .; !/ 1 everywhere on the set where 0 < < ı. Combined with
Lemma 11.13, this implies that
sc0 .r/
r D .n 1/ (11.33)
sc .r/
everywhere on Bı .p/ X fpg. This means that along each unit-speed radial geodesic
, the function u.t / D .r/j.t/ =.n 1/ D sc0 .t /=sc .t / satisfies u0 C u2 C c 0 by
direct computation. Comparing this to (11.30), we conclude that

tr HV 2r C Rc. 0 ; 0 /
c
n1

on Bı .p/Xfpg. Since
tr HV 2r 0 and Rc. 0 ; 0 / .n1/c everywhere, this is pos-
sible only if tr HV 2r vanishes identically there. Because HV 2r is positive semidefinite
and its trace is zero, it must vanish identically, which by definition of HV r means
r s 0 .r/
Hr D r D c r :
n1 sc .r/
Proposition 11.3 then shows that g has constant sectional curvature c on Bı .p/.
Now suppose .M; g/ is complete. The argument of Theorem 11.14 then shows
z !/ 1 everywhere on the set where 0 < < ı and zsc ./ > 0. In view of
that .;
z this implies in particular that ID.p/ \ Bı .0/ contains all of the
the definition of ,
points in Bı .0/ where zsc .r/ > 0. In case c 0, zsc .r/ D sc .r/ > 0 everywhere, so
ID.p/ \ Bı .0/ D Bı .0/ and therefore the metric ball of radius ı around p is actually
a geodesic ball, and the argument above applies.
In case c D 1=R2 > 0, if ı R, then zsc .r/ D sc .r/ > 0 on Bı .0/, and once
again we conclude that the metric ı-ball is a geodesic ball. On the other hand, if
ı > R, then the diameter comparison theorem (Cor. 11.18) shows that the metric
ball of radius ı is actually the entire manifold. The fact that the volume ratio is
nonincreasing implies that Vg .R/ D Vc .R/, and the argument above shows that
g has constant sectional curvature c on the metric ball of radius R. Since the
closure of that ball is all of M , the result follows by continuity. t
u
The next corollary is immediate.
Corollary 11.20. Suppose .M; g/ is a compact Riemannian manifold and there is a

positive constant c D 1=R2 such that the Ricci curvature of M satisfies Rc.v; v/
.n 1/c for all unit vectors v. Then the volume of M is no greater than the vol-
ume of the n-sphere of radius R with its round metric, and if equality holds, then
.M; g/ has constant sectional curvature c. t
u
(For explicit formulas for the volumes of n-spheres, see Problem 10-4.)
Problems
11-1. Let .M; g/ be a Riemannian manifold, and let U be a normal neighborhood

of p 2 M . Use
Corollary 6.42 to show that in every choice of polar normal
coordinates 1 ; : : : ; n1 ; r on a subset of U X fpg (see Example 6.45), the
covariant Hessian of r is given by
n1
1 X
r 2r D @r g˛ˇ d ˛ d ˇ :
2
˛;ˇ D1
11-2. Prove the following extension to Proposition 11.2: Suppose P is an embed-

ded submanifold of a Riemannian manifold .M; g/, U is a normal neigh-
borhood of P in M , and r is the radial distance function for P in U (see
Prop. 6.37). If W Œ0; b ! U is a geodesic segment with .0/ 2 P and 0 .0/
normal to P , and J is a Jacobi field along that is transverse to P in the
sense of Problem 10-14, then Dt J.t / D Hr .J.t // for all t 2 Œ0; b.
Problems 343
11-3. Let .M; g/ be a Riemannian manifold, and let f be any smooth local dis-
tance function defined on an open subset U M . Let F D grad f (so the
integral curves of F are unit-speed geodesics), and let Hf D rF (the Hes-
sian operator of f ). Show that Hf satisfies the following Riccati equation
along each integral curve of F :
Dt Hf C Hf2 C R 0 D 0; (11.34)
where R 0 .w/ D R.w; 0 / 0 . [Hint: Let W be any smooth vector field on U ,

and evaluate rF rW F in two different ways.]
11-4. Let .M; g/ be a compact Riemannian manifold. Prove that if R; L are posi-
tive numbers such that all sectional curvatures of M are less than or equal to
1=R2 and all closed geodesics have lengths greater than or equal to L, then

inj.M / min R; 12 L :
[Hint: Assume not, and use the result of Problem 10-23(b).]

11-5. T RANSVERSE JACOBI F IELD C OMPARISON T HEOREM : Let P be an em-
bedded hypersurface in a Riemannian manifold .M; g/. Suppose W Œ0; b !
M is a unit-speed geodesic segment with .0/ 2 P and 0 .0/ normal to
P , and J is a normal Jacobi field along that is transverse to P . Let
D h.J.0/; J.0//, where h is the scalar second fundamental form of P with
respect to the normal 0 .0/. Let c be a real number, and let u W R ! R be
the unique solution to the initial value problem
u00 .t / C cu.t / D 0;
u.0/ D 1; (11.35)
u0 .0/ D :
In the following statements, the principal curvatures of P are computed with

respect to the normal 0 .0/.
(a) If all sectional curvatures of M are bounded above by c, all principal
curvatures of P at .0/ are bounded below by , and u.t / ¤ 0 for t 2
.0; b/, then jJ.t /j u.t /jJ.0/j for all t 2 Œ0; b.
(b) If all sectional curvatures of M are bounded below by c, all principal
curvatures of P at .0/ are bounded above by , and J.t / ¤ 0 for t 2
.0; b/, then jJ.t /j u.t /jJ.0/j for all t 2 Œ0; b.
[Hint: Mimic the proof of Theorem 11.9, using the results of Problems 11-3
and 11-2.]
11-6. Suppose P is an embedded hypersurface in a Riemannian manifold .M; g/
and N is a unit normal vector field along P . Suppose the principal curva-
tures of P with respect to N are bounded below by a constant , and the
sectional curvatures of M are bounded above by 2 . Prove that P has no
focal points along any geodesic segment with initial velocity Np for p 2 P .
11-7. Suppose P is an embedded hypersurface in a Riemannian manifold .M; g/

and N is a unit normal vector field along P . Suppose the sectional curva-
tures of M are bounded below by a constant c, and the principal curvatures
of P with respect to N are bounded above by a constant . Let u be the
solution to the initial value problem (11.35). Prove that if b is a positive real
number such that u.b/ D 0, then P has a focal point along every geodesic
segment with initial velocity Np for some p 2 P and with length greater
than or equal to b.
Chapter 12
Curvature and Topology
In this final chapter, we bring together most of the tools we have developed so far to
prove some significant local-to-global theorems relating curvature and topology of
Riemannian manifolds.
We focus first on constant-curvature manifolds. The main result here is the
Killing–Hopf theorem, which shows that every complete, simply connected mani-
fold with constant sectional curvature is isometric to one of our frame-homogeneous
model spaces. A corollary of the theorem then shows that the ones that are not sim-
ply connected are just quotients of the models by discrete groups of isometries. The
technique used to prove this result also leads to a global characterization of sym-
metric spaces in terms of parallel curvature tensors.
Next we turn to nonpositively curved manifolds. The primary result is the
Cartan–Hadamard theorem, which topologically characterizes complete, simply
connected manifolds with nonpositive sectional curvature: they are all diffeomor-
phic to Rn . After proving the main result, we prove two other theorems, due to
Cartan and Preissman, that place severe restrictions on the fundamental groups of
complete manifolds with nonpositive or negative curvature.
Finally, we address the case of positive curvature. The main theorem is Myers’s
theorem, which says that a complete manifold with Ricci curvature bounded below
by a positive constant must be compact and have a finite fundamental group and
diameter no larger than that of the sphere with the same Ricci curvature. The bor-
derline case is addressed by Cheng’s maximal diameter theorem, which says that the
diameter bound is achieved only by a round sphere. These results are supplemented
by theorems of Milnor and Synge, which further restrict the topology of manifolds
with nonnegative Ricci and positive sectional curvature, respectively.
Manifolds of Constant Curvature

Our first major local-to-global theorem is a global characterization of complete man-
ifolds of constant curvature. If .M; g/ has constant sectional curvature, Corollary
346 12 Curvature and Topology
Fig. 12.1: Analytic continuation of a local isometry
10.15 shows that each point of M has a neighborhood that is isometric to an open
subset of one of the constant-curvature model spaces of Chapter 3. In order to turn
that into a global result, we use a technique modeled on the theory of analytic con-
tinuation in complex analysis.

Suppose .M; g/ and M ; gy are Riemannian manifolds of the same dimension
and ' W U ! M is a local isometry defined on some connected open subset U M .
If W Œ0; 1 ! M is a continuous path such that .0/ 2 U , then an analytic continua-
tion of ' along is a family of pairs f.Ut ; 't / W t 2 Œ0; 1g, where Ut is a connected
neighborhood of .t / and 't W Ut ! M is a local isometry, such that '0 D ' on
U0 \ U , and for each t 2 Œ0; 1 there exists ı > 0 such that jt t1 j < ı implies that
.t1 / 2 Ut and that 't agrees with 't1 on Ut \ Ut1 (see Fig. 12.1). Note that we
require 't and 't1 to agree where they overlap only if t and t1 are sufficiently close;
in particular, if is not injective, the values of the analytic continuation may dif-
fer at different times at which returns to the same point. However, as the next
lemma shows, all analytic continuations along the same path will end up with the same
value.

Lemma 12.1 (Uniqueness of Analytic Continuation). With .M; g/, M ; gy , and '
as above, if f.Ut ; 't /g and f.Ut0 ; 't0 /g are two analytic continuations of ' along the
same path W Œ0; 1 ! M , then '1 D '10 on a neighborhood of .1/.
Proof. Let T be the set of all t 2 Œ0; 1 such that 't D 't0 on a neighborhood of .t /.
Then 0 2 T , because both '0 and '00 agree with ' on a neighborhood of .0/. We
will show that T is open and closed in Œ0; 1, from which it follows that 1 2 T , which
proves the lemma.
To see that it is open, suppose t 2 T . By the definition of analytic continuation,
there is some ı > 0 such that if t1 2 Œ0; 1 and jt1 t j < ı, then 't1 and 't01 both agree
with 't on a neighborhood of .t /, so Œ0; 1 \ .t ı; t C ı/ T . Thus T is open.
To see that it is closed, suppose t is a limit point of T . There is a sequence ti ! t
such that ti 2 T , which means that 'ti D 't0i on a neighborhood of ti for each i .
By definition of analytic continuation, for i large enough, we have .ti / 2 Ut \ Ut0
Manifolds of Constant Curvature 347
and 't D 'ti D 't0i D 't0 on a neighborhood of .ti /. In particular, this means that
't ..ti // D 't0 ..ti // and d.'t /.ti / D d.'t0 /.ti / for each such i . By continuity,
therefore, 't ..t // D 't0 ..t // and d.'t /.t/ D d.'t0 /.t/ . Proposition 5.22 shows
that 't D 't0 on a neighborhood of .t /, so t 2 T . t
u
The previous lemma shows that all analytic continuations along the same path
end up with the same value. But when we consider analytic continuations along dif-
ferent paths, this may not be the case. The next theorem gives a sufficient condition
for analytic continuations along two different paths to result in the same value.
Theorem
12.2 (Monodromy Theorem for Local Isometries). Let .M; g/ and
M ; gy be connected Riemannian manifolds. Suppose U is a neighborhood of a
point p 2 M and ' W U ! M is a local isometry that can be analytically continued
along every path starting at p. If 0 ; 1 W Œ0; 1 ! M are path-homotopic paths from
p to a point q 2 M , then the analytic continuations along 0 and 1 agree in a
neighborhood of q.
Proof. Suppose 0 and 1 are path-homotopic, and let H W Œ0; 1 Œ0; 1 ! M be a

path homotopy from 0 to 1 . Write Hs .t / D H.t; s/, so that H0 D 0 , H1 D 1 ,
and Hs .0/ D p and Hs .1/ D q for all s 2 Œ0; 1. The hypothesis implies that for
each s 2 Œ0; 1, there is an analytic continuation of ' along Hs , which we denote by
˚ .s/ .s/
Ut ; 't W t 2 Œ0; 1 .
Consider the function P W Œ0; 1 ! M given by P .s/ D ' .s/ .q/. Given s 2 Œ0; 1,
1
if s 0 is sufficiently close to s, it follows from compactness of Hs .Œ0; 1/ that the
˚
same family Ut.s/ ; 't.s/ W t 2 Œ0; 1 serves as an analytic continuation along Hs 0 .
This means that P .s/ is locally constant as a function of s, so '1.1/ .q/ D '1.0/ .q/.
Since a path from p to q can easily be modified near q to yield a path from p to
any point q 0 sufficiently nearby, and the same analytic continuation works for the
modified path, it follows from the same argument that '1.1/ .q 0 / D '1.0/ .q 0 / for all q 0
in a neighborhood of q. t
u

Corollary 12.3. Let .M; g/ and M ; gy be simply connected, complete Riemannian
manifolds. Suppose U is a connected neighborhood of a point p 2 M and ' W U !
M is a local isometry that can be analytically continued along every path starting
at p. Then there is a global isometry ˚ W M ! M whose restriction to U agrees
with '.
Proof. Let q 2 M be arbitrary. We wish to define ˚.q/ to be the value of an analytic

continuation of ' along a path from p to q. The fact that M is simply connected
means that all paths from p to q are path-homotopic, so the monodromy theorem
implies that all such analytic continuations agree in a neighborhood of q; thus ˚
is well defined. Because it is a globally defined local isometry and M is complete,
it follows from Theorem 6.23 that ˚ is a Riemannian covering map. Since M is
simply connected, every covering of M is bijective by Corollary A.59, so ˚ is a
bijective local isometry and thus a global isometry. t
u
Fig. 12.2: Proof of the Killing–Hopf theorem
The next theorem, due to Wilhelm Killing and Heinz Hopf, is the main result of
this section.
Theorem 12.4 (Killing–Hopf). Let .M; g/ be a complete, simply connected Rie-

mannian n-manifold with constant sectional curvature, n 2. Then M is isometric
to one of the model spaces Rn , Sn .R/, and Hn .R/.

Proof. Given .M; g/ satisfying the hypotheses, let M ; gy be the Euclidean space,
sphere, or hyperbolic space with the same constant sectional curvature as M . Note
that M is simply connected in each case. Let p 2 M be arbitrary. Corollary 10.15
shows that there exist an open subset V M containing p, an open subset Vy M ,
and an isometry ' W V ! Vy . If we can show that ' can be analytically continued
along every path starting at p, then we can conclude from Corollary 12.3 that there
exists a global isometry ˚ W M ! M , thus proving the theorem.
To that end, let W Œ0; 1 ! M be a path starting at p. Corollary 10.15 shows that
for each t 2 Œ0; 1, there is a neighborhood of .t / that is isometric to an open subset
of M . After shrinking the neighborhoods if necessary, we may assume by Theorem
6.17 that each such neighborhood is a convex geodesic ball. By compactness, we
can choose a partition .t0 ; : : : ; tkC1 / of Œ0; 1 and a finite sequence of such balls
B0 ; : : : ; Bk such that .Œti ; tiC1 / Bi for each i D 0; : : : ; k. After shrinking B0 and
adding more points to the partition if necessary, we can also assume B0 V .
We wish to construct local isometries i W Bi ! M such that successive ones
agree on overlaps. Begin with 0 D 'jB0 . By our choice of B1 , there is some
isometry ˇ1 from B1 to an open subset of M . Note that B0 \ B1 is an intersec-
tion of geodesically convex sets and thus is connected, and it is nonempty because
it contains .t1 /. The composition 0 ı ˇ11 is an isometry from ˇ1 .B0 \ B1 / to
0 .B0 \ B1 /, both of which are connected open subsets of M , so by Problem

5-11(c), it is the restriction of a global isometry W M !M (Fig. 12.2). Let

1 D ı ˇ1 W B1 ! M . Our choice of guarantees that 1 jB0 \B1 D 0 jB0 \B1 .
Proceeding similarly by induction, we can define i W Bi ! M y for i D 2; : : : ; k, such

that i agrees with i1 on Bi \ Bi1 .
Now we define the analytic continuation of ' as follows: start with .Ut ; 't / D
.B0 ; 0 / for 0 t t1 , and thereafter let .Ut ; 't / D .Bi ; i / for ti < t tiC1 . It is
easy to verify that this family of maps does the job. t
u
Corollary 12.5 (Characterization of Constant-Curvature Manifolds). The com-
plete, connected, n-dimensional Riemannian manifolds of constant sectional curva-
= , where
ture are, up to isometry, exactly the Riemannian quotients of the form M

M is one of the constant-curvature n n n
model spaces R , S .R/, or H .R/, and is

a discrete subgroup of Iso M that acts freely on M
.
Proof. First suppose .M; g/ is a complete, connected Riemannian n-manifold with

constant sectional curvature, and let W M ! M be its universal covering manifold

with the pullback metric gz D g. The preceding theorem shows that M ; gz is
isometric to one of the model spaces, so we might as well take it to be one of
the models. Proposition C.20 shows that the group of covering automorphisms
acts freely and properly on M , and Corollary 2.33 shows that M is isometric to

M = . Moreover, if ' is any covering automorphism, then ı ' D , and so' gz D
' g D g D gz; thus acts isometrically, so it is a subgroup of Iso M .
To show that is discrete, suppose f'i g is an infinite set with an accumula-
tion point in Iso.M /. (Note that Problem 5-11 shows that Iso M ; gz is a Lie group
acting smoothly on M in each case, so it makes sense to talk about the topology of
the isometry group.) Since the action of is free, for every point pz 2 M the set
z is infinite, and by continuity of the action it has an accumulation point in
f'i .p/g
M . But this is impossible, since the points f'i .p/g
z all project to the same point in
M under the covering map , and so form a closed discrete set. Thus is discrete
in Iso.M /.
Conversely, suppose M is one of the model spaces and is a discrete subgroup

of Iso M that acts freely on M . Then is a closed Lie subgroup of Iso M by
Proposition C.9, so the result of Problem 6-28 shows that it acts properly on M . It

then follows from Proposition 2.32 that M = is a smooth manifold and has a unique
Riemannian metric such that the quotient map W M !M = is a Riemannian

covering, and M = is complete by Corollary 6.24. Since a Riemannian covering is
in particular a local isometry, M = has constant sectional curvature. t
u
A complete, connected Riemannian manifold with constant sectional curvature
is called a space form. A space form is called spherical, Euclidean, or hyperbolic,
depending on whether its sectional curvature is positive, zero, or negative, respec-
tively. The preceding corollary, combined with the characterization of the isometry
groups of the simply connected model spaces given in Problem 5-11, essentially
reduces the classification of space forms to the group-theoretic problem of classify-
ing the discrete subgroups of E.n/, O.n C 1/, and OC .n; 1/ that act without fixed
points. Nevertheless, the group-theoretic problem is still far from easy.
The spherical space forms were classified in 1972 by Joseph Wolf [Wol11]; the
proof is intimately connected with the representation theory of finite groups. Al-
though in even dimensions there are only spheres and projective spaces (see Problem
12-2), the situation in odd dimensions is far more complicated. Already in dimen-
sion 3 there are many interesting examples: some notable ones are the lens spaces
obtained as quotients of S3 C 2 by cyclic groups rotating the two complex coordi-
nates through different angles, and the quotients of SO.3/ (which is diffeomor-
phic to RP 3 and is therefore already a quotient of S3 ) by the dihedral groups,
the symmetry groups of regular 3-dimensional polyhedra.
The complete classification of Euclidean space forms is known only in low
dimensions. Problem 12-3 shows that every compact 2-dimensional Euclidean space
form is diffeomorphic to the torus or the Klein bottle. It turns out that there are 10
classes of nondiffeomorphic compact Euclidean space forms of dimension 3, and 75
classes in dimension 4. The fundamental groups of compact Euclidean space forms
are examples of crystallographic groups—discrete groups of Euclidean isometries
with compact quotients, which have been studied extensively by physicists as well
as geometers. (A quotient of Rn by a crystallographic group is a space form, pro-
vided it is a manifold, which is true whenever the group is torsion-free.) In higher
dimensions, the classification is still elusive. The main things that are known are
two results proved by Ludwig Bieberbach in the early twentieth century: (1) every
Euclidean space form is a quotient of a flat torus by a finite group of isometries,
and (2) in each dimension, there are only finitely many diffeomorphism classes of
Euclidean space forms. See [Wol11] for proofs of the Bieberbach theorems and a
complete survey of the state of the art.
In addition to the question of classifying Euclidean space forms up to diffeomor-
phism, there is also the question of classifying the different flat Riemannian metrics
on a given manifold up to isometry. Problem 12-5 shows, for example, that there is
a three-parameter family of nonisometric flat Riemannian metrics on the 2-torus.
Finally, the study of hyperbolic space forms is a vast and rich subject in its own
right. A good introduction is the book [Rat06].
Symmetric Spaces Revisited
The technique of analytic continuation of local isometries can also be used to derive
a fundamental global result about symmetric spaces. Theorem 10.19 showed that
locally symmetric spaces are characterized by having parallel curvature. The next
theorem is a global version of that result.
Theorem 12.6. Suppose .M; g/ is a complete, simply connected Riemannian mani-

fold with parallel Riemann curvature tensor. Then M is a symmetric space.
Proof. Let p be an arbitrary point of M . Theorem 10.19 shows that M is locally

symmetric, so there is a point reflection ' W U ! U defined on some connected
neighborhood U of p. If we can show that ' can be analytically continued along
every path starting at p, then Corollary 12.3 shows that ' extends to a global isom-
etry ˚ W M ! M ; because it agrees with ' on U , it satisfies d ˚p D d'p D Id, so
it is a point reflection.
Let W Œ0; 1 ! M be a path starting at p. By compactness, we can find a partition

.t0 ; : : : ; tkC1 / of Œ0; 1 and convex geodesic balls B0 ; : : : ; Bk such that .Œti ; tiC1 /
Bi for each i D 0; : : : ; k, with B0 chosen small enough that B0 U . As in the
proof of Theorem 12.4, we will construct local isometries i W Bi ! M such that
successive ones agree on overlaps.
Begin with 0 D 'jB0 , and let p1 D .t1 / and q1 D 0 .p1 /. Because 0 is a
local isometry, the linear map d 0 jp1 W Tp1 M ! Tq1 M satisfies d 0 jp1 gq1 D
gp1 and d 0 jp1 Rmq1 D Rmp1 . Lemma 10.18 then shows that there is a local
isometry 1 W B1 ! M that satisfies 1 .p1 / D 0 .p1 / and d 1 jp1 D d 0 jp1 , and it
follows from Proposition 5.22 that 1 and 0 agree on the connected open set B0 \
B1 . Proceeding by induction, we obtain local isometries i W Bi ! M satisfying
i jBi 1 \Bi D i1 jBi 1 \Bi . Then the analytic continuation of ' is defined just as
in the last paragraph of the proof of Theorem 12.4, thus completing the proof. t
u
Corollary 12.7. Suppose .M; g/ is a complete, connected Riemannian manifold.

The following are equivalent:
(a) M has parallel curvature tensor.
(b) M is a locally symmetric space.
(c) M is isometric to a Riemannian quotient M = , where M is a .globally/ sym-
metric space and is a discrete Lie group that acts freely, properly, and iso-
.
metrically on M
Proof. Theorem 10.19 shows that (a) , (b). To show that (a) ) (c), assume that
M has parallel curvature tensor, and let M ; gz be its universal covering manifold
with the pullback metric. Since the covering map W M ! M is a local isometry,

M also has parallel curvature tensor, and Corollary 6.24 shows that it is complete.
Then Theorem 12.6 shows that M is a symmetric space. If we let denote the covering
automorphism group with the discrete topology, then Proposition C.20 shows that
is a discrete Lie group acting smoothly, freely, and properly on M , and it acts
isometrically because the pullback metric is invariant under all covering automor-
phisms. Finally, Corollary 2.33 shows that M is isometric to M = .
Finally, we show that (c) ) (a). If M is isometric to a Riemannian quotient of the
form M = as in (c), then the quotient map M !M = is a Riemannian covering

by Proposition 2.32. Since M has parallel curvature tensor by the result of Problem
7-3, so does M . t
u
An extensive treatment of the structure and classification of symmetric spaces
can be found in [Hel01].
The analytic continuation technique used in this section was introduced in 1926
by Élie Cartan, in the same paper [Car26] in which he first defined symmetric spaces
and proved many of their properties. The technique was generalized in 1956 by War-
ren Ambrose [Amb56] to allow for isometries between more general Riemannian
manifolds with varying curvature; and then in 1959 it was generalized further by
Noel Hicks [Hic59] to manifolds with connections in their tangent bundles, not nec-
essarily Riemannian ones. The resulting theorem is known as the Cartan–Ambrose–
Hicks theorem. We do not pursue it further, but a statement and proof can be found
in [CE08].
Manifolds of Nonpositive Curvature
Our next major local-to-global theorem provides a complete topological characteri-

zation of complete, simply connected manifolds of nonpositive sectional curvature.
This was proved in 1928 by Élie Cartan, generalizing earlier proofs for surfaces by
Hans Carl Friedrich von Mangoldt and Jacques Hadamard. As with many other such
theorems, tradition has given it a title that is not entirely historically accurate, but is
generally recognized by mathematicians.
Theorem 12.8 (Cartan–Hadamard). If .M; g/ is a complete, connected Riemann-

ian manifold with nonpositive sectional curvature, then for every point p 2 M , the
map expp W Tp M ! M is a smooth covering map. Thus the universal covering
space of M is diffeomorphic to Rn , and if M is simply connected, then M itself is
diffeomorphic to Rn .
Proof. By Theorem 11.12, the assumption of nonpositive curvature guarantees that

p has no conjugate points along any geodesic. Therefore, by Proposition 10.20,
expp is a local diffeomorphism on all of Tp M .
Let gz be the (variable-coefficient) 2-tensor field expp g defined on Tp M . Be-
cause expp is a local diffeomorphism, gz is a Riemannian metric, and expp is a local

isometry from Tp M; gz to .M; g/. Note that each line t 7! t v in Tp M is a gz-
geodesic, so .Tp M; gz/ is complete by Corollary 6.20. It then follows from Theorem
6.23 that expp is a smooth covering map. The remaining statements of the theorem
follow immediately from uniqueness of the universal covering space. t
u
Because of this theorem, a complete, simply connected Riemannian manifold
with nonpositive sectional curvature is called a Cartan–Hadamard manifold. The
basic examples are Euclidean and hyperbolic spaces. The next two propositions
show that Cartan–Hadamard manifolds share many basic geometric properties with
these model spaces. A line in a Riemannian manifold is the image of a nonconstant
geodesic that is defined on all of R and restricts to a minimizing segment between
any two of its points.
Proposition 12.9 (Basic Properties of Cartan–Hadamard Manifolds). Suppose

.M; g/ is a Cartan–Hadamard manifold.
(a) The injectivity radius of M is infinite.
(b) The image of every nonconstant maximal geodesic in M is a line.
(c) Any two distinct points in M are contained in a unique line.
(d) Every open or closed metric ball in M is a geodesic ball.
(e) For every point q 2 M , the function r.x/ D dg .q; x/ is smooth on M X fqg and
r.x/2 is smooth on all of M .
Proof. These properties follow from Lemma 6.8, Proposition 6.11, and Corollaries
6.12 and 6.13, together with the fact that M is a normal neighborhood of each of its
points. t
u
Manifolds of Nonpositive Curvature 353
Suppose A; B; C are three points in a Cartan–Hadamard manifold that are non-

collinear (meaning that they are not all contained in a single line). The geodesic
triangle 4ABC determined by the points is the union of the images of the (unique)
geodesic segments connecting the three points. (In the present context, we do not
require that a geodesic triangle bound a two-dimensional region, as we did in
Chapter 9.) If 4ABC is a geodesic triangle, we denote the angle in TA M formed
by the initial velocities of the geodesic segments from A to B and A to C by †A
(or †CAB if necessary to avoid ambiguity), and similarly for the other angles.
Proposition 12.10. Suppose 4ABC is a geodesic triangle in a Cartan–Hadamard

manifold .M; g/, and let a; b; c denote the lengths of the segments opposite the ver-
tices A, B, and C , respectively. The following inequalities hold:
(a) c 2 a2 C b 2 2ab cos †C .
(b) †A C †B C †C .
If the sectional curvatures of g are all strictly negative, then strict inequality holds
in both cases.

u
Another consequence of the Cartan–Hadamard theorem is that there are strin-
gent topological restrictions on which manifolds can carry metrics of nonpositive
sectional curvature. The next corollary is immediate.
Corollary 12.11. No simply connected compact manifold admits a metric of non-

positive sectional curvature. t
u
With a little more work, we can obtain a much more powerful result.
Corollary 12.12. Suppose M and N are positive-dimensional compact, connected

smooth manifolds, at least one of which is simply connected. Then M N does not
admit any Riemannian metric of nonpositive sectional curvature.
Proof. Suppose for the sake of contradiction that M is simply connected and g is a
metricon M N with nonpositive sectional curvature. If Nz is the universal covering
manifold of N , then there is a universal covering map W M Nz ! M N , and
g is a complete metric of nonpositive sectional curvature on M Nz . The Cartan–
Hadamard theorem shows that M Nz is diffeomorphic to a Euclidean space.
Since M is simply connected, it is orientable (Cor. B.19). Let be a smooth
orientation m-form for M , where m D dim M . Then is closed because there are
no nontrivial
R .m C 1/-forms on M , but Stokes’s theorem shows that is not exact,
because M > 0. On the other hand, if we let p W M Nz ! M denote the projec-
tion on the first factor, then the form p is exact on M Nz , because every closed
m-form on a Euclidean space is exact.
Choose an arbitrary point y0 2 Nz and define W M ! M Nz by .x/ D .x; y0 /;
it is a diffeomorphism onto its image .M / D M fy0 g, which is a compact em-
bedded submanifold of M Nz . By diffeomorphism invariance of the integral,
Z Z Z Z
p D .p / D .p ı / D > 0;
.M / M M M
R
because p ı D Id. But p is exact on M Nz , which implies .M / p

D 0, a
contradiction. t
u
Example 12.13. By the preceding corollary, every metric on Sm Sn must have posi-
tive sectional curvature somewhere, provided m 2 and n 1, because both spheres
are compact and Sm is simply connected. //
With a little more algebraic topology, one can obtain more information. A topo-
logical space whose higher homotopy groups k .M / all vanish for k > 1 is called
aspherical.
Corollary 12.14. If M is a smooth manifold that admits a metric of nonpositive

sectional curvature, then M is aspherical.
Proof. Suppose M admits a nonpositively curved metric, so its universal covering

space is diffeomorphic to Rn by Cartan–Hadamard. Since Rn is contractible, it is
aspherical. Since covering maps induce isomorphisms on k for k > 1 (see [Hat02,
Prop. 4.1]), it follows that M is aspherical as well. t
u
Cartan’s Torsion Theorem
In addition to the topological restrictions on nonpositively curved manifolds im-

posed directly by the Cartan–Hadamard theorem, it turns out that there is a stringent
restriction on the fundamental groups of such manifolds (Corollary 12.18 below).
In preparation for the proof, we need the following two lemmas, both of which are
interesting in their own right.
Lemma 12.15. Suppose .M; g/ is a Cartan–Hadamard manifold. Given q 2 M , let

f W M ! Œ0; 1/ be the function f .x/ D 12 dg .x; q/2 . Then f is strictly geodesically
convex, in the sense that for every geodesic segment W Œ0; 1 ! M , the following
inequality holds for all t 2 .0; 1/:
f ..t // < .1 t /f ..0// C tf ..1//: (12.1)
Proof. We can write f .x/ D 12 r.x/2 , where r is the radial distance function from
q with respect to any normal coordinates, and Proposition 12.9 shows that f is
smooth on all of M . The Hessian comparison theorem implies that Hr .1=r/r
on M X fqg, and therefore
Hf D r.grad f / D r.r grad r/ D grad r ˝ dr C rHr grad r ˝ dr C r :
The expression on the right-hand side above is actually equal to the identity operator,
as can be checked by applying it separately to @r D grad r and an arbitrary vector
perpendicular to @r . Therefore hHf .v/; vi jvj2 for all tangent vectors to M X fqg,
and by continuity, the same holds at q as well.
Now suppose W Œ0; 1 ! M is a geodesic segment. By the chain rule,
d
f ..t // D df.t/ . 0 .t // D hgrad f j.t/ ; 0 .t /i;
dt
and thus (using the fact that 0 is parallel along )
d2
f ..t // D hr 0 .t/ grad f j.t/ ; 0 .t /i D hHf . 0 .t //; 0 .t /i > 0:
dt 2
Therefore, the function f ı is strictly convex on Œ0; 1, and (12.1) follows. t
u
Lemma 12.16. Suppose .M; g/ is a Cartan–Hadamard manifold and S is a com-

pact subset of M containing more than one point. Then there is a unique closed ball
of minimum radius containing S .
Proof. Because S is compact, it is bounded. Let ı D diam.S /, and let q0 be any

point of S , so that S Bxı .q0 /. Let c0 be the infimum of the radii of closed metric
balls containing S , and let Bxci .pi / be a sequence of closed balls containing S
such that ci & c0 as i ! 1. For i large enough that ci < 2ı, the fact that q0 2 S
Bxci .pi / implies dg .pi ; q0 / ci < 2ı, so each of the points pi lies in the compact
ball Bx2ı .q0 /. After passing to a subsequence, we may assume that pi converges to
some point p0 2 M .
To show that S Bxc0 .p0 /, suppose q 2 S , and let " > 0 be arbitrary. If i is large
enough that dg .pi ; p0 / "=2 and ci c0 C "=2, then
dg .q; p0 / dg .q; pi / C dg .pi ; p0 / ci C "=2 c0 C ":
Since " was arbitrary, this shows that dg .q; p0 / c0 , and thus S Bxc0 .p0 /.
Now we need to show that Bxc0 .p0 / is the unique closed ball of radius c0 contain-
ing S . Suppose to the contrary that Bxc0 .p00 / is another such
ball. Let W Œ0; 1 ! M
be the geodesic segment from p0 to p00 , and let m D 12 be its midpoint. Let q 2 S
be arbitrary, and let f W M ! Œ0; 1/ be the function f .x/ D 12 dg .q; x/2 . Lemma
12.15 implies that

1
d .q; m/2 D f .m/ < 12 f .p0 / C 12 f .p00 / 12 12 c02 C 12 12 c02 D 12 c02 :
2 g
Thus for every q 2 S , we have dg .q; m/ < c0 . Since S is compact, the contin-
uous function dg .; m/ takes on a maximum value b0 < c0 on S , showing that
S Bxb0 .m/. This contradicts the fact that c0 is the minimum radius of a ball con-
taining S . t
u
Let us call the center of the smallest enclosing ball the 1-center of the set S .
(The term is borrowed from optimization theory, where the “1-center problem” is
the problem of finding a location for a single production facility that minimizes
the maximum distance to any client. Some Riemannian geometry texts refer to the
1-center of a set S M as its “circumcenter” or “center of gravity,” but these terms
seem inappropriate, because the 1-center does not coincide with the classical mean-
ing of either term for finite subsets of the Euclidean plane.)
Theorem 12.17 (Cartan’s Fixed-Point Theorem). Suppose .M; g/ is a Cartan–

Hadamard manifold and G is a compact Lie group acting smoothly and isomet-
rically on M . Then G has a fixed point in M , that is, a point p0 2 M such that
' p0 D p0 for all ' 2 G.
Proof. Let q0 2 M be arbitrary, and let S D G q0 be the orbit of q0 . If S contains

only the point q0 , then q0 is a fixed point, so we may assume that the orbit contains
at least two points. Because S is the image of the continuous map from G to M
given by ' 7! ' q0 , it is compact, so by Lemma 12.16, there is a unique smallest
closed geodesic ball containing S . Let p0 be the center of this ball (the 1-center of
S ), and let c0 be its radius.
Now let '0 2 G be arbitrary, and note that '0 S is the set of all points of the
form '0 ' q0 for ' 2 G. As ' ranges over all of G, so does '0 ', so this image set
is exactly the orbit of q0 . In other words, '0 S D S .
Because G acts by isometries, '0 Bxc0 .p0 / D Bxc0 .'0 p0 /. Thus Bxc0 .'0 p0 / is
a closed ball of the same radius c0 containing '0 S D S , so by uniqueness of the
1-center we must have '0 p0 D p0 . Since '0 was arbitrary, this shows that p0 is a
fixed point of G. t
u
Cartan’s fixed-point theorem has important applications to Lie theory (where it is
used to prove that all maximal compact subgroups of a Lie group are
conjugate to each other). But our interest in the theorem is that it leads to the next
corollary, also due to Cartan. If G is a group, an element ' 2 G is called a torsion
element if ' k D 1 for some integer k > 1. A group is said to be torsion-free if the
only torsion element is the identity.
Corollary 12.18 (Cartan’s Torsion Theorem). Suppose .M; g/ is a complete, con-

nected Riemannian manifold with nonpositive sectional curvature. Then 1 .M / is
torsion-free.

Proof. Let M ;gz be the universal covering manifold of M with the pullback met-
ric. Then M ; gz is a Cartan–Hadamard manifold, and M is isometric to a Riemann-

ian quotient M = , where is subgroup of Iso M ; gz , isomorphic to 1 .M / and
acting freely on M .
Suppose ' is a torsion element of . Then the cyclic group generated by ' is
finite, and thus is a compact 0-dimensional Lie group with the discrete topology,
acting smoothly on M because isometries are smooth. Cartan’s fixed-point theorem
shows that it has a fixed point. Because acts freely, this implies that ' is the
identity. Thus is torsion-free, and the same is true of 1 .M /. t
u
Preissman’s Theorem
Taken together, the Cartan–Hadamard theorem and Cartan’s torsion theorem put
severe restrictions on the possible topologies of manifolds that can carry nonposi-
tively curved metrics. If we make the stronger assumption of strictly negative sec-
tional curvature, we can restrict the topology even further. The following theorem
was first proved in 1943 by Alexandre Preissman [Pre43].
Theorem 12.19 (Preissman). If .M; g/ is a compact, connected Riemannian man-
ifold with strictly negative sectional curvature, then every nontrivial abelian sub-
group of 1 .M / is isomorphic to Z.
Before proving Preissman’s theorem, we need two more lemmas. Suppose .M; g/
is a complete Riemannian manifold and ' W M ! M is an isometry. A geodesic
W R ! M is called an axis for ' if ' restricts to a nontrivial translation along ,
that is, if there is a nonzero constant c such that '..t // D .t C c/ for all t 2 R.
An isometry with no fixed points that has an axis is said to be axial.
Example 12.20 (Axial Isometries). In R2 with its Euclidean metric, every nontrivial
translation .x; y/ 7! .x C a; y C b/ is axial, and every line parallel to the vector
a@x C b@y is an axis. In the upper half-plane model of the hyperbolic plane, every
dilation .x; y/ 7! .cx; cy/ for c ¤ 1 is axial, and the geodesic .t / D .0; e t / is an
axis. On the other hand, the horizontal translations .x; y/ 7! .x C a; y/ are not axial
for the hyperbolic plane, because no geodesic is invariant under such an isometry. //
As the next lemma shows, there is a close relation between axes and closed
geodesics. This lemma will also prove useful in the proof of Synge’s theorem later
in the chapter.
Lemma 12.21. Suppose .M; g/ is a compact, connected Riemannian manifold, and
WM ! M is its universal covering manifold endowed with the metric gz D g.
Then every covering automorphism of has an axis, which restricts to a lift of a
closed geodesic in M that is the shortest admissible path in its free homotopy class.
Proof.
Let ' be any nontrivial covering automorphism of , so it is also an isometry
of M ; gz . Every continuous path z W Œ0; 1 ! M from a point xz 2 M to its image

' xz projects to a loop D ı z in M . We begin by showing that the set of all
loops obtained from ' in this way is exactly one free homotopy class. Suppose
0 ; 1 W Œ0; 1 ! M are two loops, so that i D ı zi for each i , where zi is
such
a path in M from xzi to ' xzi (Fig. 12.3). For i D 0; 1, let xi D xzi , so i is a
at xi .Choose
loop based z from xz0 to xz1 , and let ž D ' ı ˛
a path ˛ z , which is a path
from ' xz0 to ' xz1 , and ˛ D ı ˛ z D ı ž, a path from x0 to x1 in M. Because
M is simply connected, the two paths ˛ z1 z0 ž and z1 from xz1 to ' xz1 are path-
homotopic, and therefore so are their images ˛ 1 0 ˛ and 1 . An easy argument
shows that the loops ˛ 1 0 ˛ and 0 are freely homotopic, by a homotopy that
gradually shrinks the “tail” ˛ to a point, so it follows that 0 and 1 are freely
homotopic.
Fig. 12.3: Defining a free homotopy class
Conversely, let 0 and 1 be freely homotopic loops based at x0 and x1 in M ,

respectively, and suppose one of them, say 0 , is of the form 0 D ı z0 for some
path z0 from a point xz0 to ' xz0 in M . We need to show that 1 is also of this form. The
fact that the loops are freely homotopic means that there is a homotopy H W Œ0; 1
Œ0; 1 ! M satisfying H.s; 0/ D 0 .s/, H.s; 1/ D 1 .s/, and H.0; t / D H.1; t / for
all t . Let ˛ W Œ0; 1 ! M be the path ˛.t / D H.0; t / D H.1; t / from x0 to x1 . The
existence of such a homotopy implies that the two loops ˛ 1 0 ˛ and 1 based
at x1 are path-homotopic (see [LeeTM, Lemma 7.17]). By the monodromy theorem
(Prop. A.54(c)), their lifts starting at xz1 both end at the same point. Let z1 be the
lift of 1 starting at the point xz1 D ˛ z .1/, let ˛
z be the lift of ˛ starting at xz0 , and
let ž D 'ı ˛ z 1 z0 ž is a lift of ˛ 1 0 ˛ starting at xz1 , and it ends at
z. Then ˛
ž.1/ D ' xz1 . The monodromy theorem therefore implies that z1 is a path from xz1

to ' xz1 , as claimed.
We have shown that the covering automorphism ' determines a unique free
homotopy class in M . It is not the trivial class, because that class contains the con-
stant loop, which cannot be the image of a path from a point xz to ' xz because
' has no fixed points. Since M is compact, Proposition 6.28 shows that there is
a closed geodesic W Œ0; 1 ! M that is the shortest admissible path in this free
homotopy class. This means, in particular, that there is a lift z of that satisfies
z.1/ D ' z.0/ . Because M is complete, z extends to a geodesic in M defined
on all of R. Let z0 ; z1 W R ! M be the geodesics defined by z0 .t / D ' ı z.t / and
z1 .t / D z.t C 1/. Then z0 .0/ D z.1/ D z1 .0/, and
Fig. 12.4: Uniqueness of the axis in the negative curvature case

d z.1/ z00 .0/ D d z.1/ ı d'z.0/ z0 .0/ (definition of z0 /

D d z.0/ z0 .0/ (because ı ' D /
0
D .0/ (because ı z D /
D 0 .1/ . is a closed geodesic)

D d z.1/ z0 .1/ . ı z D again)

D d z.1/ z10 .0/ (definition of z1 /:
The fact that d z.1/ is injective then implies z00 .0/ 0

D z1 .0/, so by uniqueness of
geodesics we have z0 z1 , or in other words ' z.t/ D z.t C 1/ for all t 2 R.
Thus z is an axis for '. t
u
In general, an axis of an isometry need not be unique, as we saw in the case
of translations of R2 in Example 12.20. However, as the next lemma shows, the
situation is different in the case of negatively curved Cartan–Hadamard manifolds.
Lemma 12.22. Suppose .M; g/ is a Cartan–Hadamard manifold with strictly neg-

ative sectional curvature. If ' W M ! M is an axial isometry, then its axis is unique
up to reparametrization.
Proof. Let ' W M ! M be an axial isometry, and suppose 1 ; 2 W R ! M are both

axes for '. After reparametrizing the geodesics if necessary, we can assume that '
translates both geodesics by time 1.
Suppose first that the two axes do not intersect. Then the points A D 1 .0/, B D
1 .1/ D '.A/, C D 2 .0/, and D D 2 .1/ D '.C / are all distinct (Fig. 12.4). Let
be the geodesic segment from A to C , so that ' ı is the geodesic segment from B
to D, forming a “geodesic quadrilateral” ABDC . Because ' preserves angles, the
angles †CAB and †ABD are supplementary, as are †ACD and †CDB. Thus the
angle sum of ABDC is exactly 2.
On the other hand, Proposition 12.10 shows that the geodesic triangles 4ABC
and 4BCD have angle sums strictly less than . The angle †ACD is no larger
than the sum †ACB C †BCD: one way to see this is to note that Problem 6-2
shows that the angle between two unit vectors v; w 2 TC M , namely arccoshv; wi,
is exactly equal to the Riemannian distance between v and w regarded as points on
the unit sphere in TC M with its round metric, so we can apply the triangle inequality
for distances on the sphere. Similarly, †ABD †ABC C †CBD. Thus the sum
of the four angles of ABDC is strictly less than 2, which is a contradiction.
It follows that 1 and 2 must intersect at some point, say x D 1 .t1 / D 2 .t2 /.
But then '.x/ D 1 .t1 C1/ D 2 .t2 C1/ is another intersection point, so Proposition
12.9(c) shows that 2 must be a reparametrization of 1 . t
u
Proof of Preissman’s Theorem. Suppose .M; g/ satisfies the hypotheses of the the-

orem,
and let W M ! M be its universal covering, with the metric gz D g. Then

M ; gz is a Cartan–Hadamard manifold with strictly negative curvature. Because
the fundamental group of M based at any point is isomorphic to the group of cov-
ering automorphisms of (see Prop. C.22), it suffices to show that every nontrivial
abelian group of covering automorphisms is isomorphic to Z. Let H be such a
group.
Let ' be any non-identity element of H , and let W R ! M be the axis for ',
which we may give a unit-speed parametrization. Thus there is a nonzero constant c
such that '..t // D .t C c/ for all t 2 R. If is any other element of H , then for
all t 2 R we have
'. ..t /// D .'..t /// D ..t C c//;
which shows that ı is also an axis for '. Since axes are unique by Lemma
12.22, this means that ı is a (unit-speed) reparametrization of . It cannot be a
backward reparametrization (one satisfying ' ı .t / D .t C a/), because then
would be a nontrivial covering automorphism having .a=2/ as a fixed point, while
the only covering automorphism with a fixed point is the identity. Thus there must
be some a 2 R such that ..t // D .t C a/ for all t 2 R.
Define a map F W H ! R as follows: for each 2 H , let F . / be the unique
constant a such that ..t // D .t C a/ for all t 2 R. A simple computation shows
that F . 1 ı 2 / D F . 1 / C F . 2 /, so F is a group homomorphism. It is injective
because F . / D 0 implies that fixes every point in the image of , so it must be
the identity. Thus F .H / is an additive subgroup of R isomorphic to H .
Now let b be the infimum of the set of positive elements of F .H /. If b D 0, then
there exists 2 H such that 0 < F . / < inj.M /, and then ı is a unit-speed
geodesic in M satisfying ı .a/ D ı .0/, where a D F . /, which contradicts
the fact that 0 < a < inj.M /. Thus b > 0. If b … F .H /, there is a sequence of
elements i 2 H such that F . i / & b, and then for sufficiently large i < j we
have 0 < F . i ı j1 / < inj.M /, which leads to a contradiction as before. Thus b
is the smallest positive element of F .H /, and then it is easy to verify that F .H /
consists exactly of all integral multiples of b, so it is isomorphic to Z. t
u
The most important consequence of Preissman’s theorem is the following corol-
lary.
Corollary 12.23. No product of positive-dimensional connected compact manifolds
admits a metric of strictly negative sectional curvature.
u
Manifolds of Positive Curvature 361
Manifolds of Positive Curvature
Finally, we consider manifolds with positive curvature. The most important fact
about such manifolds is the following theorem, which was first proved in 1941 by
Sumner B. Myers [Mye41], building on earlier work of Ossian Bonnet, Heinz Hopf,
and John L. Synge.
Theorem 12.24 (Myers). Let .M; g/ be a complete, connected Riemannian n-
manifold, and suppose there is a positive constant R such that the Ricci curvature of
M satisfies Rc.v; v/ .n 1/=R2 for all unit vectors v. Then M is compact, with
diameter less than or equal to R, and its fundamental group is finite.
Proof. The diameter comparison theorem (Cor. 11.18) shows that the diameter of
M is no greater than R. To show that M is compact, we just choose a base point
p and note that every point in M can be connected to p by a geodesic segment
of length at most R. Therefore, expp W BxR .0/ ! M is surjective, so M is the
continuous image of a compact set.
To prove the statement about the fundamental group, let W M ! M denote the

universal covering space of M , with the metric gz D g. Then M ; gz is complete
by Corollary 6.24, and gz also has Ricci curvature satisfying the same lower bound,
so M is compact by the argument above. By Proposition A.61, for every p 2 M
there is a one-to-one correspondence between 1 .M; p/ and the fiber 1 .p/.
If 1 .M; p/ were infinite, therefore, 1 .p/ would be an infinite discrete closed
, contradicting the compactness of M
subset of M . Thus 1 .M; p/ is finite. t
u
In case the manifold is already known to be compact, the hypothesis can be
relaxed.
Corollary 12.25. Suppose .M; g/ is a compact, connected Riemannian n-manifold
whose Ricci tensor is positive definite everywhere. Then M has finite fundamental
group.
Proof. Because the unit tangent bundle of M is compact by Proposition 2.9, the
hypothesis implies that there is a positive constant c such that Rc.v; v/ c for all
unit tangent vectors v. Thus the hypotheses of Myers’s theorem are satisfied with
.n 1/=R2 D c. t
u
On the other hand, it is possible for complete, noncompact manifolds to have
strictly positive Ricci or even sectional curvature, as long as the curvature gets arbi-
trarily close to zero, as the following example shows.
I Exercise 12.26. Let n 2, and let M RnC1 be the paraboloid f.x 1 ; : : : ; x n ; y/ W
y D jxj2 g with the induced metric (see Problem 8-1). Show that M has strictly positive
sectional and Ricci curvatures everywhere, but is not compact.
One of the most useful applications of Myers’s theorem is to Einstein metrics.

Corollary 12.27. If .M; g/ is a complete Einstein manifold with positive scalar cur-
vature, then M is compact.
Proof. If g is Einstein with positive scalar curvature, then Rc D n1 Sg satisfies the

hypotheses of Myers’s theorem with .n 1/=R2 D n1 S . u
t
In 1975, Shiu-Yuen Cheng proved the following theorem [Che75], showing that
the diameter inequality in Myers’s theorem is an equality only for a round sphere.
Theorem 12.28 (Cheng’s Maximal Diameter Theorem). Suppose .M; g/ satisfies

the hypotheses of Myers’s theorem and diam.M / D R. Then .M; g/ is isometric
to Sn .R/; gV R .
Proof. By Myers’s theorem, M is compact, and thus by continuity of the distance

function there are points p1 ; p2 2 M such that dg .p1 ; p2 / D diam.M / D R. For
i D 1; 2 and 0 < ı R, let Bı .pi / denote the open metric ball in M of radius
ı about pi , and let BV ı denote a metric ball in Sn .R/; gV R of the same
radius.
The Bishop–Gromov theorem shows that the ratio Vol.Bı .pi //=Vol BV ı is a non-
increasing function of ı for each i , which implies the following inequality for
0 < ı < R:

Vol.Bı .pi // Vol.Bı .pi // Vol BV ı Vol BV ı
D D : (12.2)
Vol.M / Vol.BR .pi // Vol BV R Vol.Sn .R//
Now suppose ı1 ; ı2 are positive numbers such that ı1 C ı2 D R. The fact that
dg .p1 ; p2 / D R together with the triangle inequality implies that Bı1 .p1 / and
Bı2 .p2 / are disjoint, so
Vol.Bı1 .p1 // C Vol.Bı2 .p2 // Vol.M /: (12.3)

On the other hand, Vol BV ı1 C Vol BV ı2 D Vol.Sn .R// (think of balls centered at
the north and south poles), so (12.2) and (12.3) imply

Vol.Bı1 .p1 // C Vol.Bı2 .p2 // Vol BV ı1 C Vol BV ı2
1 D 1:
Vol.M / Vol.Sn .R//
Thus the inequalities above are equalities, so in particular equality holds in (12.3).
Moreover, the boundary of a metric ball has measure zero because it is contained in
the image of a sphere under the exponential map, so we also have

Vol Bxı1 .p1 / C Vol Bxı2 .p2 / D Vol.M /: (12.4)
Next we will show that dg .p1 ; q/ C dg .p2 ; q/ D dg .p1 ; p2 / D R for all q 2 M .

Suppose not: by the triangle inequality, the only way this can fail is if there is a
point q such that dg .p1 ; q/ C dg .p2 ; q/ > R. Choosing ı1 < dg .p1 ; q/ and ı2 <
dg .p2 ; q/ such that ı1 C ı2 D R, we see that q is not in the closed set Bxı1 .p1 / [
Bxı2 .p2 /, so there is a neighborhood of q disjoint from that set (Fig. 12.5). But any
such neighborhood would have positive volume, contradicting (12.4).
Suppose W Œ0; 1/ ! M is a unit-speed geodesic starting at p1 , and choose " > 0
small enough that jŒ0;" is minimizing. Then dg .p1 ; ."// D ", and there is also a
Fig. 12.5: A step in the proof of the maximal diameter theorem
minimizing unit-speed geodesic 2 from ."/ to p2 , whose length is dg .p2 ; ."// D

R ". The piecewise regular curve consisting of jŒ0;" followed by a unit-speed
reparametrization of 2 is an admissible curve from p1 to p2 of length R, and
is therefore minimizing; thus it is a (smooth) geodesic, and in fact coincides with
, which is therefore minimizing on Œ0; R. This shows that the cut time of every
unit-speed geodesic starting at p1 is at least R, so the injectivity radius inj.p1 / is
at least R; and then the injectivity radius comparison theorem (Cor. 11.17) shows
that it is exactly R. Thus M X fp2 g D BR .p1 / is a geodesic ball around p1 . An
analogous argument shows that M X fp1 g is a geodesic ball around p2 .
The upshot of these observations is that the distance functions r1 .q/ D dg .p1 ; q/
and r2 .q/ D dg .p2 ; q/ are both smooth on M X fp1 ; p2 g, and because their sum is
constant, they satisfy r1 D r2 . Elementary trigonometric identities show that
the function sc defined by (10.8) with c D 1=R satisfies sc .R t / D sc .t / and
sc0 .R t / D sc0 .t /, so the Laplacian comparison theorem (Thm. 11.15) applied to
r1 and r2 D R r1 yields
sc0 .r1 / s 0 .r2 /

r1 .n 1/ D .n 1/ c r2 D r1 :
sc .r1 / sc .r2 /
Thus both inequalities are equalities. Arguing exactly as in the last part of the proof
of the Bishop–Gromov theorem, we conclude that .M; g/ has constant sectional
2
curvature 1=R
n, and thus by Corollary 12.5 it admits a (finite-sheeted) Riemannian
covering by S .R/; gV R . Since the injectivity domain of p1 is the entire ball of ra-
dius R, on which the metric has the form (10.17) in normal coordinates, it follows
from Theorem 10.34 that the volume of M is equal to that of Sn .R/; gV R . Problem
2-14 then shows that the covering is one-sheeted, so it is a global isometry. t
u
If we weaken the hypothesis to require merely that the Ricci curvature be nonneg-
ative instead of positive definite, there are still nontrivial topological consequences.
Let be a finitely generated group. We say that has polynomial growth if there
exist a finite generating set S and positive real numbers C; k such that for every
positive integer m, the number of distinct elements of that can be expressed as
products of at most m elements of S and their inverses is no larger than C mk . The
order of polynomial growth of such a group is the infimum of such values of k. It
is easy to check that Zn has polynomial growth of order n; in fact, using the funda-
mental theorem on finitely generated abelian groups [DF04, p. 158], one can show
that every finitely generated abelian group has polynomial growth of order equal to
its free rank.
The following theorem, proved in 1968 by John Milnor [Mil68], shows in effect
that fundamental groups of complete manifolds with nonnegative Ricci curva-
ture are not too far from being abelian, at least when they are finitely generated.
(Milnor conjectured in the same paper that the fundamental group of such a mani-
fold is always finitely generated. This is true in the compact case, because compact
manifolds are homotopy equivalent to finite CW complexes [Hat02, Cor. A.12],
and these have finitely generated fundamental groups, as shown, for example, in
[LeeTM, Thm. 10.15]. But the conjecture has not been proved in the noncompact
case.) By contrast, Milnor proved in the same paper that the fundamental group of
a negatively curved compact manifold never has polynomial growth, and that was
extended in 1970 by André Avez [Ave70] to the case of a nonpositively curved
compact manifold as long as it is not flat.
Theorem 12.29 (Milnor). Suppose .M; g/ is a complete, connected Riemannian

n-manifold with nonnegative Ricci curvature. Every finitely generated subgroup of
1 .M / has polynomial growth of order at most n.

u
Corollary 12.30. If M is a connected smooth manifold whose fundamental group

is free on more than one generator, then M admits no complete Riemannian metric
with nonnegative Ricci curvature.
I Exercise 12.31. Prove this corollary.
The corollary implies, for example, that a plane with two or more punctures does
not admit a complete metric with nonnegative Gaussian curvature.
On the other hand, if we strengthen the hypothesis and assume positive sectional
curvature, then we can reach a stronger conclusion. The following theorem was
proved in 1936 by John L. Synge [Syn36].
Theorem 12.32 (Synge). Suppose .M; g/ is a compact, connected Riemannian

n-manifold with strictly positive sectional curvature.
(a) If n is even and M is orientable, then M is simply connected.
(b) If n is odd, then M is orientable.
Proof. Suppose for the sake of contradiction that the appropriate hypothesis holds
! M be the universal covering manifold of
but the conclusion is false. Let W M

M with the pullback metric gz D g. We can give M the pullback orientation in the
even-dimensional case; in the odd-dimensional case, just choose an orientation on

M . In either case, there is a nontrivial covering automorphism ' W M !M ; in the
even-dimensional case, it is orientation-preserving, and in the odd-dimensional case
we can choose it to be orientation-reversing. By Lemma 12.21, there is a geodesic
z W R ! M that is an axis for ', and D ı z restricts to a closed geodesic in M
that minimizes length in its free
homotopy class. After reparametrizing if necessary,
we may assume that
' z
.t/ Dz.t C 1/ for all t 2 R. Let xz0 D z.0/, xz1 D z.1/ D
' xz0 , and x D xz0 D xz1 .
e W Txz M
Let P ! Txz M be parallel transport along zjŒ0;1 , and let P W Tx M !
0 1
Tx M be parallel transport along jŒ0;1 . Because local isometries preserve paral-
lelism, the following diagram commutes, and all four maps are linear isometries:
Pz-
Txz0 M Txz1 M
d xz0 d xz1
? P- ?
Tx M Tx M:
Because d Vgz applied to a parallel frame is constant, PQ is orientation-preserving. In

the even-dimensional case, the vertical maps are both orientation-preserving, so P
is too. In the odd-dimensional case, the fact that d xz0 D d xz1 ı d'xz0 implies that
the two vertical maps induce opposite orientations on Tx M , so P is orientation-
reversing.
The upshot is that in each case, P is a linear isometry of Tx M whose determinant
is .1/n . Note that the determinant is the product of the eigenvalues of P , counted
with multiplicities. Since the real eigenvalues of a linear isometry are ˙1, and the
others come in conjugate pairs whose product is 1, the equation det P D .1/n
implies that the multiplicity of 1 has the same parity as n, and therefore the multi-
plicity of C1 must be even. We know that 0 .0/ is a C1-eigenvector, so there must
be another independent C1-eigenvector v 2 Tx M , which we can take to be orthog-
onal to 0 .0/. Extending v by parallel transport yields a nontrivial parallel normal
vector field V 2 X. / satisfying V .0/ D V .1/.
Consider the variation of defined by .s; t / D exp.t/ .sV .t //. Because V .0/ D
V .1/, this is a variation through admissible loops. The first variation formula (6.1)
shows that the first derivative of Lg .s / at s D 0 is zero, because is a geodesic and
the boundary terms cancel. We can then apply the general version of the second vari-
ation formula given in Problem 10-9, with the image of the geodesic s 7! expx .sv/
playing the role of M1 and M2 . Because this is a totally geodesic submanifold, its
second fundamental form vanishes, and we obtain
ˇ Z b
d 2 ˇˇ 2 0 0
L . / D jD V j Rm.V; ; ; V / dt:
ds 2 ˇsD0
g s t
a
Because V is parallel along and sec.V; 0 / > 0 for all t , this second derivative is
strictly negative, which implies that Lg .s / < Lg . / for sufficiently small s. But
this contradicts the fact that minimizes length in its free homotopy class. t
u
The next corollary shows that there are only two possibilities for fundamental
groups of positively curved compact manifolds in the even-dimensional case. (Com-
pare this to Problem 12-2, which shows that the only even-dimensional manifolds
that admit complete metrics of constant positive curvature are spheres and real pro-
jective spaces.)
Corollary 12.33. Suppose .M; g/ is a compact, connected, even-dimensional Rie-

mannian manifold with strictly positive sectional curvature. If M is orientable, then
1 .M / is trivial, and if not, then 1 .M / Š Z=2.
Proof. The orientable case is part of Synge’s theorem. If M is nonorientable, it has

a two-sheeted orientable covering manifold M ; Synge’s theorem shows that M is
simply connected, so it is the universal covering space of M . Because each fiber of
the universal covering has the same cardinality as 1 .M /, it follows that 1 .M / is
isomorphic to the two-element group Z=2. t
u
The preceding corollary implies, for example, that RP n RP n has no metric of
positive sectional curvature for any n 2, because its fundamental group is isomor-
phic to Z=2 Z=2.
For odd-dimensional compact manifolds, not many topological obstructions to
the existence of positively curved metrics are known, apart from orientability and
finiteness of 1 . And in the simply connected case, almost nothing is known. There
are many examples of simply connected compact manifolds that admit metrics of
nonnegative sectional curvature (for example, products of nonnegatively curved,
simply connected compact manifolds, by the result of Problem 8-10); but there is
not a single known example of a simply connected compact manifold that admits
a metric of nonnegative sectional curvature but not one of positive sectional curva-
ture. It was conjectured by Heinz Hopf in the early 1930s that S2 S2 admits no
positively curved metric, and it is reasonable to extend the conjecture to every prod-
uct of positive-dimensional simply connected compact manifolds that both admit
nonnegatively curved metrics; but nobody has come up with an example of such a
product for which the conjecture can be proved or disproved.
Further Results
We end the book with a brief look at some other local-to-global theorems about man-
ifolds with positive or nonnegative curvature, whose proofs are beyond our scope.
Some of the most powerful applications of comparison theory have been to prove
“pinching theorems.” Given a positive real number ı, a Riemannian manifold is said
to be ı-pinched if there exists a positive constant c such that all sectional curvatures
satisfy
ıc sec.˘ / c:
It is said to be strictly ı-pinched if at least one of the inequalities is strict. The
following celebrated theorem was originally proved by Marcel Berger and Wilhelm
Klingenberg in the early 1960s [Ber60, Kli61].
Theorem 12.34 (The Sphere Theorem). A complete, simply connected, Riemann-
ian n-manifold that is strictly 14 -pinched is homeomorphic to Sn .
This result is sharp, at least in even dimensions, because the Fubini–Study met-
rics on complex projective spaces are 14 -pinched (Problem 8-13).
For noncompact manifolds, there is the following remarkable theorem of Jeff
Cheeger and Detlef Gromoll [CG72].
Theorem 12.35 (The Soul Theorem). If .M; g/ is a complete, connected, noncom-
pact Riemannian manifold with nonnegative sectional curvature, then there exists a
compact totally geodesic submanifold S M (called a soul of M ) such that M is
diffeomorphic to the normal bundle of S in M . If M has strictly positive sectional
curvature, then S is a point and M is diffeomorphic to a Euclidean space.
Even if we assume only nonnegative Ricci curvature, something nontrivial can be
said in the noncompact case. The next theorem is also due to Cheeger and Gromoll
[CG71]. Recall that a line in a Riemannian manifold is the image of a nonconstant
geodesic defined on R whose restriction to each compact subinterval is minimizing.
Theorem 12.36 (The Splitting Theorem). If .M; g/ is a complete, connected, non-
compact Riemannian manifold with nonnegative Ricci curvature, and M contains a
line, then there is a Riemannian manifold .N; h/ with nonnegative Ricci curvature
such that M is isometric to the Riemannian product N R.
The proofs of all three of the above theorems, which can be found, for example,
in [Pet16], are elaborate applications of comparison theory.
All of our comparison theorems, and indeed most of the things we have proved
about Riemannian manifolds, are based on the analysis of ordinary differential
equations—the geodesic equation, the parallel transport equation, the Jacobi equa-
tion, and the Riccati equation. Using techniques of partial differential equations can
lead to much stronger conclusions in some cases. For instance, in 1982, Richard
Hamilton [Ham82] proved the following striking result about 3-manifolds.
Theorem 12.37 (Hamilton’s 3-Manifold Theorem). Suppose M is a simply con-
nected compact Riemannian 3-manifold with positive definite Ricci curvature. Then
M is diffeomorphic to S3 .
He followed it up four years later [Ham86] with an analogous result about 4-
manifolds, with Ricci curvature replaced by the curvature operator described in
Problem 8-33.
Theorem 12.38 (Hamilton’s 4-Manifold Theorem). Suppose M is a simply con-
nected compact Riemannian 4-manifold with positive curvature operator. Then M
is diffeomorphic to S4 .
Much more recently, Christoph Böhm and Burkhard Wilking [BW08] extended
this result to all dimensions.
Theorem 12.39 (Böhm–Wilking). Every simply connected compact Riemannian
manifold with positive curvature operator is diffeomorphic to a sphere.
The method of proof of these three sphere theorems was to start with an initial
metric g0 , and then define a one-parameter family of metrics fgt W t 0g evolving
according to the following partial differential equation, called the Ricci flow:
@
gt D 2Rct ;
@t
where Rct is the Ricci curvature of gt . Under the appropriate curvature assumption
on g0 , the metric gt can be rescaled to converge to a metric with constant positive
sectional curvature as t ! 1, and then it follows from the Killing–Hopf theorem
that the original manifold must be diffeomorphic to a sphere. Since Hamilton first
introduced it, this technique has been vastly generalized by Hamilton and others,
culminating in 2003 in a proof by Grigory Perelman of the Thurston geometrization
conjecture described in Chapter 3.
The Ricci flow has also been used to improve the original 14 -pinched sphere theo-
rem significantly. The techniques used to prove the original theorem were not strong
enough to prove that M is diffeomorphic to Sn , leaving open the possibility that
there might exist strictly 14 -pinched metrics on exotic spheres (topological spheres
with nonstandard smooth structures). In 2009, Simon Brendle and Richard Schoen
[BS09] were able to use Ricci-flow techniques to close this gap. They also were
able to prove the theorem under a weaker hypothesis: a Riemannian metric g is said
to be pointwise ı-pinched if for every p 2 M , there exists a positive number c.p/
such that
ıc.p/ sec.˘ / c.p/
for every 2-plane ˘ Tp M , and strictly pointwise ı-pinched if at least one of the
inequalities is strict at each point.
Theorem 12.40 (Differentiable Sphere Theorem). Let .M; g/ be a compact, sim-
ply connected Riemannian manifold of dimension n 4. If M is strictly pointwise
1
4
-pinched, then it is diffeomorphic to Sn with its standard smooth structure.
For a nice exposition of the proof, see the recent book [Bre10].
Problems
12-1. Suppose .M; g/ is a simply connected (but not necessarily complete) Rie-
mannian n-manifold with constant sectional curvature. Prove that there ex-
ists an isometric immersion from M into one of the model spaces Rn ,
Sn .R/, Hn .R/.
Problems 369
12-2. Prove that if n is even, then every orientation-preserving orthogonal linear map
from RnC1 to itself fixes at least one point of the unit sphere, and use this
to prove that every even-dimensional spherical space form is isometric to
either a round sphere or a real projective space with a constant multiple of
the metric defined in Example 2.34. (Used on p. 349.)
12-3. Show that every compact 2-dimensional Euclidean space form is diffeomor-
phic to the torus or the Klein bottle, and every noncompact one is diffeo-
morphic to R2 , S1 R2 , or the open Möbius band (see Example 2.35).
12-4. A lattice in Rn is an additive subgroup
Rn of the form

D fm1 v1 C C mn vn W m1 ; : : : ; mn 2 Zg;
for some basis .v1 ; : : : ; vn / of Rn . Let T n denote the n-torus.

(a) Show that if g is a flat Riemannian metric on T n , then T n ; g is iso-
metric to a Riemannian quotient of the form Rn =
for some lattice
,
acting on Rn by vector addition.
(b) Show that two such quotients Rn =
1 and Rn =
2 are isometric if and
only if
2 D A.
1 / for some A 2 O.n/.
12-5. C LASSIFICATION OF FLAT TORI : Let T 2 D S1 S1 denote the 2-torus.
Show that if g is a flat Riemannian metric on T 2 , then T 2 ; g is isometric
to one and only one Riemannian quotient R2 =
, where
is a lattice gener-
ated by a basis of the form ..a; 0/; .b; c//, with a > 0, 0 b a=2, c > 0,
and b 2 C c 2 a2 . [Hint: Given a lattice
R2 , let v1 be an element of

X f.0; 0/g of minimal norm; let v2 be an element of
X hv1 i of minimal
norm (where hv1 i is the cyclic subgroup generated by v1 ), chosen so that
the angle between v1 and v2 is less than or equal to =2; and then apply a
suitable orthogonal transformation.]
12-6. Suppose .M; g/ is a complete, connected Riemannian manifold, and p; q 2
M . Proposition 6.25 showed that every path-homotopy class of paths from
p to q contains a geodesic segment . Show that if M has nonpositive sec-
tional curvature, then is the unique geodesic segment in the given path
homotopy class.
12-7. Prove Proposition 12.10 (inequalities for triangles in Cartan–Hadamard
manifolds). [Hint: Compare with appropriate triangles in Euclidean space.]
12-8. Prove Corollary 12.23 (nonexistence of negatively curved metrics on com-
pact product manifolds).
12-9. Prove the following generalization of the Cartan–Hadamard theorem, due
to Robert Hermann [Her63]: Suppose .M; g/ is a complete, connected Rie-
mannian manifold with nonpositive sectional curvature, and P M is a
connected, properly embedded, totally geodesic submanifold. If for some
x 2 P , the homomorphism 1 .P; x/ ! 1 .M; x/ induced by the inclusion
P ,! M is surjective, then the normal exponential map E W NP ! M is a
diffeomorphism. [Hint: Mimic the proof of the Cartan–Hadamard theorem,

using the result of Problem 10-20.]
12-10. Give a counterexample to show that Myers’s theorem is false in dimensions
greater than 2 if we replace the lower bound on Ricci curvature by a positive
lower bound on scalar curvature.
12-11. Prove Theorem 12.29 (Milnor’s theorem on the fundamental group of a man-
ifold with nonnegative Ricci curvature) as follows: Let W M ! M be the
universal covering space of M , and give M the pullback metric gz D g.
Because the covering automorphism group is isomorphic to 1 .M / (Prop.
C.22), it suffices to prove the result for every finitely generated subgroup of the
automorphism group. Let be such a subgroup, and let S D f'1 ; : : : ; 'N g be a
finite generating set. Choose a base point p 2 M . For each i D 1; : : : ; N , let
di be the Riemannian distance from p to 'i .p/, and let D D max.d1 ; : : : ; dn /.
For each nonidentity element 2 , define the length of to be the small-
est integer m such can be expressed as a product of m elements of S and
their inverses.
(a) Show that there exists " > 0 such that for any two distinct elements
1 ; 2 2 , the metric balls B" . 1 .p// and B" . 2 .p// are disjoint.
(b) Show that if 2 has length m, then B" . .p// BmDC" .p/.
(c) Use the Bishop–Gromov theorem to prove that the number of distinct
elements of of length at most m is bounded by a constant times mn .
12-12. Prove that there is no compact manifold that admits both a metric of positive
definite Ricci curvature and a metric of nonpositive sectional curvature.
12-13. Suppose .M; g/ is a complete Riemannian manifold. Recall that a line in M
is the image of a nonconstant geodesic defined on all of R that is minimizing
between every pair of its points.
(a) Show that if .M; g/ has strictly positive sectional curvature, then M
contains no lines. [Hint: Given a geodesic W R ! M , let ˛ W R ! R be
a smooth positive function such that ˛.t / is smaller than the minimum
sectional curvature at .t / for each t , and let u W R ! R be the solution
to the initial value problem u00 C ˛u D 0 with u.0/ D 1 and u0 .0/ D 0.
Prove that there are numbers a < 0 < b such that u.a/ D u.b/ D 0, and
evaluate I.V; V / for a vector field of the form V .t / D u.t /E.t / along
jŒa;b , where E is a parallel unit normal vector field.]
(b) Give an example of a complete nonflat Riemannian manifold with non-
negative sectional curvature that contains a line.
Appendix A
Review of Smooth Manifolds
This book is designed for readers who already have a solid understanding of
basic topology and smooth manifold theory, at roughly the level of [LeeTM] and
[LeeSM]. In this appendix, we summarize the main ideas of these subjects that
are used throughout the book. It is included here as a review, and to establish our
notation and conventions.
Topological Preliminaries
An n-dimensional topological manifold (or simply an n-manifold) is a second-
countable Hausdorff topological space that is locally Euclidean of dimension n,
meaning that every point has a neighborhood homeomorphic to an open subset of
Rn . Given an n-manifold M , a coordinate chart for M is a pair .U; '/, where
U M is an open set and ' W U ! Uy is a homeomorphism from U to an open
subset Uy Rn . If p 2 M and .U; '/ is a chart such that p 2 U , we say that .U; '/
is a chart containing p.
We also occasionally need to consider manifolds with boundary. An n-dimen-
sional topological manifold with boundary is a second-countable Hausdorff space
in which every point has a neighborhood homeomorphic either to an open subset
of Rn or to a (relatively) open subset of the half-space RnC D f.x 1 ; : : : ; x n / 2 Rn W
x n 0g. A pair .U; '/, where U is an open subset of M and ' is a homeomorphism
from U to an open subset of Rn or RnC , is called an interior chart if '.U / is an
open subset of Rn or an open subset of RnC that does not meet @RnC D fx 2 RnC W
x n D 0g; and it is called a boundary chart if '.U / is an open subset of RnC with
'.U / \ @RnC ¤ ¿. A point p 2 M is called an interior point of M if it is in the
domain of an interior chart, and it is a boundary point of M if it is in the domain
of a boundary chart taking p to a point of @RnC .
Notice that our convention is that a manifold without further qualification is al-
ways assumed to be a manifold without boundary, and the term “manifold with
boundary” encompasses ordinary manifolds as a special case. But for clarity, we
in Mathematics 176, https://doi.org/10.1007/978-3-319-91755-9
372 Review of Smooth Manifolds
sometimes use redundant phrases such as “manifold without boundary” for an ordi-
nary manifold, and “manifold with or without boundary” when we wish to empha-
size that the discussion applies equally well in either case.
Proposition A.1. [LeeTM, Props. 4.23 & 4.64] Every topological manifold with or
without boundary is locally path-connected and locally compact.
Proposition A.2. [LeeTM, Thm. 4.77] Every topological manifold with or without
boundary is paracompact.
If M and N are topological spaces, a map F W M ! N is said to be a closed
map if for each closed subset K M , the image set F .K/ is closed in N , and an
open map if for each open subset U M , the image set F .U / is open in N . It is a
quotient map if it is surjective and V N is open if and only if F 1 .V / is open.
Proposition A.3. [LeeTM, Prop. 3.69] Suppose M and N are topological spaces,
and F W M ! N is a continuous map that is either open or closed.
(a) If F is surjective, it is a quotient map.
(b) If F is injective, it is a topological embedding.
(c) If F is bijective, it is a homeomorphism.
The next lemma often gives a convenient shortcut for proving that a map is
closed.
Lemma A.4 (Closed Map Lemma). [LeeTM, Lemma 4.50] If F is a continuous
map from a compact space to a Hausdorff space, then F is a closed map.
Here is another condition that is frequently useful for showing that a map is
closed. A map F W M ! N between topological spaces is called a proper map if
for each compact subset K N , the preimage F 1 .K/ is compact.
Proposition A.5 (Proper Continuous Maps Are Closed). [LeeTM, Thm. 4.95]
Suppose M is a topological space, N is a locally compact Hausdorff space, and
F W M ! N is continuous and proper. Then F is a closed map.
Fundamental Groups
Many of the deepest theorems in Riemannian geometry express relations between

local geometric properties and global topological properties. Because the global
properties are frequently expressed in terms of fundamental groups, we summarize
some basic definitions and properties here.
Suppose M and N are topological spaces. If F0 ; F1 W M ! N are continuous
maps, a homotopy from F0 to F1 is a continuous map H W M Œ0; 1 ! N that
satisfies H.x; 0/ D F0 .x/ and H.x; 1/ D F1 .x/ for all x 2 M . If there exists such a
homotopy, we say that F0 and F1 are homotopic.
We are most interested in homotopies in the following situation. If M is a topo-
logical space, a path in M is a continuous map ˛ W Œ0; 1 ! M . If p D ˛.0/ and
q D ˛.1/, we say that ˛ is a path from p to q. If ˛0 and ˛1 are both paths from p to
Topological Preliminaries 373
q, a path homotopy from ˛0 to ˛1 is a continuous map H W Œ0; 1 Œ0; 1 ! M that

satisfies
H.s; 0/ D ˛0 .s/ and H.s; 1/ D ˛1 .s/ for all s 2 Œ0; 1;

H.0; t / D p and H.1; t / D q for all t 2 Œ0; 1:
If there exists such a path homotopy, then ˛0 and ˛1 are said to be path-homotopic;
this is an equivalence relation on the set of all paths in M from p to q. The
equivalence class of a path ˛, called its path class, is denoted by Œ˛.
Paths ˛ and ˇ that satisfy ˛.1/ D ˇ.0/ are said to be composable; in this case,
their product is the path ˛ ˇ W Œ0; 1 ! M defined by

˛.2s/; 0 s 12 ;
˛ ˇ.s/ D
ˇ.2s 1/; 12 s 1:
Path products are well defined on path classes: if Œ˛0 D Œ˛1 and Œˇ0 D Œˇ1 , and
˛0 and ˇ0 are composable, then so are ˛1 and ˇ1 , and Œ˛0 ˇ0 D Œ˛1 ˇ1 . Thus we
obtain a well-defined product of path classes by Œ˛ Œˇ D Œ˛ ˇ.
A path from a point p to itself is called a loop based at p. If M is a topological
space and p 2 M , then any two loops based at p are composable, and the set of
path classes of loops based at p is a group under path class products, called the
fundamental group of M based at p and denoted by 1 .M; p/. The class of the
constant path cp .s/ p is the identity element, and the class of the reverse path
˛ 1 .s/ D ˛.1 s/ is the inverse of Œ˛. (Although multiplication of path classes is
associative, path products themselves are not, so a multiple product such as ˛ ˇ
is to be interpreted as .˛ ˇ/ .)
Proposition A.6 (Induced Homomorphisms). [LeeTM, Prop. 7.24 & Cor. 7.26]
Suppose M and N are topological spaces and p 2 M . If F W M ! N is a contin-
uous map, then the map F W 1 .M; p/ ! 1 .N; F .p// defined by F .Œ˛/ D ŒF ı˛
is a group homomorphism called the homomorphism induced by F . If F is a homeo-
morphism, then F is an isomorphism.
A topological space M is said to be simply connected if it is path-connected and
for some (hence every) point p 2 M , the fundamental group 1 .M; p/ is trivial.
I Exercise A.7. Show that if M is a simply connected topological space and p; q are
any two points in M , then all paths in M from p to q are path-homotopic.
A continuous map F W M ! N is said to be a homotopy equivalence if there is

a continuous map G W N ! M such that G ı F and F ı G are both homotopic to
identity maps.
Proposition A.8 (Homotopy Invariance of the Fundamental Group). [LeeTM,

Thm. 7.40] Suppose M and N are topological spaces and F W M ! N is a
homotopy equivalence. Then for every point p 2 M , the induced homomorphism
F W 1 .M; p/ ! 1 .N; F .p// is an isomorphism.
Smooth Manifolds and Smooth Maps
The setting for the study of Riemannian geometry is provided by smooth manifolds,
which are topological manifolds endowed with an extra structure that allows us to
differentiate functions and maps. First, we note that when U is an open subset of Rn ,
a map F W U ! Rk is said to be smooth (or of class C 1 ) if all of its component
functions have continuous partial derivatives of all orders. More generally, if the
domain U is an arbitrary subset of Rn , not necessarily open (such as a relatively
open subset of RnC ), then F is said to be smooth if for each x 2 U , F has a smooth
extension to a neighborhood of x in Rn . A diffeomorphism is a bijective smooth
map whose inverse is also smooth.
If M is a topological n-manifold with or without boundary, then two coordinate
charts .U; '/, .V; / for M are said to be smoothly compatible if both of the transition
maps ı ' 1 and ' ı 1 are smooth where they are defined (on '.U \ V / and
.U \ V /, respectively). Since these maps are inverses of each other, it follows that
both transition maps are in fact diffeomorphisms. An atlas for M is a collection
of coordinate charts whose domains cover M . It is called a smooth atlas if any
two charts in the atlas are smoothly compatible. A smooth structure on M is a
smooth atlas that is maximal, meaning that it is not properly contained in any larger
smooth atlas. A smooth manifold is a topological manifold endowed with a specific
smooth structure; and similarly, a smooth manifold with boundary is a topological
manifold with boundary endowed with a smooth structure. (We usually just say “M is
a smooth manifold,” or “M is a smooth manifold with boundary,” with the smooth
structure understood from the context.) If M is a set, a smooth manifold structure
on M is a second countable, Hausdorff, locally Euclidean topology together with a
smooth structure making it into a smooth manifold.
I Exercise A.9. Let M be a topological manifold with or without boundary. Show that
every smooth atlas for M is contained in a unique maximal smooth atlas, and two smooth
atlases determine the same smooth structure if and only if their union is a smooth atlas.
The manifolds in this book are always assumed to be smooth. As in most parts
of differential geometry, the theory still works under weaker differentiability as-
sumptions (such as k times continuously differentiable), but such considerations are
usually relevant only when one is treating questions of hard analysis that are beyond
our scope.
If M is a smooth manifold with or without boundary, then every coordinate chart
in the given maximal atlas is called a smooth coordinate chart for M or just a smooth
chart. The set U is called a smooth coordinate domain; if its image is an open ball
in Rn , it is called a smooth coordinate ball. If in addition ' extends
to a smooth
coordinate map ' 0 W U 0 ! Rn on an open set U 0 Ux such that ' 0 Ux is the closure
of the open ball '.U / in Rn , then U is called a regular coordinate ball. In this
case, Ux is diffeomorphic to a closed ball and is thus compact.
Proposition A.10. [LeeSM, Prop. 1.19] Every smooth manifold has a countable ba-
sis of regular coordinate balls.
Smooth Manifolds and Smooth Maps 375
Given a smooth chart .U; '/ for M , the component of' are called local
functions
coordinates for M , and are typically written as x 1 ; : : : ; x n , x i , or x, depending
on context. Although, formally speaking, a coordinate chart is a map from an open
subset U M to Rn , it is common when one is working in the domain of a specific chart
innR , and to identify a point
n
to use the coordinate map to identify U with its image
in U with its coordinate representation x ; : : : ; x 2 R .
1 n

In this book, we always write coordinates with upper indices, as in x i , and
expressions with indices are interpreted using the Einstein summation convention:
if in some monomial term the same index name appears exactly twice, once as an
upper index and once as a lower index, then that term is understood to be summed
over all possible values of that index (usually P from 1 to the dimension of the space).
Thus the expression ai vi is to be read as i ai vi . As we will see below, index posi-
tions for other sorts of objects such as vectors and covectors are chosen whenever
possible in such a way that summations that make mathematical sense obey the rule
that each repeated index appears once up and once down in each term to be summed.
Because of the result of Exercise A.9, to define a smooth structure on a manifold,
it suffices to exhibit a single smooth atlas for it. For example, Rn is a topological
n-manifold, and it has a smooth atlas consisting of the single chart Rn ; IdRn . More
generally, if V is an n-dimensional vector space, then every basis .b1 ; : : : ; bn / for
V yields a linear basis isomorphism B W Rn ! V by B.x 1 ; : : : ; x n / D x i bi , whose
inverse is a global coordinate chart, and it is easy to check that all such charts are
smoothly compatible. Thus every finite-dimensional vector space has a natural smooth
structure, which we call its standard smooth structure. We always consider Rn or
any other finite-dimensional vector space to be a smooth manifold with this structure
unless otherwise specified.
If M is a smooth n-manifold with or without boundary and W M is an open
subset, then W has a natural smooth structure consisting of all smooth charts .U; '/
for M such that U W , so every open subset of a smooth n-manifold is a smooth n-
manifold in a natural way, and similarly for manifolds with boundary. If M1 ; : : : ; Mk
are smooth manifolds of dimensions n1 ; : : : ; nk , respectively, then their Cartesian
product M1 Mk has a natural smooth atlas consisting of charts of the form
.U1 Uk ; '1 'k /, where .Ui ; 'i / is a smooth chart for Mi ; thus the prod-
uct space is a smooth manifold of dimension n1 C C nk .
Suppose M and N are smooth manifolds with or without boundary. A map
F W M ! N is said to be smooth if for every p 2 M , there exist smooth charts
.U; '/ for M containing p and .V; / for N containing F .p/ such that F .U / V
and the composite map ı F ı ' 1 is smooth from '.U / to .V /. In particular, if
N is an open subset of Rk or RkC with its standard smooth structure, we can take
to be the identity map of N , and then smoothness of F just means that each point
of M is contained in the domain of a chart .U; '/ such that F ı ' 1 is smooth.
It is an easy consequence of the definition that identity maps, constant maps, and
compositions of smooth maps are all smooth. A map F W M ! N is said to be a
diffeomorphism if it is smooth and bijective and F 1 W N ! M is also smooth.
If F W M ! N is smooth, and .U; '/ and .V; / are any smooth charts for M
and N respectively, the map Fy D ı F ı ' 1 is a smooth map between appropriate
subsets of Rn or RnC , called a coordinate representation of F . Often, in keeping

with the practice of using local coordinates to identify an open subset of a manifold
with an open subset of Rn , we identify the coordinate representation of F with (the
restriction of) F itself, and write things like “F is given in local coordinates by
F .x; y/ D .xy; x y/,” when we really mean that Fy is given by this formula for
suitable choices of charts in the domain and codomain.
We let C 1 .M; N / denote the set of all smooth maps from M to N , and
1
C .M / the vector space of all smooth functions from M to R. For every function
f W M ! R or Rk , we define the support of f , denoted by supp f , to be the closure
of the set fx 2 M W f .x/ ¤ 0g. If A M is a closed subset and U M is an
open subset containing A, then a smooth bump function for A supported in U is a
smooth function f W M ! R satisfying 0 f .x/ 1 for all x 2 M , f jA 1, and
supp f U .
If U D fU˛ g˛2A is an indexed open cover of M , then a smooth partition of
unity subordinate to U is an indexed family f ˛ g˛2A of functions ˛ 2 C 1 .M /
satisfying

0 ˛ .x/ 1 for all ˛ 2 A and all x 2 M .

supp ˛ U˛ for each ˛ 2 A.

The family of supports fsupp ˛ g˛2A is locally finite: every point of M has a
neighborhood that intersects supp ˛ for only finitely many values of ˛.
P

˛2A ˛ .x/ D 1 for all x 2 M .
Proposition A.11 (Existence of Partitions of Unity). [LeeSM, Thm. 2.23] If M is

a smooth manifold with or without boundary and U D fU˛ g˛2A is an indexed open
cover of M , then there exists a smooth partition of unity subordinate to U.
Proposition A.12 (Existence of Smooth Bump Functions). [LeeSM, Prop. 2.25]
If M is a smooth manifold with or without boundary, A M is a closed subset, and
U M is an open subset containing A, then there exists a smooth bump function
for A supported in U .
Tangent Vectors
Let M be a smooth manifold with or without boundary. There are various equiv-
alent ways of defining tangent vectors on M . The following definition is the most
convenient to work with in practice. For every point p 2 M , a tangent vector at p
is a linear map v W C 1 .M / ! R that is a derivation at p, meaning that for all
f; g 2 C 1 .M / it satisfies the product rule
v.fg/ D f .p/vg C g.p/vf: (A.1)
The set of all tangent vectors at p is denoted by Tp M and called the tangent space
at p.
Suppose M is n-dimensional and ' W U ! Uy Rn is a smooth coordinate
chart
on some open subset U M . Writing the coordinate functions of ' as x 1 ; : : : ; x n ,
Tangent Vectors 377
we define the coordinate vectors @=@x 1 jp , . . . , @=@x n jp by

ˇ ˇ
@ ˇˇ @ ˇˇ
i ˇ f D i ˇ f ı ' 1 : (A.2)
@x p @x '.p/
These vectors form a basis for Tp M , which therefore has dimension n. When there
can be no confusion about which coordinates are meant, we usually abbreviate
@=@x i jp by the notation @i jp . Thus once a smooth coordinate chart has been chosen,
every tangent vector v 2 Tp M can be written uniquely in the form
v D v i @i jp D v 1 @1 jp C C v n @n jp ; (A.3)
where the components

i v 1 ; : : : ; v n are obtained by applying v to the coordinate func-
tions: v D v x .
i
On a finite-dimensional vector space V with its standard smooth manifold struc-

ture, there is a natural (basis-independent) identification of each tangent space Tp V
with V itself, obtained by identifying a vector v 2 V with the derivation Dv jp ,
defined by ˇ
ˇ d ˇˇ
ˇ
Dv p .f / D f .p C t v/:
dt ˇtD0
i
In terms 1of the coordinates
x induced on V by any basis, this is the correspon-
P
dence v ; : : : ; v n $ i v i @i jp .
If F W M ! N is a smooth map and p is any point in M , we define a linear map
dFp W Tp M ! TF .p/ N , called the differential of F at p, by
dFp .v/f D v.f ı F /; v 2 Tp M:
I Exercise A.13. Given a smooth map F W M ! N and a point p 2 M , show that

dFp is a well-defined linear map from Tp M to TF .p/ N . Show that the differential of
the identity is the identity, and the differential of a composition is given by d.G ı F /p D
dGF .p/ ı dFp .

Once we have chosen local coordinates x i for M and y j for N , we find
by unwinding the definitions that the coordinate representation of the differential is
given by the Jacobian matrix of the coordinate representation of F , which is its
matrix of first-order partial derivatives:
ˇ ! ˇ
i @ ˇ
ˇ @Fy j i @ ˇ
ˇ
dFp v ˇ D .p/ v :
i
@x p @x i @y ˇF .p/
j
For every p 2 M , the dual vector space to Tp M is called the cotangent space
at p. This is the space Tp M D .Tp M / of real-valued linear functionals on Tp M ,
called (tangent) covectors at p. For every f 2 C 1 .M / and p 2 M , there is a
covector dfp 2 Tp M called the differential of f at p, defined by
dfp .v/ D vf for all v 2 Tp M: (A.4)

Submanifolds
The theory of submanifolds is founded on the inverse function theorem and its corol-
laries.
Theorem A.14 (Inverse Function Theorem for Manifolds). [LeeSM, Thm. 4.5]
Suppose M and N are smooth manifolds and F W M ! N is a smooth map. If the
linear map dFp is invertible at some point p 2 M , then there exist connected neigh-
borhoods U0 of p and V0 of F .p/ such that F jU0 W U0 ! V0 is a diffeomorphism.
The most useful consequence of the inverse function theorem is the following. A
smooth map F W M ! N is said to have constant rank if the linear map dFp has
the same rank at every point p 2 M .
Theorem A.15 (Rank Theorem). [LeeSM, Thm. 4.12] Suppose M and N are
smooth manifolds of dimensions m and n, respectively, and F W M ! N is a smooth
map with constant rank r. For each p 2 M there exist smooth charts .U; '/ for M
centered at p and .V; / for N centered at F .p/ such that F .U / V; in which F
has a coordinate representation of the form

Fy x 1 ; : : : ; x r ; x rC1 ; : : : ; x m D x 1 ; : : : ; x r ; 0; : : : ; 0 : (A.5)
Here are the most important types of constant-rank maps. In all of these defini-
tions, M and N are smooth manifolds, and F W M ! N is a smooth map.

F is a submersion if its differential is surjective at each point, or equivalently
if it has constant rank equal to dim N .

F is an immersion if its differential is injective at each point, or equivalently if
it has constant rank equal to dim M .

F is a local diffeomorphism if every point p 2 M has a neighborhood U such
that F jU is a diffeomorphism onto an open subset of N , or equivalently if F is
both a submersion and an immersion.

F is a smooth embedding if it is an injective immersion that is also a topologi-
cal embedding (a homeomorphism onto its image, endowed with the subspace
topology).
Theorem A.16 (Local Embedding Theorem). [LeeSM, Thm. 4.25] Every smooth
immersion is locally an embedding: if F W M ! N is a smooth immersion, then for
every p 2 M , there is a neighborhood U of p in M such that F jU is an embedding.
Theorem A.17 (Local Section Theorem). [LeeSM, Thm. 4.26] Suppose M and
N are smooth manifolds and W M ! N is a smooth map. Then is a smooth
submersion if and only if every point of M is in the image of a smooth local section
of (a map W U ! M defined on some open subset U N , with ı D IdU ).
Proposition A.18 (Properties of Submersions). [LeeSM, Prop. 4.28] Let M and

N be smooth manifolds and let W M ! N be a smooth submersion.
Submanifolds 379
(a) is an open map.

(b) If is surjective, it is a quotient map.
Proposition A.19 (Uniqueness of Smooth Quotients). [LeeSM, Thm. 4.31] Sup-

pose M , N1 , and N2 are smooth manifolds, and 1 W M ! N1 and 2 W M ! N2
are surjective smooth submersions that are constant on each other’s fibers. Then
there exists a unique diffeomorphism F W N1 ! N2 such that F ı 1 D 2 .
Theorem A.20 (Global Rank Theorem). [LeeSM, Thm. 4.14] Suppose M and N
are smooth manifolds, and F W M ! N is a smooth map of constant rank.
(a) If F is surjective, then it is a smooth submersion.
(b) If F is injective, then it is a smooth immersion.
(c) If F is bijective, then it is a diffeomorphism.
Suppose M is a smooth manifold with or without boundary. An immersed n-

dimensional submanifold of M is a subset M M endowed with a topology that
makes it into an n-dimensional topological manifold and a smooth structure such
that the inclusion map M ,! M is a smooth immersion. It is called an embedded
submanifold if the inclusion is a smooth embedding, or equivalently if the given
topology on M is the subspace topology. Immersed and embedded submanifolds
with boundary are defined similarly. In each case, the codimension of M is the
difference dim M dim M . A submanifold of codimension 1 is known as a hyper-
surface. Without further qualification, the word submanifold means an immersed
submanifold, which includes an embedded one as a special case.
An embedded submanifold with or without boundary is said to be properly em-
bedded if the inclusion map is proper.
is a smooth manifold with or without
Proposition A.21. [LeeSM, Prop. 5.5] If M
boundary, then an embedded submanifold M M is properly embedded if and only

if it is a closed subset of M .
is called
A properly embedded codimension-0 submanifold with boundary in M
a regular domain.
Proposition A.22 (Slice Coordinates). [LeeSM, Thms. 5.8 & 5.51] Let M be

a smooth m-manifold without boundary and let M M be an embedded n-
dimensional submanifold with or without boundary. Then for each p 2 M there
exist a neighborhood Uz of p in M on
and smooth coordinates x 1 ; : : : ; x m for M
z z
U such that M \ U is a set of one of the following forms:

z fx 2 Uz W x nC1 D D x m D 0g if p … @M;
M \U D
fx 2 Uz W x nC1 D D x m D 0 and x n 0g if p 2 @M:

In such a chart, x 1 ; : : : ; x n form smooth local coordinates for M .
Coordinates satisfying either of these conditions are called slice coordinates;

when it is necessary to distinguish them, coordinates satisfying the second condition
are referred to as boundary slice coordinates.
If M is an immersed submanifold, then Theorem A.16 shows that every point of
M has a neighborhood U in M that is embedded in M , so we can construct slice
coordinates for U (but they need not be slice coordinates for M ).
is a smooth manifold and M M
I Exercise A.23. Suppose M is an embedded sub-
manifold. Show that every smooth function f W M ! R can be extended to a smooth
function on a neighborhood of M in M whose restriction to M is f ; and if M is also
closed in M , then the neighborhood can be taken to be all of M . [Hint: Extend f locally

in slice coordinates by letting it be independent of x nC1 ; : : : ; x m , and patch together
using a partition of unity.]
Here is the way that most submanifolds are presented. Suppose ˚ W M ! N is

1
any map. Every subset of the form ˚ .fyg/ M for some y 2 N is called a level
set of ˚, or the fiber of ˚ over y. The simpler notation ˚ 1 .y/ is also used for a
level set when there is no likelihood of ambiguity.
Theorem A.24 (Constant-Rank Level Set Theorem). [LeeSM, Thm. 5.12] Sup-
pose M and N are smooth manifolds, and ˚ W M ! N is a smooth map with con-
stant rank r. Every level set of ˚ is a properly embedded submanifold of codimen-
.
sion r in M
Corollary A.25 (Submersion Level Set Theorem). [LeeSM, Cor. 5.13] Suppose
M and N are smooth manifolds, and ˚ W M ! N is a smooth submersion. Every
, whose codimension is
level set of ˚ is a properly embedded submanifold of M
equal to dim N .
In fact, a map does not have to be a submersion, or even to have constant rank, for
its level sets to be embedded submanifolds. If ˚ W M ! N is a smooth map, a point
p2M is called a regular point of ˚ if the linear map d ˚p W Tp M ! T˚.p/ N is
surjective, and p is called a critical point of ˚ if it is not. A point c 2 N is called a
regular value of ˚ if every point of ˚ 1 .c/ is a regular point of ˚, and a critical
value otherwise. A level set ˚ 1 .c/ is called a regular level set of ˚ if c is a regular
value of ˚.
Corollary A.26 (Regular Level Set Theorem). [LeeSM, Cor. 5.14] Let M and N
be smooth manifolds, and let ˚ W M ! N be a smooth map. Every regular level
whose codimension is equal to
set of ˚ is a properly embedded submanifold of M
dim N .
Suppose M is a smooth manifold and M M is an embedded codimension-k

submanifold. A smooth map ˚ W M ! Rk is called a defining function for M if M
is a regular level set of ˚. More generally, if ˚ W U ! Rk is a smooth map defined
on an open subset U M such that M \ U is a regular level set of ˚, then ˚ is
called a local defining function for M . The following is a partial converse to the
submersion level set theorem.
Submanifolds 381
Proposition A.27 (Existence of Local Defining Functions). [LeeSM, Prop. 5.16]

Suppose M is a smooth manifold and M M is an embedded submanifold of
and
codimension k. For each point p 2 M , there exist a neighborhood U of p in M
a smooth submersion ˚ W U ! R such that M \ U is a level set of ˚.
k
If M M is a submanifold (immersed or embedded), the inclusion map W M ,!

M induces an injective linear map d p W Tp M ! Tp M for each p 2 M . It is stan-
dard practice to identify Tp M with its image d p .Tp M / Tp M , thus regarding
Tp M as a subspace of Tp M . The next proposition gives two handy ways to identify
this subspace in the embedded case.
Proposition A.28 (Tangent Space to a Submanifold). [LeeSM, Props. 5.37 &

5.38] Suppose M is a smooth manifold, M M is an embedded submanifold, and
, the tangent space Tp M is characterized by
p 2 M . As a subspace of Tp M
˚ ˇ
Tp M D v 2 T p M W vf D 0 whenever f 2 C 1 M and f ˇ D 0 :
M
If ˚ is a local defining function for M on a neighborhood of p, then
Tp M D Ker d ˚p :
Proposition A.29 (Lagrange Multiplier Rule). Let M be a smooth m-manifold

and M M be an embedded hypersurface. Suppose f 2 C 1 M , and p 2 M is
a point at which f attains a local maximum or minimum value among points in M .
If ˚ W U ! R is a local defining function for M on a neighborhood U of p, then
there is a real number (called a Lagrange multiplier) such that dfp D d ˚p .
I Exercise A.30. Prove the preceding proposition. [Hint: One way to proceed is to begin
by showing that there are smooth functions x 1 ; : : : ; x m1 on a neighborhood of p such
that .x 1 ; : : : ; x m1 ; ˚ / are local slice coordinates for M .]
In treating manifolds with boundary, many constructions can be simplified

by viewing a manifold with boundary as being embedded in a larger manifold
without boundary. Here is one way to do that.
If M is a topological manifold with nonempty boundary, the double of M is the
quotient space D.M / formed from the disjoint union of two copies of M (say M1
and M2 ), by identifying each point in @M1 with the corresponding point in @M2 .
Proposition A.31. [LeeSM, Example 9.32] Suppose M is a smooth manifold with

nonempty boundary. Then D.M / is a topological manifold without boundary, and
it has a smooth structure such that the natural maps M ! M1 ! D.M / and M !
M2 ! D.M / are smooth embeddings. Moreover, D.M / is compact if and only if
M is compact, and connected if and only if M is connected.
Vector Bundles
Suppose M is a smooth manifold with or without boundary. The tangent bundle

of M , denoted
` by TM , is the disjoint union of the tangent spaces at all points of M :
TM D p2M Tp M . This set can be thought of both as a union of vector spaces
and as a manifold in its own right. This kind of structure, called a vector bundle,
is extremely common in differential geometry, so we take some time to review the
definitions and basic properties of vector bundles.
For any positive integer k, a smooth vector bundle of rank k is a pair of smooth
manifolds E and M with or without boundary, together with a smooth surjective
map W E ! M satisfying the following conditions:
(i) For each p 2 M , the set Ep D 1 .p/ is endowed with the structure of a
k-dimensional real vector space.
(ii) For each p 2 M , there exist a neighborhood U of p and a diffeomorphism
˚ W 1 .U / ! U Rk such that U ı ˚ D , where U W U Rk ! U is
the projection onto the first factor; and for each q 2 U , ˚ restricts to a linear
isomorphism from Eq to fqg Rk Š Rk .
The space M is called the base of the bundle, E is called its total space, and
is its projection. Each set Ep D 1 .p/ is called the fiber of E over p, and each
diffeomorphism ˚ W 1 .U / ! U Rk described above is called a smooth local
trivialization.
It frequently happens that we are presented with a collection of vector spaces
(such as tangent spaces), one for each point in a manifold, that we would like to
“glue together” to form a vector bundle. There is a shortcut for showing that such
a collection forms a vector bundle without first constructing a smooth manifold
structure on the total space. As the next lemma shows, all we need to do is to exhibit
the maps that we wish to regard as local trivializations and check that they overlap
correctly.
Lemma A.32 (Vector Bundle Chart Lemma). [LeeSM, Lemma 10.6] Let M be a
smooth manifold with or without boundary, and suppose that for each p `2 M we
are given a real vector space Ep of some fixed dimension k. Let E D p2M Ep
(the disjoint union of all the spaces Ep ), and let W E ! M be the map that takes
each element of Ep to the point p. Suppose furthermore that we are given
(i) an indexed open cover fU˛ g˛2A of M ;
(ii) for each ˛ 2 A, a bijective map ˚˛ W 1 .U˛ / ! U˛ Rk whose restriction
to each Ep is a linear isomorphism from Ep to fpg Rk Š Rk ;
(iii) for each ˛; ˇ 2 A such that U˛ \ Uˇ ¤ ¿, a smooth map ˛ˇ W U˛ \ Uˇ !
GL.k; R/ such that the composite map ˚˛ ı ˚ˇ1 from .U˛ \ Uˇ / Rk to itself
has the form
˚˛ ı ˚ˇ1 .p; v/ D .p; ˛ˇ .p/v/: (A.6)
Then E has a unique smooth manifold structure making it a smooth vector
bundle of rank k over M , with as projection and the maps ˚˛ as smooth local
trivializations.
Vector Bundles 383
The smooth GL.k; R/-valued maps ˛ˇ of this lemma are called transition func-
tions for E .
If W E ! M is a smooth vector bundle over M , then a section of E is a continuous
map W M ! E such that ı D IdM , or equivalently, .p/ 2 Ep for all p. If
U M is an open set, then a local section of E over U is a continuous map
W U ! E satisfying the analogous equation ı D IdU . A smooth (local or global)
section of E is just a section that is smooth as a map between smooth manifolds. If we
need to speak of “sections” that are not necessarily continuous, we use the following
terminology: a rough (local) section of E is a map W U ! E defined on some open
set U M , not necessarily smooth or even continuous, such that ı D IdU .
A local frame for E is an ordered k-tuple .1 ; : : : ; k / of local sections over an
open set U whose values at each p 2 U constitute a basis for Ep . It is called a global
frame if U D M . If is a (local or global) section of E and .1 ; : : : ; k / is a local
frame for E over U M , then the value of at each p 2 U can be written
.p/ D i .p/i .p/
for some functions 1 ; : : : ; k W U ! R, called the component functions of with

respect to the given frame.
The next lemma gives a criterion for smoothness that is usually easy to verify in
practice.
Lemma A.33 (Local Frame Criterion for Smoothness). [LeeSM, Prop. 10.22]
Let W E ! M be a smooth vector bundle, and let W M ! E be a rough section.
If .i / is a smooth local frame for E over an open subset U M , then is smooth
on U if and only if its component functions with respect to .i / are smooth.
If E ! M is a smooth vector bundle, then the set of smooth sections of E,

denoted by .E/, is a vector space (usually infinite-dimensional) under pointwise
addition and multiplication by constants. Its zero element is the zero section
de-
fined by
.p/ D 0 2 Ep for all p 2 M . For every section of E, the support of
is the closure of the set fp 2 M W .p/ ¤ 0g.
Given a smooth vector bundle E W E ! M , a smooth subbundle of E is a
smooth vector bundle D W D ! M , in which ˇ D is an embedded submanifold (with
or without boundary) of E and D D E ˇD , such that for each p 2 M , the subset
Dp D D \ Ep is a linear subspace of Ep , and the vector space structure on Dp is
the one inherited from Ep . Given a collection of subspaces Dp Ep , one for each
p 2 M , the following lemma gives a convenient condition for checking that their
union is a smooth subbundle.
Lemma A.34 (Local Frame Criterion for Subbundles). [LeeSM, Lemma 10.32]
Let W E ! M be a smooth vector bundle, and suppose that for each p 2 M
we are given a k-dimensional linear subspace Dp Ep . Suppose further that
each p 2 M has a neighborhood U on which there are smooth local sections
1 ; : : : ; k W S
U ! E such that 1 .q/; : : : ; k .q/ form a basis for Dq at each q 2 U .
Then D D p2M Dp E is a smooth subbundle of E.
Suppose M is a smooth manifold with or without boundary, and M M is

an immersed or embedded submanifold with or without boundary. If E ! M

is any smooth rank-k vector bundle over M , we obtain a smooth vector bundle
jM W EjM ! M of rank k over M , called the restriction of E to M , whose total
space is EjM D 1 .M /, whose fiber at each p 2 M is exactly the fiber of E, and
whose local trivializations are the restrictions of the local trivializations of E. (See
[LeeSM, Example 10.8].)
Every smooth section of E restricts to a smooth section of EjM . Conversely, the
next exercise shows that in most cases, smooth sections of EjM extend to smooth
sections of E, at least locally near M .
I Exercise A.35. Suppose M is a smooth manifold, E ! M is a smooth vector bundle,

and M M is an embedded submanifold. Show that every smooth section of the restricted
bundle E jM can be extended to a smooth section of E on a neighborhood of M in M ;
, the neighborhood can be taken to be all of M
and if M is closed in M .
The Tangent Bundle and Vector Fields
We continue to assume that M is a smooth manifold with or without boundary. In

this section, we see how the theory of vector bundles applies to the particular case
of the tangent bundle.
Given any smooth local coordinate chart U; x i for M , we can define a local
trivialization of TM over U by letting ˚ W 1 .U / ! U Rn be the map sending
v 2 Tp M to p; .v 1 ; : : : ; v n / , where v i @=@x i jp is the coordinate representation for

v. Given another such chart Uz ; xzi for M , we obtain a corresponding local triv-

ialization z̊ .v/ D p; .z v 1 ; : : : ; vzn / , where v D vzi @=@z
x i jp . The transformation law
for coordinate vectors shows that
xj
@z
vzj D .p/v i ;
@x i
so the transitionfunction between two such charts is given by the smooth matrix-
valued function @z x j =@x i . Thus the chart lemma shows that these local trivializa-
tions turn TM into a smooth vector i bundle.
1 A smooth coordinate chart x for M also yields smooth local coordinates
x ; : : : ; x n ; v 1 ; : : : ; v n on 1 .U / TM , called natural coordinates for TM , by
following
1 then local trivialization ˚ with the map U Rn ! Rn Rn that sends
p; v ; : : : ; v to x .p/; : : : ; x .p/; v 1 ; : : : ; v n .
1 n
A section of TM is called a vector field on M . To avoid confusion between the

point p 2 M at which a vector field is evaluated and the action of the vector field on
a function, we usually write the value of a vector field X at p 2 M as Xp 2 Tp M ,
or, if it is clearer (for if X itself has one or more subscripts), as X jp .
example
For example, if U; x i is a smooth coordinate chart for M , for each i we get
a smooth coordinate vector field on U , denoted by @=@x i (and often abbreviated by
The Tangent Bundle and Vector Fields 385
@i if no confusion results), whose value at p 2 U is the coordinate vector @=@x i jp

defined above. Each coordinate vector field is smooth because its coordinate rep-
resentation in natural coordinates is .x 1 ; : : : ; x n / 7! x 1 ; : : : ; x n ; 0; : : : ; 0; 1; 0; : : : ; 0 .
The vector fields @1 ; : : : ; @n form a smooth local frame for TM , called a coordinate
frame.
It follows from Lemma A.33 that a vector field is smooth if and only if its com-
ponents are smooth with respect to some smooth local frame (such as a coordinate
frame) in a neighborhood of each point. It is standard to use the notation X.M / as
another name for .TM /, the space of all smooth vector fields on M .
Now suppose M and N are smooth manifolds with or without boundary, and
F W M ! N is a smooth map. We obtain a smooth map dF W TM ! T N , called the
global differential of F , whose restriction to each tangent space Tp M is the linear
map dFp defined above. In general, the global differential does not take vector fields
to vector fields. In the special case that X 2 X.M / and Y 2 X.N / are vector fields
such that dF .Xp / D YF .p/ for all p 2 M , we say that the vector fields X and Y are
F-related.
Lemma A.36. [LeeSM, Prop. 8.19 & Cor. 8.21] Let F W M ! N be a diffeomor-
phism between smooth manifolds with or without boundary. For every X 2 X.M /,
there is a unique vector field F X 2 X.N /, called the pushforward of X , that is
F -related to X . For every f 2 C 1 .N /, it satisfies

.F X /f ı F D X.f ı F /: (A.7)
Suppose X 2 X.M /. Given a real-valued function f 2 C 1 .M /, applying X to

f yields a new function Xf 2 C 1 .M / by Xf .p/ D Xp f . The defining equation
(A.1) for tangent vectors translates into the following product rule for vector fields:
X.fg/ D f Xg C g Xf: (A.8)
A map X W C 1 .M / ! C 1 .M / is called a derivation of C 1 .M / (as opposed to a

derivation at a point) if it is linear over R and satisfies (A.8) for all f; g 2 C 1 .M /.
Lemma A.37. [LeeSM, Prop. 8.15] Let M be a smooth manifold with or without
boundary. A map D W C 1 .M / ! C 1 .M / is a derivation if and only if it is of the
form Df D Xf for some X 2 X.M /.
Given smooth vector fields X; Y 2 X.M /, define a map ŒX; Y W C 1 .M / !

1
C .M / by
ŒX; Y f D X.Yf / Y.Xf /:
A straightforward computation shows that ŒX; Y is a derivation of C 1 .M /, and
thus by Lemma A.37 it defines a smooth vector field, called the Lie bracket of X
and Y .
Proposition A.38 (Properties of Lie Brackets). [LeeSM, Prop. 8.28] Let M be a

smooth manifold with or without boundary and X; Y; Z 2 X.M /.
(a) ŒX; Y is bilinear over R as a function of X and Y .

(b) ŒX; Y D ŒY; X (antisymmetry).
(c) ŒX; ŒY; Z C ŒY; ŒZ; X C ŒZ; ŒX; Y D 0 (Jacobi identity).
(d) For f; g 2 C 1 .M /, Œf X; gY D fgŒX; Y C .f Xg/Y .gYf /X .
Proposition A.39 (Naturality of Lie Brackets). [LeeSM, Prop. 8.30 & Cor. 8.31]
Let F W M ! N be a smooth map between manifolds with or without boundary, and
let X1 ; X2 2 X.M / and Y1 ; Y2 2 X.N / be vector fields such that Xi is F -related
to Yi for i D 1; 2. Then ŒX1 ; X2 is F -related to ŒY1 ; Y2 . In particular, if F is a
diffeomorphism, then F ŒX1 ; X2 D ŒF X1 ; F X2 .
Now suppose M is a smooth manifold with or without boundary and M M

is an immersed or embedded submanifold with or without boundary. The bundle
TM to M , is called the ambient tangent bundle.
jM , obtained by restricting T M
It is a smooth bundle over M whose rank is equal to the dimension of M . The
tangent bundle TM is naturally viewed as a smooth subbundle of T M jM , and

smooth vector fields M can also be viewed as smooth sections of T M jM . A
on
vector field X 2 X M always restricts to a smooth section of T M jM , and it re-
stricts to a smooth section of TM if and only if it is tangent to M , meaning that
Xp 2 T p M T p M for each p 2 M .
Corollary A.40 (Brackets of Vector Fields Tangent to Submanifolds). [LeeSM,

Cor. 8.32] Let M be a smooth manifold and let M be an immersed submanifold
with or without boundary in M that
. If Y1 and Y2 are smooth vector fields on M
are tangent to M , then ŒY1 ; Y2 is also tangent to M .
I Exercise A.41. Let M be a smooth manifold with or without boundary and let M
M be an embedded submanifold with or without boundary. Show that a vector field X 2

X M is tangent to M if and only if Xf jM D 0 whenever f 2 C 1 M is a function
that vanishes on M .
Integral Curves and Flows
A curve in a smooth manifold M (with or without boundary) is a continuous map

W I ! M , where I R is some interval.
If is smooth, then for each t0 2 I we
obtain a vector 0 .t0 / D dt0 d=dt jt0 , called the velocity of at time t0 . It acts
on functions by
0 .t0 /f D .f ı /0 .t0 /:
In any smooth local coordinates, the coordinate expression for 0 .t0 / is exactly the
same as it would be in Rn : the components of 0 .t0 / are the ordinary t -derivatives
of the components of .
If X 2 X.M /, then a smooth curve W I ! M is called an integral curve of X if its
velocity at each point is equal to the value of X there: 0 .t / D X.t/ for each t 2 I .
The fundamental fact about vector fields (at least in the case of manifolds without
boundary) is that there exists a unique maximal integral curve starting at each point,
The Tangent Bundle and Vector Fields 387
varying smoothly as the point varies. These integral curves are all encoded into a
global object called a flow, which we now define.
Given a smooth manifold M (without boundary), a flow domain for M is an
open subset D R M with the property that for each p 2 M , the set D .p/ D
ft 2 R W .t; p/ 2 Dg is an open interval containing 0. Given a flow domain D and a
map W D ! M , for each t 2 R we let Mt D fp 2 M W .t; p/ 2 Dg, and we define
maps t W Mt ! M and .p/ W D .p/ ! M by t .p/ D .p/ .t / D .t; p/.
A flow on M is a continuous map W D ! M , where D R M is a flow
domain, that satisfies
0 D IdM ;
t ı s .p/ D tCs .p/ wherever both sides are defined.
If is a smooth flow, we obtain a smooth vector field X 2 X.M / defined by Xp D

.p/ 0
.0/, called the infinitesimal generator of .
Theorem A.42 (Fundamental Theorem on Flows). [LeeSM, Thm. 9.12] Let X

be a smooth vector field on a smooth manifold M (without boundary). There is a
unique smooth maximal flow W D ! M whose infinitesimal generator is X . This
flow has the following properties:
(a) For each p 2 M; the curve .p/ W D .p/ ! M is the unique maximal integral
curve of X starting at p. ˚
(b) If s 2 D .p/ , then D ..s;p// is the interval D .p/ s D t s W t 2 D .p/ .
(c) For each t 2 R, the set Mt is open in M , and t W Mt ! Mt is a diffeomor-
phism with inverse t .
Although the fundamental theorem guarantees only that each point lies on an
integral curve that exists for a short time, the next lemma can often be used to prove
that a particular integral curve exists for all time.
Lemma A.43 (Escape Lemma). Suppose M is a smooth manifold and X 2 X.M /.

If W I ! M is a maximal integral curve of X whose domain I has a finite least
upper bound b, then for every t0 2 I , Œt0 ; b/ is not contained in any compact
subset of M .
I Exercise A.44. Prove the escape lemma.
Proposition A.45 (Canonical Form for a Vector Field). [LeeSM, Thm. 9.22] Let
X be a smooth vector field on a smooth manifold M , and suppose p 2 M is a point
where Xp ¤ 0. There exist smooth coordinates x i on some neighborhood of p in
which X has the coordinate representation @=@x 1 .
The fundamental theorem on flows leads to a way of differentiating one vector

field along the flow of another. Suppose M is a smooth manifold, X; Y 2 X.M /,
and is the flow of X . The Lie derivative of Y with respect to X is the vector field
LX Y defined by
ˇ
d ˇˇ d.t /t .p/ Yt .p/ Yp
.LX Y /p D d.t /t .p/ Yt .p/ D lim :
dt ˇ tD0 t!0 t
This formula is useless for computation, however, because typically the flow of
a vector field is difficult or impossible to compute explicitly. Fortunately, there is
another expression for the Lie derivative that is much easier to compute.
Proposition A.46. [LeeSM, Thm. 9.38] Suppose M is a smooth manifold and

X; Y 2 X.M /. The Lie derivative of Y with respect to X is equal to the Lie bracket
ŒX; Y .
One of the most important applications of the Lie derivative is as an obstruction

to invariance under a flow. If is a smooth flow, we say that a vector field Y is
invariant under if .t / Y D Y wherever the left-hand side is defined.
Proposition A.47. [LeeSM, Thm. 9.42] Let M be a smooth manifold and X 2

X.M /. A smooth vector field is invariant under the flow of X if and only if its
Lie derivative with respect to X is identically zero.
A k-tuple of vector fields X1 ; : : : ; Xk is said to commute if ŒXi ; Xj D 0 for each

i and j .
Proposition A.48 (Canonical Form for Commuting Vector Fields). [LeeSM,

Thm. 9.46] Let M be a smooth n-manifold, and let .X1 ; : : : ; Xk / be a linearly inde-
subset W M . For
pendent k-tuple of smooth commuting vector fields onan open
each p 2 W; there exists a smooth coordinate chart U; x i centered at p such
that Xi D @=@x i for i D 1; : : : ; k.
Smooth Covering Maps

A covering map is a surjective continuous map W M ! M between connected and
locally path-connected topological spaces, for which each point of M has connected
neighborhood U that is evenly covered, meaning that each connected component of
1 .U / is mapped homeomorphically onto U by . It is called a smooth covering
map if M and M are smooth manifolds with or without boundary and each com-
ponent of 1 .U / is mapped diffeomorphically onto U . For every evenly covered
open set U M , the components of 1 .U / are called the sheets of the covering
over U .
Here are the main properties of covering maps that we need.
Proposition A.49 (Elementary Properties of Smooth Covering Maps). [LeeSM,

Prop. 4.33]
(a) Every smooth covering map is a local diffeomorphism, a smooth submersion,
an open map, and a quotient map.
Smooth Covering Maps 389
(b) An injective smooth covering map is a diffeomorphism.

(c) A topological covering map is a smooth covering map if and only if it is a local
diffeomorphism.
Proposition A.50 A covering map is a proper map if and only if it is finite-sheeted.
I Exercise A.51. Prove the preceding proposition.
Example A.52 (A Covering of the n-Torus). The n-torus is the manifold T n D

S1 S1 , regarded as the subset of R2n defined by .x 1 /2 C .x 2 /2 D D
.x 2n1 /2 C .x 2n /2 D 1. The smooth map X W Rn ! T n given by X.u1 ; : : : ; un / D
.cos u1 ; sin u1 ; : : : ; cos un ; sin un / is a smooth covering map. //
If W M ! M is a covering map and F W B ! M is a continuous map from a

topological space B into M , then a lift of F is a continuous map Fz W B ! M
such
z
that ı F D F .
! M is a smooth
Proposition A.53 (Lifts of Smooth Maps are Smooth). If W M
covering map, B is a smooth manifold with or without boundary, and F W B ! M
is a smooth map, then every lift of F is smooth.
Proof. Since is a smooth submersion, every lift Fz W B ! M

can be written locally
z
as F D ı F , where is a smooth local section of (see Thm. A.17). t
u
!M
Proposition A.54 (Lifting Properties of Covering Maps). Suppose W M
is a covering map.
(a) U NIQUE L IFTING P ROPERTY [LeeTM, Thm. 11.12] If B is a connected topo-
logical space and F W B ! M is a continuous map, then any two lifts of F
that agree at one point are identical.
(b) PATH L IFTING P ROPERTY [LeeTM, Cor. 11.14] Suppose f W Œ0; 1 ! M
is a continuous path. For every pz 2 1 .f .0//, there exists a unique lift
fz W Œ0; 1 ! M
of f such that fz.0/ D p.
z
(c) M ONODROMY T HEOREM [LeeTM, Thm. 11.15] Suppose f; g W Œ0; 1 ! M
are path-homotopic paths and fz; gz W Œ0; 1 ! M
are their lifts starting at the
same point. Then f and gz are path-homotopic and fz.1/ D gz.1/.
z
Theorem A.55 (Injectivity Theorem). [LeeTM, Thm. 11.16] If W M ! M is a

covering map, then for each point z
x 2 M , the induced fundamental group homo-

morphism W 1 M ; xz ! 1 M; .z x / is injective.
Theorem A.56 (Lifting Criterion). [LeeTM, Thm. 11.18] Suppose W M ! M is

a covering map, B is a connected and locally path-connected topological space,
and
F W B ! M is a continuous map. Given b 2 B and z
x 2 such that xz D F .b/,
M

if and only if F 1 .B; b/ 1 M
the map F has a lift to M ; xz .
Corollary A.57 (Lifting Maps from Simply Connected Spaces). [LeeTM, Cor.
11.19] Suppose W M ! M and F W B ! M satisfy the hypotheses of Theorem
A.56, and in addition B is simply connected. Then every continuous map F W B !
M has a lift to M . Given any b 2 B, the lift can be chosen to take b to any point in
the fiber over F .b/.
Corollary A.58 (Covering Map Homeomorphism Criterion). A covering map

W M ! M is a homeomorphism if and only if the induced homomorphism

W 1 M ; xz ! 1 .M; .z . A
x // is surjective for some (hence every) xz 2 M
smooth covering map is a diffeomorphism if and only if the induced homomorphism
is surjective.
Proof. By Theorem A.56, the hypothesis implies that the identity map Id W M ! M
eW M ! M
has a lift Id , which in this case is a continuous inverse for . If is a
smooth covering map, then the lift is also smooth. t
u
Corollary A.59 (Coverings of Simply Connected Spaces). [LeeTM, Cor. 11.33]

If M is a simply connected manifold with or without boundary, then every cover-
ing of M is a homeomorphism, and if M is smooth, every smooth covering is a
diffeomorphism.
Proposition A.60 (Existence of a Universal Covering Manifold). [LeeSM, Cor.

4.43] If M is a connected smooth manifold, then there exist a simply connected
smooth manifold M , called the universal covering manifold of M , and a smooth
covering map W M ! M . It is unique in the sense that if M
0 is any other simply
connected smooth manifold that admits a smooth covering map 0 W M 0 ! M , then
there exists a diffeomorphism ˚ W M !M such that ı ˚ D .
0 0
Proposition A.61. [LeeTM, Cor. 11.31] With W M ! M as in the previous propo-

sition, each fiber of has the same cardinality as the fundamental group of M .
I Exercise A.62. Suppose W M ! M is a covering map. Show that M

is compact if
and only if M is compact and is a finite-sheeted covering.
We will revisit smooth covering maps at the end of Appendix C.

Appendix B
Review of Tensors
Of all the constructions in smooth manifold theory, the ones that play the most
fundamental roles in Riemannian geometry are tensors and tensor fields. Most of
the technical machinery of Riemannian geometry is built up using tensors; indeed,
Riemannian metrics themselves are tensor fields. This appendix offers a brief review
of their definitions and properties. For a more detailed exposition of the material
summarized here, see [LeeSM].
Tensors on a Vector Space

We begin with tensors on a finite-dimensional vector space. There are many ways of
constructing tensors on a vector space, but for simplicity we will stick with the most
concrete description, as real-valued multilinear functions. The simplest tensors are
just linear functionals, also called covectors.
Covectors
Let V be an n-dimensional vector space (all of our vector spaces are assumed to be
real). When we work with bases for V , it is usually important to consider ordered
bases, so will assume that each basis comes endowed with a specific ordering. We
use parentheses to denote ordered k-tuples, and braces to denote unordered ones, so
an ordered basis is designated by either .b1 ; : : : ; bn / or .bi /, and the corresponding
unordered basis by fb1 ; : : : ; bn g or fbi g.
The dual space of V , denoted by V , is the space of linear maps from V to R.
Elements of the dual space are called covectors or linear functionals on V . Under
the operations of pointwise addition and multiplication by constants, V is a vector
space.
Suppose .b1 ; : : : ; bn / is an ordered basis for V . For each i D 1; : : : ; n, define a
covector ˇ i 2 V by
392 Review of Tensors
ˇ i .bj / D ıji ;
where ıji is the Kronecker delta symbol, defined by

1 if i D j;
ıji D (B.1)
0 if i ¤ j:

It is a standard exercise to prove that ˇ 1 ; : : : ; ˇ n is a basis for V , called the dual
basis to .bi /. Therefore, V is finite-dimensional, and its dimension is the same as
that of V . Every covector ! 2 V can thus be written in terms of the dual basis as
! D !j ˇ j ; (B.2)
where the components !j are defined by !j D !.bj /. The action of ! on an arbi-

trary vector v D v i bi 2 V is then given by
!.v/ D !j v j : (B.3)
(Here and throughout the rest of this appendix we use the Einstein summation con-
vention; see p. 375.)
Every vector v 2 V uniquely determines a linear functional on V , by ! 7! !.v/.
Because we are assuming V to be finite-dimensional, it is straightforward to show that
this correspondence gives a canonical (basis-independent) isomorphism between V
and V (the dual space of V ). Given ! 2 V and v 2 V , we can denote the natural
action of ! on v either by the traditional functional notation !.v/, or by either of the
more symmetric notations h!; vi or hv; !i. The latter notations are meant to emphasize
that it does not matter whether we think of the resulting number as the effect of the linear
functional ! acting on the vector v, or as the effect of v 2 V acting on the covector
!. Note that when applied to a vector and a covector, this pairing makes sense without
any choice of an inner product on V .
Higher-Rank Tensors on a Vector Space
Now we generalize from linear functionals to multilinear ones. If V1 ; : : : ; Vk and W

are vector spaces, a map F W V1 Vk ! W is said to be multilinear if it is linear
as a function of each variable separately, when all the others are held fixed:
F .v1 ; : : : ; avi C a0 vi0 ; : : : ; vk / D aF .v1 ; : : : ; vi ; : : : ; vk / C a0 F .v1 ; : : : ; vi0 ; : : : ; vk /:
Given a finite-dimensional vector space V , a covariant k-tensor on V is a mul-

tilinear map
™
F W V V ! R:
k copies
Similarly, a contravariant k-tensor on V is a multilinear map

Tensors on a Vector Space 393
š
F W V V ! R:
k copies
We often need to consider tensors of mixed types as well. A mixed tensor of type
(k, l), also called a k-contravariant, l-covariant tensor, is a multilinear map
š ™
F W V V V V ! R:
k copies l copies
Actually, in many cases it is necessary to consider real-valued multilinear functions

whose arguments consist of k covectors and l vectors, but not necessarily in the
order implied by the definition above; such an object is still called a tensor of type
.k; l/. For any given tensor, we will make it clear which arguments are vectors and
which are covectors.
The spaces of tensors on V of various types are denoted by
T k .V / D fcovariant k-tensors on V gI
T k .V / D fcontravariant k-tensors on V gI
T .k;l/ .V / D fmixed .k; l/-tensors on V g:
The rank of a tensor is the number of arguments (vectors and/or covectors) it takes.
By convention, a 0-tensor is just a real number. (You should be aware that the no-
tation conventions for describing the spaces of covariant, contravariant, and mixed
tensors are not universally agreed upon. Be sure to look closely at each author’s
conventions.)
Tensor Products
There is a natural product, called the tensor product, linking the various ten-
sor spaces over V : if F 2 T .k;l/ .V / and G 2 T .p;q/ .V /, the tensor F ˝ G 2
T .kCp;lCq/ .V / is defined by
F ˝ G.! 1 ; : : : ; ! kCp ; v1 ; : : : ; vlCq /

D F .! 1 ; : : : ; ! k ; v1 ; : : : ; vl /G.! kC1 ; : : : ; ! kCp ; vlC1 ; : : : ; vlCq /:
The tensor product is associative, so we can unambiguously form

tensor products of
three or more tensors on V . If .bi / is a basis for V and ˇ j is the associated dual
basis, then a basis for T .k;l/ .V / is given by the set of all tensors of the form
bi1 ˝ ˝ bik ˝ ˇ j1 ˝ ˝ ˇ jl ; (B.4)
as the indices ip , jq range from 1 to n. These tensors act on basis elements by

s
bi1 ˝ ˝ bik ˝ ˇ j1 ˝ ˝ ˇ jl .ˇ s1 ; : : : ; ˇ sk ; br1 ; : : : ; brl / D ıis11 ıikk ırj11 ırjll :
It follows that T .k;l/ .V / has dimension nkCl , where n D dim V . Every tensor F 2
T .k;l/ .V / can be written in terms of this basis (using the summation convention) as
i :::i
F D Fj11:::jkl bi1 ˝ ˝ bik ˝ ˇ j1 ˝ ˝ ˇ jl ; (B.5)
where
i :::i
Fj11:::jkl D F ˇ i1 ; : : : ; ˇ ik ; bj1 ; : : : ; bjl :
If the arguments of a mixed tensor F occur in a nonstandard order, then the
horizontal as well as vertical positions of the indices are significant and reflect which
arguments are vectors and which are covectors. For example, if A is a .1; 2/-tensor
whose first argument is a vector, second is a covector, and third is a vector, its basis
expression would be written
A D Ai j k ˇ i ˝ bj ˝ ˇ k ;
where
Ai j k D A bi ; ˇ j ; bk : (B.6)
There are obvious identifications among some of these tensor spaces:
T .0;0/ .V / D T 0 .V / D T 0 .V / D R;
T .1;0/ .V / D T 1 .V / D V;
T .0;1/ .V / D T 1 .V / D V ; (B.7)
T .k;0/ .V / D T k .V /;
T .0;k/ .V / D T k .V /:
A less obvious, but extremely important, identification is the following:
T .1;1/ .V / Š End.V /;
where End.V / denotes the space of linear maps from V to itself (also called the
endomorphisms of V ). This is a special case of the following proposition.
Proposition B.1. Let V be a finite-dimensional vector space. There is a natural

(basis-independent) isomorphism between T .kC1;l/ .V / and the space of multilinear
maps
š ™
V V V V ! V:
k copies l copies
I Exercise B.2. Prove Proposition B.1. [Hint: In the special case k D 0, l D 1, consider
the map ˚ W End.V / ! T .1;1/ .V / defined by letting ˚A be the .1; 1/-tensor defined by
˚A.!; v/ D !.Av/. The general case is similar.]
Tensors on a Vector Space 395
We can use the result of Proposition B.1 to define a natural operation called
trace or contraction, which lowers the rank of a tensor by 2. In one special case,
it is easy to describe: the operator tr W T .1;1/ .V / ! R is just the trace of F when
it is regarded as an endomorphism of V , or in other words the sum of the di-
agonal entries of any matrix representation of F . Since the trace of a linear en-
domorphism is basis-independent, this is well defined. More generally, we define
tr W T .kC1;lC1/ .V / ! T .k;l/ .V / by letting .tr F /.! 1 ; : : : ; ! k ; v1 ; : : : ; vl / be the trace
of the .1; 1/-tensor

F ! 1 ; : : : ; ! k ; ; v1 ; : : : ; vl ; 2 T .1;1/ .V /:
In terms of a basis, the components of tr F are

i :::i i :::i m
.tr F /j11 :::jkl D Fj11:::jkl m :
In other words, just set the last upper and lower indices equal and sum. Even more
generally, we can contract a given tensor on any pair of indices as long as one is
contravariant and one is covariant. There is no general notation for this operation,
so we just describe it in words each time it arises. For example, we can contract the
tensor A with components given by (B.6) on its first and second indices to obtain a
covariant 1-tensor B whose components are Bk D Ai i k .
I Exercise B.3. Show that the trace on any pair of indices (one upper and one lower) is
a well-defined linear map from T .kC1;lC1/ .V / to T .k;l/ .V /.
Symmetric Tensors
There are two classes of tensors that play particularly important roles in differential
geometry: the symmetric and alternating tensors. Here we discuss the symmetric
ones; we will take up alternating tensors later in this appendix when we discuss
differential forms.
If V is a finite-dimensional vector space, a covariant tensor F 2 T k .V / is said
to be symmetric if its value is unchanged by interchanging any pair of arguments:
F .v1 ; : : : ; vi ; : : : ; vj ; : : : ; vk / D F .v1 ; : : : ; vj ; : : : ; vi ; : : : ; vk /
whenever 1 i < j k. It follows immediately that the value is unchanged un-

der every rearrangement of the arguments. If we denote the components of F with
respect to a basis by Fi1 :::ik , then F is symmetric if and only if its components are
unchanged by every permutation of the indices i1 ; : : : ; ik . Every 0-tensor and every
1-tensor are vacuously symmetric. The set of symmetric k-tensors on V is a linear
subspace of T k .V /, which we denote by †k .V /.
A tensor product of symmetric tensors is generally not symmetric. However,
there is a natural product of symmetric tensors that does yield symmetric tensors.
Given a covariant k-tensor F , the symmetrization of F is the k-tensor Sym F de-
fined by
1 X
.SymF /.v1 ; : : : ; vk / D F v.1/ ; : : : ; v.k/ ;
kŠ
2Sk
where Sk is the group of all permutations of f1; : : : ; kg, called the symmetric group
on k elements. It is easy to check that SymF is symmetric, and SymF D F if and
only if F is symmetric. Then if F and G are symmetric tensors on V , of ranks k
and l, respectively, their symmetric product is defined to be the .k C l/-tensor F G
(denoted by juxtaposition without an explicit product symbol) given by
F G D Sym.F ˝ G/:
The most important special case is the symmetric product of two 1-tensors, which
can be characterized as follows: if ! and are covectors, then
! D 12 .! ˝ C ˝ !/:
(To prove this, just evaluate both sides on an arbitrary pair of vectors .v; w/, and
use the definition of !.) If ! is any 1-tensor, the notation ! 2 means the symmetric
product !!, which in turn is equal to ! ˝ !.
Tensor Bundles and Tensor Fields

On a smooth manifold M with or without boundary, we can perform the same linear-
algebraic constructions on each tangent space Tp M that we perform on any vector
space, yielding tensors at p. The disjoint union of tensor spaces of a particular type
at all points of the manifold yields a vector bundle, called a tensor bundle.
The most fundamental tensor bundle is the cotangent bundle, defined as
a
T M D Tp M:
p2M
More generally, the bundle of (k,l)-tensors on M is defined as

a
T .k;l/ TM D T .k;l/ .Tp M /:
p2M
As special cases, the bundle of covariant k-tensors is denoted by T k T M D

T .0;k/ TM , and the bundle of contravariant k-tensors is denoted by T k TM D
T .k;0/ TM . Similarly, the bundle of symmetric k-tensors is
a
†k T M D †k .Tp M /:
p2M
Tensor Bundles and Tensor Fields 397
There are the usual identifications among these bundles that follow from (B.7): for
example, T 1 TM D T .1;0/ TM D TM and T 1 T M D T .0;1/ TM D †1 T M D
T M .
I Exercise B.4. Show that each tensor bundle is a smooth vector bundle over M , with a
local trivialization over every open subset that admits a smooth local frame for TM .
A tensor field on M is a section of some tensor bundle over M (see p. 383 for the
definition of a section of a bundle). A section of T 1 T M D T .0;1/ TM (a covariant
1-tensor field) is also called a covector field. As we do with vector fields, we write
the value of a tensor field F at p 2 M as Fp or F jp . Because covariant tensor
fields are the most common and important tensor fields we work with, we use the
following shorthand notation for the space of all smooth covariant k-tensor fields:

T k .M / D T k T M :
The space of smooth 0-tensor fields is just C 1 .M /.

Let .Ei / D .E1 ; : : : ; En / be any smooth local frame for TM over an open sub-
set U M . Associated with such a frame is the dual coframe, which we typi-
cally denote by ."1 ; : : : ; "n /; these are smooth covector fields satisfying
"i .Ej / D
i 1 n
ıj . For example, given a coordinate frame @=@x
; : : : ; @=@x over some open
subset U M , the dual coframe is dx 1 ; : : : ; dx n , where dx i is the differential of the
coordinate function x i .
In terms of any smooth local frame .Ei / and its dual coframe ."i /, the tensor
fields Ei1 ˝ ˝ Eik ˝ "j1 ˝ ˝ "jl form
a smooth local frame for T
.k;l/
.T M /.
i
In particular, in local coordinates x , a .k; l/-tensor field F has a coordinate
expression of the form
i :::i
F D Fj11:::jkl @i1 ˝ ˝ @ik ˝ dx j1 ˝ ˝ dx jl ; (B.8)
i :::i
where each coefficient Fj11:::jkl is a smooth real-valued function on U .
I Exercise B.5. Suppose F W M ! T .k;l/ TM is a rough .k; l/-tensor field. Show that
F is smooth on an open set U M if and only if whenever ! 1 ; : : : ; ! k are smooth covec-
torfields and X1 ; : : : ; Xl are smooth vector fields defined on U , the real-valued function
F ! 1 ; : : : ; ! k ; X1 ; : : : ; Xl , defined on U by

F ! 1 ; : : : ; ! k ; X1 ; : : : ; Xl .p/ D Fp ! 1 jp ; : : : ; ! k jp ; X1 jp ; : : : ; Xl jp ;
is smooth.
An important property of tensor fields is that they

are multilinear over the space
of smooth functions. Suppose F 2 T .k;l/ TM is a smooth tensor field. Given
smooth covector fields ! 1 ; : : : ; ! k 2 T 1 .M / and smooth vector fields X 1 ; : : : ; Xl 2
X.M /, Exercise B.5 shows that the function F ! 1 ; : : : ; ! k ; X1 ; : : : ; Xl is smooth,
and thus F induces a map
œ
Fz W T 1 .M / T 1 .M / X.M / X.M / ! C 1 .M /:
k factors l factors
It is easy to check that this map is multilinear over C 1 .M /, that is, for all functions
u; v 2 C 1 .M / and smooth vector or covector fields ˛, ˇ,
Fz .: : : ; u˛ C vˇ; : : : / D uFz .: : : ; ˛; : : : / C v Fz .: : : ; ˇ; : : : /:
Even more important is the converse: as the next lemma shows, every such map that
is multilinear over C 1 .M / defines a tensor field. (This lemma is stated and proved
in [LeeSM] for covariant tensor fields, but the same argument works in the case of
mixed tensors.)
Lemma B.6 (Tensor Characterization Lemma). [LeeSM, Lemma 12.24] A map
œ
F W T 1 .M / T 1 .M / X.M / X.M / ! C 1 .M /
k factors l factors
is induced by a smooth .k; l/-tensor field as above if and only if it is multilinear over
C 1 .M /. Similarly, a map
œ
F W T 1 .M / T 1 .M / X.M / X.M / ! X.M /
k factors l factors
is induced by a smooth .k C 1; l/-tensor field as in Proposition B.1 if and only if it

is multilinear over C 1 .M /.
Because of this result, it is common to use the same symbol for both a tensor
field and the multilinear map on sections that it defines, and to refer to either of
these objects as a tensor field.
Pullbacks of Tensor Fields
Suppose F W M ! N is a smooth map and A is a covariant k-tensor field on N .

For every p 2 M , we define a tensor dFp .A/ 2 T k Tp M ; called the pointwise
pullback of A by F at p, by

dFp .A/.v1 ; : : : ; vk / D A dFp .v1 /; : : : ; dFp .vk /
for v1 ; : : : ; vk 2 Tp M ; and we define the pullback of A by F to be the tensor field

F A on M defined by
.F A/p D dFp AF .p/ :
Proposition B.7 (Properties of Tensor Pullbacks). [LeeSM, Prop. 12.25] Suppose

F W M ! N and G W P ! M are smooth maps, A and B are covariant tensor fields
on N; and f is a real-valued function on N .
Tensor Bundles and Tensor Fields 399
(a) F .f B/ D .f ı F /F B.
(b) F .A ˝ B/ D F A ˝ F B.
(c) F .A C B/ D F A C F B.
(d) F B is a (continuous) tensor field, and it is smooth if B is smooth.
(e) .F ı G/ B D G .F B/.
(f) .IdN / B D B.
If f is a continuous real-valued function (i.e., a 0-tensor field), the pullback F f

is just the composition f ı F .
The following proposition shows how pullbacks are computed in local coordi-
nates.
Proposition B.8. [LeeSM, Cor. 12.28] Let F W M ! N be smooth, and let B be a

covariant k-tensor field on N . If p 2 M and y i are smooth coordinates for N on
a neighborhood of F .p/, then F B has the following expression in a neighborhood
of p:

F Bi1 :::ik dy i1 ˝ ˝ dy ik D Bi1 :::ik ı F d y i1 ı F ˝ ˝ d y ik ı F :
Lie Derivatives of Tensor Fields
We can extend the notion of Lie derivatives to tensor fields. This can be done for
mixed tensor fields of any rank, but for simplicity we restrict attention to covariant
tensor fields. Suppose X is a smooth vector field on M and is its flow. If A is a
smooth covariant tensor field on M , the Lie derivative of A with respect to X is
the smooth covariant tensor field LX A defined by
ˇ
d ˇˇ
d.t /p At .p/ Ap
.LX A/p D . A/p D lim :
dt ˇtD0 t t!0 t
Proposition B.9. [LeeSM, Thm. 12.32] Suppose M is a smooth manifold, X 2

X.M /, and A is a smooth covariant k-tensor field on M . For all smooth vector
fields Z1 ; : : : ; Zk ,

.LX A/.Z1 ; : : : ; Zk / D X A.Z1 ; : : : ; Zk /
A.LX Z1 ; Z2 ; : : : ; Zk / A.Z1 ; LX Z2 ; : : : ; Zk / A.Z1 ; : : : ; LX Zk /:
As we did for vector fields, we say that a covariant tensor field A is invariant
under a flow if .t / A D A wherever it is defined. The next proposition is a
tensor analogue of Proposition A.47.
Proposition B.10. [LeeSM, Thm. 12.37] Let M be a smooth manifold and X 2

X.M /. A smooth covariant tensor field is invariant under the flow of X if and only
if its Lie derivative with respect to X is identically zero.
Differential Forms and Integration
In addition to symmetric tensors, the other class of tensors that play a special role
in differential geometry is that of alternating tensors, which we now define. Let
V be a finite-dimensional vector space. If F is a covariant k-tensor on V , we say
that F is alternating if its value changes sign whenever two different arguments are
interchanged:
F .v1 ; : : : ; vi ; : : : ; vj ; : : : ; vk / D F .v1 ; : : : ; vj ; : : : ; vi ; : : : ; vk /:
The set of alternating covariant k-tensors on V is a linear subspace of T k .V /,

denoted by ƒk .V /. If F 2 ƒk .V /, then the effect of an arbitrary permutation of
its arguments is given by
F .v.1/ ; : : : ; v.k/ / D .sgn /F .v1 ; : : : ; vk /;
where sgn represents the sign of the permutation 2 Sk , which is C1 if is even

(i.e., can be written as a composition of an even number of transpositions), and 1
if is odd. The components of F with respect to any basis change sign similarly
under a permutation of the indices. All 0-tensors and 1-tensors are alternating.
Analogously to the symmetrization operator defined above, if F is any covariant
k-tensor on V , we define the alternation of F by
1 X
.Alt F /.v1 ; : : : ; vk / D .sgn /F v.1/ ; : : : ; v.k/ : (B.9)
kŠ
2Sk
Again, it is easy to check that Alt F is alternating, and Alt F D F if and only if F is
alternating. Given ! 2 ƒk .V / and 2 ƒl .V /, we define their wedge product by
.k C l/Š
! ^ D Alt.! ˝ /:
kŠlŠ
It is immediate that ! ^ is an alternating .k C l/-tensor. The wedge product is
easily seen to be bilinear and anticommutative, which means that
! ^ D .1/kl ^ !; for ! 2 ƒk .V / and 2 ƒl .V /:
It is also the case, although not so easy to see, that it is associative; see [LeeSM,
Prop. 14.11] for a proof. If .bi / is a basis for V and .ˇ i / is the dual basis, the wedge
products ˇ i1 ^ ^ ˇ ik , as .i1 ; : : : ; ik / range over strictly increasing
multi-indices,

form a basis for ƒk .V /, which therefore has dimension kn D nŠ= kŠ.n k/Š ,
where n D dim V . In terms of these basis elements, the wedge product satisfies
i
ˇ i1 ^ ^ ˇ ik .bj1 ; : : : ; bjk / D det ıjpq :
More generally, if ! 1 ; : : : ; ! k are any covectors and v1 ; : : : ; vk are any vectors, then
Differential Forms and Integration 401

! 1 ^ ^ ! k .v1 ; : : : ; vk / D det ! i .vj / : (B.10)
The wedge product can be defined analogously for contravariant alternating ten-
sors ˛ 2 ƒp .V /, ˇ 2 ƒq .V /, simply by regarding ˛, ˇ, and ˛ ^ ˇ as alternating
multilinear functionals on V .
The convention we use for the wedge product is referred to in [LeeSM] as the
determinant convention. There is another convention that is also in common use,
the Alt convention, which amounts to multiplying the right-hand side of (B.10) by a
factor of 1=kŠ. The choice of which definition to use is a matter of taste, though there
are various reasons to justify each choice depending on the context. The determinant
convention is most common in introductory differential geometry texts, and is used,
for example, in [Boo86, Cha06, dC92, LeeJeff09, LeeSM, Pet16, Spi79, Tu11]. The
Alt convention is used in [KN96] and is more common in complex differential geom-
etry.
Given an alternating k-tensor ! 2 ƒk .V / and a vector v 2 V , we define an
alternating .k 1/-tensor v ³ ! by
.v ³ !/.w1 ; : : : ; wk1 / D !.v; w1 ; : : : ; wk1 /:
The operation ! 7! v ³ ! is known as interior multiplication by v, and is also

denoted by ! 7! iv !. By convention, v ³ ! D 0 when ! is a 0-tensor. Direct com-
putation shows that iv ı iv D 0, and iv satisfies the following product rule for an
alternating k-tensor ! and an alternating l-tensor :
iv .! ^ / D .iv !/ ^ C .1/k ! ^ .iv /: (B.11)
If M is a smooth manifold with or without boundary, the subbundle of T k T M

consisting of alternating tensors is denoted by ƒk T M , and an alternating tensor
field on M is called a differential k-form,
or just a k-form. The space of all smooth
k-forms is denoted by k .M / D ƒk T M . Because every 1-tensor field is alter-
nating, 1 .M / is the same as the space T 1 .M / of smooth covector fields.
Exterior Derivatives
The most important operation on differential forms is the exterior derivative, defined

as follows. Suppose M is a smooth n-manifold with or without boundary, and x i
are any smooth local coordinates on M . A smooth k-form ! can be expressed in
these coordinates as
X
!D !j1 :::jk dx j1 ^ ^ dx jk ;
j1 <<jk
and then we define the exterior derivative of !, denoted by d!, to be the .k C 1/-
form defined in coordinates by
X X
n
@!j 1 :::jk
d! D dx i ^ dx j1 ^ ^ dx jk :
@x i
j1 <<jk iD1
For a 0-form f (a smooth real-valued function), this reduces to
X
n
@f
df D dx i ;
@x i
iD1
which is exactly the differential of f defined by (A.4).
Proposition B.11 (Properties of Exterior Differentiation). [LeeSM, Thm. 14.24]

Suppose M is a smooth n-manifold with or without boundary.
(a) For each k D 0; : : : ; n, the operator d W k .M / ! kC1 .M / is well defined,
independently of coordinates.
(b) d is linear over R.
(c) d ı d 0.
(d) If ! 2 k .M / and 2 l .M /, then
d.! ^ / D d! ^ C .1/k ! ^ d:
Proposition B.12 (Exterior Derivative of a 1-form). [LeeSM, Prop. 14.29] If ! is

a smooth 1-form and X; Y are vector fields, then
d!.X; Y / D X.!.Y // Y .!.X // !.ŒX; Y /:
Proposition B.13 (Naturality of the Exterior Derivative). [LeeSM, Prop. 14.26]

If F W M ! N is a smooth map, then for each k, the pullback map F W k .N / !
k .M / commutes with d : for all ! 2 k .N /,
F .d!/ D d.F !/: (B.12)
A smooth differential form ! 2 k .M / is closed if d! D 0, and exact if there

exists a smooth .k 1/-form on M such that ! D d. The fact that d ı d D
0 implies that every exact form is closed, but the converse is not true in general.
However, the next lemma gives an important special case in which it is true. If V
is a vector space, a subset S V is said to be star-shaped with respect to a point
x 2 S if for every y 2 S , the line segment from x to y is contained in S .
Lemma B.14 (The Poincaré Lemma). [LeeSM, Thm. 17.14] If U is a star-shaped

open subset of Rn , then for k 1, every closed k-form on U is exact.
Orientations
If V is a finite-dimensional vector space, an orientation of V is an equivalence
class of ordered bases for V , where two ordered bases are considered equivalent
Differential Forms and Integration 403
if the transition matrix that expresses one basis in terms of the other has posi-
tive determinant. Every vector space has exactly two orientations. Once an orien-
tation is chosen, a basis is said to be positively oriented if it belongs to the chosen
orientation, and negatively oriented if not. The standard orientation of Rn is the one
determined by the standard basis .e1 ; : : : ; en /, where ei is the vector .0; : : : ; 1; : : : ; 0/
with 1 in the i th place and zeros elsewhere.
If M is a smooth manifold, an orientation for M is a choice of orientation for
each tangent space that is continuous in the sense that in a neighborhood of every
point there is a (continuous) local frame that determines the given orientation at
each point of the neighborhood. If there exists an orientation for M , we say that
M is orientable. An oriented manifold is a smooth orientable manifold together
with a choice of orientation.
i If M is an oriented n-manifold, then a smooth
coordinate
chart U;
x is said to be an oriented chart if the coordinate frame
@=@x 1 ; : : : ; @=@x n is positively oriented at each point. Exactly the same definitions
apply to manifolds with boundary.
Proposition B.15 (Orientation Determined by an n-Form). [LeeSM, Prop. 15.5]

Let M be an n-manifold with or without boundary. Every nonvanishing n-form
2 n .M / determines a unique orientation of M by declaring an ordered basis
.b1 ; : : : ; bn / for Tp M to be positively oriented if and only if p .b1 ; : : : ; bn / > 0. Con-
versely, if M is oriented, there is a smooth nonvanishing n-form that determines the
orientation in this way.
Because of this proposition, a nonvanishing n-form on a smooth n-manifold is

called an orientation form. If in addition M is oriented and determines the given
orientation, we say that is positively oriented.
Suppose M and N are both smooth n-manifolds with or without boundary and
F W M ! N is a local diffeomorphism. If N is oriented, we define the pullback
orientation on M induced by F to be the orientation determined by F , where
is any positively oriented orientation form for N . If both M and N are oriented,
we say that F is orientation-preserving if the pullback orientation is equal to the
given orientation on M , and orientation-reversing if the pullback orientation is the
opposite orientation.
Proposition B.16 (Orientation of a Hypersurface). [LeeSM, Prop. 15.21] Sup-

pose M is an oriented smooth n-manifold with or without boundary, S M is a
smooth immersed hypersurface, and N is a continuous vector field along S (i.e.,
a continuous map N W S ! TM such that Np 2 Tp M for each p 2 S ). If N is
nowhere tangent to S , then S has a unique orientation determined by the .n 1/-
form .N ³ /, where W S ,! M is inclusion and is any positively oriented
orientation form for M .
A special case of the preceding proposition occurs when M is a manifold with

boundary and the hypersurface in question is the boundary of M . In that case we
declare a vector field N along @M to be outward-pointing if for each p 2 @M ,
there is a smooth curve W ."; 0 ! M such that .0/ D p and 0 .0/ D Np , and
Np … Tp .@M /. We can always construct a global smooth outward-pointing vector
field by taking @=@x n in boundary coordinates in a neighborhood of each boundary

point, and gluing together with a partition of unity. Since an outward-pointing vector
field is nowhere tangent to @M , we have the following proposition.
Proposition B.17 (Induced Orientation on a Boundary). [LeeSM, Prop. 15.24]

If M is an oriented smooth manifold with boundary, then @M is orientable, and it
has a canonical orientation (called the induced orientation or Stokes orientation)
determined by .N ³ /, where W @M ,! M is inclusion, N is any outward-
pointing vector field along @M , and is any positively oriented orientation form
for M .
Proposition B.18 (Orientation Covering Theorem). [LeeSM, Thm. 15.41] If M

is a connected, nonorientable smooth manifold, then there exist an oriented smooth
manifold M and a two-sheeted smooth covering map yW M ! M , called the
orientation covering of M .
Corollary B.19. Every simply connected smooth manifold is orientable.
Proof. Corollary A.59 shows that a simply connected manifold does not admit two-
sheeted covering maps. t
u
Integration of Differential Forms
Suppose first that ! is a (continuous) n-form on an open subset U Rn or RnC . It

can be written ! D f dx 1 ^ ^ dx n for some continuous real-valued function f ,
and we define the integral of ! over U to be the ordinary multiple integral
Z Z

!D f x 1 ; : : : ; x n dx 1 dx n ;
U U
provided the integral is well defined. This will always be the case, for example, if f
is continuous and has compact support in U .
In general, if ! is a compactly supported n-form on a smooth n-manifold M
with or without boundary, we define the integral of ! over M by choosing finitely
many oriented smooth coordinate charts fUi gkiD1 whose domains cover the support
of !, together with a smooth partition of unity f 1 gkiD1 subordinate to this cover,
and defining
Z X k Z
1
!D 'i . i !/;
M y
iD1 Ui
where Uyi D 'i .Ui /, and the integrals on the right-hand side are defined as above.
Proposition 16.5 in [LeeSM] shows that this definition does not depend on the
choice of oriented charts or partition of unity.
Proposition B.20 (Properties of Integrals of Forms). [LeeSM, Prop. 16.6] Sup-

pose M and N are nonempty oriented smooth n-manifolds with or without bound-
ary, and !; are compactly supported n-forms on M .
Densities 405
(a) L INEARITY: If a; b 2 R, then

Z Z Z
a! C b D a ! Cb :
M M M
(b) O RIENTATION R EVERSAL : If M denotes M with the opposite orientation,

then Z Z
!D !:
M M
R
(c) P OSITIVITY: If ! is a positively oriented orientation form, then M ! > 0.
(d) D IFFEOMORPHISM I NVARIANCE : If F W N ! M is an orientation-preserving
or orientation-reversing diffeomorphism, then
‚ Z
Z F ! if F is orientation-preserving,
N
!D Z
F ! if F is orientation-reversing.
M

N
Theorem B.21 (Stokes’s Theorem). [LeeSM, Thm. 16.11] If M is an oriented

smooth n-manifold with boundary and ! is a compactly supported smooth .n 1/-
form on M , then Z Z
d! D !: (B.13)
M @M
In the statement of this theorem, @M is understood to have the Stokes orientation,

and the right-hand side is interpreted as the integral of the pullback of ! to @M .
The following special case is frequently useful.
Corollary B.22. [LeeSM, Cor. 16.13] Suppose M is a compact oriented smooth
n-manifold (without boundary). Then the integral of every exact n-form on M is
zero.
Densities
On an oriented n-manifold with or without boundary, n-forms are the natural objects
to integrate. But in order to integrate on a nonorientable manifold, we need closely
related objects called densities.
If V is an n-dimensional real vector space, a density on V is a function
™
W V V ! R
n copies
satisfying the following formula for every linear map T W V ! V :
.T v1 ; : : : ; T vn / D jdet T j .v1 ; : : : ; vn /: (B.14)

A density is said to be positive if .v1 ; : : : ; vn / > 0 whenever .v1 ; : : : ; vn / is a

basis of V ; it is clear from (B.14) that if this is true for some basis, then it is true
for every one. Every nonzero alternating n-tensor determines a positive density
jj by the formula
jj .v1 ; : : : ; vn / D j.v1 ; : : : ; vn /j :
The set D.V / of all densities on V is a 1-dimensional vector space, spanned by jj
for any nonzero alternating n-tensor .
When M is a smooth manifold with or without boundary, the set
a
DM D D.Tp M /
p2M
ˇis called the density

ˇ bundle of M . It is a smooth rank-1 vector bundle, with
ˇdx 1 ^ ^ dx n ˇ as a smooth local frame over any smooth coordinate chart. A den-
sity on M is a (smooth) section of DM ; in any local coordinates, it can be written
ˇ ˇ
D u ˇdx 1 ^ ^ dx n ˇ
for some locally defined smooth function u.

Under smooth maps, densities pull back in the same way as differential forms:
Suppose F W M ! N is a smooth map between n-manifolds with or without bound-
ary, and is a density on N . The pullback of is the density F on M defined by
.F /p .v1 ; : : : ; vn / D F .p/ .dFp .v1 /; : : : ; dFp .vn //:
In coordinates, this satisfies

ˇ ˇ ˇ ˇ
F u ˇdy 1 ^ ^ dy n ˇ D .u ı F / jdet DF j ˇdx 1 ^ ^ dx n ˇ ; (B.15)
where DF represents the matrix of partial derivatives of F in these coordinates.

Because the pullback formula (B.15) is exactly analogous to the change of vari-
ables formula for multiple integrals, we can define the integral of a compactly sup-
ported density on a smooth manifold with or without boundary in exactly the same
way as the integral of an n-form is defined on an oriented manifold, except that now
there is no need to have an orientation. Thus if D u jdx 1 dx n j is a smooth den-
sity that is compactly supported in an open set V Rn , we define the integral of
by Z Z
D u dx 1 dx n :
V V
It follows from (B.15) that the value of this integral is diffeomorphism-invariant, so
we can define the integral of a compactly supported density on a smooth manifold
with or without boundary by breaking it up with a partition of unity and integrating
each term in coordinates as above.
Appendix C
Review of Lie Groups
Lie groups play many important roles in Riemannian geometry, both as examples of
Riemannian manifolds and as isometry groups of other manifolds. In this appendix,
we summarize the main facts about Lie groups that are used in this book. For details,
consult [LeeSM], especially Chapters 7, 8, 20, and 21.
Definitions and Properties

A Lie group is a smooth manifold G that is also a group in the algebraic sense, with
the property that the multiplication map m W G G ! G given by m.'1 ; '2 / D '1 '2
and the inversion map i W G ! G given by i.'/ D ' 1 are smooth. For generic Lie
groups, we usually denote the identity element by e, unless there is a more specific
common notation for a particular group.
Given a Lie group G, each ' 2 G defines a map L' W G ! G, called left trans-
lation, by L' .' 0 / D ' ' 0 . It is a diffeomorphism with inverse L' 1 . Similarly, right
translation R' W G ! G is the diffeomorphism R' .' 0 / D ' 0 ', with inverse R' 1 .
A subgroup H G that is endowed with a topology and smooth structure making
it into a Lie group and an immersed or embedded submanifold of G is called a Lie
subgroup of G .
If G and H are Lie groups, a group homomorphism F W G ! H that is also a
smooth map is called a Lie group homomorphism. It is a Lie group isomorphism if
it has an inverse that is also a Lie group homomorphism. A Lie group isomorphism
from G to itself is called an automorphism of G .
Proposition C.1. [LeeSM, Thms. 7.5 & 21.27] Suppose F W G ! H is a Lie group
homomorphism.
(a) F has constant rank.
(b) The kernel of F is an embedded Lie subgroup of G.
(c) The image of F is an immersed Lie subgroup of H .

408 Review of Lie Groups
Example C.2 (Lie Groups).

(a) Every countable group with the discrete topology is a zero-dimensional Lie
group, called a discrete Lie group.
(b) Rn and C n are Lie groups under addition.
(c) The sets R , RC , and C of nonzero real numbers, positive real numbers, and
nonzero complex numbers, respectively, are Lie groups under multiplication.
(d) The unit circle S1 C is a 1-dimensional Lie group under complex multipli-
cation, called the circle group.
(e) Given Lie groups G1 ; : : : ; Gk ; their direct product group is the Lie group
whose underlying manifold is G1 Gk , and whose multiplication is given
by .g1 ; : : : ; gk / g10 ; : : : ; gk0 D g1 g10 ; : : : ; gk gk0 . For example, the n-torus is
the n-fold product group T n D S1 S1 . The 2-torus is often simply called
the torus.
(f) The general linear groups GL.n; R/ and GL.n; C/, consisting of all invert-
ible n n real or complex matrices, respectively, are Lie groups under matrix
multiplication. The identity element in both cases is the n n identity matrix,
denoted by In . More generally, if V is an n-dimensional real or complex vec-
tor space, the group GL.V / of invertible linear maps from V to itself is a Lie
group isomorphic to GL.n; R/ or GL.n; C/.
(g) The real and complex special linear groups are the subgroups SL.n; R/
GL.n; R/ and SL.n; C/ GL.n; C/ consisting of matrices of determinant 1.
(h) The orthogonal group O.n/ GL.n; R/ is the subgroup consisting of orthog-
onal matrices, those that satisfy AT A D In , where AT is the transpose of A.
The special orthogonal group is the subgroup SO.n/ D O.n/ \ SL.n; R/.
(i) Similarly, the unitary group U.n/ GL.n; C/ is the subgroup consisting of
complex unitary matrices, those that satisfy A A D In , where A D Ax T is
the conjugate transpose of A, called its adjoint. The special unitary group is
SU.n/ D U.n/ \ SL.n; C/. //
If G is a Lie group and V is a finite-dimensional vector space, a Lie group ho-

momorphism W G ! GL.V / is called a (finite-dimensional) representation of G .
It is said to be faithful if it is injective.
The Lie Algebra of a Lie Group

Suppose G is a Lie group. A vector field X on G is said to be left-invariant if it
is invariant under all left translations, meaning that .Lg / X D X for every g 2 G.
Similarly, X is right-invariant if it is invariant under all right translations.
We denote the set of all smooth left-invariant vector fields on G by Lie.G/. It
is a vector subspace of X.G/ that is closed under Lie brackets (see [LeeSM, Prop.
8.33]). Thus it is an example of a Lie algebra: a vector space g endowed with a bilin-
ear operation Œ; W g g ! g, called the bracket, that is antisymmetric (meaning
ŒX; Y D ŒY; X ) and satisfies the Jacobi identity:
The Lie Algebra of a Lie Group 409
ŒX; ŒY; Z C ŒY; ŒZ; X C ŒZ; ŒX; Y D 0:
With the Lie bracket structure that it inherits from X.G/, the Lie algebra Lie.G/ is
called the Lie algebra of G .
Proposition C.3. [LeeSM, Thm. 8.37] Let G be a Lie group with identity e. The
evaluation map X 7! Xe is a vector space isomorphism from Lie.G/ to Te G, so
Lie.G/ has the same dimension as G itself.
If g is a Lie algebra, a Lie subalgebra of g is a vector subspace h g that is

closed under Lie brackets, and is thus a Lie algebra with the restriction of the same
bracket operation. Given Lie algebras g1 and g2 , a Lie algebra homomorphism is a
linear map F W g1 ! g2 that preserves brackets: F .ŒX; Y / D ŒF .X /; F .Y /. A Lie
algebra isomorphism is an invertible Lie algebra homomorphism, and a Lie algebra
automorphism is a Lie algebra isomorphism from a Lie algebra to itself.
Example C.4 (Lie Algebras).

(a) If M is a smooth positive-dimensional manifold, the space X.M / of all smooth
vector fields on M is an infinite-dimensional Lie algebra under the Lie bracket.
(b) The vector space gl.n; R/ of all n n real matrices is an n2 -dimensional
Lie algebra under the commutator bracket ŒA; B D AB BA. The canoni-
cal identification of both gl.n; R/ and Lie.GL.n; R// with the tangent space
to GL.n; R/ at the identity yields a Lie algebra isomorphism gl.n; R/ Š
Lie.GL.n; R// [LeeSM, Prop. 8.41].
(c) Similarly, the space gl.n; C/ of n n complex matrices is a 2n2 -dimensional
(real) Lie algebra under the commutator bracket, isomorphic to the Lie algebra
of GL.n; C/ [LeeSM, Prop. 8.48].
(d) The space o.n/ of skew-symmetric n n real matrices is a Lie subalgebra of
gl.n; R/, isomorphic to the Lie algebra of O.n/ [LeeSM, Example 8.47].
(e) If V is a vector space, the space gl.V / of all linear maps from V to itself is a
Lie algebra under the commutator bracket ŒA; B D A ı B B ı A. In case V
is finite-dimensional, gl.V / is isomorphic to the Lie algebra of the Lie group
GL.V /.
(f) A Lie algebra in which all brackets are zero is called an abelian Lie algebra.
Every vector space can be made into an abelian Lie algebra by defining all
brackets to be zero. The Lie algebra of a connected Lie group is abelian if and
only if the group is abelian (see [LeeSM, Problems 8-25 and 20-7]). //
The following proposition shows that every Lie group homomorphism induces a
Lie algebra homomorphism between the respective Lie algebras.
Proposition C.5 (The Induced Lie Algebra Homomorphism). [LeeSM, Thm.

8.44] Let G and H be Lie groups, and let g and h be their Lie algebras. Given
a Lie group homomorphism F W G ! H , there is a unique Lie algebra homomor-
phism F W g ! h, called the induced Lie algebra homomorphism of F , with the
property that for each X 2 g, the vector field F X 2 h is F -related to X .
Proposition C.6 (Properties of Induced Homomorphisms). [LeeSM, Prop. 8.45]

Let G; H; K be Lie groups.
(a) .IdG / D IdLie.G/ W Lie.G/ ! Lie.G/.
(b) If F1 W G ! H and F2 W H ! K are Lie group homomorphisms, then
.F2 ı F1 / D .F2 / ı .F1 / W Lie.G/ ! Lie.K/:
(c) Isomorphic Lie groups have isomorphic Lie algebras.
The Exponential Map of a Lie Group

If G is a Lie group, a Lie group homomorphism from R (with its additive group
structure) to G is called a one-parameter subgroup of G . (Note that the term “one-
parameter subgroup” refers, confusingly, to a homomorphism, not to a Lie subgroup
of G. But the image of a one-parameter subgroup is always a Lie subgroup of di-
mension at most 1.) Theorem 20.1 of [LeeSM] shows that for every X 2 Lie.G/,
the integral curve of X starting at the identity is a one-parameter subgroup, called
the one-parameter subgroup generated by X , and every one-parameter subgroup
is of this form.
The exponential map of G is the map expG W Lie.G/ ! G defined by setting
expG .X / D .1/, where is the integral curve of X starting at e. (The exponential
map is more commonly denoted simply by exp, but we use the notation expG to
distinguish the Lie group exponential map from the Riemannian exponential map,
introduced in Chapter 5.)
Proposition C.7 (Properties of the Lie Group Exponential Map). [LeeSM, Props.
20.5 & 20.8] Let G be a Lie group and let g be its Lie algebra.
(a) expG W g ! G is smooth.
(b) For every X 2 g, the map W R ! G defined by .t / D expG .tX / is the one-
parameter subgroup generated
by X .
(c) The differential d expG 0 W T0 g ! Te G is the identity map, under the canon-
ical identifications of T0 g and Te G with g.
The exponential map is the key ingredient in the proof of the following funda-
mental result.
Theorem C.8 (Closed Subgroup Theorem). [LeeSM, Thm. 20.12 & Cor. 20.13]
Suppose G is a Lie group and H G is a subgroup in the algebraic sense. Then H
is an embedded Lie subgroup of G if and only if it is closed in the topological sense.
An important special case of the closed subgroup theorem is that of a discrete

subgroup, that is, a subgroup that is discrete in the subspace topology.
Proposition C.9. [LeeSM, Prop. 21.28] Every discrete subgroup of a Lie group is a
closed Lie subgroup of dimension zero.
Group Actions on Manifolds 411
Adjoint Representations
Let G be a Lie group and g its Lie algebra. For every ' 2 G, conjugation by
' gives a Lie group automorphism C' W G ! G, called an inner automorphism,
by C' . / D ' ' 1 . Let Ad.'/ D .C' / W g ! g be the induced Lie algebra au-
tomorphism. It follows from the definition that C'1 ı C'2 D C'1 '2 , and therefore
Ad.'1 / ı Ad.'2 / D Ad.'1 '2 /; in other words, Ad W G ! GL.g/ is a representation,
called the adjoint representation of G . Proposition 20.24 in [LeeSM] shows that
Ad is a smooth map.
Example C.10 (Adjoint Representations of Lie Groups).

(a) If G is an abelian Lie group, then C' is the identity map for every ' 2 G, so
the adjoint representation is trivial: Ad.'/ D Idg for all ' 2 G.
(b) Suppose G is a Lie subgroup of GL.n; R/. Then for each A 2 G, the conju-
gation map CA .B/ D ABA1 is the restriction to G of a linear map from the
space M.n; R/ of n n matrices to itself. Its differential, therefore, is given
by the same formula: Ad.A/X D AXA1 for every A 2 G and X 2 Lie.G/
gl.n; R/. //
There is also an adjoint representation for Lie algebras. If g is any Lie algebra, a
representation of g is a Lie algebra homomorphism from g to gl.V /, the Lie algebra
of all linear endomorphisms of some vector space V . For each X 2 g, define a
map ad.X / W g ! g by ad.X /Y D ŒX; Y . This defines ad as a map from g to gl.g/,
and a straightforward computation shows that it is a representation of g, called the
adjoint representation of g.
The next proposition shows how the two adjoint representations are related.
Proposition C.11. [LeeSM, Thm. 20.27] Let G be a Lie group and g its Lie
algebra, and let Ad W G ! GL.g/ and ad W g ! gl.g/ be their respective adjoint
representations. The induced Lie algebra representation Ad W g ! gl.g/ is given
by Ad D ad.
Group Actions on Manifolds

The most important applications of Lie groups in differential geometry involve their
actions on other manifolds.
First we consider actions by abstract groups, not necessarily endowed with a
smooth manifold structure or even a topology. Let G be a group and M a set.
A left action of G on M is a map G M ! M , usually written .'; p/ 7! ' p,
satisfying '1 .'2 p/ D .'1 '2 / p and e p D p for all '1 ; '2 2 G and p 2 M ,
where e is the identity element of G. Similarly, a right action of G on M is a map
M G ! M satisfying .p '1 / '2 D p .'1 '2 / and p e D p. In some cases, it will
be important to give a name to an action such as W G M ! M , in which case we
write ' .p/ in place of ' p, with a similar convention for right actions. Since right
actions can be converted to left actions and vice versa by setting ' p D p ' 1 ,
for most purposes we lose no real generality by restricting attention to left actions.
(But there are also situations in which right actions arise naturally.)
Given an action of G on M , for each p 2 M the isotropy subgroup at p is
the subgroup Gp G consisting of all elements that fix p: that is, Gp D f' 2 G W
' p D pg. The group action is said to be free if ' p D p for some ' 2 G and
p 2 M implies ' D e, or in other words, if Gp D feg for every p. It is said to be
effective if '1 p D '2 p for all p if and only if '1 D '2 , or equivalently, if the only
element of G that fixes every element of M is the identity. For effective actions, each
element of G is uniquely determined by the map p 7! ' p. In such cases, we will
sometimes use the same notation ' to denote either the element ' 2 G or the map
p 7! ' p. The action is said to be transitive if for every pair of points p; q 2 M ,
there exists ' 2 G such that ' p D q.
If M is a smooth manifold, an action by a group G on M is said to be an action
by diffeomorphisms if for each ' 2 G, the map p 7! ' p is a diffeomorphism
of M . If G is a Lie group, an action of G on a smooth manifold M is said to be
a smooth action if the defining map G M ! M is smooth. In this case, it is
also an action by diffeomorphisms, because each map p 7! ' p is smooth and has
p 7! ' 1 p as an inverse. If G is any countable group, an action by G on M is an
action by diffeomorphisms if and only if it is a smooth action when G is regarded
as a 0-dimensional Lie group with the discrete topology.
Example C.12 (Semidirect Products). Group actions provide an important way to

construct Lie groups out of other Lie groups. Suppose H and N are Lie groups
and W H N ! N is a smooth left action of H on N by automorphisms of N ,
meaning that for each h 2 H , the map h W N ! N given by h .n/ D h n is a
Lie group automorphism. The semidirect product of H and N determined by ,
denoted by N Ì H , is the Lie group whose underlying manifoldis the Cartesian
product N H , and whose group multiplication is .n; h/.n0 ; h0 / D nh .n0 /; hh0 . //
I Exercise C.13. Verify that N Ì H is indeed a Liegroup with the multiplication
de-
fined above, and with .e; e/ as identity and .n; h/1 D h1 .n1 /; h1 .
Now suppose G is a Lie group, and M and N are smooth manifolds endowed
with G-actions (on the left, say). A map F W M ! N is said to be equivariant with
respect to the given G actions if F .' x/ D ' F .x/ for all ' 2 G and x 2 M .
Theorem C.14 (Equivariant Rank Theorem). [LeeSM, Thm. 7.25] Suppose M; N

are smooth manifolds, and G is a Lie group acting smoothly and transitively on M and
smoothly on N . If F W M ! N is a smooth G-equivariant map, then F has con-
stant rank. Thus, if F is surjective, it is a smooth submersion; if it is injective, it is a
smooth immersion; and if it is bijective, it is a diffeomorphism.
A smooth action of G on M is said to be a proper action if the map G M !

M M defined by .'; x/ 7! .' x; x/ is a proper map, meaning that the preimage of
every compact set is compact. The following characterization is usually the easiest
way to prove that a given action is proper.
Group Actions on Manifolds 413
Proposition C.15. [LeeSM, Prop. 21.5] Suppose G is a Lie group acting smoothly
on a smooth manifold M . The action is proper if and only if the following condition
is satisfied: if .pi / is a sequence in M and .'i / is a sequence in G such that both
.pi / and .'i pi / converge, then a subsequence of .'i / converges.
Corollary C.16. Every smooth action by a compact Lie group on a smooth manifold
is proper.
Proof. If G is a compact Lie group, then every sequence in G has a convergent
subsequence, so every smooth G-action is proper by the preceding proposition. u
t
The next theorem is the most important application of proper actions. If a group
G acts on a manifold M , then each p 2 M determines a subset G p D f' p W ' 2 Gg,
called the orbit of p. Because two orbits are either identical or disjoint, the orbits
form a partition of G. The set of orbits is denoted by M=G, and with the quotient
topology it is called the orbit space of the action.
Theorem C.17 (Quotient Manifold Theorem). [LeeSM, Thm. 21.10] Suppose G
is a Lie group acting smoothly, freely, and properly on a smooth manifold M . Then
the orbit space M=G is a topological manifold whose dimension is equal to the
difference dim M dim G; and it has a unique smooth structure with the property
that the quotient map W M ! M=G is a smooth submersion.
Example C.18 (Real Projective Spaces). For each nonnegative integer n, the n-
dimensional real projective space, denoted by RP n , is defined as the set of one-
dimensional linear subspaces of RnC1 . It can be identified with the orbit space of

RnC1 Xf0g under the R of nonzero real numbers given by scalar
action of the group
multiplication: x 1 ; : : : ; x nC1 D x 1 ; : : : ; x nC1 . It is easy to check that this
action is smooth and free. To see that it is proper, we use Proposition C.15. Suppose
.xi / is a sequence in RnC1 X f0g and .i / is a sequence in R such that xi ! x 2
RnC1 X f0g and i xi ! y 2 RnC1 X f0g. Then ji j D ji xi j=jxi j converges to the
nonzero real number jyj=jxj. Thus the numbers i all lie in a compact set of the
form f W 1=C jj C g for some positive number C , so a subsequence converges
to a nonzero real number. Therefore, by the quotient manifold theorem, RP n has
a unique structure as a smooth n-dimensional manifold such that the quotient map
RnC1 X f0g ! RP n is a smooth submersion. //
Example C.19 (Complex Projective Spaces). Similarly, the n-dimensional com-
plex projective space, denoted by CP n , is the set of 1-dimensional complex sub-
spaces of C nC1 , identified with the orbit space of C nC1 X f0g under the C -action
given by z D z. The same argument as in the preceding example shows that this
action is smooth, free, and proper, so CP n is a smooth 2n-dimensional manifold
and the quotient map W C nC1 X f0g ! CP n is a smooth submersion. //
Group Actions and Covering Spaces
There is a close connection between smooth covering maps and smooth group
and M are smooth manifolds and W M
actions. To begin, suppose M !M
is a smooth covering map. A covering automorphism of is a diffeomorphism

'W M !M such that ı ' D ; the set of all covering automorphisms is a
group under
composition, called the covering automorphism group and denoted
by Aut M .

Let W M ! M be a smooth covering
Proposition C.20. [LeeSM, Prop. 21.12]
is a discrete Lie group acting smoothly,
map. With the discrete topology, Aut M
.
freely, and properly on M
Because of the requirement that a covering automorphism ' satisfy ı ' D , it

follows that ' restricts to an action on each fiber of . The next proposition describes
the conditions in which this action is transitive. A covering map W M ! M is

called a normal covering if the image of the homomorphism W z !
1 M;x

1 M; .z x / is a normal subgroup of 1 M; .z x / . (Note that every universal cov-
ering is a normal covering, because the trivial subgroup is always normal.)
Proposition C.21. [LeeTM, Cor. 12.5] If W M ! M is a smooth covering map,

then the automorphism group of acts transitively on each fiber of if and only if
is a normal covering.
The universal covering is the most important special case.
Proposition C.22 (Automorphisms of the Universal Covering). [LeeTM, Cor.

12.9] Suppose M is a connected ! M is its univer-
smooth manifold and W M

sal covering. Then Aut M is isomorphic to the fundamental group of M , and it
acts transitively on each fiber of .
The next theorem, which is an application of the quotient manifold theorem, is a

partial converse to Proposition C.20.
Proposition C.23. [LeeSM, Thm. 21.13] Suppose M is a connected smooth mani-

. Then
fold and is a discrete Lie group acting smoothly, freely, and properly on M

the orbit space M = has a unique smooth manifold structure such that the quotient
map M !M = is a smooth normal covering map.
Example C.24 (The Universal Covering of RP n ). The two-element group D

f˙1g acts smoothly, freely, and properly on Sn by multiplication, and thus Sn = is
a smooth manifold and the quotient map W Sn ! Sn = is a two-sheeted smooth
normal covering map. Note also that the quotient map RnC1 X f0g ! RP n defined
in Example C.18 restricts to a surjective smooth map q W Sn ! RP n , which is a
submersion because each point of Sn is in the image of a smooth local section. Since
q and are constant on each other’s fibers, Proposition A.19 shows that there is a
diffeomorphism Sn = ! RP n , and q is equal to the composition Sn ! Sn = !
RP n . Since this is a smooth normal covering map followed by a diffeomorphism, q
is also a smooth normal covering map. Thus Sn is the universal covering space of
RP n , and the fundamental group of RP n is isomorphic to f˙1g. //
References
[AP94] Marco Abate and Giorgio Patrizio, Finsler Metrics, a Global Approach: with Appli-
cations to Geometric Function Theory, Springer, New York, 1994.
[Amb56] Warren Ambrose, Parallel translation of Riemannian curvature, Ann. of Math. (2) 64
(1956), 337–363.
[Ave70] André Avez, Variétés riemanniennes sans points focaux, C. R. Acad. Sci. Paris Sér.
A-B 270 (1970), A188–A191 (French).
[BCS00] David Bao, Shiing-Shen Chern, and Zhongmin Shen, An Introduction to Riemann–
Finsler Geometry, Graduate Texts in Mathematics, vol. 200, Springer, New York,
2000.
[Ber60] M. Berger, Les variétés Riemanniennes (1/4)-pincées, Ann. Scuola Norm. Sup. Pisa
(3) 14 (1960), 161–170 (French).
[Ber03] Marcel Berger, A Panoramic View of Riemannian Geometry, Springer-Verlag, Berlin,
2003.
[Bes87] Arthur L. Besse, Einstein Manifolds, Springer, Berlin, 1987.
[BBBMP] Laurent Bessières, Gérard Besson, Michel Boileau, Sylvain Maillot, and Joan Porti,
Geometrisation of 3-Manifolds, EMS Tracts in Mathematics, vol. 13, European Math-
ematical Society (EMS), Züurich, 2010.
[Bis63] Richard L. Bishop, A relation between volume, mean curvature, and diameter,
Notices Amer. Math. Soc. 10 (1963), 364.
[BC64] Richard L. Bishop and Richard J. Crittenden, Geometry of Manifolds, Pure and
Applied Mathematics, Vol. XV, Academic Press, New York-London, 1964.
[BW08] Christoph Böhm and Burkhard Wilking, Manifolds with positive curvature operators
are space forms, Ann. of Math. (2) 167 (2008), no. 3, 1079–1097.
[Boo86] William M. Boothby, An Introduction to Differentiable Manifolds and Riemannian
Geometry, 2nd ed., Academic Press, Orlando, FL, 1986.
[Bre10] Simon Brendle, Ricci Flow and the Sphere Theorem, Graduate Studies in Mathemat-
ics, vol. 111, American Mathematical Society, Providence, RI, 2010.
[BS09] Simon Brendle and Richard Schoen, Manifolds with 1/4-pinched curvature are space
forms, J. Amer. Math. Soc. 22 (2009), 287–307.
[Car26] Élie Cartan, Sur une classe remarquable d’espaces de Riemann, Bull. Soc. Math.
France 54 (1926), 214–264 (French).
[Cha06] Isaac Chavel, Riemannian Geometry, 2nd ed., Cambridge Studies in Advanced Math-
ematics, vol. 98, Cambridge University Press, Cambridge, 2006. A modern introduc-
tion.
[CE08] Jeff Cheeger and David G. Ebin, Comparison Theorems in Riemannian Geometry,
AMS Chelsea Publishing, Providence, 2008. Revised reprint of the 1975 original.
[CG71] Jeff Cheeger and Detlef Gromoll, The splitting theorem for manifolds of nonnegative
Ricci curvature, J. Differential Geometry 6 (1971), 119–128.

416 REFERENCES
[CG72] Jeff Cheeger and Detlef Gromoll. On the structure of complete manifolds of nonneg-
ative curvature, Ann. Math. (2) 96 (1972), 413–443.
[Che75] Shiu-Yuen Cheng, Eigenvalue comparison theorems and its geometric applications,
Math. Z. 143 (1975), no. 3, 289–297.
[Che55] Shiing-Shen Chern, An elementary proof of the existence of isothermal parameters on
a surface, Proc. Amer. Math. Soc. 6 (1955), 771–782.
[CB09] Yvonne Choquet-Bruhat, General Relativity and the Einstein Equations, Oxford
Mathematical Monographs, Oxford University Press, Oxford, 2009.
[Cla70] Chris J. S. Clarke, On the Global Isometric Embedding of Pseudo-Riemannian Mani-
folds, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 314 (1970), no. 1518, 417–428.
[CM11] Tobias H. Colding and William P. Minicozzi II, A Course in Minimal Surfaces, Grad-
uate Studies in Mathematics, vol. 121, American Mathematical Society, Providence,
RI, 2011.
[dC92] Manfredo Perdigão do Carmo, Riemannian Geometry, Mathematics: Theory & Appli-
cations, Birkhäuser Boston, Inc., Boston, MA, 1992. Translated from the second Por-
tuguese edition by Francis Flaherty.
[DF04] David S. Dummit and Richard M. Foote, Abstract Algebra, 3rd ed., John Wiley &
Sons, Inc., Hoboken, NJ, 2004.
[Fra61] Theodore Frankel, Manifolds with positive curvature, Pacific J. Math. 11 (1961), 165–
174.
[GHL04] Sylvestre Gallot, Dominique Hulin, and Jacques Lafontaine, Riemannian Geometry,
3rd ed., Universitext, Springer-Verlag, Berlin, 2004.
[Gan73] David Gans, An Introduction to Non-Euclidean Geometry, Academic Press, New
York, 1973.
[Gau65] Carl F. Gauss, General Investigations of Curved Surfaces, Raven Press, New York,
1965.
[Gra82] Alfred Gray, Comparison theorems for the volumes of tubes as generalizations of the
Weyl tube formula, Topology 21 (1982), no. 2, 201–228.
[Gra04] Alfred Gray, Tubes, 2nd ed., Progress in Mathematics, vol. 221, Birkhäuser Verlag,
Basel, 2004. With a preface by Vicente Miquel.
[Gre93] Marvin J. Greenberg, Euclidean and Non-Euclidean Geometries: Development and
History, W. H. Freeman, New York, 1993.
[Gre70] Robert E. Greene, Isometric Embeddings of Riemannian and Pseudo-Riemannian
Manifolds, Memoirs of the Amer. Math. Soc., No. 97, Amer. Math. Soc., Providence,
1970.
[Gro07] Misha Gromov, Metric Structures for Riemannian and Non-Riemannian Spaces,
Reprint of the 2001 English edition, Modern Birkhäuser Classics, Birkhäuser Boston,
Inc., Boston, MA, 2007. Based on the 1981 French original; With appendices by M.
Katz, P. Pansu and S. Semmes; Translated from the French by Sean Michael Bates.
[Gün60] Paul Günther, Einige Sätze über das Volumenelement eines Riemannschen Raumes,
Publ. Math. Debrecen 7 (1960), 78–93 (German).
[Ham82] Richard S. Hamilton, Three-manifolds with positive Ricci curvature, J. Differential
Geom. 17 (1982), 255–306.
[Ham86] Richard S. Hamilton, Four-manifolds with positive curvature operator, J. Differential
Geom. 24 (1986), no. 2, 153–179.
[Hat02] Allen Hatcher, Algebraic Topology, Cambridge University Press, Cambridge, 2002.
[HE73] Stephen W. Hawking and George F. R. Ellis, The Large-Scale Structure of Space-
Time, Cambridge University Press, Cambridge, 1973.
[Hel01] Sigurdur Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces, Grad-
uate Studies in Mathematics, vol. 34, American Mathematical Society, Providence,
RI, 2001. Corrected reprint of the 1978 original.
[Her63] Robert Hermann, Homogeneous Riemannian manifolds of non-positive sectional cur-
vature, Nederl. Akad. Wetensch. Proc. Ser. A 66 D Indag. Math. 25 (1963), 47–56.
[Hic59] Noel J. Hicks, A theorem on affine connexions, Illinois J. Math. 3 (1959), 242–254.
REFERENCES 417
[Hil71] David Hilbert, The Foundations of Geometry, Open Court Publishing Co., Chicago,
1971. Translated from the tenth German edition by Leo Unger.
[Hop89] Heinz Hopf, Differential Geometry in the Large, 2nd ed., Lecture Notes in Mathe-
matics, vol. 1000, Springer, 1989. Notes taken by Peter Lax and John W. Gray; with
a preface by S. S. Chern; with a preface by K. Voss.
[Ive92] Birger Iversen, Hyperbolic Geometry, London Mathematical Society Student Texts,
vol. 25, Cambridge University Press, Cambridge, 1992.
[Jos17] Jürgen Jost, Riemannian Geometry and Geometric Analysis, 7th ed., Universitext,
Springer, Cham, 2017.
[JP13] Marek Jarnicki and Peter Pflug, Invariant Distances and Metrics in Complex Analysis,
Second extended edition, De Gruyter Expositions in Mathematics, vol. 9, Walter de
Gruyter GmbH & Co. KG, Berlin, 2013.
[Kaz85] Jerry Kazdan, Prescribing the Curvature of a Riemannian Manifold, CBMS Regional
Conf. Ser. in Math., Amer. Math. Soc., Providence, 1985.
[KL08] Bruce Kleiner and John Lott, Notes on Perelman’s papers, Geom. Topol. 12 (2008),
no. 5, 2587–2855.
[Kli61] Wilhelm P. A. Klingenberg, Über Riemannsche Mannigfaltigkeiten mit positiver
Krümmung, Comment. Math. Helv. 35 (1961), 47–54 (German).
[Kli95] Wilhelm P. A. Klingenberg, Riemannian Geometry, 2nd ed., De Gruyter Studies in
Mathematics, vol. 1, Walter de Gruyter & Co., Berlin, 1995.
[Kob72] Shoshichi Kobayashi, Transformation Groups in Differential Geometry, Springer,
Berlin, 1972.
[KN96] Shoshichi Kobayashi and Katsumi Nomizu, Foundations of Differential Geometry,
Wiley Classics Library, vol. I–II, John Wiley & Sons, Inc., New York, 1996. Reprint
of the 1963 and 1969 originals; A Wiley-Interscience Publication.
[LeeJeff09] Jeffrey M. Lee, Manifolds and Differential Geometry, Graduate Studies in Mathemat-
ics, vol. 107, American Mathematical Society, Providence, RI, 2009.
[LeeAG] John M. Lee, Axiomatic Geometry, Pure and Applied Undergraduate Texts, vol. 21,
American Mathematical Society, Providence, RI, 2013.
[LeeSM] John M. Lee, Introduction to Smooth Manifolds, 2nd ed., Graduate Texts in Mathe-
matics, vol. 218, Springer, New York, 2013.
[LeeTM] John M. Lee, Introduction to Topological Manifolds, 2nd ed., Graduate Texts in Math-
ematics, vol. 202, Springer, New York, 2011.
[LP87] John M. Lee and Thomas H. Parker, The Yamabe problem, Bull. Amer. Math. Soc.
(N.S.) 17 (1987), 37–91.
[MP11] William H. Meeks III and Joaqun Pérez, The classical theory of minimal surfaces,
Bull. Amer. Math. Soc. (N.S.) 48 (2011), 325–407.
[Mil63] John W. Milnor, Morse Theory, based on lecture notes by M. Spivak and R. Wells.
Annals of Mathematics Studies, No. 51, Princeton University Press, Princeton, N.J.,
1963.
[Mil68] John W. Milnor, A note on curvature and fundamental group, J. Differential Geometry
2 (1968), 1–7.
[Mil76] John W. Milnor, Curvatures of left invariant metrics on Lie groups, Advances in Math.
21 (1976), no. 3, 293–329.
[Mor98] Frank Morgan, Riemannian Geometry: A Beginner’s Guide, 2nd ed., A K Peters Ltd.,
Wellesley, MA, 1998.
[MF10] John W. Morgan and Frederick Tsz-Ho Fong, Ricci Flow and Geometrization of 3-
Manifolds, University Lecture Series, vol. 53, American Mathematical Society, Prov-
idence, RI, 2010.
[MT14] John W. Morgan and Gang Tian, The Geometrization Conjecture, Clay Mathematics
Monographs, vol. 5, American Mathematical Society, Providence, RI; Clay Mathe-
matics Institute, Cambridge, MA, 2014.
[Mun56] James Raymond Munkres, Some Applications of Triangulation Theorems, ProQuest
LLC, Ann Arbor, MI, 1956. Thesis (Ph.D.)-University of Michigan.
418 REFERENCES
[Mye41] Sumner B. Myers, Riemannian manifolds with positive mean curvature, Duke Math.
J. 8 (1941), 401–404.
[MS39] Sumner B. Myers and Norman E. Steenrod, The group of isometries of a Riemannian
manifold, Ann. of Math. (2) 40 (1939), no. 2, 400–416.
[Nas56] John Nash, The imbedding problem for Riemannian manifolds, Ann. Math. 63 (1956),
20–63.
[O’N83] Barrett O’Neill, Semi-Riemannian Geometry with Applications to General Relativity,
Academic Press, New York, 1983.
[Pet16] Peter Petersen, Riemannian Geometry, 3rd ed., Graduate Texts in Mathematics, vol.
171, Springer, Cham, 2016.
[Poo81] Walter A. Poor, Differential Geometric Structures, McGraw-Hill, 1981.
[Pre43] Alexandre Preissman, Quelques propriétés globales des espaces de Riemann, Com-
ment. Math. Helv. 15 (1943), 175–216 (French).
[Rad25] Tibor Radó, Über den Begriff der Riemannschen Fläche, Acta Litt. Sci. Szeged. 2
(1925), 101–121 (German).
[Rat06] John G. Ratcliffe, Foundations of Hyperbolic Manifolds, 2nd ed., Graduate Texts in
Mathematics, vol. 149, Springer, New York, 2006.
[Sco83] Peter Scott, The Geometries of 3-Manifolds, Bull. London Math. Soc. 15 (1983), 401–
487.
[Spi79] Michael Spivak, A Comprehensive Introduction to Differential Geometry, Vol. I–V,
Publish or Perish, Berkeley, 1979.
[Str86] Robert S. Strichartz, Sub-Riemannian geometry, J. Differential Geom. 24 (1986), no.
2, 221–263.
[Syn36] John L. Synge, On the connectivity of spaces of positive curvature, Q. J. Math. os-7
(1936), 316–320.
[Thu97] William P. Thurston, Three-Dimensional Geometry and Topology. Vol. 1, Princeton
Mathematical Series, vol. 35, Princeton University Press, Princeton, NJ, 1997. Edited
by Silvio Levy.
[Tu11] Loring W. Tu, An Introduction to Manifolds, 2nd ed., Universitext, Springer, New
York, 2011.
[Wol11] Joseph A. Wolf, Spaces of Constant Curvature, 6th ed., AMS Chelsea Publishing,
Providence, RI, 2011.
Notation Index
Symbols C
^ (wedge product), 400 C (Cotton tensor), 219
(Kulkarni–Nomizu product), 213 C 1 (infinitely differentiable), 374
q (disjoint union), 382 C 1 .M / (smooth real-valued functions),
³ (interior multiplication), 31, 401 376
] (sharp), 26–28 C 1 .M; N / (smooth maps), 376
(Hodge star operator), 50 CP n (complex projective space), 413

j j (length of a vector), 12 conv.x/ (convexity radius at x), 186

j jg (length of a vector), 12 Cut.p/ (cut locus of p), 308

Œ (path class), 373

Œ ; (commutator bracket), 409 D

Œ ; (Lie bracket), 385 ıji (Kronecker delta), 392

. ; / (global inner product on forms), 53 (Laplacian), 32

h ; i (inner product), 9, 12 @=@x i (coordinate vector field), 384

h ; i (pairing between vector and covector), @=@x i jp (coordinate vector), 377
392 @i (coordinate vector field), 384

h ; i (scalar product), 40 @i jp (coordinate vector), 377

h ; ig (inner product), 12 @r (radial vector field), 158, 179
@s (velocity of a transverse curve), 152,
197
A
@t (velocity of a main curve), 152, 197
A (adjoint of a matrix), 408
r (covariant derivative), 89
AT (transpose of a matrix), 408
r ? (normal connection), 231
A.TM / (set of connections on TM ), 94
r > (tangential connection), 93, 228
ad (adjoint representation of a Lie algebra),
r 2 u (covariant Hessian), 100
411 2
rX;Y (second total covariant derivative),
Ad (adjoint representation of a Lie group),
99
411
rF (total covariant derivative), 97
Alt (alternating
projection), 400
(covering automorphism group), rX Y (covariant derivative), 89
Aut M d (exterior derivative), 401
414 D (exterior covariant derivative), 209, 236
d (adjoint of d ), 52
B dFp (differential of a map), 377
[ (flat), 26–28 dg .p; q/ (Riemannian distance), 36
Bc .p/ (geodesic ball), 163 dg .p; S/ (distance to a subset), 174
Bx c .p/ (closed geodesic ball), 163 DM (density bundle), 406
Bn .R/ (Poincaré ball), 62 D .p/ (domain of .p/ ), 387
420 Notation Index
Ds (covariant derivative along transverse gl.V / (Lie algebra of linear maps), 409
curves), 153 ŒG1 ; G2 (group generated by commuta-
Dt (covariant derivative along a curve), 101 tors), 72
d Vg (Riemannian density), 31 g1 ˚ f 2 g2 (warped product metric), 20
d Vg (Riemannian volume form), 30 g1 ˚ g2 (product metric), 20
div X (divergence of X ), 32 GL.V / (group of invertible linear maps),
408
E grad f (gradient of f ), 27
e (identity of a Lie group), 407
E (normal exponential map), 133 H
E (domain of the exponential map), 128 h (scalar second fundamental form), 235
E jM (restriction of a bundle), 384 H (mean curvature), 238
E.n/ (Euclidean group), 57 Hn (hyperbolic space), 66
EP (domain of the normal exponential Hn .R/ (hyperbolic space of radius R), 62,
map), 133 66
Ep (domain of the restricted exponential Hn (Heisenberg group), 72
map), 128 HV r (traceless Hessian operator), 337
End.V / (space of endomorphisms), 394 Hr (Hessian operator), 320
exp (exponential map), 128 Hr;s .R/ (pseudohyperbolic space), 79
expG (exponential map of a Lie group), 410 Hol.p/ (holonomy group), 150
expp (restricted exponential map), 128 Hol0 .p/ (restricted holonomy group), 150
F I
'rz (pullback of a connection), 110 iv (interior multiplication), 401
˚ 1 .fyg/ (level set), 380 In (identity matrix), 408
˚ 1 .y/ (level set), 380 I.V; W / (index form), 301
F (induced fundamental group homomor- ID.p/ (injectivity domain), 310
phism), 373 II (second fundamental form), 227
F (induced Lie algebra homomorphism), IIN (second fundamental form in direction
409 N ), 229
F (pushforward of vector fields), 385 inj.M / (injectivity radius of M ), 166
F (pullback of a density), 406 inj.p/ (injectivity radius at p), 165
F (pullback of a tensor field), 398 Iso.M; g/ (isometry group), 13
F G (symmetric product of F and G), 396 Isop .M; g/ (isotropy group of a point), 56
G J
0 (velocity vector), 386 J./ (space of Jacobi fields), 285
J ? ./ (normal Jacobi fields), 287
0 ai˙ (one-sided velocity vectors), 34
J > ./ (tangential Jacobi fields), 287
v (geodesic with initial velocity v), 105
.E / (smooth sections of E ), 89, 383 K
k
ij (connection coefficients), 91, 123
(Cayley transform), 66
.s; t/ (family of curves), 152
(geodesic curvature), 232
gy (pullback of the round metric), 293
i (principal curvatures), 4
x (Euclidean metric), 12
g
N (signed curvature), 273
gV (round metric), 16, 58 K (Gaussian curvature), 238
gV R (round metric of radius R), 58 Kn .R/ (Beltrami–Klein model), 62
gM (hyperbolic metric), 66
gM R (hyperbolic metric of radius R), 62, 66 L
gij (metric coefficients), 13 ƒk T M (bundle of alternating tensors),
g ij (inverse of gij ), 26 401
GL.n; C/ (complex general linear group), ƒk .V / (space of alternating tensors), 400
408 Lg ./ (length of curve), 34
GL.n; R/ (general linear group), 408 Lie.G/ (Lie algebra of left-invariant vector
gl.n; R/ (matrix Lie algebra), 409 fields on G), 408
Notation Index 421
M SC (surface of revolution), 20
jj (density associated with an n-form), Sc .p/ (geodesic sphere), 163
406 Sn (unit n-sphere), 16
M (M with opposite orientation), 405 Sn .R/ (n-sphere of radius R), 58
M=G (orbit space), 413 S ? (orthogonal complement), 10, 40
M.n k; R/ (space of matrices), 48 Sr;s .R/ (pseudosphere), 79
M.n; R/ (space of n n matrices), 411 SL.n; C/ (complex special linear group),
408
N SL.n; R/ (special linear group), 408
Np M (normal space), 16 SO.n/ (special orthogonal group), 408
NM (normal bundle), 16 SU.n/ (special unitary group), 408
sec.˘ / (sectional curvature), 8, 250
O sec.v; w/ (sectional curvature), 250
!i j (connection 1-forms), 113 sgn (sign of a permutation), 400
k .M / (space of k-forms), 401 Sol (3-dimensional solvable Lie group), 72
O.M / (orthonormal bases on M ), 56 supp f (support of f ), 376
O.n/ (orthogonal group), 408 Sym (symmetrization), 395
O.n; 1/ (Lorentz group), 67
OC .n; 1/ (orthochronous Lorentz group), T
67 (torsion tensor), 112, 121
O.n C 1/ (orthogonal group), 58 T n (n-torus), 389
TM (tangent bundle), 382
P TM jM (ambient tangent bundle), 386
1 .M; p/ (fundamental group), 373 Tp M (tangent space), 376
? (normal projection), 226 >
T.t / M (space of vectors tangent to a
> (tangential projection), 226 curve), 287
P (Schouten tensor), 215 ?
T.t / M (space of vectors normal to a
Pt0 t1 (parallel transport operator), 108 curve), 287
T k .M / (space of tensor fields), 397
Q T k TM (bundle of contravariant k-
x.r;s/ (pseudo-Euclidean metric), 43, 79
q tensors), 396
.r;s/
qM R (pseudohyperbolic metric), 79 T k T M (bundle of covariant k-tensors),
.r;s/
qV R (pseudospherical metric), 79 396
T k .V / (space of contravariant k-tensors),
R 393
./ (rotation index), 264 T k .V / (space of covariant k-tensors),
r (radial distance function), 158, 179 393
R (curvature endomorphism), 196 T .k;l/ TM (bundle of mixed tensors), 396
Rn (Euclidean space), 12, 57 T .k;l/ .V / (space of mixed tensors), 393
Rr;s (pseudo-Euclidean space), 43, 79 TCL.p/ (tangent cut locus), 310
RP n (real projective space), 413 tr (trace of a tensor), 395
R.V / (space of algebraic curvature ten- trg (trace with respect to g), 28
sors), 212
R.X; Y / (curvature endomorphism), 196 U
R.X; Y / (curvature endomorphism), 205 U.n/ (unitary group), 408
Rc (Ricci tensor), 207 U n (Poincaré half-space), 62
Rm (curvature tensor), 198
V
S V (dual space of V ), 391
†k T M (bundle of symmetric tensors), jv ^ wj (norm of an alternating 2-tensor),
396 250
†k .V / (space of symmetric tensors), 395 Vk .Rn / (Stiefel manifold), 48
s (shape operator), 235 Vol.M / (volume of a Riemannian mani-
S (scalar curvature), 208 fold), 30
422 Notation Index
W X ? (normal projection of X ), 227

W (Weyl tensor), 215 X? ./ (normal vector fields along a curve),
WN (Weingarten map in direction N ), 229 287
X > (tangential projection of X ), 227
X X> ./ (tangential vector fields along a
Ì (semidirect product), 412 curve), 287
.M / (Euler characteristic), 276, 280 X./ (space of vector fields along a curve),
X.M / (space of vector fields), 385 100
Subject Index
Symbols analytic continuation of a local isometry,

.1; 1/-Hessian, 320 346
1-center, 355 angle
2-point homogeneous, 189, 261 between vectors, 10, 12
3-point homogeneous, 261 of a geodesic triangle, 353
tangent, 264, 265
A angle excess, 281
abelian Lie algebra, 409 angle-sum theorem, 2, 271, 275
absolute derivative, 102 anti-de Sitter space, 80
acceleration universal, 83
in Rn , 86 anti-self-dual, 50
of a curve in a manifold, 103 anticommutative, 400
of a plane curve, 2 antipodal points, 144
tangential, 86 arc length, parametrization by, 35
action, see group action arc-length function, 35
adapted frame, 16 area-minimizing hypersurface, 239
adjoint matrix, 408 area of a hypersurface, 239
adjoint representation aspherical, 354
of a Lie algebra, 411 asymptotically parallel, 282
of a Lie group, 68, 411 atlas, 374
admissible curve, 34 automorphism of a Lie group, 407
admissible family, 152 inner, 411
admissible loop, 150 Avez, André, 364
admissible partition axial isometry, 357
for a curve, 34 axis for an isometry, 357
for a family of curves, 152
for a vector field along a curve, 108 B
affine connection, 91 Böhm, Christoph, 368
aims at a point, 167 Böhm–Wilking theorem, 368
algebraic Bianchi identity, 203 backward reparametrization, 34
algebraic curvature tensor, 212 ball
Alt convention for wedge product, 401 geodesic, 158, 163
alternating tensor, 400 metric, 163
alternation, 400 Poincaré, 62
ambient manifold, 226 regular coordinate, 374
ambient tangent bundle, 386 smooth coordinate, 374
Ambrose, Warren, 351 volume of, 314
424 Subject Index
base of a vector bundle, 382 catenoid, 257

basis isomorphism, 375 Cauchy–Riemann manifold, 46
Beltrami–Klein model, 62 Cayley transform, 66
Berger sphere, 81, 259 central projection, 63
Berger, Marcel, 81, 367 chart, 371
bi-invariant metric, 67, 147 containing a point, 371
curvature of, 224, 259 smooth, 374
existence of, 70 Cheeger, Jeff, 367
exponential map of, 147 Cheng’s maximal diameter theorem, 362
Bianchi identity Cheng, Shiu-Yuen, 362
algebraic, 203 Chern, Shiing-Shen, 280
contracted, 209 Chern–Gauss–Bonnet theorem, 280
differential, 204 Choquet-Bruhat, Yvonne, 260
first, 203 Christoffel symbols, 123, 124
second, 204 circle classification theorem, 2
Bieberbach, Ludwig, 350 circle group, 408
bilinear form, 40 circumference theorem, 2, 271, 275
Bishop, Richard L., 340 Clarke, Chris J. S., 117
Bishop–Gromov theorem, 340 classification theorem, 2
Bochner’s formula, 223 circle, 2
Bonnet, Ossian, 361 compact surfaces, 74
boundary chart, 371
constant-curvature metrics, 8, 349, 350
boundary normal coordinates, 183
plane curve, 3, 282
boundary of a manifold with boundary, 371
closed form, 402
boundary problem, two-point, 299
closed geodesic, 173, 174, 317, 343
boundary slice coordinates, 380
closed geodesic ball, 158
bounded subset, 39
closed map, 372
bracket
closed subgroup theorem, 410
commutator, 409
Codazzi equation, 232
in a Lie algebra, 408
Brendle, Simon, 368 for a hypersurface, 236, 259
bump function, 376 Codazzi tensor, 244
bundle codimension, 379
normal, 16 coframe, 397
of tensors, 396 commutator bracket, 409
vector, 382 commuting vector fields, 388
canonical form, 388
C comparison theorem
CR manifold, 46 Bishop–Gromov, 340
Calabi–Yau manifold, 337 conjugate point, 333, 339
calculus of variations, 152 diameter, 339
canonical form Günther, 334
for a nonvanishing vector field, 387 Hessian, 327
for commuting vector fields, 388 injectivity radius, 339
Carathéodory metric, 47 Jacobi field, 329
Carnot–Carathéodory metric, 46 Laplacian, 332, 337
Cartan’s first structure equation, 113 metric, 331
Cartan’s fixed-point theorem, 356 principal curvature, 329
Cartan’s second structure equation, 222 Riccati, 324–327
Cartan’s torsion theorem, 356 volume, 334, 340
Cartan, Élie, 113, 222, 295, 351, 352 compatibility with a metric, 118
Cartan–Ambrose–Hicks theorem, 351 complete Riemannian manifold, 170
Cartan–Hadamard manifold, 352 geodesically, 131, 166
Cartan–Hadamard theorem, 8, 352 metrically, 39, 166
Subject Index 425
complex projective space, 24, 82, 258, 367, contracted Bianchi identity, 209
413 contraction, 395
component functions, 383 contravariant tensor, 392
composable paths, 373 control theory, 46
conformal diffeomorphism, 59 convex hypersurface, 259
conformal invariance convex subset, 281
of the Cotton tensor, 219 convex, geodesically, 166, 281
of the Weyl tensor, 217 convexity radius, 186
conformal Laplacian, 223 coordinate representation, 376
conformal metrics, 59 coordinate ball
conformal transformation, 59 regular, 374
of the curvature, 217 smooth, 374
of the Levi-Civita connection, 217 coordinate chart, 371
conformally equivalent metrics, 59 smooth, 374
conformally flat, locally, 59, 218 coordinate domain, 374
hyperbolic space, 66 coordinate frame, 385
spheres, 61 coordinate vector, 377
conformally related metrics, 59 coordinate vector field, 384
congruent, 2, 142 coordinates, 375
conjugate point, 298 boundary slice, 380
critical point of expp , 299 Fermi, 136
geodesic not minimizing past, 303 graph, 18
conjugate point comparison theorem, 333, have upper indices, 375
339 local, 375
connection, 89 natural, on tangent bundle, 384
Euclidean, 92 normal, 132
existence of, 93 polar Fermi, 184
flat, 224 polar normal, 184
in a vector bundle, 89 semigeodesic, 181, 182
in components, 91 slice, 379, 380
in the tangent bundle, 91 standard, on Rn , 12
Koszul, 89 cosmological constant, 211
naturality of, 125 cotangent bundle, 396
normal, 231 cotangent space, 377
on a manifold, 91 Cotton tensor, 219
on tensor bundles, 95–97 conformal invariance of, 219
Riemannian, 122, 123 covariant derivative, 89
tangential, 92–93 along a curve, 101–103
connection 1-forms, 113, 274, 275 exterior, 209, 236
connection coefficients, 91 of tensor field, 95–97
transformation law for, 92 second, 99
constant Gaussian curvature, 6 total, 98
constant mean curvature, 242 covariant Hessian, 100, 112, 320
constant rank, 378 covariant tensor, 392
constant sectional curvature, 254 covector, 377, 391
characterization of, 8 covector field, 397
classification, 349, 350 covering automorphism, 414
formula for curvature tensor, 255 covering automorphism group, 414
formula for metric, 293 covering map, 171, 388
local uniqueness, 294 homeomorphism criterion, 390
model spaces, 8 normal, 414
uniqueness, 348 Riemannian, 24, 25
constant-speed curve, 35 smooth, 388
constraint equations, Einstein, 260 universal, 390
426 Subject Index
critical point, 27, 157, 210, 380 delta, Kronecker, 10, 392
of the exponential map, 299 density, 405–406
critical value, 380 on a manifold, 406
crystallographic groups, 350 on a vector space, 405
curl, 52 pullback of, 406
curvature, 2–8, 196 Riemannian, 31
conformal transformation of, 217 density bundle, 406
constant sectional, 8, 254, 255, 294, 348, derivation, 376
349 of C 1 .M /, 385
Gaussian, 5–6, 238 determinant convention for wedge product,
geodesic, 232 401
mean, 238 diameter, 39
of a curve in a manifold, 232, 233 diameter comparison theorem, 339
of a plane curve, 2 diffeomorphism, 375
principal, 4, 237 in Rn , 374
Ricci, 207 local, 378
Riemann, 196, 198 difference tensor, 94
scalar, 208 differentiable sphere theorem, 368
sectional, 8, 250, 251 differential
signed, 3, 273 global, 385
curvature 2-forms, 222 of a function, 377, 402
curvature endomorphism, 196, 224 of a map, 377
curvature operator, 262 differential Bianchi identity, 204
curvature tensor, 196, 198 differential form, 401
determined by sectional curvatures, 252 closed, 402
is isometry invariant, 199 exact, 402
symmetries of, 202 dihedral groups, 350
curve, 33, 386 direct product of Lie groups, 408
admissible, 34 directional derivative of a vector field on
in a manifold, 33 Rn , 86
piecewise regular, 33 Dirichlet eigenvalue, 51
plane, 2 Dirichlet’s principle, 52
smooth, 33 discrete Lie group, 408
curve segment, 33 discrete subgroup, 410
curved polygon, 265, 271 disjoint union, 382
curved triangle, 276 distance function, 174
cusp vertex, 265 has geodesic integral curves, 176
cut locus, 308 has unit gradient, 175
and injectivity radius, 312 local, 176
tangent, 310 distance, Riemannian, 36
cut point, 308 on a disconnected manifold, 36, 53
cut time, 307, 308 diverge to infinity, 187
is continuous, 309 divergence, 32
cylinder, principal curvatures, 4 in coordinates, 32
in terms of covariant derivatives, 148
D of a tensor field, 149
ı-pinched, 366 divergence theorem, 32, 50
pointwise, 368 domain of the exponential map, 128
strictly, 367 dot product, 9
Darboux theorem, 193 double elliptic geometry, 144
de Sitter space, 80 double of a manifold, 381
defining function, 380 dual basis, 392
local, 248, 380 dual coframe, 397
del, 89 dual space, 391
Subject Index 427
E extension
"-tubular neighborhood, 133 of functions, 380
edge of a curved polygon, 265 of sections of vector bundles, 384
effective group action, 412 exterior angle, 265, 271
eigenfunction of the Laplacian, 51 exterior covariant derivative, 209, 236
eigenvalue of the Laplacian, 51 exterior derivative, 401–402
Dirichlet, 51 naturality of, 402
Neumann, 51 extrinsic curvature of a curve, 233
Einstein constraint equations, 260
Einstein equation, 210 F
Einstein field equation, 211, 259 F -related, 385
Einstein metric, 210, 361 faithful representation, 408
2-dimensional, 216 family of curves, 152
Einstein summation convention, 375 admissible, 152
Einstein, Albert, 43, 211 Fermi coordinates, 136, 178
Einstein–Hilbert action, 211 polar, 184
embedded submanifold, 379 Fermi, Enrico, 136
embedding, 378 fiber
isometric, 15 of a map, 380
endomorphism of a vector bundle, 382
curvature, 196 fiber metric, 29
equivalent to a .1; 1/ tensor, 394 on a tensor bundle, 29
of a vector space, 394 on differential forms, 49
energy functional, 189 figure eight, 100, 101, 112
equivariant map, 412 Finsler metric, 47
equivariant rank theorem, 412 first Bianchi identity, 203
escape lemma, 387 first fundamental form, 227
Euclidean connection, 92 first structure equation, 113
Euclidean geodesics, 137 first variation, 154
Euclidean group, 57 flat ([), 26–28
Euclidean metric, 12, 57 flat connection, 224
Euclidean space, 57 flat metric, 12, 195, 200
sectional curvature of, 254 flat vertex, 265
Euclidean space form, 349 flatness criterion, 195
Euclidean triangle, 2 flow, 387
Euclidean, locally, 371 fundamental theorem on, 387
Euler characteristic, 276, 280 flow domain, 387
Euler–Lagrange equation, 157 focal point, 317
evenly covered neighborhood, 388 form
exact form, 402 bilinear, 40
existence and uniqueness closed, 402
for linear ODEs, 106 exact, 402
of geodesics, 103 forward reparametrization, 34
of Jacobi fields, 285 frame
exponential map coordinate, 385
differential of, 128 global, 383
domain of, 128 local, 383
naturality of, 130 frame-homogeneous, 56, 79
normal, 133 locally, 74
of a bi-invariant metric, 147 Frankel, Theodore, 315
of a Lie group, 147, 410 free group action, 412
of a Riemannian manifold, 128 free homotopy class, 174
extendible Riemannian manifold, 189 trivial, 174
extendible vector fields, 100 freely homotopic, 174
428 Subject Index
Fubini–Study metric, 24, 82, 367 geodesic equation, 103

curvature of, 258 geodesic loop, 188
functional, 151 geodesic polygon, 75, 271
length, 151 geodesic segment, 104
linear, 391 in path-homotopy class, 173
fundamental form geodesic sphere, 158, 163
first, 227 geodesic triangle, 353
second, 227 geodesic vector field, 129
fundamental group, 373 geodesically complete, 131, 166
homotopy invariance, 373 equivalent to metrically complete, 169
fundamental theorem geodesically convex, 166, 281
of hypersurface theory, 245 geometrization conjecture, 77, 368
of Riemannian geometry, 122 global differential, 385
on flows, 387 global frame, 383
gradient, 27, 45
G Gram–Schmidt algorithm, 10, 16, 42, 251
Gauss equation, 230 graph coordinates, 18
for a hypersurface, 236, 259 graph of a smooth function, 18
Gauss formula, 228 graph parametrization, 18
along a curve, 229 Grassmann manifold, 48, 82
for a curve in a hypersurface, 236, 259 Grassmannian, 48, 82
for a hypersurface, 236, 259 Gray, Alfred, 136
Gauss lemma, 159 great circle, 137
for submanifolds, 179 great hyperbola, 138
Gauss map, 257 Green’s identities, 51
Gauss’s Theorema Egregium, 5, 249 Greene, Robert, 117
Gauss, Carl Friedrich, 5 Gromoll, Detlef, 367
Gauss–Bonnet formula, 273 Gromov, Misha, 340
Gauss–Bonnet theorem, 6, 277 group action, 411
Gaussian curvature, 5, 238 by diffeomorphisms, 412
constant, 6 by isometries, 23
is isometry invariant, 249 effective, 412
of an abstract 2-manifold, 250 free, 412
general linear group, 408 isometric, 23
general relativity, 43, 211 left, 411
generating curve, 18 proper, 412
genus, 278 right, 411
geodesic, 7, 103 smooth, 412
closed, 173, 174, 317, 343 transitive, 412
existence and uniqueness of, 103 Günther, Paul, 334, 340
is locally minimizing, 165 Günther’s volume comparison theorem, 334
maximal, 104
naturality of, 125 H
on Euclidean space, 105, 137 Hadamard, Jacques, 352
on hyperbolic spaces, 138 half-cylinder, principal curvatures, 5
on spheres, 137 half-space, Poincaré, 62
pseudo-Riemannian, 124 Hamilton’s 3-manifold theorem, 367
radial, 133, 161 Hamilton’s 4-manifold theorem, 367
Riemannian, 124 Hamilton, Richard, 367
with a conjugate point, 303 harmonic function, 51
with respect to a connection, 103 Heisenberg group, 72
geodesic ball, 158, 163 Hermann, Robert, 369
closed, 158 Hessian
geodesic curvature, 232 covariant, 100, 112, 320
Subject Index 429
of length functional, 302 induced metric, 15

Hessian comparison theorem, 327 on a submanifold, 15
Hessian operator, 320 induced orientation, 404
Hicks, Noel, 351 infinitesimal generator, 387
Hilbert, David, 142 injectivity domain, 310
Hodge star operator, 50, 52, 53 injectivity radius, 165, 188, 312
holonomy, 150 and cut locus, 312
restricted, 150 continuity of, 312
homogeneous injectivity radius comparison theorem, 339
2-point, 189, 261 injectivity theorem for covering maps, 389
3-point, 261 inner automorphism, 411
k-point, 189 inner product, 9
locally, 74 of tensors, 29
homogeneous Riemannian manifold, 55 inner product space, 10
homogeneous space, 72 integral
homomorphism of a function, 30
Lie algebra, 409 with respect to arc length, 53
Lie group, 407 integral curve, 386
homotopic maps, 372 integral of an n-form, 404
homotopic, freely, 174 integration by parts, 51, 149
homotopy, 372 interior angle, 2, 265, 272
homotopy equivalence, 373 interior chart, 371
Hopf, Heinz, 266, 361 interior multiplication, 31, 401
Hopf–Rinow theorem, 169 interior of a curved polygon, 265, 271
horizontal index position, 394 interior of a manifold with boundary, 371
horizontal lift, 21 interpretation of an axiomatic system, 143
horizontal tangent space, 21 intrinsic curvature of a curve, 233
horizontal vector field, 21 intrinsic property, 4
hyperbolic metric, 62–67 invariant inner product, 69
hyperbolic plane, 6, 66 invariant metric, 23
hyperbolic space, 62–67 invariant under a flow, 388, 399
distance function of, 185 invariant, local, 194
locally conformally flat, 66 inverse function theorem, 378
sectional curvature of, 254 inward-pointing normal, 273
hyperbolic space form, 349 irreducible symmetric space, 79
hyperbolic stereographic projection, 65 isometric embedding, 15
hyperboloid model, 62 isometric group action, 23
hypersurface, 234, 379 isometric immersion, 15
isometric manifolds, 12
I isometries
ideal triangle, 282 of Euclidean space, 57, 147
immersed submanifold, 379 of hyperbolic spaces, 67, 147
with boundary, 379 of spheres, 58–59, 147
immersion, 378 isometry, 4, 12
isometric, 15 axial, 357
index linear, 11
of a geodesic segment, 307 local, 171
of a scalar product, 42 metric, 186
upper and lower, 375 of a Riemannian manifold, 13
index form, 301, 302 Riemannian, 186
index positions, 375, 394 isometry group, 13
induced homomorphism of Euclidean space, 57, 147
of fundamental groups, 373 of hyperbolic spaces, 67, 147
of Lie algebras, 409 of spheres, 58–59, 147
430 Subject Index
isomorphism of Lie groups, 407 of a curve, 34

isothermal coordinates, 222 of a group element, 370
isotropic, 56 of a vector, 10, 12
at a point, 56 parameter independence of, 34
isotropy representation, 56 length functional, 151
isotropy subgroup, 56, 412 lens space, 350
level set, 380
J Levi-Civita connection, 122
Jacobi equation, 284 naturality of, 125
Jacobi field, 284 Levi-Civita, Tullio, 115
existence and uniqueness, 285 Lichnerowicz’s theorem, 223
normal, 287, 288 Lie algebra, 408
on constant-curvature manifolds, 291 abelian, 409
tangential, 287 of a Lie group, 409
transverse, 315, 317 Lie algebra automorphism, 409
Jacobi field comparison theorem, 329 Lie algebra homomorphism, 409
Jacobi identity, 386, 408 induced, 409
Jacobian matrix, 377 Lie algebra isomorphism, 409
jumps in tangent angle, 264 Lie bracket, 385
naturality of, 386
K of vectors tangent to a submanifold, 386
k-point homogeneous, 189, 261 Lie derivative, 112
Kazdan, Jerry, 279 and Lie bracket, 388
Killing vector field, 150, 190, 315 of a tensor field, 399
Killing–Hopf theorem, 348 of a vector field, 387
Klein bottle, 76 Lie group, 407
Klingenberg, Wilhelm, 367 direct product of, 408
Kobayashi metric, 47 discrete, 408
Koszul connection, 89 semidirect product of, 412
Koszul’s formula, 123 Lie group homomorphism, 407
Kronecker delta, 10, 392 Lie group isomorphism, 407
Kulkarni–Nomizu product, 213, 235 Lie subalgebra, 409
Lie subgroup, 407
L lift of a map, 389
Lagrange multiplier, 381 from simply connected spaces, 390
Laplace–Beltrami operator, 32 lifting criterion for covering maps, 389
Laplacian, 32 lifting property
conformal, 223 path, 389
Dirichlet eigenvalue of, 51 unique, 389
eigenfunction of, 51 line in a Riemannian manifold, 352, 367,
eigenvalue of, 51 370
in coordinates, 32 linear connection, 91
Neumann eigenvalue of, 51 linear functional, 391
Laplacian comparison theorem, 332, 337 linear isometry, 11
latitude circle, 18 linear ODEs, 106
lattice, 369 local coordinates, 375
law of inertia, Sylvester’s, 42 local defining function, 248, 380
left action, 411 local diffeomorphism, 378
left-invariant metric, 67 local distance function, 176
left-invariant vector field, 408 local frame, 383
left translation, 407 local invariant, 194
length local isometry, 12, 171
additivity of, 34 local parametrization, 17
isometry invariance of, 34 local section, 383
Subject Index 431
of a submersion, 378 semi-Riemannian, 42

local-to-global theorem, 2 singular Riemannian, 46
local trivialization, 382 sub-Riemannian, 46
local uniqueness of constant-curvature met- metric ball, 163
rics, 294 metric comparison theorem, 331
locally conformally flat, 59, 218 metric connection, 118
2-manifolds, 222 metric isometry, 186
hyperbolic space, 66 metric space, 39
spheres, 61 metrically complete, 39, 166
locally Euclidean, 371 equivalent to geodesically complete, 169
locally finite, 376 Milnor’s theorem on growth of the funda-
locally frame-homogeneous, 74 mental group, 364
locally homogeneous, 74 Milnor, John, 72, 364
locally minimizing curve, 157 minimal hypersurface, 242
locally symmetric space, 78, 222, 295, 350 minimal surface, 242
and parallel curvature, 297 minimizing curve, 151
loop, 174, 373 is a geodesic, 156, 165
geodesic, 188 locally, 157
Lorentz group, 67 Minkowski metric, 43
Lorentz metric, 43 Minkowski space, 43
existence of, 44 mixed tensor, 393
lowering an index, 26 Möbius band, 25
model of an axiomatic system, 143
M model Riemannian manifolds, 55
main curve, 152 monodromy theorem
manifold for local isometries, 347
smooth, 374 for paths, 389
topological, 371 Morse index theorem, 307
with boundary, 371 multilinear, 392
without boundary, 372 over C 1 .M /, 398
matrix Lie algebra, 409 multiplicity of conjugacy, 298
matrix Riccati comparison theorem, 325 Munkres, James, 77
maximal diameter theorem, 362 musical isomorphisms, 26
maximal geodesic, 104 commute with covariant derivatives, 126
mean curvature, 238 Myers’s theorem, 8, 361
meridian, 18 Myers, Sumner B., 13, 361
metric Myers–Steenrod theorem, 13
bi-invariant, 68, 147, 224, 259
Carathéodory, 47 N
Carnot–Carathéodory, 46 n-manifold, 371
Euclidean, 12, 57 n-torus, 389, 408
fiber, 29 flat metric on, 19
Finsler, 47 nabla, 89
Fubini–Study, 24, 82, 258, 367 Nash embedding theorem, 117
hyperbolic, 62–67 Nash, John, 117
induced by an immersion, 15 natural coordinates on the tangent bundle,
Kobayashi, 47 384
Lorentz, 43 naturality
on a tensor bundle, 29 of geodesics, 125
on differential forms, 49 of the exponential map, 130
product, 20 of the Levi-Civita connection, 125
pseudo-Riemannian, 42 of the Lie bracket, 386
Riemannian, 1, 11 negative definite, 40
round, 16, 58 negatively oriented basis, 403
432 Subject Index
Neumann eigenvalue, 51 of a boundary, 404

nilpotent Lie group, 72 of a curved polygon, 265
noncollinear, 353 of a hypersurface, 403
nondegenerate 2-tensor, 193 of a manifold, 403
nondegenerate bilinear form, 40 of Rn , 403
nondegenerate n-tuple of vector fields, 41 orientation covering, 404
nondegenerate subspace, 41 orientation form, 403
nonvanishing vector field, 193 orientation-preserving, 403
norm orientation-reversing, 403
on a vector space, 47 oriented basis, 403
on an inner product space, 10 oriented coordinate chart, 403
pseudo-Riemannian, 40 oriented manifold, 403
normal bundle, 16 orthochronous Lorentz group, 67
pseudo-Riemannian, 45 orthogonal complement, 10
normal connection, 231 orthogonal group, 58, 408
normal coordinates, 132 special, 408
polar, 184 orthogonal vectors, 10, 12, 40
pseudo-Riemannian, 132 orthonormal basis, 10
Riemannian, 132 standard order for, 79
normal covering map, 414 orthonormal frame, 14
normal exponential map, 133 for pseudo-Riemannian metric, 43
normal Jacobi field, 287, 288 orthonormal vectors, 10, 41
normal neighborhood osculating circle, 3, 245
of a point, 131 outward-pointing normal, 17
of a submanifold, 133 outward-pointing vector field, 403
normal projection, 17, 226 overdetermined system, 220
normal space, 16
normal variation, 301 P
normal vector, 16, 44 parallel frame, 109
outward-pointing, 17 parallel lines, 143
normal vector field, 16 parallel tensor field, 105, 110
along a curve, 287 parallel transport, 107–110
along admissible curve, 109
O determines the connection, 109
O’Neill’s formula, 224 parallel vector field, 105, 108, 110, 146
one-parameter family of curves, 152 parametrization
one-parameter subgroup, 410 by arc length, 35
one-sided velocity vectors, 34 graph, 18
open geodesic ball, 158 of a submanifold, 17
open map, 372 parametrized curve, 33
open Möbius band, 25 partition of an interval, 33
orbit, 413 partition of unity, 11, 376
orbit space, 413 path, 372
order of conjugacy, 298 path class, 373
ordered basis, 391 path-homotopic, 373
ordinary differential equation, 103 path homotopy, 373
linear, 106 path-lifting property, 171, 264, 389
ordinary vertex, 265 Perelman, Grigori, 77
orientable manifold, 403 Pfaffian, 280
orientation, 402 piecewise regular curve segment, 33
and coordinates, 403 piecewise smooth vector field, 108, 153
and n-forms, 403 pinching theorems, 366
continuous, 403 plane curve, 2
induced, 404 plane curve classification theorem, 3, 282
Subject Index 433
plane section, 250 pseudo-Riemannian normal

Poincaré ball, 62 coordinates, 132
Poincaré disk, 80 pseudo-Riemannian submanifold, 44
Poincaré half-space, 62 pseudohyperbolic space, 79, 83
point reflection, 77 pseudosphere, 79, 83
pointwise conformal metrics, 59 pullback, 398, 399
pointwise ı-pinched, 368 of a connection, 110
pointwise pullback, 398 of a density, 406
polar coordinates, 184 of an exterior derivative, 402
polar Fermi coordinates, 184 pointwise, 398
polar normal coordinates, 184 pullback orientation, 403
polarization identity, 10 pushforward of a vector field, 110, 385
polygon
curved, 265, 271 Q
geodesic, 75, 271 quadratic form, 260, 324
polynomial growth, 363 quotient manifold theorem, 413
positive density, 406 quotient map, 372
positively oriented basis, 403
positively oriented curved polygon, 265, R
271 Radó, Tibor, 276
positively oriented n-form, 403 radial distance function
Preissman’s theorem, 357 for a point, 158
Preissman, Alexandre, 357 for a submanifold, 179
principal curvature comparison theorem, radial geodesics, 133
329 are minimizing, 161
principal curvatures, 4, 237 radial vector field
principal directions, 238 for a point, 158
product metric, 20 for a submanifold, 179
warped, 20, 146 raising an index, 26
product of paths, 373 rank
product rule constant, 378
for connections, 89 of a smooth map, 378
for divergence operator, 51 of a tensor, 393
for Euclidean connection, 118 of a vector bundle, 382
product, direct, 408 rank theorem, 378
projection equivariant, 412
central, 63 global, 379
hyperbolic stereographic, 65 Rauch comparison theorem, 331
normal, 17, 226 ray in a Riemannian manifold, 189
of a vector bundle, 382 real projective space, 25, 255, 413
stereographic, 59 regular coordinate ball, 374
tangential, 17, 226 regular curve, 33
projective plane, 145 regular domain, 30, 379
projective space regular level set, 27, 380
complex, 24, 82, 258, 367, 413 regular point of a smooth map, 27, 380
real, 25, 255, 413 regular value of a smooth map, 380
proper group action, 412 related vector fields, 385
proper map, 372 relativity
proper variation, 152 general, 43, 211
proper vector field along a curve, 153 special, 43
properly embedded, 379 reparametrization, 34
pseudo-Euclidean space, 43, 79 backward, 34
pseudo-Riemannian geodesics, 124 forward, 34
pseudo-Riemannian metric, 42 of an admissible curve, 34
434 Subject Index
representation S
of a Lie algebra, 411 SSS theorem, 2
of a Lie group, 408 scalar curvature, 208
rescaling lemma, 127 geometric interpretation of, 253
restricted exponential map, 128 scalar product, 40
restricted holonomy, 150 scalar product space, 40
Riccati comparison theorem, 324–327 scalar second fundamental form, 235, 259
matrix, 325 Schoen, Richard, 211, 368
Riccati equation, 323, 343 Schouten tensor, 215
Riccati, Jacopo, 324 Schur’s lemma, 210
Ricci curvature, 207–209 secant angle function, 267
second Bianchi identity, 204
geometric interpretation of, 253
second covariant derivative, 99
Ricci decomposition of the curvature tensor,
second fundamental form, 227
216
geometric interpretation of, 233
Ricci flow, 368
scalar, 235, 259
Ricci identities, 205
second structure equation, 222
Ricci tensor, 207–209 second variation formula, 300
geometric interpretation of, 253 section
Riemann curvature endomorphism, 196 of a submersion, 378
Riemann curvature tensor, 198 of a vector bundle, 383
Riemann, Bernhard, v, 47, 255 smooth, 383
Riemannian connection, 122–125 sectional curvature, 8, 250, 251
Riemannian covering, 24, 25 constant, 254
Riemannian density, 31 determines the curvature tensor, 252
Riemannian distance, 36 of Euclidean space, 254
isometry invariance of, 37 of hyperbolic spaces, 254
on a disconnected manifold, 36, 53 of spheres, 254
Riemannian geodesics, 124 segment, curve, 33
Riemannian isometry, 186 self-dual, 50
Riemannian manifold, 1, 11 semi-Riemannian metric, 42
with boundary, 11 semicolon between indices, 98
Riemannian metric, 1, 11 semidirect product, 57, 412
existence of, 11 semigeodesic coordinates, 181, 182
Riemannian normal coordinates, 132 shape operator, 235, 259
Riemannian submanifold, 15, 225, 226 sharp (]), 26–28
of a pseudo-Riemannian manifold, 44 sheets of a covering, 388
Riemannian submersion, 22, 146, 224, 258 side of a curved polygon, 265
side-side-side theorem, 2
Riemannian symmetric space, 77, 189
sign conventions for curvature tensor, 198
locally, 78, 222, 295
sign of a permutation, 400
Riemannian volume form, 30
signature of symmetric bilinear form, 42
is parallel, 126
signed curvature, 3
right action, 411 of curved polygon, 273
right-invariant metric, 67 simple closed curve, 264
right-invariant vector field, 408 simply connected space, 373
right translation, 407 covering of, 390
rigid motion, 2 single elliptic geometry, 145
rotation index, 264 singular Riemannian metric, 46
of a curved polygon, 266, 272 slice coordinates, 379, 380
rotation index theorem, 266 boundary, 380
for a curved polygon, 272 smooth atlas, 374
rough section, 383 smooth chart, 374
round metric, 16 smooth coordinate ball, 374
Subject Index 435
smooth coordinate chart, 374 structure equation

smooth coordinate domain, 374 first, 113
smooth covering map, 388 second, 222
smooth curve, 33 sub-Riemannian metric, 46
smooth group action, 412 subalgebra, Lie, 409
smooth manifold, 374 subbundle, 383
with boundary, 374 subgroup, Lie, 407
smooth manifold structure, 374 subinterval of a partition, 33
smooth map, 374, 375 submanifold
smooth structure, 374 embedded, 379
smooth subbundle, 383 immersed, 379
smoothly compatible, 374 pseudo-Riemannian, 44
solvable Lie group, 72 Riemannian, 15, 44
soul theorem, 367 tangent space to, 381
space form, 349–350 with boundary, 379
Euclidean, 349 submersion, 378
hyperbolic, 349 Riemannian, 22, 146, 224, 258
spherical, 349 subordinate to a cover, 376
special linear group, 408 summation convention, 375
special orthogonal group, 408 support
special relativity, 43 of a function, 376
special unitary group, 408 of a section of a vector bundle, 383
spectral theorem, 237 surface of revolution, 18
speed of a curve, 35 geodesics, 145, 256
sphere, 58 Sylvester’s law of inertia, 42
Berger, 81, 259 symmetric connection, 112, 121
distance function, 185 symmetric group, 396
geodesic, 158, 163 symmetric product, 396
is locally conformally flat, 61 symmetric space, 77, 189, 350
principal curvatures of, 6 irreducible, 79
round, 16, 58 locally, 78, 222, 295, 297, 350
sectional curvature, 254 symmetric tensor, 395, 396
volume of, 314 symmetries
sphere theorem, 367 of Euclidean space, 57, 147
differentiable, 368 of hyperbolic spaces, 67, 147
spherical coordinates, 19 of spheres, 58–59, 147
spherical space form, 349 of the curvature tensor, 202
splitting theorem, 367 symmetrization, 395
standard basis of Rn , 403 symmetry lemma, 154
standard coordinates on Rn , 12 symplectic form, 193
standard order for orthonormal basis, 79 Synge’s theorem, 364
standard orientation of Rn , 403 Synge, John L., 361, 364
standard smooth structure, 375
star operator, 50, 52, 53 T
star-shaped, 128, 402 tangent angle function, 264, 265, 272
Steenrod, Norman E., 13 tangent bundle, 382
stereographic projection, 59 tangent covector, 377
hyperbolic, 65 tangent cut locus, 310
is a conformal diffeomorphism, 61 tangent space, 376
Stiefel manifold, 48 to a submanifold, 381
Stokes orientation, 404 tangent to a submanifold, 386
Stokes’s theorem, 265, 405 tangent vector, 376
stress-energy tensor, 211 tangential acceleration, 86
strictly ı-pinched, 367 tangential connection, 92–93, 117, 228
436 Subject Index
tangential directional derivative, 87 two-point boundary problem, 299

tangential projection, 17, 226
tangential vector field along a curve, 287 U
tensor Umlaufsatz, 266
alternating, 400 uniform tubular neighborhood, 133
contravariant, 392 uniformization theorem, 6
covariant, 392 for compact surfaces, 74
mixed, 393 uniformly ı-normal, 163
on a manifold, 396 uniformly normal, 163
symmetric, 395 unique lifting property, 389
tensor bundle, 396 uniqueness of constant-curvature metrics,
tensor characterization lemma, 398 294
tensor field, 397 unit-speed curve, 19, 35
along a curve, 101 unit tangent bundle, 14
tensor product, 393 unit tangent vector field, 264, 271
Theorema Egregium, 5, 249 unitary group, 408
Thurston geometrization special, 408
conjecture, 77, 368 universal anti-de Sitter space, 83
topological manifold, 371 universal covering manifold, 390
with boundary, 371 upper half-plane, 80
torsion 2-forms, 113 upper half-space, 62
torsion element of a group, 356 upper indices on coordinates, 375
torsion-free group, 356
torsion tensor, 112, 121 V
torus, 389, 408 vacuum Einstein field equation, 211
flat metric on, 19, 195, 369 variation
n-dimensional, 20, 389, 408 first, 154
total covariant derivative, 98 normal, 301
commutes with musical isomorphisms, of a curve, 152
126 proper, 152
components of, 98 second, 300
total curvature theorem, 3, 271, 276 through geodesics, 284
total scalar curvature functional, 210, 211 variation field, 153
total space of a vector bundle, 382 variational equation, 157
totally geodesic, 234 variations, calculus of, 152
trace vector bundle, 382
of a tensor, 395 section of, 383
of an endomorphism, 260 subbundle, 383
with respect to g, 28 zero section, 383
traceless Ricci tensor, 208 vector field, 384
transition function, 383 along a curve, 100
transition map, 374 along a family of curves, 152
transitive group action, 412 canonical form for, 387, 388
on fibers, 23 commuting, 388
translation, left and right, 407 coordinate, 384
transport, parallel, 107–110 Lie algebra of, 409
transverse curve, 152 nonvanishing, 193
transverse Jacobi field, 315, 317 normal, along a curve, 287
triangle piecewise smooth, 108
Euclidean, 2 proper, 153
ideal, 282 tangential, along a curve, 287
triangulation, 276 vector, tangent, 376
trivial free homotopy class, 174 velocity, 33, 86, 87, 386
tubular neighborhood, 133 vertex
Subject Index 437
cusp, 265 warped product, 20, 146

flat, 265 wedge product, 400
of a curve, 265 Alt convention, 401
ordinary, 265 determinant convention, 401
vertical action, 23 Weingarten equation, 229
vertical index position, 394 for a hypersurface, 236, 259
vertical tangent space, 21 Weingarten map, 229, 315
vertical vector field, 21 Weyl–Schouten theorem, 220
volume Weyl tensor, 215
of a ball, 314 conformal invariance of, 217
of a Riemannian manifold, 30 Wilking, Burkhard, 368
of a sphere, 313 Wolf, Joseph, 349
volume comparison theorem
Bishop–Gromov, 340 Y
Günther, 334 Yamabe equation, 223
volume form, 30
Yamabe problem, 211
von Mangoldt, Hans Carl Friedrich, 352
W Z
Warner, Frank, 279 zero section, 383

Introduction To Riemannian Manifolds

Uploaded by

Copyright:

Available Formats

Introduction To Riemannian Manifolds

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Riemannian Manifolds

Uploaded by

Copyright:

Available Formats

Graduate Texts in Mathematics

Alejandro Adem, University of British Columbia

More information about this series at http://www.springer.com/series/136

ISSN 0072-5285 ISSN 2197-5612 (electronic)

Mathematics Subject Classiﬁcation (2010): 53-01, 53C20, 53B20

Originally published with title “Riemannian Manifolds: An Introduction to Curvature”

Riemannian geometry is the study of manifolds endowed with Riemannian metrics,

Exercises and Problems

About the Second Edition

Seattle, Washington, USA John M. Lee

3 Model Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5 The Levi-Civita Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

6 Geodesics and Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

8 Riemannian Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

9 The Gauss–Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

The Gauss–Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

10 Jacobi Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

11 Comparison Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

12 Curvature and Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Appendix A: Review of Smooth Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 371

Appendix B: Review of Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391

Appendix C: Review of Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407

The Euclidean Plane

treatment of geometry as a mathematical subject began with Euclidean plane geom-

Theorem 1.4 (Circumference Theorem). The circumference of a Euclidean circle

If we want to continue our study of plane geometry beyond figures constructed

Fig. 1.1: Osculating circle

Theorem 1.5 (Plane Curve Classification Theorem). Suppose and z W Œa; b !

Theorem 1.6 (Total Curvature Theorem). If W Œa; b ! R2 is a unit-speed simple

Fig. 1.2: Computing principal curvatures

Fig. 1.3: S1 Fig. 1.4: S2

Fig. 1.5: K > 0 Fig. 1.6: K < 0

where .S / is the Euler characteristic of S .

The uniformization theorem is a classification theorem, because it replaces the

Curvature in Higher Dimensions

As we will see in Chapter 3, we again have three classes of constant (sectional)

Theorem 1.9 (Characterization of Constant-Curvature Metrics). The complete,

On the other hand, there are a number of powerful local-to-global theorems,

Theorem 1.10 (Cartan–Hadamard). Suppose M is a complete, connected Rie-

Theorem 1.11 (Myers). Suppose M is a complete, connected Riemannian mani-

© Springer International Publishing AG 2018 9

(iii) P OSITIVE D EFINITENESS : hv; vi 0, with equality if and only if v D 0.

I Exercise 2.2. Prove the preceding lemma.

The angle between two nonzero vectors v; w 2 V is defined as the unique 2

span.b1 ; : : : ; bk / D span.v1 ; : : : ; vk / for each k D 1; : : : ; n: (2.4)

Proof. The basis vectors b1 ; : : : ; bn are defined recursively by

Because v1 ¤ 0 and vj … span.b1 ; : : : ; bj 1 / for each j 2, the denominators are all

To extend these geometric ideas to abstract smooth manifolds, we define a structure

Proposition 2.4. Every smooth manifold admits a Riemannian metric.

I Exercise 2.5. Use a partition of unity to prove the preceding proposition.

We will give a number of examples of Riemannian metrics, along with several

Let g be a Riemannian metric on a smooth manifold M with or without boundary.

hv; wig D gp .v; w/:

When working with R as a Riemannian manifold, we always assume we are using

the Euclidean metric unless otherwise specified. //

A Riemannian n-manifold is said to be flat if it is locally isometric to a Euclidean

Local Representations for Metrics

For example, the Euclidean metric on Rn (Example 2.6) can be expressed in

The matrix of gx in these coordinates is thus gxij D ıij .

Theorem 1.5 (Plane Curve Classification Theorem). Suppose and z W Œa; b !

Theorem 1.6 (Total Curvature Theorem). If W Œa; b ! R2 is a unit-speed simple

where .S / is the Euler characteristic of S .

Because we usually identify Tp M with its image in Tp M under d p , and think of

Computations on a submanifold M M are usually carried out most conve-

Example 2.19 (Metrics in Graph Coordinates). If U Rn is an open set and

In particular, if is a unit-speed curve (meaning that j 0 .t /j2 D a0 .t /2 Cb 0 .t /2 1),

u D div.grad u/: (2.20)