Amol Sasane - A Mathematical Introduction To General Relativity

A Mathematical
Introduction to
General Relativity
Amol Sasane
, . , World Scientific
A Mathematical
Introduction to
General Relativity
Other World Scientific Titles by the Author
A Friendly Approach to Complex Analysis

ISBN: 978-981-4578-98-1
ISBN: 978-981-4578-99-8 (pbk)
Plain Plane Geometry

ISBN: 978-981-4740-43-2
ISBN: 978-981-4740-44-9 (pbk)
A Friendly Approach to Functional Analysis

ISBN: 978-1-78634-333-8
ISBN: 978-0-00-098763-1 (pbk)
A Mathematical
Introduction to
General Relativity
Amol Sasane
London School ofEconomicJ'~ UK
1S, World Scientific

NEW JERSEY , LONDON , SINGAPORE , BEIJING , SHANGHAI , HONG KONG , TAIPEI , CHENNAI , TOKYO
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
Library of Congress Control Number: 2021038220
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
Cover image: Spacetime Pythagoras Theorem
A MATHEMATICAL INTRODUCTION TO GENERAL RELATIVITY

Copyright © 2022 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any i,iformation storage and retrieval
system now known or to be invented, without written permission from the publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
ISBN 978-981-124-377-6 (hardcover)

ISBN 978-981-124-378-3 (ebook for institutions)
ISBN 978-981-124-379-0 (ebook for individuals)
For any available supplementary material, please visit

hUpS;//www.wotl~it:i(ifilie;~r!dsQiboobfl0,1142/l2454#tl=sUppl
Printed in Singapore
To my teachers
This page intentionally left blank
Preface
The aim of the book is to give a mathematical presentation of the theory

of general relativity (i.e., spacetime-geometry-based gravitation theory) to
advanced undergraduate or beginning graduate mathematics students. The
textbook is intended for self-study, and so the solutions to all the exercises
are included at the end of the book.
The prerequisites for reading the book are undergraduate-level mathe-
matical analysis (e.g. having seen the inverse function theorem in a mul-
tivariable context, Picard's existence and uniqueness theorem for ordinary
differential equations, etc.) and undergraduate-level linear algebra. Useful
references to look-up these things are e.g. [Apostol(1969)], [Rudin(1983)].
Familiarity with differential geometry is not assumed. A background in
physics is not needed, except for a basic university-level general physics
course (so that the reader has seen the Newtonian laws of motion and the
gravitational law). In case it is needed, we recommend [Feynman(1963)] as
reference for this part.
Mathematicians will find spacetime physics presented in the definition-
theorem-proof format familiar to them. Precise mathematical definitions
of physical notions are amenable to me, as they help avoiding pitfalls, es-
pecially in dealing with physics describing phenomena that are counter-
intuitive to everyday experiences. The definition of physical terms is given
using mathematical objects, but I have often tried to supplement these
with motivating physical ideas. Thus, for example, first the definition of
an 'inertial frame' is given as a special type of chart, and subsequently it
is shown that, in this chart, the coordinate representations of geodesics are
aflinely-parametrised straight lines (i.e., Newton's first law holds: Free-fall
motion is 'uniform' along a straight line; the 'uniform' part also comes from
the geodesy), or an 'observer' is defined as a timelike future-pointing curve,
and so on.
vii
viii A Mathematical Introduction to General Relativity
The entire book's content can be summarised as follows:

• Spacetime is a 4-dimensional Lorentzian oriented manifold with a time-
orientation.
• Spacetime curvature is governed by the matter distribution via the 'field
equation'.
• The motion of test matter in spacetime is given by the geodesic equation.
In the first part of the book, we develop the differential geometry of
Lorentzian manifolds needed to comprehend the above statements. We
choose many of our illustrating examples as the Lorentzian manifolds which
are later spacetime models. This choice of examples will serve the twofold
purpose of making the subsequent physics forthcoming in the second part
relatable, and the mathematics learnt in the first part less dry.
Then in the second part, we focus on the physics, covering much of
the essential material in the 20th century spacetime-based view of gravity:
energy-momentum tensor field of matter, field equation, examples of space-
times, Newtonian approximation, geodesics, tests of the theory, black holes,
and end with a discussion of cosmological models of the universe.
Before we proceed, it is useful to keep the following rough picture in
mind. Spacetime is the collection of all 'events'. An event is roughly a point,
'here and now'. Although spacetime (M, g) is a 4-dimensional manifold, we
have instead drawn a two-dimensional surface here:
(M,g)
/
'
What determines the spacetime (M, g)? The answer is the field equation 1:
(gcornc t, v J Ric - ½sg +Ag= 81rT ( nrn u c , J
--------------
Terms appearing here will be explained eventually:
1 Ones on the left-hand side (cap-
turing the geometry of M), will be covered in Chapter 9, while the one on the right-hand
side (describing the matter content in M) will be specified in Chapter 13.
Preface ix
where the right-hand side T describes the matter distribution (called the
'energy-momentum tensor field of matter'), and the left-hand side is ex-
pressible in terms of the metric g. The metric g is something that equips
the tangent space TpM at each point of the manifold M with a Lorentzian
scalar product, and essentially determines the 'shape' of M. So to obtain
M, one needs to find a metric g which is a solution to this field equation.
Given the metric g, the motion of 'test' matter (that is matter whose
effect on the T can be neglected) is given by the geodesic equation 2 ,
'v'v-r V.,, = 0,
where V.,, is the velocity field along 'Y, and this describes the motion of a
particle in the spacetime M by means of a curve 'Y : [a, b] -+ M, called its
'worldline'.
Does space 'evolve'? Or, what is the reality experienced by us, in terms
of our view of the spatial world, with a notion of time passing in our minds?
Firstly, we are a 'worldline', namely a curve 'Y: J-+ M, where I c JR is an
interval, thought of as the list of events we encounter in our lifetimes-the
curve merely describes the order in which we meet these events. The 'time'
we experience between two events p="((a) and p' ="f(b) along our worldline
is the 'proper time', defined by
r(p,p') := f g('Y(t))(v.,,,.,,(t), v.,,,.,,(t)) dt.
(The notation will be clarified eventually in due course.) We will see that
our worldline 'Y at each event p = 'Y(t) along it, has a 'timelike' tan-
gent vector v.,, ,P in the 4-dimensional tangent space TpM. This v.,, ,P has
a 3-dimensional orthogonal complement (v.,,,p)1_ (with respect to the scalar
product g(p)), and this is roughly the 'space' we perceive at each moment.
Thus the time experienced is 'personal' to each observer, and so is the
perceived 'space' at each moment. Different observers thus have different
ticking clocks (depending on what worldline they have in spacetime, and
also different spatial realities.
How does one use this theory of spacetime in classical (i.e. nonquan-
tum) physics? Given a matter distribution, for example a star, or even the
universe, the field equation is solved to obtain the metric g (analogous to
finding the 'gravitational potential' U by solving a Poisson equation involv-
ing the matter density distribution pin classical Newtonian gravity; minus
2 The notation which appears here will be clarified in Chapter 7, and we will discuss
geodesics in Chapter 8.
X A Mathematical Introduction to General Relativity
the gradient of the potential then gives the gravitational field g). This can
then be used to obtain the geodesic equations for test matter (analogous
to Newton's second law of motion; knowing the acceleration g, and solving
the resulting equation describes the particle trajectory).
It is well-known that one road towards the goal of mastering a subject is
to teach it, since it makes it necessary to organise what has been learnt, so
that the core principles are laid bare, and these notes were originally writ-
ten with this intention. I hope that readers of this book will find it useful.
For readers who want to delve deeper into the subject, some of the sources
that have influenced the writing of this book, are recommended. A deeper
and a more complete treatment is given in the book [Sachs and Wu(1977)],
but it assumes prior familiarity with differential geometry, while we develop
this language along the way, starting from scratch. A more modern text-
book, with an emphasis on Lorentzian geometry, is [O'Neill(1983)]. More
elementary than these two books, but also approaching the subject mathe-
matically, are the clear online lectures by Frederic Schuller [Schuller(2015)],
and the book [Oloff(2018)] (in German3 ). I have used these and many other
sources which are listed in the bibliography, but no claim to originality is
made for any of the discussed material in case there is a missing reference.
I am grateful to Jose Natario for reading the book, for pointing out
various errors and typos, and for suggesting improvements in the text and
additional exercises. Of course, any remaining errors or deficiencies are
solely my responsibility. In particular, an improved proof of Lemma 5.1,
and Exercises 5.19, 5.20, 8.8, 9.15, 12.9, 13.4, 14.5, 14.6, 14.9, 16.3, 16.6,
16. 7 were suggested by Jose. The flagged errors, and the suggestions for
improvements, are too numerous to list here.
Thanks are also due to OlofRubin for pointing out a gap in Theorem 1.2,
to Sara Sal Santos for useful feedback, including spotting a mistake in an
earlier version of the picture on page 144, and to Adam Ostaszewski for
his comments on the Preface. It is a pleasure to thank Rochelle Kronzek
and Rok Ting Tan at World Scientific for their help and enthusiasm. I am
indebted to Sara Maad Sasane, for useful discussions, detailed feedback on
the first six chapters, and encouragement.
Amol Sasane
Lund, 2021
3 0f which an English translated edition is underway.

Notation and terminology
We use N for the set {1, 2, 3, • • •} of natural numbers, Z for the set
{· • • , -1, 0, 1, • • •} of integers, and JR for the set ofreal numbers. The equiv-
alence class of an element a E S under an equivalence relation on S will be
denoted by [a]. We denote the identity map on a set S, S 3 x ...... x E S,
ids, and if Sis clear from context, simply by id. For a function f: X-> Y,
and a y E Y, the notation f = y means that for all x EX, f(x) = y. RHS
means 'right-hand side', and LHS means 'left-hand side'.
Points or vectors in JR 3 are often denoted by bold face letters, for exam-
ple, x = (x,y,z) E JR 3 . The 'vector/cross product' of vectors x,y E JR 3
is denoted by x x y E JR3 . The components of a vector x in !Rm are
often denoted by xi, 1 ~ i ~ m, in accordance with the physics liter-
ature, and if e.g. the square of xi E JR is meant, we will use paren-
thesis, writing (xi) 2 . The standard basis vectors in !Rm are given by
e1 = (1, 0, · · · , 0), · · · , em = (0, · · · , 0, 1). We will also use the nota-
tion (x, y) = x 1 y 1 + ••• + xmym for the Euclidean inner product of
x = (x1,- · · , xm) and y = (y1,- · · , ym) E !Rm, and llxll := -./(x, x) for
the Euclidean 2-norm. Given x E !Rm and r > 0, the open ball B(x, r) in
the Euclidean topology is B(x,r) = {y E !Rm: IIY - xii < r}. The closure
of a subset S of a topological space will be denoted by S. Thus the clo-
sure of the open ball B (x, r) in the Euclidean topology is the closed ball
B(x, r) = {y E !Rm : IIY - xii ~ r}. Throughout the book, unless otherwise
stated, repeated indices, mostly one appearing as a superscript and one as
a subscript, will be summed over the range of the index in question. Thus
here xiei means x 1 e 1 + •••+ xmem. The Kronecker delta symbol is denoted
by 8j, or sometimes by 8ij, and means the following:
8i = { 1 if i = j,
1 0 if i =I- j.
xi
xii A Mathematical Introduction to General Relativity
Given a real vector space V, its dual vector space consisting of all linear
maps w : V -+ IR, will be denoted by V* , and if {ei, i E J} is a basis for
V , the dual basis is denoted by {Ei, i E J}, and is defined by €i ( ej) = 8).
If Sc V, then spanS is the subspace consisting of all linear combinations
of vectors from S (with an empty sum being defined as the zero vector).
Also, -S := {-v : v E S}. An invertible linear transformation T : V -+ W
between vector spaces V, W is referred to as a linear isomorphism.
Let a~, bij , cij , 1 :,;;; i :,;;; n and 1 :,;;; j :,;;; m, be collections of numbers
or functions. Then writing A = [a~], B = [bij], C = [cij] means A , B , C
are the n x m matrices, whose entry in the ith row and jth column is a~,
respectively bij, cij. The identity matrix is denoted by I or sometimes by
Im (to emphasise its size as an m x m matrix). Thus Im = [8j] . The
transpose of a matrix M will be denoted by Mt. The determinant of a
square matrix M is denoted by <let M. !Rnxm denotes the set of n x m
matrices with real entries. GL(IR,m) stands for the group of invertible in
!Rmxm with the operation of matrix multiplication. The zero entries in a
matrix are sometimes left as blanks.
For n E N the nth order derivative of an n times differentiable function
f: JR-+ JR at a point t E JR will be denoted by J<nl(t). If Uc !Rm is an open
set, then for an f : U -+ !Rn having the component functions / 1 , · · · , fn,
and differentiable at p E !Rm , the Jacobian matrix off is given by
OU 1 1
of 1 (p) ... OUm
of (p) 1
[f'(p)] =
ror :
or
: .
OU 1 (p) ... OUm (p)
Continuously differentiable functions on U will be said to be C 1 on U. This

is equivalent to the first order partial derivatives of f being continuous on
U. If f has partial derivatives of all orders on U and they are continuous,
then f is said to be C 00 on U.
We also have the following table of physical constants. We make two
remarks: c is not reserved for the speed of light exclusively, as we will also
use it to denote a real constant number in mathematical arguments. Also,
we will often use units in which c = GN = 1.
Physical constant I Symbol I Value

Speed of light C 3 x l0 8 ms- 1
(Newton) Gravitational constant GN 6.67 x 10- 11 m3 kg- 1 s- 2
Planck constant h 6.63 x 10- 34 m 2 kgs- 1
Contents
Preface vii
Notation and terminology xi
1. Smooth Manifolds 1
1.1 Charts and atlases 1
1.2 Topology on a smooth manifold . 10
1.3 Smooth maps . . . . . . . . . . . 13
2. Co-/Tangent space 17
2.1 Tangent vectors to a surface as derivations. 18
2.2 Tangent vector definition . . . . . . . . . 22
2.3 Bump functions . . . . . . . . . . . . . . . 23
2.4 Chart-induced basis for the tangent space 26
2.5 Derivatives of smooth maps 30
2.6 Cotangent space .. 34
2. 7 Pull-back of 1-forms 37
3. Co-/Tangent bundles 39
3.1 Tangent bundle 40
3.2 Vector fields .. 41
3.3 Push forward of vector fields under diffeomorphisms 44
3.4 Lie algebra of vector fields . 46
3.5 Cotangent bundle . . . . . 48
3.6 1-form fields . . . . . . . . . 50
3.7 Pull-back of a 1-form field . 53
3.8 Integral curves and the flow of vector fields 54
xiii
xiv A Mathematical Introduction to General Relativity
4. Tensor fields 61
4.1 (r, s)-tensor fields . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Point evaluation of tensor fields . . . . . . . . . . . . . . . 63
4.3 Tensor field components, tensor product and contraction . 68
4.4 Pull-back of (0, s)-tensor fields . . . . . . . . . . . . . . . 72
5. Lorentzian manifolds 75
5.1 Scalar product 76
5.2 Semi-Riemannian, Riemannian, Lorentzian manifolds. 85
5.3 Time-orientation, observers, and proper time 90
5.4 Raising/Lowering indices or musicality . 103
6. Levi-Civita connection 107

6.1 Motivation for connections: the case of !Rm 108
6.2 Definition of a connection . . . . . . . . . . . . . . . . . 110
6.3 Locality and connection coefficients/Christoffel symbols 114
6.4 The Levi-Civita connection . . . . . 119
6.5 Covariant derivative of tensor fields . 125
6.6 Lie derivative of tensor fields 131
7. Parallel transport 137

7.1 Vector fields along curves 137
7.2 Covariant derivative along a curve 140
7.3 Parallel transport . . . . . . . . . . 148
8. Geodesics 155
8.1 Geodesic equation 155
8.2 Existence and uniqueness 158
8.3 Affine parameter . . . . . 162
8.4 Lightlike geodesics in Schwarzschild spacetime. 165
8.5 The exponential map . . . . . 170
8.6 Normal coordinates . . . . . . 174
8. 7 Geodesics as extremal paths . 176
9. Curvature 181
9.1 Ricci and scalar curvatures . . . . . . . . . . . . . . . . 188
9.2 Cosmological time and cosmological red-shift in FLRW. 192
9.3 Geodesic deviation and curvature . . . . . . . . . . . . . 195
Contents xv
9.4 Tidal forces 198
10. Form fields 209

10.1 k-forms . . . . . . 209
10.2 k-form fields . . . . 216
10.3 Exterior derivative 221
10.4 Interior multiplication 230
11. Integration 235

11.1 Orientation 236
11.2 Partitions of unity 241
11.3 Integration . . . . 245
11.4 Volume form field in semi-Riemannian manifolds 249
11.5 Hodge star . . . . . . . . . . . . . . . . . . . . . 252
12. Minkowski spacetime physics 261

12.1 Classical spacetime . . . 261
12.2 Minkowski space . . . . 264
12.3 Relative velocity addition 269
12.4 Energy and momentum 274
12.5 Electromagnetism 280
13. Matter 289

13.1 Newtonian gravity and the Poisson equation. 289
13.2 Classical dynamics from a geometric viewpoint 292
13.3 Energy-momentum tensor field . . . . . . . . . 294
14. Field equation 305

14.1 Newtonian limit 308
14.2 Some symmetries of spacetime 312
14.3 Schwarzschild spacetime 314
14.4 Perihelion precession 318
15. Black holes 327

15.1 The spacetime (MBH, g) . . . . . . . 328
15.2 Null geodesics and the event horizon 332
15.3 Kruskal extension of MBH 335
15.4 Miscellanea . . . . . . . . . . . . . . 341
xvi A Mathematical Introduction to General Relativity
16. Cosmology 345

16.1 Sectional curvatures 346
16.2 FLRW spacetime .. 352
16.3 Field equations . . . 354
16.4 Simple case solutions and the big bang . 356
Solutions 361
Bibliography 477
Index 479
Chapter 1
Smooth manifolds
As mentioned in the Preface, spacetime is, on the set-theoretic level, a

collection of points, called events. We want to do calculus on spacetime,
because firstly, the field equation determining the geometry of spacetime
is a partial differential equation on spacetime, and secondly, the geodesic
equation describing motion of matter in spacetime is an ordinary differential
equation on spacetime. It was discovered in the early twentieth century
that the appropriate structure to be used to describe spacetime is that of
a Lorentzian smooth manifold. In this chapter we will first learn about the
notion of a smooth manifold.
Roughly speaking, an m-dimensional smooth manifold is a set M which
can be covered by patches such that in each patch one can introduce coor-
dinates and use them to do calculus. As coordinates are ad-hoc, we need
to make sure that only certain 'admissible' coordinates are allowed, so that
the definition of smoothness is independent of the choice of coordinates.
It turns out that this also endows the manifold with a topology, so that
smooth manifolds are special types of topological spaces.
1.1 Charts and atlases
We have prior experience with using coordinates, for example, for the sur-
face of a sphere, one could use polar and azimuthal angles. But then we
realise that in order to obtain an injective mapping from the sphere to the
set of parameter ranges of the coordinates, we must work with patches on
the sphere, instead of the whole sphere in one go. So we anticipate that
also in the case of a manifold, it will need to be covered by patches, and
coordinates need to be set up in each patch.
1
2 A Mathematical Introduction to Geneml Relativity
Before we introduce the notion of a manifold, we explain what we mean

by putting coordinates in a patch. The aim of this section is to introduce
the notions of a 'chart' and an 'atlas'. Roughly speaking, a chart describes
the local coordinates set up in a patch of the manifold, and an atlas is
a collection of such charts so that the union of the patches of the charts
covers the whole manifold. In order that there is no conflict later on when
defining smooth objects on the manifold, we will demand that the charts
that make up an atlas are 'compatible'.
First, let us recall the notion of a topological space.
Definition 1.1. (Topological space).

A topological space (M, 0) is a set M together with a collection O of subsets
of M such that the following hold:
(Tl) 0,M E 0.
(T2) Whenever Ui, i EI, belong to 0, also LJ Ui E 0.
iel
(T3) Whenever U, VE 0, also Un VE 0.
The elements of O are called open sets, and O itself is referred to as a
topology on M.
Recall that !Rm can be equipped with its usual Euclidean topology.
Example 1. 1. (Euclidean space !Rm).

Let !Rm= {x = (x1, • • • , xm): xe E JR, 1 ~ £ ~ m} be the real vector space
with componentwise operations of vector addition and multiplication by
real scalars. For vectors x, y E !Rm, their Euclidean inner product is given
by (x, y) = x 1 y 1 + •••+ xmym. For x E !Rm, we define the ball B(x, r) with
center x and radius r > 0 by B(x, r) = {y E !Rm : (x - y,x - y) < r 2 }.
We define O to be the collection of subsets U of !Rm with the property that
whenever x EU, there exists an r > 0 such that B(x, r) c U. Then O is a
topology on !Rm. ◊
Definition 1.2. (Chart).

Let M be a set. An m-chart on Mis a pair (U, cp), where Uc M, the map
<p : U -+ !Rm is injective, and cp(U) is an open subset of !Rm.
Henceforth, we will often drop the specification 'm' in 'm-chart', and simply
refer to 'charts' for M, with the understanding that for a given M, them is
fixed. A chart allows us to talk about the coordinates of a point p E U c M,
with respect to the chart (U, <p), as the m-tuple of numbers cp(p) E !Rm.
Smooth Manifolds 3
Example 1.2. (A chart for !Rm). Let U = !Rm, and cp: !Rm-+ !Rm be the
identity map id. Then (!Rm, id) is a chart on !Rm. ◊
We will see later on that locally a manifold looks like !Rm (its tangent space).
In Example 1.1, there is a distinguished point, namely the origin, but
as nature does not provide natural coordinate systems, we also introduce
!Rm which has forgotten its origin, namely an affine space.
Example 1.3. (Affine space).

An affine space of dimension m consists of
• a set M (of 'points'),
• an m-dimensional real vector space V (whose vectors are used for
'translating'),
• a map M x V 3 (p, v) >-+ p + v E M
such that the following hold:
(Al) for all p EM, u, v EV, p + (u + v) = (p + u) + v,
(A2) for all p E M, p + 0 = p,
(A3) for all p, q EM, there exists a unique vector Vpq E V such that
q =p+vpq·
Let p EM be fixed and let {e 1 , • • • , em} be a basis for V. Given any q EM,
there exists a unique vector Vpq EV such that q = p + Vpq· This vector Vpq
can be expressed in terms of the basis vectors, giving unique coordinates
cp(q) := (x1, • • • ,xm) E !Rm. Thus q = p + Vpq = p + xiei. Clearly, cp is
injective, and with U := M, cp(U) =!Rm.Thus (M,cp) is a chart on M. ◊
Exercise 1.1. Show that if p, q, r EM, then Vpr = Vpq + Vqr•
Example 1.4. (Charts on the sphere 8 2 ). Let
8 2 := {(x,y,z) E JR3 : x 2 + y2 + z 2 = 1},
and n = (0, 0, 1), s = (0, 0, -1) denote the north and south poles in 8 2 • Set
Un= 8 2 \{n},
Us = 8 2 \{s},
and define the 'stereographic' projections
2 1 2
8 \{n} = Un 3 (x, y, z) i---+ 'Pn(x, y, z) = 1 _ z (x, y) E JR ,
1
8 2 \{s} = Us 3 (x, y, z) i---+ 'Ps(x, y, z) = -1 -(x, y) E
2
JR .
+z
See Figure 1. 1.
4 A Mathematical Introduction to General Relativity
Fig. 1.1 Charts (Un, 'Pn) and (Us, cp 8 ) on S 2 . Looking at the two similar triangles in
the left picture, we have (1- z) : 1 = x: u and (1- z) : 1 = y: v. Analogously, from the
two similar triangles in the right picture, (1 + z): 1 = x: u and (1 + z): 1 = y: v . This
holds irrespective of the sign of z.
Then (Un, 'Pn) and (Us, 'Ps) are charts on S 2 . ◊
Exercise 1.2. Show that the inverse cp;; 1 : JE. 2 -+ S2 \{n} of the map 'Pn is given by
(u , v)
'P;:; 1 (
I-----+
2u 2v u 2 +v 2
u2 + v2 + 1 ' u2 + v2 + 1 ' u2 + v2 + 1 .
-1)
. 2 2 x 2 + y2 1 - z2
Hint: u + v = (l - z)2 = (1 - z)2.
Exercise 1.3. Show that (JE., x >-+ x 3 ) is a chart on lE..
Exercise 1.4. (Cylinder).

Consider the cylinder C in JE. 3 given by C = S 1 xlE. = { (x, y, z) E JE. 3 : x 2 + y 2 = 1}.
Let Ux+ := {(x,y,z) EC: x > O}, 'Px+(x,y,z) = (y,z) . Show that (Ux+,'Px+)
is a chart on C. Similar charts (Ux-,'Px-), (Uy+,'Py+) and (Uy-,'Py-) can be
defined analogously, so that the union of Ux+,Ux- , Uy+,Uy- contains C.
A collection of charts for M (with the same m) will form an atlas provided
they cover the set M and satisfy a compatibility condition.
Definition 1.3. (Atlas) .

Let M be a set. A collection of m-charts {(Ui, 'Pi) : i E I} on M is called
an m-atlas if it has the following properties:
(Al) U Ui = M.
ie l
(A2) For all i,j E /, 'Pi(Ui n Uj) is open in JRm.
(A3) For all i, j E /, 'Pj o <p; 1 : 'Pi (Ui n Uj) ----> 'Pj (Ui n Uj) and its inverse
are C 00 . (The maps <pj o <p; 1 are called chart transition maps.)
Smooth Manifolds 5
- 1
'Pi O '{}j (Ui n Uj) is a
bijection. Recall that a function f : !Rm --> !Rn is C 00 if the components Ji
of f , 1 ~ i ~ n, have all partial derivatives of all orders with respect to the
variables xJ, 1 ~ j ~ m, which are also continuous.
The single charts in Examples 1.2, 1.3 and Exercise 1.3 are all atlases
in a trivial manner.
Example 1.5. (S 2 revisited) .
The charts (Un , 'Pn) and (Us , 'Ps) from Example 1.4 form an atlas for S 2.
Firstly, UnuUs = S 2 . Secondly, 'Pn(UnnUs) = IR 2 \ {(0, 0)} = 'Ps(UnnUs)-
Finally, IR 2\ {(0, 0)} , the chart transition
maps, are both
(u ,v)
(u , v),..... 2 2'
u +v
which is C 00 • ◊
Exercise 1.5. Show that the four charts in Exercise 1.4 form an atlas for the
cylinder C in JE. 3 .
Example 1.6. (A non-atlas). With m = 1 in Example 1.2, we get the

chart (IR,<pi), with the chart map <p 1(x) := x for x ER On the other
hand, in Exercise 1.3, we had seen that (JR, <p 2 := (x,..... x 3)) is yet another
chart on R While they individually form atlases A 1 := {(JR, <pi)} and
A2 := {(IR,<p2 )} , their union A:= A1 uA2 = {(IR,<p1), (IR,<p2 )} does not
form an atlas. Indeed, not all chart transition maps are smooth. Although
'{)2 o <p 11 = (x ,..... x 3) is smooth on JR, '{)1 o <p 21 = (x ,..... x 113) is not C 00
everywhere on JR since it is not differentiable at x = 0. ◊
The previous example motivates the following definition.

Definition 1.4. (Compatible atlases) .
Let M be a set. Two m-atlases A1, A2 are compatible if A1 u A2 is also an
m-atlas on M.
Example 1. 7. (Affine space revisited). It is clear from the chart map defi-
nition given in Example 1.3, that different choices of points p, and of bases
{e 1, • •• , em} , will lead to different coordinates. Consider points p , p' E M,
and bases B= {e1 , · · · , em}, B' = {eL · · · , e~,J for V, giving the chart maps
r.p,r.p'. Are the atlases {(M,r.p)} and {(M,r.p')} compatible? To investigate
this, we compute the chart transition map r.p' o r.p- 1 : ~m -+ ~m.
M
---+
r.p
g )
p' e'
~ 1 r.p'
---+
2
Fig. 1.3 Change of coordinates.
Given x E ~m, q := r.p- 1 (x) EM, and we wish to find the coordinates x' of
this q using the point p' and the basis B'. So we need to write q = p' +vp'q,
and find x' by expanding v p' q using the basis B'. We have v p' q = v p' P + v pq,
and as q = r.p- 1 (x), also Vpq = xiei. Introduce the change of basis matrix
A = [A{] E GL(~, m), where A; denotes the entry in the l h row and i th
column of A, defined by ei = A{ e1. Also, let b = (b1, • • • , bm) E ~m be
defined by vp'p = bJe1. Then
vp'q = vp'p + Vpq = l?e1+ xiei = bJe 1+ xi(A{e1)=(bl+ A{xi)e1,
and so the chart transition map r.p' o r.p- 1
is the affine linear map given by
r.p' o r.p- 1 (x) = b + Ax, for all x E ~m , which is C 00 . Its inverse r.p o (r.p')- 1 is
given by r.p o (r.p')- 1(x) = -A- 1b + A- 1x, x E ~m. So the atlases {(M, r.p)}
and {(M, r.p')} are compatible. ◊
Smooth Manifolds 7
Exercise 1.6. Show that the set Rm x GL(m, R) is a group with the composition
(h2,A2) · (h1,A1) = (h2+A2h1,A2A1) for (h2,A2), (h1,A1) in Rm x GL(m,R).
Exercise 1. 7. Prove that compatibility is an equivalence relation on the collection

of all atlases on a set M.
Thus to specify a manifold1, we should work with compatible atlases (so

that in hindsight, there will be no conflict when we define smooth objects
in the manifold), and given that compatibility is an equivalence relation,
we just need to commit to one particular atlas for the set M at hand. We
make this a definition.
Definition 1.5. (Smooth manifold, dimension of a manifold).

A smooth manifold (M, [A]) is a set M together with an equivalence class
[A] of compatible m-atlases on M. We call m the dimension of the smooth
manifold M. A chart from an atlas in [A] is said to be admissible for the
smooth manifold. We refer to [A] as a smooth structure on M.
Thus the atlases given in Examples 1.2, 1.3, 1.4 and Exercise 1.4 can be
used to make the respective set M into a smooth manifold. As we will
use it frequently, we will call !Rm with the atlas comprising the single chart
(!Rm, id) as the smooth manifold !Rm with the standard smooth structure.
In Example 1.6, the two atlases A 1 and A 2 are not compatible. Hence
(JR, [A1]) and (JR, [A2]) are two different smooth manifolds.
Example 1.8. (Sphere). Consider the sphere as a smooth manifold with

the smooth structure given by the atlas in Example 1.5. In this example,
we give a different compatible atlas, using one of the charts as the familiar
one with spherical polar coordinates. It can be shown that the map
(0,n) x (0,2n) 3 (0,¢) >-+ ((sin0)(cos¢), (sin0)(sin¢),cos0) E 8 2 c JR 3
is injective, and hence a bijection onto its image
U = 8 2\{(x,y,z) E JR3 : y = 0 and x ~ 0}.
The angle 0(p) is called the polar angle of the point p E 8 2 , and the angle
cp(p) is called the azimuthal angle of the point p E 8 2 . For a point p E U,
we have cp(p) := (0(p), cp(p)) E (0, n) x (0, 2n), where if p = (x, y, z), then
1 The name 'manifold' comes from the German word 'mannigfaltigkeit' used by Riemann
in his doctoral thesis, which contained, among other things, a discussion of multi-valued
complex functions and their (now called) Riemann-surfaces.
l:~-:;~:k : ::::
0(p) = cos- 1 z and
</>(p) := 2\'..(x, y) :=
21r-cos- 1 ~ ify<0.
yx2+y2
Here (-1, 1) -> (0, 1r) and
cos- 1 : (-1 , 1) -> ( - ~ , ~) are the inverse
sin- 1 :
trigonometric functions. It can be checked that </> is well-defined and that
the map JR 2 \{(x, y) E JR 2 : x;;,, O} 3 (x,y) >-+ 4-(x, y) is C 00 . Using this, it
can be checked that (U, <.p) is an admissible chart: e.g., if v > 0, then
2 2
(u, v) cpocp;; cos- 1 u 2 +v 2 -1 , cos- 1 ~ u )
1
(
I.Pn(Un n U) 3 >------>
u +v +1 vu2+v2
is C 00 , and
(U 11 Un) 3 ( 0, </>) cp~ 1 ( (sin0) cos</> , (sin0) sin</>)
<.p 1- cos0 1- cos0
is C 00 .
z
The chart U covers 8 2 except for a 'slit', namely the intersection of 8 2 with
the half plane {(x, y, z) : y = 0, x;;,, 0} . In order to cover 8 2 , we can take
another chart (V, '¢), defined in a similar manner, by taking a differently
placed slit, in a plane perpendicular to the one containing the original slit.
Then V together with U, covers 8 2 . More explicitly, V covers 8 2 except
for the intersection of 8 2 with the half plane {(x , y, z) : z = 0, x,;;; 0}. Vis
the image of the map
(0, 1r) x (0, 21r) 3 (0, </>) >-+ (-(sin 0)(cos </>), cos 0, (sin 0)(sin </>)) E 8 2 c JR3 ,
and this map is obtained by taking the polar angle with the positive y-axis,
and the azimuthal angle with the negative x-axis counterclockwise in the
y = 0 plane; see the picture above. ◊
Smooth Manifolds 9
Exercise 1.8. Consider the square S := {(x,y) ER2 : lxl+IYI = 1}. Show that the
collection A:= {(U+, 'P+ ), (U-, cp-), (V+, "P+ ), (V-, 'I{'-)} is an atlas for S, where
U+:={(x,y)ES:x>O} cp+(x,y)=y,
U-=={(x,y)eS:x<O} cp-(x,y)=y,
V+:={(x,y) ES: y>O} "P+(x,y)=x,
V-=={(x,y) ES: y<O} "P-(x,y)=x.
Thus (S, [A]) is a smooth manifold. So a smooth manifold may not necessarily
'appear' smooth.
We now discuss four 'spacetimes' by just looking at the underlying smooth

structure. Later on, our examples below will made 'Lorentzian' manifolds,
that is, smooth manifolds with some added structure. While we aren't yet
ready to specify the added structure, we nevertheless introduce these by
only describing their smooth structures. Thus these are 'pre-' spacetimes
for now, ripe for becoming legitimate spacetimes later.
Example 1.9. (Minkowski spacetime). Let V be a 4-dimensional real vec-

tor space. Suppose that M is an affine space over V. Take any p E M,
and a basis {e 1 , e 2 , e 3 , e 4 } for V, and let cp: M-+ JR4 be the corresponding
chart as in Example 1.3. Let A:= {(M, cp)}. Then (M, [A]) is a smooth
manifold, referred to as the Minkowski spacetime. ◊
Example 1.10. (Cylindrical spacetime). Let M = JR x 8 1 with the atlas

A comprising the charts defined in Exercise 1.4. Then (M, [A]) is a smooth
manifold, referred to as the cylindrical spacetime. ◊
Example 1.11. (FLRW spacetime). Let I:= (0, oo). Then M :=Ix JR3 ,
with the atlas {(Ix JR3 , idixR.3)} is a 4-dimensional smooth manifold, called
the FLRW spacetime (after Friedman-Lemaitre-Robertson-Walker). ◊
Exercise 1.9. (Product of smooth manifolds). Let M be an m-dimensional smooth

manifold, with an atlas AM = {(Ui, 'Pi), i E /}. Let N be an n-dimensional smooth
manifold, with an atlas AN= {(VJ,"Pi),j E J}. Define for i E J, j E J, the maps
'Pix "Pi : Ui x Vj -+ Rm+n by (cpi x "Pi)(p, q) = (cpi(P), "Pi(q)) for all p E Ui, q E VJ·
Show that {(Ui x VJ,'Pi x "Pi),i E J,j E J} is an atlas for M x N, making it an
(m + n )-dimensional smooth manifold.
Example 1.12. (Schwarzschild spacetime). Let m > 0, and J := (2m, oo ).

Let M = JR x I x 8 2 , where 8 2 is the unit sphere in JR3 • Taking the at-
lases {(JR, idR.) }, {(I, id1) }, and {(U, cp ), (V, '¢)} (Example 1.8), for JR, I, 8 2 ,
respectively, we see that M is a smooth manifold using the construction
based on Exercise 1.9. We call this 4-dimensional smooth manifold the
8chwarzschild spacetime. ◊
1.2 Topology on a smooth manifold
We will want to talk about continuous maps between smooth manifolds, for
example a 'worldline' in a spacetime M (Definition 1.8). The way we equip
a smooth manifold with a topology is by insisting that the chart maps are
homeomorphisms (Theorem 1.2). This is the motivation for the following
definition.
Definition 1.6. (Open set in a smooth manifold).
Let (M,[A]) be an m-dimensional smooth manifold and {(Ui, rpi), iEI}E [A].
A set Uc Mis open if for all i EI, rpi(U n Ui) is open in !Rm, where !Rm is
given its standard Euclidean topology, described by the Euclidean metric
m
d(x,y) := ~(xi-yi) 2 x=(x1,··· ,xm), y=(y1,··· ,ym)E!Rm.
Proposition 1.1. Definition 1.6 of an open set is well-defined, that is, it

does not depend on the choice of the atlas in [A].
Proof. Let A1 = {(Ui, rpi), i E I} and A2 = {(½,"Pi), j E J} be atlases in
[A]. Let Uc M, and suppose for each i EI, Ai := rpi(U n Ui) is open in
!Rm. Let j E J. We must show that "Pi(U n ½) is open in !Rm. We have
"Pi(U n ½) ='l/Ji ((U nM) n ½) ='l/Ji(U n (LJUi) n ½) ='l/Ji(LJ (U nUin ½))
i i
=LJ "Pi(U nUin ½)- (1.1)

Set Bi := rpi(Ui n ½)- Then Bi is open, since the charts (Ui, rpi) and(½, "Pi)
belong to the atlas A 1 u A 2 . The intersection of this open Bi with the open
set Ai = rpi(U n Ui), is open. Now Ai n Bi= rpi(U n Ui n ½)- (Indeed,::::> is
trivially true, and c follows from the injectivity of of <pi on Ui.) Consider
the C 00 (and in particular, continuous) map rpi o '¢7 1 : "Pi(½) -+ rpi(Ui)-
As the open set Ai n Bi = <pi (U n Ui n ½) is contained in the open set
rpi(Ui) c !Rm, it follows that (rpi o '¢;1 )- 1 (Ai n Bi) is an open subset of the
open set "Pi(½) c !Rm, that is,
(rpi o '¢7 1 )- 1 (Ai "Bi) ="Pio rp-; 1 (rpi(u" ui "½ )) = 'l/Ji(u" ui "½)
is open in !Rm. So "Pi (Un½) (:fl LJ "Pi (Un Ui n ½) is open in !Rm.
i □
We show that calling such sets 'open' is justified, as they form a topology
on the manifold.
Theorem 1.1. Let (M, [A]) be an m-dimensional smooth manifold. Then
the collection O :={Uc M: U is open in M} is a topology on M.
Smooth Manifolds 11
Proof. Let {(Ui,'Pi),i EI} E [A]. Then 0 = 'Pi(0 n Ui) is open in !Rm
for all i EI, and so 0 E 0. Also, for all i EI, 'Pi(Ui) = 'Pi(M n Ui) is open
in !Rm since (Ui, 'Pi) is a chart, and so ME 0.
Let U, V E O. Then for all i E I, 'Pi ((Un V) n Ui) = 'Pi (Un Ui) n 'Pi (V n Ui)
( c is always true for any map, and ::::, holds by the injectivity of 'Pi). Being
the intersection of open sets, 'Pi((U n V) n Ui) is open in !Rm for all i EI,
and consequently, U n V E O.
Let ½EO for all j E J. Then we have that for all i EI,
'Pi(( LJ ½) n Ui) = LJ 'Pi(½ n Ui),

i i
is open in !Rm, as it is the union of open sets 'Pi(½ n Ui) in !Rm. Hence,
U½ E 0.
i □
Definition 1.7. (Topology induced by a smooth structure).

Let (M, [A]) be an m-dimensional smooth manifold. Then the collection
0 := {U c M : U is open (Def. 1.6) in M} is called the topology induced
on M by the smooth structure [A].
Remark 1.1. Often in the literature, a smooth manifold is defined by first

defining a 'topological manifold', where one starts with a topological space
which can be covered by charts which are homeomorphisms to open subsets
of !Rm. We have not adopted this route, since such an approach forces one
to begin with a topology. But we now reconcile our definition with this
prevalent one in the following result. *
Theorem 1.2.
Let (M, [A]) be an m-dimensional smooth manifold, and let
0 := {U c M : U is open in M}
be the topology induced on M by the smooth structure [A]. Suppose that
{(Ui, 'Pi), i E I} E [A]. Then for each i EI, 'Pi : Ui -+ 'Pi(Ui) is a homeo-
morphism.
Proof. Let i E J. As (Ui, 'Pi) is a chart, we know that 'Pi(Ui) is open in

!Rm and that 'Pi : Ui -+ 'Pi(Ui) is a bijection. We only need to show the
continuity of 'Pi and of <p-; 1 . Let V c 'Pi (Ui) be open. Then cp-; 1 V c Ui.
We must show that this is an open set in M. For any j E I, we have
'Pi ( (cp-; 1 V) n Ui) = ('Pio cp-; 1 )(V n 'Pi(Ui n Ui )). As V and 'Pi(Ui n Ui) are
open in !Rm, so is their intersection. Thus ('Pi ocp-; 1 )(V n 'Pi(Ui n Ui) ), being
the inverse image under the (C 00 and hence) continuous map (cp j o cpi 1 )- 1
of the open set V n 'Pi(Ui n Uj) (c 'Pi(Ui)), is open. Hence cpj((cpi 1 V) n Uj)
is open for all j E I, that is, cpi 1 V is open in M. So 'Pi : Ui -+ 'Pi(Ui) is
continuous.
Let U c Ui be open. We want to show that 'Pi(U) = (cpi 1 )- 1 U is
open in !Rm (and hence open in 'Pi(Ui)). The fact that U is open means in
particular that 'Pi(U n Ui) is open in !Rm. But 'Pi(U n Ui) = 'Pi(U), since
Uc Ui. Thus the inverse map cpi 1 : 'Pi(Ui)-+ Ui is also continuous. □
Exercise 1.10. Let Rm be equipped with the standard smooth structure. Show
that the topology induced by this smooth structure coincides with the standard
Euclidean topology.
Exercise 1.11. Consider the double cone

C = {(x,y,z) E R 3 : x 2 + y 2 = z 2 } c R 3 .
Show that C cannot carry a smooth structure [A] making it a smooth manifold
such that the topology induced by [A] on C coincides with the subspace topology
on C (as a subset of R 3 with its standard Euclidean topology). We remark that
if we delete the point O = (0, 0, 0) from C, i.e., we consider C* := C\{O}, then we
do get a smooth manifold, for example by taking an atlas comprising two charts,
namely ({(x,y,z) E c*: z > 0},71") and ({(x,y,z) E c*: z < 0},71"), where the
chart map 71" in each case is just the restriction of the projection to the xy-plane:
R 2 3 (x,y,z) >-+ (x,y) E R 2 .
Remark 1.2. (Hausdorff and second countable assumptions on 0).

In order to do analysis, it is desirable to have two additional properties
enjoyed by the topology 0:
(H) A topology O on a set M is Hausdorff if for every p, q E M, there
exist U, V E O such that p E U, q E V, and U n V = 0- Thus
distinct points possess disjoint neighbourhoods, a type of 'separa-
tion axiom'. Such a property is quite basic, since otherwise limits
of sequences are not guaranteed to be unique.
(S) A basis for O is a collection B = {Bi : i E I} of open sets such that
every open set in O is a union of elements from B. A topology 0
on a set M is second countable if there exists a countable basis for
0. When wanting to do 'integration' on manifolds, this property
will be needed in order to construct a so-called 'partition of unity',
which will essentially mean that we can use m-charts to set up
Riemann integrals of functions defined on the manifold, and patch
these contributions to obtain an integral of the function defined on
the whole manifold.
Smooth Manifolds 13
Unfortunately, for a smooth manifold, neither of these properties are guar-

anteed to hold for the topology O from Theorem 1. 1. So, in order to proceed
without pitfalls, we will make a standing assumption that whenever we talk
of a smooth manifold in this book, we will mean in addition that the associ-
ated topology O is Hausdorff and second countable. The standard topology
of the Euclidean space !Rm generated by the 2-norm II· I satisfies the second
countability assumption since the open balls with centers all of whose com-
ponents are rational numbers, and whose radius is also a rational number,
form a countable basis. Now, if the manifold can be covered by an atlas
in the smooth structure containing countably many charts, then it follows
that (since the chart maps are homeomorphisms) the images of members
of the countable basis for !Rm under the inverse of the chart maps will form
a countable basis for the topology of the manifold. All the examples of
smooth manifolds considered in this book will be of this type. *
Exercise 1.12. Let Ube an open subset of a smooth manifold M given by an atlas
A. Let Au:= {(Un V,'1/Jlunv): (V,'1/;) EA}. Show that Au is an atlas for U.
Prove that if (W,u) is admissible for M, then (Un W,ulunw) is admissible for
(U, [Au]). U is then said to be given the smooth structure induced by (M, [A]).
In particular, if (U,cp) is an admissible chart for M, then [Au]= [{(U,cp)}].
As a spacetime M is the collection of all events, the life of a particle can be
modelled by a curve in M by stringing together all the events encountered
by the particle in its lifetime.
Definition 1.8. (Curve/Worldline).
Let I c JR be an interval and M be a smooth manifold. A continuous map
'Y : I -+ M is called a curve or a worldline.
1.3 Smooth maps
The point of the definition of a smooth manifold is to enable the consider-

ation of smooth objects on it, for example, a real-valued smooth function
(think of temperature), a 'vector field', etc.
Definition 1.9. (Smooth map).
Let M, N be smooth manifolds, with dimensions m, n, respectively. A map
f: M-+ N is said to be smooth if for all p EM,
• there exists an admissible chart (U, cp) for M such that p E U,
• there exists an admissible chart (V, '¢) for N such that J(U) c V (in
particular f (p) E V),
• 'Ip of o cp- 1 : cp(U) -+ !Rn is C 00 on cp(U) c !Rm.
If M is a smooth manifold, and JR has the standard smooth structure,

then we use the notation C 00 (M) to denote the set of all smooth maps
f:M-+R
For a smooth manifold M, the identity map idM : M -+ M is smooth.
Example 1.13. (Chart maps are smooth).

Let (U, cp) be a chart from an atlas defining the smooth manifold M. We
now consider U itself to be a smooth manifold, described by the trivial atlas
{(U,cp)}. Then cp(U) c !Rm is an open subset of!Rm. We consider cp(U)
as a smooth manifold described by the atlas comprising the single chart
(cp(U), idcp(U)). We claim that the chart map cp : U -+ cp(U) is smooth. For
each p EU, we take the admissible chart (U, cp) for U containing p, and the
admissible chart (V := cp(U), id,.,cu)) for the smooth manifold cp(U). Then
cp(U) = V. Moreover, idcp(U) o cp o cp- 1 = idcp(U) : cp(U) -+ cp(U) c !Rm, which
is clearly 0 00 • As p E U was arbitrary, cp : U -+ cp(U) is smooth. ◊
Exercise 1.13. Let M, N be smooth manifolds, and f: M-+ N be a smooth map.

Show that f is continuous.
Exercise 1.14. Let M1, M2, M3 be smooth manifolds, and let fi2 : M1 -+ M2,
h3 : M2 -+ M3 be smooth maps. Prove that h3 o fi2 : M1 -+ M3 is smooth.
Exercise 1.15. Let M, N be smooth manifolds, and M x N be the smooth manifold

described in Exercise 1.9. Let the projection map 7rM : M x N -+ M be given by
M x N 3 (p, q) >-+ p E M. Given a q EN, let the injection map iq : M -+ M x N
be given by M 3 p >-+ (p,q) EM x N.
• Show that 7rM is smooth. (Similarly, M x N 3 (p, q) >-+ q E N is smooth.)
• Show that iq is smooth. (Also, for p EM, N 3 q >-+ (p, q) EM x N is smooth.)
In particular, Exercise 1.15 has the following consequences. Firstly, given

any g E C 00 (M), the map M x N 3 (p, q) i--+ g(p) E JR, is an element
of C 00 (M x N), as it is the composition of the smooth maps g and 7rM-
Secondly, given an f E C 00 (M x N) and a q EN, the 'slice map' fq, given
by M 3 pi--+ f (p, q) E JR is smooth too, since f q = f o iq. We will use these
observations later on to show that the 'tangent space of M x N at (p, q)'
can be identified with TpM x TqN in Exercise 2.8.
Exercise 1.16. (Smoothness is a local property). Let M, N be smooth manifolds

and f : M -+ N be a map. Show that f is smooth if and only for every U open
in M, equipped with the smooth structure induced from M, Jiu : U -+ N is
smooth. Here U is given the induced smooth structure obtained from the smooth
structure of M.
Smooth Manifolds 15
Introduce the operations+,• : C 00 (M) x C 00 (M)-> C 00 (M), both defined

pointwise:
(J+g)(p) = f(p)+g(p)} for all p EM
(J-g)(p) = f(p)·g(p) .
Then it can be checked that indeed f + g, f •g E C 00 (M), and that with
these operations, (C 00 (M), +, •) forms a ring. However, C 00 (M) is not a
field, since not every nonzero C 00 (M) function 2 will have a multiplicative
inverse. Later on, we will see that the set of 'smooth vector fields' on a
manifold has the natural structure of a module over this ring C 00 (M).
We will meet geodesics later on, which will the 'straightest' possible
curves in the Lorentzian manifold, describing paths of 'freely falling' parti-
cles. The straight lines in Euclidean space and great circles on the sphere
S 2 are geodesics. In any case, they are 'smooth' curves.
Definition 1.10. (Smooth curve).

A smooth map 'Y : I -> M, where I is an open interval in R, is called a
smooth curve. If I c R is any interval, not necessarily open, then a curve
'Y : I -> M is a smooth curve if there exists an open interval J ::J I, and a
smooth curve 1': I-> M such that ::Yl1 = 'Y·
Just like in linear algebra, where one aim is to classify vector spaces up to
isomorphisms, in differential geometry, the notion analogous to an isomor-
phism is that of a diffeomorphism.
Definition 1.11. (Diffeomorphism).
Let M, N be smooth manifolds. A bijection f : M -> N such that f and
1- 1 : N-> Mare both smooth, is called a diffeomorphism, and Mand N
are said to be diffeomorphic.
Example 1.14. (Chart maps are diffeomorphisms).

Let (U, cp) be a chart from an atlas defining the smooth manifold M, and
consider U as a smooth manifold with the atlas {(U, cp)}. We had seen that
the chart map cp: U-> cp(U) is smooth. Also, it is a bijection onto the open
set cp(U). We show that its inverse cp- 1 : cp(U) -> U is smooth too. For all
cp(p) E cp(U), with p EU, we take the admissible chart (cp(U),id 40 (u)) for
cp(U) containing cp(p), and take the admissible chart (U, cp) for U, so that
cp- 1 (cp(U)) = U. Moreover, cp o cp- 1 o (id 40 (u))- 1 = idl 40 (u) : cp(U)-> cp(U),
which is clearly 0 00 • As this happens with each point in cp(U), the map
cp- 1 : cp(U) -> U is smooth. ◊
2 Consider a nonzero smooth function that has a zero at some point. In fact, in Chap-
ter 2, we'll construct nonzero functions that vanish outside a neighbourhood of a point.
Exercise 1.17. Let R be equipped with the two incompatible atlases A1 and A2
given in Example 1.6. Prove that (R, [A1]) is diffeomorphic to (R, [A2]). (From
our earlier considerations, the incompatibility of A1 with A2 can be expressed
by saying that the identity map fails to be a diffeomorphism between the smooth
manifolds (R, [A1]) and (R, [A2]). However, this exercise shows that there may
nevertheless be other maps which serve as a diffeomorphism.)
Exercise 1.18. Let M be an affine space over V, considered as a smooth manifold

in the usual way. For av EV, define 'Yv: R-+ M by ,v(t) = p + tv, t ER. Show
that 'Yv is a smooth curve.
Exercise 1.19. Let M be a smooth manifold. Show that the set

Diff( M) := {J : M -+ MI f is a diffeomorphism},
together with the operation o of composition of maps, forms a group.
Exercise 1.20. (Lie group and left translation diffeomorphisms). A Lie group is
a group (G, •) equipped with a smooth structure, such that the multiplication
map G x G 3 (p, q) >-+ p • q E G, and the inverse map G 3 q >-+ q- 1 E G, are
smooth. Given p E G, the left-translation by pis the map Lp : G-+ G defined by
G 3 q >-+ p • q. Show that Lp is a diffeomorphism for each p E G.
Before moving on, we make a final remark on some notation used. Charts
will often be denoted by (U, cp), but also by (U, x), where the understanding
is that the component functions of the map x : U -> !Rm are denoted by
xi : U -> JR, 1 ~ i ~ m. Moreover, given a function f : M -> JR, a point
p EM, and an admissible chart (U, cp), we will denote the partial derivative
off o cp- 1 : cp(U) -> JR with respect to the ith variable at the point cp(p) by
!(!u• 0 cp-1 \cp(p)).

Also, for a smooth manifold M, we will often take for granted that its
dimension is denoted by m.
Chapter 2
Co-/Tangent space
Intuitively, a tangent space at a point p on a surface in IR. 3 is the plane at p

tangential to the surface, consisting of all tangent vectors. Tangent vectors
are the 'velocities' of curves passing through p. Thus we imagine a picture
like this:
Fig. 2.1 Tangent plane TpM and tangent vector vat a point p on the surface M .
However, when we have only an abstract manifold at hand, this is no longer

an adequate definition, since there is no 'natural' 1 embedding in JR.d_ We
ought to keep the definitions 'intrinsic', not relying on any embedding, and
using just the manifold structure. The way around the obstacle is that we
begin by revisiting the familiar notion above, and write it in a way which
lends itself to an appropriate generalisation.
It will turn out that the 'right' way to think about tangent vectors at
a point p E M is as linear maps v : C 00 (M) ---> IR. that obey a Leibniz rule,
given by v(J-g) = f(p)v(g) + g(p)v(J) for all f,g E C 00 (M).
1 There exists a result of Whitney, saying that a smooth manifolds of dimension n can
be embedded in ffit 2 n, but being a purely mathematical construct, it is devoid of any
immediate physical relevance.
17
This may appear strange, but it is not too far-fetched. Indeed, imagine
what a child running in a playground experiences as its own velocity vector
in spacetime: The child perceives the changes in (the smooth functions
on spacetime such as) temperature, pressure and so on. The faster these
change, the larger the perceived velocity. Thus if 'Y: (-e:, e:) -+ M describes
the child's worldline, with 7(0) = p, then it makes sense to think of its
velocity at a point p E M as a map
f,..... :;f O 'Y) (0), f E C 00 (M),
since the above expression registers how fast f (temperature, pressure, etc)
is changing at p, as the child goes about playing!
2.1 Tangent vectors to a surface as derivations
A (parametrised) surface2 M in R 3 is a map (u, v) ,..... r( u, v) : D -+ R3

such that the 'normal vecto~
or or
n := ou x ov -I= 0 everywhere m D,
.
where the parameter domain D is an open set in R 2 . Here x is the usual
cross product of vectors in R 3 . Let r( u 0 , v0 ) = p E M. Our multivariable-
calculus understanding of a tangent vector v at p is that there is curve
(-e:,e:) 3 t ~ r(u(t),v(t)) passing through p=-y(0), such that we have
-y(0) := !fil-(0) =v. We note that
-y(0) = it(0) :: (uo, vo) + v(0) :: (uo, vo)- (2.1)

We note that several different curves may have the same tangent vector, so
it is not appropriate to associate a tangent vector with a special curve. The
tangent space TpM is the set of all tangent vectors, built by considering all
curves passing through p, and taking the corresponding tangent vectors to
these curves at p. But we see from the above expression, that by changing
the maps t,..... u(t), v(t), we can get all possible coefficients it(0) and v(0) in
the linear combination on the the right hand side of (2.1) above, and so
TpM = span { :: (uo, vo), :: (uo, vo)} .

We visualise TpM as a plane placed at p, tangential to the surface, and
having the normal
or or
np := ou (uo, vo) x ov (uo, vo)-
2See e.g. [Apostol(1969), Chap. 12].
Co-/Tangent space 19
As mentioned in the introductory remarks to this chapter, when working

in an abstract manifold, using such a definition of a tangent vector and
tangent space is not feasible. We will now see3 that the correct notion is
that of a 'derivation'.
A derivation v at p EM is a linear map v: C 00 (M)--> IR which satisfies
the Leibniz rule v(J-g) = f(p)v(g) + g(p)v(J) for all f,g E C 00 (M). In our
setting of surfaces, let us first show that every tangent vector gives rise to a
derivation. So suppose that we have a tangent vector arising from a curve
'Y: (-1:,1:)--> M such that -y(0)=p. We define
d(Jo'Y)
v,,..,,pf := dt (0), f E C 00 (M).
Then for CE JR, f, g E C 00 (M), we have
v (f+ ) = d((J+g) O "f) (0) = d(J O "f) (0) + d(g O "f) (0)
'"Y,P g dt dt dt
= v,,..,,p(J) + v,,..,,p(g),
v
(c·J)=d((c-J)o'Y)(O)=cd(Jo'Y\o)=cv (J)
'"Y,P dt dt '"Y,P ,
showing that v,,..,,P is linear. It also satisfies the Leibniz rule since
v (f·) = d((J-g)o'Y)(O) = d((Jo'Y)·(go'Y))(O)
'"Y,P g dt dt
=(JO "f)(0) :;9 0 'Y) (0) + (g O "f)(0) :? O "f) (0)
= f(p)v,,..,,p(g) + g(p)v,,..,,p(J).
Now let us show that every derivation arises from a tangent vector to some
curve. To do this we will first prove the following 'division' lemma.
Lemma 2.1. Let B be an open ball in !Rm centered at 0. Suppose that
f E C 00 (B) and f(O) = 0. Then there exist 91, · · · ,9m E C 00 (B) such that
f=x 1g1+···+xmgm in U.
Proof. For x EB, 1,:;;; k,,;; m, with 9k(x) := Jofl of

oxk (tx)dt,
f 1 of f 1 of d(txk) f1 d
Xk9k = Jo oxk(tx)xkdt = Jo oxk(tx)~dt = Jo dt(J(tx))dt
= f(x) - f(O) = f (x).

□
3 As this section is just meant to provide some motivation for our definition to follow
in the next section, we will allow some sloppiness in our arguments.
Either from the formula for 9k in the proof, or upon differentiating both
sides off= x 1 g1 + · · · + xmgm with respect to Xk, we see that
of
9k(O) = oxk (0), 1 ~ k ~ m.
We will also need the following two lemmas saying that derivations annihi-
late constant functions and that they are 'local'.
Lemma 2.2. Let M be a surface, v be a derivation at p E M, and let

f E C 00 (M) be constant. Then vf = 0.
Proof. Let 1 E C 00 (M) be the function taking value 1 everywhere. If

the function f assumes the constant value c, then f = c-1. We have that
v(l) = v(l-1) = 1 v(l) + 1 v(l) = 2 v(l), giving v(l) = 0. Consequently,
v(J) = v(cl) = cv(l) = c0 = 0. □
Lemma 2.3. Let M be a surface, and v be a derivation at p E M. If

f,g E C 00 (M) agree in a neighbourhood ofp, then vf = vg.
Proof. Suppose that f, g coincide in U. Let x be a 'bump function',

namely a function which is identically equal to 1 in a neighbourhood of p,
and is identically zero outside a neighbourhood V such that that the closure
V of Vis contained in U. That such a function exists is plausible, but we
postpone the proof until Section 2.3. Thus (J - g) • x = 0 on M, giving
0=v((J - g) · x) = (vf -vg)x(p) + (J(p) - g(p))vx= (vf - vg) 1 + 0. □
Let 4 v be a derivation at p EM. We justify below that there exists an f > 0
and a smooth curve -y: (-1:, t::) -> M such that -y(0) = p and v-y,p = v.
The map M 3 (x, y, z) = q ~ x E R is smooth, and so by acting
with v, we get a number vx. Similarly, we define vy and vz. This gives
a vector (vx, vy, vz) E R 3 . We first show that this vector belongs to the
span of the vectors g~ (
uo, vo) and g~ (
uo, vo). Suppose p = r( uo, vo). For
a point (M 3) q = r(u,v), we write (u,v) = (rp 1 (q),rp 2 (q)). Moreover, let
r(u, v) =: (X(u, v), Y(u, v), Z(u, v)). By an application of Lemma 2.1, we
can write X(u, v)-X(uo, vo) = (u-uo)e(u, v) + (v-vo)r,(u, v), for smooth
e
functions and r, that satisfy
1 2
e(rp (p),rp (p)) =e(uo,vo)=~(uo,vo)
ax and
1 2
r,( <p (p ), rp (p)) = r,( uo, vo) =
ax (uo, vo)-
a;
4 We use the notation v for the derivation here, instead of v, in order to avoid confusion
with the parameter v from (u, v) E D.
We have
x(q) - x(p) = x(r(u, v)) - x(r(uo, vo)) = X(u, v) - X(uo, vo)
= (u - uo)e(u, v) + (v - vo)'Tl(u, v)
= (,/(q) - ,/(p))e(,/(q), ,p2(q))
+ (,p2(q) - ,p2(p)),,,(,p1(q), ,p2(q)).
Operating by v on both sides, we obtain, using the Leibniz rule, that
vx - 0 = v(q,..... ,p1(q))e(,p1(p), ,p2(p))
+ (,p1(p) - ,p1(p))v(q I-+ e(,p1(q), ,p2(q)))
+v(q,..... ,p2(q))'Tl(,p1(p), ,p2(p))
+ (,p2(p) - ,p2(p))v(q,..... 'T/(</(q), ,p2(q)))
ax
= v(q,..... ,p1(q)) ou (uo, vo)
ax
+ 0 + v(q,..... ,p2(q)) a:;-(uo, vo) + 0.
With similar computations carried out for vy and vz, we get
ax ax
ou (uo, vo) a;(uo, vo)
[ ::] = ~: (uo, vo) ~: (uo, vo) [ ~] = [ !: (uo, vo) !: (uo, vo)] [ ~] ,

vz az az
ou (uo,vo) ov (uo,vo)
where U := v(q,..... cp 1 (q)) and V := v(q,..... cp 2 (q)). Since the normal vector
np is nonzero, the above U, V are the unique. Recall that (u 0 , v 0 ) E D is
such that r(uo,vo)=p. Define u(t)=uo+ Ut, v(t)=vo+ Vt fort E (-1:,1:),
where the 1: > 0 is taken small enough so that (u(t),v(t)) lies in a ball
centered at (u 0 ,v0 ) contained in D. Define -y(t) = r(u(t),v(t)), t E (-1:,1:).
Then
. (0)= dr(uo+Ut,vo+Vt)(O)= or( )U or( )V= [vxl

-Y dt ou uo,vo + ov uo,vo vy .
vz
Let Do c D be an open ball containing (uo, vo) and 8 > 0 be such that
the 'cylinder' n = {q + t np : q E r(Do), ltl < 8} is an open set in l~.3,
that is, the surface patch obtained from Do can be fattened by displacing
the surface along np. If f E C 00 (M), then we may extend f as a constant
in the direction of normal llp by setting J(q + t np) := J(q), q E r(D 0 ).
Then f E 0 00 (0). The translated set {x: x + p E O} is open and contains
the origin 0, and we can find a small ball B contained within it. Then for
x EB, define cp by cp(x) = f(x+p)-f(p). Then cp(O) = 0. By Lemma 2.1,

there are 9k, k = 1, 2, 3, such that cp = xkgk on Band
of ocp
axe (p) = axe (0) = ge(O).
For qEr(Do), f(q) = f(q-p+p) =cp(q-p)+f(p) = (qk-pk)gk(q-p)+f(p),
and so
vf = v(f(p) + (qk - pk)gk(q - p))
= v(f(p)) + (pk - pk)v(gk(· - p)) + 9k(P - p) (v(qk) - 0)
of of of
= 0 + 0 + vx ox (p) + vy oy (p) + vz oz (p). (2.2)
On the other hand,
v f = d(J o 7) (o) = [ of (p) of (p) of (p)] i'(0)

-r,P dt ox oy oz
of of of
= vx ox (p) + vy oy (p) + vz oz (p).
Consequently, vf = V-y,pf for all f E C 00 (M), that is, V = V-y,p·

Thus in the context of surfaces, we have realised that the classical view
of tangent vectors v at a point p E M as being the velocity vector of some
curve '"Y at p can be replaced by the view that tangent vectors are simply
'derivations', namely linear maps v : C 00 (M) -+ JR that obey the Leibniz
rule. This completes our discussion of the motivation for defining tangent
vectors as derivations in the abstract setting of smooth manifolds.
2.2 Tangent vector definition
Definition 2.1. (Tangent vector, tangent space).

Let M be a smooth manifold and p E M. A tangent vector v at p is a map
V: C 00 (M)-+ JR such that for all f,g E C 00 (M) and CE JR,
• (linear) v(J + cg) = v(J) + cv(g), and

• (Leibniz rule) v(J · g) = f(p)v(g) + g(p)v(J).
The set of all tangent vectors at p is the tangent space TpM at p.
Exercise 2.1. Show that TpM is a vector space with addition and scalar multi-
plication defined pointwise. (Later on, in Theorem 2.1, we will show that the
dimension of the vector space TpM is equal to the dimension of the manifold M.)
Exercise 2.2. Let v E TpM. Prove that if f E C 00 (M) is constant, then vf = 0.

Example 2.1. (Tangent vectors to curves as vectors).

Let I c JR be an open interval and 'Y : J --> M be a smooth curve such that
"f(0) = p EM. Define v7 ,p : C 00 (M) -> JR by
= d(J o 'Y) (0)
V7,p f dt . (2.3)
We claim that v7 ,p E TpM. Indeed, for f, g E C 00 (M) and c E JR, we have
+ ) = d((J + cg) o 'Y) (0) = d(J o 'Y + c(g o 'Y)) (0)
v'Y ,P (! cg dt dt
d(Jo"() d(go"()
= dt (0) + c dt (0) = v7 ,p(J) + cv7 ,p(g), and
V (!. ) = d((J. g) 0 'Y) (0) = d((J O 'Y)(g O 'Y)) (0)
'Y,P g dt dt
= (Jo 'Y)(0) :;g O 'Y) (0) +(go 'Y)(0) :;f O 'Y) (0)
= f(p)v7,p(g) + g(p)v7,p(J).
◊
Exercise 2.3. (Different curves with a common tangent vector).

Consider IR as a smooth manifold with the standard smooth structure. For t E IR,
define 71(t) = t - 1, and 72(t) = t + 3t2 - 1. Show that v"Y 1 ,-1 = v"Y2 ,-1.
Exercise 2.4. Let M be a smooth manifold. Let I, J be open subsets of IR, and
J 3 u >-+ h(u) E J be a C 00 function. Let 1' : J -+ IR be a smooth curve. Show
that the tangent vectors of the smooth curve 1' o h : J -+ M satisfy
V"Yoh,("Yoh)(s) = h(s)v"Y,"Y(h(s)) E T"Y(h(s))M, for all SE J,
where h(s) := ~!(s).
2.3 Bump functions
A technical tool which will prove to be handy in the sequel is 'bump-function

technology'. In a nutshell, it is a result on the existence of smooth functions
which are identically 1 in the neighbourhood of a point and vanish outside
a bigger neighbourhood. This will be very useful when we want to extend
smooth objects beyond a chart in a smooth manner, or while patching stuff
together on a manifold having obtained contributions on charts.
The following exercise will be used in the construction of a bump func-
tion x around a point pin a smooth manifold M (i.e., a C 00 (M) function
x which is identically 1 in a neighbourhood U of p and is zero outside a
somewhat bigger neighbourhood V ::::> U).
Exercise 2.5. Let

f(t) = { e-l/t for t >
0 fort~0
0} .
We want to show that f E C 00 (JR).
(1) Let g: JR-+ JR be continuous on JR, continuously differentiable on JR. :=JR\{0},
and such that limg'(t) exists. Show g is continuously differentiable on R
t--+O
(2) Suppose n E N. Let g: JR -+ JR be continuously differentiable n-1 times on JR,

and n times on JR., and such lim/nl(t) exists. Prove that g is n times
t--+O
continuously differentiable on R
(3) Show that f is infinitely many times differentiable.
Hint: Using induction on n, show that fort> 0, JCn) =Rnf, where Rn is a
rational function. Conclude that lim en J(t) = 0.
t'-,0
Lemma 2.4. (Existence of a bump function).

Let U c JR.m be an open set, and let O E U. Then given any R > 0 such that
B(O, R) c U, there exists a pointwise-nonnegative x E C 00 (U) such that x
vanishes outside B(O, R) and is identically 1 on a ball B(O, r) centered at
0 with radius r < R.
Proof. By composing the function f E C 00 (JR.) constructed in Exercise 2.5

with t >-+ 1 - t 2 , we get a function cp E C 00 (JR.) that vanishes outside [-1, 1].
See the left picture in the first figure below. With
'lj;(x) := J~ 00
cp(t)dt,
we have 'lj; E C 00 (JR.), 'lj; is O on (-oo, -1), and is a constant on (1, oo ).
-1 0 1 -1 0 1
By scaling and shifting, we get a function depicted in the top left-hand side
of the following figure, and by reflecting, the one in the bottom left-hand
side. Their pointwise product yields the bump function a in one variable,
shown on the right in the following picture.
-3
I
/ I
-1
I' I I'.\
I'.\.
1 3
-3
I I
-1 1 3
Defining F(x) = cr(x 1 ) • • • cr(xm) , we see that F = 1 on B(O , 1), and F = 0

outside B(O, 3-Jm) (because if (x 1 ) 2 + ••• + (xm) 2 ~ 32 m, then it cannot
be the case that each xi satisfies lxil < 3, and so for some i, lxil ~ 3, giving
o-(xi) = 0). Using suitable 'dilation' , namely x = F(>.-), with)..:= 3-Jm/R,
we have xis identically 1 on a ball B(O , r) with radius r < R, and is zero
in U outside B(O, R). □
Corollary 2.1. (Existence of a bump function). Let U c M be an open

set, and let p E U. Then there exists a pointwise-nonnegative f E C 00 ( M)
such that f is identically equal to 1 in a neighbourhood of p, and is zero
outside a neighbourhood V such that V c U.
Proof. Let (W, I.{)) be an admissible chart map such that p E W c U,
and also assume, without loss of generality, that l.()(p) = 0. Let R > 0
be such that B(O, R) c l.{)(W). By the previous Lemma 2.4, there exists
ax E C 00 (rntm) such that x = 1 in a ball B(O, r) c l.{)(W), where r < R,
and x = 0 outside the bigger ball B(O, R). Define V = 1.()- 1 B(O, R). As
the chart map I.{) is a homeomorphism, V = 1.()-l B(O, R) c W c U. Define
f = x o I.{) on W, and identically zero outside it. D
Exercise 2.6. Consider the smooth manifold IR 2 with the standard smooth struc-
ture. Construct a smooth curve 'Y : IR---+ IR 2 such that "!(0) = (0, 0), the range of
'Y is as shown below, and is traversed in the directions indicated by the arrows.
(0 ,0) X
2.4 Chart-induced basis for the tangent space
We will now show the following crucial result, which will allow tangent
vectors to act on smooth functions defined 'locally'.
Lemma 2.5. Let M be a smooth manifold, and pEM. Let vETpM. Then
if f,g E C 00 (M) agree in a neighbourhood ofp, we have vf = vg.
Proof. Without loss of generality, we may assume that f, g coincide in

an admissible chart (U, cp) for M with p EU. By Corollary 2.1, there exists
a bump function x which is identically 1 in a neighbourhood of p, and zero
outside a neighbourhood V of p such that V c U. Then (J - g)·x = 0 on
M, and so
0 = v((J - g) · x) = (v(J) - v(g)) 1 + (J(p) - g(p))vx
= v(J) - v(g) + Ovx = v(J) - v(g).
This completes the proof. □
Remark 2.1. (Extension of the definition of tangent vectors).
The above lemma can be used to extend the action of tangent vectors
v E TpM to functions f E C 00 (U), where U is any open set in M containing
the point p. Indeed, if x E C 00 (M) is a bump function which is zero outside
a neighbourhood V of p with V c U and is identically 1 in a neighbourhood
of p, then we define Fx to be x· fin U and O outside U. Then Fx E C 00 (M),
and we set v(J) := v(Fx)- We remark that the definition makes sense, for
if a is a different bump function with the same properties, then since a • f
matches with x • f in a neighbourhood of p (both being equal to !), it
follows from Lemma 2.5 that v(Fx) = v(Fa ). *
Definition 2.2. (Chart-induced tangent vectors).
Let M be a smooth manifold of dimension m, and (U, x) be a chart in a
defining atlas for M. Let p E U, and let xi be the coordinate functions,
that is x = (x1, • • • ,xm). Define Oxi,p E TpM by
o(Jlu o x- 1 )
Oxi ,vf := au•. (x(p)), f E C 00 (M).
The tangent vectors Ox1,p, · · · , Ox"',P are called the chart-induced basis tan-
gent vectors by the chart (U, x).
It is clear that Oxi,p E TpM: It is linear, since for f, g E C 00 (M) and c E JR,
1 )( (p))
::i.
ux, p(f +cg )=o((J+cg)ox-
::i . x
' uui
o(J O x- 1 ) o(g O x- 1 )
= 0ui. (x(p))+c Oui. (x(p))=(oxipf)+c(oxipg).
, ,
Theorem 2.1. Oxi,p, · · ·, Ox=,p form a basis for TpM.
Thus, the vector-space-dimension of TpM matches the smooth-manifold-

dimension of M. This makes sense, since in the case of surfaces in JR3 ,
the tangent plane at a point on the surface locally looks like the surface
around that point. In the case of spacetime M, we will later see that at
an event p E M, the instantaneous 3-dimensional space we see around, is
infinitesimally the orthogonal complement (with respect to the 'Lorentzian
scalar product g(p)' in the 4-dimensional TpM), to the tangent vector v7 ,P
to our worldline 'Y at p.
Proof. We show that {oxi,p, · · ·, Dx=,p} is independent and spans TpM.
(Independence) Let x be a bump function aroundp. Set f = xix E C 00 (M).
Then
Oxi f = :~xix) o x-1) (x(p)) = x(p) :~i o x-1) (x(p))+xi(p) :~ o x-1) (x(p))
oui . .
= 1· oui (x(p))+x'(p) · 0=6;.
If doxi,p = 0, then acting on f = xix yields Ci= 0, showing independence.
(Spanning) Let f E C 00 (M). Choose an r > 0 so that B(x(p), r) c x(U).
Define h E C 00 (B(O, r)) by h(u) = (Jox- 1 )(u+x(p))- f(p). Then h(O) =0.
By Lemma 2.1, h = uigi, for some 9i E C 00 (B(O,r)) satisfying
oh o(f o x- 1 )
9i(O) = ,i""""'(O) = 0 .
uui ui
(x(p)) = ox, vf.
'
In the neighbourhood x- 1 (B(x(p), r)) of p, f(·) = h(x(• )- x(p))+ f(p). So

vf = v(h(x(•) - x(p)) + f(p)) = v(h(x(•) - x(p))) + v(J(p))
= v(h(x(•) - x(p))) + 0 = v((x\) - xi(p))gi(x(•) - x(p)))
= (xi(p) - xi(p))v(gi(x(•) - x(p))) + 9i(x(p) - x(p))v(xi(·) - x\p))
= 0 · v(gi(x(•) - x(p))) + 9i(O) · v(xi(·) - xi(p))
= (oxi,pf)(v(xi) - v(xi(p))) = (ox;,pf)(v(xi) - 0) = (v(xi)ox;,p)f.
So v = v(xi)Oxi,p- Consequently, Ox1,p, · · · , Ox=,p span TpM. □
Example 2.2. (TpM ~ V for an affine space Mover V).

Let M be an affine space over V. For p E M and v E V, define the
smooth curve 1'v(t) = p + tv, t E R (Exercise 1.18). Then V-yv,P E TpM.
Let {e 1 , • • • , em} be a basis for V. Let v = viei. The chart defined in
Example 1.3, induces a basis {oxi,p, · · · , ox"',p} for TpM. Let us find the
coordinates of v"Yv,P with respect to this basis. From the end of the proof
of Theorem 2.1, V-yv,P = V-yv,P(xi)Oxi,p- Thus the ith component is
i) = d(xi o '}'v) (0) = d(tvi) (0) = i
V-yv ,p (X dt dt V,
which coincides with the components of v with respect to the basis
{ e 1 , • • • , em} for V. Consider the isomorphism I : V -> TpM that sends the
basis vector ei to the basis vector Oxi,p for all 1 ~ i ~ m. Then viei = v is
sent to ViOxi,p = V-yv,P· Thus our isomorphism I is the map V f-+ V-yv,P· ◊
Example 2.3. (Physicist's definition of a vector).
In the Physics literature sometimes, a vector, based at a point p in an
m-dimensional manifold M, is defined as an 'abstract object' with the fol-
lowing behaviour specified for its components: When one commits to an
admissible chart/coordinate system (U, x) on M, the abstract object pro-
duces an m-tuple (v 1 , • • • , vm), and if one has two admissible coordinate
systems containing p, say (U, x) and (U', x'), then the resulting m-tuples
are related by
Ii = OXli (i:) j
V :i . <,, V,
uxJ
where ~x'~ (e) is an abbreviation for the partial derivative

uxJ
:(~'i
uJ
o x-1) (x(p))
of the ith component of chart transition map (sometimes called 'change

of coordinates' in the literature) x' o x- 1 : x(U n U') -> x'(U n U') with
e
respect to the jth variable, at the point := x(p). We show below that this
way of thinking coincides with our notion of a vector. Thus, starting with
a tangent vector v E TpM, and using its representations obtained via the
bases {Oxi,p, · · · , Ox"',p} and {Ox,1,p, · · · , Ox'"',p}, we will establish the above
transformation rule for the components of v. Indeed, we have
v'i = v(x'i) = viOxi,p(x'i) = vi:~~,i ox-l) (x(p)),
as required.
Having chosen a basis, if we represent the vector as a column vector of

its components with respect to the basis, then the above transformation
expression can be rewritten as matrix multiplication as follows:
o(x11 o x-l) (x(p)) o(x11 O X-l) (x(p))
ou 1 oum
o(x'm O x- 1\x(p)) ... o(x'm O x- 1\x(p))

ou 1 oum
In particular, since Oxi,p = 8f Oxk,p, it follows that
k ox'i ) ox'i
Oxi,p = ( 8i oxk (~') Ox i ,p = oxi (e) Ox i ,p·
1 1
Swapping the roles of x and x', ox,i,p = ;;: (e')oxi,p, where f =x'(p). ◊
Lemma 2.6. (Vectors as the tangent vectors of curves).
Let v E TpM. Then there exists an 1:. > 0 and a smooth curve 'Y : (-1:., 1:. )-+ M
such that 7(0) = p and v'Y,P = v.
Proof. Let (U, x) be an admissible chart containing p. If Oxi,p, 1 ~ i ~ m,
denote the chart-induced tangent vectors, then we can write v = viOxi,p for
some numbers vi. Let 1:. > 0 be small enough so that x(p) + tviei E x(U)
for ltl < 1:., where e1, · · · , em are the standard basis vectors in !Rm. Set
7(t) = x- 1(x(p) + tviei), t E (-1:., 1:.). Then 7(0) = x- 1(x(p)) = p, and for
all f E C 00 (M),
v f = d(J O 'Y) (0) = d(J(x- 1(x(p) + tviei))) (0)
'Y,P dt dt
o(Jox-1) i i
= au•. (x(p))v = V Oxi ,p/ = vf.
Consequently, v'Y,P = v. □
In particular, from the above, Oxi,P is the velocity vector v'Y;,P of a locally
defined curve 'Yi passing through p = 'Yi(0): 'Yi(t) = x- 1(x(p) + tei), for
ltl < f. for a small enough 1:. > 0.
Example 2.4. (Tangent vectors to curves decomposed using chart-induced
basis). Let 'Y : J -+ M be a smooth curve, where J is an open interval, and
suppose that 7(J) c U, where (U, x) is an admissible chart for M. We
claim that
Indeed, for all f E C oc, (M), we have

d(fo-y)() d(fox- 1 oxo-y)()
V--y ,y(t) f = dt t = dt t
= o(fox- 1 )( ( ( )))d(xo-y)i() = d(xo-y)i( )o. f
oui X 'Y t dt t dt t x' ,--y(t) '
and so we see that the components of v--y ,--y(t) with respect to the chart-
induced tangent vectors are just the derivatives of components of the chart
representation of the curve. ◊
Exercise 2.7. Consider the smooth manifold IR 2 with the standard smooth struc-
ture. Consider the admissible chart (V, v,), where V = {(x,y): y > O}, vi= (r,0),
r(x,y) = ,Jx 2 +y 2 and0(x,y)=cos- 1 (x/,Jx 2 +y 2 ) . Letp=(x,y)EV. Express
Or,p and 00 ,p in terms of Ox ,p and Oy ,p•
Exercise 2.8. Let M x N be the product of the smooth manifolds M, N, and let
(p, q) E M X N. For VE TpM and w E TqN, define V EB w : C 00 (M X N) -+ IR by
(v EB w)f = v(J(- , q)) + w(J(p, ·)) for all f E C 00 (M x N). Show that v EB w belongs
to T(p ,q)(MxN). Prove that the map TpMxTqN 3 (v, w) >-+ vEBw E T(p ,q)(MxN)
is linear, injective, and surjective. Thus T(p ,q) ( M x N) is isomorphic to the direct
sum of TpM and TqN, written as T(p ,q)(MxN) "'TpMxTqN .
2.5 Derivatives of smooth maps
If 'Y : JR. --> M is a smooth curve, and f : M --> N is a smooth map, then
f o 'Y : JR.--> N is a smooth curve. For p = -y(0) E M and g E Coc,(N),
V = d(g O (f O 'Y)) (0) = d( (g O f) 0 'Y)) (0) = V ( 0 /).
f o--y,f(p)9 dt dt --Y,P g
Hence f : M --> N induces a mapping sending tangent vectors at p E M to
tangent vectors at f(p) EN. This motivates the following definition.
f
---+
Definition 2.3. (Derivative of a smooth map).

Let M, N be smooth manifolds, and f : M --> N be a smooth map. If
p E TpM, then define dfv: TvM--> Tf(p)N by (dfv(v))(g) = v(g o !), for all
g E Coc,(N) and v E TpM. The map dfv is called the derivative off at p.
For v E TpM, dfp(v) is a tangent vector at f(p) EN: For all g, h E C 00 (N)
and c Ell,
(dfp(v))(g+ch) = v((g+ch) of) =v(g of +c(h of)) =v(g of)+cv(h of)
= (dfp(v))(g) + c(dfp(v))(h), and
(dfp(v))(g-h) = v((g-h) o J) = v((g o J)-(h o J))
= (go J)(p)v(h o J) +(ho J)(p)v(g o J)
= g(J(p))(dfp(v))(h) + h(J(p))(dfp(v))(g).
Exercise 2.9. Prove that dfP : TpM -+ Tt(p)N is a linear map.
Theorem 2.2. Let M, N be smooth manifolds, f : M -+ N be a smooth

map, and p EM. Let (U, x) be an admissible chart for M containing p, and
let (V, y) be an admissible chart for N containing f (p). Then the matrix
for the linear transformation dfp: TpM-+ Tf(p)N with respect to the bases
{ox;,p, 1 ~ i ~ m} and {oyi,f(p), 1 ~ j ~ n} is given by the Jacobian matrix
of yo f o x- 1 at the point x(p).
Proof. For g E C 00 (N), we have

o(g Of O x- 1 )
(dfp(oxi,p))g = Oxi,p(g o J) = oui (x(p))
= :(~oy- 1 oyofox- 1\x(p))

u•
= :(~ 0 y-1 \y(J(p))) :(~of o x- 1 )i (x(p))
vJ u•
o(y Of O x- 1 )i
= oui (x(p))oyi,f(p)9·
o(y Of O x- 1 )i
Thus dfp(Oxi,p) = oui (x(p))oyi,f(p)· □
Exercise 2.10. Let 'Y : R -+ M be a smooth curve passing through the point
p = 7(0). Let Ot,o denote the tangent vector at t = 0 E R induced by the chart
(R, id). Let v'Y,P be defined by (2.3). Show that v'Y,P = d7o(Ot,o).
Exercise 2.11. Consider R 2 as a smooth manifold with the standard smooth struc-
ture. Fort ER, let '¢t: R 2 -+ R 2 be given by (x,y) >-+ (x-yt,y). Show that
(d'¢t)p(oy,p) = -tox,,J,,(p) + oy,,J,,(p)•
Exercise 2.12. Let M1, M2, M3 be smooth manifolds. Suppose that f: M1 -+ M2
and g: M2-+ M3 are smooth maps. Let p E M1. Show d(g o f)p = dgf(p) o dfp-
Exercise 2.13. Let f: M-+ N be a diffeomorphism between the smooth manifolds
Mand N. Let p EM. Show that dfp: TpM-+ Tt(p)N is an isomorphism.
Conclude that M and N have the same dimension.
We end this section with a useful result, namely the inverse function the-
orem, in the smooth manifold setting. First we give a C 00 version in the
Euclidean setting, as a consequence of the 'usual' continuously differentiable
version; see for example [Rudin(1983), Thm. 9.24, p. 221].
Proposition 2.1. (Euclidean+C00-inverse function theorem).
Let U, V c !Rm be open. Let f: U-+ V be C 00 , and f'(p) be invertible at
p E U. Then there exist neighbourhoods Uo c U of p and Vo c V of f(p)
such that fluo : Uo-+ Vo is a diffeomorphism.
Proof. We already know from [Rudin(1983), Thm. 9.24, p. 221] that

there are the claimed neighbourhoods such that 1- 1 : Vo-+ U0 is in C 1 on
V0 . As f oJ- 1 = idv0 , we obtain by the chain rule [Rudin(1983), Thm. 9.15,
p. 214] that ru- u-
1 (x)) · 1 )'(x) = idlR"' for all x E Vo, and so ru- 1 (x))
is an invertible matrix, and u- 1 )'(x) = (f'(J- 1 (x)))- 1 . By Cramer's rule
[Artin(1991), Thm. 5.7, p. 29], the (i,j)th entry of the matrix for (J- 1 )'(x)
is given by a polynomial combination of the partials of f, divided by the
nonzero determinant of the Jacobian matrix off, and as these are C 00 , it
follows that 1- 1 is C 00 • □
Theorem 2.3. (Smooth manifold-inverse function theorem).

Let M, N be smooth manifolds. Let f : M -+ N be smooth, and dfP be
invertible at a point p E M. Then there exist neighbourhoods Uo of p, and
Vo of J(p), such that flu0 : Uo-+ Vo is a diffeomorphism.
Proof. As dfP is bijective, M and N have the same dimension, say m.
Since f is smooth, there exist admissible charts (U, cp) for M and (V, '1/J)
for N, such that the point p E U, J(U) c V, and j = '1/J of o cp- 1 : fJ-+ V
is C 00 , where fJ = cp(U) and V = '1/J(V) are the open sets in !Rm. As cp, 'ljJ
are diffeomorphisms, dh(P) = d'I/Jt(P) o dfPod( cp- 1 )rp(p) is invertible. Now by
Proposition 2.1, there exists a neighbourhood U0 c fJ containing cp(p) and
a neighbourhood Vo c V containing '1/J(/(p)) such that flu : Uo -+
~ Vo is
a diffeomorphism. Then U0 := cp- u0 c U is a neighbourhood of p, and
0
1
Vo
Vo := 'ljJ- 1 c Vis a neighbourhood of f(p). Moreover,
-1 ~
flu" = '1/J O flu" 0 cp,
that is, flu0 : Uo -+ Vo is the composition of the diffeomorphism fluo with
the diffeomorphisms 'ljJ- 1 and ¢. So f lu0 : Uo-+ Vo is a diffeomorphism. □
Remark 2.2. We now give the justification of a technical fact, as an ap-
plication of the Euclidean C 00 inverse function theorem. This fact will be
used later on in Remark 7.1.
Fact: Let M be a smooth manifold, I be an open interval containing 0,

and 'Y : J---+ M be a smooth curve such that v-y,-y(O) -I= 0.
Then there is an open interval J c I containing 0, and admissible chart
(U, x) containing 'Y(J), such that (x o 'Y)(t) = (t, 0, • • • , 0) for all t E J.
Proof: Let (V, y) be an admissible chart containing p := 'Y(0). Then
So there exists an index i such that
Without loss of generality, let i = 1. By the inverse function theorem applied

to the smooth map "( 1 = (y o 'Y) 1 : J---+ ll, there exists an open interval Jc I
containing 0 and an open interval K c ll containing y 1 (p), such that the
smooth map (yo 'Y) 1 : J---+ K is a bijection with a smooth inverse r,: K---+ J.
Let ff be the open set ff= V n (y 1 )- 1 K, which is nonempty asp E ff. As
y : V ---+ y(V) is a diffeomorphism, the nonempty set y(ff) is open. Now
for f3 = (/31, • • • , 13m) E y(ff), we have /3 1 EK. Define g: y(ff)---+ llm by
g(/3) = (r,(/31),/32 _ 'Y2(r,(/31)), ... , 13m _ 'Ym(r,(/31)))
for all /3 = (/3 1 , · · · , 13m) E y(ff). We have that
g'(/3) = [r,'(/31) 0 ]
* Im-1
which is invertible in y(ff), and in particular at y(p). By the inverse func-

tion theorem, there exists a neighbourhood W c y(ff) of y(p) such that
g : W ---+ g(W) is a diffeomorphism. Set U = y- 1 W c ff c V, and
x = g o y : U ---+ llm. This is a diffeomorphism onto its image. As 0 E J,
and 'Y(0) = p E U, it follows by the continuity of 'Y that we can choose an
open interval J c J containing 0 such that 'Y(t) E U for all t E J. For
t E J c J, we have that
(x o 'Y)(t) = (go yo 'Y)(t)
= (r,('Yl(t)),"f2(t) _ "f2(r,('Yl(t))), ... ,"fm(t) _ "fm(r,('Yl(t))))
= (t, "(2 (t) - "( 2 (t), · · · , "fm(t) - "(m(t)) = (t, 0, · · · , 0).
This completes the proof of the claimed fact.

*
2.6 Cotangent space
In this section, we will consider the dual space (TpM)* of TpM consisting
of linear maps w : TpM -+ R Elements of (TpM)* will be called cotangent
vectors or l-forms 5 • Later on, we will also learn that the instantaneous
momentum is a 1-form. This can be intuitively understood as follows: we
know that force is the rate of change of momentum, and so as a geometric
object, the nature of the force is the same as that of momentum. We know
that work, which is a scalar, is obtained by taking the 'line integral' of the
force. But when doing so, we take the dot product of the force with a little
vectorial displacement along the curve, namely a tangent vector. So the
force/momentum acts on tangent vectors and produces scalars.
We will see later on that tangent spaces together with their duals (called
cotangent spaces) at a point p E M can be used to build the space of
tensors at the point p, and when we consider a varying p, one gets a map
pi--+ (tensor at p), which is referred to as a tensor field. The field equations
of spacetime will involve such tensor fields. The modest aim in this section,
though, is to introduce the cotangent space as the dual of the tangent space.
Definition 2.4. (1-forms, cotangent space).

The dual space (TpM)* ofTpM is called the cotangent space to Mat p EM.
Each element of TpM is called a 1-form or a cotangent vector.
Example 2.5. (Gradient of a function). Let jEC00 (M) and pEM. Define
the 1-form dfp E (TpM)* by dfp(v) = v(J) for all v E TpM. For v, w E TpM
and c E JR, dfp(v+cw) = (v+cw)(J) = v(J) + cw(!) = dfp(v) + cdfp(w).
The 1-form dfp is called the gradient off at p. ◊
Example 2.6. (Gradient revisited). Let JR be given the standard smooth

structure [{(JR, u i--+ u)}]. If q E JR, then TqJR can be identified with JR,
by mapping w = w(u) ou,q to w(u). Given an f E C 00 (M), we have the
derivative dfp : TpM-+ Tf(p)JR of the smooth map f (Definition 2.3). But
by identification of Tf(p)JR with JR, we may consider the derivative dfpv
acting on v E TpM, as a real number, rather than a vector in Tf(p)R In
fact this number is the action dfp(v) of the 1-form dfp on v, because if the
vector dfp(v) = cou,f(p), then c = (dfp(v))(u i--+ u) = v(id o J) = v(J).
So it makes sense to use the same notation dfp(v), both for the action
of derivative dfp on v (which is a vector in Tf(p)JR), and the action of the
1-form dfp on v (which is a real number). ◊
5 This terminology will become clear after we study more general k-forms in Chapter 10.
Definition 2.5. (Chart-induced I-forms).

Let M be a smooth manifold of dimension m, and (U, x) be a chart in a
defining atlas for M. Let p E U, and let xi be the coordinate functions,
that is, x = (x 1 , • • • , xm). Define (dx 1 )p, • • • (dxm)p E (TpM)* by
(dxi)p(v) = v(xi) for all v E TpM.
The I-forms (dx 1 )p, • • • , (dxm)p are called the chart-induced I-forms by the
chart (U, x).
Theorem 2.4. (dx 1 )p, • • • , (dxm)p form a basis for (TpM)*.
Proof. In fact, {(dx 1 )v,··· ,(dxm)p} is the dual basis to the chart-
induced basis {ilx1,p, · · · , ilx"',p}:
· · o(xi O x- 1 ) oui ·
(dx')p(Oxi,p) = Oxi,pX' = oui (x(p)) = oui (x(p)) = 8}.
If cidxi = 0, then by acting on Oxi,p, we obtain Ci = 8}ci = 0. Thus
(dx 1 )p, • • • (dxm)p are independent. Also, for w E (TpM)* and v E TpM, we
can write v = vi Oxi ,p, and so
(w - w(ox;,p)(dxi)p) (v) = (w - w(ox;,p)(dxi)v) (vi Oxi,p)
= viw(Oxi,p) - w(ox;,p)vi8J = 0,
showing that w = w(ox;,p)(dxi)p- Hence the I-forms (dx 1 )p, • • • (dxm)p span
(TvM)*. □
In particular, if f E C 00 (M), and (U, x) is an admissible chart for M con-

taining p, then the proof of Theorem 2.4 shows that the components of the
I-form dfp, with respect to the basis {(dx 1 )p, • • • , (dxm)p} for (TpM)*, are
given by the m-tuple (oxi,vf, · · · , Dx=,p/).
Exercise 2.14. (Physicist's definition of a 'covector').
In the Physics literature sometimes, a covector, based at a point p in an m-
dimensional manifold M, is defined as an 'abstract object' with the following
behaviour specified for its components: When one commits to an admissible
chart/coordinate system (U, x) on M, the abstract object produces an m-tuple
(w1, • • • , wm), and if one has two admissible coordinate systems, say (U, x) and
(U', x'), containing the point p, then the resulting m-tuples are related by
/ 0Xj / 0Xj / O(Xj OXl-l) / / /
W; = ~,· (e
ux •
)wj, where~,· (0 := 0 .
ux • u•
e
(x (p)) and := x (p).
The aim of this exercise is to show that this way of thinking coincides with our
notion of a 1-form. Thus, starting with a 1-form w E (TpM)*, and using its repre-
sentations in the bases {(dx 1 )p, • • • , (dxm)p} and {(dx11 )p, • • • , (dx'm)p}, establish
the above transformation rule for the components of w. We remark that, having
l
chosen a basis, if we represent the 1-form as row vector of its components with
respect to the basis, then the above transformation expression can be rewritten
as matrix multiplication as follows:
a(x 1 o x'- 1 ) (x'(p)) ___ a(x 1 o x'- 1 ) (x'(p))
au 1 au"'
[
[ w~ · · · w~ ] = [ W1 • • • Wm ] : ··. : .
1 1
a(x'"' 0 x'- ) (x'(p)) ... a(x'"' 0 x'- ) (x'(p))
au 1 au"'
In particular, as (dxi)p=8l(dxk)p, we obtain
i ( i 0Xk I ) lj 0Xi I lj
(dx )p = 8k ox'i (e) (dx )p = ox'i (e )(dx )p-
Swapping the roles of X and x', (dx'i)p = ~xii (e)(dxj)p, where

ux1
e:= x(p).
Exercise 2.15. (1-forms on IR.3: Cartesian to spherical coordinates).
Consider the smooth manifold R 3 with the standard smooth structure, and the
admissible chart (U, x), where U := R 3 \{(x, y, z) E R 3 : x~O, y= O}, x= (r, 0, </>),
x(U)=(O,oo) x (O,1r) x (O,21r), and x- 1 :x(U)-+ U is given by
x- 1 (r, 0, </>) = (r(cos </>)(sin 0), r(sin </>)(sin0), r cos 0) for (r, 0, </>) E x(U).
Let p EU be such that x(p) = (r, 0, </>). Prove that
(dx)p = (cos<f>)(sin0)(dr)p + r(cos<f>)(cos0)(d0)p - r(sin</>)(sin0)(d<p)p
(dy)p = (sin</>)(sin0)(dr)p + r(sin</>)(cos0)(d0)p + r(cos<f>)(sin0)(d<f>)p
(dz)p = (cos0)(dr)p - r(sin0)(d0)p-
Exercise 2.16. Let (U, x), (U', x') be admissible charts for a smooth manifold M.
Show that the matrices describing the change of components of tangent vectors
and 1-forms, with respect to the chart-induced bases, are inverses of each other:
If
Exercise 2.17. Let M be a smooth manifold. Let f,g EC 00 (M) and p EM. Prove
the Leibniz rule d(f-g)p = f(p)dgp + g(p)dfp-
2. 7 Pull-back of 1-forms
We have seen that the derivative of a smooth map f : M --+ N between

smooth manifolds M, N can be used to 'push forward' tangent vectors at
a point p EM to tangent vectors at f(p) EN. We now study the 'dual' of
this map, (dfp)* : (TJ(p)N)* --+ (TpM)*, which can be used to 'pull back'
1-forms.
Definition 2.6. (Pull-back of a 1-form).
Let M and N be smooth manifolds, f : M --+ N be a smooth map, and
p EM. Define (dfp)*: (TJ(p)N)*--+ (TpM)* by ((dfp)*w)(v) = w(dfp(v)) for
all w E (TJ(p)N)* and all v E TpM. The 1-form (dfp)*w E (TpM)* on M is
called the pull-back of w under f.
First we note that for v, v E TpM and c Ell,
((dfp)*w)(v + cv) = w(dfp(v + cv)) = w(dfp(v) + cdfp(v))
= w(dfp(v)) + cw(dfp(v))
= ((dfp)*w)(v) + c((dfp)*w)(v),
showing that (dfp)*w E (TpM)*. Moreover, the map (dfp)* is linear, since
if w, 0 E (TJ(p)N)* and c Ell, then for all v E TpM, we have
((dfp)*(w+c0))(v) = (w+c0)(dfp(v)) = w(dfp(v)) + c0(dfp(v))
= ((dfp)*w)(v) + c((dfp)*0)(v)
= ((dfp)*(w) + c(dfp)*(0))(v),
and so
(dfp)*(w+c0) = (dfp)*(w) + c(dfp)*(0).
Thus (dfp)*: (TJ(p)N)*--+ (TpM)* is linear.
Summarising, given a smooth map f : M --+ N,
• tangent vectors can be pushed forward, and

• 1-forms can be pulled back.
Later we will see how these can be used to define the push-forward of vector
fields or pull-back of 1-form fields, from one manifold to another using a
map linking the two manifolds.
Exercise 2.18. Let R, R 2 be equipped with their standard smooth structures, and
let f: R-+ R 2 be the smooth map given by j(t) = (cost,sint), t ER. Consider
the chart (R 2 , (x,y) >-+ (x,y)) on R 2 • Lett ER, q = f(t) E R 2 , and w E (TqR 2 )*
be the 1-form w = -y(q)(dx)q + x(q)(dy)q- Determine (dft)*w E (TtR)*.
Chapter 3
Co- /Tangent bundles
In the previous chapter, we defined the notion of a tangent vector. We now

want to talk about '(smooth) vector fields'. Intuitively, a vector field on a
manifold is thought of as a distribution of tangent vectors on the manifold
so that they change 'smoothly' from point to point. The wind velocity
on the surface of the earth is an example of a vector field. But we wish
to define vector fields 'intrinsically', i.e., built using the abstract manifold,
and moreover, the vector field should be 'smooth' (to do calculus).
The first hurdle, to intrinsically define a vector field, does not seem to be
problematic. We want to associate to each point p E M a tangent vector
VP E TpM. So the vector field must be a map V with domain M. But
what should the target space be? At the point p, we need to pick a tangent
vector in TpM, and not from some other tangent space TqM with q =Ip.
Thus we consider the 'disjoint union' of all the tangent spaces,
TM= LJ TpM = {(p,v): pE Mand v E TpM} = LJ ({p} x TpM).
pEM pEM
We think of this as a bundle of tangent vectors, aptly called 'tangent bun-

dle'. An obvious surjection is the map 7r : TM ---> M given by 1r( (p, v)) = p,
for all (p, v) E TM. Thus every 'vector' in the tangent bundle 'remembers
its base point' p EM. Then a vector field is firstly a map V: M---> TM
such that 7r o V = idM. This is referred to as a 'section' of the tangent
bundle. So the following diagram commutes.
39
In order to talk about smoothness, we would like make TM a smooth

manifold, and then a vector field would be a section from M to TM which
is also a smooth map between these smooth manifolds. Hence we want to
give an atlas for TM, that is, specify a collection of compatible charts on
TM which cover TM. Intuitively, given any point (p,v) E TM, we want
to assign coordinates to it. Now the point p, being in M, has its natural
coordinates cp(p), arising from some chart (U, cp) for M that contains p. But
this also gives natural coordinates to v, since we can write v = v(xi) ox,,p
using the chart-induced basis vectors Ox1,p, • • • , Ox=,p· Hence we can give
(p,v) the 2m coordinates (cp(p),v(x 1 ),··· ,v(xm)). It turns out that such
charts do give an atlas for TM. We will make these intuitive ideas precise
in the next section.
In this chapter, we will also define 1-form fields, that is, smooth maps
M 3 p i--+ Op E (TpM) *, by constructing, analogous to the tangent bundle, a
cotangent bundle TM*. These form the building blocks for tensor fields.
Indeed, at the end of this chapter, with TM and TM* in hand, we will be
able to talk about tensor fields in the next chapter.
3.1 Tangent bundle
Definition 3.1. (Tangent bundle).

Let (M, [A]) be a smooth manifold. Define the set
TM= lJTpM={(p,v):pEMandvETpM}= LJ ({p}xTpM).
pEM pEM
For each chart (U, cp) in an atlas A belonging to [A], define U by

U = {(p, v) : p E U and v E TpM} = LJ ({p} x TpM) ,
pEU
and rp:U-+ cp(U) x !Rm by rp(p,v)=(cp(p),v(x 1 ),··•,v(xm)) E cp(U) x !Rm,

where the component functions of cp are denoted by x 1 , • • • , xm.
We call (U, rp) the lift of the chart (U, cp). Let A be the collection
{(U, rp) : (U, cp) EA}. Define the tangent bundle to be the smooth manifold
(TM, [A]). The map 1r: TM-+ M sending (p, v) E TM top EM, is called
the canonical projection.
Co-/Tangent bundles 41
As the sets U from A cover M, the corresponding sets U from A cover TM.
Also, as cp : U -+ cp(U) is surjective, and as every m-tuple (v 1 , • • • , vm) E !Rm
results in a vector v = vi Oxi,p E TpM, <p is a surjective map. It is also
injective: If 1.p(p,v) = 1.p(q,w), then cp(p) = cp(q), so that p= q, thanks to
the injectivity of cp, and moreover, v = v(xi) Oxi,p = w(xi) Oxi,q = w. Thus
<p : U -+ cp(U) x !Rm is a bijective map onto the open subset cp(U) x !Rm of
JR 2m, and so (U,<p) is a chart for TM.
Moreover, if (U, 1.p), (V, 1/J) are lifts of the charts (U, cp), (V, '¢) for M,
then we have 1.p(U n V) = cp(U n V) x !Rm, which is open in JR2 m.
Let (U, 1.p), (V, 1/J) be lifts of admissible charts (U, cp), (V, '¢) for M. We
will show that the chart transition maps corresponding to (U, <p), (V, 1/J) are
smooth. Besides completing the verification that A is an atlas, this will also
show that the construction of the smooth structure on TM is independent
of the choice of the defining atlas A from the smooth structure [A] on M.
Let (o.,/3) E 1.p(U n V). With /3 = (/31,··· ,/3m), cp = (x1,··· ,xm) and
'¢ = (y 1 , • • • , ym), we have
(1/J O l.{)-l)(o.,/3) = 1/J(cp-la.,/3iOxi,cp-lo.)
= ( ('¢ O cp-l )( a.), /3i Oxi,cp-lo.y1, · · · , /3i Oxi,cp-lo.Ym).
The map cp(U n V) 3 a.>-+('¢ o cp- 1 )(o.) is C 00 . Also, the map
j o(yi ocp- 1 ) -l o('l/Jocp-l)j
a. f-+ Oxi cp-lo.Y = a . (cp(cp a.)) = a . (a.)
' u• u•
is a C 00 function (since the chart transition map'¢ o cp- 1 is C 00 ). It follows
that the map (a., /3) >-+ (('¢ o cp- 1)(a.), /3i Oxi,cp-l(o.)Y1, · · · , /3i Oxi,cp-l(o.)Ym)
is C 00 , as wanted.
Consequently, TM is a smooth manifold of dimension 2m, where m
is the dimension of the smooth manifold M (and also the vector space
dimension of each tangent space TpM).
Exercise 3.1. Let M be a smooth manifold. Show that the canonical projection
TM -+ M is a smooth map.
1r :
Exercise 3.2. Let M, N be smooth manifolds and let f : M -+ N be a smooth

map. Define df: TM-+ TN by df(p,v) = (f(p),dfp(v)) for all (p,v) E TM.
Show that df is smooth.
3.2 Vector fields
Now that we have a tangent bundle with the structure of a smooth manifold,
we can talk about its sections which are smooth, and these are called vector
fields.
Definition 3.2. (Vector field).

Let M be a smooth manifold. A smooth map V : M -+ TM satisfying
1r o V = idM is called a vector field. For p E M, we write V(p) = (p, Vp),
where Vp E TpM. The set of all smooth vector fields on Mis denoted 1 by
TJM.
Definition 3.3. (Vector field action on a smooth function).

Let V be a vector field and f E C 00 ( M). Then the function V f :M -+ JR
is defined by (VJ) (p) = Vp (!) for all p E M.
The following gives useful criteria for checking if a section is smooth.
Theorem 3.1. Let M be a smooth manifold. Suppose that V: M-+ TM
satisfies 1r o V = idM. Then the following are equivalent:
(1) V is smooth.
(2) For every f E C 00 (M), VJ E C 00 (M).
(3) For all charts (U, x) from an atlas defining the smooth manifold M, the
maps given by U 3 p >-+ Vpxi E JR, 1 ,:;;; i ,:;;; m, belong to C 00 (U).
Proof.
(1)=?(2): Let (U, cp) be an admissible chart for M. For o: E cp(U), we have
((VJ) ocp- 1 )(o:) = Vip-taf = Vip-1a(xi)Oxi,ip-1af.
But by virtue of the fact that f is smooth, the following map is smooth:
cp(U) Oxi ,ip-la! = 0u•.
3 0: >-+
au
o cp- 1 )
(a).
Let (U, c,o) be the lift of (U, cp). As V is smooth, it follows that the map
c,o o Vo cp- 1 : cp(U) >-+ JR 2 m is smooth, that is, the map
cp(U) 3 o: >-+ (o:, Vip-ia(x 1 ), · · • , Vip-ia(xm))
is smooth, and in particular, also the projection to any of the last m com-
ponents, o: >-+ Vip-1a(xi), is smooth. So (VJ) o cp- 1 is smooth on cp(U).
But cp : U -+ cp(U) is a diffeomorphism. Hence V flu is smooth for each
admissible chart (U,cp) for M, and by Exercise 1.16, VJ is smooth on M.
(2)=?(3): Let p E U. Let x E C 00 (M) be a bump function which is iden-
tically equal to 1 in a neighbourhood U0 c U of p, and is O outside as
slightly bigger neighbourhood U1 , with U1 c U. Then xxi E C 00 (M), and
so V(xxi) is smooth. In particular, the map given by
cp(Uo) 3 0: >-+ (V(xxi))(cp- 1 0:) = Vip-la(xxi) = Vip-lc,(xi)
1 The somewhat odd notation will become clear when we discuss tensor fields (Chap. 4).
belongs C 00 (cp(U0 )). Thus U 3 p 1--+ Vvxi E JR, 1 ~ i ~ m, belong to C 00 (U).

(3)=?(1): For o: E cp(U), (c,o o Vo cp- 1 )(o:) = (o:, Vrp-ia(x 1 ), · · · , Vrp-ia(xm))
But thanks to the smoothness of U 3 p i--+ Vpxi, it follows that the maps
o: i--+ Vrp-ia(xi), 1 ~ i ~ m, are all smooth. Hence o: i--+ (c,o o Vo cp- 1 )(o:)
is smooth. Consequently, V is smooth. □
We will henceforth identify V(p) = (p, Vp) with Vp itself, and so the vector
field V sends p E M to Vp E TpM.
Clearly, the result of pointwise adding a pair of vector fields is a new
vector field, and so is the result of scaling a vector field pointwise. This
makes the set TJ M of all smooth vector fields into a real vector space.
However, given a vector field V, rather than just producing a uniform
scaling everywhere by a factor c E JR, that is, M 3 p i--+ c Vp E TpM, we
can consider a less restrictive possibility of having the c depend on p in a
smooth manner. Thus, we want to consider not just multiplication by real
scalars, • : JR x TJ M -+ TJ M, but rather, replace JR by C 00 ( M), and have a
'scalar' multiplication with smooth functions,•: C 00 (M) x TJM-+ TJM.
The set C 00 (M) with its pointwise operations is not a field, but only a ring.
We will see that scalar multiplication by elements of C 00 ( M) makes TJ M a
C 00 (M)-module. First we give the definition of multiplication of arbitrary
vector fields by elements of the ring C 00 ( M).
Definition 3.4. (Multiplication of vector fields by smooth functions).
Let VE TJM and f E C 00 (M). Define f· VE TJM by (J-V)(p)=f(p)Vp,
for all p EM.
TJM is a module over the ring C 00 (M). Thus the usual the vector space
axioms with respect to vector addition hold, with vector addition on TJ M
defined pointwise: For V, WE TJ M, V +Wis given by (V + W)p = Vp + Wp,
for all p EM. Let 1 E C 00 (M) denote the constant function taking value 1
everywhere on M. For all f, g E C 00 (M), and all V, WE TJ M,
1-V=V
(f g) · V = f · (g · V)
(f + g) · V = f · V + g · V
f · (V + W) = f · V + f · W
Exercise 3.3. (Vector fields are derivations 2 ). Let M be a smooth manifold. Show
that V(fg) = f V g + g V f for all f, g E C 00 (M) and VE TJ M.
2 A derivation 8 on an algebra A over JR is a linear map 8 : A -+ A satisfying the Leibniz
rule 8(ab) = a8b + bfo for all a, b EA. An algebra A over JR is a real vector space with a
bilinear product • : A x A -+ A.
Example 3.1. (Chart-induced vector fields).

Let (U, x) be an admissible chart for a smooth manifold with the coordinate
functions denoted by xi, 1 ~ i ~ m. Then (U, [A]), where A= { (U, x)}, is
a smooth manifold. By Theorem 3.1, it follows that for all 1 ~ i ~ m, the
section Oxi : U-> TU given by pi--+ Oxi,p, is a vector field Oxi E TJU, since
U 3 p i--+ Oxi,p xi = 8f,
1 ~ j ~ m, are constant functions (and so trivially
smooth). ◊
Lemma 3.1. (Extension of a vector to a vector field).

Let v E TpM. Then there exists a vector field V E TJ M such that Vp = v.
Proof. Let (U, x) be an admissible chart containing p. Let x E C 00 (M)

be a bump function that is 1 in a neighbourhood of p, and identically 0
outside a neighbourhood W with W c U. Write v = viOxi,p, where Oxi.
fi h . b - { vix(q)Oxi,q if q E u,
De net e section V y Vq - 0 E TqM if q E M\U.
Then Vp = vi 1 Oxi ,P = v, and by Theorem 3.1, V is smooth. □
Exercise 3.4. Let IR.2, R be given standard smooth structures. We use the global
charts (R 2 , (x, y) I-+ (x, y)) and (R, u I-+ u). Suppose that V =xyox+x2 oy E TJR 2 '
f =x2 y E C 00 (R 2 ) and p= (1, 1) E R 2 . Determine (JV)p, (V f)(p), dfp(Vp)-
3.3 Push forward of vector fields under diffeomorphisms
We know that if / : M -> N is a smooth map, then a vector v E TpM can

be 'pushed forward' to a vector dfp(v) E TJpN. One might hope then that
a whole vector field V can also be pushed forward using this mechanism for
pushing vectors forward. However, we see that there are two problems:
• The map f : M -> N may not be surjective, and so not every point
q EN might be f(p) for some p EN. As the action of dfp only produces
vectors at Tf(p)N, we cannot define a vector field on N, since a vector
field on N has to be defined everywhere on N.
• The map f : M -> N may not be injective. This gives rise to the
problem that if /(m 1 ) = /(m 2 ), then we have a conflict: should the
evaluation of the vector field at this point be the push forward of Vm 1
or of Vm 2 ? They can be different, as shown in the following example.
Example 3.2. (Non-identical push forwards of vectors under a non-

injective map). Let JR be considered as a smooth manifold with the stan-
dard smooth structure. Let f : JR -> JR be a non-injective, but surjective
function, for example, the cubic f(x) = (x - l)x(x + 1). Note that / is
surjective, but not injective (e.g. /(0) = /(1) = 0). Let V E TJ!R be the
vector field V = u Ou- Then note that Vo = 0 E To!R, and Vi = Ou,1 E T1R
But dfo(Vo) #dfi(V1) since if we take g= [u >-+ u] E C 00 (1R), then we have
(dfo(Vo))g=O, while
d((u - l)u(u + 1)) 2
(dfi (Vi))g = Vi (go J) = du (1) = (3u - l)lu=l = 2. ◊
The problems mentioned above disappear if we have a diffeomorphism f,

and although a bijective smooth map would suffice, we will use this in the
context of the 'flow' of a vector field, where we will have diffeomorhisms.
Definition 3.5. (Push forward of a vector field under a diffeomorphism).

Let f : M -+ N be a diffeomorphism, and V E TJ M be a vector field. Then
we define the push forward f* V of V under f, by(!* V)J(p) =dfp(Vp) for all
pEM.
Exercise 3.5. Let f: M-+ N be a diffeomorphism, and VE TJ M be a vector field.

Show that f* V E TJ N.
Exercise 3.6. (Left-invariant vector field on a Lie group, Lie algebra of a Lie
group). Let G be a Lie group with identity element e E G. For p E G, recall
(from Exercise 1.20) the left translation map, G 3 q >-+ Lpq = p·q E G. A vector
field VE TJG is called left-invariant if (dLp)q Vq = Vpq for all p, q E G. Show that
V E TJG is left-invariant if and only if (LP)* V = V. The set of all left-invariant
vector fields is denoted by g. With pointwise operations, g forms a vector space,
called the Lie algebra3 of G. The aim of this exercise is to show that TeG :,, g
as vector spaces. For a vector v E TeG, define I(v) = V, where V : G-+ TG is
defined by V(p) = (p, (dLp)ev) for all p E G. It can be shown that the section V
is smooth, so that VE TJG. Show that Vis left-invariant. Prove that the linear
map I : TeG-+ g, sending v E TeG to Iv E TJG, is bijective.
Exercise 3. 7. Consider the smooth manifold ~.2 with the standard smooth struc-
ture, and the global chart (l~.2, (x, y) >-+ (x, y)). Fort ER, let "P-t : R 2 -+ R 2 be
the diffeomorphism given by (x, y) >-+ (xcost-ysin t, xsin t+ycos t). Let V = Oy.
Determine ("P-t)*V.
Exercise 3.8. Let M, N be smooth manifolds, and f : M -+ N be a diffeomor-

phism. Let VE TJ Mand cp E C 00 (N). Show that ((f* V) cp) of= V(cp of).
3 The term 'algebra' (rather than 'vector space') is used because it turns out that the
Lie bracket defines a multiplication on g; see Exercise 3.10.
3.4 Lie algebra of vector fields
In this section, we will equip TJ- M with a (anti-commutative) 'multiplica-

tion' operation, called the 'Lie bracket', which assigns to two vector fields
V, W, a new vector field [V, W]. With [·, -], the real vector space TJ-M
forms an algebra over R We will see later on that there is a geometric
motivation4 for the Lie bracket.
If V, W E TJ- M and f E C 00 (M), then W f E C 00 (M), and so for each
p E M, Vp can act on it, yielding a real number Vp (W J). It is clear that
f i--+ Vp(W J) is linear, but it turns out the Leibniz rule fails in general,
preventing this map from being a vector in TpM. However, if we subtract a
corresponding product in the other order, namely Wp (VJ), then the prob-
lematic terms cancel each other, and give rise to a vector in TpM. If we
now allow the point p to vary in M, then we obtain a vector field on M.
We have the following.
Definition 3.6. (Lie bracket).
Let M be a smooth manifold. Define the Lie bracket of V, W E TJ- M to be
the vector field [V, W] E TJ- M given by [V, W]p(J) = Vv(W J) - Wp(V J)
for all f E C 00 (M) and all p EM.
We now show that the Lie bracket of vector fields is indeed a vector field.
Lemma 3.2. If V, W E TJ- M, then [V, W] E TJ- M.
Proof. First we show that at each p E M, [V, W]p E TpM, so that [V, W]
is at least a section, and then show that it is also smooth using Theorem 3.1
(p.42). The map [V, W]p: C 00 (M)-> IR is linear: for f, g E C 00 (M), c E JR,
[V, W]vU+cg) = Vp(W(J+cg))- Wp(V(J+cg))
= Vp(Wf+cWg)- Wv(Vf +cVg)
= Vp(WJ) - Wp(VJ) + c(Vp(Wg) - Wp(Vg))
= [V, W]vf + c[V, W]v9·
To show the Leibniz rule, we have, for f,g E C 00 (M),
[V, W]p(Jg) = Vp(W(Jg)) - Wp(V(Jg))
= Vp(g(WJ) + f(Wg)) - Wp(g(VJ) + f(Vg))
= g(p)Vp(W J) + (W J)(p)Vpg + f(p)Vv(W g) + (W g)(p)Vpf
- g(p)Wp(V J)-(V J)(p)Wpg - f (p )Wp(Vg)- (V g )(p )Wvf
= g(p)[V, W]vf + f(p)[V, W]v9·
4 Propositions 3.1, 3.2, 6.6, Theorem 6.3.
Finally, we show smoothness of p >-+ [V, W]p- If/ E C 00 (M), then as V, W

are vector fields, Theorem 3.1 implies Wf, VJ E C 00 (M), and moreover
V(WJ), W(VJ) E C 00 (M) too. So [V, W]f = W(VJ) - V(WJ) E C 00 (M)
for all f E C 00 (M), and again by Theorem 3.1, it follows that [V, W] is a
smooth vector field. □
Now we show that the Lie bracket of chart-induced vector fields vanishes.
Lemma 3.3. Let (U, x) be an admissible chart for a smooth manifold, and
let Oxi E TJU, 1 ,:;;; i ,:;;; m, be the coordinate induced vector fields on U.
Then [ox;,Oxi]=O for all l,,;; i,j,,;; m.
Proof. For f E C 00 (U), we have
Oxi (Oxi J) = Oxi (P >-+ : ~

0
x-l) (x(p)))
= (p>-+ :ui(:~ox-l) oxox- 1 )(x(p)))

02(Jox- 1) ) ( 02(Jox- 1))
= (p >-+ ouii}ui (x(p)) = ouii}ui o X.
Oxi(Ox;J) is obtained by the swap i +-+ j. As f o x- 1 is C 00 , thus also

C 2 , its second order partials do not depend on the order of taking partial
derivatives (see e.g. [Apostol(1969), §8.23]). So
a2u Ox-1) a2u Ox-1)
ouii}ui
Consequently, [ox;, Oxi] = 0. □
Exercise 3.9. Let M, N be smooth manifolds, f : M -+ N be a diffeomorphism,
and X, YeTJM. Show that f*[X, Y] = [f*X,f*Y].
Exercise 3.10. Let G be a Lie group, and g its Lie algebra. Show that if V, Weg,
then [V, W] E g.
Exercise 3.11. Let M be a smooth manifold, X, YE TJM and f E C 00 (M). Show

that [f X, Y] = f[X, Y] - (Yf)X.
Consider the smooth manifold R 2 with the standard smooth structure. Let Ox, Oy
be the vector fields induced by the global chart (R 2 , (x,y) >-+ (x,y)). Show that
[(1 + y) Ox, Oy] = - Ox-
Exercise 3.12. Suppose that M is a smooth manifold, (U, x) is an admissible

chart for M, and Ox1,··· ,Ox= E TJU are the chart-induced vector fields on U.
Let V = Viox; and W= Wioxi, where Vi, Wi E C 00 (U). Show that
[V, W]=(ViOxiWj- WiOxiVj)Oxi•
The Lie bracket has the following properties.

Theorem 3.2. Let M be a smooth manifold, X, Y, Z E TJ M, and c E R
Then the fallowing hold:
(1) [X + cY, Z] = [X, Z] + c[Y, Z].

(2) [X, Y] = -[Y, X].
(3) (Jacobi identity) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y]] = 0.
Proof. (1), (2) follow immediately from the definition. We verify (3)
below. Each term on the left hand side gives 4 terms, and altogether we
get twelve terms, which will cancel in pairs, thanks to the cyclicity of the
expression on the left-hand side: For f E 0 00 ( M),
[X, [Y, Z]]f = X([Y, Z]J) - [Y, Z](XJ)
= X(Y(ZJ) - Z(YJ)) - Y(Z(XJ)) + Z(Y(XJ))
= x(Y(ZJ)) - ~-(~_c!.DJ- ~ + ~\~.(~:rn-
By cyclically permuting the vector fields, we can read off that
[Y, [Z,X]]f = Y(Z(XJ)) - Y(X(ZJ)) - Z(X(YJ)) + X(Z(YJ)),
~ --------
[Z, [X, Y]]f = Z(X(YJ)) - Z(Y(XJ)) - X(Y(ZJ)) + Y(X(ZJ)).
Adding these, we see that the result is the zero function. □
3.5 Cotangent bundle
Definition 3.7. (Cotangent bundle).

Let (M, [A]) be a smooth manifold. Define the set
TM*= LJ (TpM)* = {(p,w): p EM, and w E (TpM)*}
pEM
= LJ ({p} x (TpM)*).
pEM
For each chart (U, cp) in an atlas A belonging to [A], define U by

U = {(p,w): p EU, w E (TpM)*} = LJ ({p} x (TpM)*),
pEU
and cp: U cp(U) x !Rm by

-->
cp(p,w) = (cp(p),w(ox1,p),··· ,w(ox"',p)) E cp(U) X !Rm,

where the component functions of cp are denoted by x 1 , • • • , xm. We call
(U, cp) the lift of the chart (U, cp). Let A:= {(U, cp) : (U, cp) EA}. Define
the cotangent bundle to be the smooth manifold (TM*, [A]). The map
1r : TM* --> M sending (p, w) E TM* to p E M, is called the canonical
projection.
Since the sets U from A cover M, the corresponding U from A cover

TM*. Also, as cp : U -> cp(U) is surjective, and because every m-tuple
(w1, · · · , Wm) E !Rm results in a I-form w = Wi(dxi)p E (TpM)*, <p is a
surjective map. The map <pis also injective, since if 1.p(p,w) = 1.p(q,v),
then we have that cp(p) = cp(q), so that p = q, thanks to the injectivity of
cp, and moreover, we have w = w(ox;,p)(dxi)p = v(ox;,q)(dxi)q = v. Hence
 cp(U) x !Rm is a bijective map onto the open subset cp(U) x !Rm of
JR 2m, and so (U,<p) is a chart for TM.
Moreover, if (U, <p), (V, 1P) are the lifts of the charts (U, cp), (V, '¢) for
M, then we have 1.p(U n V) = cp(U n V) x !Rm, which is open in JR2 m.
Let (U, 1.p) and (V, -ip) be the lifts of admissible charts (U, cp) and (V, '¢)
for M. We will show that the chart transition maps corresponding to
(U, <p), (V, 1P) are smooth. Besides completing the verification that A is
an atlas, this will also show that the construction of the smooth structure
on TM* is independent of the choice of the defining atlas A from the
smooth structure [A] for M. Let the point (a,/3) E 1.p(U n V). With
/3 = (/31, · · · , f3m), cp = (x1, · · · , xm) and 'Ip= (y 1, · · · , ym), we have
(-ip o <p- 1)(0,/3)
= -ip(cp- 1 0, /3i(dxi)"'-1a.)
= (('¢ o cp- 1 )(a),/3i(dxi)"'-1a.(8y1,'l'-1a.), • • • ,/3i(dxi)"'-1a.(oy"','P-1a.)).
The map Un V 3 a>-+ (cp o '¢- 1 )(a) is C 00 • Also,
i i o(xi O 'lp-1) -1
(dx )<p-1a.(Oyi,<p-1a.) = Oyi,<p-ta.(X) = oui ('¢(cp a))
= o(cp_ o 'lp-l)i (('¢ o cp-l)(a))
ouJ
= ((:~o'¢-l)i)o('¢ocp- 1))(a).
Hence the map sending a to (dxi)"'-1a.(Oyi,<p-1a.) is a C 00 function. Thus
(a, /3) >-+ ( ( '¢ o cp- 1 ) (a), /3i (dxi)"'-•a. (8y1 ,<p-•a.), •• • , /3i (dxi)"'-•a. (oy"' ,<p-•a.))
is C 00 , as wanted.
Consequently, TM* is a smooth manifold of dimension 2m, where m
is the dimension of the smooth manifold (and also the dimension of each
cotangent space (TpM)*).
Exercise 3.13. Let M be a smooth manifold. Show that the canonical projection
TM* -+ M is a smooth map.
1r :
3.6 1-form fields
In this section, we will introduce the 'dual' object to a vector field, namely
a 'I-form field', which is, roughly speaking, a smooth distribution of 1-
forms on the manifold. We will see that, just like the C 00 (M)-module TJ M
of vector fields, the collection Tf M of all I-form fields will also have the
structure of a C 00 (M)-module. Together, TJ Mand Tf M form the building
blocks for (r, s)-tensor fields (which will be 'C 00 (M)-multilinear maps' from
r copies of Tf M ands copies of TJ M to C 00 (M), as we will learn in the
next chapter). These tensor fields form the vocabulary to discuss geometric
notions in Lorentzian geometry, such as the metric g (a (0, 2)-tensor field),
the curvature tensor field R (a (1, 3)-tensor field), etc. The field equation
for spacetime is an equality between two (0, 2)-tensor fields.
Recall that in the previous section, using the smooth manifold structure
on M, we made TM* a smooth manifold by gluing together the cotangent
spaces smoothly. Hence we can talk about smooth 'TM*-sections', which
are called I-form fields.
Definition 3.8. (I-form field).

Let M be a smooth manifold, and TM* be its cotangent bundle. A I-form
field is a smooth map O: M-> (TM)* such that 1r o O = idM. For p EM,
we write O(p) = (p, Op), where Op E (TpM)*. The set of all I-form fields
on M is denoted by Tf M.
Definition 3.9. (I-form field action on a vector field).

Let V E TJ M be a vector field and O E Tf M a I-form field. Then the
function O V: M -> JR is defined by (0 V) (p) =Op(½,) for all p E M.
Exercise 3.14. Let n E TfM. Show that the map TJM 3 V >-+ OV E C 00 (M)
is a C 00 (M)-linear map: for all V, W E TJ M and all f E C 00 (M), we have
n(v+JW) = nv+Jnw.
The following result gives useful criteria for checking if a TM* -section is
smooth.
Theorem 3.3. Let M be a smooth manifold. Let O : M -> TM* satisfy

o O = idM. Then the following are equivalent:
1r
(1) 0 is smooth.
(2) For every VE Tf M, OV E C 00 (M).
(3) For all charts (U, x) from an atlas defining the smooth manifold M, the
maps given by U 3 p >-+ Op(lJxi,p) E JR, 1,:;;; i,:;;; m, belong to C 00 (U).
0
Proof.
(1)=?(2): Let (U, cp) be an admissible chart for M. For a E cp(U), we have
((0 V) 0 cp- 1 )a = o,.,-la V,.,-la = o,.,-1a(V,.,-1a(xi)Oxi,cp-la)
= V,.,-1a(xi)0,.,-1aOxi,cp-la·
As V is smooth, a >-+ V,.,-1a (xi) is smooth on cp(U). Also, since O is
smooth, <po O o cp- 1 is smooth, that is,
0 f-+ ( a, o,.,-la (Oxl,cp-la), • • • , o,.,-1a(Ox"',cp-la))
is smooth. So the map a >-+ 0,.,-1aOxi,cp-1a is smooth for all i. Thus
a>-+ ((OV) o cp- 1 )a is smooth.
(2)=?(3): Let p E U. Let x E C 00 (M) be a bump function which is identi-

cally 1 in a neighbourhood U0 c U of p, and is O outside a slightly bigger
neighbourhood U1 with U1 c U. Then we have that XOxi E TJM, and so
O(xox;) is smooth. Hence for a E cp(U0 ), the map
0 f-+ (O(xox;))(cp- 1 a) = o,.,-1a(XOxi)cp-la = o,.,-1a(loxi,cp-la)
= o,.,-laOxi,cp-la
belongs to C 00 (cp(U0 )). So U 3 p >-+ Op(Oxi,p), l~i~m, belong to C 00 (U).
(3)=?(1): For a E cp(U),
( I.{) 0 0 0 cp- 1 )a= (a, o,.,-la (ox1,,.,-1a),""" , o,.,-1a(Ox"',cp-1a) ).
As U 3 p >-+ Op(Oxi,p) are smooth, the maps cp(U) 3 a>-+ 0,.,-1a(oxi,,.,-1a),
1 ~ i ~ m, are all smooth. Hence a>-+ (1.p o O o cp- 1 )(a) is smooth. So 0
~~~~ □
Exercise 3.15. (Gradient field). Let M be a smooth manifold, and f E C 00 (M).

Define df: M-+ TM* by (dj)(p) = (p,dfp), for p EM, where dfp E (TpM)* is
the 1-form defined in Example 2.5. Prove that df E Tf M.
We identify O(p) = (p, Op) with Op E (TpM)*, and so the 1-form field 0
sends points p EM to corresponding 1-forms Op E (TpM)*.
Clearly, the result of pointwise adding a pair of 1-form fields is a new
1-form field, and so is the result of scaling a 1-form field pointwise. This
makes the set Tf M of all 1-form fields into a vector space. However, given a
1-form field 0, rather than just uniformly producing a scaling with a factor
c E JR, that is, M 3 p >-+ c Op E (TpM)*, we can consider a less restrictive
possibility of having the c depend on pin a smooth manner. We will see
that this makes Tf M into a C 00 (M)-module. First we give the definition
of multiplication of arbitrary 1-form fields by elements of the ring C 00 (M).
Definition 3.10. (Multiplication of I-form fields by smooth functions).

Let M be a smooth manifold. Let n E Tf M and f E C 00 ( M). Define
f• !1 E TfM by
U · n)(p) = f(p) np for all p EM.

Tf M is a module over the ring C 00 ( M). In particular, besides the usual
vector space axioms with respect to vector addition in Tf M, we have that
for all f, g E C 00 (M), and all n, 8 E Tf M, the following hold:
1-n = n
(Jg)· n = f·(g· n)
(J+g)-n = f-n+ 9 -n
f-(n + e) = f-n + f-e.
Here 1 E C 00 (M) denotes the constant function equal to I everywhere.
Example 3.3. (Chart-induced I-form fields).

Let (U, x) be an admissible chart for a smooth manifold with the coordinate
functions denoted by xi, I,:;;; i,:;;; m. Then (U, [A]), where A= {(U,x)},
is a smooth manifold. Since (dxi)pOxi,p = 83,
Theorem 3.3 implies that
U 3p>-> (dxi)p, I,,;;i,,;;m, are all elements ofTfU. ◊
Lemma 3.4. (Extension of a I-form to a I-form field).

Let M be a smooth manifold. Let w E (TpM)*. Then there exists a I-form
field n ETJ M such that nP = w.
Proof. Let (U, x) be an admissible chart containing p. Let x E C 00 (M)

be a bump function that is I in a neighbourhood of p, and identically 0
outside a neighbourhood V with V c U. Write w = wi(dxi)p- Define the
TM* -section n
□
Exercise 3.16. (A 1-form field which is not a gradient field).
Consider R 2 with the standard smooth structure. Show that not every 1-form field
0 on R 2 is df for some f E C 00 (R2 ) as follows. If O = df for some f E C 00 (R 2 ),
then in the chart (R 2 ,(x,y) >-+ (x,y)), writing n = (oxf)dx +(oyf)dy, we get
Oox = oxf and Ooy = oyf, implying oy(Oox) = ox(Ooy)- Construct a 1-form field
n, which is not a gradient field, by choosing the component functions n Ox, n Oy
appropriately.
3. 7 Pull-back of a 1-form field
We had seen that given a diffeomorphism f : M --+ N between two smooth

manifolds M and N, any vector field V E TJ M can be pushed forward to a
vector field /*VE TJ N. We had also seen that a I-form w at J(p) can be
pulled back to a I-form (dfp)*w at p EM. Now we will see that we can in
fact pull back a whole I-form field in this manner, and what we need from
f is that it only be smooth (so f need not be a diffeomorphism).
Definition 3.11. (Pull-back of a I-form field).
Let M, N be a smooth manifolds, and f : M --+ N be a smooth map.
The pull-back f*O E Tf M of n E Tf N under f is defined by
(J*O)p = (dfp)*Of(p) for all p EM.
It remains to be checked that this does define a I-form field on M. It is

enough to check that for every VE TJM, we have (J*O)V E C 00 (M). We
check this locally around a point p by making use of a admissible charts
(U, x), (V, y), for M, N, respectively, containing p, f(p), respectively, and
such that J(U) c V. For all q E U, writing
Vq = Vq(xi)oxi,q and Of(q) = Of(q)(oyi,J(q))(dyi)J(q),
we have
((f*O)V)(q) = ((dfq)*Of(q))Vq
= Of(q)(dfq(Vq))
= 0 J(q) (oyi ,f(q))(dyi) f(q) (dfq(Vq(xi) ilxi,q))
= Of(q)(oyi,J(q)) · (dyi)J(q)(Vq(xi) · dfq(oxi,q))
. . k
= r,,f(q)(Oyi,J(q)). Vq(x'). (dy 1 )J(q)(((dfq)(ox;,q))(y )Oyk,J(q))
i . k
= Of(q)(Oyi,J(q)) · Vq(x) · (dy 1 )J(q)(Oxi,q(Y o J) · Oyk,J(q))
= Of(q)(oyi,J(q)) · Vq(xi) · Oxi,q(yi o !).
Since the maps
U 3 q >-+ Of(q)Oyi,J(q), Vq(xi), Oxi,q(yio!)
are smooth, so is their product.
Exercise 3.17. Let M,N be smooth manifolds, f: M--+ N be a smooth map,
g E C 00 (N), and 0, E Tf N. Prove that f*(g · 0.) =(go f)-(f*O).
Exercise 3.18. Let M, N be smooth manifolds, f: M--+ N be a smooth map, and

g E C 00 (N). Prove that f*(dg) = d(g o !).
The previous two exercises give a recipe for the computation of the pull-back
in any chart. Let p EM, and (U, x), (V, y) be admissible charts for M, N,
respectively, containing p, f(p), and such that f(U) c V. If n E TP N and
we define nj = n(oyi) E C 00 (V), then in u,
j*n = j*(nidyi) = (ni o J)-f*(dyi) = (ni o J)-d(yi o J)
= (ni o J)-((d(yi o J))(oxi))-dxi = (ni o J)-(ox;(yi o J))dxi.
Example 3.4. Let M = !R3 and N = JR2 be equipped with the standard
smooth structures. Consider the admissible charts:
(U = IR 3 , (x, y, z) >-+ (x, y, z)) for IR3, and
(V = IR 2 , (u,v) >-+ (u,v)) for JR 2 .
Let f :IR3 --+ IR 2 be the smooth map f(x, y, z) = (x +y, yz) for (x, y, z) E IR3 .
Let n = vdu + udv E Tf N. Then we have
j*n = (v o J)-(ox(u o J)dx + oy(u o J)dy + oz(u o !)dz)
+(u o J)-(ox(v o J)dx + oy(v o J)dy + oz(v o !)dz)
= (yz)·(ox(x + y)dx + oy(x + y)dy + oz(x + y)dz)
+(x + y)·(ox(yz)dx + oy(yz)dy + oz(yz)dz)
= (yz)-(dx + dy) + (x + y)(zdy + ydz)
= yzdx + (x + 2y)zdy + (x + y)ydz. ◊
3.8 Integral curves and the flow of vector fields
A vector field describes a 'flow' on a manifold. For example, consider the

wind velocity vector field on the surface of the Earth. Then the path traced
by a dust particle describes an 'integral curve' 'Y, namely, at a point p along
'Y, the velocity of the speck of dust (which is the tangent vector to the
curve), is the wind velocity at that point.
Definition 3.12. (Integral curve).
Let V E TJ M. A curve 'Y : J --+ M, where I c JR is an open interval, is
called an integral curve of V if V"Y(t) = v"Y,"Y(t) for all t E J.
Using a chart-induced b asis representation (Example 2.4), the integral
curve condition V"Y(t) = v"Y,"Y(t) (t E J) can be locally expressed as a system
of first order ordinary differential equations: if V = viOxi in a chart (U, x),
where vi = V(xi) E C 00 (U), then
i d(xo'Y)i
V ('Y(t))Oxi,")'(t) = dt (t)Oxi,")'(t), tEJ.
Thus component-wise, we have:
:t O
1')i (t) = vi(1'(t)), 1,:;;; i,:;;; m, t E J. (3.1)
Define them functions 1'i: J-> IR by 1'i(t)=xi(1'(t)) = (xo1')i(t), fort EI,

and 1,:;;; i,:;;; m. Suppose that OE I, and 1'(0) = p. Set ~i = xi(p). Writing
vi(1'(t)) = vi(x- 1 o (xo1')(t)) on the right-hand side of (3.1), we obtain the
initial value problem:
, t E J. (3.2)
Example 3.5. Consider the smooth manifold IR2 with the standard smooth
structure given by the global chart (IR2 , (x, y) i--+ (x, y)). Let V = xoy-YOx-
The differential equations describing an integral curve 1'(t) = (x(t), y(t)) of
V are given by
{ ±(t) = -y(t)
y(t) = x(t)
where· = ft-. If the curve starts at t = 0 from (xo, Yo), then the unique
solution is
[ x(t)] = [c?st -sint] [xo], tER

y(t) sm t cost Yo
The resulting curve 1' in IR 2 (thought of as a plane) describes the path of
a counterclockwise moving particle in a circular path with centre (0, 0),
starting at (xo, Yo)- ◊
Exercise 3.19. In contrast to the above example, where the integral curve existed
for all t E R, the curve may exist only on a finite interval. As an example, consider
V = u 2 Ou E TJ R on the smooth manifold R with the standard smooth structure
given by the global chart (R, u >-+ u). Show that if the integral curve starts at
t = 0 at p = -y(O) > 0, then -y(t) = p/(1 - pt) for all t < 1/p, which 'escapes to oo'
in the finite time 1/p < oo.
From the theory of differential equations, it follows that (since the vi are
all smooth) there exists a neighbourhood (-e:, e:) of O and a neighbourhood
e
N of x(p) in x(U) such that for each = (~ 1, · · · , ~m) E N, the initial value
problem for the system of ordinary differential equations (3.2), has a unique
smooth solution (-e:, e:) 3 t 1--+ (1' 1 (t; e), · · · , 1'm(t; e)). Also, the solution to
(3.2) depends smoothly on the initial condition f We will not prove this
here, but refer the interested reader to [Hartman(2002), Thm. 4.1, p.100].
Thus (3.1) has a unique smooth solution 'Y(t; x(q)) for each initial condition
q in the neighbourhood x- 1 N of p, living on the time interval (-1:, 1:), and
the map x- 1 N 3 q >-+ 'Y(t; x(q)) is smooth.
Let us suppose from now on that for our vector field on the smooth
manifold M, the integral curves exist for all t E JR, and the pathological
cases as in Exercise 3.19 do not arise. We will call such vector fields as
complete vector fields.
Definition 3.13. (Complete vector field).

Let M be a smooth manifold. A vector field V E TJ M is called a complete
vector field if each of its integral curves exist for all t E JR.
Suppose that a smooth manifold admits a complete vector field V. Then

we obtain for each frozen t E JR, a map 1Pt, given by
sending the initial condition 'Yp(O) = p to the evaluation-at-time t of the

integral curve 'Yp of V. In other words, we imagine a whole bunch of particles
starting at different places being transported along the manifold by the
vector field V, and then looking at the result (taking a snapshot) at time t.
Then p >-+ 'Yp(t) : M-+ Mis smooth, thanks to the smooth dependence of
the initial value problem solution on the initial condition. The collection of
maps {'1/Jt, t E JR} ('snapshots' of the ensemble of particles at various times
t E JR) describe the 'flow' of the vector field V.
Definition 3.14. (Flow of a vector field).
Let V E TJ M be a complete vector field on the smooth manifold M. Let
'Yp: JR-+ M be the integral curve of V satisfying 'Yp(O) = p EM. Fort E JR,
the map 1Pt : M-+ M given by '1/Jt(P) = 'Yp(t) for all p EM, is called a flow
map of V. The collection of all flow maps, Fv = {'1/Jt : M-+ M, t E JR}, is
called the flow of V.
Since for each p E M, 'Yp(O) = p, we have 'I/Jo = idM. Also the following
result makes composition a group operation on Fv.
Lemma 3.5. Let V E TJ M be a complete vector field on the smooth mani-

fold M with the flow {1Pt : M -+ M, t E JR}. Then 1Pt+s = 1Pt o 1Ps = '1/Js o 'lpt
for all t, s E R
Proof. For p EM, let "Ip be the integral curve of V such that 'Yp(O) = p.
Lets ER. Set q = "fp(s) = 1/Js(p).
______'Yq
q = 'Yp(s)
Consider the curve 'Y: R-+ M, given by 'Y(t) = 'Yp(s+t) for all t ER Then
'Y(O) = "fp(s) = q, and V7 (t) = V7 p(s+t) = v7 p,'°Yp(s+t)· But for f E C 00 (M),
_ d(J o "fp) ( t) _ 1. f('Yp(s + t + h)) - f('Yp(s + t))
v,,,,p, "",p (s+t) f - d t s + - h-+O
Im h
_ 1. f('Y(t + h)) - f('Y(t)) _ d(J o 'Y) (t) _ f

- h~ h - dt - v'Y,'Y(t) ·
Hence V7 (t) = v7 p,'°Yp(s+t) = v7 ,7 (t), that is, 'Y is an integral curve of V such
that 'Y(O) = q. By the uniqueness of solutions, 'Y = 'Yq· So
for all p EM. Thus 1Pt+s = 1Pt o 1Ps· Also, 1Pt+s = 1Ps+t = 1Ps o 1Pt· □
Composition of maps is associative. Also, idM = 1/Jo E Fv serves as the

identity element with composition. By Lemma 3.5, every 1Pt E Fv has an
inverse with respect to composition: 1/J-t 01/Jt = 1Pt 01/J-t = 1Pt-t = 1/Jo = idM.
As each 1Pt E Fv is smooth, and (1/Jt)- 1 = 1P-t, we conclude that every flow
map is a diffeomorphism.
Exercise 3.20. Consider the smooth manifold JR 2 with the standard smooth
structure given by the global chart (JR 2 , (x, y) >-+ (x, y)). Determine the flow
{'lf;t : 1R2 -+ 1R2 , t E JR} of the vector field V = (1 + y) Ox- If p = (xo, yo) E 1R2 , then
find (d'I/J-t)poy,p•
Exercise 3.21. Consider the smooth manifold JR3 with the standard smooth struc-
ture given by the global chart (JR3 , (x, y, z) >-+ (x, y, z)). Let Lx, Ly, Lz E TJJR3 be
the vector fields given by Lx =yoz - ZOy, Ly =ZOx - XOz, Lz =XOy -yOx-
Compute [La, Lb] for a, b = x, y, z. Find the flow maps for Lx, Ly, Lz.
Finally we give the geometric meaning of the Lie bracket. More generally,
we define a Lie derivative below, which measures the rate of change of the
quantity at hand (function, vector field, and later on even tensor fields)
along the flow of a vector field.
Definition 3.15. (Lie derivative).

Let M be a smooth manifold. Let VE TJ M. The Lie derivative £v f of a
function f E C 00 (M) with respect to Vis the function V f E C 00 (M). The
Lie derivative £v W of a vector field W E TJ M with respect to V is the
vector field [V, W] E TJ M.
Proposition 3.1. Let M be a smooth manifold, and V E TJ M be a com-

plete vector field with the flow {'lf't : M-+ M, t E ~}. Then for all p EM
and all f E C 00 ( M), we have
(£v f)(p) = :~u O

1PtHP)) (0).
Proof. Let ')'p(t) = 'lf't(P), t E ~, be the integral curve of V such that

')'p(0) = p. Then (f o 'lf'tHP) = f('lf'tP) = fbv(t)) = (f o ')'p)(t). Thus
d((f O -ipt)(p)) (0) = d(f O ')'p) (0) = V f = V f
dt dt ')'p,'/p(O) ')'p ,P
= Vvf = (Vf)(p) = (£vf)(p). □
Proposition 3.2. Let M be a smooth manifold, and V E TJ M be a com-

plete vector field with the flow {'lf't : M -+ M, t E ~}. Then for all vector
fields W E TJ M,
£vW = [V, W] = !~(d¢-t),i,,.W,i,,.) (0).
Here (d'lf'-t),i,,.W,i,,. is the vector field M 3 q ~ (d¢-t),i,,qW,t,,q E TqM. The

smoothness follows by noting that this vector field is (1P-t)* W: For q EM,
Thus for each p E M,

d((d¢-t),i,,pW,t,,v) (0) = lim (d¢-t),i,,vWw,v - Wv
& t~ t
makes sense as an element of TpM because both WP and (d¢-t),t,,pW,t,,p
belong to TpM (and provided we accept that the derivative exists).
Proof. Let 7q(t) = 1/Jt(q) fort ER and q EM. Let p EM. Define H by
H(t, s) = W,t,.p(/ o 1/J-s)- Let f E C 00 (M). Then
(V(WJ))(p) = Vp(WJ) = V-yp,)'p(O)(WJ) = :;(WJ) O'y'p) (0)
d((W/)(1/JtP)) (0) = d(W,t,.p/) (0) = oH (0 0).

=
dt dt ot '
With h(t) = -t for all t ER, we have {'Yq o h)(t) = 1/J-t(q), and
(V J)(q) = Vqf = V-yq,qf = :? 0

'Yq) (0) = -(-1) :? 0
'Yq) (0)
= -h(O) d(J o 'Yq) (0) = _ d(J o 'Yq oh) (0) = _ d((J o 1/J-t)(q)) (0).
dt dt dt
We have 5
(W(V J))(p) = Wp(V J) = - Wp( M 3 q ,_. :;(! 0
1P-t)(q)) (0))
(!l-:;Wp(Jo'lj;_t))(O) = -::(0,0). (3.3)
Thus
[V, W]pf = ([V, W]J)(p) = (V(WJ))(p) - (W(VJ))(p)
= oH(O 0) oH(O 0)= d(t>-+H(t,t))(O)
at ' + as ' dt
= :;W,t,.p(/ o 1P-t)) (O).
Consequently, [V, W] = :;(d1/J-t),t,•. W,t, •. ) (0).

□
Notes
The proof of Proposition 3.2 is based on [Godinho and Natario(2014), see Exer-
cise 12(a), p.33, and its solution, pp.332-333].
5 The equality ( *) in (3.3) needs justification. Formally, it is the process of taking Wp

inside the limit:
W d((f o 'P-t)(·)) (0) = W lim (f o 'P-t)(·) - f lim Wp((f o 1/J-t)(·)) - Wpf d(Wp(f o 1/J-t)) (0).
P dt P •-o t •-o t dt
We refer the reader to [Lee(2013), Thm. 9.38] for a detailed proof Proposition 3.2.
Chapter 4
Tensor fields
In this chapter, using the building blocks TJ- M and Tf M, namely the
C 00 (M)-modules of all vector fields and of all 1-form fields, we will in-
troduce (r, s)-tensor fields, which will be C 00 (M)-multilinear maps defined
on the Cartesian products of these modules to C 00 (M). As we had men-
tioned earlier, tensor fields are fundamental in Lorentzian geometry and
in spacetime physics. The Lorentzian geometry will be specified by the
metric g on a smooth manifold, where the metric will be a certain type
of a (0, 2)-tensor field. This metric will induce a curvature tensor field R
(via something called the Levi-Civita connection), and R will be a (1, 3)-
tensor field. Vector field and 1-form fields are themselves simple examples
of tensor fields: vector fields are (1, 0)-tensor fields, and 1-form fields are
(0, 1)-tensor fields.
4.1 (r, s)-tensor fields
We are familiar with multilinear maps in the context of vector spaces (for
example, the inner product on Rm is a bilinear map, and the determinant
on Rm is an m-linear form with some extra properties). An (r, s)-tensor
field is just a special type of a multilinear map, except that the field R of
scalars is replaced by the ring C 00 ( M) consisting of all smooth functions on
a smooth manifold M.
Definition 4.1. ((r, s)-tensor field).
Let M be a smooth manifold. Let r, s ~ 0 be integers. An (r, s)-tensor
field is a map
T: TfMx ·· · xTfM x TJ-Mx · · · xTJ-M-+ C 00 (M)
r copies s copies
which is C (M)-multilinear, that is, it is C 00 (M)-linear in each slot.

00
61
More explicitly,
for all 1 ~ i ~ r, for all D, 8 E Tf M , f E C 00 (M) , and
for all n 1, · · · , ni-l, [1i+l, · · · , n r Erp M , Vi,···
, Vs E TJ M,
T(D1,· ·· f8 , Di+1, . . . , nr , Vi , ··· ,Vs)
,ni- 1 , D+ (4.1)
T(D ,··· ,ni- 1 , n , ni+ 1 , ... ,nr,Vi,··· ,Vs)
1
+ J T(D1,· ·· ,ni- 1 , e , ni+1, . . . , nr , Vi,··· , Vs),

and
for all 1 ~j ~ s, for all V , W E TJ M , f E C 00 (M), and
forallD1, · ·· , DrETfM, Vi , ··· , Vj-1 , Vj+1 , ··· , V.ETJM,
T(D1, ··· ,nr,Vi,··· ,½-1 , V + JW , Vy+1 ,· ·· ,Vs) (4.2)
T(D 1, · · · , nr , Vi , · · · , ½-1, V , Vj+l , · · · , Vs)
+ J T(D 1, · · · , nr , Vi , ··· , ½-1, w, ½+1 , · · · , Vs).
The set of all (r , s)-tensor fields is denoted by M. r;
If r > 0 ands =0, then only (4.1) is required to hold (with (4.2) satisfied
vacuously). If r = 0 and s > 0, then only (4.2) is required to hold. If r = 0 = s,
then T8M:=C 00 (M).
Example 4 .1. Let M be a smooth manifold. Then T: Tf MxTJM----> M,

given by T(D, V) =DV for all DE Tf Mand all VE TJ M, is C 00 (M)-linear
in each slot. So T is a (1, 1)-tensor field. ◊
As 1-form fields act C 00 (M)-linearly on vector fields (see Exercise 3.14),

every 1-form field is a (0, 1)-tensor field. Note that the notation for the set
of (0, 1)-tensor fields is TP M, which coincides with the notation for 1-form
fields we had used in Chapter 3. Likewise, if V E TJ M is a vector field, we
can define a (1, 0)-tensor field Tv by setting Tv(D) = D(V) for all DE TP M.
Thus every vector field can be considered as a (1 , 0)-tensor field .
Exercise 4.1. Suppose that VE TJ M. Let Tv be the (1, 0)-tensor field given by
Tf M 3 !1 >--> !1(V) E C 00 (M). Suppose Tv =0 (the map Tf M 3 !1 >--> 0 E C 00 (M),
where O(p) =0 for all p EM). Prove that V =0 (the zero vector field), that is, for
all p E M, Vp = 0, the zero vector in TpM.
Example 4 .2. (A non-tensor field). Let WE TJ M be a fixed vector field

on M, and g E C 00 (M) be a fixed function. Define T: TJ(M)----> C 00 (M) by
T(V) = W (V g) for all V E TJ M. Then Tis additive, since for Vi, Vz E TJ M,
T((Vi + Vz)g) = W(Vig+ Vzg) = W(Vig) + W(Vzg) = T(Vi)+T(Vz) .
Tensor fields 63
However, T is not 0 00 ( M)-linear in general, since for V E TJ M, and for

/ E C 00 (M),
T(JV) = W(!Vg) = JW(Vg)+(Vg)Wf = JT(V)+(Vg)WJ=t=JT(V),

whenever (V g)W f =I= 0. ◊
Exercise 4.2. Let W E TJ M and g E C 00 (M). Define T : TJ M -+ C 00 (M) by

T(V) = V(W g) for all VE TJ M. Show that Tis a (0, 1)-tensor field on M.
TJ Mis a C 00 (M)-module with pointwise addition, and scalar multiplication

by C 00 (M) elements defined as follows: For TE TJ Mand/ E C 00 (M),
(J-T)(0.1,- · · ,nr, V1, · · ·, V,,) = f •(T(0.1,- · · ,nr, Vi,···, V,,)),
for all I-form fields 0. 1 , · · · , n,r E rp M, and vector fields Vi, · · · Vs E TJ M.
Note that the multiplication • on the right-hand side is the multiplication
in C 00 (M).
Exercise 4.3. ((8) and A). Let 0, 8 E Tf M. Define the 'tensor product' of the
fields O and 8, 0(8)8: TJMxTJM-+ C 00 (M), by (O®8)(V, W)=(OV)(8W)
for all V, W E TJ M. Show that O (8) 8 is a (0, 2)-tensor field. Now define the
(0, 2)-tensor field OA8 ('wedge product') by OA8 =0(8)8- 8(8)0, and similarly
also 8AO=8(8)O- 0(8)8. Show that OA8= -8AO, and that (OA8)(V, V) =0
for all VE TJM.
Exercise 4.4. (Bilinearity of®). With the same notation as in the previous ex-
ercise, show that for any C 00 (M) function f, f(O ® 8) = (f 0) ® 8 = 0 ® (!8).
Also, prove that for 01, 02, 8 E Tf M, (01 + 02) (8) 8 = 01 (8) 8 + 02 (8) 8 and
8® (01 +02) =8®01 +8®02.
Let TE TJ M with r ~ l. Let n E Tf M. Then evaluation, say, in the first

slot, with 0.,
(TPMt- 1 x(TJM)s -+ C 00 (M)
(0.1,--- ,nr- ,Vi,··· ,Vs)>-+ T(0.,0.1,---
1 ,nr- 1 ,Vi,··· ,V,,)
is an (r-1, s)-tensor field. Similarly, evaluating Ton any (1 ~) i (~r) fixed
I-form fields, and any (1 ~) j ( ~ s) fixed vector fields, gives an (r - i, s - j)-
tensor field.
4.2 Point evaluation of tensor fields

We know that a I-form field n evaluated at a point p EM gives a I-form
O.p E (TpM)*. Similarly, a vector field V evaluated at a point p gives an
element of the vector space TpM. Now we will show that an (r, s)-tensor
field at a point p gives rise to an '(r, s)-tensor'. Thus, an (r, s)-tensor field
can be viewed as a mapping which assigns to each point p of the manifold,

an (r, s )-tensor. We first define the linear-algebraic notion of a tensor 1.
Definition 4.2. ((r, s)-tensor).
Let V be an m-dimensional real vector space and V* denote its dual. Then
an (r, s)-tensor on Vis an JR-multilinear map
'T : V* X .•. X V* X V X •.• X V --> JR
'--v--' '---...,.--'
r copies s copies
that is, T is JR-linear in each slot. More explicitly,

foralll!( i !(r, forall w, 0 EV* , c EJR, and
for all w1, • • • , wi-l,wi+ 1, • • • , wr EV* , v1, •· ·, v s EV, we have
T(w1, · ·· ,wi- 1, w+ c0, wi+l, . . . , wr,v1,··· , vs) (4.3)
'T(W1,·· · ,W i-1, w ,Wi+ 1, · · · ,Wr , V1 , · · · , Vs )
1 · ·, Wi- 1, 0 ,Wi+l ,· ·· ,Wr ,V1,···,Vs,
+ C 'T(W,· )
and
for all 1 !(j !( s , for all v , w E V, c E JR, and
forallw 1,·· · ,wrEV*, w1,··· ,Vj-1,Vj+l,· · · ,Vs EV, we have
T(w 1 ,· ·· ,wr,v1, · · · ,Vj-1, v + cw , v 1+1,··· ,vs ) (4.4)
T(w1, ·· · ,wr,v1,··· , Vj - 1, v , v 1+1,·· · ,vs)
+ cT(w 1, · ·· ,wr,v1, · ·· , Vj - 1, w , v 1+1 ,· ·· , vs)-
We set TJ V = (V* )*, Tf V = V* and rgv = JR. If V = TpM, then we denote
the set of all (r, s)-tensors by r;M(p).
Since (V*)* ~ V, we identify v EV as an element of TJV with the action
v(w) = w( v) for all w E V*. With pointwise addition and scalar multiplica-
tion, r;v is a real vector space.
Exercise 4.5. Let V = R 3 . Define T : V* x V x V----> R by T(w , v, w) = w(v x w),
for w E (R 3 )* , v , w E R 3 , and where v x w denotes the cross product in R 3 of
v, w. Show that Tis a (1, 2)-tensor on R 3 .
Exercise 4.6. (Tensor product of tensors). Let V be a real vector space. Let T be
an (r, s)-tensor on V and r' be an (r', s')-tensor on V. Define their tensor product
T © 'T 1 : (V* r+r' X v s+s'-: R by
(T@T')(w\··· , wr+r,V1, ··· , Vs+s')
= T(w\· ·· ,w: v1,· ·· , Vs )T1 (wr+1, ... ,wr+r:Vs+1,··· ,Vs+s')-
Then T © r' is an (r + r', s + s')-tensor on V. Prove that the tensor product©
is associative.
1 The stress in continuum mechanics is an example, and the word 'tensor' comes from
the La tin word 'tendere ', meaning 'stretch'.
Tensor fields 65
Exercise 4. 7. Let V be an m-dimensional real vector space, with a basis

{e1, • • • , em}, and let {1::1, ••• , €m} be the corresponding dual basis for the dual
vector space V*. Show that the set
B= {ei 1 ® · · · ® eir ® €i 1 ®···®€is : 1 ~i1, · · · , ir,j1, · · · ,j. ~m}
is a basis for r;v. Conclude that dimT;V =mr+•.
We will now show that every (r, s)-tensor field T can be viewed as a map
M 3p,....... (p,T(p)) E LJT;M(p) = {(p,T) :pE Mand TET;M(p)}

pEM
= LJ ({p}xT;M(p)),
pEM
where T(p) Er; M(p) is an (r, s)-tensor on TpM. Before giving the defini-
tion of the (r, s )-tensor T (p) arising from a tensor field T, we will prove the
following result.
Theorem 4.1. Let T Er; M, and let p EM. Suppose that:

• n 1 ' ... 'nr E rp M and 8 1 ' ... ' er E rp M are such that
(ni)p = (8i)p, i = 1, · · · , r
• Vi, · · · , Vs E TJ M and W1, · · · Ws E TJ M are such that
(½)v=(Wj)p, j=l,··· ,s.
Then (T(i11,•·· ,W,Vi,··· ,Vs))(p) = (T(81,··· ,er,W1 ,··· ,Ws))(p).
Proof. Let (U, x) be an admissible chart containing p. Let x be a bump

function that is identically 1 in a neighbourhood of p, and is identically
zero outside an open set U0 , such that U0 c U. Then2 xdxi E Tf M for
1 ~ i ~ m, and XOxi E TJ M for 1 ~ j ~ m.
For ease of exposition, we will prove this just in the case when3 we have
r = S = 1. First, x2 n = x 2 n(oxk)dxk = n(xoxk)(xdxk) = /k(xdxk), with
/k := n(xoxk) E C 00 (M). Similarly, we have that x2 8 = h(xdxk), where
fk:=8(xoxk) E C 00 (M). We note that
/k(p) = (n(xoxk))(p) = (n)p(XOxk)p = (n)v(x(p)Oxk,p)
= (n)p(lOxk,p) = (n)p(Oxk,p) = (8)p(Oxk,p)
= (8)p(loxk,p) = (8)v(x(p)Oxk,p) = (e(xoxk))(p) = fk(p).
2 Here by xdxi, we mean the I-form n defined by Oq =x(q)(dxi)q if q EU, and Oq =0
if q ¢ U. The notation xi'),,; has a similar connotation.

3 The proof in the general case is the same, except notationally messier, and the factor
x2 <1 +1 ) =x 4 , used below in (4.5), should be replaced by x2 <r+•l.
Moreover, we have x 2 V = x2 V(x')c'\£ = V(xx')(xoxt) = l(xoxt), where

l := V(xx') E C 00 (M). Similarly, x 2 W = W(xx')(xoxt) = g'-(xoxt),
where 'g'- := W(xx') E C 00 (M). Note that
g'-(p) = (V(xxt))(p) = Vv(xx£) = 1 Vvx£ +O
= 1 Wvx£ + O= Wv(xx£) = (W(xxt))(p) = t(p).
So
x 2 C1 +1lr(n, V) = T(x 2 n, x 2 V) (4.5)
= T(/k(xdxk),l(xoxt))
= /kg'-T(xdxk, Xllxt).
Similarly, x 2 C1+ 1lT(0, W) = fkg£T(xdxk, Xllxt). Consequently,
(T(O, V))(p) = 1 (T(O, V))(p) = (x(p)) 2 C1 +1l(T(O, V))(p)
= (x2c1+1Jr(n, V))(p)
= (Jkg'-T(xdx\xoxt))(p)
= /k(p)/(p)(T(xdx\xoxt))(p)
= fk(p)t(p)(T(xdx\xoxt))(p)
= • • • (retrace the steps) • • •
= (T(e, W))(p).
In light of this result, we can now introduce a 'point evaluation' of a tensor
field as follows.
Definition 4.3. (Evaluation of a tensor field at a point).

Let TE r; M, and p EM. We define T(p) Er; M(p) as follows: For all
w 1 , • • • , wr E (TpM)* and all v1 , • • • , Vs E TpM, we set
T(p)(w1,--- ,wr,v1,··· ,vs)= (T(01,--- ,nr,Vi,··· ,Vs))(p),

where 01, • • • , nr E Tf M are any 1-form fields such that (Oi)p = wi, for all
1 ~ i ~ r, and Vi, • • • , Vs E TJ M are any vector fields such that (½ )p = Vj,
for all 1 ~ j ~ s.
By Theorem 4.1, the notion above is well-defined, since it depends neither
on the choice of the 1-form field extensions of the 1-forms w1 , • • • , wr, nor
on the vector field extensions of the vectors v1 , • • • , Vs- It is also clear that
T(p) is an (r, s)-tensor: additivity is immediate, and we note that constants
can be considered as constant C 00 (M) functions.
Tensor fields 67
Remark 4.1. (Tensor field as a smooth section of a tensor bundle).

One can use the language of bundles in order to define an (r, s)-tensor
bundle (analogous to the tangent/cotangent bundle), where to each point
p E M, one assigns in a smooth manner an (r, s)-tensor. Then a tensor
field can be defined as 'section' of this (r, s)-tensor bundle, and this notion
coincides with our notion of tensor fields from Definition 4.1. *
T(pJ
---~------ M
p
Proposition 4.1. Suppose for each p in the smooth manifold M, there

exists a tensor T(p) Er; M(p) such that for all S1 1, ••• , nr ETf M and all
Vi,··· , \1s E TJ-M, the map
M 3 p ~ (T(fl 1 , .. · , nr, Vi, .. · , Vs))(p)
:= T(p)((fl 1 )p, · · · , (W)p, (Vi)p, · · · , (\1s)p)
is an element of C 00 ( M). Then T E r; M.
Proof. For ease of exposition, we just consider the case when r = s = l. We
only need to show the C 00 ( M)-multilinearity of T: Tf M x TJ- M-> C 00 ( M).
For S1 1 , S1 2 E Tf M, VE TJ- M, and f E C 00 (M), we have for all p EM that
(T(fl 1 + Jfl 2 , V))(p) = T(p)((fl 1 + f fl 2 )p, Vp)
= T(p )( (fl 1 )p + f (p )(fl 2 )p, Vp)
= T(p)((fl 1 )p, Vp)+ J(p)T(p)((S1 2 )p, Vp)
= (T(S11, V))(p) + (J-T(S1 2 , V))(p)
= (T(fl 1 , V)+ f ·T(fl 2 , V))(p).
Thus
T(fl 1 + Jfl 2 , V) = T(fl1, V) + J(T(fl 2 , V)).
Similarly, for fl E Tf M, Vi, Vi E TJ-M, and f E C 00 (M),
T(fl, Vi+ !Vi)= T(fl, Vi)+ J(T(fl, Vi)).
4.3 Tensor field components, tensor product and contraction
Just like a vector field and a 1-form field have chart-induced components
that represent the field at hand, we can represent a tensor field using its
chart-induced components.
Definition 4.4. (Chart-induced tensor field components at a point).
Let (U, x) be an admissible chart for an m-dimensional smooth manifold
M, p E M, and T E r;
M. We define the chart-induced tensor components
of Tat p by
T(p)t:t = T(p)((dxi 1 )p, · · ·, (dxir)p, oxii,p, · · · , Oxis,p),
where 1 ~ i1, · · · ,ir,J1, · · · ,Js ~ m, where mis the dimension of M.
Exercise 4.8. Let (U,x) be an admissible chart for a smooth manifold M, and
TE r; M. The mr+s chart-induced tensor field components TJ;.".".t are the maps
U 3 p >-+ T(p)t::t Show that TJ;.".".t E C 00 (U).
Exercise 4.9. Let T E Tf M be the (1, 1)-tensor field on a smooth manifold M,
given by T(n, V) = nv for all n E Tf Mand all VE TJM. Suppose that (U,x)
is an admissible chart for M. Determine the chart-induced components of T.
Exercise 4.10. (Transformation rule for components).
Let M be a smooth manifold and T E r; M. Prove that if (U, x) and (U', x') are
two overlapping admissible charts, then the chart-induced tensor field components
T 1i1
ii ···ir
···is, Ti1 ···ir
ii ···is . h
wit respect to (U' , x ') , (U, x ) , respective
. 1y, trans £orm as £o11ows.
For pEU n U',
Cl li1 Cl lir c, i~ c, i~ 1 1
T'i1 •··ir ( UX ( ( ))
) UX ( ( )) UX ( '( )) UX ( '( ))Til ···ir ( )
ii ···is = OXi~ X p ... OXi~ X p ox'il X p ... ox'is X p i~ ···i~ p
p
0Xi / O(XiOXl-l) / 0X 1i O(x'iox-l)
where -;;--,. (x (p)) := 0 . (x (p)) and ~(x(p)) := 0 . (x(p)).
ux' u' ux' u'
Exercise 4.11. (Physicist's definition of a tensor field).
Let M be a smooth manifold described by an atlas A. Suppose that for each chart
(U,x) in A, there exist functions TJ;.".".t;u E C 00 (x(U)), such that the following
transformation rule holds for any two overlapping charts (U, x) and (U', x'):
···ir;U' (x'( ))
Ti1
•_-·Js
Jl p . ., .,
0X/t 1 0Xlir OX11 I OX1s I i~ •··i~;U
= - . , (x(p)) • • • -;-:,(x(p)),;--;-:-(x (p)) • • • ,;--;-:-(x (p))T., ...., (x(p)).
ox'1 ux'r ux J1 ux Js Ji 1s
Let T be given as follows: For p E (U, x), 01, · · · , nr E Tf M, Vi,··· , Vs E TJ M,

(T(01,--- ,nr,Vi,··· ,Vs))(p)
= TJ;::_J;;U (x(p))-(0 1 )p(o,,;i ,p) ... (nr)p(Oxir,p) · (dxi 1)p((Vi)p) ... (dxis)p((Vs)p)-
Then prove that T is well-defined, that is, it does not depend on the chosen chart
containing the point p E M, and that T E r; M.
Tensor fields 69
An admissible chart has the nice feature that there are available chart-
induced 1-form fields dxi and vector fields Oxi. This suggests that we
could use the tensor field component functions of T to analyse the 're-
striction TI u' to the chart U by defining, for example in the r = s = 1 case,
Tlu(nidxi, ViOxi) = TJn;Vi, where n = nidxi E TfU, V = ViOxi E TJM.
But if U is the intersection of two admissible charts, then on U there are
two available coordinate chart maps, say x and x'. Let us show that the
above is well-defined, that is, it does not depend on the choice of the chart
map: We have
So we give the following definition.
Definition 4.5. (Restriction of a tensor field to an admissible chart).

Let M be a smooth manifold. Let TE T;M. Let (U,x) be an admissible
chart for M. We define the restriction Tiu E T;u of T to U by
Tl u (d X k1 , · · · , d X kr ,uxe
::i
1 ,···
::i
,u£•
) ._
.-
Tk1·•·kr
£ •.• £.,
1
and by extending C 00 (U)-multilinearly, that is, given ni = nt dxk E TfU

(l~i ~r) and Vj = Vf oxe E TJU (l~j ~s),
Tlu(nl . . . nr Vii . . . V.)
' ' 8
'
:= nkl ... nrk
' ' 1 r
v,1£1 ... vt· T:1 ·-~kr.
8 .c.1 •·•.c.s
The following result is key to the evaluation of tensor fields in practice,

for example when the chart-induced components of the various objects are
known.
Proposition 4.2. Let M be a smooth manifold, T E T; M, (U, x) be an

admissible chart, p E U, w 1 , • • • , wr E (TpM)* and v 1 , • • • , Vs E TpM. Set
wf = wk(Oxi,p), (l~k~r, l~i~m), and
vi= (dxj)pVR, (l~e~s, l~j~ m).
If T(p)t.::~J. are the chart-induced components of T at p, then
T(p) (w 1 ' •••

'
wr ' V1 • • •
' '
V8 ) = T(p)i-Jl1... ,i,: w1
,'"' ,Js i1
• • • wr:
tr
v31° 1 • • • vi•
S
= Tlu(p)(w1,··· ,wr,v1,··· ,vs)-

Proof. For ease of exposition, we show this just for the case of a (1, 1)-
tensor field. Let x be a bump function around p, that is, a C 00 (M) element
that is identically 1 in a neighbourhood of p, and identically zero outside
Uo, with Uo c U. Then we have
T(p)(w,v) = (T(xwidxi,xvi Dxi))(p) = wivi(T(xdxi,XOxi))(p)
= WiviT(p)((dxi)p, Oxi,p) = WiviT(p);.
Also, if we define O=widxi E rpu and V =vi Oxi E TJU, then we have
Tlu(p)(w,v) = (Tlu(O, V))(p) = (Tlu(widxi,vi Dxi))(p)
= Wivi(Tlu(dx\oxi))(p) = WiviTf(p) = WiviT(p);.
In order to get the representation of the tensor field using its chart-induced
components, we will produce a (C 00 (U)-module) spanning set for r;u using
the building blocks {dx 1 , • • • , dxm} and {Oxi, • • • , Ox=}. To this end, we first
introduce the notion of the tensor product of two tensor fields.
Definition 4.6. (Tensor product of tensor fields).

Let M be a smooth manifold, T E r; M, T' E T;,' M. The tensor product
+
T®T' E r;+;, M of T and T' is defined as follows: for 0 1, • • • , nr+r E rp M
I I
and for Vi,··· , Vs+s' E TJ M,

(T®T')(O\··· ,nr+r', Vi,···, Vs+s•)
:= T(0 1, · · · , nr, V1, · · · , Vs)· T'(nr+l, · · · , nr+r', Vs+1, · · · , Vs+s' ),
where the • on the right hand side is multiplication in C 00 (M).
Exercise 4.12. Let M be a smooth manifold, T E r; M, T' E r;,' M. Recall
Exercise 4.6, where we had defined the tensor product of tensors. Show that for
allpEM, (T®T')(p)=T(p)®T'(p).
The tensor product is 'bilinear': For T, TE r; M, s, s ET;,' M, f E C 00 (M),

(T+ JT) ®S = T®S + J(T® S) and
T® (S+ f S) = T®S + f(T® S).
In particular, f (T ® S) = (! T) ® S = T ® (! S). It can be seen that the
tensor product is associative.
By the associativity of ®, if (U, x) is an admissible chart, then, even
without the use of parentheses, Oxii ® · · · ® Oxir ® dxi 1 ® · · · ® dxi• makes
sense as an element ofT;u. We show that these can be used to decompose
any tensor field locally in a chart using its chart-induced components.
Tensor fields 71
Let 01, ···,or E TfU and V1, · · ·, V,, E TJU. Then Oi=Oidxk (1,:;;;i ,,;;r)
and ½ = vfOxe E TJU (1 ,:;;; j ,,;; s), where ot
= Oi Oxk = Oxkoi and
v~/= Vj(x') = (dx1)Vj. Thus
Tlu(0 1 , ... , or, Vi, ... , Vs)
= T!l """!r 0~i1 ... 0'.'Lr v,li1 ... vis
Jl •·•Js 8
= T}:::_j:(oxi101) · · · (oxirOr)((dxi 1)V1) · · · ((dxi•)Vs)

1 """!r (o i1 '°' ... '°' i} i '°' dxi 1 '°' ... '°' dxi•) (01 . . . 0
= T!Ji···Js x \Cl \Cl x r VY \Cl \Cl , , r, Vii , . . . , V.)
s
Exercise 4.13. (Noncommutativity of®).

Consider R 2 with the standard smooth structure given by the global chart
(R 2 , (x, y) >-+ (x, y) ). Calculate (dx ® dy )( Ox, Oy) and (dy ® dx) ( Ox, Oy ).
Exercise 4.14. Let M be a smooth manifold, TE r; M, SE r;,' M, and (U, x) be

an admissible chart for M. Show that the chart-induced components of T®S are
· b (T '°' S)i' ···ir+r' _ Ti' ···ir Sir+l ···ir+r'
given y \Cl i1 ···is+s' - i1 ···is is+1 ···is+s' •
The tensor product produces new tensor fields from existing ones. We end
this section by learning about yet another operation, called 'contraction'.
We begin with the following crucial result, which gives an intrinsic function
associated with a (1, 1)-tensor field, akin to the trace of a linear transfor-
mation T : V ---+ V on a finite-dimensional vector space V.
Proposition 4.3. Let M be a smooth manifold, and T E Tf M. Define

CT : M ---+ JR by (CT)(p) = T(p)((dxi)p, Oxi,p), where (U, x) is any admis-
sible chart containing p E M. Then CT is well-defined, that is, it does not
depend on the choice of the admissible chart. Moreover, CT E C 00 (M).
Proof. With a different admissible chart (U', x') containing p, we have
The smoothness of CT follows from Exercises 4.8 and 1.16. □

Now we are ready to define the operation of contraction. Given an (r, s)-
tensor field T with r and s both at least 1, we note that if we fill in all
the I-form field slots except for one, and all the vector field slots except
for one, then we are left with a (1, 1)-tensor field. We can then apply the
contraction map C to this. The result is a C 00 function. But clearly this
depends C 00 (M)-multilinearly on the initially chosen I-form fields and the
vector fields. Thus we have a (r -1, s - 1)-tensor field. We give the precise
formulation below.
Definition 4. 7. (Contraction).
Let M be a smooth manifold. Let 1 ~ i ~ r, 1 ~j ~ s. Let TE TJ M. Define
the contraction C}T E r;~l
M with respect to the i and j indices by
(C;T)(0 1 , · · · , nr-l, V1, · · · , Vs-1)

= C(T(01, ... 'ni-1, ·, Oi+1, ... nr-1, Vi, ... '½-1, ·, ½+1, ... 'V,,)),
for all 01, · · · , nr-l E Tf M and all V1, · · · , Vs-1 E TJ-M.
Exercise 4.15. Let M be a smooth manifold. Let T E Tl M, and (U, x) be an
admissible chart for M. Show that C~T has the chart-induced components given
by (C~T)jk=Tjik·
Exercise 4.16. Consider M = Rm with the standard smooth structure given by the
global chart (Rm, id). Let T= [Tj] be an mxm matrix. Then matrix multiplication
of column vectors in Rm by T induces a (smooth) map T : Rm -+ Rm. Let the
tensor field TE T{M be defined by (T(O, V))(p) = Or(p)((dT)pVp) for p EM.
Show that CT= trace(T).
The contraction operator can be applied repeatedly. For example, if we

have TE T:j M, then C(C~T) is an element of C 00 (M), and in an admissible
chart (U,x), C(C~T)lu=TJf.
Exercise 4.17. Let M be a smooth manifold. How many different contractions
does tensor field TE T(f M have in general?
Exercise 4.18. Let M be a smooth manifold. Suppose that r, s ~ 1, 1 ~ i ~ r,

and 1 ~ j ~ s. Show that c; :
T; M -+ T;~l is C 00 (M)-linear, that is, for all
T, SE TJ M, and all f E C (M), C}(T+ f S) = (C}T) + JC}S.
00
4.4 Pull-back of (0, s)-tensor fields
Recall that given a smooth map f : M -+ N between two smooth manifolds

M and N, any I-form field n E Tf N can be pulled back under f to a I-form
Tensor fields 73
field f*D. E Tf M. Now we will see that this operation of pulling back can
be extended to (0, s)-tensor fields on N.
Definition 4.8. (Pull-back of (0, s)-tensor fields).

Let M, N be smooth manifolds, and let f : M -+ N be a smooth map. For
TE T~ N, the pull-back f*T ET~ M of T under f is defined by
for all p E M, and all Vi, · · · , V,, E TJ M.

Let us check that f*T is well-defined. First, for each p EM, the map send-
ing v1, · · · , Vs E TpM to T(J(p))(dfp(v1), · · · , dfp(vs)) is IR-multilinear, and
thus it defines a (0, s)-tensor in T~ M(p). Next, we will use Proposition 4.1
to show that f*T is a tensor field. We need to show that for all vector fields
Vi,··· ,V,, E TJM, the map p,..... T(J(p))(dfp((Vi)p),··· ,dfp((Vs)p)) is
smooth. We check this locally around a point p by making use of a admissi-
ble charts (U, x), (W, y), for M, N, respectively, containing the points p and
f(p), respectively, and such that J(U) c W. We decompose½= ½(xi)Oxi
(l~j~s), and T=Tii·••j.dyi 1 ®···®dyi•. Then for q EU, we have
((J*T)(Vi, · · · , Vs))(q)
= T(J(q))(dfq((Vi)q), · · · , dfq((V,,)q))
= Tfr··J.(J(q))·
(dy 11 ® · · · ® dy1•)(J(q))(dfq((Vi)q(xii) oxii,q), · · · , dfq((V,,)q(xi•) oxi• ,q))
= Tfr··J.(J(q)) · (Vi)q(xi 1 ) • • • (Vs)q(xi•)·
. . £ £
(dy 11 ® · • • ®dy1•)(J(q))(oxi1,q(Y 1of)oye1,f(q), 0 0
Oxis,q(Y •of)oye•,f(q))
0
,
= Tfr··Js (J(q)) · (V1)q(Xi 1 ) • • • (Vs)q(xi•) · Oxi1,q(Y11 Of)··· Oxis ,q(y1• Of),

and so U 3 q,..... ((J*T)(Vi, · · · , Vs))(q), being the product of smooth func-
tions, is smooth in U, as wanted.
Exercise 4.19. Let IR.2 be equipped with the standard smooth structure, and let
the open subsets
M = {(r,0): r>O, -f <0<f},
N = {(x,y): x>O}
be equipped with the induced smooth structure. Let f : M --+ N be the smooth
map f(r,0) = (rcos0, rsin0) for all (r,0) EM. Let TE Tf N be the (0,2)-tensor
field given by T= ~dy ® dy in the chart (N, (x, y) >-+ (x, y)). Determine f*T.
Chapter 5
Lorentzian man ifolds
So far, we have introduced notions using only the smooth structure on the
manifold. But the smooth structure is still rather flexible. Within this
picture, two spheres in JR 3 of different radii would be considered the same,
and even the surface of a potato would be the same smooth manifold as a
sphere. In this sense the smooth manifold has not yet acquired a 'shape'.
We now introduce the notion of a metric, which is roughly speaking, a
(0, 2)-tensor field g on a manifold M, whose evaluation at any point p EM
equips the tangent space TpM with a scalar product g(p) (analogous to
the dot product of vectors in the plane). This structure allows us to assign
lengths to tangent vectors and angles between them. It will also enable us
to assign lengths to curves.
In the mathematical subject of Riemannian geometry, one studies
smooth manifolds equipped with a metric, which in each tangent space
TpM gives an 'inner product' g(p), namely a 'positive definite' scalar prod-
uct. On the other hand, in spacetime physics, the metric g on spacetime
is such that the scalar product g(p) on TpM at each p EM is 'indefinite',
and in fact 'Lorentzian', which means that the scalar product in each tan-
gent space TpM has the 'index 1' (roughly speaking, there exists a vector
v E TpM so that the scalar product g(p)(v,v) < 0, and the restriction of
g(p) to the 'orthogonal complement' v1_ c TpM of v is positive definite).
We will see what this means precisely, in the course of this chapter. This
structure allows us to talk about observers, the observer's instantaneous
perception of space, and of time. We can then introduce the notion of
geodesics, which are, loosely speaking, the straightest curves. We will also
see in later chapters that the matter content in the spacetime essentially
'curves' spacetime, and determines a tensor field derived from metric g.
Test matter moves along geodesics in this curved spacetime.
75
5.1 Scalar product
We begin by introducing the notion of a scalar product on a real vector

space V. Later on, the vector space V will be replaced by particular tangent
spaces TpM, for points p belonging to a smooth manifold M. So we will
always consider only finite-dimensional, real vector spaces V.
Definition 5.1. (Scalar product, inner product).
Let V be a real vector space. A map g : V x V -+ JR is a scalar product on
V if it satisfies:
• (Symmetric) For all v, wEV, g(v, w) = g(w, v).
• (Bilinear) For all u, v, w EV and all c EJR, g(u+cv, w) = g(u, w)+cg(v, w).
• (Nondegenerate) If vEV is such that for all wEV, g(v, w) =0, then v=0.
Let Ube a subspace of V. A symmetric, bilinear map g : V x V-+ JR is
• positive definite on U if for all u E U\ {0}, g( u, u) > 0.
• negative definite on U if for all u E U\ {0}, g( u, u) < 0.
• an inner product on V if g is positive definite on V.
If g is positive (or negative) definite on U, then g is nondegenerate on U.
Example 5.1. (Euclidean inner product).

Let V = JR 2, and define g : JR 2 x JR 2 -+ JR by
g(v1, v2) = x1x2 + Y1Y2 for all v1 = (x1, Y1), v2 = (x2, Y2) E JR 2.
Then g is an inner product on JR 2 . Let q denote the 'quadratic form'
JR 2 3 v i--+ g(v, v). The only vector v E JR2 such that q(v) = 0 is v = 0.
The level sets {(x, y) E JR 2 : x 2 + y 2 = c} of the quadratic form q are circles,
as shown on the left in the following picture (p. 77). ◊
Example 5.2. (Minkowski scalar product).

Let V = JR 2, and define g : JR 2 x JR 2 -+ JR by
g(v1, v2) = x1x2 - t1t2 for all v1 = (x1, ti), v2 = (x2, t2) E JR 2.
We will check that g is a scalar product. But g cannot be an inner product,
since, for example, g(e 2,e2) = -1 < 0, where e 2 = (0,1). The bilinearity
and symmetry of g are clear. Moreover, if v = (x, t) E JR2 is such that
g(v, w) = 0 for all w E JR 2, then taking in particular w = e 1 and w = e 2,
we obtain x = 0 and t = 0, so that v = (x, t) = (0, 0) = 0. This shows
that g is nondegenerate. The level sets {(x, t) E JR2 : x 2 - t 2 = c} of the
quadratic form JR 2 3 vi--+ g(v, v), are hyperbolas, as shown on the right in
the following picture, with the degenerate case c = 0 giving the two lines at
an angle of 45° with the coordinate axes. ◊
Lorentzian manifolds 77
Euclidean plane Minkowski plane
Exercise 5.1. Let V be a finite-dimensional vector space with the scalar product
g. Let V* be the dual space of V . Show that the map V 3 v >-+ v" := g(v, •)EV*
is a linear isomorphism.
Definition 5.2. (Index of a scalar product).

Let V be a finite-dimensional vector space with the scalar product g. Define
the set Ng = {U : U is a subspace of V, and g is negative definite on U}.
The index of the scalar product g is defined by ig = max dim U.
UENg
If Ng= {O}, where O denotes the subspace {O}, and OE Vis the zero vector,
then ig=O.
Definition 5.3. (Orthogonal vectors, orthogonal complement).

Let V be a finite-dimensional vector space with the scalar product g. Two
vectors v, w E V are said to be orthogonal if g( v, w) = 0. Let Ube a subset of
V. Define the orthogonal complement of U, denoted by U l., by the subspace
u1. := {v EV : g(u, v) = 0 for all u EU}. In the case that U = span {v} for
a vector v E V, we denote U l. simply by vl..
Theorem 5.1. Let V be an m-dimensional vector space, and g be a scalar

product on V. Let U be a k-dimensional subspace of V. Then U l. is a
subspace of dimension m - k.
Proof. Set U' = {w E V* : w( u) = 0 for all u E U}. Let us first show that
U', which is a subspace of V* has dimension m - k. Let {e 1 , · · · , ek} be
a basis for U, and complete it to a basis {e 1 ,··· ,ek,ek+1,··· ,em} for V.
Let {E1 , • • • , Em} be the corresponding dual basis for V*, that is, Ei ( ej) = 8}
for all i, j. Clearly, Ek+l, • • • , Em E U'. Now let w E U', and decompose
w=a1E 1 + · · · + amEm. As 0=w(e1) = · · · =w(ek), we get 0=a1 = · · ·=ak.

So w E span{Ek+l, • • • , Em}. Consequently U' = span{Ek+l, • • • , Em}, and
dimU' =m - k. Recalling the isomorphism from Exercise 5.1, we see that
v E u1- ~ [g(v,u)=0 for all u EU]
~ [vi> (u) = 0 for all u E U]
~ [vi> EU'].
Thus the map U 1- 3 v >-+ vi> E U' is an isomorphism, and so it follows that
dimU1_= dimU'= m - k. □
Exercise 5.2. Let V be a finite-dimensional vector space with the scalar product
g. Let Ube a subspace of V. Show that U_j__j_ = U.
Theorem 5.2.
Let V be an m-dimensional vector space, and g a scalar product on V.
Let U be a k-dimensional subspace of V. Then the following are equivalent:
(1) V=U+U1_ andUnU1-={0}. (WewriteV=UEBU1-.)
(2) g : U x U ---+ JR is nondegenerate, i.e., the restriction gluxu is nonde-
generate.
Proof.
(1) ~ (2): If u E U is such that for all u' E U, g(u, u') = 0, then we have
(U3)uEU1-. SouEUnU1-={0}.
(2)~(1): Ifu EU n u1-, then for all u' EU, g(u,u')=0. As g: U x U---+ JR
is nondegenerate, u=0. So Un u1- = {0}. We have dimU1_ =m-k. Let
B= {e1, · · · , ek} be a basis for U, and B' ={Ji,··· , fm-k} be basis for u1-.
As Un u1- = {0}, it follows that B=B u B' is independent, and since there
are m elements in B, span B = V. So U + U 1- = V. □
In particular, if U E Ng, then V = U EB U 1-.
Theorem 5.3. Let V be a finite-dimensional vector space, and g a scalar

product on V. If U E Ng and dim U = ig, then g : U 1- x U 1- ---+ JR is positive
definite.
Proof. Suppose that v E u1-\{0} is such that g(v,v)~0.

Let us show we can assume with no loss of generality that g(v, v) < 0.
Suppose that g(v, v) =0. Since v#0, and as g is nondegenerate on V, there
exists aw EV such that g(v,w)#0, and we can choose1 w E u1-. For all
1 As U E N 9 , we have V = U EE) U _j_. So by decomposing w = w' + w", w' E U, w" E U _j_,
we have that O #- g(v, w) = g(v, w') + g(v, w") = 0 + g(v, w").
.A E JR, we have V>, :=.AV+ w E u1-, and

g(v>., V>.) =.A 2 g(v, v) + g(w, w) + 2..\g(v, w) =0 + g(w, w) + 2..\g(v, w).
As ,\ >-+ g(V>., V>.) a linear function of,\ with a nonzero slope, it is negative
for some ,\ E R
We now show g(v, v) <0 is impossible. Set U' = {u +av: a E JR, u EU}.
For u' =u + av E U'\{0},
g(u', u') = g(u, u) + a 2 g(v, v) + 2ag(u, v) = g(u, u) + a 2 g(v, v) < 0,
as both summands are nonpositive, and u' =I= 0 (which implies u =I= 0 or
a =I= 0). But then U' E N 9 , contradicting the definition of i9 , since we have
that dim U' = dim U + l > dim U = i9 • □
We recall the Gram-Schmidt orthonormalisation procedure for producing

an orthonormal set out of an independent set, while preserving the span.
Proposition 5.1. (Gram-Schmidt orthonormalisation).
Let V be a vector space, and let g be a scalar product on V. Let U be a
subspace of V such that g is positive definite on U. Let {u1, · · · , Uk} be an
independent set of vectors in U. With llull :=,Jg(u,u) for u EU, set
_ u1 _ UR. - g(ut, e1)e1 - · · · - g(ut, et-1)et-1 1 2 n k
e1 - - - , et - - - - - - - - - - - - - - - - ;or ~{,~ .
llu1 I llut - g( Ut, e1)e1 - · · · - g(Ut, et-1)et-1 I
Then {e1, · · · , ek} is an orthonormal set in U, that is, g(ei, ej) = 8ij for
l~i,j~k, and for all l~f,~k, span{u1,··· ,u£}=span{e1,··· ,el}.
Proof. We proceed inductively. If k = l, then, as {ui} is indepen-
dent, u1 =I= 0. Thus the vector e1 = ui/llu1II is well-defined, g(e1,e1) = 1,
and span{ui} = span{e1}. Suppose the claim holds for k = e for some e.
Let {u1, · · · , U£+1} be an independent set in U. Then also {u1, · · · , ui}
is independent. By the induction hypothesis, there exists an orthonor-
mal set {e1, · · · , el} such that span{u1, · · · , ui} = span{e1, · · · , el}. As
{u1, · · · , U£+1} is independent, U£+1 ¢ span{u1, · · · , ui} = span{e1, · · · , el}.
It follows that the vector v := U£+1-(g(ut+1, e1) e1 + · · ·+g(ut+l, et) et) =I= 0.
So et+l = v/llvll is well-defined. Trivially g(et+1,et+1) = 1. For all i ~ e,
g(v, ei) = g(ut+l, ei) - g(ut+l, ei)g(ei, ei) = 0, and so g(et+l, ei) = 0. Hence
{e1, · · · , et, et+1} is an orthonormal set. Finally, as
1
et+l = M(uH1 - I ),
espan{e1, · .. , et}
=span{u1, 00
,ui}
•
we have that
span{e1, · · · , et, et+d = span{u1, · · · , Ut, et+1} = span{u1, · · · , Ut, Ut+d-
Theorem 5.4. Each scalar product g on an m-dimensional vector space V

admits a basis {e1, · · · , em}, satisfying
- oii if i ~ ig,
g(ei,ej) = { Oij if i> ig.
Such a basis is called an orthonormal basis.
Proof. If U E Ng and dimU = ig, then we can choose an orthonor-

mal basis {e 1, • • • , e19 } with respect to the positive definite scalar product
- g : U x U-+ JR on U. Then we have g(ei, ei) = - Oij for all 1 ~ i,j ~ ig.
Moreover, we can construct an orthonormal basis {e19 +1, • • • , em} with re-
spect to the positive definite scalar product g: u1- x u1--+ JR on u1-, i.e.,
g(ei,ej)=Oij, for allig+l~i,j~m. As V=UEBU1-, {e1,··· ,em} is a
basis for V. □
Theorem 5.5. Let V be an m-dimensional vector space, and g a scalar

product on V. Let {e1, · · · , em} be a basis for V such that there exists an
i* E {0, 1, • • • , m} such that
_ ·) _ {- oii if i ~ i*,
g (e,, e1 - ;: "f . .
Uij 1 Z > Z*.
Proof. Let the subspace U E Ng be such that dimU = ig. Define the map
L: U-+ span{e1, • •· ,ei.} as follows: For u=uiei, Lu:=u 1e1 + • • • + ui*ei.·
Then L is easily seen to be linear. We now show it is injective. If Lu= 0,
that is, ui = 0 for i = 1, • • •, i*, then we have
m
g(u,u)=g(uiei,uiei)= ~ (ui)2 ~O.

i=i.+1
But since - g is positive definite on U E Ng, we can now conclude that
u = 0. Hence L is injective. It follows by the rank-nullity theorem that
ig=dimU =rank L + dimker £=rank L + 0 ~ dimspan{e 1, • • •, ei.} =i*.
Soig~i*. Letig<i*. LetW:=span{e 1,--•,e 19 }. ThendimW=ig. Also,
WE Ng, since if w=w 1e1 + • • • + w 19 e 19 , then
g(w,w)=-((w 1)2 + •·· + (w 19 )2)~0,
and g( w, w) = 0 only if w 1 = • •• = w 19 = 0, that is, w = 0. By Theorem 5.3,
g : w1- x w1- -+ JR is positive definite. However, e 19 +1 E w1- and as
ig + l~i*, g(e 19 +1,ei9 +1)=-l, a contradiction. Thus ig = i*. □
Definition 5.4. (Minkowski scalar product).

Let V be a finite-dimensional real vector space. A scalar product g on V,
with index i9 = 1 is called a Minkowski scalar product.
Besides the Minkowski plane considered earlier in Example 5.2, the follow-
ing higher-dimensional generalisation is an example of a vector space with
a Minkowski scalar product.
Example 5.3. Let(·,•) be the Euclidean inner product on Rm-l _ On the

vector space V =Rm= R x Rm- 1, define the scalar product 'T/ by
'Tf((uo, u), (vo, v)) = -uovo + (u, v).

Then 'T/ is a Minkowski scalar product on Rm. The standard basis vectors
form an orthonormal basis for Rm with respect to 'T/:
1 0 0
0 1 0
eo= O e1= O em-1=
0
0 0 1
In Example 5.5, we give many more orthonormal bases. ◊
The general case of a Minkowski scalar product can be reduced to the

above example by choosing an orthonormal basis {eo, e1, · · · , em-1} such
that g(eo, eo) = -1 and g(ei, ei) = 1 for 1 ~ i ~ m - 1. Then for
m-1 m-1
u = u 0 eo + "
LJ .
u'ei, v = v 0 eo + "
LJ .
v'ei, (5.1)
i=l i=l
with u := (u 1,··· ,um-l) E Rm-l, v := (v 1,··· ,vm-l) E Rm-l, we have
that g(u, v) = -u0 v 0 + (u, v).
Definition 5.5. (Timelike/spacelike/lightlike vector, causal vector).

Let V be a real finite-dimensional vector space, and g a Minkowski scalar
product on V. A vector v EV is called
• timelike if g( v, v) < 0
• spacelike if g( v, v) > 0
• lightlike if v =fa O and g(v, v) = 0.
A vector v E V is called causal if it is timelike or lightlike.
So V = {spacelike vectors} u {timelike vectors} u {lightlike vectors} u {O}.

Example 5.4. Let us revisit Example 5.3. We consider ~m as a topological

space with its usual Euclidean topology (this is the induced topology when
~mis considered as a smooth manifold with its standard smooth structure).
A vector (v 0 , v) E ~ x ~m-l is timelike if -(v0 ) 2 + llvll2 < 0. Thus v0 =I- 0,
and the collection of all timelike vectors with a fixed v 0 is an open ball
B(O , v0 ) = {v E ~m-l : llvll < lv0 1}. Hence the set of all timelike vectors
has two connected components C and -C, where
C= {(v 0 , v) E ~m: v0 > 0, llvf < (v 0 )2} (a solid open spherical cone).
We note that span( (-C) u C) = V and also that (-C) n C = 0. The set
of lightlike vectors is the set {(v 0 , v) E ~m : llvll = Iv 0 I =I- 0}. ◊
timelike
_ ____;_:;;..__----!_ spacelike
timelike
Exercise 5.3. Let V be a finite-dimensional vector space, and g a Minkowski scalar

product on V. Let u, v EV be two lightlike vectors. Prove that u, v are orthogonal
ifand only if there exists at E IR\{O} such that u = tv . Hint: Use the decomposition
(5.1), using an orthonormal basis for V . For the 'only if' part, use the equality
condition in the Cauchy-Schwarz inequality for the Euclidean inner product.
Lemma 5.1. Let V be a finite-dimensional vector space, and g a Minkowski

scalar product on V. Let v E V\{O} be orthogonal to a timelike vector u.
Then v is spacelike.
Proof. This follows from Theorem 5.3, by taking U = span{u}. Then as

u is timelike, U E N 9 , and moreover, dim U = l = i9 . So g : U l. x U l. ---> ~
is positive definite. As v E u1.\{0}, it follows that g(v, v) > 0, that is, v is
spacelike. □
Exercise 5.4. Let g be a Minkowski scalar product on a finite-dimensional vector

space V. On the set T of timelike vectors in V, define the relation ~ as follows:
For v,wET, v~w if g(v,w) <0. Prove that~ is an equivalence relation 2 •
Hint: For transitivity, proceed as follows. If v ~ w and w ~ x, then g( v, w) < 0 and
g(x,w)<0. So there is a k>0 such that g(v,w)=kg(x,w), i.e., g(w,v-kx)=0.
By Lemma 5.1, conclude v - kx is either O or spacelike, and in each case v ~ x.
Proposition 5.2. (Reversed Cauchy-Schwarz).

Let g be a Minkowski scalar product on a finite-dimensional vector space V.
Ifv,w EV are timelike, then (g(v,w)) 2 ~ g(v,v)g(w,w).
Equality holds if and only if v, w are dependent.
Proof. Let u = av - bw, a = g(v, w), and b = g(v, v) < 0. Then we

have g(u,v) =ag(v,v) - bg(w,v) =0. So u,v are orthogonal, and as vis
timelike, by Lemma 5.1, either u = 0 or u is spacelike. In either case, we
have g(u, u) ~0, i.e., a 2 g(v, v)+b 2 g(w, w) ~2abg(v, w) =2(g(v,w)) 2 g(v, v).
Dividing by g(v, v) < 0, we obtain (g(v, w)) 2 +g(v, v) g(w, w) ~ 2(g(v, w)) 2 ,
giving the desired inequality.
If equality holds, then a 2 =(g(v,w))2=g(v,v)g(w,w), and so
g(u, u) = a 2 g(v, v) + b2 g(w, w) - 2abg(v, w) = a 2 b + b2 g(w, w) - 2aba
= b2 g(w,w) -a2 b = (g(v,v)) 2 g(w,w)-g(v,v)g(w,w)g(v,v) = 0.
But, as observed above, u is spacelike or u = 0. But u is not spacelike since

g(u,u)=O. So u=0, that is, av-bw=O. Since b<0, v,w are dependent.
If v, w are dependent, then since each is nonzero, we have that v = tw for
some t E IR\{0}, and then
(g(v, w)) 2 = (g(tw, w)) 2 =t2 (g(w, w)) 2 = g(tw, tw) g(w, w) = g(v, v) g(w, w).
Exercise 5.5. (Reversed triangle inequality).
Let g be a Minkowski scalar product on a finite-dimensional vector space V. For
timelike w E V, define T(w) = ,J-g(w, w). Let u, v E V be timelike vectors such
that g(u, v) <0. Show that u +vis timelike and that T(u + v) ~T(u) + T(v). (In
Theorem 5.6, we will see that this inequality is related to the 'twin paradox'.)
Definition 5.6. (Lorentz transformation).

Let 'T/ be the Minkowski scalar product on !Rm given in Example 5.3. A
matrix A E !Rmxm is called a Lorentz transformation if 'T/(Au,Av) = 'T/(u,v)
for all u, v E !Rm. The set of all Lorentz transformations is denoted by
0(1,m -1).
2 We remark that in particular, ~ is transitive. After we discuss time-orientation on a
Lorentzian manifold, this will imply that if we have two future-pointing timelike vectors
at a point, then their scalar product will be negative.
Exercise 5.6. Let us denote by [77] E Rmxm the diagonal matrix with the entries
-1, 1, • • • , 1 along the diagonal. The entry in the ith row and jth column of [77]
(respectively A) is denoted by 11ii (respectively Aj ), the indices being counted
from 0, i.e., 0 ~ i,j ~ m-1. Show that the following are equivalent:
(1) A E 0(1, m-1).
(2) N[77]A = [77].
(3) 77ij = Af A}T/kl•
Exercise 5.7. Show the inclusion 0(1,m-l)cGL(m,R), that is, every Lorentz
transformation is invertible.
Exercise 5.8. Show that 0(1, m-1) is a subgroup of the group GL(m, R).
Exercise 5.9. Show that in R 4 , the following are Lorentz transformations. These
are examples of a 'boost', a spatial rotation, 'time-reversal', and a spatial reflec-
tion. Here ef>, 0 E R.
cosh ef> sinh ef>

[ sinhef> coshef> h
l
,
rl cos0 -sin01
sin0 cos0 '
1
Proposition 5.3. Let V be an m-dimensional vector space, and g a

Minkowski scalar product on V. Let {eo, · · · , em-d be an orthonormal basis
for V with respect tog such that g(eo, eo) = -1, g(ei, ei) = 1 for l ~ i ~ m-1.
Let B = {Jo,··· , fm-d be a basis for V, and let A be the transformation
matrix: Ji= AJei. Then the following are equivalent:
(l) Bis an orthonormal basis for V, g(/0 , Jo)= -1, g(f;, Ji)= 1 (1 ~i~m-1).
(2) A E 0(1, m-1).
Proof. Let [T/ij] be the diagonal matrix with the diagonal entries
-1, 1, • • • , 1. Then
g(fi, Ji)= g(A~ek, A;ee) =A~ A;g(ek, ee) =A~ A]T/kl· (5.2)
(1)=?(2): T/ij ~ g(fi, Ji) <5~l Af A]TJke, and by Exercise 5.6, AE 0(1, m -1).
(2)=?(1): g(fi, Ji)(~) Af A]TJke = %· The last equality is due to (2) and
Exercise 5.6. □
Example 5.5. In Example 5.3, we had seen that {eo, e1, · · · , em-1} forms
an orthonormal basis with respect to the Minkowski scalar product TJ on
!Rm. For any matrix A E 0(1,m -1), define the vectors /j = AJei for all
j = 0, 1, · · ·, m - 1. Then {/o, Ji,··· , fm-d is an orthonormal basis. We
note that Jo, Ji,··· , fm-1 are them columns of the matrix A. ◊
5.2 Semi-Riemannian, Riemannian, Lorentzian manifolds
Definition 5.7. (Metric, Semi-Riemannian/Lorentzian manifold).

Let M be an m-dimensional smooth manifold. A metric g on Mis a (0, 2)-
tensor field g E Tf M, such that there exists an i(g) E {O, 1, • • • , m} so that
for all points p EM, g(p) : TpM x TpM-+ JR is a scalar product with the
index ig(p) = i(g). We call i(g) the metric index.
The pair (M, g), where Mis smooth manifold, and g a metric on M, is
called a semi-Riemannian manifold. A semi-Riemannian manifold (M, g)
is said to be a
• Lorentzian manifold if i(g) = 1 and m ~ 2.
• Riemannian manifold if i(g) = 0.
-g, and i(-g) =m-i(g). So if i(g) =m-1 ~ 1,

If g is a metric, then so is
then up to a sign on the metric, we have essentially a Lorentzian manifold.
Similarly if i(g) = m, then we have Riemannian manifold. Throughout this
book, we will arrange i(g) = 1 for our Lorentzian manifolds, rather than
i(g) = m - 1 (although some literature uses this).
Exercise 5.10. Let (M,g) be a semi-Riemannian manifold and let V, VE TJM.
Suppose for all WE TJ M, g(V, W) =g(V, W). Prove that V = V.
Hint: Show pointwise equality using the nondegeneracy of g(p), and the ex-
tendibility of each w E TpM to a W E TJ M (Lemma 3.1, p.44). (This will be
used when discussing the 'Koszul formula' defining the Levi-Civita connection.)
Exercise 5.11. Suppose that (M, g) and (N, h) are semi-Riemannian manifolds.
We call a map f : M -+ N an isometry if f is a diffeomorphism, and f*h = g.
Show that Iso( M) = {J : M -+ MI f is an isometry} is a subgroup of (Diff( M), o)
(the group of diffeomorphisms; see Exercise 1.19, p.16).
Exercise 5.12. Consider R 3 as a smooth manifold with the standard smooth struc-
ture. Equip R 3 with the metric g, given in the chart (R3 , (x,y,z) >-+ (x,y,z)) by
g = dx ® dx + dy ® dy + dz ® dz. Determine g in the 'spherical coordinates' chart
(U, x) described in Exercise 2.15 (p.36). For t E R, let 'lf;t : R 3 -+ R 3 be the map
'lf;t(x, y, z) = (x, (cos t)y - (sin t)z, (sin t)y + (cos t)z) for (x, y, z) E R 3 . Show that
'lf;t is an isometry.
Example 5.6. (Minkowski spacetime). Let M be an affine space over a

4-dimensional real vector space V, which is made into a smooth manifold as
in Example 1.9. Suppose that V has been equipped with a Minkowski scalar
product g. Let {fo, f1, f2, f3} be any orthonormal basis for V with respect
to g, with g(fo, fo) = -1, g(fi, fi) = 1 (i = 1, 2, 3). Fix any p EM. This
induces an admissible chart covering all of M, and is called an inertial frame.
The chart map (x 0 , x1, x 2 , x 3 ) is such that for any q EM, q = p + xi(q)fi
(summation over i = 0, 1, 2, 3). Define the metric g by specifying its chart-
induced components:
gij = g(Oxi, Oxi) = T/ij,
where T/oo = -1, T/11 = T/22 = T/33 = 1, and whenever i =I= j, T/ij = 0. Hence
g(q): TqM x TqM-+ JR is given by
· v} 0x1,q
g(q)(ViO x-i,q, •1iJ·ViWj = -VOWO + v w + v w + v w ·
· ) = 'l'I· 1 1 2 2 3 3
By Proposition 4.1, it follows that g E TJ M. Hence g is a metric with

1,(g) = 1.
Furthermore, the metric is independent of the original choice of the
orthonormal basis and of the 'base' point p, and this can be seen as follows.
Let p EM, and let B = {fo, f1, f2, £3} be another orthonormal basis, with
g(f0 , £0 ) = -1, g(f;, £;) = 1 (i = 1, 2, 3). We denote the corresponding induced
chart map by x : M -+ JR4 • Then by Proposition 5.3, there exists a Lorentz
transformation A such that
fi = A;r;.
By Examples 1.7 (p.6) and 2.3 (p.28), if TqM 3 v= viOxi,q = 1Jiox;,q, then
1f = A;vi.
Thus if g is the metric obtained by the global chart induced by p and B,
then for v,w E TqM,
g(q)(v,w) = T/ij;;ji[;i = T/ijAivkA~w£ = T/k£Vkw£ = g(q)(v,w).
The Lorentzian manifold (M, g) thus obtained will be called the Minkowski
spacetime. ◊
For a surface M in JR3 , any tangent vector v E TpM can act on the three
smooth functions x,y,z: M-+ JR, and produce a vector (vx,vy,vz) in JR3 .
So if JR3 is equipped with a scalar product g (for example the Euclidean
inner product), then for p E M, define g(p) : TpM x TpM -+ JR by
g(p)(v, w) = g((vx, vy, vz), (wx, wy, wz)) for v, w E TpM.
Then g(p) : TpM x TpM -+ JR is bilinear, and symmetric. In particular,
g(p) is a (0, 2)-tensor on TpM. Using Proposition 4.1, it can be shown that
g E TJ M. Suppose that g(p) is nondegenerate on TpM for each p E M,
and that the index of g(p) is constant. Then g is a metric on M, called
the induced metric on M from (JR3 ,g), and (M,g) is a semi-Riemannian
manifold. We note that if we use the Euclidean inner product on JR 3 , then
g(p) will be positive definite on TpM for each p E M, and so (M, g) is a
Riemannian manifold in this case. We will now consider an example.
Example 5. 7. (Sphere).
Consider the sphere as a smooth manifold with the smooth structure given
by the atlas in Example 1.5 (p.5). We use the compatible atlas of spherical
polar coordinates described in Example 1.8 (p. 7). The image of the injective
map (0, 7r) x (0, 27r) 3 (0, <p) >-+ ((sin 0)( cos¢), (sin 0)(sin <p ), cos 0) E 8 2 c JR 3
is called U, and the inverse of this map is denoted by cp. Then (U, cp) is an
admissible chart. We equip JR 3 with the Euclidean inner product. Then the
chart-induced component g(p)00 of g at p E 8 2 , where cp(p) = (0 0, ¢ 0 ) is
g(p)00 =g(p)(o0,p, O0,p) = (O0,px) 2 + (O0,py) 2 + (O0,pz) 2
= ( o((sin 0) cos ) (0 ,I.. )~ 2 (o((sin 0) sin ) (0 ,I.. )~ 2 ( ocos 0 (0 ,I.. )~ 2
a0 o, '1'0 ') + a0 o, '1'0 ') + a0 o, '1'0 ')
= (cos 0o)2(cos ¢o) 2 + (cos 0o)2(sin¢o) 2 + (- sin0o) 2 = 1.
Similarly,
g(p)0ct, = g(p)ct,0 = g(p)(o0,p, oct,,p)
= (O0,px)(oct,,pX) + (O0,py)(oct,,pY) + (O0,pz)(oct,,pZ)
= (cos 0o)( cos </Jo)(sin 00)(- sin </Jo)
+(cos 0o)(sin </Jo)(sin 0o)( cos </Jo) + (- sin 0o)0
= 0,
g(p)ct,ct, =g(p)(oct,,p, oct,,p) = (oct,,px) 2 + (oct,,pY) 2 + (oct,,pz) 2
= (sin0o)2(-sin¢o) 2 + (sin0o)2(cos¢o) 2 + 02 = (sin0o) 2 •
Summarising, we arrange the components in a matrix, describing the metric
gin the chart (U, cp): If p E 8 2 is such that cp(p) = (0 0, ¢ 0 ), then
[ g(p)00 g(p)0ct,] [1 0 ]
g(p)ct,0 g(p)ct,ct, = 0 (sin0o) 2 ·
We see that this is a positive definite invertible matrix, as expected. The

chart U covers 8 2 except for the intersection of 8 2 with the half plane
{(x,y,z): y = 0, x~O}. We can take another chart (V,'¢), defined in a
similar manner, which, together with U, covers 8 2 : V covers 8 2 except for
the intersection of 8 2 with the half plane {(x, y, z) : z = 0, x ~ 0}, and is
the image of the map
(0, 7r) x (0, 27r) 3 (0, <p) >-+ (-(sin 0)( cos <p), cos 0, (sin 0)(sin <p)) E 8 2 c JR3 .
An expression for the component matrix of g(p) can be obtained in the
chart V in a similar manner as above, and in fact it so happens that we get
the same component matrix. We note that, as expected, the component
matrices in both charts are positive definite. We have that (8 2 , g) is a
Riemannian manifold. ◊
Exercise 5.13. Consider the surface Min l~.3, where M = {(x, y, z) E R 3 : z =xy},
as a smooth manifold with the smooth structure [A], where A consists of the
single chart M with the chart map M 3 p = (u, v, uv) >-+ ( u, v) E R 2 . Let R 3 be
equipped with the Euclidean inner product (·, •), and M be equipped with the
induced metric g from (R3 ,(·, •)). Determine the component matrix for g using
the chart-induced basis.
Example 5.8. (Cylindrical spacetime).

Consider the cylinder as a smooth manifold with the smooth structure given
by the atlas in Exercise 1.5 (p.5). It can be shown that the map
(0, 21r) xJR 3 (0, t) i--+ (cos 0, sin 0, t) E 8 1 xJR=:M c JR 3
is injective, and hence a bijection onto its image denoted by U, say. If we
denote the inverse of this map by cp, then (U, cp) is a chart, and it can be
checked that it is admissible.
We equip JR3 with the Minkowski scalar product 'T/, given by
'Tf((x1, Y1, z1), (x2, Y2, z2)) =x1x2 + Y1Y2 - z1z2
for all (x1, Y1, z1), (x2, Y2, z2) E JR3 . Then the chart-induced component
g(p)00 of g at p EM, where cp(p) = (0o, to) is given by
g(p)00 = g(p)(o0,p, 00,p) = (00,px) 2 + (00,py) 2 - (00,pz) 2
= (-sin0o) 2 + (cos00 ) 2 - 02 = 1.
Similarly,
g(p)0t = g(p)t0 = g(p)(o0,p, ot,p)
= (00,px)(ot,pX) + (00,py)(ot,pY) - (00,pz)(ot,pZ)
=(-sin0o)O+(cos0o)O-(O)1=O, and
g(p)u = g(p)(ot,p, at,p) = (ot,px)2+(at,pY) 2 - (ot,pz) 2
= 02 + 02 - 12 = -1.
Summarising, the (U, cp)-induced metric component matrix is given by
[ g(p)00 g(p)0t] [1 OJ
g(p)t0 g(p)tt = 0 -1 '
where p E U c M is such that cp(p) = (0 0, t 0 ). We see that this is an
invertible matrix. Hence g(p) is nondegenerate for all p E U. Also the
index is 1. In particular, (U, g) is a Lorentzian manifold. Taking V to be
the image of (-1r, 1r) x JR 3 (0, t) i--+ (cos 0, sin 0, t) E 8 1 x JR = M c JR3 , we
obtain another admissible chart (V, '¢), which together with U, covers M.
In a manner similar to the above, we can compute component matrix for
the induced metric on V from (R3 , r,), and in fact it turns out to be identical
to the one above for the chart (U, cp). From these component matrices, we
see that the metric has index i(g) = 1 everywhere on C, and so (M, g) is a
Lorentzian manifold. ◊
Example 5.9. (FLRW spacetime).

Consider the FLRW spacetime M =Ix R 3 , where I= (0, oo), as a smooth
manifold with the smooth structure given by the atlas comprising the single
chart (M, idixR.3 ); see Example 1.11. Let a: J-+ I be a C 00 function. For
any p EM, we have the basis B = {ot,p,ox,p,oy,p,oz,p} for TpM. Let g
respect to this B given by
[
g(p)tt g(p)tx g(p)ty g(p)tz
g(p)xt g(p)xx g(p)xy g(p)xz _
g(p)yt g(p)yx g(p)yy g(p)yz -
l [-1
be the metric on M which has the metric component matrix for g(p) with
0
0
(a(~)) 2
0
~ ~
(a(t)) 2 0
l '
g(p)zt g(p)zx g(p)zy g(p)zz 0 0 0 (a(t)) 2
where p= (t, x, y, z) EM. Then (M, g) is a Lorentzian manifold. ◊
Exercise 5.14. Let (M1,g1) be a Lorentzian manifold and (M2,g2) be a Rie-

mannian manifold. Show that the product M1 x M2 of these smooth manifolds
equipped with the metric g described below, makes (M1 x M2, g) a Lorentzian
manifold. First recall from Exercise 2.8, that for all (p1,p2) E M1 x M2, we have
T(p 1 ,p 2 J(M1 x M2):,, Tp 1 M1 x Tp 2 M2. Define the metric g by
g((p1,p2))(v1 EB w1, v2 EB w2) = g1 (p1)(v1, w1) + g2(p2)(v2, w2)
for v1 EB w1, v2 EB w2 E T(Pi ,p 2 l (M 1 x M 2). Here we use the notation v EB w from
Exercise 2.8, p.30.
Example 5.10. (Schwarzschild spacetime).

Recall from Example 1.12 (p.9), the Schwarzschild spacetime
M =Rx (2m, oo) x 8 2
as a smooth manifold. Let (U, cp) be the chart for 8 2 from Example 1.8 (p. 7).
For any p= (to, r 0 ,p0 ) EM, with t 0 ER, r0 E (2m, oo) and p0 EU c 8 2 , a
basis for TpM is given by the tangent vectors 3 Vt, Vr, v0, Vct,, given as follows.
For f E C 00 (M), using the notation from Exercise 2.8 (p.30),
Vt = Ot,t 0 EBO EBO, vtf = Ot,t 0 (/(·,ro,Po))
Vr = OEBor,ro EBO, Vrf = Ot,toU(to, ·,Po))
v0 = OEBOEB00,p0 , v0f = 00,p0 (J(to,ro,·))
Vct, = 0EBOEB8ct,,p0 , Vct,J = Oct,,p0 (/(to,ro,·)).
3 We remark that in the sequel, we use the more natural notation Ot,p, Or,p, 80,p, O<f,,p
for Vt, Vr, vo, vq,, respectively.
If g 8 2 is the (Riemannian) metric from Example 5.7, then the metric g on

Mis given by
g(p )(avt EB /3 Vr EB v, a'vt EB f3'vr EB v')
= - 2m) aa, + (1--;:;;-
(1--;:;;- 2m)- /3/3
1 ,
+ rO
2
gs2 ( Po )( v, v ') ,
where a, a', /3, (3' E JR, v, v' E Tp 0 S 2 • The metric component matrix for g(p)
in the chart JR x (2m, oo) x U, with respect to the basis vectors Vt, Vr, v0, Vct,,
l·
in this order, is given by:
~
-(1- 2/ : ) 0
[ Q (1-!;)-l
0 0 r~
0 0 0 r~ (sin 00 ) 2
We note that this has determinant -r6(sin0o) 2 =I= 0, and so the scalar
product is nondegenerate, and moreover, the index is 1, since for r 0 > 2m,
the first diagonal entry is negative, while the others are positive. We obtain
the same expression for the metric component matrix for g(p) in the chart
JR x (2m, oo) x V, where V is as in Example 1.8 (p. 7). ◊
Exercise 5.15. Let (M, g) be a Lorentzian manifold, and (U, x) be an admissible

chart for M. Let G = [gii] E Rmxm be the matrix of chart-induced component
functions gii, 1 ~ i, j ~ m. Prove that det G(p) < 0 for all p E U.
Hint: Choose an orthonormal basis B = {ei : 0 ~ i ~ m - 1} for TpM with respect
to g(p). Write Oxi,p using B, and find an expression for gij(p).
Exercise 5.16. Let (M, g) be a semi-Riemannian m-dimensional manifold. Let
(U,x) be an admissible chart for M. Let G = [gii] E Rmxm be the matrix of
chart-induced component functions gii, 1 ~ i, j ~ m. Show that for each p E U,
G(p) is symmetric, invertible. Also show that p >-+ gii = [(G(p)- 1 )]ii are elements
of C 00 (U), where [(G(p))- 1 ]ii denotes the entry in the ith row and jth column of
the matrix (G(p))- 1 . We call gii the chart-induced inverse metric components.
Exercise 5.17. Let (M, g) be a semi-Riemannian m-dimensional manifold, and
(U, x), (U', x') be admissible charts for M, with chart-induced inverse metric com-
ponents denoted by gii, g'ii, respectively. Show that
lij 0X 1i kl OX 1j OX1i . 0( X1i O X-l)
g = oxk g oxi, where oxi .= oui (x(•)).
5.3 Time-orientation, observers, and proper time
In each tangent space TpM of a Lorentzian manifold (M, g), where p EM,
the metric g(p) at p gives a scalar product with index 1, resulting in a 'causal
structure', partitioning nonzero vectors of TpM into spacelike, lightlike and
timelike vectors.
Thus we may imagine the Lorentzian manifold as a distribution of light

cones (depicted below on a surface M, with m - 2 dimensions suppressed).
(M, g)
We wish to be able to speak of 'future' light cones, where e.g. for an

observer, time 'only moves forwards' 4 . This amounts to a smoothly varying
choice of one of the components of the light cones at a point. This smoothly
varying choice is encoded in the existence of an everywhere timelike vector
field V , where we declare the timelike value Vp of the vector field at a point
p E M, to be future-pointing (thus making a choice of the cone; note that
-VP would then be deemed 'past-pointing').
Definition 5.8. (Time-orientation, future-pointing vectors).

A Lorentzian manifold (M, g) is called time-oriented if there exists a vector
field VE TJ M such that at each p E M, VP is timelike. The vector field V
is then referred to as a time-orientation for (M, g).
Let p E M. A timelike or lightlike tangent vector v E TpM is called
future-pointing if g(p)(v, Vp) < 0, and -vis then said to be past-pointing.
Example 5.11. (Minkowski spacetime).

Consider the Minkowski spacetime (M, g) from Example 5.6. Suppose that
v E V is any timelike vector. For each p E M, recall from Example 2.2
(p.28), the identification ofTpM with V via the isomorphism I: V--> TpM,
V 3 v ~ v'Yv,P' where ,v(t) = p + tv for all t ER Give (M,g) the time-
orientation5 VE TJ M, where VP := J(v), p EM. Then (M, g) is said to be
given the time-orientation induced by v. ◊
4 This is for example based on our everyday experience of thermodynamical processes,
where we can distinguish between the past and the future by observing the change in
entropy: the second law of thermodynamics implies that the entropy increases.
5 To see that V is smooth, we show that its component functions in a global chart are
smooth. Consider the chart cp = (x 0 , x 1 , x 2 , x 3 ) (Example 1.3), induced by a p. E M

and a basis {eo,e1,e2 , e3} for V . Let v = vie;. Then
v, (xi)= V
P 'Yv,P
xi= d( x i o-yv) (0) = d(bi+tvi) (0) = vi
dt dt '
where 'Yv(t) = p + tviej = p. + vp.p + tviej = p. + biej + tviej and vp,p = biej. So
V = vioxi. As the constant functions M 3 p >-+ vi are smooth, Vis smooth.
Example 5.12. (Cylindrical spacetime).

The cylindrical spacetime (M, g) from Example 5.8 can be given a time-
orientation WE TJ- M, where Wis defined as follows: In the chart (U, cp),
Wp := Dt,p (p EU) , and also in the chart (V, ¢ ), Wp :=ot,p (p EV). Then W
is a well-defined element of TJ- M , and it is timelike everywhere on M . ◊
The FLRW spacetime (M , g) from Example 5.9 can be given the time-
orientation V E TJ- M, where the vector field V is defined as follows: In the
chart (M, id1 xJR3), Vp = Dt ,p, p E M. ◊
Example 5.14. (Schwarzschild spacetime) .

The Schwarzschild spacetime (M, g) from Example 5.10 can be given the
time-orientation W E TJ- M, where W is defined as follows: For p in the
chart domain IR. x (2m, oo) x U , Wp = Ot ,p, and for pin the chart domain in
the chart IR. x (2m, oo) x V, Wp = Ot ,p· Then W is well-defined and timelike
everywhere. ◊
Example 5.15. (A non-time-orientable spacetime) .

We remark that a Lorentzian manifold does not have to be time-orientable.
A simple visual example is the following . Consider the smooth manifold
C which is the same as the smooth manifold in our cylindrical spacetime
example, but now we consider a different metric g. In our cylindrical space-
time example, the metric g was such that all the light cones had their axis
along the z-axis in IR. 3 . But now suppose that the light cones tilt as we go
along a circle I round the cylinder such that it makes a 180° rotation as it
traverses this circle ,. See the following picture.
----- --- _r_

'J-' ➔
Intuitively we expect that a smoothly varying choice of one component of

the light cone is not possible. The technical details of this are relegated to
an appendix to this chapter. ◊
Definition 5.9. (Observer, proper time).

Let (M, g) be a time-oriented Lorentzian manifold with a time-orientation
V E TJ M. An observer is a smooth curve 'Y : J -> M, where I is an interval
in R, such that for all t E J, v-y,-y(t) is timelike and future-pointing. The
proper time T-y(P, q) experienced by the observer between the events p="((a)
and q = "f(b), where a, b E J and b > a, is defined by
T-y(p,q) = f .J-g('Y(t))(v-y,-y(t),V-y,-y(t)) dt.
The following result shows that the proper time experienced is independent
of the parametrisation of the curve.
Proposition 5.4. Let 'Y: [a, b] -> M be a smooth curve such that v-y,-y(t) is
timelike for all t E [a, b]. Let cp : [c, d] -> [ a, b] be a smooth map such that
cp(c) = a, cp(d) = b, and cp'(t) >0 for all t E [c, d]. Let 7 : [c, d] -> M be given
by 7 ='Yo cp. Then r(7(c), 7(d)) = r('Y(a), "f(b)).
Proof. First we note that for any f E C 00 (M), and for c ~ t ~ d,
~~ f = d(J o 7) (t) = d(J o 'Y o cp) (t)

V-y,-y(t) dt dt
= :;f O 'Y\cp(t))cp'(t) = (cp'(t)v-y,-y(<p(t)))f.
Thus v7 ,:=y(t) = cp' (t) v-y,-y(<p(t)) · So v7 ,7(t) is timelike for all t E [c, d], thanks
to the facts that v-y,-y(t) is timelike for all t E [a, b], cp'(t) > 0 for all t E [a, b],
and the relation
g(7(t))(V7,7(t), V7,7(t)) = (cp' (t) )2g('Y( cp(t)) )(V-y,-y(<p(t)), V-y,-y(<p(t)))-

Let p="((a) =7(c) and q="((b) =7(d). Then we have
r7 (p, q) = id .J-g(7(t))(v7,7(t), v7 ,7(t)) dt
i d
-g('Y( cp( t))) (cp' (t) V-y,-y(<p(t))' cp' (t) V-y,-y(<p(t))) dt
i d
-g('Y(cp(t)))(v-y,-y(<p(t)), v-y,-y(<p(t))) cp'(t) dt (as cp'(t) >0)
s=<p(t)
=
Jb v-g('Y(s))(v-y,-y(s),V-y,-y(s))
a
.j ds = T-y(p,q).

So physically, the parametrisation of the worldline of an observer just tells

us about the order of occurence of the events along the worldline, and
does not have any real physical meaning otherwise. We can choose the
parametrisation arbitrarily/ conveniently as long as the order of events is
preserved (guaranteed by the change of variables function r.p satisfying r.p' > 0
in Proposition 5.4). On the other hand, the proper time is something
which is physically real (the time experienced/recorded by a ticking clock
carried by the observer). When discussing proper times, there is no loss of
generality in assuming that the worldlines of observers are parametrised on
a convenient interval, such as [O, 1].
Exercise 5.18. (Life ofµ: Muon decay and time dilation).
Consider the Minkowski spacetime (M, g) from Example 5.6. Let (U, x) be the
inertial frame induced by p EM, and an orthonormal basis {e 0 , e1, e2, e 3} for V
with respect to for g, such that g(e0 ,e0 ) = -1. Define the time-orientation on
(M, g) as the one induced by e 0 (Example 5.11). If d > 0 and u E (0, 1), consider
~:~(~)e='~J(i:/:+ (l-t)de1 + tteo, for all t E IR. Set P=,(0) =p+ de 1 and
s!:. 'Y
L,
u
p
p
d
Let T > 0. Show that if 1 > u > Hl, y2

l+-;p
then I is an observer and r--r(P, Q) < T.
Remark: A muon is a particle, present in cosmic radiation, which decays in its

proper time T = 2.2 x 10- 6 sec. Thus, T = 660 m, in units in which the speed c of
light is 1. Let d = 10 km, the thickness of the Earth's atmosphere, and u = 0. 998
(99.8% of the speed of light). Then the inequality above is satisfied. Hence, if a
muon just outside the Earth's atmosphere is travelling towards the Earth along
1 ('having speed u in the inertial frame (U, x) '), then it can traverse, without
decaying, the thickness of the Earth's atmosphere. Note that in our inertial
frame coordinate time, a particle can travel at most 660 m in the coordinate time
2.2 x 10- 6 sec. So how does the muon manage to travel a considerably larger
distance (bigger than the thickness d of the Earth's atmosphere)? The answer is
that for the muon decay, only its own proper time matters, not the coordinate
time in some (quite arbitrarily chosen) inertial frame. Cosmic radiation muons
can indeed be detected on the surface of the Earth. *
Theorem 5.6. (Twin paradox in Minkowski spacetime).

In Minkowski spacetime (M, g) , let p EM, and v E V be any timelike vector.
Let (M,g) be given the time-orientation VE TJM induced by v.
Consider two observers, 'YA,"/B: [O, l]--> M, where
e "f A ( t) = p + tV, 0 ~ t ~ 1,
• 'YB(O) = p = 'YA(O), and 'YB(l) = p + v = "/A(l).
Then with q := p + v, we have r,,A (p , q) ~ Tw (p, q).
Thus if 'YA, 'Y8 are imagined to be the worldlines of two twins, then twin
A ages faster. We will see later on that "/A is a 'geodesic', implying that
the motion of twin A is 'unaccelerated', and in this sense twin A can be
thought of as 'resting', while the other twin undergoes 'accelerated motion'.
The inequality in their experienced proper times can then be summarised
by saying: Resting is rusting!
Intuitively, the twin paradox inequality follows from a repeated appli-
cation of the reversed triangle inequality by a 'discretisation' process. A
picture in the Minkowski plane with a coarse discretisation is shown below.
q=p+v
Proof. We will use a basis induced by a global chart. From the proof of
Theorem 5.4, we can construct a basis B = {e0 ,e 1 ,e2 ,e 3 } for V, orthonor-
mal with respect to the Minkowski scalar product g on V, such that
V
eo = --;:=::;:====;::.
,J-g(v, v)
We use the chart (M,<p) induced by 6 p and B. For all t E [0,1], we have
'YB (t) = p + 'Y~ (t)ei, where 'Y~ := (<po "f8)i are real-valued smooth functions
oftE[0,1]. Thenvw,w(t)='Y~Oxi,w(t),where ·=~t· ForapointxEM,
recall 7 the isomorphism I : V --> TxM, which sends w ~ v,,w,x, where
6 See Example 1.3 (p.3) .
7 See Example 2.2 (p.28).
1'w(t) := x + tw, t R The time-orientation V E TJ M on M induced

E
by v is given by Vx = Iv, x E M. Since 1'B is an observer, Vw,w(t) is
future-pointing:
0 > g(1'B(t))(v')'B,")'B(t), Iv) = g(1'B(t))(i'~ Oxi,w(t), v-g(v, v) Oxo,'"YB(t))
= -v-g(v, v)'Y~,
and so i'~(t) > 0 for all t E [O, l]. We have
-g(1'B(t))(vw,w(t), Vw,w(t)) = (i'~(t)) 2 - (i'!(t)) 2 - (i';(t)) 2 - (i';(t)) 2
~ (i'~(t))2,
and so we obtain
Tw(p,p+ v) = f .J-g(1'B(t))(vw,w(t),Vw,w(t))dt
~f V(i'g(t)) 2 dt
= L\~(t)dt = 1'~(1)-1'~(0) = -g(v,eo) - 0 = v-g(v,v).
But v7A,'"YA(t) =ViOxi,'"YA(t), and so

-g(1'A(t))(v7 A,'"YA(t), v7 A,'"YA(t)) = (v 0 )2- (v 1 ) 2 - (v 2 ) 2 - (v 2 )2 = -g(v, v).
Thus,
T7 A(p,p + v) = f .J-g(1'A(t))(v7 A,'"YA(t), v7 A,'"YA(t)) dt
= f V-g(v, v) dt = V-g(v, v)-1.

Consequently, T')'B(p,p + v) ~v-g(v, v) =T')'A (p,p + v). □
One might mistakenly attribute the twin paradox, that is, the inequality
above, to be an artefact of the accelerated motion of twin B. This is
somewhat misleading, since the acceleration is only needed to make the
two twins meet again. The source of the inequality is the reversed triangle
inequality. To emphasise this, we show that the inequality can also be
brought to bear by taking two geodesics albeit in a modified spacetime,
namely the Minkowski plane made into a cylinder. The reason we need to
modify the Minkowski plane to the cylindrical spacetime is as follows. We
will see later on that geodesics in the Minkowski plane are straight lines,
and straight lines originating at a point won't ever meet again. To make
them meet again, we can 'roll up' a Minkowski plane to make a cylinder.
Theorem 5.7. (Twin paradox in cylindrical spacetime).

Let (M, g) be the cylindrical spacetime from Example 5.8, with the time-
orientation W as given in Example 5.12.
Consider the two observers "/A, "/B : [0, 271"]-+ M given by
• "/A(t) = (1,0,2t), 0::::;;t::::;;271",
• "/B(t) = (cost,sint,2t), 0::::;;t::::;;271".
Then with p= (1, 0, 0) and q= (1, 0, 471"), we have T,YA (p, q) ~ Tw (p, q).
Proof. Using the chart (V, '¢) from Example 5.8, it can be seen that the
proper time experienced by observer "/A is given by
For the observer "/B, we will use the chart (U, r.p), which covers "/Bl(o, 2,,.),
and only the endpoints p, q do not belong to U. We have
f27r
Tw(p,q) = Jo
f27r
= Jo V-(1-4)dt=v'3·(211").
We reiterate that in Minkowski space, the symmetry among the two twins
was broken because one of them was a geodesic while the other one wasn't.
So the non-resting twin 'feels' accelerated. In the case of the cylindrical
spacetime, the symmetry is broken by the fact that one of them winds
around the cylinder, while the other doesn't. However, neither twin 'feels'
accelerated! It is the global topology which distinguishes the two geodesics
and the resulting difference in the experienced proper times.
The Schwarzschild spacetime models the spacetime outside a spherically
symmetric body of mass m, in an otherwise empty universe. It turns out
that realistically, this is a good approximation near the body (where the
effect of the body is strongly felt, and the rest of the universe doesn't
matter). In particular, it can be used for modelling the spacetime geometry
near the Earth. We end this section with the discussion of a result in
Schwarzschild spacetime, which says roughly speaking that objects near
the surface of the Earth age slower, and that the light emanating from the
surface of the Earth, observed high above, is red-shifted. But first we give
the following definition.
Definition 5.10.
(Energy/frequency of a light signal measured by an observer at an event).
Let (M, g) be a Lorentzian manifold with a time-orientation. A light signal
is a smooth curve8 >. : J -> M, where I c JR is an interval, such that
for each t E I, v,\,,\(t) is lightlike. Let 'Y be an observer, and suppose that
p E M belongs to the intersection of the images of 'Y and >.. Then the
energy/frequency of>. measured by 'Y at pis defined to be
g(p)(v")',p, V,\,p)
v-g(p)(v")',p, V"Y,P)
Example 5.16. (Gravitational red-shift).

Recall the Schwarzschild spacetime (M, g) from Example 5.10, and the
time-orientation given in Example 5.14. Let r 2 > r 1 > 2m, and let p E S 2.
Consider the two observers 'Y1(t) = (t,r1,P) and 'Y2(t) = (t,r2,P), t ER
Suppose that at the event Pi = (t1, r1, p) on the worldline "/1, a 'radial light
signal' >. : [a, b] -> M is sent towards the other observer, reaching it at
the event P2 = (t2,r2,p). We write >.(r) = (t(r),r(r),p), TE [a,b], where
>.(a)=Pi and >.(b)=P2 . Since>. is the worldline of light, V,\,,\(t) is lightlike
for all t. Thus if -' = ~T, then
0 = g(>.(r))(v,\,,\(T), V,\,,\(T))
= g(.X(r) )(t' (r) Ot,,\(T) + r' (r) Or,,\(T), t' (r) Ot,,\(T) + r' (r) Or,,\(T))
= -(1- r2(m)) (t'(r))2+ l _\m (r'(r))2.
T r(t)
As the light signal travels radially outwards, we suppose that r' (T) > 0 and
t'(r) > 0. Let T := r 1 : [t1, t 2] -> [a, b] be the inverse oft: [a, b] -> [t1, t 2].
With h:=ror: [t1,t2]-> (2m,oo), we get the differential equation
dh(t) = d(r or) (t) = r'(r(t)) = l - ~ = 1 _ 2m

dt dt t'(r(t)) r(r(t)) h(t) ·
We have h(t1)=r(r(t1))=r(a)=r1, and similarly, h(t2)=r2. Integrating,
t2 -t1 = lt2 ldt lt2 ~ - ddht (t)dt

t1
=
t1
1
1- h(t)
r=h(t) lr2
=
r1
--
r
-dr.
r- 2m
This expression for t2 - t1 depends only on r1, r2 and is independent of t1.
8 The curve should be a 'geodesic', which will be defined later on in Chapter 8. Right
now we just assume that we have a special type of smooth curve given, and carry on.
t A
>- P2
A
>-
A
2m r1 r2 r
'
')'1 ')'2
Suppose another radial light signal 3: is sent by the first observer at the event
A = (t1 , r1 , p) , with t1 > t1 , towards the second observer, reaching it at the
event A= =
(t2 , r2, p). Then it follows from the above that t2 - t1 t2 - ti,
and so also t2 - t2 = t1 - t1. We have that the proper time experienced by
the first observer between the events A and Pi
is
Similarly, the proper time experienced by the second observer between P 2

and A is
~ = ~l--(t2
T,2 (P2, P2) m ~ - t2).
r2
Thanks to the equality t 2 - t 2 = t 1 - ti=: f::lt, and the inequality r 2 > r 1 we
obtain that
~ =~
T,2 (P2 , P2) - - f::lt > ~
1m 1m ~
- - f::lt = T,1 (P1 , A).
r2 r1
If A , A were the subsequent ticks of a clock carried by the first observer,
then the information about these ticks is 'seen' and recorded by the second
observer along the light signals >-, 3: at the events P 2 , A. By the above
inequality, the second observer reckons that the first observer's clock runs
slow (as compared to the clock carried by him/ her). We say that 'a clock
in a stronger gravitational field is seen as running slow by an observer in a
weaker gravitational field '. So near the surface of the Earth, we age slower
as compared to someone higher up.
Red-shift: The frequency of the outgoing light signal >. measured by the
first observer at A is
g(A)(v71,Pi, V>.,Pi)
The frequency of the incoming light signal >. measured by the second ob-
server at P2 is
g(P2)(V72,P2, V>.,P2)
V2 = ---;=:::;:=::::;::::;======;::
vf-g(P2)( V72,P2, V72,P2)
= _ g(P2)(t'(b) Ot,P2 + r'(b) Or,P2, Ot,P2) = t'(b) ✓l _ 2m.
vf-g(P2)(8t,P2, Ot,P2 ) r2
It will be shown later (see (15.6) in the proof of Proposition 15.1, p.332)
that since >. is a lightlike geodesic, there exists a constant E such that for
all TE [a, b],
So
t'(bh/1 - ~
t'(a)-Jl - 2r7
As r2 > r1, we have v2 < v1, that is, the light observed by the second
observer is 'redder' than that observed by the first observer. The red-shift
z is defined by
We say that 'light climbing out of a gravitational field is red-shifted'. ◊
Exercise 5.19. (Twin paradox in Schwarzschild spacetime).

The spacetime around Earth is well-described by the Schwarzschild spacetime
( M, g) where m = Me, the mass of Earth. Let (U, cp) be the spherical coordinate
chart given in Example 1.8 (p.7). Let the curve 'Y : R -+ M be a timelike
'geodesic' in the equatorial plane (0 = i) having a constant r-coordinate r 0 :
'Y(s) = (t(s ), ro, cp- 1 ( i, cf>(s)) ), s E R.
It can be shown9 that since 'Y is a geodesic,

(ef/)2 = "; (t')2,
ro
where ·' := ~s. Let cp('Y(O)) = c/>o, and let so> 0 be the smallest parameter value
such that cf>('Y(so)) = cf>o. Let Ta.bit be the proper time experienced by 'Y between
the events p:="f(O) and q:="f(s 0 ). Consider also the stationary timelike observer
::Y(s) = (s,ro,<p- 1(i,cf>o)), 8 E R Let Tstationa,y be the proper time experienced
by ::y between the events P, q. Show that Torbit < Tstationary• We remark that the
stationary observer ::Y is not a geodesic (Exercise 8.4), so that his/her motion
in spacetime is 'accelerated'. In contrast to Theorem 5.6, now the accelerated
twin ages more! However the proper time maximising curve is always a timelike
geodesic (Theorem 8.5, p.177), which in this case happens not to be the circular
orbit, but a radial up-and-down motion. This is explored in the next exercise.
Exercise 5.20. (Radial timelike geodesic in Schwarzschild spacetime).

In continuation of the previous exercise, now consider a radial timelike geodesic
"f., 'parametrised 10 by proper time', as described in Exercise 14.7 (p.317). Thus
'Y•: I-+ M, I 3 T >-+ (t(T),r(T),p), where p E S 2 is fixed. In Exercise 14.7, the
following equations are derived (which we accept for now):
2m) (t), 2 + ( 1--;:-
- ( 1--;:- 2m )- (r), 2 = -1,
1
t" = _ 2m
r2
(l- 2m)-\'r',
r r
,, = - r2·
m
Here •' := ~r. Using these, prove that there exists a constant E > 0 such that
(1- 2;_n)t' = E, (r') 2 =E 2 - 1+ 2 ;.n.

Show that a free-falling particle that is thrown upwards from r = r 0 , reaches a
maximum height r = r1, and falls back to r = r 0, experiences a proper time
r1
!:::..T = 2 f
1
----;===dr,
ro ✓ 2m_2m
r r1
corresponding to a 'coordinate time' lapse
Use this to show
f:::..7>f:::..tF3f!-.
(The particle experiences more proper time than the stationary observer at r = r 0 .)
Hint: Compare the derivatives of !:::..T and t::..t,Jl- 2rom with respect to r 0 using
9 Follows from the r-component of the geodesic equation, using r' =0, 0 =O; see the
1
solution to Exercise 8.8.

10 See Lemma 12.1, p.269.
Exercise 5.21. (Unruh effect).

Let (M, g) be the 2-dimensional Minkowski spacetime over V = IR2 . Let (U, x) be
the inertial frame induced by a p E M , and an orthonormal basis {e 0 , ei} for V
with respect tog:= T/ , where T/ is as in Example 5.3, and let g(e 0 ,e 0 ) = -1. The
time-orientation on (M, g) is the one induced by e 0 . Given an a > 0, consider
the observer 1 : IR -+ M given by
,(t) = p + .!(sinh(at)e0 + cosh(at)e1), t E IR.
a
The boundary iJH = {p+t(e0 - e1) :t EIR} of the half-plane
H = {p + v0 eo + v 1e1) : v 0 ?= vi}
forms a 'horizon' 11 for the observer, since an event cannot send information to
the observer from within H. Thus any light signal from the observer entering the
horizon is lost to him irretrievably. Let >-+, )._ be a 'photon pair' created at an
event ,(to) on the observer's worldline with 'energies' E , -E:
>-+(t) =,(to)+ E(t- to) (eo + e1)}
, t ?= to .
L(t) =,(to)+ E(t- to) (eo - e1)
Suppose that the photon )._ with energy - E crosses the horizon. Determine
the coordinate time L:::..t in the chart (U, x) it takes for this photon to reach the
horizon.
H
Assuming the uncertainly relation EL:::..t = %, find E, and also show that the
energy Eobs erved of the photon >-+ measured by the observer is given by
Eobs erved = na.
(This can be related to an effective temperature of the horizon perceived by the
observer,
T = Eobserved !ia
2Jrks
where ks is the Boltzmann constant.)
11 Roughly speaking, a horizon is the boundary of a region R of spacetime such that the
events from within R cannot send a light signal to observers (that is, paths whose range
lies) outside R.
5.4 Raising/Lowering indices or musicality
In the setting of a semi-Riemannian manifold, a I-form field can be con-

verted into a vector field and vice versa, using the so-called musical isomor-
phisms, as we shall see in this section. We will learn about two maps,
.I>: TJM-> Tf M ('flat') and .U: Tf M-> TJM ('sharp'),

which will be inverses of each other. If one looks at a chart-induced compo-
nents, the flat map 'lowers' the index (which we here think of as lowering
the pitch, hence 'flat'), while the sharp map 'raises' the index.
Definition 5.11. (Flat map).

Let (M, g) be a semi-Riemannian manifold. Define .I> : TJ M -> Tf M by
(Vl>)W = g(V, W), V, WE TJ M.
The C 00 (M)-linearity of v1> follows from the C 00 (M)-linearity of g in its

second slot. We now show that .I> is a C 00 (M)-module isomorphism.
Theorem 5.8. The map .I> : TJ M-> Tf M is C 00 (M)-linear and bijective.
Proof. The C 00 (M) linearity of .I> follows from the C 00 (M)-linearity of g

in its first slot.
Injectivity: Suppose that V E TJ M and VI> = 0, that is, for all W E TJ M,
0= v1>w =g(V, W). Let p EM and w E TpM. By Lemma 3.1 (p.44), there
exists a vector field W E TJ M such that Wp = w. So
0 = (g(V, W))(p) = g(p)(½,, Wv) = g(p)(Vp, w).
As w E TpM was arbitrary, and since g(p) is non-degenerate, it follows that

Vp = 0. Since p E M was arbitrary, V = 0. Thanks to the linearity of .I>,
we can conclude that .I> is injective.
Surjectivity: Suppose that n E Tf M. Let (U, x) be any admissible chart.
Then we have Olu = Oidxi, where ni = n l Jxi. Define a vector field Vlu
0
by Vlu = giiOiOxi, where we use the notation gii from Exercise 5.16 for
the chart-induced inverse metric components. As the charts cover M, this
procedure allows us to define a vector in TpM at each p E M. To check well-
definition, we note that if (U', x') were an overlapping chart, then using the
transformation properties of the I-form components, of the basis elements
of TpM, and of the inverse metric components (Exercise 5.17), we obtain
in Un U' that
VI U' -_ g tijr,t :,i

_
HiUx'1
OX 1i kf. OX 1j r, OXr OX 8 :l
.
- OXk g OXR. H r OXli OXlj Ux•

OX 1i OXr OX 1j OX 8 kf.
= oxk ox'i oxf. ox'i g OrOx•
= 8k8£gkf.0rOx• = grsnrOx• = Vlu-
In the above we also used the result from Exercise 2.16. The smoothness
of V follows from Theorem 3.1, since we know that the chart-induced com-
ponents are smooth. To complete the proof of the surjectivity, we will now
show that for this V, we have that Vb = n. Indeed, for any W E TJ M, we
have in any admissible chart (U, x) that
b · · k. · k · ·
V W = g(V, W) = gijV"W1 = gijg ·nkw1 = 8jnkw1 = njW1 = nw.
As the charts cover M, VbW =OW on M. Since WE TJM was arbitrary,
Vb =0. □
The map .U : Tf M -> TJ M, defined as the inverse of the bijective map

.b: TJM->TP M, is also seen to be a C 00 (M)-linear map as follows:
Indeed, let n, e E rp M and f E 0 00 ( M). By the bijectivity of .b, there exist

unique V, WE TJ M such that Vb= 0 and Wb = 0. The C 00 (M)-linearity
of .b implies that (JV+ W)b = f Vb+ Wb = f O + 0. Now the injectivity of .b
gives, in light of the definition of .U, that (JO + 0 )U = f V + W = f nu + eU.
Definition 5.12. (Sharp isomorphism).

Let (M, g) be a semi-Riemannian manifold. The inverse of the C 00 (M)-
linear bijective map .b : TJ M-> Tf Mis the C 00 (M)-linear bijective map
.U : Tf M -> TJ M, and is called the sharp isomorphism.
The surjectivity part of the proof of Theorem 5.8 shows that if n E TJ M,

then in any admissible chart (U,x), OUiu = giiOioxi, where Oi = Ooxi.
Exercise 5.22. (Gradient field as a vector field).
Let M be a smooth manifold and f E C 00 (M). We had defined the 1-form field
df E Tf Min Exercise 3.15 (p.51). If Mis equipped with a metric g, then using
the musical isomorphism J, we can convert the 1-form field df into a vector
field grad f E TJ M, defined by grad f = (dff Prove that for any V E TJ M,
g(grad f, V) = V f.
Now let M = Rm with the standard smooth structure, and let the metric g
be the usual Euclidean inner product (·, •). Consider the global chart (Rm, id).
Show that the chart induced components (grad f)i are given by Oxif.
Exercise 5.23. (Musicality in a vector space with a scalar product).

Let V be an m-dimensional real vector space with a scalar product g. Recall that
in Exercise 5.1, we had seen that-°: V-+ V*, where v 0 := g(v, •) (veV), is an
IR-linear isomorphism. Denote the inverse of -° by J : V* -+ V. Let {v1, • • • , Vm}
be a basis for V and let {w1, • • • , wm} be the corresponding dual basis for V*,
that is, wivi = 83 for all 1 ~ i, j ~ m. Show that (vi)° = g( Vi, Vj )wi. We conclude
that if {v1, • • • , Vm} is an orthonormal basis, then
(Vi·)"-{
-
wi if i(vi):=g(vi,Vi)=+l,
·
-w' if i(vi):=g(vi,Vi)=-1.
In this case (i.e., when {v1, • • • , vm} is an orthonormal), show that
w~ = f w(vi)g(vi, Vi)Vi-
i=I
Exercise 5.24. (Trace).

Suppose that (M, g) is a Lorentzian manifold. A (0, 2)-tensor field Tis symmetric
if T(V, W) = T(W, V) for all V, W E TJ M. For a symmetric (0, 2)-tensor field
TE Tf M, define the (1, 1)-tensor field T~ E Tf M by T~(n, V) = T(n~, V) for all
n E Tf Mand VE TJM. The trace ofT is the map traceT := C(T~) E C 00 (M).
Show that if M is 4-dimensional, then trace g = 4.
Appendix: Example 5.15
Here we give an argument justifying the impossibility of time-orientability.

First consider the admissible charts (U, x) and (V, y) covering C given by
the usual cylindrical coordinates:
• U = {p = (cos 0, sin 0, z) : 0 < 0 < 2n, z E R},
with U 3p=(cos0,sin0,z) ~ (0,z), and
• V = {q=(coscp,sincp,z): -?r< <p <n, z ER},
with V 3q=(coscp,sincp,z) .!. (cp,z).
Then the chart transition maps are given by (yo x- 1 ) ( 0, z) = (a( 0), z), and
(xoy- 1 )(cp,z) = (b(cp),z), where
0 •-{ 0 if 0<0<?r} b •-{ <p if O<<p<n}

a().- 0-2nif n<0<2?r ' (cp) .- <p + 2nif -n< <p<O .
We define g E TJ Mon C by giving its the chart-induced component matri-

ces au, av with respect to the ordered bases (ot,80), (ot,ocp), and using
Exercise 4.11 (p.68) to show well-definition. Set
au(0,z) = a(0) for (0,z) Ex(U) } [cosa sina]

v
a (cp, z) = a(cp) for (cp, z) E y(V)
, where a (a ) -_ .
sma - cos a
.
Since the Jacobian matrices of the chart transition maps are equal to the
identity matrix, and also as
au(cp + 21r, z) = au(cp, z) = av(cp, z),
av(0 - 21r, z) = av(0, z) = au(0, z),
it follows that au, av define a (0, 2)-tensor field g on C.
g is a metric: Firstly, the matrices au, av, are both pointwise symmetric
and invertible, giving the symmetry and nondegeneracy of g(p) for each
p EC. To see that the metric index of g is 1, we note that with
v = ( cos i) e1 + ( sin i) e2, w = -( sin i) e1 + ( cos i) e2,
we have vta(a)v= 1, wta(a)w= -1, vta(a)w=0.
Consider the smooth curve 'Y: (0, 21r)-> C given by
'Y(0) = (cos0,sin0,0) for all 0 E (0,21r).
Let WE TJU be the vector field
Wp = -(sin;) Ot,p + (cos;) 00,p
for all points p = (cos 0, sin 0, z) E U, 0 < 0 < 21r, z E R. Then Wp is timelike
for all p E U since g(p)(Wp, Wp) = -1 < 0. Now suppose X E TJC is a
time-orientation for (C,g), and write X = xtot + X'Pocp in V, for some
Xt, X'P E C 00 (V). Then for all p E Un V, we have
ob oz
ocp,p = ocp (y(p))o0,p + ocp (y(p))oz,p = l-00,p + O·oz,p = 00,p-
Thus in Un V, we have X = Xtot + X'Pocp = Xtot + X'Po0. Define smooth
function f: (0, 21r)-> R by
/(0) = g('Y(0))(W.,,(0),X.y(0)) for all 0 E (0,21r).
Then by the continuity of f, and the fact that W is timelike everywhere, we
must have f either everywhere positive, or everywhere negative. (Indeed,
otherwise f will be zero at some 0*, giving a point 'Y(0*) E C, where the
nonzero timelike vectors X.,,c0.), W.,,c0.) are orthogonal, a contradiction to
Lemma 5.1.) Set p = (l, 0, 0) E C. Then as Xp is timelike, we obtain
(Xt(p)) 2 - (X'P(p)) 2 < 0. This implies that X'P(p) =I- 0. But we have
. ( )-[ . 0 0 ][cos0 sin0][Xt(y- 1 (0,0))]- 'P()
J~f 0 - -sm 2 cos 2 sin0 -cos0 X'P(y- 1 (0,0)) --X P'
while
. ( )-[ . 0 0 ][cos0 sin0][Xt(y- 1 (0-21r,O))]- 'P()
hm f 0 - -sm 2 cos 2 Sln
0 /21r
. 0 - COS 0
X'P( y -1( 0 - 27r, 0)) -X P,
a contradiction to the fact that / is either everywhere positive or everywhere
negative. So (C, g) is not time-orientable.
Chapter 6
Levi-Civita connection
In this chapter we will introduce additional structure on a smooth man-

ifold, called a connection. This will allow us to calculate the directional
derivative of a vector field in a given direction. While there are infinitely
many connections definable on a smooth manifold, in the context of a semi-
Riemannian manifold, there is a certain natural one, called the Levi-Civita
connection induced by the metric g. For spacetimes (M, g), we will always
use the Levi-Civita connection induced by g.
We know that if v E TpM, and f E C 00 (M), then the directional deriva-
tive off in the 'direction' provided by the tangent vector vis simply v(J).
But now suppose that we replace f by a vector field W E TJ M. Can we
still 'differentiate W in the direction of v'?
Before we address this question, we remark that wanting a notion of a
directional derivative v' v W of a vector field W in a direction given by v
is quite natural. For example if we are moving along a curve 'Y : R -+ M,
we might want to know the rate of change v' v,,,r(tJ W of a vector field W
at 'Y(t) in the direction of the tangent vector v-y,-y(t)· In particular, we even
wish to replace the vector field W by simply the map t i--+ v-y,-y(t) so that
we have a notion of acceleration a:= v'v,,,,,(tJ v-y,-y(·)· We will then be able to
discuss the 'straight' curves as ones for which the acceleration is identically
zero along the points of the curve, and these will be called geodesics, which
is the subject of a later chapter.
Returning back to the question about the possibility of differentiating
Win the direction of v, there is a fundamental obstacle. Ideally we want
to consider a curve T (-1:, 1:)-+ M so that p="f(O), v = v-y,-y(O), and take the
limit as t -+ 0 of the difference quotient
107
The basic difficulty is that W -y (t) E T-y(t)M, while W -y (O) E T -y (o)M , since the
spaces T-y(t)M, T -y (o)M are different, we cannot make sense of W-y(t)-W-y(O)·
It would be great if we could somehow 'connect' the tangent spaces
T -y (t)M, T -y (o)M , as in the happy situation of !Rm, where (as we shall see
below), there is a standard way of doing so (using the vector space structure
of !Rm), making the above idea work. In a general manifold, however, there
is no such 'connection' available from the smooth structure alone.
W -y( t )
-----t---4. , (t)
p= , 0)
We will have to put such an additional structure 'by hand' , and this is
exactly the notion of a 'connection' , which we will soon introduce. There
are infinitely many connections available on a given smooth manifold, and
so one has to make a choice in order to be able to do the aforementioned
differentiation. However, for a semi-Riemannian manifold, there is a certain
natural choice, induced from the metric, called the Levi-Civita connection.
The Levi-Civita connection allows the definition of a geometric object called
the curvature tensor field R. The spacetime field equation involves tensor
fields derived from R .
6 .1 Motivation for connections: the case of IR,n
The smooth manifold M = !Rm , with its standard smooth structure, has
a global chart provided by the identity map, and so we can identify each
tangent space TpM with !Rm , using the map
With this identification, given a v E TpM and a W E TJ M , we can define

the derivative of W in the direction of v as follows. Let , : JR --> !Rm be any
Levi- Civita connection 109
smooth curve such that 'Y(O)=O and v"f,p=v, e.g., 'Y(t)=p+tv. Then
v'vW := lim w"((t) - w"f(O)

t-+0 t
We now show that v'vW = v(Wi)ox;, that is, the derivative of Win the
direction of v is simply obtained by taking the directional derivatives of the
chart-induced component functions of Win the direction of v. This proves
in particular that v' vW does not depend on the choice of 'Y· If W = Wiox;,
then keeping the above identification in mind,
'("'7 W _ 1· w"((t) - w"f(O) _ 1· Wi('Y(t)) - Wi('Y(O)) :i _
Vv -Im -Im U,
t-+0 t t-+0 t x ,P
d(Wi O '}') I i i
= dt t=OOxi,p = (v"Y,"f(O)w )Oxi,p = (vW )Oxi,p-
We have v' v W E TpM. If instead of v, we have a vector field at hand, and

we set (v' v W) (p) = v' Vp W, then we obtain a map
v': TJMxTJM--+ TJM

(V, W) >--+ v'vW = V(Wi)oxi-
From the formula
v'vW = V(Wi)oxi,
we obtain for c E JR, f E C 00 (M), V, W, XE TJM:

(Cl) (JR-linearity of v'v) v'v(W + cX) = v'vW + cv'vX
(C2) (Leibniz) v'v(JW) = (VJ)W + fv'vW
(C3) (C (M)-linearity of v'.W)
00 v' 1v+xW = fv'vW + v'xW.
A connection on M = !Rm is a map v' : TJ M x TJ M -+ TJ M that satisfies
(Cl)-( C3). Thus v' is a connection on !Rm, and we call it the flat connection
on !Rm. However, there are infinitely many other connections on !Rm; see
Exercise 6.3 below.
What makes v' special among all connections is that besides the manda-
tory requirements (Cl), (C2), (C3), v' also satisfies (C4) and (C5) given
below. Note that in (C5) below, we consider !Rm as being endowed with
the metric(·,•) defined in each TpM ~ !Rm to be the usual Euclidean inner
product on !Rm. Thus the formulation of (C5) needs extra structure on !Rm
(besides its standard smooth structure), namely a metric.
(C4) (Torsion-free) v'vW - v'wV - [V, W] = 0.
(C5) (Metric-compatible) V(W, X) = (v'vW,X) + (W, v'vX).
The reason behind calling (C4) the 'torsion-free' property is that the map
T 01 MxT01 M 3 (V, W) >-+
-
B(V, W):=v'vW -v'wV - [V, W]
-
E T 01 M
is C 00 (M)-bilinear (see Exercise 6.4), and so the map
T: TfMxTJMxTJM-> JR
(n, V, W) >-+ O(B(V, W)) = O(v'vW - v'wV - [V, W])
is a (1, 2)-tensor field, which is called the torsion tensor-field.
Exercise 6.1. Check (Cl)-(C5) for the flat connection V on (Rm, g).
Exercise 6.2. Consider R 3 with the standard smooth structure and the flat con-
nection. Let V =yzox+zxoy+xyoz, and W = (xy 2 +z)ox +(y 2 -x)oy+(x+z 3 )oz.
Determine VvW, VvP W, and (VvW)(p), where p= (1, 1, 1).
Exercise 6.3. Consider the smooth manifold Rm with the standard smooth struc-
ture. Suppose that rfi E C 00 (Rm) are m 3 arbitrary smooth functions on Rm. Let
V: TJMxTJM-+ TJM be given by (V, W) >-+ VvW = (ViWirfi + VWk)Oxk,
for all V = Vi Oxi, W = Wio"'; E TJRm (xi are the component functions of id!R"' ).
Show that V satisfies (Cl), (C2), (C3), and so is a connection on Rm.
Example 6.1. Consider IR 2 with its standard smooth structure, and define
the connection v' as in Exercise 6.3, where
I'"' -
xx -
rxyy -- I'Yxx -- I'Y - Q
yy - ,
rxxy -- rxyx -- 1 + x2y + y2 ' I'Y - I'Y -
xy - yx -
X
1 + x2 + y2
Then v' defines a connection on IR 2 . In Example 6.2, we will see that this
connection gives a 'non-flat saddle shape' to IR2 . ◊
6.2 Definition of a connection
Based on the considerations above, we give the following definition.
Definition 6.1. (Connection on a smooth manifold, covariant derivative).

Let M be a smooth manifold. A map
v': TJMxTJM-+ TJM
(V, W) ,___. v'vW
is a connection if it satisfies for all c E JR, f E C 00 (M), and V, W, XE TJ M:
(Cl) (IR-linearityofv'v) v'v(W+cX)=v'vW+cv'vX
(C2) (Leibniz) v'v(JW)=(VJ)W + fv'vW
(C3) (C (M)-linearity of v'.W)
00 v' 1v+xW = f v'vW + v' xW.
The vector field 'v' v W is sometimes referred to as the covariant derivative

of W with respect to the vector field 1 V.
For a fixed WE TJM, TfMxTJM 3 (0, V) >-+ 0 ('vvW) E C 00 (M) is a

(1, 1)-tensor field. (C3) provides the linearity in the vector field slot.
Exercise 6.4. (Torsion tensor field). Let M be a smooth manifold with a connec-
tion V. Define B: TJMxTJM-+ TJM by B(V, W) =v'vW-VwV- [V, W]
for V, WE TJ M. Show that B is C 00 (M)-bilinear, that is, it is C 00 (M)-linear in
each slot by verifying that
• [JV, W] = f[V, W] - (Wf)V
• B(JV + X, W) = f·B(V, W) + B(X, W)
• B(V, W) = -B(W, V).
Conclude that T defined by
T(n, V, W) = n(B(V, W)) = n(VvW - VwV - [V, W])
for all n E Tf M, and V, W E TJ M, is a (1, 2)-tensor field, called the torsion
tensor field:.
Exercise 6.5. Suppose that V, V are connections on a smooth manifold M, and

that p E C 00 (M). Prove that (1 - p)V + pV is also a connection on M.
Exercise 6.6. Let IR.2 be given the standard smooth structure, and suppose it is
also equipped with the flat connection V. Show that 2V is not a connection on
IR.2. Hint: The Leibniz rule (C2) can be shown to fail, e.g. by taking f = x and
V =Ox= W in the admissible chart (IR.2, (x, y) >-+ (x, y)).
Exercise 6.7. Suppose that V, V are two connections on a smooth manifold M.

Prove that their difference defines a (1, 2)-tensor field Ton M, where Tis given
by T(O, V, W):=O(VvW -VvW) for V, WE TJM and n E Tf M.
Exercise 6.8. (Lie derivative revisited). We had seen in Definition 3.15 (p.58) that
the Lie derivative £ : TJ M x TJ M -+ TJ M can be defined on a smooth manifold,
using simply its smooth structure, by Cv W = [V, W] for all V, W E TJ M.
Prove that£ satisfies (Cl) and (C2). However (C3) fails in general, and show this
concretely, by taking M = R 2 with its standard smooth structure, the function
f = x, and the vector fields W =Ox= V, X = 0, in the global admissible chart
(R 2 , (x,y) >-+ (x,y)).
1 For an (r, s )-tensor field, the top indices on its components are historically referred
to as contravariant, and the ones below as covariant. Fixing W, we associate v'.W
with the (1, 1)-tensor field sending (n, V) E Tf M x TJ M to n(v'v W) E C 00 (M). Thus
W,..... v'.W takes a (1, 0)-tensor field W, and makes it a (1, 1)-tensor field, increasing its
covariance index from O to 1. Thus W ,..... v'. W is referred to as 'taking the covariant
derivative'.
2 The name originates from the fact that this tensor field describes how a tangent space
twists along a curve when it is 'parallelly transported' along the curve. The notion of
parallel transport will be discussed in the chapter on geodesics.
Remark: So the Lie derivative Lv W is not the appropriate notion of the 'di-
rectional derivative of W in the direction of V' we seek. Indeed, the sought-after
directional derivative at a point should depend only on the value Vp of V at p,
and not on what V does in the vicinity of p . But the Lie derivative does take into
account how V changes. To see this, consider a V of the form V = f X. Then we
have L f x W = f Lx W - (W f)X. The presence of the extra term (W f)X shows
that, if we use two different f s that match at p, so that the Vp is the same, the
different values of f in the neighbourhood of p will have an effect, even at the
single point p being considered (since the W acts on f as a differential operator;
so although the two fs match at p , if their derivatives differ, then this will be
detected by a suitable differential operator W) . Also, one of the applications of
the connection we want to have, is to be able to define the acceleration of a curve
by calculating the derivative v' v-,,r (·> v,.,,,,.,,(·) of the velocity vector to the curve in
the direction of the velocity vector of the curve, and again the Lie derivative is
not useful since Lv V = 0 for any vector field V which is an extension of the v,.,, ,,.,,(·) ·
*
Example 6.2. Consider the parametrised surface Mc JR 3 given by
JR 2 3 (u,v) I-+ r(u , v)=(u, v , uv) c JR 3 .
This is the graph of z=f(x,y)=xy, which has a 'saddle' shape.
We had seen earlier that tangent vectors to M can be viewed as vectors in

JR 3 . We use this identification below. Let JR3 be equipped with its usual
Euclidean inner product C ·) with the induced norm denoted by II·II- Given
a vector field X E Tc} M, we can extend it to a vector field X E Tc} JR 3 . For
example, it can be extended by translating the surface along the z-axis:
Xcx,y,z+a) = xp=(x,y,z) for all p E M and all a E R So if we are given
V, WE Tc}M, we can extend them to vector fields V, WE Tc}JR 3 . Keeping
in mind that the calculation of v' v W at a point p only requires the values
of W along a curve I in M such that 1 (0) = p and v,_,,,P = Vp, the manner
of extension does not matter, and so we can determine v' v W at all points
p EM. At each such p, (v'vW)(p) E Tc}JR3 , but may not necessarily belong
to TpM. To remedy this, we subtract its component in (TpM)1-.
..... ____ ___ _____ _____ ...
Thus we define v' : TJ- M X TJ- M --> TJ- M by v' V w = v' V w - (v' V w,n) n,
where n is the 'unit normal field ' to M , given by
or or
-x-
ou OV (-v , -u , 1) 1
n= ✓1 2 2(-v, -u, l).
11(-v, -u, 1)11 +u +v
Using the properties of v' , it can be shown that v' is a connection on M,

but we will not do this here. Rather, we will first compute it explicitly, and
show that we get the same expression for the connection, in terms of the
chart-induced coordinates, as described in Example 6.1 for JR 2 . Thus we
may intuitively think of the connection as something which gives 'shape' to
a smooth manifold: at the level of the smooth structure on JR 2 , JR 2 should
not be thought of as a plane, since it does not have any shape. Only if we
endow the smooth manifold JR 2 with the flat connection does it become a
plane, and only if we endow it with the connection from Example 6.1 does
it become the saddle, etc.
Identifying tangent vectors to M as corresponding vectors in JR3 , we
obtain (with x, y, z E C 00 (M) defined by x(p) =u, y(p) =v, z(p) =uv, for all
p=(u ,v, uv) EM) that: Ou=(oux)ox+(ouy)i:ly+(ouz)oz = lox+Ooy+Yoz ,
which corresponds to the vector (1, 0, y) E JR 3 with respect to the basis
(ox, oy, oz). Similarly ov = oy + xoz = (0, 1, x) E JR 3 . So we have
v' Ou Ou= v' Ox +Y oz(ox +yoz) = v' Ox Ox + v' oJYOz) + yv' Oz Ox + yv' oJYOz) = o.
Similarly, v' 0 vov = v' oy +x 0 Joy +xoz) = 0. The nonzero contributions are
v'ovou = v'oyox + v'oy (yoz) + xv'ozox + xv'oJYoz)=O +oz+ 0 + O=oz ,

v'ouOv = v'oxOY + v'oJXoz ) + yv'ozOY + yv'oJXoz)=O +Oz + 0 + O=oz.
As at each p E M, fi.(p) is orthogonal in JR3 to :: (p) and also to :: (p),
v'vW = v'vW-(Wuvv+wvvu)((0,0,1),n.)n.
= V(Wu)ou + V(Wv)ov + (Wuvv + wvvu)oz
-(Wuvv + wvvu)((0, 0, 1), n)fi.
= V(Wu)ou + V(Wv)ov + (Wuvv + wvvu)(oz - ((0, 0, 1), fi.)fi.).
We have
oz - ((0, 0, 1), fi.)fi.
1
= (0,0,1)- 1 2 2 (-v,-u,1)= 1 12 2 (v,u,u 2 +v)
2
+u +v +u +v
So with
we have
By Exercise 6.3, v' satisfies (Cl)-(C3). Hence v' is a connection on M. ◊
6.3 Locality and connection coefficients/Christoffel symbols
In this section, we will learn that the covariant derivative can be computed
locally, and the connection information is encoded in any admissible chart
by m 3 functions I'fi called the 'connection coefficients or Christoffel sym-
bols'. We begin with the following crucial fact.
Lemma 6.1. (Locality of the connection).

Let M be a smooth manifold with a connection v' on M, p E M, and U c M
be an open set containing p. Let V, V, W, W E TJ M be such that in U, V = V
and W = W. Then (v'v W)p = (v' vW)p-
Proof. We prove this in two steps. First keeping V fixed, and then re-
placing also V. Let x E C 00 (M) be a bump function such that x = 1 in
a neighbourhood Uo of p, and x = 0 outside a neighbourhood U1 where
U1 c U. Then we have x(W - W) = 0 E TJ M. With c = 0 E JR, we have
v'v(x(W- W)) = v'v0 = v'v(c0) = cv'v0 = 0v'v0 = 0. But the left-hand

side v'v(x(W- W)) can be expanded using the Leibniz rule, and so
(Vx)(W-W) + xv'v(W-W) = o.
Evaluation at p yields (Vx)(p)·0 + l·((v'vW)p - (v'vW)v) = 0, as wanted.
Now we show (v'vW)p= (v'yW)p- With x as above, x(V- V) =0. So
~ ~ (C3) ~ ~ ~ ~
x(v'vW-v'vW) = v'x(V-v)W = v'oW = v'o.oW = 0v'oW = 0.
Evaluating at p, l((v'vW)v-(v'yW)p)=O, completing the proof. □

The result above allows us to 'restrict' a connection to open subsets of M.
Proposition 6.1. (Restriction of a connection to an open subset).

Let M be a smooth manifold with a connection v'. Let U be an open subset
of M, with the smooth structure induced from M. Then there exists a unique
connection vu on U such that for all vector fields V, W E TJ M, we have
v'v1u(Wlu) = (v'vW)lu-
Proof. Uniqueness: Suppose that y7u, vu are two such connections on

U. Let V, W E TJU. Let p E U. Take a bump function x such that
x = l in a neighbourhood Uo of p, and x = 0 outside a neighbourhood U1
where U1 c U. Then xV, xW E TJ M, and they agree with V, W in the
neighbourhood U0 c U of p. Thus by Lemma 6.1,
(v'f,W)p = (v'(xV)lu(xW)lu)p = ((v'cxvi(xW))lu)p
= (V(xV)lu(xW)lu)p = (Vf,W)p-
As the point p E U was arbitrary, and since the vector fields V, WE TJU
were arbitrary, it follows that y7u coincides with vu.
Existence: Let V, WE TJU, and p EU. To define (v'f,W)p, we first take
any bump function x as above, and define (v'f, W)p = (v' (xV) (xW) )p- This
is well-defined by Lemma 6.1. The smoothness of p >-+ (v'f, W)p E TpU
follows from the fact that for all p E Uo, we can use the same X, so that
we have (v'f,W)p = (v'cxvi(xW))p for all p E U0 , and the right-hand side
depends smoothly on pin U0 . We now check the connection axioms.
Let V, w, XE TJU, CE JR, and f E C 00 (U). Then for all p EU:
(v'f,(W +cX))p = (v'xv(x(W +cX)))v = (v'xv(xW))v + c(v'xv(xX))v
= (v'f,W +cv'f,X)p,
showing (Cl).
Also,
(v'v(JW))p = (v'x2v(x 2 JW))p
= ((x 2 V)(xf))(p)· (xW)p + (xf)(p) · (v' x2v(xW))p
= (VJ)(p)·(xW)p + (xf)(p)·x(p)·(v'xv(xW))p
= ((VJ)W + f·v'vW)p,
proving (C2). Finally,
(v'Jv +x W)p = (v' x 2 uv +x) (x 2 W))p
+ x(p)(v'xx(x 2 W))p
= (xf)(p) · (v'xv(x 2 W))p
= (xf)(p)·((xV)x)(p)·(xW)p + (xf)(p)-x(p)-(v'xv(xW))p
+l·((xX)x)(p)·(xW)p + 1-x(p)-(v'xx(xW))p
= 1-f(p)·0·Wp + f(p)-1-(v'vW)p + 1-0·Wp + 1-1-(v'~W)p
= (Jv'vW+v'~W)p,
showing (C3). This completes the proof. □
Exercise 6.9. The aim of this exercise is to show the following stronger version of
Lemma 6.1, which, roughly speaking, says that VvW is determined at a point
p by just knowing the value Vp of V at p and the values of W in any arbitrarily
small neighbourhood of p.
Let M be a smooth manifold with a connection V on M. Suppose that p E M
and U c M be an open set containing p. Let the vector fields V, V, W, W E TJ M
be such that Vp=Vp, and W=W in U. Then (VvW)p=('v\7 W)p-
Hint: Proceed as follows. Keeping V fixed, the claim about replacing W by W
was proved in Lemma 6.1. It remains to show (v'v-vW)p = 0. Set X = V-V.
Choose a chart (U, x) containing p, and write X = Xi Oxi, and note that X\p) = 0.
Using Proposition 6.1, we may use the restriction of the connection to U. Note
that (V x W)p = (v'~; 0.,, Wlu )p-
Henceforth, when considering the restriction to U, we will simply use the

notation v' instead of vu. In light of the above, we can now give the
following definition.
Definition 6.2. (Connection coefficients/Christoffel symbols).

Let M be an m-dimensional smooth manifold with a connection v', and
(U, x) be an admissible chart. The connection coefficients/Christoffel
symbols of v' with respect to the chart (U, x) are the m 3 functions I'fj,
l~i,j,k~ m, defined by
v' 0xl. Oxi = rt1· Oxk.
More explicitly, I'fj = dxk (v' a.,i Oxi ).
We can calculate the covariant derivatives using chart-induced components:

if V = Vioxi and W = WiOxi in a chart (U,x), then
Relabelling the dummy summation index i in the first summand by k, and

substituting v' o,,i Oxi = rfi Oxk in the second summand, yields
Remark 6.1. (Physicist's notation for covariant differentiation).

In a chart (U, x), given an f E C 00 (U), one sometimes meets the notation
1
f,iforOxif. IfVEToM,thenmU, · v'a,,iV= ( OxiV k +rjiVJ
k ") Oxk- So the
kth compoment of v' i'J.,; V is (v' i'J.,; v)k = vk ,i + rji Vi' which is written as
k k
V ;i = V ,i + r iik V 1. .
The notation Vk ;i for (v' 0,,; V)k may cause confusion, as the latter means we
take the kth component of the vector field v' 0,,i V, while the former suggests
that we are taking a new sort of derivative of only the kth component (and
this is clearly wrong, since (v' a,,; V)k does involve all the other components
of V too, as is evident by the right-hand side of the above display). We will
not use the notation V\i- *
Proposition 6.2. Let M be a smooth manifold with a connection v'. Let
(V, y), (U, x) be admissible charts for M, with connection coefficients de-
. l
noted by riJk·V
, riJk-U , respective y. T h en
Proof. Let us first show that for any function f E C 00 (M), we have
(oyiXr)oxrf = Oyif- Indeed, we have that the right-hand side is
which matches with the left-hand side of the equality we wanted to prove. In
order to show the claimed transformation law for the connection coefficients,
we use the definition of the connection coefficients in the chart (V, y), and
convert all the coordinate expressions relying on the (V, y) chart using the
basis elements induced by the (U, x) chart:

r:_t = dyk(v'oyi(lyi) = (dyk(oxe)dx 1)(v'coyjXr)iJ,,r((oyiX 8
)0x•))
= (oxeyk) dxe ( (Oyi xr)'\l i),,r ((OyiX 8 )0x•))
= (oxeyk)(Oyi xr) dxe ('\! i),,r ((OyiX 8 )0x•))
= (oxeyk)(OyiXr)dxe(oxr(OyiX 8 )0x• + (OyiX )r;f Ox•)
8
= (oxeyk)(OyiXr)Oxr(OyiX 8 )8! + (oxeyk)(OyiXr)(OyiX 8 )r;~u t5f

= (oxeyk)(OyiXr)Oxr(OyiXe) + (oxeyk)(OyiXr)(OyiX )r!f. 8
Using our observation above with f = OyiXe yields

r:_t = (oxeyk)(OyiXr)(OyiX )r!~u + (oxeyk)Oyi(OyiXe).
8

Thus rfj cannot be the components of some (1, 2)-tensor field. Indeed, the
presence of the second summand in the above transformation law prevents
them from being tensor field components. For a tensor field, if all the tensor
field components vanish at a point pin a chart (U, x), then thanks to the
multilinearity (the transformation law from Exercise 4.10), all the tensor
field components with respect to any other chart (V, y) covering p would
also vanish at p. But even if all the connection coefficients with respect to
(U, x) vanished at p, then the presence of the second summand above, can
prevent the connection coefficients with respect to a different overlapping
chart (V, y), from vanishing at p. The following example illustrates this.
Example 6.3. Consider IR 2 with its standard smooth structure and the flat
connection v'. Let us determine the connection coefficients with respect to
the chart (U = JR 2 , idIR2). We have
v'oxl-Oxi = 0,
for all i,j E {1, 2}, and so it follows that r:f
=0 for all i,j, k E {1, 2}. How-
ever, the connection coefficients do not necessarily vanish in other charts,
e.g. with polar coordinates, as we shall see explicitly now. Suppose that
(for ease of handling the angular coordinate) V = {(x, y) E IR2 : y > 0}, and
y(x, y) = (r, 0), where
r:=,v'x 2 +y 2 and 0:=cos- 1 x ,
y'x2+y2
and cos- 1 : (-1, 1) ---+ (0, 1r) is the inverse of the cosine function. Then,
using the above transformation rule, we have for example (where we use
the more suggestive 0, r to label the r-symbols rather than number indices)
r00 = 0 + (oxr)o0(00x) + (oyr)o0(00y)
x (-x) + y (-y) = -,Jx2+y2 = -r.
,Jx2+y2 ,Jx2+y2
Similarly,
and
Exercise 6.10. (Divergence of a vector field, given a connection).

Let v' be a connection on a smooth manifold M. Let VE TJ M. Then we know
Tf MxTJM 3 (O,X) >-+ O(v'xV) E C 00 (M)
is a (1, 1)-tensor field, say T. Contracting T then yields a C 00 (M) function
CT=: divV,
called the divergence of V. Show that in a chart (U, x), div V = o,,i Vi + Vi r;i.
(In particular, if M = R 3 , equipped with the standard smooth structure and the
flat connection, then the connection coefficients all vanish in the global chart
(R 3 , id), and so for a vector field V = V"' Ox+ VY Oy + vz Oz E TJ R 3 , we obtain
divV =o,,V"'+ OyVY+ OzVz.)
Exercise 6.11. Recall from Exercise 6.4 the definition of the torsion tensor field
corresponding to a connection v' on a smooth manifold M. Prove that if v'
satisfies the torsion freeness condition
(C4) v'vW - v'wV - [V, W] = 0 for all V, WE TJM,
then the Christoffel symbols with respect to any admissible chart satisfy rfi = rji.
Exercise 6.12. (Connection 1-forms).
Let v' be a connection on a smooth manifold M, and let (U, x) be an admissible
chart. Show that there exists a unique matrix of 1-form fields n{ E TfU such that
v'vo,,; = (O{V) o,,i for all VE TJ M. Hint: Act by dxk to obtain Of= rfi dxi.
Show that the connection coefficients can be recovered from the connection 1-form
field by the formula rfi = Of o,,i.
6.4 The Levi-Civita connection
In this section we will learn about the 'miracle' of semi-Riemannian ge-

ometry, namely that there is a unique torsion-free and metric-compatible
connection on a given smooth semi-Riemannian manifold (M, g).
Theorem 6.1. (Fundamental theorem of semi-Riemannian geometry).

Let (M, g) be a semi-Riemannian manifold. Then there exists a unique
connection v' on M that satisfies for all V, W, X E TJ M
(C4) (Torsion-free) v'vW - v'wV - [V, W] = 0
(C5) (Metric-compatible) V(g(W, X)) = g(v'vW, X) + g(W, v'vX).
Proof. Uniqueness: Suppose that such a connection v' satisfying (Cl)-
(C5) exists. Using the metric-compatibility and the torsion-freeness, we
have for any three vector fields X, Y, Z
Xg(Y, Z) = g(v'xY, Z) + g(Y, v'xZ) = g(v'xY, Z) + g(Y, v'zX + [X, Z])
= g(v' x Y, Z) + g(v' zX, Y) + g([X, Z], Y).
Cyclically permuting the X, Y, Z we have:
Xg(Y, Z) = g(v' x Y, Z) + g(v' zX, Y) + g([X, Z], Y)
Y g(Z, X) = g(v'y Z, X) + g(v' x Y, Z) + g([Y, X], Z)
Zg(X, Y) = g(v' zX, Y) + g(v'yZ, X) + g([Z, Y],X).
Subtracting the third from the sum of the first two gives
Xg(Y, Z) + Yg(Z,X) - Zg(X, Y)
= 2g(v' x Y, Z) + g([X, Z], Y) + g([Y, X], Z) - g([Z, Y], X).
Consequently, g(v' x Y, Z) is given by the following Koszul formula3 :
1
g(v'xY, Z) = 2(Xg(Y,Z) + Yg(Z,X) - Zg(X, Y)
-g([X, Z], Y) - g([Y,X], Z) + g([Z, Y],X)).
As Z E TJ M was arbitrary, this uniquely determines v' x Y (Exercise 5.10).
Existence: For fixed X, Y E TJ M, the map sending Z to the right-hand
side of Koszul's formula is C 00 (M)-linear:
1
Z,..... O(X, Y)(Z) := 2(Xg(Y, Z) + Yg(Z, X) - Zg(X, Y)
- g([X, Z], Y) - g([Y, X], Z) + g([Z, Y], X)).
Indeed, the terms -Zg(X, Y) and -g([Y, X], Z) are C 00 (M)-linear in Z.
The combination of Xg(Y, Z) with -g([X, Z], Y) is C 00 (M)-linear too: the
additivity is obvious, and we have homogeneity in scaling by f E C 00 (M):
Xg(Y, f Z) - g([X, f Z], Y)
= (Xf)g(Y,Z) + fXg(Y,Z)-g(f[X,Z]+(Xf)Z, Y)
= (Xf)g(Y, Z) + f Xg(Y, Z) -f g([X, Z], Y) - (Xf)g(Z, Y). (*)
3 After Jean-Louis Koszul (1921-2018), a French mathematician.
Similarly, Yg(Z,X) +g([Z,Y],X) = Yg(X,Z)-g([Y,Z],X) is C 00 (M)-

linear in Z. So O(X, Y) E Tf M. By Theorem 5.8, there exists a vector
field v'xY E TJM such that (v'xY/ = O(X, Y), i.e., v'xY = (O(X, Y))~.
We need to check that this v' satisfies (Cl)-(C5). This is straightforward,
albeit a bit tedious. (Cl) is immediate. To check (C2), we have firstly
O(X, JY)(Z)
1
= 2(Xg(JY,Z) + fYg(Z,X)-Zg(X,JY)
-g([X, Z], JY) - g([JY, X], Z) + g([Z, JY], X))
1
= 2((Xf)g(Y, Z) - (Zf)g(X, Y) + (Xf)g(Y, Z) + (Zf)g(Y,X))
+ JO(X, Y)(Z)
= JO(X, Y)(Z) + (Xf)g(Y, Z) = JO(X, Y)(Z) + (Xf)Yb(z).
Hence O(X, JY) = f O(X, Y) + (X !)Yb. Applying .U on both sides, we
obtain v'x(JY)=fv'xY + (Xf)Y.
For (C3), we first check that X >-+ O(X, Y)(Z) is C 00 (M)-linear. Ad-
ditivity is clear. Homogeneity in scaling by f E C 00 (M) can be seen as
follows. The terms Xg(Y, Z) and g([Z, Y], X) clearly respect the scaling
of X. The combination of Yg(Z,X) with -g([Y,X],Z) works as in(*)
above (and also similarly the combination -Zg(X, Y) - g([X, Z], Y), seen
by exchanging Y, Z):
Y g(Z, f X) - g([Y, f X], Z)
+ JY g(Z, X) - g(f[Y, X] + (Y f)X, Z)
= (Y f) g(Z, X)
= (Yf)g(Z,X) + JYg(Z, X)-fg([Y,X], Z)-(Yf)g(X, Z).
So O(JX+X, Y)(Z)=JO(X, Y)(Z) +O(X, Y)(Z) for all Z E TJM. Hence
we conclude that O(JX+X,Y)=JO(X,Y)+O(X,Y). As .U is C 00 (M)-
linear, (C3) holds.
Next, we show (C4): We have
O(X, Y)(Z) - O(Y, X)(Z)
1
= 2(Xg(Y,Z)+~-~-Xg(Z,Y)
- g([X, Z], Y) - g([Y, X], Z) + g([ Z, Y], X)
+ ~([~~]~~) + g([X, Y],Z)- ~q~::~:]_,_1:"))

= g([X, Y], Z) = [X, Y]b Z.
Thus we have O(X, Y) - O(Y,X) = [X, Y]b, and applying .U, we obtain
v'xY - v'yX = [X, Y], that is, (C4) holds.
Finally, we verify (C5): Firstly,

g(v' x Y, Z) = (v' x Y)b(z) = (O(X, Y)U)b (Z) = O(X, Y)(Z),
which is precisely the right-hand side of the Koszul formula. Hence we have
g(v' x Y, Z) + g(Y, v' x Z)
1
= -(Xg(Y,Z)+Yg(Z,X)-Zg(X,Y)
2 -~~ ~
- g([X, Z], Y) - g([Y, X], Z) + g([Z, Y], X)

+Xg(Z, Y) + ~ - Yg(X, Z)
- ~q~,-~]_,_,-~') - ~~[~2 ~]2!) + g([Y, Z], X))
= Xg(Y,Z),
proving (C5). This completes the proof. □
Proposition 6.3. Let (M, g) be a semi-Riemannian manifold, and v' de-

note the Levi-Civita connection. Then the connection coefficients with re-
spect to an admissible chart (U, x) are given by
gfk
rfi = rji = 2 (oxigje + oxigei - oxtgii).
Proof. Recall that [ox;, Oxi] =0 for all i,j (Lemma 3.3). Taking X =Ox;,
Y = Oxi, Z = Ox£ in the Koszul formula, we thus obtain,
1
g(v'a.,;oxi,Oxt) = 2(ox;g(oxi,Oxt) + Oxig(oxt,ox;)- oxtg(ox;,oxi))
1
= 2(ox;gje + Oxigfi - Oxtgij)-
Substituting v' a.,; Oxi = r;ioxr and multiplying throughout by the inverse
metric component gek (and summing over the es) yields
r~j = rji = r;i 8: = r;i gre gfk = g(r;i Oxr, Ox£) gfk
gfk
= g(v' o.,; Oxi, Ox£) gfk = 2 (ox;gje + Oxigfi - Oxtgij)-
Exercise 6.13. Let (M, g) be an m-dimensional Lorentzian manifold with the Levi-
Civita connection v'. Let (U, x) be an admissible chart for M, and define G = [gii]-
Let Sm denote the set of all permutations of {1, • • • m}. For a permutation 1r E Sm,
let sign 1rE {-1, 1} denote its signature 4 .
4 See page 210.
Using the determinant formula detG= I; (sigll7r)g1,..(1) · · · gm1r(m), show that

1'fE8m
Oxk detG = (-l)i+j det[Gij]Oxkgij•

Here Gii denotes the (i,j)th minor of G, that is, the matrix obtained by deleting
the ith row and jth column of G; see for instance [Artin(1991), §1.5].
Hint: Use the product rule to write the derivative as a sum of determinants of
matrices in which the ith row of G has been replaced by the row comprising
the derivatives of the entries of that row. Then in each determinant summand,
expand the determinant by going along the row.
Using the symmetry and invertibility of G(p) (p EU), show that
Oxk det G= (detG)gij Oxkgij,
where [gii] is the inverse of G. Hint: Use Cramer's formula [Artin(1991), §1.5].
Prove that Oxk(log.J-detG) = r;k•
Example 6.4. (Sphere). Recall Example 5.7, where we considered the
sphere S 2 as a Riemannian manifold using the metric g induced on the
tangent spaces from the usual Euclidean inner product on JR 3 . We know
that in the chart (U, cp) described there,
a-1 = [ g00 g0<1>] = [ 1 . o ] .

g<P 6 g<P<P O (sm 0)- 2
We use the letters 0, = 0 + 2 (8<t,g0<1> + 8<t,g0<1> - 00g<t,<t,)
= ½(o + 0 - 80(sin0)2) = -(sin0)(cos0).
Similarly,
Example 6.5. (Minkowski spacetime). Consider the Minkowski spacetime

(M, g) described in Example 5.6. Consider an inertial frame induced by the
choice of a p EM and an orthonormal basis {f0 , f 1 , f2 , f3 } for V with respect
tog. Then the chart-induced component matrix G = [gij] of the metric is
the constant diagonal matrix with diagonal entries -1, 1, 1, 1. Thus all the
connection coefficients rfj of the Levi-Civita connection in this chart are
zeros. ◊
Example 6.6. (Cylindrical spacetime). Consider the cylindrical spacetime

from Example 5.8 and the chart (U, cp) described there. In this chart, the
component matrix G = [gij] of the metric is the constant diagonal matrix
with diagonal entries 1, -1. So all the connection coefficients rfi of the
Levi-Civita connection in this chart are zeros. ◊
Example 6. 7. (FLRW spacetime). Consider the FLRW spacetime from

Example 5.9, namely M = Jxl~.3, where I:= (O,oo). For the admissible
global chart (M, idixIRa ), the chart-induced component matrix G = [gij] of
the metric is the diagonal matrix with diagonal entries -1, a 2 , a 2 , a 2 . Thus
a- 1 is also a diagonal matrix, with the diagonal entries -1, ,!-, , ,!, , ,!, . With
the indexing 0, 1, 2, 3 used for the variables t, x, y, z, we have, as an example,
the following connection coefficient of the Levi-Civita connection:
goo
r~l = 2(ox1g10 + Ox1g10 - Oxogu) +0+0+0
-1 2 -2aa .
= 2(0 + 0 - ooa ) = --2- = aa,
where a:= ~~. Similarly one can also calculate that for 1 ~ i, j, k ~ 3 that
rgo = 0, rgi = r?o = 0, r?j = 8ijaa, rio = 0, rij = rjo = 8ij ~' and r)k = 0.
◊
Example 6.8. (Schwarzschild spacetime). Consider the Schwarzschild

spacetime considered in Example 5.10. In the chart JR x (2m, oo) x U de-
scribed there, the metric component matrix G = [gij] is diagonal with the
diagonal entries -(1- 2;_"), (1- 2;_")- 1 , r 2 , r 2 (sin0) 2 . Thus for example, we
have the following connection coefficient of the Levi-Civita connection:
r~t = g;r (otgrt+otgrt-orgu) = -½ ( 1- 2~ )or (-( 1- 2~)) =; ( 1- 2~)-
Similarly,
t t m
rtr = rrt = r2(1- 2;."),
rr m rr ( 2m)
rr = - r 2(l-2;."), 00 = -r 1---;:- ,
0 0 1 0
rr0 = r0r = -, rcfuf, = -(sin0)(cos0),
r
</> </> cos0 cf, - cf, - 1
r 0</> = r ¢0 = ---:--0,
sm
rrcf, - rcf,r - -,
r
and all other r-symbols are zeros. ◊
Exercise 6.14. For the saddle surface M considered in Exercise 5.13, with the
induced metric g on M from the Euclidean inner product on 113 , determine the
Christoffel symbols for the Levi-Civita connection with respect to the global chart
given there. Note that these coincide with the connection coefficients determined
in Example 6.2, showing that the connection considered in Example 6.2 was in
fact the Levi-Civita connection corresponding to g.
Exercise 6.15. (Curvature tensor field).

Let M be a smooth manifold with a connection v'. For X, Y, Z E TJ M, define
R(X, Y)Z E TJM by R(X, Y)Z=v'xv'yZ - v'yv'xZ- v'[x,YJZ. Show that
• TJ M 3 X >-+ R(X, Y)Z E TJ M (for fixed Y, Z)
• TJ M 3 Y >-+ R(X, Y)Z E TJ M (for fixed Z, X)
• TJ M 3 Z >-+ R(X, Y)Z E TJ M (for fixed X, Y)
are all C 00 (M)-linear.
For any p EM, and x,y,z E TpM, set R(p)(x,y)z= (R(X,Y)Z)p, where
X, Y, Z E TJ M are any vector fields such that Xp = x, Yp = y and Zp = z. Use the
C 00 (M)-linearity of R to show that, R(p) : (TpM) 3 -+ TpM is well-defined.
Via R, we obtain a (1, 3)-tensor field R, called the curvature tensor field,
defined as follows: R(n, X, Y, Z) := O(R(X, Y)Z) for X, Y, Z E TJ M and for
n E Tf M. An admissible chart (U, x) is called an affine chart if all the connection
coefficients with respect to (U, x) are identically zero. We call M locally flat if
for every point p EM, there exists an affine chart (U, x) such that p E U. Show
that R=O for a locally flat manifold. (Later in Chapter 9, we will learn that the
converse also holds.)
Examples of locally flat semi-Riemannian manifolds are the Minkowski space-
time given in Example 5.6, and the cylindrical spacetime given in Example 5.8.
However, for the sphere S 2 considered as a Riemannian manifold as in Exam-
ple 5.7, show that in the chart (U, cp) given there, R(o0, 04>)00 = -04> #- 0, so that
R(d<f>, 00, 0<1>, 00) = -1 #- 0.
6.5 Covariant derivative of tensor fields
We started our discussion accepting that we know how to differentiate func-

tions f E C 00 (M) in a direction v E TpM, resulting in vf ER If we have a
vector field V, providing a 'distribution' of such directions, then we obtain
V f E C 00 ( M). We wanted to do a similar thing replacing f by a vector
field W. We realised that to do so we need additional structure, namely a
connection 'v' on M. How about wanting to differentiate a I-form field n
in the direction provided by V? And more generally even an (r, s)-tensor
field? It turns out that once we have a connection 'v', we are able to dif-
ferentiate any (r, s)-tensor field. The key to this is that we want a sensible
Leibniz rule, and this will motivate the definition given below.
For a (0,0)-tensor field, namely a C 00 (M) element, and a VE TJ-M, we

set v'v f = V f E C 00 (M). Note that this definition uses only the smooth
structure on M. We wish to define v'vn, where n E Tf Mis a I-form field.
Just as
for a (0, 0)-tensor field f, v'v f is a (0, 0)-tensor field,
for a (0, I)-tensor field W, v'vW is a (0, I)-tensor field,
we want v'vn to be a I-form field. We first note that the action of n on
a vector field Wis a smooth function, and so there is no problem defining
v'v(n(W)): it is V(n(W)) E C 00 (M). But now suppose that, however we
define v'vn E Tf M to be, we want the Leibniz rule:
(V(n(W)) =) v'v(n(W)) = (v'vn)(W) + n(v'vW).
Then we can 'solve for' v'vn, and obtain
(v'vn)(W) = V(n(W)) - n(v'vW).
So this seems to give the action of our desired I-form field v'vn on any
vector field W, and the right hand side is an element of C 00 ( M). In other
words, it defines the I-form field v'vn! In order to use this as the definition
ofv': Tf M-> TP M, we need to make sure that W ...... V(!1(W))-!1(v'vW)
is a I-form field. We do so below: For f E C 00 (M), W,X E TJ-M, we have
(v'vn)(JW +x)
= V(n (JW +X)) - n (v'v(JW +X))
= V(J!1W +nX) - !1((VJ)W + fv'vW +v'vX)
= (VJ)nw + JV(nw) + v(nx) - (VJ)nw - Jnv'vW - nvvx
= J(V(nw) - n(v'vW)) + v(nx) - n(v'vX)
= J(v'vn)(W) + (v'vn)(X).
Definition 6.3. (Covariant derivative of functions and of I-form fields).
Let M be a smooth manifold with a connection v', and V E TJ- M. If
f E C 00 (M), then its covariant derivative v'v f E C 00 (M) is defined by
v' v f = V f. If n E Tf M, then its covariant derivative v' vn E Tf M is
defined by (v'vi1)(W) = V(!1W) - i1v'vW, for all WE TJ-M.
Exercise 6.16. Let M be a smooth manifold with a connection v'. Let f E C 00 (M),
n, e E Tf Mand V, WeTJM. Show that v'v(fn+ 8)=(Vf)n+Jv'vn+ v'v8,
and v'tv+wO=fv'vO+v'wO.
Exercise 6.17. Let (M, g) be a semi-Riemannian manifold with the metric g
and the Levi-Civita connection v'. For vector fields V, W E TJ M, show that
v'v(W 0 ) = (v'vW) 0•
A similar strategy works with any (r, s)-tensor field T. By the desired
Leibniz rule, for 0 1, · · · , nr E TP M and V, W1, · · · , Ws E TJ- M, we have
V(T(01, · · · , or, W1, · · ·, Ws))
= v'v(T(01, · · · , nr, W1, · · ·, Ws))
= (v'vT)(01,- · ·, nr, W1, · · · , Ws)
r
+ ~ T(n1, · · · ,ni-l, v'vfl\ni+ 1, · · · ,nr, W1, · · ·, Ws)
i=l
s
+ ~T(01,--- ,nr,W1,··· ,Wj-1,v'vWJ,Wi+1,··· ,Ws),

j=l
and as before we solve for v'vT, yielding
(v'vT)(01,- · ·, nr, W1, · · · , Ws)
= V(T(01, · · · , nr, W1, · · · , Ws))
r
- ~ T(0 1, · · · ,ni-l, v'vOi,ni+ 1, · · · ,nr, W1, .. ·, Ws)
i=l
s
- ~T(01,--- ,nr,W1,··· ,Wj-1,v'vWj,Wj+1,··· ,Ws)-

i=l
Again, we must check (01, • • • , nr, W1 , • • • , W 8 ) >-+ [the RHS of the above],
is C 00 (M)-multilinear. We check this only in the case of a (1, 1)-tensor
field T, with the proof in the general case being analogous. We have, for
/ E C 00 (M), 8,0 E TfM, V, WE TJ-M, that
(v'vT)(fn+ e, W)
= V(T(JO + e, W)) -T(v'v(JO + 8), W) -T(JO + e, v'vW)
= V(JT(O, W) + T(e, W))-T((Vf)O + fv'vO + v'v8, W)
-JT(O, v'vW) -T(8, v'vW)
= (Vf)T(O, W) + JV(T(O, W)) + V(T(8, W))
-(Vf)T(O, W)
- f (T(v'vO, W)) -T(v'v8, W) - JT(O, v'vW) -T(8, v'vW)
= f (V(T(O,W)) -T(v'vO,W) -T(O, v'vW))
+ V(T(e, W)) -T(v'v8, W) -T(e, v'vW)
= f (v'vT)(n, W) + (v'vT)(e, W).
C 00 ( M)-linearity in the second slot follows similarly. The above motivates
the following.
Definition 6.4. (Covariant derivative of (r, s)-tensor fields).

Let M be a smooth manifold with a connection v'. For an (r, s)-tensor field
T, where r ~ 2 ors~ 1, and VE TJ- M, we define the covariant derivative
v'vT E T;M by
(v'vT)(01,- ··,or, W1, · · · , Ws)

= V(T(01, · · · , or, W1, · · · , Ws))
r
- ~ T(01, · · · ,ni- 1 , v'v0'.t,ni+1, · · · ,nr, W1, ·· ·, Ws)
i=l
s
- ~ T(01,··· ,nr,W1,··· ,Wj-1,v'vWj,Wj+1,··· ,Ws),
j=l
for 0 1 ' ... 'nr E Tf M and W1, ... 'Ws E TJ- M (and where an empty sum
is taken as the zero function).
From the definition we see that v'v(T + S) = v'vT + v'vS for (r, s)-tensor
fields T and S.
Exercise 6.18. Let M be a smooth manifold with a connection v'. Let R be a
(1, 3)-tensor-field such that it has the symmetry R(n, X, Y, Z) = -R(n, Y, X, Z)
for all n E Tf M and all X, Y, Z E TJ M. Show that the covariant deriva-
tive inherits this symmetry, that is, for n E Tf M and all V, X, Y, Z E TJ M,
(v'vR)(n, X, Y, Z) =-(v'vR)(n, Y,X, Z).
Exercise 6.19. Let M be a smooth manifold with a connection v'. Let f E C 00 (M),
TE T;M and VE TJM. Prove that v'v(JT)=(Vf)T + f(v'vT).
Exercise 6.20. Let M be a smooth manifold with a connection v'. Let TE r; M,

V, WE TJM and f E C 00 (M). Show that v'tv+w(T)=f(v'vT) + v'wT.
A manifestation of the metric-compatibility property (C5) of the Levi-

Civita connection on a smooth manifold with a metric, is the following
result. In the next chapter on 'parallel transport', we will see yet another
description of (C5), namely that the scalar product remains constant under
parallel transport induced by the Levi-Civita connection.
Theorem 6.2. Let (M, g) be a semi-Riemannian manifold, v' be the Levi-

Civita connection on M. Then v'.g = 0, that is, for all VE TJ-M, v'vg = 0.
Proof. For V, W, X E TJ- M,
(v' vg) (W, X) = V (g(W, X)) - g(v' v W, X) - g(W, v' v X) <~l 0.
As W, X were arbitrary, we conclude that v'vg = 0. □

If M is a smooth manifold with a connection v', and V, W TJ M, then we

E
had seen that v'vW can be found in an admissible chart (U,x) using the
components of V, W and the Christoffel symbols:
. k . ik k . ik
v'vW = (V 1 0xi W + V 3 W rij)Oxk = (V(W ) + V 3 W rij)Oxk-
A similar calculation can be done for the covariant derivative of 1-form
fields, and of more general tensor fields.
Example 6.9. (Covariant derivative of 1-form fields using chart-induced
components). Let M be a smooth manifold with a connection v', and
let (U,x) be an admissible chart. Let V E TJM and n E Tf M. Then
v'vO E Tf M, and so we can write v'vO = (v'vO)idxi in U. To determine
the components (v'vO)i, we compute
(v'vO)i = (v'vO)(ox;) = V((Okdxk)Oxi) - nkdxk(v'vio,,iOxi)
k k . . k R.
= V(Ok8i) - nkdX (V 1 v' a,,j Oxi) = vni - nk V 3 dx (rijOxe)
. R. k . k . . k
= vni - nkv1 rij81. = vni - nkv1 rij = v 1 axini - nkv1 rij" ◊
Exercise 6.21. Suppose that M is a smooth manifold with a connection v', (U, x)
is an admissible chart, VE TJ M, and TE Ti M. Show that
(v'vT);k = V(Tjk}+r~.Tjk v•- r;.T 1\ v•- r;;.TJr v•.
We now show that in a smooth manifold M with a connection v', the oper-
ations of taking the covariant derivative and that of contraction commute.
Proposition 6.4. Let M be an m-dimensional smooth manifold with a
connection v'. Let 1 ~ r, s ~ m and let T E TJ M. Then for all V E TJ M,
we have C}(v'vT) = v'v(C}T).
Proof. We will prove this only in the case when r = s = 1. The proof
in the general case is analogous. In an admissible chart (U, x), we have
v'v(CT) = V(CT) = V(T(dxi, ox;))= V(Tl). On the other hand,
C(v'vT) = (v'vT)(dxi, Oxi)
= V(T(dx\ox;))-T(v'vdxi,ox;) -T(dxi, v'vox;)
i kf.
i -T((V8ii - 81.V
= V(Ti) rjk)dx j , Oxi) -T(dx,
i V kj
rikOxi)
i ki j kj i i
= V(Ti) + V rjkTi - V rikTj = V(Ti ).
Consequently, C(v'vT) = v'v(CT). □
v' satisfies a Leibniz product rule over the tensor product ® of tensor fields.
Proposition 6.5. Suppose that M is a smooth manifold with a connection
v'. Let S E TJ M and T E r:; M. Then for all V E TJ M, we have that
v'v(S®T) = (v'vS)®T+S®v'vT.
Proof. We show this for S = n E Tf M and T E Tf M (and we will also see

this when SE TJ Min Exercise 6.22). The general case follows analogously
from these by an inductive argument (and by working in an admissible
chart (U, x), decomposing S as st
:::t oxii ® · · · ® Oxir ® dxi 1 ® · · · ® dxi•
and also using Exercise 6.19). Fore E Tf M, we have and W, XE TJ M
(v'v(O®T))(e, W,X)
= V(O(W)T(e, X)) - (n ® T)(v've, w, X)
-(O®T)(e, v'vW,X)- (O®T)(e, w, v'vX)
= V(OW)T(e, X) + O(W)V(T(e, X)) - O(W)T(v've, X)
-O(v'vW)T(e,x) - O(W)T(e, v'vX)
-------------
= (V(OW) - O(v'vW))T(e,x)
+n(W)(V(T(8,X))-T(v'v8,X)-T(e, v'vX))
= (v'vO)(W)T(e, X) + O(W)(v'vT)(e, X)
= ((v'vO) ® T)(e, w, X) + (n ® v'vT)(e, w, X)
= ((v'v0)®T+O®v'vT)(8,W,X).
Thus v'v(O®T) = (v'vO)®T+O®v'vT. □
Exercise 6.22. Let M be a smooth manifold with a connection v'. Let W E TJ M

and TE Tl M. For all VE TJ M, show v'v(W ® T) = (v'vW) ® T + W ® v'vT.
Exercise 6.23. Let (M, g) be a semi-Riemannian manifold with the Levi-Civita

connection v'. Let TE T; Mand VE TJ M. Show that v'v(g ® T) = g ® v'vT
and v'v(T®g) = (v'vT)@g.
Exercise 6.24. (v' commutes with•• and-~).

The results of this exercise will be useful later (in Proposition 9.6, p.190).
Suppose that (M, g) is a semi-Riemannian manifold with the metric g and the
Levi-Civitaconnection v'. ForT ET~ M, defineT~ E Tl M byT~(n, V) =T(n~, V)
for all n E TI' M, VE TJ M. Furthermore, for TE Tl M, define T 0 ET~ M by
T 0 (V, W)=T(V 0 , W) for V, WE TJM.
(1) Show that for all TE Tl M, (T 0 )~ =T.
(2) Show that for all T E T~ M, (T~)° = T.
(3) Show that for all TE Tl M, T 0 = C~(g ® T).
(4) Show that for all TE Tl M, and all VE TJ M, v'v(T 0 ) = (v'vT) 0 •
(5) Show that for all TE T~M, and all VE TJM, v'v(T~)=(v'vT)~.
Hint: Use (1),(2),( 4).
6.6 Lie derivative of tensor fields
Let M be a smooth manifold. Recall that if VE TJ- M, then the Lie deriva-
tive .Cv f := VJ for f E C 00 (M), and .CvW := [V, W] if W ETJ-M (Defini-
tion 3.15). We now extend this, and define the Lie derivative of arbitrary
tensor fields by demanding a Leibniz rule, akin to what was done in the
previous section for the covariant derivative. We begin by extending the
definition of the Lie derivative to 1-form fields.
Definition 6.5. (Lie derivative of functions and of 1-form fields).

Let M be a smooth manifold, and let VE TJ-M. If f E C 00 (M), then the
Lie derivative .Cv f E C 00 (M) off is defined by .Cv f = V f. If n E Tf M,
then the Lie derivative .Cvn E Tf M of n is defined by
(.Cvn)(W) = v(nw) - n[V, W] for all w E TJM.
Let us check .Cvn is a 1-formfield. Clearly, TJ-M 3 w ...... V(nW)-n[V, W]

is R-linear. The map is also C 00 (M)-linear: For f E C 00 (M), we have
(.Cvn)(JW) = V(n(JW))-n[v, JW] = V(JnW)-n(J[V, W]+(VJ)W)

= (VJ)(nw) + JV(nw) - Jn[V, w] - (vf)nw
= J(V(nw) - n[v, W]) = J(.Cvn)(W).
Exercise 6.25. Let M be a smooth manifold, and VE TJM. Let f E C 00 (M), and
n, 8 E Tf M. Prove that
C.v(fn + 8) =(C.vf)n + JC.vn + C.v8= (Vf)n + f C.vn + C.ve.
Analogous to Definition 6.4, we now give the following.
Definition 6.6. (Lie derivative of (r, s)-tensor fields).

Let M be a smooth manifold, and VE TJ-M. For an (r,s)-tensor field T,
where r ~ 2 or s ~ 1, the Lie derivative .CvT E T{ M is defined by
r
- ~ T(n1,- · · ,ni- 1,.cvn\ni+ 1, · · · ,nr, W1, .. ·, Ws)
i=l
s
- ~ T(n1,- · · , nr, W1, · · ·, Wj-1, .CvWJ, Wj+1, · · · , Ws),
j=l
for all !11, · · · , nr E Tf M and W1, · · · , Ws E TJ- M.

Again, we must check (01, • • • , nr, W1 , • • • , Ws) >-+ [the RHS of the above],
is C 00 (M)-multilinear. We check this for (1, 1)-tensor fields T. The proof
in the general case is analogous. For f E C 00 (M), e, n E Tf M, WE TJ M,
(CvT)(JO+ e, W)
= V(T(JO + e, W)) -T(Cv(JO + 8), W) -T(JO + 8,CvW)
= V(JT(O, W) + T(e, W))
-T((VJ)O + JCvO + Cve, W)-JT(O,CvW) -T(8,CvW)
= (VJ)T(O, W) + JV(T(O, W)) + V(T(e, W))
-(VJ)T(O, W) - JT(Cvn, W) -T(Cve, W)
-JT(O, CvW) -T(e, CvW)
= f(V(T(O, W)) -T(Cvn, W) -T(O, CvW))
+ V(T(e, W)) - T(Cve, W) - T(e, Cv W)
= J(CvT)(O, W) + (CvT)(e, W).
The C 00 (M)-linearity in the second slot follows similarly.
Exercise 6.26. (Killing vector field and Killing's equation).
Let (M, g) be a semi-Riemannian manifold with the Levi-Civita connection v'
induced by the metric g. A vector field V E TJ M is called a Killing" vector field
if .Cvg=O. Show that if Vis a Killing vector field, then for all X, YE TJ M,
• g([V, X], Y) + g(X, [V, Y]) = V(g(X, Y)), and
• (Killing's equation) g(v' x V, Y) + g(X, v'y V) = 0.
Exercise 6.27. Let M = IR.3 be equipped with the standard smooth structure and
the metric g described in Exercise 5.12. Show that the vector fields Lx, Ly, Lz
given in Exercise 3.21 are Killing vector fields for g.
Recall that a complete vector field V on a smooth manifold M is one whose

integral curves are defined on all of R. Then its flow is a family of diffeo-
morphisms 1Pt : M -+ M, t E R Suppose that Mis a semi-Riemannian
manifold with a metric g. Recall that isometries on (M, g) are special types
of diffeomorphisms. It is natural to ask:
Q: For what vector fields V are the flow maps isometries?
A: Killing vector fields.
We prove this below. First, we have the following analogue of Proposi-
tions 3.1 and 3.2.
5 After the German mathematician Wilhelm Killing (1847-1923) who made contribu-
tions to Lie theory.
Proposition 6.6. Let M be a smooth manifold, and V E TJ- M be a com-

plete vector field with the flow {1Pt : M ----> M, t E R}. Let T E T~ M.
Then
.CvT = lim (1Pt)*T - T.
t-+O t
Here we mean the pointwise limit, that is, the right-hand side means the
map which sends X, Y E TJ- M to the function given by
. ((('lt,t)*T - T)(X, Y))(p)
M 3 p ,__. 1I m---------.
t-+O t
That this defines an element in TJ M will follow from the proof below, since
it matches the left-hand side .CvT ET~ M.
Proof. To ease the notation, we will use
ForX,YETJM, andpEM,
IL(((1Pt)*T-T)(X, Y)(p))
= IL(T('lt,tp)((d'lt,t)pXp, (d'lt,t)vYv) -T(p)(Xp, Yp)}
= IL(T('lt,tp)((d'lt,t)pXp, (d'lt,t)vYv) -T('lt,tp)(X,t,,p, Y,i,,p))
+IL(T('lt,tp)(X,t,,p, Y,t,,v) -T(p)(Xp, Yp)}. (6.2)
We will consider the last two summands above separately. First, with
1'v (t) := 1PtP, t E R, we have v'Yp ,P = Vp as 1'v is an integral curve of V. With
f :=T(X, Y), we have for the second summand that
82 := IL(T(¢tP)(X,t,,p, Y,t,,p) - T(p)(Xp, Yp)}
d(T(','p(t))(X'Yp(t), Y'Yp(t))) (O)

dt
d(f o 1'v)
dt (0) = v'Yp,vf = Vvf = Vp(T(X, Y)).
For the first summand in (6.2), we add and subtract
We will also use 1Pt o 1P-t = 1Pt-t = ¢0 = idM, giving

So
81 := L(T('¢tP)((d'¢t)pXp, (d1/Jt)pYp) -T(1/JtP)(X,t,,p, Y,i,,p))

= L(T('¢tP)((d'¢t)pXp, (d1/Jt)pYp)} - L(T(1/JtP)(X,t,,p, (d1/Jt)pYp)}
+L(T('¢tP)(X,t,,p, (d1/Jt)pYp)} - L(T(1/JtP)(X,t,,p, Y,i,,p)}
= L(T('¢tP)((d'¢t)pXp, (d1/Jt)pYp) -T(1/JtP)(X,t,,p, (d'¢t)pYp)}
+L(T('¢tP)(X,t,,p, (d1/Jt)pYp) -T(1/JtP)(X,t,,p, Y,i,,p)}
= L(T('¢tP)((d'¢t)pXp - X,t,,p, (d1/Jt)pYp)}
+L(T('¢tP)(X,t,,p, (d1/Jt)pYp - Y,i,,p))
= L(T('¢tP)((d'¢t)p(Xp - (d1/J-t),t,,pX,t,,p), (d1/Jt)pYp)}
+L(T('¢tP)(X,t,,p, (d1/Jt)p(Yp - (d1/J-t),t,,pY,t,,p))}.
Proposition 3.2 yields 6
So
81 + 82 = Vp(T(X, Y)) - T(p)([V, X]p, Yp) - T(p)(Xp, [V, Y]p)

= (V(T(X, Y)) - T([V, X], Y) - T(X, [V, Y])}(p)
= ((.CvT)(X, Y))(p).
Thus lim (((1/Jt)*T -T)(X, Y))(p) = ((.CvT)(X, Y))(p).

t-+O t □
Finally, we will prove the following important result.
Theorem 6.3. Let (M, g) be a semi-Riemannian manifold, and V E TJ M

be a complete vector field with the flow {1/Jt : M-+ M, t ER}. Then V is a
Killing vector field if and only if for all t E R, 1Pt is an isometry.
Proof.
'If part': Suppose that each1Pt is an isometry. Then (1/Jt)*g = g, so that
Proposition 6.6 immediately yields .Cvg = 0, that is, V is a Killing vector
field.
6 At least formally, this is clear, but we do not include a detailed proof here (which can
be carried out by considering a chart containing p and all 1PtP for t close enough to 0,
and using the smoothness of gij, "/p, and using the chart representative Jacobian matrix
of (d'I/Jt)p)-
'Only if part': Let .Cvg = 0. Suppose that X, Y E TJ- M. For a fixed s E JR,
set X = ('1/Js)*X and Y = ('1/Js)*Y. For anyp EM, ((.Cvg)(X, Y))('I/JsP) = 0,
and so by Proposition 6.6,
O = lim g('l/Jt'l/JsP)((d'l/Jt),J,.pX,j,.p, (d'l/Jt),J,.pY,p.p) - g('l/J 8 p)(X,J,.p, Y,p.p)
t--+O t
= lim g('l/Jt+sP)((d'l/Jt),J,.p(d'l/Js)pXp, (d'l/Jt),J,.p(d'l/Js)pYp)-g('l/J.p)((d'l/Js)pXp, (d'l/Js)pYp)
t--+O t
= lim g('l/Jt+sP)((d('l/Jt o 'l/Js))pXp, (d('l/Jt o 'l/Js))pYp) - g('l/J.p)((d'l/Js)pXp, (d'l/Js)pYp)
t--+O t
= lim g('l/Jt+sP)((d'l/Jt+s)pXp, (d'l/Jt+s)pYp) - g('l/J.p)((d'l/Js)pXp, (d'l/Js)pYp)
t--+O t
= d(g('l/JuP)((d'l/Ju)pXp, (d'l/Ju)pYp)) ( )
~ s.
As s E JR was arbitrary, we conclude that the map
JR 3 a 1--+ g('I/JuP)((d'I/Ju)pXp, (d'I/Ju)pYp)
is constant, so that (putting a = 0)
g('I/JuP)((d'I/Ju)pXp, (d'I/Ju)pYp) = g(p)(Xp, Yp) for all a E JR.
Hence ('1/Ja)*g = g for all a ER □
Example 6.10. Consider JR 3 with the standard smooth structure, and the
Riemannian metric g, given in the global chart (JR3, (x, y, z),..... (x, y, z)) by
g = dx ® dx + dy ® dy + dz ® dz (Exercise 5.12). Recall from Exercise 3.21,
that the flow maps for the vector fields
Lx = YOz - ZOy
Ly= ZOx -XOz
Lz = XOy-YOx
are rotations about the x-, y-, z-axis, respectively, which are isometries
(Exercise 5.12). In Exercise 6.27 above, we had verified that Lx, Ly, Lz are
Killing vector fields. ◊
Notes
The proof of Theorem 6.3 is based on [O'Neill{1983), Chap. 9, Prop. 21].
Chapter 7
Parallel transport
We will now learn that a connection induces a way of 'parallel transporting'

vectors along a curve 'Y· So ifp=7(a) and q=7(b) are points along the curve
'Y, then P1b : TpM -> TqM will be a linear map which will transport vectors
from TpM to TqM 'parallelly' along the curve. We will explain the meaning
of this below. In this manner tangent spaces at different points can now be
'connected', and this is the motivation behind the name 'connection'. We
will also see that if Mis a semi-Riemannian manifold, with v' taken as the
Levi-Civita connection, then the parallel transport map is an isometry.
7.1 Vector fields along curves
The tangent vector along a curve exists only along the points of the curve.
In order to talk of the acceleration, we would like to differentiate this 'vector
field' that 'lives along the curve'. This prompts the following definition.
Definition 7.1. (Vector field along a smooth curve).

Let M be a smooth manifold, and 'Y : J -> M be a smooth curve, where I is
an open interval in R A vector field along 'Y is a smooth map V : J -> TM
such that 1r o V = 'Y, where 1r : TM -> M is the canonical projection. We
denote the set of all vector fields along 'Y by TJ7.
137
Example 7.1. (Restriction of a vector field to a curve is a vector field

along the curve). Let M be a smooth manifold, and let 'Y : I --> M be a
smooth curve, where I is an open interval. Suppose that W E TJ M. Set
V(t) = ('Y(t), W"Y(t)) fort E J. Then V = W o 'Y· As W: M ->TM, TI-> M
are smooth, so is V. Moreover, 1ro V =1ro (W o'Y) = (1ro W)o'Y=idMo'Y='Y·
So V is a vector field along 'Y. ◊
A vector field along a curve is not necessarily the restriction of a vector field
on M to 'Y(J). Indeed, the curve may intersect itself, say at p="((a) ="f(b),
where a, b E J with a # b, and if the vector field V along 'Y is such that
V"Y(a) # V"Y(b), then we cannot create a global extension W of the V to a
vector field on M, since there is a problem of specifying what Wp ought to
be. The picture below illustrates the problem when we take the tangent
vectors along 'Y (which we will see is a vector field along 'Y)- However, a
'local' extension is possible for a part of the curve in the neighbourhood of
'Y(to) if v"Y,"Y(to) #0; see Remark 7.1.
V"Y ,"Y(b)
Lemma 7 .1. Let M be a smooth manifold, and 'Y : I --> M be a smooth

curve such that 'Y(J) c U, where I is an open interval in JR. and (U, cp) is an
admissible chart for M. Suppose that V : J --> TM is such that 1r o V = 'Y.
Write V(t) = ('Y(t), Vi(t)ox' ,"Y(t)), t E J. Then V is smooth if and only if
Vi E C 00 (J) for all i E {1, · · · , m}.
Proof. We recall that the admissible chart (U, cp) for M induces an ad-
missible chart (U,<p) for TM, where U = {(p,v): p EU, v E TpM} and
<p(p,v)=(cp(p),v(x 1 ),· · · ,v(xm)). Then <p: U--> <p(U) c IR. 2 m is a diffeo-
morphism. Now suppose that V is smooth. Then <po V is smooth, that is,
I 3 t >-> (<po V)(t) = (cp('Y(t)), V 1 (t), • • • , vm(t)) is smooth. In particular,
Vi E C 00 (J).
Pamllel tmnsport 139
Conversely, if each of the Vi is smooth, then (since cp o 'Y is already smooth)

it follows from the above that cp o V is smooth. But then V = cp- 1 o (cp o V)
is smooth too. □
Proposition 7 .1. Let M be a smooth manifold defined by the atlas A, and

'Y : I -+ M be a smooth curve, where I is an open interval in R For any
chart (U, x) in A, define the open subset Iu = 7- 1 u of I. Suppose that
V : I -+ TM is such that 1r o V = 'Y. Then the following are equivalent:
(1) V is smooth.
(2) For any chart (U, x) in A such that Iu -# 0, we have Vi E C 00 (Iu) for
all i E {1, · · · , m}, where V(t) = ('Y(t), Vi(t)Oxi,-y(t)), t E Iu.
Proof.
(1)=?(2): Let (U,x) EA and t E Iu. Then 7(t) EU. By the continuity
of 'Y, there exists a small enough open interval It c Iu containing t. Then
7(It) c U. By Lemma 7.1 above, Vil1, is C 00 • As the choice oft E Iu was
arbitrary, vi E C 00 (Iu ).
(2)=?(1): Lett EI. Let (U, x) EA be such that 7(t) E U. Then t E Iu -# 0-
As Iu is open, there exists an open interval It c Iu containing t. By the
hypothesis, Vi E C 00 (Iu), and so Vi E C 00 (It) as well. By Lemma 7.1, VI It
is smooth. As t E I was arbitrary, V is smooth. □
For a vector field V E TJ7 along a curve 'Y : I -+ M, just as we have been
doing with vector fields on M, we will identify V(t) = ('Y(t), v(t)) (t E I),
where v(t) E T'Y(t)M, with v(t).
Example 7.2. (Tangent/Velocity vector field along a curve).

Let M be a smooth manifold and 'Y : I -+ M be a smooth curve, where I
is an open interval in R Fort EI, define V'Y(t)=v'Y,'Y(t) E T'Y(t)M. Then V'Y
is a vector field along 'Y· In any admissible chart (U, x),
d(xi o 7)
V'Y,'Y(t) = dt (t) Oxi,'Y(t), t E Iu,
'Y)
so that v;-= dtd(xi o
, and these are smooth on I u ( as 'Y is smooth). ◊
Let M be a smooth manifold and 'Y : I -+ M be a smooth curve, where

I is an open interval in R For V, W E TJ7, we define V + W E TJ7 by
(V + W)(t) = V(t) + W(t), for all t E I. Also, if f E C 00 (I), then we define
f ·VE TJ7 by (f-V)(t)=f(t)V(t) for all t EI. Then TJ7 is aC 00 (I)-module
(just like TJ M is a C 00 (M)-module).
7.2 Covariant derivative along a curve
Now we will learn that if v' is a connection on a smooth manifold M,

then it induces for every smooth curve 'Y : J - M, a 'covariant derivative'
v'v-r : TJ'Y - TJ"f. Roughly speaking, it allows us to differentiate a vector
field along a curve in the direction of the tangent/velocity vector field V7
along the curve 'Y· We begin with the following result, where we first work
with a chart. Below, we consider U as a smooth manifold with the atlas
given by the single chart (U, x), and the connection vu (Proposition 6.1).
For each smooth curve 'Y: J - U, we denote by TJ'Yu the vector fields along
'Y in the smooth manifold U, that is, they are smooth maps V : J - TU
such that 1r o V = idu, and 1r : TU - U is the canonical projection.
Lemma 7 .2. Let M be a smooth manifold with a connection v', and let
(U, x) be an admissible chart for M. For each smooth curve 'Y : J - U,
where I is an open interval in JR, there exists a map v'v,'Y : TJ'Yu - TJ'Yu
such that the following hold for W, X E TJ'Y u, c E JR and f E C 00 ( J) :
• (JR-linearity) v'v,'Y (c W + X) = c v'v,'Y W + v'v,

'Y
X
• (Leibniz) v'v, (! W) = ddlf W + f v'v, W
'Y t 'Y
• If WE TJU and W 7 (t) = W(t) for all t EJ, then we have

(v'v'Y W)(t) = (v'v-y,-y(t) W)7(t) for all t E J.
We note that by Exercise 6.9, the notation v' v-r,-r<•J W is legitimate, since
v' x Y at a point p depends only on Xp (and on the values of Y in a neigh-
bourhood of p). Thus, here we mean that if t 0 E J and V E TJU is any
vector field that coincides with the vector v7 ,7 (to) at the point 'Y(to) E M
(and such a vector field can be constructed using a bump function; see
Lemma 3.1), then v'v-r,-r<•l W := (v'y-W) 7 (t)·
Proof.
Existence: Given WE TJ'Y~ write W = WiOxi- Also,
TT() •i();i h ·i() d(xio'Y)()
v 7 t =v7 ,7 (t) ="f t uxi,'"l'(t), w ere 'Y t := dt t.
· dWk
Set Wk=&' and define
(v'v'Y W)(t) = (Wk(t) + rt/'Y(t))')'i(t) Wi(t))Oxk,7(t), t E J. (7.1)
As wk+(rfjo'Yhj wi E C 00 (J), we have v'v'Y WE TJ'Yu by Proposition 7.1.
We now verify that v'v,'Y has the
•
claimed
.
properties. The JR-linearity follows
.
immediately since (cW +X)'=cW'+X'.
Similarly, the Leibniz rule holds since
v'v,,(fW) =(!wk+ /Wk+ (rti O 'Yhi Jwi)Oxk = dt W + fv'v,, W

~ 1 ~ .
Let to E J and VE T0 U satisfy V--y(to) = V--y(to) =v--y,--y(to) = V'(to)Oxi,--y(to)· In
particular, if V = Viox;, then Vi('Y(t 0 ))= Vi(t 0 ). Also, writing W = Wiox;,
we have Wi('Y(t)) = Wi(t) for all t E J. We have
(v' v-,,-,(to) W)--y(to)
= (v'v(WiOxi))--y(to) = (Viv'a.,j (WiOxi))--y(to)
= ( (Vi ('Y(to))Oxi ,--y(to))(Wk) + vi ('Y(to))Wi('Y(to) )rti ('Y(to))) Oxk,--y(to)

~k
= ( V--y,--y(to)(W ) + V i (to)W i (to)rii('Y(to))
k )
Oxk,--y(to)
d(Wk O 'Y) · · k )
= ( dt (to)+ V 3 (to)W'(to)rii('Y(to)) Oxk,--y(to)
= (::Vk (to)+ Vi(to)Wi(to)rti('Y(to)) )axk,--y(to) = (v'v,, W)(to)-

This completes the proof of the 'existence' part.
Uniqueness: Suppose that there exists a map D--y : TJ'Yu -+ TJ'Yu satisfying
the given properties. Given WE TJ'Yu, write W(t) = Wi(t) Oxi,--y(t), t E J.
Then we have, using the Leibniz rule, that
- dWi -
(D--y W)(t) = (D--y(W'Oxi,--y(•)))(t) = &(t)Oxi,--y(t) + W'(t)(D--yOxi,--y(•))(t). (*)
Let t E J. To evaluate (D--yOxi,--y(-))(t), we take any vector field V E TJU

such that at this particular fixed t, we have
V--y(t) = V--y(t) = Vi(t)Oxi,--y(t) = V--y,--y(t) = 'l(t)Oxi,--y(t)·
We note that Vi(t)=iti(t). Since Oxi E TJU extends Oxi,--y(•) E TJ'Yu,
(D--yoxi,--y(•))(t) = (v' v-,,-,<•J Oxi )--y(t) = (v' voxi )--y(t) = (v'Vio,,j Oxi )--y(t)
~. ~. k
= (V3 v' O.,j Oxi )--y(t) = (V 3 rii Oxk )--y(t)
= Vi('Y(t))rt/'Y(t))Oxk,--y(t)
= Vi(t)rt/'Y(t))Oxk,--y(t) = ,yi(t)rti('Y(t))Oxk,--y(t)·
Substitution in(*) yields (7.1), showing the claimed uniqueness. □
Theorem 7.1. Let M be a smooth manifold with a connection v'. For each
smooth curve 'Y: I-> M, where I is an open interval in R, there exists a
map v'v-r : TJ-'Y-> TJ-'Y such that the following hold for W, X E TJ-'Y, c ER
and f E C 00 (I):
• (R-linearity) v'v,,(cW +X) = cv'v-r W + v'v-rX

• (Leibniz) v'v-r(JW) = dt W + fv'v-r W
• If WE TJ- M and W7(t) = W(t) for all t EI, then we have
(v'v-r W)(t) = (v' v-r,-r<•J W) 7(t) for all t EI.
If (U, x) is an admissible chart for M, then
(v'v'Y W)(t) = (Wk(t) + rtj('Y(t))'yi(t)Wi(t))oxk,')'(t) (t E Iu),
where W(t) = Wi(t)oxi,')'(t), fort E Iu := 'Y- 1 u.
Proof.
Uniqueness: Let there exist a map D7 : TJ-'Y -> TJ-'Y satisfying the given
properties. Let t 0 E I, and (U, x) be an admissible chart containing the
point 'Y(t 0 ). Let Jc I be an open interval containing t 0 such that 'Y(J) c U.
Given WE TJ-'Y, write W(t) = Wi(t) Oxi,'"l'(t), t E J. Using the Leibniz rule,
(D7W)(to) = (D7 (Wioxi,')'(·)))(to)

dWi -
= &(to)Oxi,')'(to) + W'(to)(D 7 oxi,')'(·))(to). (*)
To find (D7 oxi,'"l'(·))(to), we first take any vector field V E TJ- M extending
V7 (to), so that V7(to) = V7 (to) = Vi(to)Oxi,'"l'(to) = v7 ,7 (to) = i'i(to)Oxi,')'(to)·
We note that Vi(t 0 ) =i'i(t0 ). Thus
(D70xi,7(•))(to) = (v' v-y,-y(to)Oxi )7(to) = (v' y-Oxi )7(to) = (v' Vio,,i Oxi )7(to)
~. ~. k
= (V3 v' O,,j Oxi )7(to) = (V 3 r i j Oxk )7(to)
~. k
= V 3 ('Y(to))rij('Y(to))oxk,')'(to)
= Vi(to)rfj('Y(to))Oxk,7(to) = 7i(to)rfj('Y(to))oxk,7(to)·
Substitution in (*) now yields (D7 oxi,'"l'(·))(to) = (v'~-r W)(to), showing the
claimed uniqueness.
Existence: We will define v' V-r by using the charts from an atlas A defining
the smooth structure on M. (Then by uniqueness, it follows that the con-
struction of v'v-r does not depend on the atlas A.) Let t 0 EI, and (U, x) be
a chart from A containing the point 'Y(t0 ). Let Jc I be an interval around

to such that 'Y(J) c U. Fort E J, (v'v-r W)(t) := (v'~-r W)(t), i.e.,
(v'v'Y W)(t) = (Wk(t) + rtj ('Y(t))-yi (t) W\t)) Oxk,y(t),
where W(t) = Wi(t) Oxi,-y(t), V-y(t) =V-y,-y(t) =-/(t) Oxi,-y(t), and
wi =::vi, -l(t) = :~xio'Y) (t).
Then v' V-r as defined above has the claimed properties by the proof of the
existence part of Lemma 7.2. Also, we note that if (fi, x) is another chart
containing the point 'Y(t), then the uniqueness part of Lemma 7.2 applied
to U n ft shows that
(v'v,'Y W)(t) = (v'v,"uW)(t)
'Y
= (v'i'Y W)(t).
That I 3 t >-+ (v'v-r W)(t) is an element of TJ'Y follows from Lemma 7.1.
The last claim on the expression for v' V-r W in a chart follows from the
existence part combined with the uniqueness part. □
Exercise 7.1. Let M be a smooth manifold with a connection v'. Let I, J be open
subsets of IR, and J 3 u >-+ h( u) E J be a C 00 function. Let , : J -+ IR be a smooth
curve. From Exercise 2.4, the velocity vector field of , o h : J -+ M is given by
V-yoh(s) = h(s)V-y(h(s)) for s E J. For a given Z E TJ,, define WE TJ(, oh) by
W(s) = Z(h(s)), s E J. Show that (v'v-roh W)(s) = h(s) (v'v-rZ)(h(s)), s E J.
Exercise 7.2. The aim of this exercise is show a further strengthening of the results
in Lemma 6.1 and Exercise 6.9, namely that v'vW is determined at a point p by
just knowing
• the value Vp of V at p and
• the values of W along a curve , passing through p whose velocity vector at
p is Vp (instead of knowing W in an arbitrarily small neighbourhood of p as
in Exercise 6.9).
Let M be a smooth manifold with a connection v', p E M and v E TpM. Let
V, V E TJ M be any two vector fields such that
Vp = V = Vp.
Let , : J -+ M be a smooth curve, where I is an open interval in IR containing 0,
such that 1 (0) = p and V-r,-r(o) = v. Suppose that W, W E TJ M are such that
W-y(t) = W-y(t) (t E I).
Prove that
(v'vW)p=(v'yW)p-
1
Hint: Define the vector field X E T 0 , by X(t)
~ and consider
= W-y(t) = W-y(t)
(v'v-rX)(0).
Remark 7.1. (Extending a vector field along a curve to a locally defined

vector field). We now show the following technical fact, which will be used
later on when we discuss geodesic deviation. We had seen earlier that given
a point pin a smooth manifold M, and a vector v E TpM, there exists a
vector field VE TJ- M such that Vp = v (Lemma 3.1). Now we show that if
we have a smooth curve 'Y passing through a point p with a nonzero velocity
at p, and a smooth vector field V along 'Y, then in the vicinity of p, Vis the
restriction of a vector field on M. The precise statement is given below.
Fact: Let M be an m-dimensional smooth manifold. Suppose 'Y : I -+ M
is a smooth curve, where I is an open interval in IR. containing 0, and
v--y ,y(0) =I= 0. Let V E TJ-1. Then there exists an open interval J c I
with OE J, an admissible chart (U, x) containing 1(J) , and a vector field
VE TJ-U such that V--y(t) = V(t) for all t E J.
This can be shown as follows. By Remark 2.2, there exists an open interval
Jc I containing 0, and an admissible chart (U,x) containing 1(J), such
that (x o 1)(t) = (t,0, · · · ,0) for all t E J. Write V(t) = Vi(t)ox' ,--r(t) for
t E J. Let S be the 'strip'
S := {y = (y1, · · · , ym) E x(U) : y 1 E J} c x(U).
For 1 ,;;; i ,;;; m, define the smooth functions Vi : S -+ IR. by Vi (y) = Vi (y 1 )
for y = (y 1 , · · · ,ym) ES c x(U). Set
V(q)=(Vi ox)(q)ox,,q for q E U:=x- 1 s C ff.
Then the chart (U,xlu) is admissible, and VE TJ-U (as its components are
smooth). For all t E J, 1(t) EU, and so
V('Y(t)) = (Vio x)('Y(t)) Oxi ,,1(t) = v\t, 0, ... 'O)ox',y(t) = Vi(t)ox' ,y(t) = V(t).
This completes the justification of the fact.
*
We now give the following natural definition.
Definition 7.2. (Parallel vector field along a smooth curve).
Let M be a smooth manifold with a connection v', and 'Y : J -+ M be a
smooth curve, where I is an open interval in R A vector field W E TJ-1
along 'Y is said to be parallel along 'Y if (v' v., W) (t) = 0 for all t E J.
Parallel transport 145
Example 7.3. (Sphere). Recall Example 5.7 on page 87, where we consid-
ered the sphere S 2 as a Riemannian manifold using the metric g induced
from the Euclidean inner product on IR 3 . Let 'v denote the Levi-Civita con-
nection induced by g. Consider the meridian , with a constant longitude
</>o E (0, 27r), which in the chart (U, <p) described in Example 5.7, is given by
,(t) = ((cos</>o)(sint),(sin</>o)(sint) , cost) , for all t E (0, 7r). We claim that
the vector field W along , given by the latitudinal vector field O<f>, is not
parallel along,. Firstly, we have W(t) = o<l>,'!(t) , and so W = 1, w 0 = 0,
w
giving W = 0 = 0. Next, v"Y,'!(t) = ,:.t9 00 ,"Y(t) + ,y o<l>,"Y(t) , where
,y9 (t) = d(O O , ) (t) = dt (t) = 1 and ,y<f>(t) = d(</> 0 ,) (t) = d</>o (t) = 0.
dt dt ' dt dt
Thus, keeping track of the nonzero terms, we have that
('vv,, W)(t) = (W 0(t) + rt(,(t)hJ (t) Wi(t))o0,")'(t)

+ (W<f>(t) + rt (,(t)hj (t) wi(t)) o<t>,"Y(t)
= r: 0 (,(t)h 0 (t) w<t>(t) 00 ,")'(t) + r! 0 (,(t)h0 (t) w<t>(t) o<t>,"Y(t) ·
0 cost
But from Example 6.4 on page 123, r</> 0 =0 and r</> 0 (,(t))=-.-, so that
smt
cost cost
('vv,, W)(t) = -.-t · l · 1 o<l>,"Y(t) = -.-t o<l>,"Y(t)·
sm sm
So W is not parallel along , .
Consider the equatorµ, given by µ(t) = (cost,sint,0), for all tE (0,27r).

We claim that the vector field X along µ given by the longitudinal vector
field 00, is parallel alongµ. Firstly, we have X(t) = 00,µ(t), and so X 9 = 1,
X <t> -= o, g1vmg
· · x· 0 =- o =- x· <t> . N ext, vµ,µ(t) -- µ· 0 v0,µ(t)
::,, + µ· <t> vq,,µ(t),
::,, where
fi(t) = d(O O µ) (t) = d(~) (t) = 0 and µ<f>(t) = d(</> 0 µ) (t) = dt (t) = 1.
dt dt ' dt dt
Thus, keeping track of the nonzero terms, we have that

(v'v,,X)(t) = (X 0(t) + rfi(µ(t))µ/(t)Xi(t))o0,µ,(t)
+ (X<f>(t) + rt (µ(t)) µ/ (t) xi(t)) a<f>,µ,(t)
= r:<1>(µ(t))µ<f>(t)x 0(t)o0,µ,(t) + rt<1>(µ(t))µ<f>(t)x 0(t)oq,,µ,(t)
= 0-1-180,µ,(t) + cos01
sinO O=f ·l-loq,,µ,(t) = 0 + 0 = 0.
Here we used 0(µ(t)) = i for all t E (0, 2n). So X is parallel alongµ. ◊
Exercise 7.3. Consider the same curveµ from Example 7.3: µ(t) = (cost,sint,0),
t E (0, 211-). Take Y(t) = Oef>,µ(t), t E (0, 211-). Prove that Y is parallel alongµ.
Conclude that for a, /3 E R, aX + f3Y is also parallel along µ, where X is the
vector field along µ given in Example 7.3, namely X(t) = oo,µ(t), t E (0, 21r).
In Theorem 6.2 (p.128), we had seen a manifestation of the metric-

compatibility property (C5) of the Levi-Civita connection on a smooth
manifold with a metric g, namely v'. g = 0. We now meet another avatar
of this.
Proposition 7.2. Let (M, g) be a semi-Riemannian manifold with the met-

ric g and the Levi- Civita connection v'. Let 'Y : J -> M be a smooth curve,
where I is an open interval in R Let W, X E TJ'Y. Then
d(g(W, X)) = g(v' W X) + g(W v' X).

dt V-y ' ' V-y
Here t arguments are suppressed: :~g(W, X)) = :~g('y(t))(W(t), X(t))) etc.
Proof. It is enough to show this in admissible charts that meet 'Y(J). Let
(U, x) be such a chart. Then we have
:;g(W, X)) = (Wi xi + wi_xi) g(oxi, Oxi) + Wi xi :;g(oxi, Oxi )) .
Let us first note that

d(g(oxi, oxi )) (t) = d(g('y(t))(oxi,-,(t), Oxi,-y(t))) (t)
dt dt
= V-y,-y(t)(U 3 p >-+ g(p )(oxi,p, Oxi ,p)).
Now fix t EI, and let V be any vector field such that V-y(t) = v-y,-y(t)· Then
from the above,
Now we use the metric compatibility property (C5) of the Levi-Civita con-
nection to expand the right-hand side (V(g(ox, , Oxj)))(,(t)) , obtaining
!~g(ox,,Oxj)) (t)
= (g(v'vox,, oxj ))(,(t)) + (g(ox,, v'voxj ))(,(t))

= g(,( t) )( (v' V Oxi )-y(t), Oxj ,-y(t)) + g(,( t) )( Oxi ,-y(t), (v' VOxj ) -y(t) )
= g(,(t) )( (v' V-, Oxi )(t) , Oxj ,-y(t)) + g(,(t) )( Oxi ,-y(t) , (v' V-, Oxj )(t) ).
Thus, again suppressing the argument t , we have
d(g(W,X))
dt
= (Wi xi + Wi _xi) g( ox,, oxj) + Wi xi :~g( ox,, oxj))
= g(Wiox,, X) + g(W, _xi oxj) + WiXi (g(v'v,.,ox,, oxj) + g(ox,, v'v,.,oxj)
= g(Wiox,,X) +g(W,X1oxj) +g(Wiv'v,.,ox, , X) +g(W,X1 v'v,.,oxj)
= ~(~~~x_..:~) + ?.(~,.-\1_~:~~ + ~(~~~~::~:'~ ~) + ~~~:~_1 .?~l~:~

= g(v'v,., W, X) + g(W, v'v,.,X) (using Leibniz rule for v'v,.,).

Corollary 7.1. Let (M, g) be a semi-Riemannian manifold with the metric
g and the Levi-Civita connection v'. Let , : J ---> M be a smooth curve,
where I is an open interval in JR. Let W, XE TJ, be parallel along,. Then
d(g(W, X)) = O
dt '
that is, the scalar product g(,(t))(W(t), X(t)) = c, a constant, for all t E J.
Hence, in a Riemannian manifold, for pointwise 'unit length vector fields'

W,X along a smooth curve that are parallel along,, the 'angle' between
W(t) and X(t) stays the same.
w
7 .3 Parallel transport
In this section, we will meet the parallel transport map. Given a smooth
curve 'Y : J -> M, where M is a smooth manifold with a connection v',
and I is an open interval in R containing the points a and b the parallel
transport map
P:b : T'Y(a)M -> T"f(b)M
will transport tangent vectors from T'Y(a)M to tangent vectors in T"f(b)M,
parallelly along the curve 'Y· In this manner, it 'connects' the two 'distant'
tangent spaces using the connection v'. The transportation itself will be
done by working locally in charts, setting up a system of ordinary differen-
tial equations for components, and using a known existence and uniqueness
result from the theory of differential equations.
Proposition 7 .3.
Let I be an open interval in R and A: J-> Rmxm have C 00 entries. Then
given a E J and B E Rm, there exists a unique Y : J -> Rm such that the
entries of Y are C 00 , and
Y(t) = A(t) Y(t), t EI,
{
Y(a) = B.
Proof. This follows from a C 1 -version 1 of the result (where the A is just
assumed to be continuous), by the following argument: First view A as c 0 ,
from which we deduce that there is a unique C 1 solution Y. But then look
at the right-hand side AY, which is C1, since A is C 1 and so is Y. Hence
we conclude that Y = A Y is C 1 , which is the same as saying that Y is
C 2 . Again Y = A Y is now C 2 , since the right-hand side is the product of
A, Y E C 2 , and so Y E C 3 , and so on. □
Theorem 7.2. Let M be a smooth manifold with a connection v'. Let

'Y : J
-> M be a smooth curve, where I is an open interval in R. Let a E J
and w E T'Y(a)M. Then there exists a unique W E TJ'Y such that W is
parallel along 'Y and W (a) = w.
Proof. First suppose that 'Y(J) lies entirely within an admissible chart
(U, x). We write W = WiOxi. Then Wis parallel along 'Y if and only if for
all t EI, (v'v,, W)(t) =0, i.e.,
Wk(t) + ,yi(t)rt/'Y(t)) Wi(t) = o, 1 ~ k ~ m.
1 See for example, [Apostol(1969), §7.21].
Moreover, we have W(a) = w if and only if Wi(a) = bi, 1,( i ,( m, where

w =: bi Ox' ,-y(a)· Define the vector B = (b 1 , · · · , bm). Then the claimed W
exists if and only if the following initial value problem for a system of first
order linear differential equations, in the unknown Y = (W 1 , • • • , wm) , has
a solution:
{ Y(t) = A(t)Y(t) ,
t E J,
Y(a) = B.
Here A(·) is the matrix whose entry in the ith row and jth column is given
by the function --l(•)qkb(·)) E C 00 (J). By Proposition 7.3, there exists
a unique C 00 solution Y. Consequently, there exists a unique W E TJ---y
parallel to ')' satisfying W (a) = w.
I --y(/3)
--y(/3- !)
Now 2 suppose that there does not exist a single admissible chart covering
all of --y(J). Let (a, (3) be the maximal interval containing a on which a
W E TJ---yl(a,/3) exists which is parallel along '°Yl(a,/3) and W(a) = w. If
(3 = sup J and a= inf J, then we are done. Suppose that (3 inf J is analogous.) There is an admissible chart (V, y) such
that --y(/3) EV. We can then choose a 8 > 0 such that --y( (/3 - 8, (3 + 8)) E V.
We know that there exists a vector W(,B - !)-
Taking this as an initial
condition, there exists a unique vector field W on (/3 - 8, (3 + 8) such that
and Wis parallel to '°Yl(/3-'5,/3+.5)· But by the uniqueness, this W coincides

with Won (/3 - 8, (3). However, as W itself is defined on (/3 - 8, (3 + 8), this
means that W could have been extended beyond (3, a contradiction to the
maximality of (a, (3). □
2 We follow [Lee(2018) , Theorem 4.32].
Definition 7.3. (Parallel transport of a vector along a smooth curve).

Let M be a smooth manifold with a connection "V. Suppose that 'Y : J -> M
is a smooth curve, where I is an open interval in R. Let a E J. For a vector
w E T"Y(a)M, let PJw E TJ'Y be the unique vector field along 'Y such that
PJw is parallel along 'Y and (PJw)(a) = w. Then PJw is called the parallel
transport of w along 'Y· Define the maps PJt : T"Y(a)M-> T"Y(t)M (t E J), by
PJtw = (PJw)(t) for all w E T"Y(a)M.
Proposition 7.4. Let M be a smooth manifold with a connection "V. Let

'Y : J -> M be a smooth curve, where I is an open interval in R Let a, b E J.
Then Pdb : T"Y(a)M -> T"f(b)M is a linear isomorphism.
Proof. Let w,x E T"Y(a)M, and denote by W,X, respectively, their par-
allel transports along"(: That is, W, XE T1h, "Vv-r W = 0, W(a) = w, and
"Vv-rX = 0, X(a) = x. Let c ER Then by the R-linearity of "Vv-r, we have
"Vv-r(W + cX) = "Vv-r W + c"Vv-rX = 0 + 0 = 0,
and by the definition of addition in TJ'Y,
(W + cX)(a) = + cX(a) = w + ex.
W(a)
Hence W + cX is the parallel transport of w + ex along 'Y· In particular,
Pdb(w +ex)= (W + cX)(b) = W(b) + cX(b) = Pdbw + cPdbx.
Thus Pdb is a linear map. Next we show that Pdb : T"Y(a)M-> T"f(b)M is
invertible. If Mis m-dimensional, then since dimT"Y(a) = m = dimT"f(b)M,
it is enough to show injectivity. Let w E T"Y(a), and W denote the corre-
sponding parallel transport of w along 'Y· Then Pdb(w) = W(b). But now
suppose we want the parallel transport of the vector x := W(b) E T"f(b)M
along the curve 'Y· Clearly W already satisfies "Vv-r W =0 and W(b) =x. By
the uniqueness established in Theorem 7.2, we conclude that the parallel
transport of x along 'Y is W. In particular, P~ x = W (a) = w. Hence
(P~ 0 pdb)w = P~(P:bw) = P~x = W = idT-y(a)MW-
As w E T"Y(a)M was arbitrary, we have P~ o Pdb = idT-rcaiM• Thus Pdb is
injective. So Pdb: T"Y(a)M-> T"f(b)M is a linear isomorphism. □
Theorem 7.3. Let M be a semi-Riemannian manifold with metric g and

the Levi-Civita connection "V. Suppose that 'Y: J-> M is a smooth curve,
where I is an open interval in R Let a, b E J. Then the parallel transport
map Pdb : T"Y(a)M -> T"f(b)M is an isometry, that is, it is an invertible
linear map preserving the scalar product.
Proof. We have already seen that P:b is linear and invertible. It remains
to show that it preserves the scalar product. Let w, x E T"Y(a)M , and denote
by W, XE TJ'°Y, respectively, the corresponding parallel transports along '°Y·
Corollary 7.1 (p.147) implies that
g('Y(b))(P:bw, P:bx) = g('Y(b))(W(b) , X(b)) = g('Y(a))(W(a) , X(a))
= g('Y(a))(w , x).
Hence P:b : T"Y(a)M-> T"Y(b)M is an isometry. □
Remark 7.2. In this chapter, we have seen that once we have a connection
v' on a smooth manifold, then it induces a notion of a covariant derivative
v' v.., along each smooth curve ')', and also a notion of parallel transport
along each smooth curve ')'. One can also show3 that the parallel transport
maps determine the covariant differentiation operators along curves via
(v'v, W)(t) = lim P;t(W(r)) - W(t) , WE Th ,
-y T-->t T - t
and also the manifold's connection by
( nv W) = 1· P;o(W"Y(h)) - Wp 1
v P h~ h , p E M, V, W E T0 M,
where')' is any smooth curve such that 'Y(O) = p and v'°Y,P = VP. As we will
not need this, we will not prove this here. *
Example 7.4. (Parallel transport under the flat connection on IR.m).
Consider IR.m with the standard smooth structure and the flat connection
v'. We use the global admissible chart (IR.m, id JR= ). Then for any p E M,
we have a linear isomorphism
m,m
11'>.
( 1
3 W = W , ·· · ,W
m) >--4
l i ::i T m,m
W Vxi,p E pl1'>. •
Let '°Y : I -> IR.m be any smooth curve, where I is an open interval in JR. that
passes through the two points p, q E IR.m. Let a, b E J be such that 'Y( a) = p
and 'Y(b) = q.
3 See e.g. [Lee(2018) , Thm. 4.34, Cor. 4.35].

Claim: Pdb(1,w) = 1,w for all w E !Rm, i.e.,

Pdb(wiOxi,p) =WiOxi,q for all W = WiOxi,p E Tp!Rm.
To see this, let W be the parallel transport of w along 'Y· Decompose
W(t) = Wi(t) Oxi,'Y(t)· Then for all k, Wk(t) + ffj('Y(t)) 'Yi(t) Wi(t) = 0.
But in the chart (!Rm, idR."'), all the f-symbols are identically zero, and so
wk= 0, giving wk= ck, a constant. As Wk(a)=wk, where W=WkOxk,p,
we obtain ck =wk, 1 ~ k ~m. Hence W(t) = Wk(t)oxk,'Y(t) = wkOxk,'Y(t)· In
particular, P1bw= W(b) =WkOxk,q- ◊
In the example above, the parallel transport that maps vectors in TpM to
vectors in TqM was independent of the path 'Y connecting the two points
p, q EM. However, this was because the space was 'flat' (and in the calcu-
lation above, in fact the terms containing 'Yj vanished in our chosen chart
as the f-symbols happened to be zeros). The parallel transport between
two tangent spaces at p, q E M in general does depend on the curve used. In
fact, it is a manifestation of the 'curvature' of the manifold. The following
is an illustration of this in the case of the sphere.
Example 7.5. (Sphere). Consider the sphere S2 as a Riemannian manifold
with the metric induced from the Euclidean inner product on JR 3 , and the
corresponding Levi-Civita connection. In Example 7.3 (p.145), we had
considered the smooth curveµ, given by µ(t) = (cost,sint,0), t E (0,21r).
In the admissible chart (U, cp) considered there, the chart map gives the
usual spherical polar coordinates (0u,</P). In this chart (U,cp), let the
vector field X alongµ be given by X(t) = o0u,µ(t), t E (0, 21r). Then we had
seen that X is parallel along µ (the equator in the chart U). Let a:= ~ and
b := 3~, so that p :=µ(a)= (0, 1, 0), and q := µ(b) = (0, -1, 0). Then with
x = 00u,P = X(a), we have P::bx =X(b) = 00u,q.
Now consider another admissible chart (ff, ip), where we use new spher-
ical coordinates (0u, ¢P), measuring the polar 0u angle of a point from the
positive x-axis (instead of the usual positive z-axis), and measuring the
azimuthal angle ¢P from the negative z-axis (instead of the usual positive
x-axis). Consider the curve µ given by µ(t) = (0, sin t, - cost), t E (0, 21r),
and take Y(t) = -oc/>u,µ,(t)' t E (0,21r).
By Exercise 7.3 (or by a direct calculation), Y is parallel alongµ. With
a=~ and b=3~, we have µ(a)= (0, 1, 0) =p, and µ(b) = (0, -1, 0) =q, and
(*)
Y(a) = - oc/>u,p = o0up
, = x.
That (*) holds is convincing based on the following picture, but a justifica-
tion is given at the end of this example.
~ (•)
Referring to the picture above, P/:b x = Y(b) = - o,,,u
'f' ,q
= - o0uq
'
= - P/:b x.
Thus the result of parallel transporting x from p to q is different along the
two curves µ and µ.
Justification of ( *) and ( *):
(2.2) (p.22) implies that TpS 2 3 v >-+ v = (vx , vy , vz) E IR.3 is injective.
Thus it suffices to show (00upX,00upy,09upz)
, , ,
= - (o,,,uo/~
x,o,,,u
o/~
y,o,,,u
o/~
z)
o
in order to prove 0 u,P = - ov,v·
We have
o( (sin 0)( cos¢))
00
o( (sin 0)(sin ¢))
00
o(cos0)
~
Ul
Also,
o(cos0)
~
o( (sin 0)(sin ¢))
o,p
o(-(sin 0)(cos ¢))
o,p 0=f
t/>=t=f
Thus - o,t,fl,p = o0u,p, that is, ( *) holds. Similarly,
o( (sin 0)( cos¢))
o0
:;(sin
o(cos0)
O)(filn ¢)) - [_ l~
~ -
o( cos 0)
;~sin O)(sin,;))
o(-(sin0)(cos¢))
00 &=f o<f>
1>= 3;
justifying o'¥,,,u,q = o0 u ' q, that is, ( *) holds. ◊
Chapter 8
Geodesics
In the previous chapter, we learnt that a connection v' on a smooth manifold

M allows us to differentiate a vector field W along a curve 'Y, using the
induced operator v' V-r, where V.,, is the velocity vector field along the curve.
Thus, we can now talk about the acceleration along a curve by taking
W = V.,, itself, so that v' V-r V.,, measures how fast V.,, changes as we move
in the direction of V.,,, which exactly coincides with our intuitive notion of
acceleration. If the acceleration everywhere along the curve is zero, then
we call the curve a 'geodesic'. In the spacetime context, geodesics are the
worldlines of 'freely falling observers'.
Geometrically speaking, geodesics are the 'straightest possible curves'
in a smooth manifold M with a connection v', and are the generalisation
of straight lines from the flat Euclidean plane case. However, we will also
see that we should not think of geodesics 'Y : J -> M merely in terms of the
image 'Y(J) of points it describes, but the parametrisation is crucial, and
a reparametrisation of a geodesic may no longer be a geodesic. Roughly
speaking, only affine linear maps of the arc-length parametrisation are ad-
missible for maintaining the geodesic nature of a curve.
8.1 Geodesic equation
Definition 8.1. (Geodesic).

Let M be a smooth manifold with a connection v'. A curve 'Y : J -> M,
where I c JR is an open interval, is called a geodesic if
a(t) := (v'v-r V.,,)(t) =0 for all t E J.
We call a(t) E T.,,(t)M the instantaneous acceleration at the point 'Y(t) EM.
We refer to v' V-r V.,, = 0 as the geodesic equation.
155
Trivially, any constant curve, 'Y = p E M, satisfies the geodesic equation,

and so is a geodesic. In a chart, the geodesic equation v' V-r V7 = 0 is a second
order ordinary differential equation for the curve coordinates.
Proposition 8.1. Let M be an m-dimensional smooth manifold with a

connection v'. Let 'Y : J -> M be a smooth curve, where I c JR is an open
interval. Then 'Y is a geodesic if and only if for each admissible chart (U, x)
such that 'Y(I) n U =I= 0, for all t E Iu := 'Y- 1 u, and all 1 ~ k ~ m,
d i d"i
Here 'Yi = xi o 'Y, :Yi = d7 , -yi = d7 , and rfj are the Christoffel symbols in
(U,x).
Proof. This follows from Theorem 7.1 (p.142), as V7 (t) ="fi(t)Oxi,'"l'(t)· □
The following example shows that when we think of a geodesic 'Y : J -> M,
we should not only think of the range 'Y(J), as the parameterisation matters.
Example 8.1. Consider JR 2 with the standard smooth structure and the
flat connection v'. In the chart (JR2 , id), all the connection coefficients
vanish. If Jc JR is an open interval, then a curve I 3 t ...... 'Y(t) = (x(t), y(t))
is a geodesic if and only if we have x(t) = 0 = jj(t) for all t E J, that is,
(x(t),y(t)) = (xo, Yo)+ tv for some (xo, Yo), v E JR2 . So 'Y is a straight line.
However, not all curves whose image lies along a straight line will be a
geodesic. As an example, consider µ : JR -> JR2 , µ(t) = (t3, t 3 ) fort E R
Then V,,(t) = 3t2 Ox,µ(t) + 3t2 oy,µ(t), and so we have :yx(t) = 3t2 = lyY(t),
giving ,:;,x(t) = 6t = -yY(t). Thus the geodesic equation is not satisfied, as
-yk(t) + rfj('Y(t)) :Yi(t) lyi(t) = 6t =I= 0 fort =I= 0. Soµ is not a geodesic.
Hence, geodesics lie along straight lines, but not all curves lying along a
straight line are geodesics. ◊
Exercise 8.1. (Sphere). Consider the sphere S2 as a Riemannian manifold with

metric g induced from the Euclidean inner product on 113 • Let S2 be given the
Levi-Civita connection obtained from g. Show that the geodesic equations in the
chart (U, cp) given in Example 7.3 (p.145) are:
0-(sin0)(cos0)(¢)2 =0,
¢+2(cot0)0¢=0.
Suppose the geodesic is traversed at unit speed, i.e., g(,(t))(v"Y,"Y(t), v"Y,"Y(t)) = 1 for
all t E I (see Proposition 8.2, where it will be shown that the squared length of the
tangent vector is constant along a geodesic). Show this gives (sin 0) 2 (¢) 2+(0) 2 = 1.
Geodesics 157
Writing the cf>-component of the geodesic equation as

d · J d ·
-d log¢>=....,.= -2(cot0)0 = -2-d log(sin0),
t ¢> t
prove that ~ = .c for a constant c. Using the unit speed relation, show that
Sill 0
c2
1---
(sin 0) 2 '
if 0 > 0. Dividing the above expressions for~ and 0, prove that
d</> c( cosec 0) 2
d0 = yl - c2 - c2(cot 0) 2
Substitute t = ~ cot 0, to show that for a constant a,
vl-c2
sin(a-ef>) = ~cot0.
vl- c2
Using the trigonometric angle-addition formula for sin(a - ¢>), show that this
equation can be rearranged to read (,(t), n)JRa = 0 for a suitable fixed nonzero
vector n E IR.3. Thus, ,(t) lies in the intersection with S2 of a plane passing
through the origin with normal n, that is, , is a 'great circle' on the sphere.
Proposition 8.2. Let M be a semi-Riemannian manifold with a metric

g and the Levi-Civita connection "V. Let I c R be an open interval and
'Y : J --+ M be a geodesic. Then there exists a constant c E R such that for
all t E J, g('y(t))(v"Y,"Y(t), v"Y,"Y(t)) = c.
Proof. This is an immediate consequence of Proposition 7.2 (p.146), since
d(g(,( ·)) (v"Y,"Y(·), v"Y,"Y(·))) d(g(,( ·)) (V"Y (·), V"Y(·)))
dt dt
= g(,{))(("Vv'Y V")', V"Y) + g('y(·))(V")', "Vv'Y V"Y)
= g(,,(-))(0, VJ+ g('y(·))(V")', 0) = 0. □
Exercise 8.2. Let (M, g) be a semi-Riemannian manifold, and let X E TJ M be a

Killing vector field. Suppose that the curve , : J -+ M is a geodesic, where I c R
is an open interval, and V-y,y(t) =I= 0 for all t E /. Show that
::(,(·))(v-y,-y(•),X-y(•)) (t) = 0 (t E J).
Definition 8.2. (Spacelike, timelike, lightlike/null curves).

Let (M, g) be a Lorentzian manifold with the metric g. A smooth curve
'Y : J --+ M, where I is an open interval in R, is called
• timelike if for all t E J, v"Y,"Y(t) is timelike.
• null/lightlike if for all t E J, v"Y,"Y(t) is lightlike.
• spacelike if for all t E J, v"Y,"Y(t) is spacelike.
An arbitrary smooth curve in a Lorentzian manifold may not necessarily

have one of these 'causal characters'. However, a geodesic with respect
to the Levi-Civita connection necessarily does. In particular, a timelike
geodesic cannot change its causal character.
Exercise 8.3. Show that each geodesic in a Lorentzian manifold with the Levi-
Civita connection, possesses a causal character, i.e., it is either timelike, spacelike
or lightlike. Hint: Proposition 8.2.
Exercise 8.4. Let (M, g) be the Schwarzschild spacetime (Example 5.10, p.89),
p E S 2 , r E (2m, oo ). Consider a stationary timelike observer 'Y : R-+ M given by
-y(t) = (t, r, p), t ER. Show that 'Y is not a geodesic (that is, stationary observers
are not freely falling).
8.2 Existence and uniqueness
We now show that given any point p E M and a direction v E TpM, there
exists a geodesic 'Y : I -+ M defined on some open interval I c R containing
0, such that "f(0) = p and v"Y,"Y(o) = v, called the 'geodesic passing through
pin the direction of v'. To show this, we will use the following result, which
follows from a well-known existence and uniqueness result from the theory of
ordinary differential equations; see e.g. [Apostol(1969), Thm. 7.19, p.229].
Proposition 8.3. Let Ve Rm be an open set, Fh E C 00 (V), l~i,j,k~m,
P E V, v E Rm. Then there exists an open interval I c R containing 0, and
a map
J 3 t 1-+ Y(t) = (Y 1 (t), · · · , ym(t)) EV
which is a solution to the following initial value problem:
f-k(t) + Fij(Y(t))Yi(t)Yi(t) = 0 (l~k~m)
{ Y(0) = P (8.1)
Y(O) = V.
Moreover, any two solutions Y: I-+ V and Y: 1-+ V coincide on In I.
Proof. For all l~k~m, let us define Ak(t) = Yk(t) and Bk(t) = Yk(t).
Set A= (A1,- ••,Am) and B = (B1,- • • , Bm). Then Y satisfies the given
initial value problem if and only if (A, B) is a solution to the following
initial value proble{m, ~:: ~~;(A)B' B; ;::::::

(8.2)
A(O) = P
B(O) = V.
Geodesics 159
Let r > 0 be such that the closed ball B(P, r) with center P and radius r
is contained in V. Then the function
K := B(P, r) x B(v, r) _. JR 2 m
(a 1 · · · am b1 · · · bm) >-+ (b1 · · · bm -F-1-(a)bilJi · · · -F'!'.'(a)bilJi)
~ ' ' ' ' ' ' 1,J '' 1,J
=:a
is a Lipschitz function (since its first order partial derivatives are all
bounded on the compact set K). So by [Apostol(1969), Thm. 7.19, p.229],
there exists a C 1 solution (A, B) : J -> IR 2 m on some interval I containing
0, satisfying the initial value problem (8.2). But as the functions Fij are all
smooth, the right-hand sides of the differential equations in (8.2) are C 1 .
So (A, B) is C 1 , i.e., (A, B) is C 2 . Again as the Fij are all smooth, the
right-hand sides of the differential equations in (8.2) are C 2 , implying that
(A, B) is C 2 , i.e., (A, B) is C 3 • Continuing in this manner, (A, B) E C 00 (J).
So there exists a C 00 ( J) solution Y: = A to (8.1).
By the local uniqueness of solutions [Apostol(1969), Thm. 7.19], Y, Y
coincide in a small interval (-1:, 1:) around 0. Set
Tmax = sup{T E J" J: Y(t) = Y(t) for all t ~ r}.
We want Tmax = sup(J n J). ~ Let Tmax < sup(J n J). ~ As Y, Y~ are C 1 ,
Y(Tmax) = Y(Tmax) =: Pmax and Y(Tmax) = Y(Tmax) =: Vmax•

Considering a new initial value problem with the initial conditions
Y(Tmax) = Pmax and Y(Tmax) = Vmax, we conclude that y and y must coin-
cide in a neighbourhood of Tmax, contradicting the choice of Tmax• Hence
Y(t) = Y(t) for all nonnegative t E J n J. A similar argument shows the
agreem~nt also for nonpositive values oft E J n J. So Y(t) = Y(t) for all
t E J n J. □
Remark 8.1. (Smooth dependence on v).

We remark that with the same notation as in the theorem above, it can be
shown that there exists a neighbourhood N of v, an 1: > 0, and a C 00 map
Y : N x (-1:, 1:) -> V such that for each w EN, Y(t) := Y(w, t), t E (-1:, 1:)
solves the initial value problem
fk(t) + ffj(Y(t))Yi(t)Yi(t) = 0 (1 ~ k ~ m)
{ Y(O) = P (8.3)
Y(O) = w.
This is not contained in [Apostol(1969), Thm. 7.19, p.229] mentioned above.
Instead, we refer the reader to [Hartman(2002), Thm. 4.1, p.100]. We will
need this result in order to obtain the smoothness of the 'exponential map',
to be studied later in this chapter. *
Theorem 8.1. Let M be a smooth manifold with a connection v'. For any
p E M and any v E TpM, there exists an open interval I c R containing 0,
and a geodesic 'Y: I--+ M such that "f(0) = p and v 7 ,p = v. Moreover, any
two such geodesics 'Y : I --+ M and ~ : J--+ M coincide on I n J.
Proof. Take any admissible chart (U, x) containing p. Set V = x(U),

P = x(p) and v = (v 1 , • • • , vm), where v = viOxi,p· By Proposition 8.3,
and taking Fi1 = rfj o x- 1 , there exists an interval I containing 0, and a
Y: I--+ V, having smooth components, satisfying (8.1). Define the smooth
curve 'Y: I--+ M by 'Y(t) = (x- 1 0 Y)(t) fort E J. Then
'Yi =xi o "(=Xi o (x- 1 0 Y)= (x o x- 1 0 Y)i = yi_
So the differential equation in (8.1) implies that 'Y satisfies the geodesic
equation. Moreover, we have that 'Y(0) = x- 1 (Y(0)) = x- 1 (P) = p, and
V7,p =,yi(0) Oxi,"f(O) = Yi(O) Oxi,"f(O) = ViOxi,p = V.
The uniqueness part is immediate from Proposition 8.3 as long as one is
in a single chart. The argument in the general case is easily adapted from
the uniqueness part of the proof of Proposition 8.3. First, it follows from
Proposition 8.3 that in a small interval (-1:, 1:) around 0, the curves 'Y, ~
do coincide. Set Tmax = sup{r EI n J: 'Y(t) = ~(t) for all t ~ r}. Suppose
that Tmax <sup(J n 1). By continuity, 'Y(Tmax) =~(Tmax) =:pmax• Moreover,
v'Y,Pmax = v7,Pmax =: VPmax (since their components can be seen to be the
same by considering the limit from below Tmax)- By the uniqueness in a
single chart (containing the point Pmax), it follows that a geodesic passing
through Pmax with velocity Vmax is unique, and so 'Y and ~ must coincide in a
neighbourhood of Tmax, contradicting the choice of Tmax• Hence 'Y(t) =~(t)
for all nonnegative t E I n J. A similar argument shows the agreement also
for nonpositive t EI n J. Consequently, 'Y(t) =~(t) for all t EI n J. □
Definition 8.3. (Maximal geodesic, geodesically complete manifold).

Let M be a smooth manifold with a connection v'. A geodesic 'Y: I--+ M,
where I c R is an open interval, is called maximal if it cannot be extended
to a geodesic on a larger open interval, that is, there does not exist a
geodesic ~ : J --+ M, where J c R is an open interval such that I s;; J
and ~11 = 'Y· We call M geodesically complete if every maximal geodesic is
defined on the entire real line.
Corollary 8.1. Let M be a smooth manifold with a connection v'. For any
p E M and any v E TpM, there exists a unique maximal geodesic 'Y : I --+ M,
where I is an open interval containing 0, such that 'Y(0) = p and v 7 ,p = v.
Geodesics 161
Proof. By Theorem 8.1, we know that there exists a geodesicµ: J--> M,

where J is an open interval containing 0, such that µ(O) = p and Vµ,p = v.
Let I be the union of all such intervals J. Then J is open and contains 0.
Moreover, since any two geodesics µ , µ defined on two such intervals J, J
coincide on their overlap J n J, we can unambiguously define 'Y : J --> M
by 'Y(t) := µ(t) , where µ : J--> M is any such geodesic for which t E J. We
claim that 'Y is itself a geodesic. If t E I , then there must exist a J which
contains t, and then 'YIJ = µ is a geodesic, and so it satisfies the geodesic
equation. The maximality follows from the definition of I. □
Example 8.2. (Cylindrical spacetime). Consider the cylindrical spacetime

in Example 5.8 (p.88), and the charts (U, </>), (V, 1P) given there. Suppose
I c IR. is an open interval, and 'Y : J --> U is a geodesic. The components
0, z are given by (0(t), z(t)) := </>('Y(t)) E (0, 21r) x R Then we have that
'Y(t) = (cos(0(t)), sin(0(t)) , z(t)) , t E J. From Example 6.6, we know that the
connection coefficients in the chart (U, cp) are all 0. So the geodesic equation
in U is given, component-wise, by 0 = 0 and z = 0. Hence 0(t) = o: t + 00
and z(t) = f3t + z0 for some constants o:, /3, 00 and z 0 . So the geodesic in U
is given by 'Y( t) = (cos( o: t + 00 ), sin( o:t + 00 ), f3t +zo), t E J. These describe:
1° o: = 0 and /3 = 0 : a point
2° o: = 0 and /3 =I= 0 : a vertical line parallel to the z-axis
3° o: =I= 0 and /3 = 0 : a circular arc
4° o: =I= 0 and /3 =I= 0 : a helix.
But now take I = IR., and 'Y(t) defined by the same expression as above.
Then 'Y( t) E C, and 'Y also satisfies the geodesic equation in V ( unless o: = 0,
when it does not enter Vat all, and 'Y(IR.) c U). So 'Y is a maximal geodesic
defined on R Similarly, we can show that every geodesic with image in V
can be extended to a maximal geodesic defined on R Consequently, C is
geodesically complete. ◊
Exercise 8.5. Let M be a smooth manifold with a connection v'. Let p EM, and
'Y : I-+ M, where I is an open interval in R containing 0, be a geodesic such that
-y(0) = p and v-y,p = 0. Prove that -y(t) = p for all t EI.
Hint: Consider ::Y(t) := p for all t E R, and show that it is a maximal geodesic
passing through p with velocity 0.
Exercise 8.6. Consider M = R 2 with the standard smooth structure, and a con-
nection v' described via its connection coefficients in the global chart (R 2 , id)
(see Exercise 6.3, p.110), given by r~Y = r~x = 1, and all other r-symbols iden-
tically 0. Determine the geodesics. Given any two points p, q in the plane, is
there a geodesic that starts at p and passes through q? Plot, using the computer,
geodesics starting at p = (0, 0) in the direction v = (cos 0)ox,p + (sin 0)oy,p, for
various values of 0, say 0 = k· ~~, for k = 0, • • • , 23.
8.3 Affine parameter
We had earlier remarked that a geodesic should not merely be thought of

geometrically as the set of points in the range of 'Y, since its parametrisation
does play a role in its being a geodesic. In other words, a reparametrisation
may render a geodesic a non-geodesic. What are the allowed parametrisa-
tions? We will show that the parametrisation is unique upto an affine linear
map, that is, affine linear maps are allowed and these are the only ones.
Lemma 8.1. Let M be a smooth manifold with a connection '1. Suppose

that I, J are open subsets of R, and J 3 u >-+ h(u) E I is a C 00 function.
Let 'Y : I -+ R be a smooth curve. Then for all s E J
("Vv.., h V-yoh)(s) = h(s) V7 (h(s))
0 + (h(s))2('1v.., V7 )(h(s)).
Proof. FromExercise2.4 (p.23), V-yoh(s)=h(s)V7 (h(s)). Also, it follows
from Exercise 7.1 (p.143) that
("Vv.., h V7 (h(-))) (s) = h(s) ("Vv.., V7 )(h(s) ).
0
Consequently, using the facts above, and the Leibniz rule for the covariant
derivative along the curve 'Y o h, we obtain, for all s E J
("Vv.., h V-yoh)(s) = ("Vv.., h (h V7 (h(-))))(s)
0 0
d(h) .
= -ud (s) V7 (h(s)) + h(s)("Vv V7 (h(-)))(s)
'Y
0h
= h(s) V7 (h(s)) + h(s)h(s) ("Vv.., V7 )(h(s))

= h(s) V7 (h(s)) + (h(s))2('1v.., V7 )(h(s)).
Geodesics 163
Exercise 8. 7. Let M be a smooth manifold with a connection v', and , : (a, b) -+ M

be a geodesic. Show that if v-y,y(c) =0 for some c E (a, b), then, is constant.
(In the contrapositive form, the above says that if , is a not the constant geodesic,
then its velocity vector field is nowhere zero.)
Hint: Let J = (a -c, b -c) and h(s) = s+c, s E J. Show that, oh is constant
using Exercise 8.5.
Let 'Y : I -+ M be a geodesic in a smooth manifold with a connection v',

where I is an open interval of R The 'variable' t E I, parametrising the
range of 'Y consisting of points 'Y (t) E M, is loosely called the 'affine param-
eter' of 'Y· The justification for the name 'affine' comes from the following
result, combined with Example 8.3.
Theorem 8.2. Let M be a smooth manifold with a connection v'. Let

'Y : I -+ M be a nonconstant geodesic, where I is an open interval in R Let
J be an open interval of "IR, and J 3 u >-+ h(u) EI be a C 00 function. Then
'Yo h : J -+ M is a geodesic if and only if h( s) = as + b for some constants
a,bER
Proof. We first note that for alls E J,
(v'v-roh V"foh)(s) = h(s) V'Y(h(s)) + (h(s))2(v'v-r V'Y)(h(s))
= h(s)V"'(h(s)) + (h(s))20 = h(s)V"'(h(s)),
since (v'v-r V'Y)(h(s)) = 0 by virtue of 'Y being a geodesic.
If 'Yoh is a geodesic, then O = (v'v-roh V"foh)(s) = h(s) V'Y(h(s)) for all
s E J. But as 'Y is nonconstant, V'Y(h(s)) -# 0 for alls E J (Exercise 8.7).
So h = 0 in J, and thus h( s) = as + b for some constants a, b.
Vice versa, if h( s) = as + b, then h = 0 in J, and so we have that
(v'v-roh V"fah)(s) =h(s)V'Y(h(s)) =0 for alls E J. Thus 'Yoh is a geodesic. □
Example 8.3. (Arclength parametrisation).
Let (M, g) be a Lorentzian manifold, and let 'Y : I -+ M be a geodesic,
where I c "JR is an open interval. Suppose that 'Y is timelike or spacelike
(but not lightlike). Let t 0 EI. Define the arclength parameter along 'Y by
r(t) = t
lto
lg('Y(s))(v'Y,'Y(s),V'Y,'Y(s))I ds, t EI.
If (M, g) has a time-orientation, and if 'Y is the worldline of an observer

(a future-pointing timelike smooth curve), then the arclength parameter
measures the proper time elapsed from the event at 'Y(t0 ).
As I 3 s >-+ g('Y(s))(v"!,"f(s), v"!,"f(s)) is constant along 'Y (Proposition 8.2),
say taking value c E "JR\{O}, we have r(t) = v1cf (t - t 0 ) fort EI. Thus if
h(r) = ar+b, where a= 1/-Jicf and b = -t0 , then 'Yoh is also a geodesic,
which we refer to as the geodesic 'Y 'parametrised by arclength'. Note that
V-yoh,yoh(r) = ~~ (r) V-y,-y(h(r)) = av'Y,'Y(h(r)), so that the tangent vector of"( 0 h
has everywhere a 'unit length':
lg(('Y O h)(r))(v-yoh,-yoh(r), V-yoh,-yoh(r))I
= a 2lg('Y(h(r)))(v7 ,-y(h(r)), V-y,-y(h(r)))I = a 2icl = 1.
For lightlike geodesics 'Y: J-> M, we have that r(t) = 0, that is, the 'proper
time' experienced by a lightlike particle (photon) between any two events
along its worldline is zero. Thus to label each point along the lightlike
geodesic, the arclength parameter is to no avail. On the other hand, by
Theorem 8.2, we know that any affine linear transformation of the interval
I gives an alternative parameterisation. ◊
Example 8.4. (Minkowski spacetime, and Newton's first law in inertial

frames). Recall the Minkowski spacetime (M, g) from Example 5.6 (p.85),
and let v' be the Levi-Civita connection induced by g on M. Consider
an inertial frame (M, x) induced by the choice of a point p E M and an
orthonormal basis {fo, f1, f2, f3} for V with respect to g (the Minkowski
scalar product on V), where we suppose that f0 is timelike. We will show
that the timelike geodesics are straight lines parametrised affine-linearly
in the 'time-coordinate' (of the inertial frame) along the geodesic. Also,
the motion of a force-free particle (that is, a geodesic) is recorded in the
inertial frame as 'uniform-in-time motion, along a straight-line-in-space',
which is Newton's first law of motion. Suppose 'Y : J -> M is a geodesic,
where I is an open interval in JR containing 0. Write 7(r) = p + 'Yi(r) fi.
Because the connection coefficients in the inertial frame are identically zero
(Example 6.5), the geodesic equation gives ii(r) =0, and so 'Yi(r) =ai +bir,
i = 0, 1, 2, 3. Then -(b0 ) 2 + (b1)2 + (b2)2+ (b3)2 < 0 since 'Y is timelike. In
particular, b0 #0. The inertial frame 'time-coordinate' of a point q := 7(r)
(r E J) on the geodesic is t(q) := 7°(r) = a0 +b0 r. The affine parameter of
the geodesic at 7(r) is then given in terms of t(q) by T = (t(q) - a0 )jb0 ,
which is affine-linear in the time-coordinate t(q) recorded by the inertial
frame for q. The 'spatial position' of q recorded in the inertial frame is
('Y 1(r),7 2(r),73(r)) = (a1,a 2,a3) + r(b1,b2 ,b3 )
= (al a2 a3) + (t(q) - ao) (bl b2 b3)
' ' bO , ' '
which is affine-linear in t(q) (i.e., in the inertial frame, 'the motion is along
a straight line in space uniformly in time'). ◊
Geodesics 165
Exercise 8.8. (Circular timelike geodesic in Schwarzschild spacetime and GPS).

The spacetime around Earth is well-described by Schwarzschild spacetime (M, g)
(Example 5.10, p.89), where m=Mai~0.45cm is the mass of Earth. Let (U,cp) be
the spherical coordinate chart, given in Example 1.8 (p.7). Consider a timelike
geodesic 'Y : / -+ M in the equatorial plane (0 = i) having a constant r-coordinate
r, and parameterised by proper time:
7(T)=(t(T),r,cp- 1 G,</>(T))), TEI.
With •' := ~r, show that
t" = 0,
</>" = 0,
(</>')2 = "; (t')2,
r
(1- 3;i)ct') 2
= 1.
Let To,bit be the circular orbit's period as measured by the observer 'Y, and TE.,,h
be the period as measured by an observer at rest on the surface of the Earth. The
radius of Earth is R,ai ~ 6400 km. We ignore the rotational motion of the Earth.
Show that
To,b;t ~ l _ 3m + ~-
TE..th 2r R,ai
Hint: Assume that the light signal from the events at the start and end of the
circular orbit reaches the observer on Earth radially, and revisit Example 5.16.
For Global Positioning System (GPS) satellites, r ~ 27000km, giving
To,b;t - TEuth ~ 4 _5 X l0-10,
TEarth
and this is taken into account in the design of the GPS.
8.4 Lightlike geodesics in Schwarzschild spacetime
We now give the calculation behind one of the first observational verifica-
tions of the spacetime viewpoint of gravity. In the Newtonian viewpoint,
the path JR 3 t >-+ ( t, x( t)) E JR x JR3 of light is a straight line, or 'light travels
in a straight line'. However, a light ray, from our spacetime viewpoint, is
just a lightlike/null geodesic. The region of spacetime in the vicinity of a
star is described reasonably well by the Schwarzschild spacetime1 . We will
now consider the path of a light ray, and show that its path is 'bent' by the
star. One can first photograph a certain celestial region, recording the light
rays emanating from distant fixed stars, measured at night, and when the
Sun is far away from the path of the light rays reaching the observer (so
that the bending effect on light due to the Sun's gravity can be neglected).
1 0r we imagine that the universe contains only the star and nothing else.
See the following picture.
------------ .
Observer
*
Dista nt
fixed star Sun
One photographs the same celestial zone during a solar eclipse2 and a dif-
ferent time of the year:
* ------------------ ~
◊~ · -------------
A comparison the two photographs should reveal a change in the apparent
position of the stars due to the bending of the light ray path by the Sun's
gravity. This angular deflection can be calculated by using the geodesic
equation, and the predicted value was actually measured during the 1919
solar eclipse.
Proposition 8.4. (Schwarzschild lightlike geodesics)

Let (M=JRx(2m,oo)xS 2 ,g) be the Schwarzschild spacetime with the Levi-
Civita connection. Let (U, 'P) be the chart for S 2 from Example 5.7 (p.87).
Let (U, <p) be the chart for M given by V = JR x ( 2m, oo) x U, and the chart
map V be a lightlike geodesic, where I is
an open interval in JR, and with coordinates defined by
(t(T), r(T), 0(T), q>(T)) = <p(,(T)) (TE/).
Then with · := ~r ,
.. 2m (
2m)-1 ,
t = - - 1--
r2 r
tr
r = - ; ( 1- 2~) i2 + ; ( 1- 2~) -I i-2 + r ( 1- 2~) iP + r ( 1- 2~) (sin 0) 2 Ji
., . 2 .
0 = (sin0)(cos0)¢>2 -i-0 -
r
.. 2 . . .
</> = --r<t>- 2(cot0)0¢>
r
together with the constraint
(1- 2~)i2-(1- 2~)- r 1 2
-r2 i/2-r 2 (sin0)2J? = 0.
2 The solar eclipse is relevant here only because during a solar eclipse, the stars are
visible, and can be photographed.
Geodesics 167
Proof. The first four equations follow immediately from the geodesic
equation expressed in the chart (U, cI>) by using the connection coef-
ficients listed in Example 6.8. The constraint equation is the conse-
quence of the causal character of 'Y being lightlike, that is, for all T E J,
g('Y(r))(v7 ,7 (r), v7 ,7 (r)) = 0, expressed in coordinates, and using the com-
ponent matrix for the metric given in Example 5.10. □
If a light ray satisfies, for a To E J, 0(ro) = f and 0(ro) = 0, then by looking

at the differential equation involving 0, and the uniqueness of solutions,
we conclude that 0 = J, that is, the geodesic then stays in the equatorial
plane. Thus for such a light ray, setting 0 = J and 0 = 0 in the equations
for ¢ and f, we obtain
.. 2 .
<p=--i-¢ (8.4)
r
..
r=- 2m) -
rm2 (( 1 - ~ t -
2 ( 2m)- 1r.
1--;=- 2)
+r ( 1--;=- 2m):i, 2
'I'. (8.5)
With 0 = J and 0 = 0, the constraint equation becomes

2m) t
(1--;=- ·2
- ( 2m)-1 r
1--;=-
•2
- r 2 ¢·2 = 0.
We use this in the equation (8.5) for r above, to obtain
r= (r - 3m)¢i. (8.6)
Suppose that T >-+ ¢(T) is strictly monotone, so that ¢ is invertible, with
the inverse map h = 4>- 1 . Define the function u by q> >-+ u( <p) := (roI)(cf>).
Using the chain rule and the inverse function theorem,
i-(h(-)) 1
(r(h(-))) ¢(·).
2
Also (with argument h(·) suppressed everywhere on the right-hand side),
( 2ri- + r 2 *)r - r 2 r
r4/p2
(using (8.4))
r r-3m
= - r2¢>2 = __r_2_ (using (8.6))
1 1
= -- +3m-.
r 2 r
Thus we obtain the differential equation

d2 u
d¢ 2 (¢) = -u(¢) + 3m(u(¢)) 2 . (8.7)
We now determine much the light ray bends due to the gravity of the Sun,
for which m = M0 = 1.5km (the mass of the Sun in units where c = GN = 1),
and we consider the situation described in the following picture.
Here, a light ray starts in the equatorial plane, from a fixed distant star.
So, we assume that the light ray comes in from r = +oo and leaves with
r = +oo eventually. We assume that the point of closest approach of the
light ray to the centre of the Sun is exactly equal to the radius R 0 of the
Sun, so that the ray just grazes its surface. We choose this point of closest
approach as (-R 0 , 0), lying in the equatorial xy-plane, as shown above. In
order to compute the deflection angle, we need to compute the angles with
the x-axis of the two asymptotes, that is, the azimuthal angle ¢ at r = +oo
when coming in, and the angle ¢ at r = +oo when going out. In both cases,
r = +oo, which corresponds to u = 0. Moreover, u(1r) = r{0 and ~¢ (1r) = 0
(r-minimisation/ u-maximisation condition).
Equation (8.7), namely u" +u = 3mu 2 (where -' = ~,t,), is a nonlinear
differential equation. Note that
1 1
u = - ,( -
r R0 '
where R 0 ~ 7 x 105 km, while 3m = 3M 0 ~ 4.5 km. In units where R 0 = 1,
we have E := 3m « 1. So the term 3mu 2 = w 2 is small in comparison with
u, and we can use a perturbative method to solve the differential equation
approximately. To this end, we first consider the differential equation
u~ + u 0 = 0, with u 0 (1r) = ~ 0 and u~(1r) = 0.
Geodesics 169
The general solution is u 0 ( ¢)=A cos¢+ B sin¢, and using the above 'initial'
conditions, the solution is the vertical straight line, parallel to the y-axis:
uo(c/>)=- cos¢_
R0
Now, for our original nonlinear equation,
U 11 +U = €.U 2 , (8.8)
we seek a solution u having the form u = u0 + f.u 1 + f. 2u 2 + • • • . Substituting
this in (8.8), we obtain by comparing terms in f. that
,, 2 (cos¢) 2 l+cos(2¢)
U1 + U1 = Uo = 2 = 2
R0 2R0
The initial conditions for u1 are u1 (1r) = 0 and Ui (1r) = 0. To find a solution,
we first try to find a particular solution by making the assumption that
u1,p=a+bcos(2¢), so that Ui,p=-2bsin(2¢), and ur,p=-4bcos(2¢). Sub-
stituting this in the differential equation, and comparing the coefficients of
1 and cos(2¢), we see that a=~ and b= -~- Thus u 1 ,p = 3 -~~t
2<1>l. We
add a homogeneous solution to u 1 ,p in order to satisfy the initial conditions
u1(1r) =0 and ui(1r) =0:
. 3-cos(2¢)
u1(¢) = A cos¢+ Bsm¢ + 2
6R0
Using u1(1r) = 0 and ui(1r) = 0, we get A=~ and B = 0. So
,1..) cos¢ 3-cos(2¢)
U1 ('I' = - - 2 + 2
3R0 6R0
Hence
,1..) cos¢ (cos¢ 3-cos(2¢)) 2
U ('I' = - - - + €. - -2 + 2 + € U2 + · · · .
R0 3R0 6R0
To find the approximate value of the incoming azimuthal angle ¢in of the
light ray, we set u( ¢in) = 0, and neglect the € 2 terms, giving the following
quadratic equation in cos ¢in:
3-(2(cos¢in) 2-l)) 0
-cos¢in
- - + € (cos¢in
--2- + 2 = .
R0 3R0 6R0
Defining c5in = ~ - ¢in> 0, we obtain
c5in ~ sinc5in = cos¢in = ½( - (3~ -1) +) (~-1)2+8)

0
= 4 ( 3~0 -1 (
+ ✓ 3~0 -1)2+ 8 )-1.
O,n ~ 4 (3Ro + 3Ro )-l = -2:.__

€ € 3R0
Similarly, the angular deviation 0 ut from the vertical is given by the same
0
expression, so that the total angular deviation

4 x 1.5 . ,,
o= o,n + o out ~ 34€R 0 = -R
4M 0
0
= -
7x 1 05 radians ~ 1. 77 .
This prediction was confirmed by observation during the 1919 solar eclipse.
8.5 The exponential map
Given a smooth manifold M with a connection 'v, and a point p E M, we

know that for each tangent vector v E TpM, there is a unique maximal
geodesic "Iv : Iv -> M, defined on some open interval I v c ~ containing 0,
that passes through pin the direction of v, i.e., 'Yv(0) = p and v,,v,P(0) = v.
The exponential map expP describes the dependence of 'Yv on v, and sends
straight lines Cv := {tv : t E ~} in TpM to geodesics passing through p
in M. We will see in the next section that in the context of a Lorentzian
manifold modelling spacetime, the exponential map will give rise to 'inertial
coordinates' at a spacetime point. We begin with the following definition.
Definition 8.4. (Exponential map).

Let M be a smooth manifold with a connection 'v, and let p E M.
Given a tangent vector v E TpM, denote by "Iv : Iv -> M, where Iv c ~
is an open interval containing 0, the unique maximal geodesic such that
"fv(0) = p and v,,v,P (0) = v. Let 'Dp := {v E TpM : 1 E Iv} c TpM. We
define the exponential map expP: 'Dp-> M at p, by expP v = 'Yv(l), v E Dp.
/~v\
~
Note that expP O = p. The following example gives some motivation for the
terminology.
Geodesics 171
Example 8.5. (Exponential map for the unit circle). In the Euclidean
plane IR. 2 , consider the unit circle 8 1 = {(x , y) E IR. 2 : x 2 + y 2 = 1}. Let
U denote the circular arc which is the image of the injective map given by
(-1r, 1r) 3 0 >-+ (cos0, sin0) E 81, and denote its inverse by ¢ . Similarly, let
V be the image of the injective map (-1r, 1r) 3 0 >-+ ( - cos 0, sin 0) E 8 1 ,
and denote its inverse by '1/J . Then it can be seen that A= {(U, ¢ ), (V, '1/J)}
forms an atlas for 8 1 . We consider 8 1 as a Riemannian manifold with
metric induced from the Euclidean inner product on IR. 2 , that is, for p E 8 1
and v, w E Tp81, g(p)(v , w) = ((vx , vy) , (wx , wy) ) 111.2 , where x , y denote the
coordinate maps given by 8 1 3 (a , b) >-+ x(a, b) := a, y(a, b) := b. In (U, ¢),
g(p)00 = g(p)(oe,p, 00,p) = (00,px) 2 +(00,py) 2 = (-sin0) 2 +(cos0) 2 = 1, for
p = (cos 0, sin 0) E U. Let v' be the corresponding Levi-Ci vita connection.
Using Proposition 6.3 (p.122), the connection coefficient in (U, ¢) is 0 = 0. ri
For v E B: = ( -1 , 1) c IR., consider the curve 'Yv : (-1r, 1r) --> 8 1 defined by
'Yv(t) = (cos(vt) , sin(vt)) for t E (-1r, 1r). Then 'Yv(O) = (1 , 0) =: p E U c 8 1 .
Moreover, we have 'Y!(t) =vt, so that i'!(t) =v, and v'Y v,P=voe,p- The curve
'Yv is a geodesic because
.. e re . e . e o o o
'Yv + 00 'Yv 'Yv = + = ·
Thus {voe ,p : v E B} c VP c Tp8 1 and the exponential map at p = (1, 0)
sends voe,p, VE B, to 'Yv(l) = (cosv , sin v) E 8 1 .
0 p
Identifying IR. 2 with C, and Tp8 1 with IR. (by sending IR. 3 v >-+ voe ,p E TpM),
we write expp(v)=(cosv,sinv)=cosv+isinv= eiv _ ◊
We had seen that an affine parametric change of variables of a curve pre-

serves its geodesic nature. The following result relates the initial speeds in
the case of dilations.
Lemma 8.2. (Scaling lemma).

Let M be a smooth manifold with a connection v'. Let p E M, v E TpM,
and c > 0. Let 'Y : J --+ M be a geodesic, where I is an open interval in JR
containing 0, and such that 7(0) = p and v'Y,P = v. Thenµ : ¾I--+ M given
by µ(t) = 7(ct), t E ¾I={~ : TE J}, is a geodesic such that µ(O) = p and
Vµ,,p=CV.
Proof. Let J = (a, b). Then we have ¾I=(%,~)- Set h(t) :=ct fort E ¾I.
Then µ = 'Y o h. As 'Y is a geodesic, it follows from Theorem 8.2 that so is
µ. Moreover, µ(0)=7(c0)=7(0)=p and vµ,,p=h(O)v'Y,P=cv. □
Corollary 8.2. Suppose that M is a smooth manifold with a connection
v'. Let p EM, v E TpM and c > 0. Let 'Yv: Iv--+ M be the unique maximal
geodesic such that 'Yv(O) = p and v'Yv,P = v. Let 'Ycv : Icv --+ M be the unique
maximal geodesic such that 'Ycv(O) = p and v'Ycv,P = CV. Then elev= Iv and
for all t E Icv, 'Ycv(t) ="fv(ct).
Proof. We use the Scaling lemma, applied twice with the roles of 'Yv
and 'Ycv swapped, and the maximality of the two geodesics. First, by the
Scaling lemma, 'Yv (c •) : ¾Iv --+ M is a geodesic passing through p at t = 0
with velocity at p equal to cv. The maximality of 'Ycv : Icv --+ M implies
1 Iv c Icv and 'Ycvl1. 1 = 'Yv(c ·). Also, starting from 'Ycv, it follows by the
C cV
Scaling lemma that 'Ycv(C 1 ·) : ,;hicv --+Mis a geodesic passing through p

at t = 0, with velocity at p equal to C 1 cv = v. As 'Yv: Iv --+ M is maximal,
clcvC Iv. So clcv=lv and 'Ycv="fv(c·) on Icv=¾Iv. D
In particular, if v E Vp, that is, if 1 E Iv, then for all c E (0, 1], we have
that 1 E Icv=¼Iv and expp(cv)="fcv(l)="fv(c). Thus the exponential map
takes the straight line segment {tv: t E [0, 1]} in TpM, and maps it to the
'arc' bv(t) : t E [0, 1]} of the geodesic 'Yv·
Lemma 8.3. Let M be a smooth manifold with a connection v', and p E M.
Then Vp contains a neighbourhood W of OE TpM and expP lw : W--+ M
is smooth.
We note that TpM ~ !Rm via the isomorphism i, v=viOxi,p ...!'.. (v1, • • • ,vm),
where (U, x) is an admissible chart containing p. Thus we may endow
TpM ~ !Rm with the usual Euclidean metric topology induced by the Eu-
clidean norm II· II- Recall from Example 2.3 (p.28) that the matrix relating
the components of a vector in TpM, with respect to induced bases from two
charts, is invertible. It follows that this topology on TpM is independent of
the admissible chart (U, x) containing p used to define the isomorphism i.
Geodesics 173
Proof. Consider an admissible chart (U, x) for M containing p. If 'Y is a

geodesic through pin the direction w=wioxi,p, then Y(t) :=x('Y(t)) satisfies
fk(t) + I'f/Y(t))Yi(t)Yi(t) = 0 (1 ~ k ~ m)
{ Y(0) = x(p)
Y(O) = w
where w = (w 1 , • • • , wm). By Remark 8.1, there exists a neighbourhood
N c !Rm of 0, an 1:>0, and a C 00 map Y: N x (-1:,1:)--+ x(U) such that
for all w E N, Y(t) := Y(w, t) solves the above initial value problem. But
then iv:= i - 1 N is a neighbourhood of 0 E TpM, and for each w E iv, we
have that 'Yw(t)=x- 1 (Y(iw,t)), t E (-1:,1:) c Iw, where 'Yw is the maximal
geodesic passing through p with V-yw,P = w. Now we apply Corollary 8.2.
Let c:=1:/2. For each w = cw E civ, where w E iv, we have
1 E (-2, 2) =c- 1 (-1:, 1:) C c- 1 Iw = c- 1 Ic-lcw = c- 1 Ic-lw = Iw.
Thus the open neighbourhood W := civ ofO E TpM is contained in 'Dp. Also,
by the smoothness of Y, civ = W 3 w >-+ expp(w) ="fw(l) =x- 1 (Y(i¼w, ½))
is smooth too. □
Theorem 8.3. Let M be a smooth manifold with a connection v', and let
p E M. Then 'Dp contains a neighbourhood V of O E TpM, and there is a
neighbourhood U of p E M such that expP Iv : V --+ U is a diffeomorphism.
Proof. By the previous Lemma 8.3, there exists a neighbourhood W of
0 E TpM such that W c 'Dp and expP Iw : W --+ M is smooth. We will show
that the derivative of this smooth map at 0 E TpM is invertible, and use the
inverse function theorem (Theorem 2.3, p.32). Recall that the derivative
map at 0 E TpM is a linear transformation d(expp)o : T 0 (TpM) --+ TpM.
As dimT0 (TpM) = dimTpM, it is enough to show surjectivity of d(expp) 0 .
Let v E TpM. Consider the straight line fv : JR--+ TpM given by fv(t) =tv.
For small enough t>0, tv E W, and expp(fv(t)) =expp(tv) ="ftv(l) ="fv(t).
(Here the notation 'Yw is used for the maximal geodesic in M such that
'Yw(0)=p and V-yw,p=w.) Then for any f E C 00 (M),
d(f o expp ofv)
((d(expp))o(vev,o))(f) = vev,oU o expp) = dt (0)
d(f o 'Yv)
= dt (0) = V-yv,Pf = vf.
Hence, (d(expp))o(vev,o) =v E TpM. As the chosen v E TpM at the outset
was arbitrary, we have shown that d(expp)o is surjective (and hence invert-
ible), as wanted. So by the inverse function theorem (Theorem 2.3), there
exists a neighbourhood V c W of 0 E TpM and a neighbourhood U of
p E M such that expP Iv : V --+ U is a diffeomorphism. □
8.6 Normal coordinates
Given a smooth manifold M with a connection 'v, the tangent space at

a point p E M can be viewed as a kind of 'linearisation' of the manifold
around p, since we have seen that radial straight rays from the origin in
TpM are mapped to geodesics emanating from the point p in M. In fact,
we will see in this section that given any basis {ei : 1 ,( i ,( m} for TpM,
exponential map sets up a coordinate system in a neighbourhood of p,
called the 'normal coordinate system', such that in this coordinate system,
all the connection coefficients vanish at p (assuming 'v is torsion-free). As
the connection coefficients may be thought of as the 'components' of the
connection in a chart (see (6.1)), the normal coordinates as those which
make (M, 'v) appear 'flat' (but only) at p.
Definition 8.5. (Normal chart/neighbourhood/coordinates).

Let M be an m-dimensional smooth manifold with a connection 'v. Let
p EM. Let Ube a neighbourhood of p EM , and let V be a neighbourhood
of O E TpM, such that expP : V ----> U is a diffeomorphism. Suppose that
B = {ei : 1 ,( i ,( m} c Vis a basis for TpM. Let the map i : rntm ----> TpM
be the isomorphism rntm 3 (v 1 ,··· ,vm) ~ viei E TpM. Let X: U----> l - 1 v
be given by x = i - 1 o exp; 1 . Then the admissible 3 chart (U, x) for M
is called the normal chart (U, x) with the normal neighbourhood U, and
normal coordinates xi.
expP l l exp;l
-
/ "\
e,,L ]Rm
CE) V --+
0 el
TpM l-1
\. .)
3 The admissibility claim follows from the fact that x is a diffeomorphism (being the
composition of diffeomorphisms).
Geodesics 175
Lemma 8.4. With the notation4 from Definition 8.5, we have:
(1) xi(expp(ei))=c5f, l~i,j~m.

(2) Oxi,p=ei, l~i~m.
(3) For each q EU, exp; 1 q=xi(q)Oxi,p-
Proof.
(1) This follows immediately from the definition, since
(xi oexpp)(ei) = (xoexpp)i(ei) = (i- 1 oexp; 1 oexpp)i(ei)
= (i-l)j(ei) = c5f_
(2) Let 'Ye; be the maximal geodesic passing through p such that v-y,; ,P = ei.
We had seen that fort E [0, 1], 'Ye;(t)=expp(tei)- So
- - - d(xi o 'Ye )
(dx3 )p(ei) = ei(x3 ) = v-y,;,P(x3 ) = dt ; (0)
= d(xi o expp(tei)) (0) = d((i- 1 )i o (tei)) (0)

dt dt
= :?c5f) (0) = c5f = (dxi)p(Oxi,p)-
As j was arbitrary, this proves the claim in (2).
(3) If ei := i - 1 ei E !Rm, then it follows from the definition of i that
e 1 , • • • , em are the standard basis vectors for !Rm. Thus for q E U,
x\q)Oxi,p = xi (q)ei = xi(q)i( ei) = i(xi(q)ei) = i(x 1 (q), · · · , xm(q))
= i(x(q)) = i((i- 1 o exp; 1 )(q)) = exp; 1 q.
Theorem 8.4. Let M be a smooth manifold with a torsion-free connection
v' (that is, v' satisfies (C4): v'xY - v'yX = [X, Y] for all X, YE TJM).
Let p EM. Then in a normal chart for p, rf/p) = 0 for all l ~ i,j, k ~m.
Proof. We will use the notation from Definition 8.5. Consider a vector
V = (v1,··· ,vm) E i- 1 v C !Rm. Set V = ViOxi,p = viei. Let 'Yv be the
maximal geodesic passing through pin the direction of v. Thus 'Yv(O) = p
and V-yv,P=v. We have fort E [0, 1] that
'Yt(t) = (xi o 'Yv)(t) = xi(expp(tv)) = (x o expp)\tv)
= (i- 1 )\tv) = (i- 1 /(tviej) = tvi.
-------------
4Note in particular that the basis B = {e1, • • • , em} for TpM is chosen from within the
domain V of the map expP : V--+ U, so that expp(e;) is well-defined. In general, we can
always scale the tangent vectors in the basis to achieve this.
'Yv satisfies the geodesic equation, and so 1t + rfj('Yv(t)ht(thi(t) = 0 for

all t E [0, 1]. As
.. k - d2(tvk) - 0
'Yv - dt2 - '
we obtain rfj('Yv(t)) vivi = 0. In particular, fort= 0, rfj(P) vivi = 0 for all
l~k~m. Fix an i, and take v=ei=i- 1 (ei), to obtain
r~i(P) =0.
Next, fix distinct i,j, and take v=ei+ej =i- 1(ei+ej) to obtain
r~/p) + r~j(p) + rji(p) + rjj(p) = o.
But rfi (p) = 0 = rji (p). Thus the torsion freeness of v' (implying rfi = rji,
see Exercise 6.11, p.119) now yields 2rfj(p)=0, i.e., rfj(p)=0. □
In the context of a spacetime (M, g), that is a Lorentzian manifold with

the Levi-Civita connection v', we know that v' is torsion free. Moreover,
we can start with the basis B = {e 1, • • • , em} for TpM being orthonormal
with respect to the scalar product g(p) for TpM.
Corollary 8.3. Let (M, g) be an m-dimensional Lorentzian manifold, with

the Levi-Civita connection v'. Let p EM. For a basis B = {e1, ···,em} for
TpM as in the notation from Definition 8.5, suppose also that the basis B
is orthonormal with respect to the scalar product g(p) for TpM:
-1 if i = j = 1,
g(p)(ei, ej) = T/ij := { 1 if i = j #- 1,
0 if i #- j.
Then in the normal chart (U,x) for p, we have gij(p) = T/ij, and
r~/p) = 0 for all 1 ~ i,j,k ~ m.
8. 7 Geodesics as extremal paths
In Euclidean space JR3 (i.e., JR 3 endowed with the Euclidean inner product),
we know that the straight line segment between two points p, q, is also the
shortest path from p to q. Likewise, on the sphere in JR 3 considered as
a Riemannian manifold, the 'short arcs' of great circles give the shortest
path between the endpoints of the arc. In this section, we will give an
interpretation of timelike geodesics as being stationary for the proper time
'functional'. More precisely, we will show the following.
Geodesics 177
Theorem 8.5. (Timelike proper time maximisers are geodesics).

Suppose (M, g) is a time-oriented Lorentzian manifold with Levi-Civita con-
nection v'. Let (U, x) be any admissible chart, p, q E U, and a, b E JR, a< b,
be such that the set
Opq:={µ: [a,b]--+ U Iµ is an observer such that µ(a)=p, µ(b)=q}
is not empty. Define T : Oab --+ JR by
r(µ) = f .J-g(µ(t))(vµ,µ(t),Vµ,µ(t))dt, µ E Oab·

If the curveµ* E Oab maximises T, then there exists a diffeomorphism
t: (0, r(µ*))--+ (a, b), such thatµ* o t: (0, r(µ*))--+ U is a geodesic.
In order to prove this result, we will prove the necessity of the Euler-
Lagrange equations for maximising curves.
Proposition 8.5. (Necessity of Euler-Lagrange equations).
Suppose that a, b E JR, a< b, U c !Rm is open, and Sl c U x !Rm is open.
Let£ : {l --+ JR, U X ]Rm ::J {l 3 (o., /3) >-+ .C(o., /3) E JR, be a C 00 function.
Let p, q EU, and define the set
Opq := {µ: [a,b]--+ U I µ(a)=p, µ(b)=q}.
Define f : Opq --+ JR by
f(µ) = s: .C(µ(t), µ(t))dt, µ E Opq·

Ifµ* maximises f, then µ* satisfies the Euler-Lagrange equations:
:i(µ(t),µ(t))-1t(:i(µ(t),µ(t)) =0, tE[a,b], l~i~m.
(Here we use the notation µ= !~ and oi for the components of o. etc.)

Proof. Leth: [a, b] --+Ube any C 00 map, such that h(a) = 0 = h(b).
Then for any 1: E JR, (µ* + 1:h)(a) = p and (µ* + 1:h)(b) = q. Moreover, as
h([a,b]) is a compact subset of the open set U, it follows that for all 11:I
small enough, µ* + 1:h assumes values in U, and so µ* + 1:h E Opq. As µ*
is a maximiser for f, it follows that g, given by g(1:) := f(µ* + 1:h), has a
maximum at 1:=0, and so by elementary calculus, g'(0)=0. Thus
0 = g'(0) = :(µ*+1:b)'•=O = :€ f .C(µ*(t)+1:h(t),µ,*(t)+1:h(t))dtj•=O

= f :€ .C(µ*(t) +1:h(t), Ji.At) +1:h(t))'•=O dt
= f (:i(µ*(t),Ji,At))hi(t) + :i(µ*(t),µ*(t))b.i(t))dt. (8.9)

Integration by parts on the second summand yields
i b
a
o.C ..
o/3/µ*(t), µAt)) h'(t)dt
= - ib
a
d ( o.C ) . o.C . ,t=b
dt of3i(µAt),µ,*(t)) h'(t)dt+ of3i(µ*(t),µ,*(t))h'(t) t=a
= - f 1t(!~(µAt),µ,*(t)))hi(t)dt+O-O,
where we used the boundary conditions h(a) = 0 = h(b). Putting this ex-
pression into (8.9), we obtain that for all smooth h: [a, b] ---+ Rm satisfying
h(a) =O=h(b),
f (:i (µAt),µ,*(t)) - :t (:~ (µAt),µ,*(t)))) h\t)dt = o.

Now consider h as a column vector in which all entries are equal to the zero
function, except for the ith one, which is any smooth function h : [a, b] ---+ R
such that h(a)=O=h(b). Thus we have, for all such h, that
f (:i (µ*(t),µ,*(t)) - ! (:~ (µ*(t), µ*(t)))) h(t)dt = 0.
We will show that this implies the ith Euler-Lagrange equation. Suppose
on the contrary that there is some t 0 E [a, b] such that rp(t 0 ) # 0, where
rp(t) := :i (µ*(t), f.,,At)) - ! (!~ (µAt), µ,*(t))), t E [a, b].
Without loss of generality, suppose that rp(t 0 ) > 0. Also, by the continuity
of rp, we may choose t 0 in the open interval (a, b). Let 8 > 0 be such that
rp(t) > 'P~o) fort E (t 0 -8,t 0 +8). Take any nonnegative bump function h
such that h = 1 on (to - ! , to+!) and h = 0 outside (to - 8, to+ 8). Then we
get the contradiction that
0= ia
b
rp(t)h(t)dt =
rto+.S
Jto-.s rp(t)h(t)dt >
rto+! (t ) (t )
Jo-% rp / ldt > rp / 8 > 0.
Proof. (Of Theorem 8.5).

Let U:=x(U), and 0:= {(a, ,13) EU x Rm: gij(x- 1 (a))/3i/3j <0}.
Define 9ii = gij o x- 1 . Then
o(gij O x- 1 ) 09ij
Oxk,pgij = oak (x(p)) = oak (x(p)). (8.10)
Geodesics 179
Define .C(o.,/3)=,v'-gij(o.)j3ij3i for (o.,/3) E {l c Ux(Rm\{0}). Then .C is

smooth since x ...... y'x is 0 00 on (0, oo ). Define µ(t) = x(µ(t)) fort E [a, b].
We have vµ,µ(t) = µi(t)Oxi,µ(t), where µi := (x o µ)i = µi. It follows that
g(µ(t) )( vµ,µ(t), vµ,µ(t)) = 9ij(µ(t))µ,i(t)µ) (t), and
r(µ)= f .C(µ(t),µ,(t))dt =: J(µ),
where f is defined on Opq with p:=x(p) and q:=x(q). We have

O£, _ Ogkl /3k 13£ ✓
~ = 0°'' , where D(o.,/3) := -gij(o.)j3ij3i, and
ua• 2D
o.C 1 k £ k £ l k
i}j3i = - 2D (gkl 8i f3 + 9k£f3 8i) = - D9kif3 ·
Thus the Euler-Lagrange equation is

_ _!Ogkl~~ d g -~
2 oa' dt dt ~ _ 0
_
D +dt D - . (8.11)
Defines: [a, b]---+ R by
s(t) = i a
t di di
-gij(µ.(t)) ):• (t) dt (t) dt, tE[a,b].
Then s(a) =0, s(b) =r(µ*), and sinceµ* is timelike,
!: (t) = D(µAt), µ,*(t)) > 0 (t E [a, b]).

Thus the smooth function s is strictly increasing, and s([a, b]) = [O, r(µ*)].
In particular, sis a bijection, and has an inverse map t : [O, r(µ*)] ---+ [a, b].
Also, since the derivative of s is everywhere positive on (a, b), it follows
that the maps: (a, b)---+ (0, r(µ*)) is a diffeomorphism. Thus its inverse t
is smooth. So by the chain rule and the inverse function theorem, we have
that for any smooth h : (a, b) ---+ R,
d(h o t) 1 dh l dh
ds (s) = ~!(t(s)) dt (t(s)) = D(µAt(s)),µ,*(t(s))) dt (t(s)).
In particular, defining "Y* : (0, r(µ*))---+ U by "Y*(s) = µ*(t(s)), we have

d"'f* ( ) 1 dµ* ( ( ))
ds 8 = D(µ*(t(s)),µ,*(t(s))) dt t 8 ·
Thus (8.11) becomes
_ ~ 09k£ d"'f! d"'f! + ~ (9ki d"'f!) = 0.

2 oa• ds ds ds ds
Expanding the second summand, we obtain
d ( ( ( )) d'"'f: ( )) ogki ( ( )) d'"'f; ( ) d'"'f: ( ) ( ( )) d2'"'f: ( )

ds 9ki % s ds s = oo/- '"'I* s ds s ds s + 9ki '"'I* s ds2 s .
Thus (to get the second equality, we double the coloured middle summand
by swapping the k, £ dummy indices)
O = 9ki d2'"'f: + 09ki d'"'f! d'"'f:. _ ~ 09kf d'"'f: d'"'f;

ds 2 oaf ds ds 2 oa' ds ds
= 9ki d2'"'f: + ~ ( 09ki d'"'f; d'"'f: + 09fi d'"'f: d'"'f;) _ ~ 09kf d'"'f: d'"'f;
ds 2 2 oaf ds ds oak ds ds 2 oa' ds ds
= 9k d2'"'f: + ~ ( 09ki + ogn _ 09kf) d'"'f: d'"'f;
t ds 2 2 oaf oak oai ds ds .
Set giJ := giJ o x- 1 . Then multiplying by the evaluation of giJ on %(s),
and summing over i yields, in light of (8.10) and Proposition 6.3 (p.122),
2 k i . k f
O= lj 9ki d '"'I* + [!_:_ (09ki + 09fi _ 09kf) d'"'f* d%
ds 2 2 oaf oak ca' ds ds
2 . k f
_ d '"Yi
- ds2
rJ ( -1( ( )))d'"'f*
+ kf x '"Y* 8 ds
d'"'f*
ds ·
Consequently,'* :=x- 1 ('"'!*): (0, T(µ*))--> U satisfies the geodesic equation
in (U, x), and hence it is a geodesic. We note that
This completes the proof of Theorem 8.5. □

Essentially the same proof can be used to show that length minimisers in
Riemannian manifolds are geodesics.
Chapter 9
Curvature
The extent to which the commutator of covariant derivatives differs from

the covariant derivative along the commutator of the vector field is used
as a measure of the 'curvedness' of a manifold with a connection. In this
chapter, we will learn about a (1, 3)-tensor field R, called the curvature
tensor field, which is this object: for vector fields X, Y, Z E TJ M and a
I-form field OE Tf M,
R(X, Y)Z := [v'x, v'y]Z - v'[x,YJZ
:= v'xv'yZ - v'yv'xZ - v'[x,Y]Z
R(O, X, Y, Z) = O(R(X, Y)Z).
We had met the curvature tensor field R in Exercise 6.15 (p.125), where
we had seen that for a locally flat semi-lliemannian manifold (M, g), the
curvature tensor field R is zero. For example, R is zero for the Minkowski
space 1 and the cylindrical spacetime2 , as these are locally flat. But R-# 0
for S 2 considered as a Riemannian manifold using the metric induced on
its tangent spaces from the Euclidean inner product on JR 3 (Exercise 6.15).
In the appendix to this chapter we will also learn that if R = 0, then the
semi-Riemannian manifold is locally flat. This provides the motivation
for considering R to be a quantification of curvature. We begin with the
following definition.
Definition 9.1. (Curvature tensor field).

Let M be a smooth manifold with a connection v'. Then we define the
curvature operator
R(X, Y)Z = v'xv'yZ - v'yv'xZ - v'[x,Y]Z for X, Y, Z E TJM.

1 Example 5.6, p.85
2 Example 5.8, p.88
181
The (1, 3)-tensor field R, called the curvature tensor field, is defined by
R(O, X, Y, Z) = O(R(X, Y)Z)
= O(v'xv'yZ - v'yv'xZ - v'[x,YJZ)
for X, Y, Z E TJ M and n E Tf M. When (M, g) is a semi-Riemannian man-
ifold with the Levi-Civita connection v', we call R the Riemann curvature
tensor field.
We had seen in Exercise 6.15 that R is indeed a (1, 3)-tensor field.

Exercise 9.1. (Commutator of Lie derivatives).
Let M be a smooth manifold. By Definition 3.15 (p.58), the Lie derivative of
YE TJ M with respect to XE TJ M is .Cx Y = [X, Y] E TJ M. Show that for all
X, Y,Z E TJM, .Cx.CyZ -.Cy.CxZ - .C[x,YJZ=O. Hint: Jacobi identity.
We can express the chart-induced components of R in terms of the connec-

tion coefficients.
Proposition 9.1. Let M be a smooth manifold with a connection v', and

let (U, x) be an admissible chart. Then
Rfjk = axsfj - Oxir{i + rkjr;i - rkir;j.
Proof. We have
Rfjk = dxe(v'a.,i v'a.,jOxk -v'a.,i v'a.,;Oxk -v'[a.,;,O.,j]Oxk)
= dxe(v'a.,;(rkjOxr) - v'a.,i (rkiOxr) - v'oOxk)
= dxe((Oxirkj)Oxr + rkjv'o.,;Oxr - (Oxirki)Oxr - rkiv'a.,jOxr - 0)
= (Oxirkj)8;+ dxe(rkjr:iOx•) - (Oxirki)8; - dxe(rkir:jOx•)
= Oxir{j + rkjr;i - Oxir{i - rkir;j.
Exercise 9.2. Let M be a smooth manifold with a connection V, and let (U, x) be
an admissible chart. Show that R( Oxi' Oxi )oxk = RLkoxl.
Hint: Apply dxr on both sides.
For a 4-dimensional manifold, there are 44 = 256 components of R. Fortu-

nately, one does not have to do as many computations because of presence
of various symmetries. We begin with the following.
Lemma 9.1. Let M be a smooth manifold with a connection v'. Then for
all vector fields X, Y, Z E TJ M, we have R(X, Y)Z = -R(Y, X)Z.
Curvature 183
In terms of components of R in an admissible chart (U, x), Rfjk = - R]ik.

Proof. We have
R(X, Y)Z = v'xv'yZ - v'yv'xZ - v'[x,Y]Z

= -(-v'xv'yZ + v'yv'xZ + v' -[Y,x]Z)
= -(v'yv'xZ -v'xv'yZ -v'[Y,x]Z)
= - R(Y, X)Z. □
Proposition 9.2. (First Bianchi3 identity).

Let (M, g) be a semi-Riemannian manifold with the Levi-Civita connection.
Then for all X, Y, Z E TJ M, R(X, Y)Z + R(Y, Z)X + R(Z, X)Y = 0.
In terms of components of R in an admissible chart (U, x),
Rfjk + R]ki + Rfij = 0.

Proof. We will use the torsion-freeness of the Levi-Civita connection (for
all V, WE TJM, v'vW-v'wV = [V, W]), and the Jacobi identity. We have
R(X, Y)Z + R(Y, Z)X + R(Z, X)Y
~~~~~ -Y~.?~~ - v'[x,YJZ
+y~~."'.~ -v'zv'yX -v'[Y,z]X
+v'zv'xY- ~~~~~ -v'[z,x]Y
c~> :7!J~~J- v'[Y,z]X + Y~J~:~)- v'[z,x]Y + v'z[X, Y] - v'[x,YJZ
(~) [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y]] (Ja~bi) 0. D
In a semi-Riemannian manifold (M, g), we can use the musical isomorphism

to construct, from the Riemann curvature tensor field, a totally covariant
(0, 4)-tensor field R.
Definition 9.2. (Covariant curvature tensor field).

Let (M, g) be a semi-Riemannian manifold. The covariant curvature tensor
field RE T2 M is defined by
~
R(X, Y, Z, W) =
p
R(W , X, Y, Z) = g(W, R(X, Y)Z)
for all W,X, Y,Z E TJM.
3 After Luigi Bianchi (1856-1928), an Italian mathematician. We will meet the second
Bianchi identity in Proposition 9.4, which will involve the covariant derivatives of R.
In a chart (U, x), the covariant curvature tensor field has components
Rijk£ = g£rRijk, as
Rijk£ = R( Oxi, Oxi, Oxk, Ox£) = g( Ox£, R( Oxi, Oxi )Oxk)

= g(Oxi, RijkOxr) = Rijkg(Oxi, Oxr) = Rijkglr = glrRijk·
Exercise 9.3. (First Bianchi identity for the covariant curvature tensor field).
Let (M, g) be a semi-Riemannian manifold with the Levi-Civita connection V.
Show that R.(X, Y, Z, W) + R.(Y, Z, X, W) + R.(Z, X, Y, W) = 0.

connection V. Show that for all n E Tf M and all X, Y, Z E TJ M,
R(n, X, Y, Z) = R.(X, Y, z, n~).
Hence or otherwise, conclude that in a chart (U, x), RLk = R.ijkrgir.
Exercise 9.5. Let (M,g) be a semi-Riemannian manifold. Prove R=Ci(R®g).
We have the following symmetries for R.

Proposition 9.3. Let (M, g) be a semi-Riemannian manifold with the
Levi-Civita connection v'. Then:
For all X, Y, Z, WE TJM, In any admissible chart (U, x),

R(X, Y, z, W) = - R(Y, X, z, W) ~ ~
Rijk£ = - Rjik£
R(X, Y, z, W) = - R(X, Y, w, Z) ~ ~
Rijk£ = - Rij£k
R(X, Y, z, W) = R(Z, w, X, Y)
~ ~
Rijk£ = Rktij
In the right column above, the i, j, k, I!. are arbitrary in {1, • • • , m}.
Proof. The first claim follows from Lemma 9.1, since

R(X, Y, Z, W) = g(W, R(X, Y)Z) = g(W, -R(Y, X)Z) = - R(Y, X, Z, W).
For the second claim, we use the metric-compatibility of the Levi-Civita
connection. First,
g(W, v'xv'yZ)
= - g(v' x W, v'y Z) + X g(W, v'y Z)
= g(v'yv'xW, Z) - Yg(v'xW, Z) + XYg(W,Z) - Xg(v'yW, Z). (9.1)
Swapping X and Y, we also have
g(W, v'yv'xZ)
= g(v' x v'y W, Z) - X g(v'y W, Z) + Y X g(W, Z) - Y g(v' x W, Z).
Curvature 185
Moreover, g(W, v'[x,YJZ) = - g(v'[x,YJ W, Z) + [X, Y]g(W, Z). Subtract-

ing the last two equations from (9.1), we obtain
g(W,R(X, Y)Z) = g(v'yv'xW, Z) - g(v'xv'yW, Z) + g(v'[x,YJ W, Z)
= -g(R(X, Y)W,Z),
and so R(X, Y, Z, W) = -R(X, Y, W, Z), proving the second claim.
Finally, using the anti-symmetry in the last two components, and the
first Bianchi identity, we have
R(X, Y, z, W) = - R(X, Y, w, Z) = R(Y, w, X, Z) + R(W, X, Y, Z).
Also, R(X, Y, Z, W) = -R(Y, X, Z, W) = R(X, Z, Y, W) + R(Z, Y, X, W).
Adding these yields
2R(X, Y, z, W)
= R(Y, w, X, Z) + R(W, X, Y, Z) + R(X, z, Y, W) + R(Z, Y, X, W). (9.2)
By replacing (X, Y, Z, W) by (Z, W, X, Y), we obtain also
2R(Z, w, X, Y)
= R(W, Y, z, X) + R(Y, z, w, X) + R(Z, X, w, Y) + R(X, w, z, Y). (9.3)
Since a simultaneous swap of the first two and of the last two arguments of
R does not change its value, a comparison of the right-hand sides of (9.2)
and (9.3) reveals that the left-hand sides are also equal. □

connection V. Show that in any admissible chart (U, x), we have R~ii = R~ji·
Exercise 9.7. Recall Example 5.7 (p.87), where we considered the sphere S2 as a
Riemannian manifold using the metric g induced on the tangent spaces from the
Euclidean inner product on R 3 • In the chart (U, c.p) described there, determine
the component R:o<1> using Proposition 9.1. Also determine R.4>o4>o, and hence
determine all the 16 components of R using the symmetries of R. Find all the
components of R as well.
We now show the second Bianchi identity, which involves derivatives, and
bears some resemblance to the first Bianchi identity seen earlier. First we
note that as R is a (1, 3)-tensor field, given any V E TJ M, its covariant
derivative v'v R is a (1, 3)-tensor field. For n E Tf M and X, Y, Z E TJ M,
(v'vR)(O,X,Y,Z)
= V(R(O,X, Y,Z)) - R(v'vO,X, Y,Z)
-R(O, v'vX, Y,Z) - R(O,X, v'vY, Z) - R(O,X, Y, v'vZ)
= V(O(R(X, Y)Z)) - (v'vO)(R(X, Y)Z)
-O(R(v'vX, Y)Z) - O(R(X, v'vY)Z) - O(R(X, Y)v'vZ)
= O(v'v(R(X, Y)Z)-R(v'vX, Y)Z-R(X, v'vY)Z-R(X, Y)v'vZ).
So defining the map (v'vR)(X, Y): TJ-M-+ TJ-M by

((v'vR)(X, Y))Z
:= v'v(R(X, Y)Z)-R(v'vX, Y)Z-R(X, v'vY)Z-R(X, Y)v'vZ, (9.4)
we see that
(v'v R)(n, X, Y, Z) = O((v'v R)(X, Y)Z).
We are now ready to state the result on the second Bianchi identity.
Proposition 9.4. (Second Bianchi identity).
Let (M, g) be a semi-Riemannian manifold with the Levi-Civita connection
v'. Then with the notation in (9.4), we have for all X, Y, Z E TJ-M that
(v' x R)(Y, Z) + (v'y R)(Z, X) + (v' zR)(X, Y) = 0. (9.5)
In terms of the components of the Riemann curvature tensor field in an
admissible chart (U, x),
(v'a.,h R)ttj + (v'a.,k R)thi + (v'a.,l R)iki = 0.
By abuse of notation, the last equation is sometimes written as
Rtlj;h + Rthj;k + Rikj;l = 0
in the literature.
Proof. The left-hand side of (9.5), henceforth denoted by LHS, is the
(pointwise defined) sum of maps from TJ- M to TJ- M, and we want to show
that this is the zero map, i.e., its action on each V E TJ- M is the zero vector
field OE TJ- M. By using the definition in (9.4),
• ((v' zR)(X, Y))V
= v'z(R(X, Y)V)-R(v'zX, Y)V-R(X, v'zY)V-R(X, Y)v'zV,
• ((v' xR)(Y, Z))V

= v' x(R(Y, Z)V)-R(v' x Y, Z)V -R(Y, v' x Z)V -R(Y, Z)v' x V,
• ((v'y R)(Z, X))V

= v'y(R(Z,X)V)-R(v'yZ, X)V-R(Z, v'yX)V-R(Z,X)v'yV.
We will now add these equations. But before doing so, we note that using
the skew-symmetry of R (Lemma 9.1) with the torsion-freeness of the Levi-
Civita connection, we can combine the respective underlined terms:
-R(v'zX, Y)V - R(Y, v'xZ)V = R(Y, v'zX - v'xZ)V = R(Y, [Z,X])V
-R(X, v'zY)V - R(v'yZ,X)V = R(X, v'yZ -v'zY)V = R(X, [Y,Z])V
-R(v'xY, Z)V - R(Z, v'yX)V = R(Z, v'xY - v'yX)V = R(Z, [X, Y])V.
Curvature 187
Then we obtain the following:
LHS = '\l z(R (X , Y )V) + R (Y, [Z,X] )V- R( X , Y )'\l zV

-----------
+ V x(R(Y, Z)V) + R(X, [Y, Z])V - R(Y, Z)V x V
+ '\ly(R(Z, X)V) + R(Z, [X , Y])V - R(Z,X) '\JyV
= '\l z '\l x '\J yV - '\l z '\l y '\JxV - '\l z '\l[x ,YJV
+ V y '\/[z,x JV- V[z,xJ V YV- V[Y,[z,x]] V
- V x '\J y '\J z V + '\ly '\J x '\J z V + V [x,YJV zV
+Vx'\Jy'\lzV-Vx'\lz'\lyV-VxV[Y,zJV
+VxV[Y,zJV-V[Y,zJVxV-V[x,[Y,z]]V
-'\Jy'\lz'\lxV + '\lz'\ly'\JxV + V[Y,zJVxV
+ '\Jy'\l z '\lxV - '\ly'\Jx'\l zV- '\Jy'\l[z,xJV
+ V z '\l[x,YJV - V[ x,YJ'\l zV- V[z,[x ,YJ]V
- '\l z '\lx '\J y V + Vx'\l z '\lyV + V[z,xJ '\J y V .
The terms on the right-hand side all cancel except for the following:
LHS = - V[Y,[z,xlJ V -V[x,[Y,zlJV - V[z,[x,YJJv

- '\J[Y,[Z,X]]+[X,[Y,Z]]+[Z,[X,Y]] V
(Ja~bi) -VoV = 0,
proving the second Bianchi identity.

To obtain the identity for components in a chart, we first note that
the second Bianchi identity above implies that for all fl E Tf M and all
X, Y,Z, VE TJM,
0 = fl(((VxR)(Y,Z))V + (('\JyR)(Z,X))V + ((VzR)(X, Y))V)

= (V x R)(fl, Y, Z, V) + (Vy R)(fl, Z, X, V) + (V zR)(fl, X, Y, V).
Now taking fl=dxi E TfU, and X =Oxh, Y =Oxk, Z =oxe, V =Oxi in TJU,
we obtain
0 = (VaxhR)(dx\oxk,Oxe,Oxi)
+ (VaxkR)(dxi, Oxe, Oxh, Oxi)
+ (V oxeR)(dxi' Oxh, Oxk' Oxi)
= (VaxhR)i,j + (VaxkR)}hj + (VaxeR)hkj·
9.1 Ricci and scalar curvatures
We have seen that the curvature tensor field Risa complicated (1, 3)-tensor
field, but by contracting it, simpler tensor fields are obtained.
Definition 9.3. (Ricci curvature tensor field).

v'. The Ricci curvature tensor field, Ric E Tf M is the contraction Cf R.
More explicitly, for Y, Z E TJ M,
Ric(Y, Z)=(C~R)(Y, Z)=C(Tf MxTJM 3 (O,X),..... R(O,X, Y,Z)),

and so in an admissible chart (U, x), Ric(Y, Z) = R(dxk, Oxk, Y, Z), giving
Ricij = R(dx\Oxk,Oxi,Oxi) = R~ij·

We now show that Ric is symmetric.

Levi-Civita connection v'. For all Y, Z E TJM, Ric(Y, Z) =Ric(Z, Y).
Proof. It is enough to show that in any admissible chart, Ricij = Ricji,

which follows from the symmetry properties of R (Exercise 9.6):
k
Rkij = g
~~ ~~ ~~
Rkijl = g Rjtki = (-1)(-l)g Rtjik
= glkftljik = R~ji = R~ji· □

connection V. Show that Ric= - C~R. Hint: Use chart components and the
symmetries of :ii.
We note that Ric E Tf M can be considered as a (1, 1)-tensor field Ric~ by

'pulling up a covariant index' using the musical map, say the first one:
We can further take the 'trace' of Ric~, i.e., operate by C, and obtain a
C 00 ( M)-function.
Definition 9.4. (Scalar curvature function).

Let (M, g) be a semi-Riemannian manifold with the Levi-Civita connec-
tion v'. The scalar curvature (function), S E C 00 (M) is the contraction
C(Ric~) = C(Ric(J, ·)).
Curvature 189
Thus if (U, x) is any admissible chart, then

S = Ric((dxi)",oxi) = Ric(giioxi,Oxi) = giiRic(oxi,Oxi)
.. .. .. k
= g'3 Ricji = g'3 Ricij = g'3 Rkij•
Exercise 9.9. In continuation to Exercise 9.7, calculate the components of the
Ricci curvature tensor field in the chart (U, c.p) for the Riemannian manifold
(8 2 , g). Also determine the scalar curvature function on 8 2 •
Exercise 9.10. (Kretschmann scalar) 4 •

Let (M, g) be a semi-Riemannian manifold. Define the (4, 0)-tensor field R by
R=i'i(J, J, J, J).
The Kretschmann invariant/scalar is defined by K = C(C~(C~(C½(R (8) R.)))).
Thus, in an admissible chart (U,x), K=R.ijkl~jkl·
In continuation to Exercises 9.7 and 9.9, show that the Kretschmann scalar
K=4 for the Riemannian manifold (8 2 , g).
In Exercise 6.10 (p.119), we had defined the divergence of a vector field

on a smooth manifold equipped with a connection. In a semi-Riemannian
manifold (M, g), we can use the musical map-" to define the divergence of
(0, s)-tensor fields. We give the following definition.
Definition 9.5. (Divergence of a (0, s)-tensor field).
v'. If TE T2M is a (0,s)-tensor field, wheres~ 1, then its divergence
div T E T2_ 1 M is the (0, s -1 )-tensor field defined by
(divT)(Vi, · · ·, Vs-1)=C(Tf MxTJM 3 (0, V) I-+ (v'vT)(n", Vi,···, Vs-1))
for Vi,··· Vs-1 E TJM.
We note that for fixed Vi, · · · , Vs-1 E TJ M, the map 'I' given by
Tf MxTJM 3 (n, V) L (v'vT)(n", Vi,···, Vs_i)
is a (1, 1)-tensor field (the linearity in V follows from property (C3) of the
connection, and the linearity inn is a consequence of the C 00 (M)-linearity
of -"~and of v'vT in its first argument). So the contraction map C can act
on T.
4 After Erich Kretschmann (1887-1973), a German physicist. In Chapter 14 on the
field equation, Exercise 14.1 shows that 'vacuum spacetimes' (spacetimes for which the
energy-momentum tensor field T = 0 and the cosmological constant is 0) are 'Ricci-flat'
(Ric= 0). So the scalar curvature S = 0 for vacuum spacetimes. But the Kretschmann
scalar can then be useful since it may be still nonzero. For example, the Schwarzschild
spacetime is Ricci-flat, but the Kretchmann scalar is nonzero; see (15.4).

connection V. Suppose that T, SET~ Mare (0, s)-tensor fields, s",;::, 1, and c E IR.
Show that div(T+cS) = (divT) + cdivS.

connection V. Let VE TJ M. Prove that divV = div(V 0) E C 00 (M).
Example 9.1. (divg = 0). By Theorem 6.2 (p.128), v'.g = 0. So in any

admissible chart (U,x), we have (divg)i = (v'axJ.g)((dxi)U,ox;) = 0. ◊

Levi-Civita connection v'. Then div Ric= ½dS.
Proof. Let (U, x) be any admissible chart. Then by the second Bianchi
identity, we have (after replacing h by i, and summing over i)
(9.6)
As contraction commutes with the covariant derivative, we obtain using the
symmetry property of R (Lemma 9.1) that
C~(v' Oxk R) = v' Oxk (C~R) = -v'axk (CtR) = -ct (v' Oxk R),
and so
(v'axkRic)tj = (v'axk(CtR))tj = (CUv'axkR))tj
1 i
= -(C2(v'axkR))tj = -(v'axkR)lij•
Moreover, (v' Oxl R)~kj = (CH'v' Oxl R))kj = (v' Oxl (Cf R))kj = (v'Oxl Ric)kj·
Using these in (9.6), and multiplying by gkr, we obtain
0 = gkr(v'ax;R)tlj - gkr(v'axkRic)tj + gkr(v'axlRic)kj·
Using (v'. Ric)U = v'. (Rieu) (Exercise 6.24, p.130), we have
gkj (v' Oxl Ric)kj = gkj (v' Oxl Ric)(Oxk' Oxi) = (v'Oxl Ric)( (dxi)U' Oxi)
= (v'i)xl Ric)U (dxj' Oxi) = (v'i)xl (Rieu) )(dxj' Oxi)
= (v'axl (Rieu)){
Using also the symmetry of Ric, (v'axkRic)tj = (v'axkRic)jt, and so
(div Ric)£ = (div Ric)(oxt) = (v' axk Ric)( (dxk)U, Oxt))
= (v' i)xk Ric)(gkj Oxi, Oxl))
= gkj (v'Oxk Ric)(Oxi' Oxl) = gkj (v'Oxk Ric)(oxt, Oxi)
k.
= g 1(v'axkRic)£j· (9.7)
Curvature 191
Thus
k"3 . k"
0= g (v'a.,;R)l,ej - g 3 (v'a.,kRic)ej + g k"3 (v'a.,£Ric)kj
= gki(v'a.,;R)1ej - (divRic)e + (v'a.,£(Ricu))1
= gki(v'a.,;R)1ej - (divRic)e + Cv'a.,£(Ricu)
= gki(v'a.,;R)1ej - (divRic)e + v'a.,£C(Ricu)
= gki(v'a.,;R)1ej - (divRic)e + v'a.,£S
= gki(v'a.,;R)1ej - (divRic)e + DxtS
= gki(v'a.,;R)1ej - (divRic)e + (dS)e-
Finally, we will show that the first summand on the right-hand side of the
last equality is -(divRic)e, completing the proof. Since R = CHR ® g),
v'.R=CH(v'.R) ®g), we obtain
(v'a.,;R)pqrs = (v'a.,;R)!qrgst, and gsu(v'a.,;R)pqrs = (v'a.,;R);qr·
Using the skew-symmetry of R in its last two entries, it follows that
(v' o.,; R)kfjs = - (v' o.,; R)kfsj, and so
kj i kj is ~ kj is
g (v' o.,; Rhej = g g (v' o.,; R)kfjs = -g g (v' o.,; R)kfsj
~
= -g kj g is gjr ('7 vo.,; R)kkfs
vo.,; R)rkfs = -gis(-7
= -gis(Ctv'a.,;R)es = -gis(v'a.,; (CtR))es
= - g is ('7
v o.,;
R.lC) fs (9.=7) - (d"IV R"lC)f.

Corollary 9.1. Let (M, g) be a semi-Riemannian manifold with the Levi-
Civita connection v'. Then div (Ric - ½S g + Ag) = 0.
Proof. Let (U, x) be any admissible chart. Then
(div(Sg))i = C(TfMxTJM 3 (n, V) ,_. (v'v(Sg))(nU,ox;))
= (v'a.,k(Sg))((dxk)U,ox;)
= (OxkS)g(gkfOxt,Oxi) + S(v'a.,kg)((dxk)U,Oxi)
= (dS)kgkfg(oxt,Oxi) + S-0 (since v'.g = 0)
= (dS)k gke gei = (dS)k t5f = (dS)i.
So
div(Ric-½sg) = 0.
As div(Ag) = Adivg =AO= 0, also div(Ric - ½Sg +Ag)= 0. □
9.2 Cosmological time and cosmological red-shift in FLRW
In this section, we will consider the FLRW spacetime (M, g), as described
in Example 5.9 (p.89), where M = I x l~.3, and I= (0, oo ). In the chart
(M, idixIRa ), the chart-induced component matrix G = [gij] of the metric is
the diagonal matrix with diagonal entries -1, a 2, a 2, a 2. Now we will take
a special5 a: J-+ I, namely
2
a(t) = ta, tE (O,oo).
We will first calculate the scalar curvature, and note that the chart-
coordinate t has a geometric meaning. Then we will derive a 'cosmological
version' of the red-shift.
Scalar curvature and cosmological time when a= ti

We had seen that in an admissible chart (U, x), the curvature tensor field
R has components given in terms of the connection coefficients by
Rfjk =oxsfj - Oxir{i + rkjr;i - rkir;j.
So Ricij = R~ij = Oxkrji - Oxirjk + r;ir~k - r;kr~i• The connection
coefficients for the FLRW spacetime were given in Example 6.7 (p.124).
Using these, for example we can calculate, with (t, x, y, z) = (x 0 , x 1 , x 2, x 3 ),
Ricoo = Oxkr~o - Otr~k + r~or~k - r~kr~o
= -atr~1 - atr6 2 - atrg3 - r~ 1r~ 0 - r~ 2 r~ 0 - r~ 3 r~ 0
a a2 aa - a2 a2
= -30t- - 3- = - 3 - - - - 3 - = -3-.
a
a a 2 a 2 a 2 a
Similarly,
Rieu = Oxkrt1 - Oxirtk + rr1r~k-rrkr~1
= Otr~l - Oxlr~o + r~lr~k - r~lr~l - nor~l
~ ( ") 3 . a 2aa-
= ut aa + aa- -
. a = aa. + a-2 + 3•2
a - 2•2
a = aa.. + 2•2
a .
a a
By symmetric role played by the x 1 , x 2, x 3 variables, Ric 11 = Ric22 = Ric 33 .
The other components of Ric are all zeroes. Since [gij] is a diagonal matrix,
(Ricu)g = -Ricoo =3~,

a
(Rieu)~= (Rieu)~= (Rieu)~= _!_Ric 11 =
a2
aa+a 22a2
--------------
This choice corresponds to the so-called Einstein-de Sitter universe, which was for
5
some time believed to be the correct model for our universe. It assumes a matter-
only universe with a zero cosmological constant. The Einstein-de Sitter model is a good
approximation to our universe in the period after the radiation-dominated era but before
the dominance of the positive cosmological constant.
Curvature 193
Consequently, the scalar curvature is given by

S-(R" ~)k_3a 3aa+2a2 =6aa+a2
- IC k - ~ + a2 a2
Thus the t-coordinate in the global chart (M, id1 xJR3) can be expressed in
terms of the geometric quantity S. We call this the 'cosmological time'.
Definition 9.6. (Cosmological time for FLRW).

Let (M,g) be the FLRW spacetime with a(t)=tf The function t:M-+ JR,
t(p) = ~ f o r allpE M,
is called the 6 cosmological time.
We note that as t(p) -+ 0, S(p) -+ oo, and so the scalar curvature blows up
as we go back in cosmological time (and we think of this as approaching
the 'big-bang singularity').
Geodesics in FLRW
Let J c JR be an open interval, and suppose that 'Y : J -+ M is a lightlike
geodesic. In the global chart (J x JR3 , id), write ,y(s) = (t(s ), x(s ), y(s ), z(s) ),
s E J. As 'Y is lightlike, g('Y(s))(v.,,,-y(s),v-y,-y(s))=0 for alls E J, giving
0 = -t' 2 +(a(t))2(x' 2+y' 2 +z' 2), (9.8)

where .' = fs. The geodesic equation gives in particular (using the connec-
tion coefficient expressions from Example 6.7, p.124, and noting that r?i is
nonzero only when 0 =I= i = j, and then r?i = aa, i=l,2,3):
t" + a(t)a(t)(x' 2 +y' 2+z' 2) = 0. (9.9)
Here a(t)= ~~(t(s)), s E J. Equations (9.8), (9.9) imply that Q:=a(t)t' is
conserved along the lightlike geodesic, since
Q' = (a(t) t')' = a(t) t' t' + a(t)(t')' = a(t)(a(t) )2 (x' 2+y' 2 + z' 2) +a(t) t"
= a(t)(t" + a(t)a(t)(x' 2 +y' 2+z' 2)) = a(t)0 = 0.
We will use this fact below to derive the so-called 'cosmological red-shift'.
6 A similar definition of cosmological time can be made for many other choices of a(t).
For example, if a(t) =tP, where p > ½, then S = 6p(2p-1)/t 2 •
Now instead of a lightlike geodesic, consider for p, q E IR 3 , the curves

µp,µq: I--> M, given as follows, in the chart (Ix IR3,id):
µp(t)=(t,p), µq(t)=(t,q) , tEl=(0,oo).
Both are geodesics: As Vµp ,µp(t) = Ot,µp(t),
v' Vµp Vµp = v' Ot Ot = r~oOxk = 0.

Similarly also v'vµq V.µ q = 0. These geodesics can be thought of as the
worldlines of two freely-falling galaxies.
Cosmological redshift when a= ti

With notation as above, now consider a light ray, 'Y : J --> M, emanating
at the event A= (t, q) = "f(SA) on the worldline of µq , which reaches the
worldline of µp at the event B = (T, p) = 'Y( s B), as shown in the picture.
t
T
The energy /frequency EA of the light signal transmitted/perceived by the

galaxy µq at A is given by
EA= -g(A)(vµq,A,V, ,A )
✓- g(A)(vµq ,A, Vµq,A)
= - g(A)(ot ,A , t' (sA)Ot ,A +x'(sA)Ox,A +y'(sA)Oy,A + z'(sA)Oz,A) = t' (SA)-
✓- g(A)(ot,A, Ot,A)
Similarly, the energy /frequency EB of the light signal received/perceived by

the galaxy µP at the event B is EB = t' (s B). Since Q = a( t) t' is conserved
along the lightlike geodesic 'Y, we have
a(t(sA)) t' (sA) = a(t(sB)) t' (sB)-

Curvature 195
Consequently,
EA = t'(sA) = a(t(sB)) = a(T) = (T)f > l
EB t'(sB) a(t(sA)) a(t) t ·
Thus the frequency, when received by a 'comoving' observer along the
galaxy worldline µp (imagine this to be us), is shifted towards the red
end of the spectrum (i.e., lower frequency), as compared to what is mea-
sured by a comoving observer with galaxy µq. This reduction in frequency,
described by the quantity
EA
Z=--1=
EB
is called the cosmological red-shift.
Exercise 9.13. The age of the universe is estimated to be T = 13.8 billion years.
If the light from a distant galaxy, reaching us now, has the cosmological red-
shift z = ½, then determine how long ago the light was emitted. (This exercise
assumes that the Einstein-de Sitter model is the correct model for our universe.
As mentioned earlier, this is approximately correct from the Big Bang until today,
but not at all in the far future.)
9.3 Geodesic deviation and curvature
Intuitively we expect that the manner in which geodesics emanating from

a point 'diverge' should tell us something about the curvature.
If we imagine the geodesic as a thread on a surface passing through a point,

and we slide it across this surface through the point in such a way that each
intermediate position of the thread is still a geodesic, then the velocity with
which the points on the geodesics travel transversely, defines a vector field
along the original geodesic.
Definition 9.7. (Family of geodesics; main and transverse curves).

Let M be a smooth manifold with a connection v', I c JR be an open
interval, and E> 0. A family of geodesics is a smooth map r: (-E, E) x I---+ M
such that for each s E (-E,E) , the smooth curve fs : I---+ M , given by
r s(t) = r(s, t), t EI, is a geodesic. The geodesics rs, s E (-E, E) are called
main curves. The smooth curves I\ : (-E, E)---+ M , given by I\(s) = f(s, t),
s E ( -E, E), are called transverse curves. For s E ( -E, E) and t E J, define
V(s, t) = Vrs, r s (t) E Tr(s,t)M,
J(t, s) = Vf\,i\(s) E Tqs ,t)M.
M
i\
V.
r
We already know that the velocity vector field V(s, •) along a main curve
fs is a vector field along fs, that is, V(s, ·) E TJrs for each s E (-E, E).
Similarly, any transverse curve I\ has its velocity vector field J (·, t) E Tc} I\
for each t E J. Moreover, we have the following.
Lemma 9.2. We use the same notation as in Definition 9.7. We have
(1) V(·, t) E TJI\ for each t E J.
(2) J(s, ·) E TJrs for each s E (-E, E).
1~
(3) v' J(•,t) V(-, t) E T0 ft.
(4) v'v(s,-)J(s, ·) E TJfs.
(5) v'v(s,-)V(s, ·) = 0 for alls E (-E,E).
Proof. Below, (U, x) is any admissible chart containing the point f(t, s).
o(x O r)i o(x O r)i .
(1) We have vr.,r.(t) = ct (s, t) Oxi ,r,(s)· As ct (·, t) 1s smooth,
(-E,E) 3 s ~ V(s, t) E TJI\.
. o(x O r)i
(2) Analogous to the above, by observmg 08 (s, •) E C 00 (J).
(3) Follows from (1).
(4) Follows from (2).
(5) Since fs is a geodesic, we have v'v(s,·)V(s , ·) = 0 for alls E (-E, E). □
Curvature 197
Exercise 9.14. With the notation used in Definition 9.7, show that
(1) l 3 t >-+ (V J(•,t) V(·, t))(s) E TJr. for each s E (-E, E).
(2) (-E, E) 3 s >-+ (Vvcs,-)J(s, ·))(t) E TJi\ for each t EI.
We now establish the promised link between the behaviour of a family of

geodesics and the curvature operator R. To do so, we will assume that we
have a semi-Riemannian manifold with its Levi-Civita connection (since as
we shall see in the proof, we will use the torsion-freeness of the connection).
Theorem 9.1.
'v, I c JR. be an open interval, f > 0, and r: (-1:, 1:) x I-+ M be a family
of geodesics. Suppose that there exist vector fields J, V E TJ- M such that
Jqt,s) = J(t,s) and Vr~,-?] ': V(t,s), for all (s,t) E (-1:,1:) x I. Then we
have7 'vv'vvJ = -(R(J, V)V) or on (-1:, 1:) x I.
Proof. Let JR. 2 be given the standard smooth structure. We consider

N = (- 1:, 1:) x I as a smooth manifold with the induced smooth structure
from JR. 2 • Then [ot,os] = 0 in the global chart (N,id). Using this, we will
show that also [J, V]or=O on (-1:, 1:) xJ. Firstly, V(s, t) = (dr)(s,t)(Dt,(s,t)),
because for all/ E C 00 (M), we have
((dr)(s,t)(Ot,(s,t)))f = Ot,(t,s)U O
au or) (s, t)
r)(s, t) = ot
d(f or.)
= dt (t) = vr.,r.(t)i = V(s, t)f.
Similarly, J(s,t) = (dl')(s,t)(os,(s,t))- Secondly, for (s,t) E (-1:,1:) x I, and
f EC 00 (M),
((V f) 0 r)(s, t) = (V f)(r(s, t)) = Vqs,ti! = V(s, t)f
= ((dl')(s,t) (ot,(s,t)))(f) = ot,(s,t) (for)= (ot (for) )(s, t).
Similarly, ((]!) 0 r)(s, t) = Os,(s,t)U Or)= (o.(f Or))(s, t). Hence for all
(s, t) E (-€, t:) XI and all/ E C 00 (M),
[], V]r(s,t)f = Jqs,t)(Vf)-Vqs,t)(]f) = J(s,t)(Vf)-V(s,t)(]f)
= ((dl')(s,t)(Os,(s,t)))(Vf) - ((dl')(s,t)(Ot,(s,t)))(]f)
= Os,(s,t)((Vf) 0 r) - Ot,(s,t)((]f) 0 r)
= os,(s,t)(ot(f o r))-ot,(s,t)(o.(f or))

= oosot
2
(f Or) ( t) - o2 (f Or) ( t) = 0
s, otos s, .
7 We call this the Jacobi equation.
Consequently, [J, V] or = 0. Now we are ready to show the equation

given in the theorem statement. By the torsion-freeness of the Levi-Civita
connection, '\i' JV - v\; J = [J,V]. Thus ('v JV) 0 f = ('v V ]) 0 f. By
Exercise 7.2 (p.143), and the fact that the fs are all geodesics, we have
('v v V) or = 'v v V = 0. Recalling the definition of the operator R, we have
(R(J, V)V) or = ('v 1v v v- 'v v'v1v - 'v[J,vJ V) or
= ('v 1 o - 'v v'v vl - 'v oV) or
= 0- 'vv'vvJ - 0 = -'vv'vvJ. □
Exercise 9.15. We use the notation from Theorem 9.1. Show that
d2
dt2(g(rs(t))(J(s,t),V(s,t)))=0, sE(-E,E), tEJ.
Hint: Use Proposition 7.2 (p.146) and the Jacobi equation.
Let to E J, and J(s, to), (v'v(s,-)J(s, ·))(to) be orthogonal to vrs ,rs(to)· Conclude
that J(s, t) and (v'v(s ,-)J(s, ·))(t) are orthogonal to vrs ,rs(t) for all t E J.
9.4 Tidal forces
Theorem 9.1, relating the geodesic deviation with the curvature operator R,
can be used to explain the notion of tidal forces in the spacetime context.
We first adopt the classical Newtonian perspective. Consider any two con-
stituent particles of an extended body in an inhomogeneous gravitational
field. The resulting different accelerations lead to internal stresses in the
body, which are called tidal forces. Imagine for example a long rod falling
in the gravitational field of the Earth as shown in the picture below.
In contrast to the classical understanding, in the spacetime viewpoint the

trajectories of two particles are just geodesics, and the curvature of space-
time produces geodesic deviation, resulting in relative acceleration amongst
the particles, which is perceived by each one of the particles as a force. To
make this precise, we first define an instantaneous observer.
Curvature 199
Definition 9.8. (Instantaneous observer, space perceived).

Let (M, g) be a time-oriented Lorentzian manifold with a time-orientation
W E TJ M. An instantaneous observer at p E M is a vector v E TpM which
is timelike and future-pointing. The space perceived by an instantaneous
observer v E TpM is defined to be v1_ c TpM , where
v1_ := (span{v})_1_ = {w E TpM: g(p)(v , w) = O}.
We imagine the worldline 'Y of a 'freely-falling' observer (that is, a geodesic)
passing through the event p = "f(O), where it has the velocity v'Y ,P = v.
Note that U := span{v} ENg(p), as g(p)lu is negative definite, and so by
Theorem 5.3, g(p)lv.L is positive definite. Suppose that there is a family
of geodesics f such that f(O , ·) = 'Y , and for each8 t, J(O, t) E v~,'Y(t)" By
Taylor's formula, for small t, s, in any admissible chart (U, x) containing p,
· · o(x O f)i ·
(xof)'(s,t)-(xof)'(O, t) ~sos (0, t) = s(J(O,t))'.
Hence, formally, we can think of the spacelike vector sJ(O, t) as describing
the vector separation, at time t, in v~,'Y(t) ' of the second particle at f(s, t),
perceived by the instantaneous observer v'Y,'Y(t). So the rate of change of
this separation with the elapsed proper time is
(9v~sJ(0, ·))(t)
✓-g('Y(t))(v-y,-y(t),V-y,-y(t))'
and so the instantaneous 'acceleration' of the second particle reckoned by
the observer v is
(9v~ 9v~sJ(0, ·))(O) R(p)(x, v)v
-g(p)(v,v)g(p)(v,v) '
where x:=sJ(0,0). Here R(p)(x,v)v means (R(X, V)V)p; X, VE TJM are
any vector fields such that Xp = x and VP= v; see Exercise 6.15 (p.125).
This motivates the following definition.
A B
8 This is guaranteed if J(O,O) Ev.land ('v'v(o,-)J(O, -))(0) E v.l; see Exercise 9.15.
Definition 9.9. (Tidal force 9 ).

Let (M, g) be a time-oriented Lorentzian spacetime with a time-orientation
WE TJ-M. Let v E TpM be an instantaneous observer at p EM. Then the
tidal force is a map Fv : v_1_ -+ v_1_ given by
R(p)(x, v)v _1_
Fv (X) = g (p) (v,v ) for all X E V •
Fv is well-defined, i.e., Fv (w) does belong to v_1_ for each x E v_1_, because
g(p)(v Fvx) = g(p)(v, R(p)(x, v)v) = R(p)(x, v, v, v) = O = 0

' g(p)(v,v) g(p)(v,v) g(p)(v,v) '
where we have used the skew-symmetry of R in its last two slots. We also
note that Fv is a linear map (from the subspace v_1_ of TpM to v_1_ ).
Example 9.2. (Schwarzschild spacetime). Consider the Schwarzschild

spacetime (M, g) with the time-orientation described in Example 5.14
(p.92). Consider the chart JR x (2m,oo) x U given there, where (U,cp)
is the chart for S 2 described in Example 5.7 (p.87). Take p EM having the
coordinates (t, r, 0, ¢). Set v = Ot,p· Then v_1_ has a basis comprising the
three vectors: B = {Offior,pEBO, OEBOEB00,p, OEBOffio,t,,p}; see Example 5.10,
and we will henceforth denote these three vectors simply by Or,p, 00,p, O,t,,p·
To determine the matrix of the linear map Fv : v_1_ -+ v_1_ with respect to
this basis, we first compute
Fo _ R(p)(or,p,ot,p)ot,p _ (v'orv'o,ot-v'o,v'orot-O)p
v r,p - g(p)(ot,p,Ot,p) - -(1- 2;.") '
where we have used [or, Ot] = 0. Using the connection coefficients noted in
Example 6.8 (p.124), we have v' 0,ot = r;tor = -;!½(1- 2;,")or, and so
v' or v' o,Ot = v' or(; ( 1- 2 r; )or) =or(; ( 1- 2 r;))or+; ( r; )v 1- 2 or Or
2m 6m 2 ) m( 2m) r 2m 5m ) 2
= ( - -;y+7 or+ r 2 1---;:- rrror =
(
--;y+--;:r- or.
Also,
9 Although we call it a force, it is more appropriately an acceleration.

Curvature 201
2m 5m m )o
(-----;:F"+--;:r-,:r
2 2
1 Fvur,p=
Consequenty, :i r,p 2m :i
-(l- 2;,") =~ur,p·
Similarly, Fvo0,p = - ; 00,p and Fvoq,,p = - ; Oq,,p· Thus
[F.]B ~ [ ~ -;.'½ -;.'½ l

is the matrix of the linear map Fv : v1- -> v1- with respect to the ordered
basis B consisting of the vectors Or,p, 00,p, Oq,,p (taken in this order).
The matrix of the map Fv shows that the tidal forces of spherically
symmetric bodies stretch along the radial direction and compress along the
tangential directions; a consequence of this is that the Moon deforms the
oceans by creating two bulges along the Moon-Earth line, which gives rise
to the two daily tides as the Earth rotates. ◊
Exercise 9.16. Let (M, g) be a Lorentzian spacetime with a time-orientation given

by WE TJ M. Let v E TpM be an instantaneous observer at p EM. Show
t (F ) _ Ric(p)(v, v)
race v - g ( p )( v,v ) .
Exercise 9.17. Suppose that an astronaut can withstand an acceleration per unit
length of 100 s- 2 along his/her height. Consider the Schwarzschild spacetime,
which in this exercise, is assumed to model the spacetime in the vicinity of a
black hole. Determine the minimum mass m of a black hole, in terms of multiples
of solar masses, that a radially aligned astronaut can tolerate arbitrarily close to
the 'event horizon', i.e., r=2m+o, o>O arbitrary. Suppose that the astronaut is
an instantaneous observer with v = Ot,p E TpM as in Example 9.2. The mass M0
of the sun is M0 ~ 5µs= 5 x 10- 5 s (in units such that GN =c= 1).
Appendix: 'Locally flat' means R=O
In this final section, we will show that for a semi-Riemannian manifold

(M, g) with Levi-Civita connection v', the curvature tensor field R vanishes
if and only if M is locally flat. Recall from Exercise 6.15 (p.125) that an
admissible chart (U, x) is called an affine chart if the connection coefficients
with respect to the chart are all identically zero, and the manifold M is
called locally flat if for every point p EM, there exists an affine chart (U, x)
such that p EU. We will show 10 Theorem 9.2, for which we will need the
following result, called the Frobenius theorem 11 .
10 Thisis based on [(Moretti), §9.2].
11 Seefor example, [Frobenius(1877)] or [Shilov(1975), Chapter 2,§2.5], but we include a
proof here based on [Hakopian and Tonoyan(2004), Theorem 1.1].
Proposition 9.7. (Frobenius).

Let V c JR.m be an open set, p E V, and /ii : V x JR.n -+ JR. be smooth maps
for l ~ i ~ m, and l ~ j ~ n. Consider the following 'initial value' problem
for the system of partial differential equations in the unknown real-valued
functions yi = Yi(x1, • • • , xm), 1 ~j ~n (with x = (x1, • • • , xm) EV):
1 ~ i ~ m, 1 ~ j ~ n,
(9.10)
1 ~ j ~ n.
Suppose that the maps

(x\··· ,xm,x1, ... ,xn),__./ij=/ij(x\··· ,xm,x1, ... ,xn)
satisfy the following consistency condition in V x JR.n:
afii aJ;i f _ a/ki a/kif

axk + axe kl - axi + axe if.
Then there exists an open set U c V containing p such that there exists a
unique C 00 solution Y 1 , · • · , yn : U-+ JR. to (9.10).
Before giving the proof, we note that if there were a solution, then clearly
the Schwarz theorem on the exchangeability of the order of differentiation
for smooth functions implies
which yields the consistency condition. The content of the theorem is that
this is the only obstruction to the solvability of (9.10).
Proof. We will use induction on m. In the base case m = 1, we just have

an ordinary differential equation system with an initial condition at p E JR.:
and the consistency condition is trivially satisfied. Then the usual result 12
on existence and uniqueness of a smooth solution guarantees a solution in
some neighbourhood of p.
12 See [Apostol(1969), Thm. 7.19, p.229] for the continuously differentiable (C 1 ) version.
The C 00 version can be obtained by successively differentiating the ODE and applying
the C 1 result.
Curvature 203
Now suppose that the statement is true for m=k-1. Let m=k. We write
p = (p 1 , ... , pk). Consider the following (k-1)-variate problem in the hyper-
plane H={(x1,---,xk) E ]Rk: xk= pk}:
oYk (X 1, • · • , X k-1) =
~
f ij (X 1, · · · ,X
k-1 , Pk , yl
H, .•. ,
yn)
H (9.11)
ux'
(1,;;;i,;;;k-1, 1,;;;j,;;;n)
(1,;;;j,;;;n). (9.12)
By the induction hypothesis, this has a unique smooth solution in the neigh-
bourhood of a compact subset of H given by
K 1 ={(x 1,··· ,xk-l , pk): lxi-pil,;;;81 , 1,(i,(k-1} ,
for some 81 >0.
Next, for each (x 1, · · · , xk-l, pk) E K1 C H, consider the univariate initial

value problem on the line L = {(x 1' ... 'xk) : x 1 = x1' ... 'xk-l = xk-l }:
dYJ, ( k) f (~l ~k-l k yl yn) (9.13)

dxk x = kj x ' ... ' x 'x ' L ' ... ' L
(9.14)
By the ODE result on existence and uniqueness, this problem also has a
unique solution in a neighbourhood of
K 2 : = {( X~1 , · · · , X~k-1 , X k) : IX k - p kl ,( u2
s: } ,
for some 82 > 0. The smooth dependence on initial conditions 13 implies that
Yl, · · · ,YE depend smoothly on the parameters x1 , · · · , xk. A compactness
argument can be given to arrange a uniform 82 > 0 which works for all lines
emanating from all initial conditions (x1, · · · , xk-l ,pk) E K1. Let
U:= {(x1, · · · , xk) E ]Rm: lxk -pkl < 82, lxi -pil < 81 for 1,( i ,( k-1} 3 p.
13 See e.g. [Hartman(2002) , Thm. 4.1, p.100].
If a solution of (9.10) exists on ff, then by the uniqueness part of the

induction hypothesis, this solution must coincide with YJ, • • • , Yfj on an
open subset U1 c K 1 of H with p E U1 . Thus in view of the uniqueness
result for OD Es, it also coincides with Yf, •• • , Yf on each line L emanating
from a point in U1 cK1 . So the solution is unique on the open set
U:=ff n {(x1, • • • ,xk-l,xk) E llm: (x1, • • • ,xk-l,pk) E Ui}.
Denotethelineemanatingfrom(x 1 ,··· ,xk-l,pk)EK1 byL(x1,··· ,xk- 1),
i.e. L(x1, • • • , xk-l) = {(x1, • • • , xk-l, xk) : xk Ell}. It remains to show
Y j ( x,···,X
1 k) ._ yj ( k) (x,···,X
1 k) E ~ · ~
K' l -.:J-.::n,
.-L(xl,-.. ,xk-l)x'
is a solution of (9.10) on ff. The initial condition is satisfied, as for 1 ~j ~n,
Y j( p 1 ,··· ,Pk-1 ,Pk) = yiL(pl, .. ,pk-1) (p k) = yiH (pl ,··· ,Pk-1) = Yi •
As Yi, is already a solution of (9.13), we only need to check that

j •- oYi (1 k
T/i •- ~ - /;j x ,··· ,x ,Y ,··· ,Y
1 n)-
-0
ux'
We have
a .
oxk 'Tfi
So the vector T/i = (r,l, · · · , r,f) satisfies a homogeneous linear system of

ODEs given by
d
dxk T/i = MT/i, (9.15)
where M = [Mf] is the n x n matrix with entries Mf = ~~!. But with

x:= (xl' ... 'xk-1 ),
,,,J k
'Ii (x,p ) =
oYi k k 1 k n k
~(x,p ) - fo(x,p , Y (x,p ), • • • , Y (x,p ))
ux'
(9l4) oYt (X ) _ !· _(X, p k , ylH ( X ) , . . . , yn(
:l . iJ H X
)) (9ll) O.
ux'
Curvature 205
It now follows from (9.15), and the uniqueness of solutions to the initial
value problem for OD Es, that T/i must be the trivial solution, that is, T/i = 0,
as wanted. □
Exercise 9.18. Consider the smooth manifold M = IR.3 with the standard smooth
structure. Show that a 1-form field n E Tf M is the gradient df of some smooth
function f E C 00 (M) if and only if the 'curl of n~' is zero, that is,
oyOx = oxOy, ozOy = oyOz, oxnz = oznx,
where n is decomposed as 0=0xdx+Oydy+Ozdz in the global admissible chart
(IR.3, (x, y, z) >-+ (x, y, z)).
Theorem 9.2. Let (M, g) be a semi-Riemannian manifold with Levi-Civita

connection v'. Then M is locally flat if and only if R = 0.
Proof. The 'only if' part is trivial, and was noted in Exercise 6.15 (p.125).
We now prove the 'if' part. Let p EM, and v1 , • • • , Vm be a basis of TpM.
Step 1. We first show that there exist vector fields Vi,••• , Vm defined in
a neighbourhood U of p such that for all 1,:;;; i,:;;; m, we have (½)p = Vi,
v'.½ = 0, and for all q EU, {(V1)q, · · ·, (Vm)q} is a basis for TqM. To do
this, we first take an admissible chart (V, x) containing p, and use it to
write the as-yet-undetermined vector fields in terms of their components.
Fix an r E { 1, • • • , m}. Then the conditions v'. Vr = 0 and (Vr )p = Vr give the
following initial value problem for the component functions V/, • • • , vrm:
Dxi V/ = -r~i_V..s, 1,:;;; i,j,,;; m, (9 _16 )

{
(V;)p = Vr(x'), 1 ,:;;; i ,:;;; m.
These can be rewritten as an initial value problem differential equations in
the region x(V) x JR.m for unknown functions V/ ox- 1 , and using the Frobe-
nius theorem, it can be seen that (9.16) is guaranteed a smooth solution in
some open set U c V containing p if the following consistency conditions
are satisfied:
-(Oxkr~i)X 8 - r~i8e(-rfkxt) = -(oxS~k)X 8 - (r~k8:)(-rfixt).

for all X1, • • • , xm E R By rearranging and relabelling dummy indices,
this can be rewritten as (Oxkr~i - OxS~k + r~kr!i - rtr!k)X 8 = 0, that
is, R{; 8 X s = 0, which is true because R = 0. Hence there is a subset U c V
containing p and C 00 functions V,.1 , · • · , vrm such that with Vr := V/oxi, we
have that v'.V.. = 0 in U, and (V..)p = Vr- Repeating this with the other
r E {1, • • • , m} gives m vector fields V1 , • • • , Vm defined on some open set
U c V (taken as the intersection of each of the sets U = Ur obtained for
the various r) that contains p. Next we will show that for each q E U,
(Vi)q, • • • , (Vm)q form a basis for TqM. It is enough to show that they are
linearly independent. Let c1 , • • • , cm Ell be such that ci(½)q = 0. Consider
the vector field W = ci ½. This solves the equation v'. W = 0 with the initial
condition Wq = 0. But from the calculation done above, this is equivalent
to an initial value problem
:i
Ux'
· Wi. -- -
risi ws , l -=::,
.,,,. z,· J· .,,,.
-:::, m,
{
w~ = 0, 1 ,:;;; i ,:;;; m.
And just as done above, again the consistency conditions are satisfied due to
the vanishing of R. The Frobenius theorem then implies that the solution
is unique, implying that W = 0 in a small neighbourhood of q. In fact, the
subset ff:= {q E U : Wq = 0} is nonempty (q E ff), and (using the above
argument) open. Also the set U\ff consisting of the set of points where
the evaluation of W is a nonzero tangent vector is open (this is true for
any smooth vector field, since the component function are smooth, and in
particular, continuous). Thus we have a decomposition U = ff u (U\ff) as
a disjoint union of open sets. By shrinking U at the outset of necessary, we
could have assumed that U is connected, and thus we are able to conclude
that the nonempty set ff must be all of U. In particular, ci(½)v = 0, that
is, civi = 0, and so c1 =•••=cm= 0 (since v1 , • • • , Vm was a basis for TpM).
Step 2. In this step we will construct a new chart (U, y) containing p,
which will be the desired affine chart. To do this, we will first construct
I-form fields Oi from the ½ of Step 1, and then solve a system of differential
equations involving the components of the Oi in order to find the coordinate
functions yi. For each q EU, let (0 1 )q, • • • , (nm)q be the basis for (TqM)*
which is dual to the basis (V1 )q, • • • , (Vm)q for TqM.
Claim: For 1 ,:;;; k,,;; m, U 3 q >-+ (Ok)q E (TqM)* defines a I-form field
ni E rpu.
Firstly, since the vectors (V1 )q, • • • , (Vm)q form a basis for TqM, the matrix
M(q) := [((°Vj)q)'] is invertible for all q EU Also, as q >-+ ((Vj)ql = (V/)(q)
are smooth maps, the entries of M(•)- 1 belong to C 00 (U). Finally, we
have 8j = (Ok)q(Vj)q = (Ok)q(((Vj)qloxe,q) = ((Vj)ql(Ok)q,£, showing that
(Ok)q,t=(M(q)- 1 )r So q >-+ (Ok)q,£ belongs to C 00 (U), and nk is smooth.
Also, v'.Ok=0. To see this, we note that for all WE TJU,
o = W(8f) = v'w(Oi(½)) = (v'wOi)½ + ni(v'w½)
= (v'wOi)½ + ni(o) = (v'wOi)K
Curvature 207
But for each q, (V1 )q, • • • , (Vm)q forms a basis for TqM. Consequently,
v'wf2i =0, l~j~m.
Now we seek real-valued smooth functions yi, 1 ~ i ~ m such that
(9.17)
(We do not specify an initial condition for yi(p), 1 ~ i ~ m, which can be
taken as an arbitrary m-tuple.) Again, using the Frobenius theorem, this
is possible if Oxk n} = Oxi ni, 1 ~ i, j, k ~ m. But this is true, and can be seen
as follows. Firstly,
Oxkn} = v'a.,k(ni(Oxi)) = (v'a.,kni)(Oxi) +niv'a.,kOxi
e
= O(Oxi) + ni (rjkOxi) e i
= rjkne.
Similarly, Dxint = rf,jn~. By the torsion-freeness of the Levi-Civita con-
nection, rf,j = r;k. Hence the consistency condition in Frobenius theorem
is satisfied for all i,j, k. We will continue to denote the possibly smaller
subset of U containing p on which the functions yi exist, still by U. Define
the matrix
N := [oxiYi] l
C9 7 l [n{].
Then N is pointwise invertible in U since (!1 1 )q, • • • , (nm)q form a basis
for (TqM)* for each q EU. Hence the Jacobian of the 'transition function'
yo x- 1 is pointwise invertible on x(U). Also, as N is smooth on U, it
follows that yo x- 1 is smooth on x(U). By the inverse function theorem,
and by shrinking the neighbourhood U of p if necessary, it follows that
yo x- 1 : x(U) -+ (yo x- 1 )(x(U)) = y(U) is a diffeomorphism. Thus
y =yo x- 1 ox : U -+ JR.m is a diffeomorphism onto the open set y(U).
Hence (U, y) is an admissible chart for M. We have
k f fk k k .
8j = n (½) = nkdx (l'j Ox£)= nk l'j <>e = nk l'j = l'j OxiYJ,
i i i i i
½ = ½(yk)Oyk = ((l'jeOxi )yk)Oyk = ½e(Ox£Yk)Oyk (:J 8joyk = Oyi.

We have that in the chart (U, y),
rji = dyk (v' ayi oyi) = dyk (v' ayi ½) = dyk (o) = o.
Consequently, the admissible chart (U, y) containing pis an affine chart, as
wanted. D
Chapter 10
Form fields
In this chapter, we will discuss special types of tensor fields on M, called

'form fields' (a special but trivial example of which is a 1-form field), and
develop a useful calculus of these, after introducing the 'exterior deriva-
tive'. This will enable us to write Maxwell's equations of electromagnetism
succinctly in a later chapter. The exterior derivative generalises the grad-
div-curl vector calculus language from JR3 to smooth manifolds.
The language of form fields will also allow us to talk in particular about
a 'volume form field' on a Lorentzian manifold, enabling a coordinate-free
manner of integrating functions on a manifold. The simplest form field is a
1-form field. A one form field can be integrated along a curve: at any point
of the curve, we act the point evaluation of 1-form field on the tangent vector
to the curve, which produces a number, and we add these contributions
along the curve. The higher k-forms can be used to integrate along higher
dimensional 'submanifolds', but in this book we will only consider the full
manifold case and integrate m-forms on m-dimensional manifolds.
10.1 k-forms
An 'alternating' (0, k)-tensor on a vector space V is called a k-form.
Definition 10.1. (k-form).

Let V be a vector space and k ~ 2. Then a k-form on V is a (0, k)-tensor
w : V x • • • x V --+ JR which is alternating, that is, for all 1 ~ i < j ~ k and
all v1 , • • • , Vk E V, swapping the i th and l h entries produces a minus sign,
that is,
w(··· ,V",,;,··· ,v;,···) = -w(··· ,v;,··· ,vi,···).
The set of all k-forms on Vis denoted by /\kV*. We set /\ 1 V = V* and
/\ 0 V =R
209
Clearly the action of a k-form with k;;,:: 2 on k vectors, any two of which
are identical, is 0. This also true if the k vectors are linearly dependent. In
particular, if Vis m-dimensional, then if k > m, every k-form w is 0, and
so ;\kV*= {0} for all k>m.
We now develop notation and collect preliminary facts on permutations,
which will be needed in the proofs.
We recall that a bijective map 1r : { 1, • • • , k} --+ { 1, • • • , k} is called a
permutation. We specify a permutation by writing the image 1r(i) below
i E {1, • • • , n} in a row as follows:
Crt1) ::: 1r(n)). (10.1)
The set of all permutations on {1, • • • , k} is denoted by Sk, and it forms a

group under composition. The group is non-Abelian if k;;,:: 3:
( 123) (123) (123) (123) (123) (123)

132 ° 213 = 312 'f" 231 = 213 ° 132 .
The identity element in Sk is the identity map, {1, • • • , k} 3 ii--+ i. Trans-
positions are special permutations that just swap two different elements of
the set {1, • • • , k} while keeping the others fixed. It can be shown 1 that ev-
ery permutation can be written as a composition of transpositions. While
this decomposition is not unique, the parity of the number of factors is 2 •
That is, no matter how a permutation 1r is factored into transpositions, the
number of factors is either
• always even (and then we define sign 1r = +1), or else it is
• always odd (and then we define sign 1r = -1).
sign 1r is called the sign/signature of the permutation 1r. Clearly, for all
a,1r E Sk, we have sign(a o 1r) = (signa)(sign1r). So if {-1, 1} is considered
as a group, with the usual multiplication, then sign : Sk --+ {-1, + 1} is a
group homomorphism. Let Ak denote the alternating set, namely the the
set of all even permutations of {1, • • • , k}. Then the kernel of sign is Ak.
The inverse of a transposition is itself. So we can determine the sign of
a permutation 1r by determining the number of swaps K needed on the
second row in (10.1) to make it 1, • • • , k, and then sign1r= (-l)K.
1 See e.g. [Halmos(1987), §27, Thm. 3]. The idea is to decompose the permutation 1r
into a product of disjoint 'cycles' by operating repeatedly by 1r starting from an element,
and then writing each cycle as a product of transpositions.
2 See e.g. [Halmos(1987), §28]. The inverse of a transposition is itself. The claim then
boils down to showing that the identity permutation cannot be written as a product
of an odd number of transpositions. Each element of the set must appear in an even
number of transposition factors of the identity (to be able to return back to itself).
Form fields 211
For a k-form won V, w(v1, · · · , Vk) = (sigll7r)w(v.,,.(i), · · · , v.,,.(k)), or equiv-

alently, (sign1r)w(v1,··· ,vk) = w(v.,,.(i),··· ,V1r(k)), for all v1,··· ,vk EV.
Vice versa, a (0, k)-tensor on V having this property is seen to be an k-
form by specialising the permutations to be transpositions. The following
notation will be convenient.
Definition 10.2. (Permutation action on a (0, k)-tensor).
Let V be a vector space, k ~ l, TE TfV, and 1r E Sk. Then 1r(r) E TfV is
defined by (1r(r))(v1,··· ,vk)=r(v.,,.(i),··· ,v.,,.(k)) forallv1,··· ,vkEV.
Exercise 10.1. Suppose that Vis a vector space, k~l, TETfV, and 1r,O"E Sk.
Show that 1r(O"T) = (1r o O")T.
Any (0, k)-tensor T on V can be converted into a k-form Alt(w) by an

operation called alternation.
Definition 10.3. (Alternation).
Let V be a vector space. For T E Tf V, the alternation Alt T E /\ kV* of T is
Altr = t, ~
"-rres.
(sign1r)1rr.
Let us check that Alt T E /\kV*. Indeed, for v1, •••, Vk E V, we have
• ~~71" •
(sign 1r)(Alt r)( v.,,.(1), · · ·, V1r(k)) = ~ ~ (s1gna) r(V1r(a(l)), · · ·, V1r(a(k)))
. ues.
sign1r .
= ~ ~ (s1gna)r(v.,,. 0 u(l),···,V.,,.ou(k))
. ues.
= :, ~(sign(1roa))r(v.,,. 0 ,,.(1),···,V.,,.oa(k))
. ues.
= t, ~
. µ.es.
(signµ)r(vµ(l),···,Vµ(k))
= (Altr)(v1, · · ·, vk)-
since for eachµ E Sk, there is a unique a E Sk such thatµ= 1r o a, namely
(T = 71"-10 µ.
Example 10.1. Let V be a vector space. If w EV* =TfV, then Altw=w.

If TE T~V, then for all v, w EV, (Altr)(v, w) = -¾(r(v, w) - r(w, v)).
2.
◊
Exercise 10.2. Let V be a vector space, and T E TfV. Determine the action of
AltT on (x, y, z), where x, y, z EV, in terms of that of T.
Exercise 10.3. Let V be a vector space, and w E /\kV. Show that Altw = w.
We now list some useful properties of alternation, which will be used later.
Lemma 10.1. Let V be a vector space, w E /\kV*, 0 E A'V*. Then:

(1) Alt ((Altw) ® 0) = Alt (w ® 0).
(2) Alt (w ® (Alt0)) = Alt (w ® 0).
(3) Alt(w®0) = (-l)k£Alt(0®w).
Proof.
(1) We have, using the distributivity of® over addition, that
Alt((Altw)®0) = (k~f)! I: (sign1r)1r(t, I: (signa)(aw)@0).
1reSk+e ueSk
Let us consider a E Sk as an element (denoted again by a) of Sk+e, where
we imagine that a fixes k+l,··· ,k+e. Then (aw)®0=a(w®0) and the
sign of a, considered as an element of Sk+e, is the sign of a E Sk. Thus the
above gives
1
Alt((Altw)®0) = (k+f)! k! I: I: sign(1roa)(1roa)(w®0).
1rESk+e aESk
But for each a E Sk, the map Sk+e 3 1r >-+ 1r o a E Sk+e is a bijection, and
since Sk has k! elements, we obtain
Alt ((Altw) ® 0) = (k + ~)! k! k! I: (signµ) µ(w ® 0) = Alt (w ® 0).
µeSk+e
1 . . . e e+1 ... e+k)
(3) Suppose thatµ denotes the permutation ( k+l ... k+e 1 . . . k ·
To bring the 1 in the second row to the first position, we need e transposi-
tions (and then k + l, • • • k + e are shifted to the right by one space). Next,
to bring the 2 in the second row to the second position, we again need e
transpositions, and so on. Thus, starting fromµ, in order to move 1, • • • , k
in the second row to occupy the first k places (and in the process shift
the k + l, • • • , k + e rightwards to occupy the last e places), the number of
transpositions needed is e + ••• +e (k times) = kf, and then the resulting
new permutation sends each i to itself (i.e., it is the identity map). Hence
signµ=(-l)ke. Now for v1,···,Vk+e EV, we have
(Alt (w ® 0))(v1, · · ·, Vk+e)
= (k~f)! ~ (sign 1r)w(v1r(l), · · ·, V,r(k)) 0(v1r(k+l), ... , V,r(k+e))

n:eSk+£
= (k~f)! ~ (sign 1r) w( V,r(µ(e+l)), · · ·, V,r(µ(e+k))) 0( V,r(µ(l)), ---, V,r(µ(f)))

n:eSk+£
(signµ) .
(k+f)! ~ s1gn(1roµ)0(v(1roµ)(l),···,V(1roµ)(f))·
,reS•+' w(v (1roµ)(Hl), · · ·, v (1roµ)(f+k) ) ·
Form fields 213
The map Sk+e 3 1r >-+ 1r o µ is a bijection, and so we obtain
Alt (w © 0)(v1, · · · , Vk+e)

signµ .
= (k+f)! ~ (s1gna) 0(va(I), · · · , Va(£))w(va(£+I), · · · , Va(l+k))
ueSk+l
= (-ll£Alt(0©w)(v1,··· ,Vk+e)-
(2): Alt (w © (Alt0)) ~ (-1)k£ Alt ((Alt0) © w) ~ (-l)k£ Alt (0 © w)

~ Alt(w©0). □
The tensor product of forms is a tensor, but it may fail to be a form,
since it may not be alternating. To get an alternating form, we apply
the alternation operation. The resulting operation is called the exterior
product.
Definition 10.4. (Exterior product of forms).

Suppose that V is a vector space. Let the integers k, f ~ 1. The exterior
product" : /\kV* x /\£V* --+ /\k+eV* is defined as follows: For w E /\kV*
and 0 E /\£V*,
(k+f)!
w " 0 = ~ Alt (w ® 0).
For a, b E JR= /\0 v*, we set a" b = ab. For a E JR=/\0 v* and w E /\kV*,
we define aAw=wAa:=aw.
More explicitly, for v1, · · · , Vk+e E V,
(wA0)(v1,··· ,Vk+e) = k~f! ~ (sign1r)(w©0)(v1r(l),··· ,V1r(k+e))-

1rESk+£
It is straightforward to verify that for a E JR, w E /\ kV* and 0 E /\ £V*,
a(wA0) = (aw)A0 = wA(a0).

Also, for w,w E /\kV* and 0 E /\£V*, we have
We summarise by saying that the wedge product is JR-bilinear. We now

show that the wedge product is associative.
Proposition 10.1. Let V be a vector space, and wi E /\k;V*, i = 1,2,3.

Then w1 " (w 2 A w3 ) = (w 1 A w2 ) A w3 •
In light of the associativity of A, there is no ambiguity in using the notation

w l " • • • " w n , where w i E j\ki V* , i. = 1, • • • , n.
Lemma 10.2. Let V be a vector space, and wi E /\ ki V*, i = 1, • • • , n.

Then
l n (k1+···+kn)! l n
w A··· AW = -----Alt(w ®···®w ).
k1! ···kn!
Proof. We give an inductive argument on the number of factors in the
wedge product. For n = 1 there is nothing to prove since Altw 1 = w1, and
for n = 2, this is just the definition of the wedge product. The case n = 3
was shown in the proof of Proposition 10.1. Suppose that n > 3, and that
the result has been shown when the number of factors is ::;;; n - 1. Then
□
Form fields 215
Exercise 10.4. Let V be a vector space and €i E /\ 1V* = V*, i = 1, • • • , n.

Show that for all Vi EV, i=l,· ·· ,n, (€ 1 /\ ··· /\ €n)(v1,··· ,vn) = det[€i(vi)].
Exercise 10.5. Let V be a vector space, €1, • • • , €k E V*, and 7f E Sk be 4a permu-

tation of {1, • • • , k }. Show that €1r(l) /\ ... /\ €1r(k) = (sign 1r) €1 /\ ... /\ €k.
We have the following consequence of Lemma 10.1.(3).
Proposition 10.2. Let V be a vector space, w E ;\kV*, and 0 E ;\lV*.

Then w A0= (-l)k£(0 Aw).
Proof. We have
(k+f)! (k+f)! kl kl
w I\ 0 = JJ:e!Alt (w ® 0) = JJ:e!(-1) Alt (0 ® w) = (-1) 0 AW. □
Proposition 10.3. Let V be an m-dimensional vector space, with a basis

{e1,··· ,em}, and let {€1,··· ,€m} be the corresponding dual basis for the
dual space V*. Then
B = {€i 1 A• • • /\ €ik : 1,:;;; i1 < · · · < ik,,;; m}
is a basis for ;\kV*, and dim!\kV* = (;)-
Proof. It is clear that the elements in B are in one-to-one correspon-

dence with all k-tuples (i 1 , • • • , ik) chosen from {1, • • • , m} that are strictly
increasing, which is just the number of ways of choosing k distinct numbers
from {1, • • • , m} (since once a choice is made, they can then be arranged in
an increasing order). Thus B has (';;) elements, and once we show it is a
basis, the claim on the dimension follows.
spanB=AkV*: Let w E !\kV* c TfV, and so
As w is alternating, we have
w = Altw = w( ei 1 , • • • , eik) Alt (€i 1 ® • • • ® ik)
1 i1
= w (ei 1 , • · • , eik
) k! € /\ • · • /\ €
ik
.
Now it can be seen, using Proposition 10.2, that the summands with some
ir equal to some is, with r-# s, will be zero. In a summand where the
i1, · · · , ik are all distinct, we can write €i 1 /\ • . . A f.ik = ( -1 lf.ii /\ ... /\ f.ik
for some e E N, with j1, · · · , Jk a permutation of ii,··· , ik and such that
J1 < · · · <Jk· This shows that B spans;\ kV*.
l
Bis independent: Suppose that we have c:; 1 ··•ik 1:i 1 A· · ·A 1:ik = 0. Then acting
both sides on the tuple (ej 1 , • · • , eik), with 1 ~j1 < · · · <jk ~m, yields
l
Ei1 ( ei1) ... Ei1 ( eik)
0=ci 1 .. ,ikfi 1 A .. ·Afik(eju"•,ejk)=ci1 ... ikdet [ . : · ... :

6i1 ... 6i1 E'k ( ej1 ) .. • E'k (eik)
Jl Jk
[
= Ci 1 .. ,ik det : ·.. : .
Dik ... Dik
..._______,__.,
Jl Jk
=:t..5
We give an inductive argument (on k) to show that the determinant det b. 5
in a summand is nonzero only if J := (ii, ... ,ik) = (j1, ... ,jk) =: J. If
5
k = l, then clearly det b. =I= 0 implies I= J. Suppose that the claim is true
for some k. Then for a matrix of size k+l, the top row must be nonzero,
so that ii= JR. for some£. Expanding the determinant along the top row,
0 =I= det b.5 = (-1)£ det b.5',, where I', J' are defined by J' := (i2, · · · , ik+1),
J' := (j1, · · · ,Jt-1,JR.+1, · · · ,Jk+1)- Thus I'= J' by the induction hypoth-
esis. Now if £ > 1, then we get the contradiction that ii = JR. > j1 = i2.
So i 1 = j 1, and then together with J' = J', we obtain J = J. Moreover, if
I= J, then clearly b.5 is the identity matrix, and det b.5 = l. So we obtain
0 = Ci 1 , .. ik 1:i1 A· · · A1:ik ( eiu · · · , eik) = Cfr .. ik. Thus B is independent. □
10.2 k-form fields
A k-form field is just a (0, k)-tensor field non a smooth manifold M whose
pointwise evaluations at each point p EM is a k-form n(p) E /\k(TpM)*
on TpM. But then for all p EM, all permutations 71' of {1, • • • , k} and all
vector fields V1, · · · , vk E TJ M,
(sigll'lr)(n(V,,.(1), · · · , V,r(k)))(p) = (sigll'lr)n(p)((V,,.(1))p, · · · , W1r(k))p)
= i1(p)((Vi)p, · · · , (Vk)p)
= (!1(Vi, .. · , Vk))(p),
and so we can equivalently define a k-form field as follows.
Definition 10.5. (k-form field).
Let M be a smooth manifold. For 71' E Sk and TE T2M, define 7rT E T2M
by (7rT)(V1, .. · , Vk) =T(V1r(l), .. · , V,r(k)) for all Vi, .. · , vk E TJM.
Let k ;;,:: 2. A k-form field is a (0, k)-tensor field n E T2M which is
alternating, that is, for all permutations 71' E sk' n = (sign 71') 71' n.
Form fields 217
The set of all k-form fields on M is denoted by n,k M. We set 0 1 M = Tf M

and n,o M = C 00 (M). If OE n,k M, then k is called the degree of 0.
n,k Mis a C 00 (M)-module with the usual operations for tensor fields. Recall
that for a vector space V, I\ kV* = {0} for k > dim V. If M is m-dimensional
(so dimTpM =m for all p EM), then n,k M = {O} for all k>m.
Exercise 10.6. (Lie differentiation of a form field gives a form field). Let M be a
smooth manifold, and VE TJ M. Show that if n E nk M, then .Cvn E nk M.
Alternation can be applied to the pointwise evaluation of any (0, k)-tensor

field T to obtain a k-form field: (AltT)(p) = Alt (T(p)), p EM. By Propo-
sition 4.1, AltT E T2M, and since for each p EM, Alt (T(p)) E /\k(TpM)*,
we have AltT E O(M). Or equivalently:
Definition 10.6. (Alternation).
Let M be a smooth manifold and T E T2 M. Then the alternation of T is
the k-form field AltT E O(M) defined by
AltT = :, ~ (sigll1r)1rT.
0
1rESk
Exercise 10.7. Suppose that Mis a smooth manifold, f E C 00 (M), and TE Tf M.

Show that Alt (f T) = f Alt T.
We have the following analogue of Lemma 10.1, which can be obtained as

a consequence, or directly in the same manner.
Lemma 10.3. Let M be a smooth manifold, 0 E n,kM, 8 E n£M. Then:
(1) Alt((Alt0)©0) = Alt(0©0).
(2) Alt(O®(Alt0)) = Alt(0©0).
(3) Alt(0®8) = (-l)k£Alt(0®0).
Proof.
(1) We have for all p EM that
(Alt ((AltO) © e))(p) = Alt (((AltO) © e)(p))
= Alt (((AltO)(p)) ® 8(p))
= Alt ((Alt (O(p))) ® 8(p)) = Alt (O(p) © 0(p))
= Alt((0®8)(p)) = (Alt(0©8))(p).
So Alt (Alt (0) ® 8) = Alt (0 ® 8).
(2) Analogous to (1).

(3) For all p EM,
(Alt (0 © e))(p) = Alt ((0 © e)(p)) = Alt (O(p) © 8(p))
= (-1t£ Alt (0(p) © O(p)) = (-l)H Alt ((8 © O)(p))
= ((-1t£Alt(0©0))(p).
Thus Alt(0©0) = (-1t£Alt(0©0). □
The tensor product of form fields is a tensor field, but it may fail to be a
form field since it may not be alternating. To get an alternating form field,
we apply the alternation operation. The resulting operation is called the
exterior product.
Definition 10.7. (Exterior product).
Let M be a smooth manifold, and let the integers k, f, ~ 1. The exterior
product I\ : {lk M X n£ M --+ {lk+l M is defined as follows: For O E {lk M
and 8 E ntM,
(k+f)!
o" 0 = ~Alt(o ® 8).
For J,g E C 00 (M) = 0°(M), J "9 := Jg. For J E C 00 (M) = 0°(M) and
o E nk(M), J" o = o "J = Jo.
Clearly, for all p EM, we have (0" 0)(p) = (O(p))" (8(p)) E /\kH(TpM)*.
More explicitly, for V1, · · · , vk+l E TJ- M,
(0 A 8)(V1, ... 'vk+£) = k~f! ~ (sigll'lr) (0 © 8)(V.,,.(1), ... 'v.,,.(k+£))-

1rESk+£
Example 10.2. Let M be a smooth manifold and 0, 8 E Tf M. Then for

V,WETJ-M,
(0 " 8)(V, W) = (0 ® 8)(V, W) - (0 ® 8)(W, V)
= O(V)8(W) - e(v)o(w) = (o ® 0 - 0 ® O)(V, w).
Thus O A 8 = 0 ® 8 - 8 ® 0. ◊
It is straightforward to verify that for J E 0 00 ( M), 0 E nk M and 8 E nt M,

J(O " 0) = (JO) " 8 = 0 " (/8).
Also, for 0, nE {lk M and 8 E n£ M' we have
(0 + n) A e = 0 A 8 + n A e and e A (0 + n) = 8 A O+ 8 A n.
Thus the wedge product is bilinear over C 00 (M). We now show that the
wedge product is associative.
Form fields 219
Proposition 10.4. Suppose that M is a smooth manifold, and ni E {lk; M,

i = 1,2,3. Then i1 1A(i1 2Ai1 3) = (i1 1Ai1 2)Ai1 3.
Proof. For all p E M we have

(n1 /\ (n2 /\ n3))(p) = n1 (p) /\ ( (n2 /\ n3)(p)) = n1 (p) /\ (n2(p) /\ n3(p))
= (n1 (p) I\ n2(p)) I\ n3(p) = ( (n1 /\ n2)(p)) /\ n3(p)
= ((n1 /\ n2) /\ n3)(p).
Thus n 1/\ (!1 2 /\ !1 3 ) = (!1 1 /\ !1 2 ) /\ n3 . □
As I\ is associative, there is no ambiguity in writing !11 /\ • • • /\ nn for
ni E {lkiM, i=l,··· ,n.
Exercise 10.8. Let M be a smooth manifold and let ni E {lk; M for i = 1, • • • , n.

Show that n 1 A··· Ann= (ki + ... +kn)! Alt (!1 1 (8) ... (8) nn).
k1! •··kn!
An application of this exercise is the following result on the wedge product
of I-form fields.
Proposition 10.5. Let M be a smooth manifold and ni E 0 1 M = Tf M,

i = 1, · · · , n. Then for all ½ E TJ M, i = 1, · · · , n, we have
(!1 1 /\ · · · /\ nn)(Vi, · · · , Vn) = det[i1\Vj)].
Proof. We have
(!1 1 /\ · · · /\ nn)(V1, · · ·, Vn) = n!Alt(n 1 (8) · · -®nn)(Vi, · · ·, Vn)
= ~ (sigll7r)(i1 1® · · · (8) nn)(V,r(l), · · · , V,r(n))

1rESn
= ~ (sigll7r)i1 1(V1r(1))···!1n(V7r(n))
1rESn
= det[n\½)].
□
We have the following consequence of Lemma 10.3.(3).
Proposition 10.6. Let M be a smooth manifold, n E {lk M and 8 E n£ M.
Then n /\ 8 = (-l)k£(8 /\ !1).
Proof. We have
(k+f)! (k+f)! kl kl
n A8=/Je!Alt(i1®0)=/Je!(-1) Alt(8®i1)=(-1) e/\n. □
Exercise 10.9. Suppose that M is a smooth manifold, k is odd, and nE {lk M.

Show that n A !1=0.
Recall that in an admissible chart (U, x) for a smooth manifold M, we

can decompose a given T E Tf M as T = Ti 1.. -ik dxi 1® · · · ® dxik , for some
functions Ti1•••ik E C 00 (U). Now if n E nk M, then we have in particular
n = ni1•••ikdxi1®"" "Q9 dxik, for some functions ni1 •·•ik E C 00 (U), but
n= Alt n = ni1 ·••ikAlt (dxi 1Q9 " " • Q9 dxik) = nik;•ik dxi 1 I\ • • • I\ dxik.
Using Proposition 10.6, we can rearrange the factors of the wedge product
within a summand, and combine terms to obtain
for some functions Wii•••ik E C 00 (U). The functions Wii•·•ik are actually
ni 1 ··•ik. Indeed, recall first the definition of ni 1··•ik and note that a re-
arrangement of indices produces the sign of the permutation used, as n
is alternating. The same factor is picked while rearranging the indices in
dxi 1I\ ... I\ dxik. Finally, each term in the rearranged sum collects k! terms
in the original one. In particular, if M is m-dimensional, then every m-
form field n E nm M can be decomposed in an admissible chart (U, x) as
n = W dx 1 I\""" I\ dxm, where W = n(ox1,""" , Ox-m) E C 00 (U).
Exercise 10.10. Consider the smooth manifold M =R 3 with the standard smooth
structure, and the global admissible chart (R3 , (x, y, z) >-+ (x, y, z)). Suppose that
n := xdx+ydy+zdx E 0 1 M, 8 := ydx+zdy+xdz E 0 1 M. Determine rlA 8.
Exercise 10.11. Let M be a smooth manifold. In an admissible chart (U, x),
suppose that fl := j dxi 1 I\••• I\ dxik, for a function j E C 00 (U) and a fixed
k-tuple I= (i1, · · · ,ik), where l ,;;;;i1 < · · · <ik ,;;;;m. Show that if J = (j1, · · · ,jk),
with l ,;;;;j1 < · · · <jk ,;;;;m, then
J = I,
flil···jk =
{ Qf if
if J =f. /.
Pull-back
We had seen that if / : M -+ N is a smooth map between smooth manifolds
M and N, then a (0, k)-tensor field TE Tf N can be pulled back under f
to a (0, k)-tensor field f*T on M. If n E Ok N, then f*!1 E Ok M. Indeed,
for all Vi, ... ' vk E TJ M, any permutation 7r E sk, and all p EM, we have
((!*!1)(V1r(l), · · · , V1r(k)))(p) = !1(f(p))(dfp((V1r(1))p), · · · , dfp((V1r(k))p))
= (sign1r)n(f(p))(d/p((Vi)p),· .. ,dfp((Vk)p))
= (sign1r)((f*!1)(Vi, · · · , Vk))(p),
and so (f*i1)(V1r(l), · · ·, V1r(k)) = (sign1r) (!*!1)(Vi, · · · , Vk)- Consequently,
f*!1 E f2kM.
Form fields 221
Exercise 10.12. Let M, N be smooth manifolds, and f: M-+ N be a smooth map.

Let !1 E o,k N and 8 E gt N. Prove that f*(!l A 8) = (/*!1) A (/*8).
Exercise 10.13. Let N = IR.3 with the standard smooth structure, and M be the
open subset of N given by M = (0, oo) x (0, 1r) x (0, 21r) with the induced smooth
structure. In the admissible chart (V = N, (x, y, z) >-+ (x, y, z)) for N, consider
the 3--form field n = dx Ady A dz. Let f : M -+ N be the smooth map given
by f(r,0,cf>)=(r(sin0)(coscp),r(sin0)(sincp),rcos0), for (r,0,¢>) EM. Determine
the pull-back f*!l E 0. 3 M of n.
10.3 Exterior derivative
For a 0-form on a smooth manifold M, that is, for a smooth function

f E C 00 (M), we had defined the gradient df off as a 1-form field. Thus
d : n° M -+ 0 1 M. We now give an extension of this operation, which is
called the 'exterior derivative'.
Definition 10.8. (Exterior derivative of a k-form field).

Let M be a smooth manifold and let n E {lk M beak-form field.
The exterior derivative dO E f2k+ 1 M of n is defined by
,.... Oxini1•••ik dXI\
d H=---=----..:;.. i dXi1 /\•••/\ d xk=
i "
L..J r,
UxiHit"'ik
:l dXI\
i dX i 1 / \ • • • / \ dxk
i
k! 1 ~i1
· <··•<ik~m
·
in any admissible chart (U, x) for M.
We check this notion is well-defined. Firstly, the above gives a (k+l)-form

field in any admissible chart (U, x). So we only need to check that there is no
conflict of definition in overlapping charts. Then by Proposition 4.1 (p.67),
d n is a (0, k)-tensor field. Moreover, the alternating property holds ev-
erywhere, as it holds in every chart. Let (U, x) and (V, y) be overlapping
admissible charts for M. Then n = tin(oyit, • • • , Oyik) dyii" • • • "dyi\ and
Oyi 0( Oyit, · · · , Oyik) dyj " dyii " · · · " dyik
= (OyiXi)OxiO(oyitXi 10xi1,··· ,OyikXikOxik) ·
Oxtyidxe I\ Oxt1yi1dxe 1 I\··· I\ Oxtkyikdxek
= (Oyi xi)(OxlYj)Oxi((oyit Xi 1 ) •.• (Oyik xik) O(Oxit' . .. 'Oxik)) .
(oxt1Yj 1) """ (oxtkyik)dxe I\ dxR, 1 I\••• I\ dxek
= Oxi((oyitXi 1)···(oyikXik)fi(oxi1,··· ,Oxik))"
(oxt1Yj 1) """ (oxtkyik)dxi I\ dxR, 1 I\••• I\ dxek
= Oxi (O(ox; 1, · · · , Oxik)) dxi I\ dxi 1 I\ · · · I\ dxik + O(ox; 1, · · · , Oxik) · S,
where the product rule gives the last equality, and S is given by
k
s = Li (oxer yir )(Oxi OyirXir) dxi I\ dxi 1I\••• I\ dxir-l I\ dxlr I\ dxir+l I\•• • I\ dxik,
r=l
which we will now show is 0. We have

(Oxfr yir) (Oxi Oyir Xir) = Oxi ( (Oxfr yir) (Oyir Xir)) - (Oxi Oxfr yir) (Oyir Xir)
The first summand is Oxi((oxeryir)(oyirXir)) = 0xi(5t = 0. So
k
s= - Li (Oyir xir) (Oxi Oxfr yir) dxi I\ dxi 1I\• • • I\ dxir-l I\ dxlr I\ dxir+l I\• • • I\ dxik_
r=l
A swapping i +-+ er in dxi I\ dxi 1 I\ • • • I\ dxir-l I\ dxlr I\ dxir+l I\ • • • I\ dxik

yields an overall sign (-lt(-1t- 1 = -1, but the term OxiOxeryir does not
change under such a swapping of these dummy indices. Thus S = 0. This
shows the well-definition.
The map d : Ok M -+ f2k+ 1 M is a linear transformation, considering
Ok M, f2k+ 1 M as vector spaces.
Exercise 10.14. Let M be a smooth manifold and (U, x) be an admissible chart.

Let f E C 00 (U), and (i1, ... 'ik) be a fixed k-tuple from {1, ... , m} with distinct
entries. Let n = fdxi 1 A ••• A dxik. Show that dn = (o,,;f) dxi A dxi 1 A ••• A dxik.
By the linearity of d, it follows by the above exercise that if n E Ok M has

a decomposition in a chart given by
n- - . dxi 1 I\
w·'l.1·••'1,k • • • I\ dxik ,
for some functions Wi1•••ik E C 00 (U), then

dn = (OxiWi1 ---ik) dxi I\ dxi 1 I\ • • • I\ dxik.
The exterior derivative map also satisfies a type of Leibniz rule with respect
to the wedge product (Proposition 10. 7), and justifies calling it a derivative.
Notation 10.1. Let (U, x) be an admissible chart in a smooth manifold

M, and n E Ok M. For a k-tuple (i 1 , • • • , ik) from {1, • • • , m}, we introduce
the abbreviations J = (i 1 , • • • , ik), dx 1 = dxi 1" • • • "dxik, and
r, ni1•·•ik
~ 'I = ~ .
Then i1 = !11dx 1 .
Proposition 10. 7. Suppose that M is a smooth manifold. Let nE Ok M

and 8 E n£M. Then d(n"e) = (d!1)"8 + (-l)ki1"d8.
Form fields 223
Proof. In any admissible chart (U, x), we have

d(i1A0) = d(!11dx 1A0JdxJ) = d(!110JdX 1AdxJ)
= Oxi(n1eJ)dxiAdx 1AdxJ
. I J
= ((Oxin1)8J + !110xieJ)dx'Adx Adx
. I J k I . J
= (Oxin1)dx'Adx A0JdX + (-1) !11dx /\Oxi0Jdx'Adx
= (dn)Ae + (-1tn/\de,
where the second summand acquires the factor (-ll since dxi has to pass
through the k-form dx 1 =dxi 1 A ... Adxik_ □
In particular, we have d(fn) = d(f Ai1) = (dj)An + f Adi1.

Exercise 10.15. Consider the 1-form fields n and 8 from Exercise 10.10 again.
Show that dn = 0, and d8 = -(dxAdy + dyAdz + dzAdx).
In the previous exercise, we note that for some form fields the exterior
derivative vanishes, while it does not for others. This prompts the first of
following definitions.
Definition 10.9. (Closed and exact form fields).

Let M be a smooth manifold. A k-form field n E {lk M is called closed if
dn = 0. If k ~ l, then a k-form field n E nk M is called exact if there exists
a e E nk-lM such that i1=d0.
We now show d2 = d o d = 0. This implies every exact form field is closed,

since for an exact form field n, we have i1=d0, and so di1=d(d0)=0.
Proposition 10.8. Suppose that M is a smooth manifold, and n E {lk M.

Then d(dn) =0.
Proof. Let (U, x) be any admissible chart for M. Using Notation 10.1,
d(dn) = d((Oxin1)dxiAdx 1 )
. . I
= (OxiOxinJ)dx 3 Adx'Adx
. . I
= (OxiOxin1)dx 3 Adx'Adx
. . I
= -(OxiOxin1)dx'Adx3 Adx
= -d(dn),
where we used [ox;,Oxi]=O, and dxiAdxi = -dxiAdxi. So d(di1)=0. □
Exact form fields are closed, but the converse may not always hold.
Example 10.3. (Closed ~ exact).

Let R 2 be equipped with the standard smooth structure, and consider the
open set M := R 2 \{(0,0)} with the induced smooth structure from R 2 .
In the global admissible chart (R 2 \{(0,0)},(x,y) ...... (x,y)) for M, let the
1-form field n E Tf M be given by
-y X
n= X
2
+y
2 dx +
X
2
+y
2 dy.
Then n is closed, since

-y X
dO = oy-2-
X
- dy "dx + ox-2- -
+y 2 +y2
dx "dy
X
y2-x2 y2-x2
( X 2 +y 2 ) 2 dy" dx + (X 2 +y2 ) 2 dx" dy = 0.
However, we now argue that n is not exact. Suppose, on the contrary, that
there exists an f E C 00 (M), such that O=df =oxfdx + oydy. Then
0f = ___::J!_ and of = _x_

ox x2 +y 2 oy x2 +y 2 .
Consider the smooth curve [0, 2n] 3 t .S (cost, sint) EM. We have
0 = J((l, 0)) - J((l, 0)) = J(C(2n)) - J(C(0))
= f71" 1/(C(t))dt = f71" !J(cost,sint)dt
= f 71" (!~(cost, sin t)(- sin t) + :~(cost, sin t)(cos t) )dt
= 171" (
0
2 (- sin t)
(cost) 2 +(sint) 2
. (cost)
- - - - - - ( - s m t ) +------(cost) dt
(cost)2 + (sint) 2
~
= f71" ldt = 2n,
a contradiction. So n is not exact. ◊
If (U,x) is an admissible chart for a smooth manifold M, and n is a k-

form field of the type n = dxi 1 I\ • • • I\ dxik E nku, then it follows from
Propositions 10.7 and 10.8 that d0=0.
Exercise 10.16. Consider the smooth manifold N =R 2 with the standard smooth
structure. Show that every closed 1-form field n E 0 1 Mis exact.
Form fields 225
Remark 10.1. (de Rham cohomology).

The failure of the converse (closed~ exact) is linked to the global topology
of the manifold M. (For instance, in Example 10.3, M = R 2 \{ (0, 0)} has a
'hole' at the origin, while in Exercise 10.16, N =li2 does not.) We have the
following sequence of vector spaces and maps
0 ...3:...+ n° M ...3:...+ n1 M ...3:...+ n2 M ...3:...+ · · · ...3:...+ nm M ...3:...+ 0.
(Note that the first map on the left is just the zero map, and we have
set n- 1 M = {O}.) Also, this explains the terminology 'exact' from Defini-
tion 10.9: recall that a sequence of linear maps
A L
B __!!_. C
of vector spaces is exact at B if ran f = ker g. In order to specify which
map d we mean, we use dk for the map d: nk M-+ nk+ 1 M. Then d2 = 0
implies that ran dk-l c ker dk. The quotient vector space
HJR := (ker dk)/(randk-l)
is called the k de Rham cohomology vector space HJR of M. The asso-
th
ciation of these algebraic objects (vector spaces) to a manifold gives rise

to a dictionary for translating topological properties of M into algebraic
properties of the de Rham cohomological vector spaces. For example,
H~R = (kerd:C 00 (M)-+ n 1 M)/{0}
~ (kerd:C 00 (M)-+ n 1 M) = {J E C 00 (M): df=0}.
But if M is connected, then it can be shown that df = 0 implies that f must
be constant on M, and in general, {J E C 00 ( M) : df = 0} is the collection of
all functions that are constant on each connected component of M. Thus
H~R ~ Rbo(M), where b0 (M) is the number of connected components of M.
(The integer b0 (M) is called the 0th Betti number of M; see for instance
[Bishop and Goldberg(1980), p.186].) *
Vector calculus in R 3
In multivariable calculus, a vector-valued function V on R 3 is a smooth map
V : R 3 -+ R 3 that associates to each point p E R 3 an element V (p) E R3 .
Considering R 3 as a smooth manifold equipped with the standard smooth
structure, we have R 3 ~ TpR 3 using the global chart-induced basis. So a
vector-valued function in the multivariable-calculus-sense can be considered
a vector field in the manifold language:
v~ [E] - v~v"a,+v>a,+v•a, cTJJR'.

We also know that we can create a vector field from a smooth function
f : R 3 -+ R by taking its 'gradient'
of
ox
-+ of
v' f :=
oy
of
oz
We note that the exterior derivative off En° M is the I-form field
of of of
df = ox dx + oy dy + oz dz,
which has the same components as V f.
Let V be a vector-valued function with the component functions given
by (Vx, VY, vz). We also recall from multivariable calculus that the dif-
ferential operator V x, the curl, acts on V to produce the vector-valued
function
ovz oVY
-----
oy oz
ovx ovz
-----
oz ox
oVY ovx
-----
ox oy
If we consider the 1-form field
0:= vxdx + VYdy + vzdz
using the components of V, then its exterior derivative d n is the 2-form
field given by
dO = d(Vxdx+ VYdy+ Vzdz)
ovx ovx oVY oVY
= ajidy A dx + Tzdz A dx + ox dx Ady + Tzdz Ady
ovz ovz
+ ox dxAdz + oy dyAdz
ovz oVY ovx ovz oVY ovx
= (----)dyAdz+ (----)dzAdx+ (-----)dxAdy,
~ ~ ~ fu fu ~
which has the same components as V x V. We recall that the divergence
map V• in multivariable calculus acts on a vector-valued function V, and
produces a smooth function,
-+ -+ ovx oVY ovz
v'·V=-+-+-.
ox oy oz
Form fields 227
n = vxdy I\ dz+ VY dz I\ dx + vzdx I\ dy

If we consider the 2-form field
constructed using the components of V, then the exterior derivative dn is
the 3-form field given by
dn = d(VxdyAdz + VydzAdx + VzdxAdy)
a~ o½ o½
= ox dx I\ dy I\ dz + oy dy I\ dz I\ dx + 0 z dz I\ dx I\ dy
= (V·V)dxAdyAdz.
We know that C 00 (M) = n° M by definition, and since all 3-form fields n
on M = JR 3 can be decomposed as n = f dx I\ dy I\ dz for some function
f E C 00 (M), we can identify S1 3 M with C 00 (M) too. Furthermore, 1-form
fields n admit a decomposition n = f dx + g dy + h dz, and 2-form fields
admit a decomposition n = f dy I\ dz+ g dz I\ dx + h dx I\ dy, for some
functions f, g, h E C 00 (M), and so they can be identified with vector-valued
functions with the three component functions given by (J, g, h). Thus if
V(JR3 ) denotes the C 00 (JR3 )-module of all vector-valued functions on JR 3
(with the module operations defined component-wise in the obvious way),
then both of S1 1 JR 3 and S1 2 JR3 are isomorphic as C 00 (JR3 )-modules to V(JR3 )
under the identifications described above. Hence we obtain the following
commutative diagram.
The familiar facts from multivariable calculus saying that

• the 'curl of the gradient vanishes', V x (VJ)= 0 for smooth functions f
• the 'divergence of the curl vanishes', V •(V x V) = 0 for vector-valued
functions V,
are just manifestations of the property do d = 0 for the exterior derivative.
Pull-back
Pulling back commutes with exterior differentiation.
Proposition 10.9. Suppose M, N are smooth manifolds, and f : M --> N

is a smooth map. Ifn E nkN, then f*(dn) = d(J*n).
Proof. We work locally. For a point p EM, let (U,x) be an admissible

chart for M containing p, and (V, y) be an admissible chart for N containing
f(p) such that f(U) c V. Using Notation 10.1, we write
Then it follows from Exercises 10.12, 3.17 (p.53) and 3.18 that
J*D.=J*(D.1dyi1/\ . .. I\ dyik)=(D.10 J)d(yi10 j)A ... Ad(yiko J).
Using d2 = 0, it follows by the Leibniz rule (Proposition 10. 7) that
(10.2)
On the other hand,
dn = d(n1dyi 1 I\ ... I\ dyik) = dn1 I\ dyi 1 I\ ... I\ dyik,
and so by Exercise 10.12,
f*(dn) = J*(d0. 1 /\ dyi 1 I\ ... I\ dyik)

= f*(dD.1) I\ f*(dyi1) /\ ... /\ f*(dyik)
= d(D.10 J) I\ d(yi 10 j) I\ · · · I\ d(yik OJ) <1~ 2 ) d(J*D.).
An intrinsic expression for the exterior derivative

We had defined the exterior derivative by using charts, and checked that the
choice of chart did not matter. We now give an explicit intrinsic definition
for the exterior derivative. This alternative definition is given in the form
of a formula, and this is the content of the following result.
Theorem 10.1. Suppose that Mis a smooth manifold, and n E {lk M.

For all Vi, ... ' vk, vk+l,
k+l ~
(dD.)(Vi, · · ·, Vk+1) = ~ (-l)i-l ¼(0.(Vi, · · ·, ¼, · · ·, Vk+1))

i=l
·+· - -
+ ~ (-1)' 30.([¼,l7i],V1,···,¼,···,VJ,··•,Vk+1),
l~i<j~k+l
where-:- indicates an omission of the corresponding argument.

Form fields 229
Proof. Let us call the right-hand side 8(V1, · · · , Vk+1), and the two
summands as 81(Vi, · · · , Vk+1) and 82(V1, · · · , Vk+1)- We note that
8 : TJ-M x • • • x TJ-M -> C 00 (M) is R-multilinear. We will show that
it is in fact also C 00 (M)-multilinear. For 1:::;;£:::;;k+l, and f E C 00 (M),
81(Vi, ... '!½, ... 'Vk+1)
= (-1l- 1(J½,)(O(V1,··· ,½,··· ,Vk+1))
+ ~ (-l)i- 1V;(O(V1,··· ,f½,··· ,¼,··· ,Vk+1))
i<i~k+l
+ ~ (-l)i- 1V;(O(V1,··· ,¼,··· ,f½,··· ,Vk+1))

l~i<i
= f81(V1,··· ,½,··· ,Vk+1)+A+B,
where
A:= ~ (-l)i- 1(V;J)O(Vi,··· ,½,··· ,¼,··· ,Vk+1)
l<i~k+l
~ (-1/-1H- 1 (V;J)O(½, V1, · · · , ½, · ·· ,¼, · · ·, Vk+1)

l<i~k+l
·+t ~ -
~ (-1)3 ("Vjf)O(½,V1,··· ,½,··· ,VJ,··· ,Vk+1),
l<j,s;;k+l
(the dummy i was replaced by j to obtain the last line), and

B := ~ (-l)i- 1(V;J)O(Vi, · · ·,¼,···,Vi_,···, Vk+1)
l~i<i
i-1+£-2 ~ ~
= ~ (-1) (V;J)O(Vi,,Vi,··· ,¼,··· ,½,··· ,Vk+1)
Similarly, to compute 82(Vi, •••, J½, •••, Vk+1), we split the sum into the
parts when i=f, when j=f, and the rest (and the f factors out from this
last part). When i = e or j = e, we use [!½, ½] = ![½, ½] - (½!)½ and
[¼,J½]=f[¼, ½]+(¼!)½,. Thus
Hence the A, B terms cancel in the sum 8 = 8 1 + 8 2 , showing that 8 is

C 00 (M)-linear in the £th slot, for all 1:::;;£:::;;k+l. Now that both dO and
0 are (0, k+ 1)-tensor fields, it is enough to check that in any admissible

chart (U, x), their action on (k + 1)-tuples of chart-induced vector fields
Oxi coincide. We note that the Lie bracket [Oxi, Oxi] is zero, and so the
0 2 summand disappears. We can decompose n = !11dx1. Since both sides
depend additively on n, it is enough to check that both sides coincide for an
n of the form fdx 1 with a fixed I= (i 1 , • • •, ik), and a smooth f E C 00 (M).
l
Then di1=ox;f dxiAdx 1. So we have
(dn)(oxii, """, Oxik+1) = (ox;J)(dxi I\ dxi 1 I\ """ I\ dxik )(oxii," "", Oxik+l)
DiJl ·--6~Jk+l
6i1 ... 6~1
= (Oxi J) det [ ~1 J~+i .
6;~ ··· 6;:+l

Looking at the top row, we see that the result will be zero unless i equals
some it- So only such terms (in the sum over i) will survive, giving (by
expansion along the top row where all the entries are zeroes except for the
1 which appears in the £th place if i = it):
where the second equality follows by the Leibniz rule and by noting that
we have dx 1(Dxii, · · ·, oxi•+i) = a constant. D
So if n E S1 1 (M), then
(dn)(X, Y) =X(n(Y)) - Y(n(X)) - n([X, Y]),
for all X, YE TJM.
10.4 Interior multiplication
While the exterior derivative increases the degree of the form field, we now
learn about the operation of interior multiplication by a vector field V which
decreases the degree. Sometimes interior multiplication by V is referred to
as the interior derivative with respect to V, because it also satisfies a type
of Leibniz rule akin to the one for the exterior derivative.
Form fields 231
Definition 10.10. (Interior multiplication).

Let M be a smooth manifold, and V E TJ M. If k > 1 and n E nk M, then
the interior multiplication of n by V, denoted ivn E nk-l M, is defined by
(iv(n))(W1, · · ·, Wk-1) =i1(V, W1, · · · , Wk-1),
for all vector fields W1 , • • • , Wk-l E TJ M. If k = 1, then for a 1-form field
i1 E Tf M, we set iv(i1) =i1V E C 00 (M)=S1°M.
In particular, for any f E 0 00 ( M), iv (df) = df (V) = V f. So it is natural to

think of iv as a differentiation process.
Exercise 10.17. Suppose that M is a smooth manifold, and VE TJ M. Show that
iv: nk M-+ nk-l Mis a C 00 (M)-linear map between the C 00 (M)-modules nk M
and nk- 1 M.
An application of Proposition 10.5 yields the following local description of

the interior multiplication by V.
Lemma 10.4. Let M be a smooth manifold, and VE TJM. If (U,x) is an

admissible chart for M, then in the chart (U, x) (with".' meaning omission
of the corresponding term),
iv(dxi 1 /\ • • • Adxik)
k -
= ~(-lf- 1dxir(V)dxi 1 /\ ••• Adxir- 1 /\dxir/\dxir+l/\ ... Adxik_
r=l
Proof. For vector fields W1, · · · , Wk-1 E TJ M, by Proposition 10.5,

(iv(dxi 1 I \ · · · I\ dxik))(W1, · · · , Wk-1)
= (dxi 1 A••·/\ dxik)(V, W1, · · · , Wk-1)
= [dxh(V) dx''t,) •·· dx''(t_,)

dx'k (V) dx'k (W1 ) ••• dx'k (Wk-1)
l·
We expand the determinant down the first column, and use Proposition 10.5
again, now to convert the determinant of the submatrix obtained by deleting
the first co~mn and the rth row, into the action of the (k - 1)-form field
dxi 1 /\ • • • Adxir/\ • • • Adxik on (W1 , • • • , Wk_ 1). Thus we have
(iv(dxi 1 /\ • • • Adxik))(W1, · · ·, Wk-1)
r=l
and so the claim follows. □

We will now show the following Leibniz type of rule for iv.
Proposition 10.10. Let M be a smooth manifold, V E TJ- M, nE nk M
and e E neM. Then iv(i1A8) = (ivi1)A8 + (-1)kn/\ (iv8).
Proof. It is enough to show this locally, using an admissible chart (U, x).
Decompose f2 = !11dx 1 = !11dxi 1 I\··· I\ dxik, 8 = 8 JdXJ = 8 Jdxi 1 I\· · · I\ dxit.
Then
iv(i1A8) = !118Jiv(dxi 1 I\·•• I\ dxik I\ dxi 1 I\••• I\ dxit)
k -
= n1eJ~(-1r- 1 dxir(V)dxi 1 I\ ... Adxirl\ ... Adxikl\dXJ
r=l
f -
+ n1eJ ~ (-1)k+r- 1 dxir(V)dx 1Adxi 1 I\ . . . I\ dxirl\ ... Adxit
r=l
= (ivi1)A8+(-1tn/\(iv8).
Note that after the second equality, in the second summand the extra ( -1 )k
is accounted by the fact that the summation index runs from r = 1 to r = e,
instead of from r = k+ 1 to r = k+e. □
Just as do d = 0, we have iv o iv= 0. This is a consequence of the following

(taking V = W).
Proposition 10.11. Let M be a smooth manifold, V, WE TJ- M, k ~ 2, and
n E nk M. Then iv(iwn) = -iw(ivn).
Proof. First let k>2. For all W1, · · · , Wk-2 E TJ-M, we have
(iv(iw!1))(W1, · · · , Wk-2) = (iwi1)(V, W1, · · · , Wk-2)
= n(w, v, W1, ... , wk-2)
= -n(v, w, W1, .. •, wk-2)
= -ivi1(W, W1, · · · , Wk-2)
= - (iw(iv!1))(W1, · · · , Wk-2)
and so iv(iwi1) = -iw(ivi1). If k=2, then iwi1,ivi1 E S1 1 M, and thus
iv(iwn) = (iwn)(V) =i1(W, V) = -n(v, W) = -(ivn)(W) = -iw(ivn).
Using only the smooth structure of a smooth manifold, we have learnt two
notions of differentiation of form fields: the exterior derivative and the Lie
derivative (p.131). We will now show that although they were defined in
quite different ways, they are related via interior multiplication.
Form fields 233
Theorem 10.2. (Cartan's magic formula 3 ).

Let M be a smooth manifold, and V E TJ M. Then .Cv = do iv + iv o d.
More explicitly, for all k-form fields n E nk M, .Cvn = d(ivi1) + iv(dn).
Proof. Let n E nk M. Let V1 , • • • , Vk E TJ M be arbitrary vector fields.
Set Vo= V. Then
(iv(d!1))(V1, · · · , Vk) = (d!1)(V, V1, · · ·, Vk) = (d!1)(Vo, Vi,···, Vk)
k
= ~(-l)i½(i1(Vo,··· ,¼,··· ,Vk))
i=O
Note that in the first sum we get an extra (-1) since the i runs from i = 0
to k (instead of from i = 1 to k+ 1 as in Theorem 10.1), but in the second
sum, now as both i and j now have been reduced by 1, the factor (-l)i+i
stays the same.
(d(ivi1))(Vi, · · ·, Vk)
k
= ~(-l)i-l½((iv!1)(V1,··· ,¼,··· ,Vk))
i=l
i=l
Adding (iv(d!1))(V1, · · ·, Vk) and (d(ivi1))(Vi, · · ·, Vk), we note that only

the i = 0 term survives from both summands. Here we also use
n(Vo, [½, ½L · · ·) = -n([½, ½L Vo,···).
Consequently,
(d(ivn) + iv(dn))(V1, · · · , Vk)
= (-1) oVo(!1(V1, · · ·, Vk)) ""' (-1) o+·1 n([Vo, ½],Vi,···,½,···,
+ ~ - Vk)
O=i<j~k
k
= V(i1(Vi, · · · , Vk)) - ~ n(V1, · · · , .Cv½, · · ·, Vk) = (.Cvn)(Vi, · · ·, Vk)-
j=l
Thus d(ivi1) + iv(dn) = .Cvn. □

3 After Elie Cartan (1869-1951), French mathematician, who created the theory of form
fields.
Exercise 10.18. (Lie differentiation commutes with exterior differentiation).

Let M be a smooth manifold, and VE TJM. Show that Cvo d=do Cv.
Thus for all k-form fields n E nk M, Cv(drl) = d(Cvrl).
Hint: Use Cartan's formula and d 2 = 0.
Chapter 11
Integration
In the previous chapter, we learnt about k-form fields on smooth mani-

folds. We will learn in this chapter that these are the objects which can
be integrated over 'oriented' manifolds. If we want an intrinsic notion of
integration, we will need the notion of 'orientation'. The root cause of
this is that when performing the usual Riemann integral on R, there is a
subconscious choice made of orientation, namely
{ J(x)dx := {b f(x)dx,
J(a,b) la
where (a, b) c R is an interval, and when integrating, we go from the
smaller number a to the bigger number b. Thus the integral is set up so
that for an everywhere positive function f, this gives a positive area under
its graph. In the case of a manifold, on the other hand, there is no such
fixed coordinate system, and so an orientation has to be set up by hand.
It turns out that for arbitrary smooth manifolds, this may not always be
possible, but the ones where it is are called 'orientable', and then we have
an intrinsic notion of integration available on such manifolds.
So far, we defined a spacetime as a Lorentzian manifold (M, g) with a
time-orientation. This was reasonable based on our everyday experience
that we can only 'go forward in time'. Particle physics experiments involv-
ing weak interactions show that even 1 at the fundamental level of elemen-
tary particle processes, the universe possesses chirality, that is, 'handed-
ness', and has a preference for one type over the other2 . Thus it makes
sense to also talk about spacetime being an 'oriented' manifold.
1 Macroscopic chirality is familiar; e.g. human hearts lie on the left.
2 The weak force operating in the nucleus is what governs the production of beta rays
during radioactive decay. Beta rays are energetic rays of electrons. Electrons possess
an intrinsic spin, and hence they can be classified as right- or left-handed depending on
whether they are moving along or against their spin axis. A famous 1957 experiment
demonstrated that beta particles produced during radioactive decay have a definite chiral
asymmetry in that left-handed electrons outnumber right-handed ones.
235
11.1 Orientation
We begin with orientations in the vector space setting. Let V be an m-

dimensional real vector space, and let w E (/\ mV*)\{0}, i.e., w is a nonzero
m-form. An m-form is sometimes referred to as a 'top form'. The action of
won an m-tuple formed by the elements of any basis must be nonzero. Sup-
pose on the contrary, w(e1, ·· · ,em)=0 for a basis B={e1, ···,em}- Then
since w =I= 0, there are vectors v1, · · · , Vm E V such that w (v1 , · · · , Vm) =I= 0.
Writing Vj =c}ei, we have
If in a summand corresponding to the index tuple I = (i 1, • • • , im), we

have that ij = it, then w( eiu · · · , ei,J = 0, and that term vanishes, thanks
to the alternating nature of w. So only the summands I= (i 1, • • • , im)
where (i 1, • • • , im) is a permutation of (1, • • • , m) will survive. But then in
such a term, if 1r E Sm is the permutation 1r(l) = i1, · · · , 1r(m) = im, then
w(eiu··· ,ei,J=w(e.,,.ci),··· ,e1r(m))=(sign1r)w(e1,··· ,em)=0, and so the
term becomes zero. So we obtain w( v1, • • • , vm) = 0, a contradiction.
So in light of the above, we can use the action of w to give an orienta-
tion to V by declaring an ordered basis B = (e 1, • • • , em) to be 'positively
oriented' if the action satisfies w(e 1, • • • , em) > 0, and negatively oriented
if this action is < 0. Thus it would seem that there are as many orienta-
tions as there are nonzero 'top-forms'. But the choice is not as large as
it may seem, since any two nonzero top forms are a nonzero multiple of
each other. Suppose that w, 0 E (/\ mV*)\{0}. Let B = (e1, · · · , em) be any
ordered basis for V. As the action of w and of 0 on B is nonzero, there is
some k =i=0 such that w(e1, · · · , em)= k-0(e1, · · · , em)- Thus we also obtain
w(ei 1 , • • • , ei,J = k-0(ei 1 , • • • , ei,,J- For any vectors Vj, l ~j ~ m, in V, by
writing Vj = c}ei, we obtain that
i,n
W ( V1, · · · , Vm ) = i1
W ( Cl ei 1 , • • • , Cm ei"' ) = i1 i,n (
Cl · · · Cm W ei 1 , • • • , ei"' )
= Ci1 • • -c~ k-0(eiu · · · , ei,J = k-0(v1, · · · , Vm)-
So any two nonzero top forms are a nonzero multiple of each other. Define
the relation ~ on the set (/\ mV*)\{0}, by setting w ~ 0 if w = k 0 for a
positive k. Then ~ is an equivalence relation. Under ~, there are clearly
only two equivalence classes: [w] and [-w], where w is any nonzero top
form. Indeed, for any equivalence class [0], we have either
• w= k0 with k>0, and so [w] = [0], or
• w=k0 with k<0, i.e., 0=(-k- 1)(-w), and so 0~-w, i.e., [0]=[-w].
Integration 237
This prompts the following.

Definition 11.1. (Oriented space, positively/negatively oriented bases).
Let V be an m-dimensional vector space. The equivalence relation ~ on
(Am V*) \ {0} is defined by setting w ~ 0 if w = k 0 for a positive k. An ori-
entation on Vis an equivalence class [w] under~ , where w E (AmV*)\{0}.
After choosing an w E (AmV*)\{0} , Vis said to be oriented with the ori-
entation [w]. Then an ordered basis B= (e 1, · • • , em) is called
• positively oriented if w( e1, · · · , em)> 0
• negatively oriented if w( e1 , · · · , em)< 0.
Example 11.1. (Standard orientation on ~m).
Consider the vector space V = ~m. Define the top form w E AmV* by
w(v1 , · · ·, vm) =det[v1 · · · vm] =det[vj] , for all vectors v1 , · · ·, Vm E ~m,
where Vj = vjei. Then w is nonzero (e.g. w(e1, ···,em)= detJ = 1 > 0).
The orientation [w] on ~mis called the standard orientation on ~m. ◊
IR2
IR3
e3
e2
JR e2
e1 e1 e1
!})
Exercise 11.1. Let V be an m-dimensional vector space. Then each ordered basis
B = (e 1, · · · ,em) induces an orientation [w B] on V , called the induced orientation
on V by the basis B by setting w B = E 1 /\ • • • /\ Em, where (E 1 , • • • , Em) is the
dual basis for V* corresponding to B. Show that WB =I= 0, and (e1, ··· ,em) is
positively oriented with respect to [wB] - Prove that for another ordered basis
B' = (e~, • • • , e:,,), the following are equivalent:
(1) B' positively oriented with respect to [wB] -
(2) If ej =a.ie;, then the change of basis matrix [a}] has a positive determinant.
(3) WB ~WB'·
In the case of an m-dimensional smooth manifold M, we would like the

orientations on the tangent spaces TpM, p E M, to vary smoothly. So, in
order to fix an 'orientation on the manifold ', we want an m-form field (also
referred to as a top-form field) fl E nmM , which vanishes nowhere, so that
the evaluation Sl(p) E Am (TpM)* is nonzero, and can be used to give an
orientation to TpM.
Not every smooth manifold is 'orientable'. An example is the Mobius strip

(with the edge removed, so that it is an 'open strip', and hence a legiti-
mate 2-dimensional manifold inside ~ 3 ), and the nonorientability can be
intuitively seen as follows. We refer to Figure 11.1. Suppose M is ori-
entable, so that there exists a nowhere vanishing 2-form field n. Let Vi
be the globally defined horizontal vector field along the length of the rect-
angular strip making the Mobius strip. Take a point p on the 'equator'
C. Consider the locally defined vector field Vi which points upwards, de-
fined in a little width determined by a small arc A c C around p. Then
O(p)((Vi)p, (Vi)p) =I- 0, and without loss of generality, suppose that it is
28>0. As A 3 q >-+ O(q)((Vi)q, (Vi)q) is smooth and in particular continu-
ous, we can find a point q EA near p such that O(q)((Vi)q, (Vi)q) > 8. But
now if r is a point on the arc A between p and q, then we imagine a vertical
cut at r, and on the resulting open set U in the Mobius strip (i.e. , U is the
cut Mobius strip), we consider two vector fields: Vi as before, and V2 , which
starts at p pointing upwards just as Vi was, but by the time it comes back
to q, we have (V2)q = -(Vi)q- But Olu(Vi , V2) is smooth on U, and in par-
ticular continuous along the curve C n U. As it is also nonzero 3 , it should
not change sign. However, O(p)((Vi)p , (V2)p) = O(p)((Vi)p, (Vi)p) = 28> 0,
and O(q)((Vi)q, (V2)q) = O(q)((Vi)q, -(Vi)q) < -8 < 0, a contradiction.
Fig. 11.1 Non-orientability of the Mobius strip.
If a manifold is orientable, then does every nowhere vanishing m-form field

define a different orientation? Just as with vector spaces, the choice is
smaller than it first appears. First, we note that multiplying a nowhere
vanishing m-form field by an everywhere positive smooth function, does
not change the orientation. We introduce the relation ~ on the set of all
3 Since at any point x along the equator, (V2)x is perpendicular to (Vi )x , so that
{(V1)x , (V2)x} forms a basis for the tangent space TxU.
Integration 239
nowhere vanishing m-form fields on M by setting O ~ 8 if O = f8 for some

function f : M --> JR such that f(p) > 0 for all p E M. Then it is easily
seen that ~ is an equivalence relation on the set of all nowhere vanishing
m-form fields on M. We remark that for a connected4 orientable manifold
M, it can be shown that 5 there are only two equivalence classes under ~,
namely [O] and [-0], where n is nowhere vanishing top form.
Definition 11.2. (Orientation on a manifold, induced orientation).

A smooth manifold M is called orientable if there exists an n E nm M such
that for eachp EM, O(p) E (/\m(TpM)*)\{O}. Let~ be the equivalence re-
lation defined by setting n ~ e if n = f e for some function f > 0 everywhere
on M. If Mis orientable, then an orientation on Mis an equivalence class
[O], where n E nm M is nowhere zero. After choosing an orientation [O] for
an orientable manifold M, we say M is oriented with the orientation [O].
For p E TpM, the induced orientation is given by equivalence class [O(p)]
(under the equivalence relation ~p on (/\ m(TpM)*)\{O} defined by w ~p 0
if w=k0 for a positive k).
Now that we have the notion of orientability, we formally give the definition
of a spacetime.
Definition 11.3. (Spacetime).

A spacetime (M, g) is a connected, 4-dimensional, oriented, time-oriented
Lorentzian manifold with the metric g, equipped with the Levi-Civita con-
nection induced by g, together with a time-orientation.

Consider the FLRW spacetime M = I x JR3, where the interval I= (0, oo ),
with the smooth structure given by the atlas comprising the single chart
(M, idixIRa ), and metric g given in Example 5.9 (p.89). We give M the
orientation [O], where n = dxAdyAdzAdt E n4 M, t,x,y,z being the
components of id1x1Ra. ◊
4 Recall that a topological space X of connected if there do not exist two disjoint,
nonempty, open subsets of X whose union is X. Also, X is path-connected if every pair
of points x, y E X can be joined by a path in X. For a smooth manifold M, M is
connected if and only if it is path-connected; see e.g. [Lee(2013), Prop. 1.11, p.8].
5 We will not prove this here, as we will not use this result, but here is a sketch: If n and
e are two top form fields that are nonvanishing everywhere, then by looking at the action
on chart-induced vector fields, for admissible charts, it can be seen that there must exist
a smooth function / which is nowhere zero, such that n = /0. As M is connected, this
f must be either everywhere positive or everywhere negative. This means that either
e E [n] ore E [-n].
Orientability of a manifold is needed for integration since oriented manifolds

possess 'oriented atlases', as we shall see. First we explain what we mean
by an oriented atlas.
Definition 11.4. (Oriented atlas).

An atlas A on a set M is said to be oriented if for all charts (U, x), (V, y) E A
for which Un V =I= 0, the determinant det[(y o x- 1 )'] of the Jacobi matrix
(yo x- 1 )' of the smooth map yo x- 1 : x(U n V) --+ y(U n V) is positive
everywhere on x(U n V).

Consider the Minkowski spacetime (M, g) from Example 5.6. Let p E Mand
B = {e1 ,e2 ,e3 ,e4 } be an orthonormal basis6 for V such that g(ei,ei) = 1
for i = 1,2,3 and g(e4 ,e4 ) = -1. For q EM, writing q = p + xi(q) ei,
gives a global admissible chart (M,xp,B), where the chart map xp,B is
given by Xp,B(q) = (x 1(q),x 2(q),x 3(q),x 4(q)) for all q E M. The atlas
A={(M,xp,B)} is trivially oriented.
If we take a different point p' E M and a different orthonormal basis
B' = {eL e~, e;, e4J for V, then A u {( M, Xp' ,B')} is oriented if and only if
the Lorentz transformation taking B to B' has determinant +1 (see Propo-
sition 5.3, p.84). This follows from the chart transition map expression
derived in Example 1. 7 (p.6). Also, from the solution to Exercise 5. 7, we
know that the determinant of any Lorentz transformation is ±1. ◊
Exercise 11.2. Consider the atlas A= {(Un, 'Pn), (Us, <ps)} from Example 1.5 (p.5),
defining the smooth structure of the sphere S2 • Show that A is not oriented.
Keeping the chart (Us, <ps), modify the chart map in (Un, 'Pn) by now setting
/Pn = Ro 'Pn, where R is a reflection in the u-axis, and check that the resulting
atlas .A= {(Us,'Ps),(Un,<pn)} is oriented. Show that .Au {(U,ip)} is oriented,
where (U, ip) is the 'spherical coordinate' chart given in Example 1.8 (p.7).
Proposition 11.1.
Suppose that M is a smooth manifold with the smooth structure [A].
Then M is orientable if and only if there exists an oriented atlas A E [A].
To prove this, we will first need a result on the existence of a 'partition of

unity'.
6 Thus, in contrast to Example 5.6, we will now label ea by e4 instead. This will natu-
rally put the timelike vector Ot after spacelike vectors, and this is the usual orientation
convention used for the metric with index 1.
Integration 241
11.2 Partitions of unity
If M is a smooth manifold, and S c M, then we will denote by S the

closure of Sin M, namely the intersection of all closed sets containing S.
For f E C 00 (M), the support off, denoted by suppf, is the closure of the
set of points where f is nonzero:
suppf := {p E M:f(p) =I 0}.

A collection {Fi : i E I} of subsets of M is said to be locally finite if for
each p E M, there exists a neighbourhood U(p) of p which intersects only
finitely many sets Fi 1 , • • • , Fin(p). A collection {Ui: i EI} of open sets in M
such that U Ui = M is called an open cover of M.
iel
Definition 11.5. (Partition of unity; subordinate to a cover).

Let M be a smooth manifold. A partition of unity on M is a collection
{rpi : i EI} of smooth functions <pi E 0 00 ( M) such that
• <pi (p) ~ 0 for all p E M and i E I,

• {supp <pi : i EI} is locally finite,
• I; <pi= l. (Note that at each point, this is a finite sum.)
ieJ
Given an open cover C = {Uj: j E J}, a partition of unity {rpi: i EI} is said
to be subordinate to C if for each i E I, there exists an j(i) E J such that
SUpprpi C Uj(i)·
Lemma 11. 1. Let M be a smooth manifold. Then M has a countable basis

all of whose elements have a compact closure.
Proof. Start with a countable basis B. Let Be c B be the subcollection

of elements from B which have a compact closure. We will prove that Be is
itself a basis. Let Uc M be open, and let p E U. Choose a neighbourhood V
of p such that V c U and V has a compact closure. (To see this we can take
a small neighbourhood of p, that is contained in U, which is homeomorphic
under a (chart) map x to an open ball B(O,r) in !Rm. Now take V to be
the image x- 1 B(O, i)- Then Vis compact.) As Bis a basis, there exists an
open set B E B such that p E B c V c U. But then B c V is compact too.
Thus BE Be. So given any open U and any p EU, we found a BE Be such
that p E B c U. Thus Be is a basis, and being a subset of the countable
collection B, is also countable. □
Proposition 11.2. Let M be a manifold.

Then there exists a sequence (Vn)nel\l such that
• Vn is open and Vn is compact for all n E N
• VicV1c½c½cVicVic••·
00
• U Vn = M.
n=l
Proof. By Lemma 11.1, there exists a countable basis {Bn : n E N} with

each Bn compact. Set V1 =B1. By compactness, there exists an n1 > 1 such
that Vi cB1 u · · · uBn 1 =:½. Now 7 ½ =B1 u · · · u Bn 1 =B1 u · · · uBn 1 is
compact, and so there exists an n2 > n1 such that ½ c B1 u · · · u Bn 2 =:Vi,
and proceed inductively in this manner. Finally,
M = uBn = B1
00
n=l
U ••• U Bn1 U ••• U Bn2 U •••
00 00
= LJ(B1 U ... U Bnk)= uvk.

k-1 k-1 □
In the above construction, we note that as Vn+l \ Vn is a closed subset of
the compact set Vn+l, it is itself compact. Moreover, since Vn-1 c Vn,
we have Vn+2\Vn c Vn+2\Vn-l· Now since Vn+l c Vn+2 we also obtain
Vn+l \Vn c Vn+2\Vn. Putting these together, we have:
(compact) Vn+l \Vn C Vn+2\Vn-l (opC'n)·
Yn-1) Vn) Vn+l) Yn+2)

compact
E 3
open
Exercise 11.3. Set Vo = 0- Show that for {Vn : n EN} from Proposition 11.2,
IJ Vn+l \Vn =M.
n=O
Theorem 11. 1. Let M be a smooth manifold, and let C = {Uj : j E J} be an

open cover of M. Then there exists a partition of unity subordinate to C.
Proof. Set Vo= 0- Let {Vn: n E N} be a sequence of open sets as in

Proposition 11.2. Fix n EN. For each pin the compact set Vn+l \Vn, let
j(p) E J be an index such that p E Uj(p)· Then p belongs to the open set
Wn+2\Vn-1) r'I uj(p)·
7 Here we use the fact that the closure of a finite union is the union of the closures; see
e.g. [Bishop and Goldberg(1980), p.9].
Integration 243
Let 'I/Jp E Coc,(M) be a bump function such that 'I/Jp = 1 in an open

neighbourhood Wp of p, and with supp'!/Jp c Wn+2\Vn-i) n Uj(p)· Then
{ Wp : p E Vn+ 1 \ Vn} is an open cover of the compact set Vn+ 1 \ Vn. So
there exist finitely many points p 1, · · · ,P~(n) supplying a finite subcover
{ Wpn,
1
.. . , Wpnk(n) } , with associated bump functions 'I/Jpn,
1
.. . , 'I/Jpnk(n) satisfying
supp'lj!Pe C Wn+2\Vn-1) n uj(pe'), 1 ~ £ ~ k(n). (11.1)
As the n E N changes, we get countably many bump functions:
The collection {supp'lj!Pe : 1 ~ £ ~ k(n), n EN} of their supports is locally

finite, since firstly, using Exercise 11.3, we see that the WPe (which is con-
tained in supp'lj!Pe' and which together cover Vn+l \Vn) together cover M,
and secondly, given any n, only finitely many of the supports will intersect
WPe: (supp'!/Jpf) n WPe =0 whenever N~n+3 by (11.1).
Any point p E M is contained in Vn+l \ Vn for some n, and so p E WPe
for some£ (because {Wpn, 1
· · ·, Wpnk(n) } forms a cover for Vn+l \Vn)- Then
1Ppe'(p) = 1 >0. So
oo k(n)
'I/J := I: I: 'I/Jpe'
n-1 f-1
is pointwise > 0 on M. We remark that the above sum is well-defined

thanks to the local finiteness of {supp'lj!Pe : 1 ~ £ ~ k(n), n E N}. Now
relabel the summands occuring in the double sum as 'lj! 1 , 'lj! 2 , 'lj! 3 , · · ·, and
set 'Pk := 'I/Jk/'I/J. Then clearly ~k 'Pk= 1, and 'Pk ~ 0 pointwise for all k EN.
Finally, as 'Pk= 1PPe' for some n and some £, we have
supp 'Pk= supp 'lppe C Uj(pe)
showing the wanted subordination property. □

Oriented atlas +-+ Orientability

We now return to the task left out at the end of the previous section on
orientation, i.e., to prove Proposition 11.1, stating that a smooth manifold is
orientable if and only if it possesses an oriented atlas in its smooth structure.
Proof. (Of Proposition 11.1). Suppose that M is orientable, and that
n E nm M is a nowhere vanishing top form field. We take any atlas A from
the smooth structure such that for each chart (U, x) E A, U is connected8 .
We will alter A to make it into an oriented atlas A by changing, if necessary,
the chart maps, but not the chart domains. Consider any chart (U, x) E A,
then we can write n = f dx 1 A .•. A dxm for some f E C 00 (U). Since the
function f vanishes nowhere in the connected set U, it is either everywhere
> 0 or everywhere < 0. If f > 0 on U, then we keep the chart (U, x) EA.
But if f <0 on U, then include in A, instead of (U,x), the new admissible
chart (U,p >-+ (-x 1 (p),x 2 (p), • • • ,xm(p))=:y(p)). Then
n = f dx 1 A • • • Adxm=J((oyi1X 1 )dyi 1)A • • • A((Oyi,nXm)dyi"')
= j(Oyi 1X1 ) · • · (Oyi,,.Xm)dyi 1A · · · Adyi"'
= f ~ (oy.-(l)X 1 ) ... (oy.-(m)Xm)dy1r(l)I\ ... Ady1r(m)
1rESm.
= J det[OyiXi] dy 1 A · · · Adym = g dy 1 A··· Adym,

where g:= f det[oyixi]. Since det[oyixi] = -1 and f <0 on U, we have g>O
on U.
Now we have got a new atlas A, with the property that for any chart
(U, x) E .A, if we decompose n = f dx 1A • • • Adxm in U, then f > 0 in U. We
claim that this new atlas is oriented. To see this, let (U, x) and (V, y) be
two overlapping charts in A. Then in U n V,
n= f dx 1 A ... Adxm
= f ((oyi1 x 1 )dyi1) A •· • A ( (oyirnXm )dyi"')
= fdet[OyiXi]dy 1 A ··· Adym.
Since we also know that O=g dy 1 A •·· Adym, we obtain g=fdet[Oyixi].
But by the aforementioned property of the atlas, we know that f > 0 in U,
and g > 0 in V, so that they are in particular also > 0 in Un V. The relation
g = f det[oyixi] now implies that det[oyixi] > 0 in Un V. This completes the
proof of the 'only if' part, showing that the orientable manifold possesses
the oriented atlas A in its smooth structure [A] = [A].
8 Atlases with connected charts always exist, since a nonconnected chart can be split
into connected components.
Integration 245
It remains to show that if M admits an oriented atlas A in its smooth

structure, then M is orientable, that is, there exists a nowhere vanishing
top form field n E nm M. Let A = {(UO , x 0 ) : a E I}. Take a partition of
unity {cp 0 : a EI} subordinate to {UO : a EI}. Set
n= ~ 'Padx~ I\ •.• Adx1;:.
aEl
Let p E M. Since I; 0 cp 0 = 1, and all the cp 0 are pointwise nonnegative,

there exists an index f3 EI such that cp,a(p) > 0. Then
ael:U0 r,Up#0
~ 'Pa det[oxj x:,] dx} A · · · A dx";,

ael:U0 r,Up#0 /3
and so the action of the top form n(p) E /\m(TpM)* on the ordered basis
(ox},P' ... 'Ox;r,p) for TpM is
n(p)(ox1,P' · · · , ox 13 ,p) = ~ 'Pa(p)(det[ox~ x:,])(p) det I

ael:Uar,Up#0
~ cp,a(p)(det[oxj x1])(p) >0,

/3
where we have used det[oX/3j x~] >0 on U,a n U0 • Thus n(p) #0. Asp EM
was arbitrary, n is nonvanishing everywhere, and so M is orientable. □
Definition 11.6. (Orientation-compatible oriented atlas).

Let M be a smooth manifold with an orientation [n] for a nowhere vanishing
top-form field n E nm M. An oriented atlas A from the smooth structure
of Mis called [n]-compatible if in each chart (U, x) EA, the decomposition
n = nu dx 1 A . . . A dxm yields nu > 0 everywhere on U.
It is clear that orientation-compatibility of an atlas is independent of the
n chosen in the orientation, and so this notion, as specified above, is well-
defined.
Exercise 11.4. Let M be a smooth manifold with an orientation [O], and let A, A'
be atlases in the smooth structure of M that are both [OJ-compatible. Show that
A u A' is also an oriented atlas which is [OJ-compatible.
11.3 Integration
We first provide some heuristic motivation for the definition that is to follow.
The aim is to give intrinsic definition of the integral on a manifold.
We would like to 'integrate' a function f : M ---+ R. Assume that f is

compactly supported, and even smooth, hoping that there will not be any
problems from the analysis point of view while defining such an object.
We could use charts as long as at the end we have a chart-independent
number. Using charts means that the problem boils down to the integral
in Rm, which is already familiar to us. We recall that if / : Rm ---+ R has
compact support, then we define
i.e., the integral is defined via iterations of I-dimensional integrations. Also,

recall the change of variables formula: If : Rm ---+ Rm is a diffeormor-
phism, then
f f(u1, ... ,um)du1··•dum

JR"'
= JRJJ o - 1 )(u1,- • • , um) Idet [:~~-l )i (u1, ... , um)] Idu 1 ... dum.
An attempt at a definition of the integral of a smooth, compactly supported

function f on a smooth manifold M could be made as follows. Assume for
the moment that a smooth manifold M has global charts (M, x) and (M, y)
with bijections x, y : M ---+ Rm. Then we could try
f /:= f (Jox- 1 )(u1,···,um)du 1 ---dum,

JM JR"'
and hope that if we used y instead, then the result would be the same.
But this is clearly not true, since the above change of variables formula
shows that if M = Rm, with the standard smooth structure, and we con-
sider the admissible charts (Rm,x=id), and (Rm,y=), where is any
diffeomorphism as in the change of variables formula, then
f (fox- 1 )(u1,··· ,um)du 1 ·••dum = f f(u1,··· ,um)du 1 •··dum

JJRm JJRm
= JR"'(fo- 1 )(u1,•·· ,um)ldet[:~~-l)i(u1,•·· ,um)]ldu 1 ---dum
= f (foy- 1 )(u1,··· ,um)l(det[(oy;xi)])(y- 1 (u1,•-· ,um))ldu 1 •··dum.

JR"'
So, in general,
f (fox- 1 )(u1,··· ,um)du 1 ••·dUm"f" f (foy- 1 )(u1,•·· ,um)du 1 ··•dum.

JRm JJRm
Integration 247
We note that, firstly, the absolute value on the determinant would disappear
if we used an oriented atlas. Secondly, if we formally replace dx 1 • • • dxm
by dx 1 A • • • A dxm, then the definition of the integral of a top form field n
by setting
where n = f dx 1A • • • Adxm in a global chart (U = M, x), where x : M --+ !Rm

is a bijection, seems to be an intrinsic definition. Indeed, if (V = M, y)
is another global chart, where y : M --+ !Rm is a bijection, then writing
O=f dx 1 A · · · Adxm=gdy 1 A · · · Adym, we have g=f det[oyixi], and so
f (goy- 1 )(u1,··· ,um)du 1 •··dum

JIR=
= f (foy- 1 )(u1,··· ,um)(det[oy;xi])(y- 1 (u1,•·· ,um))du 1 •··dum
J]R=
= f (foy- 1 )(u1,--- ,um)ldet[oy;xi])(y- 1 (u1,--- ,um)ldu1 ··•dum
JIR=
= f
JJR=(fox- I
1 oxoy- 1 )(u1, ... ,um) det [o(xoy-
oui
1 )i ]I
(u1, ... ,um) du 1 ---dum
= f (fox- 1 )(u1,•··,um)du1 ·•·dum.

J]R=
When global charts are not available, we use a partition of unity to patch
local integrals.
Definition 11. 7. (Integral of top form fields).

Suppose that Mis a smooth manifold with an orientation [8].
Let A={(U"',x"'):aEA} be a [8]-compatible oriented atlas in the smooth
structure of M, and {'Pi : i E J} be a partition of unity subordinate to
{U"': a EA}. Let n be any m-form field with a compact support, that is,
there exists a compact set Kc M such that OIM\K = 0. Then we define
where nua(;) E C 00 (U<>(i)) are the functions such that
n = nuaU)dx~(i) A .•. A dx~(i)
in the chart (U<>(i),x<>(i))-

If we accept well-posedness, it is clear that the above gives a R-linear map
f M : {n E nm MI n has compact support} - R.

We need to check the well-definedness at various levels. First we note
that even with a particular [8]-compatible atlas, charts may overlap, and
we should check that in the overlap region, the right-hand side above is
the same. In the intersection region of two charts (U, x), (V, y) E A, if
n = nu dx 1 I\ ... I\ dxm = nv dy 1 I\ ... I\ dym, then nu= nv det[OxiYi]. With
<l>:=y o x- 1 , for u E x(U n V), we have
nu(x- 1 (u)) = nv(x- 1 (u))(det[OxiYi])(x- 1 (u))
o(yox-l)i]
= nv ((y- 1 o )(u)) det [ oui (u)
a<1>i ]
= nv ((y- 1 o )(u)) det [ oui (u) .
Thus for any f E C 00 (U n V), we have
f (f o x- 1 )(u) nu (x- 1 u) du 1 • • • dum
Jx(UnV)
= f (f o y- 1 o ) (u) nv( (y- 1 o ) (u)) det [ ~i (u)] du 1 ... dum
J(xoy- 1 oy)(UnV) uuJ
= f (foy- 1 o)(u)nv((y- 1 o)(u))ldet[~;(u)]ldu 1 ···dum

J(,p- 1 oy)(UnV) UU
= f (foy- 1 )(u)nv(y- 1 u)du 1 ·-•dum.

Jy(UnV)
Next, we need to check that the integral depends neither on the chosen [8]-
compatible oriented atlas from the smooth structure of M, nor on the parti-
tion of unity. Suppose that A= {(Un, xn) :a EA} and A'= {(V,a, Y,a) :/3 EB}
are two oriented [8]-compatible atlases, with corresponding subordinate
partitions of unity {'Pi: i EI} and {'l/Jj : j E J}, respectively. We have
L f,Ua(i) 'Pin= L f,Ua(i) 'Pi L 'lpjn

i i j j
(these are finite sums,
= LLf,Ua(i) 'Pi'lpjn
i j
since n has compact support)
= LL I, 'Pi1Pin (since supp(cpi'lj;j) c Ua(i) n V,a(j))

Ua(i) n V/l(j)
i j
and by symmetry, we get the same last expression also starting from
~f 'lpjn. This completes the justification of the well-posedness.
i Jv/l(j)
Integration 249
Exercise 11.5. (Baby9 Stokes' theorem).

The aim of this exercise is to prove a basic version of Stokes theorem:
If Mis an m-dimensional oriented smooth manifold, and n E nm-IM is com-
pactly supported, then
L dn =O.
Proceed as follows. Let [8] be the orientation on M, A= {(U"', x"') : a EA} be a

[8]-compatible oriented atlas, and {'Pi: i E /} be a partition of unity subordinate
to A. Then n = I:i 'Pin since I:i 'Pi= 1. If for i E J, SUPP'Pi C Ua,(i), then in Ua,(i),
justify the decomposition
'Pin= (-1t- 1 /kdX 1 A ... Adxl<A ••• Adxm,
for some smooth functions /k E C 00 (Ua(i)) having compact supports in Ua(i)·
Using this, show that d(rpin) = (oxifi + ... + Oxmfm)dx 1 A ... Adxm. Thus
J, d(rpd1) =
Uo:(i)
i
Xo:(i)Uo:(i)
(0.,1/1 + · · · + Ox=fm)(x~(i)u)du 1 ···du"'.
Show that this is O using Fubini's theorem. Conclude that J dO = 0.

M
11.4 Volume form field in semi-Riemannian manifolds
In the case of a semi-Riemannian manifold (M, g) which admits an oriented

atlas A in its smooth structure, there is a natural top form field volg which
is everywhere nonvanishing, and also such that A is [volg]-compatible. Be-
sides integrating compactly supported m-form fields using this orientation,
we can also use it to compute the integrals of compactly supported smooth
functions f on M canonically 10 by setting
f f f f·
M = M volg.
In particular, for a compact manifold, the volume of M is defined to be

fMl = J volg.
M
9 A more general version of Stokes' theorem for smooth manifolds M possessing a

'boundary' iJM exists, which roughly speaking, says that
f dO. = f 0.
JM J,'JM
for compactly supported (m-1)-form fields 0. on M. Thus in our baby version, there
is no boundary: iJM = 0- However, we will not need this more general result in the
sequel, and as it will be too much of a detour (with a necessary discussion of manifolds
with a boundary, submanifolds, and so on), we will not include this important result in
this book. We refer the interested reader to [Lee(2013), pp.411-415].
10 We could have also done this when we only had an orientation [O.], but then we would
have to make an arbitrary choice of the top form field in [O.]. In a sense, given an oriented
atlas, now the metric chooses a natural orientation from among the many available.
Definition 11.8. (Volume form field).

Let (M, g) be an m-dimensional semi-Riemannian manifold with metric g,
and which admits an oriented atlas A in its smooth structure. The volume
form field volg E nm M is defined as follows: In any chart (U, x) E A,
volglu = v'I det[g(ilxi, Oxi )] Idx 1 A · · · A dxm.
Let us first check the chart-independence of the above definition. If (U, x)
and (V, y) are overlapping charts in A, then in U n V, we have
Vldet[g(Oyi,Oyi)]ldy 1 A ... Adym
= ✓1 det[g((oy,xk)Oxk, (OyiXl)oxt)]I (ox'1Y 1 )dxi1 A ... A (ox,mYm)dxim
= ✓1 det[(oy,xk)(Oyi xi) g(Oxk, Oxt )] det[Ox•Yr] dx 1 "

I ... " dxm
ldet([oy,xk]t[g(Oxk,Oxt)][Oyixl])I det[Ox•Yr]dx 1 A ... A dxm
= V(det[oy,xi])21 det[g(Oxk, Oxt)]I det[Ox•Yr]dx 1 " ... "dxm
= vi det[g(Oxk, Oxt)]I det[oy,xi] det[Ox•Yr]dx 1 A ... A dxm
= vi det[g(Oxk, Oxt)]I det[(oyrxi)(Ox•Yr)]dx 1 A ... A dxm
= vi det[g(Oxk, Oxt)]I det[8~]dx "dxm

1" ·· ·
= vi det[g(Oxk, Oxt)]I (l)dx dxm.

1 A ... A
In the above, while extracting det[oyixk] outside the square root sign, we
have used its positivity, thanks to the fact that A is oriented. This shows
that the definition of vol g does not depend on the chosen chart from A.
So for each p EM, we have a well-defined top form vol g (p) E /\ m (TpM) *.
For vector fields Vi, · · · , Vm E TJ M, the map
p >-+ volg(p)((Vi)p, · · · , (Vm)p)
= VI det[gij(p)]I ((dx 1 )p A · · · A (dxm)p)((Vi)p, · · ·, (Vm)p)
= VI det[gij(p)]I det[Vj(xi)](p)
is smooth in each chart (U, x), and hence is smooth on M. It follows from
Proposition 4.1 that volg is a (0, m)-tensor field. As it is clearly alternating
in each chart, it belongs to nm M. We also note that for any p E (U, x),
the action ofvolg(p) E /\m(TpM)* on the ordered basis (oxi,p,··· ,ox"',p)
for TpM is given by
(volg(p))(oxi,p, · · · , Ox"',p) = Idet[g(Oxk,p, Oxt,p)]I det I> 0.
Integration 251
So volg is nowhere vanishing. Moreover, since -JI det[g(8xk,Ox£)]1 > 0 on

(U, x), it follows that A is [volg]-compatible.
Definition 11.9. (Integration of functions, volume).
Let (M, g) be an m-dimensional semi-Riemannian manifold with metric g,
and which admits an oriented atlas A in its smooth structure. Let M be
given the orientation [volg]- If/ E C 00 (M) has compact support, we define
its integral on M by
IM!= IMJ·volg,
where the integral of the right-hand side is that of the compactly supported
top form field f • vol g E nm M in the sense of Definition 11. 7. If M is
compact, we define its volume by
VolgM = IM l = IM volg.
Exercise 11.6. Consider the sphere (S 2 , g) as a Riemannian manifold with the
metric g induced on its tangent spaces from the Euclidean inner product on
R 3 , as described in Example 5. 7, p.87. Determine its volume VolgS2 with the
orientation induced by the oriented atlas .A from Exercise 11.2.
Example 11.4. (Schwarzschild spacetime).

Consider the Schwarzschild spacetime (M, g), where M =Rx (2m, oo) x 8 2
has the smooth structure and the metric g described in Example 5.10 (p.89).
Consider the sphere (8 2 , g) as a Riemannian manifold with the metric g 8 2
induced on its tangent spaces from the Euclidean inner product on R 3 , as
described in Example 5.7, and the orientation induced by the oriented atlas
A from Exercise 11.2. Denote the volume form field on (8 2 , g32) by vol g 52 .
Then
0 := r 2 dr I\ volg 52 A dt E S1 4 M (11.2)
is nowhere vanishing, and we equip (M, g) with the orientation [O]. Here
the functions t, r : M -> R are given by
t(p) = T, and
r(p) = p,
forp=(r,p,p)Elix(2m,oo)x8 2 =M. ◊
Exercise 11.7. Given the oriented atlas A*:= .Au{(U,rp)} for S 2 in Example 11.2,
consider the oriented atlas
A. = {(Rx (2m, oo) x W, p= (T,p,p) >-+ (p,u(p), T)), (W, a-) EA*}
for the Schwarzschild spacetime (M, g). Show that the volume form field volg is
the same as (11.2).
11.5 Hodge star
If V is an m-dimensional real vector space, then dim/\ kV* = (1;:) for all
O~k~m. So
dim/\m-kV* = (m1:_ k) = (;) = dim/\kV*.

So the two spaces /\ kV* and /\ m-kV* are isomorphic as vector spaces. We
will see in this section that in the context of an oriented real vector space
with a scalar product, the Hodge star operator 11 is such an isomorphism.
This operator will be used to define the magnetic field perceived by an
instantaneous observer in spacetime.
Lemma 11.2. Let V be an m-dimensional real vector space with a

scalar product g, and an orientation [w], for an w E /\mV*\{O}. Let
(e1, · · · , em), (/1, · · · , fm) be ordered bases for V that are positively oriented,
and are also orthonormal with respect tog. If fi = a;ei, l ~ j ~ m, then
det[a;] = 1.
Proof. We have (no sum over j):
Also, O=g(/i,/j)=g(afek,a;et)=afa;g(ek,et)= ~afajg(ek,ek) for i=faj.

k=l
Thus,
Taking determinants, (-1) 19 = (det[a;])2(-1) 19 , and so det[a;] E {-1, 1}.

But
1rES,n
= (det[a;])w(e1, · · · ,em),
and since w(e1, · · · , em) > 0 too, it follows that det[a;] = 1. □
11 Named after W. Hodge (1903-75), a British geometer.
Integration 253
Definition 11.10. (Volume form on a vector space with a scalar product).

Let V be an m-dimensional real vector space with a scalar product g of
index i9 , and an orientation [w], for an w E /\mV*\{0}. Let (e 1, • • • , em) be
an ordered basis which is positively oriented, and orthonormal with respect
to g. Then the top form
vol 9 := (-lY 9 (ed I \ · · · A(em)b E J\mV*
is called the volume form on V with respect tog and [w].
It might seem that vol 9 depends on the chosen basis, but it does not, since
if (/1, • • • , fm) is also an ordered basis which is positively oriented with
respect to [w], and orthonormal with respect tog, then with /j = a}ei,
ud I\ 0 0 0
I\ Uml = (ai1 ei1 )b /\ """ /\ (a~ei,,. l
= ~a;< 1l ... a~ml(sigll'lr) (ed I\··· A(em)b

1rES,n
= det[a}] (ed I \ · · · I\ (em)b = 1 (ed I \ · · · I\ (emt

This shows that vol 9 is independent of the chosen positively oriented or-
thonormal basis. Also, vol 9 E [w]: Indeed, w(e 1,··· ,em)>0 and
volg(e1, · · · , em) = (-lY 9 ((ed I \ · · · I\ (eml)(e1, · · · , em)
= (-lY 9 det[g(ei,ej)] = (-lY 9 (-lY 9 = 1 > 0.
Thus w =c -vol 9, where c:=w(e1, · · · , em) >0, and so vol 9 ~w.
Exercise 11.8. Let V be an oriented m-dimensional real vector space with a scalar
product g of index lg, and the volume form volg. Let (e1, • • • , em) be a positively
oriented ordered basis such that it is also orthonormal with respect to g, and
let (€1,··· ,€m) be the corresponding dual basis for V*. Let i(k) :=g(ek,ek),
1 ~ k~m. Show that (ek)° = i(k)€\ 1 ~ k~m. Conclude that volg =€ 1 /\ • • • /\ €m.
Exercise 11.9. Let V be an oriented m-dimensional real vector space with a scalar
product g of index lg, and the volume form volg. Let (v1, • • • , vm) be an arbitrary
(not necessarily orthonormal) positively oriented basis for V, and let (w 1, • • • , w m)
be the corresponding dual basis for V*. Show volg = -Jldet[g(vi, Vj)]lw 1 A••• I\ Wm.
Let V be an oriented m-dimensional real vector space with a scalar product

g, and the volume form vol 9 . Let w E !\kV*, and v1,· ·· ,Vm-k EV. Then
WA(v1)b I \ · · · A(Vm-k)b E J\mV*, and so using the basis {vol 9 } for J\mV*,
there is a unique f w(v1, · · · , Vm-k) E R such that
W/\(v1l I\ ... /\(Vm-k)b=fw(v1,·· · ,Vm-k)volg.
By the JR-linearity of the wedge product in each of its factors, the map
fw: vm-k-+ JR, given by vm-k 3 (v1,··· ,Vm-k) >-+ fw(v1,··· ,Vm-k)
is multilinear. Also, being a wedge product, it is also alternating in
V1, · · · , Vm-k· Hence fw E Am-kv*. Again by the linearity of the wedge
product, the map A kV* 3 w ...... fw E 1-,m-kV* is linear. This map is called
the Hodge star operator (one for each fixed O:::;; k < m).
If k = m, then we set *(w) = w(e1, · · · , em) E JR, for w E AmV*, where
(e1, • • • , em) is any positively oriented orthonormal basis. This gives a linear
map * : AmV* -+ JR= A0 v*. To show that the map does not depend on
the chosen ordered orthonormal basis, we note that since {vol 9 } is a basis
for Amv*, for w E Amv*, there exists a CE JR such that w = cvolg, and if
(e1, · · · , em), (!1, · · · , fm) are positively oriented orthonormal bases, then
w(e1, ···,em) =cvolg(e1, ···,em) =cl =cvol 9 (fi, · ··, fm) =w(J1, · · ·, fm)-
Definition 11.11. (Hodge star operator).

Let V be an oriented m-dimensional real vector space with a scalar product
g, and volume form vol 9 . The Hodge star operator * : A kV* -+ Am-kV*
is defined to be the linear map that satisfies, for all w E A kV*, and all
V1, · · · , Vm-k EV, that
WI\ (vd I\··· I\ (vm-kY' = (*w)(v1, · · · , Vm-k) vol 9.
Lemma 11.3. Let V be an oriented m-dimensional real vector space with

a scalar product g of index i9 , and the volume form vol 9 . Let w E AkV*,
v, v1, · · · , Vm-k EV. Then:
(1) *VOlg = 1.
(2) *1 = (-l)f 9 vol 9 .
(3) (*w)(v1, · · · ,Vm-k) =*(WI\ (vd I\··· I\ (vm-k/).
Proof.
(l)*vol 9 =volg(e1,··· ,em)=l. Here(e1,··· ,em)isanyorderedpositively
oriented orthonormal basis.
(2) Let w1, · · · , Wm EV. The top form w:= (w1/ /\ · · · /\ (wm)I> is equal to
w(e1, · · ·, em)vol 9 , as w=cvol 9 , and we had seen that c=w(e1, · · · , em)
for any positively oriented orthonormal basis (e1, • • • , em). We have
w(e1, · · · , em) = ((wd I\··• I\ (wm)l>)(e1, · · · , em)
= det[(wdej] =det[g(wi, ej)] = det[(ejlwi]
= ((ed I\··· I\ (em/)(w1, · · · , Wm)
= (-li9 volg(w1, · · · , Wm)-
Integration 255
Thus
(* 1) (W1, · · · , Wm) vol 9 = 1 /\ (wd I\ · · · I\ (wml
= (-1)1 9 volg(w1, · · · , wm)vol 9 .
Consequently, that * 1 = (-1 )19 vol 9 .
(3) It follows by the IR-homegeneity of* : /\ mV* -+ JR = /\ 0 v* that
*(WI\ (vd I\ · · · I\ (Vm-kY') = *( (*W )( V1, · · · , Vm-k) vol 9 )
= (*W)(v1, · · · , Vm-k) · (*VOlg)
= (*w)(v1,··· ,Vm-k)· l. □
Analogous to the interior multiplication of form fields, we introduce the
following 'pointwise' version for forms.
Definition 11.12. (Interior multiplication).
Let V be a real vector space, and v EV. If k-;;:: 1 and w E /\kV*, then the
interior multiplication ivw E /\k- 1V* of w by vis defined by
(ivw)(v1, · · · , Vk-1) =w(v, v1, · · · , Vk-1), for all v1, · · · , Vk-1 EV.
If k= 1, then for a 1-form w EV*, we set ivw=w(v) E IR=/\ 0 V*.
Exercise 11.10. Let V be a vector space. For all v E V, w1, • • • , wk E V* = I\ 1 V*,
k ~
show that iv(w 1 A··· A wk)= ~ (-1r- 1 wr(v)w 1 A··· A wr A··· A wk.
r=l
Exercise 11.11. Let V be a vector space, and v EV. For all w E /\kV*, 0 E /\lV*,
show that iv(w A0) = (ivw) A0 + (-ltw Aiv 0.
Lemma 11.4. Let V be an m-dimensional real vector space with a scalar

productg. LetwE/\kV* andv,v1,··· ,Vm-k+l EV. Then
(ivw) I\ (vd I\··· I\ (vm-k+d
m-k+I -
= ~ (-1t+rvl>(vr)WA(vd I\··· A(vr)I> I\··· A(Vm-k+d-
r=l
Proof. w I\ (vd I\··· I\ (vm-k+d E /\ m+lV* = {O}, and Exercises 11.11

and 11.10 imply
0=iv(0) = iv(wA(vd I\··· A(Vm-k+d)
=(ivw)A(vdl\ ··· A(Vm-k+d
m-k+l ---
+ (-ll ~ (-1r- 1 ((vrlv) WI\ (vd I\ ···I\ (vr)I> I\··· I\ (vm-k+d-
r=l
The proof is completed by observing that (vr)Pv = g(v,vr) = vl>(vr) and

rearranging. □

a scalar product g and the volume form vol 9 • Let v E V, and let w E /\kV*.
Then *(WI\ vi>)= iv(*w).
Proof. For v1, · · · , Vm-k-1 EV, using Lemma 11.3.(3), we have

(*(W/\Vl>))(v1,··· ,Vm-k-1) = *((wAvl>) A(vd I\··· A(Vm-k-d)
= *(W /\ (vi> I\ (vd I\ · · · I\ (vm-k-d))
= (*w)(v, v1, · · · , Vm-k-1)
= (iv(*w))(v1, · · · ,Vm-k-1).
□
a scalar product g and the volume form vol 9 • Let w E /\kV*, and let v E V.
Then *(ivw) = (-l)m-l(*w)Avl>.
Proof. We have
(-lr-l(*W) /\Vi>= (-1r- 1(-1r-kvi> /\ *W= (-lt+lvi> I\ *W.
Let *W=fIE 1 , l=(i1,··· ,im-k), where {€1,··· ,fm} is a basis for V*. For
all V1, · · · , Vm-k+l EV,
((-lr-l( *W) I\ Vb)(v1, · · · , Vm-k+l)
= (-1t+l(vl> I\ (*W))(v1, · · · , Vm-k+1)
vl>v1 · · · vl>Vm-k+l 1
= (-1t+lJidet
r Ei1_v1 Ei1v~-k+1 .
f.i=~•v1 • • · f.i=-•v~-k+l
Expanding the determinant along the top row, and using Lemma 11.4
((-lr-l(*W) /\Vb)(v1, · · · , Vm-k+l)
m-k+l
= (-lt+l ~ (-lf- 1(vi>vr)(*W)(v1,··· ,'Ur,··· ,Vm-k+l)
r=l
m-k+l -
= ~ (-1t+r(vl>vr) *(WI\ (vd I\··· I\ (vr)i> I\··· I\ (vm-k+d)
r=l
□
Exercise 11.12. Let V be an oriented m-dimensional real vector space with a
scalar product g of index 19 , and the volume form vol 9 • Let w E !\kV* and
v1, • • • , Vk EV. Prove that
(-ll(m-l)(*w) Avt A • • • Avt =*(iv.(-·· (iv 1 w) · · · )) = (-li 9w(v1, · · · , Vk)vol 9.

Integration 257
Lemma 11. 7. Let V be an oriented m-dimensional real vector space with

a scalar product g of index i9 , and the volume form vol 9. Let w E /\kV*.
Then** w = (-l)k(m-l)+igw.
Proof. For all v1, · · · , Vk EV, we have (using Exercise 11.12)

(*(*w))(v1,··· ,vk) = *((*w)A(vd I \ · · · A(Vm-kl)
= (-1t(m-l) *(w(v1,··· ,vk)(-1) 19 vol 9 )
= (-1t(m-l)(-l)i9 w(v1,· ·· ,vk) *VOlg
= (-1t(m-l)+ 19 w(v1, · · · , Vk) 1.
Thus** w = (-l)k(m-l)+igw. □
In particular, * : /\ kV* -+ I\ m-kV* is an injective linear transformation,
and since
dim/\kV* = (;) = (mr:: k) = dim/\m-kV*,
it follows that * is an isomorphism.

a scalar product g and the volume form vol 9 • Suppose that w, v E /\kV*.
Then Wl\*V = Vl\*W.
Proof. Thanks to the linearity in each of the arguments w, v on both

sides, it is enough to prove the result for w = (w 1)i, I\ • • • I\ (wk)P and for
v= (v1)P I \ · · · I\ (vk)i,, where w1, ···Wk, v1, · · · , Vk EV. We have
Wl\*V = (-ll(m-kl(*v)Aw = (-ll(m-k)(*v)A((wd I \ · · · A(wkl)
= (-1t(m-kl(-1t(m-l)+i9 v(w1, · · · , Wk)vol 9
= (-1) 19 ((vd I \ · · · A(vkl)(w1,··· ,wk)vol 9
= (-1) 19 det[(vdwj]vol 9 •
Swapping w, v, v I\ * w = ( -1 )19 det [ (Wi) i, v j] vol9 . Since taking the transpose
preserves the determinant, and as (vi)i,wj = g(vi,Wj) = (wj )i,vi, we conclude
that Wl\*V=Vl\*W. □
The following result tells us how to compute the* of a k-form, as it describes

the action on the basis vectors for /\kV*.
Definition 11.13. (Complement of a tuple, index of a tuple).

Let l =(ii,··· , ik) beak-tuple obtained from {1, · · · , m} with i1 < · · · <ik.
Then the unique (m-k)-tuple le:= (j1, · · · ,im-k), with j1 < · · · < im-k
and l u le = {1, · · · , m}, is called the complement of l.
Now let V be an m-dimensional real vector space with a scalar product g,

and let {e1, • • • , em} be an orthonormal basis with respect to g. For the
k-tuple l = (i 1, • • • , ik) of elements from {1, • • • , m}, we define the g-index
of l by i(l) := g(eiu ei 1 ) • • • g(eik, eik) = i(i1) · · · i(ik)-
Below, for a r-tuple L = (1! 1, • • • , l!r), we set EL:= /- 1 A • • • A /-r.
Proposition 11.3. Let V be an oriented m-dimensional real vector space

with a scalar product g and the volume form vol 9. Let (e1, · · · , em) be
a positively oriented ordered basis such that it is also orthonormal with
respect to g, and let (€1, • • • , €m) be the corresponding dual basis for V*. Let
l = (i1, · · · , ik) with i1 < · · · <ik, le be its complement, and let 7r11 e E Sm be
the permutation taking (l,Ic) to (1,··· ,m). Then *€ 1 = (sign1rIIJi(lc),/e.
Proof. Let J = U1, · · · ,im-k) be any (m-k)-tuple of distinct elements

of {1, • • • , m}. By Lemma 11.3.(3), and using Exercise 11.8, we have
(*i)(eju•·· ,ei,n-k) = *(€1 A(ej 1 l A··· A(ei,n-kl)
= i(j1) · · · l(jm-k) *(€1 A€J),
which is zero if l n J =I= 0- On the other hand, if J = l c, then
*(€1 A€J) = *(t./ Aie) = *((sign1rIIJvol 9 ) = (sign1rIIJ * vol 9 = (sign1rIIJ 1.
We used Exercise 10.5 to get the second equality. So if le=: (1!1, · · · , l!m-k),
then (*€1 )(eeu · · · , ee"'_k) = (sign1rIIJi(lc)- Summarising, we have shown
l
(*€ )(eju · · · 'ei,n-k) =
{Q(sign1rIIJi(lc)
if lnJ=/=0,
if J =le.
} (11.3)
We will show that this coincides with the action of (sign1rIIJ i(lc) €1e on
(ej 1 , • • • , ei,n-k). Let le= (1!1, · · · , l!m-k)- Then
€ 1e(eji,""" ,ej,n-k) = (i 1 A ... Ai,n-k)(eji,""" ,ej,,,_k) = det[/r(ej.)].
0, then at least one column of the matrix [/-r (ej.)] is zero, so
If l n J =I=
that the determinant above vanishes. On the other hand, if J = le, then
det[/r(ej.)] = det[/r(ee.)] = det[8;] = 1,
and so E1e (ej 1 , • • • , ei,n-k) = 1. Summarising, we have shown that
(11.4)
The result now follows from (11.3) and (11.4). □

Integration 259
Now suppose that (M, g) is an m-dimensional semi-Riemannian manifold

with metric g, which admits an oriented atlas A in its smooth structure. Let
M be given the orientation [vol g]. For each p E M, the tangent space TpM
is given the orientation [volg(p)]. Then we obtain the Hodge star operator
* : /\ k (TpM) * -+ I\ m-k (TpM) * corresponding to the volume form vol g(p)
belonging to the orientation [volg(p)] on TpM. In fact, for n E f!k M, the
pointwise definition
*!1 := (M 3 p >-+ *(!1(p)) E /\m-k(TpM)*)
gives an element 12 * n E nm-k M. Thus we have Hodge star operators

*: nkM-nm-kM
for O~k~m, mapping k-form fields to (m-k)-form fields.

Exercise 11.13. Let (M, g) be an m-dimensional semi-Riemannian manifold with
metric g, which admits an oriented atlas A in its smooth structure. Let M be
given the orientation [volg]- Show that * : {lk M-+ gm-k Mis a C 00 (M)-linear
map. Also show that * * n = (-l)k(m-l)+i(g)n for all n E gk M.
Exercise 11.14. Let (M, g) be a semi-Riemannian manifold with metric g, which

admits an oriented atlas A in its smooth structure. Let M be given the orienta-
tion [volg], and for p E M, TpM be given the orientation [volg(p)]. Show that
volg(p) = volg(p).

Consider the Minkowski spacetime (M, g) from Example 11.3 with the ori-
ented atlas A= {(M,Xp,B)} induced by a choice of a p E Mand an or-
thonormal basis B = {e1, e2, e3, e4} for V with respect to g. We will use the
two notations xp,B = (x 1 , x 2 , x 3 , x 4 ) and xp,B = (x, y, z, t) for the chart map
components of xp,B interchangeably. The metric g is given by specifying its
component matrix in the chart (M, Xp,B) by g(c'\, Oj) = 1 if i = j E {1, 2, 3},
g(oi, oi) = -1 if i = j = 4, and g(oi, oi) = 0 if i =I j. The volume form field
volg ESl 4 Mis given by volg = dx I\ dy I\ dz I\ dt. Let M be given the orienta-
tion [volg]- This amounts to deeming the ordered basis (ox,p, oy,p, Oz,p, Ot,p)
as positively oriented in TpM, at each p EM. We then say the Minkowski
space (M, g) has the standard orientation induced by (p, B).
We will now determine * on Sl 1 M, Sl 2 M and Sl 3 M.
12 Using an analogue of the formula given in Proposition 11.3, but for not necessar-
ily orthonormal basis, see e.g. [Marsden and Ratiu(2007), Example 7.2.14.B-D], the
smoothness of p >-+ (*(O(p)))(o,,;,,P, • • • ,o,,;=-•,p) can be established in any admissible
chart (U,x). By Proposition 4.1, *!1 is a (O,m-k)-tensor field on M. The alternating
property of * !1 follows from the pointwise alternating property.
Any f2 E 0 1 M has the decomposition f2 = ni dxi for ni E C 00 (M), and so

*n = ni * (dxi). So to describe* on 0 1 M, it is enough to determine *(dxi).
We use pointwise Proposition 11.3, giving
*dX = *dx 1 = sign G~: !) g(oy, Oy) g(oz, Oz) g(£7t, (7t) dy A dz A dt
0 0
= (1)(1)(1)(-l)dyAdzAdt = -dyAdzAdt,
*dy = *dx 2 = sign G~: !) g(ox, Ox) g(oz, Oz) g(ot, Ot) dx A dz A dt
= (-1)(1)(1)(-l)dxAdzAdt = dxAdzAdt,
*dz = *dx 3 = sign G~ ~ !) g(ox, Ox) g(oy, Oy) g(ot, Ot) dx Ady A dt
= (1)(1)(1)(-l)dxAdyAdt = -dxAdyAdt,
*dt = *dx4 = sign ( !~ ~ !)g(ox, Ox) g(oy, Oy) g(oz, Oz) dx Ady A dz
= (-l)(l)(l)(l)dxAdyAdz = -dxAdyAdz.
To determine * on 0 2 M, we compute for example
*(dxAdt)=*(dx 1 Adx4 )=sign

. (1234)
1423 g(oy,oy)g(oz,Oz)dyAdz
= (l)(l)(l)dyAdz = dyAdz.
Analogously, one obtains
*(dx Ady)= -dz Adt, *(dx Adz) =dy Adt, *(dy Adz)= -dx Adt,
*(dy Adt) = -dx Adz, *(dz Adt) =dx Ady.
For f2 E 0 2 M, * f2 can be found by decomposing
0= ~ nijdXi Adxi,
where nij E C 00 (M). Finally, we compute for example
*(dx Ady Adt) = *(dx 1 Adx 2 A dx 4 ) = sign G~ !!) g(oz, Oz) dz
= (-l)(l)dz = -dz.
Similarly *(dx Ady Adz)= -dt, *(dx Adz Adt) = dy, *(dy Adz Adt) = -dx.
These determine* on 0 3 M. We also have *(dxAdyAdzAdt) = *VOlg = 1,
the constant function equal to 1 everywhere. ◊
Exercise 11.15. In the same set up as Example 11.5, verify directly that
**n = (-l)k(m-l)+i(g)n
from Exercise 11.13 for all n E {} 3 M.
Chapter 12
Minkowski spacetime physics
In this chapter, we will study Minkowski spacetime physics which served

as a precursor to the development of the geometric theory of gravitation.
At the end of the 19th century, it was realised that the classical (Newto-
nian/Galilean) viewpoint of space and time was inadequate, as it could not
explain experimental observations involving the measurement of the 'speed
of light' c. The revision of the notions of space and time with the 1905
'special relativity theory' by Einstein gave rise to the simplest spacetime,
namely Minkowski 1 spacetime, which we will discuss in this chapter. But
before doing so, we quickly recall the familiar classical viewpoint, in order
to be able to contrast it relatively easily with the Minkowski spacetime.
We remark that most of the definitions and results in this chapter
pertain to objects in the tangent space TpM of a general time-oriented
Lorentzian manifold M at a point p. Even in the case of Lorentzian mani-
folds, the local region around a point resembles Minkowski spacetime, and
the exponential map expP maps a neighbourhood U c TpM of O onto a
neighbourhood expP U of pin M
12.1 Classical spacetime
Definition 12.1. (Classical spacetime).

Classical spacetime is a triple (M, t, r,), where
• M is an affine space over a 4-dimensional real vector space V
• t : V -+ JR is a nonzero linear map
• T/ is an inner product on the vector space ker t.
1 After Hermann Minkowski (1864-1909), who realised that space and time should be
considered as a single whole (a 4-dimensional geometric object called spacetime), rather
than two separate entities, and that special relativity, introduced by Einstein (who was
Minkowski's former student) is best understood in this mathematical setting.
261
M is thought of as the collection of all events, and t is the absolute time.

For p E M, the set of events Sp that are simultaneous with p is defined by
Sp = p + ker t = {q E M : t ( v pq) = 0}.
By the rank-nullity theorem, dim(kert) = 3. For v EV, define the smooth

curve 1'v(t) = p + tv, t E R We had seen that the map i : V -+ TpM
given by i(v) = v'Yv,P' for all v E V, is an isomorphism of vector spaces.
Henceforth we do not write down i, i - 1 explicitly.
Definition 12.2. (Instantaneous observer, space perceived).

We use the same notation as in Definition 12.1. An instantaneous observer
at p E M in classical spacetime is a vector v E TpM such that t( v) > 0. The
space perceived by an instantaneous observer v E TpM is Sp= p + ker t.
We note that Sp is the space perceived by any instantaneous observer at p.

This 'affine subspace' Sp divides the affine space Minto two parts, namely
• the 'future' of events in Sp, consisting of all events q such that t(vpq) > 0,
• the 'past' of events in Sp, consisting of all events q such that t(vpq) <0.
Definition 12.3. (Particle).

We use notation from Definition 12.1. A particle in classical spacetime is a
triple (1', m, e), where 1': J-+ Mis a smooth curve defined on an interval
I c R, such that t(v'Y,'Y(t)) >0 for all t E J, m>0 and e ER The number
m is called the mass of the particle, and e the charge of the particle.
If (1', m, e) is a particle passing through a point p E M with v'Y ,P = w E TpM,

or if w is an instantaneous observer at p, then an instantaneous observer
v E TpM will assign to w a certain relative velocity u, by observing the
motion of the particle or the other observer in its perceived space. We give
Definition 12.4. (Relative velocity).

We use notation from Definition 12.1. Let v E TpM be an instantaneous
observer in classical spacetime. Let w E TpM be such that t(w) > 0. Then
the relative velocity u E ker t of w with respect to an instantaneous observer
v is defined by
W V
u = t(w) - t(v)"
Note that t(u) =t(t~) - t~)) = 1-1=0. So the relativevelocityu Ekert.

Minkowski spacetime physics 263
Also if w = v, then the relative velocity u = 0, as expected. In the rela-

tive velocity expression, the seemingly strange factors t( w) and t( v) in the
denominators are necessary. These arise since the 'proper' time elapsed
between an event panda later event q is given by t(vpq)- Roughly speak-
ing, the curve parametrisation only keeps track of what order the events
along curve are visited, but what is physical, and is 'felt', is the proper time
elapsed between events.
t+ij•: _____ P_:~u~ w

_·_) _·p_+_kw--
So formally, referring to the picture, if the observer 'evolves' from event p

top+ v, then it does so in the time t(v). In the meantime, the particle
starting at p with v'Y,P =w will go to some point q belonging to the spatial
section Sp+v simultaneous with p + v. To determine q, we write q = p + kw,
and solve fork using t(kw) =t(v) to obtain
k = t(v)
t(w) ·
So it makes sense that the relative velocity of the particle as reckoned by
the instantaneous observer v is defined by
kw-v w V
Proposition 12.1. (Galilean addition of relative velocities).

Let (M, t, ry) be a classical spacetime, p EM. Suppose that v, v E TpM are
instantaneous observers, and w E TpM is such that t(w) > 0. Then the
u
relative velocities u, of w with respect to v and v,
respectively, are related
by u = u - x, where
V V
X=-----
t(v) t(v)
is the relative velocity of v with respect to v.
W V ~ W V
Proof. Subtract u = t(w) - t(v) from u = t(w) - t(v) · □
Thus according to the Galilean viewpoint, if v, v are observers such that

t( v) = t(v) = 1, and the relative velocity oflight with respect to the observer
v is c, then the relative velocity of light with respect to the observer v ought
to be c+v-v. However, this was not observed experimentally, notably in a
famous experiment performed by Michelson and Morley in 1887. The speed
of light is the same relative to all observers, and is a universal constant,
namely c=3x10 8 ms- 1. We will see later in this chapter that in Minkowski
spacetime, we obtain a different addition law of relative velocities, which
will automatically render c as a constant.
12.2 Minkowski space
Recall that Minkowski spacetime is a pair (M, g) , where

• M is an affine space over a 4-dimensional real vector space V ,
• M is considered 2 as a smooth manifold, and
• V has a Minkowski scalar product g, inducing3 the metric g on M.
We had introduced the notion of an instantaneous observer in a time-
oriented Lorentzian manifold in Definition 9.8: it is a timelike future-
pointing tangent vector v E TpM. The space perceived by an instantaneous
observer v E TpM is v.L = {w E TpM: g(p)(v, w) =0}.
As noted earlier, in classical spacetime, Sp , the set of all events which
are simultaneous with p, is the 3-dimensional affine space p + ker t, which
divides M into half-spaces, namely, the past of Sp, and the future of Sp·
That Sp is an affine subspace in classical spacetime, is a key difference
to what happens with Minkowski spacetime. In classical spacetime, all
instantaneous observers at an event p E M agree on the set of events that
are simultaneous with p, namely, they agree it is the affine subspace Sp .
future
past "'-
past
(__ ___:.:>
Classical spacetime Minkowski spacetime
2 See Example 1.9, p.9.

3 See Example 5.6, p.85.
On the other hand, in Minkowski spacetime, given an instantaneous ob-

server v E TpM, the set of events that are simultaneous with p is v_j_ , and
depends very much on the observer v. Hence what are the simultaneous
events to one instantaneous observer at p , may not be perceived as be-
ing simultaneous to another observer. This is not really a surprise, since
in Minkowski spacetime, there is no universal 'absolute time' that allows
a universal 'foliation' of the spacetime into 'space sections'. Instead in
Minkowski spacetime, we will see that while all instantaneous observers
agree on what is the future of p , and what is the past of p, they will dispute
on what are events simultaneous with p, as it will depend on their own
'motion' through spacetime.
Example 12.1. (Time dilation). Let (M, g) be Minkowski spacetime. Let

p E M, and eo E V be such that g(eo , eo) = -1 and let e1 E (eo)_j_ be such
that g(e 1,ei) = 1. We endow M with the time-orientation induced by e 0 .
Consider the two observers 'Y, ::Y: ~-+ M,
1(t) = p + tea,
::Y(t) = p +t(aeo+.Be1) ,
where t E ~, and a, ,B E ~ are fixed, and satisfy -a 2 + ,8 2 = -1 and ,B =I 0.
q r
We note that a> 0, since Vf."i(t) is future-pointing for all t E ~ (as ::Y is an
observer) . For some t* > 0, consider the event q = 1(t*) = p + t*e 0 . Then
the instantaneous observer v"f ,q perceives space as q + (e 0 )_!_. In this space,
there is an event r belonging to the worldline of the observer ::Y, that is,
r=::Y(s*). To determines*, we use s*(ae0 +,Be 1) - t*e 0 E (e 0 )_!_, so that
t* = s*a = s * ~ > s*. Now the proper time elapsed between p and q
for the observer 'Y is given by
Jo(· ,J-g(eo,eo)dt = Jo(· ldt = t*.

But the proper time elapsed between p and r for the observer 1 is
r•-J-g(o:ea+,8e1,o:ea+,8e1)dt= f* -J-(-o: 2 +,8 2 )dt= f* ldt = s*.
Thus, although the events q and rare deemed simultaneous by v--y ,q, we note
that 'Y has experienced more proper time to go from p to q than the proper
time experienced by 1 in going from p to r. But this is strange only if we
have not freed ourselves from the shackles of classical-spacetime-thinking.
◊
Example 12.2. (Length contraction). Let (M, g) be Minkowski spacetime.

Let p EM, and ea EV be such that g(ea , ea)= -1 and let e1 Ee~ be such
that g(e 1,ei) = 1. We endow M with the time-orientation induced by ea.
Consider the strip S = {p + aea + be1 : a E JR and lbl,;;; L}. (Imagine a 'rigid
rod' whose atoms have worldlines in M , and the union of these worldlines
constitutes the set S.) Consider the two observers 'Y, 1: JR--> M given by
'Y(t) = p + tea ,
1(t) = p + t(aea+.Be1),
where t E JR, and the fixed a, ,8 E JR satisfy -a 2 + ,8 2 = -1, a> 0, ,8 # 0.
s 'Y
'Y
At p=1(0), the instantaneous observer v1,v=aea+,8e 1 perceives space as

the set p + (aea + ,8 ei).L = {p+w : g(w, a ea+ ,8 e 1 ) = 0} c M, and deems
the rigid rod to be along the spatial section
Sn (p + (aea+,Bei).L) = {p + aea+be1: a E JR, lbl ,;;;L, -aa+b,8=0}
= {p + b(o:- 1,Bea+e1): lbl ,;;;£}.
So the observer 1' at p reckons the rod's length to be
2£ := 2£
The instantaneous observer v'Y,P at p, perceives the rod to lie along the
spatial section {p + b e1 : lbl:::;; L }, and reckons the rod's length to be 2£.
The inequality 2£ < 2£ says that the length of the rod 'measured' by 'Y at
p is strictly greater than that measured by 1' at p. ◊
Now we introduce the notion of a particle.
Definition 12.5. (Particle, mass, charge).

Let (M, g) be a Lorentzian manifold with a time-orientation VE TJ M. A
particle is a triple ('Y, m, e), where 'Y: J-+ Mis a timelike, future-pointing,
smooth curve, defined on an interval I c R, m > 0, e E R. The positive
number m is called the mass of the particle, and the number e is called the
charge of the particle.
For an instantaneous observer v, we now give the definition of the relative

velocity of a particle/another instantaneous observer, as perceived by v.
Definition 12.6. (Relative velocity).

Let (M, g) be a Lorentzian manifold with a time-orientation VE TJ M. Let
p E M and v E TpM be an instantaneous observer at p. Let w E TpM be
timelike and future-pointing. Then the relative velocity u E v1_ of w with
respect to v is defined as
,J-g(p)(v, v) v
U= ~....,.....,.....,....-,--w - ---;====
g(p)(w,v) ,J-g(p)(v,v)
Note that it follows by the reversed Cauchy-Schwarz inequality (p.83) that

since v,w are timelike, (g(p)(v,w)) 2 >0, and in particular nonzero, so that
the above formula for u is well-defined.
Exercise 12.1. We use the same notation as in Definition 12.6. Show that u E v_j_.
Also, prove that u=O if and only if v=cw for some c>O.
The rationale behind this definition is analogous to what was done in the
previous section, although the time elapsed for the instantaneous observer
in going from p top+ vis now given by y1-g(p)(v, v).
Consider the particle 'Y passing through p with w = v-y,p· By referring to the
picture above, we seek the point q = p + kw in the parallel translate of v_j_
(space perceived by v) that passes through p + v. Thus kw - v E v_j_, i.e.,
g(p)(kw-v, v) =0, and so k= :e:N:·,:)-
So it makes sense that the relative
velocity of the particle as reckoned by the instantaneous observer is
kw - v ,J-g(p)(v, v) v
~======-~----W-~=====
-J-g(p)(v, v) g(p)(w, v) -J-g(p)(v, v)
We now show there is a universal speed limit, c:= 1, for relative speeds.
Proposition 12.2. Let (M, g) be a Lorentzian manifold with a time-

orientation given by V E Tc} M. Let p E M. Let v E TpM be an instanta-
neous observer. Let w E TpM be timelike and future-pointing, and let u be
the relative velocity of w with respect to v. Then O,;;; g(p)(u, u) < l.
Proof. Let k := :f:N:·,:)). The relative velocity u of w with respect to vis

kw-v
u=-
✓,=_=g::::;::(p::::;::)::;:::(v=,v:: :;: )
Since kw - v Ev_!_, we have g(p)(kw - v , v) =0. Using this, we obtain

g(p)(kw - v, kw - v) g(p)(kw - v, kw)
g (p )( u u ) = =
' -g(p)(v, v) -g(p)(v, v)
= -k 2 g(w, w) + kg(p)(v, w) g(p)(v, v) · g(p)(w, w) +1
g(v, v) g(p)(v, v) (g(p)(v, w)) 2
{ < 0+1=1,
;:,,-1+1=0
using the fact that as v, ware timelike vectors, the reversed Cauchy-Schwarz
inequality (p.83) gives (g(p)(v, w)) 2 ;:,, g(p)(v, v)·g(p)(w, w) > 0. □
Exercise 12.2. Let us revisit Example 12.2 on length contraction. Determine the
relative velocity u of~ at p with respect to 'Y at p . Also find lul := vg(p)(u, u),
and express the ratio of lengths, ~ ~ , reckoned by the two observers at p, in terms
of lul.
12.3 Relative velocity addition
Without loss of generality, we will assume that instantaneous observers and

particles v E TpM are 'normalised', i.e., g(p)(v,v) = -1, by the following.
Lemma 12.1. (Reparametrisation by proper time).

Let (M, g) be a Lorentzian manifold, and 'Y : J -+ M be a timelike smooth
curve, where I is an open interval. Then there exists an open interval J,
a smooth bijections : J-+ J with a smooth inverse s- 1 : J-+ I such that
7:="( o s- 1 : J-+ I satisfies g(7(a))(v:::;,:::;(a), v"i,"i(u)) = -1 for all a E J.
Proof. Take any toEI. Set s(t)=i\J-g('Y(t))(v'Y,'Y(t),v'Y,Y(t))dt, tEI. So

to
~: (t) = .J-g('Y(t))(v'Y,'Y(t), v'Y,'Y(t)) > 0, t E J.
Thus with J := s(I), it follows thats : J-+ J is a smooth bijection with a

smooth inverse. Defining 7 : J -+ M by :Y(a) =('Yo s- 1 )(a) for a E J, we
have by Exercise 2.4 (p.23) that
d(s- 1 ) 1
v::y,"y(u) = ~(a)v'Y,'Y(s-lu) = ds v'Y,'Y(s-lu)
dt (s-la)
.J-g(7(a))(v'Y,"i(a),v'Y,"i(u))'
and so g((7(a))(v:::;,:::;(a),v"i,"i(o-)) = -1 for all a E J. □

For a normalised instantaneous observer, the relative velocity formula from
Definition 12.6 becomes
w
U= -----v
g(p)(w, v) ·
The following is a recovery formula for the normalised particle velocity,
knowing its relative velocity.
Lemma 12.2. Let (M, g) be a Lorentzian manifold with a time-orientation

given by V E TJ M. Let p E M. Let v E TpM be a normalised instanta-
neous observer, and w E TpM be a timelike, future-pointing vector such that
g(p)(w, w) = -1. If the relative velocity of w with respect to v is u, then
u+v
w= ----;==;::::;:::;====;=
y11 - g(p)(u, u)
Proof. We know that

w
U= _g_(p_)(-w-,v-) - v. (12.1)
Taking scalar product with u E v1_ gives

g(p)(w, u)
g(p)(u,u) = - g(p)(w,v). (12.2)
Also taking scalar product in (12.1) with w yields

1
g(p)(w, u) = ( )( ) - g(p)(w, v). (12.3)
g p w,v
Substituting for g(p)(w, u) from (12.2) into (12.3) gives, after rearranging,
2 1
(g(p)(w,v)) = 1- g(p)(u,u)
Since w, v are both timelike future-pointing, we have4 g(p)(w, v) < 0. Thus,

1
g(p)(w, v) = - y11 - g(p)(u, u)
Substituting this in (12.1), and rearranging proves the result. □

An instantaneous observer v can choose an orthonormal basis {e 1, e2, e3} for
its perceived space v1-, so that ifv =: eo is normalised, then {eo,e1,e2,e3}
forms an orthonormal basis for TpM. A relative velocity u E v1_ acquires
components u1,u 2 ,u3 with respect to the basis {e 1,e 2,e3}, obtained by
decomposing u = u 1e1 +u 2e2+u3e3.
Now suppose we have yet another normalised instantaneous observer v
with a certain nonzero relative velocity x E v1_ with respect to v. Suppose
that v chooses e 1 to be a positive multiple of x. So we then have x = f3e 1
for some f3 E (0, 1). Then by Lemma 12.2,
~ v+f3e1
v=---
../Rfi"
It is straightforward to verify that
~ /3v+e1
e1 := ../Rfi'
belong to v1_ and form an orthonormal basis for it. We have the following
result on the relative addition of velocities.
4 See Exercise 5.4, p.83.
Proposition 12.3. Suppose that (M, g) is a Lorentzian manifold with a

time-orientation V E TJ M. Let p E M. Let v, v E TpM be normalised
instantaneous observers such that the relative velocity of v with respect to v
is equal to f3e1, where e1 E TpM satisfies g(p)(e1, e1) = 1 and /3 E (0, 1). Let
e2, e3 be such that {e1, e2, e3} forms an orthonormal basis for v1_. Define
~ /3v+e1
e1:=~,
Then {e1, e2, e3} forms an orthonormal basis for ;;;1-.

Letw E TpM be timelike, future-pointing, and such that g(p)(w, w) = -1.
Suppose that the relative velocities of w with respect to v, v are, respectively,
u = u 1e1 + u 2e2 + u 3e3, and

u = u1e1 + u2e2 + u3e3.
Then
~1 Ul-/3 ~2 u2~ ~3 u3~
u = l-f3ul' u = l-f3ul ' u = l-f3ul .
Proof. By Lemma 12.2,
Thus
Using the definition of e1 , the components ui in terms of the ui, can now
be read off. □
Now we examine the limiting case where the instantaneous particle velocity
w approaches the light cone. With the same notation as in Proposition 12.3,
let w= e0 + >. e 1 . Then g(p)(w, w) = -1 + >. 2 , and so to make this vector
approach the light cone, we take the limit>.-+ 1 (from below, as wis meant
to be timelike). As we want a normalised w, we take
w eo+>.e1
w = -v-;::::_=g::;:::(p::;::::)(::::;::w=
, w;=.: :) ✓ 1 - >,2 .
The picture shows that the tips of the arrows representing the future-
pointing vector w lie along the top branch of the hyperbola
{(x,y)= ( ~ , k): l>-1<1}

(note that y 2 -x2 = 1), and as>.-+ 1, these vector tips approach the 45°
asymptote (as x/y=>.-+1).
The relative velocity u of w with respect to v is given by

w w eo+>.e1
u = - - - - - - v = - - - - - v = - - - - - - - - - e0 = >.e 1
g(p)( w , v) g(p)( w, v) g(p)( eo + >.e1 , eo) ·
Then u 2 = u 3 = 0 and so u2 = u3 = 0 as well. As u 1 =>.,we obtain
~l A-/3
u = 1-/3>..
In the limit >. -+ 1, we see that u1 -+ 1 too. So the second observer also
reckons that the photon has relative speed 1, irrespective of what /3 is.
We also remark that if the relative speeds are much smaller than the
speed of light, that is, if (3 « 1 and >. « 1, then
A-/3 = (>.-(3 )( 1+>./3+>.2 /3 2 + ... ) ~ >.-(3,
u~l = 1-/3>.
and sou= u1e 1 ~ (>.-(3)e 1 = u - x, where x=(3e 1 is the relative velocity

of v with respect to v. We note that the right-hand side, u - x, is the result
of Galilean addition of relative velocities (Proposition 12.1).
Exercise 12.3. We use the same notation as in Proposition 12.3. Suppose that the
relative velocity of w with respect to vis u = a e1 , where o: E (0, 1) . Show that
the relative velocity of w with respect to v is given by (o: ® /3)e1, where
o:®/3 := o:+/3/3 _
l+o:
(This corresponds to the situation that if a person sitting in a train moving with
speed f3 with respect to the ground, sees a fly go past at a speed o:, then a person
on the ground reckons that the speed of the fly is o: ® /3.)
Given o:, f3 E ( 0, 1) , the number o: ® f3 can be interpreted geometrically as follows.
Draw a unit circle in JR 2 with center O at (0, 0) and vertical diameter AA', where
A= (0, 1) and A'= (0, -1). Join A to (o:, O) and extend it to meet the circle at
B. Similarly, join A' to (/3, O) and extend it to meet the circle at B' . Join BB'
and let it meet the x-axis at P. Show that the x-coordinate of Pis o: ® (3.
Hint: This can be done using coordinates, but we could also use the work already
done in establishing the 3D formula given Exercise 1.2.
Exercise 12.4. Show that the operation® from Exercise 12.3 is the following:
®=((o:, /3) >-+ tanh(tanh- 1o:+tanh- 1f3)),
that is, if o:, f3 E (0, 1) are written as o: = tanh a and /3 = tanh b for some a, b > 0,
then o: ® /3 = tanh(a+b).
Exercise 12.5. A cart rolls on a table with relative speed f3 E (0, 1). A smaller cart
rolls on the first cart in the same direction with a relative speed f3 with respect
to the first cart. A third cart rolls on the second cart in the same direction with
a relative speed f3 with respect to the second cart, and so on, up to n carts.
Determine the relative speed Un of the nth cart with respect to the table.
What is lim Un?
n- oo
12.4 Energy and momentum
In classical spacetime, we think of a particle of mass m and velocity u as

having a momentum m u. When calculating the work done, we take the
inner product of force (the rate of change in momentum) with the elemental
displacement, that is,
mu(t+~t) - mu(t) . (~t)u(t)
~t
So we could view this operation also as follows: First convert u to the
corresponding 1-form by using the flat operator .I>, and act it on the velocity:
(mu(t+~tl - mu(tY')(u(t)).
With this motivation 5 , we give the following definition.
Definition 12.7. (Energy-momentum 1-form of a particle).

Let (M, g) be a Lorentzian manifold with a time-orientation given by
V E TJ- M, and p E M. Let ('Y, m, e) be a particle passing through p, with
g(p)(w, w) = -1, where w = v'Y,P· The energy-momentum 1-form p of the
particle at p is the 1-form
p = mw-P = mg(p)(w,•) E (TpM)*.
For an instantaneous observer v, the instantaneous space is v1_, and so the

momentum of the particle ought to be the restriction of p to v1_. We give
Definition 12.8. (Particle momentum/energy measured by an observer).

Let (M, g) be a Lorentzian manifold with a time-orientation given by
V E TJ- M, and p E M. Let ('Y, m, e) be a particle passing through p, with
g(p)(w, w) = -1, where w = v"f,p, and let p denote the energy-momentum
1-form of the particle at p. Let v E TpM be a normalised instantaneous
observer at p, that is, g(p)(v,v) = -1. Then the
• momentum of w measured by v is Plv_i_
• energy of w measured by vis E = -p(v) = -mg(p)(w,v).
If u E v1_ denotes the relative velocity of w with respect to v, then we have

u+v ) m
E = -mg(p)(w,v) = -mg(p) ( -Jl-g(p)(u,u)'v = -Jl-g(p)(u,u)°
5 There are also further reasons, from the Lagrangian and Hamiltonian approaches to
classical mechanics, and also from quantum mechanics, for defining the momentum using
I-forms; see e.g. [Dodson and Poston(1997), Chap. IX, §2.05].
The reason for calling this the energy can be motivated as follows. Imagine
the scenario when u is much smaller than the speed of light (so that the
classical spacetime picture is approximately valid). Consider the function
f: (-1, 1)-+ IR, given by
1
f (t) = \/1 _ t 2 , t E ( -1, 1).
By a Taylor expansion of f around the point t = 0, we obtain
1 1 2
/,7) = 1 + -2 t + h(t),
vl-t 2
where lim h(3t) =0. So if lul := -Jg(p)(u,u), then for lul«l, we have
t-+O t
E= m = m ~m+ mlul 2-
-Jl-g(p)(u,u) -JI=Tui2 2
We identify the second summand as the kinetic energy possessed by the
particle as measured by the instantaneous observer, and this motivates also
calling E as an energy. We note that if u = 0 (i.e., the particle is at rest
with respect to the instantaneous observer), then E = m. Som is sometimes
referred to as the 'rest energy'.
We remark that if we know Plv_J_ and E, then we can recover p. Choose
an orthonormal basis {e1, e2, e3} for v1_. For i = 1, 2, 3, we define
Pi:= Plv_1_(ei) = mg(p)(w,ei) = mg(p)(v1-:~;(u,u)'ei)

m m -
= ---;::==== g(p) (U, ei) = ---;::====U\
-Jl-g(p)(u,u) -Jl-g(p)(u,u)
where u = u e1 +u e2+u e3. Then for any x := x 1e1 +x 2e2+x3e3
1 2 3 E v1-,
Plv_1_(x) = x 1p1 +x 2p2+x 3p3.
With eo :=v, {eo,e1,e2,e3} forms an orthonormal basis for TpM, so that
any z E TpM has a decomposition z=z 0e0+z 1e 1+z 2e 2+z 3e3, and then
p(z) = p(z 0 v+z 1e1 +z 2e2+z3e3) = -z 0E+z 1p1 +z 2p2+z 3p3
= -z 0 E + Plv_1_(z 1e1 +z 2e2+z3e3).
We note that
-E2 + (p1)2 + (p2)2 + (p3)2
m2 m2((u1)2+(u2)2+(u3)2)
(12.4)
1-g(p)(u,u) + 1-g(p)(u,u)
m 2 2
m g(p)(u, u) 2
--,-,--,------,- + -----,----,-....,... = - m (12.5)
1-g(p)(u,u) 1-g(p)(u,u) ·
Also the calculation in (12.4) above shows that
Thus in the limiting case that the particle relative velocity approaches the
light cone, i.e. , lul = ,Jg(p)(u, u) approaches the speed of light c = 1, we
note that (p 1 ) 2 + (p 2 ) 2 + (p 3 ) 2 = E 2 . But then (12.5) shows that m --> 0,
that is, the light 'particles', i.e. , photons, must have zero mass.
Momentum of light
If we think of light as an electromagnetic wave, then the fact that it carries
momentum can be loosely justified as follows. Consider a charged particle
like an electron with a charge q encountering monochromatic light, assumed
to be an electromagnetic wave, propagating the z-direction. With e x, ey , ez
forming an orthonormal basis for IR 3 , E(x, y , z, t) = E 0 sin(kz - wt)ex and
B(x, y, z, t) = Bo sin(kz - wt)ey, where Eo = cBo, and c is the speed of
light (which is set to 1 with an appropriate choice of units). The electric
field E will move the charge up and down, along the x direction. This
moving charge (with an instantaneous velocity, say Vx, along the positive
x-direction), is subject to a perpendicular magnetic field , namely B, along
the positive y-direction. By the Lorentz force law, the charge experiences
a Lorentz force F(t) due to the magnetic field:
F() _ qvxEosin(kz-wt)
t - ------'----'-ez.
C
y
z
So when light shines on a charge, and the charge oscillates in response

to the light's electromagnetic component, and then the light's magnetic
component pushes the charge in the direction of propagation of the light
beam. This means that light possesses momentum which it transfers to the
electron.
We now give a heuristic argument to show that the pushing momentum

delivered by light per unit time equals ¾ times the energy absorbed from
light per unit time. As everything is oscillating, the time average (F) of
the force is
q(vxEo sin(kz - wt))
C
But the charge times the electric field is the force on the charge, and the
force times the velocity of the charge is the work done per unit time on
the charge. So the pushing momentum per unit time matches ¾times the
energy absorbed per unit time. This lends further credibility to E being
equal (when c = 1) to the size of the relative momentum, which we had
noted earlier).
In the quantum theory of light, one imagines light to behave like a
particle, and the energy of the light-particle is E = liw, where w = 21rv is
the 'angular frequency' of light, and v is the frequency of light (number of
oscillations per unit time if imagined to be an electromagnetic wave), and
Ii is called Planck's constant. As we know that the momentum is then the
energy divided by c, we have that the photons carry a momentum
liw =:lik.
C
One can then do kinematic computations, for example in the photoelectric
effect, Compton effect, etc.
With these preliminaries in mind, we now define the momentum of a
light signal.
Definition 12.9. (Light signal, energy-momentum of a photon).
Let (M, g) be a Lorentzian manifold with the metric g. A light signal is a
geodesic6 A : J-+ M, where I c Mis an interval and for all t E J, V>.,>.(t)
is lightlike and future-pointing. Let A : J -+ M be a lightlike signal that
passes through p E M, and let w = V>.,p• Then the energy-momentum p of
w is defined to be the 1-form p = w~ = g(p)(w,•) E (TpM)*.
As opposed to particle energy-momentum at a point, where we multiply by

m > 0, we do not have this factor now. The energy and momentum of a
light signal at a point measured by a normalised instantaneous observer at
that point are defined exactly as in Definition 12.8.
6 We know that an affine change of parameter will maintain the geodesic nature of
>.., but here when we talk about a light signal, we mean a specific curve, i.e., a spe-
cific parametrisation. This is related to what some observer records to be the fre-
quency /wavelength/energy of the light signal, which 'decides/fixes' a certain parametri-
sation among all feasible parametrisations.
Definition 12.10. (Photon momentum/energy measured by an observer).

Let (M, g) be a Lorentzian manifold with a time orientation VE TJ M.
Let p E M, and .X be a light signal passing through p, with w := V>-.,p E TpM.
Let p := wl> E (TpM)*. Let v E TpM be a normalised instantaneous observer
at p, that is, g(p)(v, v) = -1. The
• momentum of w measured by v is Plv_1_
• energy of w measured by vis E = -p(v) = -g(p)(w, v).
Particles and photons can 'collide' at an event, and can exchange energies
and momenta. In such an event, the total energy-momentum is conserved.
We make this precise below.
Definition 12.11. (Collison, law of conservation of energy-momentum).

Let (M, g) be a Lorentzian manifold with a time-orientation VE TJ M.
A collision at p E M is a collection of
• r incoming particles/photons ai : [ai, O] -+ M, with ai(O) = p, 1 ~i ~r,
• s outgoing particles/photons /3j: [O, bi]-+ M, with /3j(O) =p, 1 ~j ~s.
Additionally, denote the the energy-momentum I-form of ai at p by pi,
1 ~ i ~ r, and the energy-momentum I-form of /3j at p by qi, 1 ~j ~ s.
An allowed collision7 is one for which the following law of conservation
of energy-momentum holds:
r s
I: pi= I: qi.
i=l j=l
Example 12.3. (Compton scattering).

Compton scattering is a process where a photon interacts with an electron,
which results in a change in the photon's frequency /wavelenth. It was an
important historical milestone because it implied that light has a particle
nature. (Thomson scattering is the non-quantum, classical scattering of an
electromagnetic wave by a charged particle, which is incapable of accounting
for the change in wavelength.) We wish to derive Compton's formula for
the scattered photon energy
~ E
E = --=----- (12.6)
l + ;;.(l-cos0)'
where E is the energy of the incident photon and 0 is the 'scattering angle'.
7 We will only consider allowed collisions.
Let (M, g) be Minkowski spacetime with time-orientation V E TJ M. At

p E M, suppose an incoming photon with energy-momentum wl> scatters
off an incoming electron having energy-momentum mevi,, and it produces
an outgoing photon with energy-momentum ii} and an outgoing electron
having energy-momentum mevi,. We have g(p)(v, v) = -1 = g(p)(v, v) and
g(p)(w, w) = 0 = g(p)(w, w).
v
The Compton scattering formula gives a relation between the 'scattering

angle' of the photons as reckoned by the instantaneous observer v E TpM in
its perceived space v1_. In the observer's perceived space v1_ , the incoming
electron has relative velocity O (i.e., it appears stationary), while the photon
is deflected through an angle 0.
VJ_ A
>- o&
/VVVVVVV'o =-------+
e -
e
The 'projections' x, x of w, w, respectively, onto v1-, are the spacelike vectors

x = w + g(p)(w,v)v = w- Ev
X = w + g(p)(w,v)v = w- Ev,
where E, E are the energies of w, w, respectively, measured by v. Since
g(p)(w, w) =0 and g(p)(w, w) =0, -E2 + g(p)(x, x) =0= -E2 + g(p)(x, x).
Define the angle 0 by
cos0 := g(p)(x,x)
vg(p)(x, x)vg(p)(x, x)
So
EEcos0 = (cos0)vg(p)(x,x)vg(p)(x,x) = g(p)(x,x)
= g(p)(w-Ev, w-Ev)
= g(p)(w, w) - Eg(p)(v, w) - Eg(p)(w, v) + EEg(p)(v, v)
= g(p)(w, w) +EE+ EE - EE= g(p)(w, w) + EE. (12.7)
Now we use the law of conservation of energy-momentum to derive an

expression for g(p)(w, w): me VP +wp = me1JP +wP gives meV = mev+w-w.
Thus
-m~ = g(p)(mev, mev) = g(p)(mev+w-w, mev+w-w)
= -m~ + 2meg(p)(v,w) - 2meg(p)(v,w) + g(p)(w-w,w-w)
= -m~ - 2me(E- E) + 0 + 0- 2g(p)(w,w).
Hence we obtain g(p)(w,w) = -me(E - E). Substitution in (12.7) yields
EEcos0 = -me(E - E) +EE.Solving for E yields (12.6). ◊
Exercise 12.6. A positron is a particle with the same mass as an electron, but
with the opposite charge, i.e., if -e is the charge of the electron, then that of
the positron is +e. Show that in a Lorentzian manifold (M, g) with a time-
orientation given by VE TJ M, the annilihation of an electron-positron pair into
a photon is prohibited by the law of conservation of energy-momentum. Hint: For
future-pointing timelike vectors, the scalar product is negative; see Exercise 5.4.
12.5 Electromagnetism
The 4-dimensional viewpoint clarifies how electricity and magnetism are

really one and the same thing, described by a spacetime object called the
'Faraday tensor field' F, which is a 2-form field on M. In this section we
will see that an instantaneous observer v E TpM 'splits' F(p) into an electric
field E and a magnetic field B (and so this splitting is not universal, but very
much dependent on the observer v). The equations determining F from the
'charge-current density' Jin spacetime are the Maxwell equations. For us
now in the definitions below, the Faraday tensor field is just a 2-form field
FonM.
Definition 12.12. (Electric field perceived by an observer).
Let (M, g) be a spacetime. Let FE !1 2 M be the Faraday tensor field. Let
p E M. Let v E TpM be a normalised instantaneous observer. Then the
electric field E perceived by v is defined to be the unique vector E E v1_ such
that F(p)(w,v) = g(p)(w,E) for all w E v1-.
Note that the map v1_ 3 w ...... F(p)(w, v) is JR-linear. So F(p)(·, v) E (v1_ )*.
But then we can apply the sharp operator J : (v1_) * -+ v1_ (associated with
the vector space v1_ with inner product g(p)lv_1_) to get the vector E in v1-.
Thus E = (F(p)(·, v)lv_1_)~ or equivalently, EP = F(p)(·, v)lvL
To define the magnetic field perceived by an instantaneous observer v,
we need an orientation on v1_.
Definition 12.13. (Induced orientation).

Let (M,g) be a spacetime, with an orientation [!1], where n E f2 4 M. For
p E M, the tangent space TpM is given the induced orientation [!1(p)].
Then for an instantaneous observer v E TpM, the induced orientation [wv]
on v_1_ is given by Wv E ;\3(v_1_)*, where wv(x,y,z) = n(p)(x,y,z,v) for all
x,y,z E v_1_.
Wv is nonzero: As 8 TpM = span {v} EB v_1_, we can find an orthonormal basis

{av, e1,e2, e3}, where a= 1/y1-g(p)(v, v) #0 and {e1, e2,e3} is an orthonor-
mal basis for v_1_. Since !1(p) E /\ 4 (TpM)*\{O}, !1(p)(e1,e2,e3,av) # 0.
Hence also wv(e1, e2, e3) # 0.
Definition 12.14. (Magnetic field perceived by an observer).

Let (M, g) be a spacetime. Let FE !1 2 M be the Faraday tensor field. Let
p E M. Let v E TpM be a normalised instantaneous observer. Equip v_1_
with the induced orientation. Then the magnetic field B perceived by v is
defined to be the vector BE v_1_ such that F(p)(x,y) = (*(Bi,))(x,y) for all
x,y E v_1_, where *:/\\v_1_)*-+/\ 2(v_1_)* and _1,:v_1_-+(v_1_)*.
Let us see that the above definition makes sense. Firstly, F(p) E A2(TpM)*,
and so in particular, also the restriction of F (p) to v_1_ x v_1_ belongs to
/\ 2 (v_1_)*. We also know that the restriction of g(p) the vector space v_1_ is
an inner product by Theorem 5.3 (p.78).

Secondly, since the Hodge star operators are isomorphisms 9 , for the
oriented inner product space (v_1_, g(p) Iv_1_, [Wv]), the Hodge star operator
* : (v_1_) * = ;\ 1(v_1_) * -+ ;\2(v_1_) * is an isomorphism. Finally, the map
_i, : v_1_ -+ (v_1_) * is also an isomorphism.
Exercise 12. 7. In the same setting as Definition 12.14, show that Bis alternatively
characterised by F(p)(x,y)=volg(p)(x,y,B,v), where volg(p) is the volume form
on the vector space TpM with the orientation [O(p)] and the scalar product g(p).
Example 12.4. Let (M, g) be the Minkowski spacetime with the standard
orientation induced by (p, B) as in Example 11.5 (p.259). Let f E C 00 (M).
Suppose that F = f dz I\ dx E !1 2 M. For a q E M, consider the normalised
instantaneous observer v = Ot,q E TqM. The electric field E measured by vat
q satisfies F(q)(v,w) = g(q)(E,w) for all w E v_1_. With the decomposition
8 See Theorem 5.2 (p.78).

9 See the remark immediately following Lemma 11.7, p.257.
W=WXOx,q+wYoy,q+wzaz,q, we have
F(q)(w, v) = F(q)(w, Ot,q) = (!dz Adx)(q)(wxax,q+wYoy,q+WzOz,q, Ot,q)
= f(q)(wXdz A dx(ox, Ot)+wYdz Adx(oy, Ot) +wzdz A dx(oz, Ot))(q)
= f(q)(O + 0 + 0) = 0.
So the electric field measured by v is E = 0.
Let v' = ½( .J5ot,q+ Ox,q) be another normalised instantaneous observer.
From above, F(q)(wxax,q+wYoy,q+wzaz,q,Ot,q) = 0 for all wX,wY,wZ ER
Now if w E (v')J_, and we decompose w = WtOt,q + wxax,q+wYoy,q+wzaz,q
for some wt, wx, wY, wz E JR, then we obtain
F(q)(w, v') = 0 + ½F(q)(wxax,q+wYoy,q+wzaz,q, Ox,q)
= ½Udz Adx)(q)(wxax,q+wYoy,q+wzaz,q, Ox,q)
= ½J(q)(O + 0 + Wzdz Adx(oz, Ox))(q)
= ½J(q)wz 1 = g(q)(E', w),
where E' = ½J(q)oz,q· So the electric field measured by v' is E' #0 if f(q) #0.
Consequently, we see explicitly from the two calculations above, that the
electric field perceived at a spacetime point by an instantaneous observer
depends on the observer's state of motion. A similar thing also happens
with the magnetic field. ◊
Proposition 12.4. Let (M, g) be the Minkowski spacetime with the stan-
dard orientation induced by (p, B) as in Example 11.5. Let F E !1 2 M be
the Faraday tensor field. Let q E M. Suppose that the normalised instan-
taneous observer v = Ot,q E TqM perceives the electric field E and magnetic
field B, where
E = Exox,q + EYoy,q + Ezaz,q
B = Bxox,q + BYoy,q + Bzaz,q·
Then the Faraday tensor field at q is given by
F(q) = (ExdxAdt + EYdyAdt + EzdzAdt
+BzdxAdy + BxdyAdz + BYdzAdx)(q).
Proof. The 2-form field F(q) can be decomposed as

F(q) = (FxydxAdy+FxzdxAdz+FxtdXAdt
+FyzdyAdz +FytdyAdt +FztdzAdt)(q).
First we note that v_1_ = span {Ox,q, oy,q, Oz,q}, and the component matrix for
the inner product g(q)lv_j_ with respect to the ordered basis (ox,q, Oy,q, Oz,q)
for v1- is the identity matrix. Thus Bl>= (Bxdx+BYdy+ffdz)q- Using

Proposition 11.3 (p.258), it follows that
*(Bl>)= (BxdyAdz-BYdxAdz+BzdxAdy)(q)
= (Bxdy Adz+BYdz Adx+Bzdx Ady)(q).
Since F(q)lv_j_xv_j_ = *(Bl>), by operating both sides on (ox,q, oy,q), and in light
of the decompositions above, we obtain that F xy = Bz. Similarly, operating
on (oy,q,Oz,q), and on (oz,q,Ox,q) gives Fyz=Bx, respectively-Fxz=BY.
Next to find the F.t components, we will use F(q)(w,v) =g(q)(w,E) for
all w E v1-. Taking w = Ox,q, we obtain F xt = Ex. Similarly, setting w as
oy,q, and as Oz,q, gives F yt = EY, respectively F zt = Ez. Substituting these
components of F(q) in terms of the components of E and B, we obtain
F(q) = (BzdxAdy-BYdxAdz+ExdxAdt
+ Bx dy Adz+ EY dy Adt+ Ez dz Adt)(q). □
Definition 12.15. (Instantaneous Lorentz force).

Let (M, g) be an oriented and time-oriented spacetime, and let p EM. Let
F E !1 2 M be the Faraday tensor field. Let ('Y, m, e) be a particle passing
throughp, with g(p)(w, w) = -1, where w=v-y,p· The instantaneous Lorentz
force experienced by the particle at pis the 1-form eF(p)(·, w) E (TpM)*.
Proposition 12.5. Let (M, g) be an oriented and time-oriented spacetime,
and F E !1 2 M be the Faraday tensor field. Let p EM, and v ETpM be a nor-
malised instantaneous observer, perceiving the electric field E and magnetic
field B. Let ('Y, m, e) be a particle passing through p, with g(p)(w, w) = -1,
where w = v-y,p· If the relative velocity of w with respect to v is u, then the
instantaneous Lorentz force experienced by the particle at p is
. )I = eg(p)(•,E +(*(ul> ABl>))U)
e F(p)( ' w v_j_ ---'--'--'--'-;=============--'".;.......;...
y'l-g(p)(u, u)
Proof. For x E v1_, we have
F(p)(x,w) = F(p)(x, u+v ) = g(p)(x,E)+(*(Bl>))(x,u).

\./1-g(p)(u,u) \./1-g(p)(u,u)
By Lemma 11.3 (p.254),
(*(Bl>))(x,u) = *(Bl> /\XI> Aul>)= *(ul> ABI> Axl>)
= (*(ul> ABl>))(x) = g(p)(x, (*(ul> ABl>))U).
Thus e F(p )( x, w ) = eg(p)(x,E
.1
+ (*(ul> ABl>))U) £ 11
or a x E
1-
v .
v 1-g(p)(u, u) □
Example 12.5. (Minkowski spacetime). Consider the Minkowski space-

time (M, g) with the standard orientation 10 induced by (p, B). Let q EM.
Let v E TqM be the normalised instantaneous observer v = Ot,q· Suppose
that (1,, m, e) is a particle passing through q, w=v-y,q, and g(q)(w, w) = -1.
Let u E v1_ be the relative velocity of w with respect to v. Since we have
VJ_= span{ox,q,Oy,q,Oz,q}, we decompose u = UXOx,q+uYoy,q+uzaz,q and
also B =Bxox,q+BYoy,q+Bzoz,q· Then ub =ux(dx)q+uY(dy)q+uz(dz)q and
Bb =Bx(dx)q+BY(dy)q+Bz(dz)q- Hence ub ABb is
((uxBY-uYBx)(dx Ady)+ (uYBz-uzBY)(dy I\ dz)+ (uzBx-uxBz)(dz Adx))(q).
Thus
*(Ub /\Bb) = (uxBY-uYBx)(dz)q + (uYBz-uzBY)(dx)q + (uzBx-uxBz)(dy)q,
(*(ub I\ Bb))U = (uxBY -uYBX)oz,q + (uYBZ -uzBY)ox,q + (uzBX -uxBz)oy,q•
We recognise the components as the result of taking the usual cross product
-+-+ 3-+ -+ -+
u X Bin JR of u := (ux,uY, uz) and B := (Bx, BY,Bz). If E := (Ex,Ey,Ez),
where the electric field perceived by v is E = Ex Ox,q + EY oy,q + Ez Oz,q, then
the components with respect to the basis (dx)q, (dy)q, (dz)q for (v1_ )*, of
the instantaneous Lorentz force on the particle, are those of
e(E + ux B) E JR . 3
v'l - U---,.· U
---,. ◊
Definition 12.16. (Charge-current density).

Let (M, g) be a spacetime. A vector field J E TJ- M is called a charge-
current density if it satisfies the continuity equation d(*(Jb)) = 0, where
*: f2 1 M =TPM-> {} 3 M.
Example 12.6. (Minkowski spacetime). Consider Minkowski spacetime

(M, g) with the standard orientation induced by (p, B). Let J E TJ- M be a
charge-current density. Let us write the continuity equation in the global
chart (M,xp,B)- Decompose J =jxox+jYoy+Yoz+eot, t,jY,Y, e E C 00 (M).
Then Jb =jxdx + jYdy + Ydz - edt, and so
*(Jb) = - t dy I\ dz I\ dt + jY dx I\ dz I\ dt - Y dx I\ dy I\ dt + e dx I\ dy I\ dz.
Thus, the continuity equation becomes
0 = d(*(Jb)) = (-ox?-oyjY-ozY-ote)dxAdyAdZAdt.
Set j = (jx,jY,Y) o (xp,B)- 1 : JR3 x JR-> JR3 and p= eo (xp,B)- 1 : JR3 x JR-> R
Then we can rewrite the above continuity equation in JR 3 x JR as
-+ - op
v. j + at = o. (12.s)
◊
10 Example 11.5 (p.259)
One starts with a given J, from which the Faraday tensor field is determined
F by solving the Maxwell equations (assuming that the effect of F on the
spacetime (M, g) is negligible).
Definition 12.17. (Maxwell's equations).
Let (M, g) be a spacetime. Let J E TJ M be a charge-current density.
A (0, 2)-form field F E 0 2 M is said to satisfy the 11 Maxwell equations
corresponding to J if
• dF = 0 (closed)
• d*F = *(JP).
F is then called the Faraday tensor field.
The continuity equation is necessary for the second condition since d2 = 0.

Example 12.7. (Minkowski space). We continue with Example 12.6, and
write down the Maxwell equations for F in terms of its components. In
Proposition 12.4, the instantaneous observer v = Ot,q measured from F(q)
the electric field E(q) and the magnetic field B(q) at q. If we vary q,
and then the measurements by instantaneous observers Ot,q give the maps
q ...... E(q), B(q). By Proposition 12.4 applied to each point q EM, we have
F = ExdxAdt + EYdyAdt + EzdzAdt + BzdxAdy + BxdyAdz + BYdzAdx,
where E = Ex Ox + EY Oy + Ez Oz and B = Bx Ox + BY Oy + Bz Oz for smooth
functions Ex' EY' Ez' Bx' BY' Bz E C 00 ( M). Then
*F=ExdyAdz -EYdxAdz +EzdxAdy-BzdzAdt-BxdxAdt -BYdyAdt,
and sod( *F) is given by
~~~/\~/\~+~~~/\~/\&+~~~/\~/\~-~~~/\~/\&
+ OzEZ dx I\ dy I\ dz + OtEZ dx I\ dy I\ dt - OxBZ dx I\ dz I\ dt - oyBZ dy I\ dz I\ dt
+ Oz Bx dx I\ dz I\ dt + OyBX dx I\ dy I\ dt + Oz BY dy I\ dz I\ dt - Ox BY dx I\ dy I\ dt.
Equating this expression for d*F with that for *(JP) found in Example 12.6,
OyBz - OzBY - OtEx = jx }
OzBx - OxBz - OtEY = jY , and OxEx + OyEY + OzEz = (1. (12.9)
OxBY - OyBx - OtEz = Y
Set E=(Ex,Ey,Ez)o(:xp,B)- 1 , and B=(Bx,By,Bz)o(xp,B)- 1 . Then we
can rewrite (12.9) as
-+ -+ - oE -+ -+
v'xB =j + at' and v'·E = p. (12.10)
11 We adopt 'geometrised units', meaning that the usual constants found in the classical
version of the Maxwell PDEs have been absorbed in the charge-current density.
Next we unravel dF = 0 using chart-induced components. Firstly, dF is

-~~~A~A&-~~~A~A&+~~~A~A&-~~~A~A&
+~W~A~A&+~W~A~A&+~W~A~A~+~W~A~A&
+ OxBxdxAdy Adz+ OtBxdyAdz Adt + OyBYdxAdyAdz - OtBYdxAdzAdt.
Thus dF = 0 is equivalent to
OyEz - OzEY + OtBx = 0}
OzEx - OxEz + OtBY = 0 , and OxBx + OyBY + OzBz = 0, (12.11)
OxEY - OyEx + OtBz = 0
which can be rewritten as
-+ -+
v' x E
as
+ 7ft = 0
-+ -+
and v' · B = 0. (12.12)
We recognise (12.10) and (12.12) as the usual Maxwell equations 12 . ◊
Exercise 12.8. Suppose that E, B,j: R 3 x R-+ R 3 and p: R 3 x R-+ R satisfy

the Maxwell equations (12.10), (12.12), and (hence necessarily) the continuity
equation (12.8). Prove that the following wave equations are satisfied:
~2--+ !3- !32--+
u E -+ uj --+ u B --+ --+ -
i}t2 - ~ E = - ot - 'v p and ot2 - ~ B = 'v X j.
Here the Laplacian 13 ~ := ::2 + ::2 + ::2 . If J= 0 and p = 0, then show:

a2
at2
i ---
-~E=O and
a2
B ---
ot -~B=O.
2
Show that E(x, y, z, t) = (sin(z-t), 0, 0) and B(x, y, z, t) = (0, sin(z-t), 0) satisfy
the Maxwell equations when J= 0 and p = 0.
Exercise 12.9. (Reissner-Nordstrom spacetime).
Akin to Schwarzschild spacetime, the Reissner-Nordstrom spacetime is the
smooth manifold M = R x ( r +, oo) x S2, where r + ~ 0. Let (U, cp) be the chart for
S 2 from Example 1.8 (p. 7). Let e ~ 0 and m > 0. In the chart U :=Rx (r +, oo) x U
with the chart map id!R x idcr+,ool x cp, the metric g on Mis given by
2
1
g=-f (r ) dt®dt+ f(r/r®dr+r d0®d0+r 2(sm0
. )2
d</>®d</>,
2m e 2
wheref(r)=l--+-2 .Also,r+:=
r r
{o
m+vm--e-
~
if e 2 >m 2,
·r e2 ~m 2.
1
12 The Maxwell equations were known since the 1860s, but not in the vectorial form
that we learn nowadays. The latter were given by Oliver Heaviside in 1884, reducing
the more than 20 equations to the compactly written 4 equations familiar to us. The
Maxwell equations played an important role in the genesis of the spacetime viewpoint.
Encoded in them was the refutal of the earlier-fundamental 'instantaneous action-at-a-
distance viewpoint' of the operation of forces. Indeed, the Maxwell equations show that
the electric field and the magnetic field behave like 'waves' (see Exercise 12.8), showing
that there is a finite propagation speed, which is the speed of light!
13 ~E means~ acting component-wise on E (and a similar meaning for ~B).
We use spacetime structures on M analogous to the Schwarzschild spacetime:

the time-orientation is Ot, and the orientation is described in Example 11.4
(p.251), with r + replacing 2m. Then (M, g) models the spacetime region out-
side a charged body of mass m and electric charge e. If e = 0, then (M, g) is just
the Schwarzschild spacetime. Consider the vector fields½ E TJ M, given in U by
~ 1 1 1
Vi=vf(r)or, Vi=-00, ½=-.-00<f,, Vi= 17T::'i0t
r rsm y f(r)
and the 'dual' 1-form fields ni E Tf M, given in U by
1
n = v'i(r)dr,
1
n2 =rd0, n3 =r(sin0)d</>, n4 =v~
f(r)dt.
Show that at each p E U, ((Vi)p, (Vi)p, (½)p, (Vi)p) is a positively oriented or-
thonormal basis for TpM with the induced orientation. Define F E Tf M by
F = ~n
r2
1 /\ n4
'
and let J := 0 E TJ M. Prove that F satisfies the Maxwell equations.
Notes
The diagrammatic procedure in Exercise 12.3 was given in [Kocik(2012)]; see
also the exposition [Sasane and Ufnarovski(2016)]. Exercise 12.5 is based on
[Lightman, Press, Price and Teukolsky(1975), Problem 1.4]. The discussion on
the momentum of light is based on [Feynman(1963), §34-9]. Lemma 12.2 and
Proposition 12.3 are based on [Oloff(2018), §5.1].
Chapter 13
Matter
As we had mentioned in the Preface to the book, the field equation is
(geometry) Ric - ½sg + Ag = 81rT. (matter)
We now focus on the right hand side of this equation, and also provide
some motivation for this equation. (The real belief in the equation as the
governing equation for spacetime rests on its ability to predict physical
phenomena, and we have already met two of these tests, namely the red
shift, and the deflection of light. We will also meet a third test, on the
perihelion shift of the orbit of Mercury, where the observed discrepancy with
the Newtonian prediction 1 was correctly accounted for within the spacetime
geometric view of gravitation.) We begin by revisiting Newtonian gravity,
and recalling the Poisson equation, that shows how the matter sources
determine the 'gravitational potential'. This serves as a precursor to the
field equation above.
13.1 Newtonian gravity and the Poisson equation
We suppose that the 4--dimensional classical spacetime is described 2 by

JR x JR3 . The Newtonian gravitation law says that for point masses m, M
located at x, y, respectively, in JR3 , the instantaneous force of attraction
that m feels due to the presence of M is3
mM y-x
F(t, x) = lly-xll 2 lly-xll"
1 The Newtonian prediction was about 532 arcseconds of degree per century (due to
perturbations by the other planets), whereas the observed value was 575 arcseconds per
century, so that there was a discrepancy of 43 arcseconds per century. See §14.4.
2 After making the choice of a point p EM, a vector eo EV such that t(eo) = 1, and
(e,,, ey, ez) forming an orthonormal basis for ker t with respect to the inner product 1/·
3 In units where the universal gravitational constant is set to 1.
289
Let p: Rx R 3 --> R describe the mass density of matter, which we assume,

for the convenience of analysis below, to be a compactly supported smooth
function. The instantaneous force on m is given by the improper integral
f p(t,y) y-x
F(t, x) = m JR.a lly-xll 2 lly-xll dV(y).
Here dV (y) = dxdydz is the usual volume element in R 3 . We note that the
map y ...... lly-xll- 2 is locally integrable on R 3 . If we think of mas a 'test
mass', then the force on m is mg(t, x), where g is the gravitational field:
f p(t,y) y-x
g(t, x) = JJR3 lly-xll 2 lly-xll dV(y).
Let Vx denote the gradient operator in R 3 acting on functions of the variable
x E R 3 . With y E R 3 fixed, for x-# y, we have Vx lly~xll = lli_=-~13. Thus
f p(t,y) y-x f -+ 1
g(t, x) = JJR3 lly-xll 2 lly-xll dV(y) = JJR3 p(t, y) v'x lly-xll dV(y)
-+ f p(t, y) -+
= v'x JR.a lly-xll dV(y) = -v'x (t, x),
where is the gravitational potential, defined by the improper integral
{ p(t, y)
(t, x) = - JR.a IIY - xii dV(y).
We claim that satisfies the Poisson equation
~(t, x) = 41rp(t, x),
where~= ~ x is the Laplacian. Fix f. > 0, and let BE(O) be the ball with
center OE R 3 and radius 1:., with the sphere oBE(O) as its boundary. First,
using the substitution y' =y-x,
{ p(t, y) f p(t, y+x)
(t, x) = - JR.a lly-xll dV(y) = - JR.a IIYII dV(y).
Thus
~x(t, x)
= - f (~xp(t,y+x))-llllldV(y) - f (~xp(t,y+x))-llllldV(y).
JB, (0) Y JJR3\B, (0) Y
=:I, =:J,
(13.1)
Since p is compactly supported and smooth, we have 4
€2
IIEI ~ ll~p(t, ·)IILOO(JR3)247r. (13.2)
4 For a compactly supported smooth function f: JR.3 -+ lRm, llfllL"'(IR•) =maxllf(x)ll-
xeR3
Matter 291
For smooth f,g, V · (gV J) = (Vg) · V f +gtlf. Using the Gauss divergence
theorem 5 , for smooth f, g, either of which is compactly supported,
r gtlJdV(y)=r
JR3\B, (0)
-(vg)-vJdV(y)+r
JR3\B, (0)
v-(gvf)dV(y)
JR3\B, (0)
=- r (v g) -v I dV(y) +
JR3\B, (0)
(gv
JaB, (0)
r n .fiy ds(y),
where ny is the radially inward unit normal on the sphere oB,(O) at the
pointy, and dS(y) is the area element on the sphere. With J(y) = p(y + x)
and g(y) = IIYll- 1 , we get that J, equals
-f (llxp(t,y+x))-llllldV(y) = - f (llyp(t,y+x))-llllldV(y)
JR3\B,(O) Y JR3\B,(O) Y
= f (Vyp(t,y+x))·Vy-llllldV(y) - f -1 1 1 (Vyp(t,y+x))•fiydS(y).
JR3 \B,(O) Y JaB,(O) Y
Call these two last summand integrals as K, and L,, respectively. Firstly,
-+
IL,I:::;; llv'p(t,•)IIL (R3) 00
i oB,(0)
-1 1 lldS(y) = llv'p(t,·)IIL
Y
-+
(R3)4m,.
00 (13.3)
For finding the limiting behaviour of K, as € -> 0, we integrate by parts:
K, = r (Vyp(t,y+x))·Vy-llllldV(y)
JR3\B, (0) Y
= - f p(t,y+x)lly-llllldV(y)+ f Vy·(p(t,y+x)Vy-lllll)dV(y)
JR3\B, (0) Y JR3\B, (0) Y
= - f p(t,y+x)lly-llllldV(y)+ f
JR3\B,(O) Y JaB,(O)
1 11)•nydS(y)
p(t,y+x)(vy-11
Y
=-0+ f p(t,y+x)(vy-lllll)·nydS(y).
JaB,(O) Y
We used tly@ = 0 in JR3 \{0} to get the last equality. As Vy@ = 11;rr3'
f -y -y
K, = JaB,(o) p(t, y+x) IIYll 3 "jjyjf dS(y)
= ioB,(0)
( 1
pt, y+x)-II
l 2 dS(y)
Y
e--+O
---+ 4np(t, x). (13.4)
Taking the limit €-> 0 in (13.1), and using (13.2), (13.3), (13.4), we have
llx(t,x) = lim(J,+J,) = lim(I,+K,+L,) = 0 + 41rp(t,x) + 0 = 41rp(t,x).
E--+0 e--+O
5 See e.g. [Apostol(1969), Thm. 12.6, §12.19].

13.2 Classical dynamics from a geometric viewpoint
In the classical viewpoint, a particle of mass m at a point with coordinates

(t, x) in spacetime, in which there is a pervading gravitational potential ,
will experience the gravitational field of strength -V(t, x), and coupling
to it with its ('gravitational') mass m, feel the force
m(-V(t, x)).
We use coordinates from the previous section. Parametrise the particle
trajectory 'Y : JR-> JR x JR3 using coordinate time, i.e.,
JR 3 t >-+ (t,x(t)) E JR X JR3 .
The force, -mV(t,x(t)), is equal to the ('inertial') mass m times the
acceleration x(t), so that mx(t) = -m V(t,x(t)), that is,
x(t) + °v(t, x(t)) = o. (13.5)
So all masses move along these curves, as famously demonstrated by
Galileo's experiment in which two objects with different masses simulta-
neously hit the floor when let go from the same height.
An intriguing question then arises, namely, whether the above equation
for the trajectory is in fact a geodesic equation in curved spacetime, where
the curvature is created by the matter source p (which determines the 
which appears in the candidate for the geodesic equation). The answer to
this question is 'yes', and it shows that classical spacetime is curved by
mass sources within the framework of classical Newtonian gravity. This
viewpoint also shows that Poisson's equation is really an equation relating
geometry to matter, akin to the field equation, except that it is at odds
with what we have learnt so far in physics: Namely that matter, as well as
signals, ought to be described by future-pointing timelike or lightlike curves,
and so there is a speed limit, the speed of light. But in the Poisson equation,
there is no provision for a speed limit: if the source p changes suddenly,
then it is clear from the definition of that this effect is simultaneously
felt everywhere, no matter how far the location we take. This 'flaw' in
the geometric treatment of the Newtonian gravity will be rectified by the
field equation. Below, we will proceed heuristically and formally, to identify
(13.5) as part of the set of equations for a geodesic in the smooth manifold 6
JR x JR 3 , and then see what the Poisson equation tells us if we put on these
'geometric glasses'.
6 Here we only use the smooth manifold structure of JR x JR 3 , and proceed formally; see
[Schuller(2015), Lecture 9] or [Misner, Thorne and Wheeler(2017), Chapter 12] for a
more careful mathematical treatment.
Matter 293
With 1'0 (t) = t, 1' 1 (t) = x(t), 1'2(t) = y(t), 1'3 (t) = z(t), we obtain
·=r°(t) = 0
·/(t) + !: (1'(t))-y°(t)-y°(t) = 0
·l(t) + !:
o
b(t))-r°(t)-r°(t) = o
·l(t) + oz b(t))-r°(t)-r°(t) = o.
A naive comparison with the equations obtained in Proposition 8.1 (p.156)
suggests that the connection coefficients are all zero except the following,
where (t,x,y,z)=(x 0 ,x1,x2 ,x3 ):
2 o 3 o
roo(t,x) = oy (t,x), r 00 (t,x) = oz (t,x).
Define a corresponding connection "V as in Exercise 6.3 (p.110) with these
coefficient functions. Using Proposition 9.1 (p.182), it can be checked that
for the curvature tensor field R, we have
o2
R~oo = ox2,
Also, R8oo = 0. Thus
Ricoo = R8oo + R~oo + R~oo + R~oo =~·

So the Poisson equation can be rewritten as
Ricoo = 4'11'p.
As mentioned earlier, this equation is not appropriate in the Lorentzian
manifold case, and an appropriate generalisation is sought of (to begin with)
the right-hand side. Here only the mass appears, but we have already seen
that the energy perceived by an instantaneous observer depends on the
observer's state of motion, that is, his own normalised velocity. A wild
guess is to replace the right-hand side by an appropriate (0, 2)-tensor field
T, so that we could maybe have the aesthetically pleasing equation
Ric= 471'T.
It turns out that this train of thought is in the right direction, but it does
not work out exactly like this. We first investigate what an appropriate
generalisation of the right-hand side is.
13.3 Energy-momentum tensor field
We adopt a top-down approach, beginning with an abstract definition, and

then seeing concrete physical examples which make the abstract definition
more amenable.
Definition 13.1. (Energy-momentum tensor field).

Let (M, g) be a spacetime. An energy-momentum tensor field T is a (0, 2)-
tensor field T E TJ M that satisfies
• (symmetry) T(V, W)=T(W, V) for all V, WE TJM,

• (weak energy condition) T(V, V)~0 (pointwise) for all causal7 VETJM,
• (divergence-free) divT=0 E Tf M.
A trivial example of an energy-momentum tensor field is the (0, 2)-tensor

field that is identically zero. We refer to this as vacuum.
Remark 13.1. Where does T come from? In modern physics, the evolu-
tion equations for physical (matter) fields are the Euler-Lagrange equations
for an 'action', which is an integral over spacetime of a Lagrangian. We
have seen the Euler-Lagrange equation in a one-dimensional context in Sec-
tion 8. 7. A similar equation can also be derived in the multivariable setting.
The variation of the Lagrangian with respect to the metric then yields the
energy-momentum tensor field. The vanishing divergence condition is the
energy-momentum conservation or the continuity equation, and in the La-
grangian viewpoint, it arises as a consequence of the invariance of the action
under diffeomorphisms. We refer the reader to [Wald(1984), Appendix E]
for these matters. *
Before looking at examples, we mention that if we imagine a continuous
matter distribution in spacetime, and v E TpM is a normalised instanta-
neous observer, then for this observer, T(p)(v,v) is meant to capture phys-
ically the energy density (energy per unit 'spatial' volume in v_1_). Similar
physical interpretations8 can be given for T(p)(u, w) for u, w E TpM.
7A vector field such that for each p EM, Vp is causal, that is, either timelike or lightlike.
8 E.g.T(p)(x, v) is the momentum density in the direction of the unit vector x Ev_!_, and
T(p)(x, y) is the stress (force component in the x-direction per unit area perpendicular
to they-direction), where x, y are unit vectors in v_!_. For a detailed treatment, we refer
the reader to [Misner, Thorne and Wheeler(2017), Chap. 5].
Matter 295
Matter takes various complicated forms. For example, in the portion of

spacetime where a galaxy evolves, one has
• a swarm of particles (stars constituting the galaxy),
• an inert dust cloud,
• even if matter were absent as massive stuff, there may be a field like the
electromagnetic field, which contributes to the energy density.
The energy momentum tensor field T captures in one go 9 all of these.
Dust
We begin with a model of matter where we imagine non-colliding, non-
interacting particles which are 'freely falling' (so that their worldlines are
geodesics), and close-by particles have approximately parallel worldlines
(so that we can imagine these worldlines as being the integral curves of a
smooth vector field). We give the precise definition below.
Definition 13.2. (Dust).

Let (M, g) be a spacetime. Dust is a triple (m, v, V), where
• m > 0,
• v :M -----> [O, oo) is a smooth map,
•VE TJM satisfiesv'vV=O, and for allpE M, g(p)(Vp, Vv)=-1 with
VP future-pointing,
• div(vV) = 0.
v is called the number density, and V is called the velocity field.
To see what the number density v describes, imagine an observer moving

along an integral curve of V ('co-moving' observer), that passes through
p E M. Then the nearby particles appear stationary to the instantaneous
observer VP in the spatial section (Vp ).L c TpM.
9 In the Lagrangian viewpoint (Remark 13.1) , the energy-momentum tensor field Tis
found from the overall matter-Lagrangian obtained by adding the Lagrangians of the
various disparate matter sources.
v:J_p
In an 'infinitesimal' region R c V/ , then the number of particles in R are
L. vvolg(· , •, •, V).
Definition 13.3. (Dust energy-momentum tensor field) .

Let (M, g) be a spacetime, and (m, v , V) be dust. Then the dust energy-
momentum tensor field T E Tf M is given by T = ml/ vi, ® vi,.
Before we check that this T satisfies all the properties demanded of the
energy-momentum tensor field , let us provide some heuristic motivation for
the above definition by checking that T(p)(v, v) does give the energy density
measured by any normalised instantaneous observer v E TpM. Suppose
that v1 , v2, v 3 E v1_ are linearly independent vectors with 'small' lengths
lvil = vg(p)(vi, vi) « 1, i = 1, 2, 3, and K c V J_ be an infinitesimal
parallelepiped formed by vi, v2 , V 3 .
The 3-volume in v1_ of K is 10 defined by µ(K) = lvolg(p)(v 1 , v2, v 3 , v)I. The

number of particles in K is N = v(p)lvolg(p)(v 1 ,v2,v3 , Vp)I- The energy
measured by v for each particle is E = - mg(p)(v, Vp)- Thus the energy
density measured by v is
NE
µ(K).
10 Since volg(p)(- , •, •,v) = volg(p)(-, ·, •,v) E A3 (v_i)*; see Exercise 11.14, p.259.
Matter 297
But ½, =cv + w, for some c E JR and some vector w E span{v1, v2, v3} =v_1_.
Taking the scalar product with v, we obtain g(p)(Vp, v) = - c + 0, so that
½, = (E/m)v + w. By the linearity and skew-symmetry of volg, it follows
that lvolg(p)(v1, v2, v3, ½,)I = (E/m) lvolg(p)(v1, v2, v3, v)I- So the energy
density measured by v is
::) = v(p)!E = mv(p)(g(p)(v, ½,))2.
We note that this is independent of the choice of the parallelepiped K. On

the other hand, with T: = m V vb ® vb, we have
which matches the energy density determined above.

Now we will check that T is an energy-momentum tensor field in the
sense of Definition 13.1. First we will show the following.
Lemma 13.1. Let (M,g) be a spacetime, f E C 00 (M), and VE TJ-M.

Then div(!Vb ®Vb)= (div(JV))Vb + f(v'vV/.
Proof. For W E TJ- M, we have
(div(!Vb ® Vb))W
= C((n,x) I-+ (v'x(/Vb®Vb))(nU,w))
= C((n,x) 1-+((v'x(/Vb))®Vb)(nu, W))
+C((!1,X) ,_.(!Vb ®v'x(Vb))(nU, W))
= (VbW) C((n, X) I-+ (v' x(/Vb))nU) + f C ((n, X) I-+ Vb(nU)(v' x(Vb))W}
= (VbW)div(!Vb) + /(v'c((n,xJ ...... v•(ni)X)(Vb))W
By Exercise 9.12 (p.190), div(!Vb) = div(JV). Also, in an admissible chart
(U, x), we have that
Hence div(! Vb® Vb)= div(! V)Vb + f v'v(Vb) = div(! V)Vb + f (v'vV)b,
where the last equality follows using Exercise 6.17 (p.126). □
Proposition 13.1. Let (M, g) be a spacetime, and (m, v, V) be dust.

Then T = m v Vb ® Vb is an energy-momentum tensor field.
Proof. Clearly T E Tf M. To show symmetry, we note that for vector

fields X, YE TJM,
T(X, Y) = mv(Vi, ® Vi,)(X, Y) = mvVi,(X)Vi,(Y)
= mvVi,(Y)Vi,(X) = mv(Vi,®Vi,)(Y,X) = T(Y,X).
Also, the weak energy condition is satisfied since for all vector fields (causal
or not) X E TJ M, we have (pointwise)
T(X,X)=mv(Vi, ® Vi,)(X,X)=mv(Vi,(X)) 2 ~0,
since m > 0 and v ~ 0 pointwise. The divergence freeness of T follows
immediately from the previous lemma and the facts that v' v V = 0 and
div(vV) = 0. □
Exercise 13.1. (Dust to no dust).

Let (M, g) be Minkowski spacetime. Let V = Ot (in the notation described in
Example 11.5, p.259), and m>O, v. ~ 0 be given real numbers. Denote by v the
constant function on M assuming the value v. everywhere. Verify that (m, v, V)
is dust, and determine the component matrix [Tii] of the corresponding energy-
momentum tensor field with respect to the basis (ot, Ox, Oy, oz). Show that the
field equation is satisfied if and only if v. = 0, that is, if there is no dust.
Perfect fluid
Yet another example of an energy-momentum tensor field is that of a perfect
fluid, which is a bit more complicated than dust, because now we suppose
that there is also a pressure pat each spacetime point. This pressure arises
from the random thermal motion of the particles in the fluid, and is equal
in all spatial directions for any instantaneous observer. The fluid is 'perfect'
in the sense that there is no interparticle binding or viscosity.
Definition 13.4. (Perfect fluid).

Let (M, g) be a spacetime. A perfect fluid is a triple (p, p, V), where
• p: M--+ (0, oo) and p: M--+ [0, oo) are smooth maps,
• VE TJ Mis such that for all q EM, g(q)(Vq, Vq) = -1, and
Vq is future-pointing,
1 -((Vp)V + (dp)U).
• div(pV)=-pdivV and v'vV = - -
p+p
p is called the pressure, p is called the density, and V is called the velocity
field. The energy-momentum tensor field of a perfect fluid (p, p, V) is
T = (p + p)Vi,®Vi, +pg.
Matter 299
We note that ifp = 0, the energy-momentum tensor field of the perfect fluid
is that of dust. Intuitively, the condition on 'v' v V makes sense, because
'v'v V represents the acceleration of the fluid particles, and so the equation
roughly says that mass density times acceleration is driven by the pressure
gradient, akin to the wind flow on the surface of the Earth. The first
equation, div (p V) = - p div V can likewise be roughly interpreted as saying
that the change in energy density is driven by the work done by the pressure.
The symmetry of the T given above is clear.
Exercise 13.2. With the objects as in Definition 13.4, show that T satisfies the
weak energy condition.
We now check the divergence-freeness.
Lemma 13.2. Let (M, g) be a spacetime, and (p, p, V) be a perfect fluid.

Then divT = 0, where T = (p + p)VI> ® v1> + pg.
Proof. By Lemma 13.1, we have
divT = (div((p + p)V))VI> + (p + p)('v'vVl + div(pg)
1 -((Vp)V +(dp)Ut +div(pg)

= (-pdivV +div(p V))VI> -(p + p)-
p+p
= (-pdivV + div(pV))VI> - (Vp)VI> -dp + div(pg).
In any admissible chart (U, x), we have

div (p V) = dxi('v' a.,i (p V)) = dx\(ox;p)V + p 'v'o.,; V)
= (ox;p)Vi + pdivV = Vp + pdivV.
Also for WE TJ M,
(div(pg))W = ('v'a.,; (pg))((dxi)U, W)
.. k
= (ox;p)g(g'1 oxi, W Oxk) +0 (as 'v'.g = 0)
k .. k.
= (ox;p)W g'3gjk = (ox;p)W 81
= (ox;p)Wi = Wp = (dp)W.
So div(pg) = dp. Consequently,
divT = (-pdivV + div(p V))VI> - (Vp)VI> - dp + div(pg)
= (-pdivV + Vp + pdivV)VI> - (Vp)VI> - dp + dp = 0.
Often, there is also available a relation between the pressure and density,
of the form
p = f(p).
Such is relation is referred to 11 as an equation of state. For example, con-

sider a perfect fluid with the equation of state
p=wp,
where w is a smooth function on M. The case w = 0 corresponds to dust,

where the pressure is identically zero, as we had remarked earlier.
Remark 13.2. (Cosmological constant term arising from matter).

Consider the state equation p = wp with w = -1. Strictly speaking, this
is impossible if p and p are both supposed to be nonnegative. But we
momentarily relax 12 the positivity condition on p in Definition 13.4. Then
in the field equation Ric- ½Sg +Ag= 8nT, the term Ag on the 'geometry
side' can be brought over to the 'matter side', and thought of as the energy-
momentum tensor of a perfect fluid with the equation of state as above with
w = -l, and the pressure p = -s1'1r· *
Finally, we give a heuristic justification of the energy-momentum tensor
field for a perfect fluid. We assume that at any spacetime point q, for any
instantaneous observer v E TqM, the 'gas molecules' have uniformly dis-
tributed random speeds. Suppose that a normalised instantaneous observer
v E TqM chooses an orthonormal basis (e1, e2, e3) for v_1_, and observes a
gas molecule having the relative speed Un =u 0 (q)(n 1 e 1 +n 2 e 2 +n 3 e 3 ), where
n= (n1, n 2 , n 3 ) is a uniformly distributed random point on the unit sphere,
which decides the direction of the relative velocity of the observed molecule.
The uniform distribution assumption means that no spatial direction is pre-
ferred over any other, that is, there is 'isotropy' in the random motion of the
gas molecules. The tangent vector Wn to the worldline of the gas molecule
whose relative speed is perceived to be Un by v is (see Lemma 12.2, p.269)
V +un V + Un v+un
Wn = ---:;========= = ---:======;:==
y'l - g(q)(un, Un) y'l - (uo(q)) 2 n-n y'l-(uo(q)) 2 '
where n-n, the Euclidean inner product in l~.3, is equal to 1. Now if we
imagine each of these molecules as a dust particle (since there is no binding
11 This terminology arises from thermodynamics, where an equation of state relates 'state
variables' such as pressure, volume, temperature, etc.

12 Then the weak energy condition is not required to hold; see the solution to Exer-
cise 13.2.
Matter 301
energy with other molecules), then the energy-momentum tensor (at q)

contribution is given by mv(q) (wnY' ® (wn?, whose component matrix in
the basis (v, e1, e2, e3) for TqM is given by
1 1
y'l-(uo(q)) 2
[T(q)n] := mv(q) n 2 uo(q)
y'l-(uo(q)) 2 y'l-(uo(q)) 2
Intuitively, we expect that the 'effective' contribution of energy-momentum
tensor at q will be the expected value of the above. Denoting the 'expecta-
tion operator' by IE, we have
niuo(q) ) . (nini(uo(q)) 2 ) . .
IE ( l-(uo(q)) 2 =Ofori=l,2,3, and IE l-(uo(q)) 2 =Oforic/=J.
But since n- n = 1, it follows that 1E((n1) 2) =IE((n 2)2) =IE((n3)2) = ½- Thus
[T(q)] :=IE[T(q)n]
1
0 0 0
1-(uo(q)) 2
1 (uo(q)) 2
0 0 0
31-(uo(q)) 2
= mv(q) 1 (uo(q)) 2
0 0 0
31-(uo(q)) 2
1 (uo(q)) 2
0 0 0
3 l-(uo(q)) 2
Let us set
1 1 (uo(q)) 2 1 2
p(q):=mv(q) 1-(uo(q))2' p(q):=mv(q)31-(uo(q))2 = 3p(q)(uo(q)).
Then
[T( )] = [p~q) p~q)

q O
0 0
~ ~
O p(q)
0 p(q)
0
l .
On the other hand, we have that the components with respect to the basis
(Vq =:v, e1, e2, e3), of the (0, 2)-tensor ((p + p)VI> © v1> + pg)(q) E Tf M(q),
are given by the same matrix above. This motivates the Definition 13.4.
Remark 13.3. (Radiation or 'photon gas').

Let us consider the limiting case u 0 (q) -+ 1 in the above calculation, when
the perfect fluid particles have the relative speed 1 for the instantaneous
observer v = Vq E TqM, i.e., we think of the particles as photons. Suppose
that v behaves in such a way we have a finite limit
1
p(q) = mv(q\-(uo(q))2 -+ Po(q) as (uo(q))2-+ l.
Then necessarily
1 (uo(q)) 2 1 1
p(q) = mv(q)31- (uo(q))2 = 3p(q)(uo(q))2-+ 3Po(q) as (uo(q))2-+ 1.
In light of this, we take the energy-momentum tensor field for 'radiation'
as that of a perfect fluid with p = ½P (so that w = ½in p = wp). *
Consider the FLRW spacetime M =Ix R 3 , where I= (0, oo ), and with the
metric g, time-orientation, and orientation as in Example 5.9 (p.89), Ex-
ample 5.13 (p.92) and Example 11.2 (p.239). Recall also that in the
chart (M, idixIRa ), we had determined the Ricci curvature tensor field and
the scalar curvature function in Section 9.2. Suppose that the energy-
momentum tensor field is given by a perfect fluid (p, p, V) with V = Ot-
We will now obtain two differential equations involving a, p, p, called the
Friedmann equations 13 , which follow from the field equations, and play an
important role in cosmology. Recall that in the chart basis (ot, ox, oy, oz),
-3~, and Rieu= Ric22 = Ric33 = aa + 2a.2,

Ricoo =
a
and the other components are all zeroes. Also, the scalar curvature is
.. •2
S= 6 aa+a
a2
The 00-component of the field equation Ric - -, + Ag = 8nT then gives
a
-3~ -
1 aa a+ a2 (-1) + A(-1) = 8n((p + p)(-1)(-1) + p(-1)),
26 2
and upon simplification, this gives the first Friedmann equation,
a 2 A 8np
a2 - 3 = 3· (13.6)
The 11-component (also 22-, 33-component) of the field equation yields
aa + 2a 2 -½6 aa; a 2 (a 2) + A(a2) = 8n((p + p)(O) + p(a2)).

1 3In the 'flat spatial' case.
Matter 303
Thus
a a2
-2- - - +A= 81rp. (13.7)
a a2
·2
Solving for ~, and substituting the resulting expression in the first Fried-
mann equation (13.6), results in the second Friedmann equation,
a A 41r
- - - = --(p+3p). (13.8)
a 3 3
The other mixed components in the field equation just reduce to O = 0. ◊
Exercise 13.3. (First law of thermodynamics for FLRW spacetime).

We use notation from Example 13.1. Let p and p depend only on t. Show that
d 3 d 3
d/pa ) +pd/a ) = 0.
We remark that setting
dE = d(p ~ a 3 ) (think mass-energy change in a sphere of radius a(t)), and
dV = d( 4; a3) (think change in volume),

we can formally write the above as dE + pdV = 0, reminiscent of the first law of
thermodynamics (law of conservation of energy): The work done by pressure in
the expansion accounts for the change in the mass-energy in the volume.
Remark 13.4. (Other examples of energy-momentum tensor fields).

We mention two more commonly encountered examples of energy-
momentum tensor fields, without details.
(a) Electromagnetic energy-momentum tensor field: The Maxwell equa-

tions can be derived as the Euler-Lagrange equations for an 'electro-
magnetic action' involving a Lagrangian for the electromagnetic field,
and using the procedure explained in Remark 13.1, the corresponding
energy-momentum tensor field T can be derived, which is given in an
admissible chart (U, x) by
where F E Sl 2 M is the Faraday tensor field. The interested reader is

referred to [Sachs and Wu(1977), §3.7] and [Kriele(2001), §5.3.1] for
a detailed treatment of the electromagnetic energy-momentum tensor
field. We will consider an example in Exercise 13.4.
(b) Klein-Gordon scalar field: The Klein-Gordon equation for a scalar field
¢ : M - JR is given by
O¢-m2 ¢ = 0,
where O¢:=div((d¢)U). In an admissible chart (U,x), this becomes
.. .. k 2
g'3 OxiOxi</J - g'3 rij0xk¢ - m ¢ = 0.
The Klein-Gordon equation is the Euler-Lagrange equation for the
'Klein-Gordon action'. The corresponding energy-momentum tensor
field T is given by
T = (d¢) ® (d¢) -½(g((d¢)U, (d¢)U) + m 2 ¢ 2 ) g.

We refer the reader to [Wald(1984), §4.3] for further details. *
Exercise 13.4. (Energy-momentum tensor field in Reissner-Nordstrom spacetime).
Recall from Exercise 12.9 (p.286) the Reissner-Nordstrom spacetime (M, g), the
1-form fields ni E TI' M (i = 1, 2, 3, 4) defined there, and the Faraday tensor field
F = ~0 1 A 0 4 . The expression for the electromagnetic energy-momentum tensor
field TE T~ M given in Remark 13.4(a), for this F and g, is
2
T = _e_(n4 ® n4 - n1 ® n1 + n2 ® n2 + n3 ® n3).
81rr 4
Clearly T is symmetric. Show that T satisfies the weak energy condition.
(In Exercise 14.9, p.318, we will check that, with the cosmological constant A= 0,
the field equation Ric - ½Sg + Ag = T is satisfied. By Corollary 9.1, p.191,
we have divT = div(Ric - ½Sg) = 0. For a direct proof of divT = 0 using
the Maxwell equations, where T is the general expression for electromagnetic
energy-momentum tensor field T given in Remark 13.4(a), we refer the reader to
[Kriele(2001), Lemma 5.2.4, p.268].)
Chapter 14
Field equation
In the last chapter, we saw that the energy-momentum tensor field Ton a
spacetime (M, g) is assumed to satisfy
divT = 0.
If this is to match with a (geometric) (0, 2)-tensor field, then its divergence
ought to be zero too, and we had seen 1 one such tensor field in Chapter 9:
div (Ric-½Sg +Ag)= 0,
for any constant A. The field equation is postulated2 to be
Ric - ½sg + Ag = 8nT.

The constant A is called the cosmological constant. This equation cannot
be derived, but as with other physical theories, it is accepted in the sense of
being the most powerful unfalsified model. Indeed, it accommodates all that
the Newtonian theory has to say, but also supersedes it, since (for example)
it provides an explanation of the perihelion precession of Mercury's orbit
accurately.
Solving the field equation amounts to finding the metric g such that the
field equation is satisfied. Thus if we start with a spacetime model as a
certain smooth manifold, and we have also assumed knowledge of T (i.e., a
matter model), then in an admissible chart (U, x), the chart-induced com-
ponents gij(x- 1 (-)) of g satisfy the 10 second order nonlinear partial differ-
ential equations obtained by expressing the field equation component-wise.
As both sides of the field equation are symmetric tensor fields, there are not
16, but only 10 'independent' scalar PDEs. Determining the trajectories of
1 SeeCorollary 9.1, p.191.
2 Thiswas published first by Einstein with A= 0 in 1915. David Hilbert found the field
equations almost simultaneously with Einstein based on a variational principle. Hilbert
maintained that priority is due to Einstein, and that his investigation was founded on
Einstein's previous work. The constant A was introduced later by Einstein in 1917.
305
test particles (e.g. a planet in a solar system) amounts to determining the

geodesics. Although these are complicated problems, one usually has a lot
of symmetry at hand, allowing much simplification, and the possibility of
obtaining exact solutions, or very good approximations.
Exercise 14.1. Let the cosmological constant be A=0. Recall the definition of the
trace of a (0, 2)-tensor field from Exercise 5.24 (p.105). By the definition of the
scalar curvature, S=traceRic. Show that the field equation is equivalent to
Ric= 87r(T-½(traceT)g).
Exercise 14.2. Let T be the energy-momentum tensor field on a spacetime M.

A vector field is called lightlike (respectively timelike) if at each p E M, Vp is
lightlike (repectively timelike). Using the weak energy condition on T and the
field equation, show that Ric(V, V) ~ 0 for all lightlike vector fields V E TJ M.
This is called the null convergence condition, and is one of the hypotheses of
Penrose's singularity theorem, which establishes geodesic incompleteness in any
black hole for matter satisfying reasonable energy conditions. We say that T
satisfies the strong energy condition if for all timelike vector fields V E TJ M,
1
T(V, V) ~ 2(traceT) g(V, V).
Prove that if T satisfies the strong energy condition, and if the cosmological
constant is A= 0, then Ric(V, V) ~ 0 for all timelike vector fields V E TJ M. This
is called the timelike convergence condition, and is one of the hypotheses of the
Hawking singularity theorem, which, roughly speaking, says that any expanding
spacetime satisfying physically reasonable conditions, is 'singular', that is, it is not
geodesically complete. Prior to such singularity theorems of Penrose and Hawking
from the 1960s, it was believed that spacetimes were singular only in highly
symmetric situations, while more 'realistic' spacetimes would not be singular.
Exercise 14.3. (Plane gravitational waves).

The aim of this exercise is to consider an idealised spacetime modelling gravita-
tional radiation in a region far away from the source of the radiation (e.g. the
source may be a binary system of stars rotating about their mutual centre of
mass). We suppose that M = R 4 with the standard smooth structure. Using
the global chart (R4, id), we give M the time-orientation V = Ot, the orientation
[dx I\ dy I\ dz I\ dt], and the metric g ='I]+ 2H(x, y, t- z) (dt-dz) (8) (dt-dz),
where H : R 3 -+ R is a smooth function, and 'I] denotes the Minkowski metric,
T/ = -dt ® dt + dx ® dx + dy ® dy +dz® dz. Set Y = Ot + oz
E TJ M. Show that
g(Y, Y) = 0. Show that g is a Lorentzian metric by computing the component
matrix [g(½, ½)], where, denoting h:=H(x,y,t- z),
Vi1_- 2h+3,, 2h+l,, V _ -2h+l,, -2h+3,, V ,, TT ,,
2y'2ut+ 2y'2uz, 2-
2y'2 ut+ 2y'2 Uz, 3=ux, v4=uy.
It can be shown3 that the Riemann curvature tensor field is nonzero (so that the
3 See for example [Sachs and Wu(1977), §7.6] or [Beem, Ehrlich and Easley(1996), §13.1].
Field equation 307
spacetime (M, g) is 'not flat'), and that the Ricci curvature tensor field is given
by Ric= -(ox(oxh) +oy(oyh))(dt-dz)®(dt-dz). Show that the scalar curvature
S = 0. Assuming that the cosmological constant is A= 0, prove that the vacuum
field equation is satisfied if and only if his harmonic in the x, y variables, that is,
02 h 02 h
iJx2 + iJy2 = 0.
Let f, g : R -+ R be smooth functions such that /2 + g 2 is not identically 0. Set
f(u) 2 2
H(x,y,u) = - 2-(x -y ) + g(u)xy.
Check that the resulting h is harmonic. The spacetime (M, g) is called a plane
gravitational wave.
Exercise 14.4. (Godel spacetime).

A first example of a spacetime solution (M, g) to the field equation that allowed
the existence of closed timelike curves was given by Kurt Godel4 in 1949. The
Lorentzian manifold M is time-orientable, and the existence of timelike closed
loops means that there are worldlines along which a particle, while always trav-
elling into the future, ends up back in the past. The smooth manifold M is R 4
with the standard smooth structure. The metric g in the global chart (R4 , id) is
e2x
g = -dt®dt + dx®dx - 2 dy®dy + dz®dz -exdt®dy-exdy®dt.
By 'completing squares' in dy, dt, show that g is Lorentzian. Show that V := Ot
is timelike everywhere. We use the standard orientation [dx I\ dy I\ dz I\ dt] on
R 4 and the time-orientation is given by V = Ot. In the Godel spacetime, the
cosmological constant 5 is A=-½< 0. The matter energy-momentum tensor field
T is given by T = 8~ v• ® v•. Check that this is dust. It can be shown 6 that the
Ricci curvature tensor field is given by Ric = (Ot) • ® (Ot) •. Determine the scalar
curvature, and show that the field equation is satisfied. It can be shown 7 that
the following curve 'Y is timelike: 7(s) = (t(s), x(s), y(s), z(s)), s E (-71", 71"), where,
with d:=log(2+v'2+-J5+4v'2),
t(s) = 2-V2 tan- 1 (e-dtan~) - -V2s,
x(s) = log((coshd) + (coss)sinhd),
-V2 (sins) sinh d
y(s) = (coshd) + (coss)sinhd'
z(s) = 0.
'Y has a continuous extension to [-71", 71"]. Check that 'Y is closed by showing
7(11") = (0,d,0,0) = 7(-11"). Suppressing the z-coordinate, use the computer to
plot the curve R 3 s >-+ (t(s),x(s),y(s)) E R 3 •
4 Logician and mathematician, 1906-78, reputed for his 1931 incompleteness theorems
in mathematical logic.
5 Note that A< 0, as opposed to A> 0 in the current FLRW model of our spacetime.
6 See for example [Kriele(2001), Proposition 9.4.1].
7 See for example [Cooke(2017), p.319] and also [(Momin)].
14.1 Newtonian limit
Recall from §13.1 that in classical Newtonian spacetime8 M =Rx R 3 , test

matter moves according to the equation
x(t) + °v(t, x(t)) = o,
where
~x(t, x) = 411"p(t, x),
where p describes the mass density distribution. We wish to show that for
classical systems, so that the particles have velocities much smaller than the
speed of light, and the objects are not too massive, the geodesic equation
and the field equation reduce to the above (approximately).
We consider the spacetime (M, g), where M =Rx R 3 is equipped with
the standard smooth structure and a metric g. We use the global chart
(M,id = (t,x,y,z)). We will also use the notation (x 0 ,x1,x2 ,x3 ) for the
component functions of this (identity) chart map. In our approximation,
Ot can be regarded as a unit vector. Consider a simple matter distribution
given by T=pVl>®VI>, where V=ot. Then the chart-induced components
of T are described by the entries of the matrix
We expect that if the objects are not too massive, then the metric differs
only slightly from the flat Minkowski spacetime metric described by 'T/ from
Example 5.6 (p.85). So defining
hii = gii - 'T/ii,
we assume that lhiil « 1 and loxihjkl « 1. We also suppose that we have a
'stationary' spacetime metric where the components of g (and hence also
those of h) do not depend on t. If G= [gij], Y =[%],and H = [hij], then
Y=Y- 1 , and
a- 1 = (Y +H)- 1 = (Y(I +YH))- 1 = (I +YH)- 1 Y
= (I -Y H + (Y H) 2 - + · · ·) Y ~ Y - Y HY.
So gii = 'T/ij-'T/ik hke'T/ej- Note that traceT= T((dxi)~, Oxi) = Tjigii = pg00 .
We will need this below (when we use the field equation in the form given
in Exercise 14.1).
8M can be identified with li x li3 as described in the footnote on page 289.
Field equation 309
In the geodesic equation, which we will soon also use, we shall need the
Christoffel symbols. The Christoffel symbols are given by (suspending the
summation convention)
3 gfk 3 TJ
rtj = L 2 (Oxigje+Oxigei- Oxlgij) ~ L ;k (Oxihje+Oxihie- Oxihij)-
l=O l=O
Also, we note that thanks to the stationarity assumption, for i = 1, 2, 3,
rt0 = r~i ~ T/;£ Oxiho1,. (14.1)
Neglecting the rr terms in the expression from Proposition 9.1 (p.182) for
the curvature tensor field components, we obtain
R R,ijk ~ ::i -r£
~ Ux• ::i -r£
kj - Ux1 ki·
Hence Ricjk = RLk ~ Oxirtj - Oxirti, and in particular,
Ricoo ~ OxSbo - Ot rii = Oxirbo - 0 = Oxirbo,
where we used Ot r&i = 0 since the spacetime metric is 'stationary'.
Newtonian equation of motion from the geodesic equation

Consider, in the Newtonian sense, the motion of a slow-moving particle,
falling freely in the 'gravitational field' created by p. We can parametrise9
the particle's worldline using the x 0 = t coordinate function, and obtain a
curve 'Y: I-> M, where I c JR is an interval, and 'Y(t)=(t,x(t)), t EI. We
have ,yO(t) = 1, 71 (t) = x(t), 72 (t) = y(t), 73 (t) = z(t). By the Newtonian
'slow-moving' assumption, we mean l±(t)I, liJ(t)I, lz(t)I « 1 for all t EI. Let
T denote the proper time recorded starting from the event 'Y(c), for some
CE J:
r('Y(t)) = f ,J-g('y(t))(v,,,(t),v,,,(t)) dt ~ f v'-(-1) dt = t- c.

Thus t is aflinely related to the proper time. But as the particle is freely
falling, there exists a reparametrisation of 'Y, namely a map h : 'f -> I,
such that 1' := 'Yoh : 'f-> M is a geodesic. From Example 8.3 (p.163),
we may assume that 1' is parametrised by proper time. But because any
affine reparametrisation of 1' is still a geodesic, we conclude that 'Y itself is
a geodesic. Thus 'Y satisfies the geodesic equation
;l(t) + rtjb(t))'Y\t)'Yi(t) ~ o.
-------------
9
If µ : J M is any parametrisation of the worldline of the particle, then since the map
-+
J a ~ t(µ(a)) is strictly increasing, it has an inverse k- 1

3 : I -+ J, where I:= k(J).
Then we set"(=µ o k- 1 : I-+ M.
For i, j = 1, 2, 3, we neglect terms containing ')'i(t)'yi (t) or I'fo'Yi(t) by (14.1).

So i'k (t) + r~0 · 1- 1 ~ 0. The x-component is x( t) + rb 0 = 0, where
1 1 1
r 00 ~ 2(otho1 + oth01 - oxhoo) ~ - 2 oxhoo-
Similarly, we obtain expressions for they- and z-components. Altogether,
X. + --+(
y' - 2lh00 ) = 0.
This coincides with the Newtonian equation of motion x + V = 0, if
hoo = -2,
(up to an additive constant). So the Newtonian picture would be complete
if we manage to show that -½hoo is in fact the solution to the Poisson
equation D. = 41rp (which, in Newtonian gravity, is the equation that
describes how the matter source p produces a gravitational potential ,
whose gradient -V then gives the gravitational field). This is done below.
Newtonian gravitational potential from the field equation

By Exercise 14.1, the field equation (assuming A=0) is
Ric = 81r ( T - ½(trace T) g) . (14.2)

We had seen that trace T = p g00 . It follows that the 00-component of the
tensor field on the right-hand side is given approximately by
81r(p - ½p(-1- hoo)(-1 + hoo)) ~ 41rp.

On the other hand, the 00-component of the Ricci curvature tensor field is
Ricoo ~ OxiI'bo = OtI'8o + OxI'Ao + OyI'5o + ozrgo
~ 0 + ox ( - ½oxhoo) + oy ( - ½oyhoo) + oz ( - ½ozhoo)
= D-x( -½hoo)-
Here we used the fact that the time derivatives of the components of h are
zero (our assumption that the spacetime metric is 'stationary'). But while
looking at the equation of motion, we had identified ½h 00 = , and so we
have Ric00 = D-x. Thus the 00-component of the field equation (14.2)
above delivers the Poisson's equation in Newtonian gravity,
D-x = 41rp,
as wanted.
Field equation 311
Exercise 14.5. (Linearised field equation). Consider M=R4 as a smooth manifold

with the standard smooth structure. Suppose we have a family {g<•l: s E ( a, b)}
of Lorentzian metrics g<•l such that M x (a, b) 3 (p, s) >-+ gii(P, s) := (g<•l),;(P) is
smooth, where we use the global chart (R4,x=id). Let OE (a,b), and let g< 0 l be
the Minkowski metric, i.e., g;Jl = 'f/;;; see Example 5.6 (p.85). Define h E Tf M by
h := ogij(•, s) I dxi ® dxi.
OS s=O
Thus g<•l R< g(O) + sh for small Isl- Show that
0 (g<•l)ij
0 I = _,,,ikhkt'f/lj•
S s=O
For a fixed s E (a, b), let r<•l:i denote the Christoffel symbols corresponding to
the Levi-Civita connection on M induced by the metric g<•l. Show that
or<•J~j I ,,,kl
-"-- := -2 (Oxihjt + Ox;hu - Oxihij)- (*)
us s=O
Suppose that the cosmological constant is A= 0, the energy-momentum tensor
field is T =0, and that each of the metrics g<•l, for s E (a, b), satisfies the vacuum
field equation Ric<•l =0 (where the superscript '(s)' indicates the dependence of
the Ricci curvature tensor field on the parameter s). Differentiating this with
respect to s and using the above, one obtains, after setting s = 0, that
oRic(s) I = 0.
OS s=O
Using the expression for the Ricci curvature tensor in terms of the Christoffel
symbols, and (*) above, one obtains an equation for h, which is referred to as the
linearised field equation in vacuum:
'f/kl(OxkOxihjt + OxkOx;hu - OxiOxihkt - OxkOxthij) = 0.
Exercise 14.6. (Gauge transformations). Consider M =R4 as a smooth manifold

with the standard smooth structure. Let g be a Lorentzian metric. Let V E TJ M
be a complete vector field with a flow {'¢,. : s E R}. Define g<•l := 'lf;!g for s E R.
Show that if
V og<~) I . .
h :=~ dx'®dx 1 ,
us s=O
then h v = .Cvg. Now suppose that (M, g) is Ricci-flat, that is, it satisfies the
vacuum field equation with the cosmological constant A = 0. Then it can be
shown that (M, g<•l) is also Ricci-flat. Thus .Cvg satisfies the linearised field
equation. We conclude that a solution h to the linearised field equation is not
unique: h+.Cvg< 0 l also satisfies the linearised field equation for arbitrary complete
VE TJ M. Such transformations of hare called gauge tmnsformations.
Remark. (Gravitational waves). By an appropriate choice of gauge, one gets
a wave equation for the perturbation h of g< 0l, namely □ hij = 0, where the
wave operator □ :='f/ijOxiOxi• For time-independent perturbations one gets the
Laplace equation for h, which is consistent with the Poisson equation for p = 0.
The perturbation is thought of as a 'gravitational wave' around a nominal metric
g(O). The exact gravitational wave in Exercise 14.3 is actually a solution of the
linearised field equation. See e.g. [Misner, Thorne and Wheeler(2017), Part VIII]
and [Natario(2021), §6.4] for a detailed exposition of gravitational waves.
14.2 Some symmetries of spacetime
In this section we present a spacetime which can be thought of as being

'(spatially) spherically symmetric' and 'stationary'. We begin with the
following definitions.
Definition 14.1. (Symmetry of a spacetime; stationary metric).
Let (M, g) be a spacetime. A symmetry of (M, g) is a Killing vector field
VE TJ M for g, that is, .Cvg = 0. The spacetime metric g is called stationary
if there exists a symmetry VE TJ M of (M, g) which is everywhere timelike,
that is, g(p)(Vp, Vp) < 0 for all p EM.
A spherically symmetric stationary spacetime metric

Let us consider the smooth manifold M = R x n, where
n= {(x,y,z) E R3 I Tmin< r:=-\ix 2 +y 2 +z 2 },
where Tmin ~ 0. Then Mis diffeomorphically 10 identified with the smooth
manifold Rx(rmin,oo)x8 2 . We use the global chart (M,id). We will also
use the notation (x 0 , x1, x 2 , x 3 ) or (t, x, y, z) for the component functions
of this (identity) chart map. The smooth manifold Mis given the metric
g = -f(r)dt ©dt + (k(r)-l)dr®dr + dx®dx + dy© dy + dz ©dz,
where f (r), k(r) > 0 for all r > Tmin· If we use the usual spherical polar
coordinates in n, then the metric dx ® dx + dy © dy + dz © dz on n can
be expressed (see Exercise 5.12, p.85) in the spherical polar coordinates
(r, 0, ¢) as dr®dr +r 2 d0©d0 +r 2 (sin0) 2 dcp©d¢. Consequently, the metric
g on M in the (t, r, 0, q> )-coordinates is given by
g = -f(r)dt®dt + k(r)dr ©dr + r 2 d0®d0 + r 2 (sin0)2dcp©d¢.
It is clear that (M, g) is a Lorentzian manifold since f (r), k(r) > 0 for all
r > Tmin· The time-orientation on Mis V = Ot E TJ M.
Lemma 14.1. For (M,g) as above, .Ca.g=0, that is, the spacetime metric
is stationary.
Proof. We use the chart (M, id). Then none of the components of g have
t-dependence. Indeed, dr = (l/r)(xdx + ydy + zdz), so that
dr ® dr = r- 2 (xdx + ydy + zdz) ® (xdx + ydy + zdz).
Inserting this in the expression for g, and collecting like terms dxi ® dxi,
we see that the coefficients of these do not have any t-dependence.
10 Under the diffeomorphism M 3 (t, (x, y, z)) >-+ (t, r, (x,~,z) ), where r= -Jx 2 +y 2 +z 2 •
Field equation 313
Using Ot(gij) = 0 and [ox;, Oxi] = 0, we obtain

(.Ca,g)(oxi, oxi) = 8t(g(oxi, oxi)) - g([ 8t, oxi], oxi) - g(oxi, [ot, oxi])
=ot(gij)-0-0=0.
□
We recall from Exercise 3.21 (p.57) the three vector fields Lx, Ly, Lz, which
can be thought of the 'generators of rotations', since their flow maps were
found to be rotations about the x-, y-, z-axis, respectively. Thus in light of
Theorem 6.3 (p.134), the vanishing of the Lie derivatives of g with respect
to these three vector fields means that (M, g) is 'spherically symmetric'.
Lemma 14.2. If (M, g) is as above, and the vector fields Lx, Ly, Lz E TJ M
are given by
Lx=YOz-ZOy
Ly=ZOx-XOz
Lz = XOy -yox,
then LL.,g = LLyg = LL.g = 0.
Proof. Let X,Y E TJM, and decompose X=Xiox;, Y=Yioxi, where
Xi, yi E C 00 (M). Below we use (t,x,y,z) and (x 0 ,x1,x2 ,x3 ) interchange-
ably. For any smooth function cp E 0 00 ( M),
[Lx,X]cp
= yoz(Xiox;cp) - zoy(Xiox;cp) - Xiox;(yozcp) + Xiox;(zoycp)
= y(ozXi)oxi'P - z(oyXi)(ox;cp) - Xi(ox;y)(ozcp) + X\ox;z)(oycp)
+yXiOzOxi'P - zXioyOxi'P - XiYOxiOz'P + XiZOxiOycp
= y(ozXi)(ox;cp) - X\ox;y)(ozcp) - z(oyXi)(ox;cp) + Xi(ox;z)(oycp)
= (y(ozXi)Oxi - XYoz - z(oyXi)Oxi + Xzoy)cp
= ((LxXi)Oxi -XYoz +xzoy)cp.
Thus it follows that [Lx,X] = (LxXi)Oxi - XYoz + xzoy. We also note
3
LXiLxxi = Xi(YOz - zoy)xi = yXZ - zXY.
i=l
We have
3
(xdx+ydy+zdz)((LxXi)Oxi -XYoz+xzoy) = -zXY +yxz+ L xi LxX\
i=l
3
(xdx+ydy+zdz)Y =xYx +yYY +zYz = LXiyi_
i=l
Using dr ® dr = !, (x dx + y dy + z dz) ® (x dx + y dy + z dz), we obtain

g([Lx,X], Y)
0 0 k(r)-1 ~ . . ~ . .
= -f(r)Y LxX + 2 LJ (x'LxX' + yXz - XYz) LJ Y3x 1
r i=l j=l
3
+ ~ yi LxXi + xzyy - XYYZ
i=l 3 3
= -f(r)Y0 LxX 0 + k(r)2-l ~(xiLxXi +XiLxxi) ~Yixi
r i=l j=l
3
+ ~ yi LxXi + xzyy - XYYZ.
i=l
Also,
g(X, [Lx, Y]) = - J(r)X 0 LxY0

3
r
+ k(r)2-l
i=l
Xixi
j=l
± ±
(xi LxYi + yi Lxxi)
+ ~XiLxYi + yzxy _ yyxz,

i=l 3 3 3
Lx(g(X, Y)) = Lx( - f(r)X 0 Y 0 + k(~2-l ~Xixi ~ yixi + ~xiyi).
i=l J=l i=l
We claim Lx(J(r)) = 0 and Lx(~) = 0: For any smooth function cp,

Lx( cp(r)) = yoz( cp(-Jx 2 +y 2 + z 2 )) - z oy(cp( -Jx 2 +y 2 + z 2 ))
= yep '(r )2z
- - zcp '( r )2y
- = 0.
2r 2r
Putting all of the above together, we see that
(CL.,g)(X, Y) = Lx(g(X, Y)) - g([Lx, X], Y) - g(X, [Lx, Y]) = 0.
So CL.,g = 0. Cycling through x, y, z, also CLyg = 0 and ££.g = 0. □
14.3 Schwarzschild spacetime
The field equations were published in 1915, but besides the Minkowski
spacetime solution, no other exact solutions were known then. In 1916,
Schwarzschild derived another solution, by making the simple symmetry
assumptions of 'time-independence' and 'spherical symmetry', and that the
spacetime metric looks like the Minkowski metric far away from the source.
The resulting Schwarzschild spacetime is a simple model of a spacetime
containing a single star. It is a good model for the exterior region in the
proximity of a star or a planet. Moreover, if the matter source is sufficiently
dense, then it also gives a model of the simplest possible 'black hole'.
Field equation 315
Consider the smooth manifold M = JR x (r min, oo) x S 2 as in the previous

section, where rmin ~ 0. We have seen a metric g that is stationary and
spherically symmetric, and which, in the (t, r, 0, cp )-coordinates, is given by
g = - /(r) dt ® dt + k(r) dr ® dr + r 2 d0 ® d0 + r 2(sin 0) 2 dcp ® dcp. (14.3)
In this section we will use the field equation to solve for the functions f, k.
We label the coordinate functions (t, r, 0, cp) as (x 0 , x1, x 2, x 3). First we
note that the inverse of the diagonal matrix [gij] is again a diagonal matrix
with the diagonal entries
00 1 11 1 22 1 33 1
g =-f(r)' g =k(r)' g =r2 ' g =r2(sin0)2 "
We can now compute the Christoffel coefficients. We suppress writing the
argument r for the functions f, k below. For example,
rg 1 = r~ 0 =
gtt
2 (iltgtr + ilrgtt -
1
iltgtr) = - 2/ or (- f)) = 2f'/"
Similarly, the other nonzero r-symbols can be found to be
r1 - L
oo - 2k,
1 k'
r u = 2k,
1 r
r 22 = - k,
r1 - - r(sin 0) 2
33 - k ,
2 2 1 1
r12=f21 =-, r~3=-(sin0)(cos0), rf3=f~1 =-, r~3=f~2=cot0.
r r
Knowing the Christoffel coefficients, we can compute the components of the
Ricci curvature tensor field. For example,
Ricoo = Oxkr~o - Otf~k + roor~k - rokr~o
= Orfbo - 0 + rbo(r~o + r~l + r~2 + rf3) - rbor~o - rg1rbo

f' f' f' k' 1 1 f' f' f' f'
- or(2k) + 2k (2/ + 2k + r + r) 2k2f 2/ 2k
f" f'k' (!') 2 f'
= 2k - 4k2 - 4kf + kr'
where -' := ~r· In a similar manner we can compute the other nonzero
components of Ric:
. f" (!') 2 k' f' k'
Rieu = - 2/ + 4/2 + 4kf + kr'
. 1 r f' rk'
R1c22 = -k + 1- 2kf + 2k2'
Ric33 = (sin0) 2 Ric22-

In the exterior of the star, T = 0. We also assume that the cosmological

constant is A= 0 or in any case negligible. Then the field equation given in
Exercise 14.1 reduces simply to Ric=0. In particular, Ricii =0 for i = 0, 1,
and so also ~Ricoo +Rieu= 0, which yields
f" f'k' (!') 2 f' f" (!')2 f'k' k' . f' k'
2f - 4kf - 4f2 + rf - 2f + 4f2 + 4kf + kr = O, 1.e., f + k = O.
Hence f'k + k' f = 0, that is, (Jk)' = 0. Thus f k is constant, say C. But as
r -+ oo, we expect that the metric expression (14.3) for g approaches the
Minkowski spacetime metric (in spherical coordinates), so that
lim f(r) = 1 = lim k(r).
r--+OO r---+OO
Passing the limit r-+ oo in the relation f(r)k(r)=C, we get C=l. Thus
1
k(r) = f(r).
Substituting for k in Ric22 = 0 gives
rf'
-f +l - 2 +2
rf 2 (1)'
f = 0.
So rf'+ f =l, i.e., (rf)' =l. Thus by integrating, we obtain rf(r) = r + A,
where A E JR is an integration constant. This yields f(r) = 1 + 1-- So far,
we have not specified Tmin ~ 0. We want J(r) > 0 on (rmin, 00 ), with Tmin ~ 0
the smallest possible such number. We have the following cases:
1° A~ 0. Then Tmin := 0.
2° A< 0. Then Tmin := -A.
We will soon argue on physical grounds that A=: -2m < 0, where m > 0 is
thought of as the mass of the star. For now, given an A E JR, we just take
Tmin as specified above. Then,
f(r) =
A
1 +-;: and k(r) =
(
1 +-;:A)-1 .
We can plug these expressions back and check that indeed Ricii = 0 for
i = 0, 1, 2, 3, so that the field equations do hold (the other Ricci curvature
field components Ricij, i =I= j, were already known to be zero for arbitrary
f, k). Thus we obtain the Schwarzschild metric
g=-(1+4 )dt® dt + ( 1+4 )- dr ®dr + r d01 2 2 + r 2 (sin0) 2 d<f>®d<f>
for the spacetime region JR x (r min, oo) x S 2 ( thought of as the exterior of
a star), under the assumptions that the metric is stationary 11 , spherically
symmetric, and it approaches the Minkowski spacetime metric as r-+ oo.
11 Birkhoff's theorem says that the assumptions of stationarity and that the metric ap-
proaches the Minkowski metric away from the source, are superfluous here. For a precise
statement and a proof, we refer the reader to [Kriele(2001), Thm. 7.2.1].
Field equation 317
We now give an argument to justify that A=: -2m < 0, and that then
Tmin=-A= 2m, where m can be interpreted to be the mass of the star.
Recall that in §14.1, we had found that in the Newtonian limit, if we write
the metric g on JR x IR3 as gij = T/ij + hij, then hoo = - 2<1>, where is the
gravitational potential. But at a distance r from the source of mass m, the
Newtonian gravitational potential is
<l>=-m.
r
So for large r, writing the Schwarzschild metric as gij = T/ij + hij, we have
goo = -1 + hoo ~ -1 - 2 = -1 - 2 - ; = - ( 1 - 2~) .
Thus ifwe identify A with -2m, i.e., m:=-A/2, then this m corresponds
to the mass of the star in the classical viewpoint, and in particular m > 0.
So A=: -2m > 0, and Tmin =2m, where m has the interpretation of being
the mass of the star.
Alternatively, we could look at the geodesic equation to justify this
correspondence. We do this in the following exercise.
Exercise 14. 7. (Radial geodesic in Schwarzschild spacetime).
Let (M,g) be the Schwarzschild spacetime. Let 00 E (0,71") and ¢ 0 E (0,271") be
fixed, and p E S 2 be the point with the spherical polar coordinates (0 0 , ¢ 0 ).
Let 'Y: J-+ M, I 3 T >-+ (t(T),r(T),p), be the worldline of a free-falling particle,
parametrised by the arclength/proper time. Show that if •' := ~r, then:
-(l- 2m)
r
(t') + (l- 2m)-\r')2 = -l,
2
r
t" = _ 2m (l- 2m)-\'r', r" =
~ r
_ m.
~
Let 'Y pass through p EM. Consider a normalised instantaneous observer
v = ✓ (1- 2r; r
1
Ot,p E TpM.
Show that v perceives the relative speed of the particle 'Y to be
u= f,✓(1-o/-f 1 Or,p•
Determine the magnitude lul := -Jg(p)(u, u) of u as perceived by v.
Now suppose that r » 2m, and lul « 1 (that is, low speed as reckoned by a
'stationary observer' far away from the source). Conclude that lr'I «t', and t' R< 1.
(In physics parlance, t' R< 1 is expressed as 'coordinate time is proper time'.) Thus
if the map h denotes the inverse of the map T >-+ t(T), then 'Yoh is the map
t >-+ (t, r(h(t)), 0o, </>o). Show that
d2 m
dt2 (r oh) R< - r2.
From the Newtonian viewpoint, in the spacetime Rx R 3 , a radially freely falling
particle in the gravitational field of a mass m at the origin experiences an accel-
eration which matches with the above. So the constant m in the Schwarzschild
spacetime can be thought of as being approximately the mass of the star.
We note that the Schwarzschild metric can be used only for r > 2m, since
if r = 2m, the coefficient of dt ® dt becomes zero, and also the coefficient of
dr ® dr is not defined. For a star like our Sun, we have 2m ~ 2M0 ~ 3 km,
and so when we are outside the body of the Sun, we do have r > 2m. For
the Earth, 2m ~ 2MffJ ~ 9 mm, which is again much smaller than its own
radius. For a body of mass m, we define its Schwarzschild radius Rs to
be 2m. A body with a radius smaller than its own Schwarzschild radius is
thought of as a 'black hole', and will be discussed in the next chapter.
Exercise 14.8. Find the minimum density needed for a body with radius equal to
that of the Earth (Re= 6400km), so that its Schwarzschild radius is at least as
large as Re- Compare it with the density of water, Pwater = 1000kg/m3 •
Exercise 14.9. (Reissner-Nordstrom spacetime).

Recall the Reissner-Nordstrom spacetime (M, g) from Exercise 12.9 (p.286), and
the (0, 2)-tensor field T E Tf M given in Exercise 13.4 (p.304). Show that the
field equation is satisfied, assuming that the cosmological constant A= 0.
14.4 Perihelion precession

We had seen two effects within the spacetime-geometry viewpoint of grav-
ity, which could not be explained classically, and which have been tested
experimentally /by observation:
• the gravitational red shift (Example 5.16), and
• the deflection of light (Section 8.4).
We now learn about a third observational verification, which is on the tra-
jectories of planets in the solar system. In particular, the effect is most
pronounced for Mercury, and the amount of 'precession of its orbit' pre-
dicted matches very well with the observed value. We begin by discussing
orbits of planets within the classical framework in the idealised case when
we have a single planet around a star.
Newtonian description of planetary motion
Let us first consider planetary motion as described in Newtonian gravity.
We will learn that the trajectory of a planet like Mercury around the Sun
is in an ellipse, with the Sun at one of the foci.
Let the Sun be fixed at the origin of R 3 • It generates a gravitational field
in which the test matter, namely Mercury, moves according to the law of
gravitation. Here we assume that (since the Sun is much more massive than
Mercury) the common center of mass of the Sun and Mercury is essentially
located at the center of the Sun, that is, the origin.
Field equation 319
Thus if x(t) = (x(t), y(y), z(t)) E IR3 is the position of the planet at time
t E JR, then the equation of motion is
.. m
x = - llx ll3 x.
Here m is the mass of the Sun, and · = ~. If v = x is the velocity, then
d( ) . . ..
dt xxv =xxv+xxv=vxv+xxx
m
= 0 - llxll3 (x x x) = 0.
Thus the 'angular momentum' vector L =xx vis a constant. We assume
henceforth that L =I= 0. Then the planet is confined 12 to move in the plane
which contains the 'initial' position vector x(0) of the planet, and the initial
velocity v(0) of the planet. See the following picture.
By a suitable rotation of coordinates in IR3, we can assume that the plane of

planetary motion is the xy-plane, and we will use planar polar coordinates
(r,¢) in this plane. Then x=(rcos<p, rsin¢, 0) , and so
v = (rcos¢- r¢sin¢, rsin¢ + r¢cos¢, 0).
We know the angular momentum vector L = x x v is constant. We have
L =xx v = (rcos¢, rsin¢, 0) x (rcos<p- r¢sin¢, rsin¢ + r¢cos¢, 0)
= (0, 0, r 2 ¢).
Thus L := r 2 ¢ is a constant. We also note that by differentiating r 2 = (x, x)
with respect to time, we obtain 2rr=2(x,x)=2(x, v), so that rr =(x, v).
Differentiating again,
<
rr + (i-) 2 = (v, v) + (x, x) = llvll 2 + x, -11~3 X) = llvll2 - 1: 11. (14.4)
But llxll = r and

llvll 2 = (rcos¢- r¢sin¢) 2 + (rsin¢ + r¢cos¢) 2 = (i-) 2 + r 2 (¢) 2 .
12 Indeed, x •L = x • (xx v) = 0, showing that x(t) lies in the plane perpendicular to L.
Hence (14.4) yields rr + (i-) 2 = (i-) 2 + r 2(¢) 2 - 7 , that is,

rr = r2(¢)2 -
m = (r2¢)2 - m = £2 - m. (14.5)
r r2 r r2 r
We want r as a function of ¢ in order to see what the trajectory looks like in
the xy-plane. So let us suppose that (locally) ti--+ ¢(t) is a diffeomorphism
with an inverse h, and set
1
u(c/>) = (r o h)(c/>)
(It turns out it is more convenient to work with the reciprocal of r.) Then
du l dr dh l . 1 r(h(-))
de/> = (r(h(-))) 2 dt (h(-)) de/> = (r(h(-)))2 r(h(·)) ef>(h(·)) = - - L - .
Similarly
d2u = _..!_ d(r Oh) = _..!_r(h(·)) dh = _..!_r(h(·))-._l_
d¢2 L de/> L de/> L ¢(h(·))
= _..!_r(h(-)) (r(h(·)~)2 (r(h(·)))2r(h(·))
L (r(h(·)))2¢(h(·)) L2
Substituting this in (14.5), we obtain
d2 u m
d¢2 +u = L2· (14.6)
The general solution to the homogeneous equation

d2 u
d¢2+u=O
is given by Uh=Acos(c/>-c/>o), for some A~O and ¢0 E [0,27r). A particular
solution to (14.6) is the constant function up= v-·
Thus the general solution
to (14.6) is given by
m
u(cp) = A cos(¢ - c/>o) + £ 2.
We are interested in what this trajectory looks like. We may assume that
the xy-coordinate plane has been rotated about the z axis if necessary to
have ¢ 0 =0. Then the above gives with t=h(c/>)
£2
1 1 m k
r(t) = - - = ~ - - - = 2 = ---- (14.7)
u(¢) ; + A cos¢ 1 + AL cos¢ 1 + ecos¢'
m
where k := Lm > 0 and e := AL

2
m
~ 0. Now we will show that if O< e < 1,
2
then ¢ i--+ u{ct,) describes an ellipse.

Field equation 321
Consider an ellipse in the xy-plane with foci at (0, 0) and at (-2c, 0) where
c > 0. Then the centre C of the ellipse is at (-c, 0). If the lengths of the
major and minor axes are 2a, respectively 2b, then the ellipse is given by
(x + c)2 y2
a2 + b2 = 1. (14.8)
Recall also that the ellipse is the locus of a movable point P such that the
sum S of the distances of P to the two foci is a constant. By taking the
point P along the major axis, we obtain that S = c+a+ (a-c) = 2a. Next
taking the movable point P along the minor axis, and using Pythagoras
theorem in the right angled triangle formed by P, C and the origin, we
obtain b2 + c2 = (S/2) 2 =a 2. Thus the eccentricity e of the ellipse is
e := dista~ce be_tween foci = 2c = .::, =

maJor axis length 2a a
~
a
= r:Y'~
V1 -
E (0, l).
(- 2c, 0) C
a C
Rewriting (14.7) as r + ercoscp = k, we haver+ ex= k, i.e., r = k - ex.

Squaring, x 2 + y2 =r 2 =e 2x 2 - 2ekx + k 2, i.e., (1-e 2)x2 + 2ekx + y2 =k 2.
Completing the square in the x terms,
ek ) 2 e2k2 k2
(1-e 2) ( x+ - - +y 2 = k2 + - - = - - ,
l-e 2 l-e 2 l-e 2
that is,
ek )2
( x+--
l-e2
k 2 +
( l-e 2 )
which matches (14.8) with a= l~e 2 > b= ✓ i~e 2 > 0, and c= 1 _:_~ 2 =ae. So
1 k
cpt---->--=----
u(cp) l+ecoscp
describes an ellipse, with (0, 0) as one of its foci. Consequently, the trajec-
tory of the planet is an ellipse with the Sun its focus.
Planetary motion as a spacetime geodesic

We consider the exterior region of the Sun as a Schwarzschild spacetime
(M, g), where g is the Schwarzschild metric. We take 2m ~ 2M0 = 3km. The
planet Mercury is considered as a 'test particle' which does not influence
the geometry (so its energy-momentum contribution in the field equation
is negligible). Thus its trajectory 'Y, parametrised by proper time, is a
geodesic I 3 T >-+ (t(r), r(r), p(r)), where p(r) E S 2 has the spherical polar
coordinates (0(r), cp(r)). So 'Y satisfies the geodesic equation. In particular,
for the 0 component of 'Y, with-':= ~r' we have
0 = 0" + 2r~0 r'0' + r:<1>¢'¢' = 0" + ~r'0' - (sin0)(cos0)¢'¢'.
r
Suppose 0 E J, and consider the initial conditions 0(0) = i and ~~ (0) = 0.
Intuitively, this means that the motion begins in the equatorial plane with
the 'initial velocity' also lying in the equatorial plane. Since cos i = 0,
the above equation shows that 0 = 0 is the solution to the above equation
(thanks to uniqueness). So the planet moves in the plane, just as we had
deduced via the Newtonian viewpoint earlier in this section.
The ¢-component of the geodesic equation gives, when 0 = i, that
0 = ¢" + 2r:<1>0'4>' + 2rt</>r'¢' = ¢" + 0 + 2~r'¢',
that is ,'¥
,,_.,, = - 1r' ,1./ Hence
r '¥·
(r 2 ¢')' = 2rr'¢' +r 2 ¢" = 2rr'¢' +r 2 ( - ~r'¢') = 0.

Thus L := r 2 ¢' is a constant. The r-component of the geodesic equation
gives (with 0 = i):
0= T 11 + r;t(t') 2 + r;T(r') 2 + r99(0')2 + r;q,(¢') 2
2m) (')
= r 11 + rm2 ( 1---;:- t 2 2m)-l(r')
- rm2 ( 1---;:- 2 2m) (sm. 1r)
+0-r ( l---;:- 2
2
( ¢ ') 2
= r" + r2
m J(r)(t') 2 - ~(r') 2 - rf(r)(c/>') 2
r2 f(r)
where f(r) := 1 - 2;:1. Thus

r" = - m2 (f(r)(t') 2 - - 1-(r') 2 ) + rf(r)(¢')2. (14.9)
r J(r)
We have v')',')'(T) = t' Ot,7(r) +r' Or,7(r)+o+4>' Oq,,')'(r)· Also, as "f is parametrised
by proper time, we have g('Y(r))(v7 ,'"Y(r),v'"Y,'"Y(r))=-1, that is,
-1 = - J(r)(t') 2 + - 1-(r') 2 + 0 + r 2 (sin~) (¢') 2
f(r) 2
= -f(r)(t') 2 + ftr) (r') 2 + r 2 (¢')2. (14.10)

Field equation 323
Thus (14.9) and (14.10) together give
r" = - m (1 + r 2(¢') 2) + rf(r)(</>') 2 = - m + (- m + r(l- 2m)) (¢') 2

~ ~ r
= - ~ + (r - 3m)(¢') 2. (14.11)
r
Again we will work locally in an interval J, where the map J 3 T >-+ cp(r)
admits an inverse, and this inverse will be denoted by h. Set
1
u(cp) = (r o h)(cp)"
Then analogous to what we had done earlier, we have
du l dr dh l 1 r'(h(·))
dcp = (r(h(-))) 2 dr (h(·)) dcp = (r(h(-))) 2 r'(h(·)) cp'(h(·)) L
Similarly
d2u 1 d(r' oh) 1 "( ( )) dh l "( ( )) 1
dcp2 = - L dcp = -y,r h . dcp = -y,r h . cp'(h(·))
= _.!_ "(h(·)) (r(h(-))) 2 (r(h(·)))2r"(h(·)) r"(h(·))
Lr (r(h(</>)))2</>'(h(-)) L2
Substituting this in (14.11), and also using¢'= r 2

r
t' = r~ = L(uoh- 1 ) 2,
2 2 d2u 2 (1 ) 2 4 • d2u m 2 )
-Lu d</>2 =-mu + ~-3m Lu , 1.e., d</>2 +u= £ 2 +3mu . (14.12
Comparing with (14.6), we see that there is now an extra term 3mu 2. As
opposed to (14.6), which we could solve explicitly, now because of the extra
term, we can no longer solve the equation in an elementary manner. But
since 2m ~ 3km, and for planetary orbits r is in the order of millions 13
of kilometers, it follows that 3mu « l. As 3mu 2 = 3muu « u, in equation
(14.12), the term 3mu 2 is quite small as compared to the u on the left-hand
side. Also, assuming that the orbital speed rep' is much less than the speed
of light, we have (mu 2)/"D" = ,f? = (rc/>') 2 « 1. So we proceed perturbatively
as in §8.4. We seek a solution u = uo + 1:u1 + 1:2u2 + · · · , to
(14.13)
where 1:=3m«l, and uo="D"(l + ecoscp) is a solution to

d2u 0 m . . .. .. { uo(O) = "D"(l+e)
d,1..2 + uo = £ 2 , with the 1mtial cond1t10ns du 0 (O) = 0
'I' d<J, •
13 For example, the closest distance of Mercury to the Sun is ~ 46 million kilometers.
Substituting the series expansion for u into the differential equation (14.13),
and by comparing terms in 1: that
d2u1 2 m2 2 m2 ( e2 e2 ~
d¢2 +u1= (uo) = L4(1+ecos¢) = L 4 1+ 2 +2ecos¢+ 2 cos(2¢)1.
Set u1,p = a+ ¢(bi cos <p + b2 sin¢)+ c cos(2¢) in order to find a particular
solution u 1 ,p. Plugging this into the equation, and using the independence
of 1, sin¢, cos¢, cos(2¢), we can determine a, b1, b2, c, and obtain
m2 ( e2 e2 ~
u1,p = L 4 1 + 2 + e¢sin¢- 6 cos(2¢)).
Then :;1 ,P(O) = 0. So if u(¢) := u0 (¢) + w 1,p(</>), then, as 14 , ;: « 1,
m 3m3 ( e2 ) m du
u(0) = L 2 (l+e)+ L 4 1+ 3 ~ L 2 (l+e) =uo(0), d<p (0) =0.
So we may take u as an approximation of the solution u to (14.12) with
m du
u(0)= L 2 (1+e) and d¢(0)=0.
The perihelion of a planet's trajectory is the point along the trajectory that
is closest to the Sun. Thus r is minimised, and this corresponds to u being
maximised. In the case of the ellipse described by¢ i--+ 1/u0 (¢), where
m
uo(</>) = L 2 (1 + ecos¢),
the perihelion occurs at ¢=0, ±2n, ±4n, • • •. To determine the local maxi-
u,
mum of we seek solutions to ~~ (¢) = 0. We have
:: (¢) = - ;: sin¢+ 3;:e (sin¢+ ¢cos¢+ i sin(2¢)).
We have ~~ (0) = 0, so there is a perihelion at ¢ = 0. The next perihelion is

not at 2n (as opposed to the case of the ellipse), since
du 3m3 e
d</> (2n) = ~ 2 n =I 0.
But if we suppose that the perihelion is instead at 2n + 8, with a 181 « 1,
then this 8 satisfies approximately that
me 3m3 e ( e )
0 ~ - L 2 8 + ~ 8 + (2n+8) + 328 .
m2 6m 2 71"
Since L2 « 1, we obtain 8 ~ ~ -
14 As u 0 = -B"(l+ecos,t,), and since the eccentricity of Mercury is~ 0.2, we have that
m2 <
TI'~ .§. m
+ ecos,t,) -_ .§.
muo ,.,,,_, .!!!, ,.,,,_, 3 ,.,,,_,
10 -7 « 1.
4 mv(l 4 r 40 x 106
Field equation 325
Thus the Newtonian elliptical planetary orbit is now prevented from closing
in on itself, and the resulting trajectory resembles a rosette pattern, as
shown (much exaggerated) below, with an advancement of the perihelion
in each traversal of a 'petal'. This is called the perihelion precession 15 .
Remark 14.1. In fact, in the Newtonian viewpoint as well, the orbit of

Mercury precesses owing to the gravitational influence of the other planets;
see e.g. [Landau and Lifshitz(1976), Problem 3, p.40] . Indeed, in our New-
tonian description, we looked a test particle moving in the field created by
a single body (the Sun), while in reality we have an n-body (with n > l)
problem in our solar system, because of the presence of the other planets
which do affect the orbit of Mercury. This amounts to about 16 532 arcsec-
onds17 per century. However, the classical viewpoint could not account for
a discrepancy of about 43" with the observed value of about 575". Before
the spacetime geodesic viewpoint, it was proposed that perhaps there is
another planet, called 'Vulcan', whose orbit lies between that of Mercury
and the Sun, which explains the discrepancy from the classical value of
precession. However, Vulcan does not exist. On the other hand, the calcu-
lation from the spacetime geodesic viewpoint predicts a value of 43.03" per
century, in excellent agreement with the observed value. This early result
(1916) due to Einstein is considered as the first experimental verification of
spacetime-geometric description of gravitation. *
15 The terminology is borrowed from the mechanics of a spinning top in Earth's gravity,
where the axis of rotation itself rotates, spanning a cone, and one says the axis ' precesses'.
16 See for example [Misner, Thorne and Wheeler(2017) , p.1113].
17 An arcsecond (denoted henceforth by•") is l
3 00 of a degree, i.e., 3600 = 1° .
11
Chapter 15
Black holes
In Section 14.3, we saw that the smooth manifold M = Rx (rmin, oo) x S 2 ,

where Tmin = 2m > 0, is a spacetime 1 with the Schwarzschild metric g. In a
spherical coordinate chart described there, g is given by
g=-(1- 2~)dt©dt+ (1- 2~)- dr©dr+r d0 1 2 2 +r 2 (sin0) 2 dcp©dcp.
We had remarked that for a typical star, 2m is much smaller than its radius,
and so the above describes the exterior region of the star, where we assumed
that the cosmological constant is A= 0, and that there is vacuum, that is,
the energy-momentum tensor field is T=O.
But now let us consider the situation, where all the mass is located 'at
r = 0'. We do not think of this necessarily as a point, but rather simply
something not in our manifold (where in the manifold we have r > 0).
Let MaH := RxJxS2 , where J = (0,2m) u (2m,oo) c R. Then MaH
is a Lorentzian manifold with the same metric g given above, and again
the field equations are satisfied in MaH with A = 0 and with the vacuum
energy-momentum tensor T = 0. While Ot is timelike in the region r > 2m,
we have Or is timelike in the region r < 2m. Thus we cannot have a particle
trajectory with r = constant in the region r < 2m. The r-coordinate of the
particle in this region must necessarily increase or decrease, depending on
what time-orientation is chosen in this part.
By looking at the behaviour of lightlike geodesics, we will see that light-
rays emanating from the region 0 < r < 2m cannot travel to the exterior
region, while null geodesics are allowed to fall in. Thus no light emerges
from such an object, and it is legitimate to call it a black hole. We will only
study Schwarzschild black holes, roughly collapsed objects characterised by
a unique parameter, namely their mass m.
1 A time-orientation is given by Ot, and an orientation is induced by the orientation
[volg] on~ X ~ 3 , where volg = vldet[gij]I dxAdyAdzAdt.
327
15.1 The spacetime (Matt, g)
In this section we will show that the spacetime (MaH, g) is 'singular' with
a singularity at r = 0. But first, we will introduce the spacetime (MaH, g)
in more precise terms than the description given in the introduction.
Definition 15.1. (The spacetime (MaH, g)).
Let (MaH, g) be the smooth manifold
MBH = JR X J X S 2, where J = (O,2m) u (2m,oo) C JR,
with a Lorentzian metric g, given as follows. The smooth manifold J x S 2
is identified with {(x, y, z) E JR3 I O < r := .Jx2 +y 2 +z 2 =I= 2m}, using the
diffeomorphism (t, (x, y, z)) >-+ (t, r, (x,;,z)) E MaH· Then g is given in the
global chart (MaH, (t, x, y, z) >-+ (t, x, y, z)) by
g= - f (r) dt ® dt + ( f tr) - 1) dr ® dr + dx ® dx + dy ® dy + dz ® dz,

where f(r) = 1 - 2;". Let (U, cp) be the spherical coordinate chart for S 2
from Example 1.8 (p.7). In the chart (JR x J x U, (t, r,p) >-+ (t, r, cp(p))),
g= -(1- 2~)dt ®dt + ( 1- 2~)- 1dr ®dr + r 2 d0 2 + r 2 (sin0) 2 d¢®d¢.

An orientation is induced on MaH by the restriction to MaH of the top-form
field volg = .JI det[gii]I dx I\ dy A dz I\ dt on JR x JR 3 . A time-orientation is
given by Ot if r > 2m, and by - Or if O< r < 2m.
In the above, we have chosen 'the direction of decreasing r' in the region
0 < r < 2m, as being future-pointing. This will allow us to talk about
in-falling lightlike geodesics.
t
y
► f y
0
r
~ r y
► y
Black holes 329
The expression for the metric g seems to break down at r = 2m. But we
shall see that this is a mere artefact of the chosen coordinates. To explain
what we mean by this, let us first consider the following example.
Example 15.1. Consider the Minkowski plane JR 2 with the standard
smooth structure, and the metric
g = -dt®dt + dx®dx
in the global admissible chart (JR 2 ,(t,x) >-+ (t,x)). Then (JR 2 ,g) is a
Lorentzian manifold. Consider the admissible 2 chart
(JR 2 \{(0,x): x E JR}, (t,x) >-+ (r(t,x):=t3 ,x)).
In this new chart, the metric is given by
g = -(1/9)r- 4/ 3 dr ® dr + dx ® dx.
The coefficient function of dr ® dr above, namely
1
JR 2 \{(0,x): XE JR} 3 (t,x) = p >-+ - 9t4'
does not stay bounded asp= (t, x)-> (0, 0). ◊
How do we detect if there is a genuine 'singularity' at a point or if it is the

case that awkward coordinates have been chosen? To do this, we first need
to agree upon what we consider as a singularity for a given manifold M.
Intuitively, we could demand from a non-singular manifold that geodesics
don't stop or run into an 'obstacle' (a 'hole' or a 'missing point'), but have
their affine parameter living on the whole of JR. This motivates the following
definition.
Definition 15.2. (Singular spacetime).
A spacetime(M, g) is called singular if it is not geodesically complete.
Example 15.2. The Minkowski plane (M, g) from Example 15.1 is geodesi-
cally complete, since each geodesic is a straight line
7(t) = (xo, Yo)+ tv (t
JR),E
for some point (x 0 , y 0 ) E JR 2 and some vector v E JR2 . This follows from the
fact that the Levi-Civita connection in this case is just the flat connection
on JR 2 , and we had determined the geodesics for the flat connection in
Example 8.1 (p.156).
Also, from Example 8.4 (p.164), the 4-dimensional Minkowski spacetime
is geodesically complete. ◊
2 As opposed to Example 1.6 (p.5), where we also had the chart map t ,..... t 3 , we took
the chart domain as the whole manifold, while we now remove the problematic points
(i.e. the t-axis), so that there is no obstacle for the chart transition maps to be smooth.
One way3 to detect if the spacetime is singular is to find out if the curva-
ture goes to infinity along a geodesic as the affine parameter approaches
a finite value. In the example below we will see in this manner that the
Schwarzschild Lorentzian manifold (MaH, g) is singular, by showing that a
radial geodesic in the interior region (r < 2m) hits r = 0 in finite proper
time, while the Kretschmann scalar goes to oo.
Example 15.3. In the spacetime MaH, consider a radial timelike geodesic

'Y : J -+ MaH parametrised by proper time, with constant 0, ¢. Then we
have ¢' = 0' = 0. The connection coefficients are given by the expressions in
Example 6.8 (p.124) even in the interior region. As in Exercise 14.7 (p.317)
and its solution, the geodesic equations are
r" = - m (!(r)(t')2 - _1_(r')2), (15.1)

r2 J(r)
t" = _...l:!!!:._t'r' (15.2)
r 2 J(r)
where /(r):=1- 2;'. Since g('Y(r))(v-y,-y(T),v-y,-y(T)) = -1, also
-1 = -/(r)(t')2 + ftr) (r') 2. (15.3)
We will now construct a solution to the above equations. Let r 0 E (0, 2m),
and let r satisfy the initial value problem
r' -- - V~rm'
--:;:- r (O) = ro E (0, 2m ) .
Then r(r) = (r!- 3 :v:;f°r)f, and (r') 2 = 2;'. Lett'= f(r), and t(0) =to, i.e.,
_;j_
2r 2
t(r)=to+ fT(l- 2m 2/3)-lds, for O~T< ~ -
J0 ( ro3/2 --2-s
3v'2m ) 3v2m
We check that (15.1), (15.2), (15.3) are satisfied. Indeed, (15.3) holds, as
, 2 1 , 2 1 1 2m 1 ( 2m)
-f(r)(t) + f(r)(r) =-/(r)(J(r))2+ f(r)--:;:-=- f(r) 1---:;:- =-1.
3 However, this is not a foolproof way. The curvature tensor field may behave benignly
even for geodesically incomplete semi-Riemannian manifolds. This can happen if a point
is missing and cannot be added, or has been viciously removed. For example, consider
the double cone C, = {(x, y, z) E li3 : x 2 + y 2 = z 2 }\{(0, 0, 0)} c li3 as a smooth manifold.
Equipping C, with the metric induced from the Euclidean inner product on li3 , the
curvature tensor field of C, can be shown to be everywhere zero. But if p E C,, then the
line L(t) =tp, t E (0, oo) can be shown to be a maximal geodesic, and leaves C, at t=0.
Black holes 331
Moreover, (15.2) holds, since

2m , , 2m 1 , 2m , 1
----tr=------r=--r---
r2 J(r) r 2J(r) J(r) r 2 (J(r)) 2
= (ftr))' = (t')' = t".

Finally differentiating (r') 2 = 2m gives
r
2r'r" = - 2m r'
r2
and so using (15.3) (already proved)
r" = - m = - m (1) = - m (!(r)(t')2 - _1_(r')2).
r2 r2 r2 f(r)
Thus 'Y, with the above constructed r, t, is a geodesic. We note that r = 0
is reached in the finite proper time
l!.
T= _2r
2
_O_
3V2m.
On the other hand, it turns out 4 that the Kretschmann invariant/scalar
(Exercise 9.10, p.189) is
48m2
K= --6-, (15.4)
r
and so
limK = +oo.
r'\.0
Hence (MaH, g) is singular. ◊
Thus 'r = O' is a genuine singularity for the Lorentzian manifold (MaH, g).
What about r = 2m? From the above expression for the Kretschmann
scalar, we suspect that the problem at r = 2m for the expression for g
might very well be just a coordinate artefact. In order to see what happens
when r = 2m, we study null geodesics in (MaH, g). We refer to the 'surface'
JR x {2m} x 8 2 c JR x (0, oo) x 8 2 as the event horizon. The Schwarzschild
radius is r 8 :=2m (i.e., ~2m, after GN and c have been reinstated).
4 See e.g. [Grave and Mueller(2010), (2.2.5), p.18] or [O'Nei11(1983), Exercise 8, p.399].
We will not carry out this computation here. We remark that the scalar curvature is not
an indicator of the presence of the singularity: S = 0. This is because the Ricci curvature
tensor field Ric = 0 for the Schwarzschild spacetime. We had seen this in the exterior
region (r > 2m) while deriving the Schwarzschild metric (p.316). But since the connection
coefficients in the interior region (0<r<2m) are given by the same expressions as those
in the exterior region, the Ricci curvature tensor field is zero everywhere.
15.2 Null geodesics and the event horizon
Proposition 15.1. Let 0o E (0, 1r) and </>o E (0, 21r) be fixed, determining
a radial direction in S 2 . Let p E S 2 be the point with the spherical polar
coordinates (0o, </>o). Then the corresponding radial lightlike geodesics in
(M8 H, g) are affine reparametrisations of"(: I --> M 8 H, 1(r) = (t(r), r, p),
r E I , where the interval I c J, and J and t are given in the table below:
Region J t(r) t(r)

Interior (0, 2m) r+ log(2m-r) +c -r- log(2m-r) +c
Exterior (2m, oo) -r - log(r - 2m) + c r + log(r - 2m) + c
The constant c E IR. is arbitrary in the above table.
Proof. First consider the radial lightlike geodesic ::Y : J--> M 8 H described
by an affine parameter TE l: ::Y(T) = (t(T),r(T),p), TE J. With.'= ~7 ,
its </>- and 0-components satisfy </>' = 0 and 0' = 0. The t-component of the
geodesic equation for ::Y gives
~11 2m ~,,
t = - r2 f(r) tr
where f(r):=1- 2;?'. As"( is lightlike, g('Y(T))(v,,,(r),v,,,(r)) = 0, and so
-f(r)(t')2 + ftr) (r')2 = 0. (15.5)
We first note that E:=f(r)t' is a constant, since

~ ~ ~ 2m ~ 2m ~
(J(r)t')' = (J(r))' t' + J(r)t" = --:;:x-r' t' - f(r) r 2J(r) t'r' = 0. (15.6)
Black holes 333
Now using (15.5), -f(r)(f~~) 2 + f!r)(r')2=0. Consequently, (r') 2 =E 2 .
Sor' =a E {-E , E}. Thus r(r) = ar + b, TE 1, which is an affine map. Set

I= al+ b, and define T: I--> I by r(r) = r-:;b , r E J . Theorem 8.2 (p.163)
implies"/ =1o T: I--> M aH is a geodesic. As 1 (r) = (tt~b), r, 0o , <Po) (r E I),
d (~(
r - b))
dr t -a- = t
~, (-a-
r - b) 1
-;;: =
E 1 1 ( 2m )
f(r)-;;: = ± J(r) = ± 1 + r - 2m ·
Integrating, t(r) :=t(r~b) = ±(r + 2mlog Ir - 2ml) + c for some c ER The
interval J is determined by the range of r. □
Let us now see how things appear to an observer outside the event horizon.
Definition 15.3. (Schwarzschild observer).
Let r 0 > 2m, 00 E (0, 7r), ¢ E (0, 27r). Let p E S2 be the point with the
spherical polar coordinates (0o , <Po). The worldline "Is : IR.--> M aH given by
"/s(r) = (r)(1- 2
is called a Schwarzschild observer.
r:r 1
, ro, p) , TEIR. ,
For a Schwarzschild observer "Is, we have

v"Y,,"Y, (rl = )(1- 2r:) -1 ot,"Y, (r),
and so g("/s(r))(v"Y,,"Ys(-r), v"Y, ,"Ys(-r)) = -1 for all TE R Thus proper time for
"Is between two events p = "Is (T P) and q = "Is (Tq) along "Is is
lrp - Tql = lt(p) - t(q)I.
0
The (r, 0, ¢)-coordinates of "Is are fixed as (r0 , 00 , ¢ 0 ). Now consider a radial
infalling lightlike geodesic ,\ as shown in the picture below.
"Is
' '
'' ''
''
2m ::
0 >-------+-~-'l------r➔
The radial light geodesics emanating from the events along >. will reach the
worldline of 'Ys at later and later times, so that the Schwarzschild observer
will think that the photon never reaches the event horizon. On the other
hand, if we consider the photon parametrised5 by the affine parameter r,
then starting from the parameter value r 0 > 2m, the photon reaches the
event horizon at the finite affine parameter value r 0 - 2m.
Exercise 15.1. A similar phenomenon as above is valid with a radial timelike

geodesic. For example, we show this by constructing a geodesic similar to the one
in Example 15.3. Let r satisfy
dr =- ~ r(0) = ro > 2m.
dT y-:;:-,
-2y2m2 .2 3
So r(T) = (rJ- 3 2 mT):l".
2 For T= ~(rJ-(2m)2)
3v2m <OO, r(T) =2m. Lett satisfy
dt 1 1
t(0) =0.
dT = f(r) = 1- 2m '
( ro3/ 2 3y'2m ) 2 13
- -2-r
Let p E S 2 be the point with the spherical polar coordinates (0 0 , ¢ 0 ) . Then the
same steps as in Example 15.3 show that --y(T) := (t(T),r(T),p), TE (O,T) is a
timelike geodesic (particle) starting along the Schwarzschild observer's worldline
at (0,ro,p), and reaching the event horizon in the particle's proper time T<oo.
Set T : ( 2m, ro) --+ ( 0, T) to be the inverse of the map r, that is, r o T = id( 2 m,ro) .
Using(*) and(**), and setting t = toT,
Find t as a function of r. In the rt-plane, plot its graph for 2m = 1, and ro = 2.
No light rays emanating from the region inside the event horizon, reach
the Schwarzschild observer 'Ys, and also do not reach the events along any
timelike curve living in the exterior of the event horizon. Thus a black hole
appears black. Of course the starry light from elsewhere, e.g. 'behind' the
black hole, can travel to the observers outside the event horizon, as shown .
•
5 This fixes its energy/frequency as measured e.g. by the Schwarzschild observer at the
observer
common event >.(ro).

Black holes 335
Exercise 15.2. (Schwarzschild radius pre-Schwarzschild).

The aim of this exercise is to point out that the Schwarzschild radius already
appeared in Laplace's work in 1798. Classically, the gravitational potential energy
due to a star at distance a r from its centre, is minus the amount of work done
to bring an object from separation r to far away ('at infinity, r = oo'). By the
Newton gravitation law, the force experienced by a mass iii at a separation r is
F(r) = GNMm/r 2 where mis the star's mass. The work done to move a test mass
iii at r radially through a small distance dr is (force) • (displacement)= F(r)dr.
So the gravitational potential energy V(r) at r is given by
V(r) := -f.
r
00
GNn:_m dr
r
= _ GNmm'
r
The escape velocity v(r) at a separation r from the star's centre is defined so that
it imparts enough kinetic energy to the object in order to overcome the gravitation
potential energy V (r). Show that ~
v(r) = 2GNm.
r
Assuming that mis the mass of a star, for what radius does the escape velocity
equal c, the speed of light? This radius rs is the Schwarzschild radius. Show that
if the density of the star is p, then the condition that the star does not release
light is that its radius R satisfies ~
3 c2
R> --.
87!" GNP
Laplace concluded that the largest objects in the universe are invisible.
15.3 Kruskal extension of MaH

We have seen that while the Kretschmann scalar goes to infinity in finite
proper time as r --> 0 for a radially infalling particle starting from the region
inside the event horizon, there is no such problem at r = 2m for a radially
infalling particle starting from outside the event horizon: it reaches the
event horizon in finite proper time (Exercise 15.1) but the Kretschmann
scalar remains finite. One may wonder if the trajectory of the radially
infalling particle starting from the outside can be patched with the one
from the inside, so that although it looks discontinuous in the coordinates
chosen hitherto, in the patched-up spacetime, with better coordinates, it
appears as a single smooth curve. So we ask:
Is there an extension of MaH to a larger spacetime .MaH, in which a copy
of MaH can be diffeomorphically embedded, so that there is no singularity
corresponding to the event horizon in .MaH?
A: Yes, the Kruskal 6 spacetime. We give an abridged description here 7 .
6 AfterMartin Kruskal (1925-2006), an American mathematician and physicist.
7A detailed description can be found for example in [O'Neill(1983), Chapter 13]. The
exposition here is based on [Oloff(2018), §14.3].
Transformations in the outer region

There are four steps, in each of which we maintain the 0, <p coordinates, and
change the original pair (r, t) to get a final pair (u, v) through the sequence
(r, t) ---+ (r, t)---+ (U, V)---+ (u, v)---+ (u, v). We specify these below:
• In the first step, we keep the Schwarzschild coordinate time t, but re-
place r by r, r
given by = r + 2m log( r - 2m). This has the effect of
mapping the half-plane {(r, t) :r > 2m} to the entire plane R 2 , in such a
way that the radial incoming and outgoing lightlike geodesics become
45° straight lines, as shown in Figure 15.1: Indeed, a point (r, t) on
such a curve in the (r, t)-plane is mapped to the point (r, t), where
t = ±(r+2mlog(r-2m))+c = ±r+c.
• In the second step, the transformation is (r, t) >-+ (t- r, t + r) =: (U, V),
which is a 45° anticlockwise rotation, followed by a reflection in the
vertical axis, followed by a dilation by y'2.
• In the third step, we compress the entire planar region to the second
quadrant via the map (U, V) >-+ ( - e 4~, e ln ) =: (u, v).
• In the final step, we use again the composition of a 45° anticlockwise
rotation, a reflection in the vertical axis, and a dilation, now by 1/v'2:
(u, v) >-+½(v - u, v + u) =: (u, v).
Altogether, incoming and outgoing light rays run along the 45° half-lines
v + u = c, respectively v = u - c in the (u, v) plane. We have brought the
original r = 2m event horizon in the (r, t )-plane (along which lied the edges
of light cones in the (r, t)-plane) along a 45° line L, where u= v, in the first
quadrant of the (u, v)-plane. Soon, we will transform the interior region
0 < r < 2m in a similar manner, and glue the so-obtained new region in the
(u, v )-plane along this line L.
We now calculate the metric in the new coordinates (v, u, 0, ¢). First,
we find the overall transformation (t, r) >-+ (v, u):
r
e 4-m t t t t
= -2-(e4"' +e- 4,,.' e4"' -e- 4,,.)
= e 4:-ne½Iog(r- 2ml(cosh-t- sinh-t-)

4m' 4m
= e 4:-n-Jr-2m(cosh-t-, sinh-t-). (15.7)
4m 4m
Black holes 337
r=r + 2mlog(2m - r) ~ mlog(r - 2m)
/
1:
N
r II
.... r
l U= t - r
V = t +r
U = t -r l
V=t +r
V V
u u
l ~
U
~
_..JL
= e 4m
v =e=V
~
U = -e _..JLl
~ = e-:rm
V
V
4m
V
v
u
u
V- U v- u
U=-
2 / "= -
2
V+U V+ U
V =- V =-
2 2
Fig. 15.1 Transformations9 in the inner and outer regions of Matt -
Thus v2 - u 2 = - e2:-n (r-2m) = (2m-r)e2:-n. With f(r) := 1- 2m,

r
r
v2 - u2 = -r e 2mJ(r) . (15 .8)
--------------
This picture is based on [Oloff(2018) , Fig. 14.7, p.232].
9
An expression for t in terms of u, v is obtained by dividing v by u:

-1 V
t(u,v) = 4mtanh -.
u
Obtaining r in terms of u, vis not possible in such a straightforward manner,
as it involves solving a transcendental equation. We will need the following.
Lemma 15.1. The map h: (-1,oo)--+ (-¼,oo), given by h(x) = xex, is

bijective, smooth, and has a smooth inverse.
Proof. For x > -l, h'(x) = ex +xex = (1 +x) ex> 0, and so his strictly
increasing (and hence injective) on (-1,oo). As xexlx=-1 =-¼,and
lim xex = +oo,
X-+CX)
we have h((-1,oo))=(-¼,oo), and his onto (-¼,oo). Thus his bijective.

The map h is C 00 , and h' is positive on (-1, oo). By the inverse function
theorem, the inverse map 10 h- 1 : ( -¼, oo) --+ (-1, oo) is smooth too. □
It can be seen from (15.8) that
r(u,v) = 2m(1 + h- 1 (u::: 2

) ). (15.9)
Using (15. 7), we have

OU V ov u
ot 4m' ot 4m'
OU u l ov v l
or 4m f(r)' or 4m f(r).
Using these, we obtain
V U
Ot = 4m Ou + 4m Ov
u l V l
Or = 4m f(r) Ou+ 4m f(r) Ov-
From these two equations, using v2 - u 2 = - re 2~ f (r), we solve for Ou, Ov:
10 We remark that for y ~ 0, the unique x ~ 0 such that xex = y is sometimes denoted
by Wo(y), and the so-obtained function Wo is called the Lambert W function; see e.g.
[Corless, Gonnet, Hare, Jeffrey and Knuth (1996)].
Black holes 339
Hence the components of g in the new coordinates are given by
16m2
Similarly, g(ov, ov) = -------,,--- and g(ou, ov) = g(ov, ou) = 0. Thus,
re2m
16m2
g = ----::i::-(-dv © dv +du© du)+ r 2 d0 ® d0 + r 2 (sin0) 2 d¢ © d<j), (15.10)
re2m
where r=r(u, v), given by (15.9).
Transformations in the inner region

Again there are four steps, in each of which we maintain the 0, ¢ coordinates,
and change the original pair (r, t) to get a final pair (u, v) through the
sequence (r, t) _. (r, t) _. (U, V) _. (u, v) _. (u, v) as follows:
• In the first step, as before, we keep the Schwarzschild coordinate time
t, but replace r by r = r + 2m log(2m - r). This map sends the strip
{(r, t): O<r<2m} to the left half-plane {(r, t): r<2mlog(2m)} in the
(r, t)-plane.
• In the second step, we use the same transformation as used in the second
step for the exterior region, namely, (r, t) >-+ (t-r, t+r) =: (U, V), which
has the effect of mapping the previously obtained left half-plane to a
'lower-triangular' right half-plane in the (U, V)-plane.
• In the third step, we use transformation (U, V) >-+ (e 4~, e4':n) =: (u, v)
to compress the triangular region obtained above onto the region in the
first quadrant below the branch of the hyperbola uv
= 2m in the first
quadrant.
• In the final step, we use again a 45° anticlockwise rotation, followed by
a reflection in the vertical axis, followed by a dilation by 1/v'2, that is,
(u, v) >-+ ½(v - u, v + u) =: (u, v).
Altogether, analogous to the calculation done to obtain (15.7), we obtain
(u, v) = e4:;, v12m-r( sinh 4~, cosh 4~)-

Indeed,
1 v -U v -U 1 .!.±.f -t+r .!.±.f -t+r
(u, v) = 2(e'im -e"im, e'im +e"im) = 2(e 4 ,,, -e 4,,, , e 4,,, +e 4,,, )
= e 4 :'n ( sinh _t_ cosh-t-) = e 4:;. e½ Iog( 2m-r) ( sinh _t_ cosh-t-)
4m' 4m 4m' 4m
= e 4:;.V2m-r(sinh-t-, cosh-t-).
4m 4m
Thus again, we have v 2 - u 2 = (2m-r)e2:-n = -re2:-nf(r), and so r(u,v)
is given by (15.9). The equations after (15.9), all the way to the metric
expression in (15.10) remain the same.
Consequently, we take MBH as the smooth manifold n x 8 2 , where
n= {(u, V) E IR 2 : V > - u and v2 - u 2 < 2m}
with the metric given by
16m2
g = -----;:-(-dv ® dv +du® du) + r 2 d0 ® d0 + r 2 (sin 0)2d¢ © de/>,
re2m
where r=r(u, v), given by
r(u,v) = 2m(1 + h- 1 (u:~: 2

) ),
and h : ( -1, oo) ---+ ( -¼,

oo) is given by h(x) = xex. The interior region of
MBH, where 0<r<2m, is mapped to the region nin X 8 2, where
i1in := {(u,v) E i1: V > u}.
The exterior region of MBH, where r > 2m, is mapped to nex x 8 2, with
i1ex := {(u,v) E i1: U > v}.
As each of the maps in the four steps used for the interior/ exterior region
were diffeomorphisms, we see that (nin u nex) X 8 2 with the above metric
g can be considered to be a diffeormorphic 'copy' of (MBH, g) lying inside
MBH• But now we have also glued the original interior and exterior regions,
by mapping the event horizon r = 2m along u = v. The above expression
for g for the extended spacetime MBH has no singularity at u=v. Also, the
function h- 1 is C 00 at O (Lemma 15.1), and so g is a metric for MBH• By
continuity, the metric g solves the field equation also on the event horizon.
In the Kruskal spacetime, radially incoming free falling photons in the
exterior region are straight lines at right angles to the u = v line (corre-
sponding to the event horizon at r = 2m) in the (u, v )-plane, which simply
continue through the u = v line, and move in the interior region also as a
straight line, eventually reaching the singularity at r = 0 (corresponding to
v2 - u 2 = 2m in the (u, v )-plane). Also, it is now clear that particles starting
inside the black hole region cannot escape to the exterior region.
Black holes 341
Remark 15.1. (Maximal extension of MBH)-

The spacetime MBH is not a 'maximal' extension of MBH• There is a larger
smooth manifold in which MBH can be isometrically embedded. Formally,
this is obtained by gluing a time-reversed copy of MBH to the original MBH•
This produces a new singularity (called a white hole, since light can only
come out of it), and a corresponding new exterior region. The Schwarzschild
spacetime is used to model the exterior of a non-rotating star, which even-
tually may collapse to form a black hole. Inside the star, the spacetime
metric can't be modelled by the Schwarzschild spacetime, since the energy-
momentum tensor field is not zero there. During the star's lifetime, the
r-component r*(t) of the worldline 'Y of a particle on the surface of the
collapsing star changes with t, and eventually goes to O if a black hole is
formed. This worldline 'Y lies in MBH, and so does the region r > r*(t).
Thus the part of the maximal extension of MBH, in the complement of the
region where r > r*(t), can be discarded based on these physical grounds.
We refer the reader to [Natario(2021), §2.3] and [Wald(1984), §6.6] for a
detailed exposition.
*
15.4 Miscellanea
In this last section of this chapter on black holes, we make brief remarks
about some topics which will not be discussed here.
Existence and formation of black holes

Based on models of stellar evolution, black holes were theoretically expected
to be formed at the end of the life history of a (massive enough) star when
it has burnt up all of its nuclear fuel, and there is no outward pressure
to stop it from 'collapsing' inwards under gravity. We refer the reader to
[Misner, Thorne and Wheeler(2017), Chapter 32] for a detailed exposition.
The existence of black holes is by now widely believed, based on indirect
observational evidence. Astronomical observations also suggest that there
exists a 'supermassive' black hole at the centre of each galaxy.
Rotating black holes

The Schwarzschild solution to the field equation was found already in 1916,
and as we have seen in this chapter, it also describes a stationary black hole.
However, in 1963, Roy Kerr 11 found a generalisation, which can be thought
11 New Zealand mathematician, born in 1934.
of as a 'rotating' black hole, described by two parameters, m and a (a repre-

senting the angular momentum per unit mass). Thus when a= 0, the metric
for the Kerr spacetime reduces to that of the Schwarzschild spacetime. We
refer the interested reader to [O'Neill(2014)] for a geometrical treatment of
Kerr black holes. Kerr black holes are considered more realistic as most
stars rotate, and should eventually end up as rotating black holes.
Black hole thermodynamics

A 1975 paper by Stephen Hawking showed, based on quantum field theoretic
considerations, that black holes have a temperature, now called the Hawking
temperature, given by
TH= 8 G k
nc3
= 6.18 x 10
-1M0
-Kelvm.
.
(15.11)
7r N am m
Here, ka is the Boltzmann constant, n = h/(21r), h being the Planck con-
stant, m is the mass of the black hole, and M0 is the mass of our Sun. We
refer the interested reader to [Wald(1984), Chapter 14].
Exercise 15.3. When an object falls into a black hole, information is lost. Lack of
information is associated with entropy in statistical physics. Entropy is related
to temperature. Let us use these ideas to give a 'back of the envelope' heuristic
derivation of the Hawking temperature formula.
It was shown by Hawking in 1971 that if two black holes collide forming a
new black hole, then the area of the event horizon surface is bigger than the sum
of that of the original black holes. This result is reminiscent of the second law
of thermodynamics about entropy increasing in any dynamical process. This led
J. Bekenstein 12 to guess that the entropy of a black hole is an increasing function
of the area. But as entropy is additive, and since in the Hawking inequality
mentioned above, the right-hand side takes the sum of the areas, the simplest
guess is to take S ocA. As entropy is ka times a dimensionless quantity, we divide
the area A by the square of an appropriate 'universal' length.
The elementary particles, with kinematics governed by the constants c and
Ii, do not have a length scale, because these two constants cannot be combined
to form a quantity with the dimension of length. But adding mass, we do ob-
tain length, for example the Compton length Re := li/(mc). Similarly GN and c
also cannot be combined to give length. Adding mass again rescues the situa-
tion, since GNm/c2 has dimensions of length. Taking the geometric mean of the
lengths li/(mc) and GNm/c2, the pesky m cancels, and we obtain the universal
length scale, the so-called Planck length, given in terms of the universal physical
constants by fp=,JliGN/c3. Using this, we have S=akaA/.e;, where a is a nu-
merical proportionality constant, which was shown in Hawking's 1975 work to be
12 1947-2015, a Mexican-born Israeli-American theoretical physicist.
Black holes 343
equal to ¼. As we are only giving a heuristic derivation, let us use this numerical
constant, and see ifwe recover the rest of the terms in the formula (15.11) for TH.
A particle of mass !::!..m has 'rest energy' !::!..U = c2 !::!..m. If this enters the event
horizon, then, using T !::!..S = !::!..U, we obtain T !::!..S = c2 !::!..m. Using A= 471" ~, where
Rs := 2GNm/c2 is the Schwarzschild radius, derive (15.11).
Remark 15.2. (Emergent gravity paradigm). We have seen two instances
in spacetime physics, where the existence of a 'horizon' creates a thermo-
dynamics temperature: the Unruh effect in Exercise 5.21 and the black
hole temperature. In the classical viewpoint, the field equations of space-
time are fundamental, and the thermodynamics of spacetime is a conse-
quence. Roughly speaking, in the 'emergent gravity paradigm' one tries
to go the other way, obtaining the field equations as an effective macro-
scopic description of a more fundamental microscopic theory ('atoms of
spacetime'/associated degrees of freedom). The reader who wants to delve
deeper is referred to [Padmanabhan(2008)] and [Padmanabhan(2016)]. *
Chapter 16
Cosmology
In this chapter we will see that in the spacetime-geometry based theory of

gravitation, the field equation is able to give information about the uni-
verse as a whole. To achieve this, we ignore the local details (planets, stars,
galaxies, etc.), and instead adopt a global view by making reasonable as-
sumptions about the energy-momentum tensor field Tin the 'large scale'.
This is akin to how we think of a fluid when we model its flow, ignoring the
individual molecules or particles in the fluid. In smoothed-and-averaged
form, we will assume that the energy-momentum tensor field is that of a
perfect fluid.
We know so far only a small portion of the universe, since we only have
access to the light signals that have reached us (roughly the events in our
past light 'cone'). So in light of this lack of observational knowledge, it
is inevitable that in our cosmological model we will have to make certain
assumptions. (Then after deriving a model using these assumptions, one
makes predictions or computations, which can be compared with certain
observations to examine the validity of the model.) The astronomical ob-
servational data and the radiation we receive from the cosmos is essentially
'spatially isotropic', that is, it is the same in all spatial directions, and no
particular direction is favoured. Thus in our universe (M, g), ifwe are at the
point p EM, and we are the normalised instantaneous observer v E TpM,
then our perceived space v1_, appears the same in all directions. We will
use this as a guiding principle in the development of a cosmological model,
the FLRW spacetime. (What we hitherto called the FLRW spacetime is
really one with 'flat' spatial sections, but in this chapter, we consider more
general FLRW spacetimes.)
We assume that the spacetime M is I x S, where I c JR is an open
interval in JR, and (S, u) is a 3-dimensional connected Riemannian manifold.
345
If (U,x) is an admissible chart for S, then the metric g on Min the chart
(Ix U, (t,p) ,_. (t,x(p))) is
g = -dt ® dt + (a(t)) 2 ~ <Tijdxi © dxi.
As Ot is timelike, if we evolve along the integral curve 'Y of Ot given by

'Y(t) = (t,p), t EI, for some fixed p E S, then (v-y,y(t))1_ can be identified
with TpS. The isotropy assumption implies in particular, that there is no
preferred direction in TpS. We will see that this says something about the
curvature of S, by using the notion of sectional curvatures.
16.1 Sectional curvatures
Definition 16.1. (Sectional curvature).

Let (S,u) be a Riemannian manifold of dimension at least 2. Let p ES.
Let Gr (2, TpS) be 1 the collection of all 2-dimensional subspaces V of TpS.
The sectional curvature of S at p is the map Kp : Gr (2, TpS) -+ R, where
K (V) = u(p)(R(p)(v,w)w,v)
P u(p)(v,v)u(p)(w,w) - (u(p)(v,w)) 2 '
where V=span{v,w} E Gr(2,TpS).
Here R( X, Y) Z = v' xv' y Z - v' y v' x Z - v' [x, Y] Z is the curvature operator
(Exercise 6.15, p.125). We note that the linear independence of v, w shows
that the denominator is positive:
•- u(p)(v,w) 0
z .- v - u(p)(w, w) w =/-
thanks to the linear independence of v, w, and so
0< ( )( )= ) + (u(p)(v,w)) 2
( )( _ 2 (u(p)(v,w))
2
CT p z, Z CT p V, V
( )(
(T p w,w
)
(T
(p)( w,w )
(u(v,w)) 2
= u(p)(v, v) - (T (p )(w,w )
We also need to check that Kp is well-defined, i.e., that it does not depend
on the spanning set {v, w} chosen. We first note that
u(p)(R(p)(v,w)w,v) = vl>(R(p)(v,w)w) = R(p)(vl>,v,w,w)
= R(p)(v, w, w, v).
1 We use this notation, as it is customary to call the set consisting of all k-dimensional
subspaces of a vector space Vas the Grassmannian, and this set is denoted by Gr (k, V).
Cosmology 347
Now let v',w' E TpS =: V be such that span{v',w'} = V = span{v,w}.

Then {v, w} as well as {v', w'} form a basis for the 2-dimensional vector
space V, and so there are numbers a, b, c, d E JR such that v' = av + bw,
w' = cv + dw. Also, the change of basis matrix
is invertible, so that det A=ab-cdi= 0. Using the anti-symmetry properties

of the covariant Riemann curvature tensor field R in the first two and last
two slots, we obtain
u(p)(R(p)(v', w')w', v')
= R(p)(v',w',w',v') = R(p)(av + bw,cv + dw,cv + dw,av + bw)
= (ad - be) R(p)(av+bw, cv+dw, w, v)
= (ad - bc) 2 R(p)(v, w, w, v) = (det A)2u(p)(R(p)(v, w)w, v).
The denominator in the definition of Kp(V) also picks up the same factor:
u(p)(v' v') u(w' w')-(u(p)(v' w')) 2 = det [ u(p)( v'' v') u(p)( v'' w') ]
' ' ' u(p)(w',v') u(p)(w',w')
= det [ au(p)(v, v') +bu(p)(w, v') au(p)(v, w') +bu(p)(w, w')]
cu(p)(v, v') +du(p)(w, v') cu(p)(v, w') +du(p)(w, w')
= det ( [a b] [u(p)(v,v') u(p)(v,w')])
c d u(p)(w, v') u(p)(w, w')
= (detA)det [u(p)(v,v') u(p)(v,w')]
u(p)(w, v') u(p)(w, w')
= (det A) det [au(p)(v, v) +bu(p)(v, w) cu(p)(v, v) +du(p)(v, w) ]
au(p)(w, v)+bu(p)(w, w) cu(p)(w, v)+du(p)(w, w)
t
= (detA)det ( [u(p)(v,v) u(p)(v,w)] [a b] )
u(p)(w,v) u(p)(w,w) c d
= (detA)(det(At))det [u(p)(v,v) u(p)(v,w)]
u(p)(w,v) u(p)(w,w)
= (detA) 2 (u(p)(v,v)u(w,w) - (u(p)(v,w))2).

We now show that if at a point p, all the sectional curvatures match, and
have the common value k(p), then k(p) determines the curvature R(p) at
p. Later in the spacetime context of M = I x S, the isotropy assumption
will imply a constant sectional curvature for S at any point p E S because
otherwise there would be a plane in TpS where the sectional curvature is
the largest, producing a distinguished direction.
Proposition 16.1. Let (S, u) be a Riemannian manifold with the metric

u. Let p E S be such that at p, Kp : Gr (2, TpS) -> JR is a constant function
having the value k(p) ER Then for all x, y, z, w E TpS
R(p)(x, y, z, w) = k(p) ( u(p)(x, w) u(p)(y, z)- u(p)(x, z) u(p)(y, w)).
Proof. The bracketed expression on the right-hand side defines a (0, 4)-
tensor at p, which we call D(p) E T2S(p): For x, y, z, w E TpS,
D(p)(x,y,z,w) := u(p)(x,w)u(p)(y,z) - u(p)(x,z)u(p)(y,w).
D(p) enjoys the same symmetry properties of R.(p):
D(p)(y,x,z,w) = u(p)(y,w)u(p)(x,z) -u(p)(y,z)u(p)(x,w)
= -D(p)(x,y,z,w),
D(p)(x,y,w,z) = u(p)(x,z)u(p)(y,w) - u(p)(x,w)u(p)(y,z)
= -D(p)(x,y,z,w),
D(p)(z,w,x,y) = u(p)(z,y)u(p)(w,x) - u(p)(z,x)u(p)(w,y)
= D(p)(x,y,z,w).
D(p) also satisfies the first Bianchi identity (Exercise 9.3, p.184):
D(p)(x, y, z, w) + D(p)(y, z, x, w) + D(p)(z, x, y, w)
= u(p)(x, w) u(p)(y, z) - u(p)(x, z) u(p)(y, w)
+ u(p)(y, w) u(p)(z, x) - u(p)(y, x) u(p)(z, w)
+ u(p)(z, w) u(p)(x, y) - u(p)(z, y) u(p)(x, w)
= 0.
So the (0, 4)-tensor T(p) := R.(p) - k(p) D(p) also has the above properties.
Let v, w E TpS be any linearly independent vectors, and set V = span {v, w}.
As the sectional curvature at p has the constant value k(p),
T(p)(v,w,w,v) = R.(p)(v,w,w,v)-k(p)D(p)(v,w,w,v)
= (Kp(V) - k(p))D(p)(v,w,w,v) = 0.
If v, w are linearly dependent, then too T(p)(v, w, w, v) = 0 thanks to the
skew-symmetry of T(p) in the first two slots. So T(p)(v, w, w, v) = 0 for all
v, w E TpS. We want to show T(p) = 0, that is, T(p)(x, y, z, w) = 0 for all
x, y, z, w E TpS. First, we note that
0 = T(p)(x+z, y, y, x+z) = 0 + T(p)(x, y, y, z) + T(p)(z, y, y, x) + 0
= -T(p)(x,y,z,y)-T(p)(x,y,z,y) = -2T(p)(x,y,z,y).
Cosmology 349
So T(p)(x,y,z,y)=O for all x,y,z E TpS. Hence

0 = T(p)(x, y+w, z, y+w) = 0 + T(p)(x, y, z, w) + T(p)(x, w, z, y) + 0
= T(p)(x,y,z,w)-T(p)(y,z,x,w),
so that T(p)(x,y,z,w) =T(p)(y,z,x,w). Cycling through the (arbitrary)
x,y,z, we obtain T(p)(x,y,z,w) = T(p)(y,z,x,w) = T(p)(z,x,y,w). The
first Bianchi identity for T(p) then yields 3T(p)(x,y,z,w) = 0, completing
the proof. □
We have seen above that a constant sectional curvature at a point implies

that the point evaluation of the curvature tensor field is determined by the
value of the sectional curvature at that point. But now suppose that at each
point of S, we have a constant sectional curvature at the point. (This is a
'homogeneity' assumption in our spacetime scenario, namely that at each
spacetime point (t, p) E J x S, we assume we have isotropy for instantaneous
observers there.) Then the above result tells us that the curvature tensor
field of S ought to be governed essentially by a function p ...... k(p) : S -> R
A remarkable 'local implies global' phenomenon now happens: the function
k is in fact forced to be a constant! This is called Schur's theorem2 •
Theorem 16.1. (Schur).

Let (S,u) be a connected Riemannian manifold of dimension n ~ 3. Sup-
pose that at each p E S, Kp : Gr(2, TpS) -> R is a constant function, that
is, there exists a function k: S-> R such that for all p ES, Kp(V) = k(p)
for all V E Gr (2, TpS). Then the function k is constant. Moreover,
Ric= (n-l)ku.
Proof. We first note that k is smooth. From Proposition 16.1, R = k D,

where Dis the (0, 4)-tensor field given as follows: for all X, Y, Z, WE TJS,
D(X, Y, Z, W) := u(X, W) u(Y, Z) - u(X, Z) u(Y, W).
If span{v,w}= VE Gr(2, TpS), then we have
k(p) = Kp(V) = R(p)(v, w, w, v). (16.1)

D(p)(v, w, w, v)
Let (U,x) be an admissible chart. For p EU, and i#-j, we have v:=Oxi,p,
w:=Oxi,p are linearly independent. Consequently, D(p)(v,w,w,v)>O, and
U 3 p ...... D(p)(v, w, w, v) is smooth. It follows from (16.1) that k E C 00 (U).
As this happens with each admissible chart, k E C 00 (S).
2 After Axel Schur (1891-1930), a German mathematician.
By Proposition 16.1, R=kD, and so in an admissible chart

.
R lCij = Refij = R~ fijrO' er = k D fijrO' er = k (O'er O'ij - O'R,j O'ir ) O' er
= k (-8: O'ij - 8'; O'ir) = k (nuii - O'ij) = k (n-l) O'ij·
Hence
Ric= k(n-l)u. (16.2)
Taking the trace, we obtain S=k(n-l)n, and so
dS = (n-l)ndk. (16.3)
Taking the divergence in (16.2), we obtain
1 . (*)
2dS = (n-l)d1v(ku) = (n-l)dk. (16.4)
To check (*), we work locally in an admissible chart (U, x): for all W E TJ S,
(div(ku))(W)
= C(Tf S x TJS 3 (0, V) ,_. (v'v(ku))(O~, W))
= C(Tf S x TJS 3 (0, V) ,_. (Vk)u(O~, W) + k(v'vu)(O~, W))
= C(Tf S x TJS 3 (0, V) ,_. (Vk)u(O~, W) + kO)
= (ox;k)u((dxi)~, W) = (ox;k)u(uiioxi, Weoxt) = (ox;k)Weuiiuie
= (ox;k)We8~ = (ox;k)Wi = (Wiox;)k =Wk= (dk)W
From (16.3) and (16.4), 2(n-l)dk= (n-l)ndk, that is, (n-l)(n-2)dk=0.
As n ~ 3, this implies that dk = 0. Hence in any admissible chart (U, x),
Oxik = 0 for 1 ,::;; i ,::;; n, implying the local constancy of k. Since S is
connected, k is constant on S. □
Some constant sectional curvature manifolds in JR 3

Consider the smooth manifold S to be JR3 with its standard smooth struc-
ture. We will determine a metric u that gives constant sectional curvatures
at each point, and hence a globally constant value, say k, for the point-
wise sectional curvature. We use a chart (U, x), where the chart map x
comprises the usual spherical polar coordinates (r,0,¢) on the set
U = IR3 \{(x,y,z) E JR3 : x~O and y=O}.
Suppose that u is given in the chart (U, x) by
u = b(r) dr © dr + r 2 d0 ® d0 + r 2 (sin 0) 2 d¢ © d¢, (16.5)
Cosmology 351
where we want to determine 3 b(r). A calculation analogous to the one done

in Section 14.3 (formally replace k(r) there by b(r), and set f(r) = 0) gives
the nonzero Ricci curvature tensor field components:
• b'(r) • 1 rb'(r)
RICrr= rb(r)' R1c00= -b(r)+l+ 2(b(r)) 2 ' Ricct,ct,= (sin0) 2 Ric00-
We have Ric=k(n-l)u =2ku. In particular, Ricrr=2kurr, i.e.,
b'(r)
rb(r) = 2kb(r). (16.6)
Also, Ric00 = 2k 0-00 gives
l rb'(r) 2
- b(r) + 1 + 2(b(r)) 2 = 2kr . (16.7)
By (16.6), b'(r) =2kr(b(r)) 2. Using this in (16.7), and solving for b(r),
1
b(r) = 1- kr2
If k > 0, then we restrict the r coordinate to r E (0, 1/vk) and consider
only an open ball B = llxll < 1/vk} in
{x E R3 : R3 .
Thus we have seen that if we have a constant sectional curvature k
everywhere for a metric u of the form (16.5) in the chart (U, x), then
b(r)=(l-kr 2 )- 1 , and so the metric is
1 .
u = 1 _ kr 2dr ® dr + r 2d0 ® d0 + r 2 (sm0) 2 d¢ ® d¢. (16.8)
Vice versa, it can be checked that the sectional curvature is equal to k
everywhere for the metric above. One way to do this is to compute the
components of the covariant curvature tensor field R, and using them, show
that we have R = k D. This implies that for each p E S, KP is a constant
function assuming the value 4 k. We will not carry out this computation.
Exercise 16.1. Let u be the metric as described above in (16.8). Let 00 E (0, 1r) and
</Jo E (0, 21r) be fixed. Let I= (0, oo) if k,;;;; 0, and I= (0, 1/,./k) if k > 0. Consider
the radial curve 'Y : I -+ R 3 given by x('Y(r)) = (r, 00 , </Jo), r E I. Determine the
arclength of 'Y as a function of r E I. (This will show that if k =I 0, then we should
not interpret r as the radial distance to the origin.)
3 We use 'b' rather than 'a', since, in our discussion of the FLRW spacetime so far, the
expression for the 'spatial part of the metric' of the metric g of the full spacetime M,
already has a function a(t), and we want to avoid confusion with that function.
4 These metrics are those of the 3-sphere of radius v'k if k > 0, or Euclidean space when
k = 0, or the 'hyperbolic space of radius ~ • if k < 0. The hyperbolic space of radius
a>0 is the open half-space {(x 1 ,x 2 ,x 3 ) E JR3 : x 3 >0} equipped with the Riemannian
metric given by gij = a 2 8ij/(x 3 )2; see e.g. [Godinho and Natario(2014), Example 4.2]
or [Lee(2018), pp.62-67].
16.2 FLRW spacetime
As mentioned earlier, the isotropy of the spatial section for each instan-
taneous observer motivates the spacetime model M =Ix S, where S has
constant sectional curvature at each point. So we consider the following.
Definition 16.2. (FLRW spacetime).

Let a : J -> (0, oo) be a smooth function, where I c R is an open interval.
Let k ER, and
S := R 3 if k:::;;O,
S := {(x,y,z) E R 3 : x 2 +y 2 +z 2 <k} if k > 0,
with the standard smooth structure from R 3 • Let M be the smooth mani-
fold M =Ix S. We equip M with the Lorentzian metric g, given as follows:
In the chart (Ix U, id1 xx), where (U, x) is the spherical coordinate chart 5 ,
g = -dt ® dt + (a(t)) 2 C-~r 2 dr ® dr + r 2 d0 ® d0 + r 2 (sin0) 2 d¢ ® d<j)).

M is given the orientation [O], where n = dx I\ dy I\ dz I\ dt E Sl 4 M, where
(t,x,y,z) are the components of the chart map id1xs in the global chart
(M,id1xs). Also, Mis given the time-orientation V := Ot E TJM. Then
(M, g) is called the FLRW spacetime.
Analogous to the computation done in Example 6.7 (p.124), where k was 0,

we can determine connection coefficients in the chart Ix U, with spherical
coordinates used in U. The nonzero ones are listed below:
r~0=aar 2 r}ct, = aar 2(sin0) 2
rr kr
rr = l-kr2 q 0 =-r(l-kr2) q,cf, = -r(l-kr 2)(sin0) 2
r r0e=r00 r = -rl r:cf, = -(sin0)(cos0)

ef> -ref>..,,.r - ri
r r..,,.
_,_- ,I. -- rtcf, =r:0 =cot 0.
Here · = ¼t-- Using the above, we can calculate the components of the Ricci
curvature tensor field, and the nonzero components are given as follows:
.
R lCrr _ aa + 2a2 + 2k
Rieu= -3~, - k ,
a 1- r 2
Ric00 = (aa + 2a 2 + 2k)r2 , Ricct,ct, = (sin 0)2Ric00-
5 U := JR3 \{(x, y, z) E JR 3 : x ~ 0 and y = O} and x := (r, 0, ¢), where r, 0, <p are the usual
spherical polar coordinates.
Cosmology 353
Thus the scalar curvature is given by
S= 6 aa + a2 + k
a2
So we have now collected all the information for writing the 'geometry' side
of the field equation.
Example 16.1. (Galaxy geodesics).

In FLRW spacetime, let 'Y : J-+ M be given by "((t) = (t, x- 1 (r 0 , 00 , ¢ 0 )) for
t E I, where (U, x) is the spherical coordinate chart from Definition 16.2. We
have v"Y,r(t) = Ot,"Y(t)· So v"Y,"Y(t) is timelike and future-pointing everywhere.
Also, 'Y is an integral curve of Ot- Moreover,
and so 'Y is a geodesic. The curve 'Y is the worldline of a freely-falling galaxy
in the FLRW spacetime. ◊
Exercise 16.2. (A geodesic in a spatial slice).

Consider smooth manifold S c R 3 as in Definition 16.2, with the Riemannian
metric tr defined as follows: Let t E J, and let
tr= (a(t)) 2 C-~r dr ® dr + r d0
2
2 (8) d0 + r2 (sin0) 2 d</> ® d</>),
in the spherical coordinate chart (U, x). It can be shown that the nonzero con-
nection coefficients of the Levi-Civita connection are
r;r = 1 ~{r 2, r00 = -r(l-kr2 ), r4></.> = -r(l-kr2)(sin0)2,
0 0 1
rr0 = r0r = -,
r
r0</.></.> = -(sin0)(cos0),
rr</.> - r</.>r - r
</.> - </.> - 1
-, r0</.>
</.> </.>
= r<J.>0 = cot 0.
The aim in this exercise is show that a certain curve µ, joining the origin in S
to the point pin S with spherical coordinates (ro, 00 , </Jo), is a geodesic. We will
calculate the 'length' ofµ, which will then serve as a notion of distance between
the origin and the point p. Let so > 0, and let r be the solution to
dr(s) = -Jl - k(r(s)) 2 (O ~s~so ) ,
ds ()
at
r(O) = 0.
Let ro := r(so). Defineµ : [0, so] -+ S by µ(s) = x- 1 (r(s), 0o, </>o) for 0 ~ s ~ so.
Determine vµ,µ(s), and tr(µ(s))(vµ,µ(s),Vµ,µ(s))- Show that the length d(t) ofµ,
d(t) :=
o
iso
✓tr(µ(s))(vµ,µ(s),Vµ,µ(s)) ds =so= a(t)
1
--;:;:===;;::::::;;:dr.
o v'l - kr 2
iro
Show thatµ is a geodesic in (S,tr).
Remark: A space slice in the FLRW spacetime (M, g) is a level set oft : M --+ I.
Identifying a space slice {t} x S with S, it is reasonable to define the distance
between (t, 0), (t, p) E {t} x Sc Mas d(t). Indeed, if the distance d(t) between 0
and pis defined as the length of the shortest curve joining Oto p, then d(t) is the
length of the geodesic joining these points by an analogue of Theorem 8.5 (p.177)
in the Riemannian manifold context; see e.g. [Lee(2018), Chap. 6 and Thm. 6.4].
Exercise 16.3. (Hubble's law).

With the notation from Definition 16.2, let p ES be such that x(p) = (ro, 00 , c/>o),
where (U, x) is the spherical coordinate chart. The distance between two galaxies
')'0,1'1 in a space slice {t} x S, where ')'o(s) = (s,O) and 71(s) = (s,p), s EI, is
defined by
d(t) = a(t)i0
ro 1
~ dr.
1-kr2
Define the Hubble constant H : I --+ Ii by
6
H(t) = a(t) for t EI.

a(t)
Show that d(t) = H(t) a(t).
Remark: This is Hubble's law7 : The 'recessional speed' of a galaxy (that is, d(t))
is directly proportional to its distance (d(t)).
16.3 Field equations
We assume that on the matter side, we have a perfect fluid (p, p, V), where
V = Ot in the chart J x U described in the previous section. Here p and p
are assumed to be functions oft only. This is again due to our homogeneity
assumption that in each 'spatial slice', one point is not more special than
the other. Then the energy-momentum tensor field T= (p+p)V1,(8)V1, +pg
has the nonzero components given by
T _ p(t)(a(t)) 2
Tu=p(t), rr- l-kr2 ,
We also assume that the cosmological constant A is absorbed into the matter
side as described in Remark 13.2 (p.300). Then the field equation becomes
Ric-;g = 81rT.
6 Not really a constant, since it depends on t! The constancy is within the spatial slice.
7 After Edwin Hubble (1889-1953), an American astronomer. This was also discovered
by Georges Lemaitre (1894-1966), a Belgian physicist, two years prior to Hubble. The
law shows that the spatial slices are 'expanding'. Historically, this was an important
milestone, since ended the then-prevalent static view of the universe.
Cosmology 355
.
The tt-component gives - - 3aa + a2
- 3a 2
+ k (- 1) = 8n:p. Rearrangmg,
. s
a a
a2 k 8n: (16.9)
2a + 2a = -3 p.
The rr component of the field equation gives
aa+2a 2 +2k_ 3 aa+a 2 +k a2
l-kr 2 a2 l-kr 2
which reduces to the equation9
a a2 k
2- + - + - = -81rp. (16.10)
a a2 a2
The 00, </></> components of the field equation also give the same equation
as above, while the mixed components just reduce to O = 0.
The pair of equations (16.9) and (16.10) have three unknowns: a, p, p.
So we need one more equation if we hope to determine each of these func-
tions. The extra equation is often an equation of state, giving the pressure
as a function of the density. In a linear model p = wp, where w is a constant,
and the table below gives the interpretation of the type of matter:
matter type
I
3 radiation
0 dust
-1 cosmological constant
Exercise 16.4. Show, using (16.9) and (16.10) , that

p + 3(p+p)~ = 0. (16.11)
a
Hint: Differentiate (16.9) with respect tot , and substitute the resulting expression
for a in (16.10) .
Assuming p=wp, show p=a- 3 <1+w), where C is a constant. Hint: }i loga = !·
Thus conclude that for
• radiation, p oc a - 4 ,
• dust, p oc a- 3 ,
• cosmological constant, p is a constant.
8 This is the Friedmann equation, named after the Russian mathematical physicist
A. Friedmann (1888-1925). It was also derived independently by G. Lemaitre.
9 This is sometimes called the acceleration equation because of the presence of the double
derivative with respect to the t-coordinate.
Exercise 16.5. (de Sitter spacetime 10 ).

Suppose that the only contribution in the FLRW spacetime to the energy-
momentum tensor field is the 'matter' due to the cosmological constant, that
• A
1s, p = -p = - 8 ,...
Let A> 0, k = 1, £ = -J37X.,

I= R and a(t) = f_ cosh(t/£) for t E /. Show that the
field equations are satisfied. This spacetime is called the de Sitter spacetime11 •
Now let A<O, k=-1, f=y'-3/A, I= (O,br), and a(t) =fsin(t/£) fort E /. Show
that the field equations are satisfied 12 •
16.4 Simple case solutions and the big bang
In this section, we will assume matter to be dust, so that w = 0, i.e., the

fluid is pressureless. In this simple scenario, we can solve the Friedmann
equations explicitly. In each of the cases of constant spatial sectional curva-
tures k < 0, k = 0, k > 0, we will see that the solution a: J-+ (0, oo) living on
a 'maximal' interval I= (tmin, tmax) tends to Oas t-+ tmin· For simplicity,
by translating the t-coordinate, we make this happen at tmin = 0. We will
also see that then the scalar curvature S tends to oo as t '\. 0, showing that
there is a 'singularity' at this end point 't = O', which is referred to as the
'big bang'. We first show the following.
Lemma 16.1. C:=pa 3 is a constant.
Proof. As p=wp=0p=0, using (16.11), we obtain
:/pa3 ) = pa 3 + 3pa 2 a = a3 (iJ + 3(0+p)~)
= a3 (p + 3(p + p)~) = a 0 = 0.
3
Below, we will assume that the universe is not empty, so that pa3 =C>0.
10 After Willem de Sitter (1872-1934), a Dutch mathematician and astronomer.

11 A famous open problem in cosmology is the 1977 'cosmic no hair conjecture' due
to Gibbons and Hawking, roughly speaking saying that all solutions to the Einstein
equations with a positive cosmological constant 'eventually' look alike, in the sense that
they resemble the de Sitter spacetime. The 'no-hair' is picturesque language suggesting
'no distinguishing features'.
12 This spacetime corresponds to an open region of a bigger spacetime, namely the 'anti-
de Sitter spacetime'. We refer the reader to [Natario(2021), §2.4.3] for details.
Cosmology 357
The case k = 0
. a2 871' 871' 0 . 871'
Equation (16.9) becomes a 2 = 3 p= 3 a3 , and so aa 2 = 3 o. Thus 13
d .a. 3 C· 3 ~
dta 2 = 2 vaa= 2 y 3 0=.a. (16.12)
Solving this on I= (0, oo) such that 14 lima(t)=O, we obtain, with ,8:=ai,
t--+O
a(t) = ,Bti.
Then the scalar curvature S is given by
s = 6 aa + a2 + 0 4
3t2
a2
and so S - oo as t '\. 0. After a rescaling of the t variable, the metric is a
constant multiple of the one considered in §9.2.
The case k>O

We will see that the solution a starts from 0, grows to a maximum value,
and reduces back to O at a value tmax < oo. Rather than solving (16.9) and
(16.10), we will just check that a function given below solves them. Let
tmax = 8; 2 k~' and h : (0, 271') - (0, tmax) := I be given by
h(r) = 471' ..£.. (r - sin r) TE (0, 271').
3 k./k '
Then h' (T) = 4; kS;; (1- cos T) > 0, and so h is a diffeomorphism. Moreover,
h(0,211')=1. Set
471'0
a(t) = 3 k(l - cos(h- 1t)), t E J.
Then it follows that
a(t) = -Jk sin(h-lt) and ii(t) = k2
1- cos(h- 1 t) 4; 0(1- cos(h- 1 t)) 2 ·
Using these expressions, (16.9) holds after substituting p = O/a3:
a2 k
-
a2 +a2 -
k(sin(h- 1t)) 2 k2 k2
-,------,-.,......,...~--,--------+k---,---------
(l-cos(h-1t))2 (4; 0) 2(1-cos(h- 1 t)) 2 (4; 0) 2(1-cos(h- 1t)) 2
2k 3 871' 0 871'
=--=-p
(4;0)2(1-cos(h- t)) 1 3 3 a 3 3 ·
13 We assume that pointwise a> 0.
14 It is clear that every solution a to ~a! = a becomes O at some finite to Eli, and we
dt
choose this to = 0 by shifting the t-coordinate.
·2 k
Also (16.10) is satisfied: Using the above expression for \ + 2 , we have
a a
a a2
2-+-+-
k
a2 2 a a
-k 2 k 2k 3
=2~----- -2- ~ - - - - -1- - + ~ - - - - - -1- -3
4;C(l- cos(h- t)) 4;C(l- cos(h- t))
1 (4;C)2(1- cos(h- t))
= 0 = -81rp.
The graph oft >--> a(t) is a 'scaled version' of the cycloid. To see this, we
note that if a circle of radius R rolls on the x-axis, then the coordinates of
the point P on the circle, starting from the origin, are given by
(Rr-Rsin r , R-Rcosr) ,
where T is the angle that OP makes with the vertical, as shown in the
following picture.
(0 , 0) X
In our case, define R = 4; f Also, for t E (0, tmax) , if T E (0, 21r) is such
that h( T) = t, then
(t,a(t)) = (h(r),a(h(r))) = (),/Rr-Rsinr), R-Rcosr),
which, after a dilation of the x-axis, is a point on the cycloid. As t ----> 0, we

have T := h- 1 t----> 0, and a(t) ----> 0. Also, we compute
•· •2 k 6k 3 1
S = 6 aa + a + = --,---------- ~ oo
a2 (4; C) 2 (1 - cos(h- 1 t)) 3 ·
Thus again we have a singularity at t = 0.

Cosmology 359
The case k<O

This is similar to what we did above, except that we use hyperbolic functions
instead of sin/cos. Define h: (0,oo)-+ (0, oo)=:l by
471' C
h(r) = 3 (-k)\/-k((sinhr)-r), T > 0.
Then h'(r) = 4; (-kfH((cosh r)-1) >0. So his a diffeomorphism. Define

471' C
a(t) = 3 (-k) (cosh(h- 1 t) - 1) , t EI.
Then it follows that

.. ( ) k2
at =-4;C(cosh(h- 1 t)-1) 2 '
Using these expressions, one verifies in the same manner as in the k > 0
case, that (16.9) holds after substituting p=C/ a 3 :
a2 k 2(-k) 3
- + - = - - - - - - - - - = - - = -p
871' C 871'
a2 a2 ( 4;C) 2 (cosh(h- 1 t) -1) 3 3 a3 3 ·
Also, (16.10) is satisfied: Using the above expression for ~ + !,r, we have
ii a2 k
2-+-+- = 0 = -87l'p.
a a2 a2
We have a singularity at t = 0, since
S= 6 aii+a 2 +k=6(-k) 3 1 ~ 00
a2 ( 4; C) 2 (cosh(h- 1 t) - 1) 3 ·
Qualitative plots of a versus t in each of the three cases are shown below.
k= O
k >O
0 t
Exercise 16.6. (ACDM model).

One of currently adopted models of the universe is the FLRW spacetime model
with k=0 and A>0. We assume the presence of dust 15 (perfect fluid with p=0).
Recall from Example 13.1 (p.302) the Friedmann equations
•2
a= (81r p+
3
A) 2
3 a, and a=
.. (A3 - 41r( )~
3 p+ 3p~a.
We assume that p is a function of t alone. Also, we suppose that p, a, a have
limits as t-+ 0, and consider their continuous extensions defined on [0, oo). From
Exercise 13.3, it follows that pa 3 is a constant (since p = 0). Then M := 8; pa 3 is
a constant, assumed positive. Set H =a/a, A:=a(0), and L= Show that -J:f!A.
2 M
H = a 3
1 · 3( 1
+ L2 , and H = 2 £ 2 -
2)
H .
Since we have 16 H(O) > t,
we can locally define the function q(t) = (l/L~-H. Use
the differential equation for H to show that q satisfies the differential equation
. 3 3
q = "'iq- 2·
. () 1 ((1/L)+H(O))ef-t_((l/L)-H(O))
Solvmg for q, show that Ht =--'------'---,,--'----------'-.
L ((1/L)+H(O))eft+((l/L)-H(O))
Prove that a(t) = 2~ (( 1+ ✓1+ ~: 2

)e¾t+ (1-✓1+ ~: 2
)e-¾t) ¾.
Exercise 16. 7. (Particle horizon).

Consider the FLRW spacetime with k = 0. Suppose that we are on the integral
curve of Ot given by ")'o(t) = (t, 0), t E /. Let 0o E (0, 1r) and </Jo E (0, 21r) be fixed.
Let .X : [a, ,B] -+ M be a radial null geodesic, given by .X( s) = (t( s), x( s)), s E [ a, ,B],
where x(s) E IR.3 has spherical coordinates (r(s), 00 , </> 0 ). Moreover, suppose that
pointwise t' >0 and r' <0, and 0<t(a) =to<T=t(,B), r(a) = R, r(,B) =0 (so that
the light emitted at to and from a point with the radial coordinate R, reaches us
at time T). Prove that
R = iTto -(t dt < IT -(t dt
a
1)
O a
1) =: RH(T).
RH(T) is called the particle horizon at time T. The above inequality means that,
for us to receive a radially emitted light signal from the past, the radial coordinate
of the emitter at the moment of emission can be at most RH(T). In other words,
we can receive no signals at time T from those galaxies whose radial coordinate
in the past was always bigger than RH(T).
Notes
Section 16.1 is based on [Oloff(2018), §15.1].
15 The 'CDM' part of the abbreviation, standing for 'cold dark matter'.
16 Indeed, M>O gives a(O)>O too, which yields H(0)2 = ~+-b->-b--
Solutions
Chapter 1
Solution to Exercise 1.1

By the definition of Vpq and Vqr, we have q = p + Vpq, r = q + Vqr·
(Al)
Thus r = q + v qr = (p + v pq) + v qr = p + (v pq + v qr).
By (A3), Vpq + Vqr = Vpr·

Let (u,v) E JR 2 . We wish to find the point p= (x,y,z) E S2 \{n} such that
'Pn((x,y,z))=(u,v), i.e.,
(1:z' l~J = (u,v).

For such a triple (x, y, z),
2 2 x2 y2 x2 + y2 1- z2 1+z
u +v = ---+----'--= _ ___;__ = - - - = - -
(1-z)2 (1-z) 2 (1-z) 2 (1-z) 2 1-z'
u 2 +v 2 -1
and so z = 2 2 . Also,
u +v +1
361
Then
2 2 2 4u 2 +4v 2 +u4 +v 4 +1+2u 2v2-2u 2-2v2
x +y +z = (u2+v2+1)2
So p E S 2 • Also p =I= n, since otherwise
(
2u 2v u 2 +v 2-1) = (O O l)
+v +1' u +v +1' u 2 +v 2 +1
2 u2 2 2 ' ' '
and comparing the first two entries, we get u = 0 and v = 0, implying
u 2 +v 2 -1
- - - - = -1 =I= 1
u 2 +v 2 +1 '
a contradiction. Hence p E S2 \ { n}. Finally,
1 ( 2u 2v )
'Pn(P) = 1- u2+v2-1 u2+v2+1' u2+v2+1 = (u, v).
u 2 +v 2 +1
Hence (cp ) -1 (u v) = ( 2u 2v u 2 +v 2-1)

.
n ' u2+v2+1' u2+v2+1' u2+v2+1

Let cp(x) = x 3 , x ER As cp'(x) = 3x2 > 0 for x =I= 0, cp is strictly increasing
on (0, oo) and on (-oo, 0). It follows that cp is strictly increasing on JR, and
hence is injective. We have cp((0,oo)) = (0,oo), cp((-oo,0)) = (-oo,0), and
cp(0) = 0, so that cp(JR) = JR. Also, cp(JR) = JR is open. Thus (JR, cp) is a
chart on R

Let (x, y, z) E Ux+· Then x > 0, and as x 2 + y 2 = 1, IYI < 1. So we have
'Px+(Ux+) c (-1, 1) x R We will show that 'Px+ : Ux+ --+ (-1, 1) x JR is a
bijection.
z
Solutions 363
If 'Px+(x, Y, z) = 'Px+(x, Y, z), then (y, z) = (y, z), and soy= Y, z = z. As

(x,y,z), (x,y,z) EC, x 2 + y 2 = 1 and x2 + y2 = 1. But since x,x > 0, we
obtain x = y11 -y2 = ~ = x as well. So 'Px+ is injective.
Now let (y,z) E (-1,l)xll. Definex := ~ > 0. Thenx 2 +y 2 = 1
and x > 0, so that (x,y,z) E Ux+· Also 'Px+(x,y,z) = (y,z). Thus 'Px+ is
also surjective onto ( -1, 1) x ll. Finally, 'Px+ (Ux+) = ( -1, 1) x ll, which is
open in ll 2 •

Let
Ux+ = {(x,y,z) EC: X > 0}, 'Px+: Ux+--> ll 2 , 'Px+(x,y,z) = (y,z),
Ux- = {(x,y,z) EC: X < 0}, 'Px-: Ux---> ll 2 , 'Px-(x,y,z) = (y,z),
Uy+= {(x,y,z) EC: y > 0}, 'Py+: Uy+--> ll 2 , 'Py+(x,y,z) = (x,z),
Uy-= {(x,y,z) EC: y < 0}, 'Py-: Uy---> ll 2 , 'Py-(x,y,z) = (x,z).
Then Ux+ u Ux- u Uy+ u Uy- = C: For all (x, y, z) EC, x 2 + y 2 = 1, and
so that it cannot be that both x and y are 0, i.e., either x -# 0 (in which
case either x > 0 or x < 0) or y-# 0 (and then either y > 0 or y < 0). We
have Ux+ n Ux- = 0 and Uy+ n Uy-= 0- So we have the eight transition
maps
'Py+ ocp;;~: 'Px+(Ux+ n Uy+)-+ 'Py+(Ux+ n Uy+), (y,z) >-+ (~,z),
'Py- ocp;;~: 'Px+(Ux+ n Uy-)-+ 'Py-(Ux+ n Uy-), (y,z) >-+ (~,z),
'Py+ ocp-;;;2: 'Px-(Ux- n Uy+)-+ 'Py+(Ux- n Uy+), (y,z) >-+ (-~,z),
'Py- ocp-;;;2: 'Px-(Ux- n Uy_)-+ 'Py-(Ux- n Uy-), (y,z) >-+ (-~,z),
'Px+ o cp;~ : 'Py+(Uy+ n Ux+) ---+ 'Px+ (Uy+ n Ux+), (x, z) >-+ ( Vl - x 2, z),
'Px- o cp;~ : 'Py+(Uy+ n Ux-) ---+ 'Px-(Uy+ n Ux-), (x, z) >-+ ( Vl - x 2, z),
'Px+ ocp;2: 'Py-(Uy- n Ux+)---+ 'Px+(Uy- n Ux+), (x,z) >-+ (-~,z),
'Px- ocp;2: 'Py-(Uy- n Ux_)---+ 'Px-(Uy- n Ux-), (x,z) >-+ (-~,z),
and each of these is a C 00 map. For example, we give the details below for
the map 'Py+ o cp-;;;2. Firstly, we note that
Ux- n Uy+={(x,y,z) EC: x < 0, y > 0}.
Thus 'Px-(Ux- n Uy+)={(y,z) E ll 2 : 1 > y > 0}. Hence we have that for
(y,z) E 'Px-(Ux- n Uy+),
cp;;.!:_(y,z) = ( - ~ , y , z ) , and ('Py+ ocp;;.!:.)(y,z) = ( - ~ , z ) .
So 'Py+ o cp;;.!:_ is C 00 on 'Px-(Ux- n Uy+)= {(y,z) E ll2 : 1 > y > 0}.

• is associative: For (bi, Ai) E llm x GL(m, ll), i = 1, 2, 3, we have
(b3, A3) · ((b2, A2) · (b1, A1)) = (b3, A3)· (b2 + A2b1, A2A1)
= (b3 + A3(b2 + A2b1), A3(A2A1))
= ((b3 + A3b2) + (A3A2)b1, (A3A2)A1)
= (b3 +A3b2,A3A2)·(b1,A1)
= ((b3, A3)· (b2, A2)) · (b1, A1).
Let (b, A) E llm x GL(m, ll). With the zero vector OE llm and the identity
matrix I E GL(m, ll), (0, I)· (b, A) = (0 + I b, IA) = (b, A), and also
(b, A)· (0, I) = (b + A 0, AI) = (b, A). So (0, I) serves as the identity
element with respect to •.
Finally, for (b, A) E llm x GL(m, ll), (-A- 1 b, A- 1 ) E llm x GL(m, ll).
Moreover, (b,A)·(-A- 1b,A- 1) = (b + (A(-A- 1 b),AA- 1 ) = (0,I), and
(-A- 1 b, A- 1 )·(b, A)= (-A- 1 b + A- 1 b, A- 1 A)= (0, I). So every element
in llm x GL(m, ll) has an inverse with respect to •.
Solution to Exercise 1. 7
Denote the compatibility relation by~. Reflexivity (Au A= A is an atlas
on M for every atlas A on M) and symmetry (A1 uA2 = A2 uA1 is an
atlas on M for atlases A1,A2 on M such that A1 ~ A2) are clear. We
now show transitivity. Let A1, A2, A3 be atlases on M such that A1 ~ A2
and A2 ~ A3. Thus A1 u A2 and A2 u A3 are atlases on M. As the
charts in A 1 alone cover M, certainly the charts in A 1 u A3 cover M. So
to show A 1 u A 3 is an atlas, we only need to show that the transition
maps are C 00 • Clearly if both charts are from A 1 , or if both are from
A3, then there is nothing to prove. So let (U, cp) E A1, (V, '¢) E A3, and
Un V =I= 0- Let p E Un V. Then there exists a chart (W, a) E A 2 such
that p E W. As p E U n W, and since A1 u A2 is an atlas, the map
a o cp- 1 : cp(U n W)-> a(U n W) is C 00 • Similarly, asp E W n V, and
as A 2 u A3 is an atlas, the map '¢ o a- 1 : a(W n V) -> '¢(W n V) is
C 00 • Now the map '¢ o cp- 1 : cp(U n V n W) _. '¢(U n V n W) is C 00
because'¢ o cp- 1 = ('¢ o a- 1) o (a o cp- 1) is the composition of two C 00
maps. Hence '¢ o cp- 1 is C 00 in a neighbourhood of cp(p) E cp(U n V).
As p E U n V was arbitrary, '¢ o cp- 1 is C 00 on cp(U n V). Similarly,
cp o '¢- 1 : '¢(U n V) -> cp(U n V) is C 00 • So all the transition maps in
A1 u A3 are smooth, and so A1 u A3 forms an atlas. Thus A1 ~ A3.
Solutions 365

The maps 'P+, cp_, '¢+, 1/J- are clearly injective, and moreover we have
'P+(U+) = cp_(U_) = 1/J+(V+) = 'lj;_(V_) = (-1,1) is open in ll. Also
U+u U_u V+u V_= M, and
'P+(U+ n u_)=0, cp_(U+ n u_)=0, 'P+(U+ n V+)=(0,1),
'1/J+(U+ n V+) = (0, 1), 'P+(U+ n v_) = (-1, 0), '1/J-(U+ n v_)=(0,1),
cp_(U_ n V+) = (0, 1), '1/J+(U- n V+) = (-1, 0), cp_(U_ n v_) = (-1, 0),
'lj;_(U_ n v_) = (-1, 0), 'l/J+(V+nV_)=0, 'l/J-W+nV_)=0
are all open. As the transition maps given below are all smooth, (M, [A])
is a smooth manifold.
'l/J+ o (cp+)- 1 : 'P+(U+ n V+)---+ 'l/J+(U+ n V+) is (0, 1) 3 /3 >-+ 1 - /3 E (0, 1),
'P+ o ('l/J+)- 1 : 'l/J+(U+ n V+)---+ 'P+(U+ n V+) is (0, 1) 3 a>-+ 1 - a E (0, 1),
'l/J- o (cp+)- 1 : cp+(U+ n V_)---+ 'l/J-(U+ n V_) is (-1,0) 3 /3 >-+ 1 + /3 E (0, 1),
'P+ o ('l/J_)- 1 : 'lj;_(U+ n V_)---+ 'P+(U+ n V_) is (0, 1) 3 a>-+ a -1 E (-1, 0),
'l/J+ o (cp_)- 1 : cp_(U_ n V+)---+ 'l/J+(U_ n V+) is (0, 1) 3 /3 >-+ /3 - 1 E (-1, 0),
cp_ o ('l/J+)- 1 : 'l/J+(U_ n V+)---+ cp_(U_ n V+) is (-1, 0) 3 a>-+ 1 + a E (0, 1),
'l/J- o (cp_)- 1 : 'P-(U- n V_)---+ 'l/J-(U- n V_) is (-1,0) 3 /3 >-+ -1-/3 E (-1,0),
cp_ o ('1/J_)-l: 'l/J_(U_ n V_)---+ cp_(U_ n V_) is (-1, 0) 3 a>-+ -1 - a E (-1, 0).

Firstly:
• ui X Vj 3 ( u, V) cp;:._.,Pj ( 'PiU, 1PjV) E llm+n is injective.

Since if (cpiu, 1PjV) = (cp/u, '¢/v), then we have 'PiU = cp/u, giving u = u,
and also '¢iv = '¢/v, which implies v = v.
• (cpi x 1/Jj)(Ui x ½) is open.
(cpi X 1Pj)(Ui X ½) = {(cpiu, 1PjV), u E ui, VE ½} = 'Pi(Ui) X 1Pj(Yj) is
open in llm+n, since 'Pi(Ui) c llm and 1/Jj(l'j) c lln are open.
Thus (Ui x l'j, 'Pi x 1/Jj), i EI, j E J, are charts for M x N.

If (p, q) E M x N, then there exist a Ui E AM such that p E M and
there exists a l'j E AN such that q E l'j. So we have (p, q) E Ui x l'j. Thus
{Ui x l'j, i EI, j E J} covers M x N.
If i, i' EI, j,j' E J, then
is open in llm+n, because 'Pi (Ui n Ui,) is open in llm, and 1Pi (½ x l'j,) is
open in lln.
Now suppose that (Ui, x ½J n (Ui 2 x Vy 2 ) =I= 0- Clearly we have that if

(p,q) E (Ui, x ½J n (Ui 2 x Vy 2 ), then p E Ui, n Ui 2 and q E ½, n Vy 2 • As
the chart transition maps
cp; 2 o (cp; 1 ) - 1 : cp; 1 (U;, n U;J----> cp; 2 (U; 1 n U;J,
'Pi, 0 ('Pi2)- 1 : 'Pi2(U;, ('\ U;J----> cp;,(U;, ('\ U;J,
'lj;h O ( 'lpj, )-1 : 'lpj, (½, ('\ ½J ----> 'lj;h (½, ('\ ½J,
'lpj, o ( 'lj;jJ-1 : 'lj;h (VJ, n ½J ----> 'lpj, (VJ, n ½J ,
are C 00 , it follows that the maps
(cp;, x 'lj;j,)((U;, x ½,) n (Uh x ½2 )) 3 (a,,B)
I(-Pi2 x VJ j2) 0 (,pi1 x VJ ]l )-1
(cp; 2 x 'lj;h)((U;, x ½,) n (Uh x ½ 2)) 3 ((cp; 2 o cp~ 1)(a), ('lj;h o 'lj;;,1)(,B))
and
(cp; 2 x 'lj;h)((U;, x ½,) n (Uh x ½2 )) 3 (a,,B)
I(-Pi1 x VJ)l ) o ('Pi2 x VJ j2 )-1
(cp;, x 'lj;j,)((U;, x ½,) n (Uh x ½ 2)) 3 ((cp;, o cp1/)(a), ('lj;ii o 'lj;;,,1)(,B))

are C 00 maps.

Let O denote the topology induced on !Rm by [A], where A is the atlas
A:= {(!Rm,id)}, and let OE denote the topology on !Rm induced by the
Euclidean metric.
Let Uc !Rm be such that U E 0. Then id(U n!Rm) = U is open in
(!Rm, OE)- So O c OE.
Let U EOE. Then U is open in (!Rm,OE), and so id(U n !Rm)= U is
open in (!Rm,OE)- Thus U E 0. Hence OE c 0.

Suppose there exists an atlas A on C with the desired property. Then
0 := (0, 0, 0) E U for some chart (U, 'P) E A. Let D c 'P(U) c IR 2 be an
open disc of radius E > 0 with center 'P(O).
<p
--+
Solutions 367
Then cp- 1 D is open in C in the subspace topology. So cp- 1 D = C n V,

where V is an open subset of JR3 . But as OE cp- 1 D = C n V c V, there
exists a ball B in JR3 with center O and radius 8 > 0 such that B c V.
Then {O} s;; C n B c C n V = cp- 1 D. If 1r: JR3 ---> JR2 is the projection map
JR 3 3 (x,y,z) ...... (x,y) E IR 2 , then take any nonzero (x,y) E 1r(C n B), and
define p, q by
0-# p := (x,y,y'x 2 +y 2 ) EC n BC cp- 1 n,
0-# q := (x,y,-y'x 2 + y 2 )E C n B c cp- 1 D.
Now cp(p), cp(q) E D\{cp(0)}, and let 'Y: [0, 1] --+ D\{cp(0)} be any contin-
uous path joining cp(p) to cp(q) while avoiding cp(0). (If cp(O) does not lie
on the line segment joining cp(p) and cp(q), then their convex combination
will do. Otherwise, we first move along a concentric circular arc starting
from cp(p), and then follow a straight line segment to cp(q).) Consider the
continuous map f = z o cp- 1 o 'Y: [0, 1]---> IR. Then f is continuous, and
f(O) = (z o cp- 1 o 'Y)(0) = z o cp- 1 (cp(p)) = z(p) > 0, while
f(l) = (z o cp- 1 o 'Y)(l) = z o cp- 1 (cp(q)) = z(q) < 0.
Hence by the intermediate value theorem there exists a c E [O, 1] such that
f(c) = 0, that is, z(cp- 1 ("/(c))) = 0, and so we have cp- 1 ("/(c)) = 0. Thus
'Y(c) = cp(0), a contradiction.

Let A= { (V<>, '¢0 ) : a EA}. Since U is open in M, "P<>(U n V"') is open in
]Rm and "P<>IUnVa is injective. So (Un v<>,"P<>IUnVa) is a chart on u. These
charts cover U because U (Un V"') =Un (UV"')= U nM = U.
aEA aEA
Let a, /3 E A. Since the sets U, V,a are open in M, so is Un V,a. But
then "P<>((U n V,a) n V"') is open in !Rm, that is, "P<>((U n V"') n (Un V,a))
is open in !Rm. Also, the maps '¢,a o '¢;; 1 : "P<>((U n V"') n (Un V,a)) ___. ]Rm
and "P<> o '¢·t : '¢,a((U n V"') n (Un V,a))--+ !Rm, being the restrictions to
open subsets of the two C 00 maps '¢,a o '¢;; 1 : "P<>(V"' n V,a) --+ !Rm and
"P<> o '¢r/ : '¢,a(V"' n V,a) --+!Rm, are C 00 too. So [Au] is an atlas for U.
If (W, a) is admissible for M, then for all a E A, (V"' n W, '¢<>), (V"' n W, a)
are admissible for M. So "P<>(U n (V"' n W))='¢<>((U n V"') n (Un W)) and
a(U n (V"' n W)) = a((U n V"') n (Un W)) are open. Moreover, we have
"P<> o a- 1 : a((U n V"') n (Un W))---> !Rm, a o '¢;; 1 : "P<> ((Un V"')n (Un W))---> !Rm,
being the restrictions of the smooth maps "P<> o a- 1 : a(V"' n W) ___. ]Rm and
a o '¢;; 1 : "P<>(V"' n W) ___. ]Rm to open subsets, are C 00 too.

To show the continuity of f, we will show that each p E M has an open
neighbourhood U such that flu is continuous. As f: M--> N is smooth,
given p EM , there exists an admissible chart (U, cp) for M such that p EU ,
and an admissible chart (V, '!/J) for N such that f(U) c V , and also the map
'ljJ of o cp- 1 : cp(U) --> '!/J(V) is smooth, and in particular continuous. Since
cp : U--> cp(U) and 'ljJ : V--> '!jJ(V) are homeomorphisms, it follows that the
map ¢- 1 o (¢of o cp- 1) o cp: U--> Vis continuous, that is, flu : U--> Vis
continuous.

Let m E M1 , and
• (Ui , 'Pi) be admissible for Mi, i = 1, 2, such that we have m E U1 ,

h2(U1) C U2 , and cp2ofi2ocp1 1 E C 00 (cp1(U1)) ,
• (½ , '!/Ji ) be admissible for Mi, i = 2, 3, such that we have fi 2m E Vi ,
f23 (Vi) C Vi, and 'lp3 0 f230 ¢ 2 1 E C 00 ( ¢2 (Vi)).
We know U2 n Vi is open in M2. As fi2 is continuous, f~/(U2 n Vi) is open

in M 1 . We consider the new charts
• (U1 fl r;/(U2 fl Vi) , 'Pl luu,Ji;/(U2 nVi )) that contains m,

• (Vi , ¢ 3 ), which is admissible for M 3 , and furthermore we have that
(f23o fi2)(U1 fl r;/(U2 fl Vi)) C f23(U2 fl Vi) C h3 Vi C Vi-
Solutions 369
Finally, 'lp30 (J23ofi2)o(cp1lu,nJi:;'(U2nVi))- 1 EC 00 (cp1(U111r;/(U211 Vi))).

Indeed,
'lp3 O /23 O fi2° (cp1lu,nJ;;'(U2nV2))-l
= 'lp3 ° /23 ° ¢2 1 0 ¢2 ° 'P2 1 0 'P2 ° fi2 ° (cp1lu,nJ;;'(U2nV2))-l
is smooth, because it is the composition of the following three smooth maps:
'lp3 O /23 O 'l/J2 1 E C 00 ('!/J2(Vi)),
¢2 0 'P2 1 E C 00 (cp2(U2 11 Vi)), and
'P2 O fi2 O (cp1lu,nJi:;'(U2nVi))- 1 EC 00 (cp1(U111r;:/(U211 Vi))).

Suppose that (p, q) E M x N. Then there exist admissible charts
(U, cp), (V, '!/J) for M, N, respectively such that p EU , q EV. Consider
• (U x V, cpx ¢), which is admissible for M x N , and we have (p , q) E U x V,
• (U, cp), which is admissible for M, 7rM(U x V) = U, and moreover for
(a,/3) E (cp x '!jJ)(U x V) = (cp(U)) x ('!/J(V)), we have that
cp O 7rM O (cp X ¢)- 1(a, /3) = cpo 7rM(cp- 1(a), ¢- 1(/3)) = cp(cp- 1(a)) = a,
and so clearly we have cp o 7rM o (cp x ¢)- 1 E C 00 ((cp x 'l/J)(U x V)).
As (p, q) E M x N was arbitrary, 7rM is smooth.
Smoothness of iq: Let p E M, and (U, x) be an admissible chart for M
containing p. Let (V, y) be an admissible chart for N containing q. Then
(U x V,x x y) is an admissible chart for M x N containing (p,q). Let the
dimensions of M, N be m, n, respectively.
p u
Theniq(U)={(p,q):pE U} c UxV. Also, (xxy)oiqox- 1 :x(U)----> rntm+n
is given by
((x x y) o iq o x- 1 )(() =(xx y)(iq(x- 1(())) =(xx y)(x- 1((),q)
= (x(x- 1(()),y(q)) = ((,y(q))
for ( E xU, and the map x(U) 3 ( f--4 ((,y(q)) E rntm+n is clearly C 00 . But
p E M was arbitrary. Thus iq is smooth.

The 'if' part is clear by taking U = M. We now show the 'only if' part.
Suppose p E U. Let (U, (p) be an admissible chart for M such that p E fJ,
and (V, ¢) be an admissible chart for N such that f(U) c V and the map
¢of o rp- 1 : (p(U) -> !R.n is C 00 . Firstly, (Un U, (p) is an admissible chart for
U. Also, f(U n U) c f(U) c V , and the map¢ of o rp- 1 : (p(U n U)-> !R.n
is C 00 on (p(U n U), being the restriction of¢ of o rp- 1 : (p(U) -> !R.n to the
openset(p(fJc,U). ~ ~ )
~-~
9)
~.
0 f( U )

Suppose f, g are given by f = (x f--4 x½) : M = (IR., [A1])-> (IR., [A2]) = N,
g= (yf-4y 3 ) :N = (IR., [A2])-> (IR., [A1]) =M. Then we have
(f o g) (y) = f (y 3 ) = (y 3 ) ½= y for all y E N, and
(go J)(x)=g(x½)=(x½) 3 =x for all x EM.
Hence f is a bijection from M to N, with the inverse 1- 1 = g : N-> M.
1
f = (x >-+ x"'J)
IR.= N
l
M = IR.
'Pl= (x >-+ x) 1,.,2 = (y ...... y3)

IR. IR.
f is smooth: Let x EM. With
• (U =IR., <p 1 =id), an admissible chart for M with x EU;
• (V = IR., IR., which is C 00 .
As x was arbitrary, f is smooth.
g is smooth: Let y EN. With
• (U = IR., <p 2 = (y f--4 y 3 )), an admissible chart for N with y E U = IR.;
• (V=IR.,<p 1 =id), an admissible chart for M, g(U)=g(!R.)=!R.=V.
For /3 E <p2(U) = IR., (<p1 ogo<p21)(/3) = (ido g) (f3½) = g(f3½) = (f3½ ) 3 = id(/3).
So¢ o go <p- 1 =id: <p(U) =IR._, IR., which is C 00 .
As y was arbitrary, g is smooth. So (IR., [Ai]) is diffeomorphic to (IR., [A2]).
Solutions 371

Let t E ll. Let us take
• (U, cp) = (ll, id), an admissible chart for ll, with t Ell = U.
• Let {e 1 , • • • , em} be a basis for V, and let '¢ : M -+ llm be the map
q >-+ 'l/J(q) := (y 1 , • • • , ym), where q = p + Vpq and Vpq = yiei. Then we
had seen that (M, '¢) is an admissible chart for M. Let v = viei. We
have 'Yv(U) = 'Yv(ll) c M, and for s E cp(U) = ll,
('¢ o 'Yv o cp- 1 )(s) = '¢ o 'Yv(s) = '¢(p + sv) = 'l/J(p + sviei)
= (sv 1 , · · · , svm).
Clearly, the map ll 3 s >-+ ( sv 1 , • • • , svm) E llm is C 00 .
As the choice oft was arbitrary, it follows that 'Yv is smooth.

For I, 9 E Diff( M), we have that Io 9 is a smooth map and a bijection. Also,
(Jo 9)- 1 = 9- 1 01- 1 , which is smooth since 9- 1 and 1- 1 are smooth. Thus
Io 9 E Diff(M). Composition of maps is associative. The identity map
idM : M -+ M is a diffeomorphism, and it serves as the identity element
for (Diff(M), o) because for all IE Diff(M), we have I oidM =I= idM o I-
If I E Diff( M), then 1- 1 is also a smooth bijection, with (J- 1 )- 1 being I,
which is smooth again. Thus 1- 1 E Diff(M), and 1- 1 o I= idM =Io 1- 1 .
Consequently, (Diff(M), o) forms a group.

Let p E G. The inclusion map ip : G-+ G x G, ip(q) = (p, q) for q E G, is
smooth by Exercise 1.15. Thus the composition • o ip of this smooth map
with the multiplication map is smooth too. But this is the map Lp, since
• o ip(q) = •(p, q) = p • q = Lpq for all q E G. Asp E G was arbitrary, Lp-1
is smooth too. But Lp o Lv-1 = ida, because (withe denoting the identity
element in G)
(Lp o Lv-1 )(q) = Lp(p- 1 q) = p(p- 1 q) = (pp- 1 )q = eq = q = id 0 q.
Hence Lv-1 o Lp = Lv-1 o Lcv-1)-1 = ida. So Lp is a bijection with the
inverse (Lp)- 1 = Lp-1. As both Lp and Lp-1 are smooth, we conclude that
Lp is a diffeomorphism.
Chapter 2

If v, w E TpM, then for smooth f, g E C 00 (M) and c E JR, we have that
(v + w)(f +cg)= v(f +cg)+ w(f +cg)= vf + cvg + wf + cwg
= (v + w)f + c(v + w)g, and
(v + w)(f ·g) = v(f ·g) + w(f·g)
= J(p)vg + g(p)vf + J(p)wg + g(p)wf
= J(p)(vg + wg) + g(p)(vf + wf)
= J(p) (v + w)g + g(p)(v + w)f,
and so v + w E TpM.
If VE TpM, a E JR, then for f, g E C 00 (M), and CE JR, we have
(a·v)(f + cg) =av(!+ cg)= a(vf + cvg) = avf + acvg
= (a·v)f + c(a·v)g, and
(a·v)(f ·g) =av(f ·g) = a(f(p)vg + g(p)vf) = af(p)vg + ag(p)vf
= J(p) (a·v)g + g(p) (a·v)f,
and so a • v E TpM.

Let J(p) = c for p EM. Let 1 E C 00 (M) be given by l(p) = 1 for p EM.
Then v(l) = v(l·l) = l(p)v(l) + l(p)v(l) = lv(l) + lv(l) = 2v(l), and
so v(l) = 0. Thus v(J) = v(cl) = cv(l) = c0 = 0.

For f E C 00 (JR), we have 1'i(0) = -1, 1'2 (0) = -1, and
V f = d(fo'Yi) (0)= d(f(t-l)) (0) =f'(t-1) 11 - =f'(-1)
-n,-1 dt dt t-o '
V72,-if = :? 0
1'2 ) (0) = :?(t+ 3t 2 - l)) (0) = f'(t+3t 2 - 1) (1 +6t)lt=O = J'(-1).
Thus for all f E C 00 (M), we have v71 ,-if = v72 ,-if. Hence v71 ,-1 = v72 ,-1-

For f E C 00 (M), ands E J, we have
= d(fo('Yoh))(s)= d((fo'Y)oh)(s)= d(fo'Y)(h(s))h(s)
V7oh,('Yoh)(s) f dt dt dt
= h(s)v'Y,'Y(h(s))f-
Hence v7oh,('Yoh)(s) = h(s) v'Y,'Y(h(s)) ·
Solutions 373

(1) Let L := lim g'(t). If t =I= 0, then applying the mean value theorem for
t-+O
g on the interval with endpoints 0 and t, we obtain
g(t~ =~(O) = g'(0tt)
for some 0t E (0, 1). Given€> 0, let 8 > 0 be such that ifO < lr-01 < 8,
lg'(r) - LI < €. Thus for 0 < It - 0I < 8, we have 0 < l0t - 0I < 08 for
all 0 E (0, 1), and so
- g(O)
lg(t)t-0 - LI= lg'(0tt) - LI<€.
So g'(O) = L = lim g'(t). Thus g' is continuous at 0. Hence g is
t-+O
continuously differentiable on ll.
(2) ll --+ ll is continuous on ll and continously differentiable on
g(n-l) :
ll*. Moreover, lim(g(n-l))'(t) exists. By (1), g(n-l) is continuously
t-+O
differentiable on R So g is n times continuously differentiable on R
(3) We have fort> 0 that f'(t) = be-+ = R1(t)e-t, where R1(t) = fr·
Suppose j(n) = &e-t
Qn
for some n, where Pn, Qn are polynomials. Then
jCn+l) = (p~e-+ + frPne-+ )qn - q~pne-+
q;,
Pn + t2
Pn) ( I
I
Qn - QnPn _,! R _,!
=2 e t =: n+1e t.
qn
So by induction, j(n) = Rnf for all n.
For a fixed n, if Rn = &,
Qn
and Qn = tkQ(t) with k E {0, 1, 2, 3, · · · },
and Q a polynomial such that Q(0) =I= 0, then we have
limRnJCn) = lim Pn(t) e-t = lim Pn(t) e-t = Pn(O) 0 = 0.
t'-.0 t'-.0 Qn(t) t'-.0 Q(t) tk Q(0)
(To see that lim
t',O
T -1/t
= 0, we note that for t > 0,
1/t 11 1 1 1 1
e - 1 + 1! t + + (k+l)! tk+ 1 + > (k+l)! tk+ 1 '
_.! t',O _.!
and so 0 < 7 < t(k+l)!--+ 0. Hence lim
t',O 7
= 0.)
Thus for all n, limJCnl(t) = 0 = limJCnl(t), and so limJCnl(t) exists.

t',O t/0 t-+O
As f E C 00 (ll*), it follows from (1) and (2) that f E C 00 (ll). (Note that
f is continuous on ll: limf(t) = 0, and limf(t) = lime-+ = 0. Thus
t/0 t',O t',O
lim J(t) = 0 = f(0).)
t-+O

With fas in Exercise 2.5, we know that f E C 00 (JR). Then J(-·) E C 00 (JR),
and tj(-·), tf(·) E C 00 (JR) too. We take 'Y(t) = (tf(-t), tj(t)), t E JR. Then
-1/t
'Y is smooth. We note that fort> 0, f'(t) = 7 > 0, and so f is strictly
increasing for t > 0. Thus tf is strictly increasing for t > 0. Also, while f
approaches 1 as t--+ oo, tf--+ oo as t--+ oo, and so the range of tf is (0, oo)
for t E (0, 00 ).
We see that t ...... tf(-t) is strictly increasing for t < 0. Indeed, for
t1 < t2 < 0, we have that -ti > -t2 > 0, and so J(-t1) > J(-t2), giving
(-ti)J(-t1) > (-t2)f(-t2), i.e., tif(-t1) < td(-t2).
Moreover, the range oftf(-·) on (-oo,0) is (-oo,0). Finally, we have
'Y(0) = (Of (-0), Of (0)) = (0, 0).
The partial derivatives of the chart transition map '¢ o (id)- 1 are
or X or y
ox ,Jx2 + y2' oy ,Jx2 + y2'
80 -y 80 X
ox x2 + y2' oy x2 + y2 ·
vl . . vl
[ v2 ] ' g1vmg [ v2 ] = [
,Jx2X+ y2
y
,Jx2 + y2
l
.
and so 00,p = -yox,p + XOy,p·

Solutions 375

vEBw E T(p,q)(MxN): For all f,g E C 00 (MxN) and all c ER,
(v EB w)(J +cg) = v((J +cg)(·, q))+w((J +cg)(p, ·))
= v(J(·, q) +cg(·, q))+w(J(p, ·)+cg(p, ·))
= v(J(·, q)) + cv(g(·, q)) + w(J(p, ·)) + cw(g(p, -))
= (v EB w)(J) + c(v EB w)(g),
(v EBw)(Jg) = v((Jg)(·, q)) + w((Jg)(p, ·))
= v(J(·, q)g(·, q)) + w(J(p, ·)g(p, ·))
= f(p,q)v(g(-,q)) + g(p,q)v(J(·,q))
+ f(p, q)w(g(p, ·)) + g(p, q)w(J(p, ·))
= f(p, q)(v EB w)g + g(p, q)(v EB w)f.
Linearity of (v,w) 2: vEBw:
For V1, V2 ETpM, W1, W2 E TqN, CE R, and j E C 00 (M X N),
( (v1 EB w1) +c( v2 EB w2))f = (v1 EB w1)f +c (v2 EB w2)f
= vif(·,q)+wif(p, ·)+cv2f(·,q)+cw2f(p, ·)
= (v1 +cv2)U(·, q)) + (w1 +cw2)(J(p, ·))
= ((v1 +cv2) EB (w1 +cw2))f.
So (v1 EBw1) + c(v2 EBw2) = (v1 +cv2) EB (w1 +cw2), that is,
T(v1, w1) + cT(v2, w2) = T(v1 +cv2, w1 +cw2) = T((v1, w1)+c(v2, w2)).
Injectivity: Let (v, w)E TpM x TqN be such that T(v, w) = 0, that is,
vEBw=0 E T(p,q)(MxN). Then for all f E C 00 (MxN), (vEBw)f = 0, i.e.,
v(J(·,q))+w(J(p, -)) = 0. Let g E C 00 (M). Then f := go1rM E C 00 (MxN),
and so
0= v(J(·, q))+w(J(p, -)) =v(g o 7rM(·, q))+w(g o 7rM(P, ·))
= v(g)+w(g(p)) =v(g)+0=v(g).
Hence for all g E C 00 (M), vg = 0. Thus v = 0 E TpM. Similarly, for all
h E C 00 (N), wh = 0 (by considering f :=ho 7rN ). Sow= 0 E TqN. Hence
(v,w) = (0,0) E TpMxTqN. So Tis injective.
Surjectivity: We have
m + n = dim(TpM x TqN) = dim(ker T) + dim(ranT) = 0+dim(ranT)
using the rank-nullity theorem. Thus
dim ranT = m+n = dim(M x N) = dim T(p,q) (M x N).
As ran T c T(p,q) (M x N), ran T = T(p,q) (M x N), i.e., T is surjective.

Let v, w E TpM, CE JR, and g E C 00 (N). Then
((dfv)(v+cw))(g) = (v+cw)(g of)= v(g of)+ cw(g of)
= (dfv(v))(g) + c(dfv(w))(g) = (dfv(v) + cdfv(w))(g).
As g E C 00 (N) was arbitrary, dfv(v+cw) = dfv(v) + cdfv(w).
For f E C 00 (M),
1
(d1o(Ot,o))(f) = Ot,o(f 01) = ~~O"(oidiR \dR(O)) = ~~ 01 \0)
= v,,,(o)f = v,,vf.
As f E C (M) was arbitrary, d"fo(ot,o) = v,,p·
00
Solution to Exercise 2 .11

The Jacobian matrix of idR2 o 'l/Jt o id;} = 'l/Jt is given by
o( x - yt) ( ) o( x - yt) ( ) j [1 -tJ
r OX
oy( )
x,y oy
oy(
x,y
) = o 1 ·
OX x,y oy x,y
Thus (d'l/Jt)v(oy,p) = aox,1/J,(v) + boy,1/J,(p), where
by Theorem 2.2. Hence (d'l/Jt)p(oy,p) = -tox,1/J,(p) + oy,1/J,(p)·

Let VE TpM. Then for all I.{) E C 00 (M3),
(d(g o f)p(v))(l.f)) = v(l.f) o (go!))= v((l.f) o g) of)= ((dfv)(v))(l.f) o g)
= (dgf(p) (dfp( V)) )i.p.
Solutions 377
Hence d(g o J)p(v) = (dgf(p) o dlp)(v) for all v E TpM. Consequently, we

have d(g o J)p = dgf(p) o dip-

Considering the composition 1- 1 o I= idM : M-+ M, we have by Exer-
cise 2.12 that (dl- 1 )J(p) o dip= d(idM)p-
Claim: d(idM )p = idTpM.
Proof. Let VE TpM. For all 'PE C 00 (M),
Thus d(idM)p(v) = idTpM(v) for all v E TpM. Hence d(idM)p = idTpM· □

Consequently, (dl- 1 )J(p) o dip = d(idM)p = idTpM· Similarly, considering
Io 1- 1 = idN: N-+ N, we also conclude that dip o (dl- 1 )J(p) = idTJ(pJN•
So it follows that dip : TpM -+ Tf(p)N is an isomorphism. In particular,
dimM = dimTpM = dimTJ(p)N = dimN.

We have

With x'(p) = (x,y,z) for p = (x,y,z) EU'= !R3 , and e= x(p),
[!:~ (e) l :~:i'
= [ o x-1 \x(p)) l
(coscp)(sin0) r(coscp)(cos0) -r(sincp)(sin0)]
= [ (sincp)(sin0) r(sincp)(cos0) r(coscp)(sin0) .
cos0 -rsin0 0
So multiplying from the left by the row vectors (1, 0, 0), (0, 1, 0), (0, 0, 1),
respectively, we can read off the claimed relations.

We have the entry in the i th row and k th column of AB is given by
A;B! = :~:'io x-1\x(p)) :~: o x'-1) (x'(p))
= :~:'i o x-1 \(x o x'-l)(x'(p)) :~x: o x'-1\x'(p))
= il[i th component of u >-+ (x' o x- 1 )((x o x'- 1 )(u))] I

~ ~~
- aui I - 8i
- ouk u=x'(p)- k·
Hence AB = Im. As A is an m x m matrix, it follows that also BA = Im.

For any v E TpM,
d(J-g)p(v) = v(J-g) = f(p)vg + g(p)vf = f(p)dgpv + g(p)dfpv
= (J(p)dgp + g(p)dfp)(v).
Hence d(J-g)p = f(p)dgp + g(p)dfp-

Consider the chart (JR, u ....... u) on R By Theorem 2.2, the matrix for the
linear transformation dft : TtlR >-+ Tf(t)IR 2 with respect to the bases {Ou,t}
and {ox,f(t), oy,f(t)} is given by the Jacobian matrix oft ....... [ cost sint fat
t, which is [ - sin t cost f_ So dftilu,t = -(sin t)ox,f(t) + (cos t)oy,J(t)· Thus
( (dft) *w) (Ou,t)
= w(dft(ou,t)) = w(-(sint)ox,J(t) + (cost)oy,f(t))
= (-(sint)(dx)J(t) + (cost)(dy)J(t))(-(sint)ox,J(t) + (cost)oy,f(t))
= (-sint)(-sint)(dx)J(t)Dx,f(t) + (cost)(cost)(dy)J(t)Oy,J(t)
-(sin t)( cos t)(dx) f(t) oy,f(t) + (cost)( -(sin t) )(dy) f(t)Dx,f(t)
= (- sin t)( - sin t)ox,f(t)X + (cost)( cos t)oy,f(t)Y
-(sin t)( cos t)oy,f(t)X + (cos t)(-(sin t))Ox,J(t)Y
= (- sin t)(- sin t) 1 + (cos t)(cos t) 1 - (sin t)(cos t) 0 + (cos t)(-(sin t)) 0
= (-sint)(-sint) + (cost)(cost) = 1 = Ou,tU = (du)t(ilu,t),
and so (dft)*w = (du)t-

Solutions 379
Chapter 3

Let (p,v) E TM. Let (U,cp) be any admissible chart M such that p EU.
Let (U, cp) be the lift of (U, cp). Clearly, n(U) = U. If cp = (x1,- • • , xm),
(o., /3) E cp(U), and /3 = (/31, · · · , 13m), then
(cpo7rocp- 1)(0.,/3) = (cpon)(cp-l0.,/3iOxi,<p-1a.) = cp(cp- 10.) = 0..
So cp o ?r o cp- 1 : cp(U) -+ !Rm is the map (o., /3) ...... o., which is C 00 . Hence
?r: TM-+ Mis smooth.

Let (p, v) E TM. Since f is smooth, there are admissible charts (U, cp),
(V, '¢), for M, N, respectively, containing p, f(p), respectively, such that
f(U) c V, and'¢ of o cp- 1 E C 00 (cpU). Let (U,cp), (V,1/J) be lifts of
(U, cp), (V, '¢), respectively. Then the charts (U, cp), (V, 1/J) are admissible
for TM, respectively TN, and for each (q, w) E U, we have that
df(q,w) = (J(q), (df)qw) E {J(q)} x Tf(q)N c V.
Let M, N have dimensions m, n, respectively, and denote the component
functions of cp, '¢ by (x 1 , • • • , xm), (y 1 , • • • , yn), respectively. For o. E cp(U),
the components of the vector (df)"'-1a.Oxi,'l'-1a. with respect to the basis
Oyi,f(<p-1a.), 1 ~ j ~ n, are given by
((df)<p-1a.Oxi,<p-1a.)Yj = Oxi,<p-1a.(Yj o J) (1 ~ j ~ n).
Using the linearity of df"'-10., for (o., /3) E cp(U),
1/J o df o cp-1(0., /3)
= 'lp O df( cp- 10., /3i Oxi ,<p-10.) = 1/J(J(cp- 10. ), /3i Oxi,<p-la. (yi OJ)oyi ,f(<p-la.))
= (('¢of ocp-l)o.,/3i:~~of ocp-1)1 (o.),··· ,/3i:~~of ocp-l)n (o.)),
and so the map cp(U) 3 (o.,/3) ...... 1/J o df o cp- 1 (0.,/3) is smooth. Thus
df: TM-+ TN is smooth.

For all p EM,
(V(Jg))(p) =½,(Jg)= f(p)Vpg + g(p)Vpf = f(p)(Vg)(p) + g(p)(VJ)(p)
= (J-(Vg))(p) + (g-(VJ))(p) = (J-(Vg) + g-(VJ))(p).
Since p E M was arbitrary, V (Jg) = f · (V g) + g · (VJ).

Also, V : C 00 (M) -+ C 00 (M) is JR-linear, as V(J +cg)= V f +c V g for
f, g E C 00 (M) and CE JR: Indeed for all p EM, we have
(V(J+cg))(p) = Vp(J+cg) = Vp(J) + cVp(g)
= (VJ)(p) + c(Vg)(p) = (VJ +cVg)(p).

We have
(JV)p = (x 2 y)lc1,1) ½, = 1½, = Vp = (xy)lc1,1)0x,p + x 2 lc1,1)0y,p
= lox,p + loy,p = ox,p + oy,p·
Also, (V J)(p) = Vpf = (ox,p + Oy,p)(x 2 y) = (2xy + x 2 )lc1,1) = 2 + 1 = 3.
Finally, we have that
(df)p(Vp) = (df)p(Ox,p + Oy,p) = ((2xy)lc1,1) + (x2 )lc1,1))8u,f(p)
= (2 + l)ou,1 = 38u,1•

We will use Theorem 3.1, and check that for all g E C 00 (N), we have
(J* V)g E C 00 (N). For q EN, we have
((!* V)g)(q) = (J* V)qg = ((df)J-lq VJ-lq)(g)
= VJ-lq(g OJ) = (V(g O J))(J- 1 q)
= ((V(g o /)) o f- 1 )(q).
So (J* V)g = (V(g o J)) o 1- 1 , which is smooth, since firstly, f is a diffeo-
morphism, and secondly, as V E TJ- M and go f E C 00 ( M), by Theorem 3.1,
(V(g o /)) E C 00 (M).

Suppose that VE TJ-G is left-invariant. Then for all q E G, we have that
((LP)* V)q = ((LP)* V)Lp(p-lq) = (dLp)p-lq Vp-lq = Vp(p-lq) = Vq,
and so (LP)* V = V.
On the other hand, if (LP)* V = V, then for all q E G, we have, by
evaluating both sides at the point Lpq E G, that
(dLp)q Vq = ((LP)* V)Lpq = VLpq = ½,-q·
Solutions 381
V = (G 3 p >-+ (dLp)ev) is left-invariant: First note that Lp o Lq = Lp,q·

Indeed, for all r E G, (LpoLq)r= Lp(q·r) =p·(q·r) = (p·q)·r=Lp,qr.
By Exercise 2.12, (d(LpoLq))e = (dLp)q(dLq)e- Now V is left-invariant as
(dLp)qVq = (dLp)q(dLq)ev = (d(Lp o Lq))ev = (dLp,q)ev = Vp·q·
I is surjective: Let V E g. Then V,, E TeG, and (IVe)p = (dLp)e Ve = ½,,e
(as V is left-invariant) for all p E G. So (IV,,)p = ½,,e = Vp for all p E G,
that is, IV,, = V.
I is injective: Let v E TeG and Iv = 0 E TJ-G. In particular, we have that
(Iv)e = 0 E TeG, and so
0= (Iv)e = (dLe)ev= (d(ida))ev=idTeav=v.
(Here we used Leq = e· q = q = idaq for all q E G.)
For p E R 2 , using Theorem 2.2, we determine the matrix of the linear map
(d'l/J-t)p with respect to the ordered bases (ox,p, oy,p) and (ox,,t,_,p, oy,,t,_,p):
[(d'l/J-t)] = [c~st -sint].

P smt cost
Then, as oy,p is represented by the column vector e 2 E R 2 with respect to
the ordered basis (ox,p, oy,p), the representation of (d'l/J-t)poy,p with respect
to the basis (ox,,t,_,p, oy,,t,_,p) is
[ c~st -sint]
smt cost
[OJ=
1
[-sint].
cost
So
(('1/J-t)*V)'P-tP = (d'l/J-t)p½, = (-sint)ox,,t,_,p+(cost)oy,,P-tP
= ((- sin t) Ox+ (cost) oy),t,_,p•
Hence, as '1/J-t: R 2 -+ R 2 is surjective, ('1/J-t)*V = (-sint)ox+(cost)oy.

For all p EM, we have
(((J*V)cp) of)(p) = ((J*V)cp)(J(p)) = (J*V)J(p)'P
= ((df)pVp)cp = Vp(cp o J) = (V(cp o J))(p),
and so ((!* V)cp) of= V(cp o J).


We will use Exercise 3.8. For all r.p E C 00 (N), and q EN,
<U*[x, Y])r.pJ(q) = (U*[x, Y])r.p uu- 1 q)) = u*[x, YDiu-lq)'P
= ((df)J-lq[X, Y]1-1q)(r.p) = [X, Y]1-1q('P OJ)
= X 1-1q(Y(r.p o J)) - Y1-1q(X(r.p o J))

= X 1-1q( (J*Y)r.p) o J) - Y1-1q( (J*X)r.p) o J)
= ((df)J-lqXJ-lq)((J*Y)r.p) - ((df)J-lqYJ-lq ((J*X)r.p)
= (J*X)q((J*Y)r.p) - (J*Y)q((J*X)r.p)
= ((J*X)((J*Y)r.p) - (J*Y)((J*X)r.p) (q)
= ([f*X, f*Y]r.p)(q),
and so f*[X, Y] = [f*X, f*Y].

For all g E C 00 (M),
[f X, Y]g = f X(Yg) - Y(J(Xg)) = f X(Yg) - (YJ)(Xg) - JY(Xg)
= f(X(Y g) - Y(Xg)) - ((Y J)X)g = f[X, Y]g - ((Y J)X)g
= (J[X, Y] - (Y J)X)g.
Thus [f X, Y] = f[X, Y] - (YJ)X.
In particular, for M = IR 2 , X = Ox, Y = oy, f = l + y,
[(1 + y)ox, oy] = [f X, Y] = f[X, Y] - (YJ)X
= (1 + y)[ox, oy] - (oy(l + y))ox
= (1 + y)O - lox = -ox.

We have
[V, W]xi = V(Wxi) - W(Vxi) = V(Wioxixi) - W(Vioxixi)
= V(W\51) - W(Vi8f) = V(Wi) - W(Vi)
= Viox,Wi -Wiox;Vi.
Thus [V, W] = ([V, W]xi)oxi = (Vioxi Wi - Wioxi Vi)oxi •
Solutions 383

Suppose (p, w) E TM*. Then there exists an admissible chart (U, cp) for M
containing p. Let the chart (U, cp) for TM* be the lift of (U, cp). Clearly,
7r(U) = U. Moreover, writing cp = (x1,- • • , xm), for (o., /3) E cp(U), with
/3 = (/31, · · · , f3m), we have that
(cpo7rocp-1)(0.,/3) = (cpo7r)(cp-10.,/3i(dxi)rp-1a) = cp(cp-lo.) = o.,
and so cp o 11" o cp- 1 : cp(U) -+ !Rm is given by (o., /3) >-+ o., which is clearly
C 00 • Hence 7r : TM* -+ M is smooth.

(0 (V + JW))(p) = Op(V + f W)p = Op(Vp+ f(p)Wp)
= Op Vp+ f(p)OpWp = (OV)(p)+ f(p)(OW)(p)
= (OV + f(OW))(p).
As p E M was arbitrary, it follows that n (V + f W) = n V + f (n W).

For VE TJM and p EM, ((df)V)(p) = (df)pVp = Vpf = (VJ)(p).
As V E TJ M, the map p >-+ (VJ) (p) is smooth by Theorem 3.1.
Hence (df)V E C 00 (M), and by Theorem 3.3, we conclude that df E Tf M.

Let O=xdy E Tf!R 2 . Then we have nox=O, Ooy=X, and
ilx(Ooy)=oxx=l ~ O=oyO=oy(Oox)-
So n cannot be df for any f E C 00 ( M).

(J*(g-O))p = (dfp)*(g-0)1(p) = (dfp)*(g(J(p))01cp))
= g(J(p))·(dfp)*0 1cP) =(go J)(p)·(J*O)p
= ((gof)-(J*O))p.
So f*(g-0) = (gof)-(J*O).

For all p E M and all v E TpM, we have
(J*(dg))pv = ((dfp)*(dg)J(p))v = (dg)J(p)((dfp)v)
= ((dfp)v)(g) = v(g o J) = (d(g o J))pv.
Thus f*(dg) = d(g o J).

The ordinary differential equation describing the integral curve is
du 2
dt = u'
with the initial condition u(O) = p = 'Y(0) > 0, which gives
J ')'(t)
P
d ( 1)
- - du= -
du u
lt
0
dt.
1 1 p l 1
So - ()--=-t. Thus 'Y(t)=-- fort<-. (Ast/'-, 'Y(t)/+oo.)
'Y t p I-pt p p

Suppose that 'Yp : JR-+ JR 2 is the integral curve of V with 'Yp(O) = p.
Writing 'Yp(t) = (x(t), y(t)), t E JR, we obtain the initial value problem
x(t) = 1 + y(t), x(O) = Xo,
iJ(t) = 0, y(O) = Yo-
Integrating both sides of the second equation from 0 to t, y(t) - Yo = 0,
i.e., y(t) = y0 , t ER Inserting this in the first equation, and integrating
from 0 to t yields x(t) - xo = (1 + Yo)t, so that x(t) = xo + (1 + Yo)t. Thus
1/Jt(P) ='Yp(t) = (xo + (1 + Yo)t, Yo), t E JR. The flow of V consists of the flow
maps 1Pt : JR 2 -+ JR 2 given by 1/Jt(x, y) = (x + (1 + y)t, y) for all (x, y) E JR 2
and t E JR.
We have 1/J-t(x, y) = (x - (1 + y)t, y), t E JR. The matrix representation
of the linear transformation (d1/J-t)p : TpJR 2 -+ T,t,_,pJR 2 with respect to the
ordered basis (ox,p, oy,p) and (ox,J(p), oy,J(p)) is
[ 01 -tJ
1 .
Hence the action of (d1/J-t)p on oy,p is the multiplication of the above matrix
with the vector e2 E JR 2 . Consequently, (d1/J-t)p(oy,p) = -tox,,J,_,p + Oy,,J,_,p•
Solutions 385

For all f C 00 (JR 2 ), we have
E
[Lx,Ly]f
= (yoz - zoy)(zoxf - xozf) - (zox - XOz)(yozf - zoyf)
= y(ozz)(oxf) + yzozoxf - ~ - 0 - z 2 oyoxf - 0 + ZXOyOzf + 0
-zyoxozf - 0 + z 2 oxoyf + 0 + xyo;J + 0 - x(ozz)(oyf) - XZOzOyf
------ ~
= yoxf - xoyf = (yox - xoy)f = -Lzf.

Thus [Lx, Ly]= -Lz. Using the cycling sequence x-+ y-+ z-+ x in the
formula for Lx, we see that we obtain Lx-+ Ly-+ Lz-+ Lx, and so from
the above, [Ly,Lz] = -Lx and [Lz,Lx] = -Ly. Clearly [La,La] = 0 for
a= x, y, z, and [La, Lb]= -[Lb, La] for all a, b E {x, y, z}.
Let "/p : JR-+ JR3 be the integral curve of Lx such that "fp(O) =p= (xo, Yo, zo)-
Then with "/p(t) = (x(t), y(t), z(t)), t E JR, we obtain the initial value problem
x(t) = 0, x(O) = Xo,
y(t) = -z(t), y(O) = Yo,
i(t) = y(t), z(O) = zo.
Integrating both sides of the first equation from Oto t yields x(t) - x 0 =0,
that is, x(t) = x 0 for all t E R From the second and third differential
equations, we obtain jj = -i = -y, that is, jj + y = 0. This has the general
solution y(t) =A cost+ Bsin t for some constants A, B. Since y(O) =yo, we
have A= Yo- Also, z(t) = -y(t) = -(-Asint + B cost)= Asint - B cost.
We have zo = z(O) = -B. So "fp(t) = (xo, Yo cost - zo sin t, Yo sin t + zo cost),
t E JR. Thus the flow of Lx is given by {et: JR3 -+ JR 3 , t E JR}, where
et [: ] [ 1 c~s t - sin t ] [ : ] , for all [ : ] E JR3 .

z smt cost z z
We recognize this as a rotation about the x-axis through an angle t, in a
sense determined by the right-hand rule: If the thumb points in the positive
direction of the x-axis, then the other fingers show the direction of rotation
(i.e., looking at the yz-plane from the x > 0 half-space in JR3 , we will record
the rotation as being anticlockwise). By cyclically permuting the x, y, z, we
obtain the flows {1Jt : JR 3 -+ JR3, t E JR}, {(t : JR3 -+ JR3, t E JR}, of Ly, Lz,
respectively, where
c~st -sint l [x]

1Jt [ xl
y = [ cost l sintl [x]
y , (t [x]
y = [ smt cost y .
z -sint cost z z l z
Chapter 4

Suppose for some p E M, Vp -# 0. Let (U, x) be an admissible chart such
that p E U. Then Vp = viOxi,p, and as Vp -# 0, not all the vi are zero.
Let i* E {1, • • • , m} be such that vi•-# 0. Let x be a bump function such
that x = l in a neighbourhood of p, x = 0 outside Uo, and Uo c U.
Then f := xi•x E C 00 (M), and df E Tf M. We have with n := df that
Tvi1=!1V =df(V), and so
(Tvn)(p) = (df(V))(p) = (df)pVp = Vpf = Vp(xxi•)
= x(p)Vpxi• + xi•(p)Vpx = 1 Vpxi• + xi•(p)·O
= Vpxi• = ViOxi,pXi* = vi• -# 0.
Thus the function Tvn is not identically zero, a contradiction to Tv = 0.

For Vi, V2 E TJM, we have
T(Vi +V2) = (Vi +V2)(Wg) = Vi(Wg) + Vi(Wg) = T(Vi) + T(V2).
For VE TJ Mand f E C 00 (M), T(JV) = (JV)(W g) = f·(V(W g)) = f-T(V).
So Tis C 00 (M)-linear. Hence Tis a (0, 1)-tensor field.

For Vi, V2, WE TJM and f E C 00 (M),
(n ® 0)(V1 + JV2, W) = (!1(Vi + JV2)) (0W) = (!1(Vi)+ fi1(Vi))(0W)
= (!1(Vi))(0W) + fi1(Vi) (0W)
= (n ® 0)(V1, w) + J(n ® 0)(V2, w).
For V, W1, W2 E TJ M and f E C 00 (M),
(n ® 0)(V, W1 + JW2) = (nv) 0(W1 + JW2) = (!1V)(0W1 + f0W2)
= (!1V)0(W1) + f!1(V)0(W2)
= (n ® 0)(V, W1) + J(n ® 0)(V, W2).
So n ® 0 is a (0, 2)-tensor field.
We have i1A0=i1® 0 - 0 ® !1= -0 ® n - n ® 0= -0/\n.
For VETJM,
(n /\ 0)(V, V) = (n ® 0)(V, V) - (0 ® n)(V, V)
= (!1V)(0V) - (0V)(!1V) = 0.
Solutions 387

For V, WE TJM,
(J(0®0))(V, W) = f·((0®0)(V, W)) (!l f·(OV)-0W

= ((JO)V)-0W =((JO)® 0)(V, W)
(!l (OV) · ((J0)W) = (n ® (J0))(V, W),
and so
Also for V, WE TJ M,
((01 + 02) ®0)(V, W) = ((01 + 02)V)-0W = (01 V + 02V)-0W
= (n1 v)0w + (n2 v)0w
= (01 (8) 0)(V, W) + (02 (8) 0)(V, W)
= (01 ® 0 + 02 ® 0)(V, W), and
(0 ® (01 + 02))(V, W) = (0V)·((01 + 02)W) = (0V)-(01 W + 02W)
= (0V)01W + (0V)02W
= (0 ® 01)(V, W) + (0 ® 02)(V, W)
= (0 ® 01 + 0 ® 02)(V, W),
and so
(01 + 02) ®0 = 01 ®0 + 02 ®0 and
0 ® (n1 + n2) = 0 ® n1 + 0 ® n2.

For w, w' E (JR3 )*, v, v', w, w' E JR3 , and c E JR, we have
T(w + cw', v, w) = (w + cw')(v x w) = w(v x w) + cw'(v x w)
= T(w, v, w) + cT(w', v, w),
T(w, v + cv', w) = w((v + cv') x w) = w(vxw + cv' x w)
= w(vxw) + cw(v' xw)
= T(w, v, w) + cT(w, v', w), and
T(w, v, w +cw')= w(vx (w +cw'))= w(vxw + cvxw')
= w(vxw) + cw(vxw')
= T(w, v, w) + cT(w, v, w').

Let T,T',T" be (r,s)-, (r',s')-, (r",s")-tensors, respectively, on V. For all
w 1 , · · · , wr+r'+r" EV* and all V1, · · · , Vs+s'+s" EV, we have
(T (8) (T 1 (8)T"))(w 1 , · · · ,wr+r'+r",v1, · · · ,Vs+s'+s11 )

= T (W1 ,··· ,wr ,v1,··· ,Vs )(T I ,o,.
IC>IT ")(Wr+l ,··· ,wr+r' +r" ,Vs+1,··· ,Vs+s'+s" )
= T(w1, · · · ,wr,V1, ···,Vs)· (T'(wr+l, · · · ,wr+r',Vs+l, · · · ,Vs+s 1 )·
"( r+r' +1 r+r' +r" ))

T
W ,··· ,w ,Vs+s'+1,··· ,Vs+s'+s"
= (T(w 1, · · · , Wr, V1, · · · , Vs)· T'(wr+l, · · · , Wr+r', Vs+l, · · · , Vs+s' ))·
T "(Wr+r' +1 ,···,Wr+r' +r" ,Vs+s'+1,···,Vs+s'+s" )
= (T (8) T')(w1, · · · , wr+r', V1, · · · , Vs+s' )·
T "(Wr+r' +1,···,Wr+r' +r",Vs+s1 +1,···,Vs+s'+s" )
= ((T (8) T 1 ) (8) T 11 )(w 1, · · · , wr+r' +r", V1, · · · , Vs+s'+s" ).
Thus T (8) (T' ® T") = (T ® T') (8) T 11 •
First we show the spanning property. Let T be an (r, s)-tensor on V. For
any v E V, we have v = 1:i(v)ei, and for w E V*, w = w(ei)Ei. Thus for
w 1 , • • • , wr E V* and v1, · · · , Vs EV, we have
T(w 1 ,··· ,wr,V1,··· ,Vs)

1 . . . .
= T(w (ei1)1:' 1 ,··· ,wr(eiJ1:'r,1:11 (v1)eji,""" ,1:1•(vs)ej.)
. . 1 . .
= T(1:' 1 ,··· ,1:•r,eju•·· ,ej.)W (ei 1 )···Wr(eir)1:31 (v1)···€3"(vs)
= i
T( 1 , · · · , ir, ej1 , · · · , ej.) ·
(ei 1 (8) · · · (8) eir (8) €j 1 • • • (8) €j" )(w1, · · · , Wr, V1, · · · , Vs)-
So T = T(1:i 1, • • • , €ir, eJl,

· • • • , eJ· ) • e·i1
8
tO.
VY • • • tO.
VY e·1.r tO.
VY 1:i1 tO.
VY • • • tO.
VY €j 8 •
Next we show the independence. Let ct:t ei 1 (8) • • • (8) eir (8) 1:i1 • • • (8) 1:i• = 0.
Operating on (€k1, ... '€kr, eR,i, ... 'eR,.) E (V*t X vs'
Q = Ci_1 ·••i, e· (8) ... (8) e·tr @ €j1 ... @ €j" (1:k1 , . . .
Jl •••Js i1 ,
€kr , e•.c.1, . . . , e•.c. 8 )
= ct ::t ei 1 ( €k 1 ) · · · eir ( €kr )1:i 1 ( eR, 1 ) · · · €j" (et.)

= Ci_1 .. ,i, €k1 (ei ) ... €kr (ei )811 ... 81•
Jl •••Js 1 r .c,1 .C.s
= Ci1 •ir 8k1 ... 8kr

•r
00
£1 ... £. •1
= Ck1 •kr 00
£1 ... £•.
Solutions 389
A 1 ~ k1, · · · , kr, £1, · · · , fs ~ m were arbitrary, all the coefficients ::t ct

are zero, showing the linear independence of B. Finally, it is clear that the
number of elements in B is mr+s, since it is the number of ways of choosing
the r + s (not necessarily all distinct) numbers ii,··· , ir, j1, · · · , is from
the set {1,··· ,m}. Thus dimTJV=mr+s.

Suppose that p E U. Let x be a bump function such that x = 1 in a
neighbourhood U1 of p, x = 0 outside Uo, and Uo c U. Then for all q E U1,
T(q)t::t = T(q)((dxi 1 )q, · · ·, (dxir)q, xoxii,q, · · · , XDxi,,q)
= (T(xdxi 1 , · · · ,xdxir,xoxii,••· ,XOxi,))(q).
g . .
As q I--+ (T(xdx' 1 , · · · ,xdx'r,xoxii,••· ,XOxi,))(q) belongs to C 00 (M),
and since U1 is open, glu1 is smooth. Since smoothness is a local property,
U 3 p I--+ T(p)t:t belongs to C 00 (U).

We have T(p)} =T(p)((dxi)p, Oxi,p) = (T(xdxi, XOxi ))(p), where xis a bump
function, identically 1 in a neighbourhood of p, and O outside U0 , such that
Uo c U. Then
T(p); = (T(xdxi, XDxi ))(p) = (xdxi)p(XDxi )p = (dxi)p(XDxi )p
= x(p)Oxi,pXi = 18;.


We will first show the well-definedness. We have
=
,I/
,il ,•r J.Ji
,I/ .,
ui' · · · ui, u3." · · · u3.,,

.,
J.J B
1 r 1 B
So T is well-defined.
Since smoothness is a local property (Exercise 1.16, p.14), the map
M 3 p >-+ T(01,- • • , nr, V1, • • • , Vs) is smooth. Hence Tis a map from
(Tf Mt x (TJ M) 8 to C 00 (M). The multilinearity of Tin each component
follows from the observations that
k ~
+ JO~k )pOxik,p = (0 k )pOxik,p + f(p)-(O)pOxik,p,
(n
(dxie)p(½ + J½)p = (½ + J½)pxie = (½)pxie + J(p)·(Ve)pxie
= (dxie)p(Vi)p + J(p)·(dxie)p(Ve)p

Suppose that w 1, · · · , wr+r' E (TpM)*, and v1, · · · , Vs+s' E TpM.
Let 0 1 . . . nr+r' E T1° M be such that (0 1 ) = w 1 . . . (nr+r') = wr+r'
' ' p ' ' p '
and Vi, · · · , Vs+s' E TJ M be such that (Vi )p = V1, · · · , (Vs+s' )p = Vs+s'.
Solutions 391
Then
(T ® T')(p)(w1, · · · , wr+r', V1, · · · , Vs+s')

= ((T ® T')(01,- · · , nr+r', V1, · · · , Vs+s' ))(p)
= (T(01,--- ,nr,V1,··· ,V,,)-T'(nr+1, ... ,nr+r',Vs+1,··· ,Vs+s ))(p) 1
= (T(01,--- ,nr,V1,··· ,V,,))(p)·

(T'(nr+1, · · · , nr+r', ½+1, · · · , Vs+s' ))(p)
= T(p)((0 1)p, · · ·, (Or)p, (V1)p, · · ·, (Vs)p)·
T'(p)((nr+i )p, · · ·, (nr+r')p, (V,,+1)p, · · ·, (Vs+s' )p)
= T(p)(w1, · · · , wr, V1, · · · , V8 ) • T'(p)(wr+1, · · · , wr+r', Vs+l, · · · , Vs+s')
= (T(p) ® T'(p))(w1,- · · , wr+r', V1, · · · , Vs+s' ).
Thus (T®T')(p)=T(p)®T'(p).

We have (dx ® dy)(ox, oy) = dx(ox) ·dy(oy) = (oxx)(oyy) = 1-1 = 1, while
(dy ® dx)(ox, oy) =dy(ox) ·dx(oy) = (oxy)(oyx) = 0-0 = 0.

For p EU,
(T ® S)~: :::;.:~, (p) = (T ® S)(p)((dxi' )p, ... '(dxir+r' )p, Oxi,,p, ... 'Oxi•+•',p)
= ((T ® S)(xdxi', · · · , xdxir+r', XDxi,, · · · , xoxi•+•' ))(p),
where x E C 00 (M) is identically 1 in a neighbourhood of p, and identically

0 outside Uo with Uo c U. Thus
(T ® S)t:;_:~, (p) = ((T ® S)(xdxi', · · · , xdxir+r', XDxi,, · · · , xoxi•+•' ))(p)

= (T(xdxi', · · · , xdxir, XDxi,, · · · , XDxi• ))(p) ·
(S(xdxir+i, ... ,xdxir+r',xoxi.+1,··· ,XOxi•+•'))(p)
= T(p)((dxi' )p, · · · , (dxir )p, Dxi,,p, · · · , Dxi,,p) ·
S(p)((dx ir+l) p, ... , (dxir+r') p, UxJs+1,p,
::i . • • • ::i .
, U x's+s',p
)

For p EU, we have
(C~T)jk(P) = (C~T)(p)(Oxi,p, Oxk,p) = ((C~T)(xoxi' XOxk ))(p),
where x is a bump function, identically 1 in a neighbourhood of p, and
identically 0 outside Uo with Uo c U. But
((C~T)(XOxi,XOxk))(p) = C(T(•,XOxi, ',XOxk))(p)
= T(xdx\xoxi,XOxi,XOxk)(p)
= T(p)((dxi)p, Oxi,p, Oxi,p, Oxk,p)
= Tfik(p).

CT= tracer, because for all p E Rm, we have
(CT)(p) = T(p)((dxi)p, Oxi,p) = (T(dxi, ox;))(p)
i kj i kj i
= (dx )T(p)(Tjt5iOxk,T(p))=t5krjt5i = Ti
= tracer.

In general, there are 18 different contraction operations possible: We have
• CL c~, C½, Ci, c~, c~, which give a (1, 2)-tensor field, and
• cfcL cfct cfc½, cfc~, cfc~, cfc~, c~cL c~ct c~c½, c~ci,
C~C~, C~C~, which give a (0, 1)-tensor field.
But some of these contraction operations in the second row produce the
same (0, 1)-tensor field. To see which ones are identical, we compute com-
ponents: If r1tare the components of Tin an admissible chart (U, x), then
the contraction operations in the second row above result in tensor fields
having the following chart-induced components, respectively:
Tii Tii Tii Tii Tii Tii
ijc' J.!::,' jbi' J.!::,' ibj' ibj'
Tii Tii Tii Tii Tii Tii
ibj' aij' aji' jbi' aji' aij"
So altogether, we have 6 + 6 = 12 distinct tensor fields produced by the

contraction operations.
Solutions 393

First we show that C:TfM-+C 00 (M) is C 00 (M)-linear. Let (U,x) be an
admissible chart. Let p EU, and let x E C 00 (M) be a bump function which
is identically 1 in a neighbourhood of p and identically O outside U0 with
Uo c U. We have
(C(T+ f S))(p) = ((T+ f S)(p))((dxi)p, Oxi,p) = ((T+ f S)(xdxi,xox;))(p)
= (T(xdx\xox;) + (JS)(xdx\xox;))(p)
= T(xdx\xox;)(p) + f(p)S(xdx\xox;)(p)
= T(p)((dxi)p, Dxi,p) + J(p)S(p)((dxi)p, Dxi,p)
= (CT)(p)+f(p)(CS)(p) = (CT)(p)+(J-(CS))(p)
= ((CT)+f·(CS))(p).
Thus C(T+f S) = CT+f-CS.
For fixed n1, · · · , nr-l Erp M and Vi,··· , Vs-1 E TJ-M, we have
(c;(T+JS))(01,--- ,nr- 1,Vi,··· ,Vs-1)
= C((T+JS)(01,··· ,ni-l, ,n\-·· ,nr- 1,Vi,··· ,½-1, ,½,··· ,Vs-d)
= C(T(01,--- ,ni-l, ,n\-·· ,nr- 1,Vi,··· ,½-1, ,½,··· ,Vs-d
+f·S(01,--- ,ni-i, ,O\··· ,nr- 1,Vi,··· ,½-1, ,½,··· ,Vs-1))
= C(T(01,--- ,ni-l, ,n\-·· ,nr- 1,Vi,··· ,½-1, ,½,··· ,Vs-d)
+f·C(S(01,··· ,ni-l, ,n\-·· ,nr- 1,Vi,··· ,½-1, ,½,··· ,Vs-d)
= (Ci-T)(Ol ... nr-1 V,l ... V. 1)+J·(Ci.S)(n1 ... nr-1 V,l ... V. 1)
J ' ' ' ' ' s- J ' ' ' ' ' s-
= ((C}T) + J·(C}S))(01,- · · , nr- 1, Vi,···, Vs-d-
Consequently, c;
(T + f S) = ( T) + f ·( S). c; c;
Consider the chart (M, (r, 0) >-+ (r, 0)). Then for q= (r, 0) EM,
((J*T)(or, or ))(q) = r 2 (c~s 0) 2 ((or,qr)(or,qr)(or,qr sin0)(or,qr sin 0)
+ (Or,qr) (Or,q0) (Or,qr sin 0) (00,qr sin 0)
+ or,q(0)or,q(0)00,q(r sin0)00,q(r sin 0))

1 . 2 1 2
( ) (sm0) = 2 (tan0) .
r 2 cos 0 2 r
Similarly,
((f*T) (or, 00)) (q) = r 2 ( c~s 0) 2 ( O+ (or,qr) (00,q0)( or,qr sin 0) (00,qr sin 0) +O+O)
(
1 ) (sin0)rcos0 = !tan0,
r 2 cos 0 2 r
((f*T) (00, or)) (q) = r 2 ( c~s 0) 2 ( O+O+ (00,q0) (or,qr) (00,qr sin 0) (or,qr sin 0) +O)
(
1 ) (rcos0) sin0 =!tan 0,
r 2 cos 0 2 r
((f*T)( 00, 00)) (q) = r 2 ( c~s 0) 2 ( O+O+ O+ (00,q0) (00,q0) (00,qr sin 0)( 00,qr sin 0))
1
( )2 (rcos0)(rcos0) = 1.
r 2 cos 0
Thus
1 1
f*T = 2 (tan 0)2dr ® dr + -(tan 0)(dr ® d0 + d0 ® dr) + d0 ® d0
r r
in the chart (M, (r, 0) >-+ (r, 0)).
Solutions 395
Chapter 5

By the linearity of gin the first slot, and by the symmetry of g, g is linear
in the second slot too. Thus for each v E V, vb : V --+ JR is a linear map,
that is, vb EV*.
The map .b: V--+ V* is linear: For all v, wEV, cEIR,
(v + cw)b = g(v +cw,•)= g(v, •) + cg(w, •)=vb+ cwb.
Moreover, the map .b : V--+ V* is injective. Indeed, if v E V is such that
vb= 0 E V*, then for all w E V, vbw = 0, that is g(v, w) = 0, and as g is
nondegenerate, v = 0.
Finally, we show that .b: V--+ V*is an isomorphism. Since V, V* have the
same dimension, it follows that the injective map v i--+ vb is also surjective,
and hence it is an isomorphism.

We have by definition that u1- = {w EV : g(w, u) = 0 for all u EU}, and
(U1-)1- = {v EV: g(v,w) = 0 for all w E u1-}. Clearly ifwe take au EU,
then for all w E u1-, g(w,u) = 0, and sou E (U1-)1-. Thus Uc (U1-)1-. But
if dimU = k, then dimU1_ = m - k, and dim(U1_)1_ = m - (m - k) = k.
Now dim U = dim(U 1- )1_ and U c (U 1- )1_ imply that U = (U 1- )1_.

Let {eo, e1, · · · , em-d be an orthonormal basis for V with g(eo, eo) = -1,
and g(ei,ei)=l for l~i~m -1. Decompose
m-1 m-1
u = u 0 eo + "
Li. .
u'ei, v = v 0 eo + "
Li. .
v'ei.
i=l i=l
As u, v are lightlike, we have

0=g(u,u)=-(u0 )2+11ull 2 , and 0=g(v,v)=-(v0 ) 2 +llvll 2 ,
where u = (u1 · · · um-l) E !Rm-l v = (v 1 · · · vm-l) E !Rm-l Since u v
' ' ' ' ' . '
are orthogonal, -u0 v 0 + (u, v) = 0. So l(u, v)I = lu 0 v0 1= lu 0 11v0 1= llull llvll,
that is, the Cauchy-Schwarz inequality for the Euclidean inner product on
!Rm-l happens to be an equality. Thus u, v are linearly dependent. (Note
that as u, v are lightlike, we have u -# 0 and v -# 0. So u-# 0, since otherwise
(u 0 )2 = llull 2 =0, and then u=0, a contradiction. Similarly v-#0.)
Because u, v are linearly dependent, and each is nonzero, there exists a

tE IR\{0} such that u=tv. We have
uovo
u 0 = - - (note that (v 0 )2 = llvll 2 > 0)
v0
(u, v) t(v, v) t(v 0 ) 2 0
= -v0- = - v0
- - = -v0- =tv .
m-1 m-1
Thus u=u0 eo + I; uiei=tv 0 eo + I; tviei=tv.
i=l i=l
Conversely, if u=tv with t E IR\{0}, then
g(u, v) = g(tv, v) = tg(v, v) = t-O = 0,
where we have used g(v,v)=O (since vis lightlike).

If v ET, then g(v, v) <0 and so v~v. So~ is reflexive.
If v, w ET, and v ~ w, then g(w, v) = g(v, w) < 0, and sow~ v. Thus ~ is
symmetric.
Finally, suppose that v, w, x ET and v ~ w, w ~ x. Then g(v, w) < 0 and
g(w,x)<O. Sowithk:=;~~::~ (==) >O,wehavekg(x,w)=g(v,w),that
is, g(v-kx, w) = 0. As w is timelike, and v-kx is orthogonal to w, by
Lemma 5.1, either v-kx=O or v-kx is spacelike.
1° v-kx=O, i.e., v=kx. Then g(v,x)= g(vkkx) = g(~v) (= ~) <0. So v~x.
2° v-kx is spacelike, i.e., O<g(v-kx,v-kx)=g(v,v)-2kg(v,x)+k 2 g(x,x).
Hence we have g(v,x)< lkg(v,v)+!g(x,x)<O+O=O, and so v~x.
So ~ is transitive.
Consequently, ~ is an equivalence relation on T.

By the reversed Cauchy-Schwarz inequality,
(g(u,v)) 2 ~ g(u,u)g(v,v) = (-g(u,u))(-g(v,v)) = (r(u)) 2 (r(v))2.
So lg(u,v)l~r(u)r(v). As g(u,v)<O, this gives -g(u,v)~r(u)r(v). Thus
-g(u + v, u + v) = -g(u, u) - g(v, v) - 2g(u, v)
~ (r(u)) 2 + (r(v)) 2 + 2r(u)r(v) = (r(u) + r(v)) 2 >0.
Thus g(u + v, u + v) < 0, that is, u +vis timelike, and moreover, we obtain
(r(u + v)) 2 ~ (r(u) + r(v)) 2 , giving r(u + v) ~ r(u) + r(v) (as r(u + v),
r(u), r(v) are all positive).
Solutions 397
(1)=?(2): Let A E O(1,m -1). For all u,v E !Rm, r,(Au,Av) = r,(u,v), that
is, (Aur[r,]Av = ut[r,]v, that is, utAt[r,]Av = ut[r,]v. For 1 ~ k ~ m, let
ek E !Rm be the column vector with all entries equal to zero, except for the
kth entry, which is equal to one. Taking succesively v = ek, 1 ~ k ~ m, we
obtain utAt[r,]A = ut[r,]. Taking successively u = ek, l~k~m, we obtain
At[r,]A = [r,].
(2)=?(3): The (i,j)th entry of At[r,]A is (At)iT/ktA] = AfT/ktAf Thus
T/ij = Af A]T/k£·
(3)=?(1): A E 0(1, m - 1), because for all u, v E !Rm that
r,(Au,Av) = (Au)t[r,]Av = utAt[r,]Av = uiAfT/ktA3vi
. k £ . . .
= u"AiAjT/ktV1 = u"T/ijV1 = r,(u,v).
If A E 0(1, m - 1), then At[r,]A = [r,]. Thus taking determinants,
-1 = det[r,] = det(At[r,]A)
= (det(A t))(det[r,]) det A
= (detA)(-l)detA = -(detA) 2.
So detA E {-1, 1}. As detA#0, A is invertible. So 0(1, m-1) c GL(m, JR).

A:= Im 0(1, m - 1) because
E
r,(Au,Av) = r,(Iu,Iv) = r,(u,v) for all u,v E !Rm.

If A1,A2 E O(1,m -1), then A1A2 E O(1,m -1) since for all u,v E !Rm,
r,((A1A2)u, (A1A2)v) = r,(A1(A2u),A1(A2v)) = r,(A2u,A2v) = r,(u,v).
If A E O(1,m-1) c GL(m,IR), then A- 1 E O(1,m-1): For u,v E !Rm, with
u' := A- 1 u, v' := A- 1 v,
r,(A- 1u,A- 1v) = r,(u',v')
= r,(Au', Av') (as A E 0(1, m - 1))
= r,(AA- 1u,AA- 1v) = r,(Iu,Iv)
= r,(u,v).
Thus 0(1, m - 1) is a subgroup of GL(m, JR).

For each of the given matrices A, we will check that At[17]A [17], to
conclude that A E 0(1, 3).
We have, using (cosh¢) 2 - (sinh¢) 2 = 1, that
We have using (cos 0) 2 + (sin 0) 2 = 1 that

l
r cos0 sin0
- sin0 cos0
cos0 -sin0
sin0 cos0
We have
Finally,

For all p EM, we have (g(V, W))(p) = (g(V, W))(p), that is,
g(p)(Vp, Wv) = g(p)(Vp, Wv)-
Let w E TpM. Suppose that the vector field WE TJ Mis such that Wp = w.
(Such a vector field W exists, by Lemma 3.1, p.44.) Then the equality
g(p)(Vp, Wp) = g(p)(Vp, Wp) above gives g(p)(Vp,w) = g(p)(Vp,w), i.e.,
g(p)(½, - vv,w) = o.
As the choice of w E TpM was arbitrary, g(p)(Vp - Vp, w) = 0 for all
w E TpM. Since g(p) is nondegenerate, it follows that Vp - Vp = 0, that is,
Vp = Vp. Asp EM was arbitrary, we conclude that V = V.
Solutions 399

The identity map idM is clearly an isometry. Let f, h E Iso(M). We have
f o h- 1 E Iso(M), because for V, WE TJ-M and p EM,
(((! o h- 1 )*g)(V, W))(p)
= g((J O h- 1 )p)((d(J O h- 1 ))pVp, (d(J O h- 1 ))pWp)
= g(J(h- 1 p))((df)h-1p((dh-l )p ½,), (df)h-lp((dh- 1 )pWp))
= g(h- 1 p)((dh- 1 )pVp, (dh- 1 )pWp) (since f*g = g)
= g(hh- 1 p)((dh)h-ip(dh- 1 )pVp, (dh)h-1p(dh- 1 )pWp) (as h*g = g)
= g(p)((d(h O h- 1 ))pVp, (d(h O h- 1 ))pWp)
= g(p)((d(idM ))p Vp, (d(idM ))pWp)
= g(p)(idTpM½,,idTpMWp) = g(p)(Vp, Wp) = (g(V, W))(p).

Using the properties of the tensor product®, and Exercise 2.15,
dx ® dx = (cos¢ ) 2 (sin 0) 2 dr ® dr + r 2 (cos¢ ) 2 ( cos 0) 2 d0 ® d0
+r 2 (sin ¢) 2 (sin 0) 2 d¢ ® de/>
+r(cos ¢) 2 (sin 0)(cos 0)(dr ® d0 + d0 ® dr)
-r(cos ¢)(sin ¢)(sin 0) 2 (dr ®de/>+ de/>® dr)
-r2 (cos ¢)(sin ¢)(cos 0)(sin 0)(d0 ®de/>+ de/>® d0)
dy ® dy = (sin ¢) 2 (sin 0) 2 dr ® dr + r 2 (sin¢) 2 (cos 0)2d0 ® d0
+r 2 (cos ¢) 2 (sin0) 2 d¢ ® de/>
+r(sin ¢) 2 (sin 0)(cos 0)(dr ® d0 + d0 ® dr)
+r(sin ¢)(cos ¢)(sin 0) 2 (dr ®de/>+ de/>® dr)
+r 2 (cos ¢)(sin ¢)(cos 0)(sin 0)(d0 ®de/>+ de/>® d0)
dz® dz = (cos 0) 2 dr ® dr + r 2 (sin 0)2d0 ® d0
-r(cos0)(sin0)(dr ® d0 + d0 ® dr).
Adding these, we obtain
dx ® dx + dy ®dy + dz ®dz= dr ®dr + r 2 d0 ®d0 + r 2 (sin0) 2 d¢@d¢.
To show that 1Pt is an isometry, we must show that ('1/Jt)*g = g. For any
p= (x,y,z) E l~.3, and any vector field V = vxax + VYoy + vzaz E TJ-1~.3,
using the result from Theorem 2.2, we have
(d'I/Jt)pVp = (d'I/Jt)p(Vx(p)ox,p + VY(p)oy,p + vz(p)oz,p)
= vx(p)ox,,f.1tp + ((cost)VY(p) - (sint)Vz(p))oy,,f.,,p
+((sint)VY(p) + (cost)Vz(p))oz,,f.,,p·
Hence, for V, WE TJ!R 3 , we have
((('1/Jt)*g)(V, W))(p)
= g('l/JtP)((d'l/Jt)pVp, (d'l/Jt)pWp)
= ((dx),;,.p ® (dx),;,.p + (dy),;,.p ® (dy),;,.p
+(dz),;,.p ® (dz),;,.p)((d'l/Jt)pVp, (d'l/Jt)pWp)
= vx(p)Wx(p)
+ ((cos t)VY(p) - (sin t)Vz(p))((cos t)WY(p) - (sin t)Wz(p))
+ ((sin t)VY(p) + (cos t)Vz(p))((sin t)WY(p) + (cos t)Wz(p))
= vx(p)Wx(p) + ((cost)2 + (sint)2)VY(p)WY(p)
+((-sint) 2 + (cost)2)Vz(p)Wz(p)
= g(p)(½,, Wp) = (g(V, W))(p).
Consequently, (1Pt) *g = g, i.e., 1Pt is an isometry.

With p= (uo, vo, uovo), we have that
g(p)uu = (ou,pX) 2 + (ou,pY) 2 + (ou,pz) 2

= (!:(p)r + (!:(p)r + (!~uv)(p)r
= 12 + 02 + v5 = 1 + v5,
g(p )vv = (Ov,pX ) 2 + (Ov,pY ) 2 + (Ov,pZ ) 2
= (!: r + (!: r +
(p) (p) (!~uv) (p) r
= 02 + 12 + u5 = 1 + u5,
g(p)uv = g(p)vu = (ou,px)(ov,pX) + (ou,pY)(ov,pY) + (ou,pz)(ov,pZ)
= ( :: (p)) ( :: (p)) + ( :: (p)) ( :: (p)) + ( :~UV\p)) (:~UV \p))

= 1-0 + 0-1 + vo·uo = uovo.
Thus
[ g(p)uu g(p)uv] = [ 1 +v5 UoVo2 ] .

g(p)vu g(p)vv UoVo l+uo
Solutions 401

Tensorial property of g: Let f E C 00 (M1 X M2), X, X', X" E TJ-(M1 X M2)
and (p1,P2) E M1 x M2. Below, we will write X(p 1,p2) = Vp 1 EB Wp 2 , where
½,1 E Tp1M1, Wp 2 E Tp 2M2, etc. We have
(g(X + f X', X"))(p1,p2)
= g(p1,P2)((x + JX')<v1,v2),x<~1,v2))
= g(p1,P2)(X(p1,P2) + f(p1,P2)X(p 1,p2),X(~1,P2))
= g(p1,P2)((½,1 EB Wv2) + f(P1,P2)W;1 EB w;2), v;: EB w;2)
= g(p1,P2)((½,1 + f(p1,P2)v;J EB (Wp2 + f(p1,P2)w;J, v;: EB w;J
= g1(P1)(½,1 + f(P1,P2)v;l, v;:) + g2(P2)(Wp2 + f(P1,P2)w;2, w;J
= g1(P1)(½,i, v;J + g2(P2)(Wp2, w;J
+ f(P1,P2) (g1(P1)W;1, v;:) + g2(P2)(w;2, w;2))
= g(p1,P2)(½,1 EB Wp2, v;: EB w;J
+ f(P1,P2)g(p1,P2)W; 1 EB w; 2, v;: EB w;2)
= g(p1,P2)(X(p1,P2),X(~ 1,p2)) + f(p1,P2)g(p1,P2)(X(p1,P2)'X(~1,P2))
= g(X,X")(p1,P2) + f(p1,P2)g(X',X")(p1,p2)
= (g(X,X") + f ·g(X',X"))(p1,p2).
The C 00 (M1 x M 2 )-linearity in the second slot of g is shown similarly.
So g E TJ(M1 x M2).
g(p1,P2) is a scalar product on T(p1,p2)(M1 x M2), (p1,P2) E M1 x M2:
Bilinearity is a consequence of the fact that g is a (0, 2)-tensor field, estab-
lished above. We show symmetry:
g(p1,P2)(v1 EB w1, v2 EB w2) = g1 (p1)(v1, v2) + g2(P2)(w1, w2)
= g1 (p1)(v2, v1) + g2(P2)(w2, w1)
= g(p1, P2)(v2 EB w2, v1 EB w1).
Nondegeneracy: Suppose that for all v2 EB w2 E T(p1,p2)(M1 x M2),
g(p1,P2)(v1 EB w1, v2 EB w2) = 0.
Then in particular, for all v2 E Tp1M1, v2 EB O E T(Pi ,p2) (M1 x M2), and so
0 = g(p1,P2)(v1 EBw1,v2 EBO)
= g1(P1)(v1,v2) +g2(P2)(w1,0)
= g1(P1)(v1,v2) +O = g1(P1)(v1,v2).
Thus by the nondegeneracy of g1 (p1), v1 = 0. For all w2 E Tp2M2, we have

0 EB w2 E T(p 1 ,p2) (M1 x M2), and so
0 = g(p1,P2)(v1 EBw1,0EBw2)
= g1(P1)(v1, 0) + g2(P2)(w1, w2)
= g2(P2)(w1, w2).
So we conclude that w1 = 0, as g2(P2) is nondegenerate. Consequently, we
have v1 EB w1 = 0 EBO E T(p 1 ,p2) (M1 x M2).
Finally, we show that 1,(g) = 1. Let (p1,p2) E M1xM2. Let {e1, · · · , emi}
be an orthonormal basis for Tp 1 M1 such that g1(P1)(e1,e1) = -1 and
g1(P1)(e2, e2) = · · · =g1(P1)(emu emi) = 1. Similarly, let {/1, · · · , fmJ be an
orthonormalbasisforTp2M2. Then{e1EBO,··· ,em 1 EBO,OEB/i,··· ,OEB/mJ
spans T(Pi,P 2)(M1 x M2), and
g(p1,P2)(e1 EB 0, e1 EB 0) = g1 (p1)(e1, e1) +0 = -1,
g(p1,P2)(e2EBO,e2EBO) = ··· = g(p1,P2)(em 1 EBO,em 1 EBO)= 1,
g(p1,P2)(0EB/i,OEB/i) = ... = g(p1,P2)(0EBfm2,0EBfm2) = 1,
g(p1, P2)( ei EB 0, ej EB 0) = g1 (p1)(ei, ej) = 0 for i =I- j,
g(p1,P2)(ei EB 0, 0 EB /i) = 0 + 0 = 0 for all i,j,
g(p1,P2)(0 EB /i, 0 EB /i) = g2(P2)(/;, /j) = 0 for i =I- j.
By Theorem 5.5 (p.80), ig(p 1 ,p2) = 1. But as (p1, P2) was arbitrary, 1,(g) = 1.

Let p E U. Choose and orthonormal basis B = {e 1, • • • , em} for TpM with
respect to the scalar product g(p): g(p)(ei,ej)=Ei8ij, where
-1 if i = 1,
{
€i:= 1 if iE{2,···,m}.
Let af be such that Oxi,p = afek, 1 ~ i, k ~ m. Then A= [a{] is invertible.
We have
gij(P) = g(p)(ox;,p, Oxi,p) = g(p)(af ek, a;et)
= af a;Ek8kt = af Ek8kta; = At EA,
where E is the diagonal matrix with the diagonal entries € 1, • • • , Em- Con-
sequently, G(p) =At EA, and
det G(p) = (det(At))(-l)(l)m- 1(det A)= (det A) 2(-1) < 0.
(As A is invertible, detA=l-0.)
Solutions 403

We have gij = g(oxi,Oxi) E C 00 (U). The matrix G(p) is symmetric for all
pE u because gij = g(Oxi,Oxi) = g(Oxi,Oxi) = gji·
Let v E !Rm be such that G(p)v = 0. Let wioxi,p E TpM. Then with
wt := [ w 1 ... Wm], wtG(p)v = wto = 0, i.e., viwigji(p) = 0, and so
g(p)(wioxi,p,viox;,p)=O for all wioxi,p E TpM. As g(p) is nondegenerate,
viOxi,p=O, i.e., v=O. Thus G(p) is invertible, and detG(p)#0, p EU.
As G has C 00 (U) entries, and since C 00 (U) forms an algebra with point-
wise operations, we conclude that det G E C 00 (U). Also for all p E U,
det G(p) # 0. Thus de!a E C 00 (U). Let adj G(p) be the matrix with entry
( -1) i+i det G(p) ij in the i th row and lh column, where G(p) ij is the matrix
obtained from G(p) by deleting its ith column and jth row. Then
1
(G(p))-1 = detG(p)adjG(p).
But p >--+ [adj G(p)]ij belongs to C 00 (U) as well. Hence p >--+ [(G(p))- 1]ij
belongs to C 00 (U).

Recall (Exercise 2.16) that if
A= [AJ] = [!~'i ox-l) (x(•))], and B =[BJ]= [!~:i ox'-l) (x'(·))],

then AB=BA=Im. Since g is a (0, 2)-tensor field, by Exercise 4.10 (p.68),
I. = oxk OXR. = oxk OXR. = (Bt)i BR.
g,J OX1i OX 1 j gkf. OX 1i gkf. ox'j kgkf. J"
Thus if G = [gii] and G' = [g~i], then G' = BtGB. This gives
(G')-1 = B-1a-1(Bt)-1 = B-1a-1(B-1)t = Aa-1 At.
a ,i a ,j
Hence g'ij = [(G')-l]ij =Ai (a-lt\Atl = ...!!._gkR._::__
k J oxk OXR.
We have V-y,-y(t) = -doxi,-y(t) + ! Ot,-y(t)· Since for all t E JR,

d
g('y(t))(v-y,-y(t), Ot,-y(t)) = - ; < 0, and
g('y(t))(v-y,-y(t), v-y,-y(t)) = - ~: + d2 = -d2 ( :2 - 1) < 0,

it follows that 'Y is an observer.
We have 1'(0) =p +de1 =P, 1'(l)=p +!eo=Q, and
-r7 (P, Q) = f ✓-g(1'(t))( v7 ,7 (t), v7 ,7 (t)) dt
= f✓ ~ d2 ( - 1) dt = ✓d2( : 2 - 1)
< vd 2(1+ ~:-1) = T.
(With T~660m, d~ 10x103 m, we have that~"' 0.99783.

T2
1+-;p
So if u ~ 0.998, then I> u > ~ - Thus r,(P, Q) <T.

T2
1+-;p
In fact, -r7 (P, Q) = ✓d2(:2 -1) ~633.4m=2.11135x10- 6 sec<2.2x10- 6 sec.

So the muon survives the atmospheric transit.)
We have v'Y,'Y(s) = -t'(s) Ot,'Y(s) + <f/(s) oc/>,'Y(s)· As 1' is an observer, v'Y,'Y(s)

is future-pointing, so that t'(s) > 0 for all s E R (Here we use the time-
orientation on (M, g) given in Example 5.14.) We have
g(1'(s))(v'Y,'Y(s),v'Y,'Y(s)) = -(1- 2r7)(t'(s)) 2 +r5( sini)\<t>'(s)) 2
= -(1- 2m)(t'(s)) 2 +r5~(t'(s)) 2

ro ro
= -(1- 3r7)(t'(s))2.
Hence the proper time between p, q experienced by 1' is
In the above, the map t : M -> JR is the t-coordinate map, that is, the map
M = !Rx (2m,oo) xS 2 3 (t,r,p) >-+ t.
Solutions 405
Next, we determine the proper time experienced by 7 between the events

p, q on its worldline. We have v"'i ."i( s) = Ot,-"/( s). We note that
:Y(t(p)) = (t(p), ro, cp- 1((i, ¢0))) = P,
:Y(t(q)) = (t(q), ro, cp- 1((i, ¢0))) = q.
Thus
t(q) lt(q) ✓ 2m
T.,.,,onaxy = l ,v-g(7(s))(v7,-"/(s),v"'i,-"/(s)) ds = (1--)(1) 2 ds
t(p) t(p) ro
= ✓ (1- ~7) (t(q) - t(p)).

Consequently,
Tmbit = ✓ (1- 3r7) (t(q) - t(p)) < ✓ (1- 2r7) (t(q) - t(p)) = T.tationaxy•

We have, using the geodesic equation involving t" that
(( 1- 2m)t')' = 2mr't' + (1- 2m)t" = 2mt'r' - 2mt'r' = 0.

r r2 r r2 r2
2
So (1- ;?")t' is constant, say E. As the time-orientation is given by Ot, and
since v7 ,,7 .(r) is future-pointing, we have t' > 0 everywhere. It follows that
E>0. Moreover, multiplying the equation
2m)- 1( r ') 2 = -1
2m) (t ') 2 + ( 1--;=-
- ( 1--;=-
by 1 - 2;?1, and using (1- 2;?1 )t' = E then gives
-E2 + (r')2 = -(1 - 2~)-

So (r') 2 = E 2 - 1 + 2;?1. If r(r1) = r 1 for r 1 EI, then the r-maximisation
assumption gives r'(r1) = 0, so that 02 = E 2 - 1 + 2r';. Thus
E=vl- 2m_
r1
Hence
( dr) = E 2 _ 1 + 2m = 2m _ 2m.
2
dr r r r1
So, while r increases from r = ro tor= r1, we have
dr _. f2m 2m
dT - v-;::- - -;:;,
and on the return trip from r = r 1 tor= r 0 , we have a minus sign in front
of the square root. Consequently the total proper time to make the round
trip is
l:!..r = 2 fr' --;:=l==dr.

lo V. /2mr _ 2m
r1
Let To, T2 E J be such that r(ro) =ro =r(r2). Ifs denotes the inverse of the
(increasing) function r: (ro,r1)-+ (ro , r1) , then
. /i 2m
d(t o s) I = t'(r)-1- = V - -;:-;- 1
du r=r(T) r'(r) 1- 2m . /2m _ 2m
r V r r1
So the coordinate time lapse to go from r = r 0 to r = r 1 is given by
A similar calculation using the (decreasing) function r : (r1 , r 2 ) -+ (r 0 , r 1 )

shows that the coordinate time lapse to go from r = r 1 to r = r 0 is equal
to (l:!..t)i. Hence the coordinate time lapse for the entire round trip is
l:!..t = 2(1::!..t)i, as wanted.
Finally to show the inequality, we first note that
(17.1)
Also, we compute
d(Vl - ~l:!..t)
dr 0
Solutions 407
Using (17.1) and the above expressions, we obtain (after some algebra)
Thus, the function
ro >-+ P , f 1 t - f1r,
1s mcreasing. But the value when r 0 equals r 1 is 0, and so the value at

r 0 < r 1 must be negative, which gives the desired inequality.
(ta, t a)= B
ea .L A=,( to)
tB - l. sinh(toa)
The slope of the segment AB is -1, giving
tB -
r (t 0a) = -1, that is,
a cosh
1 etoa
tB = -(cosh(toa) + sinh(toa)) = - .
2a 2a
So the time f1t it takes for the photon ,\_ to reach the horizon in the
(U, x)-coordinates is
2a
Hence
E= Ii = Ii = lia.
2f1t 2 e- toa e-toa
2a
We have 17 V>.+,>-+(to) = eo + e1 and v-y,-y(t) = cosh(ato)eo + sinh(ato)e1. So

E _ 'Tl(E(eo + e1),cosh(ato)eo + sinh(at0 )e1)
observed - - y'-,,,(cosh(at 0 )e0 + sinh(ato)e1, cosh(ato)eo + sinh(ato)e1)
-(- cosh(ato) + sinh(ato)) e!-•~a
= ----;::::::;=:::;::===;::===;::::::;;:::===========::::;:;;:;:: = na
y'-(-(cosh(at 0 )) 2 + (sinh(at0 )) 2 ) ·

We have g(gradf, V) = (gradfY'V = ((df)U)i'V = df(V) = V f.
For the given global chart on Rm, gij = 8ij, and so gii = 8ii. We have
(df)i = df(oxi) = Dxif. Thus (gradj)i = 8ii Dxi f = Dxif.

As (vi? Ev•, and {w1,··· ,wm} forms a basis for v•, there exist Cij ER
such that (Vi )i' = Cijwi. Acting on Vk yields
~ . .
g(vi,vk) = (vi) Vk = CijW1 Vk = Cij8l = Cik·
So (vd = Cijwi = g(vi,Vj)wi.

Formula for wU when {v1, • • • , vm} is orthonormal: Writing w = Ciwi, and
operating on Vk, we get Ck = ci8i = ciwi(vk) = WVk- So (dropping the
summation convention)
m m m
w = ~ CiWi = ~(wvi)wi = ~(wvi)(i(vi)·(vd),
i=l i=l i=l
m m
giving wU = I; (wvi)g(vi,vi)((vd)U = I; (wvi)g(vi,vi)Vi-
~1 ~1

traceg = C(gU) = gU(dx\ox;) = g((dxi)U,ox;)
= g(gij Oxi, Oxi) = gij gji = 8f = 4.
17 More precisely, with p :=>..+(to), we ought to write V>.+,P = o,,o,p+ 0,,1,P = I(eo +e1),
where I : V -+ TpM is the isomorphism described in Example 2.2, p.28.
Solutions 409
Chapter 6

Let V = ViOxi, w= Wioxi, X = XkOxk, CE JR, f E C 00 (1Rm). Then:
(Cl) v'v(W +cX) = V(Wi+cXi)oxi = (VWi+cVXi)oxi
= VWiox;+cVXioxi
= v'vW +cv'vX.
(C2) v'v(JW) = V(JWi)oxi
= (V J)WiOxi + f(VWi)Oxi
= (VJ)W+fv'vW.
(C3) v'1v+xW = ((JV +X)Wi)oxi = (JVWi+XWi)oxi
= f (VWi)Oxi + (XWi)Oxi
= fv'vW +v'xW.
(C4) v'vW - v'wV = (VWi)oxi - (WVi)oxi. Also,
[V, W]f = ViOxi(WjOxif) - WjOxi(ViOxif)

= V\ox;Wi)(ilxif) + ViWioxiOxif
-Wj(Oxi Vi)(Oxif) - WjViOxiOxif
= ((V(Wi) - W(Vi))ilxi )J.
Thus [V, W] = (V(Wi) - W(Vi))oxi = v'vW - v'wV.

(C5) Finally, we have
(v'vW,X) + (W, v'vX)

= (V(Wi)Oxi,XjOxi) + (WiOxi, V(Xj)Oxi)
= V(Wi)Xi(Oxi, Oxi) + WiV(Xi)(Oxi, Oxi)
= (V(Wi)Xj + WiV(Xi))(Oxi,Oxi)
= (V(Wi Xi))(Oxi, Oxi) = (V(Wi Xi))(Oxi, Oxi) + 0
= (V(WiXi))(oxi, Dxi) + WiXiV(8ii)
= (V(Wi Xi))(oxi, ilxi) + WiXiV( (ox;, ilxi))
= V(WiXj(Oxi,Oxi)) = V(WiOxi,XjOxi)
= V(W,X).
(In the above, we used V(8ij) = 0, since 8ij is a constant.)


We have
v'vW = ((yzox + ZXOy + xyoz)(xy 2 + z))ox

+((yzox + ZXOy + xyoz)(y 2 - x))oy
+((yzox + ZXOy + xyoz)(x + z 3 ))oz
= (yz(y2)+ zx(x2y) + xy(l))ox
+(yz(-l) + zx(2y) + xy(O))oy
+(yz(l) + zx(O) + xy(3z 2 ))oz
= (zy 3 + 2x 2 yz + xy)ox + (2xyz - yz)oy + (yz + 3xyz 2 )oz.
So v'vp W = (v'vW)(p) = 4ox,v + oy,p + 4oz,v·

~ .. k ~ -
Define v'vW=V3 W'rijOxk- Then v'v=v'v+v'v.
We know that v' satisfies (Cl), (C2), (C3). We have:
~
(1) v'v(W +cX) -_Vj (W i +exi)rijOxk
k -_VjW
ikrijOxk +cV jXi rijOxk
k
=VvW+cVvX.
(2) Vv(JW) = Vi(JW)irfjOxk = JViWiI'fjOxk = JVvW.
(3) VJV+xW = (JV +X)iWiI'fjOxk = JViWiI'fjOxk + XiWiI'fjOxk
=JVvW+VxW.
Hence:
(Cl) v'v(W +cX) = Vv(W +cX)+v'v(W +cX)
~ VvW+cVvX+v'vW+cv'vX
= v'vW +cv'vX.
(C2) v'v(JW) = Vv(JW)+v'v(JW)
~ JVvW+(VJ)W+fv'vW
= fv'vW+(VJ)W.
(C3) v' JV+xW = VJV+xW + v' JV+xW
~ JVvW+VxW+fv'vW+v'xW
= fv'vW+v'xW.
So v' is a connection.
Solutions 411

First we note that if cp, f E C 00 (M) and V, WE TJ-M, then
[JV, W]cp = JV(Wcp) - W(!Vcp) = JV(Wcp) - (WJ)(Vcp) - JW(Vcp)
= f(V(Wcp) - W(Vcp)) - (WJ)(Vcp)
= J([V, W]cp) - (WJ)(Vcp) = (J[V, W] - (WJ)V)cp.
Thus [JV, W] = f[V, W]- (WJ)V. Now for f E C 00 (M), V, W,X E TJ-M,
B(JV + X, W) = v'Jv+xW -v'w(JV + X)- [JV+ X, W]

= fv'vW + v'xW - (WJ)V -fv'wV -v'wX
-f[V, W] + (WJ)V - [X, W]
= f(v'vW - v'wV - [V, W]) + v'xW - v'wX - [X, W]
= f B(V, W) + B(X, W).
Also,
B(V, W) = v'vW - v'wV - [V, W] = -(v'wV - v'vW - [W, V])

= -B(W, V).
Thus B is C 00 (M)-bilinear. Finally, Tis C 00 (M)-linear in the first slot:
T(JO + e, V, W) =(JO+ e)(B(V, W)) = JO(B(V, W)) + e(B(V, W))
= JT(O, V, W) + T(0, V, W),
and also in the latter two slots, e.g.
T(O, V,JW+X)=O(B(V,JW+X)) = 0(-B(JW+X, V))

=0(-JB(W, V)-B(X, V))=O(JB(V, W)+B(V,X))
=JO(B(V, W))+O(B(V,X))=JT(O, V, W)+T(O, V,X).
So T is a (1, 2)-tensor field.

For V, w,x E TJ-M, f E C 00 (M), c ER, and with vc:=(1- p)v' + pV:
(Cl) vcv(W +cX) = (1- p)v'v(W +cX) + pVv(W +cX)

= (1- p)(v'vW + cv'vX) + p(VvW + cVvX)
= (1- p)v'vW + pVvW + c((l - p)v'vX + pVvX)
= vcvw + cv'cvx.
(C2) v'cv(JW) = (1 - p)v'v(JW) + pVv(JW)

= (1- p)((VJ)W + fv'vW) + p((VJ)W + JVvW)
= (1- p + p)(VJ)W + J((l - p)v'vW + pVvW)
= (VJ)W + fv'cvW.
(C3) v'cJv+xW = (1- p)v'1v+xW + pVJv+xW
= (1- p)(Jv'vW + v'xW) + p(JVvW + VxW)
= /((1- p)v'vW + pVvW) + (1- p)v'xW + pVxW
= fv'cv w + vex w.
So v7c is also a connection.

With f =x, V =ox and W =ox,
2v'v(J-W) = 2v'a.,(xox)=2(oxx)ox=2ox, while
(VJ)W + f2v'vW = (oxx)ox + x2(oxl)ox=ox + O=ox.
So 2v'v(J-W)=2ox#-Ox=(VJ)W + f2v'vW.
For f E C 00 (M), V, W,X E TJM, !1,8 E TfM, we have:
T(n + Je, V, W) = (n + f8)(v'vW-VvW)
= n(v'vW-VvW) + f8(v'vW-VvW)
= T(n, V, W) + f T(e, V, W),
T(n,Jv+x, W) = n(v'1v+xW-V1v+xW)
= n(Jv'vW + v'xW-JVvW-VxW)
= Jn(v'vW-VvW) + n(v'xW-VxW)
= f T(n, V, W) + T(n, X, W).
Finally,
T(n,V,JW+X)
= n(v'v(JW +X) - Vv(JW +X))
= n((VJ)W + fv'vW
+ v'vX - (VJ)W - JVvW - VvX)
= n(J(v'vW-VvW) + v'vX-VvX)
= Jn(v'vW-VvW) + n(v'vX-VvX)=JT(n, v, W) + T(n, v,x).
So T is a (1, 2)-tensor field.
Solutions 413

Firstly, for V, WE f, cp E C 00 (M), we have
TJ M, and
[V,JW]cp = V(!Wcp) -JW(Vcp) = (VJ)(Wcp) + JV(Wcp) -JW(Vcp)
= (VJ)(Wcp) + J[V, W]cp.
So .Cv(JW) = (VJ)W + f.CvW, that is, (C2) holds. Moreover, we have
for V, W,X E TJM and c E JR
.Cv(W+cX) = [V, W+cX] = [V, W]+c[V,X] = .CvW+c.CvX,
and so (Cl) holds.
With f = x, W =Ox= V and X = 0, we have that
.CJv+xW = .CJvW = .Cxa,,Ox = [xox, ox].
For cp E C 00 (JR 2 ),
[xox, Ox]'P = XOx(Ox'P) - Ox(xoxcp)
= XOx(Ox'P) - (oxx)(oxcp) - XOxOx'P = -lOx'P·
So [xox, ox]= -ox. Thus .CJv+xW =-ox.But
.CJv+xW = -ox --I= 0 = xO + 0 = x[ox, ox]+ [0, ox]
= x.Ca,,ox + .Coox = f.CvW + .CxW
Hence (C3) fails to hold.

We have by Lemma 6.1 that (v' v W)p = (v' v W)p- We want to show that
(v'y-W)p=(v'vW)p- We have
(C3)
(v'vW)v -(v'vW)v=(v'vW-v'vW)v = (v'v-vW)v-
Let (U,x) be a chart such that p EU. Set V- V =X. Writing X =Xiox;
in U, we have Xi(p) Oxi,p = Xp = Vp - Vp = 0. So Xi(p) = 0, 1 ~ i ~ m. By
Proposition 6.1,
(v'xW)p = ((v'xW)lu)v = (v'f1uWlu)v = (v'f;a,,;Wlu)v
= (X i v'aux1. Wlu)v = X i (p)(v'aux1. Wlu)v = 0-(v'aux". Wlu)v = 0.
Thus
(v'vW)v - (v' v W)v = (v'v-v W)v= (v' xW)v =0,
that is, (v'vW)v=(v'y-W)p-
Consequently, (v' vW)p = (v'vW)p = (v'vW)p-

We have that in (U, x),
divV = dxi('\laxi V) = dxi('\lax;(Vioxi))

= dxi(oxi Vi Oxi + Vi'\J axi Oxi)
. . . k
= dx'(ox;V 1 oxi + v 1 rjiOxk)
= OxiV k 8ki + V j rji8k
k i
= oxi Vi+ vir;i•

Recall from Lemma 3.3 that [ox;, Oxi] = 0. From the torsion-freeness,
'\lax;oxi - '\laxioxi = [ox;,oxi] = 0.
So rfj = dxk('\laxiOxi) = dxk('\lax;Oxi) = rji.

Uniqueness: Let '\lvox; = (O{V)oxi• Acting by dxk, we obtain
. k k . k
(O]V)8i = dx ((O]V)oxi) = dx ('\lvox;),
that is,
k k .
Thus Oi = riidx1 .
Existence: Define Of : TJU --> 0 00 (U) by Of = rfidxi E TfU. If V E TJ M,
then Vlu E TJU, and we can decompose Vlu = vtaxl, giving
. . £
(Oi V)Oxi = r]rdxr (V Oxl )Oxi = rtr. £
V 8I Oxi
-- rjir vr UxJ
:l . -
-
vrr, :l .
V Oxr Ux-i
Recovering r-symbols from the connection I-form matrix [O;]: We have

Ofoxi = rftdxtoxi = rft8J = rfi.
Solutions 415

detG= I; (sign7r)g11r(l) · · · gm1r(m), and so by the product rule
1rES-m
m
= I; (sign 1r) I; gl7r(l) · · · g(i-l)1r(i-l) (Oxkgi,r(i)) g(i+1)1r(i+l) · · · gm1r(m)
1rES-m i=l
m
= I; I; (sign1r) gl7r(l) · · · g(i-l)1r(i-l) (Oxkgi,r(i)) g(i+1)1r(i+l) · · · gm1r(m)
i=l 1rES-m
m g(i-1)1 ... g(i-l)m mm

I; det Oxkgil · · · Oxkgim I; I; (-l)i+i(Oxkgij) detGii,
i=l g(i+l)l ... g(i+l)m i=l j=l
gml gmm
where the summation convention is not used here, and Qii is the (i,j)th
minor of G, that is the matrix obtained from G by deleting its ith row and
jth column.
As G is invertible, we have a- 1 = de!a[adj G], where adj G is the
adjugate of G (the matrix whose entry in the ith row and jth column is
(-l)i+i detGii_ Since G is a symmetric matrix, we have Qii = (Gii)t, and
so detGii =det((Gii)t) =detGii_ Hence
(- 1 )i+j
(G- 1)ii = detGii_
detG
So (not using the summation convention),
m m
Oxk detG = ~ ~(-l)i+i(Oxkgij)(detG)(G-l)ij(-l)i+j
i=l j=l
m m
= (detG)~ ~ gijOxkgij·
i=l j=l
k gkl
Recall that rii = 2 (oxigje + Oxig;e - oxtgij)-
m j m m gif
~ rij = ~ ~ 2(0xigje + Oxigie - Oxlgij)
j=l j=ll=l
m m gif m m gif
= ~ ~ 2 ox;gje +~ ~ 2 (oxigie - Oxtgij)
j=ll=l j=ll=l
m m gif
= ~ ~ 2 ox;gje + 0,
j=ll=l
where we have used gif = gfi, and exchanged the dummy indices j, e in
order to obtain the O in the line above. (Indeed, if
m m gif
S := ~ ~ 2 (c'\igu - c'\tgij),
j=lf=l
then after the j +-+ eexchange,
m m gfj m m gif
S = ~ ~ 2 (oxtgij - Oxigie) = ~ ~ 2 (-l)(Oxigie - Oxtgij) = -S,
f=lj=l j=lf=l
giving S = 0, as claimed.) On the other hand, also
Dxi (log-v'- det G) = Dxi log(-:et G) =
-2 et
~ G (- det G) f: f: gieoxigje
i=U=l
m m gif
= ~ ~ -Oxigje-
i=lf=l 2
Here we used, for h E C 00 (U) and a real-valued f smooth on an open set
containing h(U), that OxiU Oh)=(!' 0 h)ox;h: for all p Eu,
o(J oho x- 1 ) , ,
(ox;(Joh))(p) = 0u•. (x(p)) = f (h(p))oxi ,Ph=((! oh)ox;h)(p).
We recall that G = [ guu guv] = [l+v 1 UV]

2
gvu gvv UV +u2 . Thus
guu guv] 1 [l+u2 -UV]
a-1 = [ gvu gvv = (l+v2)(1+u2) - u2v2 -UV l+v2 .
Hence
u guv guu
r uu = 2 (ouguv + ouguv - ovguu) + 2 (ouguu + ouguu - ouguu)
-uv l+u 2
2(1+u2+v2) (v+v-2v) + 2(1+u2+v2) 0 = 0,
u u guu guv
r UV = r vu= 2 (Ouguv + Ovguu - Ouguv) + 2 (Ougvv + Ovguv - Ovguv)
l+u 2 2 -uv 2 v
2(1+u 2+v 2) ( v) + 2(1+u2+v 2) ( u) = l+u2+v 2 ·
By the u +-+ V symmetry, r~v = r~u = u2 2 ' and r~v = 0. Finally,
l+u +v
u guu guv
rvv = T(ovguv+ Ovguv- Ougvv)+T(ovgvv+ Ovgvv- Ovgvv)
l+u 2 -uv
----(u+u-2u) + - - - - 0 = 0,
2
2(1+u +v ) 2 2(1+u2+v 2)
and so again by the u +-+ v symmetry, r~u = 0.
Solutions 417

The additivity is clear. We simply show the C 00 (M)-homogeneity. We have
for X, Y, Z E TJ Mand f E C 00 (M) that
R(JX,Y)Z = '11x'\i'yZ- '\i'y'\i11xZ- "ux,Y]Z
= f'\i'x'\i'yZ- '\i'yf'\i'xZ- "f[X,Y]-(Yf)Xz
= f'\i'x'\i'yZ-(YJ)'\i'xZ-f'\i'y'\i'xZ-fV[x,YJZ+(YJ)'\i'xZ
= f('\i'x'\i'yZ - '\i'y'\i'xZ - '1[x,Y]Z) = f R(X, Y)Z.
Since we have the symmetry property R(X, Y)Z = -R(Y,X)Z, it follows
that R(X,JY)Z = -R(JY,X)Z = -JR(Y,X)Z = JR(X,Y)Z. Finally,
R(X,Y)JZ
= '1x'\i'y(JZ) - '\i'y'\i'x(JZ) - '1[x,YJ(JZ)
= '1x((YJ)Z+ f'\i'yZ)-'\i'y((XJ)Z+ f'\i'xZ)-([X, Y]J)Z-f'1[x,Y]Z
= X(Y J)Z +(YJ)'\i' xZ +(XJ)'\i'yZ + JV x'\i'yZ-Y(XJ)Z-(XJ)'\i'yZ
-(YJ)'\i'xZ - f'\i'y'\i'xZ - X(YJ)Z + Y(XJ)Z - f'1[x,Y]Z
= f('\i'x'\i'yZ - '\i'y'\i'xZ - '1[x,Y]Z) = f R(X, Y)Z.
To show the well-definition of R(p), we proceed as in the proof of The-
orem 4.1. Let (U, x) be an admissible chart containing p. Let x be a
bump function that is identically 1 in a neighbourhood of p, and identi-
cally zero outside an open set U0 whose closure U0 is contained in U. Let
X,X,Y,Y,Z,Z be vector fields such that Xp = Xp = x, Yp = Yp = y,
and Zp = Zp = z. We decompose X = XiOxi, etc. Then we note that
X\p) = X\p), etc. We have
(R(X, Y)Z)p = l(R(X, Y)Z)p = (x(p)) 6 (R(X, Y)Z)p = (x 6 R(X, Y)Z)p
= (R(x 2 X, x 2 Y)(x 2 Z))p = (R(x 2 XiOxi, x 2 Yj Oxi )(x2 zkaxk ))p
= (xXi)(p)(xYi)(p)(xzk)(p)(R(xoxi, XOxi )xoxk )p
= 13 Xi(p)Yi(p)Zi(p)(R(xoxi, XOxi )XOxk )p
= 13 _xi(p)Y\p)zi(p)(R(xoxi, XOxi )XOxk )p
= • • • (retrace the steps) • • • = (R(X, Y)Z)p-
In any affine chart (U,x), rfj = 0 for all i,j,k. Also using [oxi,Oxi] = 0,
dxf_(R(Oxi,Oxi)Oxk) = dxf_('\i'a,,i '\i'a,,jOxk - '\i'a,,j '\i'a,,;Oxk - '1[o,,;,i3,,j]Oxk)
= dxf_('\i' O,,i (rkjOxr) - V O,,j (rkiOx•) - 0) = dxf_(O) = 0.
Thus Rfjk = 0 in the chart (U, x), for 1 ~ i, j, k, £ ~ m. As these charts
cover M, R = 0 E T}M.
We have, in the case of the sphere with the chart (U, r.p), that:
R(o0,0¢)00 = "va 8 "va"'o0 - '\la"' "va 8 00 - "v[a 8 ,a<t>Jo0
= "v i3a (r:cf,00+ rtcf,a¢) - "v a
= -(sin0)- 2 ocp + (cot0)(r:000 + r!0ocp)

= -(sin0)- 2 o¢ + (cot0)2oc/> = -oc/>.
Hence R(dc/>,00,0¢,00) = dcp(-0¢) = -oc/>¢ = -1. (So Rtc/>0 = -1 =I- 0.)

We have for all WE TJM,
("vv(Jn))w = v(Jnw) - Jn("vvW)
= (VJ)nw + JV(nw)-Jn(VvW) = ((VJ)n+ J(Vvn))w.
Thus "vv(Jn) = (VJ) n + f ("vvn). Also,
("vv(n + e))w = V((n + e)w)-(n + e)("vvW)
= v(nw +ew)-n("vvW)-0('\lvW)
+ V(0W) - n("vvW) - e("vvW)
= V(nw)
= ("vvn)w + ("vv0)W = ("vvn + "vv0)W.
Thus "vv(n + 0) = "vvn + "vv0.
So "vv(Jn + 0) = "vv(Jn) + "vv0= (V J)n + f "vvn + "vv0.
We have for XE TJM
("v Jv+wn)(X) =(JV+ W)(nx) - n(v Jv+wX)
= JV(nx) + w(nx) - n(J"vvX + "vwX)
= JV(nx) + w(nx) - Jn("vvX) - n("vwX)
= J(V(nx) - n(VvX)) + w(nx) - n("vwX)
= f("vvn)(X) + (Vwn)(X) = (J"vvn + Vwn)(X).
Consequently, "v JV+wn = f"vvn + "vwn.

For all X E TJ M, we have, using the metric compatibility, that
("vv(Wi,))X = V(Wi, X) - Wi,("vvX) = V(g(W,X)) - g(W, "vvX)
= V(g(W,X)) + g("vvW,X)- V(g(W,X))
= g("vvW,X) = ("vvW)i,X.
Solutions 419

Using Definition 6.4, we have
(v'vR)(n,x, Y, Z)
= V(R(O,X, Y, Z)) - R(v'vO,X, Y, Z) - R(O, v'vX, Y, Z)
-R(O,X, v'vY, Z) - R(O,X, Y, v'vZ)
= V(-R(O, Y,X,Z)) + R(v'vO, Y,X,Z) + R(O, Y, v'vX, Z)
+R(O, v'vY,X, Z) + R(O, Y,X, v'vZ)
-------------
= -(V(R(O, Y, X, Z)) - R(v'vO, Y, X, Z) - R(O, v'v Y, X, Z)
-R(O, Y, v'vX, Z) - R(O, Y,X, v'vZ))
-------------
= -(v'v R)(n, Y, X, Z).

1°r~2ors~l. Forann1, ... ,nrETfMand W1,··· ,WsETJ-M,
(v'v(JT))(01,- ··,or, W1, · · ·, Ws)

= V(JT(01,- · · , or, W1, · · · , Ws))
r
-~JT(01,··· ,v'vO\··· ,nr,W1,··· ,Ws)
i=l
s
-~JT(01,··· ,nr,W1,··· ,v'vWj,••· ,Ws)

j=l
= (VJ)T(01,- ··,or, W1, · · · , Ws) + JV(T(01,- · · ,nr, W1, · · ·, Ws))
r
-f~T(01,··· ,v'vO\··· ,nr,W1,··· ,Ws)
i=l
s
-f~T(01,··· ,nr,W1,··· ,v'vWj,••· ,Ws)

j=l
= (VJ)T(0 1 ,··· ,or,W1,··· ,Ws)
+f(v'vT)(01,- ··,or, W1, · · ·, Ws)
= ((V J)T + f(v'vT)}(01, · · · , or, W1, · · · , Ws)-
Thus v'v(JT) = (V J)T + f(v'vT).
2° r=s=O. Let T=:g.
Then v'v(Jg) = V(Jg) = (V J)g+ f(V g) = (V J)g+ fv'vg.
3° r=l, s=O. Let T=:W.
Then v'v(JW)=(VJ)W+fv'vW is precisely (C2).

1° r=s=0. Let T=:g.
Then v' JV +w9 =(JV+ W)g= JV g+ W g= fv'vg+ v'wg.
2° r=l, s=O. Let T=:X.
Then v'1v+wX =fv'vX +v'wX is precisely (C3).
3° r~2 or s~l. For all n1, ... ,nr Erp Mand X1, ... ,Xs E TJM,
(v'1v+wT)(O1,--- ,nr,X1,··· ,Xs)
r
- ~T(01,--- ,v'Jv+wO\··· ,nr,X1,··· ,Xs)
i=l
s
- ~T(01,--- ,nr,X1,··· ,v'Jv+wXj,••· ,Xs)

j=l
= JV(T(O1,--- ,nr,X1,··· ,Xs)) + W(T(O1,--- ,nr,X1,··· ,Xs))
r
- ~T(01,--- ,fv'vOi+v'wO\··· ,nr,X1,··· ,Xs)
i=l
s
- ~T(01,--- ,nr,X1,··· ,fv'vXj+v'wXj,••· ,Xs)

j=l
= JV(T(O1,--- ,nr,X1,··· ,Xs)) + W(T(O1,--- ,nr,X1,··· ,Xs))
r
-f ~ T(01, · · ·, v'vO\ · · · ,nr,X1, · · · ,Xs)
i=l
r
- ~T(01,--- ,v'wO\··· ,nr,X1,··· ,Xs)
i=l
s
-f ~T(O1,--- ,nr,X1,··· ,v'vXj,••· ,Xs)

j=l
s
- ~T(01,--- ,nr,X1,··· ,v'wXj,••· ,Xs)

j=l
= f(v'vT)(O1,- ··,or, X1, · · ·, Xs)
+(v'wT)(01, · · · , nr, X1, · · · , Xs)

= (J(v'vT) + (v'wT)}(0 1, · · · , nr, X1, · · · , Xs)-
Thus v' JV+W = fv'vT + v'wT.

Solutions 421

As dxi = 8~dxr, by Example 6.9, '\lvdxi = (V8~ - 8}Vsr~s)dxr. So
('\lvT);k = ('\lvT)(dxi, Oxi, Oxk)

= V (T( dxi, Oxi, Oxk)) - T('\l vdxi, Oxi, Oxk)
-T(dxi, '\lvoxi, Oxk) -T(dxi, Oxi, '\lvoxk)
= VTJk - T((V 8; - 8~vsr~s)dxr, Oxi, Oxk)
-T(dxi, vsr;soxr, Oxk) -T(dxi, Oxi, vsq;soxr)
= VTJk + Vsr;sTjk - VsrjsT;k - VsrksT]r
= VTJk + I';sTjk Vs - r;sT;k Vs - I'ksT]r Vs.
In the above we used V(8~) = 0 since 8~ is a constant function.

For !1,0 E TfM and XE TJM,
(Vv(W ® T))(n, e, X)
= V(W(n)T(e, X)) - (W ® T)('\lvn, e, X)
- (W © T)(n, '\lv0, X) - (W © T)(n, e, '\lv X)
= V(Wn)T(e, X) + (Wn)V(T(e, X)) - W('\lvn)T(e, X)
- (Wn)T('\lv0, X) - (Wn)T(e, '\lv X)
= (V(Wn) - W('\lvn))T(e, X)
+ W(n)(V(T(0,X)) -T('\lv0,X) -T(e, '\lvX))
= (v(nw) - (Vvn)W)T(e, x) + w(n)(VvT)(e, x)
= (v(nw) - v(nw) + n(VvW))T(e,x) + (W® (VvT))(n,e,x)
= ('\lvW)(n)T(e,x) + (W®('\lvT))(n,e,x)
= (('\lvW) ®T)(n, e, X) + (W ® ('\lvT))(n, e, X)
= (('\lvW) ®T + w © ('\lvT))(n, e, X).

We have
'\lv(g®T) = ('\lvg) ©T+g© '\lvT=O®T+g© '\lvT=g© '\lvT

'\lv(T ©g) = ('\lvT) ®g+T® '\lvg= ('\lvT) ©g+T©O= ('\lvT) ® g.

(1) Fod1 E TfM, VE TJM,
(Tl>)U(o, V) = Tl>(ou, V) = T((OU)I>, V) = T(O, V).
So (Tl>)U = T.
(2) For V, WE TJM, we have
(TU)l>(V, W) = TU(vl>, W) = T((Vl>)U, W) = T(V, W).
So (TU)I> = T.
(3) Let (U, x) be an admissible chart for M.
Then (Oxi)I> = (Oxi)l>(oxk)dxk = g(Oxi,Oxk)dxk = gikdxk, and so
(Ti,)ij = Ti, (Oxi, Dxi) = T( (Oxi )1,, Dxi) = T(gikdxk, Dxi)
= gikT(dxk, Oxi) = gikTr (17.2)
For OE TP M, v,x, w E TJM, we have
(g©T)(O, V,X, W) = g(V,X)T(O, W).
Hence (C~(g©T))(V, W) = g(V,Oxk)T(dxk, W). Thus
(C~(g®T))ij = g(Oxi,Oxk)T(dx\oxi) = gikTr (17.3)
From (17.2) and (17.3), it follows that Ti,= C~(g©T).
(4) Let TE TlM. Then
v'v(Ti,) = v'v(C~(g ® T)) = C~(v'v(g © T))
= C~(g © v'vT) = (v'vTt
(5) Let TE Tf M. Then v'vT E Tf Mand
v'vT ~ v'v((TU)I>) ~ (v'v(TU)t (17.4)

For all WE TJM, we have
(.Cv(JO + e))(W)
= V((JO + e)(W)) - (JO+ e)[v, W]
= V(JOW + ew) - JO[V, W] - e[V, W]
= (V !)(OW) + JV(OW) + V(0W) - JO[V, W] - 0[V, W]
= (V !)(OW) + f(.CvO)(W) + (.Cv0)(W)
= ((VJ)O + f.CvO + .Cv0)(W).
Solutions 423

The first claim follows immediately from .Cvg = 0, since this implies that
for all X, YE TJM,
0 = (.Cvg)(X, Y)
= V(g(X, Y)) - g(.CvX, Y) - g(X,.CvY)
= V(g(X, Y)) - g([V, X], Y) - g(X, [V, Y]).
For the second claim, we start with the above, and use the torsion-freeness
(C4) and metric compatibility (C5) of v':
0 = V(g(X, Y)) - g([V, X], Y) - g(X, [V, Y])
<~l + g(X, v'v Y) - g([V, X], Y) - g(X, [V, Y])
g(v'v X, Y)
= g(v'vX - [V,X], Y) + g(X, v'vY - [V, Y])
<~l g(v'xV, Y) + g(X, v'yV),
proving the second claim.

We use the global chart (JR 3 , id), and label the coordinate maps x 1 , x 2 , x 3
by x, y, z. We have
(.CL.,g)ij = (.CL.,g)(Oxi, Oxi)
= Lx(g(oxi,Oxi))-g([Lx,ox;],oxi)-g(ox;,[Lx,oxi])
= Lx(8ij) - g([yoz - ZOy, Oxi], Oxi) - g(Oxi, [yoz - ZOy, Oxi])
= 0 - g([yoz - ZOy, Oxi], Oxi) - g(Oxi, [yoz - ZOy, Oxi ]).
Using Exercise 3.11, we have
[yoz - ZOy, Oxi] = [yoz, Oxi] - [ZOy, Oxi]
= y[oz,ox;]- (ox;y)oz - z[oy,ox;] + (ox;z)oy
= yO - 8i20z - zO + 8i30y = 8i30y - 8i20z.
So .CL.,g = 0, since for 1 ~ i,j ~ 3, we have
(.CL.,g)ij = -g(8i30y - 8i20z, Oxi) - g(Oxi, 8j30y - 8j20z)
= -8i38j2 + 8i28j3 - 8i28j3 + 8i38j2 = 0.
That .CLyg = 0 and .CL,g = 0 can be shown analogously, or by employing
the symmetry in the expression for g = dx ® dx + dy ® dy + dz ® dz and
cycling through x -> y -> z -> x to go from Lx to Ly to Lz.
Chapter 7

Let Z(t) = zi(t)oxi,"Y(t), tEI. Then
W(s)=Zi(h(s))Oxi,"Y(h(s)) (s E J).
For ans E J, let (U, x) be an admissible chart containing 'Y(h(s)). We have
• k k ,..-,.._._ . i
(v'v-rohW)(s) = (W (s) +rij('Y(h(s))) ('Yoh )1 (s)W (s))oxk,("Yoh)(s)· (17.5)
But
Wk(s) = d(Zk(h(·))) (s)
ds
dzk . . "k
= ~(h(s))h(s) = h(s)Z (h(s)). (17.6)
Also, by Exercise 2.4,
V"Yoh,("Yoh)(s) = h(s)v"Yoh,"'f(h(s)) = h(s)-i(h(s))Oxi,"Y(h(s))
= h(s)-/(h(s))Oxi,("Yoh)(s),
so that
,.......,,..___ . . .
( 'Yoh )1(s) = h(s)-yl(h(s)). (17.7)
Substituting (17.6) and (17.7) in (17.5), we obtain
(v'v-roh W)(s)
= (h(s) zk (h(s))+ rf/'Y(h(s))) h(s) ,yi (h(s)) zi(h(s))) Oxk,("Yoh)(s)
= h(s )(zk(h(s)) + rf/'Y(h(s)) )-yi (h(s)) zi(h(s))) Oxk,"Y(h(s))
= h(s) (v'v-rZ)(h(s)).
Hence for alls E J, (v'v-roh W)(s) = h(s) (v'v-rZ)(h(s)).

1 ~
Define the vector field XE To'Y by X(t) = W"Y(t) = W"Y(t) fort E J. We
have
(v'v-rX)(O) = (v'vW)"Y(o) = (v'vW)p- (17.8)
Also,
(v'v-rX)(O) = (v'yW)"Y(o) = (v'yW)p- (17.9)
From (17.8) and (17.9), it follows that (v'vW)p = (v'yW)p-
Solutions 425

Y(t) = Ocf>,µ(t), and so Y 9 = 0, ye/>= 1, giving Y 9 = 0 = ye/>_ We had seen
in Example 7.3 that fl = 0 and µcf> = l. Thus
(v'v,, Y)(t) = (Y 9 (t) + rf/µ(t)) µi (t) Yi(t)) 00,µ(t)
+ rt(µ(t))µi(t)Yi(t))acf>,µ(t)
+(Y(t)
= cr:0(µ(t))µ 9 (t)Y(t) + r:<1>(µ(t))µ<l>(t)Y(t))o0,µ(t)
+(r:e(µ(t)) µ0(t) y(t) + r:c/>(µ(t)) µ<l>(t) y(t)) acf>,µ(t)
= (0µ 9 (t)Y(t) - (sin0)(cos0)10=Jµ(t)Y(t))o0,µ(t)
+((cot0)10=1µ 9 (t)Y(t) + Oµ(t)Y(t))oc/>,µ(t)
= (0 - (1)(0)-1·1}00,µ(t) + (0-0-1 + O}oc/>,µ(t) = 0+0 = 0.
So Y is parallel along µ.
Finally, for all t E (0, 21r), we have
(v'v,, (aX + ,BY))(t) = a(v'v,,X)(t) + ,B(v'v,, Y)(t) = 0+0 = 0.
Hence aX + ,BY is parallel along µ.
Chapter 8

Let 'Y: J--> Ube a geodesic, and set (0(t), ¢(t)) = rp('Y(t)) for all t E J.
Then 'Y(t)= ( (sin 0(t))( cos ¢(t) ), (sin 0(t) )(sin ¢(t) ), cos 0(t)) for all t E J.
The geodesic equation gives:
O(t) + r: 0(t)O(t) + 2r:"'0(t)¢(t) + r!<1>¢(t)¢(t) = o
0
¢(t) + rt 0(t)O(t) + 2rt"'0(t)¢(t) + r:"'¢(t)¢(t) = o

0
that is,
0(t) - (sin0(t))(cos0(t))(¢(t)) 2 = 0
¢(t) + 2(cot0(t))0(t)¢(t) = 0.
We had seen that in (U, rp), the component matrix for g is given by
[::: ::: ] = [ ~ (si~0) 2 ] '
and so for v'"Y,'"Y(t) =000,'"Y(t) + we have

<fao<P,'"Y(t),
"2 • 2 "2 (17.10)
1 = g('Y(t))(v7 ,7 (t),v'"Y,'"Y(t)) = (0) + (sm0) (¢) .
The second geodesic equation is ¢ = -2(cot 0) 0¢, that is,
d • ¢ cos0 • d
-log¢=-:-= -2--0 = -2-log(sin0).
dt ¢ sin0 dt
Integrating from t = 0 tot, we get log ~(t) = -2 log si~ 00(t), that is,
¢(0) sm o
· ¢(0)(sin0o) 2 c
(17.11)
¢(t) = (sin0(t)) 2 = (sin0(t)) 2 '
where c:=¢(0)(sin00 ) 2. Substituting¢ from (17.11) in (17.10) gives
. 2 . 2 c2
(0) +(sm0) (sin0)4 =l.
So
c2
1---- (17.12)
(sin0)2'
where we consider the portion of the geodesic traversed in the 'increasing 0
direction', so that 0 > 0. Thus we can invert the function 0. Leth= 0- 1 .
Solutions 427
We can then view </> as a function of the '0-variable' by considering the

composition</> oh. Dividing (17.11) and (17.12) yields by the chain rule
d(<f>oh) 0 = ~(h(0)) = c 1 c(cosec0) 2

d0 ( ) 0(h(0)) (sin0) 2 ,./1 - c2/(sin0) 2 \/1 - c2 - c2(cot0) 2
Hence, using the substitution t = n 1- c2
cot B, we obtain
(0 d(<f>oh\0)d0= (0 c(cosec0)2 d0=f-vRcot0 -1 dt.

J0od0 J00,Jl-c\jl- 1 ~:,(cot0) 2 vR cot00Vl-t 2
With a slight abuse of notation, we denote ( o h)(0) simply by¢. Then
<p-<po =Sin-l(ncot0o)-sin-l(ncot0),
1-c2 1-c2
i.e., sin- 1 ( ~cot0) =a-</>, where a:=</>o+sin- 1 ( ~cot00 ). So

vl-c2 vl-c2
n c~s0B = sin(a -</>) = (sina)(cos<f>) - (cosa)sin¢.

sm
We rewrite this as
(sin0)(cos¢)(sina) - (sin0)(sin</>)(cosa) - n c o s 0 = o.
1 - c2
Thus, with ll := (sin a, - COSO'., n)1 -c2

E IR 3 , we have
(,(t), n) nP = (( (sin 0) cos</>, (sin 0) sin</>, cos 0), ( sin a , - cos a, - vl ~c 2 ))
= (sin 0)(cos </>)(sin a) - (sin 0)(sin </>)(cos a) - ~ cos 0 = 0.

vl-c2

By Proposition 7.2 (p.146), and using v'v,, V.,, =0 (as 'Y is a geodesic),
dg('Y(-))(v.,,,.,,c-), X.,,c,))
dt
= g('Y(·))(v'v,, V.,,,X.,,(,)) + g('Y(-))(v.,,,.,,(·), (v'v,,X.,,(,))(-))
= g('Y(·))(0,X.,,c,)) + g('Y(-))(v.,,,.,,c-), (v'v,,X.,,(,))(-))
= g('Y(·))(v.,,,.,,c-), (v'v,,X.,,c,))(·).
By Remark 7.1, for each 7(t), t E I, there exists an open set U and a
V E TJU which extends V.,, in U, i.e., V.,,(t) = V.,,(t) for all t E J. By the
Killing equation (Exercise 6.26),
0 = g(v'vX, V) + g(V, v'vX) = 2g(V, v'vX).
Thus g('Y(t))(v.,,,.,,(t), (v'v,,X.,,(,))(t)) = (g(V, v'v X))('Y(t)) = 0.

Let to EI, and define the point p along the geodesic by p := 7(to)- Propo-
sition 8.2 implies that the map I 3 ti-+ g('Y(t))(v.,,,.,,(t), v.,,,.,,(t)) is a constant
map. Thus for all t EI, g('Y(t))(v.,,,.,,(t),v.,,,.,,(t)) = g(p)(v.,,,p,v.,,,p)- So de-
pending on whether g(p)(v.,,,p, v.,,,p) > 0, = 0, or < 0, it follows that for all
t E I, g('Y(t))(v.,,,.,,(t), v.,,,.,,(t)) > 0, = 0, or < 0, respectively, and hence 'Y is
spacelike, lightlike, or timelike, respectively.

We take a chart as in Example 5.10 (p.89) assuming without loss of gen-
erality that p E U(c S 2 ). We have v.,,,.,,(t) = Ot,.,,(t)· For the components
(t, r, 0, ¢) of 'Y, with·= ¼t,, we have i = 1, r = 0, iJ = 0, ¢ = 0. Using the
connection coefficients from Example 6.8 (p.124), the r-component of the
geodesic equation is
0 =,;: + r~t(t) 2 + r;r(i-) 2 + I'00(0) 2 + r;cf,(¢) 2

= 0 + r~t(1) 2 + r;r(0) 2 + I'99(0) 2 + r;;,c/>(0) 2
= r~t = m
r2
(1 - 2m)'
r
so that r = 2m, a contradiction.
Solutions 429

Define 7 : JR--> M by 7(t) = p for all t ER For any f E C 00 (M),
v~ ~ f = d(J o 7) (t) = d(t I-+ J(p)) (t) = 0.
7 ,7 Ct) dt dt
Thus Vi1 = 0 for all t E R Hence
(v'v"' V7 )(t) = (v'v"'O)(t) = (v'v"'(O·O))(t) = O(v'v"'O)(t) = 0.
(In the term O•0 above, the first O is the scalar O E R) So 7 is a geodesic.
Moreover, 7(0) = p and v7 ,P = 0. Thus 7 is a maximal geodesic passing
through p with initial velocity 0. By Corollary 8.1, this is unique. Theo-
rem 8.1 implies that 711 ="f, that is, 'Y(t) =7(t) =p for all t E J.

Let 'Y: J--> JR 2 be the maximal geodesic passing through p = (a, b) E JR 2 in
the direction ofv = (a,/3) E JR2 . Then with "((t) = (x(t),y(t)), the geodesic
equation becomes
{ x + 2xyii = 0} . h h ... l d" . { x(O) = a } { y(O) = b }

= 0 , wit t e imtia con itions x(O) = a , y(O) = /3 .
They-equation gives y(t) = b+(3t, t E JR. Then the x-equation is x+2xf3 =0.
Multiplying by e2 f3t gives
:/e 213 tx) = e213 tx + 2(3e 213 tx = e213\x + 2(3x) = 0.

Integrating from O to t, we get e2 f3t x - 1 a = 0, and so x = a e - 2 f3t. Again
integrating from O to t yields
x(t) =a+ a lo
t
e
_ 213 T
dr =
{a+ at
i - e - 2 '3•
a + a ----;;:r i
if /3 = 0,
"f /3 O
=I= •
It can be checked that the y and x obtained above solve the geodesic equa-
tion and the initial conditions.
Let p = (a, b) and q = (c, d). We consider the two possible cases:
1° b = d. Take for example f3 = 0, and a = 1. We solve for T E JR so that
x(T) = c: c = x(T) =a+ aT =a+ lT yields T = c - a.
Then y(T) = b + (3T = b +OT= b = d.
2° b =I= d. Take for example f3 = 1 and T = d - b.
Theny(T) = b+f3T = b+l(d-b) = d. Wesolveforasothatx(T) = c:
2(c-a)
c = x (T) = a + a
i-e-2,ST i-e2(d-b) •
213 = a+a 2 gives a = i-e- 2 (d-b).
So given any p, q E JR 2 , there is a geodesic which passes through p and q.
The plots of a few geodesics are shown below.
-2
-3
With J = (a - c,b- c) and h(s) = s + c, s E J, we have h(J) = (a,b).
Fors E J, h(s) E (a, b), and as, is a geodesic, (v'v., V1)(h(s)) = 0, giving
(v'v., h V1oh)(s) = h(s) V1 (h(s))

0 + (h(s)) 2 (v'v., V1 )(h(s))
= O-V1 (h(s)) + (h(s)) 2 ·0 = 0 + 0 = 0.
So, o h: J-----> Mis a geodesic. Also, (, o h)(0) = ,(c) = p, and
V1o h,1oh(O) = h(O) v1,1(h(O)) = 1 v1,1(c) = 0.
By Exercise 8.5, (, o h)(s) = p for alls E J, i.e., ,(s + c) = p for alls E J.

Given any t E (a, b), t - c E J (as a -c < t -c < b -c) and we have that
,(t) = ,(h(t - c)) = (, o h)(t - c) = p. So, is constant.

Using r' = 0 and 0' = 0, and the connection coefficients from Example 6.8
(p.124), we obtain from the geodesic equation that the t- and </>-components
of , satisfy
0 = t" + 2 r~t t' r' = t" + 2 r~t t' 0 = t"

o = <t>" + 2rt<1> r' <t>' + 2rt<1> 0' <t>' = <t>" + 2rt<1> 04>' + 2rt<1> 04>' = 4>".
Solutions 431
This gives the first two equations. For the third equation, we look at the
r-component in the geodesic equation:
0 = r" + r;t(t') 2 + r~r(r') 2 + f00(0') 2 + r;1(¢') 2 = r;i(t') 2 + r;1(¢') 2
= ~(1- 2~)(t')2-r(l- 2~)( sin~)\¢') 2

2m)( t ') 2 - r ( 1 - --;-
= rm2 ( 1 - --;- 2m)( ¢ ') 2.
A rearrangement gives the third equation,
(¢')2 = n; (t')2.
r
(17.13)
As we have that v'"t ,'"t(r) = t'(r)ot ,'"t(r) + </>'(r)o1 ,'"t(r), and since 'Y is timelike
and parametrised by proper time, g('Y(r))(v'Y,'Y(r), v'"t ,'"t(r)) = -1, giving
-1 = - (t ,) 2 ( 1 - --;-
2m) 1r) 2(')
+ r 2 (.sm 2 ¢ 2 = - (')
t 2 ( 1 - --;-
2m) + r 2(')
<p 2.
Using (17.13) to replace (¢') 2 in the above yields, after a rearrangement,
(1 - 3~)(t')2 = 1.
Let P2 = 'Y(O), Pz
= 'Y(T ,b,,) E M be the events along 'Y corresponding to
0
the completion of one circular orbit. Then

t(Pz)-t(P2)= rT.,b. t'(r)dr= To,bit .
Jo -J1-3';'
Let 'Yi be the observer at rest on the surface of the Earth, that is,
'Yi(t) = (t,~,(f?- 1 (~,¢0 )), where (U,(f?) is the spherical coordinate chart
(Example 5.7), and </>o = ¢(P2) = ¢(P2). Let A , A be the events at which
radial light signals from P 2, Pz ,
respectively, reach the observer 'Yl · By
the computation done in Example 5.16 on the gravitational red-shift, it
follows that the proper time elapsed for the observer 'Yi on Earth between
the events A , A is
~ ~ ~
TEacth = v.L-Ii;;(t(A)~ - t(A)) = v1-Ii;;(t(Pz)-
~
t(Pz)) = v1-Ii;;
To,bit
✓ l- 3';' •
Using m «rand m «~'we obtain
To,b,, = ~7 ~ (l _ ! 3m) (l _ ( _ !) 2m) ~ 1 _ ~ m + ..?.:::__

TEacth V1 -_ ~ 2 r 2 ~ 2r R,qi
With the given values,
To,bit
- ---TEacth- -_ -To,bit
- - l ~m ( -1 --3 ) ~ 4.531 X
10-10 •
TEacth TEacth R,qi 2r
Chapter 9

Recall Jacobi's identity: [X, [Y, Z]] +[Y, [Z, X]]+ [Z, [X, Y]] =0. We have
.Cx .Cy Z -.Cy.Cx Z = [X, [Y, Z]] - [Y, [ X, Z]] = [X, [Y, Z]] + [Y, [ Z, X]]
= -[Z, [X, Y]] (Jacobi identity)
= [[X, Y], Z] = .C[x,YJZ·

We have that dxr(R(Oxi,Oxi)Oxk) = R(dxr,Oxi,Oxi,Oxk) = Rrjk" We also
have that dxr(Rfikoxt) = Rfik8e = Rrik" Consequently, the vector fields
R(ox;, Oxi )oxk E TJU and RfjkOxt E TJU coincide (as their components,
dxr(R(ox;,Oxi)Oxk), respectively dxr(RfjkOxt), l~r~m, coincide).

We have
R(X, Y, z, W)+R(Y, z,x, W)+R(Z,X, Y, W)
= g(W, R(X, Y)Z) +g(W, R(Y, Z)X) +g(W, R(Z, X)Y)
= g(W, R(X, Y)Z + R(Y, Z)X + R(Z, X)Y) = g(W, 0) = 0.

~ Y, Z, 0 u). Thus
We have R(O, X, Y, Z) -- R((O p) , X, Y, Z) -- R(X,
Rfjk = R(dxf' Oxi' Oxi' Oxk) = R( Oxi' Oxi' Oxk' (dxf)U)
~
= R(Oxi,Oxi,Oxk,g fr fr~ fr~
Oxr) = g R(Oxi,Oxi,Oxk,Oxr) = g Rijkr·
Alternatively, as Rijkr = grsRfjk' gfrRijkr = gfr grsRfjk = 8!Rfjk = Rfjk•

(R®g)(O,X, Y,Z, V, W)=R(O,X, Y, Z)g(V, W), and so in any admissible
chart (U,x), (CHR@g))(X, Y, Z, V)=R(dxr, X, Y, Z)g(V, oxr ). So
(C~(R®g))ijkf = (C~(R®g))(Oxi,Oxi,Oxk,Oxt)
= R(dxr, Oxi, Oxi, Oxk)g(oxt, Oxr) =Rrjkger =Rijkf•
Hence CHR@g)=R. (As g is symmetric, we also have R = C!(R®g).)
Solutions 433

We have
Recall that r:4> = -( sin 0) (cos 0) , rt¢ = cot 0 = r: 0 are the only nonzero
connection coefficients in the given chart. We have
R:01> = a1>r:0 - 00r:1> + r!0r:1> - r:1>r: 0

= 04>0- 00(-(sin0)(cos0)) + (cot0)(-(sin0)(cos0)) - r:1> 0
= 0 + (cos0) 2 - (sin0) 2 - (cos0) 2 - 0 = -(sin0) 2 .
Thus R 4>04>0 = g00R: 01> = 1 (-(sin 0)2) = -(sin 0) 2 . Then
ft1>1>1>1> = o ft 1>1>1>0 = o =o
ft 1>1>01> ~ ¢¢00 =0
~ ¢0¢¢ =0 R 4>04>0 = -(sin0) 2 R 4>004>= (sin 0) 2 ~ ¢000 = 0
~ 0¢¢4> =0 ~04>4>0 = (sin 0) 2 ~04>04> = -(sin0) 2 ~ 0¢00 = 0
R 001>1> =0 R 004>0 = 0 R 0004> = 0 R 0000 = 0
are the 16 components of the covariant Riemann curvature tensor field in
the given chart. Hence
R!e1> = 0 =1
R ! 00
Rte¢ = 0 Rt00 = 0
R: 4> =
0 -(sin0) 2 R:00 = 0
R: =04> 0 R:00 = 0.

Let (U,x) be an admissible chart for M. Then
Ricij = (C~R)ij = R(dx\oxk, Oxi ,Oxi)
k kl~ kl~
= Rkij = g Rkijl = -g R ikjl
= -gklglrR[kj = -o~R[kj = -R7kj
= -R(dx\Oxi ,Oxk,Oxi )
= -(C~R)(Oxi , Oxi ) = -(C~R)ij ·
Thus Ric = -C~R.

The components of Ric in the chart (U, cp) are given by
Ric00 = R!00 +R: 00 = 1 + 0 = 1,
Rice¢ = R! R:
0 c/> + 0 c/> = 0 + 0 = 0,
Ric¢0 = Rice¢ = 0,
Ric¢¢= R!c/>c/> + R:c/>c/> = 0 + (sin0) 2 = (sin0) 2 .
Thus the scalar curvature is given by
S= + Ric0¢g 9 cf> + Ricc/>c/>gcf>cf> + Ric¢0g 9 cf>
Ric00g 99
= Ric00·l + Ric0¢•0 + Ric¢¢•(sin0)- 2 + Ric¢0•0

= 1-1 + 0 + (sin0)2-(sin0)- 2 + 0 = 1 + 1 = 2.
Thus S = 2 on U. As every point in 8 2 \U is the limit of a sequence of

points in U, it follows by the continuity of S that S = 2 on 8 2 .

In an admissible chart, we have ftijkl = giagibgkcgldRabcd· Using the
chart (U, cp) (Exercise 9.7), and noting that the component matrix for g is
a diagonal matrix, the nonzero components of R are given by
~ cf,0cf,0 - cf,cf, 00 cf,cf, 00 ~ - 1 1 1 1 . 2 - 1
R -g g g g Rcpecpe- (sin0) 2 i(sin0) 2 i(-(sm 0) )--(sin0) 2
~ cf,00cf, - cf,cf, 00 00 cf,cf, ~ - 1 11 1 . 2 - 1
R - g g g g R¢99 c/> - (sin0) 2 i i (sin0) 2 (sm 0) - (sin0) 2
~ 0cf,cf,0 - 00 cf,cf, cf,cf, 00 ~ - 1 1 1 1 . 2 - 1
R - g g g g Re¢¢ 9 - i (sin 0) 2 (sin 0) 2 i (sm 0 ) - (sin 0) 2
~ 0cf,0cf, - 00 cf,cf, 00 cf,cf, ~ - 1 1 1 1 . 2 - 1
R - g g g g R 9 ¢ 9 c/> - i (sin0)2 i (sin0) 2 (-(sm 0) ) - - (sin0) 2 •
Consequently,
~ ..9 ..9
= R.,, .,,
~
R¢0¢0 + R.,,..99 .,,R¢00¢
~ .. ~ ....9 ~
+ R 9 .,,.,,~ 9 ..9 .. ~
R0¢¢0 + R .,, .,,R0¢0¢
~
= - (si: 0)2 (-(sin0)2)+ (si: 0)2 (sin0)2+ (si: 0)2 (sin0)2- (si:0)2 (-(sin0)2)
= 4.
Thus K = 4 on U. As every point in 8 2 \U is the limit of a sequence of

points in U, it follows by the continuity of K that K = 4 on 8 2 .
Solutions 435

We have for V1, · · · , Vs-1 E TJ-M that
(div(T+cS))(V1, · · · , Vs-1)
= C((O, V) ,_. (v'v(T+cS))(nU, Vi,···, Vs-1))
= C((O, V) ,_. ((v'vT)(nu, Vi,··· , Vs-1)+c(v'vS)(nU, Vi,··· , Vs-1)))
= C((n, V) ...... (v'vT)(nu, V1, ... , Vs-1))
+c C((O, V) ,_. (v'vS)(nU, Vi,···, Vs-1))
= (divT)(Vi, · · · , Vs-1)+c(divS)(V1, · · · , Vs-1)
= ((divT)+c divS)(Vi, · · · , Vs-1).
Consequently, div(T+cS) = (divT) + c div 8.

div(Vi,) = (v'a.,;(Vi,))((dxi)U)
= Oxi(Vi,(dxi)U) - Vi,(v'a.,;(dxi)U)
= ox;(g(V, (dxi)U)) - g(V, v'a.,; (dxi)U)
= ox;(g(V, (dxi)U)) + g(v'a.,; V, (dxi)U) - ox;(g(V, (dxi)U))
= ((dxi)Ul(v'a.,; V)
= dxi(v' a.,; V)
= C((O,X) ,_. O(v'xV)} = divV.
Here used the metric compatibility of the Levi-Civita connection to obtain

the equality in the third row.

nr
vve have l =
5 (T)i
t -1. So t he cosmo1og1ca
· 1 time
. t of em1ss1on
. . 1s .
t = (~tT ~ 0.76 · T = 0.76 · (13.8 billion years)

~ 10.5 billion years.
Hence the light was emitted 13.8 - 10.5 = 3.3 billion years ago.

(1) By the formula given in Theorem 7.1 (p.142), v' J(·,t) V(·, t) E TJ-I\ is
(v'J(•,t) V(·, t))(s)
= ( oosot
2 (x o rt k o(x o r)i
(s,t)+rij(r(s,t))ot
o2 (x o r)i ~
(s,t)os2 (s,t))Oxk,r(t,s)·
So (v'J(•,t) V(·, t))(s) = Wk(t)oxk,r.(t), where
Wk= o2(xor)k( ·) I'~-(I'( ·))o(xor)i( ·)o2(xor)i( ·)
osot s, + •J s, ot s, os2 s,
is smooth on I. Consequently, t ....... (v' J(·,t) V(·, t))(s) E TJrs.
(2) Analogous to the proof of (1). We have
(v'v(s,-)J(s, ·))(t)
o2 (x o rt o(x o r)i o2 (x o r)i ~
= ( otos
k
(s,t)+rij(r(s,t))os (s,t)ot2 (s,t))Oxk,r(t,s)·
so that (v'v(-,t)J(·, t))(s) = Xk(s)oxk,f,(s)' where
Xk = o2(x o r)k (· t) I'~- (I'(· t)) o(x o r)i (· t) o2(x o r)i (· t)
osot ' + •J ' OS ' ot2 '
1~
is smooth on (-1:, 1:). Hence s ....... (v'v(s,-)J(s, ·))(t) E T 0 I't.

As rs is a geodesic, v'v(s,-)V(s, ·) = 0. Using Proposition 7.2, we have
d
dt g(I's(t))(J(s, t), V(s, t))
= g(I's(t))((v'v(s,-)J(s, ·))(t), V(s, t))+g(I's(t))(J(s, t), (v'v(s,-) V(s, ·))(t))
= g(I's(t))((v'v(s,-)J(s, ·))(t), V(s, t)) + 0.
We now use the Jacobi equation to get the equality(*) below. We have
d2 d
dt 2 g(I's(t))(J(s, t), V(s, t)) = dt(g(I's(t))((v'v(s,-)J(s, ·))(t), V(s, t)))
= g(I's(t))((v'v(s,-)(v'v(s,-)J(s, -)))(t), V(s, t))) +0
<:J -(g(R(J, v)v, v))(r(s, t)) = -(R(J, v, v, v))(r(s, t)) = o.
To obtain the last equality, we used the skew-symmetry of R.
For vanishing initial conditions g(I's(to))(J(s, to), V(s, to))= 0, and
:;g(rs(· ))(J(s, ·), V(s, -))) (to)= g(r s (to))((v' V(s,-)J(s, ·) )(to), V(s, to))= 0,
the ordinary differential equation
d2
dt2 (g(r s (t))(J(s, t), V(s, t))) = 0 (t E I)
has the unique solution given by g(I's(t))(J(s,t), V(s,t)) = 0 for all t EI.
Solutions 437

Let U:=span{v} c TpM. Then -g(p)lu is positive definite on U, that is,
U E Ng(p)· By Theorem 5.3, g(p)lu_j_ is positive definite. Let e1, · · · , em-1
be an orthonormal basis for u1-. Then (e 1?, ••• , (em_i)i, is the dual basis
for (U1_)* since we have (ei)i,ej=g(p)(ei,ej)=8ij· Hence
traceFv = e~(R(p)(ei,v)v) (!l (C}R)(p)(v,v) = Ric(p)(v,v)
g(p)(v,v) g(p)(v,v) g(p)(v,v) ·
Justification of (*): Let (U, x) be any admissible chart containing p. We
can write (ei)P = aij (dxi )p for some numbers aij, and et= b1 Oxk ,P for some
numbers ll;. Then 8u=(ei)Pet=aiib18l=aij~- So bfaij=8j. Thus
I, k .
(ei) (R(p)(ei, v)v) = aiibi (dx 1 )p(R(p)(oxk,p, v)v)
k .
= 8j (dx 1 )p(R(p)(oxk,p, v)v)
= (dxk)p(R(p)(oxk,p,v)v)
= R(p)((dxk)p, Oxk,p, v, v)
= (C~R)(p)(v,v).
(From Example 9.2, for the instantaneous observer v = Ot,p E TpM, where
M is the Schwarzschild spacetime, we have
2m m m
traceFv = - - - - - = 0.
r3 r3 r3
So Ric(p)(v,v) =0 for all pin the chart JR x (2m,oo) x U, and the time-
like vector v = Ot,p E TpM, i.e., Rieu = 0. In fact, Ric = 0 for the
Schwarzschild spacetime, that is, it is 'Ricci-flat'.)

We 'normalize' Or,p to get a unit length vector,
e •-
orp
'
or,p
r,p .- y1g(p)(or,p, or,p)
If the astronaut's height is h, then the astronaut is represented by the vector
her,p· The acceleration experienced by the astronaut at r = 2m+8, 8 > 0, is
h (
Fv(her,p)= ,Jgrr(P)FvOr,p= ,Jgrr(p) r3
h 2m)I r=2m+o0r,p=h(2m+8)3er,p·
2m
The critical acceleration the astronaut can withstand is h(lOos- 2 ). Hence
for survival,
2 2m
h(IOos- ) > h ( 2m+<5) 3 for all 8 > 0.
_2 2m 1 . m 1 1 1 1
So 100s ~ (2m)3 = 4m2, i.e., M0 ~ 2 IOs-1 5 x IQ-6s = IQ-4 = 10000.
Consequently, the mass of the black hole should be at least ten thousand
solar masses.

We have df=(oxf)dx+(oyf)dy+(ozf)dz, and so iff/,=d/, then
nx=oxf
ny=oyf
nz=ozf.
By an application of the Schwarz theorem,
oynx = oyoxf = oxoyf = oxny,
ozny = ozoyf = oyoz/ = oynz,
oxnz = oxozf = ozoxf = oznx.
This shows the 'only if' part.
Now suppose
oynx = oxny,
ozny = oynz,
oxnz = oznx.
These are exactly the consistency conditions for the system
oxf = nx
(*) { oyf = ny
ozf = nz,
and so by the Frobenius theorem, there exists a smooth solution f satisfying
(*). But then
df = (oxf)dx + (oyf)dy + (ozf)dz = nxdx + nydy + nzdz = n,
proving the 'if' part.
Solutions 439
Chapter 10

Let wi:=v.,,.(i), l~i~k. For all v1,··· ,vk EV, we have
(1r(ar))(v1,··· ,vk) = (ar)(v.,,.(1),··· ,V1r(k)) = (ar)(w1,··· ,wk)
= T(Wu(l), ... , Wu(k)) = T(V1r(u(l)), ... , V1r(u(k)))
= T(V(1rou)(l),··· ,V(1rou)(k)) = ((1roa)r)(v1,··· ,Vk)-
Hence 1r(ar) = (1r o a)r.

We have
(Altr)(x, y, z)
1
= 3 ! (r(x, y, z)-r(x, z, y)-r(y, x, z)+r(y, z, x)-r(z, y, x)+r(z, x, y)).

For all v1, · · · , Vk, we have that
(Altw)(v1,··· ,vk) = t, ~
. 1res.
(signa)w(vu(l),··· ,Vu(k))
= t, ~
"1res.
(signa)(signa)·w(v1,··· ,vk) = t, ~
"1res.
lw(v1,··· ,vk)
(since signa E {-1, 1}, giving (signa) 2 = 1). As Sk has k! elements, we

have for all v1, · · · , Vk,
1
(Altw)(v1, · · · , vk) = k! k! w(v1, · · · , vk) = w(v1, · · · , vk)-
Thus Altw = w.

We have
(€ 1A·••/\€n)(v1,··· ,vn) =n!Alt(€ 1®···®€n)(v1,··· ,vn)
= ~ (sign1r)(€ 1@···Q9€n)(v.,,.(1), 000

,V1r(n))
1rESn
= ~ (sign1r)€ 1(v.,,.(1)) · · · €n(v.,,.(n)) = det[i(vj)].

1rESn

For all v1, · · · , Vk EV, we have
(€7r(l)/\ ··· /\€7r(k))(v1,··· ,vk)
= det[€7r(i)(vj)] = ~ (signa)(€7r(l)Vo-(l)) ... (€7r(k)Vu(k))
= (sigll7r) ~ (signr)(€ 1Vr(1)) · · · (€kVr(k))

-res.
= (sign 7r) det[ ivj] = (sign 7r)( €1/\ · · · /\ Ek)( v1, · · · , vk)-
To get the equality in the third line above, we used the 'change of vari-
able' a= To 7r: as a runs over all elements of Sk, so does T := a o 7!"- 1.
Also, for the equality in the second to last line, we note that the prod-
uct (€7r(l)vr(7r(l))) · · · (€7r(k)vr(7r(k))) contains each of the factors EiVr(i),
i E {1, • • • , k} exactly once (because 7r E Sk)-
An alternative argument can be given using the following fact: Every
permutation in Sk can be written as a product of adjacent transpositions,
namely permutations that swap only adjacent elements j, j + 1, 1 ~ j < k;
see [Halmos(1987), Exercise 5, §27]. Using this, the claim follows from
Proposition 10.2.

For all Vi ... 'vk E TJ M, and all 7l" E sk, we have
(7r(.Cv!1))(Vi, · · ·, Vk)
= (.Cvn)(V,r(l), ... 'V,r(k))

k
= V(!1(V,r(l), · · ·, V,r(k) ))- ~ n(V,r(l), · · · , V,r(i-1), .Cv V,r(i), V,r(i+l), · · · , V,r(k))
i=l
k
= V((7r!1)(Vi, ···,Vi))-~ (7r!1)(Vi, · · ·, ¼-1,.Cv¼, ¼+1, ···,Vi)
i=l
k
= V((sign 7r)!1(Vi, ···,Vi))-~ (sign 7r)!1(Vi, · · · , ¼-1, .Cv ¼, ¼+1, · · · , Vk)
i=l
= (sign7r)(.Cv!1)(Vi, · · · , Vi).
Thus 7r(.Cv0) = (sign7r) (.CvO) for all 7r E sk, and so .Cvn E nk M.
Solutions 441

We have Alt (JT) = f (Alt T), because for all Vi, · · · , Vk E TJ- M,
(Alt (JT))(Vi, · · · , Vk) = :, I; (sigll1r) (JT)(V,.,.(1), · · · , V,.,.(k))
. 1res.
= :, I:
. 1res.
(sigll1r)/T(V,.,.(1), · · · , V,.,.(k))
= f((AltT)(V1,··· ,Vk))=(!AltT)(Vi,··· ,Vk)-

We give an inductive argument on the number of factors in the wedge
product. For n = l there is nothing to prove since Alti1 1 = n1, and for
n = 2, this is just the definition of the wedge product. Suppose that we
have an n > 2, and that the result has been shown when the number of
factors is ~ n-1. Then
n 1 /\ n 2 /\ ... /\ nn = (k1 + (k2 +···+kn))! Alt (!11® (n2 /\ .. ·Ann))
k1!(k2+ ···+kn)!
1
= (k + k2 + · · · + kn)! (k2 + · · · +kn))! Alt (!11® Alt (!12® ... ® nn))
k1! (k2+ ···+kn)! k2! ···kn!
= (k1 +k2+ ···+kn)! Alt (!11® (!12® ... ® nn))
k1!k2! ···kn!
= (k1 +···+kn)! Alt (!11® ... ® nn).
k1!••·kn!
As k is odd, k 2 is odd too, and so (-1 )k 2 = -1. Proposition 10.2 implies
n/\n = (-1)k 2 n/\n = -n/\n, and so 2(i1Ai1) = 0, i.e., n/\n = 0.

We have
n/\e = (xdx+ydy+zdz)A(ydx+zdy+xdz)
= (xydxAdx+xzdxAdy+x 2 dxAdz
+y2 dy I\ dx+yzdy Ady+yxdy Adz
+zydz I\ dx+z 2 dz Ady+zxdz I\ dz)
= (0+xzdxAdy- x 2 dzAdx-y 2 dxAdy+0
+~~/\~+~~/\~-~~/\~+~
= (zx -y2 )dxAdy+(xy- z 2 )dyAdz+(yz - x 2 )dzAdx.

We have
Let the term 83~.-(1)

1 ••• r53
i_k
.-(k)
#0. Then (ii,··· ,ik)=(j1r(l),··· ,i1r(k))- Also,
as i1 < · · · <ik, we have i1r(l) < · · · <i1r(k) · Since j1 < · · · < jk, we conclude
that 1r is the identity map (trivial permutation). Thus I= J. So
f if
{o J = I,
ni1···ik = if J # I.

For all p EM and all V1, · · · , Vk+e, we have
((/*(0A8))(Vi, · · · , Vk+e))(p)
= (0 A 8)(J(p))(dfp(V1)p, · · · , dfp(Vk+e)p)
= k~f! I; (sigll1r) (0 ® 8)(J(p))(dfp(V1r(i))p, · · · , dfp(V1r(k+e))p)

1reSk+t
= k~fl I; (sigll1r)O(J(p))(d/p(V1r(i))p,··· ,dfp(V1r(k))p) ·

· ·-,,.es•+' 8(J(p))(dfp(V1r(k+1))p, · · · ,dfp(V1r(k+e))p)
= k~e, I: (sigll7r)((J*O)(V1r(l), ... 'V1r(k)))(p).
· ·-,,.es•+' ((J*8)(V1r(k+1), · · · , V1r(k+e)))(p)
= k~e! I; (sigll7r)(((J*O) ® (J*8))(V1r(l), · · · , V1r(k+e))}(p)
1reSk+t
= (((/*0) I\ (/*8))(V1r(l), · · · , V1r(k+e))}(p).
Hence J*(O I\ 8) = (/*0) /\ (!*8).

Solutions 443

We have, using Exercises 10.12 and 3.18 (p.53), that
f*(dxAdyAdz) = (J*dx) I\ (f*dy)A (J*dz)=d(x o J) Ad(y o J) Ad(z o J)
= d(r(sin 0)( cos cp )) A d(r(sin 0)(sin cp )) A d(r cos 0).
Since the gradient of a function in a chart can be computed by letting the
coordinate vector fields act on the function, we obtain
d(r(sin0)(coscp)) = (sin0)(coscp)dr + r(cos0)(coscp)d0 - r(sin0)(sincp)dcp,
d(r(sin0)(sincp)) = (sin0)(sincp)dr + r(cos0)(sincp)d0 + r(sin0)(coscp)dcp,
d(rcos0) = (cos0)dr - r(sin0)d0.
So, using dr I\ dr = 0, dr I\ d0 = -d0 I\ dr, etc., we get, after some algebraic
manipulations, that
d(r(sin 0)(sin cp )) A d(r cos 0)
= -r(sin cp )dr I\ d0-r(sin 0)( cos 0)( cos cp)dr I\ dcp+r 2 (sin 0) 2 ( cos cp )d0 I\ dcp,
and so
d(r(sin 0)(cos cp )) /\ d(r(sin 0)(sin cp)) /\ d(r cos 0)
= (r 2 (sin 0) 3 ( cos ¢) 2 +r 2 (sin 0)(cos 0) 2 (cos cp )2 +r 2 (sin 0)(sin cp) 2 )dr I\ d0 I\ dcp
= r 2 (sin0)dr Ad0 Adcp.
Thus f*(dxAdyAdz) = r 2 (sin0)drAd0Adcp.

Let 7r E sk be such that i,r(l) < ... < i,r(k)· Then
n = f dxi 1 I\ • • • I\ dxik = /(sign 1r) dxi"(l) I\ • • • I\ dxi"(k).
By Exercise 10.11, if 1 ,:;;; j 1 < • • • < Jk ,,;; m, then
nj1•••jk = { /(sign1r) ~f (!1, ... ,!k) = (~1r(l), ... ,~1r(k)),
0 1f (J1, · · · ,Jk) =/= (i1r(l), · · · ,Z1r(k))-
So by the definition of the exterior derivative,
= Ox;(J(sign1r))dxi I\ dxi"< 1 J I\••• I\ dxi"(kJ

= (ox;J)dxi I\ ((sign1r)dxi"(l) /\ ... /\ dxi"(k)}
= (Oxi J) dxi I\ dxi 1 I\ · · · I\ dxik.

We have
dO = (oxx)dxAdx + (oyx)dyAdx + (ozx)dzAdx
+ (oxy) dx I\ dy + (oyy) dy Ady+ (ozy) dz I\ dy
+(oxz)dxAdz + (oyz)dyAdz + (ozz)dzAdz
= 0,
since all summands vanish. Similarly,
de= (oxy)dxAdx + (oyy)dyAdx + (ozy)dzAdx
+ (oxz) dx I\ dy + (oyz) dy Ady+ (ozz) dz I\ dy
+(oxx)dxAdz + (oyx)dyAdz + (ozx)dzAdz
= 1 dy I\ dx + l dz I\ dy + l dx I\ dz = -( dx I\ dy + dy I\ dz + dz I\ dx).

In the global admissible chart (R2 , (x,y)....+(x,y)), write 0=0.xdx+OYdy.
So dO=oynxdyAdx+oxOYdxAdy=(oxOY- oynx)dxAdy. As n is closed,
oOY oox
ox =a:;;·
Set
f(x,y)= r nx(~,y)d~+ f O,Y(o,,,,)d,,,.
Then f E C 00 (R 2 ), and oxf =0.x+O=Ox. Also,

rx 00.x rx 00,Y
oyf= Jo By(~,y)d~+OY(O,y)= Jo
ox (~,y)~+OY(O,y)
= O,Y(x, y) - O,Y(O, y) + O,Y(O, y) = O,Y(x, y).

Thus 0=0.Xdx+OYdy=oxf dx+oyf dy=df. So O is exact.

The additivity is clear. Let k ~ 2. For all n E n,k M, f E C 00 (M), and
W1, ···,Wk-IE TJM, we have
(iv(JO))(W1, · · · , Wk-1) = (f O)(V, W1, · · · , Wk-1)
= f ·O(V, W1, · · · , Wk-1)
= f ·(ivO)(W1, · · ·, Wk-1)
= (J(ivO))(W1, · · ·, Wk-1).
Thus iv(JO)=f(ivO). For k=l, iv(/0)=(!0.)(V)=J(OV) = f(ivO).
Solutions 445

By Cartan's formula applied to dO in place of n,
.Cv(dO) = d(iv(dO)) + iv(d(dO)) = d(iv(dO)) + ivO
= d(iv(dO)) + 0 = d(iv(dO)).
Applying d to both sides ofCartan's formula .CvO=d(ivO)+iv(dO) yields

d(.CvO) = d(d(ivO) + iv(dO)) = d(d(ivO)) + d(iv(dO))
= 0 + d(iv(dO)) = d(iv(dO)).
Thus .Cv(dO) = d(iv(dO)) = d(.CvO).
Chapter 11

We have
WB(e1, · · · , em)= (t::1 I\•·• I\ €m)(e1, · · · , em) =det[i(ej )] = det[8j] = 1 > 0.
So WB =I= 0, and (e1, · · · , em) is positively oriented with respect to [wB]. We
note that
WB(e'1, · · · , e'm ) = (€ 1 A··· /\€m)(a'1.1 e·i1, · · · , ai"'e·
m i1n )
= a i11 · ·· a i,n ( 1
m € /\ • · • /\ € m) (e'l.1,
· · )
• • • , e Z,n
= ~ a;(l)···a~m\€ 1 /\ ·•• /\€m)(e.,,.(1),··· ,e1r(m))

1rE81n
= ~a;(l) • • • a~ml(sigll7r)(€ 1 /\ · · · /\€m)(e1, · · · , em)
= (det[aj])·l=det[a;].
We note that the third equality can be justified by noting that a sum-
mand will be nonzero only when (i 1, • • • , im) is a permutation of (1, • • • , m),
thanks to the skew-symmetry of w B = €1/\ • • •/\ €m.
(1)=?(2): If Bis positively oriented with respect to WB, then we have that
WB(e~,··•,e~)>0. From the above, det[a;]=wB(e~,··•,e~)>0.
(2)=?(3): If det[a;] > 0, then WB(e~, •••, e~) >0. But WB,(e~, •••, e~) > 0.
The k relating these nonzero top forms, WB' = kwB, is easily seen
to be >0 by operating on them-tuple (e~, •••, e~)- So WB ~wk.
(3)=?(1): If WB~wk, then we have WB=kwB, for a k>0. Hence we obtain
that WB(e~, · · · , e~) = kwB,(e~, · · · , e~) = kl >0. Consequently,
(e~, •••, e~) is positively oriented with respect to [wB]-

The transition maps cp8 o cp~ 1, cpno cp;;- 1 : R 2 \{(0, 0)}-+R2 \{(0, 0)} are both
given by R 2 \{(0,0)} 3 (u,v) >-+ ~u,v) 2 , and so the Jacobian matrix is
u +v
J(u, v) = [ _ (::,::~:,
(u2+v2)2
1
Thus det J(u, v) = ( 2 2) 2 < 0. So A is not oriented.
u +v
Solutions 447
In order to construct an oriented atlas .A, we keep in .A the old chart

(Us, 'Ps), and we replace the old chart (Un, 'Pn) by a new chart (Un, 'Pn),
where the new chart map 'Pn : Un - JR 2 \{(0, 0)} is defined by 'Pn =Ro 'Pn,
R being the reflection in the u-axis, i.e., R(u, v) = (u, -v). The chart
transition map 'Pn ocp;; 1 =Ro 'Pn ocp;; 1 has the Jacobian matrix R-J(u,v),
and its determinant is
(-1) 1
(detR)(detJ)=(-1)· ( 2 2 ) 2 ( 2 2)2 > o.
u +v u +v
Similarly, the chart transition map <p8 o(cpn)- 1 =<p8 o(Ro<pn)- 1 =<p8 ocp-;;_ 1 oR
has the Jacobian matrix J(u, v)-R, and so its determinant is again positive
on JR2 \{(0,0)}. Thus A is oriented.
To show that .Au {(U, cp)} is oriented, we note that U nUn = U nUs = U,
and so it is enough to show that the determinant of the Jacobian of 'Pn o <p- 1
is positive. (It will then follow that also the inverse <po cp-;;_ 1 , and the other
transition maps <po cp;; 1 = <po 'Pn o (<p8 o 'Pn)- 1 , _ (sin0)(sinef>)
[ 1-cos0 1-cos0
-sin¢> (sin0)(cosef>)'
1-cos0 1-cos0
having the determinant
-sin0
(1-cos0) 2 < O.
Consequently, the determinant of the Jacobian of 'Pn o <p- 1 is
(- sin 0) sin 0 0
(detR)· (1-cos0) 2 (1-cos0) 2 > ·
Hence Au {(U, <p)} is oriented.

00-- 00
Clearly U Vn+l \Vn c M. Let p E M. As M = U Vn, there exists a
n=O n=O
first/smallest m E N such that p E Vm.
- - - 00--
1° m = 1. Then p E Vic V1 = Vi \0 =Vi\Vo c U Vn+l \ Vn.
n=O
2° m > 1. Then p ¢ Vm-1 by the definition of m.
- 00--
Thus pE Vm \Vm-1CVm \Vm-1C LJ Vn+l \Vn.
n=O

We know that A u A' belongs to the smooth structure of M (as they are
compatible). Suppose that (U, x) E A and (V, y) E A'. We want to show
that the determinant of the Jacobian matrix of the chart transition map is
everywhere positive on U n V. In U n V, we have
nvdy 1 /\ •.• Adym=n=nudx 1 /\ •.• Adxm=nudet[oyiXi]dy1 /\ ... Adym.
So it follows that det[oyiXi] nv / nu =
> 0. Then the inverse of this =;
chart transition map also has an everywhere positive determinant for its
Jacobian matrix.

In Uo,(i), we first decompose n = nk dx 1 A • • • A dxk A • • • A dxm. So we have
'Pin = 'Pink dx 1 A • • • A dxk A • • • A dxm. Set !k := (-l)k-l'Pi nk, 1 :e::;; k ,s; m. We
obtain the decomposition 'Pin= (-1 )k-l fk dx 1 A • • • A dxk A • • • A dxm. Thus
d(cpin) = d((-1t- 1 fkdx 1 I \ · · · I\ dxk I \ · · · I\ dxm)
= (-1t- 1 (ox•fk)dxk I\ dx 1 I \ · · · I\ dxk I \ · · · I\ dxm
= (ox•fk)dx 1 I\ ... I\ dxm.
Since 9k := /k o x-;;:fi) has a compact support contained in Xo,(i)Uo,(i), by
extending it as being identically zero outside suppgk, we have
J,Xa(i)Ua(i) (ox•fk)(x:(i)u)du 1 1
···du
m
=
i o(fkox:hi)
0 k
Xa(i)Ua(i) U
1
(u)du ···du
m
=i 09 ~(u)du 1 ···dum=
Xa(i)Ua(i) OU
f
JR= OU
09~(u)du 1 ---dum
= f f og~ (u) dukdu 1 ... du1< ... dum (Fubini)

JR=-1 JR OU
= f 9k(u)lu'=oo du 1 · · · du1< · · · dum = f Odu 1 · · · du1< · · · dum = 0.
JRm-1 uk=-00 JR1n-l
As n iscompactly supported, and since {supp 'Pi : i E J} is locally finite,
I;i 'Pin is a finite sum. Hence I;i d( 'Pi n) is also a finite sum. We also note
that d(cpin), just like 'Pin, has compact support contained in Uo,(i)· Thus
IM dn = IM d(~cpin) = IM~d(cpin) = ~IM d(cpin) <!l ~La(i)d(cpin)

=Li (ox 1 fi + ... + Ox=fm)(x:hi u) du 1 ... dum
i Xa(i)Ua(i)
=LI: i i k=l Xa(i)Ua(i)

(ox•fk)(x:ti) u) du 1 · · · dum =LI: =i k=l
0 0.
Solutions 449
To see (*), we note that the support of cpjd(cpiD.) c (suppcpi) n (suppcpj)

and that I;i cpjd(cpiD.) is a finite sum, so that
f M
d(cpin) = L J,
j Ua(j)
cpjd(cpin) = L J,
j Ua(j)f"'>Ua(i)
cpjd(cpin)
= L J, cpjd( 'Pin) = J, L cpjd('Pin) = J, d( 'Pin).

j Ua(i) Ua(i) j Ua(i)

S 2 is compact. Let {0n, 0s} be partition of unity subordinate to {Un, U8 }.
Then
VolgS 2 = J,Un 0nvolg + J, Us

0s volg
= f_ 0n(cp;;: 1 (u,v))Vldet[g~ncp;:;- 1 (u,v)]ldudv

JipnUn
+ f 0s(cp; 1 (u,v))Vldet[g~•cp;;- 1 (u,v)]ldudv

JrpsUs
= f_ 0n(0;;: 1 (u,v))Vldet[g~ncp;:;- 1 (u,v)]ldudv

JipnU
+ f 0s(cp;;- 1 (u, v))VI det[g~•cp;;- 1 (u, v)] Idudv,

Jcp.U
since the contributions of the thin sets 0n(Un\U) and cp8 (Us\U) are 0. Here
(U, cp) is the spherical coordinate chart from Example 1.8 (p.7).
We now use a change of variables given by it>= IPn o cp- 1 in the first
integral, and = 'Ps o cp- 1 in the second integral. Using the transformation
relations among gg and g~n, respectively g~•,
as expected.

The chart (U,x) = (1Rx(2m,oo)xU, p = (t,r,p) ....... (r,0(p),rp(p),t))
l
belongs to A*. It is enough to show that n coincides with volg on (U, x)
(since then by continuity they coincide also on M). In (U, x), we have
(1-2;1)-l
[g(ox;, Oxi )] = [ r2 r2(sin0)2 .
-(1-2;')
So det[g(ox;,Oxi)] = -r4 (sin0) 2 , and y'ldet[g(ox;,ilxi)]I = r 2 sin0. Thus
in the chart (U, x), we have
volg = r 2 (sin0)dr A d0 I\ dv=g(v,ek)=g(Eivei,ek)=(Eiv)g(ei,ek)=1,(k)Ekv (no
sum over k), and so (ek)l>=1,(k)Ek. Hence
vol 9 = (-1) 19 (ed /\ · · · /\ (em)I> = (-lY 9 1,(l) • • • 1,(m)€1 /\ • • • /\ €m
= ( -1 )19 ( -1 y9 €1 /\ ... /\ €m = €1 /\ ... /\ €m.

Let (e 1, • • • , em) be a positively oriented ordered basis such that it is or-
thonormal with respect to g, and let (€1, • • • , Em) be the corresponding dual
basis for V*. Then wi = w~l- and Vk = viei for some w;, v;, 1 ~ i,j ~ m.
d =wi Vk =WeVk€
vve h ave uk
ur i i eei =WeVkui
i i ~e =wivk.
i i Thus the mat nces
· [wii] and
[v;] are inverses of each other. Hence
Vldet[g(vi,vi)]I w1A··· /\Wm
ldet[g(vfer,vJes)]I (wfi€i 1 )A ··· A(W4,€i"')
Idet([vf )t [g(er, es)] [vJ])I I: w!c1r · · w:Cm) €1r(l) /\ · · · /\ €1r(m)

1rES1n
,----------
(det[vf])2 Idet[g(er, es)] I I: w!c1r · · w:Cm) (sigll1r)€1 /\ · · · /\ €m
1rES,n
= (det[vj])(det[wj])vol 9 = (det J)vol 9 = (l)vol 9 = vol 9.

Solutions 451
Note that in the above, we used det[v;] > 0, which follows from the fact
that if [w] is the orientation on V, then w(e 1 , • • • , em) > 0, and also
1rES-m

This is similar to the proof of Lemma 10.4 (p.231), and we will use Exer-
cise 10.4. For v1 , · · · , vk-l EV, we have
= det [
w~v w~v 1 · · · w 1 ~k-l
.. ..
wkv wkv1 · · · WkVk-1
..
l .
We expand the determinant down the first column. Thus we have
(iv(W 1I\ · · · /\W 1))(v1, · · · ,Vk-1)

k
= ~(-1?- 1 (wrv)(w 1 I\ ··· l\w1'I\ ·•• I\Wk)(v1,··· ,Vk-1).
r=l
k
Consequently, iv(w 1 I\··· I\W 1 ) = ~ (-1r- 1 (wrv)w 1 /\ · · · Aw"/\ ···/\Wk.
r=l

Let {e 1 , • • • , em} be a basis for V, and {f1, · • • , fm} be the dual basis for
V*. Decompose w =W1f1 , where I= (i 1 , • • • , ik) and f 1 := fi 1 /\ • • • /\fik, and
similarly 0 = 0 JfJ. Then
iv(w I\ 0) = w10Jiv(fi 1 I\··· I\ fik I\ fj 1 /\•••/\fit)
k -
= WJ0J I:(-1r-lfir(v)fi1 /\ ... /\fir/\ ... /\ fik I\ fJ
r=l
f -
+w10J I:(-lt+r-lfjr(v)fl /\ fj1 /\ ... /\ fjr /\ ... /\ fjl
r=l
= (ivW)/\0 + (-llwA(iv0).

This follows immediately from Lemma 11.6 by induction:
(-lt(m-l)(*W)I\V~I\ ··· /\Vt
= (-l)(k-l)(m-l)(*(iv 1W))/\V~/\ ··· /\Vt
= (-l)(k- 2)(m-l)(*(iv2 (iv 1w)))AV~/\ ... /\Vt
= *(ivJ · · (iv 1 w) · · ·)) = *(w(v1, · · · , Vk))

= w(v1, · · · , vk)( *1) = w(v1, · · · , vk)((-1) 19 vol 9 ).

For n, e E nk Mand/ E C 00 (M), we have for all p EM that
(*(n + Je))(p) = *((n + Je)(p)) = *(n(p)+ J(p)(e(p)))
= *(n(p))+ J(p) * (e(p)) = (*n)(p)+ J(p)(*e)(p)
= (*n)(p)+U(*e))(p) = (*n + f(*e))(p).
Thus *(n + /0) =*!1 + /(*0). So*: f!k M-+ nm-k Mis C 00 (M)-linear.
For n E nk M, and for p EM, we have
(* * i1)(p) = *((*!1)(p)) = *(*(!1(p))) = (-lt(m-l)+i.,,,n(p)
= (-lt(m-l)+i(g)n(p) = ((-ll(m-l)+i(g)n)(p).
Consequently,** n = (-1t(m-l)+i(g)n.

Let (U, x) be an admissible chart for M containing p. Suppose that
(e1, • • • , em) is a positively oriented orthonormal basis for TpM. Then
volg(p)(e1, · · · , em)> 0 and also volg(p)(e1, · · · , em)= 1. It is enough to
show that the action of volg(p) and volg(p) have the same absolute value
on a basis. Let Dxi,p=a;ei. Then
gij(P) = g(p)(ox;,p,Oxi,p) = g(p)(atek,a1et) = ata1g(p)(ek,e£).
As det[g(p)(ek, et)]= (-1) 1a<,i, from above, det[gij(p)] = (det[a;]) 2 (-1) 1a<,i.
Hence
I det[a;]I = VI det[gij]I = VI det[gij(p)]I det[(dxi)p(Oxi,p)]
= ✓1 det[gij(P)]l((dx 1)p /\ · · · /\ (dxm)p)(oxi,p, ... , Oxm,p)

= volg(p)(ox1,p, · · · , ox"',v)-
Solutions 453
On the other hand, we have

volg(p)(0xi,p, · · · , Ox=,p)
= (-lie<•> ((ed I \ · · · I\ (emY')(ai1 eiu · · · , a~ei,,.)
= (-lic<•>ai1 ···a~((ed /\ ··· A(eml)(ei 1 , · •• ,ei,J
= (-lie<•> ~ a;(l) · · · a~m)((ed /\ · · · /\ (eml)(e.,,.(1), · · · , e.,,.(m))
1rESrn
= (-lie<•> ~ a;(l) · · · a~m)(sign71")((ed /\ · · · /\ (eml)(e1, · · · , em)

1rESrn
=(-lie<•>~ a;(l) • • -a~m)(sign71")det[(ed'ej] = det[a;].

1rESrn

Every n E 0 3 M can be decomposed as
n = WxdyAdzAdt + WydxAdZ/\dt + WzdxAdyAdt + WtdXAdyAdz
for some Wx,Wy,Wz,Wt E C 00 (M). Then
and so
**n = -Wx * (dx) + Wy * (dy) - Wz * (dz) - Wt* (dt)
= -wx(-dyAdzAdt) +wy(dxAdzAdt)
-wz(-dxAdyAdt)-wt(-dxAdyAdz)
= WxdyAdzAdt + wydxAdzAdt + wzdxAdyAdt + WtdXAdyAdz
= n.
With k=3, m=4, and i(g)=l, we have (-1?(m-l)+L(g)=(-1)1°=L
Chapter 12

As g(p)(u, v) = v-g(p)(v,v) g(p)(w,v) - g(p)(v,v) =0, u E v1-.
g(p)(w,v) v-g(p)(v,v)
. _ _ v-g(p)(v,v) v _
If. Let v-cw for a c>0. Then u- - (p)( ) cv - ----;:::====;:::::;::;=====;:-0.
g cv,v -J-g(p)(v,v)
Only if: If u=0, then
_v-g(p)(v,v)W_ V =0 (17.14)
g(p)(w,v) v-g(p)(v,v) ·
·
R earrangmg, v=cw, wh ere c := g(p)(v,v)
( )( ) . Clear1y c# 0 .
g p w,v
c> 0: Taking scalar product in (17.14) with the time-orientation V at p,
g(p)(v, v)
g(p)(v, Vp) = g(p)(w, v) g(p)(w, Vv),
and note that

g(p) (v, Vp) < 0 (v future-pointing),
g(p) (w, Vp) < 0 (w future-pointing),
g(p)(v,v)<0 (v timelike).
These facts show that g(p)(w,v)<0 too, and soc= gt?t'v~ ( =

g p w,v
=) >0.

We have
Thus
.
1.e., lul
1/31-- ---;::::=;::::::::;;.:. so -2£ -- ~1 -1 - ✓ - Iu 12 .
yl - lul 2 2L 1+132
Solutions 455

~1 ul-(3 ~2 u2~ ~3 u3~
We have u =a= l-ul(3' u =0= l-(3ul , u =0= l-(3ul . Hence
. u 1-(3 . 1 a+/3 . So
u 2 =0, u 3 =0. Solvmg for u 1 from a = - 1 /3, yields u = - -
13
1-u l+a
. a+/3
u = u'ei = - - e1 = (a® /3)e1.
l+a13
Setting v = 0 and u = a in the formula from Exercise 1.2 (p.4) yields
B= ( ~ a2-1)
a 2 +1' a 2 +1 ·
To get the coordinates of B', we note that the reflection B of B' in the
. . ~ 2/3 /3 2 - 1 2/3 1 - /3 2
x-axis has the coordmates B = ( 132 + 1 , 132 + 1 ), and so B = ( 132 + 1 , 132 + 1 ). So
the equation of BB' is
a 2 -1 1- 132 a2 - 1
y-~ -----
132 + 1 a2 + 1
2a 2/3 2a
x---
a2 +1 132 + 1 a2 +1
To find the x-coordinate of the intersection point of BB' with the x-axis,
we set y = 0, and solve for x, which after some algebraic manipulations,
yields
a+/3
X= - - =a@/3.
l+a/3

As a E (0, 1), and since tanh: (0, oo)-+ (0, 1) is strictly increasing and onto,
there is a unique a E (0, oo) such that tanh a= a. Similarly, let b E (0, oo)
be such that tanh b = f3. Then
sinha sinhb
a ® f3 = a+ f3 = tanh a+ tanh b ~+coshb
1 +a/3 1 + (tanh a)(tanh b) sinha sinhb
1+----
cosha coshb
= (sinha)(coshb)+(sinhb)cosha = sinh(a+b) = tanh(a+b).
(cosh a)(cosh b) + (sinh a) sinh b cosh(a+b)

Let b E (0, oo) be such that tanh b = (3. Let Uk be the relative speed of the
kth cart with respect to the table. Then
u2 = f3 ® f3 = tanh(b+b) = tanh(2b) ,
u 3 = /3®tanh(2b) = tanh(b+2b) = tanh(3b),
uk = /3®tanh((k-l)b) = tanh(b+(k-l)b) = tanh(kb).

Thus Un= tanh(nb) = tanh(ntanh- 1 /3). As b > 0,
lim un = lim tanh(ntanh- 1 /3) = lim tanh(nb) = lim sin~r:~

n--+oo n---+oo n---+oo n--+oo COS n
enb _ e-nb 1- e- nb2 1- 0
lim ----,---...,... lim -----=-..,. = - - = l.
n-+oo enb + e-nb n-+oo 1 + e- 2 nb 1+0

Let the annihilation occur at event p E M. At p, let the velocity of the elec-
tron be v_ E TpM, and that of the positron be V+ E TpM. Let the velocity
of the photon be w E TpM. Then g(p)(v_,v_) = -1 = g(p)(v+ ,v+) and
g(p)(w, w) = 0.
/
Solutions 457
By the law of conservation of energy-momentum, me(v_)IJ + me(v+ )1, =wi,.

So me(v_ + v+) =w. Hence
0 = g(p)(w,w) = g(p)(me(v_ +v+),me(v_ +v+))
= m~(-1-1 + 2g(p)(v-,v+)),
and so g(p)(v_, v+) = 1 > 0. Since v_, v+ are timelike and future-pointing,
we have that g(p)(v_, Vp) <0 and g(p)(v+, Vp) <0. Exercise 5.4 implies that
also g(p)(v-,v+)<0, which is a contradiction to g(p)(v-,v+) = 1 > 0.

Let {e1,e2,e3} be an orthonormal basis for v1_ such that (e1,e2,e3,v) is a
positively oriented basis for TpM with respect to the orientation [O(p)].
Then (e1, e2, e3) is a positively oriented basis for v1_ because
wv(e1, e2, e3) = O(p)(e1, e2, e3, v) > 0.
Let v E ;\3(v1-)* denote the volume form on v1_ induced by the orientation
[wv] and g(p)lvL Then we have that Wv=cv, where c=wv(e1,e2,e3), and
O(p) = Cvolg(p), where C = O(p)(e1, e2, e3, v) =wv(e1, e2, e3) = c. Moreover,
F(p)(x,y)=(*(Bi,))(x,y)=*(Bi, /\Xi, Ayi,)=*(kv)=k, where k is such that
Bi, /\Xi, Ayi,=kv, that is, k=(Bi, /\Xi, AY1,)(e1,e2,e3). Thus
F(p)(x,y) = (Bi, /\Xi, Al)(e1,e2,e3)
g(p)(B, e1) g(p)(B, e2) g(p)(B, ea)] [g(p)(B, e1) g(p)(x, e1) g(p)(y, e1)]
= det [ g(p)(x, e1) g(p)(x, e2) g(p)(x, ea) = det g(p)(B, e2) g(p)(x, e2) g(p)(y, e2)
g(p)(y, e1) g(p)(y, e2) g(p)(y, ea) g(p)(B, ea) g(p)(x, ea) g(p)(y, ea)
i, i, i, 1 1
=(e 1 Ae 2 Ae 3)(B,x,y) = v(B,x,y) = ~wv(B,x,y) = 0 wv(B,x,y)
1
= 0 n(p)(B,x,y,v) = volg(p)(B,x,y,v) = volg(p)(x,y,B,v).
Claim: If BE v1_ satisfies for all x, y E v1_, volg(p) (x, y, B, v) = F(p)(x, y),
then B=B.
Suppose that w := B - B =I 0. Set Ji= aw, where a= 1/y'g(p)(w, w).
Then we can find vectors h, h E v1_ so that {/1, h, h} is an orthonormal
basis for v1_. As (Ji, h, h, v) is an orthonormal basis for TpM, we have
that volg(p)(/i, h, h, v) =I 0. But
0 = F(p)(h,h)-F(p)(h,h) = volg(p)(h,h,B,v)-volg(p)(h,h,B,v)
= volg(p)(h,h,afi,v) = avolg(pJ(/1,h,h,v),
and so volg(p)(/1, h, h, v) =0, a contradiction.

18 ----+ ----+ ----+ ----+ ----+ ----+ ----+
We will use the curl-of-curl identity v' x ( v' x v ) = v' ( v' • v ) - D. v for
any smooth map v:
JR3 -> JR3 . We have
oE
2---+
o oE o ,. -+ -+
----+
oj o -+ -+ oj -+ oB- - ----+
ot2 = ot at = o/- J + v x 8 ) = - ot + o/ v x 8 ) = - ot + v x at
= -
oI
ot + v'
-+
X (-
-+
v' X
-+
E) = -
oI
ot -
-+ -+ -+
v' ( v'. E) + D. E
-+
oI -+
= - - - v'p + D.E and
-+
ot '
o2 B o oB o -+ -+ -+ oE -+ - -+ -+
ot2 = ot at = o/- v' X E) = - v' X at = - v' X (- j + v' X B)
With E(x, y, z, t) = (sin(z-t), 0, 0) and B(x, y, z, t) = (0, sin(z-t), 0),

-+-+ o
v'·E = -sin(z-t) + -0 +-0 = 0
o o
ox oy oz
v' x E = (0, cos(z-t) - 0, 0) = -(0, - cos(z-t), 0) = -at
-+-+ oB
-+-+ o o
v'·B = -0 + -sin(z-t) +-0 = 0
o
ox oy oz
v' x B = (- cos(z-t), 0, 0) = at·
-+ -+ oE

The orthonormality of (¼)p, i = 1, 2, 3, 4, is easily verified. For example,
g(p)((V4)p, (¼)p) = -f(r)(dtp ®dtp)(~ot,p, ~ot,p) = -1.
Recall the volume form field volg 52 on (S2 , g 8 2) as in Example 11.4 (p.251).
Then the ordered basis (00,p, o,t,,p) is positively oriented with the induced
orientation on TpS 2 , for all p EU, and volg 52 (p)(o0,p,o,t,,p) = sin0 > 0.
With the orientation [O] on M, where n := r2dr I\ volg 52 A dt, for all p EU,
O(p)((Vi)p, (V2)p, (½)p, (¼)p) = r 20(r) 2( ~ O) (sin0). ~ = 1 > 0.

r sm y J(r)
18 See for instance [Apostol(1969), Example 5, p.445, (12.38)].
Solutions 459
So for each p EU, ((V1)p, (V2)p, (Vi)p, (Vi)p) is a positively oriented ordered
basis for TpM.
Finally, we show that F satisfies the Maxwell equations d F = 0 and
d * F = 0. (The RHS of the second equation is 0 since J = 0.) That d F = 0
is easily seen, since
dF = d(:2dr A dt) = (or :2)dr A dr I\ dt = 0.
To show d*F=0, we first determine *Fusing Proposition 11.3 (p.258):
*F = e2 * ( ~ d r " v'i[r)dt)
r V f(r)
= ; 2 sign ( ! !)g(Vi, V2) g(Vi, Vi)
~ ~ 0 2A 0 3
= e2 (1)(1)(1)rd0A (rsin0)d¢ = e(sin0)d0 Ad¢.

r
Consequently, d * F = e (00 sin 0) d0 " d0 " d¢ = 0.
Chapter 13

The constant function v is smooth and everywhere positive. Moreover,
v'vV = v'aJ3t = rboOxi =0,
g(p)(ot,p, Ot,p) = -1, and Ot,p is future-pointing (as the time-orientation is
given by Ot)- Finally,
div(v V) = v* dxi(v' i'J.,; Ot) = v* dxi(r~iOxk) = v* dxi(Ooxk) = v* dxi(O) = 0.
T~ ~ ~ l·
Since Tii =T(oxi, oxi) =mv*g(ot, Oxi) g(8t, oxi) = mv*(-8i0)(-8jo),
[Tij] = [
0 0 0 0
As the curvature tensor field for the Minkowski spacetime is identically
zero, it follows that the Ricci curvature tensor field Ric= 0, as well as the
scalar curvature S = 0, and so the field equation becomes 0 - 0 +Ag= 81rT,
and in the chart used above, this acquires the form
This holds if and only if 0 =A= - m v*. As m > 0, the field equation holds
if and only if v* = 0.

We use the reversed Cauchy-Schwarz inequality: For all q E M, all causal
X E TJ M, and V as in Definition 13.4, we have that
(g(q)(Vq, Xq)) 2 ~ g(Vq, Vq)g(q)(Xq, Xq) = -lg(q)(Xq, Xq)-
Thus
T(q)(Xq, Xq) = (p(q) + p(q))((V~ X)(q)) 2 + p(q)g(q)(Xq, Xq)
= (p(q) + p(q))(g(q)(Vq, Xq)) 2 + p(q)g(q)(Xq, Xq)
= p(q) (g(q)(Vq, Xq))2 +p(q)((g(q)(Vq, Xq)) 2 +g(q)(Xq, Xq))
~ p(q)0 + p(q) (-g(q)(Xq, Xq) + g(q)(Xq, Xq)) = 0.
Consequently, T(X, X) ~ 0 (pointwise) for all causal X E TJ M, i.e., T
satisfies the weak energy condition.
Solutions 461

Differentiation of both sides of 3 a2- Aa 2 = 81r p a 2 (first Friedmann equation)
with respect to t gives
6aa - 2Aaa = 81r(pa 2 + 2paa). (17.15)
On the other hand, multiplying (13.7), namely
a a2
-2- - - + A=81rp,
a a2
throughout by 3aa yields
·3
-6aa - 3::..a + 3Aaa = 3(81rpaa). (17.16)
Adding (17.15) and (17.16), we obtain
-3aa(:2
·2
-
A
3) = -3: ·3
+ Aaa = 81ra(3pa + 2pa + pa). (17.17)
But, by first Friedmann equation
a2
---
A 81rp
a2 3 3
So, replacing the bracketed expression on the LHS of (17.17) by~,
. B1rp = 81ra (3 pa+2pa+pa.
- 3 aa . . . )
3
Simplifying and rearranging, it follows that pa+3pa+3pa=0. Multiplying
throughout by a 2,
0 = pa 3 + p3a2 a+ p3a 2 a= ! (pa 3 ) :t

+ p (a 3 ).

Suppose that V E TJ M is causal. We use the vector fields Vi, V2, ½, ¼
from Exercise 12.9 to decompose V = JiV;,, where f E C 00 (U) in the chart
U. As Vis causal, g(V, V),,;; 0 (pointwise). Using the 'orthonormality' of
Vi, Vi,½, V4, we obtain (/1) 2 + (/2)2 + (/3) 2 - (/4)2,:;;; 0 pointwise in U.
Now
2
T(V, V) = _e_((/4)2 _ (!1)2 + (!2)2 + (!3)2)
81rr4
2
~ 8e 4 ((/2)2 + (/3)2 + (/2)2 + (/3)2) ~ 0.
1rr
Thus T satisfies the weak energy condition.
Chapter 14

We know that traceRic=S and traceg=4. Taking the trace on both sides
of the field equation yields S - ½S 4 = 81r trace T, and so S = - 81r trace T.
Substituting this back in the field equation, Ric - ½(- 81rtraceT)g = 81rT.
Rearranging, we obtain Ric= 81r(T - ½(traceT)g).
On the other hand, if Ric= 81r(T - ½(traceT)g), then again taking
traces, S = 81r((traceT)-½(traceT)4) = - 81r traceT. Substituting this in
the assumed equation, we get Ric= 81rT+ ½(- 81rtraceT)g = 81rT+ ½Sg.
Rearranging, we obtain the field equation Ric - ½Sg = 81rT.

Let V E TJ- M be lightlike. Then V is causal, and so by the weak energy
condition, T(V, V) ~ 0. Also as V is lightlike, g(V, V) = 0. Using the field
equation,
0 ~ 81rT(V, V) = Ric(V, V) - ~Sg(V, V) + Ag(V, V) = Ric(V, V).
If T satisfies the strong energy condition, and V E TJ- M is timelike, then

by Exercise 14.1, Ric(V, V) = 81r(T(V, V) - ½(traceT)g(V, V)) ~ 0.

In the given global chart (JR4, (t, x, y, z) >-+ (t, x, y, z)), the component ma-
[g.J] - 1
1
l[
trix of g is (with h :=H(x, y, t - z)):
. . -[-1 l
+
2~
-2h
0
~ ~
0 0
O O
-2~] - [-1 + 2h
0 -
2h -2h
l
1
-2h
1+2h
l
.
To see that Y is lightlike every where, we note that using the basis
(ot, ox, oy, oz), the vector field Y is represented by the column vector
(1, 0, 0, 1), and so computing
-1 + 2h - 2h ] [ 1 ] [ 1]
[11] [ -2h 1+2h 1=[-11] 1 =-1+1=0,
we conclude that g(Y, Y) = 0. Similarly, we can represent the ½ as column

vectors using the basis (ot, ox, oy, oz), and use the component matrix of g
Solutions 463
J
given above to compute
[g(V;, V;)] ~ [-1 I I
For example, the (1, 1) entry follows by noting that
[ 2h + 3 2h + 1 ] [-1 + 2h - 2h ] [ : ~
2v'2 2v'2 -2h 1+2h 2h+l
l 2
2y'2
3
= _1
.
From the form of [g(¼, ½)], it follows that at each point p E M, the set
{ (¼)v : i = 1, 2, 3, 4} is an orthonormal basis for TpM with respect to the
scalar product g(p), and the index of g(p) is 1. Thus g is Lorentzian.
The scalar curvature is S = gii Ricij. To find [gii] = [gij J- 1, note
-1
[ -1+2h -2h] = [-1-2h -2h]
l
-2h 1+2h -2h 1-2h .
Thus
.. [-1-2h -2h
[g'J] = 1 1 .
-2h 1-2h
As the only nonzero components of Ric are the (i,j)-components with
i,j E {1,4}, we obtain, with tlh := ox(oxh) + oy(oyh), that
S = g 11 Rieu + g 14 Ric14 + g 41 Ric41 + g 44 Ric44
= (-1- 2h)(-tlh) + (-2h)(tlh) + (-2h)(tlh) + (1- 2h)(-tlh)
= (tlh)(l + 2h - 2h - 2h - 1 + 2h) = (tlh)(0) = 0.
For a harmonic (in x, y) h, Llh=0. So Ric= -(tlh) (dt-dz) ® (dt-dz) =0.
So the vacuum (T = 0) field equation is satisfied with the cosmological
constant A= 0:
Ric - ~g +Ag= 0 - ~g + 0g = 0 = 87!"0 = 811"T.
Conversely, if the vacuum field equation is satisfied, then we have
Ric= Ric - ~g + 0g = Ric - ~g +Ag= 811"T = 871"0 = 0.
So Ric=0. In particular, Ric(8t, 8t) =0, i.e., -Llh=0. Thus his harmonic.
We have Ll(x 2 -y2 ) = 2 + (-2) =0, and tl(xy) =0 + 0=0. Thus
1 1
tlh = 2/(t- z)tl(x 2 -y 2 ) + g(t- z)tl(xy) = 2/(t- z)0 + g(t- z)0 = 0.

We can rewrite the metric as
e2x
g = -(dt + exdy) ® (dt + exdy) + dx ® dx + 2 dy ® dy +dz® dz.
Define Vi= Ot, V2 = Ox, ½ = -v'2 Ot+v'2 e-x Oy, Vi= Oz. (Note that½
J
satisfies (dt + exdy)½ =0 and g(½, ½) = 1.) Then
[g(V;, V;)] ~ [-1 I I
Hence g is Lorentzian. As g(ot, Ot) = -1, V := Ot is timelike everywhere.

T is the energy-momentum tensor field of dust: We set m = 8~ > 0,
V= 1, V = Ot = V1. We have g(p)(Vp, Vp) = g(p)(ot, Ot) = -1. It follows also
that Vp is future-pointing. Furthermore, we have 'v'vV = 'v'a,ot = ritox;.
But since gtj does not depend on t, and since gu = -1, it follows that
i gij gij
rtt = 2 (otgti + otgjt - oigtt) = 2 (0 + 0 - 0) = 0.
So 'v'vV =0. We have
div(vV) = divV = dxi('v'ax;ot) = dxi(rtoxi) = rt<>J = r:i.
We have already seen r~t = 0. One can check that also rfx, rfy, rfz are 0:
For example, noting that gu = -1 and Otgyy = Ot (-e2x/2) = 0, we have
y gYi
rty = 2 (otgjy + oygtj - ojgty)
gYt gYY
= 2 (otgty + oygtt - otgty) + 2 (otgyy + oygty - oygty) = o.
Thus Tis the energy-momentum tensor field of dust (m, v, V) = ( 8~, 1, Ot)-
As Ric= (ot)i, ® (ot)i,, its components in the chart IR4 with the chart
coordinates (t, x, y, z) =: (x1, x 2, x 3 , x 4 ), are given by (using (ot)i, = glkdxk):
Ricij = (glkdxk ® gudxe) (Oxi, Oxi) = glk gie 8f 8J = g1i glj ·
The scalar curvature is S = giiRicij = gii g1i g1j = 8{ g1j = gu = -1. So
. S
Ric - 2 g + Ag = (ot) i, ® (ot) i, - -(-1) ( 1) i,
2-g + - 2 g = (ot) ® (ot)
i,
= s1r( 8~)(od ® (od = 81rT,

that is, the field equation is satisfied with the given cosmological constant
and the matter energy-momentum tensor field T.
Solutions 465
To see that the given curve is closed, we compute

lim t(s) = lim
S ---+7r S---+7r
(2v2 tan- 1( e-dtan-2s)--J2s) = 2v2 tan- 1(+00)-v'21r
= 2v2~-v27r
2
=0
'
lim t(s) =
S -+-7r
2v2 tan- 1(-oo)-v'2(-1r) = 2-J2(- ~)
2
+ v'21r = 0,
and also
x(±rr) = log((coshd) + (-l)sinhd) = loge-d = -d
y(±rr) = (sin(±rr))( .. •) = 0( .. •) = 0
z(±rr) = 0.
A plot of the curve is shown in the following picture.
-y

Let c<•l := [gi;>J, Y := [%], H := [hi1]. Then c<•l = Y + sH. As Y = y- 1 ,
for small Isl, we have
(c<•))- 1 = (Y + sH)- 1 = (Y(I + sYH))- 1 =(I+ sYH)- 1 Y
= (I - sYH + s2 YHYH - + · · · )Y.
Thus
. (G<•l)- 1 - (G< 0l)- 1 . (I-sYH + s 2 YHYH - + · · · )Y-Y
hm - - - - - - - = hm - - - - - - - - - - - - -
s -+O S s -+O S
-sYHY + s 2 YHYHY- + · ··
= lim - - - - - - - - - - - = -Y HY.
s -+O S
Recall that (g<•) )ii is the entry in the i th row and i h column of c<s). Hence
(7 (g<•) )ij I
= [-Y HY]ij = - T/ik hkf T/fj .
OS s=O

By Proposition 6.6, for all p EM, and all 1 ~ i,j ~ 4, we have
Consequently, hv = .Cvg.

We have V7,')'(r) =t'ot,7(r) + r'or,7(r)· As "f is parametrised by arclength,
we have g('Y(r))(v7 ,7 (r), v'"Y,'"Y(r)) = -1 for all TE J. Using the expression for
the Schwarzschild metric, we obtain
2m)(t ') 2 +
- ( 1---;=- ( 2m)- 1( r ') 2 + 0 + 0 = -1.
1---;=- (17.18)
The other two equations are obtained by looking at the t- and r-components
of the geodesic equation ;;/ + rf/fyi = 0. Recall the connection coefficient
formulae from Example 6.8 (p.124). The t-component of the geodesic equa-
tion , O = t" + rttt (t') 2 + 2rttr t'r' + rtrr (r') 2 , gives
2m(1 -
2m)- 1
O=t"+O+- - t'r'+O.
r 2 r
Similarly, the r-component, 0 = r" + rrt (t') 2 + 2r~tr't' + r~r(r') 2 , gives
1 (') 2
0 = r ,, + -m( l -2m)(')
- t 2 +0- -m(1 -
2m)-
- r
r2 r r2 r
1
= r ,, - rm(
2 - ( 1- --;::-
2m)(') 2
t 2m)-
+ ( 1- --;::- (r' ) 2 )
= r II - -m ( -l ) (using (17.18)).
r2
Solutions 467
Relative speed of w=t'ot,p+ r'or,p perceived by Vis U= -g(p)(w,v) - v. So
u= _ t' Ot,p + r' Or,p _ . / ( 1 _ 2m )-l Ot P
-J
g(p) ( t' Ot,p + r' Or,p, (1- ";' )- Ot,p) V
2 1 r '
-- -r'v(
t'
2m)-1
1-r- ar,p·
Thus
lul = y1g(p)(u, u) = I:: Iv( 27;)-1-Jg(p)(or,p, Or,p) = I:: I( 27; )-1.

1- 1-
For lul « 1 and r » 2m, we obtain lr'I « lt'I- Equation (17.18) now shows
that t' ~ 1. From the equation for t 11 above, we see that if we neglect the
small terms, we get t 11 ~ 0. We note that
dh __l _
and
d2h t 11 (h(·)) 1~
t 11 (h(·))
dt t'(h(-)) dt 2 (t'(h(-))) 2 (t'(h(-))) 3 .
Thus
d(r oh) dr dh , l
dt = dr dt = r (h(-)) t'(h(·))'
d2(r oh) 11 l 1 , t 11 (h(·))
dt2 = r (h(·)) t'(h(·)) t'(h(·)) - r (h(·)) (t'(h(-))) 3 .
Using t' ~ 1, lr'I « t', and r»2m, we get
d2 ( ) 11 l , 1 11 m l , 1 ( - 2m) ( 2m )- 1 , ,
dt2 roh =r (t')2-r (t')3t =-r2(t')2 -r (t')3 r2 1- r tr
= -;((t\2 -2c:r (1- 2r;)-l)

m(l1 + 2(0) 2 1 -1 0 ) =
~ - r2
m
- r2·

Reinstating the constants c = 3 x10 8 m/s ~'speed of light'), and the Newton
gravitational constant GN = 6.6 x 10- 11 k:.s2 , the mass m is m = ~: ~', and
this will deliver the answer in kilograms if Rs is specified in meters, with
the above c and GN. So
c2 Rs c2 R.,
.
dens1ty mass GN 2 GN 2 3 c2 1 4 x 10 12 -
kg
= --- = -- ~ - - = ---- ~ .
volume 41r R! 41rR! 81rGN R 2 m3
--
3
--
3
"
The minimum density of the body is p = 4 X 10 12 ~' and p : Pwater = 4 X 109 •

Recall the expressions on page 315 for the components of Ric for a spheri-
cally symmetric stationary metric g. Taking k = j, and f = l - 2;' + ~,
• ff" f' f e2 f
Ricoo = -
2
+ - r = -r 4
• f" f' e2
R1c 11 = -- - - = --
2f rf r 4f
2
Ric22 = - f + l - r J' = ~r2
• . 2 . e2(sin0) 2
R1c33 = (sm 0) R1c22 = ,
r2
and all other components are identically 0. Thus the scalar curvature is
i. • le 2f ( e2 ) 1 e2 1 e2(sin0) 2
s = g 3 R1Cij = --J-4
r +f - --:i-1
r +2
rr2 + rsm
2(. 0)2 r2 = o.
Hence the field equation with A= 0 reduces to Ric= 81rT. We recall that
e2
81rT = 4 (0 4 ® 0 4 - 0 1 ® 0 1 + 0 2 ® 0 2 + o3 ® o3).
r
Also recall the vector fields¼ from Exercise 12.9 that are 'dual' to the one
form fields OJ. We have
e2 4 2 e2 4 /"i 2 e2 f •
81rToo = 4 (0 (ot)) = 4 (0 (y JV4)) = - 4 = Ricoo
r r r
e2 1 (or)) 2 = - e2
81rT11 = --(0 - ( 0 1 ( -Vi
1 ))2
= - e2- = Rieu
r4 r4 v? r4J
e2 2 2 e2 2 2 e2 .
81rT22 = r 4 (0 (00)) = r 4 (0 (rVi)) = r 2 = R1c22
2 2 2( . 0)2
81rT33 = \ (03(8¢)) 2 = \ (03(r(sin0)Vi)) 2 = e si~ = Ric33.
r r r
The components of 81rTij for i -# j are zero, and so they coincide with the
components Ricij = 0. So the field equation is satisfied.
Solutions 469
Chapter 15

We want to evaluate the integral
~
t(r) =
ir
ro
1
-1 _ £
r
yr
Gdr = -
ya
frro
ryr
( _ )vadr,
r a a
where a:= 2m. We use the substitution r = u 2 . Then we have
~
t(r) = -
ir ryr
( _ )vadr = -
r a a
f ,./r
(
Fo u 2
u2u
_ )va2udu
a a
s; (
ro
= - ~ t~(u-lva, _ u+ly1o,))du
u2 +a+
_
-
~
c
ya
3 + Ya Y ro Yr + a log (Fo-va.
ro.jro-ryr 2 r::( ~- r.::)
c
vr+va)
c . r,;;::
yro+ya
r,;:
yr-ya
The plot with the given numerical values is shown below.
-I
Setting ~m(v(r)) 2 = - V(r) = GNmm, we obtain v(r) = ) 2 GNm.

2 r r
If the escape velocity at r = r s equals the speed of light c, then rs= 2 G;m.
C
Finally, for the star to not shine, we must have that the escape velocity
v(R) at its surface exceeds the speed of light, that is, v(R) > c. Since
m = (volume)· (density)= 4; R3 p, substituting this in the expression for the
escape velocity, the inequality v(R) > c gives
2 GN \,rR R p > c,
3 .
1.e., R >
~c 2
- -0 .
81r NP

When the mass !:lm reaches the singularity at r = 0, the mass of the black
hole becomes m + !:lm, and the Schwarzschild radius increases. Thus the
increase in area is
Consequently,
Solutions 471
Chapter 16
The arclength of 'Yl(o,r) is L(r) := I: u('y(r))(v,,,,,(r), v,,,,,(r)) dr.
dr d0o d<po
As v,,,,,(r) = dr or,,,(r) + dro0,,,(r) + dr oc/>,,,(r) = or,,,(r), we have
1
u('Y(r)) (v,,,,,(r), v,,,,,(r)) = l-kr 2 .
Thus if k = 0, then
L(r) = lr l
~ = = dr =
o ✓l-kr 2
-
o 1-0
dr = r.
lrl
If k < 0, then we use the substitution sinh u = H r E (0, oo) to obtain
L(r)= l r
~==dr=
o ✓ l-kr 2
l lsinh-i(v"='icr)
o
1 coshu sinh- 1 (Hr)
----du=--=--.
coshu R R
If k > 0, then we use the substitution sin u = v'kr E (0, 1) to obtain
fr 1 rsin-l(v'kr) 1 COSU Sin- 1 (v'kr)

L(r)= Jo ~ d r = Jo cosu v'k du= v'k .
By the mean value theorem, for x > 0, there is a Cx E (0, x) such that
sinh- 1 x - sinh- 1 0 dsinh- 1 ( ) 1

- - - - - - - = - - - Cx = ---;::===:::;:: < 1.
x-0 dx y11+(cx) 2
Uisng this, we see that if k < 0, then

1
L(r) < r-r(Hr) = r.
v-k
For x E (0, 1), there exists a ex E (0, 1) such that
sin- 1 x - sin- 1 0 _ dsin- 1 ( ) _ 1

------ - - - - Cx - ---;:===== > 1.
x- 0 dx y1l - (cx) 2
Hence, if k > 0, then

1
L(r) > v'k(Vkr) = r.

We have v µ,,µ,( s) = r' (s) Or,µ,( s), where -' = fs . Thus
u(µ(s))(vµ,,µ,(s),Vµ,,µ,(s)) = (r'(s)) 2 1 _(~([;~:)) 2 = 1.

So the length d(t) ofµ in (S, u) is
d(t) = f 0
u(µ(s))(vµ,,µ,(s),Vµ,,µ,(s))ds = f 0
Ids= so.
Also, using the differential equation for r, we have
so lso a(t)r'(s) lr(so) a(t) lro a(t)
~= l 1~= ~=~~~~= ~==~= ~==~
o o y'l-k(r(s)) 2 r(o) -v'l-kr2 o -v'l-kr 2 '
where we used the substitution r=r(s), and the given boundary values for
r. The 0- and ¢-components ofµ, being constant functions, satisfy 0' = 0
and ¢' = 0, and so
0" + 2r~00'r' + r:,t,(¢') 2 = 0 + 2r~0Or' + r:,t,(O) 2 = 0,
,,_.,, + 2r<t>r,t,'1',1..'r' + 2r<t>0,t,'I',1..'0' = 0 + 2r<t>r,t, Or'+ 2r<t>04> 00 = 0·
'I'
Finally the r-component of the geodesic equation is also satisfied, since
r"(s) + r;r(r')2+ ri;9(0')2+ r:;,,t,(4>') 2
= r"(s) + r;r(r')2 + r00(O)2 + r:;,,t,(O)2 = r"(s) + r;r(r') 2
= ~ y'l - k(r(s)) 2 + kr(s) (y'l - k(r(s)) 2 )2

ds a(t) l-k(r(s)) 2 a(t)
1 1 kr(s)
( 2 k ( ) '( ))
a(t) 2y'l-k(r(s))2 - rs r s + (a(t)) 2
__1_ 1 kr s y'l-k(r(s)) 2 + kr(s) =O
a(t) y'l-k(r(s))2 ( ) a(t) (a(t)) 2 ·

Differentiating with respect to t the relation
d(t) = a(t)
o
ro
-v'
1
l
1 - kr 2
dr,
we obtain
. lro 1
d(t) = a(t)
o
-v' 1- kr
2 dr.
Dividing these yields

d(t) = a(t) = H(t)
d(t) a(t) .
Consequently, d(t) = H(t) a(t).
Solutions 473

Diuerentiatmg
·a · · a· 2 + k = 3
81T pa 2 gives
•
. .. + 0 = 81T (pa
2 aa . 2 + 2 paa. ) ,
3
.. 81Tpa 2 +2paa . . . a a2 k .
and so a= -3 2 a. . Substitutmg m 2- + 2 + 2 = -81Tp yields
a a a
281T pa+ 2pa a 2 .!5_ _ -8
3 2 a' + a2 + a2 - 1Tp,
and using a:a + a\ = 831T pin the left-hand side of the above now gives
81Tpa+2pa 81T =-8

3 2a + 3 P 1Tp.
Rearranging, we obtain p + 3 (p + p)~ = 0, as wanted.

a
Assuming p=wp, we have p + 3(1+w)p~ =0, that is,
a
d . .
-d ((logp)+3(1+w)loga) = _e+3(1+w)~ =0.
t p a
Thus on I, we have logp+3(1+w)loga = C', a constant. Consequently,
p(t) = Ca(t)- 3 (1+w), for all t EI, where C = e0 '.
• For radiation, w= 1/3. Sop oc a- 3 (1+½) =a- 4 .
• For dust, w=0. Sop oc a- 3 (1+o) =a- 3 .
• For the cosmological constant, w= -1. So p=Ca- 3 Ci-i) =C, a constant.

If A> 0, then a= fcosh(t/e), a= sinh(t/£) and i.i = (1/£) cosh(t/£). Thus,
using (cosh(t/£)) 2 - (sinh(t/£)) 2 = 1, we obtain
a2 k (sinh(t/£)) 2
-+-=~-~~-+-----=-=-=--=-p
1 1 A 81T A 81T
a2 a 2 £2 (cosh(t/£)) 2 £2 (cosh(t/£)) 2 £2 3 3 81T 3 '
showing that (16.9) holds. Similarly, (16.10) holds since
2~ + a2 + .!5.._ = 2 (1/£) cosh(t/£) + (sinh(t/£)) 2 + 1

a a2 a2 ecosh(t/e) £2 (cosh(t/£)) 2 £2 (cosh(t/£)) 2
3
= £2 = A = - 81Tp.
If A < 0, then a = fsin(t/£), a = cos(t/£) and i.i = -(1/£) sin(t/£). Thus,

using (cos(t/£)) + (sin(t/£)) = 1, we obtain
2 2
a2 k (cos(t/£)) 2 -1 1 A 81r A 81r

a2 + a2 = £2(sin(t/£)) 2 + £2(sin(t/£)) 2 = - £2 = 3 = 3 81r = 3P,
showing that (16.9) holds. Similarly, (16.10) holds since
2~ + a2 + ~ = 2 -(1/£) sin(t/£) + (cos(t/£)) 2 + -1

a a2 a2 fsin(t/£) £2(sin(t/£)) 2 £2(sin(t/£)) 2
3
= - £2 = A = - 81rp.

Using the first Friedmann equation, and the definitions of H, M, L,
H2= a2 C:J 81r ~ = 81r a3 ~ = M ~

a2 3 p+ 3 3 p a3 + £2 a3 + £ 2·
Also, using both Friedmann equations, and noting that p = 0, we obtain
iI=~ _ a2 =~3 _41r3 p- 81r3 p-~=-~81r pC:J ~(~ -H2) =~(~ -H2)
a a2 3 2 3 2 3 2 £ 2 ·
With q := (I/L)-H' we have
. if J(p-H 2 ) Hi+H) 3 3 1 3 3
q= (i-H) 2 = (i-H) 2 = i-H = -2+£ (i-H) = Lq - 2·
Multiplying bye-ft throughout, and rearranging, we obtain

d at at. 3 at 3 at
dt (e-T q) = e-T q - Le-,; q = -2e-T .
a 3 e-ft - 1
Integrating from t = 0 tot, we obtain e--rtq(t) - q(0) = 2 3/L .
Substituting q(t) = (l/L) ~ H(t), and solving for H(t), we find
H(t) = ~ (i + H(O))eft - (t - H(O)). (17.19)

L (i + H(O))eft + (t - H(0))
• 2 M 1
Usmg H = a3 + £ 2 , we have
a(t) = W((H(t)) 2 - ; 2 )-½. (17.20)

Solutions 475
Setting t=0 in H 2 = ~ + ; 2 , we get H(0)2=: + ; 2 . So

1 2 M
£2 -(H(0)) = - A3'
1 1 ~ 1( ~)
I±H(o)=I±y-;p+v=I i±y1+A3.
Using these, and by substituting for H(t) from (17.19) into (17.20),
We have V>.,>.(s) = t'(s)c'1t,>.(s) +r'(s)or,>.(s), where .' := t-

As the geodesic A
is null, we have g(.X(s))(v>.,>.(s),V>.,>.(s))=0 for alls E (a,/3), giving
-(t'(s)) 2 + (a(t(s)))2(r'(s)) 2 = 0.
Using this, and the facts that t' > 0, r' < 0, and a> 0, we obtain
R = R - 0 = r(a) - r(/3) = - f /3 r'(s)ds = f/3 a(t(s))

0
t'(s)
0
ds
ft(/3) 1
(using the substitution t( s) = T)
= Jca) a(r) dr
= IT
to
1
--dT
a(r) ·
Since to>0, and a>0, we also get R= IT t0

1
-() dt <
at
lT -()1 dt=RH(T).
0 at
Bibliography
Apostol, T. (1969). Calculus. Volume II, 2nd edition (Wiley).

Artin, M. (1991). Algebra (Prentice-Hall).
Beem, J., Ehrlich, P., Easley, K. (1996). Global Lorentzian Geometry, 2nd edition
(Marcel Dekker).
Bishop, R. and Goldberg, S. (1980). Tensor Analysis on Manifolds (Dover).
Cooke, R. (2017). It's About Time. Elementary Mathematical Aspects of Relativity
(American Mathematical Society).
Corless, R., Gonnet, G., Hare, D., Jeffrey, D., and Knuth, D. (1996). On the
Lambert W function. Advances in Computational Mathematics 5, 4, pp.
329-359.
Dodson, C., and Poston, T. (1997). Tensor Geometry. The Geometric Viewpoint
and its Uses, 2nd edition (Springer).
Feynman, R. (1963). Lectures on Physics. Volume 1 (Addison-Wesley).
Frobenius, G. (1877). -Ober das Pfaffsche Problem. Journal fur die Reine und
Angewandte Mathematik 82, pp. 230-315.
Godinho, L., and Natario, J. (2014). An Introduction to Riemannian Geometry.
With Applications to Mechanics and Relativity (Springer).
Grave, F., and Mueller, T. (2010). Catalogue of Spacetimes
(https://arxiv.org/abs/0904.4184).
Hakopian, H., and Tonoyan, M. (2004). Partial Differential Analogs of Ordinary
Differential Equations and Systems. New York Journal of Mathematics 10,
pp. 89-116.
Halmos, P. (1987). Finite-Dimensional Vector Spaces, 2nd edition (Springer).
Hartman, R. (2002). Ordinary Differential Equations, Classics in Applied Math-
ematics 38 (SIAM).
Kocik, J. (2012). Geometric Diagram for Relativistic Addition of Velocities. Amer-
ican Journal of Physics 80, 8, p. 737.
Kriele, M. (2001). Spacetime, Corrected 2nd printing (Springer).
Landau, L., and Lifshitz, E. (1976). Mechanics. Course of Theoretical Physics.
Volume 1, 3rd edition (Elsevier).
Lang, S. (1999). Fundamentals of Differential Geometry (Springer).
Lee, J. (2013). Introduction to Smooth Manifolds, 2nd edition (Springer).
477
Lee, J. (2018). Introduction to Riemannian Manifolds, 2nd edition (Springer).

Lightman, A., Press, W., Price, R., and Teukolsky, S. (1975). Problem Book in
Relativity and Gravitation (Princeton University Press).
Marsden, J. and T. Ratiu, T. (2007). Manifolds, Tensor Analysis and Applica-
tions, 3rd edition (Springer).
Misner, C., Thorne, K., and Wheeler, J. (2017). Gravitation (Princeton University
Press).
Momin, A. The Godel Solution to the Einstein Field Equations
(http://www.math.toronto.edu/~colliand/426/Papers/A_Monin.pdf).
Moretti, V. Geometric Methods in Mathematical Physics II: Tensor Analysis on
Manifolds and General Relativity
(http://www. science. unitn. it/ ~moretti/dispense .html).
Natario, J. (2021). An Introduction to Mathematical Relativity, Latin American
Mathematical Series (Springer).
Oloff, R. (2018). Geometrie der Raumzeit. Eine mathematische Einfiihrung in die
Relativitatstheorie, 6th edition (Springer).
O'Neill, B. (1983). Semi-Riemannian Geometry. With Applications to Relativity
(Academic Press).
O'Neill, B. (2014). The Geometry of Kerr Black Holes (Dover).
Padmanabhan, T. (2008). Gravity: the Inside Story. General Relativity and Grav-
itation, 40, 13-14, pp. 2031-2036.
Padmanabhan, T. (2016). The Atoms of Space, Gravity and the Cosmological
Constant. International Journal of Modern Physics. D. Gravitation, Astro-
physics, Cosmology, 25, 7, 1630020.
Rudin, W. (1983). Principles of Mathematical Analysis, 3rd edition (McGraw-
Hill).
Sasane, A. and Ufnarovski, V. (2016). Alternative Proofs for Kocik's Geometric
Diagram for Relativistic Velocity Addition. Elemente der Mathematik 71,
3, pp. 122-130.
Sachs, R. and Wu, H. (1977). General Relativity for Mathematicians (Springer).
Schuller, F. (2015). Lectures, WE-Heraeus International Winter School on Gravity
and Light (https://gravity-and-light.herokuapp.com/lectures).
Shilov, G. (1975) Analyse Mathematique. Fonctions de Plusieurs Variables Reeles
(MIR).
Wald, R. (1984). General Relativity (University of Chicago Press).
Index
(r, s)-tensor field, 61 charge-current density, 284, 285

1-form, 34 chart, 2
1-form field, 50 chart transition map, 4
ACDM model, 360 chart-induced basis tangent vectors,
g-index of a tuple, 257 26
k-form, 209 chart-induced basis 1-forms, 35
k-form field, 216 Christoffel symbols, 116
classical spacetime, 261
absolute time, 261 closed form field, 223
acceleration equation (cosmology), closed timelike curves, 307
355 cold dark matter, 360
affine space, 3 collision, 278
algebra, 43 compatible atlases, 6
allowed collision, 278 complement of a tuple, 257
alternation, 211, 217 complete vector field, 56
anti-de Sitter spacetime, 356 Compton formula, 278
atlas, 4 Compton length, 343
azimuthal angle, 8 Compton scattering, 278
connected topological space, 239
Betti number, oth, 225 connection 1-forms, 119
Bianchi identity (first), 183 connection coefficients, 116
Bianchi identity (second), 186 connection on a smooth manifold, 110
big bang, 356 contraction, 72
black hole, 318 contravariant index, 111
cosmic no-hair conjecture, 356
canonical projection, 40 cosmological constant, 305
Cartan's magic formula, 233 cosmological time, 193
Cauchy-Schwarz inequality for the cotangent bundle, 48
Minkowski scalar product, 83 cotangent space, 34
causal vector, 81 covariant curvature tensor field, 183
causal vector field, 294 covariant derivative, 110, 126
charge, 267 covariant derivative of tensor fields,
479
128 exterior derivative of a k-form field,

covariant index, 111 221
curl of a vector-valued function, 226 exterior product, 218
curvature operator, 181 exterior product of forms, 213
curvature tensor field, 125, 181
curve, 13 family of geodesics, 196
cycloid, 358 first law of thermodynamics, 303
cylindrical spacetime, 9, 88 flat connection on Rm, 109
flat map, 103
de Rham cohomology, 225 flow of a vector field, 56
de Sitter spacetime, 356 FLRW spacetime, 9, 352
degree of a form field, 216 frequency of a light signal, 98
density of a perfect fluid, 298 Friedmann equation, 302, 303, 355
derivation, 43 fundamental theorem of
derivative of a smooth map, 30 semi-Riemannian geometry, 120
diffeomorphism, 15 future-pointing vector, 91
dimension of a manifold, 7
divergence, 226 Godel spacetime, 307
divergence of a (0, 2)-tensor field, 189 gauge transformation, 311
divergence of a vector field, 119 geodesic, 155
double cone, 12, 330
geodesic equation, 156
dust, 295
geodesically complete manifold, 160
Global Positioning System, 165
Einstein-de Sitter universe, 192
GPS, 165
electric field perceived by an observer,
gradient, 226
280
emergent gravity paradigm, 343 Grassmannian, 346
energy momentum 1-form, 274 gravitational field, 290
energy of a particle measured by an gravitational potential, 290
observer, 274 gravitational wave, 306, 311
energy of a light signal, 98 great circle on a sphere, 157
energy of a photon measured by an
observer, 278 Hausdorff topology, 12
energy-momentum of a photon, 277 Hawking singularity theorem, 306
energy-momentum tensor field, 294 Hawking temperature of a black hole,
energy-momentum tensor field for 342
dust, 296 Hodge star operator, 254
energy-momentum tensor field for horizon, 102
perfect fluid, 298 Hubble constant, 354
equation of state, 300, 355 Hubble's law, 354
Euclidean space, 2 hyperbolic space, 351
Euler-Lagrange equations, 177
event horizon, 331 index of an scalar product, 77
exact form field, 223 induced orientation, 281
exact sequence, 225 induced orientation on tangent
exponential map, 170 spaces, 239
Index 481
inertial frame in Minkowski mass, 267

spacetime, 86 maximal geodesic, 160
inner product, 76 metric, 85
instantaneous Lorentz force, 283 metric index, 85
instantaneous observer, 199 Minkowski scalar product, 81
instantaneous observer in classical Minkowski spacetime, 9, 86
spacetime, 262 momentum of a particle measured by
integral curve, 54 an observer, 274
integral of functions, 251 momentum of a photon measured by
integral of top form fields, 247 an observer, 278
interior multiplication, 231, 255
inverse metric components, 90 negatively oriented ordered basis, 237
isometry, 85 Newton's first law of motion, 164
Newton's gravitational law, 289
Jacobi equation, 197 Newtonian limit, 308
Jacobi identity, 48 normal chart, 174
normal coordinates, 174
Kerr spacetime, 342 normal neighbourhood, 174
Killing vector field, 132 normalised instantaneous observer,
Killing's equation, 132 269
Kretschmann scalar, 189 null convergence condition, 306
null curve, 157
Lambert W function, 338
law of conservation of energy, 303 observer, 93
law of conservation of open cover, 241
energy-momentum, 278 open set, 2
left translation, 16 open subset of a smooth manifold, 10
left-invariant vector field on a Lie orientable manifold, 239
group, 45 orientation on a manifold, 239
Lie algebra of a Lie group, 45 orientation-compatible atlas, 245
Lie bracket, 46 oriented atlas, 240
Lie derivative, 58, 131 oriented vector space, 237
Lie derivative of tensor fields, 131 orthogonal subspace, 77
Lie group, 16 orthogonal vectors, 77
lift of a chart, 40 orthonormal basis, 80
light signal, 98, 277
lightlike curve, 157 parallel transport of a vector along a
lightlike vector, 81 curve, 150
linearised field equation, 311 parallel vector field along a smooth
locally finite collection, 241 curve, 144
Lorentz transformation, 83 particle, 267
Lorentzian manifold, 85 particle horizon, 360
particle in classical spacetime, 262
magnetic field perceived by an partition of unity, 241
observer, 281 past-pointing vector, 91
main curve, 196 path-connected topological space, 239
perfect fluid, 298 sharp isomorphism, 104

perihelion, 324 sign/signature of a permutation, 210
permutation, 210 simultaneous, 261
permutation action, 211 singular spacetime, 329
photon gas, 302 smooth curve, 15
Planck constant, xiii, 277 smooth manifold, 7
Planck length, 343 smooth map between smooth
plane gravitational wave, 307 manifolds, 13
Poisson equation, 289, 310 smooth structure, 7
polar angle, 8 space perceived, 199, 262
positively oriented ordered basis, 237 spacelike curve, 157
positron, 280 spacelike vector, 81
pressure of a perfect fluid, 298 spacetime, 239
product of smooth manifolds, 9 spacetime symmetry, 312
proper time, 93 spatial slice, 353
pull-back of a 1-form, 37 standard orientation on Minkowski
pull-back of a 1-form field, 53 spacetime, 259
pull-back of a k-form field, 220 standard smooth structure on Rm, 7
pull-back of a covariant tensor field, stationary metric, 312
73 Stokes' theorem, 248
push forward of a vector field, 45 strong energy condition, 306
subordinate to a cover, 241
radiation, 302 support of a function, 241
red-shift, cosmological, 192, 195 symmetric tensor field, 105
red-shift, gravitational, 100
Reissner-Nordstri:im spacetime, 286, tangent bundle, 40
304, 318 tangent space, 22
relative velocity, 267 tangent vector, 22
relative velocity in classical tangent vector field along a curve, 139
spacetime, 262 tensor, 64
rest energy, 275 tensor field evaluation, 66
restriction of a tensor field to a chart, tensor product, 70
69 tensor product of tensors, 64
Ricci curvature tensor field, 188 thermodynamics of black holes, 342
Riemannian manifold, 85 Thomson scattering, 278
rotating black holes, 342 tidal force, 200
time-orientation, 91
scalar product, 76 time-oriented Lorentzian manifold, 91
scalar curvature, 188 timelike convergence condition, 306
Schur's theorem, 349 timelike curve, 157
Schwarzschild metric, 316 timelike vector, 81
Schwarzschild observer, 333 topological space, 2
Schwarzschild radius, 318, 331 topology induced by a smooth
Schwarzschild spacetime, 9 structure, 11
sectional curvature, 346 topology on a set, 2
semi-Riemannian manifold, 85 torsion tensor field, 111
Index 483
trace of a symmetric (0, 2)-tensor

field, 105
transposition, 210
transverse curve, 196
vacuum, 294
vector field, 42
vector field along a smooth curve, 137
vector-valued function in R 3 , 225
velocity field of a perfect fluid, 298
velocity vector field along a curve, 139
volume, 251
volume form field, 250
volume form on a vector space with a
scalar product, 253
weak energy condition, 294

worldline, 13

Amol Sasane - A Mathematical Introduction To General Relativity

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

Amol Sasane - A Mathematical Introduction To General Relativity

Uploaded by

Copyright:

Available Formats

A Mathematical

A Friendly Approach to Complex Analysis

Plain Plane Geometry

A Friendly Approach to Functional Analysis

1S, World Scientific

Library of Congress Control Number: 2021038220

British Library Cataloguing-in-Publication Data

Cover image: Spacetime Pythagoras Theorem

A MATHEMATICAL INTRODUCTION TO GENERAL RELATIVITY

ISBN 978-981-124-377-6 (hardcover)

For any available supplementary material, please visit

The aim of the book is to give a mathematical presentation of the theory

The entire book's content can be summarised as follows:

(gcornc t, v J Ric - ½sg +Ag= 81rT ( nrn u c , J

r(p,p') := f g('Y(t))(v.,,,.,,(t), v.,,,.,,(t)) dt.

3 0f which an English translated edition is underway.

Continuously differentiable functions on U will be said to be C 1 on U. This

Physical constant I Symbol I Value

Notation and terminology xi

6. Levi-Civita connection 107

7. Parallel transport 137

9.4 Tidal forces 198

10. Form fields 209

11. Integration 235

12. Minkowski spacetime physics 261

13. Matter 289

14. Field equation 305

15. Black holes 327

16. Cosmology 345

As mentioned in the Preface, spacetime is, on the set-theoretic level, a

1.1 Charts and atlases

Before we introduce the notion of a manifold, we explain what we mean

Definition 1.1. (Topological space).

Example 1. 1. (Euclidean space !Rm).

Definition 1.2. (Chart).

Example 1.3. (Affine space).

Then (Un, 'Pn) and (Us, 'Ps) are charts on S 2 . ◊

Exercise 1.3. Show that (JE., x >-+ x 3 ) is a chart on lE..

Exercise 1.4. (Cylinder).

Definition 1.3. (Atlas) .

Fig. 1.2 Compatible charts (U;,cp; ) and (U1 ,cp1 ).

Example 1.6. (A non-atlas). With m = 1 in Example 1.2, we get the

The previous example motivates the following definition.

Fig. 1.3 Change of coordinates.

Exercise 1. 7. Prove that compatibility is an equivalence relation on the collection

Thus to specify a manifold1, we should work with compatible atlases (so

Definition 1.5. (Smooth manifold, dimension of a manifold).

Example 1.8. (Sphere). Consider the sphere as a smooth manifold with

(0,n) x (0,2n) 3 (0,¢) >-+ ((sin0)(cos¢), (sin0)(sin¢),cos0) E 8 2 c JR 3

is injective, and hence a bijection onto its image

U = 8 2\{(x,y,z) E JR3 : y = 0 and x ~ 0}.

We now discuss four 'spacetimes' by just looking at the underlying smooth

Example 1.9. (Minkowski spacetime). Let V be a 4-dimensional real vec-

Example 1.10. (Cylindrical spacetime). Let M = JR x 8 1 with the atlas

Exercise 1.9. (Product of smooth manifolds). Let M be an m-dimensional smooth

Example 1.12. (Schwarzschild spacetime). Let m > 0, and J := (2m, oo ).

1.2 Topology on a smooth manifold

Proposition 1.1. Definition 1.6 of an open set is well-defined, that is, it

=LJ "Pi(U nUin ½)- (1.1)

'Pi(( LJ ½) n Ui) = LJ 'Pi(½ n Ui),

Definition 1.7. (Topology induced by a smooth structure).

Remark 1.1. Often in the literature, a smooth manifold is defined by first

Proof. Let i E J. As (Ui, 'Pi) is a chart, we know that 'Pi(Ui) is open in