M. Spivak - Calculus On Manifolds PDF

Michael Spivak
Brandeis University
Calculus on Manifolds
A MODERN APPROACH 1'0 CLASSICAL 1'HEOREMS
OF ADVANCED CALCULUS
•
�A'"C'..
.. o
:I
o ..
o �
1- "
".
ADDISON-WESLEY PUBLISHING COMPANY

The Advanced Book Program
Reading, Massachusetts • Menlo Park, California • New York
Don Mills, Ontario • Wokingham, England • Amsterdam • Bonn
Sydney • Singapore • Tokyo • Madrid • San Juan • Paris
SeouI • Milan • Mexico City • Taipei
A Modem Approach to Classical Theorems of Advanced Calculus
Copyright © 1965 by Addison-Wesley Publishing Company
Library of Congress Card Catalog Number 66-10910

AlI rights reserved
The manuscript was put imo production on Aprii 21,1965;

Manufactured in the United States of America
this volume was published on October 26, 1965
ISBN 0-8053-9021-9
2425262728-CRW-9998979695
Twenty-fourth printing, January 1995
Editors' Foreword
Mathematics has been expanding in ali directions at a fabulous

rate during the past half century. New fields have emerged,
the diffusion into other disciplines has proceeded apace, and
our knowledge of the classical areas has grown ever more pro
found. At the same time, one of the most striking trends in
modern mathematics is the constantly increasing interrelation
ship between its various branches. Thus the present-day
students of mathematics are faced with an immense mountain
of material. In addition to the traditional areas of mathe
matics as presented in the traditional manner-and these
presentations do abound-there are the new and often en
lightening ways of looking at these traditional areas, and also
the vast new areas teeming with potentialities. Much of this
new material is scattered indigestibly throughout the research
journals, and frequently coherently organized only in the
minds or unpublished notes of the working mathematicians.
And students desperately need to learn more and more of this
material.
This series of brief topical booklets has been conceived as a
possible means to tackle and hopefully to alleviate some of
v
vi Editors' Foreword
these pedagogical problems. They are being written by active

research mathematicians, who can Iook at the latest develop
ments, who can use these developments to clarify and con
dense the required material, who know what ideas to under
score and what techniques to stress. We hope that they will
also serve to present to the able undergraduate an introduction
to contemporary research and problems in mathematics, and
that they will be sufficiently informaI that the personal tastes
and attitudes of the leaders in modern mathematics will shine
through clearly to the readers.
The area of differential geometry is one in which recent
developments have effected great changes. That part of
differential geometry centered about Stokes' Theorem, some
times called the fundamental theorem of multivariate calculus,
is traditionally taught in advanced calculus courses (second or
third year) and is essential in engineering and physics as well
as in several current and important branches of mathematics.
However, the teaching of this material has been relatively
little affected by these modern developments; so the mathe
maticians must relearn the material in graduate school, and
other scientists are frequently altogether deprived of it. Dr.
Spivak's book should be a help to those who wish to see
Stoke's Theorem as the modern working mathematician sees
it. A student with a good course in calculus and linear algebra
behind him should find this book quite accessible.
Robert Gunning
Hugo Rossi
Princeton, New Jersey

Waltham, MassachuseUs
August 1965
Preface
This little book is especially concerned with those portions of

"advanced calculus" in which the subtlety of the concepts and
methods makes rigor difficult to attain at an elementary level.
The approach taken here uses elementary versions of modern
methods found in sophisticated mathematics. The formaI
prerequisites include only a term of linear algebra, a nodding
acquaintance with the notation of set theory, and a respectable
first-year calculus course (one which at least mentions the
least upper bound (sup) and greatest lower bound (inf) of a
set of real numbers). Beyond this a certain (perhaps latent)
rapport with abstract mathematics will be found almost
essential.
The first half of the book covers that simple part of ad
vanced calculus which generalizes elementary calculus to
higher dimensions. Chapter 1 contains preliminaries, and
Chapters 2 and 3 treat differentiation and integration.
The remainder of the book is devoted to the study of curves,
surfaces, and higher-dimensional analogues. Here the modern
and classical treatments pursue quite different routes; there are,
of course, many points of contact, and a significant encounter
vii
viii Preface
occurs in the last section. The very classical equation repro

duced on the cover appears also as the last theorem of the
book. This theorem (Stokes' Theorem) has had a curious
history and has undergone a striking metamorphosis.
The first statement of the Theorem appears as a postscript
to a letter, dated July 2, 18.50, from Sir William Thomson
(Lord Kelvin) to Stokes. It appeared publicly as question 8
on the Smith's Prize Examination for 1854. This competitive
examination, which was taken annually by the best mathe-
'matics students at Cambridge University, was set from 1849 to
1882 by Professor Stokes; by the time of his death the result
was known universal1y as Stokes' Theorem. At least three
proofs were given by his contemporaries: Thomson published
one, another appeared in Thomson and Tait's Treatise on
Natural Philosophy, and Maxwell provided another in Elec
tricity and Magnetism [13]. Since this time the name of
Stokes has been applied to much more generaI results, which
have figured so prominently in the development of certain
parts of mathematics that Stokes' Theorem may be con
sidered a case study in the value of generalizatioll.
In this book there are three forms of Stokes' Theorem.
The version known to Stokes appears in the last section, along
with its inseparable companions, Green's Theorem and the
Divergenee Theorem. These three theorems, the classical
theorems of the subtitle, are derived quite easily from a
modern Stokes' Theorem which appears earlier in Chapter 5.
What the classical theorems state for curves and surfaees, this
theorem states for the higher-dimensional analogues (mani
folds) which are studied thoroughly in the first part of Chapter
5. This study of manifolds, which eould be justified solely on
the basis of their importance in modern mathematics, actually
involves no more cffort than a eareful study of curves and sur
faces alone would require.
The reader probably suspects that the modern Stokes'
Theorem is at least as difficult as the classical theorems
derived from it. On the contrary, it is a very simpie con
sequence of yet another version of Stokes' Theorem; this very
abstract version is the final alld main result of Chapter 4.
Preface ix
It is entirely reasonable to suppose that the difficulties so far

avoided must be hidden here. Yet the proof of this theorem
is, in the mathematician's sense, an utter triviality-a straight
forward computation. On the other hand, even the statement
of this triviality cannot be understood without a horde of
difficult definitions from Chapter 4 . There are good reasons
why the theorems should aH be easy and the definitions hard.
As the evolution of Stokes' Theorem revealed, a single simple
principle can masquerade as several difficult results; the proofs
of many theorems involve merely stripping away the disguise.
The definitions, on the other hand, serve a twofold purpose:
they are rigorous replacements for vague notions, and
machinery for elegant proofs. The first two sections of
Chapter 4 define precisely, and prove the rules for manipulat
ing, what are classically described as "expressions of the form"
p dx + Q dy + R dz, or P dx dy + Q dy dz + R dz dx . Chains,
defined in the third section, and partitions of unity (already
introduced in Chapter 3) free our proofs from the necessity of
chopping manifolds up into small pieces; they reduce questions
about manifolds, where everything seems hard, to questions
about Euclidean space, where everything is easy.
Concentrating the depth of a subject in the definitions is
undeniably economical, but it is bound to produce some
difficultics for the student. I hope the reader will be encour
aged to learn Chapter 4 thoroughly by the assurance that the
results will justify the effort: the classical theorems of the last
section represent only a few, and by no means the most im
portant, applications of Chapter 4; many others appear as
problems, and further developments will be found by exploring
the bibliography.
The problems and the bibliography both deserve a few
words. Problems appear after every section and are num
bered (like the theorems) within chapters. I have starred
those problems whose results are used in the text, but this
precaution should be unnecessary-the problems are the most
important part of the book, and the reader should at least
attempt them allo It was necessary to make the bibliography
either very incomplete or unwieldy, since half the major
" Preface
branches of mathematics could legitimately be recommended

as reasonable continuations of the material in the book. I
have tried to make it incomplete but tempting.
Many criticisms and suggestions were offered during the
writing of this book. I am particularly grateful to Richard
Palais, Rugo Rossi, Robert Seeley, and Charles Stenard for
their many helpful comments.
I have used this printing as an opportunity to correct many
misprints and minor errors pointed out to me by indulgent
readers. In addition, the material following Theorem 3-1 1
has been completely revised and corrected. Other important
changes, which could not be incorporated in the text without
excessive alteration, are listed in the Addenda at the end of the
book.
Michael Spivak
WaUham, Massachusetts
March 1968
Conlenls
Editors' Foreword, v
Preface, vii
1. Functions on Euclidean Space 1

l
5
NORM AND INNER PRODUCT,
11
SUBSETS OF EUCLIDEAN SPACE,
FUNCTIONS AND CONTINUITY,
2. Differentiation 15
15
19
BASIC DEFINITIONS,
25
BASIC THEOREMS,
30
PARTIAL DERIVATIVES,
34
DERIVATIVES,
40
INVERSE FUNCTIONS,
44
IMPLICIT FUNCTIONS,
NOTATION,
xi
xii Contents
3. lntegration
46
46
50
BASIC DEFINITIONS,
52
MEASURE ZERO AND CONTENT ZF;RO,
56
INTEGRABLE FUNCTIONS,
'
63
FUDINI S THEOREM,
67
PARTITIONS OF UNITY,
GHANm; OF VARIABLE,
4. lntegration on Chains
75
75
86
ALGJo;BRAIC PRELIMINARH;S,
97
FIF.LDS AND FORMS,
100
m;OMETRIC PRl';LIMINARIES,
'l'Hl'; FUNDAfiH:NTAL THl';OREM OF CALCULUS,
5. lntegration on Manifolds 109

109
115
MANIFOLDS,
122
FH:LDS AND FORMS ON MANIFOLDS,
'
126
STOKES THEOREM ON MANIFOLDS,
134
THE VOLUME ELTèMENT,
THF. CLASSICAL THEORl';MS,
Bibliography, 139
Index, 14-1
1
Functions on Euclidean Space
NO R M AND INNER P R O D UC T
Euclidean n-space Rn is defined as the set of all n-tuples

(.-rl, . . . ,.Tn) of real numbers .Ti (a "l-tuple of numbers" is
just a number and R I = R, the set of all real num bers). An
element of Rn is often called a point in Rn, and R" R2, R3 are
often called t he line, the piane, and space, respectively. If.T
denotes an element of Rn, t hen x is an n-tuple of Ilumbers, the
it h olle of which is denoted .Ti; thus we can write
.1: = (Xl, ... ,xn).
A point 111 Rn is frequently also called a vector in Rn,
because Rn, with :r + y = (Xl + y l, . . .,xn + yn) and
ax = (axl, ...,axn), as operations, is a vector space (over
the real numbers, of dimension n) . In this vector space there
is the notion of the length of a vector x, usually called the
norm Ix!of x and defined by !x! = Y(XI)2 + ... + (xn)2.
If n = 1, then !x! is the usual absolute value of:1'. The rela
tion between the norm and the vector space structure of Rn is
very important.
1
2 Calculu8 on Manifold8
1-1 Theorem. If x,y E R n and a E R, then
(1) Ixl � O, and Ixl = O iJ and only if x O. =
(2) !�f=lXiyi! ::; Ix! '!y!; equality holds if and only if x and y
are linearly dependent.
(3) !x + y! ::; Ixl + Iyl.
(4 ) laxl = !al . Ixl.
Proof
(1) is left to the reader.

(2) If x and y are linearly dependent, equality clearly holds.
If not, then >..y - x � O for all >.. E R, so
n
O ..y - X ! 2 = I (>..yi - Xi) 2
i-l
n n n
= >.. 2 I (yi) 2 - 2>" I xiyi + I (xi) 2.
i-l i-l i-l
Therefore the right side is a quadratic equation in >.. with no
reaI solution, and its discriminant must be negative. Thus
n n n
4 ( I X iyiY - 4 I (xi) 2 . I (yi) 2 -< O.
i-l i-l i-l
(3) I x + yl 2 � i=l (Xi + yi) 2
=
= � i_ l (Xi) 2 + �i= 1 (yi) 2 + 2�i= l X iyi
::; IX! 2 + ly! 2 + 2 ! xl · ! y l by (2)

= (Ixl + ly!) . 2
(4) laxl = v'.�
-----,-,
i= 1 (axi) = v'a 2�f= 1 (xi) 2 = lal . Ix/. I
2
The quantity � i= lX iyi which appears in (2) is

called the
inner product of x and y and denoted (x,y).
The most
important properties of the inner product are the following.
1-2 Theorem. If x, Xl, X 2 and y, yt, Y 2 are vectors in R n
and a E R, then
(1) (x ,y) = (y,x) (symmetry).

Function8 on Euclidean Space
(2) (ax,y) (x,ay) a (x ,y)
= = (bilinearity).
(XI + X 2 , y) = (x I ,y) + (X 2 ,y)
(x, YI + YÙ (x, Y I ) + (X, Y 2 )
=
(3) (x,x) � O, and (x, x) O if and = (positive definiteness) .

only if x = 0
(4) Ixl = V(x,x).
Ix + Y1 2 - I x - yl 2 (polarization identity).
(5) (x,y) =
4
Proof
(1 ) (x,y) = 2:f_I Xiyi = 2:i_ l yixi = (y,x).
(2) By (1 ) it suffices to prove
(ax,y) = a(x,y),
(XI + X 2 , y) = (xI,y) + (X 2,y).
These follow from the equations
" n
(ax,y) =
L (axi)yi = a L x'y' = a(x,y),
i-I
n i-I n n
(XI + X 2 , y) =
L
i- l
(X l i + X 2i)yi = L
1rt:2 1
x!iyi + L
i-l
X 2V
= (XI ,y) + (X 2 ,y).
(3) and (4) are Ieft to the reader.
Ix + yl 2 - Ix - yl2
(5)
4
= f(x + y, x + y ) - (x - y, x - y») by (4)
f( x,x) + 2 (x,y) + (y,y) - «x,x) - 2 (x ,y) + (y,y» ]
=
= (x,y ). I
We conclude this section with some important remarks

about notation. The vector (O, ... ,o) will usually be
denoted simpIy o. The usual basÌs of R" is e I , . . . ,en ,
where e. (O, . . . ,l, . . . ,O), with the 1 in the ith pIace.
=
If T: Rn -. Rm is a Iinear transformation, the matrix of T with

respect to the usual bases of Rn and Rm is the m X n matrix
A = (aii) , where T(ei) = 2:i_Iaiiej -the coefficients of T (ei)
., Calculus on Manifolds
appeal' in the ith column of the matrix. If S: Rm � HP has

the p X m matrix B, then S o T has thc p X n matrix BA
[here S o T(x) = S(T(x» ; most books Oll linear algebra denote
S o T simply ST) . To find T(x) one computes the m X 1
matrix
t hen T(x) = (yl, ... ,ym). One notational convention

greatly simplifies many formulas: if x E Rn and y E Rm, then
(x,y) denotes
l-I. * Prove that Ixl � 4�1 Ixil.

1-2. When does equality hold in Theorem 1-1 (a)?
P roblem.,.
Hint: Re-examine
the proof; the answer is not "when x and y are linearly depend-
1-3. Prove that Ix - yl � Ixl + Iyl. When does equality hold?

ent."
1-4. Prove that I lxl - Iyl l � Ix - y

l-S. The quantity Iy - xl is ealled the distance between x and y.
l·
�
Prove and interpret geometrieally the "triangle inequality":
Iz - x l Iz - yl + Iy - xl·
�
1-6. Let f and O be integrable on [a,b].
(a) Prove that I Hf 01 (f�f)t. (H02)!.
' Hint: Consider
separately the eases O f�(f
= - ÀO) 2 for some À E R and O <
f�(f ÀO) 2 for ali À E R.
f
-
(b) If equality holds, must = ÀO for some À E R ? What if

f
and O are continuous?
1-7. A linear transformation T: R " -> R " i s nornt preserving if

(c) Show that Theorem 1-1 (2) is a speeial case of (a) .
IT(x)1 Ixl, and inner product preserving if (Tx,Ty) = (x,y).

=
(a) Prove that T is norm preserving if and only if T is inner

produet preserving.
(b) Prove that sueh a linear transformation T is 1-1 and T-l is
l-B. If x,y E R" are non-zero, the angle between x and y, denoted
of the same sorto
L(x,y), is defined as areeos «x,y)/lxl ·Iyl), whieh makes sense by

Theorem 1-1 (2). The Iinear transformation T is angle preserv
ing if T is l-l, and for x,y � O we have L(Tx,Ty) L(x,y).
=
Function8 on Euclidean Space 5
(a) Prove that if T is norm preserving, then T is angle pre

serving.
(b) If there is a basis Xl, ... ,Xn of Rn and numbers Xl, ... ,Xn
on\y ii ali I xii are equal.

sue h that TXi = XiX;' prove that T is angle preserving if and
(c) What are alI angle preserving T: Rn -> Rn?

2 2
1-9. If O :$ 8 < 11", let T: R -> R have the matrlx
•( cos 8, sin 8 .)
Show that T is angle preserving and if X F O, then L(x,Tx)
-sin 8, cos 8
l-IO. * If T: Rm -> Rn is a linear transformation, show that there is a

= 8.
number M such that IT(h)1 :$ Mlhl for h E Rm• Hint: Estimate

I T(h) I in terms of I hl and the entries in the matrix of T.
l-II. If x,y E Rn and z,w E Rm, show that «x,z),(y,w)} = (x,y) + (z,w)
and I (:r,z) \ vlxl2 + Iz12. Note that (x,z) and (y,w) denote
=
points in Rn+m.
1-12. * Let (Rn)* denote the dual space of the vector space Rn• If
=
X E Rn, define <Pz E (Rn) .. by <Pz(y) (x,y). Define T: Rn->
(R n) .. by T(x) <Pz. Show that T is a l-l linear transformation
and conclude that every <P E (Rn) is <Pr for a unique X E Rn.
..
=
1-13." If x,y E Rn, then X and y are called perpendicular (or orthog
oDal) if (x,y) -
O. If X and y are perpendicular, prove that
Ix + yl2 Ixl2 + \yI2.
=
S UB S ETS O F E UCLID E A N S P ACE
The closed intervaI [a,b] has a naturaI analogue in R2• This is

the elosed rectangle [a,b] X [c,d], defined as the collection of
alI pairs (x,y) with x E [a,b] and y E [c,d]. More general1y,
if A C Rm and B C Rn, then A X B C Rm+n is defined as
the set of alI (x,y) E Rm+n with x E A and y E B. In par
ticular, Rm+n = Rm X Rn. If A C Rm, Be Rn, and C C
RP, then ( A X B) X C = A X (B X C), and both of these
are denoted simply A X B X C; this convention is extended to
the product of any number of sets. The set [al,bI] X . . . X
[�,b ] C Rn is calIed a elosed rectangle in Rn, while the set
n
(al ,bI) X . . . X (a ,b ) C Rn is called an open rectangle.
More generalIy a setn nU C Rn is called open (Figure l-l )
if for each x E U there is an open rectangle A such that
x E A C U.
A subset C of Rn is elosed if Rn - C is open. For exam
pIe, if C contains onIy finiteIy many points, then C is ciosed.
6 Calculus on Manifolds
FI G UR E 1 -1
The reader should supply the proof that a closed reetangle in

R n is indeed a closed set.
H A C Rn and x E Rn, then one of three possibilities must
hold (Figure 1-2):
l. There is an open reetangle B sueh that x E BeA.
2. There is an open reetangle B sue h that x E B C Rn A.
-
3. H B is any open reetangle with x E B, then B contains

points of both A and Rn - A.
2 •
FI G UR E 1 -2
Functions on Euclidean Space 7
Those points satisfying ( 1) constitute the interior of A, those

satisfying (2) the exterior of A , and those satisfying ( 3) the
boundary of A. Problems 1-16 to 1-18 show that these terms
may sometimes have unexpected meanings.
It is not hard to see that the interior of any set A is open,
and the same is true for the exterior of A, which is, in fact, the
interior of Rn - A. Thus (Problem 1-14) their union is open,
and what remains, the boundary, must be closed.
A coHection 0 of open sets is an open cover of A (or, briefly,
covers A) if every point x E A is in some open set in thc
collection 0. For example, if 0 is the collection of ali open
intervals (a, a + 1) for a E R, then 0 is a cover of R. Clearly
no finite number of the open sets in 0 will cover R or, for that
matter, any unbounded subset of R. A similar situation can
also occur for bounded sets. If 0 is the collection of all open
intervals (l/n, 1 - l/n) for aH integers n > 1, then 0 is an
open cover of (0,1), but again no finite collection of sets in
o will cover (0, 1) . Although this phenomenon may not appear
particularly scandalous, sets for which this state of affairs
cannot occur are of such importance that they have received a
special designation: a set A is called cOlnpact if every open
cover 0 contains a finite subcollection of open sets which
also covers A.
A set with only finiteIy many points is obviously compact
and so is the infinite set A which contains ° and the numbers
l/n for all integers n (reason: if 0 is a cover, then ° E U for
some open set U in 0; there are onIy finiteIy many other points
of A not in U, each requiring at most one more open set).
Recognizing compact sets is greatly simplified by the follow
ing resuIts, of which only the first has any depth (i.e., uses any
facts about the real numbers) .
1-3 Theorem (Heine-Borel). The closed interval [a,b] is

compact.
Proof. If 0 is an open cover of [a,b], Iet

A = Ix: a ::; x ::; band fa,xl is covered by some finite number
of open sets in 01.
Il Calculus on Manifolds
Il
( .r .r
' )
FI G UR E 1 -3
Note that a E A and that A is clearly bounded above (by b).

We would like to show that b E A. This is done by proving
two things about a = least upper bound of A; namely, (1)
a E A and (2) b = a.
Since (9 is a cover, a E U for some U in (9. Then aH
points in some interval to the left of a are also in U (see Figure
1-3). Since a is the least upper bound of A, there is an x in
this interval such that x E A. Thus fa,xl is covered by some
finite number of open sets of (9, while [x,a] is covered by the
single set U. Hence fa , al Ìs covered by a finite number of open
sets of (9, and a E A. This proves (1).
To prove that (2) is true, suppose instead that a < b.
Then there is a point x' between a and b such that fa,x'] C U.
Since a E A, the interval fa,al is covered by finitely many
open sets of (9, while fa,x'] is covered by U. Hence x' E A,
con tradicting the fact that a is an upper bound of A. I
If B C Rm is compact and x E Rn, Ìt is easy to see that

l x l X BeRn+m is compact. However, a much stronger
assertion can be made,
1-4 Theorem. lj B is compact and (9 is an open cove1' oj

I x} X B, then there is an open set U C R n containing x such
thai U X B is covered by a finite numbe1' oj sets in (9.
Proof· Sin ce Ix} X B is compact, we can assume at the

outset that (9 is finite, and we nced only find the open set U
such that U X B is covered by (9.
For each Y E B the point (x,y) is in some open set W in (9,
Sin ce W is open, we have (x,y) E UY X Vy C W for some
open rectangle U y X V y' The sets Vy cover the compact set
B, so a finite numbcr V Yll ,V Yk also cover B. Let
U = U1I1 n . . , n Uyk' Then if (x',y') E U X B, we have
• . .
Functions on E1tclidean Space 9
u.,
FIG U R E 1 -4
y' E VYi for some i (Figure 1-4), and certainly x' E Uyi.
V
Hence (x' ,y') E UYi X Y,' which is contained in some W
in (9. I
1-5 Corollary. lf A C Rn and B C Rm are compact, then

AX Be Rn+m is compact.
Proof. If (9 is an open cover of AX B, then (9 covers (x l X B

for each x E A . By Theorem 1-4 there is an open set Ux con
taining x such that U x X B is covered by finitely many sets
in (9. Since A is compact, a finite number UXI1 • • • ,UXn of
the Ux cover A . Since finitely many sets !il (9 cover each
UXi X B, finitely many cover all of A X B. I
1-6 Corollary. Al X . . . X Ak is compact if each Ai is.

In particular, a closed rectangle in Rk is compact.
lO Calculu8 on Manifolds
1-7 Corollary. A closed bounded subset oj Rn is compact.
(The converse is also true (Problem 1-20).)
Proof. If A CRn is closed and bounded, then A CB for
some closed rectangle B. If ° is an open cover of A, then °
together with Rn - A is an open cover of B. Hence a finite
number UI, . . . , Un of sets in 0, together with Rn - A per
haps, cover B. Then U 1 , ,Un cover A. I
• • •
Problems. 1-14. · Prove that the union of any (even infinite) number
of open sets is open. Prove that the intersection of two (and hence
of finitely many) open sets is open. Give a counterexample for
infinitely many open sets.
1-15. Prove that Ix E R n: \x - a\ < r I is open (see also Problem 1-27).
1-16. Find the interior, exterior, and boundary of the sets
Ix E Rn: \x\ 91
Ix E Rn: \x\ = 1 1
I x E Rn: each xi is rationall.
1-17. Construct a set A C [0, 1 ] X [0, 1] such that A contains at most

one point on each horizontal and each vertical line but boundary
A = [0,1 ] X [0, 1]. Hint: It suffices to ensure that A contains
points in each quarter of the square [0,1] X [0,1] and al so in each
sixteenth, etc.
1-18. If A C [0, 1] is the union of open intervals (a;,bi) such that each
rational number in (0, 1 ) is contained in some (ai,bi), show that
boundary A = [O, lJ - A .
1-19. · If A is a closed set that contains every rational number r E [0,1],
show that [0,1] C A.
1-20. Prove the converse of Corollary 1-7 : A compact subset of Rn is
closed and bounded (se e al so Problem 1-28).
1-21. · (a) If A is closed and x El A, prove that there is a number
d > ° such that \y - x\ � d for ali 1/ E A.
(b) If A is closed, B is compact, and A () B = 0, prove that
there is d > ° such that \y - x\ � d for alI y E A and x E B.
this relation holds for x E U () B.

Hint: For each Il E B find an open set U containing b such that
(c) Give a counterexample in R2if A and B are closed but

neither is compact.
set D such that C C interior D and D C U .

1-22.· If U is open and C C U is compact, show that there is a compact
Functions on Euclidean Space II
FUNCTI O NS AND CO NTIN UIT Y
A function from Rn to Rm (sometimes called a (vector

valued) function of n variables) is a rule which associates to
each point in Rn some point in Rm; the point a functioll f
associates to x is denoted f(x). We write f: Rn Rm (read "f ----t
takes Rn into Rm" or "f, taking Rn into Rm," depending on con

text) to indicate that f(x) E Rm is defined for x E Rn. The
notation f: A Rm indicates that f(x) is defined only for x in
----t
the set A, which is called the domain of f. If B C A, we

define f(B) as the set of all f(x) for x E B, and if C CRm we
define f-I(C) = Ix E A: f(x) E CI. The notation f: A B ----t
indicates that f(A) CB.

A convenient representation of a function f: R2 R may ----t
be obtained by drawing a picture of its graph, the set of alI

3-tuples of the form (x,y,J(x,y»), which is actualIy a figure in
3-space (see, e.g., Figures 2-1 and 2-2 of Chapter 2).
If f,g: Rn R, the functions f + g, f - g, f· g, and f/g are
----t
defined precisely as in the one-variable case. If f: A ----t Rm

and g: B RP, where Be Rm, then the composition
----t
g of is defined by g of(x) g(f(x»; the domain of g of is

=
A rì r 1(B). If f: A Rm is 1-1, that is, if f(x) � f(y)

----t
when x � y, we define r 1 : f(A) Rn by the requirement that ----t
r 1(z) is the unique x E A with f(x) = z.

A function f: A Rm determines component functions
----t m
jl, . .. ,1 : A
m R by f(x) = (jl(x), ... ,tm(x» . If con
----t
versely, functions gl, . . . ,gm: A R are given, there

m ----t
is a unique function f: A Rm such that fi = gi, namely ----t
f(x) (gl(X), ...,gm(X». This function f will be denoted

=
(gl, ...,gm ), so that we always have f = Cf! , ... ,j).

If Rn Rn is the identity function, 1I"(x) x, then1l"i (x)
11": ----t = =
Xi ; the function 1I"i is called the ith projection function.

The notation lim f(x) = b means, as in the one-variable case,
x-+a
that we can get f(x) as close to b as desired, by choosing x suf

ficiently close to, but not equal to, a. In mathematical terms
this means that for every number � > O there is a number
� > O such that I f(x) - bi < e for all x in the domain of fwhich
12 Calculus on Mam'folds
satisfy O < I x - al < 8. A function f: A -- Rm is called con

tinuous at a E A if lim f(x ) = f(a), and i is simply called con-
tinuous if it is continuous at each a E A . One of the pleasant
surprises about the concept of continuity is that it can be
defined without using limits. It follows from the next theorem
that i: R" -- Rm is continuous if and only if r1(U) is open
whenever U C Rm is open; if the domain of i is not all of R", a
slightly more complicated condition is needed.
1-8 Theorem. If A C Rn, a function f: A Rm is contin
__
uous if and only if for every open set U C Rm there is some open
set V C R n such that r 1 ( U) = V (ì A.
Proof. Suppose f is continuous. If a E r1 ( U), then

f(a) E U. Since U is open, there is an open rectangle B with
i(a) E B C U. Since f is continuous at a, we can ensure
that f(x) E B, provided we choose x in some sufficiently
small rectangle C containing a. Do this for each a E rI(U)
and let V be the union of all such C. ClearIy rI(U) =
V (ì A. The converse is similar and is left to the reader. I

The following consequence of Theorem 1-8 is of great
importance.
1-9 Theorem. Ii f: A -- Rm is continuous, where A C Rn,
and A is compact, then i(A) C Rm is compact.
Proof. Let f) be an open cover of f(A). For each open set

U in f)
there is an open set Vu such that rI(U) Vu (ì A. =
The collection of alI Vu is an open cover of A . Since A is

compact, a finite number VUI1 • ,VUn cover A. Then
• .
UI , • ,un cover f(A) . I

. •
If f: A -- R is bounded, the extent to which f fails to be

continuous at a E A can be measured in a precise way. For
8 > O let
M(a,f,8) = sup { f(x) : x E A and I x - a l < 8},
m(a,J,8) inf { f(x) : x E A and Ix - a l < 8}.
=
Functions on Euclidean Space 13
The oscillation o (f,a) of f at a is defined by o(f,a) =

hO
lim [M(a,j,t5) - m(a,j, t5 »). This limi t always exists, since
M(a,j,t5) - m(a,j, t5) decreases as t5 decreases. There are two
important facts about o(f,a) .
1-10 Theorem. The bounded function f is continuous at a if
and only if o(f,a) O. =
Proof. Let f be continuous at a. For every number E > O

we can choose a number t5 > O so that If(x) - f(a) I < E for
alI x E A with I x - a l < 15; thus M(a,j, t5) - m(a,j, t5) � 2e.
Since this is true for every E, we have o(f,a) = O. The con
verse is similar and is left to the reader. I
1-11 Theorem. Let A C Rn be closed. lf f: A R is any -+
bounded function, and e > O, then I x E A: o(f,x) � El is

closed.
Proof. Let B = I x E A: o(f,x) � El. We wish to show
that Rn - B is open. If x E Rn - B, then either x €i A
or else x E A and o(f,x) < E. In the first case, since A is
closed, there is an open rectangle C containing x such that
C C Rn - A C Rn - B. In the second case there is a
15 > O such that M (x,j,t5) - m (x,j,t5) < E. Let C be an open
rectangle containing x such that I x - y l < 15 for aH y E C.
Then if y E C there is a 151 such that I x - z l < t5 for all z
satisfying I - y l < 151 . Thus M(y, J, t51) - m(y, J,t5 1) < E, and
z
consequently o(y, J) < E. Therefore C C Rn - B. I

Problema.
x-+o
-> a ..
ii and only ii lim fi(x) bi for i l, . . . ,m.

1-23. If f: A Rm and E A, show that Hm f(x) b
%-+0
_ =
1-24. Prove that f : A ..... R m is continuous at a if and only if each f is.

1-25. Prove that a linear transformation T: Rn ..... R m is continuous.
{ (x,y) E R 2: x > ° and ° < y < X2}.

Hint: Use Problem 1-10.
1-26. Let A =
(a) Show that every straight line through (0,0) contains an

interval around (0,0) which is in R 2 - A.
(b) Define f: R2 ..... R by f(x) ° if x Et: A and f(x)
= 1 if =
x E A. For h E R2 define glI: R -> R by g,,(t) - f(eh). Show

that each glI is continuous at O, but f is not continuous at (0,0).
14 Caleulm on Manifolda
1-27. Prove t h at {x E Rn: Ix - al < TI is open by considering the
function f: Rn ---+ R with f(x) = Ix - al.
1-28. If A C Rn is not closed, show that there is a continuous function
f: A ---+ R which is unbounded. Hint: If x E Rn - A but
x E interior (Rn - A ) , let f(y) l/ly - xl.
1-29. If A is compact, prove that every continuous function f: .ti
-=
---+R
takes on a maximum and a minimum value.
1-30. Let f: [a,b] ---+ R be an increasing function. If X l, . . . ,X" E
[a,bl are distinct, show that �;_lo(f,Xi) < f(b) - f(a).
2
Difjerentiation
B A SIC D EFINI TIONS
Recall that a function f: R R is differentiable at a E R if

--+
there is a number f'(a) such t,haI,

(1) lim f(a + h)h - f(a)
h--+O
= f'(a),
This equation eertainly makes 110 sense in the generai case of a

funct.ion f: Rn Rm, but can be l'eformulated in a way that
--+
does. If À: R R is the linear transformation dcfined by

--+
À(h) f'(a) . h, then equatioll (1) is cql1ivalent to

=
( 0,
(2) lim f(a + h) -hf(a) - À h)
;'--+0
=
Equation (2) is often interpreted as saying that À + f(a) is a

good approximatioll to f at a (see Problem 2-9). Henceforth
we focus our attention on the linear tl'ansformation À and
reformulate the definition of diffel'entiabilit,y as folIows,
15
16 CalCUlU8 on Mani/old8
funetion f: R R is differentiable at a E R if there is a
A ---+
linear transformation X: R ---+ R sueh that

lim f(a + h) -hf(a) - X(h) = O.
.\-+0
In this form the definition has a simple generalization to

higher dimensions:
A funetion f: R n Rm is differentiable at a E Rn if there
---+
is a linear transformation X: Rn ---+Rm such that

lim If(a + h) -Ihlf(a) - X(h)1 o.
.\-+0
=
Note that h is a point of Rn and f(a + h) - f(a) - X(h) a

point of Rm, so the norm signs are essential. The linear trans
formation X is denoted Df(a) and called the derivative of f at
a . The justifieation for the phrase "the linear transformation
X" is
2-1 Theorem. If f: R n ---+Rm is differentiable at a E R n

there is a unique linear lransformation X: Rn ---+Rm such thai
lim I f(a + h) -Ihlf(a) - X(h)1 o.

11-+0
=
Proof. Suppose p. : Rn ---+Rm satisfies

lim If(a + h) -Ihlf(a) - p. (h) I O.
11-+0
=
If d(h) f(a + h) - f(a) , then

=
lim IX(h) Ihl

- p.(h) I
lim IX(h) - d(h) Ihl+ d(h) - p. (h) I
=
11-+0 11-+0
< lim IX(h) Ihl- d(h) I + lim I d(h) Ihl- p. (h) I

-
10-+0 11->0
=
o.
If x E Rn, then tx ---+O as 1---+0 . Hence for x � O we have

IX(lx) - p. (tx) I IX(x) - p. (x) I .
O lim
HO
=
=
I txl Ixl
Therefore X(x) p.(x). I
=
Differentiation 17
We shall iater diseover a simple way of finding Df(a). For

the moment let us eonsider the funetion f: R2 R defined by �
f(x,y) sin x. Then Df(a,b) = X satisfies X(x,y) = (eos a) . x.

=
To prove this, note that

l. If(a + h, b + k) - f(a,b) - X(h,k) 1
1m
(h.k)--+O I (h,k) I
.
hm Isin (a + h) - I (hsin,ka) I - (eos a) . hl .

(h,k)--+O
Sinee sin'(a) eos a, we have

=
l. Isin(a + h) - sinh a - (eos a) . h l O.

�
=
h I l
Sinee l(h,k)1 � Ihl, it is also true that
lim I sin(a + h) - sin a - (eos a) . h l
hO I (h,k) l
=
o.
It is often eonvenient to eonsider the matrix of Df(a) :

Rn Rm with respeet to the usual bases of Rn and Rm•
�
This X n matrix is ealled the Jacobian matrix of f at a,

m
and denoted f'(a) . If f(x,y) = sin x, then f'(a,b) = (eos a, O).

If f: R R, then f'(a) is a 1 X 1 matrix whose single entry
�
is the number whieh is denoted f'(a) in elementary ealeulus.

The definition of Df(a) eould be made if f were defined only
in some open set eontaining a. Considering only funetions
defined on Rn streamlines the statement of theorems and
produees no real loss of generality. It is eonvenient to define
a function f: Rn Rm to be difTerentiable on A if f is differ
�
entiable at a for each a E A . If f: A Rm, then f is called

�
differentiable if f can be extended to a differentiable function

on some open set containing A .
a E Rn, the n i t i s conti nuous at a. Hint: Use Pr oble m 1-10.

Problems. 2-1.* Pr ove that i f f: Rn ---+ R m i s differe ntiable at
2-2. A function f: R 2 ---+ R is independent of the second variable if

for e ac h x E R we have f(X,Yl) f(X ,Y2) for alI Yl,Y2 E R. Show
=
that f is inde pe nde nt of the se cond var iable if and only if there is a
func tion g: R ---+ R such that f(x,y) ... g(x ). What is f'(a ,b) in
ter ms of g'?
18 CalCUlU8 on Ma nifolds
2-3. De fine whe n a function f: R 2 ---> R is inde pe nde nt of the first varia
ble and fi ndl'(a,b ) for such f. Which functions are inde pe nde nt of
the fir st var iable and also of the se cond variable?
2-4. Le t g be a continuous re al-value d funct ion on the unit circle
Ix E R2 : Ixl l} such that g(O,l) ... g(l,O) ... ° and g( x)
= -
g(x) De fine f: R2 ---> R by

- .
x � 0,
x
O. -
( a ) I f x E R2 and h: R -+ R is de fined by h(t) = f(tx), show tha t

h is diffe re ntiable.
( b) Show that f is not differe ntiable at (0,0) unle ss g O.
to be ° by consider ing
=
Hint: Fir st show that Df(O,O) would have

(h,k) with k ° and the n with h ... O.
=
{:
2-5. Le t f: R 2 ---> R be de fine d by
xlyl (x,y) � 0,
f(x,y) X 2 + y2
o.
=
(x,y) =
Show that f is a function of the kind considere d in Pr oble m 2-4,

so that f is not differe ntiable at (0,0).
2-6. Le t f: R 2 R be de fine d by f(x,y)
---> Vlxyl. H how that f is not
=
2-7. Le t f: Rn---> R be a function such that If(x)1 � Ix12. Show that

differe ntiable at (0,0).
f is differe ntiable at O.
2-8. Le tf: R---> R 2 . Pr ove that f is diffe re ntiable at a E R if and only
if p andf2 are , and that in this case
l'(a ) =
(P)/(a»).
/(a)
(f2 )
2-9. T wo functions f,g: R ---> R are equal up to nth order at a if
f(a + h) - g(a + h)
h_O
l im = O.
hn
( a) Show that f is differe ntiable at a if and only if there is a
function g of the for m g(x) = ao + al(x a) such th atf and g are
-
e qual up to fir st or der at a.

( b) If l'(a), . ,j<n)(a ) e xist, show that f and the fune tion g
de fine d by
g(x) � f(i) (a )
� -li - (x
.
i-O
= - a )'
Dijferentiation 19
are e qual up to nth orde r at a. Bint: The limit
fex) -
2:
n-l t'il (a)
-.- (x - a)'
.
'--'0_____
t.,
lim ___
(x - a)n
may be e valuate d by L' Hospital' s rule .
B A S I C THEO R EM S
2-2 Theorem (Chain Rule). Iff: Rn ---+ Rm is dijJerenti-

able at a, and g: Rm ---+ RP is dijJerentiable at f(a), then the
composition g o f: Rn---+ RP is dijJerentiable at a, and
D(g o f) (a) = Dg(f(a» o Df(a).

Remark . This equation can be written
(g o f)' (a) = g'(f(a» . f'(a).
If m = n = p = 1 , we obtain the old chain rule.
Proof. Let b = f(a), let X = Df(a) and let J.L , = Dg(f(a».
If we defÌne
(1) <p (x) f(x) - f(a) - X (x - a),
=
(2) if;(y) g(y) - g(b) - J.L( Y - b),

=
(3) p (x) = g o f(x) - g o f(a) - J.L o X(x - a) ,

then
(4) lim IxI<p-(x)alI O, =
x-+a
(5) lim lyI if;-(y)blI O '

=
y-+b
and we must show that
Ip(x) I
lim I x - al
= o.
x-+a
�ow
p (x) = g(f(x» - g(b) - J.L (X(x - a»
= g(f(x» - g(b) - J.L (f(x) - f(a) - <p(x» by (1)
=
[g(f(x» - g(b) - J.L(f(x) - f(a»] + J.L (<p(x»
=
if; (f(x» + J.L( O it follows from (5) that for some � > O we have
Ilf (f(x)) I < elf(x) - bi if V(x) - bi < �,
which is true if I x - al < � l , for a suitable 5r. Then
I lf (f(x)) I < eV(x) - bi
= el cp(x) + À (x - a) 1
:::; elcp (x)I + eMlx - al
for some M, by Problem 1-10. Equation (6) now follows
easily. I
2·3 Theore m
(1 ) If f: R n - R m is a constant function (that is, if for some
y E Rm we have f(x) = y for all x E Rn), then
Df(a) = O.
(2) If f: Rn - Rm is a linear transformation, then

Df(a) = f.
(3) If f: Rn - Rm, then f is differentiable at a E Rn if and
only if each f is, and
Df(a) = (Dfl (a) , . . . ,Dr(a)).
Thus l'(a) is the m X n matrix whose ith row is (f)'(a) .
(4) If s: R2 - R is defined by s(x,y) x + y, then =
Ds(a,b) = s.
( 5) If p: R2 - R is defined by p(x,y) = x . y, then
Dp(a,b) (x,y) = bx + ay.
Thus p'(a,b) = (b,a).
Differentiation 21
Proof
(1) lim If(a + h) Ihl- f(a) - 0 1 = lim I y - Ihly - 0 1

h-+O h-+O
= o.
(2) lim I!(a + h) -Ihlf(a) - f(h) I

h-+O
= l·
If(a) + f(h) - f(a) - f(h} I =
O
h � Ihl .
(3) H each f is differentiable at a and

À = Wj1 (a) , . . . ,Dr (a» ,
then
f(a + h) - f(a) - À(h)
= (j1 (a + h) - j 1 (a) - Dj1 (a) (h), . . . ,
rea + h) + rea) - Dfm(a)(h» .
Therefore
r If(a + h) - f(a) - À(h) I
h� Ihl
H, on the other hand, f is differentiable at a, then f =

is differentiable at a by (2) and Theorem 2-2.
1f'i o f
(4) follows from (2).
(5 ) Let À(x , y ) = bx + ay Then .
l1. m I p(a + h, b + k) - p(a,b) - À(h,k) 1

I (h,k) I
<h,k)-+O
= lim �.
(h,k)
<h,k)-+O I I
Now
<
Ihk l - { II hk ll 22 if I kl � Ihl ,
if I h l � I kl ·
Hence I hk l � Ihl 2 + I k 1 2• Therefore
so
lim �
(h.k)-+O
I (h,k) I
= O. I
2-4 Corollary. Ii i,g: R" - R are differentìable at a, then

D(f + g)(a) = Di(a) + Dg(a) ,
D(f ' g) (a) = g(a)Di(a) + i(a) Dg(a).
Ii, moreover, g(a) ;;é 0, then
g(a)Di(a) - i(a)Dg(a) .
D(fjg) (a) =
[g(a}p
Proof. We wiIl prove the first equation and leave the others
to the reader. Since i + g = s o (f,g), we have
D(f + g) (a) = Ds(f(a),g(a» o D(f,g)(a)
= s o (Di(a) ,Dg(a»
= Di(a) + Dg(a). I
We are now aHsured of the differentiability of those functions

i: Rn - Rm, whose component functions are obtained by
addition, multiplicatioll, division, and composition, from the
functions 71"; (which are Iinear transformations) and the func
tions which we can already differentiate by elementary
calculus. Finding Di(x) or 1'(x), however, may be a fairly
formidable task. For example, let i: R 2 - R be defined by
i(x, y) = si (xy 2 ). Since i = sin o (71" 1 [7I" 2J 2) , we have
n
•
1'(a,b) = sin'(ab 2 ) . [b 2 (71" 1) '(a,b) + a( [7I" 2j 2 ) '(a, b) J

= sin'(ab 2) . [b 2 (7I"1)'(a,b) + 2ab(7I" 2)'(a,b)]
= (cos(ab 2» . [b 2 (1,0) + 2ab(0,1 )]
= (h 2 cos(ab 2), 2ab cos(ab 2 » .
Fortunately, we will soon discover a much simpler method of
computing 1'.
Problems. 2-10. Use the theorerns of this section to find f' for the
following:
Dijferentiation 23
(a) l(x ,lI,z) ... xv.

(b) l(x ,lI,z) = (XV, z) .
c( ) l(x ,lI) = sin (x sin 11) .
(d) l(x ,lI,z) sin (x sin(lI sin z) .
=
(e) l(x ,lI,z) = xv'.

xV+'.
(x + 11)'.
(f) l(x ,lI,z) =
(g ) l(x ,lI,z) =
(h) l(z,lI) = sin x( ll).

(i) I(z, y) = [sin (xy) ]00. 8 .
(i) I(x , y) = (sin(x y), sin (z sin y) , XV) .
2-1 1 . Find l' for the following (wher e g: R � R isc ontinuous) :
(a) I(x, y) - S :+Yg.
(b) I(x, y) = S:' vg.
(c ) I(x,y, z) S
=
O i n ( X oin(y o i n ,»
u g.
2-12. A func tion I: Rn X Rm �
RP is bilinear if for X,X1,X2 E R",
Y,Y1.1I2 E Rm, and a E R we have
I(ax,y) a l(x , y) = I(x a, y ),
l(x 1 + X2 ,y) = I(X 1,y) + I(X2,y),
�
l(x,Y1 + Y2) = I(X,Y1) + l(x,Y2).

(a) Pr ove that if I is bilinear , then
Hm I
I(h,k) 1 o. =
(h,k)-+O I (h,k) I
(b) Pr ove that DI(a,b) (x, y) I(a, y) + I(x,b ).

=
(c) Show that the for mula for Dp(a ,b) in Theor em 2-3 is a
spec ialc ase of (b).
2-13. Define I P: R" X R" � R by I P(x , y) = (x ,1I).
(a) Find D (IP) (a ,b) and (IP)' (a ,b) .
(b) If I, g : R � R" ar e differ entiable and h : R � R is defined by
h(t) (I(t ) ,g(t» , show that
=
h'(a) = (f'(a)T,g(a» + (l(a ) ,g'(a ) T).
(Note that f'(a ) is an n X l matr ix ; its tr anspose-I ' (a )T is a l X n

matrix, whic h we consider as a member of Rn.)
(c) If I: R � R" is diff er entiable and I I(t) I = 1 for ali t, show
that (f'(t)T, J(t» = o.
(d) Exhibit a differ entiable func tion I: R R such that the
func tion I I I defined by ifl (t) = II (t) I is not differ entiable.
�
2-14. Let E;, i = 1, . . . ,k be Euclidean spac es of var ious dimensions.

A func tion I: El X . . . X Ek � RP is c aIIed multilinear if
for eac hc hoic e of Xi E Eh j � i the func tion g: Ei � RP defined by
g(x) = I(X1, . . . ,X;_l, X,X;+ 1, . . . ,Xk) is a linear tr ansfor mation.
24 Calculua on Mani!olda
(a) If! is multilinear and i � j, show that for h = (hl, . . . ,hk),
with hl E El, we have
r
O.
1!(a1, . . . ,h"� . . . ,hi. . . . ,ak) I
h�O
=
Ihl
Hint: If g(x,y) = !(a1, . . . ,x, . . . ,y, . . . ,ak), then g is
bilinear.
(b) P rove that
k
D!(a1, . . . ,ak)(x1, . . . ,Xk) = l !(a1, . . . ,ai-1,Xi,ai+1, . . . ,ak).

;=1
2-15. Regard an n X n matrix as a point in the n-fold product R" X

. . X R" by considering each row as a membe r of R".
(a) P rove that det: R" X . . . X R" -> R is diff erentiable and
,D (det) (a1, . . . ,a,,) (x1, . . . ,x,,) = �

·
det
n
I �il '
;=1
la"
(b) If aij: R -> R are diff erentiable and Jet) = det (aij (t» , show
that
l'(t) = L det
i=l
(c) If det(aij(t» � O for all t and b 1, . . . ,b,,: R -> R are dif

ferentiable, let 81, . . . ,8,, : R -> R be the functions such that
a1(1), . . . ,a,, (t) are the solutions of the equations
l aji( t)aj(t)
Il
bi (t) i - l, . ,n.
i- l
= . .
Show that ai is diff erentiable and fi nd a;'(t).

Differentiation 25
2-16. Suppose f: R" --+ R" is difl'er entiable and has IL difl'erentiahle
inverse r l : R" --+ R" . Show that (f-l)'(a) = [!'(r1(a))]-1.
Hint: f o r 1 (x) = x.
P A R TIAL D E R I VA TIVES
We begin the attack on the problem of finding derivatives

"one variable at a time." If f: Rn R and a E Rn, the limit
�
if it exists, is denoted Dd(a), and called the ith partial deriva

tive of f at a. It is important to note that DJ(a) is the ordi
nary derivative of a certain function; in fact, if g(x) =
f(a l , . . . ,x, . . . ,an) , then Dd(a) = g'(ai). This means
that DJ(a) is the slope of the tangent line at (a,j(a» to the
curve obtained by intersecting the graph of f with the pIane
xj = ai, j ;.é i (Figure 2-1). It also means that computation of
D;j(a) is a problem we can already solve. If f(x l , ,xn ) is
• . •
--- ---- x·
(a,b)
FI G UR E 2-1
given by some formula involving x l , . . . ,xn, then we find

Dd(xl, • .,xn) by differentiating the function whose value
•
at X i is given by the formula when alI xi, for j � i, are

thought of as constants. For example, if f(x,y) = sin(xy 2 ),
then Dd(x,y) = y 2 COS(xy 2) and Dzf(x,y) 2xy COS(xy 2). H,
instead, f(x,y) = xv, then Dd(x,y) = yxV-1 and Dzf(x,y) =
=
XV log x.
With a little practice (e.g., the problems at the end of this
section) you should acquire as great a facility for computing
Dd as you already have for computing ordinary derivatives.
H Dd(x) exists for all x E Rn , we obtain a function Dd:
Rn R. Thejth partial derivative of this function at that
� x,
is, Di(D;f)(x), is often denoted Di,if(x). Note that this nota

tion reverses the order of i and j. As a matter of fact, the
order is usuaI1y irrelevant, since most functions (an exceptioll is
given in the problems) satisfy DiJ = Di,;!' There are various
delicate theorems ensuring this equality; the foIIowing theorem
is quite adequate. We state it here but postpone the proof
until later (Problem 3-28) .
2-5 Theorem. lf Di.if and Dj,d are continuous anm
open set containing a, then
D iJ(a) Dj d(a).
=
,
The function DiJ is called a second-order (Illixed)

partial derivative of f. Higher-order (mixed) partial
derivatives are defined in the obvious way. Clearly Theorem
2-5 can be used to prove the equality of higher-order mixed
partial derivatives under appropriate conditions. The order
of il, . . . , ik is completely immaterial in Dii, . ,id . .
if f has continuous partial derivatives of alI orders. A function

with this property is called a C'" function. In later chapters
it will frequently be convenient to restrict our attention to C'"
functions.
Partial derivatives wiII be used in the next section to find
derivatives. They also have another important use-finding
maxima and minima of functions.
Differentiation 27
2-6 Theorem. Let A C Rn. Ij the maximum ( al" mini

mum) oj j: A � R OCCUI"S at a point a in the interioT oj A and
D,f(a) exists, then D;f(a) = O.
Proof. Let gi (X) j(a l, . . . ,x, . . . ,an) . Clearly gi

=
has a maximum (or minimum) at ai, and gi is defined in an

open interval containing ai. Hence ° = g/ (ai) = D,f(a) . I
The reader is reminded that the converse of Theorem 2-6
is false even if n = 1 (if f: R R is defined by j(x) = x3,
�
then 1' (0) = 0, but ° is not even a local maximum or mini

mum). If n > 1 , the converse of Theorem 2-6 may fail
to be true in a rather spectacular way. Suppose, far exam
pIe, that j: R2 R is defined by j(x,y) = X 2 - y2 (Figure
�
2-2). Then D d(O,O) ° because g l has a minimum at 0,

=
while DJ(O,O) = ° because g2 has a maximum at O. Clearly

(0,0) is neither a relative maximum IlOr a relative minimum.
�------ x
FIG UR E 2-2
28 Calculus on Manilolds
If Theorem 2-6 is used to find the maximum or minimum oi
f on A, the values of f at boundary points must be examined
'separately-a formidable task, since the boundary of A may
be :a11 of A ! Problem 2-27 indicates one way of doing this,
and Problem 5-1 6 states a superior method which can often
be used.
l'reble,,.. . 2-17. Find the partial derivatives of the following

fun,etions :
{a) !<Z,TI,z) - Z*.
{b) }(x, y,t) == z.
(c) j{:c, y) = sin(x:sin l).
(d) I(:c,y,t) sin(z sin(y sin t» .
==
(e) !(X,II,t) - �.
(f) f(.x.II,t) ... ztt+z•
�) !(Z.II,z) (.x + y)z.
=
(h) /(z,1/) ... sin(q).

(i) !(.x,1/) ... fsin (xy»)""" '.
.2-18. Find the partial derivative8 oC the fotlowing functionB (where
,: R -+ R il! continuous) :
(a ) !(.x,1/) 5:+//g.
'"
(b) /( ,g) ... f:,.

x
(c) I(:c,g) f:(/.

'"
(d) I(:c,y) " f

(J:,I) (/.
If I(x,g) = x" " + (log x) (arctan (arctan (arctan (sin (coB xg) -
Cl
2-19.
lo g (x + y» » ) find D2/(1,y). Bint: There is an easy way to
do thiB.
2-20. Find the partial derivatives of I in terma of the derivatives of g and
h if
(a) I(x,y) g(x)h(g) .
=
(b) I(x,y) = g(x)h(ll).

(c) I(x,y) ... g(x).
(d) I(x, y) = g(y).
(e) I(x,y) g(x + y).
=
2-21. * Let g1,g 2 : R 2 ---> R be continuous. Define /: R 2 ---> R by

� y
I(x, y) ... f g 1 (t,O)dt + f g2 (x, t)dt.

O O
(a) Show that D2/(x,y) = g2(X,y).

(b) How should 1 be defined so that Dt!(x,y) g1(X,y) ? =
(c) Find a function I: R 2 ---> R such that D1/(x,y) = x and

D2/(x,y) y. Find one such that Dt!(x,y) y and D2/(x, y) '" x.
= =
Differentiation 29
2-22. * If f: R2 --+ R and D2f 0, show that f is independent of the

- Dd
=
2-23 . * Let A {(x,y) E R2 : x < 0, or x � ° and y ;o<! O ) .

second variable. If DIf 0, show that f is constant.
=
(a) If f : A --+ R and Dlf

=
D2f =0, show that f is constant.

Hint:
=
Note that any two points in A can be connected by a
° but f is not
sequence of Iines each parallel to one of the axes.
(b) Find a function f: A --+ R such that D2f =
independent of the second variable.

2-24. Define f: R 2 --+ R by
(x,y) ;o<! 0,
f(x,y)
(x,y) = O.
(a) Show that Dd(x,O) x for ali x and

= D If(O,y) -y for
(b) Show that DI•2f(0,0) ;o<! D2.lf(0,0) .

ali y.
2-25. · Define f : R -+ R by
x ;o<! O,
f(x) = { �-"-' x = O.
Show that f is a C""
function, and f(i) (O) ° for ali i . = Hint:
lim """ii"=t can be evaluated by
e-h-' l /h
h-O h h_O e
The limit 1'(0) Hm -- = =
L 'HospitaI's rule. It is easy enough to find f'(x) for x ;o<! 0, and
h_O
1"(0) lim f'(h) /h can then be found by L'Hospital's rule.
=
2-26. * Let f(x) = { �-(,,-l)-'.e-(,,+n-' x (t= ( 1 , 1 )

x E ( - 1,1),
-
.
( - 1 , 1 ) and ° elsewhere.
(a) Show that f: R --+ R is a C"" function which is positive on
(b) Show that there is a C"" function g: R --+ [0, 1 ] sueh that
1 for x � e. Hint:
function which is positive on (O,e) and ° elsewhere, let g(x)
g (x) = ° for x � ° and g(x) = If f is a C""
=
f� f/ f � f.
(c) If a E Rn, define g : Rn --+ R by
g(x) = f([xl - alI /e) . . . . . f([xn - anI /e) .
Show that g is a C"" function which is positive on

(al - e, al + e) X . . . X (an - e, an + e)
(d) If A C Rn is open and C C A is compact, show that there is

and zero elsewhere.
a non-negative C"" function f: A --+ R such that f(x) > ° for x E C

° outside of some closed set contained in A .
(e) Show that w e can choose sueh a n f s o that f : A --+ [0,1 ] and
and f =
f(x) = 1 for x E C. Hint:

If the function f of (d) satisfies
f(x) � e for x E C, consider g o f, where g is the function of (b) .
2-27. Define g, h: {x E H2: Ixl � 1 1 -+ Ha by
g(x,y) = (x,y, V I - X 2 y2), _
h(x,y) = (x,y, - V I - X2 y2 ). _
Show that the maximum of f on {x E HZ: Ixl = 1 1 i s either the

maximum of f o g or the maximum of f o h on Ix E H 2 : Ixl � 1 1 .
D E R I VA TI VES
The reader who has compared Problems 2-10 and 2-1 7 has
probably already guessed the following.
2-7 Theorem. Ii i: Rn Rm is differentiable at a, then
-+
D jf(a) exists io ::; i ::; m, I ::; j ::; n and f'(a) is the m X n

r 1
matrix (Dif(a» .
Proof. Suppose first that m = 1, so thati: Rn -+ R. Define

h: R -t Rnby h (x) (a l , . .
= . ,x, . . . ,an), with x in the
jth pIace. Then Dif(a) (f = o h)'(ai). Hence, by Theorem
2-2,
(f h)' (af)
o = f'(a) . h'(ai )
O
= f'(a) · 1 - jth pIace.
! :J
lo
Since (f h)'(af) has the single entry D;f(a), this shows that
o
Dif(a) exists and is the jth entry of the 1 X n matrix f' (a) .
The theorem now follows for arbitrary m since, by Theorem
2-3, each fi is differentiable and the ith row of f'(a) is
(h'(a) . I
There are several examples in the problems to show that the

converse of Theorem 2-7 is false. It is true, however, if one
hypothesis is added.
Differentiation 31
2-8 Theorem. If f: Rn Rm, then Df(a) exists if all

-+
D,f(x) exist in an open set containing a and if each function

Djf is continuous at a.
(Such a function f is called continuously differentiable at a.)
Proof. As in the proof of Theorem 2-7, it suffices to consider
the case m = 1 , so that f : Rn R. Then -+
f(a + h) - f(a) = f(a l + h l , a 2, ,an) - f(a l , . . . ,an)

• • •
+ f(a l + h l , a 2 + h 2 , a3, ,an) • • •
- f(a l + h l , a 2 , ,an) • • •
+ . . .
+ f(a l + h l , . . ,an + hn )
- f(a l + h l , . . . ,an- I- + hn -l, an ).
Recall that D d is the derivative of the function g defined by
g(x) = f(x,a 2, . ,an ). Applying the mean-value theorem
•
to g we obtain
f(a l + h l , a 2 ,•,an ) - f(a l ,
• . ,an ) • . •
h l . Dd(b!, a 2, ,an) = • • •
for some b I between a l and a l + h l. Similarly the ith term

in the sum equals
h i . Dd(a l + h l , . . ,ai- I + hi- I, b i , . . . ,an ) = h iDd(Ci),
for some Ci. Then
n
!f(a + h) - f(a) - I Dd(a) . h i(

lim i- I
11--.0 Ihl
n
·
II i= 1
[Dif( Ci) - Dif(a)] . h i I
= l1m
11--. 0 Ih!
n
Ih i l
� lim I ! Dd(Ci) - Dd(a) ! . -
Ihl
1I--'° i = 1
n
� lim I I Dd(Ci) - Dd(a) I

11--. 0 i= 1
= 0,
since Dd is continuous at a. I
Although the chain rule was used in the proof of Theorem
2-7, it could easily have been eliminated. With Theorem 2-8to
provide differentiable functions, and Theorem 2-7 to provide
their derivatives, the chain rule may therefore seem . almost
superfluous. However, it has an extremely important corol
lary concerning partial derivatives.
2-9 Theorem. Let Y l , . . . ,Ym : Rn � R be continuously
dijJerentiable at a, and let f: R m � R be dijJerentiable at
(g l (a) , . . .
, Ym (a)) . Define the function F: Rn � R by
F(x) = f(g l (X), . . ,Ym (x)). Then
m
DiF(a) = l D;!(Y I (a), . . . ,Ym (a)) · DiYj(a).
j= 1
Proof. The function F is just the composition f y, where o
Y = (Y I , . . . , Ym). Since Y i is continuously differentiable at

a, it follows from Theorem 2-8 that Y is differentiable at a.
Hence by Theorem 2-2,
F'(a) = f'(y(a)) . y'(a) =
(D tf(y(a)) , . . . ,Dmf(y(a))) .
( D IYI (a), . . . ,DnY l (a)
:
)
:
D1Ym (a), . . . ,DnYm (a)
But DiF (a) is the ith entry of the left side of this equation,
while "li _ 1 Di!(Y l (a) , . . . ,Ym (a)) . D iYj (a) is the ith entry
of the right side. I
Theorem 2-9 is often called the chain rule, but is weaker
than Theorem 2-2 since Y could be differentiable without Yi
being continuously differentiable (see Problem 2-32). l\fost
computations requiring Theorem 2-9 are fairly straightforward.
A slight subtlety is required for the function F: R 2 R
defined by
�
F (x,y) = f(y(x,y) ,h(x),k(y))
Differentiation 33
where h,k: R � R. In order to apply Theorem 2-9 define

ii,k: R 2 R by
�
h(x,y) = h(x) k(x,y) = k(y) .

Then
D 1 h(x,y) = h'(x) D2ii(x,y) = O,
D 1 k(x,y) = O D2k(x,y) = k'(y),
and we can write
F (x,y) = f(y(x,y),h(x,y),1é(x,y» .
Letting a = (y(x,y),h(x),k(y» , we obtain
D 1F (x,y) = Dtf(a) . D 1y(x,y) + D 2f(a) . h'(x),
D 2F (x,y) = DI/(a) . D 2y(x,y) + D3f(a) . k' (y).
It should, of course, be unnecessary for you to actuaIIy write
down the functions ii and k.
ProblemB. 2-28. Find expressions for the partial derivatives of the
foIlowing functions :
(a) F(x, y) =f(g(x) k (y), g(x) + h(y » .
(b) F(x,y,z) f(g (x + y), h (y + z» .
=
(c) F (x,y,z)� f(xll,y·,zZ).

(d) F(x,y) =f(x,g(x),h(x,y» .
2-29. Let f: R" -+ R. For x E R", the limit
. f(a
I1m
+ tx) - f(a)
t-+O
,
t
if it exists, is denoted DJ(a), and called the directionaJ deriva

tive of f at a, in the direction x.
(a) Show that D.J(a) = D;f(a).
(b) Show that D'zf(a) tDzf(a) .
=
(c) I f f i s differentiable a t a, show that DJ(a) = Df(a) (x) and

therefore Dz+llf(a) = Dzf(a) + Dllf(a) .
alI x, but if g � O, then Dz+llf(O,O) = Dzf(O,O) + Dllf(O,O) is not

2-30. Let f b e defined a s in Problem 2-4. Show that DJ(O,O) exists for
true for alI x and· y.

2
2-31. Let f: R -+ R be defined as in Problem 1-26. Show that Dzf(O,O)
exists for alI x, although f is not even continuous at (0,0).
{:
2-32. (a) Let f: R -+ R be defined by
. 1
2
Sin ; x � O,
f(x) =
x = O.
°
34 Calculu8 0n Manifolds
Show that f is differentiable at but f' is not continuous at O.
(b) Let f: R2 � R be defined by
f(x,y) =
f\ (x2
°
+ y2) sin
VX2
1
+ y2
(x,y) � 0,
(x,y) = O.
Show that f is differentiable at (0,0) but D;j is not continuous

at (0,0) .
2-33. Show that the continuity of Dd; at a may be eliminated from the
hypothesis of Theorem 2-8.
2-34. A function f: Rn � R is homogeneou8 of degree 1n if f(tx) =
t"'f(x) for alI x. If f is also differentiable, show that

n
l xi DJ(x) = mf(x).
i- l
Hint: If g(t) f(tx), find g'(l ).
=
2-35. If f:Rn � R is differentiable and f(O) O, prove that = there exist

gi: Rn � R such that
n
f(x) xigi(X).
= l
i-l
Hint: If h,,(t) = f(tx), then f(x) = n h,,'(t)dt.
I N VER SE FUNC TIONS
Suppose that i: R R is continuously differentiable in an

�
open set containing a and f' (a) ;é O. If f'(a) > O, there is an

open interval V containing a such that i'(x) > O for x E V,
and a similar statement holds if f'(a) < O. Thus f is increas
ing (or decreasing) on V, and is therefore 1-1 with an inverse
function i defined on some open interval W containing f(a) .
-l
Moreover it is not hard to show that ) l is differentiable, and

for y E W that
(r l ) ' (Y) - ' 1 »
_
1
f (r (y
An analogous discussion in higher dimensions is much more
involved, but the result (Theorem 2-1 1 ) is very important.
We begin with a simple lemma.
Differentiation 35
2-10 Lemma. Let A C Rn be a rectangle and let f: A Rn -+
be continuously differentiable . If there is a number M such that

IDjJi (x) I � M for all x in the interior of A , then
If(x) - f(y) I � n 2 Mlx - y l
for all x,y E A.
Proof. We have
n
f(y) - f (x) = l [f (y l , . : . ,yi,xH l , . . . ,Xn)
i l
-
.
- F (y l , . . . ,yi- l ,Xi, . . . ,Xn)J.
Applying the mean-value theorem we ohtain
f (y l , . . . ,Vi, Xi+ l , . . . ,Xn ) - f (y l , . . . ,yi- l , Xi, . . . ,xn)
= (yi - Xi) . D jf(Zij)
for some Zij' The expression on the right has absolute value
less than or equal to M I yi - x'l Thus
'
"
� l l yi - xii , M � nMly - x l
I f(y) - f (x) I
i- l
since each I yi - xi i � Iy x l . Finally -
"
I i(y) - f(x) I � l l f(y) - f (x) 1 � n2M · I y - xl. I
i- I
2-11 Theorem (Inverse Function Theorem). Suppose that

f: R" Rn is continuously differentiable in an open set contain
-+
ing a, and detf' (a) ;é O. Then there is an open sel V containing

a and an open set W containing f(a) such that f: V W has a -+
continuous inverse r l : W V which is differentiable and for

-+
all y E W salisfies
(r l) ' (y) = [f' (r 1 (y» ] - 1 .
Proof. Let X be the linear transformation Df(a). Then

X is non-singular, since det f'(a) ;é O. Now D(X- l o f) (a) =
D(X- l ) (f(a» o Df(a) X- l o Df(a) is the identity linear
=
36 Calculus on Mani/olds
transformation. If the theorem is true for X- l o f, it is clearIy

true for f. Therefore we may assume at the outset that X is the
identity. Thus whenever f(a + h) = f(a), we have
V(a + h) - f(a) - X(h) 1 lM 1 _
. _
I hl - Ihl -
But
lim If(a + h) -I hlf(a) - X(h) I = O.
"... 0
This means that we cannot have f(x) = f(a) for x arbitrarily

close to, but unequal to, a. Therefore there is a closed rec
tangle U containing a in its interior such that
1. f(x) � f(a) if x E U and x � a.
Since f is continuously differentiable in an open set containing
a, we can also assume that
2. detf'(x) � O for x E U.
3. I Djf(x) - Djf(a) I < 1/2nz for all i, j, and x E U.
Note that (3) and Lemma 2-10 applied to g(x) f(x) - x =
imply for XI ,X2 E U that

I f(XI ) - Xl - (f(X2 ) - x 2) 1 ::; tl X I - x 2 1 ·
Since
IXI - x 2 1 - I f(XI ) - f(X2) I ::; I f(XI ) - Xl - (f(X2 ) - X2 ) I
::; t l x l - x z l ,
we obtain
Now f(boundary U) is a compact set which, by ( 1 ), does not

contain f(a) (Figure 2-3 ). Therefore there is a number d > O
such that I f(a) - f(x) 1 Z d for x E boundary U. Let
W I y : I y - f(a) I < d/2}. If y E W and x E boundary U,
=
then
5. Iy -
f(a) I . ""l
I
�
� ...
'<::
�
r::
�
S
;::::
.J::.
�
...
...
37
function g : U R defined by
---t
g(x) = I (y i - fi(X)) 2.
I y - f(x) 1 2 =
i= 1
This function is continuous and therefore has a minimum on

U. If x E boundary U, then, by (5) , we have g(a) < g(x).
Therefore the minimum of g does occur on the boundary not
of U. By Theorem 2-6 there is a point x E interior U such
that Djg(x) O for aH j, that is
=
"
I 2(yi - i{x) ) . Djf(x) O for all j. =
i=l
By (2) the matrix (D,l(x)) has non-zero determinant. There

fore we must have yi - fi(x) O for alI i, that is y f(x) .
= =
This proves the existence of x. Uniqueness follows immedi

ately from (4).
If V (interior U ) n r l ( W) , we have shown that the
=
function f: V W has an inverse r l : W V. We can

---t ---t
rewrite (4) as
for Yl,Y 2 E W.
This shows that r l is continuous.
Only the proof that r l is differentiable remains. Let
p. = Df(x) . We wiU show thatrl is differentiable at y f(x) =
with derivative p.- l . As in the proof of Theorem 2-2, for
X l E V, we have
where
lim Irp (X I - x) 1 = o.
:>:,--+:>: I XI x l
-
Therefore
Since every Yl E W is of the form f(Xl) for some Xl E V, this

can be written
Differentiation 39
snd it therefore suffices to show that

Hm IJl-1(I" (rIY1(Yl-) - r l (y))) I o.
1/1-+ 1/ l yl
=
Therefore (Problem 1-10) it suffices to show that

l
lim II" (r1 (Yl ) -- r (y)) I O.
1/1-+ 1/
=
IYl yl
Now
II" (r 1 (Y l) - r1 (Y)) 1
! Yl - y l
II"(r 1 (Y l) - r 1(Y)) 1 Ir 1(Y l ) - r 1(Y) 1
I f-1(Yl) - f- l (y) ! !Yl - Y!
Since r 1 is continuous, r 1(Yl) -t r l (y) as Y l -t y. There
fore the first factor approaches O. Since, by (6) , the second
factor is less than 2, the product also approaches O. I
It should be noted that an inverse function r 1 may exist
even if detf'(a) O. For example, if f: R -t R is defined by
=
f(x) x3, then 1'(0) = O but f has the inverse function

=
r 1(x) �;. One thing is certain however: if detf'(a) = O,

=
then r 1 cannot be differentiable at f(a). To prove this note

that f r 1(x) = x. If r 1 were differentiable at f(a), the
o
chain rule would givef'(a) . (r 1)'(f(a)) = I, and consequently

detf'(a) . det(r1)'(f(a)) = 1 , contradicting det f'(a) = O.
2-36.* Let A C R B be an open set and f: A
a eontinuously differentiable 1-1 funetion sue h that det f (x) � O
ProbJems. -+ RB
'
for ali x. Show thatf(A) is an open set and ,l : f(A) -+ A is differ

entiable. Show also that f(B) is open for any open set B e A .
2
not 1-1 . Hint: If, for example, Dtf(x,y) � O for ali
2-37. (a) Let f: R -+ R be a eontinuously differentiable function.
Show that f is
(x,y) in some open set A , eonsider g: A -+ R 2 defined by g(x,y) =
(j(x,y),y) .
(b) Generalize this result to the case of a eontinuously differen
tiable function f: R B -+ Rm with m < n .
2-38. (a) If f: R -+ R satisfies f'(a) � O for ali a E R, show that f is
1-1 (on aH of R).
40 CalCUlU8 on Manifold8
(b) Define f: R2 -+ R2 by f(x,y) = (eZ cos y, eZ sin
that det f'(x,y) ;é O for all (x,y) but f is not l - l .
y) . Show
{X�
2-39. Use the function f: R -+ R defined by
+ x2 Slll ;é O,
' l
; x
f(x) =
x= O,
to show ihat continuity of the derivative cannot be eliminated from

the hypothesis of Theorem 2-1 1 .
I M P L I C I T F U N C TI O NS
Consider the function f: R 2 R defined by f(x,y) = X2 +

�
y 2 l. If we choose (a,b) with f(a,b) = O and a ;é 1, - 1 ,

_
there are (Figure 2-4) open intervals A containing a and B

containing b with the following property: if x E A , there is
a unique y E B with f(x,y) O. We can therefore define
=
B b
A
-----------+------------+---��--�-----------x
a
BI
graph of gl
....
b I_
.:, -+-7"
FIG URE 2-4

Differentiation 41
a function g : A H by the condition g(x) E B and f(x,g(x»

--+
= O (if b > O, as indicated in Figure 2-4, then g(x) =
VI X 2). For the function f we are considering there is

-
another number b I such that f(a,b l ) = O. There will also be

an interval BI containing b I such that, when x E A , we
have f(X,gl(X» O for a unique gl (X) E BI (here gl(X) =
=
- VI X 2). Both g and gl are differentiable. These

-
fu�ctions are said to be defined impIicitly by the equation

f(x,y) = o.
If we choose a = l or - l it is impossible to find any such
function g defined in an open interval containing a. We
would Iike a simple criterion for deciding when, in generaI,
such a function can be found. More generally we may ask
the following: If f: Hn X H H and f(a I , . . . ,an ,b) = O,
--+
when can we find, for each (X l , . . . ,xn) near (a l , . . . ,an),

a unique y near b such that f(x I , . . . ,xn ,y) O? Even=
more generally, we can ask about the possibility of solving

m equations, depending upon parameters x . . . ,xn , in m
I,
unknowns: If
t = 1, . . . ,m
and
t = l, . ,m,
when can we find, for each (x l, . . . ,xn) near (a l , . ,an) a
unique (y I , . . . ,ym) near (bI, . . . ,bm) which satisfies
fi (X\ . . . ,xn , y\ . . . ,ym) O? The answer is provided by
=
2-12 Theorem (Implicit Function Theorem). Suppose

n
f: H X Hm Hm is conlinuously differentiable in an open set
--+
containing (a,b) and f(a,b) = O. Let M be the m X m matrix

1 � i, j � m .
If det M ;é. O, there is an open set A C H n containing a and an
open set B C Hm containing b, with the following property: for
each x E A there is a unique g(x) E B such that f(x,g(x» = o.
The function g is differentiable.
4% Ca/CU/U8 on Manifold8
Proof. Define F : Rn X Rm Rn X Rm by F(x,y) =
�
(x,j(x,y)) . Then det F'(a,b) = det M r= O. By Theorem 2-1 1

there is an open set W C Rn X Rm eontaining F(a,b) = (a,O)
and an open set in Rn X Rm eontaining (a,b), whieh we may
take to be of the form A X B, sueh that F: A X B W �
has a differentiable inverse h : W A X B. Clearly h is of

�
the form h(x,y) = (x,k(x,y» for some differentiable funetion

k (sinee F is of this form). Let 11": Rn X Rm Rm be defined
�
by 1I"(x,y) = y ; then 11" o F = f. Therefore

f(x, k{x,y» = f o h(x,y) = (11" o F) o h(x,y)
= 11" o (F o h) (x,y) = 1I"(x,y) = y.
Thus f(x,k(x,O» = O ; in other words we ean define g(x) =
k(x,O). I
Since the funetion g is known to be differentiable, it is easy

to find its derivative. In fact, since f(x,g(x» = 0, taking Dj
of both sides gives
, m
= Djf(x,g(x» l Dn+afi (x,g(x» · Djga (x)

,, - l
° +
i,j = 1, . . . ,m.
Sinee det M r= 0, these equations ean be solved for Djg"(x) .

The answer will depend on the various Djf(x,g(x» , and there
fore on g(x) . This is unavoidable, sinee the funetion g is not
unique. Reeonsidering the function f: R 2 R defined by
f(x,y) = X 2 + y 2 - 1 , we note that two possible funetions
�
satisfying f(x,g(x» = ° are g(x) = v'l X2 and g(x) =

- v'l - X 2 . Differentiating f(x,g(x»
-
° gives
=
D d(x,g(x» + Dzf(x,g(x» . g' (x) 0, =
or
2x + 2g(x) . g' (x) = 0,
g'(x) = - x/g(x) ,
whieh is indeed the case for either g(x) = v'l - X 2 or g(x) =

- � .
Differentiation 43
A generalization of the argument for Theorem 2-12 can be

given, which will be important in Chapter 5.
2-13 Theorem. Let f: Rn � RP be conttrmously differ
entiable in an open set containing a, where p � n. Ii f(a) O =
and the p X n matrix (D;!i (a» has rank p, then there is an

open set A C R" containing a and a differentiable function h:
A � R" with differentiable inverse such that
f o h (x 1 , • • • ,x") = (X"-P+l, . . . , x") .
Proof. We can consider f M a functionf: R"-P X RP � RP.

Ii det M � O, then M is the p X p matrix (D" -p+iF(a» ,
l � i, j � p, then we are precisely in the situation considered
in the proof of Theorem 2-12, and M we showed in that proof,
there is h such that f h(x t, . . . ,x") = (x"-P+ l, . . . ,x").
o
In generaI, since (D;F(a» has rank p, there will be i I <

. . . < jp such that the matrix (DiF(a» l � i � p, j =
iI, . . . ,jp has non-zero determinant. Ii g: R" � R" per
mutes the xi so that g(xl, . . . ,x") = ( . . . ,x\ . . . ,X;"),
then f g is a function of the type already considered, so
o
«(f o g) k) (xt,
o . ,x") = (x" -P + l, . . . ,xn) for some k.
Let h = g o k . I
Problems. 2-40. Use the implicit function theorem to re-do Prob
lem 2-15(c).
2-41. Let !: R X R -> R be differentiabJe. For Aach x E R define g.,:
R -> R by g.,(y) !(x, y) . Suppose that for each x there ie a
=
unique y with g.,'(y) O; Jet c(x) be this y.

=
(a) If D 2. 2!(X,y) � O for ali (x,y), show that c is differentiable

and
_ D2 ,1!(x, c (x»
c'(x) = .
D 2 . 2!(X,C(x) )
O can be written Dzf(x,y) O.
O, then for some y we have
Bint: g.,'(y) = =
(b) Show that if c'(x) =
D2 .I!(X,y) = O,
Dzf(x,y) = O.
(c ) Let !(x,y) = x (y Jog y - y) - y Jog x. Find
max ( min !(x,y» .

1:5.,:5 2 .:511:5 1
NO T A TI O N
This section is a brief and not entirely unprejudiced discussion

of classical notation connected with partial derivatives.
The partial derivative D d(x,y,z) is denoted, among devotees
of classical notation, by
af
af(x,y,z)
ax
-af or
or ax -
ax
(x,y,z) -a f(x,y,z)
or ax
or any other convenient similar symbol. This notation forces
one to write
af
- (u,v,w)
au
for D d(u,v,w), although the symbol
af(x,y,z)
ax I
I (z,I/,') = (u,v,wl
or af(x,y,z)
ax
(U,V,w)
or something similar may be used (and must be used for an

expression like D d(7, 3,2)) . Similar notation is used for Dd
and D 3f. Higher-order derivatives are denoted by symbols
like
Whenf: R � R, the symbol a automatically reverts to d; thus

sin x, not a sin x
d-- __ o
dx ax
The mere statement of Theorem 2-2 in classical notation
requires the introduction of irrelevant letters. The usual
evaluation for D (f (g,h)) runs as follows:
1 o
If f(u,v) is a function and u = g(x,y) and v = h(x,y),

then
a�
Ij� ,_
(g�(x--'-.:,y�)-,-- x�
h(�,y�
)) a_ (u,_
Ij_ v) au + a_ (u,_
Ij_ v) avo
=
ax au ax av ax
[The symbol aujax means ajax g(x,y) and ajauf(u,v) means
Differentiation 45
DI!(u,v) = DI!(g(x,y) , h(x,y)).] This equation is often written

simply
ai ai au ai av
- = - - + - _.
ax au ax av ax
Note that i means something different on the two sides of thc
equation!
The notation di/dx, always a little too tempting, has inspired
manyJusually meaningless) definitions of dx and di separately,
the sole purpose of which is to make the equation
di .
di = dx
dx
work out. If i: R 2 � R then di is defined, classically, as
ai af
di dx dy
ax + ay
=
(whatever dx and dy mean).

Chapter 4 contains rigorous definitions which enable us to
prove the abo\ée equations as theorems. It is a touchy
question whether or not these modern definitions represent a
real improvement over classical formalism; this the reader
must decide for himself.
3
I ntegration
B A SI C D EFINI TIONS
The definition of the integraI of a function f: A R, where�
A C Rn is a closed rectangIe, is so similar to that of the ordi

nary integraI that a rapid treatment will be given.
Recall that a partition P of a closed intervaI fa,bI is a
sequence to, . . . ,tk, where a = to � t I � . . . � tk = b.
The partition P divides the interval fa,bI into k subintervaIs
[ti_ l ,til. A partition of a rectangle [al,b l l X . . . X [an ,bn]
is a collection P (PI,
= • •,Pn ) , where each Pi is a par
•
tition of the interval [ai,b;J. Suppose, for example, that

P l to, . . . ,tk is a partition of [al,bl] and P2 = S o, .
= . ,SI
.
is a partition of [a2,b2]. Then the partition P =(P l ,P2) of

[a l ,bd X [a2,b2] divides the closed rectangle [a l ,bd X [a2,b21
into k . l subrectangIes, a typical one being [ti_ l ,til X [Si- l ,SiI .
In generaI, if Pi divides [ai,bil into Ni subintervals, then p =
(P!, . ,Pn ) divides [a l ,bd X . . . X [an ,bn l into N

N l · . . . . Nn subrectangles. These subrectangles will be
. . =
called subrectangles of the partition P.

Suppose now that A is a rectangIe, f: A R is a bounded
�
46
Integration 47
function, and P is a partition of A . For each subrectangle S

of the partition let
ms(f) = inflf(x) : x E S I .
Ms(f) = suplf(x) : x E S I .
and let veS) be the volume of S [the volume of a rectangle
[al,bd X . . . X [an ,bnl. and also of (al,bj) X . . . X (an ,bn ) ,
is defined as (b I - a l ) ' . . . . (bn - an )]. The lower and
upper sums of I for P are defined by
L(f,P) l ms(f) . veS) and U(f,P ) l Ms(f) . veS).

s s
= =
Clearly L(f,P) � U (f,P) , and an even stronger assertion (3-2)

is true.
3-1 Lemma. Suppose the partition p' refines P (that is,
each subrectangle 01 p' is contained in a subrectangle oj P ).
Then
L(f,P) � LU,P') and� U(f,P) . U(j,P')
Proof. Each subrectangle S of P is divided into several sub

rectangles SI, . . . ,S" of P', so VeS) = ves 1) + . . . +
veS,,) . Now ms(f) � ms,(f) , since the values j(x) for x E S
include all values j(x) for x E Si (and possibly smaller ones).
Thus
ms(f) . v eS) = ms(f) V(S l )
. + . .
. + ms(f) . v eS,, )
� m st (f) V(SI) + . . . + ms,,(f) . veS,,).
.
The sum, for al! S, of the terms on the left side is L(f,P) ,
while the sum of all the terms on the right side is L(j,P').
Hence L(j,P) � L(j,P') . The proof for upper sums is
similar. I
3-2 Corollary. Ij P and p' are any two partitions, then
L(/,P') � U(j,P ).
Proof. Let P" be a partition which refines both P and P'.
(For example, let P" = (P;', . . . ,P�) , where P;' is a par-
48 Calculu8 on Manifoids
tition of [ai,bi] which refines both Pi and P�o) Then
L(f,P') ::; L (f,P") ::; U(f,P") ::; U(f,P) o I
It folIows from CorolIary 3-2 that the least upper bound of
alI lower sums for f is less than or equal to the greatest lower
bound of all upper sums for f. A function f: A R is called
�
integrabIe on the rectangle A if fis bounded and suplL(f,P) l

= infl U (f,P) }
o This common number is then denoted fA f,
and called the integraI of f over Ao Often, the notation
fA f(x\ o o o ,xn )dx 1 o o dxn is usedo If f: [a,b] R, where
o
�
a � b, then f�f f [a,blf. A simple but useful criterion for

=
integrability is provided by
3-3 Theorem. A bounded function f: A � R is integrable
if and only if for every e > O there is a partition P of A such
that U(f,P) - L(f,P) < eo
Proof. If this condition holds, it is clear that sup lL(f,P) l

inf{ U(f,P) l and f is integrableo On the other hand, if f is
integrable, so that suplL(f,P) l = infl U(f,P) l , then for
any e > O there are partitions P and p' with U(f,P) - LU,P')
< eo If p" refines both P and P', it follows from Lemma 3-1
that U(f,P") - L(f,P") ::; U(f,P) LU,P') < eo I
-
In the following sections we will characterize the integrable

functions and discover a method of computing integralso For
the present we consider two functions, one integrable and one
noto
1 . Let f: A R be a constant function, f(x) Co Then
� =
for any partition P and subrectangle S we have ms(f) =

Ms(f) = c, so that L(f,P) = U(f,P) � sc o veS) = c v(A)o
=
o
Hence fAf c o v(A)o

=
2 0 Let f: [0,1] X [0, 1 ] R be defined by

�
f(x,y) =
{ 0 if x is rational,
1 if x is irrational.
If P is a partition, then every subrectangle S will contain
points (x,y) with x rational, alld also points (x,y) with x
lntegration 49
irrational. Hence ms (f) O and Ms(f) 1 , so = =
L(f,P) = � O . v eS) O =
and
U(f,P) = L 1 . v (S) v ( [O, l] X [0, 1 ] )
= 1.
s
Therefore f is not integrable.
Problems. 3-1 . Let f: [O, l J X [O, l J -+ R be defined by
f(x, y ) = {� if O �
if ! �
x
x
< i,
� 1.
Show that f i s integrable and [O. lX [D. l I fI

3-2. Let f : A -+ R be integrable and let g
!. =
f except at finitely rnany

=
IA I
3-3. Let f,g : A -+ R be integrable.
points. Show that g is integrable and f = A g·
(a ) For any partition P of A and subrectangle S, show that
ms(j) + ms (g) � ms(f + g) and Ms(j + g)

� Ms(j) + Ms(g)
and therefore
LCf,P) + L(g,P) � L(j + g, P) and U(j + g, P)

� U(j,P) + U(g,P).
(b) Show that f + g is integrable and IAf + g =

IA f + IA g.
(c) For any constant show that f A c, cf cIA f·
3-4. Let f: A -+ R and let P be a partition of A . Show that f is integra
=
ble if and only if for each subrectangle S the function f lS, which
consists of f restricted to S, is integrable, and that in this case
IAf �sf sflS.
3-5. Let f,g: A
=
-+ R
f A f � I Ag·
be integra bI e and suppose f � g. Show that
3-6. If f: A -+ R is integrable, show that Ifl is integrable and I IAfl �

fAlli.
3-7. Let f: [O, l J X [0, 1 1 -+ R be defined by
f x, 1j )
( = (� l /q
x irrational,
x rational, y irrational,
x rational, y p/q in lowest terrns.
=
Show that f is integrable and I [O,lIX[O. ll f = o.

MEA S UR E Z E R O A ND C O N TEN T ZERO
A subset A of Rn has (n-dimensionaI) measure O if for every

e > O there is a cover l UI, U 2, U 3, . . 1 of A by closed rec
•
tangles such that 2;;:'l V ( Ui) < e. It is obvious (but never

theless usefuI to remember) that if A has measure O and
B C A , then B has measure O. The reader may verify that
open rectangles may be used instead of closed rectangles in
the definition of measure O.
A set with only finitely many points clearly has measure O.
If A has infinitely many points which can be arranged in a
sequence al, a2, a3, . . . , then A also has measure O, for if
e > O, we can choose Ui to be a closed rectangle containing
ai with v( Ui) < e/2i . Then 2;:' l V(Ui) < 2;:' l e/2 i e. =
The set of alI rational numbers between O and 1 is an impor

tant and rather surprising example of an infinite set whose
members can be arranged in such a sequence. To see that
this is so, Iist the fractions in the following array in the order
indicated by the arrows (deleting repetitions and numbers
greater than 1 ) :
/' /' /' /'
0/1 1/1 2/1 3/1 4/1
/ / / /
0/2 1/2 2/2 3/2 4/2
/ / /
0/3 1/3 2/3 3/3 4/3
/ /
0/4
/
An important generalization of this idea can be given.

3-4 Theorem. Il A
Al U A2 U A3 U . = .
. and each
A i has measure O, then A has measure O .
Proof. Let e> O. Since A i has measure O , there is a cover

I Ui , l I U i, 2 , Ui , 3 ,
. .
i
,I of A i by closed rectangles such that
2;� l V(Ui, ;) < e/2 . Then the colIection of alI Ui,; is a cover
Integration 51
of A . By considering the array

/ / /
UI. I UI , 2 UI,a
/ / /
U2, 1 U2,2 U2,a
/ /
Ua , l Ua,2 Ua,a
/
we see that this collection can be arranged in a sequence
V I , V 2 , Va, . . . . Clearly �;=lV( Vi) < �:"I€/2i €. I =
A subset A of Rn has (n-dimensionai) content O if for every

€ > O there is a finite cover l U I , . . . , Un I of A by closed
rectangles such that ��lV( Ui) < €. Ii A has content O,
then A clearly has measure O. Again, open rectangles could
be used instead of closed rectangles in the definition.
3-5 Theorem. 1f a < b , then [a, b] C R does not have con
tent O. In fact, if l UI, . . . , Un I is a finite cover of [a , b) by
closed intervals, then �i=IV( Ui) � b a. -
Proof. Clearly we can assume that each Ui C [a,b]. Let

a = to < tI < . . . < tk = b be aIl endpoints oi alI Ui. Then
each v( Ui) is the sum oi certain ti - ti_l. Moreover, each
[ti_l,ti] lies in at Ieast one Ui ( namely , any one which contains
an interior point of [ti-l,t;]), so l;i_Iv(Ui) � l;�_l (t; - ti_l)
= b - a. I
Ii a < b, it is also true that fa ,b) does not have measure O.

This folIows from
3-6 Theorem. If A is compact and has measure O, then A
has content O.
Proof. Let € > O . Since A has measure O, there is a cover

I UI, U 2 , • 1 of A by open rectangles such that �;=IV( Ui)
• •
< €. Since A is compact, a finite number U l, . . . , Un of
the Ui also cover A and surely �� l V( Ui) < €. I
The conclusion of Theorem 3-6 is false if A is not compact.

For example, let A be the set of rational numbers between O
and 1 ; then A has measure O. Suppose, however, that
{ [a , b l, . . . ,[an,bnl l covers A . Then A is contained in
l t
the closed set [a ,b l) U . . . U [an,bn), and therefore [0, 1 ) C

l
[al,bd U . . . U [a",b,,). It folIows from Theorem 3-5 that

� i=l (bi - ai) � 1 for any such cover, and consequently A
does not have content O.
Problems. 3-8. Prove that [abb l l X . . . X [an,bnl does not have
content O if ai < bi for each i.
3-9. (a) Show that an unbounded set cannot have content O.
(b ) Give an example of a closed set of measure O which does not
have content O.
3-10. (a) If C is a set of content O, show that the boundary of C has
(b) Give an example of a bounded set C of measure O such that

content O.
3-I l . Let A be the set of Problem 1-18. If � :' I (bi - ai) < 1, show
the boundary of C does not have m easure O.
that the boundary of A does not have m easure O.

3-12. Let f: [a,bl -+ R be an increasing function. Show that Ix: f is
show that (x: o(f,x) > l /n } is finite, for each integer n.

discontinuous at x } has m easure O. Hint: Use Problem 1-30 to
3-13. * (a) Show that the collection of ali rectangles [al,bll X . . X .

(b) If A C R n is any set and 0 is an open cover of A, show that
[an ,bn l with ali ai and bi rational can be arranged in a sequence.
there is a sequence UI, U2, Ua, . . . of members of 0 which also

cover A . Hint: For each x E A there is a rectangle B [al, b ll X
. . . X [an,bnl with ali ai and bi rational such that x E B C U
=
for some U E 0.
lN TE G R A B L E F U N C TIONS
Recall that o(f,x) denotes the oscillation of f at x.

3-7 Lemma. Let A be a closed 1'ectangle and let f: A - R be
a bounded function such that o(f,x) < € fo1' all x E A . Then
the1'e is a pa1'tition P of A with U(f,P) - L (f,P) < e ' v ( A).
lntegration 53
Proof. For each x E A there is a closed rectangle Ux,

containing x in its interior, such that Mu.(f) mu. (f) < E . -
Since A is compact, a finite number U , U of the

XI > • • • Xn
sets Ux cover A . Let P be a partition for A such that each

subrectangle S of P is contained in some Ux;. Then Ms(f) -
ms(f) < E for each subrectangle S of P, so that U(f,P) -

L(f,P) 2;s[Ms(f) - ms(f)] v eS) < E . v( A ) . I
= .
3-8 Theorem. Lel A be a closed reclangle and f: A R a -t
bounded funclion. Lel B Ix: f is nol conlinuous al x l .

=
Then f is integrable if and only if B is a sel of measure o.
Proof. Suppose first that B has measure o. Let E > O and

let Be I x: o(f,x) ;::: El· Then Be C B, so that Be has
=
measure O. Since (Theorem 1-11) Be is compact, Be has con

tent O. Thus there is a finite collection UI, ,Un of • • .
closed rectangles, whose interiors cover Be, such that 2;�IV(Ui)

< E. Let P be a partition of A such that every subrectangle
S of P is in one of two groups (see Figure 3-1) :
Fl G V R E 3-1. The shaded rectangles are in S I .

54 Calculu8 0n Manifolds
which consists of subrectangles S, such that s e Ui

( 1 ) (�h,
for some i.
( 2 ) S2, which consists of subrectangles S with S (ì Be
= 0·
Let I f(x) ! < M for X E A . Then Ms (f) - ms (f) < 2 M

for every S. Therefore
n
lE . [Ms (f) - ms (f)] . v(S) < 2M 1: v( Ui) < 2Me.

S S i� I
Now, if S E S 2 , then o (f,x) < e for E S. Lemma x 3-7

implies that there is a refinement p' of P such that
") [ Ms ,(f) - ms,(f)] . v(S') < e v(S)
s s
.
'(
'
for S E S2. Then

U( f,P') - L(f, P') = ' c� [Ms,(f) - ms,(f)] . v(S')
s E .
S
+ ' cf s. [ Ms,(f) - mS'(f)] . v(S')
s E
< 2Me +
E
l
e · v(S)
S S,
:::; 2Me + e v(A ). .
Since M and v(A) are fixed, this shows that we can find a
partition p' with U(f,P') - L (f,P') as small as desired. Thus
f is integrable.
Suppose, conversely, that f is integrable. Since B =
BI U Bi U Bt U . , it suffices (Theorem 3-4) to prove

. .
that each BI/n has measure O . In fact we will show that

each B I / n has content O (since B I / n is compact, this is actuaIly
equivalent).
If e > O, Iet P be a partition of A such that U(f,P) -
L (f,P) < e/n. Let S be the collection of subrectangles S
of p which intersect B I/n' Then S is a cover of B I /n' Now if
Integration 55
S E S, then Ms (f) - ms(f) 2:: l/n. Thus

� . 2: v(S) 2: [Ms(f) - ms(f)] . v(S)
�
SES BES
� 2: [Ms(f) - ms(f) ] v(S)
.
B
e
< , -
n
and consequently � 8ESV(S) < e. I
We have thus far dealt only with the integrals of functions

over rectangles. Integrals over other sets are easily reduced
to this type. If C C R"', the eharaeteristie funetion xc
of C is defined by
xc(x) = { �
x E;è C,
x E C.
If C C A for some closed rectangle A and I: A R is --->
bounded, then fcf is defined as fAI ' xc , provided l ' Xc is

integrable. This certainly occurs (Problem 3-14) if I and
x c are integrable.
3-9 Theorem. The lunction xc: A R is integrable il and

--->
only il the boundary 01 C has measure O (and hence content O).
Proof. If x is in the interior of C, then there is an open

rectangle U with x E U C C. Thus xc = 1 on U and Xc is
clearly continuous at x. Similarly, if x is in the exterior of C,
there is an open rectangle U with x E U C R'" - C. Hence
xc O on U and xc is continuous at x. Finally, if x is in
=
the boundary of C, then for every open rectangle U containing

x, there is Yl E U lì C, so that Xc(Yl) = 1 and there is
Y 2 E U lì (R'" - C), so that Xc(Y 2) = O. Hence xc is not
continuous at x. Thus {x : xc is not continuous at xl
boundary C, and the result follows from Theorem 3-8. I
A bounded set C whose boundary has measure O is called
Jordan-measurable. The integraI f cl is called the
(n-dimensionaI) content of C, or the (n-dimensionaI) volume
of C. Naturally one-dimensionai volume is often called
length, and two-dimensionai volume, area.
Problem 3-11 shows that even an open set C may not be
Jordan-measurable, so that fcf is not necessarily defined even
if C is open and f is continuous. This unhappy state of affairs
will be rectified soon.
.f.g: -+
f · (f.
Problems. 3-14. Show that if A R are integrable, so is
3-15. Show that if C has content O, then C C A for some closed rectangle
fA XC
3-16. Give an example of a bounded set C of measure O sueh that
A and C is Jordan-measurable and = O.
fA XC
3-17. If C is a bounded set of m easure O and
does not exist.
fA XC
fA XC
exists, show that
O. Hint: Show that L(f,P)
= O for ali partitions P.
=
f Af = O, show that I x : f(x) � Ol

Use Problem 3-8.
f: A
has measure O. >
3-18. If -+ R is non-negative and
Hint: Prove that Ix: f(x) l/n l has eontent O.
except on a set of measure 0, then

3-19. Let U be the open set of Problem 3-1 1 . Show 'that if f =
xu
f is not integrable on [O, l J .

3-20. Show that a n inereasing function f: [a,bJ -+ R i s integrable on
3-21. If A is a closed reetangle, show that r: C A is Jordan-measurable

[a, bJ.
if and only if for every E > ° there is a partition

 O, show that there is

compact Jordan-measurable set C C A sueh that A-C 1 < E.
a
F UB INI ' S THE O R E M
The problem of calculating integrals is solved, in some sense,

by Theorem 3-10, which reduces the computation of integrals
over a closed rectangle in Rn, n > 1 , to the computation of
integrals over closed intervals in R. Of sufficient importance
to deserve a special designation, this theorem is usually
referred to as Fubini's theorem, although it is more or less a
lntegration 57
special case of a theorem proved by Fubini long after Theorem

3-10 was known.
The idea behind the theorem is best illustrated (Figure 3-2)
for a positive continuous function f: [a,bJ X [c,d] R. Let -
to, . . . ,tn be a partition of [a,bJ and divide [a,b] X [c,d]

inta n strips by means of the line segments I ti l X [c,dJ.
If (Ix is defined by (/x(y) = f(x,y) , then the area of the regian
under the graph of f and above I x l X [c,dJ is
d d
(Ix
f = f f(x,y)dy.
c c
The volume of the region under the graph of f and

abave [ti-l ,tiJ X [c,dJ is therefore approximately equal to
(ti - ti-d . f�f(x,y)dy, for any x E [ti_ l ,tiJ. Thus
"
f f= L f f
[a.b1 x [c.d1 i= l [t'_1.t,1 x [c.d1
18 approximately 2;� 1 ( ti - ti- l ) . f:f(Xi,y)dy, with Xi In
Jl;l'U ph or I
Il
FI G UR E 3-2
[ti_ l ,ti] '
On the other hand, sums similar to these appear in
the deflnition of J�(J�/(x,y)dy)dx. Thus, if h is defined by
h(x) = J�g", = f�/(x,y)dy, it is reasonable to hope that h is
integrable on [a,b] and that
J J h J ( J I(x,y)dy) dx.
b b d
I= =
[a,b) x [c,dl a a c
This will indeed turn out to be true when i is continuous, but

in the generaI case difficulties arise. Suppose, for example,
that the set of discontinuities of I is { xo l X [c,d] for some
Xo E [a,b]. Then I is integrable on [a,b] X [c,d] but h(xo) =
f�/(xo,y)dy may not even be defined. The statement of

Fubini's theorem therefore looks a little strange, and will be
folIowed by remarks about various special cases where simpler
statements are possibIe.
We will need one bit of terminology. If i: A R is a -+
bounded function on a cIosed rectangle, then, whether or not

I is integrable, the least upper bound of alI lower sums, and
the greatest lower bound of alI upper sums, both exist. They
are calIed the lower and upper integrals of I on A , and
denoted
and u J i,
A
respectiveIy.
3-10 Theorem (Fubini's Theorem). Let A C Rn and
B C Rm be closed rectangles, and letl: A X B -+ R be integrable.
Far x E A let g",: B -+ R be defined by g",(y) = I(x,y) and let
.e(x) = L j g", = j I(x,y)dy, L
'U(x) = u J g", u J I(x,y)dy.

=
Then .e and 'U are integrable on A and

J I J .e J ( [ l(x,Y)dY) dx,
= = L
J J 'U J (u i l(x,y)dY) dx.

AXB A A
1 = =
A XB A A
Integration 59
(The integrals on the right side are called iterated integrals

for I.)
Proof. Let PA be a partition of A and PB a partition of B .
Together they give a partition P of A X B for which any
subrectangle S is of the form SA X SB , where SA is a sub
rectangle of the partition PA , and SB is a subrectangle of the
partition PB• Thus
L (f, P) = I ms(f) . veS) = I mSAXSB(f) ' V(SA X SB)
S SA,Sa
= I (I mSAXS B(f) . V(SB» ) . V(SA).
SA Sa
Now, if X E SA , then clearly mSAXSB(f) � msB(yx) . Conse

quently, for X E SA we have
I mSAXSB(f) . V(SB) � I msB(yx) . V(SB) � L J Yx = .c(x).
Sa Sa B
Therefore
I ( I mSAXSB(f) . V(SB) ) . V(SA ) � L (.c ,PA ) '
SA S8
We thus obtain
where the proof of the last inequality is entirely analogous

to the proof of the first. Since I is integrable, sup {L (f,P) }
inf { U (f,P) } JA XBf. Hence
=
sup{L(.c ,P'A ) J inf { U (.c ,PA) J JAXB f.

= =
In other words, .c is integrable on A and JAXBI = JA.c . The

assertion for 'U follows similarly from the inequalities
L (f, P) � L(.c ,P A) � L( 'U ,P A) � U ( 'U ,P A) � U (f, P) . I
Remarks . 1. A similar proof shows that
AXB
J I = J ( L J I(x,y)dx) dy = J (u j I(x,y)dx) dy.
A
60 Calculu8 on Manifold/J
These integrala are called iterated integrals for I in the reverse
order from those of the theorem. As several problems show,
the possibility of interchanging the orders of iterated integrals
has many consequences.
2. In practice it is often the case that each gx is integrable,
so that IA XBI = IA CfBlex,y)dy)dx. This certainly occurs
if I is continuous.
3. The worst irregularity commonly encountered is that gx
is not integrable for a finite number of x E A . In this case
.c(x) = I BI(x,y)dy for alI but these finitely many x. Since
I A.c remains unchanged if .c is redefined at a finite number of
points, we can stilI write IAXBI I A (IBlex,y)dy)dx, pro
=
vided that IBI(x,y)dy is defined arbitrarily, say as o, when it

does not exist.
4. There are cases when this wilI not work and Theorem 3-10
must be used as stated. Let I: [0,11 X [0,1 ] -+ R be defined
by
if x is irrational,
J
(x,y)
� f: - l /q
if x is rational and y is irrational,
if x p!q in Iowest terms and y
=
rational.
lS
Thenf is integrable and I IO. l] X IO.lJ ! = 1 . Now nlex,y)dy = 1

if x is irrational, and does not exist if x is rational. There
fore h is not integrabIe if h(x) = nlex,y)dy is set equal to O
when the integraI does not exist.
5 . If A = [al,bd X . . . X [an ,bn] and I: A -+ R is suf
ficiently nice, we can apply Fubini's theorem repeatedly to
obtain
6. If C C A X B, Fubini's theorem can be used to evaluate

Icl, since this is by definition IA XB xci. Suppose, for exam
pIe, that
C = [ - l , l J X [ - l , l J - l (x,y) : l ex,y) 1 V1 - X2 or y < - V1 - x2,
xc (x , y)
{� otherwise.
Therefore
/-1l f(x,y) . xc (x,y)dy = J- l vr=z; f(x ,y)dy +
l
fvl _ x, !(x ,y)dy.
-
In generaI, if C C A X B, the main difficulty in deriving

expressions for Jcf will be determining C lì ( { x l X B)
for x E A . If C lì (A X { y } ) for y E B is easier to deter
mine, one should use the iterated integraI
fc f = fB (fA f(x,y) . xc (x,y)dx) dy.
3-23. Let C C A X B be a set of content O. Let
be the set of ali x E A such that { y E B: (x,y) E cl is
Problems.
A' CA
not of content O. Show that A is a set of measure O. Hint: xc is
integrable and JAXB xc J A'U JA.c, so J A'U - .c O. ...
I
= =
3-24. Let C C [0, 1 ] X [0, 1] be the union of alI ( p/q } X [O, l /q], where
p/q is a rational number in [0, 1 ] written in lowest terms. Use C
to show that the word "measure" in Problem 3-23 cannot be
replaced by "content."
3-25. Use induction on n to show that [al,bl] X . . . X [an,bn] is not a
set of measure O (or content O) if ai < bi for each i.
3-26. Let !: [a,b] - R be integrable and non-negative and let AI ...
l (x,y) : a � x � b and O � y � !(x) } . Show that AI is Jordan
measurable and has area f�.f.
3-27. If f: fa,bI X [a,b] - R is continuous, show that
fab f' f(x,y)dx dy = ia ix f(x,y)dy dx.
Hint: Compute J cf in two different ways for a suitable set

C C [a,b] X fa,bI.
3-28. · Use Fubini's theorem to give an easy proof that DI.d - D 2.1f
if these are continuous. Hint: If D I,2!(a) - D 2 , d(a) > O,
O on A .
there is a rectangle A containing a such that D1.2f - D 2 , lf >
3-29. Use Fubini's theorem to derive an expression for the volume of

a set of R8 obtained by revolving a Jordan-measurable set in the
yz-plane about the z-axis.
62 Calculua on Manilolda
3-30. Let C be the set in Problem 1-17. Show that
frO.I) (frO.I) xc(x,y)dX) dy = frO.I) (frO.I) xc(y,X)dY) dx - O

but that f [O.I) X [ O . I) xc does not exist.
3-31. If A = .
[al,bll X . . X [an,bnl and I: A -> R is continuous,
define F: A -> R by
i ra •• x') X . . . X [a,.x')
F(x) = r I.
DJi'(x), for x in the interior of A ?

.....
What is
3-32 . · Let I: [a,bl X [c,d) R be continuous and suppose Dd is con
tinuous. Define F(y) = f�/(x,y)dx. Prove Leibnuz's TUle : F'(y)
=f�D2/(x,y)dx. Hint: F(y) f�(x,y)dx f�<f:D2/(x,y)dy + = =
I(x,c» dx. (The proof wiII show that continuity of D 21 may be

replaced by considerably weaker hypotheses.)
3-33. If I: (a,b) X (c,di -+ R is continuovs and D21 is continuous, define
F(x,y) � f:/(t,lI)dt.
(a) Find DIF and D2F.
(b) If G(x) - H<x)/(t,x)dt, find G'(x).
3-34.· Let gl,g 2 : R 2 -> R be continuously differentiable and suppose
Dlg 2 D2gl. As in Problem 2-21 , let
�
x 'V
f o fo
I(x,y) = g l (t,O)dt + g 2 (x,t)dt.
Show that Dt!(x,y) = g l (X,y).

3-3�.· (a) Let g : Rft -+ Rft be a Iinear transformation of one of the fol
lowing types :
{ g(�) = ei
g(ei) = a ei
{ g(ei) =
+
ei i .= i
g(ei) = ei ek
( g (ek)
g(ei)
=
=
ek
ei
k .= i, i
g (ei) = ei.
If U is a rectangle, show that the volume of g (U) is Idet gl . v(U).
.
(b) Prove that I det gl v(U) is the volume of g(U) for any Iinear
transformation g: R" -+ R n . Hint: If det g .= O, then g is the
composition of Iinear transformations of the type considered in (a).
3-36. (CavaIieri's principle) . Let A and B b e Jordan-measurable sub
sets of R 3 . Let Ae l (x,y) : (x,y,c) E A I and define Be similarly.
=
Suppose each A e and Be are Jordan-measurable and have the same

area. Show that A and B have the same volume.
lntegration 63
PA R TITIONS O F UNI T Y
In this section we introduce a tool of extreme importance in

the theory of integration.
3-11 Theorem. Let A C R and let e be an open cover oj A .
n
Then there is a collection cf> oj C� junctions <(J defined in an open

set containing A , with the jollowing properties:
(1) For each x E A we have O � cp(x) � l .
(2) For each x E A there is an open sei V containing x such that
all but finitely many <(J E cf> are O on V.
(3) For each x E A we have �.-E4><{J(x) = 1 (by (2) for each x
this sum is finite in some open set containing x).
(4) F or each <{J E cf> there is an open set U in e such that <{J = O
outside oj some closed set contained in U.
(A collection cf> satisfying (1) to (3) is called a C� partition of
unity for A . If cf> also satisfies (4), it is said to be sub
ordinate to the cover e. In this chapter we will ollly use
continuity of the functions «J.)
Proof. Case 1. Ais compact.
Then a finite number U l , . . , Un of open sets in e cover A .
.
It clearIy suffices to construct a partition of unity subordinate

to the cover l U I , . U I We will first find compact
. . , n .
sets Di C Ui whose interiors cover A . The sets Di are con

structed inductively as folIows. Suppose that D I , . . . ,Dk
have been chosen so that l interior D I , . . . , interior Dk,
UHI, . . . , Un I covers A. Let
Ck+ l = A (int DI U . . . U int Dk U Uk+2 U . . . U Un ) .
-
Then C H I C Uk+ l is compact. Hence (Problem 1-22) we can

find a compact set Dk+I such that
Ck+ l C interior Dk+l and Dk+ l C UH 1•
Having constructed the sets D I , . . D let t.Pi be a non

. , n,
negative C� function which is positive on Di and O outside of

some closed set contained in Ui (Problem 2-26). Since
64 Calculu8 on ManifQlds
I DI, ,Dnl covers A , we have �I(x) + . . . + �n(x) > O
for all x in some open set U containing A . On U we can define
If f: U [0,1) is a C'" function which is 1 on A and O outside

---t
of some closed set in U, then I f ' CPI, ,f ' cpn l is the
= • • •
desired partition of unity.

Case 2. A = A l V A 2 V A 3 V . . . , where each Ai is
compact and Ai C interior Ai+!.
For each i let 0i consist of all U (ì (interior A i+ I - A i-2 )
for U in 0. Then 0i is an open cover of the compact set
Bi Ai interior Ai-l. By case l there is a partition of unity
<1>; for Bi, subordinate to 0i.
= -
For each x E A the sum

u(x) I cp(x)
", E 4>;, ali i
=
is a finite sum in some open set containing x, since if x E A i we

have cp(x) O for cP E <l>j with j � i + 2. For each cp in
=
each <l>i, define cp'(x) cp(x)/u(x). The collection of alI cp' is

=
the desired partition of unity.

Case 3. A is open.
Let Ai =
Ix E 11 : Ixl � i and distance from x to boundary A � l /i l ,
and apply case 2.
Case 4. A is arbitrary.
Let B be the union of aIl U in 0. By case 3 there is a par
tition of unity for B i this is also a partition of unity for A . I
An important consequence of condition (2) of the theorem
should be noted. Let C C A be compact. For each x E C
there is an open set V. containing x such that only finitely
many cp E are not O on V... Since C is compact, finitely
many such V.. cover C. Thus only finitely many cp E are
not O on C.
One important application of partitions of unity will illus
trate their main role-piecing together results obtained 10calIy.
Integration 65
An open cover e of an open set A C Rn is admissible if

each U E e is contained in A . If cI> is subordinate to e,
I: A -+ R is bounded in some open set around each point of A,
and { x : I is discontinuous at xl has measure O, then each
I A <P · 1 /1 exists . We define / to be integrabIe (in the extended
sense) if �'I' E �I A <P • iii converges (the proof of Theorem 3-1 1
shows that the <p's may be arranged in a sequence) . This
implies convergence of �'I' E �I I A ( but see Problem 3-38) .
3-12 Theore m.
Il it is another partition 01 unity, subordinate to an admis

swle cover e' 01 A, then �.y E i<I A 1/1 . i ii also converges, and
(1)
(2) Il A and I are bounded, then I is integrable in the extended

sense.
(3) Il A is Jordan-measurable and I is bounded, then this defini
tion 01 I AI agrees with the old one.
Proof
(1) Since <p . I O except on some compact set C, and there

=
are only finitely many '" which are non-zero on C, we can

write
This result, applied to III , shows the convergence of �'I' E �

�.y E +I A 1/1 • <P · 1 /1 , and hence o f �'I'E �.y E i< I I A 1/1 • <p . II ·
This absolute convergence justifies interchanging the order
of summation in the above equation ; the resulting double
sum clearly equals �.yE i<I A 1/1 . I. Finally, this result
applied to i ii proves convergence of �.y E i is finite, then
since �\P E F rp � l on A .
(3) If e >O there is ( Problem 3-22)
a compact Jordan-meas
urable C C A such that f A-cl < e. There are only
finitely many rp E «I> which are non-zero on C. If F C «I>
is any finite collection which includes these, and f AI has
ita old meaning, then
continuous function. Show that f (0.1)/ exists if and only if

Problems. 3-37. (a) Suppose that I: (0, 1) ..... R is a non-negative
lim n- SI exists.
2-0
O for x (l any A n . Show that

(b) Let A = [1 - 1 /2", 1 - 1/2"+1J . Suppose that/: (0,1) ..... R
"
satisfies fA .I = ( - 1)" /n and I( x) =
e-O
f (0.1)/ does not exist, but lim f (2.1 -2) I log 2.
=
3-38. Let An be a closed set contained in (n, n + 1). Suppose that

n O for x (l any A n .
Find two partitions o f unity cI> and 'It such that 2; 1" E �f R I(J • I and
I: R ..... R satisfies fA ./ = ( - 1) /n and I =
2; 0/- E ... fR '" . I converge absolutely to different values.
CHA N G E O F VA R IA B L E
If g : [a,b) ..... R is continuously differentiable and I: R ..... R

is continuous, then, as is wel1 known,
g (b) b
f I = f (f o g) . g '.
g (a) a
Integration 67
The proof is very simpl e : if F'f, then (F o g) ' = (f o g) . g ' ;

=
thus the left side is F(g(b» - F(g(a» , while the right side is
F o g(b) F o g(a) F(g(b»
-
= F(g (a» . -
We leave it to the reader to show that if g is 1-1, then the

above formula can be written
J f = J f o g · l g'l ·
(a,b)
g «a,b»
(Consider separately the cases where g is increasing and where

g is decreasing. ) The generalization of this formula to higher
dimensions is by no means so trivial.
3-13 Theore m. Let A C Rn be an open set and g : A � Rn

a 1-1, continuously differentiable function such that det g'(x)
� O for all x E A . If f: g(A) � R is integrable, then
Jf
g (A)
=
1 (f o g) l det g'l ·
Proof. We begin with some important reductions.
1 . Suppose there is an admissible cover (9 for A such that

for each U E (9 and any integrable f we have
J
g(U)
f =
[ (f o g) l det g'l ·
Then the theorem is true for alI of A . (Since g is auto
maticalIy 1-1 in an open set around each point, it is not sur
prising that this is the only part of the proof using the fact
that g is 1-1 on aH of A.)
Proof of (1) . The coHection of aH g ( U) is an open cover of
g(A). Let cI> be a partition of unity subordinate to this cover.
If f{J = O outside of g( U) , then, since g is 1-1, we have (f{J • f) o g
68 Calculus. on M anifollÙ
= O outside of U. Therefore the equation
f
g(U)
tp . I = l [(tp . f) o g] l det g' i ·
can be written
ulA)
( tp . 1 = f [(tp . f) o g] l det g'i ·
A
Hence
fI= L
g(A) IP E � gtA)
( tp . I = L Af [(tp . f) o g] l det g' i
IP E 4>
= L f (tp o g) ( f o g) l det g' l

IP E 4>
A
= f ( f o g) l det g' i ·
A
Remark . The theorem also follows from the assumption
that
[' I = f
g-I(V)
(f o g) l det gl i
for V in some admissible cover of g(A ) . This follows from (1)

applied to g-l .
2. It suffices to prove the theorem for the function I = 1 .

Prool 01 (2) . I f the theorem holds for I = 1 , it holds for
constant functions . Let V be a rectangle in g(A ) and P a par
tition of V. For each subrectangle S of P let Is be the con
stant function m s (f) . Then
L (f,P) = Ls ms (f) v eS) = L f Is

.
S int S
= L f (fs o g) l det g' � L f (f o g) l det g' i

S g-I(int S) S g-I (int S)
� f (f o g) l det g' l ·
g-I(V)
Since f vi is the least upper bound of alI L (f,P) , this proves

that f vi � f g -, ( v ) (f o g) l det g' i . A similar argument, letting
Is = Ms(f) , shows that f vi �
f g-, ( V) (f o g) l det g' i · The
result now follows from the above Remark.
lntegration 69
3.If the theorem is true for g : A � R " and for h : B � R",

where g(A) C B, then it is true for h o g : A R" .
�
Prooj oj (3) .
J j = J j = ( (f o h) l det h' l
Il 0 l/(A) II(I/(A» ulA)
= J [ (f o h) o g] . [ I det h'l o g] · I det g' l
A
= J j o (h o g) l det (h o g)' I .
A
4. The theorem is true if g is a linear transformation.
Prooj oj (.4-) . By (1 ) and (2) it suffices to show for any open
rectangle U that
J
I/(ll)
1 = ! I det g'l ·
This is Problem 3-35.
Observations (3)
and (4) together show that we may assume
for any particular a E A that g'(a) is the identity matrix : in
fact, if T is the linear transformation Dg(a) , then (T-l o g)'(a)
= I ; since the theorem is true for T , if it is true for T-l o g it
will be true for g .
We are now prepared to give the proof, which preceeds by
induction on n. The remarks before the statement of the
theorem, together with ( 1) and ( 2) , prove the case n = 1.
Assuming the theorem i n dimension n - l, w e prove i t in
dimension n. For each a E A we need only find an open set
U with a E U C A for which the theorem is true. Moreover
we may assume that g'(a) = I.
Define h : A � R " by h(x) = (gl(X), . . . , g" - I (X),X").
Then h'(a) = I. Hence in some open U' with a E U' C A ,
the function h is 1-1 and det h'(x) � O. We can thus
define k : h(U') � R" by k(x) = (Xl, . . . ,x"- I,g" (h- l(x)))
and g = k o h. We have thus expressed g as the composition
70
Integration 71
of two maps, each of which ehanges fewer than n eoordinates

(Figure 3-3).
We must attend to a few details to ensure tha t k is a fUllctioll
of the pro per sorto Since
we have D (gn o h- 1 ) (h(a)) D gn(a) ] , so Lhat k'(h(a) )

= =
= I.
n n
Thus in some open set V with h (a) E V C h ( U') , the
function le is 1 - 1 and det Te'(x) � O. Letting U k- 1 ( V) =
we n ow have g le o h, where h : U ---> Rn and k : V ---> Rn
=
and h ( U) C V . By (3)
it suffices t.o prove the theorcm for h
alld le. We give the proof for h ; the proof for le is sirnilar
and easier.
Let W C U be a rectangle of the form D X [a " , b L whcrc r
n
D is a rectangle in Rn- l . By Fubi ni's theorem
11= 1( 1 I dx l . . . dxn- 1 ) dx" .

h( W) [a. ,b.] h ( D X { x, ) )
Let hx" : D ---> Rn- I b e defined by hx ( x l . , . . . ,xn- l )

I .
(g l (X , •
. ,xn) , . ,gn- l (x l , . . . ,xn ) ) . Then each hx"
is clearly l-I and
det. (hx.) ' (x 1 , . . ,xn- 1 ) = det h ' (x 1 , . . . ,xn) � O.
Moreover
1 1 dX l = 1 1 dX l . . .
d.T- n- 1 •
h(D X
. . . dxn - I
I x' ) ) h,"(D)
Applying the t.heorem in the ease n - l t.herefOl'e givcs
1
1 = 1 ( 1 dx I . . . dxn- 1 ) dx" l
h( W) [a.,b ..] h.' (D)
1 ( I I det(hxu) ' (.c l , . . ,:t;n - J ) l dx " . . dxn- 1 ) dxn

[a .. ,b ..] D
.
1 ( I I det h' (x 1 , . . . ,xn) l dx J dxn- l ) dx"

ran,b,,] D
• • •
1 I det h ' l · I
W
The condition det g'(x) � O may be c1 imiuated from the

72 CalculuB on Mani/olds
hypotheses of Theorem 3-13 by using the following theorem,
which often pIays an unexpected roIe.
3-14. Theorem (Sard's Theorem). Let g: A -+ Rn be con

tinuously differentiahle, where A C Rn is open, and let B =
I x E A : del g'(x) = O } . Then g(B) has measure O.
Proof. Let U C A be a closed rectangle such that alI sides

of U have Iength l, say. Let e O. If N is sufficientIy Iarge
>
and U is divided into Nn rectangIes, with sides of Iength ilN,
then for each of these rectangIes S, if x E S we have
I Dg(x) (y - x) - g(y) - g(x) 1 < e l x - y l � e Vn (lIN)

for alI y E S. If S intersects B we can choose x E S (ì B ;
since det g' (x) = O, the set I Dg (x) (y - x) : y E S } Iies in an
(n l)-dimensionaI subspace V of Rn• Therefore the set
-
I g (y) - g(x) : y E S } Iies within e Vn (lIN) of V, so that

Ig(y) : y E S I Iies within e Vn (lIN) of the (n I)-pIane -
V + g(x) . On the other hand, by Lemma 2-10 there is a

number M such that
I g(x) - g(y) 1 < Ml x - y l � M Vn (lIN) .

Thus, if S intersects B, the set {g(y) : y E S } is contained in
a cyIinder whose height is < 2e: Vn (lIN) and whose base is an
(n - I)-dimensionaI sphere of radius < M Vn (l/N). This
cylinder has volume < C(lIN)ne for some constant C. There
are at most Nn such rectangles S, so g(U (ì B) Iies in a set of
volume < C(lIN)n . e . Nn = Cln . e . Since this is true for
alI e O, the set g ( U (ì B) has measure O. Since (ProbIem
>
3-13) we can cover all of A with a sequence of such rectangIes
U, the desired result foIIows from Theorem 3-4. I
Theorem 3-14 is actualIy onIy the easy part of Sard's

Theorem. The statement and proof of the deeper resuIt wilI
be found in [17], page 47.
the 8.Bsumption det g'(x) � o.

Problemll. 3-39. Use Theorem 3-14 to prove Theorem 3-13 without
IntegTation 73
3-40. If g: Rn ---> Rn
and det g'(x) ;é. 0,
prove that in some open set
containing x
we can write = T o
g o • • •
gn wheregl, gi
is of o
the form gi(X) = and T ' is a linear

(xl, . . . ,!i(X), . . . ,xn),
transformation. Show that we can write o • • • o if
g gn gl =
and only if g'(x)

is a diagonal matrix.
3-41. Define I: {T: T > 01 (0, 11') R2 I(T,8) (T 8, T 8).
X 2 ---> by COS sin =
l'(T,8),
° for ali is
( a) Show that 1 is l-l, compute and show that
det ( /' T,8) ;é. (T,8).
Show that I( {T: T > 01 (0,211'»
X
the set A of Problem 2-23.
(b) If r l, show that
P = = P(x,y)
where (T(X,y),8(x,y» ,
T(X,y)
8(x,y)
=
=
l
Vx2 + y2,
r a
: � ;:/a: y/x
211' + arctan y/x
x > 0, y > 0,
x < 0,
x > 0, y < 0,
x = 0, y < O.
11'/2 x 0, y > 0,
=
311'/2
( Here arctan denotes the inverse of the function tan: ( -11'/2,11'/2)

---> R.)Find P'(x,y).
The function is called the polar coor P
dinate system on A .
( c) Let G C A b e the regio n between the circles of radii TI
and the half-lines through ° which make angles of
and
T2 and with 81 82
the x-axis. If G ---> h: R
is integrable and h(x,y) = g(T(X,y) ,8(x,y» ,
show that
Tt 8'!
J h = J J Tg(T,8)d8 dT.
C r, 8,
jh
r 2..
J J Tg(T,8)d8 dT.
O O
=
(d) If Gr = [-T,TJ X [ -T,TJ, show that

J e-(x'+II') dx dy = 11'( 1 - e-r')
B,
and
dx
r
f e- (x'+II') dx dy ( f e-x' ) 2 . =
�
-r
,
(e) Prove that
lim f e-(x'+II') dx dy = lim

T� CO
f e-(x'+II') dx dy
�oo Br Cr
Calculu8 on M anifolds
and conclude that
dx
00
J e-z •
= -V;.
"A mathematician is one to whom that is as obvious as that twice

two makes four is to you. Liouville was a mathematician. "
-LORD KELVIN
4
I ntegration on Chains
A L G E B R A I C P R E L IMINA R IE S
If V i s a vector space (over R), w e will denote the k-fold

product V X . . . X V Vk . by function T: Vk ---+ R is
A
called muItilinear if for each i with 1 S i S k we have
T(VI, . . . ,Vi + v/, ,Vk ) = T(Vl, . . . ,Vi, . . . ,Vk )
+ T(VI, . . . ,V;', . . . ,Vk) ,
T(VI, . . . ,aVi, ,Vk ) = aT(vr, . . . ,Vi, . . . ,Vk ).
A multilinear function T: Vk
---+ R is called a k-tensor on V
and the set of all k-tensors, denoted ::lk (V), becomes a vector
space (over R) if for S,T E ::l\V) and a E R we define
(S + T) (VI , . . . ,Vk ) S(VI, . . . ,Vk ) + T(Vl , . . . ,Vk ) ,
=
(aS) (vI , . . . ,Vk ) a . S(VI, . . . ,Vk).
=
There is also an operation connecting the various spaces ::lk ( V).

I f S E ::lk ( V) and T E ::ll(V), w e define the tensor product
S @ T E ::lHI(V) by
S @ T(VI, . . . ,Vk ,Vk +l, . . . ,Vk +l )
= S(VI , . ,Vk ) . T(Vk + l , . . . ,Vk+l).
75
Note that the order of the factors 8 and T is crucial here since
8 @ T and T @ 8 are far from equa!. The following prop
erties of @ are left as easy exercises for the reader.
(8 1 + 8 2) @ T = 8 1 @ T + 8 2 @ T,
8 @ (T I + T 2) = 8 @ T I + 8 @ T 2,
(a8) @ T = 8 @ (aT) = a(8 @ T),
(8 @ T) @ U = 8 @ (T @ U) .
Both (8 @ T) @ U and 8 @ ( T @ U) are usually denoted
simply 8 @ T @ U ; higher-order products TI @ . . . @ T.
are defined similarly.
The reader has probably already noticed that � l (V) is j ust
the dual space V* . The operation @ allows us to express the
other vector spaces �k (V ) in terms of � l (V) .
4-1 Theorem. Let V I , . . . ,Vn be a basis far V, and let

'P I , . , 'Pn be the dual basis, 'Pi (Vj) = 8ij. Then the set of all
. •
k-fold tensor products

'Pi, @ . . @ 'Pi. •
is a basÌ8 far �k (V), which therefore has dimension nk .
Proof. Note that
. ,vi.)
= 8 i, .ù · . . . . 8i., i.
=
{ Ol if iI = il , .
otherwise.
If Wl, . . . ,Wk are k vectors with Wi = �j=l aijVi and T is in

�k (V), then
n
\'
Lt al "
,31
il, . . . .i,,= l
l
n
T (Vill ' " ,Vi.) . 'Pi, @ • • . @ 'Pi,(W l , . . . ,Wk) .

il, . . . , i.. = l
Thus T . . . ,Vi.) ' 'Pi, @ . . . @ 'Pi. ·
= �?, . . . ,i. - l T(Vil l
Consequently the 'Pi, @ . . . @ 'Pi. span �k ( V) .
Integration on ehains 77
Suppose now that there are numbers ai, . . . . . i. such that

n
ù, _
L
• • ,it - l
ai,. . . . .i <Pi,
•
.
® . . . ® <Pio = O.
Applying both sides of this equation to (Vii, . . . ,vi» yields

aj,. . . . .j. O . Thus the <Pi, ® . . . ® <Pi.
= are linearly
independent. I
One important construction, familiar for the case of dual

spaces, can also be made for tensors. If f: V � W is a linear
transformation, a linear trans{ormation f* : :Jk (W) � :Jk (V)
is defined by
f*T(V I , . . . ,Vk) = T(f(V I ), . . . ,j(Vk))
(or T E :Jk (W) and V I , ,Vk E V. I t i s easy
• . . to veri{y
that f*(8 ® T) = f*8 ® f*T.
The reader is already {amiliar with certain tensors, aside
(rom members of V*. The first example is the inner product
( ,) E :J2(Rn). On the grounds that any good mathematical
commodity is worth generalizing, we define an inner product
on V to be a 2-tensor T such that T is symmetric, that is
T(v,w) = T(w,v) (or v,w E V and such that T is positive
definite, that is , T(v,v) >
O i{ V � O. We distinguish ( ,) as
the usual inner product on Rn • The (ollowing theorem
shows that our generalization is not too generaI.
4-2 Theorem. If T is an inner product on V, there is a

basis V I , . . • ,Vn for V such that T(Vi,Vj) = �ij. (8uch a
basis is called orthonormal with respect to T.) Consequently
there is an isomorphism f: Rn V 8uch that· T(f(x),j(y)) �
(x,y) for x ,y E Rn . In other words f* T = ( ,) .

Proof. Let Wl , . . ,W", be any basis {or V. Define
etc.
78 CalCUlU8 on Mani/old8
It is easy to check that T(w/,w/)
w/ � O so = O if i � j and
that T(w/,w/) > O . Now define Vi w//,VT(w/,w/) . The =
isomorphism I may be defined by I( e i) = Vi. I
Despite its importance, the inner product plays a far lesser

role than another familiar, seemingly ubiquitous function,
the tensor det E :Jn (Rn) . In attempting to generalize this
function, we recall that interchanging two rows of a matrix
changes the sign of its determinant. This suggests the fol
lowing definition. A k-tensor w E :Jk (V) is calIed alternating
if
W(Vi, . • • ,Vi, • .,Vi>

• . ,Vk)
- W(Vi, • ,Vi l . . . ,Vi, .
• . ,Vk)
for alI Vi, • . ,Vk E V.
.
(In this equation Vi and Vj are interchanged and alI other v's
are left fixed.) The set of alI aIternating k-tensors is clearly
a subspace Ak (V) of :Jk (V). Since it requires considerable
work to produce the determinant, it is not surprising that
alternating k-tensors are difficult to write down . There is,
however, a uniform way of expressing alI of them. Recall
that the sign of a permutation 0', denoted sgn 0', is if O' is +1
even and 1 if O' is odd. If T E :Jk (V), we define Alt( T) by
-
Alt( T) (v l , . . . ,Vk) = :! 2:E S sgn O' • T (V<T( l ) , . . . ,V.( kl ),

<T .
where Sk is the set of alI permutations of the numbers 1 to k.
4-3 Theorem
(1) Il T E :Jk (V), then Alt(T) E Ak (V).

(2) Il w E A k (V ) , then Alt(w) w. =
(3) Il T E :Jk (V), then Alt ( Alt ( T)) Alt (T). =
Proo!
(1) Let ( i,j) be the permutation that interchanges i and j and

leaves alI other numbers fixed. If O' E Sk , let 0" =
O' • (id). Then
Integration on Chains 79
Alt(T)(Vl, • ,vi> . . ,Vi, . . . ,Vk)

• • .
= �! L sgn T(V�(l) , . . . ,V�(j), . .

(1 • ,V�(i) , . . , Va ( k
. )
� E S.
= �! L sgn T(v�'( l" . . . ,V�'(i) , . . . ,Va'(j) ,
(1 • . • . ,V�' ( k )
� E S.
= � L -sgn T(v�'(l), ,Va'(k)
(1
' •
• . .
a' E S.
= -Alt(T)(Vl, ,Vk). • • •
(2) If w E Ak (V), and (i,j), then W(Va(l), . . . ,Va( k ) =

(1 =
sgn W (V , . . . ,Vk). Since every is a product of per

(1 • l (1
mutations of the form (i,j), this equation holds of aB (1.
Therefore
Alt(w)(Vl, . ,Vk) ft L sgn W (Va( l ) , . . . ,Va ( k )
= (1 •
a E s.
= �! L sgn (1 ' sgn W(V
(1 ' l, . . . ,Vk )
a E S.
= W (VJ, . . . ,Vk ) .
(3) follows immediately from ( 1 ) and (2) . I
To determine the dimensions of Ak (V), we would like a
theorem analogous to Theorem 4-1. Of course, if w E Ak (V )
and TJ E A l (V), then w ® TJ is usually not in AHl(V). We
will therefore define a new product, the wedge product
w 1\ TJ E AH l (V) by
w 1\ TJ = (kk+.' l lol) ! Alt(w ® TJ) .
(The reason for the strange coefficient will appear later.) The
following properties of 1\ are left as an exercise for the reader:
( + W2 ) 1\ TJ
Wl l 1\ TJ + W 2 1\ TJ,
= W
w 1\ (TJ l + TJ 2) w 1\ TJl + W 1\ TJ 2 , =
aw 1\ TJ w 1\ aTJ a(w 1\ TJ), = =
w 1\ TJ ( - 1 ) k lTJ 1\ w, =
f* (w 1\ TJ) = f* (w) 1\ f* (TJ) .

80 CalCUlU8 on Manifolds
The equation (w 1\ TI) 1\ 8 = w 1\ (TI 1\ 8) is true but
requires more work.
4-4 Theorem
(1) liS E :lk (V) and T E :l l ( V) and Alt(S) = O, then

Alt(S @ T) = Alt( T @ S) = O.
(2) Alt(Alt(w @ TI) @ 8) = Alt(w @ TI @ 8)
Alt(w (8) Alt(TI @ 8» .=
(3) Il w E Ak (V), TI E A1(V), and 8 E Am( V) , then
(w 1\ TI) 1\ 8 = w 1\ (TI 1\ 8)
TI
(k + l + m) l
= Alt(w @ @ 8) .
kl llml
Prooj
(1)
(k + l) l Alt(S @ T) (vlJ . . . ,Vk+l)
) sgn S(vv(1), ,Vv (k» . T (Vv(k+I), . . . ,Vv(k+l ) '
v E'1, .+1
u · • . .
If G C Sk+ l consists of aH u which leave k + l , . . . ,

k + l fixed, then
l sgn u . S(Vv(I) , . . . ,Vv(k» . T(Vv(k+I) , . . . ,Vv(k+l)

"EG
= [l sgn u
' .
S(Va' ( I), . . . ,Va'(k» ] . T(Vk+l, . . . ,Vk+l )
,,' E S.
= O.
Suppose now that Uo El G. Let G . <To = l <T • uo : <T E Gl
and let Vao ( l ) , . . . ,Vao(k+l ) = Wl , . . • ,Wk+ l ' Then
) sgn <T ' S ( Va(I), . . . ,Va (k» . T (Va(k+I) , . . . ,Va (k + l»

v E'-t;·".
[sgn <TO ' l sgn <T' S ( Wa'(Il > . . ,Wa'(k» ]
,,' E G
= •
= o.
�otice that G (\ G . [25. In fact, if o- E G (\ G .

0-0 = 0- 0,
then o- = . for some u' E G and = o- . ( ) E G,

0-' Uo 0- 0 0-
' -1
a contradiction. We can then continue in this way,

breaking 8k+z up into disjoint subsets; the sum over each
subset is O, so that the sum over 8k+l is O . The relation
Alt(T 0 8) O is proved similarly.
=
(2) We have
Alt(Alt(71 0 9) - 71 0 9) Alt(71 0 8) - Alt(71 0 9) O .

= =
Hence by (1) we have

O Alt(w 0 [Alt(l1 0 9) - 71 0 9] )
=
=Alt(w 0 Alt(l1 0 9)) - A1t(w 0 11 0 9).

The other equality is proved similarly.
(k + l + m) !
(3) (w 1\ 11) 1\ 8 = (k + l) ! m !
Alt« w 1\ 11) 0 9)
_ (k + l + m) ! (k + l) !
- (k + l) ! m ! k! l! Alt(w 0 11 0 9).
The other equality is proved similarly. I
Naturally w 1\ (11 1\ 9) and (w 1\ 11) 1\ 9 are both denoted

simply w 1\ 11 1\ 9, and higher-order products W l 1\ . . . 1\ Wr
are defined similarly. If Vl, ,Vn is a basis for V and
. . •
'l' l , . . . , 'l'n is the dual basis, a basis for Ak (V) can now be
constructed quite easily.
4-5 Theorem. The set 01 all
'Pi, 1\ . . . 1\ 'l' ik
is a basis 101' Ak (V), which therelore has dimension
(n) =
n!
'
k k!(n - k) !
Proof. If w E Ak (V) C :Jk (V), then we can write

w = I ail ik 'l'il 0 . . . 0 'l'ik'
il, . . . ,ile
• . . . .
Thus
w = Al t (w) L
il, . . ,it
.
ai , . . . . . ikAlt( \Pi, @ . . . @ \Pik) '
Sin ce each Alt(\pi, @ @ \Pik) is a constant times one of the

• . .
\Pi, À . . . À \Pik ' these elements span A k (V). Linear inde

pendence is proved as in Theorem 4-1 (cf. Problem 4-1 ). I
If V has dimension n, it follows from Theorem 4-5 that
An ( v) has dimension 1 . Thus alI alternating n-tensors on V
are multiples of any non-zero one. Since the determinant is
an example of such a member of An ( Rn ) , it is not surprising
to find it in the following theorem.
4-6 Theorem. Let v!, . . . ,Vn be a basis for V, and let
w E A n (v). If Wi = 2;i_ laijVj are n vectors in V, then
7l« a l l , . . . ,a l n ) , . . . , (an i , . . ,ann ))

= w(2;a l jvi> . . . , 2;a jvj ) .
n
Clearly 71 E An (Rn) so 71 = À det for some À E R and À •
7l(e t , . . . ,en ) = W(VI, • • • ,vn ). I
Theorem 4-6 shows that a non-zero w E An ( V) splits the

bases of V into two disjoint groups, those with W (VI, ,vn) • . .
> O and those for which W(Vl, . . . ,vn) < O; if V I, . . . ,Vn

and W l , . . . ,Wn are two bases and A (aij) is defined by =
Wi 2;a ijvj , then V I , . . . ,Vn and W l , . . . ,Wn are in the

=
same group if and only if det A > O. This criterion is inde

pendent of and can always be used to divide the bases of V
w
into two disjoint groups. Either of these two groups is

called an orientation for V. The orientation to which a
basis V I , ,Vn belongs is denoted [V I ,
• • . ,vnl and the • • •
other orien tation is denoted - [VI, . . . ,vnl . In Rn we define

the usual orientation as [e l , . . . ,en l .
The fact that dim An (Rn) 1 is probabIy not new to you,
=
since det is often defined as the unique element w E An (Rn )

such that w(e" . . . ,e ) = 1 . For a generaI vector space V
there is no extra criterionn of this sort to distinguish a particular
w E A n (v). Suppose, however, that an inner pro due t T for
V is given. If VI, ,Vn and . . . ,Wn are two bases
• • . Wl,
which are orthonormal with respect to T, and the matrix

A = (aij) is defined by = �;= laijVj, thenWi
l aikajz T(Vk,V[)
k,l = ,
n
l a ikajk .
k=I
1n other words, if A T denotes the transpose of the matrix A ,

then we have A . A T = I, so det A ± 1 . It follows from =
Theorem 4-6 that if w E An ( V) satisfies W (VI, . . . ,Vn) ± 1, =
then ( w WI, . , n) ± 1 . If an orientation }J. for V has

. . W =
also been given, it follows that there is a unique w E An (V)

such that W(VI, . . . ,Vn ) = 1 whenever VI, . . . ,Vn is an
orthonormal basis such that [VI, ,vnl = }J.. This unique
• • .
w is called the volume element of V, determined by the

inner product T and orientation }J.. Note that det is the
volume element of Rn determined by the usual inner product
and usual orientation, and that I det(v I, . . . ,Vn) I is the vol
O to each
ume of the parallelipiped spanned by the line segments from
of VI, . ,Vn . . •
We conclude this section with a construction which we will

restriet to Rn . If VI, ,Vn_ 1 E Rn and <P is defined by
. . •
<p W( ) = det
then <P E A I (Rn) ; therefore there is a unique z E Rn such that
(w,z) <p(w) det

= =
This z is denoted V I X . . . X Vn-l and called the cross

product of V1, • . . ,V -l . The following properties are
n
immediate from the definition :
V1
V'C l) X .
X
..
X
X V.. (n- l)
aVi X "
=
X Vn - 1
sgn (J'
=
• V 1 X . . . X Vn - l ,
a ' (V I X ' X Vn - 1 ) ,
VI X X (Vi + vi') X . . . X Vn_ l
= VI X X V i X . . . X Vn _ 1
+ V1 X X V;' X . . . X Vn_l.
It is uncommon in mathematics to have a "product" that

depends on more than two factors. In the case of two vectors
v,w E R 3 , we obtain a more conventional looking product,
V X w E R 3 • For this reason it is sometimes maintained
that the cross product can be defined only in R3 .
Problems.
'1'1, • .
. , 'l'n be the dual basis.
(a) Show that 'l'i, /\ • • . /\ 'l'i. (ei" . ,ei.).. -
4-1. · Let e l , . . . ,en be the usual basis of R n and let
l. What
would the right s(de be if the factor (k + l) !jk!l! did not appear in
the definition of /\ ?
(b) Show that 'l'i, /\ . . . /\ 'l'i. (VI, . • . ,Vk) is the determinant
of <h. k X k min" of () obtain.d by ..I"tin. oolumn.
4-2. If
il, . . . ,ù.
f: V -+ V is a linear transformation and dim V n, then
=
r : A n (V) -+ A n (V) must be multiplication by some constant c.
Show that c det f.
=
Integration on Chain8 85
4-3. If w E A n (v) is the volume element determined by T and /A, and

Wl, o o o ,Wn E V, show that
Iw (w l , o o o ,wn ) 1 = Vdet (gij)'

where gij = T (Wi,Wj) Hint: If VI, o o o ,Vn is an orthonormal
o
basis and w; 2:j!..1 ai}1Jj, show that gij

= 2:;;_1 aikakjo
=
4-4. If w is the volume element of V determined by T and /A, and

,: R n --> V is an isomorphism such that f*T = (,l and such that
[/(el), o o , I(e n »)
o /A, show that f*w = deto
=
4-5. If c: [0,1) --> (R n ) n is continuous and each (C l (t), o ,cn (t) is

o o
a basis for R ", show that [C l (O), o o ,c n (o»)

o [cl (l), o o ,c n (l»)o
= o
Hint: Consider det o co

4-6. (a) If V E R 2, what is V X ?
(b) If V b o o o ,V n- l E R" are Iinearly independent, show
that [VI, o o o,Vn-l, VI X . . . X Vn- l) is the usual orientation of
R n.
4-7. Show that every non-zero w E A n ( V) is the volume element
determined by some illner product T and orientation /A for V.
4-8. If w E A n ( v) is a volume element, define a "cross product"
VI X o . . X Vn -l in terme of w.
4-9. * Deduce the foIIowing properties of the cross product in R 3 :
(a) el X el O
= e2 X el -ea = ea X el = e2
el X e2 = e 3 e2 X e2 = O ea X e2 = - e l
e l X ea = - e 2 e 2 X ea = el ea X ea O. =
2
(b) v X W = (v2w3 - v3w )el
+ (v 3w - v w3)e2
l l
+ (v W 2 - v w ) e .
l 2 l
3
(c) Iv X wl = Ivl · lwl · I sin 81, where 8 = L (v,w).
(v X w, v) = (v X w, w) = O.
(d) (v, w X z) = (w, z X v) = (z, v X w)
V X (w X z) (v,z)w - (v,w)z
=
(v X w) X z = (v, z)w - (w,z)v.

(e) I v X w l = V(v,v) ' (w,w) - (V,W) 2.
4-10. If W l , o
. . ,Wn-l E R n, show that
IWl X . . . X wn-l l = Vdet (gij),
where gij = (w;,Wj)o Hint: Apply Problem 4-3 to a certain
(n - I )-dimensionaI subspace of Rn•
4-1 1 . If T is an inner product on V, a linear transformation I: V --> V
for x,y E Vo If VI, • • o ,Vn is an orthonormal basis and A = (ai;)

is cali ed self-adjoint (with respect to T) if T(x,/(y» T (f(x),y) =
is the matrix of I with respect to this basis, show that aij aji. =
4-12. If /I, . . . ,In -l : Rm --> R n, define h X . . . X In -l : Rm --> R"

by /I X . . . X f,,- l( p) = h( p ) X . . . X fn -l( P ) . Use Prob-
lem 2-14 to derive a formula for D(h X . . " X 'n -l) when /I ,
. . . ,jn -l are differentiable.
FIE L D S A ND F O R MS
Ii p E R n ,the set of alI pairs (p,v), for v E Rn, is denoted

Rn and called the tangent space of Rn at p. This set is
p,
made into a vector space in the most obvious way, by defining

(p,v) + (p,w ) = (p, v + w) ,
a · ( p ,v ) = (p,av).
A vector v E Rn is often pictured as an arrow from O to v; the
vector (p,v) E Rnp may be pictured (Figure 4-1) as an arrow
with the same direction and length, but with initial point p.
This arrow goes from p to the point p + v, and we therefore
p +v
'/ p
FI G UR E 4-1
lntegration on Chains 87
define p + v to be the end point of (p,v) . We will usually

write (p,v) as vp (read: the vector v at p).
The vector space Rnp is so closely allied to Rn that many
of the structures on Rn have analogues on Rnp ' In particular
the usual inner product ( ,)p for Rnp is defined by (vp,wp)p =
(v,w), and the usual orientation for Rnp is [ ( e l )p, . . . , (en )p).
Any operation which is possible in a vector space may be
performed in each Rnp, and most of this section is merely an
elaboration of this theme. About the simplest operation in a
vector space is the selection of a vector from it. Ii such a
selection is made in each Rn p, we obtain a vector field (Figure
4-2). To be precise, a vector field is a function F such that
F( p) E Rn p for each p E Rn • For each p there are numbers
F 1 (p), . . . ,Fn (p) such that
We thus obtain n component functions Fi : Rn - R. The

vector field F is called continuous, differentiable, etc., if the
functions Fi are. Similar definitions can be made for a vector
field defined only on an open subset of Rn. Operations on
vectors yield operations on vector fields when applied at each
point separately. For example, if F and G are vector fields
FIG U R E 4-2
BB CalCUlU8 on Manifolds
and f is a function, we define
(P + G) (p) P(p) + G(p), =
W ,G)(p) = (P (p),G(p» ,
(f ' P) (p) = f( p)P(p) .
If P l, . . . ,P..-l are vector fields on Rn, then we can simi
larly define
Wl X . . . X = Pl(p) X . . . X p.. - l(p).
Pn - l ) (p)
Certain other definitions are standard and useful. We define
the divergence, div P of P, as !'f_ 1 DiPi. If we introduce
the formaI symbolism
n
V =l D i ' ei ,
;=1
we can write, symbolically, div P = (V,P). If n= 3 we

write, in conformity with this symbolism,
(V X P) (p) = (D2P3 - D 3p 2) (e l )p
+ (D 3P l - D lp 3) (e2)p
+ (DlP 2 - D2p l) (e 3 )p,
The vector field V X P is called curI P. The names "diverg

ence" and "curI" are derived from physical considerations
which are explained at the end of this book.
Many similar considerations may be applied to a function
w with w (p) E A k ( Rnp) ; such a function is called a k-form on
R", or simply a differential formo If <Pl(P), . . . , <Pn (P)
is the dual basis to (e l)p, . . . ,(en)p, then
w(p) = l W;l • • • • •ik(P) . [<Pil (P) 1\ . . . 1\ <Pik(P» )
;.< . . . <il
for certain functions Wil • • . . • ik ; the form is called continuous,

w
differentiabIe, etc., if these functions are. We shall usually

assume tacitly that forms and vector fields are differentiabIe,
and "differentiable" will henceforth mean "C�" ; this is a
simplifying assumption that eliminates the need for counting
how many times a function is differentiated in a proof. The
sum w + 71, product f . w, and wedge product w 1\ 71 are defined
lntegration on Chain8 89
in the obvious way. A function I is considered to be a O-form

and I w is also written I 1\ w.
.
If I: Rn - R is differentiable, then DI(p) E A I (Rn) . By a

minor modification we therefore obtain a l-form dI, defined by
dl(p) (vp) = DI(p) (v).
Let us consider in particular the l-forms d'Trio It is customary
to let Xi denote the lunction 'Tri. (On R 3 we often denote
x l , x 2, and x 3 by x, y, and z . ) This standard notation has
obvious disadvantages but it allows many classical results
to be expressed by formulas of equally classical appearance.
Since dxi(p)(vp) d'Tri(p) (vp) = D'Tri(p)(v) Vi, we see that
= =
dX I (p) , . . . ,dxn (p) is just the dual basis to (el)p, . . . ,(en )p.
Thus every k-form w can be written
w I Wil , ik dxit 1\
it< ' " <il:
= . . . •
The expression for dI is of particular interest.

4-7 Theorem. Il I: Rn - R is differentiable, then
dI = DJ ' dx l + . . + Dn/ ' dxn..
In classical notation,
Proof. dl(p) (vp) = DI(p)(v) = � �l Vi • D;f(p)

= �i- l dXi(p) (vp) . D;/(p)· I
If we consider now a differentiable function I: R n Rm we _
have a linear transformation DI(p) : Rn Rm• Another _
minor modification therefore produces a linear transformation

1* : Rnp - Rmf(p) defined by
This linear transformation induces a linear transformation

1*: Ak (Rml(p») - Ak (Rnp) . If w is a k-form on Rm we can
therefore define a k-forml*w on R n by (I*w)(p) = I*(w(f(p) ) ) .
90 Calculull on Manifolds
Recall this means that if VI, ,Vk E R" p, then
we have
f*W(P) (V I , . . . ,Vk) = As an
w (f(p» (f* (Vl), . . . J* (v,,) .
antidote to the abstractness of these definitions we present
a theorem, summarizing the important properties of f*, which
allows explicit calculations of f*w.
4-8 Theorem. If f: R" -- Rm is differentiable, then
afi .
��] =1 D JJ·l" · dx] ��=1 -. dx] .
. . .
( 1 ) f"' (dx')
= =
] ax]
(2) f*(Wl + W 2) f* ( WI ) + f*(W 2) .
=
(3) f *(g . w) = (g o f) . f*w.

(4) f* (w 1\ '1) f*w 1\ f*'1 .
=
Proof
(1 ) f*(dxi) (p) (vp) dxi (f(p» (f", vp)

=
= dx i(f(p » (�j� lVj . Djl (p), .
= � j_ l Vj . Djf(p)
= �i=IDjf (p) . dxj (p) (vp).
The proofs of (2) , (3) , and (4) are left to the reader. I
By repeatedly applying Theorem 4-8 we have, for example,

f* (P dX l 1\ dX 2 + Q dX 2 1\ dx3 ) =
(P o f)[f*(dxl) 1\ f*(dx 2)]
+ ( Q o f)U"' (dx 2) 1\ f* (dx3 )] .
The expression obtained by expanding out each f*(dxi) is quite

complicated. (It is helpful to remember, however, that we
have dXi 1\ dX i ( - l)dxi 1\ dxi O.) In one special case it
= =
wilI be worth our while to make an explicit evaluation.

4-9 Theorem. If f: Rn -- Rn is differentiable, then
f* (h dX l 1\ . . 1\ dxn ) =
(h o f) (detf') dx l 1\ . . . 1\ dxn .
Proof. Since
lntegralion on Chains 91
it suffices to show that

f*(dx 1 1\ . . .
(det f') dX l 1\ . . . 1\ dxn.
1\ dxn) =
Let p E R n and let A (a;j) be the matrix of f'(p) . Rere,=
and whenever convenient and not confusing, we shall omit

"p" in dX I 1\ . . . 1\ dxn (p) , etc. Then
f*(dx1 1\ . . . 1\ dx n ) (e l , . . . ,en )
= dX l 1\ 1\ dxn (f e l , . . . ,!", en )
*
by Theorem 4-6. I
An important construction associated with forms is a gen

eralization of the operator d which changes O-forms into
l-forms. If
w = I Wil, . , . , ik dXi, 1\
... it< <i.
we define a (le + l ) -form dw, the differential of w, by
dw = I dWi" . . . , ik 1\ dxi, 1\ . . 1\ dXik
it< · · · <i.
n
�
� � D (w i l, . . . , ik) . dxet 1\ dX i l
� et 1\ . . . 1\ dxit .
ù< " ' <ù a = l
4-10 Theorem
(1) d(w + 7/) = dw + d7/.

(2) If w is a k-form and 7/ is an l-form, then
d(w 1\ 7/) = dw 1\ 7/ + ( - 1) kw 1\ d7/.
(3) d(dw) O. Briefly, d 2 = O.

=
(4) If w is a k-form on Rm and f: Rn -- Rm is differentiable,

then f*(dw) = d(f*w) .
Proof
(1) Left to the reader.

(2) The formula is true if and w dXil /\ /\ dxik
/\ /\
=
7] = dxi l . . . dxiz,
since alI terms vanish. The
formula is easily checked when w is a O-formo The gen
eraI formula mf1.y be derived from (1) and these two
observations.
(3) Since
n
dw l l D",(Wil ' . . . , iI,)dx'" /\ dXil /\

'll < " ' <u a = l
= . .
we have
n n
d (dw) l l l D",,8 (Wi l, . . . . ik )dx8 /\ dx'"

'" 1 Il
=
i. < . . . <i. = = 1
In this sum the terms

D",.8 (Wil . . . . , ik )dx8 /\ dx'" /\ dxil /\
and
cancel in pairs.
(4) This is clear if w is a O-formo Suppose, indurtively, that
(4) is true when w is a k-form. It suffices to prove (4) for
a (k + l)-form of the type w /\ dxi. We have
f*(d(w /\ dXi)) = f*(dw /\ dxi + ( - 1 ) kw /\ d(dxi))
f*(dw /\ dxi) = f*(dw) /\ f*(dxi)
= d(f*w /\ f*(dx i))
=
by (2) and (3)

= d(f*(w /\ dxi)) . I
A form w is called closed if dw = O and exact if w = d7], for

some 7]. Theorem 4-10 shows that every exact form is closed,
and it is natural to ask whether, conversely, every closed form
is exact. If w is the l-form P dx + Q dy on R 2, then
dw (D1P dx + D 2P dy) /\ dx + (D 1 Q dx + D 2 Q dy) /\ dy
(D1Q - D 2P)dx /\ dy.
=
Inlegration on Chains 93
Thus, if dw = O, then D I Q = D 2P. Problems 2-21 and 3-34
show that there is a O-form I such that w = dI = Dd dx +
DJdy. If w is defined only on a subset of R2, however, such
a function may not exist. The classical exampie is the form
-y x
w= 2 dx + dy
x + y2 x2 + y2
defined on R 2 - O. This form is usually denoted dO (where
O is defined in Problem 3-4 1 ) , since (Problem 4-2 1 ) it equals dfJ
on the set { (x,y) : x < O, or x � O snd y ;é O } , where O is
defined. Note, however, that O cannot be defined continuously
on alI of R 2 - o. If w = dlfor some function/: R% 0 -+ R, -
then Dd = D I O and Dd = D 20, so I = O + constant, show

ing that such an I cannot exist.
Suppose that w = 2:i". lw, dx' is a l-formon R" and happens w
to equal dI = 2:f_ l DJ · dx'. We can clearly assume that

1(0) = o. As in Problem 2-35, we have
1 d
I(x) = J -I(tx) dt
dt
o
1 n
= Jl
O i- l
DJ(tx) · x' dt
1 n
J l w.(tx) . xi dt.
=
O .-1
This suggests that in order to find I, given w, we consider the
function Iw, defined by
l n
J l w;(tx) . x i dt.
Iw(x) =
O ;= 1
Note that the definition of Iw makes sense if w is defined only
on an open set A C Rn with the property that whenever
x E A , the line segment from O to x is contained in A ; such
an open set is called star-shaped with respect to O (Figure
4-3) . A somewhat involved calculation shows that (on a
star-shaped open set) we have w = d(lw) provided that w satis
fies the necessary condition dw = O. The calculation, as well
as the definition of Iw, may be generalized considerably:
FI G UR E 4-3
4-11 Theorem (Poincaré Lemma). If A C R n is an open

set star-shaped with respect to O, then every closed form on A
is exact.
Proof. We will define a function I from l-forms to (l - 1 )

forms (for each l), such that 1(0) = O and w = I (dw) + d(lw)
for any form w. It follows that w = d(lw) if dw = O. Let
I Wi" . , . ,il dxi, /\ . . . /\ dXil.
it< . . .
w =
<il
Since A is star-shaped we can define

l 1
I I ( _ 1 ) "'-1 (I tl-1wi" . . . ,ù(tx)dt ) Xi",
iI< ' " <il a = 1
Iw(x)
o
/\ . . . /\ /\ . . . /\
/.......
dx i, dxi", dxil .
(The symbol --... over dxia indicates that it is omitted.) The

proof that w = / (dw) + d(lw) is an elaborate computation:

We have, using Problem 3-32,
l
d ( lw) = l· l ( J tl-1Wit . . . . , il (tx)dt )
i l< . . . <il o
1\ 1\ . . . 1\ 1\ . . . 1\
/-.....
dxi dxi t dxi" dX il .
(Explain why we have the factor tI, instead of t I- l . ) We also

have
1\ 1\ . . . 1\
n
dw = ll Di (Wit, . . . . iJ . dxj dxit dxil .

ù< " ' <iz j = l
Applying / to the (l + l)-form dw, we obtain
n l
! (dw) L l ( J tlDi(Wit , . . . ,il) (tx)dt) xi
ù< " ' <ù j = l O
1\ 1\ . . . 1\ 1\ . . . 1\
/-.....
dxj dXi\ dxi" dXil .
Adding, the triple sums cancel, and we obtain
l
de/w) + ! (dw ) l l· ( J tl-lWit , . . . , il (lx)dt )
it< · · · <iz o
n l
+ l. l ( J tlxiDj (Wit , . . . ,iJ (tx) dt)
it< . · <ù j= l O
dxi t 1\ ' 1\ dxil

L Wit, • . • ,il dxit 1\ . . . 1\ dXil
il < ' " <il
= w. I
96 CalCUlU8 on Manilold8
Problema. 4-13. (a) Ii I: Rn � R'" and g: R'" � RP, show that
(g 0 /) • .. g. 0 1. and (g o f) · - r o g •.
(b) Ii l,g : Rn � R, show that d(/ ' g) = I · dg + g ' dI.
4-14. Let c be a differentiable curve in Rn, that is, a differentiable func
tion c: [D,l] � Rn. Define the tangent vector v of c at t as
c. «el)t) «c1)'(t) , . . . , (cn)'(t» C (t). If I: Rn � R"', show that
=
the tangent vector to l o c at t is I. (v).

4-15. Let I: R � R and define c: R � R2 by c(t) = (t,/(t» . Show
that the end point of the tangent vector of c at t lies on the
tangent line to the graph of I at (t,/(t» .
4-16. Let c: [D, l] � Rn be a curve such that Ic(t ) l .. l for all t. Show that
c(t) c(t) and the tangent vector tI) c at t are perpendicular.
4-17. Ii I: Rn � Rn, define a vector field f by f(p) I(p)p E Rnp. =
(a) Show that every vector field F on Rn is of the form f for

some I.
(b) Show that div f .. trace l'.
4-18. If I: Rn � R, define a vector field grad I by
(grad /) (p) ... D d(p ) . (e l )p + . . . + Dn/(p) . (en )p.

For obvious reasons we alBO write grad I '" VI. If Vf(p) .. wp,
prove that D./(p) .. (v,w) and conclude that v/(p) is the direction
in which I is changing fastest at p.
4-19. Ii F is a vector field on Ra, define the forms
"'� F1 dx + F2 dy + Fa dz,
F1 dy /I. dz + F2 dz /I. dx + F3 dx /I.
=
"'� = dy.
(a) Prove that
dI =
"'!rad/l
d(",�) =
"';url p,
d (",�) = (div F) dx /I. dy /I. dz.
(b) Use (a) to prove that
curI grad I = O,
div curI F = O.
(c) Ii F is a vector field on a star-shaped open set A and

curI F =O, show that F grad I for some function I: A � R.
=
Similarly, if div F O, show that F

= curI G for some vect;=
field G on A .
4-20. Let I: U � Rn b e a differentiable function with a differentiable
inverse r l : I(U) � Rn• If every closed form on U is exact, show
that the same is true for I(U). Bint: If d", ° and r", d'I, = =
consider (r 1) .".
Integration on ehains 97
4-21. · Prove that on the set where O is defined we have

-v
2 dx + �
+ 2
x
+
dO = dV·
x2
---
11 x V
G E O M E TR I e P R ELIMIN A R IES
A singular n-cube in A C Rn is a continuous function c :

(O,l]n A (here [O,l]n denotes the n-fold product (0,1] X . . .
�
X (0,1]). We let RO and (0,1]° both denote {O} . A singular

O-cube in A is then a functionf: {O} A or, what amounts to
�
the same thing, a point in A . A singular l-cube is often

called a curve. A particularly simple, but particularly
important example of a singular n-cube in Rn is the standard
n-cube r : [O,l]n �Rn defined by r(x) x for x E (O,l]n. =
We shall need to consider formaI sums of singular n-cubes in

A multiplied by integers, that is, expressions like
where C l , C2, Ca are singular n-cubes in A . Such a finite sum

of singular n-cubes with integer coefficients is called an
n-chain in A . In particular a singular n-cube C is also con
sidered as an n-chain 1 . c. It is clear how n-chains can be
added, and multiplied by integers. For example
(A rigorous exposition of this formalism is presented in Prob

lem 4-22.)
For each singular n-chain C in A we shall define an (n - 1)
chain in A called the boundary of C and denoted ac. The
boundary of 1 2, for example, might be defined as the sum of
four singular 1-cubes arranged counterclockwise around the
boundary of [0,1]2, as indicated in Figure 4-4(a). It is
actually much more convenient to define a1 2 as the sum, with
the indicated coefficients, of the four singular 1-cubes shown
in Figure 4-4(b). The precise definition of ar requires some
preliminary notions. For each i with 1 � i � n we define
two singular (n - l)-cubes l'(i, O) and 1'(;, I ) as follows. If
98 CalculW! on Manifolds
-1
-1 +1
+1
(a) (h)
FI G UR E 4-4
x E [0,1]"-\ then
I('i, O) (x) = . . i,Xi-l ,O,Xi, . . . ,X"- l )
[" (xl , .
(X l , '
,X -\O,Xi, . . ,X"- l), . • .
I('i, l (x) [" (x l, . . ,Xi- l ,l ,Xi, ,X"-l )

= (x l , .
= . • • •
l
,Xi- l , l ,x i, . . . ,X"-l ).
. .
We calI [('i, O) the (i,O)-face of [" and [('i, 1) the (i,l)-face

(Figure 4-5). We then define
"
!lv [" = �
� �
� ( - l ) i + a[ "(i,a) '
' � 1 a =O,1
For a generaI singuIar n-cube c: [0,1]" ---+ A we first define the

(i,a)-face,
and then define
n
ac = L L ( - l ) i+aC(i, a) '
. = 1 a =O, l
FinalIy we define the boundary of an n-chain �aiCi by
Although these few definitions suffice for alI applications in

this book, we include here the one standard property of a .
Il•. 1 l
lo
11,.0) n.ll
11..0) n.ll Il•.0)
(a) (b)
FI G UR E 4-,1
4-12 Theorem. If c is an n-chain tn A, then a(a c) = O,

Briefiy, a 2 = O,
Proof. Let i :::; j and consider ( l'(i,a» ) U ,fJ) ' If X E [O,1 I n-2 ,
then, remembering the definition of the (j,{J)-face of a singular
n-cube, we have
Similarly
( l'(i+ l. fJ» ) (i,a) = l'(i+I ,fJ) ( l(i-:a\ (X))
= I n( j+I ,fJ) (X l , . , " Xi- I ,a,X i" " ,Xn - 2 )
= I n (x l , . , . , xi-l,a,xi, ' , , ,X.i -\{3,Xj , ' , , ,Xn -2),
l'hus ( l'(i,a» )U ,fJ) ( l'(i+ l.fJ» ) (i,a) for i :::; j. (It may help to
=
verify this in Figure 4-5,) It follows easily for any singular

n-cube c that (C( i,a» ) U, /l) = (CU+l ,fJ»)( i,a) when i :::; j, Now
n
a (ac) = a (L L ( - l) i+ac(i,a» )
i = a=O,
n 1 n -1 l
= L L L L ( - l ) i+a+i+fJ(C(i,a» ) U,fJ) '
i = 1 a =O, 1 j= 1 fJ = O, 1
100 CalculU8 on M anifoldB
In this sum (C(i,a» U,�) and (cU+U» (i,a) occur with opposite
signs. Therefore aH terms cancel out in pairs and a (ac) = O.
Since the theorem is true for any singular n-cube, it is also
true for singular n-chains. I
It is natural to ask whether Theorem 4-12 has a converse: If
ac = O, is there a chain d in A such that c = ad? The answer
depends on A and is generally "no." For exampIe, define
c: [O, I J R2 - O by c(t) = (sin 21mt, cos 27rnt), where n lS
--+
a non-zero integer. Then c(l) = c(O) , so ac = O. But

(Problem 4-26) there is no 2-chain c' in R2 - O, with ac' = c.
4-22. Let S be the set of ali singular n-cubes, and Z the
integers. An n-chain is a function f: S --+ Z such that f(c)
ProblemlJ.
O =
for ali but finitely many c. Define f + g and nf by (f + g)( c) ..

n . f(c). Show that f + g and nf are
n-chains if f and g are. If c E S, let c also denote the function f
f(c) + g(c) and nf(è) =
such that f(c) = l and f(c') O for c' � c. Show that every
=
n-chain f can be written alCI + ...

+ akCk for some integers
a l,• • .
,ak and singular n-cubes C l, • .
,Ck. •
4-23. For R > O and n an integer, define the singular l-cube CR.n : [O, l) --+
R2 - O by CR,n (t) = (R cos 211'11t, R sin 211'11t) . Show that there
is a singular 2-cube c: [0,1)2 --+ R2 - O such that cR" n - CR,.n ilc. =
4-24. If c is a singular l-cube in R2 - O with c (O) = c(l), show that there

is an integer n such that c - CI.n il c2 for some 2-chain c2•
=
Hint: First partition [0, 1) so that each C([ti_l,t.)) is contained on

one side of some line through O.
THE F UN D A M E N T A L THEO R EM O F C A L C UL U S
The fact that d2 = O and a 2 = O, not to mention the typo

graphicai similarity of d and a , suggests some connection
between chains and forms. This connection is established by
integrating forms over chains. Henceforth only differentiable
singular n-cubes will be considered.
Ii w is a k-form on [O, I J\ then w = f dX l /\ . . . /\ dxk for
a unique function f. We define
J
[0,1)"
w = J f·
[0.1]"
Integra!ion on Chains 101
We could also write this as

f f dX l /\ . . . /\ dxk = f f(xl, . . . ,xk ) dx l . . . dx k ,
[O,l]k [0,1] '
one of the reasons for introducing the functions Xi.

If W is a k-form on A and c is a singular k-cube in A , we define
fw c
=
f
[O,l]k
c·w.
Note, in particular, that

/\ dXk = f ( lk) * (f dX l /\ . . . /\ dxk )
[O,l]k
f
[O,l]k
f(x \ . . . ,xk ) dx l . . . dXk .
A special definition must be made for k = O. A O-form w is

a function; if c : {O l A is a singular O-cube in A we define
�
f w = w(c(O» .
c
The integraI of w over a k-chain c = �aici is defined by
The integraI of a l-form over a l-chain is often called a Une

integraI. If P dx + Q dy is a l-form on R2 and c: [0, 1] R2 �
is a singular l-cube (a curve), then one can (but we will not)

prove that
n
J p dx + Q dy = lim L i
[Cl (ti) - C l (ti_ l )] . P (c(t »
c i-l
i
+ [C2(ti) - C2 (ti_l)] . Q (c(t »
where ,tn is a partition of [0,1] , the choice of ti in

to , . . .
[ti- l , li] is arbitrary, and the limit is taken over alI partitions
as the maximum of I ti - ti-I I goes to O. The right side is

often taken as a definition of f cP dx + Q dy. This is a natural
definition to make, since these sums are very much Iike the
sums appearing in the definition of ordinary integrals. How
ever such an expression is almost impossible to work with and
is quickIy equated with an integraI equivalent to f [ O , l ] C*(P dx
+ Q dy). AnaIogous definitions for surface integrals, that
is, integraIs of 2-forms over singular 2-cubes, are even more
complicated and difficult to use. This is one reason why we
have avoided such an approach. The other reaSOll is that the
definition given here is the one that makes sense in the more
generaI situations considered in Chapter 5 .
The relationship between forms, chains, d, and a is summed
up in the neatest possible way by Stokes' theorem, sometimes
called the fundamental theorem of calculus in higher dimen
sions (if k = 1 and c = I l , it really is the fundamental theorem
of calculus).
4-13 Theorem (Stokes' Theorem). I is a (le fw 1 )
form on an open Bet A C Rn and c is a k-chain in A, then
-
f dw = f w.
c éJc
Proof· Suppose first that c Ik and w is a (le - l)-form on

=
[O,l]k . 'l'hen w is the sum of (k l)-forms of the type

-
and it suffices to prove the theorem for each of these. This

simply invoIves a computatiOlI:
Note that
. . 1\
l.
[O, )0-1
I ri, a) * (f dx l 1\ 1\ d;i 1\ . dXk)
O if j � i,
l f f(x \ . . .
[0, 1 )0
, a, . . ,xk )dx 1
. • • •
dxk if j i.
=
lntegration on Chains 103
Therefore
k
= L L ( _ I)i+a J I�;,a) *(f dX l 1\
j - l a = O, l [O,lJ' -'
( - 1 )Hl f f(x , l
' , , ,1 , '
[O, lJ'
+ ( - I )i J f(x I ,
[O,lJ'
On the other hand,

J d (f dX l 1\ , 1\ dxi 1\
---
' , , 1\ dXk )
/'
J D d dxi 1\ dX l 1\
[O,lJ'
( _ I ) i- l f D d,
[O, W
By Fubini's theorem and the fundamental theorem of calculus

(in one dimension) we have
f d (f dX l 1\ , , l\ ;J;i 1\ , . 1\ dxk)
'
l'
1
( _ 1 ) i f
- l
, xk )dxi ) dX l
o
dxi . , . dXk
1 1
= ( _ I ) i-
l f f [f(x \ . . , , l , . . . ,x k )
o o
---
1
- f(x , ,0 , ,x k ) Jdx 1 . . . dX i . . . dXk
= ( _ 1 ) i- 1 f f(x I, . , 1, ,x k )dx 1 • • • dXk

[O,lJ'
+ ( - l)i f f(x \ . . . ,0, . . . ,xk )dx 1 . . . dxk.

[0,1)'
Thus
f dw
IO
=
a o
l w.
If c is an arbitrary singular k-cube, working through the

definitions will show that
J w = f c*w. ile ilI"
Therefore
J dw = J c* (dw) = f d(c*w) = f c*w = J w.
e l' l' ilI' ile
Finally, if C is a k-chain �aici, we have

J dw = Iai J dw = Iai J w = J w. I
C Ci aCi éJc
Stokes' theorem shares three important attributes with

many fully evolved major theorems:
1. It is trivia!'
2. It is trivial because the terms appearing in it have been
properly defined.
3. It has significant consequences.
Since this entire chapter was little more than a series of

definitions which made the statement and proof of Stokes'
theorem possible, the reader should be willing to grant the
first two of these attributes to Stokes' theorem. The rest of
the book is devoted to justifying the third.
Problems. 4-25. (lndependence 01 parameterization). Let c be a
singular k-cube and p : [O,l]k � [0, 1 ] 10 a 1 - 1 function such that
p ([O,l]k) [O,l]k and det p' (x) 2: ° for x E [O,lt
= If w is a
k-form, show that
J
c
w = J
cop
w.
feR , . dO = 211'n, and use Stokes' theorem to conclude

that CH, n � ÒC for any 2-chain C in R2
4-26. Show that
° (recall the definition of
-
CH,n in Problem 4-23).

4-27. Show that the integer n of Problem 4-24 is unique, This integer
is called the winding nu mber of C around O.
4-28. Recall that the set of complex numbers C is simply R2 with
a + bi. If al, . . . ,an E C let I: C � C be f(z) =
z n + alZ n-1 + + an ' Define the singular l-cube CB.!:
(a,b) =
. .
.
l o CR, l, and the singular 2-cube c by

t CR, n (S ) + (I - t)cR .! (S ) .
[O, IJ ---> C -O by CR.! =
c(s,t)= .
C - O if R is large enough.
(a) Show that a c CR.! - CR,n , and that c ([O,IJ X [0,1]) C
=
A lgebra: Every polynomial zn + alZn-l + . . . + an with ai E C

(b ) Using Problem 4-26, prove the Fundamental Theorem 01
has a root in C.
40-29. If w is a I-form 1 dx on [O, IJ with 1(0) = 1(1), show that there is
a unique number X sueh that w - X dx dg for some funetion g
=
with g (O) = g(I ) . Hint: Integrate w - X dx dg on [O, I J to

=
40-30. If w is a I-form on R 2 - O sueh that dw � O, prove that

find X.
w = X d9 + dg
for some X E R and g: R 2 - 0 ---> R. Hint: If
C R,l*( W) = XR dx + d( gR) ,
40-31 . If w ;.6 O, show that there is a ehain c sue h that J c w ;.6 O. Use this
show that ali numbers XR have the same value X .
faet, Stokes' theorem and a2 = O to prove d 2 O. =
4-32. ( a) Let Cl, C2 be singular I-eubes in R 2 with Cl( O)

c 2 ( 1 ) . S how that there is a singular 2-eube c sueh that ac ..
C 2( 0) and cl( l)
=
Ic,w if w is exaet.
Cl - C2 + C3 - C4, where C3 and C4 are degenerate, that is, c3( [O, I ])
and C4( [O, l]) are points. Conclude that Jc,w
Give a eounterexample on R 2 - O if w is merely closed.
=
(b) If w is a I-form on a subset of R 2 and Ic,w Ic,w for all Cl, =
C 2 with cr(O) C 2 (0) and cl(l)

= c 2 (1), show that w is exaet.
=
Hint: Consider Problems 2-21 and 3-34.

4-33. (A first course in complex variables.) If I: C ---> C, define 1 to be
differentiable at Zo E C if the limit
I(z) - I(zo)
l' (zo ) = l1' m
'--+" z - Zo
exists. (This quotient involves two complex numbers and this
definition is eompletely different from the one in Chapter 2.)
If 1 is differentiable at every point z in an open set A and f' is
ii is not (where
eontinuous on A , then 1 is ealled analytic on A .
(a) Show that I(z) z is analytie and I(z)
x + iy
= =
=x - iv). Show that the sum, product, and quotient
u + iv is analytic on A, show that u and v satisfy

of analytie funetions are analytie.
(b) If f =
the Cauchy-Riemann equations:
au av au - av
and
ax ay ay ax
106 Calculu8 on Manilold8
Hint: Use the fact that Hrn [/(z) - l(zo)J!(z - zo) must be the
Zo + (x + i . O) and z zo + (O + i · y) with
z�zo
asme for :. = =
x,y --+ O. (The converse is also true, if u and v are continuously

differentiable; this is more difficult to prove. )
(c) Let T: C --+ C b e a linear transformation (where C i s con
aidered as a vector space over R ) . If the matrix of T with respect
to the basia (l,i) is (::�) show that T is multiplication by a com

lex number if and only if a d and b - c. Part (b) shows that
= =
an analytic function I: C --+ C, considered aa a function I: R 2 --+

R2, has a derivative DI(zo) which is multiplication by a complex
number. What complex number is this?
(d) 'Define
d(w + i1/) = dw + i d1/,

Jw + i1/ = J w + i J 1/,
e e e
(w + i1/) /\ (O + iX) = w /\ O - 1/ /\ X + i(1/ /\ O + w /\ Xl,
and
di = dx + i dy.
Show that d(f · dz) O if and only if 1 satisfies the Cauchy
=
Riemann equations.
(e) Prove the Cauchy Integrai Theorem: If 1 is analytic on A ,
then J cl dz O for every closed curve c (singular l-cube with
=
c(O) = c(l» such that c ac' for some 2-chain c' in A .

=
l /z, then g . dz [or (l /z)dz in classical

notation] equals i dO + dh for some function h: C - 0 --+ R.
(f) Show that if g(z) =
Conclude that J CR (l /z)dz •• 21rin. =
(g) If 1 is analytic on { z : Izl < l } , use the fact that g(z)

I(z)/z is analytic in {z: O < Izl < l } to show that
J I�) dz = J I�) dz
CR I.n CR2.n
R-+O
if O < R l, R < 1 . Use (f) to evaluate lim J eR.J(Z) /z dz and
2
conclude :
Cauchy Integrai Formula: If 1 is analytic on {z: Izl < l } and
c is a closed curve in {z: O < Izl < l } with winding number n
around 0, thcn
n ' /(0) J I(z)

1
= - . - dz.
21rt Z
e
lntegration on Chains 1 07
Il
--------�--�- T
(a)
�.
--------+---�L---�--+_- I
!/
--------�----__�-4�_4----_+--�------ I
F I G U R E 4-6
108 Calculua on Manilolda
4-34. If F: [O, 1]2 ---> R3 and s E [0,1] define F.: [0,1] ---> R3 by F. (t) =
F (s,t) . If each F. is a closed curve, F is called a homo topy between
the closed curve Fo and the closed curve Fl. Suppose F and G are
homotopies of closed curves; if for each s the closed curves F. and
G. do not intersect, the pair (F,G) is called a homotopy between the
nonintersecting c10sed curves Fo, Go and Fl, Gl. It is intuitively
obvious that there is no such homotopy with F o, Go the pair of
curves shown in Figure 4-6 (a), and Fl, Gl the pair of (b) or (c) .
The present problem, and Problem 5-33 prove this for (h) but the
(a) If I, g: [0,1] ---> R3 are nonintersecting closed curves define

proof for (c) requires different techniques.
cJ,g: [0,1]2 ...... R3 - ° by

cJ,g (u,v) ... I(u) - g(v).
CF , G : [0,1]3 -> R3 - ° by
If (F,G) is a homotopy of nonintersecting closed curves define
CF,G (8,U,V) ... CF" G, (U,v) = F(8,U) - G(s,v).

Show that
(b) If '" is a closed 2-form on R3 - ° show that

5
Inlegralion on Manifolds
M A NIFO L D S
If U and V are open sets i n Rn, a differentiable function

h: U l
-+ V with a differentiable inverse h- : V -+ U will be
called a dift'eomorphism. ("Differentiable" henceforth

means "e""" . )
A subset M of Rn is called a k-dimensional manifold (in
Rn) if for every point x E M the following condition is
satisfied :
(M) There is an open set U containing x, an open set V C Rn,

and a diffeomorphism h : U -+ V such that
h(U lì M) = V lì (Rk X I O } )
= ty E V: yk+ 1 = . . . = yn = O}.
In other words, U lì M is, "up to diffeomorphism," simply

Rk X I O } (see Figure 5-1 ) . The two extreme cases of our
definition shouId be noted : a point in Rn is a O-dimensionaI
manifoId, and an open subset of Rn is an n-dimensionaI
manifoId.
One common example of an n-dimensionaI manifold is the
109
u
F I G U RE 5 - 1 . A one-dimensionaZ manifold in R2 and a two-dimen

sionaZ manifoZd in R3,
Integration on 1Ifanifolds 111
n-sphere sn, defined as I x E Rn+l: I x l =I l . We Ieave i t

as a n exercise for the reader t o prove t hat conditioll (111) is
satisfied. If you are unwilling to trouble yourself with the
details, you may instead use the follo,ying theorem, which
provides many exampIes of manifolds (note that sn = g - l (O) ,
where g : Rn+ l -> R i s defined by g(x) = I x l 2 - l).
5-1 Theore m. Let A C Rn be open and let g : A -> RP

be a diffe1'entiable funclion sllch that g' (x) has mnk p whenever
g(x) = O. Then g-l (O) is an (n - p)-dimensional manifold in
Rn.
Proof. This follows immediately from Theorem 2-1 3 . I
There is an alternative characterization of manifolds which

is very importan t .
5-2 Theore m. A sllbset 111 of R n i s a k-dimensional mani

fold if and only iffo1' each point x E 111 the following "coordinate
condition" is satisfied:
(C) There is an open set U containing x, an open set W C Rk,

and a l- l differentiable function f: W -> Rn such that
(1) f(W) M I\ U,
(2) f'(y) has 1'ank k fOi" each y E W,
=
( 3) r l : f(W) -> W is continuous.

[Such a function f is called a coordinate systCll1 around x
(see Figure 5-2).]
Proof. If M is a k-dimensional manifold in Rn, cl100se

h : U -> V satisfying (M) . Let W = { a E Rk : (a,O) E h (J1J) l
and define f: W -.Rn by f(a) h-1(a,O). Clearly f( TV)
= =
M (ì U and r 1 is continuous. Ii H : U -> R k is H (z) =

(h 1 (z), . . . , hk (z) , then H(f(y» = y for all y E W ; there
fore H' (f(y» . f'(y) = I and f'(y) must have rank k.
Suppose, conversely, that f: W -> R n satisfies condition (C).
Let x = f(y) . Assume that the matrix (Ddi(y» , 1 :::; i, j :::; k
has a non-zero determinant. Define g : W X R n -k -> R n by
112 Calculus o n Manifolds
FI G UR E 5-2
g(a,b) = f(a) + (O,b). Then det g'(a,b) = det (Dip(a)), so

det g'(y,O) � O. By Theorem 2-11 there is an open set V l'
containing (y,O) and an open set V 2' containing g(y,O) = X such
that g : V l ' V 2' has a differentiable inverse h : V 2 ' V l'.
---+ ---+
Since r l is continuous, { f(a) : (a,O) E V l' } U (ì f(W) for

=
some open set U. Let V 2 V 2' (ì U and V l g- l (V 2) .

=
Then V 2 (ì M is exactly I f(a) : (a,O) E V t i = { g(a,O) : (a,O)

=
E VI I , so
h(V 2 (ì M) g- 1 ( V2 (ì M) g- l ( {g(a,O) : (a,O) E V t i )

=
= =
VI (ì (R k X { O } ) . I
One consequence of the proof of Theorem 5-2 should be

noted. If fl : W l Rn and f2 : W 2 Rn are two coordinate
---+ ---+
Integration on Manifolds 113
systems, then
IS differentiable with non-singular Jacobian. If fact, J21 (y)

consists of tbe first k components of h(y).
Tbe half-space Hk C Rk is defined as I x E R k : xk � 0 1 .
A subset M of Rn is a k-dirnensional rnanifold-with
houndary (Figure 5-3) if for every point x E M eitber condi
tion (M) or the following condition is satisfied:
(M') There is an open set U containing x, an open set
V C Rn , and a diffeomorphism h : U V such that ->
h(U n M) V n (Hk x I O } )
=
= Iy E V : yk � O and yk + l .. =yn O j . = =
and h(x) has kth component = o.

It is important to note that conditions ( M) and (M')
cannot both hold for the same x. In fact, if h l : U l V l and ->
h2: U2 V2 satisfied (M) and (M') , respectively, then

->
h2 h l -l would be a differentiable map that takes an open set

o
in Rk, containing h(x), into a subset of Hk which is not open in

R k . Since det (h2 h l - l )' � O, this contradicts Problem
o
2-36. The set of alI points x E M for which condition M' is

satisfied is called the houndary of M and denoted aMo This
(a) (b)
FIG UR E 5-3. A one-dimensional and a two-dimensional manifold
with-boundary in R 3.
114 Calculus o n Manifolds
must not be confused with the boundary of a set, as defined in

Chapter 1 (see Problems 5-3 and .1-8).
Problema. 5-1 . If M is a k-dimensional manifold-with-boundary,
prove that aM is a (k - I )-dimensionai manifold and M - aM is
a k-dimensional manifold.
5-2. Find a counterexample to Theorem 5-2 if condition (3) is omitted.
Bint: Wrap an open interval into a figure six.
5-3. (a) Let A C R" be an open set such that boundary A is an (n - 1 )
dimensionai manifold. Show that N = A V boundary A is an
n-dimensionai manifold-with-boundary. (It is well to bear in mind
the following example : if A = Ix E R": JxJ < l or l < JxJ < 2 1
aN � boundary A . )
then N = A V boundary A i s a manifold-with-boundary, but
(b) Prove a similar assertion for a n open subset o f a n n-dimen

sionai manifold.
5-4. Prove a partial converse of Theorem 5-1 : If M C Rn is a k-dimen
sional manifold and x E M, then there is an open set A C R" con
taining x and a differentiable function g: A -+ Rn-k such that A (ì M
= g- l (O) and g'(y) has rank n - k when g(y)= O.
5-5. Prove that a k-dimensional (vector) subspace of R" is a k-dimen
sional manifold.
FIG U R E 5-4
5-6. I if: R n -+ Rm, the graph o f f i s I (x,y) : y f(x) I . Show that

=
the graph of f is an n-dimensionai manifold if and only if f is
Ix E R n ; xl ,xn- l > 0 1 . Ii M C Kn
differentiable.
5-7. Let Kn = O and x2,
= • • •
around the axis Xl - x..-l - O, ahow that N is a (k + 1)

is a k-dimensional manifold and N is obtained by revolving M
•
• • ==
dimensionaI manifold. Example: the torua (Figure 5-4).

5-8. (a) Ii M is a k-dimensional manifold in Rn and k < n, show that
M has measure O .
(b) Ii M is a closed n-dimensionai manifold-with-boundary in
Rn, show that the boundary of M is aMo Give a counterp.xample if
M is not closed.
(c) Ii M is a compact n-dimensional manifold-with-boundary
in Rn, show that M is Jordan-measurable.
FIEL D S A ND FO R MS ON M A NIFO L D S
Let M be a le-dimensionaI manifold in Rn and let f: W Rn �
be a coordinate system around x = f(a). Since l'(a) has rank

le, the Iinear transformation f* : Rka Rn", is l -l , andf* (Rka)
�
is a le-dimensionaI subspace of Rn",. If g: V Rn is another �
coordinate system, with x = g(b), then

g. (Rkb) f*(r1 g) . (Rkb) = f. (Rka).
= o
Thus the le-dimensionaI subspace f. (Rk ) does not depend on

a
the coordinate system f. This subspace is denoted M"" and

is called the tangent space of M at x (see Figure 5-5). In
Iater sections we will use the fact that there is a naturai inner
product T", on M"" induced by that on Rn",: if v,w E M", define
T",(v,w) = (v,w)", .
Suppose that A is an open set containing M, andF is a differ
entiable vector field on A such that F(x) E M for each x
x E M. If f:- W R n is a coordinate system, there is a

�
unique (differentiable) vector fieldG on Wsuch thatf. (G(a» =

F(f(a» for each a E W. We can also consider a function F
which merely assigns a vector F(x) E M", for each x E M i
such a function is called a: vector field on M. There is still
a unique vector field G on W such that f*(G(a» F(f(a» for =
a E W i we define F to be differentiable if G is differentiable.

Note that our definition does not depend on the coordinate
FIG UR E 5-5
system chosen : if g : V R" and g* (H(b» = F(g(b» for alI

-+
b E V, then the component functions of H (b) must equaI the

component functions of G(r1(g(b))), so H is differentiabIe
if G is.
Precisely the same considerations hoId for forms. A func
tion w which assigns w(x) E AP(Mx) for each x E M is called
a p-for.m M. If f: W R" is a coordinate system, then
on �
f*w is a p-form on W; we define w to be differentiabIe if f*w is.

A p-form on M can be written as
w
w = I W i" . . . ,ip dxi, 1\

il < . . . < ip
Rere the functions Wi" . . . , ip are defined only on M. The

definition of dw given previousIy would make no sense here,
since Dj(Wi" ) has no meaning. NevertheIess, there is a
.
. .
, ip
reasonabIe way of defining dw.

lniegratwn on Mani/olds 117
5-3 Theorem. There is a unique (p + l)-lorm dw on M

such that lor every coordinate system I: W Rn we have
�
f*(dw) = d(f*w).
Proof. If f: W Rn is
� a coordinate system with x I(a) =
and V . . . ,Vp+I E M""

I, there are unique Wl, ,Wp + l in
Rka such that j. (Wi) Vi. Define dW(X) (V I , . . . ,Vp+ l ) =
• • •
d(f*w) (a) (wI, . . . ,Wp+ I) . One can check that this definition
of dw(x) does not depend on the coordinate system I, so that
dw is well-defined. Moreover, it is clear that dw has to be
defined this way, so dw is unique. I
It is often necessary to choose an orientation Il,,, for each
tangent space M", of a manifold M. Such choices are called
consistent (Figure 5-6) provided that for every coordinate
(a)
(b)
FI G U R E 5-6. (a) Consistent and (b) inconsisient choices oJ orien
tations.
system f: W � Rn and a,b E W the relation

[f* ((el)a) , . . . ,!* ((e k)a)] = JJ. f(a)
holds if and only if
Suppose orientations JJ.x have been cho_sen consistently. Ii

f: W Rn is a coordinate system such that
�
for one, and hence for every a E W, then f is called orien

tation-preserving. Ii f is not orientation-preserving and
T : Rk � Rk is a linear transformation with det T = - 1 , then
f T is orientation-preserving. Therefore there is an orienta
o
tion-preserving coordinate system around each point. Ii f and

g are orientation-preserving and x f(a) g(b), then the
= =
relation
FIG UR E 5-7. The Mobius strip, a non-orientable manifold. A

basis begins at P, moves to the right and around, and comes back to P with
the wrong orientation.
implies that
l
so that det (g- o f) ' >
O, an important fact to remember.
A manifoid for which orientations J.Lx can be chosen con
sistentIy is called orientable, and a particuiar choice of the
J.Lx is called an orientation J.L of M. A manifold together with
an orientation Il is called an oriented manifold. The classical
example of a non-orientabie manifoid is the M6bius strip.
A model can be made by gluing together the ends of a strip of
paper which has been given a half twist (Figure 5-7).
Our definitions o f vector fields, forms, and orientations can
be made for manifolds-with-boundary also. If M is a k-dimen
sionai manifold-with-boundary and x E aM, then (aM)x is
a (k - I )-dimensionaI subspace of the k-dimensional vector
space Mx. Thus there are exactly two unit vectors in Mx
which are perpendicuiar to (aM) x ; they can be distinguished
as follows (Figure 5-8). If f: W --; Rn is a coordinate system
with WC Hk and f(O) x, then onIy one of these unit vectors
=
is f* (vo) for some Vo with vk < O. This uni t vector is called the
outward unit nonnal n (x) ; it is not hard to check that this
definition does not depend on the coordinate system f.
Suppose that J.L is an orientation of a k-dimensional manifoId
with-boundary M. If x E aM, choose VI, ,V k- l E (aM)x
,v -ll = J.Lx. If it is also true that
• . .
so that [n (x) , VI,

[n(x) , Wl, . . . ,wk- ll = J.Lx , then both [VI,
. • •
k
• •,V -l) and
•
k
[Wl, . . . , W - l) are the same orientation for (aM) x' This
k
orientation is denoted (aJ.L)x, It is easy to see that the orienta
tions (aJ.L)x, for x E aM, are consistent on aMo Thus if M is
orientabIe, aM is also orientabIe, and an orientation J.L for M
determines an orientation aJ.L for aM, called the induced
orientation. If we apply these definitions to Hk with the
usual orientation, we find that the induced orientation on
Rk-l = I x E Hk : Xk O } is ( _ l )k times the usuai orienta
=
tion. The reason for such a choice wiII become clear in the
next section.
If M is an Q1'iented (n - I )-dimensionaI manifold in Rn, a
substitute for otltward unit normai vectors can be defined,
(b)
(c)
F I G U R E 5-8. Some outward unit normal vectors of manifolds-with

boundary in R 3.
even though M is not necessariIy the boundary of an n-dimen

sionaI manifoId. If [VI, . . . , Vn- I ] fJ.x , we choose n(x) in
=
Rnx so that n(x) is a unit vector perpendicular to Mx and

[n(x), VI, . . . ,Vn-I] is the usuaI orientation of Rnx. We stili
calI n(x) the outward unit normaI to M (determined by fJ.).
The vectors n(x) vary continuousIy on M, in an obvious sense.
Conversely, if a continuous family of unit normal vectors n(x)
is defined on alI of M, then we can determine an orientation of
M. This shows that such a continuous choice of normaI
vectors is impossibIe on the l\16bius strip. In the paper modeI
of the M6bius strip the two sides of the paper (which has
thickness) may be thought of as the end points of the unit
Integration on Manifold8 121
normal vectors in both directions. The impossibility of

choosing normal vectors continuously is reflected by the
famous property of the paper model. The paper model is
one-sided (if you start to paint it on one side you end up
painting it all over) ; in other words, choosing n(x) arbitrarily
at one point, and then by the continuity requirement at other
points, eventually forces the opposite choice for n(x) at the
initial point.
of curves c in M with c (t) = x.

Problems. 5-9. Show that M" consists of the tangent vectors at t
5-10. Suppose e is a collection of coordinate systems for M such that

(l) For each x E M there is f E e which is a coordinate system
around Z; (2) if f,g E ('l, then det (1-1 0 g)' > O. Show that there
\
is a unique orientation of M such that f is orientation-preserving
5-1 1 . If M is an n-dimensionaI manifold-with-boundary in Rn, define

if f E e.
/J" as the uBual orientation of M" = Rn" (the orientation /J so
defined is the usual orientation of M). If x E aM, show that
5-12. (a) If F is a differentiable vector field on M C Rn, show that

the two definitions of n(x) given above agree.
there is an open set A :) M and a differentiable vector field F

on A with F(x) = F (x) for x E M. Hint: ])o this loeally and
(b) If M is closed, show that we ean choose A = Rn.

use partitions of unity.
5-13. Let g: A -+ RP be as in Theorem 5-1.

(a) If x E M = g- l (O), let h: U -+ Rn be the essentially unique
diffeomorphism such that g o h(y) _ (y"-P+1, . . . ,y") and
is l-l so that the n - p vectors f. « e-i)o), . . . , [* « en -p ) o) are

h (O)
= x. Define f: Rn- p -+ R n by f(a) = h(O,a). Show that f.
Iinearly independent.
(b) Show that orientations /J" can be defined consistently, so
(c) If p = l , show that the components of the outward normal

that M is orientable.
at x are some multiple of Dlg(X), . . . , Dn(J(x).

5-14. If M C R" is an orientable ( n - I)-dimensional manifold, show
that there is an open set A C R" and a differentiable g : A -+ R i SO
that M - g- l(O) and g'(z) has rank l for z E M. Hint: Prob
lem 5-4 does this loeal1y. Use the orientation to ehoose eonsistent
loeal solutions and use partitions of unity.
5-15. Let M be an (n - I )-dimensionaI manifold in R n. Let M (e) be
the set of end points of normal vectors (in both directions) of
length e and suppose e is small enough so that M (e) is also an
122 CIllculu8 on Manifolds
(n - I)-dimensionai manifold. Show that M(€) is orientable
(even if M is not) . What is M(e) if M is the Mobius strip ?
5-16. Let g: A ..... RP be as in Theorem 5-1. If f: R" ..... R is differentiable
and the maximum (or minimum) of f on g-I(O) occurs at a, show
that there are XI, . . . ,Xp E R, such that
(1) DJj(a) = L XiDjgi (a) j = 1, . . . ,n.

i-l
Hint: This equation can be written df(a) �r=IXidgi (a) and is=
obvious if g(x) = (x,.-P+t, . . . ,x").

The maximum of f on g- I (O) is sometimes calied the maximum
of f subject to the constraints g i O. One can attempt to
=
find a by solving the system of equations (1 ) . In particular, if

g: A ..... R, we must solve n + 1 equations
Djf(a) = XDjg(a),
g(a) = O,
in n + 1 unknowns al, . . . ,a", X, which is often very simple
if we leave the equation g(a) O for last. This is Lagrange's
=
method, and the useful but irrelevant X is called a Lagrangian

multiplier. The following problem gives a nice theoretical use
for Lagrangian multipliers.
5-17. (a) Let T: Rn ..... Rn be self-adjoint with matrix A = (a;j) , so
that aij = ai;' If f(x) (Tx,x) �aijxixj, show that Dkf(x)
= =
=
j
2�'j_I akjx . By considering the maximum of (Tx,x) on S,,-I
show that there is x E sn-I and X E R with Tx Xx. =
(b) If V = {y E R n : (x,y) O } , show that T(V) C V and

=
T: V ..... V is self-adjoint.
(c) Show that T has a basis of eigenvectors.
S TO KES' THE O R E M O N MA NIFO L D S
If w is a p-form on a k-dimensional manifold-with-boundary

M and c is a singular p-cube in M, we define
f w f c*w
[0.1)"
=
c
precisely as before; integrals over p-chains are also defined as

before. In the case p k it may happen that there is an
=
open set W :> [O,l]k and a coordinate system f: W Rn such -->
that c (x) = f(x) for x E [O,l)k; a k-cube in M will always be

lntegration on Manifolds 1 23
understood to be of this type. If M is oriented, the singular

k-cube c is called orientation-preserving if f is.
5-4 Theorem. If Cl,C 2 : [O,l] k --) M are two orientation
preserving singular k-cubes in the oriented k-dimensional mani
fold M and w is a k-form on M such that w = ° outside of
C l ([O, l ] k ) (ì C 2 ([O, l] k), then
f w = f W.
Cl es
Proof. We have
fw= f C l *(W) = f (C2"l o C l ) *C2*(W) .
CI [0, 1 ]' [0,1]'
( Here c2"l o C l is defined only on a subset of [O,l] k and the

second equality depends on the fact that w = ° outside of
c l ([O,l ] k) (ì c 2 ([O,1 ] k) . ) It therefore suffices to show that
f (C2"1 o C l) *C2 *(W) = f C 2 *(W) = f W.
[0,1]' [0,1]' c,
If C2 * (W) = f dX l 1\ . . . 1\ dXk and C2- 1 o Cl is denoted by g,

then by Theorem 4-9 we have
(C2 - l o Cl)*C2 *(W) = g*(f dX l 1\ . . . 1\ dxk )
= (f o g) . det g' . dXI 1\ ' 1\ dXk
= (f o g) . I det g' i . dx l 1\ ' . 1\ dxk ,
since det g' = det (c2-l o Cl)' > O . The result now follows
from Theorem 3-1 3. I
The last equation in this proof should help explain why we
have had to be so careful about orientations.
Let w be a k-form on an oriented k-dimensional manifold M.
If there is an orientation-preserving singular k-cube c in M such
that w = ° outside of c([O,l] k), we define
f w = f W.
M C
Theorem 5-4 shows fM w does not depend on the choice of c.

124 Calculu8 on Mani/olda
Suppose now that w is an arbitrary k-form on M. There is an
open cover e of M such that for each U E e there is an orienta
tion-preserving singular k-cube c with U C c([O,I]k). Let � be
a partition of unity for M subordinate to this cover. We
define
provided the sum converges as described in the discuBBion pre

ceding Theorem 3-12 thi ( s is certainly true if M is compact).
An argument similar to that in Theorem 3-12 shows that IM w
does not depend on the cover e or on �.

AH our definitions could have been given for a k-dimensional
manifold-with-boundary M with orientation IJ.. Let aM have
the induced orientation ajJ.. Let c be an orientation-preserv
ing k-cube in M such that C(k .O) lies in aM and is the only face
which has any interior points in aMo As the remarks after
the definition of alJ. show, C(k. O) is orientation-preserving if k is
even, but not if k is odd. Thus, if is a (k - l)-form on M
w
which is O outside of c([O,lJk), we have

fw=
C(k,O)
( - I)k ..[
v
W.
On the other hand, C(k. O) appears with coefficient (_I)k in ac.

Therefore
f w = f w = (-Il f w = [ w.
ac ( - l) kC(k ••) C(k ••) a
Our choice of alJ. was made to eliminate any minus signs in this
equation, and in the following theorem.
If M is a compact
w
5-5 Theorem (Stokes' Theorem).
oriented k-dimensional manifold-with-boundary and is a
(k - I )-form on M, then
f dw = f M
W.
aM
(Rere aM is given the induced orientation.)

Proof. Suppose first that there is an orientation-preserving
singular k-cube in M - aM such that w O outside of =
I ntegration on Manifolds 1 25
By Theorem 4-13 and the definition of dw we have

C( [O, 1]k) .
f dw f c* (dw) f d(c*w) f c*w = f W.
l [O, J' .[O. lJ' iJf iJc
= =
Then
M
f dw = f dw = f w = O,
c iJc
since W O on ac. On the other hand, fiJM W O since W O

= = =
on aM.
Suppose next that there is an orientation-preserving singular
k-cube in M such that C (k , O) is the onIy face in aM, and W = O
outside of c([0,1 j) k. Then
f dw = f dw = f W = f W.
M c iJc iJ M
Now consider the generaI case. There is an open cover e

of M alld a partition of unity .p for M subordinate to e such
that for each cp E .p the form cp . w is of one of the two sorts
already considered. We have
O = del) = d ( l cp) = l dcp,
.. E � .. E �
so that
Since M is compact, this is a finite sum and we have
Therefore
Problems. 5-18. If M is an n-dimensionaI manifold (or manifold

with-boundary) in Rn, with the usual orientation, show that
126 Calculus on Manilolds
LuI dx' /\ . . . /\ dx"', as defined in this section, is the same a8

J MI, as defined in Chapter 3.
5-19. (a) Show that Theorem 5-5 is false if M is not compact. Bint: If
M is a manifold-with-boundary for which 5-5 holds, then M - éJM
is also a manifold-with-houndary (with empty boundary).
(b) Show that Theorem 5-5 holds for noncompact M provided
that w vanishes outside of a compact subset of M.
5-20. If w is a (k - i ) -form on a compact k-dimensional manifold M,
prove that M dw f O. Give a counterexample if M is not
=
compact.
5-2 1 . An absolute k-tensor on V is a function 1/ : Vk -+ R of the form
Iwl for w E Ak (V) . An absolute k-form on M is a function 1/
such that 1/ (x) is an absolute k-tensor on M". Show that M 1/ f
can be defined, even if M is not orientable.
5-22. If MI C R'" is an n-dimensionai manifold-with-boundary and
M2 C Ml - éJM l is an n-dimensionai manifold-with-boundary,
and M1,M2 are compact, prove that
w here w is an (n - l )-form on M l, and éJMl and éJM2 have the ori
Find a manifold-with-boundary M such that éJM éJM l V éJM2 and

entations induced by the usual orientations of M l and M 2. Bin t :
=
such that the induced orieritation on éJM agrees with that for
éJM l on éJM l and is the negative of that for éJM 2 on éJM 2.
TRE V O L UME ELEMEN T
Let M be a k-dimensional manifold (or manifold-with-bound

ary) in Rn , with an orientation J.I.. If x E M, then J.l.x and the
inner product Tx we defÌned previously determine a volume
element w(x) E Ak( Mx) . We therefore obtain a nowhere-zero
k-form w on M, which is called the volume element on M
(determined by J.I.) and denoted dV, even though it is not gen
erally the differential of a (k - I)-form. The volume of M
is defined as f M dV, provided this integraI exists, which is
certainIy the case if M is compact. "Volume" is usually
called length or surface area for one- and two-dimensional
manifolds, and dV is denoted ds (the "element of length") or
dA [or dSJ (the "element of [surfaceJ area").
A concrete case of interest to us is the volume element of an
oriented surface (two-dimensional manifold) M in R 3• Let

n (x) be the unit outward normal at x E M. If w E A 2 (M.,)
is defined by
w(v,w) = det (: ) ,
n (x)
then w(v,w) = 1 if v and w are an orthonormal basis of M., with

[v,w] = p..,. Thus dA w . On the other hand, w(v,w) =
=
(v X w, n (x» by definition of v X w. Thus we have

dA (v,w) = (v X w, n(x» .
Since v X w is a multiple of n (x) for v,w E Mx, we conclude
that
dA(v,w) = Iv X wl
if [v,w] p..,. If we wish to compute the area of M, we must
evaluate J [0. l J 2 c* (dA) for orientation-preserving singular
=
2-cubes c. Define
1? (a) = (1)lc l (a) J 2 1- (1) lc 2 (a)] 2 1- (1)lc3(a) J 2,
F (a) = 1) lc l (a) . 1) 2c l (a)
1- 1)lc2 (a) . 1) 2c 2 (a)
1- 1) l c3(a) . 1) c3(a) ,
2
G(a) l
= (1) 2c (a)J 2 1- (1) 2c 2 (a)] 2 1- (1) 2c3(a)1 2 .
Then
c* (dA ) «e l )a,(e 2)a)
l
= dA (c* «el)a) ,C* « e 2) a»
1 (1) lC (a) ,1) lC 2 (a) , 1)lC3 (a» X (1) 2cl(a) , 1) 2C 2 (a) , 1) 2c3(a» I
= V1? (a) G (a) - F (a) 2
by Problem 4-9. Thus

f c * (dA) = f V1?G - F2•
[O,lJ' [O, l J'
Calculating surface area is clearly a foolhardy enterprise;

fortunately one seldom needs to know the area of a surface.
Moreover, there is a simple expression for dA which suffices for
theoretical considerations.
128 CalCUlU8 on Manifold8
5-6 Theorem. Let M be an oriented two-dimensional man
i/old (or mani/old-with-boundary) in R3 and let n be the unit
outward normal. Then
(1) dA = n l dy 1\ dz + n 2 dz 1\ dx + nS dx 1\ dy .
Moreover, on M we have
(2) n l dA = dy 1\ dz.
(3) n 2 dA = dz 1\ dx .
(4) nS dA = dx 1\ dy.
Praof·
Equation ( 1 ) is equivalent to the equation
dA (v,w) = det (: )
n(x)
.
This is seen by expanding the detenninant by minors along

the bottom row. To prove the other equations, let z E R 3%.
Since v X w = exn (x) for Bome ex E R, we bave
(z,n(x» . (v X w , n(x» = (z,n(x » ex = (z, cxn(x» (z,v X w). =
Choosing z el, e2, -and es we obtain (2), (3), and (4) . I

=
A word of caution: if w E A 2 (R Sa) is defined by

w = n l (a) . dy(a) 1\ dz(a)
+ n 2 (a) . dz(a) 1\ dx(a)
+ nS(a) . dx(a) 1\ dy(a),
.
it is not true, for example, that
n l (a) W = dy(a) 1\ dz(a).
The two sides give the same result only when applied to
V,w E Ma.
A few remarks should be made to justify the definition of
length and surface area we have given. If c: [0,1] Rn is �
differentiable and c([O,I]) is a one-dimensional manifold-with.

boundary, it can be shown, but the proof is messy, that the
length of c([O,I]) is indeed the least upper bound of the lengths
Integration on Manilold8 129
F l G U R E 5-9. A surlace containing 20 triangles inscribed in a por

tion 01 a cylinder. Il the number 01 triangles is increased sufficiently, by
making the bases 01 triangles 3, 4, 7, 8, etc., sufficiently small, the total area
01 the inscribed surlace can be made as large as desired.
of inscribed broken Iines. If c : [0,1] 2 � Rn , one naturaIly

hopes that the area of c([O,1] 2) wiIl be the Ieast upper bound of
the areas of surfaces made up of triangles whose vertices Iie in
c([O,l] 2 ). Amazingly enough, such a Ieast upper bound is
usuaIly nonexistent-one can find inscribed polygonal surfaces
arbitrarily cIose to c([O,1] 2) with arbitrariIy large area! This
is indicated for a cylinder in Figure 5-9. Many definitions
of surface area have been proposed, disagreeing with each
other, but alI agreeing with our definition for differentiable
surfaces. For a discussion of these difficult questions the
reader is referred to References [3] or [15].
Problems. 5-23. If M is an oriented one-dimensional manifold in
Rn and c: [0,1] � M is orientation-preserving, show that
f c*(ds) =
f V [ (c l )']2 + . . . + [ (Cn)'J2.
[0, 1 ] [0,1]
5-24. If M is an n-dimensionai manifold in Rn, with the usual orienta

tion, show thut dV dXI /\ . . . /\ dJ;n, so that the volume of
=
M, as defined in this section, is the volume as defined in Chapter 3.

(Note that this depends on the numericai factor in the definition of
w /\ ".)
5-25. Generalize Theorem 5-6 to the case of an oriented (n - I)-dimen
sionaI manifold in Rn.
5-26. (a) If f: [a, b] � R is non-negative and the graph of f in the
xy-plane is revolved around the x-axis in R3 to yield a surface M,
show that the area of M is
b
f 27rf V l + (1')2.
a
(b) Compute the area of 82•

5-27. If T: Rn � Rn is a norm preserving linear transformation and M
is a k-dimensionai manifold in Rn, show that M has the same
volume as T (M) .
5-28. (a) If M is a k-dimensionai manifold, show that an absolute
k-tensor I dvl
can be defilled, even if M is not orientable, so that
the volume of M can be defined as MldVI. f
(b) If c : [O,27r] X (- l,l) � R3 is d efi ne d b y c(u,v) =
(2 cos u + v sin (u/2)cos u, 2 sin u + v sin (u/2) sin u, v cos u/2),
show that c([O,27r] X ( - l , l » is a Mobius strip and find its area.

Inlegralion on Manifolds 1 31
5-29. If there is a nowhere-zero k-form on a k-dimensional manifold M,

show that M is orientable.
5-30. (a) If f: [O,l J -4 R is differentiable and c: [O, lJ -4 R2 is defined by
e (x) = (x,f(x» , show that e([O,l]) has length n V I + (1') 2.
(b) Show that this length is the least upper bound of lengths of
inscribed broken lines. BinI: If O = l o S II S . . . S In = 1,
then
l e (Ii) - e (li ) I
-l = VUi - li_I)2 + (fUi) - f(li_1» 2
= V (Ii - li_d2 + f' (s;) 2 (li - li_I) 2
for some Si E [/i-1, liJ .
5-31. Consider the 2-form Col defined on R3 - O by
w �
x dy 1\ dz + y dz 1\ dx + z dx 1\ dy
(x 2 + y2 + Z2)t
(a) Show that Col is closed.
(b) Show that
For T > O let S2(T) = Ix E R 3 : Ixl = T I . Show that Col restricted

to the tangent space of S2(T) is 1/T2 times the volume element,
and that I S' er) Col = 4.... Conclude that Col is not exact. Neverthe
less we denote Col by de since, as we shall see, de is the analogue of
the l-form dO on n 2 - O.
(c) If vp is a tangent vector such that v = l\p for some li. E R
show that de(p) (vp,wp) = O for ali Wp. If a two-dimensional
manifold M in R 3 is part of a generalized cone, that is, M
is the union of segments of rays through the origin, show that
I ",I de = o.
(d) Let M C R3 O be a compact two-dimensional manifold
-
with-boundary such that every ray through O intersects M at most

once (Figure 5-10). The union of those rays through O which
intersect M, is a solid cone C(M). The solid angle subtended by M
is defined as the area of C(M) (ì S2, or equivalently as 1/T2 times
the area of C(M) (ì S2(T) for T > O. Prove that the solid angle
subtended by M is I I M del . BinI: Choose T small enough so
that there is a three-dimensional manifold-with-boundary N (as in
Figure 5-10) such that aN is the union of M and C(M) (ì S2(T),
and a part of a generalized cone. (Actually, N will be a manifold
with-corners ; see the remarks at the end of the next section.)
FIG U R E 5-10
5-32. Let f, g : [0, 1 ) -- n 3 be nonintersecting closed curves.

thc linking number l(j,g) of f and g by (cf. Problem 4-34)
Definc
l(j,g) = 2
411" •
f dO.
c/.v
(a) Show that if (F,G) i� a homotopy of nonintcrsccting closed

curves, then l(Fo,Go) = l(FI,GI).
(b) H r(u,v) = I f(u) - g(v) I show that

l l
4,..
f f [r(u,v)j3 . A (u,v) du dv
-1
1
t (f,g) =
o o
where
A (u,v) = det
( l- (f ) ' (u )
(l)' (V)
(f2) , (u)
(g2)' (V)
fl (U ) gl (v) f2(u) - g2(V)
(c) Show that l (f,g) = O if f and Il both lie in the xy-plane.

The curves of Figure 4-5 (b) '1re given by f(u) = (cos u, sin u, O)
and g(v) = (1 + cos v, O, sin v). You may easily convince
yourself that calculating l(f,g) by the above integraI is hopeless in
this case. The following problem shows how to find l(f,g) without
explicit caicuiations.
5-33. (a) H (a,b,c) E R 3 define
(x - a)dy 1\ dz + (y - b)dz 1\ dx + (z - c)dx 1\

de(a.b.c)
[(x - a) 2 + (y W + (z - c) 2Jf - dy.
H M is a compact two-dimensional manifold-with-boundary In

R 3 and (a,b,c) E M define
n(a,b,c) = J de(a.b.c) .
M
Let (a,b,c) be a point on the same side of M as the outward normal

and (a',b',c') a point on the opposite side. Show that by choosing
(a,b,c) sufficiently cIose to (a',b',c') we can make n(a,b,c) -
M = aN then n(a,b,c) = -
4,.. for (a,b,c) E N
- M and n(a,b,c) =
n(a',b',c') as close to -4,.. as desired. Hinl: First show that if
O for (a,b,c) E N.
(b) Suppose f([O,l]) = aM for some compact oriented two
dimensionai manifold-with-boundary M. (H f does not intersect
itself such an M always exists, even if f is knotted, see [6], page 138.)
Suppose that whenever g intersects M at x the tangent vector v of
g is not in M z. Let n+ be the number of intersections where v
points in the same direction as the outward normal and n- the
-
number of other intersections. If n n+ - n- show that =
n = -f
4,..
l
g
dn.
134. Calculus on Manifolds
(c) Prove that
f
(y - b)dz - (z - c)dy
Dlfl(a,b,c) =
!
r3
(z - c)dx - (x - a)dz
f
!
r3
f
(x - a)dy - (y - b)dx,
D3fl(a,b,c) =
!
r3
where r(x,y,z) =/ (x,y,z) / .

(d) Show that the integer n of (b) equals the integrai of Prob
lem 5-32 (b), and use this result to show that lU,g) 1 if f and g
=
are the curves of Figure 4-6 (b) , while lU,g) O if f and g are the
=
curves of Figure 4-6 (c) . (These results were known to Gauss

[7J. The proofs outlined here are from [4J pp. 409-41 1 ; see also
[13J, Volume 2, pp. 41-43. )
THE CLA SSICA L THEOREMS
We have now prepared alI the rnachinery necessary to state and

prove the classical "Stokes' type" of theorerns. We will
indulge in a little bit of self-explanatory classical notation.
5-7 Theorem (Green's Theorem). Let M C R 2 be a com
pact two-dimensional manijold-with-boundary. Suppose that
a,(3: M R are differentiable . Then
�
=
f a dx
aM
+ (3 dy
f (D 1(3 - D 2a)dx 1\ dy
M
= ff (: - �:) dX dY.
M
(Here M is given the usual orientation, and aM the induced

orientation, also known as the counterclockwise orientation.)
Proof. This is a very special case of Theorem 5-5, since
dea dx + (3 dy) = (D 1(3 - D2a)dx 1\ dy. I
�-8 Theorem (Divergence Theore m). Let M C R 3 be a

compact three-dimensional manijold-w#h-boundary and n the
unit outward normal on aMo Let F be a differentiable vector field
on M. Then
f div F dV = { (F,n) dA .
M a lt
This equation is also wl"itten in terms oj three differentiable junc
tions ex,fJ,'Y: M � R:
Proof. Define w on M by w = F 1 dy 1\ dz + F 2 dz 1\ dx +
F3 dx 1\ dy. Then dw = div F dV. According to Theorem
5-6, on aM we have
n dA1 = dy 1\ dz,
n 2 dA = dz 1\ dx,
n 3 dA = dx 1\ dy.
Therefore on aM we have
(F,n) dA F I n I dA + F2n 2 dA + F3n 3 dA
= F I dy 1\ dz + F 2 dz 1\ dx + F 3 dx 1\ dy
=
= W.
Thus, by Theorem 5-5 we have
f div F dV = f dw = f w � f (F, n) dA . I
M M aM aM
5-9 Theorem (Stokes' Theorem). Let M C R 3 be a com

pact oriented two-dimensional manijold-with-boundary and n the
unit outward normal on M determined by the orientation oj M.
Let aM have the induced orientation. Let T be the vector field on
aM with ds( T) = 1 and let F be a differentiable vector field in
an open set conlaining M. Then
f «(V' X F) , n) dA f (F,T) ds.
M aM
This equation is sometimes written
f dx + f3 dy + dz
ex
'Y
=
òM
il [n 1 ( ay - -aaZ(3) + n2 (-aaZex - -aaX'Y) + n3 (aX-af3 - -aaya)] dS.

a
- 'Y
Proof. Define w on M by w F 1 dx + F2 dy + F3 dz. =
Since V' X F has components D 2F3 - D 3F2, D 3F 1 - D 1F3,

DIF2 - D 2FI , it follows, as in the proof of Theorem 5-8, that
on M we have
«(V' X F), n) dA = (D 2F3 - D aF 2)dy /\ dz
+ (D 3F 1 - D1F3)dz /\ dx
+ (D 1F2 - D 2F l)dx /\ dy
= dw.
On the other hand, since ds ( T) = 1, on aM we have
T I ds = dx,
T2 ds = dy,
T3 ds = dz.
(These equations may be checked by appIying both sides to
Tz, for x E aM, since Tz is a basis for (aM)z.)
Therefore on aM we have
(F,T) ds = F1T 1 ds + F2 T2 ds + F3T3 ds
F1 dx + F2 dy + F3 dz
=
= w.
Thus, by Theorem 5-5, we have

J «(V' X F),n) dA J dw = J w = = J (F, T) ds. I
M M aM aM
Theorems 5-8 and 5-9 are the basis for the names div F and
curI F. If F (x) is the velocity vector of a fiuid at x (at some
time) then I'M (F,n) dA is the amount of f1uid Hdiverging"
from M. Consequently the condition div F = O expresses
Integrat i on on Manifolds 137
the fact that the fluid is incompressible. Ii M is a disc, then

faM (F, T) ds measures the amount that the fluid curls around
the center of the disc. Ii this is zero for all discs, then V X F
= 0 , and the fluid is called irrotational.
These interpretations of div F and curI F are due to Maxwell
[13]. Maxwell actually worked with the negative of div F,
which he accordingly called the convergence. For V X F
Maxwell proposed "with great diffidence" the terminology
rotation oj F; this unfortunate term suggested the abbreviation
rot F which one occasional1y still sees.
The classical theorems of this section are usually stated in
somewhat greater generality than they are here. For exam
pIe, Green's Theorem is true for a square, and the Divergence
Theorem is true for a cube. These two particular facts can
be proved by approximating the square or cube by manifolds
with-boundary. A thorough generalization of the theorems of
this section requires the concept of manifolds-with-corners;
these are subsets of Rn which are, up to diffeomorphism,
Iocally a portion of Rk which is bounded by pieces of (k - 1)
planes. The ambitious reader will find it a challenging exer
cise to define manifolds-with-corners rigorously and to
investigate how the results of this entire chapter may be
generalized.
Problems. 5-34. Generalize the divergence theorem to the case of
an n-manifold with boundary in R".
Ix E Rn : Ixl :::; a l and F(x) = x"" find the volume of Sn-I =

5-35. Applying the generalized divergence theorem to the set M =
Ix E Rn : Ixl = I l in terms of the n-dimensionaI volume of B" �
Ix E R": Ixl :::; I l . (This volume is ","/ 2/(n/2) ! if n is even and

2 ("+1)/ 2,,, (,,-1) /2 /1 . 3 5 · . .
. .. n if n is odd.)
5-36. Define F on R3 by F(x) . = (O,O,cx3)", and let M be a compact
three-dimensional manifold-with-boundary with M C Ix: x3 :::;
O l . The vector field F may be thought of as the downward pres
sure of a fluid of density c in Ix: x3 :::; O l . Since a fluid exerts
equal pressures in ali directions, we define the buoyant force on M,
due to the fluid, as -5 àM (F,n) dA . Prove the following theorem.
Theorem (Archimedea). The buoyant force on M is equal to the
weight of the fluid displaced by M.
Bibliography
l.
Auslander and MacKenzie, Introduction to Differentiable Manifolds,
Ahlfors, Complex Analysis, McGraw-Hill, New York, 1953.
2.
McGraw-Hill, New York, 1963.
3. Cesari, Surface Area, Princeton University Press, Princeton, New
Jersey, 1956.
4. Courant, Differential and IntegraI Calculus, Volume II, Interscience,
New York, 1937.
5. Dieudonné, Foundations of Modem Analysis, Academic Press,
New York, 1960.
6. Fort, Topology of 3-Manifolds, Prentice-Hall, Englewood Clilfs,
7. Gauss, Zur mathematischen 'l'heorie der electrodynamischen Wirkungen,

New Jersey, 1962.
(4) (Nachlass) Werke V, 605.

8. Helgason, Differential Geometry and Symmetric Spaces, Academic
Press, New York, 1962.
9. Hilton and Wylie, Homology 'l'heory, Cambridge Univer� ity Press,
lO. Hu, Homotopy Theory, Academic Press, New York, 1959.

New York, 1960.
1 1 . Kelley, GeneraI Topology, Van Nostrand, Princeton, New Jersey,

1955.
139
140 Bibliography
12. Kobayashi and Nomizu, Foundations oj Differential Geometry,

Interscience, New York, 1963.
13. Maxwell, Electricity and Magnetism, Dover, New York, 1954.
14. Natanson, Theory oj Functions oj a Real Variable, Frederick Ungar,
15. Rad6, Length and Area, Volume XXX, American Mathematical

New York, 1955.
Society, Colloquium Publications, New York, 1948.

16. de Rham, Variétés Differentiables, Hermann, Paris, 1955.
17. Sternberg, Lectures on Differential Geomelry, Prentice-Hall, Engle
wood Cliffs, New Jersey, 1964.
Index
Absolute differential form, 126 Boundary, of a set, 7

Absolute tensor, 126 Buoyant force, 137
Absolute value, l
Algebra, Fundamental Theorem of, Cauchy Integrai Formula, 106
105 Cauchy Integrai Theorem, 1 06
Alternating tensor, 78 Cauchy-Riemann equations,
Analytic function, 105 105
Angle, 4 Cavalieri's principi e, 62
preserving, 4 Chain, 97, 100
solid, 131 Chain rule, 1 9, 32
Approximation, 15 Change of variable, 67-72
Archimedes, 137 Characteristic function , 5.5
Area, 56 Closed curve, 106
element of, 126 Closed diffcrential form, 92
surface, 126, 127 Closed rectangle, 5
Closed set, 5
Basis, usual for Rn, 3 Compact, 7
Bilinear function, 3, 23 Complex numbers, 1 04
Boundary Complex variables, 105
of a chain, 97, 98 Component function, 1 1, 87
of a manifold-with-boundary, Composition, 1 1
1 13 Cone, generalized, 131
141
1 42 Index
Consistent choices of orientation, Differential form, on a manifold,

1 17 117
Constant function, 20 differentiable, 1 1 7
Constraints, 1 22 Dimension
Content, 156 of a manifold, 109
Content zero, 51 of a manifold-with-boundary,
Continuous differential form , 88 1 13
Continuous function, 1 2 Directional derivative, 33
Continuous vector field, 87 Distance, 4
Continuously differentiable, 31 Divergence of a field, 88, 137
Convergence, 137 Divergence Theorem, 135
Coordinate condition, 1 1 1 Domain, l 1
Coordinate system, 1 1 1 Dual space, 5
polar, 73
Counterclockwise orientation, 134
Element of area, 1 26
Cover, 7
Element of length, 1 26
Cross product, 84
Element of volume, see Volume
Cube
element
singular, 97
End point, 87
standard n-cube, 97
Equal up to nth order, 1 8
CurI, 88, 137
Euclidean space, 1
Curve, 97
Exact differential form, 92
closed, 106
Exterior of a set, 7
differentiable, !l6
"'
C , 26
Degenerate singular cuhe, 105

Faces of a singular cube, 98
Field, see Vector field
Derivative, 16 Form, see Differential form
directional, 33 Fubini's Theorem, 58
partial, 2S Function, 1 1
higher-order (mixed) , 26 analytic, 105
second-order (mixed), 26 characteristic, 55
Diffeomorphism, 109 component, 1 1 , 87
Differentiable function, 1 5, 16, composition of, 1 1
105 constant, 20
continuously, 31 continuous, 12
Differentiable curve, 96 continuously differentiable, 31
Differentiable differential form, 88 c"', 26
on a manifold, 1 17 differentiable, 1 5, 16, 105
Differentiable vector field, 87 homogeneous, 34
on a manifold, 1 1 5 identity, 1 1
Differentiahle = C"' , 88 implicitly defined, 41
Differential, 91 see also Implicit Function
Differential form, 8S Theorem
absolute, 126 integrable, 48
closed, 92 inverse, 1 1 , 34-39
eontinuous, 88 see also Inverse Function
differentiable, 88 Theorem
exact, 92 projection, 1 1
Index 143
Fundamental Theorern of Algebra, Kelvin, 74
105
Fundamental Theorern of Calcu Lac locus, 106
lus, 100-104 Lagrange's method, 1 22
Lagrangian multiplier, 1 22
Gauss, 1 34 Leibnitz's Rule, 62
Generalized eone, 131 Length, 56, 126
Grad f, 96 elernent of, 126
Graph, 1 1 , 1 1 5 Length = norm, 1
Green's Theorern , 134 Limit, 1 1
Line, 1
Half-space, 1 13 Une integrai, 101
Hcine-Borel Theorern, 7 Linking number, 132
HomogeneouB function, 34 LiouviIIe, 74
Homotopy, 1 08 Lower integrai, 58
Lower sum, 47
Idcntity function, 1 1
Irnplicit Function Theorem, 41
Manifold, 109
ImplicitIy defined function, 41
Manifold-with-boundary, 1 13
Incompressible tluid, 137
Manifold-with-corners, 1 3 1 , 137
Independence of parameteriza-
Mathematician (old style ) , 74
tion, 104
Matrix, 1
Induced orientation, 1 19
Jacobian, 1 7
Inequality, see Triangle inequality
transpose of, 23, 83
Inner product, 2, 77
Maxima, 26-27
preserving, 4
Measure zero, 50
usual, 77, 87
Minima, 26-27
Integrable function, 48
Mobius strip, 1 19, 120, 130
Integrai, 48
Multilinear function, 23, 75
iterated, 59, 60
Multiplier, see Lagrangian multi-
line, 101
plier
lower, 58
of a forrn on a manifold,
123-124 Norrn, l
of a form over a chain, 101 Norm preserving, 4
over a set, 55 Normal, see Outward unit normal
over an open set, 65 Notation, 3, 44, 89
surface, 1 02
upper, 58 One-one ( l-l) function, 1 1
Integrai Formula, Cauchy, 106 One-sided surface, 121
Integrai Theorem, Cauchy, 106 Open cover, 7
Open set, .5
In terior of a set, 7 Open rectangle, 5
Inverse function, 1 1, 34-39
Inverse Function Theorern, 35 Orientable rnanifold, 1 1 !J
Irrotational tluid, 137 Orientation, 82, 1 19
Iterated integrai, 59, 60 consistent choices of, 1 17
counterclockwise, 134
Jacobian rnatrix, 1 7 induced, 1 19
Jordan-rneasurable, 56 usual, 83, 87, 121
144 Index
Orientation-preserving, 1 18, 123 Stokes' Theorem, 102, 1 24, 1 35

Oriented manifold, 1 1 9 Subordinate, 63
Orthogonal vectors, 5 Subrectangles of a partition, 46
Orthonormal basis, 77 Surface, 127
Oscillation, 1 3 Surface area, 126, 1 27
Outward unit normal, 1 19, 120 Surface integrai, 102
Symmetric, 2, 77
Parameterization, independence of,
104 Tangent space, 86, 1 1 5
Partial derivative, 25 Tangent vector, 96
higher-order (mixed), 26 Tensor, 75
second-order (mixed) , 26 absolute, 126
Partition alternating, 78
of a c10sed intcrval, 46 Tensor product, 75
of a closed rectangle, 46 Torus, 1 15
of unity, 63 Transpose of a matrix, 23, 83
Perpendicular, 5 Triangle inequality, 4
Piane, 1
Poincarc Lemma, 94 Unit outward normal, 1 1 9, 120
Point, 1 Upper integrai, 58
Polar coordinate system, 73 Upper Bum, 47
Polarization identity, 5 Usual, 8ee Basis, Inner product,
Positive definiteness, 3, 77 Orientation
Product, see Cross product, Inner
product, Tensor product, Variable
Wedge prod uct change of, 67-72
Projection function, 1 1 complex, 8ee Complex variables
function of n, 1 1
Rectangle (c1osed or open), 5 independent of the first, 18
Refine a partition, 47 independent of the second, 1 7
Rotation of F, 137 Vector, l
tangent, 96
Sard's Theorem, 72 Vector field, 87
Self-adjoint, 85 continuouB, 87
Sign of a permutation, 78 differentiable, 87
Singular n-cube, 97 on a manifold, 1 1 5
Solid angle, 131 continuouB, 87
Space, 1 differentiable, 1 1 5
see also Dual space, Euclidean Vector-valued function, 1 1
space, Half-space, Tangent Volume, 47, 56, 126
space Volume element, 83, 126
Sphere, 1 1 1
Standard n-cube, !l7 Wedge product, 79
Star-shaped, 93 Winding number, 104
A ddenda
1
1. It should be remarked after Theorem 2-1 1 (the Inverse
Function Theorem) that the formula for r aIlows us to con
clude that rl is actually continuously differentiable (and that
it is C· if f is). Indeed, it suffices to note that the entries of
the inverse of a matrix A are C· functions of the entries
of A. This follows from "Cramer's Rule" : (A-l)ji =
(det A iJ)/(det A), where A ii is the matrix obtained from A

by deleting row i and column j.
2. The proof of the first part of Theorem 3-8 can be simpli

fied considerably, rendering Lemma 3-7 unnecessary. It
suffices to cover B by the interiors of closed rectangles Ui with
l::'IV(Ui) < E, and to choose for each xEA -B a closed
rectangle V"" containing x in its interior, with M v. (J) -
mv. (J) < E. If every subrectangle of a partition P is con
tained in one of some finite collection of U/s and V",'s which
cover A , and I f(x) I � M for alI x in A, then U(J, P) - L(J, P)
< Ev(A) + 2ME.
The proof of the converse part contains an error, since
M,(J) - m,(J) � l/n is guaranteed only if the interior of S
intersects B 1 1 '" To compensate for this it suffices to cover the
boundaries of alI subrectangles of P with a finite collection of
rectangles with total volume < E. These, together with S,
cciver B II", and have total volume < 2E.
145
146 Addenda
3. The argument in the first part of Theorem 3-14 (Sard's

Theorem) requires a little amplification. If U C A is a closed
rectangle with sides of length l, then, because U is compact,
there is an integer N with the following property : if U is
divided into N" rectangles, with sides oi length l/N, then
I Djg' (w) - Djg'(z) I < s./n2 whenever w and z are both in one
such rectangle S. Given x E S, let f(z) =
Dg(x) (z) - g(z) .
Then, if z E S,
IDif'(Z)IIDjg' (x) - Djg'(z)I
= < s./n2.
So by Lemma 2-10, if x,y E S, then
I Dg(x) (y - x) - g(y) + g(x) I =

If(y) - f(x) 1 < s.lx - yl
5 s. Vn (l/N).
4. Finally, the notation AA:(V) appearing in this book is

incorrect, since it conflicts with the standard definition of
AA:(V) (8.8 a certain quotient of the
tensor algebra of V). For
the vector space in question (which is naturally isomorphic to
AA:(V*) for finite dimensionai vector spaces V) the notation
oA:(V) is probably on the way to becoming standard. This
substitution should be made on pa.ges 78-85, 88-89, 1 16, and
126-128.

M. Spivak - Calculus On Manifolds PDF

Uploaded by

Copyright:

Available Formats

M. Spivak - Calculus On Manifolds PDF

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

M. Spivak - Calculus On Manifolds PDF

Uploaded by

Copyright:

Available Formats

Michael Spivak

ADDISON-WESLEY PUBLISHING COMPANY

Copyright © 1965 by Addison-Wesley Publishing Company

Library of Congress Card Catalog Number 66-10910

The manuscript was put imo production on Aprii 21,1965;

this volume was published on October 26, 1965

Mathematics has been expanding in ali directions at a fabulous

these pedagogical problems. They are being written by active

Princeton, New Jersey

This little book is especially concerned with those portions of

occurs in the last section. The very classical equation repro­

It is entirely reasonable to suppose that the difficulties so far

branches of mathematics could legitimately be recommended

1. Functions on Euclidean Space 1

FUNCTIONS AND CONTINUITY,

'l'Hl'; FUNDAfiH:NTAL THl';OREM OF CALCULUS,

5. lntegration on Manifolds 109

THF. CLASSICAL THEORl';MS,

Functions on Euclidean Space

Euclidean n-space Rn is defined as the set of all n-tuples

(1) is left to the reader.

= � i_ l (Xi) 2 + �i= 1 (yi) 2 + 2�i= l X iyi

::; IX! 2 + ly! 2 + 2 ! xl · ! y l by (2)

The quantity � i= lX iyi which appears in (2) is

(1) (x ,y) = (y,x) (symmetry).

(3) (x,x) � O, and (x, x) O if and = (positive definiteness) .

We conclude this section with some important remarks

If T: Rn -. Rm is a Iinear transformation, the matrix of T with

appeal' in the ith column of the matrix. If S: Rm � HP has

t hen T(x) = (yl, ... ,ym). One notational convention

l-I. * Prove that Ixl � 4�1 Ixil.

1-3. Prove that Ix - yl � Ixl + Iyl. When does equality hold?

1-4. Prove that I lxl - Iyl l � Ix - y

(b) If equality holds, must = ÀO for some À E R ? What if

1-7. A linear transformation T: R " -> R " i s nornt preserving if

IT(x)1 Ixl, and inner product preserving if (Tx,Ty) = (x,y).

(a) Prove that T is norm preserving if and only if T is inner­

L(x,y), is defined as areeos «x,y)/lxl ·Iyl), whieh makes sense by

(a) Prove that if T is norm preserving, then T is angle pre­

on\y ii ali I xii are equal.

(c) What are alI angle preserving T: Rn -> Rn?

l-IO. * If T: Rm -> Rn is a linear transformation, show that there is a

number M such that IT(h)1 :$ Mlhl for h E Rm• Hint: Estimate

S UB S ETS O F E UCLID E A N S P ACE

The closed intervaI [a,b] has a naturaI analogue in R2• This is

The reader should supply the proof that a closed reetangle in

3. H B is any open reetangle with x E B, then B contains

Those points satisfying ( 1) constitute the interior of A, those

1-3 Theorem (Heine-Borel). The closed interval [a,b] is

Proof. If 0 is an open cover of [a,b], Iet

Note that a E A and that A is clearly bounded above (by b).

If B C Rm is compact and x E Rn, Ìt is easy to see that

1-4 Theorem. lj B is compact and (9 is an open cove1' oj

Proof· Sin ce Ix} X B is compact, we can assume at the

1-5 Corollary. lf A C Rn and B C Rm are compact, then

Proof. If (9 is an open cover of AX B, then (9 covers (x l X B

1-6 Corollary. Al X . . . X Ak is compact if each Ai is.

1-17. Construct a set A C [0, 1 ] X [0, 1] such that A contains at most

this relation holds for x E U () B.

occurs in the last section. The very classical equation repro

(a) Prove that T is norm preserving if and only if T is inner

(a) Prove that if T is norm preserving, then T is angle pre

A function from Rn to Rm (sometimes called a (vector

takes Rn into Rm" or "f, taking Rn into Rm," depending on con

that we can get f(x) as close to b as desired, by choosing x suf

satisfy O < I x - al < 8. A function f: A -- Rm is called con