AdvancedCalculusMV PDF

Multivariable Advanced Calculus
Kenneth Kuttler
December 3, 2012
2
Contents
1 Introduction 9
2 Some Fundamental Concepts 11
2.1 Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Basic Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 The Schroder Bernstein Theorem . . . . . . . . . . . . . . . . . . 13
2.1.3 Equivalence Relations . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 limsup And liminf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Double Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 Basic Linear Algebra 23
3.1 Algebra in F
n
, Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Subspaces Spans And Bases . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Linear Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Block Multiplication Of Matrices . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.1 The Determinant Of A Matrix . . . . . . . . . . . . . . . . . . . 37
3.5.2 The Determinant Of A Linear Transformation . . . . . . . . . . 48
3.6 Eigenvalues And Eigenvectors Of Linear Transformations . . . . . . . . 49
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.8 Inner Product And Normed Linear Spaces . . . . . . . . . . . . . . . . . 52
3.8.1 The Inner Product In F
n
. . . . . . . . . . . . . . . . . . . . . . 52
3.8.2 General Inner Product Spaces . . . . . . . . . . . . . . . . . . . . 53
3.8.3 Normed Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8.4 The p Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.8.5 Orthonormal Bases . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.8.6 The Adjoint Of A Linear Transformation . . . . . . . . . . . . . 59
3.8.7 Schurs Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.9 Polar Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4 Sequences 71
4.1 Vector Valued Sequences And Their Limits . . . . . . . . . . . . . . . . 71
4.2 Sequential Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3 Closed And Open Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4 Cauchy Sequences And Completeness . . . . . . . . . . . . . . . . . . . 78
4.5 Shrinking Diameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3
4 CONTENTS
5 Continuous Functions 85
5.1 Continuity And The Limit Of A Sequence . . . . . . . . . . . . . . . . . 88
5.2 The Extreme Values Theorem . . . . . . . . . . . . . . . . . . . . . . . . 89
5.3 Connected Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.4 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.5 Sequences And Series Of Functions . . . . . . . . . . . . . . . . . . . . . 94
5.6 Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.7 Sequences Of Polynomials, Weierstrass Approximation . . . . . . . . . . 99
5.7.1 The Tietze Extension Theorem . . . . . . . . . . . . . . . . . . . 104
5.8 The Operator Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.9 Ascoli Arzela Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6 The Derivative 121
6.1 Basic Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.3 The Matrix Of The Derivative . . . . . . . . . . . . . . . . . . . . . . . 123
6.4 A Mean Value Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.5 Existence Of The Derivative, C
1
Functions . . . . . . . . . . . . . . . . 126
6.6 Higher Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.7 C
k
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.7.1 Some Standard Notation . . . . . . . . . . . . . . . . . . . . . . . 132
6.8 The Derivative And The Cartesian Product . . . . . . . . . . . . . . . . 132
6.9 Mixed Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.10 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.10.1 More Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.10.2 The Case Of R
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.11 Taylors Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.11.1 Second Derivative Test . . . . . . . . . . . . . . . . . . . . . . . . 145
6.12 The Method Of Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . 147
6.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7 Measures And Measurable Functions 153
7.1 Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
7.2 An Outer Measure On P (R) . . . . . . . . . . . . . . . . . . . . . . . . 155
7.3 General Outer Measures And Measures . . . . . . . . . . . . . . . . . . 157
7.3.1 Measures And Measure Spaces . . . . . . . . . . . . . . . . . . . 157
7.4 The Borel Sets, Regular Measures . . . . . . . . . . . . . . . . . . . . . 158
7.4.1 Denition of Regular Measures . . . . . . . . . . . . . . . . . . . 158
7.4.2 The Borel Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.4.3 Borel Sets And Regularity . . . . . . . . . . . . . . . . . . . . . . 159
7.5 Measures And Outer Measures . . . . . . . . . . . . . . . . . . . . . . . 165
7.5.1 Measures From Outer Measures . . . . . . . . . . . . . . . . . . . 165
7.5.2 Completion Of Measure Spaces . . . . . . . . . . . . . . . . . . . 169
7.6 One Dimensional Lebesgue Stieltjes Measure . . . . . . . . . . . . . . . 172
7.7 Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
CONTENTS 5
8 The Abstract Lebesgue Integral 181
8.1 Denition For Nonnegative Measurable Functions . . . . . . . . . . . . . 181
8.1.1 Riemann Integrals For Decreasing Functions . . . . . . . . . . . . 181
8.1.2 The Lebesgue Integral For Nonnegative Functions . . . . . . . . 182
8.2 The Lebesgue Integral For Nonnegative Simple Functions . . . . . . . . 183
8.3 The Monotone Convergence Theorem . . . . . . . . . . . . . . . . . . . 184
8.4 Other Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.5 Fatous Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
8.6 The Righteous Algebraic Desires Of The Lebesgue Integral . . . . . . . 186
8.7 The Lebesgue Integral, L
1
. . . . . . . . . . . . . . . . . . . . . . . . . . 187
8.8 Approximation With Simple Functions . . . . . . . . . . . . . . . . . . . 191
8.9 The Dominated Convergence Theorem . . . . . . . . . . . . . . . . . . . 193
8.10 Approximation With C
c
(Y ) . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.11 The One Dimensional Lebesgue Integral . . . . . . . . . . . . . . . . . . 197
8.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
9 The Lebesgue Integral For Functions Of p Variables 207
9.1 Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
9.2 p Dimensional Lebesgue Measure And Integrals . . . . . . . . . . . . . . 208
9.2.1 Iterated Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
9.2.2 p Dimensional Lebesgue Measure And Integrals . . . . . . . . . . 209
9.2.3 Fubinis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.4 Lebesgue Measure On R
p
. . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.5 Molliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
9.6 The Vitali Covering Theorem . . . . . . . . . . . . . . . . . . . . . . . . 225
9.7 Vitali Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
9.8 Change Of Variables For Linear Maps . . . . . . . . . . . . . . . . . . . 231
9.9 Change Of Variables For C
1
Functions . . . . . . . . . . . . . . . . . . . 236
9.10 Change Of Variables For Mappings Which Are Not One To One . . . . 242
9.11 Spherical Coordinates In p Dimensions . . . . . . . . . . . . . . . . . . . 243
9.12 Brouwer Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . 246
9.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
10 Brouwer Degree 257
10.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
10.2 Denitions And Elementary Properties . . . . . . . . . . . . . . . . . . . 259
10.2.1 The Degree For C
2
_
; R
n
_
. . . . . . . . . . . . . . . . . . . . . 260
10.2.2 Denition Of The Degree For Continuous Functions . . . . . . . 266
10.3 Borsuks Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
10.5 The Product Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
10.6 Jordan Separation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 279
10.7 Integration And The Degree . . . . . . . . . . . . . . . . . . . . . . . . . 283
10.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
11 Integration Of Dierential Forms 293
11.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
11.2 Some Important Measure Theory . . . . . . . . . . . . . . . . . . . . . . 296
11.2.1 Eggoros Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 296
11.2.2 The Vitali Convergence Theorem . . . . . . . . . . . . . . . . . . 298
11.3 The Binet Cauchy Formula . . . . . . . . . . . . . . . . . . . . . . . . . 300
11.4 The Area Measure On A Manifold . . . . . . . . . . . . . . . . . . . . . 301
6 CONTENTS
11.5 Integration Of Dierential Forms On Manifolds . . . . . . . . . . . . . . 304
11.5.1 The Derivative Of A Dierential Form . . . . . . . . . . . . . . . 307
11.6 Stokes Theorem And The Orientation Of . . . . . . . . . . . . . . . 307
11.7 Greens Theorem, An Example . . . . . . . . . . . . . . . . . . . . . . . 311
11.7.1 An Oriented Manifold . . . . . . . . . . . . . . . . . . . . . . . . 311
11.7.2 Greens Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
11.8 The Divergence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 315
11.9 Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
11.10Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
12 The Laplace And Poisson Equations 325
12.1 Balls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
12.2 Poissons Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
12.2.1 Poissons Problem For A Ball . . . . . . . . . . . . . . . . . . . . 331
12.2.2 Does It Work In Case f = 0? . . . . . . . . . . . . . . . . . . . . 333
12.2.3 The Case Where f = 0, Poissons Equation . . . . . . . . . . . . 335
12.3 Properties Of Harmonic Functions . . . . . . . . . . . . . . . . . . . . . 338
12.4 Laplaces Equation For General Sets . . . . . . . . . . . . . . . . . . . . 341
12.4.1 Properties Of Subharmonic Functions . . . . . . . . . . . . . . . 341
12.4.2 Poissons Problem Again . . . . . . . . . . . . . . . . . . . . . . . 346
13 The Jordan Curve Theorem 349
14 Line Integrals 361
14.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
14.1.1 Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
14.1.2 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
14.2 The Line Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
14.3 Simple Closed Rectiable Curves . . . . . . . . . . . . . . . . . . . . . . 376
14.3.1 The Jordan Curve Theorem . . . . . . . . . . . . . . . . . . . . . 377
14.3.2 Orientation And Greens Formula . . . . . . . . . . . . . . . . . 382
14.4 Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
14.5 Interpretation And Review . . . . . . . . . . . . . . . . . . . . . . . . . 391
14.5.1 The Geometric Description Of The Cross Product . . . . . . . . 391
14.5.2 The Box Product, Triple Product . . . . . . . . . . . . . . . . . . 392
14.5.3 A Proof Of The Distributive Law For The Cross Product . . . . 393
14.5.4 The Coordinate Description Of The Cross Product . . . . . . . . 393
14.5.5 The Integral Over A Two Dimensional Surface . . . . . . . . . . 394
14.6 Introduction To Complex Analysis . . . . . . . . . . . . . . . . . . . . . 396
14.6.1 Basic Theorems, The Cauchy Riemann Equations . . . . . . . . 396
14.6.2 Contour Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
14.6.3 The Cauchy Integral . . . . . . . . . . . . . . . . . . . . . . . . . 400
14.6.4 The Cauchy Goursat Theorem . . . . . . . . . . . . . . . . . . . 403
14.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
15 Hausdor Measures And Area Formula 417
15.1 Denition Of Hausdor Measures . . . . . . . . . . . . . . . . . . . . . . 417
15.1.1 Properties Of Hausdor Measure . . . . . . . . . . . . . . . . . . 418
15.1.2 H
n
And m
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
15.2 Technical Considerations
. . . . . . . . . . . . . . . . . . . . . . . . . . 423
15.2.1 Steiner Symmetrization
. . . . . . . . . . . . . . . . . . . . . . . 425
15.2.2 The Isodiametric Inequality
. . . . . . . . . . . . . . . . . . . . 427
15.2.3 The Proper Value Of (n)
. . . . . . . . . . . . . . . . . . . . . 427
CONTENTS 7
15.2.4 A Formula For (n)
. . . . . . . . . . . . . . . . . . . . . . . . 428
15.3 Hausdor Measure And Linear Transformations . . . . . . . . . . . . . . 430
15.4 The Area Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
15.4.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . 432
15.5 The Area Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
15.6 Area Formula For Mappings Which Are Not One To One . . . . . . . . 442
15.7 The Coarea Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
15.8 A Nonlinear Fubinis Theorem . . . . . . . . . . . . . . . . . . . . . . . 454
Copyright c 2007,
8 CONTENTS
Introduction
This book is directed to people who have a good understanding of the concepts of one
variable calculus including the notions of limit of a sequence and completeness of R. It
develops multivariable advanced calculus.
In order to do multivariable calculus correctly, you must rst understand some linear
algebra. Therefore, a condensed course in linear algebra is presented rst, emphasizing
those topics in linear algebra which are useful in analysis, not those topics which are
primarily dependent on row operations.
Many topics could be presented in greater generality than I have chosen to do. I have
also attempted to feature calculus, not topology although there are many interesting
topics from topology. This means I introduce the topology as it is needed rather than
using the possibly more ecient practice of placing it right at the beginning in more
generality than will be needed. I think it might make the topological concepts more
memorable by linking them in this way to other concepts.
After the chapter on the n dimensional Lebesgue integral, you can make a choice
between a very general treatment of integration of dierential forms based on degree
theory in chapters 10 and 11 or you can follow an independent path through a proof
of a general version of Greens theorem in the plane leading to a very good version of
Stokes theorem for a two dimensional surface by following Chapters 12 and 13. This
approach also leads naturally to contour integrals and complex analysis. I got this idea
from reading Apostols advanced calculus book. Finally, there is an introduction to
Hausdor measures and the area formula in the last chapter.
I have avoided many advanced topics like the Radon Nikodym theorem, represen-
tation theorems, function spaces, and dierentiation theory. It seems to me these are
topics for a more advanced course in real analysis. I chose to feature the Lebesgue
integral because I have gone through the theory of the Riemann integral for a function
of n variables and ended up thinking it was too fussy and that the extra abstraction of
the Lebesgue integral was worthwhile in order to avoid this fussiness. Also, it seemed
to me that this book should be in some sense more advanced than my calculus book
which does contain in an appendix all this fussy theory.
9
10 INTRODUCTION
Some Fundamental Concepts
2.1 Set Theory
2.1.1 Basic Denitions
A set is a collection of things called elements of the set. For example, the set of integers,
the collection of signed whole numbers such as 1,2,-4, etc. This set whose existence will
be assumed is denoted by Z. Other sets could be the set of people in a family or
the set of donuts in a display case at the store. Sometimes parentheses, { } specify
a set by listing the things which are in the set between the parentheses. For example
the set of integers between -1 and 2, including these numbers could be denoted as
{1, 0, 1, 2}. The notation signifying x is an element of a set S, is written as x S.
Thus, 1 {1, 0, 1, 2, 3}. Here are some axioms about sets. Axioms are statements
which are accepted, not proved.
1. Two sets are equal if and only if they have the same elements.
2. To every set, A, and to every condition S (x) there corresponds a set, B, whose
elements are exactly those elements x of A for which S (x) holds.
3. For every collection of sets there exists a set that contains all the elements that
belong to at least one set of the given collection.
4. The Cartesian product of a nonempty family of nonempty sets is nonempty.
5. If A is a set there exists a set, P (A) such that P (A) is the set of all subsets of A.
This is called the power set.
These axioms are referred to as the axiom of extension, axiom of specication, axiom
of unions, axiom of choice, and axiom of powers respectively.
It seems fairly clear you should want to believe in the axiom of extension. It is
merely saying, for example, that {1, 2, 3} = {2, 3, 1} since these two sets have the same
elements in them. Similarly, it would seem you should be able to specify a new set from
a given set using some condition which can be used as a test to determine whether
the element in question is in the set. For example, the set of all integers which are
multiples of 2. This set could be specied as follows.
{x Z : x = 2y for some y Z} .
In this notation, the colon is read as such that and in this case the condition is being
a multiple of 2.
Another example of political interest, could be the set of all judges who are not
judicial activists. I think you can see this last is not a very precise condition since
11
12 SOME FUNDAMENTAL CONCEPTS
there is no way to determine to everyones satisfaction whether a given judge is an
activist. Also, just because something is grammatically correct does not mean
it makes any sense. For example consider the following nonsense.
S = {x set of dogs : it is colder in the mountains than in the winter} .
So what is a condition?
We will leave these sorts of considerations and assume our conditions make sense.
The axiom of unions states that for any collection of sets, there is a set consisting of all
the elements in each of the sets in the collection. Of course this is also open to further
consideration. What is a collection? Maybe it would be better to say set of sets or,
given a set whose elements are sets there exists a set whose elements consist of exactly
those things which are elements of at least one of these sets. If S is such a set whose
elements are sets,
{A : A S} or S
signify this union.
Something is in the Cartesian product of a set or family of sets if it consists of
a single thing taken from each set in the family. Thus (1, 2, 3) {1, 4, .2} {1, 2, 7}
{4, 3, 7, 9} because it consists of exactly one element from each of the sets which are
separated by . Also, this is the notation for the Cartesian product of nitely many
sets. If S is a set whose elements are sets,
AS
A
signies the Cartesian product.
The Cartesian product is the set of choice functions, a choice function being a func-
tion which selects exactly one element of each set of S. You may think the axiom of
choice, stating that the Cartesian product of a nonempty family of nonempty sets is
nonempty, is innocuous but there was a time when many mathematicians were ready
to throw it out because it implies things which are very hard to believe, things which
never happen without the axiom of choice.
A is a subset of B, written A B, if every element of A is also an element of B.
This can also be written as B A. A is a proper subset of B, written A B or B A
if A is a subset of B but A is not equal to B, A = B. A B denotes the intersection of
the two sets A and B and it means the set of elements of A which are also elements of
B. The axiom of specication shows this is a set. The empty set is the set which has
no elements in it, denoted as . A B denotes the union of the two sets A and B and
it means the set of all elements which are in either of the sets. It is a set because of the
axiom of unions.
The complement of a set, (the set of things which are not in the given set ) must be
taken with respect to a given set called the universal set which is a set which contains
the one whose complement is being taken. Thus, the complement of A, denoted as A
C
( or more precisely as X \ A) is a set obtained from using the axiom of specication to
write
A
C
{x X : x / A}
The symbol / means: is not an element of. Note the axiom of specication takes
place relative to a given set. Without this universal set it makes no sense to use the
axiom of specication to obtain the complement.
Words such as all or there exists are called quantiers and they must be under-
stood relative to some given set. For example, the set of all integers larger than 3. Or
there exists an integer larger than 7. Such statements have to do with a given set, in
this case the integers. Failure to have a reference set when quantiers are used turns
2.1. SET THEORY 13
out to be illogical even though such usage may be grammatically correct. Quantiers
are used often enough that there are symbols for them. The symbol is read as for
all or for every and the symbol is read as there exists. Thus could mean
for every upside down A there exists a backwards E.
DeMorgans laws are very useful in mathematics. Let S be a set of sets each of
which is contained in some universal set, U. Then
_
A
C
: A S
_
= ({A : A S})
C
and
_
A
C
: A S
_
= ({A : A S})
C
.
These laws follow directly from the denitions. Also following directly from the deni-
tions are:
Let S be a set of sets then
B {A : A S} = {B A : A S} .
and: Let S be a set of sets show
B {A : A S} = {B A : A S} .
Unfortunately, there is no single universal set which can be used for all sets. Here is
why: Suppose there were. Call it S. Then you could consider A the set of all elements
of S which are not elements of themselves, this from the axiom of specication. If A
is an element of itself, then it fails to qualify for inclusion in A. Therefore, it must not
be an element of itself. However, if this is so, it qualies for inclusion in A so it is an
element of itself and so this cant be true either. Thus the most basic of conditions you
could imagine, that of being an element of, is meaningless and so allowing such a set
causes the whole theory to be meaningless. The solution is to not allow a universal set.
As mentioned by Halmos in Naive set theory, Nothing contains everything. Always
beware of statements involving quantiers wherever they occur, even this one. This little
observation described above is due to Bertrand Russell and is called Russells paradox.
2.1.2 The Schroder Bernstein Theorem
It is very important to be able to compare the size of sets in a rational way. The most
useful theorem in this context is the Schroder Bernstein theorem which is the main
result to be presented in this section. The Cartesian product is discussed above. The
next denition reviews this and denes the concept of a function.
Denition 2.1.1 Let X and Y be sets.
X Y {(x, y) : x X and y Y }
A relation is dened to be a subset of XY . A function, f, also called a mapping, is a
relation which has the property that if (x, y) and (x, y
1
) are both elements of the f, then
y = y
1
. The domain of f is dened as
D(f) {x : (x, y) f} ,
written as f : D(f) Y .
It is probably safe to say that most people do not think of functions as a type of
relation which is a subset of the Cartesian product of two sets. A function is like a
machine which takes inputs, x and makes them into a unique output, f (x). Of course,
that is what the above denition says with more precision. An ordered pair, (x, y)
which is an element of the function or mapping has an input, x and a unique output,
y,denoted as f (x) while the name of the function is f. mapping is often a noun
meaning function. However, it also is a verb as in f is mapping A to B . That which
a function is thought of as doing is also referred to using the word maps as in: f maps
X to Y . However, a set of functions may be called a set of maps so this word might
also be used as the plural of a noun. There is no help for it. You just have to suer
with this nonsense.
The following theorem which is interesting for its own sake will be used to prove the
Schroder Bernstein theorem.
Theorem 2.1.2 Let f : X Y and g : Y X be two functions. Then there
exist sets A, B, C, D, such that
A B = X, C D = Y, A B = , C D = ,
f (A) = C, g (D) = B.
The following picture illustrates the conclusion of this theorem.
B = g(D)
A
E
'
D
C = f(A)
Y X
f
g
Proof: Consider the empty set, X. If y Y \ f (), then g (y) / because
has no elements. Also, if A, B, C, and D are as described above, A also would have this
same property that the empty set has. However, A is probably larger. Therefore, say
A
0
X satises P if whenever y Y \ f (A
0
) , g (y) / A
0
.
A {A
0
X : A
0
satises P}.
Let A = A. If y Y \ f (A), then for each A
0
A, y Y \ f (A
0
) and so g (y) / A
0
.
Since g (y) / A
0
for all A
0
A, it follows g (y) / A. Hence A satises P and is the
largest subset of X which does so. Now dene
C f (A) , D Y \ C, B X \ A.
It only remains to verify that g (D) = B.
Suppose x B = X \ A. Then A {x} does not satisfy P and so there exists
y Y \ f (A {x}) D such that g (y) A {x} . But y / f (A) and so since A
satises P, it follows g (y) / A. Hence g (y) = x and so x g (D) and This proves the
theorem.
Theorem 2.1.3 (Schroder Bernstein) If f : X Y and g : Y X are one to
one, then there exists h : X Y which is one to one and onto.
Proof: Let A, B, C, D be the sets of Theorem2.1.2 and dene
h(x)
_
f (x) if x A
g
1
(x) if x B
Then h is the desired one to one and onto mapping.
Recall that the Cartesian product may be considered as the collection of choice
functions.
2.1. SET THEORY 15
Denition 2.1.4 Let I be a set and let X
i
be a set for each i I. f is a choice
function written as
f
iI
X
i
if f (i) X
i
for each i I.
The axiom of choice says that if X
i
= for each i I, for I a set, then
iI
X
i
= .
Sometimes the two functions, f and g are onto but not one to one. It turns out that
with the axiom of choice, a similar conclusion to the above may be obtained.
Corollary 2.1.5 If f : X Y is onto and g : Y X is onto, then there exists
h : X Y which is one to one and onto.
Proof: For each y Y , f
1
(y) {x X : f (x) = y} = . Therefore, by the axiom
of choice, there exists f
1
0

yY
f
1
(y) which is the same as saying that for each
y Y , f
1
0
(y) f
1
(y). Similarly, there exists g
1
0
(x) g
1
(x) for all x X. Then
f
1
0
is one to one because if f
1
0
(y
1
) = f
1
0
(y
2
), then
y
1
= f
_
f
1
0
(y
1
)
_
= f
_
f
1
0
(y
2
)
_
= y
2
.
Similarly g
1
0
is one to one. Therefore, by the Schroder Bernstein theorem, there exists
h : X Y which is one to one and onto.
Denition 2.1.6 A set S, is nite if there exists a natural number n and a
map which maps {1, , n} one to one and onto S. S is innite if it is not nite.
A set S, is called countable if there exists a map mapping N one to one and onto
S.(When maps a set A to a set B, this will be written as : A B in the future.)
Here N {1, 2, }, the natural numbers. S is at most countable if there exists a map
: N S which is onto.
The property of being at most countable is often referred to as being countable
because the question of interest is normally whether one can list all elements of the set,
designating a rst, second, third etc. in such a way as to give each element of the set a
natural number. The possibility that a single element of the set may be counted more
than once is often not important.
Theorem 2.1.7 If X and Y are both at most countable, then X Y is also at
most countable. If either X or Y is countable, then X Y is also countable.
Proof: It is given that there exists a mapping : N X which is onto. Dene
(i) x
i
and consider X as the set {x
1
, x
2
, x
3
, }. Similarly, consider Y as the set
{y
1
, y
2
, y
3
, }. It follows the elements of XY are included in the following rectangular
array.
(x
1
, y
1
) (x
1
, y
2
) (x
1
, y
3
) Those which have x
1
in rst slot.
(x
2
, y
1
) (x
2
, y
2
) (x
2
, y
3
2
in rst slot.
(x
3
, y
1
) (x
3
, y
2
) (x
3
, y
3
3
in rst slot.
.
.
.
.
.
.
.
.
.
.
.
.
.
Follow a path through this array as follows.
(x
1
, y
1
) (x
1
, y
2
) (x
1
, y
3
)

(x
2
, y
1
) (x
2
, y
2
)

(x
3
, y
1
)
Thus the rst element of X Y is (x
1
, y
1
), the second element of X Y is (x
1
, y
2
), the
third element of X Y is (x
2
, y
1
) etc. This assigns a number from N to each element
of X Y. Thus X Y is at most countable.
It remains to show the last claim. Suppose without loss of generality that X is
countable. Then there exists : N X which is one to one and onto. Let : XY N
be dened by ((x, y))
1
(x). Thus is onto N. By the rst part there exists a
function from N onto X Y . Therefore, by Corollary 2.1.5, there exists a one to one
and onto mapping from X Y to N. This proves the theorem.
Theorem 2.1.8 If X and Y are at most countable, then X Y is at most
countable. If either X or Y are countable, then X Y is countable.
Proof: As in the preceding theorem,
X = {x
1
, x
2
, x
3
, }
and
Y = {y
1
, y
2
, y
3
, } .
Consider the following array consisting of X Y and path through it.
x
1
x
2
x
3

y
1
y
2
Thus the rst element of X Y is x
1
, the second is x
2
the third is y
1
the fourth is y
2
etc.
Consider the second claim. By the rst part, there is a map from N onto X Y .
Suppose without loss of generality that X is countable and : N X is one to one and
onto. Then dene (y) 1, for all y Y ,and (x)
1
(x). Thus, maps X Y
onto N and this shows there exist two onto maps, one mapping X Y onto N and the
other mapping N onto X Y . Then Corollary 2.1.5 yields the conclusion. This proves
the theorem.
2.1.3 Equivalence Relations
There are many ways to compare elements of a set other than to say two elements are
equal or the same. For example, in the set of people let two people be equivalent if they
have the same weight. This would not be saying they were the same person, just that
they weighed the same. Often such relations involve considering one characteristic of
the elements of a set and then saying the two elements are equivalent if they are the
same as far as the given characteristic is concerned.
Denition 2.1.9 Let S be a set. is an equivalence relation on S if it satises
the following axioms.
1. x x for all x S. (Reexive)
2.2. LIMSUP AND LIMINF 17
2. If x y then y x. (Symmetric)
3. If x y and y z, then x z. (Transitive)
Denition 2.1.10 [x] denotes the set of all elements of S which are equivalent
to x and [x] is called the equivalence class determined by x or just the equivalence class
of x.
With the above denition one can prove the following simple theorem.
Theorem 2.1.11 Let be an equivalence class dened on a set, S and let H
denote the set of equivalence classes. Then if [x] and [y] are two of these equivalence
classes, either x y and [x] = [y] or it is not true that x y and [x] [y] = .
2.2 limsup And liminf
It is assumed in all that is done that R is complete. There are two ways to describe
completeness of R. One is to say that every bounded set has a least upper bound and a
greatest lower bound. The other is to say that every Cauchy sequence converges. These
two equivalent notions of completeness will be taken as given.
The symbol, F will mean either R or C. The symbol [, ] will mean all real
numbers along with + and which are points which we pretend are at the right
and left ends of the real line respectively. The inclusion of these make believe points
makes the statement of certain theorems less trouble.
Denition 2.2.1 For A [, ] , A = sup A is dened as the least upper
bound in case A is bounded above by a real number and equals if A is not bounded
above. Similarly inf A is dened to equal the greatest lower bound in case A is bounded
below by a real number and equals in case A is not bounded below.
Lemma 2.2.2 If {A
n
} is an increasing sequence in [, ], then
sup {A
n
} = lim
n
A
n
.
Similarly, if {A
n
} is decreasing, then
inf {A
n
} = lim
n
A
n
.
Proof: Let sup ({A
n
: n N}) = r. In the rst case, suppose r < . Then letting
> 0 be given, there exists n such that A
n
(r , r]. Since {A
n
} is increasing, it
follows if m > n, then r < A
n
A
m
r and so lim
n
A
n
= r as claimed. In
the case where r = , then if a is a real number, there exists n such that A
n
> a.
Since {A
k
} is increasing, it follows that if m > n, A
m
> a. But this is what is meant
by lim
n
A
n
= . The other case is that r = . But in this case, A
n
= for all
n and so lim
n
A
n
= . The case where A
n
is decreasing is entirely similar. This
proves the lemma.
Sometimes the limit of a sequence does not exist. For example, if a
n
= (1)
n
, then
lim
n
a
n
does not exist. This is because the terms of the sequence are a distance
of 1 apart. Therefore there cant exist a single number such that all the terms of the
sequence are ultimately within 1/4 of that number. The nice thing about limsup and
liminf is that they always exist. First here is a simple lemma and denition.
Denition 2.2.3 Denote by [, ] the real line along with symbols and
. It is understood that is larger than every real number and is smaller
than every real number. Then if {A
n
} is an increasing sequence of points of [, ] ,
lim
n
A
n
equals if the only upper bound of the set {A
n
} is . If {A
n
} is bounded
above by a real number, then lim
n
A
n
is dened in the usual way and equals the
least upper bound of {A
n
}. If {A
n
} is a decreasing sequence of points of [, ] ,
lim
n
A
n
equals if the only lower bound of the sequence {A
n
} is . If {A
n
} is
bounded below by a real number, then lim
n
A
n
is dened in the usual way and equals
the greatest lower bound of {A
n
}. More simply, if {A
n
} is increasing,
lim
n
A
n
= sup {A
n
}
and if {A
n
} is decreasing then
lim
n
A
n
= inf {A
n
} .
Lemma 2.2.4 Let {a
n
} be a sequence of real numbers and let U
n
sup {a
k
: k n} .
Then {U
n
} is a decreasing sequence. Also if L
n
inf {a
k
: k n} , then {L
n
} is an
increasing sequence. Therefore, lim
n
L
n
and lim
n
U
n
both exist.
Proof: Let W
n
be an upper bound for {a
k
: k n} . Then since these sets are
getting smaller, it follows that for m < n, W
m
is an upper bound for {a
k
: k n} . In
particular if W
m
= U
m
, then U
m
is an upper bound for {a
k
: k n} and so U
m
is at
least as large as U
n
, the least upper bound for {a
k
: k n} . The claim that {L
n
} is
decreasing is similar. This proves the lemma.
From the lemma, the following denition makes sense.
Denition 2.2.5 Let {a
n
} be any sequence of points of [, ]
lim sup
n
a
n
lim
n
sup {a
k
: k n}
lim inf
n
a
n
lim
n
inf {a
k
: k n} .
Theorem 2.2.6 Suppose {a
n
} is a sequence of real numbers and that
lim sup
n
a
n
and
lim inf
n
a
n
are both real numbers. Then lim
n
a
n
exists if and only if
lim inf
n
a
n
= lim sup
n
a
n
and in this case,
lim
n
a
n
= lim inf
n
a
n
= lim sup
n
a
n
.
Proof: First note that
sup {a
k
: k n} inf {a
k
: k n}
and so from Theorem 4.1.7,
lim sup
n
a
n
lim
n
sup {a
k
: k n}
lim
n
inf {a
k
: k n}
lim inf
n
a
n
.
2.2. LIMSUP AND LIMINF 19
Suppose rst that lim
n
a
n
exists and is a real number. Then by Theorem 4.4.3 {a
n
}
is a Cauchy sequence. Therefore, if > 0 is given, there exists N such that if m, n N,
then
|a
n
a
m
| < /3.
From the denition of sup {a
k
: k N} , there exists n
1
N such that
sup {a
k
: k N} a
n1
+/3.
Similarly, there exists n
2
N such that
inf {a
k
: k N} a
n2
/3.
It follows that
sup {a
k
: k N} inf {a
k
: k N} |a
n1
a
n2
| +
2
3
< .
Since the sequence, {sup {a
k
: k N}}
N=1
is decreasing and {inf {a
k
: k N}}
N=1
is
increasing, it follows from Theorem 4.1.7
0 lim
N
sup {a
k
: k N} lim
N
inf {a
k
: k N}
Since is arbitrary, this shows
lim
N
sup {a
k
: k N} = lim
N
inf {a
k
: k N} (2.1)
Next suppose 2.1. Then
lim
N
(sup {a
k
: k N} inf {a
k
: k N}) = 0
Since sup {a
k
: k N} inf {a
k
: k N} it follows that for every > 0, there exists
N such that
sup {a
k
: k N} inf {a
k
: k N} <
Thus if m, n > N, then
|a
m
a
n
| <
which means {a
n
} is a Cauchy sequence. Since R is complete, it follows that lim
n
a
n

a exists. By the squeezing theorem, it follows
a = lim inf
n
a
n
= lim sup
n
a
n
and This proves the theorem.
With the above theorem, here is how to dene the limit of a sequence of points in
[, ].
n
} be a sequence of points of [, ] . Then lim
n
a
n
exists exactly when
lim inf
n
a
n
= lim sup
n
a
n
and in this case
lim
n
a
n
lim inf
n
a
n
= lim sup
n
a
n
.
The signicance of limsup and liminf, in addition to what was just discussed, is
contained in the following theorem which follows quickly from the denition.
n
} is a sequence of points of [, ] . Let
= lim sup
n
a
n
.
Then if b > , it follows there exists N such that whenever n N,
a
n
b.
If c < , then a
n
> c for innitely many values of n. Let
= lim inf
n
a
n
.
Then if d < , it follows there exists N such that whenever n N,
a
n
d.
If e > , it follows a
n
< e for innitely many values of n.
The proof of this theorem is left as an exercise for you. It follows directly from the
denition and it is the sort of thing you must do yourself. Here is one other simple
proposition.
Proposition 2.2.9 Let lim
n
a
n
= a > 0. Then
lim sup
n
a
n
b
n
= a lim sup
n
b
n
.
Proof: This follows from the denition. Let
n
= sup {a
k
b
k
: k n} . For all n
large enough, a
n
> a where is small enough that a > 0. Therefore,
n
sup {b
k
: k n} (a )
for all n large enough. Then
lim sup
n
a
n
b
n
= lim
n
n
lim sup
n
a
n
b
n
lim
n
(sup {b
k
: k n} (a ))
= (a ) lim sup
n
b
n
Similar reasoning shows
lim sup
n
a
n
b
n
(a +) lim sup
n
b
n
Now since > 0 is arbitrary, the conclusion follows.
2.3 Double Series
Sometimes it is required to consider double series which are of the form
k=m
j=m
a
jk

k=m
_
_
j=m
a
jk
_
_
.
In other words, rst sum on j yielding something which depends on k and then sum
these. The major consideration for these double series is the question of when
k=m
j=m
a
jk
=
j=m
k=m
a
jk
.
2.3. DOUBLE SERIES 21
In other words, when does it make no dierence which subscript is summed over rst?
In the case of nite sums there is no issue here. You can always write
M
k=m
N
j=m
a
jk
=
N
j=m
M
k=m
a
jk
because addition is commutative. However, there are limits involved with innite sums
and the interchange in order of summation involves taking limits in a dierent order.
Therefore, it is not always true that it is permissible to interchange the two sums. A
general rule of thumb is this: If something involves changing the order in which two
limits are taken, you may not do it without agonizing over the question. In general,
limits foul up algebra and also introduce things which are counter intuitive. Here is an
example. This example is a little technical. It is placed here just to prove conclusively
there is a question which needs to be considered.
Example 2.3.1 Consider the following picture which depicts some of the ordered pairs
(m, n) where m, n are positive integers.
0
0
0
0
0
0
0
-c
-c
-c
-c
-c
-c
a
c
c
c
c
c
b
0
0
0
0
0
0
0
0
0
0
0
0
0
0 0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
The numbers next to the point are the values of a
mn
. You see a
nn
= 0 for all n,
a
21
= a, a
12
= b, a
mn
= c for (m, n) on the line y = 1 + x whenever m > 1, and
a
mn
= c for all (m, n) on the line y = x 1 whenever m > 2.
Then
m=1
a
mn
= a if n = 1,
m=1
a
mn
= bc if n = 2 and if n > 2,
m=1
a
mn
=
0. Therefore,
n=1
m=1
a
mn
= a +b c.
Next observe that
n=1
a
mn
= b if m = 1,
n=1
a
mn
= a+c if m = 2, and
n=1
a
mn
=
0 if m > 2. Therefore,
m=1
n=1
a
mn
= b +a +c
and so the two sums are dierent. Moreover, you can see that by assigning dierent
values of a, b, and c, you can get an example for any two dierent numbers desired.
It turns out that if a
ij
0 for all i, j, then you can always interchange the order
of summation. This is shown next and is based on the following lemma. First, some
notation should be discussed.
Denition 2.3.2 Let f (a, b) [, ] for a A and b B where A, B
are sets which means that f (a, b) is either a number, , or . The symbol, +
is interpreted as a point out at the end of the number line which is larger than every
real number. Of course there is no such number. That is why it is called . The
symbol, is interpreted similarly. Then sup
aA
f (a, b) means sup (S
b
) where S
b

{f (a, b) : a A} .
Unlike limits, you can take the sup in dierent orders.
Lemma 2.3.3 Let f (a, b) [, ] for a A and b B where A, B are sets.
Then
sup
aA
sup
bB
f (a, b) = sup
bB
sup
aA
f (a, b) .
Proof: Note that for all a, b, f (a, b) sup
bB
sup
aA
f (a, b) and therefore, for all
a, sup
bB
f (a, b) sup
bB
sup
aA
f (a, b). Therefore,
sup
aA
sup
bB
f (a, b) sup
bB
sup
aA
f (a, b) .
Repeat the same argument interchanging a and b, to get the conclusion of the lemma.
Theorem 2.3.4 Let a
ij
0. Then
i=1
j=1
a
ij
=
j=1
i=1
a
ij
.
Proof: First note there is no trouble in dening these sums because the a
ij
are all
nonnegative. If a sum diverges, it only diverges to and so is the value of the sum.
Next note that
j=r
i=r
a
ij
sup
n
j=r
n
i=r
a
ij
because for all j,
i=r
a
ij

n
i=r
a
ij
.
Therefore,
j=r
i=r
a
ij
sup
n
j=r
n
i=r
a
ij
= sup
n
lim
m
m
j=r
n
i=r
a
ij
= sup
n
lim
m
n
i=r
m
j=r
a
ij
= sup
n
n
i=r
lim
m
m
j=r
a
ij
= sup
n
n
i=r
j=r
a
ij
= lim
n
n
i=r
j=r
a
ij
=
i=r
j=r
a
ij
Interchanging the i and j in the above argument proves the theorem.
Basic Linear Algebra
All the topics for calculus of one variable generalize to calculus of any number of variables
in which the functions can have values in m dimensional space and there is more than
one variable.
The notation, C
n
refers to the collection of ordered lists of n complex numbers. Since
every real number is also a complex number, this simply generalizes the usual notion
of R
n
, the collection of all ordered lists of n real numbers. In order to avoid worrying
about whether it is real or complex numbers which are being referred to, the symbol F
will be used. If it is not clear, always pick C.
Denition 3.0.5 Dene
F
n
{(x
1
, , x
n
) : x
j
F for j = 1, , n} .
(x
1
, , x
n
) = (y
1
, , y
n
) if and only if for all j = 1, , n, x
j
= y
j
. When
(x
1
, , x
n
) F
n
,
it is conventional to denote (x
1
, , x
n
) by the single bold face letter, x. The numbers,
x
j
are called the coordinates. The set
{(0, , 0, t, 0, , 0) : t F}
for t in the i
th
slot is called the i
th
coordinate axis. The point 0 (0, , 0) is called
the origin.
Thus (1, 2, 4i) F
3
and (2, 1, 4i) F
3
but (1, 2, 4i) = (2, 1, 4i) because, even though
the same numbers are involved, they dont match up. In particular, the rst entries are
not equal.
The geometric signicance of R
n
for n 3 has been encountered already in calculus
or in precalculus. Here is a short review. First consider the case when n = 1. Then
from the denition, R
1
= R. Recall that R is identied with the points of a line. Look
at the number line again. Observe that this amounts to identifying a point on this line
with a real number. In other words a real number determines where you are on this line.
Now suppose n = 2 and consider two lines which intersect each other at right angles as
shown in the following picture.
23
24 BASIC LINEAR ALGEBRA
2
6

(2, 6)
8
3
(8, 3)
Notice how you can identify a point shown in the plane with the ordered pair, (2, 6) .
You go to the right a distance of 2 and then up a distance of 6. Similarly, you can identify
another point in the plane with the ordered pair (8, 3) . Go to the left a distance of 8
and then up a distance of 3. The reason you go to the left is that there is a sign on the
eight. From this reasoning, every ordered pair determines a unique point in the plane.
Conversely, taking a point in the plane, you could draw two lines through the point,
one vertical and the other horizontal and determine unique points, x
1
on the horizontal
line in the above picture and x
2
on the vertical line in the above picture, such that
the point of interest is identied with the ordered pair, (x
1
, x
2
) . In short, points in the
plane can be identied with ordered pairs similar to the way that points on the real
line are identied with real numbers. Now suppose n = 3. As just explained, the rst
two coordinates determine a point in a plane. Letting the third component determine
how far up or down you go, depending on whether this number is positive or negative,
this determines a point in space. Thus, (1, 4, 5) would mean to determine the point
in the plane that goes with (1, 4) and then to go below this plane a distance of 5 to
obtain a unique point in space. You see that the ordered triples correspond to points in
space just as the ordered pairs correspond to points in a plane and single real numbers
correspond to points on a line.
You cant stop here and say that you are only interested in n 3. What if you were
interested in the motion of two objects? You would need three coordinates to describe
where the rst object is and you would need another three coordinates to describe
where the other object is located. Therefore, you would need to be considering R
6
. If
the two objects moved around, you would need a time coordinate as well. As another
example, consider a hot object which is cooling and suppose you want the temperature
of this object. How many coordinates would be needed? You would need one for the
temperature, three for the position of the point in the object and one more for the
time. Thus you would need to be considering R
5
. Many other examples can be given.
Sometimes n is very large. This is often the case in applications to business when they
are trying to maximize prot subject to constraints. It also occurs in numerical analysis
when people try to solve hard problems on a computer.
There are other ways to identify points in space with three numbers but the one
presented is the most basic. In this case, the coordinates are known as Cartesian
coordinates after Descartes
1
who invented this idea in the rst half of the seventeenth
century. I will often not bother to draw a distinction between the point in n dimensional
space and its Cartesian coordinates.
The geometric signicance of C
n
for n > 1 is not available because each copy of C
corresponds to the plane or R
2
.
1
Rene Descartes 1596-1650 is often credited with inventing analytic geometry although it seems
the ideas were actually known much earlier. He was interested in many dierent subjects, physiology,
chemistry, and physics being some of them. He also wrote a large book in which he tried to explain
the book of Genesis scientically. Descartes ended up dying in Sweden.
3.1. ALGEBRA IN F
N
, VECTOR SPACES 25
3.1 Algebra in F
n
, Vector Spaces
There are two algebraic operations done with elements of F
n
. One is addition and the
other is multiplication by numbers, called scalars. In the case of C
n
the scalars are
complex numbers while in the case of R
n
the only allowed scalars are real numbers.
Thus, the scalars always come from F in either case.
Denition 3.1.1 If x F
n
and a F, also called a scalar, then ax F
n
is
dened by
ax = a (x
1
, , x
n
) (ax
1
, , ax
n
) . (3.1)
This is known as scalar multiplication. If x, y F
n
then x +y F
n
and is dened by
x +y = (x
1
, , x
n
) + (y
1
, , y
n
)
(x
1
+y
1
, , x
n
+y
n
) (3.2)
the points in F
n
are also referred to as vectors.
With this denition, the algebraic properties satisfy the conclusions of the following
theorem. These conclusions are called the vector space axioms. Any time you have a
set and a eld of scalars satisfying the axioms of the following theorem, it is called a
vector space.
Theorem 3.1.2 For v, w F
n
and , scalars, (real numbers), the following
hold.
v +w = w+v, (3.3)
the commutative law of addition,
(v +w) +z = v+(w+z) , (3.4)
the associative law for addition,
v +0 = v, (3.5)
the existence of an additive identity,
v+(v) = 0, (3.6)
the existence of an additive inverse, Also
(v +w) = v+w, (3.7)
( +) v =v+v, (3.8)
(v) = (v) , (3.9)
1v = v. (3.10)
In the above 0 = (0, , 0).
You should verify these properties all hold. For example, consider 3.7
(v +w) = (v
1
+w
1
, , v
n
+w
n
)
= ((v
1
+w
1
) , , (v
n
+w
n
))
= (v
1
+w
1
, , v
n
+w
n
)
= (v
1
, , v
n
) + (w
1
, , w
n
)
= v +w.
As usual subtraction is dened as x y x+(y) .
3.2 Subspaces Spans And Bases
The concept of linear combination is fundamental in all of linear algebra.
Denition 3.2.1 Let {x
1
, , x
p
} be vectors in a vector space, Y having the
eld of scalars F. A linear combination is any expression of the form
p
i=1
c
i
x
i
where the c
i
are scalars. The set of all linear combinations of these vectors is called
span (x
1
, , x
n
) . If V Y, then V is called a subspace if whenever , are scalars
and u and v are vectors of V, it follows u + v V . That is, it is closed under
the algebraic operations of vector addition and scalar multiplication and is therefore, a
vector space. A linear combination of vectors is said to be trivial if all the scalars in
the linear combination equal zero. A set of vectors is said to be linearly independent if
the only linear combination of these vectors which equals the zero vector is the trivial
linear combination. Thus {x
1
, , x
n
} is called linearly independent if whenever
p
k=1
c
k
x
k
= 0
it follows that all the scalars, c
k
equal zero. A set of vectors, {x
1
, , x
p
} , is called
linearly dependent if it is not linearly independent. Thus the set of vectors is linearly
dependent if there exist scalars, c
i
, i = 1, , n, not all zero such that
p
k=1
c
k
x
k
= 0.
Lemma 3.2.2 A set of vectors {x
1
, , x
p
} is linearly independent if and only if
none of the vectors can be obtained as a linear combination of the others.
Proof: Suppose rst that {x
1
, , x
p
} is linearly independent. If
x
k
=
j=k
c
j
x
j
,
then
0 = 1x
k
+
j=k
(c
j
) x
j
,
a nontrivial linear combination, contrary to assumption. This shows that if the set is
linearly independent, then none of the vectors is a linear combination of the others.
Now suppose no vector is a linear combination of the others. Is {x
1
, , x
p
} linearly
independent? If it is not, there exist scalars, c
i
, not all zero such that
p
i=1
c
i
x
i
= 0.
Say c
k
= 0. Then you can solve for x
k
as
x
k
=
j=k
(c
j
) /c
k
x
j
contrary to assumption. This proves the lemma.
The following is called the exchange theorem.
Theorem 3.2.3 Let {x
1
, , x
r
} be a linearly independent set of vectors such
that each x
i
is in the span {y
1
, , y
s
} . Then r s.
3.2. SUBSPACES SPANS AND BASES 27
Proof: Dene span {y
1
, , y
s
} V, it follows there exist scalars, c
1
, , c
s
such
that
x
1
=
s
i=1
c
i
y
i
. (3.11)
Not all of these scalars can equal zero because if this were the case, it would follow
that x
1
= 0 and so {x
1
, , x
r
} would not be linearly independent. Indeed, if x
1
= 0,
1x
1
+
r
i=2
0x
i
= x
1
= 0 and so there would exist a nontrivial linear combination of
the vectors {x
1
, , x
r
} which equals zero.
Say c
k
= 0. Then solve (3.11) for y
k
and obtain
y
k
span
_
_
x
1
,
s-1 vectors here
..
y
1
, , y
k1
, y
k+1
, , y
s
_
_
.
Dene {z
1
, , z
s1
} by
{z
1
, , z
s1
} {y
1
, , y
k1
, y
k+1
, , y
s
}
Therefore, span {x
1
, z
1
, , z
s1
} = V because if v V, there exist constants c
1
, , c
s
such that
v =
s1
i=1
c
i
z
i
+c
s
y
k
.
Now replace the y
k
in the above with a linear combination of the vectors, {x
1
, z
1
, , z
s1
}
to obtain v span {x
1
, z
1
, , z
s1
} . The vector y
k
, in the list {y
1
, , y
s
} , has now
been replaced with the vector x
1
and the resulting modied list of vectors has the same
span as the original list of vectors, {y
1
, , y
s
} .
Now suppose that r > s and that span {x
1
, , x
l
, z
1
, , z
p
} = V where the
vectors, z
1
, , z
p
are each taken from the set, {y
1
, , y
s
} and l + p = s. This
has now been done for l = 1 above. Then since r > s, it follows that l s < r
and so l + 1 r. Therefore, x
l+1
is a vector not in the list, {x
1
, , x
l
} and since
span {x
1
, , x
l
, z
1
, , z
p
} = V there exist scalars, c
i
and d
j
such that
x
l+1
=
l
i=1
c
i
x
i
+
p
j=1
d
j
z
j
. (3.12)
Now not all the d
j
can equal zero because if this were so, it would follow that {x
1
, , x
r
}
would be a linearly dependent set because one of the vectors would equal a linear com-
bination of the others. Therefore, (3.12) can be solved for one of the z
i
, say z
k
, in terms
of x
l+1
and the other z
i
and just as in the above argument, replace that z
i
with x
l+1
to obtain
span
_
_
x
1
, x
l
, x
l+1
,
p-1 vectors here
..
z
1
, z
k1
, z
k+1
, , z
p
_
_
= V.
Continue this way, eventually obtaining
span (x
1
, , x
s
) = V.
But then x
r
span {x
1
, , x
s
} contrary to the assumption that {x
1
, , x
r
} is linearly
independent. Therefore, r s as claimed.
Here is another proof in case you didnt like the above proof.
Theorem 3.2.4 If
span (u
1
, , u
r
) span (v
1
, , v
s
) V
and {u
1
, , u
r
} are linearly independent, then r s.
Proof: Suppose r > s. Let E
p
denote a nite list of vectors of {v
1
, , v
s
} and
let |E
p
| denote the number of vectors in the list. Let F
p
denote the rst p vectors in
{u
1
, , u
r
}. In case p = 0, F
p
will denote the empty set. For 0 p s, let E
p
have
the property
span (F
p
, E
p
) = V
and |E
p
| is as small as possible for this to happen. I claim |E
p
| sp if E
p
is nonempty.
Here is why. For p = 0, it is obvious. Suppose true for some p < s. Then
u
p+1
span (F
p
, E
p
)
and so there are constants, c
1
, , c
p
and d
1
, , d
m
where m s p such that
u
p+1
=
p
i=1
c
i
u
i
+
m
j=1
d
i
z
j
for
{z
1
, , z
m
} {v
1
, , v
s
} .
Then not all the d
i
can equal zero because this would violate the linear independence
of the {u
1
, , u
r
} . Therefore, you can solve for one of the z
k
as a linear combination
of {u
1
, , u
p+1
} and the other z
j
. Thus you can change F
p
to F
p+1
and include one
fewer vector in E
p
. Thus |E
p+1
| m1 s p 1. This proves the claim.
Therefore, E
s
is empty and span (u
1
, , u
s
) = V. However, this gives a contradic-
tion because it would require
u
s+1
span (u
1
, , u
s
)
which violates the linear independence of these vectors. This proves the theorem.
Denition 3.2.5 A nite set of vectors, {x
1
, , x
r
} is a basis for a vector
space V if
span (x
1
, , x
r
) = V
and {x
1
, , x
r
} is linearly independent. Thus if v V there exist unique scalars,
v
1
, , v
r
such that v =
r
i=1
v
i
x
i
. These scalars are called the components of v with
respect to the basis {x
1
, , x
r
}.
Corollary 3.2.6 Let {x
1
, , x
r
} and {y
1
, , y
s
} be two bases
2
of F
n
. Then r =
s = n.
Proof: From the exchange theorem, r s and s r. Now note the vectors,
e
i
=
1 is in the i
th
slot
..
(0, , 0, 1, 0 , 0)
for i = 1, 2, , n are a basis for F
n
. This proves the corollary.
2
This is the plural form of basis. We could say basiss but it would involve an inordinate amount
of hissing as in The sixth shieks sixth sheep is sick. This is the reason that bases is used instead of
basiss.
3.2. SUBSPACES SPANS AND BASES 29
Lemma 3.2.7 Let {v
1
, , v
r
} be a set of vectors. Then V span (v
1
, , v
r
) is
a subspace.
Proof: Suppose , are two scalars and let
r
k=1
c
k
v
k
and
r
k=1
d
k
v
k
are two
elements of V. What about
k=1
c
k
v
k
+
r
k=1
d
k
v
k
?
Is it also in V ?
k=1
c
k
v
k
+
r
k=1
d
k
v
k
=
r
k=1
(c
k
+d
k
) v
k
V
so the answer is yes. This proves the lemma.
Denition 3.2.8 Let V be a vector space. Then dim(V ) read as the dimension
of V is the number of vectors in a basis.
Of course you should wonder right now whether an arbitrary subspace of a nite
dimensional vector space even has a basis. In fact it does and this is in the next theorem.
First, here is an interesting lemma.
Lemma 3.2.9 Suppose v / span (u
1
, , u
k
) and {u
1
, , u
k
} is linearly indepen-
dent. Then {u
1
, , u
k
, v} is also linearly independent.
Proof: Suppose
k
i=1
c
i
u
i
+ dv = 0. It is required to verify that each c
i
= 0 and
that d = 0. But if d = 0, then you can solve for v as a linear combination of the vectors,
{u
1
, , u
k
},
v =
k
i=1
_
c
i
d
_
u
i
contrary to assumption. Therefore, d = 0. But then
k
i=1
c
i
u
i
= 0 and the linear
independence of {u
1
, , u
k
} implies each c
i
= 0 also. This proves the lemma.
Theorem 3.2.10 Let V be a nonzero subspace of Y a nite dimensional vector
space having dimension n. Then V has a basis.
Proof: Let v
1
V where v
1
= 0. If span {v
1
} = V, stop. {v
1
} is a basis for V .
Otherwise, there exists v
2
V which is not in span {v
1
} . By Lemma 3.2.9 {v
1
, v
2
} is
a linearly independent set of vectors. If span {v
1
, v
2
} = V stop, {v
1
, v
2
} is a basis for
V. If span {v
1
, v
2
} = V, then there exists v
3
/ span {v
1
, v
2
} and {v
1
, v
2
, v
3
} is a larger
linearly independent set of vectors. Continuing this way, the process must stop before
n + 1 steps because if not, it would be possible to obtain n + 1 linearly independent
vectors contrary to the exchange theorem and the assumed dimension of Y . This proves
the theorem.
In words the following corollary states that any linearly independent set of vectors
can be enlarged to form a basis.
Corollary 3.2.11 Let V be a subspace of Y, a nite dimensional vector space of
dimension n and let {v
1
, , v
r
} be a linearly independent set of vectors in V . Then ei-
ther it is a basis for V or there exist vectors, v
r+1
, , v
s
such that {v
1
, , v
r
, v
r+1
, , v
s
}
is a basis for V.
Proof: This follows immediately from the proof of Theorem 3.2.10. You do exactly
the same argument except you start with {v
1
, , v
r
} rather than {v
1
}.
It is also true that any spanning set of vectors can be restricted to obtain a basis.
Theorem 3.2.12 Let V be a subspace of Y, a nite dimensional vector space of
dimension n and suppose span (u
1
, u
p
) = V where the u
i
are nonzero vectors. Then
there exist vectors, {v
1
, v
r
} such that {v
1
, v
r
} {u
1
, u
p
} and {v
1
, v
r
}
is a basis for V .
Proof: Let r be the smallest positive integer with the property that for some set,
{v
1
, v
r
} {u
1
, u
p
} ,
span (v
1
, v
r
) = V.
Then r p and it must be the case that {v
1
, v
r
} is linearly independent because if
it were not so, one of the vectors, say v
k
would be a linear combination of the others.
But then you could delete this vector from {v
1
, v
r
} and the resulting list of r 1
vectors would still span V contrary to the denition of r. This proves the theorem.
3.3 Linear Transformations
In calculus of many variables one studies functions of many variables and what is meant
by their derivatives or integrals, etc. The simplest kind of function of many variables is
a linear transformation. You have to begin with the simple things if you expect to make
sense of the harder things. The following is the denition of a linear transformation.
Denition 3.3.1 Let V and W be two nite dimensional vector spaces. A func-
tion, L which maps V to W is called a linear transformation and written as L L(V, W)
if for all scalars and , and vectors v, w,
L(v+w) = L(v) +L(w) .
An example of a linear transformation is familiar matrix multiplication, familiar if
you have had a linear algebra course. Let A = (a
ij
) be an m n matrix. Then an
example of a linear transformation L : F
n
F
m
is given by
(Lv)
i

n
j=1
a
ij
v
j
.
Here
v
_
_
_
v
1
.
.
.
v
n
_
_
_ F
n
.
In the general case, the space of linear transformations is itself a vector space. This
will be discussed next.
Denition 3.3.2 Given L, M L(V, W) dene a new element of L(V, W) ,
denoted by L +M according to the rule
(L +M) v Lv +Mv.
For a scalar and L L(V, W) , dene L L(V, W) by
L(v) (Lv) .
3.3. LINEAR TRANSFORMATIONS 31
You should verify that all the axioms of a vector space hold for L(V, W) with
the above denitions of vector addition and scalar multiplication. What about the
dimension of L(V, W)?
Before answering this question, here is a lemma.
Lemma 3.3.3 Let V and W be vector spaces and suppose {v
1
, , v
n
} is a basis
for V. Then if L : V W is given by Lv
k
= w
k
W and
L
_
n
k=1
a
k
v
k
_
k=1
a
k
Lv
k
=
n
k=1
a
k
w
k
then L is well dened and is in L(V, W) . Also, if L, M are two linear transformations
such that Lv
k
= Mv
k
for all k, then M = L.
Proof: L is well dened on V because, since {v
1
, , v
n
} is a basis, there is exactly
one way to write a given vector of V as a linear combination. Next, observe that L is
obviously linear from the denition. If L, M are equal on the basis, then if
n
k=1
a
k
v
k
is an arbitrary vector of V,
L
_
n
k=1
a
k
v
k
_
=
n
k=1
a
k
Lv
k
=
n
k=1
a
k
Mv
k
= M
_
n
k=1
a
k
v
k
_
and so L = M because they give the same result for every vector in V .
The message is that when you dene a linear transformation, it suces to tell what
it does to a basis.
Denition 3.3.4 The symbol,
ij
is dened as 1 if i = j and 0 if i = j.
Theorem 3.3.5 Let V and W be nite dimensional vector spaces of dimension
n and m respectively Then dim(L(V, W)) = mn.
Proof: Let two sets of bases be
{v
1
, , v
n
} and {w
1
, , w
m
}
for V and W respectively. Using Lemma 3.3.3, let w
i
v
j
L(V, W) be the linear
transformation dened on the basis, {v
1
, , v
n
}, by
w
i
v
k
(v
j
) w
i
jk
.
Note that to dene these special linear transformations, sometimes called dyadics, it is
necessary that {v
1
, , v
n
} be a basis since their denition requires giving the values
of the linear transformation on a basis.
Let L L(V, W). Since {w
1
, , w
m
} is a basis, there exist constants d
jr
such that
Lv
r
=
m
j=1
d
jr
w
j
Then from the above,
Lv
r
=
m
j=1
d
jr
w
j
=
m
j=1
n
k=1
d
jr
kr
w
j
=
m
j=1
n
k=1
d
jr
w
j
v
k
(v
r
)
which shows
L =
m
j=1
n
k=1
d
jk
w
j
v
k
because the two linear transformations agree on a basis. Since L is arbitrary this shows
{w
i
v
k
: i = 1, , m, k = 1, , n}
spans L(V, W).
If
i,k
d
ik
w
i
v
k
= 0,
then
0 =
i,k
d
ik
w
i
v
k
(v
l
) =
m
i=1
d
il
w
i
and so, since {w
1
, , w
m
} is a basis, d
il
= 0 for each i = 1, , m. Since l is arbitrary,
this shows d
il
= 0 for all i and l. Thus these linear transformations form a basis and
this shows the dimension of L(V, W) is mn as claimed.
Denition 3.3.6 Let V, W be nite dimensional vector spaces such that a basis
for V is
{v
1
, , v
n
}
and a basis for W is
{w
1
, , w
m
}.
Then as explained in Theorem 3.3.5, for L L(V, W) , there exist scalars l
ij
such that
L =
ij
l
ij
w
i
v
j
Consider a rectangular array of scalars such that the entry in the i
th
row and the j
th
column is l
ij
,
_
_
_
_
_
l
11
l
12
l
1n
l
21
l
22
l
2n
.
.
.
.
.
.
.
.
.
.
.
.
l
m1
l
m2
l
mn
_
_
_
_
_
This is called the matrix of the linear transformation with respect to the two bases. This
will typically be denoted by (l
ij
) . It is called a matrix and in this case the matrix is
mn because it has m rows and n columns.
Theorem 3.3.7 Let L L(V, W) and let (l
ij
) be the matrix of L with respect
to the two bases,
{v
1
, , v
n
} and {w
1
, , w
m
}.
of V and W respectively. Then for v V having components (x
1
, , x
n
) with respect
to the basis {v
1
, , v
n
}, the components of Lv with respect to the basis {w
1
, , w
m
}
are
_
_
j
l
1j
x
j
, ,
j
l
mj
x
j
_
_
Proof: From the denition of l
ij
,
Lv =
ij
l
ij
w
i
v
j
(v) =
ij
l
ij
w
i
v
j
_
k
x
k
v
k
_
=
ijk
l
ij
w
i
v
j
(v
k
) x
k
=
ijk
l
ij
w
i
jk
x
k
=
i
_
_
j
l
ij
x
j
_
_
w
i
Theorem 3.3.8 Let (V, {v
1
, , v
n
}) , (U, {u
1
, , u
m
}) , (W, {w
1
, , w
p
}) be
three vector spaces along with bases for each one. Let L L(V, U) and M L(U, W) .
Then ML L(V, W) and if (c
ij
) is the matrix of ML with respect to {v
1
, , v
n
} and
{w
1
, , w
p
} and (l
ij
) and (m
ij
) are the matrices of L and M respectively with respect
to the given bases, then
c
rj
=
m
s=1
m
rs
l
sj
.
Proof: First note that from the denition,
(w
i
u
j
) (u
k
v
l
) (v
r
) = (w
i
u
j
) u
k
lr
= w
i
jk
lr
and
w
i
v
l
jk
(v
r
) = w
i
jk
lr
which shows
(w
i
u
j
) (u
k
v
l
) = w
i
v
l
jk
(3.13)
Therefore,
ML =
_
rs
m
rs
w
r
u
s
_
_
_
ij
l
ij
u
i
v
j
_
_
=
rsij
m
rs
l
ij
(w
r
u
s
) (u
i
v
j
) =
rsij
m
rs
l
ij
w
r
v
j
is
=
rsj
m
rs
l
sj
w
r
v
j
=
rj
_
s
m
rs
l
sj
_
w
r
v
j
The relation 3.13 is a very important cancellation property which is used later as
well as in this theorem.
Theorem 3.3.9 Suppose (V, {v
1
, , v
n
}) is a vector space and a basis and
(V, {v
1
, , v
n
}) is the same vector space with a dierent basis. Suppose L L(V, V ) .
Let (l
ij
) be the matrix of L taken with respect to {v
1
, , v
n
} and let
_
l
ij
_
be the nn
matrix of L taken with respect to {v
1
, , v
n
}That is,
L =
ij
l
ij
v
i
v
j
, L =
rs
l
rs
v
r
v
s
.
Then there exist n n matrices (d
ij
) and
_
d
ij
_
satisfying
j
d
ij
d
jk
=
ik
such that
l
ij
=
rs
d
ir
l
rs
d
sj
Proof: First consider the identity map, id dened by id (v) = v with respect to the
two bases, {v
1
, , v
n
} and {v
1
, , v
n
}.
id =
tu
d
tu
v
t
v
u
, id =
ij
d
ij
v
i
v
j
(3.14)
Now it follows from 3.13
id = id id =
tuij
d
tu
d
ij
(v
t
v
u
)
_
v
i
v
j
_
=
tuij
d
tu
d
ij
iu
v
t
v
j
=
tij
d
ti
d
ij
v
t
v
j
=
tj
_
i
d
ti
d
ij
_
v
t
v
j
On the other hand,
id =
tj
tj
v
t
v
j
because id (v
k
) = v
k
and
tj
tj
v
t
v
j
(v
k
) =
tj
tj
v
jk
=
tk
v
t
= v
k
.
Therefore,
_
i
d
ti
d
ij
_
=
tj
.
Switching the order of the above products shows
_
i
d
ti
d
ij
_
=
tj
In terms of matrices, this says
_
d
ij
_
is the inverse matrix of (d
ij
) .
Now using 3.14 and the cancellation property 3.13,
L =
iu
l
iu
v
i
v
u
=
rs
l
rs
v
r
v
s
= id
rs
l
rs
v
r
v
s
id
=
ij
d
ij
v
i
v
rs
l
rs
v
r
v
tu
d
tu
v
t
v
u
=
ijturs
d
ij
l
rs
d
tu
_
v
i
v
j
_
(v
r
v
s
) (v
t
v
u
)
=
ijturs
d
ij
l
rs
d
tu
v
i
v
u
jr
st
=
iu
_
_
js
d
ij
l
js
d
su
_
_
v
i
v
u
and since the linear transformations, {v
i
v
u
} are linearly independent, this shows
l
iu
=
js
d
ij
l
js
d
su
as claimed. This proves the theorem.
Recall the following denition which is a review of important terminology about
matrices.
Denition 3.3.10 If A is an m n matrix and B is an n p matrix, A =
(A
ij
) , B = (B
ij
) , then if (AB)
ij
is the ij
th
entry of the product, then
(AB)
ij
=
k
A
ik
B
kj
An nn matrix, A is said to be invertible if there exists another nn matrix, denoted
by A
1
such that AA
1
= A
1
A = I where the ij
th
entry of I is
ij
. Recall also that
_
A
T
_
ij
A
ji
. This is called the transpose of A.
Theorem 3.3.11 The following are important properties of matrices.
1. IA = AI = A
2. (AB) C = A(BC)
3. A
1
is unique if it exists.
4. When the inverses exist, (AB)
1
= B
1
A
1
5. (AB)
T
= B
T
A
T
Proof: I will prove these things directly from the above denition but there are
more elegant ways to see these things in terms of composition of linear transformations
which is really what matrix multiplication corresponds to.
First, (IA)
ij

k

ik
A
kj
= A
ij
. The other order is similar.
Next consider the associative law of multiplication.
((AB) C)
ij

k
(AB)
ik
C
kj
=
r
A
ir
B
rk
C
kj
=
r
A
ir
k
B
rk
C
kj
=
r
A
ir
(BC)
rj
= (A(BC))
ij
Since the ij
th
entries are equal, the two matrices are equal.
Next consider the uniqueness of the inverse. If AB = BA = I, then using the
associative law,
B = IB =
_
A
1
A
_
B = A
1
(AB) = A
1
I = A
1
Thus if it acts like the inverse, it is the inverse.
Consider now the inverse of a product.
AB
_
B
1
A
1
_
= A
_
BB
1
_
A
1
= AIA
1
= I
Similarly,
_
B
1
A
1
_
AB = I. Hence from what was just shown, (AB)
1
exists and
equals B
1
A
1
.
Finally consider the statement about transposes.
_
(AB)
T
_
ij
(AB)
ji

k
A
jk
B
ki

k
_
B
T
_
ik
_
A
T
_
kj

_
B
T
A
T
_
ij
Since the ij
th
entries are the same, the two matrices are equal. This proves the theorem.
In terms of matrix multiplication, Theorem 3.3.9 says that if M

1
and M
2
are matrices
for the same linear transformation relative to two dierent bases, it follows there exists
an invertible matrix, S such that
M
1
= S
1
M
2
S
This is called a similarity transformation and is important in linear algebra but this is
as far as the theory will be developed here.
3.4 Block Multiplication Of Matrices
Consider the following problem
_
A B
C D
__
E F
G H
_
You know how to do this from the above denition of matrix mutiplication. You get
_
AE +BG AF +BH
CE +DG CF +DH
_
.
Now what if instead of numbers, the entries, A, B, C, D, E, F, G are matrices of a size
such that the multiplications and additions needed in the above formula all make sense.
Would the formula be true in this case? I will show below that this is true.
Suppose A is a matrix of the form
_
_
_
A
11
A
1m
.
.
.
.
.
.
.
.
.
A
r1
A
rm
_
_
_ (3.15)
where A
ij
is a s
i
p
j
matrix where s
i
does not depend on j and p
j
does not depend on
i. Such a matrix is called a block matrix, also a partitioned matrix. Let n =
j
p
j
and k =
i
s
i
so A is an k n matrix. What is Ax where x F
n
? From the process
of multiplying a matrix times a vector, the following lemma follows.
Lemma 3.4.1 Let A be an mn block matrix as in 3.15 and let x F
n
. Then Ax
is of the form
Ax =
_
_
_
j
A
1j
x
j
.
.
.
j
A
rj
x
j
_
_
_
where x =(x
1
, , x
m
)
T
and x
i
F
pi
.
Suppose also that B is a block matrix of the form
_
_
_
B
11
B
1p
.
.
.
.
.
.
.
.
.
B
r1
B
rp
_
_
_ (3.16)
and A is a block matrix of the form
_
_
_
A
11
A
1m
.
.
.
.
.
.
.
.
.
A
p1
A
pm
_
_
_ (3.17)
and that for all i, j, it makes sense to multiply B
is
A
sj
for all s {1, , m} and that
for each s, B
is
A
sj
is the same size so that it makes sense to write
s
B
is
A
sj
.
Theorem 3.4.2 Let B be a block matrix as in 3.16 and let A be a block matrix
as in 3.17 such that B
is
is conformable with A
sj
and each product, B
is
A
sj
is of the
same size so they can be added. Then BA is a block matrix such that the ij
th
block is
of the form
s
B
is
A
sj
. (3.18)
3.5. DETERMINANTS 37
Proof: Let B
is
be a q
i
p
s
matrix and A
sj
be a p
s
r
j
matrix. Also let x F
n
and let x = (x
1
, , x
m
)
T
and x
i
F
ri
so it makes sense to multiply A
sj
x
j
. Then from
the associative law of matrix multiplication and Lemma 3.4.1 applied twice,
_
_
_
_
_
_
B
11
B
1p
.
.
.
.
.
.
.
.
.
B
r1
B
rp
_
_
_
_
_
_
A
11
A
1m
.
.
.
.
.
.
.
.
.
A
p1
A
pm
_
_
_
_
_
_
_
_
_
x
1
.
.
.
x
m
_
_
_
=
_
_
_
B
11
B
1p
.
.
.
.
.
.
.
.
.
B
r1
B
rp
_
_
_
_
_
_
j
A
1j
x
j
.
.
.
j
A
rj
x
j
_
_
_
=
_
_
_
j
B
1s
A
sj
x
j
.
.
.
j
B
rs
A
sj
x
j
_
_
_ =
_
_
_
j
(
s
B
1s
A
sj
) x
j
.
.
.
j
(
s
B
rs
A
sj
) x
j
_
_
_
=
_
_
_
s
B
1s
A
s1

s
B
1s
A
sm
.
.
.
.
.
.
.
.
.
s
B
rs
A
s1

s
B
rs
A
sm
_
_
_
_
_
_
x
1
.
.
.
x
m
_
_
_
By Lemma 3.4.1, this shows that (BA) x equals the block matrix whose ij
th
entry is
given by 3.18 times x. Since x is an arbitrary vector in F
n
, This proves the theorem.
The message of this theorem is that you can formally multiply block matrices as
though the blocks were numbers. You just have to pay attention to the preservation of
order.
3.5 Determinants
3.5.1 The Determinant Of A Matrix
The following Lemma will be essential in the denition of the determinant.
Lemma 3.5.1 There exists a unique function, sgn
n
which maps each list of numbers
from {1, , n} to one of the three numbers, 0, 1, or 1 which also has the following
properties.
sgn
n
(1, , n) = 1 (3.19)
sgn
n
(i
1
, , p, , q, , i
n
) = sgn
n
(i
1
, , q, , p, , i
n
) (3.20)
In words, the second property states that if two of the numbers are switched, the value
of the function is multiplied by 1. Also, in the case where n > 1 and {i
1
, , i
n
} =
{1, , n} so that every number from {1, , n} appears in the ordered list, (i
1
, , i
n
) ,
sgn
n
(i
1
, , i
1
, n, i
+1
, , i
n
)
(1)
n
sgn
n1
(i
1
, , i
1
, i
+1
, , i
n
) (3.21)
where n = i
in the ordered list, (i

1
, , i
n
) .
Proof: To begin with, it is necessary to show the existence of such a function. This
is clearly true if n = 1. Dene sgn
1
(1) 1 and observe that it works. No switching
is possible. In the case where n = 2, it is also clearly true. Let sgn
2
(1, 2) = 1 and
sgn
2
(2, 1) = 1 while sgn
2
(2, 2) = sgn
2
(1, 1) = 0 and verify it works. Assuming such a
function exists for n, sgn
n+1
will be dened in terms of sgn
n
. If there are any repeated
numbers in (i
1
, , i
n+1
) , sgn
n+1
(i
1
, , i
n+1
) 0. If there are no repeats, then n+1
appears somewhere in the ordered list. Let be the position of the number n+1 in the
list. Thus, the list is of the form (i
1
, , i
1
, n + 1, i
+1
, , i
n+1
) . From 3.21 it must
be that
sgn
n+1
(i
1
, , i
1
, n + 1, i
+1
, , i
n+1
)
(1)
n+1
sgn
n
(i
1
, , i
1
, i
+1
, , i
n+1
) .
It is necessary to verify this satises 3.19 and 3.20 with n replaced with n+1. The rst
of these is obviously true because
sgn
n+1
(1, , n, n + 1) (1)
n+1(n+1)
sgn
n
(1, , n) = 1.
If there are repeated numbers in (i
1
, , i
n+1
) , then it is obvious 3.20 holds because
both sides would equal zero from the above denition. It remains to verify 3.20 in the
case where there are no numbers repeated in (i
1
, , i
n+1
) . Consider
sgn
n+1
_
i
1
, ,
r
p, ,
s
q, , i
n+1
_
,
where the r above the p indicates the number, p is in the r
th
position and the s above
the q indicates that the number, q is in the s
th
position. Suppose rst that r < < s.
Then
sgn
n+1
_
i
1
, ,
r
p, ,
n + 1, ,
s
q, , i
n+1
_
(1)
n+1
sgn
n
_
i
1
, ,
r
p, ,
s1
q , , i
n+1
_
while
sgn
n+1
_
i
1
, ,
r
q, ,
n + 1, ,
s
p, , i
n+1
_
=
(1)
n+1
sgn
n
_
i
1
, ,
r
q, ,
s1
p , , i
n+1
_
and so, by induction, a switch of p and q introduces a minus sign in the result. Similarly,
if > s or if < r it also follows that 3.20 holds. The interesting case is when = r or
= s. Consider the case where = r and note the other case is entirely similar.
sgn
n+1
_
i
1
, ,
r
n + 1, ,
s
q, , i
n+1
_
=
(1)
n+1r
sgn
n
_
i
1
, ,
s1
q , , i
n+1
_
(3.22)
while
sgn
n+1
_
i
1
, ,
r
q, ,
s
n + 1, , i
n+1
_
=
(1)
n+1s
sgn
n
_
i
1
, ,
r
q, , i
n+1
_
. (3.23)
By making s 1 r switches, move the q which is in the s 1
th
position in 3.22 to the
r
th
position in 3.23. By induction, each of these switches introduces a factor of 1 and
so
sgn
n
_
i
1
, ,
s1
q , , i
n+1
_
= (1)
s1r
sgn
n
_
i
1
, ,
r
q, , i
n+1
_
.
Therefore,
sgn
n+1
_
i
1
, ,
r
n + 1, ,
s
q, , i
n+1
_
= (1)
n+1r
sgn
n
_
i
1
, ,
s1
q , , i
n+1
_
= (1)
n+1r
(1)
s1r
sgn
n
_
i
1
, ,
r
q, , i
n+1
_
= (1)
n+s
sgn
n
_
i
1
, ,
r
q, , i
n+1
_
= (1)
2s1
(1)
n+1s
sgn
n
_
i
1
, ,
r
q, , i
n+1
_
= sgn
n+1
_
i
1
, ,
r
q, ,
s
n + 1, , i
n+1
_
.
This proves the existence of the desired function.
To see this function is unique, note that you can obtain any ordered list of distinct
numbers from a sequence of switches. If there exist two functions, f and g both satisfying
3.19 and 3.20, you could start with f (1, , n) = g (1, , n) and applying the same
sequence of switches, eventually arrive at f (i
1
, , i
n
) = g (i
1
, , i
n
) . If any numbers
are repeated, then 3.20 gives both functions are equal to zero for that ordered list. This
proves the lemma.
In what follows sgn will often be used rather than sgn
n
because the context supplies
the appropriate n.
Denition 3.5.2 Let f be a real valued function which has the set of ordered
lists of numbers from {1, , n} as its domain. Dene
(k1, ,kn)
f (k
1
k
n
)
to be the sum of all the f (k
1
k
n
) for all possible choices of ordered lists (k
1
, , k
n
)
of numbers of {1, , n} . For example,
(k1,k2)
f (k
1
, k
2
) = f (1, 2) +f (2, 1) +f (1, 1) +f (2, 2) .
Denition 3.5.3 Let (a
ij
) = A denote an n n matrix. The determinant of
A, denoted by det (A) is dened by
det (A)
(k1, ,kn)
sgn (k
1
, , k
n
) a
1k1
a
nkn
where the sum is taken over all ordered lists of numbers from {1, , n}. Note it suces
to take the sum over only those ordered lists in which there are no repeats because if
there are, sgn (k
1
, , k
n
) = 0 and so that term contributes 0 to the sum.
Let A be an n n matrix, A = (a
ij
) and let (r
1
, , r
n
) denote an ordered list of n
numbers from {1, , n}. Let A(r
1
, , r
n
) denote the matrix whose k
th
row is the r
k
row of the matrix, A. Thus
det (A(r
1
, , r
n
)) =
(k1, ,kn)
sgn (k
1
, , k
n
) a
r1k1
a
rnkn
(3.24)
and
A(1, , n) = A.
Proposition 3.5.4 Let
(r
1
, , r
n
)
be an ordered list of numbers from {1, , n}. Then
sgn (r
1
, , r
n
) det (A)
=
(k1, ,kn)
sgn (k
1
, , k
n
) a
r1k1
a
rnkn
(3.25)
= det (A(r
1
, , r
n
)) . (3.26)
Proof: Let (1, , n) = (1, , r, s, , n) so r < s.
det (A(1, , r, , s, , n)) = (3.27)
(k1, ,kn)
sgn (k
1
, , k
r
, , k
s
, , k
n
) a
1k1
a
rkr
a
sks
a
nkn
,
and renaming the variables, calling k
s
, k
r
and k
r
, k
s
, this equals
=
(k1, ,kn)
sgn (k
1
, , k
s
, , k
r
, , k
n
) a
1k1
a
rks
a
skr
a
nkn
=
(k1, ,kn)
sgn
_
_
k
1
, ,
These got switched
..
k
r
, , k
s
, , k
n
_
_
a
1k1
a
skr
a
rks
a
nkn
= det (A(1, , s, , r, , n)) . (3.28)
Consequently,
det (A(1, , s, , r, , n)) =
det (A(1, , r, , s, , n)) = det (A)
Now letting A(1, , s, , r, , n) play the role of A, and continuing in this way,
switching pairs of numbers,
det (A(r
1
, , r
n
)) = (1)
p
det (A)
where it took p switches to obtain(r
1
, , r
n
) from (1, , n). By Lemma 3.5.1, this
implies
det (A(r
1
, , r
n
)) = (1)
p
det (A) = sgn (r
1
, , r
n
) det (A)
and proves the proposition in the case when there are no repeated numbers in the ordered
list, (r
1
, , r
n
). However, if there is a repeat, say the r
th
row equals the s
th
row, then
the reasoning of 3.27 -3.28 shows that A(r
1
, , r
n
) = 0 and also sgn (r
1
, , r
n
) = 0
so the formula holds in this case also.
Observation 3.5.5 There are n! ordered lists of distinct numbers from {1, , n} .
With the above, it is possible to give a more symmetric description of the determinant
from which it will follow that det (A) = det
_
A
T
_
.
Corollary 3.5.6 The following formula for det (A) is valid.
det (A) =
1
n!
(r1, ,rn)
(k1, ,kn)
sgn (r
1
, , r
n
) sgn (k
1
, , k
n
) a
r1k1
a
rnkn
. (3.29)
And also det
_
A
T
_
= det (A) where A
T
is the transpose of A. (Recall that for A
T
=
_
a
T
ij
_
, a
T
ij
= a
ji
.)
Proof: From Proposition 3.5.4, if the r
i
are distinct,
det (A) =
(k1, ,kn)
sgn (r
1
, , r
n
) sgn (k
1
, , k
n
) a
r1k1
a
rnkn
.
Summing over all ordered lists, (r
1
, , r
n
) where the r
i
are distinct, (If the r
i
are not
distinct, sgn (r
1
, , r
n
) = 0 and so there is no contribution to the sum.)
n! det (A) =
(r1, ,rn)
(k1, ,kn)
sgn (r
1
, , r
n
) sgn (k
1
, , k
n
) a
r1k1
a
rnkn
.
This proves the corollary. since the formula gives the same number for A as it does
for A
T
.
Corollary 3.5.7 If two rows or two columns in an nn matrix, A, are switched, the
determinant of the resulting matrix equals (1) times the determinant of the original
matrix. If A is an n n matrix in which two rows are equal or two columns are equal
then det (A) = 0. Suppose the i
th
row of A equals (xa
1
+yb
1
, , xa
n
+yb
n
). Then
det (A) = xdet (A
1
) +y det (A
2
)
where the i
th
row of A
1
is (a
1
, , a
n
) and the i
th
row of A
2
is (b
1
, , b
n
) , all other
rows of A
1
and A
2
coinciding with those of A. In other words, det is a linear function
of each row A. The same is true with the word row replaced with the word column.
Proof: By Proposition 3.5.4 when two rows are switched, the determinant of the
resulting matrix is (1) times the determinant of the original matrix. By Corollary
3.5.6 the same holds for columns because the columns of the matrix equal the rows of
the transposed matrix. Thus if A
1
is the matrix obtained from A by switching two
columns,
det (A) = det
_
A
T
_
= det
_
A
T
1
_
= det (A
1
) .
If A has two equal columns or two equal rows, then switching them results in the same
matrix. Therefore, det (A) = det (A) and so det (A) = 0.
It remains to verify the last assertion.
det (A)
(k1, ,kn)
sgn (k
1
, , k
n
) a
1k1
(xa
ki
+yb
ki
) a
nkn
= x
(k1, ,kn)
sgn (k
1
, , k
n
) a
1k1
a
ki
a
nkn
+y
(k1, ,kn)
sgn (k
1
, , k
n
) a
1k1
b
ki
a
nkn
xdet (A
1
) +y det (A
2
) .
The same is true of columns because det
_
A
T
_
= det (A) and the rows of A
T
are the
columns of A.
The following corollary is also of great use.
Corollary 3.5.8 Suppose A is an n n matrix and some column (row) is a linear
combination of r other columns (rows). Then det (A) = 0.
Proof: Let A =
_
a
1
a
n
_
be the columns of A and suppose the condition
that one column is a linear combination of r of the others is satised. Then by us-
ing Corollary 3.5.7 you may rearrange the columns to have the n
th
column a linear
combination of the rst r columns. Thus a
n
=
r
k=1
c
k
a
k
and so
det (A) = det
_
a
1
a
r
a
n1
r
k=1
c
k
a
k
_
.
By Corollary 3.5.7
det (A) =
r
k=1
c
k
det
_
a
1
a
r
a
n1
a
k
_
= 0.
The case for rows follows from the fact that det (A) = det
_
A
T
_
. This proves the corol-
lary.
Recall the following denition of matrix multiplication.
Denition 3.5.9 If A and B are n n matrices, A = (a
ij
) and B = (b
ij
),
AB = (c
ij
) where
c
ij

n
k=1
a
ik
b
kj
.
One of the most important rules about determinants is that the determinant of a
product equals the product of the determinants.
Theorem 3.5.10 Let A and B be n n matrices. Then
det (AB) = det (A) det (B) .
Proof: Let c
ij
be the ij
th
entry of AB. Then by Proposition 3.5.4,
det (AB) =
(k1, ,kn)
sgn (k
1
, , k
n
) c
1k1
c
nkn
=
(k1, ,kn)
sgn (k
1
, , k
n
)
_
r1
a
1r1
b
r1k1
_

_
rn
a
nrn
b
rnkn
_
=
(r1 ,rn)
(k1, ,kn)
sgn (k
1
, , k
n
) b
r1k1
b
rnkn
(a
1r1
a
nrn
)
=
(r1 ,rn)
sgn (r
1
r
n
) a
1r1
a
nrn
det (B) = det (A) det (B) .
This proves the theorem.
Lemma 3.5.11 Suppose a matrix is of the form
M =
_
A
0 a
_
(3.30)
or
M =
_
A 0
a
_
(3.31)
where a is a number and A is an (n 1)(n 1) matrix and denotes either a column
or a row having length n1 and the 0 denotes either a column or a row of length n1
consisting entirely of zeros. Then
det (M) = a det (A) .
Proof: Denote M by (m
ij
) . Thus in the rst case, m
nn
= a and m
ni
= 0 if i = n
while in the second case, m
nn
= a and m
in
= 0 if i = n. From the denition of the
determinant,
det (M)
(k1, ,kn)
sgn
n
(k
1
, , k
n
) m
1k1
m
nkn
Letting denote the position of n in the ordered list, (k
1
, , k
n
) then using the earlier
conventions used to prove Lemma 3.5.1, det (M) equals
(k1, ,kn)
(1)
n
sgn
n1
_
k
1
, , k
1
,
k
+1
, ,
n1
k
n
_
m
1k1
m
nkn
Now suppose 3.31. Then if k
n
= n, the term involving m
nkn
in the above expression
equals zero. Therefore, the only terms which survive are those for which = n or in
other words, those for which k
n
= n. Therefore, the above expression reduces to
a
(k1, ,kn1)
sgn
n1
(k
1
, k
n1
) m
1k1
m
(n1)kn1
= a det (A) .
To get the assertion in the situation of 3.30 use Corollary 3.5.6 and 3.31 to write
det (M) = det
_
M
T
_
= det
__
A
T
0
a
__
= a det
_
A
T
_
= a det (A) .
This proves the lemma.
In terms of the theory of determinants, arguably the most important idea is that of
Laplace expansion along a row or a column. This will follow from the above denition
of a determinant.
Denition 3.5.12 Let A = (a
ij
) be an nn matrix. Then a new matrix called
the cofactor matrix, cof (A) is dened by cof (A) = (c
ij
) where to obtain c
ij
delete the
i
th
row and the j
th
column of A, take the determinant of the (n 1) (n 1) matrix
which results, (This is called the ij
th
minor of A. ) and then multiply this number by
(1)
i+j
. To make the formulas easier to remember, cof (A)
ij
will denote the ij
th
entry
of the cofactor matrix.
The following is the main result.
Theorem 3.5.13 Let A be an n n matrix where n 2. Then
det (A) =
n
j=1
a
ij
cof (A)
ij
=
n
i=1
a
ij
cof (A)
ij
. (3.32)
The rst formula consists of expanding the determinant along the i
th
row and the second
expands the determinant along the j
th
column.
Proof: Let (a
i1
, , a
in
) be the i
th
row of A. Let B
j
be the matrix obtained from A
by leaving every row the same except the i
th
row which in B
j
equals (0, , 0, a
ij
, 0, , 0) .
Then by Corollary 3.5.7,
det (A) =
n
j=1
det (B
j
)
Denote by A
ij
the (n 1) (n 1) matrix obtained by deleting the i
th
row and the
j
th
column of A. Thus cof (A)
ij
(1)
i+j
det
_
A
ij
_
. At this point, recall that from
Proposition 3.5.4, when two rows or two columns in a matrix, M, are switched, this
results in multiplying the determinant of the old matrix by 1 to get the determinant
of the new matrix. Therefore, by Lemma 3.5.11,
det (B
j
) = (1)
nj
(1)
ni
det
__
A
ij
0 a
ij
__
= (1)
i+j
det
__
A
ij
0 a
ij
__
= a
ij
cof (A)
ij
.
Therefore,
det (A) =
n
j=1
a
ij
cof (A)
ij
which is the formula for expanding det (A) along the i
th
row. Also,
det (A) = det
_
A
T
_
=
n
j=1
a
T
ij
cof
_
A
T
_
ij
=
n
j=1
a
ji
cof (A)
ji
which is the formula for expanding det (A) along the i
th
column. This proves the
theorem.
Note that this gives an easy way to write a formula for the inverse of an nn matrix.
Theorem 3.5.14 A
1
exists if and only if det(A) = 0. If det(A) = 0, then
A
1
=
_
a
1
ij
_
where
a
1
ij
= det(A)
1
cof (A)
ji
for cof (A)
ij
the ij
th
cofactor of A.
Proof: By Theorem 3.5.13 and letting (a
ir
) = A, if det (A) = 0,
n
i=1
a
ir
cof (A)
ir
det(A)
1
= det(A) det(A)
1
= 1.
Now consider
n
i=1
a
ir
cof (A)
ik
det(A)
1
when k = r. Replace the k
th
column with the r
th
column to obtain a matrix, B
k
whose
determinant equals zero by Corollary 3.5.7. However, expanding this matrix along the
k
th
column yields
0 = det (B
k
) det (A)
1
=
n
i=1
a
ir
cof (A)
ik
det (A)
1
Summarizing,
n
i=1
a
ir
cof (A)
ik
det (A)
1
=
rk
.
Using the other formula in Theorem 3.5.13, and similar reasoning,
n
j=1
a
rj
cof (A)
kj
det (A)
1
=
rk
This proves that if det (A) = 0, then A
1
exists with A
1
=
_
a
1
ij
_
, where
a
1
ij
= cof (A)
ji
det (A)
1
.
Now suppose A
1
exists. Then by Theorem 3.5.10,
1 = det (I) = det
_
AA
1
_
= det (A) det
_
A
1
_
so det (A) = 0. This proves the theorem.
The next corollary points out that if an nn matrix, A has a right or a left inverse,
then it has an inverse.
Corollary 3.5.15 Let A be an nn matrix and suppose there exists an nn matrix,
B such that BA = I. Then A
1
exists and A
1
= B. Also, if there exists C an n n
matrix such that AC = I, then A
1
exists and A
1
= C.
Proof: Since BA = I, Theorem 3.5.10 implies
det Bdet A = 1
and so det A = 0. Therefore from Theorem 3.5.14, A
1
exists. Therefore,
A
1
= (BA) A
1
= B
_
AA
1
_
= BI = B.
The case where CA = I is handled similarly.
The conclusion of this corollary is that left inverses, right inverses and inverses are
all the same in the context of n n matrices.
Theorem 3.5.14 says that to nd the inverse, take the transpose of the cofactor
matrix and divide by the determinant. The transpose of the cofactor matrix is called
the adjugate or sometimes the classical adjoint of the matrix A. It is an abomination
to call it the adjoint although you do sometimes see it referred to in this way. In words,
A
1
is equal to one over the determinant of A times the adjugate matrix of A.
In case you are solving a system of equations, Ax = y for x, it follows that if A
1
exists,
x =
_
A
1
A
_
x = A
1
(Ax) = A
1
y
thus solving the system. Now in the case that A
1
exists, there is a formula for A
1
given above. Using this formula,
x
i
=
n
j=1
a
1
ij
y
j
=
n
j=1
1
det (A)
cof (A)
ji
y
j
.
By the formula for the expansion of a determinant along a column,
x
i
=
1
det (A)
det
_
_
_
y
1

.
.
.
.
.
.
.
.
.
y
n

_
_
_,
where here the i
th
column of A is replaced with the column vector, (y
1
, y
n
)
T
, and
the determinant of this modied matrix is taken and divided by det (A). This formula
is known as Cramers rule.
Denition 3.5.16 A matrix M, is upper triangular if M
ij
= 0 whenever i > j.
Thus such a matrix equals zero below the main diagonal, the entries of the form M
ii
as
shown.
_
_
_
_
_
_

0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0
_
_
_
_
_
_
A lower triangular matrix is dened similarly as a matrix for which all entries above
the main diagonal are equal to zero.
With this denition, here is a simple corollary of Theorem 3.5.13.
Corollary 3.5.17 Let M be an upper (lower) triangular matrix. Then det (M) is
obtained by taking the product of the entries on the main diagonal.
Denition 3.5.18 A submatrix of a matrix A is the rectangular array of num-
bers obtained by deleting some rows and columns of A. Let A be an mn matrix. The
determinant rank of the matrix equals r where r is the largest number such that some
r r submatrix of A has a non zero determinant. The row rank is dened to be the
dimension of the span of the rows. The column rank is dened to be the dimension of
the span of the columns.
Theorem 3.5.19 If A has determinant rank, r, then there exist r rows of the
matrix such that every other row is a linear combination of these r rows.
Proof: Suppose the determinant rank of A = (a
ij
) equals r. If rows and columns
are interchanged, the determinant rank of the modied matrix is unchanged. Thus rows
and columns can be interchanged to produce an r r matrix in the upper left corner of
the matrix which has non zero determinant. Now consider the r +1 r +1 matrix, M,
_
_
_
_
_
a
11
a
1r
a
1p
.
.
.
.
.
.
.
.
.
a
r1
a
rr
a
rp
a
l1
a
lr
a
lp
_
_
_
_
_
where C will denote the r r matrix in the upper left corner which has non zero
determinant. I claim det (M) = 0.
There are two cases to consider in verifying this claim. First, suppose p > r. Then
the claim follows from the assumption that A has determinant rank r. On the other
hand, if p < r, then the determinant is zero because there are two identical columns.
Expand the determinant along the last column and divide by det (C) to obtain
a
lp
=
r
i=1
cof (M)
ip
det (C)
a
ip
.
Now note that cof (M)
ip
does not depend on p. Therefore the above sum is of the form
a
lp
=
r
i=1
m
i
a
ip
which shows the l
th
row is a linear combination of the rst r rows of A. Since l is
arbitrary, This proves the theorem.
Corollary 3.5.20 The determinant rank equals the row rank.
Proof: From Theorem 3.5.19, the row rank is no larger than the determinant rank.
Could the row rank be smaller than the determinant rank? If so, there exist p rows for
p < r such that the span of these p rows equals the row space. But this implies that
the r r submatrix whose determinant is nonzero also has row rank no larger than p
which is impossible if its determinant is to be nonzero because at least one row is a
linear combination of the others.
Corollary 3.5.21 If A has determinant rank, r, then there exist r columns of the
matrix such that every other column is a linear combination of these r columns. Also
the column rank equals the determinant rank.
Proof: This follows from the above by considering A
T
. The rows of A
T
are the
columns of A and the determinant rank of A
T
and A are the same. Therefore, from
Corollary 3.5.20, column rank of A = row rank of A
T
= determinant rank of A
T
=
determinant rank of A.
The following theorem is of fundamental importance and ties together many of the
ideas presented above.
Theorem 3.5.22 Let A be an nn matrix. Then the following are equivalent.
1. det (A) = 0.
2. A, A
T
are not one to one.
3. A is not onto.
Proof: Suppose det (A) = 0. Then the determinant rank of A = r < n. Therefore,
there exist r columns such that every other column is a linear combination of these
columns by Theorem 3.5.19. In particular, it follows that for some m, the m
th
column
is a linear combination of all the others. Thus letting A =
_
a
1
a
m
a
n
_
where the columns are denoted by a
i
, there exists scalars,
i
such that
a
m
=
k=m
k
a
k
.
Now consider the column vector, x
_

1
1
n
_
T
. Then
Ax = a
m
+
k=m
k
a
k
= 0.
Since also A0 = 0, it follows A is not one to one. Similarly, A
T
is not one to one by the
same argument applied to A
T
. This veries that 1.) implies 2.).
Now suppose 2.). Then since A
T
is not one to one, it follows there exists x = 0 such
that
A
T
x = 0.
Taking the transpose of both sides yields
x
T
A = 0
where the 0 is a 1 n matrix or row vector. Now if Ay = x, then
|x|
2
= x
T
(Ay) =
_
x
T
A
_
y = 0y = 0
contrary to x = 0. Consequently there can be no y such that Ay = x and so A is not
onto. This shows that 2.) implies 3.).
Finally, suppose 3.). If 1.) does not hold, then det (A) = 0 but then from Theorem
3.5.14 A
1
exists and so for every y F
n
there exists a unique x F
n
such that Ax = y.
In fact x = A
1
y. Thus A would be onto contrary to 3.). This shows 3.) implies 1.)
and proves the theorem.
Corollary 3.5.23 Let A be an n n matrix. Then the following are equivalent.
1. det(A) = 0.
2. A and A
T
are one to one.
3. A is onto.
Proof: This follows immediately from the above theorem.
3.5.2 The Determinant Of A Linear Transformation
One can also dene the determinant of a linear transformation.
Denition 3.5.24 Let L L(V, V ) and let {v
1
, , v
n
} be a basis for V .
Thus the matrix of L with respect to this basis is (l
ij
) M
L
where
L =
ij
l
ij
v
i
v
j
Then dene
det (L) det ((l
ij
)) .
Proposition 3.5.25 The above denition is well dened.
Proof: Suppose {v
1
, , v
n
} is another basis for V and
_
l
ij
_
M
L
is the matrix
of L with respect to this basis. Then by Theorem 3.3.9,
M
L
= S
1
M
L
S
for some matrix, S. Then by Theorem 3.5.10,
det (M
L
) = det
_
S
1
_
det (M
L
) det (S)
= det
_
S
1
S
_
det (M
L
) = det (M
L
)
because S
1
S = I and det (I) = 1. This shows the denition is well dened.
Also there is an equivalence just as in the case of matrices between various properties
of L and the nonvanishing of the determinant.
Theorem 3.5.26 Let L L(V, V ) for V a nite dimensional vector space.
Then the following are equivalent.
1. det (L) = 0.
2. L is not one to one.
3. L is not onto.
3.6. EIGENVALUES AND EIGENVECTORS OF LINEAR TRANSFORMATIONS 49
Proof: Suppose 1.). Let {v
1
, , v
n
} be a basis for V and let (l
ij
) be the matrix
of L with respect to this basis. By denition, det ((l
ij
)) = 0 and so (l
ij
) is not one to
one. Thus there is a nonzero vector x F
n
such that
j
l
ij
x
j
= 0 for each i. Then
letting v
n
j=1
x
j
v
j
,
Lv =
rs
l
rs
v
r
v
s
_
_
n
j=1
x
j
v
j
_
_
=
rs
l
rs
v
r
sj
x
j
=
r
_
_
j
l
rj
x
j
_
_
v
r
= 0
Thus L is not one to one because L0 = 0 and Lv = 0.
Suppose 2.). Thus there exists v = 0 such that Lv = 0. Say
v =
i
x
i
v
i
.
Then if {Lv
i
}
n
i=1
were linearly independent, it would follow that
0 = Lv =
i
x
i
Lv
i
and so all the x
i
would equal zero which is not the case. Hence these vectors cannot be
linearly independent so they do not span V . Hence there exists
w V \ span (Lv
1
, ,Lv
n
)
and therefore, there is no u V such that Lu = w because if there were such a u, then
u =
i
x
i
v
i
and so Lu =
i
x
i
Lv
i
span (Lv
1
, ,Lv
n
) .
Finally suppose L is not onto. Then (l
ij
) also cannot be onto F
n
. Therefore,
det ((l
ij
)) det (L) = 0. Why cant (l
ij
) be onto? If it were, then for any y F
n
,
there exists x F
n
such that y
i
=
j
l
ij
x
j
. Thus
k
y
k
v
k
=
rs
l
rs
v
r
v
s
_
_
j
x
j
v
j
_
_
= L
_
_
j
x
j
v
j
_
_
but the expression on the left in the above formula is that of a general element of V
and so L would be onto. This proves the theorem.
3.6 Eigenvalues And Eigenvectors Of Linear Trans-
formations
Let V be a nite dimensional vector space. For example, it could be a subspace of C
n
or
R
n
. Also suppose A L(V, V ) .
Denition 3.6.1 The characteristic polynomial of A is dened as q () det (id A)
where id is the identity map which takes every vector in V to itself. The zeros of q ()
in C are called the eigenvalues of A.
Lemma 3.6.2 When is an eigenvalue of A which is also in F, the eld of scalars,
then there exists v = 0 such that Av = v.
Proof: This follows from Theorem 3.5.26. Since F,
id A L(V, V )
and since it has zero determinant, it is not one to one so there exists v = 0 such that
(id A) v = 0.
The following lemma gives the existence of something called the minimal polynomial.
It is an interesting application of the notion of the dimension of L(V, V ).
Lemma 3.6.3 Let A L(V, V ) where V is either a real or a complex nite dimen-
sional vector space of dimension n. Then there exists a polynomial of the form
p () =
m
+c
m1
m1
+ +c
1
+c
0
such that p (A) = 0 and m is as small as possible for this to occur.
Proof: Consider the linear transformations, I, A, A
2
, , A
n
2
. There are n
2
+ 1 of
these transformations and so by Theorem 3.3.5 the set is linearly dependent. Thus there
exist constants, c
i
F (either R or C) such that
c
0
I +
n
2
k=1
c
k
A
k
= 0.
This implies there exists a polynomial, q () which has the property that q (A) = 0.
In fact, one example is q () c
0
+
n
2
k=1
c
k
k
. Dividing by the leading term, it can
be assumed this polynomial is of the form
m
+ c
m1
m1
+ + c
1
+ c
0
, a monic
polynomial. Now consider all such monic polynomials, q such that q (A) = 0 and pick
the one which has the smallest degree. This is called the minimial polynomial and will
be denoted here by p () . This proves the lemma.
Theorem 3.6.4 Let V be a nonzero nite dimensional vector space of dimension
n with the eld of scalars equal to F which is either R or C. Suppose A L(V, V ) and
for p () the minimal polynomial dened above, let F be a zero of this polynomial.
Then there exists v = 0, v V such that
Av = v.
If F = C, then A always has an eigenvector and eigenvalue. Furthermore, if {
1
, ,
m
}
are the zeros of p () in F, these are exactly the eigenvalues of A for which there exists
an eigenvector in V.
Proof: Suppose rst is a zero of p () . Since p () = 0, it follows
p () = ( ) k ()
where k () is a polynomial having coecients in F. Since p has minimal degree, k (A) =
0 and so there exists a vector, u = 0 such that k (A) u v = 0. But then
(AI) v = (AI) k (A) (u) = 0.
The next claim about the existence of an eigenvalue follows from the fundamental
theorem of algebra and what was just shown.
3.7. EXERCISES 51
It has been shown that every zero of p () is an eigenvalue which has an eigenvector
in V . Now suppose is an eigenvalue which has an eigenvector in V so that Av = v
for some v V, v = 0. Does it follow is a zero of p ()?
0 = p (A) v = p () v
and so is indeed a zero of p (). This proves the theorem.
In summary, the theorem says the eigenvalues which have eigenvectors in V are
exactly the zeros of the minimal polynomial which are in the eld of scalars, F.
The idea of block multiplication turns out to be very useful later. For now here is an
interesting and signicant application which has to do with characteristic polynomials.
In this theorem, p
M
(t) denotes the polynomial, det (tI M) . Thus the zeros of this
polynomial are the eigenvalues of the matrix, M.
Theorem 3.6.5 Let A be an m n matrix and let B be an n m matrix for
m n. Then
p
BA
(t) = t
nm
p
AB
(t) ,
so the eigenvalues of BA and AB are the same including multiplicities except that BA
has n m extra zero eigenvalues.
Proof: Use block multiplication to write
_
AB 0
B 0
__
I A
0 I
_
=
_
AB ABA
B BA
_
_
I A
0 I
__
0 0
B BA
_
=
_
AB ABA
B BA
_
.
Therefore,
_
I A
0 I
_
1
_
AB 0
B 0
__
I A
0 I
_
=
_
0 0
B BA
_
Since the two matrices above are similar it follows that
_
0 0
B BA
_
and
_
AB 0
B 0
_
have the same characteristic polynomials. Therefore, noting that BA is an nn matrix
and AB is an mm matrix,
t
m
det (tI BA) = t
n
det (tI AB)
and so det (tI BA) = p
BA
(t) = t
nm
det (tI AB) = t
nm
p
AB
(t) . This proves the
theorem.
3.7 Exercises
1. Let M be an n n matrix. Thus letting Mx be dened by ordinary matrix
multiplication, it follows M L(C
n
, C
n
) . Show that all the zeros of the mini-
mal polynomial are also zeros of the characteristic polynomial. Explain why this
requires the minimal polynomial to divide the characteristic polynomial. Thus
q () = p () k () for some polynomial k () where q () is the characteristic poly-
nomial. Now explain why q (M) = 0. That every n n matrix satises its char-
acteristic polynomial is the Cayley Hamilton theorem. Can you extend this to a
result about L L(V, V ) for V an n dimensional real or complex vector space?
2. Give examples of subspaces of R
n
and examples of subsets of R
n
which are not
subspaces.
3. Let L L(V, W) . Dene ker L {v V : Lv = 0} . Determine whether ker L is
a subspace.
4. Let L L(V, W) . Then L(V ) denotes those vectors in W such that for some
v,Lv = w. Show L(V ) is a subspace.
5. Let L L(V, W) and suppose {w
1
, , w
k
} are linearly independent and that
Lz
i
= w
i
. Show {z
1
, , z
k
} is also linearly independent.
6. If L L(V, W) and {z
1
, , z
k
} is linearly independent, what is needed in order
that {Lz
1
, , Lz
k
} be linearly independent? Explain your answer.
7. Let L L(V, W). The rank of L is dened as the dimension of L(V ) . The nullity
of L is the dimension of ker (L) . Show
dim(V ) = rank +nullity.
8. Let L L(V, W) and let M L(W, Y ) . Show
rank (ML) min (rank (L) , rank (M)) .
9. Let M (t) = (b
1
(t) , , b
n
(t)) where each b
k
(t) is a column vector whose com-
ponent functions are dierentiable functions. For such a column vector,
b(t) = (b
1
(t) , , b
n
(t))
T
,
dene
b
(t) (b
1
(t) , , b
n
(t))
T
Show
det (M (t))
=
n
i=1
det M
i
(t)
where M
i
(t) has all the same columns as M (t) except the i
th
column is replaced
with b
i
(t).
10. Let A = (a
ij
) be an n n matrix. Consider this as a linear transformation using
ordinary matrix multiplication. Show
A =
ij
a
ij
e
i
e
j
where e
i
is the vector which has a 1 in the i
th
place and zeros elsewhere.
11. Let {w
1
, , w
n
} be a basis for the vector space, V. Show id, the identity map is
given by
id =
ij
ij
w
i
w
j
3.8 Inner Product And Normed Linear Spaces
3.8.1 The Inner Product In F
n
To do calculus, you must understand what you mean by distance. For functions of
one variable, the distance was provided by the absolute value of the dierence of two
numbers. This must be generalized to F
n
and to more general situations.
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 53
Denition 3.8.1 Let x, y F
n
. Thus x = (x
1
, , x
n
) where each x
k
F and
a similar formula holding for y. Then the dot product of these two vectors is dened to
be
x y
j
x
j
y
j
x
1
y
1
+ +x
n
y
n
.
This is also often denoted by (x, y) and is called an inner product. I will use either
notation.
Notice how you put the conjugate on the entries of the vector, y. It makes no
dierence if the vectors happen to be real vectors but with complex vectors you must
do it this way. The reason for this is that when you take the dot product of a vector
with itself, you want to get the square of the length of the vector, a positive number.
Placing the conjugate on the components of y in the above denition assures this will
take place. Thus
x x =
j
x
j
x
j
=
j
|x
j
|
2
0.
If you didnt place a conjugate as in the above denition, things wouldnt work out
correctly. For example,
(1 +i)
2
+ 2
2
= 4 + 2i
and this is not a positive number.
The following properties of the dot product follow immediately from the denition
and you should verify each of them.
Properties of the dot product:
1. u v = v u.
2. If a, b are numbers and u, v, z are vectors then (au +bv) z = a (u z) +b (v z) .
3. u u 0 and it equals 0 if and only if u = 0.
Note this implies (xy) = (x y) because
(xy) = (y x) = (y x) = (x y)
The norm is dened as follows.
Denition 3.8.2 For x F
n
,
|x|
_
n
k=1
|x
k
|
2
_
1/2
= (x x)
1/2
3.8.2 General Inner Product Spaces
Any time you have a vector space which possesses an inner product, something satisfying
the properties 1 - 3 above, it is called an inner product space.
Here is a fundamental inequality called the Cauchy Schwarz inequality which
holds in any inner product space. First here is a simple lemma.
Lemma 3.8.3 If z F there exists F such that z = |z| and || = 1.
Proof: Let = 1 if z = 0 and otherwise, let =
z
|z|
. Recall that for z = x+iy, z =
x iy and zz = |z|
2
. In case z is real, there is no change in the above.
Theorem 3.8.4 (Cauchy Schwarz)Let H be an inner product space. The fol-
lowing inequality holds for x and y H.
|(x y)| (x x)
1/2
(y y)
1/2
(3.33)
Equality holds in this inequality if and only if one vector is a multiple of the other.
Proof: Let F such that || = 1 and
(x y) = |(x y)|
Consider p (t)
_
x +ty x +ty
_
where t R. Then from the above list of properties
of the dot product,
0 p (t) = (x x) +t (x y) +t (y x) +t
2
(y y)
= (x x) +t (x y) +t(x y) +t
2
(y y)
= (x x) + 2t Re ( (x y)) +t
2
(y y)
= (x x) + 2t |(x y)| +t
2
(y y) (3.34)
and this must hold for all t R. Therefore, if (y y) = 0 it must be the case that
|(x y)| = 0 also since otherwise the above inequality would be violated. Therefore, in
this case,
|(x y)| (x x)
1/2
(y y)
1/2
.
On the other hand, if (y y) = 0, then p (t) 0 for all t means the graph of y = p (t) is
a parabola which opens up and it either has exactly one real zero in the case its vertex
touches the t axis or it has no real zeros. From the quadratic formula this happens
exactly when
4 |(x y)|
2
4 (x x) (y y) 0
which is equivalent to 3.33.
It is clear from a computation that if one vector is a scalar multiple of the other that
equality holds in 3.33. Conversely, suppose equality does hold. Then this is equivalent
to saying 4 |(x y)|
2
4 (x x) (y y) = 0 and so from the quadratic formula, there exists
one real zero to p (t) = 0. Call it t
0
. Then
p (t
0
)
_
x +t
0
y, x +t
0
y
_
=
x +ty
2
= 0
and so x = t
0
y. This proves the theorem.
Note that in establishing the inequality, I only used part of the above properties of
the dot product. It was not necessary to use the one which says that if (x x) = 0 then
x = 0.
Now the length of a vector can be dened.
Denition 3.8.5 Let z H. Then |z| (z z)
1/2
.
Theorem 3.8.6 For length dened in Denition 3.8.5, the following hold.
|z| 0 and |z| = 0 if and only if z = 0 (3.35)
If is a scalar, |z| = || |z| (3.36)
|z +w| |z| +|w| . (3.37)
Proof: The rst two claims are left as exercises. To establish the third,
|z +w|
2
(z +w, z +w)
= z z +w w+w z +z w
= |z|
2
+|w|
2
+ 2 Re w z
|z|
2
+|w|
2
+ 2 |w z|
|z|
2
+|w|
2
+ 2 |w| |z| = (|z| +|w|)
2
.
3.8.3 Normed Vector Spaces
The best sort of a norm is one which comes from an inner product. However, any vector
space, V which has a function, |||| which maps V to [0, ) is called a normed vector
space if |||| satises 3.35 - 3.37. That is
||z|| 0 and ||z|| = 0 if and only if z = 0 (3.38)
If is a scalar, ||z|| = || ||z|| (3.39)
||z +w|| ||z|| +||w|| . (3.40)
The last inequality above is called the triangle inequality. Another version of this is
|||z|| ||w||| ||z w|| (3.41)
To see that 3.41 holds, note
||z|| = ||z w+w|| ||z w|| +||w||
which implies
||z|| ||w|| ||z w||
and now switching z and w, yields
||w|| ||z|| ||z w||
which implies 3.41.
3.8.4 The p Norms
Examples of norms are the p norms on C
n
.
Denition 3.8.7 Let x C
n
. Then dene for p 1,
||x||
p

_
n
i=1
|x
i
|
p
_
1/p
The following inequality is called Holders inequality.
Proposition 3.8.8 For x, y C
n
,
n
i=1
|x
i
| |y
i
|
_
n
i=1
|x
i
|
p
_
1/p
_
n
i=1
|y
i
|
p
_
1/p
The proof will depend on the following lemma.

Lemma 3.8.9 If a, b 0 and p
is dened by
1
p
+
1
p
= 1, then
ab
a
p
p
+
b
p
.
Proof of the Proposition: If x or y equals the zero vector there is nothing to
prove. Therefore, assume they are both nonzero. Let A = (
n
i=1
|x
i
|
p
)
1/p
and B =
_
n
i=1
|y
i
|
p
_
1/p
. Then using Lemma 3.8.9,

n
i=1
|x
i
|
A
|y
i
|
B

n
i=1
_
1
p
_
|x
i
|
A
_
p
+
1
p
_
|y
i
|
B
_
p
_
=
1
p
1
A
p
n
i=1
|x
i
|
p
+
1
p
1
B
p
n
i=1
|y
i
|
p
=
1
p
+
1
p
= 1
and so
n
i=1
|x
i
| |y
i
| AB =
_
n
i=1
|x
i
|
p
_
1/p
_
n
i=1
|y
i
|
p
_
1/p
.
This proves the proposition.
Theorem 3.8.10 The p norms do indeed satisfy the axioms of a norm.
Proof: It is obvious that ||||
p
does indeed satisfy most of the norm axioms. The
only one that is not clear is the triangle inequality. To save notation write |||| in place
of ||||
p
in what follows. Note also that
p
p
= p 1. Then using the Holder inequality,

||x +y||
p
=
n
i=1
|x
i
+y
i
|
p
i=1
|x
i
+y
i
|
p1
|x
i
| +
n
i=1
|x
i
+y
i
|
p1
|y
i
|
=
n
i=1
|x
i
+y
i
|
p
p
|x
i
| +
n
i=1
|x
i
+y
i
|
p
p
|y
i
|
_
n
i=1
|x
i
+y
i
|
p
_
1/p
_
_
_
n
i=1
|x
i
|
p
_
1/p
+
_
n
i=1
|y
i
|
p
_
1/p
_
_
= ||x +y||
p/p
_
||x||
p
+||y||
p
_
so dividing by ||x +y||
p/p
, it follows
||x +y||
p
||x +y||
p/p
= ||x +y|| ||x||

p
+||y||
p
_
p
p
p
= p
_
1
1
p
_
= p
1
p
= 1.
_
. This proves the theorem.
It only remains to prove Lemma 3.8.9.
Proof of the lemma: Let p
= q to save on notation and consider the following

picture:
b
a
x
t
x = t
p1
t = x
q1
ab
a
0
t
p1
dt +
b
0
x
q1
dx =
a
p
p
+
b
q
q
.
Note equality occurs when a
p
= b
q
.
Alternate proof of the lemma: Let
f (t)
1
p
(at)
p
+
1
q
_
b
t
_
q
, t > 0
You see right away it is decreasing for a while, having an assymptote at t = 0 and then
reaches a minimum and increases from then on. Take its derivative.
f
(t) = (at)
p1
a +
_
b
t
_
q1
_
b
t
2
_
Set it equal to 0. This happens when
t
p+q
=
b
q
a
p
. (3.42)
Thus
t =
b
q/(p+q)
a
p/(p+q)
and so at this value of t,
at = (ab)
q/(p+q)
,
_
b
t
_
= (ab)
p/(p+q)
.
Thus the minimum of f is
1
p
_
(ab)
q/(p+q)
_
p
+
1
q
_
(ab)
p/(p+q)
_
q
= (ab)
pq/(p+q)
but recall 1/p + 1/q = 1 and so pq/ (p +q) = 1. Thus the minimum value of f is ab.
Letting t = 1, this shows
ab
a
p
p
+
b
q
q
.
Note that equality occurs when the minimum value happens for t = 1 and this indicates
from 3.42 that a
p
= b
q
. This proves the lemma.
3.8.5 Orthonormal Bases
Not all bases for an inner product space H are created equal. The best bases are
orthonormal.
Denition 3.8.11 Suppose {v
1
, , v
k
} is a set of vectors in an inner product
space H. It is an orthonormal set if
v
i
v
j
=
ij
=
_
1 if i = j
0 if i = j
Every orthonormal set of vectors is automatically linearly independent.
Proposition 3.8.12 Suppose {v
1
, , v
k
} is an orthonormal set of vectors. Then
it is linearly independent.
Proof: Suppose
k
i=1
c
i
v
i
= 0. Then taking dot products with v
j
,
0 = 0 v
j
=
i
c
i
v
i
v
j
=
i
c
i
ij
= c
j
.
Since j is arbitrary, this shows the set is linearly independent as claimed.
It turns out that if X is any subspace of H, then there exists an orthonormal basis
for X.
Lemma 3.8.13 Let X be a subspace of dimension n whose basis is {x
1
, , x
n
} .
Then there exists an orthonormal basis for X, {u
1
, , u
n
} which has the property that
for each k n, span(x
1
, , x
k
) = span (u
1
, , u
k
) .
Proof: Let {x
1
, , x
n
} be a basis for X. Let u
1
x
1
/ |x
1
| . Thus for k = 1,
span (u
1
) = span (x
1
) and {u
1
} is an orthonormal set. Now suppose for some k <
n, u
1
, , u
k
have been chosen such that (u
j
, u
l
) =
jl
and span (x
1
, , x
k
) =
span (u
1
, , u
k
). Then dene
u
k+1

x
k+1
k
j=1
(x
k+1
u
j
) u
j
x
k+1
k
j=1
(x
k+1
u
j
) u
j
, (3.43)
where the denominator is not equal to zero because the x
j
form a basis and so
x
k+1
/ span (x
1
, , x
k
) = span (u
1
, , u
k
)
Thus by induction,
u
k+1
span (u
1
, , u
k
, x
k+1
) = span (x
1
, , x
k
, x
k+1
) .
Also, x
k+1
span (u
1
, , u
k
, u
k+1
) which is seen easily by solving 3.43 for x
k+1
and
it follows
span (x
1
, , x
k
, x
k+1
) = span (u
1
, , u
k
, u
k+1
) .
If l k, then denoting by C the scalar
x
k+1
k
j=1
(x
k+1
u
j
) u
j
1
,
(u
k+1
u
l
) = C
_
_
(x
k+1
u
l
)
k
j=1
(x
k+1
u
j
) (u
j
u
l
)
_
_
= C
_
_
(x
k+1
u
l
)
k
j=1
(x
k+1
u
j
)
lj
_
_
= C ((x
k+1
u
l
) (x
k+1
u
l
)) = 0.
The vectors, {u
j
}
n
j=1
, generated in this way are therefore an orthonormal basis because
each vector has unit length.
The process by which these vectors were generated is called the Gram Schmidt
process.
3.8.6 The Adjoint Of A Linear Transformation
There is a very important collection of ideas which relates a linear transformation to the
inner product in an inner product space. In order to discuss these ideas, it is necessary
to prove a simple and very interesting lemma about linear transformations which map
an inner product space H to the eld of scalars, F. This is sometimes called the Riesz
representation theorem.
Theorem 3.8.14 Let H be a nite dimensional inner product space and let
L L(H, F) . Then there exists a unique z H such that for all x H,
Lx =(x z) .
Proof: By the Gram Schmidt process, there exists an orthonormal basis for H,
{e
1
, , e
n
} .
First note that if x is arbitrary, there exist unique scalars, x
i
such that
x =
n
i=1
x
i
e
i
Taking the dot product of both sides with e
k
yields
(x e
k
) =
_
n
i=1
x
i
e
i
e
k
_
=
n
i=1
x
i
(e
i
e
k
) =
n
i=1
x
i
ik
= x
k
which shows that
x =
n
i=1
(x e
i
) e
i
and so by the properties of the dot product,
Lx =
n
i=1
(x e
i
) Le
i
=
_
x
n
i=1
e
i
Le
i
_
so let z =
n
i=1
e
i
Le
i
. It only remains to verify z is unique. However, this is obvious
because if (x z
1
) = (x z
2
) = Lx for all x, then
(x z
1
z
2
) = 0
for all x and in particular for x = z
1
z
2
which requires z
1
= z
2
. This proves the
theorem.
Now with this theorem, it becomes easy to dene something called the adjoint of
a linear operator. Let L L(H
1
, H
2
) where H
1
and H
2
are nite dimensional inner
product spaces. Then letting ( )
i
denote the inner product in H
i
,
x (Lx y)
2
is in L(H
1
, F) and so from Theorem 3.8.14 there exists a unique element of H
1
, denoted
by L
y such that for all x H

1
,
(Lx y)
2
= (xL
y)
1
Thus L
y H
1
when y H
2
. Also L
is linear. This is because by the properties of

the dot product,
(xL
(y +z))
1
(Lxy +z)
2
= (Lx y)
2
+ (Lx z)
2
= (xL
y)
1
+ (xL
z)
1
= (xL
y)
1
+ (xL
z)
1
and
(x L
y +L
z)
1
= (xL
y)
1
+ (xL
z)
1
Since
(xL
(y +z))
1
= (x L
y +L
z)
1
for all x, this requires
L
(y +z) = L
y +L
z.
In simple words, when you take it across the dot, you put a star on it. More precisely,
here is the denition.
Denition 3.8.15 Let H
1
and H
2
be nite dimensional inner product spaces
and let L L(H
1
, H
2
) . Then L
L(H
2
, H
1
) is dened by the formula
(Lx y)
2
= (xL
y)
1
.
In the case where H
1
= H
2
= H, an operator L L(H, H) is said to be self adjoint if
L = L
. This is also called Hermitian.

The following diagram might help.
H
1
L
H
2
H
1
L
H
2
I will not bother to place subscripts on the symbol for the dot product in the future. I
will be clear from context which inner product is meant.
Proposition 3.8.16 The adjoint has the following properties.
1. (xLy) = (L
x y) , (Lx y) = (xL
y)
2. (L
= L
3. (aL +bM)
= aL
+bM
4. (ML)
= L
Proof: Consider the rst claim.

(xLy) = (Ly x) = (yL
x) = (L
x y)
This does the rst claim. The second part was discussed earlier when the adjoint was
dened.
Consider the second claim. From the rst claim,
(Lx y) = (xL
y) =
_
(L
x y
_
and since this holds for all y, it follows Lx = (L
x.
Consider the third claim.
_
x (aL +bM)
y
_
= ((aL +bM) x y) = a (Lx y) +b (Mx y)
and
_
x
_
aL
+bM
_
y
_
= a (xL
y) +b (xM
y) = a (Lx y) +b (Mx y)
and since
_
x (aL +bM)
y
_
=
_
x
_
aL
+bM
_
y
_
for all x, it must be that
(aL +bM)
y =
_
aL
+bM
_
y
for all y which yields the third claim.
Consider the fourth.
_
x (ML)
y
_
= ((ML) x y) = (LxM
y) = (xL
y)
Since this holds for all x, y the conclusion follows as above. This proves the theorem.
Here is a very important example.

Example 3.8.17 Suppose F L(H
1
, H
2
) . Then FF
L(H
2
, H
2
) and is self adjoint.
To see this is so, note it is the composition of linear transformations and is therefore
linear as stated. To see it is self adjoint, Proposition 3.8.16 implies
(FF
= (F
= FF
In the case where A L(F

n
, F
m
) , considering the matrix of A with respect to the
usual bases, there is no loss of generality in considering A to be an mn matrix,
(Ax)
i
=
j
A
ij
x
j
.
Then in terms of components of the matrix, A,
(A
)
ij
= A
ji
.
You should verify this is so from the denition of the usual inner product on F
k
. The
following little proposition is useful.
Proposition 3.8.18 Suppose A is an mn matrix where m n. Also suppose
det (AA
) = 0.
Then A has m linearly independent rows and m independent columns.
Proof: Since det (AA
) = 0, it follows the m m matrix AA
has m independent
rows. If this is not true of A, then there exists x a 1 m matrix such that
xA = 0.
Hence
xAA
= 0
and this contradicts the independence of the rows of AA
. Thus the row rank of A

equals m and by Corollary 3.5.20 this implies the column rank of A also equals m. This
proves the proposition.
3.8.7 Schurs Theorem
Recall that for a linear transformation, L L(V, V ) , it could be represented in the
form
L =
ij
l
ij
v
i
v
j
where {v
1
, , v
n
} is a basis. Of course dierent bases will yield dierent matrices,
(l
ij
) . Schurs theorem gives the existence of a basis in an inner product space such that
(l
ij
) is particularly simple.
Denition 3.8.19 Let L L(V, V ) where V is vector space. Then a subspace
U of V is L invariant if L(U) U.
Theorem 3.8.20 Let L L(H, H) for H a nite dimensional inner product
space such that the restriction of L
to every L invariant subspace has its eigenvalues in

F. Then there exist constants, c
ij
for i j and an orthonormal basis, {w
i
}
n
i=1
such
that
L =
n
j=1
j
i=1
c
ij
w
i
w
j
The constants, c
ii
are the eigenvalues of L.
Proof: If dim(H) = 1 let H = span (w) where |w| = 1. Then Lw = kw for some
k. Then
L = kww
because by denition, ww(w) = w. Therefore, the theorem holds if H is 1 dimensional.
Now suppose the theorem holds for n 1 = dim(H) . By Theorem 3.6.4 and the
assumption, there exists w
n
, an eigenvector for L
. Dividing by its length, it can be

assumed |w
n
| = 1. Say L
w
n
= w
n
. Using the Gram Schmidt process, there exists an
orthonormal basis for H of the form {v
1
, , v
n1
, w
n
} . Then
(Lv
k
w
n
) = (v
k
L
w
n
) = (v
k
w
n
) = 0,
which shows
L : H
1
span (v
1
, , v
n1
) span (v
1
, , v
n1
) .
Denote by L
1
the restriction of L to H
1
. Since H
1
has dimension n 1, the induction
hypothesis yields an orthonormal basis, {w
1
, , w
n1
} for H
1
such that
L
1
=
n1
j=1
j
i=1
c
ij
w
i
w
j
. (3.44)
Then {w
1
, , w
n
} is an orthonormal basis for H because every vector in
span (v
1
, , v
n1
)
has the property that its dot product with w
n
is 0 so in particular, this is true for the
vectors {w
1
, , w
n1
}. Now dene c
in
to be the scalars satisfying
Lw
n

n
i=1
c
in
w
i
(3.45)
and let
B
n
j=1
j
i=1
c
ij
w
i
w
j
.
Then by 3.45,
Bw
n
=
n
j=1
j
i=1
c
ij
w
i
nj
=
n
j=1
c
in
w
i
= Lw
n
.
If 1 k n 1,
Bw
k
=
n
j=1
j
i=1
c
ij
w
i
kj
=
k
i=1
c
ik
w
i
while from 3.44,
Lw
k
= L
1
w
k
=
n1
j=1
j
i=1
c
ij
w
i
jk
=
k
i=1
c
ik
w
i
.
Since L = B on the basis {w
1
, , w
n
} , it follows L = B.
It remains to verify the constants, c
kk
are the eigenvalues of L, solutions of the
equation, det (I L) = 0. However, the denition of det (I L) is the same as
det (I C)
where C is the upper triangular matrix which has c
ij
for i j and zeros elsewhere.
This equals 0 if and only if is one of the diagonal entries, one of the c
kk
. This proves
the theorem.
There is a technical assumption in the above theorem about the eigenvalues of re-
strictions of L
being in F, the eld of scalars. If F = C this is no restriction. There is

also another situation in which F = R for which this will hold.
Lemma 3.8.21 Suppose H is a nite dimensional inner product space and
{w
1
, , w
n
}
is an orthonormal basis for H. Then
(w
i
w
j
)
= w
j
w
i
Proof: It suces to verify the two linear transformations are equal on {w
1
, , w
n
} .
Then
_
w
p
(w
i
w
j
)
w
k
_
((w
i
w
j
) w
p
w
k
) = (w
i
jp
w
k
) =
jp
ik
(w
p
(w
j
w
i
) w
k
) = (w
p
w
j
ik
) =
ik
jp
Since w
p
is arbitrary, it follows from the properties of the inner product that
_
x (w
i
w
j
)
w
k
_
= (x (w
j
w
i
) w
k
)
for all x H and hence (w
i
w
j
)
w
k
= (w
j
w
i
) w
k
. Since w
k
is arbitrary, This proves
the lemma.
Lemma 3.8.22 Let L L(H, H) for H an inner product space. Then if L = L
so
L is self adjoint, it follows all the eigenvalues of L are real.
Proof: Let {w
1
, , w
n
} be an orthonormal basis for H and let (l
ij
) be the matrix
of L with respect to this orthonormal basis. Thus
L =
ij
l
ij
w
i
w
j
, id =
ij
ij
w
i
w
j
Denote by M
L
the matrix whose ij
th
entry is l
ij
. Then by denition of what is meant
by the determinant of a linear transformation,
det (id L) = det (I M
L
)
and so the eigenvalues of L are the same as the eigenvalues of M
L
. However, M
L

L(C
n
, C
n
) with M
L
x determined by ordinary matrix multiplication. Therefore, by the
fundamental theorem of algebra and Theorem 3.6.4, if is an eigenvalue of L it follows
there exists a nonzero x C
n
such that M
L
x = x. Since L is self adjoint, it follows
from Lemma 3.8.21
L =
ij
l
ij
w
i
w
j
= L
ij
l
ij
w
j
w
i
=
ij
l
ji
w
i
w
j
which shows l
ij
= l
ji
.
Then
|x|
2
= (x x) = (x x) = (M
L
x x) =
ij
l
ij
x
j
x
i
=
ij
l
ij
x
j
x
i
=
ij
l
ji
x
j
x
i
= (M
L
x x) = (xM
L
x) = |x|
2
showing = . This proves the lemma.
If L is a self adjoint operator on H, either a real or complex inner product space,
it follows the condition about the eigenvalues of the restrictions of L
to L invariant
subspaces of H must hold because these restrictions are self adjoint. Here is why. Let
x, y be in one of those invariant subspaces. Then since L
= L,
(L
x y) = (xLy) = (xL
y)
so by the above lemma, the eigenvalues are real and are therefore, in the eld of scalars.
Now with this lemma, the following theorem is obtained. This is another major
theorem. It is equivalent to the theorem in matrix theory which states every self adjoint
matrix can be diagonalized.
Theorem 3.8.23 Let H be a nite dimensional inner product space, real or
complex, and let L L(H, H) be self adjoint. Then there exists an orthonormal basis
{w
1
, , w
n
} and real scalars,
k
such that
L =
n
k=1
k
w
k
w
k
.
The scalars are the eigenvalues and w
k
is an eigenvector for
k
for each k.
Proof: By Theorem 3.8.20, there exists an orthonormal basis, {w
1
, , w
n
} such
that
L =
n
j=1
n
i=1
c
ij
w
i
w
j
3.9. POLAR DECOMPOSITIONS 65
where c
ij
= 0 if i > j. Now using Lemma 3.8.21 and Proposition 3.8.16 along with the
assumption that L is self adjoint,
L =
n
j=1
n
i=1
c
ij
w
i
w
j
= L
=
n
j=1
n
i=1
c
ij
w
j
w
i
=
n
i=1
n
j=1
c
ji
w
i
w
j
If i < j, then this shows c
ij
= c
ji
and the second number equals zero because j > i.
Thus c
ij
= 0 if i < j and it is already known that c
ij
= 0 if i > j. Therefore, let
k
= c
kk
and the above reduces to
L =
n
j=1
j
w
j
w
j
=
n
j=1
j
w
j
w
j
showing that
j
=
j
so the eigenvalues are all real. Now
Lw
k
=
n
j=1
j
w
j
w
j
(w
k
) =
n
j=1
j
w
j
jk
=
k
w
k
which shows all the w
k
are eigenvectors. This proves the theorem.
3.9 Polar Decompositions
An application of Theorem 3.8.23, is the following fundamental result, important in
geometric measure theory and continuum mechanics. It is sometimes called the right
polar decomposition. When the following theorem is applied in continuum mechanics,
F is normally the deformation gradient, the derivative, discussed later, of a nonlinear
map from some subset of three dimensional space to three dimensional space. In this
context, U is called the right Cauchy Green strain tensor. It is a measure of how a body
is stretched independent of rigid motions. First, here is a simple lemma.
Lemma 3.9.1 Suppose R L(X, Y ) where X, Y are nite dimensional inner prod-
uct spaces and R preserves distances,
|Rx|
Y
= |x|
X
.
Then R
R = I.
Proof: Since R preserves distances, |Rx| = |x| for every x. Therefore from the
axioms of the dot product,
|x|
2
+|y|
2
+ (x y) + (y x)
= |x +y|
2
= (R(x +y) R(x +y))
= (RxRx) + (RyRy) + (Rx Ry) + (Ry Rx)
= |x|
2
+|y|
2
+ (R
Rx y) + (y R
Rx)
and so for all x, y,
(R
Rx x y) + (yR
Rx x) = 0
Hence for all x, y,
Re (R
Rx x y) = 0
Now for x, y given, choose C such that
(R
Rx x y) = |(R
Rx x y)|
Then
0 = Re (R
Rx xy) = Re (R
Rx x y)
= |(R
Rx x y)|
Thus |(R
Rx x y)| = 0 for all x, y because the given x, y were arbitrary. Let y =

R
Rx x to conclude that for all x,

R
Rx x = 0
which says R
R = I since x is arbitrary. This proves the lemma.

Denition 3.9.2 In case R L(X, X) for X a real or complex inner product
space of dimension n, R is said to be unitary if it preserves distances. Thus, from the
above lemma, unitary transformations are those which satisfy
R
R = RR
= id
where id is the identity map on X.
Theorem 3.9.3 Let X be a real or complex inner product space of dimension n,
let Y be a real or complex inner product space of dimension m n and let F L(X, Y ).
Then there exists R L(X, Y ) and U L(X, X) such that
F = RU, U = U
, (U is Hermitian),
all eigenvalues of U are non negative,
U
2
= F
F, R
R = I,
and |Rx| = |x| .
Proof: (F
F)
= F
F and so by Theorem 3.8.23, there is an orthonormal basis of

eigenvectors for X, {v
1
, , v
n
} such that
F
F =
n
i=1
i
v
i
v
i
, F
Fv
k
=
k
v
k
.
It is also clear that
i
0 because
i
(v
i
v
i
) = (F
Fv
i
v
i
) = (Fv
i
Fv
i
) 0.
Let
U
n
i=1
1/2
i
v
i
v
i
.
so U maps X to X and is self adjoint. Then from 3.13,
U
2
=
ij
1/2
i

1/2
j
(v
i
v
i
) (v
j
v
j
)
=
ij
1/2
i

1/2
j
v
i
v
j
ij
=
n
i=1
i
v
i
v
i
= F
F
3.9. POLAR DECOMPOSITIONS 67
Let {Ux
1
, , Ux
r
} be an orthonormal basis for U (X) . Extend this using the Gram
Schmidt procedure to an orthonormal basis for X,
{Ux
1
, , Ux
r
, y
r+1
, , y
n
} .
Next note that {Fx
1
, , Fx
r
} is also an orthonormal set of vectors in Y because
(Fx
k
Fx
j
) = (F
Fx
k
x
j
) =
_
U
2
x
k
x
j
_
= (Ux
k
Ux
j
) =
jk
.
Now extend {Fx
1
, , Fx
r
} to an orthonormal basis for Y,
{Fx
1
, , Fx
r
, z
r+1
, , z
m
} .
Since m n, there are at least as many z
k
as there are y
k
.
Now dene R as follows. For x X, there exist unique scalars, c
k
and d
k
such that
x =
r
k=1
c
k
Ux
k
+
n
k=r+1
d
k
y
k
.
Then
Rx
r
k=1
c
k
Fx
k
+
n
k=r+1
d
k
z
k
. (3.46)
Thus, since {Fx
1
, , Fx
r
, z
r+1
, , z
m
} is orthonormal, a short computation shows
|Rx|
2
=
r
k=1
|c
k
|
2
+
n
k=r+1
|d
k
|
2
= |x|
2
.
Now I need to verify RUx = Fx. Since {Ux
1
, , Ux
r
} is an orthonormal basis for
UX, there exist scalars, b
k
such that
Ux =
r
k=1
b
k
Ux
k
(3.47)
and so from the denition of R given in 3.46,
RUx
r
k=1
b
k
Fx
k
= F
_
r
k=1
b
k
x
k
_
.
RU = F is shown if F (
r
k=1
b
k
x
k
) = F (x).
_
F
_
r
k=1
b
k
x
k
_
F (x) F
_
r
k=1
b
k
x
k
_
F (x)
_
=
_
F
F
_
r
k=1
b
k
x
k
x
_
k=1
b
k
x
k
x
_
=
_
U
2
_
r
k=1
b
k
x
k
x
_
_
r
k=1
b
k
x
k
x
__
=
_
U
_
r
k=1
b
k
x
k
x
_
U
_
r
k=1
b
k
x
k
x
__
=
_
r
k=1
b
k
Ux
k
Ux
r
k=1
b
k
Ux
k
Ux
_
= 0
by 3.47.
Since |Rx| = |x| , it follows R
R = I from Lemma 3.9.1. This proves the theorem.
The following corollary follows as a simple consequence of this theorem. It is called

the left polar decomposition.
Corollary 3.9.4 Let F L(X, Y ) and suppose n m where X is a inner product
space of dimension n and Y is a inner product space of dimension m. Then there exists
a Hermitian U L(X, X) , and an element of L(X, Y ) , R, such that
F = UR, RR
= I.
Proof: Recall that L
= L and (ML)
= L
. Now apply Theorem 3.9.3 to

F
L(X, Y ). Thus,
F
= R
U
where R
and U satisfy the conditions of that theorem. Then

F = UR
and RR
= R
= I. This proves the corollary.

This is a good place to consider a useful lemma.
Lemma 3.9.5 Let X be a nite dimensional inner product space of dimension n and
let R L(X, X) be unitary. Then |det (R)| = 1.
Proof: Let {w
k
} be an orthonormal basis for X. Then to take the determinant it
suces to take the determinant of the matrix, (c
ij
) where
R =
ij
c
ij
w
i
w
j
.
Rw
k
=
i
c
ik
w
i
and so
(Rw
k
, w
l
) = c
lk
.
and hence
R =
lk
(Rw
k
, w
l
) w
l
w
k
Similarly
R
ij
(R
w
j
, w
i
) w
i
w
j
.
Since R is given to be unitary,
RR
= id =
lk
ij
(Rw
k
, w
l
) (R
w
j
, w
i
) (w
l
w
k
) (w
i
w
j
)
=
ijkl
(Rw
k
, w
l
) (R
w
j
, w
i
)
ki
w
l
w
j
=
jl
_
i
(Rw
i
, w
l
) (R
w
j
, w
i
)
_
w
l
w
j
Hence
i
(R
w
j
, w
i
) (Rw
i
, w
l
) =
jl
=
i
(Rw
i
, w
j
) (Rw
i
, w
l
) (3.48)
3.10. EXERCISES 69
because
id =
jl
jl
w
l
w
j
Thus letting M be the matrix whose ij
th
entry is (Rw
i
, w
l
) , det (R) is dened as
det (M) and 3.48 says
i
_
M
T
_
ji
M
il
=
jl
.
It follows
1 = det (M) det
_
M
T
_
= det (M) det
_
M
_
= det (M) det (M) = |det (M)|
2
.
Thus |det (R)| = |det (M)| = 1 as claimed.
3.10 Exercises
1. For u, v vectors in F
3
, dene the product, u v u
1
v
1
+2u
2
v
2
+3u
3
v
3
. Show the
axioms for a dot product all hold for this funny product. Prove
|u v| (u u)
1/2
(v v)
1/2
.
2. Suppose you have a real or complex vector space. Can it always be considered
as an inner product space? What does this mean about Schurs theorem? Hint:
Start with a basis and decree the basis is orthonormal. Then dene an inner
product accordingly.
3. Show that (a b) =
1
4
_
|a +b|
2
|a b|
2
_
.
4. Prove from the axioms of the dot product the parallelogram identity, |a +b|
2
+
|a b|
2
= 2 |a|
2
+ 2 |b|
2
.
5. Suppose f, g are two Darboux Stieltjes integrable functions dened on [0, 1] . Dene
(f g) =
1
0
f (x) g (x)dF.
Show this dot product satises the axioms for the inner product. Explain why the
Cauchy Schwarz inequality continues to hold in this context and state the Cauchy
Schwarz inequality in terms of integrals. Does the Cauchy Schwarz inequality still
hold if
(f g) =
1
0
f (x) g (x)p (x) dF
where p (x) is a given nonnegative function? If so, what would it be in terms of
integrals.
6. If A is an nn matrix considered as an element of L(C
n
, C
n
) by ordinary matrix
multiplication, use the inner product in C
n
to show that (A
)
ij
= A
ji
. In words,
the adjoint is the transpose of the conjugate.
7. A symmetric matrix is a real n n matrix A which satises A
T
= A. Show every
symmetric matrix is self adjoint and that there exists an orthonormal set of real
vectors {x
1
, , x
n
} such that
A =
k
x
k
x
k
8. A normal matrix is an n n matrix, A such that A
A = AA
. Show that for a

normal matrix there is an orthonormal basis of C
n
, {x
1
, , x
n
} such that
A =
i
a
i
x
i
x
i
That is, with respect to this basis the matrix of A is diagonal. Hint: This is a
harder version of what was done to prove Theorem 3.8.23. Use Schurs theorem
to write A =
n
j=1
n
i=1
B
ij
w
i
w
j
where B
ij
is an upper triangular matrix. Then
use the condition that A is normal and eventually get an equation
k
B
ik
B
lk
=
k
B
ki
B
kl
Next let i = l and consider rst l = 1, then l = 2, etc. If you are careful, you will
nd B
ij
= 0 unless i = j.
9. Suppose A L(H, H) where H is an inner product space and
A =
i
a
i
w
i
w
i
where the vectors {w
1
, , w
n
} are an orthonormal set. Show that A must be
normal. In other words, you cant represent A L(H, H) in this very convenient
way unless it is normal.
10. If L is a self adjoint operator dened on an inner product space, H such that
L has all only nonnegative eigenvalues. Explain how to dene L
1/n
and show
why what you come up with is indeed the n
th
root of the operator. For a
self adjoint operator L on an inner product space, can you dene sin (L)
k=0
(1)
k
L
2k+1
/ (2k + 1)!? What does the innite series mean? Can you make
some sense of this using the representation of L given in Theorem 3.8.23?
11. If L is a self adjoint linear transformation dened on L(H, H) for H an inner
product space which has all eigenvalues nonnegative, show the square root is
unique.
12. Using Problem 11 show F L(H, H) for H an inner product space is normal if
and only if RU = UR where F = RU is the right polar decomposition dened
above. Recall R preserves distances and U is self adjoint. What is the geometric
signicance of a linear transformation being normal?
13. Suppose you have a basis, {v
1
, , v
n
} in an inner product space, X. The Gram-
mian matrix is the n n matrix whose ij
th
entry is (v
i
v
j
) . Show this matrix is
invertible. Hint: You might try to show that the inner product of two vectors,
k
a
k
v
k
and
k
b
k
v
k
has something to do with the Grammian.
14. Suppose you have a basis, {v
1
, , v
n
} in an inner product space, X. Show there
exists a dual basis
_
v
1
, , v
n
_
which satises v
k
v
j
=
k
j
, which equals 0 if
j = k and equals 1 if j = k.
Sequences
4.1 Vector Valued Sequences And Their Limits
Functions dened on the set of integers larger than a given integer which have values
in a vector space are called vector valued sequences. I will always assume the vector
space is a normed vector space. Actually, it will specialized even more to F
n
, although
everything can be done for an arbitrary vector space and when it creates no diculties,
I will state certain denitions and easy theorems in the more general context and use
the symbol |||| to refer to the norm. Other than this, the notation is almost the same as
it was when the sequences had values in C. The main dierence is that certain variables
are placed in bold face to indicate they are vectors. Even this is not really necessary
but it is conventional to do it.The concept of subsequence is also the same as it was for
sequences of numbers. To review,
n
} be a sequence and let n
1
< n
2
< n
3
, be any
strictly increasing list of integers such that n
1
is at least as large as the rst number in
the domain of the function. Then if b
k
a
n
k
, {b
k
} is called a subsequence of {a
n
} .
Example 4.1.2 Let a
n
=
_
n + 1, sin
_
1
n
__
. Then {a
n
}
n=1
is a vector valued sequence.
The denition of a limit of a vector valued sequence is given next. It is just like the
denition given for sequences of scalars. However, here the symbol || refers to the usual
norm in F
n
. In a general normed vector space, it will be denoted by |||| .
Denition 4.1.3 A vector valued sequence {a
n
}
n=1
converges to a in a normed
vector space V, written as
lim
n
a
n
= a or a
n
a
if and only if for every > 0 there exists n
such that whenever n n
,
||a
n
a|| < .
In words the denition says that given any measure of closeness , the terms of the
sequence are eventually this close to a. Here, the word eventually refers to n being
suciently large.
Theorem 4.1.4 If lim
n
a
n
= a and lim
n
a
n
= a
1
then a
1
= a.
Proof: Suppose a
1
= a. Then let 0 < < ||a
1
a|| /2 in the denition of the limit.
It follows there exists n
such that if n n
, then ||a
n
a|| < and |a
n
a
1
| < .
Therefore, for such n,
||a
1
a|| ||a
1
a
n
|| +||a
n
a||
< + < ||a
1
a|| /2 +||a
1
a|| /2 = ||a
1
a|| ,
a contradiction.
71
72 SEQUENCES
n
} and {b
n
} are vector valued sequences and that
lim
n
a
n
= a and lim
n
b
n
= b.
Also suppose x and y are scalars in F. Then
lim
n
xa
n
+yb
n
= xa +yb (4.1)
Also,
lim
n
(a
n
b
n
) = (a b) (4.2)
If {x
n
} is a sequence of scalars in F converging to x and if {a
n
} is a sequence of vectors
in F
n
converging to a, then
lim
n
x
n
a
n
= xa. (4.3)
Also if {x
k
} is a sequence of vectors in F
n
then x
k
x, if and only if for each j,
lim
k
x
j
k
= x
j
. (4.4)
where here
x
k
=
_
x
1
k
, , x
n
k
_
, x =
_
x
1
, , x
n
_
.
Proof: Consider the rst claim. By the triangle inequality
||xa +yb(xa
n
+yb
n
)|| |x| ||a a
n
|| +|y| ||b b
n
|| .
By denition, there exists n
such that if n n
,
||a a
n
|| , ||b b
n
|| <

2 (1 +|x| +|y|)
so for n > n
,
||xa +yb(xa
n
+yb
n
)|| < |x|

2 (1 +|x| +|y|)
+|y|

2 (1 +|x| +|y|)
.
Now consider the second. Let > 0 be given and choose n
1
such that if n n
1
then
|a
n
a| < 1.
For such n, it follows from the Cauchy Schwarz inequality and properties of the inner
product that
|a
n
b
n
a b| |(a
n
b
n
) (a
n
b)| +|(a
n
b) (a b)|
|a
n
| |b
n
b| +|b| |a
n
a|
(|a| + 1) |b
n
b| +|b| |a
n
a| .
Now let n
2
be large enough that for n n
2
,
|b
n
b| <

2 (|a| + 1)
, and |a
n
a| <

2 (|b| + 1)
.
Such a number exists because of the denition of limit. Therefore, let
n
> max (n
1
, n
2
) .
4.1. VECTOR VALUED SEQUENCES AND THEIR LIMITS 73
For n n
,
|a
n
b
n
a b| (|a| + 1) |b
n
b| +|b| |a
n
a|
< (|a| + 1)

2 (|a| + 1)
+|b|

2 (|b| + 1)
.
This proves 4.2. The claim, 4.3 is left for you to do.
Finally consider the last claim. If 4.4 holds, then from the denition of distance in
F
n
,
lim
k
|x x
k
| lim
k
_
n
j=1
_
x
j
x
j
k
_
2
= 0.
On the other hand, if lim
k
|x x
k
| = 0, then since
x
j
k
x
j
|x x
k
| , it follows
from the squeezing theorem that
lim
k
x
j
k
x
j
= 0.
An important theorem is the one which states that if a sequence converges, so does
every subsequence. You should review Denition 4.1.1 at this point. The proof is
identical to the one involving sequences of numbers.
n
} be a vector valued sequence with lim
n
x
n
= x and
let {x
n
k
} be a subsequence. Then lim
k
x
n
k
= x.
Proof: Let > 0 be given. Then there exists n
such that if n > n
, then ||x
n
x|| <
. Suppose k > n
. Then n
k
k > n
and so
||x
n
k
x|| <
showing lim
k
x
n
k
= x as claimed.
n
} be a sequence of real numbers and suppose each x
n
l
( l)and lim
n
x
n
= x. Then x l ( l) . More generally, suppose {x
n
} and {y
n
} are
two sequences such that lim
n
x
n
= x and lim
n
y
n
= y. Then if x
n
y
n
for all n
suciently large, then x y.
Proof: Let > 0 be given. Then for n large enough,
l x
n
> x
and so
l + x.
Since > 0 is arbitrary, this requires l x. The other case is entirely similar or else
you could consider l and {x
n
} and apply the case just considered.
Consider the last claim. There exists N such that if n N then x
n
y
n
and
|x x
n
| +|y y
n
| < /2.
Then considering n > N in what follows,
x y x
n
+/2 (y
n
/2) = x
n
y
n
+ .
Since was arbitrary, it follows x y 0. This proves the theorem.
74 SEQUENCES
n
} be a sequence vectors and suppose each ||x
n
|| l
( l)and lim
n
x
n
= x. Then x l ( l) . More generally, suppose {x
n
} and {y
n
}
are two sequences such that lim
n
x
n
= x and lim
n
y
n
= y. Then if ||x
n
|| ||y
n
||
for all n suciently large, then ||x|| ||y|| .
Proof: It suces to just prove the second part since the rst part is similar. By
the triangle inequality,
|||x
n
|| ||x||| ||x
n
x||
and for large n this is given to be small. Thus {||x
n
||} converges to ||x|| . Similarly
{||y
n
||} converges to ||y||. Now the desired result follows from Theorem 4.1.7. This
proves the theorem.
4.2 Sequential Compactness
The following is the denition of sequential compactness. It is a very useful notion
which can be used to prove existence theorems.
Denition 4.2.1 A set, K V, a normed vector space is sequentially compact
if whenever {a
n
} K is a sequence, there exists a subsequence, {a
n
k
} such that this
subsequence converges to a point of K.
First of all, it is convenient to consider the sequentially compact sets in F.
Lemma 4.2.2 Let I
k
=
_
a
k
, b
k
and suppose that for all k = 1, 2, ,

I
k
I
k+1
.
Then there exists a point, c R which is an element of every I
k
.
Proof: Since I
k
I
k+1
, this implies
a
k
a
k+1
, b
k
b
k+1
. (4.5)
Consequently, if k l,
a
l
a
l
b
l
b
k
. (4.6)
Now dene
c sup
_
a
l
: l = 1, 2,
_
By the rst inequality in 4.5, and 4.6
a
k
c = sup
_
a
l
: l = k, k + 1,
_
b
k
(4.7)
for each k = 1, 2 . Thus c I
k
for every k and This proves the lemma. If this went
too fast, the reason for the last inequality in 4.7 is that from 4.6, b
k
is an upper bound
to
_
a
l
: l = k, k + 1,
_
. Therefore, it is at least as large as the least upper bound.
Theorem 4.2.3 Every closed interval, [a, b] is sequentially compact.
Proof: Let {x
n
} [a, b] I
0
. Consider the two intervals
_
a,
a+b
2
and
_
a+b
2
, b
each
of which has length (b a) /2. At least one of these intervals contains x
n
for innitely
many values of n. Call this interval I
1
. Now do for I
1
what was done for I
0
. Split it
in half and let I
2
be the interval which contains x
n
for innitely many values of n.
Continue this way obtaining a sequence of nested intervals I
0
I
1
I
2
I
3
where
the length of I
n
is (b a) /2
n
. Now pick n
1
such that x
n1
I
1
, n
2
such that n
2
> n
1
4.3. CLOSED AND OPEN SETS 75
and x
n2
I
2
, n
3
such that n
3
> n
2
and x
n3
I
3
, etc. (This can be done because in each
case the intervals contained x
n
for innitely many values of n.) By the nested interval
lemma there exists a point, c contained in all these intervals. Furthermore,
|x
n
k
c| < (b a) 2
k
and so lim
k
x
n
k
= c [a, b] . This proves the theorem.
Theorem 4.2.4 Let
I =
n
k=1
K
k
where K
k
is a sequentially compact set in F. Then I is a sequentially compact set in F
n
.
Proof: Let {x
k
}
k=1
be a sequence of points in I. Let
x
k
=
_
x
1
k
, , x
n
k
_
Thus
_
x
i
k
_
k=1
is a sequence of points in K
i
. Since K
i
is sequentially compact, there exists
a subsequence of {x
k
}
k=1
denoted by {x
1k
} such that
_
x
1
1k
_
converges to x
1
for some
x
1
K
1
. Now there exists a further subsequence, {x
2k
} such that
_
x
1
2k
_
converges to x
1
,
because by Theorem 4.1.6, subsequences of convergent sequences converge to the same
limit as the convergent sequence, and in addition,
_
x
2
2k
_
converges to some x
2
K
2
.
Continue taking subsequences such that for {x
jk
}
k=1
, it follows
_
x
r
jk
_
converges to
some x
r
K
r
for all r j. Then {x
nk
}
k=1
is the desired subsequence such that the
sequence of numbers in F obtained by taking the j
th
component of this subsequence
converges to some x
j
K
j
. It follows from Theorem 4.1.5 that x
_
x
1
, , x
n
_
I
and is the limit of {x
nk
}
k=1
Corollary 4.2.5 Any box of the form
[a, b] +i [c, d] {x +iy : x [a, b] , y [c, d]}
is sequentially compact in C.
Proof: The given box is essentially [a, b] [c, d] .
{x
k
+iy
k
}
k=1
[a, b] +i [c, d]
is the same as saying (x
k
, y
k
) [a, b] [c, d] . Therefore, there exists (x, y) [a, b] [c, d]
such that x
k
x and y
k
y. In other words x
k
+iy
k
x+iy and x+iy [a, b]+i [c, d] .
This proves the corollary.
4.3 Closed And Open Sets
The denition of open and closed sets is next.
Denition 4.3.1 Let U be a set of points in a normed vector space, V . A point,
p U is said to be an interior point if whenever ||x p|| is suciently small, it follows
x U also. The set of points, x which are closer to p than is denoted by
B(p, ) {x V : ||x p|| < } .
This symbol, B(p, ) is called an open ball of radius . Thus a point, p is an interior
point of U if there exists > 0 such that p B(p, ) U. An open set is one for which
every point of the set is an interior point. Closed sets are those which are complements
of open sets. Thus H is closed means H
C
is open.
76 SEQUENCES
Theorem 4.3.2 The intersection of any nite collection of open sets is open.
The union of any collection of open sets is open. The intersection of any collection of
closed sets is closed and the union of any nite collection of closed sets is closed.
Proof: To see that any union of open sets is open, note that every point of the
union is in at least one of the open sets. Therefore, it is an interior point of that set
and hence an interior point of the entire union.
Now let {U
1
, , U
m
} be some open sets and suppose p
m
k=1
U
k
. Then there exists
r
k
> 0 such that B(p, r
k
) U
k
. Let 0 < r min (r
1
, r
2
, , r
m
) . Then B(p, r)
m
k=1
U
k
and so the nite intersection is open. Note that if the nite intersection is
empty, there is nothing to prove because it is certainly true in this case that every point
in the intersection is an interior point because there arent any such points.
Suppose {H
1
, , H
m
} is a nite set of closed sets. Then
m
k=1
H
k
is closed if its
complement is open. However, from DeMorgans laws,
(
m
k=1
H
k
)
C
=
m
k=1
H
C
k
,
a nite intersection of open sets which is open by what was just shown.
Next let C be some collection of closed sets. Then
(C)
C
=
_
H
C
: H C
_
,
a union of open sets which is therefore open by the rst part of the proof. Thus C is
closed. This proves the theorem.
Next there is the concept of a limit point which gives another way of characterizing
closed sets.
Denition 4.3.3 Let A be any nonempty set and let x be a point. Then x is
said to be a limit point of A if for every r > 0, B(x, r) contains a point of A which is
not equal to x.
Example 4.3.4 Consider A = B(x, ) , an open ball in a normed vector space. Then
every point of B(x, ) is a limit point. There are more general situations than normed
vector spaces in which this assertion is false.
If z B(x, ) , consider z+
1
k
(x z) w
k
for k N. Then
||w
k
x|| =
z+
1
k
(x z) x
_
1
1
k
_
z
_
1
1
k
_
x
=
k 1
k
||z x|| <
and also
||w
k
z||
1
k
||x z|| < /k
so w
k
z. Furthermore, the w
k
are distinct. Thus z is a limit point of A as claimed.
This is because every ball containing z contains innitely many of the w
k
and since
they are all distinct, they cant all be equal to z.
Similarly, the following holds in any normed vector space.
Theorem 4.3.5 Let A be a nonempty set in V, a normed vector space. A point
a is a limit point of A if and only if there exists a sequence of distinct points of A, {a
n
}
which converges to a. Also a nonempty set, A is closed if and only if it contains all its
limit points.
4.3. CLOSED AND OPEN SETS 77
Proof: Suppose rst a is a limit point of A. There exists a
1
B(a, 1) A such
that a
1
= a. Now supposing distinct points, a
1
, , a
n
have been chosen such that none
are equal to a and for each k n, a
k
B(a, 1/k) , let
0 < r
n+1
< min
_
1
n + 1
, ||a a
1
|| , , ||a a
n
||
_
.
Then there exists a
n+1
B(a, r
n+1
) A with a
n+1
= a. Because of the denition of
r
n+1
, a
n+1
is not equal to any of the other a
k
for k < n+1. Also since ||a a
m
|| < 1/m,
it follows lim
m
a
m
= a. Conversely, if there exists a sequence of distinct points of A
converging to a, then B(a, r) contains all a
n
for n large enough. Thus B(a, r) contains
innitely many points of A since all are distinct. Thus at least one of them is not equal
to a. This establishes the rst part of the theorem.
Now consider the second claim. If A is closed then it is the complement of an open
set. Since A
C
is open, it follows that if a A
C
, then there exists > 0 such that
B(a, ) A
C
and so no point of A
C
can be a limit point of A. In other words, every
limit point of A must be in A. Conversely, suppose A contains all its limit points. Then
A
C
does not contain any limit points of A. It also contains no points of A. Therefore, if
a A
C
, since it is not a limit point of A, there exists > 0 such that B(a, ) contains
no points of A dierent than a. However, a itself is not in A because a A
C
. Therefore,
B(a, ) is entirely contained in A
C
. Since a A
C
was arbitrary, this shows every point
of A
C
is an interior point and so A
C
is open. This proves the theorem.
Closed subsets of sequentially compact sets are sequentially compact.
Theorem 4.3.6 If K is a sequentially compact set in a normed vector space and
if H is a closed subset of K then H is sequentially compact.
Proof: Let {x
n
} H. Then since K is sequentially compact, there is a subsequence,
{x
n
k
} which converges to a point, x K. If x / H, then since H
C
is open, it follows
there exists B(x, r) such that this open ball contains no points of H. However, this is
a contradiction to having x
n
k
x which requires x
n
k
B(x, r) for all k large enough.
Thus x H and this has shown H is sequentially compact.
Denition 4.3.7 A set S V, a normed vector space is bounded if there is
some r > 0 such that S B(0, r) .
Theorem 4.3.8 Every closed and bounded set in F
n
is sequentially compact.
Conversely, every sequentially compact set in F
n
is closed and bounded.
Proof: Let H be a closed and bounded set in F
n
. Then H B(0, r) for some r.
Therefore, if x H, x =(x
1
, , x
n
) , it must be that
_
n
i=1
|x
i
|
2
< r
and so each x
i
[r, r] + i [r, r] R
r
, a sequentially compact set by Corollary 4.2.5.
Thus H is a closed subset of
n
i=1
R
r
which is a sequentially compact set by Theorem 4.2.4. Therefore, by Theorem 4.3.6 it
follows H is sequentially compact.
Conversely, suppose K is a sequentially compact set in F
n
. If it is not bounded, then
there exists a sequence, {k
m
} such that k
m
K but k
m
/ B(0, m) for m = 1, 2, .
78 SEQUENCES
However, this sequence cannot have any convergent subsequence because if k
m
k
k,
then for large enough m, k B(0,m) D(0, m) and k
m
k
B(0, m)
C
for all k large
enough and this is a contradiction because there can only be nitely many points of the
sequence in B(0, m) . If K is not closed, then it is missing a limit point. Say k
is a
limit point of K which is not in K. Pick k
m
B
_
k
,
1
m
_
. Then {k
m
} converges to
k
and so every subsequence also converges to k
by Theorem 4.1.6. Thus there is no

point of K which is a limit of some subsequence of {k
m
} , a contradiction. This proves
the theorem.
What are some examples of closed and bounded sets in a general normed vector
space and more specically F
n
?
Proposition 4.3.9 Let D(z, r) denote the set of points,
{w V : ||wz|| r}
Then D(z, r) is closed and bounded. Also, let S (z,r) denote the set of points
{w V : ||wz|| = r}
Then S (z, r) is closed and bounded. It follows that if V = F
n
,then these sets are
sequentially compact.
Proof: First note D(z, r) is bounded because
D(z, r) B(0, ||z|| + 2r)
Here is why. Let x D(z, r) . Then ||x z|| r and so
||x|| ||x z|| +||z|| r +||z|| < 2r +||z|| .
It remains to verify it is closed. Suppose then that y / D(z, r) . This means ||y z|| > r.
Consider the open ball B(y, ||y z|| r) . If x B(y, ||y z|| r) , then
||x y|| < ||y z|| r
and so by the triangle inequality,
||z x|| ||z y|| ||y x|| > ||x y|| +r ||x y|| = r
Thus the complement of D(z, r) is open and so D(z, r) is closed.
For the second type of set, note S (z, r)
C
= B(z, r) D(z, r)
C
, the union of two
open sets which by Theorem 4.3.2 is open. Therefore, S (z, r) is a closed set which is
clearly bounded because S (z, r) D(z, r).
4.4 Cauchy Sequences And Completeness
The concept of completeness is that every Cauchy sequence converges. Cauchy sequences
are those sequences which have the property that ultimately the terms of the sequence
are bunching up. More precisely,
Denition 4.4.1 {a
n
} is a Cauchy sequence in a normed vector space, V if for
all > 0, there exists n
such that whenever n, m n
,
||a
n
a
m
|| < .
4.4. CAUCHY SEQUENCES AND COMPLETENESS 79
Theorem 4.4.2 The set of terms (values) of a Cauchy sequence in a normed
vector space V is bounded.
Proof: Let = 1 in the denition of a Cauchy sequence and let n > n
1
. Then from
the denition,
||a
n
a
n1
|| < 1.
It follows that for all n > n
1
,
||a
n
|| < 1 +||a
n1
|| .
Therefore, for all n,
||a
n
|| 1 +||a
n1
|| +
n1
k=1
||a
k
|| .
Theorem 4.4.3 If a sequence {a
n
} in V, a normed vector space converges, then
the sequence is a Cauchy sequence.
Proof: Let > 0 be given and suppose a
n
a. Then from the denition of
convergence, there exists n
such that if n > n
, it follows that
||a
n
a|| <

2
Therefore, if m, n n
+ 1, it follows that
||a
n
a
m
|| ||a
n
a|| +||a a
m
|| <

2
+

2
=
showing that, since > 0 is arbitrary, {a
n
} is a Cauchy sequence.
The following theorem is very useful. It is identical to an earlier theorem. All that
is required is to put things in bold face to indicate they are vectors.
n
} is a Cauchy sequence in any normed vector space
and there exists a subsequence, {a
n
k
} which converges to a. Then {a
n
} also converges
to a.
Proof: Let > 0 be given. There exists N such that if m, n > N, then
||a
m
a
n
|| < /2.
Also there exists K such that if k > K, then
||a a
n
k
|| < /2.
Then let k > max (K, N) . Then for such k,
||a
k
a|| ||a
k
a
n
k
|| +||a
n
k
a||
< /2 +/2 = .
Denition 4.4.5 If V is a normed vector space having the property that every
Cauchy sequence converges, then V is called complete. It is also referred to as a Banach
space.
Example 4.4.6 R is given to be complete. This is a fundamental axiom on which
calculus is developed.
80 SEQUENCES
Given R is complete, the following lemma is easily obtained.
Lemma 4.4.7 C is complete.
Proof: Let {x
k
+iy
k
}
k=1
be a Cauchy sequence in C. This requires {x
k
} and {y
k
}
are both Cauchy sequences in R. This follows from the obvious estimates
|x
k
x
m
| , |y
k
y
m
| |(x
k
+iy
k
) (x
m
+iy
m
)| .
By completeness of R there exists x R such that x
k
x and similarly there exists
y R such that y
k
y. Therefore, since
|(x
k
+iy
k
) (x +iy)|
(x
k
x)
2
+ (y
k
y)
2
|x
k
x| +|y
k
y|
it follows (x
k
+iy
k
) (x +iy) .
A simple generalization of this idea yields the following theorem.
Theorem 4.4.8 F
n
is complete.
Proof: By 4.4.7, F is complete. Now let {a
m
} be a Cauchy sequence in F
n
. Then
by the denition of the norm
a
j
m
a
j
k
|a
m
a
k
|
where a
j
m
denotes the j
th
component of a
m
. Thus for each j = 1, 2, , n,
_
a
j
m
_
m=1
is
a Cauchy sequence. It follows from Theorem 4.4.7, the completeness of F, there exists
a
j
such that
lim
m
a
j
m
= a
j
Theorem 4.1.5 implies that lim
m
a
m
= a where
a =
_
a
1
, , a
n
_
.
4.5 Shrinking Diameters
It is useful to consider another version of the nested interval lemma. This involves a
sequence of sets such that set (n + 1) is contained in set n and such that their diameters
converge to 0. It turns out that if the sets are also closed, then often there exists a unique
point in all of them.
Denition 4.5.1 Let S be a nonempty set in a normed vector space, V . Then
diam(S) is dened as
diam(S) sup{||x y|| : x, y S} .
This is called the diameter of S.
Theorem 4.5.2 Let {F
n
}
n=1
be a sequence of closed sets in F
n
such that
lim
n
diam(F
n
) = 0
and F
n
F
n+1
for each n. Then there exists a unique p
k=1
F
k
.
4.6. EXERCISES 81
Proof: Pick p
k
F
k
. This is always possible because by assumption each set is
nonempty. Then {p
k
}
k=m
F
m
and since the diameters converge to 0, it follows
{p
k
} is a Cauchy sequence. Therefore, it converges to a point, p by completeness of
F
n
discussed in Theorem 4.4.8. Since each F
k
is closed, p F
k
for all k. This is because
it is a limit of a sequence of points only nitely many of which are not in the closed set
F
k
. Therefore, p
k=1
F
k
. If q
k=1
F
k
, then since both p, q F
k
,
|p q| diam(F
k
) .
It follows since these diameters converge to 0, |p q| for every . Hence p = q. This
proves the theorem.
A sequence of sets {G
n
} which satises G
n
G
n+1
for all n is called a nested
sequence of sets.
4.6 Exercises
1. For a nonempty set, S in a normed vector space, V, dene a function
x dist (x,S) inf {||x y|| : y S} .
Show
||dist (x, S) dist (y, S)|| ||x y|| .
2. Let A be a nonempty set in F
n
or more generally in a normed vector space. Dene
the closure of A to equal the intersection of all closed sets which contain A. This
is usually denoted by A. Show A = A A
where A
consists of the set of limit

points of A. Also explain why A is closed.
3. The interior of a set was dened above. Tell why the interior of a set is always an
open set. The interior of a set A is sometimes denoted by A
0
.
4. Given an example of a set A whose interior is empty but whose closure is all of
R
n
.
5. A point, p is said to be in the boundary of a nonempty set, A if for every r > 0,
B(p, r) contains points of A as well as points of A
C
. Sometimes this is denoted
as A. In a normed vector space, is it always the case that AA = A? Prove or
disprove.
6. Give an example of a nite dimensional normed vector space where the eld of
scalars is the rational numbers which is not complete.
7. Explain why as far as the theorems of this chapter are concerned, C
n
is essentially
the same as R
2n
.
8. A set, A R
n
is said to be convex if whenever x, y A it follows tx+(1 t) y A
whenever t [0, 1]. Show B(z, r) is convex. Also show D(z,r) is convex. If A is
convex, does it follow A is convex? Explain why or why not.
9. Let A be any nonempty subset of R
n
. The convex hull of A, usually denoted by
co (A) is dened as the set of all convex combinations of points in A. A convex
combination is of the form
p
k=1
t
k
a
k
where each t
k
0 and
k
t
k
= 1. Note
that p can be any nite number. Show co (A) is convex.
82 SEQUENCES
10. Suppose A R
n
and z co (A) . Thus z =
p
k=1
t
k
a
k
for t
k
0 and
k
t
k
=
1. Show there exists n + 1 of the points {a
1
, , a
p
} such that z is a convex
combination of these n +1 points. Hint: Show that if p > n +1 then the vectors
{a
k
a
1
}
p
k=2
must be linearly dependent. Conclude from this the existence of
scalars {
i
} such that
p
i=1
i
a
i
= 0. Now for s R, z =
p
k=1
(t
k
+s
i
) a
k
.
Consider small s and adjust till one or more of the t
k
+s
k
vanish. Now you are
in the same situation as before but with only p1 of the a
k
. Repeat the argument
till you end up with only n + 1 at which time you cant repeat again.
11. Show that any uncountable set of points in F
n
must have a limit point.
12. Let V be any nite dimensional vector space having a basis {v
1
, , v
n
} . For
x V, let
x =
n
k=1
x
k
v
k
so that the scalars, x
k
are the components of x with respect to the given basis.
Dene for x, y V
(x y)
n
i=1
x
i
y
i
Show this is a dot product for V satisfying all the axioms of a dot product presented
earlier.
13. In the context of Problem 12 let |x| denote the norm of x which is produced by
this inner product and suppose |||| is some other norm on V . Thus
|x|
_
i
|x
i
|
2
_
1/2
where
x =
k
x
k
v
k
. (4.8)
Show there exist positive numbers < independent of x such that
|x| ||x|| |x|
This is referred to by saying the two norms are equivalent. Hint: The top half is
easy using the Cauchy Schwarz inequality. The bottom half is somewhat harder.
Argue that if it is not so, there exists a sequence {x
k
} such that |x
k
| = 1 but
k
1
|x
k
| = k
1
||x
k
|| and then note the vector of components of x
k
is on
S (0, 1) which was shown to be sequentially compact. Pass to a limit in 4.8 and
use the assumed inequality to get a contradiction to {v
1
, , v
n
} being a basis.
14. It was shown above that in F
n
, the sequentially compact sets are exactly those
which are closed and bounded. Show that in any nite dimensional normed vector
space, V the closed and bounded sets are those which are sequentially compact.
15. Two norms on a nite dimensional vector space, ||||
1
and ||||
2
are said to be
equivalent if there exist positive numbers < such that
||x||
1
||x||
2
||x
1
||
1
.
Show the statement that two norms are equivalent is an equivalence relation.
Explain using the result of Problem 13 why any two norms on a nite dimensional
vector space are equivalent.
4.6. EXERCISES 83
16. A normed vector space, V is separable if there is a countable set {w
k
}
k=1
such
that whenever B(x, ) is an open ball in V, there exists some w
k
in this open ball.
Show that F
n
is separable. This set of points is called a countable dense set.
17. Let V be any normed vector space with norm ||||. Using Problem 13 show that
V is separable.
18. Suppose V is a normed vector space. Show there exists a countable set of open
balls B {B(x
k
, r
k
)}
k=1
having the remarkable property that any open set, U is
the union of some subset of B. This collection of balls is called a countable basis.
Hint: Use Problem 17 to get a countable dense dense set of points, {x
k
}
k=1
and
then consider balls of the form B
_
x
k
,
1
r
_
where r N. Show this collection of
balls is countable and then show it has the remarkable property mentioned.
19. Suppose S is any nonempty set in V a nite dimensional normed vector space.
Suppose C is a set of open sets such that C S. (Such a collection of sets is
called an open cover.) Show using Problem 18 that there are countably many
sets from C, {U
k
}
k=1
such that S
k=1
U
k
. This is called the Lindelo property
when every open cover can be reduced to a countable sub cover.
20. A set, H in a normed vector space is said to be compact if whenever C is a set of
open sets such that C H, there are nitely many sets of C, {U
1
, , U
p
} such
that
H
p
i=1
U
i
.
Show using Problem 19 that if a set in a normed vector space is sequentially
compact, then it must be compact. Next show using Problem 14 that a set in a
normed vector space is compact if and only if it is closed and bounded. Explain
why the sets which are compact, closed and bounded, and sequentially compact
are the same sets in any nite dimensional normed vector space
84 SEQUENCES
Continuous Functions
Continuous functions are dened as they are for a function of one variable.
Denition 5.0.1 Let V, W be normed vector spaces. A function f: D(f )
V W is continuous at x D(f ) if for each > 0 there exists > 0 such that
whenever y D(f ) and
||y x||
V
<
it follows that
||f (x) f (y)||
W
< .
A function, f is continuous if it is continuous at every point of D(f ) .
There is a theorem which makes it easier to verify certain functions are continuous
without having to always go to the above denition. The statement of this theorem is
purposely just a little vague. Some of these things tend to hold in almost any context,
certainly for any normed vector space.
Theorem 5.0.2 The following assertions are valid
1. The function, af+bg is continuous at x when f, g are continuous at x D(f )
D(g) and a, b F.
2. If and f and g have values in F
n
and they are each continuous at x, then f g is
continuous at x. If g has values in F and g (x) = 0 with g continuous, then f /g is
continuous at x.
3. If f is continuous at x, f (x) D(g) , and g is continuous at f (x) ,then g f is
continuous at x.
4. If V is any normed vector space, the function f : V R, given by f (x) = ||x|| is
continuous.
5. f is continuous at every point of V if and only if whenever U is an open set in
W, f
1
(W) is open.
Proof: First consider 1.) Let > 0 be given. By assumption, there exist
1
> 0 such
that whenever |x y| <
1
, it follows |f (x) f (y)| <

2(|a|+|b|+1)
and there exists
2
> 0
such that whenever |x y| <
2
, it follows that |g (x) g (y)| <

2(|a|+|b|+1)
. Then let
0 < min (
1
,
2
) . If |x y| < , then everything happens at once. Therefore, using
the triangle inequality
|af (x) +bf (x) (ag (y) +bg (y))|
85
86 CONTINUOUS FUNCTIONS
|a| |f (x) f (y)| +|b| |g (x) g (y)|
< |a|
_

2 (|a| +|b| + 1)
_
+|b|
_

2 (|a| +|b| + 1)
_
< .
Now consider 2.) There exists
1
> 0 such that if |y x| <
1
, then |f (x) f (y)| <
1. Therefore, for such y,
|f (y)| < 1 +|f (x)| .
It follows that for such y,
|f g (x) f g (y)| |f (x) g (x) g (x) f (y)| +|g (x) f (y) f (y) g (y)|
|g (x)| |f (x) f (y)| +|f (y)| |g (x) g (y)|
(1 +|g (x)| +|f (y)|) [|g (x) g (y)| +|f (x) f (y)|]
(2 +|g (x)| +|f (x)|) [|g (x) g (y)| +|f (x) f (y)|]
Now let > 0 be given. There exists
2
such that if |x y| <
2
, then
|g (x) g (y)| <

2 (2 +|g (x)| +|f (x)|)
,
and there exists
3
such that if |x y| <
3
, then
|f (x) f (y)| <

2 (2 +|g (x)| +|f (x)|)
Now let 0 < min (
1
,
2
,
3
) . Then if |x y| < , all the above hold at once and so
|f g (x) f g (y)|
(2 +|g (x)| +|f (x)|) [|g (x) g (y)| +|f (x) f (y)|]
< (2 +|g (x)| +|f (x)|)
_

2 (2 +|g (x)| +|f (x)|)
+

2 (2 +|g (x)| +|f (x)|)
_
= .
This proves the rst part of 2.) To obtain the second part, let
1
be as described above
and let
0
> 0 be such that for |x y| <
0
,
|g (x) g (y)| < |g (x)| /2
and so by the triangle inequality,
|g (x)| /2 |g (y)| |g (x)| |g (x)| /2
which implies |g (y)| |g (x)| /2, and |g (y)| < 3 |g (x)| /2.
Then if |x y| < min (
0
,
1
) ,
f (x)
g (x)

f (y)
g (y)
f (x) g (y) f (y) g (x)

g (x) g (y)
|f (x) g (y) f (y) g (x)|

_
|g(x)|
2
2
_
=
2 |f (x) g (y) f (y) g (x)|
|g (x)|
2
87
2
|g (x)|
2
[|f (x) g (y) f (y) g (y) +f (y) g (y) f (y) g (x)|]
2
|g (x)|
2
[|g (y)| |f (x) f (y)| +|f (y)| |g (y) g (x)|]
2
|g (x)|
2
_
3
2
|g (x)| |f (x) f (y)| + (1 +|f (x)|) |g (y) g (x)|
_
2
|g (x)|
2
(1 + 2 |f (x)| + 2 |g (x)|) [|f (x) f (y)| +|g (y) g (x)|]
M [|f (x) f (y)| +|g (y) g (x)|]
where M is dened by
M
2
|g (x)|
2
(1 + 2 |f (x)| + 2 |g (x)|)
Now let
2
be such that if |x y| <
2
, then
|f (x) f (y)| <

2
M
1
and let
3
be such that if |x y| <
3
, then
|g (y) g (x)| <

2
M
1
.
Then if 0 < min (
0
,
1
,
2
,
3
) , and |x y| < , everything holds and
f (x)
g (x)

f (y)
g (y)
M [|f (x) f (y)| +|g (y) g (x)|]

< M
_
2
M
1
+

2
M
1
_
= .
This completes the proof of the second part of 2.)
Note that in these proofs no eort is made to nd some sort of best . The problem
is one which has a yes or a no answer. Either it is or it is not continuous.
Now consider 3.). If f is continuous at x, f (x) D(g) , and g is continuous at
f (x) ,then g f is continuous at x. Let > 0 be given. Then there exists > 0 such
that if |y f (x)| < and y D(g) , it follows that |g (y) g (f (x))| < . From
continuity of f at x, there exists > 0 such that if |x z| < and z D(f ) , then
|f (z) f (x)| < . Then if |x z| < and z D(g f ) D(f ) , all the above hold and
so
|g (f (z)) g (f (x))| < .
This proves part 3.)
To verify part 4.), let > 0 be given and let = . Then if ||x y|| < , the triangle
inequality implies
|f (x) f (y)| = |||x|| ||y|||
||x y|| < = .
This proves part 4.)
Next consider 5.) Suppose rst f is continuous. Let U be open and let x f
1
(U) .
This means f (x) U. Since U is open, there exists > 0 such that B(f (x) , ) U. By
continuity, there exists > 0 such that if y B(x, ) , then f (y) B(f (x) , ) and so
this shows B(x,) f
1
(U) which implies f
1
(U) is open since x is an arbitrary point
of f
1
(U) . Next suppose the condition about inverse images of open sets are open. Then
apply this condition to the open set B(f (x) , ) . The condition says f
1
(B(f (x) , )) is
open and since x f
1
(B(f (x) , )) , it follows x is an interior point of f
1
(B(f (x) , ))
so there exists > 0 such that B(x, ) f
1
(B(f (x) , )) . This says f (B(x, ))
B(f (x) , ) . In other words, whenever ||y x|| < , ||f (y) f (x)|| < which is the
condition for continuity at the point x. Since x is arbitrary, This proves the theorem.
5.1 Continuity And The Limit Of A Sequence

There is a very useful way of thinking of continuity in terms of limits of sequences found
in the following theorem. In words, it says a function is continuous if it takes convergent
sequences to convergent sequences whenever possible.
Theorem 5.1.1 A function f : D(f ) W is continuous at x D(f ) if and only
if, whenever x
n
x with x
n
D(f ) , it follows f (x
n
) f (x) .
Proof: Suppose rst that f is continuous at x and let x
n
x. Let > 0 be given.
By continuity, there exists > 0 such that if ||y x|| < , then ||f (x) f (y)|| < .
However, there exists n
such that if n n
, then ||x
n
x|| < and so for all n this
large,
||f (x) f (x
n
)|| <
which shows f (x
n
) f (x) .
Now suppose the condition about taking convergent sequences to convergent se-
quences holds at x. Suppose f fails to be continuous at x. Then there exists > 0 and
x
n
D(f ) such that ||x x
n
|| <
1
n
, yet
||f (x) f (x
n
)|| .
But this is clearly a contradiction because, although x
n
x, f (x
n
) fails to converge to
f (x) . It follows f must be continuous after all. This proves the theorem.
Theorem 5.1.2 Suppose f : D(f ) R is continuous at x D(f ) and suppose
||f (x
n
)|| l ( l)
where {x
n
} is a sequence of points of D(f ) which converges to x. Then
||f (x)|| l ( l) .
Proof: Since ||f (x
n
)|| l and f is continuous at x, it follows from the triangle
inequality, Theorem 4.1.8 and Theorem 5.1.1,
||f (x)|| = lim
n
||f (x
n
)|| l.
The other case is entirely similar. This proves the theorem.
Another very useful idea involves the automatic continuity of the inverse function
under certain conditions.
Theorem 5.1.3 Let K be a sequentially compact set and suppose f : K f (K)
is continuous and one to one. Then f
1
must also be continuous.
5.2. THE EXTREME VALUES THEOREM 89
Proof: Suppose f (k
n
) f (k) . Does it follow k
n
k? If this does not happen,
then there exists > 0 and a subsequence still denoted as {k
n
} such that
|k
n
k| (5.1)
Now since K is compact, there exists a further subsequence, still denoted as {k
n
} such
that
k
n
k
K
However, the continuity of f requires
f (k
n
) f (k
)
and so f (k
) = f (k). Since f is one to one, this requires k
= k, a contradiction to 5.1.
5.2 The Extreme Values Theorem
The extreme values theorem says continuous functions achieve their maximum and
minimum provided they are dened on a sequentially compact set.
The next theorem is known as the max min theorem or extreme value theorem.
Theorem 5.2.1 Let K F
n
be sequentially compact. Thus K is closed and
bounded, and let f : K R be continuous. Then f achieves its maximum and its
minimum on K. This means there exist, x
1
, x
2
K such that for all x K,
f (x
1
) f (x) f (x
2
) .
Proof: Let = sup {f (x) : x K} . Next let {
k
} be an increasing sequence which
converges to but each
k
< . Therefore, for each k, there exists x
k
K such that
f (x
k
) >
k
.
Since K is sequentially compact, there exists a subsequence, {x
k
l
} such that lim
l
x
k
l
=
x K. Then by continuity of f,
f (x) = lim
l
f (x
k
l
) lim
l
k
l
=
which shows f achieves its maximum on K. To see it achieves its minimum, you could
repeat the argument with a minimizing sequence or else you could consider f and
apply what was just shown to f, f having its minimum when f has its maximum.
5.3 Connected Sets
Stated informally, connected sets are those which are in one piece. In order to dene
what is meant by this, I will rst consider what it means for a set to not be in one piece.
Denition 5.3.1 Let A be a nonempty subset of V a normed vector space. Then
A is dened to be the intersection of all closed sets which contain A. Note the whole
space, V is one such closed set which contains A.
Lemma 5.3.2 Let A be a nonempty set in a normed vector space V. Then A is a
closed set and
A = A A
where A
denotes the set of limit points of A.

Proof: First of all, denote by C the set of closed sets which contain A. Then
A = C
and this will be closed if its complement is open. However,
A
C
=
_
H
C
: H C
_
.
Each H
C
is open and so the union of all these open sets must also be open. This is
because if x is in this union, then it is in at least one of them. Hence it is an interior
point of that one. But this implies it is an interior point of the union of them all which
is an even larger set. Thus A is closed.
The interesting part is the next claim. First note that from the denition, A A so
if x A, then x A. Now consider y A
but y / A. If y / A, a closed set, then there

exists B(y, r) A
C
. Thus y cannot be a limit point of A, a contradiction. Therefore,
A A
A
Next suppose x A and suppose x / A. Then if B(x, r) contains no points of A
dierent than x, since x itself is not in A, it would follow that B(x,r) A = and so
recalling that open balls are open, B(x, r)
C
is a closed set containing A so from the
denition, it also contains A which is contrary to the assertion that x A. Hence if
x / A, then x A
and so
A A
A
Now that the closure of a set has been dened it is possible to dene what is meant
by a set being separated.
Denition 5.3.3 A set, S in a normed vector space is separated if there exist
sets A, B such that
S = A B, A, B = , and A B = B A = .
In this case, the sets A and B are said to separate S. A set is connected if it is not
separated. Remember A denotes the closure of the set A.
Note that the concept of connected sets is dened in terms of what it is not. This
makes it somewhat dicult to understand. One of the most important theorems about
connected sets is the following.
Theorem 5.3.4 Suppose U and V are connected sets having nonempty intersec-
tion. Then U V is also connected.
Proof: Suppose U V = AB where AB = BA = . Consider the sets AU
and B U. Since
(A U) (B U) = (A U)
_
B U
_
= ,
It follows one of these sets must be empty since otherwise, U would be separated. It
follows that U is contained in either A or B. Similarly, V must be contained in either
A or B. Since U and V have nonempty intersection, it follows that both V and U are
contained in one of the sets A, B. Therefore, the other must be empty and this shows
U V cannot be separated and is therefore, connected.
5.3. CONNECTED SETS 91
The intersection of connected sets is not necessarily connected as is shown by the
following picture.
U
V
Theorem 5.3.5 Let f : X Y be continuous where Y is a normed vector space
and X is connected. Then f (X) is also connected.
Proof: To do this you show f (X) is not separated. Suppose to the contrary that
f (X) = A B where A and B separate f (X) . Then consider the sets f
1
(A) and
f
1
(B) . If z f
1
(B) , then f (z) B and so f (z) is not a limit point of A. Therefore,
there exists an open set, U containing f (z) such that UA = . But then, the continuity
of f and Theorem 5.0.2 implies that f
1
(U) is an open set containing z such that
f
1
(U) f
1
(A) = . Therefore, f
1
(B) contains no limit points of f
1
(A) . Similar
reasoning implies f
1
(A) contains no limit points of f
1
(B). It follows that X is
separated by f
1
(A) and f
1
(B) , contradicting the assumption that X was connected.
An arbitrary set can be written as a union of maximal connected sets called con-
nected components. This is the concept of the next denition.
Denition 5.3.6 Let S be a set and let p S. Denote by C
p
the union of
all connected subsets of S which contain p. This is called the connected component
determined by p.
Theorem 5.3.7 Let C
p
be a connected component of a set S in a normed vector
space. Then C
p
is a connected set and if C
p
C
q
= , then C
p
= C
q
.
Proof: Let C denote the connected subsets of S which contain p. If C
p
= A B
where
A B = B A = ,
then p is in one of A or B. Suppose without loss of generality p A. Then every set
of C must also be contained in A since otherwise, as in Theorem 5.3.4, the set would
be separated. But this implies B is empty. Therefore, C
p
is connected. From this, and
Theorem 5.3.4, the second assertion of the theorem is proved.
This shows the connected components of a set are equivalence classes and partition
the set.
A set, I is an interval in R if and only if whenever x, y I then (x, y) I. The
following theorem is about the connected sets in R.
Theorem 5.3.8 A set, C in R is connected if and only if C is an interval.
Proof: Let C be connected. If C consists of a single point, p, there is nothing
to prove. The interval is just [p, p] . Suppose p < q and p, q C. You need to show
(p, q) C. If
x (p, q) \ C
let C (, x) A, and C (x, ) B. Then C = A B and the sets A and B
separate C contrary to the assumption that C is connected.
Conversely, let I be an interval. Suppose I is separated by A and B. Pick x A
and y B. Suppose without loss of generality that x < y. Now dene the set,
S {t [x, y] : [x, t] A}
and let l be the least upper bound of S. Then l A so l / B which implies l A. But
if l / B, then for some > 0,
(l, l +) B =
contradicting the denition of l as an upper bound for S. Therefore, l B which implies
l / A after all, a contradiction. It follows I must be connected.
This yields a generalization of the intermediate value theorem from one variable
calculus.
Corollary 5.3.9 Let E be a connected set in a normed vector space and suppose
f : E R and that y (f (e
1
) , f (e
2
)) where e
i
E. Then there exists e E such that
f (e) = y.
Proof: From Theorem 5.3.5, f (E) is a connected subset of R. By Theorem 5.3.8
f (E) must be an interval. In particular, it must contain y. This proves the corollary.
The following theorem is a very useful description of the open sets in R.

Theorem 5.3.10 Let U be an open set in R. Then there exist countably many
disjoint open sets {(a
i
, b
i
)}
i=1
such that U =
i=1
(a
i
, b
i
) .
Proof: Let p U and let z C
p
, the connected component determined by p. Since
U is open, there exists, > 0 such that (z , z +) U. It follows from Theorem
5.3.4 that
(z , z +) C
p
.
This shows C
p
is open. By Theorem 5.3.8, this shows C
p
is an open interval, (a, b)
where a, b [, ] . There are therefore at most countably many of these connected
components because each must contain a rational number and the rational numbers are
countable. Denote by {(a
i
, b
i
)}
i=1
the set of these connected components. This proves
the theorem.
Denition 5.3.11 A set E in a normed vector space is arcwise connected if for
any two points, p, q E, there exists a closed interval, [a, b] and a continuous function,
: [a, b] E such that (a) = p and (b) = q.
An example of an arcwise connected topological space would be any subset of R
n
which is the continuous image of an interval. Arcwise connected is not the same as
connected. A well known example is the following.
__
x, sin
1
x
_
: x (0, 1]
_
{(0, y) : y [1, 1]} (5.2)
You can verify that this set of points in the normed vector space R
2
is not arcwise
connected but is connected.
Lemma 5.3.12 In a normed vector space, B(z,r) is arcwise connected.
5.3. CONNECTED SETS 93
Proof: This is easy from the convexity of the set. If x, y B(z, r) , then let
(t) = x +t (y x) for t [0, 1] .
||x +t (y x) z|| = ||(1 t) (x z) +t (y z)||
(1 t) ||x z|| +t ||y z||
< (1 t) r +tr = r
showing (t) stays in B(z, r).
Proposition 5.3.13 If X is arcwise connected, then it is connected.
Proof: Let X be an arcwise connected set and suppose it is separated. Then
X = AB where A, B are two separated sets. Pick p A and q B. Since X is given
to be arcwise connected, there must exist a continuous function : [a, b] X such
that (a) = p and (b) = q. But then ([a, b]) = ( ([a, b]) A) ( ([a, b]) B) and
the two sets ([a, b]) A and ([a, b]) B are separated thus showing that ([a, b]) is
separated and contradicting Theorem 5.3.8 and Theorem 5.3.5. It follows that X must
be connected as claimed.
Theorem 5.3.14 Let U be an open subset of a normed vector space. Then U
is arcwise connected if and only if U is connected. Also the connected components of an
open set are open sets.
Proof: By Proposition 5.3.13 it is only necessary to verify that if U is connected
and open in the context of this theorem, then U is arcwise connected. Pick p U. Say
x U satises P if there exists a continuous function, : [a, b] U such that (a) = p
and (b) = x.
A {x U such that x satises P.}
If x A, then Lemma 5.3.12 implies B(x, r) U is arcwise connected for small
enough r. Thus letting y B(x, r) , there exist intervals, [a, b] and [c, d] and continuous
functions having values in U, , such that (a) = p, (b) = x, (c) = x, and (d) =
y. Then let
1
: [a, b +d c] U be dened as
1
(t)
_
(t) if t [a, b]
(t +c b) if t [b, b +d c]
Then it is clear that
1
is a continuous function mapping p to y and showing that
B(x, r) A. Therefore, A is open. A = because since U is open there is an open set,
B(p, ) containing p which is contained in U and is arcwise connected.
Now consider B U \ A. I claim this is also open. If B is not open, there exists
a point z B such that every open set containing z is not contained in B. Therefore,
letting B(z, ) be such that z B(z, ) U, there exist points of A contained in
B(z, ) . But then, a repeat of the above argument shows z A also. Hence B is open
and so if B = , then U = B A and so U is separated by the two sets B and A
contradicting the assumption that U is connected.
It remains to verify the connected components are open. Let z C
p
where C
p
is
the connected component determined by p. Then picking B(z, ) U, C
p
B(z, )
is connected and contained in U and so it must also be contained in C
p
. Thus z is an
interior point of C
p
As an application, consider the following corollary.
Corollary 5.3.15 Let f : Z be continuous where is a connected open set in
a normed vector space. Then f must be a constant.
Proof: Suppose not. Then it achieves two dierent values, k and l = k. Then
= f
1
(l) f
1
({m Z : m = l}) and these are disjoint nonempty open sets which
separate . To see they are open, note
f
1
({m Z : m = l}) = f
1
_
m=l
_
m
1
6
, m+
1
6
__
which is the inverse image of an open set while f
1
(l) = f
1
__
l
1
6
, l +
1
6
__
also an
open set.
5.4 Uniform Continuity
The concept of uniform continuity is also similar to the one dimensional concept.
Denition 5.4.1 Let f be a function. Then f is uniformly continuous if for
every > 0, there exists a depending only on such that if ||x y|| < then
||f (x) f (y)|| < .
Theorem 5.4.2 Let f : K F be continuous where K is a sequentially compact
set in F
n
or more generally a normed vector space. Then f is uniformly continuous on
K.
Proof: If this is not true, there exists > 0 such that for every > 0 there exists a
pair of points, x
and y
such that even though ||x
|| < , ||f (x
) f (y
)|| .
Taking a succession of values for equal to 1, 1/2, 1/3, , and letting the exceptional
pair of points for = 1/n be denoted by x
n
and y
n
,
||x
n
y
n
|| <
1
n
, ||f (x
n
) f (y
n
)|| .
Now since K is sequentially compact, there exists a subsequence, {x
n
k
} such that x
n
k

z K. Now n
k
k and so
||x
n
k
y
n
k
|| <
1
k
.
Hence
||y
n
k
z|| ||y
n
k
x
n
k
|| +||x
n
k
z||
<
1
k
+||x
n
k
z||
Consequently, y
n
k
z also. By continuity of f and Theorem 5.1.2,
0 = ||f (z) f (z)|| = lim
k
||f (x
n
k
) f (y
n
k
)|| ,
an obvious contradiction. Therefore, the theorem must be true.
Recall the closed and bounded subsets of F
n
are those which are sequentially com-
pact.
5.5 Sequences And Series Of Functions
Now it is an easy matter to consider sequences of vector valued functions.
Denition 5.5.1 A sequence of functions is a map dened on N or some set of
integers larger than or equal to a given integer, m which has values which are functions.
It is written in the form {f
n
}
n=m
where f
n
is a function. It is assumed also that the
domain of all these functions is the same.
5.5. SEQUENCES AND SERIES OF FUNCTIONS 95
Here the functions have values in some normed vector space.
The denition of uniform convergence is exactly the same as earlier only now it is
not possible to draw representative pictures so easily.
Denition 5.5.2 Let {f
n
} be a sequence of functions. Then the sequence con-
verges pointwise to a function f if for all x D, the domain of the functions in the
sequence,
f (x) = lim
n
f
n
(x)
Thus you consider for each x D the sequence of numbers {f
n
(x)} and if this
sequence converges for each x D, the thing it converges to is called f (x).
n
} be a sequence of functions dened on D. Then {f
n
}
is said to converge uniformly to f if it converges pointwise to f and for every > 0 there
exists N such that for all n N
||f (x) f
n
(x)|| <
for all x D.
Theorem 5.5.4 Let {f
n
} be a sequence of continuous functions dened on D
and suppose this sequence converges uniformly to f . Then f is also continuous on D. If
each f
n
is uniformly continuous on D, then f is also uniformly continuous on D.
Proof: Let > 0 be given and pick z D. By uniform convergence, there exists N
such that if n > N, then for all x D,
||f (x) f
n
(x)|| < /3. (5.3)
Pick such an n. By assumption, f
n
is continuous at z. Therefore, there exists > 0 such
that if ||z x|| < then
||f
n
(x) f
n
(z)|| < /3.
It follows that for ||x z|| < ,
||f (x) f (z)|| ||f (x) f
n
(x)|| +||f
n
(x) f
n
(z)|| +||f
n
(z) f (z)||
< /3 +/3 +/3 =
which shows that since was arbitrary, f is continuous at z.
In the case where each f
n
is uniformly continuous, and using the same f
n
for which
5.3 holds, there exists a > 0 such that if ||y z|| < , then
||f
n
(z) f
n
(y)|| < /3.
Then for ||y z|| < ,
||f (y) f (z)|| ||f (y) f
n
(y)|| +||f
n
(y) f
n
(z)|| +||f
n
(z) f (z)||
< /3 +/3 +/3 =
This shows uniform continuity of f . This proves the theorem.
n
} be a sequence of functions dened on D. Then the
sequence is said to be uniformly Cauchy if for every > 0 there exists N such that
whenever m, n N,
||f
m
(x) f
n
(x)|| <
for all x D.
Then the following theorem follows easily.
n
} be a uniformly Cauchy sequence of functions dened
on D having values in a complete normed vector space such as F
n
for example. Then
there exists f dened on D such that {f
n
} converges uniformly to f .
Proof: For each x D, {f
n
(x)} is a Cauchy sequence. Therefore, it converges to
some vector f (x). Let > 0 be given and let N be such that if n, m N,
||f
m
(x) f
n
(x)|| < /2
for all x D. Then for any x D, pick n N and it follows from Theorem 4.1.8
||f (x) f
n
(x)|| = lim
m
||f
m
(x) f
n
(x)|| /2 < .
Corollary 5.5.7 Let {f
n
} be a uniformly Cauchy sequence of functions continuous
on D having values in a complete normed vector space like F
n
. Then there exists f
dened on D such that {f
n
} converges uniformly to f and f is continuous. Also, if each
f
n
is uniformly continuous, then so is f .
Proof: This follows from Theorem 5.5.6 and Theorem 5.5.4. This proves the corol-
lary.
Here is one more fairly obvious theorem.
n
} be a sequence of functions dened on D having values
in a complete normed vector space like F
n
. Then it converges pointwise if and only if
the sequence {f
n
(x)} is a Cauchy sequence for every x D. It converges uniformly if
and only if {f
n
} is a uniformly Cauchy sequence.
Proof: If the sequence converges pointwise, then by Theorem 4.4.3 the sequence
{f
n
(x)} is a Cauchy sequence for each x D. Conversely, if {f
n
(x)} is a Cauchy se-
quence for each x D, then {f
n
(x)} converges for each x D because of completeness.
Now suppose {f
n
} is uniformly Cauchy. Then from Theorem 5.5.6 there exists f
such that {f
n
} converges uniformly on D to f . Conversely, if {f
n
} converges uniformly
to f on D, then if > 0 is given, there exists N such that if n N,
|f (x) f
n
(x)| < /2
for every x D. Then if m, n N and x D,
|f
n
(x) f
m
(x)| |f
n
(x) f (x)| +|f (x) f
m
(x)| < /2 +/2 = .
Thus {f
n
} is uniformly Cauchy.
Once you understand sequences, it is no problem to consider series.
n
} be a sequence of functions dened on D. Then
_

k=1
f
k
_
(x) lim
n
n
k=1
f
k
(x) (5.4)
whenever the limit exists. Thus there is a new function denoted by
k=1
f
k
(5.5)
5.5. SEQUENCES AND SERIES OF FUNCTIONS 97
and its value at x is given by the limit of the sequence of partial sums in 5.4. If for all
x D, the limit in 5.4 exists, then 5.5 is said to converge pointwise.
k=1
f
k
is said
to converge uniformly on D if the sequence of partial sums,
_
n
k=1
f
k
_
converges uniformly.If the indices for the functions start at some other value than 1,
you make the obvious modication to the above denition.
n
} be a sequence of functions dened on D which have
values in a complete normed vector space like F
n
. The series
k=1
f
k
converges pointwise
if and only if for each > 0 and x D, there exists N
,x
which may depend on x as
well as such that when q > p N
,x
,
k=p
f
k
(x)
<
The series
k=1
f
k
converges uniformly on D if for every > 0 there exists N
such
that if q > p N then
k=p
f
k
(x)
< (5.6)
for all x D.
Proof: The rst part follows from Theorem 5.5.8. The second part follows from
observing the condition is equivalent to the sequence of partial sums forming a uniformly
Cauchy sequence and then by Theorem 5.5.6, these partial sums converge uniformly to
a function which is the denition of
k=1
f
k
Is there an easy way to recognize when 5.6 happens? Yes, there is. It is called the
Weierstrass M test.
n
} be a sequence of functions dened on D having val-
ues in a complete normed vector space like F
n
. Suppose there exists M
n
such that
sup {|f
n
(x)| : x D} < M
n
and
n=1
M
n
converges. Then
n=1
f
n
converges uni-
formly on D.
Proof: Let z D. Then letting m < n and using the triangle inequality
k=1
f
k
(z)
m
k=1
f
k
(z)
k=m+1
||f
k
(z)||
k=m+1
M
k
<
whenever m is large enough because of the assumption that
n=1
M
n
converges. There-
fore, the sequence of partial sums is uniformly Cauchy on D and therefore, converges
uniformly to
k=1
f
k
on D. This proves the theorem.
Theorem 5.5.12 If {f
n
} is a sequence of continuous functions dened on D
and
k=1
f
k
converges uniformly, then the function,
k=1
f
k
must also be continuous.
Proof: This follows from Theorem 5.5.4 applied to the sequence of partial sums of
the above series which is assumed to converge uniformly to the function,
k=1
f
k
.
5.6 Polynomials
General considerations about what a function is have already been considered earlier.
For functions of one variable, the special kind of functions known as a polynomial has
a corresponding version when one considers a function of many variables. This is found
in the next denition.
Denition 5.6.1 Let be an n dimensional multi-index. This means
= (
1
, ,
n
)
where each
i
is a positive integer or zero. Also, let
||
n
i=1
|
i
|
Then x
means
x
x
1
1
x
2
2
x
n
3
where each x
j
F. An n dimensional polynomial of degree m is a function of the form
p (x) =
||m
d
.
where the d
are complex or real numbers. Rational functions are dened as the quotient
of two polynomials. Thus these functions are dened on F
n
.
For example, f (x) = x
1
x
2
2
+ 7x
4
3
x
1
is a polynomial of degree 5 and
x
1
x
2
2
+ 7x
4
3
x
1
+x
3
2
4x
3
1
x
2
2
+ 7x
2
3
x
1
x
3
2
is a rational function.
Note that in the case of a rational function, the domain of the function might not
be all of F
n
. For example, if
f (x) =
x
1
x
2
2
+ 7x
4
3
x
1
+x
3
2
x
2
2
+ 3x
2
1
4
,
the domain of f would be all complex numbers such that x
2
2
+ 3x
2
1
= 4.
By Theorem 5.0.2 all polynomials are continuous. To see this, note that the function,
k
(x) x
k
is a continuous function because of the inequality
|
k
(x)
k
(y)| = |x
k
y
k
| |x y| .
Polynomials are simple sums of scalars times products of these functions. Similarly,
by this theorem, rational functions, quotients of polynomials, are continuous at points
where the denominator is non zero. More generally, if V is a normed vector space,
consider a V valued function of the form
f (x)
||m
d
where d
V , sort of a V valued polynomial. Then such a function is continuous

by application of Theorem 5.0.2 and the above observation about the continuity of the
functions
k
.
Thus there are lots of examples of continuous functions. However, it is even better
than the above discussion indicates. As in the case of a function of one variable, an
arbitrary continuous function can typically be approximated uniformly by a polynomial.
This is the n dimensional version of the Weierstrass approximation theorem.
5.7. SEQUENCES OF POLYNOMIALS, WEIERSTRASS APPROXIMATION 99
5.7 Sequences Of Polynomials, Weierstrass Approx-
imation
Just as an arbitrary continuous function dened on an interval can be approximated
uniformly by a polynomial, there exists a similar theorem which is just a generalization
of the earlier one which will hold for continuous functions dened on a box or more
generally a closed and bounded set. The proof is based on the following lemma.
Lemma 5.7.1 The following estimate holds for x [0, 1] and m 2.
m
k=0
_
m
k
_
(k mx)
2
x
k
(1 x)
mk
1
4
m
Proof: First of all, from the binomial theorem
m
k=0
_
m
k
_
(tx)
k
(1 x)
mk
= (1 x +tx)
m
Take a derivative and then let t = 1.
m
k=0
_
m
k
_
k (tx)
k1
x(1 x)
mk
= mx(tx x + 1)
m1
m
k=0
_
m
k
_
k (x)
k
(1 x)
mk
= mx
Then also,
m
k=0
_
m
k
_
k (tx)
k
(1 x)
mk
= mxt (tx x + 1)
m1
Take another time derivative of both sides.
m
k=0
_
m
k
_
k
2
(tx)
k1
x(1 x)
mk
= mx
_
(tx x + 1)
m1
tx(tx x + 1)
m2
+mtx(tx x + 1)
m2
_
Plug in t = 1.
m
k=0
_
m
k
_
k
2
x
k
(1 x)
mk
= mx(mx x + 1)
Then it follows
m
k=0
_
m
k
_
(k mx)
2
x
k
(1 x)
mk
=
m
k=0
_
m
k
_
_
k
2
2kmx +x
2
m
2
_
x
k
(1 x)
mk
and from what was just shown along with the binomial theorem again, this equals
x
2
m
2
x
2
m+mx 2mx(mx) +x
2
m
2
= x
2
m+mx =
m
4
m
_
x
1
2
_
2
.
Thus the expression is maximized when x = 1/2 and yields m/4 in this case. This
proves the lemma.
Now let f be a continuous function dened on [0, 1] . Let p
n
be the polynomial
dened by
p
n
(x)
n
k=0
_
n
k
_
f
_
k
n
_
x
k
(1 x)
nk
. (5.7)
Now for f a continuous function dened on [0, 1]
n
and for x = (x
1
, , x
n
) ,consider
the polynomial,
p
m
(x)
m1
k1=0

mn
kn=0
_
m
1
k
1
__
m
2
k
2
_

_
m
n
k
n
_
x
k1
1
(1 x
1
)
m1k1
x
k2
2
(1 x
2
)
m2k2
x
kn
n
(1 x
n
)
mnkn
f
_
k
1
m
1
, ,
k
n
m
n
_
. (5.8)
Also dene if I is a set in R
n
||h||
I
sup {|h(x)| : x I} .
Let
min (m) min {m
1
, , m
n
} , max (m) max {m
1
, , m
n
}
Denition 5.7.2 Dene p
m
converges uniformly to f on a set, I if
lim
min(m)
||p
m
f||
I
= 0.
To simplify the notation, let k = (k
1
, , k
n
) where each k
i
[0, m
i
],
k
m

_
k1
m1
, ,
kn
mn
_
,
and let
_
m
k
_
_
m
1
k
1
__
m
2
k
2
_

_
m
n
k
n
_
.
Also dene for k = (k
1
, , k
n
)
k m if 0 k
i
m
i
for each i
x
k
(1 x)
mk
x
k1
1
(1 x
1
)
m1k1
x
k2
2
(1 x
2
)
m2k2
x
kn
n
(1 x
n
)
mnkn
.
Thus in terms of this notation,
p
m
(x) =
km
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
This is the n dimensional version of the Bernstein polynomials which is what results in
the case where n = 1. Also
k
m

_
k
1
m
1
, ,
k
n
m
n
_
.
Lemma 5.7.3 For x [0, 1]
n
, f a continuous F valued function dened on [0, 1]
n
,
and p
m
given in 5.8, p
m
converges uniformly to f on [0, 1]
n
as m . More generally,
one can have f a continuous function with values in an arbitrary real or complex normed
linear space. There is no change in the conclusions. You just write instead of ||.
Proof: The function f is uniformly continuous because it is continuous on a se-
quentially compact set [0, 1]
n
. Therefore, there exists > 0 such that if |x y| < ,
then
|f (x) f (y)| < .
Denote by G the set of k such that (k
i
m
i
x
i
)
2
<
2
m
2
for each i where = /
n.
Note this condition is equivalent to saying that for each i,
ki
mi
x
i
< and
k
m
x
<
A short computation shows that by the binomial theorem,
km
_
m
k
_
x
k
(1 x)
mk
= 1
and so for x [0, 1]
n
,
|p
m
(x) f (x)|
km
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
f (x)
kG
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
f (x)
kG
C
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
f (x)
(5.9)
Now for k G it follows that for each i
k
i
m
i
x
i
<

n
(5.10)
and so
f
_
k
m
_
f (x)
< because the above implies
k
m
x
< . Therefore, the rst

sum on the right in 5.9 is no larger than
kG
_
m
k
_
x
k
(1 x)
mk

km
_
m
k
_
x
k
(1 x)
mk
= .
Letting M max {|f (x)| : x [0, 1]
n
} it follows that for some j,
k
j
m
j
x
j
n
, (k
j
m
j
x
j
)
2
m
2
j
2
n
by Lemma 5.7.1,
|p
m
(x) f (x)|
+ 2M
kG
C
_
m
k
_
x
k
(1 x)
mk
+ 2Mn
kG
C
_
m
k
_
(k
j
m
j
x
j
)
2
2
m
2
j
x
k
(1 x)
mk
+ 2Mn
1
2
m
2
j
1
4
m
j
= +
1
2
M
n
2
m
j
+
1
2
M
n
2
min (m)
(5.11)
Therefore, since the right side does not depend on x, it follows that
||p
m
f||
[0,1]
n 2
and since is arbitrary, this shows lim
min(m)
||p
m
f||
[0,1]
n = 0. This proves the
lemma.
These Bernstein polynomials are very remarkable approximations. It turns out that
if f is C
1
([0, 1]
n
) , then
lim
min(m)
p
mxi
(x) f
xi
(x) uniformly on [0, 1]
n
.
We show this rst for the case that n = 1. From this, it is obvious for the general case.
Lemma 5.7.4 Let f C
1
([0, 1]) and let
p
m
(x)
m
k=0
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
be the m
th
Bernstein polynomial. Then in addition to p
m
f
[0,1]
0, it also follows
that
p
m
f
[0,1]
0
Proof: From simple computations,
p
m
(x) =
m
k=1
_
m
k
_
kx
k1
(1 x)
mk
f
_
k
m
_
m1
k=0
_
m
k
_
x
k
(mk) (1 x)
m1k
f
_
k
m
_
=
m
k=1
m(m1)!
(mk)! (k 1)!
x
k1
(1 x)
mk
f
_
k
m
_
m1
k=0
_
m
k
_
x
k
(mk) (1 x)
m1k
f
_
k
m
_
=
m1
k=0
m(m1)!
(m1 k)!k!
x
k
(1 x)
m1k
f
_
k + 1
m
_
m1
k=0
m(m1)!
(m1 k)!k!
x
k
(1 x)
m1k
f
_
k
m
_
=
m1
k=0
m(m1)!
(m1 k)!k!
x
k
(1 x)
m1k
_
f
_
k + 1
m
_
f
_
k
m
__
=
m1
k=0
_
m1
k
_
x
k
(1 x)
m1k
_
f
_
k+1
m
_
f
_
k
m
_
1/m
_
By the mean value theorem,
f
_
k+1
m
_
f
_
k
m
_
1/m
= f
(x
k,m
) , x
k,m

_
k
m
,
k + 1
m
_
Now the desired result follows as before from the uniform continuity of f
on [0, 1]. Let

> 0 be such that if
|x y| < , then |f
(x) f
(y)| <
and let m be so large that 1/m < /2. Then if
x
k
m
< /2, it follows that

|x x
k,m
| < and so
|f
(x) f
(x
k,m
)| =
(x)
f
_
k+1
m
_
f
_
k
m
_
1/m
< .
Now as before, letting M |f
(x)| for all x,

|p
m
(x) f
(x)|
m1
k=0
_
m1
k
_
x
k
(1 x)
m1k
|f
(x
k,m
) f
(x)|
{x:|x
k
m
|<
2
}
_
m1
k
_
x
k
(1 x)
m1k
+M
m1
k=0
_
m1
k
_
4 (k mx)
2
m
2
2
x
k
(1 x)
m1k
+ 4M
1
4
m
1
m
2
2
= +M
1
m
2
< 2
whenever m is large enough. Thus this proves uniform convergence.
Now consider the case where n 1. Applying the same manipulations to the sum
which corresponds to the i
th
variable,
p
mxi
(x)
m1
k1=0

mi1
ki=0

mn
kn=0
_
m
1
k
1
_

_
m
i
1
k
i
_

_
m
n
k
n
_
x
k1
1
(1 x
1
)
m1k1

x
ki
i
(1 x
i
)
mi1ki
x
kn
n
(1 x
n
)
mnkn
f
_
k1
m1
,
ki+1
mi
,
kn
mn
_
f
_
k1
m1
,
ki
mi
,
kn
mn
_
1/m
i
By the mean value theorem, the dierence quotient is of the form
f
xi
(x
k,m
) ,
the i
th
component of x
k,m
being between
ki
mi
and
ki+1
mi
. Therefore, a repeat of the above
argument involving splitting the sum into two pieces, one for which k/m is close to x,
hence close to x
k,m
and one for which some k
j
/m
j
is not close to x
j
for some j yields
the same conclusion about uniform convergence on [0, 1]
n
. This has essentially proved
the following lemma.
Lemma 5.7.5 Let f be in C
k
([0, 1]
n
) . Then there exists a sequence of polynomi-
als p
m
(x) such that each partial derivative up to order k converges uniformly to the
corresponding partial derivative of f.
Proof: It was shown above that letting m = (m
1
, m
2
, , m
n
) ,
lim
min(m)
p
m
f
[0,1]
n = 0, lim
min(m)
p
mxi
f
xi
[0,1]
n = 0
for each x
i
. Extending to higher derivatives is just a technical generalization of what
was just shown.
Theorem 5.7.6 Let f be a continuous function dened on
R
n
k=1
[a
k
, b
k
] .
Then there exists a sequence of polynomials {p
m
} converging uniformly to f on R as
min (m) . If f is C
k
(R) , then the partial derivatives of p
m
up to order k converge
uniformly to the corresponding partial derivatives of f.
Proof: Let g
k
: [0, 1] [a
k
, b
k
] be linear, one to one, and onto and let
x = g (y) (g
1
(y
1
) , g
2
(y
2
) , , g
n
(y
n
)) .
Thus g : [0, 1]
n

n
k=1
[a
k
, b
k
] is one to one, onto, and each component function is
linear. Then f g is a continuous function dened on [0, 1]
n
. It follows from Lemma
5.7.3 there exists a sequence of polynomials, {p
m
(y)} each dened on [0, 1]
n
which
converges uniformly to f g on [0, 1]
n
. Therefore,
_
p
m
_
g
1
(x)
__
converges uniformly
to f (x) on R. But
y = (y
1
, , y
n
) =
_
g
1
1
(x
1
) , , g
1
n
(x
n
)
_
and each g
1
k
is linear. Therefore,
_
p
m
_
g
1
(x)
__
is a sequence of polynomials. As to
the partial derivatives, it was shown above that
lim
min(m)
Dp
m
D(f g)
[0,1]
n = 0
Now the chain rule implies that
D
_
p
m
g
1
_
(x) = Dp
m
_
g
1
(x)
_
Dg
1
(x)
Therefore, the following convergences are uniform in x R.
lim
min(m)
D
_
p
m
g
1
_
(x)
= lim
min(m)
Dp
m
_
g
1
(x)
_
Dg
1
(x)
= lim
min(m)
D(f g)
_
g
1
(x)
_
Dg
1
(x)
= lim
min(m)
Df
_
g
_
g
1
(x)
__
Dg
_
g
1
(x)
_
Dg
1
(x)
= Df (x)
The claim about higher order derivatives is more technical but follows in the same way.
There is a more general version of this theorem which is easy to get. It depends on
the Tietze extension theorem, a wonderful little result which is interesting for its own
sake.
5.7.1 The Tietze Extension Theorem
To generalize the Weierstrass approximation theorem I will give a special case of the
Tietze extension theorem, a very useful result in topology. When this is done, it will
be possible to prove the Weierstrass approximation theorem for functions dened on a
closed and bounded subset of R
n
rather than a box.
Lemma 5.7.7 Let S R
n
be a nonempty subset. Dene
dist (x, S) inf {|x y| : y S} .
Then x dist (x, S) is a continuous function satisfying the inequality,
|dist (x, S) dist (y, S)| |x y| . (5.12)
Proof: The continuity of x dist (x, S) is obvious if the inequality 5.12 is estab-
lished. So let x, y R
n
. Without loss of generality, assume dist (x, S) dist (y, S) and
pick z S such that |y z| < dist (y, S) . Then
|dist (x, S) dist (y, S)| = dist (x, S) dist (y, S)
|x z| (|y z| )
|z y| +|x y| |y z| + = |x y| +.
Since is arbitrary, this proves 5.12.
Lemma 5.7.8 Let H, K be two nonempty disjoint closed subsets of R
n
. Then there
exists a continuous function, g : R
n
[1, 1] such that g (H) = 1/3, g (K) =
1/3, g (R
n
) [1/3, 1/3] .
Proof: Let
f (x)
dist (x, H)
dist (x, H) + dist (x, K)
.
The denominator is never equal to zero because if dist (x, H) = 0, then x H because
H is closed. (To see this, pick h
k
B(x, 1/k) H. Then h
k
x and since H is closed,
x H.) Similarly, if dist (x, K) = 0, then x K and so the denominator is never zero
as claimed. Hence f is continuous and from its denition, f = 0 on H and f = 1 on K.
Now let g (x)
2
3
_
f (x)
1
2
_
. Then g has the desired properties.
Denition 5.7.9 For f a real or complex valued bounded continuous function
dened on M R
n
.
||f||
M
sup {|f (x)| : x M} .
Lemma 5.7.10 Suppose M is a closed set in R
n
where R
n
and suppose f : M
[1, 1] is continuous at every point of M. Then there exists a function, g which is dened
and continuous on all of R
n
such that ||f g||
M
<
2
3
, g (R
n
) [1/3, 1/3] .
Proof: Let H = f
1
([1, 1/3]) , K = f
1
([1/3, 1]) . Thus H and K are disjoint
closed subsets of M. Suppose rst H, K are both nonempty. Then by Lemma 5.7.8 there
exists g such that g is a continuous function dened on all of R
n
and g (H) = 1/3,
g (K) = 1/3, and g (R
n
) [1/3, 1/3] . It follows ||f g||
M
< 2/3. If H = , then f
has all its values in [1/3, 1] and so letting g 1/3, the desired condition is obtained.
If K = , let g 1/3. This proves the lemma.
Lemma 5.7.11 Suppose M is a closed set in R
n
and suppose f : M [1, 1] is
continuous at every point of M. Then there exists a function, g which is dened and
continuous on all of R
n
such that g = f on M and g has its values in [1, 1] .
Proof: Using Lemma 5.7.10, let g
1
be such that g
1
(R
n
) [1/3, 1/3] and
||f g
1
||
M

2
3
.
Suppose g
1
, , g
m
have been chosen such that g
j
(R
n
) [1/3, 1/3] and
f
m
i=1
_
2
3
_
i1
g
i
M
<
_
2
3
_
m
. (5.13)
This has been done for m = 1. Then
_
3
2
_
m
_
f
m
i=1
_
2
3
_
i1
g
i
_
M
1
and so
_
3
2
_
m
_
f
m
i=1
_
2
3
_
i1
g
i
_
can play the role of f in the rst step of the proof.
Therefore, there exists g
m+1
dened and continuous on all of R
n
such that its values
are in [1/3, 1/3] and
_
3
2
_
m
_
f
m
i=1
_
2
3
_
i1
g
i
_
g
m+1
2
3
.
Hence
_
f
m
i=1
_
2
3
_
i1
g
i
_
_
2
3
_
m
g
m+1
_
2
3
_
m+1
.
It follows there exists a sequence, {g
i
} such that each has its values in [1/3, 1/3] and
for every m 5.13 holds. Then let
g (x)
i=1
_
2
3
_
i1
g
i
(x) .
It follows
|g (x)|
i=1
_
2
3
_
i1
g
i
(x)
i=1
_
2
3
_
i1
1
3
1
and
_
2
3
_
i1
g
i
(x)
_
2
3
_
i1
1
3
so the Weierstrass M test applies and shows convergence is uniform. Therefore g must
be continuous. The estimate 5.13 implies f = g on M.
The following is the Tietze extension theorem.
Theorem 5.7.12 Let M be a closed nonempty subset of R
n
and let f : M
[a, b] be continuous at every point of M. Then there exists a function, g continuous on
all of R
n
which coincides with f on M such that g (R
n
) [a, b] .
Proof: Let f
1
(x) = 1 +
2
ba
(f (x) b) . Then f
1
satises the conditions of Lemma
5.7.11 and so there exists g
1
: R
n
[1, 1] such that g is continuous on R
n
and equals
f
1
on M. Let g (x) = (g
1
(x) 1)
_
ba
2
_
+b. This works.
With the Tietze extension theorem, here is a better version of the Weierstrass ap-
proximation theorem.
Theorem 5.7.13 Let K be a closed and bounded subset of R
n
and let f : K R
be continuous. Then there exists a sequence of polynomials {p
m
} such that
lim
m
(sup {|f (x) p
m
(x)| : x K}) = 0.
In other words, the sequence of polynomials converges uniformly to f on K.
Proof: By the Tietze extension theorem, there exists an extension of f to a con-
tinuous function g dened on all R
n
such that g = f on K. Now since K is bounded,
there exist intervals, [a
k
, b
k
] such that
K
n
k=1
[a
k
, b
k
] = R
Then by the Weierstrass approximation theorem, Theorem 5.7.6 there exists a sequence
of polynomials {p
m
} converging uniformly to g on R. Therefore, this sequence of poly-
nomials converges uniformly to g = f on K as well. This proves the theorem.
By considering the real and imaginary parts of a function which has values in C one
can generalize the above theorem.
Corollary 5.7.14 Let K be a closed and bounded subset of R
n
and let f : K F
be continuous. Then there exists a sequence of polynomials {p
m
} such that
lim
m
(sup {|f (x) p
m
(x)| : x K}) = 0.
In other words, the sequence of polynomials converges uniformly to f on K.
5.8. THE OPERATOR NORM 107
5.8 The Operator Norm
It is important to be able to measure the size of a linear operator. The most convenient
way is described in the next denition.
Denition 5.8.1 Let V, W be two nite dimensional normed vector spaces hav-
ing norms ||||
V
and ||||
W
respectively. Let L L(V, W) . Then the operator norm of
L, denoted by ||L|| is dened as
||L|| sup {||Lx||
W
: ||x||
V
1} .
Then the following theorem discusses the main properties of this norm. In the future,
I will dispense with the subscript on the symbols for the norm because it is clear from
the context which norm is meant. Here is a useful lemma.
Lemma 5.8.2 Let V be a normed vector space having a basis {v
1
, , v
n
}. Let
A =
_
a F
n
:
k=1
a
k
v
k
1
_
where a = (a
1
, , a
n
) . Then A is a closed and bounded subset of F
n
.
Proof: First suppose a / A. Then
k=1
a
k
v
k
> 1.
Then for b = (b
1
, , b
n
) , and using the triangle inequality,
k=1
b
k
v
k
k=1
(a
k
(a
k
b
k
)) v
k
k=1
a
k
v
k
k=1
|a
k
b
k
| ||v
k
||
and now it is apparent that if |a b| is suciently small so that each |a
k
b
k
| is small
enough, this expression is larger than 1. Thus there exists > 0 such that B(a, ) A
C
showing that A
C
is open. Therefore, A is closed.
Next consider the claim that A is bounded. Suppose this is not so. Then there exists
a sequence {a
k
} of points of A,
a
k
=
_
a
1
k
, , a
n
k
_
,
such that lim
k
|a
k
| = . Then from the denition of A,
j=1
a
j
k
|a
k
|
v
j
1
|a
k
|
. (5.14)
Let
b
k
=
_
a
1
k
|a
k
|
, ,
a
n
k
|a
k
|
_
Then |b
k
| = 1 so b
k
is contained in the closed and bounded set, S (0, 1) which is
sequentially compact in F
n
. It follows there exists a subsequence, still denoted by {b
k
}
such that it converges to b S (0,1) . Passing to the limit in 5.14 using the following
inequality,
j=1
a
j
k
|a
k
|
v
j

n
j=1
b
j
v
j
j=1
a
j
k
|a
k
|
b
j
||v
j
||
to see that the sum converges to
n
j=1
b
j
v
j
, it follows
n
j=1
b
j
v
j
= 0
and this is a contradiction because {v
1
, , v
n
} is a basis and not all the b
j
can equal
zero. Therefore, A must be bounded after all. This proves the lemma.
Theorem 5.8.3 The operator norm has the following properties.
1. ||L|| <
2. For all x X, ||Lx|| ||L|| ||x|| and if L L(V, W) while M L(W, Z) , then
||ML|| ||M|| ||L||.
3. |||| is a norm. In particular,
(a) ||L|| 0 and ||L|| = 0 if and only if L = 0, the linear transformation which
sends every vector to 0.
(b) ||aL|| = |a| ||L|| whenever a F
(c) ||L +M|| ||L|| +||M||
4. If L L(V, W) for V, W normed vector spaces, L is continuous, meaning that
L
1
(U) is open whenever U is an open set in W.
Proof: First consider 1.). Let A be as in the above lemma. Then
||L|| sup
_
_
_
L
_
_
n
j=1
a
j
v
j
_
_
: a A
_
_
_
= sup
_
_
_
j=1
a
j
L(v
j
)
: a A
_
_
_
<
because a
n
j=1
a
j
L(v
j
)
is a real valued continuous function dened on a sequen-

tially compact set and so it achieves its maximum.
Next consider 2.). If x = 0 there is nothing to show. Assume x = 0. Then from the
denition of ||L|| ,
L
_
x
||x||
_
||L||
and so, since L is linear, you can multiply on both sides by ||x|| and conclude
||L(x)|| ||L|| ||x|| .
For the other claim,
||ML|| sup {||ML(x)|| : ||x|| 1}
||M|| sup {||Lx|| : ||x|| 1} ||M|| ||L|| .
5.8. THE OPERATOR NORM 109
Finally consider 3.) If ||L|| = 0 then from 2.), ||Lx|| 0 and so Lx = 0 for every x
which is the same as saying L = 0. If Lx = 0 for every x, then L = 0 by denition. Let
a F. Then from the properties of the norm, in the vector space,
||aL|| sup {||aLx|| : ||x|| 1}
= sup {|a| ||Lx|| : ||x|| 1}
= |a| sup {||Lx|| : ||x|| 1} |a| ||L||
Finally consider the triangle inequality.
||L +M|| sup {||Lx +Mx|| : ||x|| 1}
sup {||Mx|| +||Lx|| : ||x|| 1}
sup {||Lx|| : ||x|| 1} + sup {||Mx|| : ||x|| 1}
because ||Lx|| sup {||Lx|| : ||x|| 1} with a similar inequality holding for M. There-
fore, by denition,
||L +M|| ||L|| +||M|| .
Finally consider 4.). Let L L(V, W) and let U be open in W and v L
1
(U) .
Thus since U is open, there exists > 0 such that
L(v) B(L(v) , ) U.
Then if w V,
||L(v w)|| = ||L(v) L(w)|| ||L|| ||v w||
and so if ||v w|| is suciently small, ||v w|| < / ||L|| , then L(w) B(L(v) , )
which shows B(v, / ||L||) L
1
(U) and since v L
1
(U) was arbitrary, this shows
L
1
(U) is open. This proves the theorem.
The operator norm will be very important in the chapter on the derivative.
Part 1.) of Theorem 5.8.3 says that if L L(V, W) where V and W are two normed
vector spaces, then there exists K such that for all v V,
||Lv||
W
K||v||
V
An obvious case is to let L = id, the identity map on V and let there be two dierent
norms on V, ||||
1
and ||||
2
. Thus (V, ||||
1
) is a normed vector space and so is (V, ||||
2
) .
Then Theorem 5.8.3 implies that
||v||
2
= ||id (v)||
2
K
2
||v||
1
(5.15)
while the same reasoning implies there exists K
1
such that
||v||
1
K
1
||v||
2
. (5.16)
This leads to the following important theorem.
Theorem 5.8.4 Let V be a nite dimensional vector space and let ||||
1
and ||||
2
be two norms for V. Then these norms are equivalent which means there exist constants,
, such that for all v V
||v||
1
||v||
2
||v||
1
A set, K is sequentially compact if and only if it is closed and bounded. Also every nite
dimensional normed vector space is complete. Also any closed and bounded subset of a
nite dimensional normed vector space is sequentially compact.
Proof: From 5.15 and 5.16
||v||
1
K
1
||v||
2
K
1
K
2
||v||
1
and so
1
K
1
||v||
1
||v||
2
K
2
||v||
1
.
Next consider the claim that all closed and bounded sets in a normed vector space
are sequentially compact. Let L : F
n
V be dened by
L(a)
n
k=1
a
k
v
k
where {v
1
, , v
n
} is a basis for V . Thus L L(F
n
, V ) and so by Theorem 5.8.3 this
is a continuous function. Hence if K is a closed and bounded subset of V it follows
L
1
(K) = F
n
\ L
1
_
K
C
_
= F
n
\ (an open set) = a closed set.
Also L
1
(K) is bounded. To see this, note that L
1
is one to one onto V and so
L
1
L(V, F
n
). Therefore,
L
1
(v)
L
1
||v||
L
1
r
where K B(0, r) . Since K is bounded, such an r exists. Thus L
1
(K) is a closed
and bounded subset of F
n
and is therefore sequentially compact. It follows that if
{v
k
}
k=1
K, there is a subsequence {v
k
l
}
l=1
such that
_
L
1
v
k
l
_
converges to a
point, a L
1
(K) . Hence by continuity of L,
v
k
l
= L
_
L
1
(v
k
l
)
_
La K.
Conversely, suppose K is sequentially compact. I need to verify it is closed and
bounded. If it is not closed, then it is missing a limit point, k
0
. Since k
0
is a limit point,
there exists k
n
B
_
k
0
,
1
n
_
such that k
n
= k
0
. Therefore, {k
n
} has no limit point in
K because k
0
/ K. It follows K must be closed. If K is not bounded, then you could
pick k
n
K such that k
m
/ B(0, m) and it follows {k
k
} cannot have a subsequence
which converges because if k K, then for large enough m, k B(0, m/2) and so if
_
k
kj
_
is any subsequence, k
kj
/ B(0, m) for all but nitely many j. In other words,
for any k K, it is not the limit of any subsequence. Thus K must also be bounded.
Finally consider the claim about completeness. Let {v
k
}
k=1
be a Cauchy sequence
in V . Since L
1
, dened above is in L(V, F
n
) , it follows
_
L
1
v
k
_
k=1
is a Cauchy
sequence in F
n
. This follows from the inequality,
L
1
v
k
L
1
v
l
L
1
||v
k
v
l
|| .
therefore, there exists a F
n
such that L
1
v
k
a and since L is continuous,
v
k
= L
_
L
1
(v
k
)
_
L(a) .
Next suppose K is a closed and bounded subset of V and let {x
k
}
k=1
be a sequence
of vectors in K. Let {v
1
, , v
n
} be a basis for V and let
x
k
=
n
j=1
x
j
k
v
j
5.9. ASCOLI ARZELA THEOREM 111
Dene a norm for V according to
||x||
2
j=1
x
j
2
, x =
n
j=1
x
j
v
j
It is clear most axioms of a norm hold. The triangle inequality also holds because by
the triangle inequality for F
n
,
||x +y||
_
_
n
j=1
x
j
+y
j
2
_
_
1/2
_
_
n
j=1
x
j
2
_
_
1/2
+
_
_
n
j=1
y
j
2
_
_
1/2
||x|| +||y|| .
By the rst part of this theorem, this norm is equivalent to the norm on V . Thus K is
closed and bounded with respect to this new norm. It follows that for each j,
_
x
j
k
_
k=1
is a bounded sequence in F and so by the theorems about sequential compactness in F
it follows upon taking subsequences n times, there exists a subsequence x
k
l
such that
for each j,
lim
l
x
j
k
l
= x
j
for some x
j
. Hence
lim
l
x
k
l
= lim
l
n
j=1
x
j
k
l
v
j
=
n
j=1
x
j
v
j
K
because K is closed. This proves the theorem.
Example 5.8.5 Let V be a vector space and let {v
1
, , v
n
} be a basis. Dene a norm
on V as follows. For v =
n
k=1
a
k
v
k
,
||v|| max {|a
k
| : k = 1, , n}
In the above example, this is a norm on the vector space, V . It is clear ||av|| = |a| ||v||
and that ||v|| 0 and equals 0 if and only if v = 0. The hard part is the triangle
inequality. Let v =
n
k=1
a
k
v
k
and w = v =
n
k=1
b
k
v
k
.
||v +w|| max
k
{|a
k
+b
k
|} max
k
{|a
k
| +|b
k
|}
max
k
|a
k
| + max
k
|b
k
| ||v|| +||w|| .
This shows this is indeed a norm.
5.9 Ascoli Arzela Theorem
Let {f
k
}
k=1
be a sequence of functions dened on a compact set which have values in a
nite dimensional normed vector space V. The following denition will be of the norm
of such a function.
Denition 5.9.1 Let f : K V be a continuous function which has values in
a nite dimensional normed vector space V where here K is a compact set contained in
some normed vector space. Dene
||f || sup {||f (x)||
V
: x K} .
Denote the set of such functions by C (K; V ) .
Proposition 5.9.2 The above denition yields a norm and in fact C (K; V ) is a
complete normed linear space.
Proof: This is obviously a vector space. Just verify the axioms. The main thing
to show it that the above is a norm. First note that ||f || = 0 if and only if f = 0
and ||f || = || ||f || whenever F, the eld of scalars, C or R. As to the triangle
inequality,
||f +g|| sup {||(f +g) (x)|| : x K}
sup {||f (x)||
V
: x K} + sup {||g (x)||
V
: x K}
||g|| +||f ||
Furthermore, the function x ||f (x)||
V
is continuous thanks to the triangle inequality
which implies
|||f (x)||
V
||f (y)||
V
| ||f (x) f (y)||
V
.
Therefore, ||f || is a well dened nonnegative real number.
It remains to verify completeness. Suppose then {f
k
} is a Cauchy sequence with
respect to this norm. Then from the denition it is a uniformly Cauchy sequence and
since by Theorem 5.8.4 V is a complete normed vector space, it follows from Theorem
5.5.6, there exists f C (K; V ) such that {f
k
} converges uniformly to f . That is,
lim
k
||f f
k
|| = 0.
Theorem 5.8.4 says that closed and bounded sets in a nite dimensional normed vec-
tor space V are sequentially compact. This theorem typically doesnt apply to C (K; V )
because generally this is not a nite dimensional vector space although as shown
above it is a complete normed vector space. It turns out you need more than closed and
bounded in order to have a subset of C (K; V ) be sequentially compact.
Denition 5.9.3 Let F C (K; V ) . Then F is equicontinuous if for every
> 0 there exists > 0 such that whenever |x y| < , it follows
|f (x) f (y)| <
for all f F. F is bounded if there exists C such that
||f || C
for all f F.
Lemma 5.9.4 Let K be a sequentially compact nonempty subset of a nite dimen-
sional normed vector space. Then there exists a countable set D {k
i
}
i=1
such that for
every > 0 and for every x K,
B(x, ) D = .
5.9. ASCOLI ARZELA THEOREM 113
Proof: Let n N. Pick k
n
1
K. If B
_
k
n
1
,
1
n
_
K, stop. Otherwise pick
k
n
2
K \ B
_
k
n
1
,
1
n
_
Continue this way till the process ends. It must end because if it didnt, there would
exist a convergent subsequence which would imply two of the k
n
j
would have to be closer
than 1/n which is impossible from the construction. Denote this collection of points
by D
n
. Then D
n=1
D
n
. This must work because if > 0 is given and x K, let
1/n < /3 and the construction implies x B(k
n
i
, 1/n) for some k
n
i
D
n
D. Then
k
n
i
B(x, ) .
D is countable because it is the countable union of nite sets. This proves the lemma.
Denition 5.9.5 More generally, if K is any subset of a normed vector space

and there exists D such that D is countable and for all x K,
B(x, ) D =
then K is called separable.
Now here is another remarkable result about equicontinuous functions.
Lemma 5.9.6 Suppose {f
k
}
k=1
is equicontinuous and the functions are dened on
a sequentially compact set K. Suppose also for each x K,
lim
k
f
k
(x) = f (x) .
Then in fact f is continuous and the convergence is uniform. That is
lim
k
||f
k
f || = 0.
Proof: Uniform convergence would say that for every > 0, there exists n
such
that if k, l n
, then for all x K,

||f
k
(x) f
l
(x)|| < .
Thus if the given sequence does not converge uniformly, there exists > 0 such that for
all n, there exists k, l n and x
n
K such that
||f
k
(x
n
) f
l
(x
n
)||
Since K is sequentially compact, there exists a subsequence, still denoted by {x
n
} such
that lim
n
x
n
= x K. Then letting k, l be associated with n as just described,
||f
k
(x
n
) f
l
(x
n
)||
V
||f
k
(x
n
) f
k
(x)||
V
+||f
k
(x) f
l
(x)||
V
+||f
l
(x) f
l
(x
n
)||
V
By equicontinuity, if n is large enough, this implies
<

3
+||f
k
(x) f
l
(x)||
V
+

3
and now taking n still larger if necessary, the middle term on the right in the above is
also less than /3 which yields a contradiction. Hence convergence is uniform and so it
follows from Theorem 5.5.6 the function f is actually continuous and
lim
k
||f f
k
|| = 0.
The Ascoli Arzela theorem is the following.
Theorem 5.9.7 Let K be a closed and bounded subset of a nite dimensional
normed vector space and let F C (K; V ) where V is a nite dimensional normed
vector space. Suppose also that F is bounded and equicontinuous. Then if {f
k
}
k=1
F,
there exists a subsequence {f
k
l
}
l=1
which converges to a function f C (K; V ) in the
sense that
lim
l
||f f
k
l
||
Proof: Denote by
_
f
(k,n)
_
n=1
a subsequence of
_
f
(k1,n)
_
n=1
where the index de-
noted by (k 1, k 1) is always less than the index denoted by (k, k) . Also let the
countable dense subset of Lemma 5.9.4 be D = {d
k
}
k=1
. Then consider the following
diagram.
f
(1,1)
, f
(1,2)
, f
(1,3)
, f
(1,4)
, d
1
f
(2,1)
, f
(2,2)
, f
(2,3)
, f
(2,4)
, d
1
, d
2
f
(3,1)
, f
(3,2)
, f
(3,3)
, f
(3,4)
, d
1
, d
2
, d
3
f
(4,1)
, f
(4,2)
, f
(4,3)
, f
(4,4)
, d
1
, d
2
, d
3
, d
4
.
.
.
The meaning is as follows.
_
f
(1,k)
_
k=1
is a subsequence of the original sequence which
converges at d
1
. Such a subsequence exists because {f
k
(d
1
)}
k=1
is contained in a
bounded set so a subsequence converges by Theorem 5.8.4. (It is given to be in a
bounded set and so the closure of this bounded set is both closed and bounded, hence
weakly compact.) Now
_
f
(2,k)
_
k=1
is a subsequence of the rst subsequence which
converges, at d
2
. Then by Theorem 4.1.6 this new subsequence continues to converge at
d
1
. Thus, as indicated by the diagram, it converges at both b
1
and b
2
. Continuing this
way explains the meaning of the diagram. Now consider the subsequence of the original
sequence
_
f
(k,k)
_
k=1
. For k n, this subsequence is a subsequence of the subsequence
{f
n,k
}
k=1
and so it converges at d
1
, d
2
, d
3
, d
4
. This being true for all n, it follows
_
f
(k,k)
_
k=1
converges at every point of D. To save on notation, I shall simply denote
this as {f
k
} .
Then letting d D,
||f
k
(x) f
l
(x)||
V
||f
k
(x) f
k
(d)||
V
+||f
k
(d) f
l
(d)||
V
+||f
l
(d) f
l
(x)||
V
Picking d close enough to x and applying equicontinuity,
||f
k
(x) f
l
(x)||
V
< 2/3 +||f
k
(d) f
l
(d)||
V
Thus for k, l large enough, the right side is less than . This shows that for each x K,
{f
k
(x)}
k=1
is a Cauchy sequence and so by completeness of V this converges. Let f (x)
be the thing to which it converges. Then f is continuous and the convergence is uniform
by Lemma 5.9.6. This proves the theorem.
5.10 Exercises
1. In Theorem 5.7.6 it is assumed f has values in F. Show there is no change if f
has values in V, a normed vector space provided you redene the denition of a
polynomial to be something of the form
||m
a
where a
V .
2. How would you generalize the conclusion of Corollary 5.7.14 to include the situa-
tion where f has values in a nite dimensional normed vector space?
5.10. EXERCISES 115
3. Recall the Bernstein polynomials
p
m
(x) =
km
_
m
k
_
x
k
(1 x)
mk
f
_
k
m
_
It was shown that these converge uniformly to f provided min (m) . Explain
why it suces to have max (m) . See the estimate which was derived.
4. If {f
n
} and {g
n
} are sequences of F
n
valued functions dened on D which converge
uniformly, show that if a, b are constants, then af
n
+bg
n
also converges uniformly.
If there exists a constant, M such that |f
n
(x)| , |g
n
(x)| < M for all n and for all
x D, show {f
n
g
n
} converges uniformly. Let f
n
(x) 1/ |x| for x B(0,1)
and let g
n
(x) (n 1) /n. Show {f
n
} converges uniformly on B(0,1) and {g
n
}
converges uniformly but {f
n
g
n
} fails to converge uniformly.
5. Formulate a theorem for series of functions of n variables which will allow you to
conclude the innite series is uniformly continuous based on reasonable assump-
tions about the functions in the sum.
6. If f and g are real valued functions which are continuous on some set, D, show
that
min (f, g) , max (f, g)
are also continuous. Generalize this to any nite collection of continuous functions.
Hint: Note max (f, g) =
|fg|+f+g
2
. Now recall the triangle inequality which can
be used to show || is a continuous function.
7. Find an example of a sequence of continuous functions dened on R
n
such that
each function is nonnegative and each function has a maximum value equal to 1
but the sequence of functions converges to 0 pointwise on R
n
\ {0} , that is, the
set of vectors in R
n
excluding 0.
8. Theorem 5.3.14 says an open subset U of R
n
is arcwise connected if and only if U is
connected. Consider the usual Cartesian coordinates relative to axes x
1
, , x
n
.
A square curve is one consisting of a succession of straight line segments each
of which is parallel to some coordinate axis. Show an open subset U of R
n
is
connected if and only if every two points can be joined by a square curve.
9. Let x h(x) be a bounded continuous function. Show the function f (x) =
n=1
h(nx)
n
2
is continuous.
10. Let S be a any countable subset of R
n
. Show there exists a function, f dened
on R
n
which is discontinuous at every point of S but continuous everywhere else.
Hint: This is real easy if you do the right thing. It involves the Weierstrass M
test.
11. By Theorem 5.7.13 there exists a sequence of polynomials converging uniformly to
f (x) = |x| on R
n
k=1
[M, M] . Show there exists a sequence of polynomials,
{p
n
} converging uniformly to f on R which has the additional property that for
all n, p
n
(0) = 0.
12. If f is any continuous function dened on K a sequentially compact subset of R
n
,
show there exists a series of the form
k=1
p
k
, where each p
k
is a polynomial,
which converges uniformly to f on [a, b]. Hint: You should use the Weierstrass
approximation theorem to obtain a sequence of polynomials. Then arrange it so
the limit of this sequence is an innite sum.
13. A function f is Holder continuous if there exists a constant, K such that
|f (x) f (y)| K|x y|
for some 1 for all x, y. Show every Holder continuous function is uniformly
continuous.
14. Consider f (x) dist (x,S) where S is a nonempty subset of R
n
. Show f is
uniformly continuous.
15. Let K be a sequentially compact set in a normed vector space V and let f : V
W be continuous where W is also a normed vector space. Show f (K) is also
sequentially compact.
16. If f is uniformly continuous, does it follow that |f | is also uniformly continuous?
If |f | is uniformly continuous does it follow that f is uniformly continuous? An-
swer the same questions with uniformly continuous replaced with continuous.
Explain why.
17. Let f : D R be a function. This function is said to be lower semicontinuous
1
at x D if for any sequence {x
n
} D which converges to x it follows
f (x) lim inf
n
f (x
n
) .
Suppose D is sequentially compact and f is lower semicontinuous at every point
of D. Show that then f achieves its minimum on D.
18. Let f : D R be a function. This function is said to be upper semicontinuous at
x D if for any sequence {x
n
} D which converges to x it follows
f (x) lim sup
n
f (x
n
) .
Suppose D is sequentially compact and f is upper semicontinuous at every point
of D. Show that then f achieves its maximum on D.
19. Show that a real valued function dened on D R
n
is continuous if and only if
it is both upper and lower semicontinuous.
20. Show that a real valued lower semicontinuous function dened on a sequentially
compact set achieves its minimum and that an upper semicontinuous function
dened on a sequentially compact set achieves its maximum.
21. Give an example of a lower semicontinuous function dened on R
n
which is not
continuous and an example of an upper semicontinuous function which is not
continuous.
22. Suppose {f
: } is a collection of continuous functions. Let

F (x) inf {f
(x) : }
Show F is an upper semicontinuous function. Next let
G(x) sup {f
(x) : }
Show G is a lower semicontinuous function.
1
The notion of lower semicontinuity is very important for functions which are dened on innite
dimensional sets. In more general settings, one formulates the concept dierently.
5.10. EXERCISES 117
23. Let f be a function. epi (f) is dened as
{(x, y) : y f (x)} .
It is called the epigraph of f. We say epi (f) is closed if whenever (x
n
, y
n
) epi (f)
and x
n
x and y
n
y, it follows (x, y) epi (f) . Show f is lower semicontinuous
if and only if epi (f) is closed. What would be the corresponding result equivalent
to upper semicontinuous?
24. The operator norm was dened for L(V, W) above. This is the usual norm used
for this vector space of linear transformations. Show that any other norm used on
L(V, W) is equivalent to the operator norm. That is, show that if ||||
1
is another
norm, there exist scalars , such that
||L|| ||L||
1
||L||
for all L L(V, W) where here |||| denotes the operator norm.
25. One alternative norm which is very popular is as follows. Let L L(V, W) and
let (l
ij
) denote the matrix of L with respect to some bases. Then the Frobenius
norm is dened by
_
_
ij
|l
ij
|
2
_
_
1/2
||L||
F
.
Show this is a norm. Other norms are of the form
_
_
ij
|l
ij
|
p
_
_
1/p
where p 1 or even
||L||
= max
ij
|l
ij
| .
Show these are also norms.
26. Explain why L(V, W) is always a complete normed vector space whenever V, W
are nite dimensional normed vector spaces for any choice of norm for L(V, W).
Also explain why every closed and bounded subset of L(V, W) is sequentially
compact for any choice of norm on this space.
27. Let L L(V, V ) where V is a nite dimensional normed vector space. Dene
e
L
k=1
L
k
k!
Explain the meaning of this innite sum and show it converges in L(V, V ) for any
choice of norm on this space. Now tell how to dene sin (L).
28. Let X be a nite dimensional normed vector space, real or complex. Show that X
is separable. Hint: Let {v
i
}
n
i=1
be a basis and dene a map from F
n
to X, , as
follows. (
n
k=1
x
k
e
k
)
n
k=1
x
k
v
k
. Show is continuous and has a continuous
inverse. Now let D be a countable dense set in F
n
and consider (D).
29. Let B(X; R
n
) be the space of functions f , mapping X to R
n
such that
sup{|f (x)| : x X} < .
Show B(X; R
n
) is a complete normed linear space if we dene
||f || sup{|f (x)| : x X}.
30. Let (0, 1]. Dene, for X a compact subset of R
p
,
C
(X; R
n
) {f C (X; R
n
) :
(f ) +||f || ||f ||
< }
where
||f || sup{|f (x)| : x X}
and
(f ) sup{
|f (x) f (y)|
|x y|
: x, y X, x = y}.
Show that (C
(X; R
n
) , ||||
) is a complete normed linear space. This is called a

Holder space. What would this space consist of if > 1?
31. Let {f
n
}
n=1
C
(X; R
n
) where X is a compact subset of R
p
and suppose
||f
n
||
M
for all n. Show there exists a subsequence, n
k
, such that f
n
k
converges in C (X; R
n
).
The given sequence is precompact when this happens. (This also shows the em-
bedding of C
(X; R
n
) into C (X; R
n
) is a compact embedding.) Hint: You might
want to use the Ascoli Arzela theorem.
32. This problem is for those who know about the derivative and the integral of a
function of one variable. Let f :R R
n
R
n
be continuous and bounded and let
x
0
R
n
. If
x : [0, T] R
n
and h > 0, let
h
x(s)
_
x
0
if s h,
x(s h) , if s > h.
For t [0, T], let
x
h
(t) = x
0
+
t
0
f (s,
h
x
h
(s)) ds.
Show using the Ascoli Arzela theorem that there exists a sequence h 0 such
that
x
h
x
in C ([0, T] ; R
n
). Next argue
x(t) = x
0
+
t
0
f (s, x(s)) ds
and conclude the following theorem. If f :R R
n
R
n
is continuous and bounded,
and if x
0
R
n
is given, there exists a solution to the following initial value prob-
lem.
x
= f (t, x) , t [0, T]
x(0) = x
0
.
This is the Peano existence theorem for ordinary dierential equations.
5.10. EXERCISES 119
33. Let D(x
0
, r) be the closed ball in R
n
,
{x : |x x
0
| r}
where this is the usual norm coming from the dot product. Let P : R
n
D(x
0
, r)
be dened by
P (x)
_
x if x D(x
0
, r)
x
0
+r
xx
0
|xx
0
|
if x / D(x
0
, r)
Show that |Px Py| |x y| for all x R
n
.
34. Use Problem 32 to obtain local solutions to the initial value problem where f is
not assumed to be bounded. It is only assumed to be continuous. This means
there is a small interval whose length is perhaps not T such that the solution to
the dierential equation exists on this small interval.
The Derivative
6.1 Basic Denitions
The concept of derivative generalizes right away to functions of many variables. How-
ever, no attempt will be made to consider derivatives from one side or another. This is
because when you consider functions of many variables, there isnt a well dened side.
However, it is certainly the case that there are more general notions which include such
things. I will present a fairly general notion of the derivative of a function which is
dened on a normed vector space which has values in a normed vector space. The case
of most interest is that of a function which maps F
n
to F
m
but it is no more trouble
to consider the extra generality and it is sometimes useful to have this extra generality
because sometimes you want to consider functions dened, for example on subspaces
of F
n
and it is nice to not have to trouble with ad hoc considerations. Also, you might
want to consider F
n
with some norm other than the usual one.
In what follows, X, Y will denote normed vector spaces. Thanks to Theorem 5.8.4
all the denitions and theorems given below work the same for any norm given on the
vector spaces.
Let U be an open set in X, and let f : U Y be a function.
Denition 6.1.1 A function g is o(v) if
lim
||v||0
g (v)
||v||
= 0 (6.1)
A function f : U Y is dierentiable at x U if there exists a linear transformation
L L(X, Y ) such that
f (x +v) = f (x) +Lv +o(v)
This linear transformation L is the denition of Df (x). This derivative is often called
the Frechet derivative.
Note that from Theorem 5.8.4 the question whether a given function is dierentiable
is independent of the norm used on the nite dimensional vector space. That is, a
function is dierentiable with one norm if and only if it is dierentiable with another
norm.
The denition 6.1 means the error,
f (x +v) f (x) Lv
converges to 0 faster than ||v||. Thus the above denition is equivalent to saying
lim
||v||0
||f (x +v) f (x) Lv||
||v||
= 0 (6.2)
121
122 THE DERIVATIVE
or equivalently,
lim
yx
||f (y) f (x) Df (x) (y x)||
||y x||
= 0. (6.3)
The symbol, o(v) should be thought of as an adjective. Thus, if t and k are con-
stants,
o(v) = o(v) +o(v) , o(tv) = o(v) , ko(v) = o(v)
and other similar observations hold.
Theorem 6.1.2 The derivative is well dened.
Proof: First note that for a xed vector, v, o(tv) = o(t). This is because
lim
t0
o(tv)
|t|
= lim
t0
||v||
o(tv)
||tv||
= 0
Now suppose both L
1
and L
2
work in the above denition. Then let v be any vector
and let t be a real scalar which is chosen small enough that tv +x U. Then
f (x +tv) = f (x) +L
1
tv +o(tv) , f (x +tv) = f (x) +L
2
tv +o(tv) .
Therefore, subtracting these two yields (L
2
L
1
) (tv) = o(tv) = o(t). Therefore, di-
viding by t yields (L
2
L
1
) (v) =
o(t)
t
. Now let t 0 to conclude that (L
2
L
1
) (v) =
0. Since this is true for all v, it follows L
2
= L
1
Lemma 6.1.3 Let f be dierentiable at x. Then f is continuous at x and in fact,
there exists K > 0 such that whenever ||v|| is small enough,
||f (x +v) f (x)|| K||v||
Proof: From the denition of the derivative,
f (x +v) f (x) = Df (x) v +o(v) .
Let ||v|| be small enough that
o(||v||)
||v||
< 1 so that ||o(v)|| ||v||. Then for such v,
||f (x +v) f (x)|| ||Df (x) v|| +||v||
(||Df (x)|| + 1) ||v||
This proves the lemma with K = ||Df (x)|| + 1.
Here ||Df (x)|| is the operator norm of the linear transformation, Df (x).
6.2 The Chain Rule
With the above lemma, it is easy to prove the chain rule.
Theorem 6.2.1 (The chain rule) Let U and V be open sets U X and V Y .
Suppose f : U V is dierentiable at x U and suppose g : V F
q
is dierentiable
at f (x) V . Then g f is dierentiable at x and
D(g f ) (x) = D(g (f (x))) D(f (x)) .
6.3. THE MATRIX OF THE DERIVATIVE 123
Proof: This follows from a computation. Let B(x,r) U and let r also be small
enough that for ||v|| r, it follows that f (x +v) V . Such an r exists because f is
continuous at x. For ||v|| < r, the denition of dierentiability of g and f implies
g (f (x +v)) g (f (x)) =
Dg (f (x)) (f (x +v) f (x)) +o(f (x +v) f (x))
= Dg (f (x)) [Df (x) v +o(v)] +o(f (x +v) f (x))
= D(g (f (x))) D(f (x)) v +o(v) +o(f (x +v) f (x)) . (6.4)
It remains to show o(f (x +v) f (x)) = o(v).
By Lemma 6.1.3, with K given there, letting > 0, it follows that for ||v|| small
enough,
||o(f (x +v) f (x))|| (/K) ||f (x +v) f (x)|| (/K) K||v|| = ||v|| .
Since > 0 is arbitrary, this shows o(f (x +v) f (x)) = o(v) because whenever ||v||
is small enough,
||o(f (x +v) f (x))||
||v||
.
By 6.4, this shows
g (f (x +v)) g (f (x)) = D(g (f (x))) D(f (x)) v +o(v)
which proves the theorem.
6.3 The Matrix Of The Derivative
Let X, Y be normed vector spaces, a basis for X being {v
1
, , v
n
} and a basis for Y
being {w
1
, , w
m
} . First note that if
i
: X F is dened by
i
v x
i
where v =
k
x
k
v
k
,
then
i
L(X, F) and so by Theorem 5.8.3, it follows that
i
is continuous and if
lim
st
g (s) = L, then |
i
g (s)
i
L| ||
i
|| ||g (s) L|| and so the i
th
components
converge also.
Suppose that f : U Y is dierentiable. What is the matrix of Df (x) with respect
to the given bases? That is, if
Df (x) =
ij
J
ij
(x) w
i
v
j
,
what is J
ij
(x)?
D
v
k
f (x) lim
t0
f (x +tv
k
) f (x)
t
= lim
t0
Df (x) (tv
k
) +o(tv
k
)
t
= Df (x) (v
k
) =
ij
J
ij
(x) w
i
v
j
(v
k
) =
ij
J
ij
(x) w
i
jk
=
i
J
ik
(x) w
i
124 THE DERIVATIVE
It follows
lim
t0
j
_
f (x +tv
k
) f (x)
t
_
lim
t0
f
j
(x +tv
k
) f
j
(x)
t
D
v
k
f
j
(x)
=
j
_
i
J
ik
(x) w
i
_
= J
jk
(x)
Thus J
ik
(x) = D
v
k
f
i
(x).
In the case where X = R
n
and Y = R
m
and v is a unit vector, D
v
f
i
(x) is the
familiar directional derivative in the direction v of the function, f
i
.
Of course the case where X = F
n
and f : U F
n
F
m
, is dierentiable and the
basis vectors are the usual basis vectors is the case most commonly encountered. What
is the matrix of Df (x) taken with respect to the usual basis vectors? Let e
i
denote the
vector of F
n
which has a one in the i
th
entry and zeroes elsewhere. This is the standard
basis for F
n
. Denote by J
ij
(x) the matrix with respect to these basis vectors. Thus
Df (x) =
ij
J
ij
(x) e
i
e
j
.
Then from what was just shown,
J
ik
(x) = D
e
k
f
i
(x) lim
t0
f
i
(x +te
k
) f
i
(x)
t
f
i
x
k
(x) f
i,x
k
(x) f
i,k
(x)
where the last several symbols are just the usual notations for the partial derivative of
the function, f
i
with respect to the k
th
variable where
f (x)
m
i=1
f
i
(x) e
i
.
In other words, the matrix of Df (x) is nothing more than the matrix of partial deriva-
tives. The k
th
column of the matrix (J
ij
) is
f
x
k
(x) = lim
t0
f (x +te
k
) f (x)
t
D
e
k
f (x) .
Thus the matrix of Df (x) with respect to the usual basis vectors is the matrix of
the form
_
_
_
f
1,x1
(x) f
1,x2
(x) f
1,xn
(x)
.
.
.
.
.
.
.
.
.
f
m,x1
(x) f
m,x2
(x) f
m,xn
(x)
_
_
_
where the notation g
,x
k
denotes the k
th
partial derivative given by the limit,
lim
t0
g (x +te
k
) g (x)
t

g
x
k
.
The above discussion is summarized in the following theorem.
Theorem 6.3.1 Let f : F
n
F
m
and suppose f is dierentiable at x. Then all
the partial derivatives
fi(x)
xj
exist and if Jf (x) is the matrix of the linear transformation,
Df (x) with respect to the standard basis vectors, then the ij
th
entry is given by
fi
xj
(x)
also denoted as f
i,j
or f
i,xj
.
6.4. A MEAN VALUE INEQUALITY 125
Denition 6.3.2 In general, the symbol
D
v
f (x)
is dened by
lim
t0
f (x +tv) f (x)
t
where t F. This is often called the Gateaux derivative.
What if all the partial derivatives of f exist? Does it follow that f is dierentiable?
Consider the following function, f : R
2
R,
f (x, y) =
_
xy
x
2
+y
2
if (x, y) = (0, 0)
0 if (x, y) = (0, 0)
.
Then from the denition of partial derivatives,
lim
h0
f (h, 0) f (0, 0)
h
= lim
h0
0 0
h
= 0
and
lim
h0
f (0, h) f (0, 0)
h
= lim
h0
0 0
h
= 0
However f is not even continuous at (0, 0) which may be seen by considering the behavior
of the function along the line y = x and along the line x = 0. By Lemma 6.1.3 this
implies f is not dierentiable. Therefore, it is necessary to consider the correct denition
of the derivative given above if you want to get a notion which generalizes the concept
of the derivative of a function of one variable in such a way as to preserve continuity
whenever the function is dierentiable.
6.4 A Mean Value Inequality
The following theorem will be very useful in much of what follows. It is a version of the
mean value theorem as is the next lemma.
Lemma 6.4.1 Let Y be a normed vector space and suppose h : [0, 1] Y is dier-
entiable and satises
||h
(t)|| M.
Then
||h(1) h(0)|| M.
Proof: Let > 0 be given and let
S {t [0, 1] : for all s [0, t] , ||h(s) h(0)|| (M +) s}
Then 0 S. Let t = sup S. Then by continuity of h it follows
||h(t) h(0)|| = (M +) t (6.5)
Suppose t < 1. Then there exist positive numbers, h
k
decreasing to 0 such that
||h(t +h
k
) h(0)|| > (M +) (t +h
k
)
and now it follows from 6.5 and the triangle inequality that
||h(t +h
k
) h(t)|| +||h(t) h(0)||
= ||h(t +h
k
) h(t)|| + (M +) t > (M +) (t +h
k
)
126 THE DERIVATIVE
and so
||h(t +h
k
) h(t)|| > (M +) h
k
Now dividing by h
k
and letting k
||h
(t)|| M +,
a contradiction. This proves the lemma.
Theorem 6.4.2 Suppose U is an open subset of X and f : U Y has the
property that Df (x) exists for all x in U and that, x + t (y x) U for all t [0, 1].
(The line segment joining the two points lies in U.) Suppose also that for all points on
this line segment,
||Df (x+t (y x))|| M.
Then
||f (y) f (x)|| M|y x| .
Proof: Let
h(t) f (x +t (y x)) .
Then by the chain rule,
h
(t) = Df (x +t (y x)) (y x)
and so
||h
(t)|| = ||Df (x +t (y x)) (y x)||

M||y x||
by Lemma 6.4.1
||h(1) h(0)|| = ||f (y) f (x)|| M||y x|| .
6.5 Existence Of The Derivative, C
1
Functions
There is a way to get the dierentiability of a function from the existence and continuity
of the Gateaux derivatives. This is very convenient because these Gateaux derivatives
are taken with respect to a one dimensional variable. The following theorem is the main
result.
Theorem 6.5.1 Let X be a normed vector space having basis {v
1
, , v
n
} and
let Y be another normed vector space having basis {w
1
, , w
m
} . Let U be an open set
in X and let f : U Y have the property that the Gateaux derivatives,
D
v
k
f (x) lim
t0
f (x +tv
k
) f (x)
t
exist and are continuous functions of x. Then Df (x) exists and
Df (x) v =
n
k=1
D
v
k
f (x) a
k
where
v =
n
k=1
a
k
v
k
.
6.5. EXISTENCE OF THE DERIVATIVE, C
1
FUNCTIONS 127
Furthermore, x Df (x) is continuous; that is
lim
yx
||Df (y) Df (x)|| = 0.
Proof: Let v =
n
k=1
a
k
v
k
. Then
f (x +v) f (x) = f
_
x+
n
k=1
a
k
v
k
_
f (x) .
Then letting
0
k=1
0, f (x +v) f (x) is given by
n
k=1
_
_
f
_
_
x+
k
j=1
a
j
v
j
_
_
f
_
_
x+
k1
j=1
a
j
v
j
_
_
_
_
=
n
k=1
[f (x +a
k
v
k
) f (x)] +
n
k=1
_
_
_
_
f
_
_
x+
k
j=1
a
j
v
j
_
_
f (x +a
k
v
k
)
_
_
_
_
f
_
_
x+
k1
j=1
a
j
v
j
_
_
f (x)
_
_
_
_
(6.6)
Consider the k
th
term in 6.6. Let
h(t) f
_
_
x+
k1
j=1
a
j
v
j
+ta
k
v
k
_
_
f (x +ta
k
v
k
)
for t [0, 1] . Then
h
(t) = a
k
lim
h0
1
a
k
h
_
_
f
_
_
x+
k1
j=1
a
j
v
j
+ (t +h) a
k
v
k
_
_
f (x + (t +h) a
k
v
k
)
_
_
f
_
_
x+
k1
j=1
a
j
v
j
+ta
k
v
k
_
_
f (x +ta
k
v
k
)
_
_
_
_
and this equals
_
_
D
v
k
f
_
_
x+
k1
j=1
a
j
v
j
+ta
k
v
k
_
_
D
v
k
f (x +ta
k
v
k
)
_
_
a
k
(6.7)
Now without loss of generality, it can be assumed the norm on X is given by that of
Example 5.8.5,
||v|| max
_
_
_
|a
k
| : v =
n
j=1
a
k
v
k
_
_
_
because by Theorem 5.8.4 all norms on X are equivalent. Therefore, from 6.7 and the
assumption that the Gateaux derivatives are continuous,
||h
(t)|| =
_
_
D
v
k
f
_
_
x+
k1
j=1
a
j
v
j
+ta
k
v
k
_
_
D
v
k
f (x +ta
k
v
k
)
_
_
a
k
|a
k
| ||v||
128 THE DERIVATIVE
provided ||v|| is suciently small. Since is arbitrary, it follows from Lemma 6.4.1 the
expression in 6.6 is o(v) because this expression equals a nite sum of terms of the form
h(1) h(0) where ||h
(t)|| ||v|| . Thus

f (x +v) f (x) =
n
k=1
[f (x +a
k
v
k
) f (x)] +o(v)
=
n
k=1
D
v
k
f (x) a
k
+
n
k=1
[f (x +a
k
v
k
) f (x) D
v
k
f (x) a
k
] +o(v) .
Consider the k
th
term in the second sum.
f (x +a
k
v
k
) f (x) D
v
k
f (x) a
k
= a
k
_
f (x +a
k
v
k
) f (x)
a
k
D
v
k
f (x)
_
where the expression in the parentheses converges to 0 as a
k
0. Thus whenever ||v||
is suciently small,
||f (x +a
k
v
k
) f (x) D
v
k
f (x) a
k
|| |a
k
| ||v||
which shows the second sum is also o(v). Therefore,
f (x +v) f (x) =
n
k=1
D
v
k
f (x) a
k
+o(v) .
Dening
Df (x) v
n
k=1
D
v
k
f (x) a
k
where v =
k
a
k
v
k
, it follows Df (x) L(X, Y ) and is given by the above formula.
It remains to verify x Df (x) is continuous.
||(Df (x) Df (y)) v||
k=1
||(D
v
k
f (x) D
v
k
f (y)) a
k
||
max {|a
k
| , k = 1, , n}
n
k=1
||D
v
k
f (x) D
v
k
f (y)||
= ||v||
n
k=1
||D
v
k
f (x) D
v
k
f (y)||
and so
||Df (x) Df (y)||
n
k=1
||D
v
k
f (x) D
v
k
f (y)||
which proves the continuity of Df because of the assumption the Gateaux derivatives
are continuous. This proves the theorem.
This motivates the following denition of what it means for a function to be C
1
.
Denition 6.5.2 Let U be an open subset of a normed nite dimensional vector
space, X and let f : U Y another nite dimensional normed vector space. Then f is
said to be C
1
if there exists a basis for X, {v
1
, , v
n
} such that the Gateaux derivatives,
D
v
k
f (x)
exist on U and are continuous.
6.6. HIGHER ORDER DERIVATIVES 129
Here is another denition of what it means for a function to be C
1
.
Denition 6.5.3 Let U be an open subset of a normed nite dimensional vector
space, X and let f : U Y another nite dimensional normed vector space. Then f
is said to be C
1
if f is dierentiable and x Df (x) is continuous as a map from U to
L(X, Y ).
Now the following major theorem states these two denitions are equivalent.
Theorem 6.5.4 Let U be an open subset of a normed nite dimensional vector
space, X and let f : U Y another nite dimensional normed vector space. Then the
two denitions above are equivalent.
Proof: It was shown in Theorem 6.5.1 that Denition 6.5.2 implies 6.5.3. Suppose
then that Denition 6.5.3 holds. Then if v is any vector,
lim
t0
f (x +tv) f (x)
t
= lim
t0
Df (x) tv +o(tv)
t
= Df (x) v+ lim
t0
o(tv)
t
= Df (x) v
Thus D
v
f (x) exists and equals Df (x) v. By continuity of x Df (x) , this establishes
continuity of x D
v
f (x) and proves the theorem.
Note that the proof of the theorem also implies the following corollary.
Corollary 6.5.5 Let U be an open subset of a normed nite dimensional vector
space, X and let f : U Y another nite dimensional normed vector space. Then if
there is a basis of X, {v
1
, , v
n
} such that the Gateaux derivatives, D
v
k
f (x) exist and
are continuous. Then all Gateaux derivatives, D
v
f (x) exist and are continuous for all
v X.
From now on, whichever denition is more convenient will be used.
6.6 Higher Order Derivatives
If f : U X Y for U an open set, then
x Df (x)
is a mapping from U to L(X, Y ), a normed vector space. Therefore, it makes perfect
sense to ask whether this function is also dierentiable.
Denition 6.6.1 The following is the denition of the second derivative.
D
2
f (x) D(Df (x)) .
Thus,
Df (x +v) Df (x) = D
2
f (x) v +o(v) .
This implies
D
2
f (x) L(X, L(X, Y )) , D
2
f (x) (u) (v) Y,
and the map
(u, v) D
2
f (x) (u) (v)
is a bilinear map having values in Y . In other words, the two functions,
u D
2
f (x) (u) (v) , v D
2
f (x) (u) (v)
130 THE DERIVATIVE
are both linear.
The same pattern applies to taking higher order derivatives. Thus,
D
3
f (x) D
_
D
2
f (x)
_
and D
3
f (x) may be considered as a trilinear map having values in Y . In general D
k
f (x)
may be considered a k linear map. This means the function
(u
1
, , u
k
) D
k
f (x) (u
1
) (u
k
)
has the property
u
j
D
k
f (x) (u
1
) (u
j
) (u
k
)
is linear.
Also, instead of writing
D
2
f (x) (u) (v) , or D
3
f (x) (u) (v) (w)
the following notation is often used.
D
2
f (x) (u, v) or D
3
f (x) (u, v, w)
with similar conventions for higher derivatives than 3. Another convention which is
often used is the notation
D
k
f (x) v
k
instead of
D
k
f (x) (v, , v) .
Note that for every k, D
k
f maps U to a normed vector space. As mentioned above,
Df (x) has values in L(X, Y ) , D
2
f (x) has values in L(X, L(X, Y )) , etc. Thus it makes
sense to consider whether D
k
f is continuous. This is described in the following denition.
Denition 6.6.2 Let U be an open subset of X, a normed vector space and let
f : U Y. Then f is C
k
(U) if f and its rst k derivatives are all continuous. Also,
D
k
f (x) when it exists can be considered a Y valued multilinear function.
6.7 C
k
Functions
Recall that for a C
1
function, f
Df (x) v =
j
D
vj
f (x) a
j
=
ij
D
vj
f
i
(x) w
i
a
j
=
ij
D
vj
f
i
(x) w
i
v
j
_
k
a
k
v
k
_
=
ij
D
vj
f
i
(x) w
i
v
j
(v)
where
k
a
k
v
k
= v and
f (x) =
i
f
i
(x) w
i
. (6.8)
This is because
w
i
v
j
_
k
a
k
v
k
_
k
a
k
w
i
jk
= w
i
a
j
.
Thus
Df (x) =
ij
D
vj
f
i
(x) w
i
v
j
6.7. C
K
FUNCTIONS 131
I propose to iterate this observation, starting with f and then going to Df and then
D
2
f and so forth. Hopefully it will yield a rational way to understand higher order
derivatives in the same way that matrices can be used to understand linear transforma-
tions. Thus beginning with the derivative,
Df (x) =
ij1
D
vj
1
f
i
(x) w
i
v
j1
.
Then letting w
i
v
j1
play the role of w
i
in 6.8,
D
2
f (x) =
ij1j2
D
vj
2
_
D
vj
1
f
i
_
(x) w
i
v
j1
v
j2
ij1j2
D
vj
1
vj
2
f
i
(x) w
i
v
j1
v
j2
Then letting w
i
v
j1
v
j2
play the role of w
i
in 6.8,
D
3
f (x) =
ij1j2j3
D
vj
3
_
D
vj
1
vj
2
f
i
_
(x) w
i
v
j1
v
j2
v
j3
ij1j2j3
D
vj
1
vj
2
vj
3
f
i
(x) w
i
v
j1
v
j2
v
j3
etc. In general, the notation,
w
i
v
j1
v
j2
v
jn
denes an appropriate linear transformation given by
w
i
v
j1
v
j2
v
jn
(v
k
) = w
i
v
j1
v
j2
v
jn1
kjn
.
The following theorem is important.
Theorem 6.7.1 The function x D
k
f (x) exists and is continuous for k p
if and only if there exists a basis for X, {v
1
, , v
n
} and a basis for Y, {w
1
, , w
m
}
such that for
f (x)
i
f
i
(x) w
i
,
it follows that for each i = 1, 2, , m all Gateaux derivatives,
D
vj
1
vj
2
vj
k
f
i
(x)
for any choice of v
j1
v
j2
v
j
k
and for any k p exist and are continuous.
Proof: This follows from a repeated application of Theorems 6.5.1 and 6.5.4 at each
new dierentiation.
Denition 6.7.2 Let X, Y be nite dimensional normed vector spaces and let
U be an open set in X and f : U Y be a function,
f (x) =
i
f
i
(x) w
i
where {w
1
, , w
m
} is a basis for Y. Then f is said to be a C
n
(U) function if for every
k n, D
k
f (x) exists for all x U and is continuous. This is equivalent to the other
condition which states that for each i = 1, 2, , m all Gateaux derivatives,
D
vj
1
vj
2
vj
k
f
i
(x)
for any choice of v
j1
v
j2
v
j
k
where {v
1
, , v
n
} is a basis for X and for any k n
exist and are continuous.
132 THE DERIVATIVE
6.7.1 Some Standard Notation
In the case where X = R
n
and the basis chosen is the standard basis, these Gateaux
derivatives are just the partial derivatives. Recall the notation for partial derivatives in
the following denition.
Denition 6.7.3 Let g : U X. Then
g
x
k
(x)
g
x
k
(x) lim
h0
g (x +he
k
) g (x)
h
Higher order partial derivatives are dened in the usual way.
g
x
k
x
l
(x)

2
g
x
l
x
k
(x)
and so forth.
A convenient notation which is often used which helps to make sense of higher order
partial derivatives is presented in the following denition.
Denition 6.7.4 = (
1
, ,
n
) for
1

n
positive integers is called a
multi-index. For a multi-index, ||
1
+ +
n
and if x X,
x =(x
1
, , x
n
),
and f a function, dene
x
x
1
1
x
2
2
x
n
n
, D
f (x)

||
f (x)
x
1
1
x
2
2
x
n
n
.
Then in this special case, the following denition is equivalent to the above as a
denition of what is meant by a C
k
function.
Denition 6.7.5 Let U be an open subset of R
n
and let f : U Y. Then for k
a nonnegative integer, f is C
k
if for every || k, D
f exists and is continuous.

6.8 The Derivative And The Cartesian Product
There are theorems which can be used to get dierentiability of a function based on
existence and continuity of the partial derivatives. A generalization of this was given
above. Here a function dened on a product space is considered. It is very much like
what was presented above and could be obtained as a special case but to reinforce
the ideas, I will do it from scratch because certain aspects of it are important in the
statement of the implicit function theorem.
The following is an important abstract generalization of the concept of partial deriva-
tive presented above. Insead of taking the derivative with respect to one variable, it is
taken with respect to several but not with respect to others. This vague notion is made
precise in the following denition. First here is a lemma.
Lemma 6.8.1 Suppose U is an open set in X Y. Then the set, U
y
dened by
U
y
{x X : (x, y) U}
is an open set in X. Here XY is a nite dimensional vector space in which the vector
space operations are dened componentwise. Thus for a, b F,
a (x
1
, y
1
) +b (x
2
, y
2
) = (ax
1
+bx
2
, ay
1
+by
2
)
and the norm can be taken to be
||(x, y)|| max (||x|| , ||y||)
6.8. THE DERIVATIVE AND THE CARTESIAN PRODUCT 133
Proof: Recall by Theorem 5.8.4 it does not matter how this norm is dened and
the denition above is convenient. It obviously satises most axioms of a norm. The
only one which is not obvious is the triangle inequality. I will show this now.
||(x, y) + (x
1
, y
1
)|| = ||(x +x
1
, y +y
1
)|| max (||x +x
1
|| , ||y +y
1
||)
max (||x|| +||x
1
|| , ||y|| +||y
1
||)
suppose then that ||x|| +||x
1
|| ||y|| +||y
1
|| . Then the above equals
||x|| +||x
1
|| max (||x|| , ||y||) + max (||x
1
|| , ||y
1
||) ||(x, y)|| +||(x
1
, y
1
)||
In case ||x|| +||x
1
|| < ||y|| +||y
1
|| , the argument is similar.
Let x U
y
. Then (x, y) U and so there exists r > 0 such that
B((x, y) , r) U.
This says that if (u, v) X Y such that ||(u, v) (x, y)|| < r, then (u, v) U. Thus
if
||(u, y) (x, y)|| = ||u x|| < r,
then (u, y) U. This has just said that B(x,r), the ball taken in X is contained in U
y
.
Or course one could also consider
U
x
{y : (x, y) U}
in the same way and conclude this set is open in Y . Also, the generalization to many
factors yields the same conclusion. In this case, for x
n
i=1
X
i
, let
||x|| max
_
||x
i
||
Xi
: x = (x
1
, , x
n
)
_
Then a similar argument to the above shows this is a norm on
n
i=1
X
i
.
Corollary 6.8.2 Let U
n
i=1
X
i
and let
U
(x1, ,xi1,xi+1, ,xn)

_
x F
ri
:
_
x
1
, , x
i1
, x, x
i+1
, , x
n
_
U
_
.
Then U
(x1, ,xi1,xi+1, ,xn)
is an open set in F
ri
.
The proof is similar to the above.
Denition 6.8.3 Let g : U
n
i=1
X
i
Y , where U is an open set. Then the
map
z g
_
x
1
, , x
i1
, z, x
i+1
, , x
n
_
is a function from the open set in X
i
,
_
z : x =
_
x
1
, , x
i1
, z, x
i+1
, , x
n
_
U
_
to Y . When this map is dierentiable, its derivative is denoted by D
i
g (x). To aid in
the notation, for v X
i
, let
i
v
n
i=1
X
i
be the vector (0, , v, , 0) where the v
is in the i
th
slot and for v
n
i=1
X
i
, let v
i
denote the entry in the i
th
slot of v. Thus,
by saying
z g
_
x
1
, , x
i1
, z, x
i+1
, , x
n
_
is dierentiable is meant that for v X
i
suciently small,
g (x +
i
v) g (x) = D
i
g (x) v +o(v) .
Note D
i
g (x) L(X
i
, Y ) .
134 THE DERIVATIVE
Denition 6.8.4 Let U X be an open set. Then f : U Y is C
1
(U) if f is
dierentiable and the mapping
x Df (x) ,
is continuous as a function from U to L(X, Y ).
With this denition of partial derivatives, here is the major theorem.
Theorem 6.8.5 Let g, U,
n
i=1
X
i
, be given as in Denition 6.8.3. Then g is
C
1
(U) if and only if D
i
g exists and is continuous on U for each i. In this case, g is
dierentiable and
Dg (x) (v) =
k
D
k
g (x) v
k
(6.9)
where v = (v
1
, , v
n
) .
Proof: Suppose then that D
i
g exists and is continuous for each i. Note that
k
j=1
j
v
j
= (v
1
, , v
k
, 0, , 0) .
Thus
n
j=1
j
v
j
= v and dene
0
j=1
j
v
j
0. Therefore,
g (x +v) g (x) =
n
k=1
_
_
g
_
_
x+
k
j=1
j
v
j
_
_
g
_
_
x +
k1
j=1
j
v
j
_
_
_
_
(6.10)
Consider the terms in this sum.
g
_
_
x+
k
j=1
j
v
j
_
_
g
_
_
x +
k1
j=1
j
v
j
_
_
= g (x+
k
v
k
) g (x) + (6.11)
_
_
g
_
_
x+
k
j=1
j
v
j
_
_
g (x+
k
v
k
)
_
_
_
_
g
_
_
x +
k1
j=1
j
v
j
_
_
g (x)
_
_
(6.12)
and the expression in 6.12 is of the form h(v
k
) h(0) where for small w X
k
,
h(w) g
_
_
x+
k1
j=1
j
v
j
+
k
w
_
_
g (x +
k
w) .
Therefore,
Dh(w) = D
k
g
_
_
x+
k1
j=1
j
v
j
+
k
w
_
_
D
k
g (x +
k
w)
and by continuity, ||Dh(w)|| < provided ||v|| is small enough. Therefore, by Theorem
6.4.2, whenever ||v|| is small enough,
||h(v
k
) h(0)|| ||v
k
|| ||v||
which shows that since is arbitrary, the expression in 6.12 is o(v). Now in 6.11
g (x+
k
v
k
) g (x) = D
k
g (x) v
k
+o(v
k
) = D
k
g (x) v
k
+o(v) .
6.9. MIXED PARTIAL DERIVATIVES 135
Therefore, referring to 6.10,
g (x +v) g (x) =
n
k=1
D
k
g (x) v
k
+o(v)
which shows Dg (x) exists and equals the formula given in 6.9.
Next suppose g is C
1
. I need to verify that D
k
g (x) exists and is continuous. Let
v X
k
suciently small. Then
g (x +
k
v) g (x) = Dg (x)
k
v +o(
k
v)
= Dg (x)
k
v +o(v)
since ||
k
v|| = ||v||. Then D
k
g (x) exists and equals
Dg (x)
k
Now x Dg (x) is continuous. Since
k
is linear, it follows from Theorem 5.8.3 that
k
: X
k

n
i=1
X
i
is also continuous, This proves the theorem.
The way this is usually used is in the following corollary, a case of Theorem 6.8.5
obtained by letting X
i
= F in the above theorem.
Corollary 6.8.6 Let U be an open subset of F
n
and let f :U F
m
be C
1
in the sense
that all the partial derivatives of f exist and are continuous. Then f is dierentiable
and
f (x +v) = f (x) +
n
k=1
f
x
k
(x) v
k
+o(v) .
6.9 Mixed Partial Derivatives
Continuing with the special case where f is dened on an open set in F
n
, I will next
consider an interesting result due to Euler in 1734 about mixed partial derivatives. It
turns out that the mixed partial derivatives, if continuous will end up being equal.
Recall the notation
f
x
=
f
x
= D
e1
f
and
f
xy
=

2
f
yx
= D
e1e2
f.
Theorem 6.9.1 Suppose f : U F
2
R where U is an open set on which
f
x
, f
y
, f
xy
and f
yx
exist. Then if f
xy
and f
yx
are continuous at the point (x, y) U, it
follows
f
xy
(x, y) = f
yx
(x, y) .
Proof: Since U is open, there exists r > 0 such that B((x, y) , r) U. Now let
|t| , |s| < r/2, t, s real numbers and consider
(s, t)
1
st
{
h(t)
..
f (x +t, y +s) f (x +t, y)
h(0)
..
(f (x, y +s) f (x, y))}. (6.13)
Note that (x +t, y +s) U because
|(x +t, y +s) (x, y)| = |(t, s)| =
_
t
2
+s
2
_
1/2
_
r
2
4
+
r
2
4
_
1/2
=
r
2
< r.
136 THE DERIVATIVE
As implied above, h(t) f (x +t, y +s) f (x +t, y). Therefore, by the mean value
theorem and the (one variable) chain rule,
(s, t) =
1
st
(h(t) h(0)) =
1
st
h
(t) t
=
1
s
(f
x
(x +t, y +s) f
x
(x +t, y))
for some (0, 1) . Applying the mean value theorem again,
(s, t) = f
xy
(x +t, y +s)
where , (0, 1).
If the terms f (x +t, y) and f (x, y +s) are interchanged in 6.13, (s, t) is unchanged
and the above argument shows there exist , (0, 1) such that
(s, t) = f
yx
(x +t, y +s) .
Letting (s, t) (0, 0) and using the continuity of f
xy
and f
yx
at (x, y) ,
lim
(s,t)(0,0)
(s, t) = f
xy
(x, y) = f
yx
(x, y) .
The following is obtained from the above by simply xing all the variables except
for the two of interest.
Corollary 6.9.2 Suppose U is an open subset of X and f : U R has the property
that for two indices, k, l, f
x
k
, f
x
l
, f
x
l
x
k
, and f
x
k
x
l
exist on U and f
x
k
x
l
and f
x
l
x
k
are
both continuous at x U. Then f
x
k
x
l
(x) = f
x
l
x
k
(x) .
By considering the real and imaginary parts of f in the case where f has values in
C you obtain the following corollary.
Corollary 6.9.3 Suppose U is an open subset of F
n
and f : U F has the property
x
k
, f
x
l
, f
x
l
x
k
, and f
x
k
x
l
exist on U and f
x
k
x
l
and f
x
l
x
k
are
both continuous at x U. Then f
x
k
x
l
(x) = f
x
l
x
k
(x) .
Finally, by considering the components of f you get the following generalization.
Corollary 6.9.4 Suppose U is an open subset of F
n
and f : U F
m
has the property
x
k
, f
x
l
, f
x
l
x
k
, and f
x
k
x
l
exist on U and f
x
k
x
l
and f
x
l
x
k
are both
continuous at x U. Then f
x
k
x
l
(x) = f
x
l
x
k
(x) .
It is necessary to assume the mixed partial derivatives are continuous in order to
assert they are equal. The following is a well known example [3].
Example 6.9.5 Let
f (x, y) =
_
xy(x
2
y
2
)
x
2
+y
2
if (x, y) = (0, 0)
0 if (x, y) = (0, 0)
From the denition of partial derivatives it follows immediately that f
x
(0, 0) =
f
y
(0, 0) = 0. Using the standard rules of dierentiation, for (x, y) = (0, 0) ,
f
x
= y
x
4
y
4
+ 4x
2
y
2
(x
2
+y
2
)
2
, f
y
= x
x
4
y
4
4x
2
y
2
(x
2
+y
2
)
2
6.10. IMPLICIT FUNCTION THEOREM 137
Now
f
xy
(0, 0) lim
y0
f
x
(0, y) f
x
(0, 0)
y
= lim
y0
y
4
(y
2
)
2
= 1
while
f
yx
(0, 0) lim
x0
f
y
(x, 0) f
y
(0, 0)
x
= lim
x0
x
4
(x
2
)
2
= 1
showing that although the mixed partial derivatives do exist at (0, 0) , they are not equal
there.
6.10 Implicit Function Theorem
The following lemma is very useful.
Lemma 6.10.1 Let A L(X, X) where X is a nite dimensional normed vector
space and suppose ||A|| r < 1. Then
(I A)
1
exists (6.14)
and

(I A)
1
(1 r)
1
. (6.15)
Furthermore, if
I
_
A L(X, X) : A
1
exists
_
the map A A
1
is continuous on I and I is an open subset of L(X, X).
Proof: Let ||A|| r < 1. If (I A) x = 0, then x = Ax and so if x = 0,
||x|| = ||Ax|| ||A|| ||x|| < r ||x||
which is a contradiction. Therefore, (I A) is one to one. Hence it maps a basis of X
to a basis of X and is therefore, onto. Here is why. Let {v
1
, , v
n
} be a basis for X
and suppose
n
k=1
c
k
(I A) v
k
= 0.
Then
(I A)
_
n
k=1
c
k
v
k
_
= 0
and since (I A) is one to one, it follows
n
k=1
c
k
v
k
= 0
138 THE DERIVATIVE
which requires each c
k
= 0 because the {v
k
} are independent. Hence {(I A) v
k
}
n
k=1
is a basis for X because there are n of these vectors and every basis has the same size.
Therefore, if y X, there exist scalars, c
k
such that
y =
n
k=1
c
k
(I A) v
k
= (I A)
_
n
k=1
c
k
v
k
_
so (I A) is onto as claimed. Thus (I A)
1
L(X, X) and it remains to estimate
its norm.
||x Ax|| ||x|| ||Ax|| ||x|| ||A|| ||x|| ||x|| (1 r)
Letting y = x Ax so x = (I A)
1
y, this shows, since (I A) is onto that for all
y X,
||y||
(I A)
1
y
(1 r)
and so
(I A)
1
(1 r)
1
. This proves the rst part.
To verify the continuity of the inverse map, let A I. Then
B = A
_
I A
1
(AB)
_
and so if
A
1
(AB)
< 1 which, by Theorem 5.8.3, happens if

||AB|| < 1/
A
1
,
it follows from the rst part of this proof that
_
I A
1
(AB)
_
1
exists and so
B
1
=
_
I A
1
(AB)
_
1
A
1
which shows I is open. Also, if
A
1
(AB)
r < 1, (6.16)
B
1
A
1
(1 r)
1
Now for such B this close to A such that 6.16 holds,
B
1
A
1
B
1
(AB) A
1
||AB||
B
1
A
1
||AB||
A
1
2
(1 r)
1
which shows the map which takes a linear transformation in I to its inverse is continuous.
The next theorem is a very useful result in many areas. It will be used in this
section to give a short proof of the implicit function theorem but it is also useful in
studying dierential equations and integral equations. It is sometimes called the uniform
contraction principle.
Theorem 6.10.2 Let X, Y be nite dimensional normed vector spaces. Also let
E be a closed subset of X and F a closed subset of Y. Suppose for each (x, y) E F,
T(x, y) E and satises
||T(x, y) T(x
, y)|| r ||x x
|| (6.17)
where 0 < r < 1 and also
||T(x, y) T(x, y
)|| M||y y
|| . (6.18)
Then for each y F there exists a unique xed point for T(, y) , x E, satisfying
T(x, y) = x (6.19)
and also if x(y) is this xed point,
||x(y) x(y
)||
M
1 r
||y y
|| . (6.20)
Proof: First consider the claim there exists a xed point for the mapping, T(, y).
For a xed y, let g (x) T(x, y). Now pick any x
0
E and consider the sequence,
x
1
= g (x
0
) , x
k+1
= g (x
k
) .
Then by 6.17,
||x
k+1
x
k
|| = ||g (x
k
) g (x
k1
)|| r ||x
k
x
k1
||
r
2
||x
k1
x
k2
|| r
k
||g (x
0
) x
0
|| .
Now by the triangle inequality,
||x
k+p
x
k
||
p
i=1
||x
k+i
x
k+i1
||
i=1
r
k+i1
||g (x
0
) x
0
||
r
k
||g (x
0
) x
0
||
1 r
.
Since 0 < r < 1, this shows that {x
k
}
k=1
is a Cauchy sequence. Therefore, by com-
pleteness of E it converges to a point x E. To see x is a xed point, use the continuify
of g to obtain
x lim
k
x
k
= lim
k
x
k+1
= lim
k
g (x
k
) = g (x) .
This proves 6.19. To verify 6.20,
||x(y) x(y
)|| = ||T(x(y) , y) T(x(y
) , y
)||
||T(x(y) , y) T(x(y) , y
)|| +||T(x(y) , y
) T(x(y
) , y
)||
M||y y
|| +r ||x(y) x(y
)|| .
Thus
(1 r) ||x(y) x(y
)|| M||y y
|| .
This also shows the xed point for a given y is unique. This proves the theorem.
The implicit function theorem deals with the question of solving, f (x, y) = 0 for x
in terms of y and how smooth the solution is. It is one of the most important theorems
in mathematics. The proof I will give holds with no change in the context of innite
dimensional complete normed vector spaces when suitable modications are made on
what is meant by L(X, Y ) . There are also even more general versions of this theorem
than to normed vector spaces.
Recall that for X, Y normed vector spaces, the norm on X Y is of the form
||(x, y)|| = max (||x|| , ||y||) .
140 THE DERIVATIVE
Theorem 6.10.3 (implicit function theorem) Let X, Y, Z be nite dimensional
normed vector spaces and suppose U is an open set in X Y . Let f : U Z be in
C
1
(U) and suppose
f (x
0
, y
0
) = 0, D
1
f (x
0
, y
0
)
1
L(Z, X) . (6.21)
Then there exist positive constants, , , such that for every y B(y
0
, ) there exists a
unique x(y) B(x
0
, ) such that
f (x(y) , y) = 0. (6.22)
Furthermore, the mapping, y x(y) is in C
1
(B(y
0
, )).
Proof: Let T(x, y) x D
1
f (x
0
, y
0
)
1
f (x, y). Therefore,
D
1
T(x, y) = I D
1
f (x
0
, y
0
)
1
D
1
f (x, y) . (6.23)
by continuity of the derivative which implies continuity of D
1
T, it follows there exists
> 0 such that if ||(x x
0
, y y
0
)|| < , then
||D
1
T(x, y)|| <
1
2
. (6.24)
Also, it can be assumed is small enough that
D
1
f (x
0
, y
0
)
1
||D
2
f (x, y)|| < M (6.25)
where M >
D
1
f (x
0
, y
0
)
1
||D
2
f (x
0
, y
0
)||. By Theorem 6.4.2, whenever x, x

B(x
0
, ) and y B(y
0
, ),
||T(x, y) T(x
, y)||
1
2
||x x
|| . (6.26)
Solving 6.23 for D
1
f (x, y) ,
D
1
f (x, y) = D
1
f (x
0
, y
0
) (I D
1
T(x, y)) .
By Lemma 6.10.1 and the assumption that D
1
f (x
0
, y
0
)
1
exists, it follows, D
1
f (x, y)
1
exists and equals
(I D
1
T(x, y))
1
D
1
f (x
0
, y
0
)
1
By the estimate of Lemma 6.10.1 and 6.24,
D
1
f (x, y)
1
D
1
f (x
0
, y
0
)
1
. (6.27)
Next more restrictions are placed on y to make it even closer to y
0
. Let
0 < < min
_
,

3M
_
.
Then suppose x B(x
0
, ) and y B(y
0
, ). Consider
x D
1
f (x
0
, y
0
)
1
f (x, y) x
0
= T(x, y) x
0
g (x, y) .
D
1
g (x, y) = I D
1
f (x
0
, y
0
)
1
D
1
f (x, y) = D
1
T(x, y) ,
and
D
2
g (x, y) = D
1
f (x
0
, y
0
)
1
D
2
f (x, y) .
Also note that T(x, y) = x is the same as saying f (x
0
, y
0
) = 0 and also g (x
0
, y
0
) = 0.
Thus by 6.25 and Theorem 6.4.2, it follows that for such (x, y) B(x
0
, ) B(y
0
, ),
||T(x, y) x
0
|| = ||g (x, y)|| = ||g (x, y) g (x
0
, y
0
)||
||g (x, y) g (x, y
0
)|| +||g (x, y
0
) g (x
0
, y
0
)||
M||y y
0
|| +
1
2
||x x
0
|| <

2
+

3
=
5
6
< . (6.28)
Also for such (x, y
i
) , i = 1, 2, Theorem 6.4.2 and 6.25 implies
||T(x, y
1
) T(x, y
2
)|| =
D
1
f (x
0
, y
0
)
1
(f (x, y
2
) f (x, y
1
))
M||y
2
y
1
|| . (6.29)
From now on assume ||x x
0
|| < and ||y y
0
|| < so that 6.29, 6.27, 6.28, 6.26, and
6.25 all hold. By 6.29, 6.26, 6.28, and the uniform contraction principle, Theorem 6.10.2
applied to E B
_
x
0
,
5
6
_
and F B(y
0
, ) implies that for each y B(y
0
, ), there
exists a unique x(y) B(x
0
, ) (actually in B
_
x
0
,
5
6
_
) such that T(x(y) , y) = x(y)
which is equivalent to
f (x(y) , y) = 0.
Furthermore,
||x(y) x(y
)|| 2M||y y
|| . (6.30)
This proves the implicit function theorem except for the verication that y x(y)
is C
1
. This is shown next. Letting v be suciently small, Theorem 6.8.5 and Theorem
6.4.2 imply
0 = f (x(y +v) , y +v) f (x(y) , y) =
D
1
f (x(y) , y) (x(y +v) x(y)) +
+D
2
f (x(y) , y) v +o ((x(y +v) x(y) , v)) .
The last term in the above is o (v) because of 6.30. Therefore, using 6.27, solve the
above equation for x(y +v) x(y) and obtain
x(y +v) x(y) = D
1
(x(y) , y)
1
D
2
f (x(y) , y) v +o (v)
Which shows that y x(y) is dierentiable on B(y
0
, ) and
Dx(y) = D
1
f (x(y) , y)
1
D
2
f (x(y) , y) . (6.31)
Now it follows from the continuity of D
2
f , D
1
f , the inverse map, 6.30, and this formula
for Dx(y)that x() is C
1
(B(y
0
, )). This proves the theorem.
The next theorem is a very important special case of the implicit function theo-
rem known as the inverse function theorem. Actually one can also obtain the implicit
function theorem from the inverse function theorem. It is done this way in [28] and in
[2].
Theorem 6.10.4 (inverse function theorem) Let x
0
U, an open set in X ,
and let f : U Y where X, Y are nite dimensional normed vector spaces. Suppose
f is C
1
(U) , and Df (x
0
)
1
L(Y, X). (6.32)
142 THE DERIVATIVE
Then there exist open sets W, and V such that
x
0
W U, (6.33)
f : W V is one to one and onto, (6.34)
f
1
is C
1
, (6.35)
Proof: Apply the implicit function theorem to the function
F(x, y) f (x) y
where y
0
f (x
0
). Thus the function y x(y) dened in that theorem is f
1
. Now
let
W B(x
0
, ) f
1
(B(y
0
, ))
and
V B(y
0
, ) .
6.10.1 More Derivatives
In the implicit function theorem, suppose f is C
k
. Will the implicitly dened function
also be C
k
? It was shown above that this is the case if k = 1. In fact it holds for any
positive integer k.
First of all, consider D
2
f (x(y) , y) L(Y, Z) . Let {w
1
, , w
m
} be a basis for Y
and let {z
1
, , z
n
} be a basis for Z. Then D
2
f (x(y) , y) has a matrix with respect
to these bases. Thus conserving on notation, denote this matrix by
_
D
2
f (x(y) , y)
ij
_
.
Thus
D
2
f (x(y) , y) =
ij
D
2
f (x(y) , y)
ij
z
i
w
j
The scalar valued entries of the matrix of D
2
f (x(y) , y) have the same dierentiabil-
ity as the function y D
2
f (x(y) , y) . This is because the linear projection map,
ij
mapping L(Y, Z) to F given by
ij
L L
ij
, the ij
th
entry of the matrix of L with
respect to the given bases is continuous thanks to Theorem 5.8.3. Similar considera-
tions apply to D
1
f (x(y) , y) and the entries of its matrix, D
1
f (x(y) , y)
ij
taken with
respect to suitable bases. From the formula for the inverse of a matrix, Theorem 3.5.14,
the ij
th
entries of the matrix of D
1
f (x(y) , y)
1
, D
1
f (x(y) , y)
1
ij
also have the same
dierentiability as y D
1
f (x(y) , y).
Now consider the formula for the derivative of the implicitly dened function in 6.31,
Dx(y) = D
1
f (x(y) , y)
1
D
2
f (x(y) , y) . (6.36)
The above derivative is in L(Y, X) . Let {w
1
, , w
m
} be a basis for Y and let {v
1
, , v
n
}
be a basis for X. Letting x
i
be the i
th
component of x with respect to the basis for
X, it follows from Theorem 6.7.1, y x(y) will be C
k
if all such Gateaux derivatives,
D
wj
1
wj
2
wjr
x
i
(y) exist and are continuous for r k and for any i. Consider what is
required for this to happen. By 6.36,
D
wj
x
i
(y) =
k
_
D
1
f (x(y) , y)
1
_
ik
(D
2
f (x(y) , y))
kj
G
1
(x(y) , y) (6.37)
6.11. TAYLORS FORMULA 143
where (x, y) G
1
(x, y) is C
k1
because it is assumed f is C
k
and one derivative has
been taken to write the above. If k 2, then another Gateaux derivative can be taken.
D
wjw
k
x
i
(y) lim
t0
G
1
(x(y +tw
k
) , y +tw
k
) G
1
(x(y) , y)
t
= D
1
G
1
(x(y) , y) Dx(y) w
k
+D
2
G
1
(x(y) , y)
G
2
(x(y) , y,Dx(y))
Since a similar result holds for all i and any choice of w
j
, w
k
, this shows x is at least
C
2
. If k 3, then another Gateaux derivative can be taken because then (x, y, z)
G
2
(x, y, z) is C
1
and it has been established Dx is C
1
. Continuing this way shows
D
wj
1
wj
2
wjr
x
i
(y) exists and is continuous for r k. This proves the following corollary
to the implicit and inverse function theorems.
Corollary 6.10.5 In the implicit and inverse function theorems, you can replace
C
1
with C
k
in the statements of the theorems for any k N.
6.10.2 The Case Of R
n
In many applications of the implicit function theorem,
f : U R
n
R
m
R
n
and f (x
0
, y
0
) = 0 while f is C
1
. How can you recognize the condition of the implicit
function theorem which says D
1
f (x
0
, y
0
)
1
exists? This is really not hard. You recall
the matrix of the transformation D
1
f (x
0
, y
0
) with respect to the usual basis vectors is
_
_
_
f
1,x1
(x
0
, y
0
) f
1,xn
(x
0
, y
0
)
.
.
.
.
.
.
f
n,x1
(x
0
, y
0
) f
n,xn
(x
0
, y
0
)
_
_
_
and so D
1
f (x
0
, y
0
)
1
exists exactly when the determinant of the above matrix is
nonzero. This is the condition to check. In the general case, you just need to verify
D
1
f (x
0
, y
0
) is one to one and this can also be accomplished by looking at the matrix
of the transformation with respect to some bases on X and Z.
6.11 Taylors Formula
First recall the Taylor formula with the Lagrange form of the remainder. It will only
be needed on [0, 1] so that is what I will show.
Theorem 6.11.1 Let h : [0, 1] R have m + 1 derivatives. Then there exists
t (0, 1) such that
h(1) = h(0) +
m
k=1
h
(k)
(0)
k!
+
h
(m+1)
(t)
(m+ 1)!
.
Proof: Let K be a number chosen such that
h(1)
_
h(0) +
m
k=1
h
(k)
(0)
k!
+K
_
= 0
144 THE DERIVATIVE
Now the idea is to nd K. To do this, let
F (t) = h(1)
_
h(t) +
m
k=1
h
(k)
(t)
k!
(1 t)
k
+K (1 t)
m+1
_
Then F (1) = F (0) = 0. Therefore, by Rolles theorem there exists t between 0 and 1
such that F
(t) = 0. Thus,
0 = F
(t) = h
(t) +
m
k=1
h
(k+1)
(t)
k!
(1 t)
k
k=1
h
(k)
(t)
k!
k (1 t)
k1
K (m+ 1) (1 t)
m
And so
= h
(t) +
m
k=1
h
(k+1)
(t)
k!
(1 t)
k
m1
k=0
h
(k+1)
(t)
k!
(1 t)
k
K (m+ 1) (1 t)
m
= h
(t) +
h
(m+1)
(t)
m!
(1 t)
m
h
(t) K (m+ 1) (1 t)
m
and so
K =
h
(m+1)
(t)
(m+ 1)!
.
Now let f : U R where U X a normed vector space and suppose f C
m
(U).
Let x U and let r > 0 be such that
B(x,r) U.
Then for ||v|| < r consider
f (x+tv) f (x) h(t)
for t [0, 1]. Then by the chain rule,
h
(t) = Df (x+tv) (v) , h
(t) = D
2
f (x+tv) (v) (v)
and continuing in this way,
h
(k)
(t) = D
(k)
f (x+tv) (v) (v) (v) D
(k)
f (x+tv) v
k
.
It follows from Taylors formula for a function of one variable given above that
f (x +v) = f (x) +
m
k=1
D
(k)
f (x) v
k
k!
+
D
(m+1)
f (x+tv) v
m+1
(m+ 1)!
. (6.38)
This proves the following theorem.
Theorem 6.11.2 Let f : U R and let f C
m+1
(U). Then if
B(x,r) U,
and ||v|| < r, there exists t (0, 1) such that 6.38 holds.
6.11. TAYLORS FORMULA 145
6.11.1 Second Derivative Test
Now consider the case where U R
n
and f : U R is C
2
(U). Then from Taylors
theorem, if v is small enough, there exists t (0, 1) such that
f (x +v) = f (x) +Df (x) v+
D
2
f (x+tv) v
2
2
. (6.39)
Consider
D
2
f (x+tv) (e
i
) (e
j
) D(D(f (x+tv)) e
i
) e
j
= D
_
f (x +tv)
x
i
_
e
j
=

2
f (x +tv)
x
j
x
i
where e
i
are the usual basis vectors. Letting
v =
n
i=1
v
i
e
i
,
the second derivative term in 6.39 reduces to
1
2
i,j
D
2
f (x+tv) (e
i
) (e
j
) v
i
v
j
=
1
2
i,j
H
ij
(x+tv) v
i
v
j
where
H
ij
(x+tv) = D
2
f (x+tv) (e
i
) (e
j
) =

2
f (x+tv)
x
j
x
i
.
Denition 6.11.3 The matrix whose ij
th
entry is

2
f(x)
xjxi
is called the Hessian
matrix, denoted as H(x).
From Theorem 6.9.1, this is a symmetric real matrix, thus self adjoint. By the
continuity of the second partial derivative,
f (x +v) = f (x) +Df (x) v+
1
2
v
T
H (x) v+
1
2
_
v
T
(H (x+tv) H (x)) v
_
. (6.40)
where the last two terms involve ordinary matrix multiplication and
v
T
= (v
1
v
n
)
for v
i
the components of v relative to the standard basis.
Denition 6.11.4 Let f : D R where D is a subset of some normed vector
space. Then f has a local minimum at x D if there exists > 0 such that for all
y B(x, )
f (y) f (x) .
f has a local maximum at x D if there exists > 0 such that for all y B(x, )
f (y) f (x) .
146 THE DERIVATIVE
Theorem 6.11.5 If f : U R where U is an open subset of R
n
and f is C
2
,
suppose Df (x) = 0. Then if H (x) has all positive eigenvalues, x is a local minimum.
If the Hessian matrix H (x) has all negative eigenvalues, then x is a local maximum.
If H (x) has a positive eigenvalue, then there exists a direction in which f has a local
minimum at x, while if H (x) has a negative eigenvalue, there exists a direction in which
H (x) has a local maximum at x.
Proof: Since Df (x) = 0, formula 6.40 holds and by continuity of the second deriva-
tive, H (x) is a symmetric matrix. Thus H (x) has all real eigenvalues. Suppose rst
that H (x) has all positive eigenvalues and that all are larger than
2
> 0. Then by
Theorem 3.8.23, H (x) has an orthonormal basis of eigenvectors, {v
i
}
n
i=1
and if u is an
arbitrary vector, such that u =
n
j=1
u
j
v
j
where u
j
= u v
j
, then
u
T
H (x) u =
n
j=1
u
j
v
T
j
H (x)
n
j=1
u
j
v
j
=
n
j=1
u
2
j
j

2
n
j=1
u
2
j
=
2
|u|
2
.
From 6.40 and the continuity of H, if v is small enough,
f (x +v) f (x) +
1
2
2
|v|
2
1
4
2
|v|
2
= f (x) +

2
4
|v|
2
.
This shows the rst claim of the theorem. The second claim follows from similar rea-
soning. Suppose H (x) has a positive eigenvalue
2
. Then let v be an eigenvector for
this eigenvalue. Then from 6.40,
f (x+tv) = f (x) +
1
2
t
2
v
T
H (x) v+
1
2
t
2
_
v
T
(H (x+tv) H (x)) v
_
which implies
f (x+tv) = f (x) +
1
2
t
2
2
|v|
2
+
1
2
t
2
_
v
T
(H (x+tv) H (x)) v
_
f (x) +
1
4
t
2
2
|v|
2
whenever t is small enough. Thus in the direction v the function has a local minimum
at x. The assertion about the local maximum in some direction follows similarly. This
proves the theorem.
This theorem is an analogue of the second derivative test for higher dimensions. As
in one dimension, when there is a zero eigenvalue, it may be impossible to determine
from the Hessian matrix what the local qualitative behavior of the function is. For
example, consider
f
1
(x, y) = x
4
+y
2
, f
2
(x, y) = x
4
+y
2
.
Then Df
i
(0, 0) = 0 and for both functions, the Hessian matrix evaluated at (0, 0) equals
_
0 0
0 2
_
but the behavior of the two functions is very dierent near the origin. The second has
a saddle point while the rst has a minimum there.
6.12. THE METHOD OF LAGRANGE MULTIPLIERS 147
6.12 The Method Of Lagrange Multipliers
As an application of the implicit function theorem, consider the method of Lagrange
multipliers from calculus. Recall the problem is to maximize or minimize a function
subject to equality constraints. Let f : U R be a C
1
function where U R
n
and let
g
i
(x) = 0, i = 1, , m (6.41)
be a collection of equality constraints with m < n. Now consider the system of nonlinear
equations
f (x) = a
g
i
(x) = 0, i = 1, , m.
x
0
is a local maximum if f (x
0
) f (x) for all x near x
0
which also satises the
constraints 6.41. A local minimum is dened similarly. Let F : U R R
m+1
be
dened by
F(x,a)
_
_
_
_
_
f (x) a
g
1
(x)
.
.
.
g
m
(x)
_
_
_
_
_
. (6.42)
Now consider the m + 1 n Jacobian matrix, the matrix of the linear transformation,
D
1
F(x, a) with respect to the usual basis for R
n
and R
m+1
.
_
_
_
_
_
f
x1
(x
0
) f
xn
(x
0
)
g
1x1
(x
0
) g
1xn
(x
0
)
.
.
.
.
.
.
g
mx1
(x
0
) g
mxn
(x
0
)
_
_
_
_
_
.
If this matrix has rank m + 1 then some m + 1 m + 1 submatrix has nonzero deter-
minant. It follows from the implicit function theorem that there exist m+ 1 variables,
x
i1
, , x
im+1
such that the system
F(x,a) = 0 (6.43)
species these m+1 variables as a function of the remaining n (m+ 1) variables and
a in an open set of R
nm
. Thus there is a solution (x,a) to 6.43 for some x close to x
0
whenever a is in some open interval. Therefore, x
0
cannot be either a local minimum or
a local maximum. It follows that if x
0
is either a local maximum or a local minimum,
then the above matrix must have rank less than m + 1 which, by Corollary 3.5.20,
requires the rows to be linearly dependent. Thus, there exist m scalars,
1
, ,
m
,
and a scalar , not all zero such that
_
_
_
f
x1
(x
0
)
.
.
.
f
xn
(x
0
)
_
_
_ =
1
_
_
_
g
1x1
(x
0
)
.
.
.
g
1xn
(x
0
)
_
_
_+ +
m
_
_
_
g
mx1
(x
0
)
.
.
.
g
mxn
(x
0
)
_
_
_. (6.44)
If the column vectors
_
_
_
g
1x1
(x
0
)
.
.
.
g
1xn
(x
0
)
_
_
_,
_
_
_
g
mx1
(x
0
)
.
.
.
g
mxn
(x
0
)
_
_
_ (6.45)
148 THE DERIVATIVE
are linearly independent, then, = 0 and dividing by yields an expression of the form
_
_
_
f
x1
(x
0
)
.
.
.
f
xn
(x
0
)
_
_
_ =
1
_
_
_
g
1x1
(x
0
)
.
.
.
g
1xn
(x
0
)
_
_
_+ +
m
_
_
_
g
mx1
(x
0
)
.
.
.
g
mxn
(x
0
)
_
_
_ (6.46)
at every point x
0
which is either a local maximum or a local minimum. This proves the
following theorem.
Theorem 6.12.1 Let U be an open subset of R
n
and let f : U R be a C
1
function. Then if x
0
U is either a local maximum or local minimum of f subject to
the constraints 6.41, then 6.44 must hold for some scalars ,
1
, ,
m
not all equal to
zero. If the vectors in 6.45 are linearly independent, it follows that an equation of the
form 6.46 holds.
6.13 Exercises
1. Suppose L L(X, Y ) and suppose L is one to one. Show there exists r > 0 such
that for all x X,
||Lx|| r ||x|| .
Hint: You might argue that |||x||| ||Lx|| is a norm.
2. Show every polynomial,
||k
d
is C
k
for every k.
3. If f : U R where U is an open set in X and f is C
2
, show the mixed Gateaux
derivatives, D
v1v2
f (x) and D
v2v1
f (x) are equal.
4. Give an example of a function which is dierentiable everywhere but at some
point it fails to have continuous partial derivatives. Thus this function will be an
example of a dierentiable function which is not C
1
.
5. The existence of partial derivatives does not imply continuity as was shown in an
example. However, much more can be said than this. Consider
f (x, y) =
_
(x
2
y
4
)
2
(x
2
+y
4
)
2
if (x, y) = (0, 0) ,
1 if (x, y) = (0, 0) .
Show each Gateaux derivative, D
v
f (0) exists and equals 0 for every v. Also show
each Gateaux derivative exists at every other point in R
2
. Now consider the curve
x
2
= y
4
and the curve y = 0 to verify the function fails to be continuous at (0, 0).
This is an example of an everywhere Gateaux dierentiable function which is not
dierentiable and not continuous.
6. Let f be a real valued function dened on R
2
by
f (x, y)
_
x
3
y
3
x
2
+y
2
if (x, y) = (0, 0)
0 if (x, y) = (0, 0)
Determine whether f is continuous at (0, 0). Find f
x
(0, 0) and f
y
(0, 0) . Are the
partial derivatives of f continuous at (0, 0)? Find D
(u,v)
f ((0, 0)) , lim
t0
f(t(u,v))
t
.
Is the mapping (u, v) D
(u,v)
f ((0, 0)) linear? Is f dierentiable at (0, 0)?
6.13. EXERCISES 149
7. Let f : V R where V is a nite dimensional normed vector space. Suppose f is
convex which means
f (tx + (1 t) y) tf (x) + (1 t) f (y)
whenever t [0, 1]. Suppose also that f is dierentiable. Show then that for every
x, y V,
(Df (x) Df (y)) (x y) 0.
8. Suppose f : U V F where U is an open subset of V, a nite dimensional inner
product space with the inner product denoted by (, ) . Suppose f is dierentiable.
Show there exists a unique vector v (x) V such that
(u v (x)) = Df (x) u.
This special vector is called the gradient and is usually denoted by f (x) . Hint:
You might review the Riesz representation theorem presented earlier.
9. Suppose f : U Y where U is an open subset of X, a nite dimensional normed
vector space. Suppose that for all v X, D
v
f (x) exists. Show that whenever
a F D
av
f (x) = aD
v
f (x). Explain why if x D
v
f (x) is continuous then v
D
v
f (x) is linear. Show that if f is dierentiable at x, then D
v
f (x) = Df (x) v.
10. Suppose B is an open ball in X and f : B Y is dierentiable. Suppose also
there exists L L(X, Y ) such that
||Df (x) L|| < k
for all x B. Show that if x
1
, x
2
B,
|f (x
1
) f (x
2
) L(x
1
x
2
)| k |x
1
x
2
| .
Hint: Consider Tx = f (x)Lx and argue ||DT (x)|| < k. Then consider Theorem
6.4.2.
11. Let U be an open subset of X, f : U Y where X, Y are nite dimensional
normed vector spaces and suppose f C
1
(U) and Df (x
0
) is one to one. Then
show f is one to one near x
0
. Hint: Show using the assumption that f is C
1
that
there exists > 0 such that if
x
1
, x
2
B(x
0
, ) ,
then
|f (x
1
) f (x
2
) Df (x
0
) (x
1
x
2
)|
r
2
|x
1
x
2
| (6.47)
then use Problem 1.
12. Suppose M L(X, Y ) and suppose M is onto. Show there exists L L(Y, X)
such that
LMx =Px
where P L(X, X), and P
2
= P. Also show L is one to one and onto. Hint:
Let {y
1
, , y
m
} be a basis of Y and let Mx
i
= y
i
. Then dene
Ly =
m
i=1
i
x
i
where y =
m
i=1
i
y
i
.
150 THE DERIVATIVE
Show {x
1
, , x
m
} is a linearly independent set and show you can obtain {x
1
, , x
m
, , x
n
},
a basis for X in which Mx
j
= 0 for j > m. Then let
Px
m
i=1
i
x
i
where
x =
m
i=1
i
x
i
.
13. This problem depends on the result of Problem 12. Let f : U X Y, f is
C
1
, and Df (x) is onto for each x U. Then show f maps open subsets of U
onto open sets in Y . Hint: Let P = LDf (x) as in Problem 12. Argue L maps
open sets from Y to open sets of the vector space X
1
PX and L
1
maps open
sets from X
1
to open sets of Y. Then Lf (x +v) = Lf (x) + LDf (x) v +o(v) .
Now for z X
1
, let h(z) = Lf (x +z) Lf (x) . Then h is C
1
on some small
open subset of X
1
containing 0 and Dh(0) = LDf (x) which is seen to be one
to one and onto and in L(X
1
, X
1
) . Therefore, if r is small enough, h(B(0,r))
equals an open set in X
1
, V. This is by the inverse function theorem. Hence
L(f (x +B(0,r)) f (x)) = V and so f (x +B(0,r)) f (x) = L
1
(V ) , an open
set in Y.
14. Suppose U R
2
is an open set and f : U R
3
is C
1
. Suppose Df (s
0
, t
0
) has
rank two and
f (s
0
, t
0
) =
_
_
x
0
y
0
z
0
_
_
.
Show that for (s, t) near (s
0
, t
0
), the points f (s, t) may be realized in one of the
following forms.
{(x, y, (x, y)) : (x, y) near (x
0
, y
0
)},
{((y, z) , y, z) : (y, z) near (y
0
, z
0
)},
or
{(x, (x, z) , z, ) : (x, z) near (x
0
, z
0
)}.
This shows that parametrically dened surfaces can be obtained locally in a par-
ticularly simple form.
15. Let f : U Y , Df (x) exists for all x U, B(x
0
, ) U, and there exists
L L(X, Y ), such that L
1
L(Y, X), and for all x B(x
0
, )
||Df (x) L|| <
r
||L
1
||
, r < 1.
Show that there exists > 0 and an open subset of B(x
0
, ) , V , such that
f : V B(f (x
0
) , ) is one to one and onto. Also Df
1
(y) exists for each y
B(f (x
0
) , ) and is given by the formula
Df
1
(y) =
_
Df
_
f
1
(y)
_
1
.
Hint: Let
T
y
(x) T (x, y) xL
1
(f (x) y)
for |y f (x
0
)| <
(1r)
2||L
1
||
, consider {T
n
y
(x
0
)}. This is a version of the inverse
function theorem for f only dierentiable, not C
1
.
6.13. EXERCISES 151
16. Recall the n
th
derivative can be considered a multilinear function dened on X
n
with values in some normed vector space. Now dene a function denoted as
w
i
v
j1
v
jn
which maps X
n
Y in the following way
w
i
v
j1
v
jn
(v
k1
, , v
kn
) w
i
j1k1
j2k2

jnkn
(6.48)
and w
i
v
j1
v
jn
is to be linear in each variable. Thus, for
_
n
k1=1
a
k1
v
k1
, ,
n
kn=1
a
kn
v
kn
_
X
n
,
w
i
v
j1
v
jn
_
n
k1=1
a
k1
v
k1
, ,
n
kn=1
a
kn
v
kn
_
k1k2kn
w
i
(a
k1
a
k2
a
kn
)
j1k1
j2k2

jnkn
= w
i
a
j1
a
j2
a
jn
(6.49)
Show each w
i
v
j1
v
jn
is an n linear Y valued function. Next show the set of n
linear Y valued functions is a vector space and these special functions, w
i
v
j1
v
jn
for all choices of i and the j
k
is a basis of this vector space. Find the dimension
of the vector space.
17. Minimize
n
j=1
x
j
subject to the constraint
n
j=1
x
2
j
= a
2
. Your answer should
be some function of a which you may assume is a positive number.
18. Find the point, (x, y, z) on the level surface, 4x
2
+ y
2
z
2
= 1which is closest to
(0, 0, 0) .
19. A curve is formed from the intersection of the plane, 2x + 3y + z = 3 and the
cylinder x
2
+y
2
= 4. Find the point on this curve which is closest to (0, 0, 0) .
20. A curve is formed from the intersection of the plane, 2x + 3y + z = 3 and the
sphere x
2
+y
2
+z
2
= 16. Find the point on this curve which is closest to (0, 0, 0) .
21. Find the point on the plane, 2x+3y +z = 4 which is closest to the point (1, 2, 3) .
22. Let A = (A
ij
) be an n n matrix which is symmetric. Thus A
ij
= A
ji
and recall
(Ax)
i
= A
ij
x
j
where as usual sum over the repeated index. Show

xi
(A
ij
x
j
x
i
) =
2A
ij
x
j
. Show that when you use the method of Lagrange multipliers to maximize
the function, A
ij
x
j
x
i
subject to the constraint,
n
j=1
x
2
j
= 1, the value of which
corresponds to the maximum value of this functions is such that A
ij
x
j
= x
i
.
Thus Ax = x. Thus is an eigenvalue of the matrix, A.
23. Let x
1
, , x
5
be 5 positive numbers. Maximize their product subject to the
constraint that
x
1
+ 2x
2
+ 3x
3
+ 4x
4
+ 5x
5
= 300.
24. Let f (x
1
, , x
n
) = x
n
1
x
n1
2
x
1
n
. Then f achieves a maximum on the set,
S
_
x R
n
:
n
i=1
ix
i
= 1 and each x
i
0
_
.
If x S is the point where this maximum is achieved, nd x
1
/x
n
.
152 THE DERIVATIVE
25. Let (x, y) be a point on the ellipse, x
2
/a
2
+y
2
/b
2
= 1 which is in the rst quadrant.
Extend the tangent line through (x, y) till it intersects the x and y axes and let
A(x, y) denote the area of the triangle formed by this line and the two coordinate
axes. Find the minimum value of the area of this triangle as a function of a and
b.
26. Maximize
n
i=1
x
2
i
( x
2
1
x
2
2
x
2
3
x
2
n
) subject to the constraint,
n
i=1
x
2
i
=
r
2
. Show the maximum is
_
r
2
/n
_
n
. Now show from this that
_
n
i=1
x
2
i
_
1/n
1
n
n
i=1
x
2
i
and nally, conclude that if each number x
i
0, then
_
n
i=1
x
i
_
1/n
1
n
n
i=1
x
i
and there exist values of the x
i
for which equality holds. This says the geometric
mean is always smaller than the arithmetic mean.
27. Maximize x
2
y
2
subject to the constraint
x
2p
p
+
y
2q
q
= r
2
where p, q are real numbers larger than 1 which have the property that
1
p
+
1
q
= 1.
show the maximum is achieved when x
2p
= y
2q
and equals r
2
. Now conclude that
if x, y > 0, then
xy
x
p
p
+
y
q
q
and there are values of x and y where this inequality is an equation.
Measures And Measurable
Functions
The integral to be discussed next is the Lebesgue integral. This integral is more general
than the Riemann integral of beginning calculus. It is not as easy to dene as this
integral but is vastly superior in every application. In fact, the Riemann integral has
been obsolete for over 100 years. There exist convergence theorems for this integral
which are not available for the Riemann integral and unlike the Riemann integral, the
Lebesgue integral generalizes readily to abstract settings used in probability theory.
Much of the analysis done in the last 100 years applies to the Lebesgue integral. For
these reasons, and because it is very easy to generalize the Lebesgue integral to functions
of many variables I will present the Lebesgue integral here. First it is convenient to
discuss outer measures, measures, and measurable function in a general setting.
7.1 Compact Sets
This is a good place to put an important theorem about compact sets. The denition
of what is meant by a compact set follows.
Denition 7.1.1 Let U denote a collection of open sets in a normed vector
space. Then U is said to be an open cover of a set K if K U. Let K be a subset of
a normed vector space. Then K is compact if whenever U is an open cover of K there
exist nitely many sets of U, {U
1
, , U
m
} such that
K
m
k=1
U
k
.
In words, every open cover admits a nite subcover.
It was shown earlier that in any nite dimensional normed vector space the closed
and bounded sets are those which are sequentially compact. The next theorem says that
in any normed vector space, sequentially compact and compact are the same.
1
First
here is a very interesting lemma about the existence of something called a Lebesgue
number, the number r in the next lemma.
Lemma 7.1.2 Let K be a sequentially compact set in a normed vector space and let
U be an open cover of K. Then there exists r > 0 such that if x K, then B(x, r) is a
subset of some set of U.
1
Actually, this is true more generally than for normed vector spaces. It is also true for metric spaces,
those on which there is a distance dened.
153
154 MEASURES AND MEASURABLE FUNCTIONS
Proof: Suppose no such r exists. Then in particular, 1/n does not work for each
n N. Therefore, there exists x
n
K such that B(x
n
, r) is not a subset of any of the
sets of U. Since K is sequentially compact, there exists a subsequence, {x
n
k
} converging
to a point x of K. Then there exists r > 0 such that B(x, r) U U because U is an
open cover. Also x
n
k
B(x,r/2) for all k large enough and also for all k large enough,
1/n
k
< r/2. Therefore, there exists x
n
k
B(x,r/2) and 1/n
k
< r/2. But this is a
contradiction because
B(x
n
k
, 1/n
k
) B(x, r) U
contrary to the choice of x
n
k
which required B(x
n
k
, 1/n
k
) is not contained in any set
of U. This proves the lemma.
Theorem 7.1.3 Let K be a set in a normed vector space. Then K is compact if
and only if K is sequentially compact. In particular if K is a closed and bounded subset
of a nite dimensional normed vector space, then K is compact.
Proof: Suppose rst K is sequentially compact and let U be an open cover. Let r
be a Lebesgue number as described in Lemma 7.1.2. Pick x
1
K. Then B(x
1
, r) U
1
for some U
1
U. Suppose {B(x
i
, r)}
m
i=1
have been chosen such that
B(x
i
, r) U
i
U.
If their union contains K then {U
i
}
m
i=1
is a nite subcover of U. If {B(x
i
, r)}
m
i=1
does
not cover K, then there exists x
m+1
/
m
i=1
B(x
i
, r) and so B(x
m+1
, r) U
m+1
U.
This process must stop after nitely many choices of B(x
i
, r) because if not, {x
k
}
k=1
would have a subsequence which converges to a point of K which cannot occur because
whenever i = j,
||x
i
x
j
|| > r
Therefore, eventually
K
m
k=1
B(x
k
, r)
m
k=1
U
k
.
this proves one half of the theorem.
Now suppose K is compact. I need to show it is sequentially compact. Suppose it
is not. Then there exists a sequence, {x
k
} which has no convergent subsequence. This
requires that {x
k
} have no limit point for if it did have a limit point, x, then B(x, 1/n)
would contain innitely many distinct points of {x
k
} and so a subsequence of {x
k
}
converging to x could be obtained. Also no x
k
is repeated innitely often because if
there were such, a convergent subsequence could be obtained. Hence
k=m
{x
k
} C
m
is a closed set, closed because it contains all its limit points. (It has no limit points so
it contains them all.) Then letting U
m
= C
C
m
, it follows {U
m
} is an open cover of K
which has no nite subcover. Thus K must be sequentially compact after all.
If K is a closed and bounded set in a nite dimensional normed vector space, then K
is sequentially compact by Theorem 5.8.4. Therefore, by the rst part of this theorem,
it is sequentially compact. This proves the theorem.
Summarizing the above theorem along with Theorem 5.8.4 yields the following corol-
lary which is often called the Heine Borel theorem.
Corollary 7.1.4 Let X be a nite dimensional normed vector space and let K X.
Then the following are equivalent.
1. K is closed and bounded.
2. K is sequentially compact.
3. K is compact.
7.2. AN OUTER MEASURE ON P (R) 155
7.2 An Outer Measure On P (R)
A measure on R is like length. I will present something more general because it is no
trouble to do so and the generalization is useful in many areas of mathematics such as
probability. Recall that P (S) denotes the set of all subsets of S.
Theorem 7.2.1 Let F be an increasing function dened on R, an integrator
function. There exists a function : P (R) [0, ] which satises the following
properties.
1. If A B, then 0 (A) (B) , () = 0.
2. (
k=1
A
i
)
i=1
(A
i
)
3. ([a, b]) = F (b+) F (a) ,
4. ((a, b)) = F (b) F (a+)
5. ((a, b]) = F (b+) F (a+)
6. ([a, b)) = F (b) F (a) where
F (b+) lim
tb+
F (t) , F (b) lim
ta
F (t) .
Proof: First it is necessary to dene the function, . This is contained in the
following denition.
Denition 7.2.2 For A R,
(A) = inf
_
_
_
j=1
(F (b
i
) F (a
i
+)) : A
i=1
(a
i
, b
i
)
_
_
_
In words, you look at all coverings of A with open intervals. For each of these
open coverings, you add the lengths of the individual open intervals and you take the
inmum of all such numbers obtained.
Then 1.) is obvious because if a countable collection of open intervals covers B then
it also covers A. Thus the set of numbers obtained for B is smaller than the set of
numbers for A. Why is () = 0? Pick a point of continuity of F. Such points exist
because F is increasing and so it has only countably many points of discontinuity. Let
a be this point. Then (a , a +) and so () 2 for every > 0.
Consider 2.). If any (A
i
) = , there is nothing to prove. The assertion simply is
. Assume then that (A
i
) < for all i. Then for each m N there exists a
countable set of open intervals, {(a
m
i
, b
m
i
)}
i=1
such that
(A
m
) +

2
m
>
i=1
(F (b
m
i
) F (a
m
i
+)) .
Then using Theorem 2.3.4 on Page 22,
(
m=1
A
m
)
im
(F (b
m
i
) F (a
m
i
+))
=
m=1
i=1
(F (b
m
i
) F (a
m
i
+))
m=1
(A
m
) +

2
m
=
m=1
(A
m
) +
and since is arbitrary, this establishes 2.).
Next consider 3.). By denition, there exists a sequence of open intervals, {(a
i
, b
i
)}
i=1
whose union contains [a, b] such that
([a, b]) +
i=1
(F (b
i
) F (a
i
+))
By Theorem 7.1.3, nitely many of these intervals also cover [a, b]. It follows there exists
nitely many of these intervals, {(a
i
, b
i
)}
n
i=1
which overlap such that a (a
1
, b
1
) , b
1

(a
2
, b
2
) , , b (a
n
, b
n
) . Therefore,
([a, b])
n
i=1
(F (b
i
) F (a
i
+))
It follows
n
i=1
(F (b
i
) F (a
i
+)) ([a, b])
i=1
(F (b
i
) F (a
i
+))
F (b+) F (a)
([a, b]) F (b+) F (a)
but also, from the denition, the following inequality holds for all > 0.
([a, b]) F ((b +) ) F ((a ) +) F (b +) F (a )
Therefore, letting 0 yields
([a, b]) F (b+) F (a)
This establishes 3.).
Consider 4.). For small > 0,
([a +, b ]) ((a, b)) ([a, b]) .
Therefore, from 3.) and the denition of ,
F ((b )) F ((a +)) F ((b ) +) F ((a +) )
= ([a +, b ]) ((a, b)) F (b) F (a+)
Now letting decrease to 0 it follows
F (b) F (a+) ((a, b)) F (b) F (a+)
This shows 4.)
Consider 5.). From 3.) and 4.), for small > 0,
F (b+) F ((a +))
F (b+) F ((a +) )
= ([a +, b]) ((a, b])
((a, b +)) = F ((b +) ) F (a+)
F (b +) F (a+) .
7.3. GENERAL OUTER MEASURES AND MEASURES 157
Now let converge to 0 from above to obtain
F (b+) F (a+) = ((a, b]) = F (b+) F (a+) .
This establishes 5.) and 6.) is entirely similar to 5.). This proves the theorem.
Denition 7.2.3 Let be a nonempty set. A function mapping P () [0, ]
is called an outer measure if it satises the conditions 1.) and 2.) in Theorem 7.2.1.
7.3 General Outer Measures And Measures
First the general concept of a measure will be presented. Then it will be shown how
to get a measure from any outer measure. Using the outer measure just obtained,
this yields Lebesgue Stieltjes measure on R. Then an abstract Lebesgue integral and
its properties will be presented. After this the theory is specialized to the situation
of R and the outer measure in Theorem 7.2.1. This will yield the Lebesgue Stieltjes
integral on R along with spectacular theorems about its properties. The generalization
to Lebesgue integration on R
n
turns out to be very easy.
7.3.1 Measures And Measure Spaces
First here is a denition of a measure.
Denition 7.3.1 S P () is called a algebra , pronounced sigma algebra,
if
, S,
If E S then E
C
S
and
If E
i
S, for i = 1, 2, , then
i=1
E
i
S.
A function : S [0, ] where S is a algebra is called a measure if whenever
{E
i
}
i=1
S and the E
i
are disjoint, then it follows
j=1
E
j
_
=
j=1
(E
j
) .
The triple (, S, ) is often called a measure space. Sometimes people refer to (, S) as
a measurable space, making no reference to the measure. Sometimes (, S) may also be
called a measure space.
Theorem 7.3.2 Let {E
m
}
m=1
be a sequence of measurable sets in a measure
space (, F, ). Then if E
n
E
n+1
E
n+2
,
(
i=1
E
i
) = lim
n
(E
n
) (7.1)
and if E
n
E
n+1
E
n+2
and (E
1
) < , then
(
i=1
E
i
) = lim
n
(E
n
). (7.2)
Stated more succinctly, E
k
E implies (E
k
) (E) and E
k
E with (E
1
) <
implies (E
k
) (E).
Proof: First note that
i=1
E
i
= (
i=1
E
C
i
)
C
F so
i=1
E
i
is measurable. Also
note that for A and B sets of F, A \ B
_
A
C
B
_
C
F. To show 7.1, note that 7.1
is obviously true if (E
k
) = for any k. Therefore, assume (E
k
) < for all k. Thus
(E
k+1
\ E
k
) +(E
k
) = (E
k+1
)
and so
(E
k+1
\ E
k
) = (E
k+1
) (E
k
).
Also,
k=1
E
k
= E
1
k=1
(E
k+1
\ E
k
)
and the sets in the above union are disjoint. Hence
(
i=1
E
i
) = (E
1
) +
k=1
(E
k+1
\ E
k
) = (E
1
)
+
k=1
(E
k+1
) (E
k
)
= (E
1
) + lim
n
n
k=1
(E
k+1
) (E
k
) = lim
n
(E
n+1
).
This shows part 7.1.
To verify 7.2,
(E
1
) = (
i=1
E
i
) +(E
1
\
i=1
E
i
)
since (E
1
) < , it follows (
i=1
E
i
) < . Also, E
1
\
n
i=1
E
i
E
1
\
i=1
E
i
and so by
7.1,
(E
1
) (
i=1
E
i
) = (E
1
\
i=1
E
i
) = lim
n
(E
1
\
n
i=1
E
i
)
= (E
1
) lim
n
(
n
i=1
E
i
) = (E
1
) lim
n
(E
n
),
Hence, subtracting (E
1
) from both sides,
lim
n
(E
n
) = (
i=1
E
i
).
The following denition is important.
Denition 7.3.3 If something happens except for on a set of measure zero, then
it is said to happen a.e. almost everywhere. For example, {f
k
(x)} is said to converge
to f (x) a.e. if there is a set of measure zero, N such that if x N, then f
k
(x) f (x).
7.4 The Borel Sets, Regular Measures
7.4.1 Denition of Regular Measures
It is important to consider the interaction between measures and open and compact
sets. This involves the concept of a regular measure.
Denition 7.4.1 Let Y be a closed subset of X a nite dimensional normed
vector space. The closed sets in Y are the intersections of closed sets in X with Y . The
open sets in Y are intersections of open sets of X with Y . Now let F be a algebra of
sets of Y and let be a measure dened on F. Then is said to be a regular measure
if the following two conditions hold.
7.4. THE BOREL SETS, REGULAR MEASURES 159
1. For every F F
(F) = sup {(K) : K F and K is compact} (7.3)
2. For every F F
(F) = inf {(V ) : V F and V is open in Y } (7.4)
The rst of the above conditions is called inner regularity and the second is called
outer regularity.
Proposition 7.4.2 In the above situation, a set, K Y is compact in Y if and
only if it is compact in X.
Proof: If K is compact in X and K Y, let U be an open cover of K of sets open
in Y. This means U = {Y V : V V} where V is an open cover of K consisting of
sets open in X. Therefore, V admits a nite subcover, {V
1
, , V
m
} and consequently,
{Y V
1
, , Y V
m
} is a nite subcover from U. Thus K is compact in Y.
Now suppose K is compact in Y . This means that if U is an open cover of sets
open in Y it admitts a nite subcover. Now let V be any open cover of K, consisting
of sets open in X. Then U {V Y : V V} is a cover consisting of sets open in Y
and by denition, this admitts a nite subcover, {Y V
1
, , Y V
m
} but this implies
{V
1
, , V
m
} is also a nite subcover consisting of sets of V. This proves the proposition.
7.4.2 The Borel Sets

If Y is a closed subset of X, a normed vector space, denote by B (Y ) the smallest
algebra of subsets of Y which contains all the open sets of Y . To see such a smallest
algebra exists, let H denote the set of all algebras which contain the open sets P (Y ),
the set of all subsets of Y is one such algebra. Dene B (Y ) H. Then B (Y ) is a
algebra because , Y are both open sets in Y and so they are in each algebra of H.
If F B (Y ), then F is a set of every algebra of H and so F
C
is also a set of every
algebra of H. Thus F
C
B (Y ). If {F
i
} is a sequence of sets of B (Y ), then {F
i
} is
a sequence of sets of every algebra of H and so
i
F
i
is a set in every algebra of H
which implies
i
F
i
B (Y ) so B (Y ) is a algebra as claimed. From its denition, it is
the smallest algebra which contains the open sets.
7.4.3 Borel Sets And Regularity
To illustrate how nice the Borel sets are, here are some interesting results about regu-
larity. The rst Lemma holds for any algebra, not just the Borel sets. Here is some
notation which will be used. Let
S (0,r) {x Y : ||x|| = r}
D(0,r) {x Y : ||x|| r}
B(0,r) {x Y : ||x|| < r}
Thus S (0,r) is a closed set as is D(0,r) while B(0,r) is an open set. These are closed
or open as stated in Y . Since S (0,r) and D(0,r) are intersections of closed sets Y and
a closed set in X, these are also closed in X. Of course B(0, r) might not be open in
X. This would happen if Y has empty interior in X for example. However, S (0,r) and
D(0,r) are compact.
Lemma 7.4.3 Let Y be a closed subset of X a nite dimensional normed vector
space and let S be a algebra of sets of Y containing the open sets of Y . Suppose is
a measure dened on S and suppose also (K) < whenever K is compact. Then if
7.4 holds, so does 7.3.
Proof: It is desired to show that in this setting outer regularity implies inner
regularity. First suppose F D(0, n) where n N and F S. The following diagram
will help to follow the technicalities. In this picture, V is the material between the two
dotted curves, F is the inside of the solid curve and D(0,n) is inside the larger solid
curve.
D(0, n) \ F
F F V V
The idea is to use outer regularity on D(0, n) \F to come up with V approximating
this set as suggested in the picture. Then V
C
D(0, n) is a compact set contained in F
which approximates F. On the picture, the error is represented by the material between
the small dotted curve and the smaller solid curve which is less than the error between
V and D(0, n) \ F as indicated by the picture. If you need the details, they follow.
Otherwise the rest of the proof starts at
Taking complements with respect to Y
D(0, n) \ F = D(0, n) F
C
=
_
D(0, n)
C
F
_
C
S
because it is given that S contains the open sets. By 7.4 there exists an open set,
V D(0, n) \ F such that
(D(0, n) \ F) + > (V ) . (7.5)
Since is a measure,
(V \ (D(0, n) \ F)) +(D(0, n) \ F) = (V )
and so from 7.5
(V \ (D(0, n) \ F)) < (7.6)
Note
V \ (D(0, n) \ F) = V
_
D(0, n) F
C
_
C
=
_
V D(0, n)
C
_
(V F)
and by 7.6,
(V \ (D(0, n) \ F)) <
so in particular,
(V F) < .
Now
V D(0, n) F
C
and so
V
C
D(0, n)
C
F
which implies
V
C
D(0, n) F D(0, n) = F
Since F D(0, n) ,
_
F \
_
V
C
D(0, n)
__
=
_
F
_
V
C
D(0, n)
_
C
_
=
_
(F V )
_
F D(0, n)
C
__
= (F V ) <
showing the compact set, V
C
D(0, n) is contained in F and
_
V
C
D(0, n)
_
+ > (F) .
In the general case where F is only given to be in S, let F
n
= B(0, n) F. Then by
7.1, if l < (F) is given, then for all suciently small,
l + < (F
n
)
provided n is large enough. Now it was just shown there exists K a compact subset of
F
n
such that (F
n
) < (K) +. Then K F and
l + < (F
n
) < (K) +
and so whenever l < (F) , it follows there exists K a compact subset of F such that
l < (K)
and This proves the lemma.
The following is a useful result which will be used in what follows.
Lemma 7.4.4 Let X be a normed vector space and let S be any nonempty subset of
X. Dene
dist (x, S) inf {||x y|| : y S}
Then
|dist (x
1
, S) dist (x
2
, S)| ||x
1
x
2
|| .
Proof: Suppose dist (x
1
, S) dist (x
2
, S) . Then let y S such that
dist (x
2
, S) + > ||x
2
y||
Then
|dist (x
1
, S) dist (x
2
, S)| = dist (x
1
, S) dist (x
2
, S)
dist (x
1
, S) (||x
2
y|| )
||x
1
y|| ||x
2
y|| +
|||x
1
y|| ||x
2
y||| +
||x
1
x
2
|| +.
Since is arbitrary, this proves the lemma in case dist (x
1
, S) dist (x
2
, S) . The case
where dist (x
2
, S) dist (x
1
, S) is entirely similar. This proves the lemma.
The next lemma says that regularity comes free for nite measures dened on the
Borel sets. Actually, it only almost says this. The following theorem will say it. This
lemma deals with closed in place of compact.
Lemma 7.4.5 Let be a nite measure dened on B (Y ) where Y is a closed subset
of X, a nite dimensional normed vector space. Then for every F B (Y ) ,
(F) = sup {(K) : K F, K is closed }
(F) = inf {(V ) : V F, V is open}
Proof: For convenience, I will call a measure which satises the above two conditions
almost regular. It would be regular if closed were replaced with compact. First note
every open set is the countable union of closed sets and every closed set is the countable
intersection of open sets. Here is why. Let V be an open set and let
K
k

_
x V : dist
_
x, V
C
_
1/k
_
.
Then clearly the union of the K
k
equals V and each is closed because xdist (x, S) is
always a continuous function whenever S is any nonempty set. Next, for K closed let
V
k
{x Y : dist (x, K) < 1/k} .
Clearly the intersection of the V
k
equals K because if x / K, then since K is closed,
B(x, r) has empty intersection with K and so for k large enough that 1/k < r, V
k
excludes x. Thus the only points in the intersection of the V
k
are those in K and in
addition each point of K is in this intersection.
Therefore from what was just shown, letting V denote an open set and K a closed
set, it follows from Theorem 7.3.2 that
(V ) = sup {(K) : K V and K is closed}
(K) = inf {(V ) : V K and V is open} .
Also since V is open and K is closed,
(V ) = inf {(U) : U V and V is open}
(K) = sup {(L) : L K and L is closed}
In words, is almost regular on open and closed sets. Let
F {F B (Y ) such that is almost regular on F} .
Then F contains the open sets. I want to show F is a algebra and then it will follow
F = B (Y ).
First I will show F is closed with respect to complements. Let F F. Then since
is nite and F is inner regular, there exists K F such that
(F \ K) = (F) (K) < .
But K
C
\ F
C
= F \ K and so
_
K
C
\ F
C
_
=
_
K
C
_
_
F
C
_
<
showing that is outer regular on F
C
. I have just approximated the measure of F
C
with
the measure of K
C
, an open set containing F
C
. A similar argument works to show F
C
is inner regular. You start with V F such that (V \ F) < , note F
C
\ V
C
= V \ F,
and then conclude
_
F
C
\ V
C
_
< , thus approximating F
C
with the closed subset,
V
C
.
Next I will show F is closed with respect to taking countable unions. Let {F
k
} be
a sequence of sets in F. Then since F
k
F, there exist {K
k
} such that K
k
F
k
and
(F
k
\ K
k
) < /2
k+1
. First choose m large enough that
((
k=1
F
k
) \ (
m
k=1
F
k
)) <

2
.
Then
((
m
k=1
F
k
) \ (
m
k=1
K
k
)) (
m
k=1
(F
k
\ K
k
))
k=1
2
k+1
<

2
and so
((
k=1
F
k
) \ (
m
k=1
K
k
)) ((
k=1
F
k
) \ (
m
k=1
F
k
))
+((
m
k=1
F
k
) \ (
m
k=1
K
k
))
<

2
+

2
=
Since is outer regular on F
k
, there exists V
k
such that (V
k
\ F
k
) < /2
k
. Then
((
k=1
V
k
) \ (
k=1
F
k
)) (
k=1
(V
k
\ F
k
))
k=1
(V
k
\ F
k
)
<
k=1
2
k
=
and this completes the demonstration that F is a algebra. This proves the lemma.
The next theorem is the main result. It shows regularity is automatic if (K) <
for all compact K.
Theorem 7.4.6 Let be a nite measure dened on B (Y ) where Y is a closed
subset of X, a nite dimensional normed vector space. Then is regular. If is not
necessarily nite but is nite on compact sets, then is regular.
Proof: From Lemma 7.4.5 is outer regular. Now let F B (Y ). Then since is
nite, it follows from Lemma 7.4.5 there exists H F such that H is closed, H F,
and
(F) < (H) +.
Then let K
k
H B(0, k). Thus K
k
is a closed and bounded, hence compact set and
k=1
K
k
= H. Therefore by Theorem 7.3.2, for all k large enough,
(F)
< (K
k
) +
< sup {(K) : K F and K compact} +
(F) +
Since was arbitrary, it follows
sup {(K) : K F and K compact} = (F) .
This establishes is regular if is nite.
Now suppose it is only known that is nite on compact sets. Consider outer
regularity. There are at most nitely many r [0, R] such that (S (0,r)) > >
0. If this were not so, then (D(0,R)) = contrary to the assumption that is
nite on compact sets. Therefore, there are at most countably many r [0, R] such
that (S (0,r)) > 0. Here is why. Let S
k
denote those values of r [0, R] such that
(S (0,r)) > 1/k. Then the values of r such that (S (0,r)) > 0 equals
m=1
S
m
, a
countable union of nite sets which is at most countable.
It follows there are at most countably many r (0, ) such that (S (0,r)) > 0.
Therefore, there exists an increasing sequence {r
k
} such that lim
k
r
k
= and
(S (0,r
k
)) = 0. This is easy to see by noting that (n, n+1] contains uncountably many
points and so it contains at least one r such that (S (0,r)) = 0.
S (0,r) =
k=1
(B(0, r + 1/k) D(0,r 1/k))
a countable intersection of open sets which are decreasing as k . Since (B(0, r)) <
by assumption, it follows from Theorem 7.3.2 that for each r
k
there exists an open
set, U
k
S (0,r
k
) such that
(U
k
) < /2
k+1
.
Let (F) < . There is nothing to show if (F) = . Dene nite measures,
k
as follows.
1
(A) (B(0, 1) A) ,
2
(A) ((B(0, 2) \ D(0, 1)) A) ,
3
(A) ((B(0, 3) \ D(0, 2)) A)
etc. Thus
(A) =
k=1
k
(A)
and each
k
is a nite measure. By the rst part there exists an open set V
k
such that
V
k
F (B(0, k) \ D(0, k 1))
and
k
(V
k
) <
k
(F) +/2
k+1
Without loss of generality V
k
(B(0, k) \ D(0, k 1)) since you can take the intersec-
tion of V
k
with this open set. Thus
k
(V
k
) = ((B(0, k) \ D(0, k 1)) V
k
) = (V
k
)
and the V
k
are disjoint. Then let V =
k=1
V
k
and U =
k=1
U
k
. It follows V U is an
open set containing F and
(F) =
k=1
k
(F) >
k=1
k
(V
k
)

2
k+1
=
k=1
(V
k
)

2
= (V )

2
(V ) +(U)

2

2
(V U)
which shows is outer regular. Inner regularity can be obtained from Lemma 7.4.3.
Alternatively, you can use the above construction to get it right away. It is easier than
the outer regularity.
First assume (F) < . By the rst part, there exists a compact set,
K
k
F (B(0, k) \ D(0, k 1))
7.5. MEASURES AND OUTER MEASURES 165
such that
k
(K
k
) +/2
k+1
>
k
(F (B(0, k) \ D(0, k 1)))
=
k
(F) = (F (B(0, k) \ D(0, k 1))) .
Since K
k
is a subset of F (B(0, k) \ D(0, k 1)) it follows
k
(K
k
) = (K
k
). There-
fore,
(F) =
k=1
k
(F) <
k=1
k
(K
k
) +/2
k
<
_

k=1
k
(K
k
)
_
+/2 <
N
k=1
(K
k
) +
provided N is large enough. The K
k
are disjoint and so letting K =
N
k=1
K
k
, this says
K F and
(F) < (K) +.
Now consider the case where (F) = . If l < , it follows from Theorem 7.3.2
(F B(0, m)) > l
whenever m is large enough. Therefore, letting
m
(A) (A B(0, m)) , there exists
a compact set, K F B(0, m) such that
(K) =
m
(K) >
m
(F B(0, m)) = (F B(0, m)) > l
7.5 Measures And Outer Measures
7.5.1 Measures From Outer Measures
Earlier an outer measure on P (R) was constructed. This can be used to obtain a
measure dened on R. However, the procedure for doing so is a special case of a general
approach due to Caratheodory in about 1918.
Denition 7.5.1 Let be a nonempty set and let : P() [0, ] be an
outer measure. For E , E is measurable if for all S ,
(S) = (S \ E) +(S E). (7.7)
To help in remembering 7.7, think of a measurable set, E, as a process which divides
a given set into two pieces, the part in E and the part not in E as in 7.7. In the Bible,
there are several incidents recorded in which a process of division resulted in more stu
than was originally present.
2
Measurable sets are exactly those which are incapable of
such a miracle. You might think of the measurable sets as the nonmiraculous sets. The
idea is to show that they form a algebra on which the outer measure, is a measure.
First here is a denition and a lemma.
2
1 Kings 17, 2 Kings 4, Mathew 14, and Mathew 15 all contain such descriptions. The stu involved
was either oil, bread, our or sh. In mathematics such things have also been done with sets. In the
book by Bruckner Bruckner and Thompson there is an interesting discussion of the Banach Tarski
paradox which says it is possible to divide a ball in R
3
into ve disjoint pieces and assemble the pieces
to form two disjoint balls of the same size as the rst. The details can be found in: The Banach Tarski
Paradox by Wagon, Cambridge University press. 1985. It is known that all such examples must involve
the axiom of choice.
Denition 7.5.2 (S)(A) (S A) for all A . Thus S is the name
of a new outer measure, called restricted to S.
The next lemma indicates that the property of measurability is not lost by consid-
ering this restricted measure.
Lemma 7.5.3 If A is measurable, then A is S measurable.
Proof: Suppose A is measurable. It is desired to to show that for all T ,
(S)(T) = (S)(T A) + (S)(T \ A).
Thus it is desired to show
(S T) = (T A S) +(T S A
C
). (7.8)
But 7.8 holds because A is measurable. Apply Denition 7.5.1 to S T instead of S.
If A is S measurable, it does not follow that A is measurable. Indeed, if you

believe in the existence of non measurable sets, you could let A = S for such a non
measurable set and verify that S is S measurable. In fact there do exist nonmeasurable
sets but this is a topic for a more advanced course in analysis and will not be needed in
this book.
The next theorem is the main result on outer measures which shows that starting
with an outer measure you can obtain a measure.
Theorem 7.5.4 Let be a set and let be an outer measure on P (). The
collection of measurable sets S, forms a algebra and
If F
i
S, F
i
F
j
= , then (
i=1
F
i
) =
i=1
(F
i
). (7.9)
If F
n
F
n+1
, then if F =
n=1
F
n
and F
n
S, it follows that
(F) = lim
n
(F
n
). (7.10)
If F
n
F
n+1
, and if F =
n=1
F
n
for F
n
S then if (F
1
) < ,
(F) = lim
n
(F
n
). (7.11)
This measure space is also complete which means that if (F) = 0 for some F S then
if G F, it follows G S also.
Proof: First note that and are obviously in S. Now suppose A, B S. I will
show A\ B A B
C
is in S. To do so, consider the following picture.
S
A
C
B
C
S
A
C
B
S
B
S
B
C
A
B
S
Since is subadditive,
(S)
_
S A B
C
_
+(A B S) +
_
S B A
C
_
+
_
S A
C
B
C
_
.
Now using A, B S,
(S)
_
S A B
C
_
+(S A B) +
_
S B A
C
_
+
_
S A
C
B
C
_
= (S A) +
_
S A
C
_
= (S)
It follows equality holds in the above. Now observe, using the picture if you like, that
(A B S)
_
S B A
C
_
_
S A
C
B
C
_
= S \ (A\ B)
and therefore,
(S) =
_
S A B
C
_
+(A B S) +
_
S B A
C
_
+
_
S A
C
B
C
_
(S (A\ B)) +(S \ (A\ B)) .
Therefore, since S is arbitrary, this shows A\ B S.
Since S, this shows that A S if and only if A
C
S. Now if A, B S,
A B = (A
C
B
C
)
C
= (A
C
\ B)
C
S. By induction, if A
1
, , A
n
S, then so is
n
i=1
A
i
. If A, B S, with A B = ,
(A B) = ((A B) A) +((A B) \ A) = (A) +(B).
By induction, if A
i
A
j
= and A
i
S,
(
n
i=1
A
i
) =
n
i=1
(A
i
). (7.12)
Now let A =
i=1
A
i
where A
i
A
j
= for i = j.
i=1
(A
i
) (A) (
n
i=1
A
i
) =
n
i=1
(A
i
).
Since this holds for all n, you can take the limit as n and conclude,
i=1
(A
i
) = (A)
which establishes 7.9.
Consider part 7.10. Without loss of generality (F
k
) < for all k since otherwise
there is nothing to show. Suppose {F
k
} is an increasing sequence of sets of S. Then
letting F
0
, {F
k+1
\ F
k
}
k=0
is a sequence of disjoint sets of S since it was shown
above that the dierence of two sets of S is in S. Also note that from 7.12
(F
k+1
\ F
k
) +(F
k
) = (F
k+1
)
and so if (F
k
) < , then
(F
k+1
\ F
k
) = (F
k+1
) (F
k
) .
Therefore, letting
F
k=1
F
k
which also equals
k=1
(F
k+1
\ F
k
) ,
it follows from part 7.9 just shown that
(F) =
k=1
(F
k+1
\ F
k
) = lim
n
n
k=1
(F
k+1
\ F
k
)
= lim
n
n
k=1
(F
k+1
) (F
k
) = lim
n
(F
n+1
) .
In order to establish 7.11, let the F
n
be as given there. Then, since (F
1
\ F
n
)
increases to (F
1
\ F), 7.10 implies
lim
n
((F
1
) (F
n
)) = (F
1
\ F) .
Now (F
1
\ F) +(F) (F
1
) and so (F
1
\ F) (F
1
) (F). Hence
lim
n
((F
1
) (F
n
)) = (F
1
\ F) (F
1
) (F)
which implies
lim
n
(F
n
) (F) .
But since F F
n
,
(F) lim
n
(F
n
)
and this establishes 7.11. Note that it was assumed (F
1
) < because (F
1
) was
subtracted from both sides.
It remains to show S is closed under countable unions. Recall that if A S, then
A
C
S and S is closed under nite unions. Let A
i
S, A =
i=1
A
i
, B
n
=
n
i=1
A
i
.
Then
(S) = (S B
n
) +(S \ B
n
) (7.13)
= (S)(B
n
) + (S)(B
C
n
).
By Lemma 7.5.3 B
n
is (S) measurable and so is B
C
n
. I want to show (S) (S \
A) +(S A). If (S) = , there is nothing to prove. Assume (S) < . Then apply
Parts 7.11 and 7.10 to the outer measure, S in 7.13 and let n . Thus
B
n
A, B
C
n
A
C
and this yields (S) = (S)(A) + (S)(A
C
) = (S A) +(S \ A).
Therefore A S and this proves Parts 7.9, 7.10, and 7.11.
It only remains to verify the assertion about completeness. Letting G and F be as
described above, let S . I need to verify
(S) (S G) +(S \ G)
However,
(S G) +(S \ G) (S F) +(S \ F) +(F \ G)
= (S F) +(S \ F) = (S)
because by assumption, (F \ G) (F) = 0.This proves the theorem.
7.5.2 Completion Of Measure Spaces
Suppose (, F, ) is a measure space. Then it is always possible to enlarge the algebra
and dene a new measure on this larger algebra such that
_
, F,
_
is a complete
measure space. Recall this means that if N N
F and (N
) = 0, then N F. The
following theorem is the main result. The new measure space is called the completion
of the measure space.
Denition 7.5.5 A measure space, (, F, ) is called nite if there exists a
sequence {
n
} F such that
n
n
= and (
n
) < .
For example, if X is a nite dimensional normed vector space and is a measure
dened on B (X) which is nite on compact sets, then you could take
n
= B(0, n) .
Theorem 7.5.6 Let (, F, ) be a nite measure space. Then there exists a
unique measure space,
_
, F,
_
satisfying
1.
_
, F,
_
is a complete measure space.
2. = on F
3. F F
4. For every E F there exists G F such that G E and (G) = (E) .
In addition to this,
5. For every E F there exists F F such that F E and (F) = (E) .
Also for every E F there exist sets G, F F such that G E F and
(G\ F) = (G\ F) = 0 (7.14)
Proof: First consider the claim about uniqueness. Suppose (, F
1
,
1
) and (, F
2
,
2
)
both satisfy 1.) - 4.) and let E F
1
. Also let (
n
) < ,
n

n+1
,
and
n=1
n
= . Dene E
n
E
n
. Then there exists G
n
E
n
such that
(G
n
) =
1
(E
n
) , G
n
F and G
n

n
. I claim there exists F
n
F such that
G
n
E
n
F
n
and (G
n
\ F
n
) = 0. To see this, look at the following diagram.
G
n
\ E
n
F
n
E
n
H
n
H
n
In this diagram, there exists H
n
F containing G
n
\ E
n
, represented in the picture
as the set between the dotted lines, such that (H
n
) = (G
n
\ E
n
) . Then dene
F
n
H
C
n
G
n
. This set is in F, is contained in E
n
and as shown in the diagram,
(E
n
) (F
n
) (H
n
) (G
n
\ E
n
) = 0.
Therefore, since is a measure,
(G
n
\ F
n
) = (G
n
\ E
n
) +(E
n
\ F
n
)
= (G
n
) (E
n
) +(E
n
) (F
n
) = 0
Then letting G =
n
G
n
, F
n
F
n
, it follows G E F and
(G\ F) (
n
(G
n
\ F
n
))
n
(G
n
\ F
n
) = 0.
Thus
i
(G\ F) = 0 for i = 1, 2. Now E \ F G\ F and since (, F
2
,
2
) is complete,
it follows E \ F F
2
. Since F F
2
, it follows E = (E \ F) F F
2
. Thus F
1
F
2
.
Similarly F
2
F
1
.
Now it only remains to verify
1
=
2
. Thus let E F
1
= F
2
and let G and F be
as just described. Since
i
= on F,
(F)
1
(E)
=
1
(E \ F) +
1
(F)

1
(G\ F) +
1
(F)
=
1
(F) = (F)
Similarly
2
(E) = (F) . This proves uniqueness. The construction has also veried
7.14.
Next dene an outer measure, on P () as follows. For S ,
(S) inf {(E) : E F} .
Then it is clear is increasing. It only remains to verify is subadditive. Then let
S =
i=1
S
i
. If any (S
i
) = , there is nothing to prove so suppose (S
i
) < for
each i. Then there exist E
i
F such that E
i
S
i
and
(S
i
) +/2
i
> (E
i
) .
Then
(S) = (
i
S
i
)
(
i
E
i
)
i
(E
i
)
i
_
(S
i
) +/2
i
_
=
i
(S
i
) +.
Since is arbitrary, this veries is subadditive and is an outer measure as claimed.
Denote by F the algebra of measurable sets in the sense of Caratheodory. Then it
follows from the Caratheodory procedure, Theorem 7.5.4, that
_
, F,
_
is a complete
measure space. This veries 1.
Now let E F. Then from the denition of , it follows
(E) inf {(F) : F F and F E} (E) .
If F E and F F, then (F) (E) and so (E) is a lower bound for all such
(F) which shows that
(E) inf {(F) : F F and F E} (E) .
This veries 2.
Next consider 3. Let E F and let S be a set. I must show
(S) (S \ E) +(S E) .
If (S) = there is nothing to show. Therefore, suppose (S) < . Then from the
denition of there exists G S such that G F and (G) = (S) . Then from the
denition of ,
(S) (S \ E) +(S E)
(G\ E) +(G E)
= (G) = (S)
This veries 3.
Claim 4 comes by the denition of as used above. The other case is when (S) =
. However, in this case, you can let G = .
It only remains to verify 5. Let the
n
be as described above and let E F such
that E
n
. By 4 there exists H F such that H
n
, H
n
\ E, and
(H) = (
n
\ E) . (7.15)
Then let F
n
H
C
. It follows F E and
E \ F = E F
C
= E
_
H
C
n
_
= E H = H \ (
n
\ E)
Hence from 7.15
(E \ F) = (H \ (
n
\ E)) = 0.
It follows
(E) = (F) = (F) .
In the case where E F is arbitrary, not necessarily contained in some
n
, it follows
from what was just shown that there exists F
n
F such that F
n
E
n
and
(F
n
) = (E
n
) .
Letting F
n
F
n
(E \ F) (
n
(E
n
\ F
n
))
n
(E
n
\ F
n
) = 0.
Therefore, (E) = (F) and this proves 5. This proves the theorem.
Here is another observation about regularity which follows from the above theorem.
Theorem 7.5.7 Suppose is a regular measure dened on B (X) where X is a
nite dimensional normed vector space. Then denoting by
_
X, B (X),
_
the completion
of (X, B (X) , ) , it follows is also regular. Furthermore, if a algebra, F B (X) and
(X, F, ) is a complete measure space such that for every F F there exists G B (X)
such that (F) = (G) and G F, then F = B (X) and = .
Proof: Let F B (X) with (F) < . By Theorem 7.5.6 there exists G B (X)
such that
(G) = (G) = (F) .
Now by regularity of there exists an open set, V G F such that
(F) + = (G) + > (V ) = (V )
Therefore, is outer regular. If (F) = , there is nothing to show.
Now take F B (X). By Theorem 7.5.6 there exists H F with H B (X) and
(H) = (F). If l < (F) = (H) , it follows from regularity of there exists K a
compact subset of H such that
l < (K) = (K)
Thus is also inner regular. The last assertion follows from the uniqueness part of
Theorem 7.5.6 and This proves the theorem.
A repeat of the above argument yields the following corollary.
Corollary 7.5.8 The conclusion of the above theorem holds for X replaced with Y
where Y is a closed subset of X.
7.6 One Dimensional Lebesgue Stieltjes Measure
Now with these major results about measures, it is time to specialize to the outer
measure of Theorem 7.2.1. The next theorem gives Lebesgue Stieltjes measure on R.
Theorem 7.6.1 Let S denote the algebra of Theorem 7.5.4 applied to the outer
measure in Theorem 7.2.1 on which is a measure. Then every open interval is in
S. So are all open and closed sets. Furthermore, if E is any set in S
(E) = sup {(K) : K is a closed and bounded set, K E} (7.16)
(E) = inf {(V ) : V is an open set, V E} (7.17)
7.6. ONE DIMENSIONAL LEBESGUE STIELTJES MEASURE 173
Proof: The rst task is to show (a, b) S. I need to show that for every S R,
(S) (S (a, b)) +
_
S (a, b)
C
_
(7.18)
Suppose rst S is an open interval, (c, d) . If (c, d) has empty intersection with (a, b) or
is contained in (a, b) there is nothing to prove. The above expression reduces to nothing
more than (S) = (S). Suppose next that (c, d) (a, b) . In this case, the right side
of the above reduces to
((a, b)) +((c, a] [b, d))
F (b) F (a+) +F (a+) F (c+) +F (d) F (b)
= F (d) F (c+) = ((c, d))
The only other cases are c a < d b or a c < d b. Consider the rst of these
cases. Then the right side of 7.18 for S = (c, d) is
((a, d)) +((c, a]) = F (d) F (a+) +F (a+) F (c+)
= F (d) F (c+) = ((c, d))
The last case is entirely similar. Thus 7.18 holds whenever S is an open interval. Now
it is clear 7.18 also holds if (S) = . Suppose then that (S) < and let
S
k=1
(a
k
, b
k
)
such that
(S) + >
k=1
(F (b
k
) F (a
k
+)) =
k=1
((a
k
, b
k
)) .
Then since is an outer measure, and using what was just shown,
(S (a, b)) +
_
S (a, b)
C
_
(
k=1
(a
k
, b
k
) (a, b)) +
_
k=1
(a
k
, b
k
) (a, b)
C
_
k=1
((a
k
, b
k
) (a, b)) +
_
(a
k
, b
k
) (a, b)
C
_
k=1
((a
k
, b
k
)) (S) +.
Since is arbitrary, this shows 7.18 holds for any S and so any open interval is in S.
It follows any open set is in S. This follows from Theorem 5.3.10 which implies that
if U is open, it is the countable union of disjoint open intervals. Since each of these
open intervals is in S and S is a algebra, their union is also in S. It follows every
closed set is in S also. This is because S is a algebra and if a set is in S then so is its
complement. The closed sets are those which are complements of open sets.
Thus the algebra of measurable sets F includes B (R). Consider the completion of
the measure space, (R, B (R) , ) ,
_
R, B (R),
_
. By the uniqueness assertion in Theorem
7.5.6 and the fact that (R, F, ) is complete, this coincides with (R, F, ) because the
construction of implies is outer regular and for every F F, there exists G B (R)
containing F such that (F) = (G) . In fact, you can take G to equal a countable
intersection of open sets. By Theorem 7.4.6 is regular on every set of B (R) , this
because is nite on compact sets. Therefore, by Theorem 7.5.7 = is regular on F
which veries the last two claims. This proves the theorem.
7.7 Measurable Functions
The integral will be dened on measurable functions which is the next topic considered.
It is sometimes convenient to allow functions to take the value +. You should think
of +, usually referred to as as something out at the right end of the real line and
its only importance is the notion of sequences converging to it. x
n
exactly when
for all l R, there exists N such that if n N, then
x
n
> l.
This is what it means for a sequence to converge to . Dont think of as a number.
It is just a convenient symbol which allows the consideration of some limit operations
more simply. Similar considerations apply to but this value is not of very great
interest. In fact the set of most interest for the values of a function, f is the complex
numbers or more generally some normed vector space.
Recall the notation,
f
1
(A) {x : f (x) A} [f (x) A]
in whatever context the notation occurs.
Lemma 7.7.1 Let f : (, ] where F is a algebra of subsets of . Then
the following are equivalent.
f
1
((d, ]) F for all nite d,
f
1
((, d)) F for all nite d,
f
1
([d, ]) F for all nite d,
f
1
((, d]) F for all nite d,
f
1
((a, b)) F for all a < b, < a < b < .
Proof: First note that the rst and the third are equivalent. To see this, observe
f
1
([d, ]) =
n=1
f
1
((d 1/n, ]),
and so if the rst condition holds, then so does the third.
f
1
((d, ]) =
n=1
f
1
([d + 1/n, ]),
and so if the third condition holds, so does the rst.
Similarly, the second and fourth conditions are equivalent. Now
f
1
((, d]) = (f
1
((d, ]))
C
so the rst and fourth conditions are equivalent. Thus the rst four conditions are
equivalent and if any of them hold, then for < a < b < ,
f
1
((a, b)) = f
1
((, b)) f
1
((a, ]) F.
Finally, if the last condition holds,
f
1
([d, ]) =
_
k=1
f
1
((k +d, d))
_
C
F
and so the third condition holds. Therefore, all ve conditions are equivalent. This
proves the lemma.
This lemma allows for the following denition of a measurable function having values
in (, ].
7.7. MEASURABLE FUNCTIONS 175
Denition 7.7.2 Let (, F, ) be a measure space and let f : (, ].
Then f is said to be F measurable if any of the equivalent conditions of Lemma 7.7.1
hold.
Theorem 7.7.3 Let f
n
and f be functions mapping to (, ] where F is a
algebra of measurable sets of . Then if f
n
is measurable, and f() = lim
n
f
n
(),
it follows that f is also measurable. (Pointwise limits of measurable functions are mea-
surable.)
Proof: The idea is to show f
1
((a, b)) F. Let V
m

_
a +
1
m
, b
1
m
_
and V
m
=
_
a +
1
m
, b
1
m
. Then for all m, V

m
(a, b) and
(a, b) =
m=1
V
m
=
m=1
V
m
.
Note that V
m
= for all m large enough. Since f is the pointwise limit of f
n
,
f
1
(V
m
) { : f
k
() V
m
for all k large enough} f
1
(V
m
).
You should note that the expression in the middle is of the form
n=1
k=n
f
1
k
(V
m
).
Therefore,
f
1
((a, b)) =
m=1
f
1
(V
m
)
m=1
n=1
k=n
f
1
k
(V
m
)

m=1
f
1
(V
m
) = f
1
((a, b)).
It follows f
1
((a, b)) F because it equals the expression in the middle which is mea-
surable. This shows f is measurable.
Proposition 7.7.4 Let (, F, ) be a measure space and let f : (, ].
Then f is F measurable if and only if f
1
(U) F whenever U is an open set in R.
Proof: If f
1
(U) F whenever U is an open set in R then it follows from the last
condition of Lemma 7.7.1 that f is measurable. Next suppose f is measurable so this
last condition of Lemma 7.7.1 holds. Then by Theorem 5.3.10 if U is any open set in
R, it is the countable union of open intervals, U =
k=1
(a
k
, b
k
) . Hence
f
1
(U) =
k=1
f
1
((a
k
, b
k
)) F
because F is a algebra.
From this proposition, it follows one can generalize the denition of a measurable
function to those which have values in any normed vector space as follows.
Denition 7.7.5 Let (, F, ) be a measure space and let f : X where X
is a normed vector space. Then f is measurable means f
1
(U) F whenever U is an
open set in X.
Now here is an important theorem which shows that you can do lots of things to
measurable functions and still have a measurable function.
Theorem 7.7.6 Let (, F, ) be a measure space and let X, Y be normed vector
spaces and g : X Y continuous. Then if f : X is F measurable, it follows g f
is also F measurable.
Proof: From the denition, it suces to show (g f )
1
(U) F whenever U is an
open set in Y. However, since g is continuous, it follows g
1
(U) is open and so
(g f )
1
(U) = f
1
_
g
1
(U)
_
= f
1
(an open set) F.
This theorem implies for example that if f is a measurable X valued function, then
||f || is a measurable R valued function. It also implies that if f is an X valued function,
then if {v
1
, , v
n
} is a basis for X and
k
is the projection onto the k
th
component,
then
k
f is a measurable F valued function. Does it go the other way? That is, if
it is known that
k
f is measurable for each k, does it follow f is measurable? The
following technical lemma is interesting for its own sake.
Lemma 7.7.7 Let ||x|| max {|x
i
| , i = 1, 2, , n} for x F
n
. Then every set U
which is open in F
n
is the countable union of balls of the form B(x,r) where the open
ball is dened in terms of the above norm.
Proof: By Theorem 5.8.3 if you consider the two normed vector spaces (F
n
, ||)
and (F
n
, ||||) , the identity map is continuous in both directions. Therefore, if a set, U
is open with respect to || it follows it is open with respect to |||| and the other way
around. The other thing to notice is that there exists a countable dense subset of F.
The rationals will work if F = R and if F = C, then you use Q + iQ. Letting D be
a countable dense subset of F, D
n
is a countable dense subset of F
n
. It is countable
because it is a nite Cartesian product of countable sets and you can use Theorem 2.1.7
of Page 15 repeatedly. It is dense because if x F
n
, then by density of D, there exists
d
j
D such that
|d
j
x
j
| <
then d (d
1
, , d
n
) is such that ||d x|| < .
Now consider the set of open balls,
B {B(d, r) : d D
n
, r Q} .
This collection of open balls is countable by Theorem 2.1.7 of Page 15. I claim every
open set is the union of balls from B. Let U be an open set in F
n
and x U. Then there
exists > 0 such that B(x, ) U. There exists d D
n
B(x, /5) . Then pick rational
number /5 < r < 2/5. Consider the set of B, B(d, r) . Then x B(d, r) because
r > /5. However, it is also the case that B(d, r) B(x, ) because if y B(d, r) then
||y x|| ||y d|| +||d x||
<
2
5
+

5
< .
Corollary 7.7.8 Let (, F, ) be a measure space and let X be a normed vector space
with basis {v
1
, , v
n
} . Let
k
be the k
th
projection map onto the k
th
component. Thus
k
x x
k
where x =
n
i=1
x
i
v
i
.
Then each
k
f is a measurable F valued function if and only if f is a measurable X
valued function.
7.7. MEASURABLE FUNCTIONS 177
Proof: The if part has already been noted. Suppose that each
k
f is an F valued
measurable function. Let g : X F
n
be given by
g (x) (
1
x, ,
n
x) .
Thus g is linear, one to one, and onto. By Theorem 5.8.3 both g and g
1
are continuous.
Therefore, every open set in X is of the form g
1
(U) where U is an open set in F
n
. To
see this, start with V open set in X. Since g
1
is continuous, g (V ) is open in F
n
and
so V = g
1
(g (V )) . Therefore, it suces to show that for every U an open set in F
n
,
f
1
_
g
1
(U)
_
= (g f )
1
(U) F.
By Lemma 7.7.7 there are countably many open balls of the form B(x
j
, r
j
) such that
U is equal to the union of these balls. Thus
(g f )
1
(U) = (g f )
1
(
k=1
B(x
k
, r
k
))
=
k=1
(g f )
1
(B(x
k
, r
k
)) (7.19)
Now from the denition of the norm,
B(x
k
, r
k
) =
n
j=1
(x
kj
, x
kj
+)
and so
(g f )
1
(B(x
k
, r
k
)) =
n
j=1
(
j
f )
1
((x
kj
, x
kj
+)) F.
It follows 7.19 is the countable union of sets in F and so it is also in F. This proves the
corollary.
Note that if {f
i
}
n
i=1
are measurable functions dened on (, F, ) having values in
F then letting f (f
1
, , f
n
) , it follows f is a measurable F
n
valued function. Now let
: F
n
F be given by (x)
n
k=1
a
k
x
k
. Then is linear and so by Theorem 5.8.3
it follows is continuous. Hence by Theorem 7.7.6, (f ) is an F valued measurable
function. Thus linear combinations of measurable functions are measurable. By similar
reasoning, products of measurable functions are measurable. In general, it seems like
you can start with a collection of measurable functions and do almost anything you like
with them and the result, if it is a function will be measurable. This is in stark contrast
to the functions which are generalized Riemann integrable.
The following theorem considers the case of functions which have values in a normed
vector space.
n
} be a sequence of measurable functions mapping to
X where X is a normed vector space and (, F) is a measure space. Suppose also that
f () = lim
n
f
n
() for all . Then f is also a measurable function.
Proof: It is required to show f
1
(U) is measurable for all U open. Let
V
m

_
x U : dist
_
x, U
C
_
>
1
m
_
.
Thus
V
m

_
x U : dist
_
x, U
C
_
1
m
_
and V
m
V
m
V
m+1
and
m
V
m
= U. Then since V
m
is open, it follows that if
f () V
m
then for all suciently large k, it must be the case f
k
() V
m
also. That
is, f
1
k
(V
m
) for all suciently large k. Thus
f
1
(V
m
) =
n=1
k=n
f
1
k
(V
m
)
and so
f
1
(U) =
m=1
f
1
(V
m
)
=
m=1
n=1
k=n
f
1
k
(V
m
)

m=1
f
1
_
V
m
_
= f
1
(U)
which shows f
1
(U) is measurable. The step from the second to the last line follows
because if
n=1

k=n
f
1
k
(V
m
) , this says f
k
() V
m
for all k large enough.
Therefore, the point of X to which the sequence {f
k
()} converges must be in V
m
which equals V
m
V
m
, the limit points of V
m
Now here is a simple observation involving something called simple functions. It
uses the following notation.
Notation 7.7.10 For E a set let X
E
() be dened by
X
E
(x) =
_
1 if E
0 if / E
Theorem 7.7.11 Let f : X where X is some normed vector space. Sup-
pose
f () =
m
k=1
x
k
X
A
k
()
where each x
k
X and the A
k
are disjoint measurable sets. (Such functions are often
referred to as simple functions.) Then f is measurable.
Proof: Letting U be open, f
1
(U) = {A
k
: x
k
U} , a nite union of measurable
sets.
In the Lebesgue integral, the simple functions play a role similar to step functions
in the theory of the Riemann integral. Also there is a fundamental theorem about
measurable functions and simple functions which says essentially that the measurable
functions are those which are pointwise limits of simple functions.
Theorem 7.7.12 Let f 0 be measurable with respect to the measure space
(, F, ). Then there exists a sequence of nonnegative simple functions {s
n
} satisfying
0 s
n
() (7.20)
s
n
() s
n+1
()
f() = lim
n
s
n
() for all . (7.21)
If f is bounded the convergence is actually uniform.
Proof : First note that
f
1
([a, b)) = f
1
((, a))
C
f
1
((, b))
=
_
f
1
((, a)) f
1
((, b))
C
_
C
F.
Letting I { : f () = } , dene
t
n
() =
2
n
k=0
k
n
X
f
1
([k/n,(k+1)/n))
() +nX
I
().
7.8. EXERCISES 179
Then t
n
() f() for all and lim
n
t
n
() = f() for all . This is because
t
n
() = n for I and if f () [0,
2
n
+1
n
), then
0 f () t
n
()
1
n
. (7.22)
Thus whenever / I, the above inequality will hold for all n large enough. Let
s
1
= t
1
, s
2
= max (t
1
, t
2
) , s
3
= max (t
1
, t
2
, t
3
) , .
Then the sequence {s
n
} satises 7.20-7.21.
To verify the last claim, note that in this case the term nX
I
() is not present.
Therefore, for all n large enough that 2
n
n f () for all , 7.22 holds for all . Thus
the convergence is uniform. This proves the theorem.
7.8 Exercises
1. Let C be a set whose elements are algebras of subsets of . Show C is a
algebra also.
2. Let be any set. Show P () , the set of all subsets of is a algebra. Now let
L denote some subset of P () . Consider all algebras which contain L. Show
the intersection of all these algebras which contain L is a algebra containing
L and it is the smallest algebra containing L, denoted by (L). When is a
normed vector space, and L consists of the open sets (L) is called the algebra
of Borel sets.
3. Consider = [0, 1] and let S denote all subsets of [0, 1] , F such that either F
C
or F is countable. Note the empty set must be countable. Show S is a algebra.
(This is a sick algebra.) Now let : S [0, ] be dened by (F) = 1 if F
C
is countable and (F) = 0 if F is countable. Show is a measure on S.
4. Let = N, the positive integers and let a algebra be given by F = P (N), the
set of all subsets of N. What are the measurable functions having values in C?
Let (E) be the number of elements of E where E is a subset of N. Show is a
measure.
5. Let F be a algebra of subsets of and suppose F has innitely many elements.
Show that F is uncountable. Hint: You might try to show there exists a count-
able sequence of disjoint sets of F, {A
i
}. It might be easiest to verify this by
contradiction if it doesnt exist rather than a direct construction however, I have
seen this done several ways. Once this has been done, you can dene a map, ,
from P (N) into F which is one to one by (S) =
iS
A
i
. Then argue P (N) is
uncountable and so F is also uncountable.
6. A probability space is a measure space, (, F, P) where the measure, P has the
property that P () = 1. Such a measure is called a probability measure. Random
vectors are measurable functions, X, mapping a probability space, (, F, P) to
R
n
. Thus X() R
n
for each and P is a probability measure dened on
the sets of F, a algebra of subsets of . For E a Borel set in R
n
, dene
(E) P
_
X
1
(E)
_
probability that X E.
Show this is a well dened probability measure on the Borel sets of R
n
. Thus
(E) = P (X() E) . It is called the distribution. Explain why must be
regular.
7. Suppose (, S, ) is a measure space which may not be complete. Show that
another way to complete the measure space is to dene S to consist of all sets of
the form E where there exists F S such that (F \ E) (E \ F) N for some
N S which has measure zero and then let (E) =
1
(F)? Explain.
The Abstract Lebesgue
Integral
The general Lebesgue integral requires a measure space, (, F, ) and, to begin with,
a nonnegative measurable function. I will use Lemma 2.3.3 about interchanging two
supremums frequently. Also, I will use the observation that if {a
n
} is an increasing
sequence of points of [0, ] , then sup
n
a
n
= lim
n
a
n
which is obvious from the
denition of sup.
8.1 Denition For Nonnegative Measurable Functions
8.1.1 Riemann Integrals For Decreasing Functions
First of all, the notation
[g < f]
is short for
{ : g () < f ()}
with other variants of this notation being similar. Also, the convention, 0 = 0 will
be used to simplify the presentation whenever it is convenient to do so.
Denition 8.1.1 For f a nonnegative decreasing function dened on a nite
interval [a, b] , dene
b
a
f () d lim
M
b
a
M f () d = sup
M
b
a
M f () d
where a b means the minimum of a and b. Note that for f bounded,
sup
M
b
a
M f () d =
b
a
f () d
where the integral on the right is the usual Riemann integral because eventually M > f.
For f a nonnegative decreasing function dened on [0, ),

0
fd lim
R
R
0
fd = sup
R>1
R
0
fd = sup
R
sup
M>0
R
0
f Md
Since decreasing bounded functions are Riemann integrable, the above denition is
well dened. Now here are some obvious properties.
181
182 THE ABSTRACT LEBESGUE INTEGRAL
Lemma 8.1.2 Let f be a decreasing nonnegative function dened on an interval
[a, b] . Then if [a, b] =
m
k=1
I
k
where I
k
[a
k
, b
k
] and the intervals I
k
are non overlap-
ping, it follows
b
a
fd =
m
k=1
b
k
a
k
fd.
Proof: This follows from the computation,
b
a
fd lim
M
b
a
f Md
= lim
M
m
k=1
b
k
a
k
f Md =
m
k=1
b
k
a
k
fd
Note both sides could equal +.
8.1.2 The Lebesgue Integral For Nonnegative Functions
Here is the denition of the Lebesgue integral of a function which is measurable and
has values in [0, ].
Denition 8.1.3 Let (, F, ) be a measure space and suppose f : [0, ]
is measurable. Then dene
fd

0
([f > ]) d
which makes sense because ([f > ]) is nonnegative and decreasing.
Lemma 8.1.4 In the situation of the above denition,
fd = sup
h>0
i=1
([f > hi]) h
Proof:
fd

0
([f > ]) d = sup
M
sup
R>1
R
0
([f > ]) Md
= sup
M
sup
R>1
sup
h>0
M(R)
k=1
h(([f > kh]) M)
where M (R) is such that R h M (R) h R. The sum is just a lower sum for the
integral. Hence this equals
= sup
R>1
sup
h>0
sup
M
M(R)
k=1
h(([f > kh]) M)
= sup
R>1
sup
h>0
lim
M
M(R)
k=1
h(([f > kh]) M)
= sup
h>0
sup
R>1
M(R)
k=1
h([f > kh]) = sup
h>0
k=1
h([f > kh])
8.2. THE LEBESGUE INTEGRAL FOR NONNEGATIVE SIMPLE FUNCTIONS 183

8.2 The Lebesgue Integral For Nonnegative Simple
Functions
To begin with, here is a useful lemma.
Lemma 8.2.1 If f () = 0 for all > a, where f is a decreasing nonnegative func-
tion, then

0
f () d =
a
0
f () d.
Proof: From the denition,

0
f () d = lim
R
R
0
f () d = sup
R>1
R
0
f () d
= sup
R>1
sup
M
R
0
f () Md
= sup
M
sup
R>1
R
0
f () Md
= sup
M
sup
R>1
a
0
f () Md
= sup
M
a
0
f () Md
a
0
f () d.
Now the Lebesgue integral for a nonnegative function has been dened, what does
it do to a nonnegative simple function? Recall a nonnegative simple function is one
which has nitely many nonnegative values which it assumes on measurable sets. Thus
a simple function can be written in the form
s () =
n
i=1
c
i
X
Ei
()
where the c
i
are each nonnegative real numbers, the distinct values of s.
Lemma 8.2.2 Let s () =
p
i=1
a
i
X
Ei
() be a nonnegative simple function where
the E
i
are distinct but the a
i
might not be. Then
sd =
p
i=1
a
i
(E
i
) . (8.1)
Proof: Without loss of generality, assume 0 a
0
< a
1
a
2
a
p
and that
(E
i
) < , i > 0. Here is why. If (E
i
) = , then letting a (a
i1
, a
i
) , by Lemma
8.2.1, the left side would be
ap
0
([s > ]) d
ai
a0
([s > ]) d
sup
M
ai
0
([s > ]) Md
= sup
M
Ma
i
=
and so both sides are equal to . Thus it can be assumed for each i, (E
i
) < . Then
it follows from Lemma 8.2.1 and Lemma 8.1.2,

0
([s > ]) d =
ap
0
([s > ]) d =
p
k=1
a
k
a
k1
([s > ]) d
=
p
k=1
p
i=k
(a
k
a
k1
) (E
i
) =
p
i=1
(E
i
)
i
k=1
(a
k
a
k1
) =
p
i=1
a
i
(E
i
)
Lemma 8.2.3 If a, b 0 and if s and t are nonnegative simple functions, then
as +btd = a
sd +b
td.
Proof: Let
s() =
n
i=1
i
X
Ai
(), t() =
m
i=1
j
X
Bj
()
where
i
are the distinct values of s and the
j
are the distinct values of t. Clearly as+bt
is a nonnegative simple function because it has nitely many values on measurable sets
In fact,
(as +bt)() =
m
j=1
n
i=1
(a
i
+b
j
)X
AiBj
()
where the sets A
i
B
j
are disjoint and measurable. By Lemma 8.2.2,
as +btd =
m
j=1
n
i=1
(a
i
+b
j
)(A
i
B
j
)
=
n
i=1
a
m
j=1
i
(A
i
B
j
) +b
m
j=1
n
i=1
j
(A
i
B
j
)
= a
n
i=1
i
(A
i
) +b
m
j=1
j
(B
j
)
= a
sd +b
td.
8.3 The Monotone Convergence Theorem

The following is called the monotone convergence theorem. This theorem and related
convergence theorems are the reason for using the Lebesgue integral.
Theorem 8.3.1 (Monotone Convergence theorem) Let f have values in [0, ]
and suppose {f
n
} is a sequence of nonnegative measurable functions having values in
[0, ] and satisfying
lim
n
f
n
() = f() for each .
f
n
() f
n+1
()
Then f is measurable and
fd = lim
n
f
n
d.
8.4. OTHER DEFINITIONS 185
Proof: By Lemma 8.1.4
lim
n
f
n
d = sup
n
f
n
d
= sup
n
sup
h>0
k=1
([f
n
> kh]) h = sup
h>0
sup
N
sup
n
N
k=1
([f
n
> kh]) h
= sup
h>0
sup
N
N
k=1
([f > kh]) h = sup
h>0
k=1
([f > kh]) h =
fd
To illustrate what goes wrong without the Lebesgue integral, consider the following
example.
Example 8.3.2 Let {r
n
} denote the rational numbers in [0, 1] and let
f
n
(t)
_
1 if t / {r
1
, , r
n
}
0 otherwise
Then f
n
(t) f (t) where f is the function which is one on the rationals and zero on
the irrationals. Each f
n
is Riemann integrable (why?) but f is not Riemann integrable.
Therefore, you cant write
fdx = lim
n
f
n
dx.
A meta-mathematical observation related to this type of example is this. If you can
choose your functions, you dont need the Lebesgue integral. The Riemann Darboux
integral is just ne. It is when you cant choose your functions and they come to you as
pointwise limits that you really need the superior Lebesgue integral or at least something
more general than the Riemann integral. The Riemann integral is entirely adequate for
evaluating the seemingly endless lists of boring problems found in calculus books.
8.4 Other Denitions
To review and summarize the above, if f 0 is measurable,
fd

0
([f > ]) d (8.2)
another way to get the same thing for
fd is to take an increasing sequence of non-

negative simple functions, {s
n
} with s
n
() f () and then by monotone convergence
theorem,
fd = lim
n
s
n
where if s
n
() =
m
j=1
c
i
X
Ei
() ,
s
n
d =
m
i=1
c
i
(E
i
) .
Similarly this also shows that for such nonnegative measurable function,
fd = sup
_
s : 0 s f, s simple
_
Here is an equivalent denition of the integral of a nonnegative measurable function.
The fact it is well dened has been discussed above.
Denition 8.4.1 For s a nonnegative simple function,
s () =
n
k=1
c
k
X
E
k
() ,
s =
n
k=1
c
k
(E
k
) .
For f a nonnegative measurable function,
fd = sup
_
s : 0 s f, s simple
_
.
8.5 Fatous Lemma
The next theorem, known as Fatous lemma is another important theorem which justies
the use of the Lebesgue integral.
Theorem 8.5.1 (Fatous lemma) Let f
n
be a nonnegative measurable function
with values in [0, ]. Let g() = liminf
n
f
n
(). Then g is measurable and
gd lim inf
n
f
n
d.
In other words,
_
lim inf
n
f
n
_
d lim inf
n
f
n
d
Proof: Let g
n
() = inf{f
k
() : k n}. Then
g
1
n
([a, ]) =
k=n
f
1
k
([a, ])
=
_
k=n
f
1
k
([a, ])
C
_
C
F.
Thus g
n
is measurable by Lemma 7.7.1. Also g() = lim
n
g
n
() so g is measurable
because it is the pointwise limit of measurable functions. Now the functions g
n
form an
increasing sequence of nonnegative measurable functions so the monotone convergence
theorem applies. This yields
gd = lim
n
g
n
d lim inf
n
f
n
d.
The last inequality holding because
g
n
d
f
n
d.
(Note that it is not known whether lim
n
f
n
d exists.) This proves the theorem.
8.6 The Righteous Algebraic Desires Of The Lebesgue

Integral
The monotone convergence theorem shows the integral wants to be linear. This is the
essential content of the next theorem.
Theorem 8.6.1 Let f, g be nonnegative measurable functions and let a, b be non-
negative numbers. Then af +bg is measurable and
(af +bg) d = a
fd +b
gd. (8.3)
8.7. THE LEBESGUE INTEGRAL, L
1
187
Proof: By Theorem 7.7.12 on Page 178 there exist increasing sequences of nonneg-
ative simple functions, s
n
f and t
n
g. Then af + bg, being the pointwise limit
of the simple functions as
n
+ bt
n
, is measurable. Now by the monotone convergence
theorem and Lemma 8.2.3,
(af +bg) d = lim

n
as
n
+bt
n
d
= lim
n
_
a
s
n
d +b
t
n
d
_
= a
fd +b
gd.
As long as you are allowing functions to take the value +, you cannot consider
something like f +(g) and so you cant very well expect a satisfactory statement about
the integral being linear until you restrict yourself to functions which have values in a
vector space. This is discussed next.
8.7 The Lebesgue Integral, L
1
The functions considered here have values in C, a vector space.
Denition 8.7.1 Let (, S, ) be a measure space and suppose f : C. Then
f is said to be measurable if both Re f and Imf are measurable real valued functions.
Denition 8.7.2 A complex simple function will be a function which is of the
form
s () =
n
k=1
c
k
X
E
k
()
where c
k
C and (E
k
) < . For s a complex simple function as above, dene
I (s)
n
k=1
c
k
(E
k
) .
Lemma 8.7.3 The denition, 8.7.2 is well dened. Furthermore, I is linear on the
vector space of complex simple functions. Also the triangle inequality holds,
|I (s)| I (|s|) .
Proof: Suppose
n
k=1
c
k
X
E
k
() = 0. Does it follow that
k
c
k
(E
k
) = 0? The
supposition implies
n
k=1
Re c
k
X
E
k
() = 0,
n
k=1
Imc
k
X
E
k
() = 0. (8.4)
Choose large and positive so that +Re c
k
0. Then adding
k
X
E
k
to both sides
of the rst equation above,
n
k=1
( + Re c
k
) X
E
k
() =
n
k=1
X
E
k
and by Lemma 8.2.3 on Page 184, it follows upon taking
of both sides that

n
k=1
( + Re c
k
) (E
k
) =
n
k=1
(E
k
)
which implies
n
k=1
Re c
k
(E
k
) = 0. Similarly,
n
k=1
Imc
k
(E
k
) = 0
and so
n
k=1
c
k
(E
k
) = 0. Thus if
j
c
j
X
Ej
=
k
d
k
X
F
k
then
j
c
j
X
Ej
+
k
(d
k
) X
F
k
= 0 and so the result just established veries
j
c
j
(E
j
)
k
d
k
(F
k
) = 0
which proves I is well dened.
That I is linear is now obvious. It only remains to verify the triangle inequality.
Let s be a simple function,
s =
j
c
j
X
Ej
Then pick C such that I (s) = |I (s)| and || = 1. Then from the triangle inequality
for sums of complex numbers,
|I (s)| = I (s) = I (s) =
j
c
j
(E
j
)
=
j
c
j
(E
j
)
j
|c
j
| (E
j
) = I (|s|) .
Note that for any simple function s =
n
k=1
c
k
X
E
k
where c
k
> 0, (E
k
) < , it
follows from Lemma 8.2.2 that
sd = I (s) since they both equal
n
k=1
c
k
X
E
k
.
With this lemma, the following is the denition of L
1
() .
Denition 8.7.4 f L
1
() means there exists a sequence of complex simple
functions, {s
n
} such that
s
n
() f () for all
lim
m,n
I (|s
n
s
m
|) = lim
n,m
|s
n
s
m
| d = 0
(8.5)
Then
I (f) lim
n
I (s
n
) . (8.6)
Lemma 8.7.5 Denition 8.7.4 is well dened. Also L
1
() is a vector space.
Proof: There are several things which need to be veried. First suppose 8.5. Then
by Lemma 8.7.3
|I (s
n
) I (s
m
)| = |I (s
n
s
m
)| I (|s
n
s
m
|)
8.7. THE LEBESGUE INTEGRAL, L
1
189
and for m, n large enough, this last is given to be small so {I (s
n
)} is a Cauchy sequence
in C and so it converges. This veries the limit in 8.6 at least exists. It remains to
consider another sequence {t
n
} having the same properties as {s
n
} and verifying I (f)
determined by this other sequence is the same. By Lemma 8.7.3 and Fatous lemma,
Theorem 8.5.1 on Page 186,
|I (s
n
) I (t
n
)| I (|s
n
t
n
|) =
|s
n
t
n
| d
|s
n
f| +|f t
n
| d
lim inf
k
|s
n
s
k
| d + lim inf
k
|t
n
t
k
| d <
whenever n is large enough. Since is arbitrary, this shows the limit from using the t
n
is the same as the limit from using s
n
.
Why is L
1
() a vector space? Let f, g be in L
1
() and let a, b C. Then let {s
n
}
and {t
n
} be sequences of complex simple functions associated with f and g respectively
as described in Denition 8.7.4. Consider {as
n
+bt
n
} , another sequence of complex
simple functions. Then as
n
() + bt
n
() af () + bg () for each . Also, from
Theorem 8.6.1,
|as
n
+bt
n
(as
m
+bt
m
)| d |a|
|s
n
s
m
| d +|b|
|t
n
t
m
| d
and the sum of the two terms on the right converge to zero as m, n . Thus af +bg
L
1
().
Now here is another characterization for a function to be in L
1
().
Corollary 8.7.6 Let (, S, ) be a measure space and let f : C. Then f
L
1
() if and only if f is measurable and
|f| d < .
Proof: First suppose f L
1
. Then there exists a sequence {s
n
} of the sort de-
scribed above attached to f. It follows that f is measurable because it is the limit of
these measurable functions. Also for the same reasoning |f| = lim
n
|s
n
| so |f| is
measurable as a real valued function. Now from I being linear,
|s
n
| d
|s
m
| d
=
|I (|s
n
|) I (|s
m
|)| = |I (|s
n
| |s
m
|)| I (||s
n
| |s
m
||)
=
||s
n
| |s
m
|| d
|s
n
s
m
| d
which is small whenever n, m are large. As to
|f| d being nite, this follows from

Fatous lemma.
|f| d lim inf
n
|s
n
| d <
Next suppose f is measurable and absolutely integrable. First suppose f 0. Then
by the approximation theorem involving simple functions, Theorem 7.7.12, there exists
a sequence of nonnegative simple functions s
n
which increases pointwise to f. Each of
these must be nonzero only on a set of nite measure because
fd < . Note that
2f (f s
n
) d +
f s
n
d =
2f
and so

2f (f s
n
) d =
2fd
(f s
n
) d
Then by the monotone convergence theorem,
2f (f s
n
) d =
2fd
(f s
n
) d
2f
which shows that
|f s
n
| d 0. It follows that
I (|s
n
s
m
|) =
|s
n
s
m
| d
|s
n
f| d +
|f s
m
| d
both of which converge to 0. Thus there exists the right sort of sequence attached to
f and this shows f L
1
() as claimed. Now in the case where f has complex values,
just write
f = Re f
+
Re f
+i
_
Imf
+
Imf
_
for h
+
1
2
(|h| +h) and h

1
2
(|h| h). Each of the above is nonnegative, measurable
with nite integral and so from the above argument, each is in L
1
() from what was
just shown. Therefore, by Lemma 8.7.5 so is f.
Consider the following picture. I have just given a denition of an integral for
functions having values in C. However, [0, ) C.
complex valued functions values in [0, ]
What if f has values in [0, )? Earlier
fd was dened for such functions and now

I (f) has been dened. Are they the same? If so, I can be regarded as an extension of
d to a larger class of functions.

Lemma 8.7.7 Suppose f has values in [0, ) and f L
1
() . Then f is measurable
and
I (f) =
fd.
Proof: Since f is the pointwise limit of a sequence of complex simple functions,
{s
n
} having the properties described in Denition 8.7.4, it follows
f () = lim
n
Re s
n
()
Also it is always the case that if a, b are real numbers,
a
+
b
+
|a b|
8.8. APPROXIMATION WITH SIMPLE FUNCTIONS 191
and so
(Re s
n
)
+
(Re s
m
)
+
|Re s
n
Re s
m
| d
|s
n
s
m
| d
where x
+
1
2
(|x| +x) , the positive part of the real number x.
1
Thus there is no loss
of generality in assuming {s
n
} is a sequence of complex simple functions having values
in [0, ). By Corollary 8.7.6,
fd < .
Therefore, there exists a nonnegative simple function t f such that
fd
td +.
Then since, for such nonnegative complex simple functions, I (s) =
sd,
I (f)
fd
I (f)
td
+ |I (f) I (s
n
)|
+
s
n
d
td
+ = |I (f) I (s
n
)| +|I (s
n
) I (t)| +
+
|s
n
t| d + +
|s
n
f| d +
|f t| d +
3 + lim inf
k
|s
n
s
k
| d < 4
whenever n is large enough. Since is arbitrary, this shows I (f) =
fd as claimed.
As explained above, I can be regarded as an extension of
d, so from now on, the

usual symbol,
d will be used. It is now easy to verify
d is linear on the vector

space L
1
() .
8.8 Approximation With Simple Functions
The next theorem says the integral as dened above is linear and also gives a way to
compute the integral in terms of real and imaginary parts. In addition, functions in L
1
can be approximated with simple functions.
Theorem 8.8.1

d is linear on L
1
() and L
1
() is a complex vector space.
If f L
1
() , then Re f, Imf, and |f| are all in L
1
() . Furthermore, for f L
1
() ,
fd =
(Re f)
+
d
(Re f)
d +i
_
(Imf)
+
d
(Imf)
d
_
,
and the triangle inequality holds,
fd
|f| d
Also for every f L
1
() , for every > 0 there exists a simple function s such that
|f s| d < .
1
The negative part of the real number x is dened to be x

1
2
(|x| x) . Thus |x| = x
+
+x
and
x = x
+
x
. .
Proof: Why is the integral linear? Let {s
n
} and {t
n
} be sequences of simple
functions attached to f and g respectively according to the denition.
(af +bg) d lim

n
(as
n
+bt
n
) d
= lim
n
_
a
s
n
d +b
t
n
d
_
= a lim
n
s
n
d +b lim
n
t
n
d
= a
fd +b
gd.
The fact that
is linear makes the triangle inequality easy to verify. Let f L

1
()
and let C such that || = 1 and
fd =
fd
. Then
fd
fd =
Re (f) d =
Re (f)
+
Re (f)
Re (f)
+
d
|Re (f)| d
|f| d
Now the last assertion follows from the denition. There exists a sequence of simple
functions {s
n
} converging pointwise to f such that for all m, n large enough,
2
>
|s
n
s
m
| d
Fix such an m and let n . By Fatous lemma
>

2
lim inf
n
|s
n
s
m
| d
|f s
m
| d.
Let s = s
m
.
Recall that it has been shown that in computing the integrals on the right in ??,
then for g one of those integrands,
gd =

0
([g > t]) dt = sup
h>0
k=1
([g > kh]) h
= sup
_
sd : s g, s a nonnegative simple function
_
One of the major theorems in this theory is the dominated convergence theorem.
Before presenting it, here is a technical lemma about limsup and liminf .
Lemma 8.8.2 Let {a
n
} be a sequence in [, ] . Then lim
n
a
n
exists if and
only if
lim inf
n
a
n
= lim sup
n
a
n
and in this case, the limit equals the common value of these two numbers.
Proof: Suppose rst lim
n
a
n
= a R. Then, letting > 0 be given, a
n

(a , a +) for all n large enough, say n N. Therefore, both inf {a
k
: k n} and
sup {a
k
: k n} are contained in [a , a +] whenever n N. It follows limsup
n
a
n
and liminf
n
a
n
are both in [a , a +] , showing
lim inf
n
a
n
lim sup
n
a
n
< 2.
8.9. THE DOMINATED CONVERGENCE THEOREM 193
Since is arbitrary, the two must be equal and they both must equal a. Next suppose
lim
n
a
n
= . Then if l R, there exists N such that for n N,
l a
n
and therefore, for such n,
l inf {a
k
: k n} sup {a
k
: k n}
and this shows, since l is arbitrary that
lim inf
n
a
n
= lim sup
n
a
n
= .
The case for is similar.
Conversely, suppose liminf
n
a
n
= limsup
n
a
n
= a. Suppose rst that a R.
Then, letting > 0 be given, there exists N such that if n N,
sup {a
k
: k n} inf {a
k
: k n} <
therefore, if k, m > N, and a
k
> a
m
,
|a
k
a
m
| = a
k
a
m
sup {a
k
: k n} inf {a
k
: k n} <
showing that {a
n
} is a Cauchy sequence. Therefore, it converges to a R, and as in the
rst part, the liminf and limsup both equal a. If liminf
n
a
n
= limsup
n
a
n
= ,
then given l R, there exists N such that for n N,
inf
n>N
a
n
> l.
Therefore, lim
n
a
n
= . The case for is similar. This proves the lemma.
8.9 The Dominated Convergence Theorem
The dominated convergence theorem is one of the most important theorems in the
theory of the integral. It is one of those big theorems which justies the study of the
Lebesgue integral.
Theorem 8.9.1 (Dominated Convergence theorem) Let f
n
L
1
() and suppose
f() = lim
n
f
n
(),
and there exists a measurable function g, with values in [0, ],
2
such that
|f
n
()| g() and
g()d < .
Then f L
1
() and
0 = lim
n
|f
n
f| d = lim
n
fd
f
n
d
2
Note that, since g is allowed to have the value , it is not known that g L
1
() .
Proof: f is measurable by Theorem 7.7.3. Since |f| g, it follows that
f L
1
() and |f f
n
| 2g.
By Fatous lemma (Theorem 8.5.1),
2gd lim inf

n
2g |f f
n
|d
=
2gd lim sup

n
|f f
n
|d.
Subtracting
2gd,
0 lim sup
n
|f f
n
|d.
Hence
0 lim sup
n
_
|f f
n
|d
_
lim inf
n
_
|f f
n
|d
_
fd
f
n
d
0.
This proves the theorem by Lemma 8.8.2 because the limsup and liminf are equal.
Corollary 8.9.2 Suppose f
n
L
1
() and f () = lim
n
f
n
() . Suppose also
there exist measurable functions, g
n
, g with values in [0, ] such that lim
n
g
n
d =
gd, g
n
() g () a.e. and both
g
n
d and
gd are nite. Also suppose

|f
n
()| g
n
() . Then
lim
n
|f f
n
| d = 0.
Proof: It is just like the above. This time g +g
n
|f f
n
| 0 and so by Fatous
lemma,
2gd lim sup

n
|f f
n
| d =
lim inf
n
(g
n
+g) d lim sup
n
|f f
n
| d
= lim inf
n
((g
n
+g) |f f
n
|) d
2gd
and so limsup
n
|f f
n
| d 0. Thus
0 lim sup
n
_
|f f
n
|d
_
lim inf
n
_
|f f
n
|d
_
fd
f
n
d
0.
Denition 8.9.3 Let E be a measurable subset of .
E
fd
fX
E
d.
If L
1
(E) is written, the algebra is dened as
{E A : A F}
and the measure is restricted to this smaller algebra. Clearly, if f L
1
(), then
fX
E
L
1
(E)
and if f L
1
(E), then letting

f be the 0 extension of f o of E, it follows

f L
1
().
8.10. APPROXIMATION WITH CC (Y ) 195
8.10 Approximation With C
c
(Y )
Let (Y, F, ) be a measure space where Y is a closed subset of X a nite dimensional
normed vector space and F B (Y ) , the Borel sets in Y . Suppose also that (K) <
whenever K is a compact set in Y . By Theorem 7.4.6 it follows is regular. This
regularilty of implies an important approximation result valid for any f L
1
(Y ) .
It turns out that in this situation, for all > 0, there exists g a continuous function
dened on Y with g equal to 0 outside some compact set and
|f g| d < .
Denition 8.10.1 Let f : X Y where X is a normed vector space. Then
the support of f , denoted by spt (f ) is the closure of the set where f is not equal to zero.
Thus
spt (f ) {x : f (x) = 0}
Also, if U is an open set, f C
c
(U) means f is continuous on U and spt (f ) U.
Similarly f C
m
c
(U) if f has m continuous derivatives and spt (f ) U and f C
c
(U)
if spt (f ) U and f has continuous derivatives of every order on U.
Lemma 8.10.2 Let Y be a closed subset of X a nite dimensional normed vector
space. Let K V where K is compact in Y and V is open in Y . Then there exists a
continuous function f : Y [0, 1] such that spt (f) V , f (x) = 1 for all x K. If
(Y, F, ) is a measure space with F B (Y ) and (K) < , for every compact K, then
if (E) < where E F, there exists a sequence of functions in C
c
(Y ) {f
k
} such
that
lim
k
Y
|f
k
(x) X
E
(x)| d = 0.
Proof: For each x K, there exists r
x
such that
D(x, r
x
) {y Y : ||x y|| r
x
} V.
Since K is compact, there are nitely many balls, {B(x
k
, r
x
k
)}
m
k=1
which cover K. Let
W =
m
k=1
B(x
k
, r
x
k
) . Since there are only nitely many of these,
W =
m
k=1
D(x, r
x
)
and W is a compact subset of V because it is closed and bounded, being the nite union
of closed and bounded sets. Now dene
f (x)
dist
_
x, W
C
_
dist (x, W
C
) + dist (x, K)
The denominator is never equal to 0 because if dist (x, K) = 0 then since K is closed,
x K and so since K W, an open set, dist
_
x, W
C
_
> 0. Therefore, f is continuous.
When x K, f (x) = 1. If x / W, then f (x) = 0 and so spt (f) W V. In the above
situation the following notation is often used.
K f V. (8.7)
It remains to prove the last assertion. By Theorem 7.4.6, is regular and so there
exist compact sets {K
k
} and open sets {V
k
} such that V
k
V
k+1
, K
k
K
k+1
for all
k, and
K
k
E V
k
, (V
k
\ K
k
) < 2
k
.
From the rst part of the lemma, there exists a sequence {f
k
} such that
K
k
f
k
V
k
.
Then f
k
(x) converges to X
E
(x) a.e. because if convergence fails to take place, then x
must be in innitely many of the sets V
k
\ K
k
. Thus x is in
m=1
k=m
V
k
\ K
k
and for each p
(
m=1
k=m
V
k
\ K
k
)
_
k=p
V
k
\ K
k
_
k=p
(V
k
\ K
k
)
<
k=p
1
2
k
2
(p1)
Now the functions are all bounded above by 1 and below by 0 and are equal to zero o
V
1
, a set of nite measure so by the dominated convergence theorem,
lim
k
|X
E
(x) f
k
(x)| d = 0,
the dominating function being X
E
(x) +X
V1
(x) . This proves the lemma.
With this lemma, here is an important major theorem.
Theorem 8.10.3 Let Y be a closed subset of X a nite dimensional normed
vector space. Let (Y, F, ) be a measure space with F B (Y ) and (K) < , for
every compact K in Y . Let f L
1
(Y ) and let > 0 be given. Then there exists
g C
c
(Y ) such that
Y
|f (x) g (x)| d < .
Proof: By considering separately the positive and negative parts of the real and
imaginary parts of f it suces to consider only the case where f 0. Then by Theorem
7.7.12 and the monotone convergence theorem, there exists a simple function,
s (x)
p
m=1
c
m
X
Em
(x) , s (x) f (x)
such that
|f (x) s (x)| d < /2.

By Lemma 8.10.2, there exists {h
mk
}
k=1
be functions in C
c
(Y ) such that
lim
k
Y
|X
Em
f
mk
| d = 0.
Let
g
k
(x)
p
m=1
c
m
h
mk
.
8.11. THE ONE DIMENSIONAL LEBESGUE INTEGRAL 197
Thus for k large enough,
|s (x) g
k
(x)| d =
m=1
c
m
(X
Em
h
mk
)
m=1
c
m
|X
Em
h
mk
| d < /2
Thus for k this large,
|f (x) g
k
(x)| d
|f (x) s (x)| d +
|s (x) g
k
(x)| d
< /2 +/2 = .
People think of this theorem as saying that f is approximated by g
k
in L
1
(Y ). It is
customary to consider functions in L
1
(Y ) as vectors and the norm of such a vector is
given by
||f||
1

|f (x)| d.
You should verify this mostly satises the axioms of a norm. The problem comes in
asserting f = 0 if ||f|| = 0 which strictly speaking is false. However, the other axioms
of a norm do hold.
8.11 The One Dimensional Lebesgue Integral
Let F be an increasing function dened on R. Let be the Lebesgue Stieltjes measure
dened in Theorems 7.6.1 and 7.2.1. The conclusions of these theorems are reviewed
here.
Theorem 8.11.1 Let F be an increasing function dened on R, an integrator
function. There exists a function : P (R) [0, ] which satises the following
properties.
1. If A B, then 0 (A) (B) , () = 0.
2. (
k=1
A
i
)
i=1
(A
i
)
3. ([a, b]) = F (b+) F (a) ,
4. ((a, b)) = F (b) F (a+)
5. ((a, b]) = F (b+) F (a+)
6. ([a, b)) = F (b) F (a) where
F (b+) lim
tb+
F (t) , F (b) lim
ta
F (t) .
There also exists a algebra S of measurable sets on which is a measure which
contains the open sets and also satises the regularity conditions,
(E) = sup {(K) : K is a closed and bounded set, K E} (8.8)
(E) = inf {(V ) : V is an open set, V E} (8.9)
whenever E is a set in S.
The Lebesgue integral taken with respect to this measure, is called the Lebesgue
Stieltjes integral. Note that any real valued continuous function is measurable with
respect to S. This is because if f is continuous, inverse images of open sets are open
and open sets are in S. Thus f is measurable because f
1
((a, b)) S. Similarly if
f has complex values this argument applied to its real and imaginary parts yields the
conclusion that f is measurable.
For f a continuous function, how does the Lebesgue Stieltjes integral compare with
the Darboux Stieltjes integral? To answer this question, here is a technical lemma.
Lemma 8.11.2 Let D be a countable subset of R and suppose a, b / D. Also suppose
f is a continuous function dened on [a, b] . Then there exists a sequence of functions
{s
n
} of the form
s
n
(x)
mn
k=1
f
_
z
n
k1
_
X
[z
n
k1
,z
n
k
)
(x)
such that each z
n
k
/ D and
sup {|s
n
(x) f (x)| : x [a, b]} < 1/n.
Proof: First note that D contains no intervals. To see this let D = {d
k
}
k=1
. If D
has an interval of length 2, let I
k
be an interval centered at d
k
which has length /2
k
.
Therefore, the sum of the lengths of these intervals is no more than
k=1
2
k
= .
Thus D cannot contain an interval of length 2. Since is arbitrary, D cannot contain
any interval.
Since f is continuous, it follows from Theorem 5.4.2 on Page 94 that f is uniformly
continuous. Therefore, there exists > 0 such that if |x y| 3, then
|f (x) f (y)| < 1/n
Now let {x
0
, , x
mn
} be a partition of [a, b] such that |x
i
x
i1
| < for each i. For
k = 1, 2, , m
n
1, let z
n
k
/ D and |z
n
k
x
k
| < . Then
z
n
k
z
n
k1
|z
n
k
x
k
| +|x
k
x
k1
| +
x
k1
z
n
k1
< 3.
It follows that for each x [a, b]
mn
k=1
f
_
z
n
k1
_
X
[z
n
k1
,z
n
k
)
(x) f (x)
< 1/n.
Proposition 8.11.3 Let f be a continuous function dened on R. Also let F be an
increasing function dened on R. Then whenever c, d are not points of discontinuity of
F and [a, b] [c, d] ,
b
a
fX
[c,d]
dF =
fd
Here is the Lebesgue Stieltjes measure dened above.
8.11. THE ONE DIMENSIONAL LEBESGUE INTEGRAL 199
Proof: Since F is an increasing function it can have only countably many disconti-
nuities. The reason for this is that the only kind of discontinuity it can have is where
F (x+) > F (x) . Now since F is increasing, the intervals (F (x) , F (x+)) for x a
point of discontinuity are disjoint and so since each must contain a rational number and
the rational numbers are countable, and therefore so are these intervals.
Let D denote this countable set of discontinuities of F. Then if l, r / D, [l, r] [a, b] ,
it follows quickly from the denition of the Darboux Stieltjes integral that
b
a
X
[l,r)
dF = F (r) F (l) = F (r) F (l)
= ([l, r)) =
X
[l,r)
d.
Now let {s
n
} be the sequence of step functions of Lemma 8.11.2 such that these step
functions converge uniformly to f on [c, d] Then
_
X
[c,d]
f X
[c,d]
s
n
_
d
X
[c,d]
(f s
n
)
d
1
n
([c, d])
and
b
a
_
X
[c,d]
f X
[c,d]
s
n
_
dF
b
a
X
[c,d]
|f s
n
| dF <
1
n
(F (b) F (a)) .
Also if s
n
is given by the formula of Lemma 8.11.2,
X
[c,d]
s
n
d =
mn
k=1
f
_
z
n
k1
_
X
[z
n
k1
,z
n
k
)
d
=
mn
k=1
f
_
z
n
k1
_
X
[z
n
k1
,z
n
k
)
d
=
mn
k=1
f
_
z
n
k1
_
_
[z
n
k1
, z
n
k
)
_
=
mn
k=1
f
_
z
n
k1
_ _
F (z
n
k
) F
_
z
n
k1
__
=
mn
k=1
f
_
z
n
k1
_ _
F (z
n
k
) F
_
z
n
k1
__
=
mn
k=1
b
a
f
_
z
n
k1
_
X
[z
n
k1
,z
n
k
)
dF =
b
a
s
n
dF.
Therefore,
X
[c,d]
fd
b
a
X
[c,d]
fdF
X
[c,d]
fd
X
[c,d]
s
n
d
X
[c,d]
s
n
d
b
a
s
n
dF
b
a
s
n
dF
b
a
X
[c,d]
fdF
1
n
([c, d]) +
1
n
(F (b) F (a))
and since n is arbitrary, this shows
fd
b
a
fdF = 0.
In particular, in the special case where F is continuous and f is continuous,
b
a
fdF =
X
[a,b]
fd.
Thus, if F (x) = x so the Darboux Stieltjes integral is the usual integral from calculus,
b
a
f (t) dt =
X
[a,b]
fd
where is the measure which comes from F (x) = x as described above. This measure
is often denoted by m. Thus when f is continuous
b
a
f (t) dt =
X
[a,b]
fdm
and so there is no problem in writing
b
a
f (t) dt
for either the Lebesgue or the Riemann integral. Furthermore, when f is continuous,
you can compute the Lebesgue integral by using the fundamental theorem of calculus
because in this case, the two integrals are equal.
8.12 Exercises
1. Let = N ={1, 2, }. Let F = P(N), the set of all subsets of N, and let (S) =
number of elements in S. Thus ({1}) = 1 = ({2}), ({1, 2}) = 2, etc. Show
(, F, ) is a measure space. It is called counting measure. What functions are
measurable in this case? For a nonnegative function, f dened on N, show
N
fd =
k=1
f (k)
What do the monotone convergence and dominated convergence theorems say
about this example?
2. For the measure space of Problem 1, give an example of a sequence of nonneg-
ative measurable functions {f
n
} converging pointwise to a function f, such that
inequality is obtained in Fatous lemma.
3. If (, F, ) is a measure space and f 0 is measurable, show that if g () = f ()
a.e. and g 0, then
gd =
fd. Show that if f, g L

1
() and g () = f ()
a.e. then
gd =
fd.
8.12. EXERCISES 201
4. An algebra A of subsets of is a subset of the power set such that is in the
algebra and for A, B A, A \ B and A B are both in A. Let C {E
i
}
i=1
be
a countable collection of sets and let
1

i=1
E
i
. Show there exists an algebra
of sets A, such that A C and A is countable. Note the dierence between this
problem and Problem 5. Hint: Let C
1
denote all nite unions of sets of C and
1
. Thus C
1
is countable. Now let B
1
denote all complements with respect to
1
of sets of C
1
. Let C
2
denote all nite unions of sets of B
1
C
1
. Continue in this
way, obtaining an increasing sequence C
n
, each of which is countable. Let
A
i=1
C
i
.
5. Let A P () where P () denotes the set of all subsets of . Let (A) denote
the intersection of all algebras which contain A, one of these being P (). Show
(A) is also a algebra.
6. We say a function g mapping a normed vector space, to a normed vector space
is Borel measurable if whenever U is open, g
1
(U) is a Borel set. (The Borel
sets are those sets in the smallest algebra which contains the open sets.) Let
f : X and let g : X Y where X is a normed vector space and Y equals
C, R, or (, ] and F is a algebra of sets of . Suppose f is measurable and
g is Borel measurable. Show g f is measurable.
7. Let (, F, ) be a measure space. Dene : P() [0, ] by
(A) = inf{(B) : B A, B F}.
Show satises
() = 0, if A B, (A) (B),
(
i=1
A
i
)
i=1
(A
i
), (A) = (A) if A F.
If satises these conditions, it is called an outer measure. This shows every
measure determines an outer measure on the power set.
8. Let {E
i
} be a sequence of measurable sets with the property that
i=1
(E
i
) < .
Let S = { such that E
i
for innitely many values of i}. Show (S) = 0
and S is measurable. This is part of the Borel Cantelli lemma. Hint: Write S
in terms of intersections and unions. Something is in S means that for every n
there exists k > n such that it is in E
k
. Remember the tail of a convergent series
is small.
9. Let {f
n
} , f be measurable functions with values in C. {f
n
} converges in measure
if
lim
n
(x : |f(x) f
n
(x)| ) = 0
for each xed > 0. Prove the theorem of F. Riesz. If f
n
converges to f in
measure, then there exists a subsequence {f
n
k
} which converges to f a.e. Hint:
Choose n
1
such that
(x : |f(x) f
n1
(x)| 1) < 1/2.
Choose n
2
> n
1
such that
(x : |f(x) f
n2
(x)| 1/2) < 1/2
2
,
n
3
> n
2
such that
(x : |f(x) f
n3
(x)| 1/3) < 1/2
3
,
etc. Now consider what it means for f
n
k
(x) to fail to converge to f(x). Then use
Problem 8.
10. Suppose (, ) is a nite measure space (() < ) and S L
1
(). Then S is
said to be uniformly integrable if for every > 0 there exists > 0 such that if E
is a measurable set satisfying (E) < , then
E
|f| d <
for all f S. Show S is uniformly integrable and bounded in L
1
() if there
exists an increasing function h which satises
lim
t
h(t)
t
= , sup
_
h(|f|) d : f S
_
< .
S is bounded if there is some number, M such that
|f| d M
for all f S.
11. Let (, F, ) be a measure space and suppose f, g : (, ] are measurable.
Prove the sets
{ : f() < g()} and { : f() = g()}
are measurable. Hint: The easy way to do this is to write
{ : f() < g()} =
rQ
[f < r] [g > r] .
Note that l (x, y) = xy is not continuous on (, ] so the obvious idea doesnt
work.
12. Let {f
n
} be a sequence of real or complex valued measurable functions. Let
S = { : {f
n
()} converges}.
Show S is measurable. Hint: You might try to exhibit the set where f
n
converges
in terms of countable unions and intersections using the denition of a Cauchy
sequence.
13. Suppose u
n
(t) is a dierentiable function for t (a, b) and suppose that for t
(a, b),
|u
n
(t)|, |u
n
(t)| < K
n
where
n=1
K
n
< . Show
(
n=1
u
n
(t))
n=1
u
n
(t).
Hint: This is an exercise in the use of the dominated convergence theorem and
the mean value theorem.
8.12. EXERCISES 203
14. Let E be a countable subset of R. Show m(E) = 0. Hint: Let the set be {e
i
}
i=1
and let e
i
be the center of an open interval of length /2
i
.
15. If S is an uncountable set of irrational numbers, is it necessary that S has
a rational number as a limit point? Hint: Consider the proof of Problem 14
when applied to the rational numbers. (This problem was shown to me by Lee
Erlebach.)
16. Suppose {f
n
} is a sequence of nonnegative measurable functions dened on a
measure space, (, S, ). Show that
k=1
f
k
d =
k=1
f
k
d.
Hint: Use the monotone convergence theorem along with the fact the integral is
linear.
17. The integral
f (t) dt will denote the Lebesgue integral taken with respect to

one dimensional Lebesgue measure as discussed earlier. Show that for > 0, t
e
at
2
is in L
1
(R). The gamma function is dened for x > 0 as
(x)

0
e
t
t
x1
dt
Show t e
t
t
x1
is in L
1
(R) for all x > 0. Also show that
(x + 1) = x(x) , (1) = 1.
How does (n) for n an integer compare with (n 1)!?
18. This problem outlines a treatment of Stirlings formula which is a very useful
approximation to n! based on a section in [34]. It is an excellent application of
the monotone convergence theorem. Follow and justify the following steps using
the convergence theorems for the Lebesgue integral as needed. Here x > 0.
(x + 1) =

0
e
t
t
x
dt
First change the variables letting t = x(1 +u) to get (x + 1) =
e
x
x
x+1

1
_
e
u
(1 +u)
_
x
du
Next make the change of variables u = s
2
x
to obtain (x + 1) =
2e
x
x
x+(1/2)
x
2
_
e
s
2
x
_
1 +s
2
x
__
x
ds
The integrand is increasing in x. This is most easily seen by taking ln of the
integrand and then taking the derivative with respect to x. This derivative is
positive. Next show the limit of the integrand as x is e
s
2
. This isnt too
bad if you take ln and then use LHospitals rule. Consider the integral. Explain
why it must be increasing in x. Next justify the following assertion. Remember
the monotone convergence theorem applies to a sequence of functions.
lim
x
x
2
_
e
s
2
x
_
1 +s
2
x
__
x
ds =
e
s
2
ds
Now Stirlings formula is
lim
x
(x + 1)
2e
x
x
x+(1/2)
=
e
s
2
ds
where this last improper integral equals a well dened constant (why?). It is
very easy, when you know something about multiple integrals of functions of more
than one variable to verify this constant is

but the necessary mathematical
machinery has not yet been presented. It can also be done through much more
dicult arguments in the context of functions of only one variable. See [34] for
these clever arguments.
19. To show you the power of Stirlings formula, nd whether the series
n=1
n!e
n
n
n
converges. The ratio test falls at but you can try it if you like. Now explain why,
if n is large enough
n!
1
2
_

e
s
2
ds
_
2e
n
n
n+(1/2)
c
2e
n
n
n+(1/2)
.
Use this.
20. The Riemann integral is only dened for functions which are bounded which are
also dened on a bounded interval. If either of these two criteria are not satised,
then the integral is not the Riemann integral. Suppose f is Riemann integrable
on a bounded interval, [a, b]. Show that it must also be Lebesgue integrable with
respect to one dimensional Lebesgue measure and the two integrals coincide.
21. Give a theorem in which the improper Riemann integral coincides with a suitable
Lebesgue integral. (There are many such situations just nd one.)
22. Note that
0
sin x
x
dx is a valid improper Riemann integral dened by
lim
R
R
0
sin x
x
dx
but this function, sin x/x is not in L
1
([0, )). Why?
23. Let f be a nonnegative strictly decreasing function dened on [0, ). For 0 y
f (0), let f
1
(y) = x where y [f (x+) , f (x)]. (Draw a picture. f could have
jump discontinuities.) Show that f
1
is nonincreasing and that

0
f (t) dt =
f(0)
0
f
1
(y) dy.
Hint: Use the distribution function description.
24. Consider the following nested sequence of compact sets {P
n
}. We let P
1
= [0, 1],
P
2
=
_
0,
1
3
_
2
3
, 1
, etc. To go from P
n
to P
n+1
, delete the open interval which
is the middle third of each closed interval in P
n
. Let P =
n=1
P
n
. Since P is the
intersection of nested nonempty compact sets, it follows from advanced calculus
that P = . Show m(P) = 0. Show there is a one to one onto mapping of [0, 1] to
P. The set P is called the Cantor set. Thus, although P has measure zero, it has
the same number of points in it as [0, 1] in the sense that there is a one to one and
onto mapping from one to the other. Hint: There are various ways of doing this
last part but the most enlightenment is obtained by exploiting the construction
of the Cantor set.
8.12. EXERCISES 205
25. Consider the sequence of functions dened in the following way. Let f
1
(x) = x on
[0, 1]. To get from f
n
to f
n+1
, let f
n+1
= f
n
on all intervals where f
n
is constant. If
f
n
is nonconstant on [a, b], let f
n+1
(a) = f
n
(a), f
n+1
(b) = f
n
(b), f
n+1
is piecewise
linear and equal to
1
2
(f
n
(a) + f
n
(b)) on the middle third of [a, b]. Sketch a few
of these and you will see the pattern. The process of modifying a nonconstant
section of the graph of this function is illustrated in the following picture.
Show {f
n
} converges uniformly on [0, 1]. If f(x) = lim
n
f
n
(x), show that
f(0) = 0, f(1) = 1, f is continuous, and f
(x) = 0 for all x / P where P is

the Cantor set. This function is called the Cantor function.It is a very important
example to remember. Note it has derivative equal to zero a.e. and yet it succeeds
in climbing from 0 to 1. Thus
1
0
f
(t) dt = 0 = f (1) f (0) .

Is this somehow contradictory to the fundamental theorem of calculus? Hint:
This isnt too hard if you focus on getting a careful estimate on the dierence
between two successive functions in the list considering only a typical small interval
in which the change takes place. The above picture should be helpful.
26. Let m(W) > 0, W is measurable, W [a, b]. Show there exists a nonmeasurable
subset of W. Hint: Let x y if x y Q. Observe that is an equivalence
relation on R. See Denition 2.1.9 on Page 16 for a review of this terminology. Let
C be the set of equivalence classes and let D {C W : C C and C W = }.
By the axiom of choice, there exists a set, A, consisting of exactly one point from
each of the nonempty sets which are the elements of D. Show
W
rQ
A+r (a.)
A+r
1
A+r
2
= if r
1
= r
2
, r
i
Q. (b.)
Observe that since A [a, b], then A + r [a 1, b + 1] whenever |r| < 1. Use
this to show that if m(A) = 0, or if m(A) > 0 a contradiction results.Show there
exists some set, S such that m(S) < m(S A) +m(S \ A) where m is the outer
measure determined by m.
27. This problem gives a very interesting example found in the book by McShane
[31]. Let g(x) = x + f(x) where f is the strange function of Problem 25. Let
P be the Cantor set of Problem 24. Let [0, 1] \ P =
j=1
I
j
where I
j
is open
and I
j
I
k
= if j = k. These intervals are the connected components of the
complement of the Cantor set. Show m(g(I
j
)) = m(I
j
) so
m(g(
j=1
I
j
)) =
j=1
m(g(I
j
)) =
j=1
m(I
j
) = 1.
Thus m(g(P)) = 1 because g([0, 1]) = [0, 2]. By Problem 26 there exists a set,
A g (P) which is non measurable. Dene (x) = X
A
(g(x)). Thus (x) = 0
unless x P. Tell why is measurable. (Recall m(P) = 0 and Lebesgue measure
is complete.) Now show that X
A
(y) = (g
1
(y)) for y [0, 2]. Tell why g
1
is
continuous but g
1
is not measurable. (This is an example of measurable
continuous = measurable.) Show there exist Lebesgue measurable sets which are
not Borel measurable. Hint: The function, is Lebesgue measurable. Now show
that Borel measurable = measurable.
28. If A is mS measurable, it does not follow that A is m measurable. Give an
example to show this is the case.
29. If f is a nonnegative Lebesgue measurable function, show there exists g a Borel
measurable function such that g (x) = f (x) a.e.
The Lebesgue Integral For
Functions Of p Variables
9.1 Systems
The approach to p dimensional Lebesgue measure will be based on a very elegant idea
due to Dynkin.
Denition 9.1.1 Let be a set and let K be a collection of subsets of . Then
K is called a system if , K and whenever A, B K, it follows A B K.
For example, if R
p
= , an example of a system would be the set of all open sets.
Another example would be sets of the form
p
k=1
A
k
where A
k
is a Lebesgue measurable
set.
The following is the fundamental lemma which shows these systems are useful.
Lemma 9.1.2 Let K be a system of subsets of , a set. Also let G be a collection
of subsets of which satises the following three properties.
1. K G
2. If A G, then A
C
G
3. If {A
i
}
i=1
is a sequence of disjoint sets from G then
i=1
A
i
G.
Then G (K) , where (K) is the smallest algebra which contains K.
Proof: First note that if
H {G : 1 - 3 all hold}
then H yields a collection of sets which also satises 1 - 3. Therefore, I will assume in
the argument that G is the smallest collection of sets satisfying 1 - 3, the intersection of
all such collections. Let A K and dene
G
A
{B G : A B G} .
I want to show G
A
satises 1 - 3 because then it must equal G since G is the smallest
collection of subsets of which satises 1 - 3. This will give the conclusion that for
A K and B G, A B G. This information will then be used to show that if
A, B G then A B G. From this it will follow very easily that G is a algebra
which will imply it contains (K). Now here are the details of the argument.
207
208 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF P VARIABLES
Since K is given to be a system, K G
A
. Property 3 is obvious because if {B
i
} is
a sequence of disjoint sets in G
A
, then
A
i=1
B
i
=
i=1
A B
i
G
because A B
i
G and the property 3 of G.
It remains to verify Property 2 so let B G
A
. I need to verify that B
C
G
A
. In
other words, I need to show that A B
C
G. However,
A B
C
=
_
A
C
(A B)
_
C
G
Here is why. Since B G
A
, AB G and since A K G it follows A
C
G. It follows
the union of the disjoint sets A
C
and (A B) is in G and then from 2 the complement
of their union is in G. Thus G
A
satises 1 - 3 and this implies since G is the smallest
such, that G
A
G. However, G
A
is constructed as a subset of G and so G = G
A
. This
proves that for every B G and A K, A B G. Now pick B G and consider
G
B
{A G : A B G} .
I just proved K G
B
. The other arguments are identical to show G
B
satises 1 - 3 and
is therefore equal to G. This shows that whenever A, B G it follows A B G.
This implies G is a algebra. To show this, all that is left is to verify G is closed
under countable unions because then it follows G is a algebra. Let {A
i
} G. Then
let A
1
= A
1
and
A
n+1
A
n+1
\ (
n
i=1
A
i
)
= A
n+1
n
i=1
A
C
i
_
=
n
i=1
_
A
n+1
A
C
i
_
G
because nite intersections of sets of G are in G. Since the A
i
are disjoint, it follows
i=1
A
i
=
i=1
A
i
G
Therefore, G (K) because it is a algebra which contains K and This proves the
lemma.
9.2 p Dimensional Lebesgue Measure And Integrals
9.2.1 Iterated Integrals
Let m denote one dimensional Lebesgue measure. That is, it is the Lebesgue Stieltjes
measure which comes from the integrator function, F (x) = x. Also let the algebra of
measurable sets be denoted by F. Recall this algebra contained the open sets. Also
from the construction given above,
m([a, b]) = m((a, b)) = b a
Denition 9.2.1 Let f be a function of p variables and consider the symbol
f (x
1
, , x
p
) dx
i1
dx
ip
. (9.1)
where (i
1
, , i
p
) is a permutation of the integers {1, 2, , p} . The symbol means to
rst do the Lebesgue integral
f (x
1
, , x
p
) dx
i1
9.2. P DIMENSIONAL LEBESGUE MEASURE AND INTEGRALS 209
yielding a function of the other p 1 variables given above. Then you do
_
f (x
1
, , x
p
) dx
i1
_
dx
i2
and continue this way. The iterated integral is said to make sense if the process just
described makes sense at each step. Thus, to make sense, it is required
x
i1
f (x
1
, , x
p
)
can be integrated. Either the function has values in [0, ] and is measurable or it is a
function in L
1
. Then it is required
x
i2

f (x
1
, , x
p
) dx
i1
can be integrated and so forth. The symbol in 9.1 is called an iterated integral.
With the above explanation of iterated integrals, it is now time to dene p dimen-
sional Lebesgue measure.
9.2.2 p Dimensional Lebesgue Measure And Integrals
With the Lemma about systems given above and the monotone convergence theorem,
it is possible to give a very elegant and fairly easy denition of the Lebesgue integral of
a function of p real variables. This is done in the following proposition.
Proposition 9.2.2 There exists a algebra of sets of R
p
which contains the open
sets, F
p
and a measure m
p
dened on this algebra such that if f : R
p
[0, ) is
measurable with respect to F
p
then for any permutation (i
1
, , i
p
) of {1, , p} it
follows
R
p
fdm
p
=
f (x
1
, , x
p
) dx
i1
dx
ip
(9.2)
In particular, this implies that if A
i
is Lebesgue measurable for each i = 1, , p then
m
p
_
p
i=1
A
i
_
=
p
i=1
m(A
i
) .
Proof: Dene a system as
K
_
p
i=1
A
i
: A
i
is Lebesgue measurable
_
Also let R
n
[n, n]
p
, the p dimensional rectangle having sides [n, n]. A set F R
p
will be said to satisfy property P if for every n N and any two permutations of
{1, 2, , p}, (i
1
, , i
p
) and (j
1
, , j
p
) the two iterated integrals
X
RnF
dx
i1
dx
ip
,
X
RnF
dx
j1
dx
jp
make sense and are equal. Now dene G to be those subsets of R
p
which have property
P.
Thus K G because if (i
1
, , i
p
) is any permutation of {1, 2, , p} and
A =
p
i=1
A
i
K
then
X
RnA
dx
i1
dx
ip
=
p
i=1
m([n, n] A
i
) .
Now suppose F G and let (i
1
, , i
p
) and (j
1
, , j
p
) be two permutations. Then
R
n
=
_
R
n
F
C
_
(R
n
F)
and so
X
RnF
Cdx
i1
dx
ip
=
(X
Rn
X
RnF
) dx
i1
dx
ip
Since R
n
G the iterated integrals on the right and hence on the left make sense. Then
continuing with the expression on the right and using that F G, it equals
(2n)
p
X
RnF
dx
i1
dx
ip
= (2n)
p
X
RnF
dx
j1
dx
jp
=
(X
Rn
X
RnF
) dx
j1
dx
jp
=
X
RnF
Cdx
j1
dx
jp
which shows that if F G then so is F
C
.
Next suppose {F
i
}
i=1
is a sequence of disjoint sets in G. Let F =
i=1
F
i
. I need to
show F G. Since the sets are disjoint,
X
RnF
dx
i1
dx
ip
=
k=1
X
RnF
k
dx
i1
dx
ip
=
lim
N
N
k=1
X
RnF
k
dx
i1
dx
ip
Do the iterated integrals make sense? Note that the iterated integral makes sense for
N
k=1
X
RnF
k
as the integrand because it is just a nite sum of functions for which the
iterated integral makes sense. Therefore,
x
i1

k=1
X
RnF
k
(x)
is measurable and by the monotone convergence theorem,
k=1
X
RnF
k
(x) dx
i1
= lim
N
k=1
X
RnF
k
dx
i1
Now each of the functions,
x
i2

k=1
X
RnF
k
dx
i1
is measurable and so the limit of these functions,
k=1
X
RnF
k
(x) dx
i1
is also measurable. Therefore, one can do another integral to this function. Continu-
ing this way using the monotone convergence theorem, it follows the iterated integral
makes sense. The same reasoning shows the iterated integral makes sense for any other
permutation.
Now applying the monotone convergence theorem as needed,
X
RnF
dx
i1
dx
ip
=
k=1
X
RnF
k
dx
i1
dx
ip
=
lim
N
N
k=1
X
RnF
k
dx
i1
dx
ip
=
lim
N
N
k=1
X
RnF
k
dx
i1
dx
ip
=
lim
N
N
k=1

X
RnF
k
dx
i1
dx
ip

= lim
N
N
k=1
X
RnF
k
dx
i1
dx
ip
= lim
N
N
k=1
X
RnF
k
dx
j1
dx
jp
the last step holding because each F
k
G. Then repeating the steps above in the
opposite order, this equals
k=1
X
RnF
k
dx
j1
dx
jp
=
X
RnF
dx
j1
dx
jp
Thus F G. By Lemma 9.1.2 G (K) .
Let F
p
= (K). Each set of the form
p
k=1
U
k
where U
k
is an open set is in K. Also
every open set in R
p
is a countable union of open sets of this form. This follows from
Lemma 7.7.7 on Page 176. Therefore, every open set is in F
p
.
For F F
p
dene
m
p
(F) lim
n
X
RnF
dx
j1
dx
jp
where (j
1
, , j
p
) is a permutation of {1, , p} . It doesnt matter which one. It was
shown above they all give the same result. I need to verify m
p
is a measure. Let {F
k
}
be a sequence of disjoint sets of F
p
.
m
p
(
k=1
F
k
) = lim
n
k=1
X
RnF
k
dx
j1
dx
jp
.
Using the monotone convergence theorem repeatedly as in the rst part of the argument,
this equals
k=1
lim
n
X
RnF
k
dx
j1
dx
jp

k=1
m
p
(F
k
) .
Thus m
p
is a measure. Now letting A
k
be a Lebesgue measurable set,
m
p
_
p
k=1
A
k
_
= lim
n
k=1
X
[n,n]A
k
(x
k
) dx
j1
dx
jp
= lim
n
p
k=1
m([n, n] A
k
) =
p
k=1
m(A
k
) .
It only remains to prove 9.2.
It was shown above that for F F it follows
R
p
X
F
dm
p
= lim
n
X
RnF
dx
j1
dx
jp
Applying the monotone convergence theorem repeatedly on the right, this yields that
the iterated integral makes sense and
R
p
X
F
dm
p
=
X
F
dx
j1
dx
jp
It follows 9.2 holds for every nonnegative simple function in place of f because these are
just linear combinations of functions, X
F
. Now taking an increasing sequence of non-
negative simple functions, {s
k
} which converges to a measurable nonnegative function
f
R
p
fdm
p
= lim
k
R
p
s
k
dm
p
= lim
k
s
k
dx
j1
dx
jp
=
fdx
j1
dx
jp
9.2.3 Fubinis Theorem
Formula 9.2 is often called Fubinis theorem. So is the following theorem. In general,
people tend to refer to theorems about the equality of iterated integrals as Fubinis
theorem and in fact Fubini did produce such theorems but so did Tonelli and some of
these theorems presented here and above should be called Tonellis theorem.
Theorem 9.2.3 Let m
p
be dened in Proposition 9.2.2 on the algebra of
sets F
p
given there. Suppose f L
1
(R
p
) . Then if (i
1
, , i
p
) is any permutation
of {1, , p} ,
R
p
fdm
p
=
f (x) dx
i1
dx
ip
.
In particular, iterated integrals for any permutation of {1, , p} are all equal.
Proof: It suces to prove this for f having real values because if this is shown the
general case is obtained by taking real and imaginary parts. Since f L
1
(R
p
) ,
R
p
|f| dm
p
<
and so both
1
2
(|f| +f) and
1
2
(|f| f) are in L
1
(R
p
) and are each nonnegative. Hence
from Proposition 9.2.2,
R
p
fdm
p
=
R
p
_
1
2
(|f| +f)
1
2
(|f| f)
_
dm
p
=
R
p
1
2
(|f| +f) dm
p
R
p
1
2
(|f| f) dm
p
=
1
2
(|f (x)| +f (x)) dx
i1
dx
ip
1
2
(|f (x)| f (x)) dx
i1
dx
ip
=
1
2
(|f (x)| +f (x))
1
2
(|f (x)| f (x)) dx
i1
dx
ip
=
f (x) dx
i1
dx
ip
The following corollary is a convenient way to verify the hypotheses of the above
theorem.
Corollary 9.2.4 Suppose f is measurable with respect to F
p
and suppose for some
permutation, (i
1
, , i
p
)
|f (x)| dx
i1
dx
ip
<
Then f L
1
(R
p
) .
Proof: By Proposition 9.2.2,
R
p
|f| dm
p
=
|f (x)| dx
i1
dx
ip
<
and so f is in L
1
(R
p
) by Corollary 8.7.6. This proves the corollary.
The following theorem is a summary of the above specialized to Borel sets along
with an assertion about regularity.
Theorem 9.2.5 Let B (R
p
) be the Borel sets on R
p
. There exists a measure m
p
dened on B (R
p
) such that if f is a nonnegative Borel measurable function,
R
p
fdm
p
=
f (x) dx
i1
dx
ip
(9.3)
whenever (i
1
, , i
p
) is a permutation of {1. , p}. If f L
1
(R
p
) and f is Borel
measurable, then the above equation holds for f and all integrals make sense. If f is
Borel measurable and for some (i
1
, , i
p
) a permutation of {1. , p}
|f (x)| dx
i1
dx
ip
< ,
then f L
1
(R
p
). The measure m
p
is both inner and outer regular on the Borel sets.
That is, if E B (R
p
),
m
p
(E) = sup {m
p
(K) : K E and K is compact}
m
p
(E) = inf {m
p
(V ) : V E and V is open} .
Also if A
k
is a Borel set in R then
p
k=1
A
k
is a Borel set in R
p
and
m
p
_
p
k=1
A
k
_
=
p
k=1
m(A
k
) .
Proof: Most of it was shown earlier since B (R
p
) F
p
. The two assertions about
regularity follow from observing that m
p
is nite on compact sets and then using Theo-
rem 7.4.6. It remains to show the assertion about the product of Borel sets. If each A
k
is open, there is nothing to show because the result is an open set. Suppose then that
whenever A
1
, , A
m
, m p are open, the product,
p
k=1
A
k
is a Borel set. Let K be
the open sets in R and let G be those Borel sets such that if A
m
G it follows
p
k=1
A
k
is Borel. Then K is a system and is contained in G. Now suppose F G. Then
_
m1
k=1
A
k
F
p
k=m+1
A
k
_
_
m1
k=1
A
k
F
C
k=m+1
A
k
_
=
_
m1
k=1
A
k
R
p
k=m+1
A
k
_
and by assumption this is of the form
B A = D.
where B, A are disjoint and B and D are Borel. Therefore, A = D\ B which is a Borel
set. Thus G is closed with respect to complements. If {F
i
} is a sequence of disjoint
elements of G
_
m1
k=1
A
k

i
F
i
k=m+1
A
k
_
=
i=1
_
m1
k=1
A
k
F
i
k=m+1
A
k
_
which is a countable union of Borel sets and is therefore, Borel. Hence G is also closed
with respect to countable unions of disjoint sets. Thus by the Lemma on systems
G (K) = B (R) and this shows that A
m
can be any Borel set. Thus the assertion
about the product is true if only A
1
, , A
m1
are open while the rest are Borel.
Continuing this way shows the assertion remains true for each A
i
being Borel. Now the
nal formula about the measure of a product follows from 9.3.
R
p
X
p
k=1
A
k
dm
p
=
p
k=1
A
k
(x) dx
1
dx
p
=
k=1
X
A
k
(x
k
) dx
1
dx
p
=
p
k=1
m(A
k
) .
Of course iterated integrals can often be used to compute the Lebesgue integral.
Sometimes the iterated integral taken in one order will allow you to compute the
Lebesgue integral and it does not work well in the other order. Here is a simple example.
9.3. EXERCISES 215
Example 9.2.6 Find the iterated integral
1
0
1
x
sin (y)
y
dydx
Notice the limits. The iterated integral equals
R
2
X
A
(x, y)
sin (y)
y
dm
2
where
A = {(x, y) : x y where x [0, 1]}
Fubinis theorem can be applied because the function (x, y) sin (y) /y is continuous
except at y = 0 and can be redened to be continuous there. The function is also
bounded so
(x, y) X
A
(x, y)
sin (y)
y
clearly is in L
1
_
R
2
_
. Therefore,
R
2
X
A
(x, y)
sin (y)
y
dm
2
=

X
A
(x, y)
sin (y)
y
dxdy
=
1
0
y
0
sin (y)
y
dxdy
=
1
0
sin (y) dy = 1 cos (1)
9.3 Exercises
1. Find
2
0
62z
0
3z
1
2
x
(3 z) cos
_
y
2
_
dy dxdz.
2. Find
1
0
183z
0
6z
1
3
x
(6 z) exp
_
y
2
_
dy dxdz.
3. Find
2
0
244z
0
6z
1
4
y
(6 z) exp
_
x
2
_
dxdy dz.
4. Find
1
0
124z
0
3z
1
4
y
sin x
x
dxdy dz.
5. Find
20
0
1
0
5z
1
5
y
sin x
x
dxdz dy +
25
20
5
1
5
y
0
5z
1
5
y
sin x
x
dxdz dy. Hint: You might
try doing it in the order, dy dxdz
6. Explain why for each t > 0, x e
tx
is a function in L
1
(R) and

0
e
tx
dx =
1
t
.
Thus
R
0
sin (t)
t
dt =
R
0

0
sin (t) e
tx
dxdt
Now explain why you can change the order of integration in the above iterated
integral. Then compute what you get. Next pass to a limit as R and show

0
sin (t)
t
dt =
1
2

7. Explain why
a
f (t) dt lim
r
r
a
f (t) dt whenever f L
1
(a, ) ; that is
fX
[a,)
L
1
(R).
8. B(p, q) =
1
0
x
p1
(1 x)
q1
dx, (p) =
0
e
t
t
p1
dt for p, q > 0. The rst of
these is called the beta function, while the second is the gamma function. Show
a.) (p + 1) = p(p); b.) (p)(q) = B(p, q)(p + q).Explain why the gamma
function makes sense for any p > 0.
9. Let f (y) = g (y) = |y|
1/2
if y (1, 0) (0, 1) and f (y) = g (y) = 0 if y /
(1, 0) (0, 1). For which values of x does it make sense to write the integral
R
f (x y) g (y) dy?
10. Let {a
n
} be an increasing sequence of numbers in (0, 1) which converges to 1.
Let g
n
be a nonnegative function which equals zero outside (a
n
, a
n+1
) such that
g
n
dx = 1. Now for (x, y) [0, 1) [0, 1) dene
f (x, y)
k=1
g
n
(y) (g
n
(x) g
n+1
(x)) .
Explain why this is actually a nite sum for each such (x, y) so there are no
convergence questions in the innite sum. Explain why f is a continuous function
on [0, 1) [0, 1). You can extend f to equal zero o [0, 1) [0, 1) if you like. Show
the iterated integrals exist but are not equal. In fact, show
1
0
1
0
f (x, y) dydx = 1 = 0 =
1
0
1
0
f (x, y) dxdy.
Does this example contradict the Fubini theorem? Explain why or why not.
9.4 Lebesgue Measure On R
p
The algebra of Lebesgue measurable sets is larger than the above algebra of Borel
sets or of the earlier algebra which came from an application of the system lemma. It
is convenient to use this larger algebra, especially when considering change of variables
formulas, although it is certainly true that one can do most interesting theorems with
the Borel sets only. However, it is in some ways easier to consider the more general
situation and this will be done next.
Denition 9.4.1 The completion of (R
p
, B (R
p
) , m
p
) is the Lebesgue measure
space. Denote this by (R
p
, F
p
, m
p
) .
Thus for each E F
p
,
m
p
(E) = inf {m
p
(F) : F E and F B (R
p
)}
It follows that for each E F
p
there exists F B (R
p
) such that F E and
m
p
(E) = m
p
(F) .
Theorem 9.4.2 m
p
is regular on F
p
. In fact, if E F
p
, there exist sets in
B (R
p
) , F, G such that
F E G,
F is a countable union of compact sets, G is a countable intersection of open sets and
m
p
(G\ F) = 0.
9.4. LEBESGUE MEASURE ON R
P
217
If A
k
is Lebesgue measurable then
p
k=1
A
k
F
p
and
m
p
_
p
k=1
A
k
_
=
p
k=1
m(A
k
) .
In addition to this, m
p
is translation invariant. This means that if E F
p
and x R
p
,
then
m
p
(x+E) = m
p
(E) .
The expression x +E means {x +e : e E} .
Proof: m
p
is regular on B (R
p
) by Theorem 7.4.6 because it is nite on compact
sets. Then from Theorem 7.5.7, it follows m
p
is regular on F
p
. This is because the
regularity of m
p
on the Borel sets and the denition of the completion of a measure
given there implies the uniqueness part of Theorem 7.5.6 can be used to conclude
(R
p
, F
p
, m
p
) =
_
R
p
, B (R
p
), m
p
_
Now for E F
p
, having nite measure, there exists F B (R
p
) such that m
p
(F) =
m
p
(E) and F E. Thus m
p
(F \ E) = 0. By regularity of m
p
on the Borel sets, there
exists G a countable intersection of open sets such that G F and m
p
(G) = m
p
(F) =
m
p
(E) . Thus m
p
(G\ E) = 0. If E does not have nite measure, then letting
E
m
(B(0, m) \ B(0, m1)) E,
it follows there exists G
m
, an intersection of open sets such that m
p
(G
m
\ E
m
) = 0.
Hence if G =
m
G
m
, it follows m
p
(G\ E)

m
m
p
(G
m
\ E
m
) = 0. Thus G is a
countable intersection of open sets.
To obtain F a countable union of compact sets contained in E such that m
p
(E \ F) =
0, consider the closed sets A
m
= B(0,m) \ B(0, m1) and let E
m
= A
m
E. Then
from what was just shown, there exists G
m
(A
m
\ E
m
) such that
m
p
(G
m
\ (A
m
\ E
m
)) = 0.
and G
m
is the countable intersection of open sets. The set on the inside equals
_
G
m
A
C
m
_
(G
m
E
m
) . Also
G
C
m
A
C
m
E
m
so G
C
m
A
m
E
m
and G
C
m
A
m
is the countable union of closed sets. Also
m
p
_
E
m
\
_
G
C
m
A
m
__
= m
p
_
_
_(E
m
G
m
)
_
_
_
=
..
E
m
A
C
m
_
_
_
_
_
_
m
p
__
G
m
A
C
m
_
(G
m
E
m
)
_
= 0.
Denote this set G
C
m
A
m
by F
m
. It is a countable union of closed sets and m
p
(E
m
\ F
m
) =
0. Let F =
m=1
F
m
. Then F is a countable union of compact sets and
m
p
(E \ F)
m=1
m
p
(E
m
\ F
m
) = 0.
Consider the next assertion about the measure of a Cartesian product. By regularity
of m there exists B
k
, C
k
B (R
p
) such that B
k
A
k
C
k
and m(B
k
) = m(A
k
) =
m(C
k
). In fact, you can have B
k
equal a countable intersection of open sets and C
k
a
countable union of compact sets. Then
p
k=1
m(A
k
) =
p
k=1
m(C
k
) m
p
_
p
k=1
C
k
_
m
p
_
p
k=1
A
k
_
m
p
_
p
k=1
B
k
_
=
p
k=1
m(B
k
) =
p
k=1
m(A
k
) .
It remains to prove the claim about the measure being translation invariant.
Let K denote all sets of the form
p
k=1
U
k
where each U
k
is an open set in R. Thus K is a system.
x +
p
k=1
U
k
=
p
k=1
(x
k
+U
k
)
which is also a nite Cartesian product of nitely many open sets. Also,
m
p
_
x +
p
k=1
U
k
_
= m
p
_
p
k=1
(x
k
+U
k
)
_
=
p
k=1
m(x
k
+U
k
)
=
p
k=1
m(U
k
) = m
p
_
p
k=1
U
k
_
The step to the last line is obvious because an arbitrary open set in R is the disjoint
union of open intervals and the lengths of these intervals are unchanged when they are
slid to another location.
Now let G denote those Borel sets E with the property that for each n N
m
p
(x +E (n, n)
p
) = m
p
(E (n, n)
p
)
and the set, x +E (n, n)
p
is a Borel set. Thus K G. If E G then
_
x +E
C
(n, n)
p
_
(x +E (n, n)
p
) = x + (n, n)
p
which implies x +E
C
(n, n)
p
is a Borel set since it equals a dierence of two Borel
sets. Now consider the following.
m
p
_
x +E
C
(n, n)
p
_
+m
p
(E (n, n)
p
)
= m
p
_
x +E
C
(n, n)
p
_
+m
p
(x +E (n, n)
p
)
= m
p
(x + (n, n)
p
) = m
p
((n, n)
p
)
= m
p
_
E
C
(n, n)
p
_
+m
p
(E (n, n)
p
)
9.5. MOLLIFIERS 219
which shows
m
p
_
x +E
C
(n, n)
p
_
= m
p
_
E
C
(n, n)
p
_
showing that E
C
G.
If {E
k
} is a sequence of disjoint sets of G,
m
p
(x +
k=1
E
k
(n, n)
p
) = m
p
(
k=1
x +E
k
(n, n)
p
)
Now the sets {x +E
k
(p, p)
n
} are also disjoint and so the above equals
k
m
p
(x +E
k
(n, n)
p
) =
k
m
p
(E
k
(n, n)
p
)
= m
p
(
k=1
E
k
(n, n)
p
)
Thus G is also closed with respect to countable disjoint unions. It follows from the
lemma on systems that G (K) . But from Lemma 7.7.7 on Page 176, every open
set is a countable union of sets of K and so (K) contains the open sets. Therefore,
B (R
p
) G (K) B (K) which shows G = B (R
p
).
I have just shown that for every E B (R
p
) , and any n N,
m
p
(x +E (n, n)
p
) = m
p
(E (n, n)
p
)
Taking the limit as n yields
m
p
(x +E) = m
p
(E) .
This proves translation invariance on Borel sets.
Now suppose m
p
(S) = 0 so that S is a set of measure zero. From outer regularity,
there exists a Borel set, F such that F S and m
p
(F) = 0. Therefore from what was
just shown,
m
p
(x +S) m
p
(x +F) = m
p
(F) = m
p
(S) = 0
which shows that if m
p
(S) = 0 then so does m
p
(x +S) . Let F be any set of F
p
. By
regularity, there exists E F where E B (R
p
) and m
p
(E \ F) = 0. Then
m
p
(F) = m
p
(E) = m
p
(x +E) = m
p
(x + (E \ F) F)
= m
p
(x +E \ F) +m
p
(x +F) = m
p
(x +F) .
9.5 Molliers
From Theorem 8.10.3, every function in L
1
(R
p
) can be approximated by one in C
c
(R
p
)
but even more incredible things can be said. In fact, you can approximate an arbitrary
function in L
1
(R
p
) with one which is innitely dierentiable having compact support.
This is very important in partial dierential equations. I am just giving a short intro-
duction to this concept here. Consider the following example.
Example 9.5.1 Let U = B(z, 2r)
(x) =
_
_
_
exp
_
_
|x z|
2
r
2
_
1
_
if |x z| < r,
0 if |x z| r.
Then a little work shows C
c
(U). The following also is easily obtained.
You show this by verifying the partial derivatives all exist and are continuous. The
only place this is hard is when |x z| = r. It is left as an exercise. You might consider
a simpler example,
f (x) =
_
e
1/x
2
if x = 0
0 if x = 0
and reduce the above to a consideration of something like this simpler case.
Lemma 9.5.2 Let U be any open set. Then C
c
(U) = .
Proof: Pick z U and let r be small enough that B(z, 2r) U. Then let
C
c
(B(z, 2r)) C
c
(U) be the function of the above example.
Denition 9.5.3 Let U = {x R
p
: |x| < 1}. A sequence {
m
} C
c
(U) is
called a mollier if
m
(x) 0,
m
(x) = 0, if |x|
1
m
,
and

m
(x) = 1. Sometimes it may be written as {
} where
satises the above

conditions except
(x) = 0 if |x| . In other words, takes the place of 1/m and in

everything that follows 0 instead of m .
f(x, y)dm
p
(y) will mean x is xed and the function y f(x, y) is being inte-
grated. To make the notation more familiar, dx is written instead of dm
p
(x).
Example 9.5.4 Let
C
c
(B(0, 1)) (B(0, 1) = {x : |x| < 1})
with (x) 0 and
dm = 1. Let
m
(x) = c
m
(mx) where c
m
is chosen in such a
way that

m
dm = 1.
Denition 9.5.5 A function, f, is said to be in L
1
loc
(R
p
) if f is Lebesgue
measurable and if |f|X
K
L
1
(R
p
) for every compact set, K. If f L
1
loc
(R
p
), and
g C
c
(R
p
),
f g(x)
f(y)g(x y)dy.
This is called the convolution of f and g.
The following is an important property of the convolution.
Proposition 9.5.6 Let f and g be as in the above denition. Then
f(y)g(x y)dy =
f(x y)g(y)dy
Proof: This follows right away from the change of variables formula. In the left,
let x y u. Then the left side equals
f (x u) g (u) du
because the absolute value of the determinant of the derivative is 1. Now replace u with
y and This proves the proposition.
The following lemma will be useful in what follows. It says among other things that
one of these very unregular functions in L
1
loc
(R
p
) is smoothed out by convolving with
a mollier.
9.5. MOLLIFIERS 221
Lemma 9.5.7 Let f L
1
loc
(R
p
), and g C
c
(R
p
). Then f g is an innitely
dierentiable function. Also, if {
m
} is a mollier and U is an open set and f
C
0
(U) L
1
loc
(R
p
) , then at every x U,
lim
m
f
m
(x) = f (x) .
If f C
1
(U) L
1
loc
(R
p
) and x U,
(f
m
)
xi
(x) = f
xi

m
(x) .
Proof: Consider the dierence quotient for calculating a partial derivative of f g.
f g (x +te
j
) f g (x)
t
=
f(y)
g(x +te
j
y) g (x y)
t
dy.
Using the fact that g C
c
(R
p
), the quotient,
g(x +te
j
y) g (x y)
t
,
is uniformly bounded. To see this easily, use Theorem 6.4.2 on Page 126 to get the
existence of a constant, M depending on
max {||Dg (x)|| : x R
p
}
such that
|g(x +te
j
y) g (x y)| M|t|
for any choice of x and y. Therefore, there exists a dominating function for the inte-
grand of the above integral which is of the form C |f (y)| X
K
where K is a compact set
depending on the support of g. It follows from the dominated convergence theorem the
limit of the dierence quotient above passes inside the integral as t 0 and so
x
j
(f g) (x) =
f(y)

x
j
g (x y) dy.
Now letting

xj
g play the role of g in the above argument, a repeat of the above
reasoning shows partial derivatives of all orders exist. A similar use of the dominated
convergence theorem shows all these partial derivatives are also continuous.
It remains to verify the claim about the mollier. Let x U and let m be large
enough that B
_
x,
1
m
_
U. Then
|f g (x) f (x)|
B(0,
1
m
)
|f (x y) f (x)|
m
(y) dy
By continuity of f at x, for all m suciently large, the above is dominated by
B(0,
1
m
)
m
(y) dy =
and this proves the claim.
Now consider the formula in the case where f C
1
(U). Using Proposition 9.5.6,
f
m
(x +he
i
) f
m
(x)
h
=
1
h
_
B(0,
1
m
)
f (x+he
i
y)
m
(y) dy
B(0,
1
m
)
f (x y)
m
(y) dy
_
=
B(0,
1
m
)
(f (x+he
i
y) f (x y))
h

m
(y) dy
Now letting m be small enough that B
_
0,
1
m
_
is contained in U along with the continuity
of the partial derivatives, it follows the dierence quotients are uniformly bounded for
all h suciently small and so one can apply the dominated convergence theorem and
pass to the limit obtaining
R
p
f
xi
(x y)
m
(y) dy f
xi

m
(x)
Theorem 9.5.8 Let K be a compact subset of an open set, U. Then there exists
a function, h C
c
(U), such that h(x) = 1 for all x K and h(x) [0, 1] for all x.
Also there exists an open set W such that
K W W U
such that W is compact.
Proof: Let r > 0 be small enough that K+B(0, 3r) U. The symbol, K+B(0, 3r)
means
{k +x : k K and x B(0, 3r)} .
Thus this is simply a way to write
{B(k, 3r) : k K} .
Think of it as fattening up the set, K. Let K
r
= K + B(0, r). A picture of what is
happening follows.
K K
r
U
Consider X
Kr
m
where
m
is a mollier. Let m be so large that
1
m
< r. Then from
the denition of what is meant by a convolution, and using that
m
has support in
B
_
0,
1
m
_
, X
Kr

m
= 1 on K and its support is in K +B(0, 3r), a bounded set. Now
using Lemma 9.5.7, X
Kr
m
is also innitely dierentiable. Therefore, let h = X
Kr
m
.
As to the existence of the open set W, let it equal the closed and bounded set
h
1
__
1
2
, 1
_
The following is the remarkable theorem mentioned above. First, here is some no-
tation.
Denition 9.5.9 Let g be a function dened on a vector space. Then g
y
(x)
g (x y) .
Theorem 9.5.10 C
c
(R
p
) is dense in L
1
(R
p
). Here the measure is Lebesgue
measure.
Proof: Let f L
1
(R
p
) and let > 0 be given. Choose g C
c
(R
p
) such that
|g f| dm
p
< /2
9.5. MOLLIFIERS 223
This can be done by using Theorem 8.10.3. Now let
g
m
(x) = g
m
(x)
g (x y)
m
(y) dm
p
(y) =
g (y)
m
(x y) dm
p
(y)
where {
m
} is a mollier. It follows from Lemma 9.5.7 g
m
C
c
(R
p
). It vanishes if
x / spt(g) +B(0,
1
m
).
|g g
m
| dm
p
=
|g(x)
g(x y)
m
(y)dm
p
(y)|dm
p
(x)
|g(x) g(x y)|

m
(y)dm
p
(y))dm
p
(x)

|g(x) g(x y)|dm
p
(x)
m
(y)dm
p
(y)
=
B(0,
1
m
)
|g g
y
|dm
p
(x)
m
(y)dm
p
(y) <

2
whenever m is large enough. This follows because since g has compact support, it is
uniformly continuous on R
p
and so if > 0 is given, then whenever |y| is suciently
small,
|g (x) g (x y)| <
for all x. Thus, since g has compact support, if y is small enough, it follows
|g g
y
|dm
p
(x) < /2.
There is no measurability problem in the use of Fubinis theorem because the function
(x, y) |g(x) g(x y)|
m
(y)
is continuous. Thus when m is large enough,
|f g
m
| dm
p

|f g| dm
p
+
|g g
m
| dm
p
<

2
+

2
= .
Another important application of Theorem 9.5.8 has to do with a partition of unity.
Denition 9.5.11 A collection of sets H is called locally nite if for every x,
there exists r > 0 such that B(x, r) has nonempty intersection with only nitely many
sets of H. Of course every nite collection of sets is locally nite. This is the case of
most interest in this book but the more general notion is interesting.
The thing about locally nite collection of sets is that the closure of their union
equals the union of their closures. This is clearly true of a nite collection.
Lemma 9.5.12 Let H be a locally nite collection of sets of a normed vector space
V . Then
H =
_
H : H H
_
.
Proof: It is obvious holds in the above claim. It remains to go the other way.
Suppose then that p is a limit point of H and p / H. There exists r > 0 such
that B(p, r) has nonempty intersection with only nitely many sets of H say these are
H
1
, , H
m
. Then I claim p must be a limit point of one of these. If this is not so, there
would exist r
such that 0 < r
< r with B(p, r
) having empty intersection with each

of these H
i
. But then p would fail to be a limit point of H. Therefore, p is contained
in the right side. It is clear H is contained in the right side and so This proves the
lemma.
A good example to consider is the rational numbers each being a set in R. This
is not a locally nite collection of sets and you note that Q = R = {x : x Q} . By
contrast, Z is a locally nite collection of sets, the sets consisting of individual integers.
The closure of Z is equal to Z because Z has no limit points so it contains them all.
Notation 9.5.13 I will write V to symbolize C
c
(V ) , has values in [0, 1] ,
and has compact support in V . I will write K V for K compact and V open to
symbolize is 1 on K and has values in [0, 1] with compact support contained in V .
Lemma 9.5.14 Let K be a closed set and let {V
i
}
i=1
be a locally nite list of bounded
open sets whose union contains K. Then there exist functions,
i
C
c
(V
i
) such that
for all x K,
1 =
i=1
i
(x)
and the function f (x) given by
f (x) =
i=1
i
(x)
is in C
(R
p
) .
Proof: Let K
1
= K \
i=2
V
i
. Thus K
1
is compact because K
1
V
1
. Let W
1
be
an open set having compact closure which satises
K
1
W
1
W
1
V
1
Thus W
1
, V
2
, , V
n
covers K and W
1
V
1
. Suppose W
1
, , W
r
have been dened
such that W
i
V
i
for each i, and W
1
, , W
r
, V
r+1
, , V
n
covers K. Then let
K
r+1
K \ (
_
i=r+2
V
i
_
r
j=1
W
j
_
).
It follows K
r+1
is compact because K
r+1
V
r+1
. Let W
r+1
satisfy
K
r+1
W
r+1
W
r+1
V
r+1
, W
r+1
is compact
Continuing this way denes a sequence of open sets {W
i
}
i=1
having compact closures
with the property
W
i
V
i
, K
i=1
W
i
.
Note {W
i
}
i=1
is locally nite because the original list, {V
i
}
i=1
was locally nite. Now
let U
i
be open sets which satisfy
W
i
U
i
U
i
V
i
, U
i
is compact.
Similarly, {U
i
}
i=1
is locally nite.
W
i
U
i
V
i
9.6. THE VITALI COVERING THEOREM 225
Since the sets {W
i
}
i=1
are locally nite, it follows
i=1
W
i
=
i=1
W
i
and so it is
possible to dene
i
and , innitely dierentiable functions having compact support
such that
U
i

i
V
i
,
i=1
W
i

i=1
U
i
.
Now dene
i
(x) =
_
(x)
i
(x)/
j=1
j
(x) if
j=1
j
(x) = 0,
0 if
j=1
j
(x) = 0.
If x is such that
j=1
j
(x) = 0, then x /
i=1
U
i
because
i
equals one on U
i
.
Consequently (y) = 0 for all y near x thanks to the fact that
i=1
U
i
is closed and
so
i
(y) = 0 for all y near x. Hence
i
is innitely dierentiable at such x. If
j=1
j
(x) = 0, this situation persists near x because each
j
is continuous and so
i
is innitely dierentiable at such points also. Therefore
i
is innitely dierentiable. If
x K, then (x) = 1 and so
j=1
j
(x) = 1. Clearly 0
i
(x) 1 and spt(
j
) V
j
.
The functions, {
i
} are called a C
partition of unity.
The method of proof of this lemma easily implies the following useful corollary.
Corollary 9.5.15 If H is a compact subset of V
i
for some V
i
there exists a partition
of unity such that
i
(x) = 1 for all x H in addition to the conclusion of Lemma
9.5.14.
Proof: Keep V
i
the same but replace V
j
with

V
j
V
j
\ H. Now in the proof above,
applied to this modied collection of open sets, if j = i,
j
(x) = 0 whenever x H.
Therefore,
i
(x) = 1 on H.
9.6 The Vitali Covering Theorem
The Vitali covering theorem is a profound result about coverings of a set in R
p
with
open balls. The balls can be dened in terms of any norm for R
p
. For example, the
norm could be
||x|| max {|x
k
| : k = 1, , p}
or the usual norm
|x| =
k
|x
k
|
2
or any other. The proof given here is from Basic Analysis [27]. It rst considers the
case of open balls and then generalizes to balls which may be neither open nor closed.
Lemma 9.6.1 Let F be a countable collection of balls satisfying
> M sup{r : B(p, r) F} > 0
and let k (0, ) . Then there exists G F such that
If B(p, r) G then r > k, (9.4)
If B
1
, B
2
G then B
1
B
2
= , (9.5)
G is maximal with respect to 9.4 and 9.5. (9.6)
By this is meant that if H is a collection of balls satisfying 9.4 and 9.5, then H cannot
properly contain G.
Proof: If no ball of F has radius larger than k, let G = . Assume therefore, that
some balls have radius larger than k. Let F {B
i
}
i=1
. Now let B
n1
be the rst ball
in the list which has radius greater than k. If every ball having radius larger than k
intersects this one, then stop. The maximal set is {B
n1
} . Otherwise, let B
n2
be the
next ball having radius larger than k which is disjoint from B
n1
. Continue this way
obtaining {B
ni
}
i=1
, a nite or innite sequence of disjoint balls having radius larger
than k. Then let G {B
ni
}. To see G is maximal with respect to 9.4 and 9.5, suppose
B F, B has radius larger than k, and G {B} satises 9.4 and 9.5. Then at some
point in the process, B would have been chosen because it would be the ball of radius
larger than k which has the smallest index. Therefore, B G and this shows G is
maximal with respect to 9.4 and 9.5.
For an open ball, B = B(x, r) , denote by

B the open ball, B(x, 4r) .
Lemma 9.6.2 Let F be a collection of open balls, and let
A {B : B F} .
Suppose
> M sup {r : B(p, r) F} > 0.
Then there exists G F such that G consists of disjoint balls and
A {
B : B G}.
Proof: Without loss of generality assume F is countable. This is because there is
a countable subset of F, F
such that F
= A. To see this, consider the set of balls

having rational radii and centers having all components rational. This is a countable
set of balls and you should verify that every open set is the union of balls of this form.
Therefore, you can consider the subset of this set of balls consisting of those which are
contained in some open set of F, G so G = A and use the axiom of choice to dene a
subset of F consisting of a single set from F containing each set of G. Then this is F
.
The union of these sets equals A . Then consider F
instead of F. Therefore, assume

at the outset F is countable.
By Lemma 9.6.1, there exists G
1
F which satises 9.4, 9.5, and 9.6 with k =
2M
3
.
Suppose G
1
, , G
m1
have been chosen for m 2. Let
F
m
= {B F : B R
p
\
union of the balls in these Gj
..
{G
1
G
m1
} }
and using Lemma 9.6.1, let G
m
be a maximal collection of disjoint balls from F
m
with
the property that each ball has radius larger than
_
2
3
_
m
M. Let G
k=1
G
k
. Let
x B(p, r) F. Choose m such that
_
2
3
_
m
M < r
_
2
3
_
m1
M
Then B(p, r) must have nonempty intersection with some ball fromG
1
G
m
because
if it didnt, then G
m
would fail to be maximal. Denote by B(p
0
, r
0
) a ball in G
1
G
m
which has nonempty intersection with B(p, r) . Thus
r
0
>
_
2
3
_
m
M.
Consider the picture, in which w B(p
0
, r
0
) B(p, r) .
9.7. VITALI COVERINGS 227
w '
r
0
p
0
c
r
p
x
Then
|x p
0
| |x p| +|p w| +
<r0
..
|wp
0
|
< r +r +r
0
2
<
3
2
r0
..
_
2
3
_
m1
M +r
0
< 2
_
3
2
_
r
0
+r
0
= 4r
0
.
This proves the lemma since it shows B(p, r) B(p
0
, 4r
0
).
With this Lemma consider a version of the Vitali covering theorem in which the
balls do not have to be open. In this theorem, B will denote an open ball, B(x, r) along
with either part or all of the points where ||x|| = r and |||| is any norm for R
p
.
Denition 9.6.3 Let B be a ball centered at x having radius r. Denote by

B
the open ball, B(x, 5r).
Theorem 9.6.4 (Vitali) Let F be a collection of balls, and let
A {B : B F} .
Suppose
> M sup {r : B(p, r) F} > 0.
Then there exists G F such that G consists of disjoint balls and
A {
B : B G}.
Proof: For B one of these balls, say B(x, r) B B(x, r), denote by B
1
, the
open ball B
_
x,
5r
4
_
. Let F
1
{B
1
: B F} and let A
1
denote the union of the balls in
F
1
. Apply Lemma 9.6.2 to F
1
to obtain
A
1
{
B
1
: B
1
G
1
}
where G
1
consists of disjoint balls from F
1
. Now let G {B F : B
1
G
1
}. Thus G
consists of disjoint balls from F because they are contained in the disjoint open balls,
G
1
. Then
A A
1
{
B
1
: B
1
G
1
} = {
B : B G}
because for B
1
= B
_
x,
5r
4
_
, it follows

B
1
= B(x, 5r) =

B. This proves the theorem.
9.7 Vitali Coverings
There is another version of the Vitali covering theorem which is also of great importance.
In this one, disjoint balls from the original set of balls almost cover the set, leaving out
only a set of measure zero. It is like packing a truck with stu. You keep trying to ll
in the holes with smaller and smaller things so as to not waste space. It is remarkable
that you can avoid wasting any space at all when you are dealing with balls of any sort
provided you can use arbitrarily small balls.
Denition 9.7.1 Let F be a collection of balls that cover a set, E, which have
the property that if x E and > 0, then there exists B F, diameter of B < and
x B. Such a collection covers E in the sense of Vitali.
In the following covering theorem, m
p
denotes the outer measure determined by p
dimensional Lebesgue measure. Thus, letting F denote the Lebesgue measurable sets,
m
p
(S) inf
_

k=1
m
p
(E
k
) : S
k
E
k
, E
k
F
_
Recall that from this denition, if S R
p
there exists E
1
S such that m
p
(E
1
) =
m
p
(S). To see this, note that it suces to assume in the above denition of m
p
that
the E
k
are also disjoint. If not, replace with the sequence given by
F
1
= E
1
, F
2
E
2
\ F
1
, , F
m
E
m
\ F
m1
,
etc. Then for each l > m
p
(S) , there exists {E
k
} such that
l >
k
m
p
(E
k
)
k
m
p
(F
k
) = m
p
(
k
E
k
) m
p
(S) .
If m
p
(S) = , let E
1
= R
p
. Otherwise, there exists G
k
F such that
m
p
(S) m
p
(G
k
) m
p
(S) + 1/k.
then let E
1
=
k
G
k
.
Note this implies that if m
p
(S) = 0 then S must be in F because of completeness
of Lebesgue measure.
Theorem 9.7.2 Let E R
p
and suppose 0 < m
p
(E) < where m
p
is the
outer measure determined by m
p
, p dimensional Lebesgue measure, and let F be a
collection of closed balls of bounded radii such that F covers E in the sense of Vitali.
Then there exists a countable collection of disjoint balls from F, {B
j
}
j=1
, such that
m
p
(E \
j=1
B
j
) = 0.
Proof: From the denition of outer measure there exists a Lebesgue measurable set,
E
1
E such that m
p
(E
1
) = m
p
(E). Now by outer regularity of Lebesgue measure,
there exists U, an open set which satises
m
p
(E
1
) > (1 10
p
)m
p
(U), U E
1
.
E
1
U
Each point of E is contained in balls of F of arbitrarily small radii and so there
exists a covering of E with balls of F which are themselves contained in U. Therefore,
by the Vitali covering theorem, there exist disjoint balls, {B
i
}
i=1
F such that
E
j=1
B
j
, B
j
U.
9.7. VITALI COVERINGS 229
Therefore,
m
p
(E
1
) = m
p
(E) m
p
_
j=1
B
j
_
j
m
p
_
B
j
_
= 5
p
j
m
p
(B
j
) = 5
p
m
p
_
j=1
B
j
_
Then
m
p
(E
1
) > (1 10
p
)m
p
(U)
(1 10
p
)[m
p
(E
1
\
j=1
B
j
) +m
p
(
j=1
B
j
)]
(1 10
p
)[m
p
(E
1
\
j=1
B
j
) + 5
p
=mp(E1)
..
m
p
(E) ].
and so
_
1
_
1 10
p
_
5
p
_
m
p
(E
1
) (1 10
p
)m
p
(E
1
\
j=1
B
j
)
which implies
m
p
(E
1
\
j=1
B
j
)
(1 (1 10
p
) 5
p
)
(1 10
p
)
m
p
(E
1
)
Now a short computation shows
0 <
(1 (1 10
p
) 5
p
)
(1 10
p
)
< 1
Hence, denoting by
p
a number such that
(1 (1 10
p
) 5
p
)
(1 10
p
)
<
p
< 1,
m
p
_
E \
j=1
B
j
_
m
p
(E
1
\
j=1
B
j
) <
p
m
p
(E
1
) =
p
m
p
(E)
Now using Theorem 7.3.2 on Page 157 there exists N
1
large enough that
p
m
p
(E) m
p
(E
1
\
N1
j=1
B
j
) m
p
(E \
N1
j=1
B
j
) (9.7)
Let F
1
= {B F : B
j
B = , j = 1, , N
1
}. If E \
N1
j=1
B
j
= , then F
1
=
and
m
p
_
E \
N1
j=1
B
j
_
= 0
Therefore, in this case let B
k
= for all k > N
1
. Consider the case where
E \
N1
j=1
B
j
= .
In this case, since the balls are closed and F is a Vitali cover, F
1
= and covers
E \
N1
j=1
B
j
in the sense of Vitali. Repeat the same argument, letting E \
N1
j=1
B
j
play
the role of E. (You pick a dierent E
1
whose measure equals the outer measure of
E \
N1
j=1
B
j
and proceed as before.) Then choosing B
j
for j = N
1
+1, , N
2
as in the
above argument,
p
m
p
(E \
N1
j=1
B
j
) m
p
(E \
N2
j=1
B
j
)
and so from 9.7,
2
p
m
p
(E) m
p
(E \
N2
j=1
B
j
).
Continuing this way
k
p
m
p
(E) m
p
_
E \
N
k
j=1
B
j
_
.
If it is ever the case that E \
N
k
j=1
B
j
= , then as in the above argument,
m
p
_
E \
N
k
j=1
B
j
_
= 0.
Otherwise, the process continues and
m
p
_
E \
j=1
B
j
_
m
p
_
E \
N
k
j=1
B
j
_

k
p
m
p
(E)
for every k N. Therefore, the conclusion holds in this case also because
p
< 1. This
proves the theorem.
There is an obvious corollary which removes the assumption that 0 < m
p
(E).
Corollary 9.7.3 Let E R
p
and suppose m
p
(E) < where m
p
is the outer mea-
sure determined by m
p
, p dimensional Lebesgue measure, and let F, be a collection of
closed balls of bounded radii such thatF covers E in the sense of Vitali. Then there exists
a countable collection of disjoint balls from F, {B
j
}
j=1
, such that m
p
(E\
j=1
B
j
) = 0.
Proof: If 0 = m
p
(E) you simply pick any ball from F for your collection of disjoint
balls.
It is also not hard to remove the assumption that m
p
(E) < .
p
and let F, be a collection of closed balls of bounded
radii such that F covers E in the sense of Vitali. Then there exists a countable collection
of disjoint balls from F, {B
j
}
j=1
, such that m
p
(E \
j=1
B
j
) = 0.
Proof: Let R
m
(m, m)
p
be the open rectangle having sides of length 2m which
is centered at 0 and let R
0
= . Let H
m
R
m
\ R
m
. Since both R
m
and R
m
have the
same measure, (2m)
p
, it follows m
p
(H
m
) = 0. Now for all k N, R
k
R
k
R
k+1
.
Consider the disjoint open sets U
k
R
k+1
\ R
k
. Thus R
p
=
k=0
U
k
N where N is a
set of measure zero equal to the union of the H
k
. Let F
k
denote those balls of F which
are contained in U
k
and let E
k
U
k
E. Then from Theorem 9.7.2, there exists a
sequence of disjoint balls, D
k

_
B
k
i
_
i=1
of F
k
such that m
p
(E
k
\
j=1
B
k
j
) = 0. Letting
{B
i
}
i=1
be an enumeration of all the balls of
k
D
k
, it follows that
m
p
(E \
j=1
B
j
) m
p
(N) +
k=1
m
p
(E
k
\
j=1
B
k
j
) = 0.
Also, you dont have to assume the balls are closed.

p
and let F, be a collection of open balls of bounded
radii such that F covers E in the sense of Vitali. Then there exists a countable collection
of disjoint balls from F, {B
j
}
j=1
, such that m
p
(E \
j=1
B
j
) = 0.
Proof: Let F be the collection of closures of balls in F. Then F covers E in the
sense of Vitali and so from Corollary 9.7.4 there exists a sequence of disjoint closed balls
from F satisfying m
p
_
E \
i=1
B
i
_
= 0. Now boundaries of the balls, B
i
have measure
zero and so {B
i
} is a sequence of disjoint open balls satisfying m
p
(E \
i=1
B
i
) = 0.
The reason for this is that
(E \
i=1
B
i
) \
_
E \
i=1
B
i
_

i=1
B
i
\
i=1
B
i

i=1
B
i
\ B
i
,
a set of measure zero. Therefore,
E \
i=1
B
i

_
E \
i=1
B
i
_
i=1
B
i
\ B
i
_
9.8. CHANGE OF VARIABLES FOR LINEAR MAPS 231
and so
m
p
(E \
i=1
B
i
) m
p
_
E \
i=1
B
i
_
+m
p
_
i=1
B
i
\ B
i
_
= m
p
_
E \
i=1
B
i
_
= 0.
This implies you can ll up an open set with balls which cover the open set in the
sense of Vitali.
Corollary 9.7.6 Let U R
p
be an open set and let F be a collection of closed or
even open balls of bounded radii contained in U such that F covers U in the sense of
Vitali. Then there exists a countable collection of disjoint balls from F, {B
j
}
j=1
, such
that m
p
(U \
j=1
B
j
) = 0.
9.8 Change Of Variables For Linear Maps
To begin with certain kinds of functions map measurable sets to measurable sets. It
will be assumed that U is an open set in R
p
and that h : U R
p
satises
Dh(x) exists for all x U, (9.8)
Note that if
h(x) = Lx
where L L(R
p
, R
p
) , then L is included in 9.8 because
L(x +v) = L(x) +L(v) +o(v)
In fact, o(v) = 0.
It is convenient in the following lemma to use the norm on R
p
given by
||x|| = max {|x
k
| : k = 1, 2, , p} .
Thus B(x, r) is the open box,
p
k=1
(x
k
r, x
k
+r)
and so m
p
(B(x, r)) = (2r)
p
.
Lemma 9.8.1 Let h satisfy 9.8. If T U and m
p
(T) = 0, then m
p
(h(T)) = 0.
Proof: Let
T
k
{x T : ||Dh(x)|| < k}
and let > 0 be given. Now by outer regularity, there exists an open set V , containing
T
k
which is contained in U such that m
p
(V ) < . Let x T
k
. Then by dierentiability,
h(x +v) = h(x) +Dh(x) v +o (v)
and so there exist arbitrarily small r
x
< 1 such that B(x,5r
x
) V and whenever
||v|| 5r
x
, ||o (v)|| < k ||v|| . Thus
h(B(x, 5r
x
)) B(h(x) , 6kr
x
) .
From the Vitali covering theorem, there exists a countable disjoint sequence of these
balls, {B(x
i
, r
i
)}
i=1
such that {B(x
i
, 5r
i
)}
i=1
=
_
B
i
_
i=1
covers T
k
Then letting m
p
denote the outer measure determined by m
p
,
m
p
(h(T
k
)) m
p
_
h
_
i=1
B
i
__
i=1
m
p
_
h
_
B
i
__
i=1
m
p
(B(h(x
i
) , 6kr
xi
))
=
i=1
m
p
(B(x
i
, 6kr
xi
)) = (6k)
p
i=1
m
p
(B(x
i
, r
xi
))
(6k)
p
m
p
(V ) (6k)
p
.
Since > 0 is arbitrary, this shows m
p
(h(T
k
)) = 0. Now
m
p
(h(T)) = lim
k
m
p
(h(T
k
)) = 0.
Lemma 9.8.2 Let h satisfy 9.8. If S is a Lebesgue measurable subset of U, then

h(S) is Lebesgue measurable.
Proof: By Theorem 9.4.2 there exists F which is a countable union of compact sets
F =
k=1
K
k
such that
F S, m
p
(S \ F) = 0.
Then since h is continuous
h(F) =
k
h(K
k
) B (R
p
)
because the continuous image of a compact set is compact. Also, h(S \ F) is a set of
measure zero by Lemma 9.8.1 and so
h(S) = h(F) h(S \ F) F
p
because it is the union of two sets which are in F
p
In particular, this proves the following corollary.
Corollary 9.8.3 Suppose A L(R
p
, R
p
). Then if S is a Lebesgue measurable set,
it follows AS is also a Lebesgue measurable set.
In the next lemma, the norm used for dening balls will be the usual norm,
|x| =
_
p
k=1
|x
k
|
2
_
1/2
.
Thus a unitary transformation preserves distances measured with respect to this norm.
In particular, if R is unitary, (R
R = RR
= I) then
R(B(0, r)) = B(0, r) .
Lemma 9.8.4 Let R be unitary and let V be a an open set. Then m
p
(RV ) =
m
p
(V ) .
Proof: First assume V is a bounded open set. By Corollary 9.7.6 there is a disjoint
sequence of closed balls, {B
i
} such that U =
i=1
B
i
N where m
p
(N) = 0. Denote by
x
i
the center of B
i
and let r
i
be the radius of B
i
. Then by Lemma 9.8.1 m
p
(RV ) =
i=1
m
p
(RB
i
) . Now by invariance of translation of Lebesgue measure, this equals
i=1
m
p
(RB
i
Rx
i
) =
i=1
m
p
(RB(0, r
i
)) .
Since R is unitary, it preserves all distances and so RB(0, r
i
) = B(0, r
i
) and therefore,
m
p
(RV ) =
i=1
m
p
(B(0, r
i
)) =
i=1
m
p
(B
i
) = m
p
(V ) .
This proves the lemma in the case that V is bounded. Suppose now that V is just an
open set. Let V
k
= V B(0, k) . Then m
p
(RV
k
) = m
p
(V
k
) . Letting k , this yields
the desired conclusion. This proves the lemma in the case that V is open.
Lemma 9.8.5 Let E be Lebesgue measurable set in R
p
and let R be unitary. Then
m
p
(RE) = m
p
(E) .
Proof: Let K be the open sets. Thus K is a system. Let G denote those Borel
sets F such that for each n N,
m
p
(R(F (n, n)
p
)) = m
n
(F (n, n)
p
) .
Thus G contains K from Lemma 9.8.4. It is also routine to verify G is closed with respect
to complements and countable disjoint unions. Therefore from the systems lemma,
G (K) = B (R
p
) G
and this proves the lemma whenever E B (R
p
). If E is only in F
p
, it follows from
Theorem 9.4.2
E = F N
where m
p
(N) = 0 and F is a countable union of compact sets. Thus by Lemma 9.8.1
m
p
(RE) = m
p
(RF) +m
p
(RN) = m
p
(RF) = m
p
(F) = m
p
(E) .
Lemma 9.8.6 Let D L(R
p
, R
p
) be of the form
D =
j
d
j
e
j
e
j
where d
j
0 and {e
j
} is the usual orthonormal basis of R
p
. Then for all E F
p
m
p
(DE) = |det (D)| m
p
(E) .
Proof: Let K consist of open sets of the form
p
k=1
(a
k
, b
k
)
_
p
k=1
x
k
e
k
such that x
k
(a
k
, b
k
)
_
Hence
D
_
p
k=1
(a
k
, b
k
)
_
=
_
p
k=1
d
k
x
k
e
k
such that x
k
(a
k
, b
k
)
_
=
p
k=1
(d
k
a
k
, d
k
b
k
) .
It follows
m
p
_
D
_
p
k=1
(a
k
, b
k
)
__
=
_
p
k=1
d
k
__
p
k=1
(b
k
a
k
)
_
= |det (D)| m
p
_
p
k=1
(a
k
, b
k
)
_
.
Now let G consist of Borel sets F with the property that
m
p
(D(F (n, n)
p
)) = |det (D)| m
p
(F (n, n)
p
) .
Thus K G.
Suppose now that F G and rst assume D is one to one. Then
m
p
_
D
_
F
C
(n, n)
p
__
+m
p
(D(F (n, n)
p
)) = m
p
(D(n, n)
p
)
and so
m
p
_
D
_
F
C
(n, n)
p
__
+|det (D)| m
p
(F (n, n)
p
) = |det (D)| m
p
((n, n)
p
)
which shows
m
p
_
D
_
F
C
(n, n)
p
__
= |det (D)| [m
p
((n, n)
p
) m
p
(F (n, n)
p
)]
= |det (D)| m
p
_
F
C
(n, n)
p
_
In case D is not one to one, it follows some d
j
= 0 and so |det (D)| = 0 and
0 m
p
_
D
_
F
C
(n, n)
p
__
m
p
(D(n, n)
p
) =
p
i=1
(d
i
p +d
i
p) = 0
= |det (D)| m
p
_
F
C
(n, n)
p
_
so F
C
G.
If {F
k
} is a sequence of disjoint sets of G and D is one to one
m
p
(D(
k=1
F
k
(n, n)
p
)) =
k=1
m
p
(D(F
k
(n, n)
p
))
= |det (D)|
k=1
m
p
(F
k
(n, n)
p
)
= |det (D)| m
p
(
k
F
k
(n, n)
p
) .
If D is not one to one, then det (D) = 0 and so the right side of the above equals 0.
The left side is also equal to zero because it is no larger than
m
p
(D(n, n)
p
) = 0.
Thus G is closed with respect to complements and countable disjoint unions. Hence it
contains (K) , the Borel sets. But also G B (R
p
) and so G equals B (R
p
) . Letting
p yields the conclusion of the lemma in case E B (R
p
).
Now for E F
p
arbitrary, it follows from Theorem 9.4.2
E = F N
where N is a set of measure zero and F is a countable union of compact sets. Hence as
before,
m
p
(D(E)) = m
p
(DF DN) m
p
(DF) +m
p
(DN)
= |det (D)| m
p
(F) = |det (D)| m
p
(E)
Also from Theorem 9.4.2 there exists G Borel such that
G = E S
where S is a set of measure zero. Therefore,
|det (D)| m
p
(E) = |det (D)| m
p
(G) = m
p
(DG)
= m
p
(DE DS) m
p
(DE) +m
p
(DS)
= m
p
(DE)
The main result follows.
Theorem 9.8.7 Let E F
p
and let A L(R
p
, R
p
). Then
m
p
(AV ) = |det (A)| m
p
(V ) .
Proof: Let RU be the right polar decomposition (Theorem 3.9.3 on Page 66) of A.
Thus R is unitary and
U =
k
d
k
w
k
w
k
where each d
k
0. It follows |det (A)| = |det (U)| because
|det (A)| = |det (R) det (U)| = |det (R)| |det (U)| = |det (U)| .
Recall from Lemma 3.9.5 on Page 68 the determinant of a unitary transformation has
absolute value equal to 1. Then from Lemma 9.8.5,
m
p
(AE) = m
p
(RUE) = m
p
(UE) .
Let
Q =
j
w
j
e
j
and so by Lemma 3.8.21 on Page 63,
Q
k
e
k
w
k
.
Thus Q and Q
are both unitary and a simple computation shows

U = Q
i
d
i
e
i
e
i
Q
QDQ
.
Do both sides to w
k
and observe both sides give d
k
w
k
. Since the two linear operators
agree on a basis, they must be the same. Thus
|det (D)| = |det (U)| = |det (A)| .
Therefore, from Lemma 9.8.5 and Lemma 9.8.6
m
p
(AE) = m
p
(QDQ
E) = m
p
(DQ
E)
= |det (D)| m
p
(Q
E) = |det (A)| m
p
(E) .
9.9 Change Of Variables For C
1
Functions
In this section theorems are proved which yield change of variables formulas for C
1
functions. More general versions can be seen in Kuttler [27], Kuttler [28], and Rudin
[35]. You can obtain more by exploiting the Radon Nikodym theorem and the Lebesgue
fundamental theorem of calculus, two topics which are best studied in a real analysis
course. Instead, I will present some good theorems using the Vitali covering theorem
directly.
A basic version of the theorems to be presented is the following. If you like, let the
balls be dened in terms of the norm
||x|| max {|x
k
| : k = 1, , p}
Lemma 9.9.1 Let U and V be bounded open sets in R
p
and let h, h
1
be C
1
functions
such that h(U) = V . Also let f C
c
(V ) . Then
V
f (y) dm
p
=
U
f (h(x)) |det (Dh(x))| dm
p
Proof: First note h
1
(spt (f)) is a closed subset of the bounded set, U and so it is
compact. Thus x f (h(x)) |det (Dh(x))| is bounded and continuous.
Let x U. By the assumption that h and h
1
are C
1
,
h(x +v) h(x) = Dh(x) v +o(v)
= Dh(x)
_
v +Dh
1
(h(x)) o(v)
_
= Dh(x) (v +o(v))
and so if r > 0 is small enough then B(x, r) is contained in U and
h(B(x, r)) h(x) =
h(x+B(0,r)) h(x) Dh(x) (B(0, (1 +) r)) . (9.9)
Making r still smaller if necessary, one can also obtain
|f (y) f (h(x))| < (9.10)
for any y h(B(x, r)) and also
|f (h(x
1
)) |det (Dh(x
1
))| f (h(x)) |det (Dh(x))|| < (9.11)
whenever x
1
B(x, r) . The collection of such balls is a Vitali cover of U. By Corollary
9.7.6 there is a sequence of disjoint closed balls {B
i
} such that U =
i=1
B
i
N where
9.9. CHANGE OF VARIABLES FOR C
1
FUNCTIONS 237
m
p
(N) = 0. Denote by x
i
the center of B
i
and r
i
the radius. Then by Lemma 9.8.1,
the monotone convergence theorem, and 9.9 - 9.11,
V
f (y) dm
p
=
i=1
h(Bi)
f (y) dm
p
m
p
(V ) +
i=1
h(Bi)
f (h(x
i
)) dm
p
m
p
(V ) +
i=1
f (h(x
i
)) m
p
(h(B
i
))
m
p
(V ) +
i=1
f (h(x
i
)) m
p
(Dh(x
i
) (B(0, (1 +) r
i
)))
= m
p
(V ) + (1 +)
p
i=1
Bi
f (h(x
i
)) |det (Dh(x
i
))| dm
p
m
p
(V ) + (1 +)
p
i=1
_
Bi
p
+m
p
(B
i
)
_
m
p
(V ) + (1 +)
p
i=1
Bi
p
+ (1 +)
p
m
p
(U)
= m
p
(V ) + (1 +)
p
U
p
+ (1 +)
p
m
p
(U)
Since > 0 is arbitrary, this shows
V
f (y) dm
p

U
p
(9.12)
whenever f C
c
(V ) . Now x f (h(x)) |det (Dh(x))| is in C
c
(U) and so using the
same argument with U and V switching roles and replacing h with h
1
,
U
p
V
f
_
h
_
h
1
(y)
__
det
_
Dh
_
h
1
(y)
__
det
_
Dh
1
(y)
_
dm
p
=
V
f (y) dm
p
by the chain rule. This with 9.12 proves the lemma.
The next task is to relax the assumption that f is continuous.
Corollary 9.9.2 Let U and V be bounded open sets in R
p
and let h, h
1
be C
1
functions such that h(U) = V . Also let E V be measurable. Then
V
X
E
(y) dm
p
=
U
X
E
(h(x)) |det (Dh(x))| dm
p
.
Proof: First suppose E H V where H is compact. By regularity, there exist
compact sets K
k
and a decreasing sequence of open sets G
k
V such that
K
k
E G
k
and m
p
(G
k
\ K
k
) < 2
k
. By Lemma 8.10.2, there exist f
k
such that K
k
f
k
G
k
.
Then f
k
(y) X
E
(y) a.e. because if y is such that convergence fails, it must be the
case that y is in G
k
\ K
k
for innitely many k and
k
m
p
(G
k
\ K
k
) < . This set
equals
N =
m=1
k=m
G
k
\ K
k
and so for each m N
m
p
(N) m
p
(
k=m
G
k
\ K
k
)
k=m
m
p
(G
k
\ K
k
) <
k=m
2
k
= 2
(m1)
showing m
p
(N) = 0.
Then f
k
(h(x)) must converge to X
E
(h(x)) for all x / h
1
(N) , a set of measure
zero by Lemma 9.8.1. Thus
V
f
k
(y) dm
p
=
U
f
k
p
.
Since V is bounded, G
1
is compact. Therefore, |det (Dh(x))| is bounded independent
of k and so, by the dominated convergence theorem, using a dominating function, X
V
in the integral on the left and X
G1
|det (Dh)| on the right, it follows
V
X
E
(y) dm
p
=
U
X
E
p
.
For an arbitrary measurable E, let E
k
= H
k
E replace E in the above with E
k
and
use the monotone convergence theorem letting k .
You dont need to assume the open sets are bounded.
Corollary 9.9.3 Let U and V be open sets in R
p
and let h, h
1
be C
1
functions
such that h(U) = V . Also let E V be measurable. Then
V
X
E
(y) dm
p
=
U
X
E
p
.
Proof: For each x U, there exists r
x
such that B(x, r
x
) U and r
x
< 1. Then by
the mean value inequality Theorem 6.4.2, it follows h(B(x, r
x
)) is also bounded. These
balls, B(x, r
x
) give a Vitali cover of U and so by Corollary 9.7.6 there is a sequence of
these balls, {B
i
} such that they are disjoint, h(B
i
) is bounded and
m
p
(U \
i
B
i
) = 0.
It follows from Lemma 9.8.1 that h(U \
i
B
i
) also has measure zero. Then from Corol-
lary 9.9.2
V
X
E
(y) dm
p
=
h(Bi)
X
Eh(Bi)
(y) dm
p
=
Bi
X
E
p
=
U
X
E
p
.
With this corollary, the main theorem follows.
Theorem 9.9.4 Let U and V be open sets in R
p
and let h, h
1
be C
1
functions
such that h(U) = V. Then if g is a nonnegative Lebesgue measurable function,
V
g (y) dm
p
=
U
g (h(x)) |det (Dh(x))| dm
p
. (9.13)
Proof: From Corollary 9.9.3, 9.13 holds for any nonnegative simple function in place
of g. In general, let {s
k
} be an increasing sequence of simple functions which converges
to g pointwise. Then from the monotone convergence theorem
V
g (y) dm
p
= lim
k
V
s
k
dm
p
= lim
k
U
s
k
p
=
U
p
.
1
FUNCTIONS 239
Of course this theorem implies the following corollary by splitting up the function
into the positive and negative parts of the real and imaginary parts.
Corollary 9.9.5 Let U and V be open sets in R
p
and let h, h
1
be C
1
functions
such that h(U) = V. Let g L
1
(V ) . Then
V
g (y) dm
p
=
U
p
.
This is a pretty good theorem but it isnt too hard to generalize it. In particular, it
is not necessary to assume h
1
is C
1
.
Lemma 9.9.6 Suppose V is an p1 dimensional subspace of R
p
and K is a compact
subset of V . Then letting
K

xK
B(x,) = K +B(0, ) ,
it follows that
m
p
(K
) 2
p
(diam(K) +)
p1
.
Proof: Using the Gram Schmidt procedure, there exists an orthonormal basis for
V , {v
1
, , v
p1
} and let
{v
1
, , v
p1
, v
p
}
be an orthonormal basis for R
p
. Now dene a linear transformation, Q by Qv
i
= e
i
.
Thus QQ
= Q
Q = I and Q preserves all distances and is a unitary transformation

because
i
a
i
e
i
2
=
i
a
i
v
i
2
=
i
|a
i
|
2
=
i
a
i
e
i
2
.
Thus m
p
(K
) = m
p
(QK
). Letting k
0
K, it follows K B(k
0
, diam(K)) and so,
QK B
p1
(Qk
0
, diam(QK)) = B
p1
(Qk
0
, diam(K))
where B
p1
refers to the ball taken with respect to the usual norm in R
p1
. Every
point of K
is within of some point of K and so it follows that every point of QK
is
within of some point of QK. Therefore,
QK
B
p1
(Qk
0
, diam(QK) +) (, ) ,
To see this, let x QK
. Then there exists k QK such that |k x| < . Therefore,

|(x
1
, , x
p1
) (k
1
, , k
p1
)| < and |x
p
k
p
| < and so x is contained in the set
on the right in the above inclusion because k
p
= 0. However, the measure of the set on
the right is smaller than
[2 (diam(QK) +)]
p1
(2) = 2
p
[(diam(K) +)]
p1
.
Note this is a very sloppy estimate. You can certainly do much better but this
estimate is sucient to prove Sards lemma which follows.
Denition 9.9.7 If T, S are two nonempty sets in a normed vector space,
dist (S, T) inf {||s t|| : s S, t T} .
Lemma 9.9.8 Let h be a C
1
function dened on an open set, U R
p
and let K be
a compact subset of U. Then if > 0 is given, there exists r
1
> 0 such that if |v| r
1
,
then for all x K,
|h(x +v) h(x) Dh(x) v| < |v| .
Proof: Let 0 < < dist
_
K, U
C
_
. Such a positive number exists because if there
exists a sequence of points in K, {k
k
} and points in U
C
, {s
k
} such that |k
k
s
k
| 0,
then you could take a subsequence, still denoted by k such that k
k
k K and then
s
k
k also. But U
C
is closed so k K U
C
, a contradiction. Then for |v| < it
follows that for every x K,
x+tv U
and
|h(x +v) h(x) Dh(x) v|
|v|

1
0
Dh(x +tv) vdt Dh(x) v
|v|
1
0
|Dh(x +tv) v Dh(x) v| dt
|v|
.
The integral in the above involves integrating componentwise. Thus t Dh(x +tv) v
is a function having values in R
p
_
_
_
Dh
1
(x+tv) v
.
.
.
Dh
p
(x+tv) v
_
_
_
and the integral is dened by
_
_
_
1
0
Dh
1
(x+tv) vdt
.
.
.
1
0
Dh
p
(x+tv) vdt
_
_
_
Now from uniform continuity of Dh on the compact set, {x : dist (x,K) } it follows
there exists r
1
< such that if |v| r
1
, then ||Dh(x +tv) Dh(x)|| < for every
x K. From the above formula, it follows that if |v| r
1
,
|h(x +v) h(x) Dh(x) v|
|v|

1
0
|Dh(x +tv) v Dh(x) v| dt
|v|
<
1
0
|v| dt
|v|
= .
The following is Sards lemma. In the proof, it does not matter which norm you use
in dening balls.
Lemma 9.9.9 (Sard) Let U be an open set in R
p
and let h : U R
p
be C
1
. Let
Z {x U : det Dh(x) = 0} .
Then m
p
(h(Z)) = 0.
1
FUNCTIONS 241
Proof: Let {U
k
}
k=1
be an increasing sequence of open sets whose closures are com-
pact and whose union equals U and let Z
k
Z U
k
. To obtain such a sequence,
let
U
k
=
_
x U : dist
_
x,U
C
_
<
1
k
_
B(0, k) .
First it is shown that h(Z
k
) has measure zero. Let W be an open set contained in U
k+1
which contains Z
k
and satises
m
p
(Z
k
) + > m
p
(W)
where here and elsewhere, < 1. Let
r = dist
_
U
k
, U
C
k+1
_
and let r
1
> 0 be a constant as in Lemma 9.9.8 such that whenever x U
k
and
0 < |v| r
1
,
|h(x +v) h(x) Dh(x) v| < |v| . (9.14)
Now the closures of balls which are contained in W and which have the property that
their diameters are less than r
1
yield a Vitali covering of W. Therefore, by Corollary
9.7.6 there is a disjoint sequence of these closed balls,
_
B
i
_
such that
W =
i=1
B
i
N
where N is a set of measure zero. Denote by {B
i
} those closed balls in this sequence
which have nonempty intersection with Z
k
, let d
i
be the diameter of B
i
, and let z
i
be a
point in B
i
Z
k
. Since z
i
Z
k
, it follows Dh(z
i
) B(0,d
i
) = D
i
where D
i
is contained
in a subspace, V which has dimension p 1 and the diameter of D
i
is no larger than
2C
k
d
i
where
C
k
max {||Dh(x)|| : x Z
k
}
Then by 9.14, if z B
i
,
h(z) h(z
i
) D
i
+B(0, d
i
) D
i
+B(0,d
i
) .
Thus
h(B
i
) h(z
i
) +D
i
+B(0,d
i
)
By Lemma 9.9.6
m
p
(h(B
i
)) 2
p
(2C
k
d
i
+d
i
)
p1
d
i
d
p
i
_
2
p
[2C
k
+]
p1
_
C
p,k
m
p
(B
i
) .
Therefore, by Lemma 9.8.1
m
p
(h(Z
k
)) m
p
(W) =
i
m
p
(h(B
i
)) C
p,k
i
m
p
(B
i
)
C
p,k
m
p
(W) C
p,k
(m
p
(Z
k
) +)
Since is arbitrary, this shows m
p
(h(Z
k
)) = 0 and so 0 = lim
k
m
p
(h(Z
k
)) =
m
p
(h(Z)).
With this important lemma, here is a generalization of Theorem 9.9.4.
Theorem 9.9.10 Let U be an open set and let h be a 1 1, C
1
(U) function
with values in R
p
. Then if g is a nonnegative Lebesgue measurable function,
h(U)
g (y) dm
p
=
U
p
. (9.15)
Proof: Let Z = {x : det (Dh(x)) = 0} , a closed set. Then by the inverse function
theorem, h
1
is C
1
on h(U \ Z) and h(U \ Z) is an open set. Therefore, from Lemma
9.9.9, h(Z) has measure zero and so by Theorem 9.9.4,
h(U)
g (y) dm
p
=
h(U\Z)
g (y) dm
p
=
U\Z
p
=
U
p
.
Of course the next generalization considers the case when h is not even one to one.
9.10 Change Of Variables For Mappings Which Are
Not One To One
Now suppose h is only C
1
, not necessarily one to one. For
U
+
{x U : |det Dh(x)| > 0}
and Z the set where |det Dh(x)| = 0, Lemma 9.9.9 implies m
p
(h(Z)) = 0. For x U
+
,
the inverse function theorem implies there exists an open set B
x
U
+
, such that h is
one to one on B
x
.
Let {B
i
} be a countable subset of {B
x
}
xU+
such that U
+
=
i=1
B
i
. Let E
1
= B
1
.
If E
1
, , E
k
have been chosen, E
k+1
= B
k+1
\
k
i=1
E
i
. Thus
i=1
E
i
= U
+
, h is one to one on E
i
, E
i
E
j
= ,
and each E
i
is a Borel set contained in the open set B
i
. Now dene
n(y)
i=1
X
h(Ei)
(y) +X
h(Z)
(y).
The set, h(E
i
) , h(Z) are measurable by Lemma 9.8.2. Thus n() is measurable.
Lemma 9.10.1 Let F h(U) be measurable. Then
h(U)
n(y)X
F
(y)dm
p
=
U
X
F
(h(x))| det Dh(x)|dm
p
.
9.11. SPHERICAL COORDINATES IN P DIMENSIONS 243
Proof: Using Lemma 9.9.9 and the Monotone Convergence Theorem
h(U)
n(y)X
F
(y)dm
p
=
h(U)
_
_
_
i=1
X
h(Ei)
(y) +
mp(h(Z))=0
..
X
h(Z)
(y)
_
_
_X
F
(y)dm
p
=
i=1
h(U)
X
h(Ei)
(y)X
F
(y)dm
p
=
i=1
h(Bi)
X
h(Ei)
(y)X
F
(y)dm
p
=
i=1
Bi
X
Ei
(x)X
F
p
=
i=1
U
X
Ei
(x)X
F
p
=
i=1
X
Ei
(x)X
F
p
=
U+
X
F
p
=
U
X
F
p
.
Denition 9.10.2 For y h(U), dene a function, #, according to the for-
mula
#(y) number of elements in h
1
(y).
Observe that
#(y) = n(y) a.e. (9.16)
because n(y) = #(y) if y / h(Z), a set of measure 0. Therefore, # is a measurable
function because of completeness of Lebesgue measure.
Theorem 9.10.3 Let g 0, g measurable, and let h be C
1
(U). Then
h(U)
#(y)g(y)dm
p
=
U
g(h(x))| det Dh(x)|dm
p
. (9.17)
Proof: From 9.16 and Lemma 9.10.1, 9.17 holds for all g, a nonnegative simple
function. Approximating an arbitrary measurable nonnegative function, g, with an
increasing pointwise convergent sequence of simple functions and using the monotone
convergence theorem, yields 9.17 for an arbitrary nonnegative measurable function, g.
9.11 Spherical Coordinates In p Dimensions
Sometimes there is a need to deal with spherical coordinates in more than three di-
mensions. In this section, this concept is dened and formulas are derived for these
coordinate systems. Recall polar coordinates are of the form
y
1
= cos
y
2
= sin
where > 0 and R. Thus these transformation equations are not one to one but they
are one to one on (0, )[0, 2). Here I am writing in place of r to emphasize a pattern
which is about to emerge. I will consider polar coordinates as spherical coordinates in
two dimensions. I will also simply refer to such coordinate systems as polar coordinates
regardless of the dimension. This is also the reason I am writing y
1
and y
2
instead of
the more usual x and y. Now consider what happens when you go to three dimensions.
The situation is depicted in the following picture.
(x
1
, x
2
, x
3
)
R
2
R
From this picture, you see that y
3
= cos
1
. Also the distance between (y
1
, y
2
) and
(0, 0) is sin (
1
) . Therefore, using polar coordinates to write (y
1
, y
2
) in terms of and
this distance,
y
1
= sin
1
cos ,
y
2
= sin
1
sin ,
y
3
= cos
1
.
where
1
R and the transformations are one to one if
1
is restricted to be in [0, ] .
What was done is to replace with sin
1
and then to add in y
3
= cos
1
. Having
done this, there is no reason to stop with three dimensions. Consider the following
picture:
(x
1
, x
2
, x
3
, x
4
)
R
3
R
From this picture, you see that y
4
= cos
2
. Also the distance between (y
1
, y
2
, y
3
)
and (0, 0, 0) is sin (
2
) . Therefore, using polar coordinates to write (y
1
, y
2
, y
3
) in terms
of ,
1
, and this distance,
y
1
= sin
2
sin
1
cos ,
y
2
= sin
2
sin
1
sin ,
y
3
= sin
2
cos
1
,
y
4
= cos
2
where
2
R and the transformations will be one to one if
2
,
1
[0, ] , [0, 2), (0, ) .
Continuing this way, given spherical coordinates in R
p
, to get the spherical coordi-
nates in R
p+1
, you let y
p+1
= cos
p1
and then replace every occurance of with
sin
p1
to obtain y
1
y
p
in terms of
1
,
2
, ,
p1
,, and .
It is always the case that measures the distance from the point in R
p
to the origin
in R
p
, 0. Each
i
R and the transformations will be one to one if each
i
[0, ] ,
and [0, 2). It can be shown using math induction that these coordinates map
p2
i=1
[0, ] [0, 2) (0, ) one to one onto R
p
\ {0} .
9.11. SPHERICAL COORDINATES IN P DIMENSIONS 245
Theorem 9.11.1 Let y = h(, , ) be the spherical coordinate transformations
in R
p
. Then letting A =
p2
i=1
[0, ] [0, 2), it follows h maps A (0, ) one to one
onto R
p
\ {0} . Also |det Dh(, , )| will always be of the form
|det Dh(, , )| =
p1
(, ) . (9.18)
where is a continuous function of and .
1
Furthermore whenever f is Lebesgue
measurable and nonnegative,
R
p
f (y) dy =
p1
A
f (h(, , )) (, ) ddd (9.19)
where here dd denotes dm
p1
on A. The same formula holds if f L
1
(R
p
) .
Proof: Formula 9.18 is obvious from the denition of the spherical coordinates.
The rst claim is also clear from the denition and math induction. It remains to verify
9.19. Let A
0

p2
i=1
(0, ) (0, 2) . Then it is clear that (A\ A
0
) (0, ) N is a
set of measure zero in R
p
. Therefore, from Lemma 9.8.1 it follows h(N) is also a set of
measure zero. Therefore, using the change of variables theorem, Fubinis theorem, and
Sards lemma,
R
p
f (y) dy =
R
p
\{0}
f (y) dy =
R
p
\({0}h(N))
f (y) dy
=
A0(0,)
f (h(, , ))
p1
(, ) dm
p
=
X
A(0,)
(, , ) f (h(, , ))
p1
(, ) dm
p
=
p1
_
A
f (h(, , )) (, ) dd
_
d.
Now the claim about f L
1
follows routinely from considering the positive and negative
parts of the real and imaginary parts of f in the usual way. This proves the theorem.
Notation 9.11.2 Often this is written dierently. Note that from the spherical coor-
dinate formulas, f (h(, , )) = f () where || = 1. Letting S
p1
denote the unit
sphere, { R
p
: || = 1} , the inside integral in the above formula is sometimes written
as

S
p1
f () d
where is a measure on S
p1
. See [27] for another description of this measure. It isnt
an important issue here. Either 9.19 or the formula
p1
_
S
p1
f () d
_
d
will be referred to as polar coordinates and is very useful in establishing estimates. Here
_
S
p1
_
A
(, ) dd.
Example 9.11.3 For what values of s is the integral
B(0,R)
_
1 +|x|
2
_
s
dy bounded
independent of R? Here B(0, R) is the ball, {x R
p
: |x| R} .
1
Actually it is only a function of the rst but this is not important in what follows.
I think you can see immediately that s must be negative but exactly how negative?
It turns out it depends on p and using polar coordinates, you can nd just exactly what
is needed. From the polar coordinates formula above,
B(0,R)
_
1 +|x|
2
_
s
dy =
R
0
S
p1
_
1 +
2
_
s
p1
dd
= C
p
R
0
_
1 +
2
_
s
p1
d
Now the very hard problem has been reduced to considering an easy one variable problem
of nding when
R
0
p1
_
1 +
2
_
s
d
is bounded independent of R. You need 2s + (p 1) < 1 so you need s < p/2.
9.12 Brouwer Fixed Point Theorem
The Brouwer xed point theorem is one of the most signicant theorems in mathematics.
There exist relatively easy proofs of this important theorem. The proof I am giving here
is the one given in Evans [14]. I think it is one of the shortest and easiest proofs of this
important theorem. It is based on the following lemma which is an interesting result
about cofactors of a matrix.
Recall that for A an p p matrix, cof (A)
ij
is the determinant of the matrix which
results from deleting the i
th
row and the j
th
column and multiplying by (1)
i+j
. In the
proof and in what follows, I am using Dg to equal the matrix of the linear transformation
Dg taken with respect to the usual basis on R
p
. Thus
Dg (x) =
ij
(Dg)
ij
e
i
e
j
and recall that (Dg)
ij
= g
i
/x
j
where g =
i
g
i
e
i
.
Lemma 9.12.1 Let g : U R
p
be C
2
where U is an open subset of R
p
. Then
p
j=1
cof (Dg)
ij,j
= 0,
where here (Dg)
ij
g
i,j

gi
xj
. Also, cof (Dg)
ij
=
det(Dg)
gi,j
.
Proof: From the cofactor expansion theorem,
det (Dg) =
p
i=1
g
i,j
cof (Dg)
ij
and so
det (Dg)
g
i,j
= cof (Dg)
ij
(9.20)
which shows the last claim of the lemma. Also
kj
det (Dg) =
i
g
i,k
(cof (Dg))
ij
(9.21)
9.12. BROUWER FIXED POINT THEOREM 247
because if k = j this is just the cofactor expansion of the determinant of a matrix in
which the k
th
and j
th
columns are equal. Dierentiate 9.21 with respect to x
j
and sum
on j. This yields
r,s,j
kj
(det Dg)
g
r,s
g
r,sj
=
ij
g
i,kj
(cof (Dg))
ij
+
ij
g
i,k
cof (Dg)
ij,j
.
Hence, using
kj
= 0 if j = k and 9.20,
rs
(cof (Dg))
rs
g
r,sk
=
rs
g
r,ks
(cof (Dg))
rs
+
ij
g
i,k
cof (Dg)
ij,j
.
Subtracting the rst sum on the right from both sides and using the equality of mixed
partials,
i
g
i,k
_
_
j
(cof (Dg))
ij,j
_
_
= 0.
If det (g
i,k
) = 0 so that (g
i,k
) is invertible, this shows
j
(cof (Dg))
ij,j
= 0. If
det (Dg) = 0, let
g
k
(x) = g (x) +
k
x
where
k
0 and det (Dg +
k
I) det (Dg
k
) = 0. Then
j
(cof (Dg))
ij,j
= lim
k
j
(cof (Dg
k
))
ij,j
= 0
and This proves the lemma.
Denition 9.12.2 Let h be a function dened on an open set, U R
p
. Then
h C
k
_
U
_
if there exists a function g dened on an open set, W containng U such
that g = h on U and g is C
k
(W) .
In the following lemma, you could use any norm in dening the balls and everything
would work the same but I have in mind the usual norm.
Lemma 9.12.3 There does not exist h C
2
_
B(0, R)
_
such that h :B(0, R)
B(0, R) which also has the property that h(x) = x for all x B(0, R) . Such a
function is called a retraction.
Proof: Suppose such an h exists. Let [0, 1] and let p
(x) x +(h(x) x) .
This function, p
is called a homotopy of the identity map and the retraction, h. Let

I ()
B(0,R)
det (Dp
(x)) dx.
Then using the dominated convergence theorem,
I
() =
B(0,R)
i.j
det (Dp
(x))
p
i,j
p
ij
(x)
dx
=
B(0,R)
j
det (Dp
(x))
p
i,j
(h
i
(x) x
i
)
,j
dx
=
B(0,R)
j
cof (Dp
(x))
ij
(h
i
(x) x
i
)
,j
dx
Now by assumption, h
i
(x) = x
i
on B(0, R) and so one can form iterated integrals
and integrate by parts in each of the one dimensional integrals to obtain
I
() =
B(0,R)
j
cof (Dp
(x))
ij,j
(h
i
(x) x
i
) dx = 0.
Therefore, I () equals a constant. However,
I (0) = m
p
(B(0, R)) > 0
but
I (1) =
B(0,1)
det (Dh(x)) dm
p
=
B(0,1)
#(y) dm
p
= 0
because from polar coordinates or other elementary reasoning, m
p
(B(0, 1)) = 0. This
proves the lemma.
The following is the Brouwer xed point theorem for C
2
maps.
Lemma 9.12.4 If h C
2
_
B(0, R)
_
and h : B(0, R) B(0, R), then h has a
xed point, x such that h(x) = x.
Proof: Suppose the lemma is not true. Then for all x, |x h(x)| = 0. Then dene
g (x) = h(x) +
x h(x)
|x h(x)|
t (x)
where t (x) is nonnegative and is chosen such that g (x) B(0, R) . This mapping is
illustrated in the following picture.
f (x)
x
g(x)
If x t (x) is C
2
near B(0, R), it will follow g is a C
2
retraction onto B(0, R)
contrary to Lemma 9.12.3. Now t (x) is the nonnegative solution, t to
H (x, t) = |h(x)|
2
+ 2
_
h(x) ,
x h(x)
|x h(x)|
_
t +t
2
= R
2
(9.22)
Then
H
t
(x, t) = 2
_
h(x) ,
x h(x)
|x h(x)|
_
+ 2t.
If this is nonzero for all x near B(0, R), it follows from the implicit function theorem
that t is a C
2
function of x. From 9.22
2t = 2
_
h(x) ,
x h(x)
|x h(x)|
_
4
_
h(x) ,
x h(x)
|x h(x)|
_
2
4
_
|h(x)|
2
R
2
_
9.13. EXERCISES 249
and so
H
t
(x, t) = 2t + 2
_
h(x) ,
x h(x)
|x h(x)|
_
=
4
_
R
2
|h(x)|
2
_
+ 4
_
h(x) ,
x h(x)
|x h(x)|
_
2
If |h(x)| < R, this is nonzero. If |h(x)| = R, then it is still nonzero unless
(h(x) , x h(x)) = 0.
But this cannot happen because the angle between h(x) and x h(x) cannot be /2.
Alternatively, if the above equals zero, you would need
(h(x) , x) = |h(x)|
2
= R
2
which cannot happen unless x = h(x) which is assumed not to happen. Therefore,
x t (x) is C
2
near B(0, R) and so g (x) given above contradicts Lemma 9.12.3. This
proves the lemma.
Now it is easy to prove the Brouwer xed point theorem.
Theorem 9.12.5 Let f : B(0, R) B(0, R) be continuous. Then f has a xed
point.
Proof: If this is not so, there exists > 0 such that for all x B(0, R),
|x f (x)| > .
By the Weierstrass approximation theorem, there exists h, a polynomial such that
max
_
|h(x) f (x)| : x B(0, R)
_
<

2
.
Then for all x B(0, R),
|x h(x)| |x f (x)| |h(x) f (x)| >

2
=

2
contradicting Lemma 9.12.4. This proves the theorem.
9.13 Exercises
1. Recall the denition of f
y
. Prove that if f L
1
(R
p
) , then
lim
y0
R
p
|f f
y
| dm
p
= 0
This is known as continuity of translation. Hint: Use the theorem about being
able to approximate an arbitrary function in L
1
(R
p
) with a function in C
c
(R
p
).
2. Show that if a, b 0 and if p, q > 0 such that
1
p
+
1
q
= 1
then
ab
a
p
p
+
b
q
q
Hint: You might consider for xed a 0, the function h(b)
a
p
p
+
b
q
q
ab and
nd its minimum.
3. In the context of the previous problem, prove Holders inequality. If f, g measur-
able functions, then
|f| |g| d
_
|f|
p
d
_
1/p
_
|g|
q
d
_
1/q
Hint: If either of the factors on the right equals 0, explain why there is nothing
to show. Now let a = |f| /
_
|f|
p
d
_
1/p
and b = |g| /
_
|g|
q
d
_
1/q
. Apply the
inequality of the previous problem.
4. Let E be a Lebesgue measurable set in R. Suppose m(E) > 0. Consider the set
E E = {x y : x E, y E}.
Show that E E contains an interval. Hint: Let
f(x) =
X
E
(t)X
E
(x +t)dt.
Explain why f is continuous at 0 and f(0) > 0 and use continuity of translation
in L
1
.
5. If f L
1
(R
p
) , show there exists g L
1
(R
p
) such that g is also Borel measurable
such that g (x) = f (x) for a.e. x.
6. Suppose f, g L
1
(R
p
) . Dene f g (x) by
f (x y) g (y) dm
p
(y) .
Show this makes sense for a.e. x and that in fact for a.e. x
|f (x y)| |g (y)| dm
p
(y)
Next show
|f g (x)| dm
p
(x)
|f| dm
p
|g| dm
p
.
Hint: Use Problem 5. Show rst there is no problem if f, g are Borel measurable.
The reason for this is that you can use Fubinis theorem to write

|f (x y)| |g (y)| dm
p
(y) dm
p
(x)
=

|f (x y)| |g (y)| dm
p
(x) dm
p
(y)
=
|f (z)| dm
p
|g (y)| dm
p
.
Explain. Then explain why if f and g are replaced by functions which are equal
to f and g a.e. but are Borel measurable, the convolution is unchanged.
7. In the situation of Problem 6 Show x f g (x) is continuous whenever g is also
bounded. Hint: Use Problem 1.
8. Let f : [0, ) R be in L
1
(R, m). The Laplace transform is given by

f(x) =
0
e
xt
f(t)dt. Let f, g be in L
1
(R, m), and let h(x) =
x
0
f(x t)g(t)dt. Show
h L
1
, and

h =

f g.
9.13. EXERCISES 251
9. Suppose A is covered by a nite collection of Balls, F. Show that then there exists
a disjoint collection of these balls, {B
i
}
p
i=1
, such that A
p
i=1
B
i
where

B
i
has
the same center as B
i
but 3 times the radius. Hint: Since the collection of balls
is nite, they can be arranged in order of decreasing radius.
10. Let f be a function dened on an interval, (a, b). The Dini derivates are dened
as
D
+
f (x) lim inf
h0+
f (x +h) f (x)
h
,
D
+
f (x) lim sup
h0+
f (x +h) f (x)
h
D
f (x) lim inf

h0+
f (x) f (x h)
h
,
D
f (x) lim sup

h0+
f (x) f (x h)
h
.
Suppose f is continuous on (a, b) and for all x (a, b), D
+
f (x) 0. Show that
then f is increasing on (a, b). Hint: Consider the function, H (x) f (x) (d c)
x(f (d) f (c)) where a < c < d < b. Thus H (c) = H (d). Also it is easy to see
that H cannot be constant if f (d) < f (c) due to the assumption that D
+
f (x) 0.
If there exists x
1
(a, b) where H (x
1
) > H (c), then let x
0
(c, d) be the point
where the maximum of f occurs. Consider D
+
f (x
0
). If, on the other hand,
H (x) < H (c) for all x (c, d), then consider D
+
H (c).
11. Suppose in the situation of the above problem we only know
D
+
f (x) 0 a.e.
Does the conclusion still follow? What if we only know D
+
f (x) 0 for every
x outside a countable set? Hint: In the case of D
+
f (x) 0,consider the bad
function in the exercises for the chapter on the construction of measures which
was based on the Cantor set. In the case where D
+
f (x) 0 for all but countably
many x, by replacing f (x) with

f (x) f (x) + x, consider the situation where
D
+
f (x) > 0 for all but countably many x. If in this situation,

f (c) >

f (d) for
some c < d, and y
_
f (d) ,

f (c)
_
,let
z sup
_
x [c, d] :

f (x) > y
0
_
.
Show that

f (z) = y
0
and D
+
f (z) 0. Conclude that if

f fails to be increasing,
then D
+
f (z) 0 for uncountably many points, z. Now draw a conclusion about

f.
12. Let f : [a, b] R be increasing. Show
m
_
_
_
Npq
..
_
D
+
f (x) > q > p > D
+
f (x)
_
_
_ = 0 (9.23)
and conclude that aside from a set of measure zero, D
+
f (x) = D
+
f (x). Similar
reasoning will show D
f (x) = D
f (x) a.e. and D

+
f (x) = D
f (x) a.e. and so

o some set of measure zero, we have
D
f (x) = D
f (x) = D
+
f (x) = D
+
f (x)
which implies the derivative exists and equals this common value. Hint: To show
9.23, let U be an open set containing N
pq
such that m(N
pq
) + > m(U). For
each x N
pq
there exist y > x arbitrarily close to x such that
f (y) f (x) < p (y x) .
Thus the set of such intervals, {[x, y]} which are contained in U constitutes a
Vitali cover of N
pq
. Let {[x
i
, y
i
]} be disjoint and
m(N
pq
\
i
[x
i
, y
i
]) = 0.
Now let V
i
(x
i
, y
i
). Then also we have
m
_
_
_N
pq
\
=V
..
i
(x
i
, y
i
)
_
_
_ = 0.
and so m(N
pq
V ) = m(N
pq
). For each x N
pq
V , there exist y > x arbitrarily
close to x such that
f (y) f (x) > q (y x) .
Thus the set of such intervals, {[x
, y
]} which are contained in V is a Vitali cover

of N
pq
V . Let {[x
i
, y
i
]} be disjoint and
m(N
pq
V \
i
[x
i
, y
i
]) = 0.
Then verify the following:
i
f (y
i
) f (x
i
) > q
i
(y
i
x
i
) qm(N
pq
V ) = qm(N
pq
)
pm(N
pq
) > p (m(U) ) p
i
(y
i
x
i
) p
i
(f (y
i
) f (x
i
)) p
i
f (y
i
) f (x
i
) p
and therefore, (q p) m(N
pq
) p. Since > 0 is arbitrary, this proves that
there is a right derivative a.e. A similar argument does the other cases.
13. Suppose f is a function in L
1
(R) and f is innitely dierentiable. Does it follow
that f
L
1
(R)? Hint: What if C
c
(0, 1) and f (x) = (2
p
(x p)) for
x (p, p + 1) , f (x) = 0 if x < 0?
14. For a function f L
1
(R
p
), the Fourier transform, Ff is given by
Ff (t)
1
R
p
e
itx
f (x) dx
and the so called inverse Fourier transform, F
1
f is dened by
Ff (t)
1
R
p
e
itx
f (x) dx
Show that if f L
1
(R
p
) , then lim
|x|
Ff (x) = 0. Hint: You might try to
show this rst for f C
c
(R
p
).
15. Prove Lemma 9.8.1 which says a C
1
function maps a set of measure zero to a set
of measure zero using Theorem 9.10.3.
9.13. EXERCISES 253
16. For this problem dene
a
f (t) dt lim
r
r
a
f (t) dt. Note this coincides with
the Lebesgue integral when f L
1
(a, ). Show
(a)
0
sin(u)
u
du =

2
(b) lim
r
sin(ru)
u
du = 0 whenever > 0.
(c) If f L
1
(R), then lim
r
R
sin (ru) f (u) du = 0.
Hint: For the rst two, use
1
u
=
0
e
ut
dt and apply Fubinis theorem to
R
0
sin u
R
e
ut
dtdu. For the last part, rst establish it for f C
c
(R) and
then use the density of this set in L
1
(R) to obtain the result. This is called the
Riemann Lebesgue lemma.
17. Suppose that g L
1
(R) and that at some x > 0, g is locally Holder continuous
from the right and from the left. This means
lim
r0+
g (x +r) g (x+)
exists,
lim
r0+
g (x r) g (x)
exists and there exist constants K, > 0 and r (0, 1] such that for |x y| < ,
|g (x+) g (y)| < K|x y|
r
for y > x and
|g (x) g (y)| < K|x y|
r
for y < x. Show that under these conditions,
lim
r
2

0
sin (ur)
u
_
g (x u) +g (x +u)
2
_
du
=
g (x+) +g (x)
2
.
18. Let g L
1
(R) and suppose g is locally Holder continuous from the right and
from the left at x. Show that then
lim
R
1
2
R
R
e
ixt
e
ity
g (y) dydt =
g (x+) +g (x)
2
.
This is very interesting. This shows F
1
(Fg) (x) =
g(x+)+g(x)
2
, the midpoint of
the jump in g at the point, x provided Fg is in L
1
. Hint: Show the left side of
the above equation reduces to
2

0
sin (ur)
u
_
g (x u) +g (x +u)
2
_
du
and then use Problem 17 to obtain the result.
19. A measurable function g dened on (0, ) has exponential growth if |g (t)|
Ce
t
for some . For Re (s) > , dene the Laplace Transform by
Lg (s)

0
e
su
g (u) du.
Assume that g has exponential growth as above and is Holder continuous from
the right and from the left at t. Pick > . Show that
lim
R
1
2
R
R
e
t
e
iyt
Lg ( +iy) dy =
g (t+) +g (t)
2
.
This formula is sometimes written in the form
1
2i
+i
i
e
st
Lg (s) ds
and is called the complex inversion integral for Laplace transforms. It can be used
to nd inverse Laplace transforms. Hint:
1
2
R
R
e
t
e
iyt
Lg ( +iy) dy =
1
2
R
R
e
t
e
iyt

0
e
(+iy)u
g (u) dudy.
Now use Fubinis theorem and do the integral from R to R to get this equal to
e
t
e
u
g (u)
sin (R(t u))
t u
du
where g is the zero extension of g o [0, ). Then this equals
e
t
e
(tu)
g (t u)
sin (Ru)
u
du
which equals
2e
t

0
g (t u) e
(tu)
+g (t +u) e
(t+u)
2
sin (Ru)
u
du
and then apply the result of Problem 17.
20. Let K be a nonempty closed and convex subset of R
p
. Recall K is convex means
that if x, y K, then for all t [0, 1] , tx + (1 t) y K. Show that if x R
p
there exists a unique z K such that
|x z| = min {|x y| : y K} .
This z will be denoted as Px. Hint: First note you do not know K is compact.
Establish the parallelogram identity if you have not already done so,
|u v|
2
+|u +v|
2
= 2 |u|
2
+ 2 |v|
2
.
Then let {z
k
} be a minimizing sequence,
lim
k
|z
k
x|
2
= inf {|x y| : y K} .
Now using convexity, explain why
z
k
z
m
2
2
+
x
z
k
+z
m
2
2
= 2
x z
k
2
2
+ 2
x z
m
2
2
and then use this to argue {z
k
} is a Cauchy sequence. Then if z
i
works for i = 1, 2,
consider (z
1
+z
2
) /2 to get a contradiction.
9.13. EXERCISES 255
21. In Problem 20 show that Px satises the following variational inequality.
(xPx) (yPx) 0
for all y K. Then show that |Px
1
Px
2
| |x
1
x
2
|. Hint: For the rst part
note that if y K, the function t |x(Px +t (yPx))|
2
achieves its minimum
on [0, 1] at t = 0. For the second part,
(x
1
Px
1
) (Px
2
Px
1
) 0, (x
2
Px
2
) (Px
1
Px
2
) 0.
Explain why
(x
2
Px
2
(x
1
Px
1
)) (Px
2
Px
1
) 0
and then use a some manipulations and the Cauchy Schwarz inequality to get the
desired inequality.
22. Establish the Brouwer xed point theorem for any convex compact set in R
p
.
Hint: If K is a compact and convex set, let R be large enough that the closed
ball, D(0, R) K. Let P be the projection onto K as in Problem 21 above. If f
is a continuous map from K to K, consider f P. You want to show f has a xed
point in K.
23. In the situation of the implicit function theorem, suppose f (x
0
, y
0
) = 0 and
assume f is C
1
. Show that for (x, y) B(x
0
, ) B(y
0
, r) where , r are small
enough, the mapping
x T
y
(x) xD
1
f (x
0
, y
0
)
1
f (x, y)
is continuous and maps B(x
0
, ) to B(x
0
, /2) B(x
0
, ). Apply the Brouwer
xed point theorem to obtain a shorter proof of the implicit function theorem.
24. Here is a really interesting little theorem which depends on the Brouwer xed point
theorem. It plays a prominent role in the treatment of the change of variables
formula in Rudins book, [35] and is useful in other contexts as well. The idea is
that if a continuous function mapping a ball in R
k
to R
k
doesnt move any point
very much, then the image of the ball must contain a slightly smaller ball.
Lemma: Let B = B(0, r), a ball in R
k
and let F : B R
k
be continuous and
suppose for some < 1,
|F(v) v| < r (9.24)
for all v B. Then
F(B) B(0, r (1 )) .
Hint: Suppose a B(0, r (1 )) \ F(B) so it didnt work. First explain why
a = F(v) for all v B. Now letting G :B B, be dened by G(v)
r(aF(v))
|aF(v)|
,it
follows G is continuous. Then by the Brouwer xed point theorem, G(v) = v
for some v B. Explain why |v| = r. Then take the inner product with v and
explain the following steps.
(G(v) , v) = |v|
2
= r
2
=
r
|a F(v)|
(a F(v) , v)
=
r
|a F(v)|
(a v +v F(v) , v)
=
r
|a F(v)|
[(a v, v) +(v F(v) , v)]
=
r
|a F(v)|
_
(a, v) |v|
2
+(v F(v) , v)
_
r
|a F(v)|
_
r
2
(1 ) r
2
+r
2
= 0.
25. Using Problem 24 establish the following interesting result. Suppose f : U R
p
is dierentiable. Let
S = {x U : det Df (x) = 0}.
Show f (U \ S) is an open set.
26. Let K be a closed, bounded and convex set in R
p
and let f : K R
p
be continuous
and let y R
p
. Show using the Brouwer xed point theorem there exists a point
x K such that P (y f (x) +x) = x. Next show that (y f (x) , z x) 0
for all z K. The existence of this x is known as Browders lemma and it has
great signicance in the study of certain types of nolinear operators. Now suppose
f : R
p
R
p
is continuous and satises
lim
|x|
(f (x) , x)
|x|
= .
Show using Browders lemma that f is onto.
Brouwer Degree
This chapter is on the Brouwer degree, a very useful concept with numerous and impor-
tant applications. The degree can be used to prove some dicult theorems in topology
such as the Brouwer xed point theorem, the Jordan separation theorem, and the in-
variance of domain theorem. It also is used in bifurcation theory and many other areas
in which it is an essential tool. This is an advanced calculus course so the degree will be
developed for R
n
. When this is understood, it is not too dicult to extend to versions
of the degree which hold in Banach space. There is more on degree theory in the book
by Deimling [9] and much of the presentation here follows this reference.
To give you an idea what the degree is about, consider a real valued C
1
function
dened on an interval, I, and let y f (I) be such that f
(x) = 0 for all x f

1
(y). In
this case the degree is the sum of the signs of f
(x) for x f
1
(y), written as d (f, I, y).
y
In the above picture, d (f, I, y) is 0 because there are two places where the sign is 1
and two where it is 1.
The amazing thing about this is the number you obtain in this simple manner is
a specialization of something which is dened for continuous functions and which has
nothing to do with dierentiability.
There are many ways to obtain the Brouwer degree. The method I will use here
is due to Heinz [23] and appeared in 1959. It involves rst studying the degree for
functions in C
2
and establishing all its most important topological properties with the
aid of an integral. Then when this is done, it is very easy to extend to general continuous
functions.
When you have the topological degree, you can get all sorts of amazing theorems
like the invariance of domain theorem and others.
10.1 Preliminary Results
In this chapter will refer to a bounded open set.
257
258 BROUWER DEGREE
Denition 10.1.1 For a bounded open set, denote by C
_
_
the set of func-
tions which are continuous on and by C
m
_
_
, m the space of restrictions of
functions in C
m
c
(R
n
) to . The norm in C
_
_
is dened as follows.
||f||
= ||f||
C()
sup
_
|f (x)| : x
_
.
If the functions take values in R
n
write C
m
_
; R
n
_
or C
_
; R
n
_
for these functions
if there is no dierentiability assumed. The norm on C
_
; R
n
_
is dened in the same
way as above,
||f ||
= ||f ||
C(;R
n
)
sup
_
|f (x)| : x
_
.
Also, C (; R
n
) consists of functions which are continuous on that have values in R
n
and C
m
(; R
n
) denotes the functions which have m continuous derivatives dened on
.
Theorem 10.1.2 Let be a bounded open set in R
n
and let f C
_
_
. Then
there exists g C
_
with ||g f||
C()
< . In fact, g can be assumed to equal a
polynomial for all x .
Proof: This follows immediately from the Weierstrass approximation theorem, The-
orem 5.7.13. Pick a polynomial, p such that ||p f||
C()
< . Now p / C
_
because
it does not vanish outside some compact subset of R
n
so let g equal p multiplied by
some function C
c
(R
n
) where = 1 on . See Theorem 9.5.8.
Applying this result to the components of a vector valued function yields the follow-
ing corollary.
Corollary 10.1.3 If f C
_
; R
n
_
for a bounded subset of R
n
, then for all > 0,
there exists g C
_
; R
n
_
such that
||g f ||
< .
Lemma 9.12.1 on Page 246 will also play an important role in the denition of the
Brouwer degree. Earlier it made possible an easy proof of the Brouwer xed point
theorem. Later in this chapter, it is used to show the denition of the degree is well
dened. For convenience, here it is stated again.
Lemma 10.1.4 Let g : U R
n
be C
2
n
. Then
n
j=1
cof (Dg)
ij,j
= 0,
where here (Dg)
ij
g
i,j

gi
xj
. Also, cof (Dg)
ij
=
det(Dg)
gi,j
.
Another simple result which will be used whenever convenient is the following lemma,
stated in somewhat more generality than needed.
Lemma 10.1.5 Let K be a compact set and C a closed set in a complete normed
vector space such that K C = . Then
dist (K, C) > 0.
10.2. DEFINITIONS AND ELEMENTARY PROPERTIES 259
Proof: Let
d inf {||k c|| : k K, c C}
Let {k
n
} , {c
n
} be such that
d +
1
n
> ||k
n
c
n
|| .
Since K is compact, there is a subsequence still denoted by {k
n
} such that k
n
k K.
Then also
||c
n
c
m
|| ||c
n
k
n
|| +||k
n
k
m
|| +||c
m
k
m
||
If d = 0, then as m, n it follows ||c
n
c
m
|| 0 and so {c
n
} is a Cauchy sequence
which must converge to some c C. But then ||c k|| = lim
n
||c
n
k
n
|| = 0 and so
c = k C K, a contradiction to these sets being disjoint. This proves the lemma.
In particular the distance between a point and a closed set is always positive if the
point is not in the closed set. Of course this is obvious even without the above lemma.
10.2 Denitions And Elementary Properties
In this section, f : R
n
will be a continuous map. It is always assumed that f ()
misses the point y where d (f , , y) is the topological degree which is being dened.
Also, it is assumed is a bounded open set.
Denition 10.2.1 U
y

_
f C
_
; R
n
_
: y / f ()
_
. (Recall that = \
) For two functions,
f , g U
y
,
f g if there exists a continuous function,
h : [0, 1] R
n
such that h(x, 1) = g (x) and h(x, 0) = f (x) and x h(x,t) U
y
for all t
[0, 1] (y / h(, t)). This function, h, is called a homotopy and f and g are homotopic.
Denition 10.2.2 For W an open set in R
n
and g C
1
(W; R
n
) y is called a
regular value of g if whenever x g
1
(y), det (Dg (x)) = 0. Note that if g
1
(y) = ,
it follows that y is a regular value from this denition. Denote by S
g
the set of singular
values of g, those y such that det (Dg (x)) = 0 for some x g
1
(y).
Lemma 10.2.3 The relation is an equivalence relation and, denoting by [f ] the
equivalence class determined by f , it follows that [f ] is an open subset of
U
y

_
f C
_
; R
n
_
: y / f ()
_
.
Furthermore, U
y
is an open set in C
_
; R
n
_
and if f U
y
and > 0, there exists
g [f ] C
2
_
; R
n
_
for which y is a regular value of g and ||f g|| < .
Proof: In showing that is an equivalence relation, it is easy to verify that f f
and that if f g, then g f . To verify the transitive property for an equivalence
relation, suppose f g and g k, with the homotopy for f and g, the function, h
1
and the homotopy for g and k, the function h
2
. Thus h
1
(x,0) = f (x), h
1
(x,1) = g (x)
and h
2
(x,0) = g (x), h
2
(x,1) = k(x). Then dene a homotopy of f and k as follows.
h(x,t)
_
h
1
(x,2t) if t
_
0,
1
2
h
2
(x,2t 1) if t
_
1
2
, 1
.
260 BROUWER DEGREE
It is obvious that U
y
is an open subset of C
_
; R
n
_
. Next consider the claim that
[f ] is also an open set. If f U
y
, There exists > 0 such that B(y, 2) f () = .
Let f
1
C
_
; R
n
_
with ||f
1
f ||
< . Then if t [0, 1], and x

|f (x) +t (f
1
(x) f (x)) y| |f (x) y| t ||f f
1
||
> 2 t > 0.
Therefore, B(f ,) [f ] because if f
1
B(f , ), this shows that, letting h(x,t)
f (x) +t (f
1
(x) f (x)), f
1
f .
It remains to verify the last assertion of the lemma. Since [f ] is an open set, it
follows from Theorem 10.1.2 there exists g [f ] C
2
_
; R
n
_
and ||g f || < /2. If y
is a regular value of g, leave g unchanged. The desired function has been found. In the
other case, let be small enough that B(y, 2) g () = . Next let
S
_
x : det Dg (x) = 0
_
By Sards lemma, Lemma 9.9.9 on Page 240, g (S) is a set of measure zero and so in
particular contains no open ball and so there exists a regular values of g arbitrarily close
to y. Let y be one of these regular values, |y y| < /2, and consider
g
1
(x) g (x) +y y.
It follows g
1
(x) = y if and only if g (x) = y and so, since Dg (x) = Dg
1
(x), y is a
regular value of g
1
. Then for t [0, 1] and x ,
|g (x) +t (g
1
(x) g (x)) y| |g (x) y| t |y y| > 2 t > 0.
provided |y y| is small enough. It follows g
1
g and so g
1
f . Also provided |y y|
is small enough,
||f g
1
|| ||f g|| +||g g
1
||
< /2 +/2 = .
The main conclusion of this lemma is that for f U
y
, there always exists a function
g of C
2
_
; R
n
_
which is uniformly close to f , homotopic to f and also such that y is a
regular value of g.
10.2.1 The Degree For C
2
_
; R
n
_
Here I will give a denition of the degree which works for all functions in C
2
_
; R
n
_
.
Denition 10.2.4 Let g C
2
_
; R
n
_
U
y
where is a bounded open set.
Also let
be a mollier.
c
(B(0, )) ,
0,
dx = 1.
Then
d (g, , y) lim
0
(g (x) y) det Dg (x) dx

Lemma 10.2.5 The above denition is well dened. In particular the limit exists.
In fact

does not depend on whenenver is small enough. If y is a regular value for g then
for all small enough,
_
sgn (det Dg (x)) : x g
1
(y)
_
(10.1)
If f , g are two functions in C
2
_
; R
n
_
such that for all x
y / tf (x) + (1 t) g (x) (10.2)
for all t [0, 1] , then for each > 0,
(f (x) y) det Df (x) dx

=
(g (x) y) det Dg (x) dx (10.3)

If g, f U
y
C
2
_
; R
n
_
, and 10.2 holds, then
d (f , , y) = d (g, , y)
Proof:
The case where y is a regular value
First consider the case where y is a regular value of g. I will show that in this case,
the integral expression is eventually constant for small > 0 and equals the right side
of 10.1. I claim the right side of this equation is actually a nite sum. This follows from
the inverse function theorem because g
1
(y) is a closed, hence compact subset of due
to the assumption that y / g (). If g
1
(y) had innitely many points in it, there
would exist a sequence of distinct points {x
k
} g
1
(y). Since is bounded, some
subsequence {x
k
l
} would converge to a limit point x
. By continuity of g, it follows
x
g
1
(y) also and so x
. Therefore, since y is a regular value, there is an

open set, U
x
, containing x
such that g is one to one on this open set contradicting

the assertion that lim
l
x
k
l
= x
. Therefore, this set is nite and so the sum is well

dened.
Thus the right side of 10.1 is nite when y is a regular value. Next I need to show
the left side of this equation is eventually constant. By what was just shown, there are
nitely many points, {x
i
}
m
i=1
= g
1
(y). By the inverse function theorem, there exist
disjoint open sets U
i
with x
i
U
i
, such that g is one to one on U
i
with det (Dg (x))
having constant sign on U
i
and g (U
i
) is an open set containing y. Then let be small
enough that B(y, )
m
i=1
g (U
i
) and let V
i
g
1
(B(y, )) U
i
.
g(U
2
)
g(U
3
)
g(U
1
) y
x
1
x
2
x
3
V
1
V
2
V
3
262 BROUWER DEGREE
Therefore, for any this small,
(g (x) y) det Dg (x) dx =

m
i=1
Vi

The reason for this is as follows. The integrand on the left is nonzero only if g (x) y
B(0, ) which occurs only if g (x) B(y, ) which is the same as x g
1
(B(y, )).
Therefore, the integrand is nonzero only if x is contained in exactly one of the disjoint
sets V
i
. Now using the change of variables theorem,
=
m
i=1
g(Vi)y
(z) det Dg
_
g
1
(y +z)
_
det Dg
1
(y +z)
dz
By the chain rule, I = Dg
_
g
1
(y +z)
_
Dg
1
(y +z) and so
det Dg
_
g
1
(y +z)
_
det Dg
1
(y +z)
= sgn
_
det Dg
_
g
1
(y +z)
__
det Dg
_
g
1
(y +z)
_
det Dg
1
(y +z)
= sgn
_
det Dg
_
g
1
(y +z)
__
= sgn (det Dg (x)) = sgn (det Dg (x
i
)) .
Therefore, this reduces to
m
i=1
sgn (det Dg (x
i
))
g(Vi)y
(z) dz =
m
i=1
sgn (det Dg (x
i
))
B(0,)
(z) dz =
m
i=1
sgn (det Dg (x
i
)) .
In case g
1
(y) = , there exists > 0 such that g
_
_
B(y, ) = and so for this
small,
(g (x) y) det Dg (x) dx = 0.

Showing the integral is constant for small
With this done it is necessary to show that the integral in the denition of the degree
is constant for small enough even if y is not a regular value. To do this, I will rst
show that if 10.2 holds, then 10.3 holds. This particular part of the argument is the
trick which makes surprising things happen. This is where the fact the functions are
twice continuously dierentiable is used. Suppose then that f , g satisfy 10.2. Also let
> 0 be such that for all t [0, 1] ,
B(y, ) (f +t (g f )) () = (10.4)
Dene for t [0, 1],
H (t)
(f y +t (g f )) det (D(f +t (g f ))) dx.

Then if t (0, 1),
H
(t) =
,
(f (x) y +t (g (x) f (x)))
(g
(x) f
(x)) det D(f +t (g f )) dx

+
(f y +t (g f ))
,j
det D(f +t (g f ))
,j
(g
)
,j
dx A+B.
In this formula, the function det is considered as a function of the n
2
entries in the nn
matrix and the , j represents the derivative with respect to the j
th
entry. Now as in
the proof of Lemma 9.12.1 on Page 246,
det D(f +t (g f ))
,j
= (cof D(f +t (g f )))
j
and so
B =
(f y +t (g f ))
(cof D(f +t (g f )))
j
(g
)
,j
dx.
By hypothesis
x
(f (x) y +t (g (x) f (x)))

(cof D(f (x) +t (g (x) f (x))))
j
is in C
1
c
() because if x , it follows by 10.4 that for all t [0, 1]
f (x) y +t (g (x) f (x)) / B(0, )
and so
(f (x) y +t (g (x) f (x))) = 0. Furthermore this situation persists for x near

. Therefore, integrate by parts and write
B =
x
j
(
(f y +t (g f )))
j
(g
) dx+
(f y +t (g f ))
j,j
(g
) dx.
The second term equals zero by Lemma 10.1.4. Simplifying the rst term yields
B =
,
(f y +t (g f ))
(f
,j
+t (g
,j
f
,j
)) (cof D(f +t (g f )))
j
(g
) dx
=
,
(f y +t (g f ))
det (D(f +t (g f ))) (g
) dx
=
,
(f y +t (g f ))
det (D(f +t (g f ))) (g
) dx
264 BROUWER DEGREE
= A.
Therefore, H
(t) = 0 and so H is a constant.

Now let g U
y
C
2
_
; R
n
_
. By Sards lemma, Lemma 9.9.9 there exists a regular
value y
1
of g which is very close to y. This is because, by this lemma, the set of points
which are not regular values has measure zero so this set of points must have empty
interior. Let
g
1
(x) g (x) +y y
1
and let y
1
y be so small that
y / (1 t) g
1
+gt () g
1
+t (g g
1
) () for all t [0, 1] .
Then g
1
(x) = y if and only if g (x) = y
1
which is a regular value. Note also D(g (x)) =
D(g
1
(x)). Then from what was just shown, letting f = g and g = g
1
in the above and
using g y
1
= g
1
y,
(g (x) y
1
) det (D(g (x))) dx
=
(g
1
(x) y) det (D(g (x))) dx
=
(g (x) y) det (D(g (x))) dx

Since y
1
is a regular value of g it follows from the rst part of the argument that the
rst integral in the above is eventually constant for small enough . It follows the last
integral is also eventually constant for small enough . This proves the claim about
the limit existing and in fact being constant for small . The last claim follows right
away from the above. Suppose 10.2 holds. Then choosing small enough, it follows
d (f , , y) = d (g, , y) because the two integrals dening the degree for small are
equal. This proves the lemma.
Next I will show that if f g where f , g U
y
C
2
_
; R
n
_
then d (f , , y) =
d (g, , y) . In the special case where
h(x,t) = tf (x) + (1 t) g (x)
this has already been done in the above lemma. In the following lemma the two functions
k, l are only assumed to be continuous.
Lemma 10.2.6 Suppose k l. Then there exists a sequence of functions of U
y
,
{g
i
}
m
i=1
,
such that g
i
C
2
_
; R
n
_
, and dening g
0
k and g
m+1
l, there exists > 0 such
that for i = 1, , m+ 1,
B(y, ) (tg
i
+ (1 t) g
i1
) () = , for all t [0, 1] . (10.5)
Proof: This lemma is not really very surprising. By Lemma 10.2.3, [k] is an open
set and since everything in [k] is homotopic to k, it is also connected because it is path
connected. The Lemma merely asserts there exists a piecewise linear curve joining k
and l which stays within the open connected set, [k] in such a way that the vertices
of this curve are in the dense set C
2
_
; R
n
_
. This is the abstract idea. Now here is a
more down to earth treatment.
Let h : [0, 1] R
n
be a homotopy of k and l with the property that y / h(, t)
for all t [0, 1]. Such a homotopy exists because k l. Now let 0 = t
0
< t
1
< <
t
m+1
= 1 be such that for i = 1, , m+ 1,
||h(, t
i
) h(, t
i1
)||
< (10.6)
where > 0 is small. By Lemma 10.2.3, for each i {1, , m} , there exists g
i

U
y
C
2
_
; R
n
_
such that
||g
i
h(, t
i
)||
< (10.7)
Thus
||g
i
g
i1
||
||g
i
h(, t
i
)||
+||h(, t
i
) h(, t
i1
)||
+||g
i1
h(, t
i1
)||
< 3.
(Recall g
0
k in case i = 1.)
It was just shown that for each x ,
g
i
(x) B(g
i1
(x) , 3)
Also for each x
|g
j
(x) y| |h(x, t
j
) y| |g
j
(x) h(x, t
j
)|
|h(x, t
j
) y|
For x
|tg
i
(x) + (1 t) g
i1
(x) y|
= |g
i1
(x) +t (g
i
(x) g
i1
(x)) y|
|g
i1
(x) y| t |g
i
(x) g
i1
(x)|
(|h(x, t
i1
) y| ) t3
|h(x, t
i1
) y| 4 >
provided was small enough that
B(y, 5) h( [0, 1]) = (10.8)
so that for all t [0, 1] , |h(x, t) y| > 5. This proves the lemma.
With this lemma the homotopy invariance of the degree on C
2
_
; R
n
_
is easy to
obtain. The following theorem gives this homotopy invariance and summarizes one of
the results of Lemma 10.2.5.
Theorem 10.2.7 Let f , g U
y
C
2
_
; R
n
_
and suppose f g. Then
d (f ,, y) = d (g,, y) .
When f U
y
C
2
_
; R
n
_
and y is a regular value of g with f g,
d (f ,, y) =
_
1
(y)
_
.
The degree is an integer. Also
y d (f ,, y)
is continuous on R
n
\ f () and y d (f ,, y) is constant on every connected compo-
nent of R
n
\ f ().
266 BROUWER DEGREE
Proof: From Lemma 10.2.6 there exists a sequence of functions in C
2
_
; R
n
_
having
the properties listed there. Then from Lemma 10.2.5
d (f ,, y) = d (g
1
,, y) = d (g
2
,, y) = = d (g,, y) .
The second assertion follows from Lemma 10.2.5. Finally consider the claim the degree
is an integer. This is obvious if y is a regular point. If y is not a regular point, let
g
1
(x) g (x) +y y
1
where y
1
is a regular point of g and |y y
1
| is so small that
y / (tg
1
+ (1 t) g) () .
From Lemma 10.2.5
d (g
1
, , y) = d (g, , y) .
But since g
1
y = g y
1
and det Dg (x) = det Dg
1
(x) ,
d (g
1
, , y) = lim
0
(g
1
(x) y) det Dg (x) dx
= lim
0
(g (x) y
1
) det Dg (x) dx
which by Lemma 10.2.5 equals
_
1
(y
1
)
_
, an integer.
What about the continuity assertion and being constant on connected components?
Let U be a connected component of R
n
\ f () and let y, y
1
be two points of U. The
argument is similar to some of the above. Let y R
n
\ f () and suppose y
1
is very
close to y, close enough that for
f
1
(x) f (x) +y y
1
it follows
y / tf +(1 t) f
1
()
Then from Lemma 10.2.5
d (f , , y) = d (f
1
, , y) = lim
0
(f
1
(x) y) Df (x) dx
= lim
0
(f (x) y
1
) Df (x) dx d (f , , y
1
)
which shows that y d (f , , y) is continuous. Since it has integer values, it follows
from Corollary 5.3.15 on Page 93 that this function must be constant on every connected
component. This proves the theorem.
10.2.2 Denition Of The Degree For Continuous Functions
With the above results, it is now possible to extend the denition of the degree to con-
tinuous functions which have no dierentiability. It is desired to preserve the homotopy
invariance. This requires the following denition.
Denition 10.2.8 Let f U
y
where y R
n
\ f () . Then
d (f , , y) f (g, , y)
where g U
y
C
2
_
; R
n
_
and f g.
Theorem 10.2.9 The denition of the degree given in Denition 10.2.8 is well
dened, equals an integer, and satises the following properties. In what follows, id (x) =
x.
1. d (id, , y) = 1 if y .
2. If
i
,
i
open, and
1
2
= and if y / f
_
\ (
1

2
)
_
, then d (f ,
1
, y)+
d (f ,
2
, y) = d (f , , y).
3. If y / f
_
\
1
_
and
1
is an open subset of , then
d (f , , y) = d (f ,
1
, y) .
4. For y R
n
\ f () , if d (f , , y) = 0 then f
1
(y) = .
5. If f , g are homotopic with a homotopy, h : [0, 1] for which h(, t) does not
contain y, then d (g,, y) = d (f , , y).
6. d (, , y) is dened and constant on
_
g C
_
; R
n
_
: ||g f ||
< r
_
where r = dist (y, f ()).
7. d (f , , ) is constant on every connected component of R
n
\ f ().
8. d (g,, y) = d (f , , y) if g|
= f |
.
Proof: First it is necessary to show the denition is well dened. There are two
parts to this. First I need to show there exists g with the desired properties and then
I need to show that it doesnt matter which g I happen to pick. The rst part is easy.
Let be small enough that
B(y, ) f () = .
Then by Lemma 10.2.3 there exists g C
2
_
; R
n
_
such that ||g f ||
< . It follows
that for t [0, 1] ,
y / (tg + (1 t) f ) ()
and so g f . This does the rst part. Now consider the second part. Suppose g f
and g
1
f . Then by Lemma 10.2.3 again
g g
1
and by Theorem 10.2.7 it follows d (g,, y) = d (g
1
,, y) which shows the denition is
well dened. Also d (f ,, y) must be an integer because it equals d (g,, y) which is an
integer.
Now consider the properties. The rst one is obvious from Theorem 10.2.7 since y
is a regular point of id.
Consider the second property. The assumption implies
y / f () f (
1
) f (
2
)
Let g C
2
_
; R
n
_
such that ||f g||
is small enough that

y / g
_
\ (
1

2
)
_
(10.9)
and also small enough that
y / (tg + (1 t) f ) () , y / (tg + (1 t) f ) (
1
)
y / (tg + (1 t) f ) (
2
) (10.10)
268 BROUWER DEGREE
for all t [0, 1] . Then it follows from Lemma 10.2.5, for all small enough,
d (g, , y) =
(g (x) y) Dg (x) dx
From 10.9 there is a positive distance between the compact set
g
_
\ (
1

2
)
_
and y. Therefore, making still smaller if necessary,
(g (x) y) = 0 if x /
1

2
Therefore, using the denition of the degree and 10.10,
d (f , , y) = d (g, , y) = lim
0
(g (x) y) Dg (x) dx
= lim
0
_
1
(g (x) y) Dg (x) dx+
(g (x) y) Dg (x) dx
_
= d (g,
1
, y) +d (g,
2
, y)
= d (f ,
1
, y) +d (f ,
2
, y)
This proves the second property.
Consider the third. This really follows from the second property. You can take
2
= . I leave the details to you. To be more careful, you can modify the proof of
property 2 slightly.
The fourth property is very important because it can be used to deduce the existence
of solutions to a nonlinear equation. Suppose f
1
(y) = . I will show this requires
d (f , , y) = 0. It is assumed y / f () and so if f
1
(y) = , then y / f
_
_
.
Choosing g C
2
_
; R
n
_
such that ||f g||
is suciently small, it can be assumed

y / g
_
_
, y / (tf + (1 t) g) () for all t [0, 1] .
Then it follows from the denition of the degree
d (f , , y) = d (g, , y) lim
0
(g (x) y) Dg (x) dx = 0
because eventually is smaller than the distance from y to g
_
_
and so
(g (x) y) =
0 for all x .
Property 5 follows from the denition of the degree.
Consider the sixth property. If g is in the specied set, and t [0, 1] with x
tg (x) + (1 t) f (x) = f (x) +t (g (x) f (x))
and so
|tg (x) + (1 t) f (x) y|
|f (x) y| t |g (x) f (x)| > r tr 0
and from the denition of the degree
d (f , , y) = d (g, , y)
10.3. BORSUKS THEOREM 269
The seventh claim is done already for the case where f C
2
_
; R
n
_
in Theorem
10.2.7. It remains to verify this for the case where f is only continuous. This will be
done by showing y d (f ,, y) is continuous. Let y
0
R
n
\ f () and let be small
enough that
B(y
0
, 4) f () = .
Now let g C
2
_
; R
n
_
such that ||g f ||
< . Then for x , t [0, 1] , and

y B(y
0
, ) ,
|(tg+(1 t) f ) (x) y| |f (x) y| t |g (x) f (x)|
|f (x) y
0
| |y
0
y| ||g f ||
4 > 0.
Therefore, for all such y B(y
0
, )
d (f , , y) = d (g, , y)
and it was shown in Theorem 10.2.7 that y d (g, , y) is continuous. In particular
d (f , , ) is continuous at y
0
. Since y
0
was arbitrary, this shows y d (f , , y) is
continuous. Therefore, since it has integer values, this function is constant on every
connected component of R
n
\ f () by Corollary 5.3.15.
Consider the eighth claim about the degree in which f = g on . This one is easy
because for
y R
n
\ f () = R
n
\ g () ,
and x ,
tf (x) + (1 t) g (x) y = f (x) y = 0
for all t [0, 1] and so by the fth claim, d (f , , y) = d (g, , y) and This proves the
theorem.
10.3 Borsuks Theorem
In this section is an important theorem which can be used to verify that d (f , , y) = 0.
This is signicant because when this is known, it follows from Theorem 10.2.9 that
f
1
(y) = . In other words there exists x such that f (x) = y.
Denition 10.3.1 A bounded open set, is symmetric if = . A contin-
uous function, f : R
n
is odd if f (x) = f (x).
Suppose is symmetric and g C
2
_
; R
n
_
is an odd map for which 0 is a regular
value. Then the chain rule implies Dg (x) = Dg (x) and so d (g, , 0) must equal
an odd integer because if x g
1
(0), it follows that x g
1
(0) also and since
Dg (x) = Dg (x), it follows the overall contribution to the degree from x and x
must be an even integer. Also 0 g
1
(0) and so the degree equals an even integer
added to sgn (det Dg (0)), an odd integer, either 1 or 1. It seems reasonable to expect
that something like this would hold for an arbitrary continuous odd function dened on
symmetric . In fact this is the case and this is next. The following lemma is the key
result used. This approach is due to Gromes [20]. See also Deimling [9] which is where
I found this argument.
The idea is to start with a smooth odd map and approximate it with a smooth odd
map which also has 0 a regular value.
270 BROUWER DEGREE
Lemma 10.3.2 Let g C
2
_
; R
n
_
be an odd map. Then for every > 0, there
exists h C
2
_
; R
n
_
such that h is also an odd map, ||h g||
< , and 0 is a regular

value of h.
Proof: In this argument > 0 will be a small positive number and C will be a
constant which depends only on the diameter of . Let h
0
(x) = g (x) +x where is
chosen such that det Dh
0
(0) = 0. Now let
i
{x : x
i
= 0}. In other words, leave
out the plane x
i
= 0 from in order to obtain
i
. A succession of modications is
about to take place on
1
,
1

2
, etc. Finally a function will be obtained on
n
j=1
j
which is everything except 0.
Dene h
1
(x) h
0
(x)y
1
x
3
1
where
y
1
< and y
1
=
_
y
1
1
, , y
1
n
_
is a regular value
of the function, x
h0(x)
x
3
1
for x
1
. the existence of y
1
follows from Sards lemma
because this function is in C
2
(
1
; R
n
). Thus h
1
(x) = 0 if and only if y
1
=
h0(x)
x
3
1
.
Since y
1
is a regular value, it follows that for such x,
det
_
h
0i,j
(x) x
3
1

xj
_
x
3
1
_
h
0i
(x)
x
6
1
_
=
det
_
h
0i,j
(x) x
3
1

xj
_
x
3
1
_
y
1
i
x
3
1
x
6
1
_
= 0
implying that
det
_
h
0i,j
(x)

x
j
_
x
3
1
_
y
1
i
_
= det (Dh
1
(x)) = 0.
This shows 0 is a regular value of h
1
on the set
1
and it is clear h
1
is an odd map in
C
2
_
; R
n
_
and ||h
1
g||
C where C depends only on the diameter of .

Now suppose for some k such that 1 k < n there exists an odd mapping h
k
in C
2
_
; R
n
_
such that 0 is a regular value of h
k
on
k
i=1
i
and ||h
k
g||
C.
Sards theorem implies there exists y
k+1
a regular value of the function x h
k
(x) /x
3
k+1
dened on
k+1
such that
y
k+1
< and let h

k+1
(x) h
k
(x)y
k+1
x
3
k+1
. As before,
h
k+1
(x) = 0 if and only if h
k
(x) /x
3
k+1
= y
k+1
, a regular value of x h
k
(x) /x
3
k+1
.
Consider such x for which h
k+1
(x) = 0. First suppose x
k+1
. Then
det
_
h
ki,j
(x) x
3
k+1

xj
_
x
3
k+1
_
y
k+1
i
x
3
k+1
x
6
k+1
_
= 0
which implies that whenever h
k+1
(x) = 0 and x
k+1
,
det
_
h
ki,j
(x)

x
j
_
x
3
k+1
_
y
k+1
i
_
= det (Dh
k+1
(x)) = 0. (10.11)
However, if x
k
i=1
k
but x /
k+1
, then x
k+1
= 0 and so the left side of 10.11
reduces to det (h
ki,j
(x)) which is not zero because 0 is assumed a regular value of h
k
.
Therefore, 0 is a regular value for h
k+1
on
k+1
i=1
k
. (For x
k+1
i=1
k
, either x
k+1
or
x /
k+1
. If x
k+1
0 is a regular value by the construction above. In the other case,
0 is a regular value by the induction hypothesis.) Also h
k+1
is odd and in C
2
_
; R
n
_
,
and ||h
k+1
g||
C.
Let h h
n
. Then 0 is a regular value of h for x
n
j=1
j
. The point of which is
not in
n
j=1
j
is 0. If x = 0, then from the construction, Dh(0) = Dh
0
(0) and so 0 is
a regular value of h for x . By choosing small enough, it follows ||h g||
< .
10.4. APPLICATIONS 271
Theorem 10.3.3 (Borsuk) Let f C
_
; R
n
_
be odd and let be symmetric
with 0 / f (). Then d (f , , 0) equals an odd integer.
Proof: Let be small enough that B(0,3) f () = . Let
g
1
C
2
_
; R
n
_
be such that ||f g
1
||
< and let g denote the odd part of g

1
. Thus
g (x)
1
2
(g
1
(x) g
1
(x)) .
Then since f is odd, it follows that for x ,
|f (x) g (x)| =
1
2
(f (x) f (x))
1
2
(g
1
(x) g
1
(x))
1
2
|f (x) g
1
(x)| +
1
2
|f (x) g
1
(x)| <
Thus ||f g||
< also. By Lemma 10.3.2 there exists odd h C

2
_
; R
n
_
for which
0 is a regular value and ||h g||
< . Therefore,
||f h||
||f g||
+||g h||
< 2
and so for t [0, 1] and x ,
|th(x) + (1 t) f (x) 0| |f (x) 0| t |h(x) f (x)|
3 > 0
and so, from the from the denition of the degree, d (f , , 0) = d (h, , 0).
Since 0 is a regular point of h, h
1
(0) = {x
i
, x
i
, 0}
m
i=1
, and since h is odd,
Dh(x
i
) = Dh(x
i
) and so
d (h, , 0)
m
i=1
sgn det (Dh(x
i
)) +
m
i=1
sgn det (Dh(x
i
)) + sgn det (Dh(0)) ,
an odd integer.
10.4 Applications
With these theorems it is possible to give easy proofs of some very important and
dicult theorems.
Denition 10.4.1 If f : U R
n
R
n
where U is an open set. Then f is
locally one to one if for every x U, there exists > 0 such that f is one to one on
B(x, ).
As a rst application, consider the invariance of domain theorem. This result says
that a one to one continuous map takes open sets to open sets. It is an amazing result
which is essential to understand if you wish to study manifolds. In fact, the following
theorem only requires f to be locally one to one. First here is a lemma which has the
main idea.
272 BROUWER DEGREE
Lemma 10.4.2 Let g : B(0,r) R
n
be one to one and continuous where here
B(0,r) is the ball centered at 0 of radius r in R
n
. Then there exists > 0 such that
g (0) +B(0, ) g (B(0,r)) .
The symbol on the left means: {g (0) +x : x B(0, )} .
Proof: For t [0, 1] , let
h(x, t) g
_
x
1 +t
_
g
_
tx
1 +t
_
Then for x B(0, r) , h(0, t) = 0 because if this were so, the fact g is one to one
implies
x
1 +t
=
tx
1 +t
and this requires x = 0 which is not the case. Since B(0,r) [0, 1] is compact, there
exists > 0 such that for all t [0, 1] and x B(0,r) ,
B(0, ) h(x,t) = .
In particular, when t = 0, B(0,) is contained in a single component of
R
n
\ (g g (0)) (B(0,r))
and so
d (g g (0) , B(0,r) , z)
is constant for z B(0,) . Therefore, from the properties of the degree in Theorem
10.2.9,
d (g g (0) , B(0,r) , z) = d (g g (0) , B(0,r) , 0)
= d (h(, 0) , B(0,r) , 0)
= d (h(, 1) , B(0,r) , 0) = 0
the last assertion following from Borsuks theorem, Theorem 10.3.3 and the observation
that h(, 1) is odd. From Theorem 10.2.9 again, it follows that for all z B(0, ) there
exists x B(0,r) such that
z = g (x) g (0)
which shows g (0) +B(0, ) g (B(0, r)) and This proves the lemma.
Now with this lemma, it is easy to prove the very important invariance of domain
theorem.
Theorem 10.4.3 (invariance of domain)Let be any open subset of R
n
and
let f : R
n
be continuous and locally one to one. Then f maps open subsets of to
open sets in R
n
.
Proof: Let B(x
0
, r) U where f is one to one on B(x
0
, r) and U is an open
subset of . Let g be dened on B(0, r) given by
g (x) f (x +x
0
)
Then g satises the conditions of Lemma 10.4.2, being one to one and continuous. It
follows from that lemma there exists > 0 such that
f (U) f (B(x
0
, r)) = f (x
0
+B(0, r))
= g (B(0,r)) g (0) +B(0, )
= f (x
0
) +B(0,) = B(f (x
0
) , )
This shows that for any x
0
U, f (x
0
) is an interior point of f (U) which shows f (U) is
open. This proves the theorem.
10.4. APPLICATIONS 273
Corollary 10.4.4 If n > m there does not exist a continuous one to one map from
R
n
to R
m
.
Proof: Suppose not and let f be such a continuous map,
f (x) (f
1
(x) , , f
m
(x))
T
.
Then let g (x) (f
1
(x) , , f
m
(x) , 0, , 0)
T
where there are n m zeros added in.
Then g is a one to one continuous map from R
n
to R
n
and so g (R
n
) would have to be
open from the invariance of domain theorem and this is not the case. This proves the
corollary.
Corollary 10.4.5 If f is locally one to one and continuous, f : R
n
R
n
, and
lim
|x|
|f (x)| = ,
then f maps R
n
onto R
n
.
Proof: By the invariance of domain theorem, f (R
n
) is an open set. It is also true
that f (R
n
) is a closed set. Here is why. If f (x
k
) y, the growth condition ensures
that {x
k
} is a bounded sequence. Taking a subsequence which converges to x R
n
and using the continuity of f , it follows f (x) = y. Thus f (R
n
) is both open and closed
which implies f must be an onto map since otherwise, R
n
would not be connected.
The next theorem is the famous Brouwer xed point theorem.
Theorem 10.4.6 (Brouwer xed point) Let B = B(0, r) R
n
and let f : B
B be continuous. Then there exists a point, x B, such that f (x) = x.
Proof: Consider h(x, t) tf (x) x for t [0, 1] . Then if there is no xed point
in B for f , it follows that 0 / h(B, t) for all t. When t = 1, this follows from having
no xed point for f . If t < 1, then if this were not so, then for some x B,
tf (x) = x
and taking norms,
r > tr = |x| = r
a contradiction.
Therefore, by the homotopy invariance,
0 = d (f id, B, 0) = d (id, B, 0) = (1)
n
,
a contradiction. This proves the theorem.
Denition 10.4.7 f is a retraction of B(0, r) onto B(0, r) if f is continuous,
f
_
B(0, r)
_
B(0,r), and f (x) = x for all x B(0,r).
Theorem 10.4.8 There does not exist a retraction of B(0, r) onto its bound-
ary, B(0, r).
Proof: Suppose f were such a retraction. Then for all x B(0, r), f (x) = x and
so from the properties of the degree, the one which says if two functions agree on ,
then they have the same degree,
1 = d (id, B(0, r) , 0) = d (f , B(0, r) , 0)
274 BROUWER DEGREE
which is clearly impossible because f
1
(0) = which implies d (f , B(0, r) , 0) = 0. This
proves the theorem.
You should now use this theorem to give another proof of the Brouwer xed point
theorem.
The proofs of the next two theorems make use of the Tietze extension theorem,
Theorem 5.7.12.
Theorem 10.4.9 Let be a symmetric open set in R
n
such that 0 and let
f : V be continuous where V is an m dimensional subspace of R
n
, m < n. Then
f (x) = f (x) for some x .
Proof: Suppose not. Using the Tietze extension theorem, extend f to all of ,
f
_
_
V . (Here the extended function is also denoted by f .) Let g (x) = f (x)f (x).
Then 0 / g () and so for some r > 0, B(0,r) R
n
\ g (). For z B(0,r),
d (g, , z) = d (g, , 0) = 0
because B(0,r) is contained in a component of R
n
\g () and Borsuks theorem implies
that d (g, , 0) = 0 since g is odd. Hence
V g () B(0,r)
and this is a contradiction because V is m dimensional. This proves the theorem.
This theorem is called the Borsuk Ulam theorem. Note that it implies there exist
two points on opposite sides of the surface of the earth which have the same atmospheric
pressure and temperature, assuming the earth is symmetric and that pressure and tem-
perature are continuous functions. The next theorem is an amusing result which is like
combing hair. It gives the existence of a cowlick.
Theorem 10.4.10 Let n be odd and let be an open bounded set in R
n
with
0 . Suppose f : R
n
\ {0} is continuous. Then for some x and = 0,
f (x) = x.
Proof: Using the Tietze extension theorem, extend f to all of . Also denote the
extended function by f . Suppose for all x , f (x) = x for all R. Then
0 / tf (x) + (1 t) x, (x, t) [0, 1]
0 / tf (x) (1 t) x, (x, t) [0, 1] .
Thus there exists a homotopy of f and id and a homotopy of f and id. Then by the
homotopy invariance of degree,
d (f , , 0) = d (id, , 0) , d (f , , 0) = d (id, , 0) .
But this is impossible because d (id, , 0) = 1 but d (id, , 0) = (1)
n
= 1. This
proves the theorem.
10.5 The Product Formula
This section is on the product formula for the degree which is used to prove the Jordan
separation theorem. To begin with here is a lemma which is similar to an earlier result
except here there are r points.
Lemma 10.5.1 Let y
1
, , y
r
be points not in f () and let > 0. Then there
exists
f C
2
_
; R
n
_
such that
f f
< and y
i
is a regular value for
f for each i.
10.5. THE PRODUCT FORMULA 275
Proof: Let f
0
C
2
_
; R
n
_
, ||f
0
f ||
<

2
. Let y
1
be a regular value for f
0
and
| y
1
y
1
| <

3r
. Let f
1
(x) f
0
(x) + y
1
y
1
. Thus y
1
is a regular value of f
1
because
Df
1
(x) = Df
0
(x) and if f
1
(x) = y
1
, this is the same as having f
0
(x) = y
1
where y
1
is
a regular value of f
0
. Then also
||f f
1
||
||f f
0
||
+||f
0
f
1
||
= ||f f
0
||
+| y
1
y
1
|
<

3r
+

2
.
Suppose now there exists f
k
C
2
_
; R
n
_
with each of the y
i
for i = 1, , k a regular
value of f
k
and
||f f
k
||
<

2
+
k
r
_
3
_
.
Then letting S
k
denote the singular values of f
k
, Sards theorem implies there exists
y
k+1
such that
| y
k+1
y
k+1
| <

3r
and
y
k+1
/ S
k

k
i=1
(S
k
+y
k+1
y
i
) . (10.12)
Let
f
k+1
(x) f
k
(x) +y
k+1
y
k+1
. (10.13)
If f
k+1
(x) = y
i
for some i k, then
f
k
(x) +y
k+1
y
i
= y
k+1
and so f
k
(x) is a regular value for f
k
since by 10.12, y
k+1
/ S
k
+ y
k+1
y
i
and
so f
k
(x) / S
k
. Therefore, for i k, y
i
is a regular value of f
k+1
since by 10.13,
Df
k+1
= Df
k
. Now suppose f
k+1
(x) = y
k+1
. Then
y
k+1
= f
k
(x) +y
k+1
y
k+1
so f
k
(x) = y
k+1
implying that f
k
(x) = y
k+1
/ S
k
. Hence det Df
k+1
(x) = det Df
k
(x) =
0. Thus y
k+1
is also a regular value of f
k+1
. Also,
||f
k+1
f || ||f
k+1
f
k
|| +||f
k
f ||

3r
+

2
+
k
r
_
3
_
=

2
+
k + 1
r
_
3
_
.
Let
f f
r
. Then
f f
<

2
+
_
3
_
<
and each of the y
i
is a regular value of
f . This proves the lemma.

Denition 10.5.2 Let the connected components of R
n
\ f () be denoted by
K
i
. From the properties of the degree listed in Theorem 10.2.9, d (f , , ) is constant on
each of these components. Denote by d (f , , K
i
) the constant value on the component,
K
i
.
276 BROUWER DEGREE
The product formula considers the situation depicted in the following diagram in
which y / g (f ()) and the K
i
are the connected components of R
n
\ f ().
f
f
_
_
R
n
\f ()=iKi
g
R
n
y
The following diagram may be helpful in remembering what it says.
K
1
K
2
K
3
E
g
E
f
y
Lemma 10.5.3 Let f C
_
; R
n
_
, g C
2
(R
n
, R
n
) , and y / g (f ()). Suppose
also that y is a regular value of g. Then the following product formula holds where K
i
are the bounded components of R
n
\ f ().
d (g f , , y) =
i=1
d (f , , K
i
) d (g, K
i
, y) .
All but nitely many terms in the sum are zero.
Proof: First note that if K
i
is unbounded, d (f , , K
i
) = 0 because there exists
a point, z K
i
such that f
1
(z) = due to the fact that f
_
_
is compact and is
consequently bounded. Thus it makes no dierence in the above formula whether the
K
i
are arbitrary components or only bounded components. Let
_
x
i
j
_
mi
j=1
denote the
points of g
1
(y) which are contained in K
i
, the i
th
bounded component of R
n
\ f ().
Then m
i
< because if not, there would exist a limit point x for this sequence. Then
g (x) = y and so x / f (). Thus det (Dg (x)) = 0 and so by the inverse function
theorem, g would be one to one on an open ball containing x which contradicts having
x a limit point.
Note also that g
1
(y) f
_
_
is a compact set covered by the components of R
n
\
f () because g
1
(y) f () = . It follows g
1
(y) f
_
_
is covered by nitely
many of these components.
The only terms in the above sum which are nonzero are those corresponding to K
i
having nonempty intersection with g
1
(y) f
_
_
. The other components contribute
0 to the above sum because if K
i
g
1
(y) = , it follows from Theorem 10.2.9 that
d (g, K
i
, y) = 0. If K
i
does not intersect f
_
_
, then d (f , , K
i
) = 0. Therefore, the
above sum is actually a nite sum since g
1
(y) f
_
_
, being a compact set, is covered
by nitely many of the K
i
. Thus there are no convergence problems. Now let > 0 be
small enough that
B(y, 3) g (f ()) = ,
and for each x
i
j
g
1
(y)
B
_
x
i
j
, 3
_
f () = .
10.5. THE PRODUCT FORMULA 277
By uniform continuity of g on the compact set f
_
_
, there exists > 0, < such that if
z
1
and z
2
are any two points of f
_
_
with |z
1
z
2
| < , it follows that |g (z
1
) g (z
2
)| <
.
Now using Lemma 10.5.1, choose
f C
2
_
; R
n
_
such that
f f
< and each

point, x
i
j
is a regular value of
f . From the properties of the degree in Theorem 10.2.9

d (f , , K
i
) = d
_
f , , x
i
j
_
for each j = 1, , m
i
. For x , and t [0, 1],
f (x) +t
_
f (x) f (x)
_
x
i
j
f (x) x
i
j
f (x) f (x)
> 3 t > 0
and so by the homotopy invariance of the degree,
d
_
f , , x
i
j
_
= d
_
f , , x
i
j
_
d (f , , K
i
) (10.14)
independent of j. Also for x , and t [0, 1],
g (f (x)) +t
_
g
_
f (x)
_
g (f (x))
_
y
3 t > 0
and so by the homotopy invariance of the degree,
d (g f , , y) = d
_
g
f , , y
_
. (10.15)
Now

f
1
_
x
i
j
_
is a nite set because

f
1
_
x
i
j
_
, a bounded open set and x
i
j
is a
regular value. It follows from 10.15
d (g f , , y) = d
_
g
f , , y
_
=
i=1
mi
j=1
f
1
(x
i
j
)
sgn det Dg
_
_
_
_
x
i
j
..
f (z)
_
_
_
_
sgn det D
f (z)
=
i=1
mi
j=1
sgn det Dg
_
x
i
j
_
d
_
f , , x
i
j
_
=
i=1
d (g, K
i
, y) d
_
f , , x
i
j
_
=
i=1
d (g, K
i
, y) d (f , , K
i
) .
With this lemma, the following is the product formula.
Theorem 10.5.4 (product formula) Let {K
i
}
i=1
be the bounded components of
R
n
\ f () for f C
_
; R
n
_
, let g C (R
n
, R
n
), and suppose that y / g (f ()).
Then
d (g f , , y) =
i=1
d (g, K
i
, y) d (f , , K
i
) . (10.16)
All but nitely many terms in the sum are zero.
278 BROUWER DEGREE
Proof: Let B(y,3) g (f ()) = and let g C
2
(R
n
, R
n
) be such that
sup
_
| g (z) g (z)| : z f
_
__
<
And also y is a regular value of g. Then from the above inequality, if x and
t [0, 1],
|g (f (x)) +t ( g (f (x)) g (f (x))) y| |g (f (x)) y| t | g (f (x)) g (f (x))|
3 t > 0.
It follows that
d (g f , , y) = d ( g f , , y) . (10.17)
Now also, K
i
f () and so if z K
i
, then g (z) g (f ()). Consequently, for
such z,
|g (z) +t ( g (z) g (z)) y| |g (z) y| t > 3 t > 0
which shows that, by homotopy invariance,
d (g, K
i
, y) = d ( g, K
i
, y) . (10.18)
Therefore, by Lemma 10.5.3,
d (g f , , y) = d ( g f , , y) =
i=1
d ( g, K
i
, y) d (f , , K
i
)
=
i=1
d (g, K
i
, y) d (f , , K
i
)
and the sum has only nitely many non zero terms. This proves the product formula.
Note there are no convergence problems because these sums are actually nite sums
because, as in the previous lemma, g
1
(y) f
_
_
is a compact set covered by the
components of R
n
\ f () and so it is covered by nitely many of these components.
For the other components, d (f , , K
i
) = 0 or else d (g, K
i
, y) = 0.
The following theorem is the Jordan separation theorem, a major result. A home-
omorphism is a function which is one to one onto and continuous having continuous
inverse. Before the theorem, here is a helpful lemma.
Lemma 10.5.5 Let be a bounded open set in R
n
, f C
_
; R
n
_
, and suppose
{
i
}
i=1
are disjoint open sets contained in such that
y / f
_
\
j=1
j
_
Then
d (f , , y) =
j=1
d (f ,
j
, y)
where the sum has all but nitely many terms equal to 0.
Proof: By assumption, the compact set f
1
(y) has empty intersection with
\
j=1
j
and so this compact set is covered by nitely many of the
j
, say {
1
, ,
n1
} and
y / f
_
j=n
j
_
10.6. JORDAN SEPARATION THEOREM 279
By Theorem 10.2.9 and letting O =
j=n
j
,
d (f , , y) =
n1
j=1
d (f ,
j
, y) +d (f ,O, y) =
j=1
d (f ,
j
, y)
because d (f ,O, y) = 0 as is d (f ,
j
, y) for every j n. This proves the lemma.
Lemma 10.5.6 Dene U to be those points x with the property that for every r > 0,
B(x, r) contains points of U and points of U
C
. Then for U an open set,
U = U \ U
Let C be a closed subset of R
n
and let K denote the set of components of R
n
\ C. Then
if K is one of these components, it is open and
K C
Proof: Let x U \ U. If B(x, r) contains no points of U, then x / U. If B(x, r)
contains no points of U
C
, then x U and so x / U \ U. Therefore, U \ U U. Now
let x U. If x U, then since U is open there is a ball containing x which is contained
in U contrary to x U. Therefore, x / U. If x is not a limit point of U, then some
ball containing x contains no points of U contrary to x U. Therefore, x U \ U
which shows the two sets are equal.
Why is K open for K a component of R
n
\ C? This is obvious because in R
n
an
open ball is connected. Thus if k K,letting B(k, r) C
C
, it follows K B(k, r) is
connected and contained in C
C
. Thus K B(k, r) is connected, contained in C
C
, and
therefore is contained in K because K is maximal with respect to being connected and
contained in C
C
.
Now for K a component of R
n
\ C, why is K C? Let x K. If x / C, then
x K
1
, some component of R
n
\ C. If K
1
= K then x cannot be a limit point of K
and so it cannot be in K. Therefore, K = K
1
but this also is a contradiction because
if x K then x / K and This proves the lemma.
10.6 Jordan Separation Theorem
Theorem 10.6.1 (Jordan separation theorem) Let f be a homeomorphism of C
and f (C) where C is a compact set in R
n
. Then R
n
\ C and R
n
\ f (C) have the same
number of connected components.
Proof : Denote by K the bounded components of R
n
\ C and denote by L, the
bounded components of R
n
\ f (C). Also, using the Tietze extension theorem, there
exists f an extension of f to all of R
n
which maps into a bounded set and let f
1
be an
extension of f
1
to all of R
n
which also maps into a bounded set. Pick K K and take
y K. Then
y / f
1
_
f (K)
_
because by Lemma 10.5.6, K C and on C, f = f . Thus the right side is of the form
f
1
_
_
_
f (C)
..
f (K)
_
_
_ = f
1
(f (K)) C
280 BROUWER DEGREE
and y / C. Since f
1
f equals the identity id on K, it follows from the properties of
the degree that
1 = d (id, K, y) = d
_
f
1
f , K, y
_
.
Let H denote the set of bounded components of R
n
\ f (K). By the product formula,
1 = d
_
f
1
f , K, y
_
=
HH
d
_
f , K, H
_
d
_
f
1
, H, y
_
, (10.19)
the sum being a nite sum from the product formula. It might help to consult the
following diagram.
R
n
\ C
f
f
1
R
n
\ f (C)
K L
K R
n
\ f (K)
y K H, H
1
H
L
H
Now letting x L L, if S is a connected set containing x and contained in R
n
\f (C),
then it follows S is contained in R
n
\ f (K) because K C. Therefore, every set of
L is contained in some set of H. Furthermore, if any L L has nonempty intersection
with H H then it must be contained in H. This is because
L = (L H) (L H)
_
L H
C
_
.
Now by Lemma 10.5.6,
L H L f (K) L f (C) = .
Since L is connected, LH
C
= . Letting L
H
denote those sets of L which are contained
in H equivalently having nonempty intersection with H, if p H \ L
H
= H \ L,
then p H f (C) and so
H = (L
H
) (H f (C)) (10.20)
Claim 1:
H \ L
H
f (C) .
Proof of the claim: Suppose p H \ L
H
but p / f (C). Then p L L.
It must be the case that L has nonempty intersection with H since otherwise p could
not be in H. However, as shown above, this requires L H and now by 10.20 and
p / L
H
, it follows p f (C) after all. This proves the claim.
Claim 2: y / f
1
_
H \ L
H
_
. Recall y K K the bounded components of
R
n
\ C.
Proof of the claim: If not, then f
1
(z) = y where z H \ L
H
f (C) and
so z = f (w) for some w C and so y = f
1
(f (w)) = w C contrary to y K, a
component of R
n
\ C.
Now every set of L is contained in some set of H. What about those sets of H which
contain no set of L so that L
H
= ? From 10.20 it follows H f (C). Therefore,
d
_
f
1
, H, y
_
= d
_
f
1
, H, y
_
= 0
10.6. JORDAN SEPARATION THEOREM 281
because y K a component of R
n
\ C. Therefore, letting H
1
denote those sets of H
which contain some set of L, 10.19 is of the form
1 =
HH1
d
_
f , K, H
_
d
_
f
1
, H, y
_
.
and it is still a nite sum because the terms in the sum are 0 for all but nitely many
H H
1
. I want to expand d
_
f
1
, H, y
_
as a sum of the form
LLH
d
_
f
1
, L, y
_
using Lemma 10.5.5. Therefore, I must verify
y / f
1
_
H \ L
H
_
but this is just Claim 2. By Lemma 10.5.5, I can write the above sum in place of
d
_
f
1
, H, y
_
. Therefore,
1 =
HH1
d
_
f , K, H
_
d
_
f
1
, H, y
_
=
HH1
d
_
f , K, H
_

LLH
d
_
f
1
, L, y
_
where there are only nitely many H which give a nonzero term and for each of these,
there are only nitely many L in L
H
which yield d
_
f
1
, L, y
_
= 0. Now the above
equals
=
HH1
LLH
d
_
f , K, H
_
d
_
f
1
, L, y
_
. (10.21)
By denition,
d
_
f , K, H
_
= d
_
f , K, x
_
where x is any point of H. In particular d
_
f , K, H
_
= d
_
f , K, L
_
for any L L
H
.
Therefore, the above reduces to
=
LL
d
_
f , K, L
_
d
_
f
1
, L, y
_
(10.22)
Here is why. There are nitely many H H
1
for which the term in the double sum of
10.21 is not zero, say H
1
, , H
m
. Then the above sum in 10.22 equals
m
k=1
LL
H
k
d
_
f , K, L
_
d
_
f
1
, L, y
_
+
L\
m
k=1
LH
k
d
_
f , K, L
_
d
_
f
1
, L, y
_
The second sum equals 0 because those L are contained in some H H for which
0 = d
_
f , K, H
_
d
_
f
1
, H, y
_
= d
_
f , K, H
_

LLH
d
_
f
1
, L, y
_
=
LL
H
d
_
f , K, L
_
d
_
f
1
, L, y
_
.
Therefore, the sum in 10.22 reduces to
m
k=1
LL
H
k
d
_
f , K, L
_
d
_
f
1
, L, y
_
282 BROUWER DEGREE
which is the same as the sum in 10.21. Therefore, 10.22 does follow. Then the sum in
10.22 reduces to
=
LL
d
_
f , K, L
_
d
_
f
1
, L, K
_
and all but nitely many terms in the sum are 0.
By the same argument,
1 =
KK
d
_
f , K, L
_
d
_
f
1
, L, K
_
and all but nitely many terms in the sum are 0. Letting |K| denote the number of
elements in K, similar for L,
|K| =
KK
1 =
KK
_
LL
d
_
f , K, L
_
d
_
f
1
, L, K
_
_
|L| =
LL
1 =
LL
_
KK
d
_
f , K, L
_
d
_
f
1
, L, K
_
_
Suppose |K| < . Then you can switch the order of summation in the double sum for
|K| and so
|K| =
KK
_
LL
d
_
f , K, L
_
d
_
f
1
, L, K
_
_
=
LL
_
KK
d
_
f , K, L
_
d
_
f
1
, L, K
_
_
= |L|
It follows that if either |K| or |L| is nite, then they are equal. Thus if one is innite, so
is the other. This proves the theorem because if n > 1 there is exactly one unbounded
component to both R
n
\C and R
n
\f (C) and if n = 1 there are exactly two unbounded
components.
As an application, here is a very interesting little result. It has to do with d (f , , f (x))
in the case where f is one to one and is connected. You might imagine this should
equal 1 or 1 based on one dimensional analogies. In fact this is the case and it is a
nice application of the Jordan separation theorem and the product formula.
Proposition 10.6.2 Let be an open connected bounded set in R
n
, n 1 such
that R
n
\ consists of two, three if n = 1, connected components. Let f C
_
; R
n
_
be continuous and one to one. Then f () is the bounded component of R
n
\ f () and
for y f () , d (f , , y) either equals 1 or 1.
Proof: First suppose n 2. By the Jordan separation theorem, R
n
\f () consists
of two components, a bounded component B and an unbounded component U. Using
the Tietze extention theorem, there exists g dened on R
n
such that g = f
1
on f
_
_
.
Thus on , g f = id. It follows from this and the product formula that
1 = d (id, , g (y)) = d (g f , , g (y))
= d (g, B, g (y)) d (f , , B) +d (f , , U) d (g, U, g (y))
= d (g, B, g (y)) d (f , , B)
Therefore, d (f , , B) = 0 and so for every z B, it follows z f () . Thus B f () .
On the other hand, f () cannot have points in both U and B because it is a connected
10.7. INTEGRATION AND THE DEGREE 283
set. Therefore f () B and this shows B = f (). Thus d (f , , B) = d (f , , y) for
each y B and the above formula shows this equals either 1 or 1 because the degree
is an integer. In the case where n = 1, the argument is similar but here you have 3
components in R
1
\ f () so there are more terms in the above sum although two of
them give 0. This proves the proposition.
10.7 Integration And The Degree
There is a very interesting application of the degree to integration [18]. Recall Lemma
10.2.5. I want to generalize this to the case where h :R
n
R
n
is only C
1
(R
n
; R
n
),
vanishing outside a bounded set. In the following proposition, let
m
be a symmetric
nonnegative mollier,
m
(x) m
n
(mx) , spt B(0, 1)
and let
be a mollier as 0
(x)
_
1
_
n
_
x
_
, spt B(0, 1)
will be a bounded open set.
Proposition 10.7.1 Let S h()
C
such that
dist (S, h()) > 0
where is a bounded open set and also let h be C
1
(R
n
; R
n
), vanishing outside some
bounded set. Then there exists
0
> 0 such that whenever 0 < <
0
d (h, , y) =
(h(x) y) det Dh(x) dx

for all y S.
Proof: Let
0
> 0 be small enough that for all y S,
B(y, 5
0
) h() = .
Now let
m
be a mollier as m with support in B
_
0, m
1
_
and let
h
m
h
m
.
Thus h
m
C
_
; R
n
_
and h
m
converges uniformly to h while Dh
m
converges uni-
formly to Dh. Denote by ||||
this norm dened by

||h||
sup {|h(x)| : x R
n
}
||Dh||
sup
_
max
i,j
h
i
(x)
x
j
: x R
n
_
It is nite because it is given that h vanishes o a bounded set. Choose M such that
for m M,
||h
m
h||
<
0
. (10.23)
Thus h
m
U
y
C
2
_
; R
n
_
for all y S where U
y
is dened on Page 259 and consists
of those functions f in C
_
; R
n
_
for which y / f ().
284 BROUWER DEGREE
For y S, let z B(y, ) where <
0
and suppose x , and k, m M. Then
for t [0, 1] ,
|(1 t) h
m
(x) +h
k
(x) t z| |h
m
(x) z| t |h
k
(x) h
m
(x)|
> 2
0
t2
0
0
showing that for each y S, B(y, ) ((1 t) h
m
+th
k
) () = . By Lemma 10.2.5,
for all y S,
(h
m
(x) y) det (Dh
m
(x)) dx =
(h
k
(x) y) det (Dh
k
(x)) dx (10.24)
for all k, m M. By this lemma again, which says that for small enough the integral
is constant and the denition of the degree in Denition 10.2.4,
d (y,, h
m
) =
(h
m
(x) y) det (Dh
m
(x)) dx (10.25)
for all small enough. For x , y S, and t [0, 1],
|(1 t) h(x) +h
m
(x) t y| |h(x) y| t |h(x) h
m
(x)|
> 3
0
t2
0
> 0
and so by Theorem 10.2.9, the part about homotopy, for each y S,
d (y,, h) = d (y,, h
m
) =
(h
m
(x) y) det (Dh
m
(x)) dx
whenever is small enough. Fix such an <
0
and use 10.24 to conclude the right side
of the above equation is independent of m > M. Now from the uniform convergence
noted above,
d (y,, h) = lim
m
(h
m
(x) y) det (Dh
m
(x)) dx
=
(h(x) y) det (Dh(x)) dx.

The next lemma is quite interesting. It says a C
1
function maps sets of measure
zero to sets of measure zero. This was proved earlier in Lemma 9.8.1 but I am stating
a special case here for convenience.
Lemma 10.7.2 Let h C
1
(R
n
; R
n
) and h vanishes o a bounded set. Let m
n
(A) =
0. Then h(A) also has measure zero.
Next is an interesting change of variables theorem. Let be a bounded open set
with the property that has measure zero and let h be C
1
and vanish o a bounded
set. Then from Lemma 10.7.2, h() also has measure zero.
Now suppose f C
c
_
h()
C
_
. By compactness, there are nitely many compo-
nents of h()
C
which have nonempty intersection with spt (f). From the Proposition
above,
f (y) d (y, , h) dy =
f (y) lim
0
(h(x) y) det Dh(x) dxdy

Actually, there exists an small enough that for all y spt (f) ,
lim
0
(h(x) y) det Dh(x) dx =
(h(x) y) det Dh(x) dx

= d (y, , h)
This is because spt (f) is at a positive distance from the compact set h()
C
so this
follows from Proposition 10.7.1. Therefore, for all small enough,
f (y) d (y, , h) dy =

f (y)
(h(x) y) det Dh(x) dxdy

=
det Dh(x)
f (y)
(h(x) y) dydx
=
det Dh(x) f (h(x)) dx +
det Dh(x)
(f (y) f (h(x)))
(h(x) y) dydx
Using the uniform continuity of f, you can now pass to a limit and obtain
f (y) d (y, , h) dy =
f (h(x)) det Dh(x) dx

This has essentially proved the following interesting Theorem.
Theorem 10.7.3 Let f C
c
_
h()
C
_
for a bounded open set where h is
in C
1
_
; R
n
_
. Suppose has measure zero. Then
f (y) d (y, , h) dy =
det (Dh(x)) f (h(x)) dx.

Proof: Both sides depend only on h restricted to and so the above results apply
and give the formula of the theorem for any h C
1
(R
n
; R
n
) which vanishes o a
bounded set which coincides with h on . This proves the lemma.
Lemma 10.7.4 If h C
1
_
, R
n
_
for a bounded connected open set with
having measure zero and h is one to one on . Then for any E a Borel set,
h()
X
E
(y) d (y, , h) dy =
det (Dh(x)) X
E
(h(x)) dx
Furthermore, o a set of measure zero, det (Dh(x)) has constant sign equal to the sign
of d (y, , h).
Proof: First here is a simple observation about the existence of a sequence of
nonnegative continuous functions which increase to X
O
for O open.
Let O be an open set and let H
j
denote those points y of O such that
dist
_
y, O
C
_
1/j, j = 1, 2, .
Then dene K
j
B(0,j) H
j
and
W
j
K
j
+B
_
0,
1
2j
_
.
where this means
_
k +b : k K
j
, b B
_
0,
1
2j
__
.
286 BROUWER DEGREE
Let
f
j
(y)
dist
_
y, W
C
j
_
dist (y, K
j
) + dist
_
y, W
C
j
_
Thus f
j
is nonnegative, increasing in j, has compact support in W
j
, is continuous, and
eventually f
j
(y) = 1 for all y O and f
j
(y) = 0 for all y / O. Thus lim
j
f
j
(y) =
X
O
(y).
Now let O h()
C
. Then from the above, let f
j
be as described above for the
open set O h() . (By invariance of domain, h() is open.)
h()
f
j
(y) d (y, , h) dy =
det (Dh(x)) f
j
(h(x)) dx
From Proposition 10.6.2, d (y, , h) either equals 1 or 1 for all y h(). Then by
the monotone convergence theorem on the left, using the fact that d (y, , h) is either
always 1 or always 1 and the dominated convergence theorem on the right, it follows
h()
X
O
(y) d (y, , h) dy =
det (Dh(x)) X
Oh()
(h(x)) dx
=
det (Dh(x)) X
O
(h(x)) dx
If y h() , then since h() h() is assumed to be empty, it follows y / h().
Therefore, the above formula holds for any O open.
Now let G denote those Borel sets of R
n
, E such that
h()
X
E
(y) d (y, , h) dy =
det (Dh(x)) X
E
(h(x)) dx
Then as shown above, G contains the system of open sets. Since |det (Dh(x))| is
bounded uniformly, it follows easily that if E G then E
C
G. This is because, since
R
n
is an open set,
h()
X
E
(y) d (y, , h) dy +
h()
X
E
C (y) d (y, , h) dy
=
h()
d (y, , h) dy =
det (Dh(x)) dx
=
det (Dh(x)) X
E
C (h(x)) dx +
det (Dh(x)) X
E
(h(x)) dx
Now cancelling
det (Dh(x)) X
E
(h(x))
from the right with
h()
X
E
(y) d (y, , h) dy
from the left, since E G, yields the desired result.
In addition, if {E
i
}
i=1
is a sequence of disjoint sets of G, the monotone convergence
and dominated convergence theorems imply the union of these disjoint sets is in G. By
the Lemma on systems, Lemma 9.1.2 it follows G equals the Borel sets.
Now consider the last claim. Suppose d (y, , h) = 1. Consider the compact set
E
{x : det (Dh(x)) }
and f (y) X
h(E)
(y) . Then from the rst part,
0
h()
X
h(E)
(y) (1) dy =
det Dh(x) X
E
(x) dx
m
n
(E
)
and so m
n
(E
) = 0. Therefore, if E is the set of x where det (Dh(x)) > 0, it

equals
k=1
E
1/k
, a set of measure zero. Thus o this set det Dh(x) 0. Similarly, if
d (y, , h) = 1, det Dh(x) 0 o a set of measure zero. This proves the lemma.
Theorem 10.7.5 Let f 0 and measurable for a bounded open connected
set where h is in C
1
_
; R
n
_
and is one to one on . Then
h()
f (y) d (y, , h) dy =
det (Dh(x)) f (h(x)) dx.

which amounts to the same thing as
h()
f (y) dy =
|det (Dh(x))| f (h(x)) dx.

Proof: First suppose has measure zero. From Proposition 10.6.2, d (y, , h)
either equals 1 or 1 for all y h() and det (Dh(x)) has the same sign as the sign of
the degree a.e. Suppose d (y, , h) = 1. By Theorem 9.9.10, if f 0 and is Lebesgue
measurable,
h()
f (y) dy =
|det (Dh(x))| f (h(x)) dx =
(1) det (Dh(x)) f (h(x)) dx

and so multiplying both sides by 1,
det (Dh(x)) f (h(x)) dx =
h()
f (y) (1) dy =
h()
f (y) d (y, , h) dy
The case where d (y, , h) = 1 is similar. This proves the theorem when has measure
zero.
Next I show it is not necessary to assume has measure zero. By Corollary 9.7.6
there exists a sequence of disjoint balls {B
i
} such that
m
n
( \
i=1
B
i
) = 0.
Since B
i
has measure zero, the above holds for each B
i
and so
h(Bi)
f (y) d (y, B
i
, h) dy =
Bi
det (Dh(x)) f (h(x)) dx
Since h is one to one, if y h(B
i
) , then y / h( \ B
i
) . Therefore, it follows from
Theorem 10.2.9 that
d (y, B
i
, h) = d (y, , h)
Thus
h(Bi)
f (y) d (y, , h) dy =
Bi
From Lemma 10.7.4 det (Dh(x)) has the same sign o a set of measure zero as the
constant sign of d (y, , h) . Therefore, using the monotone convergence theorem and
that h is one to one,
i=1
Bi
288 BROUWER DEGREE
=
i=1
Bi
i=1
h(Bi)
f (y) d (y, , h) dy
=
i=1
h(Bi)
f (y) d (y, , h) dy =
h(
i=1
Bi)
f (y) d (y, , h) dy
=
h()
f (y) d (y, , h) dy,
the last line following from Lemma 10.7.2. This proves the theorem.
10.8 Exercises
1. Show the Brouwer xed point theorem is equivalent to the nonexistence of a
continuous retraction onto the boundary of B(0, r).
2. Using the Jordan separation theorem, prove the invariance of domain theorem.
Hint: You might consider B(x, r) and show f maps the inside to one of two
components of R
n
\ f (B(x, r)) . Thus an open ball goes to some open set.
3. Give a version of Proposition 10.6.2 which is valid for the case where n = 1.
4. Suppose n > m. Does there exists a continuous one to one map, f which maps R
m
onto R
n
? This is a very interesting question because there do exist continuous maps
of [0, 1] which cover a square for example. Hint: First show that if K is compact
and f : K f (K) is one to one, then f
1
must be continuous. Now consider
the increasing sequence of compact sets
_
B(0, k)
_
k=1
whose union is R
m
and the
increasing sequence
_
f
_
B(0, k)
__
k=1
which you might assume covers R
n
. You
might use this to argue that if such a function exists, then f
1
must be continuous
and then apply Corollary 10.4.4.
5. Can there exists a one to one onto continuous map, f which takes the unit interval
to the unit disk? Hint: Think in terms of invariance of domain and use the hint
to Problem 4.
6. Consider the unit disk,
_
(x, y) : x
2
+y
2
1
_
D
and the annulus
_
(x, y) :
1
2
x
2
+y
2
1
_
A
Is it possible there exists a one to one onto continuous map f such that f (D) = A?
Thus D has no holes and A is really like D but with one hole punched out. Can
you generalize to dierent numbers of holes? Hint: Consider the invariance of
domain theorem. The interior of D would need to be mapped to the interior of A.
Where do the points of the boundary of A come from? Consider Theorem 5.3.5.
7. Suppose C is a compact set in R
n
which has empty interior and f : C R
n
is
one to one onto and continuous with continuous inverse. Could have nonempty
interior? Show also that if f is one to one and onto then if it is continuous, so
is f
1
.
10.8. EXERCISES 289
8. Let C denote the unit circle,
_
(x, y) R
2
: x
2
+y
2
= 1
_
. Suppose f : C R
2
is one to one and onto having continuous inverse. The Jordan curve theorem
says that under these conditions R
2
\ consists of two components, a bounded
component (called the inside) and an unbounded component (called the outside).
Prove the Jordan curve theorem using the Jordan separation theorem. Also show
that the boundary of each of these two components of R
2
\ is . Hint: Let U
1
and U
2
be the two components of R
2
\ . Explain why none of the limit points of
U
1
are are in U
2
so they must all be in . Similarly no limit point of U
2
can be in
U
1
. Use Problem 7 to show has empty interior. Use this observation to argue
U
1
U
2
= R
2
. Suppose p U
1
\ U
2
. Then explain why, since p / U
2
, it is not
a limit point of U
2
and so there is a ball B(p, r) which contains no points of U
2
.
Now where is B(p, r)? Doesnt this contradict p U
1
? Similarly U
2
\ U
1
=
and so U
1
= U
2
. Now justify the statement
U
1
U
2
= U
1
= U
2
.
Thus U
i
= .
9. Let K be a nonempty closed and convex subset of R
n
. Recall K is convex means
that if x, y K, then for all t [0, 1] , tx + (1 t) y K. Show that if x R
n
there exists a unique z K such that
|x z| = min {|x y| : y K} .
This z will be denoted as Px. Hint: First note you do not know K is compact.
Establish the parallelogram identity if you have not already done so,
|u v|
2
+|u +v|
2
= 2 |u|
2
+ 2 |v|
2
.
Then let {z
k
} be a minimizing sequence,
lim
k
|z
k
x|
2
= inf {|x y| : y K} .
Now using convexity, explain why
z
k
z
m
2
2
+
x
z
k
+z
m
2
2
= 2
x z
k
2
2
+ 2
x z
m
2
2
and then use this to argue {z
k
} is a Cauchy sequence. Then if z
i
works for i = 1, 2,
consider (z
1
+z
2
) /2 to get a contradiction.
10. In Problem 9 show that Px satises the following variational inequality.
(xPx) (yPx) 0
for all y K. Then show that |Px
1
Px
2
| |x
1
x
2
|. Hint: For the rst part
note that if y K, the function t |x(Px +t (yPx))|
2
achieves its minimum
on [0, 1] at t = 0. For the second part,
(x
1
Px
1
) (Px
2
Px
1
) 0, (x
2
Px
2
) (Px
1
Px
2
) 0.
Explain why
(x
2
Px
2
(x
1
Px
1
)) (Px
2
Px
1
) 0
and then use a some manipulations and the Cauchy Schwarz inequality to get the
desired inequality.
290 BROUWER DEGREE
11. Establish the Brouwer xed point theorem for any convex compact set in R
n
.
Hint: If K is a compact and convex set, let R be large enough that the closed
ball, D(0, R) K. Let P be the projection onto K as in Problem 10 above. If f
is a continuous map from K to K, consider f P. You want to show f has a xed
point in K.
12. Suppose D is a set which is homeomorphic to B(0, 1). This means there exists a
continuous one to one map, h such that h
_
B(0, 1)
_
= D such that h
1
is also
one to one. Show that if f is a continuous function which maps D to D then f has
a xed point. Now show that it suces to say that h is one to one and continuous.
In this case the continuity of h
1
is automatic. Sets which have the property that
continuous functions taking the set to itself have at least one xed point are said
to have the xed point property. Work Problem 6 using this notion of xed point
property. What about a solid ball and a donut?
13. There are many dierent proofs of the Brouwer xed point theorem. Let l be
a line segment. Label one end with A and the other end B. Now partition the
segment into n little pieces and label each of these partition points with either A
or B. Show there is an odd number of little segments with one end labeled with A
and the other labeled with B. If f :l l is continuous, use the fact it is uniformly
continuous and this little labeling result to give a proof for the Brouwer xed
point theorem for a one dimensional segment. Next consider a triangle. Label the
vertices with A, B, C and subdivide this triangle into little triangles, T
1
, , T
m
in such a way that any pair of these little triangles intersects either along an entire
edge or a vertex. Now label the unlabeled vertices of these little triangles with
either A, B, or C in any way. Show there is an odd number of little triangles
having their vertices labeled as A, B, C. Use this to show the Brouwer xed point
theorem for any triangle. This approach generalizes to higher dimensions and you
will see how this would take place if you are successful in going this far. This is an
outline of the Sperners lemma approach to the Brouwer xed point theorem. Are
there other sets besides compact convex sets which have the xed point property?
14. Using the denition of the derivative and the Vitali covering theorem, show that
if f C
1
_
U,R
n
_
and U has n dimensional measure zero then f (U) also has
measure zero. (This problem has little to do with this chapter. It is a review.)
15. Suppose is any open bounded subset of R
n
which contains 0 and that f : R
n
is continuous with the property that
f (x) x 0
for all x . Show that then there exists x such that f (x) = 0. Give a
similar result in the case where the above inequality is replaced with . Hint:
You might consider the function
h(t, x) tf (x) + (1 t) x.
16. Suppose is an open set in R
n
containing 0 and suppose that f : R
n
is
continuous and |f (x)| |x| for all x . Show f has a xed point in . Hint:
Consider h(t, x) t (x f (x)) +(1 t) x for t [0, 1] . If t = 1 and some x
is sent to 0, then you are done. Suppose therefore, that no xed point exists on
. Consider t < 1 and use the given inequality.
17. Let be an open bounded subset of R
n
and let f , g : R
n
both be continuous
such that
|f (x)| |g (x)| > 0
10.8. EXERCISES 291
for all x . Show that then
d (f g, , 0) = d (f , , 0)
Show that if there exists x f
1
(0) , then there exists x (f g)
1
(0). Hint:
You might consider h(t, x) (1 t) f (x)+t (f (x) g (x)) and argue 0 / h(t, )
for t [0, 1].
18. Let f : C C where C is the eld of complex numbers. Thus f has a real and
imaginary part. Letting z = x +iy,
f (z) = u(x, y) +iv (x, y)
Recall that the norm in C is given by |x +iy| =
x
2
+y
2
and this is the usual
norm in R
2
for the ordered pair (x, y) . Thus complex valued functions dened on
C can be considered as R
2
valued functions dened on some subset of R
2
. Such a
complex function is said to be analytic if the usual denition holds. That is
f
(z) = lim
h0
f (z +h) f (z)
h
.
In other words,
f (z +h) = f (z) +f
(z) h +o (h) (10.26)

at a point z where the derivative exists. Let f (z) = z
n
where n is a positive
integer. Thus z
n
= p (x, y) + iq (x, y) for p, q suitable polynomials in x and y.
Show this function is analytic. Next show that for an analytic function and u and
v the real and imaginary parts, the Cauchy Riemann equations hold.
u
x
= v
y
, u
y
= v
x
.
In terms of mappings show 10.26 has the form
_
u(x +h
1
, y +h
2
)
v (x +h
1
, y +h
2
)
_
=
_
u(x, y)
v (x, y)
_
+
_
u
x
(x, y) u
y
(x, y)
v
x
(x, y) v
y
(x, y)
__
h
1
h
2
_
+o(h)
=
_
u(x, y)
v (x, y)
_
+
_
u
x
(x, y) v
x
(x, y)
v
x
(x, y) u
x
(x, y)
__
h
1
h
2
_
+o(h)
where h =(h
1
, h
2
)
T
and h is given by h
1
+ih
2
. Thus the determinant of the above
matrix is always nonnegative. Letting B
r
denote the ball B(0, r) = B((0, 0) , r)
show
d (f, B
r
, 0) = n.
where f (z) = z
n
. In terms of mappings on R
2
,
f (x, y) =
_
u(x, y)
v (x, y)
_
.
Thus show
d (f , B
r
, 0) = n.
Hint: You might consider
g (z)
n
j=1
(z a
j
)
where the a
j
are small real distinct numbers and argue that both this function
and f are analytic but that 0 is a regular value for g although it is not so for f .
However, for each a
j
small but distinct d (f , B
r
, 0) = d (g, B
r
, 0).
292 BROUWER DEGREE
19. Using Problem 18, prove the fundamental theorem of algebra as follows. Let p (z)
be a nonconstant polynomial of degree n,
p (z) = a
n
z
n
+a
n1
z
n1
+
Show that for large enough r, |p (z)| > |p (z) a
n
z
n
| for all z B(0, r). Now
from Problem 17 you can conclude d (p, B
r
, 0) = d (f, B
r
, 0) = n where f (z) =
a
n
z
n
.
20. Generalize Theorem 10.7.5 to the situation where is not necessarily a connected
open set. You may need to make some adjustments on the hypotheses.
21. Suppose f : R
n
R
n
satises
|f (x) f (y)| |x y| , > 0,
Show that f must map R
n
onto R
n
. Hint: First show f is one to one. Then
use invariance of domain. Next show, using the inequality, that the points not in
f (R
n
) must form an open set because if y is such a point, then there can be no
sequence {f (x
n
)} converging to it. Finally recall that R
n
is connected.
Integration Of Dierential
Forms
11.1 Manifolds
Manifolds are sets which resemble R
n
locally. A manifold with boundary resembles half
of R
n
locally. To make this concept of a manifold more precise, here is a denition.
Denition 11.1.1 Let R
m
. A set, U, is open in if it is the intersection
of an open set from R
m
with . Equivalently, a set, U is open in if for every point,
x U, there exists > 0 such that if |x y| < and y , then y U. A set, H, is
closed in if it is the intersection of a closed set from R
m
with . Equivalently, a set,
H, is closed in if whenever, y is a limit point of H and y , it follows y H.
Recall the following denition.
Denition 11.1.2 Let V R
n
. C
k
_
V ; R
m
_
is the set of functions which are
restrictions to V of some function dened on R
n
which has k continuous derivatives
and compact support which has values in R
m
. When k = 0, it means the restriction to
V of continuous functions with compact support.
Denition 11.1.3 A closed and bounded subset of R
m
, , will be called an n
dimensional manifold with boundary, n 1, if there are nitely many sets U
i
, open in
and continuous one to one on U
i
functions, R
i
C
0
_
U
i
, R
n
_
such that R
i
U
i
is relatively
open in R
n
{u R
n
: u
1
0} , R
1
i
is continuous and one to one on R
i
_
U
i
_
. These
mappings, R
i
, together with the relatively open sets U
i
, are called charts and the totality
of all the charts, (U
i
, R
i
) just described is called an atlas for the manifold. Dene
int () {x : for some i, R
i
x R
n
<
}
where R
n
<
{u R
n
: u
1
< 0}. Also dene
{x : for some i, R
i
x R
n
0
}
where
R
n
0
{u R
n
: u
1
= 0}
and is called the boundary of . Note that if n = 1, R
n
0
is just the single point 0.
By convention, we will consider the boundary of such a 0 dimensional manifold to be
empty.
293
294 INTEGRATION OF DIFFERENTIAL FORMS
Theorem 11.1.4 Let and int () be as dened above. Then int () is open
in and is closed in . Furthermore, int () = , = int (), and for
n 1, is an n 1 dimensional manifold for which () = . The property of
being in int () or does not depend on the choice of atlas.
Proof: It is clear that = int (). First consider the claim that int () =
. Suppose this does not happen. Then there would exist x int (). Therefore,
there would exist two mappings R
i
and R
j
such that R
j
x R
n
0
and R
i
x R
n
<
with
x U
i
U
j
. Now consider the map, R
j
R
1
i
, a continuous one to one map from R
n
to R
n
having a continuous inverse. By continuity, there exists r > 0 small enough that,
R
1
i
B(R
i
x,r) U
i
U
j
.
Therefore, R
j
R
1
i
(B(R
i
x,r)) R
n
and contains a point on R

n
0
, R
j
x. However,
this cannot occur because it contradicts the theorem on invariance of domain, Theorem
10.4.3, which requires that R
j
R
1
i
(B(R
i
x,r)) must be an open subset of R
n
and
this one isnt because of the point on R
n
0
. Therefore, int () = as claimed. This
same argument shows that the property of being in int () or does not depend on
the choice of the atlas.
To verify that () = , let S
i
be the restriction of R
i
to U
i
. Thus
S
i
(x) = (0, (R
i
x)
2
, , (R
i
x)
n
)
and the collection of such points for x U
i
is an open bounded subset of
{u R
n
: u
1
= 0} ,
identied with R
n1
. S
i
( U
i
) is bounded because S
i
is the restriction of a contin-
uous function dened on R
m
and U
i
V
i
is contained in the compact set . Thus
if S
i
is modied slightly, to be of the form
S
i
(x) = ((R
i
x)
2
k
i
, , (R
i
x)
n
)
where k
i
is chosen suciently large enough that (R
i
(V
i
))
2
k
i
< 0, it follows that
{(V
i
, S
i
)} is an atlas for as an n 1 dimensional manifold such that every point of
is sent to to R
n1
<
and none gets sent to R
n1
0
. It follows is an n1 dimensional
manifold with empty boundary. In case n = 1, the result follows by denition of the
boundary of a 0 dimensional manifold.
Next consider the claim that int () is open in . If x int () , are all points of
which are suciently close to x also in int ()? If this were not true, there would
exist {x
n
} such that x
n
and x
n
x. Since there are only nitely many charts
of interest, this would imply the existence of a subsequence, still denoted by x
n
and a
single map, R
i
such that R
i
(x
n
) R
n
0
. But then R
i
(x
n
) R
i
(x) and so R
i
(x) R
n
0
showing x , a contradiction to int () = . Now it follows that is closed
in because = \ int (). This proves the theorem.
Denition 11.1.5 An n dimensional manifold with boundary, is a C
k
man-
ifold with boundary for some k 0 if
R
j
R
1
i
C
k
_
R
i
(U
i
U
j
); R
n
_
and R
1
i
C
k
_
R
i
U
i
; R
m
_
. It is called a continuous manifold with boundary if the
mappings, R
j
R
1
i
, R
1
i
, R
i
are continuous. In the case where is a C
k
, k 1
11.1. MANIFOLDS 295
manifold, it is called orientable if in addition to this there exists an atlas, (U
r
, R
r
),
such that whenever U
i
U
j
= ,
det
_
D
_
R
j
R
1
i
__
(u) > 0 for all u R
i
(U
i
U
j
) (11.1)
The mappings, R
i
R
1
j
are called the overlap maps. In the case where k = 0, the R
i
are only assumed continuous so there is no dierentiability available and in this case,
the manifold is oriented if whenever A is an open connected subset of int (R
i
(U
i
U
j
))
whose boundary has measure zero and separates R
n
into two components,
d
_
y,A, R
j
R
1
i
_
{1, 0} (11.2)
depending on whether y R
j
R
1
i
(A). An atlas satisfying 11.1 or more generally 11.2
is called an oriented atlas. By Lemma 10.7.4 and Proposition 10.6.2, this denition in
terms of the degree when applied to the situation of a C
k
manifold, gives the same thing
as the earlier denition in terms of the determinant of the derivative.
The advantage of using the degree in the above denition to dene orientation is
that it does not depend on any kind of dierentiability and since I am trying to relax
smoothness of the boundary, this is a good idea.
In calculus, you probably looked at piecewise smooth curves. The following is an
attempt to generalize this to the present situation.
Denition 11.1.6 In the above context, I will call a PC
1
manifold if it
is a C
0
manifold with charts (R
i
, U
i
) and there exists a closed set L such that
R
i
(L U
i
) is closed in R
i
(U
i
) and has m
n
measure zero, R
i
(L U
i
) is closed in
R
n1
and has m
n1
measure zero, and the following conditions hold.
R
1
i
C
1
(R
i
(U
i
\ L) ; R
m
) (11.3)
R
j
R
1
i
C
1
(R
i
((U
i
U
j
) \ L) ; R
n
) (11.4)
sup
_
DR
1
i
(u)
F
: u R
i
(U
i
\ L)
_
< (11.5)
where the norm is the Frobenius norm
||M||
F

_
_
i,j
|M
ij
|
2
_
_
1/2
.
The study of manifolds is really a generalization of something with which every-
one who has taken a normal calculus course is familiar. We think of a point in three
dimensional space in two ways. There is a geometric point and there are coordinates
associated with this point. There are many dierent coordinate systems which describe
a point. There are spherical coordinates, cylindrical coordinates and rectangular coor-
dinates to name the three most popular coordinate systems. These coordinates are like
the vector u. The point, x is like the geometric point although it is always assumed
here x has rectangular coordinates in R
m
for some m. Under fairly general conditions,
it can be shown there is no loss of generality in making such an assumption.
Now it will be convenient to use the following equivalent denition of orientable in
the case of a PC
1
manifold.
Proposition 11.1.7 Let be a PC
1
n dimensional manifold with boundary. Then
it is an orientable manifold if and only if there exists an atlas {R
i
, U
i
} such that for
each i, j
det D
_
R
i
R
1
j
_
(u) 0 a.e. u int
_
R
i
R
1
j
(U
j
U
i
)
_
(11.6)
If v = R
i
R
1
j
(u) , I will often write
(v
1
v
n
)
(u
1
u
n
)
det DR
i
R
1
j
(u)
Thus in this situation
(v1vn)
(u1un)
0.
Proof: Suppose rst the chart is an oriented chart so
d
_
v,A, R
i
R
1
j
_
= 1
whenever v int R
j
(A) where A is an open ball contained in R
i
R
1
j
(U
i
U
j
\ L) .
Then by Theorem 10.7.5, if E A is any Borel measurable set,
0
RiR
1
j
(A)
X
RiR
1
j
(E)
(v) 1dv =
A
det
_
D
_
R
i
R
1
j
_
(u)
_
X
E
(u) du
Since this is true for arbitrary E A, it follows det
_
D
_
R
i
R
1
j
_
(u)
_
0 a.e. u A
because if not so, then you could take E

_
u : det
_
D
_
R
i
R
1
j
_
(u)
_
<
_
and
for some > 0 this would have positive measure. Then the right side of the above is
negative while the left is nonnegative. By the Vitali covering theorem Corollary 9.7.6,
and the assumptions of PC
1
, there exists a sequence of disjoint open balls contained in
R
i
R
1
j
(U
i
U
j
\ L) , {A
k
} such that
int
_
R
i
R
1
j
(U
j
U
i
)
_
= L
k=1
A
k
and from the above, there exist sets of measure zero N
k
A
k
such that det D
_
R
i
R
1
j
_
(u)
0 for all u A
k
\ N
k
. Then det D
_
R
i
R
1
j
_
(u) 0 on int
_
R
i
R
1
j
(U
j
U
i
)
_
\
(L
k=1
N
k
) . This proves one direction. Now consider the other direction.
Suppose the condition det D
_
R
i
R
1
j
_
(u) 0 a.e. Then by Theorem 10.7.5
RiR
1
j
(A)
d
_
v, A, R
i
R
1
j
_
dv =
A
det
_
D
_
R
i
R
1
j
_
(u)
_
du 0
The degree is constant on the connected open set R
i
R
1
j
(A) . By Proposition 10.6.2,
the degree equals either 1 or 1. The above inequality shows it cant equal 1 and so
it must equal 1. This proves the proposition.
This shows it would be ne to simply use 11.6 as the denition of orientable in
the case of a PC
1
manifold and not bother with the deniton in terms of the degree.
However, that one is more general because it does not depend on any dierentiability.
I will use this result whenever convenient in what follows.
11.2 Some Important Measure Theory
11.2.1 Eggoros Theorem
Eggoros theorem says that if a sequence converges pointwise, then it almost converges
uniformly in a certain sense.
Theorem 11.2.1 (Egoro) Let (, F, ) be a nite measure space,
(() < )
11.2. SOME IMPORTANT MEASURE THEORY 297
and let f
n
, f be complex valued functions such that Re f
n
, Imf
n
are all measurable and
lim
n
f
n
() = f()
for all / E where (E) = 0. Then for every > 0, there exists a set,
F E, (F) < ,
such that f
n
converges uniformly to f on F
C
.
Proof: First suppose E = so that convergence is pointwise everywhere. It follows
then that Re f and Imf are pointwise limits of measurable functions and are therefore
measurable. Let E
km
= { : |f
n
() f()| 1/m for some n > k}. Note that
|f
n
() f ()| =
(Re f
n
() Re f ())
2
+ (Imf
n
() Imf ())
2
and so,
_
|f
n
f|
1
m
_
is measurable. Hence E
km
is measurable because
E
km
=
n=k+1
_
|f
n
f|
1
m
_
.
For xed m,
k=1
E
km
= because f
n
converges to f . Therefore, if there exists
k such that if n > k, |f
n
() f ()| <
1
m
which means / E
km
. Note also that
E
km
E
(k+1)m
.
Since (E
1m
) < , Theorem 7.3.2 on Page 157 implies
0 = (
k=1
E
km
) = lim
k
(E
km
).
Let k(m) be chosen such that (E
k(m)m
) < 2
m
and let
F =
m=1
E
k(m)m
.
Then (F) < because
(F)
m=1
_
E
k(m)m
_
<
m=1
2
m
=
Now let > 0 be given and pick m
0
such that m
1
0
< . If F
C
, then

m=1
E
C
k(m)m
.
Hence E
C
k(m0)m0
so
|f
n
() f()| < 1/m
0
<
for all n > k(m
0
). This holds for all F
C
and so f
n
converges uniformly to f on F
C
.
Now if E = , consider {X
E
Cf
n
}
n=1
. Each X
E
Cf
n
has real and imaginary parts
measurable and the sequence converges pointwise to X
E
f everywhere. Therefore, from
the rst part, there exists a set of measure less than , F such that on F
C
, {X
E
Cf
n
}
converges uniformly to X
E
Cf. Therefore, on (E F)
C
, {f
n
} converges uniformly to f.
11.2.2 The Vitali Convergence Theorem
The Vitali convergence theorem is a convergence theorem which in the case of a nite
measure space is superior to the dominated convergence theorem.
Denition 11.2.2 Let (, F, ) be a measure space and let S L
1
(). S is
uniformly integrable if for every > 0 there exists > 0 such that for all f S
|
E
fd| < whenever (E) < .
Lemma 11.2.3 If S is uniformly integrable, then |S| {|f| : f S} is uniformly
integrable. Also S is uniformly integrable if S is nite.
Proof: Let > 0 be given and suppose S is uniformly integrable. First suppose the
functions are real valued. Let be such that if (E) < , then
E
fd
<

2
for all f S. Let (E) < . Then if f S,
E
|f| d
E[f0]
(f) d +
E[f>0]
fd
=
E[f0]
fd
E[f>0]
fd
<

2
+

2
= .
In general, if Sis a uniformly integrable set of complex valued functions, the inequalities,
E
Re fd
E
fd
E
Imfd
E
fd
,
imply Re S {Re f : f S} and ImS {Imf : f S} are also uniformly integrable.
Therefore, applying the above result for real valued functions to these sets of functions,
it follows |S| is uniformly integrable also.
For the last part, is suces to verify a single function in L
1
() is uniformly inte-
grable. To do so, note that from the dominated convergence theorem,
lim
R
[|f|>R]
|f| d = 0.
Let > 0 be given and choose R large enough that
[|f|>R]
|f| d <

2
. Now let
(E) <

2R
. Then
E
|f| d =
E[|f|R]
|f| d +
E[|f|>R]
|f| d
< R(E) +

2
<

2
+

2
= .
The following gives a nice way to identify a uniformly integrable set of functions.
11.2. SOME IMPORTANT MEASURE THEORY 299
Lemma 11.2.4 Let S be a subset of L
1
(, ) where () < . Let t h(t) be a
continuous function which satises
lim
t
h(t)
t
=
Then S is uniformly integrable and bounded in L
1
() if
sup
_
h(|f|) d : f S
_
= N < .
Proof: First I show S is bounded in L
1
(; ) which means there exists a constant
M such that for all f S,
|f| d M.
From the properties of h, there exists R
n
such that if t R
n
, then h(t) nt. Therefore,
|f| d =
[|f|Rn]
|f| d +
[|f|<Rn]
|f| d
Letting n = 1,
|f| d
[|f|R1]
h(|f|) d +R
1
([|f| < R
1
])
N +R
1
() M.
Next let E be a measurable set. Then for every f S,
E
|f| d =
[|f|Rn]E
|f| d +
[|f|<Rn]E
|f| d
1
n
|f| d +R
n
(E)
N
n
+R
n
(E)
and letting n be large enough, this is less than
/2 +R
n
(E)
Now if (E) < /2R
n
, it follows that for all f S,
E
|f| d <
Letting h(t) = t
2
, it follows that if all the functions in S are bounded, then the
collection of functions is uniformly integrable.
The following theorem is Vitalis convergence theorem.
n
} be a uniformly integrable set of complex valued func-
tions, () < , and f
n
(x) f(x) a.e. where f is a measurable complex valued
function. Then f L
1
() and
lim
n
|f
n
f|d = 0. (11.7)
Proof: First it will be shown that f L
1
(). By uniform integrability, there exists
> 0 such that if (E) < , then
E
|f
n
| d < 1
for all n. By Egoros theorem, there exists a set, E of measure less than such that
on E
C
, {f
n
} converges uniformly. Therefore, for p large enough, and n > p,
E
C
|f
p
f
n
| d < 1
which implies
E
C
|f
n
| d < 1 +
|f
p
| d.
Then since there are only nitely many functions, f
n
with n p, there exists a constant,
M
1
such that for all n,
E
C
|f
n
| d < M
1
.
But also,
|f
m
| d =
E
C
|f
m
| d +
E
|f
m
|
M
1
+ 1 M.
Therefore, by Fatous lemma,
|f| d lim inf

n
|f
n
| d M,
showing that f L
1
as hoped.
Now S {f} is uniformly integrable so there exists
1
> 0 such that if (E) <
1
,
then
E
|g| d < /3 for all g S {f}.
By Egoros theorem, there exists a set, F with (F) <
1
such that f
n
converges
uniformly to f on F
C
. Therefore, there exists N such that if n > N, then
F
C
|f f
n
| d <

3
.
It follows that for n > N,
|f f
n
| d
F
C
|f f
n
| d +
F
|f| d +
F
|f
n
| d
<

3
+

3
+

3
= ,
which veries 11.7.
11.3 The Binet Cauchy Formula
The Binet Cauchy formula is a generalization of the theorem which says the determinant
of a product is the product of the determinants. The situation is illustrated in the
following picture where A, B are matrices.
B A
11.4. THE AREA MEASURE ON A MANIFOLD 301
Theorem 11.3.1 Let A be an n m matrix with n m and let B be a mn
matrix. Also let A
i
i = 1, , C (n, m)
be the m m submatrices of A which are obtained by deleting n m rows and let B
i
be the m m submatrices of B which are obtained by deleting corresponding n m
columns. Then
det (BA) =
C(n,m)
k=1
det (B
k
) det (A
k
)
Proof: This follows from a computation. By Corollary 3.5.6 on Page 40, det (BA) =
1
m!
(i1im)
(j1jm)
sgn (i
1
i
m
) sgn (j
1
j
m
) (BA)
i1j1
(BA)
i2j2
(BA)
imjm
1
m!
(i1im)
(j1jm)
sgn (i
1
i
m
) sgn (j
1
j
m
)
n
r1=1
B
i1r1
A
r1j1
n
r2=1
B
i2r2
A
r2j2

n
rm=1
B
imrm
A
rmjm
Now denote by I
k
one of the r subsets of {1, , n} . Thus there are C (n, m) of these.
=
C(n,m)
k=1
{r1, ,rm}=I
k
1
m!
(i1im)
(j1jm)
sgn (i
1
i
m
) sgn (j
1
j
m
)
B
i1r1
A
r1j1
B
i2r2
A
r2j2
B
imrm
A
rmjm
=
C(n,m)
k=1
{r1, ,rm}=I
k
1
m!
(i1im)
sgn (i
1
i
m
) B
i1r1
B
i2r2
B
imrm

(j1jm)
sgn (j
1
j
m
) A
r1j1
A
r2j2
A
rmjm
=
C(n,m)
k=1
{r1, ,rm}=I
k
1
m!
sgn (r
1
r
m
)
2
det (B
k
) det (A
k
) B
=
C(n,m)
k=1
det (B
k
) det (A
k
)
since there are m! ways of arranging the indices {r
1
, , r
m
}.
11.4 The Area Measure On A Manifold
It is convenient to specify a surface measure on a manifold. This concept is a little
easier because you dont have to worry about orientation. It will involve the following
denition.
Denition 11.4.1 Let (U
i
, R
i
) be an atlas for an n dimensional PC
1
manifold
. Also let {
i
}
p
i=1
be a C
partition of unity, spt

i
U
i
. Then for E a Borel subset
of , dene
n
(E)
p
i=1
RiUi
i
_
R
1
i
(u)
_
X
E
_
R
1
i
(u)
_
J
i
(u) du
where
J
i
(u)
_
det
_
DR
1
i
(u)
DR
1
i
(u)
__
1/2
By the Binet Cauchy theorem, this equals
_
_

i1, ,in
_
(x
i1
x
in
)
(u
1
u
n
)
(u)
_
2
_
_
1/2
where the sum is taken over all increasing strings of n indices (i
1
, , i
n
) and
(x
i1
x
in
)
(u
1
u
n
)
(u) det
_
_
_
_
_
x
i1,u1
x
i1,u2
x
i1,un
x
i2,u1
x
i2,u2
x
i2,u2
.
.
.
.
.
.
.
.
.
.
.
.
x
in,u1
x
in,u2
x
in,un
_
_
_
_
_
(u) (11.8)
Suppose (V
j
, S
j
) is another atlas and you have
v = S
j
R
1
i
(u) (11.9)
for u R
i
((V
j
U
i
) \ L) . Then R
1
i
= S
1
j

_
S
j
R
1
i
_
and so by the chain rule,
DS
1
j
(v) D
_
S
j
R
1
i
_
(u) = DR
1
i
(u)
Therefore,
J
i
(u) =
_
det
_
DR
1
i
(u)
DR
1
i
(u)
__
1/2
=
_
_
_det
_
_
_
nn
..
D
_
S
j
R
1
i
_
(u)
nn
..
DS
1
j
(v)
DS
1
j
(v)
nn
..
D
_
S
j
R
1
i
_
(u)
_
_
_
_
_
_
1/2
=
det
_
D
_
S
j
R
1
i
_
(u)
_
J
j
(v) (11.10)
Similarly
J
j
(v) =
det
_
D
_
R
i
S
1
j
_
(v)
_
J
i
(u) . (11.11)
In the situation of 11.9, it is convenient to use the notation
(v
1
v
n
)
(u
1
u
n
)
det
_
D
_
S
j
R
1
i
_
(u)
_
and this will be used occasionally below.
It is necessary to show the above denition is well dened.
Theorem 11.4.2 The above denition of surface measure is well dened. That
is, suppose is an n dimensional orientable PC
1
manifold with boundary and let
{(U
i
, R
i
)}
p
i=1
and {(V
j
, S
j
)}
q
j=1
be two atlass of . Then for E a Borel set, the com-
putation of
n
(E) using the two dierent atlass gives the same thing. This denes a
Borel measure on . Furthermore, if E U
i
,
n
(E) reduces to
RiUi
X
E
_
R
1
i
(u)
_
J
i
(u) du
Also
n
(L) = 0.
11.4. THE AREA MEASURE ON A MANIFOLD 303
Proof: Let {
i
} be a partition of unity as described in Lemma 11.5.3 which is
associated with the atlas (U
i
, R
i
) and let {
i
} be a partition of unity associated in the
same manner with the atlas (V
i
, S
i
). First note the following.
RiUi
i
_
R
1
i
(u)
_
X
E
_
R
1
i
(u)
_
J
i
(u) du
=
q
j=1
Ri(UiVj)
j
_
R
1
i
(u)
_
i
_
R
1
i
(u)
_
X
E
_
R
1
i
(u)
_
J
i
(u) du
=
q
j=1
int Ri(UiVj\L)
j
_
R
1
i
(u)
_
i
_
R
1
i
(u)
_
X
E
_
R
1
i
(u)
_
J
i
(u) du
The reason this can be done is that points not on the interior of R
i
(U
i
V
j
) are on
the plane u
1
= 0 which is a set of measure zero and by assumption R
i
(L U
i
V
j
)
has measure zero. Of course the above determinants in the denition of J
i
(u) in the
integrand are not even dened on the set of measure zero R
i
(L U
i
) . It follows the
denition of
n
(E) in terms of the atlas {(U
i
, R
i
)}
p
i=1
and specied partition of unity
is
p
i=1
q
j=1
int Ri(UiVj\L)
j
_
R
1
i
(u)
_
i
_
R
1
i
(u)
_
X
E
_
R
1
i
(u)
_
J
i
(u) du
By the change of variables formula, Theorem 9.9.4, this equals
p
i=1
q
j=1
(SjR
1
i
)(int Ri(UiVj\L))
j
_
S
1
j
(v)
_
i
_
S
1
j
(v)
_
X
E
_
S
1
j
(v)
_
J
i
(u)
det
_
R
i
S
1
j
(v)
_
dv
The integral is unchanged if it is taken over S
j
(U
i
V
j
) and so, by 11.11, this equals
p
i=1
q
j=1
Sj(UiVj)
j
_
S
1
j
(v)
_
i
_
S
1
j
(v)
_
X
E
_
S
1
j
(v)
_
J
j
(v) dv
=
q
j=1
Sj(UiVj)
j
_
S
1
j
(v)
_
X
E
_
S
1
j
(v)
_
J
j
(v) dv
which equals the denition of
n
(E) taken with respect to the other atlas and partition
of unity. Thus the denition is well dened. This also has shown that the partition of
unity can be picked at will.
It remains to verify the claim. First suppose E = K a compact subset of U
i
. Then
using Lemma 11.5.3 there exists a partition of unity such that
k
= 1 on K. Consider
the sum used to dene
n
(K) ,
p
i=1
RiUi
i
_
R
1
i
(u)
_
X
K
_
R
1
i
(u)
_
J
i
(u) du
Unless R
1
i
(u) K, the integrand equals 0. Assume then that R
1
i
(u) K. If
i = k,
i
_
R
1
i
(u)
_
= 0 because
k
_
R
1
i
(u)
_
= 1 and these functions sum to 1.
Therefore, the above sum reduces to
R
k
U
k
X
K
_
R
1
k
(u)
_
J
k
(u) du.
Next consider the general case. By Theorem 7.4.6 the Borel measure
n
is regular. Also
Lebesgue measure is regular. Therefore, there exists an increasing sequence of compact
subsets of E, {K
r
}
k=1
such that
lim
r
n
(K
r
) =
n
(E)
Then letting F =
r=1
K
r
, the monotone convergence theorem implies
n
(E) = lim
r
n
(K
r
) =
R
k
U
k
X
F
_
R
1
k
(u)
_
J
k
(u) du
R
k
U
k
X
E
_
R
1
k
(u)
_
J
k
(u) du
Next take an increasing sequence of compact sets contained in R
k
(E) such that
lim
r
m
n
(K
r
) = m
n
(R
k
(E)) .
Thus
_
R
1
k
(K
r
)
_
r=1
is an increasing sequence of compact subsets of E. Therefore,
n
(E) lim
r
n
_
R
1
k
(K
r
)
_
= lim
r
R
k
U
k
X
Kr
(u) J
k
(u) du
=
R
k
U
k
X
R
k
(E)
(u) J
k
(u) du
=
R
k
U
k
X
E
_
R
1
k
(u)
_
J
k
(u) du
Thus
n
(E) =
R
k
U
k
X
E
_
R
1
k
(u)
_
J
k
(u) du
as claimed.
So what is the measure of L? By denition it equals
n
(L)
p
i=1
RiUi
i
_
R
1
i
(u)
_
X
L
_
R
1
i
(u)
_
J
i
(u) du
=
p
i=1
RiUi
i
_
R
1
i
(u)
_
X
Ri(LUi)
(u) J
i
(u) du = 0
because by assumption, the measure of each R
i
(L U
i
) is zero. This proves the theo-
rem.
11.5 Integration Of Dierential Forms On Manifolds
This section presents the integration of dierential forms on manifolds. This topic is a
higher dimensional version of what is done in calculus in nding the work done by a
force eld on an object which moves over some path. There you evaluated line integrals.
Dierential forms are just a higher dimensional version of this idea and it turns out they
are what it makes sense to integrate on manifolds. The following lemma, on Page 246
used in establishing the denition of the degree and in giving a proof of the Brouwer xed
point theorem is also a fundamental result in discussing the integration of dierential
forms.
11.5. INTEGRATION OF DIFFERENTIAL FORMS ON MANIFOLDS 305
Lemma 11.5.1 Let g : U V be C
2
where U and V are open subsets of R
n
. Then
n
j=1
(cof (Dg))
ij,j
= 0,
where here (Dg)
ij
g
i,j

gi
xj
.
Recall Proposition 10.6.2.
Proposition 11.5.2 Let be an open connected bounded set in R
n
such that R
n
\
consists of two, three if n = 1, connected components. Let f C
_
; R
n
_
be con-
tinuous and one to one. Then f () is the bounded component of R
n
\ f () and for
y f () , d (f , , y) either equals 1 or 1.
Also recall the following fundamental lemma on partitions of unity in Lemma 9.5.15
and Corollary 9.5.14.
Lemma 11.5.3 Let K be a compact set in R
n
and let {U
i
}
m
i=1
be an open cover of
K. Then there exist functions,
k
C
c
(U
i
) such that
i
U
i
and for all x K,
m
i=1
i
(x) = 1.
If K is a compact subset of U
1
(U
i
)there exist such functions such that also
1
(x) = 1
(
i
(x) = 1) for all x K.
With the above, what follows is the denition of what a dierential form is and how
to integrate one.
Denition 11.5.4 Let I denote an ordered list of n indices taken from the set,
{1, , m}. Thus I = (i
1
, , i
n
). It is an ordered list because the order matters. A
dierential form of order n in R
m
is a formal expression,
=
I
a
I
(x) dx
I
where a
I
is at least Borel measurable dx
I
is short for the expression
dx
i1
dx
in
,
and the sum is taken over all ordered lists of indices taken from the set, {1, , m}.
For an orientable n dimensional manifold with boundary, let {(U
i
, R
i
)} be an oriented
atlas for . Each U
i
is the intersection of an open set in R
m
, O
i
, with and so there
exists a C
partition of unity subordinate to the open cover, {O

i
} which sums to 1 on
. Thus
i
C
c
(O
i
), has values in [0, 1] and satises
i
(x) = 1 for all x .
Dene
i=1
RiUi
i
_
R
1
i
(u)
_
a
I
_
R
1
i
(u)
_
(x
i1
x
in
)
(u
1
u
n
)
du (11.12)
Note
(x
i1
x
in
)
(u
1
u
n
)
,
given by 11.8 is not dened on R
i
(U
i
L) but this is assumed a set of measure zero so
it is not important in the integral.
Of course there are all sorts of questions related to whether this denition is well
dened. What if you had a dierent atlas and a dierent partition of unity? Would
change? In general, the answer is yes. However, there is a sense in which the integral of
a dierential form is well dened. This involves the concept of orientation which looks
a lot like the concept of an oriented manifold.
Denition 11.5.5 Suppose is an n dimensional orientable manifold with
boundary and let (U
i
, R
i
) and (V
i
, S
i
) be two oriented atlass of . They have the same
orientation if for all open connected sets A S
j
(V
j
U
i
) with A having measure zero
and separating R
n
into two components,
d
_
u, R
i
S
1
j
, A
_
{0, 1}
depending on whether u R
i
S
1
j
(A).
The above denition of
is well dened in the sense that any two atlass which

have the same orientation deliver the same value for this symbol.
Theorem 11.5.6 Suppose is an n dimensional orientable PC
1
manifold with
boundary and let (U
i
, R
i
) and (V
i
, S
i
) be two oriented atlass of . Suppose the two atlass
have the same orientation. Then if
is computed with respect to the two atlass the

same number is obtained. Assume each a
I
is Borel measurable and bounded.
1
Proof: Let {
i
} be a partition of unity as described in Lemma 11.5.3 which is
associated with the atlas (U
i
, R
i
) and let {
i
} be a partition of unity associated in the
same manner with the atlas (V
i
, S
i
). Then the denition using the atlas {(U
i
, R
i
)} is
p
i=1
RiUi
i
_
R
1
i
(u)
_
a
I
_
R
1
i
(u)
_
(x
i1
x
in
)
(u
1
u
n
)
du (11.13)
=
p
i=1
q
j=1
Ri(UiVj)
j
_
R
1
i
(u)
_
i
_
R
1
i
(u)
_
a
I
_
R
1
i
(u)
_
(x
i1
x
in
)
(u
1
u
n
)
du
=
p
i=1
q
j=1
int Ri(UiVj\L)
j
_
R
1
i
(u)
_
i
_
R
1
i
(u)
_
a
I
_
R
1
i
(u)
_
(x
i1
x
in
)
(u
1
u
n
)
du
The reason this can be done is that points not on the interior of R
i
(U
i
V
j
) are on
the plane u
1
= 0 which is a set of measure zero and by assumption R
i
(L U
i
V
j
) has
measure zero. Of course the above determinant in the integrand is not even dened on
R
i
(L U
i
V
j
) . By the change of variables formula Theorem 9.9.10 and Proposition
11.1.7, this equals
p
i=1
q
j=1
int Sj(UiVj\L)
j
_
S
1
j
(v)
_
i
_
S
1
j
(v)
_
a
I
_
S
1
j
(v)
_
(x
i1
x
in
)
(v
1
v
n
)
dv
=
p
i=1
q
j=1
Sj(UiVj)
j
_
S
1
j
(v)
_
i
_
S
1
j
(v)
_
a
I
_
S
1
j
(v)
_
(x
i1
x
in
)
(v
1
v
n
)
dv
=
q
j=1
Sj(Vj)
j
_
S
1
j
(v)
_
a
I
_
S
1
j
(v)
_
(x
i1
x
in
)
(v
1
v
n
)
dv
which is the denition
using the other atlas {(V

j
, S
j
)} and partition of unity. This
proves the theorem.
1
This is so issues of existence for the various integrals will not arrise. This is leading to Stokes
theorem in which even more will be assumed on a
I
.
11.6. STOKES THEOREM AND THE ORIENTATION OF 307
11.5.1 The Derivative Of A Dierential Form
The derivative of a dierential form is dened next.
Denition 11.5.7 Let =
I
a
I
(x) dx
i1
dx
in1
be a dierential form
of order n1 where a
I
is C
1
. Then dene d, a dierential form of order n by replacing
a
I
(x) with
da
I
(x)
m
k=1
a
I
(x)
x
k
dx
k
(11.14)
and putting a wedge after the dx
k
. Therefore,
d
I
m
k=1
a
I
(x)
x
k
dx
k
dx
i1
dx
in1
. (11.15)
11.6 Stokes Theorem And The Orientation Of
Here will be an n dimensional orientable PC
1
manifold with boundary in R
m
. Let
an oriented atlas for it be {U
i
, R
i
}
p
i=1
and let a C
partition of unity be {
i
}
p
i=1
. Also
let
=
I
a
I
(x) dx
i1
dx
in1
be a dierential form such that a
I
is C
1
_
_
. Since
i
(x) = 1 on ,
d =
I
m
k=1
p
j=1
j
a
I
_
x
k
(x) dx
k
dx
i1
dx
in1
because the right side equals
I
m
k=1
p
j=1
j
x
k
a
I
(x) dx
k
dx
i1
dx
in1
+
I
m
k=1
p
j=1
a
I
x
k
j
(x) dx
k
dx
i1
dx
in1
I
a
I
(x)
m
k=1
x
k
_
_
_
_
_
=1
..
p
j=1
j
_
_
_
_
_
dx
k
dx
i1
dx
in1
+
I
m
k=1
a
I
x
k
p
j=1
j
(x) dx
k
dx
i1
dx
in1
=
I
m
k=1
a
I
x
k
(x) dx
k
dx
i1
dx
in1
d
It follows
d =
I
m
k=1
p
j=1
Rj(Uj)
j
a
I
_
x
k
_
R
1
j
(u)
_
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
du
=
I
m
k=1
p
j=1
Rj(Uj)
j
a
I
_
x
k
_
R
1
j
(u)
_
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
du+
I
m
k=1
p
j=1
Rj(Uj)
j
a
I
_
x
k
_
R
1
j
(u)
_
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
du
I
m
k=1
p
j=1
Rj(Uj)
j
a
I
_
x
k
_
R
1
j
(u)
_
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
du (11.16)
Here
R
1
j
(u) R
1
j

(u)
for
a mollier and x
i
is the i
th
component mollied. Thus by Lemma 9.5.7, this
function with the subscript is innitely dierentiable. The last two expressions in
11.16 sum to e () which converges to 0 as 0. Here is why.
j
a
I
_
x
k
_
R
1
j
(u)
_
j
a
I
_
x
k
_
R
1
j
(u)
_
a.e.
because of the pointwise convergence of R
1
j
to R
1
j
which follows from Lemma 9.5.7.
In addition to this,
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)

_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
a.e.
because of this lemma used again on each of the component functions. This convergence
happens on R
j
(U
j
\ L) for each j. Thus e () is a nite sum of integrals of integrands
which converges to 0 a.e. By assumption 11.5, these integrands are uniformly integrable
and so it follows from the Vitali convergence theorem, Theorem 11.2.5, the integrals
converge to 0.
Then 11.16 equals
=
I
m
k=1
p
j=1
Rj(Uj)
j
a
I
_
x
k
_
R
1
j
(u)
_
m
l=1
x
k
u
l
A
1l
du +e ()
where A
1l
is the 1l
th
cofactor for the determinant
_
x
k
, x
i1
x
in1
_
(u
1
, , u
n
)
which is determined by a particular I. I am suppressing the and I for the sake of
notation. Then the above reduces to
=
I
p
j=1
Rj(Uj)
n
l=1
A
1l
m
k=1
j
a
I
_
x
k
_
R
1
j
(u)
_
x
k
u
l
du +e ()
=
I
p
j=1
n
l=1
Rj(Uj)
A
1l
u
l
_
j
a
I
R
1
j
_
(u) du +e () (11.17)
(Note l goes up to n not m.) Recall R
j
(U
j
) is relatively open in R
n
. Consider the
integral where l > 1. Integrate rst with respect to u
l
. In this case the boundary term
vanishes because of
j
and you get
Rj(Uj)
A
1l,l
_
j
a
I
R
1
j
_
(u) du (11.18)
11.6. STOKES THEOREM AND THE ORIENTATION OF 309
Next consider the case where l = 1. Integrating rst with respect to u
1
, the term reduces
to
RjVj
j
a
I
R
1
j
(0, u
2
, , u
n
) A
11
du
1
Rj(Uj)
A
11,1
_
j
a
I
R
1
j
_
(u) du (11.19)
where R
j
V
j
is an open set in R
n1
consisting of
_
(u
2
, , u
n
) R
n1
: (0, u
2
, , u
n
) R
j
(U
j
)
_
and du
1
represents du
2
du
3
du
n
on R
j
V
j
for short. Thus V
j
is just the part of
which is in U
j
and the mappings S
1
j
given on R
j
V
j
= R
j
(U
j
) by
S
1
j
(u
2
, , u
n
) R
1
j
(0, u
2
, , u
n
)
are such that {(S
j
, V
j
)} is an atlas for . Then if 11.18 and 11.19 are placed in 11.17,
it follows from Lemma 11.5.1 that 11.17 reduces to
I
p
j=1
RjVj
j
a
I
R
1
j
(0, u
2
, , u
n
) A
11
du
1
+e ()
Now as before, each x
s
/u
r
converges pointwise a.e. to x
s
/u
r
, o R
j
(V
j
L)
assumed to be a set of measure zero, and the integrands are bounded. Using the Vitali
convergence theorem again, pass to a limit as 0 to obtain
I
p
j=1
RjVj
j
a
I
R
1
j
(0, u
2
, , u
n
) A
11
du
1
=
I
p
j=1
SjVj
j
a
I
S
1
j
(u
2
, , u
n
) A
11
du
1
=
I
p
j=1
SjVj
j
a
I
S
1
j
(u
2
, , u
n
)
_
x
i1
x
in1
_
(u
2
, , u
n
)
(0, u
2
, , u
n
) du
1
(11.20)
This of course is the denition of
provided {S
j
, V
j
} is an oriented atlas. Note
the integral is well dened because of the assumption that R
i
(L U
i
) has m
n1
measure zero. That is orientable and that this atlas is an oriented atlas is shown
next. I will write u
1
(u
2
, , u
n
).
What if spt a
I
K U
i
U
j
for each I? Then using Lemma 11.5.3 it follows that
d =
Sj(VjVj)
a
I
S
1
j
(u
2
, , u
n
)
_
x
i1
x
in1
_
(u
2
, , u
n
)
(0, u
2
, , u
n
) du
1
This is done by using a partition of unity which has the property that
j
equals 1 on
K which forces all the other
k
to equal zero there. Using the same trick involving a
judicious choice of the partition of unity,
d is also equal to
Si(VjVj)
a
I
S
1
i
(v
2
, , v
n
)
_
x
i1
x
in1
_
(v
2
, , v
n
)
(0, v
2
, , v
n
) dv
1
Since S
i
(L U
i
) , S
j
(L U
j
) have measure zero, the above integrals may be taken over
S
j
(V
j
V
j
\ L) , S
i
(V
j
V
j
\ L)
respectively. Also these are equal, both being
d. To simplify the notation, let

I
denote the projection onto the components corresponding to I. Thus if I = (i
1
, , i
n
) ,
I
x (x
i1
, , x
in
) .
Then writing in this simpler notation, the above would say
Sj(VjVj\L)
a
I
S
1
j
(u
1
) det D
I
S
1
j
(u
1
) du
1
=
Si(VjVj\L)
a
I
S
1
i
(v
1
) det D
I
S
1
i
(v
1
) dv
1
and both equal to
d. Thus using the change of variables formula, Theorem 9.9.10,

it follows the second of these equals
Sj(VjVj\L)
a
I
S
1
j
(u
1
) det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
_
det D
_
S
i
S
1
j
_
(u
1
)
du
1
(11.21)
I want to argue det D
_
S
i
S
1
j
_
(u
1
) 0. Let A be the open subset of S
j
(V
j
V
j
\ L)
on which for > 0,
det D
_
S
i
S
1
j
_
(u
1
) < (11.22)
I want to show A = so assume A is nonempty. If this is the case, we could consider
an open ball contained in A. To simplify notation, assume A is an open ball. Letting
f
I
be a smooth function which vanishes o a compact subset of S
1
j
(A) the above
argument and the chain rule imply
Sj(VjVj\L)
f
I
S
1
j
(u
1
) det D
I
S
1
j
(u
1
) du
1
=
A
f
I
S
1
j
(u
1
) det D
I
S
1
j
(u
1
) du
1
A
f
I
S
1
j
(u
1
) det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
_
det D
_
S
i
S
1
j
_
(u
1
) du
1
Now from 11.21, this equals
=
A
f
I
S
1
j
(u
1
) det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
_
det D
_
S
i
S
1
j
_
(u
1
) du
1
and consequently
0 = 2
A
f
I
S
1
j
(u
1
) det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
_
det D
_
S
i
S
1
j
_
(u
1
) du
1
Now for each I, let
_
f
k
I
S
1
j
_
k=1
be a sequence of bounded functions having compact
support in A which converge pointwise to det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
_
. Then it follows
from the Vitali convergence theorem, one can pass to the limit and obtain
0 = 2
I
_
det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
__
2
det D
_
S
i
S
1
j
_
(u
1
) du
1
2
I
_
det D
I
S
1
i
_
S
i
S
1
j
(u
1
)
__
2
du
1
11.7. GREENS THEOREM, AN EXAMPLE 311
Since the integrand is continuous, this would require
det D
I
S
1
i
(v
1
) 0 (11.23)
for each I and for each v
1
S
i
S
1
j
(A), an open set which must have positive measure.
But since it has positive measure, it follows from the change of variables theorem and
the chain rule,
D
_
S
j
S
1
i
_
(v
1
) =
nm
..
DS
j
_
S
1
i
(v
1
)
_
mn
..
DS
1
i
(v
1
)
cannot be identically 0. By the Binet Cauchy theorem, at least some
D
I
S
1
i
(v
1
) = 0
contradicting 11.23. Thus A = and since > 0 was arbitrary, this shows
det D
_
S
i
S
1
j
_
(u
1
) 0.
Hence this is an oriented atlas as claimed. This proves the theorem.
Theorem 11.6.1 Let be an oriented PC
1
manifold and let
=
I
a
I
(x) dx
i1
dx
in1
.
where each a
I
is C
1
_
_
. For {U
j
, R
j
}
p
j=0
an oriented atlas for where R
j
(U
j
) is a
relatively open set in
{u R
n
: u
1
0} ,
dene an atlas for , {V
j
, S
j
} where V
j
U
j
and S
j
is just the restriction of R
j
to V
j
. Then this is an oriented atlas for and
d
where the two integrals are taken with respect to the given oriented atlass.
11.7 Greens Theorem, An Example
Greens theorem is a well known result in calculus and it pertains to a region in the
plane. I am going to generalize to an open set in R
n
with suciently smooth boundary
using the methods of dierential forms described above.
11.7.1 An Oriented Manifold
A bounded open subset, , of R
n
, n 2 has PC
1
boundary and lies locally on one side
of its boundary if it satises the following conditions.
For each p \ , there exists an open set, Q, containing p, an open interval
(a, b), a bounded open set B R
n1
, and an orthogonal transformation R such that
det R = 1,
(a, b) B = RQ,
and letting W = Q ,
RW = {u R
n
: a < u
1
< g (u
2
, , u
n
) , (u
2
, , u
n
) B}
g (u
2
, , u
n
) < b for (u
2
, , u
n
) B. Also g vanishes outside some compact set in
R
n1
and g is continuous.
R( Q) = {u R
n
: u
1
= g (u
2
, , u
n
) , (u
2
, , u
n
) B} .
Note that nitely many of these sets Q cover because is compact. Assume there
exists a closed subset of , L such that the closed set S
Q
dened by
{(u
2
, , u
n
) B : (g (u
2
, , u
n
) , u
2
, , u
n
) R(L Q)} (11.24)
has m
n1
measure zero. g C
1
(B \ S
Q
) and all the partial derivatives of g are uni-
formly bounded on B \ S
Q
. The following picture describes the situation. The pointy
places symbolize the set L.
x
W
Q
-
R
R(W)
a
b
R(Q)
u
Dene P
1
: R
n
R
n1
by
P
1
u (u
2
, , u
n
)
and : R
n
R
n
given by
u u g (P
1
u) e
1
u g (u
2
, , u
n
) e
1
(u
1
g (u
2
, , u
n
) , u
2
, , u
n
)
Thus is invertible and
1
u = u +g (P
1
u) e
1
(u
1
+g (u
2
, , u
n
) , u
2
, , u
n
)
For x Q, it follows the rst component of Rx is g (P
1
(Rx)) . Now dene R :W
R
n
as
u Rx Rx g (P
1
(Rx)) e
1
Rx
and so it follows
R
1
= R
1
.
These mappings R involve rst a rotation followed by a variable sheer in the direction
of the u
1
axis. From the above description, R(L Q) = 0 S
Q
, a set of m
n1
measure
zero. This is because
(u
2
, , u
n
) S
Q
if and only if
(g (u
2
, , u
n
) , u
2
, , u
n
) R(L Q)
if and only if
(0, u
2
, , u
n
) (g (u
2
, , u
n
) , u
2
, , u
n
)
R(L Q) R(L Q) .
Since is compact, there are nitely many of these open sets Q
1
, , Q
p
which
cover . Let the orthogonal transformations and other quantities described above
11.7. GREENS THEOREM, AN EXAMPLE 313
also be indexed by k for k = 1, , p. Also let Q
0
be an open set with Q
0
and
is covered by Q
0
, Q
1
, , Q
p
. Let u R
0
x x ke
1
where k is large enough
that R
0
Q
0
R
n
<
. Thus in this case, the orthogonal transformation R
0
equals I and
0
x x ke
1
. I claim is an oriented manifold with boundary and the charts are
(W
i
, R
i
) .
To see this is an oriented atlas for the manifold, note that for a.e. points of
R
i
(W
i
W
j
) the function g is dierentiable. Then using the above notation, at these
points R
j
R
1
i
is of the form
j
R
j
R
1
i
and it is a one to one mapping. What is the determinant of its derivative? By the chain
rule,
D
_
j
R
j
R
1
i
_
= D
j
_
R
j
R
1
i
_
DR
j
_
R
1
i
_
DR
i
_
1
i
_
D
1
i
However,
det (D
j
) = 1 = det
_
D
1
j
_
and det (R
i
) = det (R
i
) = 1 by assumption. Therefore, for a.e. u
_
R
j
R
1
i
_
(A) ,
det
_
D
_
R
j
R
1
i
_
(u)
_
> 0.
By Proposition 11.1.7 is indeed an oriented manifold with the given atlas.
11.7.2 Greens Theorem
The general Greens theorem is the following. It follows from Stokes theorem above.
Theorem 11.6.1.
First note that since R
n
, there is no loss of generality in writing
=
n
k=1
a
k
(x) dx
1

dx
k
dx
n
where the hat indicates the dx
k
is omitted. Therefore,
d =
n
k=1
n
j=1
a
k
(x)
x
j
dx
j
dx
1

dx
k
dx
n
=
n
k=1
a
k
(x)
x
k
(1)
k1
dx
1
dx
n
This follows from the denition of integration of dierential forms. If there is a repeat
in the dx
j
then this will lead to a determinant of a matrix which has two equal columns
in the denition of the integral of the dierential form. Also, from the denition, and
again from the properties of the determinant, when you switch two dx
k
it changes the
sign because it is equivalent to switching two columns in a determinant. Then with
these observations, Greens theorem follows.
Theorem 11.7.1 Let be a bounded open set in R
n
, n 2 and let it have
PC
1
boundary and lie locally on one side of its boundary as described above. Also let
=
n
k=1
a
k
(x) dx
1

dx
k
dx
n
be a dierential form where a
I
is assumed to be in C
1
_
_
. Then
k=1
a
k
(x)
x
k
(1)
k1
dm
n
Proof: From the denition and using the usual technique of ignoring the exceptional
set of measure zero,
d
p
i=1
n
k=1
RiWi
a
k
_
R
1
i
(u)
_
x
k
(1)
k1
i
_
R
1
i
(u)
_
(x
1
x
n
)
(u
1
u
n
)
du
Now from the above description of R
1
i
, the determinant in the above integrand equals
1. Therefore, the change of variables theorem applies and the above reduces to
p
i=1
n
k=1
Wi
a
k
(x)
x
k
(1)
k1
i
(x) dx =
i=1
n
k=1
a
k
(x)
x
k
(1)
k1
i
(x) dm
n
=
k=1
a
k
(x)
x
k
(1)
k1
dm
n
This Greens theorem may appear very general because it is an n dimensional the-
orem. However, the best versions of this theorem in the plane are considerably more
general in terms of smoothness of the boundary. Later what is probably the best Greens
theorem is discussed. The following is a specialization to the familiar calculus theorem.
Example 11.7.2 The usual Greens theorem follows from the above specialized to the
case of n = 2.
P (x, y) dx +Q(x, y) dy =
(Q
x
P
y
) dxdy
This follows because the dierential form on the left is of the form
Pdx
dy +Q
dx dy
and so, as above, the derivative of this is
P
y
dy dx +Q
x
dx dy = (Q
x
P
y
) dx dy
It is understood in all the above that the oriented atlas for is the one described there
where
(x, y)
(u
1
, u
2
)
= 1.
Saying is PC
1
reduces in this case to saying is a piecewise C
1
closed curve
which is often the case stated in calculus courses. From calculus, the orientation of
was dened not in the abstract manner described above but by saying that motion
around the curve takes place in the counter clockwise direction. This is really a little
vague although later it will be made very precise. However, it is not hard to see that
this is what is taking place in the above formulation. To do this, consider the following
pictures representing rst the rotation and then the shear which were used to specify
an atlas for .
Q
W
E
R R(W) E
R(W)
T
u
2
11.8. THE DIVERGENCE THEOREM 315
The vertical arrow at the end indicates the direction of increasing u
2
. The vertical side
of R(W) shown there corresponds to the curved side in R(W) which corresponds to
the part of which is selected by Q as shown in the picture. Here R is an orthogonal
transformation which has determinant equal to 1. Now the shear which goes from the
diagram on the right to the one on its left preserves the direction of motion relative to
the surface the curve is bounding. This is geometrically clear. Similarly, the orthogonal
transformation R
which goes from the curved part of the boundary of R(W) to the
corresponding part of preserves the direction of motion relative to the surface. This
is because orthogonal transformations in R
2
whose determinants are 1 correspond to
rotations. Thus increasing u
2
corresponds to counter clockwise motion around R(W)
along the vertical side of R(W) which corresponds to counter clockwise motion around
R(W) along the curved side of R(W) which corresponds to counter clockwise motion
around in the sense that the direction of motion along the curve is always such that
if you were walking in this direction, your left hand would be over the surface. In other
words this agrees with the usual calculus conventions.
11.8 The Divergence Theorem
From Greens theorem, one can quickly obtain a general Divergence theorem for as
described above in Section 11.7.1. First note that from the above description of the R
j
,
_
x
k
, x
i1
, x
in1
_
(u
1
, , u
n
)
= sgn (k, i
1
, i
n1
) .
Let F(x) be a C
1
_
; R
n
_
vector eld. Say F = (F
1
, , F
n
) . Consider the dierential
form
(x)
n
k=1
F
k
(x) (1)
k1
dx
1

dx
k
dx
n
where the hat means dx
k
is being left out. Then
d (x) =
n
k=1
n
j=1
F
k
x
j
(1)
k1
dx
j
dx
1

dx
k
dx
n
=
n
k=1
F
k
x
k
dx
1
dx
k
dx
n
div (F) dx
1
dx
k
dx
n
The assertion between the rst and second lines follows right away from properties of
determinants and the denition of the integral of the above wedge products in terms
of determinants. From Greens theorem and the change of variables formula applied to
the individual terms in the description of
div (F) dx =
p
j=1
Bj
n
k=1
(1)
k1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
_
j
F
k
_
R
1
j
(0, u
2
, , u
n
) du
1
,
du
1
short for du
2
du
3
du
n
.
I want to write this in a more attractive manner which will give more insight. The
above involves a particular partition of unity, the functions being the
i
. Replace F in
the above with
s
F. Next let
_
j
_
be a partition of unity
j
Q
j
such that
s
= 1 on
spt
s
. This partition of unity exists by Lemma 11.5.3. Then
div (
s
F) dx =
p
j=1
Bj
n
k=1
(1)
k1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
_
s
F
k
_
R
1
j
(0, u
2
, , u
n
) du
1
=
Bs
n
k=1
(1)
k1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
(
s
F
k
) R
1
s
(0, u
2
, , u
n
) du
1
(11.25)
because since
s
= 1 on spt
s
, it follows all the other
j
equal zero there.
Consider the vector N dened for u
1
R
s
(W
s
\ L) R
n
0
whose k
th
component is
N
k
= (1)
k1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
= (1)
k+1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
(11.26)
Suppose you dot this vector with a tangent vector R
1
s
/u
i
. This yields
k
(1)
k+1
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
x
k
u
i
= 0
because it is the expansion of
x
1,i
x
1,2
x
1,n
x
2,i
x
2,2
x
2,n
.
.
.
.
.
.
.
.
.
.
.
.
x
n,i
x
n,2
x
n,n
,
a determinant with two equal columns provided i 2. Thus this vector is at least in
some sense normal to . If i = 1, then the above dot product is just
(x
1
x
n
)
(u
1
u
n
)
= 1
This vector is called an exterior normal.
The important thing is the existence of the vector, but does it deserve to be called
an exterior normal? Consider the following picture of R
s
(W
s
)
R
s
(W
s
)
u
1
u
2
, , u
n
E
e
1
We got this by rst doing a rotation of a piece of and then a shear in the direction of
e
1
. Also it was shown above
R
1
s
(u) = R
s
(u
1
+g (u
2
, , u
n
) , u
2
, , u
n
)
T
where R
s
is a rotation, an orthogonal transformation whose determinant is 1. Letting
x = R
1
s
(u) , the above discussion shows x/u
1
N = 1 > 0. Thus it is also the case
that for h small and positive,
x/u1
..
x(u he
1
) x(u)
h
N < 0
11.8. THE DIVERGENCE THEOREM 317
Hence if is the angle between N and
x(uhe1)x(u)
h
, it must be the case that > /2.
However,
x(u he
1
) x(u)
h
points in to for small h > 0 because x(u he
1
) W
s
while x(u) is on the boundary
of W
s
. Therefore, N should be pointing away from at least at the points where
u x(u) is dierentiable. Thus it is geometrically reasonable to use the word exterior
on this vector N.
One could normalize N given in 11.26 by dividing by its magnitude. Then it would
be the unit exterior normal n. The norm of this vector is
_
n
k=1
_
(x
1
, x
k
, x
n
)
(u
2
, , u
n
)
_
2
_
1/2
and by the Binet Cauchy theorem this equals
det
_
DR
1
s
(u
1
)
DR
1
s
(u
1
)
_
1/2
J (u
1
)
where as usual u
1
= (u
2
, , u
n
). Thus the expression in 11.25 reduces to
Bs
_
s
F R
1
s
(u
1
)
_
n
_
R
1
s
(u
1
)
_
J (u
1
) du
1
.
The integrand
u
1

_
s
F R
1
s
(u
1
)
_
n
_
R
1
s
(u
1
)
_
is Borel measurable and bounded. Writing as a sum of positive and negative parts and
using Theorem 7.7.12, there exists a sequence of bounded simple functions {s
k
} which
converges pointwise a.e. to this function. Also the resulting integrands are uniformly
integrable. Then by the Vitali convergence theorem, and Theorem 11.4.2 applied to
these approximations,
Bs
_
s
F R
1
s
(u
1
)
_
n
_
R
1
s
(u
1
)
_
J (u
1
) du
1
= lim
k
Bs
s
k
_
R
s
_
R
1
s
(u
1
)
__
J (u
1
) du
1
= lim
k
Ws
s
k
(R
s
(x)) d
n1
=
Ws
s
(x) F(x) n(x) d
n1
=
s
(x) F(x) n(x) d
n1
Recall the exceptional set on has
n1
measure zero. Upon summing over all s using
that the
s
add to 1,
s
(x) F(x) n(x) d
n1
=
F(x) n(x) d
n1
On the other hand, from 11.25, the left side of the above equals
div (
s
F) dx =
i=1
s,i
F
i
+
s
F
i,i
dx
=
div (F) dx +
n
i=1
_
s
_
,i
F
i
=
div (F) dx
This proves the following general divergence theorem.

Theorem 11.8.1 Let be a bounded open set having PC
1
boundary as de-
scribed above. Also let F be a vector eld with the property that for F
k
a component
function of F, F
k
C
1
_
; R
n
_
. Then there exists an exterior normal vector n which is
dened
n1
a.e. (o the exceptional set L) on such that
F nd
n1
=
div (F) dx
It is worth noting that everything above will work if you relax the requirement in
PC
1
which requires the partial derivatives be bounded o an exceptional set. Instead,
it would suce to say that for some p > n all integrals of the form
Ri(Ui)
x
k
u
j
p
du
are bounded. Here x
k
is the k
th
component of R
1
i
. This is because this condition will
suce to use the Vitali convergence theorem. This would have required more work to
show however so I have not included it. This is also a reason for featuring the Vitali
convergence theorem rather than the dominated convergence theorem which could have
been used in many of the steps in the above presentation.
All of the above can be done more elegantly and in greater generality if you have
Rademachers theorem which gives the almost everywhere dierentiability of Lipschitz
functions. In fact, some of the details become a little easier. However, this approach
requires more real analysis than I want to include in this book, but the main ideas are
all the same. You convolve with a mollier and then do the hard computations with
the mollied function exploiting equality of mixed partial derivatives and then pass to
the limit.
11.9 Spherical Coordinates
Consider the following picture.
0
W
I
(E)
11.9. SPHERICAL COORDINATES 319
Denition 11.9.1 The symbol W
I
(E) represents the piece of a wedge between
the two concentric spheres such that the points x W
I
(E) have the property that
x/ |x| E, a subset of the unit sphere in R
n
, S
n1
and |x| I, an interval on the real
line which does not contain 0.
Now here are some technical results which are interesting for their own sake. The
rst gives the existence of a countable basis for R
n
. This is a countable set of open sets
which has the property that every open set is the union of these special open sets.
Lemma 11.9.2 Let B denote the countable set of all balls in R
n
which have centers
x Q
n
and rational radii. Then every open set is the union of sets of B.
Proof: Let U be an open set and let y U. Then B(y, R) U for some R > 0.
Now by density of Q
n
in R
n
, there exists x B(y, R/10) Q
n
. Now let r Q and
satisfy R/10 < r < R/3. Then y B(x, r) B(y, R) U. This proves the lemma.
With the above countable basis, the following theorem is very easy to obtain. It is
called the Lindelof property.
Theorem 11.9.3 Let C be any collection of open sets and let U = C. Then
there exist countably many sets of C whose union is also equal to U.
Proof: Let B
denote those sets of B in Lemma 11.9.2 which are contained in some

set of C. By this lemma, it follows B
= U. Now use axiom of choice to select for each

B
a single set of C containing it. Denote the resulting countable collection C
. Then
U = B
U
Now consider all the open subsets of R
n
\ {0} . If U is any such open set, it is clear
that if y U, then there exists a set open in S
n1
, E and an open interval I such that
y W
I
(E) U. It follows from Theorem 11.9.3 that every open set which does not
contain 0 is the countable union of the sets of the form W
I
(E) for E open in S
n1
.
The divergence theorem and Greens theorem hold for sets W
I
(E) whenever E is
the intersection of S
n1
with a nite intersection of balls. This is because the resulting
set has PC
1
boundary. Therefore, from the divergence theorem and letting I = (0, 1)
WI(E)
div (x) dx =
E
x
x
|x|
d +
straight part
x nd
where I am going to denote by the measure on S
n1
which corresponds to the diver-
gence theorem and other theorems given above. On the straight parts of the boundary
of W
I
(E) , the vector eld x is parallel to the surface while n is perpendicular to it, all
this o a set of measure zero of course. Therefore, the integrand vanishes and the above
reduces to
nm
n
(W
I
(E)) = (E)
Now let G denote those Borel sets of S
n1
such that the above holds for I = (0, 1) ,
both sides making sense because both E and W
I
(E) are Borel sets in S
n1
and R
n
\{0}
respectively. Then G contains the system of sets which are the nite intersection of
balls with S
n1
. Also if {E
i
} are disjoint sets in G, then W
I
(
i=1
E
i
) =
i=1
W
I
(E
i
)
and so
nm
n
(W
I
(
i=1
E
i
)) = nm
n
(
i=1
W
I
(E
i
))
= n
i=1
m
n
(W
I
(E
i
))
=
i=1
(E
i
) = (
i=1
E
i
)
and so G is closed with respect to countable disjoint unions. Next let E G. Then
nm
n
_
W
I
_
E
C
__
+nm
n
(W
I
(E)) = nm
n
_
W
I
_
S
n1
__
=
_
S
n1
_
= (E) +
_
E
C
_
Now subtracting the equal quantities nm
n
(W
I
(E)) and (E) from both sides yields
E
C
G also. Therefore, by the Lemma on systems Lemma 9.1.2, it follows G contains
the algebra generated by these special sets E the intersection of nitely many open
balls with S
n1
. Therefore, since any open set is the countable union of balls, it follows
the sets open in S
n1
are contained in this algebra. Hence G equals the Borel sets.
This has proved the following important theorem.
Theorem 11.9.4 Let be the Borel measure on S
n1
which goes with the di-
vergence theorems and other theorems like Greens and Stokes theorem. Then for all E
Borel,
(E) = nm
n
(W
I
(E))
where I = (0, 1). Furthermore, W
I
(E) is Borel for any interval. Also
m
n
_
W
[a,b]
(E)
_
= m
n
_
W
(a,b)
(E)
_
= (b
n
a
n
) m
n
_
W
(0,1)
(E)
_
Proof: To show W
I
(E) is Borel for any I rst suppose I is open of the form (0, r).
Then
W
I
(E) = rW
(0,1)
(E)
and this mapping x rx is continuous with continuous inverse so it maps Borel sets
to Borel sets. If I = (0, r],
W
I
(E) =
n=1
W
(0,
1
n
+r)
(E)
and so it is Borel.
W
[a,b]
(E) = W
(0,b]
(E) \ W
(0,a)
(E)
so this is also Borel. Similarly W
(a,b]
(E) is Borel. The last assertion is obvious and
follows from the change of variables formula. This proves the theorem.
Now with this preparation, it is possible to discuss polar coordinates (spherical
coordinates) a dierent way than before.
Note that if = |x| and x/ |x| , then x = . Also the map which takes
(0, ) S
n1
to R
n
\ {0} given by (, ) = x is one to one and onto and
continuous. In addition to this, it follows right away from the denition that if I is any
interval and E S
n1
,
X
WI(E)
() = X
E
() X
I
()
Lemma 11.9.5 For any Borel set F R
n
,
m
n
(F) =
S
n1
n1
X
F
() dd
and the iterated integral on the right makes sense.
Proof: First suppose F = W
I
(E) where E is Borel in S
n1
and I is an interval
having endpoints a b. Then
S
n1
n1
X
WI(E)
() dd =
S
n1
n1
X
E
() X
I
() dd
11.10. EXERCISES 321
=
b
a
n1
(E) d =
b
a
n1
nm
n
_
W
(0,1)
(E)
_
d
= (b
n
a
n
) m
n
_
W
(0,1)
(E)
_
and by Theorem 11.9.4, this equals m
n
(W
I
(E)) . If I is an interval which contains 0,
the above conclusion still holds because both sides are unchanged if 0 is included on the
left and = 0 is included on the right. In particular, the conclusion holds for B(0, r)
in place of F.
Now let G be those Borel sets F such that the desired conclusion holds for F
B(0, M).This contains the system of sets of the form W
I
(E) and is closed with
respect to countable unions of disjoint sets and complements. Therefore, it equals the
Borel sets. Thus
m
n
(F B(0, M)) =
S
n1
n1
X
FB(0,M)
() dd
Now let M and use the monotone convergence theorem. This proves the lemma.
The lemma implies right away that for s a simple function
R
n
sdm
n
=
S
n1
n1
s () dd
Now the following polar coordinates theorem follows.
Theorem 11.9.6 Let f 0 be Borel measurable. Then
R
n
fdm
n
=
S
n1
n1
f () dd
and the iterated integral on the right makes sense.
Proof: By Theorem 7.7.12 there exists a sequence of nonnegative simple functions
{s
k
} which increases to f. Therefore, from the monotone convergence theorem and
what was shown above,
R
n
fdm
n
= lim
k
R
n
s
k
dm
n
= lim
k
S
n1
n1
s
k
() dd
=
S
n1
n1
f () dd
11.10 Exercises
1. Let
(x)
I
a
I
(x) dx
I
be a dierential form where x R
m
and the I are increasing lists of n indices
taken from 1, , m. Also assume each a
I
(x) has the property that all mixed
partial derivatives are equal. For example, from Corollary 6.9.2 this happens if
the function is C
2
. Show that under this condition, d (d ()) = 0. To show this,
rst explain why
dx
i
dx
j
dx
I
= dx
j
dx
i
dx
I
When you integrate one you get 1 times the integral of the other. This is the
sense in which the above formula holds. When you have a dierential form with
the property that d = 0 this is called a closed form. If = d, then is called
exact. Thus every closed form is exact provided you have sucient smoothness
on the coecients of the dierential form.
2. Recall that in the denition of area measure, you use
J (u) = det
_
DR
1
(u)
DR
1
(u)
_
1/2
Now in the special case of the manifold of Greens theorem where
R
1
(u
2
, , u
n
) = R
(g (u
2
, , u
n
) , u
2
, , u
n
) ,
show
J (u) =
1 +
_
g
u
2
_
2
+ +
_
g
u
n
_
2
3. Let u
1
, , u
p
be vectors in R
n
. Show det M 0 where M
ij
u
i
u
j
. Hint:
Show this matrix has all nonnegative eigenvalues and then use the theorem which
says the determinant is the product of the eigenvalues. This matrix is called the
Grammian matrix. The details follow from noting that M is of the form
U
U
_
_
_
u
1
.
.
.
u
p
_
_
_
_
u
1
u
p
_
and then showing that U
U has all nonnegative eigenvalues.

4. Suppose {v
1
, , v
n
} are n vectors in R
m
for m n. Show that the only appro-
priate denition of the volume of the n dimensional parallelepiped determined by
these vectors,
_
_
_
n
j=1
s
j
v
j
: s
j
[0, 1]
_
_
_
is
det (M
M)
1/2
where M is the mn matrix which has columns v
1
, , v
n
. Hint: Show this is
clearly true if n = 1 because the above just yields the usual length of the vector.
Now suppose the formula gives the right thing for n1 vectors and argue it gives
the right thing for n vectors. In doing this, you might want to show that a vector
which is perpendicular to the span of v
1
, , v
n1
is
det
_
_
_
_
_
u
1
u
2
u
n
v
11
v
12
v
1n
.
.
.
.
.
.
.
.
.
v
n1,1
v
n1,2
v
n1,n
_
_
_
_
_
where {u
1
, , u
n
} is an orthonormal basis for span (v
1
, , v
n
) and v
ij
is the
j
th
component of v
i
with respect to this orthonormal basis. Then argue that
if you replace the top line with v
n1
, , v
nn
, the absolute value of the resulting
determinant is the appropriate denition of the volume of the parallelepiped. Next
note you could get this number by taking the determinant of the transpose of the
above matrix times that matrix and then take a square root. After this, identify
this product with a Grammian matrix and then the desired result follows.
11.10. EXERCISES 323
5. Why is the denition of area on a manifold given above reasonable and what is
its geometric meaning? Each function
u
i
R
1
(u
1
, , u
n
)
yields a curve which lies in . Thus R
1
,ui
is a vector tangent to this curve and
R
1
,ui
du
i
is an innitesimal vector tangent to the curve. Now use the previous
problem to see that when you nd the area of a set on , you are essentially
summing the volumes of innitesimal parallelepipeds which are tangent to .
6. Let be a bounded open set in R
n
with PC
1
boundary or more generally one for
which the divergence theorem holds. Let u, v C
2
_
_
. Then
(vu uv) dx =
_
v
u
n
u
v
n
_
d
n1
Here
u
n
u n
where n is the unit outer normal described above. Establish this formula which is
known as Greens identity. Hint: You might establish the following easy identity.
(vu) vu = v u.
Recall u
n
k=1
u
x
k
x
k
and u = (u
x1
, , u
xn
) while
F f
1x1
+ +f
nxn
div (F)
The Laplace And Poisson
Equations
This material is mostly in the book by Evans [14] which is where I got it. It is really
partial dierential equations but it is such a nice illustration of the divergence theorem
and other advanced calculus theorems, that I am including it here even if it is somewhat
out of place and would normally be encountered in a partial dierential equations course.
12.1 Balls
Recall, B(x, r) denotes the set of all y R
n
such that |y x| < r. By the change of
variables formula for multiple integrals or simple geometric reasoning, all balls of radius
r have the same volume. Furthermore, simple reasoning or change of variables formula
will show that the volume of the ball of radius r equals
n
r
n
where
n
will denote the
volume of the unit ball in R
n
. With the divergence theorem, it is now easy to give a
simple relationship between the surface area of the ball of radius r and the volume. By
the divergence theorem,
B(0,r)
div x dx =
B(0,r)
x
x
|x|
d
n1
because the unit outward normal on B(0, r) is
x
|x|
. Therefore,
n
n
r
n
= r
n1
(B(0, r))
and so
n1
(B(0, r)) = n
n
r
n1
.
You recall the surface area of S
2
_
x R
3
: |x| = r
_
is given by 4r
2
while the volume
of the ball, B(0, r) is
4
3
r
3
. This follows the above pattern. You just take the derivative
with respect to the radius of the volume of the ball of radius r to get the area of the
surface of this ball. Let
n
denote the area of the sphere S
n1
= {x R
n
: |x| = 1} . I
just showed that
n
= n
n
. (12.1)
I want to nd
n
now and also to get a relationship between
n
and
n1
. Consider
the following picture of the ball of radius seen on the side.
325
326 THE LAPLACE AND POISSON EQUATIONS
y
r
R
n1
Taking slices at height y as shown and using that these slices have n1 dimensional
area equal to
n1
r
n1
, it follows from Fubinis theorem
n
= 2
n1
_
2
y
2
_
(n1)/2
dy (12.2)
Lemma 12.1.1 (1/2) =

Proof:
_
1
2
_

0
e
t
t
1/2
dt
Now change the variables letting t = s
2
so dt = 2sds and the integral becomes
2

0
e
s
2
ds =
e
s
2
ds
Thus
_
1
2
_
=
e
x
2
dx so
_
1
2
_
2
=
e
(x
2
+y
2
)
dxdy and by polar coordi-
nates and changing the variables, this is just
2
0

0
e
r
2
rdrd =
Therefore,
_
1
2
_
=

as claimed. This proves the lemma.
Theorem 12.1.2
n
=

n/2
(
n
2
+1)
where denotes the gamma function, dened
for > 0 by
()

0
e
t
t
1
dt.
Proof: Recall that ( + 1) = () . (Establish this by integrating by parts.)
This is proved by induction using 12.2. When n = 1, the right answer should be 2
because in this case the ball is just (1, 1) . Is this what is obtained from the formula?
Is
1
2 =

1/2
(3/2)
?
Using the identity ( + 1) = () , the above equals
1/2
(1/2) (1/2)
= 2
from the above lemma. Now suppose the theorem is true for n. Then letting = 1,
12.2 implies
n+1
= 2

n/2
_
n
2
+ 1
_
1
0
_
1 y
2
_
n/2
dy
12.2. POISSONS PROBLEM 327
Now change variables. Let u = y
2
. Then
n+1
=

n/2
_
n
2
+ 1
_
1
0
u
1/2
(1 u)
1
2
n
du
=

n/2
_
n
2
+ 1
_
1
0
u
(1/2)1
(1 u)
n+2
2
1
du
=

n/2
_
n
2
+ 1
_B
_
1
2
,
n + 2
2
_
At this point, use the result of Problem 8 on Page 216 to simplify the messy integral
which equals the beta function. Thus the above equals
n/2
_
n
2
+ 1
_
(1/2)
_
n+2
2
_
_
n+2
2
+
1
2
_
=
n/2

1/2
_
1
2
n +
3
2
_ =

(n+1)/2
_
n+1
2
+ 1
_
and this gives the correct formula for
n+1
12.2 Poissons Problem
The Poisson problem is to nd u satisfying the two conditions
u = f, in U, u = g on U. (12.3)
Here U is an open bounded set for which the divergence theorem holds. For example,
it could be a PC
1
manifold. When f = 0 this is called Laplaces equation and the
boundary condition given is called a Dirichlet boundary condition. When u = 0,
the function, u is said to be a harmonic function. When f = 0, it is called Poissons
equation. I will give a way of representing the solution to these problems. When this
has been done, great and marvelous conclusions may be drawn about the solutions.
Before doing anything else however, it is wise to prove a fundamental result called the
weak maximum principle.
Theorem 12.2.1 Suppose U is an open bounded set and
u C
2
(U) C
_
U
_
and
u 0 in U.
Then
max
_
u(x) : x U
_
= max {u(x) : x U} .
Proof: Suppose not. Then there exists x
0
U such that
u(x
0
) > max {u(x) : x U} .
Consider w
(x) u(x) + |x|

2
. I claim that for small enough > 0, the function w
also has this property. If not, there exists x
U such that w
(x
) w
(x) for all

x U. But since U is bounded, it follows the points, x
are in a compact set and so

there exists a subsequence, still denoted by x
such that as 0, x
x
1
U. But
then for any x U,
u(x
0
) w
(x
0
) w
(x
)
and taking a limit as 0 yields
u(x
0
) u(x
1
)
contrary to the property of x
0
above. It follows that my claim is veried. Pick such an
. Then w
assumes its maximum value in U say at x

2
. Then by the second derivative
test,
w
(x
2
) = u(x
2
) + 2 0
which requires u(x
2
) 2, contrary to the assumption that u 0. This proves
the theorem.
The theorem makes it very easy to verify the following uniqueness result.
Corollary 12.2.2 Suppose U is an open bounded set and
u C
2
(U) C
_
U
_
and
u = 0 in U, u = 0 on U.
Then u = 0.
Proof: From the weak maximum principle, u 0. Now apply the weak maximum
principle to u which satises the same conditions as u. Thus u 0 and so u 0.
Therefore, u = 0 as claimed. This proves the corollary.
Dene
r
n
(x)
_
ln |x| if n = 2
1
|x|
n2
if n > 2
.
Then it is fairly routine to verify the following Lemma.
Lemma 12.2.3 For r
n
given above,
r
n
= 0.
Proof: I will verify the case where n 3 and leave the other case for you.
D
xi
_
n
i=1
x
2
i
_
(n2)/2
= (n 2) x
i
_
_
j
x
2
j
_
_
n/2
Therefore,
D
xi
(D
xi
(r
n
)) =
_
_
j
x
2
j
_
_
(n+2)/2
(n 2)
_
_
nx
2
i

n
j=1
x
2
j
_
_
.
It follows
r
n
=
_
_
j
x
2
j
_
_
(n+2)
2
(n 2)
_
_
n
n
i=1
x
2
i

n
i=1
n
j=1
x
2
j
_
_
= 0.
From now on assume U is PC
1
to be specic.
Now let U
be as indicated in the following picture. I have taken out a ball of radius

which is centered at the point, x U.
B(x, ) = B
Then the divergence theorem will continue to hold for U
(why?) and so I can use

Greens identity, Problem 6 on Page 323 to write the following for u, v C
2
_
U
_
.
U
(uv vu) dx =
U
_
u
v
n
v
u
n
_
d
B
_
u
v
n
v
u
n
_
d (12.4)
Now, letting x U, I will pick for v the function,
v (y) r
n
(y x)
x
(y) (12.5)
where
x
is a function which is chosen such that on U,
x
(y) = r
n
(y x)
so that 12.5 vanishes for y U and
x
is in C
2
_
U
_
and also satises
x
= 0.
The existence of such a function is another issue. For now, assume such a
function exists.
1
Then assuming such a function exists, 12.4 reduces to
U
vudx =
U
u
v
n
d
B
_
u
v
n
v
u
n
_
d. (12.6)
The idea now is to let 0 and see what happens. Consider the term
B
v
u
n
d.
The area is O
_
n1
_
while the integrand is O
_
(n2)
_
in the case where n 3. In the
case where n = 2, the area is O() and the integrand is O(|ln |||) . Now you know that
lim
0
ln || = 0 and so in the case n = 2, this term converges to 0 as 0. In the
case that n 3, it also converges to zero because in this case the integral is O() .
Next consider the term
B
u
v
n
d =
B
u(y)
_
r
n
n
(y x)

x
n
(y)
_
d.
1
In fact, if the boundary of U is smooth enough, such a function will always exist, although this
requires more work to show but this is not the point. The point is to explicitly nd it and this will
only be possible for certain simple choices of U.
This term does not disappear as 0. First note that since
x
has bounded derivatives,
lim
0
B
u(y)
_
r
n
n
(y x)

x
n
(y)
_
d = lim
0
_
B
u(y)
r
n
n
(y x) d
_
(12.7)
and so it is just this last item which is of concern.
First consider the case that n = 2. In this case,
r
2
(y) =
_
y
1
|y|
2
,
y
2
|y|
2
_
Also, on B
, the exterior unit normal, n, equals

1
(y
1
x
1
, y
2
x
2
) .
It follows that on B
,
r
2
n
(y x) =
1
(y
1
x
1
, y
2
x
2
)
_
y
1
x
1
|y x|
2
,
y
2
x
2
|y x|
2
_
=
1
.
Therefore, this term in 12.7 converges to
u(x) 2. (12.8)
Next consider the case where n 3. In this case,
r
n
(y) = (n 2)
_
y
1
|y|
n
, ,
y
n
|y|
_
and the unit outer normal, n, equals
1
(y
1
x
1
, , y
n
x
n
) .
Therefore,
r
n
n
(y x) =
(n 2)
|y x|
2
|y x|
n
=
(n 2)
n1
.
Letting
n
denote the n1 dimensional surface area of the unit sphere, S
n1
, it follows
that the last term in 12.7 converges to
u(x) (n 2)
n
(12.9)
Finally consider the integral,
B
vudx.
B
|vu| dx C
B
|r
n
(y x)
x
(y)| dy
C
B
|r
n
(y x)| dy +O(
n
)
Using polar coordinates to evaluate this improper integral in the case where n 3,
C
B
|r
n
(y x)| dx = C
S
n1
1
n2
n1
dd
= C
S
n1
dd
which converges to 0 as 0. In the case where n = 2
C
B
|r
n
(y x)| dx = C
S
n1
ln () dd
which also converges to 0 as 0. Therefore, returning to 12.6 and using the above
limits, yields in the case where n 3,
U
vudx =
U
u
v
n
d +u(x) (n 2)
n
, (12.10)
and in the case where n = 2,
U
vudx =
U
u
v
n
d u(x) 2. (12.11)
These two formulas show that it is possible to represent the solutions to Poissons
problem provided the function,
x
can be determined. I will show you can determine
this function in the case that U = B(0, r) .
12.2.1 Poissons Problem For A Ball
Lemma 12.2.4 When |y| = r and x = 0,
y |x|
r

rx
|x|
= |x y| ,
and for |x| , |y| < r, x = 0,
y |x|
r

rx
|x|
= 0.
Proof: Suppose rst that |y| = r. Then
y |x|
r

rx
|x|
2
=
_
y |x|
r

rx
|x|
_
_
y |x|
r

rx
|x|
_
=
|x|
2
r
2
|y|
2
2y x +r
2
|x|
2
|x|
2
= |x|
2
2x y +|y|
2
= |x y|
2
.
This proves the rst claim. Next suppose |x| , |y| < r and suppose, contrary to what is
claimed, that
y |x|
r

rx
|x|
= 0.
Then
y |x|
2
= r
2
x
and so |y| |x|
2
= r
2
|x| which implies
|y| |x| = r
2
contrary to the assumption that |x| , |y| < r.
Let
x
(y)
_
_
_
y|x|
r

rx
|x|
(n2)
, r
(n2)
for x = 0 if n 3
ln
y|x|
r

rx
|x|
, ln (r) if x = 0 if n = 2
Note that
lim
x0
y |x|
r

rx
|x|
= r.
Then
x
(y) = r
n
(y x) if |y| = r, and
x
= 0. This last claim is obviously true if
x = 0. If x = 0, then
0
(y) equals a constant and so it is also obvious in this case that
x
= 0.
The following lemma is easy to obtain.
Lemma 12.2.5 Let
f (y) =
_
|y x|
(n2)
if n 3
ln |y x| if n = 2
.
Then
f (y) =
_
(n2)(yx)
|yx|
n if n 3
yx
|yx|
2
if n = 2
.
Also, the outer normal on B(0, r) is y/r.
From Lemma 12.2.5 it follows easily that for v (y) = r
n
(y x)
x
(y) and y
B(0, r) , then for n 3,
v
n
=
y
r

_
_
(n 2) (y x)
|y x|
n
+
_
|x|
r
_
(n2)
(n 2)
_
y
r
2
|x|
2
x
_
y
r
2
|x|
2
x
n
_
_
=
(n 2)
r
_
r
2
y x
_
|y x|
n
+
|x|
2
r
2
_
|x|
r
_
n
(n 2)
r
_
r
2
r
2
|x|
2
x y
_
y
r
2
|x|
2
x
n
=
(n 2)
r
_
r
2
y x
_
|y x|
n
+
(n 2)
r
_
|x|
2
r
2
r
2
x y
_
|x|
r
y
r
|x|
x
n
which by Lemma 12.2.4 equals
(n 2)
r
_
r
2
y x
_
|y x|
n
+
(n 2)
r
_
|x|
2
r
2
r
2
x y
_
|y x|
n
=
(n 2)
r
r
2
|y x|
n
+
(n 2)
r
|x|
2
|y x|
n
=
(n 2)
r
|x|
2
r
2
|y x|
n
.
In the case where n = 2, and |y| = r, then Lemma 12.2.4 implies
v
n
=
y
r

_
_
(y x)
|y x|
2

_
|x|
r
_
_
y|x|
r

rx
|x|
_
y|x|
r

rx
|x|
2
_
_
=
y
r

_
_
(y x)
|y x|
2

_
y|x|
2
r
2
x
_
|y x|
2
_
_
=
1
r
r
2
|x|
2
|y x|
2
.
Referring to 12.10 and 12.11, we would hope a solution, u to Poissons problem
satises for n 3
U
(r
n
(y x)
x
(y)) f (y) dy
=
U
g (y)
_
(n 2)
r
|x|
2
r
2
|y x|
n
_
d (y) +u(x) (n 2)
n
.
Thus
u(x) =
1
n
(n 2)
U
(
x
(y) r
n
(y x)) f (y) dy +
U
g (y)
_
(n 2)
r
r
2
|x|
2
|y x|
n
_
d (y)
_
. (12.12)
In the case where n = 2,
U
(r
2
(y x)
x
(y)) f (y) dx =
U
g (y)
_
1
r
r
2
|x|
2
|y x|
2
_
d (y) u(x) 2
and so in this case,
u(x) =
1
2
_
U
(r
2
(y x)
x
(y)) f (y) dx +
U
g (y)
_
1
r
r
2
|x|
2
|y x|
2
_
d (y)
_
.
(12.13)
12.2.2 Does It Work In Case f = 0?
It turns out these formulas work better than you might expect. In particular, they work
in the case where g is only continuous. In deriving these formulas, more was assumed
on the function than this. In particular, it would have been the case that g was equal
to the restriction of a function in C
2
(R
n
) to B(0,r) . The problem considered here is
u = 0 in U, u = g on U
From 12.12 it follows that if u solves the above problem, known as the Dirichlet problem,
then
u(x) =
r
2
|x|
2
n
r
U
g (y)
1
|y x|
n
d (y) .
I have shown this in case u C
2
_
U
_
which is more specic than to say u C
2
(U)
C
_
U
_
. Nevertheless, it is enough to give the following lemma.
Lemma 12.2.6 The following holds for n 3.
1 =
U
r
2
|x|
2
r
n
|y x|
n
d (y) .
For n = 2,
1 =
U
1
2r
r
2
|x|
2
|y x|
2
d (y) .
Proof: Consider the problem
u = 0 in U, u = 1 on U.
I know a solution to this problem which is in C
2
_
U
_
, namely u 1. Therefore, by
Corollary 12.2.2 this is the only solution and since it is in C
2
_
U
_
, it follows from 12.12
that in case n 3,
1 = u(x) =
U
r
2
|x|
2
r
n
|y x|
n
d (y)
and in case n = 2, the other formula claimed above holds.
Theorem 12.2.7 Let U = B(0, r) and let g C (U) . Then there exists a
unique solution u C
2
(U) C
_
U
_
to the problem
u = 0 in U, u = g on U.
This solution is given by the formula,
u(x) =
1
n
r
U
g (y)
r
2
|x|
2
|y x|
n
d (y) (12.14)
for every n 2. Here
2
2.
Proof: That u = 0 in U follows from the observation that the dierence quotients
used to compute the partial derivatives converge uniformly in y U for any given
x U. To see this note that for y U, the partial derivatives of the expression,
r
2
|x|
2
|y x|
n
taken with respect to x
k
are uniformly bounded and continuous. In fact, this is true
of all partial derivatives. Therefore you can take the dierential operator inside the
integral and write
x
1
n
r
U
g (y)
r
2
|x|
2
|y x|
n
d (y) =
1
n
r
U
g (y)
x
_
r
2
|x|
2
|y x|
n
_
d (y) = 0.
It only remains to verify that it achieves the desired boundary condition. Let x
0
U.
From Lemma 12.2.6,
|g (x
0
) u(x)|
1
n
r
U
|g (y) g (x
0
)|
_
r
2
|x|
2
|y x|
n
_
d (y) (12.15)
n
r
[|yx
0
|<]
|g (y) g (x
0
)|
_
r
2
|x|
2
|y x|
n
_
d (y) +(12.16)
1
n
r
[|yx
0
|]
|g (y) g (x
0
)|
_
r
2
|x|
2
|y x|
n
_
d (y) (12.17)
where is a positive number. Letting > 0 be given, choose small enough that if
|y x
0
| < , then |g (y) g (x
0
)| <

2
. Then for such ,
1
n
r
[|yx
0
|<]
|g (y) g (x
0
)|
_
r
2
|x|
2
|y x|
n
_
d (y)
n
r
[|yx
0
|<]
2
_
r
2
|x|
2
|y x|
n
_
d (y)
n
r
2
_
r
2
|x|
2
|y x|
n
_
d (y) =

2
.
Denoting by M the maximum value of g on U, the integral in 12.17 is dominated by
2M
n
r
[|yx
0
|]
_
r
2
|x|
2
|y x|
n
_
d (y)
2M
n
r
[|yx
0
|]
_
r
2
|x|
2
[|y x
0
| |x x
0
|]
n
_
d (y)
2M
n
r
[|yx
0
|]
_
r
2
|x|
2
_

2
n
_
d (y)
2M
n
r
_
2
_
n
U
_
r
2
|x|
2
_
d (y)
when |x x
0
| is suciently small. Then taking |x x
0
| still smaller, if necessary, this
last expression is less than /2 because |x
0
| = r and so lim
xx0
_
r
2
|x|
2
_
= 0. This
proves lim
xx0
u(x) = g (x
0
) and this proves the existence part of this theorem. The
uniqueness part follows from Corollary 12.2.2.
Actually, I could have said a little more about the boundary values in Theorem
12.2.7. Since g is continuous on U, it follows g is uniformly continuous and so the
above proof shows that actually lim
xx0
u(x) = g (x
0
) uniformly for x
0
U.
Not surprisingly, it is not necessary to have the ball centered at 0 for the above to
work.
Corollary 12.2.8 Let U = B(x
0
, r) and let g C (U) . Then there exists a unique
solution u C
2
(U) C
_
U
_
to the problem
u = 0 in U, u = g on U.
u(x) =
1
n
r
U
g (y)
r
2
|x x
0
|
2
|y x|
n
d (y) (12.18)
for every n 2. Here
2
= 2.
This corollary implies the following.
Corollary 12.2.9 Let u be a harmonic function dened on an open set, U R
n
and let B(x
0
, r) U. Then
u(x
0
) =
1
n
r
n1
B(x0,r)
u(y) d
The representation formula, 12.14 is called Poissons integral formula. I have now
shown it works better than you had a right to expect for the Laplace equation. What
happens when f = 0?
12.2.3 The Case Where f = 0, Poissons Equation
I will verify the results for the case n 3. The case n = 2 is entirely similar. This is
still in the context that U = B(0, r) . Thus
x
(y)
_
_
_
y|x|
r

rx
|x|
(n2)
, r
(n2)
for x = 0 if n 3
ln
y|x|
r

rx
|x|
, ln (r) if x = 0 if n = 2
Recall that r
n
(y x) =
x
(y) whenever y U.
_
U
_
or in L
p
(U) for p > n/2
2
. Then for x U, and
x
0
U,
lim
xx0
1
n
(n 2)
U
(
x
(y) r
n
(y x)) f (y) dy = 0.
Proof: There are two parts to this lemma. First the following claim is shown in
which an integral is taken over B(x
0
, ). After this, the integral over U \ B(x
0
, ) will
be considered. First note that
lim
xx0
x
(y) r
n
(y x) = 0
Claim:
lim
0
B(x0,)
x
(y) |f (y)| dy = 0, lim
0
B(x0,)
r
n
(y x) |f (y)| dy = 0.
Proof of the claim: Using polar coordinates,
B(x0,)
x
(y) |f (y)| dy
=
B(0,)
r
n
_
(x
0
+z) |x|
r

rx
|x|
_
|f (x
0
+z)| dz
=
S
n1
r
n
_
(x
0
+w) |x|
r

rx
|x|
_
|f (x
0
+w)|
n1
dd
Now from the formula for r
n
, there exists
0
> 0 such that for [0,
0
] ,
r
n
_
(x
0
+w) |x|
r

rx
|x|
_
n2
is bounded. Therefore,
B(x0,)
x
(y) |f (y)| dy C
S
n1
|f (x
0
+w)| dd.
If f is continuous, this is dominated by an expression of the form
C
S
n1
dd
which converges to 0 as 0.
If f L
p
(U) , then by Holders inequality, (Problem 3 on Page 250) for
1
p
+
1
q
= 1,
S
n1
|f (x
0
+w)| dd
=
S
n1
|f (x
0
+w)|
2n
n1
dd
S
n1
|f (x
0
+w)|
p
n1
dd
_
1/p
S
n1
_
2n
_
q
n1
dd
_
1/q
C ||f||
L
p
(U)
.
2
This means f is measurable and |f|
p
has nite integral
Similar reasoning shows that
lim
0
B(x0,)
|r
n
(y x)| |f (y)| dy = 0.
This proves the claim.
Let > 0 be given and choose > 0 such that r/2 > > 0 where r is the radius of
the ball U and small enough that
B(x0,2)
|f (y)| |
x
(y) r
n
(y x)| dy < .
Then consider x B(x
0
, ) so that for y / B(x
0
, 2) , |y x| > and so for such y,
|
x
(y) r
n
(y x)| C
(n2)
for some constant C. Thus the integrand in
U
f (y) (
x
(y) r
n
(y x)) dy
U\B(x0,2)
|f (y)| |(
x
(y) r
n
(y x))| dy
+
B(x0,2)
|f (y)| |
x
(y) r
n
(y x)| dy
U\B(x0,2)
f (y) (
x
(y) r
n
(y x)) dy +
Now apply the dominated convergence theorem in this last integral to conclude it con-
verges to 0 as x x
0
The following lemma follows from this one and Theorem 12.2.7.
_
U
_
or in L
p
(U) for p > n/2 and let g C (U) . Then
if u is given by 12.12 in the case where n 3 or by 12.13 in the case where n = 2, then
if x
0
U,
lim
xx0
u(x) = g (x
0
) .
Not surprisingly, you can relax the condition that g C (U) but I wont do so here.
The next question is about the partial dierential equation satised by u for u given
by 12.12 in the case where n 3 or by 12.13 for n = 2. This is going to introduce a
new idea. I will just sketch the main ideas and leave you to work out the details, most
of which have already been considered in a similar context.
Let C
c
(U) and let x U. Let U
denote the open set which has B(y, )

deleted from it, much as was done earlier. In what follows I will denote with a subscript
of x things for which x is the variable. Then denoting by G(y, x) the expression
x
(y) r
n
(y x) , it is easy to verify that
x
G(y, x) = 0 and so by Fubinis theorem,
U
1
n
(n 2)
_
U
(
x
(y) r
n
(y x)) f (y) dy
_
x
(x) dx
= lim
0
U
1
n
(n 2)
_
U
(
x
(y) r
n
(y x)) f (y) dy
_
x
(x) dx
= lim
0
U
_
U
1
n
(n 2)
(
x
(y) r
n
(y x))
x
(x) dx
_
f (y) dy
= lim
0
U
_
_
_
U
1
n
(n 2)
_
_
_
G(y,x)
..
x
(y) r
n
(y x)
_
_
_
x
(x) dx
_
_
_f (y) dy
= lim
0
1
n
(n 2)
U
f (y)
_
B(y,)
_
G

n
x
G
n
x
_
d (x)
_
dy
= lim
0
1
n
(n 2)
U
f (y)
B(y,)
G
n
x
d (x) dy
Now x
x
(y) and its partial derivatives are continuous and so the above reduces to
= lim
0
1
n
(n 2)
U
f (y)
B(y,)
r
n
n
x
(x y) d (x) dy
= lim
0
1
U
f (y)
B(y,)
n1
d (x) dy =
U
f (y) (y) dy.
Similar but easier reasoning shows that
U
_
1
n
r
U
g (y)
r
2
|x|
2
|y x|
n
d (y)
_
x
(x) dx = 0.
Therefore, if n 3, and u is given by 12.12, then whenever C
c
(U) ,
U
udx =
U
fdx. (12.19)
The same result holds for n = 2.
Denition 12.2.12 u = f on U in the weak sense or in the sense of distri-
butions if for all C
c
(U) , 12.19 holds.
This with Lemma 12.2.11 proves the following major theorem.
Theorem 12.2.13 Let f C
_
U
_
or in L
p
(U) for p > n/2 and let g C (U) .
Then if u is given by 12.12 in the case where n 3 or by 12.13 in the case where n = 2,
then u solves the dierential equation of the Poisson problem in the sense of distributions
along with the boundary conditions.
12.3 Properties Of Harmonic Functions
Consider the problem for g C (U) .
u = 0 in U, u = g on U.
When U = B(x
0
, r) , it has now been shown there exists a unique solution to the above
problem satisfying u C
2
(U) C
_
U
_
and it is given by the formula
u(x) =
r
2
|x x
0
|
2
n
r
B(x0,r)
g (y)
|y x|
n
d (y) (12.20)
It was also noted that this formula implies the mean value property for harmonic func-
tions,
u(x
0
) =
1
n
r
n1
B(x0,r)
u(y) d (y) . (12.21)
The mean value property can also be formulated in terms of an integral taken over
B(x
0
, r) .
12.3. PROPERTIES OF HARMONIC FUNCTIONS 339
Lemma 12.3.1 Let u be harmonic and C
2
on an open set, V and let B(x
0
, r) V.
Then
u(x
0
) =
1
m
n
(B(x
0
, r))
B(x0,r)
u(y) dy
where here m
n
(B(x
0
, r)) denotes the volume of the ball.
Proof: From the method of polar coordinates and the mean value property given
in 12.21, along with the observation that m
n
(B(x
0
, r)) =
n
n
r
n
,
B(x0,r)
u(y) dy =
r
0
S
n1
u(x
0
+y)
n1
d (y) d
=
r
0
B(0,)
u(x
0
+y) d (y) d
= u(x
0
)
r
0
n1
d = u(x
0
)

n
n
r
n
= u(x
0
) m
n
(B(x
0
, r)) .
There is a very interesting theorem which says roughly that the values of a nonneg-
ative harmonic function are all comparable. It is known as Harnacks inequality.
Theorem 12.3.2 Let U be an open set and let u C
2
(U) be a nonnegative
harmonic function. Also let U
1
be a connected open set which is bounded and satises
U
1
U. Then there exists a constant, C, depending only on U
1
such that
max
_
u(x) : x U
1
_
C min
_
u(x) : x U
1
_
Proof: There is a positive distance between U
1
and U
C
because of compactness
of U
1
. Therefore there exists r > 0 such that whenever x U
1
, B(x, 2r) U. Then
consider x U
1
and let |x y| < r. Then from Lemma 12.3.1
u(x) =
1
m
n
(B(x, 2r))
B(x,2r)
u(z) dz
=
1
2
n
m
n
(B(x, r))
B(x,2r)
u(z) dz
1
2
n
m
n
(B(y, r))
B(y,r)
u(z) dz =
1
2
n
u(y) .
The fact that u 0 is used in going to the last line. Since U
1
is compact, there exist
nitely many balls having centers in U
1
, {B(x
i
, r)}
m
i=1
such that
U
1

m
i=1
B(x
i
, r/2) .
Furthermore each of these balls must have nonempty intersection with at least one of the
others because if not, it would follow that U
1
would not be connected. Letting x, y U
1
,
there must be a sequence of these balls, B
1
, B
2
, , B
k
such that x B
1
, y B
k
, and
B
i
B
i+1
= for i = 1, 2, , k 1. Therefore, picking a point, z
i+1
B
i
B
i+1
, the
above estimate implies
u(x)
1
2
n
u(z
2
) , u(z
2
)
1
2
n
u(z
3
) , u(z
3
)
1
2
n
u(z
4
) , , u(z
k
)
1
2
n
u(y) .
Therefore,
u(x)
_
1
2
n
_
k
u(y)
_
1
2
n
_
m
u(y) .
Therefore, for all x U
1
,
sup {u(y) : y U
1
} (2
n
)
m
u(x)
and so
max
_
u(x) : x U
1
_
= sup {u(y) : y U
1
}
(2
n
)
m
inf {u(x) : x U
1
} = (2
n
)
m
min
_
u(x) : x U
1
_
.
This proves the inequality.
The next theorem comes from the representation formula for harmonic functions
given above.
Theorem 12.3.3 Let U be an open set and suppose u C
2
(U) and u is har-
monic. Then in fact, u C
(U) . That is, u possesses all partial derivatives and they

are all continuous.
Proof: Let B(x
0
,r) U. I will show that u C
(B(x
0
,r)) . From 12.20, it follows
that for x B(x
0
,r) ,
r
2
|x x
0
|
2
n
r
B(x0,r)
u(y)
|y x|
n
d (y) = u(x) .
It is obvious that x
r
2
|xx
0
|
2
nr
is innitely dierentiable. Therefore, consider
x
B(x0,r)
u(y)
|y x|
n
d (y) . (12.22)
Take x B(x
0
, r) and consider a dierence quotient for t = 0.
_
B(x0,r)
u(y)
1
t
_
1
|y(x +te
k
)|
n

1
|y x|
n
_
d (y)
_
Then by the mean value theorem, the term
1
t
_
1
|y(x +te
k
)|
n

1
|y x|
n
_
equals
n|x+t (t) e
k
y|
(n+2)
(x
k
+ (t) t y
k
)
and as t 0, this converges uniformly for y B(x
0
, r) to
n|x y|
(n+2)
(x
k
y
k
) .
This uniform convergence implies you can take a partial derivative of the function of x
given in 12.22 obtaining the partial derivative with respect to x
k
equals
B(x0,r)
n(x
k
y
k
) u(y)
|y x|
n+2
d (y) .
Now exactly the same reasoning applies to this function of x yielding a similar formula.
The continuity of the integrand as a function of x implies continuity of the partial
derivatives. The idea is there is never any problem because y B(x
0
, r) and x is a
given point not on this boundary. This proves the theorem.
Liouvilles theorem is a famous result in complex variables which asserts that an
entire bounded function is constant. A similar result holds for harmonic functions.
12.4. LAPLACES EQUATION FOR GENERAL SETS 341
Theorem 12.3.4 (Liouvilles theorem) Suppose u is harmonic on R
n
and is
bounded. Then u is constant.
Proof: From the Poisson formula
r
2
|x|
2
n
r
B(0,r)
u(y)
|y x|
n
d (y) = u(x) .
Now from the discussion above,
u(x)
x
k
=
2x
k
n
r
B(x0,r)
u(y)
|y x|
n
d (y) +
r
2
|x|
2
n
r
B(0,r)
u(y) (y
k
x
k
)
|y x|
n+2
d (y)
Therefore, letting |u(y)| M for all y R
n
,
u(x)
x
k

2 |x|
n
r
B(x0,r)
M
(r|x|)
n
d (y) +
_
r
2
|x|
2
_
M
n
r
B(0,r)
1
(r|x|)
n+1
d (y)
=
2 |x|
n
r
M
(r|x|)
n
n
r
n1
+
_
r
2
|x|
2
_
M
n
r
1
(r|x|)
n+1
n
r
n1
and these terms converge to 0 as r . Since the inequality holds for all r > |x| , it
follows
u(x)
x
k
= 0. Similarly all the other partial derivatives equal zero as well and so u
is a constant. This proves the theorem.
12.4 Laplaces Equation For General Sets
Here I will consider the Laplace equation with Dirichlet boundary conditions on a general
bounded open set, U. Thus the problem of interest is
u = 0 on U, and u = g on U.
I will be presenting Perrons method for this problem. This method is based on exploit-
ing properties of subharmonic functions which are functions satisfying the following
denition.
Denition 12.4.1 Let U be an open set and let u be a function dened on U.
Then u is subharmonic if it is continuous and for all x U,
u(x)
1
nr
n1
B(x,r)
u(y) d (12.23)
whenever r is small enough.
Compare with Corollary 12.2.9.
12.4.1 Properties Of Subharmonic Functions
The rst property is a maximum principle. Compare to Theorem 12.2.1.
Theorem 12.4.2 Suppose U is a bounded open set and u is subharmonic on U
and continuous on U. Then
max
_
u(y) : y U
_
= max {u(y) : y U} .
Proof: Suppose x U and u(x) = max
_
u(y) : y U
_
M. Let V denote
the connected component of U which contains x. Then since u is subharmonic on
V, it follows that for all small r > 0, u(y) = M for all y B(x, r) . Therefore,
there exists some r
0
> 0 such that u(y) = M for all y B(x, r
0
) and this shows
{x V : u(x) = M} is an open subset of V. However, since u is continuous, it is also a
closed subset of V. Therefore, since V is connected,
{x V : u(x) = M} = V
and so by continuity of u, it must be the case that u(y) = M for all y V U.
This proves the theorem because M = u(y) for some y U.
As a simple corollary, the proof of the above theorem shows the following startling
result.
Corollary 12.4.3 Suppose U is a connected open set and that u is subharmonic on
U. Then either
u(x) < sup {u(y) : y U}
for all x U or
u(x) sup {u(y) : y U}
for all x U.
The next result indicates that the maximum of any nite list of subharmonic func-
tions is also subharmonic.
Lemma 12.4.4 Let U be an open set and let u
1
, u
2
, , u
p
be subharmonic functions
dened on U. Then letting
v max (u
1
, u
2
, , u
p
) ,
it follows that v is also subharmonic.
Proof: Let x U. Then whenever r is small enough to satisfy the subharmonicity
condition for each u
i
.
v (x) = max (u
1
(x) , u
2
(x) , , u
p
(x))
max
_
1
n
r
n1
B(x,r)
u
1
(y) d (y) , ,
1
n
r
n1
B(x,r)
u
p
(y) d (y)
_
n
r
n1
B(x,r)
max (u
1
, u
2
, , u
p
) (y) d (y) =
1
n
r
n1
B(x,r)
v (y) d (y) .
The next lemma concerns modifying a subharmonic function on an open ball in such
a way as to make the new function harmonic on the ball. Recall Corollary 12.2.8 which
I will list here for convenience.
Corollary 12.4.5 Let U = B(x
0
, r) and let g C (U) . Then there exists a unique
solution u C
2
(U) C
_
U
_
to the problem
u = 0 in U, u = g on U.
u(x) =
1
n
r
U
g (y)
r
2
|x x
0
|
2
|y x|
n
d (y) (12.24)
for every n 2. Here
2
= 2.
Denition 12.4.6 Let U be an open set and let u be subharmonic on U. Then
for B(x
0
,r) U dene
u
x0,r
(x)
_
u(x) if x / B(x
0
, r)
1
nr
B(x0,r)
u(y)
r
2
|xx
0
|
2
|yx|
n d (y) if x B(x
0
, r)
Thus u
x0,r
is harmonic on B(x
0
, r) , and equals to u o B(x
0
, r) . The wonderful
thing about this is that u
x0,r
is still subharmonic on all of U. Also note that from
Corollary 12.2.9 on Page 335 every harmonic function is subharmonic.
Lemma 12.4.7 Let U be an open set and B(x
0
,r) U as in the above denition.
Then u
x0,r
is subharmonic on U and u u
x0,r
.
Proof: First I show that u u
x0,r
. This follows from the maximum principle. Here
is why. The function uu
x0,r
is subharmonic on B(x
0
, r) and equals zero on B(x
0
, r) .
Here is why: For z B(x
0
, r) ,
u(z) u
x0r
(z) = u(z)
1
n1
B(z,)
u
x0,r
(y) d (y)
for all small enough. This is by the mean value property of harmonic functions and
the observation that u
x0r
is harmonic on B(x
0
, r) . Therefore, from the fact that u is
subharmonic,
u(z) u
x0r
(z)
1
n1
B(z,)
(u(y) u
x0,r
(y)) d (y)
Therefore, for all x B(x
0
, r) ,
u(x) u
x0,r
(x) 0.
The two functions are equal o B(x
0
, r) .
The condition for being subharmonic is clearly satised at every point, x / B(x
0
,r).
It is also satised at every point of B(x
0
,r) thanks to the mean value property, Corollary
12.2.9 on Page 335. It is only at the points of B(x
0
,r) where the condition needs to
be checked. Let z B(x
0
,r) . Then since u is given to be subharmonic, it follows that
for all r small enough,
u
x0,r
(z) = u(z)
1
n
r
n1
B(x0,r)
u(y) d
n
r
n1
B(x0,r)
u
x0,r
(y) d.
Denition 12.4.8 For U a bounded open set and g C (U), dene
w
g
(x) sup {u(x) : u S
g
}
where S
g
consists of those functions u which are subharmonic with u(y) g (y) for all
y U and u(y) min {g (y) : y U} m.
Note that S
g
= because u(x) m is a member of S
g
. Also all functions in S
g
have
values between m and max {g (y) : y U}. The fundamental result is the following
absolutely amazing incredible result.
Proposition 12.4.9 Let U be a bounded open set and let g C (U). Then w
g
S
g
and in addition to this, w
g
is harmonic.
Proof: Let B(x
0
, 2r) U and let {x
k
}
k=1
denote a countable dense subset of
B(x
0
, r). Let {u
1k
} denote a sequence of functions of S
g
with the property that
lim
k
u
1k
(x
1
) = w
g
(x
1
) .
By Lemma 12.4.7, it can be assumed each u
1k
is a harmonic function in B(x
0
, 2r) since
otherwise, you could use the process of replacing u with u
x0,2r
. Similarly, for each l,
there exists a sequence of harmonic functions in S
g
, {u
lk
} with the property that
lim
k
u
lk
(x
l
) = w
g
(x
l
) .
Now dene
w
k
= (max (u
1k
, , u
kk
))
x0,2r
.
Then each w
k
S
g
, each w
k
is harmonic in B(x
0
, 2r), and for each x
l
,
lim
k
w
k
(x
l
) = w
g
(x
l
) .
For x B(x
0
, r)
w
k
(x) =
1
n
2r
B(x0,2r)
w
k
(y)
r
2
|x x
0
|
2
|y x|
n
d (y) (12.25)
and so there exists a constant, C which is independent of k such that for all i =
1, 2, , n and x B(x
0
, r),
w
k
(x)
x
i
C
Therefore, this set of functions, {w
k
} is equicontinuous on B(x
0
, r) as well as being
uniformly bounded and so by the Ascoli Arzela theorem, it has a subsequence which
converges uniformly on B(x
0
, r) to a continuous function I will denote by w which has
the property that for all k,
w(x
k
) = w
g
(x
k
) (12.26)
Also since each w
k
is harmonic,
w
k
(x) =
1
n
r
B(x0,r)
w
k
(y)
r
2
|x x
0
|
2
|y x|
n
d (y) (12.27)
Passing to the limit in 12.27 using the uniform convergence, it follows
w(x) =
1
n
r
B(x0,r)
w(y)
r
2
|x x
0
|
2
|y x|
n
d (y) (12.28)
which shows that w is also harmonic. I have shown that w = w
g
on a dense set. Also, it
follows that w(x) w
g
(x) for all x B(x
0
, r). It remains to verify these two functions
are in fact equal.
Claim: w
g
is lower semicontinuous on U.
Proof of claim: Suppose z
k
z. I need to verify that
lim inf
k
w
g
(z
k
) w
g
(z) .
Let > 0 be given and pick u S
g
such that w
g
(z) < u(z) . Then
w
g
(z) < u(z) = lim inf
k
u(z
k
) lim inf
k
w
g
(z
k
) .
Since is arbitrary, this proves the claim.
Using the claim, let x B(x
0
, r) and pick x
k
l
x where {x
k
l
} is a subsequence of
the dense set, {x
k
} . Then
w
g
(x) w(x) = lim inf
l
w(x
k
l
) = lim inf
l
w
g
(x
k
l
) w
g
(x) .
This proves w = w
g
and since w is harmonic, so is w
g
. This proves the proposition.
It remains to consider whether the boundary values are assumed. This requires an
additional assumption on the set, U. It is a remarkably mild assumption, however.
Denition 12.4.10 A bounded open set, U has the barrier condition at z U,
if there exists a function, b
z
called a barrier function which has the property that b
z
is
subharmonic on U, b
z
(z) = 0, and for all x U \ {z} , b
z
(x) < 0.
The main result is the following remarkable theorem.
Theorem 12.4.11 Let U be a bounded open set which has the barrier condition
at z U and let g C (U) . Then the function, w
g
, dened above is in C
2
(U) and
satises
w
g
= 0 in U,
lim
xz
w
g
(x) = g (z) .
Proof: From Proposition 12.4.9 it follows w
g
= 0. Let z U and let b
z
be the
barrier function at z. Then letting > 0 be given, the function
u
(x) max (g (z) +Kb

z
(x) , m)
is subharmonic for all K > 0.
Claim: For K large enough, g (z) +Kb
z
(x) g (x) for all x U.
Proof of claim: Let > 0 and let B
= max {b
z
(x) : x U \ B(z, )} . Then
B
< 0 by assumption and the compactness of U \ B(z, ) . Choose > 0 small

enough that if |x z| < , then g (x) g (z) + > 0. Then for |x z| < ,
b
z
(x)
g (x) g (z) +
K
for any choice of positive K. Now choose K large enough that B
<
g(x)g(z)+
K
for all
x U. This can be done because B
< 0. It follows the above inequality holds for all

x U. This proves the claim.
Let K be large enough that the conclusion of the above claim holds. Then, for all
x, u
(x) g (x) for all x U and so u
S
g
which implies u
w
g
and so
g (z) +Kb
z
(x) w
g
(x) . (12.29)
This is a very nice inequality and I would like to say
lim
xz
g (z) +Kb
z
(x) = g (z)
lim inf
xz
w
g
(x)
lim sup
xz
w
g
(x) = w
g
(z) g (z)
but this would be wrong because I do not know that w
g
is continuous at a boundary
point. I only have shown that it is harmonic in U. Therefore, a little more is required.
Let
u
+
(x) g (z) + Kb
z
(x) .
Then u
+
is subharmonic and also if K is large enough, it follows from reasoning similar
to that of the above claim that
u
+
(x) = g (z) +Kb
z
(x) g (x)
on U. Therefore, letting u S
g
, u u
+
is a subharmonic function which satises for
x U,
u(x) u
+
(x) g (x) g (x) = 0.
Consequently, the maximum principle implies u u
+
and so since this holds for every
u S
g
, it follows
w
g
(x) u
+
(x) = g (z) + Kb
z
(x) .
It follows that
g (z) +Kb
z
(x) w
g
(x) g (z) + Kb
z
(x)
and so,
g (z) lim inf
xz
w
g
(x) lim sup
xz
w
g
(x) g (z) +.
lim
xz
w
g
(x) = g (z) .
12.4.2 Poissons Problem Again
Corollary 12.4.12 Let U be a bounded open set which has the barrier condition and
let f C
_
U
_
, g C (U). Then there exists at most one solution, u C
2
(U) C
_
U
_
to Poissons problem. If there is a solution, then it is of the form
u(x) =
1
(n 2)
n
_
U
G(x, y) f (y) dy +
U
g (y)
G
n
y
(x, y) d (y)
_
, if n 3, (12.30)
u(x) =
1
2
_
U
g (y)
G
n
y
(x, y) d +
U
G(x, y) f (y) dx
_
, if n = 2 (12.31)
for G(x, y) = r
n
(y x)
x
(y) where
x
is a function which satises
x
C
2
(U)
C
_
U
_

x
= 0,
x
(y) = r
n
(x y) for y U.
Furthermore, if u is given by the above representations, then u is a weak solution to
Poissons problem.
Proof: Uniqueness follows from Corollary 12.2.2 on Page 328. If u
1
and u
2
both
solve the Poisson problem, then their dierence, w satises
w = 0, in U, w = 0 on U.
The same arguments used earlier show that the representations in 12.30 and 12.31 both
yield a weak solution to Poissons problem.
The function, G in the above representation is called Greens function. Much more
can be said about the Greens function.
How can you recognize that a bounded open set, U has the barrier condition? One
way would be to check the following condition.
Condition 12.4.13 For each z U, there exists x
z
/ U such that |x
z
z| < |x
z
y|
for every y U \ {z} .
Proposition 12.4.14 Suppose Condition 12.4.13 holds. Then U satises the bar-
rier condition.
Proof: For n 3, let b
z
(y) r
n
(y x
z
) r
n
(z x
z
). Then b
z
(z) = 0 and if y
U with y = z, then clearly b
z
(y) < 0. For n = 2, let b
z
(y) = ln |y x
z
|+ln |z x
z
| .
This works out the same way.
Here is a picture of a domain which satises the barrier condition.
In fact, you have to have a fairly pathological example in order to nd something
which does not satisfy the barrier condition. You might try to think of some examples.
Think of B(0, 1) \ {z axis} for example. The points on the z axis which are in B(0, 1)
become boundary points of this new set. Thus this set cant satisfy the above condition.
Could this set have the barrier property?
The Jordan Curve Theorem
This short chapter is devoted to giving an elementary proof of the Jordan curve theorem
which is independent of the chapter on degree theory. I am following lecture notes from
a topology course given by Fernley at BYU in the 1970s. The ideas used in this
presentation are elementary and also lead to more general notions in algebraic topology.
In addition to this, these techniques are very useful in complex analysis.
Denition 13.0.15 A grating G is a nite set of horizontal and vertical lines,
each of which separate the plane. The grating divides the plane into two dimensional
domains the closures of which are called 2 cells of G. The 1 cells of G are the edges of
the 2 cells and the 0 cells of G are the end points of the 1 cells.
2 cell
T
1 cell
T
2 cell
T 2 cell
0 cell
For k = 0, 1, 2, one speaks of k chains. For {a

j
}
n
j=1
a set of k cells, the k chain is
denoted as a formal sum
C = a
1
+a
2
+ +a
n
where the sum is taken modulo 2. The sums are just formal expressions like the above.
Thus for a a k cell, a +a = 0, 0 +a = a, the summation sign is commutative. In other
words, if a k cell is repeated an even number of times in the formal sum, it disappears
resulting in 0 dened by 0 +a = a +0 = a. For a a k cell, |a| denotes the points of the
plane which are contained in a. For a k chain, C as above,
|C| {x : x |a
j
| for some a
j
}
so |C| is the union of the k cells in the sum remembering that when a k cell occurs
twice, it is gone and does not contribute to |C|.
The following picture illustrates the above denition. The following is a picture of
the 2 cells in a 2 chain. The dotted lines indicate the lines in the grating.
349
350 THE JORDAN CURVE THEOREM
Now the following is a picture of the 1 chain consisting of the sum of the 1 cells
which are the edges of the above 2 cells. Remember when a 1 cell is added to itself, it
disappears from the chain. Thus if you add up the 1 cells which are the edges of the
above 2 cells, lots of them cancel o. In fact all the edges which are shared between two
2 cells disappear. The following is what results.
Denition 13.0.16 Next the boundary operator is dened. This is denoted by
. takes k cells to k 1 chains. If a is a 2 cell, then a consists of the edges of a.
If a is a 1 cell, then a consists of the ends of the 1 cell. If a is a 0 cell, then a 0.
This extends in a natural way to k chains. For
C = a
1
+a
2
+ +a
n
,
C a
1
+a
2
+ +a
n
A k chain C is called a cycle if C = 0.
In the second of the above pictures, you have a 1 cycle. Here is a picture of another
one in which the boundary of another 2 cell has been included over on the right.
This 1 cycle shown above is the boundary of exactly two 2 chains. What are they?
C
1
consists of the 2 cells in the rst picture above along with the 2 cell whose boundary
is the 1 cycle over on the right. C
2
is all the other 2 cells of the grating. You see this
clearly works. Could you make that 2 cell on the right be in C
2
? No, you couldnt do
it. This is because the 1 cells which are shown would disappear, being listed twice.
This illustrates the fundamental lemma of the plane which comes next.
Lemma 13.0.17 If C is a bounded 1 cycle (C = 0), then there are exactly two 2
chains D
1
, D
2
such that
C = D
1
= D
2
.
Proof : The lemma is vacuously true unless there are at least two vertical lines and
at least two horizontal lines in the grating G. It is also obviously true if there are exactly
two vertical lines and two horizontal lines in G. Suppose the theorem is true for n lines
in G. Then as just mentioned, there is nothing to prove unless there are either 2 or more
351
vertical lines and two or more horizontal lines. Suppose without loss of generality there
are at least as many veritical lines are there are horizontal lines and that this number
is at least 3. If it is only two, there is nothing left to show. Let l be the second vertical
line from the left. Let {e
1
, , e
m
} be the 1 cells of C with the property that |e
j
| l.
Note that e
j
occurs only once in C since if it occurred twice, it would disappear because
of the rule for addition. Pick one of the 2 cells adjacent to e
j
, b
j
and add in b
j
which
is a 1 cycle. Thus
C +
j
b
j
is a bounded 1 cycle and it has the property that it has no 1 cells contained in l. Thus
you could eliminate l from the grating G and all the 1 cells of the above 1 chain are edges
of the grating G \ {l}. By induction, there are exactly two 2 chains D
1
, D
2
composed
of 2 cells of G\ {l} such that for i = 1, 2,
D
i
= C +
j
b
j
(13.1)
Since none of the 2 cells of D
i
have any edges on l, one can add l back in and regard D
1
and D
2
as 2 chains in G. Therefore, adding
j
b
j
to both sides of the above yields
C = D
i
+
j
b
j
=
_
_
D
i
+
j
b
j
_
_
, i = 1, 2.
and this shows there exist two 2 chains which have C as the boundary. If D
i
= C,
then
D
i
+
j
b
j
=
_
_
D
i
+
j
b
j
_
_
= C +
j
b
j
and by induction, there are exactly two 2 chains which D
i
+
j
b
j
can equal. Thus
adding
j
b
j
there are exactly two 2 chains which D
i
can equal.
Here is another proof which is not by induction. This proof also gives an algorithm
for identifying the two 2 chains. The 1 cycle is bounded and so every 1 cell in it is part
of the boundary of a 2 cell which is bounded. For the unbounded 2 cells on the left,
label them all as A. Now starting from the left and moving toward the right, toggle
between A and B every time you hit a vertical 1 cell of C. This will label every 2 cell
with either A or B. Next, starting at the top, label all the unbounded 2 cells as A and
move down and toggle between A and B every time you encounter a horizontal 1 cell
of C. This also labels every 2 cell as either A or B. Suppose there is a contradiction
in the labeling. Pick the rst column in which a contradiction occurs and then pick the
top contradictory 2 cell in this column. There are various cases which can occur, each
leading to the existence of a vertex of C which is contained in an odd number of 1 cells
of C, thus contradicting the conclusion that C is a 1 cycle. In the following picture,
AB will mean the labeling from the left to right gives A and the labeling from top to
bottom yields B with similar modication for AA and BB.
BB
AB
AA
BB
AA
AB
AA
AA
A solid line indicates the corresponding 1 cell is in C. It is there because a change
took place either from top to bottom or from left to right. Note that in both of those
situations the vertex right in the middle of the crossed lines will occur in C and so C is
not a 1 cycle. There are 8 similar pictures you can draw and in each case this happens.
The vertex in the center gets added in an odd number of times. You can also notice
that if you start with the contradictory 2 cell and move counter clockwise, crossing 1
cells as you go and starting with B, you must end up at A as a result of crossing 1 cells
of C and this requires crossing either one or three of these 1 cells of C.
AB
Thus that center vertex is a boundary point of C and so C is not a 1 cycle after
all. Similar considerations would hold if the contradictory 2 cell were labeled BA. Thus
there can be no contradiction in the two labeling schemes. They label the 2 cells in G
either A or B in an unambiguous manner.
The labeling algorithm encounters every 1 cell of C (in fact of G) and gives a label
to every 2 cell of G. Dene the two 2 chains as A and B where A consists of those
labeled as A and B those labeled as B. The 1 cells which cause a change to take place
in the labeling are exactly those in C and each is contained in one 2 cell from A and one
2 cell from B. Therefore, each of these 1 cells of C appears in A and B which shows
C A and C B. On the other hand, if l is a 1 cell in A, then it can only occur
in a single 2 cell of A and so the 2 cell adjacent to that one along l must be in B and so
l is one of the 1 cells of C by denition. As to uniqueness, in moving from left to right,
you must assign adjacent 2 cells joined at a 1 cell of C to dierent 2 chains or else the
1 cell would not appear when you take the boundary of either A or B since it would be
added in twice. Thus there are exactly two 2 chains with the desired property.
The next lemma is interesting because it gives the existence of a continuous curve
joining two points.
y
x
Lemma 13.0.18 Let C be a bounded 1 chain such that C = x +y. Then both x, y
are contained in a continuous curve which is a subset of |C|.
Proof : There are an odd number of 1 cells of C which have x at one end. Otherwise
C = x + y. Begin at x and move along an edge leading away from x. Continue till
there is no new edge to travel along. You must be at y since otherwise, you would have
found another boundary point of C. This point would be in either one or three one cells
of C. It cant be x because x is contained in either one or three one cells of C. Thus,
there is always a way to leave x if the process returns to it. IT follows that there is a
continuous curve in |C| joining x to y.
353
The next lemma gives conditions under which you can go around a couple of closed
sets. It is called Alexanders lemma. The following picture is a rough illustration of the
situation. Roughly, it says that if you can mis F
1
and you can mis F
2
in going from x
to y, then you can mis both F
1
and F
2
by climbing around F
1
.
F
1 F
2
y
x
C
1
C
2
Lemma 13.0.19 Let F
1
be compact and F
2
closed. Suppose C
1
, C
2
are two bounded
1 chains in a grating which has no unbounded two cells having nonempty intersection
with F
1
.Suppose C
i
= x + y where x, y / F
1
F
2
. Suppose C
2
does not intersect F
2
and C
1
does not intersect F
1
. Also suppose the 1 cycle C
1
+ C
2
bounds a 2 chain D
for which |D| F
1
F
2
= . Then there exists a 1 chain C such that C = x + y
and |C| (F
1
F
2
) = . In particular x, y cannot be in dierent components of the
complement of F
1
F
2
.
Proof: Let a
1
, a
2
, , a
m
be the 2 cells of D which intersect the compact set F
1
.
Consider
C C
2
+
k
a
k
.
This is a 1 chain and C = x+y because a
k
= 0. Then |a
k
| F
2
= . This is because
|a
k
| F
1
= and none of the 2 cells of D intersect both F
1
and F
2
by assumption.
Therefore, C is a bounded 1 chain which avoids intersecting F
2
.
Does it also avoid F
1
? Suppose to the contrary that l is a one cell of C which does
intersect F
1
. If |l| |C
1
+C
2
| , then it would be an edge of some 2 cell of D and would
have to be a 1 cell of C
2
since it intersects F
1
so it would have been added twice, once
from C
2
and once from
k
a
k
and therefore could not be a summand in C. Therefore, |l| is not in |C
1
+C
2
|. It follows
l must be an edge of some a
k
D and it is not a 1 cell of C
1
+C
2
. Therefore, if b is the
2 cell adjacent to a
k
, it must follow b D since otherwise l would be a 1 cell of C
1
+C
2
the boundary of D and this was just ruled out. But now it would follow that l would
occur twice in the above sum so l cannot be a summand of C. Therefore C misses F
1
also.
Here is another argument. Suppose |l| F
1
= . l C = C
2
+
k
a
k
. First note
that l / C
1
since |C
1
| F
1
= .
Case 1: l C
2
.
In this case it is in C
1
+ C
2
because, as just noted it is not in C
1
. Therefore, there
exists a D such that l is an edge of a and is not in the two cell adjacent to a. But
this would require l to disappear since it would occur in both C
2
and
k
a
k
. Hence
l / C
2
.
Case 2: In this case l / C
2
. Then l is the edge of some a D which intersects F
1
.
Letting b be the two cell adjacent to a sharing l, then b cannot be in D since otherwise
l would occur twice in the above sum and would then disappear. Hence b / D and
l D = C
1
+C
2
but this cannot happen because l / C
1
and in this case l / C
2
either.

Lemma 13.0.20 Let C be a bounded 1 cycle such that |C| H = where H is a
connected set. Also let D, E be the 2 chains with D = C = E. Then either |H| |E|
or |H| |D| .
Proof: If p is a limit point of |E| and p |D| , then p must be contained in an
edge of some 2 cell of D since otherwise it could not be a limit point, being contained in
an open set whose intersection with |E| is empty. If p is a point of an even number of
edges of 2 cells of D, then it is likewise an interior point of |D| which cannot be a limit
point of |E| . Therefore, if p is a limit point of |E| , it must be the case that p |C| . A
similar observation holds for the case where p |E| and is a limit point of |D|. Thus if
H |D| and H |E| are both nonempty, then they separate the connected set H and
so H must be a subset of one of |D| or |E|.
Denition 13.0.21 A Jordan arc is a set of points of the form r ([a, b])
where r is a one to one map from [a, b] to the plane. For p, q , say p < q if
p = r (t
1
) , q = r (t
2
) for t
1
< t
2
. Also let pq denote the arc r ([t
1
, t
2
]).
Theorem 13.0.22 Let be a Jordan arc. Then its complement is connected.
Proof: Suppose this is not so. Then there exists x, y points in
C
which are in
dierent components of
C
. Let G be a grating having x, y as points of intersection of a
horizontal line and a vertical line of G and let p, q be the points at the ends of the Jordan
arc. Also let G be such that no unbounded two cell has nonempty intersection with .
Let p = r (a) and q = r (b) . Now let z = r
_
a+b
2
_
and consider the two arcs pz and zq.
If C = x + y then it is required |C| = since otherwise these two points would
not be in dierent components. Suppose there exists C
1
, C
1
= x +y and |C
1
| zq =
and C
2
, C
2
= x + y but |C
2
| pz = . Then C
1
+ C
2
is a 1 cycle and so by Lemma
13.0.17 there are exactly two 2 chains whose boundaries are C
1
+ C
2
. Since z / |C
i
| ,
it follows z = pz zq can only be in one of these 2 chains because it is a single point.
Then by Lemma 13.0.19, Alexanders lemma, there exists C a 1 chain with C = x +y
and |C| (pz zq) = so by Lemma 13.0.18 x, y are not in dierent components of
C
contrary to the assumption they are in dierent components. Hence one of pz, zq has
the property that every 1 chain, C = x + y goes through it. Say every such 1 chain
goes through zq. Then let zq play the role of pq and conclude every 1 chain C such that
C = x +y goes through either zw or wq there
w = r
__
a +b
2
+b
_
1
2
_
Thus, continuing this way, there is a sequence of Jordan arcs p
k
q
k
where r (t
k
) = q
k
and r (s
k
) = p
k
with |t
k
s
k
| <
ba
2
k
, [s
k
, t
k
] [a, b] such that every C with C = x +y
has nonempty intersection with p
k
q
k
. The intersection of these arcs is r (s) where s =
k=1
[s
k
, t
k
]. Then all such C must go through r (s) because such C with C = x + y
must intersect p
k
q
k
for each k and their intersection is r (s). But now there is an obvious
contradiction to having every 1 chain whose boundary is x +y intersecting r (s).
r(s)
q
p
y
x
Pick a 1 chain whose boundary is x+y. Let D be the two chain of at most four 2 cells
consisting of those two cells which have r (s) on some edge. Then (C +D) = C =
355
x +y but r (s) / |C +D| . Therefore, this contradiction shows
C
must be connected
after all.
The other important observation about a Jordan arc is that it has no interior points.
This will follow later from a harder result but it is also easy to prove.
Lemma 13.0.23 Let = r ([a, b]) be a Jordan arc where r is as above, one to one,
onto and continuous. Then has no interior points.
Proof : Suppose to the contrary that has an interior point p. Then for some r > 0,
B(p, r) .
Consider the circles of radius < r centered at p. Denoting as C
one of these, it follows

the C
are disjoint. Therefore, since r is one to one, the sets r

1
(C
) are also disjoint.

Now r is continuous and one to one mapping to a compact set. Therefore, r
1
is also
continuous. It follows r
1
(C
) is connected and compact. Thus by Theorem 5.3.8 each

of these sets is a closed interval of positive length since r is one to one. It follows there
exist disjoint open nonempty intervals consisting of the interiors of r
1
(C
) , {I
}
<r
.
This is a contradiction to the density of Q and the fact that Q is at most countable.
Denition 13.0.24 Let r map [a, b] to the plane such that r is one to one on
[a, b) and (a, b] but r (a) = r (b). Then J = r ([a, b]) is called a simple closed curve. It is
also called a Jordan curve. Also since the term boundary has been given a specialized
meaning relative to chains of various sizes, we say x is in the frontier of S if every open
ball containing x contains points of S as well as points of S
C
.
Note that if J is a Jordan curve, then it is the union of two Jordan arcs whose
intersection is two distinct points of J. You could pick z (a, b) and consider r ([a, z])
and r ([z, b]) as the two Jordan arcs.
The next lemma gives a probably more convenient way of thinking about a Jordan
curve. It says essentially that a Jordan curve is a wriggly circle. First consider the
following simple lemma.
Lemma 13.0.25 Let K be a compact set in R
n
and let f : K R
m
be continuous
and one to one. Then f
1
: f (K) K is also continuous.
Proof: Suppose {f (k
n
)} is a convergent sequence in f (K) converging to f (k). Does
it follow that k
n
k? If not, there exists a subsequence {k
n
k
} which converges as
k to l = k. Then by continuity of f it follows f (k
n
k
) f (l) . Hence f (l) = f (k)
which violates the condition that f is one to one.
Lemma 13.0.26 J is a simple closed curve if and only if there exists a mapping
: S
1
J where S
1
is the unit circle
_
(x, y) : x
2
+y
2
= 1
_
,
such that is one to one and continuous.
Proof : Suppose that J is a simple closed curve so there is a parameterization r
and an interval [a, b] such that r is continuous and one to one on [a, b) and (a, b] with
r (a) = r (b) . Let C
0
= r ((a, b)) , C
= r ([a +, b ]) , and let S

1
denote the unit
circle. Let l be a linear one to one map from [a, b] onto [0, 2]. Consider the following
diagram.
[a, b]
l
[0, 2]
r R
C S
1
where R() (cos , sin ) . Then clearly R is continuous. It is also the case that,
from the above lemma, r
1
is continuous on C
. Therefore, since > 0 is arbitrary,

R l r
1
is a one to one and onto mapping from C
0
to S
1
\ (1, 0). Also, letting
p = r (a) = r (b) , it follows that (p) = (1, 0). It remains to verify that is continuous
at p. Suppose then that r (x
n
) p = r (a) = r (b). If (r (x
n
)) fails to converge
to (1, 0) = (p) , then there is a subsequence, still denoted as x
n
and > 0 such
that | (r (x
n
)) (p)| . In particular x
n
/ {a, b}. By the above lemma, r
1
is
continuous on r ([a, b)) since this is true for r ([a, b ]) for each > 0. Since p = r (a) ,
it follows that
| (r (x
n
)) (r (a))| = |R l (x
n
) R l (a)|
Hence there is some > 0 such that |x
n
a|
1
. Similarly, |x
n
b|
2
> 0. Letting
= min (
1
,
2
) , it follows that x
n
[a +, b ]. Taking a convergent subsequence,
still denoted as {x
n
} , there exists x [a +, b ] such that x
n
x. However, this
implies that r (x
n
) r (x) and so r (x) = r (a) = p, a contradiction to the fact that r
is one to one on [a, b).
Next suppose J is the image of the unit circle as just explained. Then let R :
[0, 2] S
1
be dened as R(t) (cos (t) , sin (t)) . Then consider r (t) (R(t)). r
is one to one on [0, 2) and (0, 2] with r (0) = r (2) and is continuous, being the
composition of continuous functions.
Before the proof of the Jordan curve theorem, recall Theorem 5.3.14 which says that
the connected components of an open sets are open and that an open connected set is
arcwise connected. If J is a Jordan curve then it is the continuous image of the compact
set S
1
and so J is also compact. Therefore, its complement is open and the connected
components of J
C
are connected. The following lemma is a fairly obvious conclusion
of this. A square curve is a continuous curve which consists entirely of line segments
which are either horizontal or vertical.
Lemma 13.0.27 Let U be a connected open set and let x, y be points of U. Then
there is a square curve which joins x and y.
Proof: Let V denote those points of U which can be joined to x by a square curve.
Then if z V, there exists B(z, r) U. It is clear that every point of B(z, r) can be
joined to z by a square curve. Also V
C
must be open since if z V
C
, B(z, r) U for
some r. Then if any w B(z, r) is in V, one could join w to z by a square curve and
conclude that z V after all. The fact that both V, V
C
are both open would result in
a contradiction unless both x, y V since otherwise, U is separated by V, V
C
.
Theorem 13.0.28 Let J be a Jordan curve in the plane. Then J
C
consists of
exactly two components, a bounded component, and an unbounded component, and J is
the frontier of both of these components. Furthermore, J has empty interior.
Proof : To begin with consider the claim there are no more than two components.
Suppose this is not so. Then there exist x, y, z each of which is in a dierent component
of J
C
. Let J = H K where H and K are two Jordan arcs joined at the points a and
b. If the Jordan curve is r ([c, d]) where r (c) = r (d) as described above, you could take
H = r
__
c,
c+d
2
_
and K = r
__
c+d
2
, d
_
. Thus the points on the Jordan curve illustrated
in the following picture could be
a = r (c) , b = r
_
c +d
2
_
357
K
a
H
b
First we show that there is at most two components in J
C
. Suppose to the contrary
that there exists x, y, z, each in a dierent component. By the Jordan arc theorem
above, and the above lemma about square curves, there exists a square curve C
xyH
such that C
xyH
= x +y and |C
x,yH
| H = . Using the same notation in relation to
the other points, there exist square curves in the following list.
C
xyH
, C
xyH
= x +y, C
yzH
, C
yzH
= y +z
C
xyK
, C
xyH
= x +y, C
yzK
, C
yzK
= y +z
Let these square curves be part of a grating which includes all vertices of all these
square curves and contains the compact set J in the bounded two cells. First note that
C
xyH
+C
xyK
is a one cycle and that
|C
xyH
+C
xyK
| (H K) =
Also note that H K = {a, b} since r is one to one on [c, d) and (c, d]. Therefore, there
exist unique two chains D, E such that D = E = C
xyH
+C
xyK
. Now if one of these
two chains contains both a, b then then the other two chain does not contain either a
nor b. Then by Alexanders lemma, Lemma 13.0.19, there would exist a square curve C
such that |C| (H K) = |C| J = and C = x+y which is assumed not to happen.
Therefore, one of the two chains contains a and the other contains b. Say a |D| and
b |E|. Similarly there exist unique two chains P, Q such that
P = Q = C
yzH
+C
yzK
where a |P| and b |Q|. Now consider
(D +Q) = C
xyH
+C
xyK
+C
yzH
+C
yzK
= (C
xyH
+C
yzH
) + (C
xyK
+C
yzK
)
This is a one cycle because its boundary is x + y + y + z + x + y + y + z = 0. By
Lemma 13.0.17, the fundamental lemma of the plane, there are exactly two two chains
whose boundaries equal this one cycle. Therefore, D + Q must be one of them. Also
b |Q| and is not in |D| . Hence b |D +Q|. Similarly a |D +Q|. It follows that the
other two chain whose boundary equals the above one cycle contains neither a nor b. In
addition to this, C
xyH
+C
yzH
misses H and C
xyK
+C
yzK
misses K. Both of these one
chains have boundary equal to x + z. By Alexanders lemma, there exists a one chain
C which misses both H and K (all of J) such that C = x + z which contradicts the
assertion that x, z are in dierent components. This proves the assertion that there are
only two components to J
C
.
Next, why are there at least two components in J
C
? Suppose there is only one and
let a, b be the points of J described above and H, K also as above. Let Q be a small
square 1 cycle which encloses a on its inside such that b is not inside Q. Thus a is on
the inside of Q and b is on the outside of Q as shown in the picture.
K
a
Q
H
b
Now let G be a grating which has the corners of Q as points of intersection of
horizontal and vertical lines and also has all the 2 cells so small that none of them
can intersect both of the disjoint compact sets H |Q| and |Q| K. Let P be the
1 cells contained in Q which have nonempty intersection with H. Some of them must
have nonempty intersection with H because if not, then H would fail to be connected,
having points inside Q, a, and points outside Q, b, but no points on Q. Similarly some
of these one cells in Q have nonempty intersection with K. Let P = x
1
+ + x
m
.
Then it follows each x
k
/ H. Could P = 0? Suppose P = 0. If l is a one cell of
P, then since its ends are not in P, the two adjacent one cells to l which are in Q
must also intersect H. Moving counterclockwise around Q, it would follow that all the
one cells contained in Q would intersect H. However, at least one must intersect K
because if not, a is a point of K inside the square Q while b is a point of K outside
Q thus separating K which is a connected set. However, this contradicts the choice of
the grating. Therefore, P = 0. Now this violates the assumption that no 2 cell of G
can intersect both of those disjoint compact sets H |Q| and |Q| K. Starting with
a one cell of Q which does not intersect H, move counter clockwise till you obtain the
rst one which intersects H. This will produce a point of P. Then the next point of
P will occur when the rst one cell of P which does not intersect H is encountered.
Thus a pair of points in P are obtained. Now you are in the same position as before,
continue moving counter clockwise and obtaining pairs of points of P till there are no
more one cells of Q which intersect H. You must have encountered an even number of
points for P.
Since it is assumed there is only one component of J
C
, it follows upon rening G
if necessary, there exist 1 chains B
k
contained in J
C
such that B
k
= x
1
+ x
k
and
it is the existence of these B
k
which will give the desired contradiction. Let
B = B
2
+ +B
m
. Then P +B is a 1 cycle which misses K. It is a one cycle because
m is even.
(P +B) =
m=2l
k=2
x
1
+x
k
+
2l
k=1
x
k
=
2l
k=1
x
k
+
2l
k=2
x
k
+x
k
= 0
It misses K because B misses J and all the 1 cells of P in the original grating G intersect
H |Q| so they cannot intersect K. Also P +Q+B is a 1 cycle which misses H. This
is because B misses J and every 1 cell of P which intersects H disappears because
P + Q causes them to be added twice. Since H and K are connected, it follows from
Lemma 13.0.20, 13.0.17 that P + B bounds a 2 chain D which is contained entirely in
K
C
(the one which does not contain K). Similarly P +Q+B bounds a 2 chain E which
is contained in H
C
(the 2 chain which does not contain H). Thus D + E is a 2 chain
which does not contain either a or b. (D misses K and E misses H and {a, b} = HK)
However,
(D +E) = P +B +P +Q+B = Q
359
and so D + E is one of the 2 chains of Lemma 13.0.17 which have Q as boundary.
However, Q bounds the 2 chain of 2 cells which are inside Q which contains a and the
2 chain of 2 cells which are outside Q which contains b. This is a contradiction because
neither of these 2 chains miss both a and b and this shows there are two components of
J
C
.
In the above argument, if each pair {x
1
, x
i
} can be joined by a square curve B
i
which lies in J
C
, then the contradiction was obtained. Therefore, there must exist a
pair {x
1
, x
i
} which cant be joined by any square curve in J
C
and this requires these
points to be in dierent components by Lemma 13.0.27 above. Since they are both on
Q and Q could be as small as desired, this shows a is in the frontier of both components
of J
C
. Furthermore, a was arbitrary so every point of J is a frontier point of both the
components of J
C
. These are the only frontier points because the components of J
C
are open.
By Lemma 13.0.26, J is the continuous image of the compact set S
1
so it follows J is
bounded. The unbounded component of J
C
is the one which contains the connected set
B(0, R)
C
where J B(0, R). Thus there are two components for J
C
, the unbounded
one which contains B(0, R)
C
and the bounded one which must be contained in B(0, R) .
Line Integrals
14.1 Basic Properties
14.1.1 Length
I will give a discussion of what is meant by a line integral which is independent of the
earlier material on Lebesgue integration. Line integrals are of fundamental importance
in physics and in the theory of functions of a complex variable.
Denition 14.1.1 Let : [a, b] R
n
be a function. Then is of bounded
variation if
sup
_
n
i=1
| (t
i
) (t
i1
)| : a = t
0
< < t
n
= b
_
V (, [a, b]) <
where the sums are taken over all possible lists, {a = t
0
< < t
n
= b} . The set of
points traced out will be denoted by
([a, b]). The function is called a param-

eterization of
. The set of points
is called a rectiable curve. If a set of points
= ([a, b]) where is continuous and is one to one on [a, b) and also one to one
on (a, b], then
is called a simple curve. A closed curve is one which has a parame-

terization dened on an interval [a, b] such that (a) = (b). It is a simple closed
curve if there is a parameterization g such that is one to one on [a, b) and one to one
on (a, b] with (a) = (b).
The case of most interest is for simple curves. It turns out that in this case, the above
concept of length is a property which
possesses independent of the parameterization

used to describe the set of points
. To show this, it is helpful to use the following

lemma.
Lemma 14.1.2 Let : [a, b] R be a continuous function and suppose is 11 on
(a, b). Then is either strictly increasing or strictly decreasing on [a, b] . Furthermore,
1
is continuous.
Proof: First it is shown that is either strictly increasing or strictly decreasing on
(a, b) .
If is not strictly decreasing on (a, b), then there exists x
1
< y
1
, x
1
, y
1
(a, b) such
that
((y
1
) (x
1
)) (y
1
x
1
) > 0.
If for some other pair of points, x
2
< y
2
with x
2
, y
2
(a, b) , the above inequality does
not hold, then since is 1 1,
((y
2
) (x
2
)) (y
2
x
2
) < 0.
361
362 LINE INTEGRALS
Let x
t
tx
1
+(1 t) x
2
and y
t
ty
1
+(1 t) y
2
. Then x
t
< y
t
for all t [0, 1] because
tx
1
ty
1
and (1 t) x
2
(1 t) y
2
with strict inequality holding for at least one of these inequalities since not both t and
(1 t) can equal zero. Now dene
h(t) ((y
t
) (x
t
)) (y
t
x
t
) .
Since h is continuous and h(0) < 0, while h(1) > 0, there exists t (0, 1) such that
h(t) = 0. Therefore, both x
t
and y
t
are points of (a, b) and (y
t
) (x
t
) = 0 contra-
dicting the assumption that is one to one. It follows is either strictly increasing or
strictly decreasing on (a, b) .
This property of being either strictly increasing or strictly decreasing on (a, b) carries
over to [a, b] by the continuity of . Suppose is strictly increasing on (a, b) , a similar
argument holding for strictly decreasing on (a, b) . If x > a, then pick y (a, x) and
from the above, (y) < (x) . Now by continuity of at a,
(a) = lim
xa+
(z) (y) < (x) .
Therefore, (a) < (x) whenever x (a, b) . Similarly (b) > (x) for all x (a, b).
It only remains to verify
1
is continuous. Suppose then that s
n
s where s
n
and s are points of ([a, b]) . It is desired to verify that
1
(s
n
)
1
(s) . If this
does not happen, there exists > 0 and a subsequence, still denoted by s
n
such that
1
(s
n
)
1
(s)
. Using the sequential compactness of [a, b] there exists a further

subsequence, still denoted by n, such that
1
(s
n
) t
1
[a, b] , t
1
=
1
(s) . Then by
continuity of , it follows s
n
(t
1
) and so s = (t
1
) . Therefore, t
1
=
1
(s) after
all. This proves the lemma.
Now suppose and are two parameterizations of the simple curve
as described
above. Thus ([a, b]) =
= ([c, d]) and the two continuous functions , are one

to one on their respective open intervals. I need to show the two denitions of length
yield the same thing with either parameterization. Since
is compact, it follows
from Theorem 5.1.3 on Page 88, both
1
and
1
are continuous. Thus
1
:
[c, d] [a, b] is continuous. It is also uniformly continuous because [c, d] is compact. Let
P {t
0
, , t
n
} be a partition of [a, b] , t
0
< t
1
< < t
n
such that for L < V (, [a, b]) ,
L <
n
k=1
| (t
k
) (t
k1
)| V (, [a, b])
Note the sums approximating the total variation are all no larger than the total variation
because when another point is added in to the partition, it is an easy exercise in the
triangle inequality to show the corresponding sum either becomes larger or stays the
same.
Let
1
(s
k
) = t
k
so that {s
0
, , s
n
} is a partition of [c, d] . By the lemma, the
s
k
are either strictly decreasing or strictly increasing as a function of k, depending on
whether
1
is increasing or decreasing. Thus (t
k
) = (s
k
) and so
L <
n
k=1
| (s
k
) (s
k1
)| V (, [a, b])
It follows that whenever L < V (, [a, b]) , there exists a partition of [c, d] , {s
0
, , s
n
}
such that
L <
n
k=1
| (s
k
) (s
k1
)|
14.1. BASIC PROPERTIES 363
It follows that for every L < V (, [a, b]) , V (, [c, d]) L which requires V (, [c, d])
V (, [a, b]). Turning the argument around, it follows
V (, [c, d]) = V (, [a, b]) .
This proves the following fundamental theorem.
Theorem 14.1.3 Let be a simple curve and let be a parameterization for
where is one to one on (a, b), continuous on [a, b] and of bounded variation. Then
the total variation
V (, [a, b])
can be used as a denition for the length of in the sense that if = ([c, d]) where
is a continuous function which is one to one on (c, d) with ([c, d]) = ,
V (, [a, b]) = V (, [c, d]) .
This common value can be denoted by V () and is called the length of .
The length is not dependent on parameterization. Simple curves which have such
parameterizations are called rectiable.
14.1.2 Orientation
There is another notion called orientation. For simple rectiable curves, you can think
of it as a direction of motion over the curve but what does this really mean for a wriggly
curve? A precise description is needed.
Denition 14.1.4 Let , be continuous one to one parameterizations for a
simple rectiable curve. If
1
is increasing, then and are said to be equivalent
parameterizations and this is written as . It is also said that the two parameteri-
zations give the same orientation for the curve when .
When the parameterizations are equivalent, they preserve the direction of motion
along the curve and this also shows there are exactly two orientations of the curve
since either
1
is increasing or it is decreasing thanks to Lemma 14.1.2. In simple
language, the message is that there are exactly two directions of motion along a simple
curve.
Lemma 14.1.5 The following hold for .
, (14.1)
If then , (14.2)
If and , then . (14.3)
Proof: Formula 14.1 is obvious because
1
(t) = t so it is clearly an increasing
function. If then
1
is increasing. Now
1
must also be increasing because
it is the inverse of
1
. This veries 14.2. To see 14.3,
1
=
_
1

_
1

_
and so since both of these functions are increasing, it follows
1
is also increasing.
Denition 14.1.6 Let be a simple rectiable curve and let be a parame-
terization for . Denoting by [] the equivalence class of parameterizations determined
by the above equivalence relation, the following pair will be called an oriented curve.
(, [])
In simple language, an oriented curve is one which has a direction of motion specied.
364 LINE INTEGRALS
Actually, people usually just write and there is understood a direction of motion
or orientation on . How can you identify which orientation is being considered?
Proposition 14.1.7 Let (, []) be an oriented simple curve and let p, q be any
two distinct points of . Then [] is determined by the order of
1
(p) and
1
(q).
This means that [] if and only if
1
(p) and
1
(q) occur in the same order as
1
(p) and
1
(q).
Proof: Suppose
1
(p) <
1
(q) and let [] . Is it true that
1
(p) <
1
(q)? Of course it is because
1
is increasing. Therefore, if
1
(p) >
1
(q)
it would follow
1
(p) =
1

_
1
(p)
_
>
1

_
1
(q)
_
=
1
(q)
which is a contradiction. Thus if
1
(p) <
1
(q) for one [] , then this is true
for all [].
Now suppose is a parameterization for dened on [c, d] which has the property
that
1
(p) <
1
(q)
Does it follow []? Is
1
increasing? By Lemma 14.1.2 it is either increasing
or decreasing. Thus it suces to test it on two points of [c, d] . Pick the two points
1
(p) ,
1
(q) . Is
1

_
1
(p)
_
<
1

_
1
(q)
_
?
Yes because these reduce to
1
(p) on the left and
1
(q) on the right. It is given
that
1
(p) <
1
(q) . This proves the lemma.
This shows that the direction of motion on the curve is determined by any two
points and the determination of which is encountered rst by any parameterization in
the equivalence class of parameterizations which determines the orientation. Sometimes
people indicate this direction of motion by drawing an arrow.
Now here is an interesting observation relative to two simple closed rectiable curves.
The situation is illustrated by the following picture.
1
() =
2
()
1
() =
2
()
2
l
1
(a) =
1
(b)
1
([a, ])
1
([, b])
1
([, ])
2
([, ])
2
([, d])
2
([c, ])
2
(c) =
2
(d)
Proposition 14.1.8 Let
1
and
2
be two simple closed rectiable oriented curves
and let their intersection be l. Suppose also that l is itself a simple curve. Also suppose
the orientation of l when considered a part of
1
is opposite its orientation when consid-
ered a part of
2
. Then if the open segment (l except for its endpoints) of l is removed,
the result is a simple closed rectiable curve . This curve has a parameterization
with the property that on
1
j
(
j
) ,
1
j
is increasing. In other words, has an
orientation consistent with that of
1
and
2
. Furthermore, if has such a consistent
14.1. BASIC PROPERTIES 365
orientation, then the orientations of l as part of the two simple closed curves,
1
and
2
are opposite.
1
() =
2
()
1
() =
2
()
2
Proof: Let
1
=
1
([a, b]) ,
1
(a) =
1
(b) , and
2
=
2
([c, d]) ,
2
(c) =
2
(d) ,
with l =
1
([, ]) =
2
([, ]). (Recall continuous images of connected sets are con-
nected and the connected sets on the real line are intervals.) By the assumption the two
orientations are opposite, something can be said about the relationship of , , , . Sup-
pose without loss of generality that < . Then because of this assumption it follows
2
() =
1
() ,
2
() =
1
(). The following diagram might be useful to summarize
what was just said.
a
1
b
c
2
d
Note the rst of the interval [, d] matches the last of the interval [a, ] and the rst
of [, b] matches the last of [c, ] , all this in terms of where these points are sent.
Now I need to describe the parameterization of
1

2
. To verify it is a
simple closed curve, I must produce an interval and a mapping from this interval to
which satises the conditions needed for to be a simple closed rectiable curve. The
following is the denition as well as a description of which part of
j
is being obtained.
It is helpful to look at the above picture and the following picture in which there are
intervals placed next to each other. Above each is where the left end point starts o
followed by its length and nally where it ends up.
1
(a), a,
1
()
2
(), d ,
2
(d)
2
(c), c,
2
()
1
(), b ,
1
(b)
Note it ends up where it started, at
1
(a) =
1
(b). The following involved description
is nothing but the above picture with the edges of the little intervals computed along
with a description of which corresponds to the above picture.
Then (t) is given by
(t)
_
1
(t) , t [a, ] ,
1
(a)
1
() =
2
()
2
(t + ) , t [, +d ] ,
2
()
2
(d) =
2
(c)
2
(t +c d +) , t [ +d , +d + c] ,
2
(c) =
2
(d)
2
() =
1
()
1
(t d + +c +) , t [ +d + c, +d + c +b ] ,
1
()
1
(b) =
1
(a)
The construction shows is one to one on
(a, +d + c +b )
366 LINE INTEGRALS
and if t is in this open interval, then
(t) = (a) =
1
(a)
and
(t) = ( +d + c +b ) =
1
(b) .
Also
(a) =
1
(a) = ( +d + c +b ) =
1
(b)
so it is a simple closed curve. The claim about preserving the orientation is also obvious
from the formula. Note that t is never subtracted.
It only remains to prove the last claim. Suppose then that it is not so and l has
the same orientation as part of each
j
. Then from a repeat of the above argument,
you could change the orientation of l relative to
2
and obtain an orientation of
which is consistent with that of
1
and
2
. Call a parameterization which has this
new orientation
n
while is the one which is assumed to exist. This new orientation
of l changes the orientation of
2
because there are two points in l. Therefore on
1
2
(
2
),
1
n

2
is decreasing while
1
2
is assumed to be increasing. Hence
and
n
are not equivalent. However, the above construction would leave the orientation
of both
1
([a, ]) and
1
([, b]) unchanged and at least one of these must have at least
two points. Thus the orientation of must be the same for
n
as for . That is,
n
.
This is a contradiction. This proves the proposition.
There is a slightly dierent aspect of the above proposition which is interesting. It
involves using the shared segment to orient the simple closed curve .
Corollary 14.1.9 Let the intersection of simple closed rectiable curves,
1
and
2
consist of the simple curve l. Then place opposite orientations on l, and use these two
dierent orientations to specify orientations of
1
and
2
. Then letting denote the
simple closed curve which is obtained from deleting the open segment of l, there exists
an orientation for which is consistent with the orientations of
1
and
2
obtained
from the given specication of opposite orientations on l.
14.2 The Line Integral
Now I will return to considering the more general notion of bounded variation parame-
terizations without worrying about whether is one to one on the open interval. The
line integral and its properties are presented next.
Denition 14.2.1 Let : [a, b] R
n
be of bounded variation and let f :

R
n
. Letting P {t
0
, , t
n
} where a = t
0
< t
1
< < t
n
= b, dene
||P|| max {|t
j
t
j1
| : j = 1, , n}
and the Riemann Stieltjes sum by
S (P)
n
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
where
j
[t
j1
, t
j
] . (Note this notation is a little sloppy because it does not identify
the specic point,
j
used. It is understood that this point is arbitrary.) Dene
f d
as the unique number which satises the following condition. For all > 0 there exists
a > 0 such that if ||P|| , then
f d S (P)
< .
14.2. THE LINE INTEGRAL 367
Sometimes this is written as
f d lim
||P||0
S (P) .
Then
is a set of points in R
n
and as t moves from a to b, (t) moves from (a)
to (b) . Thus
has a rst point and a last point. (In the case of a closed curve these
are the same point.) If : [c, d] [a, b] is a continuous nondecreasing function, then
: [c, d] R
n
is also of bounded variation and yields the same set of points in R
n
with the same rst and last points.
Theorem 14.2.2 Let and be as just described. Then assuming that
f d
exists, so does
f d ( )
and
f d =
f d ( ) . (14.4)
Proof: There exists > 0 such that if P is a partition of [a, b] such that ||P|| < ,
then
f d S (P)
< .
By continuity of , there exists > 0 such that if Q is a partition of [c, d] with ||Q|| <
, Q = {s
0
, , s
n
} , then |(s
j
) (s
j1
)| < . Thus letting P denote the points in
[a, b] given by (s
j
) for s
j
Q, it follows that ||P|| < and so
f d
n
j=1
f ( ((
j
))) ( ((s
j
)) ((s
j1
)))
<
where
j
[s
j1
, s
j
] . Therefore, from the denition 14.4 holds and
f d ( )
exists. This proves the theorem.
This theorem shows that
f d is independent of the particular parameterization

used in its computation to the extent that if is any nondecreasing continuous function
from another interval, [c, d] , mapping to [a, b] , then the same value is obtained by
replacing with . In other words, this line integral depends only on
and the
order in which (t) encounters the points of
as t moves from one end to the other of

the interval. For the case of an oriented rectiable curve this shows the line integral
is dependent only on the set of points and the orientation of .
The fundamental result in this subject is the following theorem.
Theorem 14.2.3 Let f :
R
n
be continuous and let : [a, b] R
n
be
continuous and of bounded variation. Then
f d exists. Also letting

m
> 0 be such
that |t s| <
m
implies |f ( (t)) f ( (s))| <
1
m
,
f d S (P)
2V (, [a, b])
m
whenever ||P|| <
m
.
368 LINE INTEGRALS
Proof: The function, f , is uniformly continuous because it is dened on a
compact set. Therefore, there exists a decreasing sequence of positive numbers, {
m
}
such that if |s t| <
m
, then
|f ( (t)) f ( (s))| <
1
m
.
Let
F
m
{S (P) : ||P|| <
m
}.
Thus F
m
is a closed set. (The symbol, S (P) in the above denition, means to include
all sums corresponding to P for any choice of
j
.) It is shown that
diam(F
m
)
2V (, [a, b])
m
(14.5)
and then it will follow there exists a unique point, I
m=1
F
m
. This is because R is
complete. It will then follow I =
f (t) d (t) . To verify 14.5, it suces to verify that

whenever P and Q are partitions satisfying ||P|| <
m
and ||Q|| <
m
,
|S (P) S (Q)|
2
m
V (, [a, b]) . (14.6)
Suppose ||P|| <
m
and Q P. Then also ||Q|| <
m
. To begin with, suppose that
P {t
0
, , t
p
, , t
n
} and Q {t
0
, , t
p1
, t
, t
p
, , t
n
} . Thus Q contains only
one more point than P. Letting S (Q) and S (P) be Riemann Stieltjes sums,
S (Q)
p1
j=1
f ( (
j
)) ( (t
j
) (t
j1
)) +f ( (
)) ( (t
) (t
p1
))
+f ( (
)) ( (t
p
) (t
)) +
n
j=p+1
f ( (
j
)) ( (t
j
) (t
j1
)) ,
S (P)
p1
j=1
f ( (
j
)) ( (t
j
) (t
j1
)) +
=f ((p))((tp)(tp1))
..
f ( (
p
)) ( (t
) (t
p1
)) +f ( (
p
)) ( (t
p
) (t
))
+
n
j=p+1
f ( (
j
)) ( (t
j
) (t
j1
)) .
Therefore,
|S (P) S (Q)|
p1
j=1
1
m
| (t
j
) (t
j1
)| +
1
m
| (t
) (t
p1
)| +
1
m
| (t
p
) (t
)| +
n
j=p+1
1
m
| (t
j
) (t
j1
)|
1
m
V (, [a, b]) . (14.7)
Clearly the extreme inequalities would be valid in 14.7 if Q had more than one extra
point. You simply do the above trick more than one time. Let S (P) and S (Q) be
Riemann Stieltjes sums for which ||P|| and ||Q|| are less than
m
and let R P Q.
Then from what was just observed,
|S (P) S (Q)| |S (P) S (R)| +|S (R) S (Q)|
2
m
V (, [a, b]) .
and this shows 14.6 which proves 14.5. Therefore, there exists a unique number, I
m=1
F
m
which satises the denition of
f d. This proves the theorem.

Note this is a general sort of result. It is not assumed that is one to one anywhere
in the proof. The following theorem follows easily from the above denitions and
theorem. This theorem is used to establish estimates.
Theorem 14.2.4 Let f be a continuous function dened on
, denoted as f
C (
) where : [a, b] R
n
is of bounded variation and continuous. Let
M max {|f (t)| : t [a, b]} . (14.8)
Then
f d
MV (, [a, b]) . (14.9)

Also if {f
m
} is a sequence of functions of C (
) which is converging uniformly to the

function, f on
, then
lim
m
f
m
d =
f d. (14.10)
In case (a) = (b) so the curve is a closed curve and for f
k
the k
th
component of f ,
m
k
f
k
(x) M
k
for all x
, it also follows
f d
1
2
_
n
k=1
(M
k
m
k
)
2
_
1/2
V (, [a, b]) (14.11)
Proof: Let 14.8 hold. From the proof of Theorem 14.2.3, when ||P|| <
m
,
f d S (P)
2
m
V (, [a, b])
and so
f d
|S (P)| +
2
m
V (, [a, b])
Using the Cauchy Schwarz inequality and the above estimate in S (P) ,
j=1
M| (t
j
) (t
j1
)| +
2
m
V (, [a, b])
MV (, [a, b]) +
2
m
V (, [a, b]) .
This proves 14.9 since m is arbitrary.
To verify 14.10 use the above inequality to write
f d
f
m
d
(f f
m
) d (t)
max {|f (t) f

m
(t)| : t [a, b]} V (, [a, b]) .
Since the convergence is assumed to be uniform, this proves 14.10.
Claim: Let be closed bounded variation curve. Then if c is a constant vector,
cd = 0
370 LINE INTEGRALS
Proof of the claim: Let P {t
0
, , t
p
} be a partition with the property that
cd
p
k=1
c ( (t
k
) (t
k1
))
< .
Consider the sum. It is of the form
p
k=1
c (t
k
)
p
k=1
c (t
k1
) =
p
k=1
c (t
k
)
p1
k=0
c (t
k
)
= c (t
p
) c (t
0
) = 0
because it is given that since
is a closed curve, (t
0
) = (t
p
) . This shows the claim.
It only remains to verify 14.11. In this case (a) = (b) and so for each vector c
f d =
(f c) d
for any constant vector c. Let
c
k
=
1
2
(M
k
+m
k
)
Then for t [a, b]
|f ( (t)) c|
2
=
n
k=1
f
k
( (t))
1
2
(M
k
+m
k
)
k=1
_
1
2
(M
k
m
k
)
_
2
=
1
4
n
k=1
(M
k
m
k
)
2
Then with this choice of c, it follows from 14.9 that
f d
(f c) d
1
2
_
n
k=1
(M
k
m
k
)
2
_
1/2
V (, [a, b])
It turns out to be much easier to evaluate line integrals in the case where there exists
a parameterization which is in C
1
([a, b]) . The following theorem about approximation
will be very useful but rst here is an easy lemma.
Lemma 14.2.5 Let : [a, b] R
n
be in C
1
([a, b]) . Then V (, [a, b]) < so is
of bounded variation.
Proof: This follows from the following
n
j=1
| (t
j
) (t
j1
)| =
n
j=1
tj
tj1
(s) ds
j=1
tj
tj1
|
(s)| ds
j=1
tj
tj1
||
||
ds
= ||
||
(b a) .
where
||
||
max {|
(t)| : t [a, b]}

which exists because
is given to be continuous. Therefore it follows V (, [a, b])

||
||
(b a) . This proves the lemma.

The following is a useful theorem for reducing bounded variation curves to ones
which have a C
1
parameterization.
Theorem 14.2.6 Let : [a, b] R
n
be continuous and of bounded variation.
Let be an open set containing
and let f : R
n
be continuous, and let > 0
be given. Then there exists : [a, b] R
n
such that (a) = (a) , (b) = (b) ,
C
1
([a, b]) , and
|| || < , (14.12)
where || || max {| (t) (t)| : t [a, b]} . Also
f d
f d
< , (14.13)
V (, [a, b]) V (, [a, b]) , (14.14)
Proof: Extend to be dened on all R according to the rule (t) = (a) if t < a
and (t) = (b) if t > b. Now dene
h
(t)
1
2h
t+
2h
(ba)
(ta)
2h+t+
2h
(ba)
(ta)
(s) ds.
where the integral is dened in the obvious way, that is componentwise. Since is
continuous, this is certainly possible. Then
h
(b)
1
2h
b+2h
b
(s) ds =
1
2h
b+2h
b
(b) ds = (b) ,
h
(a)
1
2h
a
a2h
(s) ds =
1
2h
a
a2h
(a) ds = (a) .
Also, because of continuity of and the fundamental theorem of calculus,
h
(t) =
1
2h
_
_
t +
2h
b a
(t a)
__
1 +
2h
b a
_
_
2h +t +
2h
b a
(t a)
__
1 +
2h
b a
__
and so
h
C
1
([a, b]) . The following lemma is signicant.
Lemma 14.2.7 V (
h
, [a, b]) V (, [a, b]) .
Proof: Let a = t
0
< t
1
< < t
n
= b. Then using the denition of
h
and changing
the variables to make all integrals over [0, 2h] ,
n
j=1
|
h
(t
j
)
h
(t
j1
)| =
n
j=1
1
2h
2h
0
_
_
s 2h +t
j
+
2h
b a
(t
j
a)
_
372 LINE INTEGRALS
_
s 2h +t
j1
+
2h
b a
(t
j1
a)
__
1
2h
2h
0
n
j=1
_
s 2h +t
j
+
2h
b a
(t
j
a)
_
_
s 2h +t
j1
+
2h
b a
(t
j1
a)
_
ds.
For a given s [0, 2h] , the points, s 2h + t
j
+
2h
ba
(t
j
a) for j = 1, , n form an
increasing list of points in the interval [a 2h, b + 2h] and so the integrand is bounded
above by V (, [a 2h, b + 2h]) = V (, [a, b]) . It follows
n
j=1
|
h
(t
j
)
h
(t
j1
)| V (, [a, b])
which proves the lemma.
With this lemma the proof of the theorem can be completed without too much
trouble. Let H be an open set containing
such that H is a compact subset of . Let

0 < < dist
_
, H
C
_
. Then there exists
1
such that if h <
1
, then for all t,
| (t)
h
(t)|
1
2h
t+
2h
(ba)
(ta)
2h+t+
2h
(ba)
(ta)
| (s) (t)| ds
<
1
2h
t+
2h
(ba)
(ta)
2h+t+
2h
(ba)
(ta)
ds = (14.15)
due to the uniform continuity of . This proves 14.12.
Using the estimate from Theorem 14.2.3, 14.5, the uniform continuity of f on H,
and the above lemma, there exists such that if ||P|| < , then
f d (t) S (P)
<

3
,
h
f d
h
(t) S
h
(P)
<

3
for all h < 1. Here S (P) is a Riemann Stieltjes sum of the form
n
i=1
f ( (
i
)) ( (t
i
) (t
i1
))
and S
h
(P) is a similar Riemann Stieltjes sum taken with respect to
h
instead of .
Because of 14.15
h
(t) has values in H . Therefore, x the partition P, and choose
h small enough that in addition to this, the following inequality is valid.
|S (P) S
h
(P)| <

3
This is possible because of 14.15 and the uniform continuity of f on H. It follows
f d (t)
h
f d
h
(t)
f d (t) S (P)
+|S (P) S
h
(P)|
+
S
h
(P)
h
f d
h
(t)
< .
Let
h
. Formula 14.14 follows from the lemma. This proves the theorem.
This is a very useful theorem because if is C
1
([a, b]) , it is easy to calculate
f d
and the above theorem allows a reduction to the case where is C
1
. The next theorem
shows how easy it is to compute these integrals in the case where is C
1
. First note
that if f is continuous and C
1
([a, b]) , then by Lemma 14.2.5 and the fundamental
existence theorem, Theorem 14.2.3,
f d exists.
Theorem 14.2.8 If f :
X is continuous and : [a, b] R

n
is in
C
1
([a, b]) and is a parameterization, then
f d =
b
a
f ( (t))
(t) dt. (14.16)

Proof: Let P be a partition of [a, b], P = {t
0
, , t
n
} and ||P|| is small enough
that whenever |t s| < ||P|| ,
|f ( (t)) f ( (s))| < (14.17)
and

f d
n
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
< .
Now
n
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
=
b
a
n
j=1
f ( (
j
)) X
[tj1,tj]
(s)
(s) ds
where here
X
[p,q]
(s)
_
1 if s [p, q]
0 if s / [p, q]
.
Also,
b
a
f ( (s))
(s) ds =
b
a
n
j=1
f ( (s)) X
[tj1,tj]
(s)
(s) ds
and thanks to 14.17,
n
j=1
f ((j))((tj)(tj1))
..
b
a
n
j=1
f ( (
j
)) X
[tj1,tj]
(s)
(s) ds
b
a
f ((s))
(s)ds
..
b
a
n
j=1
f ( (s)) X
[tj1,tj]
(s)
(s) ds
374 LINE INTEGRALS
j=1
tj
tj1
|f ( (
j
)) f ( (s))| |
(s)| ds
||
||
j
(t
j
t
j1
)
= ||
||
(b a) .
It follows that
f d
b
a
f ( (s))
(s) ds
f d
n
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
b
a
f ( (s))
(s) ds
||
||
(b a) +.
Since is arbitrary, this veries 14.16.
You can piece bounded variation curves together to get another bounded variation
curve. You can also take the integral in the opposite direction along a given curve.
There is also something called a potential.
Denition 14.2.9 A function f : R
n
for an open set in R
n
has a
potential if there exists a function, F, the potential, such that F = f . Also if
k
:
[a
k
, b
k
] R
n
is continuous and of bounded variation, for k = 1, , m and
k
(b
k
) =
k+1
(a
k
) , dene
m
k=1
k
f d
k

m
k=1
k
f d
k
. (14.18)
In addition to this, for : [a, b] R
n
, dene : [a, b] R
n
by (t) (b +a t) .
Thus simply traces out the points of
in the opposite order.

The following lemma is useful and follows quickly from Theorem 14.2.2.
Lemma 14.2.10 In the above denition, there exists a continuous bounded variation
function, dened on some closed interval, [c, d] , such that ([c, d]) =
m
k=1
k
([a
k
, b
k
])
and (c) =
1
(a
1
) while (d) =
m
(b
m
) . Furthermore,
f d =
m
k=1
k
f d
k
.
If : [a, b] R
n
is of bounded variation and continuous, then
f d =
f d.
The following theorem shows that it is very easy to compute a line integral when
the function has a potential.
Theorem 14.2.11 Let : [a, b] R
n
be continuous and of bounded variation.
Also suppose F = f on , an open set containing
and f is continuous on . Then
f d = F ( (b)) F ( (a)) .
Proof: By Theorem 14.2.6 there exists C
1
([a, b]) such that (a) = (a) , and
(b) = (b) such that
f d
f d
< .
Then from Theorem 14.2.8, since is in C
1
([a, b]) , it follows from the chain rule and
the fundamental theorem of calculus that
f d =
b
a
f ( (t))
(t) dt =
b
a
d
dt
F ( (t)) dt
= F ( (b)) F ( (a)) = F ( (b)) F ( (a)) .
Therefore,
(F ( (b)) F ( (a)))
f d
<
and since > 0 is arbitrary, This proves the theorem.
Corollary 14.2.12 If : [a, b] R
n
is continuous, has bounded variation, is a
closed curve, (a) = (b) , and
where is an open set on which F = f , then
f d = 0.
Theorem 14.2.13 Let be a connected open set and let f : R
n
be con-
tinuous. Then f has a potential F if and only if
f d
is path independent for all a bounded variation curve such that
is contained in .
This means the above line integral depends only on (a) and (b).
Proof: The rst part was proved in Theorem 14.2.11. It remains to verify the
existence of a potential in the situation of path independence.
Let x
0
be xed. Let S be the points x of which have the property there is
a bounded variation curve joining x
0
to x. Let
x0x
denote such a curve. Note rst
that S is nonempty. To see this, B(x
0
, r) for r small enough. Every x B(x
0
, r)
is in S. Then S is open because if x S, then B(x, r) for small enough r and if
y B(x, r) , you could go take
x0x
and from x follow the straight line segment joining
x to y. In addition to this, \ S must also be open because if x \ S, then choosing
B(x, r) , no point of B(x, r) can be in S because then you could take the straight
line segment from that point to x and conclude that x S after all. Therefore, since
is connected, it follows \ S = . Thus for every x S, there exists
x0x
, a bounded
variation curve from x
0
to x.
Dene
F (x)
x
0
x
f d
x0x
376 LINE INTEGRALS
F is well dened by assumption. Now let l
x(x+te
k
)
denote the linear segment from x to
x + te
k
. Thus to get to x + te
k
you could rst follow
x0x
to x and from there follow
l
x(x+te
k
)
to x +te
k
. Hence
F (x+te
k
) F (x)
t
=
1
t
l
x(x+te
k
)
f dl
x(x+te
k
)
=
1
t
t
0
f (x +se
k
) e
k
ds f
k
(x)
by continuity of f . Thus F = f and This proves the theorem.
Corollary 14.2.14 Let be a connected open set and f : R
n
. Then f has a
potential if and only if every closed, (a) = (b) , bounded variation curve contained
in has the property that
f d = 0
Proof: Using Lemma 14.2.10, this condition about closed curves is equivalent to
the condition that the line integrals of the above theorem are path independent. This
proves the corollary.
Such a vector valued function is called conservative.
14.3 Simple Closed Rectiable Curves
There are examples of space lling continuous curves. However, bounded variation
curves are not like this. In fact, one can even say the two dimensional Lebesgue measure
of a bounded variation curve is 0.
Theorem 14.3.1 Let : [a, b]
R
n
where n 2 is a continuous bounded
variation curve. Then
m
n
(
) = 0
where m
n
denotes n dimensional Lebesgue measure.
Proof: Let > 0 be given. Let t
0
a and if t
0
, , t
k
have been chosen, let t
k+1
be the rst number larger than t
k
such that
| (t
k+1
) (t
k
)| = .
If the set of t such that | (t) (t
k
)| = is nonempty, then this set is clearly closed
and so such a t
k+1
exists until k is such that

k
j=0
B( (t
j
) , )
Let m be the last index of this process where t
m+1
does not exist. How large is m? This
can be estimated because
V (, [a, b])
m
k=0
| (t
k+1
) (t
k
)| = m
and so m V (, [a, b]) /. Since

m
j=0
B( (t
j
) , ) ,
m
n
(
)
m
j=0
m
n
(B( (t
j
) , ))
V (, [a, b])
c
n
n
= c
n
V (, [a, b])
n1
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 377
Since was arbitrary, This proves the theorem.
Since a ball has positive measure, this proves the following corollary.
Corollary 14.3.2 Let : [a, b]
R
n
where n 2 is a continuous bounded
variation curve. Then
has empty interior.

Lemma 14.3.3 Let be a simple closed curve. Then there exists a mapping :
S
1
where S
1
is the unit circle
_
(x, y) : x
2
+y
2
= 1
_
,
such that is one to one and continuous.
Proof: Since is a simple closed curve, there is a parameterization and an interval
[a, b] such that is continuous and one to one on [a, b) and (a, b] with (a) = (b) .
Dene
1
: S
1
by
1
(x)
_
cos
_
2
b a
_
1
(x) a
_
_
, sin
_
2
b a
_
1
(x) a
_
__
Note that
1
is onto S
1
. The function is well dened because it sends the point (a) =
(b) to the same point, (1, 0) . It is also one to one. To see this note
1
is one
to one on \ { (a) , (b)} . What about the case where x = (a) = (b)? Could
1
(x) =
1
( (a))? In this case,
1
(x) is in (a, b) while
1
( (a)) = a so
1
(x) =
1
( (a)) = (1, 0) .
Thus
1
is one to one on .
Why is
1
continuous? Suppose x
n
(a) = (b) rst. Why does
1
(x
n
)
(1, 0) =
1
( (a))? Let {x
n
} denote any subsequence of the given sequence. Then by
compactness of [a, b] there exists a further subsequence, still denoted by x
n
such that
1
(x
n
) t [a, b]
Hence by continuity of , x
n
(t) and so (t) must equal (a) = (b) . It follows
from the assumption of what a simple curve is that t {a, b} . Hence
1
(x
n
) converges
to either
_
cos
_
2
b a
(a a)
_
, sin
_
2
b a
(a a)
__
or
_
cos
_
2
b a
(b a)
_
, sin
_
2
b a
(b a)
__
but these are the same point. This has shown that if x
n
(a) = (b) , there is a
subsequence such that
1
(x
n
)
1
( (a)) . Thus
1
is continuous at (a) = (b).
Next suppose x
n
x = (a) p. Then there exists B(p, r) such that for all n
large enough, x
n
and x are contained in the compact set \ B(p, r) K. Then is
continuous and one to one on the compact set
1
(K) (a, b) and so by Theorem 5.1.3
1
is continuous on K. In particular it is continuous at x so
1
(x
n
)
1
(x). This
proves the lemma.
14.3.1 The Jordan Curve Theorem
The following theorem includes the Jordan Curve theorem, a major result for simple
closed curves in the plane. In this theorem and in what follows U
i
will denote the inside
of a simple closed curve and U
o
will denote the outside. This theorem is proved from
the Jordan separation theorem. If you have read the elementary chapter on the Jordan
curve theorem you can skip this. See Theorem 13.0.28.
378 LINE INTEGRALS
Theorem 14.3.4 Let C denote the unit circle,
_
(x, y) R
2
: x
2
+y
2
= 1
_
. Sup-
pose : C R
2
is one to one onto and continuous. Then R
2
\ consists of two
components, a bounded component (called the inside) U
i
and an unbounded component
(called the outside), U
o
. Also the boundary of each of these two components of R
2
\
is and has empty interior.
Proof: That R
2
\ consists of two components, U
o
and U
i
follows from the Jordan
separation theorem. There is exactly one unbounded component because is bounded
and so U
i
is dened as the bounded component. It remains to verify the assertion about
being the boundary. Let x be a limit point of U
i
. Then it cant be in U
o
because these
are both open sets. Therefore, all the limit points of U
i
are in U
i
. Similarly all the
limit points of U
o
are in U
o
. Thus U
i
and U
o
.
I claim has empty interior. This follows because by Theorem 5.1.3 on Page 88,
and
1
must both be continuous since C is compact. Thus if B is an open ball
contained in , it follows from invariance of domain that
1
(B) is an open set in
R
2
. But this needs to be contained in C which is a contradiction because C has empty
interior obviously.
Now let x R
n
such that x / U
o
U
i
. Then x and must be a limit point of
either U
o
or U
i
since if this were not so, would be forced to have nonempty interior.
Hence U
1
U
2
= R
2
. Next I will show U
i
= U
o
. Suppose then that
p U
i
\ U
o
Then p / U
o
because p U
i
which is disjoint from U
o
. Thus p is not in U
o
because it is given to not be in U
o
. Hence there is a ball centered at p, B(p, r) which
contains no points of U
o
. Thus B(p, r) U
i
and so p is an interior point of U
i
which
implies p is actually in U
i
, a contradiction to p U
i
. Thus U
i
\U
o
= and a similar
argument shows U
o
\ U
i
= . Thus U
i
= U
o
and so if x , then it is in either
U
i
or U
o
and these are equal. Thus
U
i
= U
o
U
i
U
o
= U
i
= U
o
.
The following lemma will be of importance in what follows. To say two sets are
homeomorphic is to say there is a one to one continuous onto mapping from one to
the other which also has continuous inverse. Clearly the statement that two sets are
homeomorphic denes an equivalence relation. Then from Lemma 13.0.26 a Jordan
curve is just a curve which is homeomorphic to a circle.
Lemma 14.3.5 In the situation of Theorem 14.3.4 or Theorem 13.0.28, let be a
simple closed curve and let
be a straight line segment such that the open segment,
o
,
without its endpoints, is contained in U

i
such that the intersection of
with
equals {p, q} . Then this line segment divides U
i
into two connected open sets U
1i
, U
2i
which are the insides of two simple closed curves such that
U
i
= U
1i

o
U
2i
Proof: Denote by C the unit circle and let : C be continuous one to one and
onto. Say (a) = p and (b) = q. Let C
j
, j = 1, 2 denote the two circular arcs joining
a and b. Thus letting
j
(C
j
) it follows
1
,
2
are simple curves whose union is
which intersect at the points p and q. Letting
j

J
j
it follows J
j
is a simple
closed curve. Here is why. Dene
h
1
(x)
_
(x) if x C
1
f
2
(x) if x C
2
where f
j
is a continuous one to one onto mapping from C
j
to
. Then h
1
is continuous
and maps C one to one and onto J
1
. Dene h
2
similarly. Denote by U
ji
the inside and
U
jo
the outside of J
j
.
p
q
1
U
2i
U
1i
Claim 1: U
1i
, U
2i
U
i
, U
1i
, U
2i
contain no points of J
2
J
1
.
Proof: First consider the claim that U
1i
, U
2i
contain no points of J
2
J
1
. If
x
2
\ {p, q} , then near x there are points of U
i
and U
o
. Therefore, if x U
1i
,
there would be points of both U
i
and U
o
in U
1i
. Now U
o
contains no points of
by
assumption and it contains no points of by denition. Therefore, since U
o
is connected,
it must be contained in U
1i
but this is impossible because U
o
is unbounded and U
1i
is
bounded. Thus x / U
1i
. Similarly x / U
2i
. Similar reasoning applied to x
1
\{p, q}
implies no point of \ {p, q} can be in either U
1i
or U
2i
. If x
then by denition it
is not in either U
1i
or U
2i
and so neither U
1i
not U
2i
contain any points of J
2
J
1
. If
U
1i
U
o
= , then the whole connected set U
1i
is contained in U
o
since otherwise U
1i
would contain points of which, by what was just shown, is not the case. But now
this is a contradiction to the open segment of
being contained in U
i
. A point x on
this open segment must have points of U
1i
near it by the Jordan curve theorem but
if U
1i
U
o
, then this cannot happen because a small ball centered at x contains only
points of U
i
and none of U
o
.
Similarly U
2i
U
i
. Letting
o
denote the open segment of
, it follows
U
i
U
1i

o
U
2i
(14.19)
Claim 2:U
1i
U
2i
= , U
2i
U
1o
, U
1i
U
2o
.
Proof: If x is a point of U
1i
which is in U
2i
then U
1i
must be contained in U
2i
.
This is because U
1i
is connected and, as noted above, U
1i
contains no boundary points
of U
2i
because these points are all contained in J
2
J
2
. Similarly U
2i
would need to be
contained in U
1i
and so these two interior components would need to coincide. But now
consider x
2
\ {p, q} . Let r be small enough that B(x, r) J
1
= . Then this ball
contains points of U
2i
= U
1i
and points of U
2o
. However, this ball is contained in one
of the complementary components of J
1
and since it contains points of U
1i
, this forces
B(x, r) to be contained in U
1i
. But now the same contradiction as above holds namely:
There exists a point of U
o
in U
1i
which forces U
o
to be contained in U
1i
. Therefore,
U
1i
U
2i
= . Also U
2i
U
1o
because if not, there would be points of U
1i
in U
2i
and it
was just shown this doesnt happen. Similarly U
1i
U
2o
. This shows Claim 2.
Next I need to verify that equality holds in 14.19. First I will argue
U
1i

o
U
2i
is an open set. Let y
o
and let B(y, r) = .
380 LINE INTEGRALS
p
q
1
U
2i
U
1i
y
Thus
o
divides B(y, r) into two halves, H
1
, H
2
. The ball contains points of U
2i
by the Jordan curve theorem. Say H
2
contains some of these points. Then I claim H
2
cannot contain any points of U
1i
. This is because if it did, there would be a segment
joining a point of U
1i
with a point of U
2i
which is contained in H
2
which is a connected
open set which is therefore contained in a single component of J
C
1
. This is a contradiction
because as shown above, U
2i
U
1o
. Could H
2
contain any points of U
2o
? No because
then there would be a segment joining a point of U
2o
to a point of U
2i
which is contained
in the same component of J
C
2
. Therefore, H
2
consists entirely of points of U
2i
. Similarly
H
1
consists entirely of points of U
1i
. Therefore, U
1i

o
U
2i
is an open set because
the only points which could possibly fail to be interior points, those on
0
are interior
points of U
1i

o
U
2i
.
Suppose equality does not hold in 14.19. Then there exists w U
i
\(U
1i

o
U
2i
) .
Let x U
1i

o
U
2i
. Then since U
i
is connected and open, there exists a continuous
mapping r : [0, 1] U
i
such that r (0) = x and r (1) = w. Since U
1i

o
U
2i
is open,
there exists a rst point in the closed set r
1
_
(U
1i

o
U
2i
)
C
_
, s. Thus r (s) is a
limit point of U
1i

o
U
2i
but is not in this set which implies it is in U
1o
U
2o
. It
follows r (s) is a limit point of either U
1i
or U
2i
because each point of
o
is a limit point
of U
1i
and U
2i
. Also, r (s) cannot be in
0
because it is not in U
1i

o
U
2i
. Suppose
without loss of generality it is a limit point of U
1i
. Then every ball containing r (s)
must contain points of U
1o
U
2o
U
1o
as well as points U
1i
. But by the Jordan curve
theorem, this implies r (s) is in J
1
but is not in
o
. Therefore, r (s) is a point of and
this contradicts r (s) U
i
. Therefore, equality must hold in 14.19 after all. This proves
the lemma.
The following lemma has to do with decomposing the inside and boundary of a simple
closed rectiable curve into small pieces. The argument is like one given in Apostol
[3]. In doing this I will refer to a region as the union of a connected open set with its
boundary. Also, two regions will be said to be non overlapping if they either have empty
intersection or the intersection is contained in the intersection of their boundaries.The
height of a set A equals sup {|y
1
y
2
| : (x
1
, y
1
) , (x
2
, y
2
) A} . The width of A will be
dened similarly.
Lemma 14.3.6 Let be a simple closed rectiable curve. Also let > 0 be given
such that 2 is smaller than both the height and width of . Then there exist nitely
many non overlapping regions {R
k
}
n
k=1
consisting of simple closed rectiable curves
along with their interiors whose union equals U
i
. These regions consist of two kinds,
those contained in U
i
and those with nonempty intersection with . These latter regions
are called border regions. The boundary of a border region consists of straight line
segments parallel to the coordinate axes of the form x = m or y = k for m, k integers
along with arcs from . The regions contained in U
i
consist of rectangles. Thus all
of these regions have boundaries which are rectiable simple closed curves. Also each
region is contained in a square having sides of length no more than 2. There are at
most
4
_
V ()
+ 1
_
border regions. The construction also yields an orientation for and for all these
regions, and the orientations for any segment shared by two regions are opposite.
Proof: Let = ([a, b]) where = (
1
,
2
) . Let
y
1
max {
2
(t) : t [a, b]}
and let
y
2
min {
2
(t) : t [a, b]} .
Thus (x
1
, y
1
) , x
1

1
_
1
2
(y
1
)
_
is the top point of while (x
2
, y
2
) is the bottom
point of . Consider the lines y = y
1
and y = y
2
. By assumption |y
1
y
2
| > 2.
Consider the line l given by y = m where m is chosen to make m as close as possible
to (y
1
+y
2
) /2. Thus y
1
> m > y
2
. By Theorem 14.3.4 or 13.0.28(x
j
, y
j
) j = 1, 2,
being on are both limit points of U
i
so there exist points p
j
U
i
such that p
1
is
above l and p
2
is below l. (Simply pick p
j
very close to (x
j
, y
j
) and yet in U
i
and this
will take place.) The horizontal line l must have nonempty intersection with U
i
because
U
i
is connected. If it had empty intersection it would be possible to separate U
i
into
two nonempty open sets, one containing p
1
and the other containing p
2
.
Let q be a point of U
i
which is also in l. Then there exists a maximal segment
of the line l containing q which is contained in U
i
. This segment,
satises the
conditions of Lemma 14.3.5 and so it divides U
i
into disjoint open connected sets whose
boundaries are simple rectiable closed curves. Note the line segment has nite length.
Letting
j
be the simple closed curve which contains p
j
, orient
as part of
2
such
that motion is from right to left. As part of
1
the motion along
is from left to right.

By Proposition 14.1.7 this provides an orientation to each
j
. By Proposition 14.1.8
there exists an orientation for which is consistent with these two orientations on the
j
.
Now do the same process to the two simple closed curves just obtained and continue
till all regions have height less than 2. Each application of the process yields two new
non overlapping regions of the desired sort in place of an earlier region of the desired
sort except possibly the regions might have excessive height. The orientation of a new
line segment in the construction is determined from the orientations of the simple closed
curves obtained earlier. By Proposition 14.1.7 the orientations of the segments shared
by two regions are opposite so eventually the line integrals over these segments cancel.
Eventually this process ends because all regions have height less than 2. The reason for
this is that if it did not end, the curve could not have nite total variation because
there would exist an arbitrarily large number of non overlapping regions each of which
have a pair of points which are farther apart than 2. This takes care of nding the
subregions so far as height is concerned.
Now follow the same process just described on each of the non overlapping short
regions just obtained using vertical rather than horizontal lines, letting the orientation
of the vertical edges be determined from the orientation already obtained, but this time
feature width instead of height and let the lines be vertical of the form x = k where k
is an integer.
How many border regions are there? Denote by V () the length of . Now de-
compose into N arcs of length with maybe one having length less than . Thus
N 1
V ()
and so
N
V ()
+ 1
382 LINE INTEGRALS
The resulting regions are each contained in a box having sides of length no more than
2 in length. Each of these N arcs cant intersect any more than four of these boxes
because of their short length. Therefore, at most 4N boxes of the construction can
intersect . Thus there are no more than
4
_
V ()
+ 1
_
border regions. This proves the lemma.
Note that this shows that the region of U
i
which is included by the boundary boxes
has area at most equal to
4
_
V ()
+ 1
_
_
16
2
_
which converges to 0 as 0.
14.3.2 Orientation And Greens Formula
How do you describe the orientation of a simple closed rectiable curve analytically? The
above process did it but I want another way to identify this which is more geometrically
appealing. For simple examples, this is not too hard but it becomes less obvious when
you consider the general case. The problem is the simple closed curve could be very
wriggly.
The orientation of a rectiable simple closed curve will be dened in terms of a very
important formula known as Greens formula. First I will present Greens formula for a
rectangle. In this lemma, it is very easy to understand the orientation of the bounding
curve. The direction of motion is counter clockwise. As described in Proposition 14.1.7
it suces to describe a direction of motion along the curve using any two points.
Lemma 14.3.7 Let R = [a, b] [c, d] be a rectangle and let P, Q be functions which
are C
1
in some open set containing R. Orient the boundary of R as shown in the fol-
lowing picture. This is called the counter clockwise direction or the positive orientation
Then letting denote the oriented boundary of R as shown,
R
(Q
x
(x, y) P
y
(x, y)) dm
2
=
f d
where
f (x, y) (P (x, y) , Q(x, y)) .
In this context the line integral is usually written using the notation
R
Pdx +Qdy.
Proof: This follows from direct computation. A parameterization for the bottom
line of R is
B
(t) = (a +t (b a) , c) , t [0, 1]
A parameterization for the top line of R with the given orientation is
T
(t) = (b +t (a b) , d) , t [0, 1]
A parameterization for the line on the right side is
R
(t) = (b, c +t (d c)) , t [0, 1]
and a parameterization for the line on the left side is
L
(t) = (a, d +t (c d)) , t [0, 1]
Now it is time to do the computations using Theorem 14.2.8.
f d =
1
0
P (a +t (b a) , c) (b a) dt
+
1
0
P (b +t (a b) , d) (a b) dt
+
1
0
Q(b, c +t (d c)) (d c) dt +
1
0
Q(a, d +t (c d)) (c d) dt
Changing the variables and combining the integrals, this equals
=
b
a
P (x, c) dx
b
a
P (x, d) dx +
d
c
Q(b, y) dy
d
c
Q(a, y) dy
=
b
a
d
c
P
y
(x, y) dydx +
d
c
b
a
Q
x
(x, y) dxdy
=
R
(Q
x
P
y
) dm
2
By Fubinis theorem, Theorem 9.2.3 on Page 212. (To use this theorem you can extend
the functions to equal 0 o R.) This proves the lemma.
Note that if the rectangle were oriented in the opposite way, you would get
f d =
R
(P
y
Q
x
) dm
2
With this lemma, it is possible to prove Greens theorem and also give an ana-
lytic criterion which will distinguish between dierent orientations of a simple closed
rectiable curve. First here is a discussion which amounts to a computation.
Let be a rectiable simple closed curve with inside U
i
and outside U
o
. Let {R
k
}
n
k=1
denote the non overlapping regions of Lemma 14.3.6 all oriented as explained there and
let also be oriented as explained there. It could be shown that all the regions contained
in U
i
have positive orientation but this will not be fussed over here. What can be said
with no fussing is that since the shared edges have opposite orientations, all these interior
regions are either oriented positively or they are all oriented negatively.
Let B
be the set of border regions and let I
be the rectangles contained in U

i
. Thus
in taking the sum of the line integrals over the boundaries of the interior rectangles, the
384 LINE INTEGRALS
integrals over the interior edges cancel out and you are left with a line integral over
the exterior edges of a polygon which is composed of the union of the squares in I
.
Now let f (x, y) = (P (x, y) , Q(x, y)) be a vector eld which is C
1
on U
i
, and suppose
also that both P
y
and Q
x
are in L
1
(U
i
) (Absolutely integrable) and that P, Q are
continuous on U
i
. (An easy way to get all this to happen is to let P, Q be restrictions
to U
i
of functions which are C
1
on some open set containing U
i
.) Note that
>0
{R : R I
} = U
i
and that for
I
{R : R I
} ,
the following pointwise convergence holds.
lim
0
X
I
(x) = X
Ui
(x) .
By the dominated convergence theorem,
lim
0
(Q
x
P
y
) dm
2
=
Ui
(Q
x
P
y
) dm
2
lim
0
(P
y
Q
x
) dm
2
=
Ui
(P
y
Q
x
) dm
2
Let R denote the boundary of R for R one of these regions of Lemma 14.3.6 oriented
as described. Let w
(R)
2
denote
(max {Q(x) : x R} min {Q(x) : x R})
2
+(max {P (x) : x R} min {P (x) : x R})
2
By uniform continuity of P, Q on the compact set U
i
, if is small enough, w
(R) <
for all R B
. Then for R B
, it follows from Theorem 14.2.4
R
f d
1
2
w
(R) (V (R)) < (V (R)) (14.20)

whenever is small enough. Always let be this small.
Also since the line integrals cancel on shared edges
RI
R
f d +
RB
R
f d =
f d (14.21)
Consider the second sum on the left. From 14.20
RB
R
f d
RB
R
f d
RB
(V (R))
Denote by
R
the part of which is contained in R B
and V (
R
) is its length. Then
the above sum equals
_

RB
V (
R
) +B
_
= (V () +B
)
where B
is the sum of the lengths of the straight edges. This is easy to estimate. Recall
from 14.3.6 there are no more than
4
_
V ()
+ 1
_
of these border regions. Furthermore, the sum of the lengths of all four edges of one of
these is no more than 8 and so
B
4
_
V ()
+ 1
_
8 = 32V () + 32.
Thus the absolute value of the second sum on the right in 14.21 is dominated by
(33V () + 32)
Since was arbitrary, this formula implies with Greens theorem proved above for
squares,
f d = lim
0
RI
R
f d + lim
0
RB
R
f d
= lim
0
RI
R
f d = lim
0
(Q
x
P
y
) dm
2
=
Ui
(Q
x
P
y
) dm
2
where the adusts for whether the interior rectangles are all oriented positively or all
oriented negatively.
This has proved the general form of Greens theorem which is stated in the following
theorem.
Theorem 14.3.8 Let be a rectiable simple closed curve in R
2
having inside
U
i
and outside U
o
. Let P, Q be functions with the property that
Q
x
, P
y
L
1
(U
i
)
and P, Q are C
1
on U
i
. Assume also P, Q are continuous on U
i
. Then there exists
an orientation for (Remember there are only two.) such that for
f (x, y) = (P (x, y) , Q(x, y)) ,
f d =
Ui
(Q
x
P
y
) dm
2
.
Proof: In the construction of the regions, an orientation was imparted to . The
above computation shows
f d =
Ui
(Q
x
P
y
) dm
2
If the area integral equals
Ui
(Q
x
P
y
) dm
2
,
just take the other orientation for . This proves the theorem.
With this wonderful theorem, it is possible to give an analytic description of the two
dierent orientations of a rectiable simple closed curve. The positive orientation is the
one for which Greens theorem holds and the other one, called the negative orientation
is the one for which
f d =
Ui
(P
y
Q
x
) dm
2
.
There are other regions for which Greens theorem holds besides just the inside and
boundary of a simple closed curve. For a simple closed curve and U
i
its inside, lets
refer to U
i
as a Jordan region. When you have two non overlapping Jordan regions
which intersect in a nite number of simple curves, you can delete the interiors of these
386 LINE INTEGRALS
simple curves and what results will also be a region for which Greens theorem holds.
This is illustrated in the following picture.
U
1i
U
i2
There are two Jordan regions here with insides U
1i
and U
2i
and these regions intersect
in three simple curves. As indicated in the picture, opposite orientations are given to
each of these three simple curves. Then the line integrals over these cancel. The area
integrals add. Recall the two dimensional area of a bounded variation curve equals 0.
Denote by the curve on the outside of the whole thing and
1
and
2
the oriented
boundaries of the two holes which result when the curves of intersection are removed,
the orientations as shown. Then letting f (x, y) = (P (x, y) , Q(x, y)) , and
U = U
1i
U
2i
{Open segments of intersection}
as shown in the following picture,
2
U
it follows from applying Greens theorem to both of the Jordan regions,
f d+
1
f d
1
+
2
f d
2
=
U1iU2i
(Q
x
P
y
) dm
2
=
U
(Q
x
P
y
) dm
2
To make this simpler, just write it in the form
U
f d =
U
(Q
x
P
y
) dm
2
where U is oriented as indicated in the picture and involves the three oriented curves
,
1
,
2
.
14.4 Stokes Theorem
Stokes theorem is usually presented in calculus courses under far more restrictive as-
sumptions than will be used here. It turns out that all the hard questions are related
to Greens theorem and that when you have the general version of Greens theorem
this can be used to obtain a general version of Stokes theorem using a simple identity.
14.4. STOKES THEOREM 387
This is because Stokes theorem is really just a three dimensional version of the two
dimensional Greens theorem. This will be made more precise below.
To begin with suppose is a rectiable curve in R
2
having parameterization :
[a, b] for a continuous function. Let R : U R
n
be a C
1
function where U
contains
. Then one could dene a curve

(t) R((t)) , t [a, b] .
Lemma 14.4.1 The curve
where is as just described is a rectiable curve. If

F is dened and continuous on
then
F d =
((F R) R
u
, (F R) R
v
) d
where R
u
signies the partial derivative of R with respect to the variable u.
Proof: Let
K
_
y R
2
: dist (y,
) r
_
where r is small enough that K U. This is easily done because
is compact. Let
C
K
max {||DR(x)|| : x K}
Consider
n1
j=0
|R((t
j+1
)) R((t
j
))| (14.22)
where {t
0
, , t
n
} is a partition of [a, b] . Since is continuous, there exists a such
that if ||P|| < , then the segment
{(t
j
) +s ((t
j+1
) (t
j
)) : s [0, 1]}
is contained in K. Therefore, by the mean value inequality, Theorem 6.4.2,
n1
j=0
|R((t
j+1
)) R((t
j
))|
n1
j=0
C
K
|(t
j+1
) (t
j
)|
Now if P is any partition, 14.22 can always be made larger by adding in points to P till
||P|| < and so this shows
V (, [a, b]) C
K
V (, [a, b]) .
This proves the rst part.
Next consider the claim about the integral. Let
G(v, x) R(x +v) R(x) DR(x) (v) .
Then
D
1
G(v, x) = DR(x +v) DR(x)
and so by uniform continuity of DR on the compact set K, it follows there exists > 0
such that if |v| < , then for all x
,
||DR(x +v) DR(x)|| = ||D
1
G(v, x)|| < .
By Theorem 6.4.2 again it follows that for all x
and |v| < ,

|G(v, x)| = |R(x +v) R(x) DR(x) (v)| |v| (14.23)
388 LINE INTEGRALS
Letting ||P|| be small enough, it follows from the continuity of that
|(t
j+1
) (t
j
)| <
Therefore for such P,
n1
j=0
F( (t
j
)) ( (t
j+1
) (t
j
))
=
n1
j=0
F(R((t
j
))) (R((t
j+1
)) R((t
j
)))
=
n1
j=0
F(R((t
j
))) [DR((t
j
)) ((t
j+1
) (t
j
)) +o((t
j+1
) (t
j
))]
where
o((t
j+1
) (t
j
)) = R((t
j+1
)) R((t
j
)) DR((t
j
)) ((t
j+1
) (t
j
))
and by 14.23,
|o((t
j+1
) (t
j
))| < |(t
j+1
) (t
j
)|
It follows
n1
j=0
F( (t
j
)) ( (t
j+1
) (t
j
))
n1
j=0
F(R((t
j
))) DR((t
j
)) ((t
j+1
) (t
j
))
(14.24)
n1
j=0
|o((t
j+1
) (t
j
))|
n1
j=0
|(t
j+1
) (t
j
)| V (, [a, b])
Consider the second sum in 14.24. A term in the sum equals
F(R((t
j
))) (R
u
((t
j
)) (
1
(t
j+1
)
1
(t
j
)) +R
v
((t
j
)) (
2
(t
j+1
)
2
(t
j
)))
= (F(R((t
j
))) R
u
((t
j
)) , F(R((t
j
))) R
v
((t
j
))) ((t
j+1
) (t
j
))
By continuity of F, R
u
and R
v
, it follows that sum converges as ||P|| 0 to
((F R) R
u
, (F R) R
v
) d
Therefore, taking the limit as ||P|| 0 in 14.24
F d
((F R) R
u
, (F R) R
v
) d
< V (, [a, b]) .

Since > 0 is arbitrary, This proves the lemma.
The following is a little identity which will allow a proof of Stokes theorem to follow
from Greens theorem. First recall the following denition from calculus of the curl of
a vector eld and the cross product of two vectors from calculus.
Denition 14.4.2 Let u (a, b, c) and v (d, e, f) be two vectors in R
3
. Then
u v
i j k
a b c
d e f
14.4. STOKES THEOREM 389

where the determinant is expanded formally along the top row. Let f : U R
3
for
U R
3
denote a vector eld. The curl of the vector eld yields another vector eld
and it is dened as follows.
(curl (f ) (x))
i
(f (x))
i
where here
j
means the partial derivative with respect to x
j
and the subscript of i in
(curl (f ) (x))
i
means the i
th
Cartesian component of the vector, curl (f ) (x) . Thus the
curl is evaluated by expanding the following determinant along the top row.
i j k
z
f
1
(x, y, z) f
2
(x, y, z) f
3
(x, y, z)
.
Note the similarity with the cross product. More precisely and less evocatively,
f (x, y, z) =
_
F
3
y

F
2
z
_
i+
_
F
1
z

F
3
x
_
j+
_
F
2
x

F
1
y
_
k.
In the above, i = e
1
, j = e
2
, and k = e
3
the standard unit basis vectors for R
3
.
With this denition, here is the identity.
Lemma 14.4.3 Let R : U V R
3
2
and V is an
open subset of R
3
. Suppose R is C
2
and let F be a C
1
vector eld dened in V.
(R
u
R
v
) (F) (R(u, v)) = ((F R)
u
R
v
(F R)
v
R
u
) (u, v) . (14.25)
Proof: Letting x, y, z denote the components of R(u) and f
1
, f
2
, f
3
denote the
components of F, and letting a subscripted variable denote the partial derivative with
respect to that variable, the left side of 14.25 equals
i j k
x
u
y
u
z
u
x
v
y
v
z
v
i j k
x

y

z
f
1
f
2
f
3
= (f
3y
f
2z
) (y
u
z
v
z
u
y
v
) + (f
1z
f
3x
) (z
u
x
v
x
u
z
v
) + (f
2x
f
1y
) (x
u
y
v
y
u
x
v
)
= f
3y
y
u
z
v
+f
2z
z
u
y
v
+f
1z
z
u
x
v
+f
3x
x
u
z
v
+f
2x
x
u
y
v
+f
1y
y
u
x
v
(f
2z
y
u
z
v
+f
3y
z
u
y
v
+f
1z
x
u
z
v
+f
3x
z
u
x
v
+f
2x
y
u
x
v
+f
1y
x
u
y
v
)
= f
1y
y
u
x
v
+f
1z
z
u
x
v
+f
2x
x
u
y
v
+f
2z
z
u
y
v
+f
3x
x
u
z
v
+f
3y
y
u
z
v
(f
1y
y
v
x
u
+f
1z
z
v
x
u
+f
2x
x
v
y
u
+f
2z
z
v
y
u
+f
3x
x
v
z
u
+f
3y
y
v
z
u
)
At this point add in and subtract o certain terms. Then the above equals
= f
1x
x
u
x
v
+f
1y
y
u
x
v
+f
1z
z
u
x
v
+f
2x
x
u
y
v
+f
2x
y
u
y
v
+f
2z
z
u
y
v
+f
3x
x
u
z
v
+f
3y
y
u
z
v
+f
3z
z
u
z
v
_
f
1x
x
v
x
u
+f
1y
y
v
x
u
+f
1z
z
v
x
u
+f
2x
x
v
y
u
+f
2x
y
v
y
u
+f
2z
z
v
y
u
+f
3x
x
v
z
u
+f
3y
y
v
z
u
+f
3z
z
v
z
u
_
=
f
1
R(u, v)
u
x
v
+
f
2
R(u, v)
u
y
v
+
f
3
R(u, v)
u
z
v
_
f
1
R(u, v)
v
x
u
+
f
2
R(u, v)
v
y
u
+
f
3
R(u, v)
v
z
u
_
= ((F R)
u
R
v
(F R)
v
R
u
) (u, v) .
390 LINE INTEGRALS
Let U be a region in R
2
for which Greens theorem holds. Thus Greens theorem
says that for P, Q continuous on U
i
, P
v
, Q
u
L
1
(U
i
) , P, Q being C
1
on U
i
,
U
(Q
u
P
v
) dm
2
=
U
f d
where U consists of some simple closed rectiable oriented curves as explained above.
Here the u and v axes are in the same relation as the x and y axes.
Theorem 14.4.4 (Stokes Theorem) Let U be any region in R
2
for which the
conclusion of Greens theorem holds. Let R C
2
_
U, R
3
_
be a one to one function. Let
j
= R
j
,
where the
j
are parameterizations for the oriented curves making up the boundary of
U such that the conclusion of Greens theorem holds. Let S denote the surface,
S {R(u, v) : (u, v) U} ,
Then for F a C
1
vector eld dened near S,
n
i=1
i
F d
i
=
U
(R
u
(u, v) R
v
(u, v)) (F(R(u, v))) dm
2
Proof: By Lemma 14.4.1,
n
j=1
j
F d
j
=
n
j=1
j
((F R) R
u
, (F R) R
v
) d
j
By the assumption that the conclusion of Greens theorem holds for U, this equals
U
[((F R) R
v
)
u
((F R) R
u
)
v
] dm
2
=
U
[(F R)
u
R
v
+ (F R) R
vu
(F R) R
uv
(F R)
v
R
u
] dm
2
=
U
[(F R)
u
R
v
(F R)
v
R
u
] dm
2
the last step holding by equality of mixed partial derivatives, a result of the assumption
that R is C
2
. Now by Lemma 14.4.3, this equals
U
(R
u
(u, v) R
v
(u, v)) (F(R(u, v))) dm
2
This proves Stokes theorem.
With approximation arguments one can remove the assumption that R is C
2
and
replace this condition with weaker conditions. This is not surprising because in the nal
result, only rst derivatives of R occur.
14.5. INTERPRETATION AND REVIEW 391
14.5 Interpretation And Review
To understand the interpretation of Stokes theorem in terms of an integral over the
surface S, it is necessary to either do more theoretical development or to review some
beginning calculus. I will do the latter here. First of all, it is important to understand
the geometrical properties of the cross product. Those who have had a typical calculus
course will probably not have seen this so I will present it here. It is elementary material
which is a little out of place in an advanced calculus book but it is nevertheless useful
and important and if you have not seen it, you should.
14.5.1 The Geometric Description Of The Cross Product
The cross product is a way of multiplying two vectors in R
3
. It is very dierent from
the dot product in many ways. First the geometric meaning is discussed and then a
description in terms of coordinates is given. Both descriptions of the cross product
are important. The geometric description is essential in order to understand the ap-
plications to physics and geometry while the coordinate description is the only way to
practically compute the cross product. In this presentation a vector is something which
is characterized by direction and magnitude.
Denition 14.5.1 Three vectors, a, b, c form a right handed system if when
you extend the ngers of your right hand along the vector, a and close them in the
direction of b, the thumb points roughly in the direction of c.
For an example of a right handed system of vectors, see the following picture.
y
%
a
b
c
In this picture the vector c points upwards from the plane determined by the other
two vectors. You should consider how a right hand system would dier from a left hand
system. Try using your left hand and you will see that the vector, c would need to point
in the opposite direction as it would for a right hand system.
From now on, the vectors, i, j, k will always form a right handed system. To repeat,
if you extend the ngers of your right hand along i and close them in the direction j,
the thumb points in the direction of k. Recall these are the basis vectors e
1
, e
2
, e
3
.
The following is the geometric description of the cross product. It gives both the
direction and the magnitude and therefore species the vector.
Denition 14.5.2 Let a and b be two vectors in R
3
. Then a b is dened by
the following two rules.
1. |a b| = |a| |b| sin where is the included angle.
2. a b a = 0, a b b = 0, and a, b, a b forms a right hand system.
Note that |a b| is the area of the parallelogram spanned by a and b.
392 LINE INTEGRALS
Q
E
b
a
|b|sin()
%
The cross product satises the following properties.
a b = (b a) , a a = 0, (14.26)
For a scalar,
(a) b = (a b) = a(b) , (14.27)
For a, b, and c vectors, one obtains the distributive laws,
a(b +c) = a b +a c, (14.28)
(b +c) a = b a +c a. (14.29)
Formula 14.26 follows immediately from the denition. The vectors a b and b a
have the same magnitude, |a| |b| sin , and an application of the right hand rule shows
they have opposite direction. Formula 14.27 is also fairly clear. If is a nonnegative
scalar, the direction of (a) b is the same as the direction of a b,(a b) and
a(b) while the magnitude is just times the magnitude of a b which is the same
as the magnitude of (a b) and a(b) . Using this yields equality in 14.27. In the
case where < 0, everything works the same way except the vectors are all pointing in
the opposite direction and you must multiply by || when comparing their magnitudes.
The distributive laws are much harder to establish but the second follows from the rst
quite easily. Thus, assuming the rst, and using 14.26,
(b +c) a = a(b +c)
= (a b +a c)
= b a +c a.
To verify the distributive law one can consider something called the box product.
14.5.2 The Box Product, Triple Product
Denition 14.5.3 A parallelepiped determined by the three vectors, a, b, and
c consists of
{ra+sb +tc : r, s, t [0, 1]} .
That is, if you pick three numbers, r, s, and t each in [0, 1] and form ra+sb +tc, then
the collection of all such points is what is meant by the parallelepiped determined by
these three vectors.
The following is a picture of such a thing.
-
3
a
b
c
6
a b
You notice the area of the base of the parallelepiped, the parallelogram determined
by the vectors, a and b has area equal to |a b| while the altitude of the parallelepiped
is |c| cos where is the angle shown in the picture between c and a b. Therefore,
the volume of this parallelepiped is the area of the base times the altitude which is just
|a b| |c| cos = a b c.
This expression is known as the box product and is sometimes written as [a, b, c] . You
should consider what happens if you interchange the b with the c or the a with the c.
You can see geometrically from drawing pictures that this merely introduces a minus
sign. In any case the box product of three vectors always equals either the volume of
the parallelepiped determined by the three vectors or else minus this volume. From
geometric reasoning like this you see that
a b c = a b c.
In other words, you can switch the and the .
14.5.3 A Proof Of The Distributive Law For The Cross Product
Here is a proof of the distributive law for the cross product. Let x be a vector. From
the above observation,
x a(b +c) = (x a) (b +c)
= (x a) b+(x a) c
= x a b +x a c
= x (a b +a c) .
Therefore,
x [a(b +c) (a b +a c)] = 0
for all x. In particular, this holds for x = a(b +c) (a b +a c) showing that
a(b +c) = a b +a c and this proves the distributive law for the cross product.
14.5.4 The Coordinate Description Of The Cross Product
Now from the properties of the cross product and its denition,
i j = k j i = k
k i = j i k = j
j k = i k j = i
394 LINE INTEGRALS
With this information, the following gives the coordinate description of the cross prod-
uct.
Proposition 14.5.4 Let a = a
1
i +a
2
j +a
3
k and b = b
1
i +b
2
j +b
3
k be two vectors.
Then
a b = (a
2
b
3
a
3
b
2
) i+(a
3
b
1
a
1
b
3
) j+
+ (a
1
b
2
a
2
b
1
) k. (14.30)
Proof: From the above table and the properties of the cross product listed,
(a
1
i +a
2
j +a
3
k) (b
1
i +b
2
j +b
3
k) =
a
1
b
2
i j +a
1
b
3
i k +a
2
b
1
j i +a
2
b
3
j k+
+a
3
b
1
k i +a
3
b
2
k j
= a
1
b
2
k a
1
b
3
j a
2
b
1
k +a
2
b
3
i +a
3
b
1
j a
3
b
2
i
= (a
2
b
3
a
3
b
2
) i+(a
3
b
1
a
1
b
3
) j+(a
1
b
2
a
2
b
1
) k (14.31)
The easy way to remember the above formula is to write it as follows.
a b =
i j k
a
1
a
2
a
3
b
1
b
2
b
3
(14.32)
where you expand the determinant along the top row. This yields
(a
2
b
3
a
3
b
2
) i(a
1
b
3
a
3
b
1
) j+(a
1
b
2
a
2
b
1
) k (14.33)
which is the same as 14.31.
14.5.5 The Integral Over A Two Dimensional Surface
First it is good to dene what is meant by a smooth surface.
Denition 14.5.5 Let S be a subset of R
3
. Then S is a smooth surface
if there exists an open set, U R
2
and a C
1
function, R dened on U such that
R(U) = S, R is one to one, and for all (u, v) U,
R
u
R
v
= 0. (14.34)
This last condition ensures that there is always a well dened normal on S. This func-
tion, R is called a parameterization of the surface. It is just like a parameterization of
a curve but here there are two parameters, u, v.
One way to think of this is that there is a piece of rubber occupying U in the plane
and then it is taken and stretched in three dimensions. This gives S.
Denition 14.5.6 Let u
1
, u
2
be vectors in R
3
. The 2 dimensional parallelogram
determined by these vectors will be denoted by P (u
1
, u
2
) and it is dened as
P (u
1
, u
2
)
_
_
_
2
j=1
s
j
u
j
: s
j
[0, 1]
_
_
_
.
Then the area of this parallelogram is
area P (u
1
, u
2
) |u
1
u
2
| .
Suppose then that x = R(u) where u U, a subset of R
2
and x is a point in V, a
subset of 3 dimensional space. Thus, letting the Cartesian coordinates of x be given by
x = (x
1
, x
2
, x
3
)
T
, each x
i
being a function of u, an innitesimal rectangle located at
u
0
corresponds to an innitesimal parallelogram located at R(u
0
) which is determined
by the 2 vectors
_
R(u0)
u
du,
R(u0)
v
dv
_
, each of which is tangent to the surface dened
by x = R(u) . This is a very vague and unacceptable description. What exactly is
an innitesimal rectangle? However, it can all be made precise later and this is good
motivation for the real thing.
dV
u
0
du
2
du
1
:
f
u2
(u
0
)du
2
f
u1
(u
0
)du
1
f (dV )
+
From Denition 14.5.6, the volume of this innitesimal parallelepiped located at
R(u
0
) is given by
R(u
0
)
u
du
R(u
0
)
v
dv
R(u
0
)
u

R(u
0
)
v
dudv (14.35)
= |R
u
R
v
| dudv (14.36)
This motivates the following denition of what is meant by the integral over a paramet-
rically dened surface in R
3
.
Denition 14.5.7 Suppose U is a subset of R
2
and suppose R : U R(U) =
S R
3
is a one to one and C
1
function. Then if h : R(U) R, dene the 2
dimensional surface integral,
R(U)
h(x) dS according to the following formula.
S
h(x) dS
U
h(R(u)) |R
u
(u) R
v
(u)| dudv.
With this understanding, it becomes possible to interpret the meaning of Stokes
theorem. This is stated in the following theorem. Note that slightly more is assumed
here than earlier. In particular, it is assumed that R
u
R
v
= 0. This allows the
denition of a well dened normal vector which varies continuously over the surface, S.
Theorem 14.5.8 (Stokes Theorem) Let U be any region in R
2
for which the
conclusion of Greens theorem holds. Let R C
2
_
U
i
, R
3
_
be a one to one function such
that R
u
R
v
= 0 on U. Let
j
= R
j
,
where the
j
are parameterizations for the oriented bounded variation curves bounding
the region U oriented such that the conclusion of Greens theorem holds. Let S denote
the surface,
S {R(u, v) : (u, v) U} ,
396 LINE INTEGRALS
Then for F a C
1
vector eld dened near S,
n
j=1
j
F d
i
=
U
(R
u
R
v
) (F) (R(u, v)) dm
2
(14.37)
=
S
(F) ndS (14.38)
Proof: Formula 14.37 was established in Theorem 14.4.4. The unit normal of the
point R(u, v) of S is (R
u
R
v
) / |R
u
R
v
| and from the denition of the integral over
the surface, Denition 14.5.7, Formula 14.38 follows.
14.6 Introduction To Complex Analysis
14.6.1 Basic Theorems, The Cauchy Riemann Equations
With Greens theorem and the technique of proof used in proving it, it is possible to
present the most important parts of complex analysis almost eortlessly. I will do this
here and leave some of the other parts for the exercises. Recall the complex numbers
should be considered as points in the plane. Thus a complex number is of the form
x +iy where i
2
= 1. The complex conjugate is dened by
x +iy x iy
and for z a complex number,
|z| (zz)
1/2
=
x
2
+y
2
.
Thus when x +iy is considered an ordered pair (x, y) R
2
the magnitude of a complex
number is nothing more than the usual norm of the ordered pair. Also for z = x+iy, w =
u +iv,
|z w| =
(x u)
2
+ (y v)
2
so in terms of all topological considerations, R
2
is the same as C. Thus to say z f (z)
is continuous, is the same as saying
(x, y) u(x, y) , (x, y) v (x, y)
are continuous where f (z) u(x, y) + iv (x, y) with u and v being called the real and
imaginary parts of f. The only new thing is that writing an ordered pair (x, y) as x+iy
with the convention i
2
= 1 makes C into a eld. Now here is the denition of what it
means for a function to be analytic.
Denition 14.6.1 Let U be an open subset of C (R
2
) and let f : U C be a
function. Then f is said to be analytic on U if for every z U,
lim
z0
f (z + z) f (z)
z
f
(z)
exists and is a continuous function of z U. For a function having values in C denote
by u(x, y) the real part of f and v (x, y) the imaginary part. Both u and v have real
values and
f (x +iy) f (z) u(x, y) +iv (x, y)
14.6. INTRODUCTION TO COMPLEX ANALYSIS 397
Proposition 14.6.2 Let U be an open subset of C . Then f : U C is analytic if
and only if for
f (x +iy) u(x, y) +iv (x, y)
u(x, y) , v (x, y) being the real and imaginary parts of f, it follows
u
x
(x, y) = v
y
(x, y) , u
y
(x, y) = v
x
(x, y)
and all these partial derivatives, u
x
, u
y
, v
x
, v
y
are continuous on U. (The above equations
are called the Cauchy Riemann equations.)
Proof: First suppose f is analytic. First let z = ih and take the limit of the
dierence quotient as h 0 in the denition. Thus from the denition,
f
(z) lim
h0
f (z +ih) f (z)
ih
= lim
h0
u(x, y +h) +iv (x, y +h) (u(x, y) +iv (x, y))
ih
= lim
h0
1
i
(u
y
(x, y) +iv
y
(x, y)) = iu
y
(x, y) +v
y
(x, y)
Next let z = h and take the limit of the dierence quotient as h 0.
f
(z) lim
h0
f (z +h) f (z)
h
= lim
h0
u(x +h, y) +iv (x +h, y) (u(x, y) +iv (x, y))
h
= u
x
(x, y) +iv
x
(x, y) .
Therefore, equating real and imaginary parts,
u
x
= v
y
, v
x
= u
y
and this yields the Cauchy Riemann equations. Since z f
(z) is continuous, it follows

the real and imaginary parts of this function must also be continuous. Thus from the
above formulas for f
(z) , it follows from the continuity of z f
(z) all the partial

derivatives of the real and imaginary parts are continuous.
Next suppose the Cauchy Riemann equations hold and these partial derivatives are
all continuous. For z = h +ik,
f (z + z) f (z) = u(x +h, y +k) +iv (x +h, y +k) (u(x, y) +iv (x, y))
= u
x
(x, y) h +u
y
(x, y) k +i (v
x
(x, y) h +v
y
(x, y) k) +o ((h, k))
= u
x
(x, y) h +u
y
(x, y) k +i (v
x
(x, y) h +v
y
(x, y) k) +o (z)
This follows from Theorem 6.5.1 which says that C
1
implies dierentiable along with
the denition of the norm (absolute value) in C. By the Cauchy Riemann equations
this equals
= u
x
(x, y) h v
x
(x, y) k +i (v
x
(x, y) h +u
x
(x, y) k) +o (z)
= u
x
(x, y) (h +ik) +iv
x
(x, y) (h +ik) +o (z)
= u
x
(x, y) z +iv
x
(x, y) z +o (z)
Dividing by z and taking a limit yields f
(z) exists and equals u

x
(x, y) + iv
x
(x, y)
which are assumed to be continuous. This proves the proposition.
398 LINE INTEGRALS
14.6.2 Contour Integrals
The most important tools in complex analysis are Cauchys theorem in some form and
Cauchys formula for an analytic function. I will give one of the very best versions of
these theorems. They all involve something called a contour integral. Now a contour
integral is just a sort of line integral. Here is the denition.
Denition 14.6.3 Let : [a, b] C be of bounded variation and let f :

C. Letting P {t
0
, , t
n
} where a = t
0
< t
1
< < t
n
= b, dene
||P|| max {|t
j
t
j1
| : j = 1, , n}
and the Riemann Stieltjes sum by
S (P)
n
j=1
f ( (
j
)) ( (t
j
) (t
j1
))
where
j
[t
j1
, t
j
] . (Note this notation is a little sloppy because it does not identify the
specic point,
j
used. It is understood that this point is arbitrary.) Dene
f (z) dz
as the unique number which satises the following condition. For all > 0 there exists
a > 0 such that if ||P|| , then
f (z) dz S (P)
< .
Sometimes this is written as
f (z) dz lim
||P||0
S (P) .
You note that this is essentially the same denition given earlier for the line integral
only this time the function has values in C rather than R
n
and there is no dot product
involved. Instead, you multiply by the complex number (t
j
) (t
j1
) in the Riemann
Stieltjes sum. To tie this in with the line integral even more, consider a typical term in
the sum for S (P). Let (t) =
1
(t) +i
2
(t) . Then letting u be the real part of f and
v the imaginary part, S (P) equals
n
j=1
(u(
1
(
j
) ,
2
(
j
)) +iv (
1
(
j
) ,
2
(
j
)))
(
1
(t
j
)
1
(t
j1
) +i (
2
(t
j
)
2
(t
j1
)))
=
n
j=1
u(
1
(
j
) ,
2
(
j
)) (
1
(t
j
)
1
(t
j1
))
j=1
v (
1
(
j
) ,
2
(
j
)) (
2
(t
j
)
2
(t
j1
))
+i
n
j=1
v (
1
(
j
) ,
2
(
j
)) (
1
(t
j
)
1
(t
j1
))
+i
n
j=1
u(
1
(
j
) ,
2
(
j
)) (
2
(t
j
)
2
(t
j1
))
Combining these leads to
n
j=1
(u( (
j
)) , v ( (
j
))) ( (t
j
) (t
j1
))
+i
n
j=1
(v ( (
j
)) , u( (
j
))) ( (t
j
) (t
j1
)) (14.39)
Since the functions u and v are continuous, the limit as ||P|| 0 of the above equals
(u, v) d +i
(v, u) d
This proves most of the following lemma.

Lemma 14.6.4 Let be a rectiable curve in C having parameterization which is
continuous with bounded variation. Also let f : C be continuous. Then the contour
integral
f (z) dz exists and is given by the sum of the following line integrals.
f (z) dz =
(u, v) d +i
(v, u) d (14.40)
Proof: The existence of the two line integrals as limits of S (P) as ||P|| 0
follows from continuity of u, v and Theorem 14.2.3 along with the above discussion
which decomposes the sum for the contour integral into the expression of 14.39 for
which the two sums converge to the line integrals in the above formula. This proves the
lemma.
The lemma implies all the algebraic properties for line integrals hold in the same
way for contour integrals. In particular, if is C
1
, then
f (z) dz =
b
a
f ( (t))
(t) dt.
Another important observation is the following.
Proposition 14.6.5 Suppose F
(z) = f (z) for all z , an open set containing
where : [a, b] C is a continuous bounded variation curve. Then
f (z) dz = F ( (b)) F ( (a)) .

Proof: Letting u and v be real and imaginary parts of f, it follows from Lemma
14.6.4
f (z) dz =
(u, v) d +i
(v, u) d (14.41)
Consider the real valued function
G(x, y)
1
2
_
F (x +iy) +F (x +iy)
_
Re F (x +iy)
By assumption,
F
(x +iy) = f (x +iy) = u(x, y) +iv (x, y) .

Thus it is routine to verify G = (u, v). Next let the real valued function H be
dened by
H (x, y)
1
2i
_
F (x +iy) F (x +iy)
_
ImF (x +iy)
400 LINE INTEGRALS
Then H = (v, u) and so from 14.41 and Theorem 14.2.11
f (z) dz = G( (b)) G( (a)) +i (H ( (b)) H ( (a)))

= F ( (b)) F ( (a)) .
A function F such that F
= f is called a primitive of f. See how it acts a lot like a

potential, the dierence being that a primitive has complex, not real values. In calculus,
in the context of a function of one real variable, this is often called an antiderivative
and every continuous function has one thanks to the fundamental theorem of calculus.
However, it will be shown below that the situation is not at all the same for functions
of a complex variable.
14.6.3 The Cauchy Integral
The following is the rst form of the Cauchy integral theorem.
Lemma 14.6.6 Let U be an open set in C and let be a simple closed rectiable
curve contained in U having parameterization . Also let f be analytic in U. Then
f (z) dz = 0.
Proof: This follows right away from the Cauchy Riemann equations and the formula
14.40. Assume without loss of generality the orientation of is the positive orientation.
If not, the argument is the same. Then from formula 14.40,
f (z) dz =
(u, v) d +i
(v, u) d
and by Greens theorem and U
i
the inside of this equals
Ui
(v
x
u
y
) dm
2
+i
(u
x
v
y
) dm
2
= 0
by the Cauchy Riemann equations. This proves the lemma.
It is easy to improve on this result using the argument for proving Greens theorem.
You only need continuity on the bounding curve. You also dont need to make any
assumption about the functions u
x
, etc. being in L
1
(U). The following is a very
general version of the Cauchy integral theorem.
Theorem 14.6.7 Let U
i
be the inside of a simple closed rectiable curve
having parameterization . Also let f be analytic in U
i
and continuous on U
i
. Then
f (z) dz = 0.
Proof: Let B
, I
be those regions of Lemma 14.3.6 where as earlier I
are those
which have empty intersection with and B
are the border regions. Without loss of

generality, assume is positively oriented. As in the proof of Greens theorem you can
apply the same argument to the line integrals on the right of 14.40 to obtain, just as in
the proof of Greens theorem
RI
R
f (z) dz +
RB
R
f (z) dz =
f (z) dz
In this case the rst sum on the left in the above formula equals 0 from Lemma 14.6.6
for any > 0. Now just as in the proof of Greens theorem, you can choose small
enough that
RB
R
f (z) dz
< .
Since is arbitrary This proves the theorem.
With this really marvelous theorem it is time to consider the Cauchy integral formula
which represents the value of an analytic function at a point on the inside in terms of
its values on the boundary. First here are some lemmas.
Lemma 14.6.8 Let R be a rectangle such that R is positively oriented. Recall this
means the direction of motion is counter clockwise.
Then if z is on the inside of R,
1
2i
R
1
w z
dw = 1
while if z is on the outside of R, the above integral equals 0.
Proof: This follows from a routine computation and is left to you. In the case
where z is on the outside of R, the conclusion follows from the Cauchy integral formula
Theorem 14.6.7 as you can verify by noting that f (w) 1/ (w z) is analytic on an
open set containing R and that in fact its derivative equals what you would think,
1/ (w z)
2
.
Now with this little lemma, here is the Cauchy integral formula.
Theorem 14.6.9 Let be a positively oriented simple closed rectiable curve
having parameterization and let z U
i
, the inside of . Also let f be analytic on U
i
,
and continuous on U
i
. Then
f (z) =
1
2i
f (w)
w z
dw.
In particular, letting f (z) 1,
1
2i
1
w z
dw = 1.
Proof: In constructing the special regions in the proof of Greens theorem, always
choose such that the point z is not on any of the lines m = y and x = k. This makes
it possible to avoid thinking about the case where z is not on the interior of any of the
rectangles of I
. Pick small enough that I
= and z is contained in some R

0
I
.
From Lemma 14.6.8 it follows for each R I
1
2i
R
f (w)
w z
dw f (z) =
1
2i
R
f (w) f (z)
w z
dw
402 LINE INTEGRALS
Then as in the proof of Theorem 14.6.7
1
2i
RI
R
f (w) f (z)
w z
dw +
1
2i
RB
R
f (w) f (z)
w z
dw
=
1
2i
f (w) f (z)
w z
dw
By Theorem 14.6.7, all these integrals on the left equal 0 except for R
0
, the one which
contains z on its interior.
Thus the above reduces to
1
2i
R0
f (w) f (z)
w z
dw =
1
2i
f (w) f (z)
w z
dw
The integrand of the left converges to f
(z) as 0 and the length of R

0
also converges
to 0 so it follows from Theorem 14.2.4 that the limit as 0 in the above exists and
yields
0 =
1
2i
f (w) f (z)
w z
dw =
1
2i
f (w)
w z
dw f (z)
1
2i
1
w z
dw. (14.42)
Consider the last integral above.
RT
R
1
w z
dw +
RB
R
1
w z
dw =
1
w z
dw (14.43)
As in the proof of Greens theorem, choosing small enough the second sum on the left
in the above satises
RB
R
1
w z
dw
RB
R
1
w z
dw
< .
By Lemma 14.6.8, the rst sum on the left in 14.43 equals
R0
1
w z
dw
where R
0
is the rectangle for which z is on the inside of R
0
. Then by this lemma
again, this equals 2i. Therefore for such small , 14.43 reduces to
2i
1
w z
dw
<
1
2i
1
w z
dw = 1.
Now using this, 14.42 implies the claimed formula of the theorem. This proves the
theorem.
Theorem 14.6.10 Let : [a, b] C be of bounded variation. Let f be contin-
uous on
. For z /
, dene
g (z)
f (w)
w z
dw
Then g is innitely dierentiable. Futhermore,
g
(n)
(z) = n!
f (w)
(w z)
n+1
Proof:
(g (z +h) g (z)) /h =
1
h
_
f (w)
(w z h)

f (w)
(w z)
_
dw
=
1
h
f (w)
_
h
(w z h) (w z)
_
dw =
f (w)
_
1
(w z h) (w z)
_
dw
Consider only h C such that 2 |h| < dist (z,
). The integrand converges to 1/ (w z)

2
.
Then for these values of h,
1
(w z h) (w z)

1
(w z)
2
h
(w z h) (w z)
2
|h|
dist (z,
)
3
/2
=
2 |h|
dist (z,
)
3
and so the convergence of the integrand to
f (w) / (w z)
2
is uniform for |h| < dist (z,
) /2 . Using Theorem 14.2.4, it follows

g
(z) = lim
h0
g (z +h) g (z)
h
= lim
h0
1
h
_
f (w)
(w z h)

f (w)
(w z)
_
dw
=
f (w)
(w z)
2
dw.
One can then dierentiate the above expression using the same arguments. Contin-
uing this way results in the following formula.
g
(n)
(z) = n!
f (w)
(w z)
n+1
dw
It turns out that in the denition of what it means for a function dened on an open
set, U to be analytic it is not necessary to say that z f
(z) is continuous. In fact,

this comes for free. The statement that z f
(z) is continuous is REDUNDANT! The

key to understanding this is the Cauchy Goursat theorem.
14.6.4 The Cauchy Goursat Theorem
If you have two points in C, z
1
and z
2
, you can consider (t) z
1
+ t (z
2
z
1
) for
t [0, 1] to obtain a continuous bounded variation curve from z
1
to z
2
. More generally,
if z
1
, , z
m
are points in C you can obtain a continuous bounded variation curve from
z
1
to z
m
which consists of rst going from z
1
to z
2
and then from z
2
to z
3
and so
on, till in the end one goes from z
m1
to z
m
. Denote this piecewise linear curve as
(z
1
, , z
m
) . Now let T be a triangle with vertices z
1
, z
2
and z
3
encountered in the
404 LINE INTEGRALS
counter clockwise direction as shown.
z
1
z
2
z
3
Denote by
T
f (z) dz, the expression,
(z1,z2,z3,z1)
f (z) dz. Consider the following
picture.
E
E
T T
1
1
T
1
2
T
1
3
T
1
4
s
s
' E
s
z
1
z
2
z
3
Thus
T
f (z) dz =
4
k=1
T
1
k
f (z) dz. (14.44)
On the inside lines the integrals cancel because there are two integrals going in op-
posite directions for each of these inside lines.
Theorem 14.6.11 (Cauchy Goursat) Let f : C have the property that
f
(z) exists for all z and let T be a triangle contained in . Then
T
f (w) dw = 0.
Proof: Suppose not. Then
T
f (w) dw
= = 0.
From 14.44 it follows

4
k=1
T
1
k
f (w) dw
and so for at least one of these T

1
k
, denoted from now on as T
1
,
T1
f (w) dw

4
.
Now let T
1
play the same role as T. Subdivide as in the above picture, and obtain T
2
such that

T2
f (w) dw

4
2
.
Continue in this way, obtaining a sequence of triangles,
T
k
T
k+1
, diam(T
k
) diam(T) 2
k
,
and

T
k
f (w) dw

4
k
.
Then let z
k=1
T
k
and note that by assumption, f
(z) exists. Therefore, for all k

large enough,
T
k
f (w) dw =
T
k
(f (z) +f
(z) (w z) +g (w)) dw
where |g (w)| < |w z| . Now observe that w f (z) +f
(z) (w z) has a primitive,

namely,
F (w) = f (z) w +f
(z) (w z)
2
/2.
Therefore, by Proposition 14.6.5,
T
k
f (w) dw =
T
k
g (w) dw.
From Theorem 14.2.4 applied to contour integrals or the denition of the contour inte-
gral,
4
k

T
k
g (w) dw
diam(T
k
) (length of T
k
)
2
k
(length of T) diam(T) 2
k
,
and so
(length of T) diam(T) .
Since is arbitrary, this shows = 0, a contradiction. Thus
T
f (w) dw = 0 as
claimed.
This fundamental result yields the following important theorem.
Theorem 14.6.12 (Morera
1
) Let be an open set and let f

(z) exist for all
z . Let D B(z
0
, r) . Then there exists > 0 such that f has a primitive on
B(z
0
, r +). (Recall this is a function F such that F
(z) = f (z) .)
Proof: Choose > 0 small enough that B(z
0
, r +) . Then for w B(z
0
, r +) ,
dene
F (w)
(z0,w)
f (u) du.
Then by the Cauchy Goursat theorem, Theorem 14.6.11, and w B(z
0
, r +) , it
follows that for |h| small enough,
F (w +h) F (w)
h
=
1
h
(w,w+h)
f (u) du
=
1
h
1
0
f (w +th) hdt =
1
0
f (w +th) dt
which converges to f (w) due to the continuity of f at w. This proves the theorem.
The following is a slight generalization of the above theorem which is also referred to
as Moreras theorem. It contains the proof that the condition of continuity of z f
(z)
is redundant.
1
Giancinto Morera 1856-1909. This theorem or one like it dates from around 1886
406 LINE INTEGRALS
Corollary 14.6.13 Let be an open set and suppose that whenever
(z
1
, z
2
, z
3
, z
1
)
is a closed curve bounding a triangle T, which is contained in , and f is a continuous
function dened on , it follows that
(z1,z2,z3,z1)
f (z) dz = 0,
then f is analytic on . Also, if f
(z) exists for z , then z f
(z) is continuous.
Proof: As in the proof of Moreras theorem, let B(z
0
, r) and use the given
condition to construct a primitive, F for f on B(z
0
, r) . (As just shown in Theorem
14.6.12, the given condition is satised whenever f
(z) exists for all z .) Then F is

analytic and so by the Cauchy integral formula, for z B(z
0
, r)
F (z) =
1
2i
B(z0,r)
F (w)
w z
dw.
It follows from Theorem 14.6.10 that F and hence f have innitely many derivatives,
implying that f is analytic on B(z
0
, r) . Since z
0
is arbitrary, this shows f is analytic on
. In particular z f
(z) is continuous because actually this function is dierentiable.

This shows that an equivalent denition of what it means for a function to be analytic
is the following denition.
Denition 14.6.14 Let U be an open set in C and suppose f
(z) exists for all

z U. Then f is called analytic.
These theorems form the foundation for the study of functions of a complex variable.
Some important theorems will be discussed in the exercises.
14.7 Exercises
1. Suppose f : [a, b] [c, d] is continuous and one to one on (a, b) . For s (c, d) ,
show
d (f, (a, b) , s) = 1
show it is 1 if f is increasing and 1 if f is decreasing. How can this be used to
relate the degree to orientation?
2. In dening a simple curve the assumption was made that (t) = (a) and (t) =
(b) if t (a, b) . Is this fussy condition really necessary? Which theorems and
lemmas hold with simply assuming is one to one on (a, b)? Does the fussy
condition follow from assuming is one to one on (a, b)?
3. Show that for many open sets in R
2
, Area of U =
U
xdy, and Area of U =
U
ydx and Area of U =
1
2
U
ydx +xdy. Hint: Use Greens theorem.
4. A closed polygon in the plane starts at (x
0
, y
0
) , goes to (x
1
, y
1
) , to (x
2
, y
2
) to
(x
n
, y
n
) = (x
0
, y
0
) . Suppose the line segments never cross so that you have a
simple closed curve. Using Greens theorem nd a simple formula for the area
of the parallelogram. You can use Problem 3. Get the area using a line integral
obtained by adding the line integrals corresponding to the vertices of the polygon.
14.7. EXERCISES 407
5. Let be a simple C
1
oriented curve having parameterization where t is the time
and suppose f is a force dened on . Then the work done by f on an object of
mass m as it moves over the curve is dened by
f d
Newtons second law states that f = m
dv
dt
where v
(t). Let v
b
=
(b) with
v
a
dened similarly. Thus these are the nal and initial velocities. Show the work
equals
1
2
m|v
b
|
2
1
2
m|v
a
|
2
.
6. In the situation of the above problem, show that if f (x) = F (x) where F is a
potential, then if the motion is governed by the Newtons law it follows that for
(t) the motion,
F ( (t)) +
1
2
m|
(t)|
2
is constant.
7. Generalize Stokes theorem, Theorem 14.4.4 to the case where R is only assumed
C
1
.
8. Given an example of a simple closed rectiable curve and a horizontal line which
intersects this curve in innitely many points.
9. Let be a simple closed rectiable curve and let U
i
be its inside. Show you can
remove any nite number of circular disks from U
i
and what remains will still
be a region for which Greens theorem holds. Hint: You might get some ideas
from looking at the proof of Lemma 14.3.6. This is much harder than it looks
because you only know is a simple closed rectiable curve. Begin by punching
one circular hole and go from there.
10. Let : [a, b] R be of bounded variation. Show there exist increasing functions
f (t) and g (t) such that
(t) = f (t) g (t) .
Hint: You might let f (t) = V (; [a, t]) . Show this is increasing and then consider
g (t) = f (t) (t) .
11. Using Problem 10 describe another way to obtain the integral
fd for f a real
valued function and a real valued curve of bounded variation as just described
using the theory of Lebesgue integration. What exactly is this integral in this
simple case? Next extend to the case where has values in R
n
and f :
R
n
.
What are some advantages of using this other approach?
12. Suppose f is continuous but not analytic and a function of z U C. Show f has
no primitive. When functions of real variables are considered, there are function
spaces C
m
(U) which specify how many continuous derivatives the function has.
Why are such function spaces irrelevant when considering functions of a complex
variable?
13. Analytic functions are all just long polynomials. Prove this disappointing result.
More precisely prove the following. If f : U C is analytic where U is an open
set and if B(z
0
, r) U, then
f (z) =
n=0
a
n
(z z
0
)
n
(14.45)
408 LINE INTEGRALS
for all |z z
0
| < r. Furthermore,
a
n
=
f
(n)
(z
0
)
n!
. (14.46)
Hint: You use the Cauchy integral formula. For z B(z
0
, r) and C the positively
oriented boundary,
f (z) =
1
2i
C
f (w)
w z
=
1
2i
C
f (w)
w z
0
1
1
zz0
wz0
dw
=
1
2i
n=0
f (w)
(w z
0
)
n+1
(z z
0
)
n
dw
Now explain why you can switch the sum and the integral. You will need to argue
the sum converges uniformly which is what will justify this manipulation. Next
use the result of Theorem 14.6.10.
14. Prove the following amazing result about the zeros of an analytic function. Let
be a connected open set (region) and let f : X be analytic. Then the
following are equivalent.
(a) f (z) = 0 for all z
(b) There exists z
0
such that f
(n)
(z
0
) = 0 for all n.
(c) There exists z
0
which is a limit point of the set,
Z {z : f (z) = 0} .
Hint: From Problem 13, if c.) holds, then for z near z
0
f (z) =
n=m
f
(n)
(z
0
)
n!
(z z
0
)
n
Say f
(n)
(z
0
) = 0. Then consider
f (z)
(z z
0
)
m
=
f
(m)
(z
0
)
m!
+
n=m+1
f
(n)
(z
0
)
n!
(z z
0
)
nm
Now let z
n
z
0
, z
n
= z
0
but f (z
n
) = 0. What does this say about f
(m)
(z
0
)?
Clearly the rst two conditions are equivalent and they imply the third.
15. You want to dene e
z
for z complex such that it is analytic on C. Using Problem
14 explain why there is at most one way to do it and still have it coincide with e
x
when z = x +i0. Then show using the Cauchy Riemann equations that
e
z
e
x
(cos (y) +i sin (y))
is analytic and agrees with e
x
when z = x +i0. Also show
d
dz
e
z
= e
z
.
Hint: For the rst part, suppose two functions, f, g work. Then consider f g.
this is analytic and has a zero set, R.
14.7. EXERCISES 409
16. Do the same thing as Problem 15 for sin (z) , cos (z) . Also explain with a very short
argument why all identities for these functions continue to hold for the extended
functions. This argument shouldnt require any computations at all. Why is
sin (z) no longer bounded if z is allowed to be complex? Hint: You might try
something involving the above formula for e
z
to get the denition.
17. Show that if f is analytic on C and f
(z) = 0 for all z, then f (z) c for some

constant c C. You might want to use Problem 14 to do this really quickly. Now
using Theorem 14.6.10 prove Liouvilles theorem which states that a function
which is analytic on all of C which is also bounded is constant. Hint: By that
theorem,
f
(z) =
1
2i
Cr
f (w)
(w z)
2
dw
where C
r
is the positively oriented circle of radius r which is centered at z. Now
consider what happens as r . You might use the corresponding version of
Theorem 14.2.4 applied to contour integrals and note the total length of C
r
is 2r.
18. Using Problem 15 prove the fundamental theorem of algebra which says every
nonconstant polynomial having complex coecients has at least one zero in C.
(This is the very best way to prove the fundamental theorem of algebra.) Hint:
If p (z) has no zeros, consider 1/p (z) and prove it must then be bounded and
analytic on all of C.
19. Let f be analytic on U
i
, the inside of , a rectiable simple closed curve positively
oriented with parameterization . Suppose also there are no zeros of f on .
Show then that the number of zeros, of f contained in U
i
counted according to
multiplicity is given by the formula
1
2i
(z)
f (z)
dz
Hint: You ought to rst show f (z) =
m
k=1
(z z
k
) g (z) where the z
k
are the
zeros of f in U
i
and g (z) is an analytic function which never vanishes in U
i
. In
the above product there might be some repeats corresponding to repeated zeros.
20. An open connected set U is said to be star shaped if there exists a point z
0
U
called a star center such that the for all z U, (z
0
, z)
as described in before
the proof of the Cauchy Goursat theorem is contained in U. For example, pick
any complex number and consider everything left after leaving out the ray
{t : t 0} . Show this is star shaped with a star center t for t < 0. Now for U
a star shaped open connected set, suppose g is analytic on U and g (z) = 0 for all
z U. Show there exists an analytic function h dened on U such that
e
h(z)
= g (z) .
This function h(z) is like log (g (z)). Hint: Use an argument like that used to
prove Moreras theorem and the Cauchy Goursat theorem to obtain a primitive
for g
/g, h
1
. Next consider the function
ge
h1
Using the chain rule and the product rule, show
d
dz
_
ge
h1
_
= 0. Using one of the
results of Problem 17 show
g = ce
h1
for some constant c. Tell why c can be written as e
a+ib
. Then let h = h
1
+a +ib.
410 LINE INTEGRALS
21. One of the most amazing theorems is the open mapping theorem. Let U be an
open connected set in C and suppose f : U C is analytic. Then f (U) is either
a point or an open connected set. In the case where f (U) is an open connected
set, it follows that for each z
0
U, there exists an open set, V containing z
0
and
m N such that for all z V,
f (z) = f (z
0
) +(z)
m
(14.47)
where : V B(0, ) is one to one, analytic and onto, (z
0
) = 0,
(z) = 0 on
V and
1
analytic on B(0, ) . If f is one to one then m = 1 for each z
0
and
f
1
: f (U) U is analytic. Consider the real valued function f (x) = x
2
. f (R)
is neither a point nor an open connected set. This is a strictly complex analysis
phenomenon. Hint:Work out the details of the following outline. Suppose f (U)
is not a point. Then using Problem 14 about the zeros of an analytic function
there exists r > 0 such that for z B(z
0
, r) \ {z
0
} ,
f (z) f (z
0
) = 0.
Explain why there exists g (z) analytic and nonzero on B(z
0
, r) such that for some
positive integer m,
f (z) f (z
0
) = (z z
0
)
m
g (z)
Next one tries to take the m
th
root of g (z) . Using Problem 20 there exists h
analytic such that
g (z) = e
h(z)
, g (z) =
_
e
h(z)/m
_
m
Now let (z) = (z z
0
) e
h(z)/m
. This yields the formula 14.47. Also
(z
0
) =
e
h(z0)/m
= 0. Now consider
(x +iy) = u(x, y) +iv (x, y)
and the map
_
x
y
_
_
u(x, y)
v (x, y)
_
Here u, v are C
1
because is analytic. Use the Cauchy Riemann equations to
verify the Jacobian of this transformation at (x
0
, y
0
) is nonzero. This is where
you use
(z
0
) = 0. Use inverse function theorem to verify maps some open
set V containing z
0
one to one and onto B(0, ) . Thus also
m
maps V onto
B(0,
m
). Explain why it follows from 14.47 and the fact that z
0
is arbitrary that
f is an open map. Since f is continuous and U is connected, so is f (U). However,
if m > 1, this mapping, f cant be one to one. To verify this,
e
i2/m
(z
1
) = (z
1
)
but both are in B(0, ). Hence there exists z
2
= z
1
such that (z
2
) = e
i2/m
(z
1
)
( is one to one) but f (z
2
) = f (z
1
). If f is one to one, then the above shows
that f
1
is continuous and for each z, the m in the above is always 1 so f
(z) =
e
h(z)/1
= 0. Hence
_
f
1
_
(f (z)) = lim
f(z1)f(z)
f
1
(f (z
1
)) f
1
(f (z))
f (z
1
) f (z)
= lim
z1z
z
1
z
f (z
1
) f (z)
=
1
f
(z)
14.7. EXERCISES 411
22. Let U be what is left when you leave out the ray t for t 0. This is a star
shaped open set and g (z) = z is nonzero on this set. Therefore, there exists h(z)
such that z = e
h(z)
by Problem 20. Explain why h(z) is analytic on U. When
= 1 this is called the principle branch of the logarithm. In this case dene
Arg (z) (, ) such that the given z equals |z| e
i
. Explain why this
principle branch of the logarithm is
log (z) = ln (|z|) +iArg (z)
Note it follows from the open mapping theorem this is an analytic function on U.
You dont have to fuss with any tedium in order to show this.
23. Suppose is a simple closed curve and let U
i
be the inside. Suppose f is analytic
on U
i
and continuous on U
i
. Consider the function z |f (z)|. This is a
continuous function. Show that if it achieves its maximum at any point of U
i
then
f must be a constant. Hint: You might use the open mapping theorem.
24. Let f, g be analytic on U
i
, the inside of , a rectiable simple closed curve positively
oriented with parameterization . Suppose either
|f (z) +g (z)| < |f (z)| +|g (z)| on
or
|f (z) g (z)| < |f (z)| on
Let Z
f
denote the number of zeros in U
i
and let Z
g
denote the number of zeros of
g in U
i
. Then neither f, g, nor f/g can equal zero anywhere on and Z
f
= Z
g
.
Hint: The rst condition implies for all z ,
f (z)
g (z)
C \ [0, )
Show there exists a primitive F for
(f/g)
f/g
.
and argue
0 =
(f/g)
f/g
dz =
dz
g
dz = Z
f
Z
g
.
You could consider F = L(f/g) where L is the analytic function dened on
C \ [0, ) with the property that
e
L(z)
= z.
Thus
e
L(z)
L
(z) = 1, L
(z) = 1/z.
In the second case, show g/f / (, 0] and so a similar thing can be done. This
problem is a case of Rouches theorem.
25. Use the result of Problem 24 to give another proof of the fundamental theorem of
algebra as follows. Let g (z) be a polynomial of degree n, a
n
z
n
+a
n1
z
n1
+ +
a
1
z +a
0
where a
n
= 0. Now let f (z) = a
n
z
n
. Let be a big circle, large enough
that |f (z) g (z)| < |f (z)| on this circle. Then tell why g and f have the same
number of zeros where they are counted according to multiplicity.
412 LINE INTEGRALS
26. Let p (z) = z
7
+11z
3
5z
2
+5. Identify a ball B(0, r) which must contain all the
zeros of p (z) . Try to make r reasonably small. Use Problem 24.
27. Here is another approach to the open mapping theorem which I think might be a
little easier and shorter which is based on Rouches theorem and makes no reference
to real variable techniques. Let f : U C where U is an open connected set.
Then f (U) is either an open connected set or a point. Hint: Suppose f (U) is
not a point. Then explain why for any z
0
U there exists r > 0 such that
f (z) f (z
0
) = g (z) (z z
0
)
m
where g is analytic and nonzero on B(z
0
, r). Now consider the function z
f (z) w. I would like to use Rouches theorem to claim this function has the
same number of zeros, namely m as the function z f (z) f (z
0
). Let
= min {|f (z) f (z
0
)| : |z z
0
| = r}
Then if |w f (z
0
)| < ,
|w f (z
0
)| = |f (z) f (z
0
) (f (z) w)| < |f (z) f (z
0
)|
for each z B(z
0
, r) and so you can apply Rouches theorem. What does this
say about when f is one to one? Why is f (U) open? Why is f (U) connected?
28. Let : [a, b] C be of bounded variation, (a) = (b) and suppose z /
.
Dene
n(, z)
1
2i
dw
w z
.
This is called the winding number. When
is positively oriented and a simple

closed curve, this number equals 1 by the Cauchy integral formula. However, it is
always an integer. Furthermore, z n(, z) is continuous and so is constant on
every component of C\
. For z in the unbounded component, n(, z) = 0. Most

modern treatments of complex analysis feature the winding number extensively in
the statement of all the major theorems. This is because it makes possible the most
general form of the theorems. Prove the above properties of the winding number.
Hint: The continuity is easy. It follows right away from a simple estimate and
Theorem 14.2.4 applied to contour integrals. The tricky part is in showing it is an
integer. This is where it is convenient to use Theorem 14.2.6 applied to contour
integrals. There exists : [a, b] C which is C
1
on [a, b] and
max {| (t) (t)| : t [a, b]} < ,
(a) = (b) = (a) = (b)
1
2i
dw
w z

1
2i
dw
w z
<
where < dist (z,
) . Thus z /
. Consider the contour integral which involves

and show it is an integer. Then there exists a sequence of these C
1
contours
{
k
} such that
1
2i
dw
w z

1
2i
k
dw
w z
0.
Consequently, for all k large enough there can be no change in
1
2i
k
dw
w z
14.7. EXERCISES 413
which shows
1
2i
dw
w z
is an integer as claimed. So how do you show the contour integral involving
yields an integer? As mentioned above,
1
2i
k
dw
w z
=
1
2i
b
a
(t)
(t) z
dt
Let
g (t)
t
a
(s)
(s) z
ds
Formally this is a lot like some sort of log ( (s) z) (recall beginning calculus) so
it is reasonable to consider
_
e
g(t)
(t) z
_
.
Show this equals 0. Explain why this requires the function which is dierentiated
must be constant. Thus
e
g(a)
(a) z
=
e
g(b)
(b) z
Now (a) = (b) , g (a) = 0, and so e
g(a)
= 1 = e
g(b)
. Explain why this requires
g (b) = 2mi for m an integer. Now this gives the desired result.
29. Let
B
(a, r) {z C such that 0 < |z a| < r} .

Thus this is the usual ball without the center. A function is said to have an
isolated singularity at the point a C if f is analytic on B
(a, r) for some r > 0.

An isolated singularity of f is said to be removable if there exists an analytic
function, g analytic at a and near a such that f = g at all points near a. A major
theorem is the following.
Theorem 14.7.1 Let f : B
(a, r) X be analytic. Thus f has an iso-

lated singularity at a. Suppose also that
lim
za
f (z) (z a) = 0.
Then there exists a unique analytic function, g : B(a, r) X such that g = f on
B
(a, r) . Thus the singularity at a is removable.

Prove this theorem. Hint: Let h(z) = f (z) (z a)
2
. Then h(a) = 0 and h
(a)
exists and equals 0. Show this. Also h is analytic near a. Therefore,
h(z) =
k=2
a
k
(z a)
k
Maybe consider g (z) = h(z) / (z a)
2
. Argue g is analytic and equals f for z
near a.
30. Another really amazing theorem in complex analysis is the Casorati Weierstrass
theorem.
414 LINE INTEGRALS
Theorem 14.7.2 Let a be an isolated singularity and suppose for some
r > 0, f (B
(a, r)) is not dense in C. Then either a is a removable singularity

or there exist nitely many b
1
, , b
M
for some nite number, M such that for z
near a,
f (z) = g (z) +
M
k=1
b
k
(z a)
k
(14.48)
where g (z) is analytic near a. When the above formula holds, f is said to have a
pole of order M at a.
Prove this theorem. Hint: Suppose a is not removable and B(z
0
, ) has no points
of f (B
(a, r)) . Such a ball must exist if f (B
(a, r)) is not dense in the plane.

this means that for all 0 < |z a| < r,
|f (z) z
0
| > 0
Hence
lim
za
1
f (z) z
0
(z a) = 0
and so 1/ (f (z) z
0
) has a removable sincularity at a. See Problem 29. Let g (z)
be analytic at and near a and agree with this function. Thus
g (z) =
n=0
a
n
(z a)
n
.
There are two cases, g (a) = 0 and g (a) = 0. First suppose g (a) = 0. Then explain
why
g (z) = h(z) (z a)
m
where h(z) is analytic and non zero near a. Then
f (z) z
0
=
1
h(z)
1
(z a)
m
Show this yields the desired conclusion. Next suppose g (a) = 0. Then explain why
g (z) = 0 near a and this would contradict the assertion that a is not removable.
31. One of the very important techniques in complex analysis is the method of residues.
When a is a pole the residue of f at a denoted by res (f, a) , is dened as b
1
in
14.48. Suppose a is a pole and is a simple closed rectiable curve containing a
on the inside with no other singular points on or anywhere else inside . Show
that under these conditions,
f (z) dz = 2i (res (f, a))

Also describe a way to nd res (f, a) by multiplying by (z a)
m
and dierentiat-
ing. Hint: You should show
1
(za)
mdz = 0 whenever m > 1. This is because
the function has a primitive.
32. Using Problem 9 give a holy version of the Cauchy integral theorem. This is
it. Let be a positively oriented rectiable simple closed curve with inside U
i
and remove nitely many open discs B(z
j
, r
j
) from U
i
. Thus the result is a holy
14.7. EXERCISES 415
region. Suppose f is analytic on some open set containing U
i
\
n
j=1
B(z
j
, r
j
) .
Then letting
j
denote the negatively oriented boundary of B(z
j
, r
j
) , show
0 =
f (z) dz +
n
j=1
j
f (z) dz
where
j
is a parameterization for
j
. Hint: The proof is the same as given
earlier. You just use Greens theorem.
33. Let be a simple closed curve and suppose on its inside there are nitely many
poles for a function f which is analytic near . Call these poles {z
k
}
n
k=1
. Then
f (z) dz = 2i
n
j=1
res (f, z
j
)
This is the very important residue theorem for computing line integrals. Hint:
You should use Problem 32 and Problem 30, the Casorati Weierstrass theorem.
416 LINE INTEGRALS
Hausdor Measures And Area
Formula
15.1 Denition Of Hausdor Measures
This chapter is on Hausdor measures. First I will discuss some outer measures. In all
that is done here, (n) will be the volume of the ball in R
n
which has radius 1. This
volume is the usual Lebesgue measure and the balls will be determined by the usual
norm on R
n
.
Denition 15.1.1 For a set, E, denote by r (E) the number which is half the
diameter of E. Thus
r (E)
1
2
sup {|x y| : x, y E}
1
2
diam(E)
Let E R
n
.
H
s
(E) inf{
j=1
(s)(r (C
j
))
s
: E
j=1
C
j
, r (C
j
) }
H
s
(E) lim
0
H
s
(E).
In the above denition, (s) is an appropriate positive constant depending on s.
Later I will tell what this constant is but it is not important for now. It will be chosen
in such a way that whenever n is a positive integer, H
n
([0, 1)
n
) = 1 = m
n
([0, 1)
n
) . In
fact, this is all you need to know about it.
Lemma 15.1.2 H
s
and H
s
are outer measures.

Proof: It is clear that H
s
() = 0 and if A B, then H
s
(A) H
s
(B) with similar
assertions valid for H
s
. Suppose E =
i=1
E
i
and H
s
(E
i
) < for each i. Let {C
i
j
}
j=1
be a covering of E
i
with
j=1
(s)(r(C
i
j
))
s
/2
i
< H
s
(E
i
)
417
418 HAUSDORFF MEASURES AND AREA FORMULA
and diam(C
i
j
) . Then
H
s
(E)
i=1
j=1
(s)(r(C
i
j
))
s
i=1
H
s
(E
i
) +/2
i
+
i=1
H
s
(E
i
).
It follows that since > 0 is arbitrary,
H
s
(E)
i=1
H
s
(E
i
)
which shows H
s
is an outer measure. Now notice that H

s
(E) is increasing as 0.
Picking a sequence
k
decreasing to 0, the monotone convergence theorem implies
H
s
(E)
i=1
H
s
(E
i
).
The outer measure H
s
is called s dimensional Hausdor measure when restricted to
the algebra of H
s
measurable sets. Recall these are the sets E such that for all S,
H
s
(S) = H
s
(S E) +H
s
(S \ E) .
Next I will show the algebra of H
s
measurable sets includes the Borel sets. This
is done by the following very interesting condition known as Caratheodorys criterion.
15.1.1 Properties Of Hausdor Measure
Denition 15.1.3 For two sets A, B in a metric space, dene
dist (A, B) inf {||x y|| : x A, y B} .
Theorem 15.1.4 Let be an outer measure on the subsets of X, a closed subset
of a normed vector space and suppose
(A B) = (A) +(B)
whenever dist(A, B) > 0, then the algebra of measurable sets contains the Borel sets.
Proof: It suces to show that closed sets are in F, the -algebra of measurable
sets, because then the open sets are also in F and consequently F contains the Borel
sets. Let K be closed and let S be a subset of . Is (S) (S K) + (S \ K)? It
suces to assume (S) < . Let
K
n
{x : dist(x, K)
1
n
}
By Lemma 7.4.4 on Page 161, x dist (x, K) is continuous and so K
n
is closed. By
the assumption of the theorem,
(S) ((S K) (S \ K
n
)) = (S K) +(S \ K
n
) (15.1)
15.1. DEFINITION OF HAUSDORFF MEASURES 419
since S K and S \ K
n
are a positive distance apart. Now
(S \ K
n
) (S \ K) (S \ K
n
) +((K
n
\ K) S). (15.2)
If lim
n
((K
n
\ K) S) = 0 then the theorem will be proved because this limit
along with 15.2 implies lim
n
(S \ K
n
) = (S \ K) and then taking a limit in 15.1,
(S) (S K) +(S \ K) as desired. Therefore, it suces to establish this limit.
Since K is closed, a point, x / K must be at a positive distance from K and so
K
n
\ K =
k=n
K
k
\ K
k+1
.
Therefore
(S (K
n
\ K))
k=n
(S (K
k
\ K
k+1
)). (15.3)
If
k=1
(S (K
k
\ K
k+1
)) < , (15.4)
then (S (K
n
\ K)) 0 because it is dominated by the tail of a convergent series so
it suces to show 15.4.
M
k=1
(S (K
k
\ K
k+1
)) =
k even, kM
(S (K
k
\ K
k+1
)) +
k odd, kM
(S (K
k
\ K
k+1
)). (15.5)
By the construction, the distance between any pair of sets S (K
k
\ K
k+1
) for dierent
even values of k is positive and the distance between any pair of sets S (K
k
\ K
k+1
)
for dierent odd values of k is positive. Therefore,
k even, kM
(S (K
k
\ K
k+1
)) +
k odd, kM
(S (K
k
\ K
k+1
))
(
k even
S (K
k
\ K
k+1
)) +(
k odd
S (K
k
\ K
k+1
)) 2(S) <
and so for all M,
M
k=1
(S (K
k
\ K
k+1
)) 2(S) showing 15.4 and proving the
theorem.
The next theorem applies the Caratheodory criterion above to H
s
.
Theorem 15.1.5 The algebra of H
s
measurable sets contains the Borel sets
and H
s
has the property that for all E R
n
, there exists a Borel set F E such that
H
s
(F) = H
s
(E).
Proof: Let dist(A, B) = 2
0
> 0. Is it the case that
H
s
(A) +H
s
(B) = H
s
(A B)?
This is what is needed to use Caratheodorys criterion.
Let {C
j
}
j=1
be a covering of A B such that r(C
j
) <
0
/2 for each j and
H
s
(A B) + >
j=1
(s)(r (C
j
))
s
.
Thus
H
s
(A B
) + >
jJ1
(s)(r (C
j
))
s
+
jJ2
(s)(r (C
j
))
s
where
J
1
= {j : C
j
A = }, J
2
= {j : C
j
B = }.
Recall dist(A, B) = 2
0
and so J
1
J
2
= . It follows
H
s
(A B) + > H
s
(A) +H
s
(B).
Letting 0, and noting > 0 was arbitrary, yields
H
s
(A B) H
s
(A) +H
s
(B).
Equality holds because H
s
is an outer measure. By Caratheodorys criterion, H
s
is a
Borel measure.
To verify the second assertion, note rst there is no loss of generality in letting
H
s
(E) < . Let
E
j=1
C
j
, r(C
j
) < ,
and
H
s
(E) + >
j=1
(s)(r (C
j
))
s
.
Let
F
j=1
C
j
.
Thus F
E and
H
s
(E) H
s
(F
j=1
(s)(r
_
C
j
_
)
s
=
j=1
(s)(r (C
j
))
s
< +H
s
(E).
Let
k
0 and let F =
k=1
F
k
. Then F E and
H
s
k
(E) H
s
k
(F) H
s
k
(F
)
k
+H
s
k
(E).
Letting k ,
H
s
(E) H
s
(F) H
s
(E)
A measure satisfying the rst conclusion of Theorem 15.1.5 is sometimes called a
Borel regular measure.
15.1.2 H
n
And m
n
Next I will compare H
n
and m
n
. To do this, recall the following covering theorem which
is a summary of Corollaries 9.7.5 and 9.7.4 found on Page 230.
Theorem 15.1.6 Let E R
n
and let F, be a collection of balls of bounded radii
such that F covers E in the sense of Vitali. Then there exists a countable collection of
disjoint balls from F, {B
j
}
j=1
, such that m
n
(E \
j=1
B
j
) = 0.
In the next lemma, the balls are the usual balls taken with respect to the usual
distance in R
n
.
15.1. DEFINITION OF HAUSDORFF MEASURES 421
Lemma 15.1.7 If m
n
(S) = 0 then H
n
(S) = H
n
(S) = 0. Also, there exists a

constant, k such that H
n
(E) km
n
(E) for all E Borel. Also, if Q
0
[0, 1)
n
, the unit
cube, then H
n
([0, 1)
n
) > 0.
Proof: Suppose rst m
n
(S) = 0. First suppose S is bounded. Then by outer
regularity, there exists a bounded open V containing S and m
n
(V ) < . For each
x S, there exists a ball B
x
such that

B
x
V and > r
_
B
x
_
. By the Vitali covering
theorem there is a sequence of disjoint balls {B
k
} such that
_
B
k
_
covers S. Then letting
(n) be the Lebesgue measure of the unit ball in R
n
H
n
(S)
k
(n) r
_
B
k
_
n
=
(n)
(n)
5
n
k
(n) r (B
k
)
n
(n)
(n)
5
n
m
n
(V ) <
(n)
(n)
5
n
Since is arbitrary, this shows H

n
(S) = 0 and now it follows H

n
(S) = 0. In case S is not
bounded, let S
m
= B(0,m) S. Then H
n
(S
m
) = 0 and so letting m , H
n
(S) = 0
also. Then as before, H
n
(S) = 0.
Letting U be an open set and > 0, consider all balls, B contained in U which
have diameters less than . This is a Vitali covering of U and therefore by Theorem
15.1.6, there exists {B
i
} , a sequence of disjoint balls of radii less than contained in
U such that
i=1
B
i
diers from U by a set of Lebesgue measure zero. Let (n) be the
Lebesgue measure of the unit ball in R
n
. Then from what was just shown,
H
n
(U) = H
n
(
i
B
i
)
i=1
(n) r (B
i
)
n
=
(n)
(n)
i=1
(n) r (B
i
)
n
=
(n)
(n)
i=1
m
n
(B
i
) =
(n)
(n)
m
n
(U) km
n
(U) .
Now letting E be Borel, it follows from the outer regularity of m
n
there exists a de-
creasing sequence of open sets {V
i
} containing E such such that m
n
(V
i
) m
n
(E) .
Then from the above,
H
n
(E) lim
i
H
n
(V
i
) lim
i
km
n
(V
i
) = km
n
(E) .
Since > 0 is arbitrary, it follows that also
H
n
(E) km
n
(E) .
This proves the rst part of the lemma.
To verify the second part, note that it is obvious H
n
and H
n
are translation invariant
because diameters of sets do not change when translated. Therefore, if H
n
([0, 1)
n
) = 0,
it follows H
n
(R
n
) = 0 because R
n
is the countable union of translates of Q
0
[0, 1)
n
.
Since each H
n
is no larger than H
n
, the same must hold for H
n
. Therefore, there exists

a sequence of sets {C
i
} each having diameter less than such that the union of these
sets equals R
n
but
1 >
i=1
(n) r (C
i
)
n
.
Now let B
i
be a ball having radius equal to diam(C
i
) = 2r (C
i
) which contains C
i
. It
follows
m
n
(B
i
) = (n) 2
n
r (C
i
)
n
=
(n) 2
n
(n)
(n) r (C
i
)
n
which implies
1 >
i=1
(n) r (C
i
)
n
=
i=1
(n)
(n) 2
n
m
n
(B
i
) = ,
a contradiction. This proves the lemma.
Lemma 15.1.8 Every open set in R
n
is the countable disjoint union of half open
boxes of the form
n
i=1
(a
i
, a
i
+ 2
k
]
where a
i
= l2
k
for some integers, l, k. The sides of these boxes are of equal length.
One could also have half open boxes of the form
n
i=1
[a
i
, a
i
+ 2
k
)
and the conclusion would be unchanged.
Proof: Let
C
k
= {All half open boxes
n
i=1
(a
i
, a
i
+ 2
k
] where
a
i
= l2
k
for some integer l.}
Thus C
k
consists of a countable disjoint collection of boxes whose union is R
n
. This is
sometimes called a tiling of R
n
. Think of tiles on the oor of a bathroom and you will
get the idea. Note that each box has diameter no larger than 2
k
n. This is because if
x, y
n
i=1
(a
i
, a
i
+ 2
k
],
then |x
i
y
i
| 2
k
. Therefore,
|x y|
_
n
i=1
_
2
k
_
2
_
1/2
= 2
k
n.
Let U be open and let B
1
all sets of C
1
which are contained in U. If B
1
, , B
k
have
been chosen, B
k+1
all sets of C
k+1
contained in
U \
_
k
i=1
B
i
_
.
Let B
i=1
B
i
. In fact B
= U. Clearly B
U because every box of every B

i
is contained in U. If p U, let k be the smallest integer such that p is contained in a
box from C
k
which is also a subset of U. Thus
p B
k
B
.
Hence B
is the desired countable disjoint collection of half open boxes whose union
is U. The last assertion about the other type of half open rectangle is obvious. This
proves the lemma.
Theorem 15.1.9 By choosing (n) properly, one can obtain H
n
= m
n
on all
Lebesgue measurable sets.
15.2. TECHNICAL CONSIDERATIONS
423
Proof: I will show H
n
is a positive multiple of m
n
for any choice of (n) . Dene
k =
m
n
(Q
0
)
H
n
(Q
0
)
where Q
0
= [0, 1)
n
is the half open unit cube in R
n
. I will show kH
n
(E) = m
n
(E) for
any Lebesgue measurable set. When this is done, it will follow that by adjusting (n)
the multiple can be taken to be 1.
Let Q =
n
i=1
[a
i
, a
i
+ 2
k
) be a half open box where a
i
= l2
k
. Thus Q
0
is the
union of
_
2
k
_
n
of these identical half open boxes. By translation invariance, of H
n
and
m
n
_
2
k
_
n
H
n
(Q) = H
n
(Q
0
) =
1
k
m
n
(Q
0
) =
1
k
_
2
k
_
n
m
n
(Q) .
Therefore, kH
n
(Q) = m
n
(Q) for any such half open box and by translation invariance,
for the translation of any such half open box. It follows from Lemma 15.1.8 that
kH
n
(U) = m
n
(U) for all open sets. It follows immediately, since every compact set
is the countable intersection of open sets that kH
n
= m
n
on compact sets. Therefore,
they are also equal on all closed sets because every closed set is the countable union of
compact sets. Now let F be an arbitrary Lebesgue measurable set. I will show that F
is H
n
measurable and that kH
n
(F) = m
n
(F). Let F
l
= B(0, l) F. Then there exists
H a countable union of compact sets and G a countable intersection of open sets such
that
H F
l
G (15.6)
and m
n
(G\ H) = 0 which implies by Lemma 15.1.7
m
n
(G\ H) = kH
n
(G\ H) = 0. (15.7)
To do this, let {G
i
} be a decreasing sequence of bounded open sets containing F
l
and
let {H
i
} be an increasing sequence of compact sets contained in F
l
such that
kH
n
(G
i
\ H
i
) = m
n
(G
i
\ H
i
) < 2
i
Then letting G =
i
G
i
and H =
i
H
i
this establishes 15.6 and 15.7. Then by com-
pleteness of H
n
it follows F
l
is H
n
measurable and
kH
n
(F
l
) = kH
n
(H) = m
n
(H) = m
n
(F
l
) .
Now taking l , it follows F is H
n
measurable and kH
n
(F) = m
n
(F). Therefore,
adjusting (n) it can be assumed the constant, k is 1. This proves the theorem.
The exact determination of (n) is more technical. You can skip it if you want.
Just remember (n) is chosen such that H
n
([0, 1)
n
) = 1. It turns out this will require
(n) = (n) where (n) is the volume of the unit ball taken with respect to the usual
norm. The optional sections are starred.
15.2 Technical Considerations
Let (n) be the volume of the unit ball in R

n
. Thus the volume of B(0, r) in R
n
is
(n)r
n
from the change of variables formula. There is a very important and interesting
inequality known as the isodiametric inequality which says that if A is any set in R
n
,
then
m(A) (n)(2
1
diam(A))
n
.
This inequality may seem obvious at rst but it is not really. The reason it is not is
that there are sets which are not subsets of any sphere having the same diameter as the
set. For example, consider an equilateral triangle.
Lemma 15.2.1 Let f : R
n1
[0, ) be Borel measurable and let
S = {(x,y) :|y| < f(x)}.
Then S is a Borel set in R
n
.
Proof: Set s
k
be an increasing sequence of Borel measurable functions converging
pointwise to f.
s
k
(x) =
N
k
m=1
c
k
m
X
E
k
m
(x).
Let
S
k
=
N
k
m=1
E
k
m
(c
k
m
, c
k
m
).
Then (x,y) S
k
if and only if f(x) > 0 and |y| < s
k
(x) f(x). It follows that
S
k
S
k+1
and
S =
k=1
S
k
.
But each S
k
is a Borel set and so S is also a Borel set. This proves the lemma.
Let P
i
be the projection onto
span (e
1
, , e
i1
, e
i+1
, , e
n
)
where the e
k
are the standard basis vectors in R
n
, e
k
being the vector having a 1 in the
k
th
slot and a 0 elsewhere. Thus P
i
x
j=i
x
j
e
j
. Also let
A
Pix
{x
i
: (x
1
, , x
i
, , x
n
) A}
x
A
Pix
P
i
x span{e
1
, , e
i1
e
i+1
, , e
n
}.
Lemma 15.2.2 Let A R
n
be a Borel set. Then P
i
x m(A
Pix
) is a Borel
measurable function dened on P
i
(R
n
).
Proof: Let K be the system consisting of sets of the form
n
j=1
A
j
where A
i
is
Borel. Also let G denote those Borel sets of R
n
such that if A G then
P
i
x m((A R
k
)
Pix
) is Borel measurable.
where R
k
= (k, k)
n
. Thus K G. If A G
P
i
x m
_
_
A
C
R
k
_
Pix
_
is Borel measurable because it is of the form
m
_
(R
k
)
Pix
_
m
_
(A R
k
)
Pix
_
and these are Borel measurable functions of P
i
x. Also, if {A
i
} is a disjoint sequence of
sets in G then
m
_
(
i
A
i
R
k
)
Pix
_
=
i
m
_
(A
i
R
k
)
Pix
_
425
and each function of P
i
x is Borel measurable. Thus by the lemma on systems G =
B (R
n
) and This proves the lemma.
Now let A R
n
be Borel. Let P
i
be the projection onto
span
_
e
1
, , e
i1
, e
i+1
, , e
n
_
and as just described,
A
Pix
= {y R : P
i
x +ye
i
A}
Thus for x = (x
1
, , x
n
),
A
Pix
= {y R : (x
1
, , x
i1
, y, x
i+1
, , x
n
) A}.
Since A is Borel, it follows from Lemma 15.2.1 that
P
i
x m(A
Pix
)
is a Borel measurable function on P
i
R
n
= R
n1
.
15.2.1 Steiner Symmetrization
Dene
S(A, e
i
) {x =P
i
x +ye
i
: |y| < 2
1
m(A
Pix
)}
Lemma 15.2.3 Let A be a Borel subset of R
n
. Then S(A, e
i
) satises
P
i
x +ye
i
S(A, e
i
) if and only if P
i
x ye
i
S(A, e
i
),
S(A, e
i
) is a Borel set in R
n
,
m
n
(S(A, e
i
)) = m
n
(A), (15.8)
diam(S(A, e
i
)) diam(A). (15.9)
Proof : The rst assertion is obvious from the denition. The Borel measurability
of S(A, e
i
) follows from the denition and Lemmas 15.2.2 and 15.2.1. To show Formula
15.8,
m
n
(S(A, e
i
)) =
PiR
n
2
1
m(AP
i
x)
2
1
m(A
P
i
x
)
dx
i
dx
1
dx
i1
dx
i+1
dx
n
=
PiR
n
m(A
Pix
)dx
1
dx
i1
dx
i+1
dx
n
= m(A).
Now suppose x
1
and x
2
S(A, e
i
)
x
1
= P
i
x
1
+y
1
e
i
, x
2
= P
i
x
2
+y
2
e
i
.
For x A dene
l(x) =sup{y : P
i
x+ye
i
A}.
g(x) =inf{y : P
i
x+ye
i
A}.
Then it is clear that
l(x
1
) g(x
1
) m(A
Pix1
) 2|y
1
|, (15.10)
l(x
2
) g(x
2
) m(A
Pix2
) 2|y
2
|. (15.11)
Claim: |y
1
y
2
| |l(x
1
) g(x
2
)| or |y
1
y
2
| |l(x
2
) g(x
1
)|.
Proof of Claim: If not,
2|y
1
y
2
| > |l(x
1
) g(x
2
)| +|l(x
2
) g(x
1
)|
|l(x
1
) g(x
1
) +l(x
2
) g(x
2
)|
= l(x
1
) g(x
1
) +l(x
2
) g(x
2
).
2 |y
1
| + 2 |y
2
|
by 15.10 and 15.11 contradicting the triangle inequality.
Now suppose |y
1
y
2
| |l(x
1
) g(x
2
)|. From the claim,
|x
1
x
2
| = (|P
i
x
1
P
i
x
2
|
2
+|y
1
y
2
|
2
)
1/2
(|P
i
x
1
P
i
x
2
|
2
+|l(x
1
) g(x
2
)|
2
)
1/2
(|P
i
x
1
P
i
x
2
|
2
+ (|z
1
z
2
| + 2)
2
)
1/2
diam(A) +O(
)
where z
1
and z
2
are such that P
i
x
1
+z
1
e
i
A, P
i
x
2
+z
2
e
i
A, and
|z
1
l(x
1
)| < and |z
2
g(x
2
)| < .
If |y
1
y
2
| |l(x
2
) g(x
1
)|, then we use the same argument but let
|z
1
g(x
1
)| < and |z
2
l(x
2
)| < ,
Since x
1
, x
2
are arbitrary elements of S(A, e
i
) and is arbitrary, this proves 15.9.
The next lemma says that if A is already symmetric with respect to the j
th
direction,
then this symmetry is not destroyed by taking S (A, e
i
).
Lemma 15.2.4 Suppose A is a Borel set in R
n
such that P
j
x +e
j
x
j
A if and
only if P
j
x+(x
j
)e
j
A. Then if i = j, P
j
x +e
j
x
j
S(A, e
i
) if and only if
P
j
x+(x
j
)e
j
S(A, e
i
).
Proof : By denition,
P
j
x +e
j
x
j
S(A, e
i
)
if and only if
|x
i
| < 2
1
m(A
Pi(Pjx+e
j
xj)
).
Now
x
i
A
Pi(Pjx+e
j
xj)
if and only if
x
i
A
Pi(Pjx+(xj)ej)
by the assumption on A which says that A is symmetric in the e
j
direction. Hence
P
j
x +e
j
x
j
S(A, e
i
)
if and only if
|x
i
| < 2
1
m(A
Pi(Pjx+(xj)ej)
)
if and only if
P
j
x+(x
j
)e
j
S(A, e
i
).
427
15.2.2 The Isodiametric Inequality
The next theorem is called the isodiametric inequality. It is the key result used to
compare Lebesgue and Hausdor measures.
Theorem 15.2.5 Let A be any Lebesgue measurable set in R
n
. Then
m
n
(A) (n)(r (A))
n
.
Proof: Suppose rst that A is Borel. Let A
1
= S(A, e
1
) and let A
k
= S(A
k1
, e
k
).
Then by the preceding lemmas, A
n
is a Borel set, diam(A
n
) diam(A), m
n
(A
n
) =
m
n
(A), and A
n
is symmetric. Thus x A
n
if and only if x A
n
. It follows that
A
n
B(0, r (A
n
)).
(If x A
n
\B(0, r (A
n
)), then x A
n
\B(0, r (A
n
)) and so diam(A
n
) 2|x| >diam(A
n
).)
Therefore,
m
n
(A
n
) (n)(r (A
n
))
n
(n)(r (A))
n
.
It remains to establish this inequality for arbitrary measurable sets. Letting A be such
a set, let {K
n
} be an increasing sequence of compact subsets of A such that
m(A) = lim
k
m(K
k
).
Then
m(A) = lim
k
m(K
k
) lim sup
k
(n)(r (K
k
))
n
(n)(r (A))
n
.
15.2.3 The Proper Value Of (n)
I will show that the proper determination of (n) is (n), the volume of the unit ball.
Since (n) has been adjusted such that k = 1, m
n
(B(0, 1)) = H
n
(B(0, 1)). There
exists a covering of B(0,1) of sets of radii less than , {C
i
}
i=1
such that
H
n
(B(0, 1)) + >
i
(n) r (C
i
)
n
Then by Theorem 15.2.5, the isodiametric inequality,
H
n
(B(0, 1)) + >
i
(n) r (C
i
)
n
=
(n)
(n)
i
(n) r
_
C
i
_
n
(n)
(n)
i
m
n
_
C
i
_
(n)
(n)
m
n
(B(0, 1)) =
(n)
(n)
H
n
(B(0, 1))
Now taking the limit as 0,
H
n
(B(0, 1)) +
(n)
(n)
H
n
(B(0, 1))
and since > 0 is arbitrary, this shows (n) (n).
By the Vitali covering theorem, there exists a sequence of disjoint balls, {B
i
} such
that B(0, 1) = (
i=1
B
i
) N where m
n
(N) = 0. Then H
n
(N) = 0 can be concluded

because H
n
H
n
and Lemma 15.1.7. Using m
n
(B(0, 1)) = H
n
(B(0, 1)) again,
H
n
(B(0, 1)) = H
n
(
i
B
i
)
i=1
(n) r (B
i
)
n
=
(n)
(n)
i=1
(n) r (B
i
)
n
=
(n)
(n)
i=1
m
n
(B
i
)
=
(n)
(n)
m
n
(
i
B
i
) =
(n)
(n)
m
n
(B(0, 1)) =
(n)
(n)
H
n
(B(0, 1))
which implies (n) (n) and so the two are equal. This proves that if (n) = (n) ,
then the H
n
= m
n
on the measurable sets of R
n
.
This gives another way to think of Lebesgue measure which is a particularly nice
way because it is coordinate free, depending only on the notion of distance.
For s < n, note that H
s
is not a Radon measure because it will not generally
be nite on compact sets. For example, let n = 2 and consider H
1
(L) where L is a
line segment joining (0, 0) to (1, 0). Then H
1
(L) is no smaller than H
1
(L) when L is
considered a subset of R
1
, n = 1. Thus by what was just shown, H
1
(L) 1. Hence
H
1
([0, 1] [0, 1]) = . The situation is this: L is a one-dimensional object inside R
2
and H
1
is giving a one-dimensional measure of this object. In fact, Hausdor measures
can make such heuristic remarks as these precise. Dene the Hausdor dimension of a
set, A, as
dim(A) = inf{s : H
s
(A) = 0}
15.2.4 A Formula For (n)
What is (n)? Recall the gamma function which makes sense for all p > 0.
(p)

0
e
t
t
p1
dt.
Lemma 15.2.6 The following identities hold.
p(p) = (p + 1),
(p)(q) =
_
1
0
x
p1
(1 x)
q1
dx
_
(p +q),
_
1
2
_
=
Proof: Using integration by parts,

(p + 1) =

0
e
t
t
p
dt = e
t
t
p
|
0
+p

0
e
t
t
p1
dt
= p(p)
429
Next
(p) (q) =

0
e
t
t
p1
dt

0
e
s
s
q1
ds
=

0
e
(t+s)
t
p1
s
q1
dtds
=

s
e
u
(u s)
p1
s
q1
duds
=
u
0
e
u
(u s)
p1
s
q1
dsdu
=
1
0
e
u
(u ux)
p1
(ux)
q1
udxdu
=
1
0
e
u
u
p+q1
(1 x)
p1
x
q1
dxdu
= (p +q)
_
1
0
x
p1
(1 x)
q1
dx
_
.
It remains to nd
_
1
2
_
.
_
1
2
_
=

0
e
t
t
1/2
dt =

0
e
u
2 1
u
2udu = 2

0
e
u
2
du
Now
_

0
e
x
2
dx
_
2
=

0
e
x
2
dx

0
e
y
2
dy =

0
e
(x
2
+y
2
)
dxdy
=
/2
0
e
r
2
rddr =
1
4
and so
_
1
2
_
= 2

0
e
u
2
du =

Next let n be a positive integer.
Theorem 15.2.7 (n) =
n/2
((n/2+1))
1
where (s) is the gamma function
(s) =

0
e
t
t
s1
dt.
Proof: First let n = 1.
(
3
2
) =
1
2
_
1
2
_
=
2
.
Thus
1/2
((1/2 + 1))
1
=
2
= 2 = (1) .
and this shows the theorem is true if n = 1.
Assume the theorem is true for n and let B
n+1
be the unit ball in R
n+1
. Then by
the result in R
n
,
m
n+1
(B
n+1
) =
1
1
(n)(1 x
2
n+1
)
n/2
dx
n+1
= 2(n)
1
0
(1 t
2
)
n/2
dt.
Doing an integration by parts and using Lemma 15.2.6
= 2(n)n
1
0
t
2
(1 t
2
)
(n2)/2
dt
= 2(n)n
1
2
1
0
u
1/2
(1 u)
n/21
du
= n(n)
1
0
u
3/21
(1 u)
n/21
du
= n(n)(3/2)(n/2)(((n + 3)/2))
1
= n
n/2
((n/2 + 1))
1
(((n + 3)/2))
1
(3/2)(n/2)
= n
n/2
((n/2)(n/2))
1
(((n + 1)/2 + 1))
1
(3/2)(n/2)
= 2
n/2
(3/2)(((n + 1)/2 + 1))
1
=
(n+1)/2
(((n + 1)/2 + 1))
1
.
From now on, in the denition of Hausdor measure, it will always be the case that
(s) = (s) . As shown above, this is the right thing to have (s) to equal if s is a
positive integer because this yields the important result that Hausdor measure is the
same as Lebesgue measure. Note the formula,
s/2
((s/2 + 1))
1
makes sense for any
s 0.
15.3 Hausdor Measure And Linear Transformations
Hausdor measure makes possible a unied development of n dimensional area including
in one theory length and surface area. Imagine the boundary of an open set in R
3
. You
would tend to think of this as something two dimensional. The way to measure it is
with H
2
. Length can be measured by H
1
and the boundary of an open set in R
4
is
measured in terms of H
3
etc.
As in the case of Lebesgue measure, the rst step in this is to understand basic
considerations related to linear transformations. Recall that for L L
_
R
k
, R
l
_
, L
is
dened by
(Lu, v) = (u, L
v) .
Also recall the right polar decomposition, Theorem 3.9.3 on Page 66. This theorem says
you can write a linear transformation as the composition of two linear transformations,
one which preserves length and the other which distorts, the right polar decomposition.
The one which distorts is the one which will have a nontrivial interaction with Hausdor
measure while the one which preserves lengths does not change Hausdor measure.
These ideas are behind the following theorems and lemmas.
Lemma 15.3.1 Let R L(R
n
, R
m
), n m, and R
R = I. Then if A R
n
,
H
n
(RA) = H
n
(A).
In fact, if P : R
n
R
m
satises |Px Py| = |x y| , then
H
n
(PA) = H
n
(A) .
15.3. HAUSDORFF MEASURE AND LINEAR TRANSFORMATIONS 431
Proof: Note that
|R(x y)|
2
=(R(x y) , R(x y)) = (R
R(x y) , x y) = |x y|
2
Thus R preserves lengths.
Now let P be an arbitrary mapping which preserves lengths and let A be bounded,
P(A)
j=1
C
j
, r(C
j
) < , and
H
n
(PA) + >
j=1
(n)(r(C
j
))
n
.
Since P preserves lengths, it follows P is one to one on P (R
n
) and P
1
also preserves
lengths on P (R
n
) . Replacing each C
j
with C
j
(PA),
H
n
(PA) + >
j=1
(n)r(C
j
(PA))
n
=
j=1
(n)r
_
P
1
(C
j
(PA))
_
n
H
n
(A).
Thus H
n
(PA) H
n
(A).
Now let A
j=1
C
j
, diam(C
j
) , and
H
n
(A) +
j=1
(n) (r (C
j
))
n
Then
H
n
(A) +
j=1
(n) (r (C
j
))
n
=
j=1
(n) (r (PC
j
))
n
H
n
(PA).
Hence H
n
(PA) = H
n
(A). Letting 0 yields the desired conclusion in the case where

A is bounded. For the general case, let A
r
= A B(0, r). Then H
n
(PA
r
) = H
n
(A
r
).
Now let r . This proves the lemma.
Lemma 15.3.2 Let F L(R
n
, R
m
), n m, and let F = RU where R and U are
described in Theorem 3.9.3 on Page 66. Then if A R
n
is Lebesgue measurable,
H
n
(FA) = det(U)m
n
(A).
Proof: Using Theorem 9.8.7 on Page 235 and Theorem 15.1.9,
H
n
(FA) = H
n
(RUA)
= H
n
(UA) = m
n
(UA) = det(U)m
n
(A).
Denition 15.3.3 Dene J to equal det(U). Thus
J = det((F
F)
1/2
) = (det(F
F))
1/2
.
15.4 The Area Formula
15.4.1 Preliminary Results
It was shown in Lemma 15.3.2 that for F L(R
n
, R
m
) , m n
H
n
(FA) = det(U)m
n
(A)
where F = RU with R preserving distances and U L(R
n
, R
n
) having all positive
eigenvalues. The area formula gives a generalization of this simple relationship to the
case where F is replaced by a nonlinear mapping, h. It contains as a special case the
earlier change of variables formula. The area formula has to do with n dimensional
measure on a set in R
m
where m > n. Thus it includes notions of length and area of
curves or surfaces in higher dimensional space. For example, you can use these ideas to
consider the two dimensional area of a surface in R
3
or even in R
8
and it can all be done
in a unied and rational way. In addition, the area formula will not require integration
over open sets. Measurable sets are good enough.
Assume m n and h maps an open set in R
n
to R
m
. Also suppose
Dh(x) exists for all x V, (15.12)
Lemma 15.4.1 If T V where V is an open set in R
n
and m
n
(T) = 0, then
H
n
(h(T)) = 0. If h is one to one, V and h(V ) are both bounded, N h(V ) , H
n
(N) =
0, and for Dh(x) = R(x) U (x) the right polar decomposition, U (x)
1
exists for all
x V, then m
n
_
h
1
(N)
_
= 0.
Proof: Let
T
k
{x T : ||Dh(x)|| < k} .
Thus T =
k
T
k
. I will show h(T
k
) has H
n
measure zero and then it will follow that
h(T) =
k=1
h(T
k
)
must also have measure zero.
Let > 0 be given. By outer regularity, there exists an open set, W, containing
T
k
which is contained in V such that m
n
(W) <

k
n
6
n
. For x T
k
it follows from
dierentiability,
h(x +v) = h(x) +Dh(x) v +o (v)
and so whenever r
x
is small enough, B(x,5r
x
) W and whenever |v| < 5r
x
, |o (v)| <
kr
x
. Therefore, if |v| < 5r
x
,
Dh(x) v +o (v) B(0, 5kr
x
) +B(0,kr
x
) B(0, 6kr
x
)
and so
h(B(x, 5r
x
)) B(h(x) , 6kr
x
).
Letting > 0 be given, the Vitali covering theorem implies there exists a sequence of
disjoint balls {B
i
}, B
i
= B(x
i
, r
xi
), which are contained in W such that the sequence
of enlarged balls,
_
B
i
_
, having the same center but 5 times the radius, covers T
k
and
6kr
xi
< . Then
H
n
(h(T
k
)) H
n
_
h
_
i=1
B
i
__
i=1
H
n
_
h
_
B
i
__
i=1
H
n
(B(h(x
i
) , 6kr
xi
))
15.4. THE AREA FORMULA 433
i=1
(n) (6kr
xi
)
n
= (6k)
n
i=1
(n) r
n
xi
= (6k)
n
i=1
m
n
(B(x
i
, r
xi
))
(6k)
n
m
n
(W) (6k)
n

k
n
6
n
= .
Since > 0 is arbitrary, this shows H
n
(h(T
k
)) = 0. Since is arbitrary, this implies
H
n
(h(T
k
)) = 0. Now
H
n
(h(T)) = lim
k
H
n
(h(T
k
)) = 0.
It remains to verify the last claim. Recall
U (x) =
n
k=1
a
k
v
k
v
k
where {v
k
} is an orthonormal basis and since U (x)
1
is given to exist, each a
k
> 0.
Therefore, (U (x) w, w) (x) |w|
2
where (x) > 0.
Next I claim h
1
is continuous on h(V ) . Suppose then that h(x
k
) h(x) . Then
let the sequentially compact set B(x, r) V. Without loss of generality all x
k
may be
assumed to lie in B(x, r). If {x
k
} fails to converge to x, then since B(x, r) is sequentially
compact, there exists a subsequence {x
k
l
} converging to z = x. But then
h(x) = lim
l
h(x
k
l
) = h(z)
a contradiction to h being one to one. Thus x
k
x and so h is continuous.
For > 0 let
N

_
y N :
_
h
1
(y)
_

_
.
Then for y N
,
(y +v) y = h
_
h
1
(y +v)
_
h
_
h
1
(y)
_
= R
_
h
1
(y)
_
U
_
h
1
(y)
_ _
h
1
(y +v) h
1
(y)
_
+o
_
h
1
(y +v) h
1
(y)
_
therefore
R
_
h
1
(y)
_
v =U
_
h
1
(y)
_ _
h
1
(y +v) h
1
(y)
_
+o
_
h
1
(y +v) h
1
(y)
_
(15.13)
Using continuity of h
1
, it follows that if v is small enough,
o
_
h
1
(y +v) h
1
(y)
_
h
1
(y +v) h
1
(y)
Taking the inner product of both sides of 15.13 with h

1
(y +v) h
1
(y) yields
_
h
1
(y)
_
v
h
1
(y +v) h
1
(y)
h
1
(y +v) h
1
(y)
h
1
(y +v) h
1
(y)
2
.
Now since R preserves distances and R
R = I
(R
v, R
v) = (v, RR
v) |v| |RR
v| = |v| |R
v|
and so
|R
v| |v| . (15.14)
Thus the above formula implies
|v|

2
h
1
(y +v) h
1
(y)
. (15.15)
Since N
has H
n
measure zero, there exist {C
k
} covering N
such that r (C
i
) < and
n
/4
n
>
k=1
(n) r (C
k
)
n
.
Without loss of generality each C
k
has nonempty intersection with N
, containing y
k
.
Now
_
h
1
(C
k
)
_
covers h
1
(N
) and from 15.15

diam
_
h
1
(C
k
)
_
2 diam(C
k
)
and so
H
n
4/
_
h
1
(N
)
_

k
(n)
_
2
_
n
(diam(C
k
)
n
)
4
n
k
(n) r (C
k
)
n
<

n
4
n
4
n
n
=
Since is arbitrary, H
n
4/
_
h
1
(N
)
_
= 0 and letting 0 yields
H
n
_
h
1
(N
)
_
= m
n
_
h
1
(N
)
_
= 0
because by Theorem 15.1.9 Lebesgue and Hausdor measure are the same on the
Lebesgue measurable sets of R
n
. Now take
k
0
m
n
_
h
1
(N)
_
= lim
k
m
n
_
h
1
_
N
k
__
= 0.
Lemma 15.4.2 If S is a Lebesgue measurable subset of an open set V R
n
, on
which h is dened and C
1
, then h(S) is H
n
measurable.
Proof: Let S
k
= S B(0, k) , k N. By inner regularity of Lebesgue measure,
there exists a set, F, which is the countable union of compact sets and a set T with
m
n
(T) = 0 such that
F T = S
k
.
Then h(F) h(S
k
) h(F) h(T). By continuity of h, h(F) is a countable union of
compact sets and so it is Borel. By Lemma 15.4.1, H
n
(h(T)) = 0 and so h(S
k
) is H
n
measurable because of completeness of Hausdor measure, which comes from H
n
being
obtained from an outer measure. Now h(S) =
k=1
h(S
k
) and so it is also true that
h(S) is H
n
measurable. This proves the lemma.
The following lemma, depending on the Brouwer xed point theorem and found
in Rudin [35], will be important for the following arguments. The idea is that if a
continuous function mapping a ball in R
k
to R
k
doesnt move any point very much,
then the image of the ball must contain a slightly smaller ball.
Lemma 15.4.3 Let B = B(0, r), a ball in R
k
and let F : B R
k
be continuous
and suppose for some < 1,
|F(v) v| < r (15.16)
for all v B. Then
F(B) B(0, r (1 )) .
Proof: Suppose a B(0, r (1 )) \ F(B) .
I claim that a = F(v) for all v B, not just for v B. Here is why. By the
assumption a / F(B), if F(v) = a, then |v| = r and so
|F(v) v| = |a v| |v| |a| > r r (1 ) = r,
a contradiction to 15.16.
Now letting G :B B, be dened by
G(v)
r (a F(v))
|a F(v)|
,
it follows from what was just shown G is continuous. Then by the Brouwer xed point
theorem, G(v) = v for some v B. Using the formula for G, it follows
|v| = |G(v)| =
r (a F(v))
|a F(v)|
= r
Taking the inner product with v,
(G(v) , v) = |v|
2
= r
2
=
r
|a F(v)|
(a F(v) , v)
=
r
|a F(v)|
(a v +v F(v) , v)
=
r
|a F(v)|
[(a v, v) +(v F(v) , v)]
=
r
|a F(v)|
_
(a, v) |v|
2
+(v F(v) , v)
_
r
|a F(v)|
_
r
2
(1 ) r
2
+r
2
= 0,
a contradiction to |v| = r. Therefore, B(0, r (1 )) \ F(B) = and This proves the
lemma.
Lemma 15.4.4 If |Px Py| L|x y| , then for E a set,
H
n
(PE) L
n
H
n
(E) .
Proof: Without loss of generality, assume H
n
(E) < . Let > 0 and let {C
i
}
i=1
be a covering of E such that r (C
i
) for each i and
i=1
(n) r (C
i
)
n
H
n
(E) +.
Then {PC
i
}
i=1
is a covering of PE such that r (PC
i
) L. Therefore,
H
n
L
(PE)
i=1
(n) r (PC
i
)
n
L
n
i=1
(n) r (C
i
)
n
L
n
H
n
(E) +L
n
L
n
H
n
(E) +L
n
.
Letting 0,
H
n
(PE) L
n
H
n
(E) +L
n
and since > 0 is arbitrary, This proves the lemma.

Then the following corollary follows from 15.14.
Corollary 15.4.5 Let R : R
n
R
m
where m n and R is linear and preserves
distances. Let T R
m
. Then
H
n
(T) H
n
(RR
T) = H
n
(R
T) .
Now let V be an open set in R
n
and let h : V R
m
be a C
1
(V ) function. For
Dh(x) = R(x) U (x) the right polar decomposition, suppose U (x)
1
exists for all
x U. By denition of the derivative,
h(x +v) h(x) = R(x) U (x) v +o(v)
= Dh(x) v +o(v) (15.17)
and therefore letting > 0 be given,
U
1
R
(h(x +v) h(x)) v
< |v|
whenever v is small enough. Thus by Lemma 15.4.3, if r
x
is small enough,
U
1
R
(h(x+B(0,r
x
)) h(x)) B(0, r
x
(1 ))
which implies
h(x+B(0,r
x
)) RUB(0, r
x
(1 )) +h(x)
or in other words,
h(B(x,r
x
)) Dh(x) B(0, r
x
(1 )) +h(x) . (15.18)
Referring to 15.17 again,
R
(x) (h(x +v) h(x)) = U (x) v +o(v)

= U (x)
_
v +U
1
(x) o(v)
_
= U (x) (v +o(v))
It follows that if r
x
is suciently small,
R
(x) (h(x+B(0, r
x
)) h(x)) U (B(0,r
x
) +B(0, r
x
))
UB(0,r
x
(1 +))
and so
h(x+B(0, r
x
)) =
h(B(x,r
x
)) Dh(x) B(0,r
x
(1 +)) +h(x) (15.19)
15.5 The Area Formula
The following lemma is the rst approximation to the area formula.
Lemma 15.5.1 Let V be a bounded open set in R
n
and let h C
1
(V ) , h : V R
m
for m n be one to one such that for
Dh(x) = R(x) U (x) ,
the right polar decomposition, U (x) is one to one. Assume also
|det U (x)| M, x V
Also assume H
n
(h(V )) < and h(V ) is bounded. Let W be a Borel set in R
m
and
let A be a Lebesgue measurable subset of V . Then
h(A)
X
W
(y) dH
n
=
A
X
W
(h(x)) |det (U (x))| dm
n
Proof: Recall
det (U (x)) = det
_
Dh(x)
Dh(x)
_
1/2
and so the function, x det (U (x)) is continuous.
Also let O be an open set containing h
1
(W)Asuch that m
n
_
O \
_
h
1
(W) A
__
<
and H
n
_
h
_
O \
_
h
1
(W) A
___
< . To do this, let {O
m
} be a decreasing sequence
of open sets containing h
1
(W) A such that m
n
(O
m
) m
n
_
h
1
(W) A
_
. Thus
m
n
_
(
m
O
m
) \ h
1
(W) A
_
= 0
and so
H
n
_
h
_
(
m
O
m
) \ h
1
(W) A
__
= 0
by Lemma 15.4.1. Therefore, if m is large enough, letting O = O
m
gives the desired
open set.
Let x h
1
(W) A. First note h
1
(W) is a Borel set because
S
_
E B (R
m
) : h
1
(E) B (R
n
)
_
is a algebra which contains the open sets due to the fact h is continuous. Therefore,
S = B (R
m
). Thus h
1
(W) A is measurable.
There exists 1 > r
x
> 0 small enough that 15.19 and 15.18 both hold. There exists
a possibly smaller r
x
such that
B(x, r
x
) O (15.20)
and
||det (U (x
1
))| |det (U (x))|| < (15.21)
whenever x
1
B(x, r
x
) .
The collection of such balls is a Vitali cover of h
1
(W)A. By Corollary 9.7.5 there
is a sequence of disjoint balls {B
i
} such that for
N h
1
(W) A\
i=1
B
i
,
m
n
(N) = 0. Therefore, renaming N to equal
h
1
(W) A\
i=1
B
i
h
1
(W) A,
m
n
(N) = 0 and by Lemma 15.4.1,
H
n
(h(N)) = 0.
h
1
(W) A =
_
i=1
B
i
A h
1
(W)
_
N
W h(A) = (
i=1
h(B
i
A) W) h(N) (15.22)
where m
n
(N) = H
n
(h(N)) = 0.
Denote by x
i
the center of B
i
and r
i
the radius. Using 15.18, Lemma 15.4.1 which
says h takes sets of Lebesgue measure zero to sets of H
n
measure zero, the translation
invariance of H
n
, Lemma 15.3.2 which gives the rule for taking a linear transformation
outside the Hausdor measure of something, 15.21, and the assumption that h is one
to one,
h(A)
X
W
(y) dH
n
=
h(A)W
dH
n
=
i=1
h(BiA)W
dH
n
=
h(
i=1
BiAh
1
(W))
dH
n
h(
i=1
Bi)
dH
n

i=1
H
n
(h(B
i
))
i=1
H
n
(Dh(x
i
) (B(0, (1 ) r
i
)))
=
i=1
|det (U (x
i
))| m
n
(B(0, (1 ) r
i
))
= (1 )
n
i=1
|det (U (x
i
))| m
n
(B(x
i
,r
i
))
(1 )
n
i=1
_
Bi
|det (U (x))| dm
n
m
n
(B
i
)
_
(1 )
n
i=1
BiAh
1
(W)
|det (U (x))| dm
n
(1 )
n
m
n
(V )
= (1 )
n
V
X
i=1
h(BiAh
1
(W))
(h(x)) |det (U (x))| dm
n
(1 )
n
m
n
(V )
= (1 )
n
V
X
Wh(A)
(h(x)) |det (U (x))| dm
n
(1 )
n
m
n
(V )
(1 )
n
V
X
N
(x) |det (U (x))| dm
n
= (1 )
n
A
X
W
(h(x)) |det (U (x))| dm
n
(1 )
n
m
n
(V )
The last three lines follows from 15.22. Recall m
n
(N) = 0. Since > 0 is arbitrary,
this shows
h(A)
X
W
(y) dH
n
A
X
W
n
The opposite inequality can be established in exactly the same way using 15.19
instead of 15.18 and turning all the inequalities around featuring (1 +) instead of
(1 ) , much as was done in the proof of Lemma 9.9.1. Thus
h(A)
X
W
(y) dH
n
=
h(A)W
dH
n
=
i=1
h(BiA)W
dH
n
=
i=1
H
n
(h(B
i
A) W)
i=1
H
n
(h(B
i
))
i=1
H
n
(Dh(x
i
) (B(0, (1 +) r
i
)))
=
i=1
|det (U (x
i
))| m
n
(B(0, (1 +) r
i
))
= (1 +)
n
i=1
|det (U (x
i
))| m
n
(B(x
i
,r
i
))
(1 +)
n
i=1
_
Bi
|det (U (x))| dm
n
+m
n
(B
i
)
_
(1 +)
n
i=1
BiAh
1
(W)
|det (U (x))| dm
n
+ (1 +)
n
m
n
(V )
+(1 +)
n
O\(Ah
1
(W))
|det (U (x))| dm
n
(1 +)
n
V
X
i=1
h(BiA)W
(h(x)) |det (U (x))| dm
n
+(1 +)
n
m
n
(V ) + (1 +)
n
M
= (1 +)
n
V
X
Wh(A)
(h(x)) |det (U (x))| dm
n
+ (1 +)
n
m
n
(V )
+ (1 +)
n
M (1 +)
n
V
X
N
(x) |det (U (x))| dm
n
= (1 +)
n
A
X
W
(h(x)) |det (U (x))| dm
n
+ (1 +)
n
m
n
(V ) + (1 +)
n
M
Since is arbitrary, This proves the lemma.
Next the Borel sets will be enlarged to H
n
measurable sets.
Lemma 15.5.2 Let V be a bounded open set in R
n
and let h C
1
(V ) , h : V R
m
for m n be one to one such that for
Dh(x) = R(x) U (x) ,
the right polar decomposition, U (x) is one to one. Assume also
|det U (x)| M, x V
Also assume H
n
(h(V )) < and h(V ) is bounded. Let W be a H
n
measurable set in
R
m
and let A be a Lebesgue measurable subset of V . Then
h(A)
X
W
(y) dH
n
=
A
X
W
(h(x)) |det (U (x))| dm
n
(15.23)
Proof: By Theorem 15.1.5 there exists a Borel set, F containing W such that
H
n
(F \ W) = 0. By Lemma 15.4.1 m
n
_
h
1
((F \ W) h(A))
_
= 0. Therefore, from
Lemma 15.5.1
h(A)
X
W
(y) dH
n
=
h(A)
X
F
(y) dH
n
=
A
X
F
(h(x)) |det (U (x))| dm
n
=
A
X
F\W
(h(x)) |det (U (x))| dm
n
+
A
X
W
(h(x)) |det (U (x))| dm
n
=
A
X
W
(h(x)) |det (U (x))| dm
n
.
Note that
X
W
(h(x)) |det (U (x))| +X
F\W
(h(x)) |det (U (x))| = X
F
(h(x)) |det (U (x))|
and the second term on the left equals X
h
1
(F\W)
(x) and is measurable because h
1
(F \ W)
has Lebesgue measure zero by Lemma 15.4.1 while the term on the right is Lebesgue
measurable because F is Borel so X
F
h is measurable because it is a Borel measurable
function composed with a continuous function (why?). Therefore, x X
W
(h(x)) |det (U (x))|
is also measurable. This proves the lemma.
You dont need to assume the open sets are bounded and you dont need to assume
a bound on |det U (x)|.
Corollary 15.5.3 Let V be an open set in R
n
and let h C
1
(V ) be one to one and
also for Dh(x) = R(x) U (x) the right polar decomposition, U (x) is one to one. Let E
be H
n
measurable and let A V be Lebesgue measurable. Then
h(A)
X
E
(y) dH
n
=
A
X
E
(h(x)) |det (U (x))| dm
n
.
Proof: For each x A, there exists r
x
such that B(x, r
x
) V and r
x
< 1. Then by
the mean value inequality Theorem 6.4.2 and the observation that ||Dh(x)|| is bounded
on the compact set B(x, r
x
), it follows h(B(x, r
x
)) is also bounded. Also |det U (x)| is
bounded on the compact set B(x, r
x
). These balls are a Vitali cover of A. By Corollary
9.7.5 there is a sequence of these disjoint balls {B
i
} such that m
n
(A\
i=1
B
i
) = 0 and
so
A =
i=1
(B
i
A) N
where N = A\
i=1
B
i
has measure zero.
It follows from Lemma 15.4.1 that h(N) also has measure zero. Then from Lemma
15.5.2 applied to B
i
,
h(A)
X
E
(y) dH
n
=
h(BiA)
X
Eh(BiA)
(y) dH
n
=
BiA
X
E
(h(x)) |det (U (x))| dm
n
=
A
X
E
(h(x)) |det (U (x))| dm
n
.
With this corollary, the main theorem follows.
Theorem 15.5.4 Let V be an open set in R
n
and A is a Lebesgue measurable
subset of V. Let h C
1
(V ) be one to one and also for Dh(x) = R(x) U (x) the right
polar decomposition, U (x) is one to one. Then if g is any nonnegative H
n
measurable
function,
h(A)
g (y) dH
n
=
A
g (h(x)) |det (U (x))| dm
n
. (15.24)
Proof: From Corollary 15.5.3, 15.24 holds for any nonnegative simple function in
place of g. In general, let {s
k
} be an increasing sequence of simple functions which
converges to g pointwise. Then from the monotone convergence theorem
h(A)
g (y) dH
n
= lim
k
h(A)
s
k
dH
n
= lim
k
A
s
k
(h(x)) |det (U (x))| dm
n
=
A
n
.
Of course this theorem implies the following corollary obtained by splitting the
function into the positive and negative parts of the real and imaginary parts.
n
and A is a Lebesgue measurable subset
of V. Let h C
1
(V ) be one to one and also for Dh(x) = R(x) U (x) the right polar
decomposition, U (x) is one to one. Then if g is any function in L
1
(h(V )),
h(A)
g (y) dH
n
=
A
n
.
You dont need to assume U (x) is one to one. The following lemma is like Sards
lemma presented earlier. However, it might seem a little easier and if so, it is because
the area formula above is available and Hausdor measures are in some ways easier to
work with since they only depend on distance.
Lemma 15.5.6 Let V be an open set in R
n
and let h : V R
m
be a C
1
function
such that the right polar decomosition of the derivative is Dh(x) = R(x) U (x). Let E
denote the set
E
_
x V : U (x)
1
does not exist
_
Then H
n
(h(E)) = 0.
Proof: Recall the notation J (x) |det (U (x))| discussed earlier. Modify it slightly
as
Jh(x) |det (U (x))|
where Dh(x) = R(x) U (x). First suppose V is bounded and so is ||Dh(x)||. Dene
k
: R
n
R
m
R
n
k
(x)
_
h(x)
x
_
Thus
k
(x +v) k
(x) =
_
h(x +v) h(x)
v
_
=
_
Dh(x) v
v
_
+o(v)
and so
Dk
(x) =
_
Dh(x)
id
_
L(R
n
, R
m
R
n
)
It is left as an exercise to explain how
Dk
(x)
=
_
Dh(x)
id
_
L(R
m
R
n
, R
n
)
and
Jk
(x)
2
det
_
Dk
(x)
Dk
(x)
_
= det
_
Dh
(x) Dh(x) +
2
id
_
Now there is an orthonormal basis, {v
k
} for R
n
such that
Dh
(x) Dh(x) =
n
k=1
a
k
v
k
v
k
,
2
id =
2
n
k=1
v
k
v
k
.
where each a
k
0 and on E, at least one equals 0. Thus the determinant above is the
determinant of a diagonal matrix which has all positive entries on the diagonal but at
least one of them is
2
. Since ||Dh(x)|| is bounded, this shows there exists a constant
C independent of x V such that on E, Jk
(x) C. Now by the earlier area formula,

Theorem 15.5.4,
k(V )
X
k(E)
(z) dH
n
=
V
X
k(E)
(k
(x)) Jk
(x) dm
n
=
V
X
E
(x) Jk
(x) dm
n
Thus
H
n
(k
(E)) Cm
n
(E)
However, |h(x) h(x
1
)| |k
(x) k
(x
1
)| and so by Lemma 15.4.4 H
n
(k
(E))
H
n
(h(E)) . Let P
_
y x
_
y. Thus
H
n
(h(E)) Cm
n
(E)
and since is arbitrary, this shows H
n
(h(E)) = 0.
In the general case when V might not be bounded dene for each k N suciently
large that the sets are nonempty
V
k
B(0,k)
_
x V : dist
_
x,V
C
_
> 1/k
_
Then V
k
is compact and if E
k
E V
k
and V
k
plays the role of V above, it fol-
lows ||Dh(x)|| is bounded on V
k
and H
n
(h(E
k
)) = 0. Now let k to conclude
H
n
(h(E)) = 0. This proves the lemma.
Now with this lemma it is easy to give a fairly general version of the area formula.
Theorem 15.5.7 Let V be an open set in R
n
and A is a Lebesgue measurable
subset of V . Let h C
1
(V ) be one to one. Then if g is any nonnegative H
n
measurable
function,
h(V )
g (y) dH
n
=
V
n
. (15.25)
Proof: Let E =
_
x V : U (x)
1
does not exist
_
which is the same as the set
where det U (x) = 0. Then by Lemma 15.5.6 H
n
(h(E)) = 0 and so
h(V )
g (y) dH
n
=
h(V )
X
h(V \E)
(y) g (y) dH
n
=
V
X
h(V \E)
(h(x)) g (h(x)) |det (U (x))| dm
n
=
V
X
V \E
(x) g (h(x)) |det (U (x))| dm
n
=
V
n
Note that if m = n, this reduces to the usual change of variables formula because
H
n
= m
n
on R
n
as was shown above.
As before, there is an obvious corollary obtained by splitting up the function in L
1
into positive and negative parts of real and imaginary parts, noting the desired result
holds for each of these pieces and then adding them together.
n
and A is a Lebesgue measurable subset
of V and let h C
1
(V ) be one to one. Then if g is any function in L
1
(h(A)),
h(A)
g (y) dH
n
=
A
n
.
15.6 Area Formula For Mappings Which Are Not
One To One
Now suppose h is only C
1
, not necessarily one to one. For
V
+
{x V : |det U (x)| > 0}
and Z the set where |det U (x)| = 0, Lemma 15.5.6 implies H
n
(h(Z)) = 0.
Now the following lemma is very interesting for its own sake.
15.6. AREA FORMULA FOR MAPPINGS WHICH ARE NOT ONE TO ONE 443
Lemma 15.6.1 For x V
+
, there exists an open ball B
x
V
+
such that h is one
to one on B
x
.
Proof: Let Dh(x) = R(x) U (x) be the right polar decomposition. Recall that
U (x) is self adjoint and satises U (x) v v |v|
2
for some > 0 where is the
smallest eigenvalue of U (x) , the square root of the smallest eigenvalue of U (x)
2
=
Dh(x)
Dh(x) . Let r > 0 such that B(x, r) V

+
. Then for y B(x, r), the continuity
of y U (y)
2
, resulting for the assumption that h is C
1
, implies
U (y)
2
v v = U (x)
2
v v +
_
U (y)
2
U (x)
2
_
v v

2
|v|
2

2
2
|v|
2
=

2
2
|v|
2
provided r is suciently small. Thus for small enough r, the eigenvalues of U (y)
2
for
y B(x, r) are at least as large as
2
/2 and so the eigenvalues of U (y) for these values
of y are lat least as large as /
2. Thus for y B(x, r),

U (y) v v

2
|v|
2
.
Now make r still smaller if necessary such that for y, z B(x, r),
||Dh(y) Dh(z)|| <

2
.
Then for any y, z of this sort,
|h(z) h(y) Dh(y) (z y)| <

2
|z y| . (15.26)
This follows from the mean value inequality, Theorem 6.4.2 because if you dene for
such a xed y B(x, r)
F(z) h(z) h(y) Dh(y) (z y) ,
it follows F(y) = 0 and DF(z) = Dh(z) Dh(y) .
Then for y, z B(x, r),
R
(y) (h(z) h(y)) = U (y) (z y) +R
o(z y) (15.27)
where o(z y) is of the form
h(z) h(y) Dh(y) (z y)
Then from 15.26,
|R
o(z y)| |o(z y)|

2
|z y| .
If h(z) = h(y) , then taking the inner product of both sides of 15.27 with z y,
0 (U (y) (z y) , (z y))

2
|z y|
2

2
_
|z y|
2
showing z = y. This proves the lemma.
Let {B
i
} be a countable subset of {B
x
}
xV+
such that V
+
=
i=1
B
i
. Let E
1
= B
1
.
If E
1
, , E
k
have been chosen, E
k+1
= B
k+1
\
k
i=1
E
i
. Thus
i=1
E
i
= V
+
, h is one to one on E
i
, E
i
E
j
= ,
and each E
i
is a Borel set contained in the open set B
i
. Now dene
n(y)
i=1
X
h(EiA)
(y) +X
h(Z)
(y).
Thus
i=1
X
h(EiA)
(y) = X
A+
where A
+
V
+
A. The set, h(E
i
) , h(Z) are H
n
measurable by Lemma 15.4.2. Thus
n() is H
n
measurable.
Lemma 15.6.2 Let V be an open set, F h(V ) be H
n
measurable and let A be a
Lebesgue measurable subset of V . Then
h(A)
n(y)X
F
(y)dm
n
=
A
X
F
n
.
Proof: Using Lemma 15.5.6 and the Monotone Convergence Theorem
h(A)
n(y)X
F
(y)dH
n
=
h(A)
_
_
_
i=1
X
h(Ei)
(y) +
H
n
(h(Z))=0
..
X
h(Z)
(y)
_
_
_X
F
(y)dH
n
=
i=1
h(A)
X
h(Ei)
(y)X
F
(y)dH
n
=
i=1
h(BiA)
X
h(EiA)
(y)X
F
(y)dH
n
=
i=1
BiA
X
EiA
(x)X
F
(h(x))| det U (x))|dm
n
=
i=1
A
X
EiA
(x)X
F
(h(x))| det U (x) |dm
n
=
i=1
X
EiA
(x)X
F
n
=
A+
X
F
n
=
A
X
F
n
.
The integrand in the integrand on the left was shown to be Lebesgue measurable
from the above argument. Therefore, the integrand of the integral on the right is also
Lebesgue measurable because it equals
X
F
(h(x))| det U (x) |X
A+
+ 0X
Z
(x)
and both functions in the sum are measurable. This proves the lemma.
Denition 15.6.3 For y h(A), dene a function, #, according to the for-
mula
#(y) number of elements in h
1
(y).
Observe that
#(y) = n(y) H
n
a.e. (15.28)
because n(y) = #(y) if y / h(Z), a set of H
n
measure 0. Therefore, # is a measurable
function because of completeness of H
n
.
15.7. THE COAREA FORMULA 445
Theorem 15.6.4 Let g 0, g is H
n
measurable, and let h be C
1
(V ) and A is
a Lebesgue measurable subset of V . Then
h(A)
#(y)g(y)dm
n
=
A
g(h(x))| det Dh(x)|dm
n
. (15.29)
The integrand on the right is Lebesgue measurable.
Proof: From 15.28 and Lemma 15.6.2, 15.29 holds for all g, a nonnegative simple
function. Approximating an arbitrary measurable nonnegative function, g, with an
increasing pointwise convergent sequence of simple functions and using the monotone
convergence theorem, yields 15.29 for an arbitrary nonnegative measurable function, g.
15.7 The Coarea Formula
In the coarea formula, h maps V R
n
to R
m
where m n. It is possible to obtain
this formula from the area formula and some interesting linear algebra.
Lemma 15.7.1 Let A L(R
n
, R
m
) . Then the nonzero eigenvalues of AA
and A
A
are the same and occur with the same algebraic multiplicities.
Proof: This follows from Theorem 3.6.5 on Page 51 applied to the matrices of A
and A
.
Corollary 15.7.2 Let A L(R
n
, R
m
) . Then
det (id +A
A) = det (id +AA
) .
Proof: This follows from Lemma because
id +A
A =
r
k=1
(1 +a
k
) w
k
w
k
+
n
k=r+1
1w
k
w
k
while id +AA
is of the form
r
k=1
(1 +a
k
) v
k
v
k
+
m
k=r+1
1v
k
v
k
.
Therefore, both of these determinants equal
r
k=1
(1 +a
k
) .
Next is a lemma which leads to some conclusions about measurability.
Lemma 15.7.3 Let h : V R
p
R
m
be continuous and > 0. Then if A R
p
is
either open or compact,
y H
s
_
A h
1
(y)
_
is Borel measurable.
Proof: Suppose rst that A is compact and suppose for > 0,
H
s
_
A h
1
(y)
_
< t
Then there exist sets S
i
, satisfying
r (S
i
) < , A h
1
(y)
i=1
S
i
,
and
i=1
(s) (r (S
i
))
s
< t.
I claim these sets can be taken to be open sets. Choose > 1 but close enough to 1
that
i=1
(s) (r (S
i
))
s
< t
Replace S
i
with S
i
+B(0,
i
) where
i
is small enough that
diam(S
i
) + 2
i
< diam(S
i
) .
Then
diam(S
i
+B(0,
i
)) diam(S
i
)
and so r (S
i
+B(0,
i
)) r (S
i
) . Thus
i=1
(s) r (S
i
+B(0,
i
))
s
< t.
Hence you could replace S
i
with S
i
+ B(0,
i
) and so one can assume the sets S
i
are
open.
Claim: If z is close enough to y, then A h
1
(z)
i=1
S
i
.
Proof: If not, then there exists a sequence {z
k
} such that
z
k
y,
and
x
k
(A h
1
(z
k
)) \
i=1
S
i
.
By compactness of A, there exists a subsequence still denoted by k such that
z
k
y, x
k
x A\
i=1
S
i
.
Hence
h(x) = lim
k
h(x
k
) = lim
k
z
k
= y.
But x /
i=1
S
i
contrary to the assumption that A h
1
(y)
i=1
S
i
.
It follows from this claim that whenever z is close enough to y,
H
s
_
A h
1
(z)
_
< t.
This shows
_
z R
p
: H
s
_
A h
1
(z)
_
< t
_
is an open set and so y H
s
_
A h
1
(y)
_
is Borel measurable whenever A is compact.
Now let V be an open set and let
A
k
V, A
k
compact.
Then
H
s
_
V h
1
(y)
_
= lim
k
H
s
_
A
k
h
1
(y)
_
so y H
s
_
V h
1
(y)
_
is Borel measurable for all V open. This proves the lemma.
Lemma 15.7.4 Let h : V R
p
R
m
be Lipschitz continuous, satisfying an in-
equality of the form
|h(x) h(y)| Lip (h) |x y|
where Lip (h) is a positive constant. Suppose A is either open or compact in R
p
. Then
y H
s
_
A h
1
(y)
_
is also Borel measurable and
R
m
H
s
_
A h
1
(y)
_
dm
m
2
m
(Lip (h))
m
(s) (m)
(s +m)
H
s+m
(A)
In particular, if s = n m and p = n
R
m
H
nm
_
A h
1
(y)
_
dm
m
2
m
(Lip (h))
m
(n m) (m)
(n)
m
n
(A) (15.30)
Proof: From Lemma 15.7.3 y H
s
_
A h
1
(y)
_
is Borel measurable for each
> 0. Without loss of generality, H
s+m
(A) < . Now let C
i
be closed sets with
r (C
i
) < , A
i=1
C
i
, and
H
s+m
(A) + >
i=1
(s +m) r (C
i
)
s+m
.
Note each C
i
is compact so y H
s
_
C
i
h
1
(y)
_
is Borel measurable. Thus
R
m
H
s
_
A h
1
(y)
_
dm
m
R
m
i
H
s
_
C
i
h
1
(y)
_
dm
m
=
R
m
H
s
_
C
i
h
1
(y)
_
dm
m
h(Ci)
H
s
(C
i
) dm
m
=
i
m
m
(h(C
i
)) H
s
(C
i
)
Now h(C
i
) is contained in a ball of radius 2r (C
i
) and so
i
(Lip (h))
m
2
m
(m) r (C
i
)
m
(s) r (C
i
)
s
= (Lip (h))
m
(m) (s)
(m+s)
2
m
i
(s +m) r (C
i
)
m+s
(Lip (h))
m
(m) (s)
(m+s)
2
m
_
H
s+m
(A) +
_
Since > 0 is arbitrary,
R
m
H
s
_
A h
1
(y)
_
dm
m
(Lip (h))
m
(m) (s)
(m+s)
2
m
H
s+m
(A)
Taking a limit as 0 This proves the lemma.
Next I will show that whenever A is Lebesgue measurable,
y H
nm
_
A h
1
(y)
_
is m
m
measurable and the above estimate holds.
Lemma 15.7.5 Let A be a Lebesgue measurable subset of V an open set and let
h : V R
n
R
m
be Lipschitz continuous,
|h(x) h(y)| Lip (h) |x y| .
Then
y H
nm
_
A h
1
(y)
_
is Lebesgue measurable. Furthermore
R
m
H
nm
_
A h
1
(y)
_
dm
m
2
m
(Lip (h))
m
(n m) (m)
(n)
m
n
(A)
Proof: Let A be a bounded Lebesgue measurable set in R
n
. Then by inner and
outer regularity of Lebesgue measure there exists an increasing sequence of compact
sets {K
k
} contained in A and a decreasing sequence of open sets {V
k
} containing A
such that m
n
(V
k
\ K
k
) < 2
k
. Thus m
n
(V
1
) m
n
(A) + 1. By Lemma 15.7.4
R
m
H
nm
_
V
1
h
1
(y)
_
dm
m
< 2
m
(Lip (h))
m
(n m) (m)
(n)
(m
n
(A) + 1) .
Also
H
nm
_
K
k
h
1
(y)
_
H
nm
_
A h
1
(y)
_
H
nm
_
V
k
h
1
(y)
_
(15.31)
By Lemma 15.7.4
=
R
m
_
H
nm
_
V
k
h
1
(y)
_
H
nm
_
K
k
h
1
(y)
__
dm
m
=
R
m
H
nm
_
(V
k
K
k
) h
1
(y)
_
dm
m
2
m
(Lip (h))
m
(n m) (m)
(n)
m
n
(V
k
\ K
k
)
< 2
m
(Lip (h))
m
(n m) (m)
(n)
2
k
Let the Borel measurable functions, g and f be dened by
g (y) lim
k
H
nm
_
V
k
h
1
(y)
_
, f (y) lim
k
H
nm
_
K
k
h
1
(y)
_
It follows from the dominated convergence theorem using H
nm
_
V
1
h
1
(y)
_
as a
dominating function and 15.31 that
f (y) H
nm
_
A h
1
(y)
_
g (y)
and
R
m
(g (y) f (y)) dm
m
= 0.
By completness of m
m
, this establishes y H
nm
_
A h
1
(y)
_
is Lebesgue measur-
able. Then by Lemma 15.7.4 again,
R
m
H
nm
_
A h
1
(y)
_
dm
m
2
m
(Lip (h))
m
(n m) (m)
(n)
m
n
(A) .
Letting 0 and using the monotone convergence theorem yields the desired inequality
for H
nm
_
A h
1
(y)
_
.
The case where A is not bounded can be handled by considering A
r
= A B(0, r)
and letting r . This proves the lemma.
By fussing with the isodiametric inequality one can remove the factor of 2
m
in the
above inequalities obtaining more attractive formulas. This is done in [13]. See also
[28] which follows [13] and [16]. This last reference probably has the most complete
treatment of these topics.
With these lemmas, it is now possible to give a proof of the coarea formula.
Dene (n, m) as all possible ordered lists of m numbers taken from {1, 2, , n} .
Lemma 15.7.6 Let A be a Lebesgue measurable set in V and let h : V R
n
R
m
where m n is C
1
and also a Lipschitz map and for which
Jh(x) det
_
Dh(x) Dh(x)
_
1/2
= 0.
Then the following formula holds along with all measurability assertions needed for it to
make sense.

R
m
H
nm
_
A h
1
(y)
_
dy =
A
Jh(x) dx (15.32)
Proof: For x R
n
, and i (n, m), with i =(i
1
, , i
m
), dene x
i
(x
i1
, ,
x
im
), and
i
x x
i
. Also for i (n, m), let i
c
(n, n m) consist of the re-
maining indices taken in order. For h : R
n
R
m
where m n, dene Jh(x)
det
_
Dh(x) Dh(x)
_
1/2
. For each i (n, m), dene h
i
: R
n
R
m
R
nm
by
h
i
(x)
_
h(x)
x
ic
_
.
As in Lemma 15.6.1, since h
i
is in C
1
(V ) there exist Borel sets
_
F
i
k
_
which are disjoint,
F
i
k
B
i
k
, h
i
is one to one on the open ball B
i
k
with a C
1
inverse dened on h
i
_
B
i
k
_
and
k
F
i
k
=
_
x R
n
: det Dh
i
(x) = 0
_
.
By assumption, for each x V there exists i such that det (D
x
i
h(x)) = 0 which implies
det Dh
i
(x) = 0. This follows from Proposition 3.8.18 which says that Dh(x) has m
independent columns. Hence
i,j
F
i
j
= V
Thus
i,j
F
i
j
A = A
The problem is the F
i
j
might not be disjoint. Let
_
E
i
j
_
be measurable sets such that
E
i
j
F
i
k
A for some k, the sets E
i
j
are disjoint, and their union equals A. Then
A
Jh(x) dx =
i (n,m)
j=1
E
i
j
det
_
Dh(x) Dh(x)
_
1/2
dx. (15.33)
Now each E
i
j
is contained in some B
i
k
and so h
i
has an inverse on h
i
_
B
i
k
_
which I will
denote by g. Thus letting
ic
x x
ic
and using the denition of g
g (h(x) , x
ic
) = x.
By the chain rule,
Dh
i
(g (y)) Dg (y) = I
on h
i
_
E
i
j
_
. Changing the variables using the area formula, the expression in 15.33
equals
A
Jh(x) dx =
i (n,m)
j=1
h
i
(E
i
j
)
det
_
Dh(g (y)) Dh(g (y))
_
1/2
|det Dg (y)| dy =
i (n,m)
j=1
h
i
(E
i
j
)
det
_
Dh(g (y)) Dh(g (y))
_
1/2
det Dh
i
(g (y))
1
dy. (15.34)
Note the integrands are all Borel measurable functions because they are continuous
functions of the entries of matrices which entries come from taking limits of dierence
quotients of continuous functions. Thus from 15.33,
E
i
j
det
_
Dh(x) Dh(x)
_
1/2
dx =
R
n
X
h
i
(E
i
j
)
(y) det
_
Dh(g (y)) Dh(g (y))
_
1/2
det Dh
i
(g (y))
1
dy (15.35)
Next this integral is split using Fubinis theorem. Let y
1
R
m
be xed and now it is
necessary to decide where y
2
is. I need
(y
1
, y
2
) h
i
_
E
i
j
_
=
_
h
_
E
i
j
_
,
ic
_
E
i
j
__
This requires y
2

ic
_
E
i
j
_
. However, it is also the case that y
1
is given. Now y
1
= h(x)
and so
x =(x
i
, x
ic
) = (x
i
, y
2
) h
1
(y
1
)
which implies
ic
x = y
2

ic
h
1
(y
1
) . Thus
y
2

ic
_
h
1
(y
1
) E
i
j
_
and by Fubinis theorem, the integral in 15.35 is
R
m
ic
(h
1
(y1)E
i
j
)
det
_
Dh(g (y)) Dh(g (y))
_
1/2
|det D
x
i
h(g (y))|
1
dy
2
dy
1
(15.36)
Now consider the inner integral in 15.36 in which y
1
is xed. The integrand equals
det
_
_
D
x
i
h(g (y)) D
x
ic
h(g (y))
_
_
D
x
i
h(g (y))
D
x
ic
h(g (y))
__
1/2
|det D
x
i
h(g (y))|
1
.
(15.37)
I want to massage the above expression slightly. Since y
1
is xed, and
y
1
= h(
i
g (y) ,
ic
g (y)) = h(g (y)) ,
it follows
0 = D
x
i
h(g (y)) D
y2
i
g (y) +D
x
ic
h(g (y)) D
y2
ic
g (y)
= D
x
i
h(g (y)) D
y2
i
g (y) +D
x
ic
h(g (y)).
because by denition of g and h
i
,
ic
g (y) = y
2
so D
y2
ic
g (y) = id . Letting A
D
x
i
h(g (y)) and B D
y2
i
g (y) and using the above formula, 15.37 is of the form
det
_
_
A AB
_
_
A
__
1/2
|det A|
1
= det [AA
+ABB
]
1/2
|det A|
1
= det [A(I +BB
) A
]
1/2
|det A|
1
= (det (A) det (A
))
1/2
det (I +BB
)
1/2
|det A|
1
= det (I +BB
)
1/2
,
which, by Corollary 15.7.2, equals det (I +B
B)
1/2
. (Note the size of the identity
changes in these two expressions, the rst being an mm matrix and the second being
a n mn m matrix.)
Since
ic
g (y) = y
2
,
det (I +B
B)
1/2
= det
_
_
B
I
_
_
B
I
__
1/2
= det
_
_
D
y2
i
g (y)
D
y2
ic
g (y)
_
_
D
y2
i
g (y)
D
y2
ic
g (y)
__
1/2
= det
_
D
y2
g (y)
D
y2
g (y)
_
1/2
.
Therefore, 15.36 reduces to
E
i
j
det
_
Dh(x) Dh(x)
_
1/2
dx =
R
m
ic
(h
1
(y1)E
i
j
)
det
_
D
y2
g (y)
D
y2
g (y)
_
1/2
dy
2
dy
1
. (15.38)
Recall g (y
1
, y
2
) = g (h(x) , y
2
) = x. Thus
g (y
1
,
ic
x) = g (y
1
, y
2
) = x.
Thus
g
_
y
1
,
ic
_
h
1
(y
1
) E
i
j
__
= h
1
(y
1
) E
i
j
and so the area formula applied to the inside integral in 15.38 yields
h
1
(y1)E
i
j
dH
n
=
ic
(h
1
(y1)E
i
j
)
det
_
D
y2
g (y)
D
y2
g (y)
_
1/2
dy
2
dy
1
and so this integral equals
H
nm
_
h
1
(y
1
) E
i
j
_
.
It follows
E
i
j
det
_
Dh(x) Dh(x)
_
1/2
dx
=
R
m
H
nm
_
h
1
(y
1
) E
i
j
_
dy
1
.
Therefore, summing the terms over all i and j,
A
det
_
Dh(x) Dh(x)
_
1/2
dx =
R
m
H
nm
_
h
1
(y) A
_
dy.
You dont need to assume h is Lipschitz.
Corollary 15.7.7 Let A be a Lebesgue measurable set in V, an open set and let
h : V R
n
R
m
where m n is C
1
and for which
Jh(x) det
_
Dh(x) Dh(x)
_
1/2
= 0.
Then the following formula holds along with all measurability assertions needed for it to
make sense.

R
m
H
nm
_
A h
1
(y)
_
dy =
A
Jh(x) dx (15.39)
Proof: Consider for each x A, a ball, B(x, r
x
) B(x, r
x
) V such that r
x
< 1.
Then by the mean value inequality, Theorem 6.4.2, h is Lipschitz on B
i
. Letting {B
i
}
be an open covering of A, let C
i
B
i
A. Then let {A
i
} be such that A
i
C
i
but the
A
i
are disjoint and
i
A
i
= A. The conclusion of Lemma 15.7.6 applies with B
i
playing
the role of V in that lemma and one can write
R
m
H
nm
_
A
i
h
1
(y)
_
dy =
Ai
Jh(x) dx
By Lemma 15.4.1 H
nm
Then using the monotone convergence theorem,
R
m
H
nm
_
A h
1
(y)
_
dy =
R
m
H
nm
_
i=1
A
i
h
1
(y)
_
dy =
R
m
i=1
H
nm
_
A
i
h
1
(y)
_
dy =
R
m
lim
M
M
i=1
H
nm
_
A
i
h
1
(y)
_
dy
= lim
M
R
m
M
i=1
H
nm
_
A
i
h
1
(y)
_
dy = lim
M
M
i=1
R
m
H
nm
_
A
i
h
1
(y)
_
dy
= lim
M
M
i=1
X
Ai
(x) Jh(x) dx =
i=1
X
Ai
(x) Jh(x) dx
=
i=1
Ai
(x) Jh(x) dx =
X
A
(x) Jh(x) dx
You dont need to assume Jh(x) det
_
Dh(x) Dh(x)
_
1/2
= 0.
Corollary 15.7.8 Let A V an open set be measurable and let h : V R
m
be C
1
where m n. Then the following formula holds along with all measurability assertions
needed for it to make sense.
R
m
H
nm
_
A h
1
(y)
_
dy =
A
Jh(x) dx (15.40)
where Jh(x) det
_
Dh(x) Dh(x)
_
1/2
.
Proof: By Corollary 15.7.7, this formula is true for all measurable A contained in
the open set R
n
\ S. It remains to verify the formula for all measurable sets A, whether
or not they intersect S.
Consider the case where
A S {x : J (Dh(x)) = 0}.
Let A be compact so that by Lemma 15.7.4, y H
nm
_
A h
1
(y)
_
is Borel measur-
able. For > 0, dene k, p : R
n
R
m
R
m
by
k(x, y) h(x) +y, p(x, y) y.
Then
Dk(x, y) = (Dh(x) , I)
and so
Jk
2
= det
_
(Dh(x) , I)
_
Dh
I
__
= det
_
Dh(x) Dh(x)
+
2
I
_
Now
Dh(x) Dh(x)
is bounded on A because A is compact and Dh is continuous.

The linear operator is also of the form
m
k=1
k
z
k
z
k
for each
k
0 because it is self adjoint and
Dh(x) Dh(x)
x x =Dh(x)
xDh(x)
x 0.
Since A S, at least one of these
k
equals zero. However, they are all bounded by some
constant, C for all x A due to the existence of an upper bound for
Dh(x) Dh(x)
.
Thus
Jk
2
=
m
i=1
_
2
i
+
2
_
[
2m
, C
2
2
] (15.41)
since one of the
i
equals 0. Since Jk = 0, 15.41 implies
Cm
n+m
_
AB(0,1)
_
AB(0,1)
|Jk| dm
n+m
=
R
m
H
n
_
k
1
(y) AB(0,1)
_
dy
Which by Lemma 15.7.4 is at least as large as
C
nm
R
m
R
m
H
nm
_
k
1
(y) p
1
(w) AB(0,1)
_
dwdy (15.42)
where C
nm
=
(n)
(nm)(m)
. It is formula 15.30 applied to the situation where h = p. It
is clear p is Lipschitz continuous with Lipschitz constant 1 since p is a projection.
Claim:
H
nm
_
k
1
(y) p
1
(w) AB(0,1)
_
X
B(0,1)
(w) H
nm
_
h
1
(y w) A
_
.
Proof of the claim: If w / B(0,1), there is nothing to prove so assume w
B(0,1). For such w,
(x, w
1
) k
1
(y) p
1
(w) AB(0,1)
if and only if
k(x, w
1
) = h(x) +w
1
= y, p(x, w
1
) = w
1
= w, x A,
if and only if
(x, w
1
) h
1
(y w) A{w}.
Therefore for w B(0,1),
H
nm
_
k
1
(y) p
1
(w) AB(0,1)
_
H
nm
_
h
1
(y w) A{w}
_
= H
nm
_
h
1
(y w) A
_
.
(Actually equality holds in the claim.) From the claim, 15.42 is at least as large as
C
nm
R
m
B(0,1)
H
nm
_
h
1
(y w) A
_
dwdy (15.43)
= C
nm
B(0,1)
R
m
H
nm
_
h
1
(y w) A
_
dydw
= C
nm
m
m
_
B(0,1)
_
R
m
H
nm
_
h
1
(y) A
_
dy. (15.44)
The use of Fubinis theorem is justied because the integrand is Borel measurable.
Now by 15.44 this has shown
Cm
n+m
_
AB(0,1)
_
C
nm
m
m
_
B(0,1)
_
R
m
H
nm
_
h
1
(y) A
_
dy.
it follows since > 0 is arbitrary,
R
m
H
nm
_
A h
1
(y)
_
dy = 0 =
A
Jh(x) dx.
Since this holds for arbitrary compact sets in S, it follows from Lemma 15.7.5 and inner
regularity of Lebesgue measure that the equation holds for all measurable subsets of S.
Thus if A is any measurable set contained in V
R
m
H
nm
_
A h
1
(y)
_
dy =
R
m
H
nm
_
A S h
1
(y)
_
dy
+
R
m
H
nm
_
(A\ S) h
1
(y)
_
dy
=
R
m
H
nm
_
(A\ S) h
1
(y)
_
dy
=
A\S
Jh(x) dx =
A
Jh(x) dx
This completes the proof of the coarea formula.
15.8 A Nonlinear Fubinis Theorem
Coarea formula holds for h : V R
n
R
m
, n m if whenever A is a Lebesgue
measurable subset of the open set V,the following formula is valid.
R
m
H
nm
_
A h
1
(y)
_
dy =
A
Jh(x) dx (15.45)
where
Jh(x) = det
_
Dh(x) Dh(x)
_
1/2
.
15.8. A NONLINEAR FUBINIS THEOREM 455
Note this is the same as
A
Jh(x))dx =
h(A)
H
nm
_
A h
1
(y)
_
dy
because if y / h(A) , then h
1
(y) = . Now let
s (x) =
p
i=1
c
i
X
Ei
(x)
where E
i
is a Lebesgue measurable subset of V and c
i
0. Then
V
s (x) Jh(x) dx =
p
i=1
c
i
Ei
Jh(x) dx
=
p
i=1
c
i
h(Ei)
H
nm
_
E
i
h
1
(y)
_
dy
=
h(V )
p
i=1
c
i
H
nm
_
E
i
h
1
(y)
_
dy
=
h(V )
_
h
1
(y)
s dH
nm
_
dy
=
h(V )
_
h
1
(y)
s dH
nm
_
dy. (15.46)
Theorem 15.8.1 Let g 0 be Lebesgue measurable and let V be an open subset
of R
n
.
h : V R
m
, n m
where h is C
1
.Then
V
g (x) J ((Dh(x))) dx =
h(V )
_
h
1
(y)
g dH
nm
_
dy.
Proof: Let s
i
g where s
i
is a simple function satisfying 15.46. Then let i and
use the monotone convergence theorem to replace s
i
with g. This proves the nonlinear
version of Fubinis theorem.
Note that this formula is a nonlinear version of Fubinis theorem. The n m di-
mensional surface, h
1
(y), plays the role of R
nm
and H
nm
is like nm dimensional
Lebesgue measure. The term, J ((Dh(x))), corrects for the error occurring because of
the lack of atness of h
1
(y) .
Bibliography
[1] Apostol, T. M., Calculus second edition, Wiley, 1967.
[2] Apostol T.M. Calculus Volume II Second edition, Wiley 1969.
[3] Apostol, T. M., Mathematical Analysis, Addison Wesley Publishing Co., 1974.
[4] Baker, Roger, Linear Algebra, Rinton Press 2001.
[5] Bartle R.G., A Modern Theory of Integration, Grad. Studies in Math., Amer.
Math. Society, Providence, RI, 2000.
[6] Bartle R. G. and Sherbert D.R. Introduction to Real Analysis third edition,
Wiley 2000.
[7] Chahal J. S. , Historical Perspective of Mathematics 2000 B.C. - 2000 A.D.
[8] Davis H. and Snider A., Vector Analysis Wm. C. Brown 1995.
[9] Deimling K. Nonlinear Functional Analysis, Springer-Verlag, 1985.
[10] DAngelo, J. and West D. Mathematical Thinking Problem Solving and Proofs,
Prentice Hall 1997.
[11] Edwards C.H. Advanced Calculus of several Variables, Dover 1994.
[12] Euclid, The Thirteen Books of the Elements, Dover, 1956.
[13] Evans L.C. and Gariepy, Measure Theory and Fine Properties of Functions,
CRC Press, 1992.
[14] Evans L.C. Partial Dierential Equations, Berkeley Mathematics Lecture Notes.
1993.
[15] Fitzpatrick P. M., Advanced Calculus a course in Mathematical Analysis, PWS
Publishing Company 1996.
[16] Federer H., Geometric Measure Theory, Springer-Verlag, New York, 1969.
[17] Fleming W., Functions of Several Variables, Springer Verlag 1976.
[18] Fonesca I. and Gangbo W. Degree theory in analysis and applications Clarendon
Press 1995.
[19] Greenberg, M. Advanced Engineering Mathematics, Second edition, Prentice
Hall, 1998
[20] Gromes W. Ein einfacher Beweis des Satzes von Borsuk. Math. Z. 178, pp. 399
-400 (1981)
457
458 BIBLIOGRAPHY
[21] Gurtin M. An introduction to continuum mechanics, Academic press 1981.
[22] Hardy G., A Course Of Pure Mathematics, Tenth edition, Cambridge University
Press 1992.
[23] Heinz, E.An elementary analytic theory of the degree of mapping in n dimensional
space. J. Math. Mech. 8, 231-247 1959
[24] Henstock R. Lectures on the Theory of Integration, World Scientic Publiching
Co. 1988.
[25] Horn R. and Johnson C. matrix Analysis, Cambridge University Press, 1985.
[26] Karlin S. and Taylor H. A First Course in Stochastic Processes, Academic
Press, 1975.
[27] Kuttler K. L., Basic Analysis, Rinton
[28] Kuttler K.L., Modern Analysis CRC Press 1998.
[29] Lang S. Real and Functional analysis third edition Springer Verlag 1993. Press,
2001.
[30] McLeod R. The Generalized Riemann Integral, Mathematical Association of
America, Carus Mathematical Monographs number 20 1980
[31] McShane E. J. Integration, Princeton University Press, Princeton, N.J. 1944.
[32] Nobel B. and Daniel J. Applied Linear Algebra, Prentice Hall, 1977.
[33] Rose, David, A., The College Math Journal, vol. 22, No.2 March 1991.
[34] Rudin, W., Principles of mathematical analysis, McGraw Hill third edition 1976
[35] Rudin W., Real and Complex Analysis, third edition, McGraw-Hill, 1987.
[36] Salas S. and Hille E., Calculus One and Several Variables, Wiley 1990.
[37] Sears and Zemansky, University Physics, Third edition, Addison Wesley 1963.
[38] Tierney John, Calculus and Analytic Geometry, fourth edition, Allyn and Bacon,
Boston, 1969.
[39] Yosida K., Functional Analysis, Springer Verlag, 1978.
Index
C
c
, 195
C
m
c
, 195
systems, 207
algebra, 157
a.e., 158
adjugate, 45
almost everywhere, 158
approximate identity, 220
area formula, 442
functions which are not one to one,
444
area of a parallelogram, 391
arithmetic mean, 152
at most countable, 15
atlas, 293
axiom of choice, 11, 15, 205
axiom of extension, 11
axiom of specication, 11
axiom of unions, 11
barallelepiped
volume, 392
barrier condition, 345
beta function, 216
Binet Cauchy formula, 300
block matrix, 36
Borel Cantelli lemma, 201
Borel measurable, 205
Borel regular, 420
Borsuk, 270
Borsuk Ulam theorem, 274
boundary operator, 350
bounded, 77
bounded variation, 361
box product, 393
Brouwer degree, 260
Borsuk theorem, 270
Brouwer xed point theorem, 273
Browders lemma, 256
Cantor function, 205
Cantor set, 204
Caratheodorys criterion, 418
Caratheodorys procedure, 166
Cartesian coordinates, 24
Casorati Weierstrass theorem, 413
Cauchy integral theorem, 400
Cauchy Riemann equations, 397
Cauchy Schwarz inequality, 53
Cayley Hamilton theorem, 51
chain rule, 122
change of variables, 455
change of variables general case, 243,
445
characteristic polynomial, 49
chart, 293
coarea formula, 449
cofactor, 43
compact, 153
completion of measure space, 169
components of a vector, 28
connected, 90
connected component, 91
connected components, 91
conservative, 376
contour integral, 398
convergence in measure, 201
convex hull, 81
convolution, 220
Coordinates, 23
countable, 15
countable basis, 83, 319
Cramers rule, 45
cross product, 391
area of parallelogram, 391
coordinate description, 394
geometric description, 391
derivatives, 122
determinant, 39
product, 42
transpose, 40
diameter of a set, 80
dierential equations
Peano existence theorem, 118
dierential form, 305
closed, 322
459
460 INDEX
derivative, 307
exact, 322
integral, 305
dierential forms, 304
Dini derivates, 251
dominated convergence theorem, 193
dot product, 53
dual basis, 70
Egoro theorem, 296
eigenvalue, 151
eigenvalues, 49
equality of mixed partial derivatives, 137
equicontinuous, 112
equivalence class, 17
equivalence relation, 16
equivalent norms, 82
exchange theorem, 26
exponential growth, 253
extreme value theorem, 89
Fatous lemma, 186
xed point property, 290
Frechet derivative, 121
frontier, 355
Fubinis theorem, 212
function, 13
uniformly continuous, 94
Gamma function, 429
gamma function, 203, 216
Gateaux derivative, 125
geometric mean, 152
gradient, 149
Gram Schmidt process, 59
Grammian, 70
Grammian matrix, 322
grating, 349
harmonic function, 327
Haursdor measures, 417
Hausdor and Lebesgue measure, 428,
430
Hausdor dimension, 428
Hausdor measure
translation invariant, 421
Hausdor measures, 417
Heine Borel, 74
Heine Borel theorem, 154
Hermitian, 60
Hessian matrix, 145
higher order derivatives, 130
Holder, 116
Holders inequality, 55
homotopic, 259
homotopy, 247, 259
imaginary part, 396
implicit function theorem, 140
inner product, 53
inner regularity, 159
interior point, 75
invariance of domain, 271
invariant, 62
inverse function theorem, 141, 150
inverses and determinants, 44
isodiametric inequality, 423, 427
isolated singularity, 413
iterated integral, 209
Jordan arc, 354
Jordan curve, 355
Jordan curve theorem, 289
Jordan Separation theorem, 278
Jordan separation theorem, 279
Lagrange multipliers, 147, 148
Laplace expansion, 43
Laplace transform, 250, 254
Laplaces equation, 327
Lebesgue number, 153
length, 363
limit point, 76
Lindelof property, 319
linear combination, 26
linear transformation, 30
linearly dependent, 26
linearly independent, 26
Liouvilles theorem, 409
local maximum, 145
local minimum, 145
locally nite, 223
lower semicontinuous, 116
manifolds
boundary, 293
interior, 293
orientable, 295
smooth, 295
matrix
left inverse, 45
lower triangular, 46
right inverse, 45
upper triangular, 46
matrix of a linear transformation, 32
max. min.theorem, 89
INDEX 461
measurable, 165
measurable function, 175
pointwise limits, 175
measurable functions
Borel, 201
measurable sets, 165
measure, 157
measure space, 157
minimal polynomial, 50
minor, 43
mixed partial derivatives, 135
mollier, 220
monotone convergence theorem, 184
monotone functions
dierentiable, 252
multi - index, 98
multi-index, 132
nested interval lemma, 74
nonlinear Fubinis theorem, 455
nonmeasurable set, 205
norm
p norm, 55
odd map, 269
open cover, 153
open set, 75
orientable manifold, 295
orientation, 306, 363
oriented curve, 363
orthonormal, 58
outer measure, 157, 201
outer regularity, 159
parallelepiped, 392
partial derivatives, 124
pi systems, 207
pointwise convergence
sequence, 95
series, 97
Poissons equation, 327
Poissons integral formula, 335
Poissons problem, 327
polar decomposition
left, 68
right, 65
potential, 374
power set, 11
precompact, 118
primitive, 400
probability measure, 179
probability space, 179
product formula, 277
rank of a matrix, 46
rational function, 98
real part, 396
rectiable, 361
rectiable curve, 361
regular measure, 159
regular values, 259
relative topology, 293
retraction, 247
right Cauchy Green strain tensor, 65
right handed system, 391
Rouche theorem, 411
Russells paradox, 13
Sards lemma, 240
scalars, 25
Schroder Bernstein theorem, 14
Schurs theorem, 62
second derivative test, 146
self adjoint, 60
separable, 83
separated, 90
sets, 11
sigma algebra, 157
simple curve, 361
simple functions, 178
singular values, 259
smooth surface, 394
span, 26
Steiner symetrization, 425
Stirlings formula, 203
Stokes theorem, 390, 395
Stokes theorem, 311
subharmonic, 341
subspace, 26
support, 195
Taylors formula, 144
Tietze extension theorem, 106
triangle inequality, 55
trivial, 26
uniform contractions, 138
uniform convergence
sequence, 95
series, 97
uniformly Cauchy
sequence, 95
uniformly continuous, 94
uniformly integrable, 202, 298
unitary, 66
upper semicontinuous, 116
vector space axioms, 25
462 INDEX
vectors, 25
Vitali
convergence theorem, 298
Vitali convergence theorem, 299
Vitali covering theorem, 227, 228, 230
Vitali coverings, 228, 230
volume of unit ball, 429
weak maximum principle, 327
Weierstrass M test, 97
work, 407

AdvancedCalculusMV PDF

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

AdvancedCalculusMV PDF

Încărcat de

Drepturi de autor:

Formate disponibile

Multivariable Advanced Calculus

In terms of matrix multiplication, Theorem 3.3.9 says that if M

in the ordered list, (i

The proof will depend on the following lemma.

. Then using Lemma 3.8.9,

= p 1. Then using the Holder inequality,

= ||x +y|| ||x||

= q to save on notation and consider the following

y such that for all x H

is linear. This is because by the properties of

. This is also called Hermitian.

Proof: Consider the rst claim.

Here is a very important example.

In the case where A L(F

) = 0, it follows the m m matrix AA

. Thus the row rank of A

to every L invariant subspace has its eigenvalues in

. Dividing by its length, it can be

being in F, the eld of scalars. If F = C this is no restriction. There is

Rx x y)| = 0 for all x, y because the given x, y were arbitrary. Let y =

Rx x to conclude that for all x,

R = I since x is arbitrary. This proves the lemma.

F and so by Theorem 3.8.23, there is an orthonormal basis of

R = I from Lemma 3.9.1. This proves the theorem.

The following corollary follows as a simple consequence of this theorem. It is called

. Now apply Theorem 3.9.3 to

and U satisfy the conditions of that theorem. Then

= I. This proves the corollary.

. Show that for a

such that whenever n n

such that if n > n

and suppose that for all k = 1, 2, ,

and so every subsequence also converges to k

by Theorem 4.1.6. Thus there is no

such that whenever n, m n

such that if n > n

consists of the set of limit

f (x) g (y) f (y) g (x)

|f (x) g (y) f (y) g (x)|

M [|f (x) f (y)| +|g (y) g (x)|]

5.1 Continuity And The Limit Of A Sequence

) = f (k). Since f is one to one, this requires k

denotes the set of limit points of A.

but y / A. If y / A, a closed set, then there

The following theorem is a very useful description of the open sets in R.

such that even though ||x

V , sort of a V valued polynomial. Then such a function is continuous

< because the above implies

< . Therefore, the rst

on [0, 1]. Let

< /2, it follows that

(x)| for all x,

is a real valued continuous function dened on a sequen-

Denition 5.9.5 More generally, if K is any subset of a normed vector space

, then for all x K,

: } is a collection of continuous functions. Let

) is a complete normed linear space. This is called a

(t)|| = ||Df (x +t (y x)) (y x)||

(t)|| ||v|| . Thus

f exists and is continuous.

< 1 which, by Theorem 5.8.3, happens if

)|| = ||T(x(y) , y) T(x(y

(t) = Df (x+tv) (v) , h