Multivariable Advanced Calculus

Kenneth Kuttler

October 12, 2009

2
Contents

1 Introduction 9

2 Some Fundamental Concepts 11

2.1 Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 The Schroder Bernstein Theorem . . . . . . . . . . . . . . . . . . 13
2.1.3 Equivalence Relations . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 lim sup And lim inf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Double Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Basic Linear Algebra 23

3.1 Algebra in Fn , 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 Fn . . . . . . . . . . . . . . . . . . . . . . 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 Schur’s 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 . . . . . . . . . . . . . . . . . . . 102
5.8 The Operator Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.9 Ascoli Arzela Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6 The Derivative 119

6.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.2 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.3 The Matrix Of The Derivative . . . . . . . . . . . . . . . . . . . . . . . 121
6.4 A Mean Value Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.5 Existence Of The Derivative, C 1 Functions . . . . . . . . . . . . . . . . 124
6.6 Higher Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.7 C k Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.7.1 Some Standard Notation . . . . . . . . . . . . . . . . . . . . . . . 130
6.8 The Derivative And The Cartesian Product . . . . . . . . . . . . . . . . 130
6.9 Mixed Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.10 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.10.1 More Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.10.2 The Case Of Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.11 Taylor’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.11.1 Second Derivative Test . . . . . . . . . . . . . . . . . . . . . . . . 143
6.12 The Method Of Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . 145
6.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

7 Measures And Measurable Functions 151

7.1 Compact Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
7.2 An Outer Measure On P (R) . . . . . . . . . . . . . . . . . . . . . . . . 153
7.3 General Outer Measures And Measures . . . . . . . . . . . . . . . . . . 155
7.3.1 Measures And Measure Spaces . . . . . . . . . . . . . . . . . . . 155
7.4 The Borel Sets, Regular Measures . . . . . . . . . . . . . . . . . . . . . 156
7.4.1 Definition of Regular Measures . . . . . . . . . . . . . . . . . . . 156
7.4.2 The Borel Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.4.3 Borel Sets And Regularity . . . . . . . . . . . . . . . . . . . . . . 157
7.5 Measures And Outer Measures . . . . . . . . . . . . . . . . . . . . . . . 163
7.5.1 Measures From Outer Measures . . . . . . . . . . . . . . . . . . . 163
7.5.2 Completion Of Measure Spaces . . . . . . . . . . . . . . . . . . . 167
7.6 One Dimensional Lebesgue Stieltjes Measure . . . . . . . . . . . . . . . 170
7.7 Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
CONTENTS 5

8 The Abstract Lebesgue Integral 179

8.1 Definition For Nonnegative Measurable Functions . . . . . . . . . . . . . 179
8.1.1 Riemann Integrals For Decreasing Functions . . . . . . . . . . . . 179
8.1.2 The Lebesgue Integral For Nonnegative Functions . . . . . . . . 180
8.2 The Lebesgue Integral For Nonnegative Simple Functions . . . . . . . . 180
8.3 The Monotone Convergence Theorem . . . . . . . . . . . . . . . . . . . 183
8.4 Other Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
8.5 Fatou’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.6 The Righteous Algebraic Desires Of The Lebesgue Integral . . . . . . . 185
8.7 The Lebesgue Integral, L1 . . . . . . . . . . . . . . . . . . . . . . . . . . 186
8.8 Approximation With Simple Functions . . . . . . . . . . . . . . . . . . . 189
8.9 The Dominated Convergence Theorem . . . . . . . . . . . . . . . . . . . 192
8.10 Approximation With Cc (Y ) . . . . . . . . . . . . . . . . . . . . . . . . . 194
8.11 The One Dimensional Lebesgue Integral . . . . . . . . . . . . . . . . . . 196
8.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

9 The Lebesgue Integral For Functions Of n Variables 207

9.1 π Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
9.2 n Dimensional Lebesgue Measure And Integrals . . . . . . . . . . . . . . 208
9.2.1 Iterated Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
9.2.2 n Dimensional Lebesgue Measure And Integrals . . . . . . . . . . 209
9.2.3 Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
9.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.4 Lebesgue Measure On Rn . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.5 Mollifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 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 n Dimensions . . . . . . . . . . . . . . . . . . . 243
9.12 Brouwer Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . 246
9.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

10 Brouwer Degree 257

10.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
10.2 Definitions And Elementary¡ Properties
¢ . . . . . . . . . . . . . . . . . . . 259
10.2.1 The Degree For C 2 Ω; Rn . . . . . . . . . . . . . . . . . . . . . 260
10.2.2 Definition Of The Degree For Continuous Functions . . . . . . . 266
10.3 Borsuk’s 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 Differential Forms 293

11.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
11.2 Some Important Measure Theory . . . . . . . . . . . . . . . . . . . . . . 296
11.2.1 Eggoroff’s 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 Differential Forms On Manifolds . . . . . . . . . . . . . . 304

11.5.1 The Derivative Of A Differential Form . . . . . . . . . . . . . . . 307
11.6 Stoke’s Theorem And The Orientation Of ∂Ω . . . . . . . . . . . . . . . 307
11.7 Green’s Theorem, An Example . . . . . . . . . . . . . . . . . . . . . . . 311
11.7.1 An Oriented Manifold . . . . . . . . . . . . . . . . . . . . . . . . 311
11.7.2 Green’s 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 Poisson’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
12.2.1 Poisson’s Problem For A Ball . . . . . . . . . . . . . . . . . . . . 331
12.2.2 Does It Work In Case f = 0? . . . . . . . . . . . . . . . . . . . . 333
12.2.3 The Case Where f 6= 0, Poisson’s Equation . . . . . . . . . . . . 335
12.3 Properties Of Harmonic Functions . . . . . . . . . . . . . . . . . . . . . 338
12.4 Laplace’s Equation For General Sets . . . . . . . . . . . . . . . . . . . . 341
12.4.1 Properties Of Subharmonic Functions . . . . . . . . . . . . . . . 341
12.4.2 Poisson’s Problem Again . . . . . . . . . . . . . . . . . . . . . . . 346

13 The Jordan Curve Theorem 349

14 Line Integrals 359

14.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
14.1.1 Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
14.1.2 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
14.2 The Line Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
14.3 Simple Closed Rectifiable Curves . . . . . . . . . . . . . . . . . . . . . . 374
14.3.1 The Jordan Curve Theorem . . . . . . . . . . . . . . . . . . . . . 375
14.3.2 Orientation And Green’s Formula . . . . . . . . . . . . . . . . . 380
14.4 Stoke’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
14.5 Interpretation And Review . . . . . . . . . . . . . . . . . . . . . . . . . 389
14.5.1 The Geometric Description Of The Cross Product . . . . . . . . 389
14.5.2 The Box Product, Triple Product . . . . . . . . . . . . . . . . . . 390
14.5.3 A Proof Of The Distributive Law For The Cross Product . . . . 391
14.5.4 The Coordinate Description Of The Cross Product . . . . . . . . 391
14.5.5 The Integral Over A Two Dimensional Surface . . . . . . . . . . 392
14.6 Introduction To Complex Analysis . . . . . . . . . . . . . . . . . . . . . 394
14.6.1 Basic Theorems, The Cauchy Riemann Equations . . . . . . . . 394
14.6.2 Contour Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
14.6.3 The Cauchy Integral . . . . . . . . . . . . . . . . . . . . . . . . . 398
14.6.4 The Cauchy Goursat Theorem . . . . . . . . . . . . . . . . . . . 401
14.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404

15 Hausdorff Measures And Area Formula 415

15.1 Definition Of Hausdorff Measures . . . . . . . . . . . . . . . . . . . . . . 415
15.1.1 Properties Of Hausdorff Measure . . . . . . . . . . . . . . . . . . 416
15.1.2 Hn And mn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
15.2 Technical Considerations∗ . . . . . . . . . . . . . . . . . . . . . . . . . . 421
15.2.1 Steiner Symmetrization∗ . . . . . . . . . . . . . . . . . . . . . . . 423
15.2.2 The Isodiametric Inequality∗ . . . . . . . . . . . . . . . . . . . . 425
∗
15.2.3 The Proper Value Of β (n) . . . . . . . . . . . . . . . . . . . . . 425
CONTENTS 7

∗
15.2.4 A Formula For α (n) . . . . . . . . . . . . . . . . . . . . . . . . 426
15.3 Hausdorff Measure And Linear Transformations . . . . . . . . . . . . . . 428
15.4 The Area Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
15.4.1 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . 430
15.5 The Area Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
15.6 Area Formula For Mappings Which Are Not One To One . . . . . . . . 440
15.7 The Coarea Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
15.8 A Nonlinear Fubini’s Theorem . . . . . . . . . . . . . . . . . . . . . . . 452
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 first understand some linear
algebra. Therefore, a condensed course in linear algebra is presented first, 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 efficient 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 differential 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 Green’s theorem in the plane leading to a very good version of
Stoke’s 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 Apostol’s advanced calculus book. Finally, there is an introduction to
Hausdorff 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 differentiation 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 Definitions
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 specification, 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 specified 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 everyone’s 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 finitely many
sets. If S is a set whose elements are sets,
Y
A
A∈S

signifies 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 6= 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 specification 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 AC
( or more precisely as X \ A) is a set obtained from using the axiom of specification to
write
AC ≡ {x ∈ X : x ∈ / A}
The symbol ∈ / means: “is not an element of”. Note the axiom of specification takes
place relative to a given set. Without this universal set it makes no sense to use the
axiom of specification to obtain the complement.
Words such as “all” or “there exists” are called quantifiers 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 quantifiers are used turns
2.1. SET THEORY 13

out to be illogical even though such usage may be grammatically correct. Quantifiers
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.
DeMorgan’s 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
∪ AC : A ∈ S = (∩ {A : A ∈ S})

and © ª C
∩ AC : A ∈ S = (∪ {A : A ∈ S}) .
These laws follow directly from the definitions. Also following directly from the defini-
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 specification. 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 qualifies for inclusion in A so it is an
element of itself and so this can’t 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 quantifiers wherever they occur, even this one. This little
observation described above is due to Bertrand Russell and is called Russell’s 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 definition reviews this and defines the concept of a function.

Definition 2.1.1 Let X and Y be sets.

X × Y ≡ {(x, y) : x ∈ X and y ∈ Y }

A relation is defined to be a subset of X × Y . A function, f, also called a mapping, is a

relation which has the property that if (x, y) and (x, y1 ) are both elements of the f , then
y = y1 . The domain of f is defined 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,
14 SOME FUNDAMENTAL CONCEPTS

that is what the above definition 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 suffer
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.

X Y
f
A - C = f (A)

g
B = g(D) ¾ D

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
A0 ⊆ X satisfies P if whenever y ∈ Y \ f (A0 ) , g (y) ∈
/ A0 .
A ≡ {A0 ⊆ X : A0 satisfies P}.
Let A = ∪A. If y ∈ Y \ f (A), then for each A0 ∈ A, y ∈ Y \ f (A0 ) and so g (y) ∈
/ A0 .
Since g (y) ∈
/ A0 for all A0 ∈ A, it follows g (y) ∈
/ A. Hence A satisfies P and is the
largest subset of X which does so. Now define
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
satisfies 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 define
½
f (x) if x ∈ A
h (x) ≡
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

Definition 2.1.4 Let I be a set and let Xi be a set for each i ∈ I. f is a choice
function written as
Y
f∈ Xi
i∈I

if f (i) ∈ Xi for each i ∈ I.

The axiom of choice says that if Xi 6= ∅ for each i ∈ I, for I a set, then
Y
Xi 6= ∅.
i∈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 −1Q(y) ≡ {x ∈ X : f (x) = y} 6= ∅. Therefore, by the axiom

of choice, there exists f0−1 ∈ y∈Y f −1 (y) which is the same as saying that for each
y ∈ Y , f0−1 (y) ∈ f −1 (y). Similarly, there exists g0−1 (x) ∈ g −1 (x) for all x ∈ X. Then
f0−1 is one to one because if f0−1 (y1 ) = f0−1 (y2 ), then
¡ ¢ ¡ ¢
y1 = f f0−1 (y1 ) = f f0−1 (y2 ) = y2 .

Similarly g0−1 is one to one. Therefore, by the Schroder Bernstein theorem, there exists
h : X → Y which is one to one and onto.

Definition 2.1.6 A set S, is finite if there exists a natural number n and a

map θ which maps {1, · · ·, n} one to one and onto S. S is infinite if it is not finite.
A set S, is called countable if there exists a map θ mapping N one to one and onto
S.(When θ maps a set A to a set B, this will be written as θ : A → B in the future.)
Here N ≡ {1, 2, · · ·}, the natural numbers. S is at most countable if there exists a map
θ : 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 first, 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 η : N → X which is onto. Define

η (i) ≡ xi and consider X as the set {x1 , x2 , x3 , · · ·}. Similarly, consider Y as the set
{y1 , y2 , y3 , · · ·}. It follows the elements of X ×Y are included in the following rectangular
array.

(x1 , y1 ) (x1 , y2 ) (x1 , y3 ) · · · ← Those which have x1 in first slot.

(x2 , y1 ) (x2 , y2 ) (x2 , y3 ) · · · ← Those which have x2 in first slot.
(x3 , y1 ) (x3 , y2 ) (x3 , y3 ) · · · ← Those which have x3 in first slot. .
.. .. .. ..
. . . .
16 SOME FUNDAMENTAL CONCEPTS

Follow a path through this array as follows.

(x1 , y1 ) → (x1 , y2 ) (x1 , y3 ) →

. %
(x2 , y1 ) (x2 , y2 )
↓ %
(x3 , y1 )

Thus the first element of X × Y is (x1 , y1 ), the second element of X × Y is (x1 , y2 ), the
third element of X × Y is (x2 , y1 ) 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 α : N → X which is one to one and onto. Let β : X×Y → N
be defined by β ((x, y)) ≡ α−1 (x). Thus β is onto N. By the first 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 = {x1 , x2 , x3 , · · ·}

and
Y = {y1 , y2 , y3 , · · ·} .
Consider the following array consisting of X ∪ Y and path through it.

x1 → x2 x3 →
. %
y1 → y2

Thus the first element of X ∪ Y is x1 , the second is x2 the third is y1 the fourth is y2
etc.
Consider the second claim. By the first part, there is a map from N onto X × Y .
Suppose without loss of generality that X is countable and α : N → X is one to one and
onto. Then define β (y) ≡ 1, for all y ∈ Y ,and β (x) ≡ α−1 (x). Thus, β 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.

Definition 2.1.9 Let S be a set. ∼ is an equivalence relation on S if it satisfies

the following axioms.

1. x ∼ x for all x ∈ S. (Reflexive)

2.2. LIM SUP AND LIM INF 17

2. If x ∼ y then y ∼ x. (Symmetric)

3. If x ∼ y and y ∼ z, then x ∼ z. (Transitive)

Definition 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 definition one can prove the following simple theorem.

Theorem 2.1.11 Let ∼ be an equivalence class defined 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 lim sup And lim inf

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.

Definition 2.2.1 For A ⊆ [−∞, ∞] , A 6= ∅ sup A is defined 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 defined 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 {An } is an increasing sequence in [−∞, ∞], then

sup {An } = lim An .
n→∞

Similarly, if {An } is decreasing, then

inf {An } = lim An .

n→∞

Proof: Let sup ({An : n ∈ N}) = r. In the first case, suppose r < ∞. Then letting
ε > 0 be given, there exists n such that An ∈ (r − ε, r]. Since {An } is increasing, it
follows if m > n, then r − ε < An ≤ Am ≤ r and so limn→∞ An = r as claimed. In
the case where r = ∞, then if a is a real number, there exists n such that An > a.
Since {Ak } is increasing, it follows that if m > n, Am > a. But this is what is meant
by limn→∞ An = ∞. The other case is that r = −∞. But in this case, An = −∞ for all
n and so limn→∞ An = −∞. The case where An is decreasing is entirely similar. This
proves the lemma. ¥
n
Sometimes the limit of a sequence does not exist. For example, if an = (−1) , then
limn→∞ an does not exist. This is because the terms of the sequence are a distance
of 1 apart. Therefore there can’t exist a single number such that all the terms of the
sequence are ultimately within 1/4 of that number. The nice thing about lim sup and
lim inf is that they always exist. First here is a simple lemma and definition.
18 SOME FUNDAMENTAL CONCEPTS

Definition 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 {An } is an increasing sequence of points of [−∞, ∞] ,
limn→∞ An equals ∞ if the only upper bound of the set {An } is ∞. If {An } is bounded
above by a real number, then limn→∞ An is defined in the usual way and equals the
least upper bound of {An }. If {An } is a decreasing sequence of points of [−∞, ∞] ,
limn→∞ An equals −∞ if the only lower bound of the sequence {An } is −∞. If {An } is
bounded below by a real number, then limn→∞ An is defined in the usual way and equals
the greatest lower bound of {An }. More simply, if {An } is increasing,
lim An = sup {An }
n→∞

and if {An } is decreasing then

lim An = inf {An } .
n→∞

Lemma 2.2.4 Let {an } be a sequence of real numbers and let Un ≡ sup {ak : k ≥ n} .
Then {Un } is a decreasing sequence. Also if Ln ≡ inf {ak : k ≥ n} , then {Ln } is an
increasing sequence. Therefore, limn→∞ Ln and limn→∞ Un both exist.
Proof: Let Wn be an upper bound for {ak : k ≥ n} . Then since these sets are
getting smaller, it follows that for m < n, Wm is an upper bound for {ak : k ≥ n} . In
particular if Wm = Um , then Um is an upper bound for {ak : k ≥ n} and so Um is at
least as large as Un , the least upper bound for {ak : k ≥ n} . The claim that {Ln } is
decreasing is similar. This proves the lemma. ¥
From the lemma, the following definition makes sense.
Definition 2.2.5 Let {an } be any sequence of points of [−∞, ∞]
lim sup an ≡ lim sup {ak : k ≥ n}
n→∞ n→∞

lim inf an ≡ lim inf {ak : k ≥ n} .

n→∞ n→∞

Theorem 2.2.6 Suppose {an } is a sequence of real numbers and that

lim sup an
n→∞

and
lim inf an
n→∞
are both real numbers. Then limn→∞ an exists if and only if
lim inf an = lim sup an
n→∞ n→∞

and in this case,

lim an = lim inf an = lim sup an .
n→∞ n→∞ n→∞

Proof: First note that

sup {ak : k ≥ n} ≥ inf {ak : k ≥ n}
and so from Theorem 4.1.7,
lim sup an ≡ lim sup {ak : k ≥ n}
n→∞ n→∞

≥ lim inf {ak : k ≥ n}

n→∞
≡ lim inf an .
n→∞
2.2. LIM SUP AND LIM INF 19

Suppose first that limn→∞ an exists and is a real number. Then by Theorem 4.4.3 {an }
is a Cauchy sequence. Therefore, if ε > 0 is given, there exists N such that if m, n ≥ N,
then
|an − am | < ε/3.
From the definition of sup {ak : k ≥ N } , there exists n1 ≥ N such that

sup {ak : k ≥ N } ≤ an1 + ε/3.

Similarly, there exists n2 ≥ N such that

inf {ak : k ≥ N } ≥ an2 − ε/3.

It follows that
2ε
sup {ak : k ≥ N } − inf {ak : k ≥ N } ≤ |an1 − an2 | + < ε.
3
∞ ∞
Since the sequence, {sup {ak : k ≥ N }}N =1 is decreasing and {inf {ak : k ≥ N }}N =1 is
increasing, it follows from Theorem 4.1.7

0 ≤ lim sup {ak : k ≥ N } − lim inf {ak : k ≥ N } ≤ ε

N →∞ N →∞

Since ε is arbitrary, this shows

lim sup {ak : k ≥ N } = lim inf {ak : k ≥ N } (2.1)

N →∞ N →∞

Next suppose 2.1. Then

lim (sup {ak : k ≥ N } − inf {ak : k ≥ N }) = 0

N →∞

Since sup {ak : k ≥ N } ≥ inf {ak : k ≥ N } it follows that for every ε > 0, there exists
N such that
sup {ak : k ≥ N } − inf {ak : k ≥ N } < ε
Thus if m, n > N, then
|am − an | < ε
which means {an } is a Cauchy sequence. Since R is complete, it follows that limn→∞ an ≡
a exists. By the squeezing theorem, it follows

a = lim inf an = lim sup an

n→∞ n→∞

and This proves the theorem. ¥

With the above theorem, here is how to define the limit of a sequence of points in
[−∞, ∞].

Definition 2.2.7 Let {an } be a sequence of points of [−∞, ∞] . Then limn→∞ an

exists exactly when
lim inf an = lim sup an
n→∞ n→∞

and in this case

lim an ≡ lim inf an = lim sup an .
n→∞ n→∞ n→∞

The significance of lim sup and lim inf, in addition to what was just discussed, is
contained in the following theorem which follows quickly from the definition.
20 SOME FUNDAMENTAL CONCEPTS

Theorem 2.2.8 Suppose {an } is a sequence of points of [−∞, ∞] . Let

λ = lim sup an .
n→∞

Then if b > λ, it follows there exists N such that whenever n ≥ N,

an ≤ b.
If c < λ, then an > c for infinitely many values of n. Let
γ = lim inf an .
n→∞

Then if d < γ, it follows there exists N such that whenever n ≥ N,

an ≥ d.
If e > γ, it follows an < e for infinitely many values of n.
The proof of this theorem is left as an exercise for you. It follows directly from the
definition and it is the sort of thing you must do yourself. Here is one other simple
proposition.
Proposition 2.2.9 Let limn→∞ an = a > 0. Then
lim sup an bn = a lim sup bn .
n→∞ n→∞

Proof: This follows from the definition. Let λn = sup {ak bk : k ≥ n} . For all n
large enough, an > a − ε where ε is small enough that a − ε > 0. Therefore,
λn ≥ sup {bk : k ≥ n} (a − ε)
for all n large enough. Then
lim sup an bn = lim λn ≡ lim sup an bn
n→∞ n→∞ n→∞
≥ lim (sup {bk : k ≥ n} (a − ε))
n→∞
= (a − ε) lim sup bn
n→∞

Similar reasoning shows

lim sup an bn ≤ (a + ε) lim sup bn
n→∞ n→∞

Now since ε > 0 is arbitrary, the conclusion follows.

2.3 Double Series

Sometimes it is required to consider double series which are of the form
 
X∞ X ∞ X∞ X∞
ajk ≡  ajk  .
k=m j=m k=m j=m

In other words, first 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
∞ X
X ∞ ∞ X
X ∞
ajk = ajk .
k=m j=m j=m k=m
2.3. DOUBLE SERIES 21

In other words, when does it make no difference which subscript is summed over first?
In the case of finite sums there is no issue here. You can always write

M X
X N N X
X M
ajk = ajk
k=m j=m j=m k=m

because addition is commutative. However, there are limits involved with infinite sums
and the interchange in order of summation involves taking limits in a different 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 c 0 -c

0 0 0 0 c 0 -c 0

0 0 0 c 0 -c 0 0

0 0 c 0 -c 0 0 0

0 c 0 -c 0 0 0 0

b 0 -c 0 0 0 0 0

0 a 0 0 0 0 0 0

The numbers next to the point are the values of amn . You see ann = 0 for all n,
a21 = a, a12 = b, amn = c for (m, n) on the line y = 1 + x whenever m > 1, and
amn = −c for all (m, n) on the line y = x − 1 whenever m > 2.
P∞ P∞ P∞
Then m=1 amn = a if n = 1, m=1 amn = b−c if n = 2 and if n > 2, m=1 amn =
0. Therefore,
∞ X
X ∞
amn = a + b − c.
n=1 m=1
22 SOME FUNDAMENTAL CONCEPTS

P∞ P∞ P∞
Next observe that n=1 amn = b if m = 1, n=1 amn = a+c if m = 2, and n=1 amn =
0 if m > 2. Therefore,
X∞ X ∞
amn = b + a + c
m=1 n=1
and so the two sums are different. Moreover, you can see that by assigning different
values of a, b, and c, you can get an example for any two different numbers desired.
It turns out that if aij ≥ 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.
Definition 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 supa∈A f (a, b) means sup (Sb ) where Sb ≡
{f (a, b) : a ∈ A} .
Unlike limits, you can take the sup in different orders.
Lemma 2.3.3 Let f (a, b) ∈ [−∞, ∞] for a ∈ A and b ∈ B where A, B are sets.
Then
sup sup f (a, b) = sup sup f (a, b) .
a∈A b∈B b∈B a∈A

Proof: Note that for all a, b, f (a, b) ≤ supb∈B supa∈A f (a, b) and therefore, for all
a, supb∈B f (a, b) ≤ supb∈B supa∈A f (a, b). Therefore,
sup sup f (a, b) ≤ sup sup f (a, b) .
a∈A b∈B b∈B a∈A

Repeat the same argument interchanging a and b, to get the conclusion of the lemma.
Theorem 2.3.4 Let aij ≥ 0. Then
∞ X
X ∞ ∞ X
X ∞
aij = aij .
i=1 j=1 j=1 i=1

Proof: First note there is no trouble in defining these sums because the aij are all
nonnegative. If a sum diverges, it only diverges to ∞ and so ∞ is the value of the sum.
Next note that
X∞ X ∞ X∞ Xn
aij ≥ sup aij
n
j=r i=r j=r i=r
because for all j,
∞
X n
X
aij ≥ aij .
i=r i=r
Therefore,
∞ X
X ∞ ∞ X
X n m X
X n
aij ≥ sup aij = sup lim aij
n n m→∞
j=r i=r j=r i=r j=r i=r

n X
X m n
X m
X
= sup lim aij = sup lim aij
n m→∞ n m→∞
i=r j=r i=r j=r
n
XX∞ n
XX∞ ∞ X
X ∞
= sup aij = lim aij = aij
n n→∞
i=r j=r i=r j=r i=r j=r

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, Cn 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 Rn , 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.

Definition 3.0.5 Define

Fn ≡ {(x1 , · · ·, xn ) : xj ∈ F for j = 1, · · ·, n} .

(x1 , · · ·, xn ) = (y1 , · · ·, yn ) if and only if for all j = 1, · · ·, n, xj = yj . When

(x1 , · · ·, xn ) ∈ Fn ,

it is conventional to denote (x1 , · · ·, xn ) by the single bold face letter, x. The numbers,
xj are called the coordinates. The set

{(0, · · ·, 0, t, 0, · · ·, 0) : t ∈ F}

for t in the ith slot is called the ith coordinate axis. The point 0 ≡ (0, · · ·, 0) is called
the origin.

Thus (1, 2, 4i) ∈ F3 and (2, 1, 4i) ∈ F3 but (1, 2, 4i) 6= (2, 1, 4i) because, even though
the same numbers are involved, they don’t match up. In particular, the first entries are
not equal.
The geometric significance of Rn 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 definition, R1 = R. Recall that R is identified 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

6 · (2, 6)

(−8, 3) · 3
2

−8

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, x1 on the horizontal
line in the above picture and x2 on the vertical line in the above picture, such that
the point of interest is identified with the ordered pair, (x1 , x2 ) . In short, points in the
plane can be identified with ordered pairs similar to the way that points on the real
line are identified with real numbers. Now suppose n = 3. As just explained, the first
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 can’t 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 first object is and you would need another three coordinates to describe
where the other object is located. Therefore, you would need to be considering R6 . 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 R5 . 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 profit 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 Descartes1 who invented this idea in the first 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 significance of Cn for n > 1 is not available because each copy of C
corresponds to the plane or R2 .

1 René 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 different 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 scientifically. Descartes ended up dying in Sweden.
3.1. ALGEBRA IN FN , VECTOR SPACES 25

3.1 Algebra in Fn , Vector Spaces

There are two algebraic operations done with elements of Fn . One is addition and the
other is multiplication by numbers, called scalars. In the case of Cn the scalars are
complex numbers while in the case of Rn the only allowed scalars are real numbers.
Thus, the scalars always come from F in either case.

Definition 3.1.1 If x ∈ Fn and a ∈ F, also called a scalar, then ax ∈ Fn is

defined by
ax = a (x1 , · · ·, xn ) ≡ (ax1 , · · ·, axn ) . (3.1)
n n
This is known as scalar multiplication. If x, y ∈ F then x + y ∈ F and is defined by

x + y = (x1 , · · ·, xn ) + (y1 , · · ·, yn )
≡ (x1 + y1 , · · ·, xn + yn ) (3.2)

the points in Fn are also referred to as vectors.

With this definition, 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 field of scalars satisfying the axioms of the following theorem, it is called a
vector space.

Theorem 3.1.2 For v, w ∈ Fn and α, β 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

α (v + w) = αv+αw, (3.7)

(α + β) v =αv+βv, (3.8)
α (βv) = αβ (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

α (v + w) = α (v1 + w1 , · · ·, vn + wn )
= (α (v1 + w1 ) , · · ·, α (vn + wn ))
= (αv1 + αw1 , · · ·, αvn + αwn )
= (αv1 , · · ·, αvn ) + (αw1 , · · ·, αwn )
= αv + αw.

As usual subtraction is defined as x − y ≡ x+ (−y) .

26 BASIC LINEAR ALGEBRA

3.2 Subspaces Spans And Bases

The concept of linear combination is fundamental in all of linear algebra.

Definition 3.2.1 Let {x1 , · · ·, xp } be vectors in a vector space, Y having the

field of scalars F. A linear combination is any expression of the form
p
X
ci xi
i=1

where the ci are scalars. The set of all linear combinations of these vectors is called
span (x1 , · · ·, xn ) . If V ⊆ Y, then V is called a subspace if whenever α, β are scalars
and u and v are vectors of V, it follows αu + βv ∈ 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 {x1 , · · ·, xn } is called linearly independent if whenever
p
X
ck xk = 0
k=1

it follows that all the scalars, ck equal zero. A set of vectors, {x1 , · · ·, xp } , is called
linearly dependent if it is not linearly independent. Thus the set of vectors Pp is linearly
dependent if there exist scalars, ci , i = 1, · · ·, n, not all zero such that k=1 ck xk = 0.

Lemma 3.2.2 A set of vectors {x1 , · · ·, xp } is linearly independent if and only if

none of the vectors can be obtained as a linear combination of the others.

Proof: Suppose first that {x1 , · · ·, xp } is linearly independent. If

X
xk = cj xj ,
j6=k

then X
0 = 1xk + (−cj ) xj ,
j6=k

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 {x1 , · · ·, xp } linearly
independent? If it is not, there exist scalars, ci , not all zero such that
p
X
ci xi = 0.
i=1

Say ck 6= 0. Then you can solve for xk as

X
xk = (−cj ) /ck xj
j6=k

contrary to assumption. This proves the lemma. ¥

The following is called the exchange theorem.

Theorem 3.2.3 Let {x1 , · · ·, xr } be a linearly independent set of vectors such

that each xi is in the span {y1 , · · ·, ys } . Then r ≤ s.
3.2. SUBSPACES SPANS AND BASES 27

Proof: Define span {y1 , · · ·, ys } ≡ V, it follows there exist scalars, c1 , · · ·, cs such

that
Xs
x1 = ci yi . (3.11)
i=1

Not all of these scalars can equal zero because if this were the case, it would follow
that x1P= 0 and so {x1 , · · ·, xr } would not be linearly independent. Indeed, if x1 = 0,
r
1x1 + i=2 0xi = x1 = 0 and so there would exist a nontrivial linear combination of
the vectors {x1 , · · ·, xr } which equals zero.
Say ck 6= 0. Then solve (3.11) for yk and obtain
 
s-1 vectors here
z }| {
yk ∈ span x1 , y1 , · · ·, yk−1 , yk+1 , · · ·, ys  .

Define {z1 , · · ·, zs−1 } by

{z1 , · · ·, zs−1 } ≡ {y1 , · · ·, yk−1 , yk+1 , · · ·, ys }

Therefore, span {x1 , z1 , · · ·, zs−1 } = V because if v ∈ V, there exist constants c1 , · · ·, cs

such that
s−1
X
v= ci zi + cs yk .
i=1

Now replace the yk in the above with a linear combination of the vectors, {x1 , z1 , · · ·, zs−1 }
to obtain v ∈ span {x1 , z1 , · · ·, zs−1 } . The vector yk , in the list {y1 , · · ·, ys } , has now
been replaced with the vector x1 and the resulting modified list of vectors has the same
span as the original list of vectors, {y1 , · · ·, ys } .
Now suppose that r > s and that span {x1 , · · ·, xl , z1 , · · ·, zp } = V where the vectors,
z1 , ···, zp are each taken from the set, {y1 , · · ·, ys } 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,
xl+1 is a vector not in the list, {x1 , · · ·, xl } and since span {x1 , · · ·, xl , z1 , · · ·, zp } = V
there exist scalars, ci and dj such that
l
X p
X
xl+1 = ci xi + dj zj . (3.12)
i=1 j=1

Now not all the dj can equal zero because if this were so, it would follow that {x1 , · · ·, xr }
would be a linearly dependent set because one of the vectors would equal a linear
combination of the others. Therefore, (3.12) can be solved for one of the zi , say zk , in
terms of xl+1 and the other zi and just as in the above argument, replace that zi with
xl+1 to obtain
 
p-1 vectors here
z }| {
span x1 , · · ·xl , xl+1 , z1 , · · ·zk−1 , zk+1 , · · ·, zp  = V.

Continue this way, eventually obtaining

span (x1 , · · ·, xs ) = V.

But then xr ∈ span {x1 , · · ·, xs } contrary to the assumption that {x1 , · · ·, xr } is linearly
independent. Therefore, r ≤ s as claimed.
Here is another proof in case you didn’t like the above proof.
28 BASIC LINEAR ALGEBRA

Theorem 3.2.4 If

span (u1 , · · ·, ur ) ⊆ span (v1 , · · ·, vs ) ≡ V

and {u1 , · · ·, ur } are linearly independent, then r ≤ s.

Proof: Suppose r > s. Let Ep denote a finite list of vectors of {v1 , · · ·, vs } and
let |Ep | denote the number of vectors in the list. Let Fp denote the first p vectors in
{u1 , · · ·, ur }. In case p = 0, Fp will denote the empty set. For 0 ≤ p ≤ s, let Ep have
the property
span (Fp , Ep ) = V
and |Ep | is as small as possible for this to happen. I claim |Ep | ≤ s−p if Ep is nonempty.
Here is why. For p = 0, it is obvious. Suppose true for some p < s. Then

up+1 ∈ span (Fp , Ep )

and so there are constants, c1 , · · ·, cp and d1 , · · ·, dm where m ≤ s − p such that

p
X m
X
up+1 = ci ui + di zj
i=1 j=1

for
{z1 , · · ·, zm } ⊆ {v1 , · · ·, vs } .
Then not all the di can equal zero because this would violate the linear independence
of the {u1 , · · ·, ur } . Therefore, you can solve for one of the zk as a linear combination
of {u1 , · · ·, up+1 } and the other zj . Thus you can change Fp to Fp+1 and include one
fewer vector in Ep . Thus |Ep+1 | ≤ m − 1 ≤ s − p − 1. This proves the claim.
Therefore, Es is empty and span (u1 , · · ·, us ) = V. However, this gives a contradiction
because it would require
us+1 ∈ span (u1 , · · ·, us )
which violates the linear independence of these vectors. This proves the theorem. ¥

Definition 3.2.5 A finite set of vectors, {x1 , · · ·, xr } is a basis for a vector

space V if
span (x1 , · · ·, xr ) = V
and {x1 , · · ·, xr } is linearly
Pr independent. Thus if v ∈ V there exist unique scalars,
v1 , · · ·, vr such that v = i=1 vi xi . These scalars are called the components of v with
respect to the basis {x1 , · · ·, xr }.

Corollary 3.2.6 Let {x1 , · · ·, xr } and {y1 , · · ·, ys } be two bases2 of Fn . Then r =

s = n.

Proof: From the exchange theorem, r ≤ s and s ≤ r. Now note the vectors,

1 is in the ith slot

z }| {
ei = (0, · · ·, 0, 1, 0 · ··, 0)

for i = 1, 2, · · ·, n are a basis for Fn . 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 shiek’s 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 {v1 , · · ·, vr } be a set of vectors. Then V ≡ span (v1 , · · ·, vr ) is a

subspace.
Pr Pr
Proof: Suppose α, β are two scalars and let k=1 ck vk and k=1 dk vk are two
elements of V. What about
r
X r
X
α ck vk + β dk vk ?
k=1 k=1

Is it also in V ?
r
X r
X r
X
α ck vk + β dk vk = (αck + βdk ) vk ∈ V
k=1 k=1 k=1

so the answer is yes. This proves the lemma. ¥

Definition 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 finite
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 (u1 , · · ·, uk ) and {u1 , · · ·, uk } is linearly indepen-

dent. Then {u1 , · · ·, uk , v} is also linearly independent.
Pk
Proof: Suppose i=1 ci ui + dv = 0. It is required to verify that each ci = 0 and
that d = 0. But if d 6= 0, then you can solve for v as a linear combination of the vectors,
{u1 , · · ·, uk },
Xk ³ ´
ci
v=− ui
i=1
d
Pk
contrary to assumption. Therefore, d = 0. But then i=1 ci ui = 0 and the linear
independence of {u1 , · · ·, uk } implies each ci = 0 also. This proves the lemma. ¥

Theorem 3.2.10 Let V be a nonzero subspace of Y a finite dimensional vector

space having dimension n. Then V has a basis.

Proof: Let v1 ∈ V where v1 6= 0. If span {v1 } = V, stop. {v1 } is a basis for V .

Otherwise, there exists v2 ∈ V which is not in span {v1 } . By Lemma 3.2.9 {v1 , v2 } is
a linearly independent set of vectors. If span {v1 , v2 } = V stop, {v1 , v2 } is a basis for
V. If span {v1 , v2 } 6= V, then there exists v3 ∈
/ span {v1 , v2 } and {v1 , v2 , v3 } 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 finite dimensional vector space of di-

mension n and let {v1 , · · ·, vr } be a linearly independent set of vectors in V . Then either
it is a basis for V or there exist vectors, vr+1 , · · ·, vs such that {v1 , · · ·, vr , vr+1 , · · ·, vs }
is a basis for V.
30 BASIC LINEAR ALGEBRA

Proof: This follows immediately from the proof of Theorem 3.2.10. You do exactly
the same argument except you start with {v1 , · · ·, vr } rather than {v1 }.
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 finite dimensional vector space of

dimension n and suppose span (u1 · ··, up ) = V where the ui are nonzero vectors. Then
there exist vectors, {v1 · ··, vr } such that {v1 · ··, vr } ⊆ {u1 · ··, up } and {v1 · ··, vr } is a
basis for V .

Proof: Let r be the smallest positive integer with the property that for some set,
{v1 · ··, vr } ⊆ {u1 · ··, up } ,
span (v1 · ··, vr ) = V.
Then r ≤ p and it must be the case that {v1 · ··, vr } is linearly independent because if
it were not so, one of the vectors, say vk would be a linear combination of the others.
But then you could delete this vector from {v1 · ··, vr } and the resulting list of r − 1
vectors would still span V contrary to the definition 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 definition of a linear transformation.

Definition 3.3.1 Let V and W be two finite 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 α and β, and vectors v, w,

L (αv+βw) = αL (v) + βL (w) .

An example of a linear transformation is familiar matrix multiplication, familiar if

you have had a linear algebra course. Let A = (aij ) be an m × n matrix. Then an
example of a linear transformation L : Fn → Fm is given by
n
X
(Lv)i ≡ aij vj .
j=1

Here  
v1
 
v ≡  ...  ∈ Fn .
vn
In the general case, the space of linear transformations is itself a vector space. This
will be discussed next.

Definition 3.3.2 Given L, M ∈ L (V, W ) define a new element of L (V, W ) ,

denoted by L + M according to the rule

(L + M ) v ≡ Lv + M v.

For α a scalar and L ∈ L (V, W ) , define αL ∈ L (V, W ) by

αL (v) ≡ α (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 definitions 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 {v1 , · · ·, vn } is a basis for
V. Then if L : V → W is given by Lvk = wk ∈ W and
à n ! n n
X X X
L ak vk ≡ ak Lvk = ak wk
k=1 k=1 k=1

then L is well defined and is in L (V, W ) . Also, if L, M are two linear transformations
such that Lvk = M vk for all k, then M = L.

Proof: L is well defined on V because, since {v1 , · · ·, vn } is a basis, there is exactly

one way to write a given vector of V as a linear combination. Next, observe Pthat L is
n
obviously linear from the definition. If L, M are equal on the basis, then if k=1 ak vk
is an arbitrary vector of V,
à n ! n
X X
L ak vk = ak Lvk
k=1 k=1
n
à n
!
X X
= ak M vk = M ak vk
k=1 k=1

and so L = M because they give the same result for every vector in V .
The message is that when you define a linear transformation, it suffices to tell what
it does to a basis.

Definition 3.3.4 The symbol, δ ij is defined as 1 if i = j and 0 if i 6= j.

Theorem 3.3.5 Let V and W be finite dimensional vector spaces of dimension

n and m respectively Then dim (L (V, W )) = mn.

Proof: Let two sets of bases be

{v1 , · · ·, vn } and {w1 , · · ·, wm }

for V and W respectively. Using Lemma 3.3.3, let wi vj ∈ L (V, W ) be the linear
transformation defined on the basis, {v1 , · · ·, vn }, by

wi vk (vj ) ≡ wi δ jk .

Note that to define these special linear transformations, sometimes called dyadics, it is
necessary that {v1 , · · ·, vn } be a basis since their definition requires giving the values of
the linear transformation on a basis.
Let L ∈ L (V, W ). Since {w1 , · · ·, wm } is a basis, there exist constants djr such that
m
X
Lvr = djr wj
j=1

Then from the above,

m
X m X
X n m X
X n
Lvr = djr wj = djr δ kr wj = djr wj vk (vr )
j=1 j=1 k=1 j=1 k=1
32 BASIC LINEAR ALGEBRA

which shows
m X
X n
L= djk wj vk
j=1 k=1

because the two linear transformations agree on a basis. Since L is arbitrary this shows

{wi vk : i = 1, · · ·, m, k = 1, · · ·, n}

spans L (V, W ).
If X
dik wi vk = 0,
i,k

then
X m
X
0= dik wi vk (vl ) = dil wi
i,k i=1

and so, since {w1 , · · ·, wm } is a basis, dil = 0 for each i = 1, · · ·, m. Since l is arbitrary,
this shows dil = 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.

Definition 3.3.6 Let V, W be finite dimensional vector spaces such that a basis
for V is
{v1 , · · ·, vn }
and a basis for W is
{w1 , · · ·, wm }.
Then as explained in Theorem 3.3.5, for L ∈ L (V, W ) , there exist scalars lij such that
X
L= lij wi vj
ij

Consider a rectangular array of scalars such that the entry in the ith row and the j th
column is lij ,
 
l11 l12 · · · l1n
 l21 l22 · · · l2n 
 
 .. .. .. .. 
 . . . . 
lm1 lm2 ··· lmn
This is called the matrix of the linear transformation with respect to the two bases. This
will typically be denoted by (lij ) . It is called a matrix and in this case the matrix is
m × n because it has m rows and n columns.

Theorem 3.3.7 Let L ∈ L (V, W ) and let (lij ) be the matrix of L with respect
to the two bases,
{v1 , · · ·, vn } and {w1 , · · ·, wm }.
of V and W respectively. Then for v ∈ V having components (x1 , · · ·, xn ) with respect
to the basis {v1 , · · ·, vn }, the components of Lv with respect to the basis {w1 , · · ·, wm }
are  
X X
 l1j xj , · · ·, lmj xj 
j j
3.3. LINEAR TRANSFORMATIONS 33

Proof: From the definition of lij ,

à !
X X X
Lv = lij wi vj (v) = lij wi vj xk vk
ij ij k
 
X X X X
= lij wi vj (vk ) xk = lij wi δ jk xk =  lij xj  wi
ijk ijk i j

and This proves the theorem. ¥

Theorem 3.3.8 Let (V, {v1 , · · ·, vn }) , (U, {u1 , · · ·, um }) , (W, {w1 , · · ·, wp }) be three
vector spaces along with bases for each one. Let L ∈ L (V, U ) and M ∈ L (U, W ) . Then
M L ∈ L (V, W ) and if (cij ) is the matrix of M L with respect to {v1 , · · ·, vn } and
{w1 , · · ·, wp } and (lij ) and (mij ) are the matrices of L and M respectively with respect
to the given bases, then
m
X
crj = mrs lsj .
s=1

Proof: First note that from the definition,

(wi uj ) (uk vl ) (vr ) = (wi uj ) uk δ lr = wi δ jk δ lr

and
wi vl δ jk (vr ) = wi δ jk δ lr
which shows
(wi uj ) (uk vl ) = wi vl δ jk (3.13)
Therefore,
à ! 
X X
ML = mrs wr us  lij ui vj 
rs ij
X X
= mrs lij (wr us ) (ui vj ) = mrs lij wr vj δ is
rsij rsij
à !
X X X
= mrs lsj wr vj = mrs lsj wr vj
rsj rj s

and This proves the theorem. ¥

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, {v1 , · · ·, vn }) is a vector space and a basis and
(V, {v10 , · · ·, vn0 }) is the same vector space with a different basis. Suppose
¡ 0 ¢L ∈ L (V, V ) .
Let (lij ) be the matrix of L taken with respect to {v1 , · · ·, vn } and let lij be the n × n
matrix of L taken with respect to {v10 , · · ·, vn0 }That is,
X X
0
L= lij vi vj , L = lrs vr0 vs0 .
ij rs
¡ ¢
Then there exist n × n matrices (dij ) and d0ij satisfying
X
dij d0jk = δ ik
j
34 BASIC LINEAR ALGEBRA

such that X
0
lij = dir lrs d0sj
rs

Proof: First consider the identity map, id defined by id (v) = v with respect to the
two bases, {v1 , · · ·, vn } and {v10 , · · ·, vn0 }.
X X
id = d0tu vt0 vu , id = dij vi vj0 (3.14)
tu ij

Now it follows from 3.13

X ¡ ¢ X 0
id = id ◦ id = d0tu dij (vt0 vu ) vi vj0 = dtu dij δ iu vt0 vj0
tuij tuij
à !
X X X
= d0ti dij vt0 vj0 = d0ti dij vt0 vj0
tij tj i

On the other hand, X

id = δ tj vt0 vj0
tj

because id (vk0 ) = vk0

and
X X X
δ tj vt0 vj0 (vk0 ) = δ tj vt0 δ jk = δ tk vt0 = vk0 .
tj tj t

Therefore, Ã !
X
d0ti dij = δ tj .
i
Switching the order of the above products shows
à !
X
0
dti dij = δ tj
i
¡ ¢
In terms of matrices, this says d0ij is the inverse matrix of (dij ) .
Now using 3.14 and the cancellation property 3.13,
X X X
0
L = liu vi vu = lrs vr0 vs0 = id 0
lrs vr0 vs0 id
iu rs rs
X X X
= dij vi vj0 0
lrs vr0 vs0 d0tu vt0 vu
ij rs tu
X ¡ ¢
0
= dij lrs d0tu vi vj0 (vr0 vs0 ) (vt0 vu )
ijturs
 
X X X
= 0
dij lrs d0tu vi vu δ jr δ st =  0
dij ljs d0su  vi vu
ijturs iu js

and since the linear transformations, {vi vu } are linearly independent, this shows
X
0
liu = dij ljs d0su
js

as claimed. This proves the theorem. ¥

Recall the following definition which is a review of important terminology about
matrices.
3.3. LINEAR TRANSFORMATIONS 35

Definition 3.3.10 If A is an m × n matrix and B is an n × p matrix, A =

(Aij ) , B = (Bij ) , then if (AB)ij is the ij th entry of the product, then
X
(AB)ij = Aik Bkj
k

An n × n matrix, A is said to be invertible if there exists another n × n matrix, denoted

−1 −1 −1 th
by
¡ TA¢ such that AA = A A = I where the ij entry of I is δ ij . Recall also that
A ij ≡ Aji . 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.
−1
4. When the inverses exist, (AB) = B −1 A−1
T
5. (AB) = B T AT
Proof: I will prove these things directly from the above definition but there are
more elegant ways to see these things in terms of composition of linear transformations
which is really whatPmatrix multiplication corresponds to.
First, (IA)ij ≡ k δ ik Akj = Aij . The other order is similar.
Next consider the associative law of multiplication.
X XX
((AB) C)ij ≡ (AB)ik Ckj = Air Brk Ckj
k k r
X X X
= Air Brk Ckj = Air (BC)rj = (A (BC))ij
r k r

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
¡ ¢ −1
Similarly, B −1 A−1 AB = I. Hence from what was just shown, (AB) exists and
equals B −1 A−1 .
Finally consider the statement about transposes.
³ ´ X X¡ ¢ ¡ ¢ ¡ ¢
T
(AB) ≡ (AB)ji ≡ Ajk Bki ≡ B T ik AT kj ≡ B T AT ij
ij
k k
th
Since the ij 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 M1 and M2 are matrices
for the same linear transformation relative to two different bases, it follows there exists
an invertible matrix, S such that
M1 = S −1 M2 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.
36 BASIC LINEAR ALGEBRA

3.4 Block Multiplication Of Matrices

Consider the following problem
µ ¶µ ¶
A B E F
C D G H

You know how to do this from the above definition of matrix mutiplication. You get
µ ¶
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
 
A11 · · · A1m
 .. .. .. 
 . . .  (3.15)
Ar1 ··· Arm

where Aij is a si × pj matrix where si does not depend on j and pj does not depend
P on
matrix is called a block matrix, also a partitioned matrix. Let n = j pj
i. Such a P
and k = i si so A is an k × n matrix. What is Ax where x ∈ Fn ? From the process
of multiplying a matrix times a vector, the following lemma follows.

Lemma 3.4.1 Let A be an m × n block matrix as in 3.15 and let x ∈ Fn . Then Ax

is of the form  P 
A1j xj j
 .. 
Ax =  
P .
A x
j rj j
T
where x = (x1 , · · ·, xm ) and xi ∈ Fpi .

Suppose also that B is a block matrix of the form

 
B11 · · · B1p
 .. .. .. 
 . . .  (3.16)
Br1 ··· Brp

and A is a block matrix of the form

 
A11 ··· A1m
 .. .. .. 
 . . .  (3.17)
Ap1 ··· Apm

and that for all i, j, it makes sense to multiply Bis Asj for all s ∈ P
{1, · · ·, m} and that
for each s, Bis Asj is the same size so that it makes sense to write s Bis Asj .

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 Bis is conformable with Asj and each product, Bis Asj 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 X
Bis Asj . (3.18)
s
3.5. DETERMINANTS 37

Proof: Let Bis be a qi × ps matrix and Asj be a ps × rj matrix. Also let x ∈ Fn

T
and let x = (x1 , · · ·, xm ) and xi ∈ Fri so it makes sense to multiply Asj xj . Then from
the associative law of matrix multiplication and Lemma 3.4.1 applied twice,
    
B11 · · · B1p A11 · · · A1m x1
 .. .. ..   .. .. ..   .. 
 . . .  . . .   . 
Br1 ··· Brp Ap1Apm ··· xm
  P 
B11 · · · B1p j A1j xj
 ..   
=  ... ..
. .   P ..
.

Br1 · · · Brp j A rj x j
 P P   P P 
s j B1s Asj xj j( s B1s Asj ) xj
 ..   .. 
= = 
P P . P P .
s j Brs Asj xj j( s Brs Asj ) xj
 P P  
s B1s As1 ··· s B1s Asm x1
 .. .. ..   .. 
= . . .  . 
P P
B
s rs s1A · · · B A
s rs sm xm
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 Fn , 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 definition of the determinant.

Lemma 3.5.1 There exists a unique function, sgnn 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.
sgnn (1, · · ·, n) = 1 (3.19)
sgnn (i1 , · · ·, p, · · ·, q, · · ·, in ) = − sgnn (i1 , · · ·, q, · · ·, p, · · ·, in ) (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 {i1 , · · ·, in } =
{1, · · ·, n} so that every number from {1, · · ·, n} appears in the ordered list, (i1 , · · ·, in ) ,

sgnn (i1 , · · ·, iθ−1 , n, iθ+1 , · · ·, in ) ≡

n−θ
(−1) sgnn−1 (i1 , · · ·, iθ−1 , iθ+1 , · · ·, in ) (3.21)
where n = iθ in the ordered list, (i1 , · · ·, in ) .

Proof: To begin with, it is necessary to show the existence of such a function. This
is clearly true if n = 1. Define sgn1 (1) ≡ 1 and observe that it works. No switching
is possible. In the case where n = 2, it is also clearly true. Let sgn2 (1, 2) = 1 and
sgn2 (2, 1) = −1 while sgn2 (2, 2) = sgn2 (1, 1) = 0 and verify it works. Assuming such a
function exists for n, sgnn+1 will be defined in terms of sgnn . If there are any repeated
38 BASIC LINEAR ALGEBRA

numbers in (i1 , · · ·, in+1 ) , sgnn+1 (i1 , · · ·, in+1 ) ≡ 0. If there are no repeats, then n + 1
appears somewhere in the ordered list. Let θ be the position of the number n + 1 in the
list. Thus, the list is of the form (i1 , · · ·, iθ−1 , n + 1, iθ+1 , · · ·, in+1 ) . From 3.21 it must
be that
sgnn+1 (i1 , · · ·, iθ−1 , n + 1, iθ+1 , · · ·, in+1 ) ≡
n+1−θ
(−1) sgnn (i1 , · · ·, iθ−1 , iθ+1 , · · ·, in+1 ) .
It is necessary to verify this satisfies 3.19 and 3.20 with n replaced with n + 1. The first
of these is obviously true because
n+1−(n+1)
sgnn+1 (1, · · ·, n, n + 1) ≡ (−1) sgnn (1, · · ·, n) = 1.

If there are repeated numbers in (i1 , · · ·, in+1 ) , then it is obvious 3.20 holds because
both sides would equal zero from the above definition. It remains to verify 3.20 in the
case where there are no numbers repeated in (i1 , · · ·, in+1 ) . Consider
³ r s
´
sgnn+1 i1 , · · ·, p, · · ·, q, · · ·, in+1 ,

where the r above the p indicates the number, p is in the rth position and the s above
the q indicates that the number, q is in the sth position. Suppose first that r < θ < s.
Then µ ¶
r θ s
sgnn+1 i1 , · · ·, p, · · ·, n + 1, · · ·, q, · · ·, in+1 ≡
³ r s−1
´
n+1−θ
(−1) sgnn i1 , · · ·, p, · · ·, q , · · ·, in+1

while µ ¶
r θ s
sgnn+1 i1 , · · ·, q, · · ·, n + 1, · · ·, p, · · ·, in+1 =
³ r s−1
´
n+1−θ
(−1) sgnn i1 , · · ·, q, · · ·, p , · · ·, in+1

and so, by induction, a switch of p and q introduces a minus sign in the result. Similarly,
if θ > s or if θ < r it also follows that 3.20 holds. The interesting case is when θ = r or
θ = s. Consider the case where θ = r and note the other case is entirely similar.
³ r s
´
sgnn+1 i1 , · · ·, n + 1, · · ·, q, · · ·, in+1 =
³ s−1
´
n+1−r
(−1) sgnn i1 , · · ·, q , · · ·, in+1 (3.22)

while ³ ´
r s
sgnn+1 i1 , · · ·, q, · · ·, n + 1, · · ·, in+1 =
³ r
´
n+1−s
(−1) sgnn i1 , · · ·, q, · · ·, in+1 . (3.23)

By making s − 1 − r switches, move the q which is in the s − 1th position in 3.22 to the
rth position in 3.23. By induction, each of these switches introduces a factor of −1 and
so ³ ´ ³ ´
s−1 s−1−r r
sgnn i1 , · · ·, q , · · ·, in+1 = (−1) sgnn i1 , · · ·, q, · · ·, in+1 .

Therefore,
³ r s
´ ³ s−1
´
n+1−r
sgnn+1 i1 , · · ·, n + 1, · · ·, q, · · ·, in+1 = (−1) sgnn i1 , · · ·, q , · · ·, in+1
3.5. DETERMINANTS 39
³ r
´
n+1−r s−1−r
= (−1) (−1) sgnn i1 , · · ·, q, · · ·, in+1
³ r
´ ³ r
´
n+s 2s−1 n+1−s
= (−1) sgnn i1 , · · ·, q, · · ·, in+1 = (−1) (−1) sgnn i1 , · · ·, q, · · ·, in+1
³ r s ´
= − sgnn+1 i1 , · · ·, q, · · ·, n + 1, · · ·, in+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 (i1 , · · ·, in ) = g (i1 , · · ·, in ) . 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 sgnn because the context supplies
the appropriate n.
Definition 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. Define
X
f (k1 · · · kn )
(k1 ,···,kn )

to be the sum of all the f (k1 · · · kn ) for all possible choices of ordered lists (k1 , · · ·, kn )
of numbers of {1, · · ·, n} . For example,
X
f (k1 , k2 ) = f (1, 2) + f (2, 1) + f (1, 1) + f (2, 2) .
(k1 ,k2 )

Definition 3.5.3 Let (aij ) = A denote an n × n matrix. The determinant of

A, denoted by det (A) is defined by
X
det (A) ≡ sgn (k1 , · · ·, kn ) a1k1 · · · ankn
(k1 ,···,kn )

where the sum is taken over all ordered lists of numbers from {1, · · ·, n}. Note it suffices
to take the sum over only those ordered lists in which there are no repeats because if
there are, sgn (k1 , · · ·, kn ) = 0 and so that term contributes 0 to the sum.
Let A be an n × n matrix, A = (aij ) and let (r1 , · · ·, rn ) denote an ordered list of n
numbers from {1, · · ·, n}. Let A (r1 , · · ·, rn ) denote the matrix whose k th row is the rk
row of the matrix, A. Thus
X
det (A (r1 , · · ·, rn )) = sgn (k1 , · · ·, kn ) ar1 k1 · · · arn kn (3.24)
(k1 ,···,kn )

and
A (1, · · ·, n) = A.
Proposition 3.5.4 Let
(r1 , · · ·, rn )
be an ordered list of numbers from {1, · · ·, n}. Then
sgn (r1 , · · ·, rn ) det (A)
X
= sgn (k1 , · · ·, kn ) ar1 k1 · · · arn kn (3.25)
(k1 ,···,kn )

= det (A (r1 , · · ·, rn )) . (3.26)

40 BASIC LINEAR ALGEBRA

Proof: Let (1, · · ·, n) = (1, · · ·, r, · · ·s, · · ·, n) so r < s.

det (A (1, · · ·, r, · · ·, s, · · ·, n)) = (3.27)

X
sgn (k1 , · · ·, kr , · · ·, ks , · · ·, kn ) a1k1 · · · arkr · · · asks · · · ankn ,
(k1 ,···,kn )

and renaming the variables, calling ks , kr and kr , ks , this equals

X
= sgn (k1 , · · ·, ks , · · ·, kr , · · ·, kn ) a1k1 · · · arks · · · askr · · · ankn
(k1 ,···,kn )

 These got switched 

X z }| {
= − sgn k1 , · · ·, kr , · · ·, ks , · · ·, kn  a1k1 · · · askr · · · arks · · · ankn
(k1 ,···,kn )

= − 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,
p
det (A (r1 , · · ·, rn )) = (−1) det (A)

where it took p switches to obtain(r1 , · · ·, rn ) from (1, · · ·, n). By Lemma 3.5.1, this
implies
p
det (A (r1 , · · ·, rn )) = (−1) det (A) = sgn (r1 , · · ·, rn ) det (A)
and proves the proposition in the case when there are no repeated numbers in the ordered
list, (r1 , · · ·, rn ). However, if there is a repeat, say the rth row equals the sth row, then
the reasoning of 3.27 -3.28 shows that A (r1 , · · ·, rn ) = 0 and also sgn (r1 , · · ·, rn ) = 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 AT .

Corollary 3.5.6 The following formula for det (A) is valid.

1
det (A) = ·
n!
X X
sgn (r1 , · · ·, rn ) sgn (k1 , · · ·, kn ) ar1 k1 · · · arn kn . (3.29)
(r1 ,···,rn ) (k1 ,···,kn )
¡ T¢
And also
¡ T¢ T det A = det (A) where AT is the transpose of A. (Recall that for AT =
aij , aij = aji .)

Proof: From Proposition 3.5.4, if the ri are distinct,

X
det (A) = sgn (r1 , · · ·, rn ) sgn (k1 , · · ·, kn ) ar1 k1 · · · arn kn .
(k1 ,···,kn )
3.5. DETERMINANTS 41

Summing over all ordered lists, (r1 , · · ·, rn ) where the ri are distinct, (If the ri are not
distinct, sgn (r1 , · · ·, rn ) = 0 and so there is no contribution to the sum.)

n! det (A) =
X X
sgn (r1 , · · ·, rn ) sgn (k1 , · · ·, kn ) ar1 k1 · · · arn kn .
(r1 ,···,rn ) (k1 ,···,kn )

This proves the corollary. ¥ since the formula gives the same number for A as it does
for AT .

Corollary 3.5.7 If two rows or two columns in an n×n 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 ith row of A equals (xa1 + yb1 , · · ·, xan + ybn ). Then

det (A) = x det (A1 ) + y det (A2 )

where the ith row of A1 is (a1 , · · ·, an ) and the ith row of A2 is (b1 , · · ·, bn ) , all other
rows of A1 and A2 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 A1 is the matrix obtained from A by switching two
columns, ¡ ¢ ¡ ¢
det (A) = det AT = − det AT1 = − det (A1 ) .
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.
X
det (A) ≡ sgn (k1 , · · ·, kn ) a1k1 · · · (xaki + ybki ) · · · ankn
(k1 ,···,kn )

X
=x sgn (k1 , · · ·, kn ) a1k1 · · · aki · · · ankn
(k1 ,···,kn )
X
+y sgn (k1 , · · ·, kn ) a1k1 · · · bki · · · ankn
(k1 ,···,kn )

≡ x det (A1 ) + y det (A2 ) .

¡ ¢
The same is true of columns because det AT = det (A) and the rows of AT 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 = a1 · · · an be the columns of A and suppose the condition
that one column is a linear combination of r of the others is satisfied. Then by us-
th
ing Corollary 3.5.7 you may rearrange the columns
Pr to have the n column a linear
combination of the first r columns. Thus an = k=1 ck ak and so
¡ Pr ¢
det (A) = det a1 · · · ar · · · an−1 k=1 ck ak .
42 BASIC LINEAR ALGEBRA

By Corollary 3.5.7
r
X ¡ ¢
det (A) = ck det a1 ··· ar ··· an−1 ak = 0.
k=1
¡ ¢
The case for rows follows from the fact that det (A) = det AT . This proves the corol-
lary. ¥
Recall the following definition of matrix multiplication.

Definition 3.5.9 If A and B are n × n matrices, A = (aij ) and B = (bij ),

AB = (cij ) where
n
X
cij ≡ aik bkj .
k=1

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 cij be the ij th entry of AB. Then by Proposition 3.5.4,

det (AB) =

X
sgn (k1 , · · ·, kn ) c1k1 · · · cnkn
(k1 ,···,kn )
à ! à !
X X X
= sgn (k1 , · · ·, kn ) a1r1 br1 k1 ··· anrn brn kn
(k1 ,···,kn ) r1 rn
X X
= sgn (k1 , · · ·, kn ) br1 k1 · · · brn kn (a1r1 · · · anrn )
(r1 ···,rn ) (k1 ,···,kn )
X
= sgn (r1 · · · rn ) a1r1 · · · anrn det (B) = det (A) det (B) .
(r1 ···,rn )

This proves the theorem. ¥

Lemma 3.5.11 Suppose a matrix is of the form

µ ¶
A ∗
M= (3.30)
0 a
or µ ¶
A 0
M= (3.31)
∗ a
where a is a number and A is an (n − 1) × (n − 1) matrix and ∗ denotes either a column
or a row having length n − 1 and the 0 denotes either a column or a row of length n − 1
consisting entirely of zeros. Then

det (M ) = a det (A) .

3.5. DETERMINANTS 43

Proof: Denote M by (mij ) . Thus in the first case, mnn = a and mni = 0 if i 6= n
while in the second case, mnn = a and min = 0 if i 6= n. From the definition of the
determinant, X
det (M ) ≡ sgnn (k1 , · · ·, kn ) m1k1 · · · mnkn
(k1 ,···,kn )

Letting θ denote the position of n in the ordered list, (k1 , · · ·, kn ) then using the earlier
conventions used to prove Lemma 3.5.1, det (M ) equals
X µ ¶
θ n−1
n−θ
(−1) sgnn−1 k1 , · · ·, kθ−1 , kθ+1 , · · ·, kn m1k1 · · · mnkn
(k1 ,···,kn )

Now suppose 3.31. Then if kn 6= n, the term involving mnkn in the above expression
equals zero. Therefore, the only terms which survive are those for which θ = n or in
other words, those for which kn = n. Therefore, the above expression reduces to
X
a sgnn−1 (k1 , · · ·kn−1 ) m1k1 · · · m(n−1)kn−1 = a det (A) .
(k1 ,···,kn−1 )

To get the assertion in the situation of 3.30 use Corollary 3.5.6 and 3.31 to write
µµ T ¶¶
¡ ¢ A 0 ¡ ¢
det (M ) = det M T = det = a det AT = a det (A) .
∗ 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 definition
of a determinant.

Definition 3.5.12 Let A = (aij ) be an n × n matrix. Then a new matrix called

the cofactor matrix, cof (A) is defined by cof (A) = (cij ) where to obtain cij delete the
ith 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
i+j
(−1) . 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

n
X n
X
det (A) = aij cof (A)ij = aij cof (A)ij . (3.32)
j=1 i=1

The first formula consists of expanding the determinant along the ith row and the second
expands the determinant along the j th column.

Proof: Let (ai1 , · · ·, ain ) be the ith row of A. Let Bj be the matrix obtained from A
by leaving every row the same except the ith row which in Bj equals (0, · · ·, 0, aij , 0, · · ·, 0) .
Then by Corollary 3.5.7,
Xn
det (A) = det (Bj )
j=1

Denote by A the (n − 1) × (n − 1) matrix obtained by deleting the ith row and the
ij
i+j ¡ ¢
j th column of A. Thus cof (A)ij ≡ (−1) det Aij . At this point, recall that from
44 BASIC LINEAR ALGEBRA

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,
µµ ij ¶¶
n−j n−i A ∗
det (Bj ) = (−1) (−1) det
0 aij
µµ ij ¶¶
i+j A ∗
= (−1) det = aij cof (A)ij .
0 aij
Therefore,
n
X
det (A) = aij cof (A)ij
j=1

which is the formula for expanding det (A) along the ith row. Also,
n
¡ ¢ X ¡ ¢
det (A) = det AT = aTij cof AT ij
j=1
n
X
= aji cof (A)ji
j=1

which is the formula for expanding det (A) along the ith column. This proves the
theorem. ¥
Note that this gives an easy way to write a formula for the inverse of an n×n matrix.

Theorem
¡ ¢ 3.5.14 A−1 exists if and only if det(A) 6= 0. If det(A) 6= 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 (air ) = A, if det (A) 6= 0,

n
X
air cof (A)ir det(A)−1 = det(A) det(A)−1 = 1.
i=1

Now consider
n
X
air cof (A)ik det(A)−1
i=1

when k 6= r. Replace the k column with the rth column to obtain a matrix, Bk whose
th

determinant equals zero by Corollary 3.5.7. However, expanding this matrix along the
k th column yields
n
X
−1 −1
0 = det (Bk ) det (A) = air cof (A)ik det (A)
i=1

Summarizing,
n
X −1
air cof (A)ik det (A) = δ rk .
i=1

Using the other formula in Theorem 3.5.13, and similar reasoning,

n
X −1
arj cof (A)kj det (A) = δ rk
j=1
3.5. DETERMINANTS 45
¡ ¢
This proves that if det (A) 6= 0, then A−1 exists with A−1 = a−1
ij , where

−1
a−1
ij = cof (A)ji det (A) .

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) 6= 0. This proves the theorem. ¥

The next corollary points out that if an n × n matrix, A has a right or a left inverse,
then it has an inverse.

Corollary 3.5.15 Let A be an n×n matrix and suppose there exists an n×n 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 B det A = 1

and so det A 6= 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 find 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,
n
X n
X 1
xi = a−1
ij yj = cof (A)ji yj .
j=1 j=1
det (A)

By the formula for the expansion of a determinant along a column,

 
∗ ··· y1 ··· ∗
1  .. .. ..  ,
xi = det  . . . 
det (A)
∗ ··· yn ··· ∗

T
where here the ith column of A is replaced with the column vector, (y1 · · · ·, yn ) , and
the determinant of this modified matrix is taken and divided by det (A). This formula
is known as Cramer’s rule.
46 BASIC LINEAR ALGEBRA

Definition 3.5.16 A matrix M , is upper triangular if Mij = 0 whenever i > j.

Thus such a matrix equals zero below the main diagonal, the entries of the form Mii as
shown.  
∗ ∗ ··· ∗
 .. . 
 0 ∗
 . .. 

 . . 
 .. .. ... ∗ 
0 ··· 0 ∗
A lower triangular matrix is defined similarly as a matrix for which all entries above
the main diagonal are equal to zero.

With this definition, 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.

Definition 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 m × n 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 defined to be the
dimension of the span of the rows. The column rank is defined 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 = (aij ) equals r. If rows and columns
are interchanged, the determinant rank of the modified 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,
 
a11 · · · a1r a1p
 .. .. .. 
 . . . 
 
 ar1 · · · arr arp 
al1 · · · alr alp

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
r
X cof (M )ip
alp = − aip .
i=1
det (C)

Now note that cof (M )ip does not depend on p. Therefore the above sum is of the form
r
X
alp = mi aip
i=1

which shows the lth row is a linear combination of the first r rows of A. Since l is
arbitrary, This proves the theorem. ¥
3.5. DETERMINANTS 47

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 AT . The rows of AT are the
columns of A and the determinant rank of AT and A are the same. Therefore, from
Corollary 3.5.20, column rank of A = row rank of AT = determinant rank of AT =
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 n × n matrix. Then the following are equivalent.

1. det (A) = 0.

2. A, AT 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
th
columns by Theorem 3.5.19. In particular, it follows that for¡ some m, the m column¢
is a linear combination of all the others. Thus letting A = a1 · · · am · · · an
where the columns are denoted by ai , there exists scalars, αi such that
X
am = α k ak .
k6=m

¡ ¢T
Now consider the column vector, x ≡ α1 ··· −1 ··· αn . Then
X
Ax = −am + αk ak = 0.
k6=m

Since also A0 = 0, it follows A is not one to one. Similarly, AT is not one to one by the
same argument applied to AT . This verifies that 1.) implies 2.).
Now suppose 2.). Then since AT is not one to one, it follows there exists x 6= 0 such
that
AT x = 0.
Taking the transpose of both sides yields

xT A = 0

where the 0 is a 1 × n matrix or row vector. Now if Ay = x, then

2 ¡ ¢
|x| = xT (Ay) = xT A y = 0y = 0
48 BASIC LINEAR ALGEBRA

contrary to x 6= 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) 6= 0 but then from Theorem
3.5.14 A−1 exists and so for every y ∈ Fn there exists a unique x ∈ Fn 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) 6= 0.

2. A and AT 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 define the determinant of a linear transformation.

Definition 3.5.24 Let L ∈ L (V, V ) and let {v1 , · · ·, vn } be a basis for V . Thus
the matrix of L with respect to this basis is (lij ) ≡ ML where
X
L= lij vi vj
ij

Then define
det (L) ≡ det ((lij )) .

Proposition 3.5.25 The above definition is well defined.

¡0 ¢
Proof: Suppose {v10 , · · ·, vn0 } is another basis for V and lij ≡ ML0 is the matrix of
L with respect to this basis. Then by Theorem 3.3.9,

ML0 = S −1 ML S

for some matrix, S. Then by Theorem 3.5.10,

¡ ¢
det (ML0 ) = det S −1 det (ML ) det (S)
¡ ¢
= det S −1 S det (ML ) = det (ML )

because S −1 S = I and det (I) = 1. This shows the definition is well defined.
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 finite 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 {v1 , · · ·, vn } be a basis for V and let (lij ) be the matrix of
L with respect to this basis. By definition, det ((lijP
)) = 0 and so (lij ) is not one to one.
Thus there is a nonzero vector x ∈ Fn such that j lij xj = 0 for each i. Then letting
Pn
v ≡ j=1 xj vj ,
 
X n
X XX
Lv = lrs vr vs  x j vj  = lrs vr δ sj xj
rs j=1 j rs
 
X X
=  lrj xj  vr = 0
r j

Thus L is not one to one because L0 = 0 and Lv = 0.

Suppose 2.). Thus there exists v 6= 0 such that Lv = 0. Say
X
v= xi vi .
i

n
Then if {Lvi }i=1 were linearly independent, it would follow that
X
0 = Lv = xi Lvi
i

and so all the xi 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 (Lv1 , · · ·,Lvn )

and therefore, there is no u ∈ V such that Lu = w because if there were such a u, then
X
u= xi vi
i
P
and so Lu = i xi Lvi ∈ span (Lv1 , · · ·,Lvn ) .
Finally suppose L is not onto. Then (lij ) also cannot be onto Fn . Therefore,
n
det ((lij )) ≡ det (L) = 0. Why can’t
P (lij ) be onto? If it were, then for any y ∈ F ,
there exists x ∈ Fn such that yi = j lij xj . Thus
   
X X X X
yk vk = lrs vr vs  x j vj  = L  xj vj 
k rs j 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 finite dimensional vector space. For example, it could be a subspace of Cn or
Rn . Also suppose A ∈ L (V, V ) .

Definition 3.6.1 The characteristic polynomial of A is defined as q (λ) ≡ det (λ id −A)

where id is the identity map which takes every vector in V to itself. The zeros of q (λ)
in C are called the eigenvalues of A.
50 BASIC LINEAR ALGEBRA

Lemma 3.6.2 When λ is an eigenvalue of A which is also in F, the field of scalars,

then there exists v 6= 0 such that Av = λv.

Proof: This follows from Theorem 3.5.26. Since λ ∈ F,

λ id −A ∈ L (V, V )

and since it has zero determinant, it is not one to one so there exists v 6= 0 such that
(λ 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 finite dimen-
sional vector space of dimension n. Then there exists a polynomial of the form

p (λ) = λm + cm−1 λm−1 + · · · + c1 λ + c0

such that p (A) = 0 and m is as small as possible for this to occur.

2
Proof: Consider the linear transformations, I, A, A2 , · · ·, An . There are n2 + 1 of
these transformations and so by Theorem 3.3.5 the set is linearly dependent. Thus there
exist constants, ci ∈ F (either R or C) such that
2
n
X
c0 I + ck Ak = 0.
k=1

This implies there exists a polynomial, q (λ) which has the property that q (A) = 0.
Pn2
In fact, one example is q (λ) ≡ c0 + k=1 ck λk . Dividing by the leading term, it can
be assumed this polynomial is of the form λm + cm−1 λm−1 + · · · + c1 λ + c0 , 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 (λ) . This proves the lemma. ¥

Theorem 3.6.4 Let V be a nonzero finite dimensional vector space of dimension

n with the field of scalars equal to F which is either R or C. Suppose A ∈ L (V, V ) and
for p (λ) the minimal polynomial defined above, let µ ∈ F be a zero of this polynomial.
Then there exists v 6= 0, v ∈ V such that

Av = µv.

If F = C, then A always has an eigenvector and eigenvalue. Furthermore, if {λ1 , · · ·, λm }

are the zeros of p (λ) in F, these are exactly the eigenvalues of A for which there exists
an eigenvector in V.

Proof: Suppose first µ is a zero of p (λ) . Since p (µ) = 0, it follows

p (λ) = (λ − µ) k (λ)

where k (λ) is a polynomial having coefficients in F. Since p has minimal degree, k (A) 6=
0 and so there exists a vector, u 6= 0 such that k (A) u ≡ v 6= 0. But then

(A − µI) v = (A − µI) 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 (λ) is an eigenvalue which has an eigenvector
in V . Now suppose µ is an eigenvalue which has an eigenvector in V so that Av = µv
for some v ∈ V, v 6= 0. Does it follow µ is a zero of p (λ)?

0 = p (A) v = p (µ) v

and so µ is indeed a zero of p (λ). 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 field of scalars, F.
The idea of block multiplication turns out to be very useful later. For now here is an
interesting and significant application which has to do with characteristic polynomials.
In this theorem, pM (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
pBA (t) = tn−m pAB (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

µ ¶µ ¶ µ ¶
AB 0 I A AB ABA
=
B 0 0 I B BA
µ ¶µ ¶ µ ¶
I A 0 0 AB ABA
= .
0 I B BA B BA
Therefore,
µ ¶−1 µ ¶µ ¶ µ ¶
I A AB 0 I A 0 0
=
0 I B 0 0 I BA B
µ ¶ µ ¶
0 0 AB 0
Since the two matrices above are similar it follows that and
B BA B 0
have the same characteristic polynomials. Therefore, noting that BA is an n × n matrix
and AB is an m × m matrix,

tm det (tI − BA) = tn det (tI − AB)

and so det (tI − BA) = pBA (t) = tn−m det (tI − AB) = tn−m pAB (t) . This proves the
theorem. ¥

3.7 Exercises
1. Let M be an n × n matrix. Thus letting M x be defined by ordinary matrix
multiplication, it follows M ∈ L (Cn , Cn ) . 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 (λ) = p (λ) k (λ) for some polynomial k (λ) where q (λ) is the characteristic poly-
nomial. Now explain why q (M ) = 0. That every n × n matrix satisfies 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 Rn and examples of subsets of Rn which are not
subspaces.
52 BASIC LINEAR ALGEBRA

3. Let L ∈ L (V, W ) . Define 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 {w1 , · · ·, wk } are linearly independent and that
Lzi = wi . Show {z1 , · · ·, zk } is also linearly independent.

6. If L ∈ L (V, W ) and {z1 , · · ·, zk } is linearly independent, what is needed in order

that {Lz1 , · · ·, Lzk } be linearly independent? Explain your answer.

7. Let L ∈ L (V, W ). The rank of L is defined 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 (M L) ≤ min (rank (L) , rank (M )) .

9. Let M (t) = (b1 (t) , · · ·, bn (t)) where each bk (t) is a column vector whose com-
ponent functions are differentiable functions. For such a column vector,
T
b (t) = (b1 (t) , · · ·, bn (t)) ,

define
T
b0 (t) ≡ (b01 (t) , · · ·, b0n (t))
Show
n
X
0
det (M (t)) = det Mi (t)
i=1

where Mi (t) has all the same columns as M (t) except the ith column is replaced
with b0i (t).

10. Let A = (aij ) be an n × n matrix. Consider this as a linear transformation using

ordinary matrix multiplication. Show
X
A= aij ei ej
ij

where ei is the vector which has a 1 in the ith place and zeros elsewhere.

11. Let {w1 , · · ·, wn } be a basis for the vector space, V. Show id, the identity map is
given by X
id = δ ij wi wj
ij

3.8 Inner Product And Normed Linear Spaces

3.8.1 The Inner Product In Fn
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 difference of two
numbers. This must be generalized to Fn and to more general situations.
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 53

Definition 3.8.1 Let x, y ∈ Fn . Thus x = (x1 , · · ·, xn ) where each xk ∈ F and

a similar formula holding for y. Then the dot product of these two vectors is defined to
be X
x·y ≡ xj yj ≡ x1 y1 + · · · + xn yn .
j

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
difference 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 definition assures this will
take place. Thus X X 2
x·x= xj xj = |xj | ≥ 0.
j j

If you didn’t place a conjugate as in the above definition, things wouldn’t work out
correctly. For example,
2
(1 + i) + 22 = 4 + 2i
and this is not a positive number.
The following properties of the dot product follow immediately from the definition
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 (x·αy) = α (x · y) because

(x·αy) = (αy · x) = α (y · x) = α (x · y)

The norm is defined as follows.

Definition 3.8.2 For x ∈ Fn ,

à n
!1/2
X 2 1/2
|x| ≡ |xk | = (x · x)
k=1

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 θ ∈ F such that θz = |z| and |θ| = 1.

z
Proof: Let θ = 1 if z = 0 and otherwise, let θ = . Recall that for z = x + iy, z =
|z|
2
x − iy and zz = |z| . In case z is real, there is no change in the above.
54 BASIC LINEAR ALGEBRA

Theorem 3.8.4 (Cauchy Schwarz)Let H be an inner product space. The fol-

lowing inequality holds for x and y ∈ H.
1/2 1/2
|(x · y)| ≤ (x · x) (y · y) (3.33)

Equality holds in this inequality if and only if one vector is a multiple of the other.

Proof: Let θ ∈ F such that |θ| = 1 and

θ (x · y) = |(x · y)|
¡ ¢
Consider p (t) ≡ x + θty · x + tθy where t ∈ R. Then from the above list of properties
of the dot product,

0 ≤ p (t) = (x · x) + tθ (x · y) + tθ (y · x) + t2 (y · y)
= (x · x) + tθ (x · y) + tθ(x · y) + t2 (y · y)
= (x · x) + 2t Re (θ (x · y)) + t2 (y · y)
= (x · x) + 2t |(x · y)| + t2 (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,
1/2 1/2
|(x · y)| ≤ (x · x) (y · y) .
On the other hand, if (y · y) 6= 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
2
4 |(x · y)| − 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
2
to saying 4 |(x · y)| −4 (x · x) (y · y) = 0 and so from the quadratic formula, there exists
one real zero to p (t) = 0. Call it t0 . Then
¡ ¢ ¯ ¯2
p (t0 ) ≡ x + θt0 y, x + t0 θy = ¯x + θty¯ = 0

and so x = −θt0 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 defined.

Definition 3.8.5 Let z ∈ H. Then |z| ≡ (z · z)

1/2
.

Theorem 3.8.6 For length defined in Definition 3.8.5, the following hold.

|z| ≥ 0 and |z| = 0 if and only if z = 0 (3.35)

If α is a scalar, |αz| = |α| |z| (3.36)

|z + w| ≤ |z| + |w| . (3.37)

3.8. INNER PRODUCT AND NORMED LINEAR SPACES 55

Proof: The first two claims are left as exercises. To establish the third,
2
|z + w| ≡ (z + w, z + w)
= z·z+w·w+w·z+z·w
2 2
= |z| + |w| + 2 Re w · z
2 2
≤ |z| + |w| + 2 |w · z|
2 2 2
≤ |z| + |w| + 2 |w| |z| = (|z| + |w|) .

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 ||·|| satisfies 3.35 - 3.37. That is

||z|| ≥ 0 and ||z|| = 0 if and only if z = 0 (3.38)

If α is a scalar, ||αz|| = |α| ||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 Cn .

Definition 3.8.7 Let x ∈ Cn . Then define for p ≥ 1,

à n !1/p
X p
||x||p ≡ |xi |
i=1

The following inequality is called Holder’s inequality.

Proposition 3.8.8 For x, y ∈ Cn ,

n
à n
!1/p à n !1/p0
X X p
X p 0
|xi | |yi | ≤ |xi | |yi |
i=1 i=1 i=1

The proof will depend on the following lemma.

56 BASIC LINEAR ALGEBRA

Lemma 3.8.9 If a, b ≥ 0 and p0 is defined by 1

p + 1
p0 = 1, then

0
ap bp
ab ≤ + 0.
p p

Proof of the Proposition: If x or y equals the zero vector there is nothing to

Pn p 1/p
prove. Therefore, assume they are both nonzero. Let A = ( i=1 |xi | ) and B =
³P ´1/p 0
n p0
i=1 |yi | . Then using Lemma 3.8.9,

n n
" µ ¶p µ ¶p0 #
X |xi | |yi | X 1 |xi | 1 |yi |
≤ + 0
i=1
A B i=1
p A p B
n n
1 1 X p 1 1 X p0
= p
|xi | + 0 p
|yi |
p A i=1 p B i=1
1 1
= + =1
p p0

and so
n
à n
!1/p à n !1/p0
X X p
X p0
|xi | |yi | ≤ AB = |xi | |yi | .
i=1 i=1 i=1

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 pp0 = p − 1. Then using the Holder inequality,

n
X
p p
||x + y|| = |xi + yi |
i=1
n
X n
X
p−1 p−1
≤ |xi + yi | |xi | + |xi + yi | |yi |
i=1 i=1
Xn n
X
p p
= |xi + yi | p0 |xi | + |xi + yi | p0 |yi |
i=1 i=1
à n !1/p0 à n !1/p à n !1/p 
X p
X p
X p
≤ |xi + yi |  |xi | + |yi | 
i=1 i=1 i=1
³ ´
p/p0
= ||x + y|| ||x||p + ||y||p

p/p0
so dividing by ||x + y|| , it follows

p −p/p0
||x + y|| ||x + y|| = ||x + y|| ≤ ||x||p + ||y||p
³ ³ ´ ´
p 1
p− p0 =p 1− p0 = p p1 = 1. . This proves the theorem. ¥
It only remains to prove Lemma 3.8.9.
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 57

Proof of the lemma: Let p0 = q to save on notation and consider the following
picture:

x
b
x = tp−1
t = xq−1

t
a

Z a Z b
ap bq
ab ≤ tp−1 dt + xq−1 dx = + .
0 0 p q
Note equality occurs when ap = bq .
Alternate proof of the lemma: Let
µ ¶q
1 p 1 b
f (t) ≡ (at) + , t>0
p q t

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.
µ ¶q−1 µ ¶
p−1 b −b
f 0 (t) = (at) a+
t t2

Set it equal to 0. This happens when

bq
tp+q = . (3.42)
ap
Thus
bq/(p+q)
t=
ap/(p+q)
and so at this value of t,
µ ¶
q/(p+q) b p/(p+q)
at = (ab) , = (ab) .
t

Thus the minimum of f is

1³ q/(p+q)
´p 1 ³
p/(p+q)
´q
pq/(p+q)
(ab) + (ab) = (ab)
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
ap bq
ab ≤ + .
p q
Note that equality occurs when the minimum value happens for t = 1 and this indicates
from 3.42 that ap = bq . This proves the lemma. ¥
58 BASIC LINEAR ALGEBRA

3.8.5 Orthonormal Bases

Not all bases for an inner product space H are created equal. The best bases are
orthonormal.
Definition 3.8.11 Suppose {v1 , · · ·, vk } is a set of vectors in an inner product
space H. It is an orthonormal set if
½
1 if i = j
vi · vj = δ ij =
0 if i 6= j
Every orthonormal set of vectors is automatically linearly independent.
Proposition 3.8.12 Suppose {v1 , · · ·, vk } is an orthonormal set of vectors. Then
it is linearly independent.
Pk
Proof: Suppose i=1 ci vi = 0. Then taking dot products with vj ,
X X
0 = 0 · vj = ci vi · vj = ci δ ij = cj .
i i

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 {x1 , · · ·, xn } .
Then there exists an orthonormal basis for X, {u1 , · · ·, un } which has the property that
for each k ≤ n, span(x1 , · · ·, xk ) = span (u1 , · · ·, uk ) .
Proof: Let {x1 , · · ·, xn } be a basis for X. Let u1 ≡ x1 / |x1 | . Thus for k = 1,
span (u1 ) = span (x1 ) and {u1 } is an orthonormal set. Now suppose for some k <
n, u1 , · · ·, uk have been chosen such that (uj , ul ) = δ jl and span (x1 , · · ·, xk ) =
span (u1 , · · ·, uk ). Then define
Pk
xk+1 − j=1 (xk+1 · uj ) uj
uk+1 ≡ ¯¯ Pk ¯,
¯ (3.43)
¯xk+1 − j=1 (xk+1 · uj ) uj ¯

where the denominator is not equal to zero because the xj form a basis and so
xk+1 ∈
/ span (x1 , · · ·, xk ) = span (u1 , · · ·, uk )
Thus by induction,
uk+1 ∈ span (u1 , · · ·, uk , xk+1 ) = span (x1 , · · ·, xk , xk+1 ) .
Also, xk+1 ∈ span (u1 , · · ·, uk , uk+1 ) which is seen easily by solving 3.43 for xk+1 and
it follows
span (x1 , · · ·, xk , xk+1 ) = span (u1 , · · ·, uk , uk+1 ) .
¯ Pk ¯−1
¯ ¯
If l ≤ k, then denoting by C the scalar ¯xk+1 − j=1 (xk+1 · uj ) uj ¯ ,
 
k
X
(uk+1 · ul ) = C (xk+1 · ul ) − (xk+1 · uj ) (uj · ul )
j=1
 
k
X
= C (xk+1 · ul ) − (xk+1 · uj ) δ lj 
j=1

= C ((xk+1 · ul ) − (xk+1 · ul )) = 0.
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 59

n
The vectors, {uj }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 field of scalars, F. This is sometimes called the Riesz
representation theorem.

Theorem 3.8.14 Let H be a finite 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,

{e1 , · · ·, en } .

First note that if x is arbitrary, there exist unique scalars, xi such that
n
X
x= x i ei
i=1

Taking the dot product of both sides with ek yields

à n ! n n
X X X
(x · ek ) = xi ei · ek = xi (ei · ek ) = xi δ ik = xk
i=1 i=1 i=1

which shows that

n
X
x= (x · ei ) ei
i=1

and so by the properties of the dot product,

n
à n
!
X X
Lx = (x · ei ) Lei = x· ei Lei
i=1 i=1
Pn
so let z = i=1 ei Lei . It only remains to verify z is unique. However, this is obvious
because if (x · z1 ) = (x · z2 ) = Lx for all x, then

(x · z1 − z2 ) = 0

for all x and in particular for x = z1 − z2 which requires z1 = z2 . This proves the
theorem. ¥
Now with this theorem, it becomes easy to define something called the adjoint of
a linear operator. Let L ∈ L (H1 , H2 ) where H1 and H2 are finite dimensional inner
product spaces. Then letting ( · )i denote the inner product in Hi ,

x → (Lx · y)2
60 BASIC LINEAR ALGEBRA

is in L (H1 , F) and so from Theorem 3.8.14 there exists a unique element of H1 , denoted
by L∗ y such that for all x ∈ H1 ,

(Lx · y)2 = (x·L∗ y)1

Thus L∗ y ∈ H1 when y ∈ H2 . Also L∗ is linear. This is because by the properties of

the dot product,

(x·L∗ (αy + βz))1 ≡ (Lx·αy + βz)2

= α (Lx · y)2 + β (Lx · z)2
= α (x·L∗ y)1 + β (x·L∗ z)1
= α (x·L∗ y)1 + β (x·L∗ z)1

and
(x · αL∗ y + βL∗ z)1 = α (x·L∗ y)1 + β (x·L∗ z)1
Since
(x·L∗ (αy + βz))1 = (x · αL∗ y + βL∗ z)1
for all x, this requires
L∗ (αy + βz) = αL∗ y + βL∗ z.
In simple words, when you take it across the dot, you put a star on it. More precisely,
here is the definition.

Definition 3.8.15 Let H1 and H2 be finite dimensional inner product spaces

and let L ∈ L (H1 , H2 ) . Then L∗ ∈ L (H2 , H1 ) is defined by the formula

(Lx · y)2 = (x·L∗ y)1 .

In the case where H1 = H2 = 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.

L∗
H1 ← H2
L
H1 → H2

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. (x·Ly) = (L∗ x · y) , (Lx · y) = (x·L∗ y)
∗
2. (L∗ ) = L
∗
3. (aL + bM ) = aL∗ + bM ∗
∗
4. (M L) = L∗ M ∗

Proof: Consider the first claim.

(x·Ly) = (Ly · x) = (y·L∗ x) = (L∗ x · y)

This does the first claim. The second part was discussed earlier when the adjoint was
defined.
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 61

Consider the second claim. From the first claim,

¡ ∗ ¢
(Lx · y) = (x·L∗ 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 (M x · y)
and ¡ ¡ ¢ ¢
x· aL∗ + bM ∗ y = a (x·L∗ y) + b (x·M ∗ y) = a (Lx · y) + b (M x · 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· (M L) y = ((M L) x · y) = (Lx·M ∗ y) = (x·L∗ M ∗ 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 (H1 , H2 ) . Then F F ∗ ∈ L (H2 , H2 ) 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
∗ ∗
(F F ∗ ) = (F ∗ ) F ∗ = F F ∗
In the case where A ∈ L (Fn , Fm ) , considering the matrix of A with respect to the
usual bases, there is no loss of generality in considering A to be an m × n matrix,
X
(Ax)i = Aij xj .
j

Then in terms of components of the matrix, A,

(A∗ )ij = Aji .

You should verify this is so from the definition of the usual inner product on Fk . The
following little proposition is useful.
Proposition 3.8.18 Suppose A is an m × n matrix where m ≤ n. Also suppose
det (AA∗ ) 6= 0.
Then A has m linearly independent rows and m independent columns.
Proof: Since det (AA∗ ) 6= 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. ¥
62 BASIC LINEAR ALGEBRA

3.8.7 Schur’s Theorem

Recall that for a linear transformation, L ∈ L (V, V ) , it could be represented in the
form X
L= lij vi vj
ij

where {v1 , · · ·, v2 } is a basis. Of course different bases will yield different matrices,
(lij ) . Schur’s theorem gives the existence of a basis in an inner product space such that
(lij ) is particularly simple.

Definition 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 finite dimensional inner product

space such that the restriction of L∗ to every L invariant subspace has its eigenvalues in
n
F. Then there exist constants, cij for i ≤ j and an orthonormal basis, {wi }i=1 such
that
X n Xj
L= cij wi wj
j=1 i=1

The constants, cii 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 definition, 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 wn , an eigenvector for L∗ . Dividing by its length, it can be
assumed |wn | = 1. Say L∗ wn = µwn . Using the Gram Schmidt process, there exists an
orthonormal basis for H of the form {v1 , · · ·, vn−1 , wn } . Then

(Lvk · wn ) = (vk · L∗ wn ) = (vk · µwn ) = 0,

which shows
L : H1 ≡ span (v1 , · · ·, vn−1 ) → span (v1 , · · ·, vn−1 ) .
Denote by L1 the restriction of L to H1 . Since H1 has dimension n − 1, the induction
hypothesis yields an orthonormal basis, {w1 , · · ·, wn−1 } for H1 such that
n−1
XX j
L1 = cij wi wj . (3.44)
j=1 i=1

Then {w1 , · · ·, wn } is an orthonormal basis for H because every vector in

span (v1 , · · ·, vn−1 )

has the property that its dot product with wn is 0 so in particular, this is true for the
vectors {w1 , · · ·, wn−1 }. Now define cin to be the scalars satisfying
n
X
Lwn ≡ cin wi (3.45)
i=1

and let
j
n X
X
B≡ cij wi wj .
j=1 i=1
3.8. INNER PRODUCT AND NORMED LINEAR SPACES 63

Then by 3.45,
X j
n X n
X
Bwn = cij wi δ nj = cin wi = Lwn .
j=1 i=1 j=1

If 1 ≤ k ≤ n − 1,
X j
n X k
X
Bwk = cij wi δ kj = cik wi
j=1 i=1 i=1

while from 3.44,

n−1
XX j k
X
Lwk = L1 wk = cij wi δ jk = cik wi .
j=1 i=1 i=1

Since L = B on the basis {w1 , · · ·, wn } , it follows L = B.

It remains to verify the constants, ckk are the eigenvalues of L, solutions of the
equation, det (λI − L) = 0. However, the definition of det (λI − L) is the same as

det (λI − C)

where C is the upper triangular matrix which has cij for i ≤ j and zeros elsewhere.
This equals 0 if and only if λ is one of the diagonal entries, one of the ckk . 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 field 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 finite dimensional inner product space and

{w1 , · · ·, wn }

is an orthonormal basis for H. Then

∗
(wi wj ) = wj wi

Proof: It suffices to verify the two linear transformations are equal on {w1 , · · ·, wn } .
Then
¡ ∗ ¢
wp · (wi wj ) wk ≡ ((wi wj ) wp · wk ) = (wi δ jp · wk ) = δ jp δ ik
(wp · (wj wi ) wk ) = (wp · wj δ ik ) = δ ik δ jp

Since wp is arbitrary, it follows from the properties of the inner product that
¡ ∗ ¢
x · (wi wj ) wk = (x · (wj wi ) wk )
∗
for all x ∈ H and hence (wi wj ) wk = (wj wi ) wk . Since wk 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 {w1 , · · ·, wn } be an orthonormal basis for H and let (lij ) be the matrix
of L with respect to this orthonormal basis. Thus
X X
L= lij wi wj , id = δ ij wi wj
ij ij
64 BASIC LINEAR ALGEBRA

Denote by ML the matrix whose ij th entry is lij . Then by definition of what is meant
by the determinant of a linear transformation,

det (λ id −L) = det (λI − ML )

and so the eigenvalues of L are the same as the eigenvalues of ML . However, ML ∈

L (Cn , Cn ) with ML x determined by ordinary matrix multiplication. Therefore, by the
fundamental theorem of algebra and Theorem 3.6.4, if λ is an eigenvalue of L it follows
there exists a nonzero x ∈ Cn such that ML x = λx. Since L is self adjoint, it follows
from Lemma 3.8.21
X X X
L= lij wi wj = L∗ = lij wj wi = lji wi wj
ij ij ij

which shows lij = lji .

Then
2
X
λ |x| = λ (x · x) = (λx · x) = (ML x · x) = lij xj xi
ij
X X 2
= lij xj xi = lji xj xi = (ML x · x) = (x·ML x) = λ |x|
ij ij

showing λ = λ. 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) = (x·Ly) = (x·L∗ y)

so by the above lemma, the eigenvalues are real and are therefore, in the field 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 finite dimensional inner product space, real or

complex, and let L ∈ L (H, H) be self adjoint. Then there exists an orthonormal basis
{w1 , · · ·, wn } and real scalars, λk such that
n
X
L= λk wk wk .
k=1

The scalars are the eigenvalues and wk is an eigenvector for λk for each k.

Proof: By Theorem 3.8.20, there exists an orthonormal basis, {w1 , · · ·, wn } such

that
Xn X n
L= cij wi wj
j=1 i=1

where cij = 0 if i > j. Now using Lemma 3.8.21 and Proposition 3.8.16 along with the
assumption that L is self adjoint,
n X
X n n X
X n n X
X n
L= cij wi wj = L∗ = cij wj wi = cji wi wj
j=1 i=1 j=1 i=1 i=1 j=1
3.9. POLAR DECOMPOSITIONS 65

If i < j, then this shows cij = cji and the second number equals zero because j > i.
Thus cij = 0 if i < j and it is already known that cij = 0 if i > j. Therefore, let λk = ckk
and the above reduces to
n
X n
X
L= λj wj wj = λj wj wj
j=1 j=1

showing that λj = λj so the eigenvalues are all real. Now

n
X n
X
Lwk = λj wj wj (wk ) = λj wj δ jk = λk wk
j=1 j=1

which shows all the wk 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 finite 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,
2 2
|x| + |y| + (x · y) + (y · x)
2
= |x + y|
= (R (x + y) · R (x + y))
= (Rx·Rx) + (Ry·Ry) + (Rx · Ry) + (Ry · Rx)
2 2
= |x| + |y| + (R∗ Rx · y) + (y · R∗ Rx)

and so for all x, y,

(R∗ Rx − x · y) + (y·R∗ Rx − x) = 0
Hence for all x, y,
Re (R∗ Rx − x · y) = 0
Now for x, y given, choose α ∈ C such that

α (R∗ Rx − x · y) = |(R∗ Rx − x · y)|

Then

0 = Re (R∗ Rx − x·αy) = Re α (R∗ Rx − x · y)

= |(R∗ Rx − x · y)|
66 BASIC LINEAR ALGEBRA

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. ¥

Definition 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, {v1 , · · ·, vn } such that
n
X
F ∗F = λi vi vi , F ∗ F vk = λk vk .
i=1

It is also clear that λi ≥ 0 because

λi (vi · vi ) = (F ∗ F vi · vi ) = (F vi · F vi ) ≥ 0.

Let
n
X 1/2
U≡ λi vi vi .
i=1

so U maps X to X and is self adjoint. Then from 3.13,

X 1/2 1/2
U2 = λi λj (vi vi ) (vj vj )
ij
X n
X
1/2 1/2
= λi λj vi vj δ ij = λi vi vi = F ∗ F
ij i=1

Let {U x1 , · · ·, U xr } be an orthonormal basis for U (X) . Extend this using the Gram
Schmidt procedure to an orthonormal basis for X,

{U x1 , · · ·, U xr , yr+1 , · · ·, yn } .

Next note that {F x1 , · · ·, F xr } is also an orthonormal set of vectors in Y because

¡ ¢
(F xk · F xj ) = (F ∗ F xk · xj ) = U 2 xk · xj = (U xk · U xj ) = δ jk .
3.9. POLAR DECOMPOSITIONS 67

Now extend {F x1 , · · ·, F xr } to an orthonormal basis for Y,

{F x1 , · · ·, F xr , zr+1 , · · ·, zm } .

Since m ≥ n, there are at least as many zk as there are yk .

Now define R as follows. For x ∈ X, there exist unique scalars, ck and dk such that
r
X n
X
x= ck U xk + dk yk .
k=1 k=r+1

Then
r
X n
X
Rx ≡ ck F xk + dk zk . (3.46)
k=1 k=r+1

Thus, since {F x1 , · · ·, F xr , zr+1 , · · ·, zm } is orthonormal, a short computation shows

r
X n
X
2 2 2 2
|Rx| = |ck | + |dk | = |x| .
k=1 k=r+1

Now I need to verify RU x = F x. Since {U x1 , · · ·, U xr } is an orthonormal basis for

U X, there exist scalars, bk such that
r
X
Ux = bk U xk (3.47)
k=1

and so from the definition of R given in 3.46,

r
à r
!
X X
RU x ≡ bk F xk = F bk xk .
k=1 k=1
Pr
RU = F is shown if F ( k=1 bk xk ) = F (x).
à à r ! à r ! !
X X
F bk xk − F (x) · F bk xk − F (x)
k=1 k=1

à à r
! r
!
X X
∗
= F F bk xk − x · bk xk − x
k=1 k=1
à à r ! à r !!
X X
= U2 bk xk − x · bk xk − x
k=1 k=1
à à r ! à r !!
X X
= U bk xk − x ·U bk xk − x
k=1 k=1
à r r
!
X X
= bk U xk − U x · bk U xk − U x =0
k=1 k=1

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.
68 BASIC LINEAR ALGEBRA

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 = U R, RR∗ = I.
∗
Proof: Recall that L∗∗ = L and (M L) = L∗ M ∗ . 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 R = I. This proves the corollary. ¥
This is a good place to consider a useful lemma.
Lemma 3.9.5 Let X be a finite dimensional inner product space of dimension n and
let R ∈ L (X, X) be unitary. Then |det (R)| = 1.
Proof: Let {wk } be an orthonormal basis for X. Then to take the determinant it
suffices to take the determinant of the matrix, (cij ) where
X
R= cij wi wj .
ij
P
Rwk = i cik wi and so
(Rwk , wl ) = clk .
and hence X
R= (Rwk , wl ) wl wk
lk
Similarly X
R∗ = (R∗ wj , wi ) wi wj .
ij
Since R is given to be unitary,
XX
RR∗ = id = (Rwk , wl ) (R∗ wj , wi ) (wl wk ) (wi wj )
lk ij
X
= (Rwk , wl ) (R∗ wj , wi ) δ ki wl wj
ijkl
à !
X X
∗
= (Rwi , wl ) (R wj , wi ) wl wj
jl i

Hence X X
(R∗ wj , wi ) (Rwi , wl ) = δ jl = (Rwi , wj ) (Rwi , wl ) (3.48)
i i
because X
id = δ jl wl wj
jl

Thus letting M be the matrix whose ij th entry is (Rwi , wl ) , det (R) is defined as
det (M ) and 3.48 says
X³ ´
MT Mil = δ jl .
ji
i
It follows
³ T´ ¡ ¢ 2
1 = det (M ) det M = det (M ) det M = det (M ) det (M ) = |det (M )| .

Thus |det (R)| = |det (M )| = 1 as claimed.

3.10. EXERCISES 69

3.10 Exercises
1. For u, v vectors in F3 , define the product, u ∗ v ≡ u1 v1 + 2u2 v2 + 3u3 v3 . Show the
axioms for a dot product all hold for this funny product. Prove
1/2 1/2
|u ∗ v| ≤ (u ∗ u) (v ∗ v) .

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 Schur’s theorem? Hint:
Start with a basis and decree the basis is orthonormal. Then define an inner
product accordingly.
h i
2 2
3. Show that (a · b) = 14 |a + b| − |a − b| .
2
4. Prove from the axioms of the dot product the parallelogram identity, |a + b| +
2 2 2
|a − b| = 2 |a| + 2 |b| .
5. Suppose f, g are two Darboux Stieltjes integrable functions defined on [0, 1] . Define
Z 1
(f · g) = f (x) g (x)dF.
0

Show this dot product satisfies 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 Z 1
(f · g) = f (x) g (x)p (x) dF
0
where p (x) is a given nonnegative function? If so, what would it be in terms of
integrals.
6. If A is an n × n matrix considered as an element of L (Cn , Cn ) by ordinary matrix
multiplication, use the inner product in Cn to show that (A∗ )ij = Aji . In words,
the adjoint is the transpose of the conjugate.
7. A symmetric matrix is a real n × n matrix A which satisfies AT = A. Show every
symmetric matrix is self adjoint and that there exists an orthonormal set of real
vectors {x1 , · · ·, xn } such that
X
A= λk xk xk
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 Cn , {x1 , · · ·, xn } such that
X
A= ai xi xi
i

That is, with respect to this basis the matrix of A is diagonal. Hint: This is a
Pn was done to prove Theorem 3.8.23. Use Schur’s theorem
harder versionPof what
n
to write A = j=1 i=1 Bij wi wj where Bij is an upper triangular matrix. Then
use the condition that A is normal and eventually get an equation
X X
Bik Blk = Bki Bkl
k k

Next let i = l and consider first l = 1, then l = 2, etc. If you are careful, you will
find Bij = 0 unless i = j.
70 BASIC LINEAR ALGEBRA

9. Suppose A ∈ L (H, H) where H is an inner product space and

X
A= ai wi wi
i

where the vectors {w1 , · · ·, wn } are an orthonormal set. Show that A must be
normal. In other words, you can’t represent A ∈ L (H, H) in this very convenient
way unless it is normal.
10. If L is a self adjoint operator defined on an inner product space, H such that
L has all only nonnegative eigenvalues. Explain how to define L1/n and show
why what you come up with is indeed the nth root of the operator. For a
self adjoint operator L on an inner product space, can you define sin (L) ≡
P∞ k 2k+1
k=0 (−1) L / (2k + 1)!? What does the infinite 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 defined 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 = U R where F = RU is the right polar decomposition defined
above. Recall R preserves distances and U is self adjoint. What is the geometric
significance of a linear transformation being normal?
13. Suppose you have a basis, {v1 , · · ·, vn } in an inner product space, X. The Gram-
mian matrix is the n × n matrix whose ij th entry is (vi · vj ) . Show this matrix is
invertible.
P Hint:
P You might try to show that the inner product of two vectors,
a
k k kv and k bk vk has something to do with the Grammian.

14. Suppose you have a basis,

© {v1 , · ·ª·, vn } in an inner product space, X. Show there
exists a “dual basis” v1 , · · ·, vn which satisfies vk · vj = δ kj , which equals 0 if
j 6= k and equals 1 if j = k.
Sequences

4.1 Vector Valued Sequences And Their Limits

Functions defined 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 Fn , although
everything can be done for an arbitrary vector space and when it creates no difficulties,
I will state certain definitions 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 difference 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,
Definition 4.1.1 Let {an } be a sequence and let n1 < n2 < n3 , · · · be any
strictly increasing list of integers such that n1 is at least as large as the first number in
the domain of the function. Then if bk ≡ ank , {bk } is called a subsequence of {an } .
¡ ¡ ¢¢ ∞
Example 4.1.2 Let an = n + 1, sin n1 . Then {an }n=1 is a vector valued sequence.
The definition of a limit of a vector valued sequence is given next. It is just like the
definition given for sequences of scalars. However, here the symbol |·| refers to the usual
norm in Fn . In a general normed vector space, it will be denoted by ||·|| .
Definition 4.1.3 ∞
A vector valued sequence {an }n=1 converges to a in a normed
vector space V, written as
lim an = a or an → a
n→∞
if and only if for every ε > 0 there exists nε such that whenever n ≥ nε ,
||an − a|| < ε.
In words the definition says that given any measure of closeness ε, the terms of the
sequence are eventually this close to a. Here, the word “eventually” refers to n being
sufficiently large.
Theorem 4.1.4 If limn→∞ an = a and limn→∞ an = a1 then a1 = a.
Proof: Suppose a1 6= a. Then let 0 < ε < ||a1 − a|| /2 in the definition of the limit.
It follows there exists nε such that if n ≥ nε , then ||an − a|| < ε and |an − a1 | < ε.
Therefore, for such n,
||a1 − a|| ≤ ||a1 − an || + ||an − a||
< ε + ε < ||a1 − a|| /2 + ||a1 − a|| /2 = ||a1 − a|| ,
a contradiction.

71
72 SEQUENCES

Theorem 4.1.5 Suppose {an } and {bn } are vector valued sequences and that

lim an = a and lim bn = b.

n→∞ n→∞

Also suppose x and y are scalars in F. Then

lim xan + ybn = xa + yb (4.1)

n→∞

Also,
lim (an · bn ) = (a · b) (4.2)
n→∞

If {xn } is a sequence of scalars in F converging to x and if {an } is a sequence of vectors

in Fn converging to a, then
lim xn an = xa. (4.3)
n→∞

Also if {xk } is a sequence of vectors in Fn then xk → x, if and only if for each j,

lim xjk = xj . (4.4)

k→∞

where here ¡ ¢ ¡ ¢
xk = x1k , · · ·, xnk , x = x1 , · · ·, xn .

Proof: Consider the first claim. By the triangle inequality

||xa + yb− (xan + ybn )|| ≤ |x| ||a − an || + |y| ||b − bn || .

By definition, there exists nε such that if n ≥ nε ,

ε
||a − an || , ||b − bn || <
2 (1 + |x| + |y|)

so for n > nε ,
ε ε
||xa + yb− (xan + ybn )|| < |x| + |y| ≤ ε.
2 (1 + |x| + |y|) 2 (1 + |x| + |y|)

Now consider the second. Let ε > 0 be given and choose n1 such that if n ≥ n1 then

|an − a| < 1.

For such n, it follows from the Cauchy Schwarz inequality and properties of the inner
product that

|an · bn − a · b| ≤ |(an · bn ) − (an · b)| + |(an · b) − (a · b)|

≤ |an | |bn − b| + |b| |an − a|
≤ (|a| + 1) |bn − b| + |b| |an − a| .

Now let n2 be large enough that for n ≥ n2 ,

ε ε
|bn −b| < , and |an −a| < .
2 (|a| + 1) 2 (|b| + 1)

Such a number exists because of the definition of limit. Therefore, let

nε > max (n1 , n2 ) .

4.1. VECTOR VALUED SEQUENCES AND THEIR LIMITS 73

For n ≥ nε ,
|an · bn − a · b| ≤ (|a| + 1) |bn − b| + |b| |an − a|
ε ε
< (|a| + 1) + |b| ≤ ε.
2 (|a| + 1) 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 definition of distance in
Fn , v
uX
u n ³ ´2
lim |x − xk | ≡ lim t xj − xjk = 0.
k→∞ k→∞
j=1
¯ ¯
¯ ¯
On the other hand, if limk→∞ |x − xk | = 0, then since ¯xjk − xj ¯ ≤ |x − xk | , it follows
from the squeezing theorem that
¯ ¯
¯ ¯
lim ¯xjk − xj ¯ = 0.
k→∞

This proves the theorem. ¥

An important theorem is the one which states that if a sequence converges, so does
every subsequence. You should review Definition 4.1.1 at this point. The proof is
identical to the one involving sequences of numbers.
Theorem 4.1.6 Let {xn } be a vector valued sequence with limn→∞ xn = x and
let {xnk } be a subsequence. Then limk→∞ xnk = x.
Proof: Let ε > 0 be given. Then there exists nε such that if n > nε , then ||xn −x|| <
ε. Suppose k > nε . Then nk ≥ k > nε and so
||xnk −x|| < ε
showing limk→∞ xnk = x as claimed.

Theorem 4.1.7 Let {xn } be a sequence of real numbers and suppose each xn ≤ l
(≥ l)and limn→∞ xn = x. Then x ≤ l (≥ l) . More generally, suppose {xn } and {yn } are
two sequences such that limn→∞ xn = x and limn→∞ yn = y. Then if xn ≤ yn for all n
sufficiently large, then x ≤ y.
Proof: Let ε > 0 be given. Then for n large enough,
l ≥ xn > x − ε
and so
l + ε ≥ x.
Since ε > 0 is arbitrary, this requires l ≥ x. The other case is entirely similar or else
you could consider −l and {−xn } and apply the case just considered.
Consider the last claim. There exists N such that if n ≥ N then xn ≤ yn and
|x − xn | + |y − yn | < ε/2.
Then considering n > N in what follows,
x − y ≤ xn + ε/2 − (yn − ε/2) = xn − yn + ε ≤ ε.
Since ε was arbitrary, it follows
x − y ≤ 0.
This proves the theorem. ¥
74 SEQUENCES

Theorem 4.1.8 Let {xn } be a sequence vectors and suppose each ||xn || ≤ l
(≥ l)and limn→∞ xn = x. Then x ≤ l (≥ l) . More generally, suppose {xn } and {yn }
are two sequences such that limn→∞ xn = x and limn→∞ yn = y. Then if ||xn || ≤ ||yn ||
for all n sufficiently large, then ||x|| ≤ ||y|| .

Proof: It suffices to just prove the second part since the first part is similar. By
the triangle inequality,
|||xn || − ||x||| ≤ ||xn − x||
and for large n this is given to be small. Thus {||xn ||} converges to ||x|| . Similarly
{||yn ||} 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 definition of sequential compactness. It is a very useful notion
which can be used to prove existence theorems.

Definition 4.2.1 A set, K ⊆ V, a normed vector space is sequentially compact

if whenever {an } ⊆ K is a sequence, there exists a subsequence, {ank } 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 Ik = ak , bk and suppose that for all k = 1, 2, · · ·,
Ik ⊇ Ik+1 .

Then there exists a point, c ∈ R which is an element of every Ik .

Proof: Since Ik ⊇ Ik+1 , this implies

ak ≤ ak+1 , bk ≥ bk+1 . (4.5)

Consequently, if k ≤ l,
al ≤ al ≤ bl ≤ bk . (4.6)
Now define © ª
c ≡ sup al : l = 1, 2, · · ·
By the first inequality in 4.5, and 4.6
© ª
ak ≤ c = sup al : l = k, k + 1, · · · ≤ bk (4.7)

for each k = 1, 2 · · · . Thus c ∈ Ik for every k and This proves the lemma. ¥ If this went
too©fast, the reason for ªthe last inequality in 4.7 is that from 4.6, bk is an upper bound
to al : 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 {xn } ⊆ [a, b] ≡ I0 . 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 xn for infinitely
many values of n. Call this interval I1 . Now do for I1 what was done for I0 . Split it
in half and let I2 be the interval which contains xn for infinitely many values of n.
Continue this way obtaining a sequence of nested intervals I0 ⊇ I1 ⊇ I2 ⊇ I3 · ·· where
the length of In is (b − a) /2n . Now pick n1 such that xn1 ∈ I1 , n2 such that n2 > n1
4.3. CLOSED AND OPEN SETS 75

and xn2 ∈ I2 , n3 such that n3 > n2 and xn3 ∈ I3 , etc. (This can be done because in each
case the intervals contained xn for infinitely many values of n.) By the nested interval
lemma there exists a point, c contained in all these intervals. Furthermore,
|xnk − c| < (b − a) 2−k
and so limk→∞ xnk = c ∈ [a, b] . This proves the theorem. ¥

Theorem 4.2.4 Let

n
Y
I= Kk
k=1
where Kk is a sequentially compact set in F. Then I is a sequentially compact set in Fn .
∞
Proof: Let {xk }k=1 be a sequence of points in I. Let
¡ ¢
xk = x1k , · · ·, xnk
© ª∞
Thus xik k=1 is a sequence of points in Ki . Since Ki is sequentially compact, there exists
∞ © ª
a subsequence of {xk }k=1 denoted by {x1k } such that x11k converges 1
© 1 ª to x for some
x ∈ K1 . Now there exists a further subsequence, {x2k } such that x2k converges to x1 ,
1

because by Theorem 4.1.6, subsequences of convergent © sequences

ª converge to the same
limit as the convergent sequence, and in addition, x22k convergesn to o some x2 ∈ K2 .
∞
Continue taking subsequences such that for {xjk }k=1 , it follows xrjk converges to
r ∞
some x ∈ Kr for all r ≤ j. Then {xnk }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 xj ∈ Kj . It follows from Theorem 4.1.5 that x ≡ x1 , · · ·, xn ∈ I
∞
and is the limit of {xnk }k=1 . This proves the theorem. ¥
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] .
∞
{xk + iyk }k=1 ⊆ [a, b] + i [c, d]
is the same as saying (xk , yk ) ∈ [a, b] × [c, d] . Therefore, there exists (x, y) ∈ [a, b] × [c, d]
such that xk → x and yk → y. In other words xk +iyk → x+iy and x+iy ∈ [a, b]+i [c, d] .
This proves the corollary. ¥

4.3 Closed And Open Sets

The definition of open and closed sets is next.
Definition 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 sufficiently small, it follows
x ∈ U also. The set of points, x which are closer to p than δ is denoted by
B (p, δ) ≡ {x ∈ V : ||x − p|| < δ} .
This symbol, B (p, δ) is called an open ball of radius δ. Thus a point, p is an interior
point of U if there exists δ > 0 such that p ∈ B (p, δ) ⊆ 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 finite 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 finite 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 {U1 , · · ·, Um } be some open sets and suppose p ∈ ∩mk=1 Uk . Then there exists
rk > 0 such that B (p, rk ) ⊆ Uk . Let 0 < r ≤ min (r1 , r2 , · · ·, rm ) . Then B (p, r) ⊆
∩mk=1 Uk and so the finite intersection is open. Note that if the finite 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 aren’t any such points.
Suppose {H1 , · · ·, Hm } is a finite set of closed sets. Then ∪m k=1 Hk is closed if its
complement is open. However, from DeMorgan’s laws,
C
(∪m m C
k=1 Hk ) = ∩k=1 Hk ,

a finite intersection of open sets which is open by what was just shown.
Next let C be some collection of closed sets. Then
C © ª
(∩C) = ∪ H C : H ∈ C ,

a union of open sets which is therefore open by the first 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.

Definition 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, δ) , an open ball in a normed vector space. Then
every point of B (x, δ) is a limit point. There are more general situations than normed
vector spaces in which this assertion is false.

If z ∈ B (x, δ) , consider z+ k1 (x − z) ≡ wk for k ∈ N. Then

¯¯ ¯¯
¯¯ 1 ¯¯
||wk − x|| = ¯¯z+ (x − z) − x¯¯¯¯
¯ ¯
k
¯¯µ ¶ µ ¶ ¯¯
¯¯ 1 1 ¯¯
¯ ¯
= ¯¯ 1 − z− 1 − x¯¯¯¯
k k
k−1
= ||z − x|| < δ
k
and also
1
||wk − z|| ≤ ||x − z|| < δ/k
k
so wk → z. Furthermore, the wk are distinct. Thus z is a limit point of A as claimed.
This is because every ball containing z contains infinitely many of the wk and since
they are all distinct, they can’t 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, {an }
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 first a is a limit point of A. There exists a1 ∈ B (a, 1) ∩ A such that
a1 6= a. Now supposing distinct points, a1 , · · ·, an have been chosen such that none are
equal to a and for each k ≤ n, ak ∈ B (a, 1/k) , let
½ ¾
1
0 < rn+1 < min , ||a − a1 || , · · ·, ||a − an || .
n+1

Then there exists an+1 ∈ B (a, rn+1 ) ∩ A with an+1 6= a. Because of the definition of
rn+1 , an+1 is not equal to any of the other ak for k < n+1. Also since ||a − am || < 1/m,
it follows limm→∞ am = a. Conversely, if there exists a sequence of distinct points of A
converging to a, then B (a, r) contains all an for n large enough. Thus B (a, r) contains
infinitely many points of A since all are distinct. Thus at least one of them is not equal
to a. This establishes the first part of the theorem.
Now consider the second claim. If A is closed then it is the complement of an open
set. Since AC is open, it follows that if a ∈ AC , then there exists δ > 0 such that
B (a, δ) ⊆ AC and so no point of AC 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
AC does not contain any limit points of A. It also contains no points of A. Therefore, if
a ∈ AC , since it is not a limit point of A, there exists δ > 0 such that B (a, δ) contains
no points of A different than a. However, a itself is not in A because a ∈ AC . Therefore,
B (a, δ) is entirely contained in AC . Since a ∈ AC was arbitrary, this shows every point
of AC is an interior point and so AC 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 {xn } ⊆ H. Then since K is sequentially compact, there is a subsequence,

{xnk } 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 xnk → x which requires xnk ∈ B (x, r) for all k large enough.
Thus x ∈ H and this has shown H is sequentially compact.

Definition 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 Fn is sequentially compact.

Conversely, every sequentially compact set in Fn is closed and bounded.

Proof: Let H be a closed and bounded set in Fn . Then H ⊆ B (0, r) for some r.
Therefore, if x ∈ H, x = (x1 , · · ·, xn ) , it must be that
v
u n
uX
t 2
|xi | < r
i=1

and so each xi ∈ [−r, r] + i [−r, r] ≡ Rr , a sequentially compact set by Corollary 4.2.5.

Thus H is a closed subset of
Yn
Rr
i=1

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 Fn . If it is not bounded, then
there exists a sequence, {km } such that km ∈ K but km ∈ / B (0, m) for m = 1, 2, · · ·.
78 SEQUENCES

However, this sequence cannot have any convergent subsequence because if kmk → k,
C
then for large enough m, k ∈ B (0,m) ⊆ D (0, m) and kmk ∈ B (0, m) for all k large
enough and this is a contradiction because there can only be finitely 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
1
limit point of K which is not in K. Pick km ∈ B k∞ , m . Then {km } 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 {km } , a contradiction. This proves
the theorem. ¥
What are some examples of closed and bounded sets in a general normed vector
space and more specifically Fn ?

Proposition 4.3.9 Let D (z, r) denote the set of points,

{w ∈ V : ||w − z|| ≤ r}

Then D (z, r) is closed and bounded. Also, let S (z,r) denote the set of points

{w ∈ V : ||w − z|| = r}

Then S (z, r) is closed and bounded. It follows that if V = Fn ,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.

C C
For the second type of set, note S (z, r) = B (z, r) ∪ D (z, r) , 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,

Definition 4.4.1 {an } is a Cauchy sequence in a normed vector space, V if for

all ε > 0, there exists nε such that whenever n, m ≥ nε ,

||an −am || < ε.

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 ε = 1 in the definition of a Cauchy sequence and let n > n1 . Then from
the definition,
||an −an1 || < 1.
It follows that for all n > n1 ,
||an || < 1 + ||an1 || .
Therefore, for all n,
n1
X
||an || ≤ 1 + ||an1 || + ||ak || .
k=1

This proves the theorem. ¥

Theorem 4.4.3 If a sequence {an } in V, a normed vector space converges, then

the sequence is a Cauchy sequence.

Proof: Let ε > 0 be given and suppose an → a. Then from the definition of
convergence, there exists nε such that if n > nε , it follows that
ε
||an −a|| <
2
Therefore, if m, n ≥ nε + 1, it follows that
ε ε
||an −am || ≤ ||an −a|| + ||a − am || < + =ε
2 2
showing that, since ε > 0 is arbitrary, {an } 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.

Theorem 4.4.4 Suppose {an } is a Cauchy sequence in any normed vector space
and there exists a subsequence, {ank } which converges to a. Then {an } also converges
to a.

Proof: Let ε > 0 be given. There exists N such that if m, n > N, then

||am −an || < ε/2.

Also there exists K such that if k > K, then

||a − ank || < ε/2.

Then let k > max (K, N ) . Then for such k,

||ak −a|| ≤ ||ak −ank || + ||ank −a||

< ε/2 + ε/2 = ε.

This proves the theorem. ¥

Definition 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 {xk + iyk }k=1 be a Cauchy sequence in C. This requires {xk } and {yk }
are both Cauchy sequences in R. This follows from the obvious estimates

|xk − xm | , |yk − ym | ≤ |(xk + iyk ) − (xm + iym )| .

By completeness of R there exists x ∈ R such that xk → x and similarly there exists

y ∈ R such that yk → y. Therefore, since
q
2 2
|(xk + iyk ) − (x + iy)| ≤ (xk − x) + (yk − y)
≤ |xk − x| + |yk − y|

it follows (xk + iyk ) → (x + iy) .¥

A simple generalization of this idea yields the following theorem.

Theorem 4.4.8 Fn is complete.

Proof: By 4.4.7, F is complete. Now let {am } be a Cauchy sequence in Fn . Then

by the definition of the norm
¯ ¯
¯ j j¯
¯am − ak ¯ ≤ |am − ak |
© ª∞
where ajm denotes the j th component of am . Thus for each j = 1, 2, · · ·, n, ajm m=1 is
a Cauchy sequence. It follows from Theorem 4.4.7, the completeness of F, there exists
aj such that
lim ajm = aj
m→∞

Theorem 4.1.5 implies that limm→∞ am = a where

¡ ¢
a = a1 , · · ·, an .

This proves the theorem. ¥

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.

Definition 4.5.1 Let S be a nonempty set in a normed vector space, V . Then

diam (S) is defined as

diam (S) ≡ sup {||x − y|| : x, y ∈ S} .

This is called the diameter of S.

Theorem 4.5.2 ∞
Let {Fn }n=1 be a sequence of closed sets in Fn such that

lim diam (Fn ) = 0

n→∞

and Fn ⊇ Fn+1 for each n. Then there exists a unique p ∈ ∩∞

k=1 Fk .
4.6. EXERCISES 81

Proof: Pick pk ∈ Fk . This is always possible because by assumption each set is

∞
nonempty. Then {pk }k=m ⊆ Fm and since the diameters converge to 0, it follows
{pk } is a Cauchy sequence. Therefore, it converges to a point, p by completeness of
Fn discussed in Theorem 4.4.8. Since each Fk is closed, p ∈ Fk for all k. This is because
it is a limit of a sequence of points only finitely many of which are not in the closed set
Fk . Therefore, p ∈ ∩∞ ∞
k=1 Fk . If q ∈ ∩k=1 Fk , then since both p, q ∈ Fk ,

|p − q| ≤ diam (Fk ) .

It follows since these diameters converge to 0, |p − q| ≤ ε for every ε. Hence p = q. This

proves the theorem. ¥
A sequence of sets, {Gn } which satisfies Gn ⊇ Gn+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, define 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 Fn or more generally in a normed vector space. Define

the closure of A to equal the intersection of all closed sets which contain A. This
is usually denoted by A. Show A = A ∪ A0 where A0 consists of the set of limit
points of A. Also explain why A is closed.

3. The interior of a set was defined above. Tell why the interior of a set is always an
open set. The interior of a set A is sometimes denoted by A0 .

4. Given an example of a set A whose interior is empty but whose closure is all of
Rn .

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 AC . Sometimes this is denoted
as ∂A. In a normed vector space, is it always the case that A ∪ ∂A = A? Prove or
disprove.

6. Give an example of a finite dimensional normed vector space where the field of
scalars is the rational numbers which is not complete.

7. Explain why as far as the theorems of this chapter are concerned, Cn is essentially
the same as R2n .

8. A set, A ⊆ Rn 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 Rn . The convex hull of A, usually denoted by

co (A) is defined as the set of
Ppall convex combinations of pointsPin A. A convex
combination is of the form k=1 tk ak where each tk ≥ 0 and k tk = 1. Note
that p can be any finite number. Show co (A) is convex.
82 SEQUENCES

Pp P
10. Suppose A ⊆ Rn and z ∈ co (A) . Thus z = k=1 tk ak for tk ≥ 0 and k tk =
1. Show there exists n + 1 of the points {a1 , · · ·, ap } such that z is a convex
combination of these n + 1 points. Hint: Show that if p > n + 1 then the vectors
p
{ak − a1 }k=2 must be P linearly dependent. Conclude from this Pp the existence of
p
scalars {αi } such that i=1 αi ai = 0. Now for s ∈ R, z = k=1 (tk + sαi ) ak .
Consider small s and adjust till one or more of the tk + sαk vanish. Now you are
in the same situation as before but with only p − 1 of the ak . Repeat the argument
till you end up with only n + 1 at which time you can’t repeat again.
11. Show that any uncountable set of points in Fn must have a limit point.
12. Let V be any finite dimensional vector space having a basis {v1 , · · ·, vn } . For
x ∈ V, let
Xn
x= xk vk
k=1

so that the scalars, xk are the components of x with respect to the given basis.
Define for x, y ∈ V
n
X
(x · y) ≡ xi yi
i=1

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
à !1/2
X 2
|x| ≡ |xi |
i

where X
x= xk vk . (4.8)
k

Show there exist positive numbers δ < ∆ independent of x such that

δ |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 {xk } such that |xk | = 1 but
k −1 |xk | = k −1 ≥ ||xk || and then note the vector of components of xk 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 {v1 , · · ·, vn } being a basis.
14. It was shown above that in Fn , the sequentially compact sets are exactly those
which are closed and bounded. Show that in any finite dimensional normed vector
space, V the closed and bounded sets are those which are sequentially compact.
15. Two norms on a finite dimensional vector space, ||·||1 and ||·||2 are said to be
equivalent if there exist positive numbers δ < ∆ such that

δ ||x||1 ≤ ||x||2 ≤ ∆ ||x1 ||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 finite dimensional
vector space are equivalent.
4.6. EXERCISES 83

∞
16. A normed vector space, V is separable if there is a countable set {wk }k=1 such
that whenever B (x, δ) is an open ball in V, there exists some wk in this open ball.
Show that Fn 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 (xk , rk )}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, {xk }k=1 and
1
then consider balls of the form B xk , 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 finite 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, {Uk }k=1 such that S ⊆ ∪∞
k=1 Uk . This is called the Lindeloff 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 finitely many sets of C, {U1 , · · ·, Up } such
that
H ⊆ ∪pi=1 Ui .
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 finite dimensional normed vector space
84 SEQUENCES
Continuous Functions

Continuous functions are defined as they are for a function of one variable.

Definition 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 ε > 0 there exists δ > 0 such that
whenever y ∈ D (f ) and
||y − x||V < δ
it follows that
||f (x) −f (y)||W < ε.
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 definition. 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 Fn and they are each continuous at x, then f · g is
continuous at x. If g has values in F and g (x) 6= 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 ε > 0 be given. By assumption, there exist δ 1 > 0 such
ε
that whenever |x − y| < δ 1 , it follows |f (x) − f (y)| < 2(|a|+|b|+1) and there exists δ 2 > 0
ε
such that whenever |x − y| < δ 2 , it follows that |g (x) − g (y)| < 2(|a|+|b|+1) . Then let
0 < δ ≤ min (δ 1 , δ 2 ) . If |x − y| < δ, 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)|

µ ¶ µ ¶
ε ε
< |a| + |b| < ε.
2 (|a| + |b| + 1) 2 (|a| + |b| + 1)

Now consider 2.) There exists δ 1 > 0 such that if |y − x| < δ 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 ε > 0 be given. There exists δ 2 such that if |x − y| < δ 2 , then
ε
|g (x) − g (y)| < ,
2 (2 + |g (x)| + |f (x)|)

and there exists δ 3 such that if |x − y| < δ 3 , then

ε
|f (x) − f (y)| <
2 (2 + |g (x)| + |f (x)|)

Now let 0 < δ ≤ min (δ 1 , δ 2 , δ 3 ) . Then if |x − y| < δ, 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)|]

µ ¶
ε ε
< (2 + |g (x)| + |f (x)|) + = ε.
2 (2 + |g (x)| + |f (x)|) 2 (2 + |g (x)| + |f (x)|)

This proves the first part of 2.) To obtain the second part, let δ 1 be as described above
and let δ 0 > 0 be such that for |x − y| < δ 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 (δ 0 , δ 1 ) ,
¯ ¯ ¯ ¯
¯ f (x) f (y) ¯¯ ¯¯ f (x) g (y) − f (y) g (x) ¯¯
¯ − =
¯ g (x) g (y) ¯ ¯ g (x) g (y) ¯
|f (x) g (y) − f (y) g (x)|
≤ ³ 2
´
|g(x)|
2

2 |f (x) g (y) − f (y) g (x)|

= 2
|g (x)|
 87

2
≤ 2 [|f (x) g (y) − f (y) g (y) + f (y) g (y) − f (y) g (x)|]
|g (x)|
2
≤ 2 [|g (y)| |f (x) − f (y)| + |f (y)| |g (y) − g (x)|]
|g (x)|
· ¸
2 3
≤ 2 |g (x)| |f (x) − f (y)| + (1 + |f (x)|) |g (y) − g (x)|
|g (x)| 2
2
≤ 2 (1 + 2 |f (x)| + 2 |g (x)|) [|f (x) − f (y)| + |g (y) − g (x)|]
|g (x)|
≡ M [|f (x) − f (y)| + |g (y) − g (x)|]

where M is defined by
2
M≡ 2 (1 + 2 |f (x)| + 2 |g (x)|)
|g (x)|

Now let δ 2 be such that if |x − y| < δ 2 , then

ε −1
|f (x) − f (y)| < M
2
and let δ 3 be such that if |x − y| < δ 3 , then
ε −1
|g (y) − g (x)| < M .
2
Then if 0 < δ ≤ min (δ 0 , δ 1 , δ 2 , δ 3 ) , and |x − y| < δ, everything holds and
¯ ¯
¯ f (x) f (y) ¯¯
¯
¯ g (x) − g (y) ¯ ≤ M [|f (x) − f (y)| + |g (y) − g (x)|]
hε ε i
< M M −1 + M −1 = ε.
2 2
This completes the proof of the second part of 2.)
Note that in these proofs no effort is made to find some sort of “best” δ. 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 ε > 0 be given. Then there exists η > 0 such
that if |y − f (x)| < η and y ∈ D (g) , it follows that |g (y) − g (f (x))| < ε. From
continuity of f at x, there exists δ > 0 such that if |x − z| < δ and z∈ D (f ) , then
|f (z) − f (x)| < η. Then if |x − z| < δ and z ∈ D (g ◦ f ) ⊆ D (f ) , all the above hold and
so
|g (f (z)) − g (f (x))| < ε.
This proves part 3.)
To verify part 4.), let ε > 0 be given and let δ = ε. Then if ||x − y|| < δ, the triangle
inequality implies

|f (x) − f (y)| = |||x|| − ||y|||

≤ ||x − y|| < δ = ε.

This proves part 4.)

Next consider 5.) Suppose first f is continuous. Let U be open and let x ∈ f −1 (U ) .
This means f (x) ∈ U. Since U is open, there exists ε > 0 such that B (f (x) , ε) ⊆ U. By
continuity, there exists δ > 0 such that if y ∈ B (x, δ) , then f (y) ∈ B (f (x) , ε) and so
this shows B (x,δ) ⊆ f −1 (U ) which implies f −1 (U ) is open since x is an arbitrary point
88 CONTINUOUS FUNCTIONS

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) , ε) . The condition says f −1 (B (f (x) , ε)) is
open and since x ∈ f −1 (B (f (x) , ε)) , it follows x is an interior point of f −1 (B (f (x) , ε))
so there exists δ > 0 such that B (x, δ) ⊆ f −1 (B (f (x) , ε)) . This says f (B (x, δ)) ⊆
B (f (x) , ε) . In other words, whenever ||y − x|| < δ, ||f (y) − f (x)|| < ε 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 xn → x with xn ∈ D (f ) , it follows f (xn ) → f (x) .

Proof: Suppose first that f is continuous at x and let xn → x. Let ε > 0 be given.
By continuity, there exists δ > 0 such that if ||y − x|| < δ, then ||f (x) − f (y)|| < ε.
However, there exists nδ such that if n ≥ nδ , then ||xn − x|| < δ and so for all n this
large,
||f (x) − f (xn )|| < ε
which shows f (xn ) → 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 ε > 0 and
xn ∈ D (f ) such that ||x − xn || < n1 , yet

||f (x) − f (xn )|| ≥ ε.

But this is clearly a contradiction because, although xn → x, f (xn ) 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 (xn )|| ≤ l (≥ l)

where {xn } is a sequence of points of D (f ) which converges to x. Then

||f (x)|| ≤ l (≥ l) .

Proof: Since ||f (xn )|| ≤ 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 ||f (xn )|| ≤ l.

n→∞

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 (kn ) → f (k) . Does it follow kn → k? If this does not happen,
then there exists ε > 0 and a subsequence still denoted as {kn } such that

|kn − k| ≥ ε (5.1)
Now since K is compact, there exists a further subsequence, still denoted as {kn } such
that
kn → k0 ∈ K
However, the continuity of f requires
f (kn ) → f (k0 )

and so f (k0 ) = f (k). Since f is one to one, this requires k0 = k, a contradiction to 5.1.
This proves the theorem. ¥

5.2 The Extreme Values Theorem

The extreme values theorem says continuous functions achieve their maximum and
minimum provided they are defined 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 ⊆ Fn 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, x1 , x2 ∈ K such that for all x ∈ K,
f (x1 ) ≤ f (x) ≤ f (x2 ) .
Proof: Let λ = sup {f (x) : x ∈ K} . Next let {λk } be an increasing sequence which
converges to λ but each λk < λ. Therefore, for each k, there exists xk ∈ K such that
f (xk ) > λk .

Since K is sequentially compact, there exists a subsequence, {xkl } such that liml→∞ xkl =
x ∈ K. Then by continuity of f,

f (x) = lim f (xkl ) ≥ lim λkl = λ

l→∞ 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.
This proves the theorem. ¥

5.3 Connected Sets

Stated informally, connected sets are those which are in one piece. In order to define
what is meant by this, I will first consider what it means for a set to not be in one piece.
Definition 5.3.1 Let A be a nonempty subset of V a normed vector space. Then
A is defined 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 ∪ A0
where A0 denotes the set of limit points of A.
90 CONTINUOUS FUNCTIONS

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,

C © ª
A = ∪ HC : 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 definition, A ⊆ A so
if x ∈ A, then x ∈ A. Now consider y ∈ A0 but y ∈ / A. If y ∈
/ A, a closed set, then there
C
exists B (y, r) ⊆ A . Thus y cannot be a limit point of A, a contradiction. Therefore,

A ∪ A0 ⊆ A

Next suppose x ∈ A and suppose x ∈ / A. Then if B (x, r) contains no points of A

different than x, since x itself is not in A, it would follow that B (x,r) ∩ A = ∅ and so
C
recalling that open balls are open, B (x, r) is a closed set containing A so from the
definition, it also contains A which is contrary to the assertion that x ∈ A. Hence if
x∈ / A, then x ∈ A0 and so
A ∪ A0 ⊇ A
This proves the lemma. ¥
Now that the closure of a set has been defined it is possible to define what is meant
by a set being separated.

Definition 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 6= ∅, 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 defined in terms of what it is not. This
makes it somewhat difficult 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 = A ∪ B where A ∩ B = B ∩ A = ∅. Consider the sets, A ∩ U

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.

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 U ∩A = ∅. 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 definition.
Definition 5.3.6 Let S be a set and let p ∈ S. Denote by Cp 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 Cp be a connected component of a set S in a normed vector
space. Then Cp is a connected set and if Cp ∩ Cq 6= ∅, then Cp = Cq .
Proof: Let C denote the connected subsets of S which contain p. If Cp = 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, Cp 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
92 CONTINUOUS FUNCTIONS

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 define 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 δ > 0,
(l, l + δ) ∩ B = ∅
contradicting the definition 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 (e1 ) , f (e2 )) where ei ∈ 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, {(ai , bi )}i=1 such that U = ∪∞i=1 (ai , bi ) .

Proof: Let p ∈ U and let z ∈ Cp , the connected component determined by p. Since

U is open, there exists, δ > 0 such that (z − δ, z + δ) ⊆ U. It follows from Theorem
5.3.4 that
(z − δ, z + δ) ⊆ Cp .
This shows Cp is open. By Theorem 5.3.8, this shows Cp 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 {(ai , bi )}i=1 the set of these connected components. This proves
the theorem. ¥

Definition 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,
γ : [a, b] → E such that γ (a) = p and γ (b) = q.

An example of an arcwise connected topological space would be any subset of Rn

which is the continuous image of an interval. Arcwise connected is not the same as
connected. A well known example is the following.
½µ ¶ ¾
1
x, sin : x ∈ (0, 1] ∪ {(0, y) : y ∈ [−1, 1]} (5.2)
x

You can verify that this set of points in the normed vector space R2 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
γ (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 γ (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 = A ∪ B 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 γ : [a, b] → X such
that γ (a) = p and γ (b) = q. But then γ ([a, b]) = (γ ([a, b]) ∩ A) ∪ (γ ([a, b]) ∩ B) and
the two sets, γ ([a, b]) ∩ A and γ ([a, b]) ∩ B are separated thus showing that γ ([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 satisfies P if there exists a continuous function, γ : [a, b] → U such that γ (a) = p
and γ (b) = x.
A ≡ {x ∈ U such that x satisfies 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 , γ, η such that γ (a) = p, γ (b) = x, η (c) = x, and η (d) =
y. Then let γ 1 : [a, b + d − c] → U be defined as
½
γ (t) if t ∈ [a, b]
γ 1 (t) ≡
η (t + c − b) if t ∈ [b, b + d − c]

Then it is clear that γ 1 is a continuous function mapping p to y and showing that

B (x, r) ⊆ A. Therefore, A is open. A 6= ∅ because since U is open there is an open set,
B (p, δ) 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, δ) be such that z ∈ B (z, δ) ⊆ U, there exist points of A contained in
B (z, δ) . But then, a repeat of the above argument shows z ∈ A also. Hence B is open
and so if B 6= ∅, 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 ∈ Cp where Cp is
the connected component determined by p. Then picking B (z, δ) ⊆ U, Cp ∪ B (z, δ)
is connected and contained in U and so it must also be contained in Cp . Thus z is an
interior point of Cp . This proves the theorem. ¥
As an application, consider the following corollary.

Corollary 5.3.15 Let f : Ω → Z be continuous where Ω is a connected open set in

a normed vector space. Then f must be a constant.
94 CONTINUOUS FUNCTIONS

Proof: Suppose not. Then it achieves two different values, k and l 6= k. Then
Ω = f −1 (l) ∪ f −1 ({m ∈ Z : m 6= l}) and these are disjoint nonempty open sets which
separate Ω. To see they are open, note
µ µ ¶¶
−1 −1 1 1
f ({m ∈ Z : m 6= l}) = f ∪m6=l m − , m +
6 6
¡¡ ¢¢
which is the inverse image of an open set while f −1 (l) = f −1 l − 61 , l + 16 also an
open set. ¥

5.4 Uniform Continuity

The concept of uniform continuity is also similar to the one dimensional concept.
Definition 5.4.1 Let f be a function. Then f is uniformly continuous if for
every ε > 0, there exists a δ depending only on ε such that if ||x − y|| < δ then
||f (x) − f (y)|| < ε.

Theorem 5.4.2 Let f : K → F be continuous where K is a sequentially compact

set in Fn or more generally a normed vector space. Then f is uniformly continuous on
K.
Proof: If this is not true, there exists ε > 0 such that for every δ > 0 there exists a
pair of points, xδ and yδ such that even though ||xδ − yδ || < δ, ||f (xδ ) − f (yδ )|| ≥ ε.
Taking a succession of values for δ equal to 1, 1/2, 1/3, · · ·, and letting the exceptional
pair of points for δ = 1/n be denoted by xn and yn ,
1
||xn − yn || < , ||f (xn ) − f (yn )|| ≥ ε.
n
Now since K is sequentially compact, there exists a subsequence, {xnk } such that xnk →
z ∈ K. Now nk ≥ k and so
1
||xnk − ynk || < .
k
Hence
||ynk − z|| ≤ ||ynk − xnk || + ||xnk − z||
1
< + ||xnk − z||
k
Consequently, ynk → z also. By continuity of f and Theorem 5.1.2,
0 = ||f (z) − f (z)|| = lim ||f (xnk ) − f (ynk )|| ≥ ε,
k→∞

an obvious contradiction. Therefore, the theorem must be true. ¥

Recall the closed and bounded subsets of Fn 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.
Definition 5.5.1 A sequence of functions is a map defined 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 {fn }n=m where fn 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 definition of uniform convergence is exactly the same as earlier only now it is
not possible to draw representative pictures so easily.

Definition 5.5.2 Let {fn } 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 fn (x)
n→∞

Thus you consider for each x ∈ D the sequence of numbers {fn (x)} and if this
sequence converges for each x ∈ D, the thing it converges to is called f (x).

Definition 5.5.3 Let {fn } be a sequence of functions defined on D. Then {fn }

is said to converge uniformly to f if it converges pointwise to f and for every ε > 0 there
exists N such that for all n ≥ N

||f (x) − fn (x)|| < ε

for all x ∈ D.

Theorem 5.5.4 Let {fn } be a sequence of continuous functions defined on D

and suppose this sequence converges uniformly to f . Then f is also continuous on D. If
each fn is uniformly continuous on D, then f is also uniformly continuous on D.

Proof: Let ε > 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) − fn (x)|| < ε/3. (5.3)

Pick such an n. By assumption, fn is continuous at z. Therefore, there exists δ > 0 such

that if ||z − x|| < δ then
||fn (x) − fn (z)|| < ε/3.
It follows that for ||x − z|| < δ,

||f (x) − f (z)|| ≤ ||f (x) − fn (x)|| + ||fn (x) − fn (z)|| + ||fn (z) − f (z)||
< ε/3 + ε/3 + ε/3 = ε

which shows that since ε was arbitrary, f is continuous at z.

In the case where each fn is uniformly continuous, and using the same fn for which
5.3 holds, there exists a δ > 0 such that if ||y − z|| < δ, then

||fn (z) − fn (y)|| < ε/3.

Then for ||y − z|| < δ,

||f (y) − f (z)|| ≤ ||f (y) − fn (y)|| + ||fn (y) − fn (z)|| + ||fn (z) − f (z)||
< ε/3 + ε/3 + ε/3 = ε

This shows uniform continuity of f . This proves the theorem. ¥

Definition 5.5.5 Let {fn } be a sequence of functions defined on D. Then the

sequence is said to be uniformly Cauchy if for every ε > 0 there exists N such that
whenever m, n ≥ N,
||fm (x) − fn (x)|| < ε
for all x ∈ D.
96 CONTINUOUS FUNCTIONS

Then the following theorem follows easily.

Theorem 5.5.6 Let {fn } be a uniformly Cauchy sequence of functions defined

on D having values in a complete normed vector space such as Fn for example. Then
there exists f defined on D such that {fn } converges uniformly to f .

Proof: For each x ∈ D, {fn (x)} is a Cauchy sequence. Therefore, it converges to

some vector f (x). Let ε > 0 be given and let N be such that if n, m ≥ N,

||fm (x) − fn (x)|| < ε/2

for all x ∈ D. Then for any x ∈ D, pick n ≥ N and it follows from Theorem 4.1.8

||f (x) − fn (x)|| = lim ||fm (x) − fn (x)|| ≤ ε/2 < ε.

m→∞

This proves the theorem. ¥

Corollary 5.5.7 Let {fn } be a uniformly Cauchy sequence of functions continuous

on D having values in a complete normed vector space like Fn . Then there exists f
defined on D such that {fn } converges uniformly to f and f is continuous. Also, if each
fn 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.

Theorem 5.5.8 Let {fn } be a sequence of functions defined on D having values

in a complete normed vector space like Fn . Then it converges pointwise if and only if
the sequence {fn (x)} is a Cauchy sequence for every x ∈ D. It converges uniformly if
and only if {fn } is a uniformly Cauchy sequence.

Proof: If the sequence converges pointwise, then by Theorem 4.4.3 the sequence
{fn (x)} is a Cauchy sequence for each x ∈ D. Conversely, if {fn (x)} is a Cauchy se-
quence for each x ∈ D, then {fn (x)} converges for each x ∈ D because of completeness.
Now suppose {fn } is uniformly Cauchy. Then from Theorem 5.5.6 there exists f
such that {fn } converges uniformly on D to f . Conversely, if {fn } converges uniformly
to f on D, then if ε > 0 is given, there exists N such that if n ≥ N,

|f (x) − fn (x)| < ε/2

for every x ∈ D. Then if m, n ≥ N and x ∈ D,

|fn (x) − fm (x)| ≤ |fn (x) − f (x)| + |f (x) − fm (x)| < ε/2 + ε/2 = ε.

Thus {fn } is uniformly Cauchy. ¥

Once you understand sequences, it is no problem to consider series.

Definition 5.5.9 Let {fn } be a sequence of functions defined on D. Then

̰ ! n
X X
fk (x) ≡ lim fk (x) (5.4)
n→∞
k=1 k=1

whenever the limit exists. Thus there is a new function denoted by

∞
X
fk (5.5)
k=1
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
P5.4. If for all
∞
x ∈ D, the limit in 5.4 exists, then 5.5 is said to converge pointwise. k=1 fk is said
to converge uniformly on D if the sequence of partial sums,
( n )
X
fk
k=1

converges uniformly.If the indices for the functions start at some other value than 1,
you make the obvious modification to the above definition.

Theorem 5.5.10 Let {fn } be a sequence of functions P∞defined on D which have

values in a complete normed vector space like Fn . The series k=1 fk converges pointwise
if and only if for each ε > 0 and x ∈ D, there exists Nε,x which may depend on x as
well as ε such that when q > p ≥ Nε,x ,
¯¯ ¯¯
¯¯ X ¯¯
¯¯ q ¯¯
¯¯ fk (x)¯¯¯¯ < ε
¯¯
¯¯k=p ¯¯
P∞
The series k=1 fk converges uniformly on D if for every ε > 0 there exists Nε such
that if q > p ≥ N then ¯¯ ¯¯
¯¯ q ¯¯
¯¯ X ¯¯
¯¯ f (x)¯¯ < ε (5.6)
¯¯ k ¯¯
¯¯k=p ¯¯

for all x ∈ D.

Proof: The first 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 P∞5.5.6, these partial sums converge uniformly to
a function which is the definition of k=1 fk . This proves the theorem. ¥
Is there an easy way to recognize when 5.6 happens? Yes, there is. It is called the
Weierstrass M test.

Theorem 5.5.11 Let {fn } be a sequence of functions defined on D having val-

space like Fn . Suppose there
ues in a complete normed vector P P∞exists Mn such that
∞
sup {|fn (x)| : x ∈ D} < Mn and n=1 Mn converges. Then n=1 fn converges uni-
formly on D.

Proof: Let z ∈ D. Then letting m < n and using the triangle inequality
¯¯ ¯¯
¯¯ X
n m
X ¯¯ n
X X∞
¯¯ ¯¯
¯¯ fk (z) − fk (z)¯¯ ≤ ||fk (z)|| ≤ Mk < ε
¯¯ ¯¯
k=1 k=1 k=m+1 k=m+1
P∞
whenever m is large enough because of the assumption that n=1 Mn converges. There-
fore, the sequence
P∞ of partial sums is uniformly Cauchy on D and therefore, converges
uniformly to k=1 fk on D. This proves the theorem. ¥

Theorem
P∞ 5.5.12 If {fn } is a sequence of continuous
P∞ functions defined on D
and k=1 fk 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
P∞ sums of
the above series which is assumed to converge uniformly to the function, k=1 fk . ¥
98 CONTINUOUS FUNCTIONS

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 definition.
Definition 5.6.1 Let α be an n dimensional multi-index. This means
α = (α1 , · · ·, αn )
where each αi is a positive integer or zero. Also, let
n
X
|α| ≡ |αi |
i=1

Then xα means
xα ≡ xα 1 α2 αn
1 x2 · · · x3
where each xj ∈ F. An n dimensional polynomial of degree m is a function of the form
X
p (x) = dα xα.
|α|≤m

where the dα are complex or real numbers. Rational functions are defined as the quotient
of two polynomials. Thus these functions are defined on Fn .
For example, f (x) = x1 x22 + 7x43 x1 is a polynomial of degree 5 and
x1 x22 + 7x43 x1 + x32
4x31 x22 + 7x23 x1 − x32
is a rational function.
Note that in the case of a rational function, the domain of the function might not
be all of Fn . For example, if
x1 x22 + 7x43 x1 + x32
f (x) = ,
x22 + 3x21 − 4
the domain of f would be all complex numbers such that x22 + 3x21 6= 4.
By Theorem 5.0.2 all polynomials are continuous. To see this, note that the function,
π k (x) ≡ xk
is a continuous function because of the inequality
|π k (x) − π k (y)| = |xk − yk | ≤ |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
X
f (x) ≡ dα xα
|α|≤m

where dα ∈ 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 π 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 defined 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 defined 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 µ ¶
X m 2 m−k 1
(k − mx) xk (1 − x) ≤ m
k 4
k=0

Proof: First of all, from the binomial theorem

X m µ ¶
m m−k
kxk (1 − x)
k
k=0
X m µ ¶
m m−k
= kxk (1 − x)
k
k=1
m
X (m − 1)! m−k
= mx xk−1 (1 − x)
(k − 1)! (m − k)!
k=1
Xm
(m − 1)! m−1−k
= mx xk (1 − x)
(k − 1)! (m − 1 − k)!
k=1
m−1
= mx (1 + 1 − x) = mx.

Next, using what was just shown and the binomial theorem again,
Xm µ ¶ Xm µ ¶
m 2 k m−k m m−k
k x (1 − x) = k (k − 1) xk (1 − x)
k k
k=0 k=1
Xm µ ¶
m m−k
+ kxk (1 − x)
k
k=0

m µ ¶
X m m−k
= k (k − 1) xk (1 − x) + mx
k
k=2

Xµ
m−2
m
¶
m−2−k
= (k + 2) (k + 1) xk+2 (1 − x) + mx
k+2
k=0

Xµ
m−2 ¶
m − 2 k+2 m−2−k
= m (m − 1) x (1 − x) + mx
k
k=0
X µm − 2¶
m−2
m−2−k
2
= x m (m − 1) xk (1 − x) + mx
k
k=0
= x2 m (m − 1) + mx = x2 m2 − x2 m + mx

It follows
m µ ¶
X m 2 m−k
(k − mx) xk (1 − x)
k
k=0
100 CONTINUOUS FUNCTIONS

m µ ¶
X m ¡ 2 ¢ m−k
= k − 2kmx + x2 m2 xk (1 − x)
k
k=0

and from what was just shown along with the binomial theorem again, this equals
µ ¶2
m 1
x2 m2 − x2 m + mx − 2mx (mx) + x2 m2 = −x2 m + mx = −m x− .
4 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 defined on [0, 1] . Let pn be the polynomial
defined by
n µ ¶ µ ¶
X n k n−k
pn (x) ≡ f xk (1 − x) . (5.7)
k n
k=0
n
Now for f a continuous function defined on [0, 1] and for x = (x1 , · · ·, xn ) ,consider the
polynomial,
m
X m µ ¶µ ¶
X µ ¶
m m m k1 m−k1 k2 m−k2
pm (x) ≡ ··· ··· x (1 − x1 ) x2 (1 − x2 )
k1 k2 kn 1
k1 =1 kn =1

µ ¶
m−kn k1 kn
··· xknn (1 − xn ) f , · · ·, . (5.8)
m m
Also define if I is a set in Rn

||h||I ≡ sup {|h (x)| : x ∈ I} .

Thus pm converges uniformly to f on a set, I if

lim ||pm − f ||I = 0.

m→∞

¡ k1 ¢
To simplify the notation, let k = (k1 , · · ·, kn ) where each ki ∈ [0, m], k
m ≡ m,· · ·, kmn ,
and let µ ¶ µ ¶µ ¶ µ ¶
m m m m
≡ ··· .
k k1 k2 kn
Also define
||k||∞ ≡ max {ki , i = 1, 2, · · ·, n}
m−k m−k1 m−k2 m−kn
xk (1 − x) ≡ xk11 (1 − x1 ) xk22 (1 − x2 ) · · · xknn (1 − xn ) .

Thus in terms of this notation,

X µ ¶ µ ¶
m k m−k k
pm (x) = x (1 − x) f
k m
||k||∞ ≤m

This is the n dimensional version of the Bernstein polynomials which is what results n
the case where n=1.

Lemma 5.7.2 For x ∈ [0, 1]n , f a continuous F valued function defined on [0, 1]n ,
n
and pm given in 5.8, pm converges uniformly to f on [0, 1] as m → ∞.
5.7. SEQUENCES OF POLYNOMIALS, WEIERSTRASS APPROXIMATION 101

Proof: The function, f is uniformly continuous because it is continuous on a se-

n
quentially compact set, [0, 1] . Therefore, there exists δ > 0 such that if |x − y| < δ,
then
|f (x) − f (y)| < ε.
2 √
Denote by G the set of k such that (ki − mxi ) < η 2 m 2
¯ k for each
¯ i where η = δ/ n. Note
this condition is equivalent to saying that for each i, ¯ mi − xi ¯ < η. A short computation
shows that by the binomial theorem,
X µm¶ m−k
xk (1 − x) =1
k
||k||∞ ≤m

n
and so for x ∈ [0, 1] ,
X µ ¶ ¯ µ ¶ ¯
m k m−k ¯¯ k ¯
|pm (x) − f (x)| ≤ x (1 − x) ¯ f − f (x)¯¯
k m
||k||∞ ≤m

X µm¶ ¯ µ ¶ ¯
k m−k ¯¯ k ¯
¯
≤ x (1 − x) ¯f m − f (x)¯
k
k∈G
X µm¶ ¯ µ ¶ ¯
m−k ¯¯ k ¯
¯
+ xk (1 − x) ¯ f − f (x) ¯ (5.9)
C
k m
k∈G

Now for k ∈ G it follows that for each i

¯ ¯
¯ ki ¯
¯ − xi ¯ < √δ (5.10)
¯m ¯ n
¯ ¡k¢ ¯ ¯k ¯
and so ¯f m − f (x)¯ < ε because the above implies ¯ m − x¯ < δ. Therefore, the first
sum on the right in 5.9 is no larger than
X µm¶ m−k
X µm¶ m−k
k
x (1 − x) ε≤ xk (1 − x) ε = ε.
k k
k∈G ||k||∞ ≤m

n
Letting M ≥ max {|f (x)| : x ∈ [0, 1] } it follows

|pm (x) − f (x)|

X µm¶ m−k
≤ ε + 2M xk (1 − x)
k
k∈GC
µ ¶n X µ ¶ Y n
1 m 2 m−k
≤ ε + 2M (kj − mxj ) xk (1 − x)
η 2 m2 k j=1
k∈GC
µ ¶n X µ ¶ Y n
1 m 2 m−k
≤ ε + 2M (kj − mxj ) xk (1 − x)
η 2 m2 k j=1
||k||∞ ≤m

because on GC ,
2
(kj − mxj )
< 1, j = 1, · · ·, n.
η 2 m2
Now by Lemma 5.7.1,
µ ¶n ³
1 m ´n
|pm (x) − f (x)| ≤ ε + 2M .
η 2 m2 4
102 CONTINUOUS FUNCTIONS

Therefore, since the right side does not depend on x, it follows that for all m sufficiently
large,
||pm − f ||[0,1]n ≤ 2ε
and since ε is arbitrary, this shows limm→∞ ||pm − f ||[0,1]n = 0. This proves the lemma.
¥

Theorem 5.7.3 Let f be a continuous function defined on

n
Y
R≡ [ak , bk ] .
k=1

Then there exists a sequence of polynomials {pm } converging uniformly to f on R.

Proof: Let gk : [0, 1] → [ak , bk ] be linear, one to one, and onto and let

x = g (y) ≡ (g1 (y1 ) , g2 (y2 ) , · · ·, gn (yn )) .

n Qn
Thus g : [0, 1] → k=1 [ak , bk ] is one to one, onto, and each component function is
n
linear. Then f ◦ g is a continuous function defined on [0, 1] . It follows from Lemma
n
5.7.2 there exists a sequence of polynomials, {pm©(y)}¡ each defined ¢ª on [0, 1] which
n −1
converges uniformly to f ◦ g on [0, 1] . Therefore, pm g (x) converges uniformly
to f (x) on R. But
¡ ¢
y = (y1 , · · ·, yn ) = g1−1 (x1 ) , · · ·, gn−1 (xn )
© ¡ ¢ª
and each gk−1 is linear. Therefore, pm g−1 (x) is a sequence of polynomials. This
proves the theorem. ¥
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 defined on a
closed and bounded subset of Rn rather than a box.

Lemma 5.7.4 Let S ⊆ Rn be a nonempty subset. Define

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| . (5.11)

Proof: The continuity of x → dist (x, S) is obvious if the inequality 5.11 is estab-
lished. So let x, y ∈ Rn . Without loss of generality, assume dist (x, S) ≥ dist (y, S) and
pick z ∈ S such that |y − z| − ε < dist (y, S) . Then

|dist (x, S) − dist (y, S)| = dist (x, S) − dist (y, S)

≤ |x − z| − (|y − z| − ε)
≤ |z − y| + |x − y| − |y − z| + ε = |x − y| + ε.

Since ε is arbitrary, this proves 5.11. ¥

5.7. SEQUENCES OF POLYNOMIALS, WEIERSTRASS APPROXIMATION 103

Lemma 5.7.5 Let H, K be two nonempty disjoint closed subsets of Rn . Then there
exists a continuous function, g : Rn → [−1, 1] such that g (H) = −1/3, g (K) =
1/3, g (Rn ) ⊆ [−1/3, 1/3] .

Proof: Let
dist (x, H)
f (x) ≡ .
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 hk ∈ B (x, 1/k) ∩ H. Then hk → 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 definition, f = 0 on H and f = 1 on K.
Now let g (x) ≡ 23 f (x) − 12 . Then g has the desired properties. ¥

Definition 5.7.6 For f a real or complex valued bounded continuous function

defined on M ⊆ Rn .
||f ||M ≡ sup {|f (x)| : x ∈ M } .

Lemma 5.7.7 Suppose M is a closed set in Rn where Rn and suppose f : M →

[−1, 1] is continuous at every point of M. Then there exists a function, g which is
defined and continuous on all of Rn such that ||f − g||M < 23 , g (Rn ) ⊆ [−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 first H, K are both nonempty. Then by Lemma 5.7.5 there
exists g such that g is a continuous function defined on all of Rn and g (H) = −1/3,
g (K) = 1/3, and g (Rn ) ⊆ [−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 5.7.8 Suppose M is a closed set in Rn and suppose f : M → [−1, 1] is

continuous at every point of M. Then there exists a function, g which is defined and
continuous on all of Rn such that g = f on M and g has its values in [−1, 1] .

Proof: Using Lemma 5.7.7, let g1 be such that g1 (Rn ) ⊆ [−1/3, 1/3] and
2
||f − g1 ||M ≤ .
3
Suppose g1 , · · ·, gm have been chosen such that gj (Rn ) ⊆ [−1/3, 1/3] and
¯¯ ¯¯
¯¯ m µ ¶i−1 ¯¯
X µ ¶m
¯¯ 2 ¯¯ 2
¯¯f − gi ¯¯ < . (5.12)
¯¯ 3 ¯¯ 3
i=1 M

This has been done for m = 1. Then

¯¯µ ¶ à m µ ¶i−1
!¯¯
¯¯ 3 m X 2 ¯¯
¯¯ ¯¯
¯¯ f− gi ¯¯ ≤1
¯¯ 2 3 ¯¯
i=1 M

¡ 3 ¢m ³ Pm ¡ 2 ¢i−1 ´
and so 2 f − i=1 3 gi can play the role of f in the first step of the proof.
Therefore, there exists gm+1 defined and continuous on all of Rn such that its values
are in [−1/3, 1/3] and
¯¯µ ¶ à m µ ¶i−1
! ¯¯
¯¯ 3 m X 2 ¯¯ 2
¯¯ ¯¯
¯¯ f− gi − gm+1 ¯¯ ≤ .
¯¯ 2 3 ¯¯ 3
i=1 M
104 CONTINUOUS FUNCTIONS

Hence ¯¯Ã ! µ ¶ ¯¯
¯¯ m µ ¶i−1
X m ¯¯ µ ¶m+1
¯¯ 2 2 ¯¯ 2
¯¯ f − gi − gm+1 ¯¯ ≤ .
¯¯ 3 3 ¯¯ 3
i=1 M
It follows there exists a sequence, {gi } such that each has its values in [−1/3, 1/3] and
for every m 5.12 holds. Then let
X∞ µ ¶i−1
2
g (x) ≡ gi (x) .
i=1
3

It follows ¯∞ µ ¶ ¯
¯X 2 i−1 ¯ X m µ ¶i−1
¯ ¯ 2 1
|g (x)| ≤ ¯ gi (x)¯ ≤ ≤1
¯ 3 ¯ 3 3
i=1 i=1
and ¯µ ¶ ¯ µ ¶
¯ 2 i−1 ¯ 2
i−1
1
¯ ¯
¯ gi (x)¯ ≤
¯ 3 ¯ 3 3
so the Weierstrass M test applies and shows convergence is uniform. Therefore g must
be continuous. The estimate 5.12 implies f = g on M . ¥
The following is the Tietze extension theorem.
Theorem 5.7.9 Let M be a closed nonempty subset of Rn and let f : M → [a, b]
be continuous at every point of M. Then there exists a function, g continuous on all of
Rn which coincides with f on M such that g (Rn ) ⊆ [a, b] .
2
Proof: Let f1 (x) = 1 + b−a (f (x) − b) . Then f1 satisfies the conditions of Lemma
5.7.8 and so there exists g1 : Rn → [−1,
¡ b−a
n
¢ 1] such that g is continuous on R and equals
f1 on M . Let g (x) = (g1 (x) − 1) 2 + b. This works. ¥
With the Tietze extension theorem, here is a better version of the Weierstrass ap-
proximation theorem.
Theorem 5.7.10 Let K be a closed and bounded subset of Rn and let f : K → R
be continuous. Then there exists a sequence of polynomials {pm } such that
lim (sup {|f (x) − pm (x)| : x ∈ K}) = 0.
m→∞

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 defined on all Rn such that g = f on K. Now since K is bounded,
there exist intervals, [ak , bk ] such that
n
Y
K⊆ [ak , bk ] = R
k=1

Then by the Weierstrass approximation theorem, Theorem 5.7.3 there exists a sequence
of polynomials {pm } 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 5.7.11 Let K be a closed and bounded subset of Rn and let f : K → F
be continuous. Then there exists a sequence of polynomials {pm } such that
lim (sup {|f (x) − pm (x)| : x ∈ K}) = 0.
m→∞

In other words, the sequence of polynomials converges uniformly to f on K.

5.8. THE OPERATOR NORM 105

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 definition.

Definition 5.8.1 Let V, W be two finite 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 defined 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 {v1 , · · ·, vn }. Let
( ¯¯ n ¯¯ )
¯¯ X ¯¯
n ¯¯ ¯¯
A= a ∈ F : ¯¯ ak vk ¯¯ ≤ 1
¯¯ ¯¯
k=1

where a = (a1 , · · ·, an ) . Then A is a closed and bounded subset of Fn .

Proof: First suppose a ∈

/ A. Then
¯¯ ¯¯
¯¯ X
n ¯¯
¯¯ ¯¯
¯¯ ak vk ¯¯ > 1.
¯¯ ¯¯
k=1

Then for b = (b1 , · · ·, bn ) , and using the triangle inequality,

¯¯ ¯¯ ¯¯ ¯¯
¯¯ Xn ¯¯ ¯¯ X n ¯¯
¯¯ ¯¯ ¯¯ ¯¯
¯¯ bk vk ¯¯ = ¯¯ (ak − (ak − bk )) vk ¯¯
¯¯ ¯¯ ¯¯ ¯¯
k=1 k=1
¯¯ ¯¯
¯¯ X
n ¯¯ X n
¯¯ ¯¯
≥ ¯¯ ak vk ¯¯ − |ak − bk | ||vk ||
¯¯ ¯¯
k=1 k=1

and now it is apparent that if |a − b| is sufficiently small so that each |ak − bk | is small
enough, this expression is larger than 1. Thus there exists δ > 0 such that B (a, δ) ⊆ AC
showing that AC is open. Therefore, A is closed.
Next consider the claim that A is bounded. Suppose this is not so. Then there exists
a sequence {ak } of points of A,
¡ ¢
ak = a1k , · · ·, ank ,

such that limk→∞ |ak | = ∞. Then from the definition of A,

¯¯ ¯¯
¯¯ n ¯¯
¯¯X ajk ¯¯ 1
¯¯ v ¯¯
j ¯¯ ≤ . (5.13)
¯¯ |a | |a k|
¯¯ j=1 k ¯¯

Let µ ¶
a1k an
bk = , · · ·, k
|ak | |ak |
Then |bk | = 1 so bk is contained in the closed and bounded set, S (0, 1) which is
sequentially compact in Fn . It follows there exists a subsequence, still denoted by {bk }
106 CONTINUOUS FUNCTIONS

such that it converges to b ∈ S (0,1) . Passing to the limit in 5.13 using the following
inequality, ¯¯ ¯¯
¯¯ n ¯¯ X ¯
n ¯ j
¯
¯¯X ajk Xn
¯¯ a ¯
¯¯ ¯ ¯ ¯ k ¯
¯¯ v j − b v
j j ¯¯ ≤ ¯ − bj ¯ ||vj ||
¯¯ j=1 |a k | ¯¯ j=1 ¯ |a k | ¯
j=1
Pn
to see that the sum converges to j=1 bj vj , it follows
n
X
bj vj = 0
j=1

and this is a contradiction because {v1 , · · ·, vn } is a basis and not all the bj 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
||M L|| ≤ ||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

¯¯   ¯¯ 
¯¯¯¯ Xn ¯¯
¯¯ 
||L|| ≡ sup ¯¯¯¯L  aj vj ¯¯¯¯ : a ∈ A
¯¯ ¯¯ 
j=1
¯¯ ¯¯ 
¯¯¯¯Xn ¯¯
¯¯ 
= sup ¯¯¯¯ aj L (vj )¯¯¯¯ : a ∈ A < ∞
¯¯ ¯¯ 
j=1
¯¯P ¯¯
¯¯ n ¯¯
because a → ¯¯ j=1 aj L (vj )¯¯ is a real valued continuous function defined 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 6= 0. Then from the
definition of ||L|| , ¯¯ µ ¶¯¯
¯¯ x ¯¯¯¯
¯¯L ≤ ||L||
¯¯ ||x|| ¯¯
and so, since L is linear, you can multiply on both sides by ||x|| and conclude

||L (x)|| ≤ ||L|| ||x|| .

For the other claim,

||M L|| ≡ sup {||M L (x)|| : ||x|| ≤ 1}

≤ ||M || sup {||Lx|| : ||x|| ≤ 1} ≡ ||M || ||L|| .
5.8. THE OPERATOR NORM 107

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 definition. 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 + M x|| : ||x|| ≤ 1}

≤ sup {||M x|| + ||Lx|| : ||x|| ≤ 1}

≤ sup {||Lx|| : ||x|| ≤ 1} + sup {||M x|| : ||x|| ≤ 1}
because ||Lx|| ≤ sup {||Lx|| : ||x|| ≤ 1} with a similar inequality holding for M. There-
fore, by definition,
||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 δ > 0 such that

L (v) ∈ B (L (v) , δ) ⊆ U.

Then if w ∈ V,
||L (v − w)|| = ||L (v) − L (w)|| ≤ ||L|| ||v − w||
and so if ||v − w|| is sufficiently small, ||v − w|| < δ/ ||L|| , then L (w) ∈ B (L (v) , δ)
which shows B (v, δ/ ||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 different
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 ≤ K2 ||v||1 (5.14)

while the same reasoning implies there exists K1 such that

||v||1 ≤ K1 ||v||2 . (5.15)

This leads to the following important theorem.

Theorem 5.8.4 Let V be a finite dimensional vector space and let ||·||1 and ||·||2
be two norms for V. Then these norms are equivalent which means there exist constants,
δ, ∆ such that for all v ∈ V

δ ||v||1 ≤ ||v||2 ≤ ∆ ||v||1

A set, K is sequentially compact if and only if it is closed and bounded. Also every finite
dimensional normed vector space is complete. Also any closed and bounded subset of a
finite dimensional normed vector space is sequentially compact.
108 CONTINUOUS FUNCTIONS

Proof: From 5.14 and 5.15

||v||1 ≤ K1 ||v||2 ≤ K1 K2 ||v||1

and so
1
||v||1 ≤ ||v||2 ≤ K2 ||v||1 .
K1
Next consider the claim that all closed and bounded sets in a normed vector space
are sequentially compact. Let L : Fn → V be defined by
n
X
L (a) ≡ ak vk
k=1

where {v1 , · · ·, vn } is a basis for V . Thus L ∈ L (Fn , 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) = Fn \ L−1 K C = Fn \ (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, Fn ). Therefore,
¯ −1 ¯ ¯¯ ¯¯ ¯¯ ¯¯
¯L (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 Fn and is therefore sequentially compact.
© ªIt follows that if
∞ ∞
{vk }k=1 ⊆ K, there is a subsequence {vkl }l=1 such that L−1 vkl converges to a
point, a ∈ L−1 (K) . Hence by continuity of L,
¡ ¢
vkl = L L−1 (vkl ) → 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, k0 . Since k0 is a limit point,
there exists kn ∈ B k0 , n1 such that kn 6= k0 . Therefore, {kn } has no limit point in
K because k0 ∈ / K. It follows K must be closed. If K is not bounded, then you could
pick kn ∈ K such that km ∈ / B (0, m) and it follows {kk } cannot have a subsequence
which
© ª converges because if k ∈ K, then for large enough m, k ∈ B (0, m/2) and so if
kkj is any subsequence, kkj ∈ / B (0, m) for all but finitely 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 {vk }k=1 © −1 be aªCauchy sequence
−1 n ∞
in V . Since L , defined above is in L (V, F ) , it follows L vk k=1 is a Cauchy
sequence in Fn . This follows from the inequality,
¯ −1 ¯ ¯¯ ¯¯
¯L vk − L−1 vl ¯ ≤ ¯¯L−1 ¯¯ ||vk − vl || .

therefore, there exists a ∈ Fn such that L−1 vk → a and since L is continuous,

¡ ¢
vk = L L−1 (vk ) → L (a) .
∞
Next suppose K is a closed and bounded subset of V and let {xk }k=1 be a sequence
of vectors in K. Let {v1 , · · · , vn } be a basis for V and let
n
X
xk = xjk vj
j=1
5.9. ASCOLI ARZELA THEOREM 109

Define a norm for V according to

n
X n
X
¯ j ¯2
2
||x|| ≡ ¯x ¯ , x = xj vj
j=1 j=1

It is clear most axioms of a norm hold. The triangle inequality also holds because by
the triangle inequality for Fn ,
 1/2
n
X ¯ j ¯ 2
||x + y|| ≡  ¯x + y j ¯ 
j=1

 1/2  1/2
n
X Xn
¯ j ¯2 ¯ j ¯2
≤ ¯x ¯  +  ¯y ¯  ≡ ||x|| + ||y|| .
j=1 j=1

By the first part of this theorem, this norm is equivalent to the norm on V . Thus
n oK∞ is
closed and bounded with respect to this new norm. It follows that for each j, xjk
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 xkl such that
for each j,
lim xjkl = xj
l→∞

for some xj . Hence

n
X n
X
lim xkl = lim xjkl vj = xj vj ∈ K
l→∞ l→∞
j=1 j=1

because K is closed. This proves the theorem. ¥

Example 5.8.5 Let V beP a vector space and let {v1 , · · ·, vn } be a basis. Define a norm
n
on V as follows. For v = k=1 ak vk ,

||v|| ≡ max {|ak | : 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 P
and equals 0 if and onlyPif v = 0. The hard part is the triangle
n n
inequality. Let v = k=1 ak vk and w = v = k=1 bk vk .

||v + w|| ≡ max {|ak + bk |} ≤ max {|ak | + |bk |}

k k
≤ max |ak | + max |bk | ≡ ||v|| + ||w|| .
k k

This shows this is indeed a norm.

5.9 Ascoli Arzela Theorem

∞
Let {fk }k=1 be a sequence of functions defined on a compact set which have values in a
finite dimensional normed vector space V. The following definition will be of the norm
of such a function.
110 CONTINUOUS FUNCTIONS

Definition 5.9.1 Let f : K → V be a continuous function which has values in

a finite dimensional normed vector space V where here K is a compact set contained in
some normed vector space. Define

||f || ≡ sup {||f (x)||V : x ∈ K} .

Denote the set of such functions by C (K; V ) .

Proposition 5.9.2 The above definition 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 ||αf || = |α| ||f || whenever α ∈ F, the field 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 defined nonnegative real number.
It remains to verify completeness. Suppose then {fk } is a Cauchy sequence with
respect to this norm. Then from the definition 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 {fk } converges uniformly to f . That is,

lim ||f − f k || = 0.
k→∞

This proves the proposition. ¥

Theorem 5.8.4 says that closed and bounded sets in a finite dimensional normed vec-
tor space V are sequentially compact. This theorem typically doesn’t apply to C (K; V )
because generally this is not a finite 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.

Definition 5.9.3 Let F ⊆ C (K; V ) . Then F is equicontinuous if for every

ε > 0 there exists δ > 0 such that whenever |x − y| < δ, it follows

|f (x) − f (y)| < ε

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 finite dimen-

∞
sional normed vector space. Then there exists a countable set D ≡ {ki }i=1 such that for
every ε > 0 and for every x ∈ K,

B (x, ε) ∩ D 6= ∅.
5.9. ASCOLI ARZELA THEOREM 111
¡ ¢
Proof: Let n ∈ N. Pick k1n ∈ K. If B k1n , n1 ⊇ K, stop. Otherwise pick
µ ¶
n n 1
k2 ∈ K \ B k1 ,
n
Continue this way till the process ends. It must end because if it didn’t, there would
exist a convergent subsequence which would imply two of the kjn would have to be closer
than 1/n which is impossible from the construction. Denote this collection of points
by Dn . Then D ≡ ∪∞ n=1 Dn . This must work because if ε > 0 is given and x ∈ K, let
1/n < ε/3 and the construction implies x ∈ B (kin , 1/n) for some kin ∈ Dn ∪ D. Then
kin ∈ B (x, ε) .
D is countable because it is the countable union of finite sets. This proves the lemma.
¥
Definition 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, ε) ∩ D 6= ∅
then K is called separable.
Now here is another remarkable result about equicontinuous functions.
Lemma 5.9.6 Suppose {fk }∞
k=1 is equicontinuous and the functions are defined on
a sequentially compact set K. Suppose also for each x ∈ K,
lim fk (x) = f (x) .
k→∞

Then in fact f is continuous and the convergence is uniform. That is

lim ||fk − f || = 0.
k→∞

Proof: Uniform convergence would say that for every ε > 0, there exists nε such
that if k, l ≥ nε , then for all x ∈ K,
||fk (x) − fl (x)|| < ε.
Thus if the given sequence does not converge uniformly, there exists ε > 0 such that for
all n, there exists k, l ≥ n and xn ∈ K such that
||fk (xn ) − fl (xn )|| ≥ ε
Since K is sequentially compact, there exists a subsequence, still denoted by {xn } such
that limn→∞ xn = x ∈ K. Then letting k, l be associated with n as just described,
ε ≤ ||fk (xn ) − fl (xn )||V ≤ ||fk (xn ) − fk (x)||V
+ ||fk (x) − fl (x)||V + ||fl (x) − fl (xn )||V
By equicontinuity, if n is large enough, this implies
ε ε
ε < + ||fk (x) − fl (x)||V +
3 3
and now taking n still larger if necessary, the middle term on the right in the above is
also less than ε/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 ||f − f k || = 0.
k→∞

This proves the lemma. ¥

The Ascoli Arzela theorem is the following.
112 CONTINUOUS FUNCTIONS

Theorem 5.9.7 Let K be a closed and bounded subset of a finite dimensional

normed vector space and let F ⊆ C (K; V ) where V is a finite dimensional normed
∞
vector space. Suppose also that F is bounded and equicontinuous. Then if {fk }k=1 ⊆ F,
∞
there exists a subsequence {fkl }l=1 which converges to a function f ∈ C (K; V ) in the
sense that
lim ||f − f kl ||
l→∞

© ª∞ © ª∞
Proof: Denote by f(k,n) n=1 a subsequence of f(k−1,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 = {dk }k=1 . Then consider the following
diagram.
f(1,1) , f(1,2) , f(1,3) , f(1,4) , · · · → d1
f(2,1) , f(2,2) , f(2,3) , f(2,4) , · · · → d1 , d2
f(3,1) , f(3,2) , f(3,3) , f(3,4) , · · · → d1 , d2 , d3
f(4,1) , f(4,2) , f(4,3) , f(4,4) , · · · → d1 , d2 , d3 , d4
..
.
© ª∞
The meaning is as follows. f(1,k) k=1 is a subsequence of the original sequence which
∞
converges at d1 . Such a subsequence exists because {fk (d1 )}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
© ªof∞this bounded set is both closed and bounded, hence
weakly compact.) Now f(2,k) k=1 is a subsequence of the first subsequence which
converges, at d2 . Then by Theorem 4.1.6 this new subsequence continues to converge at
d1 . Thus, as indicated by the diagram, it converges at both b1 and b2 . Continuing this
way explains
© theª∞ 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
∞
{fn,k }k=1 and so it converges at d1 , d2 , d3 · · · , d4 . This being true for all n, it follows
© ª∞
f(k,k) k=1 converges at every point of D. To save on notation, I shall simply denote
this as {fk } .
Then letting d ∈ D,

||fk (x) − fl (x)||V ≤ ||fk (x) − fk (d)||V

+ ||fk (d) − fl (d)||V + ||fl (d) − fl (x)||V

Picking d close enough to x and applying equicontinuity,

||fk (x) − fl (x)||V < 2ε/3 + ||fk (d) − fl (d)||V

Thus for k, l large enough, the right side is less than ε. This shows that for each x ∈ K,
∞
{fk (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 5.7.3 it is assumed f has values in F. Show there is no change if f
has values in V, a normed vector space provided
P you redefine the definition of a
polynomial to be something of the form |α|≤m aα xα where aα ∈ V .

2. How would you generalize the conclusion of Corollary 5.7.11 to include the situa-
tion where f has values in a finite dimensional normed vector space?
5.10. EXERCISES 113

3. If {fn } and {gn } are sequences of Fn valued functions defined on D which converge
uniformly, show that if a, b are constants, then afn + bgn also converges uniformly.
If there exists a constant, M such that |fn (x)| , |gn (x)| < M for all n and for all
x ∈ D, show {fn · gn } converges uniformly. Let fn (x) ≡ 1/ |x| for x ∈ B (0,1)
and let gn (x) ≡ (n − 1) /n. Show {fn } converges uniformly on B (0,1) and {gn }
converges uniformly but {fn gn } fails to converge uniformly.

4. Formulate a theorem for series of functions of n variables which will allow you to
conclude the infinite series is uniformly continuous based on reasonable assump-
tions about the functions in the sum.

5. 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 finite collection of continuous functions.
Hint: Note max (f, g) = |f −g|+f
2
+g
. Now recall the triangle inequality which can
be used to show |·| is a continuous function.

6. Find an example of a sequence of continuous functions defined on Rn 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 Rn \ {0} , that is, the
set of vectors in Rn excluding 0.

7. Theorem 5.3.14 says an open subset U of Rn is arcwise connected if and only if U is

connected. Consider the usual Cartesian coordinates relative to axes x1 , · · · , xn .
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 Rn is
connected if and only if every two points can be joined by a square curve.

8. Let x → h (x) be a bounded continuous function. Show the function f (x) =

P∞ h(nx)
n=1 n2 is continuous.

9. Let S be a any countable subset of Rn . Show there exists a function, f defined

on Rn 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.

10. By Theorem 5.7.10 there

Qn exists a sequence of polynomials converging uniformly to
f (x) = |x| on R ≡ k=1 [−M, M ] . Show there exists a sequence of polynomials,
{pn } converging uniformly to f on R which has the additional property that for
all n, pn (0) = 0.

11. If f is any continuous function defined onP∞K a sequentially compact subset of Rn ,

show there exists a series of the form k=1 pk , where each pk 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 infinite sum.

12. A function f is Holder continuous if there exists a constant, K such that

α
|f (x) − f (y)| ≤ K |x − y|

for some α ≤ 1 for all x, y. Show every Holder continuous function is uniformly
continuous.
114 CONTINUOUS FUNCTIONS

13. Consider f (x) ≡ dist (x,S) where S is a nonempty subset of Rn . Show f is

uniformly continuous.
14. 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.
15. 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.
16. Let f : D → R be a function. This function is said to be lower semicontinuous1
at x ∈ D if for any sequence {xn } ⊆ D which converges to x it follows
f (x) ≤ lim inf f (xn ) .
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.
17. Let f : D → R be a function. This function is said to be upper semicontinuous at
x ∈ D if for any sequence {xn } ⊆ D which converges to x it follows
f (x) ≥ lim sup f (xn ) .
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.
18. Show that a real valued function defined on D ⊆ Rn is continuous if and only if
it is both upper and lower semicontinuous.
19. Show that a real valued lower semicontinuous function defined on a sequentially
compact set achieves its minimum and that an upper semicontinuous function
defined on a sequentially compact set achieves its maximum.
20. Give an example of a lower semicontinuous function defined on Rn which is not
continuous and an example of an upper semicontinuous function which is not
continuous.
21. Suppose {fα : α ∈ Λ} is a collection of continuous functions. Let
F (x) ≡ inf {fα (x) : α ∈ Λ}
Show F is an upper semicontinuous function. Next let
G (x) ≡ sup {fα (x) : α ∈ Λ}
Show G is a lower semicontinuous function.
22. Let f be a function. epi (f ) is defined as
{(x, y) : y ≥ f (x)} .
It is called the epigraph of f . We say epi (f ) is closed if whenever (xn , yn ) ∈ epi (f )
and xn → x and yn → 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?
1 The notion of lower semicontinuity is very important for functions which are defined on infinite
dimensional sets. In more general settings, one formulates the concept differently.
5.10. EXERCISES 115

23. The operator norm was defined 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 δ, ∆ such that

δ ||L|| ≤ ||L||1 ≤ ∆ ||L||

for all L ∈ L (V, W ) where here ||·|| denotes the operator norm.

24. One alternative norm which is very popular is as follows. Let L ∈ L (V, W ) and
let (lij ) denote the matrix of L with respect to some bases. Then the Frobenius
norm is defined by
 1/2
X 2
 |lij |  ≡ ||L||F .
ij

Show this is a norm. Other norms are of the form

 1/p
X p
 |lij | 
ij

where p ≥ 1 or even
||L||∞ = max |lij | .
ij

Show these are also norms.

25. Explain why L (V, W ) is always a complete normed vector space whenever V, W
are finite 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.

26. Let L ∈ L (V, V ) where V is a finite dimensional normed vector space. Define
∞
X Lk
eL ≡
k!
k=1

Explain the meaning of this infinite sum and show it converges in L (V, V ) for any
choice of norm on this space. Now tell how to define sin (L).

27. Let X be a finite dimensional normed vector space, real or complex. Show that X
n n
is separable.
PnHint: Let {vPi }i=1 be a basis and define a map from F to X, θ, as
n
follows. θ ( k=1 xk ek ) ≡ k=1 xk vk . Show θ is continuous and has a continuous
inverse. Now let D be a countable dense set in Fn and consider θ (D).

28. Let B (X; Rn ) be the space of functions f , mapping X to Rn such that

sup{|f (x)| : x ∈ X} < ∞.

Show B (X; Rn ) is a complete normed linear space if we define

||f || ≡ sup{|f (x)| : x ∈ X}.

29. Let α ∈ (0, 1]. Define, for X a compact subset of Rp ,

C α (X; Rn ) ≡ {f ∈ C (X; Rn ) : ρα (f ) + ||f || ≡ ||f ||α < ∞}

116 CONTINUOUS FUNCTIONS

where
||f || ≡ sup{|f (x)| : x ∈ X}
and
|f (x) − f (y)|
ρα (f ) ≡ sup{ α : x, y ∈ X, x 6= y}.
|x − y|
Show that (C α (X; Rn ) , ||·||α ) is a complete normed linear space. This is called a
Holder space. What would this space consist of if α > 1?
30. Let {fn }∞ α n p
n=1 ⊆ C (X; R ) where X is a compact subset of R and suppose

||fn ||α ≤ M

for all n. Show there exists a subsequence, nk , such that fnk converges in C (X; Rn ).
The given sequence is precompact when this happens. (This also shows the em-
bedding of C α (X; Rn ) into C (X; Rn ) is a compact embedding.) Hint: You might
want to use the Ascoli Arzela theorem.
31. This problem is for those who know about the derivative and the integral of a
function of one variable. Let f :R × Rn → Rn be continuous and bounded and let
x0 ∈ Rn . If
x : [0, T ] → Rn
and h > 0, let ½
x0 if s ≤ h,
τ h x (s) ≡
x (s − h) , if s > h.
For t ∈ [0, T ], let
Z t
xh (t) = x0 + f (s, τ h xh (s)) ds.
0
Show using the Ascoli Arzela theorem that there exists a sequence h → 0 such
that
xh → x
in C ([0, T ] ; Rn ). Next argue
Z t
x (t) = x0 + f (s, x (s)) ds
0

and conclude the following theorem. If f :R × Rn → Rn is continuous and bounded,

and if x0 ∈ Rn is given, there exists a solution to the following initial value prob-
lem.

x0 = f (t, x) , t ∈ [0, T ]
x (0) = x0 .

This is the Peano existence theorem for ordinary differential equations.

32. Let D (x0 , r) be the closed ball in Rn ,

{x : |x − x0 | ≤ r}

where this is the usual norm coming from the dot product. Let P : Rn → D (x0 , r)
be defined by ½
x if x ∈ D (x0 , r)
P (x) ≡ x−x0
x0 + r |x−x 0|
if x ∈
/ D (x0 , r)
Show that |P x − P y| ≤ |x − y| for all x ∈ Rn .
5.10. EXERCISES 117

33. Use Problem 31 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 differential equation exists on this small interval.
118 CONTINUOUS FUNCTIONS
The Derivative

6.1 Basic Definitions

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 isn’t a well defined 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
defined 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 Fn to Fm 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 defined, for example on subspaces
of Fn and it is nice to not have to trouble with ad hoc considerations. Also, you might
want to consider Fn 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 definitions 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.

Definition 6.1.1 A function g is o (v) if

g (v)
lim =0 (6.1)
||v||→0 ||v||

A function f : U → Y is differentiable 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 definition 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 differentiable
is independent of the norm used on the finite dimensional vector space. That is, a
function is differentiable with one norm if and only if it is differentiable with another
norm.
The definition 6.1 means the error,

f (x + v) − f (x) − Lv

converges to 0 faster than ||v||. Thus the above definition is equivalent to saying

||f (x + v) − f (x) − Lv||

lim =0 (6.2)
||v||→0 ||v||

119
120 THE DERIVATIVE

or equivalently,
||f (y) − f (x) − Df (x) (y − x)||
lim = 0. (6.3)
y→x ||y − x||
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 defined.

Proof: First note that for a fixed vector, v, o (tv) = o (t). This is because

o (tv) o (tv)
lim = lim ||v|| =0
t→0 |t| t→0 ||tv||

Now suppose both L1 and L2 work in the above definition. 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) + L1 tv + o (tv) , f (x + tv) = f (x) + L2 tv + o (tv) .

Therefore, subtracting these two yields (L2 − L1 ) (tv) = o (tv) = o (t). Therefore, di-
viding by t yields (L2 − L1 ) (v) = o(t)t . Now let t → 0 to conclude that (L2 − L1 ) (v) =
0. Since this is true for all v, it follows L2 = L1 . This proves the theorem. ¥

Lemma 6.1.3 Let f be differentiable 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 definition of the derivative,

f (x + v) − f (x) = Df (x) v + o (v) .

o(||v||)
Let ||v|| be small enough that ||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 differentiable at x ∈ U and suppose g : V → Fq is differentiable
at f (x) ∈ V . Then g ◦ f is differentiable at x and

D (g ◦ f ) (x) = D (g (f (x))) D (f (x)) .

6.3. THE MATRIX OF THE DERIVATIVE 121

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 definition of differentiability 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 ε > 0, it follows that for ||v|| small
enough,

||o (f (x + v) − f (x))|| ≤ (ε/K) ||f (x + v) − f (x)|| ≤ (ε/K) K ||v|| = ε ||v|| .

Since ε > 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))||
≤ ε.
||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 {v1 , · · ·, vn } and a basis for Y
being {w1 , · · ·, wm } . First note that if π i : X → F is defined by
X
π i v ≡ xi where v = xk vk ,
k

then π i ∈ L (X, F) and so by Theorem 5.8.3, it follows that π i is continuous and if

lims→t g (s) = L, then |π i g (s) − π i L| ≤ ||π i || ||g (s) − L|| and so the ith components
converge also.
Suppose that f : U → Y is differentiable. What is the matrix of Df (x) with respect
to the given bases? That is, if
X
Df (x) = Jij (x) wi vj ,
ij

what is Jij (x)?

f (x + tvk ) − f (x) Df (x) (tvk ) + o (tvk )

Dvk f (x) ≡ lim = lim
t→0 t t→0 t
X X
= Df (x) (vk ) = Jij (x) wi vj (vk ) = Jij (x) wi δ jk
ij ij
X
= Jik (x) wi
i
122 THE DERIVATIVE

It follows
µ ¶
f (x + tvk ) − f (x)
lim π j
t→0 t
fj (x + tvk ) − fj (x)
≡ lim ≡ Dvk fj (x)
t→0
à t !
X
= πj Jik (x) wi = Jjk (x)
i

Thus Jik (x) = Dvk fi (x).

In the case where X = Rn and Y = Rm and v is a unit vector, Dv fi (x) is the
familiar directional derivative in the direction v of the function, fi .
Of course the case where X = Fn and f : U ⊆ Fn → Fm , is differentiable 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 ei denote the
vector of Fn which has a one in the ith entry and zeroes elsewhere. This is the standard
basis for Fn . Denote by Jij (x) the matrix with respect to these basis vectors. Thus
X
Df (x) = Jij (x) ei ej .
ij

Then from what was just shown,

fi (x + tek ) − fi (x)
Jik (x) = Dek fi (x) ≡ lim
t→0 t
∂fi
≡ (x) ≡ fi,xk (x) ≡ fi,k (x)
∂xk
where the last several symbols are just the usual notations for the partial derivative of
the function, fi with respect to the k th variable where
m
X
f (x) ≡ fi (x) ei .
i=1

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 (Jij ) is
∂f f (x + tek ) − f (x)
(x) = lim ≡ Dek f (x) .
∂xk t→0 t
Thus the matrix of Df (x) with respect to the usual basis vectors is the matrix of
the form  
f1,x1 (x) f1,x2 (x) · · · f1,xn (x)
 .. .. .. 
 . . . 
fm,x1 (x) fm,x2 (x) · · · fm,xn (x)
where the notation g,xk denotes the k th partial derivative given by the limit,
g (x + tek ) − g (x) ∂g
lim ≡ .
t→0 t ∂xk
The above discussion is summarized in the following theorem.
Theorem 6.3.1 Let f : Fn → Fm and suppose f is differentiable at x. Then all
∂fi (x)
the partial derivatives ∂xj exist and if Jf (x) is the matrix of the linear transformation,
∂fi
Df (x) with respect to the standard basis vectors, then the ij th entry is given by ∂xj (x)
also denoted as fi,j or fi,xj .
6.4. A MEAN VALUE INEQUALITY 123

Definition 6.3.2 In general, the symbol

Dv f (x)

is defined by
f (x + tv) − f (x)
lim
t→0 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 differentiable?
Consider the following function, f : R2 → R,
½ xy
f (x, y) = x2 +y 2 if (x, y) 6= (0, 0) .
0 if (x, y) = (0, 0)
Then from the definition of partial derivatives,
f (h, 0) − f (0, 0) 0−0
lim = lim =0
h→0 h h→0 h
and
f (0, h) − f (0, 0) 0−0
lim = lim =0
h→0 h h→0 h
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 differentiable. Therefore, it is necessary to consider the correct definition
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 differentiable.

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 differ-
entiable and satisfies
||h0 (t)|| ≤ M.
Then
||h (1) − h (0)|| ≤ M.

Proof: Let ε > 0 be given and let

S ≡ {t ∈ [0, 1] : for all s ∈ [0, t] , ||h (s) − h (0)|| ≤ (M + ε) s}

Then 0 ∈ S. Let t = sup S. Then by continuity of h it follows

||h (t) − h (0)|| = (M + ε) t (6.5)

Suppose t < 1. Then there exist positive numbers, hk decreasing to 0 such that

||h (t + hk ) − h (0)|| > (M + ε) (t + hk )

and now it follows from 6.5 and the triangle inequality that

||h (t + hk ) − h (t)|| + ||h (t) − h (0)||

= ||h (t + hk ) − h (t)|| + (M + ε) t > (M + ε) (t + hk )
124 THE DERIVATIVE

and so
||h (t + hk ) − h (t)|| > (M + ε) hk
Now dividing by hk and letting k → ∞

||h0 (t)|| ≥ M + ε,

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,

h0 (t) = Df (x + t (y − x)) (y − x)

and so

||h0 (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|| .

This proves the theorem. ¥

6.5 Existence Of The Derivative, C 1 Functions

There is a way to get the differentiability 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 {v1 , · · ·, vn } and
let Y be another normed vector space having basis {w1 , · · ·, wm } . Let U be an open set
in X and let f : U → Y have the property that the Gateaux derivatives,
f (x + tvk ) − f (x)
Dvk f (x) ≡ lim
t→0 t
exist and are continuous functions of x. Then Df (x) exists and
n
X
Df (x) v = Dvk f (x) ak
k=1

where
n
X
v= ak vk .
k=1
6.5. EXISTENCE OF THE DERIVATIVE, C 1 FUNCTIONS 125

Furthermore, x → Df (x) is continuous; that is

lim ||Df (y) − Df (x)|| = 0.

y→x
Pn
Proof: Let v = k=1 ak vk . Then
à n
!
X
f (x + v) − f (x) = f x+ ak vk − f (x) .
k=1
P0
Then letting k=1 ≡ 0, f (x + v) − f (x) is given by
    
Xn Xk k−1
X
f x+ aj vj  − f x+ aj vj 
k=1 j=1 j=1

n
X
= [f (x + ak vk ) − f (x)] +
k=1
       
n
X k
X k−1
X
f x+ aj vj  − f (x + ak vk ) − f x+ aj vj  − f (x) (6.6)
k=1 j=1 j=1

Consider the k th term in 6.6. Let

 
k−1
X
h (t) ≡ f x+ aj vj + tak vk  − f (x + tak vk )
j=1

for t ∈ [0, 1] . Then

  
k−1
X
1  
h0 (t) = ak lim f x+ aj vj + (t + h) ak vk  − f (x + (t + h) ak vk )
h→0 ak h
j=1
   
k−1
X
− f x+ aj vj + tak vk  − f (x + tak vk )
j=1

and this equals

   
k−1
X
Dvk f x+ aj vj + tak vk  − Dvk f (x + tak vk ) ak (6.7)
j=1

Now without loss of generality, it can be assumed the norm on X is given by that of
Example 5.8.5,  
 n
X 
||v|| ≡ max |ak | : v = ak vk
 
j=1

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,
¯¯    ¯¯
¯¯ k−1 ¯¯
¯ ¯ X ¯¯
0 ¯ ¯  
||h (t)|| = ¯¯ Dvk f x+ aj vj + tak vk − Dvk f (x + tak vk ) ak ¯¯¯¯
 
¯¯ j=1 ¯¯
≤ ε |ak | ≤ ε ||v||
126 THE DERIVATIVE

provided ||v|| is sufficiently small. Since ε is arbitrary, it follows from Lemma 6.4.1 the
expression in 6.6 is o (v) because this expression equals a finite sum of terms of the form
h (1) − h (0) where ||h0 (t)|| ≤ ε ||v|| . Thus
n
X
f (x + v) − f (x) = [f (x + ak vk ) − f (x)] + o (v)
k=1
n
X n
X
= Dvk f (x) ak + [f (x + ak vk ) − f (x) − Dvk f (x) ak ] + o (v) .
k=1 k=1

Consider the k th term in the second sum.

µ ¶
f (x + ak vk ) − f (x)
f (x + ak vk ) − f (x) − Dvk f (x) ak = ak − Dvk f (x)
ak
where the expression in the parentheses converges to 0 as ak → 0. Thus whenever ||v||
is sufficiently small,
||f (x + ak vk ) − f (x) − Dvk f (x) ak || ≤ ε |ak | ≤ ε ||v||
which shows the second sum is also o (v). Therefore,
n
X
f (x + v) − f (x) = Dvk f (x) ak + o (v) .
k=1

Defining
n
X
Df (x) v ≡ Dvk f (x) ak
k=1
P
where v = k ak vk , 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||
X n
≤ ||(Dvk f (x) − Dvk f (y)) ak ||
k=1
n
X
≤ max {|ak | , k = 1, · · ·, n} ||Dvk f (x) − Dvk f (y)||
k=1
n
X
= ||v|| ||Dvk f (x) − Dvk f (y)||
k=1

and so
n
X
||Df (x) − Df (y)|| ≤ ||Dvk f (x) − Dvk f (y)||
k=1
which proves the continuity of Df because of the assumption the Gateaux derivatives
are continuous. This proves the theorem. ¥
This motivates the following definition of what it means for a function to be C 1 .
Definition 6.5.2 Let U be an open subset of a normed finite dimensional vector
space, X and let f : U → Y another finite dimensional normed vector space. Then f is
said to be C 1 if there exists a basis for X, {v1 , · · ·, vn } such that the Gateaux derivatives,
Dvk f (x)
exist on U and are continuous.
6.6. HIGHER ORDER DERIVATIVES 127

Here is another definition of what it means for a function to be C 1 .

Definition 6.5.3 Let U be an open subset of a normed finite dimensional vector

space, X and let f : U → Y another finite dimensional normed vector space. Then f
is said to be C 1 if f is differentiable and x →Df (x) is continuous as a map from U to
L (X, Y ).

Now the following major theorem states these two definitions are equivalent.

Theorem 6.5.4 Let U be an open subset of a normed finite dimensional vector

space, X and let f : U → Y another finite dimensional normed vector space. Then the
two definitions above are equivalent.

Proof: It was shown in Theorem 6.5.1 that Definition 6.5.2 implies 6.5.3. Suppose
then that Definition 6.5.3 holds. Then if v is any vector,
f (x + tv) − f (x) Df (x) tv + o (tv)
lim = lim
t→0 t t→0 t
o (tv)
= Df (x) v+ lim = Df (x) v
t→0 t
Thus Dv f (x) exists and equals Df (x) v. By continuity of x → Df (x) , this establishes
continuity of x → Dv 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 finite dimensional vector

space, X and let f : U → Y another finite dimensional normed vector space. Then if
there is a basis of X, {v1 , · · ·, vn } such that the Gateaux derivatives, Dvk f (x) exist and
are continuous. Then all Gateaux derivatives, Dv f (x) exist and are continuous for all
v ∈ X.

From now on, whichever definition 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 differentiable.

Definition 6.6.1 The following is the definition of the second derivative.

D2 f (x) ≡ D (Df (x)) .

Thus,
Df (x + v) − Df (x) = D2 f (x) v + o (v) .
This implies
D2 f (x) ∈ L (X, L (X, Y )) , D2 f (x) (u) (v) ∈ Y,
and the map
(u, v) → D2 f (x) (u) (v)
is a bilinear map having values in Y . In other words, the two functions,

u → D2 f (x) (u) (v) , v → D2 f (x) (u) (v)

128 THE DERIVATIVE

are both linear.

The same pattern applies to taking higher order derivatives. Thus,
¡ ¢
D3 f (x) ≡ D D2 f (x)

and D3 f (x) may be considered as a trilinear map having values in Y . In general Dk f (x)
may be considered a k linear map. This means the function

(u1 , · · ·, uk ) → Dk f (x) (u1 ) · · · (uk )

has the property

uj → Dk f (x) (u1 ) · · · (uj ) · · · (uk )
is linear.
Also, instead of writing

D2 f (x) (u) (v) , or D3 f (x) (u) (v) (w)

the following notation is often used.

D2 f (x) (u, v) or D3 f (x) (u, v, w)

with similar conventions for higher derivatives than 3. Another convention which is
often used is the notation
Dk f (x) vk
instead of
Dk f (x) (v, · · ·, v) .
Note that for every k, Dk f maps U to a normed vector space. As mentioned above,
Df (x) has values in L (X, Y ) , D2 f (x) has values in L (X, L (X, Y )) , etc. Thus it makes
sense to consider whether Dk f is continuous. This is described in the following definition.

Definition 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 first k derivatives are all continuous. Also,
Dk 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
X X
Df (x) v = Dvj f (x) aj = Dvj fi (x) wi aj
j ij
à !
X X X
= Dvj fi (x) wi vj ak vk = Dvj fi (x) wi vj (v)
ij k ij
P
where k ak vk = v and X
f (x) = fi (x) wi . (6.8)
i

This is because à !
X X
wi vj ak vk ≡ ak wi δ jk = wi aj .
k k

Thus X
Df (x) = Dvj fi (x) wi vj
ij
6.7. C K FUNCTIONS 129

I propose to iterate this observation, starting with f and then going to Df and then
D2 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,
X
Df (x) = Dvj1 fi (x) wi vj1 .
ij1

Then letting wi vj1 play the role of wi in 6.8,

X ¡ ¢
D2 f (x) = Dvj2 Dvj1 fi (x) wi vj1 vj2
ij1 j2
X
≡ Dvj1 vj2 fi (x) wi vj1 vj2
ij1 j2

Then letting wi vj1 vj2 play the role of wi in 6.8,

X ¡ ¢
D3 f (x) = Dvj3 Dvj1 vj2 fi (x) wi vj1 vj2 vj3
ij1 j2 j3
X
≡ Dvj1 vj2 vj3 fi (x) wi vj1 vj2 vj3
ij1 j2 j3

etc. In general, the notation,

wi vj1 vj2 · · · vjn
defines an appropriate linear transformation given by
wi vj1 vj2 · · · vjn (vk ) = wi vj1 vj2 · · · vjn−1 δ kjn .
The following theorem is important.
Theorem 6.7.1 The function x → Dk f (x) exists and is continuous for k ≤ p
if and only if there exists a basis for X, {v1 , · · ·, vn } and a basis for Y, {w1 , · · ·, wm }
such that for X
f (x) ≡ fi (x) wi ,
i
it follows that for each i = 1, 2, · · ·, m all Gateaux derivatives,
Dvj1 vj2 ···vjk fi (x)
for any choice of vj1 vj2 · · · vjk 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 differentiation. ¥
Definition 6.7.2 Let X, Y be finite dimensional normed vector spaces and let
U be an open set in X and f : U → Y be a function,
X
f (x) = fi (x) wi
i

where {w1 , · · ·, wm } is a basis for Y. Then f is said to be a C n (U ) function if for every

k ≤ n, Dk 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,
Dvj1 vj2 ···vjk fi (x)
for any choice of vj1 vj2 · · · vjk where {v1 , · · ·, vn } is a basis for X and for any k ≤ n
exist and are continuous.
130 THE DERIVATIVE

6.7.1 Some Standard Notation

In the case where X = Rn 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 definition.
Definition 6.7.3 Let g : U → X. Then
∂g g (x + hek ) − g (x)
gxk (x) ≡ (x) ≡ lim
∂xk h→0 h
Higher order partial derivatives are defined in the usual way.
∂2g
gxk xl (x) ≡ (x)
∂xl ∂xk
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 definition.
Definition 6.7.4 α = (α1 , · · ·, αn ) for α1 · · · αn positive integers is called a
multi-index. For α a multi-index, |α| ≡ α1 + · · · + αn and if x ∈ X,
x = (x1 , · · ·, xn ),
and f a function, define
∂ |α| f (x)
xα ≡ xα 1 α2 αn α
1 x2 · · · xn , D f (x) ≡ .
∂xα α2
1 ∂x2
1
· · · ∂xα
n
n

Then in this special case, the following definition is equivalent to the above as a
definition of what is meant by a C k function.
Definition 6.7.5 Let U be an open subset of Rn and let f : U → Y. Then for k
a nonnegative integer, f is C k if for every |α| ≤ k, Dα f exists and is continuous.

6.8 The Derivative And The Cartesian Product

There are theorems which can be used to get differentiability of a function based on
existence and continuity of the partial derivatives. A generalization of this was given
above. Here a function defined 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 definition. First here is a lemma.
Lemma 6.8.1 Suppose U is an open set in X × Y. Then the set, Uy defined by
Uy ≡ {x ∈ X : (x, y) ∈ U }
is an open set in X. Here X × Y is a finite dimensional vector space in which the vector
space operations are defined componentwise. Thus for a, b ∈ F,
a (x1 , y1 ) + b (x2 , y2 ) = (ax1 + bx2 , ay1 + by2 )
and the norm can be taken to be
||(x, y)|| ≡ max (||x|| , ||y||)
6.8. THE DERIVATIVE AND THE CARTESIAN PRODUCT 131

Proof: Recall by Theorem 5.8.4 it does not matter how this norm is defined and
the definition above is convenient. It obviously satisfies most axioms of a norm. The
only one which is not obvious is the triangle inequality. I will show this now.

||(x, y) + (x1 , y1 )|| = ||(x + x1 , y + y1 )|| ≡ max (||x + x1 || , ||y + y1 ||)

≤ max (||x|| + ||x1 || , ||y|| + ||y1 ||)

suppose then that ||x|| + ||x1 || ≥ ||y|| + ||y1 || . Then the above equals

||x|| + ||x1 || ≤ max (||x|| , ||y||) + max (||x1 || , ||y1 ||) ≡ ||(x, y)|| + ||(x1 , y1 )||

In case ||x|| + ||x1 || < ||y|| + ||y1 || , the argument is similar.

Let x ∈ Uy . 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 Uy .
This proves the lemma. ¥
Or course one could also consider

Ux ≡ {y : (x, y) ∈ U }

in the same way and conclude this set is open in Y . Also, Qn the generalization to many
factors yields the same conclusion. In this case, for x ∈ i=1 Xi , let
¡ ¢
||x|| ≡ max ||xi ||Xi : x = (x1 , · · ·, xn )
Qn
Then a similar argument to the above shows this is a norm on i=1 Xi .
Q
Corollary 6.8.2 Let U ⊆ ni=1 Xi and let
© ¡ ¢ ª
U(x1 ,···,xi−1 ,xi+1 ,···,xn ) ≡ x ∈ Fri : x1 , · · ·, xi−1 , x, xi+1 , · · ·, xn ∈ U .

Then U(x1 ,···,xi−1 ,xi+1 ,···,xn ) is an open set in Fri .

The proof is similar to the above.

Qn
Definition 6.8.3 Let g : U ⊆ i=1 Xi → Y , where U is an open set. Then the
map ¡ ¢
z → g x1 , · · ·, xi−1 , z, xi+1 , · · ·, xn
is a function from the open set in Xi ,
© ¡ ¢ ª
z : x = x1 , · · ·, xi−1 , z, xi+1 , · · ·, xn ∈ U

to Y . When this map is differentiable,

Qn its derivative is denoted by Di g (x). To aid in
the notation, for v ∈ Xi , letQθi v ∈ i=1 Xi be the vector (0, · · ·, v, · · ·, 0) where the v
n
is in the ith slot and for v ∈ i=1 Xi , let vi denote the entry in the ith slot of v. Thus,
by saying ¡ ¢
z → g x1 , · · ·, xi−1 , z, xi+1 , · · ·, xn
is differentiable is meant that for v ∈ Xi sufficiently small,

g (x + θi v) − g (x) = Di g (x) v + o (v) .

Note Di g (x) ∈ L (Xi , Y ) .

132 THE DERIVATIVE

Definition 6.8.4 Let U ⊆ X be an open set. Then f : U → Y is C 1 (U ) if f is

differentiable and the mapping
x →Df (x) ,
is continuous as a function from U to L (X, Y ).

With this definition of partial derivatives, here is the major theorem.

Qn
Theorem 6.8.5 Let g, U, i=1 Xi , be given as in Definition 6.8.3. Then g is
1
C (U ) if and only if Di g exists and is continuous on U for each i. In this case, g is
differentiable and X
Dg (x) (v) = Dk g (x) vk (6.9)
k

where v = (v1 , · · ·, vn ) .

Proof: Suppose then that Di g exists and is continuous for each i. Note that
k
X
θj vj = (v1 , · · ·, vk , 0, · · ·, 0) .
j=1

Pn P0
Thus j=1 θj vj = v and define j=1 θj vj ≡ 0. Therefore,
    
n
X k
X k−1
X
g (x + v) − g (x) = g x+ θj vj  − g x + θ j vj  (6.10)
k=1 j=1 j=1

Consider the terms in this sum.

   
X k k−1
X
g x+ θj vj  − g x + θj vj  = g (x+θk vk ) − g (x) + (6.11)
j=1 j=1

       
k
X k−1
X
g x+ θj vj  − g (x+θk vk ) − g x + θj vj  − g (x) (6.12)
j=1 j=1

and the expression in 6.12 is of the form h (vk ) − h (0) where for small w ∈ Xk ,
 
k−1
X
h (w) ≡ g x+ θj vj + θk w − g (x + θk w) .
j=1

Therefore,  
k−1
X
Dh (w) = Dk g x+ θj vj + θk w − Dk g (x + θk w)
j=1

and by continuity, ||Dh (w)|| < ε provided ||v|| is small enough. Therefore, by Theorem
6.4.2, whenever ||v|| is small enough,

||h (vk ) − h (0)|| ≤ ε ||vk || ≤ ε ||v||

which shows that since ε is arbitrary, the expression in 6.12 is o (v). Now in 6.11

g (x+θk vk ) − g (x) = Dk g (x) vk + o (vk ) = Dk g (x) vk + o (v) .

6.9. MIXED PARTIAL DERIVATIVES 133

Therefore, referring to 6.10,

n
X
g (x + v) − g (x) = Dk g (x) vk + o (v)
k=1

which shows Dg (x) exists and equals the formula given in 6.9.
Next suppose g is C 1 . I need to verify that Dk g (x) exists and is continuous. Let
v ∈ Xk sufficiently small. Then
g (x + θk v) − g (x) = Dg (x) θk v + o (θk v)
= Dg (x) θk v + o (v)
since ||θk v|| = ||v||. Then Dk g (x) exists and equals
Dg (x) ◦ θk
Now x → DgQn (x) is continuous. Since θk is linear, it follows from Theorem 5.8.3 that
θk : Xk → i=1 Xi 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 Xi = F in the above theorem.
Corollary 6.8.6 Let U be an open subset of Fn and let f :U → Fm be C 1 in the sense
that all the partial derivatives of f exist and are continuous. Then f is differentiable
and
Xn
∂f
f (x + v) = f (x) + (x) vk + o (v) .
∂xk
k=1

6.9 Mixed Partial Derivatives

Continuing with the special case where f is defined on an open set in Fn , 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
fx = = De1 f
∂x
and
∂2f
fxy = = De1 e2 f.
∂y∂x

Theorem 6.9.1 Suppose f : U ⊆ F2 → R where U is an open set on which

fx , fy , fxy and fyx exist. Then if fxy and fyx are continuous at the point (x, y) ∈ U , it
follows
fxy (x, y) = fyx (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
h(t) h(0)
1 z }| { z }| {
∆ (s, t) ≡ {f (x + t, y + s) − f (x + t, y) − (f (x, y + s) − f (x, y))}. (6.13)
st
Note that (x + t, y + s) ∈ U because
¡ ¢1/2
|(x + t, y + s) − (x, y)| = |(t, s)| = t2 + s2
µ 2 ¶1/2
r r2 r
≤ + = √ < r.
4 4 2
134 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,
1 1
∆ (s, t) = (h (t) − h (0)) = h0 (αt) t
st st
1
= (fx (x + αt, y + s) − fx (x + αt, y))
s
for some α ∈ (0, 1) . Applying the mean value theorem again,

∆ (s, t) = fxy (x + αt, y + βs)

where α, β ∈ (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 γ, δ ∈ (0, 1) such that

∆ (s, t) = fyx (x + γt, y + δs) .

Letting (s, t) → (0, 0) and using the continuity of fxy and fyx at (x, y) ,

lim ∆ (s, t) = fxy (x, y) = fyx (x, y) .

(s,t)→(0,0)

This proves the theorem. ¥

The following is obtained from the above by simply fixing 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, fxk , fxl , fxl xk , and fxk xl exist on U and fxk xl and fxl xk are
both continuous at x ∈ U. Then fxk xl (x) = fxl xk (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 Fn and f : U → F has the property

that for two indices, k, l, fxk , fxl , fxl xk , and fxk xl exist on U and fxk xl and fxl xk are
both continuous at x ∈ U. Then fxk xl (x) = fxl xk (x) .

Finally, by considering the components of f you get the following generalization.

Corollary 6.9.4 Suppose U is an open subset of Fn and f : U → F m has the property

that for two indices, k, l, fxk , fxl , fxl xk , and fxk xl exist on U and fxk xl and fxl xk are both
continuous at x ∈ U. Then fxk xl (x) = fxl xk (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

(
xy (x2 −y 2 )
f (x, y) = x2 +y 2 if (x, y) 6= (0, 0)
0 if (x, y) = (0, 0)

From the definition of partial derivatives it follows immediately that fx (0, 0) =

fy (0, 0) = 0. Using the standard rules of differentiation, for (x, y) 6= (0, 0) ,

x4 − y 4 + 4x2 y 2 x4 − y 4 − 4x2 y 2
fx = y 2 , fy = x 2
(x2 + y 2 ) (x2 + y 2 )
6.10. IMPLICIT FUNCTION THEOREM 135

Now
fx (0, y) − fx (0, 0)
fxy (0, 0) ≡ lim
y→0 y
−y 4
= lim = −1
y→0 (y 2 )2

while
fy (x, 0) − fy (0, 0)
fyx (0, 0) ≡ lim
x→0 x
x4
= lim =1
x→0 (x2 )2

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 finite dimensional normed vector

space and suppose ||A|| ≤ r < 1. Then
−1
(I − A) exists (6.14)

and ¯¯ ¯¯
¯¯ −1 ¯¯ −1
¯¯(I − A) ¯¯ ≤ (1 − r) . (6.15)

Furthermore, if
© ª
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 6= 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 {v1 , · · ·, vn } be a basis for X
and suppose
Xn
ck (I − A) vk = 0.
k=1

Then à !
n
X
(I − A) ck vk =0
k=1

and since (I − A) is one to one, it follows

n
X
ck vk = 0
k=1
136 THE DERIVATIVE

n
which requires each ck = 0 because the {vk } are independent. Hence {(I − A) vk }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, ck such that
n
à n !
X X
y= ck (I − A) vk = (I − A) ck vk
k=1 k=1

−1
so (I − A) is onto as claimed. Thus (I − A) ∈ L (X, X) and it remains to estimate
its norm.

||x − Ax|| ≥ ||x|| − ||Ax|| ≥ ||x|| − ||A|| ||x|| ≥ ||x|| (1 − r)

−1
Letting y = x − Ax so x = (I − A) y, this shows, since (I − A) is onto that for all
y ∈ X, ¯¯ ¯¯
¯¯ −1 ¯¯
||y|| ≥ ¯¯(I − A) y¯¯ (1 − r)
¯¯ ¯¯
¯¯ −1 ¯¯ −1
and so ¯¯(I − A) ¯¯ ≤ (1 − r) . This proves the first part.
To verify the continuity of the inverse map, let A ∈ I. Then
¡ ¢
B = A I − A−1 (A − B)
¯¯ ¯¯
and so if ¯¯A−1 (A − B)¯¯ < 1 which, by Theorem 5.8.3, happens if
¯¯ ¯¯
||A − B|| < 1/ ¯¯A−1 ¯¯ ,
¡ ¢−1
it follows from the first part of this proof that I − A−1 (A − B) exists and so
¡ ¢−1 −1
B −1 = I − A−1 (A − B) A

which shows I is open. Also, if

¯¯ −1 ¯¯
¯¯A (A − B)¯¯ ≤ r < 1, (6.16)
¯¯ −1 ¯¯ ¯¯ −1 ¯¯
¯¯B ¯¯ ≤ ¯¯A ¯¯ (1 − r)−1
Now for such B this close to A such that 6.16 holds,
¯¯ −1 ¯¯ ¯¯ ¯¯ ¯¯ ¯¯ ¯¯ ¯¯
¯¯B − A−1 ¯¯ = ¯¯B −1 (A − B) A−1 ¯¯ ≤ ||A − B|| ¯¯B −1 ¯¯ ¯¯A−1 ¯¯
¯¯ ¯¯2
≤ ||A − B|| ¯¯A−1 ¯¯ (1 − r)
−1

which shows the map which takes a linear transformation in I to its inverse is continuous.
This proves the lemma. ¥
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 differential equations and integral equations. It is sometimes called the uniform
contraction principle.

Theorem 6.10.2 Let X, Y be finite 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 satisfies

||T (x, y) − T (x0 , y)|| ≤ r ||x − x0 || (6.17)

where 0 < r < 1 and also

||T (x, y) − T (x, y0 )|| ≤ M ||y − y0 || . (6.18)

6.10. IMPLICIT FUNCTION THEOREM 137

Then for each y ∈ F there exists a unique “fixed point” for T (·, y) , x ∈ E, satisfying

T (x, y) = x (6.19)

and also if x (y) is this fixed point,

M
||x (y) − x (y0 )|| ≤ ||y − y0 || . (6.20)
1−r

Proof: First consider the claim there exists a fixed point for the mapping, T (·, y).
For a fixed y, let g (x) ≡ T (x, y). Now pick any x0 ∈ E and consider the sequence,

x1 = g (x0 ) , xk+1 = g (xk ) .

Then by 6.17,

||xk+1 − xk || = ||g (xk ) − g (xk−1 )|| ≤ r ||xk − xk−1 || ≤

r2 ||xk−1 − xk−2 || ≤ · · · ≤ rk ||g (x0 ) − x0 || .

Now by the triangle inequality,
p
X
||xk+p − xk || ≤ ||xk+i − xk+i−1 ||
i=1

p
X rk ||g (x0 ) − x0 ||
≤ rk+i−1 ||g (x0 ) − x0 || ≤ .
i=1
1−r
∞
Since 0 < r < 1, this shows that {xk }k=1 is a Cauchy sequence. Therefore, by com-
pleteness of E it converges to a point x ∈ E. To see x is a fixed point, use the continuify
of g to obtain
x ≡ lim xk = lim xk+1 = lim g (xk ) = g (x) .
k→∞ k→∞ k→∞

This proves 6.19. To verify 6.20,

||x (y) − x (y0 )|| = ||T (x (y) , y) − T (x (y0 ) , y0 )|| ≤

||T (x (y) , y) − T (x (y) , y0 )|| + ||T (x (y) , y0 ) − T (x (y0 ) , y0 )||

≤ M ||y − y0 || + r ||x (y) − x (y0 )|| .

Thus
(1 − r) ||x (y) − x (y0 )|| ≤ M ||y − y0 || .
This also shows the fixed 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 infinite
dimensional complete normed vector spaces when suitable modifications 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||) .

138 THE DERIVATIVE

Theorem 6.10.3 (implicit function theorem) Let X, Y, Z be finite 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
−1
f (x0 , y0 ) = 0, D1 f (x0 , y0 ) ∈ L (Z, X) . (6.21)

Then there exist positive constants, δ, η, such that for every y ∈ B (y0 , η) there exists a
unique x (y) ∈ B (x0 , δ) such that

f (x (y) , y) = 0. (6.22)

Furthermore, the mapping, y → x (y) is in C 1 (B (y0 , η)).

−1
Proof: Let T (x, y) ≡ x − D1 f (x0 , y0 ) f (x, y). Therefore,
−1
D1 T (x, y) = I − D1 f (x0 , y0 ) D1 f (x, y) . (6.23)

by continuity of the derivative which implies continuity of D1 T, it follows there exists

δ > 0 such that if ||(x − x0 , y − y0 )|| < δ, then

1
||D1 T (x, y)|| < . (6.24)
2
Also, it can be assumed δ is small enough that
¯¯ ¯¯
¯¯ −1 ¯¯
¯¯D1 f (x0 , y0 ) ¯¯ ||D2 f (x, y)|| < M (6.25)
¯¯ ¯¯
¯¯ −1 ¯¯
where M > ¯¯D1 f (x0 , y0 ) ¯¯ ||D2 f (x0 , y0 )||. By Theorem 6.4.2, whenever x, x0 ∈
B (x0 , δ) and y ∈ B (y0 , δ),

1
||T (x, y) − T (x0 , y)|| ≤ ||x − x0 || . (6.26)
2
Solving 6.23 for D1 f (x, y) ,

D1 f (x, y) = D1 f (x0 , y0 ) (I − D1 T (x, y)) .

−1 −1
By Lemma 6.10.1 and the assumption that D1 f (x0 , y0 ) exists, it follows, D1 f (x, y)
exists and equals
−1 −1
(I − D1 T (x, y)) D1 f (x0 , y0 )
By the estimate of Lemma 6.10.1 and 6.24,
¯¯ ¯¯ ¯¯ ¯¯
¯¯ −1 ¯¯ ¯¯ −1 ¯¯
¯¯D1 f (x, y) ¯¯ ≤ 2 ¯¯D1 f (x0 , y0 ) ¯¯ . (6.27)

Next more restrictions are placed on y to make it even closer to y0 . Let

µ ¶
δ
0 < η < min δ, .
3M

Then suppose x ∈ B (x0 , δ) and y ∈ B (y0 , η). Consider

−1
x − D1 f (x0 , y0 ) f (x, y) − x0 = T (x, y) − x0 ≡ g (x, y) .
−1
D1 g (x, y) = I − D1 f (x0 , y0 ) D1 f (x, y) = D1 T (x, y) ,
6.10. IMPLICIT FUNCTION THEOREM 139

and
−1
D2 g (x, y) = −D1 f (x0 , y0 ) D2 f (x, y) .
Also note that T (x, y) = x is the same as saying f (x0 , y0 ) = 0 and also g (x0 , y0 ) = 0.
Thus by 6.25 and Theorem 6.4.2, it follows that for such (x, y) ∈ B (x0 , δ) × B (y0 , η),

||T (x, y) − x0 || = ||g (x, y)|| = ||g (x, y) − g (x0 , y0 )||

≤ ||g (x, y) − g (x, y0 )|| + ||g (x, y0 ) − g (x0 , y0 )||

1 δ δ 5δ
≤ M ||y − y0 || + ||x − x0 || < + = < δ. (6.28)
2 2 3 6
Also for such (x, yi ) , i = 1, 2, Theorem 6.4.2 and 6.25 implies
¯¯ ¯¯
¯¯ −1 ¯¯
||T (x, y1 ) − T (x, y2 )|| = ¯¯D1 f (x0 , y0 ) (f (x, y2 ) − f (x, y1 ))¯¯

≤ M ||y2 − y1 || . (6.29)
From now on assume ||x − x0 || < δ and ||y − y0 || < η 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 x0 , 5δ 6 and F ≡ B (y0 , η) implies that for each y ∈ B (y0 , η), there
¡ ¢
exists a unique x (y) ∈ B (x0 , δ) (actually in B x0 , 5δ6 ) such that T (x (y) , y) = x (y)
which is equivalent to
f (x (y) , y) = 0.
Furthermore,
||x (y) − x (y0 )|| ≤ 2M ||y − y0 || . (6.30)
This proves the implicit function theorem except for the verification that y → x (y)
is C 1 . This is shown next. Letting v be sufficiently small, Theorem 6.8.5 and Theorem
6.4.2 imply
0 = f (x (y + v) , y + v) − f (x (y) , y) =
D1 f (x (y) , y) (x (y + v) − x (y)) +
+D2 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
−1
x (y + v) − x (y) = −D1 (x (y) , y) D2 f (x (y) , y) v + o (v)

Which shows that y → x (y) is differentiable on B (y0 , η) and

−1
Dx (y) = −D1 f (x (y) , y) D2 f (x (y) , y) . (6.31)

Now it follows from the continuity of D2 f , D1 f , the inverse map, 6.30, and this formula
for Dx (y)that x (·) is C 1 (B (y0 , η)). 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 x0 ∈ U , an open set in X ,

and let f : U → Y where X, Y are finite dimensional normed vector spaces. Suppose

f is C 1 (U ) , and Df (x0 )−1 ∈ L(Y, X). (6.32)

140 THE DERIVATIVE

Then there exist open sets, W , and V such that

x0 ∈ 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 y0 ≡ f (x0 ). Thus the function y → x (y) defined in that theorem is f −1 . Now
let
W ≡ B (x0 , δ) ∩ f −1 (B (y0 , η))
and
V ≡ B (y0 , η) .
This proves the theorem. ¥

6.10.1 More Derivatives

In the implicit function theorem, suppose f is C k . Will the implicitly defined 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 D2 f (x (y) , y) ∈ L (Y, Z) . Let {w1 , · · ·, wm } be a basis for Y
and let {z1 , · · ·, zn } be a basis for Z. Then D2 f (x (y) , y) has a matrix
³ with respect ´to
these bases. Thus conserving on notation, denote this matrix by D2 f (x (y) , y)ij .
Thus X
D2 f (x (y) , y) = D2 f (x (y) , y)ij zi wj
ij

The scalar valued entries of the matrix of D2 f (x (y) , y) have the same differentiabil-
ity as the function y →D2 f (x (y) , y) . This is because the linear projection map, π ij
mapping L (Y, Z) to F given by π ij L ≡ Lij , 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 D1 f (x (y) , y) and the entries of its matrix, D1 f (x (y) , y)ij taken with
respect to suitable bases. From the formula for the inverse of a matrix, Theorem 3.5.14,
−1 −1
the ij th entries of the matrix of D1 f (x (y) , y) , D1 f (x (y) , y)ij also have the same
differentiability as y →D1 f (x (y) , y).
Now consider the formula for the derivative of the implicitly defined function in 6.31,
−1
Dx (y) = −D1 f (x (y) , y) D2 f (x (y) , y) . (6.36)

The above derivative is in L (Y, X) . Let {w1 , · · ·, wm } be a basis for Y and let {v1 , · · ·, vn }
be a basis for X. Letting xi be the ith 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,
Dwj1 wj2 ···wjr xi (y) exist and are continuous for r ≤ k and for any i. Consider what is
required for this to happen. By 6.36,
X³ −1
´
Dwj xi (y) = −D1 f (x (y) , y) (D2 f (x (y) , y))kj
ik
k
≡ G1 (x (y) , y) (6.37)
6.11. TAYLOR’S FORMULA 141

where (x, y) → G1 (x, y) is C k−1 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.

G1 (x (y + twk ) , y + twk ) − G1 (x (y) , y)

Dwj wk xi (y) ≡ lim
t→0 t
= D1 G1 (x (y) , y) Dx (y) wk + D2 G1 (x (y) , y)
≡ G2 (x (y) , y,Dx (y))

Since a similar result holds for all i and any choice of wj , wk , this shows x is at least
C 2 . If k ≥ 3, then another Gateaux derivative can be taken because then (x, y, z) →
G2 (x, y, z) is C 1 and it has been established Dx is C 1 . Continuing this way shows
Dwj1 wj2 ···wjr xi (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 Rn

In many applications of the implicit function theorem,

f : U ⊆ Rn × Rm → Rn

and f (x0 , y0 ) = 0 while f is C 1 . How can you recognize the condition of the implicit
−1
function theorem which says D1 f (x0 , y0 ) exists? This is really not hard. You recall
the matrix of the transformation D1 f (x0 , y0 ) with respect to the usual basis vectors is
 
f1,x1 (x0 , y0 ) · · · f1,xn (x0 , y0 )
 .. .. 
 . . 
fn,x1 (x0 , y0 ) · · · fn,xn (x0 , y0 )
−1
and so D1 f (x0 , y0 ) 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
D1 f (x0 , y0 ) 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 Taylor’s 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
m
X h(k) (0) h(m+1) (t)
h (1) = h (0) + + .
k! (m + 1)!
k=1

Proof: Let K be a number chosen such that

à m
!
X h(k) (0)
h (1) − h (0) + +K =0
k!
k=1
142 THE DERIVATIVE

Now the idea is to find K. To do this, let

à m
!
X h(k) (t) k m+1
F (t) = h (1) − h (t) + (1 − t) + K (1 − t)
k!
k=1

Then F (1) = F (0) = 0. Therefore, by Rolle’s theorem there exists t between 0 and 1
such that F 0 (t) = 0. Thus,
m
X h(k+1) (t) k
0 = −F 0 (t) = h0 (t) + (1 − t)
k!
k=1
m
X h(k) (t) k−1 m
− k (1 − t) − K (m + 1) (1 − t)
k!
k=1

And so
m
X m−1
X
h(k+1) (t) k h(k+1) (t) k
= h0 (t) + (1 − t) − (1 − t)
k! k!
k=1 k=0
m
−K (m + 1) (1 − t)

h(m+1) (t) m m
= h0 (t) + (1 − t) − h0 (t) − K (m + 1) (1 − t)
m!
and so
h(m+1) (t)
K= .
(m + 1)!
This proves the theorem. ¥
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,

h0 (t) = Df (x+tv) (v) , h00 (t) = D2 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) vk .

It follows from Taylor’s formula for a function of one variable given above that
m
X D(k) f (x) vk D(m+1) f (x+tv) vm+1
f (x + v) = f (x) + + . (6.38)
k! (m + 1)!
k=1

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. TAYLOR’S FORMULA 143

6.11.1 Second Derivative Test

Now consider the case where U ⊆ Rn and f : U → R is C 2 (U ). Then from Taylor’s
theorem, if v is small enough, there exists t ∈ (0, 1) such that

D2 f (x+tv) v2
f (x + v) = f (x) + Df (x) v+ . (6.39)
2
Consider

D2 f (x+tv) (ei ) (ej ) ≡ D (D (f (x+tv)) ei ) ej

µ ¶
∂f (x + tv)
= D ej
∂xi
∂ 2 f (x + tv)
=
∂xj ∂xi

where ei are the usual basis vectors. Letting

n
X
v= vi ei ,
i=1

the second derivative term in 6.39 reduces to

1X 2 1X
D f (x+tv) (ei ) (ej ) vi vj = Hij (x+tv) vi vj
2 i,j 2 i,j

where
∂ 2 f (x+tv)
Hij (x+tv) = D2 f (x+tv) (ei ) (ej ) = .
∂xj ∂xi

Definition 6.11.3 The matrix whose ij th entry is ∂ 2 f (x)

∂xj ∂xi 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,
1
f (x + v) = f (x) + Df (x) v+ vT H (x) v+
2
1¡ T ¢
v (H (x+tv) −H (x)) v . (6.40)
2
where the last two terms involve ordinary matrix multiplication and

vT = (v1 · · · vn )

for vi the components of v relative to the standard basis.

Definition 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 δ > 0 such that for all
y ∈ B (x, δ)
f (y) ≥ f (x) .
f has a local maximum at x ∈ D if there exists δ > 0 such that for all y ∈ B (x, δ)

f (y) ≤ f (x) .
144 THE DERIVATIVE

Theorem 6.11.5 If f : U → R where U is an open subset of Rn 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 first
that H (x) has all positive eigenvalues and that all are larger than δ 2 > 0. Then by
n
Theorem 3.8.23, H (x) has an orthonormal
Pn basis of eigenvectors, {vi }i=1 and if u is an
arbitrary vector, such that u = j=1 uj vj where uj = u · vj , then
n
X n
X
uT H (x) u = uj vjT H (x) uj vj
j=1 j=1

n
X n
X 2
= u2j λj ≥ δ 2 u2j = δ 2 |u| .
j=1 j=1

From 6.40 and the continuity of H, if v is small enough,

1 2 1 2 δ2 2
f (x + v) ≥ f (x) + δ 2 |v| − δ 2 |v| = f (x) + |v| .
2 4 4
This shows the first claim of the theorem. The second claim follows from similar rea-
soning. Suppose H (x) has a positive eigenvalue λ2 . Then let v be an eigenvector for
this eigenvalue. Then from 6.40,
1
f (x+tv) = f (x) + t2 vT H (x) v+
2
1 2¡ T ¢
t v (H (x+tv) −H (x)) v
2
which implies
1 2 1 ¡ ¢
f (x+tv) = f (x) + t2 λ2 |v| + t2 vT (H (x+tv) −H (x)) v
2 2
1 2 2 2
≥ f (x) + t λ |v|
4
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
f1 (x, y) = x4 + y 2 , f2 (x, y) = −x4 + y 2 .
Then Dfi (0, 0) = 0 and for both functions, the Hessian matrix evaluated at (0, 0) equals
µ ¶
0 0
0 2

but the behavior of the two functions is very different near the origin. The second has
a saddle point while the first has a minimum there.
6.12. THE METHOD OF LAGRANGE MULTIPLIERS 145

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 ⊆ Rn and let

gi (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
gi (x) = 0, i = 1, · · ·, m.

x0 is a local maximum if f (x0 ) ≥ f (x) for all x near x0 which also satisfies the
constraints 6.41. A local minimum is defined similarly. Let F : U × R → Rm+1 be
defined by  
f (x) − a
 g1 (x) 
 
F (x,a) ≡  .. . (6.42)
 . 
gm (x)
Now consider the m + 1 × n Jacobian matrix, the matrix of the linear transformation,
D1 F (x, a) with respect to the usual basis for Rn and Rm+1 .
 
fx1 (x0 ) · · · fxn (x0 )
 g1x1 (x0 ) · · · g1xn (x0 ) 
 
 .. .. .
 . . 
gmx1 (x0 ) ··· gmxn (x0 )

If this matrix has rank m+1 then some m+1×m+1 submatrix has nonzero determinant.
It follows from the implicit function theorem that there exist m + 1 variables, xi1 , · ·
·, xim+1 such that the system
F (x,a) = 0 (6.43)
specifies these m + 1 variables as a function of the remaining n − (m + 1) variables and
a in an open set of Rn−m . Thus there is a solution (x,a) to 6.43 for some x close to x0
whenever a is in some open interval. Therefore, x0 cannot be either a local minimum or
a local maximum. It follows that if x0 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,

λ1 , · · ·, λm ,

and a scalar µ, not all zero such that

     
fx1 (x0 ) g1x1 (x0 ) gmx1 (x0 )
 ..   ..   .. 
µ .  = λ1  .  + · · · + λm  . . (6.44)
fxn (x0 ) g1xn (x0 ) gmxn (x0 )

If the column vectors

   
g1x1 (x0 ) gmx1 (x0 )
 ..   .. 
 . ,· · · .  (6.45)
g1xn (x0 ) gmxn (x0 )
146 THE DERIVATIVE

are linearly independent, then, µ 6= 0 and dividing by µ yields an expression of the form
     
fx1 (x0 ) g1x1 (x0 ) gmx1 (x0 )
 ..   ..   .. 
 .  = λ1  .  + · · · + λm  .  (6.46)
fxn (x0 ) g1xn (x0 ) gmxn (x0 )

at every point x0 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 Rn and let f : U → R be a C 1

function. Then if x0 ∈ 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 µ, λ1 , · · ·, λm 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.
P
2. Show every polynomial, |α|≤k dα xα 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, Dv1 v2 f (x) and Dv2 v1 f (x) are equal.

4. Give an example of a function which is differentiable everywhere but at some

point it fails to have continuous partial derivatives. Thus this function will be an
example of a differentiable 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
( 2
(x2 −y4 )
f (x, y) = if (x, y) 6= (0, 0) ,
(x2 +y 4 )2
1 if (x, y) = (0, 0) .

Show each Gateaux derivative, Dv f (0) exists and equals 0 for every v. Also show
each Gateaux derivative exists at every other point in R2 . Now consider the curve
x2 = 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 differentiable function which is not
differentiable and not continuous.

6. Let f be a real valued function defined on R2 by

( 3 3
x −y
f (x, y) ≡ x2 +y 2 if (x, y) 6= (0, 0)
0 if (x, y) = (0, 0)

Determine whether f is continuous at (0, 0). Find fx (0, 0) and fy (0, 0) . Are the
partial derivatives of f continuous at (0, 0)? Find D(u,v) f ((0, 0)) , limt→0 f (t(u,v))
t .
Is the mapping (u, v) → D(u,v) f ((0, 0)) linear? Is f differentiable at (0, 0)?
6.13. EXERCISES 147

7. Let f : V → R where V is a finite 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 differentiable. 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 finite dimensional inner

product space with the inner product denoted by (·, ·) . Suppose f is differentiable.
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 finite dimensional normed

vector space. Suppose that for all v ∈ X, Dv f (x) exists. Show that whenever
a ∈ F Dav f (x) = aDv f (x). Explain why if x → Dv f (x) is continuous then v →
Dv f (x) is linear. Show that if f is differentiable at x, then Dv f (x) = Df (x) v.

10. Suppose B is an open ball in X and f : B → Y is differentiable. Suppose also

there exists L ∈ L (X, Y ) such that

||Df (x) − L|| < k

for all x ∈ B. Show that if x1 , x2 ∈ B,

|f (x1 ) − f (x2 ) − L (x1 − x2 )| ≤ k |x1 − x2 | .

Hint: Consider T x = 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 finite dimensional

normed vector spaces and suppose f ∈ C 1 (U ) and Df (x0 ) is one to one. Then
show f is one to one near x0 . Hint: Show using the assumption that f is C 1 that
there exists δ > 0 such that if

x1 , x2 ∈ B (x0 , δ) ,

then
r
|f (x1 ) − f (x2 ) − Df (x0 ) (x1 − x2 )| ≤ |x1 − x2 | (6.47)
2
then use Problem 1.

12. Suppose M ∈ L (X, Y ) and suppose M is onto. Show there exists L ∈ L (Y, X)
such that
LM x =P x
where P ∈ L (X, X), and P 2 = P . Also show L is one to one and onto. Hint:
Let {y1 , · · ·, ym } be a basis of Y and let M xi = yi . Then define
m
X m
X
Ly = αi xi where y = α i yi .
i=1 i=1
148 THE DERIVATIVE

Show {x1 , · · ·, xm } is a linearly independent set and show you can obtain {x1 , · ·
·, xm , · · ·, xn }, a basis for X in which M xj = 0 for j > m. Then let
m
X
Px ≡ α i xi
i=1

where
m
X
x= αi xi .
i=1

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 X1 ≡ P X and L−1 maps open
sets from X1 to open sets of Y. Then Lf (x + v) = Lf (x) + LDf (x) v + o (v) .
Now for z ∈ X1 , let h (z) = Lf (x + z) − Lf (x) . Then h is C 1 on some small
open subset of X1 containing 0 and Dh (0) = LDf (x) which is seen to be one
to one and onto and in L (X1 , X1 ) . Therefore, if r is small enough, h (B (0,r))
equals an open set in X1 , 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 ⊆ R2 is an open set and f : U → R3 is C 1 . Suppose Df (s0 , t0 ) has
rank two and  
x0
f (s0 , t0 ) =  y0  .
z0
Show that for (s, t) near (s0 , t0 ), the points f (s, t) may be realized in one of the
following forms.
{(x, y, φ (x, y)) : (x, y) near (x0 , y0 )},
{(φ (y, z) , y, z) : (y, z) near (y0 , z0 )},
or
{(x, φ (x, z) , z, ) : (x, z) near (x0 , z0 )}.
This shows that parametrically defined surfaces can be obtained locally in a par-
ticularly simple form.
15. Let f : U → Y , Df (x) exists for all x ∈ U , B (x0 , δ) ⊆ U , and there exists
L ∈ L (X, Y ), such that L−1 ∈ L (Y, X), and for all x ∈ B (x0 , δ)
r
||Df (x) − L|| < , r < 1.
||L−1 ||

Show that there exists ε > 0 and an open subset of B (x0 , δ) , V , such that
f : V →B (f (x0 ) , ε) is one to one and onto. Also Df −1 (y) exists for each y ∈
B (f (x0 ) , ε) and is given by the formula
£ ¡ ¢¤−1
Df −1 (y) = Df f −1 (y) .

Hint: Let
Ty (x) ≡ T (x, y) ≡ x−L−1 (f (x) − y)
(1−r)δ n
for |y − f (x0 )| < 2||L−1 || , consider {Ty (x0 )}. This is a version of the inverse
1
function theorem for f only differentiable, not C .
6.13. EXERCISES 149

16. Recall the nth derivative can be considered a multilinear function defined on X n
with values in some normed vector space. Now define a function denoted as
wi vj1 · · · vjn which maps X n → Y in the following way

wi vj1 · · · vjn (vk1 , · · ·, vkn ) ≡ wi δ j1 k1 δ j2 k2 · · · δ jn kn (6.48)

and wi vj1 · · · vjn is to be linear in each variable. Thus, for

à n n
!
X X
ak1 vk1 , · · ·, akn vkn ∈ X n ,
k1 =1 kn =1
à n n
!
X X
wi vj1 · · · vjn ak1 vk1 , · · ·, akn vkn
k1 =1 kn =1

X
≡ wi (ak1 ak2 · · · akn ) δ j1 k1 δ j2 k2 · · · δ jn kn
k1 k2 ···kn
= wi aj1 aj2 · · · ajn (6.49)

Show each wi vj1 · · · vjn 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, wi vj1 ···vjn
for all choices of i and the jk is a basis of this vector space. Find the dimension
of the vector space.
Pn Pn
17. Minimize j=1 xj subject to the constraint j=1 x2j = a2 . 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, 4x2 + 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 x2 + 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 x2 + 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 = (Aij ) be an n × n matrix which is symmetric. Thus Aij = Aji and recall
∂
(Ax)i = Aij xj where as usual sum over the repeated index. Show ∂x i
(Aij xj xi ) =
2Aij xj . Show that when you use the method of P Lagrange multipliers to maximize
n
the function, Aij xj xi subject to the constraint, j=1 x2j = 1, the value of λ which
corresponds to the maximum value of this functions is such that Aij xj = λxi .
Thus Ax = λx. Thus λ is an eigenvalue of the matrix, A.
23. Let x1 , · · ·, x5 be 5 positive numbers. Maximize their product subject to the
constraint that
x1 + 2x2 + 3x3 + 4x4 + 5x5 = 300.

24. Let f (x1 , · · ·, xn ) = xn1 x2n−1 · · · x1n . Then f achieves a maximum on the set,
( n
)
X
n
S≡ x∈R : ixi = 1 and each xi ≥ 0 .
i=1

If x ∈ S is the point where this maximum is achieved, find x1 /xn .

150 THE DERIVATIVE

25. Let (x, y) be a point on the ellipse, x2 /a2 +y 2 /b2 = 1 which is in the first 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.
Qn Pn
26. Maximize i=1 x2i (≡ x¡21 ×x22 ×x
¢
2 2
3 ×···×xn ) subject to the constraint,
2 2
i=1 xi = r .
2 n
Show the maximum is r /n . Now show from this that
à n
!1/n n
Y 1X 2
x2i ≤ x
i=1
n i=1 i

and finally, conclude that if each number xi ≥ 0, then

à n
!1/n n
Y 1X
xi ≤ xi
i=1
n i=1

and there exist values of the xi for which equality holds. This says the “geometric
mean” is always smaller than the arithmetic mean.
27. Maximize x2 y 2 subject to the constraint

x2p y 2q
+ = r2
p q
where p, q are real numbers larger than 1 which have the property that
1 1
+ = 1.
p q

show the maximum is achieved when x2p = y 2q and equals r2 . Now conclude that
if x, y > 0, then
xp yq
xy ≤ +
p 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 define 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 definition
of what is meant by a compact set follows.

Definition 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 finitely many sets of U, {U1 , · · ·, Um } such that

K ⊆ ∪m
k=1 Uk .

In words, every open cover admits a finite subcover.

It was shown earlier that in any finite 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 defined.

151
152 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 xn ∈ K such that B (xn , r) is not a subset of any of the
sets of U. Since K is sequentially compact, there exists a subsequence, {xnk } 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 xnk ∈ B (x,r/2) for all k large enough and also for all k large enough,
1/nk < r/2. Therefore, there exists xnk ∈ B (x,r/2) and 1/nk < r/2. But this is a
contradiction because
B (xnk , 1/nk ) ⊆ B (x, r) ⊆ U
contrary to the choice of xnk which required B (xnk , 1/nk ) 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 finite dimensional normed vector space, then K is compact.

Proof: Suppose first 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 x1 ∈ K. Then B (x1 , r) ⊆ U1
m
for some U1 ∈ U. Suppose {B (xi , r)}i=1 have been chosen such that

B (xi , r) ⊆ Ui ∈ U .
m m
If their union contains K then {Ui }i=1 is a finite subcover of U. If {B (xi , r)}i=1 does
not cover K, then there exists xm+1 ∈ / ∪mi=1 B (xi , r) and so B (xm+1 , r) ⊆ Um+1 ∈ U.
∞
This process must stop after finitely many choices of B (xi , r) because if not, {xk }k=1
would have a subsequence which converges to a point of K which cannot occur because
whenever i 6= j,
||xi − xj || > r
Therefore, eventually
K ⊆ ∪m m
k=1 B (xk , r) ⊆ ∪k=1 Uk .

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, {xk } which has no convergent subsequence. This
requires that {xk } have no limit point for if it did have a limit point, x, then B (x, 1/n)
would contain infinitely many distinct points of {xk } and so a subsequence of {xk }
converging to x could be obtained. Also no xk is repeated infinitely often because if
there were such, a convergent subsequence could be obtained. Hence ∪∞ k=m {xk } ≡ Cm
is a closed set, closed because it contains all its limit points. (It has no limit points so
C
it contains them all.) Then letting Um = Cm , it follows {Um } is an open cover of K
which has no finite subcover. Thus K must be sequentially compact after all.
If K is a closed and bounded set in a finite dimensional normed vector space, then K
is sequentially compact by Theorem 5.8.4. Therefore, by the first 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 finite 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) 153

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 defined on R, an integrator
function. There exists a function µ : P (R) → [0, ∞] which satisfies the following
properties.
1. If A ⊆ B, then 0 ≤ µ (A) ≤ µ (B) , µ (∅) = 0.
P∞
2. µ (∪∞
k=1 Ai ) ≤ i=1 µ (Ai )

3. µ ([a, b]) = F (b+) − F (a−) ,

4. µ ((a, b)) = F (b−) − F (a+)
5. µ ((a, b]) = F (b+) − F (a+)
6. µ ([a, b)) = F (b−) − F (a−) where
F (b+) ≡ lim F (t) , F (b−) ≡ lim F (t) .
t→b+ t→a−

Proof: First it is necessary to define the function, µ. This is contained in the

following definition.
Definition 7.2.2 For A ⊆ R,
 
X ∞ 
µ (A) = inf (F (bi −) − F (ai +)) : A ⊆ ∪∞
i=1 (a ,
i ib )
 
j=1

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
infimum 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 µ (∅) = 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 − δ, a + δ) and so µ (∅) ≤ 2δ for every δ > 0.
Consider 2.). If any µ (Ai ) = ∞, there is nothing to prove. The assertion simply is
∞ ≤ ∞. Assume then that µ (Ai ) < ∞ for all i. Then for each m ∈ N there exists a
m ∞
countable set of open intervals, {(am
i , bi )}i=1 such that
X∞
ε
µ (Am ) + > (F (bm m
i −) − F (ai +)) .
2m i=1

Then using Theorem 2.3.4 on Page 22,

X
µ (∪∞
m=1 Am ) ≤ (F (bm m
i −) − F (ai +))
im
∞ X
X ∞
= (F (bm m
i −) − F (ai +))
m=1 i=1
X∞
ε
≤ µ (Am ) +
m=1
2m
X∞
= µ (Am ) + ε
m=1
154 MEASURES AND MEASURABLE FUNCTIONS

and since ε is arbitrary, this establishes 2.).

∞
Next consider 3.). By definition, there exists a sequence of open intervals, {(ai , bi )}i=1
whose union contains [a, b] such that
∞
X
µ ([a, b]) + ε ≥ (F (bi −) − F (ai +))
i=1

By Theorem 7.1.3, finitely many of these intervals also cover [a, b]. It follows there exists
n
finitely many of these intervals, {(ai , bi )}i=1 which overlap such that a ∈ (a1 , b1 ) , b1 ∈
(a2 , b2 ) , · · ·, b ∈ (an , bn ) . Therefore,
n
X
µ ([a, b]) ≤ (F (bi −) − F (ai +))
i=1

It follows
n
X
(F (bi −) − F (ai +)) ≥ µ ([a, b])
i=1
n
X
≥ (F (bi −) − F (ai +)) − ε
i=1
≥ F (b+) − F (a−) − ε
Since ε is arbitrary, this shows
µ ([a, b]) ≥ F (b+) − F (a−)
but also, from the definition, the following inequality holds for all δ > 0.
µ ([a, b]) ≤ F ((b + δ) −) − F ((a − δ) −) ≤ F (b + δ) − F (a − δ)
Therefore, letting δ → 0 yields
µ ([a, b]) ≤ F (b+) − F (a−)
This establishes 3.).
Consider 4.). For small δ > 0,
µ ([a + δ, b − δ]) ≤ µ ((a, b)) ≤ µ ([a, b]) .
Therefore, from 3.) and the definition of µ,
F ((b − δ)) − F ((a + δ)) ≤ F ((b − δ) +) − F ((a + δ) −)
= µ ([a + δ, b − δ]) ≤ µ ((a, b)) ≤ F (b−) − F (a+)
Now letting δ decrease to 0 it follows
F (b−) − F (a+) ≤ µ ((a, b)) ≤ F (b−) − F (a+)
This shows 4.)
Consider 5.). From 3.) and 4.)
F (b+) − F ((a + δ))
≤ F (b+) − F ((a + δ) −)
= µ ([a + δ, b]) ≤ µ ((a, b])
≤ µ ((a, b + δ)) = F ((b + δ) −) − F (a+)
≤ F (b + δ) − F (a+) .
7.3. GENERAL OUTER MEASURES AND MEASURES 155

Now let δ converge to 0 from above to obtain

F (b+) − F (a+) = µ ((a, b]) = F (b+) − F (a+) .

This establishes 5.) and 6.) is entirely similar to 5.). This proves the theorem. ¥

Definition 7.2.3 Let Ω be a nonempty set. A function mapping P (Ω) → [0, ∞]

is called an outer measure if it satisfies 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 Rn turns out to be very easy.

7.3.1 Measures And Measure Spaces

First here is a definition of a measure.

Definition 7.3.1 S ⊆ P (Ω) is called a σ algebra , pronounced “sigma algebra”,

if
∅, Ω ∈ S,
If E ∈ S then E C ∈ S
and
If Ei ∈ S, for i = 1, 2, · · ·, then ∪∞
i=1 Ei ∈ S.

A function µ : S → [0, ∞] where S is a σ algebra is called a measure if whenever

∞
{Ei }i=1 ⊆ S and the Ei are disjoint, then it follows
∞
¡ ¢ X
µ ∪∞ E
j=1 j = µ (Ej ) .
j=1

The triple (Ω, S, µ) is often called a measure space. Sometimes people refer to (Ω, S) as
a measurable space, making no reference to the measure. Sometimes (Ω, S) may also be
called a measure space.

Theorem 7.3.2 Let {Em }∞m=1 be a sequence of measurable sets in a measure

space (Ω, F, µ). Then if · · ·En ⊆ En+1 ⊆ En+2 ⊆ · · ·,

µ(∪∞
i=1 Ei ) = lim µ(En ) (7.1)
n→∞

and if · · ·En ⊇ En+1 ⊇ En+2 ⊇ · · · and µ(E1 ) < ∞, then

µ(∩∞
i=1 Ei ) = lim µ(En ). (7.2)
n→∞

Stated more succinctly, Ek ↑ E implies µ (Ek ) ↑ µ (E) and Ek ↓ E with µ (E1 ) < ∞
implies µ (Ek ) ↓ µ (E).
156 MEASURES AND MEASURABLE FUNCTIONS

Proof: First note that ∩∞ ∞ C C ∞

i=1 Ei = (∪i=1 Ei ) ∈ F so ∩i=1 Ei is measurable. Also
¡ C ¢C
note that for A and B sets of F, A \ B ≡ A ∪ B ∈ F. To show 7.1, note that 7.1
is obviously true if µ(Ek ) = ∞ for any k. Therefore, assume µ(Ek ) < ∞ for all k. Thus
µ(Ek+1 \ Ek ) + µ(Ek ) = µ(Ek+1 )
and so
µ(Ek+1 \ Ek ) = µ(Ek+1 ) − µ(Ek ).
Also,
∞
[ ∞
[
Ek = E1 ∪ (Ek+1 \ Ek )
k=1 k=1
and the sets in the above union are disjoint. Hence
∞
X
µ(∪∞
i=1 Ei ) = µ(E1 ) + µ(Ek+1 \ Ek ) = µ(E1 )
k=1
∞
X
+ µ(Ek+1 ) − µ(Ek )
k=1
n
X
= µ(E1 ) + lim µ(Ek+1 ) − µ(Ek ) = lim µ(En+1 ).
n→∞ n→∞
k=1
This shows part 7.1.
To verify 7.2,
µ(E1 ) = µ(∩∞ ∞
i=1 Ei ) + µ(E1 \ ∩i=1 Ei )

since µ(E1 ) < ∞, it follows µ(∩∞ n ∞

i=1 Ei ) < ∞. Also, E1 \ ∩i=1 Ei ↑ E1 \ ∩i=1 Ei and so by
7.1,
µ(E1 ) − µ(∩∞ ∞ n
i=1 Ei ) = µ(E1 \ ∩i=1 Ei ) = lim µ(E1 \ ∩i=1 Ei )
n→∞

= µ(E1 ) − lim µ(∩ni=1 Ei ) = µ(E1 ) − lim µ(En ),

n→∞ n→∞

Hence, subtracting µ (E1 ) from both sides,

lim µ(En ) = µ(∩∞
i=1 Ei ).
n→∞

This proves the theorem. ¥

The following definition is important.
Definition 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, {fk (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 fk (x) → f (x).

7.4 The Borel Sets, Regular Measures

7.4.1 Definition 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.
Definition 7.4.1 Let Y be a closed subset of X a finite 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 σ algebra of
sets of Y and let µ be a measure defined on F. Then µ is said to be a regular measure
if the following two conditions hold.
7.4. THE BOREL SETS, REGULAR MEASURES 157

1. For every F ∈ F

µ (F ) = sup {µ (K) : K ⊆ F and K is compact} (7.3)

2. For every F ∈ F

µ (F ) = inf {µ (V ) : V ⊇ F and V is open in Y } (7.4)

The first 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 finite subcover, {V1 , · · ·, Vm } and consequently,
{Y ∩ V1 , · · ·, Y ∩ Vm } is a finite 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 finite 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 definition, this admitts a finite subcover, {Y ∩ V1 , · · ·, Y ∩ Vm } but this implies
{V1 , · · ·, Vm } is also a finite 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 σ
algebra of subsets of Y which contains all the open sets of Y . To see such a smallest σ
algebra exists, let H denote the set of all σ algebras which contain the open sets P (Y ),
the set of all subsets of Y is one such σ algebra. Define B (Y ) ≡ ∩H. Then B (Y ) is a
σ algebra because ∅, Y are both open sets in Y and so they are in each σ algebra of H.
If F ∈ B (Y ), then F is a set of every σ algebra of H and so F C is also a set of every
σ algebra of H. Thus F C ∈ B (Y ). If {Fi } is a sequence of sets of B (Y ), then {Fi } is
a sequence of sets of every σ algebra of H and so ∪i Fi is a set in every σ algebra of H
which implies ∪i Fi ∈ B (Y ) so B (Y ) is a σ algebra as claimed. From its definition, it is
the smallest σ 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 first Lemma holds for any σ 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.
158 MEASURES AND MEASURABLE FUNCTIONS

Lemma 7.4.3 Let Y be a closed subset of X a finite dimensional normed vector

space and let S be a σ algebra of sets of Y containing the open sets of Y . Suppose µ is
a measure defined on S and suppose also µ (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.

V F F V

D(0, n) \ F

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
³ ´C
C
D (0, n) \ F = D (0, n) ∩ F C = D (0, n) ∪ F ∈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

µ (D (0, n) \ F ) + ε > µ (V ) . (7.5)

Since µ is a measure,

µ (V \ (D (0, n) \ F )) + µ (D (0, n) \ F ) = µ (V )

and so from 7.5

µ (V \ (D (0, n) \ F )) < ε (7.6)
Note
¡ ¢C ³ C
´
V \ (D (0, n) \ F ) = V ∩ D (0, n) ∩ F C = V ∩ D (0, n) ∪ (V ∩ F )

and by 7.6,
µ (V \ (D (0, n) \ F )) < ε
7.4. THE BOREL SETS, REGULAR MEASURES 159

so in particular,
µ (V ∩ F ) < ε.
Now
V ⊇ D (0, n) ∩ F C
and so
C
V C ⊆ D (0, n) ∪ F
which implies
V C ∩ D (0, n) ⊆ F ∩ D (0, n) = F
Since F ⊆ D (0, n) ,
¡ ¡ ¢¢ ³ ¡ ¢C ´
µ F \ V C ∩ D (0, n) = µ F ∩ V C ∩ D (0, n)
³ ³ ´´
C
= µ (F ∩ V ) ∪ F ∩ D (0, n)
= µ (F ∩ V ) < ε

showing the compact set, V C ∩ D (0, n) is contained in F and

¡ ¢
µ V C ∩ D (0, n) + ε > µ (F ) .

In the general case where F is only given to be in S, let Fn = B (0, n) ∩ F. Then by

7.1, if l < µ (F ) is given, then for all ε sufficiently small,

l + ε < µ (Fn )

provided n is large enough. Now it was just shown there exists K a compact subset of
Fn such that µ (Fn ) < µ (K) + ε. Then K ⊆ F and

l + ε < µ (Fn ) < µ (K) + ε

and so whenever l < µ (F ) , it follows there exists K a compact subset of F such that

l < µ (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. Define
dist (x, S) ≡ inf {||x − y|| : y ∈ S}
Then
|dist (x1 , S) − dist (x2 , S)| ≤ ||x1 −x2 || .

Proof: Suppose dist (x1 , S) ≥ dist (x2 , S) . Then let y ∈ S such that

dist (x2 , S) + ε > ||x2 − y||

Then

|dist (x1 , S) − dist (x2 , S)| = dist (x1 , S) − dist (x2 , S)

≤ dist (x1 , S) − (||x2 − y|| − ε)
≤ ||x1 − y|| − ||x2 − y|| + ε
≤ |||x1 − y|| − ||x2 − y||| + ε
≤ ||x1 − x2 || + ε.
160 MEASURES AND MEASURABLE FUNCTIONS

Since ε is arbitrary, this proves the lemma in case dist (x1 , S) ≥ dist (x2 , S) . The case
where dist (x2 , S) ≥ dist (x1 , S) is entirely similar. This proves the lemma. ¥
The next lemma says that regularity comes free for finite measures defined 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 µ be a finite measure defined on B (Y ) where Y is a closed subset

of X, a finite dimensional normed vector space. Then for every F ∈ B (Y ) ,

µ (F ) = sup {µ (K) : K ⊆ F, K is closed }

µ (F ) = inf {µ (V ) : V ⊇ F, V is open}

Proof: For convenience, I will call a measure which satisfies 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
© ¡ ¢ ª
Kk ≡ x ∈ V : dist x, V C ≥ 1/k .

Then clearly the union of the Kk equals V and each is closed because x→ dist (x, S) is
always a continuous function whenever S is any nonempty set. Next, for K closed let

Vk ≡ {x ∈ Y : dist (x, K) < 1/k} .

Clearly the intersection of the Vk 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, Vk
excludes x. Thus the only points in the intersection of the Vk 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

µ (V ) = sup {µ (K) : K ⊆ V and K is closed}

µ (K) = inf {µ (V ) : V ⊇ K and V is open} .

Also since V is open and K is closed,

µ (V ) = inf {µ (U ) : U ⊇ V and V is open}

µ (K) = sup {µ (L) : L ⊆ K and L is closed}

In words, µ is almost regular on open and closed sets. Let

F ≡ {F ∈ B (Y ) such that µ is almost regular on F } .

Then F contains the open sets. I want to show F is a σ 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
µ is finite and F is inner regular, there exists K ⊆ F such that

µ (F \ K) = µ (F ) − µ (K) < ε.

But K C \ F C = F \ K and so
¡ ¢ ¡ ¢ ¡ ¢
µ KC \ F C = µ KC − µ F C < ε

showing that µ 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
7.4. THE BOREL SETS, REGULAR MEASURES 161

C C
is inner regular. You start
¡ C with ¢V ⊇ F such that µ (V \ F ) <Cε, note F \ V = V \ F,
C
and then conclude µ F \ V < ε, thus approximating F with the closed subset,
V C.
Next I will show F is closed with respect to taking countable unions. Let {Fk } be
a sequence of sets in F. Then since Fk ∈ F, there exist {Kk } such that Kk ⊆ Fk and
µ (Fk \ Kk ) < ε/2k+1 . First choose m large enough that
ε
µ ((∪∞ m
k=1 Fk ) \ (∪k=1 Fk )) < .
2
Then

µ ((∪m m
k=1 Fk ) \ (∪k=1 Kk )) ≤ µ (∪m
k=1 (Fk \ Kk ))
Xm
ε ε
≤ <
2k+1 2
k=1

and so

µ ((∪∞ m
k=1 Fk ) \ (∪k=1 Kk )) ≤ µ ((∪∞ m
k=1 Fk ) \ (∪k=1 Fk ))
+µ ((∪m m
k=1 Fk ) \ (∪k=1 Kk ))
ε ε
< + =ε
2 2
Since µ is outer regular on Fk , there exists Vk such that µ (Vk \ Fk ) < ε/2k . Then

µ ((∪∞ ∞
k=1 Vk ) \ (∪k=1 Fk )) ≤ µ (∪∞
k=1 (Vk \ Fk ))
X∞
≤ µ (Vk \ Fk )
k=1
X∞
ε
< =ε
2k
k=1

and this completes the demonstration that F is a σ algebra. This proves the lemma. ¥
The next theorem is the main result. It shows regularity is automatic if µ (K) < ∞
for all compact K.

Theorem 7.4.6 Let µ be a finite measure defined on B (Y ) where Y is a closed

subset of X, a finite dimensional normed vector space. Then µ is regular. If µ is not
necessarily finite but is finite on compact sets, then µ is regular.

Proof: From Lemma 7.4.5 µ is outer regular. Now let F ∈ B (Y ). Then since µ is
finite, it follows from Lemma 7.4.5 there exists H ⊆ F such that H is closed, H ⊆ F,
and
µ (F ) < µ (H) + ε.
Then let Kk ≡ H ∩ B (0, k). Thus Kk is a closed and bounded, hence compact set and
∪∞
k=1 Kk = H. Therefore by Theorem 7.3.2, for all k large enough,

µ (F )
< µ (Kk ) + ε
< sup {µ (K) : K ⊆ F and K compact} + ε
≤ µ (F ) + ε

Since ε was arbitrary, it follows

sup {µ (K) : K ⊆ F and K compact} = µ (F ) .

162 MEASURES AND MEASURABLE FUNCTIONS

This establishes µ is regular if µ is finite.

Now suppose it is only known that µ is finite on compact sets. Consider outer
regularity. There are at most finitely many r ∈ [0, R] such that µ (S (0,r)) > δ >
0. If this were not so, then µ (D (0,R)) = ∞ contrary to the assumption that µ is
finite on compact sets. Therefore, there are at most countably many r ∈ [0, R] such
that µ (S (0,r)) > 0. Here is why. Let Sk denote those values of r ∈ [0, R] such that
µ (S (0,r)) > 1/k. Then the values of r such that µ (S (0,r)) > 0 equals ∪∞ m=1 Sm , a
countable union of finite sets which is at most countable.
It follows there are at most countably many r ∈ (0, ∞) such that µ (S (0,r)) > 0.
Therefore, there exists an increasing sequence {rk } such that limk→∞ rk = ∞ and
µ (S (0,rk )) = 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 µ (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 µ (B (0, r)) <
∞ by assumption, it follows from Theorem 7.3.2 that for each rk there exists an open
set, Uk ⊇ S (0,rk ) such that
µ (Uk ) < ε/2k+1 .
Let µ (F ) < ∞. There is nothing to show if µ (F ) = ∞. Define finite measures, µk
as follows.

µ1 (A) ≡ µ (B (0, 1) ∩ A) ,
µ2 (A) ≡ µ ((B (0, 2) \ D (0, 1)) ∩ A) ,
µ3 (A) ≡ µ ((B (0, 3) \ D (0, 2)) ∩ A)

etc. Thus
∞
X
µ (A) = µk (A)
k=1
and each µk is a finite measure. By the first part there exists an open set Vk such that

Vk ⊇ F ∩ (B (0, k) \ D (0, k − 1))

and
µk (Vk ) < µk (F ) + ε/2k+1
Without loss of generality Vk ⊆ (B (0, k) \ D (0, k − 1)) since you can take the intersec-
tion of Vk with this open set. Thus

µk (Vk ) = µ ((B (0, k) \ D (0, k − 1)) ∩ Vk ) = µ (Vk )

and the Vk are disjoint. Then let V = ∪∞ ∞

k=1 Vk and U = ∪k=1 Uk . It follows V ∪ U is an
open set containing F and
∞
X ∞
X ∞
X
ε ε
µ (F ) = µk (F ) > µk (Vk ) − µ (Vk ) −
=
2k+1 2
k=1 k=1 k=1
ε ε ε
= µ (V ) − ≥ µ (V ) + µ (U ) − − ≥ µ (V ∪ U ) − ε
2 2 2
which shows µ 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 µ (F ) < ∞. By the first part, there exists a compact set,

Kk ⊆ F ∩ (B (0, k) \ D (0, k − 1))

7.5. MEASURES AND OUTER MEASURES 163

such that

µk (Kk ) + ε/2k+1 > µk (F ∩ (B (0, k) \ D (0, k − 1)))

= µk (F ) = µ (F ∩ (B (0, k) \ D (0, k − 1))) .

Since Kk is a subset of F ∩ (B (0, k) \ D (0, k − 1)) it follows µk (Kk ) = µ (Kk ). There-

fore,
∞
X ∞
X
µ (F ) = µk (F ) < µk (Kk ) + ε/2k
k=1 k=1
à ∞
! N
X X
< µk (Kk ) + ε/2 < µ (Kk ) + ε
k=1 k=1

provided N is large enough. The Kk are disjoint and so letting K = ∪N

k=1 Kk , this says
K ⊆ F and
µ (F ) < µ (K) + ε.
Now consider the case where µ (F ) = ∞. If l < ∞, it follows from Theorem 7.3.2

µ (F ∩ B (0, m)) > l

whenever m is large enough. Therefore, letting µm (A) ≡ µ (A ∩ B (0, m)) , there exists
a compact set, K ⊆ F ∩ B (0, m) such that

µ (K) = µm (K) > µm (F ∩ B (0, m)) = µ (F ∩ B (0, m)) > l

This proves the theorem. ¥

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 defined on R. However, the procedure for doing so is a special case of a general
approach due to Caratheodory in about 1918.

Definition 7.5.1 Let Ω be a nonempty set and let µ : P(Ω) → [0, ∞] be an

outer measure. For E ⊆ Ω, E is µ measurable if for all S ⊆ Ω,

µ(S) = µ(S \ E) + µ(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 stuff
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 σ algebra on which the outer measure, µ is a measure.
First here is a definition and a lemma.
2 1 Kings 17, 2 Kings 4, Mathew 14, and Mathew 15 all contain such descriptions. The stuff involved

was either oil, bread, flour or fish. 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 R3 into five disjoint pieces and assemble the pieces
to form two disjoint balls of the same size as the first. 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.
164 MEASURES AND MEASURABLE FUNCTIONS

Definition 7.5.2 (µbS)(A) ≡ µ(S ∩ A) for all A ⊆ Ω. Thus µbS is the name
of a new outer measure, called µ 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 µ measurable, then A is µbS measurable.

Proof: Suppose A is µ measurable. It is desired to to show that for all T ⊆ Ω,

(µbS)(T ) = (µbS)(T ∩ A) + (µbS)(T \ A).

Thus it is desired to show

µ(S ∩ T ) = µ(T ∩ A ∩ S) + µ(T ∩ S ∩ AC ). (7.8)

But 7.8 holds because A is µ measurable. Apply Definition 7.5.1 to S ∩ T instead of S.

¥
If A is µbS measurable, it does not follow that A is µ measurable. Indeed, if you
believe in the existence of non measurable sets, you could let A = S for such a µ non
measurable set and verify that S is µbS 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 Ω be a set and let µ be an outer measure on P (Ω). The
collection of µ measurable sets, S, forms a σ algebra and

∞
X
If Fi ∈ S, Fi ∩ Fj = ∅, then µ(∪∞
i=1 Fi ) = µ(Fi ). (7.9)
i=1

If · · ·Fn ⊆ Fn+1 ⊆ · · ·, then if F = ∪∞

n=1 Fn and Fn ∈ S, it follows that

µ(F ) = lim µ(Fn ). (7.10)

n→∞

If · · ·Fn ⊇ Fn+1 ⊇ · · ·, and if F = ∩∞

n=1 Fn for Fn ∈ S then if µ(F1 ) < ∞,

µ(F ) = lim µ(Fn ). (7.11)

n→∞

This measure space is also complete which means that if µ (F ) = 0 for some F ∈ S then
if G ⊆ F, it follows G ∈ S also.

Proof: First note that ∅ and Ω 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.
7.5. MEASURES AND OUTER MEASURES 165

S
T CT C
S A B

T T
S AC B
T T
S A BC
T T
S A B
B

Since µ is subadditive,
¡ ¢ ¡ ¢ ¡ ¢
µ (S) ≤ µ S ∩ A ∩ B C + µ (A ∩ B ∩ S) + µ S ∩ B ∩ AC + µ S ∩ AC ∩ B C .

Now using A, B ∈ S,
¡ ¢ ¡ ¢ ¡ ¢
µ (S) ≤ µ S ∩ A ∩ B C + µ (S ∩ A ∩ B) + µ S ∩ B ∩ AC + µ S ∩ AC ∩ B C
¡ ¢
= µ (S ∩ A) + µ S ∩ AC = µ (S)

It follows equality holds in the above. Now observe, using the picture if you like, that
¡ ¢ ¡ ¢
(A ∩ B ∩ S) ∪ S ∩ B ∩ AC ∪ S ∩ AC ∩ B C = S \ (A \ B)

and therefore,
¡ ¢ ¡ ¢ ¡ ¢
µ (S) = µ S ∩ A ∩ B C + µ (A ∩ B ∩ S) + µ S ∩ B ∩ AC + µ S ∩ AC ∩ B C
≥ µ (S ∩ (A \ B)) + µ (S \ (A \ B)) .

Therefore, since S is arbitrary, this shows A \ B ∈ S.

Since Ω ∈ S, this shows that A ∈ S if and only if AC ∈ S. Now if A, B ∈ S,
A ∪ B = (AC ∩ B C )C = (AC \ B)C ∈ S. By induction, if A1 , · · ·, An ∈ S, then so is
∪ni=1 Ai . If A, B ∈ S, with A ∩ B = ∅,

µ(A ∪ B) = µ((A ∪ B) ∩ A) + µ((A ∪ B) \ A) = µ(A) + µ(B).

By induction, if Ai ∩ Aj = ∅ and Ai ∈ S,
n
X
µ(∪ni=1 Ai ) = µ(Ai ). (7.12)
i=1

Now let A = ∪∞
i=1 Ai where Ai ∩ Aj = ∅ for i 6= j.
∞
X n
X
µ(Ai ) ≥ µ(A) ≥ µ(∪ni=1 Ai ) = µ(Ai ).
i=1 i=1
166 MEASURES AND MEASURABLE FUNCTIONS

Since this holds for all n, you can take the limit as n → ∞ and conclude,
∞
X
µ(Ai ) = µ(A)
i=1

which establishes 7.9.

Consider part 7.10. Without loss of generality µ (Fk ) < ∞ for all k since otherwise
there is nothing to show. Suppose {Fk } is an increasing sequence of sets of S. Then
∞
letting F0 ≡ ∅, {Fk+1 \ Fk }k=0 is a sequence of disjoint sets of S since it was shown
above that the difference of two sets of S is in S. Also note that from 7.12

µ (Fk+1 \ Fk ) + µ (Fk ) = µ (Fk+1 )

and so if µ (Fk ) < ∞, then

µ (Fk+1 \ Fk ) = µ (Fk+1 ) − µ (Fk ) .

Therefore, letting
F ≡ ∪∞
k=1 Fk

which also equals

∪∞
k=1 (Fk+1 \ Fk ) ,

it follows from part 7.9 just shown that

∞
X n
X
µ (F ) = µ (Fk+1 \ Fk ) = lim µ (Fk+1 \ Fk )
n→∞
k=1 k=1
n
X
= lim µ (Fk+1 ) − µ (Fk ) = lim µ (Fn+1 ) .
n→∞ n→∞
k=1

In order to establish 7.11, let the Fn be as given there. Then, since (F1 \ Fn )
increases to (F1 \ F ), 7.10 implies

lim (µ (F1 ) − µ (Fn )) = µ (F1 \ F ) .

n→∞

Now µ (F1 \ F ) + µ (F ) ≥ µ (F1 ) and so µ (F1 \ F ) ≥ µ (F1 ) − µ (F ). Hence

lim (µ (F1 ) − µ (Fn )) = µ (F1 \ F ) ≥ µ (F1 ) − µ (F )

n→∞

which implies
lim µ (Fn ) ≤ µ (F ) .
n→∞

But since F ⊆ Fn ,
µ (F ) ≤ lim µ (Fn )
n→∞

and this establishes 7.11. Note that it was assumed µ (F1 ) < ∞ because µ (F1 ) was
subtracted from both sides.
It remains to show S is closed under countable unions. Recall that if A ∈ S, then
AC ∈ S and S is closed under finite unions. Let Ai ∈ S, A = ∪∞ n
i=1 Ai , Bn = ∪i=1 Ai .
Then

µ(S) = µ(S ∩ Bn ) + µ(S \ Bn ) (7.13)

= (µbS)(Bn ) + (µbS)(BnC ).
7.5. MEASURES AND OUTER MEASURES 167

By Lemma 7.5.3 Bn is (µbS) measurable and so is BnC . I want to show µ(S) ≥ µ(S \
A) + µ(S ∩ A). If µ(S) = ∞, there is nothing to prove. Assume µ(S) < ∞. Then apply
Parts 7.11 and 7.10 to the outer measure, µbS in 7.13 and let n → ∞. Thus

Bn ↑ A, BnC ↓ AC

and this yields µ(S) = (µbS)(A) + (µbS)(AC ) = µ(S ∩ A) + µ(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 ⊆ Ω. I need to verify

µ (S) ≥ µ (S ∩ G) + µ (S \ G)

However,

µ (S ∩ G) + µ (S \ G) ≤ µ (S ∩ F ) + µ (S \ F ) + µ (F \ G)
= µ (S ∩ F ) + µ (S \ F ) = µ (S)

because by assumption, µ (F \ G) ≤ µ (F ) = 0.This proves the theorem. ¥

7.5.2 Completion Of Measure Spaces

Suppose (Ω, F, µ) is a measure space. Then it is always possible to
¡ enlarge
¢ the σ algebra
and define a new measure µ on this larger σ algebra such that Ω, F, µ is a complete
measure space. Recall this means that if N ⊆ N 0 ∈ F and µ (N 0 ) = 0, then N ∈ F. The
following theorem is the main result. The new measure space is called the completion
of the measure space.

Definition 7.5.5 A measure space, (Ω, F, µ) is called σ finite if there exists a

sequence {Ωn } ⊆ F such that ∪n Ωn = Ω and µ (Ωn ) < ∞.

For example, if X is a finite dimensional normed vector space and µ is a measure

defined on B (X) which is finite on compact sets, then you could take Ωn = B (0, n) .

Theorem 7.5.6¡ Let (Ω,¢ F, µ) be a σ finite measure space. Then there exists a
unique measure space, Ω, F, µ satisfying
¡ ¢
1. Ω, F, µ is a complete measure space.

2. µ = µ on F

3. F ⊇ F

4. For every E ∈ F there exists G ∈ F such that G ⊇ E and µ (G) = µ (E) .

In addition to this,

5. For every E ∈ F there exists F ∈ F such that F ⊆ E and µ (F ) = µ (E) .

Also for every E ∈ F there exist sets G, F ∈ F such that G ⊇ E ⊇ F and

µ (G \ F ) = µ (G \ F ) = 0 (7.14)
168 MEASURES AND MEASURABLE FUNCTIONS

Proof: First consider the claim about uniqueness. Suppose (Ω, F1 , ν 1 ) and (Ω, F2 , ν 2 )
both satisfy 1.) - 4.) and let E ∈ F1 . Also let µ (Ωn ) < ∞, · · ·Ωn ⊆ Ωn+1 · ··,
and ∪∞ n=1 Ωn = Ω. Define En ≡ E ∩ Ωn . Then there exists Gn ⊇ En such that
µ (Gn ) = ν 1 (En ) , Gn ∈ F and Gn ⊆ Ωn . I claim there exists Fn ∈ F such that
Gn ⊇ En ⊇ Fn and µ (Gn \ Fn ) = 0. To see this, look at the following diagram.

Hn En Fn Hn

Gn \ En

In this diagram, there exists Hn ∈ F containing Gn \ En , represented in the picture

as the set between the dotted lines, such that µ (Hn ) = µ (Gn \ En ) . Then define
Fn ≡ HnC ∩ Gn . This set is in F, is contained in En and as shown in the diagram,
µ (En ) − µ (Fn ) ≤ µ (Hn ) − µ (Gn \ En ) = 0.
Therefore, since µ is a measure,
µ (Gn \ Fn ) = µ (Gn \ En ) + µ (En \ Fn )
= µ (Gn ) − µ (En ) + µ (En ) − µ (Fn ) = 0
Then letting G = ∪n Gn , F ≡ ∪n Fn , it follows G ⊇ E ⊇ F and
µ (G \ F ) ≤ µ (∪n (Gn \ Fn ))
X
≤ µ (Gn \ Fn ) = 0.
n

Thus ν i (G \ F ) = 0 for i = 1, 2. Now E \ F ⊆ G \ F and since (Ω, F2 , ν 2 ) is complete,

it follows E \ F ∈ F2 . Since F ∈ F2 , it follows E = (E \ F ) ∪ F ∈ F2 . Thus F1 ⊆ F2 .
Similarly F2 ⊆ F1 .
Now it only remains to verify ν 1 = ν 2 . Thus let E ∈ F1 = F2 and let G and F be
as just described. Since ν i = µ on F,
µ (F ) ≤ ν 1 (E)
= ν 1 (E \ F ) + ν 1 (F )
≤ ν 1 (G \ F ) + ν 1 (F )
= ν 1 (F ) = µ (F )
Similarly ν 2 (E) = µ (F ) . This proves uniqueness. The construction has also verified
7.14.
Next define an outer measure, µ on P (Ω) as follows. For S ⊆ Ω,
µ (S) ≡ inf {µ (E) : E ∈ F} .
7.5. MEASURES AND OUTER MEASURES 169

Then it is clear µ is increasing. It only remains to verify µ is subadditive. Then let

S = ∪∞ i=1 Si . If any µ (Si ) = ∞, there is nothing to prove so suppose µ (Si ) < ∞ for
each i. Then there exist Ei ∈ F such that Ei ⊇ Si and

µ (Si ) + ε/2i > µ (Ei ) .

Then

µ (S) = µ (∪i Si )
X
≤ µ (∪i Ei ) ≤ µ (Ei )
i
X¡ ¢ X
≤ µ (Si ) + ε/2i = µ (Si ) + ε.
i i

Since ε is arbitrary, this verifies µ is subadditive and is an outer measure as claimed.

Denote by F the σ algebra of measurable sets in the sense of¡Caratheodory.
¢ Then it
follows from the Caratheodory procedure, Theorem 7.5.4, that Ω, F, µ is a complete
measure space. This verifies 1.
Now let E ∈ F. Then from the definition of µ, it follows

µ (E) ≡ inf {µ (F ) : F ∈ F and F ⊇ E} ≤ µ (E) .

If F ⊇ E and F ∈ F , then µ (F ) ≥ µ (E) and so µ (E) is a lower bound for all such
µ (F ) which shows that

µ (E) ≡ inf {µ (F ) : F ∈ F and F ⊇ E} ≥ µ (E) .

This verifies 2.
Next consider 3. Let E ∈ F and let S be a set. I must show

µ (S) ≥ µ (S \ E) + µ (S ∩ E) .

If µ (S) = ∞ there is nothing to show. Therefore, suppose µ (S) < ∞. Then from the
definition of µ there exists G ⊇ S such that G ∈ F and µ (G) = µ (S) . Then from the
definition of µ,

µ (S) ≤ µ (S \ E) + µ (S ∩ E)
≤ µ (G \ E) + µ (G ∩ E)
= µ (G) = µ (S)

This verifies 3.
Claim 4 comes by the definition of µ as used above. The other case is when µ (S) =
∞. However, in this case, you can let G = Ω.
It only remains to verify 5. Let the Ωn be as described above and let E ∈ F such
that E ⊆ Ωn . By 4 there exists H ∈ F such that H ⊆ Ωn , H ⊇ Ωn \ E, and

µ (H) = µ (Ωn \ E) . (7.15)

Then let F ≡ Ωn ∩ H C . It follows F ⊆ E and

¡ ¢
E\F = E ∩ F C = E ∩ H ∪ ΩC
n
= E ∩ H = H \ (Ωn \ E)

Hence from 7.15

µ (E \ F ) = µ (H \ (Ωn \ E)) = 0.
170 MEASURES AND MEASURABLE FUNCTIONS

It follows
µ (E) = µ (F ) = µ (F ) .
In the case where E ∈ F is arbitrary, not necessarily contained in some Ωn , it follows
from what was just shown that there exists Fn ∈ F such that Fn ⊆ E ∩ Ωn and

µ (Fn ) = µ (E ∩ Ωn ) .

Letting F ≡ ∪n Fn
X
µ (E \ F ) ≤ µ (∪n (E ∩ Ωn \ Fn )) ≤ µ (E ∩ Ωn \ Fn ) = 0.
n

Therefore, µ (E) = µ (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 µ is a regular measure defined

³ on B (X)
´ where X is a
finite dimensional normed vector space. Then denoting by X, B (X), µ the completion
of (X, B (X) , µ) , it follows µ is also regular. Furthermore, if a σ algebra, F ⊇ B (X) and
(X, F, µ) is a complete measure space such that for every F ∈ F there exists G ∈ B (X)
such that µ (F ) = µ (G) and G ⊇ F , then F = B (X) and µ = µ.

Proof: Let F ∈ B (X) with µ (F ) < ∞. By Theorem 7.5.6 there exists G ∈ B (X)
such that
µ (G) = µ (G) = µ (F ) .
Now by regularity of µ there exists an open set, V ⊇ G ⊇ F such that

µ (F ) + ε = µ (G) + ε > µ (V ) = µ (V )

Therefore, µ is outer regular. If µ (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
µ (H) = µ (F ). If l < µ (F ) = µ (H) , it follows from regularity of µ there exists K a
compact subset of H such that

l < µ (K) = µ (K)

Thus µ 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 σ algebra of Theorem 7.5.4 applied to the outer
measure µ in Theorem 7.2.1 on which µ 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

µ (E) = sup {µ (K) : K is a closed and bounded set, K ⊆ E} (7.16)

µ (E) = inf {µ (V ) : V is an open set, V ⊇ E} (7.17)

7.6. ONE DIMENSIONAL LEBESGUE STIELTJES MEASURE 171

Proof: The first task is to show (a, b) ∈ S. I need to show that for every S ⊆ R,
³ ´
C
µ (S) ≥ µ (S ∩ (a, b)) + µ S ∩ (a, b) (7.18)

Suppose first 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 µ (S) = µ (S). Suppose next that (c, d) ⊇ (a, b) . In this case, the right side
of the above reduces to

µ ((a, b)) + µ ((c, a] ∪ [b, d))

≤ F (b−) − F (a+) + F (a+) − F (c+) + F (d−) − F (b−)
= F (d−) − F (c+) = µ ((c, d))

The only other cases are c ≤ a < d ≤ b or a ≤ c < d ≤ b. Consider the first of these
cases. Then the right side of 7.18 for S = (c, d) is

µ ((a, d)) + µ ((c, a]) = F (d−) − F (a+) + F (a+) − F (c+)

= F (d−) − F (c+) = µ ((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 µ (S) = ∞. Suppose then that µ (S) < ∞ and let

S ⊆ ∪∞
k=1 (ak , bk )

such that
∞
X ∞
X
µ (S) + ε > (F (bk −) − F (ak +)) = µ ((ak , bk )) .
k=1 k=1

Then since µ is an outer measure, and using what was just shown,
³ ´
C
µ (S ∩ (a, b)) + µ S ∩ (a, b)
³ ´
C
≤ µ (∪∞ ∞
k=1 (ak , bk ) ∩ (a, b)) + µ ∪k=1 (ak , bk ) ∩ (a, b)
∞
X ³ ´
C
≤ µ ((ak , bk ) ∩ (a, b)) + µ (ak , bk ) ∩ (a, b)
k=1
X∞
≤ µ ((ak , bk )) ≤ µ (S) + ε.
k=1

Since ε 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 σ algebra, their union is also in S. It follows every
closed set is in S also. This is because S is a σ 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 σ algebra of µ measurable sets,³ F includes
´ B (R). Consider the comple-
tion of the measure space, (R, B (R) , µ) , R, B (R), µ . By the uniqueness assertion in
Theorem 7.5.6 and the fact that (R, F, µ) is complete, this coincides with (R, F, µ) be-
cause the construction of µ implies µ is outer regular and for every F ∈ F , there exists
G ∈ B (R) containing F such that µ (F ) = µ (G) . In fact, you can take G to equal a
countable intersection of open sets. By Theorem 7.4.6 µ is regular on every set of B (R) ,
this because µ is finite on compact sets. Therefore, by Theorem 7.5.7 µ = µ is regular
on F which verifies the last two claims. This proves the theorem. ¥
172 MEASURES AND MEASURABLE FUNCTIONS

7.7 Measurable Functions

The integral will be defined 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. xn → ∞ exactly when
for all l ∈ R, there exists N such that if n ≥ N, then

xn > l.

This is what it means for a sequence to converge to ∞. Don’t 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 : Ω → (−∞, ∞] where F is a σ algebra of subsets of Ω. Then

the following are equivalent.

f −1 ((d, ∞]) ∈ F for all finite d,

f −1 ((−∞, d)) ∈ F for all finite d,

f −1 ([d, ∞]) ∈ F for all finite d,
f −1 ((−∞, d]) ∈ F for all finite d,
f −1 ((a, b)) ∈ F for all a < b, −∞ < a < b < ∞.

Proof: First note that the first and the third are equivalent. To see this, observe

f −1 ([d, ∞]) = ∩∞
n=1 f
−1
((d − 1/n, ∞]),

and so if the first 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 first.

Similarly, the second and fourth conditions are equivalent. Now

f −1 ((−∞, d]) = (f −1 ((d, ∞]))C

so the first and fourth conditions are equivalent. Thus the first 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,

¡ ¢C
f −1 ([d, ∞]) = ∪∞
k=1 f
−1
((−k + d, d)) ∈ F

and so the third condition holds. Therefore, all five conditions are equivalent. This
proves the lemma. ¥
This lemma allows for the following definition of a measurable function having values
in (−∞, ∞].
7.7. MEASURABLE FUNCTIONS 173

Definition 7.7.2 Let (Ω, F, µ) be a measure space and let f : Ω → (−∞, ∞].
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 fn and f be functions mapping Ω to (−∞, ∞] where F is a

σ algebra of measurable sets of Ω. Then if fn is measurable, and f (ω) = limn→∞ fn (ω),
it follows that f is also measurable. (Pointwise limits of measurable functions are mea-
surable.)

−1
¡ 1 1
¢
£ Proof: The¤ idea is to show f ((a, b)) ∈ F. Let V m ≡ a+ m, b − m and V m =
1 1
a + m , b − m . Then for all m, Vm ⊆ (a, b) and

(a, b) = ∪∞ ∞
m=1 Vm = ∪m=1 V m .

Note that Vm 6= ∅ for all m large enough. Since f is the pointwise limit of fn ,

f −1 (Vm ) ⊆ {ω : fk (ω) ∈ Vm for all k large enough} ⊆ f −1 (V m ).

You should note that the expression in the middle is of the form
−1
∪∞ ∞
n=1 ∩k=n fk (Vm ).

Therefore,
−1
f −1 ((a, b)) = ∪∞
m=1 f
−1
(Vm ) ⊆ ∪∞ ∞ ∞
m=1 ∪n=1 ∩k=n fk (Vm )

⊆ ∪∞
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 (Ω, F, µ) be a measure space and let f : Ω → (−∞, ∞].
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 (ak , bk ) . Hence

f −1 (U ) = ∪∞
k=1 f
−1
((ak , bk )) ∈ F

because F is a σ algebra.
From this proposition, it follows one can generalize the definition of a measurable
function to those which have values in any normed vector space as follows.

Definition 7.7.5 Let (Ω, F, µ) be a measure space and let f : Ω → 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 (Ω, F, µ) be a measure space and let X, Y be normed vector
spaces and g : X → Y continuous. Then if f : Ω → X is F measurable, it follows g ◦ f
is also F measurable.
174 MEASURES AND MEASURABLE FUNCTIONS

−1
Proof: From the definition, it suffices to show (g ◦ f ) (U ) ∈ F whenever U is an
open set in Y. However, since g is continuous, it follows g−1 (U ) is open and so
−1 ¡ ¢
(g ◦ f ) (U ) = f −1 g−1 (U ) = f −1 (an open set) ∈ F.

This proves the theorem. ¥

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 {v1 , · · ·, vn } is a basis for X and π k is the projection onto the k th component,
then π k ◦ f is a measurable F valued function. Does it go the other way? That is, if
it is known that π 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 {|xi | , i = 1, 2, · · ·, n} for x ∈ Fn . Then every set U
which is open in Fn is the countable union of balls of the form B (x,r) where the open
ball is defined in terms of the above norm.

Proof: By Theorem 5.8.3 if you consider the two normed vector spaces (Fn , |·|)
and (Fn , ||·||) , 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, Dn is a countable dense subset of Fn . It is countable
because it is a finite Cartesian product of countable sets and you can use Theorem 2.1.7
of Page 15 repeatedly. It is dense because if x ∈ Fn , then by density of D, there exists
dj ∈ D such that
|dj − xj | < ε
then d ≡ (d1 , · · ·, dn ) is such that ||d − x|| < ε.
Now consider the set of open balls,

B ≡ {B (d, r) : d ∈ Dn , 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 Fn and x ∈ U . Then there
exists δ > 0 such that B (x, δ) ⊆ U. There exists d ∈ Dn ∩B (x, δ/5) . Then pick rational
number δ/5 < r < 2δ/5. Consider the set of B, B (d, r) . Then x ∈ B (d, r) because
r > δ/5. However, it is also the case that B (d, r) ⊆ B (x, δ) because if y ∈ B (d, r) then

||y − x|| ≤ ||y − d|| + ||d − x||

2δ δ
< + < δ.
5 5
This proves the lemma. ¥

Corollary 7.7.8 Let (Ω, F, µ) be a measure space and let X be a normed vector
space with basis {v1 , · · ·, vn } . Let π k be the k th projection map onto the k th component.
Thus
Xn
π k x ≡ xk where x = xi vi .
i=1

Then each π k ◦ f is a measurable F valued function if and only if f is a measurable X

valued function.
7.7. MEASURABLE FUNCTIONS 175

Proof: The if part has already been noted. Suppose that each π k ◦ f is an F valued
measurable function. Let g : X → Fn be given by
g (x) ≡ (π 1 x, · · ·,π 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 Fn . To
see this, start with V open set in X. Since g−1 is continuous, g (V ) is open in Fn and
so V = g−1 (g (V )) . Therefore, it suffices to show that for every U an open set in Fn ,
¡ ¢ −1
f −1 g−1 (U ) = (g ◦ f ) (U ) ∈ F.
By Lemma 7.7.7 there are countably many open balls of the form B (xj , rj ) such that
U is equal to the union of these balls. Thus
−1 −1
(g ◦ f ) (U ) = (g ◦ f ) (∪∞
k=1 B (xk , rk ))
−1
= ∪∞
k=1 (g ◦ f ) (B (xk , rk )) (7.19)
Now from the definition of the norm,
n
Y
B (xk , rk ) = (xkj − δ, xkj + δ)
j=1

and so
−1 −1
(g ◦ f ) (B (xk , rk )) = ∩nj=1 (π j ◦ f ) ((xkj − δ, xkj + δ)) ∈ F.
It follows 7.19 is the countable union of sets in F and so it is also in F. This proves the
corollary. ¥
n
Note that if {fi }i=1 are measurable functions defined on (Ω, F, µ) having values in
follows f is a measurable Fn valued function. Now let
F then letting f ≡ (f1 , · · ·, fn ) , itP
n
Σ : Fn → F be given by Σ (x) ≡ k=1 ak xk . Then Σ is linear and so by Theorem 5.8.3
it follows Σ is continuous. Hence by Theorem 7.7.6, Σ (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.
Theorem 7.7.9 Let {fn } be a sequence of measurable functions mapping Ω to
X where X is a normed vector space and (Ω, F) is a measure space. Suppose also that
f (ω) = limn→∞ fn (ω) for all ω ∈ Ω. Then f is also a measurable function.
Proof: It is required to show f −1 (U ) is measurable for all U open. Let
½ ¾
¡ C
¢ 1
Vm ≡ x ∈ U : dist x, U > .
m
Thus ½ ¾
¡ ¢ 1
Vm ⊆ x ∈ U : dist x, U C ≥
m
and Vm ⊆ Vm ⊆ Vm+1 and ∪m Vm = U. Then since Vm is open, it follows that if
f (ω) ∈ Vm then for all sufficiently large k, it must be the case fk (ω) ∈ Vm also. That
is, ω ∈ fk−1 (Vm ) for all sufficiently large k. Thus
−1
f −1 (Vm ) = ∪∞ ∞
n=1 ∩k=n fk (Vm )
176 MEASURES AND MEASURABLE FUNCTIONS

and so

f −1 (U ) = ∪∞
m=1 f
−1
(Vm )
−1
= ∪m=1 ∪n=1 ∩∞
∞ ∞
k=n fk (Vm )
¡ ¢
⊆ ∪∞
m=1 f
−1
Vm = f −1 (U )

which shows f −1 (U ) is measurable. The step from the second to the last line follows
−1
because if ω ∈ ∪∞ ∞
n=1 ∩k=n fk (Vm ) , this says fk (ω) ∈ Vm for all k large enough.
Therefore, the point of X to which the sequence {fk (ω)} converges must be in Vm
which equals Vm ∪ Vm0 , the limit points of Vm . This proves the theorem. ¥
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 XE (ω) be defined by

½
1 if ω ∈ E
XE (x) =
0 if ω ∈
/E

Theorem 7.7.11 Let f : Ω → X where X is some normed vector space. Sup-

pose
m
X
f (ω) = xk XAk (ω)
k=1

where each xk ∈ X and the Ak 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 ) = ∪ {Ak : xk ∈ U } , a finite 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

(Ω, F, µ). Then there exists a sequence of nonnegative simple functions {sn } satisfying

0 ≤ sn (ω) (7.20)

· · · sn (ω) ≤ sn+1 (ω) · ··

f (ω) = lim sn (ω) for all ω ∈ Ω. (7.21)
n→∞

If f is bounded the convergence is actually uniform.

Proof : First note that

C
f −1 ([a, b)) = f −1 ((−∞, a)) ∩ f −1 ((−∞, b))
³ ´C
C
= f −1 ((−∞, a)) ∪ f −1 ((−∞, b)) ∈ F.

Letting I ≡ {ω : f (ω) = ∞} , define

n
2
X k
tn (ω) = X −1 (ω) + nXI (ω).
n f ([k/n,(k+1)/n))
k=0
7.8. EXERCISES 177

Then tn (ω) ≤ f (ω) for all ω and limn→∞ tn (ω) = f (ω) for all ω. This is because
n
tn (ω) = n for ω ∈ I and if f (ω) ∈ [0, 2 n+1 ), then
1
0 ≤ f (ω) − tn (ω) ≤ . (7.22)
n
Thus whenever ω ∈
/ I, the above inequality will hold for all n large enough. Let

s1 = t1 , s2 = max (t1 , t2 ) , s3 = max (t1 , t2 , t3 ) , · · ·.

Then the sequence {sn } satisfies 7.20-7.21.

To verify the last claim, note that in this case the term nXI (ω) is not present.
Therefore, for all n large enough that 2n n ≥ f (ω) for all ω, 7.22 holds for all ω. Thus
the convergence is uniform. This proves the theorem. ¥

7.8 Exercises
1. Let C be a set whose elements are σ algebras of subsets of Ω. Show ∩C is a σ
algebra also.
2. Let Ω be any set. Show P (Ω) , the set of all subsets of Ω is a σ algebra. Now let
L denote some subset of P (Ω) . Consider all σ algebras which contain L. Show
the intersection of all these σ algebras which contain L is a σ algebra containing
L and it is the smallest σ algebra containing L, denoted by σ (L). When Ω is a
normed vector space, and L consists of the open sets, σ (L) is called the σ algebra
of Borel sets.
3. Consider Ω = [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 σ algebra.
(This is a sick σ algebra.) Now let µ : S → [0, ∞] be defined by µ (F ) = 1 if F C
is countable and µ (F ) = 0 if F is countable. Show µ is a measure on S.
4. Let Ω = N, the positive integers and let a σ algebra be given by F = P (N), the
set of all subsets of N. What are the measurable functions having values in C?
Let µ (E) be the number of elements of E where E is a subset of N. Show µ is a
measure.
5. Let F be a σ algebra of subsets of Ω and suppose F has infinitely many elements.
Show that F is uncountable. Hint: You might try to show there exists a count-
able sequence of disjoint sets of F, {Ai }. It might be easiest to verify this by
contradiction if it doesn’t exist rather than a direct construction however, I have
seen this done several ways. Once this has been done, you can define a map, θ,
from P (N) into F which is one to one by θ (S) = ∪i∈S Ai . Then argue P (N) is
uncountable and so F is also uncountable.
6. A probability space is a measure space, (Ω, F, P ) where the measure, P has the
property that P (Ω) = 1. Such a measure is called a probability measure. Random
vectors are measurable functions, X, mapping a probability space, (Ω, F, P ) to
Rn . Thus X (ω) ∈ Rn for each ω ∈ Ω and P is a probability measure defined on
the sets of F, a σ algebra of subsets of Ω. For E a Borel set in Rn , define
¡ ¢
µ (E) ≡ P X−1 (E) ≡ probability that X ∈ E.

Show this is a well defined probability measure on the Borel sets of Rn . Thus
µ (E) = P (X (ω) ∈ E) . It is called the distribution. Explain why µ must be
regular.
178 MEASURES AND MEASURABLE FUNCTIONS

7. Suppose (Ω, S, µ) is a measure space which may not be complete. Show that
another way to complete the measure space is to define 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 µ (E) = µ1 (F )? Explain.
The Abstract Lebesgue
Integral

The general Lebesgue integral requires a measure space, (Ω, F, µ) 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 {an } is an increasing
sequence of points of [0, ∞] , then supn an = limn→∞ an which is obvious from the
definition of sup.

8.1 Definition For Nonnegative Measurable Functions

8.1.1 Riemann Integrals For Decreasing Functions
First of all, the notation
[g < f ]
is short for
{ω ∈ Ω : g (ω) < f (ω)}
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.

Definition 8.1.1 For f a nonnegative decreasing function defined on a finite

interval [a, b] , define
Z b Z b Z b
f (λ) dλ ≡ lim M ∧ f (λ) dλ = sup M ∧ f (λ) dλ
a M →∞ a M a

where a ∧ b means the minimum of a and b. Note that for f bounded,

Z b Z b
sup M ∧ f (λ) dλ = f (λ) dλ
M a a

where the integral on the right is the usual Riemann integral because eventually M > f .
For f a nonnegative decreasing function defined on [0, ∞),
Z ∞ Z R Z R
f dλ ≡ lim f dλ = sup f dλ
0 R→∞ R−1 R>1 R−1

Since decreasing bounded functions are Riemann integrable, the above definition is
well defined. Now here are some obvious properties.

179
180 THE ABSTRACT LEBESGUE INTEGRAL

Lemma 8.1.2 Let f be a decreasing nonnegative function defined on an interval

[a, b] . Then if [a, b] = ∪m
k=1 Ik where Ik ≡ [ak , bk ] and the intervals Ik are non overlap-
ping, it follows
Z b m Z bk
X
f dλ = f dλ.
a k=1 ak

Proof: This follows from the computation,

Z b Z b
f dλ ≡ lim f ∧ M dλ
a M →∞ a
m Z bk
X m Z
X bk
= lim f ∧ M dλ = f dλ
M →∞ ak ak
k=1 k=1

Note both sides could equal +∞. This proves the lemma. ¥

8.1.2 The Lebesgue Integral For Nonnegative Functions

Here is the definition of the Lebesgue integral of a function which is measurable and
has values in [0, ∞].

Definition 8.1.3 Let (Ω, F, µ) be a measure space and suppose f : Ω → [0, ∞]

is measurable. Then define
Z Z ∞
f dµ ≡ µ ([f > λ]) dλ
0

which makes sense because λ → µ ([f > λ]) is nonnegative and decreasing.

8.2 The Lebesgue Integral For Nonnegative Simple

Functions
To begin with, here is a useful lemma.

Lemma 8.2.1 If f (λ) = 0 for all λ > a, where f is a decreasing nonnegative func-
tion, then Z Z
∞ a
f (λ) dλ = f (λ) dλ.
0 0

Proof: From the definition,

Z ∞ Z R Z R
f (λ) dλ = lim f (λ) dλ = sup f (λ) dλ
0 R→∞ R−1 R>1 R−1
Z R
= sup sup f (λ) ∧ M dλ
R>1 M R−1
Z R
= sup sup f (λ) ∧ M dλ
M R>1 R−1
Z a
= sup sup f (λ) ∧ M dλ
M R>1 R−1
Z a Z a
= sup f (λ) ∧ M dλ ≡ f (λ) dλ.
M 0 0
8.2. THE LEBESGUE INTEGRAL FOR NONNEGATIVE SIMPLE FUNCTIONS 181

This proves the lemma. ¥

Now the Lebesgue integral for a nonnegative function has been defined, what does
it do to a nonnegative simple function? Recall a nonnegative simple function is one
which has finitely many nonnegative values which it assumes on measurable sets. Thus
a simple function can be written in the form
n
X
s (ω) = ci XEi (ω)
i=1

where the ci are each nonnegative real numbers, the distinct values of s.
Pp
Lemma 8.2.2 Let s (ω) = i=1 ai XEi (ω) be a nonnegative simple function with the
ai the distinct non zero values of s. Then
Z p
X
sdµ = ai µ (Ei ) . (8.1)
i=1

Proof: Without loss of generality, assume 0 ≡ a0 < a1 < a2 < · · · < ap and that
µ (Ei ) < ∞. Here is why. If µ (Ei ) = ∞, then letting a ∈ (ai−1 , ai ) by Lemma 8.2.1,
the left side would be
Z ap Z ap
µ ([s > λ]) dλ ≥ µ ([ai XEi > λ]) dλ
0 0
Z a
= sup µ ([ai XEi > λ]) ∧ M dλ
M 0
= sup aM = ∞
M

and so both sides are equal to ∞. Thus it can be assumed for each i, µ (Ei ) < ∞. Then
letting a0 ≡ 0, it follows from Lemma 8.2.1 and Lemma 8.1.2,
Z ∞ Z ap p Z
X ak
µ ([s > λ]) dλ = µ ([s > λ]) dλ = µ ([s > λ]) dλ
0 0 k=1 ak−1

p Z
X ak p
X p
X p
X
= µ (Ej ) dλ = (ak − ak−1 ) µ (Ej )
k=1 ak−1 j=k k=1 j=k

p
X p
X p
X p
X
= ak µ (Ej ) − ak−1 µ (Ej )
k=1 j=k k=1 j=k

p
X p
X p−1
X p
X
= ak µ (Ej ) − ak µ (Ej )
k=1 j=k k=0 j=k+1

 
p−1
X p
X p
X
= ap µ (Ep ) − 0 + ak  µ (Ej ) − µ (Ej )
k=1 j=k j=k+1
p−1
X p
X
= ap µ (Ep ) + ak µ (Ek ) = ak µ (Ek ) .
k=1 k=1

This proves the lemma. ¥

182 THE ABSTRACT LEBESGUE INTEGRAL

Lemma 8.2.3 Let the nonnegative simple function, s be defined as

n
X
s (ω) = ci XEi (ω)
i=1

where the ci are not necessarily distinct but the Ei are disjoint. It follows that
Z n
X
sdµ = ci µ (Ei ) .
i=1

Proof: Let the values of s be {a1 , · · · , am }. Therefore, since the Ei are disjoint,
each ai equal to one of the cj . Let Ai ≡ ∪ {Ej : cj = ai }. Then from Lemma 8.2.2 it
follows that
Z m
X Xm X
sdµ = ai µ (Ai ) = ai µ (Ej )
i=1 i=1 {j:cj =ai }
m
X X n
X
= cj µ (Ej ) = ci µ (Ei ) .
i=1 {j:cj =ai } i=1

This proves the

R lemma. ¥ R
Note that sdµ could equal +∞ if µ (Ak ) = ∞ and ak > 0 for some k, but sdµ is
well defined because s ≥ 0. Recall that 0 · ∞ = 0.

Lemma 8.2.4 If a, b ≥ 0 and if s and t are nonnegative simple functions, then

Z Z Z
as + btdµ = a sdµ + b tdµ.

Proof: Let
n
X m
X
s(ω) = αi XAi (ω), t(ω) = β j XBj (ω)
i=1 i=1

where αi are the distinct values of s and the β j are the distinct values of t. Clearly as+bt
is a nonnegative simple function because it has finitely many values on measurable sets
In fact,
Xm X n
(as + bt)(ω) = (aαi + bβ j )XAi ∩Bj (ω)
j=1 i=1

where the sets Ai ∩ Bj are disjoint and measurable. By Lemma 8.2.3,

Z m X
X n
as + btdµ = (aαi + bβ j )µ(Ai ∩ Bj )
j=1 i=1
n
X m
X m X
X n
= a αi µ(Ai ∩ Bj ) + b β j µ(Ai ∩ Bj )
i=1 j=1 j=1 i=1
Xn m
X
= a αi µ(Ai ) + b β j µ(Bj )
i=1 j=1
Z Z
= a sdµ + b tdµ.

This proves the lemma. ¥

8.3. THE MONOTONE CONVERGENCE THEOREM 183

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. Here is a lemma
which will make things work out.
Lemma 8.3.1 Let f : Ω → [0, ∞] be measurable. Then
Z R
µ ([f > λ]) dλ
R−1
k
X
= sup µ ([f > ihk ]) hk
k∈N i=1
¡ ¢
where hk ≡ R − R−1 /k.
Proof: First,
Z R k
X
(µ ([f > λ]) ∧ M ) dλ = sup hk (µ ([f > ihk ]) ∧ M )
R−1 k∈N i=1

because the sum on the right is a lower sum for the Riemann integral on the left. Now
from the definition,
Z R Z R
µ ([f > λ]) dλ ≡ sup (µ ([f > λ]) ∧ M ) dλ
R−1 M R−1
k
X
= sup sup hk (µ ([f > ihk ]) ∧ M )
M k∈N i=1

k
X k
X
= sup lim hk (µ ([f > ihk ]) ∧ M ) = sup hk µ ([f > ihk ])
k∈N M →∞ i=1 k∈N i=1

Note this last equality holds even if µ ([f > ihk ]) = ∞. This proves the lemma. ¥

Theorem 8.3.2 (Monotone Convergence theorem) Let f have values in [0, ∞]

and suppose {fn } is a sequence of nonnegative measurable functions having values in
[0, ∞] and satisfying
lim fn (ω) = f (ω) for each ω.
n→∞
· · · fn (ω) ≤ fn+1 (ω) · · ·
Then f is measurable and Z Z
f dµ = lim fn dµ.
n→∞

Proof: By Lemma 8.3.1

Z R k
X
µ ([f > λ]) dλ = sup µ ([f > ihk ]) hk
R−1 k∈N i=1
¡ ¢
where hk ≡ R − R−1 /k, and so since µ ([fn > ihk ]) increases in n to µ ([f > ihk ]) ,
k
X k
X
= sup sup µ ([fn > ihk ]) hk = sup lim µ ([fn > ihk ]) hk
k∈N i=1 n k∈N i=1 n→∞

k
X k
X
= sup lim µ ([fn > ihk ]) hk = sup sup µ ([fn > ihk ]) hk
k∈N n→∞ i=1 k∈N n i=1
184 THE ABSTRACT LEBESGUE INTEGRAL

k
X Z R
= sup sup µ ([fn > ihk ]) hk = sup µ ([fn > λ]) dλ,
n k∈N n R−1
i=1
the sequence {µ ([fn > λ])}being increasing in n. Then from the above,
Z Z ∞ Z R
f dµ ≡ µ ([f > λ]) dλ ≡ sup µ ([f > λ]) dλ
0 R R−1
Z R Z R
= sup sup µ ([fn > λ]) dλ = sup sup µ ([fn > λ]) dλ
R n R−1 n R R−1
Z ∞ Z
≡ sup µ ([fn > λ]) dλ = lim fn dµ
n 0 n→∞

This proves the theorem. ¥

To illustrate what goes wrong without the Lebesgue integral, consider the following
example.
Example 8.3.3 Let {rn } denote the rational numbers in [0, 1] and let
½
1 if t ∈
/ {r1 , · · · , rn }
fn (t) ≡
0 otherwise
Then fn (t) ↑ f (t) where f is the function which is one on the rationals and zero on
the irrationals. Each fn is RRiemann integrable
R (why?) but f is not Riemann integrable.
Therefore, you can’t write f dx = limn→∞ fn dx.
A meta-mathematical observation related to this type of example is this. If you can
choose your functions, you don’t need the Lebesgue integral. The Riemann Darboux
integral is just fine. It is when you can’t 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 Definitions

To review and summarize the above, if f ≥ 0 is measurable,
Z Z ∞
f dµ ≡ µ ([f > λ]) dλ (8.2)
0
R
another way to get the same thing for f dµ is to take an increasing sequence of non-
negative simple functions, {sn } with sn (ω) → f (ω) and then by monotone convergence
theorem, Z Z
f dµ = lim sn
n→∞
Pm
where if sn (ω) = j=1 ci XEi (ω) ,
Z m
X
sn dµ = ci µ (Ei ) .
i=1

Similarly this also shows that for such nonnegative measurable function,
Z ½Z ¾
f dµ = sup s : 0 ≤ s ≤ f, s simple

Here is an equivalent definition of the integral of a nonnegative measurable function.

The fact it is well defined has been discussed above.
8.5. FATOU’S LEMMA 185

Definition 8.4.1 For s a nonnegative simple function,

Xn Z X n
s (ω) = ck XEk (ω) , s = ck µ (Ek ) .
k=1 k=1

For f a nonnegative measurable function,

Z ½Z ¾
f dµ = sup s : 0 ≤ s ≤ f, s simple .

8.5 Fatou’s Lemma

The next theorem, known as Fatou’s lemma is another important theorem which justifies
the use of the Lebesgue integral.
Theorem 8.5.1 (Fatou’s lemma) Let fn be a nonnegative measurable function
with values in [0, ∞]. Let g(ω) = lim inf n→∞ fn (ω). Then g is measurable and
Z Z
gdµ ≤ lim inf fn dµ.
n→∞

In other words, Z ³ Z
´
lim inf fn dµ ≤ lim inf fn dµ
n→∞ n→∞

Proof: Let gn (ω) = inf{fk (ω) : k ≥ n}. Then

−1
gn−1 ([a, ∞]) = ∩∞
k=n fk ([a, ∞])
¡ ∞ −1 ¢C
= ∪k=n fk ([a, ∞])C ∈ F.
Thus gn is measurable by Lemma 7.7.1. Also g(ω) = limn→∞ gn (ω) so g is measurable
because it is the pointwise limit of measurable functions. Now the functions gn form an
increasing sequence of nonnegative measurable functions so the monotone convergence
theorem applies. This yields
Z Z Z
gdµ = lim gn dµ ≤ lim inf fn dµ.
n→∞ n→∞

The last inequality holding because

Z Z
gn dµ ≤ fn dµ.
R
(Note that it is not known whether limn→∞ fn dµ 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
Z Z Z
(af + bg) dµ = a f dµ + b gdµ. (8.3)
186 THE ABSTRACT LEBESGUE INTEGRAL

Proof: By Theorem 7.7.12 on Page 176 there exist increasing sequences of nonneg-
ative simple functions, sn → f and tn → g. Then af + bg, being the pointwise limit
of the simple functions asn + btn , is measurable. Now by the monotone convergence
theorem and Lemma 8.2.4,
Z Z
(af + bg) dµ = lim asn + btn dµ
n→∞
µ Z Z ¶
= lim a sn dµ + b tn dµ
n→∞
Z Z
= a f dµ + b gdµ.

This proves the theorem. ¥

As long as you are allowing functions to take the value +∞, you cannot consider
something like f +(−g) and so you can’t 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, L1

The functions considered here have values in C, a vector space.

Definition 8.7.1 Let (Ω, S, µ) be a measure space and suppose f : Ω → C. Then

f is said to be measurable if both Re f and Im f are measurable real valued functions.

Definition 8.7.2 A complex simple function will be a function which is of the

form
n
X
s (ω) = ck XEk (ω)
k=1

where ck ∈ C and µ (Ek ) < ∞. For s a complex simple function as above, define
n
X
I (s) ≡ ck µ (Ek ) .
k=1

Lemma 8.7.3 The definition, 8.7.2 is well defined. Furthermore, I is linear on the
vector space of complex simple functions. Also the triangle inequality holds,

|I (s)| ≤ I (|s|) .
Pn P
Proof: Suppose k=1 ck XEk (ω) = 0. Does it follow that k ck µ (Ek ) = 0? The
supposition implies
n
X n
X
Re ck XEk (ω) = 0, Im ck XEk (ω) = 0. (8.4)
k=1 k=1
P
Choose λ large and positive so that λ + Re ck ≥ 0. Then adding k λXEk to both sides
of the first equation above,
n
X n
X
(λ + Re ck ) XEk (ω) = λXEk
k=1 k=1
8.7. THE LEBESGUE INTEGRAL, L1 187

R
and by Lemma 8.2.4 on Page 182, it follows upon taking of both sides that
n
X n
X
(λ + Re ck ) µ (Ek ) = λµ (Ek )
k=1 k=1
Pn
which implies k=1 Re ck µ (Ek ) = 0. Similarly,
n
X
Im ck µ (Ek ) = 0
k=1
Pn
and so k=1 ck µ (Ek ) = 0. Thus if
X X
cj XEj = dk XFk
j k
P P
then j cj XEj + k (−dk ) XFk = 0 and so the result just established verifies
X X
cj µ (Ej ) − dk µ (Fk ) = 0
j k

which proves I is well defined.

That I is linear is now obvious. It only remains to verify the triangle inequality.
Let s be a simple function, X
s= cj XEj
j

Then pick θ ∈ C such that θI (s) = |I (s)| and |θ| = 1. Then from the triangle inequality
for sums of complex numbers,
X
|I (s)| = θI (s) = I (θs) = θcj µ (Ej )
j
¯ ¯
¯ ¯
¯X ¯ X
= ¯ θc µ (E )¯≤ |θcj | µ (Ej ) = I (|s|) .
¯ j j ¯
¯ j ¯ j

This proves the lemma. ¥

With this lemma, the following is the definition of L1 (Ω) .

Definition 8.7.4 f ∈ L1 (Ω) means there exists a sequence of complex simple

functions, {sn } such that

sn (ω) → f (ω) for all ωR ∈ Ω

(8.5)
limm,n→∞ I (|sn − sm |) = limn,m→∞ |sn − sm | dµ = 0

Then
I (f ) ≡ lim I (sn ) . (8.6)
n→∞

Lemma 8.7.5 Definition 8.7.4 is well defined.

Proof: There are several things which need to be verified. First suppose 8.5. Then
by Lemma 8.7.3

|I (sn ) − I (sm )| = |I (sn − sm )| ≤ I (|sn − sm |)

188 THE ABSTRACT LEBESGUE INTEGRAL

and for m, n large enough this last is given to be small so {I (sn )} is a Cauchy sequence
in C and so it converges. This verifies the limit in 8.6 at least exists. It remains to
consider another sequence {tn } having the same properties as {sn } and verifying I (f )
determined by this other sequence is the same. By Lemma 8.7.3 and Fatou’s lemma,
Theorem 8.5.1 on Page 185,
Z
|I (sn ) − I (tn )| ≤ I (|sn − tn |) = |sn − tn | dµ
Z
≤ |sn − f | + |f − tn | dµ
Z Z
≤ lim inf |sn − sk | dµ + lim inf |tn − tk | dµ < ε
k→∞ k→∞

whenever n is large enough. Since ε is arbitrary, this shows the limit from using the tn
is the same as the limit from using sn . This proves the lemma. ¥
Consider the following picture. I have just given a definition of an integral for
functions having values in C. However, [0, ∞) ⊆ C.

complex valued functions values in [0, ∞]

R
What if f has values in [0, ∞)? Earlier f dµ was defined for such functions and now
IR (f ) has been defined. Are they the same? If so, I can be regarded as an extension of
dµ to a larger class of functions.

Lemma 8.7.6 Suppose f has values in [0, ∞) and f ∈ L1 (Ω) . Then f is measurable
and Z
I (f ) = f dµ.

Proof: Since f is the pointwise limit of a sequence of complex simple functions,

{sn } having the properties described in Definition 8.7.4, it follows

f (ω) = lim Re sn (ω)

n→∞

and so f is measurable. Also it is always the case that if a, b are real numbers,
¯ + ¯
¯a − b+ ¯ ≤ |a − b|

and so
Z ¯ ¯ Z Z
¯ + +¯
¯(Re sn ) − (Re sm ) ¯ dµ ≤ |Re sn − Re sm | dµ ≤ |sn − sm | dµ
8.8. APPROXIMATION WITH SIMPLE FUNCTIONS 189

where x+ ≡ 12 (|x| + x) , the positive part of the real number, x. 1 Thus there is no loss
of generality in assuming {sn } is a sequence of complex simple functions
R having values
in [0, ∞). Then since for such complex simple functions, I (s) = sdµ,
¯ Z ¯ ¯Z Z ¯ Z
¯ ¯ ¯ ¯
¯I (f ) − f dµ¯ ≤ |I (f ) − I (sn )| + ¯ sn dµ − f dµ¯ < ε + |sn − f | dµ
¯ ¯ ¯ ¯

whenever n is large enough. But by Fatou’s lemma, Theorem 8.5.1 on Page 185, the
last term is no larger than
Z
lim inf |sn − sk | dµ < ε
k→∞

R
whenever n is large enough. Since ε is arbitrary, this shows I (f ) = f dµ as claimed.
¥ R
As explained
R above, I can be regarded as an extension R of dµ so from now on, the
usual symbol, dµ will be used. It is now easy to verify dµ is linear on L1 (Ω) .

8.8 Approximation With Simple Functions

The next theorem says the integral as defined above is linear and also gives a way to
compute the integral in terms of real and imaginary parta. In addition, functions in L1
can be approximated with simple functions.
R
Theorem 8.8.1 dµ is linear on L1 (Ω) and L1 (Ω) is a complex vector space.
If f ∈ L (Ω) , then Re f, Im f, and |f | are all in L1 (Ω) . Furthermore, for f ∈ L1 (Ω) ,
1

Z Z Z µZ Z ¶
+ − + −
f dµ = (Re f ) dµ − (Re f ) dµ + i (Im f ) dµ − (Im f ) dµ ,

and the triangle inequality holds,

¯Z ¯ Z
¯ ¯
¯ f dµ¯ ≤ |f | dµ
¯ ¯

Also for every f ∈ L1 (Ω) , for every ε > 0 there exists a simple function s such that
Z
|f − s| dµ < ε.

Proof: First it is necessary to verify that L1 (Ω) is really a vector space because
it makes no sense to speak of linear maps without having these maps defined on a
vector space. Let f, g be in L1 (Ω) and let a, b ∈ C. Then let {sn } and {tn } be
sequences of complex simple functions associated with f and g respectively as described
in Definition 8.7.4. Consider {asn + btn } , another sequence of complex simple functions.
Then asn (ω) + btn (ω) → af (ω) + bg (ω) for each ω. Also, from Lemma 8.7.3
Z Z Z
|asn + btn − (asm + btm )| dµ ≤ |a| |sn − sm | dµ + |b| |tn − tm | dµ

1 The 1
negative part of the real number x is defined to be x− ≡ 2
(|x| − x) . Thus |x| = x+ + x− and
x= x+ − x− . .
190 THE ABSTRACT LEBESGUE INTEGRAL

and the sum of the two terms on the right converge to zero as m, n → ∞. Thus af +bg ∈
L1 (Ω) . Also
Z Z
(af + bg) dµ ≡ lim (asn + btn ) dµ
n→∞
µ Z Z ¶
= lim a sn dµ + b tn dµ
n→∞
Z Z
= a lim sn dµ + b lim tn dµ
n→∞ n→∞
Z Z
= a f dµ + b gdµ.

If {sn } is a sequence of complex simple functions described in Definition 8.7.4 cor-

responding to f , then {|sn |} is a sequence of complex simple functions satisfying the
conditions of Definition 8.7.4 corresponding to |f | . This is because |sn (ω)| → |f (ω)|
and Z Z
||sn | − |sm || dµ ≤ |sm − sn | dµ

with this last expression converging to 0 as m, n → ∞. Thus |f | ∈ L1 (Ω). Also, by

similar reasoning, {Re sn } and {Im sn } correspond to Re f and Im f respectively in the
manner described by Definition 8.7.4 showing that Re f and Im f are in L1 (Ω). Now

+ 1
(Re f ) = (|Re f | + Re f )
2
and
− 1
(Re f ) = (|Re f | − Re f )
2
+
so both of these functions are in L1 (Ω) . Similar formulas establish that (Im f ) and
−
(Im f ) are in L1 (Ω) .
The formula follows from the observation that
³ ´
+ − + −
f = (Re f ) − (Re f ) + i (Im f ) − (Im f )
R
and the fact shown first that f → f dµ is linear.
To verify the triangle inequality, let {sn } be complex simple functions for f as in
Definition 8.7.4. Then
¯Z ¯ ¯Z ¯ Z Z
¯ ¯ ¯ ¯
¯ f dµ¯ = lim ¯ sn dµ¯ ≤ lim |s | dµ = |f | dµ.
¯ ¯ n→∞ ¯ ¯ n→∞
n

Now the last assertion follows from the definition. There exists a sequence of simple
functions {sn } converging pointwise to f such that for all m, n large enough,
Z
ε
> |sn − sm | dµ
2
Fix m and let n → ∞. By Fatou’s lemma
Z Z
ε
ε > ≥ lim inf |sn − sm | dµ ≥ |f − sm | dµ.
2 n→∞

Let s = sm . This proves the theorem. ¥

Now here is an equivalent description of L1 (Ω) which is the version which will be
used more often than not.
8.8. APPROXIMATION WITH SIMPLE FUNCTIONS 191

Corollary 8.8.2 Let (Ω, S, µ) be a measure

R space and let f : Ω → C. Then f ∈
L1 (Ω) if and only if f is measurable and |f | dµ < ∞.

Proof: Suppose f ∈ L1 (Ω) . Then from Definition 8.7.4, it follows both real and
imaginary parts of f are measurable. Just take real and imaginary parts of sn and
observe the real and imaginary parts of f are limits of the real and imaginary parts of
sn respectively. By Theorem 8.8.1 this shows the only if part.
R The more interesting part is the if part. Suppose then that f is measurable and
|f | dµ < ∞. Suppose first that f has values in [0, ∞). It is necessary to obtain the
sequence of complex simple functions. By Theorem 7.7.12, there exists an increasing
sequence of nonnegative simple functions, {sn } such that sn (ω) ↑ f (ω). Then by the
monotone convergence theorem,
Z Z
lim (2f − (f − sn )) dµ = 2f dµ
n→∞

and so Z
lim (f − sn ) dµ = 0.
n→∞
R
Letting m be large enough, it follows (f − sm ) dµ < ε and so if n > m
Z Z
|sm − sn | dµ ≤ |f − sm | dµ < ε.

Therefore, f ∈ L1 (Ω) because {sn } is a suitable sequence.

The general case follows from considering positive and negative parts of real and
imaginary parts of f. These are each measurable and nonnegative and their integrals
are finite so each is in L1 (Ω) by what was just shown. Since
¡ ¢
f = Re f + − Re f − + i Im f + − Im f − ,

it follows f ∈ L1 (Ω). This proves the corollary. ¥

One of the major theorems in this theory is the dominated convergence theorem.
Before presenting it, here is a technical lemma about lim sup and lim inf .

Lemma 8.8.3 Let {an } be a sequence in [−∞, ∞] . Then limn→∞ an exists if and
only if
lim inf an = lim sup an
n→∞ n→∞

and in this case, the limit equals the common value of these two numbers.

Proof: Suppose first limn→∞ an = a ∈ R. Then, letting ε > 0 be given, an ∈

(a − ε, a + ε) for all n large enough, say n ≥ N. Therefore, both inf {ak : k ≥ n} and
sup {ak : k ≥ n} are contained in [a − ε, a + ε] whenever n ≥ N. It follows lim supn→∞ an
and lim inf n→∞ an are both in [a − ε, a + ε] , showing
¯ ¯
¯ ¯
¯lim inf an − lim sup an ¯ < 2ε.
¯ n→∞ n→∞
¯

Since ε is arbitrary, the two must be equal and they both must equal a. Next suppose
limn→∞ an = ∞. Then if l ∈ R, there exists N such that for n ≥ N,

l ≤ an

and therefore, for such n,

l ≤ inf {ak : k ≥ n} ≤ sup {ak : k ≥ n}

192 THE ABSTRACT LEBESGUE INTEGRAL

and this shows, since l is arbitrary that

lim inf an = lim sup an = ∞.

n→∞ n→∞

The case for −∞ is similar.

Conversely, suppose lim inf n→∞ an = lim supn→∞ an = a. Suppose first that a ∈ R.
Then, letting ε > 0 be given, there exists N such that if n ≥ N,

sup {ak : k ≥ n} − inf {ak : k ≥ n} < ε

therefore, if k, m > N, and ak > am ,

|ak − am | = ak − am ≤ sup {ak : k ≥ n} − inf {ak : k ≥ n} < ε

showing that {an } is a Cauchy sequence. Therefore, it converges to a ∈ R, and as in the

first part, the lim inf and lim sup both equal a. If lim inf n→∞ an = lim supn→∞ an = ∞,
then given l ∈ R, there exists N such that for n ≥ N,

inf an > l.
n>N

Therefore, limn→∞ an = ∞. 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 justifies the study of the
Lebesgue integral.

Theorem 8.9.1 (Dominated Convergence theorem) Let fn ∈ L1 (Ω) and suppose

f (ω) = lim fn (ω),

n→∞

and there exists a measurable function g, with values in [0, ∞],2 such that
Z
|fn (ω)| ≤ g(ω) and g(ω)dµ < ∞.

Then f ∈ L1 (Ω) and

Z ¯Z Z ¯
¯ ¯
0 = lim |fn − f | dµ = lim ¯ f dµ − fn dµ¯¯
¯
n→∞ n→∞

Proof: f is measurable by Theorem 7.7.3. Since |f | ≤ g, it follows that

f ∈ L1 (Ω) and |f − fn | ≤ 2g.

By Fatou’s lemma (Theorem 8.5.1),

Z Z
2gdµ ≤ lim inf 2g − |f − fn |dµ
n→∞
Z Z
= 2gdµ − lim sup |f − fn |dµ.
n→∞

2 Note that, since g is allowed to have the value ∞, it is not known that g ∈ L1 (Ω) .
8.9. THE DOMINATED CONVERGENCE THEOREM 193

R
Subtracting 2gdµ,
Z
0 ≤ − lim sup |f − fn |dµ.
n→∞

Hence
µZ ¶
0 ≥ lim sup |f − fn |dµ
n→∞
µZ ¶ ¯Z Z ¯
¯ ¯
≥ lim inf |f − fn |dµ ≥ ¯ f dµ − fn dµ¯¯ ≥ 0.
¯
n→∞

This proves the theorem by Lemma 8.8.3 because the lim sup and lim inf are equal. ¥

Corollary 8.9.2 Suppose fn ∈ L1 (Ω) and f (ω) = limn→∞ fn (ω) . Suppose

R also
Rthere exist measurable functions, gn , g withR values in [0,R∞] such that limn→∞ gn dµ =
gdµ, gn (ω) → g (ω) µ a.e. and both gn dµ and gdµ are finite. Also suppose
|fn (ω)| ≤ gn (ω) . Then
Z
lim |f − fn | dµ = 0.
n→∞

Proof: It is just like the above. This time g + gn − |f − fn | ≥ 0 and so by Fatou’s

lemma, Z Z
2gdµ − lim sup |f − fn | dµ =
n→∞

Z Z
lim inf (gn + g) dµ − lim sup |f − fn | dµ
n→∞ n→∞
Z Z
= lim inf ((gn + g) − |f − fn |) dµ ≥ 2gdµ
n→∞

R
and so − lim supn→∞ |f − fn | dµ ≥ 0. Thus
µZ ¶
0 ≥ lim sup |f − fn |dµ
n→∞
µZ ¶ ¯Z Z ¯
¯ ¯
≥ lim inf |f − fn |dµ ≥ ¯ f dµ − fn dµ¯¯ ≥ 0.
¯
n→∞

This proves the corollary. ¥

Definition 8.9.3 Let E be a measurable subset of Ω.

Z Z
f dµ ≡ f XE dµ.
E

If L1 (E) is written, the σ algebra is defined as

{E ∩ A : A ∈ F }

and the measure is µ restricted to this smaller σ algebra. Clearly, if f ∈ L1 (Ω), then

f XE ∈ L1 (E)

and if f ∈ L1 (E), then letting f˜ be the 0 extension of f off of E, it follows f˜ ∈ L1 (Ω).

194 THE ABSTRACT LEBESGUE INTEGRAL

8.10 Approximation With Cc (Y )

Let (Y, F, µ) be a measure space where Y is a closed subset of X a finite dimensional
normed vector space and F ⊇ B (Y ) , the Borel sets in Y . Suppose also that µ (K) < ∞
whenever K is a compact set in Y . By Theorem 7.4.6 it follows µ is regular. This
regularilty of µ implies an important approximation result valid for any f ∈ L1 (Y ) .
It turns out that in this situation, for all ε > 0, there exists g a continuous function
defined on Y with g equal to 0 outside some compact set and
Z
|f − g| dµ < ε.

Definition 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) 6= 0}
Also, if U is an open set, f ∈ Cc (U ) means f is continuous on U and spt (f ) ⊆ U .
Similarly f ∈ Ccm (U ) if f has m continuous derivatives and spt (f ) ⊆ U and f ∈ Cc∞ (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 finite 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, µ) is a measure space with F ⊇ B (Y ) and µ (K) < ∞, for every compact K, then
if µ (E) < ∞ where E ∈ F , there exists a sequence of functions in Cc (Y ) {fk } such
that Z
lim |fk (x) − XE (x)| dµ = 0.
k→∞ Y

Proof: For each x ∈ K, there exists rx such that

D (x, rx ) ≡ {y ∈ Y : ||x − y|| ≤ rx } ⊆ V.

m
Since K is compact, there are finitely many balls, {B (xk , rxk )}k=1 which cover K. Let
W = ∪m k=1 B (xk , rxk ) . Since there are only finitely many of these,

W = ∪m
k=1 D (x, rx )

and W is a compact subset of V because it is closed and bounded, being the finite union
of closed and bounded sets. Now define
¡ ¢
dist x, W C
f (x) ≡
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, µ is regular and so there
exist compact sets, {Kk } and open sets {Vk } such that Vk ⊇ Vk+1 , Kk ⊆ Kk+1 for all
k, and
Kk ⊆ E ⊆ Vk , µ (Vk \ Kk ) < 2−k .
8.10. APPROXIMATION WITH CC (Y ) 195

From the first part of the lemma, there exists a sequence {fk } such that

K k ≺ f k ≺ Vk .

Then fk (x) converges to XE (x) a.e. because if convergence fails to take place, then x
must be in infinitely many of the sets Vk \ Kk . Thus x is in

∩∞ ∞
m=1 ∪k=m Vk \ Kk

and for each p

¡ ¢
µ (∩∞ ∞
m=1 ∪k=m Vk \ Kk ) ≤ µ ∪∞k=p Vk \ Kk
X∞
≤ µ (Vk \ Kk )
k=p
X∞
1
< ≤ 2−(p−1)
2k
k=p

Now the functions are all bounded above by 1 and below by 0 and are equal to zero off
V1 , a set of finite measure so by the dominated convergence theorem,
Z
lim |XE (x) − fk (x)| dµ = 0,
k→∞

the dominating function being XE (x) + XV1 (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 finite dimensional normed

vector space. Let (Y, F, µ) be a measure space with F ⊇ B (Y ) and µ (K) < ∞, for
every compact K in Y . Let f ∈ L1 (Y ) and let ε > 0 be given. Then there exists
g ∈ Cc (Y ) such that
Z
|f (x) − g (x)| dµ < ε.
Y

Proof: By considering separately the positive and negative parts of the real and
imaginary parts of f it suffices to consider only the case where f ≥ 0. Then by Theorem
7.7.12 and the monotone convergence theorem, there exists a simple function,
p
X
s (x) ≡ cm XEm (x) , s (x) ≤ f (x)
m=1

such that Z
|f (x) − s (x)| dµ < ε/2.

∞
By Lemma 8.10.2, there exists {hmk }k=1 be functions in Cc (Y ) such that
Z
lim |XEm − fmk | dµ = 0.
k→∞ Y

Let
p
X
gk (x) ≡ cm hmk .
m=1
196 THE ABSTRACT LEBESGUE INTEGRAL

Thus for k large enough,

Z Z ¯¯ X p
¯
¯
¯ ¯
|s (x) − gk (x)| dµ = ¯ cm (XEm − hmk )¯ dµ
¯ ¯
m=1
Xp Z
≤ cm |XEm − hmk | dµ < ε/2
m=1

Thus for k this large,

Z Z Z
|f (x) − gk (x)| dµ ≤ |f (x) − s (x)| dµ + |s (x) − gk (x)| dµ

< ε/2 + ε/2 = ε.

This proves the theorem. ¥

People think of this theorem as saying that f is approximated by gk in L1 (Y ). It is
customary to consider functions in L1 (Y ) as vectors and the norm of such a vector is
given by Z
||f ||1 ≡ |f (x)| dµ.

You should verify this mostly satisfies 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 defined on R. Let µ be the Lebesgue Stieltjes measure
defined 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 defined on R, an integrator

function. There exists a function µ : P (R) → [0, ∞] which satisfies the following
properties.

1. If A ⊆ B, then 0 ≤ µ (A) ≤ µ (B) , µ (∅) = 0.

P∞
2. µ (∪∞
k=1 Ai ) ≤ i=1 µ (Ai )

3. µ ([a, b]) = F (b+) − F (a−) ,

4. µ ((a, b)) = F (b−) − F (a+)
5. µ ((a, b]) = F (b+) − F (a+)
6. µ ([a, b)) = F (b−) − F (a−) where

F (b+) ≡ lim F (t) , F (b−) ≡ lim F (t) .

t→b+ t→a−

There also exists a σ algebra S of measurable sets on which µ is a measure which

contains the open sets and also satisfies the regularity conditions,

µ (E) = sup {µ (K) : K is a closed and bounded set, K ⊆ E} (8.8)

µ (E) = inf {µ (V ) : V is an open set, V ⊇ E} (8.9)

whenever E is a set in S.
8.11. THE ONE DIMENSIONAL LEBESGUE INTEGRAL 197

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 defined on [a, b] . Then there exists a sequence of functions
{sn } of the form
mn
X ¡ n ¢
sn (x) ≡ f zk−1 X[zk−1
n n ) (x)
,zk
k=1

such that each zkn ∈

/ D and

sup {|sn (x) − f (x)| : x ∈ [a, b]} < 1/n.

∞
Proof: First note that D contains no intervals. To see this let D = {dk }k=1 . If D
has an interval of length 2ε, let Ik be an interval centered at dk which has length ε/2k .
Therefore, the sum of the lengths of these intervals is no more than
∞
X ε
= ε.
2k
k=1

Thus D cannot contain an interval of length 2ε. Since ε 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 δ > 0 such that if |x − y| ≤ 3δ, then

|f (x) − f (y)| < 1/n

Now let {x0 , · · · , xmn } be a partition of [a, b] such that |xi − xi−1 | < δ for each i. For
k = 1, 2, · · · , mn − 1, let zkn ∈
/ D and |zkn − xk | < δ. Then
¯ n ¯ ¯ ¯
¯zk − zk−1
n ¯
≤ |zkn − xk | + |xk − xk−1 | + ¯xk−1 − zk−1
n ¯
< 3δ.

It follows that for each x ∈ [a, b]

¯m ¯
¯X n
¡ n ¢ ¯
¯ ¯
¯ f zk−1 X[zk−1
n n ) (x) − f (x)¯ < 1/n.
,zk
¯ ¯
k=1

This proves the lemma. ¥

Proposition 8.11.3 Let f be a continuous function defined on R. Also let F be an

increasing function defined on R. Then whenever c, d are not points of discontinuity of
F and [a, b] ⊇ [c, d] ,
Z b Z
f X[c,d] dF = f dµ
a

Here µ is the Lebesgue Stieltjes measure defined above.

198 THE ABSTRACT LEBESGUE INTEGRAL

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 definition of the Darboux Stieltjes integral that
Z b
X[l,r) dF = F (r) − F (l) = F (r−) − F (l−)
a
Z
= µ ([l, r)) = X[l,r) dµ.

Now let {sn } be the sequence of step functions of Lemma 8.11.2 such that these step
functions converge uniformly to f on [c, d] Then
¯Z ¯ Z
¯ ¡ ¢ ¯ ¯ ¯ 1
¯ X[c,d] f − X[c,d] sn dµ¯¯ ≤ ¯X[c,d] (f − sn )¯ dµ ≤ µ ([c, d])
¯ n
and
¯Z ¯ Z
¯ b¡ ¢ ¯¯ b
1
¯
¯ X[c,d] f − X[c,d] sn dF ¯ ≤ X[c,d] |f − sn | dF < (F (b) − F (a)) .
¯ a ¯ a n

Also if sn is given by the formula of Lemma 8.11.2,

Z Z X
mn
¡ n ¢
X[c,d] sn dµ = f zk−1 X[zk−1
n n ) dµ
,zk
k=1
mn Z
X ¡ n ¢
= f zk−1 X[zk−1
n n ) dµ
,zk
k=1

mn
X ¡ n ¢ ¡ n ¢
= f zk−1 µ [zk−1 , zkn )
k=1
mn
X ¡ n ¢¡ ¡ n ¢¢
= f zk−1 F (zkn −) − F zk−1 −
k=1

mn
X ¡ n ¢¡ ¡ n ¢¢
= f zk−1 F (zkn ) − F zk−1
k=1
mn Z b
X Z b
¡ n ¢
= f zk−1 X[zk−1
n n ) dF =
,zk sn dF.
k=1 a a

Therefore, ¯Z ¯
¯ Z b ¯
¯ ¯
¯ X[c,d] f dµ − X[c,d] f dF ¯
¯ a ¯
¯Z Z ¯
¯ ¯
≤ ¯¯ X[c,d] f dµ − X[c,d] sn dµ¯¯
¯Z Z b ¯ ¯Z Z b ¯
¯ ¯ ¯ b ¯
¯ ¯ ¯ ¯
+ ¯ X[c,d] sn dµ − sn dF ¯ + ¯ sn dF − X[c,d] f dF ¯
¯ a ¯ ¯ a a ¯
8.12. EXERCISES 199

1 1
≤ µ ([c, d]) + (F (b) − F (a))
n n
and since n is arbitrary, this shows
Z Z b
f dµ − f dF = 0.
a

This proves the theorem. ¥

In particular, in the special case where F is continuous and f is continuous,
Z b Z
f dF = X[a,b] f dµ.
a

Thus, if F (x) = x so the Darboux Stieltjes integral is the usual integral from calculus,
Z b Z
f (t) dt = X[a,b] f dµ
a

where µ is the measure which comes from F (x) = x as described above. This measure
is often denoted by m. Thus when f is continuous
Z b Z
f (t) dt = X[a,b] f dm
a

and so there is no problem in writing

Z b
f (t) dt
a

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 Ω = N ={1, 2, · · ·}. Let F = P(N), the set of all subsets of N, and let µ(S) =
number of elements in S. Thus µ({1}) = 1 = µ({2}), µ({1, 2}) = 2, etc. Show
(Ω, F, µ) is a measure space. It is called counting measure. What functions are
measurable in this case? For a nonnegative function, f defined on N, show
Z ∞
X
f dµ = f (k)
N k=1

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 {fn } converging pointwise to a function f , such that
inequality is obtained in Fatou’s lemma.

3. If (Ω, F, µ) is a measure
R space and
R f ≥ 0 is measurable, show that if g (ω) = f (ω)
1
a.e. ω andRg ≥ 0, then
R gdµ = f dµ. Show that if f, g ∈ L (Ω) and g (ω) = f (ω)
a.e. then gdµ = f dµ.
200 THE ABSTRACT LEBESGUE INTEGRAL

4. An algebra A of subsets of Ω is a subset of the power set such that Ω is in the

∞
algebra and for A, B ∈ A, A \ B and A ∪ B are both in A. Let C ≡ {Ei }i=1 be
a countable collection of sets and let Ω1 ≡ ∪∞ i=1 Ei . Show there exists an algebra
of sets, A, such that A ⊇ C and A is countable. Note the difference between this
problem and Problem 5. Hint: Let C1 denote all finite unions of sets of C and
Ω1 . Thus C1 is countable. Now let B1 denote all complements with respect to Ω1
of sets of C1 . Let C2 denote all finite unions of sets of B1 ∪ C1 . Continue in this
way, obtaining an increasing sequence Cn , each of which is countable. Let

A ≡ ∪∞
i=1 Ci .

5. Let A ⊆ P (Ω) where P (Ω) denotes the set of all subsets of Ω. Let σ (A) denote
the intersection of all σ algebras which contain A, one of these being P (Ω). Show
σ (A) is also a σ algebra.
6. We say a function g mapping a normed vector space, Ω 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 σ algebra which contains the open sets.) Let
f : Ω → X and let g : X → Y where X is a normed vector space and Y equals
C, R, or (−∞, ∞] and F is a σ algebra of sets of Ω. Suppose f is measurable and
g is Borel measurable. Show g ◦ f is measurable.
7. Let (Ω, F, µ) be a measure space. Define µ : P(Ω) → [0, ∞] by

µ(A) = inf{µ(B) : B ⊇ A, B ∈ F}.

Show µ satisfies

µ(∅) = 0, if A ⊆ B, µ(A) ≤ µ(B),

X∞
µ(∪∞
i=1 Ai ) ≤ µ(Ai ), µ (A) = µ (A) if A ∈ F.
i=1

If µ satisfies these conditions, it is called an outer measure. This shows every

measure determines an outer measure on the power set.
8. Let {Ei } be a sequence of measurable sets with the property that
∞
X
µ(Ei ) < ∞.
i=1

Let S = {ω ∈ Ω such that ω ∈ Ei for infinitely many values of i}. Show µ(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 Ek . Remember the tail of a convergent series
is small.
9. ↑ Let {fn } , f be measurable functions with values in C. {fn } converges in measure
if
lim µ(x ∈ Ω : |f (x) − fn (x)| ≥ ε) = 0
n→∞

for each fixed ε > 0. Prove the theorem of F. Riesz. If fn converges to f in

measure, then there exists a subsequence {fnk } which converges to f a.e. Hint:
Choose n1 such that

µ(x : |f (x) − fn1 (x)| ≥ 1) < 1/2.

8.12. EXERCISES 201

Choose n2 > n1 such that

µ(x : |f (x) − fn2 (x)| ≥ 1/2) < 1/22,

n3 > n2 such that

µ(x : |f (x) − fn3 (x)| ≥ 1/3) < 1/23,
etc. Now consider what it means for fnk (x) to fail to converge to f (x). Then use
Problem 8.
10. Suppose (Ω, µ) is a finite measure space (µ (Ω) < ∞) and S ⊆ L1 (Ω). Then S is
said to be uniformly integrable if for every ε > 0 there exists δ > 0 such that if E
is a measurable set satisfying µ (E) < δ, then
Z
|f | dµ < ε
E

for all f ∈ S. Show S is uniformly integrable and bounded in L1 (Ω) if there

exists an increasing function h which satisfies
½Z ¾
h (t)
lim = ∞, sup h (|f |) dµ : f ∈ S < ∞.
t→∞ t Ω

S is bounded if there is some number, M such that

Z
|f | dµ ≤ M

for all f ∈ S.
11. Let (Ω, F, µ) be a measure space and suppose f, g : Ω → (−∞, ∞] are measurable.
Prove the sets
{ω : f (ω) < g(ω)} and {ω : f (ω) = g(ω)}
are measurable. Hint: The easy way to do this is to write

{ω : f (ω) < g(ω)} = ∪r∈Q [f < r] ∩ [g > r] .

Note that l (x, y) = x−y is not continuous on (−∞, ∞] so the obvious idea doesn’t
work.
12. Let {fn } be a sequence of real or complex valued measurable functions. Let

S = {ω : {fn (ω)} converges}.

Show S is measurable. Hint: You might try to exhibit the set where fn converges
in terms of countable unions and intersections using the definition of a Cauchy
sequence.
13. Suppose un (t) is a differentiable function for t ∈ (a, b) and suppose that for t ∈
(a, b),
|un (t)|, |u0n (t)| < Kn
P∞
where n=1 Kn < ∞. Show
∞
X ∞
X
( un (t))0 = u0n (t).
n=1 n=1

Hint: This is an exercise in the use of the dominated convergence theorem and
the mean value theorem.
202 THE ABSTRACT LEBESGUE INTEGRAL

∞
14. Let E be a countable subset of R. Show m(E) = 0. Hint: Let the set be {ei }i=1
and let ei be the center of an open interval of length ε/2i .
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 {fn } is a sequence of nonnegative measurable functions defined on a
measure space, (Ω, S, µ). Show that
Z X∞ ∞ Z
X
fk dµ = fk dµ.
k=1 k=1

Hint: Use the monotone convergence theorem along with the fact the integral is
linear.
R∞
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 α > 0, t →
2
e−at is in L1 (R). The gamma function is defined for x > 0 as
Z ∞
Γ (x) ≡ e−t tx−1 dt
0
−t x−1 1
Show t → e t is in L (R) for all x > 0. Also show that
Γ (x + 1) = xΓ (x) , Γ (1) = 1.
How does Γ (n) for n an integer compare with (n − 1)!?
18. This problem outlines a treatment of Stirling’s 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.
Z ∞
Γ (x + 1) = e−t tx dt
0

First change the variables letting t = x (1 + u) to get Γ (x + 1) =

Z ∞
¡ −u ¢x
e−x xx+1 e (1 + u) du
−1
q
2
Next make the change of variables u = s x to obtain Γ (x + 1) =
Z Ã Ã r !!x
√ ∞ √2 2
−x x+(1/2) −s
2e x √x e x 1+s ds
− 2
x

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
2
positive. Next show the limit of the integrand as x → ∞ is e−s . This isn’t too
bad if you take ln and then use L’Hospital’s 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.
Z ∞ Ã √ Ã r !!x Z ∞
−s x2 2 2
lim e 1+s ds = e−s ds
x→∞ −√ x x −∞
2
8.12. EXERCISES 203

Now Stirling’s formula is

Z ∞
Γ (x + 1) 2
lim √ = e−s ds
x→∞ 2e−x xx+(1/2) −∞

where this last improper integral equals a well defined 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 π but the necessary mathematical
machinery has not yet been presented. It can also be done through much more
difficult arguments in the context of functions of only one variable. See [34] for
these clever arguments.
19. To show you the power of Stirling’s formula, find whether the series
X∞
n!en
n=1
nn
converges. The ratio test falls flat but you can try it if you like. Now explain why,
if n is large enough
µZ ∞ ¶
1 −s2
√ −n n+(1/2) √
n! ≥ e ds 2e n ≡ c 2e−n nn+(1/2) .
2 −∞

Use this.
20. The Riemann integral is only defined for functions which are bounded which are
also defined on a bounded interval. If either of these two criteria are not satisfied,
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 find one.)
R∞
22. Note that 0 sinx x dx is a valid improper Riemann integral defined by
Z R
sin x
lim dx
R→∞ 0 x
but this function, sin x/x is not in L1 ([0, ∞)). Why?
23. Let f be a nonnegative strictly decreasing function defined 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
Z ∞ Z f (0)
f (t) dt = f −1 (y) dy.
0 0

Hint: Use the distribution function description.

24. Consider
£ the¤ following
£ ¤ nested sequence of compact sets, {Pn }. We let P1 = [0, 1],
P2 = 0, 31 ∪ 23 , 1 , etc. To go from Pn to Pn+1 , delete the open interval which
is the middle third of each closed interval in Pn . Let P = ∩∞n=1 Pn . Since P is the
intersection of nested nonempty compact sets, it follows from advanced calculus
that P 6= ∅. 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.
204 THE ABSTRACT LEBESGUE INTEGRAL

25. ↑ Consider the sequence of functions defined in the following way. Let f1 (x) = x on
[0, 1]. To get from fn to fn+1 , let fn+1 = fn on all intervals where fn is constant. If
fn is nonconstant on [a, b], let fn+1 (a) = fn (a), fn+1 (b) = fn (b), fn+1 is piecewise
linear and equal to 12 (fn (a) + fn (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 {fn } converges uniformly on [0, 1]. If f (x) = limn→∞ fn (x), show that
f (0) = 0, f (1) = 1, f is continuous, and f 0 (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
Z 1
f 0 (t) dt = 0 6= f (1) − f (0) .
0

Is this somehow contradictory to the fundamental theorem of calculus? Hint:

This isn’t too hard if you focus on getting a careful estimate on the difference
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 Definition 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 6= ∅}.
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 ⊆ ∪r∈Q A + r (a.)

A + r1 ∩ A + r2 = ∅ if r1 6= r2 , ri ∈ 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 Ij where Ij is open
and Ij ∩ Ik = ∅ if j 6= k. These intervals are the connected components of the
complement of the Cantor set. Show m(g(Ij )) = m(Ij ) so
∞
X ∞
X
m(g(∪∞
j=1 Ij )) = m(g(Ij )) = m(Ij ) = 1.
j=1 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. Define φ(x) = XA (g(x)). Thus φ(x) = 0
unless x ∈ P . Tell why φ is measurable. (Recall m(P ) = 0 and Lebesgue measure
is complete.) Now show that XA (y) = φ(g −1 (y)) for y ∈ [0, 2]. Tell why g −1 is
8.12. EXERCISES 205

continuous but φ ◦ g −1 is not measurable. (This is an example of measurable ◦

continuous 6= measurable.) Show there exist Lebesgue measurable sets which are
not Borel measurable. Hint: The function, φ is Lebesgue measurable. Now show
that Borel ◦ measurable = measurable.
28. If A is mbS 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.
206 THE ABSTRACT LEBESGUE INTEGRAL
The Lebesgue Integral For
Functions Of n Variables

9.1 π Systems
The approach to n dimensional Lebesgue measure will be based on a very elegant idea
due to Dynkin.

Definition 9.1.1 Let Ω be a set and let K be a collection of subsets of Ω. Then

K is called a π system if ∅, Ω ∈ K and whenever A, B ∈ K, it follows A ∩ B ∈ K.

For example, if Rn = Ω, an example of aQπ system would be the set of all open sets.
n
Another example would be sets of the form k=1 Ak where Ak is a Lebesgue measurable
set.
The following is the fundamental lemma which shows these π systems are useful.

Lemma 9.1.2 Let K be a π system of subsets of Ω, a set. Also let G be a collection

of subsets of Ω which satisfies the following three properties.

1. K ⊆ G

2. If A ∈ G, then AC ∈ G
∞
3. If {Ai }i=1 is a sequence of disjoint sets from G then ∪∞
i=1 Ai ∈ G.

Then G ⊇ σ (K) , where σ (K) is the smallest σ algebra which contains K.

Proof: First note that if

H ≡ {G : 1 - 3 all hold}

then ∩H yields a collection of sets which also satisfies 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 define

GA ≡ {B ∈ G : A ∩ B ∈ G} .

I want to show GA satisfies 1 - 3 because then it must equal G since G is the smallest
collection of subsets of Ω which satisfies 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 σ algebra
which will imply it contains σ (K). Now here are the details of the argument.

207
208 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Since K is given to be a π system, K ⊆ G A . Property 3 is obvious because if {Bi } is

a sequence of disjoint sets in GA , then

A ∩ ∪∞ ∞
i=1 Bi = ∪i=1 A ∩ Bi ∈ G

because A ∩ Bi ∈ G and the property 3 of G.

It remains to verify Property 2 so let B ∈ GA . I need to verify that B C ∈ GA . In
other words, I need to show that A ∩ B C ∈ G. However,
¡ ¢C
A ∩ B C = AC ∪ (A ∩ B) ∈ G

Here is why. Since B ∈ GA , A∩B ∈ G and since A ∈ K ⊆ G it follows AC ∈ G. It follows

the union of the disjoint sets, AC and (A ∩ B) is in G and then from 2 the complement
of their union is in G. Thus GA satisfies 1 - 3 and this implies since G is the smallest
such, that GA ⊇ G. However, GA is constructed as a subset of G and so G = GA . This
proves that for every B ∈ G and A ∈ K, A ∩ B ∈ G. Now pick B ∈ G and consider

GB ≡ {A ∈ G : A ∩ B ∈ G} .

I just proved K ⊆ GB . The other arguments are identical to show GB satisfies 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 σ algebra. To show this, all that is left is to verify G is closed
under countable unions because then it follows G is a σ algebra. Let {Ai } ⊆ G. Then
let A01 = A1 and

A0n+1 ≡ An+1 \ (∪ni=1 Ai )

¡ ¢
= An+1 ∩ ∩ni=1 AC i
¡ ¢
= ∩ni=1 An+1 ∩ AC i ∈G

because finite intersections of sets of G are in G. Since the A0i are disjoint, it follows

∪∞ ∞ 0
i=1 Ai = ∪i=1 Ai ∈ G

Therefore, G ⊇ σ (K) because it is a σ algebra which contains K and This proves the
lemma. ¥

9.2 n 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 σ algebra of
measurable sets be denoted by F. Recall this σ algebra contained the open sets. Also
from the construction given above,

m ([a, b]) = m ((a, b)) = b − a

Definition 9.2.1 Let f be a function of n variables and consider the symbol

Z Z
· · · f (x1 , · · ·, xn ) dxi1 · · · dxin . (9.1)

where (i1 , · · ·, in ) is a permutation of the integers {1, 2, · · ·, n} . The symbol means to

first do the Lebesgue integral
Z
f (x1 , · · ·, xn ) dxi1
9.2. N DIMENSIONAL LEBESGUE MEASURE AND INTEGRALS 209

yielding a function of the other n − 1 variables given above. Then you do

Z µZ ¶
f (x1 , · · ·, xn ) dxi1 dxi2

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

xi1 → f (x1 , · · ·, xn )

can be integrated. Either the function has values in [0, ∞] and is measurable or it is a
function in L1 . Then it is required
Z
xi2 → f (x1 , · · ·, xn ) dxi1

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 define n dimen-
sional Lebesgue measure.

9.2.2 n Dimensional Lebesgue Measure And Integrals

With the Lemma about π systems given above and the monotone convergence theorem,
it is possible to give a very elegant and fairly easy definition of the Lebesgue integral of
a function of n real variables. This is done in the following proposition.

Proposition 9.2.2 There exists a σ algebra of sets of Rn which contains the open
sets, F n and a measure mn defined on this σ algebra such that if f : Rn → [0, ∞)
is measurable with respect to F n then for any permutation (i1 , · · ·, in ) of {1, · · ·, n} it
follows Z Z Z
f dmn = ··· f (x1 , · · ·, xn ) dxi1 · · · dxin (9.2)
Rn
In particular, this implies that if Ai is Lebesgue measurable for each i = 1, · · ·, n then
à n ! n
Y Y
mn Ai = m (Ai ) .
i=1 i=1

Proof: Define a π system as

( n )
Y
K≡ Ai : Ai is Lebesgue measurable
i=1
n
Also let Rp ≡ [−p, p] , the n dimensional rectangle having sides [−p, p]. A set F ⊆ Rn
will be said to satisfy property P if for every p ∈ N and any two permutations of
{1, 2, · · ·, n}, (i1 , · · ·, in ) and (j1 , · · ·, jn ) the two iterated integrals
Z Z Z Z
· · · XRp ∩F dxi1 · · · dxin , · · · XRp ∩F dxj1 · · · dxjn

make sense and are equal. Now define G to be those subsets of Rn which have property
P.
Thus K ⊆ G because if (i1 , · · ·, in ) is any permutation of {1, 2, · · ·, n} and
n
Y
A= Ai ∈ K
i=1
210 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

then Z Z n
Y
··· XRp ∩A dxi1 · · · dxin = m ([−p, p] ∩ Ai ) .
i=1

Now suppose F ∈ G and let (i1 , · · ·, in ) and (j1 , · · ·, jn ) be two permutations. Then
¡ ¢
Rp = Rp ∩ F C ∪ (Rp ∩ F )

and so
Z Z Z Z
¡ ¢
··· XRp ∩F C dxi1 · · · dxin = ··· XRp − XRp ∩F dxi1 · · · dxin

Since Rp ∈ 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
Z Z
n
(2p) − · · · XRp ∩F dxi1 · · · dxin
Z Z
n
= (2p) − · · · XRp ∩F dxj1 · · · dxjn
Z Z
¡ ¢
= ··· XRp − XRp ∩F dxj1 · · · dxjn
Z Z
= · · · XRp ∩F C dxj1 · · · dxjn

which shows that if F ∈ G then so is F C .

∞
Next suppose {Fi }i=1 is a sequence of disjoint sets in G. Let F = ∪∞ i=1 Fi . I need to
show F ∈ G. Since the sets are disjoint,
Z Z Z Z X∞
· · · XRp ∩F dxi1 · · · dxin = · · · XRp ∩Fk dxi1 · · · dxin
k=1

Z Z N
X
= ··· lim XRp ∩Fk dxi1 · · · dxin
N →∞
k=1
Do
PNthe iterated integrals make sense? Note that the iterated integral makes sense for
k=1 XRp ∩Fk as the integrand because it is just a finite sum of functions for which the
iterated integral makes sense. Therefore,
∞
X
xi1 → XRp ∩Fk (x)
k=1

is measurable and by the monotone convergence theorem,

Z X
∞ Z X
N
XRp ∩Fk (x) dxi1 = lim XRp ∩Fk dxi1
N →∞
k=1 k=1

Now each of the functions,

Z X
N
xi2 → XRp ∩Fk dxi1
k=1

is measurable and so the limit of these functions,

Z X∞
XRp ∩Fk (x) dxi1
k=1
9.2. N DIMENSIONAL LEBESGUE MEASURE AND INTEGRALS 211

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,
Z Z Z Z X
∞
··· XRp ∩F dxi1 · · · dxin = ··· XRp ∩Fk dxi1 · · · dxin
k=1

Z Z N
X
= ··· lim XRp ∩Fk dxi1 · · · dxin
N →∞
k=1
Z Z N Z
X
= ··· lim XRp ∩Fk dxi1 · · · dxin
N →∞
k=1

Z N Z Z
X
= · · · lim XRp ∩Fk dxi1 · · · dxin · · ·
N →∞
k=1
N Z
X Z
= lim ··· XRp ∩Fk dxi1 · · · dxin
N →∞
k=1
N Z
X Z
= lim ··· XRp ∩Fk dxj1 · · · dxjn
N →∞
k=1

the last step holding because each Fk ∈ G. Then repeating the steps above in the
opposite order, this equals
Z Z X
∞ Z Z
··· XRp ∩Fk dxj1 · · · dxjn = ··· XRp ∩F dxj1 · · · dxjn
k=1

Thus F ∈ G. By Lemma 9.1.2 G ⊇ σ (K) . Q

n
Let F n = σ (K). Each set of the form k=1 Uk where Uk is an open set is in K.
n
Also every open set in R is a countable union of open sets of this form. This follows
from Lemma 7.7.7 on Page 174. Therefore, every open set is in F n .
For F ∈ F n define
Z Z
mn (F ) ≡ lim · · · XRp ∩F dxj1 · · · dxjn
p→∞

where (j1 , · · ·, jn ) is a permutation of {1, · · ·, n} . It doesn’t matter which one. It was

shown above they all give the same result. I need to verify mn is a measure. Let {Fk }
be a sequence of disjoint sets of F n .
Z Z X
∞
mn (∪∞
k=1 Fk ) = lim ··· XRp ∩Fk dxj1 · · · dxjn .
p→∞
k=1

Using the monotone convergence theorem repeatedly as in the first part of the argument,
this equals
X∞ Z Z ∞
X
lim · · · XRp ∩Fk dxj1 · · · dxjn ≡ mn (Fk ) .
p→∞
k=1 k=1
212 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Thus mn is a measure. Now letting Ak be a Lebesgue measurable set,

à n ! Z Z Y
n
Y
mn Ak = lim ··· X[−p,p]∩Ak (xk ) dxj1 · · · dxjn
p→∞
k=1 k=1
n
Y n
Y
= lim m ([−p, p] ∩ Ak ) = m (Ak ) .
p→∞
k=1 k=1

It only remains to prove 9.2.

It was shown above that for F ∈ F it follows
Z Z Z
XF dmn = lim · · · XRp ∩F dxj1 · · · dxjn
Rn p→∞

Applying the monotone convergence theorem repeatedly on the right, this yields that
the iterated integral makes sense and
Z Z Z
XF dmn = · · · XF dxj1 · · · dxjn
Rn

It follows 9.2 holds for every nonnegative simple function in place of f because these are
just linear combinations of functions, XF . Now taking an increasing sequence of non-
negative simple functions, {sk } which converges to a measurable nonnegative function
f
Z Z
f dmn = lim sk dmn
Rn k→∞ Rn
Z Z
= lim · · · sk dxj1 · · · dxjn
k→∞
Z Z
= · · · f dxj1 · · · dxjn

This proves the proposition. ¥

9.2.3 Fubini’s Theorem

Formula 9.2 is often called Fubini’s theorem. So is the following theorem. In general,
people tend to refer to theorems about the equality of iterated integrals as Fubini’s
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 Tonelli’s theorem.

Theorem 9.2.3 Let mn be defined in Proposition 9.2.2 on the σ algebra of

sets F n given there. Suppose f ∈ L1 (Rn ) . Then if (i1 , · · ·, in ) is any permutation
of {1, · · ·, n} ,
Z Z Z
f dmn = · · · f (x) dxi1 · · · dxin .
Rn

In particular, iterated integrals for any permutation of {1, · · ·, n} are all equal.

Proof: It suffices 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 ∈ L1 (Rn ) ,
Z
|f | dmn < ∞
Rn
9.2. N DIMENSIONAL LEBESGUE MEASURE AND INTEGRALS 213

and so both 12 (|f | + f ) and 12 (|f | − f ) are in L1 (Rn ) and are each nonnegative. Hence
from Proposition 9.2.2,
Z Z · ¸
1 1
f dmn = (|f | + f ) − (|f | − f ) dmn
Rn n 2 2
ZR Z
1 1
= (|f | + f ) dmn − (|f | − f ) dmn
Rn 2 Rn 2

Z Z
1
= ··· (|f (x)| + f (x)) dxi1 · · · dxin
2
Z Z
1
− ··· (|f (x)| − f (x)) dxi1 · · · dxin
2
Z Z
1 1
= ··· (|f (x)| + f (x)) − (|f (x)| − f (x)) dxi1 · · · dxin
2 2
Z Z
= ··· f (x) dxi1 · · · dxin

This proves the theorem. ¥

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 n and suppose for some
permutation, (i1 , · · ·, in )
Z Z
··· |f (x)| dxi1 · · · dxin < ∞

Then f ∈ L1 (Rn ) .

Proof: By Proposition 9.2.2,

Z Z Z
|f | dmn = · · · |f (x)| dxi1 · · · dxin < ∞
Rn

and so f is in L1 (Rn ) by Corollary 8.8.2. 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 (Rn ) be the Borel sets on Rn . There exists a measure mn
defined on B (Rn ) such that if f is a nonnegative Borel measurable function,
Z Z Z
f dmn = · · · f (x) dxi1 · · · dxin (9.3)
Rn

whenever (i1 , · · ·, in ) is a permutation of {1. · ··, n}. If f ∈ L1 (Rn ) 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 (i1 , · · ·, in ) a permutation of {1. · ··, n}
Z Z
· · · |f (x)| dxi1 · · · dxin < ∞,

then f ∈ L1 (Rn ). The measure mn is both inner and outer regular on the Borel sets.
That is, if E ∈ B (Rn ),

mn (E) = sup {mn (K) : K ⊆ E and K is compact}

214 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

mn (E) = inf {mn (V ) : V ⊇ E and V is open} .

Also if Ak is a Borel set in R then
n
Y
Ak
k=1

is a Borel set in Rn and à !

n
Y n
Y
mn Ak = m (Ak ) .
k=1 k=1

Proof: Most of it was shown earlier since B (Rn ) ⊆ F n . The two assertions about
regularity follow from observing that mn is finite 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 Ak
is open, there is nothing to show because the result isQan open set. Suppose then that
n
whenever A1 , · · ·, Am , m ≤ n are open, the product, k=1 Ak is a Borel set. Let
Qn K be
the open sets in R and let G be those Borel sets such that if Am ∈ G it follows k=1 Ak
is Borel. Then K is a π system and is contained in G. Now suppose F ∈ G. Then
Ãm−1 n
! Ãm−1 n
!
Y Y Y Y
C
Ak × F × Ak ∪ Ak × F × Ak
k=1 k=m+1 k=1 k=m+1
Ãm−1 n
!
Y Y
= Ak × R × Ak
k=1 k=m+1

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 {Fi } is a sequence of disjoint
elements of G
Ãm−1 n
! Ãm−1 n
!
Y Y Y Y
∞
Ak × ∪i Fi × Ak = ∪i=1 Ak × Fi × Ak
k=1 k=m+1 k=1 k=m+1

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 π systems
G ⊇ σ (K) = B (R) and this shows that Am can be any Borel set. Thus the assertion
about the product is true if only A1 , · · ·, Am−1 are open while the rest are Borel.
Continuing this way shows the assertion remains true for each Ai being Borel. Now the
final formula about the measure of a product follows from 9.3.
Z Z Z
X k=1 Ak dmn = · · · XQnk=1 Ak (x) dx1 · · · dxn
Q n

Rn

Z Z Y
n n
Y
= ··· XAk (xk ) dx1 · · · dxn = m (Ak ) .
k=1 k=1

This proves the theorem. ¥

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

Z 1Z 1
sin (y)
dydx
0 x y
Notice the limits. The iterated integral equals
Z
sin (y)
XA (x, y) dm2
R2 y
where
A = {(x, y) : x ≤ y where x ∈ [0, 1]}
Fubini’s theorem can be applied because the function (x, y) → sin (y) /y is continuous
except at y = 0 and can be redefined to be continuous there. The function is also
bounded so
sin (y)
(x, y) → XA (x, y)
y
1
¡ 2¢
clearly is in L R . Therefore,
Z Z Z
sin (y) sin (y)
XA (x, y) dm2 = XA (x, y) dxdy
R2 y y
Z 1Z y
sin (y)
= dxdy
0 0 y
Z 1
= sin (y) dy = 1 − cos (1)
0

9.3 Exercises
R 2 R 6−2z R 3−z ¡ ¢
1. Find 0 0 1 (3 − z) cos y 2 dy dx dz.
2x

R 1 R 18−3z R 6−z ¡ ¢
2. Find 0 0 1 (6 − z) exp y 2 dy dx dz.
3x

R 2 R 24−4z R 6−z ¡ ¢
3. Find 0 0 1 (6 − z) exp x2 dx dy dz.
4y

R 1 R 12−4z R 3−z sin x

4. Find 0 0 1
x dx dy dz.
4y

R 20 R 1 R 5−z R 25 R 5− 1 y R 5−z
5. Find 0 0 1 y sinx x dx dz dy + 20 0 5 1 y sin x
x dx dz dy. Hint: You might
5 5
try doing it in the order, dy dx dz
6. Explain why for each t > 0, x → e−tx is a function in L1 (R) and
Z ∞
1
e−tx dx = .
0 t
Thus Z Z Z
R R ∞
sin (t)
dt = sin (t) e−tx dxdt
0 t 0 0
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
Z ∞
sin (t) 1
dt = π
0 t 2
216 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

R∞ Rr
7. Explain why a f (t) dt ≡ limr→∞ a f (t) dt whenever f ∈ L1 (a, ∞) ; that is
f X[a,∞) ∈ L1 (R).
R1 R∞
8. B(p, q) = 0 xp−1 (1 − x)q−1 dx, Γ(p) = 0 e−t tp−1 dt for p, q > 0. The first of
these is called the beta function, while the second is the gamma function. Show
a.) Γ(p + 1) = pΓ(p); b.) Γ(p)Γ(q) = B(p, q)Γ(p + q).Explain why the gamma
function makes sense for any p > 0.
−1/2
9. Let f (y) = g (y) = |y| if y ∈ (−1, 0) ∪ (0, 1) and f (y) = g (y) = 0 if y ∈/
(−1,
R 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 {an } be an increasing sequence of numbers in (0, 1) which converges to 1.
Let
R gn be a nonnegative function which equals zero outside (an , an+1 ) such that
gn dx = 1. Now for (x, y) ∈ [0, 1) × [0, 1) define
∞
X
f (x, y) ≡ gn (y) (gn (x) − gn+1 (x)) .
k=1

Explain why this is actually a finite sum for each such (x, y) so there are no
convergence questions in the infinite sum. Explain why f is a continuous function
on [0, 1) × [0, 1). You can extend f to equal zero off [0, 1) × [0, 1) if you like. Show
the iterated integrals exist but are not equal. In fact, show
Z 1Z 1 Z 1Z 1
f (x, y) dydx = 1 6= 0 = f (x, y) dxdy.
0 0 0 0

Does this example contradict the Fubini theorem? Explain why or why not.

9.4 Lebesgue Measure On Rn

The σ algebra of Lebesgue measurable sets is larger than the above σ algebra of Borel
sets or of the earlier σ algebra which came from an application of the π system lemma. It
is convenient to use this larger σ 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.

Definition 9.4.1 The completion of (Rn , B (Rn ) , mn ) is the Lebesgue measure

n
space. Denote this by (R , Fn , mn ) .

Thus for each E ∈ Fn ,

mn (E) = inf {mn (F ) : F ⊇ E and F ∈ B (Rn )}

It follows that for each E ∈ Fn there exists F ∈ B (Rn ) such that F ⊇ E and

mn (E) = mn (F ) .

Theorem 9.4.2 mn is regular on Fn . In fact, if E ∈ Fn there exist sets in

B (Rn ) , F, G such that
F ⊆ E ⊆ G,
F is a countable union of compact sets, and G is a countable intersection of open sets
and
mn (G \ F ) = 0.
9.4. LEBESGUE MEASURE ON RN 217

Qn
If Ak is Lebesgue measurable then k=1 Ak ∈ Fn and
à n
! n
Y Y
mn Ak = m (Ak ) .
k=1 k=1

In addition to this, mn is translation invariant. This means that if E ∈ Fn and x ∈ Rn ,

then
mn (x+E) = mn (E) .
The expression x + E means {x + e : e ∈ E} .

Proof: mn is regular on B (Rn ) by Theorem 7.4.6 because it is finite on compact

sets. Then from Theorem 7.5.7, it follows mn is regular on Fn . This is because the
regularity of mn on the Borel sets and the definition of the completion of of a measure
given there implies the uniqueness part of Theorem 7.5.6 can be obtained to conclude
³ ´
(Rn , Fn , mn ) = Rn , B (Rn ), mn

Now for E ∈ Fn , let

Em ≡ (B (0, m) \ B (0, m − 1)) ∩ E.
∞
It follows from regularity there exists a sequence of open sets, {Vmk }k=1 such that

Vmk ⊇ Em

and
¡ ¢¡ ¢
mn (Vmk ) − mn (Em ) = mn (Vmk \ Em ) < 2−k 2−m .
Then
E = ∪∞ ∞
m=1 Em ⊆ ∪m=1 Vmk ≡ Vk

and
¡ ¢
∩kl=1 Vl \ E ⊆ ∪∞
m=1 Vmk \ E

so
∞ ∞
¡¡ ¢ ¢ X X
mn ∩kl=1 Vl \ E ≤ mn (Vmk \ E) < 2−k 2−m ≤ 2−k .
m=1 m=1

Let G ≡ ∩∞ k
k=1 ∩l=1 Vl . Then from the above, and passing to the limit, it follows

mn (G \ E) = 0.

To obtain F a countable union of compact sets

¡ contained
¢ in E such that mn (E \ F ) = 0,
let Kmk ⊆ Em such that mn (Em \ Kmk ) < 2−k (2−m ) .

mn (E \ ∪∞ ∞
m=1 Kmk ) ≤ mn (∪m=1 (Em \ Kmk )) < 2
−k

Let F = ∪∞ ∞
k=1 ∪m=1 Kmk . Then

mn (E \ F ) ≤ mn (E \ ∪∞
m=1 Kmk ) < 2
−k

for every k and so mn (E \ F ) = 0.

Consider the next assertion about the measure of a Cartesian product. By regularity
of m there exists Bk , Ck ∈ B (Rn ) such that Bk ⊇ Ak ⊇ Ck and m (Bk ) = m (Ak ) =
218 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

m (Ck ). In fact, you can have Bk equal a countable intersection of open sets and Ck a
countable union of compact sets. Then
n n
à n !
Y Y Y
m (Ak ) = m (Ck ) ≤ mn Ck
k=1 k=1 k=1
à n
! Ã n
!
Y Y
≤ mn Ak ≤ mn Bk
k=1 k=1
n
Y n
Y
= m (Bk ) = m (Ak ) .
k=1 k=1

It remains to prove the claim about the measure being translation invariant.
Let K denote all sets of the form
n
Y
Uk
k=1

where each Uk is an open set in R. Thus K is a π system.

n
Y n
Y
x+ Uk = (xk + Uk )
k=1 k=1

which is also a finite Cartesian product of finitely many open sets. Also,
à n
! Ã n !
Y Y
mn x + Uk = mn (xk + Uk )
k=1 k=1
n
Y
= m (xk + Uk )
k=1
n
à n
!
Y Y
= m (Uk ) = mn Uk
k=1 k=1

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 p ∈ N
n n
mn (x + E ∩ (−p, p) ) = mn (E ∩ (−p, p) )
n
and the set, x + E ∩ (−p, p) is a Borel set. Thus K ⊆ G. If E ∈ G then
¡ n¢ n n
x + E C ∩ (−p, p) ∪ (x + E ∩ (−p, p) ) = x + (−p, p)
n
which implies x + E C ∩ (−p, p) is a Borel set since it equals a difference of two Borel
sets. Now consider the following.
¡ n¢ n
mn x + E C ∩ (−p, p) + mn (E ∩ (−p, p) )
¡ n ¢ n
= mn x + E C ∩ (−p, p) + mn (x + E ∩ (−p, p) )
n n
= mn (x + (−p, p) ) = mn ((−p, p) )
¡ n¢ n
= mn E C ∩ (−p, p) + mn (E ∩ (−p, p) )
which shows ¡ n¢ ¡ n¢
mn x + E C ∩ (−p, p) = mn E C ∩ (−p, p)
9.5. MOLLIFIERS 219

showing that E C ∈ G.
If {Ek } is a sequence of disjoint sets of G,
n n
mn (x + ∪∞ ∞
k=1 Ek ∩ (−p, p) ) = mn (∪k=1 x + Ek ∩ (−p, p) )

n
Now the sets {x + Ek ∩ (−p, p) } are also disjoint and so the above equals
X n
X n
mn (x + Ek ∩ (−p, p) ) = mn (Ek ∩ (−p, p) )
k k
n
= mn (∪∞
k=1 Ek ∩ (−p, p) )

Thus G is also closed with respect to countable disjoint unions. It follows from the
lemma on π systems that G ⊇ σ (K) . But from Lemma 7.7.7 on Page 174, every open
set is a countable union of sets of K and so σ (K) contains the open sets. Therefore,
B (Rn ) ⊇ G ⊇ σ (K) ⊇ B (K) which shows G = B (Rn ).
I have just shown that for every E ∈ B (Rn ) , and any p ∈ N,
n n
mn (x + E ∩ (−p, p) ) = mn (E ∩ (−p, p) )

Taking the limit as p → ∞ yields

mn (x + E) = mn (E) .

This proves translation invariance on Borel sets.

Now suppose mn (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 mn (F ) = 0. Therefore from what was
just shown,
mn (x + S) ≤ mn (x + F ) = mn (F ) = mn (S) = 0
which shows that if mn (S) = 0 then so does mn (x + S) . Let F be any set of Fn . By
regularity, there exists E ⊇ F where E ∈ B (Rn ) and mn (E \ F ) = 0. Then

mn (F ) = mn (E) = mn (x + E) = mn (x + (E \ F ) ∪ F )

= mn (x + E \ F ) + mn (x + F ) = mn (x + F ) .
This proves the theorem. ¥

9.5 Mollifiers
From Theorem 8.10.3, every function in L1 (Rn ) can be approximated by one in Cc (Rn )
but even more incredible things can be said. In fact, you can approximate an arbitrary
function in L1 (Rn ) with one which is infinitely differentiable having compact support.
This is very important in partial differential 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)

 ·³ ´−1 ¸
 2
exp |x − z| − r2 if |x − z| < r,
ψ (x) =

0 if |x − z| ≥ r.

Then a little work shows ψ ∈ Cc∞ (U ). The following also is easily obtained.
220 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

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, ½ 2
e−1/x if x 6= 0
f (x) =
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 Cc∞ (U ) 6= ∅.

Proof: Pick z ∈ U and let r be small enough that B (z, 2r) ⊆ U . Then let ψ ∈
Cc∞ (B (z, 2r)) ⊆ Cc∞ (U ) be the function of the above example. ¥

Definition 9.5.3 Let U = {x ∈ Rn : |x| < 1}. A sequence {ψ m } ⊆ Cc∞ (U ) is

called a mollifier if
1
ψ m (x) ≥ 0, ψ m (x) = 0, if |x| ≥,
m
R
and ψ m (x) = 1. Sometimes it may be written as {ψ ε } where ψ ε satisfies the above
conditions except ψ ε (x) = 0 if |x| ≥ ε. In other words, ε takes the place of 1/m and in
everything that follows ε → 0 instead of m → ∞.
R
f (x, y)dmn (y) will mean x is fixed and the function y → f (x, y) is being inte-
grated. To make the notation more familiar, dx is written instead of dmn (x).

Example 9.5.4 Let

ψ ∈ Cc∞ (B(0, 1)) (B(0, 1) = {x : |x| < 1})

R
with ψ(x)R ≥ 0 and ψdm = 1. Let ψ m (x) = cm ψ(mx) where cm is chosen in such a
way that ψ m dm = 1.

Definition 9.5.5 A function, f , is said to be in L1loc (Rn ) if f is Lebesgue

measurable and if |f |XK ∈ L1 (Rn ) for every compact set, K. If f ∈ L1loc (Rn ), and
g ∈ Cc (Rn ), Z
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 definition. Then

Z Z
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
Z
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 L1loc (Rn ) is smoothed out by convolving with
a mollifier.
9.5. MOLLIFIERS 221

Lemma 9.5.7 Let f ∈ L1loc (Rn ), and g ∈ Cc∞ (Rn ). Then f ∗ g is an infinitely
differentiable function. Also, if {ψ m } is a mollifier and U is an open set and f ∈
C 0 (U ) ∩ L1loc (Rn ) , then at every x ∈ U,

lim f ∗ ψ m (x) = f (x) .

m→∞

If f ∈ C 1 (U ) ∩ L1loc (Rn ) and x ∈ U,

(f ∗ ψ m )xi (x) = fxi ∗ ψ m (x) .

Proof: Consider the difference quotient for calculating a partial derivative of f ∗ g.

Z
f ∗ g (x + tej ) − f ∗ g (x) g(x + tej − y) − g (x − y)
= f (y) dy.
t t
Using the fact that g ∈ Cc∞ (Rn ), the quotient,
g(x + tej − y) − g (x − y)
,
t
is uniformly bounded. To see this easily, use Theorem 6.4.2 on Page 124 to get the
existence of a constant, M depending on

max {||Dg (x)|| : x ∈ Rn }

such that
|g(x + tej − 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)| XK where K is a compact set
depending on the support of g. It follows from the dominated convergence theorem the
limit of the difference quotient above passes inside the integral as t → 0 and so
Z
∂ ∂
(f ∗ g) (x) = f (y) g (x − y) dy.
∂xj ∂xj
∂
Now letting ∂x j
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 mollifier. Let x ∈ U and let m be large
1
enough that B x, m ⊆ U. Then
Z
|f ∗ g (x) − f (x)| ≤ |f (x − y) − f (x)| ψ m (y) dy
B (0, m
1
)
By continuity of f at x, for all m sufficiently large, the above is dominated by
Z
ε ψ m (y) dy = ε
B (0, m
1
)
and this proves the claim.
Now consider the formula in the case where f ∈ C 1 (U ). Using Proposition 9.5.6,
f ∗ ψ m (x + hei ) − f ∗ ψ m (x)
=
h
ÃZ Z !
1
f (x+hei −y) ψ m (y) dy − f (x − y) ψ m (y) dy
h B (0, m
1
) B (0, m
1
)
222 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Z
(f (x+hei −y) − f (x − y))
= ψ m (y) dy
B( 1
0, m) h
¡ 1¢
Now letting m be small enough that B 0, m is contained in U along with the continuity
of the partial derivatives, it follows the difference quotients are uniformly bounded for
all h sufficiently small and so one can apply the dominated convergence theorem and
pass to the limit obtaining
Z
fxi (x − y) ψ m (y) dy ≡ fxi ∗ ψ m (x)
Rn

This proves the lemma. ¥

Theorem 9.5.8 Let K be a compact subset of an open set, U . Then there exists
a function, h ∈ Cc∞ (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 Kr = K + B(0, r). A picture of what is

happening follows.

K Kr U

1
Consider XKr ∗ ψ m where ψ m is a mollifier. Let m be so large that m < r. Then from
the¡ definition
¢ of what is meant by a convolution, and using that ψ m has support in
1
B 0, m , XKr ∗ ψ m = 1 on K and its support is in K + B (0, 3r), a bounded set. Now
using Lemma 9.5.7, XKr ∗ψ m is also infinitely differentiable. Therefore, let h = XKr ∗ψ m .
As
¡£ to ¤¢the existence of the open set W, let it equal the closed and bounded set
h−1 12 , 1 . This proves the theorem. ¥
The following is the remarkable theorem mentioned above. First, here is some no-
tation.

Definition 9.5.9 Let g be a function defined on a vector space. Then gy (x) ≡

g (x − y) .

Theorem 9.5.10 Cc∞ (Rn ) is dense in L1 (Rn ). Here the measure is Lebesgue
measure.

Proof: Let f ∈ L1 (Rn ) and let ε > 0 be given. Choose g ∈ Cc (Rn ) such that
Z
|g − f | dmn < ε/2
9.5. MOLLIFIERS 223

This can be done by using Theorem 8.10.3. Now let

Z Z
gm (x) = g ∗ ψ m (x) ≡ g (x − y) ψ m (y) dmn (y) = g (y) ψ m (x − y) dmn (y)

where {ψ m } is a mollifier. It follows from Lemma 9.5.7 gm ∈ Cc∞ (Rn ). It vanishes if

1
x∈
/ spt(g) + B(0, m ).
Z Z Z
|g − gm | dmn = |g(x) − g(x − y)ψ m (y)dmn (y)|dmn (x)
Z Z
≤ ( |g(x) − g(x − y)|ψ m (y)dmn (y))dmn (x)
Z Z
≤ |g(x) − g(x − y)|dmn (x)ψ m (y)dmn (y)
Z Z
ε
= |g − gy |dmn (x)ψ m (y)dmn (y) <
1
B(0, m ) 2

whenever m is large enough. This follows because since g has compact support, it is
uniformly continuous on Rn and so if η > 0 is given, then whenever |y| is sufficiently
small,
|g (x) − g (x − y)| < η
for all x. Thus, since g has compact support, if y is small enough, it follows
Z
|g − gy |dmn (x) < ε/2.

There is no measurability problem in the use of Fubini’s theorem because the function

(x, y) → |g(x) − g(x − y)|ψ m (y)

is continuous. Thus when m is large enough,

Z Z Z
ε ε
|f − gm | dmn ≤ |f − g| dmn + |g − gm | dmn < + = ε.
2 2
This proves the theorem. ¥
Another important application of Theorem 9.5.8 has to do with a partition of unity.

Definition 9.5.11 A collection of sets H is called locally finite if for every x,

there exists r > 0 such that B (x, r) has nonempty intersection with only finitely many
sets of H. Of course every finite collection of sets is locally finite. This is the case of
most interest in this book but the more general notion is interesting.

The thing about locally finite collection of sets is that the closure of their union
equals the union of their closures. This is clearly true of a finite collection.

Lemma 9.5.12 Let H be a locally finite collection of sets of a normed vector space
V . Then © ª
∪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 finitely many sets of H say these are
H1 , · · · , Hm . Then I claim p must be a limit point of one of these. If this is not so, there
would exist r0 such that 0 < r0 < r with B (p, r0 ) having empty intersection with each
224 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

of these Hi . 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 finite collection of sets and you note that Q = R 6= ∪ {x : x ∈ Q} . By
contrast, Z is a locally finite 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 φ ≺ V to symbolize φ ∈ Cc (V ) , φ has values in [0, 1] ,

and φ has compact support in V . I will write K ≺ φ ≺ V for K compact and V open to
symbolize φ is 1 on K and φ has values in [0, 1] with compact support contained in V .

Lemma 9.5.14 Let K be a closed set and let {Vi }∞

i=1 be a locally finite list of bounded
open sets whose union contains K. Then there exist functions, ψ i ∈ Cc∞ (Vi ) such that
for all x ∈ K,
∞
X
1= ψ i (x)
i=1

and the function f (x) given by

∞
X
f (x) = ψ i (x)
i=1

is in C ∞ (Rn ) .

Proof: Let K1 = K \ ∪∞ i=2 Vi . Thus K1 is compact because K1 ⊆ V1 . Let W1 be

an open set having compact closure which satisfies

K1 ⊆ W1 ⊆ W 1 ⊆ V1

Thus W1 , V2 , · · · , Vn covers K and W 1 ⊆ V1 . Suppose W1 , · · · , Wr have been defined

such that Wi ⊆ Vi for each i, and W1 , · · · , Wr , Vr+1 , · · · , Vn covers K. Then let
¡ ¢ ¡ r ¢
Kr+1 ≡ K \ ( ∪∞
i=r+2 Vi ∪ ∪j=1 Wj ).

It follows Kr+1 is compact because Kr+1 ⊆ Vr+1 . Let Wr+1 satisfy

Kr+1 ⊆ Wr+1 ⊆ W r+1 ⊆ Vr+1 , W r+1 is compact

∞
Continuing this way defines a sequence of open sets, {Wi }i=1 having compact closures
with the property
Wi ⊆ Vi , K ⊆ ∪∞i=1 Wi .
∞ ∞
Note {Wi }i=1 is locally finite because the original list, {Vi }i=1 was locally finite. Now
let Ui be open sets which satisfy

W i ⊆ Ui ⊆ U i ⊆ Vi , U i is compact.
∞
Similarly, {Ui }i=1 is locally finite.

Wi Ui Vi
9.6. THE VITALI COVERING THEOREM 225

∞ ∞
Since the sets, {Wi }i=1 are locally finite, it follows ∪∞
i=1 Wi = ∪i=1 Wi and so it is
possible to define φi and γ, infinitely differentiable functions having compact support
such that
U i ≺ φi ≺ Vi , ∪∞ ∞
i=1 W i ≺ γ ≺ ∪i=1 Ui .

Now define ½ P∞ P∞
γ(x)φi (x)/ j=1 φj (x) if j=1 φj (x) 6= 0,
ψ i (x) = P∞
0 if j=1 φj (x) = 0.
P∞
If x is such that j=1 φj (x) = 0, then x ∈ / ∪∞i=1 Ui because φi equals one on Ui .
Consequently γ (y) = 0 for all y near x thanks to the fact that ∪∞ i=1 Ui is closed and
so
P∞ ψ i (y) = 0 for all y near x. Hence ψ i is infinitely differentiable at such x. If
j=1 φj (x) 6= 0, this situation persists near x because each φj is continuous and so ψ i
is infinitely differentiable at such
P∞points also. Therefore ψ i is infinitely differentiable. If
x ∈ K, then γ (x) = 1 and so j=1 ψ j (x) = 1. Clearly 0 ≤ ψ i (x) ≤ 1 and spt(ψ j ) ⊆ Vj .
This proves the theorem. ¥
The functions, {ψ 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 Vi for some Vi there exists a partition

of unity such that ψ i (x) = 1 for all x ∈ H in addition to the conclusion of Lemma
9.5.14.

Proof: Keep Vi the same but replace Vj with V fj ≡ Vj \ H. Now in the proof above,
applied to this modified collection of open sets, if j 6= i, φj (x) = 0 whenever x ∈ H.
Therefore, ψ 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 Rn with
open balls. The balls can be defined in terms of any norm for Rn . For example, the
norm could be
||x|| ≡ max {|xk | : k = 1, · · ·, n}
or the usual norm sX
2
|x| = |xk |
k

or any other. The proof given here is from Basic Analysis [27]. It first 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 B1 , B2 ∈ G then B1 ∩ B2 = ∅, (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.
226 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

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 ≡ {Bi }i=1 . Now let Bn1 be the first 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 {Bn1 } . Otherwise, let Bn2 be the
next ball having radius larger than k which is disjoint from Bn1 . Continue this way
∞
obtaining {Bni }i=1 , a finite or infinite sequence of disjoint balls having radius larger
than k. Then let G ≡ {Bni }. 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} satisfies 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 e 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
e : B ∈ G}.
A ⊆ ∪{B

Proof: Without loss of generality assume F is countable. This is because there is

a countable subset of F, F 0 such that ∪F 0 = 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 define a
subset of F consisting of a single set from F containing each set of G. Then this is F 0 .
The union of these sets equals A . Then consider F 0 instead of F. Therefore, assume
at the outset F is countable.
By Lemma 9.6.1, there exists G1 ⊆ F which satisfies 9.4, 9.5, and 9.6 with k = 2M 3 .
Suppose G1 , · · ·, Gm−1 have been chosen for m ≥ 2. Let
union of the balls in these Gj
z }| {
n
Fm = {B ∈ F : B ⊆ R \ ∪{G1 ∪ · · · ∪ Gm−1 } }

and using Lemma 9.6.1, let Gm be a maximal collection

¡ ¢of disjoint balls from Fm with
m
the property that each ball has radius larger than 32 M. Let G ≡ ∪∞ k=1 Gk . Let
x ∈ B (p, r) ∈ F. Choose m such that
µ ¶m µ ¶m−1
2 2
M <r≤ M
3 3

Then B (p, r) must have nonempty intersection with some ball from G1 ∪···∪Gm because
if it didn’t, then Gm would fail to be maximal. Denote by B (p0 , r0 ) a ball in G1 ∪···∪Gm
which has nonempty intersection with B (p, r) . Thus
µ ¶m
2
r0 > M.
3

Consider the picture, in which w ∈ B (p0 , r0 ) ∩ B (p, r) .

9.7. VITALI COVERINGS 227

r0 x
¾ w p
p0 r
?
Then
<r0
z }| {
|x − p0 | ≤ |x − p| + |p − w| + |w − p0 |
< 32 r0
z }| {
µ ¶m−1
2
< r + r + r0 ≤ 2 M + r0
3
µ ¶
3
< 2 r0 + r0 = 4r0 .
2
This proves the lemma since it shows B (p, r) ⊆ B (p0 , 4r0 ). ¥
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 Rn .

Definition 9.6.3 b
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
b : B ∈ G}.
A ⊆ ∪{B

Proof: For
¡ B¢ one of these balls, say B (x, r) ⊇ B ⊇ B (x, r), denote by B1 , the
open ball B x, 5r
4 . Let F1 ≡ {B1 : B ∈ F} and let A1 denote the union of the balls in
F1 . Apply Lemma 9.6.2 to F1 to obtain
f1 : B1 ∈ G1 }
A1 ⊆ ∪{B

where G1 consists of disjoint balls from F1 . Now let G ≡ {B ∈ F : B1 ∈ G1 }. Thus G

consists of disjoint balls from F because they are contained in the disjoint open balls,
G1 . Then
A ⊆ A1 ⊆ ∪{B f1 : B1 ∈ G1 } = ∪{B
b : B ∈ G}
¡ 5r ¢
because for B1 = B x, 4 , it follows B f1 = B (x, 5r) = B.
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 stuff. You keep trying to fill
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.
228 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Definition 9.7.1 Let F be a collection of balls that cover a set, E, which have
the property that if x ∈ E and ε > 0, then there exists B ∈ F, diameter of B < ε and
x ∈ B. Such a collection covers E in the sense of Vitali.

In the following covering theorem, mn denotes the outer measure determined by n

dimensional Lebesgue measure. Thus, letting F denote the Lebesgue measurable sets,
(∞ )
X
mn (S) ≡ inf mn (Ek ) : S ⊆ ∪k Ek , Ek ∈ F
k=1

Recall that from this definition, if S ⊆ Rn there exists E1 ⊇ S such that mn (E1 ) =
mn (S). To see this, note that it suffices to assume in the above definition of mn that
the Ek are also disjoint. If not, replace with the sequence given by

F1 = E1 , F2 ≡ E2 \ F1 , · · ·, Fm ≡ Em \ Fm−1 ,

etc. Then for each l > mn (S) , there exists {Ek } such that
X X
l> mn (Ek ) ≥ mn (Fk ) = mn (∪k Ek ) ≥ mn (S) .
k k

If mn (S) = ∞, let E1 = Rn . Otherwise, there exists Gk ∈ F such that

mn (S) ≤ mn (Gk ) ≤ mn (S) + 1/k.

then let E1 = ∩k Gk .
Note this implies that if mn (S) = 0 then S must be in F because of completeness
of Lebesgue measure.

Theorem 9.7.2 Let E ⊆ Rn and suppose 0 < mn (E) < ∞ where mn is the
outer measure determined by mn , n 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, {Bj }∞j=1 , such that
mn (E \ ∪∞j=1 B j ) = 0.

Proof: From the definition of outer measure there exists a Lebesgue measurable set,
E1 ⊇ E such that mn (E1 ) = mn (E). Now by outer regularity of Lebesgue measure,
there exists U , an open set which satisfies

mn (E1 ) > (1 − 10−n )mn (U ), U ⊇ E1 .

E1

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, {Bi }i=1 ⊆ F such that

E ⊆ ∪∞ b
j=1 Bj , Bj ⊆ U.
9.7. VITALI COVERINGS 229

Therefore,
³ ´ X ³ ´
mn (E1 ) = mn (E) ≤ mn ∪∞ b
j=1 Bj ≤
bj
mn B
j
X ¡ ¢
= 5n mn (Bj ) = 5n mn ∪∞
j=1 Bj
j

Then
mn (E1 ) > (1 − 10−n )mn (U )
≥ (1 − 10−n )[mn (E1 \ ∪∞ ∞
j=1 Bj ) + mn (∪j=1 Bj )]

=mn (E1 )
z }| {
−n
≥ (1 − 10 )[mn (E1 \ ∪∞
j=1 Bj ) +5 −n
mn (E) ].
and so ¡ ¡ ¢ ¢
1 − 1 − 10−n 5−n mn (E1 ) ≥ (1 − 10−n )mn (E1 \ ∪∞
j=1 Bj )

which implies
(1 − (1 − 10−n ) 5−n )
mn (E1 \ ∪∞
j=1 Bj ) ≤ mn (E1 )
(1 − 10−n )
Now a short computation shows

(1 − (1 − 10−n ) 5−n )
0< <1
(1 − 10−n )
Hence, denoting by θn a number such that

(1 − (1 − 10−n ) 5−n )
< θn < 1,
(1 − 10−n )
¡ ¢
mn E \ ∪ ∞ ∞
j=1 Bj ≤ mn (E1 \ ∪j=1 Bj ) < θ n mn (E1 ) = θ n mn (E)

Now using Theorem 7.3.2 on Page 155 there exists N1 large enough that

θn mn (E) ≥ mn (E1 \ ∪N N1
j=1 Bj ) ≥ mn (E \ ∪j=1 Bj )
1
(9.7)

Let F1 = {B ∈ F : Bj ∩ B = ∅, j = 1, · · ·, N1 }. If E \ ∪N
j=1 Bj = ∅, then F1 = ∅ and
1

³ ´
mn E \ ∪N j=1 j = 0
1
B

Therefore, in this case let Bk = ∅ for all k > N1 . Consider the case where

E \ ∪N
j=1 Bj 6= ∅.
1

In this case, since the balls are closed and F is a Vitali cover, F1 6= ∅ and covers
E \ ∪N N1
j=1 Bj in the sense of Vitali. Repeat the same argument, letting E \ ∪j=1 Bj play
1

the role of E. (You pick a different E1 whose measure equals the outer measure of
N1
E \ ∪j=1 Bj and proceed as before.) Then choosing Bj for j = N1 + 1, · · ·, N2 as in the
above argument,
θn mn (E \ ∪N N2
j=1 Bj ) ≥ mn (E \ ∪j=1 Bj )
1

and so from 9.7,

θ2n mn (E) ≥ mn (E \ ∪N
j=1 Bj ).
2

Continuing this way ³ ´

θkn mn (E) ≥ mn E \ ∪N
j=1 j .
k
B
230 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

If it is ever the case that E \ ∪N

j=1 Bj = ∅, then as in the above argument,
k

³ ´
mn E \ ∪ N k
j=1 Bj = 0.

Otherwise, the process continues and

¡ ¢ ³ ´
Nk k
mn E \ ∪∞
j=1 Bj ≤ mn E \ ∪j=1 Bj ≤ θ n mn (E)

for every k ∈ N. Therefore, the conclusion holds in this case also because θn < 1. This
proves the theorem. ¥
There is an obvious corollary which removes the assumption that 0 < mn (E).

Corollary 9.7.3 Let E ⊆ Rn and suppose mn (E) < ∞ where mn is the outer mea-
sure determined by mn , n 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, {Bj }∞ ∞
j=1 , such that mn (E \ ∪j=1 Bj ) = 0.

Proof: If 0 = mn (E) you simply pick any ball from F for your collection of disjoint
balls. ¥
It is also not hard to remove the assumption that mn (E) < ∞.

Corollary 9.7.4 Let E ⊆ Rn 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, {Bj }∞ ∞
j=1 , such that mn (E \ ∪j=1 Bj ) = 0.

n
Proof: Let Rm ≡ (−m, m) be the open rectangle having sides of length 2m which
is centered at 0 and let R0 = ∅. Let Hm ≡ Rm \ Rm . Since both Rm and Rm have the
n
same measure, (2m) , it follows mn (Hm ) = 0. Now for all k ∈ N, Rk ⊆ Rk ⊆ Rk+1 .
Consider the disjoint open sets, Uk ≡ Rk+1 \ Rk . Thus Rn = ∪∞ k=0 Uk ∪ N where N is
a set of measure zero equal to the union of the Hk . Let Fk denote those balls of F
which are contained in Uk and let Ek© ≡ U
ªk∞∩ E. Then from Theorem 9.7.2, there exists
a sequence of disjoint balls, Dk ≡ Bik i=1 of Fk such that mn (Ek \ ∪∞ k
j=1 j = 0.
B )
∞
Letting {Bi }i=1 be an enumeration of all the balls of ∪k Dk , it follows that
∞
X
mn (E \ ∪∞
j=1 Bj ) ≤ mn (N ) + mn (Ek \ ∪∞ k
j=1 Bj ) = 0.
k=1

¥
Also, you don’t have to assume the balls are closed.

Corollary 9.7.5 Let E ⊆ Rn 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, {Bj }∞ ∞
j=1 , such that mn (E \ ∪j=1 Bj ) = 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 mn E \ ∪∞ i=1 Bi = 0. Now boundaries of the balls, Bi have measure
zero and so {Bi } is a sequence of disjoint open balls satisfying mn (E \ ∪∞ i=1 Bi ) = 0.
The reason for this is that
¡ ¢
(E \ ∪∞ ∞ ∞ ∞ ∞
i=1 Bi ) \ E \ ∪i=1 Bi ⊆ ∪i=1 Bi \ ∪i=1 Bi ⊆ ∪i=1 Bi \ Bi ,

a set of measure zero. Therefore,

¡ ¢ ¡ ∞ ¢
E \ ∪∞ ∞
i=1 Bi ⊆ E \ ∪i=1 Bi ∪ ∪i=1 Bi \ Bi
9.8. CHANGE OF VARIABLES FOR LINEAR MAPS 231

and so
¡ ¢ ¡ ∞ ¢
mn (E \ ∪∞
i=1 Bi ) ≤ mn E \ ∪∞
i=1 Bi + mn ∪i=1 Bi \ Bi
¡ ¢
= mn E \ ∪∞
i=1 Bi = 0.

¥
This implies you can fill up an open set with balls which cover the open set in the
sense of Vitali.

Corollary 9.7.6 Let U ⊆ Rn 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, {Bj }∞
j=1 , such
that mn (U \ ∪∞
j=1 Bj ) = 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 Rn and that h : U → Rn satisfies

Dh (x) exists for all x ∈ U, (9.8)

Note that if
h (x) = Lx
where L ∈ L (Rn , Rn ) , 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 Rn given by

||x|| = max {|xk | : k = 1, 2, · · ·, n} .

Thus B (x, r) is the open box,

n
Y
(xk − r, xk + r)
k=1

n
and so mn (B (x, r)) = (2r) .

Lemma 9.8.1 Let h satisfy 9.8. If T ⊆ U and mn (T ) = 0, then mn (h (T )) = 0.

Proof: Let
Tk ≡ {x ∈ T : ||Dh (x)|| < k}
and let ε > 0 be given. Now by outer regularity, there exists an open set, V , containing
Tk which is contained in U such that mn (V ) < ε. Let x ∈ Tk . Then by differentiability,

h (x + v) = h (x) + Dh (x) v + o (v)

and so there exist arbitrarily small rx < 1 such that B (x,5rx ) ⊆ V and whenever
||v|| ≤ rx , ||o (v)|| < k ||v|| . Thus

h (B (x, rx )) ⊆ B (h (x) , 2krx ) .

232 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

From the Vitali covering theorem there exists ancountable

o∞ disjoint sequence of these
∞ ∞
balls, {B (xi , ri )}i=1 such that {B (xi , 5ri )}i=1 = Bci covers Tk Then letting mn
i=1
denote the outer measure determined by mn ,
³ ³ ´´
mn (h (Tk )) ≤ mn h ∪∞ bi
B
i=1

∞
X ³ ³ ´´ X∞
≤ bi
mn h B ≤ mn (B (h (xi ) , 2krxi ))
i=1 i=1

∞
X ∞
X
n
= mn (B (xi , 2krxi )) = (2k) mn (B (xi , rxi ))
i=1 i=1
n n
≤ (2k) mn (V ) ≤ (2k) ε.

Since ε > 0 is arbitrary, this shows mn (h (Tk )) = 0. Now

mn (h (T )) = lim mn (h (Tk )) = 0.
k→∞

This proves the lemma. ¥

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 Kk such that
F ⊆ S, mn (S \ F ) = 0.
Then since h is continuous

h (F ) = ∪k h (Kk ) ∈ B (Rn )

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 ) ∈ Fn

because it is the union of two sets which are in Fn . This proves the lemma. ¥
In particular, this proves the following corollary.

Corollary 9.8.3 Suppose A ∈ L (Rn , Rn ). 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 defining balls will be the usual norm,
à n
!1/2
X 2
|x| = |xk | .
k=1

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 mn (RV ) =
mn (V ) .
9.8. CHANGE OF VARIABLES FOR LINEAR MAPS 233

Proof: First assume V is a bounded open set. By Corollary 9.7.6 there is a disjoint
sequence of closed balls, {Bi } such that U = ∪∞ i=1 Bi ∪ N where mn (N ) = 0. Denote by
x
P∞i the center of B i and let ri be the radius of Bi . Then by Lemma 9.8.1 mn (RV ) =
i=1 mn (RBi ) . Now by invariance of translation of Lebesgue measure, this equals

∞
X ∞
X
mn (RBi − Rxi ) = mn (RB (0, ri )) .
i=1 i=1

Since R is unitary, it preserves all distances and so RB (0, ri ) = B (0, ri ) and therefore,
∞
X ∞
X
mn (RV ) = mn (B (0, ri )) = mn (Bi ) = mn (V ) .
i=1 i=1

This proves the lemma in the case that V is bounded. Suppose now that V is just
an open set. Let Vk = V ∩ B (0, k) . Then mn (RVk ) = mn (Vk ) . 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 Rn and let R be unitary. Then
mn (RE) = mn (E) .

Proof: Let K be the open sets. Thus K is a π system. Let G denote those Borel
sets F such that for each p ∈ N,
n n
mn (R (F ∩ (−p, p) )) = mn (F ∩ (−p, 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 π systems lemma,

G ⊇ σ (K) = B (Rn ) ⊇ G

and this proves the lemma whenever E ∈ B (Rn ). If E is only in Fn , it follows from
Theorem 9.4.2
E =F ∪N
where mn (N ) = 0 and F is a countable union of compact sets. Thus by Lemma 9.8.1

mn (RE) = mn (RF ) + mn (RN ) = mn (RF ) = mn (F ) = mn (E) .

This proves the theorem. ¥

Lemma 9.8.6 Let D ∈ L (Rn , Rn ) be of the form

X
D= dj ej ej
j

where dj ≥ 0 and {ej } is the usual orthonormal basis of Rn . Then for all E ∈ Fn

mn (DE) = |det (D)| mn (E) .

Proof: Let K consist of open sets of the form

n
( n )
Y X
(ak , bk ) ≡ xk ek such that xk ∈ (ak , bk )
k=1 k=1
234 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Hence
à n
! ( n
)
Y X
D (ak , bk ) = dk xk ek such that xk ∈ (ak , bk )
k=1 k=1
n
Y
= (dk ak , dk bk ) .
k=1

It follows
à à n
!! Ã n
!Ã n
!
Y Y Y
mn D (ak , bk ) = dk (bk − ak )
k=1 k=1 k=1
à n
!
Y
= |det (D)| mn (ak , bk ) .
k=1

Now let G consist of Borel sets F with the property that

n n
mn (D (F ∩ (−p, p) )) = |det (D)| mn (F ∩ (−p, p) ) .

Thus K ⊆ G.
Suppose now that F ∈ G and first assume D is one to one. Then
¡ ¡ n ¢¢ n n
mn D F C ∩ (−p, p) + mn (D (F ∩ (−p, p) )) = mn (D (−p, p) )

and so
¡ ¡ n ¢¢ n n
mn D F C ∩ (−p, p) + |det (D)| mn (F ∩ (−p, p) ) = |det (D)| mn ((−p, p) )

which shows
¡ ¡ n ¢¢ n n
mn D F C ∩ (−p, p) = |det (D)| [mn ((−p, p) ) − mn (F ∩ (−p, p) )]
¡ n¢
= |det (D)| mn F C ∩ (−p, p)

In case D is not one to one, it follows some dj = 0 and so |det (D)| = 0 and
n
Y
¡ ¡ n ¢¢ n
0 ≤ mn D F C ∩ (−p, p) ≤ mn (D (−p, p) ) = (di p + di p) = 0
i=1
¡ n¢
= |det (D)| mn F C ∩ (−p, p)

so F C ∈ G.
If {Fk } is a sequence of disjoint sets of G and D is one to one
∞
X
n n
mn (D (∪∞
k=1 Fk ∩ (−p, p) )) = mn (D (Fk ∩ (−p, p) ))
k=1
∞
X n
= |det (D)| mn (Fk ∩ (−p, p) )
k=1
n
= |det (D)| mn (∪k Fk ∩ (−p, 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
n
mn (D (−p, p) ) = 0.
9.8. CHANGE OF VARIABLES FOR LINEAR MAPS 235

Thus G is closed with respect to complements and countable disjoint unions. Hence it
contains σ (K) , the Borel sets. But also G ⊆ B (Rn ) and so G equals B (Rn ) . Letting
p → ∞ yields the conclusion of the lemma in case E ∈ B (Rn ).
Now for E ∈ Fn 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,

mn (D (E)) = mn (DF ∪ DN ) ≤ mn (DF ) + mn (DN )

= |det (D)| mn (F ) = |det (D)| mn (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)| mn (E) = |det (D)| mn (G) = mn (DG)

= mn (DE ∪ DS) ≤ mn (DE) + mn (DS)
= mn (DE)

This proves the theorem. ¥

The main result follows.

Theorem 9.8.7 Let E ∈ Fn and let A ∈ L (Rn , Rn ). Then

mn (AV ) = |det (A)| mn (V ) .

Proof: Let RU be the right polar decomposition (Theorem 3.9.3 on Page 66) of A.
Thus R is unitary and X
U= dk wk wk
k

where each dk ≥ 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,

mn (AE) = mn (RU E) = mn (U E) .

Let X
Q= wj ej
j

and so by Lemma 3.8.21 on Page 63,

X
Q∗ = ek wk .
k

Thus Q and Q∗ are both unitary and a simple computation shows

X
U =Q di ei ei Q∗ ≡ QDQ∗ .
i
236 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Do both sides to wk and observe both sides give dk wk . 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

mn (AE) = mn (QDQ∗ E) = mn (DQ∗ E)

= |det (D)| mn (Q∗ E) = |det (A)| mn (E) .

This proves the theorem. ¥

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 defined in terms of the norm

||x|| ≡ max {|xk | : k = 1, · · ·, n}

Lemma 9.9.1 Let U and V be bounded open sets in Rn and let h, h−1 be C 1 func-
tions such that h (U ) = V . Also let f ∈ Cc (V ) . Then
Z Z
f (y) dmn = f (h (x)) |det (Dh (x))| dmn
V U

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 + ε) r)) . (9.9)

Making r still smaller if necessary, one can also obtain

|f (y) − f (h (x))| < ε (9.10)

for any y ∈ h (B (x, r)) and also

|f (h (x1 )) |det (Dh (x1 ))| − f (h (x)) |det (Dh (x))|| < ε (9.11)

whenever x1 ∈ 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 {Bi } such that U = ∪∞ i=1 Bi ∪ N where
9.9. CHANGE OF VARIABLES FOR C 1 FUNCTIONS 237

mn (N ) = 0. Denote by xi the center of Bi and ri the radius. Then by Lemma 9.8.1,

the monotone convergence theorem, and 9.9 - 9.11,
R P∞ R
V
f (y) dmn = i=1 h(Bi ) f (y) dmn
P∞ R
≤ εmn (V ) + i=1 h(Bi ) f (h (xi )) dmn
P∞
≤ εmn (V ) + i=1 f (h (xi )) mn (h (Bi ))
P∞
≤ εmn (V ) + i=1 f (h (xi )) mn (Dh (xi ) (B (0, (1 + ε) ri )))
n P∞ R
= εmn (V ) + (1 + ε) i=1 Bi f (h (xi )) |det (Dh (xi ))| dmn
P ³R ´
n ∞
≤ εmn (V ) + (1 + ε) i=1 Bi
f (h (x)) |det (Dh (x))| dmn + εmn (Bi )
n P∞ R n
≤ εmn (V ) + (1 + ε) i=1 Bi f (h (x)) |det (Dh (x))| dmn + (1 + ε) εmn (U )
nR n
= εmn (V ) + (1 + ε) U f (h (x)) |det (Dh (x))| dmn + (1 + ε) εmn (U )

Since ε > 0 is arbitrary, this shows

Z Z
f (y) dmn ≤ f (h (x)) |det (Dh (x))| dmn (9.12)
V U

whenever f ∈ Cc (V ) . Now x →f (h (x)) |det (Dh (x))| is in Cc (U ) and so using the

same argument with U and V switching roles and replacing h with h−1 ,
Z
f (h (x)) |det (Dh (x))| dmn
ZU
¡ ¡ ¢¢ ¯ ¡ ¡ ¢¢¯ ¯ ¡ ¢¯
≤ f h h−1 (y) ¯det Dh h−1 (y) ¯ ¯det Dh−1 (y) ¯ dmn
ZV
= f (y) dmn
V

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 Rn and let h, h−1 be C 1
functions such that h (U ) = V . Also let E ⊆ V be measurable. Then
Z Z
XE (y) dmn = XE (h (x)) |det (Dh (x))| dmn .
V U

Proof: By regularity, there exist compact sets, Kk and a decreasing sequence of

open sets Gk such that
Kk ⊆ E ⊆ G k
and mn (Gk \ Kk ) < 2−k . By Lemma 8.10.2, there exist fk such that Kk ≺ fk ≺ Gk .
Then fk (y) → XE (y) a.e. because if y is such that P
convergence fails, it must be the
case that y is in Gk \ Kk for infinitely many k and k mn (Gk \ Kk ) < ∞. This set
equals
N = ∩∞ ∞
m=1 ∪k=m Gk \ Kk

and so for each m ∈ N

mn (N ) ≤ mn (∪∞
k=m Gk \ Kk )
X∞ ∞
X
≤ mn (Gk \ Kk ) < 2−k = 2−(m−1)
k=m k=m
238 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

showing mn (N ) = 0.
Then fk (h (x)) must converge to XE (h (x)) for all x ∈/ h−1 (N ) , a set of measure
zero by Lemma 9.8.1. Thus
Z Z
fk (y) dmn = fk (h (x)) |det (Dh (x))| dmn .
V U

Since U is bounded G1 is compact. Therefore, |det (Dh (x))| is bounded independent

of k and so, by the dominated convergence theorem, using a dominating function, XV
in the integral on the left and XG1 |det (Dh)| on the right, it follows
Z Z
XE (y) dmn = XE (h (x)) |det (Dh (x))| dmn .
V U

This proves the corollary. ¥

You don’t need to assume the open sets are bounded.

Corollary 9.9.3 Let U and V be open sets in Rn and let h, h−1 be C 1 functions
such that h (U ) = V . Also let E ⊆ V be measurable. Then
Z Z
XE (y) dmn = XE (h (x)) |det (Dh (x))| dmn .
V U

Proof: For each x ∈ U, there exists rx such that B (x, rx ) ⊆ U and rx < 1. Then by
the mean value inequality Theorem 6.4.2, it follows h (B (x, rx )) is also bounded. These
balls, B (x, rx ) give a Vitali cover of U and so by Corollary 9.7.6 there is a sequence of
these balls, {Bi } such that they are disjoint, h (Bi ) is bounded and

mn (U \ ∪i Bi ) = 0.

It follows from Lemma 9.8.1 that h (U \ ∪i Bi ) also has measure zero. Then from Corol-
lary 9.9.2
Z XZ
XE (y) dmn = XE∩h(Bi ) (y) dmn
V i h(Bi )
XZ
= XE (h (x)) |det (Dh (x))| dmn
i Bi
Z
= XE (h (x)) |det (Dh (x))| dmn .
U

This proves the corollary. ¥

With this corollary, the main theorem follows.

Theorem 9.9.4 Let U and V be open sets in Rn and let h, h−1 be C 1 functions
such that h (U ) = V. Then if g is a nonnegative Lebesgue measurable function,
Z Z
g (y) dmn = g (h (x)) |det (Dh (x))| dmn . (9.13)
V U

Proof: From Corollary 9.9.3, 9.13 holds for any nonnegative simple function in place
of g. In general, let {sk } be an increasing sequence of simple functions which converges
to g pointwise. Then from the monotone convergence theorem
Z Z Z
g (y) dmn = lim sk dmn = lim sk (h (x)) |det (Dh (x))| dmn
V k→∞ V k→∞ U
Z
= g (h (x)) |det (Dh (x))| dmn .
U
9.9. CHANGE OF VARIABLES FOR C 1 FUNCTIONS 239

This proves the theorem. ¥

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 Rn and let h, h−1 be C 1 functions
such that h (U ) = V. Let g ∈ L1 (V ) . Then
Z Z
g (y) dmn = g (h (x)) |det (Dh (x))| dmn .
V U

This is a pretty good theorem but it isn’t 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 n−1 dimensional subspace of Rn and K is a compact

subset of V . Then letting

Kε ≡ ∪x∈K B (x,ε) = K + B (0, ε) ,

it follows that
n−1
mn (Kε ) ≤ 2n ε (diam (K) + ε) .

Proof: Using the Gram Schmidt procedure, there exists an orthonormal basis for
V , {v1 , · · ·, vn−1 } and let
{v1 , · · ·, vn−1 , vn }
be an orthonormal basis for Rn . Now define a linear transformation, Q by Qvi = ei .
Thus QQ∗ = Q∗ Q = I and Q preserves all distances and is a unitary transformation
because ¯ ¯2 ¯ ¯2 ¯ ¯2
¯ X ¯ ¯X ¯ X ¯X ¯
¯ ¯ ¯ ¯ 2 ¯ ¯
¯Q ai ei ¯ = ¯ ai vi ¯ = |ai | = ¯ ai ei ¯ .
¯ ¯ ¯ ¯ ¯ ¯
i i i i

Thus mn (Kε ) = mn (QKε ). Letting k0 ∈ K, it follows K ⊆ B (k0 , diam (K)) and so,

QK ⊆ B n−1 (Qk0 , diam (QK)) = B n−1 (Qk0 , diam (K))

where B n−1 refers to the ball taken with respect to the usual norm in Rn−1 . Every
point of Kε is within ε of some point of K and so it follows that every point of QKε is
within ε of some point of QK. Therefore,

QKε ⊆ B n−1 (Qk0 , diam (QK) + ε) × (−ε, ε) ,

To see this, let x ∈ QKε . Then there exists k ∈ QK such that |k − x| < ε. Therefore,
|(x1 , · · ·, xn−1 ) − (k1 , · · ·, kn−1 )| < ε and |xn − kn | < ε and so x is contained in the set
on the right in the above inclusion because kn = 0. However, the measure of the set on
the right is smaller than
n−1 n−1
[2 (diam (QK) + ε)] (2ε) = 2n [(diam (K) + ε)] ε.

This proves the lemma. ¥

Note this is a very sloppy estimate. You can certainly do much better but this
estimate is sufficient to prove Sard’s lemma which follows.

Definition 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 } .

240 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Lemma 9.9.8 Let h be a C 1 function defined on an open set, U ⊆ Rn and let K be

a compact subset of U. Then if ε > 0 is given, there exists r1 > 0 such that if |v| ≤ r1 ,
then for all x ∈ K,
|h (x + v) − h (x) − Dh (x) v| < ε |v| .
¡ ¢
Proof: Let 0 < δ < dist K, U C . Such a positive number exists because if there
exists a sequence of points in K, {kk } and points in U C , {sk } such that |kk − sk | → 0,
then you could take a subsequence, still denoted by k such that kk → k ∈ K and then
sk → k also. But U C is closed so k ∈ K ∩ U C , a contradiction. Then for |v| < δ it
follows that for every x ∈ K,
x+tv ∈ U

and
¯R ¯
¯ 1 ¯
|h (x + v) − h (x) − Dh (x) v| ¯ 0 Dh (x + tv) vdt − Dh (x) v¯
≤
|v| |v|
R1
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 Rn
 
Dh1 (x+tv) v
 .. 
 . 
Dhn (x+tv) v

and the integral is defined by

 R1 
0
Dh1 (x+tv) vdt
 .. 
 . 
R1
0
Dhn (x+tv) vdt

Now from uniform continuity of Dh on the compact set, {x : dist (x,K) ≤ δ} it follows
there exists r1 < δ such that if |v| ≤ r1 , then ||Dh (x + tv) − Dh (x)|| < ε for every
x ∈ K. From the above formula, it follows that if |v| ≤ r1 ,
R1
|h (x + v) − h (x) − Dh (x) v| 0
|Dh (x + tv) v − Dh (x) v| dt
≤
|v| |v|
R1
0
ε |v| dt
< = ε.
|v|

This proves the lemma. ¥

The following is Sard’s lemma. In the proof, it does not matter which norm you use
in defining balls.

Lemma 9.9.9 (Sard) Let U be an open set in Rn and let h : U → Rn be C 1 . Let

Z ≡ {x ∈ U : det Dh (x) = 0} .

Then mn (h (Z)) = 0.
9.9. CHANGE OF VARIABLES FOR C 1 FUNCTIONS 241

∞
Proof: Let {Uk }k=1 be an increasing sequence of open sets whose closures are com-
pact and whose union equals U and let Zk ≡ Z ∩ Uk . To obtain such a sequence,
let ½ ¾
¡ C
¢ 1
Uk = x ∈ U : dist x,U < ∩ B (0, k) .
k
First it is shown that h (Zk ) has measure zero. Let W be an open set contained in Uk+1
which contains Zk and satisfies

mn (Zk ) + ε > mn (W )

where here and elsewhere, ε < 1. Let

¡ C
¢
r = dist Uk , Uk+1

and let r1 > 0 be a constant as in Lemma 9.9.8 such that whenever x ∈ Uk and
0 < |v| ≤ r1 ,
|h (x + v) − h (x) − Dh (x) v| < ε |v| . (9.14)
Now the closures of balls which are contained in W and which have the property that
their diameters are less than r1 yield a Vitali covering n ofoW. Therefore, by Corollary
9.7.6 there is a disjoint sequence of these closed balls, Bei such that

W = ∪∞ e
i=1 Bi ∪ N

where N is a set of measure zero. Denote by {Bi } those closed balls in this sequence
which have nonempty intersection with Zk , let di be the diameter of Bi , and let zi be a
point in Bi ∩ Zk . Since zi ∈ Zk , it follows Dh (zi ) B (0,di ) = Di where Di is contained
in a subspace, V which has dimension n − 1 and the diameter of Di is no larger than
2Ck di where
Ck ≥ max {||Dh (x)|| : x ∈ Zk }
Then by 9.14, if z ∈ Bi ,

h (z) − h (zi ) ∈ Di + B (0, εdi ) ⊆ Di + B (0,εdi ) .

Thus
h (Bi ) ⊆ h (zi ) + Di + B (0,εdi )
By Lemma 9.9.6
n−1
mn (h (Bi )) ≤ 2n (2Ck di + εdi ) εdi
³ ´
n−1
≤ dni 2n [2Ck + ε] ε
≤ Cn,k mn (Bi ) ε.

Therefore, by Lemma 9.8.1

X X
mn (h (Zk )) ≤ mn (W ) = mn (h (Bi )) ≤ Cn,k ε mn (Bi )
i i
≤ εCn,k mn (W ) ≤ εCn,k (mn (Zk ) + ε)

Since ε is arbitrary, this shows mn (h (Zk )) = 0 and so 0 = limk→∞ mn (h (Zk )) =

mn (h (Z)). ¥
With this important lemma, here is a generalization of Theorem 9.9.4.
242 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Theorem 9.9.10 Let U be an open set and let h be a 1 − 1, C 1 (U ) function

n
with values in R . Then if g is a nonnegative Lebesgue measurable function,
Z Z
g (y) dmn = g (h (x)) |det (Dh (x))| dmn . (9.15)
h(U ) U

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,
Z Z Z
g (y) dmn = g (y) dmn = g (h (x)) |det (Dh (x))| dmn
h(U ) h(U \Z) U \Z
Z
= g (h (x)) |det (Dh (x))| dmn .
U

This proves the theorem. ¥

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 mn (h(Z)) = 0. For x ∈ U+ ,
the inverse function theorem implies there exists an open set Bx ⊆ U+ , such that h is
one to one on Bx .
Let {Bi } be a countable subset of {Bx }x∈U+ such that U+ = ∪∞i=1 Bi . Let E1 = B1 .
If E1 , · · ·, Ek have been chosen, Ek+1 = Bk+1 \ ∪ki=1 Ei . Thus

∪∞
i=1 Ei = U+ , h is one to one on Ei , Ei ∩ Ej = ∅,

and each Ei is a Borel set contained in the open set Bi . Now define

∞
X
n(y) ≡ Xh(Ei ) (y) + Xh(Z) (y).
i=1

The set, h (Ei ) , h (Z) are measurable by Lemma 9.8.2. Thus n (·) is measurable.

Lemma 9.10.1 Let F ⊆ h(U ) be measurable. Then

Z Z
n(y)XF (y)dmn = XF (h(x))| det Dh(x)|dmn .
h(U ) U
9.11. SPHERICAL COORDINATES IN N DIMENSIONS 243

Proof: Using Lemma 9.9.9 and the Monotone Convergence Theorem

 
mn (h(Z))=0
Z Z X∞ z }| {
 
n(y)XF (y)dmn =  Xh(Ei ) (y) + Xh(Z) (y)  XF (y)dmn
h(U ) h(U ) i=1

∞ Z
X
= Xh(Ei ) (y)XF (y)dmn
i=1 h(U )
X∞ Z
= Xh(Ei ) (y)XF (y)dmn
i=1 h(Bi )
X∞ Z
= XEi (x)XF (h(x))| det Dh(x)|dmn
i=1 Bi
X∞ Z
= XEi (x)XF (h(x))| det Dh(x)|dmn
i=1 U
Z X
∞
= XEi (x)XF (h(x))| det Dh(x)|dmn
U i=1

Z Z
= XF (h(x))| det Dh(x)|dmn = XF (h(x))| det Dh(x)|dmn .
U+ U

This proves the lemma. ¥

Definition 9.10.2 For y ∈ h(U ), define 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

Z Z
#(y)g(y)dmn = g(h(x))| det Dh(x)|dmn . (9.17)
h(U ) U

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.
This proves the theorem. ¥

9.11 Spherical Coordinates In n Dimensions

Sometimes there is a need to deal with spherical coordinates in more than three di-
mensions. In this section, this concept is defined and formulas are derived for these
coordinate systems. Recall polar coordinates are of the form

y1 = ρ cos θ
y2 = ρ sin θ
244 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

where ρ > 0 and θ ∈ R. Thus these transformation equations are not one to one but they
are one to one on (0, ∞)×[0, 2π). Here I am writing ρ 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 y1 and y2 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.

R (x1 , x2 , x3 )
φ1
ρ

R2

From this picture, you see that y3 = ρ cos φ1 . Also the distance between (y1 , y2 ) and
(0, 0) is ρ sin (φ1 ) . Therefore, using polar coordinates to write (y1 , y2 ) in terms of θ and
this distance,
y1 = ρ sin φ1 cos θ,
y2 = ρ sin φ1 sin θ,
y3 = ρ cos φ1 .
where φ1 ∈ R and the transformations are one to one if φ1 is restricted to be in [0, π] .
What was done is to replace ρ with ρ sin φ1 and then to add in y3 = ρ cos φ1 . Having
done this, there is no reason to stop with three dimensions. Consider the following
picture:

R (x1 , x2 , x3 , x4 )
φ2
ρ
R3

From this picture, you see that y4 = ρ cos φ2 . Also the distance between (y1 , y2 , y3 )
and (0, 0, 0) is ρ sin (φ2 ) . Therefore, using polar coordinates to write (y1 , y2 , y3 ) in terms
of θ, φ1 , and this distance,

y1 = ρ sin φ2 sin φ1 cos θ,

y2 = ρ sin φ2 sin φ1 sin θ,
y3 = ρ sin φ2 cos φ1 ,
y4 = ρ cos φ2

where φ2 ∈ R and the transformations will be one to one if

φ2 , φ1 ∈ [0, π] , θ ∈ [0, 2π), ρ ∈ (0, ∞) .

Continuing this way, given spherical coordinates in Rn , to get the spherical coordi-
nates in Rn+1 , you let yn+1 = ρ cos φn−1 and then replace every occurance of ρ with
ρ sin φn−1 to obtain y1 · · · yn in terms of φ1 , φ2 , · · ·, φn−1 ,θ, and ρ.
It is always the case that ρ measures the distance from the point in Rn to the origin
in Rn , 0. Each φi ∈ R and the transformations will be one to one if each φi ∈ [0, π] ,
and θ ∈ [0, 2π). It can be shown using math induction that these coordinates map
Qn−2 n
i=1 [0, π] × [0, 2π) × (0, ∞) one to one onto R \ {0} .
9.11. SPHERICAL COORDINATES IN N DIMENSIONS 245

Theorem 9.11.1QLet y = h (φ, θ, ρ) be the spherical coordinate transformations

n−2
in Rn . Then letting A = i=1 [0, π] × [0, 2π), it follows h maps A × (0, ∞) one to one
n
onto R \ {0} . Also |det Dh (φ, θ, ρ)| will always be of the form

|det Dh (φ, θ, ρ)| = ρn−1 Φ (φ, θ) . (9.18)

where Φ is a continuous function of φ and θ.1 Furthermore whenever f is Lebesgue

measurable and nonnegative,
Z Z ∞ Z
n−1
f (y) dy = ρ f (h (φ, θ, ρ)) Φ (φ, θ) dφ dθdρ (9.19)
Rn 0 A

where here dφ dθ denotes dmn−1 on A. The same formula holds if f ∈ L1 (Rn ) .

Proof: Formula 9.18 is obvious from the definition of the spherical coordinates.
The first claim isQalso clear from the definition and math induction. It remains to verify
n−2
9.19. Let A0 ≡ i=1 (0, π) × (0, 2π) . Then it is clear that (A \ A0 ) × (0, ∞) ≡ N is a
set of measure zero in Rn . Therefore, from Lemma 9.8.1 it follows h (N ) is also a set of
measure zero. Therefore, using the change of variables theorem, Fubini’s theorem, and
Sard’s lemma,
Z Z Z
f (y) dy = f (y) dy = f (y) dy
Rn Rn \{0} Rn \({0}∪h(N ))
Z
= f (h (φ, θ, ρ)) ρn−1 Φ (φ, θ) dmn
A0 ×(0,∞)
Z
= XA×(0,∞) (φ, θ, ρ) f (h (φ, θ, ρ)) ρn−1 Φ (φ, θ) dmn
Z ∞ µZ ¶
n−1
= ρ f (h (φ, θ, ρ)) Φ (φ, θ) dφ dθ dρ.
0 A

Now the claim about f ∈ L1 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 differently. Note that from the spherical coor-
dinate formulas, f (h (φ, θ, ρ)) = f (ρω) where |ω| = 1. Letting S n−1 denote the unit
sphere, {ω ∈ Rn : |ω| = 1} , the inside integral in the above formula is sometimes written
as Z
f (ρω) dσ
S n−1

where σ is a measure on S n−1 . See [27] for another description of this measure. It isn’t
an important issue here. Either 9.19 or the formula
Z ∞ µZ ¶
ρn−1 f (ρω) dσ dρ
0 S n−1

will
¡ be referred
¢ R to as polar coordinates and is very useful in establishing estimates. Here
σ S n−1 ≡ A Φ (φ, θ) dφ dθ.
R ³ ´s
2
Example 9.11.3 For what values of s is the integral B(0,R) 1 + |x| dy bounded
n
independent of R? Here B (0, R) is the ball, {x ∈ R : |x| ≤ R} .
1 Actually it is only a function of the first but this is not important in what follows.
246 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

I think you can see immediately that s must be negative but exactly how negative?
It turns out it depends on n and using polar coordinates, you can find just exactly what
is needed. From the polar coordinates formula above,
Z ³ ´s Z R Z
2 ¡ ¢s
1 + |x| dy = 1 + ρ2 ρn−1 dσdρ
B(0,R) 0 S n−1
Z R ¡ ¢s
= Cn 1 + ρ2 ρn−1 dρ
0

Now the very hard problem has been reduced to considering an easy one variable problem
of finding when
Z R
¡ ¢s
ρn−1 1 + ρ2 dρ
0

is bounded independent of R. You need 2s + (n − 1) < −1 so you need s < −n/2.

9.12 Brouwer Fixed Point Theorem

The Brouwer fixed point theorem is one of the most significant 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 n × n matrix, cof (A)ij is the determinant of the matrix which
i+j
results from deleting the ith row and the j th column and multiplying by (−1) . 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 Rn . Thus
X
Dg (x) = (Dg)ij ei ej
ij
P
and recall that (Dg)ij = ∂gi /∂xj where g = i gi ei .

Lemma 9.12.1 Let g : U → Rn be C 2 where U is an open subset of Rn . Then

n
X
cof (Dg)ij,j = 0,
j=1

∂gi ∂ det(Dg)
where here (Dg)ij ≡ gi,j ≡ ∂xj . Also, cof (Dg)ij = ∂gi,j .

Proof: From the cofactor expansion theorem,

n
X
det (Dg) = gi,j cof (Dg)ij
i=1

and so
∂ det (Dg)
= cof (Dg)ij (9.20)
∂gi,j
which shows the last claim of the lemma. Also
X
δ kj det (Dg) = gi,k (cof (Dg))ij (9.21)
i
9.12. BROUWER FIXED POINT THEOREM 247

because if k 6= j this is just the cofactor expansion of the determinant of a matrix in

which the k th and j th columns are equal. Differentiate 9.21 with respect to xj and sum
on j. This yields
X ∂ (det Dg) X X
δ kj gr,sj = gi,kj (cof (Dg))ij + gi,k cof (Dg)ij,j .
r,s,j
∂gr,s ij ij

Hence, using δ kj = 0 if j 6= k and 9.20,

X X X
(cof (Dg))rs gr,sk = gr,ks (cof (Dg))rs + gi,k cof (Dg)ij,j .
rs rs ij

Subtracting the first sum on the right from both sides and using the equality of mixed
partials,  
X X
gi,k  (cof (Dg))ij,j  = 0.
i j
P
If det (gi,k ) 6= 0 so that (gi,k ) is invertible, this shows j (cof (Dg))ij,j = 0. If
det (Dg) = 0, let
gk (x) = g (x) + εk x
where εk → 0 and det (Dg + εk I) ≡ det (Dgk ) 6= 0. Then
X X
(cof (Dg))ij,j = lim (cof (Dgk ))ij,j = 0
k→∞
j j

and This proves the lemma. ¥

Definition
¡ ¢ 9.12.2 Let h be a function defined on an open set, U ⊆ Rn . Then
k
h ∈ C U if there exists a function g defined 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 defining the balls and everything
would work the same but I have in mind the usual norm.
³ ´
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 λ ∈ [0, 1] and let pλ (x) ≡ x + λ (h (x) − x) .
This function, pλ is called a homotopy of the identity map and the retraction, h. Let
Z
I (λ) ≡ det (Dpλ (x)) dx.
B(0,R)

Then using the dominated convergence theorem,

Z X ∂ det (Dpλ (x)) ∂pλij (x)
I 0 (λ) = dx
B(0,R) i.j ∂pλi,j ∂λ
Z X X ∂ det (Dpλ (x))
= (hi (x) − xi ),j dx
B(0,R) i j
∂pλi,j
Z XX
= cof (Dpλ (x))ij (hi (x) − xi ),j dx
B(0,R) i j
248 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Now by assumption, hi (x) = xi on ∂B (0, R) and so one can form iterated integrals
and integrate by parts in each of the one dimensional integrals to obtain
XZ X
I 0 (λ) = − cof (Dpλ (x))ij,j (hi (x) − xi ) dx = 0.
i B(0,R) j

Therefore, I (λ) equals a constant. However,

I (0) = mn (B (0, R)) > 0

but Z Z
I (1) = det (Dh (x)) dmn = # (y) dmn = 0
B(0,1) ∂B(0,1)

because from polar coordinates or other elementary reasoning, mn (∂B (0, 1)) = 0. This
proves the lemma. ¥
The following is the Brouwer fixed point theorem for C 2 maps.
³ ´
Lemma 9.12.4 If h ∈ C 2 B (0, R) and h : B (0, R) → B (0, R), then h has a
fixed point, x such that h (x) = x.

Proof: Suppose the lemma is not true. Then for all x, |x − h (x)| 6= 0. Then define

x − h (x)
g (x) = h (x) + t (x)
|x − h (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
µ ¶
2 x − h (x)
H (x, t) = |h (x)| + 2 h (x) , t + t2 = R 2 (9.22)
|x − h (x)|

Then µ ¶
x − h (x)
Ht (x, t) = 2 h (x) , + 2t.
|x − h (x)|

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
µ ¶
x − h (x)
2t = −2 h (x) ,
|x − h (x)|
s µ ¶2
x − h (x) ³ ´
2
± 4 h (x) , − 4 |h (x)| − R2
|x − h (x)|
9.13. EXERCISES 249

and so
µ ¶
x − h (x)
Ht (x, t) = 2t + 2 h (x) ,
|x − h (x)|
s µ ¶2
³ ´ x − h (x)
2 2
= ± 4 R − |h (x)| + 4 h (x) ,
|x − h (x)|
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 π/2.
Alternatively, if the above equals zero, you would need
2
(h (x) , x) = |h (x)| = R2
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 fixed point theorem.
Theorem 9.12.5 Let f : B (0, R) → B (0, R) be continuous. Then f has a fixed
point.
Proof: If this is not so, there exists ε > 0 such that for all x ∈ B (0, R),
|x − f (x)| > ε.
By the Weierstrass approximation theorem, there exists h, a polynomial such that
n o ε
max |h (x) − f (x)| : x ∈ B (0, R) < .
2
Then for all x ∈ B (0, R),
ε ε
|x − h (x)| ≥ |x − f (x)| − |h (x) − f (x)| > ε − =
2 2
contradicting Lemma 9.12.4. This proves the theorem. ¥

9.13 Exercises
1. Recall the definition of fy . Prove that if f ∈ L1 (Rn ) , then
Z
lim |f − fy | dmn = 0
y→0 Rn

This is known as continuity of translation. Hint: Use the theorem about being
able to approximate an arbitrary function in L1 (Rn ) with a function in Cc (Rn ).
2. Show that if a, b ≥ 0 and if p, q > 0 such that
1 1
+ =1
p q
then
ap bq
ab ≤ +
p q
ap bq
Hint: You might consider for fixed a ≥ 0, the function h (b) ≡ p + q − ab and
find its minimum.
250 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

3. In the context of the previous problem, prove Holder’s inequality. If f, g measur-

able functions, then
Z µZ ¶1/p µZ ¶1/q
p q
|f | |g| dµ ≤ |f | dµ |g| dµ

Hint: If either of the factors on the right equals 0, explain why there is nothing
¡R p ¢1/p ¡R q ¢1/q
to show. Now let a = |f | / |f | dµ and b = |g| / |g| dµ . 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

Z
f (x) = XE (t)XE (x + t)dt.

Explain why f is continuous at 0 and f (0) > 0 and use continuity of translation
in L1 .
5. If f ∈ L1 (Rn ) , show there exists g ∈ L1 (Rn ) such that g is also Borel measurable
such that g (x) = f (x) for a.e. x.
6. Suppose f, g ∈ L1 (Rn ) . Define f ∗ g (x) by
Z
f (x − y) g (y) dmn (y) .

Show this makes sense for a.e. x and that in fact for a.e. x
Z
|f (x − y)| |g (y)| dmn (y)

Next show Z Z Z
|f ∗ g (x)| dmn (x) ≤ |f | dmn |g| dmn .

Hint: Use Problem 5. Show first there is no problem if f, g are Borel measurable.
The reason for this is that you can use Fubini’s theorem to write
Z Z
|f (x − y)| |g (y)| dmn (y) dmn (x)
Z Z
= |f (x − y)| |g (y)| dmn (x) dmn (y)
Z Z
= |f (z)| dmn |g (y)| dmn .

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 : [0, ∞) → R be in L1 (R, m). The Laplace transform b
R ∞ f−xt 1
R x is given by f (x) =
0
e f (t)dt. Let f, g be in L (R, m), and let h(x) = 0 f (x − t)g(t)dt. Show
h ∈ L1 , and b
h = fbgb.
9.13. EXERCISES 251

9. Suppose A is covered by a finite collection of Balls, F. Show that then there exists
p bi where B
a disjoint collection of these balls, {Bi }i=1 , such that A ⊆ ∪pi=1 B bi has
the same center as Bi but 3 times the radius. Hint: Since the collection of balls
is finite, they can be arranged in order of decreasing radius.
10. Let f be a function defined on an interval, (a, b). The Dini derivates are defined
as
f (x + h) − f (x)
D+ f (x) ≡ lim inf ,
h→0+ h
f (x + h) − f (x)
D+ f (x) ≡ lim sup
h→0+ h

f (x) − f (x − h)
D− f (x) ≡ lim inf ,
h→0+ h
f (x) − f (x − h)
D− f (x) ≡ lim sup .
h→0+ 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 x1 ∈ (a, b) where H (x1 ) > H (c), then let x0 ∈ (c, d) be the point
where the maximum of f occurs. Consider D+ f (x0 ). 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 fe(x) ≡ f (x) + εx, consider the situation where
D+ fe(x) > 0 for all but
³ countably
e e
´ many x. If in this situation, f (c) > f (d) for
some c < d, and y ∈ fe(d) , fe(c) ,let
n o
z ≡ sup x ∈ [c, d] : fe(x) > y0 .

Show that fe(z) = y0 and D+ fe(z) ≤ 0. Conclude that if fe fails to be increasing,

then D+ fe(z) ≤ 0 for uncountably many points, z. Now draw a conclusion about
f.
12. ↑ Let f : [a, b] → R be increasing. Show
 Npq

z£ }| {¤
 
m  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
off some set of measure zero, we have
D− f (x) = D− f (x) = D+ f (x) = D+ f (x)
252 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

which implies the derivative exists and equals this common value. Hint: To show
9.23, let U be an open set containing Npq such that m (Npq ) + ε > m (U ). For
each x ∈ Npq 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 Npq . Let {[xi , yi ]} be disjoint and

m (Npq \ ∪i [xi , yi ]) = 0.

Now let V ≡ ∪i (xi , yi ). Then also we have

 
=V
 z }| { 
m Npq \ ∪i (xi , yi ) = 0.

and so m (Npq ∩ V ) = m (Npq ). For each x ∈ Npq ∩V , there exist y > x arbitrarily
close to x such that
f (y) − f (x) > q (y − x) .
Thus the set of such intervals, {[x0 , y 0 ]} which are contained in V is a Vitali cover
of Npq ∩ V . Let {[x0i , yi0 ]} be disjoint and

m (Npq ∩ V \ ∪i [x0i , yi0 ]) = 0.

Then verify the following:

X X
f (yi0 ) − f (x0i ) > q (yi0 − x0i ) ≥ qm (Npq ∩ V ) = qm (Npq )
i i
X
≥ pm (Npq ) > p (m (U ) − ε) ≥ p (yi − xi ) − pε
i
X X
≥ (f (yi ) − f (xi )) − pε ≥ f (yi0 ) − f (x0i ) − pε
i i

and therefore, (q − p) m (Npq ) ≤ pε. Since ε > 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 L1 (R) and f is infinitely differentiable. Does it follow
that f 0 ∈ L1 (R)? Hint: What if φ ∈ Cc∞ (0, 1) and f (x) = φ (2n (x − n)) for
x ∈ (n, n + 1) , f (x) = 0 if x < 0?
14. For a function f ∈ L1 (Rn ), the Fourier transform, F f is given by
Z
1
F f (t) ≡ √ e−it·x f (x) dx
2π R n

and the so called inverse Fourier transform, F −1 f is defined by

Z
1
F f (t) ≡ √ eit·x f (x) dx
2π Rn
Show that if f ∈ L1 (Rn ) , then lim|x|→∞ F f (x) = 0. Hint: You might try to
show this first for f ∈ Cc∞ (Rn ).
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

R∞ Rr
16. For this problem define a f (t) dt ≡ limr→∞ a f (t) dt. Note this coincides with
the Lebesgue integral when f ∈ L1 (a, ∞). Show
R∞ sin(u) π
(a) 0 u du = 2
R∞
(b) limr→∞ δ sin(ru)
u du= 0 whenever δ > 0.
1
R
(c) If f ∈ L (R), then limr→∞ R sin (ru) f (u) du = 0.
R∞
Hint: For the first two, use u1 = 0 e−ut dt and apply Fubini’s theorem to
RR R
0
sin u R e−ut dtdu. For the last part, first establish it for f ∈ Cc∞ (R) and
then use the density of this set in L1 (R) to obtain the result. This is called the
Riemann Lebesgue lemma.

17. ↑Suppose that g ∈ L1 (R) and that at some x > 0, g is locally Holder continuous
from the right and from the left. This means

lim g (x + r) ≡ g (x+)
r→0+

exists,
lim g (x − r) ≡ g (x−)
r→0+

exists and there exist constants K, δ > 0 and r ∈ (0, 1] such that for |x − y| < δ,
r
|g (x+) − g (y)| < K |x − y|

for y > x and

r
|g (x−) − g (y)| < K |x − y|
for y < x. Show that under these conditions,
Z µ ¶
2 ∞ sin (ur) g (x − u) + g (x + u)
lim du
r→∞ π 0 u 2

g (x+) + g (x−)
= .
2
18. ↑ Let g ∈ L1 (R) and suppose g is locally Holder continuous from the right and
from the left at x. Show that then
Z R Z ∞
1 g (x+) + g (x−)
lim eixt e−ity g (y) dydt = .
R→∞ 2π −R −∞ 2

This is very interesting. This shows F −1 (F g) (x) = g(x+)+g(x−)

2 , the midpoint of
the jump in g at the point, x provided F g is in L1 . Hint: Show the left side of
the above equation reduces to
Z µ ¶
2 ∞ sin (ur) g (x − u) + g (x + u)
du
π 0 u 2

and then use Problem 17 to obtain the result.

19. ↑ A measurable function g defined on (0, ∞) has exponential growth if |g (t)| ≤

Ceηt for some η. For Re (s) > η, define the Laplace Transform by
Z ∞
Lg (s) ≡ e−su g (u) du.
0
254 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

Assume that g has exponential growth as above and is Holder continuous from
the right and from the left at t. Pick γ > η. Show that
Z R
1 g (t+) + g (t−)
lim eγt eiyt Lg (γ + iy) dy = .
R→∞ 2π −R 2
This formula is sometimes written in the form
Z γ+i∞
1
est Lg (s) ds
2πi γ−i∞
and is called the complex inversion integral for Laplace transforms. It can be used
to find inverse Laplace transforms. Hint:
Z R
1
eγt eiyt Lg (γ + iy) dy =
2π −R
Z R Z ∞
1
eγt eiyt e−(γ+iy)u g (u) dudy.
2π −R 0
Now use Fubini’s theorem and do the integral from −R to R to get this equal to
Z
eγt ∞ −γu sin (R (t − u))
e g (u) du
π −∞ t−u

where g is the zero extension of g off [0, ∞). Then this equals
Z
eγt ∞ −γ(t−u) sin (Ru)
e g (t − u) du
π −∞ u
which equals
Z ∞
2eγt g (t − u) e−γ(t−u) + g (t + u) e−γ(t+u) sin (Ru)
du
π 0 2 u
and then apply the result of Problem 17.
20. Let K be a nonempty closed and convex subset of Rn . 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 ∈ Rn
there exists a unique z ∈ K such that

|x − z| = min {|x − y| : y ∈ K} .

This z will be denoted as P x. Hint: First note you do not know K is compact.
Establish the parallelogram identity if you have not already done so,
2 2 2 2
|u − v| + |u + v| = 2 |u| + 2 |v| .

Then let {zk } be a minimizing sequence,

2
lim |zk − x| = inf {|x − y| : y ∈ K} ≡ λ.
k→∞

Now using convexity, explain why

¯ ¯ ¯ ¯2 ¯ ¯2 ¯ ¯2
¯ zk − zm ¯2 ¯ ¯ ¯ ¯ ¯ ¯
¯ ¯ + ¯x− zk + zm ¯ = 2 ¯ x − zk ¯ + 2 ¯ x − zm ¯
¯ 2 ¯ ¯ 2 ¯ ¯ 2 ¯ ¯ 2 ¯
and then use this to argue {zk } is a Cauchy sequence. Then if zi works for i = 1, 2,
consider (z1 + z2 ) /2 to get a contradiction.
9.13. EXERCISES 255

21. In Problem 20 show that P x satisfies the following variational inequality.

(x−P x) · (y−P x) ≤ 0
for all y ∈ K. Then show that |P x1 − P x2 | ≤ |x1 − x2 |. Hint: For the first part
2
note that if y ∈ K, the function t → |x− (P x + t (y−P x))| achieves its minimum
on [0, 1] at t = 0. For the second part,
(x1 −P x1 ) · (P x2 −P x1 ) ≤ 0, (x2 −P x2 ) · (P x1 −P x2 ) ≤ 0.
Explain why
(x2 −P x2 − (x1 −P x1 )) · (P x2 −P x1 ) ≥ 0
and then use a some manipulations and the Cauchy Schwarz inequality to get the
desired inequality.
22. Establish the Brouwer fixed point theorem for any convex compact set in Rn .
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 fixed
point in K.
23. In the situation of the implicit function theorem, suppose f (x0 , y0 ) = 0 and
assume f is C 1 . Show that for (x, y) ∈ B (x0 , δ) × B (y0 , r) where δ, r are small
enough, the mapping
−1
x →Ty (x) ≡ x−D1 f (x0 , y0 ) f (x, y)
is continuous and maps B (x0 , δ) to B (x0 , δ/2) ⊆ B (x0 , δ). Apply the Brouwer
fixed 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 fixed point
theorem. It plays a prominent role in the treatment of the change of variables
formula in Rudin’s book, [35] and is useful in other contexts as well. The idea is
that if a continuous function mapping a ball in Rk to Rk doesn’t 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 Rk and let F : B → Rk be continuous and
suppose for some ε < 1,
|F (v) −v| < εr (9.24)
for all v ∈ B. Then
F (B) ⊇ B (0, r (1 − ε)) .
Hint: Suppose a ∈ B (0, r (1 − ε)) \ F (B) so it didn’t work. First explain why
a 6= F (v) for all v ∈ B. Now letting G :B → B, be defined by G (v) ≡ r(a−F(v))
|a−F(v)| ,it
follows G is continuous. Then by the Brouwer fixed 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.
2 r
(G (v) , v) = |v| = r2 = (a − F (v) , v)
|a − F (v)|
r
= (a − v + v − F (v) , v)
|a − F (v)|
r
= [(a − v, v) + (v − F (v) , v)]
|a − F (v)|
r h i
2
= (a, v) − |v| + (v − F (v) , v)
|a − F (v)|
r £ 2 ¤
≤ r (1 − ε) − r2 +r2 ε = 0.
|a − F (v)|
256 THE LEBESGUE INTEGRAL FOR FUNCTIONS OF N VARIABLES

25. Using Problem 24 establish the following interesting result. Suppose f : U → Rn

is differentiable. 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 Rn and let f : K → Rn be continuous
and let y ∈ Rn . Show using the Brouwer fixed 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 Browder’s lemma and it has
great significance in the study of certain types of nolinear operators. Now suppose
f : Rn → Rn is continuous and satisfies
(f (x) , x)
lim = ∞.
|x|→∞ |x|

Show using Browder’s 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 difficult theorems in topology
such as the Brouwer fixed 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 Rn . When this is understood, it is not too difficult 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
defined on an interval, I, and let y ∈ f (I) be such that f 0 (x) 6= 0 for all x ∈ f −1 (y). In
this case the degree is the sum of the signs of f 0 (x) for x ∈ f −1 (y), written as d (f, I, 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 defined for continuous functions and which has
nothing to do with differentiability.
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 first 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 Ω will refer to a bounded open set.

257
258 BROUWER DEGREE

¡ ¢
Definition 10.1.1 For Ω a bounded open set, denote by C Ω the set of func-
¡ ¢
tions which are continuous on Ω and by C m Ω , m ≤ ∞ the space of restrictions of
¡ ¢
functions in Ccm (Rn ) to Ω. The norm in C Ω is defined as follows.
© ª
||f ||∞ = ||f ||C (Ω) ≡ sup |f (x)| : x ∈ Ω .
¡ ¢ ¡ ¢
If the functions take values in Rn write C m Ω; Rn or C Ω; Rn for these functions
¡ ¢
if there is no differentiability assumed. The norm on C Ω; Rn is defined in the same
way as above,
© ª
||f ||∞ = ||f ||C (Ω;Rn ) ≡ sup |f (x)| : x ∈ Ω .

Also, C (Ω; Rn ) consists of functions which are continuous on Ω that have values in Rn
and C m (Ω; Rn ) denotes the functions which have m continuous derivatives defined on
Ω.
¡ ¢
Theorem 10.1.2
¡ ¢
Let Ω be a bounded open set in Rn and let f ∈ C Ω . Then
there exists g ∈ C ∞ Ω with ||g − f ||C (Ω) < ε. In fact, g can be assumed to equal a
polynomial for all x ∈ Ω.

Proof: This follows immediately from the Weierstrass approximation theorem,

¡ ¢ The-
/ C ∞ Ω because
orem 5.7.10. Pick a polynomial, p such that ||p − f ||C (Ω) < ε. Now p ∈
it does not vanish outside some compact subset of Rn so let g equal p multiplied by
some function ψ ∈ Cc∞ (Rn ) where ψ = 1 on Ω. 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 Ω; Rn for Ω a bounded subset of Rn , then for all ε > 0,
¢
∞ n
there exists g ∈ C Ω; R such that

||g − f ||∞ < ε.

Lemma 9.12.1 on Page 246 will also play an important role in the definition of the
Brouwer degree. Earlier it made possible an easy proof of the Brouwer fixed point
theorem. Later in this chapter, it is used to show the definition of the degree is well
defined. For convenience, here it is stated again.

Lemma 10.1.4 Let g : U → Rn be C 2 where U is an open subset of Rn . Then

n
X
cof (Dg)ij,j = 0,
j=1

∂gi ∂ det(Dg)
where here (Dg)ij ≡ gi,j ≡ ∂xj . Also, cof (Dg)ij = ∂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 {kn } , {cn } be such that
1
d+ > ||kn − cn || .
n
Since K is compact, there is a subsequence still denoted by {kn } such that kn → k ∈ K.
Then also
||cn − cm || ≤ ||cn − kn || + ||kn − km || + ||cm − km ||
If d = 0, then as m, n → ∞ it follows ||cn − cm || → 0 and so {cn } is a Cauchy sequence
which must converge to some c ∈ C. But then ||c − k|| = limn→∞ ||cn − kn || = 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 Definitions And Elementary Properties

In this section, f : Ω → Rn will be a continuous map. It is always assumed that f (∂Ω)
misses the point y where d (f , Ω, y) is the topological degree which is being defined.
Also, it is assumed Ω is a bounded open set.
© ¡ ¢ ª
Definition 10.2.1 Uy ≡ f ∈ C Ω; Rn : y ∈/ f (∂Ω) . (Recall that ∂Ω = Ω \
Ω) For two functions,
f , g ∈ Uy ,
f ∼ g if there exists a continuous function,

h : Ω × [0, 1] → Rn

such that h (x, 1) = g (x) and h (x, 0) = f (x) and x → h (x,t) ∈ Uy for all t ∈
[0, 1] (y ∈
/ h (∂Ω, t)). This function, h, is called a homotopy and f and g are homotopic.

Definition 10.2.2 For W an open set in Rn and g ∈ C 1 (W ; Rn ) y is called a

regular value of g if whenever x ∈ g−1 (y), det (Dg (x)) 6= 0. Note that if g−1 (y) = ∅,
it follows that y is a regular value from this definition. Denote by Sg 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
© ¡ ¢ ª
Uy ≡ f ∈ C Ω; Rn : y ∈ / f (∂Ω) .
¡ ¢
Furthermore, Uy is an open set in C Ω; Rn and if f ∈ Uy and ε > 0, there exists
¡ ¢
g ∈ [f ] ∩ C 2 Ω; Rn for which y is a regular value of g and ||f − g|| < ε.

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, h1
and the homotopy for g and k, the function h2 . Thus h1 (x,0) = f (x), h1 (x,1) = g (x)
and h2 (x,0) = g (x), h2 (x,1) = k (x). Then define a homotopy of f and k as follows.
½ £ ¤
h1 (x,2t) if t ∈ 0, 21£ ¤
h (x,t) ≡
h2 (x,2t − 1) if t ∈ 21 , 1 .
260 BROUWER DEGREE

¡ ¢
It is obvious that Uy is an open subset of C Ω; Rn . Next consider the claim that
[f ] is also an
¡ open¢ set. If f ∈ Uy , There exists δ > 0 such that B (y, 2δ) ∩ f (∂Ω) = ∅.
Let f1 ∈ C Ω; Rn with ||f1 − f ||∞ < δ. Then if t ∈ [0, 1], and x ∈ ∂Ω

|f (x) + t (f1 (x) − f (x)) − y| ≥ |f (x) − y| − t ||f − f 1 ||∞ > 2δ − tδ > 0.

Therefore, B (f ,δ) ⊆ [f ] because if f1 ∈ B (f , δ), this shows that, letting h (x,t) ≡

f (x) + t (f1 (x) − f (x)), f1 ∼ 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 Ω; Rn and ||g − f || < ε/2. If y
is a regular value of g, leave g unchanged. The desired function has been found. In the
other case, let δ be small enough that B (y, 2δ) ∩ g (∂Ω) = ∅. Next let
© ª
S ≡ x ∈ Ω : det Dg (x) = 0

By Sard’s 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
e be one of these regular values, |y−e
to y. Let y y| < ε/2, and consider

g1 (x) ≡ g (x) + y−e

y.

It follows g1 (x) = y if and only if g (x) = y e and so, since Dg (x) = Dg1 (x), y is a
regular value of g1 . Then for t ∈ [0, 1] and x ∈ ∂Ω,

|g (x) + t (g1 (x) − g (x)) − y| ≥ |g (x) − y| − t |y−e

y| > 2δ − tδ ≥ δ > 0.

provided |y−ey| is small enough. It follows g1 ∼ g and so g1 ∼ f . Also provided |y−e

y|
is small enough,

||f − g1 || ≤ ||f − g|| + ||g − g1 ||

< ε/2 + ε/2 = ε.

This proves the lemma. ¥

The ¡main conclusion
¢ of this lemma is that for f ∈ Uy , there always exists a function
g of C 2 Ω; Rn 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 Ω; Rn
¡ ¢
Here I will give a definition of the degree which works for all functions in C 2 Ω; Rn .
¡ ¢
Definition 10.2.4 Let g ∈ C 2 Ω; Rn ∩ Uy where Ω is a bounded open set.
Also let φε be a mollifier.
Z
φε ∈ Cc∞ (B (0, ε)) , φε ≥ 0, φε dx = 1.

Then Z
d (g, Ω, y) ≡ lim φε (g (x) − y) det Dg (x) dx
ε→0 Ω

Lemma 10.2.5 The above definition is well defined. In particular the limit exists.
In fact Z
φε (g (x) − y) det Dg (x) dx
Ω
10.2. DEFINITIONS AND ELEMENTARY PROPERTIES 261

does not depend on ε whenenver ε is small enough. If y is a regular value for g then
for all ε small enough,
Z
φε (g (x) − y) det Dg (x) dx ≡
Ω
X© ª
sgn (det Dg (x)) : x ∈ g−1 (y) (10.1)
¡ ¢
If f , g are two functions in C 2 Ω; Rn such that for all x ∈ ∂Ω

y∈
/ tf (x) + (1 − t) g (x) (10.2)

for all t ∈ [0, 1] , then for each ε > 0,

Z
φε (f (x) − y) det Df (x) dx
Ω
Z
= φε (g (x) − y) det Dg (x) dx (10.3)
Ω
¡ ¢
If g, f ∈ Uy ∩ C 2 Ω; Rn , and 10.2 holds, then

d (f , Ω, y) = d (g, Ω, 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 ε > 0 and equals the right side
of 10.1. I claim the right side of this equation is actually a finite sum. This follows from
the inverse function theorem because g−1 (y) is a closed, hence compact subset of Ω due
to the assumption that y ∈ / g (∂Ω). If g−1 (y) had infinitely many points in it, there
would exist a sequence of distinct points {xk } ⊆ g−1 (y). Since Ω is bounded, some
subsequence {xkl } would converge to a limit point x∞ . By continuity of g, it follows
x∞ ∈ g−1 (y) also and so x∞ ∈ Ω. Therefore, since y is a regular value, there is an
open set, Ux∞ , containing x∞ such that g is one to one on this open set contradicting
the assertion that liml→∞ xkl = x∞ . Therefore, this set is finite and so the sum is well
defined.
Thus the right side of 10.1 is finite 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
m
finitely many points, {xi }i=1 = g−1 (y). By the inverse function theorem, there exist
disjoint open sets, Ui with xi ∈ Ui , such that g is one to one on Ui with det (Dg (x))
having constant sign on Ui and g (Ui ) is an open set containing y. Then let ε be small
enough that B (y, ε) ⊆ ∩m i=1 g (Ui ) and let Vi ≡ g
−1
(B (y, ε)) ∩ Ui .

g(U3 ) g(U2 )
x3 V1 x1

V3 ε y g(U1 )
ª
V2 x2
262 BROUWER DEGREE

Therefore, for any ε this small,

Z m Z
X
φε (g (x) − y) det Dg (x) dx = φε (g (x) − y) det Dg (x) dx
Ω i=1 Vi

The reason for this is as follows. The integrand on the left is nonzero only if g (x) − y ∈
B (0, ε) which occurs only if g (x) ∈ B (y, ε) which is the same as x ∈ g−1 (B (y, ε)).
Therefore, the integrand is nonzero only if x is contained in exactly one of the disjoint
sets, Vi . Now using the change of variables theorem,
Xm Z
¡ ¢¯ ¯
= φε (z) det Dg g−1 (y + z) ¯det Dg−1 (y + z)¯ dz
i=1 g(Vi )−y
¡ ¢
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 (xi )) .

Therefore, this reduces to
m
X Z
sgn (det Dg (xi )) φε (z) dz =
i=1 g(Vi )−y

m
X Z m
X
sgn (det Dg (xi )) φε (z) dz = sgn (det Dg (xi )) .
i=1 B(0,ε) i=1
¡ ¢
In case g−1 (y) = ∅, there exists ε > 0 such that g Ω ∩ B (y, ε) = ∅ and so for ε this
small, Z
φε (g (x) − y) det Dg (x) dx = 0.
Ω

Showing the integral is constant for small ε

With this done it is necessary to show that the integral in the definition of the degree
is constant for small enough ε even if y is not a regular value. To do this, I will first
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 differentiable is used. Suppose then that f , g satisfy 10.2. Also let
ε > 0 be such that for all t ∈ [0, 1] ,

B (y, ε) ∩ (f +t (g − f )) (∂Ω) = ∅ (10.4)

Define for t ∈ [0, 1],

Z
H (t) ≡ φε (f − y + t (g − f )) det (D (f + t (g − f ))) dx.
Ω

Then if t ∈ (0, 1),

Z X
0
H (t) = φε,α (f (x) − y + t (g (x) − f (x))) ·
Ω α
10.2. DEFINITIONS AND ELEMENTARY PROPERTIES 263

(gα (x) − fα (x)) det D (f + t (g − f )) dx

Z
+ φε (f − y + t (g − f )) ·
Ω
X
det D (f + t (g − f )),αj (gα − fα ),j dx ≡ A + B.
α,j

In this formula, the function det is considered as a function of the n2 entries in the n × n
matrix and the , αj represents the derivative with respect to the αj th entry. Now as in
the proof of Lemma 9.12.1 on Page 246,

det D (f + t (g − f )),αj = (cof D (f +t (g − f )))αj

and so Z XX
B= φε (f − y + t (g − f )) ·
Ω α j

(cof D (f +t (g − f )))αj (gα − fα ),j dx.

By hypothesis

x → φε (f (x) −y + t (g (x) −f (x))) ·

(cof D (f (x) +t (g (x) −f (x))))αj

is in Cc1 (Ω) because if x ∈ ∂Ω, it follows by 10.4 that for all t ∈ [0, 1]

f (x) −y + t (g (x) −f (x)) ∈

/ B (0, ε)

and so φε (f (x) −y + t (g (x) −f (x))) = 0. Furthermore this situation persists for x near
∂Ω. Therefore, integrate by parts and write
Z XX
∂
B=− (φε (f − y + t (g − f ))) ·
Ω α j
∂x j

(cof D (f +t (g − f )))αj (gα − fα ) dx+

Z XX
− φε (f − y + t (g − f ))
Ω α j
· (cof D (f +t (g − f )))αj,j (gα − fα ) dx.

The second term equals zero by Lemma 10.1.4. Simplifying the first term yields
Z XXX
B=− φε,β (f − y + t (g − f )) ·
Ω α j β

(fβ,j + t (gβ,j − fβ,j )) (cof D (f +t (g − f )))αj (gα − fα ) dx

Z XX
= − φε,β (f − y + t (g − f )) δ βα ·
Ω α β
det (D (f +t (g − f ))) (gα − fα ) dx

Z X
= − φε,α (f − y + t (g − f ))
Ω α

· det (D (f +t (g − f ))) (gα − fα ) dx

264 BROUWER DEGREE

= −A.
0
Therefore, H (t) = 0 and¡ so H ¢is a constant.
Now let g ∈ Uy ∩ C 2 Ω; Rn . By Sard’s lemma, Lemma 9.9.9 there exists a regular
value y1 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
g1 (x) ≡ g (x) + y − y1
and let y1 − y be so small that

y∈
/ (1 − t) g1 + gt (∂Ω) ≡ g1 + t (g − g1 ) (∂Ω) for all t ∈ [0, 1] .

Then g1 (x) = y if and only if g (x) = y1 which is a regular value. Note also D (g (x)) =
D (g1 (x)). Then from what was just shown, letting f = g and g = g1 in the above and
using g − y1 = g1 − y,
Z
φε (g (x) − y1 ) det (D (g (x))) dx
ZΩ
= φε (g1 (x) − y) det (D (g (x))) dx
ZΩ
= φε (g (x) − y) det (D (g (x))) dx
Ω

Since y1 is a regular value of g it follows from the first part of the argument that the
first integral in the above is eventually constant for small enough ε. It follows the last
integral is also eventually constant for small enough ε. This proves the claim about
the limit existing and in fact being constant for small ε. The last claim follows right
away from the above. Suppose 10.2 holds. Then choosing ε small enough, it follows
d (f , Ω, y) = d (g, Ω, y) because the two integrals defining the degree for small ε are
equal. This proves the lemma. ¥ ¡ ¢
Next I will show that if f ∼ g where f , g ∈ Uy ∩ C 2 Ω; Rn then d (f , Ω, y) =
d (g, Ω, 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 Uy ,

m
{gi }i=1 ,
¡ ¢
such that gi ∈ C 2 Ω; Rn , and defining g0 ≡ k and gm+1 ≡ l, there exists δ > 0 such
that for i = 1, · · ·, m + 1,

B (y, δ) ∩ (tgi + (1 − t) gi−1 ) (∂Ω) = ∅, 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 Ω; Rn . This is the abstract idea. Now here is a
more down to earth treatment.
10.2. DEFINITIONS AND ELEMENTARY PROPERTIES 265

Let h : Ω×[0, 1] → Rn be a homotopy of k and l with the property that y ∈

/ h (∂Ω, t)
for all t ∈ [0, 1]. Such a homotopy exists because k ∼ l. Now let 0 = t0 < t1 < · · · <
tm+1 = 1 be such that for i = 1, · · ·, m + 1,

||h (·, ti ) − h (·, ti−1 )||∞ < δ (10.6)

where δ ¡> 0 is¢ small. By Lemma 10.2.3, for each i ∈ {1, · · ·, m} , there exists gi ∈
Uy ∩ C 2 Ω; Rn such that
||gi − h (·, ti )||∞ < δ (10.7)
Thus
||gi − gi−1 ||∞ ≤ ||gi − h (·, ti )||∞
+ ||h (·, ti ) − h (·, ti−1 )||∞ + ||gi−1 − h (·, ti−1 )||∞ < 3δ.
(Recall g0 ≡ k in case i = 1.)
It was just shown that for each x ∈ ∂Ω,

gi (x) ∈ B (gi−1 (x) , 3δ)

Also for each x ∈ ∂Ω

|gj (x) − y| ≥ |h (x, tj ) − y| − |gj (x) − h (x, tj )|

≥ |h (x, tj ) − y| − δ

For x ∈ ∂Ω

|tgi (x) + (1 − t) gi−1 (x) − y|

= |gi−1 (x) + t (gi (x) − gi−1 (x)) − y|

≥ |gi−1 (x) − y| − t |gi (x) − gi−1 (x)|

≥ (|h (x, ti−1 ) − y| − δ) − t3δ

≥ |h (x, ti−1 ) − y| − 4δ > δ

provided δ was small enough that

B (y, 5δ) ∩ h (∂Ω × [0, 1]) = ∅ (10.8)

so that for all t ∈ [0, 1] , |h (x, t) − y| > 5δ. This proves the lemma. ¥
¡ ¢
With this lemma the homotopy invariance of the degree on C 2 Ω; Rn 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 ∈ Uy ∩ C 2 Ω; Rn and suppose f ∼ g. Then
d (f ,Ω, y) = d (g,Ω, y) .
¡ ¢
When f ∈ Uy ∩ C Ω; Rn and y is a regular value of g with f ∼ g,
2

X© ª
d (f ,Ω, y) = sgn (det Dg (x)) : x ∈ g−1 (y) .

The degree is an integer. Also

y → d (f ,Ω, y)
is continuous on Rn \ f (∂Ω) and y → d (f ,Ω, y) is constant on every connected compo-
nent of Rn \ f (∂Ω).
266 BROUWER DEGREE

¡ ¢
Proof: From Lemma 10.2.6 there exists a sequence of functions in C 2 Ω; Rn having
the properties listed there. Then from Lemma 10.2.5

d (f ,Ω, y) = d (g1 ,Ω, y) = d (g2 ,Ω, y) = · · · = d (g,Ω, 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

g1 (x) ≡ g (x) + y − y1

where y1 is a regular point of g and |y − y1 | is so small that

y∈
/ (tg1 + (1 − t) g) (∂Ω) .

From Lemma 10.2.5

d (g1 , Ω, y) = d (g, Ω, y) .
But since g1 − y = g − y1 and det Dg (x) = det Dg1 (x) ,
Z
d (g1 , Ω, y) = lim φε (g1 (x) − y) det Dg (x) dx
ε→0 Ω
Z
= lim φε (g (x) − y1 ) det Dg (x) dx
ε→0 Ω
P© ª
which by Lemma 10.2.5 equals sgn (det Dg (x)) : x ∈ g−1 (y1 ) , an integer.
What about the continuity assertion and being constant on connected components?
Let U be a connected component of Rn \ f (∂Ω) and let y, y1 be two points of U. The
argument is similar to some of the above. Let y ∈ Rn \ f (∂Ω) and suppose y1 is very
close to y, close enough that for

f1 (x) ≡ f (x) + y − y1

it follows
y∈
/ tf + (1 − t) f1 (∂Ω)
Then from Lemma 10.2.5
Z
d (f , Ω, y) = d (f1 , Ω, y) = lim φε (f1 (x) − y) Df (x) dx
ε→0 Ω
Z
= lim φε (f (x) − y1 ) Df (x) dx ≡ d (f , Ω, y1 )
ε→0 Ω

which shows that y → d (f , Ω, 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 Definition Of The Degree For Continuous Functions

With the above results, it is now possible to extend the definition of the degree to con-
tinuous functions which have no differentiability. It is desired to preserve the homotopy
invariance. This requires the following definition.

Definition 10.2.8 Let f ∈ Uy where y ∈ Rn \ f (∂Ω) . Then

d (f , Ω, y) ≡ f (g, Ω, y)
¡ ¢
where g ∈ Uy ∩ C 2 Ω; Rn and f ∼ g.
10.2. DEFINITIONS AND ELEMENTARY PROPERTIES 267

Theorem 10.2.9 The definition of the degree given in Definition 10.2.8 is well
defined, equals an integer, and satisfies the following properties. In what follows, id (x) =
x.

1. d (id, Ω, y) = 1 if y ∈ Ω.
¡ ¢
2. If Ωi ⊆ Ω, Ωi open, and Ω1 ∩Ω2 = ∅ and if y ∈
/ f Ω \ (Ω1 ∪ Ω2 ) , then d (f , Ω1 , y)+
d (f , Ω2 , y) = d (f , Ω, y).
¡ ¢
3. If y ∈ / f Ω \ Ω1 and Ω1 is an open subset of Ω, then

d (f , Ω, y) = d (f , Ω1 , y) .

4. For y ∈ Rn \ f (∂Ω) , if d (f , Ω, y) 6= 0 then f −1 (y) ∩ Ω 6= ∅.

5. If f , g are homotopic with a homotopy, h : Ω × [0, 1] for which h (∂Ω, t) does not
contain y, then d (g,Ω, y) = d (f , Ω, y).
6. d (·, Ω, y) is defined and constant on
© ¡ ¢ ª
g ∈ C Ω; Rn : ||g − f ||∞ < r

where r = dist (y, f (∂Ω)).

7. d (f , Ω, ·) is constant on every connected component of Rn \ f (∂Ω).
8. d (g,Ω, y) = d (f , Ω, y) if g|∂Ω = f |∂Ω .

Proof: First it is necessary to show the definition is well defined. 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 doesn’t matter which g I happen to pick. The first part is easy.
Let δ be small enough that
B (y, δ) ∩ f (∂Ω) = ∅.
¡ ¢
Then by Lemma 10.2.3 there exists g ∈ C 2 Ω; Rn such that ||g − f ||∞ < δ. It follows
that for t ∈ [0, 1] ,
y∈/ (tg + (1 − t) f ) (∂Ω)
and so g ∼ f . This does the first part. Now consider the second part. Suppose g ∼ f
and g1 ∼ f . Then by Lemma 10.2.3 again

g ∼ g1

and by Theorem 10.2.7 it follows d (g,Ω, y) = d (g1 ,Ω, y) which shows the definition is
well defined. Also d (f ,Ω, y) must be an integer because it equals d (g,Ω, y) which is an
integer.
Now consider the properties. The first 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 (∂Ω) ∪ f (∂Ω1 ) ∪ f (∂Ω2 )

2
¡ ¢
Let g ∈ C Ω; Rn such that ||f − g||∞ is small enough that
¡ ¢
y∈/ g Ω \ (Ω1 ∪ Ω2 ) (10.9)

and also small enough that

y ∈
/ (tg + (1 − t) f ) (∂Ω) , y ∈
/ (tg + (1 − t) f ) (∂Ω1 )
y ∈
/ (tg + (1 − t) f ) (∂Ω2 ) (10.10)
268 BROUWER DEGREE

for all t ∈ [0, 1] . Then it follows from Lemma 10.2.5, for all ε small enough,
Z
d (g, Ω, y) = φε (g (x) − y) Dg (x) dx
Ω

From 10.9 there is a positive distance between the compact set

¡ ¢
g Ω \ (Ω1 ∪ Ω2 )

and y. Therefore, making ε still smaller if necessary,

φε (g (x) − y) = 0 if x ∈
/ Ω1 ∪ Ω2

Therefore, using the definition of the degree and 10.10,

Z
d (f , Ω, y) = d (g, Ω, y) = lim φε (g (x) − y) Dg (x) dx
ε→0 Ω

µZ
= lim φε (g (x) − y) Dg (x) dx+
ε→0 Ω1
Z ¶
φε (g (x) − y) Dg (x) dx
Ω2
= d (g, Ω1 , y) + d (g, Ω2 , y)
= d (f , Ω1 , y) + d (f , Ω2 , y)

This proves the second property.

Consider the third. This really follows from the second property. You can take
Ω2 = ∅. 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) ∩ Ω = ∅. I will show this requires
¡ ¢
d (f , Ω, y) = 0. It is assumed y ∈/ f (∂Ω) and so if f −1 (y) ∩ Ω = ∅, then y ∈ / f Ω .
¡ ¢
Choosing g ∈ C 2 Ω; Rn such that ||f − g||∞ is sufficiently small, it can be assumed
¡ ¢
y∈ /g Ω , y∈ / (tf + (1 − t) g) (∂Ω) for all t ∈ [0, 1] .

Then it follows from the definition of the degree

Z
d (f , Ω, y) = d (g, Ω, y) ≡ lim φε (g (x) − y) Dg (x) dx = 0
ε→0 Ω
¡ ¢
because eventually ε is smaller than the distance from y to g Ω and so φε (g (x) − y) =
0 for all x ∈ Ω.
Property 5 follows from the definition of the degree.
Consider the sixth property. If g is in the specified set, and t ∈ [0, 1] with x ∈ ∂Ω

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 definition of the degree

d (f , Ω, y) = d (g, Ω, y)
10.3. BORSUK’S THEOREM 269
¡ ¢
The seventh claim is done already for the case where f ∈ C 2 Ω; Rn 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 ,Ω, y) is continuous. Let y0 ∈ Rn \ f (∂Ω) and let δ be small
enough that
B (y0 , 4δ) ∩ f (∂Ω) = ∅.
2
¡ n
¢
Now let g ∈ C Ω; R such that ||g − f ||∞ < δ. Then for x ∈ ∂Ω, t ∈ [0, 1] , and
y ∈ B (y0 , δ) ,

|(tg+ (1 − t) f ) (x) − y| ≥ |f (x) − y| − t |g (x) − f (x)|

≥ |f (x) − y0 | − |y0 − y| − ||g − f ||∞

≥ 4δ − δ − δ > 0.

Therefore, for all such y ∈ B (y0 , δ)

d (f , Ω, y) = d (g, Ω, y)

and it was shown in Theorem 10.2.7 that y → d (g, Ω, y) is continuous. In particular

d (f , Ω, ·) is continuous at y0 . Since y0 was arbitrary, this shows y → d (f , Ω, y) is
continuous. Therefore, since it has integer values, this function is constant on every
connected component of Rn \ f (∂Ω) by Corollary 5.3.15.
Consider the eighth claim about the degree in which f = g on ∂Ω. This one is easy
because for
y ∈ Rn \ f (∂Ω) = Rn \ g (∂Ω) ,
and x ∈ ∂Ω,
tf (x) + (1 − t) g (x) − y = f (x) − y 6= 0
for all t ∈ [0, 1] and so by the fifth claim, d (f , Ω, y) = d (g, Ω, y) and This proves the
theorem. ¥

10.3 Borsuk’s Theorem

In this section is an important theorem which can be used to verify that d (f , Ω, y) 6= 0.
This is significant because when this is known, it follows from Theorem 10.2.9 that
f −1 (y) 6= ∅. In other words there exists x ∈ Ω such that f (x) = y.

Definition 10.3.1 A bounded open set, Ω is symmetric if −Ω = Ω. A contin-

uous function, f : Ω → Rn is odd if f (−x) = −f (x).
¡ ¢
Suppose Ω is symmetric and g ∈ C 2 Ω; Rn is an odd map for which 0 is a regular
value. Then the chain rule implies Dg (−x) = Dg (x) and so d (g, Ω, 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 defined on
symmetric Ω. 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 Ω; Rn be an odd map. Then for every ε > 0, there
¢
2 n
exists h ∈ C Ω; R such that h is also an odd map, ||h − g||∞ < ε, and 0 is a regular
value of h.

Proof: In this argument η > 0 will be a small positive number and C will be a
constant which depends only on the diameter of Ω. Let h0 (x) = g (x) + ηx where η is
chosen such that det Dh0 (0) 6= 0. Now let Ωi ≡ {x ∈ Ω : xi 6= 0}. In other words, leave
out the plane xi = 0 from Ω in order to obtain Ωi . A succession of modifications is
about to take place on Ω1 , Ω1 ∪ Ω2 , etc. Finally a function will be obtained on ∪nj=1 Ωj
which is everything except 0. ¯ ¯ ¡ ¢
Define h1 (x) ≡ h0 (x)−y1 x31 where ¯y1 ¯ < η and y1 = y11 , · · ·, yn1 is a regular value
of the function, x → h0x(x)
3 for x ∈ Ω1 . the existence of y1 follows from Sard’s lemma
1
h0 (x)
because this function is in C 2 ( Ω1 ; Rn ). Thus h1 (x) = 0 if and only if y1 = x31
.
Since y1 is a regular value, it follows that for such x,
à ∂
¡ 3¢ !
h0i,j (x) x31 − ∂x j
x1 h0i (x)
det =
x61
à ∂
¡ 3¢ 1 3 !
h0i,j (x) x31 − ∂xj x1 yi x1
det 6= 0
x61
implying that
µ ¶
∂ ¡ 3¢ 1
det h0i,j (x) − x y = det (Dh1 (x)) 6= 0.
∂xj 1 i
This
¡ shows¢ 0 is a regular value of h1 on the set Ω1 and it is clear h1 is an odd map in
C 2 Ω; Rn and ||h1 − g||∞ ≤ Cη where C depends only on the diameter of Ω.
Now
¡ suppose
¢ for some k such that 1 ≤ k < n there exists an odd mapping hk
in C 2 Ω; Rn such that 0 is a regular value of hk on ∪ki=1 Ωi and ||hk − g||∞ ≤ Cη.
k+1
Sard’s theorem implies there
¯¯ exists
¯¯ y a regular value of the function x → hk (x) /x3k+1
defined on Ωk+1 such that y¯ ¯ k+1 ¯¯
< η and let hk+1 (x) ≡ hk (x)−yk+1 x3k+1 . As before,
hk+1 (x) = 0 if and only if hk (x) /x3k+1 = yk+1 , a regular value of x → hk (x) /x3k+1 .
Consider such x for which hk+1 (x) = 0. First suppose x ∈ Ωk+1 . Then
à ∂
¡ 3 ¢ k+1 3 !
hki,j (x) x3k+1 − ∂x j
xk+1 yi xk+1
det 6 6= 0
xk+1

which implies that whenever hk+1 (x) = 0 and x ∈ Ωk+1 ,

µ ¶
∂ ¡ 3 ¢ k+1
det hki,j (x) − x y = det (Dhk+1 (x)) 6= 0. (10.11)
∂xj k+1 i

However, if x ∈ ∪ki=1 Ωk but x ∈ / Ωk+1 , then xk+1 = 0 and so the left side of 10.11
reduces to det (hki,j (x)) which is not zero because 0 is assumed a regular value of hk .
Therefore, 0 is a regular value for hk+1 on ∪k+1 k+1
i=1 Ωk . (For x ∈ ∪i=1 Ωk , either x ∈ Ωk+1 or
x∈/ Ωk+1 . If x ∈ Ωk+1 0 is a regular value by the construction above. In the other ¡ case,
¢
0 is a regular value by the induction hypothesis.) Also hk+1 is odd and in C 2 Ω; Rn ,
and ||hk+1 − g||∞ ≤ Cη.
Let h ≡ hn . Then 0 is a regular value of h for x ∈ ∪nj=1 Ωj . The point of Ω which is
not in ∪nj=1 Ωj is 0. If x = 0, then from the construction, Dh (0) = Dh0 (0) and so 0 is
a regular value of h for x ∈ Ω. By choosing η small enough, it follows ||h − g||∞ < ε.
This proves the lemma. ¥
10.4. APPLICATIONS 271
¡ ¢
Theorem 10.3.3 (Borsuk) Let f ∈ C Ω; Rn be odd and let Ω be symmetric
with 0 ∈
/ f (∂Ω). Then d (f , Ω, 0) equals an odd integer.

Proof: Let δ be small enough that B (0,3δ) ∩ f (∂Ω) = ∅. Let

¡ ¢
g1 ∈ C 2 Ω; Rn

be such that ||f − g1 ||∞ < δ and let g denote the odd part of g1 . Thus

1
g (x) ≡ (g1 (x) − g1 (−x)) .
2

Then since f is odd, it follows that for x ∈ Ω,

|f (x) − g (x)| =
¯ ¯
¯1 ¯
¯ (f (x) − f (−x)) − 1 (g1 (x) − g1 (−x))¯
¯2 2 ¯
1 1
≤ |f (x) − g1 (x)| + |f (−x) − g1 (−x)| < δ
2 2
¡ ¢
Thus ||f − g||∞ < δ also. By Lemma 10.3.2 there exists odd h ∈ C 2 Ω; Rn for which
0 is a regular value and ||h − g||∞ < δ. Therefore,

||f − h||∞ ≤ ||f − g||∞ + ||g − h||∞ < 2δ

and so for t ∈ [0, 1] and x ∈ ∂Ω,

|th (x) + (1 − t) f (x) − 0| ≥ |f (x) − 0| − t |h (x) − f (x)|

≥ 3δ − δ > 0

and so, from the from the definition of the degree, d (f , Ω, 0) = d (h, Ω, 0).
m
Since 0 is a regular point of h, h−1 (0) = {xi , −xi , 0}i=1 , and since h is odd,
Dh (−xi ) = Dh (xi ) and so
m
X m
X
d (h, Ω, 0) ≡ sgn det (Dh (xi )) + sgn det (Dh (−xi )) + sgn det (Dh (0)) ,
i=1 i=1

an odd integer. ¥

10.4 Applications
With these theorems it is possible to give easy proofs of some very important and
difficult theorems.

Definition 10.4.1 If f : U ⊆ Rn → Rn where U is an open set. Then f is

locally one to one if for every x ∈ U , there exists δ > 0 such that f is one to one on
B (x, δ).

As a first 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) → Rn be one to one and continuous where here
B (0,r) is the ball centered at 0 of radius r in Rn . Then there exists δ > 0 such that
g (0) + B (0, δ) ⊆ g (B (0,r)) .
The symbol on the left means: {g (0) + x : x ∈ B (0, δ)} .
Proof: For t ∈ [0, 1] , let
µ ¶ µ ¶
x −tx
h (x, t) ≡ g −g
1+t 1+t
Then for x ∈ ∂B (0, r) , h (0, t) 6= 0 because if this were so, the fact g is one to one
implies
x −tx
=
1+t 1+t
and this requires x = 0 which is not the case. Since ∂B (0,r) × [0, 1] is compact, there
exists δ > 0 such that for all t ∈ [0, 1] and x ∈ ∂B (0,r) ,
B (0, δ) ∩ h (x,t) = ∅.
In particular, when t = 0, B (0,δ) is contained in a single component of
Rn \ (g − g (0)) (∂B (0,r))
and so
d (g − g (0) , B (0,r) , z)
is constant for z ∈ B (0,δ) . 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) 6= 0
the last assertion following from Borsuk’s 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, δ) there
exists x ∈ B (0,r) such that
z = g (x) − g (0)
which shows g (0) + B (0, δ) ⊆ 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 Ω be any open subset of Rn and
n
let f : Ω → R be continuous and locally one to one. Then f maps open subsets of Ω to
open sets in Rn .
Proof: Let B (x0 , r) ⊆ U ⊆ Ω where f is one to one on B (x0 , r) and U is an open
subset of Ω. Let g be defined on B (0, r) given by
g (x) ≡ f (x + x0 )
Then g satisfies the conditions of Lemma 10.4.2, being one to one and continuous. It
follows from that lemma there exists δ > 0 such that
f (U ) ⊇ f (B (x0 , r)) = f (x0 + B (0, r))
= g (B (0,r)) ⊇ g (0) + B (0, δ)
= f (x0 ) + B (0,δ) = B (f (x0 ) , δ)
This shows that for any x0 ∈ U, f (x0 ) 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
Rn to Rm .

Proof: Suppose not and let f be such a continuous map,

T
f (x) ≡ (f1 (x) , · · ·, fm (x)) .
T
Then let g (x) ≡ (f1 (x) , · · ·, fm (x) , 0, · · ·, 0) where there are n − m zeros added in.
Then g is a one to one continuous map from Rn to Rn and so g (Rn ) 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 : Rn → Rn , and

lim |f (x)| = ∞,
|x|→∞

then f maps Rn onto Rn .

Proof: By the invariance of domain theorem, f (Rn ) is an open set. It is also true
that f (Rn ) is a closed set. Here is why. If f (xk ) → y, the growth condition ensures
that {xk } is a bounded sequence. Taking a subsequence which converges to x ∈ Rn
and using the continuity of f , it follows f (x) = y. Thus f (Rn ) is both open and closed
which implies f must be an onto map since otherwise, Rn would not be connected. ¥
The next theorem is the famous Brouwer fixed point theorem.

Theorem 10.4.6 (Brouwer fixed point) Let B = B (0, r) ⊆ Rn 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 fixed point
in B for f , it follows that 0 ∈/ h (∂B, t) for all t. When t = 1, this follows from having
no fixed 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,
n
0 = d (f − id, B, 0) = d (− id, B, 0) = (−1) ,

a contradiction. This proves the theorem. ¥

Definition
³ ´ 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 ∂Ω,
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 fixed point
theorem.
The proofs of the next two theorems make use of the Tietze extension theorem,
Theorem 5.7.9.

Theorem 10.4.9 Let Ω be a symmetric open set in Rn such that 0 ∈ Ω and let
f : ∂Ω → V be continuous where V is an m dimensional subspace of Rn , m < n. Then
f (−x) = f (x) for some x ∈ ∂Ω.

¡ Proof:
¢ Suppose not. Using the Tietze extension theorem, extend f to all of Ω,
f Ω ⊆ V . (Here the extended function is also denoted by f .) Let g (x) = f (x)−f (−x).
/ g (∂Ω) and so for some r > 0, B (0,r) ⊆ Rn \ g (∂Ω). For z ∈ B (0,r),
Then 0 ∈

d (g, Ω, z) = d (g, Ω, 0) 6= 0

because B (0,r) is contained in a component of Rn \g (∂Ω) and Borsuk’s theorem implies

that d (g, Ω, 0) 6= 0 since g is odd. Hence

V ⊇ g (Ω) ⊇ 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 Ω be an open bounded set in Rn with
0 ∈ Ω. Suppose f : ∂Ω → Rn \ {0} is continuous. Then for some x ∈ ∂Ω and λ 6= 0,
f (x) = λx.

Proof: Using the Tietze extension theorem, extend f to all of Ω. Also denote the
extended function by f . Suppose for all x ∈ ∂Ω, f (x) 6= λx for all λ ∈ R. Then

0∈
/ tf (x) + (1 − t) x, (x, t) ∈ ∂Ω × [0, 1]

0∈
/ tf (x) − (1 − t) x, (x, t) ∈ ∂Ω × [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 , Ω, 0) = d (id, Ω, 0) , d (f , Ω, 0) = d (− id, Ω, 0) .
n
But this is impossible because d (id, Ω, 0) = 1 but d (− id, Ω, 0) = (−1) = −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 y1 , · · ·,¯y¯ r be ¯points

¯ not in f (∂Ω) and let δ > 0. Then there
¡ ¢ ¯¯e ¯¯
e
exists f ∈ C Ω; R such that ¯¯f − f ¯¯ < δ and yi is a regular value for e
2 n
f for each i.
∞
10.5. THE PRODUCT FORMULA 275
¡ ¢
Proof: Let f0 ∈ C 2 Ω; Rn , ||f0 − f ||∞ < 2δ . Let ye1 be a regular value for f0 and
δ
|e e1 . Thus y1 is a regular value of f1 because
y1 − y1 | < 3r . Let f1 (x) ≡ f0 (x) + y1 − y
e1 where y
Df1 (x) = Df0 (x) and if f1 (x) = y1 , this is the same as having f0 (x) = y e1 is
a regular value of f0 . Then also

||f − f 1 ||∞ ≤
||f − f 0 ||∞ + ||f0 − f1 ||∞
=
||f − f 0 ||∞ + |e
y1 − y1 |
δ δ
< + .
3r 2
¡ ¢
Suppose now there exists fk ∈ C 2 Ω; Rn with each of the yi for i = 1, · · ·, k a regular
value of fk and
µ ¶
δ k δ
||f − f k ||∞ < + .
2 r 3
Then letting Sk denote the singular values of fk , Sard’s theorem implies there exists
ek+1 such that
y
δ
|e
yk+1 − yk+1 | <
3r
and
y / Sk ∪ ∪ki=1 (Sk + yk+1 − yi ) .
ek+1 ∈ (10.12)

Let
ek+1 .
fk+1 (x) ≡ fk (x) + yk+1 − y (10.13)

If fk+1 (x) = yi for some i ≤ k, then

ek+1
fk (x) + yk+1 − yi = y

ek+1 ∈
and so fk (x) is a regular value for fk since by 10.12, y / Sk + yk+1 − yi and
so fk (x) ∈
/ Sk . Therefore, for i ≤ k, yi is a regular value of fk+1 since by 10.13,
Dfk+1 = Dfk . Now suppose fk+1 (x) = yk+1 . Then

ek+1
yk+1 = fk (x) + yk+1 − y

ek+1 implying that fk (x) = y

so fk (x) = y ek+1 ∈/ Sk . Hence det Dfk+1 (x) = det Dfk (x) 6=
0. Thus yk+1 is also a regular value of fk+1 . Also,

||fk+1 − f || ≤ ||fk+1 − fk || + ||fk − f ||

µ ¶ µ ¶
δ δ k δ δ k+1 δ
≤ + + = + .
3r 2 r 3 2 r 3

Let e
f ≡ fr . Then
¯¯ ¯¯ µ ¶
¯¯e ¯¯ δ δ
¯¯f − f ¯¯ < + <δ
∞ 2 3

and each of the yi is a regular value of e

f . This proves the lemma. ¥

Definition 10.5.2 Let the connected components of Rn \ f (∂Ω) be denoted by

Ki . From the properties of the degree listed in Theorem 10.2.9, d (f , Ω, ·) is constant on
each of these components. Denote by d (f , Ω, Ki ) the constant value on the component,
Ki .
276 BROUWER DEGREE

The product formula considers the situation depicted in the following diagram in
/ g (f (∂Ω)) and the Ki are the connected components of Rn \ f (∂Ω).
which y ∈
f ¡ ¢ g
Ω → f Ω → Rn
y
Rn \f (∂Ω)=∪i Ki

The following diagram may be helpful in remembering what it says.

K3

f g
Ω - K1 - y

K2

¡ ¢
Lemma 10.5.3 Let f ∈ C Ω; Rn , g ∈ C 2 (Rn , Rn ) , and y ∈/ g (f (∂Ω)). Suppose
also that y is a regular value of g. Then the following product formula holds where Ki
are the bounded components of Rn \ f (∂Ω).
∞
X
d (g ◦ f , Ω, y) = d (f , Ω, Ki ) d (g, Ki , y) .
i=1

All but finitely many terms in the sum are zero.

Proof: First note that if Ki is unbounded, d (f , Ω, Ki ) = ¡0 because

¢ there exists
a point, z ∈ Ki such that f −1 (z) = ∅ due to the fact that f Ω is compact and is
consequently bounded. Thus it makes no difference in the above formula
© ªmi whether the
Ki are arbitrary components or only bounded components. Let xij j=1 denote the
points of g−1 (y) which are contained in Ki , the ith bounded component of Rn \ f (∂Ω).
Then mi < ∞ because if not, there would exist a limit point x for this sequence. Then
g (x) = y and so x ∈ / f (∂Ω). Thus det (Dg (x)) 6= 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 Ω is a compact set covered by the components of Rn \
¡ ¢
f (∂Ω) because g−1 (y) ∩ f (∂Ω) = ∅. It follows g−1 (y) ∩ f Ω is covered by finitely
many of these components.
The only terms in the above sum which are¡nonzero¢ are those corresponding to Ki
having nonempty intersection with g−1 (y) ∩ f Ω . The other components contribute
0 to the above sum because if Ki ∩ g−1 (y)¡ =¢∅, it follows from Theorem 10.2.9 that
d (g, Ki , y) = 0. If Ki does not intersect f Ω , then d (f , Ω, Ki ) = 0. Therefore, the
¡ ¢
above sum is actually a finite sum since g−1 (y) ∩ f Ω , being a compact set, is covered
by finitely many of the Ki . Thus there are no convergence problems. Now let ε > 0 be
small enough that
B (y, 3ε) ∩ g (f (∂Ω)) = ∅,
and for each xij ∈ g−1 (y)
¡ ¢
B xij , 3ε ∩ f (∂Ω) = ∅.
10.5. THE PRODUCT FORMULA 277
¡ ¢
By uniform continuity of g on the compact set f Ω , there exists δ > 0, δ < ε such that if
¡ ¢
z1 and z2 are any two points of f Ω with |z1 − z2 | < δ, it follows that |g (z1 ) − g (z2 )| <
ε. ¡ ¢ ¯¯ ¯¯
¯¯ ¯¯
Now using Lemma 10.5.1, choose e f ∈ C 2 Ω; Rn such that ¯¯e f − f ¯¯ < δ and each
∞
point, xij is a regular value of e
f . From the properties of the degree in Theorem 10.2.9
¡ ¢
d (f , Ω, Ki ) = d f , Ω, xij

for each j = 1, · · ·, mi . For x ∈ ∂Ω, and t ∈ [0, 1],

¯ ³ ´ ¯ ¯ ¯ ¯ ¯
¯ ¯ ¯ ¯
¯f (x) + t e f (x) − f (x) − xij ¯ ≥ ¯f (x) − xij ¯ − t ¯ef (x) − f (x)¯
> 3ε − tε > 0

and so by the homotopy invariance of the degree,

³ ´ ¡ ¢
d e f , Ω, xij = d f , Ω, xij ≡ d (f , Ω, Ki ) (10.14)

independent of j. Also for x ∈ ∂Ω, and t ∈ [0, 1],

¯ ³ ³ ´ ´ ¯
¯ ¯
¯g (f (x)) + t g ef (x) − g (f (x)) − y¯ ≥ 3ε − tε > 0

and so by the homotopy invariance of the degree,

³ ´
d (g ◦ f , Ω, y) = d g ◦ e
f , Ω, y . (10.15)
¡ ¢ ¡ ¢
Now e f −1 xij is a finite set because e f −1 xij ⊆ Ω, a bounded open set and xij is a
regular value. It follows from 10.15
³ ´
d (g ◦ f , Ω, y) = d g ◦ e
f , Ω, y

 xij

X mi
∞ X X  z}|{ 
= sgn det Dg e  e
f (z) sgn det Df (z)
i=1 j=1 z∈e
f −1 (xij )

X mi
∞ X
¡ ¢ ³ ´ X∞ ³ ´
= sgn det Dg xij d e
f , Ω, xij = d (g, Ki , y) d ef , Ω, xij
i=1 j=1 i=1

∞
X
= d (g, Ki , y) d (f , Ω, Ki ) .
i=1

This proves the lemma. ¥

With this lemma, the following is the product formula.

Theorem 10.5.4
¡ (product
¢
∞
formula) Let {Ki }i=1 be the bounded components of
R \ f (∂Ω) for f ∈ C Ω; Rn , let g ∈ C (Rn , Rn ), and suppose that y ∈
n
/ g (f (∂Ω)).
Then
∞
X
d (g ◦ f , Ω, y) = d (g, Ki , y) d (f , Ω, Ki ) . (10.16)
i=1

All but finitely many terms in the sum are zero.

278 BROUWER DEGREE

Proof: Let B (y,3δ) ∩ g (f (∂Ω)) = ∅ and let g e ∈ C 2 (Rn , Rn ) be such that

© ¡ ¢ª
sup |e g (z) − g (z)| : z ∈ f Ω < δ

e. Then from the above inequality, if x ∈ ∂Ω and

And also y is a regular value of g
t ∈ [0, 1],

|g (f (x)) + t (e
g (f (x)) − g (f (x))) − y| ≥ |g (f (x)) − y| − t |e
g (f (x)) − g (f (x))|
≥ 3δ − tδ > 0.

It follows that
d (g ◦ f , Ω, y) = d (e
g ◦ f , Ω, y) . (10.17)
Now also, ∂Ki ⊆ f (∂Ω) and so if z ∈ ∂Ki , then g (z) ∈ g (f (∂Ω)). Consequently, for
such z,
|g (z) + t (e
g (z) − g (z)) − y| ≥ |g (z) − y| − tδ > 3δ − tδ > 0
which shows that, by homotopy invariance,

d (g, Ki , y) = d (e
g, Ki , y) . (10.18)

Therefore, by Lemma 10.5.3,

∞
X
d (g ◦ f , Ω, y) = d (e
g ◦ f , Ω, y) = d (e
g, Ki , y) d (f , Ω, Ki )
i=1
∞
X
= d (g, Ki , y) d (f , Ω, Ki )
i=1

and the sum has only finitely many non zero terms. This proves the product formula.
Note there are no convergence problems because ¡ ¢these sums are actually finite sums
because, as in the previous lemma, g−1 (y) ∩ f Ω is a compact set covered by the
components of Rn \ f (∂Ω) and so it is covered by finitely many of these components.
For the other components, d (f , Ω, Ki ) = 0 or else d (g, Ki , 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 Ω be a bounded open set in Rn , f ∈ C Ω; Rn , and suppose
∞
{Ωi }i=1 are disjoint open sets contained in Ω such that
¡ ¢
y∈/ f Ω \ ∪∞ j=1 Ωj

Then
∞
X
d (f , Ω, y) = d (f , Ωj , y)
j=1

where the sum has all but finitely many terms equal to 0.

Proof: By assumption, the compact set f −1 (y) has empty intersection with

Ω \ ∪∞
j=1 Ωj

and so this compact set is covered by finitely many of the Ωj , say {Ω1 , · · · , Ωn−1 } and
¡ ¢
y∈/ f ∪∞ j=n Ωj
10.6. JORDAN SEPARATION THEOREM 279

By Theorem 10.2.9 and letting O = ∪∞

j=n Ωj ,

n−1
X ∞
X
d (f , Ω, y) = d (f , Ωj , y) + d (f ,O, y) = d (f , Ωj , y)
j=1 j=1

because d (f ,O, y) = 0 as is d (f , Ωj , y) for every j ≥ n. This proves the lemma. ¥

Lemma 10.5.6 Define ∂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 Rn and let K denote the set of components of Rn \ 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 Rn \ C? This is obvious because in Rn 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 Rn \ C, why is ∂K ⊆ C? Let x ∈ ∂K. If x ∈ / C, then
x ∈ K1 , some component of Rn \ C. If K1 6= K then x cannot be a limit point of K
and so it cannot be in ∂K. Therefore, K = K1 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 Rn . Then Rn \ C and Rn \ f (C) have the same
number of connected components.

Proof : Denote by K the bounded components of Rn \ C and denote by L, the

bounded components of Rn \ f (C). Also, using the Tietze extension theorem, there
exists f an extension of f to all of Rn which maps into a bounded set and let f −1 be an
extension of f −1 to all of Rn 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 (C)
z }| {
f −1 f (∂K) = f −1 (f (∂K)) ⊆ C
280 BROUWER DEGREE

/ C. Since f −1 ◦ f equals the identity id on ∂K, it follows from the properties of

and y ∈
the degree that ³ ´
1 = d (id, K, y) = d f −1 ◦ f , K, y .

Let H denote the set of bounded components of Rn \ f (∂K). By the product formula,
³ ´ X ¡ ¢ ³ ´
1 = d f −1 ◦ f , K, y = d f , K, H d f −1 , H, y , (10.19)
H∈H

the sum being a finite sum from the product formula. It might help to consult the
following diagram.
f
→
Rn \ C Rn \ f (C)
f −1
←
K L
K Rn \ f (∂K)
y∈K H, H1
H
LH
Now letting x ∈ L ∈ L, if S is a connected set containing x and contained in Rn \ f (C),
then it follows S is contained in Rn \ 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
³ C
´
L = (L ∩ H) ∪ (L ∩ ∂H) ∪ L ∩ H .

Now by Lemma 10.5.6,

L ∩ ∂H ⊆ L ∩ f (∂K) ⊆ L ∩ f (C) = ∅.
C
Since L is connected, L∩H = ∅. Letting LH denote those sets of L which are contained
in H equivalently having nonempty intersection with H, if p ∈ H \ ∪LH = H \ ∪L,
then p ∈ H ∩ f (C) and so

H = (∪LH ) ∪ (H ∩ f (C)) (10.20)

Claim 1:
H \ ∪LH ⊆ f (C) .

Proof of the claim: Suppose p ∈ H \ ∪LH 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∈ / ∪LH , it follows p ∈ f¡ (C) after¢all. This proves the claim.
Claim 2: y ∈ / f −1 H \ ∪LH . Recall y ∈ K ∈ K the bounded components of
Rn \ C.
Proof of the claim: If not, then f −1 (z) = y where z ∈ H \ ∪LH ⊆ 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 Rn \ 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 LH = ∅? From 10.20 it follows H ⊆ f (C). Therefore,
³ ´ ¡ ¢
d f −1 , H, y = d f −1 , H, y = 0
10.6. JORDAN SEPARATION THEOREM 281

because y ∈ K a component of Rn \ C. Therefore, letting H1 denote those sets of H

which contain some set of L, 10.19 is of the form
X ¡ ¢ ³ ´
1= d f , K, H d f −1 , H, y .
H∈H1

and it is still a finite sum because

³ the terms
´ in the sum are 0 for all but finitely many
−1
H ∈ H1 . I want to expand d f , H, y as a sum of the form
X ³ ´
d f −1 , L, y
L∈LH

using Lemma 10.5.5. Therefore, I must verify

¡ ¢
y∈/ f −1 H \ ∪LH

but
³ this is just
´ Claim 2. By Lemma 10.5.5, I can write the above sum in place of
−1
d f , H, y . Therefore,
X ¡ ¢ ³ ´ X ¡ ¢ X ³ ´
1= d f , K, H d f −1 , H, y = d f , K, H d f −1 , L, y
H∈H1 H∈H1 L∈LH

where there are only finitely many H which give a nonzero

³ term´ and for each of these,
−1
there are only finitely many L in LH which yield d f , L, y 6= 0. Now the above
equals X X ¡ ¢ ³ ´
= d f , K, H d f −1 , L, y . (10.21)
H∈H1 L∈LH

By definition, ¡ ¢ ¡ ¢
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 ∈ LH .
Therefore, the above reduces to
X ¡ ¢ ³ ´
= d f , K, L d f −1 , L, y (10.22)
L∈L

Here is why. There are finitely many H ∈ H1 for which the term in the double sum of
10.21 is not zero, say H1 , · · · , Hm . Then the above sum in 10.22 equals
m
X X ¡ ¢ ³ ´ X ¡ ¢ ³ ´
d f , K, L d f −1 , L, y + d f , K, L d f −1 , L, y
k=1 L∈LHk L\∪m
k=1 LHk

The second sum equals 0 because those L are contained in some H ∈ H for which
¡ ¢ ³ ´ ¡ ¢ X ³ ´
0 = d f , K, H d f −1 , H, y = d f , K, H d f −1 , L, y
L∈LH
X ¡ ¢ ³ ´
= d f , K, L d f −1 , L, y .
L∈LH

Therefore, the sum in 10.22 reduces to

m
X X ¡ ¢ ³ ´
d f , K, L d f −1 , L, y
k=1 L∈LHk
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 X ¡ ¢ ³ ´
= d f , K, L d f −1 , L, K
L∈L

and all but finitely many terms in the sum are 0.

By the same argument,
X ¡ ¢ ³ ´
1= d f , K, L d f −1 , L, K
K∈K

and all but finitely many terms in the sum are 0. Letting |K| denote the number of
elements in K, similar for L,
à !
X X X ¡ ¢ ³ ´
|K| = 1= d f , K, L d f −1 , L, K
K∈K K∈K L∈L
à !
X X X ¡ ¢ ³ ´
|L| = 1= d f , K, L d f −1 , L, K
L∈L L∈L K∈K

Suppose |K| < ∞. Then you can switch the order of summation in the double sum for
|K| and so
à !
X X ¡ ¢ ³ ´
|K| = −1
d f , K, L d f , L, K
K∈K L∈L
à !
X X ¡ ¢ ³ ´
= d f , K, L d f −1 , L, K = |L|
L∈L K∈K

It follows that if either |K| or |L| is finite, then they are equal. Thus if one is infinite, so
is the other. This proves the theorem because if n > 1 there is exactly one unbounded
component to both Rn \ C and Rn \ 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 , Ω, f (x))
in the case where f is one to one and Ω 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 Ω be an open connected bounded set in Rn , n ≥¡1 such¢

that Rn \ ∂Ω consists of two, three if n = 1, connected components. Let f ∈ C Ω; Rn
be continuous and one to one. Then f (Ω) is the bounded component of Rn \ f (∂Ω) and
for y ∈ f (Ω) , d (f , Ω, y) either equals 1 or −1.

Proof: First suppose n ≥ 2. By the Jordan separation theorem, Rn \ f (∂Ω) consists

of two components, a bounded component B and an unbounded component U . Using ¡ ¢
the Tietze extention theorem, there exists g defined on Rn such that g = f −1 on f Ω .
Thus on ∂Ω, g ◦ f = id. It follows from this and the product formula that

1 = d (id, Ω, g (y)) = d (g ◦ f , Ω, g (y))

= d (g, B, g (y)) d (f , Ω, B) + d (f , Ω, U ) d (g, U, g (y))
= d (g, B, g (y)) d (f , Ω, B)

Therefore, d (f , Ω, B) 6= 0 and so for every z ∈ B, it follows z ∈ f (Ω) . Thus B ⊆ f (Ω) .

On the other hand, f (Ω) cannot have points in both U and B because it is a connected
10.7. INTEGRATION AND THE DEGREE 283

set. Therefore f (Ω) ⊆ B and this shows B = f (Ω). Thus d (f , Ω, B) = d (f , Ω, 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 R1 \ f (∂Ω) 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 :Rn → Rn is only C 1 (Rn ; Rn ),
vanishing outside a bounded set. In the following proposition, let ψ m be a symmetric
nonnegative mollifier,

ψ m (x) ≡ mn ψ (mx) , spt ψ ⊆ B (0, 1)

and let φε be a mollifier as ε → 0

µ ¶n ³ ´
1 x
φε (x) ≡ φ , spt φ ⊆ B (0, 1)
ε ε

Ω will be a bounded open set.

Proposition 10.7.1 Let S ⊆ h(∂Ω)C such that

dist (S, h (∂Ω)) > 0

where Ω is a bounded open set and also let h be C 1 (Rn ; Rn ), vanishing outside some
bounded set. Then there exists ε0 > 0 such that whenever 0 < ε < ε0
Z
d (h, Ω, y) = φε (h (x) − y) det Dh (x) dx
Ω

for all y ∈ S.

Proof: Let ε0 > 0 be small enough that for all y ∈ S,

B (y, 5ε0 ) ∩ h (∂Ω) = ∅.

¡ ¢
Now let ψ m be a mollifier as m → ∞ with support in B 0, m−1 and let

hm ≡ h∗ψ m .
∞
¡ ¢
Thus hm ∈ C Ω; Rn and hm converges uniformly to h while Dhm converges uni-
formly to Dh. Denote by ||·||∞ this norm defined by

||h||∞ ≡ sup {|h (x)| : x ∈ Rn }

½ ¯ ¯ ¾
¯ ∂hi (x) ¯
||Dh||∞ ≡ sup max ¯ ¯ ¯ :x∈R n
i,j ∂xj ¯
It is finite because it is given that h vanishes off a bounded set. Choose M such that
for m ≥ M ,
||hm − h||∞ < ε0 . (10.23)
2
¡ n
¢
Thus hm ∈ Uy ∩ C Ω; R for all y ∈ S where Uy is defined on Page 259 and consists
¡ ¢
of those functions f in C Ω; Rn for which y ∈ / f (∂Ω).
284 BROUWER DEGREE

For y ∈ S, let z ∈ B (y, ε) where ε < ε0 and suppose x ∈ ∂Ω, and k, m ≥ M . Then
for t ∈ [0, 1] ,

|(1 − t) hm (x) + hk (x) t − z| ≥ |hm (x) − z| − t |hk (x) − hm (x)|

> 2ε0 − t2ε0 ≥ 0

showing that for each y ∈ S, B (y, ε) ∩ ((1 − t) hm + thk ) (∂Ω) = ∅. By Lemma 10.2.5,
for all y ∈ S, Z
φε (hm (x) − y) det (Dhm (x)) dx =
Ω
Z
φε (hk (x) − y) det (Dhk (x)) dx (10.24)
Ω
for all k, m ≥ M . By this lemma again, which says that for small enough ε the integral
is constant and the definition of the degree in Definition 10.2.4,
Z
d (y,Ω, hm ) = φε (hm (x) − y) det (Dhm (x)) dx (10.25)
Ω

for all ε small enough. For x ∈ ∂Ω, y ∈ S, and t ∈ [0, 1],

|(1 − t) h (x) + hm (x) t − y| ≥ |h (x) − y| − t |h (x) − hm (x)|

> 3ε0 − t2ε0 > 0

and so by Theorem 10.2.9, the part about homotopy, for each y ∈ S,

d (y,Ω, h) = d (y,Ω, hm ) =
Z
φε (hm (x) − y) det (Dhm (x)) dx
Ω
whenever ε is small enough. Fix such an ε < ε0 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,
Z
d (y,Ω, h) = lim φε (hm (x) − y) det (Dhm (x)) dx
m→∞ Ω
Z
= φε (h (x) − y) det (Dh (x)) dx.
Ω

This proves the proposition. ¥

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 (Rn ; Rn ) and h vanishes off a bounded set. Let mn (A) =
0. Then h (A) also has measure zero.

Next is an interesting change of variables theorem. Let Ω be a bounded open set

with the property that ∂Ω has measure zero and let h be C 1 and vanish off a bounded
set. Then from Lemma 10.7.2,
³ h (∂Ω)
´ also has measure zero.
C
Now suppose f ∈ Cc h (∂Ω) . By compactness, there are finitely many compo-
C
nents of h (∂Ω) which have nonempty intersection with spt (f ). From the Proposition
above,
Z Z Z
f (y) d (y, Ω, h) dy = f (y) lim φε (h (x) − y) det Dh (x) dxdy
ε→0 Ω
10.7. INTEGRATION AND THE DEGREE 285

Actually, there exists an ε small enough that for all y ∈ spt (f ) ,

Z Z
lim φδ (h (x) − y) det Dh (x) dx = φε (h (x) − y) det Dh (x) dx
δ→0 Ω Ω
= d (y, Ω, h)
C
This is because spt (f ) is at a positive distance from the compact set h (∂Ω) so this
follows from Proposition 10.7.1. Therefore, for all ε small enough,
Z Z Z
f (y) d (y, Ω, h) dy = f (y) φε (h (x) − y) det Dh (x) dxdy
Z Ω Z
= det Dh (x) f (y) φε (h (x) − y) dydx
Ω
Z Z Z
= det Dh (x) f (h (x)) dx + det Dh (x) (f (y) − f (h (x))) φε (h (x) − y) dydx
Ω Ω
Using the uniform continuity of f, you can now pass to a limit and obtain
Z Z
f (y) d (y, Ω, h) dy = f (h (x)) det Dh (x) dx
Ω

This has essentially proved the following interesting Theorem.

³ ´
Theorem 10.7.3 Let f ∈ Cc h (∂Ω)C for Ω a bounded open set where h is
¡ ¢
in C 1 Ω; Rn . Suppose ∂Ω has measure zero. Then
Z Z
f (y) d (y, Ω, h) dy = det (Dh (x)) f (h (x)) dx.
Ω

Proof: Both sides depend only on h restricted to Ω and so the above results apply
and give the formula of the theorem for any h ∈ C 1 (Rn ; Rn ) which vanishes off a
bounded set which coincides with h on Ω. This proves the lemma. ¥
¡ ¢
Lemma 10.7.4 If h ∈ C 1 Ω, Rn for Ω a bounded connected open set with ∂Ω
having measure zero and h is one to one on Ω. Then for any E a Borel set,
Z Z
XE (y) d (y, Ω, h) dy = det (Dh (x)) XE (h (x)) dx
h(Ω) Ω

Furthermore, off a set of measure zero, det (Dh (x)) has constant sign equal to the sign
of d (y, Ω, h).

Proof: First here is a simple observation about the existence of a sequence of

nonnegative continuous functions which increase to XO for O open.
Let O be an open set and let Hj denote those points y of O such that
¡ ¢
dist y, OC ≥ 1/j, j = 1, 2, · · · .

Then define Kj ≡ B (0,j) ∩ Hj and

µ ¶
1
Wj ≡ Kj + B 0, .
2j
where this means ½ µ ¶¾
1
k + b : k ∈ Kj , b ∈ B 0, .
2j
286 BROUWER DEGREE

Let ¡ ¢
dist y, WjC
fj (y) ≡ ¡ ¢
dist (y, Kj ) + dist y, WjC
Thus fj is nonnegative, increasing in j, has compact support in Wj , is continuous, and
eventually fj (y) = 1 for all y ∈ O and fj (y) = 0 for all y ∈
/ O. Thus limj→∞ fj (y) =
XO (y).
C
Now let O ⊆ h (∂Ω) . Then from the above, let fj be as described above for the
open set O ∩ h (Ω) . (By invariance of domain, h (Ω) is open.)
Z Z
fj (y) d (y, Ω, h) dy = det (Dh (x)) fj (h (x)) dx
h(Ω) Ω

From Proposition 10.6.2, d (y, Ω, h) either equals 1 or −1 for all y ∈ h (Ω). Then by
the monotone convergence theorem on the left, using the fact that d (y, Ω, h) is either
always 1 or always −1 and the dominated convergence theorem on the right, it follows
Z Z
XO (y) d (y, Ω, h) dy = det (Dh (x)) XO∩h(Ω) (h (x)) dx
h(Ω) Ω
Z
= det (Dh (x)) XO (h (x)) dx
Ω

If y ∈ h (∂Ω) , then since h (Ω) ∩ h (∂Ω) is assumed to be empty, it follows y ∈

/ h (Ω).
Therefore, the above formula holds for any O open.
Now let G denote those Borel sets of Rn , E such that
Z Z
XE (y) d (y, Ω, h) dy = det (Dh (x)) XE (h (x)) dx
h(Ω) Ω

Then as shown above, G contains the π 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
Rn is an open set,
Z Z
XE (y) d (y, Ω, h) dy + XE C (y) d (y, Ω, h) dy
h(Ω) h(Ω)

Z Z
= d (y, Ω, h) dy = det (Dh (x)) dx
h(Ω) Ω
Z Z
= det (Dh (x)) XE C (h (x)) dx + det (Dh (x)) XE (h (x)) dx
Ω Ω

Now cancelling Z
det (Dh (x)) XE (h (x))
Ω
from the right with Z
XE (y) d (y, Ω, h) dy
h(Ω)

from the left, since E ∈ G, yields the desired result.

∞
In addition, if {Ei }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 π systems, Lemma 9.1.2 it follows G equals the Borel sets.
Now consider the last claim. Suppose d (y, Ω, h) = −1. Consider the compact set

Eε ≡ {x ∈ Ω : det (Dh (x)) ≥ ε}

10.7. INTEGRATION AND THE DEGREE 287

and f (y) ≡ Xh(Eε ) (y) . Then from the first part,

Z Z
0≥ Xh(Eε ) (y) (−1) dy = det Dh (x) XEε (x) dx
h(Ω) Ω

≥ εmn (Eε )
and so mn (Eε ) = 0. Therefore, if E is the set of x ∈ Ω where det (Dh (x)) > 0, it
equals ∪∞ k=1 E1/k , a set of measure zero. Thus off this set det Dh (x) ≤ 0. Similarly, if
d (y, Ω, h) = 1, det Dh (x) ≥ 0 off a set of measure zero. This proves the lemma. ¥

Theorem 10.7.5
¡ Let
¢ f ≥ 0 and measurable for Ω a bounded open connected
set where h is in C 1 Ω; Rn and is one to one on Ω. Then
Z Z
f (y) d (y, Ω, h) dy = det (Dh (x)) f (h (x)) dx.
h(Ω) Ω

which amounts to the same thing as

Z Z
f (y) dy = |det (Dh (x))| f (h (x)) dx.
h(Ω) Ω

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

and so multiplying both sides by −1,

Z Z Z
det (Dh (x)) f (h (x)) dx = f (y) (−1) dy = f (y) d (y, Ω, h) dy
Ω h(Ω) h(Ω)

The case where d (y, Ω, h) = 1 is similar. This proves the theorem when ∂Ω has measure
zero.
Next I show it is not necessary to assume ∂Ω has measure zero. By Corollary 9.7.6
there exists a sequence of disjoint balls {Bi } such that
mn (Ω \ ∪∞
i=1 Bi ) = 0.

Since ∂Bi has measure zero, the above holds for each Bi and so
Z Z
f (y) d (y, Bi , h) dy = det (Dh (x)) f (h (x)) dx
h(Bi ) Bi

Since h is one to one, if y ∈ h (Bi ) , then y ∈/ h (Ω \ Bi ) . Therefore, it follows from

Theorem 10.2.9 that
d (y, Bi , h) = d (y, Ω, h)
Thus Z Z
f (y) d (y, Ω, h) dy = det (Dh (x)) f (h (x)) dx
h(Bi ) Bi

From Lemma 10.7.4 det (Dh (x)) has the same sign off a set of measure zero as the
constant sign of d (y, Ω, h) . Therefore, using the monotone convergence theorem and
that h is one to one,
Z Z
det (Dh (x)) f (h (x)) dx = det (Dh (x)) f (h (x)) dx
Ω ∪∞
i=1 Bi
288 BROUWER DEGREE

∞ Z
X ∞ Z
X
= det (Dh (x)) f (h (x)) dx = f (y) d (y, Ω, h) dy
i=1 Bi i=1 h(Bi )
Z Z
= f (y) d (y, Ω, h) dy = f (y) d (y, Ω, h) dy
∪∞
i=1 h(Bi ) h(∪∞
i=1 Bi )

Z
= f (y) d (y, Ω, h) dy,
h(Ω)

the last line following from Lemma 10.7.2. This proves the theorem. ¥

10.8 Exercises
1. Show the Brouwer fixed 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 Rn \ 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 Rm
onto Rn ? 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, thennf −1 must o be continuous. Now consider
∞
the increasing sequence of compact sets B (0, k) whose union is Rm and the
n ³ ´o∞ k=1
increasing sequence f B (0, k) which you might assume covers Rn . You
k=1
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,

© ª
(x, y) : x2 + y 2 ≤ 1 ≡ D

and the annulus ½ ¾

1
(x, y) : ≤ x2 + y 2 ≤ 1 ≡A
2
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 different 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 Rn which has empty interior and f : C → Γ ⊆ Rn is

one to one onto and continuous with continuous inverse. Could Γ have nonempty
interior? Show also that if f is one to one and onto Γ then if it is continuous, so
is f −1 .
10.8. EXERCISES 289
© ª
8. Let C denote the unit circle, (x, y) ∈ R2 : x2 + y 2 = 1 . Suppose f : C → Γ ⊆ R2
is one to one and onto having continuous inverse. The Jordan curve theorem
says that under these conditions R2 \ Γ 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 R2 \ Γ is Γ. Hint: Let U1
and U2 be the two components of R2 \ Γ. Explain why none of the limit points of
U1 are are in U2 so they must all be in Γ. Similarly no limit point of U2 can be in
U1 . Use Problem 7 to show Γ has empty interior. Use this observation to argue
U1 ∪ U2 = R2 . Suppose p ∈ ∂U1 \ ∂U2 . Then explain why, since p ∈ / ∂U2 , it is not
a limit point of U2 and so there is a ball B (p, r) which contains no points of U2 .
Now where is B (p, r)? Doesn’t this contradict p ∈ ∂U1 ? Similarly ∂U2 \ ∂U1 = ∅
and so ∂U1 = ∂U2 . Now justify the statement

Γ ⊆ ∂U1 ∪ ∂U2 = ∂U1 = ∂U2 ⊆ Γ.

Thus ∂Ui = Γ.
9. Let K be a nonempty closed and convex subset of Rn . 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 ∈ Rn
there exists a unique z ∈ K such that

|x − z| = min {|x − y| : y ∈ K} .

This z will be denoted as P x. Hint: First note you do not know K is compact.
Establish the parallelogram identity if you have not already done so,
2 2 2 2
|u − v| + |u + v| = 2 |u| + 2 |v| .

Then let {zk } be a minimizing sequence,

2
lim |zk − x| = inf {|x − y| : y ∈ K} ≡ λ.
k→∞

Now using convexity, explain why

¯ ¯ ¯ ¯2 ¯ ¯2 ¯ ¯2
¯ zk − zm ¯2 ¯ ¯ ¯ ¯ ¯ ¯
¯ ¯ + ¯x− zk + zm ¯ = 2 ¯ x − zk ¯ + 2 ¯ x − zm ¯
¯ 2 ¯ ¯ 2 ¯ ¯ 2 ¯ ¯ 2 ¯

and then use this to argue {zk } is a Cauchy sequence. Then if zi works for i = 1, 2,
consider (z1 + z2 ) /2 to get a contradiction.
10. In Problem 9 show that P x satisfies the following variational inequality.

(x−P x) · (y−P x) ≤ 0

for all y ∈ K. Then show that |P x1 − P x2 | ≤ |x1 − x2 |. Hint: For the first part
2
note that if y ∈ K, the function t → |x− (P x + t (y−P x))| achieves its minimum
on [0, 1] at t = 0. For the second part,

(x1 −P x1 ) · (P x2 −P x1 ) ≤ 0, (x2 −P x2 ) · (P x1 −P x2 ) ≤ 0.

Explain why
(x2 −P x2 − (x1 −P x1 )) · (P x2 −P x1 ) ≥ 0
and then use a some manipulations and the Cauchy Schwarz inequality to get the
desired inequality.
290 BROUWER DEGREE

11. Establish the Brouwer fixed point theorem for any convex compact set in Rn .
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 fixed
point in K.
12. Suppose D is a set which is homeomorphic ³to 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 fixed point. Now show that it suffices 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 fixed point are said
to have the fixed point property. Work Problem 6 using this notion of fixed point
property. What about a solid ball and a donut?
13. There are many different proofs of the Brouwer fixed 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 fixed
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, T1 , · · · , Tm
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 fixed 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 Sperner’s lemma approach to the Brouwer fixed point theorem. Are
there other sets besides compact convex sets which have the fixed point property?
14. Using the ¡definition
¢ of the derivative and the Vitali covering theorem, show that
if f ∈ C 1 U ,Rn 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 Ω is any open bounded subset of Rn which contains 0 and that f : Ω → Rn
is continuous with the property that
f (x) · x ≥ 0
for all x ∈ ∂Ω. Show that then there exists x ∈ Ω 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 Ω is an open set in Rn containing 0 and suppose that f : Ω → Rn is

continuous and |f (x)| ≤ |x| for all x ∈ ∂Ω. Show f has a fixed point in Ω. Hint:
Consider h (t, x) ≡ t (x − f (x)) + (1 − t) x for t ∈ [0, 1] . If t = 1 and some x ∈ ∂Ω
is sent to 0, then you are done. Suppose therefore, that no fixed point exists on
∂Ω. Consider t < 1 and use the given inequality.
17. Let Ω be an open bounded subset of Rn and let f , g : Ω → Rn both be continuous
such that
|f (x)| − |g (x)| > 0
10.8. EXERCISES 291

for all x ∈ ∂Ω. Show that then

d (f − g, Ω, 0) = d (f , Ω, 0)
−1
Show that if there exists x ∈ f −1 (0) , then there exists x ∈ (f − g) (0). Hint:
You might consider h (t, x) ≡ (1 − t) f (x)+t (f (x) − g (x)) and argue 0 ∈
/ h (t, ∂Ω)
for t ∈ [0, 1].
18. Let f : C → C where C is the field of complex numbers. Thus f has a real and
imaginary part. Letting z = x + iy,
f (z) = u (x, y) + iv (x, y)
p
Recall that the norm in C is given by |x + iy| = x2 + y 2 and this is the usual
norm in R2 for the ordered pair (x, y) . Thus complex valued functions defined on
C can be considered as R2 valued functions defined on some subset of R2 . Such a
complex function is said to be analytic if the usual definition holds. That is
f (z + h) − f (z)
f 0 (z) = lim .
h→0 h
In other words,
f (z + h) = f (z) + f 0 (z) h + o (h) (10.26)
n
at a point z where the derivative exists. Let f (z) = z 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.
ux = vy , uy = −vx .
In terms of mappings show 10.26 has the form
µ ¶
u (x + h1 , y + h2 )
v (x + h1 , y + h2 )
µ ¶ µ ¶µ ¶
u (x, y) ux (x, y) uy (x, y) h1
= + + o (h)
v (x, y) vx (x, y) vy (x, y) h2
µ ¶ µ ¶µ ¶
u (x, y) ux (x, y) −vx (x, y) h1
= + + o (h)
v (x, y) vx (x, y) ux (x, y) h2
T
where h = (h1 , h2 ) and h is given by h1 +ih2 . Thus the determinant of the above
matrix is always nonnegative. Letting Br denote the ball B (0, r) = B ((0, 0) , r)
show
d (f, Br , 0) = n.
where f (z) = z n . In terms of mappings on R2 ,
µ ¶
u (x, y)
f (x, y) = .
v (x, y)
Thus show
d (f , Br , 0) = n.
Hint: You might consider
n
Y
g (z) ≡ (z − aj )
j=1

where the aj 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 aj small but distinct d (f , Br , 0) = d (g, Br , 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) = an z n + an−1 z n−1 + · · ·

Show that for large enough r, |p (z)| > |p (z) − an z n | for all z ∈ ∂B (0, r). Now
from Problem 17 you can conclude d (p, Br , 0) = d (f, Br , 0) = n where f (z) =
an z n .
20. Generalize Theorem 10.7.5 to the situation where Ω is not necessarily a connected
open set. You may need to make some adjustments on the hypotheses.
Integration Of Differential
Forms

11.1 Manifolds
Manifolds are sets which resemble Rn locally. A manifold with boundary resembles half
of Rn locally. To make this concept of a manifold more precise, here is a definition.

Definition 11.1.1 Let Ω ⊆ Rm . A set, U, is open in Ω if it is the intersection

m
of an open set from R with Ω. Equivalently, a set, U is open in Ω if for every point,
x ∈ U, there exists δ > 0 such that if |x − y| < δ and y ∈ Ω, then y ∈ U . A set, H, is
closed in Ω if it is the intersection of a closed set from Rm with Ω. Equivalently, a set,
H, is closed in Ω if whenever, y is a limit point of H and y ∈ Ω, it follows y ∈ H.

Recall the following definition.

¡ ¢
Definition 11.1.2 Let V ⊆ Rn . C k V ; Rm is the set of functions which are
restrictions to V of some function defined on Rn which has k continuous derivatives
and compact support which has values in Rm . When k = 0, it means the restriction to
V of continuous functions with compact support.

Definition 11.1.3 A closed and bounded subset of Rm , Ω, will be called an n

dimensional manifold with boundary, n ≥ 1, if there are ¡ finitely
¢ many sets, Ui , open in Ω
and continuous one to one on Ui functions, Ri ∈ C 0 Ui , Rn such that Ri Ui is relatively
¡ ¢
open in Rn≤ ≡ {u ∈ Rn : u1 ≤ 0} , R−1 i is continuous and one to one on Ri Ui . These
mappings, Ri , together with the relatively open sets Ui , are called charts and the totality
of all the charts, (Ui , Ri ) just described is called an atlas for the manifold. Define

int (Ω) ≡ {x ∈ Ω : for some i, Ri x ∈ Rn< }

where Rn< ≡ {u ∈ Rn : u1 < 0}. Also define

∂Ω ≡ {x ∈ Ω : for some i, Ri x ∈ Rn0 }

where
Rn0 ≡ {u ∈ Rn : u1 = 0}
and ∂Ω is called the boundary of Ω. Note that if n = 1, Rn0 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 ∂Ω and int (Ω) be as defined above. Then int (Ω) is open
in Ω and ∂Ω is closed in Ω. Furthermore, ∂Ω ∩ int (Ω) = ∅, Ω = ∂Ω ∪ int (Ω), and for
n ≥ 1, ∂Ω is an n − 1 dimensional manifold for which ∂ (∂Ω) = ∅. The property of
being in int (Ω) or ∂Ω does not depend on the choice of atlas.

Proof: It is clear that Ω = ∂Ω∪int (Ω). First consider the claim that ∂Ω∩int (Ω) =
∅. Suppose this does not happen. Then there would exist x ∈ ∂Ω ∩ int (Ω). Therefore,
there would exist two mappings Ri and Rj such that Rj x ∈ Rn0 and Ri x ∈ Rn< with
x ∈ Ui ∩ Uj . Now consider the map, Rj ◦ R−1
i , a continuous one to one map from R≤
n
n
to R≤ having a continuous inverse. By continuity, there exists r > 0 small enough that,

R−1
i B (Ri x,r) ⊆ Ui ∩ Uj .

Therefore, Rj ◦ R−1 n n
i (B (Ri x,r)) ⊆ R≤ and contains a point on R0 , Rj x. However,
this cannot occur because it contradicts the theorem on invariance of domain, Theorem
10.4.3, which requires that Rj ◦ R−1 n
i (B (Ri x,r)) must be an open subset of R and
n
this one isn’t because of the point on R0 . Therefore, ∂Ω ∩ int (Ω) = ∅ as claimed. This
same argument shows that the property of being in int (Ω) or ∂Ω does not depend on
the choice of the atlas.
To verify that ∂ (∂Ω) = ∅, let Si be the restriction of Ri to ∂Ω ∩ Ui . Thus

Si (x) = (0, (Ri x)2 , · · · , (Ri x)n )

and the collection of such points for x ∈ ∂Ω ∩ Ui is an open bounded subset of

{u ∈ Rn : u1 = 0} ,

identified with Rn−1 . Si (∂Ω ∩ Ui ) is bounded because Si is the restriction of a contin-

uous function defined on Rm and ∂Ω ∩ Ui ≡ Vi is contained in the compact set Ω. Thus
if Si is modified slightly, to be of the form

S0i (x) = ((Ri x)2 − ki , · · · , (Ri x)n )

where ki is chosen sufficiently large enough that (Ri (Vi ))2 − ki < 0, it follows that
{(Vi , S0i )} is an atlas for ∂Ω as an n − 1 dimensional manifold such that every point of
∂Ω is sent to to Rn−1 < and none gets sent to Rn−1
0 . It follows ∂Ω is an n − 1 dimensional
manifold with empty boundary. In case n = 1, the result follows by definition of the
boundary of a 0 dimensional manifold.
Next consider the claim that int (Ω) is open in Ω. If x ∈ int (Ω) , are all points of
Ω which are sufficiently close to x also in int (Ω)? If this were not true, there would
exist {xn } such that xn ∈ ∂Ω and xn → x. Since there are only finitely many charts
of interest, this would imply the existence of a subsequence, still denoted by xn and a
single map, Ri such that Ri (xn ) ∈ Rn0 . But then Ri (xn ) → Ri (x) and so Ri (x) ∈ Rn0
showing x ∈ ∂Ω, a contradiction to int (Ω) ∩ ∂Ω = ∅. Now it follows that ∂Ω is closed
in Ω because ∂Ω = Ω \ int (Ω). This proves the theorem. ¥

Definition 11.1.5 An n dimensional manifold with boundary, Ω is a C k man-

ifold with boundary for some k ≥ 0 if
³ ´
Rj ◦ R−1
i ∈ C k Ri (Ui ∩ Uj ); Rn
¡ ¢
and R−1
i ∈ C k Ri Ui ; Rm . It is called a continuous manifold with boundary if the
mappings, Rj ◦ R−1 −1 k
i , Ri , Ri are continuous. In the case where Ω is a C , k ≥ 1
11.1. MANIFOLDS 295

manifold, it is called orientable if in addition to this there exists an atlas, (Ur , Rr ),

such that whenever Ui ∩ Uj 6= ∅,
¡ ¡ ¢¢
det D Rj ◦ R−1 i (u) > 0 for all u ∈ Ri (Ui ∩ Uj ) (11.1)

The mappings, Ri ◦ R−1 j are called the overlap maps. In the case where k = 0, the Ri
are only assumed continuous so there is no differentiability available and in this case,
the manifold is oriented if whenever A is an open connected subset of int (Ri (Ui ∩ Uj ))
whose boundary has measure zero and separates Rn into two components,
¡ ¢
d y,A, Rj ◦ R−1
i ∈ {1, 0} (11.2)

depending on whether y ∈ Rj ◦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 definition in
terms of the degree when applied to the situation of a C k manifold, gives the same thing
as the earlier definition in terms of the determinant of the derivative.

The advantage of using the degree in the above definition to define orientation is
that it does not depend on any kind of differentiability 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.

Definition 11.1.6 In the above context, I will call Ω a P C 1 manifold if it

0
is a C manifold with charts (Ri , Ui ) and there exists a closed set L ⊆ Ω such that
Ri (L ∩ Ui ) is closed in Ri (Ui ) and has mn measure zero, Ri (L ∩ Ui ∩ ∂Ω) is closed in
Rn−1 and has mn−1 measure zero, and the following conditions hold.

R−1
i ∈ C 1 (Ri (Ui \ L) ; Rm ) (11.3)
Rj ◦ R−1
i ∈ C 1 (Ri ((Ui ∩ Uj ) \ L) ; Rn ) (11.4)
©¯¯ ¯¯ ª
sup ¯¯DR−1 ¯¯
i (u) F : u ∈ Ri (Ui \ L) < ∞ (11.5)
where the norm is the Frobenius norm
 1/2
X 2
||M ||F ≡  |Mij |  .
i,j

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 different 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 Rm 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 definition of orientable in
the case of a P C 1 manifold.

Proposition 11.1.7 Let Ω be a P C 1 n dimensional manifold with boundary. Then

it is an orientable manifold if and only if there exists an atlas {Ri , Ui } such that for
each i, j ¡ ¢ ¡ ¢
det D Ri ◦ R−1
j (u) ≥ 0 a.e. u ∈ int Ri ◦ R−1
j (Uj ∩ Ui ) (11.6)
296 INTEGRATION OF DIFFERENTIAL FORMS

If v = Ri ◦ R−1
j (u) , I will often write

∂ (v1 · · · vn )
≡ det DRi ◦ R−1
j (u)
∂ (u1 · · · un )
∂(v1 ···vn )
Thus in this situation ∂(u1 ···un ) ≥ 0.

Proof: Suppose first the chart is an oriented chart so

¡ ¢
d v,A, Ri ◦ R−1j =1

whenever v ∈ int Rj (A) where A is an open ball contained in Ri ◦ R−1

j (Ui ∩ Uj \ L) .
Then by Theorem 10.7.5, if E ⊆ A is any Borel measurable set,
Z Z
¡ ¡ ¢ ¢
0≤ XRi ◦R−1 (E) (v) 1dv = det D Ri ◦ R−1j (u) XE (u) du
j
Ri ◦R−1
j (A) A

¡ ¡ ¢ ¢
Since this is true for arbitrary E ⊆ A, it follows det D Ri ◦ R−1 j (u) ≥ 0 a.e. u ∈ A
© ¡ ¡ ¢ ¢ ª
because if not so, then you could take Eδ ≡ u : det D Ri ◦ R−1 j (u) < −δ and
for some δ > 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 P C 1 , there exists a sequence of disjoint open balls contained in
Ri ◦ R−1
j (Ui ∩ Uj \ L) , {Ak } such that
¡ ¢
int Ri ◦ R−1 ∞
j (Uj ∩ Ui ) = L ∪ ∪k=1 Ak
¡ ¢
and from the above, there exist sets of measure zero Nk ⊆ Ak such that det D Ri ◦ R−1 j¢ (u) ≥
¡ ¢ ¡
0 for all u ∈ Ak \ Nk . Then det D Ri ◦ R−1 j (u) ≥ 0 on int Ri ◦ R−1 j (Uj ∩ Ui ) \
∞
(L ∪ ∪k=1 Nk ) . This proves one direction.
¡ Now
¢ consider the other direction.
Suppose the condition det D Ri ◦ R−1 j (u) ≥ 0 a.e. Then by Theorem 10.7.5
Z Z
¡ ¢ ¡ ¡ ¢ ¢
d v, A, Ri ◦ R−1
j dv = det D Ri ◦ R−1
j (u) du ≥ 0
Ri ◦R−1
j (A) A

The degree is constant on the connected open set Ri ◦ R−1j (A) . By Proposition 10.6.2,
the degree equals either −1 or 1. The above inequality shows it can’t equal −1 and so
it must equal 1. This proves the proposition. ¥
This shows it would be fine to simply use 11.6 as the definition of orientable in
the case of a P C 1 manifold and not bother with the definiton in terms of the degree.
However, that one is more general because it does not depend on any differentiability.
I will use this result whenever convenient in what follows.

11.2 Some Important Measure Theory

11.2.1 Eggoroff ’s Theorem
Eggoroff’s theorem says that if a sequence converges pointwise, then it almost converges
uniformly in a certain sense.

Theorem 11.2.1 (Egoroff ) Let (Ω, F, µ) be a finite measure space,

(µ(Ω) < ∞)
11.2. SOME IMPORTANT MEASURE THEORY 297

and let fn , f be complex valued functions such that Re fn , Im fn are all measurable and
lim fn (ω) = f (ω)
n→∞

for all ω ∈
/ E where µ(E) = 0. Then for every ε > 0, there exists a set,
F ⊇ E, µ(F ) < ε,
such that fn converges uniformly to f on F C .
Proof: First suppose E = ∅ so that convergence is pointwise everywhere. It follows
then that Re f and Im f are pointwise limits of measurable functions and are therefore
measurable. Let Ekm = {ω ∈ Ω : |fn (ω) − f (ω)| ≥ 1/m for some n > k}. Note that
q
2 2
|fn (ω) − f (ω)| = (Re fn (ω) − Re f (ω)) + (Im fn (ω) − Im f (ω))
and so, · ¸
1
|fn − f | ≥
m
is measurable. Hence Ekm is measurable because
· ¸
1
Ekm = ∪∞n=k+1 |fn − f | ≥ .
m
For fixed m, ∩∞k=1 Ekm = ∅ because fn converges to f . Therefore, if ω ∈ Ω there exists
1
k such that if n > k, |fn (ω) − f (ω)| < m which means ω ∈
/ Ekm . Note also that
Ekm ⊇ E(k+1)m .
Since µ(E1m ) < ∞, Theorem 7.3.2 on Page 155 implies
0 = µ(∩∞
k=1 Ekm ) = lim µ(Ekm ).
k→∞

Let k(m) be chosen such that µ(Ek(m)m ) < ε2−m and let
∞
[
F = Ek(m)m .
m=1

Then µ(F ) < ε because

∞
X ∞
¡ ¢ X
µ (F ) ≤ µ Ek(m)m < ε2−m = ε
m=1 m=1

Now let η > 0 be given and pick m0 such that m−1 C

0 < η. If ω ∈ F , then
∞
\
C
ω∈ Ek(m)m .
m=1

C
Hence ω ∈ Ek(m 0 )m0
so
|fn (ω) − f (ω)| < 1/m0 < η
for all n > k(m0 ). This holds for all ω ∈ F C and so fn converges uniformly to f on F C .
∞
Now if E 6= ∅, consider {XE C fn }n=1 . Each XE C fn has real and imaginary parts
measurable and the sequence converges pointwise to XE f everywhere. Therefore, from
the first part, there exists a set of measure less than ε, F such that on F C , {XE C fn }
C
converges uniformly to XE C f. Therefore, on (E ∪ F ) , {fn } converges uniformly to f .
This proves the theorem. ¥
298 INTEGRATION OF DIFFERENTIAL FORMS

11.2.2 The Vitali Convergence Theorem

The Vitali convergence theorem is a convergence theorem which in the case of a finite
measure space is superior to the dominated convergence theorem.

Definition 11.2.2 Let (Ω, F, µ) be a measure space and let S ⊆ L1 (Ω). S is

uniformly integrable if for every ε > 0 there exists δ > 0 such that for all f ∈ S
Z
| f dµ| < ε whenever µ(E) < δ.
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 finite.

Proof: Let ε > 0 be given and suppose S is uniformly integrable. First suppose the
functions are real valued. Let δ be such that if µ (E) < δ, then
¯Z ¯
¯ ¯
¯ f dµ¯ < ε
¯ ¯ 2
E

for all f ∈ S. Let µ (E) < δ. Then if f ∈ S,

Z Z Z
|f | dµ ≤ (−f ) dµ + f dµ
E E∩[f ≤0] E∩[f >0]
¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
¯ ¯ ¯ ¯
= ¯ f dµ¯ + ¯ f dµ¯
¯ E∩[f ≤0] ¯ ¯ E∩[f >0] ¯
ε ε
< + = ε.
2 2
In general, if S is a uniformly integrable set of complex valued functions, the inequalities,
¯Z ¯ ¯Z ¯ ¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
¯ Re f dµ¯ ≤ ¯ f dµ¯ , ¯ Im f dµ¯ ≤ ¯ f dµ¯ ,
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
E E E E

imply Re S ≡ {Re f : f ∈ S} and Im S ≡ {Im f : 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 suffices to verify a single function in L1 (Ω) is uniformly inte-
grable. To do so, note that from the dominated convergence theorem,
Z
lim |f | dµ = 0.
R→∞ [|f |>R]

R ε
Let ε > 0 be given and choose R large enough that [|f |>R]
|f | dµ < 2. Now let
ε
µ (E) < 2R . Then
Z Z Z
|f | dµ = |f | dµ + |f | dµ
E E∩[|f |≤R] E∩[|f |>R]
ε ε ε
< Rµ (E) + < + = ε.
2 2 2
This proves the lemma. ¥
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 L1 (Ω, µ) where µ (Ω) < ∞. Let t → h (t) be a
continuous function which satisfies

h (t)
lim =∞
t→∞ t

Then S is uniformly integrable and bounded in L1 (Ω) if

½Z ¾
sup h (|f |) dµ : f ∈ S = N < ∞.
Ω

Proof: First I show S is bounded in L1 (Ω; µ) which means there exists a constant
M such that for all f ∈ S, Z
|f | dµ ≤ M.
Ω

From the properties of h, there exists Rn such that if t ≥ Rn , then h (t) ≥ nt. Therefore,
Z Z Z
|f | dµ = |f | dµ + |f | dµ
Ω [|f |≥Rn ] [|f |<Rn ]

Letting n = 1,
Z Z
|f | dµ ≤ h (|f |) dµ + R1 µ ([|f | < R1 ])
Ω [|f |≥R1 ]
≤ N + R1 µ (Ω) ≡ M.

Next let E be a measurable set. Then for every f ∈ S,

Z Z Z
|f | dµ = |f | dµ + |f | dµ
E [|f |≥Rn ]∩E [|f |<Rn ]∩E

Z
1 N
≤ |f | dµ + Rn µ (E) ≤ + Rn µ (E)
n Ω n
and letting n be large enough, this is less than

ε/2 + Rn µ (E)

Now if µ (E) < ε/2Rn , it follows that for all f ∈ S,

Z
|f | dµ < ε
E

This proves the lemma. ¥

Letting h (t) = t2 , it follows that if all the functions in S are bounded, then the
collection of functions is uniformly integrable.
The following theorem is Vitali’s convergence theorem.

Theorem 11.2.5 Let {fn } be a uniformly integrable set of complex valued func-
tions, µ(Ω) < ∞, and fn (x) → f (x) a.e. where f is a measurable complex valued
function. Then f ∈ L1 (Ω) and
Z
lim |fn − f |dµ = 0. (11.7)
n→∞ Ω
300 INTEGRATION OF DIFFERENTIAL FORMS

Proof: First it will be shown that f ∈ L1 (Ω). By uniform integrability, there exists
δ > 0 such that if µ (E) < δ, then
Z
|fn | dµ < 1
E

for all n. By Egoroff’s theorem, there exists a set, E of measure less than δ such that
on E C , {fn } converges uniformly. Therefore, for p large enough, and n > p,
Z
|fp − fn | dµ < 1
EC

which implies Z Z
|fn | dµ < 1 + |fp | dµ.
EC Ω
Then since there are only finitely many functions, fn with n ≤ p, there exists a constant,
M1 such that for all n, Z
|fn | dµ < M1 .
EC
But also,
Z Z Z
|fm | dµ = |fm | dµ + |fm |
Ω EC E
≤ M1 + 1 ≡ M.

Therefore, by Fatou’s lemma,

Z Z
|f | dµ ≤ lim inf |fn | dµ ≤ M,
Ω n→∞

showing that f ∈ L1 as hoped.

Now
R S ∪ {f } is uniformly integrable so there exists δ 1 > 0 such that if µ (E) < δ 1 ,
then E |g| dµ < ε/3 for all g ∈ S ∪ {f }.
By Egoroff’s theorem, there exists a set, F with µ (F ) < δ 1 such that fn converges
uniformly to f on F C . Therefore, there exists N such that if n > N , then
Z
ε
|f − fn | dµ < .
FC 3
It follows that for n > N ,
Z Z Z Z
|f − fn | dµ ≤ |f − fn | dµ + |f | dµ + |fn | dµ
Ω FC F F
ε ε ε
< + + = ε,
3 3 3
which verifies 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 m × n

matrix. Also let Ai
i = 1, · · · , C (n, m)
be the m × m submatrices of A which are obtained by deleting n − m rows and let Bi
be the m × m submatrices of B which are obtained by deleting corresponding n − m
columns. Then
C(n,m)
X
det (BA) = det (Bk ) det (Ak )
k=1

Proof: This follows from a computation. By Corollary 3.5.6 on Page 40

det (BA) =

1 X X
sgn (i1 · · · im ) sgn (j1 · · · jm )
m! i ···i j1 ···jm
1 m

· (BA)i1 j1 (BA)i2 j2 · · · (BA)im jm

X 1 X X
= sgn (i1 · · · im ) sgn (j1 · · · jm ) ·
r1 ,··· ,rm
m! i ···i j1 ···jm
1 m

Bi1 r1 Ar1 j1 Bi2 r2 Ar2 j2 · · · Bim rm Arm jm

X 1 X
= sgn (i1 · · · im ) Bi1 r1 Bi2 r2 · · · Bim rm ·
r1 ,··· ,rm
m! i ···i
1 m
X
sgn (j1 · · · jm ) Ar1 j1 Ar2 j2 · · · Arm jm
j1 ···jm

X 1 2
= sgn (r1 · · · rn ) det (B (r1 · · · rm )) det (A (r1 · · · rm ))
r1 ,··· ,rm
m!

where A (r1 · · · rm ) denotes the submatrix of A which consists of rows r1 · · · rm taken in

that order and B (r1 · · · rm ) denotes the submatrix of B consisting of columns r1 · · · rm
taken in that order. Since there are m! ways of permuting m rows (columns), this
reduces to the claimed result of the theorem. ¥

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 don’t have to worry about orientation. It will involve the following
definition.

Definition p11.4.1 Let (Ui , Ri ) be an atlas for an n dimensional P C 1 manifold

∞
Ω. Also let {ψ i }i=1 be a C partition of unity, spt ψ i ⊆ Ui . Then for E a Borel subset
of Ω, define
p Z
X ¡ ¢ ¡ −1 ¢
σ n (E) ≡ ψ i R−1
i (u) XE Ri (u) Ji (u) du
i=1 Ri Ui

where ¡ ¡ ¢¢1/2
∗
Ji (u) ≡ det DR−1 −1
i (u) DRi (u)
302 INTEGRATION OF DIFFERENTIAL FORMS

By the Binet Cauchy theorem, this equals

 
X µ ∂ (xi · · · xi ) ¶2 1/2
 1 n
(u) 
i ,··· ,i
∂ (u1 · · · un )
1 n

where the sum is taken over all increasing strings of n indices (i1 , · · · , in ) and
 
xi1 ,u1 xi1 ,u2 · · · xi1 ,un
 · · · xi2 ,u2 
∂ (xi1 · · · xin )  xi2 ,u1 xi2 ,u2 
(u) ≡ det  .. .. .. ..  (u) (11.8)
∂ (u1 · · · un )  . . . . 
xin ,u1 xin ,u2 · · · xin ,un

Suppose (Vj , Sj ) is another atlas and you have

v = Sj ◦ R−1
i (u) (11.9)
¡ ¢
for u ∈ Ri ((Vj ∩ Ui ) \ L) . Then R−1
i = S−1
j ◦ Sj ◦ Ri
−1
and so by the chain rule,
¡ ¢
DS−1 j (v) D Sj ◦ Ri
−1
(u) = DR−1i (u)

Therefore,
¡ ¡ ∗ ¢¢1/2
Ji (u) = det DR−1 −1
i (u) DRi (u)
  1/2
n×n n×n n×n
z ¡ }| ¢ {z }| {z ¡ }| ¢ {
  ∗ ∗ 
= det D Sj ◦ R−1
i (u) DS−1 −1 −1
j (v) DSj (v)D Sj ◦ Ri (u)

¯ ¡ ¡ ¢ ¢¯
= ¯det D Sj ◦ R−1
i (u) ¯ Jj (v) (11.10)
Similarly ¯ ¡ ¡ ¢ ¢¯
Jj (v) = ¯det D Ri ◦ S−1
j (v) ¯ Ji (u) . (11.11)
In the situation of 11.9, it is convenient to use the notation
∂ (v1 · · · vn ) ¡ ¡ ¢ ¢
≡ det D Sj ◦ R−1
i (u)
∂ (u1 · · · un )
and this will be used occasionally below.
It is necessary to show the above definition is well defined.

Theorem 11.4.2 The above definition of surface measure is well defined. That
is, suppose Ω is an n dimensional orientable P C 1 manifold with boundary and let
p q
{(Ui , Ri )}i=1 and {(Vj , Sj )}j=1 be two atlass of Ω. Then for E a Borel set, the com-
putation of σ n (E) using the two different atlass gives the same thing. This defines a
Borel measure on Ω. Furthermore, if E ⊆ Ui , σ n (E) reduces to
Z
¡ ¢
XE R−1 i (u) Ji (u) du
Ri Ui

Also σ n (L) = 0.

Proof: Let {ψ i } be a partition of unity as described in Lemma 11.5.3 which is

associated with the atlas (Ui , Ri ) and let {η i } be a partition of unity associated in the
same manner with the atlas (Vi , Si ). First note the following.
Z
¡ ¢ ¡ −1 ¢
ψ i R−1 i (u) XE Ri (u) Ji (u) du
R i Ui
11.4. THE AREA MEASURE ON A MANIFOLD 303

q Z
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢
= η j R−1
i (u) ψ i Ri (u) XE Ri (u) Ji (u) du
j=1 Ri (Ui ∩Vj )

q Z
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢
= η j R−1
i (u) ψ i Ri (u) XE Ri (u) Ji (u) du
j=1 int Ri (Ui ∩Vj \L)

The reason this can be done is that points not on the interior of Ri (Ui ∩ Vj ) are on
the plane u1 = 0 which is a set of measure zero and by assumption Ri (L ∩ Ui ∩ Vj )
has measure zero. Of course the above determinants in the definition of Ji (u) in the
integrand are not even defined on the set of measure zero Ri (L ∩ Ui ) . It follows the
p
definition of σ n (E) in terms of the atlas {(Ui , Ri )}i=1 and specified partition of unity
is
p X
X q Z
¡ ¢ ¡ −1 ¢ ¡ −1 ¢
η j R−1
i (u) ψ i Ri (u) XE Ri (u) Ji (u) du
i=1 j=1 int Ri (Ui ∩Vj \L)

By the change of variables formula, Theorem 9.9.4, this equals

q Z
p X
X ¡ ¢ ¡ −1 ¢
η j S−1
j (v) ψ i Sj (v) ·
(
i=1 j=1 )(int Ri (Ui ∩Vj \L))
Sj ◦R−1
i
¡ −1 ¢ ¯ ¡ ¢¯
XE Sj (v) Ji (u) ¯det Ri ◦ S−1 ¯
j (v) dv

The integral is unchanged if it is taken over Sj (Ui ∩ Vj ) and so, by 11.11, this equals
q Z
p X
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢
η j S−1
j (v) ψ i Sj (v) XE Sj (v) Jj (v) dv
i=1 j=1 Sj (Ui ∩Vj )

Xq Z
¡ ¢ ¡ −1 ¢
= η j S−1
j (v) XE Sj (v) Jj (v) dv
j=1 Sj (Ui ∩Vj )

which equals the definition of σ n (E) taken with respect to the other atlas and partition
of unity. Thus the definition is well defined. 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 Ui . Then
using Lemma 11.5.3 there exists a partition of unity such that ψ k = 1 on K. Consider
the sum used to define σ n (K) ,
p Z
X ¡ ¢ ¡ −1 ¢
ψ i R−1
i (u) XK Ri (u) Ji (u) du
i=1 Ri Ui

Unless R−1 i¡ (u) ∈ ¢K, the integrand equals

−1
¡ −1 0. ¢ Assume then that Ri (u) ∈ K. If
−1
i 6= k, ψ i Ri (u) = 0 because ψ k Ri (u) = 1 and these functions sum to 1.
Therefore, the above sum reduces to
Z
¡ ¢
XK R−1k (u) Jk (u) du.
Rk Uk

Next consider the general case. By Theorem 7.4.6 the Borel measure σ n is regular. Also
Lebesgue measure is regular. Therefore, there exists an increasing sequence of compact
∞
subsets of E, {Kr }k=1 such that

lim σ n (Kr ) = σ n (E)

r→∞
304 INTEGRATION OF DIFFERENTIAL FORMS

Then letting F = ∪∞ r=1 Kr , the monotone convergence theorem implies

Z
¡ ¢
σ n (E) = lim σ n (Kr ) = XF R−1k (u) Jk (u) du
r→∞ Rk Uk
Z
¡ −1 ¢
≤ XE Rk (u) Jk (u) du
R k Uk

Next take an increasing sequence of compact sets contained in Rk (E) such that

lim mn (Kr ) = mn (Rk (E)) .

r→∞
© ª∞
Thus R−1
k (Kr ) r=1 is an increasing sequence of compact subsets of E. Therefore,
Z
¡ ¢
σ n (E) ≥ lim σ n R−1 k (K r ) = lim XKr (u) Jk (u) du
r→∞ r→∞ R U
Z k k

= XRk (E) (u) Jk (u) du

Rk Uk
Z
¡ ¢
= XE R−1k (u) Jk (u) du
Rk Uk

Thus Z
¡ ¢
σ n (E) = XE R−1
k (u) Jk (u) du
Rk Uk

as claimed.
So what is the measure of L? By definition it equals

p Z
X ¡ ¢ ¡ −1 ¢
σ n (L) ≡ ψ i R−1
i (u) XL Ri (u) Ji (u) du
i=1 Ri Ui
Xp Z
¡ ¢
= ψ i R−1
i (u) XRi (L∩Ui ) (u) Ji (u) du = 0
i=1 Ri Ui

because by assumption, the measure of each Ri (L ∩ Ui ) is zero. This proves the theo-
rem. ¥

11.5 Integration Of Differential Forms On Manifolds

This section presents the integration of differential forms on manifolds. This topic is a
higher dimensional version of what is done in calculus in finding the work done by a
force field on an object which moves over some path. There you evaluated line integrals.
Differential 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 definition of the degree and in giving a proof of the Brouwer fixed
point theorem is also a fundamental result in discussing the integration of differential
forms.

Lemma 11.5.1 Let g : U → V be C 2 where U and V are open subsets of Rn . Then

n
X
(cof (Dg))ij,j = 0,
j=1

∂gi
where here (Dg)ij ≡ gi,j ≡ ∂xj .
11.5. INTEGRATION OF DIFFERENTIAL FORMS ON MANIFOLDS 305

Recall Proposition 10.6.2.

Proposition 11.5.2 Let Ω be an open connected bounded set in Rn¡ such ¢that Rn \
∂Ω consists of two, three if n = 1, connected components. Let f ∈ C Ω; Rn be con-
tinuous and one to one. Then f (Ω) is the bounded component of Rn \ f (∂Ω) and for
y ∈ f (Ω) , d (f , Ω, 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 Rn and let {Ui }m

i=1 be an open cover of
K. Then there exist functions, ψ k ∈ Cc∞ (Ui ) such that ψ i ≺ Ui and for all x ∈ K,
m
X
ψ i (x) = 1.
i=1

If K is a compact subset of U1 (Ui )there exist such functions such that also ψ 1 (x) = 1
(ψ i (x) = 1) for all x ∈ K.

With the above, what follows is the definition of what a differential form is and how
to integrate one.

Definition 11.5.4 Let I denote an ordered list of n indices taken from the set,
{1, · · · , m}. Thus I = (i1 , · · · , in ). It is an ordered list because the order matters. A
differential form of order n in Rm is a formal expression,
X
ω= aI (x) dxI
I

where aI is at least Borel measurable dxI is short for the expression

dxi1 ∧ · · · ∧ dxin ,

and the sum is taken over all ordered lists of indices taken from the set, {1, · · · , m}.
For Ω an orientable n dimensional manifold with boundary, let {(Ui , Ri )} be an oriented
atlas for Ω. Each Ui is the intersection of an open set in Rm , Oi , with Ω and so there
exists a C ∞ partition of unity subordinate to the open cover,
P {Oi } which sums to 1 on
Ω. Thus ψ i ∈ Cc∞ (Oi ), has values in [0, 1] and satisfies i ψ i (x) = 1 for all x ∈ Ω.
Define
Z p XZ
X ¡ ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
ω≡ ψ i R−1
i (u) aI Ri (u) du (11.12)
Ω i=1 Ri Ui ∂ (u1 · · · un )
I

Note
∂ (xi1 · · · xin )
,
∂ (u1 · · · un )
given by 11.8 is not defined on Ri (Ui ∩ 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 definition is Rwell
defined. What if you had a different atlas and a different partition of unity? Would Ω ω
change? In general, the answer is yes. However, there is a sense in which the integral of
a differential form is well defined. This involves the concept of orientation which looks
a lot like the concept of an oriented manifold.
306 INTEGRATION OF DIFFERENTIAL FORMS

Definition 11.5.5 Suppose Ω is an n dimensional orientable manifold with

boundary and let (Ui , Ri ) and (Vi , Si ) be two oriented atlass of Ω. They have the same
orientation if for all open connected sets A ⊆ Sj (Vj ∩ Ui ) with ∂A having measure zero
and separating Rn into two components,
¡ ¢
d u, Ri ◦ S−1
j , A ∈ {0, 1}

depending on whether u ∈ Ri ◦ S−1

j (A).

R
The above definition of Ω ω is well defined in the sense that any two atlass which
have the same orientation deliver the same value for this symbol.

Theorem 11.5.6 Suppose Ω is an n dimensional orientable P C 1 manifold with

boundary and let (Ui , Ri ) and (Vi , Si )R be two oriented atlass of Ω. Suppose the two atlass
have the same orientation. Then if Ω ω is computed with respect to the two atlass the
same number is obtained. Assume each aI is Borel measurable and bounded.1

Proof: Let {ψ i } be a partition of unity as described in Lemma 11.5.3 which is

associated with the atlas (Ui , Ri ) and let {η i } be a partition of unity associated in the
same manner with the atlas (Vi , Si ). Then the definition using the atlas {(Ui , Ri )} is
p XZ
X ¡ ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
ψ i R−1
i (u) aI Ri (u) du (11.13)
i=1 Ri Ui ∂ (u1 · · · un )
I

q XZ
p X
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
= η j R−1i (u) ψ i Ri (u) aI Ri (u) du
i=1 j=1 I Ri (Ui ∩Vj ) ∂ (u1 · · · un )
Xp X q XZ
¡ ¢ ¡ −1 ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
= η j R−1
i (u) ψ i Ri (u) aI Ri (u) du
i=1 j=1 int Ri (Ui ∩Vj \L) ∂ (u1 · · · un )
I

The reason this can be done is that points not on the interior of Ri (Ui ∩ Vj ) are on
the plane u1 = 0 which is a set of measure zero and by assumption Ri (L ∩ Ui ∩ Vj ) has
measure zero. Of course the above determinant in the integrand is not even defined on
Ri (L ∩ Ui ∩ Vj ) . By the change of variables formula Theorem 9.9.10 and Proposition
11.1.7, this equals
q XZ
p X
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
η j S−1
j (v) ψ i Sj (v) aI Sj (v) dv
i=1 j=1 int Sj (Ui ∩Vj \L) ∂ (v1 · · · vn )
I

q XZ
p X
X ¡ ¢ ¡ −1 ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
= η j S−1
j (v) ψ i Sj (v) aI Sj (v) dv
i=1 j=1 Sj (Ui ∩Vj ) ∂ (v1 · · · vn )
I

q XZ
X ¡ ¢ ¡ −1 ¢ ∂ (xi1 · · · xin )
= η j S−1
j (v) aI Sj (v) dv
j=1 Sj (Vj ) ∂ (v1 · · · vn )
I
R
which is the definition Ω
ω using the other atlas {(Vj , Sj )} and partition of unity. This
proves the theorem. ¥
1 Thisis so issues of existence for the various integrals will not arrise. This is leading to Stoke’s
theorem in which even more will be assumed on aI .
11.6. STOKE’S THEOREM AND THE ORIENTATION OF ∂Ω 307

11.5.1 The Derivative Of A Differential Form

The derivative of a differential form is defined next.
P
Definition 11.5.7 Let ω = I aI (x) dxi1 ∧ · · · ∧ dxin−1 be a differential form
of order n−1 where aI is C 1 . Then define dω, a differential form of order n by replacing
aI (x) with
Xm
∂aI (x)
daI (x) ≡ dxk (11.14)
∂xk
k=1

and putting a wedge after the dxk . Therefore,

m
XX ∂aI (x)
dω ≡ dxk ∧ dxi1 ∧ · · · ∧ dxin−1 . (11.15)
∂xk
I k=1

11.6 Stoke’s Theorem And The Orientation Of ∂Ω

Here Ω will be an n dimensional orientable P C 1 manifold with boundary in Rm . Let
p p
an oriented atlas for it be {Ui , Ri }i=1 and let a C ∞ partition of unity be {ψ i }i=1 . Also
let X
ω= aI (x) dxi1 ∧ · · · ∧ dxin−1
I
¡ ¢ P
be a differential form such that aI is C 1 Ω . Since ψ i (x) = 1 on Ω,
m Xp ¡ ¢
XX ∂ ψ j aI
dω = (x) dxk ∧ dxi1 ∧ · · · ∧ dxin−1
j=1
∂xk
I k=1

because the right side equals

p
m X
XX ∂ψ j
aI (x) dxk ∧ dxi1 ∧ · · · ∧ dxin−1
∂xk
I k=1 j=1
m X
XX p
∂aI
+ ψ j (x) dxk ∧ dxi1 ∧ · · · ∧ dxin−1
j=1
∂x k
I k=1

=1  
z }| {
∂  
X m
X p
X 
aI (x)  ψ j  dxk ∧ dxi1 ∧ · · · ∧ dxin−1
∂xk  j=1 
I k=1

m
XX p
∂aI X
+ ψ (x) dxk ∧ dxi1 ∧ · · · ∧ dxin−1
∂xk j=1 j
I k=1

m
XX ∂aI
= (x) dxk ∧ dxi1 ∧ · · · ∧ dxin−1 ≡ dω
∂xk
I k=1

It follows
Z p Z
m X ¡ ¢ ¡ ¢
XX ∂ ψ j aI ¡ −1 ¢ ∂ xk , xi1 · · · xin−1
dω = Rj (u) du
Rj (Uj ) ∂xk ∂ (u1 , · · · , un )
I k=1 j=1
308 INTEGRATION OF DIFFERENTIAL FORMS

p Z
m X ¡ ¢ ¡ ¢
XX ∂ ψ j aI ¡ −1 ¢ ∂ xkε , xi1 ε · · · xin−1 ε
= Rjε (u) du+
∂xk ∂ (u1 , · · · , un )
I k=1 j=1 Rj (Uj )
m X p Z ¡ ¢ ¡ ¢
XX ∂ ψ j aI ¡ −1 ¢ ∂ xk , xi1 · · · xin−1
Rj (u) du
∂xk ∂ (u1 , · · · , un )
I k=1 j=1 Rj (Uj )
m Xp Z ¡ ¢ ¡ ¢
XX ∂ ψ j aI ¡ −1 ¢ ∂ xkε , xi1 ε · · · xin−1 ε
− Rjε (u) du (11.16)
j=1 Rj (Uj )
∂xk ∂ (u1 , · · · , un )
I k=1

Here
R−1 −1
jε (u) ≡ Rj ∗ φε (u)

for φε a mollifier and xiε is the ith component mollified. Thus by Lemma 9.5.7, this
function with the subscript ε is infinitely differentiable. The last two expressions in
11.16 sum to e (ε) which converges to 0 as ε → 0. Here is why.
¡ ¢ ¡ ¢
∂ ψ j aI ¡ −1 ¢ ∂ ψ j aI ¡ −1 ¢
Rjε (u) → Rj (u) a.e.
∂xk ∂xk

because of the pointwise convergence of R−1 jε to Rj

−1
which follows from Lemma 9.5.7.
In addition to this,
¡ ¢ ¡ ¢
∂ xkε , xi1 ε · · · xin−1 ε ∂ xk , xi1 · · · xin−1
→ a.e.
∂ (u1 , · · · , un ) ∂ (u1 , · · · , un )
because of this lemma used again on each of the component functions. This convergence
happens on Rj (Uj \ L) for each j. Thus e (ε) is a finite 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
m X p Z ¡ ¢ m
XX ∂ ψ j aI ¡ −1 ¢X ∂xkε
= Rjε (u) A1l du + e (ε)
j=1 Rj (Uj )
∂x k ∂ul
I k=1 l=1

where A1l is the 1lth cofactor for the determinant

¡ ¢
∂ xkε , xi1 ε · · · xin−1 ε
∂ (u1 , · · · , un )
which is determined by a particular I. I am suppressing the ε and I for the sake of
notation. Then the above reduces to
p Z n m ¡ ¢
XX X X ∂ ψ j aI ¡ −1 ¢ ∂xkε
= A1l Rjε (u) du + e (ε)
∂x k ∂ul
I j=1 Rj (Uj ) l=1 k=1
XX p Xn Z
∂ ¡ ¢
= A1l ψ j aI ◦ R−1
jε (u) du + e (ε) (11.17)
j=1 Rj (Uj ) ∂ul
I l=1

(Note l goes up to n not m.) Recall Rj (Uj ) is relatively open in Rn≤ . Consider the
integral where l > 1. Integrate first with respect to ul . In this case the boundary term
vanishes because of ψ j and you get
Z
¡ ¢
− A1l,l ψ j aI ◦ R−1
jε (u) du (11.18)
Rj (Uj )
11.6. STOKE’S THEOREM AND THE ORIENTATION OF ∂Ω 309

Next consider the case where l = 1. Integrating first with respect to u1 , the term reduces
to
Z Z
−1
¡ ¢
ψ j aI ◦ Rjε (0, u2 , · · · , un ) A11 du1 − A11,1 ψ j aI ◦ R−1
jε (u) du (11.19)
Rj Vj Rj (Uj )

where Rj Vj is an open set in Rn−1 consisting of

© ª
(u2 , · · · , un ) ∈ Rn−1 : (0, u2 , · · · , un ) ∈ Rj (Uj )

and du1 represents du2 du3 · · · dun on Rj Vj for short. Thus Vj is just the part of ∂Ω
which is in Uj and the mappings S−1 j given on Rj Vj = Rj (Uj ∩ ∂Ω) by

S−1 −1
j (u2 , · · · , un ) ≡ Rj (0, u2 , · · · , un )

are such that {(Sj , Vj )} is an atlas for ∂Ω. Then if 11.18 and 11.19 are placed in 11.17,
it follows from Lemma 11.5.1 that 11.17 reduces to
XX p Z
ψ j aI ◦ R−1
jε (0, u2 , · · · , un ) A11 du1 + e (ε)
I j=1 Rj Vj

Now as before, each ∂xsε /∂ur converges pointwise a.e. to ∂xs /∂ur , off Rj (Vj ∩ 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 ε → 0 to obtain
p Z
XX
ψ j aI ◦ R−1
j (0, u2 , · · · , un ) A11 du1
I j=1 Rj Vj
p Z
XX
= ψ j aI ◦ S−1
j (u2 , · · · , un ) A11 du1
I j=1 Sj Vj

p Z ¡ ¢
XX ∂ xi1 · · · xin−1
= ψ j aI ◦ S−1
(u2 , · · · , un )
j (0, u2 , · · · , un ) du1 (11.20)
∂ (u2 , · · · , un )
I j=1 Sj Vj
R
This of course is the definition of ∂Ω ω provided {Sj , Vj } is an oriented atlas. Note
the integral is well defined because of the assumption that Ri (L ∩ Ui ∩ ∂Ω) has mn−1
measure zero. That ∂Ω is orientable and that this atlas is an oriented atlas is shown
next. I will write u1 ≡ (u2 , · · · , un ).
R What if spt aI ⊆ K ⊆ Ui ∩ Uj for each I? Then using Lemma 11.5.3 it follows that
dω =
¡ ¢
XZ ∂ xi1 · · · xin−1
−1
aI ◦ Sj (u2 , · · · , un ) (0, u2 , · · · , un ) du1
Sj (Vj ∩Vj ) ∂ (u2 , · · · , un )
I

This is done by using a partition of unity which has the property that ψ j equals 1 on
K which forces all the other ψ k to equal Rzero there. Using the same trick involving a
judicious choice of the partition of unity, dω is also equal to
¡ ¢
XZ ∂ xi1 · · · xin−1
aI ◦ S−1
i (v2 , · · · , v n ) (0, v2 , · · · , vn ) dv1
Si (Vj ∩Vj ) ∂ (v2 , · · · , vn )
I

Since Si (L ∩ Ui ) , Sj (L ∩ Uj ) have measure zero, the above integrals may be taken over

Sj (Vj ∩ Vj \ L) , Si (Vj ∩ Vj \ L)
310 INTEGRATION OF DIFFERENTIAL FORMS

R
respectively. Also these are equal, both being dω. To simplify the notation, let π I
denote the projection onto the components corresponding to I. Thus if I = (i1 , · · · , in ) ,

π I x ≡ (xi1 , · · · , xin ) .

Then writing in this simpler notation, the above would say

XZ
aI ◦ S−1 −1
j (u1 ) det Dπ I Sj (u1 ) du1
I Sj (Vj ∩Vj \L)

XZ
= aI ◦ S−1 −1
i (v1 ) det Dπ I Si (v1 ) dv1
I Si (Vj ∩Vj \L)
R
and both equal to dω. Thus using the change of variables formula, Theorem 9.9.10,
it follows the second of these equals
XZ ¡ ¢¯ ¡ ¢ ¯
aI ◦ S−1 −1
Si ◦ S−1 ¯ −1
(u1 )¯ du1
j (u1 ) det Dπ I Si j (u1 ) det D Si ◦ Sj
I Sj (Vj ∩Vj \L)

¡ ¢ (11.21)
I want to argue det D Si ◦ S−1
j (u1 ) ≥ 0. Let A be the open subset of Sj (Vj ∩ Vj \ L)
on which for δ > 0, ¡ ¢
det D Si ◦ S−1j (u1 ) < −δ (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
fI be a smooth function which vanishes off a compact subset of S−1 j (A) the above
argument and the chain rule imply
XZ
fI ◦ S−1 −1
j (u1 ) det Dπ I Sj (u1 ) du1
I Sj (Vj ∩Vj \L)
XZ
= fI ◦ S−1 −1
j (u1 ) det Dπ I Sj (u1 ) du1
I A

XZ ¡ ¢ ¡ ¢
fI ◦ S−1 −1
j (u1 ) det Dπ I Si Si ◦ S−1 −1
j (u1 ) det D Si ◦ Sj (u1 ) du1
I A

Now from 11.21, this equals

XZ ¡ ¢ ¡ ¢
=− fI ◦ S−1 −1
j (u1 ) det Dπ I Si Si ◦ S−1 −1
j (u1 ) det D Si ◦ Sj (u1 ) du1
I A

and consequently
XZ ¡ ¢ ¡ ¢
0=2 fI ◦ S−1 −1
j (u1 ) det Dπ I Si Si ◦ S−1 −1
j (u1 ) det D Si ◦ Sj (u1 ) du1
I A

© ª∞
Now for each I, let fIk ◦ S−1
j be a sequence of bounded functions having compact
k=1 ¡ ¢
support in A which converge pointwise to det Dπ I S−1 i Si ◦ S−1
j (u1 ) . Then it follows
from the Vitali convergence theorem, one can pass to the limit and obtain
Z X
¡ ¡ ¢¢2 ¡ ¢
0 = 2 det Dπ I S−1
i Si ◦ S−1
j (u1 ) det D Si ◦ S−1
j (u1 ) du1
A
ZI X
¡ ¡ ¢¢2
≤ −2δ det Dπ I S−1
i Si ◦ S−1
j (u1 ) du1
A I
11.7. GREEN’S THEOREM, AN EXAMPLE 311

Since the integrand is continuous, this would require

det Dπ I S−1
i (v1 ) ≡ 0 (11.23)

for each I and for each v1 ∈ Si ◦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,
n×m m×n
¡ ¢ z ¡ }| ¢{z −1
}| {
−1 −1
D Sj ◦ Si (v1 ) = DSj Si (v1 ) DSi (v1 )
cannot be identically 0. By the Binet Cauchy theorem, at least some

Dπ I S−1
i (v1 ) 6= 0

contradicting 11.23. Thus A = ∅ and since δ > 0 was arbitrary, this shows
¡ ¢
det D Si ◦ S−1
j (u1 ) ≥ 0.

Hence this is an oriented atlas as claimed. This proves the theorem. ¥

Theorem 11.6.1 Let Ω be an oriented P C 1 manifold and let

X
ω= aI (x) dxi1 ∧ · · · ∧ dxin−1 .
I
¡ ¢ p
where each aI is C 1 Ω . For {Uj , Rj }j=0 an oriented atlas for Ω where Rj (Uj ) is a
relatively open set in
{u ∈ Rn : u1 ≤ 0} ,
define an atlas for ∂Ω, {Vj , Sj } where Vj ≡ ∂Ω ∩ Uj and Sj is just the restriction of Rj
to Vj . Then this is an oriented atlas for ∂Ω and
Z Z
ω= dω
∂Ω Ω

where the two integrals are taken with respect to the given oriented atlass.

11.7 Green’s Theorem, An Example

Green’s 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 Rn with sufficiently smooth boundary
using the methods of differential forms described above.

11.7.1 An Oriented Manifold

A bounded open subset, Ω, of Rn , n ≥ 2 has P C 1 boundary and lies locally on one side
of its boundary if it satisfies the following conditions.
For each p ∈ ∂Ω ≡ Ω \ Ω, there exists an open set, Q, containing p, an open interval
(a, b), a bounded open set B ⊆ Rn−1 , and an orthogonal transformation R such that
det R = 1,
(a, b) × B = RQ,
and letting W = Q ∩ Ω,

RW = {u ∈ Rn : a < u1 < g (u2 , · · · , un ) , (u2 , · · · , un ) ∈ B}

312 INTEGRATION OF DIFFERENTIAL FORMS

g (u2 , · · · , un ) < b for (u2 , · · · , un ) ∈ B. Also g vanishes outside some compact set in
Rn−1 and g is continuous.

R (∂Ω ∩ Q) = {u ∈ Rn : u1 = g (u2 , · · · , un ) , (u2 , · · · , un ) ∈ B} .

Note that finitely many of these sets Q cover ∂Ω because ∂Ω is compact. Assume there
exists a closed subset of ∂Ω, L such that the closed set SQ defined by

{(u2 , · · · , un ) ∈ B : (g (u2 , · · · , un ) , u2 , · · · , un ) ∈ R (L ∩ Q)} (11.24)

has mn−1 measure zero. g ∈ C 1 (B \ SQ ) and all the partial derivatives of g are uni-
formly bounded on B \ SQ . The following picture describes the situation. The pointy
places symbolize the set L.

Q
R(W )
u
W R - R(Q)
x a b
Define P1 : Rn → Rn−1 by

P1 u ≡ (u2 , · · · , un )

and Σ : Rn → Rn given by
Σu ≡ u − g (P1 u) e1
≡ u − g (u2 , · · · , un ) e1
≡ (u1 − g (u2 , · · · , un ) , u2 , · · · , un )
Thus Σ is invertible and
Σ−1 u = u + g (P1 u) e1
≡ (u1 + g (u2 , · · · , un ) , u2 , · · · , un )
For x ∈ ∂Ω∩Q, it follows the first component of Rx is g (P1 (Rx)) . Now define R :W →
Rn≤ as
u ≡ Rx ≡ Rx − g (P1 (Rx)) e1 ≡ ΣRx
and so it follows
R−1 = R∗ Σ−1 .
These mappings R involve first a rotation followed by a variable sheer in the direction
of the u1 axis. From the above description, R (L ∩ Q) = 0 × SQ , a set of mn−1 measure
zero. This is because
(u2 , · · · , un ) ∈ SQ
if and only if
(g (u2 , · · · , un ) , u2 , · · · , un ) ∈ R (L ∩ Q)
if and only if

(0, u2 , · · · , un ) ≡ Σ (g (u2 , · · · , un ) , u2 , · · · , un )
∈ ΣR (L ∩ Q) ≡ R (L ∩ Q) .

Since ∂Ω is compact, there are finitely many of these open sets, Q1 , · · · , Qp which
cover ∂Ω. Let the orthogonal transformations and other quantities described above
11.7. GREEN’S THEOREM, AN EXAMPLE 313

also be indexed by k for k = 1, · · · , p. Also let Q0 be an open set with Q0 ⊆ Ω and

Ω is covered by Q0 , Q1 , · · · , Qp . Let u ≡ R0 x ≡ x − ke1 where k is large enough
that R0 Q0 ⊆ Rn< . Thus in this case, the orthogonal transformation R0 equals I and
Σ0 x ≡ x − ke1 . I claim Ω is an oriented manifold with boundary and the charts are
(Wi , Ri ) .
To see this is an oriented atlas for the manifold, note that for a.e. points of
Ri (Wi ∩ Wj ) the function g is differentiable. Then using the above notation, at these
points Rj ◦ R−1
i is of the form
Σj Rj Ri∗ Σ−1
i
and it is a one to one mapping. What is the determinant of its derivative? By the chain
rule,
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
D Σj Rj Ri∗ Σ−1
i = DΣj Rj Ri∗ Σ−1
i DRj Ri∗ Σ−1
i DRi∗ Σ−1i DΣ−1
i

However, ¡ ¢
det (DΣj ) = 1 = det DΣ−1
j
¡ ¢
and det (Ri ) = det (Ri∗ ) = 1 by assumption. Therefore, for a.e. u ∈ Rj ◦ R−1
i (A) ,
¡ ¡ −1
¢ ¢
det D Rj ◦ Ri (u) > 0.
By Proposition 11.1.7 Ω is indeed an oriented manifold with the given atlas.

11.7.2 Green’s Theorem

The general Green’s theorem is the following. It follows from Stoke’s theorem above.
Theorem 11.6.1.
First note that since Ω ⊆ Rn , there is no loss of generality in writing
n
X
ω= dk ∧ · · · ∧ dxn
ak (x) dx1 ∧ · · · ∧ dx
k=1

where the hat indicates the dxk is omitted. Therefore,

n X
X n
∂ak (x) dk ∧ · · · ∧ dxn
dω = dxj ∧ dx1 ∧ · · · ∧ dx
∂xj
k=1 j=1
n
X ∂ak (x) k−1
= (−1) dx1 ∧ · · · ∧ dxn
∂xk
k=1
This follows from the definition of integration of differential forms. If there is a repeat
in the dxj then this will lead to a determinant of a matrix which has two equal columns
in the definition of the integral of the differential form. Also, from the definition, and
again from the properties of the determinant, when you switch two dxk it changes the
sign because it is equivalent to switching two columns in a determinant. Then with
these observations, Green’s theorem follows.
Theorem 11.7.1 Let Ω be a bounded open set in Rn , n ≥ 2 and let it have
1
P C boundary and lie locally on one side of its boundary as described above. Also let
n
X
ω= dk ∧ · · · ∧ dxn
ak (x) dx1 ∧ · · · ∧ dx
k=1
¡ ¢
be a differential form where aI is assumed to be in C 1 Ω . Then
Z Z X n
∂ak (x) k−1
ω= (−1) dmn
∂Ω Ω ∂x k
k=1
314 INTEGRATION OF DIFFERENTIAL FORMS

Proof: From the definition and using the usual technique of ignoring the exceptional
set of measure zero,
Z p X n Z ¡ ¢
X ∂ak R−1i (u) k−1 ¡ ¢ ∂ (x1 · · · xn )
dω ≡ (−1) ψ i R−1
i (u) du
Ω i=1 Ri Wi ∂xk ∂ (u1 · · · un )
k=1

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
Xp X n Z Z X p X n
∂ak (x) k−1 ∂ak (x) k−1
(−1) ψ i (x) dx = (−1) ψ i (x) dmn
i=1 k=1 Wi
∂x k Ω i=1 k=1 ∂x k
Z X n
∂ak (x) k−1
= (−1) dmn
Ω ∂xk
k=1

This proves the theorem. ¥

This Green’s 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 Green’s
theorem is discussed. The following is a specialization to the familiar calculus theorem.
Example 11.7.2 The usual Green’s theorem follows from the above specialized to the
case of n = 2. Z Z
P (x, y) dx + Q (x, y) dy = (Qx − Py ) dxdy
∂Ω Ω

This follows because the differential form on the left is of the form
c + Qdx
P dx ∧ dy c ∧ dy

and so, as above, the derivative of this is

Py dy ∧ dx + Qx dx ∧ dy = (Qx − Py ) dx ∧ dy
It is understood in all the above that the oriented atlas for Ω is the one described there
where
∂ (x, y)
= 1.
∂ (u1 , u2 )
Saying Ω is P C 1 reduces in this case to saying ∂Ω is a piecewise C 1 closed curve
which is often the case stated in calculus courses. From calculus, the orientation of
∂Ω was defined 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 first the rotation and then the shear which were used to specify
an atlas for Ω.

Q u2
6
Σ
R R(W ) -
W - R(W )
Ω
11.8. THE DIVERGENCE THEOREM 315

The vertical arrow at the end indicates the direction of increasing u2 . The vertical side
of R (W ) shown there corresponds to the curved side in R (W ) which corresponds to
the part of ∂Ω 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 ∂Ω preserves the direction of motion relative to the surface. This
is because orthogonal transformations in R2 whose determinants are 1 correspond to
rotations. Thus increasing u2 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 Ω 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 Green’s theorem, one can quickly obtain a general Divergence theorem for Ω as
described above in Section 11.7.1. First note that from the above description of the Rj ,
¡ ¢
∂ xk , xi1 , · · · xin−1
= sgn (k, i1 · · · , in−1 ) .
∂ (u1 , · · · , un )
¡ ¢
Let F (x) be a C 1 Ω; Rn vector field. Say F = (F1 , · · · , Fn ) . Consider the differential
form
X n
ω (x) ≡ Fk (x) (−1)
k−1 dk ∧ · · · ∧ dxn
dx1 ∧ · · · ∧ dx
k=1

where the hat means dxk is being left out. Then

n X
X n
∂Fk k−1 dk ∧ · · · ∧ dxn
dω (x) = (−1) dxj ∧ dx1 ∧ · · · ∧ dx
∂xj
k=1 j=1
Xn
∂Fk
= dx1 ∧ · · · ∧ dxk ∧ · · · ∧ dxn
∂xk
k=1
≡ div (F) dx1 ∧ · · · ∧ dxk ∧ · · · ∧ dxn

The assertion between the first and second lines follows right away from properties of
determinants and the definition of the integral of the above wedge products in terms
of determinants. From Green’s theorem and R the change of variables formula applied to
the individual terms in the description of Ω dω
Z
div (F) dx =
Ω

p Z
X n
X k−1 ck · · · , xn ) ¡
∂ (x1 , · · · x ¢
(−1) ψ j Fk ◦ R−1
j (0, u2 , · · · , un ) du1 ,
j=1 Bj k=1 ∂ (u2 , · · · , un )

du1 short for du2 du3 · · · dun .

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 ψ i . Replace F in
316 INTEGRATION OF DIFFERENTIAL FORMS

© ª
the above with ψ s F. Next let η j be a partition of unity η j ≺ Qj such that η s = 1 on
spt ψ s . This partition of unity exists by Lemma 11.5.3. Then
Z
div (ψ s F) dx =
Ω
p Z
X n
X k−1 ck · · · , xn ) ¡
∂ (x1 , · · · x ¢
(−1) η j ψ s Fk ◦ R−1
j (0, u2 , · · · , un ) du1
j=1 Bj k=1 ∂ (u2 , · · · , un )
Z n
X k−1 ∂ (x1 , · · · x
ck · · · , xn )
= (−1) (ψ s Fk ) ◦ R−1
s (0, u2 , · · · , un ) du1 (11.25)
Bs k=1 ∂ (u2 , · · · , un )
because since η s = 1 on spt ψ s , it follows all the other η j equal zero there.
Consider the vector N defined for u1 ∈ Rs (Ws \ L) ∩ Rn0 whose k th component is

k−1 ∂ (x1 , · · · x
ck · · · , xn ) k+1 ∂ (x1 , · · · x
ck · · · , xn )
Nk = (−1) = (−1) (11.26)
∂ (u2 , · · · , un ) ∂ (u2 , · · · , un )
Suppose you dot this vector with a tangent vector ∂R−1
s /∂ui . This yields
X k+1 ck · · · , xn ) ∂xk
∂ (x1 , · · · x
(−1) =0
∂ (u2 , · · · , un ) ∂ui
k

because it is the expansion of

¯ ¯
¯ x1,i x1,2 ··· x1,n ¯
¯ ¯
¯ x2,i x2,2 ··· x2,n ¯
¯ ¯
¯ .. .. .. .. ¯,
¯ . . . . ¯
¯ ¯
¯ xn,i xn,2 ··· xn,n ¯

a determinant with two equal columns provided i ≥ 2. Thus this vector is at least in
some sense normal to ∂Ω. If i = 1, then the above dot product is just
∂ (x1 · · · xn )
=1
∂ (u1 · · · un )
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 Rs (Ws )
u2 , · · · , un

Rs (Ws ) -
e1

u1
We got this by first doing a rotation of a piece of Ω and then a shear in the direction of
e1 . Also it was shown above
T
R−1 ∗
s (u) = Rs (u1 + g (u2 , · · · , un ) , u2 , · · · , un )

where Rs∗ is a rotation, an orthogonal transformation whose determinant is 1. Letting

x = R−1s (u) , the above discussion shows ∂x/∂u1 · N = 1 > 0. Thus it is also the case
that for h small and positive,
≈−∂x/∂u1
z }| {
x (u − he1 ) − x (u)
·N<0
h
11.8. THE DIVERGENCE THEOREM 317

Hence if θ is the angle between N and x(u−heh1 )−x(u) , it must be the case that θ > π/2.
However,
x (u − he1 ) − x (u)
h
points in to Ω for small h > 0 because x (u − he1 ) ∈ Ws while x (u) is on the boundary
of Ws . Therefore, N should be pointing away from Ω at least at the points where
u → x (u) is differentiable. 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
à n µ ¶2 !1/2
X ∂ (x1 , · · · x
ck · · · , xn )
∂ (u2 , · · · , un )
k=1

and by the Binet Cauchy theorem this equals

¡ ∗ ¢1/2
det DR−1 −1
s (u1 ) DRs (u1 ) ≡ J (u1 )

where as usual u1 = (u2 , · · · , un ). Thus the expression in 11.25 reduces to

Z
¡ ¢ ¡ −1 ¢
ψ s F ◦ R−1
s (u1 ) · n Rs (u1 ) J (u1 ) du1 .
Bs

The integrand
¡ ¢ ¡ −1 ¢
u1 → ψ s F ◦ R−1
s (u1 ) · n Rs (u1 )

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 {sk } 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,
Z
¡ ¢ ¡ −1 ¢
ψ s F ◦ R−1
s (u1 ) · n Rs (u1 ) J (u1 ) du1
Bs
Z
¡ ¡ ¢¢
= lim sk Rs R−1
s (u1 ) J (u1 ) du1
k→∞ Bs

Z
= lim sk (Rs (x)) dσ n−1
k→∞ Ws ∩∂Ω
Z
= ψ s (x) F (x) · n (x) dσ n−1
Ws ∩∂Ω
Z
= ψ s (x) F (x) · n (x) dσ n−1
∂Ω

Recall the exceptional set on ∂Ω has σ n−1 measure zero. Upon summing over all s using
that the ψ s add to 1,
XZ Z
ψ s (x) F (x) · n (x) dσ n−1 = F (x) · n (x) dσ n−1
s ∂Ω ∂Ω
318 INTEGRATION OF DIFFERENTIAL FORMS

On the other hand, from 11.25, the left side of the above equals
XZ XZ X n
div (ψ s F) dx = ψ s,i Fi + ψ s Fi,i dx
s Ω s Ω i=1
Z n
à !
X X
= div (F) dx + ψs Fi
Ω i=1 s ,i
Z
= div (F) dx
Ω

¥
This proves the following general divergence theorem.
Theorem 11.8.1 Let Ω be a bounded open set having P C 1 boundary as de-
scribed above. Also let ¡F be a¢ vector field with the property that for Fk a component
function of F, Fk ∈ C 1 Ω; Rn . Then there exists an exterior normal vector n which is
defined σ n−1 a.e. (off the exceptional set L) on ∂Ω such that
Z Z
F · ndσ n−1 = div (F) dx
∂Ω Ω

It is worth noting that everything above will work if you relax the requirement in
P C 1 which requires the partial derivatives be bounded off an exceptional set. Instead,
it would suffice to say that for some p > n all integrals of the form
Z ¯ ¯
¯ ∂xk ¯p
¯ ¯ du
¯ ¯
Ri (Ui ) ∂uj

are bounded. Here xk is the k th component of R−1 i . This is because this condition will
suffice 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
Rademacher’s theorem which gives the almost everywhere differentiability 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 mollifier and then do the hard computations with
the mollified function exploiting equality of mixed partial derivatives and then pass to
the limit.

11.9 Spherical Coordinates

Consider the following picture.

WI (E)
0
11.9. SPHERICAL COORDINATES 319

Definition 11.9.1 The symbol WI (E) represents the piece of a wedge between
the two concentric spheres such that the points x ∈ WI (E) have the property that
x/ |x| ∈ E, a subset of the unit sphere in Rn , S n−1 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
first gives the existence of a countable basis for Rn . 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 Rn which have centers
x ∈ Qn 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 Qn in Rn , there exists x ∈ B (y, R/10) ∩ Qn . 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 Lindelöf 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 B0 denote those sets of B in Lemma 11.9.2 which are contained in some
set of C. By this lemma, it follows ∪B 0 = U . Now use axiom of choice to select for each
B0 a single set of C containing it. Denote the resulting countable collection C 0 . Then
U = ∪B0 ⊆ ∪C 0 ⊆ U
This proves the theorem.
Now consider all the open subsets of Rn \ {0} . If U is any such open set, it is clear
that if y ∈ U, then there exists a set open in S n−1 , E and an open interval I such that
y ∈ WI (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 WI (E) for E open in S n−1 .
The divergence theorem and Green’s theorem hold for sets WI (E) whenever E is
the intersection of S n−1 with a finite intersection of balls. This is because the resulting
set has P C 1 boundary. Therefore, from the divergence theorem and letting I = (0, 1)
Z Z Z
x
div (x) dx = x· dσ + x · ndσ
WI (E) E |x| straight part

where I am going to denote by σ the measure on S n−1 which corresponds to the diver-
gence theorem and other theorems given above. On the straight parts of the boundary
of WI (E) , the vector field x is parallel to the surface while n is perpendicular to it, all
this off a set of measure zero of course. Therefore, the integrand vanishes and the above
reduces to
nmn (WI (E)) = σ (E)
Now let G denote those Borel sets of S n−1 such that the above holds for I = (0, 1) ,
both sides making sense because both E and WI (E) are Borel sets in S n−1 and Rn \{0}
respectively. Then G contains the π system of sets which are the finite intersection of
balls with S n−1 . Also if {Ei } are disjoint sets in G, then WI (∪∞ ∞
i=1 Ei ) = ∪i=1 WI (Ei )
and so
nmn (WI (∪∞
i=1 Ei )) = nmn (∪∞
i=1 WI (Ei ))
∞
X
= n mn (WI (Ei ))
i=1
∞
X
= σ (Ei ) = σ (∪∞
i=1 Ei )
i=1
320 INTEGRATION OF DIFFERENTIAL FORMS

and so G is closed with respect to countable disjoint unions. Next let E ∈ G. Then
¡ ¡ ¢¢ ¡ ¡ ¢¢
nmn WI E C + nmn (WI (E)) = nmn WI S n−1
¡ ¢ ¡ ¢
= σ S n−1 = σ (E) + σ E C

Now subtracting the equal quantities nmn (WI (E)) and σ (E) from both sides yields
E C ∈ G also. Therefore, by the Lemma on π systems Lemma 9.1.2, it follows G contains
the σ algebra generated by these special sets E the intersection of finitely many open
balls with S n−1 . Therefore, since any open set is the countable union of balls, it follows
the sets open in S n−1 are contained in this σ algebra. Hence G equals the Borel sets.
This has proved the following important theorem.

Theorem 11.9.4 Let σ be the Borel measure on S n−1 which goes with the di-
vergence theorems and other theorems like Green’s and Stoke’s theorem. Then for all E
Borel,
σ (E) = nmn (WI (E))
where I = (0, 1). Furthermore, WI (E) is Borel for any interval. Also
¡ ¢ ¡ ¢ ¡ ¢
mn W[a,b] (E) = mn W(a,b) (E) = (bn − an ) mn W(0,1) (E)

Proof: To show WI (E) is Borel for any I first suppose I is open of the form (0, r).
Then
WI (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],

WI (E) = ∩∞
n=1 W(0, 1 +r ) (E)
n

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 different way than before.
Note that if ρ = |x| and ω ≡ x/ |x| , then x = ρω. Also the map which takes
(0, ∞) × S n−1 to Rn \ {0} given by (ρ, ω) → ρω = x is one to one and onto and
continuous. In addition to this, it follows right away from the definition that if I is any
interval and E ⊆ S n−1 ,
XWI (E) (ρω) = XE (ω) XI (ρ)

Lemma 11.9.5 For any Borel set F ⊆ Rn ,

Z ∞ Z
mn (F ) = ρn−1 XF (ρω) dσdρ
0 S n−1

and the iterated integral on the right makes sense.

Proof: First suppose F = WI (E) where E is Borel in S n−1 and I is an interval

having endpoints a ≤ b. Then
Z ∞Z Z ∞Z
n−1
ρ XWI (E) (ρω) dσdρ = ρn−1 XE (ω) XI (ρ) dσdρ
0 S n−1 0 S n−1
11.10. EXERCISES 321

Z b Z b ¡ ¢
= ρn−1 σ (E) dρ = ρn−1 nmn W(0,1) (E) dρ
a a
n
¡ n
¢
= (b − a ) mn W(0,1) (E)
and by Theorem 11.9.4, this equals mn (WI (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 ρ = 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 π system of sets of the form WI (E) and is closed with
respect to countable unions of disjoint sets and complements. Therefore, it equals the
Borel sets. Thus
Z ∞Z
mn (F ∩ B (0, M )) = ρn−1 XF ∩B(0,M ) (ρω) dσdρ
0 S n−1

Now let M → ∞ and use the monotone convergence theorem. This proves the lemma.
The lemma implies right away that for s a simple function
Z Z ∞Z
sdmn = ρn−1 s (ρω) dσdρ
Rn 0 S n−1

Now the following polar coordinates theorem follows.

Theorem 11.9.6 Let f ≥ 0 be Borel measurable. Then

Z Z ∞Z
f dmn = ρn−1 f (ρω) dσdρ
Rn 0 S n−1

and the iterated integral on the right makes sense.

Proof: By Theorem 7.7.12 there exists a sequence of nonnegative simple functions

{sk } which increases to f . Therefore, from the monotone convergence theorem and
what was shown above,
Z Z
f dmn = lim sk dmn
Rn k→∞ Rn
Z ∞Z
= lim ρn−1 sk (ρω) dσdρ
k→∞ 0 S n−1
Z ∞Z
= ρn−1 f (ρω) dσdρ
0 S n−1

This proves the theorem.

11.10 Exercises
1. Let X
ω (x) ≡ aI (x) dxI
I
m
be a differential form where x ∈ R and the I are increasing lists of n indices
taken from 1, · · · , m. Also assume each aI (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 (ω)) = 0. To show this,
first explain why
dxi ∧ dxj ∧ dxI = −dxj ∧ dxi ∧ dxI
322 INTEGRATION OF DIFFERENTIAL FORMS

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 differential form ω with
the property that dω = 0 this is called a closed form. If ω = dα, then ω is called
exact. Thus every closed form is exact provided you have sufficient smoothness
on the coefficients of the differential form.
2. Recall that in the definition of area measure, you use
¡ ∗ ¢1/2
J (u) = det DR−1 (u) DR−1 (u)
Now in the special case of the manifold of Green’s theorem where
R−1 (u2 , · · · , un ) = R∗ (g (u2 , · · · , un ) , u2 , · · · , un ) ,
show s µ ¶2 µ ¶2
∂g ∂g
J (u) = 1+ + ··· +
∂u2 ∂un
3. Let u1 , · · · , up be vectors in Rn . Show det M ≥ 0 where Mij ≡ ui · uj . 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
 ∗ 
u1
 ¡ ¢
U ∗ U ≡  ...  u1 · · · up
u∗p
and then showing that U ∗ U has all nonnegative eigenvalues.
4. Suppose {v1 , · · · , vn } are n vectors in Rm for m ≥ n. Show that the only appro-
priate definition of the volume of the n dimensional parallelepiped determined by
these vectors,  
X n 
sj vj : sj ∈ [0, 1]
 
j=1

is
1/2
det (M ∗ M )
where M is the m × n matrix which has columns v1 , · · · , vn . 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 n − 1 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 v1 , · · · , vn−1 is
 
u1 u2 ··· un
 v11 v12 ··· v1n 
 
det  .. .. .. 
 . . . 
vn−1,1 vn−1,2 · · · vn−1,n
where {u1 , · · · , un } is an orthonormal basis for span (v1 , · · · , vn ) and vij is the
j th component of vi with respect to this orthonormal basis. Then argue that
if you replace the top line with vn1 , · · · , vnn , the absolute value of the resulting
determinant is the appropriate definition 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 definition of area on a manifold given above reasonable and what is
its geometric meaning? Each function

ui → R−1 (u1 , · · · , un )

yields a curve which lies in Ω. Thus R−1 ,ui is a vector tangent to this curve and
R−1
,ui dui is an “infinitesimal” vector tangent to the curve. Now use the previous
problem to see that when you find the area of a set on Ω, you are essentially
summing the volumes of infinitesimal parallelepipeds which are “tangent” to Ω.

6. Let Ω be a bounded open set in Rn with P C 1 boundary

¡ ¢ or more generally one for
which the divergence theorem holds. Let u, v ∈ C 2 Ω . Then
Z Z µ ¶
∂u ∂v
(v∆u − u∆v) dx = v −u dσ n−1
Ω ∂Ω ∂n ∂n

Here
∂u
≡ ∇u · n
∂n
where n is the unit outer normal described above. Establish this formula which is
known as Green’s identity. Hint: You might establish the following easy identity.

∇ · (v∇u) − v∆u = ∇v · ∇u.

Pn
Recall ∆u ≡ k=1 uxk xk and ∇u = (ux1 , · · · , uxn ) while

∇ · F ≡ f1x1 + · · · + fnxn ≡ div (F)

324 INTEGRATION OF DIFFERENTIAL FORMS
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 differential 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 differential equations course.

12.1 Balls
Recall, B (x, r) denotes the set of all y ∈ Rn 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 αn rn where αn will denote the
volume of the unit ball in Rn . 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,
Z Z
x
div x dx = x· dσ n−1
B(0,r) ∂B(0,r) |x|

x
because the unit outward normal on ∂B (0, r) is |x| . Therefore,

nαn rn = rσ n−1 (∂B (0, r))

and so
σ n−1 (∂B (0, r)) = nαn rn−1 .
© ª
You recall the surface area of S 2 ≡ x ∈ R3 : |x| = r is given by 4πr2 while the volume
of the ball, B (0, r) is 43 πr3 . 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 ω n denote the area of the sphere S n−1 = {x ∈ Rn : |x| = 1} . I
just showed that
ω n = nαn . (12.1)

I want to find αn now and also to get a relationship between ω n and ω n−1 . Consider
the following picture of the ball of radius ρ seen on the side.

325
326 THE LAPLACE AND POISSON EQUATIONS

r
y ρ Rn−1

Taking slices at height y as shown and using that these slices have n − 1 dimensional
area equal to αn−1 rn−1 , it follows from Fubini’s theorem
Z ρ
n
¡ ¢(n−1)/2
αn ρ = 2 αn−1 ρ2 − y 2 dy (12.2)
0
√
Lemma 12.1.1 Γ (1/2) = π
Proof: µ ¶ Z ∞
1
Γ ≡ e−t t−1/2 dt
2 0

Now change the variables letting t = s2 so dt = 2sds and the integral becomes
Z ∞ Z ∞
2 2
2 e−s ds = e−s ds
0 −∞

¡ ¢ R∞ ¡ ¢2 R∞ R∞
Thus Γ 12 = −∞ e−x dx so Γ 12 = −∞ −∞ e−(x +y ) dxdy and by polar coordi-
2 2 2

nates and changing the variables, this is just

Z 2π Z ∞
2
e−r rdrdθ = π
0 0
¡1¢ √
Therefore, Γ 2 = π as claimed. This proves the lemma. ¥

Theorem 12.1.2 αn = π n/2

Γ( n
where Γ denotes the gamma function, defined
2 +1)
for α > 0 by Z ∞
Γ (α) ≡ e−t tα−1 dt.
0

Proof: Recall that Γ (α + 1) = αΓ (α) . (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
π 1/2
α1 ≡ 2 = ?
Γ (3/2)
Using the identity Γ (α + 1) = αΓ (α) , the above equals

π 1/2
=2
(1/2) Γ (1/2)

from the above lemma. Now suppose the theorem is true for n. Then letting ρ = 1,
12.2 implies
Z 1
π n/2 ¡ ¢n/2
αn+1 = 2 ¡ n ¢ 1 − y2 dy
Γ 2 +1 0
12.2. POISSON’S PROBLEM 327

Now change variables. Let u = y 2 . Then

Z 1
π n/2 1
n
αn+1 = ¡n ¢ u−1/2 (1 − u) 2 du
Γ 2 +1 0
Z 1
π n/2 n+2
−1
= ¡n ¢ u(1/2)−1 (1 − u) 2 du
Γ 2 +1 0
µ ¶
π n/2 1 n+2
= ¡ ¢ B ,
Γ n2 + 1 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
¡ ¢
π n/2 Γ (1/2) Γ n+2
¡n ¢ ¡ n+2 12¢
Γ 2 +1 Γ 2 + 2
π 1/2 π (n+1)/2
= π n/2 ¡ 1 ¢ = ¡ ¢
Γ 2 n + 32 Γ n+1
2 +1

and this gives the correct formula for αn+1 . This proves the theorem. ¥

12.2 Poisson’s Problem

The Poisson problem is to find 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 P C 1 manifold. When f = 0 this is called Laplace’s 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 6= 0, it is called Poisson’s
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 © ª
max u (x) : x ∈ U = max {u (x) : x ∈ ∂U } .

Proof: Suppose not. Then there exists x0 ∈ U such that

u (x0 ) > max {u (x) : x ∈ ∂U } .

2
Consider wε (x) ≡ u (x) + ε |x| . I claim that for small enough ε > 0, the function wε
also has this property. If not, there exists xε ∈ ∂U such that wε (xε ) ≥ wε (x) for all
x ∈ U. But since U is bounded, it follows the points, xε are in a compact set and so
there exists a subsequence, still denoted by xε such that as ε → 0, xε → x1 ∈ ∂U. But
then for any x ∈ U,
u (x0 ) ≤ wε (x0 ) ≤ wε (xε )
328 THE LAPLACE AND POISSON EQUATIONS

and taking a limit as ε → 0 yields

u (x0 ) ≤ u (x1 )

contrary to the property of x0 above. It follows that my claim is verified. Pick such an
ε. Then wε assumes its maximum value in U say at x2 . Then by the second derivative
test,
∆wε (x2 ) = ∆u (x2 ) + 2ε ≤ 0
which requires ∆u (x2 ) ≤ −2ε, 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 satisfies the same conditions as u. Thus −u ≤ 0 and so u ≥ 0.
Therefore, u = 0 as claimed. This proves the corollary. ¥
Define ½
ln |x| if n = 2
rn (x) ≡ 1
if n > 2 .
|x|n−2

Then it is fairly routine to verify the following Lemma.

Lemma 12.2.3 For rn given above,

∆rn = 0.

Proof: I will verify the case where n ≥ 3 and leave the other case for you.
à n !−(n−2)/2  −n/2
X X
Dxi x2i = − (n − 2) xi  x2j 
i=1 j

Therefore,
 −(n+2)/2  
X n
X
Dxi (Dxi (rn )) =  x2j  (n − 2) nx2i − x2j  .
j j=1

It follows
  −(n+2)
2
 
X n
X n X
X n
∆rn =  x2j  (n − 2) n x2i − x2j  = 0.
j i=1 i=1 j=1

This proves the lemma. ¥

From now on assume ∂U is P C 1 to be specific.
Now let Uε be as indicated in the following picture. I have taken out a ball of radius
ε which is centered at the point, x ∈ U .
12.2. POISSON’S PROBLEM 329

εµ
x Uε
B(x, ε) = B²

Then the divergence theorem will continue to hold for Uε (why?) and so¡ I ¢can use
Green’s identity, Problem 6 on Page 323 to write the following for u, v ∈ C 2 U .
Z µ Z ¶ Z µ ¶
∂v ∂u ∂v ∂u
(u∆v − v∆u) dx = u −v dσ − u −v dσ (12.4)
Uε ∂U ∂n ∂n ∂Bε ∂n ∂n

Now, letting x ∈ U, I will pick for v the function,

v (y) ≡ rn (y − x) − ψ x (y) (12.5)

where ψ x is a function which is chosen such that on ∂U,

ψ x (y) = rn (y − x)
¡ ¢
so that 12.5 vanishes for y ∈ ∂U and ψ x is in C 2 U and also satisfies

∆ψ 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
Z Z Z µ ¶
∂v ∂v ∂u
− v∆udx = u dσ − u −v dσ. (12.6)
Uε ∂U ∂n ∂Bε ∂n ∂n

The idea now is to let ε → 0 and see what happens. Consider the term
Z
∂u
v dσ.
∂Bε ∂n
¡ ¢ ¡ ¢
The area is O εn−1 while the integrand is O ε−(n−2) in the case where n ≥ 3. In the
case where n = 2, the area is O (ε) and the integrand is O (|ln |ε||) . Now you know that
limε→0 ε ln |ε| = 0 and so in the case n = 2, this term converges to 0 as ε → 0. In the
case that n ≥ 3, it also converges to zero because in this case the integral is O (ε) .
Next consider the term
Z Z µ ¶
∂v ∂rn ∂ψ x
− u dσ = − u (y) (y − x) − (y) dσ.
∂Bε ∂n ∂Bε ∂n ∂n
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 find it and this will
only be possible for certain simple choices of U .
330 THE LAPLACE AND POISSON EQUATIONS

This term does not disappear as ε → 0. First note that since ψ x has bounded derivatives,
Z µ ¶ µ Z ¶
∂rn ∂ψ x ∂rn
lim − u (y) (y − x) − (y) dσ = lim − u (y) (y − x) dσ
ε→0 ∂Bε ∂n ∂n ε→0 ∂Bε ∂n
(12.7)
and so it is just this last item which is of concern.
First consider the case that n = 2. In this case,
à !
y1 y2
∇r2 (y) = 2, 2
|y| |y|
Also, on ∂Bε , the exterior unit normal, n, equals
1
(y1 − x1 , y2 − x2 ) .
ε
It follows that on ∂Bε ,
à !
∂r2 1 y1 − x1 y2 − x2 1
(y − x) = (y1 − x1 , y2 − x2 ) · 2, 2 = .
∂n ε |y − x| |y − x| ε

Therefore, this term in 12.7 converges to

− u (x) 2π. (12.8)
Next consider the case where n ≥ 3. In this case,
µ ¶
y1 yn
∇rn (y) = − (n − 2) n , · · · ,
|y| |y|
and the unit outer normal, n, equals
1
(y1 − x1 , · · · , yn − xn ) .
ε
Therefore,
2
∂rn (n − 2) |y − x| − (n − 2)
(y − x) = − n = .
∂n ε |y − x| εn−1
Letting ω n denote the n − 1 dimensional surface area of the unit sphere, S n−1 , it follows
that the last term in 12.7 converges to
u (x) (n − 2) ω n (12.9)
Finally consider the integral, Z
v∆udx.
Bε

Z Z
|v∆u| dx ≤ C |rn (y − x) − ψ x (y)| dy
Bε Bε
Z
≤ C |rn (y − x)| dy + O (εn )
Bε

Using polar coordinates to evaluate this improper integral in the case where n ≥ 3,
Z Z εZ
1 n−1
C |rn (y − x)| dx = C n−2
ρ dσdρ
Bε 0 S n−1 ρ
Z εZ
= C ρdσdρ
0 S n−1
12.2. POISSON’S PROBLEM 331

which converges to 0 as ε → 0. In the case where n = 2

Z Z εZ
C |rn (y − x)| dx = C ln (ρ) ρdσdρ
Bε 0 S n−1

which also converges to 0 as ε → 0. Therefore, returning to 12.6 and using the above
limits, yields in the case where n ≥ 3,
Z Z
∂v
− v∆udx = u dσ + u (x) (n − 2) ω n , (12.10)
U ∂U ∂n
and in the case where n = 2,
Z Z
∂v
− v∆udx = u dσ − u (x) 2π. (12.11)
U ∂U ∂n
These two formulas show that it is possible to represent the solutions to Poisson’s
problem provided the function, ψ x can be determined. I will show you can determine
this function in the case that U = B (0, r) .

12.2.1 Poisson’s Problem For A Ball

Lemma 12.2.4 When |y| = r and x 6= 0,
¯ ¯
¯ y |x| rx ¯¯
¯
¯ r − |x| ¯ = |x − y| ,

and for |x| , |y| < r, x 6= 0, ¯ ¯

¯ y |x| rx ¯¯
¯ − 6= 0.
¯ r |x| ¯

Proof: Suppose first that |y| = r. Then

¯ ¯2 µ ¶ µ ¶
¯ y |x| rx ¯¯ y |x| rx y |x| rx
¯
¯ r − |x| ¯ =
r
−
|x|
·
r
−
|x|
2 2
|x| 2 |x|
= 2
|y| − 2y · x + r2 2
r |x|
2 2 2
= |x| − 2x · y + |y| = |x − y| .

This proves the first claim. Next suppose |x| , |y| < r and suppose, contrary to what is
claimed, that
y |x| rx
− = 0.
r |x|
Then
2
y |x| = r2 x
2
and so |y| |x| = r2 |x| which implies

|y| |x| = r2

contrary to the assumption that |x| , |y| < r. ¥

Let  ¯ ¯−(n−2)
 ¯¯ y|x| − rx ¯¯ , r−(n−2) for x = 0 if n ≥ 3
x r¯ |x| ¯
ψ (y) ≡
 ln ¯¯ y|x| − rx ¯¯ , ln (r) if x = 0 if n = 2
r |x|
332 THE LAPLACE AND POISSON EQUATIONS

Note that ¯ ¯
¯ y |x| rx ¯¯
lim ¯¯ − = r.
x→0 r |x| ¯
Then ψ x (y) = rn (y − x) if |y| = r, and ∆ψ x = 0. This last claim is obviously true if
x 6= 0. If x = 0, then ψ 0 (y) equals a constant and so it is also obvious in this case that
∆ψ x = 0.
The following lemma is easy to obtain.

Lemma 12.2.5 Let

½ −(n−2)
|y − x| if n ≥ 3
f (y) = .
ln |y − x| if n = 2

Then ( −(n−2)(y−x)
|y−x|n if n ≥3
∇f (y) = y−x .
|y−x|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) = rn (y − x) − ψ x (y) and y ∈
∂B (0, r) , then for n ≥ 3,
 ³ ´
µ ¶−(n−2) y− |x| r2
∂v y  − (n − 2) (y − x) |x| 2x
= · n + (n − 2) ¯ ¯n 
∂n r |y − x| r ¯ r2 ¯
¯y− |x| 2 x¯
³ ´
¡ ¢ r2
− (n − 2) r2 − y · x
|x|2
2 (n − 2) r2 − |x| 2x · y
= n + ³ r ´n ¯ ¯n
r |y − x| |x| r ¯ r2 ¯
r ¯ y− |x| 2 x¯
³ 2 ´
¡ ¢ |x| 2
− (n − 2) r2 − y · x (n − 2) r2 r − x · y
= n + ¯ ¯n
r |y − x| r ¯ |x| r ¯
¯ r y − |x| x¯

which by Lemma 12.2.4 equals

³ ´
¡ ¢ |x|2 2
r2 r −x·y
2
− (n − 2) r − y · x (n − 2)
n + n
r |y − x| r |y − x|
2
− (n − 2) r2 (n − 2) |x|
= n + n
r |y − x| r |y − x|
2
(n − 2) |x| − r2
= n.
r |y − x|
In the case where n = 2, and |y| = r, then Lemma 12.2.4 implies
 ³ ´
µ ¶ y|x| − rx
∂v y  (y − x) |x| r |x| 
= · 2 − ¯ ¯2 
∂n r |y − x| r ¯ y|x| rx ¯
¯ r − |x| ¯
 ³ ´ 
y|x|2
y  (y − x) r2 − x
= · − 
r 2 2
|y − x| |y − x|
2
1 r2 − |x|
= .
r |y − x|2
12.2. POISSON’S PROBLEM 333

Referring to 12.10 and 12.11, we would hope a solution, u to Poisson’s problem

satisfies for n ≥ 3
Z
− (rn (y − x) − ψ x (y)) f (y) dy
U
Z Ã !
2
(n − 2) |x| − r2
= g (y) n dσ (y) + u (x) (n − 2) ω n .
∂U r |y − x|

Thus
1
u (x) = ·
ω n (n − 2)
"Z Z Ã ! #
2
x (n − 2) r2 − |x|
(ψ (y) − rn (y − x)) f (y) dy + g (y) n dσ (y) . (12.12)
U ∂U r |y − x|

In the case where n = 2,

Z Z Ã !
2
x 1 r2 − |x|
− (r2 (y − x) − ψ (y)) f (y) dx = g (y) dσ (y) − u (x) 2π
U ∂U r |y − x|2

and so in this case,

"Z Z Ã ! #
2
1 x 1 r2 − |x|
u (x) = (r2 (y − x) − ψ (y)) f (y) dx + g (y) dσ (y) .
2π U ∂U r |y − x|2
(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 (Rn ) 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
2 Z
r2 − |x| 1
u (x) = g (y) n dσ (y) .
ωn r ∂U |y − x|
¡ ¢
I have shown this in case u ∈ C 2 U which is more specific 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.

Z 2
r2 − |x|
1= n dσ (y) .
∂U rω n |y − x|

For n = 2,
Z 2
1 r2 − |x|
1= dσ (y) .
∂U 2πr |y − x|2
334 THE LAPLACE AND POISSON EQUATIONS

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,
Z 2
r2 − |x|
1 = u (x) = n dσ (y)
∂U rω n |y − x|
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,
Z 2
1 r2 − |x|
u (x) = g (y) n dσ (y) (12.14)
ω n r ∂U |y − x|
for every n ≥ 2. Here ω 2 ≡ 2π.
Proof: That ∆u = 0 in U follows from the observation that the difference 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,
2
r2 − |x|
n
|y − x|
taken with respect to xk are uniformly bounded and continuous. In fact, this is true
of all partial derivatives. Therefore you can take the differential operator inside the
integral and write
Z Z Ã !
2 2
1 r2 − |x| 1 r2 − |x|
∆x g (y) n dσ (y) = g (y) ∆x n dσ (y) = 0.
ω n r ∂U |y − x| ω n r ∂U |y − x|
It only remains to verify that it achieves the desired boundary condition. Let x0 ∈ ∂U.
From Lemma 12.2.6,
Z Ã !
2
1 r2 − |x|
|g (x0 ) − u (x)| ≤ |g (y) − g (x0 )| n dσ (y) (12.15)
ω n r ∂U |y − x|
Z Ã !
2
1 r2 − |x|
≤ |g (y) − g (x0 )| n dσ (y) +(12.16)
ω n r [|y−x0 |<δ] |y − x|
Z Ã !
2
1 r2 − |x|
|g (y) − g (x0 )| n dσ (y) (12.17)
ω n r [|y−x0 |≥δ] |y − x|
where δ is a positive number. Letting ε > 0 be given, choose δ small enough that if
|y − x0 | < δ, then |g (y) − g (x0 )| < 2ε . Then for such δ,
Z Ã !
2
1 r2 − |x|
|g (y) − g (x0 )| n dσ (y)
ω n r [|y−x0 |<δ] |y − x|
Z Ã !
2
1 ε r2 − |x|
≤ dσ (y)
ω n r [|y−x0 |<δ] 2 |y − x|n
Z Ã !
2
1 ε r2 − |x| ε
≤ dσ (y) = .
ω n r ∂U 2 |y − x|n 2
12.2. POISSON’S PROBLEM 335

Denoting by M the maximum value of g on ∂U, the integral in 12.17 is dominated by

Z Ã !
2
2M r2 − |x|
dσ (y)
ω n r [|y−x0 |≥δ] |y − x|n
Z Ã !
2
2M r2 − |x|
≤ dσ (y)
ω n r [|y−x0 |≥δ] [|y − x0 | − |x − x0 |]n
Z Ã !
2
2M r2 − |x|
≤ £ ¤n dσ (y)
ω n r [|y−x0 |≥δ] δ − 2δ
µ ¶n Z ³ ´
2M 2 2
≤ r2 − |x| dσ (y)
ωn r δ ∂U

when |x − x0 | is sufficiently small. Then taking |x − x0 | still smaller,

³ if necessary,
´ this
2 2
last expression is less than ε/2 because |x0 | = r and so limx→x0 r − |x| = 0. This
proves limx→x0 u (x) = g (x0 ) 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 limx→x0 u (x) = g (x0 ) uniformly for x0 ∈ ∂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 (x0 , r) and let g ∈ C (∂U ) . Then there exists a unique
2
solution u ∈ C (U ) ∩ C U to the problem
∆u = 0 in U, u = g on ∂U.
This solution is given by the formula,
Z 2
1 r2 − |x − x0 |
u (x) = g (y) n dσ (y) (12.18)
ω n r ∂U |y − x|
for every n ≥ 2. Here ω 2 = 2π.
This corollary implies the following.
Corollary 12.2.9 Let u be a harmonic function defined on an open set, U ⊆ Rn
and let B (x0 , r) ⊆ U. Then
Z
1
u (x0 ) = u (y) dσ
ω n rn−1 ∂B(x0 ,r)

The representation formula, 12.14 is called Poisson’s integral formula. I have now
shown it works better than you had a right to expect for the Laplace equation. What
happens when f 6= 0?

12.2.3 The Case Where f 6= 0, Poisson’s 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
 ¯ ¯−(n−2)
 ¯¯ y|x| − rx ¯¯ , r−(n−2) for x = 0 if n ≥ 3
x r¯ |x| ¯
ψ (y) ≡
 ln ¯¯ y|x| − rx ¯¯ , ln (r) if x = 0 if n = 2
r |x|

Recall that rn (y − x) = ψ x (y) whenever y ∈ ∂U .

336 THE LAPLACE AND POISSON EQUATIONS

¡ ¢
Lemma 12.2.10 Let f ∈ C U or in Lp (U ) for p > n/2 2 . Then for x ∈ U, and
x0 ∈ ∂U, Z
1
lim (ψ x (y) − rn (y − x)) f (y) dy = 0.
x→x0 ω n (n − 2) U

Proof: There are two parts to this lemma. First the following claim is shown in
which an integral is taken over B (x0 , δ). After this, the integral over U \ B (x0 , δ) will
be considered. First note that
lim ψ x (y) − rn (y − x) = 0
x→x0

Claim:
Z Z
lim ψ x (y) |f (y)| dy = 0, lim rn (y − x) |f (y)| dy = 0.
δ→0 B(x0 ,δ) δ→0 B(x0 ,δ)

Proof of the claim: Using polar coordinates,

Z
ψ x (y) |f (y)| dy
B(x0 ,δ)
Z µ ¶
(x0 + z) |x| rx
= rn − |f (x0 + z)| dz
B(0,δ) r |x|
Z δZ µ ¶
(x0 + ρw) |x| rx
= rn − |f (x0 + ρw)| ρn−1 dσdρ
0 S n−1 r |x|
Now from the formula for rn , there exists δ 0 > 0 such that for ρ ∈ [0, δ 0 ] ,
µ ¶
(x0 + ρw) |x| rx
rn − ρn−2
r |x|
is bounded. Therefore,
Z Z δ Z
ψ x (y) |f (y)| dy ≤ C |f (x0 + ρw)| ρdσdρ.
B(x0 ,δ) 0 S n−1

If f is continuous, this is dominated by an expression of the form

Z δZ
C0 ρdσdρ
0 S n−1

which converges to 0 as δ → 0.
1 1
If f ∈ Lp (U ) , then by Holder’s inequality, (Problem 3 on Page 250) for p + q = 1,
Z δ Z
|f (x0 + ρw)| ρdσdρ
0 S n−1
Z δ Z
= |f (x0 + ρw)| ρ2−n ρn−1 dσdρ
0 S n−1
ÃZ Z !1/p
δ
p n−1
≤ |f (x0 + ρw)| ρ dσdρ ·
0 S n−1
ÃZ Z !1/q
δ ¡ ¢
2−n q
ρ ρn−1 dσdρ
0 S n−1

≤ C ||f ||Lp (U ) .
2 This means f is measurable and |f |p has finite integral
12.2. POISSON’S PROBLEM 337

Similar reasoning shows that

Z
lim |rn (y − x)| |f (y)| dy = 0.
δ→0 B(x0 ,δ)

This proves the claim.

Let ε > 0 be given and choose δ > 0 such that r/2 > δ > 0 where r is the radius of
the ball U and small enough that
Z
|f (y)| |ψ x (y) − rn (y − x)| dy < ε.
B(x0 ,2δ)

Then consider x ∈ B (x0 , δ) so that for y ∈

/ B (x0 , 2δ) , |y − x| > δ and so for such y,

|ψ x (y) − rn (y − x)| ≤ Cδ −(n−2)

for some constant C. Thus the integrand in

¯Z ¯
¯ ¯
¯ f (y) (ψ x (y) − rn (y − x)) dy ¯
¯ ¯
ZU
≤ |f (y)| |(ψ x (y) − rn (y − x))| dy
U \B(x0 ,2δ)
Z
+ |f (y)| |ψ x (y) − rn (y − x)| dy
B(x0 ,2δ)
Z
≤ f (y) (ψ x (y) − rn (y − x)) dy + ε
U \B(x0 ,2δ)

Now apply the dominated convergence theorem in this last integral to conclude it con-
verges to 0 as x → x0 . This proves the lemma. ¥
The following lemma follows from this one and Theorem 12.2.7.
¡ ¢
Lemma 12.2.11 Let f ∈ C U or in Lp (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 x0 ∈ ∂U,
lim u (x) = g (x0 ) .
x→x0

Not surprisingly, you can relax the condition that g ∈ C (∂U ) but I won’t do so here.
The next question is about the partial differential equation satisfied 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 φ ∈ Cc∞ (U ) and let x ∈ U. Let Uε denote the open set which has B (y, ε)
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
ψ x (y) − rn (y − x) , it is easy to verify that ∆x G (y, x) = 0 and so by Fubini’s theorem,
Z ·Z ¸
1
(ψ x (y) − rn (y − x)) f (y) dy ∆x φ (x) dx
U ω n (n − 2) U

Z ·Z ¸
1
= lim (ψ x (y) − rn (y − x)) f (y) dy ∆x φ (x) dx
ε→0 U ω n (n − 2) U
Z εµZ ¶
1
= lim (ψ x (y) − rn (y − x)) ∆x φ (x) dx f (y) dy
ε→0 U Uε ω n (n − 2)
338 THE LAPLACE AND POISSON EQUATIONS

   
G(y,x)
Z Z z }| {
 1  x  
= lim  ψ (y) − rn (y − x) ∆x φ (x) dx f (y) dy
ε→0 U Uε ω n (n − 2)

Z " Z µ ¶ #
1 ∂φ ∂G
= lim f (y) − G −φ dσ (x) dy
ε→0 ω n (n − 2) U ∂B(y,ε) ∂nx ∂nx
Z Z
1 ∂G
= lim f (y) φ dσ (x) dy
ε→0 ω n (n − 2) U ∂B(y,ε) ∂nx
x
Now x →ψ (y) and its partial derivatives are continuous and so the above reduces to
Z Z
1 ∂rn
= lim f (y) φ (x − y) dσ (x) dy
ε→0 ω n (n − 2) U ∂B(y,ε) ∂nx
Z Z Z
1 1
= lim f (y) φ n−1 dσ (x) dy = f (y) φ (y) dy.
ε→0 ω n U ∂B(y,ε) ε U

Similar but easier reasoning shows that

Z Ã Z 2
!
1 r2 − |x|
g (y) n dσ (y) ∆x φ (x) dx = 0.
U ω n r ∂U |y − x|

Therefore, if n ≥ 3, and u is given by 12.12, then whenever φ ∈ Cc∞ (U ) ,

Z Z
u∆φdx = f φdx. (12.19)
U U

The same result holds for n = 2.

Definition 12.2.12 ∆u = f on U in the weak sense or in the sense of distri-
butions if for all φ ∈ Cc∞ (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 Lp (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 differential 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 (x0 , 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
2 Z
r2 − |x − x0 | g (y)
u (x) = n dσ (y) (12.20)
ωn r ∂B(x0 ,r) |y − x|
It was also noted that this formula implies the mean value property for harmonic func-
tions, Z
1
u (x0 ) = u (y) dσ (y) . (12.21)
ω n rn−1 ∂B(x0 ,r)
The mean value property can also be formulated in terms of an integral taken over
B (x0 , 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 (x0 , r) ⊆ V.
Then Z
1
u (x0 ) = u (y) dy
mn (B (x0 , r)) B(x0 ,r)

where here mn (B (x0 , 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 mn (B (x0 , r)) = ωnn rn ,
Z Z rZ
u (y) dy = u (x0 + y) ρn−1 dσ (y) dρ
B(x0 ,r) 0 S n−1
Z r Z
= u (x0 + y) dσ (y) dρ
0 ∂B(0,ρ)
Z r
ωn n
= u (x0 ) ω n ρn−1 dρ = u (x0 ) r
0 n
= u (x0 ) mn (B (x0 , r)) .
This proves the lemma. ¥
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 Harnack’s inequality.
Theorem 12.3.2 Let U be an open set and let u ∈ C 2 (U ) be a nonnegative
harmonic function. Also let U1 be a connected open set which is bounded and satisfies
U1 ⊆ U. Then there exists a constant, C, depending only on U1 such that
© ª © ª
max u (x) : x ∈ U1 ≤ C min u (x) : x ∈ U1

Proof: There is a positive distance between U1 and U C because of compactness

of U1 . Therefore there exists r > 0 such that whenever x ∈ U1 , B (x, 2r) ⊆ U. Then
consider x ∈ U1 and let |x − y| < r. Then from Lemma 12.3.1
Z
1
u (x) = u (z) dz
mn (B (x, 2r)) B(x,2r)
Z
1
= u (z) dz
2n mn (B (x, r)) B(x,2r)
Z
1 1
≥ n
u (z) dz = n u (y) .
2 mn (B (y, r)) B(y,r) 2

The fact that u ≥ 0 is used in going to the last line. Since U1 is compact, there exist
m
finitely many balls having centers in U1 , {B (xi , r)}i=1 such that
U1 ⊆ ∪m
i=1 B (xi , 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 U1 would not be connected. Letting x, y ∈ U1 ,
there must be a sequence of these balls, B1 , B2 , · · · , Bk such that x ∈ B1 , y ∈ Bk , and
Bi ∩ Bi+1 6= ∅ for i = 1, 2, · · · , k − 1. Therefore, picking a point, zi+1 ∈ Bi ∩ Bi+1 , the
above estimate implies
1 1 1 1
u (x) ≥ n
u (z2 ) , u (z2 ) ≥ n u (z3 ) , u (z3 ) ≥ n u (z4 ) , · · · , u (zk ) ≥ n u (y) .
2 2 2 2
Therefore,
µ ¶k µ ¶m
1 1
u (x) ≥ u (y) ≥ u (y) .
2n 2n
340 THE LAPLACE AND POISSON EQUATIONS

Therefore, for all x ∈ U1 ,

m
sup {u (y) : y ∈ U1 } ≤ (2n ) u (x)

and so © ª
max u (x) : x ∈ U1 = sup {u (y) : y ∈ U1 }
m m © ª
≤ (2n ) inf {u (x) : x ∈ U1 } = (2n ) min u (x) : x ∈ U1 .
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 (x0 ,r) ⊆ U. I will show that u ∈ C ∞ (B (x0 ,r)) . From 12.20, it follows
that for x ∈ B (x0 ,r) ,
2 Z
r2 − |x − x0 | u (y)
n dσ (y) = u (x) .
ωn r ∂B(x0 ,r) |y − x|
2
−|x−x0 |2
It is obvious that x → r ωn r is infinitely differentiable. Therefore, consider
Z
u (y)
x→ n dσ (y) . (12.22)
∂B(x0 ,r) |y − x|

Take x ∈ B (x0 , r) and consider a difference quotient for t 6= 0.

ÃZ µ ¶ !
1 1 1
u (y) − dσ (y)
∂B(x0 ,r) t |y− (x + tek )|n |y − x|
n

Then by the mean value theorem, the term

µ ¶
1 1 1
−
t |y− (x + tek )|n |y − x|
n

equals
−(n+2)
−n |x+tθ (t) ek − y| (xk + θ (t) t − yk )
and as t → 0, this converges uniformly for y ∈ ∂B (x0 , r) to
−(n+2)
−n |x − y| (xk − yk ) .

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 xk equals
Z
−n (xk − yk ) u (y)
n+2 dσ (y) .
∂B(x0 ,r) |y − x|
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 (x0 , r) and x is a
given point not on this boundary. This proves the theorem. ¥
Liouville’s 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. LAPLACE’S EQUATION FOR GENERAL SETS 341

Theorem 12.3.4 (Liouville’s theorem) Suppose u is harmonic on Rn and is

bounded. Then u is constant.

Proof: From the Poisson formula

2 Z
r2 − |x| u (y)
n dσ (y) = u (x) .
ωn r ∂B(0,r) |y − x|

Now from the discussion above,

Z 2 Z
∂u (x) −2xk u (y) r2 − |x| u (y) (yk − xk )
= n dσ (y) + n+2 dσ (y)
∂xk ωn r ∂B(x0 ,r) |y − x| ωn r ∂B(0,r) |y − x|

Therefore, letting |u (y)| ≤ M for all y ∈ Rn ,

³ ´
¯ ¯ Z MZ
r2 − |x|
2
¯ ∂u (x) ¯ 2 |x| M 1
¯ ¯ ≤ n dσ (y) +
¯ ∂xk ¯ ωn r ωn r n+1 dσ (y)
∂B(x0 ,r) (r− |x|) ∂B(0,r) (r− |x|)
³ ´
2
2 |x| M r2 − |x| M 1
n−1 n−1
= n ωn r + n+1 ω n r
ω n r (r− |x|) ωn r (r− |x|)

and these terms converge to 0 as r → ∞. Since the inequality holds for all r > |x| , it
follows ∂u(x)
∂xk = 0. Similarly all the other partial derivatives equal zero as well and so u
is a constant. This proves the theorem. ¥

12.4 Laplace’s 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 Perron’s method for this problem. This method is based on exploit-
ing properties of subharmonic functions which are functions satisfying the following
definition.

Definition 12.4.1 Let U be an open set and let u be a function defined on U.

Then u is subharmonic if it is continuous and for all x ∈ U,
Z
1
u (x) ≤ ω rn−1 u (y) dσ (12.23)
n
∂B(x,r)

whenever r is small enough.

Compare with Corollary 12.2.9.

12.4.1 Properties Of Subharmonic Functions

The first 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
© ª
max u (y) : y ∈ U = max {u (y) : y ∈ ∂U } .
342 THE LAPLACE AND POISSON EQUATIONS

© ª
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 r0 > 0 such that u (y) = M for all y ∈ B (x, r0 ) 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 finite list of subharmonic func-
tions is also subharmonic.
Lemma 12.4.4 Let U be an open set and let u1 , u2 , · · · , up be subharmonic functions
defined on U. Then letting
v ≡ max (u1 , u2 , · · · , up ) ,
it follows that v is also subharmonic.
Proof: Let x ∈ U. Then whenever r is small enough to satisfy the subharmonicity
condition for each ui .
v (x) = max (u1 (x) , u2 (x) , · · · , up (x))
à Z Z !
1 1
≤ max u1 (y) dσ (y) , · · · , up (y) dσ (y)
ω n rn−1 ∂B(x,r) ω n rn−1 ∂B(x,r)
Z Z
1 1
≤ max (u 1 , u 2 , · · · , up ) (y) dσ (y) = v (y) dσ (y) .
ω n rn−1 ∂B(x,r) ω n rn−1 ∂B(x,r)
This proves the lemma. ¥
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 (x0 , r) and let g ∈ C (∂U ) . Then there exists a unique
2
solution u ∈ C (U ) ∩ C U to the problem
∆u = 0 in U, u = g on ∂U.
This solution is given by the formula,
Z 2
1 r2 − |x − x0 |
u (x) = g (y) n dσ (y) (12.24)
ω n r ∂U |y − x|
for every n ≥ 2. Here ω 2 = 2π.
12.4. LAPLACE’S EQUATION FOR GENERAL SETS 343

Definition 12.4.6 Let U be an open set and let u be subharmonic on U. Then

for B (x0 ,r) ⊆ U define
(
u (x) if x ∈/ B (x0 , r)
ux0 ,r (x) ≡ 1
R r 2 −|x−x0 |2
ω n r ∂B(x0 ,r) u (y) |y−x|n dσ (y) if x ∈ B (x0 , r)

Thus ux0 ,r is harmonic on B (x0 , r) , and equals to u off B (x0 , r) . The wonderful
thing about this is that ux0 ,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 (x0 ,r) ⊆ U as in the above definition.
Then ux0 ,r is subharmonic on U and u ≤ ux0 ,r .

Proof: First I show that u ≤ ux0 ,r . This follows from the maximum principle. Here
is why. The function u−ux0 ,r is subharmonic on B (x0 , r) and equals zero on ∂B (x0 , r) .
Here is why: For z ∈ B (x0 , r) ,
Z
1
u (z) − ux0 r (z) = u (z) − ux ,r (y) dσ (y)
ωρn−1 ∂B(z,ρ) 0

for all ρ small enough. This is by the mean value property of harmonic functions and
the observation that ux0 r is harmonic on B (x0 , r) . Therefore, from the fact that u is
subharmonic,
Z
1
u (z) − ux0 r (z) ≤ (u (y) − ux0 ,r (y)) dσ (y)
ωρn−1 ∂B(z,ρ)

Therefore, for all x ∈ B (x0 , r) ,

u (x) − ux0 ,r (x) ≤ 0.

The two functions are equal off B (x0 , r) .

The condition for being subharmonic is clearly satisfied at every point, x ∈ / B (x0 ,r).
It is also satisfied at every point of B (x0 ,r) thanks to the mean value property, Corollary
12.2.9 on Page 335. It is only at the points of ∂B (x0 ,r) where the condition needs to
be checked. Let z ∈ ∂B (x0 ,r) . Then since u is given to be subharmonic, it follows that
for all r small enough,
Z
1
ux0 ,r (z) = u (z) ≤ u (y) dσ
ω n rn−1 ∂B(x0 ,r)
Z
1
≤ ux ,r (y) dσ.
ω n rn−1 ∂B(x0 ,r) 0

This proves the lemma. ¥

Definition 12.4.8 For U a bounded open set and g ∈ C (∂U ), define

wg (x) ≡ sup {u (x) : u ∈ Sg }

where Sg 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 Sg 6= ∅ because u (x) ≡ m is a member of Sg . Also all functions in Sg have

values between m and max {g (y) : y ∈ ∂U }. The fundamental result is the following
absolutely amazing incredible result.
344 THE LAPLACE AND POISSON EQUATIONS

Proposition 12.4.9 Let U be a bounded open set and let g ∈ C (∂U ). Then wg ∈ Sg
and in addition to this, wg is harmonic.
∞
Proof: Let B (x0 , 2r) ⊆ U and let {xk }k=1 denote a countable dense subset of
B (x0 , r). Let {u1k } denote a sequence of functions of Sg with the property that

lim u1k (x1 ) = wg (x1 ) .

k→∞

By Lemma 12.4.7, it can be assumed each u1k is a harmonic function in B (x0 , 2r) since
otherwise, you could use the process of replacing u with ux0 ,2r . Similarly, for each l,
there exists a sequence of harmonic functions in Sg , {ulk } with the property that

lim ulk (xl ) = wg (xl ) .

k→∞

Now define
wk = (max (u1k , · · · , ukk ))x0 ,2r .
Then each wk ∈ Sg , each wk is harmonic in B (x0 , 2r), and for each xl ,

lim wk (xl ) = wg (xl ) .

k→∞

For x ∈ B (x0 , r)
Z 2
1 r2 − |x − x0 |
wk (x) = wk (y) n dσ (y) (12.25)
ω n 2r ∂B(x0 ,2r) |y − x|

and so there exists a constant, C which is independent of k such that for all i =
1, 2, · · · , n and x ∈ B (x0 , r), ¯ ¯
¯ ∂wk (x) ¯
¯ ¯
¯ ∂xi ¯ ≤ C

Therefore, this set of functions, {wk } is equicontinuous on B (x0 , r) as well as being

uniformly bounded and so by the Ascoli Arzela theorem, it has a subsequence which
converges uniformly on B (x0 , r) to a continuous function I will denote by w which has
the property that for all k,
w (xk ) = wg (xk ) (12.26)
Also since each wk is harmonic,
Z 2
1 r2 − |x − x0 |
wk (x) = wk (y) n dσ (y) (12.27)
ωn r ∂B(x0 ,r) |y − x|

Passing to the limit in 12.27 using the uniform convergence, it follows

Z 2
1 r2 − |x − x0 |
w (x) = w (y) n dσ (y) (12.28)
ωn r ∂B(x0 ,r) |y − x|

which shows that w is also harmonic. I have shown that w = wg on a dense set. Also, it
follows that w (x) ≤ wg (x) for all x ∈ B (x0 , r). It remains to verify these two functions
are in fact equal.
Claim: wg is lower semicontinuous on U.
Proof of claim: Suppose zk → z. I need to verify that

lim inf wg (zk ) ≥ wg (z) .

k→∞
12.4. LAPLACE’S EQUATION FOR GENERAL SETS 345

Let ε > 0 be given and pick u ∈ Sg such that wg (z) − ε < u (z) . Then

wg (z) − ε < u (z) = lim inf u (zk ) ≤ lim inf wg (zk ) .

k→∞ k→∞

Since ε is arbitrary, this proves the claim.

Using the claim, let x ∈ B (x0 , r) and pick xkl → x where {xkl } is a subsequence of
the dense set, {xk } . Then

wg (x) ≥ w (x) = lim inf w (xkl ) = lim inf wg (xkl ) ≥ wg (x) .

l→∞ l→∞

This proves w = wg and since w is harmonic, so is wg . 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.

Definition 12.4.10 A bounded open set, U has the barrier condition at z ∈ ∂U,
if there exists a function, bz called a barrier function which has the property that bz is
subharmonic on U, bz (z) = 0, and for all x ∈ ∂U \ {z} , bz (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, wg , defined above is in C 2 (U ) and
satisfies
∆wg = 0 in U,
lim wg (x) = g (z) .
x→z

Proof: From Proposition 12.4.9 it follows ∆wg = 0. Let z ∈ ∂U and let bz be the
barrier function at z. Then letting ε > 0 be given, the function

u− (x) ≡ max (g (z) − ε + Kbz (x) , m)

is subharmonic for all K > 0.

Claim: For K large enough, g (z) − ε + Kbz (x) ≤ g (x) for all x ∈ ∂U.
Proof of claim: Let δ > 0 and let Bδ = max {bz (x) : x ∈ ∂U \ B (z, δ)} . Then
Bδ < 0 by assumption and the compactness of ∂U \ B (z, δ) . Choose δ > 0 small
enough that if |x − z| < δ, then g (x) − g (z) + ε > 0. Then for |x − z| < δ,

g (x) − g (z) + ε
bz (x) ≤
K

for any choice of positive K. Now choose K large enough that Bδ < g(x)−g(z)+ε
K for all
x ∈ ∂U. This can be done because Bδ < 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− ∈ Sg which implies u− ≤ wg and so

g (z) − ε + Kbz (x) ≤ wg (x) . (12.29)

This is a very nice inequality and I would like to say

lim g (z) − ε + Kbz (x) = g (z) − ε

x→z
≤ lim inf wg (x)
x→z
≤ lim sup wg (x) = wg (z) ≤ g (z)
x→z
346 THE LAPLACE AND POISSON EQUATIONS

but this would be wrong because I do not know that wg 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) + ε − Kbz (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) − ε + Kbz (x) ≤ −g (x)
on ∂U. Therefore, letting u ∈ Sg , u − u+ is a subharmonic function which satisfies 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 ∈ Sg , it follows
wg (x) ≤ u+ (x) = g (z) + ε − Kbz (x) .
It follows that
g (z) − ε + Kbz (x) ≤ wg (x) ≤ g (z) + ε − Kbz (x)
and so,
g (z) − ε ≤ lim inf wg (x) ≤ lim sup wg (x) ≤ g (z) + ε.
x→z x→z
Since ε is arbitrary, this shows
lim wg (x) = g (z) .
x→z

This proves the theorem. ¥

12.4.2 Poisson’s 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 Poisson’s problem. If there is a solution, then it is of the form
·Z Z ¸
−1 ∂G
u (x) = G (x, y) f (y) dy + g (y) (x, y) dσ (y) , if n(12.30)
≥ 3,
(n − 2) ω n U ∂U ∂ny
·Z Z ¸
1 ∂G
u (x) = g (y) (x, y) dσ + G (x, y) f (y) dx , if n = 2 (12.31)
2π ∂U ∂ny U

for¡ G¢(x, y) = rn (y − x) − ψ x (y) where ψ x is a function which satisfies ψ x ∈ C 2 (U ) ∩

C U
∆ψ x = 0, ψ x (y) = rn (x − y) for y ∈ ∂U.
Furthermore, if u is given by the above representations, then u is a weak solution to
Poisson’s problem.
Proof: Uniqueness follows from Corollary 12.2.2 on Page 328. If u1 and u2 both
solve the Poisson problem, then their difference, w satisfies
∆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 Poisson’s problem. ¥
The function, G in the above representation is called Green’s function. Much more
can be said about the Green’s function.
How can you recognize that a bounded open set, U has the barrier condition? One
way would be to check the following condition.
12.4. LAPLACE’S EQUATION FOR GENERAL SETS 347

Condition 12.4.13 For each z ∈ ∂U, there exists xz ∈

/ U such that |xz − z| < |xz −y|
for every y ∈ ∂U \ {z} .

Proposition 12.4.14 Suppose Condition 12.4.13 holds. Then U satisfies the bar-
rier condition.

Proof: For n ≥ 3, let bz (y) ≡ rn (y − xz ) − rn (z − xz ). Then bz (z) = 0 and if y ∈

∂U with y 6= z, then clearly bz (y) < 0. For n = 2, let bz (y) = − ln |y − xz |+ln |z − xz | .
This works out the same way. ¥
Here is a picture of a domain which satisfies the barrier condition.

In fact, you have to have a fairly pathological example in order to find 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 can’t satisfy the above condition.
Could this set have the barrier property?
348 THE LAPLACE AND POISSON EQUATIONS
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 1970’s. 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.

Definition 13.0.15 A grating G is a finite 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.

0 cell
R

6 6 2 cell

6 ª
2 cell 1 cell 2 cell
n
For k = 0, 1, 2, one speaks of k chains. For {aj }j=1 a set of k cells, the k chain is
denoted as a formal sum
C = a1 + a2 + · · · + an

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 defined 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 ∈ |aj | for some aj }

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 definition. 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 off. In fact all the edges which are shared between two
2 cells disappear. The following is what results.

Definition 13.0.16 Next the boundary operator is defined. 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 = a1 + a2 + · · · + an ,

∂C ≡ ∂a1 + ∂a2 + · · · + ∂an

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?
C1 consists of the 2 cells in the first picture above along with the 2 cell whose boundary
is the 1 cycle over on the right. C2 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 C2 ? No, you couldn’t 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 D1 , D2 such that
C = ∂D1 = ∂D2 .

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 {e1 , · · · , em } be the 1 cells of C with the property that |ej | ⊆ l.
Note that ej 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 ej , bj and add in ∂bj which
is a 1 cycle. Thus X
C+ ∂bj
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 D1 , D2 composed
of 2 cells of G \ {l} such that for i = 1, 2,
X
∂Di = C + ∂bj (13.1)
j

Since none of the 2 cells of Di have any edgesP

on l, one can add l back in and regard D1
and D2 as 2 chains in G. Therefore, adding j ∂bj to both sides of the above yields
 
X X
C = ∂Di + ∂bj = ∂ Di + bj  , i = 1, 2.
j j

and this shows there exist two 2 chains which have C as the boundary. If ∂Di0 = C,
then  
X X X
∂Di0 + ∂bj = ∂ Di0 + bj  = C + ∂bj
j j j
P
and by P induction, there are exactly two 2 chains which Di0 + j bj can equal. Thus
adding j bj there are exactly two 2 chains which Di0 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, 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 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 first 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 modification for AA
and BB.

AA BB AA AA

BB AB AA AB
352 THE JORDAN CURVE THEOREM

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. Define 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 definition. As to uniqueness, in moving from left to right,
you must assign adjacent 2 cells joined at a 1 cell of C to different 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.

Lemma 13.0.18 Let C be a 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 6= 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. Thus there is a continuous curve in |C| joining x to y.
¥
The next lemma gives conditions under which you can go around a couple of closed
sets. It is called Alexander’s lemma. The following picture is a rough illustration of the
situation.
 353

x
C1
C2

F1 F2

Lemma 13.0.19 Let F1 be compact and F2 closed. Suppose C1 , C2 are two 1 chains,
∂Ci = x + y where x, y ∈ / F1 ∪ F2 . Suppose C2 does not intersect F2 and C1 does not
intersect F1 . Also suppose the 1 cycle C1 + C2 bounds a 2 chain D for which |D| ∩
F1 ∩F2 = ∅. Then there exists a 1 chain C such that ∂C = x+y and |∂C|∩(F1 ∪ F2 ) = ∅.
In particular x, y cannot be in different components of the complement of F1 ∪ F2 .

Proof: Let a1 , a2 , · · · , am be the 2 cells of D which intersect F1 . Consider

X
C ≡ C2 + ∂ak .
k

This is a 1 chain and ∂C = x + y because ∂∂ak = 0. Also |ak | ∩ F2 = ∅. This is because

|ak | ∩ F1 6= ∅ and none of the 2 cells of D intersect both F1 and F2 . Therefore, C is a
1 chain which avoids intersecting F2 . Does it also avoid F1 ? Suppose to the contrary
that l is a one cell of C which does intersect F1 . If |l| ⊆ |C1 + C2 | , then it would be an
edge of some 2 cell of D and would have to be a 1 cell of C2 since it intersects F1 so it
would have been added twice, once from C2 and once from
X
∂ak
k

and therefore could not be a summand in C. It follows l must be an edge of some

ak ∈ D and it is not a 1 cell of C1 + C2 . Therefore, if b is the 2 cell adjacent to a, it
must follow b ∈ D since otherwise l would be a 1 cell of C1 + C2 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 F1 also. This proves
the lemma. ¥

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|. ¥

Definition 13.0.21 A Jordan arc is a set of points of the form Γ ≡ r ([a, b])
where r is a one to one map from [a, b] to the plane. For p, q ∈ Γ, say p < q if
p = r (t1 ) , q = r (t2 ) for t1 < t2 . Also let pq denote the arc r ([t1 , t2 ]).

Theorem 13.0.22 Let Γ be a Jordan arc. Then its complement is connected.

354 THE JORDAN CURVE THEOREM

Proof: Suppose this is not so. Then there exists x, y points in ΓC which are in
different components of ΓC . 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. 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| ∩ Γ 6= ∅ since otherwise these two points would
not be in different components. Suppose there exists C1 , ∂C1 = x + y and |C1 | ∩ zq = ∅
and C2 , ∂C2 = x + y but |C2 | ∩ pz = ∅. Then C1 + C2 is a 1 cycle and so by Lemma
13.0.17 there are exactly two 2 chains whose boundaries are C1 + C2 . Since z ∈ / |Ci | ,
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, Alexander’s 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 different components of ΓC
contrary to the assumption they are in different 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
µµ ¶ ¶
a+b 1
w=r +b
2 2
Thus, continuing this way, there is a sequence of Jordan arcs pk qk where r (tk ) = qk
and r (sk ) = pk with |tk − sk | < b−a
2k
, [sk , tk ] ⊆ [a, b] such that every C with ∂C = x + y
has nonempty intersection with pk qk . The intersection of these arcs is r (s) where s =
∩∞k=1 [sk , tk ]. Then all such C must go through r (s) because such C with ∂C = x + y
must intersect pk qk 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). 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 = x + y but
/ |C + ∂D| . Therefore, this contradiction shows ΓC must be connected after all.
r (s) ∈
¥
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 Γ = r ([a, b]) be a Jordan arc where r is as above, one to one,
onto and continuous. Then Γ has no interior points.
Proof : Suppose to the contrary that Γ has an interior point p. Then for some r > 0,
B (p, r) ⊆ Γ.
Consider the circles of radius δ < r centered at p. Denoting as Cδ one of these, it follows
the Cδ are disjoint. Therefore, since r is one to one, the sets r−1 (Cδ ) 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 (Cδ ) 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 (Cδ ) , {Iδ }δ<r .
This is a contradiction to the density of Q and the fact that Q is at most countable. ¥
Definition 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.
 355

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.

Lemma 13.0.25 J is a simple closed curve if and only if there exists a mapping
θ : S 1 → J where S 1 is the unit circle
© ª
(x, y) : x2 + y 2 = 1 ,

such that θ is one to one and continuous.

Proof : First suppose J is the image of the unit circle as just explained. Then let
p : [0, 2π] → S 1 be defined as p (t) ≡ (cos (t) , sin (t)) . Then consider r (t) ≡ θ (p (t)).
r is one to one on [0, 2π) and (0, 2π] with r (0) = r (2π) and is continuous, being the
composition of continuous functions.
Suppose now 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) . Define θ −1 : J → S 1 by
µ µ ¶ µ ¶¶
2π ¡ −1 ¢ 2π ¡ −1 ¢
θ −1 (x) ≡ cos r (x) − a , sin r (x) − a
b−a b−a

Note that θ −1 is onto S 1 . The function is well defined because it sends the point r (a) =
r (b) to the same point, (1, 0) . It is also one to one. To see this note r−1 is one to one
on J \ {r (a) , r (b)} . What about the case where x 6= r (a) = r (b)? Could θ −1 (x) =
θ −1 (r (a))? In this case, r−1 (x) is in (a, b) while r−1 (r (a)) = a so

θ −1 (x) 6= θ −1 (r (a)) = (1, 0) .

Thus θ −1 is one to one on J.

Why is θ −1 continuous? Suppose xn → r (a) = r (b) first. Why does θ −1 (xn ) →
(1, 0) = θ −1 (r (a))? Let {xn } denote any subsequence of the given sequence. Then by
compactness of [a, b] there exists a further subsequence, still denoted by xn such that

r−1 (xn ) → t ∈ [a, b]

Hence by continuity of r, xn → r (t) and so r (t) must equal r (a) = r (b) . It follows from
the assumption of what a simple curve is that t ∈ {a, b} . Hence θ −1 (xn ) converges to
either µ µ ¶ µ ¶¶
2π 2π
cos (a − a) , sin (a − a)
b−a b−a
or µ µ ¶ µ ¶¶
2π 2π
cos (b − a) , sin (b − a)
b−a b−a
but these are the same point. This has shown that if xn → r (a) = r (b) , there is a
subsequence such that θ −1 (xn ) → θ −1 (r (a)) . Thus θ −1 is continuous at r (a) = r (b).
Next suppose xn → x 6= r (a) ≡ p. Then there exists B (p, r) such that for all n
large enough, xn and x are contained in the compact set J \ B (p, r) ≡ K. Then r is
continuous and one to one on the compact set r−1 (K) ⊆ (a, b) and so by Theorem 5.1.3
r−1 is continuous on K. In particular it is continuous at x so θ −1 (xn ) → θ −1 (x). This
proves the lemma. ¥
Now with this preparation, here is the main result, the Jordan curve theorem.

Theorem 13.0.26 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.
356 THE JORDAN CURVE THEOREM

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 different component
of J C . Let J = H ∪ K where H and K are two Jordan arcs joined at the points a and
b.
a
H K

Let C1 be a square shaped curve (having vertical and horizontal sides) which joins
x and y which misses H. Then let C2 be such a square shaped curve which joins x and
y and misses K. This is possible by Theorem 13.0.22 which says the complements of
these Jordan arcs are connected and Problem 7 on Page 113 which says an open set is
connected if and only if every pair of points can be joined by a square curve. Similarly,
let C3 and C4 be square curves joining y and z such that C3 misses H and C4 misses
K. Let G be a grating which has each of the corners of all these square curves as
intersections of horizonal and vertical lines of G as well as each of x, y, z.
Thus ∂C1 = x + y = ∂C2 and ∂C3 = y + z = ∂C4 . Then C1 + C2 is a cycle and so
by Lemma 13.0.17 there are exactly two 2 chains D, E such that

∂E = ∂D = C1 + C2 .

I claim that neither |D| nor |E| can contain both a and b. Here is why. If they were
both in |D| for example, then E would be a 2 chain whose boundary is C1 + C2 which
misses {a, b} ≡ H ∩ K. (Note {a, b} = H ∩ K and so C1 + C2 does not contain either a
or b.) Therefore, by Lemma 13.0.19 there exists a 1 chain C such that ∂C = x + y which
misses both H and K which along with Lemma 13.0.18 then contradicts the assertion
that x, y are in different components. Similarly, not both a, b can be in |E| . Say D is
the two chain such that a ∈ |D| and b ∈ |E|. Similarly there are exactly two 2 chains
P, Q such that ∂P = ∂Q = C3 + C4 and by similar reasoning neither |P | nor |Q| can
contain both a and b. Let a ∈ |P | and b ∈ |Q|. Now (C1 + C3 ) + (C2 + C4 ) is a 1 cycle
because
∂ (C1 + C3 + C2 + C4 ) = x + y + y + z + x + y + y + z = 0

and ∂ (C1 + C3 ) = x + z while ∂ (C2 + C4 ) = x + z. Also |C1 + C3 | misses H while

|C2 + C4 | misses K. In particular, neither a nor b is contained in any of the |Ci |. Then
by Lemma 13.0.17 there are exactly two 2 chains of which C1 + C3 + C2 + C4 is the
boundary. However,
∂ (D + Q) = C3 + C4 + C1 + C2

and so D + Q is one of these two chains. Therefore, |C1 + C3 | misses H and |C2 + C4 |
misses K and both a and b are in |D + Q| . It follows the other two chain of which
C3 + C4 + C1 + C2 is the boundary has empty intersection with {a, b} = H ∩ K and by
Alexander’s lemma there exists a 1 chain C such that x + z = ∂C and so x, z are not in
different components after all.Thus there are at most 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.
 357

a
Q

H K

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 = x1 + · · · + xm .
Then it follows each xk ∈ / H. Could ∂P = 0? The answer is no. If every 1 cell of Q had
both ends in H, then consider one of them which intersects K. This must take place
for some 1 cell of Q because K is connected. Now this violates the assumption that no
2 cell of G can intersect both of those disjoint compact sets, H ∩ |Q| and |Q| ∩ K.
Since it is assumed there is only one component of J C , it follows upon refining G
if necessary, there exist 1 chains Bk contained in J C such that ∂Bk = xk + xk+1 and
it is the existence of these Bk which will give the desired contradiction. Let
B = B1 + · · · + Bm . Then P + B is a 1 cycle which misses K. The reason for this is
that 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. However,
∂ (D + E) = P + B + P + Q + B = Q
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
JC.
In the above argument, if each pair {xi , xi+1 } can be joined by a square curve Bi
which lies in J C , then the contradiction was obtained. Therefore, there must exist a
pair {xi , xi+1 } which can’t be joined by any square curve in J C and this requires these
points to be in different components by Problem 7 on Page 113. 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 13.0.25, 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
358 THE JORDAN CURVE THEOREM

C
B (0, R) where J ⊆ B (0, R). Thus there are two components for J C , the unbounded
C
one which contains B (0, R) and the bounded one which must be contained in B (0, R) .
This proves the theorem. ¥
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.

Definition 14.1.1 Let γ : [a, b] → Rn be a function. Then γ is of bounded

variation if
( n )
X
sup |γ (ti ) − γ (ti−1 )| : a = t0 < · · · < tn = b ≡ V (γ, [a, b]) < ∞
i=1

where the sums are taken over all possible lists, {a = t0 < · · · < tn = b} . The set of
points traced out will be denoted by γ ∗ ≡ γ ([a, b]). The function γ is called a param-
eterization of γ ∗ . The set of points γ ∗ is called a rectifiable curve. If a set of points
γ ∗ = γ ([a, b]) where γ is continuous and γ is one to one on [a, b) and also one to one
on (a, b], then γ ∗ is called a simple curve. A closed curve is one which has a parame-
terization γ defined on an interval [a, b] such that γ (a) = γ (b). It is a simple closed
curve if there is a parameterization g such that γ is one to one on [a, b) and one to one
on (a, b] with γ (a) = γ (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 γ ∗ possesses independent of the parameterization
γ used to describe the set of points γ ∗ . To show this, it is helpful to use the following
lemma.

Lemma 14.1.2 Let φ : [a, b] → R be a continuous function and suppose φ is 1−1 on

(a, b). Then φ is either strictly increasing or strictly decreasing on [a, b] . Furthermore,
φ−1 is continuous.

Proof: First it is shown that φ is either strictly increasing or strictly decreasing on

(a, b) .
If φ is not strictly decreasing on (a, b), then there exists x1 < y1 , x1 , y1 ∈ (a, b) such
that
(φ (y1 ) − φ (x1 )) (y1 − x1 ) > 0.
If for some other pair of points, x2 < y2 with x2 , y2 ∈ (a, b) , the above inequality does
not hold, then since φ is 1 − 1,

(φ (y2 ) − φ (x2 )) (y2 − x2 ) < 0.

359
360 LINE INTEGRALS

Let xt ≡ tx1 + (1 − t) x2 and yt ≡ ty1 + (1 − t) y2 . Then xt < yt for all t ∈ [0, 1] because

tx1 ≤ ty1 and (1 − t) x2 ≤ (1 − t) y2

with strict inequality holding for at least one of these inequalities since not both t and
(1 − t) can equal zero. Now define

h (t) ≡ (φ (yt ) − φ (xt )) (yt − xt ) .

Since h is continuous and h (0) < 0, while h (1) > 0, there exists t ∈ (0, 1) such that
h (t) = 0. Therefore, both xt and yt are points of (a, b) and φ (yt ) − φ (xt ) = 0 contra-
dicting the assumption that φ is one to one. It follows φ 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 φ. Suppose φ is strictly increasing on (a, b) , a similar
argument holding for φ strictly decreasing on (a, b) . If x > a, then pick y ∈ (a, x) and
from the above, φ (y) < φ (x) . Now by continuity of φ at a,

φ (a) = lim φ (z) ≤ φ (y) < φ (x) .

x→a+

Therefore, φ (a) < φ (x) whenever x ∈ (a, b) . Similarly φ (b) > φ (x) for all x ∈ (a, b).
It only remains to verify φ−1 is continuous. Suppose then that sn → s where sn
and s are points of φ ([a, b]) . It is desired to verify that φ−1 (sn ) → φ−1 (s) . If this
¯does not happen, there¯ exists ε > 0 and a subsequence, still denoted by sn such that
¯φ−1 (sn ) − φ−1 (s)¯ ≥ ε. Using the sequential compactness of [a, b] there exists a further
subsequence, still denoted by n, such that φ−1 (sn ) → t1 ∈ [a, b] , t1 6= φ−1 (s) . Then by
continuity of φ, it follows sn → φ (t1 ) and so s = φ (t1 ) . Therefore, t1 = φ−1 (s) after
all. This proves the lemma. ¥
Now suppose γ and η are two parameterizations of the simple curve γ ∗ as described
above. Thus γ ([a, b]) = γ ∗ = η ([c, d]) and the two continuous functions γ, η are one
to one on their respective open intervals. I need to show the two definitions of length
yield the same thing with either parameterization. Since γ ∗ is compact, it follows
from Theorem 5.1.3 on Page 88, both γ −1 and η −1 are continuous. Thus γ −1 ◦ η :
[c, d] → [a, b] is continuous. It is also uniformly continuous because [c, d] is compact. Let
P ≡ {t0 , · · ·, tn } be a partition of [a, b] , t0 < t1 < · · · < tn such that for L < V (γ, [a, b]) ,
n
X
L< |γ (tk ) − γ (tk−1 )| ≤ V (γ, [a, b])
k=1

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 γ −1 ◦ η (sk ) = tk so that {s0 , · · ·, sn } is a partition of [c, d] . By the lemma, the
sk are either strictly decreasing or strictly increasing as a function of k, depending on
whether γ −1 ◦ η is increasing or decreasing. Thus γ (tk ) = η (sk ) and so
n
X
L< |η (sk ) − η (sk−1 )| ≤ V (γ, [a, b])
k=1

It follows that whenever L < V (γ, [a, b]) , there exists a partition of [c, d] , {s0 , · · ·, sn }
such that
X n
L< |η (sk ) − η (sk−1 )|
k=1
14.1. BASIC PROPERTIES 361

It follows that for every L < V (γ, [a, b]) , V (η, [c, d]) ≥ L which requires V (η, [c, d]) ≥
V (γ, [a, b]). Turning the argument around, it follows
V (η, [c, d]) = V (γ, [a, b]) .
This proves the following fundamental theorem.
Theorem 14.1.3 Let Γ be a simple curve and let γ be a parameterization for
Γ where γ is one to one on (a, b), continuous on [a, b] and of bounded variation. Then
the total variation
V (γ, [a, b])
can be used as a definition for the length of Γ in the sense that if Γ = η ([c, d]) where
η is a continuous function which is one to one on (c, d) with η ([c, d]) = Γ,
V (γ, [a, b]) = V (η, [c, d]) .
This common value can be denoted by V (Γ) and is called the length of Γ.
The length is not dependent on parameterization. Simple curves which have such
parameterizations are called rectifiable.

14.1.2 Orientation
There is another notion called orientation. For simple rectifiable 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.
Definition 14.1.4 Let η, γ be continuous one to one parameterizations for a
simple rectifiable curve. If η −1 ◦ γ is increasing, then γ and η are said to be equivalent
parameterizations and this is written as γ ∼ η. It is also said that the two parameteri-
zations give the same orientation for the curve when γ ∼ η.
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 η −1 ◦ γ 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 ∼.
γ ∼ γ, (14.1)
If γ ∼ η then η ∼ γ, (14.2)
If γ ∼ η and η ∼ θ, then γ ∼ θ. (14.3)
Proof: Formula 14.1 is obvious because γ −1 ◦ γ (t) = t so it is clearly an increasing
function. If γ ∼ η then γ −1 ◦η is increasing. Now η −1 ◦γ must also be ¡ increasing
¢ ¡ because¢
it is the inverse of γ −1 ◦η. This verifies 14.2. To see 14.3, γ −1 ◦θ = γ −1 ◦ η ◦ η −1 ◦ θ
and so since both of these functions are increasing, it follows γ −1 ◦ θ is also increasing.
This proves the lemma. ¥
Definition 14.1.6 Let Γ be a simple rectifiable curve and let γ be a parame-
terization for Γ. Denoting by [γ] the equivalence class of parameterizations determined
by the above equivalence relation, the following pair will be called an oriented curve.
(Γ, [γ])
In simple language, an oriented curve is one which has a direction of motion specified.
362 LINE INTEGRALS

Actually, people usually just write Γ and there is understood a direction of motion
or orientation on Γ. How can you identify which orientation is being considered?

Proposition 14.1.7 Let (Γ, [γ]) be an oriented simple curve and let p, q be any
two distinct points of Γ. Then [γ] is determined by the order of γ −1 (p) and γ −1 (q).
This means that η ∈ [γ] if and only if η −1 (p) and η −1 (q) occur in the same order as
γ −1 (p) and γ −1 (q).

Proof: Suppose γ −1 (p) < γ −1 (q) and let η ∈ [γ] . Is it true that η −1 (p) <
η (q)? Of course it is because γ −1 ◦ η is increasing. Therefore, if η −1 (p) > η −1 (q)
−1

it would follow
¡ ¢ ¡ ¢
γ −1 (p) = γ −1 ◦ η η −1 (p) > γ −1 ◦ η η −1 (q) = γ −1 (q)

which is a contradiction. Thus if γ −1 (p) < γ −1 (q) for one γ ∈ [γ] , then this is true
for all η ∈ [γ].
Now suppose η is a parameterization for Γ defined on [c, d] which has the property
that
η −1 (p) < η −1 (q)

Does it follow η ∈ [γ]? Is γ −1 ◦ η increasing? By Lemma 14.1.2 it is either increasing

or decreasing. Thus it suffices to test it on two points of [c, d] . Pick the two points
η −1 (p) , η −1 (q) . Is
¡ ¢ ¡ ¢
γ −1 ◦ η η −1 (p) < γ −1 ◦ η η −1 (q) ?

Yes because these reduce to γ −1 (p) on the left and γ −1 (q) on the right. It is given
that γ −1 (p) < γ −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 first 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 rectifiable curves.
The situation is illustrated by the following picture.

γ 1 (α) = γ 2 (β)

Γ2 l Γ1

γ 1 (δ) = γ 2 (θ)

Proposition 14.1.8 Let Γ1 and Γ2 be two simple closed rectifiable 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 Γ1 is opposite its orientation when consid-
ered a part of Γ2 . Then if the open segment (l except for its endpoints) of l is removed,
the result is a simple closed rectifiable curve Γ. This curve has a parameterization γ
with the property that on γ −1
j (Γ ∩ Γj ) , γ
−1
γ j is increasing. In other words, Γ has an
orientation consistent with that of Γ1 and Γ2 . Furthermore, if Γ has such a consistent
orientation, then the orientations of l as part of the two simple closed curves, Γ1 and
Γ2 are opposite.
14.1. BASIC PROPERTIES 363

γ 1 (α) = γ 2 (β)

Γ2 Γ1

γ 1 (δ) = γ 2 (θ)

Proof: Let Γ1 = γ 1 ([a, b]) , γ 1 (a) = γ 1 (b) , and Γ2 = γ 2 ([c, d]) , γ 2 (c) = γ 2 (d) ,
with l = γ 1 ([α, δ]) = γ 2 ([θ, β]). (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 α, δ, θ, β. Sup-
pose without loss of generality that α < δ. Then because of this assumption it follows
γ 2 (θ) = γ 1 (δ) , γ 2 (β) = γ 1 (α). The following diagram might be useful to summarize
what was just said.
a α δ b γ
1

c θ β d γ
2

Note the first of the interval [β, d] matches the last of the interval [a, α] and the first
of [δ, b] matches the last of [c, θ] , all this in terms of where these points are sent.
Now I need to describe the parameterization of Γ ≡ Γ1 ∪ Γ2 . To verify it is a
simple closed curve, I must produce an interval and a mapping from this interval to Γ
which satisfies the conditions needed for γ to be a simple closed rectifiable curve. The
following is the definition as well as a description of which part of Γj 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 off
followed by its length and finally where it ends up.
γ 1 (a), α − a, γ 1 (α) γ 2 (β), d − β, γ 2 (d) γ 2 (c), θ − c, γ 2 (θ) γ 1 (δ), b − δ, γ 1 (b)

Note it ends up where it started, at γ 1 (a) = γ 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 γ which corresponds to the above picture.
Then γ (t) is given by
γ (t) ≡


 γ 1 (t) , t ∈ [a, α] , γ 1 (a) → γ 1 (α) = γ 2 (β)



 γ (t + β − α) , t ∈ [α, α + d − β] , γ 2 (β) → γ 2 (d) = γ 2 (c)
 2
γ 2 (t + c − α − d + β) , t ∈ [α + d − β, α + d − β + θ − c] ,

 γ 2 (c) = γ 2 (d) → γ 2 (θ) = γ 1 (δ)



 γ (t − α − d + β − θ + c + δ) , t ∈ [α + d − β + θ − c, α + d − β + θ − c + b − δ] ,
 1
γ 1 (δ) → γ 1 (b) = γ 1 (a)
The construction shows γ is one to one on
(a, α + d − β + θ − c + b − δ)
and if t is in this open interval, then
γ (t) 6= γ (a) = γ 1 (a)
364 LINE INTEGRALS

and
γ (t) 6= γ (α + d − β + θ − c + b − δ) = γ 1 (b) .
Also
γ (a) = γ 1 (a) = γ (α + d − β + θ − c + b − δ) = γ 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 Γj . Then from a repeat of the above argument,
you could change the orientation of l relative to Γ2 and obtain an orientation of Γ
which is consistent with that of Γ1 and Γ2 . Call a parameterization which has this
new orientation γ n while γ is the one which is assumed to exist. This new orientation
of l changes the orientation of Γ2 because there are two points in l. Therefore on
γ 2−1 (Γ ∩ Γ2 ), γ −1
n γ 2 is decreasing while γ
−1
γ 2 is assumed to be increasing. Hence γ
and γ n are not equivalent. However, the above construction would leave the orientation
of both γ 1 ([a, α]) and γ 1 ([δ, b]) unchanged and at least one of these must have at least
two points. Thus the orientation of Γ must be the same for γ n as for γ. That is, γ ∼ γ n .
This is a contradiction. This proves the proposition. ¥
There is a slightly different aspect of the above proposition which is interesting. It
involves using the shared segment to orient the simple closed curve Γ.
Corollary 14.1.9 Let the intersection of simple closed rectifiable curves, Γ1 and Γ2
consist of the simple curve l. Then place opposite orientations on l, and use these two
different orientations to specify orientations of Γ1 and Γ2 . Then letting Γ denote the
simple closed curve which is obtained from deleting the open segment of l, there exists
an orientation for Γ which is consistent with the orientations of Γ1 and Γ2 obtained
from the given specification 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 γ is one to one on the open interval. The
line integral and its properties are presented next.
Definition 14.2.1 Let γ : [a, b] → Rn be of bounded variation and let f : γ ∗ →
n
R . Letting P ≡ {t0 , · · ·, tn } where a = t0 < t1 < · · · < tn = b, define
||P|| ≡ max {|tj − tj−1 | : j = 1, · · ·, n}
and the Riemann Stieltjes sum by
n
X
S (P) ≡ f (γ (τ j )) · (γ (tj ) − γ (tj−1 ))
j=1

where τ j ∈ [tj−1 , tj ] . (Note this notation is a little sloppy because it does not identify
R
the specific point, τ j used. It is understood that this point is arbitrary.) Define γ f ·dγ
as the unique number which satisfies the following condition. For all ε > 0 there exists
a δ > 0 such that if ||P|| ≤ δ, then
¯Z ¯
¯ ¯
¯ f ·dγ − S (P)¯ < ε.
¯ ¯
γ

Sometimes this is written as

Z
f ·dγ ≡ lim S (P) .
γ ||P||→0
14.2. THE LINE INTEGRAL 365

Then γ ∗ is a set of points in Rn and as t moves from a to b, γ (t) moves from γ (a)
to γ (b) . Thus γ ∗ has a first point and a last point. (In the case of a closed curve these
are the same point.) If φ : [c, d] → [a, b] is a continuous nondecreasing function, then
γ ◦ φ : [c, d] → Rn is also of bounded variation and yields the same set of points in Rn
with the same first and last points.

Theorem 14.2.2 Let φ and γ be as just described. Then assuming that

Z
f ·dγ
γ

exists, so does Z
f ·d (γ ◦ φ)
γ◦φ

and Z Z
f ·dγ = f ·d (γ ◦ φ) . (14.4)
γ γ◦φ

Proof: There exists δ > 0 such that if P is a partition of [a, b] such that ||P|| < δ,
then ¯Z ¯
¯ ¯
¯ f ·dγ − S (P)¯ < ε.
¯ ¯
γ

By continuity of φ, there exists σ > 0 such that if Q is a partition of [c, d] with ||Q|| <
σ, Q = {s0 , · · ·, sn } , then |φ (sj ) − φ (sj−1 )| < δ. Thus letting P denote the points in
[a, b] given by φ (sj ) for sj ∈ Q, it follows that ||P|| < δ and so
¯ ¯
¯Z n ¯
¯ X ¯
¯ f ·dγ − f (γ (φ (τ j ))) · (γ (φ (sj )) − γ (φ (sj−1 )))¯¯ < ε
¯
¯ γ j=1 ¯

where τ j ∈ [sj−1 , sj ] . Therefore, from the definition 14.4 holds and

Z
f ·d (γ ◦ φ)
γ◦φ

exists. This proves the theorem.

R ¥
This theorem shows that γ f ·dγ is independent of the particular parameterization γ
used in its computation to the extent that if φ is any nondecreasing continuous function
from another interval, [c, d] , mapping to [a, b] , then the same value is obtained by
replacing γ with γ ◦ φ. In other words, this line integral depends only on γ ∗ and the
order in which γ(t) encounters the points of γ ∗ as t moves from one end to the other of
the interval. For the case of an oriented rectifiable curve Γ this shows the line integral
is dependent only on the set of points and the orientation of Γ. ¥
The fundamental result in this subject is the following theorem.

Theorem 14.2.3 Let f : γ ∗ → Rn Rbe continuous and let γ : [a, b] → Rn be

continuous and of bounded variation. Then γ f ·dγ exists. Also letting δ m > 0 be such
1
that |t − s| < δ m implies |f (γ (t)) − f (γ (s))| < m ,
¯Z ¯
¯ ¯
¯ f dγ − S (P)¯ ≤ 2V (γ, [a, b])
¯ ¯ m
γ

whenever ||P|| < δ m .

366 LINE INTEGRALS

Proof: The function, f ◦ γ , is uniformly continuous because it is defined on a

compact set. Therefore, there exists a decreasing sequence of positive numbers, {δ m }
such that if |s − t| < δ m , then
1
|f (γ (t)) − f (γ (s))| < .
m
Let
Fm ≡ {S (P) : ||P|| < δ m }.
Thus Fm is a closed set. (The symbol, S (P) in the above definition, means to include
all sums corresponding to P for any choice of τ j .) It is shown that
2V (γ, [a, b])
diam (Fm ) ≤ (14.5)
m
and then it will follow there existsR a unique point, I ∈ ∩∞m=1 Fm . This is because R is
complete. It will then follow I = γ f (t) dγ (t) . To verify 14.5, it suffices to verify that
whenever P and Q are partitions satisfying ||P|| < δ m and ||Q|| < δ m ,
2
|S (P) − S (Q)| ≤ V (γ, [a, b]) . (14.6)
m
Suppose ||P|| < δ m and Q ⊇ P. Then also ||Q|| < δ m . To begin with, suppose that
P ≡ {t0 , · · ·, tp , · · ·, tn } and Q ≡ {t0 , · · ·, tp−1 , t∗ , tp , · · ·, tn } . Thus Q contains only one
more point than P. Letting S (Q) and S (P) be Riemann Stieltjes sums,
p−1
X
S (Q) ≡ f (γ (σ j )) · (γ (tj ) − γ (tj−1 )) + f (γ (σ ∗ )) (γ (t∗ ) − γ (tp−1 ))
j=1

n
X
∗ ∗
+f (γ (σ )) · (γ (tp ) − γ (t )) + f (γ (σ j )) · (γ (tj ) − γ (tj−1 )) ,
j=p+1
p−1
X
S (P) ≡ f (γ (τ j )) · (γ (tj ) − γ (tj−1 )) +
j=1

=f (γ(τ p ))·(γ(tp )−γ(tp−1 ))

z }| {
f (γ (τ p )) · (γ (t∗ ) − γ (tp−1 )) + f (γ (τ p )) · (γ (tp ) − γ (t∗ ))
n
X
+ f (γ (τ j )) · (γ (tj ) − γ (tj−1 )) .
j=p+1

Therefore,
p−1
X 1 1
|S (P) − S (Q)| ≤ |γ (tj ) − γ (tj−1 )| + |γ (t∗ ) − γ (tp−1 )| +
j=1
m m

Xn
1 1 1
|γ (tp ) − γ (t∗ )| + |γ (tj ) − γ (tj−1 )| ≤ V (γ, [a, b]) . (14.7)
m j=p+1
m m

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 δ m and let R ≡ P ∪ Q.
Then from what was just observed,
2
|S (P) − S (Q)| ≤ |S (P) − S (R)| + |S (R) − S (Q)| ≤ V (γ, [a, b]) .
m
14.2. THE LINE INTEGRAL 367

and this shows 14.6 which proves 14.5. Therefore,

R there exists a unique number, I ∈
∩∞m=1 Fm which satisfies the definition of γ f ·dγ. This proves the theorem. ¥
Note this is a general sort of result. It is not assumed that γ is one to one anywhere
in the proof. The following theorem follows easily from the above definitions and
theorem. This theorem is used to establish estimates.

Theorem 14.2.4 Let f be a continuous function defined on γ ∗ , denoted as f ∈

C (γ ) where γ : [a, b] → Rn is of bounded variation and continuous. Let
∗

M ≥ max {|f ◦ γ (t)| : t ∈ [a, b]} . (14.8)

Then ¯Z ¯
¯ ¯
¯ f ·dγ ¯ ≤ M V (γ, [a, b]) . (14.9)
¯ ¯
γ

Also if {fm } is a sequence of functions of C (γ ∗ ) which is converging uniformly to the

function, f on γ ∗ , then Z Z
lim fm · dγ = f ·dγ. (14.10)
m→∞ γ γ

In case γ (a) = γ (b) so the curve is a closed curve and for fk the k th component of f ,

mk ≤ fk (x) ≤ Mk

for all x ∈ γ ∗ , it also follows

¯Z ¯ Ã n !1/2
¯ ¯ 1 X
¯ f · dγ ¯ ≤ (Mk − mk )
2
V (γ, [a, b]) (14.11)
¯ ¯ 2
γ k=1

Proof: Let 14.8 hold. From the proof of Theorem 14.2.3, when ||P|| < δ m ,
¯Z ¯
¯ ¯
¯ f ·dγ − S (P)¯ ≤ 2 V (γ, [a, b])
¯ ¯ m
γ

and so ¯Z ¯
¯ ¯
¯ f ·dγ ¯ ≤ |S (P)| + 2 V (γ, [a, b])
¯ ¯ m
γ

Using the Cauchy Schwarz inequality and the above estimate in S (P) ,
n
X 2
≤ M |γ (tj ) − γ (tj−1 )| + V (γ, [a, b])
j=1
m
2
≤ M V (γ, [a, b]) + V (γ, [a, b]) .
m
This proves 14.9 since m is arbitrary.
To verify 14.10 use the above inequality to write
¯Z Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
¯ f ·dγ − fm · dγ ¯ = ¯ (f − fm ) · dγ (t)¯¯
¯ ¯
¯
γ γ γ

≤ max {|f ◦ γ (t) − fm ◦ γ (t)| : t ∈ [a, b]} V (γ, [a, b]) .

Since the convergence is assumed to be uniform, this proves 14.10.
Claim: Let γ be closed bounded variation curve. Then if c is a constant vector,
Z
c·dγ = 0
γ
368 LINE INTEGRALS

Proof of the claim: Let P ≡ {t0 , · · · , tp } be a partition with the property that
¯Z ¯
¯ Xp ¯
¯ ¯
¯ c·dγ− c· (γ (tk ) − γ (tk−1 ))¯ < ε.
¯ γ ¯
k=1

Consider the sum. It is of the form

p
X p
X p
X p−1
X
c · γ (tk ) − c · γ (tk−1 ) = c · γ (tk ) − c · γ (tk )
k=1 k=1 k=1 k=0
= c · γ (tp ) − c · γ (t0 ) = 0
because it is given that since γ ∗ is a closed curve, γ (t0 ) = γ (tp ) . This shows the claim.
It only remains to verify 14.11. In this case γ (a) = γ (b) and so for each vector c
Z Z
f ·dγ = (f − c) · dγ
γ γ

for any constant vector c. Let

1
ck = (Mk + mk )
2
Then for t ∈ [a, b]
Xn ¯ ¯2
¯ 1 ¯
|f (γ (t)) − c|
2
= ¯ ¯
¯fk (γ (t)) − 2 (Mk + mk )¯
k=1
Xn µ ¶2 n
1 1X 2
≤ (Mk − mk ) = (Mk − mk )
2 4
k=1 k=1

Then with this choice of c, it follows from 14.9 that

¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
¯ f ·dγ ¯ = ¯ (f − c) · dγ ¯
¯ ¯ ¯ ¯
γ γ
à n
!1/2
1 X 2
≤ (Mk − mk ) V (γ, [a, b])
2
k=1
This proves the lemma. ¥
It turns out to be much easier to evaluate line integrals in the case where there exists
a parameterization γ which is in C 1 ([a, b]) . The following theorem about approximation
will be very useful but first here is an easy lemma.
Lemma 14.2.5 Let γ : [a, b] → Rn be in C 1 ([a, b]) . Then V (γ, [a, b]) < ∞ so γ is
of bounded variation.
Proof: This follows from the following
¯
n ¯Z t j
¯
n
X X ¯
¯ ¯
|γ (tj ) − γ (tj−1 )| = ¯ γ 0 (s) ds¯
¯ tj−1 ¯
j=1 j=1
Xn Z tj
≤ |γ 0 (s)| ds
j=1 tj−1
Xn Z tj
≤ ||γ 0 ||∞ ds
j=1 tj−1
0
= ||γ ||∞ (b − a) .
14.2. THE LINE INTEGRAL 369

where
||γ 0 ||∞ ≡ max {|γ 0 (t)| : t ∈ [a, b]}
which exists because γ 0 is given to be continuous. Therefore it follows V (γ, [a, b]) ≤
||γ 0 ||∞ (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 γ : [a, b] → Rn be continuous and of bounded variation.

Let Ω be an open set containing γ ∗ and let f : Ω → Rn be continuous, and let ε > 0
be given. Then there exists η : [a, b] → Rn such that η (a) = γ (a) , γ (b) = η (b) ,
η ∈ C 1 ([a, b]) , and
||γ − η|| < ε, (14.12)
where ||γ − η|| ≡ max {|γ (t) − η (t)| : t ∈ [a, b]} . Also
¯Z Z ¯
¯ ¯
¯ f · dγ − f · dη ¯ < ε, (14.13)
¯ ¯
γ η

V (η, [a, b]) ≤ V (γ, [a, b]) , (14.14)

Proof: Extend γ to be defined on all R according to the rule γ (t) = γ (a) if t < a
and γ (t) = γ (b) if t > b. Now define
Z 2h
t+ (b−a) (t−a)
1
γ h (t) ≡ γ (s) ds.
2h 2h
−2h+t+ (b−a) (t−a)

where the integral is defined in the obvious way, that is componentwise. Since γ is
continuous, this is certainly possible. Then
Z b+2h Z b+2h
1 1
γ h (b) ≡ γ (s) ds = γ (b) ds = γ (b) ,
2h b 2h b
Z a Z a
1 1
γ h (a) ≡ γ (s) ds = γ (a) ds = γ (a) .
2h a−2h 2h a−2h
Also, because of continuity of γ and the fundamental theorem of calculus,
½ µ ¶µ ¶
1 2h 2h
γ 0h (t) = γ t+ (t − a) 1+ −
2h b−a b−a
µ ¶µ ¶¾
2h 2h
γ −2h + t + (t − a) 1+
b−a b−a
and so γ h ∈ C 1 ([a, b]) . The following lemma is significant.

Lemma 14.2.7 V (γ h , [a, b]) ≤ V (γ, [a, b]) .

Proof: Let a = t0 < t1 < ··· < tn = b. Then using the definition of γ h and changing
the variables to make all integrals over [0, 2h] ,
n
X
|γ h (tj ) − γ h (tj−1 )| =
j=1
¯ Z
Xn ¯ 2h · µ ¶
¯ 1 2h
¯ γ s − 2h + tj + (tj − a) −
¯ 2h 0 b−a
j=1
370 LINE INTEGRALS

µ ¶¸¯
2h ¯
γ s − 2h + tj−1 + (tj−1 − a) ¯¯
b−a
n ¯ µ
Z 2h X ¶
1 ¯
≤ ¯γ s − 2h + tj + 2h (tj − a) −
2h 0 j=1 ¯ b−a
µ ¶¯
2h ¯
γ s − 2h + tj−1 + (tj−1 − a) ¯¯ ds.
b−a
2h
For a given s ∈ [0, 2h] , the points, s − 2h + tj + b−a (tj − 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 (γ, [a − 2h, b + 2h]) = V (γ, [a, b]) . It follows
n
X
|γ h (tj ) − γ h (tj−1 )| ≤ V (γ, [a, b])
j=1

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 γ ∗ such that H is a compact subset of Ω. Let
0 < ε < dist γ ∗ , H C . Then there exists δ 1 such that if h < δ 1 , then for all t,
Z 2h
t+ (b−a) (t−a)
1
|γ (t) − γ h (t)| ≤ |γ (s) − γ (t)| ds
2h 2h
−2h+t+ (b−a) (t−a)
Z 2h
t+ (b−a) (t−a)
1
< εds = ε (14.15)
2h 2h
−2h+t+ (b−a) (t−a)

due to the uniform continuity of γ. 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 δ such that if ||P|| < δ, then
¯Z ¯ ¯Z ¯
¯ ¯ ε ¯¯ ¯ ε
¯
¯ f · dγ (t) − S (P)¯ < , ¯ f · dγ (t) − S (P) ¯<
¯ ¯ 3 ¯ h h
¯ 3
γ γh

for all h < 1. Here S (P) is a Riemann Stieltjes sum of the form
n
X
f (γ (τ i )) · (γ (ti ) − γ (ti−1 ))
i=1

and Sh (P) is a similar Riemann Stieltjes sum taken with respect to γ h instead of γ.
Because of 14.15 γ h (t) has values in H ⊆ Ω. Therefore, fix the partition P, and choose
h small enough that in addition to this, the following inequality is valid.
ε
|S (P) − Sh (P)| <
3

This is possible because of 14.15 and the uniform continuity of f on H. It follows

¯Z Z ¯
¯ ¯
¯ ¯
¯ f · dγ (t) − f · dγ h (t)¯ ≤
¯ γ γh ¯
¯Z ¯
¯ ¯
¯ f · dγ (t) − S (P)¯ + |S (P) − Sh (P)|
¯ ¯
γ
14.2. THE LINE INTEGRAL 371
¯ Z ¯
¯ ¯
¯ ¯
+ ¯Sh (P) − f · dγ h (t)¯ < ε.
¯ γh ¯

Let η ≡ γ h . Formula 14.14 follows from the lemma. This proves the theorem. ¥R
This is a very useful theorem because if γ is C 1 ([a, b]) , it is easy to calculate γ f dγ
and the above theorem allows a reduction to the case where γ is C 1 . The next theorem
shows how easy it is to compute these integrals in the case where γ is C 1 . First note
that if f is continuous and γ ∈ C 1 ([a,
R b]) , then by Lemma 14.2.5 and the fundamental
existence theorem, Theorem 14.2.3, γ f ·dγ exists.

Theorem 14.2.8 If f : γ ∗ → X is continuous and γ : [a, b] → Rn is in

1
C ([a, b]) and is a parameterization, then
Z Z b
f ·dγ = f (γ (t)) · γ 0 (t) dt. (14.16)
γ a

Proof: Let P be a partition of [a, b], P = {t0 , · · ·, tn } and ||P|| is small enough that
whenever |t − s| < ||P|| ,
|f (γ (t)) − f (γ (s))| < ε (14.17)
and ¯ ¯
¯Z n ¯
¯ X ¯
¯ f ·dγ − f (γ (τ )) · (γ (t ) − γ (t )) ¯
j−1 ¯ < ε.
¯ j j
¯ γ j=1 ¯

Now
n
X
f (γ (τ j )) · (γ (tj ) − γ (tj−1 ))
j=1
Z n
bX
= f (γ (τ j )) · X[tj−1 ,tj ] (s) γ 0 (s) ds
a j=1

where here ½
1 if s ∈ [p, q]
X[p,q] (s) ≡ .
0 if s ∈
/ [p, q]
Also,
Z b Z n
bX
f (γ (s)) · γ 0 (s) ds = f (γ (s)) · X[tj−1 ,tj ] (s) γ 0 (s) ds
a a j=1

and thanks to 14.17,

¯ P
= n
¯ j=1 f (γ(τ j ))·(γ(tj )−γ(tj−1 ))
¯z }| {
¯Z b n
¯ X
¯ 0
f (γ (τ j )) · X[tj−1 ,tj ] (s) γ (s) ds
¯
¯ a j=1
¯
¯
R ¯
= ab f (γ(s))·γ 0 (s)ds ¯
z }| {¯¯
Z bX n ¯
¯
− f (γ (s)) · X[tj−1 ,tj ] (s) γ 0 (s) ds¯
a j=1 ¯
¯
¯
¯
372 LINE INTEGRALS

n Z
X tj
≤ |f (γ (τ j )) − f (γ (s))| |γ 0 (s)| ds
j=1 tj−1
X
≤ ||γ 0 ||∞ ε (tj − tj−1 )
j

= ε ||γ 0 ||∞ (b − a) .

It follows that
¯Z Z b ¯
¯ ¯
¯ ¯
¯ f ·dγ − f (γ (s)) · γ 0 (s) ds¯
¯ γ a ¯
¯ ¯
¯Z Xn ¯
¯ ¯
≤ ¯ f ·dγ − f (γ (τ j )) · (γ (tj ) − γ (tj−1 ))¯¯
¯
¯ γ j=1 ¯

¯ ¯
¯X Z b ¯
¯ n ¯
¯
+¯ f (γ (τ j )) · (γ (tj ) − γ (tj−1 )) − f (γ (s)) · γ (s) ds¯¯
0
¯ j=1 a ¯
≤ ε ||γ 0 ||∞ (b − a) + ε.

Since ε is arbitrary, this verifies 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.

Definition 14.2.9 A function f : Ω → Rn for Ω an open set in Rn has a

potential if there exists a function, F, the potential, such that ∇F = f . Also if γ k :
[ak , bk ] → Rn is continuous and of bounded variation, for k = 1, · · ·, m and γ k (bk ) =
γ k+1 (ak ) , define
Z m Z
X
P
f · dγ k ≡ f · dγ k . (14.18)
m
k=1 γk k=1 γk

In addition to this, for γ : [a, b] → Rn , define −γ : [a, b] → Rn by −γ (t) ≡ γ (b + a − t) .

Thus γ simply traces out the points of γ ∗ in the opposite order.

The following lemma is useful and follows quickly from Theorem 14.2.2.

Lemma 14.2.10 In the above definition, there exists a continuous bounded variation
function, γ defined on some closed interval, [c, d] , such that γ ([c, d]) = ∪m
k=1 γ k ([ak , bk ])
and γ (c) = γ 1 (a1 ) while γ (d) = γ m (bm ) . Furthermore,
Z m Z
X
f · dγ = f · dγ k .
γ k=1 γk

If γ : [a, b] → Rn is of bounded variation and continuous, then

Z Z
f · dγ = − f · dγ.
γ −γ

The following theorem shows that it is very easy to compute a line integral when
the function has a potential.
14.2. THE LINE INTEGRAL 373

Theorem 14.2.11 Let γ : [a, b] → Rn be continuous and of bounded variation.

Also suppose ∇F = f on Ω, an open set containing γ ∗ and f is continuous on Ω. Then
Z
f · dγ = F (γ (b)) − F (γ (a)) .
γ

Proof: By Theorem 14.2.6 there exists η ∈ C 1 ([a, b]) such that γ (a) = η (a) , and
γ (b) = η (b) such that ¯Z ¯
Z
¯ ¯
¯ f · dγ − f · dη ¯¯ < ε.
¯
γ η
1
Then from Theorem 14.2.8, since η is in C ([a, b]) , it follows from the chain rule and
the fundamental theorem of calculus that
Z Z b Z b
d
f · dη = f (η (t)) η 0 (t) dt = F (η (t)) dt
η a a dt
= F (η (b)) − F (η (a)) = F (γ (b)) − F (γ (a)) .

Therefore, ¯¯ Z ¯¯
¯¯ ¯¯
¯¯(F (γ (b)) − F (γ (a))) − f (z) dz ¯¯¯¯ < ε
¯¯
γ

and since ε > 0 is arbitrary, This proves the theorem. ¥

Corollary 14.2.12 If γ : [a, b] → Rn is continuous, has bounded variation, is a

closed curve, γ (a) = γ (b) , and γ ∗ ⊆ Ω where Ω is an open set on which ∇F = f , then
Z
f · dγ = 0.
γ

Theorem 14.2.13 Let Ω be a connected open set and let f : Ω → Rn be con-

tinuous. Then f has a potential F if and only if
Z
f · dγ
γ

is path independent for all γ a bounded variation curve such that γ ∗ is contained in Ω.
This means the above line integral depends only on γ (a) and γ (b).

Proof: The first part was proved in Theorem 14.2.11. It remains to verify the
existence of a potential in the situation of path independence.
Let x0 ∈ Ω be fixed. Let S be the points x of Ω which have the property there is
a bounded variation curve joining x0 to x. Let γ x0 x denote such a curve. Note first
that S is nonempty. To see this, B (x0 , r) ⊆ Ω for r small enough. Every x ∈ B (x0 , r)
is in S. Then S is open because if x ∈ S, then B (x, r) ⊆ Ω for small enough r and if
y ∈ B (x, r) , you could go take γ x0 x and from x follow the straight line segment joining
x to y. In addition to this, Ω \ S must also be open because if x ∈ Ω \ S, then choosing
B (x, r) ⊆ Ω, 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 Ω
is connected, it follows Ω \ S = ∅. Thus for every x ∈ S, there exists γ x0 x , a bounded
variation curve from x0 to x.
Define Z
F (x) ≡ f · dγ x0 x
γ x0 x
374 LINE INTEGRALS

F is well defined by assumption. Now let lx(x+tek ) denote the linear segment from x to
x + tek . Thus to get to x + tek you could first follow γ x0 x to x and from there follow
lx(x+tek ) to x + tek . Hence
Z
F (x+tek ) − F (x) 1
= f · dlx(x+tek )
t t lx x+te
( k)

Z
1 t
= f (x + sek ) · ek ds → fk (x)
t 0
by continuity of f . Thus ∇F = f and This proves the theorem. ¥
Corollary 14.2.14 Let Ω be a connected open set and f : Ω → Rn . Then f has a
potential if and only if every closed, γ (a) = γ (b) , bounded variation curve contained
in Ω has the property that Z
f · dγ = 0
γ

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 Rectifiable Curves

There are examples of space filling 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 γ : [a, b] → γ ∗ ⊆ Rn where n ≥ 2 is a continuous bounded
variation curve. Then
mn (γ ∗ ) = 0
where mn denotes n dimensional Lebesgue measure.
Proof: Let ε > 0 be given. Let t0 ≡ a and if t0 , · · · , tk have been chosen, let tk+1
be the first number larger than tk such that
|γ (tk+1 ) − γ (tk )| = ε.
If the set of t such that |γ (t) − γ (tk )| = ε is nonempty, then this set is clearly closed
and so such a tk+1 exists until k is such that
γ ∗ ⊆ ∪kj=0 B (γ (tj ) , ε)
Let m be the last index of this process where tm+1 does not exist. How large is m? This
can be estimated because
m
X
V (γ, [a, b]) ≥ |γ (tk+1 ) − γ (tk )| = mε
k=0

and so m ≤ V (γ, [a, b]) /ε. Since γ ∗ ⊆ ∪m

j=0 B (γ (tj ) , ε) ,
m
X
mn (γ ∗ ) ≤ mn (B (γ (tj ) , ε))
j=0
V (γ, [a, b])
≤ cn εn = cn V (γ, [a, b]) εn−1
ε
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 375

Since ε was arbitrary, This proves the theorem. ¥

Since a ball has positive measure, this proves the following corollary.
Corollary 14.3.2 Let γ : [a, b] → γ ∗ ⊆ Rn where n ≥ 2 is a continuous bounded
variation curve. Then γ ∗ has empty interior.
Lemma 14.3.3 Let Γ be a simple closed curve. Then there exists a mapping θ :
S 1 → Γ where S 1 is the unit circle
© ª
(x, y) : x2 + y 2 = 1 ,
such that θ is one to one and continuous.
Proof: Since Γ is a simple closed curve, there is a parameterization γ and an interval
[a, b] such that γ is continuous and one to one on [a, b) and (a, b] with γ (a) = γ (b) .
Define θ −1 : Γ → S 1 by
µ µ ¶ µ ¶¶
−1 2π ¡ −1 ¢ 2π ¡ −1 ¢
θ (x) ≡ cos γ (x) − a , sin γ (x) − a
b−a b−a

Note that θ −1 is onto S 1 . The function is well defined because it sends the point γ (a) =
γ (b) to the same point, (1, 0) . It is also one to one. To see this note γ −1 is one
to one on Γ \ {γ (a) , γ (b)} . What about the case where x 6= γ (a) = γ (b)? Could
θ −1 (x) = θ −1 (γ (a))? In this case, γ −1 (x) is in (a, b) while γ −1 (γ (a)) = a so
θ −1 (x) 6= θ −1 (γ (a)) = (1, 0) .
Thus θ −1 is one to one on Γ.
Why is θ −1 continuous? Suppose xn → γ (a) = γ (b) first. Why does θ −1 (xn ) →
(1, 0) = θ −1 (γ (a))? Let {xn } denote any subsequence of the given sequence. Then by
compactness of [a, b] there exists a further subsequence, still denoted by xn such that
γ −1 (xn ) → t ∈ [a, b]
Hence by continuity of γ, xn → γ (t) and so γ (t) must equal γ (a) = γ (b) . It follows
from the assumption of what a simple curve is that t ∈ {a, b} . Hence θ −1 (xn ) converges
to either µ µ ¶ µ ¶¶
2π 2π
cos (a − a) , sin (a − a)
b−a b−a
or µ µ ¶ µ ¶¶
2π 2π
cos (b − a) , sin (b − a)
b−a b−a
but these are the same point. This has shown that if xn → γ (a) = γ (b) , there is a
subsequence such that θ −1 (xn ) → θ −1 (γ (a)) . Thus θ −1 is continuous at γ (a) = γ (b).
Next suppose xn → x 6= γ (a) ≡ p. Then there exists B (p, r) such that for all n
large enough, xn and x are contained in the compact set Γ \ B (p, r) ≡ K. Then γ is
continuous and one to one on the compact set γ −1 (K) ⊆ (a, b) and so by Theorem 5.1.3
γ −1 is continuous on K. In particular it is continuous at x so θ −1 (xn ) → θ −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 Ui will denote the inside
of a simple closed curve and Uo 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 13.0.26.
376 LINE INTEGRALS

© ª
Theorem 14.3.4 Let C denote the unit circle, (x, y) ∈ R2 : x2 + y 2 = 1 . Sup-
pose γ : C → Γ ⊆ R2 is one to one onto and continuous. Then R2 \ Γ consists of two
components, a bounded component (called the inside) Ui and an unbounded component
(called the outside), Uo . Also the boundary of each of these two components of R2 \ Γ
is Γ and Γ has empty interior.

Proof: That R2 \ Γ consists of two components, Uo and Ui follows from the Jordan
separation theorem. There is exactly one unbounded component because Γ is bounded
and so Ui is defined as the bounded component. It remains to verify the assertion about
Γ being the boundary. Let x be a limit point of Ui . Then it can’t be in Uo because these
are both open sets. Therefore, all the limit points of Ui are in Ui ∪ Γ. Similarly all the
limit points of Uo are in Uo ∪ Γ. Thus ∂Ui ⊆ Γ and ∂Uo ⊆ Γ.
I claim Γ has empty interior. This follows because by Theorem 5.1.3 on Page 88,
γ and γ −1 must both be continuous since C is compact. Thus if B is an open ball
contained in Γ, it follows from invariance of domain that γ −1 (B) is an open set in
R2 . But this needs to be contained in C which is a contradiction because C has empty
interior obviously.
Now let x ∈ Rn such that x ∈ / Uo ∪ Ui . Then x ∈ Γ and must be a limit point of
either Uo or Ui since if this were not so, Γ would be forced to have nonempty interior.
Hence U1 ∪ U2 = R2 . Next I will show ∂Ui = ∂Uo . Suppose then that

p ∈ ∂Ui \ ∂Uo

Then p ∈ / Uo because p ∈ ∂Ui ⊆ Γ which is disjoint from Uo . Thus p is not in Uo

because it is given to not be in ∂Uo . Hence there is a ball centered at p, B (p, r) which
contains no points of Uo . Thus B (p, r) ⊆ Ui and so p is an interior point of Ui which
implies p is actually in Ui , a contradiction to p ∈ ∂Ui . Thus ∂Ui \ ∂Uo = ∅ and a similar
argument shows ∂Uo \ ∂Ui = ∅. Thus ∂Ui = ∂Uo and so if x ∈ Γ, then it is in either
∂Ui or ∂Uo and these are equal. Thus

∂Ui = ∂Uo ⊆ Γ ⊆ ∂Ui ∪ ∂Uo = ∂Ui = ∂Uo .

This proves the theorem. ¥

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 defines an equivalence relation. Then from Lemma 13.0.25 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 13.0.26, let Γ be a

simple closed curve and let γ ∗ be a straight line segment such that the open segment,
γ ∗o , γ ∗ without its endpoints, is contained in Ui such that the intersection of γ ∗ with Γ
equals {p, q} . Then this line segment divides Ui into two connected open sets, U1i , U2i
which are the insides of two simple closed curves such that

Ui = U1i ∪ γ ∗o ∪ U2i

Proof: Denote by C the unit circle and let θ : C → Γ be continuous one to one and
onto. Say θ (a) = p and θ (b) = q. Let Cj , j = 1, 2 denote the two circular arcs joining
a and b. Thus letting Γj ≡ θ (Cj ) it follows Γ1 , Γ2 are simple curves whose union is Γ
which intersect at the points p and q. Letting Γj ∪ γ ∗ ≡ Jj it follows Jj is a simple
closed curve. Here is why. Define
½
θ (x) if x ∈ C1
h1 (x) ≡
f2 (x) if x ∈ C2
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 377

where fj is a continuous one to one onto mapping from Cj to γ ∗ . Then h1 is continuous

and maps C one to one and onto J1 . Define h2 similarly. Denote by Uji the inside and
Ujo the outside of Jj .

p
Γ

Γ1 U1i U2i Γ2
γ∗

Claim 1: U1i , U2i ⊆ Ui , U1i , U2i contain no points of J2 ∪ J1 .

Proof: First consider the claim that U1i , U2i contain no points of J2 ∪ J1 . If
x ∈ Γ2 \ {p, q} , then near x there are points of Ui and Uo . Therefore, if x ∈ U1i ,
there would be points of both Ui and Uo in U1i . Now Uo contains no points of γ ∗ by
assumption and it contains no points of Γ by definition. Therefore, since Uo is connected,
it must be contained in U1i but this is impossible because Uo is unbounded and U1i is
bounded. Thus x ∈ / U1i . Similarly x ∈
/ U2i . Similar reasoning applied to x ∈ Γ1 \ {p, q}
implies no point of Γ \ {p, q} can be in either U1i or U2i . If x ∈ γ ∗ then by definition it
is not in either U1i or U2i and so neither U1i not U2i contain any points of J2 ∪ J1 . If
U1i ∩ Uo 6= ∅, then the whole connected set U1i is contained in Uo since otherwise U1i
would contain points of Γ which, by what was just shown, is not the case. But now
this is a contradiction to the open segment of γ ∗ being contained in Ui . A point x on
this open segment must have points of U1i near it by the Jordan curve theorem but
if U1i ⊆ Uo , then this cannot happen because a small ball centered at x contains only
points of Ui and none of Uo .
Similarly U2i ⊆ Ui . Letting γ ∗o denote the open segment of γ ∗ , it follows

Ui ⊇ U1i ∪ γ ∗o ∪ U2i (14.19)

Claim 2:U1i ∩ U2i = ∅, U2i ⊆ U1o , U1i ⊆ U2o .

Proof: If x is a point of U1i which is in U2i then U1i must be contained in U2i .
This is because U1i is connected and, as noted above, U1i contains no boundary points
of U2i because these points are all contained in J2 ∪ J2 . Similarly U2i would need to be
contained in U1i and so these two interior components would need to coincide. But now
consider x ∈ Γ2 \ {p, q} . Let r be small enough that B (x, r) ∩ J1 = ∅. Then this ball
contains points of U2i = U1i and points of U2o . However, this ball is contained in one
of the complementary components of J1 and since it contains points of U1i , this forces
B (x, r) to be contained in U1i . But now the same contradiction as above holds namely:
There exists a point of Uo in U1i which forces Uo to be contained in U1i . Therefore,
U1i ∩ U2i = ∅. Also U2i ⊆ U1o because if not, there would be points of U1i in U2i and it
was just shown this doesn’t happen. Similarly U1i ⊆ U2o . This shows Claim 2.
Next I need to verify that equality holds in 14.19. First I will argue

U1i ∪ γ ∗o ∪ U2i

is an open set. Let y ∈ γ ∗o and let B (y, r) ∩ Γ = ∅.

378 LINE INTEGRALS

p
Γ
y
Γ1 U1i U2i Γ2
γ∗

Thus γ ∗o divides B (y, r) into two halves, H1 , H2 . The ball contains points of U2i
by the Jordan curve theorem. Say H2 contains some of these points. Then I claim H2
cannot contain any points of U1i . This is because if it did, there would be a segment
joining a point of U1i with a point of U2i which is contained in H2 which is a connected
open set which is therefore contained in a single component of J1C . This is a contradiction
because as shown above, U2i ⊆ U1o . Could H2 contain any points of U2o ? No because
then there would be a segment joining a point of U2o to a point of U2i which is contained
in the same component of J2C . Therefore, H2 consists entirely of points of U2i . Similarly
H1 consists entirely of points of U1i . Therefore, U1i ∪ γ ∗o ∪ U2i is an open set because
the only points which could possibly fail to be interior points, those on γ ∗0 are interior
points of U1i ∪ γ ∗o ∪ U2i .
Suppose equality does not hold in 14.19. Then there exists w ∈ Ui \(U1i ∪ γ ∗o ∪ U2i ) .
Let x ∈ U1i ∪ γ ∗o ∪ U2i . Then since Ui is connected and open, there exists a continuous
mapping r : [0, 1] → Ui such that r (0) = x and³r (1) = w. Since U´1i ∪ γ ∗o ∪ U2i is open,
C
there exists a first point in the closed set r−1 (U1i ∪ γ ∗o ∪ U2i ) , s. Thus r (s) is a
limit point of U1i ∪ γ ∗o ∪ U2i but is not in this set which implies it is in U1o ∩ U2o . It
follows r (s) is a limit point of either U1i or U2i because each point of γ ∗o is a limit point
of U1i and U2i . Also, r (s) cannot be in γ ∗0 because it is not in U1i ∪ γ ∗o ∪ U2i . Suppose
without loss of generality it is a limit point of U1i . Then every ball containing r (s)
must contain points of U1o ∩ U2o ⊆ U1o as well as points U1i . But by the Jordan curve
theorem, this implies r (s) is in J1 but is not in γ ∗o . Therefore, r (s) is a point of Γ and
this contradicts r (s) ∈ Ui . 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 rectifiable 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 {|y1 − y2 | : (x1 , y1 ) , (x2 , y2 ) ∈ A} . The width of A will be
defined similarly.

Lemma 14.3.6 Let Γ be a simple closed rectifiable curve. Also let δ > 0 be given
such that 2δ is smaller than both the height and width of Γ. Then there exist finitely
n
many non overlapping regions {Rk }k=1 consisting of simple closed rectifiable curves
along with their interiors whose union equals Ui ∪ Γ. These regions consist of two kinds,
those contained in Ui and those with nonempty intersection with Γ. 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 = mδ or y = kδ for m, k integers
along with arcs from Γ. The regions contained in Ui consist of rectangles. Thus all of
these regions have boundaries which are rectifiable simple closed curves. Also all regions
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 379

are contained in a square having sides of length no more than 2δ. There are at most
µ ¶
V (Γ)
4 +1
δ
border regions. The construction also yields an orientation for Γ and for all these
regions, and the orientations for any segment shared by two regions are opposite.

Proof: Let Γ = γ ([a, b]) where γ = (γ 1 , γ 2 ) . Let

y1 ≡ max {γ 2 (t) : t ∈ [a, b]}

and let
y2 ≡ min {γ 2 (t) : t ∈ [a, b]} .
Thus (x1 , y1 ) is the “top” point of Γ while (x2 , y2 ) is the “bottom” point of Γ. Consider
the lines y = y1 and y = y2 . By assumption |y1 − y2 | > 2δ. Consider the line l given
by y = mδ where m is chosen to make mδ as close as possible to (y1 + y2 ) /2. Thus
y1 > mδ > y2 . By Theorem 14.3.4 or 13.0.26(xj , yj ) j = 1, 2, being on Γ are both
limit points of Ui so there exist points pj ∈ Ui such that p1 is above l and p2 is below
l. (Simply pick pj very close to (xj , yj ) and yet in Ui and this will take place.) The
horizontal line l must have nonempty intersection with Ui because Ui is connected. If
it had empty intersection it would be possible to separate Ui into two nonempty open
sets, one containing p1 and the other containing p2 .
Let q be a point of Ui which is also in l. Then there exists a maximal segment
of the line l containing q which is contained in Ui ∪ Γ. This segment, γ ∗ satisfies the
conditions of Lemma 14.3.5 and so it divides Ui into disjoint open connected sets whose
boundaries are simple rectifiable closed curves. Note the line segment has finite length.
Letting Γj be the simple closed curve which contains pj , orient γ ∗ as part of Γ2 such
that motion is from right to left. As part of Γ1 the motion along the curve is from left
to right. By Proposition 14.1.7 this provides an orientation to each Γj . By Proposition
14.1.8 there exists an orientation for Γ which is consistent with these two orientations
on the Γj .
Now do the same process to the two simple closed curves just obtained and continue
till all regions have height less than 2δ. 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 2δ. The reason for
this is that if it did not end, the curve Γ could not have finite 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 2δ. This takes care of finding 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 = kδ where k
is an integer.
How many border regions are there? Denote by V (Γ) the length of Γ. Now de-
compose Γ into N arcs of length δ with maybe one having length less than δ. Thus
N − 1 ≤ V (Γ)δ and so
V (Γ)
N≤ +1
δ
380 LINE INTEGRALS

The resulting regions are each contained in a box having sides of length no more than
2δ in length. Each of these N arcs can’t intersect any more than four of these boxes.
Therefore, at most 4N boxes of the construction can intersect Γ. Thus there are no
more than µ ¶
V (Γ)
4 +1
δ
border regions. This proves the lemma. ¥

14.3.2 Orientation And Green’s Formula

How do you describe the orientation of a simple closed rectifiable 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 rectifiable simple closed curve will be defined in terms of a very
important formula known as Green’s formula. First I will present Green’s 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 suffices 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 γ denote the oriented boundary of R as shown,

Z Z
(Qx (x, y) − Py (x, y)) dm2 = f ·dγ
R γ

where
f (x, y) ≡ (P (x, y) , Q (x, y)) .
In this context the line integral is usually written using the notation
Z
P dx + Qdy.
∂R

Proof: This follows from direct computation. A parameterization for the bottom
line of R is
γ B (t) = (a + t (b − a) , c) , t ∈ [0, 1]
A parameterization for the top line of R with the given orientation is

γ T (t) = (b + t (a − b) , d) , t ∈ [0, 1]
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 381

A parameterization for the line on the right side is

γ R (t) = (b, c + t (d − c)) , t ∈ [0, 1]

and a parameterization for the line on the left side is

γ L (t) = (a, d + t (c − d)) , t ∈ [0, 1]

Now it is time to do the computations using Theorem 14.2.8.

Z Z 1
f · dγ = P (a + t (b − a) , c) (b − a) dt
γ 0
Z 1
+ P (b + t (a − b) , d) (a − b) dt
0
Z 1 Z 1
+ Q (b, c + t (d − c)) (d − c) dt + Q (a, d + t (c − d)) (c − d) dt
0 0
Changing the variables and combining the integrals, this equals
Z b Z b Z d Z d
= P (x, c) dx − P (x, d) dx + Q (b, y) dy − Q (a, y) dy
a a c c

Z bZ d Z dZ b
= − Py (x, y) dydx + Qx (x, y) dxdy
a c c a
Z
= (Qx − Py ) dm2
R

By Fubini’s theorem, Theorem 9.2.3 on Page 212. (To use this theorem you can extend
the functions to equal 0 off R.) This proves the lemma. ¥
Note that if the rectangle were oriented in the opposite way, you would get
Z Z
f · dγ = (Py − Qx ) dm2
γ R

With this lemma, it is possible to prove Green’s theorem and also give an ana-
lytic criterion which will distinguish between different orientations of a simple closed
rectifiable curve. First here is a discussion which amounts to a computation.
nδ
Let Γ be a rectifiable simple closed curve with inside Ui and outside Uo . Let {Rk }k=1
denote the non overlapping regions of Lemma 14.3.6 all oriented as explained there and
let Γ also be oriented as explained there. It could be shown that all the regions contained
in Ui 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 Bδ be the set of border regions and let Iδ be the rectangles contained in Ui . Thus
in taking the sum of the line integrals over the boundaries of the interior rectangles, the
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 Iδ .
Now let f (x, y) = (P (x, y) , Q (x, y)) be a vector field which is C 1 on Ui , and suppose
also that both Py and Qx are in L1 (Ui ) (Absolutely integrable) and that P, Q are
continuous on Ui ∪Γ. (An easy way to get all this to happen is to let P, Q be restrictions
to Ui ∪ Γ of functions which are C 1 on some open set containing Ui ∪ Γ.) Note that

∪δ>0 {R : R ∈ Iδ } = Ui
382 LINE INTEGRALS

and that for

Iδ ≡ ∪ {R : R ∈ Iδ } ,
the following pointwise convergence holds.

lim XIδ (x) = XUi (x) .

δ→0

By the dominated convergence theorem,

Z Z
lim (Qx − Py ) dm2 = (Qx − Py ) dm2
δ→0 I Ui
Z δ
Z
lim (Py − Qx ) dm2 = (Py − Qx ) dm2
δ→0 Iδ Ui

Let ∂R denote the boundary of R for R one of these regions of Lemma 14.3.6 oriented
2
as described. Let wδ (R) denote
2
(max {Q (x) : x ∈ ∂R} − min {Q (x) : x ∈ ∂R})
2
+ (max {P (x) : x ∈ ∂R} − min {P (x) : x ∈ ∂R})

By uniform continuity of P, Q on the compact set Ui ∪Γ, if δ is small enough, wδ (R) < ε
for all R ∈ Bδ . Then for R ∈ Bδ , it follows from Theorem 14.2.4
¯Z ¯
¯ ¯ 1
¯ f · dγ ¯¯ ≤ wδ (R) (V (∂R)) < ε (V (∂R)) (14.20)
¯ 2
∂R

whenever δ is small enough. Always let δ be this small.

Also since the line integrals cancel on shared edges
X Z X Z Z
f · dγ + f · dγ = f · dγ (14.21)
R∈Iδ ∂R R∈Bδ ∂R Γ

Consider the second sum on the left. From 14.20

¯ ¯
¯X Z
¯
¯
¯ X ¯¯Z ¯
¯ X
¯ f · dγ ¯ ≤ ¯ f · dγ ¯¯ ≤ ε (V (∂R))
¯ ¯ ¯
R∈Bδ ∂R ∂R
R∈Bδ R∈Bδ

Denote by ΓR the part of Γ which is contained in R ∈ Bδ and V (ΓR ) is its length. Then
the above sum equals
à !
X
ε V (ΓR ) + Bδ = ε (V (Γ) + Bδ )
R∈Bδ

where Bδ 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
µ ¶
V (Γ)
4 +1
δ

of these border regions. Furthermore, the sum of the lengths of all four edges of one of
these is no more than 8δ and so
µ ¶
V (Γ)
Bδ ≤ 4 + 1 8δ = 32V (Γ) + 32δ.
δ
14.3. SIMPLE CLOSED RECTIFIABLE CURVES 383

Thus the absolute value of the second sum on the right in 14.21 is dominated by

ε (33V (Γ) + 32δ)

Since ε was arbitrary, this formula implies with Green’s theorem proved above for
squares, Z X Z X Z
f · dγ = lim f · dγ + lim f · dγ
Γ δ→0 ∂R δ→0 ∂R
R∈Iδ R∈Bδ

X Z Z Z
= lim f · dγ = lim ± (Qx − Py ) dm2 = ± (Qx − Py ) dm2
δ→0 ∂R δ→0 Iδ Ui
R∈Iδ

where the ± adusts for whether the interior rectangles are all oriented positively or all
oriented negatively. ¥
This has proved the general form of Green’s theorem which is stated in the following
theorem.

Theorem 14.3.8 Let Γ be a rectifiable simple closed curve in R2 having inside

Ui and outside Uo . Let P, Q be functions with the property that

Qx , Py ∈ L1 (Ui )

and P, Q are C 1 on Ui . Assume also P, Q are continuous on Γ ∪ Ui . Then there exists

an orientation for Γ (Remember there are only two.) such that for

f (x, y) = (P (x, y) , Q (x, y)) ,

Z Z
f · dγ = (Qx − Py ) dm2 .
Γ Ui

Proof: In the construction of the regions, an orientation was imparted to Γ. The

above computation shows
Z Z
f · dγ = ± (Qx − Py ) dm2
Γ Ui

If the area integral equals Z

− (Qx − Py ) dm2 ,
Ui

just take the other orientation for Γ. This proves the theorem. ¥
With this wonderful theorem, it is possible to give an analytic description of the two
different orientations of a rectifiable 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 Z Z
f · dγ = (Py − Qx ) dm2 .
Γ Ui

There are other regions for which Green’s theorem holds besides just the inside and
boundary of a simple closed curve. For Γ a simple closed curve and Ui its inside, lets
refer to Ui ∪ Γ as a Jordan region. When you have two non overlapping Jordan regions
which intersect in a finite number of simple curves, you can delete the interiors of these
simple curves and what results will also be a region for which Green’s theorem holds.
This is illustrated in the following picture.
384 LINE INTEGRALS

Ui2

U1i

There are two Jordan regions here with insides U1i and U2i 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 Γ the curve on the outside of the whole thing and Γ1 and Γ2 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 = U1i ∪ U2i ∪ {Open segments of intersection}

as shown in the following picture,

Γ2 U Γ1
Γ

it follows from applying Green’s theorem to both of the Jordan regions,

Z Z Z Z
f · dγ+ f · dγ 1 + f · dγ 2 = (Qx − Py ) dm2
Γ Γ1 Γ2
ZU1i ∪U2i
= (Qx − Py ) dm2
U

To make this simpler, just write it in the form

Z Z
f · dγ = (Qx − Py ) dm2
∂U U

where ∂U is oriented as indicated in the picture and involves the three oriented curves
Γ, Γ1 , Γ2 .

14.4 Stoke’s 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 Green’s theorem and that when you have the general version of Green’s theorem
this can be used to obtain a general version of Stoke’s theorem using a simple identity.
This is because Stoke’s theorem is really just a three dimensional version of the two
dimensional Green’s theorem. This will be made more precise below.
14.4. STOKE’S THEOREM 385

To begin with suppose Γ is a rectifiable curve in R2 having parameterization α :

[a, b] → Γ for α a continuous function. Let R : U → Rn be a C 1 function where U
contains α∗ . Then one could define a curve

γ (t) ≡ R (α (t)) , t ∈ [a, b] .

Lemma 14.4.1 The curve γ ∗ where γ is as just described is a rectifiable curve. If

F is defined and continuous on γ ∗ then
Z Z
F · dγ = ((F ◦ R) · Ru , (F ◦ R) · Rv ) · dα
γ α

where Ru signifies the partial derivative of R with respect to the variable u.

Proof: Let © ª
K ≡ y ∈ R2 : dist (y, α∗ ) ≤ r
where r is small enough that K ⊆ U . This is easily done because α∗ is compact. Let

CK ≡ max {||DR (x)|| : x ∈ K}

Consider
n−1
X
|R (α (tj+1 )) − R (α (tj ))| (14.22)
j=0

where {t0 , · · · , tn } is a partition of [a, b] . Since α is continuous, there exists a δ such

that if ||P|| < δ, then the segment

{α (tj ) + s (α (tj+1 ) − α (tj )) : s ∈ [0, 1]}

is contained in K. Therefore, by the mean value inequality, Theorem 6.4.2,

n−1
X n−1
X
|R (α (tj+1 )) − R (α (tj ))| ≤ CK |α (tj+1 ) − α (tj )|
j=0 j=0

Now if P is any partition, 14.22 can always be made larger by adding in points to P till
||P|| < δ and so this shows

V (γ, [a, b]) ≤ CK V (α, [a, b]) .

This proves the first part.

Next consider the claim about the integral. Let

G (v, x) ≡ R (x + v) − R (x) − DR (x) (v) .

Then
D1 G (v, x) = DR (x + v) − DR (x)
and so by uniform continuity of DR on the compact set K, it follows there exists δ > 0
such that if |v| < δ, then for all x ∈ α∗ ,

||DR (x + v) − DR (x)|| = ||D1 G (v, x)|| < ε.

By Theorem 6.4.2 again it follows that for all x ∈ α∗ and |v| < δ,

|G (v, x)| = |R (x + v) − R (x) − DR (x) (v)| ≤ ε |v| (14.23)

386 LINE INTEGRALS

Letting ||P|| be small enough, it follows from the continuity of α that

|α (tj+1 ) − α (tj )| < δ

Therefore for such P,

n−1
X
F (γ (tj )) · (γ (tj+1 ) − γ (tj ))
j=0

n−1
X
= F (R (α (tj ))) · (R (α (tj+1 )) − R (α (tj )))
j=0

n−1
X
= F (R (α (tj ))) · [DR (α (tj )) (α (tj+1 ) − α (tj )) + o (α (tj+1 ) − α (tj ))]
j=0

where

o (α (tj+1 ) − α (tj )) = R (α (tj+1 )) − R (α (tj )) − DR (α (tj )) (α (tj+1 ) − α (tj ))

and by 14.23,
|o (α (tj+1 ) − α (tj ))| < ε |α (tj+1 ) − α (tj )|
It follows
¯ ¯
¯n−1 n−1 ¯
¯X X ¯
¯ F (γ (tj )) · (γ (tj+1 ) − γ (tj )) − F (R (α (tj ))) · DR (α (tj )) (α (tj+1 ) − α (tj ))¯¯
¯
¯ j=0 j=0 ¯
(14.24)
n−1
X n−1
X
≤ |o (α (tj+1 ) − α (tj ))| ≤ ε |α (tj+1 ) − α (tj )| ≤ εV (α, [a, b])
j=0 j=0

Consider the second sum in 14.24. A term in the sum equals

F (R (α (tj ))) · (Ru (α (tj )) (α1 (tj+1 ) − α1 (tj )) + Rv (α (tj )) (α2 (tj+1 ) − α2 (tj )))

= (F (R (α (tj ))) · Ru (α (tj )) , F (R (α (tj ))) · Rv (α (tj ))) · (α (tj+1 ) − α (tj ))

By continuity of F, Ru and Rv , it follows that sum converges as ||P|| → 0 to
Z
((F ◦ R) · Ru , (F ◦ R) · Rv ) · dα
α

Therefore, taking the limit as ||P|| → 0 in 14.24

¯Z Z ¯
¯ ¯
¯ F · dγ − ((F ◦ R) · Ru , (F ◦ R) · Rv ) · dα¯¯ < εV (α, [a, b]) .
¯
γ α

Since ε > 0 is arbitrary, This proves the lemma. ¥

The following is a little identity which will allow a proof of Stoke’s theorem to follow
from Green’s theorem. First recall the following definition from calculus of the curl of
a vector field and the cross product of two vectors from calculus.

Definition 14.4.2 Let u ≡ (a, b, c) and v ≡ (d, e, f ) be two vectors in R3 . Then

¯ ¯
¯ i j k ¯
¯ ¯
u × v ≡ ¯¯ a b c ¯¯
¯ d e f ¯
14.4. STOKE’S THEOREM 387

where the determinant is expanded formally along the top row. Let f : U → R3 for
U ⊆ R3 denote a vector field. The curl of the vector field yields another vector field
and it is defined as follows.

(curl (f ) (x))i ≡ (∇ × f (x))i

where here ∂j means the partial derivative with respect to xj and the subscript of i in
(curl (f ) (x))i means the ith 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 ¯
¯ ¯
¯ ∂ ∂ ∂ ¯.
¯ ∂x ∂y ∂z ¯
¯ f1 (x, y, z) f2 (x, y, z) f3 (x, y, z) ¯

Note the similarity with the cross product. More precisely and less evocatively,
µ ¶ µ ¶ µ ¶
∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
∇ × f (x, y, z) =≡ − i+ − j+ − k.
∂y ∂z ∂z ∂x ∂x ∂y

In the above, i = e1 , j = e2 , and k = e3 the standard unit basis vectors for R3 .

With this definition, here is the identity.

Lemma 14.4.3 Let R : U → V ⊆ R3 where U is an open subset of R2 and V is an

open subset of R3 . Suppose R is C 2 and let F be a C 1 vector field defined in V.

(Ru × Rv ) · (∇ × F) (R (u, v)) = ((F ◦ R)u · Rv − (F ◦ R)v · Ru ) (u, v) . (14.25)

Proof: Letting x, y, z denote the components of R (u) and f1 , f2 , f3 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 ¯¯ ¯¯ i j k ¯¯
¯
¯ xu yu zu ¯ · ¯ ∂x ∂y ∂z ¯
¯ ¯ ¯ ¯
¯ xv yv zv ¯ ¯ f1 f2 f3 ¯

= (f3y − f2z ) (yu zv − zu yv ) + (f1z − f3x ) (zu xv − xu zv ) + (f2x − f1y ) (xu yv − yu xv )

= f3y yu zv + f2z zu yv + f1z zu xv + f3x xu zv + f2x xu yv + f1y yu xv
− (f2z yu zv + f3y zu yv + f1z xu zv + f3x zu xv + f2x yu xv + f1y xu yv )
= f1y yu xv + f1z zu xv + f2x xu yv + f2z zu yv + f3x xu zv + f3y yu zv
− (f1y yv xu + f1z zv xu + f2x xv yu + f2z zv yu + f3x xv zu + f3y yv zu )
At this point add in and subtract off certain terms. Then the above equals

= f1x xu xv + f1y yu xv + f1z zu xv + f2x xu yv + f2x yu yv

µ +f2z zu yv + f3x xu zv + f3y yu zv + f3z zu zv ¶
f1x xv xu + f1y yv xu + f1z zv xu + f2x xv yu + f2x yv yu
−
+f2z zv yu + f3x xv zu + f3y yv zu + f3z zv zu

∂f1 ◦ R (u, v) ∂f2 ◦ R (u, v) ∂f3 ◦ R (u, v)

= xv + yv + zv
∂u ∂u ∂u
µ ¶
∂f1 ◦ R (u, v) ∂f2 ◦ R (u, v) ∂f3 ◦ R (u, v)
− xu + yu + zu
∂v ∂v ∂v

= ((F ◦ R)u · Rv − (F ◦ R)v · Ru ) (u, v) .

388 LINE INTEGRALS

This proves the lemma. ¥

Let U be a region in R2 for which Green’s theorem holds. Thus Green’s theorem
says that for P, Q continuous on Ui ∪ Γ, Pv , Qu ∈ L1 (Ui ∪ Γ) , P, Q being C 1 on Ui ,
Z Z
(Qu − Pv ) dm2 = f · dα
U ∂U

where ∂U consists of some simple closed rectifiable 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 (Stoke’s Theorem) Let ¡U be any 2

¢ region in R for which the
2 3
conclusion of Green’s theorem holds. Let R ∈ C U , R be a one to one function. Let

γ j = R ◦ αj ,

where the αj are parameterizations for the oriented curves making up the boundary of
U such that the conclusion of Green’s theorem holds. Let S denote the surface,

S ≡ {R (u, v) : (u, v) ∈ U } ,

Then for F a C 1 vector field defined near S,

n Z
X Z
F · dγ i = (Ru (u, v) × Rv (u, v)) · (∇ × F (R (u, v))) dm2
i=1 γi U

Proof: By Lemma 14.4.1,

n Z
X
F · dγ j =
j=1 γj

n Z
X
((F ◦ R) · Ru , (F ◦ R) · Rv ) · dαj
j=1 αj

By the assumption that the conclusion of Green’s theorem holds for U , this equals
Z
[((F ◦ R) · Rv )u − ((F ◦ R) · Ru )v ] dm2
ZU
= [(F ◦ R)u · Rv + (F ◦ R) · Rvu − (F ◦ R) · Ruv − (F ◦ R)v · Ru ] dm2
ZU
= [(F ◦ R)u · Rv − (F ◦ R)v · Ru ] dm2
U

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
Z
(Ru (u, v) × Rv (u, v)) · (∇ × F (R (u, v))) dm2
U

This proves Stoke’s 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 final
result, only first derivatives of R occur.
14.5. INTERPRETATION AND REVIEW 389

14.5 Interpretation And Review

To understand the interpretation of Stoke’s 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 R3 . It is very different 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.

Definition 14.5.1 Three vectors, a, b, c form a right handed system if when

you extend the fingers 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.

±
c

y a

b
¼

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 differ 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 fingers 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 e1 , e2 , e3 .
The following is the geometric description of the cross product. It gives both the
direction and the magnitude and therefore specifies the vector.

Definition 14.5.2 Let a and b be two vectors in R3 . Then a × b is defined by

the following two rules.

1. |a × b| = |a| |b| sin θ where θ is the included angle.

390 LINE INTEGRALS

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.

3
|b|sin(θ)
b ¼

θ a -

The cross product satisfies the following properties.

a × b = − (b × a) , a × a = 0, (14.26)

For α a scalar,
(αa) ×b = α (a × b) = a× (αb) , (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 definition. The vectors a × b and b × a
have the same magnitude, |a| |b| sin θ, and an application of the right hand rule shows
they have opposite direction. Formula 14.27 is also fairly clear. If α is a nonnegative
scalar, the direction of (αa) ×b is the same as the direction of a × b,α (a × b) and
a× (αb) while the magnitude is just α times the magnitude of a × b which is the same
as the magnitude of α (a × b) and a× (αb) . Using this yields equality in 14.27. In the
case where α < 0, everything works the same way except the vectors are all pointing in
the opposite direction and you must multiply by |α| when comparing their magnitudes.
The distributive laws are much harder to establish but the second follows from the first
quite easily. Thus, assuming the first, 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

Definition 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.
14.5. INTERPRETATION AND REVIEW 391

The following is a picture of such a thing.

a×b
Á

c
3
θ b
a -

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 θ where θ 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 θ = 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 definition,
i × j = k j × i = −k
k × i = j i × k = −j
j × k = i k × j = −i
392 LINE INTEGRALS

With this information, the following gives the coordinate description of the cross prod-
uct.
Proposition 14.5.4 Let a = a1 i + a2 j + a3 k and b = b1 i + b2 j + b3 k be two vectors.
Then
a × b = (a2 b3 − a3 b2 ) i+ (a3 b1 − a1 b3 ) j+
+ (a1 b2 − a2 b1 ) k. (14.30)
Proof: From the above table and the properties of the cross product listed,
(a1 i + a2 j + a3 k) × (b1 i + b2 j + b3 k) =
a1 b2 i × j + a1 b3 i × k + a2 b1 j × i + a2 b3 j × k+
+a3 b1 k × i + a3 b2 k × j
= a1 b2 k − a1 b3 j − a2 b1 k + a2 b3 i + a3 b1 j − a3 b2 i
= (a2 b3 − a3 b2 ) i+ (a3 b1 − a1 b3 ) j+ (a1 b2 − a2 b1 ) k (14.31)
This proves the proposition. ¥
The easy way to remember the above formula is to write it as follows.
¯ ¯
¯ i j k ¯¯
¯
a × b = ¯¯ a1 a2 a3 ¯¯ (14.32)
¯ b1 b2 b3 ¯
where you expand the determinant along the top row. This yields
(a2 b3 − a3 b2 ) i− (a1 b3 − a3 b1 ) j+ (a1 b2 − a2 b1 ) 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 define what is meant by a smooth surface.
Definition 14.5.5 Let S be a subset of R3 . Then S is a smooth surface
if there exists an open set, U ⊆ R2 and a C 1 function, R defined on U such that
R (U ) = S, R is one to one, and for all (u, v) ∈ U,
Ru × Rv 6= 0. (14.34)
This last condition ensures that there is always a well defined 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.
Definition 14.5.6 Let u1 , u2 be vectors in R3 . The 2 dimensional parallelogram
determined by these vectors will be denoted by P (u1 , u2 ) and it is defined as
 
X 2 
P (u1 , u2 ) ≡ sj uj : sj ∈ [0, 1] .
 
j=1

Then the area of this parallelogram is

area P (u1 , u2 ) ≡ |u1 × u2 | .
14.5. INTERPRETATION AND REVIEW 393

Suppose then that x = R (u) where u ∈ U, a subset of R2 and x is a point in V, a

subset of 3 dimensional space. Thus, letting the Cartesian coordinates of x be given by
T
x = (x1 , x2 , x3 ) , each xi being a function of u, an infinitesimal rectangle located at
u0 corresponds to n an infinitesimal parallelogram
o located at R (u0 ) which is determined
∂R(u0 ) ∂R(u0 )
by the 2 vectors ∂u du, ∂v dv , each of which is tangent to the surface defined
by x = R (u) . This is a very vague and unacceptable description. What exactly is
an infinitesimal rectangle? However, it can all be made precise later and this is good
motivation for the real thing.

Á f (dV )

du2 fu2 (u0 )du2

dV +
fu1 (u0 )du1 :
u0 du1

From Definition 14.5.6, the volume of this infinitesimal parallelepiped located at

R (u0 ) is given by
¯ ¯ ¯ ¯
¯ ∂R (u0 ) ∂R (u0 ) ¯¯ ¯ ∂R (u0 ) ∂R (u0 ) ¯¯
¯ ¯
¯ ∂u du × ∂v
dv ¯ = ¯
∂u
×
∂v ¯ dudv (14.35)

= |Ru × Rv | dudv (14.36)

This motivates the following definition of what is meant by the integral over a paramet-
rically defined surface in R3 .

Definition 14.5.7 Suppose U is a subset of R2 and suppose R : U → R (U ) =

3 1
S ⊆ R is a one to one and R C function. Then if h : R (U ) → R, define the 2
dimensional surface integral, R(U ) h (x) dS according to the following formula.
Z Z
h (x) dS ≡ h (R (u)) |Ru (u) × Rv (u)| dudv.
S U

With this understanding, it becomes possible to interpret the meaning of Stoke’s

theorem. This is stated in the following theorem. Note that slightly more is assumed
here than earlier. In particular, it is assumed that Ru × Rv 6= 0. This allows the
definition of a well defined normal vector which varies continuously over the surface, S.

Theorem 14.5.8 (Stoke’s Theorem) Let¡U be any 2

¢ region in R for which the
2 3
conclusion of Green’s theorem holds. Let R ∈ C Ui , R be a one to one function such
that Ru × Rv 6= 0 on U . Let
γ j = R ◦ αj ,

where the αj are parameterizations for the oriented bounded variation curves bounding
the region U oriented such that the conclusion of Green’s theorem holds. Let S denote
the surface,
S ≡ {R (u, v) : (u, v) ∈ U } ,
394 LINE INTEGRALS

Then for F a C 1 vector field defined near S,

n Z
X Z
F · dγ i = (Ru × Rv ) · (∇ × F) (R (u, v)) dm2 (14.37)
j=1 γj U
Z
= (∇ × F) · ndS (14.38)
S

Proof: Formula 14.37 was established in Theorem 14.4.4. The unit normal of the
point R (u, v) of S is (Ru × Rv ) / |Ru × Rv | and from the definition of the integral over
the surface, Definition 14.5.7, Formula 14.38 follows. ¥

14.6 Introduction To Complex Analysis

14.6.1 Basic Theorems, The Cauchy Riemann Equations
With Green’s theorem and the technique of proof used in proving it, it is possible to
present the most important parts of complex analysis almost effortlessly. 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 i2 = −1. The complex conjugate is defined by

x + iy ≡ x − iy

and for z a complex number,

1/2
p
|z| ≡ (zz) = x2 + y 2 .

Thus when x + iy is considered an ordered pair (x, y) ∈ R2 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,
q
2 2
|z − w| = (x − u) + (y − v)

so in terms of all topological considerations, R2 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 i2 = −1 makes C into a field. Now here is the definition of what it
means for a function to be analytic.

Definition 14.6.1 Let U be an open subset of C (R2 ) and let f : U → C be a

function. Then f is said to be analytic on U if for every z ∈ U,

f (z + ∆z) − f (z)
lim ≡ f 0 (z)
∆z→0 ∆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 395

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

ux (x, y) = vy (x, y) , uy (x, y) = −vx (x, y)

and all these partial derivatives, ux , uy , vx , vy 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
difference quotient as h → 0 in the definition. Thus from the definition,

f (z + ih) − f (z)
f 0 (z) ≡ lim
h→0 ih
u (x, y + h) + iv (x, y + h) − (u (x, y) + iv (x, y))
= lim
h→0 ih
1
= lim (uy (x, y) + ivy (x, y)) = −iuy (x, y) + vy (x, y)
h→0 i

Next let ∆z = h and take the limit of the difference quotient as h → 0.

f (z + h) − f (z)
f 0 (z) ≡ lim
h→0 h
u (x + h, y) + iv (x + h, y) − (u (x, y) + iv (x, y))
= lim
h→0 h
= ux (x, y) + ivx (x, y) .

Therefore, equating real and imaginary parts,

ux = vy , vx = −uy

and this yields the Cauchy Riemann equations. Since z → f 0 (z) is continuous, it follows
the real and imaginary parts of this function must also be continuous. Thus from the
above formulas for f 0 (z) , it follows from the continuity of z → f 0 (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))

= ux (x, y) h + uy (x, y) k + i (vx (x, y) h + vy (x, y) k) + o ((h, k))

= ux (x, y) h + uy (x, y) k + i (vx (x, y) h + vy (x, y) k) + o (∆z)
This follows from Theorem 6.5.1 which says that C 1 implies differentiable along with
the definition of the norm (absolute value) in C. By the Cauchy Riemann equations
this equals

= ux (x, y) h − vx (x, y) k + i (vx (x, y) h + ux (x, y) k) + o (∆z)

= ux (x, y) (h + ik) + ivx (x, y) (h + ik) + o (∆z)
= ux (x, y) ∆z + ivx (x, y) ∆z + o (∆z)

Dividing by ∆z and taking a limit yields f 0 (z) exists and equals ux (x, y) + ivx (x, y)
which are assumed to be continuous. This proves the proposition. ¥
396 LINE INTEGRALS

14.6.2 Contour Integrals

The most important tools in complex analysis are Cauchy’s theorem in some form and
Cauchy’s 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 definition.

Definition 14.6.3 Let γ : [a, b] → C be of bounded variation and let f : γ ∗ →

C. Letting P ≡ {t0 , · · ·, tn } where a = t0 < t1 < · · · < tn = b, define

||P|| ≡ max {|tj − tj−1 | : j = 1, · · ·, n}

and the Riemann Stieltjes sum by

n
X
S (P) ≡ f (γ (τ j )) (γ (tj ) − γ (tj−1 ))
j=1

where τ j ∈ [tj−1 , tj ] . (Note this notation is a little sloppy because it does not identify
R the
specific point, τ j used. It is understood that this point is arbitrary.) Define γ f (z) dz
as the unique number which satisfies the following condition. For all ε > 0 there exists
a δ > 0 such that if ||P|| ≤ δ, then
¯Z ¯
¯ ¯
¯ f (z) dz − S (P)¯ < ε.
¯ ¯
γ

Sometimes this is written as

Z
f (z) dz ≡ lim S (P) .
γ ||P||→0

You note that this is essentially the same definition given earlier for the line integral
only this time the function has values in C rather than Rn and there is no dot product
involved. Instead, you multiply by the complex number γ (tj ) − γ (tj−1 ) 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 γ (t) = γ 1 (t) + iγ 2 (t) . Then letting u be the real part of f and
v the imaginary part, S (P) equals
n
X
(u (γ 1 (τ j ) , γ 2 (τ j )) + iv (γ 1 (τ j ) , γ 2 (τ j )))
j=1
(γ 1 (tj ) − γ 1 (tj−1 ) + i (γ 2 (tj ) − γ 2 (tj−1 )))

n
X
= u (γ 1 (τ j ) , γ 2 (τ j )) (γ 1 (tj ) − γ 1 (tj−1 ))
j=1
n
X
− v (γ 1 (τ j ) , γ 2 (τ j )) (γ 2 (tj ) − γ 2 (tj−1 ))
j=1
Xn
+i v (γ 1 (τ j ) , γ 2 (τ j )) (γ 1 (tj ) − γ 1 (tj−1 ))
j=1
Xn
+i u (γ 1 (τ j ) , γ 2 (τ j )) (γ 2 (tj ) − γ 2 (tj−1 ))
j=1
14.6. INTRODUCTION TO COMPLEX ANALYSIS 397

Combining these leads to

n
X
(u (γ (τ j )) , −v (γ (τ j ))) · (γ (tj ) − γ (tj−1 ))
j=1
n
X
+i (v (γ (τ j )) , u (γ (τ j ))) · (γ (tj ) − γ (tj−1 )) (14.39)
j=1

Since the functions u and v are continuous, the limit as ||P|| → 0 of the above equals
Z Z
(u, −v) · dγ + i (v, u) · dγ
γ γ

¥
This proves most of the following lemma.

Lemma 14.6.4 Let Γ be a rectifiable curve in C having parameterization γ which is

continuous
R with bounded variation. Also let f : Γ → C be continuous. Then the contour
integral γ f (z) dz exists and is given by the sum of the following line integrals.
Z Z Z
f (z) dz = (u, −v) · dγ + i (v, u) · dγ (14.40)
γ γ γ

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 γ is C 1 , then
Z Z b
f (z) dz = f (γ (t)) γ 0 (t) dt.
γ a

Another important observation is the following.

Proposition 14.6.5 Suppose F 0 (z) = f (z) for all z ∈ Ω, an open set containing
γ ∗ where γ : [a, b] → C is a continuous bounded variation curve. Then
Z
f (z) dz = F (γ (b)) − F (γ (a)) .
γ

Proof: Letting u and v be real and imaginary parts of f, it follows from Lemma
14.6.4 Z Z Z
f (z) dz = (u, −v) · dγ + i (v, u) · dγ (14.41)
γ γ γ

Consider the real valued function

1³ ´
G (x, y) ≡ F (x + iy) + F (x + iy) ≡ Re F (x + iy)
2
By assumption,
F 0 (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
defined by
1 ³ ´
H (x, y) ≡ F (x + iy) − F (x + iy) ≡ Im F (x + iy)
2i
398 LINE INTEGRALS

Then ∇H = (v, u) and so from 14.41 and Theorem 14.2.11

Z
f (z) dz = G (γ (b)) − G (γ (a)) + i (H (γ (b)) − H (γ (a)))
γ
= F (γ (b)) − F (γ (a)) .

This proves the proposition. ¥

A function F such that F 0 = f is called a primitive of f. See how it acts a lot like a
potential, the difference 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 first form of the Cauchy integral theorem.

Lemma 14.6.6 Let U be an open set in C and let Γ be a simple closed rectifiable
curve contained in U having parameterization γ. Also let f be analytic in U . Then
Z
f (z) dz = 0.
γ

Proof: This follows right away from the Cauchy Riemann equations and the formula
14.40. Assume without loss of generality the orientation of Γ is the positive orientation.
If not, the argument is the same. Then from formula 14.40,
Z Z Z
f (z) dz = (u, −v) · dγ + i (v, u) · dγ
γ γ γ

and by Green’s theorem and Ui the inside of Γ this equals

Z Z
(−vx − uy ) dm2 + i (ux − vy ) dm2 = 0
Ui γ

by the Cauchy Riemann equations. This proves the lemma. ¥

It is easy to improve on this result using the argument for proving Green’s theorem.
You only need continuity on the bounding curve. You also don’t need to make any
assumption about the functions ux , etc. being in L1 (U ). The following is a very
general version of the Cauchy integral theorem.

Theorem 14.6.7 Let Ui be the inside of Γ a simple closed rectifiable curve

having parameterization γ. Also let f be analytic in Ui and continuous on Ui ∪ Γ. Then
Z
f (z) dz = 0.
γ

Proof: Let Bδ , Iδ be those regions of Lemma 14.3.6 where as earlier Iδ are those
which have empty intersection with Γ and Bδ are the border regions. Without loss of
generality, assume Γ is positively oriented. As in the proof of Green’s 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 Green’s theorem
X Z X Z Z
f (z) dz + f (z) dz = f (z) dz
R∈Iδ ∂R R∈Bδ ∂R γ
14.6. INTRODUCTION TO COMPLEX ANALYSIS 399

In this case the first sum on the left in the above formula equals 0 from Lemma 14.6.6
for any δ > 0. Now just as in the proof of Green’s theorem, you can choose δ small
enough that
X ¯¯Z ¯
¯
¯ f (z) dz ¯¯ < ε.
¯
R∈Bδ ∂R

Since ε 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,

Z
1 1
dw = 1
2πi ∂R w−z

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,
2
−1/ (w − z) .

This proves the lemma. ¥

Now with this little lemma, here is the Cauchy integral formula.

Theorem 14.6.9 Let Γ be a positively oriented simple closed rectifiable curve

having parameterization γ and let z ∈ Ui , the inside of Γ. Also let f be analytic on Ui ,
and continuous on Ui ∪ Γ. Then
Z
1 f (w)
f (z) = dw.
2πi γ w − z

In particular, letting f (z) ≡ 1,

Z
1 1
dw = 1.
2πi γ w−z

Proof: In constructing the special regions in the proof of Green’s theorem, always
choose δ such that the point z is not on any of the lines mδ = y and x = kδ. This makes
it possible to avoid thinking about the case where z is not on the interior of any of the
rectangles of Iδ . Pick δ small enough that Iδ 6= ∅ and z is contained in some R0 ∈ Iδ .
From Lemma 14.6.8 it follows for each R ∈ Iδ
Z Z
1 f (w) 1 f (w) − f (z)
dw − f (z) = dw
2πi R w − z 2πi R w−z
400 LINE INTEGRALS

Then as in the proof of Theorem 14.6.7

Z Z
1 X f (w) − f (z) 1 X f (w) − f (z)
dw + dw
2πi ∂R w−z 2πi ∂R w−z
R∈Iδ R∈Bδ
Z
1 f (w) − f (z)
= dw
2πi γ w−z
By Theorem 14.6.7, all these integrals on the left equal 0 except for R0 , the one which
contains z on its interior.
Thus the above reduces to
Z Z
1 f (w) − f (z) 1 f (w) − f (z)
dw = dw
2πi ∂R0 w−z 2πi γ w−z
The integrand of the left converges to f 0 (z) as δ → 0 and the length of R0 also converges
to 0 so it follows from Theorem 14.2.4 that the limit as δ → 0 in the above exists and
yields
Z Z Z
1 f (w) − f (z) 1 f (w) 1 1
0= dw = dw − f (z) dw. (14.42)
2πi γ w−z 2πi γ w − z 2πi γ w − z
Consider the last integral above.
X Z 1 X Z 1
Z
1
dw + dw = dw (14.43)
R∈Tδ ∂R w − z ∂R w − z
R∈Bδ γ w−z

As in the proof of Green’s theorem, choosing δ small enough the second sum on the left
in the above satisfies
¯ ¯
X ¯¯Z
¯X Z ¯ ¯
¯ 1 ¯ 1 ¯
¯ dw¯ ≤ ¯ dw ¯ < ε.
¯ ¯ ¯ ¯
R∈Bδ ∂R w − z ∂R w − z
R∈Bδ

By Lemma 14.6.8, the first sum on the left in 14.43 equals

Z
1
dw
∂R0 w − z

where R0 is the rectangle for which z is on the inside of ∂R0 . Then by this lemma
again, this equals 2πi. Therefore for such small δ, 14.43 reduces to
¯ Z ¯
¯ 1 ¯
¯2πi − dw ¯<ε
¯ ¯
γ w−z

Since ε is arbitrary, this shows

Z
1 1
dw = 1.
2πi γ w−z
Now using this, 14.42 implies the claimed formula of the theorem. This proves the
theorem. ¥
Theorem 14.6.10 Let γ : [a, b] → C be of bounded variation. Let f be contin-
uous on γ ∗ . For z ∈
/ γ ∗ , define
Z
f (w)
g (z) ≡ dw
γ w −z
Then g is infinitely differentiable. Futhermore,
Z
(n) f (w)
g (z) = n! n+1
γ (w − z)
14.6. INTRODUCTION TO COMPLEX ANALYSIS 401

Proof:
Z µ ¶
1 f (w) f (w)
(g (z + h) − g (z)) /h = − dw
h γ (w − z − h) (w − z)

Z µ ¶ Z µ ¶
1 h 1
= f (w) dw = f (w) dw
h γ (w − z − h) (w − z) γ (w − z − h) (w − z)
2
Consider only h ∈ C such that 2 |h| < dist (z, γ ∗ ). The integrand converges to 1/ (w − z) .
Then for these values of h,
¯ ¯ ¯ ¯
¯ 1 1 ¯ ¯ h ¯
¯ ¯ ¯ ¯
¯ − ¯ = ¯ ¯
¯ (w − z − h) (w − z) (w − z)2 ¯ ¯ (w − z − h) (w − z)2 ¯

|h| 2 |h|
≤ 3 = 3
dist (z, γ ∗ ) /2 dist (z, γ ∗ )

and so the convergence of the integrand to

2
f (w) / (w − z)

is uniform for |h| < dist (z, γ ∗ ) /2 . Using Theorem 14.2.4, it follows

g (z + h) − g (z)
g 0 (z) = lim
h→0 h
Z µ ¶
1 f (w) f (w)
= lim − dw
h→0 h γ (w − z − h) (w − z)
Z
f (w)
= 2 dw.
γ (w − z)

One can then differentiate the above expression using the same arguments. Contin-
uing this way results in the following formula.
Z
(n) f (w)
g (z) = n! n+1 dw
γ (w − z)

This proves the theorem. ¥

It turns out that in the definition of what it means for a function defined on an open
set, U to be analytic it is not necessary to say that z → f 0 (z) is continuous. In fact,
this comes for free. The statement that z → f 0 (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, z1 and z2 , you can consider γ (t) ≡ z1 + t (z2 − z1 ) for
t ∈ [0, 1] to obtain a continuous bounded variation curve from z1 to z2 . More generally,
if z1 , ···, zm are points in C you can obtain a continuous bounded variation curve from z1
to zm which consists of first going from z1 to z2 and then from z2 to z3 and so on, till in
the end one goes from zm−1 to zm . Denote this piecewise linear curve as γ (z1 , · · ·, zm ) .
Now let T be a triangle with vertices z1 , z2 and z3 encountered in the counter clockwise
402 LINE INTEGRALS

direction as shown.
z3

z1 z2
R R
Denote by ∂T
f (z) dz, the expression, γ(z1 ,z2 ,z3 ,z1 )
f (z) dz. Consider the following
picture.
z3
ª
T11 I T
- ¾
R T21 ª
ª
T31 T41 I
- Iµ -
z1 z2

Thus
Z 4 Z
X
f (z) dz = f (z) dz. (14.44)
∂T k=1 ∂Tk1

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 : Ω → C have the property that

f 0 (z) exists for all z ∈ Ω and let T be a triangle contained in Ω. Then
Z
f (w) dw = 0.
∂T

Proof: Suppose not. Then

¯Z ¯
¯ ¯
¯ f (w) dw¯¯ = α 6= 0.
¯
∂T

From 14.44 it follows ¯ ¯

4 ¯Z
X ¯
¯ ¯
α≤ ¯ f (w) dw¯
¯ ∂T 1 ¯
k=1 k

and so for at least one of these Tk1 , denoted from now on as T1 ,

¯Z ¯
¯ ¯ α
¯ f (w) dw ¯≥ .
¯ ¯ 4
∂T1

Now let T1 play the same role as T . Subdivide as in the above picture, and obtain T2
such that ¯Z ¯
¯ ¯ α
¯ f (w) dw¯¯ ≥ 2 .
¯ 4
∂T2

Continue in this way, obtaining a sequence of triangles,

Tk ⊇ Tk+1 , diam (Tk ) ≤ diam (T ) 2−k ,

14.6. INTRODUCTION TO COMPLEX ANALYSIS 403

and ¯Z ¯
¯ ¯ α
¯ f (w) dw¯¯ ≥ k .
¯ 4
∂Tk

Then let z ∈ ∩∞ 0
k=1 Tk and note that by assumption, f (z) exists. Therefore, for all k
large enough,
Z Z
f (w) dw = (f (z) + f 0 (z) (w − z) + g (w)) dw
∂Tk ∂Tk

where |g (w)| < ε |w − z| . Now observe that w → f (z) + f 0 (z) (w − z) has a primitive,
namely,
2
F (w) = f (z) w + f 0 (z) (w − z) /2.
Therefore, by Proposition 14.6.5,
Z Z
f (w) dw = g (w) dw.
∂Tk ∂Tk

From Theorem 14.2.4 applied to contour integrals or the definition of the contour inte-
gral,
¯Z ¯
α ¯ ¯
≤ ¯ ¯ g (w) dw¯¯ ≤ ε diam (Tk ) (length of ∂Tk )
4 k
∂Tk

≤ ε2−k (length of T ) diam (T ) 2−k ,

and so
α ≤ ε (length of T ) diam (T ) .
R
Since ε is arbitrary, this shows α = 0, a contradiction. Thus ∂T f (w) dw = 0 as
claimed. ¥
This fundamental result yields the following important theorem.

Theorem 14.6.12 (Morera1 ) Let Ω be an open set and let f 0 (z) exist for all
z ∈ Ω. Let D ≡ B (z0 , r) ⊆ Ω. Then there exists ε > 0 such that f has a primitive on
B (z0 , r + ε). (Recall this is a function F such that F 0 (z) = f (z) .)

Proof: Choose ε > 0 small enough that B (z0 , r + ε) ⊆ Ω. Then for w ∈ B (z0 , r + ε) ,
define Z
F (w) ≡ f (u) du.
γ(z0 ,w)

Then by the Cauchy Goursat theorem, Theorem 14.6.11, and w ∈ B (z0 , r + ε) , it

follows that for |h| small enough,
Z
F (w + h) − F (w) 1
= f (u) du
h h γ(w,w+h)
Z 1 Z 1
1
= f (w + th) hdt = f (w + th) dt
h 0 0

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 Morera’s theorem. It contains the proof that the condition of continuity of z → f 0 (z)
is redundant.
1 Giancinto Morera 1856-1909. This theorem or one like it dates from around 1886
404 LINE INTEGRALS

Corollary 14.6.13 Let Ω be an open set and suppose that whenever

γ (z1 , z2 , z3 , z1 )

is a closed curve bounding a triangle T, which is contained in Ω, and f is a continuous

function defined on Ω, it follows that
Z
f (z) dz = 0,
γ(z1 ,z2 ,z3 ,z1 )

then f is analytic on Ω. Also, if f 0 (z) exists for z ∈ Ω, then z → f 0 (z) is continuous.

Proof: As in the proof of Morera’s theorem, let B (z0 , r) ⊆ Ω and use the given
condition to construct a primitive, F for f on B (z0 , r) . (As just shown in Theorem
14.6.12, the given condition is satisfied whenever f 0 (z) exists for all z ∈ Ω.) Then F is
analytic and so by the Cauchy integral formula, for z ∈ B (z0 , r)
Z
1 F (w)
F (z) = dw.
2πi ∂B(z0 ,r) w − z

It follows from Theorem 14.6.10 that F and hence f have infinitely many derivatives,
implying that f is analytic on B (z0 , r) . Since z0 is arbitrary, this shows f is analytic on
Ω. In particular z → f 0 (z) is continuous because actually this function is differentiable.
This proves the corollary. ¥
This shows that an equivalent definition of what it means for a function to be analytic
is the following definition.

Definition 14.6.14 Let U be an open set in C and suppose f 0 (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 defining a simple curve the assumption was made that γ (t) 6= γ (a) and γ (t) 6=
γ (b) if t ∈ (a, b) . Is this fussy condition really necessary? Which theorems and
lemmas hold with simply assuming γ is one to one on (a, b)? Does the fussy
condition follow from assuming γ is one to one on (a, b)?
2
R
3. RShow that for many open sets R in R , Area of U = ∂U
xdy, and Area of U =
1
∂U
−ydx and Area of U = 2 ∂U −ydx + xdy. Hint: Use Green’s theorem.

4. A closed polygon in the plane starts at (x0 , y0 ) , goes to (x1 , y1 ) , to (x2 , y2 ) to

· · · (xn , yn ) = (x0 , y0 ) . Suppose the line segments never cross so that you have a
simple closed curve. Using Green’s theorem find 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 405

5. Let Γ be a simple C 1 oriented curve having parameterization γ where t is the time

and suppose f is a force defined on Γ. Then the work done by f on an object of
mass m as it moves over the curve is defined by
Z
f · dγ
γ

Newton’s second law states that f = m dv 0 0

dt where v ≡ γ (t). Let vb = γ (b) with
va defined similarly. Thus these are the final and initial velocities. Show the work
equals
1 2 1 2
m |vb | − m |va | .
2 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 Newton’s law it follows that for
γ (t) the motion,
1 2
−F (γ (t)) + m |γ 0 (t)|
2
is constant.
7. Generalize Stoke’s theorem, Theorem 14.4.4 to the case where R is only assumed
C 1.
8. Given an example of a simple closed rectifiable curve Γ and a horizontal line which
intersects this curve in infinitely many points.
9. Let Γ be a simple closed rectifiable curve and let Ui be its inside. Show you can
remove any finite number of circular disks from Ui and what remains will still
be a region for which Green’s 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 Γ is a simple closed rectifiable curve. Begin by punching
one circular hole and go from there.
10. Let γ : [a, b] → R be of bounded variation. Show there exist increasing functions
f (t) and g (t) such that
γ (t) = f (t) − g (t) .
Hint: You might let f (t) = V (γ; [a, t]) . Show this is increasing and then consider
g (t) = f (t) − γ (t) .
R
11. Using Problem 10 describe another way to obtain the integral γ f dγ for f a real
valued function and γ 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 γ has values in Rn and f : γ ∗ → Rn .
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 (z0 , r) ⊆ U, then
∞
X n
f (z) = an (z − z0 ) (14.45)
n=0
406 LINE INTEGRALS

for all |z − z0 | < r. Furthermore,

f (n) (z0 )
an = . (14.46)
n!
Hint: You use the Cauchy integral formula. For z ∈ B (z0 , r) and C the positively
oriented boundary,
Z Z
1 f (w) 1 f (w) 1
f (z) = = z−z0 dw
2πi C w − z 2πi C w − z0 1 − w−z 0
Z X ∞
1 f (w) n
= (z − z0 ) dw
2πi C n=0 (w − z0 )n+1

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
Ω be a connected open set (region) and let f : Ω → X be analytic. Then the
following are equivalent.

(a) f (z) = 0 for all z ∈ Ω

(b) There exists z0 ∈ Ω such that f (n) (z0 ) = 0 for all n.
(c) There exists z0 ∈ Ω which is a limit point of the set,

Z ≡ {z ∈ Ω : f (z) = 0} .

Hint: From Problem 13, if c.) holds, then for z near z0

X∞
f (n) (z0 ) n
f (z) = (z − z0 )
n=m
n!

Say f (n) (z0 ) 6= 0. Then consider

X∞
f (z) f (m) (z0 ) f (n) (z0 ) n−m
m = + (z − z0 )
(z − z0 ) m! n=m+1
n!

Now let zn → z0 , zn 6= z0 but f (zn ) = 0. What does this say about f (m) (z0 )?
Clearly the first two conditions are equivalent and they imply the third.

15. You want to define ez 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 ex
when z = x + i0. Then show using the Cauchy Riemann equations that

ez ≡ ex (cos (y) + i sin (y))

is analytic and agrees with ex when z = x + i0. Also show

d z
e = ez .
dz
Hint: For the first part, suppose two functions, f, g work. Then consider f − g.
this is analytic and has a zero set, R.
14.7. EXERCISES 407

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 shouldn’t 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 ez to get the definition.
17. Show that if f is analytic on C and f 0 (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 Liouville’s theorem which states that a function
which is analytic on all of C which is also bounded is constant. Hint: By that
theorem, Z
0 1 f (w)
f (z) = dw
2πi Cr (w − z)2
where Cr 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 Cr is 2πr.
18. Using Problem 15 prove the fundamental theorem of algebra which says every
nonconstant polynomial having complex coefficients 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 Ui , the inside of Γ, a rectifiable simple closed curve positively
oriented with parameterization γ. Suppose also there are no zeros of f on Γ.
Show then that the number of zeros, of f contained in Ui counted according to
multiplicity is given by the formula
Z 0
1 f (z)
dz
2πi γ f (z)
Qm
Hint: You ought to first show f (z) = k=1 (z − zk ) g (z) where the zk are the
zeros of f in Ui and g (z) is an analytic function which never vanishes in Ui ∪ Γ. 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 z0 ∈ U
∗
called a star center such that the for all z ∈ U, γ (z0 , z) as described in before
the proof of the Cauchy Goursat theorem is contained in U . For example, pick
any complex number α and consider everything left after leaving out the ray
{tα : t ≥ 0} . Show this is star shaped with a star center tα for t < 0. Now for U
a star shaped open connected set, suppose g is analytic on U and g (z) 6= 0 for all
z ∈ U. Show there exists an analytic function h defined on U such that

eh(z) = g (z) .

This function h (z) is like log (g (z)). Hint: Use an argument like that used to
prove Morera’s theorem and the Cauchy Goursat theorem to obtain a primitive
for g 0 /g, h1 . Next consider the function

ge−h1
d
¡ −h ¢
Using the chain rule and the product rule, show dz ge 1 = 0. Using one of the
results of Problem 17 show
g = ceh1
for some constant c. Tell why c can be written as ea+ib . Then let h = h1 + a + ib.
408 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 z0 ∈ U, there exists an open set, V containing z0 and
m ∈ N such that for all z ∈ V,
m
f (z) = f (z0 ) + φ (z) (14.47)

where φ : V → B (0, δ) is one to one, analytic and onto, φ (z0 ) = 0, φ0 (z) 6= 0 on

V and φ−1 analytic on B (0, δ) . If f is one to one then m = 1 for each z0 and
f −1 : f (U ) → U is analytic. Consider the real valued function f (x) = x2 . 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 (z0 , r) \ {z0 } ,

f (z) − f (z0 ) 6= 0.

Explain why there exists g (z) analytic and nonzero on B (z0 , r) such that for some
positive integer m,
m
f (z) − f (z0 ) = (z − z0 ) g (z)
Next one tries to take the mth root of g (z) . Using Problem 20 there exists h
analytic such that ³ ´m
g (z) = eh(z) , g (z) = eh(z)/m

Now let φ (z) = (z − z0 ) eh(z)/m . This yields the formula 14.47. Also φ0 (z0 ) =
eh(z0 )/m 6= 0. Now consider

φ (x + iy) = u (x, y) + iv (x, y)

and the map µ ¶ µ ¶

x u (x, y)
→
y v (x, y)
Here u, v are C 1 because φ is analytic. Use the Cauchy Riemann equations to
verify the Jacobian of this transformation at (x0 , y0 ) is nonzero. This is where
you use φ0 (z0 ) 6= 0. Use inverse function theorem to verify φ maps some open
set V containing z0 one to one and onto B (0, δ) . Thus also φm maps V onto
B (0, δ m ). Explain why it follows from 14.47 and the fact that z0 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 can’t be one to one. To verify this,

ei2π/m φ (z1 ) 6= φ (z1 )

but both are in B (0, δ). Hence there exists z2 6= z1 such that φ (z2 ) = ei2π/m φ (z1 )
(φ is one to one) but f (z2 ) = f (z1 ). 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 0 (z) =
eh(z)/1 6= 0. Hence

¡ ¢0 f −1 (f (z1 )) − f −1 (f (z))
f −1 (f (z)) = lim
f (z1 )→f (z) f (z1 ) − f (z)
z1 − z 1
= lim = 0
z1 →z f (z1 ) − f (z) f (z)
14.7. EXERCISES 409

22. Let U be what is left when you leave out the ray tα 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 = eh(z) by Problem 20. Explain why h (z) is analytic on U. When
α = −1 this is called the principle branch of the logarithm. In this case define
Arg (z) ≡ θ ∈ (−π, π) such that the given z equals |z| eiθ . 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 don’t have to fuss with any tedium in order to show this.

23. Suppose Γ is a simple closed curve and let Ui be the inside. Suppose f is analytic
on Ui and continuous on Ui ∪ Γ. Consider the function z → |f (z)|. This is a
continuous function. Show that if it achieves its maximum at any point of Ui then
f must be a constant. Hint: You might use the open mapping theorem.

24. Let f, g be analytic on Ui , the inside of Γ, a rectifiable simple closed curve positively
oriented with parameterization γ. Suppose either

|f (z) + g (z)| < |f (z)| + |g (z)| on Γ

or
|f (z) − g (z)| < |f (z)| on Γ
Let Zf denote the number of zeros in Ui and let Zg denote the number of zeros of
g in Ui . Then neither f , g, nor f /g can equal zero anywhere on Γ and Zf = Zg .
Hint: The first condition implies for all z ∈ Γ,

f (z)
∈ C \ [0, ∞)
g (z)

Show there exists a primitive F for

0
(f /g)
.
f /g

and argue
Z 0 Z Z
(f /g) f0 g0
0= dz = dz − dz = Zf − Zg .
γ f /g γ g0 γ g
You could consider F = L (f /g) where L is the analytic function defined on
C \ [0, ∞) with the property that

eL(z) = z.

Thus
eL(z) L0 (z) = 1, L0 (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 Rouche’s 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, an z n + an−1 z n−1 + · · · +
a1 z + a0 where an 6= 0. Now let f (z) = an z n . Let Γ 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.
410 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 Rouche’s 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 z0 ∈ U there exists r > 0 such that
m
f (z) − f (z0 ) = g (z) (z − z0 )

where g is analytic and nonzero on B (z0 , r). Now consider the function z →
f (z) − w. I would like to use Rouche’s theorem to claim this function has the
same number of zeros, namely m as the function z → f (z) − f (z0 ). Let

δ = min {|f (z) − f (z0 )| : |z − z0 | = r}

Then if |w − f (z0 )| < δ,

|w − f (z0 )| = |f (z) − f (z0 ) − (f (z) − w)| < δ ≤ |f (z) − f (z0 )|

for each z ∈ ∂B (z0 , r) and so you can apply Rouche’s theorem. What does this
say about when f is one to one? Why is f (U ) open? Why is f (U ) connected?
/ γ∗.
28. Let γ : [a, b] → C be of bounded variation, γ (a) = γ (b) and suppose z ∈
Define Z
1 dw
n (γ, z) ≡ .
2πi γ w − z
This is called the winding number. When γ ∗ 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 (γ, z) is continuous and so is constant on
every component of C \ γ ∗ . For z in the unbounded component, n (γ, 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 η : [a, b] → C which is C 1 on [a, b] and

max {|η (t) − γ (t)| : t ∈ [a, b]} < ε,

η (a) = η (b) = γ (a) = γ (b)

¯ Z Z ¯
¯ 1 dw 1 dw ¯¯
¯ − <ε
¯ 2πi w−z 2πi w − z¯
γ η
∗ ∗
where ε < dist (z, γ ) . Thus z ∈
/ η . Consider the contour integral which involves
η and show it is an integer. Then there exists a sequence of these C 1 contours
{η k } such that ¯ ¯
¯ 1 Z dw 1
Z
dw ¯¯
¯
¯ − ¯ → 0.
¯ 2πi γ w − z 2πi ηk w − z ¯
Consequently, for all k large enough there can be no change in
Z
1 dw
2πi ηk w − z
14.7. EXERCISES 411

which shows Z
1 dw
2πi γ w−z
is an integer as claimed. So how do you show the contour integral involving η
yields an integer? As mentioned above,
Z Z b
1 dw 1 η 0 (t)
= dt
2πi ηk w−z 2πi a η (t) − z

Let Z t
η 0 (s)
g (t) ≡ ds
a η (s) − z
Formally this is a lot like some sort of log (η (s) − z) (recall beginning calculus) so
it is reasonable to consider µ g(t) ¶0
e
.
η (t) − z
Show this equals 0. Explain why this requires the function which is differentiated
must be constant. Thus
eg(a) eg(b)
=
η (a) − z η (b) − z
Now η (a) = η (b) , g (a) = 0, and so eg(a) = 1 = eg(b) . Explain why this requires
g (b) = 2mπi for m an integer. Now this gives the desired result.

29. Let
B 0 (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 0 (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 0 (a, r) → X be analytic. Thus f has an iso-

lated singularity at a. Suppose also that

lim f (z) (z − a) = 0.
z→a

Then there exists a unique analytic function, g : B (a, r) → X such that g = f on

B 0 (a, r) . Thus the singularity at a is removable.

2
Prove this theorem. Hint: Let h (z) = f (z) (z − a) . Then h (a) = 0 and h0 (a)
exists and equals 0. Show this. Also h is analytic near a. Therefore,
∞
X k
h (z) = ak (z − a)
k=2

2
Maybe consider g (z) = h (z) / (z − a) . Argue g is analytic and equals f for z
near a.

30. Another really amazing theorem in complex analysis is the Casorati Weierstrass
theorem.
412 LINE INTEGRALS

Theorem 14.7.2 Let a be an isolated singularity and suppose for some

r > 0, f (B 0 (a, r)) is not dense in C. Then either a is a removable singularity
or there exist finitely many b1 , · · ·, bM for some finite number, M such that for z
near a,
XM
b−k
f (z) = g (z) + k
(14.48)
k=1 (z − a)

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 (z0 , δ) has no points
of f (B 0 (a, r)) . Such a ball must exist if f (B 0 (a, r)) is not dense in the plane.
this means that for all 0 < |z − a| < r,

|f (z) − z0 | ≥ δ > 0

Hence
1
lim (z − a) = 0
z→a f (z) − z0
and so 1/ (f (z) − z0 ) has a removable sincularity at a. See Problem 29. Let g (z)
be analytic at and near a and agree with this function. Thus
∞
X n
g (z) = an (z − a) .
n=0

There are two cases, g (a) = 0 and g (a) 6= 0. First suppose g (a) = 0. Then explain
why
m
g (z) = h (z) (z − a)
where h (z) is analytic and non zero near a. Then

1 1
f (z) − z0 =
h (z) (z − a)m

Show this yields the desired conclusion. Next suppose g (a) 6= 0. Then explain why
g (z) 6= 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 defined as b−1 in
14.48. Suppose a is a pole and Γ is a simple closed rectifiable curve containing a
on the inside with no other singular points on Γ or anywhere else inside Γ. Show
that under these conditions,
Z
f (z) dz = 2πi (res (f, a))
Γ

m
Also describe a way to find resR(f, a) by multiplying by (z − a) and differentiat-
1
ing. Hint: You should show Γ (z−a) m dz = 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 Γ be a positively oriented rectifiable simple closed curve with inside Ui
and remove finitely many open discs B (zj , rj ) from Ui . Thus the result is a holy
14.7. EXERCISES 413

region. Suppose f is analytic on some open set containing Ui \ ∪nj=1 B (zj , rj ) .

Then letting Γj denote the negatively oriented boundary of B (zj , rj ) , show
Z n Z
X
0= f (z) dz + f (z) dz
γ j=1 γj

where γ j is a parameterization for Γj . Hint: The proof is the same as given

earlier. You just use Green’s theorem.
33. Let Γ be a simple closed curve and suppose on its inside there are finitely many
n
poles for a function f which is analytic near Γ. Call these poles {zk }k=1 . Then
Z n
X
f (z) dz = 2πi res (f, zj )
γ j=1

This is the very important residue theorem for computing line integrals. Hint:
You should use Problem 32 and Problem 30, the Casorati Weierstrass theorem.
414 LINE INTEGRALS
Hausdorff Measures And Area
Formula

15.1 Definition Of Hausdorff Measures

This chapter is on Hausdorff measures. First I will discuss some outer measures. In all
that is done here, α (n) will be the volume of the ball in Rn which has radius 1. This
volume is the usual Lebesgue measure and the balls will be determined by the usual
norm on Rn .

Definition 15.1.1 For a set, E, denote by r (E) the number which is half the
diameter of E. Thus

1 1
r (E) ≡ sup {|x − y| : x, y ∈ E} ≡ diam (E)
2 2

Let E ⊆ Rn .
∞
X
Hδs (E) ≡ inf{ β(s)(r (Cj ))s : E ⊆ ∪∞
j=1 Cj , r (Cj ) ≤ δ}
j=1

Hs (E) ≡ lim Hδs (E).

δ→0

In the above definition, β (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
n
in such a way that whenever n is a positive integer, Hn ([0, 1) ) = 1 = mn ([0, 1)n ) . In
fact, this is all you need to know about it.

Lemma 15.1.2 Hs and Hδs are outer measures.

Proof: It is clear that Hs (∅) = 0 and if A ⊆ B, then Hs (A) ≤ Hs (B) with similar
assertions valid for Hδs . Suppose E = ∪∞ s i ∞
i=1 Ei and Hδ (Ei ) < ∞ for each i. Let {Cj }j=1
be a covering of Ei with

∞
X
β(s)(r(Cji ))s − ε/2i < Hδs (Ei )
j=1

415
416 HAUSDORFF MEASURES AND AREA FORMULA

and diam(Cji ) ≤ δ. Then

∞ X
X ∞
Hδs (E) ≤ β(s)(r(Cji ))s
i=1 j=1
∞
X
≤ Hδs (Ei ) + ε/2i
i=1
∞
X
≤ ε+ Hδs (Ei ).
i=1

It follows that since ε > 0 is arbitrary,

∞
X
Hδs (E) ≤ Hδs (Ei )
i=1

which shows Hδs is an outer measure. Now notice that Hδs (E) is increasing as δ → 0.
Picking a sequence δ k decreasing to 0, the monotone convergence theorem implies
∞
X
Hs (E) ≤ Hs (Ei ).
i=1

This proves the lemma. ¥

The outer measure Hs is called s dimensional Hausdorff measure when restricted to
the σ algebra of Hs measurable sets. Recall these are the sets, E such that for all S,

Hs (S) = Hs (S ∩ E) + Hs (S \ E) .

Next I will show the σ algebra of Hs measurable sets includes the Borel sets. This
is done by the following very interesting condition known as Caratheodory’s criterion.

15.1.1 Properties Of Hausdorff Measure

Definition 15.1.3 For two sets, A, B in a metric space, define

dist (A, B) ≡ inf {||x − y|| : x ∈ A, y ∈ B} .

Theorem 15.1.4 Let µ be an outer measure on the subsets of X, a closed subset

of a normed vector space and suppose

µ(A ∪ B) = µ(A) + µ(B)

whenever dist(A, B) > 0, then the σ algebra of measurable sets contains the Borel sets.

Proof: It suffices to show that closed sets are in F, the σ-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 Ω. Is µ(S) ≥ µ(S ∩ K) + µ(S \ K)? It
suffices to assume µ(S) < ∞. Let
1
Kn ≡ {x : dist(x, K) ≤ }
n
By Lemma 7.4.4 on Page 159, x → dist (x, K) is continuous and so Kn is closed. By
the assumption of the theorem,

µ(S) ≥ µ((S ∩ K) ∪ (S \ Kn )) = µ(S ∩ K) + µ(S \ Kn ) (15.1)

15.1. DEFINITION OF HAUSDORFF MEASURES 417

since S ∩ K and S \ Kn are a positive distance apart. Now

µ(S \ Kn ) ≤ µ(S \ K) ≤ µ(S \ Kn ) + µ((Kn \ K) ∩ S). (15.2)

If limn→∞ µ((Kn \ K) ∩ S) = 0 then the theorem will be proved because this limit
along with 15.2 implies limn→∞ µ (S \ Kn ) = µ (S \ K) and then taking a limit in 15.1,
µ(S) ≥ µ(S ∩ K) + µ(S \ K) as desired. Therefore, it suffices to establish this limit.
Since K is closed, a point, x ∈
/ K must be at a positive distance from K and so

Kn \ K = ∪∞
k=n Kk \ Kk+1 .

Therefore
∞
X
µ(S ∩ (Kn \ K)) ≤ µ(S ∩ (Kk \ Kk+1 )). (15.3)
k=n

If
∞
X
µ(S ∩ (Kk \ Kk+1 )) < ∞, (15.4)
k=1

then µ(S ∩ (Kn \ K)) → 0 because it is dominated by the tail of a convergent series so
it suffices to show 15.4.
XM
µ(S ∩ (Kk \ Kk+1 )) =
k=1
X X
µ(S ∩ (Kk \ Kk+1 )) + µ(S ∩ (Kk \ Kk+1 )). (15.5)
k even, k≤M k odd, k≤M

By the construction, the distance between any pair of sets, S ∩ (Kk \ Kk+1 ) for different
even values of k is positive and the distance between any pair of sets, S ∩ (Kk \ Kk+1 )
for different odd values of k is positive. Therefore,
X X
µ(S ∩ (Kk \ Kk+1 )) + µ(S ∩ (Kk \ Kk+1 )) ≤
k even, k≤M k odd, k≤M

[ [
µ( S ∩ (Kk \ Kk+1 )) + µ( S ∩ (Kk \ Kk+1 )) ≤ 2µ (S) < ∞
k even k odd
PM
and so for all M, k=1 µ(S ∩ (Kk \ Kk+1 )) ≤ 2µ (S) showing 15.4 and proving the
theorem. ¥
The next theorem applies the Caratheodory criterion above to Hs .

Theorem 15.1.5 The σ algebra of Hs measurable sets contains the Borel sets
and H has the property that for all E ⊆ Rn , there exists a Borel set F ⊇ E such that
s

Hs (F ) = Hs (E).

Proof: Let dist(A, B) = 2δ 0 > 0. Is it the case that

Hs (A) + Hs (B) = Hs (A ∪ B)?

This is what is needed to use Caratheodory’s criterion.

Let {Cj }∞
j=1 be a covering of A ∪ B such that r(Cj ) ≤ δ < δ 0 /2 for each j and

∞
X
Hδs (A ∪ B) + ε > β(s)(r (Cj ))s.
j=1
418 HAUSDORFF MEASURES AND AREA FORMULA

Thus X X
Hδs (A ∪ B )˙ + ε > β(s)(r (Cj ))s + β(s)(r (Cj ))s
j∈J1 j∈J2

where
J1 = {j : Cj ∩ A 6= ∅}, J2 = {j : Cj ∩ B 6= ∅}.
Recall dist(A, B) = 2δ 0 and so J1 ∩ J2 = ∅. It follows

Hδs (A ∪ B) + ε > Hδs (A) + Hδs (B).

Letting δ → 0, and noting ε > 0 was arbitrary, yields

Hs (A ∪ B) ≥ Hs (A) + Hs (B).

Equality holds because Hs is an outer measure. By Caratheodory’s criterion, Hs is a

Borel measure.
To verify the second assertion, note first there is no loss of generality in letting
Hs (E) < ∞. Let
E ⊆ ∪∞j=1 Cj , r(Cj ) < δ,

and
∞
X
Hδs (E) + δ > β(s)(r (Cj ))s.
j=1

Let
Fδ = ∪∞
j=1 Cj .

Thus Fδ ⊇ E and
∞
X ¡ ¢
Hδs (E) ≤ Hδs (Fδ ) ≤ β(s)(r Cj )s
j=1
∞
X
= β(s)(r (Cj ))s < δ + Hδs (E).
j=1

Let δ k → 0 and let F = ∩∞

k=1 Fδ k . Then F ⊇ E and

Hδsk (E) ≤ Hδsk (F ) ≤ Hδsk (Fδ ) ≤ δ k + Hδsk (E).

Letting k → ∞,
Hs (E) ≤ Hs (F ) ≤ Hs (E)
This proves the theorem. ¥
A measure satisfying the first conclusion of Theorem 15.1.5 is sometimes called a
Borel regular measure.

15.1.2 Hn And mn
Next I will compare Hn and mn . 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 ⊆ Rn 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, {Bj }∞ ∞
j=1 , such that mn (E \ ∪j=1 Bj ) = 0.

In the next lemma, the balls are the usual balls taken with respect to the usual
distance in Rn .
15.1. DEFINITION OF HAUSDORFF MEASURES 419

Lemma 15.1.7 If mn (S) = 0 then Hn (S) = Hδn (S) = 0. Also, there exists a
constant, k such that Hn (E) ≤ kmn (E) for all E Borel. Also, if Q0 ≡ [0, 1)n , the unit
cube, then Hn ([0, 1)n ) > 0.

Proof: Suppose first mn (S) = 0. First suppose S is bounded. Then by outer

regularity, there exists a bounded open V containing S ³and´mn (V ) < ε. For each
x ∈ S, there exists a ball Bx such that Bcx ⊆ V and δ > r B cx . By the Vitali covering
n o
theorem there is a sequence of disjoint balls {Bk } such that Bck covers S. Then letting
α (n) be the Lebesgue measure of the unit ball in Rn
X ³ ´n X
Hδn (S) ≤ ck = β (n) 5n
β (n) r B α (n) r (Bk )
n
α (n)
k k
β (n) n β (n) n
≤ 5 mn (V ) < 5 ε
α (n) α (n)
Since ε is arbitrary, this shows Hδn (S) = 0 and now it follows Hn (S) = 0. In case S is not
bounded, let Sm = B (0,m) ∩ S. Then Hδn (Sm ) = 0 and so letting m → ∞, Hδn (S) = 0
also. Then as before, Hn (S) = 0.
Letting U be an open set and δ > 0, consider all balls, B contained in U which
have diameters less than δ. This is a Vitali covering of U and therefore by Theorem
15.1.6, there exists {Bi } , a sequence of disjoint balls of radii less than δ contained in
U such that ∪∞ i=1 Bi differs from U by a set of Lebesgue measure zero. Let α (n) be the
Lebesgue measure of the unit ball in Rn . Then from what was just shown,
∞
X ∞
n β (n) X n
Hδn (U ) = Hδn (∪i Bi ) ≤ β (n) r (Bi ) = α (n) r (Bi )
i=1
α (n) i=1
∞
β (n) X β (n)
= mn (Bi ) = mn (U ) ≡ kmn (U ) .
α (n) i=1 α (n)

Now letting E be Borel, it follows from the outer regularity of mn there exists a de-
creasing sequence of open sets, {Vi } containing E such such that mn (Vi ) → mn (E) .
Then from the above,

Hδn (E) ≤ lim Hδn (Vi ) ≤ lim kmn (Vi ) = kmn (E) .
i→∞ i→∞

Since δ > 0 is arbitrary, it follows that also

Hn (E) ≤ kmn (E) .

This proves the first part of the lemma.

To verify the second part, note that it is obvious Hδn and Hn are translation invariant
because diameters of sets do not change when translated. Therefore, if Hn ([0, 1)n ) = 0,
it follows Hn (Rn ) = 0 because Rn is the countable union of translates of Q0 ≡ [0, 1)n .
Since each Hδn is no larger than Hn , the same must hold for Hδn . Therefore, there exists
a sequence of sets, {Ci } each having diameter less than δ such that the union of these
sets equals Rn but
∞
X n
1> β (n) r (Ci ) .
i=1

Now let Bi be a ball having radius equal to diam (Ci ) = 2r (Ci ) which contains Ci . It
follows
n α (n) 2n n
mn (Bi ) = α (n) 2n r (Ci ) = β (n) r (Ci )
β (n)
420 HAUSDORFF MEASURES AND AREA FORMULA

which implies
∞
X X∞
n β (n)
1> β (n) r (Ci ) = mn (Bi ) = ∞,
i=1 i=1
α (n) 2n
a contradiction. This proves the lemma. ¥

Lemma 15.1.8 Every open set in Rn is the countable disjoint union of half open
boxes of the form
n
Y
(ai , ai + 2−k ]
i=1
−k
where ai = l2 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
Y
[ai , ai + 2−k )
i=1

and the conclusion would be unchanged.

Proof: Let
n
Y
Ck = {All half open boxes (ai , ai + 2−k ] where
i=1

ai = l2−k for some integer l.}

Thus Ck consists of a countable disjoint collection of boxes whose union is Rn . This is
sometimes called a tiling of Rn . Think of tiles on the floor 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
n
Y
x, y ∈ (ai , ai + 2−k ],
i=1

then |xi − yi | ≤ 2−k . Therefore,

à n
!1/2
X ¡ ¢ √
−k 2
|x − y| ≤ 2 = 2−k n.
i=1

Let U be open and let B1 ≡ all sets of C1 which are contained in U . If B1 , · · ·, Bk have
been chosen, Bk+1 ≡ all sets of Ck+1 contained in
¡ ¢
U \ ∪ ∪ki=1 Bi .

Let B∞ = ∪∞ i=1 Bi . In fact ∪B∞ = U . Clearly ∪B∞ ⊆ U because every box of every Bi
is contained in U . If p ∈ U , let k be the smallest integer such that p is contained in a
box from Ck which is also a subset of U . Thus

p ∈ ∪Bk ⊆ ∪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 β (n) properly, one can obtain Hn = mn on all

Lebesgue measurable sets.
15.2. TECHNICAL CONSIDERATIONS∗ 421

Proof: I will show Hn is a positive multiple of mn for any choice of β (n) . Define
mn (Q0 )
k=
Hn (Q0 )
where Q0 = [0, 1)n is the half open unit cube in Rn . I will show kHn (E) = mn (E) for
any Lebesgue measurable set. When this is done, it will follow that by adjusting β (n)
the multiple can
Qn be taken to be 1.
Let Q¡ =¢ i=1 [ai , ai + 2−k ) be a half open box where ai = l2−k . Thus Q0 is the
n
union of 2k of these identical half open boxes. By translation invariance, of Hn and
mn
¡ k ¢n n 1 1 ¡ k ¢n
2 H (Q) = Hn (Q0 ) = mn (Q0 ) = 2 mn (Q) .
k k
Therefore, kHn (Q) = mn (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
kHn (U ) = mn (U ) for all open sets. It follows immediately, since every compact set
is the countable intersection of open sets that kHn = mn 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 Hn measurable and that kHn (F ) = mn (F ). Let Fl = 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 ⊆ Fl ⊆ G (15.6)
and mn (G \ H) = 0 which implies by Lemma 15.1.7

mn (G \ H) = kHn (G \ H) = 0. (15.7)

To do this, let {Gi } be a decreasing sequence of bounded open sets containing Fl and
let {Hi } be an increasing sequence of compact sets contained in Fl such that

kHn (Gi \ Hi ) = mn (Gi \ Hi ) < 2−i

Then letting G = ∩i Gi and H = ∪i Hi this establishes 15.6 and 15.7. Then by com-
pleteness of Hn it follows Fl is Hn measurable and

kHn (Fl ) = kHn (H) = mn (H) = mn (Fl ) .

Now taking l → ∞, it follows F is Hn measurable and kHn (F ) = mn (F ). Therefore,

adjusting β (n) it can be assumed the constant, k is 1. This proves the theorem. ¥
The exact determination of β (n) is more technical. You can skip it if you want.
Just remember β (n) is chosen such that Hn ([0, 1)n ) = 1. It turns out this will require
β (n) = α (n) where α (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 α(n) be the volume of the unit ball in Rn . Thus the volume of B(0, r) in Rn is
α(n)rn 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 Rn ,
then
m(A) ≤ α(n)(2−1 diam(A))n.
This inequality may seem obvious at first 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.
422 HAUSDORFF MEASURES AND AREA FORMULA

Lemma 15.2.1 Let f : Rn−1 → [0, ∞) be Borel measurable and let

S = {(x,y) :|y| < f (x)}.

Then S is a Borel set in Rn .

Proof: Set sk be an increasing sequence of Borel measurable functions converging

pointwise to f .
Nk
X
sk (x) = ckm XEm
k (x).

m=1

Let
Sk = ∪N k k k
m=1 Em × (−cm , cm ).
k

Then (x,y) ∈ Sk if and only if f (x) > 0 and |y| < sk (x) ≤ f (x). It follows that
Sk ⊆ Sk+1 and
S = ∪∞
k=1 Sk .

But each Sk is a Borel set and so S is also a Borel set. This proves the lemma. ¥
Let Pi be the projection onto

span (e1 , · · ·, ei−1 , ei+1 , · · ·, en )

n
where the ek are the standard basis vectors
P in R , ek being the vector having a 1 in the
th
k slot and a 0 elsewhere. Thus Pi x ≡ j6=i xj ej . Also let

APi x ≡ {xi : (x1 , · · ·, xi , · · ·, xn ) ∈ A}

APi x x

Pi x ∈ span{e1 , · · ·, ei−1 ei+1 , · · ·, en }.

Lemma 15.2.2 Let A ⊆ Rn be a Borel set. Then Pi x → m(APi x ) is a Borel

measurable function defined on Pi (Rn ).
Qn
Proof: Let K be the π system consisting of sets of the form j=1 Aj where Ai is
Borel. Also let G denote those Borel sets of Rn such that if A ∈ G then

Pi x → m((A ∩ Rk )Pi x ) is Borel measurable.

where Rk = (−k, k)n . Thus K ∈ G. If A ∈ G

³¡ ¢ ´
Pi x → m AC ∩ Rk Pi x

is Borel measurable because it is of the form

¡ ¢ ¡ ¢
m (Rk )Pi x − m (A ∩ Rk )Pi x

and these are Borel measurable functions of Pi x. Also, if {Ai } is a disjoint sequence of
sets in G then ¡ ¢ X ¡ ¢
m (∪i Ai ∩ Rk )Pi x = m (Ai ∩ Rk )Pi x
i
15.2. TECHNICAL CONSIDERATIONS∗ 423

and each function of Pi x is Borel measurable. Thus by the lemma on π systems G =

B (Rn ) and This proves the lemma. ¥
Now let A ⊆ Rn be Borel. Let Pi be the projection onto
¡ ¢
span e1 , · · ·, ei−1 , ei+1 , · · ·, en

and as just described,

APi x = {y ∈ R : Pi x + yei ∈ A}
Thus for x = (x1 , · · ·, xn ),

APi x = {y ∈ R : (x1 , · · ·, xi−1 , y, xi+1 , · · ·, xn ) ∈ A}.

Since A is Borel, it follows from Lemma 15.2.1 that

Pi x → m(APi x )

is a Borel measurable function on Pi Rn = Rn−1.

15.2.1 Steiner Symmetrization∗

Define
S(A, ei ) ≡ {x =Pi x + yei : |y| < 2−1 m(APi x )}

Lemma 15.2.3 Let A be a Borel subset of Rn . Then S(A, ei ) satisfies

Pi x + yei ∈ S(A, ei ) if and only if Pi x − yei ∈ S(A, ei ),

S(A, ei ) is a Borel set in Rn,

mn (S(A, ei )) = mn (A), (15.8)
diam(S(A, ei )) ≤ diam(A). (15.9)

Proof : The first assertion is obvious from the definition. The Borel measurability
of S(A, ei ) follows from the definition and Lemmas 15.2.2 and 15.2.1. To show Formula
15.8,
Z Z 2−1 m(APi x )
mn (S(A, ei )) = dxi dx1 · · · dxi−1 dxi+1 · · · dxn
P i Rn −2−1 m(APi x )
Z
= m(APi x )dx1 · · · dxi−1 dxi+1 · · · dxn
P i Rn
= m(A).

Now suppose x1 and x2 ∈ S(A, ei )

x1 = Pi x1 + y1 ei , x2 = Pi x2 + y2 ei .

For x ∈ A define
l(x) = sup{y : Pi x+yei ∈ A}.
g(x) = inf{y : Pi x+yei ∈ A}.
Then it is clear that
l(x1 ) − g(x1 ) ≥ m(APi x1 ) ≥ 2|y1 |, (15.10)
l(x2 ) − g(x2 ) ≥ m(APi x2 ) ≥ 2|y2 |. (15.11)
Claim: |y1 − y2 | ≤ |l(x1 ) − g(x2 )| or |y1 − y2 | ≤ |l(x2 ) − g(x1 )|.
424 HAUSDORFF MEASURES AND AREA FORMULA

Proof of Claim: If not,

2|y1 − y2 | > |l(x1 ) − g(x2 )| + |l(x2 ) − g(x1 )|

≥ |l(x1 ) − g(x1 ) + l(x2 ) − g(x2 )|

= l(x1 ) − g(x1 ) + l(x2 ) − g(x2 ).
≥ 2 |y1 | + 2 |y2 |
by 15.10 and 15.11 contradicting the triangle inequality.
Now suppose |y1 − y2 | ≤ |l(x1 ) − g(x2 )|. From the claim,

|x1 − x2 | = (|Pi x1 − Pi x2 |2 + |y1 − y2 |2 )1/2

≤ (|Pi x1 − Pi x2 |2 + |l(x1 ) − g(x2 )|2 )1/2
≤ (|Pi x1 − Pi x2 |2 + (|z1 − z2 | + 2ε)2 )1/2
√
≤ diam(A) + O( ε)

where z1 and z2 are such that Pi x1 + z1 ei ∈ A, Pi x2 + z2 ei ∈ A, and

|z1 − l(x1 )| < ε and |z2 − g(x2 )| < ε.

If |y1 − y2 | ≤ |l(x2 ) − g(x1 )|, then we use the same argument but let

|z1 − g(x1 )| < ε and |z2 − l(x2 )| < ε,

Since x1 , x2 are arbitrary elements of S(A, ei ) and ε 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, ei ).

Lemma 15.2.4 Suppose A is a Borel set in Rn such that Pj x + ej xj ∈ A if and

only if Pj x+(−xj )ej ∈ A. Then if i 6= j, Pj x + ej xj ∈ S(A, ei ) if and only if
Pj x+(−xj )ej ∈ S(A, ei ).

Proof : By definition,
Pj x + ej xj ∈ S(A, ei )
if and only if
|xi | < 2−1 m(APi (Pj x+ej xj ) ).
Now
xi ∈ APi (Pj x+ej xj )
if and only if
xi ∈ APi (Pj x+(−xj )ej )
by the assumption on A which says that A is symmetric in the ej direction. Hence

Pj x + ej xj ∈ S(A, ei )

if and only if
|xi | < 2−1 m(APi (Pj x+(−xj )ej ) )
if and only if
Pj x+(−xj )ej ∈ S(A, ei ).
This proves the lemma. ¥
15.2. TECHNICAL CONSIDERATIONS∗ 425

15.2.2 The Isodiametric Inequality∗

The next theorem is called the isodiametric inequality. It is the key result used to
compare Lebesgue and Hausdorff measures.

Theorem 15.2.5 Let A be any Lebesgue measurable set in Rn . Then

mn (A) ≤ α(n)(r (A))n.

Proof: Suppose first that A is Borel. Let A1 = S(A, e1 ) and let Ak = S(Ak−1 , ek ).
Then by the preceding lemmas, An is a Borel set, diam(An ) ≤ diam(A), mn (An ) =
mn (A), and An is symmetric. Thus x ∈ An if and only if −x ∈ An . It follows that

An ⊆ B(0, r (An )).

(If x ∈ An \B(0, r (An )), then −x ∈ An \B(0, r (An )) and so diam (An ) ≥ 2|x| > diam(An ).)
Therefore,
mn (An ) ≤ α(n)(r (An ))n ≤ α(n)(r (A))n .
It remains to establish this inequality for arbitrary measurable sets. Letting A be such
a set, let {Kn } be an increasing sequence of compact subsets of A such that

m(A) = lim m(Kk ).

k→∞

Then

m(A) = lim m(Kk ) ≤ lim sup α(n)(r (Kk ))n

k→∞ k→∞
n
≤ α(n)(r (A)) .

This proves the theorem. ¥

15.2.3 The Proper Value Of β (n)∗

I will show that the proper determination of β (n) is α (n), the volume of the unit ball.
Since β (n) has been adjusted such that k = 1, mn (B (0, 1)) = Hn (B (0, 1)). There
∞
exists a covering of B (0,1) of sets of radii less than δ, {Ci }i=1 such that
X n
Hδn (B (0, 1)) + ε > β (n) r (Ci )
i

Then by Theorem 15.2.5, the isodiametric inequality,

X n β (n) X ¡ ¢n
Hδn (B (0, 1)) + ε > β (n) r (Ci ) = α (n) r C i
i
α (n) i
β (n) X ¡ ¢ β (n) β (n) n
≥ mn C i ≥ mn (B (0, 1)) = H (B (0, 1))
α (n) i α (n) α (n)

Now taking the limit as δ → 0,

β (n) n
Hn (B (0, 1)) + ε ≥ H (B (0, 1))
α (n)

and since ε > 0 is arbitrary, this shows α (n) ≥ β (n).

426 HAUSDORFF MEASURES AND AREA FORMULA

By the Vitali covering theorem, there exists a sequence of disjoint balls, {Bi } such
that B (0, 1) = (∪∞ n
i=1 Bi ) ∪ N where mn (N ) = 0. Then Hδ (N ) = 0 can be concluded
because Hδ ≤ H and Lemma 15.1.7. Using mn (B (0, 1)) = Hn (B (0, 1)) again,
n n

∞
X n
Hδn (B (0, 1)) = Hδn (∪i Bi ) ≤ β (n) r (Bi )
i=1
∞ ∞
β (n) X n β (n) X
= α (n) r (Bi ) = mn (Bi )
α (n) i=1
α (n) i=1
β (n) β (n) β (n) n
= mn (∪i Bi ) = mn (B (0, 1)) = H (B (0, 1))
α (n) α (n) α (n)

which implies α (n) ≤ β (n) and so the two are equal. This proves that if α (n) = β (n) ,
then the Hn = mn on the measurable sets of Rn .
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 Hs is not a Radon measure because it will not generally
be finite on compact sets. For example, let n = 2 and consider H1 (L) where L is a
line segment joining (0, 0) to (1, 0). Then H1 (L) is no smaller than H1 (L) when L is
considered a subset of R1 , n = 1. Thus by what was just shown, H1 (L) ≥ 1. Hence
H1 ([0, 1] × [0, 1]) = ∞. The situation is this: L is a one-dimensional object inside R2
and H1 is giving a one-dimensional measure of this object. In fact, Hausdorff measures
can make such heuristic remarks as these precise. Define the Hausdorff dimension of a
set, A, as
dim(A) = inf{s : Hs (A) = 0}

15.2.4 A Formula For α (n)∗

What is α(n)? Recall the gamma function which makes sense for all p > 0.
Z ∞
Γ (p) ≡ e−t tp−1 dt.
0

Lemma 15.2.6 The following identities hold.

pΓ(p) = Γ(p + 1),

µZ 1 ¶
p−1 q−1
Γ(p)Γ(q) = x (1 − x) dx Γ(p + q),
0

µ ¶
1 √
Γ = π
2

Proof: Using integration by parts,

Z ∞ Z ∞
Γ (p + 1) = e−t tp dt = −e−t tp |∞
0 +p e−t tp−1 dt
0 0
= pΓ (p)
15.2. TECHNICAL CONSIDERATIONS∗ 427

Next
Z ∞ Z ∞
−t p−1
Γ (p) Γ (q) = e t dt e−s sq−1 ds
Z0 ∞ Z ∞ 0

= e−(t+s) tp−1 sq−1 dtds

Z0 ∞ Z0 ∞
p−1 q−1
= e−u (u − s) s duds
0 s
Z ∞ Z u
p−1 q−1
= e−u (u − s) s dsdu
0 0
Z ∞ Z 1
p−1 q−1
= e−u (u − ux) (ux) udxdu
0 0
Z ∞ Z 1
p−1
= e−u up+q−1 (1 − x) xq−1 dxdu
0 0
µZ 1 ¶
p−1 q−1
= Γ (p + q) x (1 − x) dx .
0
¡1¢
It remains to find Γ 2 .
µ ¶ Z ∞ Z ∞ Z ∞
1 −t −1/2 −u2 1 2
Γ = e t dt = e 2udu = 2 e−u du
2 0 0 u 0

Now
µZ ∞ ¶2 Z ∞ Z ∞ Z ∞ Z ∞
e−(x +y 2 )
2 2 2 2
e−x dx = e−x dx e−y dy = dxdy
0 0 0 0 0
Z ∞ Z π/2
2 1
= e−r rdθdr = π
0 0 4
and so µ ¶ Z ∞
1 2 √
Γ =2 e−u du = π
2 0

This proves the lemma. ¥

Next let n be a positive integer.

Theorem 15.2.7 α(n) = π n/2 (Γ(n/2+1))−1 where Γ(s) is the gamma function
Z ∞
Γ(s) = e−t ts−1 dt.
0

Proof: First let n = 1.

µ ¶ √
3 1 1 π
Γ( ) = Γ = .
2 2 2 2

Thus
2 √
π 1/2 (Γ(1/2 + 1))−1 = √ π = 2 = α (1) .
π
and this shows the theorem is true if n = 1.
Assume the theorem is true for n and let Bn+1 be the unit ball in Rn+1 . Then by
the result in Rn , Z 1
mn+1 (Bn+1 ) = α(n)(1 − x2n+1 )n/2 dxn+1
−1
428 HAUSDORFF MEASURES AND AREA FORMULA

Z 1
= 2α(n) (1 − t2 )n/2 dt.
0

Doing an integration by parts and using Lemma 15.2.6

Z 1
= 2α(n)n t2 (1 − t2 )(n−2)/2 dt
0
Z
1 1 1/2
= 2α(n)n u (1 − u)n/2−1 du
2 0
Z 1
= nα(n) u3/2−1 (1 − u)n/2−1 du
0
= nα(n)Γ(3/2)Γ(n/2)(Γ((n + 3)/2))−1
= nπ n/2 (Γ(n/2 + 1))−1 (Γ((n + 3)/2))−1 Γ(3/2)Γ(n/2)
= nπ n/2 (Γ(n/2)(n/2))−1 (Γ((n + 1)/2 + 1))−1 Γ(3/2)Γ(n/2)
= 2π n/2 Γ(3/2)(Γ((n + 1)/2 + 1))−1
= π (n+1)/2 (Γ((n + 1)/2 + 1))−1 .

This proves the theorem. ¥

From now on, in the definition of Hausdorff measure, it will always be the case that
β (s) = α (s) . As shown above, this is the right thing to have β (s) to equal if s is a
positive integer because this yields the important result that Hausdorff measure is the
same as Lebesgue measure. Note the formula, π s/2 (Γ(s/2 + 1))−1 makes sense for any
s ≥ 0.

15.3 Hausdorff Measure And Linear Transformations

Hausdorff measure makes possible a unified development of n dimensional area including
in one theory length and surface area. Imagine the boundary of an open set in R3 . You
would tend to think of this as something two dimensional. The way to measure it is
with H2 . Length can be measured by H1 and the boundary of an open set in R4 is
measured in terms of H3 etc.
As in the case of Lebesgue measure, the first step in this is to understand
¡ ¢ basic
considerations related to linear transformations. Recall that for L ∈ L Rk , Rl , L∗ is
defined 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 Hausdorff
measure while the one which preserves lengths does not change Hausdorff measure.
These ideas are behind the following theorems and lemmas.

Lemma 15.3.1 Let R ∈ L(Rn , Rm ), n ≤ m, and R∗ R = I. Then if A ⊆ Rn ,

Hn (RA) = Hn (A).

In fact, if P : Rn → Rm satisfies |P x − P y| = |x − y| , then

Hn (P A) = Hn (A) .
15.3. HAUSDORFF MEASURE AND LINEAR TRANSFORMATIONS 429

Proof: Note that

2 2
|R(x − y)| = (R (x − y) , R (x − y)) = (R∗ R (x − y) , x − y) = |x − y|

Thus R preserves lengths.

Now let P be an arbitrary mapping which preserves lengths and let A be bounded,
P (A) ⊆ ∪∞
j=1 Cj , r(Cj ) < δ, and

∞
X
Hδn (P A) + ε > α(n)(r(Cj ))n .
j=1

Since P preserves lengths, it follows P is one to one on P (Rn ) and P −1 also preserves
lengths on P (Rn ) . Replacing each Cj with Cj ∩ (P A),
∞
X
Hδn (P A) + ε > α(n)r(Cj ∩ (P A))n
j=1
∞
X ¡ ¢n
= α(n)r P −1 (Cj ∩ (P A))
j=1
≥ Hδn (A).

Thus Hδn (P A) ≥ Hδn (A).

Now let A ⊆ ∪∞ j=1 Cj , diam(Cj ) ≤ δ, and

∞
X n
Hδn (A) + ε ≥ α(n) (r (Cj ))
j=1

Then
∞
X n
Hδn (A) + ε ≥ α(n) (r (Cj ))
j=1
X∞
n
= α(n) (r (P Cj ))
j=1
≥ Hδn (P A).

Hence Hδn (P A) = Hδn (A). Letting δ → 0 yields the desired conclusion in the case where
A is bounded. For the general case, let Ar = A ∩ B (0, r). Then Hn (P Ar ) = Hn (Ar ).
Now let r → ∞. This proves the lemma. ¥

Lemma 15.3.2 Let F ∈ L(Rn , Rm ), n ≤ m, and let F = RU where R and U are

described in Theorem 3.9.3 on Page 66. Then if A ⊆ Rn is Lebesgue measurable,

Hn (F A) = det(U )mn (A).

Proof: Using Theorem 9.8.7 on Page 235 and Theorem 15.1.9,

Hn (F A) = Hn (RU A)

= Hn (U A) = mn (U A) = det(U )mn (A).

Definition 15.3.3 Define J to equal det(U ). Thus

J = det((F ∗ F )1/2 ) = (det(F ∗ F ))1/2 .

430 HAUSDORFF MEASURES AND AREA FORMULA

15.4 The Area Formula

15.4.1 Preliminary Results
It was shown in Lemma 15.3.2 that for F ∈ L (Rn , Rm ) , m ≥ n

Hn (F A) = det(U )mn (A)

where F = RU with R preserving distances and U ∈ L (Rn , Rn ) 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 Rm 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 R3 or even in R8 and it can all be done
in a unified 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 Rn to Rm . 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 Rn and mn (T ) = 0, then

Hn (h (T )) = 0. If h is one to one, V and h (V ) are both bounded, N ⊆ h (V ) , Hn (N ) =
−1
0, and for Dh (x)
¡ −1= R (x)¢ U (x) the right polar decomposition, U (x) exists for all
x ∈ V, then mn h (N ) = 0.

Proof: Let
Tk ≡ {x ∈ T : ||Dh (x)|| < k} .
Thus T = ∪k Tk . I will show h (Tk ) has Hn measure zero and then it will follow that

h (T ) = ∪∞
k=1 h (Tk )

must also have measure zero.

Let ε > 0 be given. By outer regularity, there exists an open set, W , containing
Tk which is contained in V such that mn (W ) < knε6n . For x ∈ Tk it follows from
differentiability,
h (x + v) = h (x) + Dh (x) v + o (v)
and so whenever rx is small enough, B (x,5rx ) ⊆ W and whenever |v| < 5rx , |o (v)| <
krx . Therefore, if |v| < 5rx ,

Dh (x) v + o (v) ∈ B (0, 5krx ) + B (0,krx ) ⊆ B (0, 6krx )

and so
h (B (x, 5rx )) ⊆ B (h (x) , 6krx ).
Letting δ > 0 be given, the Vitali covering theorem implies there exists a sequence of
disjoint balls {Bi },n Bio= B (xi , rxi ), which are contained in W such that the sequence
of enlarged balls, B bi , having the same center but 5 times the radius, covers Tk and
6krxi < δ. Then ³ ³ ´´
Hδn (h (Tk )) ≤ Hδn h ∪∞
i=1
b
B i

∞
X ³ ³ ´´ X∞
≤ Hδn b
h Bi ≤ Hδn (B (h (xi ) , 6krxi ))
i=1 i=1
15.4. THE AREA FORMULA 431

∞
X ∞
X
n n
≤ α (n) (6krxi ) = (6k) α (n) rxni
i=1 i=1
∞
X
n
= (6k) mn (B (xi , rxi ))
i=1
n n ε
≤ (6k) mn (W ) ≤ (6k) = ε.
k n 6n
Since ε > 0 is arbitrary, this shows Hδn (h (Tk )) = 0. Since δ is arbitrary, this implies
Hn (h (Tk )) = 0. Now

Hn (h (T )) = lim Hn (h (Tk )) = 0.
k→∞

It remains to verify the last claim. Recall

n
X
U (x) = ak vk vk
k=1

−1
where {vk } is an orthonormal basis and since U (x) is given to exist, each ak > 0.
2
Therefore, (U (x) w, w) ≥ δ (x) |w| where δ (x) > 0.
−1
Next I claim h is continuous on h (V ) . Suppose then that h (xk ) → h (x) . Then
let the sequentially compact set B (x, r) ⊆ V. Without loss of generality all xk may be
assumed to lie in B (x, r). If {xk } fails to converge to x, then since B (x, r) is sequentially
compact, there exists a subsequence {xkl } converging to z 6= x. But then

h (x) = lim h (xkl ) = h (z)

l→∞

a contradiction to h being one to one. Thus xk → x and so h is continuous.

For η > 0 let © ¡ ¢ ª
Nη ≡ y ∈ N : δ h−1 (y) ≥ η .
Then for y ∈ Nη ,
¡ ¢ ¡ ¢
(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,
¯ ¡ −1 ¢¯ ¯ ¯
¯o h (y + v) − h−1 (y) ¯ ≤ η ¯h−1 (y + v) − h−1 (y)¯
2
Taking the inner product of both sides of 15.13 with h−1 (y + v) − h−1 (y) yields
¯ ∗ ¡ −1 ¢ ¯¯ ¯
¯R h (y) v¯ ¯h−1 (y + v) − h−1 (y)¯
¯ ¯2 η ¯ ¯2
≥ η ¯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|

432 HAUSDORFF MEASURES AND AREA FORMULA

and so
|R∗ v| ≤ |v| . (15.14)
Thus the above formula implies
η ¯¯ −1 ¯
|v| ≥ h (y + v) − h−1 (y)¯ . (15.15)
2
Since Nη has Hn measure zero, there exist {Ck } covering Nη such that r (Ci ) < δ and
∞
X n
εη n /4n > α (n) r (Ck ) .
k=1

Without
© loss ofªgenerality each Ck has nonempty intersection with Nη , containing yk .
Now h−1 (Ck ) covers h−1 (Nη ) and from 15.15

¡ ¢ 2
diam h−1 (Ck ) ≤ × 2 × diam (Ck )
η

and so
X µ ¶n
n
¡ −1 ¢ 2 n
H4δ/η h (Nη ) ≤ α (n) (diam (Ck ) )
η
k
4n X n εη n 4n
≤ α (n) r (C k ) < =ε
ηn 4n η n
k

n
¡ ¢
Since ε is arbitrary, H4δ/η h−1 (Nη ) = 0 and letting δ → 0 yields
¡ ¢ ¡ ¢
Hn h−1 (Nη ) = mn h−1 (Nη ) = 0

because by Theorem 15.1.9 Lebesgue and Hausdorff measure are the same on the
Lebesgue measurable sets of Rn . Now take η k → 0
¡ ¢ ¡ ¡ ¢¢
mn h−1 (N ) = lim mn h−1 Nηk = 0.
k→∞

This proves the lemma. ¥

Lemma 15.4.2 If S is a Lebesgue measurable subset of an open set V ⊆ Rn , on

which h is defined and C 1 , then h (S) is Hn measurable.

Proof: Let Sk = 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
mn (T ) = 0 such that
F ∪ T = Sk .
Then h (F ) ⊆ h (Sk ) ⊆ 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, Hn (h (T )) = 0 and so h (Sk ) is Hn
measurable because of completeness of Hausdorff measure, which comes from Hn being
obtained from an outer measure. Now h (S) = ∪∞ k=1 h (Sk ) and so it is also true that
h (S) is Hn measurable. This proves the lemma. ¥
The following lemma, depending on the Brouwer fixed 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 Rk to Rk doesn’t move any point very much,
then the image of the ball must contain a slightly smaller ball.
15.4. THE AREA FORMULA 433

Lemma 15.4.3 Let B = B (0, r), a ball in Rk and let F : B → Rk be continuous

and suppose for some ε < 1,
|F (v) −v| < εr (15.16)
for all v ∈ B. Then
F (B) ⊇ B (0, r (1 − ε)) .
Proof: Suppose a ∈ B (0, r (1 − ε)) \ F (B) .
I claim that a 6= 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 − ε) = rε,
a contradiction to 15.16.
Now letting G :B → B, be defined by
r (a − F (v))
G (v) ≡ ,
|a − F (v)|
it follows from what was just shown G is continuous. Then by the Brouwer fixed point
theorem, G (v) = v for some v ∈ B. Using the formula for G, it follows
¯ ¯
¯ r (a − F (v)) ¯
|v| = |G (v)| = ¯¯ ¯=r
|a − F (v)| ¯
Taking the inner product with v,
2 r
(G (v) , v) = |v| = r2 = (a − F (v) , v)
|a − F (v)|
r
= (a − v + v − F (v) , v)
|a − F (v)|
r
= [(a − v, v) + (v − F (v) , v)]
|a − F (v)|
r h i
2
= (a, v) − |v| + (v − F (v) , v)
|a − F (v)|
r £ 2 ¤
≤ r (1 − ε) − r2 +r2 ε = 0,
|a − F (v)|
a contradiction to |v| = r. Therefore, B (0, r (1 − ε)) \ F (B) = ∅ and This proves the
lemma. ¥
Lemma 15.4.4 If |P x − P y| ≤ L |x − y| , then for E a set,
Hn (P E) ≤ Ln Hn (E) .
∞
Proof: Without loss of generality, assume Hn (E) < ∞. Let δ > 0 and let {Ci }i=1
be a covering of E such that r (Ci ) ≤ δ for each i and
∞
X n
α (n) r (Ci ) ≤ Hδn (E) + ε.
i=1
∞
Then {P Ci }i=1 is a covering of P E such that r (P Ci ) ≤ Lδ. Therefore,
∞
X
n n
HLδ (P E) ≤ α (n) r (P Ci )
i=1
∞
X n
≤ Ln α (n) r (Ci ) ≤ Ln Hδn (E) + Ln ε
i=1
≤ L H (E) + Ln ε.
n n
434 HAUSDORFF MEASURES AND AREA FORMULA

Letting δ → 0,
Hn (P E) ≤ Ln Hn (E) + Ln ε

and since ε > 0 is arbitrary, This proves the lemma. ¥

Then the following corollary follows from 15.14.

Corollary 15.4.5 Let R : Rn → Rm where m ≥ n and R is linear and preserves

distances. Let T ⊆ Rm . Then

Hn (T ) ≥ Hn (RR∗ T ) = Hn (R∗ T ) .

Now let V be an open set in Rn and let h : V → Rm be a C 1 (V ) function. For

−1
Dh (x) = R (x) U (x) the right polar decomposition, suppose U (x) exists for all
x ∈ U. By definition 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 ε > 0 be given,

¯ −1 ∗ ¯
¯U R (h (x + v) − h (x)) − v¯ < ε |v|

whenever v is small enough. Thus by Lemma 15.4.3, if rx is small enough,

U −1 R∗ (h (x+B (0,rx )) − h (x)) ⊇ B (0, rx (1 − ε))

which implies
h (x+B (0,rx )) ⊇ RU B (0, rx (1 − ε)) + h (x)

or in other words,

h (B (x,rx )) ⊇ Dh (x) B (0, rx (1 − ε)) + 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 rx is sufficiently small,

R∗ (x) (h (x+B (0, rx )) − h (x)) ⊆ U (B (0,rx ) + B (0, εrx ))

⊆ U B (0,rx (1 + ε))

and so
h (x+B (0, rx )) =

h (B (x,rx )) ⊆ Dh (x) B (0,rx (1 + ε)) + h (x) (15.19)

15.5 The Area Formula

The following lemma is the first approximation to the area formula.
15.5. THE AREA FORMULA 435

Lemma 15.5.1 Let V be a bounded open set in Rn and let h ∈ C 1 (V ) , h : V → Rm

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 Hn (h (V )) < ∞ and h (V ) is bounded. Let W be a Borel set in Rm and

let A be a Lebesgue measurable subset of V . Then
Z Z
XW (y) dHn = XW (h (x)) |det (U (x))| dmn
h(A) A

Proof: Recall ¡ ¢1/2

∗
det (U (x)) = det Dh (x) Dh (x)
and so the function, x → det (U (x)) is continuous. ¡ ¡ ¢¢
Also let¡ O¡be an
¡ open set containing
¢¢¢ h−1 (W )∩A such that mn O \ h−1 (W ) ∩ A <
ε and Hn h O \ h−1 (W ) ∩ A < ε. To do this, let {Om } be¡ a decreasing¢sequence
of open sets containing h−1 (W ) ∩ A such that mn (Om ) → mn h−1 (W ) ∩ A . Thus
¡ ¢
mn (∩m Om ) \ h−1 (W ) ∩ A = 0

and so ¡ ¡ ¢¢
Hn h (∩m Om ) \ h−1 (W ) ∩ A = 0
by Lemma 15.4.1. Therefore, if m is large enough, letting O = Om gives the desired
open set.
Let x ∈ h−1 (W ) ∩ A. First note h−1 (W ) is a Borel set because
© ª
S ≡ E ∈ B (Rm ) : h−1 (E) ∈ B (Rn )

is a σ algebra which contains the open sets due to the fact h is continuous. Therefore,
S = B (Rm ). Thus h−1 (W ) ∩ A is measurable.
There exists 1 > rx > 0 small enough that 15.19 and 15.18 both hold. There exists
a possibly smaller rx such that
B (x, rx ) ⊆ O (15.20)
and
||det (U (x1 ))| − |det (U (x))|| < ε (15.21)
whenever x1 ∈ B (x, rx ) .
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 {Bi } such that for

N ≡ h−1 (W ) ∩ A \ ∪∞
i=1 Bi ,

mn (N ) = 0. Therefore, renaming N to equal

h−1 (W ) ∩ A \ ∪∞
i=1 Bi ∩ h
−1
(W ) ∩ A,

mn (N ) = 0 and by Lemma 15.4.1,

Hn (h (N )) = 0.
¡ ¢
h−1 (W ) ∩ A = ∪∞i=1 Bi ∩ A ∩ h
−1
(W ) ∪ N
436 HAUSDORFF MEASURES AND AREA FORMULA

W ∩ h (A) = (∪∞
i=1 h (Bi ∩ A) ∩ W ) ∪ h (N ) (15.22)

where mn (N ) = Hn (h (N )) = 0.
Denote by xi the center of Bi and ri the radius. Using 15.18, Lemma 15.4.1 which
says h takes sets of Lebesgue measure zero to sets of Hn measure zero, the translation
invariance of Hn , Lemma 15.3.2 which gives the rule for taking a linear transformation
outside the Hausdorff measure of something, 15.21, and the assumption that h is one
to one,

R R R
XW (y) dHn = h(A)∩W dHn = ∪∞ h(Bi ∩A)∩W dHn
h(A) i=1
R R
= h(∪∞ Bi ∩A∩h−1 (W )) dH ≥ h(∪∞ Bi ) dHn − ε ≥
n
i=1 i=1
P∞ n
P∞ n
i=1 H (h (Bi )) − ε ≥ i=1 H (Dh (xi ) (B (0, (1 − ε) ri ))) − ε
P∞
= i=1 |det (U (xi ))| mn (B (0, (1 − ε) ri )) − ε
n P∞
= (1 − ε) i=1 |det (U (xi ))| mn (B (xi ,ri )) − ε
³ R ´
n P∞
≥ (1 − ε) i=1 Bi
|det (U (x))| dm n − εm n (B i ) −ε

n P∞ R n
≥ (1 − ε) i=1 Bi ∩A∩h−1 (W ) |det (U (x))| dmn − (1 − ε) εmn (V ) − ε
nR n
= (1 − ε) V X∪∞ i=1 h(Bi ∩A∩h
−1 (W )) (h (x)) |det (U (x))| dmn − (1 − ε) εmn (V ) − ε

nR n
= (1 − ε) V XW ∩h(A) (h (x)) |det (U (x))| dmn − (1 − ε) εmn (V )
nR
− (1 − ε) V XN (x) |det (U (x))| dmn − ε
nR n
= (1 − ε) A XW (h (x)) |det (U (x))| dmn − (1 − ε) εmn (V ) − ε

The last three lines follows from 15.22. Recall mn (N ) = 0. Since ε > 0 is arbitrary,
this shows
Z Z
XW (y) dHn ≥ XW (h (x)) |det (Dh (x))| dmn
h(A) A

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 + ε) instead of
(1 − ε) , much as was done in the proof of Lemma 9.9.1. Thus

R R P∞ R
h(A)
XW (y) dHn = h(A)∩W dHn = i=1 h(Bi ∩A)∩W dHn
P∞ P∞
= i=1 Hn (h (Bi ∩ A) ∩ W ) ≤ i=1 Hn (h (Bi ))
P∞
≤ i=1 Hn (Dh (xi ) (B (0, (1 + ε) ri )))
P∞
= i=1 |det (U (xi ))| mn (B (0, (1 + ε) ri ))
n P∞
= (1 + ε) i=1 |det (U (xi ))| mn (B (xi ,ri ))
P ³R ´
n ∞
≤ (1 + ε) i=1 Bi
|det (U (x))| dmn + εmn (Bi )
15.5. THE AREA FORMULA 437

n P∞ R n
≤ (1 + ε) i=1 Bi ∩A∩h −1 (W ) |det (U (x))| dmn + (1 + ε) εmn (V )
nR
+ (1 + ε) O\(A∩h−1 (W )) |det (U (x))| dmn
n R
≤ (1 + ε) V
X∪∞
i=1 h(Bi ∩A)∩W
(h (x)) |det (U (x))| dmn
n n
+ (1 + ε) εmn (V ) + ε (1 + ε) M
n R n
= (1 + ε) XW ∩h(A) (h (x)) |det (U (x))| dmn + (1 + ε) εmn (V )
V
n nR
+ε (1 + ε) M − (1 + ε) V XN (x) |det (U (x))| dmn
n R n n
= (1 + ε) A
XW (h (x)) |det (U (x))| dmn + (1 + ε) εmn (V ) + ε (1 + ε) M
Since ε is arbitrary, This proves the lemma. ¥
Next the Borel sets will be enlarged to Hn measurable sets.

Lemma 15.5.2 Let V be a bounded open set in Rn and let h ∈ C 1 (V ) , h : V → Rm

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 Hn (h (V )) < ∞ and h (V ) is bounded. Let W be a Hn measurable set in

Rm and let A be a Lebesgue measurable subset of V . Then
Z Z
XW (y) dHn = XW (h (x)) |det (U (x))| dmn (15.23)
h(A) A

Proof: By Theorem 15.1.5 there exists

¡ a Borel set, F containing
¢ W such that
Hn (F \ W ) = 0. By Lemma 15.4.1 mn h−1 ((F \ W ) ∩ h (A)) = 0. Therefore, from
Lemma 15.5.1
Z Z
XW (y) dHn = XF (y) dHn
h(A) h(A)
Z
= XF (h (x)) |det (U (x))| dmn
A
Z
= XF \W (h (x)) |det (U (x))| dmn
A
Z
+ XW (h (x)) |det (U (x))| dmn
Z A
= XW (h (x)) |det (U (x))| dmn .
A

Note that

XW (h (x)) |det (U (x))| + XF \W (h (x)) |det (U (x))| = XF (h (x)) |det (U (x))|

and the second term on the left equals Xh−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 XF ◦ h is measurable because it is a Borel measurable
function composed with a continuous function (why?). Therefore, x →XW (h (x)) |det (U (x))|
is also measurable. This proves the lemma. ¥
You don’t need to assume the open sets are bounded and you don’t need to assume
a bound on |det U (x)|.
438 HAUSDORFF MEASURES AND AREA FORMULA

Corollary 15.5.3 Let V be an open set in Rn 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 Hn measurable and let A ⊆ V be Lebesgue measurable. Then
Z Z
n
XE (y) dH = XE (h (x)) |det (U (x))| dmn .
h(A) A

Proof: For each x ∈ A, there exists rx such that B (x, rx ) ⊆ V and rx < 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, rx ), it follows h (B (x, rx )) is also bounded. Also |det U (x)| is
bounded on the compact set B (x, rx ). These balls are a Vitali cover of A. By Corollary
∞
9.7.5 there is a sequence of these disjoint balls {Bi } such that mn (A \ ∪i=1 Bi ) = 0 and
so
A = ∪∞ i=1 (Bi ∩ A) ∪ N

where N = A \ ∪∞ i=1 Bi 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 Bi ,
Z XZ
XE (y) dHn = XE∩h(Bi ∩A) (y) dHn
h(A) i h(Bi ∩A)
XZ
= XE (h (x)) |det (U (x))| dmn
i Bi ∩A
Z
= XE (h (x)) |det (U (x))| dmn .
A

This proves the corollary. ¥

With this corollary, the main theorem follows.

Theorem 15.5.4 Let V be an open set in Rn and A is a Lebesgue measurable

1
subset of V. Let h ∈ C (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 Hn measurable
function, Z Z
g (y) dHn = g (h (x)) |det (U (x))| dmn . (15.24)
h(A) A

Proof: From Corollary 15.5.3, 15.24 holds for any nonnegative simple function in
place of g. In general, let {sk } be an increasing sequence of simple functions which
converges to g pointwise. Then from the monotone convergence theorem
Z Z Z
g (y) dHn = lim sk dHn = lim sk (h (x)) |det (U (x))| dmn
h(A) k→∞ h(A) k→∞ A
Z
= g (h (x)) |det (U (x))| dmn .
A

This proves the theorem. ¥

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.

Corollary 15.5.5 Let V be an open set in Rn 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 L1 (h (V )),
Z Z
n
g (y) dH = g (h (x)) |det (U (x))| dmn .
h(A) A
15.5. THE AREA FORMULA 439

You don’t need to assume U (x) is one to one. The following lemma is like Sard’s
lemma presented earlier. However, it might seem a little easier and if so, it is because
the area formula above is available and Hausdorff 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 Rn and let h : V → Rm 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 n o
−1
E ≡ x ∈ V : U (x) does not exist

Then Hn (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)||. Define
kε : Rn → Rm × Rn µ ¶
h (x)
kε (x) ≡
εx
Thus µ ¶
h (x + v) − h (x)
kε (x + v) − kε (x) =
εv
µ ¶
Dh (x) v
= + o (v)
εv
and so µ ¶
Dh (x)
Dkε (x) = ∈ L (Rn , Rm × Rn )
ε id
It is left as an exercise to explain how
∗ ¡ ∗ ¢
Dkε (x) = Dh (x) ε id ∈ L (Rm × Rn , Rn )

and ¡ ¢ ¡ ¢
2 ∗
Jkε (x) ≡ det Dkε (x) Dkε (x) = det Dh∗ (x) Dh (x) + ε2 id
Now there is an orthonormal basis, {vk } for Rn such that
n
X n
X
Dh∗ (x) Dh (x) = ak vk vk , ε2 id = ε2 vk vk .
k=1 k=1

where each ak ≥ 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 ε2 . Since ||Dh (x)|| is bounded, this shows there exists a constant
C independent of x ∈ V such that on E, Jkε (x) ≤ εC. Now by the earlier area formula,
Theorem 15.5.4,
Z Z
n
Xkε (E) (z) dH = Xkε (E) (kε (x)) Jkε (x) dmn
kε (V ) V
Z
= XE (x) Jkε (x) dmn
V

Thus
Hn (kε (E)) ≤ εCmn (E)
440 HAUSDORFF MEASURES AND AREA FORMULA

However, |h (x) − h¡ (x1 )| ≤¢ |kε (x) − kε (x1 )| and so by Lemma 15.4.4 Hn (kε (E)) ≥
Hn (h (E)) . Let P y x ≡ y. Thus
Hn (h (E)) ≤ εCmn (E)
and since ε is arbitrary, this shows Hn (h (E)) = 0.
In the general case when V might not be bounded define for each k ∈ N sufficiently
large that the sets are nonempty
© ¡ ¢ ª
Vk ≡ B (0,k) ∩ x ∈ V : dist x,V C > 1/k

Then Vk is compact and if Ek ≡ E ∩ Vk and Vk plays the role of V above, it fol-

lows ||Dh (x)|| is bounded on Vk and Hn (h (Ek )) = 0. Now let k → ∞ to conclude
Hn (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 Rn and A is a Lebesgue measurable
subset of V . Let h ∈ C (V ) be one to one. Then if g is any nonnegative Hn measurable
1

function, Z Z
g (y) dHn = g (h (x)) |det (U (x))| dmn . (15.25)
h(V ) V
n o
−1
Proof: Let E = x ∈ V : U (x) does not exist which is the same as the set
where det U (x) = 0. Then by Lemma 15.5.6 Hn (h (E)) = 0 and so
Z Z
g (y) dHn = Xh(V \E) (y) g (y) dHn
h(V ) h(V )
Z
= Xh(V \E) (h (x)) g (h (x)) |det (U (x))| dmn
V
Z
= XV \E (x) g (h (x)) |det (U (x))| dmn
ZV
= g (h (x)) |det (U (x))| dmn
V

This proves the theorem. ¥

Note that if m = n, this reduces to the usual change of variables formula because
Hn = mn on Rn as was shown above.
As before, there is an obvious corollary obtained by splitting up the function in L1
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.
Corollary 15.5.8 Let V be an open set in Rn 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 L1 (h (A)),
Z Z
g (y) dHn = g (h (x)) |det (U (x))| dmn .
h(A) A

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 Hn (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 441

Lemma 15.6.1 For x ∈ V+ , there exists an open ball Bx ⊆ V+ such that h is one
to one on Bx .

Proof: Let Dh (x) = R (x) U (x) be the right polar decomposition. Recall that
2
U (x) is self adjoint and satisfies U (x) v · v ≥ δ |v| for some δ > 0 where δ is the
2
smallest eigenvalue of U (x) , the square root of the smallest eigenvalue of U (x) =
∗
Dh (x) Dh (x) . Let r > 0 such that B (x, r) ⊆ V+ . Then for y ∈ B (x, r), the continuity
2
of y → U (y) , resulting for the assumption that h is C 1 , implies
³ ´
2 2 2 2
U (y) v · v = U (x) v · v + U (y) − U (x) v · v

2 δ2 2 δ2 2
≥ δ 2 |v| − |v| = |v|
2 2
2
provided r is sufficiently small. Thus for small enough r, the eigenvalues of U (y) for
y ∈ B (x, r) are at least as large√as δ 2 /2 and so the eigenvalues of U (y) for these values
of y are lat least as large as δ/ 2. Thus for y ∈ B (x, r),
δ 2
U (y) v · v ≥ √ |v| .
2

Now make r still smaller if necessary such that for y, z ∈ B (x, r),
δ
||Dh (y) − Dh (z)|| < .
2
Then for any y, z of this sort,
δ
|h (z) − h (y) − Dh (y) (z − y)| < |z − y| . (15.26)
2
This follows from the mean value inequality, Theorem 6.4.2 because if you define for
such a fixed 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,

δ
|R∗ o (z − y)| ≤ |o (z − y)| ≤ |z − y| .
2
If h (z) = h (y) , then taking the inner product of both sides of 15.27 with z − y,
µ ¶
δ 2 δ δ 2
0 ≥ (U (y) (z − y) , (z − y)) − |z − y| ≥ √ − |z − y|
2 2 2
showing z = y. This proves the lemma. ¥
Let {Bi } be a countable subset of {Bx }x∈V+ such that V+ = ∪∞i=1 Bi . Let E1 = B1 .
If E1 , · · ·, Ek have been chosen, Ek+1 = Bk+1 \ ∪ki=1 Ei . Thus

∪∞
i=1 Ei = V+ , h is one to one on Ei , Ei ∩ Ej = ∅,
442 HAUSDORFF MEASURES AND AREA FORMULA

and each Ei is a Borel set contained in the open set Bi . Now define
∞
X
n(y) ≡ Xh(Ei ∩A) (y) + Xh(Z) (y).
i=1

Thus
∞
X
Xh(Ei ∩A) (y) = XA+
i=1
where A+ ≡ V+ ∩ A. The set, h (Ei ) , h (Z) are Hn measurable by Lemma 15.4.2. Thus
n (·) is Hn measurable.
Lemma 15.6.2 Let V be an open set, F ⊆ h(V ) be Hn measurable and let A be a
Lebesgue measurable subset of V . Then
Z Z
n(y)XF (y)dmn = XF (h(x))| det Dh(x)|dmn .
h(A) A

Proof: Using Lemma 15.5.6 and the Monotone Convergence Theorem

 
Hn (h(Z))=0
Z Z ∞
X z }| {
 
n(y)XF (y)dHn =  Xh(Ei ) (y) + Xh(Z) (y)  XF (y)dHn
h(A) h(A) i=1

∞ Z
X
= Xh(Ei ) (y)XF (y)dHn
i=1 h(A)
∞ Z
X
= Xh(Ei ∩A) (y)XF (y)dHn
i=1 h(Bi ∩A)
∞ Z
X
= XEi ∩A (x)XF (h(x))| det U (x))|dmn
i=1 Bi ∩A
X∞ Z
= XEi ∩A (x)XF (h(x))| det U (x) |dmn
i=1 A
Z X
∞
= XEi ∩A (x)XF (h(x))| det U (x) |dmn
A i=1
Z Z
= XF (h(x))| det U (x) |dmn = XF (h(x))| det U (x) |dmn .
A+ A

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
XF (h(x))| det U (x) |XA+ + 0XZ (x)
and both functions in the sum are measurable. This proves the lemma. ¥
Definition 15.6.3 For y ∈ h(A), define a function, #, according to the for-
mula
#(y) ≡ number of elements in h−1 (y).
Observe that
#(y) = n(y) Hn a.e. (15.28)
/ h(Z), a set of Hn measure 0. Therefore, # is a measurable
because n(y) = #(y) if y ∈
function because of completeness of Hn .
15.7. THE COAREA FORMULA 443

Theorem 15.6.4 Let g ≥ 0, g is Hn measurable, and let h be C 1 (V ) and A is

a Lebesgue measurable subset of V . Then
Z Z
#(y)g(y)dmn = g(h(x))| det Dh(x)|dmn . (15.29)
h(A) A

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.
This proves the theorem. ¥

15.7 The Coarea Formula

In the coarea formula, h maps V ⊆ Rn to Rm 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 (Rn , Rm ) . 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 (Rn , Rm ) . Then

det (id +A∗ A) = det (id +AA∗ ) .

Proof: This follows from Lemma because

r
X n
X
id +A∗ A = (1 + ak ) wk wk + 1wk wk
k=1 k=r+1

while id +AA∗ is of the form

r
X m
X
(1 + ak ) vk vk + 1vk vk .
k=1 k=r+1

Therefore, both of these determinants equal

r
Y
(1 + ak ) .
k=1

This proves the corollary. ¥

Next is a lemma which leads to some conclusions about measurability.

Lemma 15.7.3 Let h : V ⊆ Rp → Rm be continuous and δ > 0. Then if A ⊆ Rp is

either open or compact,
¡ ¢
y → Hδs A ∩ h−1 (y)
is Borel measurable.
444 HAUSDORFF MEASURES AND AREA FORMULA

Proof: Suppose first that A is compact and suppose for δ > 0,

¡ ¢
Hδs A ∩ h−1 (y) < t
Then there exist sets Si , satisfying
r (Si ) < δ, A ∩ h−1 (y) ⊆ ∪∞
i=1 Si ,

and
∞
X s
α (s) (r (Si )) < t.
i=1
I claim these sets can be taken to be open sets. Choose λ > 1 but close enough to 1
that
X∞
s
α (s) (λr (Si )) < t
i=1
Replace Si with Si + B (0, η i ) where η i is small enough that
diam (Si ) + 2η i < λ diam (Si ) .
Then
diam (Si + B (0, η i )) ≤ λ diam (Si )
and so r (Si + B (0, η i )) ≤ λr (Si ) . Thus
∞
X s
α (s) r (Si + B (0, η i )) < t.
i=1

Hence you could replace Si with Si + B (0, η i ) and so one can assume the sets Si are
open.
Claim: If z is close enough to y, then A ∩ h−1 (z) ⊆ ∪∞ i=1 Si .
Proof: If not, then there exists a sequence {zk } such that
zk → y,
and
xk ∈ (A ∩ h−1 (zk )) \ ∪∞
i=1 Si .
By compactness of A, there exists a subsequence still denoted by k such that
zk → y, xk → x ∈ A \ ∪∞
i=1 Si .

Hence
h (x) = lim h (xk ) = lim zk = y.
k→∞ k→∞

But x ∈/ ∪∞
i=1 Si contrary to the assumption that A ∩ h−1 (y) ⊆ ∪∞ i=1 Si .
It follows from this claim that whenever z is close enough to y,
¡ ¢
Hδs A ∩ h−1 (z) < t.
This shows © ¡ ¢ ª
z ∈ Rp : Hδs A ∩ h−1 (z) < t
¡ ¢
is an open set and so y →Hδs A ∩ h−1 (y) is Borel measurable whenever A is compact.
Now let V be an open set and let
Ak ↑ V, Ak compact.
Then ¡ ¢ ¡ ¢
Hδs V ∩ h−1 (y) = lim Hδs Ak ∩ h−1 (y)
k→∞
s
¡ −1
¢
so y →Hδ V ∩ h (y) is Borel measurable for all V open. This proves the lemma. ¥
15.7. THE COAREA FORMULA 445

Lemma 15.7.4 Let h : V ⊆ Rp → Rm be Lipschitz continuous, satisfying an in-

equality of the form
|h (x) − h (y)| ≤ Lip (h) |x − y|
where Lip
¡ (h) is−1a positive
¢ constant. Suppose A is either open or compact in Rp . Then
s
y → H A ∩ h (y) is also Borel measurable and
Z
¡ ¢ m α (s) α (m) s+m
Hs A ∩ h−1 (y) dmm ≤ 2m (Lip (h)) H (A)
R m α (s + m)
In particular, if s = n − m and p = n
Z
¡ ¢ m α (n − m) α (m)
Hn−m A ∩ h−1 (y) dmm ≤ 2m (Lip (h)) mn (A) (15.30)
Rm α (n)
¡ ¢
Proof: From Lemma 15.7.3 y → Hδs A ∩ h−1 (y) is Borel measurable for each
δ > 0. Without loss of generality, Hs+m (A) < ∞. Now let Ci be closed sets with
r (Ci ) < δ, A ⊆ ∪∞i=1 Ci , and
∞
X s+m
Hδs+m (A) + ε > α (s + m) r (Ci ) .
i=1
¡ ¢
Note each Ci is compact so y → Hδs Ci ∩ h−1 (y) is Borel measurable. Thus
Z
¡ ¢
Hδs A ∩ h−1 (y) dmm
m
ZR X
¡ ¢
≤ Hδs Ci ∩ h−1 (y) dmm
Rm i
XZ ¡ ¢
= Hδs Ci ∩ h−1 (y) dmm
i Rm
XZ
≤ Hδs (Ci ) dmm
i h(Ci )
X
= mm (h (Ci )) Hδs (Ci )
i

Now h (Ci ) is contained in a ball of radius 2r (Ci ) and so

X m m s
≤ (Lip (h)) 2m α (m) r (Ci ) α (s) r (Ci )
i

m α (m) α (s) m X m+s

= (Lip (h)) 2 α (s + m) r (Ci )
α (m + s) i

m α (m) α (s) m ¡ s+m ¢

≤ (Lip (h)) 2 Hδ (A) + ε
α (m + s)
Since ε > 0 is arbitrary,
Z
¡ ¢ m α (m) α (s) m s+m
Hδs A ∩ h−1 (y) dmm ≤ (Lip (h)) 2 Hδ (A)
R m α (m + s)
Taking a limit as δ → 0 This proves the lemma. ¥
Next I will show that whenever A is Lebesgue measurable,
¡ ¢
y → Hn−m A ∩ h−1 (y)

is mm measurable and the above estimate holds.

446 HAUSDORFF MEASURES AND AREA FORMULA

Lemma 15.7.5 Let A be a Lebesgue measurable subset of V an open set and let
h : V ⊆ Rn → Rm be Lipschitz continuous,

|h (x) − h (y)| ≤ Lip (h) |x − y| .

Then ¡ ¢
y → Hn−m A ∩ h−1 (y)
is Lebesgue measurable. Furthermore
Z
¡ ¢ m α (n − m) α (m)
Hn−m A ∩ h−1 (y) dmm ≤ 2m (Lip (h)) mn (A)
Rm α (n)
Proof: Let A be a bounded Lebesgue measurable set in Rn . Then by inner and
outer regularity of Lebesgue measure there exists an increasing sequence of compact
sets, {Kk } contained in A and a decreasing sequence of open sets, {Vk } containing A
such that mn (Vk \ Kk ) < 2−k . Thus mn (V1 ) ≤ mn (A) + 1. By Lemma 15.7.4
Z
¡ ¢ m α (n − m) α (m)
Hδn−m V1 ∩ h−1 (y) dmm < 2m (Lip (h)) (mn (A) + 1) .
Rm α (n)
Also ¡ ¢
Hδn−m Kk ∩ h−1 (y) ≤
¡ ¢ ¡ ¢
Hδn−m A ∩ h−1 (y) ≤ Hδn−m Vk ∩ h−1 (y) (15.31)
By Lemma 15.7.4
Z
¡ ¡ ¢ ¡ ¢¢
= Hδn−m Vk ∩ h−1 (y) − Hδn−m Kk ∩ h−1 (y) dmm
m
ZR
¡ ¢
= Hδn−m (Vk − Kk ) ∩ h−1 (y) dmm
Rm
m α (n − m) α (m)
≤ 2m (Lip (h)) mn (Vk \ Kk )
α (n)
m α (n − m) α (m) −k
< 2m (Lip (h)) 2
α (n)
Let the Borel measurable functions, g and f be defined by
¡ ¢ ¡ ¢
g (y) ≡ lim Hδn−m Vk ∩ h−1 (y) , f (y) ≡ lim Hδn−m Kk ∩ h−1 (y)
k→∞ k→∞
¡ ¢
It follows from the dominated convergence theorem using Hδn−m V1 ∩ h−1 (y) as a
dominating function and 15.31 that
¡ ¢
f (y) ≤ Hδn−m A ∩ h−1 (y) ≤ g (y)

and Z
(g (y) − f (y)) dmm = 0.
Rm
¡ ¢
By completness of mm , this establishes y →Hδn−m A ∩ h−1 (y) is Lebesgue measur-
able. Then by Lemma 15.7.4 again,
Z
¡ ¢ m α (n − m) α (m)
Hδn−m A ∩ h−1 (y) dmm ≤ 2m (Lip (h)) mn (A) .
Rm α (n)
Letting δ →
¡ 0 and using¢ the monotone convergence theorem yields the desired inequality
for Hn−m A ∩ h−1 (y) .
15.7. THE COAREA FORMULA 447

The case where A is not bounded can be handled by considering Ar = A ∩ B (0, r)

and letting r → ∞. This proves the lemma. ¥
By fussing with the isodiametric inequality one can remove the factor of 2m 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.
Define Λ (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 ⊆ Rn → Rm

where m ≤ n is C 1 and also a Lipschitz map and for which
¡ ∗ ¢1/2
Jh (x) ≡ det Dh (x) Dh (x) 6= 0.

Then the following formula holds along with all measurability assertions needed for it to
make sense. Z Z
¡ ¢
Hn−m A ∩ h−1 (y) dy = Jh (x) dx (15.32)
Rm A

Proof: For x ∈ Rn , and i ∈ Λ (n, m), with i = (i1 , · · ·, im ), define xi ≡ (xi1 ,

· · ·, xim ), and π i x ≡ xi . Also for i ∈ Λ (n, m), let ic ∈ Λ (n, n − m) consist of
the remaining indices taken in order. For h : Rn → Rm where m ≤ n, define
¡ ∗ ¢1/2
Jh (x) ≡ det Dh (x) Dh (x) . For each i ∈ Λ (n, m), define hi : Rn → Rm × Rn−m
by µ ¶
i h (x)
h (x) ≡ .
xic
© ª
As in Lemma 15.6.1, since hi is in C 1 (V ) there exist Borel sets Fki which are¡ disjoint,
¢
Fki ⊆ Bki , hi is one to one on the open ball Bki with a C 1 inverse defined on hi Bki and
© ª
∪k Fki = x ∈ Rn : det Dhi (x) 6= 0 .

By assumption, for each x ∈ V there exists i such that det (Dxi h (x)) 6= 0 which implies
det Dhi (x) 6= 0. This follows from Proposition 3.8.18 which says that Dh (x) has m
independent columns. Hence
∪i,j Fji = V
Thus
∪i,j Fji ∩ A = A
© ª
The problem is the Fji might not be disjoint. Let Eji be measurable sets such that
Eji ⊆ Fki ∩ A for some k, the sets, Eji are disjoint, and their union equals A. Then
Z X ∞ Z
X ¡ ∗ ¢1/2
Jh (x) dx = det Dh (x) Dh (x) dx. (15.33)
A i ∈Λ(n,m) j=1 Eji

¡ ¢
Now each Eji is contained in some Bki and so hi has an inverse on hi Bki which I will
denote by g. Thus letting π ic x ≡ xic and using the definition of g

g (h (x) , xic ) = x.

By the chain rule,

Dhi (g (y)) Dg (y) = I
448 HAUSDORFF MEASURES AND AREA FORMULA

¡ ¢
on hi Eji . Changing the variables using the area formula, the expression in 15.33
equals Z
Jh (x) dx =
A

X ∞ Z
X ¡ ∗ ¢1/2
det Dh (g (y)) Dh (g (y)) |det Dg (y)| dy =
i ∈Λ(n,m) j=1 hi (Eji )

X ∞ Z
X ¡ ∗ ¢1/2 ¯¯ ¯−1
det Dh (g (y)) Dh (g (y)) det Dhi (g (y))¯ dy. (15.34)
i ∈Λ(n,m) j=1 hi (Eji )

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 difference
quotients of continuous functions. Thus from 15.33,
Z
¡ ∗ ¢1/2
det Dh (x) Dh (x) dx =
Eji

Z
¡ ∗ ¢1/2 ¯¯ ¯−1
Xhi (E i ) (y) det Dh (g (y)) Dh (g (y)) det Dhi (g (y))¯ dy (15.35)
j
Rn

Next this integral is split using Fubini’s theorem. Let y1 ∈ Rm be fixed and now it is
necessary to decide where y2 is. I need
¡ ¢ ¡ ¡ ¢ ¡ ¢¢
(y1 , y2 ) ∈ hi Eji = h Eji , π ic Eji
¡ ¢
This requires y2 ∈ π ic Eji . However, it is also the case that y1 is given. Now y1 = h (x)
and so
x = (xi , xic ) = (xi , y2 ) ∈ h−1 (y1 )
which implies π ic x = y2 ∈ π ic h−1 (y1 ) . Thus
¡ ¢
y2 ∈ π ic h−1 (y1 ) ∩ Eji

and by Fubini’s theorem, the integral in 15.35 is

Z Z
¡ ∗ ¢1/2 −1
det Dh (g (y)) Dh (g (y)) |det Dxi h (g (y))| dy2 dy1
Rm π ic (h−1 (y1 )∩Eji )
(15.36)
Now consider the inner integral in 15.36 in which y1 is fixed. The integrand equals
· µ ∗ ¶¸1/2
¡ ¢ Dxi h (g (y)) −1
det Dxi h (g (y)) Dxic h (g (y)) ∗ |det Dxi h (g (y))| .
Dxic h (g (y))
(15.37)
I want to massage the above expression slightly. Since y1 is fixed, and

y1 = h (π i g (y) , π ic g (y)) = h (g (y)) ,

it follows

0 = Dxi h (g (y)) Dy2 π i g (y) + Dxic h (g (y)) Dy2 π ic g (y)

= Dxi h (g (y)) Dy2 π i g (y) + Dxic h (g (y)).
15.7. THE COAREA FORMULA 449

because by definition of g and hi , π ic g (y) = y2 so Dy2 π ic g (y) = id . Letting A ≡

Dxi h (g (y)) and B ≡ Dy2 π i g (y) and using the above formula, 15.37 is of the form
· µ ¶¸1/2
¡ ¢ A∗ −1
det A −AB |det A|
−B ∗ A∗
1/2 −1
= det [AA∗ + ABB ∗ A∗ ] |det A|
1/2 −1
= det [A (I + BB ∗ ) A∗ ] |det A|
1/2 1/2 −1
= (det (A) det (A∗ )) det (I + BB ∗ ) |det A|
1/2
= det (I + BB ∗ ) ,
1/2
which, by Corollary 15.7.2, equals det (I + B ∗ B) . (Note the size of the identity
changes in these two expressions, the first being an m × m matrix and the second being
a n − m × n − m matrix.)
Since π ic g (y) = y2 ,
· µ ¶¸1/2
1/2 ¡ ∗ ¢ B
det (I + B ∗ B) = det B I
I
· µ ¶¸1/2
¡ ∗ ∗ ¢ Dy2 π i g (y)
= det Dy2 π i g (y) Dy2 π ic g (y)
Dy2 π ic g (y)
¡ ∗ ¢1/2
= det Dy2 g (y) Dy2 g (y) .
Therefore, 15.36 reduces to
Z
¡ ∗ ¢1/2
det Dh (x) Dh (x) dx =
Eji
Z Z
¡ ∗ ¢1/2
det Dy2 g (y) Dy2 g (y) dy2 dy1 . (15.38)
Rm ( )
π ic h−1 (y1 )∩Eji

Recall g (y1 , y2 ) = g (h (x) , y2 ) = x. Thus

g (y1 , π ic x) = g (y1 , y2 ) = x.
Thus ¡ ¡ ¢¢
g y1 , π ic h−1 (y1 ) ∩ Eji = h−1 (y1 ) ∩ Eji
and so the area formula applied to the inside integral in 15.38 yields
Z Z
¡ ∗ ¢1/2
dHn = det Dy2 g (y) Dy2 g (y) dy2 dy1
h−1 (y1 )∩Eji π ic (h−1 (y1 )∩Eji )

and so this integral equals ¡ ¢

Hn−m h−1 (y1 ) ∩ Eji .
It follows
Z
¡ ∗ ¢1/2
det Dh (x) Dh (x) dx
Eji
Z
¡ ¢
= Hn−m h−1 (y1 ) ∩ Eji dy1 .
Rm
Therefore, summing the terms over all i and j,
Z Z
¡ ∗ ¢1/2 ¡ ¢
det Dh (x) Dh (x) dx = Hn−m h−1 (y) ∩ A dy.
A Rm
This proves the lemma. ¥
You don’t need to assume h is Lipschitz.
450 HAUSDORFF MEASURES AND AREA FORMULA

Corollary 15.7.7 Let A be a Lebesgue measurable set in V, an open set and let
h : V ⊆ Rn → Rm where m ≤ n is C 1 and for which
¡ ∗ ¢1/2
Jh (x) ≡ det Dh (x) Dh (x) 6= 0.

Then the following formula holds along with all measurability assertions needed for it to
make sense. Z Z
¡ ¢
Hn−m A ∩ h−1 (y) dy = Jh (x) dx (15.39)
Rm A

Proof: Consider for each x ∈ A, a ball, B (x, rx ) ⊆ B (x, rx ) ⊆ V such that rx < 1.
Then by the mean value inequality, Theorem 6.4.2, h is Lipschitz on Bi . Letting {Bi }
be an open covering of A, let Ci ≡ Bi ∩ A. Then let {Ai } be such that Ai ⊆ Ci but the
Ai are disjoint and ∪i Ai = A. The conclusion of Lemma 15.7.6 applies with Bi playing
the role of V in that lemma and one can write
Z Z
n−m
¡ −1
¢
H Ai ∩ h (y) dy = Jh (x) dx
Rm Ai

By Lemma 15.4.1 Hn−m Then using the monotone convergence theorem,

Z Z
n−m
¡ −1
¢ ¡ ¢
H A ∩ h (y) dy = Hn−m ∪∞ i=1 Ai ∩ h
−1
(y) dy =
Rm Rm

Z ∞
X Z M
X
n−m
¡ −1
¢ ¡ ¢
H Ai ∩ h (y) dy = lim Hn−m Ai ∩ h−1 (y) dy
Rm Rm M →∞
i=1 i=1
Z M
X M Z
X
¡ ¢ ¡ ¢
= lim Hn−m Ai ∩ h−1 (y) dy = lim Hn−m Ai ∩ h−1 (y) dy
M →∞ Rm i=1 M →∞ Rm
i=1

M Z
X Z X
∞
= lim XAi (x) Jh (x) dx = XAi (x) Jh (x) dx
M →∞
i=1 i=1
Z Z
= X ∪∞
i=1 Ai
(x) Jh (x) dx = XA (x) Jh (x) dx

This proves the theorem. ¥

¡ ∗ ¢1/2
You don’t need to assume Jh (x) ≡ det Dh (x) Dh (x) 6= 0.

Corollary 15.7.8 Let A ⊆ V an open set be measurable and let h : V → Rm be C 1

where m ≤ n. Then the following formula holds along with all measurability assertions
needed for it to make sense.
Z Z
¡ ¢
Hn−m A ∩ h−1 (y) dy = Jh (x) dx (15.40)
Rm A

¡ ∗ ¢1/2
where Jh (x) ≡ det Dh (x) Dh (x) .

Proof: By Corollary 15.7.7, this formula is true for all measurable A contained in
the open set Rn \ 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}.

15.7. THE COAREA FORMULA 451
¡ ¢
Let A be compact so that by Lemma 15.7.4, y →Hn−m A ∩ h−1 (y) is Borel measur-
able. For ε > 0, define k, p : Rn × Rm → Rm by

k (x, y) ≡ h (x) + εy, p (x, y) ≡ y.

Then
Dk (x, y) = (Dh (x) , εI)
and so
µ µ ¶¶
Dh∗ ¡ ∗ ¢
Jk2 = det (Dh (x) , εI) = det Dh (x) Dh (x) + ε2 I
εI
¯¯ ¯ ¯
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
X
λk zk zk
k=1

for each λk ≥ 0 because it is self adjoint and

∗ ∗ ∗
Dh (x) Dh (x) x · x =Dh (x) x·Dh (x) x ≥ 0.

Since A ⊆ S, at least one of these λk 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
m
Y ¡ 2 ¢
Jk2 = λi + ε2 ∈ [ε2m , C 2 ε2 ] (15.41)
i=1

since one of the λi equals 0. Since Jk 6= 0, 15.41 implies

³ ´ Z
εCmn+m A × B (0,1) ≥ |Jk| dmn+m
A×B(0,1)
Z ³ ´
= Hn k−1 (y) ∩ A × B (0,1) dy
Rm
Which by Lemma 15.7.4 is at least as large as
Z Z ³ ´
Cnm Hn−m k−1 (y) ∩ p−1 (w) ∩ A × B (0,1) dwdy (15.42)
Rm Rm

α(n)
where Cnm = α(n−m)α(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: ³ ´
Hn−m k−1 (y) ∩ p−1 (w) ∩ A × B (0,1)
¡ ¢
≥ XB(0,1) (w) Hn−m h−1 (y − εw) ∩ 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, w1 ) ∈ k−1 (y) ∩ p−1 (w) ∩ A × B (0,1)

if and only if

k (x, w1 ) = h (x) + εw1 = y, p (x, w1 ) = w1 = w, x ∈ A,

452 HAUSDORFF MEASURES AND AREA FORMULA

if and only if
(x, w1 ) ∈ h−1 (y − εw) ∩ A × {w}.
Therefore for w ∈ B (0,1),
³ ´
Hn−m k−1 (y) ∩ p−1 (w) ∩ A × B (0,1)
¡ ¢ ¡ ¢
≥ Hn−m h−1 (y − εw) ∩ A × {w} = Hn−m h−1 (y − εw) ∩ A .
(Actually equality holds in the claim.) From the claim, 15.42 is at least as large as
Z Z
¡ ¢
Cnm Hn−m h−1 (y − εw) ∩ A dwdy (15.43)
Rm B(0,1)
Z Z
¡ ¢
= Cnm Hn−m h−1 (y − εw) ∩ A dydw
B(0,1) Rm
³ ´Z ¡ ¢
= Cnm mm B (0,1) Hn−m h−1 (y) ∩ A dy. (15.44)
Rm
The use of Fubini’s theorem is justified because the integrand is Borel measurable.
Now by 15.44 this has shown
³ ´ ³ ´Z ¡ ¢
εCmn+m A × B (0,1) ≥ Cnm mm B (0,1) Hn−m h−1 (y) ∩ A dy.
Rm

it follows since ε > 0 is arbitrary,

Z Z
¡ ¢
Hn−m A ∩ h−1 (y) dy = 0 = Jh (x) dx.
Rm A

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
Z Z
¡ ¢ ¡ ¢
Hn−m A ∩ h−1 (y) dy = Hn−m A ∩ S ∩ h−1 (y) dy
Rm Rm
Z
¡ ¢
+ Hn−m (A \ S) ∩ h−1 (y) dy
m
Z R
¡ ¢
= Hn−m (A \ S) ∩ h−1 (y) dy
m
ZR Z
= Jh (x) dx = Jh (x) dx
A\S A

This completes the proof of the coarea formula. ¥

15.8 A Nonlinear Fubini’s Theorem

Coarea formula holds for h : V ⊆ Rn → Rm , n ≥ m if whenever A is a Lebesgue
measurable subset of the open set V,the following formula is valid.
Z Z
n−m
¡ −1
¢
H A ∩ h (y) dy = Jh (x) dx (15.45)
Rm A

where ¡ ∗ ¢1/2
Jh (x) = det Dh (x) Dh (x) .
15.8. A NONLINEAR FUBINI’S THEOREM 453

Note this is the same as

Z Z
¡ ¢
Jh (x))dx = Hn−m A ∩ h−1 (y) dy
A h(A)

/ h (A) , then h−1 (y) = ∅. Now let

because if y ∈
p
X
s (x) = ci XEi (x)
i=1

where Ei is a Lebesgue measurable subset of V and ci ≥ 0. Then

Z p
X Z
s (x) Jh (x) dx = ci Jh (x) dx
V i=1 Ei

Xp Z
¡ ¢
= ci Hn−m Ei ∩ h−1 (y) dy
i=1 h(Ei )
Z p
X ¡ ¢
= ci Hn−m Ei ∩ h−1 (y) dy
h(V ) i=1
Z "Z #
n−m
= s dH dy
h(V ) h−1 (y)

Z "Z #
n−m
= s dH dy. (15.46)
h(V ) h−1 (y)

Theorem 15.8.1 Let g ≥ 0 be Lebesgue measurable and let V be an open subset

of Rn .
h : V → Rm , n ≥ m
where h is C 1 .Then
Z Z "Z #
n−m
g (x) J ((Dh (x))) dx = g dH dy.
V h(V ) h−1 (y)

Proof: Let si ↑ g where si is a simple function satisfying 15.46. Then let i → ∞ and
use the monotone convergence theorem to replace si with g. This proves the nonlinear
version of Fubini’s theorem. ¥
Note that this formula is a nonlinear version of Fubini’s theorem. The “n − m di-
mensional surface”, h−1 (y), plays the role of Rn−m and Hn−m is like n−m dimensional
Lebesgue measure. The term, J ((Dh (x))), corrects for the error occurring because of
the lack of flatness of h−1 (y) .
454 HAUSDORFF MEASURES AND AREA FORMULA
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] D’Angelo, 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 Differential 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)

455
456 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 Scientific 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

Cc∞ , 194 Caratheodory’s procedure, 164

Ccm , 194 Cartesian coordinates, 24
π systems, 207 Casorati Weierstrass theorem, 411
σ algebra, 155 Cauchy integral theorem, 398
Cauchy Riemann equations, 395
a.e., 156 Cauchy Schwarz inequality, 53
adjugate, 45 Cayley Hamilton theorem, 51
almost everywhere, 156 chain rule, 120
approximate identity, 220 change of variables, 453
area formula, 440 change of variables general case, 243,
functions which are not one to one, 443
442 characteristic polynomial, 49
area of a parallelogram, 390 chart, 293
arithmetic mean, 150 coarea formula, 447
at most countable, 15 cofactor, 43
atlas, 293 compact, 151
axiom of choice, 11, 15, 204 completion of measure space, 167
axiom of extension, 11 components of a vector, 28
axiom of specification, 11 connected, 90
axiom of unions, 11 connected component, 91
connected components, 91
barallelepiped conservative, 374
volume, 390 contour integral, 396
barrier condition, 345 convergence in measure, 200
beta function, 216 convex hull, 81
Binet Cauchy formula, 300 convolution, 220
block matrix, 36 Coordinates, 23
Borel Cantelli lemma, 200 countable, 15
Borel measurable, 204 countable basis, 83, 319
Borel regular, 418 Cramer’s rule, 45
Borsuk, 270 cross product, 389
Borsuk Ulam theorem, 274 area of parallelogram, 390
boundary operator, 350 coordinate description, 392
bounded, 77 geometric description, 389
bounded variation, 359
box product, 391 derivatives, 120
Brouwer degree, 260 determinant, 39
Borsuk theorem, 270 product, 42
Brouwer fixed point theorem, 273 transpose, 40
Browder’s lemma, 256 diameter of a set, 80
differential equations
Cantor function, 204 Peano existence theorem, 116
Cantor set, 203 differential form, 305
Caratheodory’s criterion, 416 closed, 322

457
458 INDEX

derivative, 307 Holder’s inequality, 55

exact, 322 homotopic, 259
integral, 305 homotopy, 247, 259
differential forms, 304
Dini derivates, 251 imaginary part, 394
dominated convergence theorem, 192 implicit function theorem, 138
dot product, 53 inner product, 53
dual basis, 70 inner regularity, 157
interior point, 75
Egoroff theorem, 296 invariance of domain, 271
eigenvalue, 149 invariant, 62
eigenvalues, 49 inverse function theorem, 139, 148
equality of mixed partial derivatives, 135 inverses and determinants, 44
equicontinuous, 110 isodiametric inequality, 421, 425
equivalence class, 17 isolated singularity, 411
equivalence relation, 16 iterated integral, 209
equivalent norms, 82
exchange theorem, 26 Jordan arc, 353
exponential growth, 253 Jordan curve, 354
extreme value theorem, 89 Jordan curve theorem, 289
Jordan Separation theorem, 278
Fatou’s lemma, 185 Jordan separation theorem, 279
fixed point property, 290
Frechet derivative, 119 Lagrange multipliers, 145, 146
frontier, 354 Laplace expansion, 43
Fubini’s theorem, 212 Laplace transform, 250, 254
function, 13 Laplace’s equation, 327
uniformly continuous, 94 Lebesgue number, 151
length, 361
Gamma function, 427 limit point, 76
gamma function, 202, 216 Lindelof property, 319
Gateaux derivative, 123 linear combination, 26
geometric mean, 150 linear transformation, 30
gradient, 147 linearly dependent, 26
Gram Schmidt process, 59 linearly independent, 26
Grammian, 70 Liouville’s theorem, 407
Grammian matrix, 322 local maximum, 143
grating, 349 local minimum, 143
locally finite, 223
harmonic function, 327 lower semicontinuous, 114
Haursdorff measures, 415
Hausdorff and Lebesgue measure, 426, manifolds
428 boundary, 293
Hausdorff dimension, 426 interior, 293
Hausdorff measure orientable, 295
translation invariant, 419 smooth, 295
Hausdorff measures, 415 matrix
Heine Borel, 74 left inverse, 45
Heine Borel theorem, 152 lower triangular, 46
Hermitian, 60 right inverse, 45
Hessian matrix, 143 upper triangular, 46
higher order derivatives, 128 matrix of a linear transformation, 32
Holder, 113 max. min.theorem, 89
INDEX 459

measurable, 163 rank of a matrix, 46

measurable function, 173 rational function, 98
pointwise limits, 173 real part, 394
measurable functions rectifiable, 359
Borel, 200 rectifiable curve, 359
measurable sets, 163 regular measure, 157
measure, 155 regular values, 259
measure space, 155 relative topology, 293
minimal polynomial, 50 retraction, 247
minor, 43 right Cauchy Green strain tensor, 65
mixed partial derivatives, 133 right handed system, 389
mollifier, 220 Rouche theorem, 409
monotone convergence theorem, 183 Russell’s paradox, 13
monotone functions
differentiable, 252 Sard’s lemma, 240
multi - index, 98 scalars, 25
multi-index, 130 Schroder Bernstein theorem, 14
Schur’s theorem, 62
nested interval lemma, 74 second derivative test, 144
nonlinear Fubini’s theorem, 453 self adjoint, 60
nonmeasurable set, 204 separable, 83
norm separated, 90
p norm, 55 sets, 11
sigma algebra, 155
odd map, 269 simple curve, 359
open cover, 151 simple functions, 176
open set, 75 singular values, 259
orientable manifold, 295 smooth surface, 392
orientation, 305, 361 span, 26
oriented curve, 361 Steiner symetrization, 423
orthonormal, 58 Stirling’s formula, 202
outer measure, 155, 200 Stoke’s theorem, 388, 393
outer regularity, 157 Stokes theorem, 311
subharmonic, 341
parallelepiped, 390 subspace, 26
partial derivatives, 122 support, 194
pi systems, 207
pointwise convergence Taylor’s formula, 142
sequence, 95 Tietze extension theorem, 104
series, 97 triangle inequality, 55
Poisson’s equation, 327 trivial, 26
Poisson’s integral formula, 335 uniform contractions, 136
Poisson’s problem, 327 uniform convergence
polar decomposition sequence, 95
left, 67 series, 97
right, 65 uniformly Cauchy
potential, 372 sequence, 95
power set, 11 uniformly continuous, 94
precompact, 116 uniformly integrable, 201, 298
primitive, 398 unitary, 66
probability measure, 177 upper semicontinuous, 114
probability space, 177
product formula, 277 vector space axioms, 25
460 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, 427

weak maximum principle, 327

Weierstrass M test, 97
work, 405