Basic Analysis

Kenneth Kuttler
April 16, 2001

Contents
1 Basic Set theory
1.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 The Schroder Bernstein theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9
10
11
14

2 Linear Algebra
2.1 Vector Spaces . . . . . . . . .
2.2 Linear Transformations . . .
2.3 Inner product spaces . . . . .
2.4 Exercises . . . . . . . . . . .
2.5 Determinants . . . . . . . . .
2.6 The characteristic polynomial
2.7 The rank of a matrix . . . . .
2.8 Exercises . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

15
15
18
22
30
31
38
40
41

3 General topology
3.1 Compactness in metric space
3.2 Connected sets . . . . . . . .
3.3 The Tychonoff theorem . . .
3.4 Exercises . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

43
49
52
55
57

4 Spaces of Continuous Functions

61
4.1 Compactness in spaces of continuous functions . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Stone Weierstrass theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5 Abstract measure and Integration
5.1 Algebras . . . . . . . . . . . . . .
5.2 Monotone classes and algebras . .
5.3 Exercises . . . . . . . . . . . . . .
5.4 The Abstract Lebesgue Integral . .
5.5 The space L1 . . . . . . . . . . . .
5.6 Double sums of nonnegative terms
5.7 Vitali convergence theorem . . . .
5.8 The ergodic theorem . . . . . . . .
5.9 Exercises . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
3

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

71
71
71
78
80
83
87
88
90
95

4
6 The
6.1
6.2
6.3

CONTENTS
Construction Of Measures
97
Outer measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Positive linear functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7 Lebesgue Measure
7.1 Lebesgue measure . . . .
7.2 Iterated integrals . . . . .
7.3 Change of variables . . . .
7.4 Polar coordinates . . . . .
7.5 The Lebesgue integral and
7.6 Exercises . . . . . . . . .

.
.
.
.
.
.

113
113
117
120
124
126
127

8 Product Measure
8.1 Measures on infinite products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 A strong ergodic theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

131
136
141
145

9 Fourier Series
9.1 Definition and basic properties . . . . . . . . .
9.2 Pointwise convergence of Fourier series . . . . .
9.2.1 Dinis criterion . . . . . . . . . . . . . .
9.2.2 Jordans criterion . . . . . . . . . . . . .
9.2.3 The Fourier cosine series . . . . . . . . .
9.3 The Cesaro means . . . . . . . . . . . . . . . .
9.4 Gibbs phenomenon . . . . . . . . . . . . . . .
9.5 The mean square convergence of Fourier series
9.6 Exercises . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

147
147
150
152
154
157
158
160
160
164

10 The
10.1
10.2
10.3
10.4
10.5
10.6

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

169
169
175
182
184
192
195

. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
the Riemann integral
. . . . . . . . . . . . .

Frechet derivative
Norms for finite dimensional
The Derivative . . . . . . .
Higher order derivatives . .
Implicit function theorem .
Taylors formula . . . . . .
Exercises . . . . . . . . . .

vector
. . . .
. . . .
. . . .
. . . .
. . . .

spaces
. . . .
. . . .
. . . .
. . . .
. . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

11 Change of variables for C 1 maps

199
11.1 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
11.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12 The
12.1
12.2
12.3
12.4
12.5
12.6

Lp Spaces
Basic inequalities and properties . . . . .
Density of simple functions . . . . . . . .
Continuity of translation . . . . . . . . . .
Separability . . . . . . . . . . . . . . . . .
Mollifiers and density of smooth functions
Exercises . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

209
209
213
214
215
215
218

CONTENTS

13 Fourier Transforms
13.1 The Schwartz class . . . . . . . .
13.2 Fourier transforms of functions in
13.3 Tempered distributions . . . . . .
13.4 Exercises . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

223
223
228
233
238

14 Banach Spaces
14.1 Baire category theorem . .
14.2 Uniform boundedness closed
14.3 Hahn Banach theorem . . .
14.4 Exercises . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .
graph and open mapping theorems
. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

243
243
246
248
253

15 Hilbert Spaces
15.1 Basic theory . . . . . . . .
15.2 Orthonormal sets . . . . .
15.3 The Fredholm alternative
15.4 Sturm Liouville problems
15.5 Exercises . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

. . . . .
L2 (Rn )
. . . . .
. . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.

.
.
.
.
.

16 Brouwer Degree
16.1 Preliminary results . . . . . . . . . . . . . .
16.2 Definitions and elementary properties . . .
16.3 Applications . . . . . . . . . . . . . . . . . .
16.4 The Product formula and Jordan separation
16.5 Integration and the degree . . . . . . . . . .
17 Differential forms
17.1 Manifolds . . . . . . . . . . . . . . . .
17.2 The integration of differential forms on
17.3 Some examples of orientable manifolds
17.4 Stokes theorem . . . . . . . . . . . . .
17.5 A generalization . . . . . . . . . . . .
17.6 Surface measures . . . . . . . . . . . .
17.7 Divergence theorem . . . . . . . . . .
17.8 Exercises . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

257
257
261
266
268
274

. . . . .
. . . . .
. . . . .
theorem
. . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

277
277
278
287
289
293

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

297
297
298
302
304
306
308
311
316

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

. . . . . .
manifolds
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .

.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

18 Representation Theorems
18.1 Radon Nikodym Theorem . . . . . . . . . . . . . . . .
18.2 Vector measures . . . . . . . . . . . . . . . . . . . . .
18.3 Representation theorems for the dual space of Lp . . .
18.4 Riesz Representation theorem for non finite measure
18.5 The dual space of C (X) . . . . . . . . . . . . . . . . .
18.6 Weak convergence . . . . . . . . . . . . . . . . . . .
18.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . .

. . . .
. . . .
. . . .
spaces
. . . .
. . . .
. . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

319
319
321
325
330
335
338
339

19 Weak Derivatives
19.1 Test functions and weak derivatives .
19.2 Weak derivatives in Lploc . . . . . . .
19.3 Morreys inequality . . . . . . . . . .
19.4 Rademachers theorem . . . . . . . .
19.5 Exercises . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

345
345
348
350
352
355

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

CONTENTS

20 Fundamental Theorem of Calculus

20.1 The Vitali covering theorem . . . . . . . . . . . . .
20.2 Differentiation with respect to Lebesgue measure .
20.3 The change of variables formula for Lipschitz maps
20.4 Mappings that are not one to one . . . . . . . . . .
20.5 Differential forms on Lipschitz manifolds . . . . . .
20.6 Some examples of orientable Lipschitz manifolds .
20.7 Stokes theorem on Lipschitz manifolds . . . . . . .
20.8 Surface measures on Lipschitz manifolds . . . . . .
20.9 The divergence theorem for Lipschitz manifolds . .
20.10 Exercises . . . . . . . . . . . . . . . . . . . . . . .
21 The
21.1
21.2
21.3

.
.
.
.
.
.
.
.
.
.

359
359
361
364
372
375
376
378
379
383
386

complex numbers
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The extended complex plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

393
396
397
398

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

22 Riemann Stieltjes integrals

399
22.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
23 Analytic functions
409
23.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
23.2 Examples of analytic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
23.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
24 Cauchys formula for a disk
415
24.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
25 The
25.1
25.2
25.3

general Cauchy integral formula

423
The Cauchy Goursat theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
The Cauchy integral formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432

26 The
26.1
26.2
26.3
26.4
26.5

open mapping theorem

Zeros of an analytic function . . . . . . . .
The open mapping theorem . . . . . . . . .
Applications of the open mapping theorem
Counting zeros . . . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

433
433
434
436
437
440

27 Singularities
443
27.1 The Laurent series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
27.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
28 Residues and evaluation of integrals
28.1 The argument principle and Rouches theorem .
28.2 Exercises . . . . . . . . . . . . . . . . . . . . . .
28.3 The Poisson formulas and the Hilbert transform
28.4 Exercises . . . . . . . . . . . . . . . . . . . . . .
28.5 Infinite products . . . . . . . . . . . . . . . . . .
28.6 Exercises . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

451
460
461
462
466
467
473

CONTENTS

29 The Riemann mapping theorem

477
29.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
30 Approximation of analytic functions
483
30.1 Runges theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
30.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
A The Hausdorff Maximal theorem
489
A.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492

CONTENTS

Basic Set theory

We think of a set as a collection of things called elements of the set. For example, we may consider the set
of integers, the collection of signed whole numbers such as 1,2,-4, etc. This set which we will believe in 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 we use parentheses, { } to specify a set. When we do this, we list 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} . We say x is an element of a set S, and write x S if x is one of the things
in S. Thus, 1 {1, 0, 1, 2, 3} . Here are some axioms about sets. Axioms are statements we will agree to
believe.
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.
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 we 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 we
would want 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, we could consider 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. Of course, there could be questions about what constitutes a condition. 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} .
We will leave these sorts of considerations however and assume our conditions make sense. The axiom of
unions states that if we have 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
9

10

BASIC SET THEORY

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, we write the union of all these sets in the following way.
{A : A S}
or sometimes as
S.
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, we could write
Y
A
AS

for the Cartesian product. We can think of the Cartesian product as the set of choice functions, a choice
function being a function 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.
We say A is a subset of B and write A B if every element of A is also an element of B. This can also
be written as B A. We say A is a proper subset of B and write A B or B A if A is a subset of B but
A is not equal to B, A 6= B. The intersection of two sets is a set denoted as A 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 . The union of two sets is denoted as A B and it means
the set of all elements which are in either of the sets. We know this is a set by 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, if we want to take the complement of a set A, we can say its complement, denoted as AC ( or more
precisely as X \ A) is a set by using the axiom of specification to write
AC {x X : x
/ A}
The symbol
/ is read as is not an element of. Note the axiom of specification takes place relative to a
given set which we believe exists. Without this universal set we cannot use the axiom of specification to
speak of the complement.
Words such as all or there exists are called quantifiers and they must be understood relative to
some given set. Thus we can speak of the set of all integers larger than 3. Or we can say 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 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.

1.1

Exercises

1. There is no set of all sets. This was not always known and was pointed out by Bertrand Russell. Here
is what he observed. Suppose there were. Then we could use the axiom of specification to consider
the set of all sets which are not elements of themselves. Denoting this set by S, determine whether S
is an element of itself. Either it is or it isnt. Show there is a contradiction either way. This is known
as Russells paradox.

1.2. THE SCHRODER BERNSTEIN THEOREM

11

2. The above problem shows there is no universal set. Comment on the statement Nothing contains
everything. What does this show about the precision of standard English?
3. Do you believe each person who has ever lived on this earth has the right to do whatever he or she
wants? (Note the use of the universal quantifier with no set in sight.) If you believe this, do you really
believe what you say you believe? What of those people who want to deprive others their right to do
what they want? Do people often use quantifiers this way?
4. DeMorgans laws are very useful in mathematics. Let S be a set of sets each of which is contained in
some universal set, U . Show


C
AC : A S = ( {A : A S})
and


C
AC : A S = ( {A : A S}) .
5. Let S be a set of sets show
B {A : A S} = {B A : A S} .
6. Let S be a set of sets show
B {A : A S} = {B A : A S} .

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. To aid
in this endeavor and because it is important for its own sake, we give the following definition.
Definition 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 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 } .
and we write 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 a mapping, sort of a machine which takes inputs,
x and makes them into a unique output, f (x) . Of course, that is what the above definition says with more
precision. An ordered pair, (x, y) which is an element of the function has an input, x and a unique output,
y which we denote as f (x) while the name of the function is f.
The following theorem which is interesting for its own sake will be used to prove the Schroder Bernstein
theorem.
Theorem 1.2 Let f : X Y and g : Y X be two mappings. 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.

12

BASIC SET THEORY

The following picture illustrates the conclusion of this theorem.
X

Y
f
A

B = g(D) 

- C = f (A)

g
D

Proof: We will say A0 X satisfies P if whenever y Y \ f (A0 ) , g (y)

/ A0 . Note satisfies P.
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. Define
C f (A) , D Y \ C, B X \ A.
Thus all conditions of the theorem are satisfied except for g (D) = B and we verify this condition now.
Suppose x B = X \ A. Then A {x} does not satisfy P because this set is larger than A.Therefore
there exists
y Y \ f (A {x}) Y \ f (A) D
such that g (y) A {x}. But g (y)
/ A because y Y \ f (A) and A satisfies P. Hence g (y) = x and this
proves the theorem.
Theorem 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 Theorem1.2 and define

f (x) if x A
h (x)
g 1 (x) if x B
It is clear h is one to one and onto.
Recall that the Cartesian product may be considered as the collection of choice functions. We give a
more precise description next.
Definition 1.4 Let I be a set and let Xi be a set for each i I. We say that f is a choice function and
write
Y
f
Xi
iI

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= .
iI

The symbol above denotes the collection of all choice functions. Using the axiom of choice, we can obtain
the following interesting corollary to the Schroder Bernstein theorem.

1.2. THE SCHRODER BERNSTEIN THEOREM

13

Corollary 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, let
f01 (y) f 1 (y) {x X : f (x) = y}
and similarly let g01 (x) g 1 (x) . We used the axiom of choice to pick a single element, f01 (y) in f 1 (y)
and similarly for g 1 (x) . Then f01 and g01 are one to one so by the Schroder Bernstein theorem, there
exists h : X Y which is one to one and onto.
Definition 1.6 We say a set S, is finite if there exists a natural number n and a map which maps {1, , n}
one to one and onto S. We say S is infinite if it is not finite. A set S, is called countable if there exists a
map mapping N one to one and onto S.(When maps a set A to a set B, we will write : A B in the
future.) Here N {1, 2, }, the natural numbers. If there exists a map : N S which is onto, we say that
S is at most countable.
In the literature, the property of being at most countable is often referred to as being countable. When
this is done, there is usually no harm incurred from this sloppiness 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 exhaust
the entire set. The possibility that a single element of the set may occur more than once in the list is often
not important.
Theorem 1.7 If X and Y are both at most countable, then X Y is also at most countable.
Proof: We know there exists a mapping : N X which is onto. If we define (i) xi we may consider
X as the sequence {xi }
i=1 , written in the traditional way. Similarly, we may consider Y as the sequence
{yj }
j=1 . It follows we can represent all elements of X Y by the following infinite rectangular array.
(x1 , y1 ) (x1 , y2 ) (x1 , y3 )
(x2 , y1 ) (x2 , y2 ) (x2 , y3 )
(x3 , y1 ) (x3 , y2 ) (x3 , y3 ) .
..
..
..
.
.
.
We follow a path through this array as follows.
(x1 , y1 ) (x1 , y2 )
.
%
(x2 , y1 )
(x2 , y2 )

%
(x3 , y1 )

(x1 , y3 )

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. In this way we see that we can assign a number from N to each element of X Y. In
other words there exists a mapping from N onto X Y. This proves the theorem.
Corollary 1.8 If either X or Y is countable, then X Y is also countable.
Proof: By Theorem 1.7, there exists a mapping : N X Y which is onto. Suppose without loss of
generality that X is countable. Then there exists : N X which is one to one and onto. Let : X Y N
be defined by ((x, y)) 1 (x) . Then by Corollary 1.5, there is a one to one and onto mapping from
X Y to N. This proves the corollary.

14

BASIC SET THEORY

Theorem 1.9 If X and Y are at most countable, then X Y is at most countable.

Proof: Let X = {xi }

i=1 , Y = {yj }j=1 and consider the following array consisting of X Y and path
through it.

x1
y1

x2
.
y2

x3

Thus the first element of X Y is x1 , the second is x2 the third is y1 the fourth is y2 etc. This proves the
theorem.
Corollary 1.10 If either X or Y are countable, then X Y is countable.
Proof: There is a map from N onto X Y. Suppose without loss of generality that X is countable and
: N X is one to one and onto. Then define (y) 1, for all y Y,and (x) 1 (x) . Thus, maps
X Y onto N and applying Corollary 1.5 yields the conclusion and proves the corollary.

1.3

Exercises

1. Show the rational numbers, Q, are countable.

2. We say a number is an algebraic number if it is the solution of an equation of the form
an xn + + a1 x + a0 = 0

where all the aj are integers and all exponents are also integers. Thus 2 is an algebraic number
because it is a solution of the equation x2 2 = 0. Using the fundamental theorem of algebra which
implies that such equations or order n have at most n solutions, show the set of all algebraic numbers
is countable.
3. Let A be a set and let P (A) be its power set, the set of all subsets of A. Show there does not exist
any function f, which maps A onto P (A) . Thus the power set is always strictly larger than the set
from which it came. Hint: Suppose f is onto. Consider S {x A : x
/ f (x)}. If f is onto, then
f (y) = S for some y A. Is y f (y)? Note this argument holds for sets of any size.
4. The empty set is said to be a subset of every set. Why? Consider the statement: If pigs had wings,
then they could fly. Is this statement true or false?
5. If S = {1, , n} , show P (S) has exactly 2n elements in it. Hint: You might try a few cases first.
6. Show the set of all subsets of N, the natural numbers, which have 3 elements, is countable. Is the set
of all subsets of N which have finitely many elements countable? How about the set of all subsets of
N?

Linear Algebra
2.1

Vector Spaces

A vector space is an Abelian group of vectors satisfying the axioms of an Abelian group,
v + w = w + v,
the commutative law of addition,
(v + w) + z = v+ (w + z) ,
the associative law for addition,
v + 0 = v,
the existence of an additive identity,
v+ (v) = 0,
the existence of an additive inverse, along with a field of scalars, F which are allowed to multiply the
vectors according to the following rules. (The Greek letters denote scalars.)
(v + w) = v+w,

(2.1)

( + ) v =v+v,

(2.2)

(v) = (v) ,

(2.3)

1v = v.

(2.4)

The field of scalars will always be assumed to be either R or C and the vector space will be called real or
complex depending on whether the field is R or C. A vector space is also called a linear space.
For example, Rn with the usual conventions is an example of a real vector space and Cn is an example
of a complex vector space.
Definition 2.1 If {v1 , , vn } V, a vector space, then
span (v1 , , vn ) {

n
X

i vi : i F}.

i=1

A subset, W V is said to be a subspace if it is also a vector space with the same field of scalars. Thus
W V is a subspace if u + v W whenever , F and u, v W. The span of a set of vectors as just
described is an example of a subspace.
15

16

LINEAR ALGEBRA

Definition 2.2 If {v1 , , vn } V, we say the set of vectors is linearly independent if

n
X

i vi = 0

i=1

implies
1 = = n = 0
and we say {v1 , , vn } is a basis for V if
span (v1 , , vn ) = V
and {v1 , , vn } is linearly independent. We say the set of vectors is linearly dependent if it is not linearly
independent.
Theorem 2.3 If
span (u1 , , um ) = span (v1 , , vn )
and {u1 , , um } are linearly independent, then m n.
Proof: Let V span (v1 , , vn ) . Then
u1 =

n
X

ci vi

i=1

and one of the scalars ci is non zero. This must occur because of the assumption that {u1 , , um } is
linearly independent. (We cannot have any of these vectors the zero vector and still have the set be linearly
independent.) Without loss of generality, we assume c1 6= 0. Then solving for v1 we find
v1 span (u1 , v2 , , vn )
and so
V = span (u1 , v2 , , vn ) .
Thus, there exist scalars c1 , , cn such that
u2 = c1 u1 +

n
X

ck vk .

k=2

By the assumption that {u1 , , um } is linearly independent, we know that at least one of the ck for k 2
is non zero. Without loss of generality, we suppose this scalar is c2 . Then as before,
v2 span (u1 , u2 , v3 , , vn )
and so V = span (u1 , u2 , v3 , , vn ) . Now suppose m > n. Then we can continue this process of replacement
till we obtain
V = span (u1 , , un ) = span (u1 , , um ) .
Thus, for some choice of scalars, c1 cn ,
um =

n
X

ci ui

i=1

which contradicts the assumption of linear independence of the vectors {u1 , , un }. Therefore, m n and
this proves the Theorem.

2.1. VECTOR SPACES

17

Corollary 2.4 If {u1 , , um } and {v1 , , vn } are two bases for V, then m = n.
Proof: By Theorem 2.3, m n and n m.
Definition 2.5 We say a vector space V is of dimension n if it has a basis consisting of n vectors. This
is well defined thanks to Corollary 2.4. We assume here that n < and say such a vector space is finite
dimensional.
Theorem 2.6 If V = span (u1 , , un ) then some subset of {u1 , , un } is a basis for V. Also, if {u1 ,
, uk } V is linearly independent and the vector space is finite dimensional, then the set, {u1 , , uk }, can
be enlarged to obtain a basis of V.
Proof: Let
S = {E {u1 , , un } such that span (E) = V }.
For E S, let |E| denote the number of elements of E. Let
m min{|E| such that E S}.
Thus there exist vectors
{v1 , , vm } {u1 , , un }
such that
span (v1 , , vm ) = V
and m is as small as possible for this to happen. If this set is linearly independent, it follows it is a basis
for V and the theorem is proved. On the other hand, if the set is not linearly independent, then there exist
scalars,
c1 , , cm
such that
0=

m
X

ci vi

i=1

and not all the ci are equal to zero. Suppose ck 6= 0. Then we can solve for the vector, vk in terms of the
other vectors. Consequently,
V = span (v1 , , vk1 , vk+1 , , vm )
contradicting the definition of m. This proves the first part of the theorem.
To obtain the second part, begin with {u1 , , uk }. If
span (u1 , , uk ) = V,
we are done. If not, there exists a vector,
uk+1
/ span (u1 , , uk ) .
Then {u1 , , uk , uk+1 } is also linearly independent. Continue adding vectors in this way until the resulting
list spans the space V. Then this list is a basis and this proves the theorem.

18

LINEAR ALGEBRA

2.2

Linear Transformations

Definition 2.7 Let V and W be two finite dimensional vector spaces. We say
L L (V, W )
if for all scalars and , and vectors v, w,
L (v+w) = L (v) + L (w) .
We will sometimes write Lv when it is clear that L (v) is meant.
An example of a linear transformation is familiar matrix multiplication. Let A = (aij ) be an m n
matrix. Then we may define a linear transformation L : Fn Fm by
(Lv)i

n
X

aij vj .

j=1

Here

v1

v ... .
vn
Also, if V is an n dimensional vector space and {v1 , , vn } is a basis for V, there exists a linear map
q : Fn V
defined as
q (a)

n
X

ai vi

i=1

where
a=

n
X

ai ei ,

i=1

for ei the standard basis vectors for Fn consisting of

0
..
.

ei
1
.
..
0

where the one is in the ith slot. It is clear that q defined in this way, is one to one, onto, and linear. For
v V, q 1 (v) is a list of scalars called the components of v with respect to the basis {v1 , , vn }.
Definition 2.8 Given a linear transformation L, mapping V to W, where {v1 , , vn } is a basis of V and
{w1 , , wm } is a basis for W, an m n matrix A = (aij )is called the matrix of the transformation L with
respect to the given choice of bases for V and W , if whenever v V, then multiplication of the components
of v by (aij ) yields the components of Lv.

2.2. LINEAR TRANSFORMATIONS

19

The following diagram is descriptive of the definition. Here qV and qW are the maps defined above with
reference to the bases, {v1 , , vn } and {w1 , , wm } respectively.
{v1 , , vn }

L
V W
qV qW
Fn Fm
A

{w1 , , wm }
(2.5)

Letting b Fn , this requires

X

aij bj wi = L

i,j

bj vj =

bj Lvj .

Now
Lvj =

cij wi

(2.6)

for some choice of scalars cij because {w1 , , wm } is a basis for W. Hence
X
X X
X
aij bj wi =
bj
cij wi =
cij bj wi .
i,j

i,j

It follows from the linear independence of {w1 , , wm } that

X
X
aij bj =
cij bj
j

for any choice of b Fn and consequently

aij = cij
where cij is defined by (2.6). It may help to write (2.6) in the form




Lv1 Lvn = w1 wm C = w1

wm

(2.7)

where C = (cij ) , A = (aij ) .

Example 2.9 Let
V { polynomials of degree 3 or less},
W { polynomials of degree 2 or less},
and L D where D is the differentiation operator. A basis for V is {1,x, x2 , x3 } and a basis for W is {1, x,
x2 }.
What is the matrix of this linear transformation with respect to this basis? Using (2.7),




0 1 2x 3x2 = 1 x x2 C.
It follows from this that

0
C= 0
0

1
0
0

0
2
0

0
0 .
3

20

LINEAR ALGEBRA

Now consider the important case where V = Fn , W = Fm , and the basis chosen is the standard basis
of vectors ei described above. Let L be a linear transformation from Fn to Fm and let A be the matrix of
the transformation with respect to these bases. In this case the coordinate maps qV and qW are simply the
identity map and we need
i (Lb) = i (Ab)
where i denotes the map which takes a vector in Fm and returns the ith entry in the vector, the ith
component of the vector with respect to the standard basis vectors. Thus, if the components of the vector
in Fn with respect to the standard basis are (b1 , , bn ) ,

T X
b = b 1 bn
=
b i ei ,
i

then
i (Lb) (Lb)i =

aij bj .

What about the situation where different pairs of bases are chosen for V and W ? How are the two matrices
with respect to these choices related? Consider the following diagram which illustrates the situation.
Fn
q2
V
q1
Fn

m
A

2 F
p2
L
W

p1
m
A

1 F

In this diagram qi and pi are coordinate maps as described above. We see from the diagram that
1
p1
1 p2 A2 q2 q1 = A1 ,

where q21 q1 and p1

1 p2 are one to one, onto, and linear maps.
In the special case where V = W and only one basis is used for V = W, this becomes
q11 q2 A2 q21 q1 = A1 .
Letting S be the matrix of the linear transformation q21 q1 with respect to the standard basis vectors in Fn ,
we get
S 1 A2 S = A1 .
When this occurs, we say that A1 is similar to A2 and we call A S 1 AS a similarity transformation.
Theorem 2.10 In the vector space of n n matrices, we say
AB
if there exists an invertible matrix S such that
A = S 1 BS.
Then is an equivalence relation and A B if and only if whenever V is an n dimensional vector space,
there exists L L (V, V ) and bases {v1 , , vn } and {w1 , , wn } such that A is the matrix of L with respect
to {v1 , , vn } and B is the matrix of L with respect to {w1 , , wn }.

2.2. LINEAR TRANSFORMATIONS

21

Proof: A A because S = I works in the definition. If A B , then B A, because

A = S 1 BS
implies
B = SAS 1 .
If A B and B C, then
A = S 1 BS, B = T 1 CT
and so
1

A = S 1 T 1 CT S = (T S)

CT S

which implies A C. This verifies the first part of the conclusion.

Now let V be an n dimensional vector space, A B and pick a basis for V,
{v1 , , vn }.
Define L L (V, V ) by
Lvi

aji vj

where A = (aij ) . Then if B = (bij ) , and S = (sij ) is the matrix which provides the similarity transformation,
A = S 1 BS,
between A and B, it follows that
Lvi =

sir brs s1

r,s,j

sj

vj .

(2.8)

Now define
wi

s1

s1

ij

vj .

Then from (2.8),

X
i

s1

ki

Lvi =

i,j,r,s

s b
ki ir rs

s1

sj

vj

and so
Lwk =

bks ws .

This proves the theorem because the if part of the conclusion was established earlier.

22

LINEAR ALGEBRA

2.3

Inner product spaces

Definition 2.11 A vector space X is said to be a normed linear space if there exists a function, denoted by
|| : X [0, ) which satisfies the following axioms.
1. |x| 0 for all x X, and |x| = 0 if and only if x = 0.
2. |ax| = |a| |x| for all a F.
3. |x + y| |x| + |y| .
Note that we are using the same notation for the norm as for the absolute value. This is because the
norm is just a generalization to vector spaces of the concept of absolute value. However, the notation ||x|| is
also often used. Not all norms are created equal. There are many geometric properties which they may or
may not possess. There is also a concept called an inner product which is discussed next. It turns out that
the best norms come from an inner product.
Definition 2.12 A mapping (, ) : V V F is called an inner product if it satisfies the following axioms.
1. (x, y) = (y, x).
2. (x, x) 0 for all x V and equals zero if and only if x = 0.
3. (ax + by, z) = a (x, z) + b (y, z) whenever a, b F.
Note that 2 and 3 imply (x, ay + bz) = a(x, y) + b(x, z).
We will show that if (, ) is an inner product, then
1/2

(x, x)

|x|

defines a norm.
Definition 2.13 A normed linear space in which the norm comes from an inner product as just described
is called an inner product space. A Hilbert space is a complete inner product space. Recall this means that
every Cauchy sequence,{xn } , one which satisfies
lim |xn xm | = 0,

n,m

converges.
Example 2.14 Let V = Cn with the inner product given by
(x, y)

n
X

xk y k .

k=1

. This is an example of a complex Hilbert space.

Example 2.15 Let V = Rn,
(x, y) = x y

n
X
j=1

. This is an example of a real Hilbert space.

xj yj .

2.3. INNER PRODUCT SPACES

23

Theorem 2.16 (Cauchy Schwartz) In any inner product space

|(x, y)| |x||y|.
1/2

where |x| (x, x)

Proof: Let C, || = 1, and (x, y) = |(x, y)| = Re(x, y). Let

F (t) = (x + ty, x + ty).
If y = 0 there is nothing to prove because
(x, 0) = (x, 0 + 0) = (x, 0) + (x, 0)
and so (x, 0) = 0. Thus, we may assume y 6= 0. Then from the axioms of the inner product,
F (t) = |x|2 + 2t Re(x, y) + t2 |y|2 0.
This yields
|x|2 + 2t|(x, y)| + t2 |y|2 0.
Since this inequality holds for all t R, it follows from the quadratic formula that
4|(x, y)|2 4|x|2 |y|2 0.
This yields the conclusion and proves the theorem.
Earlier it was claimed that the inner product defines a norm. In this next proposition this claim is proved.
1/2

Proposition 2.17 For an inner product space, |x| (x, x)

does specify a norm.

Proof: All the axioms are obvious except the triangle inequality. To verify this,
2

|x + y|

(x + y, x + y) |x| + |y| + 2 Re (x, y)

2

|x| + |y| + 2 |(x, y)|

|x| + |y| + 2 |x| |y| = (|x| + |y|) .

The best norms of all are those which come from an inner product because of the following identity which
is known as the parallelogram identity.
1/2

Proposition 2.18 If (V, (, )) is an inner product space then for |x| (x, x)
2

, the following identity holds.

|x + y| + |x y| = 2 |x| + 2 |y| .
It turns out that the validity of this identity is equivalent to the existence of an inner product which
determines the norm as described above. These sorts of considerations are topics for more advanced courses
on functional analysis.
Definition 2.19 We say a basis for an inner product space, {u1 , , un } is an orthonormal basis if

1 if k = j
(uk , uj ) = kj
.
0 if k 6= j

24

LINEAR ALGEBRA

Note that if a list of vectors satisfies the above condition for being an orthonormal set, then the list of
vectors is automatically linearly independent. To see this, suppose
n
X

cj uj = 0

j=1

Then taking the inner product of both sides with uk , we obtain

0=

n
X

cj (uj , uk ) =

j=1

n
X

cj jk = ck .

j=1

Lemma 2.20 Let X be a finite dimensional inner product space of dimension n. Then there exists an
orthonormal basis for X, {u1 , , un } .
Proof: Let {x1 , , xn } be a basis for X. Let u1 x1 / |x1 | . Now suppose for some k < n, u1 , , uk
have been chosen such that (uj , uk ) = jk . Then we define
Pk
xk+1 j=1 (xk+1 , uj ) uj
,
uk+1 
Pk

xk+1 j=1 (xk+1 , uj ) uj 
where the numerator is not equal to zero because the xj form a basis. Then if l k,

k
X
(uk+1 , ul ) = C (xk+1 , ul )
(xk+1 , uj ) (uj , ul )
j=1

= C (xk+1 , ul )

k
X
j=1

(xk+1 , uj ) lj

= C ((xk+1 , ul ) (xk+1 , ul )) = 0.
n

The vectors, {uj }j=1 , generated in this way are therefore an orthonormal basis.
The process by which these vectors were generated is called the Gramm Schmidt process.
n

Lemma 2.21 Suppose {uj }j=1 is an orthonormal basis for an inner product space X. Then for all x X,
x=

n
X

(x, uj ) uj .

j=1

Proof: By assumption that this is an orthonormal basis,

n
X

(x, uj ) (uj , ul ) = (x, ul ) .

j=1

Letting y =

Pn

j=1

(x, uj ) uj , it follows (x y, uj ) = 0 for all j. Hence, for any choice of scalars, c1 , , cn ,

n
X
x y,
cj uj = 0
j=1

and so (x y, z) = 0 for all z X. Thus this holds in particular for z = x y. Therefore, x = y and this
proves the theorem.
The next theorem is one of the most important results in the theory of inner product spaces. It is called
the Riesz representation theorem.

2.3. INNER PRODUCT SPACES

25

Theorem 2.22 Let f L (X, F) where X is a finite dimensional inner product space. Then there exists a
unique z X such that for all x X,
f (x) = (x, z) .
Proof: First we verify uniqueness. Suppose zj works for j = 1, 2. Then for all x X,
0 = f (x) f (x) = (x, z1 z2 )
and so z1 = z2 .
n
It remains to verify existence. By Lemma 2.20, there exists an orthonormal basis, {uj }j=1 . Define
z

n
X

f (uj )uj .

j=1

Then using Lemma 2.21,

(x, z)

= x,

n
X
j=1

f (uj )uj =

n
X

f (uj ) (x, uj )

j=1

n
X
= f
(x, uj ) uj = f (x) .
j=1

This proves the theorem.

Corollary 2.23 Let A L (X, Y ) where X and Y are two finite dimensional inner product spaces. Then
there exists a unique A L (Y, X) such that
(Ax, y)Y = (x, A y)X
for all x X and y Y.
Proof: Let fy L (X, F) be defined as
fy (x) (Ax, y)Y .
Then by the Riesz representation theorem, there exists a unique element of X, A (y) such that
(Ax, y)Y = (x, A (y))X .
It only remains to verify that A is linear. Let a and b be scalars. Then for all x X,
(x, A (ay1 + by2 ))X a (Ax, y1 ) + b (Ax, y2 )
a (x, A (y1 )) + b (x, A (y2 )) = (x, aA (y1 ) + bA (y2 )) .
By uniqueness, A (ay1 + by2 ) = aA (y1 ) + bA (y2 ) which shows A is linear as claimed.
The linear map, A is called the adjoint of A. In the case when A : X X and A = A , we call A a self
adjoint map. The next theorem will prove useful.
Theorem 2.24 Suppose V is a subspace of Fn having dimension p n. Then there exists a Q L (Fn , Fn )
such that QV Fp and |Qx| = |x| for all x. Also
Q Q = QQ = I.

26

LINEAR ALGEBRA
p

Proof: By Lemma 2.20 there exists an orthonormal basis for V, {vi }i=1 . By using the Gramm Schmidt
process we may extend this orthonormal basis to an orthonormal basis of the whole space, Fn ,
{v1 , , vp , vp+1 , , vn } .
Pn
Now define Q L (Fn , Fn ) by Q (vi ) ei and extend linearly. If i=1 xi vi is an arbitrary element of Fn ,


2
2
!2  n
n
n
n


X

X

X
X






2
xi vi  = 
xi ei  =
|xi | = 
xi vi  .
Q






i=1

i=1

i=1

i=1

It remains to verify that Q Q = QQ = I. To do so, let x, y F . Then

(Q (x + y) , Q (x + y)) = (x + y, x + y) .
Thus
2

|Qx| + |Qy| + Re (Qx,Qy) = |x| + |y| + Re (x, y)

and since Q preserves norms, it follows that for all x, y Fn ,
Re (Qx,Qy) = Re (x,Q Qy) = Re (x, y) .
Therefore, since this holds for all x, it follows that Q Qy = y showing that Q Q = I. Now
Q = Q (Q Q) = (QQ ) Q.
Since Q is one to one, this implies
I = QQ
and proves the theorem.
This case of a self adjoint map turns out to be very important in applications. It is also easy to discuss
the eigenvalues and eigenvectors of such a linear map. For A L (X, X) , we give the following definition of
eigenvalues and eigenvectors.
Definition 2.25 A non zero vector, y is said to be an eigenvector for A L (X, X) if there exists a scalar,
, called an eigenvalue, such that
Ay = y.
The important thing to remember about eigenvectors is that they are never equal to zero. The following
theorem is about the eigenvectors and eigenvalues of a self adjoint operator. The proof given generalizes to
the situation of a compact self adjoint operator on a Hilbert space and leads to many very useful results. It
is also a very elementary proof because it does not use the fundamental theorem of algebra and it contains
a way, very important in applications, of finding the eigenvalues. We will use the following notation.
Definition 2.26 Let X be an inner product space and let S X. Then
S {x X : (x, s) = 0 for all s S} .
Note that even if S is not a subspace, S is.
Definition 2.27 Let X be a finite dimensional inner product space and let A L (X, X). We say A is self
adjoint if A = A.

2.3. INNER PRODUCT SPACES

27

Theorem 2.28 Let A L (X, X) be self adjoint. Then there exists an orthonormal basis of eigenvectors,
n
{uj }j=1 .
Proof: Consider (Ax, x) . This quantity is always a real number because
(Ax, x) = (x, Ax) = (x, A x) = (Ax, x)
thanks to the assumption that A is self adjoint. Now define
1 inf {(Ax, x) : |x| = 1, x X1 X} .
Claim: 1 is finite and there exists v1 X with |v1 | = 1 such that (Av1 , v1 ) = 1 .
n
Proof: Let {uj }j=1 be an orthonormal basis for X and for x X, let (x1 , , xn ) be defined as the
components of the vector x. Thus,
x=

n
X

xj uj .

j=1

Since this is an orthonormal basis, it follows from the axioms of the inner product that
2

|x| =

n
X

|xj | .

j=1

Thus

(Ax, x) =

n
X

xk Auk ,

X
j=1

k=1

xj uj =

xk xj (Auk , uj ) ,

k,j

a continuous function of (x1 , , xn ). Thus this function achieves its minimum on the closed and bounded
subset of Fn given by
{(x1 , , xn ) Fn :

n
X

|xj | = 1}.

j=1

Pn

Then v1 j=1 xj uj where (x1 , , xn ) is the point of Fn at which the above function achieves its minimum.
This proves the claim.

Continuing with the proof of the theorem, let X2 {v1 } and let
2 inf {(Ax, x) : |x| = 1, x X2 }

As before, there exists v2 X2 such that (Av2 , v2 ) = 2 . Now let X2 {v1 , v2 } and continue in this way.
n
This leads to an increasing sequence of real numbers, {k }k=1 and an orthonormal set of vectors, {v1 , ,
vn }. It only remains to show these are eigenvectors and that the j are eigenvalues.
Consider the first of these vectors. Letting w X1 X, the function of the real variable, t, given by
f (t)

(A (v1 + tw) , v1 + tw)

2

|v1 + tw|

(Av1 , v1 ) + 2t Re (Av1 , w) + t2 (Aw, w)

2

|v1 | + 2t Re (v1 , w) + t2 |w|

28

LINEAR ALGEBRA

achieves its minimum when t = 0. Therefore, the derivative of this function evaluated at t = 0 must equal
zero. Using the quotient rule, this implies
2 Re (Av1 , w) 2 Re (v1 , w) (Av1 , v1 )
= 2 (Re (Av1 , w) Re (v1 , w) 1 ) = 0.
Thus Re (Av1 1 v1 , w) = 0 for all w X. This implies Av1 = 1 v1 . To see this, let w X be arbitrary
and let be a complex number with || = 1 and
|(Av1 1 v1 , w)| = (Av1 1 v1 , w) .
Then


|(Av1 1 v1 , w)| = Re Av1 1 v1 , w = 0.
Since this holds for all w, Av1 = 1 v1 . Now suppose Avk = k vk for all k < m. We observe that A : Xm Xm
because if y Xm and k < m,
(Ay, vk ) = (y, Avk ) = (y, k vk ) = 0,

showing that Ay {v1 , , vm1 } Xm . Thus the same argument just given shows that for all w Xm ,
(Avm m vm , w) = 0.

(2.9)

For arbitrary w X.
w=

m1
X

(w, vk ) vk

k=1

m1
X

(w, vk ) vk w + wm

k=1

and the term in parenthesis is in {v1 , , vm1 } Xm while the other term is contained in the span of the
vectors, {v1 , , vm1 }. Thus by (2.9),
(Avm m vm , w) = (Avm m vm , w + wm )
= (Avm m vm , wm ) = 0
because
A : Xm Xm {v1 , , vm1 }

and wm span (v1 , , vm1 ) . Therefore, Avm = m vm for all m. This proves the theorem.
When a matrix has such a basis of eigenvectors, we say it is non defective.
There are more general theorems of this sort involving normal linear transformations. We say a linear
transformation is normal if AA = A A. It happens that normal matrices are non defective. The proof
involves the fundamental theorem of algebra and is outlined in the exercises.
As an application of this theorem, we give the following fundamental result, important in geometric
measure theory and continuum mechanics. It is sometimes called the right polar decomposition.
Theorem 2.29 Let F L (Rn , Rm ) where m n. Then there exists R L (Rn , Rm ) and U L (Rn , Rn )
such that
F = RU, U = U ,
all eigen values of U are non negative,
U 2 = F F, R R = I,
and |Rx| = |x| .

2.3. INNER PRODUCT SPACES

29

Proof: (F F ) = F F and so by linear algebra there is an orthonormal basis of eigenvectors, {v1 , , vn }

such that
F F vi = i vi .
It is also clear that i 0 because
i (vi , vi ) = (F F vi , vi ) = (F vi , F vi ) 0.
Now if u, v Rn , we define the tensor product u v L (Rn , Rn ) by
u v (w) (w, v) u.
Then F F =

Pn

i=1

i vi vi because both linear transformations agree on the basis {v1 , , vn }. Let

U

n
X

1/2

i vi vi .

i=1

on
n
1/2
Then U 2 = F F, U = U , and the eigenvalues of U, i

are all nonnegative.

i=1

Now R is defined on U (R ) by
RU x F x.
This is well defined because if U x1 = U x2 , then U 2 (x1 x2 ) = 0 and so
2

0 = (F F (x1 x2 ) , x1 x2 ) = |F (x1 x2 )| .
2

Now |RU x| = |U x| because

2

|RU x| = |F x| = (F x,F x) = (F F x, x)


2
= U 2 x, x = (U x,U x) = |U x| .
Let {x1 , , xr } be an orthonormal basis for

U (Rn ) {x Rn : (x, z) = 0 for all z U (Rn )}

and let {y1 , , yp } be an orthonormal basis for F (Rn ) . Then p r because if {F (zi )}i=1 is an orthonormal
s
basis for F (Rn ), it follows that {U (zi )}i=1 is orthonormal in U (Rn ) because


(U zi , U zj ) = U 2 zi , zj = (F F zi , zj ) = (F zi , F zj ).
Therefore,
p + s = m n = r + dim U (Rn ) r + s.
Now define R L (Rn , Rm ) by Rxi yi , i = 1, , r. Thus

2
!2  r
r
r
X



X
X




2
2
ci yi + F v =
|ci | + |F v|
ci xi + U v  = 
R




i=1

i=1

r
X
i=1

i=1

 r
2
X



2
2
|ci | + |U v| = 
ci xi + U v ,


i=1

30

LINEAR ALGEBRA

and so |Rz| = |z| which implies that for all x, y,

2

|x| + |y| + 2 (x, y) = |x + y| = |R (x + y)|

2

= |x| + |y| + 2 (Rx,Ry).

Therefore,
(x, y) = (R Rx, y)
for all x, y and so R R = I as claimed. This proves the theorem.

2.4

Exercises

1. Show (A ) = A and (AB) = B A .

2. Suppose A : X X, an inner product space, and A 0. By this we mean (Ax, x) 0 for all x X and
n
A = A . Show that A has a square root, U, such that U 2 = A. Hint: Let {uk }k=1 be an orthonormal
basis of eigenvectors with Auk = k uk . Show each k 0 and consider
U

n
X

1/2

k u k uk

k=1

3. In the context of Theorem 2.29, suppose m n. Show

F = UR
where
U L (Rm , Rm ) , R L (Rn , Rm ) , U = U
, U has all non negative eigenvalues, U 2 = F F , and RR = I. Hint: This is an easy corollary of
Theorem 2.29.
4. Show that if X is the inner product space Fn , and A is an n n matrix, then
A = AT .
5. Show that if A is an n n matrix and A = A then all the eigenvalues and eigenvectors are real
and that eigenvectors associated with distinct eigenvalues are orthogonal, (their inner product is zero).
Such matrices are called Hermitian.
n

6. Let the orthonormal basis of eigenvectors of F F be denoted by {vi }i=1 where {vi }i=1 are those
n
whose eigenvalues are positive and {vi }i=r are those whose eigenvalues equal zero. In the context of
n
the RU decomposition for F, show {Rvi }i=1 is also an orthonormal basis. Next verify that there exists a
r

solution, x, to the equation, F x = b if and only if b span {Rvi }i=1 if and only is b (ker F ) . Here

ker F {x : F x = 0} and (ker F ) {y : (y, x) = 0 for all x ker F } . Hint: Show that F x = 0
n
n
if and only if U R x = 0 if and only if R x span {vi }i=r+1 if and only if x span {Rvi }i=r+1 .
7. Let A and B be n n matrices and let the columns of B be
b1 , , bn

2.5. DETERMINANTS

31

and the rows of A are

aT1 , , aTn .
Show the columns of AB are
Ab1 Abn
and the rows of AB are
aT1 B aTn B.
8. Let v1 , , vn be an orthonormal basis for Fn . Let Q be a matrix whose ith column is vi . Show
Q Q = QQ = I.
such a matrix is called an orthogonal matrix.
9. Show that a matrix, Q is orthogonal if and only if it preserves distances. By this we mean |Qv| = |v| .
1/2
Here |v| (v v)
for the dot product defined above.
10. Suppose {v1 , , vn } and {w1 , , wn } are two orthonormal bases for Fn and suppose Q is an n n
matrix satisfying Qvi = wi . Then show Q is orthogonal. If |v| = 1, show there is an orthogonal
transformation which maps v to e1 .
11. Let A be a Hermitian matrix so A = A and suppose all eigenvalues of A are larger than 2 . Show
2

(Av, v) 2 |v|
Where here, the inner product is
(v, u)

n
X

vj uj .

j=1

12. Let L L (Fn , Fm ) . Let {v1 , , vn } be a basis for V and let {w1 , , wm } be a basis for W. Now
define w v L (V, W ) by the rule,
w v (u) (u, v) w.
Show w v L (Fn , Fm ) as claimed and that
{wj vi : i = 1, , n, j = 1, , m}
is a basis for L (Fn , Fm ) . Conclude the dimension of L (Fn , Fm ) is nm.
2

13. Let X be an inner product space. Show |x + y| + |x y| = 2 |x| + 2 |y| . This is called the parallelogram identity.

2.5

Determinants

Here we give a discussion of the most important properties of determinants. There are more elegant ways
to proceed and the reader is encouraged to consult a more advanced algebra book to read these. Another
very good source is Apostol [2]. The goal here is to present all of the major theorems on determinants with
a minimum of abstract algebra as quickly as possible. In this section and elsewhere F will denote the field
of scalars, usually R or C. To begin with we make a simple definition.

32

LINEAR ALGEBRA

Definition 2.30 Let (k1 , , kn ) be an ordered list of n integers. We define

Y
(k1 , , kn )
{(ks kr ) : r < s} .
In words, we consider all terms of the form (ks kr ) where ks comes after kr in the ordered list and then
multiply all these together. We also make the following definition.

1 if (k1 , , kn ) > 0
1 if (k1 , , kn ) < 0
sgn (k1 , , kn )

0 if (k1 , , kn ) = 0


This is called the sign of the permutation k1n
in the case when there are no repeats in the ordered list,
1 kn
(k1 , , kn ) and {k1 , , kn } = {1, , n}.
Lemma 2.31 Let (k1 , ki , kj , , kn ) be a list of n integers. Then
(k1 , , ki , kj , , kn ) = (k1 , , kj , ki , , kn )
and
sgn (k1 , , ki , kj , , kn ) = sgn (k1 , , kj , ki , , kn )
In words, if we switch two entries the sign changes.
Proof: The two lists are
(k1 , , ki , , kj , , kn )

(2.10)

(k1 , , kj , , ki , , kn ) .

(2.11)

and

Suppose there are r 1 numbers between ki and kj in the first list and consequently r 1 numbers between
kj and ki in the second list. In computing (k1 , , ki , , kj , , kn ) we have r 1 terms of the form
(kj kp ) where the kp are those numbers occurring in the list between ki and kj . Corresponding to these
terms we have r 1 terms of the form (kp kj ) in the computation of (k1 , , kj , , ki , , kn ). These
r1
differences produce a (1)
in going from (k1 , , ki , , kj , , kn ) to (k1 , , kj , , ki , , kn ) . We
also have the r 1 terms (kp ki ) in computing (k1 , , ki , kj , , kn ) and the r 1 terms, (ki kp ) in
r1
computing (k1 , , kj , , ki , , kn ), producing another (1)
. Thus, in considering the differences in
, we see these terms just considered do not change the sign. However, we have (kj ki ) in the first product
and (ki kj ) in the second and all other factors in the computation of match up in the two computations
so it follows (k1 , , ki , , kj , , kn ) = (k1 , , kj , , ki , , kn ) as claimed.
Corollary 2.32 Suppose (k1 , , kn ) is obtained by making p switches in the ordered list, (1, , n) . Then
p

(1) = sgn (k1 , , kn ) .

(2.12)

Proof: We observe that sgn (1, , n) = 1 and according to Lemma 2.31, each time we switch two
p
p
entries we multiply by (1) . Therefore, making the p switches, we obtain (1) = (1) sgn (1, , n) =
sgn (k1 , , kn ) as claimed.
We now are ready to define the determinant of an n n matrix.
Definition 2.33 Let (aij ) = A denote an n n matrix. We define
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 ordererd lists in which there are no repeats because if there are, we know sgn (k1 , , kn ) = 0.

2.5. DETERMINANTS

33

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
(2.13)
(k1 ,,kn )

and
A (1, , n) = A.
Proposition 2.34 Let (r1 , , rn ) be an ordered list of numbers from {1, , n}. Then
X
sgn (r1 , , rn ) det (A) =
sgn (k1 , , kn ) ar1 k1 arn kn

(2.14)

(k1 ,,kn )

det (A (r1 , , rn )) .

(2.15)

In words, if we take the determinant of the matrix obtained by letting the pth row be the rp row of A, then
the determinant of this modified matrix equals the expression on the left in (2.14).
Proof: Let (1, , n) = (1, , r, s, , n) so r < s.
det (A (1, , r, s, , n)) =

(2.16)

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

(k1 ,,kn )

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

(k1 ,,kn )

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

(k1 ,,kn )

= det (A (1, , s, , r, , n)) .

(2.17)

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, we eventually arrive at
the conclusion
p

det (A (r1 , , rn )) = (1) det (A)

where it took p switches to obtain(r1 , , rn ) from (1, , n) . By Corollary 2.32 this implies
det (A (r1 , , rn )) = 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 (2.16) -(2.17) shows
that A (r1 , , rn ) = 0 and we also know that sgn (r1 , , rn ) = 0 so the formula holds in this case also.

34

LINEAR ALGEBRA

Corollary 2.35 We have the following formula for det (A) .

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

(2.18)

(r1 ,,rn ) (k1 ,,kn )

And also det AT = det (A) where AT is the transpose of A. Thus if AT = aTij , we have aTij = aji .
Proof: From Proposition 2.34, if the ri are distinct,
X
det (A) =
sgn (r1 , , rn ) sgn (k1 , , kn ) ar1 k1 arn kn .
(k1 ,,kn )

Summing over all ordered lists, (r1 , , rn ) where the ri are distinct, (If the ri are not distinct, we know
sgn (r1 , , rn ) = 0 and so there is no contribution to the sum.) we obtain
X
X
sgn (r1 , , rn ) sgn (k1 , , kn ) ar1 k1 arn kn .
n! det (A) =
(r1 ,,rn ) (k1 ,,kn )

This proves the corollary.

Corollary 2.36 If we switch two rows or two columns in an nn matrix, A, 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.
Proof: By Proposition 2.34 when we switch two rows the determinant of the resulting matrix is (1)
times the determinant of the original matrix. By Corollary 2.35 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, then




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.
Definition 2.37 If A and B are n n matrices, A = (aij ) and B = (bij ) , we form the product, AB = (cij )
by defining
cij

n
X

aik bkj .

k=1

This is just the usual rule for matrix multiplication.

One of the most important rules about determinants is that the determinant of a product equals the
product of the determinants.
Theorem 2.38 Let A and B be n n matrices. Then det (AB) = det (A) det (B) .
Proof: We will denote by cij the ij th entry of AB. Thus by Proposition 2.34,
X
det (AB) =
sgn (k1 , , kn ) c1k1 cnkn
(k1 ,,kn )

sgn (k1 , , kn )

r1

(k1 ,,kn )

a1r1 br1 k1

anrn brn kn

rn

sgn (k1 , , kn ) br1 k1 brn kn (a1r1 anrn )

(r1 ,rn ) (k1 ,,kn )

(r1 ,rn )

sgn (r1 rn ) a1r1 anrn det (B) = det (A) det (B) .

2.5. DETERMINANTS

35

This proves the theorem.

In terms of the theory of determinants, arguably the most important idea is that of Laplace expansion
along a row or a column.
Definition 2.39 Let A = (aij ) be an n n matrix. Then we define a new matrix, cof (A) by cof (A) = (cij )
where to obtain cij we delete the ith row and the j th column of A, take the determinant of the n 1 n 1
i+j
matrix which results and then multiply this number by (1) . The determinant of the n 1 n 1 matrix
th
just described is called the ij minor of A. To make the formulas easier to remember, we shall write cof (A)ij
for the ij th entry of the cofactor matrix.
The main result is the following monumentally important theorem. It states that you can expand an
n n matrix along any row or column. This is often taken as a definition in elementary courses but how
anyone in their right mind could believe without a proof that you always get the same answer by expanding
along any row or column is totally beyond my powers of comprehension.
Theorem 2.40 Let A be an n n matrix. Then
det (A) =

n
X

aij cof (A)ij =

j=1

n
X

aij cof (A)ij .

(2.19)

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: We will prove this by using the definition and then doing a computation and verifying that we
have what we want.
X
det (A) =
sgn (k1 , , kr , , kn ) a1k1 arkr ankn
(k1 ,,kn )

n
X

kr =1

(k1 ,,kr ,,kn )

sgn (k1 , , kr , , kn ) a1k1 a(r1)k(r1) a(r+1)k(r+1) ankn arkr

n
X

r1

(1)

j=1

(k1 ,,j,,kn )

sgn (j, k1 , , kr1 , kr+1 , kn ) a1k1 a(r1)k(r1) a(r+1)k(r+1) ankn arj .

We need to consider for fixed j the term

X
sgn (j, k1 , , kr1 , kr+1 , kn ) a1k1 a(r1)k(r1) a(r+1)k(r+1) ankn .

(2.20)

(2.21)

(k1 ,,j,,kn )

We may assume all the indices in (k1 , , j, , kn ) are distinct. We define (l1 , , ln1 ) as follows. If
k < j, then l k . If k > j, then l k 1. Thus every choice of the ordered list, (k1 , , j, , kn ) ,
corresponds to an ordered list, (l1 , , ln1 ) of indices from {1, , n 1}. Now define

ak if < r,
bl
a(+1)k if n 1 > r

36

LINEAR ALGEBRA

where here k corresponds to l as just described. Thus (b ) is the n 1 n 1 matrix which results from
deleting the rth row and the j th column. In computing
(j, k1 , , kr1 , kr+1 , kn ) ,
we note there are exactly j 1 of the ki which are less than j. Therefore,
j1

sgn (k1 , , kr1 , kr+1 , kn ) (1)

= sgn (j, k1 , , kr1 , kr+1 , kn ) .

But it also follows from the definition that

sgn (k1 , , kr1 , kr+1 , kn ) = sgn (l1 , ln1 )
and so the term in (2.21) equals
X

j1

(1)

sgn (l1 , , ln1 ) b1l1 b(n1)l(n1)

(l1 ,,ln1 )

Using this in (2.20) we see

det (A)

n
X

r1

(1)

j1

(1)

j=1

(l1 ,,ln1 )

n
X

(1)

n
X

arj cof (A)rj

r+j

j=1

sgn (l1 , , ln1 ) b1l1 b(n1)l(n1) arj

(l1 ,,ln1 )

sgn (l1 , , ln1 ) b1l1 b(n1)l(n1) arj

j=1

as claimed. Now to get the second half of (2.19), we can apply the first part to AT and write for AT = aTij
det (A)

n

X


= det AT =
aTij cof AT ij
j=1

n
X
j=1

aji cof (A)ji =

n
X

aij cof (A)ij .

i=1

This proves the theorem. We leave it as an exercise to show that cof AT ij = cof (A)ji .
Note that this gives us an easy way to write a formula for the inverse of an n n matrix.
Definition 2.41 We say an n n matrix, A has an inverse, A1 if and only if AA1 = A1 A = I where
I = ( ij ) for

1 if i = j
ij
0 if i 6= j


Theorem 2.42 A1 exists if and only if det(A) 6= 0. If det(A) 6= 0, then A1 = a1
where
ij
1
a1
Cji
ij = det(A)

for Cij the ij th cofactor of A.

2.5. DETERMINANTS

37

Proof: By Theorem 2.40 and letting (air ) = A, if we assume det (A) 6= 0,

n
X

air Cir det(A)1 = det(A) det(A)1 = 1.

i=1

Now we consider
n
X

air Cik det(A)1

i=1

when k 6= r. We replace the k th column with the rth column to obtain a matrix, Bk whose determinant
equals zero by Corollary 2.36. However, expanding this matrix along the k th column yields
1

0 = det (Bk ) det (A)

n
X

air Cik det (A)

i=1

Summarizing,
n
X

air Cik det (A)

= rk .

i=1

Using the other formula in Theorem 2.40, we can also write using similar reasoning,
n
X

arj Ckj det (A)

= rk

j=1

This proves that if det (A) 6= 0, then A1 exists and if A1 = a1
,
ij
1

a1
ij = Cji det (A)

Now suppose A1 exists. Then by Theorem 2.38,

1 = det (I) = det AA1 = det (A) det A1
so det (A) 6= 0. This proves the theorem.
This theorem says that to find the inverse, we can 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. The term, adjoint, should be reserved
for something much more interesting which will be discussed later. In words, A1 is equal to one over the
determinant of A times the adjugate matrix of A.
In case we are solving a system of equations,
Ax = y
for x, it follows that if A1 exists, we can write


x = A1 A x = A1 (Ax) = A1 y
thus solving the system. Now in the case that A1 exists, we just presented a formula for A1 . Using this
formula, we see
xi =

n
X
j=1

a1
ij yj =

n
X
j=1

1
cof (A)ji yj .
det (A)

38

LINEAR ALGEBRA

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

y1
1
..
..
xi =
det .
.
det (A)
yn

.. ,
.

where here we have replaced the ith column of A with the column vector, (y1 , yn ) , taken its determinant
and divided by det (A) . This formula is known as Cramers rule.
Definition 2.43 We say 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, we give the following corollary of Theorem 2.40.
Corollary 2.44 Let M be an upper (lower) triangular matrix. Then det (M ) is obtained by taking the
product of the entries on the main diagonal.

2.6

The characteristic polynomial

Definition 2.45 Let A be an n n matrix. The characteristic polynomial is defined as

pA (t) det (tI A) .
A principal submatrix of A is one lying in the same

set of k rows and columns and a principal minor is the
determinant of a principal submatrix. There are nk principal minors of A. How do we get a typical principal
submatrix? We pick k rows, say r1 , , rk and consider the k k matrix which results from using exactly
those entries of these k rows which are also in one of the r1 , , rk columns. We denote by Ek (A) the sum
of the principal k k minors of A.
We write a formula for the characteristic polynomial in terms of the Ek (A) .
X
pA (t) =
sgn (k1 , , kn ) (t 1k1 a1k1 ) (t 1kn a1kn )
(k1 ,,kn )

Consider the terms which are multiplied by tr . A typical such term would be
X
nr
tr (1)
sgn (k1 , , kn ) m1 km1 mr kmr as1 ks1 as(nr) ks(nr)

(2.22)

(k1 ,,kn )

where {m1 , , mr , s1 , , snr } = {1, , n} .

expression is the determinant of a matrix like

From the definition of determinant, the sum in the above

0

2.6. THE CHARACTERISTIC POLYNOMIAL

39

where the starred rows are simply the original rows of the matrix, A. Using the row operation which involves
replacing a row with a multiple of another row added to itself, we can use the ones to zero out everything
above them and below them, obtaining a modified matrix which has the same determinant (See Problem 2).
In the given example this would result in a matrix of the form

1 0 0 0 0
0 0 0

0 0 1 0 0

0 0 0
0 0 0 0 1
and so the sum in (2.22)
is just the principal minor corresponding to the subset {m1 , , mr } of {1, , n} .


For each of the nr such choices, there is such a term equal to the principal minor determined in this
way and so the sum of these equals the coefficient of the tr term. Therefore, the coefficient of tr equals
nr
(1)
Enr (A) . It follows
pA (t)

n
X

nr

tr (1)

Enr (A)

r=0

n1

(1) En (A) + (1)

tEn1 (A) + + (1) tn1 E1 (A) + tn .

Definition 2.46 The solutions to pA (t) = 0 are called the eigenvalues of A.

We know also that
pA (t) =

n
Y

(t k )

k=1

where k are the roots of the equation, pA (t) = 0. (Note these might be complex numbers.) Therefore,
expanding the above polynomial,
Ek (A) = Sk (1 , , n )
where Sk (1 , , n ) , called the k th elementary symmetric function of the numbers 1 , , n , is defined as
the sum of all possible products of k elements of {1 , , n } . Therefore,
pA (t) = tn S1 (1 , , n ) tn1 + S2 (1 , , n ) tn2 + Sn (1 , , n ) .
A remarkable and profound theorem is the Cayley Hamilton theorem which states that every matrix
satisfies its characteristic equation. We give a simple proof of this theorem using the following lemma.
Lemma 2.47 Suppose for all || large enough, we have
A0 + A1 + + Am m = 0,
where the Ai are n n matrices. Then each Ai = 0.
Proof: Multiply by m to obtain
A0 m + A1 m+1 + + Am1 1 + Am = 0.
Now let || to obtain Am = 0. With this, multiply by to obtain
A0 m+1 + A1 m+2 + + Am1 = 0.
Now let || to obtain Am1 = 0. Continue multiplying by and letting to obtain that all the
Ai = 0. This proves the lemma.
With the lemma, we have the following simple corollary.

40

LINEAR ALGEBRA

Corollary 2.48 Let Ai and Bi be n n matrices and suppose

A0 + A1 + + Am m = B0 + B1 + + Bm m
for all || large enough. Then Ai = Bi for all i.
Proof: Subtract and use the result of the lemma.
With this preparation, we can now give an easy proof of the Cayley Hamilton theorem.
Theorem 2.49 Let A be an n n matrix and let p () det (I A) be the characteristic polynomial.
Then p (A) = 0.
Proof: Let C () equal the transpose of the cofactor matrix of (I A) for || large. (If || is large
1
enough, then cannot be in the finite list of eigenvalues of A and so for such , (I A) exists.) Therefore,
by Theorem 2.42 we may write
1

C () = p () (I A)

Note that each entry in C () is a polynomial in having degree no more than n 1. Therefore, collecting
the terms, we may write
C () = C0 + C1 + + Cn1 n1
for Cj some n n matrix. It follows that for all || large enough,


(A I) C0 + C1 + + Cn1 n1 = p () I
and so we are in the situation of Corollary 2.48. It follows the matrix coefficients corresponding to equal
powers of are equal on both sides of this equation.Therefore, we may replace with A and the two will be
equal. Thus


0 = (A A) C0 + C1 A + + Cn1 An1 = p (A) I = p (A) .
This proves the Cayley Hamilton theorem.

2.7

The rank of a matrix

Definition 2.50 Let A be an m n matrix. Then the row rank is the dimension of the span of the rows in
Fn , the column rank is the dimension of the span of the columns, and the determinant rank equals r where r
is the largest number such that some r r submatrix of A has a non zero determinant.Note the column rank
of A is nothing more than the dimension of A (Fn ) .
Theorem 2.51 The determinant rank, row rank, and column rank coincide.
Proof: Suppose the determinant rank of A = (aij ) equals r. First note that if rows and columns are
interchanged, the row, column, and determinant ranks of the modified matrix are unchanged. Thus we may
assume without loss of generality that there is an r r matrix in the upper left corner of the matrix which
has non zero determinant. Consider the matrix

a11 a1r a1p

..

..
..
.

.
.

ar1 arr arp

al1 alr alp

2.8. EXERCISES

41

where we will denote by C the r r matrix which has non zero determinant. The above matrix has
determinant equal to zero. 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
alp =

r
X
i=1

Cip
aip ,
det (C)

where Cip is the cofactor of aip . Now note that Cip does not depend on p. Therefore the above sum is of the
form
alp =

r
X

mi aip

i=1

which shows the lth row is a linear combination of the first r rows of A. Thus the first r rows of A are linearly
independent and span the row space so the row rank equals r. It follows from this that
column rank of A = row rank of AT =
= determinant rank of AT = determinant rank of A = row rank of A.
This proves the theorem.

2.8

Exercises

1. Let A L (V, V ) where V is a vector space of dimension n. Show using the fundamental theorem of
algebra which states that every non constant polynomial has a zero in the complex numbers, that A
has an eigenvalue and eigenvector. Recall that (, v) is an eigen pair if v 6= 0 and (A I) (v) = 0.
2. Show that if we replace a row (column) of an n n matrix A with itself added to some multiple of
another row (column) then the new matrix has the same determinant as the original one.
3. Let A be an n n matrix and let
Av1 = 1 v1 , |v1 | = 1.
Show there exists an orthonormal basis, {v1 , , vn } for Fn . Let Q0 be a matrix whose ith column is
vi . Show Q0 AQ0 is of the form

1
0

..

A
1

where A1 is an n 1 n 1 matrix.
e 1 such that Q
e A1 Q
e 1 is of the
4. Using the result of problem 3, show there exists an orthogonal matrix Q
1
form

2
0

..
.
.

A
2

42

LINEAR ALGEBRA
Now let Q1 be the n n matrix of the form

Show Q1 Q0 AQ0 Q1 is of the form

1 0
e1
0 Q

1
0
0
..
.

2
0
..
.

A2

where A2 is an n 2 n 2 matrix. Continuing in this way, show there exists an orthogonal matrix
Q such that
Q AQ = T
where T is upper triangular. This is called the Schur form of the matrix.
5. Let A be an m n matrix. Show the column rank of A equals the column rank of A A. Next verify
column rank of A A is no larger than column rank of A . Next justify the following inequality to
conclude the column rank of A equals the column rank of A .
rank (A) = rank (A A) rank (A )
= rank (AA ) rank (A) .
r

Hint: Start with an orthonormal basis, {Axj }j=1 of A (Fn ) and verify {A Axj }j=1 is a basis for
A A (Fn ) .
6. Show the i on the main diagonal of T in problem 4 are the eigenvalues of A.
7. We say A is normal if
A A = AA .
Show that if A = A, then A is normal. Show that if A is normal and Q is an orthogonal matrix, then
Q AQ is also normal. Show that if T is upper triangular and normal, then T is a diagonal matrix.
Conclude the Shur form of every normal matrix is diagonal.
8. If A is such that there exists an orthogonal matrix, Q such that
Q AQ = diagonal matrix,
is it necessary that A be normal? (We know from problem 7 that if A is normal, such an orthogonal
matrix exists.)

General topology
This chapter is a brief introduction to general topology. Topological spaces consist of a set and a subset of
the set of all subsets of this set called the open sets or topology which satisfy certain axioms. Like other
areas in mathematics the abstraction inherent in this approach is an attempt to unify many different useful
examples into one general theory.
For example, consider Rn with the usual norm given by
|x|

n
X

|xi |

!1/2

i=1

We say a set U in Rn is an open set if every point of U is an interior point which means that if x U ,
there exists > 0 such that if |y x| < , then y U . It is easy to see that with this definition of open sets,
the axioms (3.1) - (3.2) given below are satisfied if is the collection of open sets as just described. There
are many other sets of interest besides Rn however, and the appropriate definition of open set may be
very different and yet the collection of open sets may still satisfy these axioms. By abstracting the concept
of open sets, we can unify many different examples. Here is the definition of a general topological space.
Let X be a set and let be a collection of subsets of X satisfying
, X ,

(3.1)

If C , then C
If A, B , then A B .

(3.2)

Definition 3.1 A set X together with such a collection of its subsets satisfying (3.1)-(3.2) is called a topological space. is called the topology or set of open sets of X. Note P(X), the set of all subsets of X,
also called the power set.
Definition 3.2 A subset B of is called a basis for if whenever p U , there exists a set B B such
that p B U . The elements of B are called basic open sets.
The preceding definition implies that every open set (element of ) may be written as a union of basic
open sets (elements of B). This brings up an interesting and important question. If a collection of subsets B
of a set X is specified, does there exist a topology for X satisfying (3.1)-(3.2) such that B is a basis for ?
Theorem 3.3 Let X be a set and let B be a set of subsets of X. Then B is a basis for a topology if and
only if whenever p B C for B, C B, there exists D B such that p D C B and B = X. In this
case consists of all unions of subsets of B.
43

44

GENERAL TOPOLOGY

Proof: The only if part is left to the reader. Let consist of all unions of sets of B and suppose B satisfies
the conditions of the proposition. Then because B. X because B = X by assumption. If
C then clearly C . Now suppose A, B , A = S, B = R, S, R B. We need to show
A B . If A B = , we are done. Suppose p A B. Then p S R where S S, R R. Hence
there exists U B such that p U S R. It follows, since p A B was arbitrary, that A B = union
of sets of B. Thus A B . Hence satisfies (3.1)-(3.2).
Definition 3.4 A topological space is said to be Hausdorff if whenever p and q are distinct points of X,
there exist disjoint open sets U, V such that p U, q V .

Hausdorff
Definition 3.5 A subset of a topological space is said to be closed if its complement is open. Let p be a
point of X and let E X. Then p is said to be a limit point of E if every open set containing p contains a
point of E distinct from p.
Theorem 3.6 A subset, E, of X is closed if and only if it contains all its limit points.
Proof: Suppose first that E is closed and let x be a limit point of E. We need to show x E. If x
/ E,
then E C is an open set containing x which contains no points of E, a contradiction. Thus x E. Now
suppose E contains all its limit points. We need to show the complement of E is open. But if x E C , then
x is not a limit point of E and so there exists an open set, U containing x such that U contains no point of
E other than x. Since x
/ E, it follows that x U E C which implies E C is an open set.
Theorem 3.7 If (X, ) is a Hausdorff space and if p X, then {p} is a closed set.
C

Proof: If x 6= p, there exist open sets U and V such that x U, p V and U V = . Therefore, {p}
is an open set so {p} is closed.
Note that the Hausdorff axiom was stronger than needed in order to draw the conclusion of the last
theorem. In fact it would have been enough to assume that if x 6= y, then there exists an open set containing
x which does not intersect y.
Definition 3.8 A topological space (X, ) is said to be regular if whenever C is a closed set and p is a point
not in C, then there exist disjoint open sets U and V such that p U, C V . The topological space, (X, )
is said to be normal if whenever C and K are disjoint closed sets, there exist disjoint open sets U and V
such that C U, K V .

C
Regular

V
C

K
Normal

45
is defined to be the smallest closed set containing E. Note that
Definition 3.9 Let E be a subset of X. E
this is well defined since X is closed and the intersection of any collection of closed sets is closed.
Theorem 3.10 E = E {limit points of E}.
Proof: Let x E and suppose that x
/ E. If x is not a limit point either, then there exists an open
set, U,containing x which does not intersect E. But then U C is a closed set which contains E which does
not contain x, contrary to the definition that E is the intersection of all closed sets containing E. Therefore,
x must be a limit point of E after all.
Now E E so suppose x is a limit point of E. We need to show x E. If H is a closed set containing
E, which does not contain x, then H C is an open set containing x which contains no points of E other than
x negating the assumption that x is a limit point of E.
Definition 3.11 Let X be a set and let d : X X [0, ) satisfy
d(x, y) = d(y, x),

(3.3)

d(x, y) + d(y, z) d(x, z), (triangle inequality)

d(x, y) = 0 if and only if x = y.

(3.4)

Such a function is called a metric. For r [0, ) and x X, define

B(x, r) = {y X : d(x, y) < r}
This may also be denoted by N (x, r).
Definition 3.12 A topological space (X, ) is called a metric space if there exists a metric, d, such that the
sets {B(x, r), x X, r > 0} form a basis for . We write (X, d) for the metric space.
Theorem 3.13 Suppose X is a set and d satisfies (3.3)-(3.4). Then the sets {B(x, r) : r > 0, x X} form
a basis for a topology on X.
Proof: We observe that the union of these balls includes the whole space, X. We need to verify the
condition concerning the intersection of two basic sets. Let p B (x, r1 ) B (z, r2 ) . Consider
r min (r1 d (x, p) , r2 d (z, p))
and suppose y B (p, r) . Then
d (y, x) d (y, p) + d (p, x) < r1 d (x, p) + d (x, p) = r1
and so B (p, r) B (x, r1 ) . By similar reasoning, B (p, r) B (z, r2 ) . This verifies the conditions for this set
of balls to be the basis for some topology.
Theorem 3.14 If (X, ) is a metric space, then (X, ) is Hausdorff, regular, and normal.
Proof: It is obvious that any metric space is Hausdorff. Since each point is a closed set, it suffices to
verify any metric space is normal. Let H and K be two disjoint closed nonempty sets. For each h H, there
exists rh > 0 such that B (h, rh ) K = because K is closed. Similarly, for each k K there exists rk > 0
such that B (k, rk ) H = . Now let
U {B (h, rh /2) : h H} , V {B (k, rk /2) : k K} .
then these open sets contain H and K respectively and have empty intersection for if x U V, then
x B (h, rh /2) B (k, rk /2) for some h H and k K. Suppose rh rk . Then
d (h, k) d (h, x) + d (x, k) < rh ,
a contradiction to B (h, rh ) K = . If rk rh , the argument is similar. This proves the theorem.

46

GENERAL TOPOLOGY

Definition 3.15 A metric space is said to be separable if there is a countable dense subset of the space.
This means there exists D = {pi }
i=1 such that for all x and r > 0, B(x, r) D 6= .
Definition 3.16 A topological space is said to be completely separable if it has a countable basis for the
topology.
Theorem 3.17 A metric space is separable if and only if it is completely separable.
Proof: If the metric space has a countable basis for the topology, pick a point from each of the basic
open sets to get a countable dense subset of the metric space.
Now suppose the metric space, (X, d) , has a countable dense subset, D. Let B denote all balls having
centers in D which have positive rational radii. We will show this is a basis for the topology. It is clear it
is a countable set. Let U be any open set and let z U. Then there exists r > 0 such that B (z, r) U. In
B (z, r/3) pick a point from D, x. Now let r1 be a positive rational number in the interval (r/3, 2r/3) and
consider the set from B, B (x, r1 ) . If y B (x, r1 ) then
d (y, z) d (y, x) + d (x, z) < r1 + r/3 < 2r/3 + r/3 = r.
Thus B (x, r1 ) contains z and is contained in U. This shows, since z is an arbitrary point of U that U is the
union of a subset of B.
We already discussed Cauchy sequences in the context of Rp but the concept makes perfectly good sense
in any metric space.
Definition 3.18 A sequence {pn }
n=1 in a metric space is called a Cauchy sequence if for every > 0 there
exists N such that d(pn , pm ) < whenever n, m > N . A metric space is called complete if every Cauchy
sequence converges to some element of the metric space.
Example 3.19 Rn and Cn are complete metric spaces for the metric defined by d(x, y) |x y| (
yi |2 )1/2 .

Pn

i=1

|xi

Not all topological spaces are metric spaces and so the traditional  definition of continuity must be
modified for more general settings. The following definition does this for general topological spaces.
Definition 3.20 Let (X, ) and (Y, ) be two topological spaces and let f : X Y . We say f is continuous
at x X if whenever V is an open set of Y containing f (x), there exists an open set U such that x U
and f (U ) V . We say that f is continuous if f 1 (V ) whenever V .
Definition 3.21 Let (X, ) and (Y, ) be two topological spaces. X Y is the Cartesian product. (X Y =
{(x, y) : x X, y Y }). We can define a product topology as follows. Let B = {(A B) : A , B }.
B is a basis for the product topology.
Theorem 3.22 B defined above is a basis satisfying the conditions of Theorem 3.3.
More generally we have the following definition which considers any finite Cartesian product of topological
spaces.
Definition
Qn 3.23 If (Xi , i ) is a topological space, we make
basis be i=1 Ai where Ai i .

Qn

i=1

Xi into a topological space by letting a

Theorem 3.24 Definition 3.23 yields a basis for a topology.

The proof of this theorem is almost immediate from the definition and is left for the reader.
The definition of compactness is also considered for a general topological space. This is given next.

47
Definition 3.25 A subset, E, of a topological space (X, ) is said to be compact if whenever C and
E C, there exists a finite subset of C, {U1 Un }, such that E ni=1 Ui . (Every open covering admits
a finite subcovering.) We say E is precompact if E is compact. A topological space is called locally compact
if it has a basis B, with the property that B is compact for each B B. Thus the topological space is locally
compact if it has a basis of precompact open sets.
In general topological spaces there may be no concept of bounded. Even if there is, closed and bounded
is not necessarily the same as compactness. However, we can say that in any Hausdorff space every compact
set must be a closed set.
Theorem 3.26 If (X, ) is a Hausdorff space, then every compact subset must also be a closed set.
Proof: Suppose p
/ K. For each x X, there exist open sets, Ux and Vx such that
x Ux , p Vx ,
and
Ux Vx = .
Since K is assumed to be compact, there are finitely many of these sets, Ux1 , , Uxm which cover K. Then
let V m
i=1 Vxi . It follows that V is an open set containing p which has empty intersection with each of the
Uxi . Consequently, V contains no points of K and is therefore not a limit point. This proves the theorem.
Lemma 3.27 Let (X, ) be a topological space and let B be a basis for . Then K is compact if and only if
every open cover of basic open sets admits a finite subcover.
The proof follows directly from the definition and is left to the reader. A very important property enjoyed
by a collection of compact sets is the property that if it can be shown that any finite intersection of this
collection has non empty intersection, then it can be concluded that the intersection of the whole collection
has non empty intersection.
Definition 3.28 If every finite subset of a collection of sets has nonempty intersection, we say the collection
has the finite intersection property.
Theorem 3.29 Let K be a set whose elements are compact subsets of a Hausdorff topological space, (X, ) .
Suppose K has the finite intersection property. Then =
6 K.
Proof: Suppose to the contrary that = K. Then consider


C KC : K K .
It follows C is an open cover of K0 where K0 is any particular element of K. But then there are finitely many
K K, K1 , , Kr such that K0 ri=1 KiC implying that ri=0 Ki = , contradicting the finite intersection
property.
It is sometimes important to consider the Cartesian product of compact sets. The following is a simple
example of the sort of theorem which holds when this is done.
Theorem 3.30 Let X and Y be topological spaces, and K1 , K2 be compact sets in X and Y respectively.
Then K1 K2 is compact in the topological space X Y .
Proof: Let C be an open cover of K1 K2 of sets A B where A and B are open sets. Thus C is a open
cover of basic open sets. For y Y , define
Cy = {A B C : y B}, Dy = {A : A B Cy }

48

GENERAL TOPOLOGY

Claim: Dy covers K1 .
Proof: Let x K1 . Then (x, y) K1 K2 so (x, y) A B C. Therefore A B Cy and so
x A Dy .
Since K1 is compact,
{A1 , , An(y) } Dy
covers K1 . Let
n(y)

By = i=1 Bi
Thus {A1 , , An(y) } covers K1 and Ai By Ai Bi Cy .
Since K2 is compact, there is a finite list of elements of K2 , y1 , , yr such that
{By1 , , Byr }
covers K2 . Consider
n(y ) r

{Ai Byl }i=1l l=1 .

If (x, y) K1 K2 , then y Byj for some j {1, , r}. Then x Ai for some i {1, , n(yj )}. Hence
(x, y) Ai Byj . Each of the sets Ai Byj is contained in some set of C and so this proves the theorem.
Another topic which is of considerable interest in general topology and turns out to be a very useful
concept in analysis as well is the concept of a subbasis.
Definition 3.31 S is called a subbasis for the topology if the set B of finite intersections of sets of S
is a basis for the topology, .
Recall that the compact sets in Rn with the usual topology are exactly those that are closed and bounded.
We will have use of the following simple result in the following chapters.
Theorem 3.32 Let U be an open set in Rn . Then there exists a sequence of open sets, {Ui } satisfying
Ui Ui Ui+1
and
U =
i=1 Ui .
Proof: The following lemma will be interesting for its own sake and in addition to this, is exactly what
is needed for the proof of this theorem.
Lemma 3.33 Let S be any nonempty subset of a metric space, (X, d) and define
dist (x,S) inf {d (x, s) : s S} .
Then the mapping, x dist (x, S) satisfies
|dist (y, S) dist (x, S)| d (x, y) .
Proof of the lemma: One of dist (y, S) , dist (x, S) is larger than or equal to the other. Assume without
loss of generality that it is dist (y, S). Choose s1 S such that
dist (x, S) +  > d (x, s1 )

3.1. COMPACTNESS IN METRIC SPACE

49

Then
|dist (y, S) dist (x, S)| = dist (y, S) dist (x, S)
d (y, s1 ) d (x, s1 ) +  d (x, y) + d (x, s1 ) d (x, s1 ) + 
= d (x, y) + .
Since  is arbitrary, this proves the lemma.
If U = Rn it is clear that U =
i=1 B (0, i) and so, letting Ui = B (0, i),
B (0, i) = {x Rn : dist (x, {0}) < i}
and by continuity of dist (, {0}) ,
B (0, i) = {x Rn : dist (x, {0}) i} .
Therefore, the Heine Borel theorem applies and we see the theorem is true in this case.


Now we use this lemma to finish the proof in the case where U is not all of Rn . Since x dist x,U C is
continuous, the set,



1
C
Ui x U : dist x,U
> and |x| < i ,
i
is an open set. Also U =
i=1 Ui and these sets are increasing. By the lemma,



1
Ui = x U : dist x,U C and |x| i ,
i
a compact set by the Heine Borel theorem and also, Ui Ui Ui+1 .

3.1

Compactness in metric space

Many existence theorems in analysis depend on some set being compact. Therefore, it is important to be
able to identify compact sets. The purpose of this section is to describe compact sets in a metric space.
Definition 3.34 In any metric space, we say a set E is totally bounded if for every  > 0 there exists a
finite set of points {x1 , , xn } such that
E ni=1 B (xi , ).
This finite set of points is called an  net.
The following proposition tells which sets in a metric space are compact.
Proposition 3.35 Let (X, d) be a metric space. Then the following are equivalent.
(X, d) is compact,

(3.5)

(X, d) is sequentially compact,

(3.6)

(X, d) is complete and totally bounded.

(3.7)

50

GENERAL TOPOLOGY

Recall that X is sequentially compact means every sequence has a convergent subsequence converging
so an element of X.
Proof: Suppose (3.5) and let {xk } be a sequence. Suppose {xk } has no convergent subsequence. If this
is so, then {xk } has no limit point and no value of the sequence is repeated more than finitely many times.
Thus the set
Cn = {xk : k n}
is a closed set and if
Un = CnC ,
then
X =
n=1 Un
but there is no finite subcovering, contradicting compactness of (X, d).
Now suppose (3.6) and let {xn } be a Cauchy sequence. Then xnk x for some subsequence. Let  > 0
be given. Let n0 be such that if m, n n0 , then d (xn , xm ) < 2 and let l be such that if k l then
d (xnk , x) < 2 . Let n1 > max (nl , n0 ). If n > n1 , let k > l and nk > n0 .
d (xn , x) d(xn , xnk ) + d (xnk , x)


<
+ = .
2 2
Thus {xn } converges to x and this shows (X, d) is complete. If (X, d) is not totally bounded, then there
exists  > 0 for which there is no  net. Hence there exists a sequence {xk } with d (xk , xl )  for all l 6= k.
This contradicts (3.6) because this is a sequence having no convergent subsequence. This shows (3.6) implies
(3.7).
n
n
Now suppose (3.7). We show this implies (3.6). Let {pn } be a sequence and let {xni }m
net for
i=1 be a 2
n = 1, 2, . Let


Bn B xnin , 2n
be such that Bn contains pk for infinitely many values of k and Bn Bn+1 6= . Let pnk be a subsequence
having
pnk B k .
Then if k l,
d (pnk , pnl )

k1
X

d pni+1 , pni

<

k1
X

2(i1) < 2(l2).

i=l

i=l

Consequently {pnk } is a Cauchy sequence. Hence it converges. This proves (3.6).

Now suppose (3.6) and (3.7). Let Dn be a n1 net for n = 1, 2, and let
D =
n=1 Dn .
Thus D is a countable dense subset of (X, d). The set of balls
B = {B (q, r) : q D, r Q (0, )}

3.1. COMPACTNESS IN METRIC SPACE

51

is a countable basis for (X, d). To see this, let p B (x, ) and choose r Q (0, ) such that
 d (p, x) > 2r.
Let q B (p, r) D. If y B (q, r), then
d (y, x) d (y, q) + d (q, p) + d (p, x)
< r + r +  2r = .
Hence p B (q, r) B (x, ) and this shows each ball is the union of balls of B. Now suppose C is any open
cover of X. Let Be denote the balls of B which are contained in some set of C. Thus
Be = X.

e pick U C such that U B. Let Ce be the resulting countable collection of sets. Then Ce is
For each B B,
e
a countable open cover of X. Say Ce = {Un }
n=1 . If C admits no finite subcover, then neither does C and we
n
can pick pn X \ k=1 Uk . Then since X is sequentially compact, there is a subsequence {pnk } such that
{pnk } converges. Say
p = lim pnk .
k

All but finitely many points of {pnk } are in X \

nk=1 Uk .

Therefore p X \ nk=1 Uk for each n. Hence

p
/
k=1 Uk
contradicting the construction of {Un }
n=1 . Hence X is compact. This proves the proposition.
Next we apply this very general result to a familiar example, Rn . In this setting totally bounded and
bounded are the same. This will yield another proof of the Heine Borel theorem.
Lemma 3.36 A subset of Rn is totally bounded if and only if it is bounded.
Proof: Let A be totally bounded. We need to show it is bounded. Let x1 , , xp be a 1 net for A. Now
consider the ball B (0, r + 1) where r > max (||xi || : i = 1, , p) . If z A, then z B (xj , 1) for some j and
so by the triangle inequality,
||z 0|| ||z xj || + ||xj || < 1 + r.
Thus A B (0,r + 1) and so A is bounded.
Now suppose A is bounded and suppose A is not totally bounded. Then there exists  > 0 such that
there is no  net for A. Therefore, there exists a sequence of points {ai } with ||ai aj ||  if i 6= j. Since
A is bounded, there exists r > 0 such that
A [r, r)n.
(x [r, r)n means xi [r, r) for each i.) Now define S to be all cubes of the form
n
Y

[ak , bk )

k=1

where
ak = r + i2p r, bk = r + (i + 1) 2p r,

n
for i {0, 1, , 2p+1 1}. Thus S is a collectionof 2p+1 nonoverlapping cubes whose union equals
[r, r)n and whose diameters are all equal to 2p r n. Now choose p large enough that the diameter of
these cubes is less than . This yields a contradiction because one of the cubes must contain infinitely many
points of {ai }. This proves the lemma.
The next theorem is called the Heine Borel theorem and it characterizes the compact sets in Rn .

52

GENERAL TOPOLOGY

Theorem 3.37 A subset of Rn is compact if and only if it is closed and bounded.

Proof: Since a set in Rn is totally bounded if and only if it is bounded, this theorem follows from
Proposition 3.35 and the observation that a subset of Rn is closed if and only if it is complete. This proves
the theorem.
The following corollary is an important existence theorem which depends on compactness.
Corollary 3.38 Let (X, ) be a compact topological space and let f : X R be continuous.
max {f (x) : x X} and min {f (x) : x X} both exist.

Then

Proof: Since f is continuous, it follows that f (X) is compact. From Theorem 3.37 f (X) is closed and
bounded. This implies it has a largest and a smallest value. This proves the corollary.

3.2

Connected sets

Stated informally, connected sets are those which are in one piece. More precisely, we give the following
definition.
Definition 3.39 We say a set, S in a general topological 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. We say a set is connected if it is not separated.
One of the most important theorems about connected sets is the following.
Theorem 3.40 Suppose U and V are connected sets having nonempty intersection. 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.
The intersection of connected sets is not necessarily connected as is shown by the following picture.
U
V

Theorem 3.41 Let f : X Y be continuous where X and Y are topological spaces and X is connected.
Then f (X) is also connected.

3.2. CONNECTED SETS

53

Proof: We 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 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 connected components. This
is the concept of the next definition.
Definition 3.42 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 3.43 Let Cp be a connected component of a set S in a general topological 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 also since otherwise, as in Theorem 3.40, the set would be separated. But this implies B is
empty. Therefore, Cp is connected. From this, and Theorem 3.40, 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 3.44 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. We need to show (p, q) C. If
x (p, q) \ C
let C (, x) A, and C (x, ) B. Then C = A B and the sets, A and B separate C contrary to
the assumption that C is connected.
Conversely, let I be an interval. Suppose I is separated by A and B. Pick x A and y B. Suppose
without loss of generality that x < y. Now 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 > 0,
(l, l + ) 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.
The following theorem is a very useful description of the open sets in R.

Theorem 3.45 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 ) .

54

GENERAL TOPOLOGY

Proof: Let p U and let z Cp , the connected component determined by p. Since U is open, there
exists, > 0 such that (z , z + ) U. It follows from Theorem 3.40 that
(z , z + ) Cp .
This shows Cp is open. By Theorem 3.44, 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 3.46 We say a topological space, E is arcwise connected if for any two points, p, q E, there
exists a closed interval, [a, b] and a continuous function, : [a, b] E such that (a) = p and (b) = q. We
say E is locally connected if it has a basis of connected open sets. We say E is locally arcwise connected if
it has a basis of arcwise connected open sets.
An example of an arcwise connected topological space would be the any subset of Rn which is the
continuous image of an interval. Locally connected is not the same as connected. A well known example is
the following.



1
: x (0, 1] {(0, y) : y [1, 1]}
(3.8)
x, sin
x
We leave it as an exercise to verify that this set of points considered as a metric space with the metric from
R2 is not locally connected or arcwise connected but is connected.
Proposition 3.47 If a topological space is arcwise connected, then it is connected.
Proof: Let X be an arcwise connected space 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 : [a, b] X such that (a) = p and (b) = q. But then we would have
([a, b]) = ( ([a, b]) A) ( ([a, b]) B) and the two sets, ([a, b]) A and ([a, b]) B are separated thus
showing that ([a, b]) is separated and contradicting Theorem 3.44 and Theorem 3.41. It follows that X
must be connected as claimed.
Theorem 3.48 Let U be an open subset of a locally arcwise connected topological space, X. Then U is
arcwise connected if and only if U if connected. Also the connected components of an open set in such a
space are open sets, hence arcwise connected.
Proof: By Proposition 3.47 we only need to verify that if U is connected and open in the context of this
theorem, then U is arcwise connected. Pick p U . We will say x U satisfies P if there exists a continuous
function, : [a, b] U such that (a) = p and (b) = x.
A {x U such that x satisfies P.}
If x A, there exists, according to the assumption that X is locally arcwise connected, an open set, V,
containing x and contained in U which is arcwise connected. Thus letting y V, there exist intervals, [a, b]
and [c, d] and continuous functions having values in U , , such that (a) = p, (b) = x, (c) = x, and
(d) = y. Then let 1 : [a, b + d c] U be defined as

(t) if t [a, b]
1 (t)
(t) if t [b, b + d c]
Then it is clear that 1 is a continuous function mapping p to y and showing that V A. Therefore, A is
open. We also know that A 6= because there is an open set, V containing p which is contained in U and is
arcwise connected.

3.3. THE TYCHONOFF THEOREM

55

Now consider B U \ A. We will verify that this is also open. If B is not open, there exists a point
z B such that every open set conaining z is not contained in B. Therefore, letting V be one of the basic
open sets chosen such that z V U, we must have points of A contained in V. 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.
We need to verify the connected components are open. Let z Cp where Cp is the connected component
determined by p. Then picking V an arcwise connected open set which contains z and is contained in U,
Cp V is connected and contained in U and so it must also be contained in Cp . This proves the theorem.

3.3

The Tychonoff theorem

This section might be omitted on a first reading of the book. It is on the very important theorem about
products of compact topological spaces. In order to prove this theorem we need to use a fundamental result
about partially ordered sets which we describe next.
Definition 3.49 Let F be a nonempty set. F is called a partially ordered set if there is a relation, denoted
here by , such that
x x for all x F.
If x y and y z then x z.
C F is said to be a chain if every two elements of C are related. By this we mean that if x, y C, then
either x y or y x. Sometimes we call a chain a totally ordered set. C is said to be a maximal chain if
whenever D is a chain containing C, D = C.
The most common example of a partially ordered set is the power set of a given set with being the
relation. The following theorem is equivalent to the axiom of choice. For a discussion of this, see the appendix
on the subject.
Theorem 3.50 (Hausdorff Maximal Principle) Let F
exists a maximal chain.

be a nonempty partially ordered set. Then there

The main tool in the study of products of compact topological spaces is the Alexander subbasis theorem
which we present next.
Theorem 3.51 Let (X, ) be a topological space and let S be a subbasis for . (Recall this means that
finite intersections of sets of S form a basis for .) Then if H X, H is compact if and only if every open
cover of H consisting entirely of sets of S admits a finite subcover.
Proof: The only if part is obvious. To prove the other implication, first note that if every basic open
cover, an open cover composed of basic open sets, admits a finite subcover, then H is compact. Now suppose
that every subbasic open cover of H admits a finite subcover but H is not compact. This implies that there
exists a basic open cover of H, O, which admits no finite subcover. Let F be defined as
{O : O is a basic open cover of H which admits no finite subcover}.
Partially order F by set inclusion and use the Hausdorff maximal principle to obtain a maximal chain, C, of
such open covers. Let
D = C.

56

GENERAL TOPOLOGY

Then it follows that D is an open cover of H which is maximal with respect to the property of being a basic
open cover having no finite subcover of H. (If D admits a finite subcover, then since C is a chain and the
finite subcover has only finitely many sets, some element of C would also admit a finite subcover, contrary
to the definition of F.) Thus if D0 % D and D0 is a basic open cover of H, then D0 has a finite subcover of
H. One of the sets of D, U , has the property that
U = m
i=1 Bi , Bi S
and no Bi is in D. If not, we could replace each set in D with a subbasic set also in D containing it
and thereby obtain a subbasic cover which would, by assumption, admit a finite subcover, contrary to the
properties of D. Thus D {B i } admits a finite subcover,
V1i , , Vmi i , Bi
for each i = 1, , m. Consider
{U, Vji , j = 1, , mi , i = 1, , m}.
If p H \ {Vji }, then p Bi for each i and so p U . This is therefore a finite subcover of D contradicting
the properties of D. This proves the theorem.
Let I be a set and suppose for each i I, (Xi , i ) is a nonempty topological space. The Cartesian
product of the Xi , denoted by
Y
Xi ,
iI

consists of the set of all choice functions defined on I which select a single element of each Xi . Thus
Y
f
Xi
iI

means for every i I, f (i) Xi . The axiom of choice says

Pj (A) =

iI

Xi is nonempty. Let

Bi

iI

where Bi = Xi if i 6= j and Bj = A. A subbasis for a topology on the product space consists of all sets
Pj (A) where A j . (These sets have an open set in the j th slot and the whole space in the other slots.)
Thus a basis consists of finite intersections of these sets. It is easy to see that finite intersections
Q do form a
basis for a topology. This topology is called the product topology and we will denote it by i . Next we
use the Alexander subbasis theorem to prove the Tychonoff theorem.
Q
Q
Theorem 3.52 If (Xi i ) is compact, then so is ( iI Xi , i ).
Proof: By the Alexander subbasis theorem, we will establish compactness of the product space if we
show
every subbasic open cover admits a finite subcover. Therefore, let O be a subbasic open cover of
Q
X
i . Let
iI
Oj = {Q O : Q = Pj (A) for some A j }.
Let
j Oj = {A : Pj (A) Oj }.

3.4. EXERCISES

57

If no j Oj covers Xj , then pick

f

Xi \ i Oi

iI

so f (j)
/ j Oj and so f
/ O contradicting O is an open cover. Hence, for some j,
Xj = j Oj
and so there exist A1 , , Am , sets in j such that

and Pj (Ai ) O. Therefore, {Pj (Ai )}m

i=1

3.4

Xj m
i=1 Ai
Q
covers iI Xi .

Exercises

1. Prove the definition

in Rn or Cn satisfies (3.3)-(3.4). In addition to this, prove that ||||
Pnof distance
2 1/2
given by ||x|| = ( i=1 |xi | )
is a norm. This means it satisfies the following.
||x|| 0, ||x|| = 0 if and only if x = 0.
||x|| = ||||x|| for a number.
||x + y|| ||x|| + ||y||.
2. Completeness of R is an axiom. Using this, show Rn and Cn are complete metric spaces with respect
to the distance given by the usual norm.
3. Prove Urysohns lemma. A Hausdorff space, X, is normal if and only if whenever K and H are disjoint
nonempty closed sets, there exists a continuous function f : X [0, 1] such that f (k) = 0 for all k K
and f (h) = 1 for all h H.
4. Prove that f : X Y is continuous if and only if f is continuous at every point of X.
5. Suppose (X, d), and (Y, ) are metric spaces and let f : X Y . Show f is continuous at x X if and
only if whenever xn x, f (xn ) f (x). (Recall that xn x means that for all  > 0, there exists n
such that d (xn , x) <  whenever n > n .)
6. If (X, d) is a metric space, give an easy proof independent of Problem 3 that whenever K, H are
disjoint non empty closed sets, there exists f : X [0, 1] such that f is continuous, f (K) = {0}, and
f (H) = {1}.
7. Let (X, ) (Y, )be topological spaces with (X, ) compact and let f : X Y be continuous. Show
f (X) is compact.
8. (An example ) Let X = [, ] and consider B defined by sets of the form (a, b), [, b), and (a, ].
Show B is the basis for a topology on X.
9. Show (X, ) defined in Problem 8 is a compact Hausdorff space.
10. Show (X, ) defined in Problem 8 is completely separable.
11. In Problem 8, show sets of the form [, b) and (a, ] form a subbasis for the topology described
in Problem 8.

58

GENERAL TOPOLOGY

12. Let (X, ) and (Y, ) be topological spaces and let f : X Y . Also let S be a subbasis for . Show
f is continuous if and only if f 1 (V ) for all V S. Thus, it suffices to check inverse images of
subbasic sets in checking for continuity.
13. Show the usual topology of Rn is the same as the product topology of
n
Y

R R R R.

i=1

Do the same for Cn .

14. If M is a separable metric space and T M , then T is separable also.
15. Prove the Heine
Qn Borel theorem as follows. First show [a, b] is compact in R. Next use Theorem 3.30
to show that i=1 [ai , bi ] is compact. Use this to verify that compact sets are exactly those which are
closed and bounded.
16. Show the rational numbers, Q, are countable.
17. Verify that the set of (3.8) is connected but not locally connected or arcwise connected.
18. Let be an n dimensional multi-index. This means
= (1 , , n )
where each i is a natural number or zero. Also, we let
||

n
X

|i |

i=1

When we write x , we mean

n
1 2
x x
1 x2 x3 .

An n dimensional polynomial of degree m is a function of the form

X
d x.
||m

Let R be all n dimensional polynomials whose coefficients d come from the rational numbers, Q.
Show R is countable.
19. Let (X, d) be a metric space where d is a bounded metric. Let C denote the collection of closed subsets
of X. For A, B C, define
(A, B) inf { > 0 : A B and B A}
where for a set S,
S {x : dist (x, S) inf {d (x, s) : s S} } .
Show S is a closed set containing S. Also show that is a metric on C. This is called the Hausdorff
metric.

3.4. EXERCISES

59

20. Using 19, suppose (X, d) is a compact metric space. Show (C, ) is a complete metric space. Hint:
Show first that if Wn W where Wn is closed, then (Wn , W ) 0. Now let {An } be a Cauchy
sequence in C. Then if  > 0 there exists N such that when m, n N, then (An , Am ) < . Therefore,
for each n N,
(An )
k=n Ak .

Let A
n=1 k=n Ak . By the first part, there exists N1 > N such that for n N1 ,

k=n Ak , A < , and (An ) k=n Ak .

Therefore, for such n, A Wn An and (Wn ) (An ) A because

(An )
k=n Ak A.
21. In the situation of the last two problems, let X be a compact metric space. Show (C, ) is compact.
Hint: Let Dn be a 2n net for X. Let Kn denote finite unions of sets of the form B (p, 2n ) where
p Dn . Show Kn is a 2(n1) net for (C, ) .

60

GENERAL TOPOLOGY

Spaces of Continuous Functions

This chapter deals with vector spaces whose vectors are continuous functions.

4.1

Compactness in spaces of continuous functions

Let (X, ) be a compact space and let C (X; Rn ) denote the space of continuous Rn valued functions. For
f C (X; Rn ) let
||f || sup{|f (x) | : x X}
where the norm in the parenthesis refers to the usual norm in Rn .
The following proposition shows that C (X; Rn ) is an example of a Banach space.
Proposition 4.1 (C (X; Rn ) , || || ) is a Banach space.
Proof: It is obvious || || is a norm because (X, ) is compact. Also it is clear that C (X; Rn ) is a linear
space. Suppose {fr } is a Cauchy sequence in C (X; Rn ). Then for each x X, {fr (x)} is a Cauchy sequence
in Rn . Let
f (x) lim fk (x).
k

Therefore,
sup |f (x) fk (x) | = sup lim |fm (x) fk (x) |
xX m

xX

lim sup ||fm fk || < 

m

for all k large enough. Thus,

lim sup |f (x) fk (x) | = 0.

k xX

It only remains to show that f is continuous. Let

sup |f (x) fk (x) | < /3
xX

whenever k k0 and pick k k0 .

|f (x) f (y) |

|f (x) fk (x) | + |fk (x) fk (y) | + |fk (y) f (y) |

< 2/3 + |fk (x) fk (y) |
61

62

SPACES OF CONTINUOUS FUNCTIONS

Now fk is continuous and so there exists U an open set containing x such that if y U , then
|fk (x) fk (y) | < /3.
Thus, for all y U , |f (x) f (y) | <  and this shows that f is continuous and proves the proposition.
This space is a normed linear space and so it is a metric space with the distance given by d (f, g)
||f g|| . The next task is to find the compact subsets of this metric space. We know these are the subsets
which are complete and totally bounded by Proposition 3.35, but which sets are those? We need another way
to identify them which is more convenient. This is the extremely important Ascoli Arzela theorem which is
the next big theorem.
Definition 4.2 We say F C (X; Rn ) is equicontinuous at x0 if for all  > 0 there exists U , x0 U ,
such that if x U , then for all f F,
|f (x) f (x0 ) | < .
If F is equicontinuous at every point of X, we say F is equicontinuous. We say F is bounded if there exists
a constant, M , such that ||f || < M for all f F.
Lemma 4.3 Let F C (X; Rn ) be equicontinuous and bounded and let  > 0 be given. Then if {fr } F,
there exists a subsequence {gk }, depending on , such that
||gk gm || < 
whenever k, m are large enough.
Proof: If x X there exists an open set Ux containing x such that for all f F and y Ux ,
|f (x) f (y) | < /4.

(4.1)

Since X is compact, finitely many of these sets, Ux1 , , Uxp , cover X. Let {f1k } be a subsequence of
{fk } such that {f1k (x1 )} converges. Such a subsequence exists because F is bounded. Let {f2k } be a
subsequence of {f1k } such that {f2k (xi )} converges for i = 1, 2. Continue in this way and let {gk } = {fpk }.
Thus {gk (xi )} converges for each xi . Therefore, if  > 0 is given, there exists m such that for k, m > m,
max {|gk (xi ) gm (xi )| : i = 1, , p} <


.
2

Now if y X, then y Uxi for some xi . Denote this xi by xy . Now let y X and k, m > m . Then by
(4.1),
|gk (y) gm (y)| |gk (y) gk (xy )| + |gk (xy ) gm (xy )| + |gm (xy ) gm (y)|

<



+ max {|gk (xi ) gm (xi )| : i = 1, , p} + < .
4
4

It follows that for such k, m,

||gk gm || < 
and this proves the lemma.
Theorem 4.4 (Ascoli Arzela) Let F C (X; Rn ). Then F is compact if and only if F is closed, bounded,
and equicontinuous.

4.2. STONE WEIERSTRASS THEOREM

63

Proof: Suppose F is closed, bounded, and equicontinuous. We will show this implies F is totally
bounded. Then since F is closed, it follows that F is complete and will therefore be compact by Proposition
3.35. Suppose F is not totally bounded. Then there exists  > 0 such that there is no  net. Hence there
exists a sequence {fk } F such that
||fk fl || 
for all k 6= l. This contradicts Lemma 4.3. Thus F must be totally bounded and this proves half of the
theorem.
Now suppose F is compact. Then it must be closed and totally bounded. This implies F is bounded.
It remains to show F is equicontinuous. Suppose not. Then there exists x X such that F is not
equicontinuous at x. Thus there exists  > 0 such that for every open U containing x, there exists f F
such that |f (x) f (y)|  for some y U .
Let {h1 , , hp } be an /4 net for F. For each z, let Uz be an open set containing z such that for all
y Uz ,
|hi (z) hi (y)| < /8
for all i = 1, , p. Let Ux1 , , Uxm cover X. Then x Uxi for some xi and so, for some y Uxi ,there exists
f F such that |f (x) f (y)| . Since {h1 , , hp } is an /4 net, it follows that for some j, ||f hj || < 4
and so
 |f (x) f (y)| |f (x) hj (x)| + |hj (x) hj (y)| +
|hi (y) f (y)| /2 + |hj (x) hj (y)| /2 +
|hj (x) hj (xi )| + |hj (xi ) hj (y)| 3/4,
a contradiction. This proves the theorem.

4.2

Stone Weierstrass theorem

In this section we give a proof of the important approximation theorem of Weierstrass and its generalization
by Stone. This theorem is about approximating an arbitrary continuous function uniformly by a polynomial
or some other such function.
Definition 4.5 We say A is an algebra of functions if A is a vector space and if whenever f, g A then
f g A.
We will assume that the field of scalars is R in this section unless otherwise indicated. The approach
to the Stone Weierstrass depends on the following estimate which may look familiar to someone who has
taken a probability class. The left side of the following estimate is the variance of a binomial distribution.
However, it is not necessary to know anything about probability to follow the proof below although what is
being done is an application of the moment generating function technique to find the variance.
Lemma 4.6 The following estimate holds for x [0, 1].
n  
X
n

k=0

nk

(k nx) xk (1 x)

2n

64

SPACES OF CONTINUOUS FUNCTIONS

Proof: By the Binomial theorem,
n  
X

k

n
n
nk
et x (1 x)
= 1 x + et x .
k
k=0

Differentiating both sides with respect to t and then evaluating at t = 0 yields

n  
X
n
nk
kxk (1 x)
= nx.
k
k=0

Now doing two derivatives with respect to t yields

n  
X

n2 2t 2
n 2 t
k
nk
k e x (1 x)
= n (n 1) 1 x + et x
e x
k
k=0

n1 t
+n 1 x + et x
xe .
Evaluating this at t = 0,
n  
X
n

k=0

nk

k 2 (x) (1 x)

= n (n 1) x2 + nx.

Therefore,
n  
X
n

k=0

nk

(k nx) xk (1 x)

= n (n 1) x2 + nx 2n2 x2 + n2 x2


= n x x2 2n.

This proves the lemma.

Definition 4.7 Let f C ([0, 1]). Then the following polynomials are known as the Bernstein polynomials.
n    
X
n
k
nk
xk (1 x)
.
pn (x)
f
n
k
k=0

Theorem 4.8 Let f C ([0, 1]) and let pn be given in Definition 4.7. Then
lim ||f pn || = 0.

Proof: Since f is continuous on the compact [0, 1], it follows f is uniformly continuous there and so if
 > 0 is given, there exists > 0 such that if
|y x| ,
then
|f (x) f (y)| < /2.
By the Binomial theorem,
f (x) =

n  
X
n

k=0

nk

f (x) xk (1 x)

4.2. STONE WEIERSTRASS THEOREM

65

and so

n    
X

n 
k
nk
f
|pn (x) f (x)|
f (x) xk (1 x)

k
n
k=0


   

n 
k
nk
f
f (x) xk (1 x)
+

k
n


   

n 
k
nk
f
f (x) xk (1 x)

k
n

|k/nx|>

|k/nx|

<

/2 + 2 ||f ||

n 
X

(knx)2 >n2

2 ||f ||
n2 2

k=0

 
n k
nk
x (1 x)
k
2


n
2
nk
(k nx) xk (1 x)
+ /2.
k

By the lemma,

4 ||f ||
+ /2 < 
2 n

whenever n is large enough. This proves the theorem.

The next corollary is called the Weierstrass approximation theorem.
Corollary 4.9 The polynomials are dense in C ([a, b]).
Proof: Let f C ([a, b]) and let h : [0, 1] [a, b] be linear and onto. Then f h is a continuous function
defined on [0, 1] and so there exists a polynomial, pn such that
|f (h (t)) pn (t)| < 
for all t [0, 1]. Therefore for all x [a, b],



f (x) pn h1 (x)  < .

Since h is linear pn h1 is a polynomial. This proves the theorem.

The next result is the key to the profound generalization of the Weierstrass theorem due to Stone in
which an interval will be replaced by a compact or locally compact set and polynomials will be replaced with
elements of an algebra satisfying certain axioms.
Corollary 4.10 On the interval [M, M ], there exist polynomials pn such that
pn (0) = 0
and
lim ||pn |||| = 0.

66

SPACES OF CONTINUOUS FUNCTIONS

Proof: Let pn || uniformly and let
pn pn pn (0).

This proves the corollary.

The following generalization is known as the Stone Weierstrass approximation theorem. First, we say
an algebra of functions, A defined on A, annihilates no point of A if for all x A, there exists g A such
that g (x) 6= 0. We say the algebra separates points if whenever x1 6= x2 , then there exists g A such that
g (x1 ) 6= g (x2 ).
Theorem 4.11 Let A be a compact topological space and let A C (A; R) be an algebra of functions which
separates points and annihilates no point. Then A is dense in C (A; R).
Proof: We begin by proving a simple lemma.
Lemma 4.12 Let c1 and c2 be two real numbers and let x1 6= x2 be two points of A. Then there exists a
function fx1 x2 such that
fx1 x2 (x1 ) = c1 , fx1 x2 (x2 ) = c2 .
Proof of the lemma: Let g A satisfy
g (x1 ) 6= g (x2 ).
Such a g exists because the algebra separates points. Since the algebra annihilates no point, there exist
functions h and k such that
h (x1 ) 6= 0, k (x2 ) 6= 0.
Then let
u gh g (x2 ) h, v gk g (x1 ) k.
It follows that u (x1 ) 6= 0 and u (x2 ) = 0 while v (x2 ) 6= 0 and v (x1 ) = 0. Let
fx1 x2

c1 u
c2 v
+
.
u (x1 ) v (x2 )

This proves the lemma. Now we continue with the proof of the theorem.
First note that A satisfies the same axioms as A but in addition to these axioms, A is closed. Suppose
f A and suppose M is large enough that
||f || < M.
Using Corollary 4.10, let pn be a sequence of polynomials such that
||pn |||| 0, pn (0) = 0.
It follows that pn f A and so |f | A whenever f A. Also note that
max (f, g) =

|f g| + (f + g)
2

min (f, g) =

(f + g) |f g|
.
2

4.3. EXERCISES

67

Therefore, this shows that if f, g A then

max (f, g) , min (f, g) A.
By induction, if fi , i = 1, 2, , m are in A then
max (fi , i = 1, 2, , m) , min (fi , i = 1, 2, , m) A.
Now let h C (A; R) and use Lemma 4.12 to obtain fxy , a function of A which agrees with h at x and
y. Let  > 0 and let x A. Then there exists an open set U (y) containing y such that
fxy (z) > h (z)  if z U (y).
Since A is compact, let U (y1 ) , , U (yl ) cover A. Let
fx max (fxy1 , fxy2 , , fxyl ).
Then fx A and
fx (z) > h (z) 
for all z A and fx (x) = h (x). Then for each x A there exists an open set V (x) containing x such that
for z V (x),
fx (z) < h (z) + .
Let V (x1 ) , , V (xm ) cover A and let
f min (fx1 , , fxm ).
Therefore,
f (z) < h (z) + 
for all z A and since each
fx (z) > h (z) ,
it follows
f (z) > h (z) 
also and so
|f (z) h (z)| < 
for all z. Since  is arbitrary, this shows h A and proves A = C (A; R). This proves the theorem.

4.3

Exercises

1. Let (X, ) , (Y, ) be topological spaces and let A X be compact. Then if f : X Y is continuous,
show that f (A) is also compact.
2. In the context of Problem 1, suppose R = Y where the usual topology is placed on R. Show f
achieves its maximum and minimum on A.

68

SPACES OF CONTINUOUS FUNCTIONS

3. Let V be an open set in Rn . Show there is an increasing sequence of compact sets, Km , such that
V =
m=1 Km . Hint: Let




1
Cm x Rn : dist x,V C
m
where
dist (x,S) inf {|y x| such that y S}.
Consider Km Cm B (0,m).
4. 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
||f || sup{|f (x)| : x X}.
5. Let [0, 1]. We define, for X a compact subset of Rp ,
C (X; Rn ) {f C (X; Rn ) : (f ) + ||f || ||f || < }
where
||f || sup{|f (x)| : x X}
and
(f ) sup{

|f (x) f (y)|
: x, y X, x 6= y}.

|x y|

Show that (C (X; Rn ) , |||| ) is a complete normed linear space.

n
p
6. Let {fn }
n=1 C (X; R ) where X is a compact subset of R and suppose

||fn || M
for all n. Show there exists a subsequence, nk , such that fnk converges in C (X; Rn ). We say the
given sequence is precompact when this happens. (This also shows the embedding of C (X; Rn ) into
C (X; Rn ) is a compact embedding.)
7. Let f :R Rn Rn be continuous and bounded and let x0 Rn . If
x : [0, T ] Rn
and h > 0, let
h x (s)

x0 if s h,
x (s h) , if s > h.

For t [0, T ], let

xh (t) = x0 +

Z
0

f (s, h xh (s)) ds.

4.3. EXERCISES

69

Show using the Ascoli Arzela theorem that there exists a sequence h 0 such that
xh x
in C ([0, T ] ; Rn ). Next argue
x (t) = x0 +

f (s, x (s)) ds

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 problem.
x0 = f (t, x) , t [0, T ]
x (0) = x0 .
This is the Peano existence theorem for ordinary differential equations.
8. Show the set of polynomials R described in Problem 18 of Chapter 3 is dense in the space C (A; R)
when A is a compact subset of Rn . Conclude from this other problem that C (A; R) is separable.
9. Let H and K be disjoint closed sets in a metric space, (X, d), and let
g (x)

2
1
h (x)
3
3

where
h (x)

dist (x, H)
.
dist (x, H) + dist (x, K)



Show g (x) 13 , 13 for all x X, g is continuous, and g equals
it necessary to be in a metric space to do this?

1
3

on H while g equals

1
3

on K. Is

10. Suppose M is a closed set in X where X is the metric space of problem 9 and suppose f : M [1, 1]
is continuous. Show there exists g : X [1, 1] such that g is continuous and g = f on M . Hint:
Show there exists


1 1
, ,
g1 C (X) , g1 (x)
3 3
and |f (x) g1 (x)|

2
3

for all x H. To do this, consider the disjoint closed sets





1
1
1
1
Hf
1,
, Kf
,1
3
3

and use Problem 9 if the two sets are nonempty. When this has been done, let
3
(f (x) g1 (x))
2
play the role of f and let g2 be like g1 . Obtain

  
n  i1
n


X
2
2


gi (x)
f (x)


3
3
i=1

and consider

g (x)

 i1
X
2
i=1

Is it necessary to be in a metric space to do this?

gi (x).

70

SPACES OF CONTINUOUS FUNCTIONS

11. Let M be a closed set in a metric space (X, d) and suppose f C (M ). Show there exists g C (X)
such that g (x) = f (x) for all x M and if f (M ) [a, b], then g (X) [a, b]. This is a version of the
Tietze extension theorem. Is it necessary to be in a metric space for this to work?
12. Let X be a compact topological space and suppose {fn } is a sequence of functions continuous on X
having values in Rn . Show there exists a countable dense subset of X, {xi } and a subsequence of {fn },
{fnk }, such that {fnk (xi )} converges for each xi . Hint: First get a subsequence which converges at
x1 , then a subsequence of this subsequence which converges at x2 and a subsequence of this one which
converges at x3 and so forth. Thus the second of these subsequences converges at both x1 and x2
while the third converges at these two points and also at x3 and so forth. List them so the second
is under the first and the third is under the second and so forth thus obtaining an infinite matrix of
entries. Now consider the diagonal sequence and argue it is ultimately a subsequence of every one of
these subsequences described earlier and so it must converge at each xi . This procedure is called the
Cantor diagonal process.
13. Use the Cantor diagonal process to give a different proof of the Ascoli Arzela theorem than that
presented in this chapter. Hint: Start with a sequence of functions in C (X; Rn ) and use the Cantor
diagonal process to produce a subsequence which converges at each point of a countable dense subset
of X. Then show this sequence is a Cauchy sequence in C (X; Rn ).
14. What about the case where C0 (X) consists of complex valued functions and the field of scalars is C
rather than R? In this case, suppose A is an algebra of functions in C0 (X) which separates the points,
annihilates no point, and has the property that if f A, then f A. Show that A is dense in C0 (X).
Hint: Let Re A {Re f : f A}, Im A {Im f : f A}. Show A = Re A + i Im A = Im A + i Re A.
Then argue that both Re A and Im A are real algebras which annihilate no point of X and separate
the points of X. Apply the Stone Weierstrass theorem to approximate Re f and Im f with functions
from these real algebras.
15. Let (X, d) be a metric space where d is a bounded metric. Let C denote the collection of closed subsets
of X. For A, B C, define
(A, B) inf { > 0 : A B and B A}
where for a set S,
S {x : dist (x, S) inf {d (x, s) : s S} } .
Show x dist (x, S) is continuous and that therefore, S is a closed set containing S. Also show that
is a metric on C. This is called the Hausdorff metric.
16. Suppose (X, d) is a compact metric space. Show (C, ) is a complete metric space. Hint: Show first
that if Wn W where Wn is closed, then (Wn , W ) 0. Now let {An } be a Cauchy sequence in
C. Then if  > 0 there exists N such that when m, n N, then (An , Am ) < . Therefore, for each
n N,
(An )
k=n Ak .

Let A
n=1 k=n Ak . By the first part, there exists N1 > N such that for n N1 ,

k=n Ak , A < , and (An ) k=n Ak .

Therefore, for such n, A Wn An and (Wn ) (An ) A because

(An )
k=n Ak A.
17. Let X be a compact metric space. Show (C, ) is compact. Hint: Let Dn be a 2n net for X. Let Kn
denote finite unions of sets of the form B (p, 2n ) where p Dn . Show Kn is a 2(n1) net for (C, ) .

Abstract measure and Integration

5.1

Algebras

This chapter is on the basics of measure theory and integration. A measure is a real valued mapping from
some subset of the power set of a given set which has values in [0, ] . We will see that many apparently
different things can be considered as measures and also that whenever we are in such a measure space defined
below, there is an integral defined. By discussing this in terms of axioms and in a very abstract setting,
we unify many topics into one general theory. For example, it will turn out that sums are included as an
integral of this sort. So is the usual integral as well as things which are often thought of as being in between
sums and integrals.
Let be a set and let F be a collection of subsets of satisfying
F, F,

(5.1)

E F implies E C \ E F,

If {En }
n=1 F, then n=1 En F.

(5.2)

Definition 5.1 A collection of subsets of a set, , satisfying Formulas (5.1)-(5.2) is called a algebra.
As an example, let be any set and let F = P(), the set of all subsets of (power set). This obviously
satisfies Formulas (5.1)-(5.2).
Lemma 5.2 Let C be a set whose elements are algebras of subsets of . Then C is a algebra also.
Example 5.3 Let denote the collection of all open sets in Rn and let ( ) intersection of all algebras
that contain . ( ) is called the algebra of Borel sets .
This is a very important algebra and it will be referred to frequently as the Borel sets. Attempts to
describe a typical Borel set are more trouble than they are worth and it is not easy to do so. Rather, one
uses the definition just given in the example. Note, however, that all countable intersections of open sets
and countable unions of closed sets are Borel sets. Such sets are called G and F respectively.

5.2

Monotone classes and algebras

Definition 5.4 A is said to be an algebra of subsets of a set, Z if Z A, A, and when E, F A, E F

and E \ F are both in A.
It is important to note that if A is an algebra, then it is also closed under finite intersections. Thus,
E F = (E C F C )C A because E C = Z \ E A and similarly F C A.
71

72

ABSTRACT MEASURE AND INTEGRATION

Definition 5.5 M P(Z) is called a monotone class if

a.) En En+1 , E =
n=1 En, and En M, then E M.
b.) En En+1 , E =
n=1 En , and En M, then E M.
(In simpler notation, En E and En M implies E M. En E and En M implies E M.)
How can we easily identify algebras? The following lemma is useful for this problem.
Lemma 5.6 Suppose that R and E are subsets of P(Z) such that E is defined as the set of all finite disjoint
unions of sets of R. Suppose also that
, Z R
A B R whenever A, B R,
A \ B E whenever A, B R.
Then E is an algebra of sets of Z.
Proof: Note first that if A R, then AC E because AC = Z \ A. Now suppose that E1 and E2 are in
E,
n
E1 = m
i=1 Ri , E2 = j=1 Rj

where the Ri are disjoint sets in R and the Rj are disjoint sets in R. Then
n
E1 E2 = m
i=1 j=1 Ri Rj

which is clearly an element of E because no two of the sets in the union can intersect and by assumption
they are all in R. Thus finite intersections of sets of E are in E. If E = ni=1 Ri
E C = ni=1 RiC = finite intersection of sets of E
which was just shown to be in E. Thus if E1 , E2 E,
E1 \ E2 = E1 E2C E
and
E1 E2 = (E1 \ E2 ) E2 E
from the definition of E. This proves the lemma.
Corollary 5.7 Let (Z1 , R1 , E1 ) and (Z2 , R2 , E2 ) be as described in Lemma 5.6. Then (Z1 Z2 , R, E) also
satisfies the conditions of Lemma 5.6 if R is defined as
R {R1 R2 : Ri Ri }
and
E { finite disjoint unions of sets of R}.
Consequently, E is an algebra of sets.

5.2. MONOTONE CLASSES AND ALGEBRAS

73

Proof: It is clear , Z1 Z2 R. Let R11 R21 and R12 R22 be two elements of R.
R11 R21 R12 R22 = R11 R12 R21 R22 R
by assumption.


R11 R21 \ R12 R22 =






R11 R21 \ R22 R11 \ R12 R22 R21
= R11 A2 A1 R2
where A2 E2 , A1 E1 , and R2 R2 .

R22
R21

R12
R11

Since the two sets in the above expression on the right do not intersect, and each Ai is a finite union
of disjoint elements of Ri , it follows the above expression is in E. This proves the corollary. The following
example will be referred to frequently.
Example 5.8 Consider for R, sets of the form I = (a, b] (, ) where a [, ] and b [, ].
Then, clearly, , (, ) R and it is not hard to see that all conditions for Corollary 5.7 are satisfied.
Applying this corollary repeatedly, we find that for
( n
)
Y
Ii : Ii = (ai , bi ] (, )
R
i=1

and E is defined as finite disjoint unions of sets of R,

(Rn , R, E)
satisfies the conditions of Corollary 5.7 and in particular E is an algebra of sets of Rn . It is clear that the
same would hold if I were of the form [a, b) (, ).
Theorem 5.9 (Monotone Class theorem) Let A be an algebra of subsets of Z and let M be a monotone
class containing A. Then M (A), the smallest -algebra containing A.
Proof: We may assume M is the smallest monotone class containing A. Such a smallest monotone class
exists because the intersection of monotone classes containing A is a monotone class containing A. We show
that M is a -algebra. It will then follow M (A). For A A, define
MA {B M such that A B M}.
Clearly MA is a monotone class containing A. Hence MA = M because M is the smallest such monotone
class. This shows that A B M whenever A A and B M. Now pick B M and define
MB {D M such that D B M}.

74

ABSTRACT MEASURE AND INTEGRATION

We just showed A MB . It is clear that MB is a monotone class. Thus MB = M and it follows that
D B M whenever D M and B M.
A similar argument shows that D \ B M whenever D, B M. (For A A, let
MA = {B M such that B \ A and A \ B M}.
Argue MA is a monotone class containing A, etc.)
Thus M is both a monotone class and an algebra. Hence, if E M then Z \ E M. We want to show
M is a -algebra. But if Ei M and Fn = ni=1 Ei , then Fn M and Fn
i=1 Ei . Since M is a monotone
class,
i=1 Ei M and so M is a -algebra. This proves the theorem.
Definition 5.10 Let F be a algebra of sets of and let : F [0, ]. We call a measure if
(

Ei ) =

i=1

(Ei )

(5.3)

i=1

whenever the Ei are disjoint sets of F. The triple, (, F, ) is called a measure space and the elements of
F are called the measurable sets. We say (, F, ) is a finite measure space when () < .
Theorem 5.11 Let {Em }
m=1 be a sequence of measurable sets in a measure space (, F, ). Then if
En En+1 En+2 ,
(
i=1 Ei ) = lim (En )

(5.4)

and if En En+1 En+2 and (E1 ) < , then

(
i=1 Ei ) = lim (En ).

(5.5)

C C
Proof: First note that
F so
i=1 Ei = (i=1 Ei )
i=1 Ei is measurable. To show (5.4), note that
(5.4) is obviously true if (Ek ) = for any k. Therefore, assume (Ek ) < for all k. Thus

(Ek+1 \ Ek ) = (Ek+1 ) (Ek ).

Hence by (5.3),
(
i=1 Ei ) = (E1 ) +

(Ek+1 \ Ek ) = (E1 )

k=1

(Ek+1 ) (Ek )

k=1

= (E1 ) + lim

n
X

(Ek+1 ) (Ek ) = lim (En+1 ).

n

k=1

This shows part (5.4). To verify (5.5), since (E1 ) < ,

n
(E1 ) (
i=1 Ei ) = (E1 \ i=1 Ei ) = lim (E1 \ i=1 Ei )
n

= (E1 ) lim (ni=1 Ei ) = (E1 ) lim (En ),

n

where the second equality follows from part (5.4). Hence

lim (En ) = (
i=1 Ei ).

This proves the theorem.

5.2. MONOTONE CLASSES AND ALGEBRAS

75

Definition 5.12 Let (, F, ) be a measure space and let (X, ) be a topological space. A function f : X
is said to be measurable if f 1 (U ) F whenever U . (Inverse images of open sets are measurable.)
Note the analogy with a continuous function for which inverse images of open sets are open.
Definition 5.13 Let {an }
n=1 X, (X, ) where X and are described above. Then
lim an = a

means that whenever a U , there exists n0 such that if n > n0 , then an U . (Every open set containing
a also contains an for all but finitely many values of n.) Note this agrees with the definition given earlier
for Rp , and C while also giving a definition of what is meant for convergence in general topological spaces.
Recall that (X, ) has a countable basis if there is a countable subset of , B, such that every set of
is the union of sets of B. We observe that for X given as either R, C, or [0, ] with the definition of
described earlier (a subbasis for the topology of [0, ] is sets of the form [0, b) and sets of the form (b, ]),
the following hold.
(X, ) has a countable basis, B.

(5.6)

Whenever U B, there exists a sequence of open sets, {Vm }

m=1 , such that
Vm V m Vm+1 , U =

Vm .

(5.7)

m=1

Recall S is defined as the union of the set S with all its limit points.
Theorem 5.14 Let fn and f be functions mapping to X where F is a algebra of measurable sets
of and (X, ) is a topological space satisfying Formulas (5.6) - (5.7). Then if fn is measurable, and
f () = limn fn (), it follows that f is also measurable. (Pointwise limits of measurable functions are
measurable.)
Proof: Let B be the countable basis of (5.6) and let U B. Let {Vm } be the sequence of (5.7). Since
f is the pointwise limit of fn ,
f 1 (Vm ) { : fk () Vm for all k large enough} f 1 (V m ).
Therefore,
1
1

f 1 (U ) =
(Vm )
m=1 f
m=1 n=1 k=n fk (Vm )

1

(Vm ) = f 1 (U ).
m=1 f

It follows f 1 (U ) F because it equals the expression in the middle which is measurable. Now let W .
Since B is countable, W =
n=1 Un for some sets Un B. Hence
1
f 1 (W ) =
(Un ) F.
n=1 f

This proves the theorem.

Example 5.15 Let X = [, ] and let a basis for a topology, , be sets of the form [, a), (a, b), and
(a, ]. Then it is clear that (X, ) satisfies Formulas (5.6) - (5.7) with a countable basis, B, given by sets
of this form but with a and b rational.

76

ABSTRACT MEASURE AND INTEGRATION

Definition 5.16 Let fn : [, ].

lim sup fn () = lim (sup{fk () : k n}).

(5.8)

lim inf fn () = lim (inf{fk () : k n}).

(5.9)

Note that in [, ] with the topology just described, every increasing sequence converges and every
decreasing sequence converges. This follows from Definition 5.13. Also, if
An () = inf{fk () : k n}, Bn () = sup{fk () : k n}.
It is clear that Bn () is decreasing while An () is increasing. Therefore, Formulas (5.8) and (5.9) always
make sense unlike the limit.
Lemma 5.17 Let f : [, ] where F is a algebra of subsets of . Then f is measurable if any of
the following hold.
f 1 ((d, ]) F for all finite d,
f 1 ([, c)) F for all finite c,
f 1 ([d, ]) F for all finite d,
f 1 ([, c]) F for all finite c.
Proof: First note that the first and the third are equivalent. To see this, note
1
f 1 ([d, ]) =
((d 1/n, ]),
n=1 f
1
f 1 ((d, ]) =
([d + 1/n, ]).
n=1 f

Similarly, the second and fourth conditions are equivalent.

f 1 ([, c]) = (f 1 ((c, ]))C
so the first and fourth conditions are equivalent. Thus all four conditions are equivalent and if any of them
hold,
f 1 ((a, b)) = f 1 ([, b)) f 1 ((a, ]) F.
Thus f 1 (B) F whenever B is a basic open set described in Example 5.15. Since every open set can be
obtained as a countable union of these basic open sets, it follows that if any of the four conditions hold, then
f is measurable. This proves the lemma.
Theorem 5.18 Let fn : [, ] be measurable with respect to a algebra, F, of subsets of . Then
lim supn fn and lim inf n fn are measurable.
Proof : Let gn () = sup{fk () : k n}. Then
1
gn1 ((c, ]) =
k=n fk ((c, ]) F.

Therefore gn is measurable.
lim sup fn () = lim gn ()
n

and so by Theorem 5.14 lim supn fn is measurable. Similar reasoning shows lim inf n fn is measurable.

5.2. MONOTONE CLASSES AND ALGEBRAS

77

Theorem 5.19 Let fi , i = 1, , n be a measurable function mapping to the topological space (X,
Qn) and
T
suppose that has a countable basis, B. Then
f
=
(f

f
)
is
a
measurable
function
from

to
1
n
i=1 X.
Qn
(Here it is understood that the topology of i=1 X is the standard product topology and that F is the
algebra of measurable subsets of .)
Qn
Proof: First we observe that sets of the form i=1 Bi , Bi B form a countable basis for the product
topology. Now
f 1 (

n
Y

Bi ) = ni=1 fi1 (Bi ) F.

i=1

Since every open set is a countable union of these sets, it follows f 1 (U ) F for all open U .
Theorem 5.20 Let (, F) be a measure space and let fQ
i , i = 1, , n be measurable functions mapping to
n
(X, ), a topological space with a countable basis. Let g : i=1 X X be continuous and let f = (f1 fn )T .
Then g f is a measurable function.
Proof: Let U be open.
(g f )1 (U ) = f 1 (g 1 (U )) = f 1 (open set) F
by Theorem 5.19.
Example 5.21 Let X = (, ] with a basis for the topology given by sets of the form (a, b) and (c, ], a, b, c
rational numbers. Let + : X X X be given by +(x, y) = x + y. Then + is continuous; so if f, g are
measurable functions mapping to X, we may conclude by Theorem 5.20 that f + g is also measurable.
Also, if a, b are positive real numbers and l(x, y) = ax + by, then l : X X X is continuous and so
l(f, g) = af + bg is measurable.
Note that the basis given in this example provides the usual notions of convergence in (-, ]. Theorems
5.19 and 5.20 imply that under appropriate conditions, sums, products, and, more generally, continuous
functions of measurable functions are measurable. The following is also interesting.
Theorem 5.22 Let f : X be measurable. Then f 1 (B) F for every Borel set, B, of (X, ).
Proof: Let S {B X such that f 1 (B) F}. S contains all open sets. It is also clear that S is
a algebra. Hence S contains the Borel sets because the Borel sets are defined as the intersection of all
algebras containing the open sets.
The following theorem is often very useful when dealing with sequences of measurable functions.
Theorem 5.23 (Egoroff ) Let (, F, ) be a finite measure space
(() < )
and let fn , f be complex valued measurable functions such that
lim fn () = f ()

for all
/ E where (E) = 0. Then for every > 0, there exists a set,
F E, (F ) < ,
such that fn converges uniformly to f on F C .

78

ABSTRACT MEASURE AND INTEGRATION

Proof: Let Ekm = { E C : |fn () f ()| 1/m for some n > k}. By Theorems 5.19 and 5.20,
{ E C : |fn () f ()| 1/m}

is measurable. Hence Ekm is measurable because

C
Ekm =
n=k+1 { E : |fn () f ()| 1/m}.

For fixed m,
k=1 Ekm = and so it has measure 0. Note also that
Ekm E(k+1)m .
Since (E1m ) < ,
0 = (
k=1 Ekm ) = lim (Ekm )
k

by Theorem 5.11. Let k(m) be chosen such that (Ek(m)m ) < 2m . Let
F =E

Ek(m)m .

m=1

Then (F ) < because

(F ) (E) +

m=1

Ek(m)m .

C
Now let > 0 be given and pick m0 such that m1
0 < . If F , then

C
Ek(m)m
.

m=1
C
Hence Ek(m
so
0 )m0

|fn () f ()| < 1/m0 <

for all n > k(m0 ). This holds for all F C and so fn converges uniformly to f on F C . This proves the
theorem.
We conclude this chapter with a comment about notation. We say that something happens for a.e. and
say almost everywhere if there exists a set E with (E) = 0 and the thing takes place for all
/ E. Thus
f () = g() a.e. if f () = g() for all
/ E where (E) = 0.
We also say a measure space, (, F, ) is finite if there exist measurable sets, n such that (n ) <
and =
n=1 n .

5.3

Exercises

1. Let = N ={1, 2, }. Let F = P(N) and let (S) = number of elements in S. Thus ({1}) = 1 =
({2}), ({1, 2}) = 2, etc. Show (, F, ) is a measure space. It is called counting measure.
2. Let be any uncountable set and let F = {A : either A or AC is countable}. Let (A) = 1 if A
is uncountable and (A) = 0 if A is countable. Show (, F, ) is a measure space.
3. Let F be a algebra of subsets of and suppose F has infinitely many elements. Show that F is
uncountable.

5.3. EXERCISES

79

4. Prove Lemma 5.2.

5. We say g is Borel measurable if whenever U is open, g 1 (U ) is Borel. Let f : X and let g : X Y
where X, Y are topological spaces and F is a algebra of sets of . Suppose f is measurable and g is
Borel measurable. Show g f is measurable.
6. Let (, F) be a measure space and suppose f : C. Show f is measurable if and only if Re f and
Im f are measurable real-valued functions.
7. Let (, F, ) be a measure space. Define : P() [0, ] by
(A) = inf{(B) : B A, B F}.
Show satisfies
() = 0, if A B, (A) (B), (
i=1 Ai )

(Ai ).

i=1

If 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

(Ei ) < .

i=1

Let S = { such that Ei for infinitely many values of i}. Show (S) = 0 and S is measurable.
This is part of the Borel Cantelli lemma.
9. Let fn , f be measurable functions with values in C. We say that fn converges in measure if
lim (x : |f (x) fn (x)| ) = 0

for each fixed > 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
(x : |f (x) fn1 (x)| 1) < 1/2.
Choose n2 > n1 such that
(x : |f (x) fn2 (x)| 1/2) < 1/22,
n3 > n2 such that
(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 remember Problem 8.

10. Let C {Ei }i=1 be a countable collection of sets and let 1

i=1 Ei . Show there exists an algebra
of sets, A, such that A C and A is countable. Hint: Let C1 denote all finite unions of sets of C and
1 . Thus C1 is countable. Now let B1 denote all complements with respect to 1 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 .

80

ABSTRACT MEASURE AND INTEGRATION

5.4

The Abstract Lebesgue Integral

In this section we develop the Lebesgue integral and present some of its most important properties. In all
that follows will be a measure defined on a algebra F of subsets of . We always define 0 = 0. This
may seem somewhat arbitrary and this is so. However, a little thought will soon demonstrate that this is the
right definition for this meaningless expression in the context of measure theory. To see this, consider the
zero function defined on R. What do we want the integral of this function to be? Obviously, by an analogy
with the Riemann integral, we would want this to equal zero. Formally, it is zero times the length of the set
or infinity. The following notation will be used.
For a set E,

1 if E,
XE () =
0 if
/ E.
This is called the characteristic function of E.
Definition 5.24 A function, s, is called simple if it is measurable and has only finitely many values. These
values will never be .
Definition 5.25 If s(x) 0 and s is simple,
Z

m
X

ai (Ai )

i=1

where Ai = { : s(x) = ai } and a1 , , am are the distinct values of s.

R
R
Note that s could equal + if (Ak ) = and ak > 0 for some k, but s is well defined because s 0
and we use the convention that 0 = 0.
Lemma 5.26 If a, b 0 and if s and t are nonnegative simple functions, then
Z
Z
Z
as + bt a s + b t.
Proof: Let
s() =

n
X

i XAi (), t() =

i=1

m
X

j XBj ()

i=1

where i are the distinct values of s and the j are the distinct values of t. Clearly as + bt is a nonnegative
simple function. Also,
(as + bt)() =

n
m X
X

(ai + b j )XAi Bj ()

j=1 i=1

where the sets Ai Bj are disjoint. Now we dont know that all the values ai + b j P
are distinct, but we
r
note that if E1 , , Er are disjoint measurable sets whose union is E, then (E) = i=1 (Ei ). Thus
Z
m X
n
X
as + bt =
(ai + b j )(Ai Bj )
j=1 i=1
n
X

= a

i (Ai ) + b

i=1

= a
This proves the lemma.

s+b

m
X
j=1

t.

j (Bj )

5.4. THE ABSTRACT LEBESGUE INTEGRAL

Corollary 5.27 Let s =

Pn

i=1

81

ai XEi where ai 0 and the Ei are not necessarily disjoint. Then

Z

s=

n
X

ai (Ei ).

i=1

R
Proof: aXEi = a(Ei ) so this follows from Lemma 5.26.
Now we are ready to define the Lebesgue integral of a nonnegative measurable function.
Definition 5.28 Let f : [0, ] be measurable. Then
Z
Z
f d sup{ s : 0 s f, s simple}.

Lemma 5.29 If s 0 is a nonnegative simple function,

R

sd =

s. Moreover, if f 0, then

f d 0.

Proof:
The second claim is obvious. To verify the first, suppose 0 t s and t is simple. Then clearly
R
t s and so
Z
Z
Z
sd = sup{ t : 0 t s, t simple} s.

R
R
But s s and s is simple so sd s.
The next theorem is one of the big results that justifies the use of the Lebesgue integral.
Theorem 5.30 (Monotone Convergence theorem) Let f 0 and suppose {fn } is a sequence of nonnegative
measurable functions satisfying
lim fn () = f () for each .

fn () fn+1 ()

(5.10)

Then f is measurable and

Z

f d = lim

fn d.

Proof: First note that f is measurable

it is the limit of measurable functions.
R by Theorem 5.14 since
R
It is also clear from (5.10) that limn fn d exists because { fn d} forms an increasing sequence. This
limit may be + but in any case,
Z
Z
lim
fn d f d
n

R
R
because fn d f d.
Let (0, 1) and let s be a simple function with
0 s() f (), s() =

r
X

i XAi ().

i=1

Then (1 )s() f () for all with strict inequality holding whenever f () > 0. Let
En = { : fn () (1 )s()}

(5.11)

82

ABSTRACT MEASURE AND INTEGRATION

Then
En En+1 , and
n=1 En = .
Therefore
lim

sXEn =

s.

This follows from Theorem 5.11 which implies that i (En Ai ) i (Ai ). Thus, from (5.11)
Z
Z
Z
Z
f d fn d fn XEn d ( sXEn d)(1 ).

(5.12)

Letting n in (5.12) we see that

Z

(5.13)

f d lim

fn d (1 )

s.

Now let 0 in (5.13) to obtain

Z

f d lim

fn d

s.

Now s was an arbitrary simple function less than or equal to f . Hence,

Z
Z
Z
Z
f d lim
fn d sup{ s : 0 s f, s simple} f d.
n

This proves the theorem.

The next theorem will be used frequently. It says roughly that measurable functions are pointwise limits
of simple functions. This is similar to continuous functions being the limit of step functions.
Theorem 5.31 Let f 0 be measurable. Then there exists a sequence of simple functions {sn } satisfying
0 sn ()

(5.14)

sn () sn+1 ()
f () = lim sn () for all .
n

Before proving this, we give a definition.

Definition 5.32 If f, g are functions having values in [0, ],
f g = max(f, g), f g = min(f, g).
Note that if f, g have finite values,
f g = 21 (f + g + |f g|), f g = 21 (f + g |f g|).
From this observation, the following lemma is obvious.
Lemma 5.33 If s, t are nonnegative simple functions, then
s t, s t
are also simple functions. (Recall + is not a value of either s or t.)

(5.15)

5.5. THE SPACE L1

83

Proof of Theorem 5.31: Let

I = {x : f (x) = +}.
Let Enk = f 1 ([ nk , k+1
n )). Let
n

2
X
k
tn () =
XE () + nXI ().
n nk
k=0

Then tn () f () for all and limn tn () = f () for all . This is because tn () = n for I and if
n
f () [0, 2 n+1 ), then
0 fn () tn ()

1
.
n

Thus whenever
/ I, the above inequality will hold for all n large enough. Let
s1 = t1 , s2 = t1 t2 , s3 = t1 t2 t3 , .
Then the sequence {sn } satisfies Formulas (5.14)-(5.15) and this proves the theorem.
Next we show that the integral is linear on nonnegative functions. Roughly speaking, it shows the
integral is trying to be linear and is only prevented from being linear at this point by not yet being defined
on functions which could be negative or complex valued. We will define the integral for these functions soon
and then this lemma will be the key to showing the integral is linear.
Lemma 5.34 Let f, g 0 be measurable. Let a, b 0 be constants. Then
Z
Z
Z
(af + bg)d = a f d + b gd.
Proof: Let {sn } and {
sn } be increasing sequences of simple functions such that
lim sn () = f (), lim sn () = g().

Then by the monotone convergence theorem and Lemma 5.26,

Z
Z
(af + bg)d = lim
(asn + b
sn )d
n
Z
Z
Z
= lim
asn + b
sn = lim a sn + b sn
n
n
Z
Z
Z
Z
= lim a sn d + b sn d = a f d + b gd.
n

This proves the lemma.

5.5

The space L1

Now suppose f has complex values and is measurable. We need to define what is meant by the integral of
such functions. First some theorems about measurability need to be shown.
Theorem 5.35 Let f = u + iv where u, v are real-valued functions. Then f is a measurable C valued
function if and only if u and v are both measurable R valued functions.

84

ABSTRACT MEASURE AND INTEGRATION

Proof: Suppose first that f is measurable. Let V R be open.
u1 (V ) = { : u() V } = { : f () V + iR} F,
v 1 (V ) = { : v() V } = { : f () R+iV } F.

Now suppose u and v are real and measurable.

f 1 ((a, b) + i(c, d)) = u1 (a, b) v 1 (c, d) F.
Since every open set in C may be written as a countable union of open sets of the form (a, b) + i(c, d), it
follows that f 1 (U ) F whenever U is open in C. This proves the theorem.
Definition 5.36 L1 () is the space of complex valued measurable functions, f , satisfying
Z
|f ()|d < .
We also write the symbol, ||f ||L1 to denote

|f ()| d.

Note that if f : C is measurable, then by Theorem 5.20, |f | : R is also measurable.

Definition 5.37 If u is real-valued,
u+ u 0, u (u 0).
Thus u+ and u are both nonnegative and
u = u+ u , |u| = u+ + u.
Definition 5.38 Let f = u + iv where u, v are real-valued. Suppose f L1 (). Then
Z
Z
Z
Z
Z
f d u+ d u d + i[ v + d v d].

Note that all this is well defined because

Z

u+ d,

|f |d < and so
u d,

v + d,

v d

are all finite. The next theorem shows the integral is linear on L1 ().
Theorem 5.39 L1 () is a complex vector space and if a, b C and
f, g L1 (),
then
Z

af + bgd = a

f d + b

gd.

(5.16)

5.5. THE SPACE L1

85

Proof: First suppose f, g are real-valued and in L1 (). We note that

h+ = 21 (h + |h|), h = 21 (|h| h)
whenever h is real-valued. Consequently,
f + + g + (f + g ) = (f + g)+ (f + g) = f + g.
Hence
f + + g + + (f + g) = (f + g)+ + f + g .
From Lemma 5.34,
Z

f + d +

g + d +

(f + g) d =

f d +

g d +

(5.17)

(f + g)+ d.

Since all integrals are finite,

Z
Z
Z
(f + g)d
(f + g)+ d (f + g) d
Z
Z
Z
Z
+
+

=
f d + g d ( f d + g d)
Z
Z

f d + gd.

(5.18)

(5.19)

Now suppose that c is a real constant and f is real-valued. Note

(cf ) = cf + if c < 0, (cf ) = cf if c 0.
(cf )+ = cf if c < 0, (cf )+ = cf + if c 0.
If c < 0, we use the above and Lemma 5.34 to write
Z
Z
Z
cf d
(cf )+ d (cf ) d
Z
Z
Z

+
= c f d + c f d c f d.
Similarly, if c 0,
Z

(cf ) d (cf ) d
Z
Z
Z
= c f + d c f d c f d.

cf d

This shows (5.16) holds if f, g, a, and b are all real-valued. To conclude, let a = + i, f = u + iv and use
the preceding.
Z
Z
af d =
( + i)(u + iv)d
Z
=
(u v) + i(u + v)d
Z
Z
Z
Z
= ud vd + i ud + i vd
Z
Z
Z
= ( + i)( ud + i vd) = a f d.

86

ABSTRACT MEASURE AND INTEGRATION

Thus (5.16) holds whenever f, g, a, and b are complex valued. It is obvious that L1 () is a vector space.
This proves the theorem.
The next theorem, known as Fatous lemma is another important theorem which justifies the use of the
Lebesgue integral.
Theorem 5.40 (Fatous lemma) Let fn be a nonnegative measurable function with values in [0, ]. Let
g() = lim inf n fn (). Then g is measurable and
Z
Z
gd lim inf
fn d.
n

Proof: Let gn () = inf{fk () : k n}. Then

1
gn1 ([a, ]) =
k=n fk ([a, ]) F.

Thus gn is measurable by Lemma 5.17. Also g() = limn gn () 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
gd = lim
gn d lim inf
fn d.
n

The last inequality holding because

Z

gn d

fn d.

This proves the Theorem.

Theorem 5.41 (Triangle inequality) Let f L1 (). Then
Z
Z
| f d| |f |d.
Proof:

R
R
R
f d C so there exists C, || = 1 such that | f d| = f d = f d. Hence
Z
Z
Z
| f d| =
f d = (Re(f ) + i Im(f ))d
Z
Z
Z
=
Re(f )d = (Re(f ))+ d (Re(f )) d
Z
Z
Z
+

(Re(f )) + (Re(f )) d |f |d = |f |d

which proves the theorem.

Theorem 5.42 (Dominated Convergence theorem) Let fn L1 () and suppose
f () = lim fn (),
n

and there exists a measurable function g, with values in [0, ], such that
Z
|fn ()| g() and
g()d < .
Then f L1 () and
Z

f d = lim

fn d.

5.6. DOUBLE SUMS OF NONNEGATIVE TERMS

87

Proof: f is measurable by Theorem 5.14. Since |f | g, it follows that

f L1 () and |f fn | 2g.
By Fatous lemma (Theorem 5.40),
Z
Z
2gd lim inf
2g |f fn |d
n
Z
Z
=
2gd lim sup
|f fn |d.
n

Subtracting

2gd,
0 lim sup

|f fn |d.

Hence
Z
Z
Z
0 lim sup ( |f fn |d) lim sup | f d fn d|
n

which proves the theorem.

Definition 5.43 Let E be a measurable subset of .
Z
Z
f d f XE d.
E

Also we may refer to L1 (E). The algebra in this case is just

{E A : A F}
and the measure is restricted to this smaller algebra. Clearly, if f L1 (), then
f XE L1 (E)
and if f L1 (E), then letting f be the 0 extension of f off of E, we see that f L1 ().

5.6

Double sums of nonnegative terms

The definition of the Lebesgue integral and the monotone convergence theorem imply that the order of
summation of a double sum of nonnegative terms can be interchanged and in fact the terms can be added in
any order. To see this, let = NN and let be counting measure defined on the set of all subsets of N N.
Thus, (E) = the number of elements of E. Then (, , P ()) is a measure space and if a : [0, ],
then a is a measurable function. Following the usual notation, aij a (i, j).
Theorem 5.44 Let a : [0, ]. Then
X

X
i=1 j=1

aij =

aij =

j=1 i=1

where is any one to one and onto map from N to .

ad =

k=1

a ( (k))

88

ABSTRACT MEASURE AND INTEGRATION

Proof: By the definition of the integral,
n X
l
X

aij

ad

j=1 i=1

for any n, l. Therefore, by the definition of what is meant by an infinite sum,

X

aij

ad.

j=1 i=1

Now let s a and s is a nonnegative simple function. If s (i, j) > 0 for infinitely many values of (i, j) ,
then
Z
Z
X

X
s = = ad =
aij .
j=1 i=1

Therefore, it suffices to assume s (i, j) > 0 for only finitely many values of (i, j) N N. Hence, for some
n > 1,
Z

n X
n
X

aij

j=1 i=1

aij .

j=1 i=1

Since s is an arbitrary nonnegative simple function, this shows

Z

ad

aij

X
X

aij .

j=1 i=1

and so
Z

ad =

j=1 i=1

The same argument holds if i and j are interchanged which verifies the first two equalities in the conclusion
of the theorem. The last equation in the conclusion of the theorem follows from the monotone convergence
theorem.

5.7

Vitali convergence theorem

In this section we consider a remarkable convergence theorem which, in the case of finite measure spaces
turns out to be better than the dominated convergence theorem.
Definition 5.45 Let (, F, ) be a measure space and let S L1 (). We say that S is uniformly integrable
if for every > 0 there exists > 0 such that for all f S
Z
|
f d| < whenever (E) < .
E

Lemma 5.46 If S is uniformly integrable, then |S| {|f | : f S} is uniformly integrable. Also S is
uniformly integrable if S is finite.

5.7. VITALI CONVERGENCE THEOREM

89

Proof: Let > 0 be given and suppose S is uniformly integrable. First suppose the functions are real
valued. Let be such that if (E) < , then
Z



 f d <

 2
E
for all f S. Let (E) < . Then if f S,
Z
Z
Z
|f | d
(f ) d +
f d
E
E[f 0]
E[f >0]
Z
 Z


 


 

= 
f d + 
f d
 E[f 0]
  E[f >0]


<
+ = .
2 2

In the above, [f > 0] is short for { : f () > 0} with a similar definition holding for [f 0] . In general,
if S is uniformly integrable, then Re S {Re f : f S} and Im S {Im f : f S} are easily seen to be
uniformly integrable. Therefore, applying the above result for real valued functions to these sets of functions,
it is routine to verify that |S| is uniformly integrable also.
For the last part, is suffices to verify a single function in L1 () is uniformly integrable. To do so, note
that from the dominated convergence theorem,
Z
lim
|f | d = 0.
R

[|f |>R]

Let > 0 be given and choose R large enough that

Z

[|f |>R]

|f | d < 2 . Now let (E) <

|f | d +

<

R (E) +

<

+ = .
2 2

|f | d =

E[|f |R]

2R .

Then

|f | d

E[|f |>R]

This proves the lemma.

The following theorem is Vitalis convergence theorem.
Theorem 5.47 Let {fn } be a uniformly integrable set of complex valued functions, () < , and fn (x)
f (x) a.e. where f is a measurable complex valued function. Then f L1 () and
Z
lim
|fn f |d = 0.
(5.20)
n

Proof: First we show that f L () . By uniform integrability, there exists > 0 such that if (E) < ,
then
Z
|fn | d < 1
E

for all n. By Egoroffs theorem, there exists a set, E of measure less than such that on E C , {fn } converges
uniformly. Therefore, if we pick p large enough, and let n > p,
Z
|fp fn | d < 1
EC

90

ABSTRACT MEASURE AND INTEGRATION

which implies
Z

|fn | d < 1 +

EC

|fp | d.

Then since there are only finitely many functions, fn with n p, we have the existence of a constant, M1
such that for all n,
Z
|fn | d < M1 .
EC

But also, we have

Z

|fm | d =

|fm | d +

EC

|fm |

M1 + 1 M.
Therefore, by Fatous lemma,
Z

|f | d lim inf

|fn | d M,

showing that f L1 as hoped.

R
Now S {f } is uniformly integrable so there exists 1 > 0 such that if (E) < 1 , then E |g| d < /3
for all g S {f }. Now by Egoroffs theorem, there exists a set, F with (F ) < 1 such that fn converges
uniformly to f on F C . Therefore, there exists N such that if n > N, then
Z

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

|f fn | d

|f fn | d +

FC

|f | d +

|fn | d

+ + = ,
3 3 3

<
which verifies (5.20).

5.8

The ergodic theorem

This section deals with a fundamental convergence theorem known as the ergodic theorem. It will only be
used in one place later in the book so you might omit this topic on a first reading or pick it up when you
need it later. I am putting it here because it seems to fit in well with the material of this chapter.
In this section (, F, ) will be a probability measure space. This means that () = 1. The mapping,
T : will satisfy the following condition.
T 1 (A) F whenever A F, T is one to one.
Lemma 5.48 If T satisfies (5.21), then if f is measurable, f T is measurable.
Proof: Let U be an open set. Then
1

(f T )

(U ) = T 1 f 1 (U ) F

(5.21)

5.8. THE ERGODIC THEOREM

91

by (5.21).
Now suppose that in addition to (5.21) T also satisfies


T 1 A = (A) ,

(5.22)

for all A F. In words, T 1 is measure preserving. Then for T satisfying (5.21) and (5.22), we have the
following simple lemma.
Lemma 5.49 If T satisfies (5.21) and (5.22) then whenever f is nonnegative and mesurable,
Z
Z
f () d =
f (T ) d.

(5.23)

Also (5.23) holds whenever f L1 () .

Proof: Let f 0 and f is measurable. By Theorem 5.31, let sn be an increasing sequence of simple
functions converging pointwise to f. Then by (5.22) it follows
Z
Z
sn () d = sn (T ) d
and so by the Monotone convergence theorem,
Z
Z
f () d = lim
sn () d
n

= lim

sn (T ) d =

f (T ) d.

Splitting f L1 into real and imaginary parts we apply the above to the positive and negative parts of these
and obtain (5.23) in this case also.
Definition 5.50 A measurable function, f, is said to be invariant if
f (T ) = f () .
A set, A F is said to be invariant if XA is an invariant function. Thus a set is invariant if and only if
T 1 A = A.
The following theorem, the individual ergodic theorem, is the main result.
Theorem 5.51 Let (, F, ) be a probability space and let T : satisfy (5.21) and (5.22). Then if
f L1 () having real or complex values and
Sn f ()

n
X

k=1

f T k1 , S0 f () 0,

(5.24)

it follows there exists a set of measure zero, N, and an invariant function g such that for all
/ N,
lim

1
Sn f () = g () .
n

and also
lim

1
Sn f = g in L1 ()
n

(5.25)

92

ABSTRACT MEASURE AND INTEGRATION

Proof: The proof of this theorem will make use of the following functions.
Mn f () sup {Sk f () : 0 k n}

(5.26)

M f () sup {Sk f () : 0 k} .

(5.27)

We will also define the following for h a measurable real valued function.
[h > 0] { : h () > 0} .
Now if A is an invariant set,
Sn (XA f ) ()

n
X

k=1

= XA ()

n
X

f T k1 XA T k1

f T k1 = XA () Sn f () .

k=1

Therefore, for such an invariant set,

Mn (XA f ) () = XA () Mn f () , M (XA f ) () = XA () M f () .

(5.28)

Let < a < b < and define

Nab


1
: < lim inf
Sn f () < a
n n

1
Sn f () < .
n n

< b < lim sup

(5.29)

Observe that from the definition, if |f ()| =

6 ,
lim inf

1
1
Sn f () = lim inf
Sn f (T )
n
n
n

and
1
1
Sn f () = lim sup Sn f (T ) .
n n
n n

lim sup

Therefore, T Nab = Nab so Nab is an invariant set because T is one to one. Also,
Nab [M (f b) > 0] [M (a f ) > 0] .
Consequently,
Z

(f () b) d =

Nab

[XNab M (f b)>0]

[M (XNab (f b))>0]

XNab () (f () b) d

XNab () (f () b) d

(5.30)

5.8. THE ERGODIC THEOREM

93

and
Z

(a f ()) d =

[XNab M (af )>0]

Nab

[M (XNab (af ))>0]

XNab () (a f ()) d

XNab () (a f ()) d.

(5.31)

We will complete the proof with the aid of the following lemma which implies the last terms in (5.30) and
(5.31) are nonnegative.
Lemma 5.52 Let f L1 () . Then
Z

f d 0.

[M f >0]

We postpone the proof of this lemma till we have completed the proof of the ergodic theorem. From
(5.30), (5.31), and Lemma 5.52,
Z
a (Nab )
f d b (Nab ) .
(5.32)
Nab

Since a < b, it follows that (Nab ) = 0. Now let

N {Nab : a < b, a, b Q} .
Since f L1 () and has complex values, it follows that (N ) = 0. Now T Na,b = Na,b and so
T (N ) = a,b T (Na,b ) = a,b Na,b = N.
Therefore, N is measurable and has measure zero. Also, T n N = N for all n N and so
n
N
n=1 T N.

For
/ N, limn n1 Sn f () exists. Now let
g ()

0 if N
.
limn n1 Sn f () if
/N

Then it is clear g satisfies the conditions of the theorem because if N, then T N also and so in this
case, g (T ) = g () . On the other hand, if
/ N, then
g (T ) = lim

1
1
Sn f (T ) = lim Sn f () = g () .
n n
n



The last claim follows from the Vitali convergence theorem if we verify the sequence, n1 Sn f n=1 is
uniformly integrable. To see this is the case, we know
f L1 () and so if  > 0 is given, there exists > 0
R

such that whenever B F and (B) , then B f () d < . Now by approximating the positive and
negative parts of f with simple functions we see that
Z
Z


k1
f T
d =
f () d.
A

T (k1) A

94

ABSTRACT MEASURE AND INTEGRATION

Taking (A) < , it follows


Z
  n Z

 1 X


1
k1


f T
d
 n Sn f () d  n

A
A
k=1


 n Z

n Z
n

1 X
X

1 X 


< 1
f () d
f
()
d
=
=
 (k1)
 n
 n
n
T (k1) A
T
A
k=1

k=1

k=1

because T (k1) A = (A) by assumption. This proves the above sequence is uniformly integrable and
so, by the Vitali convergence theorem,


 1



lim  Sn f g  = 0.
n n
L1

This proves the theorem.

It remains to prove the lemma.
Proof of Lemma 5.52: First note that Mn f () 0 for all n and . This follows easily from the
observation that by definition, S0 f () = 0 and so Mn f () is at least as large. Also note that the sets,
[Mn f > 0] are increasing in n and their union is [M f > 0] . Therefore, it suffices to show that for all n > 0,
Z
f d 0.
[Mn f >0]

Let T h h T. Thus T maps measurable functions to measurable functions by Lemma 5.48. It is also
clear that if h 0, then T h 0 also. Therefore,
Sk f () = f () + T Sk1 f () f () + T Mn f
and therefore,
Mn f () f () + T Mn f () .
Now
Z

Mn f () d =

Mn f () d

[Mn f >0]

f () d +

[Mn f >0]

T Mn f () d

f () d +

[Mn f >0]

Mn f () d

by Lemma 5.49. This proves the lemma.

The following is a simple corollary which follows from the above theorem.
Corollary 5.53 The conclusion of Theorem 5.51 holds if is only assumed to be a finite measure.
Definition 5.54 We say a set, A F is invariant if T (A) = A. We say the above mapping, T, is ergodic,
if the only invariant sets have measure 0 or 1.
If the map, T is ergodic, the following corollary holds.

5.9. EXERCISES

95

Corollary 5.55 In the situation of Theorem 5.51, if T is ergodic, then

Z
g () = f () d
for a.e. .
Proof: Let g be the function of Theorem 5.51 and let R1 be a rectangle in R2 = C of the form [a, a]
[a, a] such that g 1 (R1 ) has measure greater than 0. This set is invariant because the function, g is
invariant and so it must have measure 1. Divide R1 into four equal rectangles, R10 , R20 , R30 , R40 . Then one
of these, renamed R2 has the property that g 1 (R2 ) has positive measure. Therefore, since the set is
invariant, it must have measure 1. Continue in this way obtaining a sequence of closed rectangles, {Ri }
such that the diamter of Ri converges to zero and g 1 (Ri ) has measure 1. Then let c =
j=1 Rj . We know




1
1
g (c) = limn g (Ri ) = 1. It follows that g () = c for a.e. . Now from Theorem 5.51,
c=

1
n n

cd = lim

Sn f d =

f d.

This proves the corollary.

5.9

Exercises

1. Let = N = {1, 2, } and (S) = number of elements in S. If

f :C
what do we mean by

f d? Which functions are in L1 ()?

2. Give an example of a measure space, (, , F), and a sequence of nonnegative measurable functions
{fn } converging pointwise to a function f , such that inequality is obtained in Fatous lemma.
3. Fill in all the details of the proof of Lemma 5.46.
4. Suppose (, ) is a finite measure space and S L1 (). Show S is uniformly integrable and bounded
in L1 () if there exists an increasing function h which satisfies
h (t)
lim
= , sup
t t

Z


h (|f |) d : f S < .

When we say S is bounded we mean there is some number, M such that

Z
|f | d M
for all f S.
5. Let (, F, ) be a measure space and suppose f L1 () has the property that whenever (E) > 0,
Z
1
|
f d| C.
(E) E
Show |f ()| C a.e.

96

ABSTRACT MEASURE AND INTEGRATION

6. Let {an }, {bn } be sequences in [, ]. Show
lim sup (an ) = lim inf (an )
n

lim sup (an + bn ) lim sup an + lim sup bn

n

provided no sum is of the form . Also show strict inequality can hold in the inequality. State
and prove corresponding statements for lim inf.
7. Let (, F, ) be a measure space and suppose f, g : [, ] are measurable. Prove the sets
{ : f () < g()} and { : f () = g()}
are measurable.
8. Let {fn } be a sequence of real or complex valued measurable functions. Let
S = { : {fn ()} converges}.
Show S is measurable.
9. In the monotone convergence theorem
0 fn () fn+1 () .
The sequence of functions is increasing. In what way can increasing be replaced by decreasing?
10. Let (, F, ) be a measure space and suppose fn converges uniformly to f and that fn is in L1 ().
When can we conclude that
Z
Z
lim
fn d = f d?
n

11. Suppose un (t) is a differentiable function for t (a, b) and suppose that for t (a, b),
|un (t)|, |u0n (t)| < Kn
where

n=1

Kn < . Show
(

n=1

un (t))0 =

n=1

u0n (t).

The Construction Of Measures

6.1

Outer measures

We have impressive theorems about measure spaces and the abstract Lebesgue integral but a paucity of
interesting examples. In this chapter, we discuss the method of outer measures due to Caratheodory (1918).
This approach shows how to obtain measure spaces starting with an outer measure. This will then be used
to construct measures determined by positive linear functionals.
Definition 6.1 Let be a nonempty set and let : P() [0, ] satisfy
() = 0,
If A B, then (A) (B),
(
i=1 Ei )

(Ei ).

i=1

Such a function is called an outer measure. For E , we say E is measurable if for all S ,
(S) = (S \ E) + (S E).

(6.1)

To help in remembering (6.1), think of a measurable set, E, as a knife which is used to divide an arbitrary
set, S, into the pieces, S \ E and S E. If E is a sharp knife, the amount of stuff after cutting is the same
as the amount you started with. The measurable sets are like sharp knives. The idea is to show that the
measurable sets form a algebra. First we give a definition and a lemma.
Definition 6.2 (bS)(A) (S A) for all A . Thus bS is the name of a new outer measure.
Lemma 6.3 If A is measurable, then A is bS measurable.
Proof: Suppose A is measurable. We need to show that for all T ,
(bS)(T ) = (bS)(T A) + (bS)(T \ A).
Thus we need to show
(S T ) = (T A S) + (T S AC ).

(6.2)

But we know (6.2) holds because A is measurable. Apply Definition 6.1 to S T instead of S.
The next theorem is the main result on outer measures. It is a very general result which applies whenever
one has an outer measure on the power set of any set. This theorem will be referred to as Caratheodorys
procedure in the rest of the book.
97

98

THE CONSTRUCTION OF MEASURES

Theorem 6.4 The collection of measurable sets, S, forms a algebra and

If Fi S, Fi Fj = , then

(
i=1 Fi )

(Fi ).

(6.3)

i=1

If Fn Fn+1 , then if F =
n=1 Fn and Fn S, it follows that
(F ) = lim (Fn ).

(6.4)

If Fn Fn+1 , and if F =
n=1 Fn for Fn S then if (F1 ) < , we may conclude that
(F ) = lim (Fn ).

(6.5)

Also, (S, ) is complete. By this we mean that if F S and if E with (E \ F ) + (F \ E) = 0, then

E S.
Proof: First note that and are obviously in S. Now suppose that A, B S. We show A \ B = A B C
is in S. Using the assumption that B S in the second equation below, in which S A plays the role of S
in the definition for B being measurable,
(S (A B C )) + (S \ (A B C )) = (S A B C ) + (S (AC B))
=(SAB C )
}|
{
z
= (S (AC B)) + (S A) (S A B).

(6.6)

The following picture of S (AC B) may be of use.

S
B

A
From the picture, and the measurability of A, we see that (6.6) is no larger than
(S(AC B))
=(SAB C )
z
}|
{ z
}|
{
(S A B) + (S \ A) + (S A) (S A B)
= (S \ A) + (S A) = (S) .

This has shown that if A, B S, then A \ B S. Since 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 = ,
(A B) = ((A B) A) + ((A B) \ A) = (A) + (B).
Pn
By induction, if Ai Aj = and Ai S, (ni=1 Ai ) = i=1 (Ai ).
Now let A =
i=1 Ai where Ai Aj = for i 6= j.

X
i=1

(Ai ) (A) (ni=1 Ai ) =

n
X
i=1

(Ai ).

6.1. OUTER MEASURES

99

Since this holds for all n, we can take the limit as n and conclude,

(Ai ) = (A)

i=1

which establishes (6.3). Part (6.4) follows from part (6.3) just as in the proof of Theorem 5.11.
In order to establish (6.5), let the Fn be as given there. Then, since (F1 \ Fn ) increases to (F1 \ F ) , we
may use part (6.4) to conclude
lim ( (F1 ) (Fn )) = (F1 \ F ) .

Now (F1 \ F ) + (F ) (F1 ) and so (F1 \ F ) (F1 ) (F ) . Hence

lim ( (F1 ) (Fn )) = (F1 \ F ) (F1 ) (F )

which implies
lim (Fn ) (F ) .

But since F Fn , we also have

(F ) lim (Fn )
n

and this establishes (6.5).

It remains to show S is closed under countable unions. We already know that if A S, then AC S and
n
S is closed under finite unions. Let Ai S, A =
i=1 Ai , Bn = i=1 Ai . Then
(S) = (S Bn ) + (S \ Bn )
= (bS)(Bn ) + (bS)(BnC ).

(6.7)

By Lemma 6.3 we know Bn is (bS) measurable and so is BnC . We want to show (S) (S \ A) + (S A).
If (S) = , there is nothing to prove. Assume (S) < . Then we apply Parts (6.5) and (6.4) to (6.7)
and let n . Thus
Bn A, BnC AC
and this yields (S) = (bS)(A) + (bS)(AC ) = (S A) + (S \ A).
Thus A S and this proves Parts (6.3), (6.4), and (6.5).
Let F S and let (E \ F ) + (F \ E) = 0. Then
(S)
=

(S E) + (S \ E)
(S E F ) + (S E F C ) + (S E C )
(S F ) + (E \ F ) + (S \ F ) + (F \ E)
(S F ) + (S \ F ) = (S).

Hence (S) = (S E) + (S \ E) and so E S. This shows that (S, ) is complete.

Where do outer measures come from? One way to obtain an outer measure is to start with a measure
, defined on a algebra of sets, S, and use the following definition of the outer measure induced by the
measure.
Definition 6.5 Let be a measure defined on a algebra of sets, S P (). Then the outer measure
induced by , denoted by is defined on P () as
(E) = inf{(V ) : V S and V E}.
We also say a measure space, (S, , ) is finite if there exist measurable sets, i with (i ) < and
=
i=1 i .

100

THE CONSTRUCTION OF MEASURES

The following lemma deals with the outer measure generated by a measure which is finite. It says that
if the given measure is finite and complete then no new measurable sets are gained by going to the induced
outer measure and then considering the measurable sets in the sense of Caratheodory.
Lemma 6.6 Let (, S, ) be any measure space and let : P() [0, ] be the outer measure induced
by . Then is an outer measure as claimed and if S is the set of measurable sets in the sense of
Caratheodory, then S S and = on S. Furthermore, if is finite and (, S, ) is complete, then
S = S.
Proof: It is easy to see that is an outer measure. Let E S. We need to show E S and (E) = (E).
Let S . We need to show
(S) (S E) + (S \ E).

(6.8)

If (S) = , there is nothing to prove, so assume (S) < . Thus there exists T S, T S, and
(S) > (T ) = (T E) + (T \ E)
(T E) + (T \ E)
(S E) + (S \ E) .
Since is arbitrary, this proves (6.8) and verifies S S. Now if E S and V E,
(E) (V ).
Hence, taking inf,
(E) (E).
But also (E) (E) since E S and E E. Hence
(E) (E) (E).
Now suppose (, S, ) is complete. Thus if E, D S, and (E \ D) = 0, then if D F E, it follows
F S

(6.9)

because
F \ D E \ D S,
a set of measure zero. Therefore,
F \D S
and so F = D (F \ D) S.
We know already that S S so let F S. Using the assumption that the measure space is finite, let
{Bn } S, Bn = , Bn Bm = , (Bn ) < . Let
En F Bn , (En ) =
(F Bn ),

(6.10)

Hn Bn \ F = Bn F C , (Hn ) =
(Bn \ F ),

(6.11)

where En S, and let

6.2. POSITIVE LINEAR FUNCTIONALS

101

where Hn S. The following picture may be helpful in visualizing this situation.


Bn F

En

F Bn

6
Hn
Thus
Hn B n F C
and so
HnC BnC F
which implies
HnC Bn F Bn .
We have
HnC Bn F Bn En , HnC Bn , En S.

(6.12)

Claim: If A, B, D S and if A B with (A \ B) = 0. Then (A D) = (B D) .

Proof of claim: This follows from the observation that (A D) \ (B D) A \ B.
Now from (6.10) and (6.11) and this claim,









En \ HnC Bn = (F Bn ) \ HnC Bn = F Bn BnC Hn




= (F Hn Bn ) = F Bn F C Bn = () = 0.
Therefore, from (6.9) and (6.12) F Bn S. Therefore,
F =
n=1 F Bn S.
This proves the lemma.
Note that it was not necessary to assume was finite in order to consider and conclude that =
on S. This is sometimes referred to as the process of completing a measure because is a complete measure
and extends .

6.2

Positive linear functionals

One of the most important theorems related to the construction of measures is the Riesz. representation
theorem. The situation is that there exists a positive linear functional defined on the space Cc () where
is a topological space of some sort and is said to be a positive linear functional if it satisfies the following
definition.

102

THE CONSTRUCTION OF MEASURES

Definition 6.7 Let be a topological space. We say f : C is in Cc () if f is continuous and

spt (f ) {x : f (x) 6= 0}
is a compact set. (The symbol, spt (f ) is read as support of f .) If we write Cc (V ) for V an open set,
we mean that spt (f ) V and We say is a positive linear functional defined on Cc () if is linear,
(af + bg) = af + bg
for all f, g Cc () and a, b C. It is called positive because
f 0 whenever f (x) 0 for all x .
The most general versions of the theory about to be presented involve locally compact Hausdorff spaces
but here we will assume the topological space is a metric space, (, d) which is also compact, defined
below, and has the property that the closure of any open ball, B (x, r) is compact.
Definition 6.8 We say a topological space, , is compact if =
k=1 k where k is a compact subset
of .
To begin with we need some technical results and notation. In all that follows, will be a compact
metric space with the property that the closure of any open ball is compact. An obvious example of such a
thing is any closed subset of Rn or Rn itself and it is these cases which interest us the most. The terminology
of metric spaces is used because it is convenient and contains all the necessary ideas for the proofs which
follow while being general enough to include the cases just described.
Definition 6.9 If K is a compact subset of an open set, V , we say K V if
Cc (V ), (K) = {1}, () [0, 1].
Also for Cc (), we say K if
() [0, 1] and (K) = 1.
We say V if
() [0, 1] and spt() V.
The next theorem is a very important result known as the partition of unity theorem. Before we present
it, we need a simple lemma which will be used repeatedly.
Lemma 6.10 Let K be a compact subset of the open set, V. Then there exists an open set, W such that W
is a compact set and
K W W V.
Also, if K and V are as just described there exists a continuous function, such that K V.
Proof: For each k K, let B (k, rk ) Bk be such that Bk V. Since K is compact, finitely many of
these balls, Bk1 , , Bkl cover K. Let W li=1 Bki . Then it follows that W = li=1 Bki and satisfies the
conclusion of the lemma. Now we define as


dist x, W C
(x)
.
dist (x, W C ) + dist (x, K)
Note the denominator is never equal to zero because if dist (x, K) = 0, then x W and so is at a positive
distance from W C because W is open. This proves the lemma. Also note that spt () = W .

6.2. POSITIVE LINEAR FUNCTIONALS

103

Theorem 6.11 (Partition of unity) Let K be a compact subset of and suppose

K V = ni=1 Vi , Vi open.
Then there exist i Vi with
n
X

i (x) = 1

i=1

for all x K.
Proof: Let K1 = K \ ni=2 Vi . Thus K1 is compact and K1 V1 . By the above lemma, we let
K1 W 1 W 1 V 1
with W 1 compact and f be such that K1 f V1 with
W1 {x : f (x) 6= 0} .
Thus W1 , V2 , , Vn covers K and W 1 V1 . Let
K2 = K \ (ni=3 Vi W1 ).
Then K2 is compact and K2 V2 . Let K2 W2 W 2 V2 , W 2 compact. Continue this way finally
obtaining W1 , , Wn , K W1 Wn , and W i Vi W i compact. Now let W i Ui U i Vi , U i
compact.

Wi Ui Vi

By the lemma again, we may define i and such that

U i i Vi , ni=1 W i ni=1 Ui .
Now define
i (x) =

Pn
Pn
(x)i (x)/ j=1 j (x) if j=1 j (x) 6= 0,
Pn
0 if
j=1 j (x) = 0.

Pn
If x is such that j=1 j (x) = 0, then x
/ ni=1 U i . Consequently (y) = 0 for all y near x and so
Pn
i (y) = 0 for all y near x. Hence i is continuous at such x. If j=1 j (x) 6= 0, this situation persists
near
Pn x and so i is continuous at such points. Therefore i is continuous. If x K, then (x) = 1 and so
j=1 j (x) = 1. Clearly 0 i (x) 1 and spt( j ) Vj . This proves the theorem.
We dont need the following corollary at this time but it is useful later.
Corollary 6.12 If H is a compact subset of Vi , we can pick our partition of unity in such a way that
i (x) = 1 for all x H in addition to the conclusion of Theorem 6.11.
fj Vj \ H. Now in the proof above, applied to this
Proof: Keep Vi the same but replace Vj with V
modified collection of open sets, we see that if j 6= i, j (x) = 0 whenever x H. Therefore, i (x) = 1 on H.
Next we consider a fundamental theorem known as Caratheodorys criterion which gives an easy to check
condition which, if satisfied by an outer measure defined on the power set of a metric space, implies that the
algebra of measurable sets contains the Borel sets.

104

THE CONSTRUCTION OF MEASURES

Definition 6.13 For two sets, A, B in a metric space, we define

dist (A, B) inf {d (x, y) : x A, y B} .
Theorem 6.14 Let be an outer measure on the subsets of (X, d), a metric space. If
(A B) = (A) + (B)
whenever dist(A, B) > 0, then the algebra of measurable sets contains the Borel sets.
Proof: We only need show that closed sets are in S, the -algebra of measurable sets, because then the
open sets are also in S and so S Borel sets. Let K be closed and let S be a subset of . We need to show
(S) (S K) + (S \ K). Therefore, we may assume without loss of generality that (S) < . Let
Kn = {x : dist(x, K)

1
} = closed set
n

(Recall that x dist (x, K) is continuous.)

(S) ((S K) (S \ Kn )) = (S K) + (S \ Kn )

(6.13)

by assumption, since S K and S \ Kn are a positive distance apart. Now

(S \ Kn ) (S \ K) (S \ Kn ) + ((Kn \ K) S).

(6.14)

We look at ((Kn \ K) S). Note that 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
(S (Kn \ K))

(S (Kk \ Kk+1 )).

(6.15)

k=n

Now

(S (Kk \ Kk+1 )) =

k=1

(S (Kk \ Kk+1 )) +

k even

(S (Kk \ Kk+1 )).

(6.16)

k odd

Note that if A =
i=1 Ai and the distance between any pair of sets is positive, then
(A) =

(Ai ),

i=1

because

X
i=1

(Ai ) (A) (ni=1 Ai ) =

n
X
i=1

(Ai ).

6.2. POSITIVE LINEAR FUNCTIONALS

105

Therefore, from (6.16),

(S (Kk \ Kk+1 ))

k=1

= (

S (Kk \ Kk+1 )) + (

k even

S (Kk \ Kk+1 ))

k odd

< 2(S) < .

Therefore from (6.15)
lim (S (Kn \ K)) = 0.

From (6.14)
0 (S \ K) (S \ Kn ) (S (Kn \ K))
and so
lim (S \ Kn ) = (S \ K).

From (6.13)
(S) (S K) + (S \ K).
This shows K S and proves the theorem.
The following technical lemma will also prove useful in what follows.
Lemma 6.15 Suppose is a measure defined on a algebra, S of sets of , where (, d) is a metric space
having the property that =
k=1 k where k is a compact set and for all k, k k+1 . Suppose that S
contains the Borel sets and is finite on compact sets. Suppose that also has the property that for every
E S,
(E) = inf { (V ) : V E, V open} .

(6.17)

(E) = sup { (K) : K E, K compact} .

(6.18)

Then it follows that for all E S

Proof: Let E S and let l < (E) . By Theorem 5.11 we may choose k large enough that
l < (E k ) .
Now let F k \ E. Thus F (E k ) = k . By assumption, there is an open set, V containing F with
(V ) (F ) = (V \ F ) < (E k ) l.
We define the compact set, K V C k . Then K E k and
E k \ K = E k V C
k

= E k V k F C V V \ F.

106

THE CONSTRUCTION OF MEASURES

Therefore,
(E k ) (K) = ((E k ) \ K)
(V \ F ) < (E k ) l
which implies
l < (K) .
This proves the lemma because l < (E) was arbitrary.
Definition 6.16 We say a measure which satisfies (6.17) for all E measurable, is outer regular and a
measure which satisfies (6.18) for all E measurable is inner regular. A measure which satisfies both is called
regular.
Thus Lemma 6.15 gives a condition under which outer regular implies inner regular.
With this preparation we are ready to prove the Riesz representation theorem for positive linear functionals.
Theorem 6.17 Let (, d) be a compact metric space with the property that the closures of balls are compact
and let be a positive linear functional on Cc () . Then there exists a unique algebra and measure, ,
such that
is complete, Borel, and regular,

(6.19)

(K) < for all K compact,

(6.20)

f =

f d for all f Cc () .

(6.21)

Such measures satisfying (6.19) and (6.20) are called Radon measures.
Proof: First we deal with the question of existence and then we will consider uniqueness. In all that
follows V will denote an open set and K will denote a compact set. Define
(V ) sup { (f ) : f V } , () 0,

(6.22)

and for an arbitrary set, T,

(T ) inf { (V ) : V T } .
We need to show first that this is well defined because there are two ways of defining (V ) .
Lemma 6.18 is a well defined outer measure on P () .
Proof: First we consider the question of whether is well defined. To clarify the argument, denote by
1 the first definition for open sets given in (6.22).
(V ) inf {1 (U ) : U V } 1 (V ) .
But also, whenever U V, 1 (U ) 1 (V ) and so
(V ) 1 (V ) .

6.2. POSITIVE LINEAR FUNCTIONALS

107

This proves that is well defined. Next we verify is an outer measure. It is clear that if A B then
(A) (B) . First we verify countable subadditivity for open sets. Thus let V =
i=1 Vi and let l < (V ) .
Then there exists f V such that f > l. Now spt (f ) is a compact subset of V and so there exists m such
m
that {Vi }i=1 covers spt (f ) . Then, letting i be a partition of unity from Theorem 6.11 with spt ( i ) Vi ,
it follows that
l < (f ) =

n
X

( i f )

i=1

(Vi ) .

i=1

Since l < (V ) is arbitrary, it follows that

(V )

(Vi ) .

i=1

Now we must verify that for any sets, Ai ,

(
i=1 Ai )

(Ai ) .

i=1

It suffices to consider the case that (Ai ) < for all i. Let Vi Ai and (Ai ) +
countable subadditivity on open sets,

2i

> (Vi ) . Then from

(
i=1 Ai ) (i=1 Vi )

X
i=1

(Vi )

X
i=1

(Ai ) +

+
(Ai ) .
2i
i=1

Since is arbitrary, this proves the lemma.

We will denote by S the algebra of measurable sets.
Lemma 6.19 The outer measure, is finite on all compact sets and in fact, if K g, then
(K) (g)

(6.23)

Also S Borel sets so is a Borel measure.

Proof: Let V {x : g (x) > } where (0, 1) is arbitrary. Now let h V . Thus h (x) 1 and
equals zero off V while 1 g (x) 1 on V . Therefore,


1 g (h) .
Since h V was arbitrary, this shows 1 (g) (V ) (K) . Letting 1 yields the formula (6.23).
Next we verify that S Borel sets. First suppose that V1 and V2 are disjoint open sets with (V1 V2 ) <
. Let fi Vi be such that (fi ) + > (Vi ) . Then
(V1 V2 ) (f1 + f2 ) = (f1 ) + (f2 ) (V1 ) + (V2 ) 2.
Since is arbitrary, this shows that (V1 V2 ) = (V1 ) + (V2 ) .
Now suppose that dist (A, B) = r > 0 and (A B) < . Let
o
n  r
o
n  r
: a A , Ve2 B b,
:bB .
Ve1 B a,
2
2

108

THE CONSTRUCTION OF MEASURES

Now let W be an open set containing A B such that (A B) + > (W ) . Now define Vi W Vei and
V V1 V2 . Then
(A B) + > (W ) (V ) = (V1 ) + (V2 ) (A) + (B) .
Since is arbitrary, the conditions of Caratheodorys criterion are satisfied showing that S Borel sets.
This proves the lemma.
It is now easy to verify condition (6.19) and (6.20). Condition (6.20) and that is Borel is proved in
Lemma 6.19. The measure space just described is complete because it comes from an outer measure using
the Caratheodory procedure. The construction of shows outer regularity and the inner regularity follows
from (6.20), shown in Lemma 6.19, and Lemma 6.15. It only remains to verify Condition (6.21), that the
measure reproduces the functional in the desired manner and that the given measure and algebra is unique.
R
Lemma 6.20 f d = f for all f Cc ().
Proof: It suffices to verify this for f Cc (), f real-valued. Suppose f is such a function and
f () [a, b]. Choose t0 < a and let t0 < t1 < < tn = b, ti ti1 < . Let
Ei = f 1 ((ti1 , ti ]) spt(f ).

(6.24)

Note that ni=1 Ei is a closed set and in fact

ni=1 Ei = spt(f )

(6.25)

since = ni=1 f 1 ((ti1 , ti ]). From outer regularity and continuity of f , let Vi Ei , Vi is open and let Vi
satisfy
f (x) < ti + for all x Vi ,

(6.26)

(Vi \ Ei ) < /n.

By Theorem 6.11 there exists hi Cc () such that
hi V i ,

n
X

hi (x) = 1 on spt(f ).

i=1

Now note that for each i,

f (x)hi (x) hi (x)(ti + ).
(If x Vi , this follows from (6.26). If x
/ Vi both sides equal 0.) Therefore,
f

n
n
X
X
hi (ti + ))
f hi ) (
(

n
X

i=1

i=1
n
X
i=1

i=1

(ti + )(hi )
(|t0 | + ti + )(hi ) |t0 |

n
X

hi .

i=1

Now note that |t0 | + ti + 0 and so from the definition of and Lemma 6.19, this is no larger than
n
X
i=1

(|t0 | + ti + )(Vi ) |t0 |(spt(f ))

6.2. POSITIVE LINEAR FUNCTIONALS

n
X

109

(|t0 | + ti + )((Ei ) + /n) |t0 |(spt(f ))

i=1

|t0 |

n
X

(Ei ) + |t0 | +

i=1

n
X

ti (Ei ) + (|t0 | + |b|)

i=1

n
X

(Ei ) + 2 |t0 |(spt(f )).

i=1

From (6.25) and (6.24), the first and last terms cancel. Therefore this is no larger than
(2|t0 | + |b| + (spt(f )) + ) +

n
X

ti1 (Ei ) + (spt(f ))

i=1

f d + (2|t0 | + |b| + 2(spt(f )) + ).

Since > 0 is arbitrary,

f

f d

(6.27)

R
R
for all Rf Cc (), f real. Hence equality holds in (6.27) because (f ) f d so (f ) f d. Thus
f = f d for all f Cc (). Just apply the result for real functions to the real and imaginary parts of f .
This proves the Lemma.
Now that we have shown that satisfies the conditions of the Riesz representation theorem, we show
that is the only measure that does so.
Lemma 6.21 The measure and algebra of Theorem 6.17 are unique.
Proof: If (1 , S1 ) and (2 , S2 ) both work, let
K V, K f V.
Then
1 (K)

f d1 = f =

f d2 2 (V ).

Thus 1 (K) 2 (K) because of the outer regularity of 2 . Similarly, 1 (K) 2 (K) and this shows that
1 = 2 on all compact sets. It follows from inner regularity that the two measures coincide on all open sets
as well. Now let E S1 , the algebra associated with 1 , and let En = E n . By the regularity of the
measures, there exist sets G and H such that G is a countable intersection of decreasing open sets and H is
a countable union of increasing compact sets which satisfy
G En H, 1 (G \ H) = 0.
Since the two measures agree on all open and compact sets, it follows that 2 (G) = 1 (G) and a similar
equation holds for H in place of G. Therefore 2 (G \ H) = 1 (G \ H) = 0. By completeness of 2 , En S2 ,
the algebra associated with 2 . Thus E S2 since E =
n=1 En , showing that S1 S2 . Similarly S2 S1 .

110

THE CONSTRUCTION OF MEASURES

Since the two algebras are equal and the two measures are equal on every open set, regularity of these
measures shows they coincide on all measurable sets and this proves the theorem.
The following theorem is an interesting application of the Riesz representation theorem for measures
defined on subsets of Rn .
Let M be a closed subset of Rn . Then we may consider M as a metric space which has closures of balls
compact if we let the topology on M consist of intersections of open sets from the standard topology of Rn
with M or equivalently, use the usual metric on Rn restricted to M .
Proposition 6.22 Let be the relative topology of M consisting of intersections of open sets of Rn with M
and let B be the Borel sets of the topological space (M, ). Then
B = S {E M : E is a Borel set of Rn }.
Proof: It is clear that S defined above is a algebra containing and so S B. Now define
M {E Borel in Rn such that E M B} .
Then M is clearly a algebra which contains the open sets of Rn . Therefore, M Borel sets of Rn which
shows S B. This proves the proposition.
Theorem 6.23 Suppose is a measure defined on the Borel sets of M where M is a closed subset of Rn .
Suppose also that is finite on compact sets. Then , the outer measure determined by , is a Radon
measure on a algebra containing the Borel sets of (M, ) where is the relative topology described above.
Proof: Since is Borel and finite on compact sets, we may define a positive linear functional on Cc (M )
as
Lf

f d.

By the Riesz representation theorem, there exists a unique Radon measure and algebra, 1 and S (1 ) respectively,
such that for all f Cc (M ),
Z
Z
f d =
f d1 .
M

Let R and E be as described in Example 5.8 and let Qr = (r, r]n . Then if R R, it follows that R Qr
has the form,
R Qr =

n
Y

(ai , bi ].

i=1

Let fk (x1 , xn ) =

Qn

i=1

hki (xi ) where hki is given by the following graph.

1


hki

B
B





ai ai + k 1

B
bi + k 1

bi

Then spt (fk ) [r, r + 1]n K. Thus

Z
Z
fk d =
KM

fk d =

Z
M

fk d1 =

Z
KM

fk d1 .

6.3. EXERCISES

111

Since fk XRQr pointwise and both measures are finite, on K M , it follows from the dominated
convergence theorem that
(R Qr M ) = 1 (R Qr M )
and so it follows that this holds for R replaced with any A E. Now define
M {E Borel : (E Qr M ) = 1 (E Qr M )}.
Then E M and it is clear that M is a monotone class. By the theorem on monotone classes, it follows
that M (E), the smallest algebra containing E. This algebra contains the open sets because
n
Y

(ai , bi ) =
k=1

i=1

n
Y

(ai , bi k 1 ] (E)

i=1

and every open set can be written as a countable union of sets of the form

Qn

i=1

(ai , bi ). Therefore,

M (E) Borel sets M.

Thus,
(E Qr M ) = 1 (E Qr M )
for all E Borel. Letting r , it follows that (E M ) = 1 (E M ) for all E a Borel set of Rn . By
Proposition 6.22 (F ) = 1 (F ) for all F Borel in (M, ). Consequently,
1 (S)
=
=
=

inf {1 (E) : E S, E S (1 )}
inf {1 (F ) : F S, F Borel}
inf { (F ) : F S, F Borel}
(S).

Therefore, by Lemma 6.6, the measurable sets consist of S (1 ) and = 1 on S (1 ) and this shows is
regular as claimed. This proves the theorem.

6.3

Exercises

1. Let = N, the natural numbers and let d (p, q) = |p q| , the usual distance
in R. Show that (, d) is
P
compact and the closures of the balls are compact. Now let f k=1 f (k) whenever f Cc () .
Show this is a well defined positive linear functional on the space Cc () . Describe the measure of the
Riesz representation theorem which results from this positive linear functional. What if (f ) = f (1)?
What measure would result from this functional?
R
2. Let F : R R be increasing and right continuous. Let f f dF where the integral is the Riemann
Stieltjes integral of f . Show the measure from the Riesz representation theorem satisfies
([a, b]) = F (b) F (a) , ((a, b]) = F (b) F (a) ,
([a, a]) = F (a) F (a) .
3. Let be a compact metric space with the closed balls compact and suppose is a measure defined
on the Borel sets of which is finite on compact sets. Show there exists a unique Radon measure,
which equals on the Borel sets.

112

THE CONSTRUCTION OF MEASURES

4. Random vectors are measurable functions, X, mapping a probability space, (, P, F) to Rn . Thus

X () Rn for each and P is a probability measure defined on the sets of F, a algebra of
subsets of . For E a Borel set in Rn , define


(E) P X1 (E) probability that X E.
Show this is a well defined measure on the Borel sets of Rn and use Problem 3 to obtain a Radon
measure, X defined on a algebra of sets of Rn including the Borel sets such that for E a Borel set,
X (E) =Probability that (X E) .
5. Let (X, dX ) and (Y, dY ) be metric spaces and make X Y into a metric space in the following way.
dXY ((x, y) , (x1 , y1 )) max (dX (x, x1 ) , dY (y, y1 )) .
Show this is a metric space.
6. Show (X Y, dXY ) is also a compact metric space having closed balls compact if both X and Y
are compact metric spaces having the closed balls compact. Let
A {E F : E is a Borel set in X, F is a Borel set in Y } .
Show (A) , the smallest algebra containing A contains the Borel sets. Hint: Show every open set
in a compact metric space can be obtained as a countable union of compact sets. Next show this
implies every open set can be obtained as a countable union of open sets of the form U V where U
is open in X and V is open in Y.
7. Let and be Radon measures on X and Y respectively. Define for
f Cc (X Y ) ,
the linear functional given by the iterated integral,
Z Z
f
f (x, y) dd.
X

Show this is well defined and yields a positive linear functional on Cc (X Y ) . Let be the Radon
measure representing . Show for f 0 and Borel measurable, that
Z Z
Z Z
Z
f (x, y) dd =
f (x, y) dd =
f d ( )
Y

XY

Lebesgue Measure
7.1

Lebesgue measure

In this chapter, n dimensional Lebesgue measure and many of its properties are obtained from the Riesz
representation theorem. This is done by using a positive linear functional familiar to anyone who has had a
course in calculus. The positive linear functional is
Z
Z
f

f (x1 , , xn )dx1 dxn

(7.1)

for f Cc (Rn ). This is the ordinary Riemann iterated integral and we need to observe that it makes sense.
Lemma 7.1 Let f Cc (Rn ) for n 2. Then
Z
Z
h(xn )

f (x1 xn1 xn )dx1 dxn1

is well defined and h Cc (R).

Proof: Assume this is true for all 2 k n 1. Then fixing xn ,
Z
Z
xn1

f (x1 xn2 , xn1 , xn )dx1 dxn2

is a function in Cc (R). Therefore, it makes sense to write

Z Z
Z
h(xn )

f (x1 xn2 , xn1 , xn )dx1 dxn1 .

We need to verify h Cc (R). Since f vanishes whenever |x| is large enough, it follows h(xn ) = 0 whenever
|xn | is large enough. It only remains to show h is continuous. But f is uniformly continuous, so if > 0 is
given there exists a such that
|f (x1 ) f (x)| <
whenever |x1 x| < . Thus, letting |xn x
n | < ,
|h(xn ) h(
xn )|
Z

|f (x1 xn1 , xn ) f (x1 xn1 , x

n )|dx1 dxn1

113

114

LEBESGUE MEASURE
(b a)n1

where spt(f ) [a, b]n [a, b] [a, b]. This argument also shows the lemma is true for n = 2. This
proves the lemma.
From Lemma 7.1 it is clear that (7.1) makes sense and also that is a positive linear functional for
n = 1, 2, .
Definition 7.2 mn is the unique Radon measure representing . Thus for all f Cc (Rn ),
Z
f = f dmn .
Qn
Qn
LetQR = i=1 [ai , bi ], R0 = i=1 (ai , bi ). What are mn (R) and mn (R0 )? We show that both of these
n
equal i=1 (bi ai ). To see this is the case, let k be large enough that
ai + 1/k < bi 1/k
for i = 1, , n. Consider functions gik and fik having the following graphs.
fik

1





ai

B
bi 1/k

B
B

ai + 1/k

bi
gik

1




ai 1/k

B
B
ai

bi

bi + 1/k

Let
k

g (x) =

n
Y

gik (xi ),

f (x) =

i=1

n
Y

fik (xi ).

i=1

Then
n
Y

(bi ai + 2/k) g =

g k dmn mn (R) mn (R0 )

i=1

f k dmn = f k

n
Y

i=1

(bi ai 2/k).

7.1. LEBESGUE MEASURE

115

Letting k , it follows that

mn (R) = mn (R0 ) =

n
Y

(bi ai )

i=1

as expected.
We say R is a half open box if
R=

n
Y

[ai , ai + r).

i=1

Lemma 7.3 Every open set in Rn is the countable disjoint union of half open boxes of the form
n
Y

(ai , ai + 2k ]

i=1

where ai = l2k for some integers, l, k.

Proof: Let
Ck = {All half open boxes

n
Y

(ai , ai + 2k ] where

i=1

ai = l2k for some integer l.}

Thus Ck consists of a countable disjoint collection
of boxes whose union is Rn . This is sometimes called a
n
k
n. Let U be open and let B1 all sets of C1 which are
tiling of R . Note that each box has diameter 2
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 . We claim B = U . Clearly B 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 . This proves the
lemma.
Lebesgue measure is translation invariant. This means roughly that if you take a Lebesgue measurable
set, and slide it around, it remains Lebesgue measurable and the measure does not change.
Theorem 7.4 Lebesgue measure is translation invariant, i.e.,
mn (v+E) = mn (E),
for E Lebesgue measurable.
Proof: First note that if E is Borel, then so is v + E. To show this, let
S = {E Borel sets such that v + E is Borel}.
Then from Lemma 7.3, S contains the open sets and is easily seen to be a algebra, so S = Borel sets. Now
let E be a Borel set. Choose V open such that
mn (V ) < mn (E B(0, k)) + , V E B(0, k).

116

LEBESGUE MEASURE

Then
mn (v + E B(0, k)) mn (v + V ) = mn (V )
mn (E B(0, k)) + .
The equal sign is valid because the conclusion of Theorem 7.4 is clearly true for all open sets thanks to
Lemma 7.3 and the simple observation that the theorem is true for boxes. Since is arbitrary,
mn (v + E B(0, k)) mn (E B(0, k)).
Letting k ,
mn (v + E) mn (E).
Since v is arbitrary,
mn (v + (v + E)) mn (E + v).
Hence
mn (v + E) mn (E) mn (v + E)
proving the theorem in the case where E is Borel. Now suppose that mn (S) = 0. Then there exists E S, E
Borel, and mn (E) = 0.
mn (E+v) = mn (E) = 0
Now S + v E + v and so by completeness of the measure, S + v is Lebesgue measurable and has measure
zero. Thus,
mn (S) = mn (S + v) .
Now let F be an arbitrary Lebesgue measurable set and let Fr = F B (0, r). Then there exists a Borel set
E, E Fr , and mn (E \ Fr ) = 0. Then since
(E + v) \ (Fr + v) = (E \ Fr ) + v,
it follows
mn ((E + v) \ (Fr + v)) = mn ((E \ Fr ) + v) = mn (E \ Fr ) = 0.
By completeness of mn , Fr + v is Lebesgue measurable and mn (Fr + v) = mn (E + v). Hence
mn (Fr ) = mn (E) = mn (E + v) = mn (Fr + v).
Letting r , we obtain
mn (F ) = mn (F + v)
and this proves the theorem.

7.2. ITERATED INTEGRALS

7.2

117

Iterated integrals

The positive linear functional used to define Lebesgue measure was an iterated integral. Of course one could
take the iterated integral in another order. What would happen to the resulting Radon measure if another
order was used? This question will be considered in this section. First, here is a simple lemma.
Lemma 7.5 If and are two Radon measures defined on algebras, S and S , of subsets of Rn and if
(V ) = (V ) for all V open, then = and S = S .
Proof: Let and be the outer measures determined by and . Then if E is a Borel set,
(E)

= inf { (V ) : V E and V is open}

= inf { (V ) : V E and V is open} = (E).

Now if S is any subset of Rn ,

(S) inf { (E) : E S, E S }
= inf { (F ) : F S, F is Borel}
= inf { (F ) : F S, F is Borel}
= inf { (E) : E S, E S } (S)
where the second and fourth equalities follow from the outer regularity which is assumed to hold for both
measures. Therefore, the two outer measures are identical and so the measurable sets determined in the sense
of Caratheodory are also identical. By Lemma 6.6 of Chapter 6 this implies both of the given algebras are
equal and = .
Lemma 7.6 If and are two Radon measures on Rn and = on every half open box, then = .
Proof: From Lemma 7.3, (U ) = (U ) for all U open. Therefore, by Lemma 7.5, the two measures
coincide with their respective algebras.
Corollary 7.7 Let {1, 2, , n} = {k1 , k2 , , kn } and the ki are distinct. For f Cc (Rn ) let
Z
Z
=
f

f (x1 , , xn )dxk1 dxkn ,

an iterated integral in a different order. Then for all f Cc (Rn ),

f = f,
e and mn is the Radon measure representing mn , then mn =
and if m
e n is the Radon measure representing
m
e n.

Then clearly mn = m
Proof: Let m
n be the Radon measure representing .
n on every half open box.
By Lemma 7.6, mn = m
n . Thus,
Z
Z

f = f dmn = f dm
n = f.
This Corollary 7.7 is pretty close to Fubinis theorem for Riemann integrable functions. Now we will
generalize it considerably by allowing f to only be Borel measurable. To begin with, we consider the question
of existence of the iterated integrals for XE where E is a Borel set.

118

LEBESGUE MEASURE

Lemma
7.8 Let E be a Borel set and let {k1 , k2 , , kn } distinct integers from {1, , n} and if Qk
Qn
(p,
p]. Then for each r = 2, , n 1, the function,
i=1
r integrals

xkr+1

z }| Z{
Z
XEQp (x1 , , xn ) dm (xk1 ) dm (xkr )

(7.2)

is Lebesgue measurable. Thus we can add another iterated integral and write
r integrals

Z zZ }| Z{


XEQp (x1 , , xn ) dm (xk1 ) dm (xkr ) dm xkr+1 .
Here the notation dm (xki ) means we integrate the function of xki with respect to one dimensional Lebesgue
measure.
Proof: If E is an element of E, the algebra of Example 5.8, we leave the conclusion of this lemma to
the reader. If M is the collection of Borel sets such that (7.2) holds, then the dominated convergence and
monotone convergence theorems show that M is a monotone class. Therefore, M equals the Borel sets by
the monotone class theorem. This proves the lemma.
The following lemma is just a generalization of this one.
Lemma 7.9 Let f be any nonnegative Borel measurable function. Then for each r = 2, , n 1, the
function,
r integrals

xkr+1

zZ }| Z{
f (x1 , , xn ) dm (xk1 ) dm (xkr )

(7.3)

is one dimensional Lebesgue measurable.

Proof: Letting p in the conclusion of Lemma 7.8 we see the conclusion of this lemma holds without
the intersection with Qp . Thus, if s is a nonnegative Borel measurable function (7.3) holds with f replaced
with s. Now let sn be an increasing sequence of nonnegative Borel measurable functions which converge
pointwise to f. Then by the monotone convergence theorem applied r times,
r integrals

zZ }| Z{
sn (x1 , , xn ) dm (xk1 ) dm (xkr )
lim

r integrals

zZ }| Z{
f (x1 , , xn ) dm (xk1 ) dm (xkr )

and so we may draw the desired conclusion because the given function of xkr+1 is a limit of a sequence of
measurable functions of this variable.
To summarize this discussion, we have shown that if f is a nonnegative Borel measurable function, we
may take the iterated integrals in any order and everything which needs to be measurable in order for the
expression to make sense, is. The next lemma shows that different orders of integration in the iterated
integrals yield the same answer.
Lemma 7.10 Let E be any Borel set. Then
Z
Z
Z
XE (x) dmn = XE (x1 , , xn ) dm (x1 ) dm (xn )
Rn

7.2. ITERATED INTEGRALS

=

119

XE (x1 , , xn ) dm (xk1 ) dm (xkn )

(7.4)

where {k1 , , kn } = {1, , n} and everything which needs to be measurable is. Here the notation involving
the iterated integrals refers to one-dimensional Lebesgue integrals.
Proof: Let Qk = (k, k]n and let
M {Borel sets, E, such that (7.4) holds for E Qk
and there are no measurability problems} .
If E is the algebra of Example 5.8, we see easily that for all such sets, A of E, (7.4) holds for A Qk and so
M E. Now the theorem on monotone classes implies M (E) which, by Lemma 7.3, equals the Borel
sets. Therefore, M = Borel sets. Letting k , and using the monotone convergence theorem, yields the
conclusion of the lemma.
The next theorem is referred to as Fubinis theorem. Although a more abstract version will be presented
later, the version in the next theorem is particularly useful when dealing with Lebesgue measure.
Theorem 7.11 Let f 0 be Borel measurable. Then everything which needs to be measurable in order to
write the following formulae is measurable, and
Z
Z
Z
f (x) dmn = f (x1 , , xn ) dm (x1 ) dm (xn )
Rn

f (x1 , , xn ) dm (xk1 ) dm (xkn )

where {k1 , , kn } = {1, , n}.

Proof: This follows from the previous lemma since the conclusion of this lemma holds for nonnegative
simple functions in place of XE and we may obtain f as the pointwise limit of an increasing sequence of
nonnegative simple functions. The conclusion follows from the monotone convergence theorem.
Corollary 7.12 Suppose f is complex valued and for some
{k1 , , kn } = {1, , n},
it follows that
Z

|f (x1 , , xn )| dm (xk1 ) dm (xkn ) < .

Then f L1 (Rn , mn ) and if {l1 , , ln } = {1, , n}, then

Z
Z
Z
f dmn = f (x1 , , xn ) dm (xl1 ) dm (xln ).

(7.5)

(7.6)

Rn

Proof: Applying Theorem 7.11 to the positive and negative parts of the real and imaginary parts of
f, (7.5) implies all these integrals are finite and all iterated integrals taken in any order are equal for these
functions. Therefore, the definition of the integral implies (7.6) holds.

120

7.3

LEBESGUE MEASURE

Change of variables

In this section we show that if F L (Rn , Rn ) , then mn (F (E)) = F mn (E) whenever E is Lebesgue
measurable. The constant F will also be shown to be |det (F )|. In order to prove this theorem, we recall
Theorem 2.29 which is listed here for convenience.
Theorem 7.13 Let F L (Rn , Rm ) where m n. Then there exists R L (Rn , Rm ) and U L (Rn , Rn )
such that U = U , all eigenvalues of U are nonnegative,
U 2 = F F, R R = I, F = RU,
and |Rx| = |x|.
The following corollary follows as a simple consequence of this theorem.
Corollary 7.14 Let F L (Rn , Rm ) and suppose n m. Then there exists a symmetric nonnegative
element of L(Rm , Rm ), U, and an element of L(Rn , Rm ), R, such that
F = U R, RR = I.

Proof: We recall that if M, L L(Rs , Rp ), then L = L and (M L) = L M . Now apply Theorem

2.29 to F L(Rm , Rn ). 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.
The next few lemmas involve the consideration of F (E) where E is a measurable set. They show that
F (E) is Lebesgue measurable. We will have occasion to establish similar theorems in other contexts later
in the book. In each case, the overall approach will be to show the mapping in question takes sets of
measure zero to sets of measure zero and then to exploit the continuity of the mapping and the regularity
and completeness of some measure to obtain the final result. The next lemma gives the first part of this
procedure here. First we give a simple definition.
Definition 7.15 Let F L (Rn , Rm ) . We define ||F || max {|F (x)| : |x| 1} . This number exists because
the closed unit ball is compact.
Now we note that from this definition, if v is any nonzero vector, then
 

 


v
v


|F (v)| = F
|v|  = |v| F
||F || |v| .
|v|
|v|

Lemma 7.16 Let F L (Rn , Rn ), and suppose E is a measurable set having finite measure. Then

n
mn (F (E)) 2 ||F || n mn (E)
where mn () refers to the outer measure,
mn (S) inf {mn (E) : E S and E is measurable} .

7.3. CHANGE OF VARIABLES

121

Proof: Let  > 0 be given and let E V, an open set with mn (V ) < mn (E) + . Then let V =
i=1 Qi
where Qi is ahalf open box all of whose sides have length 2l for some l N and Qi Qj = if i 6= j. Then
diam (Qi ) = nai where ai is the length of the sides of Qi . Thus, if yi is the center of Qi , then
B (F yi , ||F || diam (Qi )) F Qi .
Let Qi denote the cube with sides of length 2 ||F || diam (Qi ) and center at F yi . Then
Qi B (F yi , ||F || diam (Qi )) F Qi
and so
mn (F (E)) mn (F (V ))

(2 ||F || diam (Qi ))

i=1

n

n X
(mn (Qi )) = 2 ||F || n mn (V )
2 ||F || n
i=1

n
2 ||F || n (mn (E) + ).
Since  > 0 is arbitrary, this proves the lemma.
Lemma 7.17 If E is Lebesgue measurable, then F (E) is also Lebesgue measurable.
Proof: First note that if K is compact, F (K) is also compact and is therefore a Borel set. Also, if V is
an open set, it can be written as a countable union of compact sets, {Ki } , and so
F (V ) =
i=1 F (Ki )
which shows that F (V ) is a Borel set also. Now take any Lebesgue measurable set, E, which is bounded
and use regularity of Lebesgue measure to obtain open sets, Vi , and compact sets, Ki , such that
Ki E Vi ,
Vi Vi+1 , Ki Ki+1 , Ki E Vi , and mn (Vi \ Ki ) < 2i . Let

Q
i=1 F (Vi ), P i=1 F (Ki ).

Thus both Q and P are Borel sets (hence Lebesgue) measurable. Observe that
Q \ P i,j (F (Vi ) \F (Kj ))

i=1 F (Vi ) \ F (Ki ) i=1 F (Vi \ Ki )

which means, by Lemma 7.16,

n
mn (Q \ P ) mn (F (Vi \ Ki )) 2 ||F || n 2i
which implies Q \ P is a set of Lebesgue measure zero since i is arbitrary. Also,
P F (E) Q.
By completeness of Lebesgue measure, this shows F (E) is Lebesgue measurable.
If E is not bounded but is measurable, consider E B (0, k). Then
F (E) =
k=1 F (E B (0, k))
and, thus, F (E) is measurable. This proves the lemma.

122

LEBESGUE MEASURE

Lemma 7.18 Let Q0 [0, 1)n and let F mn (F (Q0 )). Then if Q is any half open box whose sides are
of length 2k , k N, and F is one to one, it follows
mn (F Q) = mn (Q) F
and if F is not one to one we can say
mn (F Q) mn (Q) F.


k n

n
Proof: There are 2
l nonintersecting half open boxes, Qi , each having measure 2k whose
union equals Q0 . If F is one to one, translation invariance of Lebesgue measure and the assumption F is
linear imply
2k

n

mn (F (Q)) =

l
X

mn (F (Qi )) =

i=1

mn li=1 F (Qi ) = mn (F (Q0 )) = F.

()

Therefore,
mn (F (Q)) = 2k

n

F = mn (Q) F.

If F is not one to one, the sets F (Qi ) are not necessarily disjoint and so the second equality sign in ()
should be . This proves the lemma.
Theorem 7.19 If E is any Lebesgue measurable set, then
mn (F (E)) = F mn (E) .
If R R = I and R preserves distances, then
R = 1.
Also, if F, G L (Rn , Rn ) , then
(F G) = F G.

(7.7)

Proof: Let V be any open set and let {Qi } be half open disjoint boxes of the sort discussed earlier whose
union is V and suppose first that F is one to one. Then
mn (F (V )) =

mn (F (Qi )) = F

i=1

mn (Qi ) = F mn (V ).

(7.8)

i=1

Now let E be an arbitrary bounded measurable set and let Vi be a decreasing sequence of open sets containing
E with
i1 mn (Vi \ E).
Then let S
i=1 F (Vi )

S \ F (E) =
i=1 (F (Vi ) \ F (E)) i=1 F (Vi \ E)

and so from Lemma 7.16,

mn (S \ F (E)) lim sup mn (F (Vi \ E))
i

7.3. CHANGE OF VARIABLES

123

n
lim sup 2 ||F || n i1 = 0.
i

Thus
mn (F (E)) = mn (S) = lim mn (F (Vi ))
i

= lim F mn (Vi ) = F mn (E) .

i

(7.9)

If E is not bounded, apply the above to E B (0, k) and let k .

To see the second claim of the theorem,
Rmn (B (0, 1)) = mn (RB (0, 1)) = mn (B (0, 1)) .
Now suppose F is not one to one. Then let {v1 , , vn } be an orthonormal basis of Rn such that for
some r < n, {v1 , , vr } is an orthonormal basis for F (Rn ). Let Rvi ei where the ei are the standard
unit basis vectors. Then RF (Rn ) span (e1 , , er ) and so by what we know about the Lebesgue measure
of boxes, whose sides are parallel to the coordinate axes,
mn (RF (Q)) = 0
whenever Q is a box. Thus,
mn (RF (Q)) = Rmn (F (Q)) = 0
and this shows that in the case where F is not one to one, mn (F (Q)) = 0 which shows from Lemma 7.18
that mn (F (Q)) = F mn (Q) even if F is not one to one. Therefore, (7.8) continues to hold even if F is
not one to one and this implies (7.9). (7.7) follows from this. This proves the theorem.
Lemma 7.20 Suppose U = U and U has all nonnegative eigenvalues, {i }. Then
U =

n
Y

i .

i=1

Proof: Suppose U0

Pn

i=1

i ei ei . Note that
( n
)
X
Q0 =
ti ei : ti [0, 1) .
i=1

Thus
U0 (Q0 ) =

( n
X

i ti ei : ti [0, 1)

i=1

and so
U0 mn (U0 Q0 ) =

n
Y

i=1

Now by linear algebra, since U = U ,

U=

n
X
i=1

i vi vi

i .

124

LEBESGUE MEASURE

where {vi } is an orthonormal basis of eigenvectors. Define R L (Rn ,Rn ) such that
Rvi = ei .
Then R preserves distances and RU R = U0 where U0 is given above. Therefore, if E is any measurable set,
n
Y

i mn (E) = U0 mn (E) = mn (U0 (E)) = mn (RU R (E))

i=1

= RU R mn (E) = U mn (E).
Hence

Qn

i=1

i = U as claimed. This proves the theorem.

Theorem 7.21 Let F L (Rn , Rn ). Then F = |det (F )|. Thus

mn (F (E)) = |det (F )| mn (E)
for all E Lebesgue measurable.
Proof: By Theorem 2.29, F = RU where R and U are described in that theorem. Then
F = RU = U = det (U ).


2
2
Now F F = U 2 and so (det (U )) = det U 2 = det (F F ) = (det (F )) . Therefore,
det (U ) = |det F |
and this proves the theorem.

7.4

Polar coordinates

One of the most useful of all techniques in establishing estimates which involve integrals taken with respect
to Lebesgue measure on Rn is the technique of polar coordinates. This section presents the polar coordinate
formula. To begin with we give a general lemma.
Lemma 7.22 Let X and Y be topological spaces. Then if E is a Borel set in X and F is a Borel set in Y,
then E F is a Borel set in X Y .
Proof: Let E be an open set in X and let
SE {F Borel in Y such that E F is Borel in X Y } .
Then SE contains the open sets and is clearly closed with respect to countable unions. Let F SE . Then
E F C E F = E Y = a Borel set.
Therefore, since E F is Borel, it follows E F C is Borel. Therefore, SE is a algebra. It follows SE =
Borel sets, and so, we have shown- open Borel = Borel. Now let F be a fixed Borel set in Y and define
SF {E Borel in X such that E F is Borel in X Y }.
The same argument which was just used shows SF is a algebra containing the open sets. Therefore, SF =
the Borel sets, and this proves the lemma since F was an arbitrary Borel set.

7.4. POLAR COORDINATES

125

Now we define the unit sphere in Rn , S n1 , by

S n1 {w Rn : |w| = 1}.
Then S n1 is a compact metric space using the usual metric on Rn . We define a map
: S n1 (0, ) Rn \ {0}
by
(w,) w.
It is clear that is one to one and onto with a continuous inverse. Therefore, if B1 is the set of Borel sets
in S n1 (0, ), and B are the Borel sets in Rn \ {0}, it follows
B = { (F ) : F B1 }.

(7.10)

Observe also that the Borel sets of S n1 satisfy the conditions of Lemma 5.6 with Z defined as S n1 and
the same is true of the sets (a, b] (0, ) where 0 a, b if Z is defined as (0, ). By Corollary 5.7,
finite disjoint unions of sets of the form

E I : E is Borel in S n1
and I = (a, b] (0, ) where 0 a, b }
form an algebra of sets, A. It is also clear that (A) contains the open sets and so (A) = B1 because every
set in A is in B1 thanks to Lemma 7.22. Let Ar S n1 (0, r] and let

Z
M F B1 :
X(F Ar ) dmn
Rn

S n1

(0,)

X(F Ar ) (w) n1 ddm ,

where for E a Borel set in S n1 ,

(E) nmn ( (E (0, 1))).

(7.11)

E
6
(E (0, 1))

Then if F A, say F = E (a, b], we can show F M. This follows easily from the observation that
Z
Z
Z
X(F ) dmn =
X(E(0,b]) (y) dmn
X(E(0,a]) (y) dmn
Rn

Rn

Rn

= mn ((E (0, 1)) bn mn ((E (0, 1)) an = (E)

(bn an )
,
n

126

LEBESGUE MEASURE

a consequence of the change of variables theorem applied to y =ax, and

Z
Z
Z bZ
X(E(a,b]) (w) n1 ddm =
n1 dd
S n1

(0,)

= (E)

(bn an )
.
n

Since it is clear that M is a monotone class, it follows from the monotone class theorem that M = B 1 .
Letting r , we may conclude that for all F B1 ,
Z
Z
Z
X(F ) dmn =
X(F ) (w) n1 ddm.
Rn

(0,)

S n1

By (7.10), if A is any Borel set in Rn , then A \ {0} = (F ) for some F B1 . Thus

Z
Z
XA dmn =
X(F ) dmn =
Rn

(0,)

Rn

X(F ) (w)

n1

ddm =

S n1

(0,)

XA (w) n1 ddm.

(7.12)

S n1

With this preparation, it is easy to prove the main result which is the following theorem.
Theorem 7.23 Let f 0 and f is Borel measurable on Rn . Then
Z
Z
Z
f (y) dmn =
f (w) n1 ddm
Rn

(0,)

(7.13)

S n1

where is defined by (7.11) and y =w, for w S n1 .

Proof: From (7.12), (7.13) holds for f replaced with a nonnegative simple function. Now the monotone
convergence theorem applied to a sequence of simple functions increasing to f yields the desired conclusion.

7.5

The Lebesgue integral and the Riemann integral

We assume the reader is familiar with the Riemann integral of a function of one variable. It is natural to ask
how this is related to the Lebesgue integral where the Lebesgue integral is taken with respect to Lebesgue
measure. The following gives the essential result.
Lemma 7.24 Suppose f is a non negative Riemann integrable function defined on [a, b] . Then X[a,b] f is
Rb
R
Lebesgue measurable and a f dx = f X[a,b] dm where the first integral denotes the usual Riemann integral
and the second integral denotes the Lebesgue integral taken with respect to Lebesgue measure.
Proof: Since f is Riemann integral, there exist step functions, un and ln of the form
n
X

ci X[ti1 ,ti ) (t)

i=1

such that un f ln and

Z

un dx
a

Z
Z b
 b



un dx
ln dx


 a
a

1
2n

f dx

Z
a

ln dx,

7.6. EXERCISES

127

Rb
Rb
Here a un dx is an upper sum and a ln dx is a lower sum. We also note that these step functions are Borel
Rb
R
measurable simple functions and so a un dx = un dm with a similar formula holding for ln . Replacing ln with
max (l1 , , ln ) if necessary, we may assume that the functions, ln are increasing and similarly, we may assume
the un are decreasing. Therefore, we may define a Borel measurable function, g, by g (x) = limn ln (x)
and a Borel measurable function, h (x) by h (x) = limn un (x) . We claim that f (x) = g (x) a.e. To see
this note that h (x) f (x) g (x) for all x and by the dominated convergence theorem, we can say that
Z
0 = lim
(un vn ) dm
n [a,b]
Z
=
(h g) dm
[a,b]

Therefore, h = g a.e. and so off a Borel set of measure zero, f = g = h. By completeness

of Lebesgue
R
measure, it follows that f is Lebesgue measurable and that both the Lebesgue integral, [a,b] f dm and the
Rb
Riemann integral, a f dx are contained in the interval of length 2n ,
"Z
a

ln dx,

un dx

showing that these two integrals are equal.

7.6

Exercises

1. If A is the algebra of sets of Example 5.8, show (A), the smallest algebra containing the algebra,
is the Borel sets.
2. Consider the following nested sequence of compact sets, {Pn }. The set Pn consists of 2n disjoint closed
intervals contained in [0, 1]. The first interval, P1 , equals [0, 1] and Pn is obtained from Pn1 by deleting
the open interval which is the middle third of each closed interval in Pn . Let P =
n=1 Pn . Show
P 6= , m(P ) = 0, P [0, 1].
(There is a 1-1 onto mapping of [0, 1] to P.) The set P is called the Cantor set.
3. Consider the sequence of functions defined in the following way. We 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 21 (fn (a) + fn (b)) on the
middle third of [a, b]. Sketch a few of these and you will see the pattern. Show
{fn } converges uniformly on [0, 1].

(a.)

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 of Problem 2. This function is called the Cantor function.

(b.)

128

LEBESGUE MEASURE

4. Suppose X, Y are two locally compact, compact, metric spaces. Let A be the collection of finite
disjoint unions of sets of the form E F where E and F are Borel sets. Show that A is an algebra
and that the smallest algebra containing A, (A), contains the Borel sets of X Y . Hint: Show
X Y , with the usual product topology, is a compact metric space. Next show every open set can
be written as a countable union of compact sets. Using this, show every open set can be written as a
countable union of open sets of the form U V where U is an open set in X and V is an open set in
Y.
5. Suppose X, Y are two locally compact, compact, metric spaces and let and be Radon measures
on X and Y respectively. Define for f Cc (X Y ),
Z Z
f
f (x, y) dd.
X

Show this is well defined and is a positive linear functional on

Cc (X Y ) .
Let ( )be the measure representing . Show that for f 0, and f Borel measurable,
Z
Z Z
Z Z
f d ( ) =
f (x, y) dvd =
f (x, y) dd.
XY

Hint: First show, using the dominated convergence theorem, that if E F is the Cartesian product
of two Borel sets each of whom have finite measure, then
Z Z
( ) (E F ) = (E) (F ) =
XEF (x, y) dd.
X


k
2
6. Let f :Rn R be defined by f (x) 1 + |x| . Find the values of k for which f is in L1 (Rn ). Hint:
This is easy and reduces to a one-dimensional problem if you use the formula for integration using
polar coordinates.
7. Let B be a Borel set in Rn and let v be a nonzero vector in Rn . Suppose B has the following property.
For each x Rn , m({t : x + tv B}) = 0. Then show mn (B) = 0. Note the condition on B says
roughly that B is thin in one direction.

8. Let f (y) = X(0,1] (y) sin(1/y)

and let g (y) = X(0,1] (y) 1y . For which values of x does it make sense to
|y|
R
write the integral R f (x y) g (y) dy?
9. If f : Rn [0, ] is Lebesgue measurable, show there exists g : Rn [0, ] such that g = f a.e. and
g is Borel measurable.
10. Let f L1 (R), g L1 (R). Whereever the integral makes sense, define
Z
(f g)(x)
f (x y)g(y)dy.
R

Show the above integral makes sense for a.e. x and that if we define f g (x) 0 at every point where
the above integral does not make sense, it follows that |(f g)(x)| < a.e. and
Z
||f g||L1 ||f ||L1 ||g||L1 . Here ||f ||L1 |f |dx.
Hint: If f is Lebesgue measurable, there exists g Borel measurable with g(x) = f (x) a.e.

7.6. EXERCISES

129

R
11. Let f : [0, ) R be in L1 (R, m). The Laplace transform is given by fb(x) = 0 ext f (t)dt. Let
Rx
f, g be in L1 (R, m), and let h(x) = 0 f (x t)g(t)dt. Show h L1 , and b
h = fbgb.
RA
R
12. Show limA 0 sinx x dx = 2 . Hint: Use x1 = 0 ext dt and Fubinis theorem. This limit is sometimes called the Cauchy principle value. Note that the function sin (x) /x is not in L1 so we are not
finding a Lebesgue integral.
13. Let D consist of functions, g Cc (Rn ) which are of the form
g (x)

n
Y

gi (xi )

i=1

where each gi Cc (R). Show that if f Cc (Rn ) , then there exists a sequence of functions, {gk } in
D which satisfies
lim sup {|f (x) gk (x)| : x Rn } = 0.

(*)

Now for g D given as above, let

0 (g)

Z Y
n

gi (xi ) dm (x1 ) dm (xn ),

i=1

and define, for arbitrary f Cc (Rn ) ,

f lim 0 gk
k

where * holds. Show this is a well-defined positive linear functional which yields Lebesgue measure.
Establish all theorems in this chapter using this as a basis for the definition of Lebesgue measure. Note
this approach is arguably less fussy than the presentation in the chapter. Hint: You might want to
use the Stone Weierstrass theorem.
14. If f : Rn [0, ] is Lebesgue measurable, show there exists g : Rn [0, ] such that g = f a.e. and
g is Borel measurable.

15. Let E be 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 /2i .
16. 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 15 when applied to the rational numbers. (This
problem was shown to me by Lee Earlbach.)

130

LEBESGUE MEASURE

Product Measure
There is a general procedure for constructing a measure space on a algebra of subsets of the Cartesian
product of two given measure spaces. This leads naturally to a discussion of iterated integrals. In calculus,
we learn how to obtain multiple integrals by evaluation of iterated integrals. We are asked to believe that
the iterated integrals taken in the different orders give the same answer. The following simple example shows
that sometimes when iterated integrals are performed in different orders, the results differ.
Example 8.1 Let 0 < 1 < 2 < R < n < 1, limn n = 1. Let gn be a real continuous function with
1
gn = 0 outside of ( n , n+1 ) and 0 gn (x)dx = 1 for all n. Define
f (x, y) =

(gn (x) gn+1 (x))gn (y).

n=1

Then you can show the following:

a.) f is continuous on [0, 1) [0, 1)
R1R1
R1R1
b.) 0 0 f (x, y)dydx = 1 , 0 0 f (x, y)dxdy = 0.
Nevertheless, it is often the case that the iterated integrals are equal and give the value of an appropriate
multiple integral. The best theorems of this sort are to be found in the theory of Lebesgue integration and
this is what will be discussed in this chapter.
Definition 8.2 A measure space (X, F, ) is said to be finite if
X =
n=1 Xn , Xn F, (Xn ) < .
In the rest of this chapter, unless otherwise stated, (X, S, ) and (Y, F, ) will be two finite measure
spaces. Note that a Radon measure on a compact, locally compact space gives an example of a finite
space. In particular, Lebesgue measure is finite.
Definition 8.3 A measurable rectangle is a set A B X Y where A S, B F. An elementary set
will be any subset of X Y which is a finite union of disjoint measurable rectangles. S F will denote the
smallest algebra of sets in P(X Y ) containing all elementary sets.
Example 8.4 It follows from Lemma 5.6 or more easily from Corollary 5.7 that the elementary sets form
an algebra.
Definition 8.5 Let E X Y,
Ex = {y Y : (x, y) E},

E y = {x X : (x, y) E}.
These are called the x and y sections.
131

132

PRODUCT MEASURE

Ex
X
x

Theorem 8.6 If E S F, then Ex F and E y S for all x X and y Y .

Proof: Let
M = {E S F such that Ex F,
E y S for all x X and y Y.}
Then M contains all measurable rectangles. If Ei M,

(
i=1 Ei )x = i=1 (Ei )x F.
y

Similarly, (
i=1 Ei ) S. M is thus closed under countable unions. If E M,
EC

= (Ex ) F.

y
Similarly, E C S. Thus M is closed under complementation. Therefore M is a -algebra containing the
elementary sets. Hence, M S F. But M S F. Therefore M = S F and the theorem is proved.
It follows from Lemma 5.6 that the elementary sets form an algebra because clearly the intersection of
two measurable rectangles is a measurable rectangle and
(A B) \ (A0 B0 ) = (A \ A0 ) B (A A0 ) (B \ B0 ),
an elementary set. We use this in the next theorem.
Theorem 8.7 If (X, S, ) and (Y, F, ) are both finite measure spaces ((X), (Y ) < ), then for every
E S F,
a.) Rx (Ex ) is R measurable, y (E y ) is measurable
b.) X (Ex )d = Y (E y )d.
Proof: Let M = {E S F such that both a.) and b.) hold}. Since and are both finite, the
monotone convergence and dominated convergence theorems imply that M is a monotone class. Clearly M
contains the algebra of elementary sets. By the monotone class theorem, M S F.
Theorem 8.8 If (X, S, ) and (Y, F, ) are both -finite measure spaces, then for every E S F,
a.) Rx (Ex ) is R measurable, y (E y ) is measurable.
b.) X (Ex )d = Y (E y )d.

Proof: Let X =
n=1 Xn , Y = n=1 Yn where,

Xn Xn+1 , Yn Yn+1 , (Xn ) < , (Yn ) < .

133
Let
Sn = {A Xn : A S}, Fn = {B Yn : B F}.
Thus (Xn , Sn , ) and (Yn , Fn , ) are both finite measure spaces.
Claim: If E S F, then E (Xn Yn ) Sn Fn .
Proof: Let Mn = {E S F : E (Xn Yn ) Sn Fn }. Clearly Mn contains the algebra of elementary sets. It is also clear that Mn is a monotone class. Thus Mn = S F.
Now let E S F. By Theorem 8.7,
Z
Z
((E (Xn Yn ))x )d =
((E (Xn Yn ))y )d
(8.1)
Xn

Yn

where the integrands are measurable. Now

(E (Xn Yn ))x =
if x
/ Xn and a similar observation holds for the second integrand in (8.1). Therefore,
Z
Z
((E (Xn Yn ))x )d =
((E (Xn Yn ))y )d.
X

Then letting n , we use the monotone convergence theorem to get b.). The measurability assertions of
a.) are valid because the limit of a sequence of measurable functions is measurable.
R
R
Definition 8.9 For E S F and (X, S, ), (Y, F, ) -finite, ( )(E) X (Ex )d = Y (E y )d.
This definition is well defined because of Theorem 8.8. We also have the following theorem.
Theorem 8.10 If A S, B F, then ( )(A B) = (A)(B), and is a measure on S F called
product measure.
The proof of Theorem 8.10 is obvious and is left to the reader. Use the Monotone Convergence theorem.
The next theorem is one of several theorems due to Fubini and Tonelli. These theorems all have to do with
interchanging the order of integration in a multiple integral. The main ideas are illustrated by the next
theorem which is often referred to as Fubinis theorem.
Theorem 8.11 Let f : X Y [0, ] be measurable with respect to S F and suppose and are
-finite. Then
Z
Z Z
Z Z
f d( ) =
f (x, y)dd =
f (x, y)dd
(8.2)
XY

and all integrals make sense.

Proof: For E S F, we note
Z
Z
XE (x, y)d = (Ex ),
XE (x, y)d = (E y ).
Y

Thus from Definition 8.9, (8.2) holds if f = XE . It follows that (8.2) holds for every nonnegative simple function. By Theorem 5.31, there exists an increasing sequence, {fn }, of simple functions converging
pointwise to f . Then
Z
Z
f (x, y)d = lim
fn (x, y)d,
Y

134

PRODUCT MEASURE
Z

f (x, y)d = lim

fn (x, y)d.

This follows from the monotone convergence theorem. Since

Z
x
fn (x, y)d
Y

R
is measurable with respect to S, it follows that
x Y f (x, y)d is also measurable with respect to S. A
R
similar conclusion can be drawn about y X f (x, y)d. Thus the two iterated integrals make sense. Since
(8.2) holds for fn, another application of the Monotone Convergence theorem shows (8.2) holds for f . This
proves the theorem.
R R
R R
Corollary 8.12 Let f : X Y C be S F measurable. Suppose either X Y |f | dd or Y X |f | dd <
. Then f L1 (X Y, ) and
Z
Z Z
Z Z
f d( ) =
f dd =
f dd
(8.3)
XY

with all integrals making sense.

Proof : Suppose first that f is real-valued. Apply Theorem 8.11 to f + and f . (8.3) follows from
observing that f = f + f ; and that all integrals are finite. If f is complex valued, consider real and
imaginary parts. This proves the corollary.
How can we tell if f is S F measurable? The following theorem gives a convenient way for many
examples.
Theorem 8.13 If X and Y are topological spaces having a countable basis of open sets and if S and F both
contain the open sets, then S F contains the Borel sets.
Proof: We need to show S F contains the open sets in X Y . If B is a countable basis for the
topology of X and if C is a countable basis for the topology of Y , then
{B C : B B, C C}
is a countable basis for the topology of X Y . (Remember a basis for the topology of X Y is the collection
of sets of the form U V where U is open in X and V is open in Y .) Thus every open set is a countable
union of sets B C where B B and C C. Since B C is a measurable rectangle, it follows that every
open set in X Y is in S F. This proves the theorem.
The importance of this theorem is that we can use it to assert that a function is product measurable if
it is Borel measurable. For an example of how this can sometimes be done, see Problem 5 in this chapter.
Theorem 8.14 Suppose S and F are Borel, and are regular on S and F respectively, and S F is
Borel. Then is regular on S F. (Recall Theorem 8.13 for a sufficient condition for S F to be
Borel.)
Proof: Let (Xn ) < , (Yn ) < , and Xn X, Yn Y . Let Rn = Xn Yn and define
Gn = {S S F : is regular on S Rn }.
By this we mean that for S Gn
( )(S Rn ) = inf{( )(V ) : V is open and V S Rn }

135
and
( )(S Rn ) =
sup{( )(K) : K is compact and K S Rn }.
If P Q is a measurable rectangle, then
(P Q) Rn = (P Xn ) (Q Yn ).
Let Kx (P Xn ) and Ky (Q Yn ) be such that
(Kx ) + > (P Xn )
and
(Ky ) + > (Q Yn ).
By Theorem 3.30 Kx Ky is compact and from the definition of product measure,
( )(Kx Ky ) = (Kx )(Ky )
(P Xn )(Q Yn ) ((Q Yn ) + (P Xn )) + 2 .
Since is arbitrary, this verifies that ( ) is inner regular on S Rn whenever S is an elementary set.
Similarly, () is outer regular on S Rn whenever S is an elementary set. Thus Gn contains the elementary
sets.
Next we show that Gn is a monotone class. If Sk S and Sk Gn , let Kk be a compact subset of Sk Rn
with
( )(Kk ) + 2k > ( )(Sk Rn ).
Let K =
k=1 Kk . Then
S R n \ K
k=1 (Sk Rn \ Kk ).
Therefore
( )(S Rn \ K)

k=1

( )(Sk Rn \ Kk )
2k = .

k=1

Now let Vk Sk Rn , Vk is open and

( )(Sk Rn ) + > ( )(Vk ).
Let k be large enough that
( )(Sk Rn ) < ( )(S Rn ).
Then ( )(S Rn ) + 2 > ( )(Vk ). This shows Gn is closed with respect to intersections of decreasing
sequences of its elements. The consideration of increasing sequences is similar. By the monotone class
theorem, Gn = S F.

136

PRODUCT MEASURE

Now let S S F and let l < ( )(S). Then l < ( )(S Rn ) for some n. It follows from the
first part of this proof that there exists a compact subset of S Rn , K, such that ( )(K) > l. It follows
that ( ) is inner regular on S F. To verify that the product measure is outer regular on S F, let Vn
be an open set such that
Vn S Rn , ( )(Vn \ (S Rn )) < 2n .
Let V =
n=1 Vn . Then V S and
V \ S
n=1 Vn \ (S Rn ).
Thus,
( )(V \ S)

2n =

n=1

and so
( )(V ) + ( )(S).
This proves the theorem.

8.1

Measures on infinite products

It is important in some applications to consider measures on infinite, even uncountably many products.
In order to accomplish this, we first give a simple and fundamental theorem of Caratheodory called the
Caratheodory extension theorem.
Definition 8.15 Let E be an algebra of sets of and let 0 be a finite measure on E. By this we mean
that 0 is finitely additive and if Ei , E are sets of E with the Ei disjoint and
E =
i=1 Ei ,
then
0 (E) =

(Ei )

i=1

while 0 () < .
In this definition, 0 is trying to be a measure and acts like one whenever possible. Under these conditions,
we can show that 0 can be extended uniquely to a complete measure, , defined on a algebra of sets
containing E and such that agrees with 0 on E. We will prove the following important theorem which is
the main result in this section.
Theorem 8.16 Let 0 be a measure on an algebra of sets, E, which satisfies 0 () < . Then there exists
a complete measure space (, S, ) such that
(E) = 0 (E)
for all E E. Also if is any such measure which agrees with 0 on E, then = on (E) , the algebra
generated by E.

8.1. MEASURES ON INFINITE PRODUCTS

137

Proof: We define an outer measure as follows.

(
)
X

(S) inf
0 (Ei ) : S i=1 Ei , Ei E
i=1

Claim 1: is an outer measure.

Proof of Claim 1: Let S

i=1 Si and let Si j=1 Eij , where

(Si ) +

X


(Eij ) .
2i
j=1

Then
(S)

XX
i

(Eij ) =

X

(Si ) +

 X
=
(Si ) + .
2i
i

Since  is arbitrary, this shows is an outer measure as claimed.

By the Caratheodory procedure, there exists a unique algebra, S, consisting of the measurable sets
such that
(, S, )
is a complete measure space. It remains to show extends 0 .
Claim 2: If S is the algebra of measurable sets, S E and = 0 on E.
Proof of Claim 2: First we observe that if A E, then (A) 0 (A) by definition. Letting
(A) +  >

0 (Ei ) ,
i=1 Ei A,

0 (Ei A) 0 (A)

i=1

it follows
(A) +  >

i=1

since A =
i=1 Ei A. Therefore, = 0 on E.
Next we need show E S. Let A E and let S be any set. There exist sets {Ei } E such that

i=1 Ei S but
(S) +  >

(Ei ) .

i=1

Then
(S) (S A) + (S \ A)

(
i=1 Ei \ A) + (i=1 (Ei A))

(Ei \A) +

i=1

Since  is arbitrary, this shows A S.

X
i=1

(Ei A) =

X
i=1

(Ei ) < (S) + .

138

PRODUCT MEASURE

With these two claims, we have established the existence part of the theorem. To verify uniqueness, Let
M {E (E) : (E) = (E)} .
Then M is given to contain E and is obviously a monotone class. Therefore by the theorem on monotone
classes, M = (E) and this proves the lemma.
With this result we are ready to consider the Kolmogorov theorem about measures on infinite product
spaces. One can consider product measure for infinitely many factors. The Caratheodory extension theorem
above implies an important theorem due to Kolmogorov which we will use to give an interesting application
of the individual ergodic theorem. The situation involves a probability space, (, S, P ), an index set, I
possibly infinite, even uncountably infinite, and measurable functions, {Xt }tI , Xt : R. These
measurable functions are called random variables in this context. It is convenient to consider the topological
space
Y
I
[, ]
[, ]
tI

with the product topology where a subbasis for the topology of [, ] consists of sets of the form [, b)
and (a, ] where a, b R. Thus [, ] is a compact set with respect to this topology and by Tychonoffs
I
theorem, so is [, ] . Q
Let J I. Then if E tI Et , we define
J E

Ft

tI

where
Ft =

Et if t J
[, ] if t
/J

Thus J E leaves alone Et for t J and changes the other Et into [, ] . Also we define for J a finite
subset of I,
Y
J x
xt
tJ

so J is a continuous mapping from [, ] to [, ] .

J E

Et .

tJ

Note that for J a finite subset of I,

Y

[, ] = [, ]

tJ

is a compact metric space with respect to the metric,

d (x, y) = max {dt (xt , yt ) : t J} ,
where
dt (x, y) |arctan (x) arctan (y)| , for x, y [, ] .

8.1. MEASURES ON INFINITE PRODUCTS

139

We leave this assertion as an exercise for the reader. You can show that this is a metric and that the metric
just described delivers the usual product topology which will prove the assertion. Now we define for J a
finite subset of I,
(
)
Y
J
RJ
E=
Et : J E = E, Et a Borel set in [, ]
tI

{RJ : J I, and J finite}

I

Thus R consists of those sets of [, ] for which every slot is filled with [, ] except for a finite set,
J I where the slots are filled with a Borel set, Et . We define E as finite disjoint unions of sets of R. In
fact E is an algebra of sets.
Lemma 8.17 The sets, E defined above form an algebra of sets of
I

[, ] .
I

Proof: Clearly and [, ] are both in E. Suppose A, B R. Then for some finite set, J,
J A = A, J B = B.
Then
J (A \ B) = A \ B E, J (A B) = A B R.
By Lemma 5.6 this shows E is an algebra.
Let XJ = (Xt1 , , Xtm ) where {t1 , , tm } = J. We may define a Radon probability measure, XJ on
J
J
a algebra of sets of [, ] as follows. For E a Borel set in [, ] ,
bX (E) P ({ : XJ () E}).

J
bX is
(Remember the random variables have values in R so they do not take the value .) Now since
J
J
a probability measure, it is certainly finite on the compact sets of [, ] . Also note that if B denotes


J
J
the Borel sets of [, ] , then B E (, ) : E B is a algebra of sets of (, ) which
J
contains the open sets. Therefore, B contains the Borel sets of (, ) and so we can apply Theorem 6.23
bX which will be denoted as X .
to conclude there is a unique Radon measure extending
J
For E R, with J E = E we define
J (E) XJ ( J E)
Theorem 8.18 (Kolmogorov) There exists a complete probability measure space,


I
[, ] , S,
such that if E R and J E = E,
(E) = J (E) .
I

The measure is unique on (E) , the smallest algebra containing E. If A [, ] is a set having
measure 1, then
J ( J A) = 1
for every finite J I.

140

PRODUCT MEASURE

Proof: We first describe a measure, 0 on E and then extend this measure using the Caratheodory
extension theorem. If E R is such that J (E) = E, then 0 is defined as
0 (E) J (E) .
Note that if J J1 and J (E) = E and J1 (E) = E, then J (E) = J1 (E) because

Y
1
J E
[, ] = X1
X1
J ( J E) .
J1 ( J1 E) = XJ1
J1 \J

Therefore 0 is finitely additive on E and is well defined. We need to verify that 0 is actually a measure on
E. Let An where An E and Jn (An ) = An . We need to show that
0 (An ) 0.
Suppose to the contrary that
0 (An )  > 0.
By regularity of the Radon measure, XJn , there exists a compact set,
K n Jn A n
such that
XJn (( Jn An ) \ Kn ) <


.
2n+1

(8.4)

Claim: We can assume Hn 1

Jn Kn is in E in addition to (8.4).
Proof of the claim: Since An E, we know An is a finite disjoint union of sets of R. Therefore, it
suffices to verify the claim under the assumption that An R. Suppose then that
Y
Jn (An ) =
Et
tJn

where the Et are Borel sets. Let Kt be defined by

Kt xt : for some y =

ys , xt y Kn

s6=t,sJn

Then using the compactness of Kn , it is easy to see that Kt is a closed subset of [, ] and is therefore,
compact. Now let
Y
K0n
Kt .
tJn

0
It follows K0n Kn is a compact subset of tJn Et = Jn An and 1
Jn Kn R E. It follows we could
0
have picked Kn . This proves the claim.
I
Let Hn 1
Jn (Kn ) . By the Tychonoff theorem, Hn is compact in [, ] because it is a closed subset
I
of [, ] , a compact space. (Recall that J is continuous.) Also Hn is a set of E. Thus,

0 (An \ Hn ) = XJn (( Jn An ) \ Kn ) <


.
2n+1

8.2. A STRONG ERGODIC THEOREM

141

It follows {Hn } has the finite intersection property for if m

k=1 Hk = , then
 0 (Am \ m
k=1 Hk )

m
X

0 (Ak \ Hk ) <

k=1

k=1



= ,
2k+1
2

a contradiction. Now since these sets have the finite intersection property, it follows

k=1 Ak k=1 Hk 6= ,

a contradiction.
Now we show this implies 0 is a measure. If {Ei } are disjoint sets of E with
E =
i=1 Ei , E, Ei E,
n

then E \ i=1 Ei decreases to the empty set and so

0 (E \

n
i=1 Ei )

= 0 (E)

n
X

0 (Ei ) 0.

i=1

P
Thus 0 (E) = k=1 0 (Ek ) . Now an application of the Caratheodory extension theorem proves the main
part of the theorem. It remains to verify the last assertion.


1
To verify this assertion, let A be a set of measure 1. We note that 1
J ( J A) A, J J ( J A) =


1
1
J ( J A) , and J J ( J A) = J A. Therefore,





1
1 = (A) 1
= J ( J A) 1.
J ( J A) = J J J ( J A)
This proves the theorem.

8.2

A strong ergodic theorem

Here we give an application of the individual ergodic theorem, Theorem 5.51. Let Z denote the integers and
let {Xn }nZ be a sequence of real valued random variables defined on a probability space, (, F, P ). Let
Z
Z
T : [, ] [, ] be defined by
Y
T (x)n = xn+1 , where x =
{xn }
nZ

Thus T slides the sequence to the left one slot. By the Kolmogorov theorem there exists a unique measure,
Z
defined on (E) the smallest algebra of sets of [, ] , which contains E, the algebra of sets described in
the proof of Kolmogorovs theorem. We give conditions next which imply that the mapping, T, just defined
satisfies the conditions of Theorem 5.51.
Definition 8.19 We say the sequence of random variables, {Xn }nZ is stationary if
(Xn ,,Xn+p ) = (X0 ,,Xp )
for every n Z and p 0.
Theorem 8.20 Let {Xn } be stationary and let T be the shift operator just described. Also let be the
Z
measure of the Kolmogorov existence theorem defined on (E), the smallest algebra of sets of [, ]
containing E. Then T is one to one and
T 1 A, T A (E) whenever A (E)


(T A) = T 1 A = (A) for all A (E)
and T is ergodic.

142

PRODUCT MEASURE

Proof: It is obvious that T is one 

to one. We need to show T and T 1 map (E) to (E). It is clear
both these maps take E to E. Let F E (E) : T 1 (E) (E) then F is a algebra and so it equals
(E) . If E R with J E = E, and J = {n, n + 1, , n + p} , then
J E =

n+p
Y

Ej , Ek = [, ] , k
/ J.

j=n

Then
TE = F =

Fn

nZ

where Fk1 = Ek . Therefore,

(E) = (Xn ,,Xn+p ) (En En+p ) ,
and
(T E) = (Xn1 ,,Xn+p1 ) (En En+p ) .
By the assumption these random variables are stationary, (T E) = (E) . A similar assertion holds for T 1 .
It follows by a monotone class argument that this holds for all E (E) . It is routine to verify T is ergodic.
This proves the theorem.
Next we give an interesting lemma which we use in what follows.
Lemma 8.21 In the context of Theorem 8.20 suppose A (E) and (A) = 1. Then there exists F
such that P () = 1 and
)
(

Y
Xk () A
= :
k=

Proof: Let Jn {n, , n} and let n { : XJn () Jn (A)} . Here

XJn () (Xn () , , Xn ()) .
Then from the definition of Jn , we see that
1
A =
n=1 Jn ( Jn A)

and the sets, 1

Jn ( Jn A) are decreasing in n. Let n n . Then if ,
n
Y

Xk () Jn A

k=n

for all n and so for each n, we have

Xk () 1
Jn ( Jn A)

k=

and consequently,

k=

Xk () n 1
Jn ( Jn A) = A

8.2. A STRONG ERGODIC THEOREM

143

showing that

Xk () A

k=

Now suppose

k=

Xk () A. We need to verify that . We know

n
Y

Xk () Jn A

k=n

for each n and so n for each n. Therefore,

n=1 n . This proves the lemma.
The following theorem is the main result.
Theorem 8.22 Let {Xi }iZ be stationary and let J = {0, , p}. Then if f J is a function in L1 () , it
follows that
k
1X
f (Xn+i1 () , , Xn+p+i1 ()) = m
k k
i=1

lim

in L1 (P ) where
m

f (X0 , , Xp ) dP

and for a.e. .

Proof: Let f L1 () where f is (E) measurable. Then by Theorem 5.51 it follows that
n


1
1X
Sn f
f T k1 ()
n
n

f (x) d m

k=1

pointwise a.e. and in L1 () .

Now suppose f is of the form, f (x) = f ( J (x)) where J is a finite subset of Z,
J = {n, n + 1, , n + p} .
Thus,
m=

f ( J x) d(Xn ,,Xn+p ) =

f (Xn , , Xn+p ) dP.

(To verify the equality between the two integrals, verify for simple functions and then take limits.) Now
1
Sk f (x)
k

k


1X
f T i1 (x)
k i=1

k



1X
f J T i1 (x)
k i=1

k
1X
f (xn+i1 , , xn+p+i1 ) .
k i=1

144

PRODUCT MEASURE

Also, by the assumption that the sequence of random variables is stationary,


Z 
1

 Sk f (Xn () , , Xn+p ()) m dP =

k


Z  X
k

1

f (Xn+i1 () , , Xn+p+i1 ()) m dP =


k
i=1

RJ



k
1 X



f (xn+i1 , , xn+p+i1 ) m d(Xn ,,Xn+p ) =

k

i=1

Z 
Z
1



=  Sk f ( J ()) m d =
k



1

 Sk f () m d
k


and this last expression converges to 0 as k from the above.

By the individual ergodic theorem, we know that
1
Sk f (x) m
k

pointwise a.e. with respect to , say for x Q

A where (A) = 1. Now by Lemma 8.21, there exists a set of

measure 1 in F, , such that for all , k= Xk () A. Therefore, for such ,

k
1
1X
Sk f ( J (X ())) =
f (Xn+i1 () , , Xn+p+i1 ())
k
k i=1

converges to m. This proves the theorem.

An important example of a situation in which the random variables are stationary is the case when they
are identically distributed and independent.

Definition 8.23 We say the random variables, {Xj }j= are independent if whenever J = {i1 , , im } is
 m
a finite subset of I, and Eij j=1 are Borel sets in [, ] ,

XJ

m
Y

j=1

E ij =

n
Y

j=1

Xij Eij .

We say the random variables, {Xj }j= are identically distributed if whenever i, j Z and E [, ]
is a Borel set,
Xi (E) = Xj (E) .
As a routine lemma we obtain the following.

Lemma 8.24 Suppose {Xj }j= are independent and identically distributed. Then {Xj }j= are stationary.
The following corollary is called the strong law of large numbers.

8.3. EXERCISES

145

Corollary 8.25 Suppose {Xj }j= are independent and identically distributed random variables which are
in L1 (, P ). Then
Z
k
1X
Xi1 () = m
X0 () dP
lim
k k
R
i=1

(8.5)

a.e. and in L1 (P ) .
Proof: Let
f (x) = f ( 0 (x)) = x0
so f (x) = 0 (x) .
Z

|f (x)| d =

|x| dX0 =

|X0 ()| dP < .

Therefore, from the above strong ergodic theorem, we obtain (8.5).

8.3

Exercises

1. Let A be an algebra of sets in P(Z) and suppose and are two finite measures on (A), the
-algebra generated by A. Show that if = on A, then = on (A).
2. Extend Problem 1 to the case where , are finite with
Z =
n=1 Zn , Zn A
and (Zn ) < .
R
RA
3. Show limA 0 sinx x dx = 2 . Hint: Use x1 = 0 ext dt and Fubinis theorem. This limit is sometimes called the Cauchy principle value. Note that the function sin (x) /x is not in L1 so we are not
finding a Lebesgue integral.
4. Suppose g : Rn R has the property that g is continuous in each variable. Can we conclude that g is
continuous? Hint: Consider
 xy
x2 +y 2 if (x, y) 6= (0, 0) ,
g (x, y)
0 if (x, y) = (0, 0) .
5. Suppose g : Rn R is continuous in every variable. Show that g is the pointwise limit of some
sequence of continuous functions. Conclude that if g is continuous in each variable, then g is Borel
measurable. Give an example of a Borel measurable function on Rn which is not continuous in each
variable. Hint: In the case of n = 2 let
ai

i
,i Z
n

and for (x, y) [ai1 , ai ) R, we let

gn (x, y)

ai x
x ai1
g (ai1 , y) +
g (ai , y).
ai ai1
ai ai1

Show gn converges to g and is continuous. Now use induction to verify the general result.

146

PRODUCT MEASURE

6. Show (R2 , m m, S S) where S is the set of Lebesgue measurable sets is not a complete measure
space. Show there exists A S S and E A such that (m m)(A) = 0, but E
/ S S.
7. Recall that for
E S F, ( )(E) =

(Ex )d =

(E y )d.

Why is a measure on S F?
Rx
Rx
8. Suppose G (x) = G (a) + a g (t) dt where g L1 and suppose F (x) = F (a) + a f (t) dt where f L1 .
Show the usual formula for integration by parts holds,
Z

f Gdx = F G|ba

F gdx.

Hint: You might try replacing G (x) with G (a) +

using Fubinis theorem.

Rx
a

g (t) dt in the first integral on the left and then

9. Let f : [0, ) be measurable where (, S, ) is a finite measure space. Let : [0, ) [0, )
satisfy: is increasing. Show
Z
Z
(f (x))d =
0 (t)(x : f (x) > t)dt.
X

The function t (x : f (x) > t) is called the distribution function. Hint:

Z
Z Z
(f (x))d =
X[0,f (x)) 0 (t)dtdx.
X

Now try to use Fubinis theorem. Be sure to check that everything is appropriately measurable. In
doing so, you may want to first consider f (x) a nonnegative simple function. Is it necessary to assume
(, S, ) is finite?
10. In the Kolmogorov extension theorem, could Xt be a random vector with values in an arbitrary locally
compact Hausdorff space?
11. Can you generalize the strong ergodic theorem to the case where the random variables have values not
in R but Rk for some k?

Fourier Series
9.1

Definition and basic properties

A Fourier series is an expression of the form

ck eikx

k=

where we understand this symbol to mean

lim

n
X

ck eikx .

k=n

Obviously such a sequence of partial sums may or may not converge at a particular value of x.
These series have been important in applied math since the time of Fourier who was an officer in
Napoleans army. He was interested in studying the flow of heat in cannons and invented the concept
to aid him in his study. Since that time, Fourier series and the mathematical problems related to their convergence have motivated the development of modern methods in analysis. As recently as the mid 1960s a
problem related to convergence of Fourier series was solved for the first time and the solution of this problem
was a big surprise. In this chapter, we will discuss the classical theory of convergence of Fourier series. We
will use standard notation for the integral also. Thus,
Z

f (x) dx

X[a,b] (x) f (x) dm.

Definition 9.1 We say a function, f defined on R is a periodic function of period T if f (x + T ) = f (x)

for all x.
Fourier series are useful for representing periodic functions and no other kind of function. To see this,
note that the partial sums are periodic of period 2. Therefore, it is not reasonable to expect to represent a
function which is not periodic of period 2 on R by a series of the above form. However, if we are interested
in periodic functions, there is no loss of generality in studying only functions which are periodic of period
2. Indeed, if
f is a function which has period T, we can study this function in terms of the function,
g (x) f T2x and g is periodic of period 2.
Definition 9.2 For f L1 (, ) and f periodic on R, we define the Fourier series of f as

ck eikx ,

k=

147

(9.1)

148

FOURIER SERIES

where
ck

1
2

f (y) eiky dy.

(9.2)

We also define the nth partial sum of the Fourier series of f by

n
X

Sn (f ) (x)

ck eikx .

(9.3)

k=n

It may be interesting to see where this formula came from. Suppose then that

f (x) =

ck eikx

k=

and we multiply both sides by eimx and take the integral,

f (x) eimx dx =

, so that

ck eikx eimx dx.

k=

Now we switch the sum and the integral on the right side even though we have absolutely no reason to
believe this makes any sense. Then we get
Z
Z

X
f (x) eimx dx =
ck
eikx eimx dx

k=
Z

= cm

1dx = 2ck

R
because eikx eimx dx = 0 if k 6= m. It is formal manipulations of the sort just presented which suggest
that Definition 9.2 might be interesting.
In case f is real valued, we see that ck = ck and so
Z
n
X


1
Sn f (x) =
f (y) dy +
2 Re ck eikx .
2
k=1

Letting ck k + i k
1
Sn f (x) =
2

f (y) dy +

n
X

2 [k cos kx k sin kx]

k=1

where
ck =

1
2

f (y) eiky dy =

1
2

f (y) (cos ky i sin ky) dy

which shows that

k =

1
2

f (y) cos (ky) dy, k =

1
2

f (y) sin (ky) dy

Therefore, letting ak = 2k and bk = 2 k , we see that

Z
Z
1
1
ak =
f (y) cos (ky) dy, bk =
f (y) sin (ky) dy

9.1. DEFINITION AND BASIC PROPERTIES

149

and
n

Sn f (x) =

a0 X
+
ak cos kx + bk sin kx
2

(9.4)

k=1

This is often the way Fourier series are presented in elementary courses where it is only real functions which
are to be approximated. We will stick with Definition 9.2 because it is easier to use.
The partial sums of a Fourier series can be written in a particularly simple form which we present next.
Sn f (x)

n
X

ck eikx

k=n
n 
X

k=n

1
2


f (y) eiky dy eikx

n
1 X  ik(xy) 
e
f (y) dy
2
k=n

Dn (x y) f (y) dy.

(9.5)

The function,
Dn (t)

n
1 X ikt
e
2
k=n

is called the Dirichlet Kernel

Theorem 9.3 The function, Dn satisfies the following:
R
1. Dn (t) dt = 1
2. Dn is periodic of period 2
1
1 sin(n+ 2 )t
.
sin( 2t )

3. Dn (t) = (2)

R iky
1
Proof: Part 1 is obvious because 2
e
dy = 0 whenever k 6= 0 and it equals 1 if k = 0. Part 2 is

also obvious because t eikt is periodic of period 2. It remains to verify Part 3.

2Dn (t) =

n
X

eikt , 2eit Dn (t) =

k=n

n
X

n+1
X

ei(k+1)t =

k=n

k=n+1

and so subtracting we get

2Dn (t) 1 eit = eint ei(n+1)t .
Therefore,


1
1
2Dn (t) eit/2 eit/2 = ei(n+ 2 )t ei(n+ 2 )t
and so


t
2Dn (t) 2i sin
2

1
= 2i sin n +
2

eikt

150

FOURIER SERIES

which establishes Part 3. This proves the theorem.

It is not reasonable to expect a Fourier series to converge to the function at every point. To see this,
change the value of the function at a single point in (, ) and extend to keep the modified function
periodic. Then the Fourier series of the modified function is the same as the Fourier series of the original
function and so if pointwise convergence did take place, it no longer does. However, it is possible to prove
an interesting theorem about pointwise convergence of Fourier series. This is done next.

9.2

Pointwise convergence of Fourier series

The following lemma will make possible a very easy proof of the very important Riemann Lebesgue lemma,
the big result which makes possible the study of pointwise convergence of Fourier series.
Lemma 9.4 Let f L1 (a, b) where (a, b) is some interval, possibly an infinite interval like (, ) and
let > 0. Then there exists an interval of finite length, [a1 , b1 ] (a, b) and a function, g Cc (a1 , b1 ) such
that also g 0 Cc (a1 , b1 ) which has the property that
Z b
|f g| dx <
(9.6)
a

Proof: Without loss of generality we may assume that f (x) 0 for all x since we can always consider the
positive and negative parts of the real and imaginary parts of f. Letting a < an < bn < b with limn an = a
and limn bn = b, we may use the dominated convergence theorem to conclude
Z b


f (x) f (x) X[a ,b ] (x) dx = 0.
lim
n n
n

Therefore, there exist c > a and d < b such that if h = f X(c,d) , then
Z b

|f h| dx < .
4
a

(9.7)

Now from Theorem 5.31 on the pointwise convergence of nonnegative simple functions to nonnegative measurable functions and the monotone convergence theorem, there exists a simple function,
s (x) =

p
X

ci XEi (x)

i=1

with the property that

Z

|s h| dx <

(9.8)

and we may assume each Ei (c, d) . Now by regularity of Lebesgue measure, there are compact sets, Ki
and open sets, Vi such that Ki Ei Vi , Vi (c, d) , and m (Vi \ Ki ) < , where is a positive number
which is arbitrary. Now we let Ki ki Vi . Thus ki is a continuous function whose support is contained in
Vi , which is nonnegative, and equal to one on Ki . Then let
g1 (x)

p
X

ci ki (x) .

i=1

We see that g1 is a continuous function whose support is contained in the compact set, [c, d] (a, b) . Also,
Z b
p
p
X
X
|g1 s| dx
ci m (Vi \ Ki ) <
ci .
a

i=1

i=1

9.2. POINTWISE CONVERGENCE OF FOURIER SERIES

151

Choosing small enough, we may assume

b

|g1 s| dx <

.
4

(9.9)

Now choose r small enough that [c r, d + r] (a, b) and for 0 < h < r, let
gh (x)

1
2h

x+h

g1 (t) dt.

xh

Then gh is a continuous function whose derivative is also continuous and for which both gh and gh0 have
support in [c r, d + r] (a, b) . Now let [a1 , b1 ] [c r, d + r] .
Z

Z x+h
1
|g1 (x) g1 (t)| dtdx
a1 2h xh

(b1 a1 ) =
4 (b1 a1 )
4

|g1 gh | dx

<

b1

(9.10)

whenever h is small enough due to the uniform continuity of g1 . Let g = gh for such h. Using (9.7) - (9.10),
we obtain
Z b
Z b
Z b
Z b
|f g| dx
|f h| dx +
|h s| dx +
|s g1 | dx
+

a
b

|g1 g| dx

+ + + = .
4 4 4 4

<

This proves the lemma.

With this lemma, we are ready to prove the Riemann Lebesgue lemma.
Lemma 9.5 (Riemann Lebesgue) Let f L1 (a, b) where (a, b) is some interval, possibly an infinite interval
like (, ). Then
lim

f (t) sin (t + ) dt = 0.

(9.11)

Proof: Let > 0 be given and use Lemma 9.4 to obtain g such that g and g 0 are both continuous, the
support of both g and g 0 are contained in [a1 , b1 ] , and
Z
a

|g f | dx <

.
2

Then
Z

 b



f (t) sin (t + ) dt

 a


Z

Z b
 b




f (t) sin (t + ) dt
g (t) sin (t + ) dt
 a

a
Z

 b



+
g (t) sin (t + ) dt
 a

Z

Z b
 b




|f g| dx + 
g (t) sin (t + ) dt


a
a
Z


 b1

<
+
g (t) sin (t + ) dt .

2  a1

(9.12)

152

FOURIER SERIES

We integrate the last term by parts obtaining

Z

b1

a1

cos (t + )
g (t) sin (t + ) dt =
g (t) |ba11 +

b1

a1

cos (t + ) 0
g (t) dt,

an expression which converges to zero since g 0 is bounded. Therefore, taking large enough, we see
Z

 b



f (t) sin (t + ) dt < + =

 a
 2 2
and this proves the lemma.

9.2.1

Dinis criterion

Now we are ready to prove a theorem about the convergence of the Fourier series to the midpoint of the
jump of the function. The condition given for convergence in the following theorem is due to Dini. [3].
Theorem 9.6 Let f be a periodic function of period 2 which is in L1 (, ) . Suppose at some x, we have
f (x+) and f (x) both exist and that the function
y

f (x y) f (x) + f (x + y) f (x+)
y

(9.13)

is in L1 (0, ) for some > 0. Then

lim Sn f (x) =

f (x+) + f (x)
.
2

(9.14)

Proof:
Sn f (x) =

Dn (x y) f (y) dy

We change variables x y y and use the periodicity of f and Dn to write this as

Z
Sn f (x) =
Dn (y) f (x y)

Dn (y) f (x y) dy +

Dn (y) f (x y) dy

Dn (y) [f (x y) + f (x + y)] dy





1 sin n + 12 y f (x y) + f (x + y)


dy.
2
sin y2
0
0

Also
f (x+) + f (x)

Dn (y) [f (x+) + f (x)] dy

=
=

2
Z

Dn (y) [f (x+) + f (x)] dy

1 sin n + 12 y


[f (x+) + f (x)] dy

sin y2

(9.15)

9.2. POINTWISE CONVERGENCE OF FOURIER SERIES

153

and so




Sn f (x) f (x+) + f (x)  =


2

Now the function

Z

 1 sin n + 1
y
 f (x y) f (x) + f (x + y) f (x+)  


2


dy  .

 0

2
sin y2

(9.16)

f (x y) f (x) + f (x + y) f (x+)


2 sin y2

(9.17)

is in L1 (0, ) . To see this, note the numerator is in L1 because f
is. Therefore, this function is
in L1 (, )
where is given in the Hypotheses of this theorem because sin y2 is bounded below by sin 2 for such y.
The function is in L1 (0, ) because
f (x y) f (x) + f (x + y) f (x+)
y
f (x y) f (x) + f (x + y) f (x+)


=
y
y
2 sin 2
2 sin

y
2

,

and y/2 sin y2 is bounded on [0, ] . Thus the function in (9.17) is in L1 (0, ) as claimed. It follows from
the Riemann Lebesgue lemma, that (9.16) converges to zero as n . This proves the theorem.
The following corollary is obtained immediately from the above proof with minor modifications.
Corollary 9.7 Let f be a periodic function of period 2 which is in L1 (, ) . Suppose at some x, we have
the function
y

f (x y) + f (x + y) 2s
y

(9.18)

is in L1 (0, ) for some > 0. Then

lim Sn f (x) = s.

(9.19)

As pointed out by Apostol, [3], this is a very remarkable result because even though the Fourier coeficients
depend on the values of the function on all of [, ] , the convergence properties depend in this theorem on
very local behavior of the function. Dinis condition is rather like a very weak smoothness condition on f at
the point, x. Indeed, in elementary treatments of Fourier series, the assumption given is that the one sided
derivatives of the function exist. The following corollary gives an easy to check condition for the Fourier
series to converge to the mid point of the jump.
Corollary 9.8 Let f be a periodic function of period 2 which is in L1 (, ) . Suppose at some x, we have
f (x+) and f (x) both exist and there exist positive constants, K and such that whenever 0 < y <
|f (x y) f (x)| Ky , |f (x + y) f (x+)| < Ky

(9.20)

where (0, 1]. Then

lim Sn f (x) =

f (x+) + f (x)
.
2

Proof: The condition (9.20) clearly implies Dinis condition, (9.13).

(9.21)

154

9.2.2

FOURIER SERIES

Jordans criterion

We give a different condition under which the Fourier series converges to the mid point of the jump. In order
to prove the theorem, we need to give some introductory lemmas which are interesting for their own sake.
Lemma 9.9 Let G be an increasing function defined on [a, b] . Thus G (x) G (y) whenever x < y. Then
G (x) = G (x+) = G (x) for every x except for a countable set of exceptions.
Proof: We let S {x [a, b] : G (x+) > G (x)} . Then there is a rational number in each interval,
(G (x) , G (x+)) and also, since G is increasing, these intervals are disjoint. It follows that there are only
contably many such intervals. Therefore, S is countable and if x
/ S, we have G (x+) = G (x) showing
that G is continuous on S C and the claimed equality holds.
The next lemma is called the second mean value theorem for integrals.
Lemma 9.10 Let G be an increasing function defined on [a, b] and let f be a continuous function defined
on [a, b] . Then there exists t0 [a, b] such that
!
Z t0

Z b
Z b
G (s) f (s) ds = G (a)
f (s) ds + G (b)
f (s) ds .
(9.22)
a

t0

Proof: Letting h > 0 we define

1
Gh (t) 2
h

th

G (r) drds

sh

where we understand G (x) = G (a) for all x < a. Thus Gh (a) = G (a) for all h. Also, from the fundamental theorem of calculus, we see that G0h (t) 0 and is a continuous function of t. Also it is clear that
Rt
limh Gh (t) = G (t) for all t [a, b] . Letting F (t) a f (s) ds,
Z b
Z b
b
Gh (s) f (s) ds = F (t) Gh (t) |a
F (t) G0h (t) dt.
(9.23)
a

Now letting m = min {F (t) : t [a, b]} and M = max {F (t) : t [a, b]} , since G0h (t) 0, we have
Z b
Z b
Z b
0
0
m
Gh (t) dt
F (t) Gh (t) dt M
G0h (t) dt.
a

Therefore, if

Rb
a

G0h (t) dt 6= 0,
m

Rb
a

F (t) G0h (t) dt

M
Rb 0
Gh (t) dt
a

and so by the intermediate value theorem from calculus,

Rb
F (t) G0h (t) dt
F (th ) = a R b
G0h (t) dt
a
Rb
for some th [a, b] . Therefore, substituting for a F (t) G0h (t) dt in (9.23) we have
"
#
Z
Z
b

Gh (s) f (s) ds

G0h (t) dt

= F (t) Gh (t) |ba F (th )

= F (b) Gh (b) F (th ) Gh (b) + F (th ) Gh (a)

!
Z th

Z b
=
f (s) ds Gh (b) +
f (s) ds G (a) .
th

9.2. POINTWISE CONVERGENCE OF FOURIER SERIES

155

Now selecting a subsequence, still denoted by h which converges to zero, we can assume th t0 [a, b].
Therefore, using the dominated convergence theorem, we may obtain the following from the above lemma.
Z

G (s) f (s) ds

G (s) f (s) ds
!
Z

f (s) ds G (b) +

Z

t0

t0

f (s) ds G (a) .

and this proves the lemma.

Definition 9.11 Let f : [a, b] C be a function. We say f is of bounded variation if
)
( n
X
sup
|f (ti ) f (ti1 )| : a = t0 < < tn = b V (f, [a, b]) <
i=1

where the sums are taken over all possible lists, {a = t0 < < tn = b} . The symbol, V (f, [a, b]) is known
as the total variation on [a, b] .
Lemma 9.12 A real valued function, f, defined on an interval, [a, b] is of bounded variation if and only if
there are increasing functions, H and G defined on [a, b] such that f (t) = H (t) G (t) . A complex valued
function is of bounded variation if and only if the real and imaginary parts are of bounded variation.
Proof: For f a real valued function of bounded variation, we may define an increasing function, H (t)
V (f, [a, t]) and then note that
G(t)

z
}|
{
f (t) = H (t) [H (t) f (t)].
It is routine to verify that G (t) is increasing. Conversely, if f (t) = H (t) G (t) where H and G are
increasing, we can easily see the total variation for H is just H (b) H (a) and the total variation for G is
G (b) G (a) . Therefore, the total variation for f is bounded by the sum of these.
The last claim follows from the observation that
|f (ti ) f (ti1 )| max (|Re f (ti ) Re f (ti1 )| , |Im f (ti ) Im f (ti1 )|)
and
|Re f (ti ) Re f (ti1 )| + |Im f (ti ) Im f (ti1 )| |f (ti ) f (ti1 )| .
With this lemma, we can now prove the Jordan criterion for pointwise convergence of the Fourier series.
Theorem 9.13 Suppose f is 2 periodic and is in L1 (, ) . Suppose also that for some > 0, f is of
bounded variation on [x , x + ] . Then
lim Sn f (x) =

f (x+) + f (x)
.
2

(9.24)

Proof: First note that from Definition 9.11, limyx Re f (y) exists because Re f is the difference of two
increasing functions. Similarly this limit will exist for Im f by the same reasoning and limits of the form
limyx+ will also exist. Therefore, the expression on the right in (9.24) exists. If we can verify (9.24) for
real functions which are of bounded variation on [x , x + ] , we can apply this to the real and imaginary

156

FOURIER SERIES

parts of f and obtain the desired result for f. Therefore, we assume without loss of generality that f is real
valued and of bounded variation on [x , x + ] .


Z
f (x+) + f (x)
f (x+) + f (x)
Sn f (x)
=
dy
Dn (y) f (x y)
2
2

Dn (y) [(f (x + y) f (x+)) + (f (x y) f (x))] dy.

Now the Dirichlet kernel, Dn (y) is a constant multiple of sin ((n + 1/2) y) / sin (y/2) and so we can apply
the Riemann Lebesgue lemma to conclude that
Z
lim
Dn (y) [(f (x + y) f (x+)) + (f (x y) f (x))] dy = 0
n

and so it suffices to show that

Z
lim
Dn (y) [(f (x + y) f (x+)) + (f (x y) f (x))] dy = 0.
n

(9.25)

Now y (f (x + y) f (x+)) + (f (x y) f (x)) = h (y) is of bounded variation for y [0, ] and

limy0+ h (y) = 0. Therefore, we can write h (y) = H (y) G (y) where H and G are increasing and for
F = G, H, limy0+ F (y) = F (0) = 0. It suffices to show (9.25) holds with f replaced with either of G or H.
Letting > 0 be given, we choose 1 < such that H ( 1 ) , G ( 1 ) < . Now
Z
Z
Z 1
Dn (y) G (y) dy =
Dn (y) G (y) dy +
Dn (y) G (y) dy
0

and we see from the Riemann Lebesgue lemma that the first integral on the right converges to 0 for any
choice of 1 (0, ) . Therefore, we estimate the second integral on the right. Using the second mean value
theorem, Lemma 9.10, we see there exists n [0, 1 ] such that
Z



Z 1
 1







Dn (y) G (y) dy  = G ( 1 )
Dn (y) dy 

 0



n
Z

 1




Dn (y) dy  .
 n


Now


Z

Z


 1

 1
sin n + 12 y 
y



D (y) = C 
dt

 n n

 n sin (y/2)

y

and for small 1 , y/ sin (y/2) is approximately equal to 2. Therefore, the expression on the right will be
bounded if we can show that
Z

 1 sin n + 1
y 


2
dt

 n

y

is bounded independent of choice of n [0, 1 ] . Changing variables, we see this is equivalent to showing
that
Z

 b sin y 


dy 

 a y


9.2. POINTWISE CONVERGENCE OF FOURIER SERIES

157

is bounded independent of the choice of a, b. But this follows from the convergence of the Cauchy principle
value integral given by
Z A
sin y
lim
dy
A 0
y
which was considered in Problem 3 of Chapter 8 or Problem 12 of Chapter 7. Using the above argument for
H as well as G, this shows that there exists a constant, C independent of such that
Z





lim sup 
Dn (y) [(f (x + y) f (x+)) + (f (x y) f (x))] dy  C.

n  0
Since was arbitrary, this proves the theorem.
It is known that neither the Jordan criterion nor the Dini criterion implies the other.

9.2.3

The Fourier cosine series

Suppose now that f is a real valued function which is defined on the interval [0, ] . Then we can define f on
the interval, [, ] according to the rule, f (x) = f (x) . Thus the resulting function, still denoted by f is
an even function. We can now extend this even function to the whole real line by requiring f (x + 2) = f (x)
obtaining a 2 periodic function. Note that if f is continuous, then this periodic function defined on the
whole line is also continuous. What is the Fourier series of the extended function f ? Since f is an even
function, the nth coefficient is of the form
Z
Z
1
1
inx
f (x) e
dx =
f (x) cos (nx) dx if n 6= 0
cn
2
0
Z
1
c0 =
f (x) dx if n = 0.
0
Thus cn = cn and we see the Fourier series of f is of the form
Z
 X

 Z
1
2
f (x) dx +
f (y) cos ky cos kx

0
0

(9.26)

k=1

= c0 +

2ck cos kx.

(9.27)

k=1

Definition 9.14 If f is a function defined on [0, ] then (9.26) is called the Fourier cosine series of f.
Observe that Fourier series of even 2 periodic functions yield Fourier cosine series.
We have the following corollary to Theorem 9.6 and Theorem 9.13.
Corollary 9.15 Let f be an even function defined on R which has period 2 and is in L1 (0, ). Then at
every point, x, where f (x+) and f (x) both exist and the function
y

f (x y) f (x) + f (x + y) f (x+)
y

(9.28)

is in L1 (0, ) for some > 0, or for which f is of bounded variation near x we have
lim a0 +

n
X

ak cos kx =

k=1

f (x+) + f (x)
2

(9.29)

Here
1
a0 =

Z
0

2
f (x) dx, an =

Z
0

f (x) cos (nx) dx.

(9.30)

158

FOURIER SERIES

There is another way of approximating periodic piecewise continuous functions as linear combinations of
the functions eiky which is clearly superior in terms of pointwise and uniform convergence. This other way
does not depend on any hint of smoothness of f near the point in question.

9.3

The Cesaro means

In this section we define the notion of the Cesaro mean and show these converge to the midpoint of the jump
under very general conditions.
Definition 9.16 We define the nth Cesaro mean of a periodic function which is in L1 (, ), n f (x) by
the formula
n

n f (x)

1 X
Sk f (x) .
n+1
k=0

Thus the nth Cesaro mean is just the average of the first n + 1 partial sums of the Fourier series.
Just as in the case of the Sn f, we can write the Cesaro means in terms of convolution of the function
with a suitable kernel, known as the Fejer kernel. We want to find a formula for the Fejer kernel and obtain
some of its properties. First we give a simple formula which follows from elementary trigonometry.
n
X

k=0



1
sin k +
y
2

n
1 X
(cos ky cos (k + 1) y)
2 sin y2

1 cos ((n + 1) y)
.
2 sin y2

k=0

(9.31)

Lemma 9.17 There exists a unique function, Fn (y) with the following properties.
1. n f (x) =

Fn (x y) f (y) dy,

2. Fn is periodic of period 2,
3. Fn (y) 0 and if > |y| r > 0, then limn Fn (y) = 0,
R
4. Fn (y) dy = 1,
5. Fn (y) =

1cos((n+1)y)
4(n+1) sin2 ( y2 )

Proof: From the definition of n , it follows that

#
Z "
n
1 X
n f (x) =
Dk (x y) f (y) dy.
n + 1
k=0

Therefore,
n

Fn (y) =

1 X
Dk (y) .
n+1

(9.32)

k=0

That Fn is periodic of period 2 follows from this formula and the fact, established earlier that Dk is periodic
of period 2. Thus we have established parts 1 and 2. Part 4 also follows immediately from the fact that

9.3. THE CESARO MEANS

R

159

Dk (y) dy = 1. We now establish Part 5 and Part 3. From (9.32) and (9.31),
Fn (y)


 
n
X
1
1
1


sin
k
+
y
2 (n + 1) sin y2
2
k=0


1
1 cos ((n + 1) y)


2 sin y2
2 (n + 1) sin y2

=
=

1 cos ((n + 1) y)

.
4 (n + 1) sin2 y2

(9.33)

This verifies Part 5 and also shows that Fn (y) 0, the first part of Part 3. If |y| > r,
|Fn (y)|

2
4 (n + 1) sin2

r
2

(9.34)

and so the second part of Part 3 holds. This proves the lemma.
The following theorem is called Fejers theorem
Theorem 9.18 Let f be a periodic function with period 2 which is in L1 (, ) . Then if f (x+) and
f (x) both exist,
lim n f (x) =

f (x+) + f (x)
.
2

(9.35)

If f is everywhere continuous, then n f converges uniformly to f on all of R.

Proof: As before, we may use the periodicity of f and Fn to write
Z
n f (x) =
Fn (y) [f (x y) + f (x + y)] dy


Z0
f (x y) + f (x + y)
=
2Fn (y)
dy.
2
0
R
From the formula for Fn , we see that Fn is even and so 0 2Fn (y) dy = 1. Also


Z
f (x) + f (x+)
f (x) + f (x+)
=
2Fn (y)
dy.
2
2
0
Therefore,




 n f (x) f (x) + f (x+)  =


2
Z




 

f (x) + f (x+) f (x y) + f (x + y)
2Fn (y)

dy 
2
2


Z
0

2Fn (y) dy +

2Fn (y) Cdy

where r is chosen small enough that



 f (x) + f (x+) f (x y) + f (x + y) 
<




2
2

(9.36)

160

FOURIER SERIES

for all 0 < y r. Now using the estimate (9.34) we obtain



Z


1
 n f (x) f (x) + f (x+)  + C


2

(n
+
1)
sin2
r
e
C
+
.
n+1

r
2

dy

and so, letting n , we obtain the desired result.

In case that f is everywhere continuous, then since it is periodic, it must also be uniformly continuous.
It follows that f (x) = f (x) and by the uniform continuity of f, we may choose r small enough that (9.36)
holds for all x whenever y < r. Therefore, we obtain uniform convergence as claimed.

9.4

Gibbs phenomenon

The Fourier series converges to the mid point of the jump in the function under various conditions including
those given above. However, in doing so the convergence cannot be uniform due to the discontinuity of the
function to which it converges. In this section we show there is a small bump in the partial sums of the
Fourier series on either side of the jump which does not disappear as the number of terms in the Fourier series
increases. The small bump just gets narrower. To illustrate this phenomenon, known as Gibbs phenomenon,
we consider a function, f, which equals 1 for x < 0 and 1 for x > 0. Thus the nth partial sum of the Fourier
series is
Sn f (x) =

n
4 X sin ((2k 1) x)
.

2k 1
k=1

We consider the value of this at the point

2n .

This equals

n
n

4 X sin (2k 1) 2n
2 X sin (2k 1) 2n

=

2k 1

n
(2k 1) 2n
k=1
k=1
R
which is seen to be a Riemann sum for the integral 2 0 siny y dy. This integral is a positive constant approximately equal to 1. 179. Therefore, although the value of the function equals 1 for all x > 0, we see that for
large n, the value of the nth partial sum of the Fourier series at points near x = 0 equals approximately
1.179. To illustrate this phenomenon we graph the Fourier series of this function for large n, say n = 10.
P10
The following is the graph of the function, S10 f (x) = 4 k=1 sin((2k1)x)
2k1
You see the little blip near the jump which does not disappear. So you will see this happening for even
P20
larger n, we graph this for n = 20. The following is the graph of S20 f (x) = 4 k=1 sin((2k1)x)
2k1
As you can observe, it looks the same except the wriggles are a little closer together. Nevertheless, it
still has a bump near the discontinuity.

9.5

The mean square convergence of Fourier series

We showed that in terms of pointwise convergence, Fourier series are inferior to the Cesaro means. However,
there is a type of convergence that Fourier series do better than any other sequence of linear combinations
of the functions, eikx . This convergence is often called mean square convergence. We describe this next.
Definition 9.19 We say f L2 (, ) if f is Lebesgue measurable and
Z
2
|f (x)| dx < .

9.5. THE MEAN SQUARE CONVERGENCE OF FOURIER SERIES

161

We say a sequence of functions, {fn } converges to a function, f in the mean square sense if
Z
2
lim
|fn f | dx = 0.
n

Lemma 9.20 If f L2 (, ) , then f L1 (, ) .

2

Proof: We use the inequality ab a2 + b2 whenever a, b 0, which follows from the inequality
2
(a b) 0.
Z
Z
Z
2
|f (x)|
1
|f (x)| dx
dx +
dx < .
2

2
This proves the lemma.
From this lemma, we see we can at least discuss the Fourier series of a function in L2 (, ) . The
following theorem is the main result which shows the superiority of the Fourier series in terms of mean
square convergence.
Theorem 9.21 For ck complex numbers, the choice of ck which minimizes the expression
2
Z 
n

X

ikx 
ck e  dx
f (x)



(9.37)

k=n

is for ck to equal the Fourier coefficient, k where

Z
1
k =
f (x) eikx dx.
2

(9.38)

Also we have Bessels inequality,

1
2

|f | dx

n
X

|k | =

k=n

1
2

|Sn f | dx

(9.39)

where k denotes the kth Fourier coefficient,

k =

1
2

f (x) eikx dx.

(9.40)

Proof: It is routine to obtain that the expression in (9.37) equals

Z
Z
Z
n
n
n
X
X
X
2
2
ikx
ikx
|f | dx
ck
f (x) e dx
ck
f (x) e
dx + 2
|ck |

|f | dx 2

where k =

1
2

k=n
n
X

k=n

ck k 2

n
X

k=n

k=n

ck k + 2

n
X

k=n

|ck |

k=n

f (x) eikx dx, the kth Fourier coefficient. Now

2

ck k ck k + |ck | = |ck k | |k |
Therefore,


2
"
#
Z
n
n
n


X
X
X
1


2
2
2
ck eikx  dx = 2
|f | dx
|k | +
|k ck | 0.
f (x)

2


k=n

k=n

k=n

162

FOURIER SERIES

It is clear from this formula that the minimum occurs when k = ck and that Bessels inequality holds. It
only remains to verify the equal sign in (9.39).
!
!
Z
Z
n
n
X
X
1
1
2
ikx
ilx
|Sn f | dx =
k e
l e
dx
2
2
k=n
l=n
Z X
n
n
X
1
2
2
=
|k | dx =
|k | .
2
k=n

k=n

This proves the theorem.

This theorem has shown that if we measure the distance between two functions in the mean square sense,
d (f, g) =

Z

1/2
2
|f g| dx
,

then the partial sums of the Fourier series do a better job approximating the given function than any other
linear combination of the functions eikx for n k n. We show now that
Z
2
lim
|f (x) Sn f (x)| dx = 0
n

whenever f L2 (, ) . To begin with we need the following lemma.

Lemma 9.22 Let > 0 and let f L2 (, ) . Then there exists g Cc (, ) such that
Z
2
|f g| dx < .

Proof: We can use the dominated convergence theorem to conclude

Z


f f X(+r,r) 2 dx = 0.
lim
r0

Therefore, picking r small enough, we may define k f X(+r,r) and have

Z

|f k| dx <

.
9

(9.41)

Now let k = h+ h + i (l+ l ) where the functions, h and l are nonnegative. We may then use Theorem
5.31 on the pointwise convergence of nonnegative simple functions to nonnegative measurable functions and
the dominated convergence theorem to obtain a nonnegative simple function, s+ h+ such that
Z
 +

h s+ 2 dx < .
144

Similarly, we may obtain simple functions, s , t+ , and t such that

Z
Z


+

h s 2 dx < ,
l t+ 2 dx < ,
144
144



l t 2 dx < .
144

9.5. THE MEAN SQUARE CONVERGENCE OF FOURIER SERIES

163

Letting s s+ s + i (t+ t ) , and using the inequality,

2
(a + b + c + d) 4 a2 + b2 + c2 + d2 ,
we see that

|k s| dx








 
h+ s+ 2 + h s 2 + l+ t+ 2 + l t 2 dx
4


+
+
+
= .
144 144 144 144
9

(9.42)

Pn
Let s (x) = i=1 ci XEi (x) where the Ei are disjoint and Ei ( + r, r) . Let > 0 be given. Using
the regularity of Lebesgue measure, we can get compact sets, Ki and open sets, Vi such that
Ki Ei Vi ( + r, r)
Pn
and m (Vi \ Ki ) < . Then letting Ki ri Vi and g (x) i=1 ci ri (x)
Z

|s g| dx

n
X

|ci | m (Vi \ Ki )

i=1

n
X
i=1

|ci | <

(9.43)

provided we choose small enough. Thus g Cc (, ) and from (9.41) - (9.42),

Z
Z 

2
2
2
2
|f g| dx 3
|f k| + |k s| + |s g| dx < .

This proves the lemma.

With this lemma, we are ready to prove a theorem about the mean square convergence of Fourier series.
Theorem 9.23 Let f L2 (, ) . Then
lim

|f Sn f | dx = 0.

(9.44)

Proof: From Lemma 9.22 there exists g Cc (, ) such that

Z
2
|f g| dx < .

Extend g to make the extended function 2 periodic. Then from Theorem 9.18, n g converges uniformly to
g on all of R. In particular, this uniform convergence occurs on (, ) . Therefore,
Z
2
lim
| n g g| dx = 0.
n

Also note that n f is a linear combination of the functions eikx for |k| n. Therefore,
Z
Z
2
2
| n g g| dx
|Sn g g| dx

164

FOURIER SERIES

which implies
lim

|Sn g g| dx = 0.

Also from (9.39),

Z

|Sn (g f )| dx

|g f | dx.

Therefore, if n is large enough, this shows

Z
Z
2
|f Sn f | dx 3

|f g| dx +

|g Sn g| dx


|Sn (g f )| dx
2

3 ( + + ) = 9.
This proves the theorem.

9.6

Exercises

1. Let f be a continuous function defined on [, ] . Show there exists a polynomial, p such that
||p f || < where ||g|| sup {|g (x)| : x [, ]} . Extend this result to an arbitrary interval. This is
called the Weierstrass approximation theorem. Hint: First find a linear function, ax + b = y such that
f y has the property that it has the same value at both ends of [, ] . Therefore, you may consider
this as the restriction
to [, ] of a continuous periodic function, F. Now find a trig polynomial,
Pn
(x) a0 + k=1 ak cos kx + bk sin kx such that || F || < 3 . Recall (9.4). Now consider the power
series of the trig functions.
2. Show that neither the Jordan nor the Dini criterion for pointwise convergence implies the other criterion. That is, find an example of a function for which Jordans condition implies pointwise convergence
but not Dinis and then find a function for which Dini works but Jordan does not. Hint: You might
1
try considering something like y = [ln (1/x)] for x > 0 to get something for which Jordan works but
Dini does not. For the other part, try something like x sin (1/x) .
R
3. If f L2 (, ) show using Bessels inequality that limn f (x) einx dx = 0. Can this be used
to give a proof of the Riemann Lebesgue lemma for the case where f L2 ?
4. Let f (x) = x for x (, ) and extend to make the resulting function defined on R and periodic of
period 2. Find the Fourier series of f. Verify the Fourier series converges to the midpoint of the jump
and use this series to find a nice formula for 4 . Hint: For the last part consider x = 2 .
5. Let f (x) = x2 on (, ) and extend to form a 2 periodic function defined on R. Find the Fourier
2
series of f. Now obtain a famous formula for 6 by letting x = .
6. Let f (x) = cos x for x (0, ) and define f (x) cos x for x (, 0) . Now extend this function to
make it 2 periodic. Find the Fourier series of f.

9.6. EXERCISES

165

7. Show that for f L2 (, ) ,

Z

f (x) Sn f (x)dx = 2

n
X

|k |

k=n

where the k are the Fourier coefficients of f. Use this and the theorem about mean square convergence,
Theorem 9.23, to show that
1
2

|f (x)| dx =

|k | lim

k=

n
X

|k |

k=n

8. Suppose f, g L2 (, ) . Show using Problem 7

1
2

f gdx =

k k ,

k=

where k are the Fourier coefficients of f and k are the Fourier coefficients of g.




Pn
Pn
Pn
9. Find a formula for k=1 sin kx. Hint: Let Sn = k=1 sin kx. The sin x2 Sn = k=1 sin kx sin x2 .
Now use a Trig. identity to write the terms of this series as a difference of cosines.
Pq
Pq1
10. Prove the Dirichlet formula which says that k=p ak bk = Aq bq Ap1 bp + k=p Ak (bk bk+1 ) . Here
Pq
Aq k=1 ak .
11. Let {an } be a sequence of positive numbers having the propertyPthat limn nan = 0 and nan

(n + 1) an+1 . Show that if this is so, it follows that the series,

k=1 an sin nx converges uniformly
on R. This is a variation of a very interesting problem found in Apostols book, [3]. Hint: Use
the Dirichlet formula of Problem 10 and consider the Fourier series for the 2 periodic extension of
the function f (x) = x on (0, 2) . Show the partial sums for this Fourier
Pn series are uniformly
bounded for x R. To do this it might be of use to maximize the series k=1 sinkkx using methods
of
Pnelementary calculus. Thus you would find the maximum of this function among the points where
k=1 cos (kx) = 0. This sum can be expressed in a simple closed form using techniques similar to those
in
10. Then, having found the value of x at which the maximum is achieved, plug it in to
PnProblem
sin kx
and observe you have a Riemann sum for a certain finite integral.
k=1
k

12. The problem in Apostols book mentioned in Problem 11 is as follows. Let {ak }k=1 be a decreasing
sequence of nonnegative numbers which satisfies limn nan = 0. Then

ak sin (kx)

k=1

converges uniformly on R. Hint: (Following Jones [18]) First show that for p < q, and x (0, ) ,


X

x
 q


ak sin (kx) ap csc
.

2
k=p

To do this, use summation by parts and establish the formula



q
X
cos p 21 x cos


sin (kx) =
2 sin x2
k=p

q+

1
2

x

166

FOURIER SERIES
Next show that if ak

C
k

and {ak } is decreasing, then



n
X



ak sin (kx) 5C.



k=1

To do this, establish that on (0, ) sin x

and for any integer, k, |sin (kx)| |kx| and then write
This equals 0 if m=n

 n

X



ak sin (kx)




k=1

m
X

k=1
m
X
k=1

}|
{
z
n
 X



ak sin (kx)
ak |sin (kx)| + 


k=m+1

x
C
|kx| + am+1 csc
k
2

Cmx +

C
.
m+1x

(9.45)

Now consider two cases, x 1/n and x > 1/n. In the first case, let m = n and in the second, choose
m such that
n>

1
1
m > 1.
x
x

Finally, establish the desired result by modifying ak making it equal to ap for all k p and then
writing


X

 q



a
sin
(kx)
k


k=p


 p
  q

 X
X

 

ak sin (kx) + 
ak sin (kx) 10e (p)


 

k=1

k=1

where e (p) sup {nan : n p} . This will verify uniform convergence on (0, ) . Now explain why this
yields uniform convergence on all of R.
P
13. Suppose P
f (x) = k=1 ak sin kx and that the convergence is uniform. Is it reasonable to suppose that

f 0 (x) = k=1 ak k cos kx? Explain.

14. Suppose |uk (x)| Kk for all x D where

Kk = lim

k=

n
X

Kk < .

k=n

P
Show that k= uk (x) converges converges uniformly on D in the sense that for all > 0, there
exists N such that whenever n > N,


n
 X

X


uk (x)
uk (x) <



k=

k=n

for all x D. This is called the Weierstrass M test.

9.6. EXERCISES

167

15. Suppose f is a differentiable function of period 2 and suppose that both f and f 0 are in L2 (, )
such that for all x (, ) and y sufficiently small,
f (x + y) f (x) =

x+y

f 0 (t) dt.

Show that the Fourier series of f convergesPuniformly to f. Hint: First show using the Dini criterion

that Sn f (x) f (x) for all x. Next let k= ck eikx be the Fourier series for f. Then from the
1 0
ck where c0k is the Fourier coefficient of f 0 . Now use the
definition of ck , show that for k 6= 0, ck = ik
P
0 2
Bessels inequality to argue that k= |ck | < and use the Cauchy Schwarz inequality to obtain
P
|ck | < . Then using the version of the Weierstrass M test given in Problem 14 obtain uniform
convergence of the Fourier series to f.
16. Suppose f L2 (, ) and that E is a measurable subset of (, ) . Show that
Z
Z
lim
Sn f (x) dx =
f (x) dx.
n

Can you conclude that

Z
E

f (x) dx =

k=

ck

Z
E

eikx dx?

168

FOURIER SERIES

The Frechet derivative

10.1

Norms for finite dimensional vector spaces

This chapter is on the derivative of a function defined on a finite dimensional normed vector space. In this
chapter, X and Y are finite dimensional vector spaces which have a norm. We will say a set, U X is open
if for every p U, there exists > 0 such that
B (p, ) {x : ||x p|| < } U.
Thus, a set is open if every point of the set is an interior point. To begin with we give an important inequality
known as the Cauchy Schwartz inequality.
Theorem 10.1 The following inequality holds for ai and bi C.
 n

X



ai bi 




n
X

i=1

Proof: Let t R and define

h (t)

n
X

|ai |

!1/2

n
X

i=1

|bi |

!1/2

(10.1)

i=1

(ai + tbi ) (ai + tbi ) =

i=1

n
X

|ai | + 2t Re

i=1

n
X

ai bi + t2

i=1

n
X

|bi | .

i=1

Now h (t) 0 for all t R. If all bi equal 0, then the inequality (10.1) clearly holds so assume this does not
happen. Then the graph of y = h (t) is a parabola which opens up and intersects the t axis in at most one
point. Thus there is either one real zero or none. Therefore, from the quadratic formula,
4 Re

n
X

ai bi

!2

i=1

n
X

|ai |

i=1

n
X

|bi |

i=1

which shows


n
 X



ai bi 
Re


i=1

n
X

|ai |

!1/2

i=1

n
X

|bi |

i=1

To get the desired result, let C be such that || = 1 and

n
X
i=1

ai bi =

n
X
i=1



n
X



ai bi = 
ai bi  .



169

i=1

!1/2

(10.2)

170

THE FRECHET DERIVATIVE

Then apply (10.2) replacing ai with ai . Then

 n

!1/2
n
n
X

X
X


2
ai bi  = Re
ai bi
|ai |



i=1
i=1
i=1
!1/2 n
!1/2
n
X
X 2
2
=
|ai |
|bi |
.
i=1

n
X

|bi |

!1/2

i=1

i=1

This proves the theorem.

Recall that a linear space X is a normed linear space if there is a norm defined on X, |||| satisfying
||x|| 0, ||x|| = 0 if and only if x = 0,
||x + y|| ||x|| + ||y|| ,
||cx|| = |c| ||x||
whenever c is a scalar.
Definition 10.2 We say a normed linear space, (X, ||||) is a Banach space if it is complete. Thus, whenever,
{xn } is a Cauchy sequence, there exists x X such that limn ||x xn || = 0.
Let X be a finite dimensional normed linear space with norm ||||where the field of scalars is denoted by
F and is understood to be either R or C. Let {v1 , , vn } be a basis for X. If x X, we will denote by xi
the ith component of x with respect to this basis. Thus
x=

n
X

xi vi .

i=1

Definition 10.3 For x X and {v1 , , vn } a basis, we define a new norm by

|x|

n
X

|xi |

!1/2

i=1

Similarly, for y Y with basis {w1 , , wm }, and yi its components with respect to this basis,
|y|

m
X

|yi |

!1/2

i=1

For A L (X, Y ) , the space of linear mappings from X to Y,

||A|| sup{|Ax| : |x| 1}.

(10.3)

We also say that a set U is an open set if for all x U, there exists r > 0 such that
B (x,r) U
where
B (x,r) {y : |y x| < r}.
Another way to say this is that every point of U is an interior point. The first thing we will show is that
these two norms, |||| and || , are equivalent. This means the conclusion of the following theorem holds.

10.1. NORMS FOR FINITE DIMENSIONAL VECTOR SPACES

171

Theorem 10.4 Let (X, ||||) be a finite dimensional normed linear space and let || be described above relative
to a given basis, {v1 , , vn }. Then || is a norm and there exist constants , > 0 independent of x such
that
||x|| |x| ||x|| .

(10.4)

Proof: All of the above properties of a norm are obvious except the second, the triangle inequality. To
establish this inequality, we use the Cauchy Schwartz inequality to write
2

|x + y|

n
X

n
X

|xi + yi |

i=1

|xi | +

i=1

n
X

|x| + |y| + 2

|xi |

n
X

|yi | + 2 Re

i=1
!1/2

i=1

n
X

|yi |

n
X

xi y i

i=1
!1/2

i=1
2

= |x| + |y| + 2 |x| |y| = (|x| + |y|)

and this proves the second property above.

It remains to show the equivalence of the two norms. By the Cauchy Schwartz inequality again,
 n

!1/2
n
n
X
 X
X


2
xi vi 
|xi | ||vi || |x|
||vi ||
||x|| 



i=1
1

i=1

i=1

|x| .

This proves the first half of the inequality.

Suppose the second half of the inequality is not valid. Then there exists a sequence xk X such that
 k
  
x  > k xk  , k = 1, 2, .

Then define

yk

xk
.
|xk |

It follows
 k
 
 
y  = 1, yk  > k yk  .

(10.5)

Letting yik be the components of yk with respect to the given basis, it follows the vector


y1k , , ynk
is a unit vector in Fn . By the Heine Borel theorem, there exists a subsequence, still denoted by k such that


y1k , , ynk (y1 , , yn ) .
It follows from (10.5) and this that for
y=

n
X

yi vi ,

i=1

 n
 

n
X
 X

 k 







0 = lim y  = lim 
yik vi  = 
yi vi 
 

k
k 
i=1

i=1

but not all the yi equal zero. This contradicts the assumption that {v1 , , vn } is a basis and this proves
the second half of the inequality.

172

THE FRECHET DERIVATIVE

Corollary 10.5 If (X, ||||) is a finite dimensional normed linear space with the field of scalars F = C or R,
then X is complete.
Proof: Let {xk } be a Cauchy sequence. Then letting the components of xk with respect to the given
basis be
xk1 , , xkn ,
it follows from Theorem 10.4, that
xk1 , , xkn

is a Cauchy sequence in Fn and so

xk1 , , xkn (x1 , , xn ) Fn .
Thus,
xk =

n
X

xki vi

i=1

n
X

xi vi X.

i=1

This proves the corollary.

Corollary 10.6 Suppose X is a finite dimensional linear space with the field of scalars either C or R and
|||| and |||||| are two norms on X. Then there exist positive constants, and , independent of x X such
that
|||x||| ||x|| |||x||| .
Thus any two norms are equivalent.
Proof: Let {v1 , , vn } be a basis for X and let || be the norm taken with respect to this basis which
was described earlier. Then by Theorem 10.4, there are positive constants 1 , 1 , 2 , 2 , all independent of
x X such that
2 |||x||| |x| 2 |||x||| ,
1 ||x|| |x| 1 ||x|| .
Then
2 |||x||| |x| 1 ||x||

1
1 2
|x|
|||x|||
1
1

and so
2
2
|||x||| ||x||
|||x|||
1
1
which proves the corollary.
Definition 10.7 Let X and Y be normed linear spaces with norms ||||X and ||||Y respectively. Then
L (X, Y ) denotes the space of linear transformations, called bounded linear transformations, mapping X to
Y which have the property that
||A|| sup {||Ax||Y : ||x||X 1} < .
Then ||A|| is referred to as the operator norm of the bounded linear transformation, A.

10.1. NORMS FOR FINITE DIMENSIONAL VECTOR SPACES

173

We leave it as an easy exercise to verify that |||| is a norm on L (X, Y ) and it is always the case that
||Ax||Y ||A|| ||x||X .
Theorem 10.8 Let X and Y be finite dimensional normed linear spaces of dimension n and m respectively
and denote by |||| the norm on either X or Y . Then if A is any linear function mapping X to Y, then
A L (X, Y ) and (L (X, Y ) , ||||) is a complete normed linear space of dimension nm with
||Ax|| ||A|| ||x|| .
Proof: We need to show the norm defined on linear transformations really is a norm. Again the first
and third properties listed above for norms are obvious. We need to show the second and verify ||A|| < .
Letting {v1 , , vn } be a basis and || defined with respect to this basis as above, there exist constants
, > 0 such that
||x|| |x| ||x|| .
Then,
||A + B|| sup{||(A + B) (x)|| : ||x|| 1}
sup{||Ax|| : ||x|| 1} + sup{||Bx|| : ||x|| 1}
||A|| + ||B|| .
Next we verify that ||A|| < . This follows from

!
n
n

 X
X


||A (x)|| = A
xi vi 
|xi | ||A (vi )||


i=1

|x|

n
X

||A (vi )||

!1/2

i=1

||x||

n
X

||A (vi )||

!1/2

< .

i=1

i=1

P
1/2
n
2
Thus ||A||
.
i=1 ||A (vi )||
Next we verify the assertion about the dimension of L (X, Y ) . Let the two sets of bases be
{v1 , , vn } and {w1 , , wm }
for X and Y respectively. Let wi vk L (X, Y ) be defined by

0 if l 6= k
wi vk vl
wi if l = k
and let L L (X, Y ) . Then
Lvr =

m
X

djr wj

j=1

for some djk . Also

m X
n
X

j=1 k=1

djk wj vk (vr ) =

m
X
j=1

djr wj .

174

THE FRECHET DERIVATIVE

It follows that
L=

m X
n
X

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 (X, Y ) . If
X

dik wi vk = 0,

i,k

then
0=

dik wi vk (vl ) =

m
X

dil wi

i=1

i,k

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 (X, Y ) is
mn as claimed. By Corollary 10.5 (L (X, Y ) , ||||) is complete. If x 6= 0,


  x 
1


 ||A||
= A
||Ax||
||x||  ||x|| 

This proves the theorem.

An interesting application of the notion of equivalent norms on Rn is the process of giving a norm on a
finite Cartesian product of normed linear spaces.
Definition 10.9 Let Xi , i = 1, , n be normed linear spaces with norms, ||||i . For
x (x1 , , xn )

n
Y

Xi

i=1

define :

Qn

i=1

Xi Rn by

(x) (||x1 ||1 , , ||xn ||n )

Qn
Then if |||| is any norm on Rn , we define a norm on i=1 Xi , also denoted by |||| by
||x|| ||x|| .

The following theorem follows immediately from Corollary 10.6.

Qn
Theorem 10.10 Let Xi and ||||i be given in the above definition and Q
consider the norms on i=1 Xi
n
described there in terms of norms on Rn . Then any two of these norms on i=1 Xi obtained in this way are
equivalent.
For example, we may define
||x||1

n
X

|xi | ,

i=1

||x|| max {||xi ||i , i = 1, , n} ,

or
||x||2 =

n
X
i=1

and all three are equivalent norms on

Qn

i=1

Xi .

2
||xi ||i

!1/2

10.2. THE DERIVATIVE

10.2

175

The Derivative

Let U be an open set in X, a normed linear space and let f : U Y be a function.

Definition 10.11 We say a function g is o (v) if
lim

||v||0

g (v)
=0
||v||

(10.6)

We say 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) , the derivative sometimes called the Frechet derivative.
Note that in finite dimensional normed linear spaces, it does not matter which norm we use in this
definition because of Theorem 10.4 and Corollary 10.6. The definition means that the error,
f (x + v) f (x) Lv
converges to 0 faster than ||v|| . The term o (v) is notation that is descriptive of the behavior in (10.6) and
it is only this behavior that concerns us. Thus,
o (v) = o (v) + o (v) , o (tv) = o (v) , ko (v) = o (v)
and other similar observations hold. This notation is both sloppy and useful because it neglects details which
are not important.
Theorem 10.12 The derivative is well defined.
Proof: 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 (t) .
Note that o (tv) = o (t) for fixed v. Therefore, dividing by t yields
(L2 L1 ) (v) =

o (t)
.
t

Now let t 0 to conclude that (L2 L1 ) (v) = 0. This proves the theorem.
Lemma 10.13 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||

176

THE FRECHET DERIVATIVE

Proof:
f (x + v) f (x) = Df (x) v + o (v) .
Let ||v|| be small enough that
||o (v)|| ||v||
Then
||f (x + v) f (x)|| ||Df (x) v|| + ||v||
(||Df (x)|| + 1) ||v||
This proves the lemma with K = ||Df (x)|| + 1.
Theorem 10.14 (The chain rule) Let X, Y, and Z be normed linear spaces, and let U X be an open set
and let V Y also be an open set. Suppose f : U V is differentiable at x and suppose g : V Z is
differentiable at f (x) . Then g f is differentiable at x and
D (g f ) (x) = D (g (f (x))) D (f (x)) .
Proof: This follows from a computation. Let B (x,r) U and let r also be small enough that for
||v|| r,
f (x + v) V. For such v, using the definition of differentiability of g and f ,
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)) .

(10.7)

Now by Lemma 10.13, letting  > 0, it follows that for ||v|| small enough,
||o (f (x + v) f (x))||  ||f (x + v) f (x)|| K ||v|| .
Since  > 0 is arbitrary, this shows o (f (x + v) f (x)) = o (v) . By (10.7), this shows
g (f (x + v)) g (f (x)) = D (g (f (x))) D (f (x)) v + o (v)
which proves the theorem.
We have defined the derivative as a linear transformation. This means that we can consider the matrix
of the linear transformation with respect to various bases on X and Y . In the case where X = Rn and
Y = Rm ,we shall denote the matrix taken with respect to the standard basis vectors ei , the vector with a
1 in the ith slot and zeros elsewhere, by Jf (x) . Thus, if the components of v with respect to the standard
basis vectors are vi ,
X
Jf (x)ij vj = i (Df (x) v)
(10.8)
j

where i is the projection onto the ith component of a vector in Y = Rm . What are the entries of Jf (x)?
Letting
f (x) =

m
X
i=1

fi (x) ei ,

10.2. THE DERIVATIVE

177
fi (x + v) fi (x) = i (Df (x) v) + o (v) .

Thus, letting t be a small scalar,

fi (x+tej ) fi (x) = t i (Df (x) ej ) + o (t) .
Dividing by t, and letting t 0,
fi (x)
= i (Df (x) ej ) .
xj
Thus, from (10.8),
Jf (x)ij =

fi (x)
.
xj

(10.9)

This proves the following theorem

Theorem 10.15 In the case where X = Rn and Y = Rm , if f is differentiable at x then all the partial
derivatives
fi (x)
xj
exist and if Jf (x) is the matrix of the linear transformation with respect to the standard basis vectors, then
the ijth entry is given by (10.9).
What if all the partial derivatives of f exist? Does it follow that f is differentiable? Consider the following
function. f : R2 R.
 xy
x2 +y 2 if (x, y) 6= (0, 0) .
f (x, y) =
0 if (x, y) = (0, 0)
Then from the definition of partial derivatives, this function has both partial derivatives at (0, 0). However
f is not even continuous at (0, 0) which may be seen by considering the behavior of the function along the
line y = x and along the line x = 0. By Lemma 10.13 this implies f is not differentiable.
Lemma 10.16 Suppose X = Rn , f : U R and all the partial derivatives of f exist and are continuous
in U . Then f is differentiable in U .
Proof: Let B (x, r) U and let ||v|| < r. Then,

i1
n
i
X
X
X
f x +
f (x + v) f (x) =
vj ej f x +
vj ej
i=1

j=1

j=1

where
0
X

vj ej 0.

i=1

By the one variable mean value theorem,

f (x + v) f (x) =



Pi1
n f x +
v
e
+

v
e
X
i i i
j=1 j j
i=1

xi

vi

178

THE FRECHET DERIVATIVE

where j [0, 1] . Therefore,

f (x + v) f (x) =

n
X
f (x)
i=1

n
X
i=1

xi



Pi1
f x + j=1 vj ej + i vi ei
xi

vi +

f (x)

vi .
xi

Consider the last term.





Pi1
X

f x + j=1 vj ej + j vj ej
 n
f (x) 


vi 

xi
xi
 i=1


n
X

|vi |

!1/2

i=1


 
2 1/2
Pi1

n  f x +
v
e
+

v
e
X
j
j
j
j
j

j=1
f (x) 



xi
xi

i=1 
and so it follows from continuity of the partial derivatives that this last term is o (v) . Therefore, we define
Lv

n
X
f (x)
i=1

xi

vi

where
v=

n
X

vi ei .

i=1

Then L is a linear transformation which satisfies the conditions needed for it to equal Df (x) and this proves
the lemma.
Theorem 10.17 Suppose X = Rn , Y = Rm and f : U Y and suppose the partial derivatives,
fi
xj
all exist and are continuous in U. Then f is differentiable in U.
Proof: From Lemma 10.16,
fi (x + v) fi (x) = Dfi (x) v+o (v) .
Letting
(Df (x) v)i Dfi (x) v,
we see that
f (x + v) f (x) = Df (x) v+o (v)
and this proves the theorem.
When all the partial derivatives exist and are continuous we say the function is a C 1 function. More
generally, we give the following definition.

10.2. THE DERIVATIVE

179

Definition 10.18 In the case where X and Y are normed linear spaces, and U X is an open set, we say
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 ) .
The following is an important abstract generalization of the concept of partial derivative defined above.
Definition 10.19 Let X and Y be normed linear spaces. Then we can make X Y into a normed linear
space by defining a norm,
||(x, y)|| max (||x||X , ||y||Y ) .
Now let g : U X Y Z, where U is an open set and X , Y, and Z are normed linear spaces, and denote
an element of X Y by (x, y) where x X and y Y. Then the map x g (x, y) is a function from the
open set in X,
{x : (x, y) U }
to Z. When this map is differentiable, we denote its derivative by
D1 g (x, y) , or sometimes by Dx g (x, y) .
Thus,
g (x + v, y) g (x, y) = D1 g (x, y) v + o (v) .
A similar definition holds for the symbol Dy g or D2 g.
The following theorem will be very useful in much of what follows. It is a version of the mean value
theorem.
Theorem 10.20 Suppose X and Y are Banach spaces, 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
S {t [0, 1] : for all s [0, t] ,
||f (x + s (y x)) f (x)|| (M + ) s ||y x||} .
Then 0 S and by continuity of f , it follows that if t sup S, then t S and if t < 1,
||f (x + t (y x)) f (x)|| = (M + ) t ||y x|| .

If t < 1, then there exists a sequence of positive numbers, {hk }k=1 converging to 0 such that
||f (x + (t + hk ) (y x)) f (x)|| > (M + ) (t + hk ) ||y x||

(10.10)

180

THE FRECHET DERIVATIVE

which implies that

||f (x + (t + hk ) (y x)) f (x + t (y x))||
+ ||f (x + t (y x)) f (x)|| > (M + ) (t + hk ) ||y x|| .
By (10.10), this inequality implies
||f (x + (t + hk ) (y x)) f (x + t (y x))|| > (M + ) hk ||y x||
which yields upon dividing by hk and taking the limit as hk 0,
||Df (x + t (y x)) (y x)|| > (M + ) ||y x|| .
Now by the definition of the norm of a linear operator,
M ||y x|| ||Df (x + t (y x))|| ||y x|| > (M + ) ||y x|| ,
a contradiction. Therefore, t = 1 and so
||f (x + (y x)) f (x)|| (M + ) ||y x|| .
Since  > 0 is arbitrary, this proves the theorem.
The next theorem is a version of the theorem presented earlier about continuity of the partial derivatives
implying differentiability, presented in a more general setting. In the proof of this theorem, we will take
||(u, v)|| max (||u||X , ||v||Y )
and always we will use the operator norm for linear maps.
Theorem 10.21 Let g, U, X, Y, and Z be given as in Definition 10.19. Then g is C 1 (U ) if and only if D1 g
and D2 g both exist and are continuous on U. In this case we have the formula,
Dg (x, y) (u, v) = D1 g (x, y) u+D2 g (x, y) v.
Proof: Suppose first that g C 1 (U ) . Then if (x, y) U,
g (x + u, y) g (x, y) = Dg (x, y) (u, 0) + o (u) .
Therefore, D1 g (x, y) u =Dg (x, y) (u, 0) . Then
||(D1 g (x, y) D1 g (x0 , y0 )) (u)|| =
||(Dg (x, y) Dg (x0 , y0 )) (u, 0)||
||Dg (x, y) Dg (x0 , y0 )|| ||(u, 0)|| .
Therefore,
||D1 g (x, y) D1 g (x0 , y0 )|| ||Dg (x, y) Dg (x0 , y0 )|| .
A similar argument applies for D2 g and this proves the continuity of the function, (x, y) Di g (x, y) for
i = 1, 2. The formula follows from
Dg (x, y) (u, v)

= Dg (x, y) (u, 0) + Dg (x, y) (0, v)

D1 g (x, y) u+D2 g (x, y) v.

10.2. THE DERIVATIVE

181

Now suppose D1 g (x, y) and D2 g (x, y) exist and are continuous.

g (x + u, y + v) g (x, y) = g (x + u, y + v) g (x, y + v)
+g (x, y + v) g (x, y)
= g (x + u, y) g (x, y) + g (x, y + v) g (x, y) +
[g (x + u, y + v) g (x + u, y) (g (x, y + v) g (x, y))]

= D1 g (x, y) u + D2 g (x, y) v + o (v) + o (u) +

[g (x + u, y + v) g (x + u, y) (g (x, y + v) g (x, y))] .

(10.11)

Let h (x, u) g (x + u, y + v) g (x + u, y) . Then the expression in [ ] is of the form,

h (x, u) h (x, 0) .
Also
Du h (x, u) = D1 g (x + u, y + v) D1 g (x + u, y)
and so, by continuity of (x, y) D1 g (x, y) ,
||Du h (x, u)|| <
whenever ||(u, v)|| is small enough. By Theorem 10.20, there exists > 0 such that if ||(u, v)|| < , the
norm of the last term in (10.11) satisfies the inequality,
||g (x + u, y + v) g (x + u, y) (g (x, y + v) g (x, y))|| < ||u|| .

(10.12)

Therefore, this term is o ((u, v)) . It follows from (10.12) and (10.11) that
g (x + u, y + v) =

g (x, y) + D1 g (x, y) u + D2 g (x, y) v+o (u) + o (v) + o ((u, v))

= g (x, y) + D1 g (x, y) u + D2 g (x, y) v + o ((u, v))

Showing that Dg (x, y) exists and is given by
Dg (x, y) (u, v) = D1 g (x, y) u + D2 g (x, y) v.
The continuity of (x, y) Dg (x, y) follows from the continuity of (x, y) Di g (x, y) . This proves the
theorem.

182

THE FRECHET DERIVATIVE

10.3

Higher order derivatives

If f : U Y, then
x Df (x)
is a mapping from U to L (X, Y ) , a normed linear space.
Definition 10.22 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. The same pattern applies to taking higher order derivatives. Thus,


D3 f (x) D D2 f (x)
and we can consider D3 f (x) as a tri linear map. Also, instead of writing
D2 f (x) (u) (v) ,
we sometimes write
D2 f (x) (u, v) .
We say f is C k (U ) if f and its first k derivatives are all continuous. For example, for f to be C 2 (U ) ,
x D2 f (x)
would have to be continuous as a map from U to L (X, L (X, Y )) . The following theorem deals with the
question of symmetry of the map D2 f .
This next lemma is a finite dimensional result but a more general result can be proved using the Hahn
Banach theorem which will also be valid for an infinite dimensional setting. We leave this to the interested
reader who has had some exposure to functional analysis. We are primarily interested in finite dimensional
situations here, although most of the theorems and proofs given so far carry over to the infinite dimensional
case with no change.
Lemma 10.23 If z Y, there exists L L (Y, F) such that
2

Lz = |z| , |L| |z| .

2

Here |z|

Pm

i=1

|zi | where z =

Pm

i=1 zi wi ,

for {w1 , , wm }a basis for Y and

|L| sup {|Lx| : |x| 1} ,

the operator norm for L with respect to this norm.

10.3. HIGHER ORDER DERIVATIVES

183

Proof of the lemma: Let

Lx

m
X

xi z i

i=1

where

Pm

i=1 zi

wi = z. Then
L (z)

m
X

zi z i =

i=1

m
X

|zi | = |z| .

i=1

Also
m

X



|Lx| = 
xi z i  |x| |z|


i=1

and so |L| |z| . This proves the lemma.

Actually, the following lemma is valid but its proof involves the Hahn Banach theorem. Infinite dimensional versions of the following theorem will need this version of the lemma.
Lemma 10.24 If z (Y, || ||) a normed linear space, there exists L L (Y, F) such that
2

Lz = ||z|| , ||L|| ||z|| .

Theorem 10.25 Suppose f : U X Y where X and Y are normed linear spaces, D2 f (x) exists for all
x U and D2 f is continuous at x U. Then
D2 f (x) (u) (v) = D2 f (x) (v) (u) .
Proof: Let B (x,r) U and let t, s (0, r/2]. Now let L L (Y, F) and define
(s, t)

Re L
{f (x+tu+sv) f (x+tu) (f (x+sv) f (x))}.
st

(10.13)

Let h (t) = Re L (f (x+sv+tu) f (x+tu)) . Then by the mean value theorem,

(s, t)

=
=

1
1
(h (t) h (0)) = h0 (t) t
st
st
1
(Re LDf (x+sv+tu) u Re LDf (x+tu) u) .
s

Applying the mean value theorem again,

(s, t) = Re LD2 f (x+sv+tu) (v) (u)
where , (0, 1) . If the terms f (x+tu) and f (x+sv) are interchanged in (10.13), (s, t) is also unchanged
and the above argument shows there exist , (0, 1) such that
(s, t) = Re LD2 f (x+sv+tu) (u) (v) .
Letting (s, t) (0, 0) and using the continuity of D2 f at x,
lim

(s, t) = Re LD2 f (x) (u) (v) = Re LD2 f (x) (v) (u) .

(s,t)(0,0)

By Lemma 10.23, there exists L L (Y, F) such that for some norm on Y, || ,


L D2 f (x) (u) (v) D2 f (x) (v) (u) =

184

THE FRECHET DERIVATIVE

 2

D f (x) (u) (v) D2 f (x) (v) (u)2

and


|L| D2 f (x) (u) (v) D2 f (x) (v) (u) .
For this L,


0 = Re L D2 f (x) (u) (v) D2 f (x) (v) (u)


= L D2 f (x) (u) (v) D2 f (x) (v) (u)

2
= D2 f (x) (u) (v) D2 f (x) (v) (u)
and this proves the theorem in the case where the vector spaces are finite dimensional. We leave the general
case to the reader. Use the second version of the above lemma, the one which depends on the Hahn Banach
theorem in the last step of the proof where an auspicious choice is made for L.
Consider the important special case when X = Rn and Y = R. If ei are the standard basis vectors, what
is
D2 f (x) (ei ) (ej )?
To see what this is, use the definition to write
D2 f (x) (ei ) (ej ) = t1 s1 D2 f (x) (tei ) (sej )
= t1 s1 (Df (x+tei ) Df (x) + o (t)) (sej )
= t1 s1 (f (x+tei + sej ) f (x+tei )
+o (s) (f (x+sej ) f (x) + o (s)) + o (t) s) .
First let s 0 to get
t1


f
f
(x+tei )
(x) + o (t)
xj
xj

and then let t 0 to obtain

D2 f (x) (ei ) (ej ) =

2f
(x)
xi xj

(10.14)

Thus the theorem asserts that in this special case the mixed partial derivatives are equal at x if they are
defined near x and continuous at x.

10.4

Implicit function theorem

The following lemma is very useful.

10.4. IMPLICIT FUNCTION THEOREM

185

Lemma 10.26 Let A L (X, X) where X is a Banach space, (complete normed linear space), and suppose
||A|| r < 1. Then
1

(I A)

exists

(10.15)

and



1 
1
(I A)  (1 r) .

Furthermore, if

(10.16)



I A L (X, X) : A1 exists
the map A A1 is continuous on I and I is an open subset of L (X, X) .
Proof: Consider
Bk

k
X

Ai .

i=0

Then if N < l < k,

||Bk Bl ||

k
k
X
 i  X
rN
i
 A  
.
||A||
1r

i=N

i=N

It follows Bk is a Cauchy sequence and so it converges to B L (X, X) . Also,

(I A) Bk = I Ak+1 = Bk (I A)
and so
I = lim (I A) Bk = (I A) B, I = lim Bk (I A) = B (I A) .
k

Thus
1

(I A)

=B=

Ai .

i=0

It follows


 X
 i  X

1 
i
A 
||A|| =
(I A) 
i=1

i=0

1
.
1r

To verify the continuity of the inverse map, let A I. Then

B = A I A1 (A B)



1 1
and so if A1 (A B) < 1 it follows B 1 = I A1 (A B)
A which shows I is open. Now for
such B this close to A,

 1
 


B A1  =  I A1 (A B) 1 A1 A1 




1


=  I A1 (A B)
I A1 

186

THE FRECHET DERIVATIVE




 X
 1
 


k 1  X

1
A (A B)k A1 
= 
A (A B) A 


k=1

k=1


 1
A (A B)
||A1

 1 
A 

1
(A B)||
 1

which shows that if ||A B|| is small, so is B A1  . 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 10.27 Let (Y, ) and (X, d) be complete metric spaces and suppose for each (x, y) X Y,
T (x, y) X and satisfies
d (T (x, y) , T (x0 , y)) rd (x, x0 )

(10.17)

d (T (x, y) , T (x, y 0 )) M (y, y 0 ) .

(10.18)

where 0 < r < 1 and also

Then for each y Y there exists a unique fixed point for T (, y) , x X, satisfying
T (x, y) = x

(10.19)

and also if x (y) is this fixed point,

d (x (y) , x (y 0 ))

M
(y, y 0 ) .
1r

(10.20)

Proof: First we show there exists a fixed point for the mapping, T (, y) . For a fixed y, let g (x) T (x, y) .
Now pick any x0 X and consider the sequence,
x1 = g (x0 ) , xk+1 = g (xk ) .
Then by (10.17),
d (xk+1 , xk ) = d (g (xk ) , g (xk1 )) rd (xk , xk1 )
r2 d (xk1 , xk2 ) rk d (g (x0 ) , x0 ) .
Now by the triangle inequality,
d (xk+p , xk )

p
X

d (xk+i , xk+i1 )

i=1

p
X

rk+i1 d (x0 , g (x0 ))

i=1

rk d (x0 , g (x0 ))
.
1r

Since 0 < r < 1, this shows that {xk }k=1 is a Cauchy sequence. Therefore, it converges to a point in X, x.
To see x is a fixed point,
x = lim xk = lim xk+1 = lim g (xk ) = g (x) .
k

10.4. IMPLICIT FUNCTION THEOREM

187

This proves (10.19). To verify (10.20),

d (x (y) , x (y 0 )) = d (T (x (y) , y) , T (x (y 0 ) , y 0 ))
d (T (x (y) , y) , T (x (y) , y 0 )) + d (T (x (y) , y 0 ) , T (x (y 0 ) , y 0 ))
M (y, y 0 ) + rd (x (y) , x (y 0 )) .
Thus (1 r) d (x (y) , x (y 0 )) M (y, y 0 ) . This also shows the fixed point for a given y is unique. This proves
the theorem.
The implicit function theorem is one of the most important results in Analysis. It provides the theoretical
justification for such procedures as implicit differentiation taught in Calculus courses and has surprising
consequences in many other areas. It deals with the question of solving, f (x, y) = 0 for x in terms of y
and how smooth the solution is. We give a proof of this theorem next. The proof we give will apply with
no change to the case where the linear spaces are infinite dimensional once the necessary changes are made
in the definition of the derivative. In a more general setting one assumes the derivative is what is called
a bounded linear transformation rather than just a linear transformation as in the finite dimensional case.
Basically, this means we assume the operator norm is defined. In the case of finite dimensional spaces, this
boundedness of a linear transformation can be proved. We will use the norm for X Y given by,
||(x, y)|| max {||x|| , ||y||} .
Theorem 10.28 (implicit function theorem) Let X, Y, Z be complete normed linear 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) .

(10.21)

Then there exist positive constants, , , such that for every y B (y0 , ) there exists a unique x (y)
B (x0 , ) such that
f (x (y) , y) = 0.

(10.22)

Futhermore, the mapping, y x (y) is in C 1 (B (y0 , )) .

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) .

(10.23)

by continuity of the derivative and Theorem 10.21, it follows that there exists > 0 such that if ||(x x0 , y y0 )|| <
, then
||D1 T (x, y)|| <

1
,
2




1 
D1 f (x0 , y0 )  ||D2 f (x, y)|| < M

(10.24)




1 
where M > D1 f (x0 , y0 )  ||D2 f (x0 , y0 )|| . By Theorem 10.20, whenever x, x0 B (x0 , ) and y
B (y0 , ) ,
||T (x, y) T (x0 , y)||

1
||x x0 || .
2

(10.25)

188

THE FRECHET DERIVATIVE

Solving (10.23) for D1 f (x, y) ,

D1 f (x, y) = D1 f (x0 , y0 ) (I D1 T (x, y)) .
1

By Lemma 10.26 and (10.24), D1 f (x, y) exists and







1 
1 
D1 f (x, y)  2 D1 f (x0 , y0 )  .

(10.26)

Now we will restrict y some more. Let 0 < < min , 3M

. Then suppose x B (x0 , ) and y
B (y0 , ). Consider 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) ,

and
1

D2 g (x, y) = D1 f (x0 , y0 )

D2 f (x, y) .

Thus by (10.24), (10.21) saying that f (x0 , y0 ) = 0, and Theorems 10.20 and (10.11), it follows that for such
(x, y) ,
||T (x, y) x0 || = ||g (x, y)|| = ||g (x, y) g (x0 , y0 )||

5
||x x0 || + M ||y y0 || < + =
< .
2
2 3
6

(10.27)

Also for such (x, yi ) , i = 1, 2, we can use Theorem 10.20 and (10.24) to obtain




1
||T (x, y1 ) T (x, y2 )|| = D1 f (x0 , y0 ) (f (x, y2 ) f (x, y1 ))
M ||y2 y1 || .

(10.28)

From now on we assume ||x x0 || < and ||y y0 || < so that (10.28), (10.26), (10.27), (10.25), and
(10.24) all hold. By (10.28), (10.25), (10.27), and the uniform contraction principle, Theorem 10.27 applied to


X B x0 , 5
and Y B (y0 , ) implies that for each y B (y0 , ) , there exists a unique x (y) B (x0 , )
6


(actually in B x0 , 5
) such that T (x (y) , y) = x (y) which is equivalent to
6
f (x (y) , y) = 0.
Furthermore,
||x (y) x (y0 )|| 2M ||y y0 || .

(10.29)

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 10.21 and Theorem 10.20 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)) .

10.4. IMPLICIT FUNCTION THEOREM

189

The last term in the above is o (v) because of (10.29). Therefore, by (10.26), we can 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 , ) and

1

Dx (y) = D1 (x (y) , y)

D2 f (x (y) , y) .

Now it follows from the continuity of D2 f , D1 f , the inverse map, (10.29), and this formula for Dx (y)that
x () is C 1 (B (y0 , )). This proves the theorem.
The next theorem is a very important special case of the implicit function theorem known as the inverse
function theorem. Actually one can also obtain the implicit function theorem from the inverse function
theorem.
Theorem 10.29 (Inverse Function Theorem) Let x0 U, an open set in X , and let f : U Y . Suppose
f is C 1 (U ) , and Df (x0 )1 L(Y, X).

(10.30)

Then there exist open sets, W, and V such that

x0 W U,

(10.31)

f : W V is 1 1 and onto,

(10.32)

f 1 is C 1 ,

(10.33)

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 , ) f 1 (B (y0 , ))
and
V B (y0 , ) .
This proves the theorem.
Lemma 10.30 Let
O {A L (X, Y ) : A1 L (Y, X)}

(10.34)

I : O L (Y, X) , IA A1 .

(10.35)

and let

Then O is open and I is in C m (O) for all m = 1, 2, Also

DI (A) (B) = I (A) (B) I (A) .

(10.36)

190

THE FRECHET DERIVATIVE

Proof: Let A O and let B L (X, Y ) with

||B||

1  1 1
A
.
2

Then
 1   1 
A B  A  ||B|| 1
2
and so by Lemma 10.26,
I + A1 B

1

L (X, X) .

Thus
1

(A + B)

(1)

n=0

A1 B

= I + A1 B

n

1

A1 =



A1 = I A1 B + o (B) A1

which shows that O is open and also,

I (A + B) I (A) =

(1)

A1 B

n=0

n

A1 A1

= A1 BA1 + o (B)
= I (A) (B) I (A) + o (B)
which demonstrates (10.36). It follows from this that we can continue taking derivatives of I. For ||B1 ||
small,
[DI (A + B1 ) (B) DI (A) (B)]
= I (A + B1 ) (B) I (A + B1 ) I (A) (B) I (A)
= I (A + B1 ) (B) I (A + B1 ) I (A) (B) I (A + B1 ) +
I (A) (B) I (A + B1 ) I (A) (B) I (A)
= [I (A) (B1 ) I (A) + o (B1 )] (B) I (A + B1 )
+I (A) (B) [I (A) (B1 ) I (A) + o (B1 )]
=



[I (A) (B1 ) I (A) + o (B1 )] (B) A1 A1 B1 A1 +
I (A) (B) [I (A) (B1 ) I (A) + o (B1 )]

= I (A) (B1 ) I (A) (B) I (A) + I (A) (B) I (A) (B1 ) I (A) + o (B1 )
and so
D2 I (A) (B1 ) (B) =
I (A) (B1 ) I (A) (B) I (A) + I (A) (B) I (A) (B1 ) I (A)
which shows I is C 2 (O) . Clearly we can continue in this way, which shows I is in C m (O) for all m = 1, 2,

10.4. IMPLICIT FUNCTION THEOREM

191

Corollary 10.31 In the inverse or implicit function theorems, assume

f C m (U ) , m 1.
Then
f 1 C m (V )
in the case of the inverse function theorem. In the implicit function theorem, the function
y x (y)
is C m .
Proof: We consider the case of the inverse function theorem.



Df 1 (y) = I Df f 1 (y) .
Now by Lemma 10.30, and the chain rule,
D2 f 1 (y) (B) =








I Df f 1 (y) (B) I Df f 1 (y) D2 f f 1 (y) Df 1 (y)






= I Df f 1 (y) (B) I Df f 1 (y)





D2 f f 1 (y) I Df f 1 (y) .
Continuing in this way we see that it is possible to continue taking derivatives up to order m. Similar
reasoning applies in the case of the implicit function theorem. This proves the corollary.
As an application of the implicit function theorem, we consider the method of Lagrange multipliers from
calculus. Recall the problem is to maximize or minimize a function subject to equality constraints. Let
x Rn
and let f : U R be a C 1 function. Also let
gi (x) = 0, i = 1, , m

(10.37)

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.
We say x0 is a local maximum if f (x0 ) f (x) for all x near x0 which also satisfies the constraints (10.37).
A local minimum is defined similarly. Let F : U R Rm+1 be defined by

f (x) a
g1 (x)

(10.38)
F (x,a)
.
..

.
gm (x)

192

THE FRECHET DERIVATIVE

Now consider the m + 1 n Jacobian matrix,

fx1 (x0 )
g1x1 (x0 )

..

fxn (x0 )
g1xn (x0 )
..
.

gmx1 (x0 )

gmxn (x0 )

If this matrix has rank m + 1 then it follows from the implicit function theorem that we can select m + 1
variables, xi1 , , xim+1 such that the system
F (x,a) = 0

(10.39)

specifies these m + 1 variables as a function of the remaining n (m + 1) variables and a in an open set of
Rnm . Thus there is a solution (x,a) to (10.39) 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 requires the
rows to be linearly dependent. Thus, there exist m scalars,
1 , , m ,
and a scalar such that

g1x1 (x0 )
gmx1 (x0 )
fx1 (x0 )

..
..
..

= 1
+ + m
.
.
.
.
fxn (x0 )
g1xn (x0 )
gmxn (x0 )

(10.40)

If the column vectors

g1x1 (x0 )
gmx1 (x0 )

..
..

.
.
g1xn (x0 )
gmxn (x0 )

(10.41)

are linearly independent, then, 6= 0 and dividing by yields an expression of the form

g1x1 (x0 )
gmx1 (x0 )
fx1 (x0 )

..
..
..

= 1
+ + m

.
.
.
fxn (x0 )

g1xn (x0 )

(10.42)

gmxn (x0 )

at every point x0 which is either a local maximum or a local minimum. This proves the following theorem.
Theorem 10.32 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 (10.37), then (10.40) must hold for
some scalars , 1 , , m not all equal to zero. If the vectors in (10.41) are linearly independent, it follows
that an equation of the form (10.42) holds.

10.5

Taylors formula

First we recall the Taylor formula with the Lagrange form of the remainder. Since we will only need this on
a specific interval, we will state it for this interval.
Theorem 10.33 Let h : (, 1 + ) R have m + 1 derivatives. Then there exists t [0, 1] such that
h (1) = h (0) +

m
X
h(k) (0)

k=1

k!

h(m+1) (t)
.
(m + 1)!

10.5. TAYLORS FORMULA

193

Now let f : U R where U X a normed linear 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 we consider
f (x+tv) f (x) h (t)
for t [0, 1] . Then
h0 (t) = Df (x+tv) (v) , h00 (t) = D2 f (x+tv) (v) (v)
and continuing in this way, we see that
h(k) (t) = D(k) f (x+tv) (v) (v) (v) D(k) f (x+tv) vk .
It follows from Taylors formula for a function of one variable that
f (x + v) = f (x) +

m
X
D(k) f (x) vk

k!

k=1

D(m+1) f (x+tv) vm+1

.
(m + 1)!

(10.43)

This proves the following theorem.

Theorem 10.34 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 (10.43) holds.
Now we consider the case where U Rn and f : U R is C 2 (U ) . Then from Taylors theorem, if v is
small enough, there exists t (0, 1) such that
f (x + v) = f (x) + Df (x) v+

D2 f (x+tv) v2
.
2

Letting
v=

n
X

vi ei ,

i=1

where ei are the usual basis vectors, the second derivative term 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
Hij (x+tv) = D2 f (x+tv) (ei ) (ej ) =

2 f (x+tv)
,
xj xi

the Hessian matrix. From Theorem 10.25, this is a symmetric matrix. By the continuity of the second partial
derivative and this,
1
f (x + v) = f (x) + Df (x) v+ vT H (x) v+
2

194

THE FRECHET DERIVATIVE

1 T
v (H (x+tv) H (x)) v .
2
where the last two terms involve ordinary matrix multiplication and

(10.44)

vT = (v1 vn )
for vi the components of v relative to the standard basis.
Theorem 10.35 In the above situation, suppose Df (x) = 0. Then if H (x) has all positive eigenvalues, x
is a local minimum. If 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 (10.44) holds and by continuity of the second derivative, we know
H (x) is a symmetric matrix. Thus H (x) has all real eigenvalues. Suppose first that H (x) has all positive
n
eigenvalues and that all are larger than 2 > 0. Then
PnH (x) has an orthonormal basis of eigenvectors, {vi }i=1
and if u is an arbitrary vector, we can write u = j=1 uj vj where uj = u vj . Thus
T

u H (x) u =

n
X

uj vjT H

j=1

n
X
j=1

u2j j 2

(x)

n
X

uj vj

j=1

n
X

u2j = 2 |u| .

j=1

From (10.44) and the continuity of H, if v is small enough,

1
2
1
2
2
2
|v| .
f (x + v) f (x) + 2 |v| 2 |v| = f (x) +
2
4
4
This shows the first claim of the theorem. The second claim follows from similar reasoning. Suppose H (x)
has a positive eigenvalue 2 . Then let v be an eigenvector for this eigenvalue. Then from (10.44),
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
1
2
= f (x) + t2 2 |v| + t2 vT (H (x+tv) H (x)) v
2
2
1
2
f (x) + t2 2 |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 prove the theorem.
This theorem is an analogue of the second derivative test for higher dimensions. As in one dimension,
when there is a zero eigenvalue, it may be impossible to determine from the Hessian matrix what the local
qualitative behavior of the function is. For example, consider
f (x+tv)

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


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.

10.6. EXERCISES

10.6

195

Exercises

1. Suppose L L (X, Y ) where X and Y are two finite dimensional vector spaces and suppose L is one
to one. Show there exists r > 0 such that for all x X,
|Lx| r |x| .
Hint: Define |x|1 |Lx| , observe that ||1 is a norm and then use the theorem proved earlier that all
norms are equivalent in a finite dimensional normed linear space.
2. Let U be an open subset of X, f : U Y where X, Y are finite dimensional normed linear 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 > 0 such that if
x1 , x2 B (x0 , ) ,
then
|f (x1 ) f (x2 ) Df (x0 ) (x1 x2 )|

r
|x1 x2 |
2

then use Problem 1.

3. Suppose M L (X, Y ) where X and Y are finite dimensional linear spaces 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. Hint: Let {y1 yn } be a basis of Y and let M xi = yi . Then
define
Ly =

n
X

i xi where y =

i=1

n
X

i yi .

i=1

Show {x1 , , xn } is a linearly independent set and show you can obtain {x1 , , xn , , xm }, a basis
for X in which M xj = 0 for j > n. Then let
Px

n
X

i xi

i=1

where
x=

m
X

i xi .

i=1

4. Let f : U Y, f C 1 (U ) , and Df (x1 ) is onto. Show there exists ,  > 0 such that f (B (x1 , ))
B (f (x1 ) , ) . Hint:Let
L L (Y, X) , LDf (x1 ) x =P x,
and let X1 P X where P 2 = P , x1 X1 , and let U1 X1 U. Now apply the inverse function
theorem to f restricted to X1 .
5. Let f : U 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.

196

THE FRECHET DERIVATIVE

6. 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, (x, y)) : (x, y) near (x0 , y0 )},
{( (y, z) y, z) : (y, z) near (y0 , z0 )},
or
{(x, (x, z) , z, ) : (x, z) near (x0 , z0 )}.
7. 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
||f (x1 + t (x2 x1 )) f (x1 ) tL (x2 x1 )||
and let
S {t [0, 1] : ||f (x1 + t (x2 x1 )) f (x1 ) tL (x2 x1 )||
(k + ) t ||x2 x1 ||} .
Now imitate the proof of Theorem 10.20.
8. Let f : U Y, Df (x) exists for all x U, B (x0 , ) U, and there exists L L (X, Y ) , such that
L1 L (Y, X) , and for all x B (x0 , )
||Df (x) L|| <

r
, r < 1.
||L1 ||

Show that there exists  > 0 and an open subset of B (x0 , ) , V, such that f : V B (f (x0 ) , ) is one
to one and onto. Also Df 1 (y) exists for each y B (f (x0 ) , ) and is given by the formula


1
Df 1 (y) = Df f 1 (y)
.
Hint: Let
Ty (x) T (x, y) x L1 (f (x) y)
(1r)
n
for |y f (x0 )| < 2||L
1 || , consider {Ty (x0 )}. This is a version of the inverse function theorem for f
1
only differentiable, not C .

10.6. EXERCISES

197

9. Denote by C ([0, T ] : Rn ) the space of functions which are continuous having values in Rn and define
a norm on this linear space as follows.


||f || max |f (t)| et : t [0, T ] .
Show for each R, this is a norm and that C ([0, T ] ; Rn ) is a complete normed linear space with
this norm.
10. Let f : R Rn Rn be continuous and suppose f satisfies a Lipschitz condition,
|f (t, x) f (t, y)| K |x y|
and let x0 Rn . Show there exists a unique solution to the Cauchy problem,
x0 = f (t, x) , x (0) = x0 ,
for t [0, T ] . Hint: Consider the map
G : C ([0, T ] ; Rn ) C ([0, T ] ; Rn )
defined by
Gx (t) x0 +

f (s, x (s)) ds,

where the integral is defined componentwise. Show G is a contraction map for |||| given in Problem 9
for a suitable choice of and that therefore, it has a unique fixed point in C ([0, T ] ; Rn ) . Next argue,
using the fundamental theorem of calculus, that this fixed point is the unique solution to the Cauchy
problem.
11. Let (X, d) be a complete metric space and let T : X X be a mapping which satisfies
d (T n x, T n y) rd (x, y)
for some r < 1 whenever n is sufficiently large. Show T has a unique fixed point. Can you give another
proof of Problem 10 using this result?

198

THE FRECHET DERIVATIVE

Change of variables for C 1 maps

In this chapter we will give theorems for the change of variables in Lebesgue integrals in which the mapping
relating the two sets of variables is not linear. To begin with we will assume U is a nonempty open set and
h :U V h (U ) is C 1 , one to one, and det Dh (x) 6= 0 for all x U. Note that this implies by the inverse
function theorem that V is also open. Using Theorem 3.32, there exist open sets, U1 and U2 which satisfy
=
6 U1 U1 U2 U2 U,
and U2 is compact. Then




0 < r dist U1 , U2C inf ||x y|| : x U1 and y U2C .
In this section |||| will be defined as
||x|| max {|xi | , i = 1, , n}
T

where
Qn x (x1 , , xn ) . We do this because with this definition, B (x,r) is just an open n dimensional cube,
i=1 (xi r, xi + r) whose center is at x and whose sides are of length 2r. This is not the most usual norm
used and this norm has dreadful geometric properties but it is very convenient here because of the way we
can fill open sets up with little n cubes of this sort. By Corollary 10.6 there are constants and depending
on n such that
|x| ||x|| |x| .
Therefore, letting B0 (x,r) denote the ball taken with respect to the usual norm, || , we obtain




B (x,r) B0 x, 1 r B x, 1 r .

(11.1)

Thus we can use this norm in the definition of differentiability.

Recall that for A L (Rn , Rn ) ,
||A|| sup {||Ax|| : ||x|| 1} ,
is called the operator norm of A taken with respect to the norm ||||. Theorem 10.8 implied |||| is a norm
on L (Rn , Rn ) which satisfies
||Ax|| ||A|| ||x||
and is equivalent to the operator norm of A taken with respect to the usual norm, || because, by this
theorem, L (Rn , Rn ) is a finite dimensional vector space and Corollary 10.6 implies any two norms on this
finite dimensional vector space are equivalent.
We will also write dy or dx to denote the integration with respect to Lebesgue measure. This is done
because the notation is standard and yields formulae which are easy to remember.
199

CHANGE OF VARIABLES FOR C 1 MAPS

200

Lemma 11.1 Let  > 0 be given. There exists r1 (0, r) such that whenever x U1 ,
||h (x + v) h (x) Dh (x) v||
<
||v||
for all ||v|| < r1 .
Proof: The above expression equals
R


 1
 0 (Dh (x+tv) vDh (x) v) dt
||v||

which is no larger than

R1
0

||Dh (x+tv) Dh (x)|| dt ||v||

=
||v||
Z

||Dh (x+tv) Dh (x)|| dt.

(11.2)

Now x Dh (x) is uniformly continuous on the compact set U2 . Therefore, there exists 1 > 0 such that if
||x y|| < 1 , then
||Dh (x) Dh (y)|| < .
Let 0 < r1 < min ( 1 , r) . Then if ||v|| < r1 , the expression in (11.2) is smaller than  whenever x U1 .
Now let Dp consist of all rectangles
n
Y

(ai , bi ] (, )

i=1

where ai = l2p or and bi = (l + 1) 2p or for k, l, p integers with p > 0. The following lemma is
the key result in establishing a change of variable theorem for the above mappings, h.
Lemma 11.2 Let R Dp . Then
Z
Z
Xh(RU1 ) (y) dy XRU1 (x) |det Dh (x)| dx

(11.3)

and also the integral on the left is well defined because the integrand is measurable.
n o
fi
fj Dq where q > p
Proof: The set, U1 is the countable union of disjoint sets, R
of rectangles, R
i=1
is chosen large enough that for r1 described in Lemma 11.1,
2q < r1 , ||det Dh (x)| |det Dh (y)|| < 
if ||x y|| 2q , x, y U1 . Each of these sets,
with R. Denote by

{Ri }i=1

n o
fi
R

i=1

is either a subset of R or has empty intersection

those which are contained in R. Thus

R U1 =
i=1 Ri .

201
The set, h (Ri ) , is a Borel set because for
Ri

n
Y

(ai , bi ],

i=1

h (Ri ) equals
!
n
Y


1
ai + k , bi ,

k=1 h

i=1

a countable union of compact sets. Then h (R U1 ) =

i=1 h (Ri ) and so it is also a Borel set. Therefore,
the integrand in the integral on the left of (11.3) is measurable.
Z

Z X

Xh(RU1 ) (y) dy =

Xh(Ri ) (y) dy =

i=1

Z
X

Xh(Ri ) (y) dy.

i=1

Now by Lemma 11.1 if xi is the center of interior(Ri ) = B (xi , ) , then since all these rectangles are chosen
so small, we have for ||v|| ,


1
h (xi + v) h (xi ) = Dh (xi ) v+Dh (xi ) o (v)
where ||o (v)|| <  ||v|| . Therefore,
 




1 
h (Ri ) h (xi ) + Dh (xi ) B 0, 1 +  Dh (xi ) 


h (xi ) + Dh (xi ) B (0, (1 + C))
where

It follows from Theorem 7.21 that


n
o

1 
C = max Dh (x)  : x U1 .
n

mn (h (Ri )) |det Dh (xi )| mn (Ri ) (1 + C)

and hence
Z

dy

h(RU1 )

i=1

Z
X
i=1

Z
X
i=1

Z
X
i=1

Ri

|det Dh (xi )| mn (Ri ) (1 + C)

|det Dh (xi )| dx (1 + C)

Ri

(|det Dh (x)| + ) dx (1 + C)

Ri

|det Dh (x)| dx (1 + C) + mn (U1 ) (1 + C)

CHANGE OF VARIABLES FOR C 1 MAPS

202
=

Z



n
XRU1 (x) |det Dh (x)| dx + mn (U1 ) (1 + C) .

Since  is arbitrary, this proves the Lemma.

Borel measurable functions have a very nice feature. When composed with a continuous map, the result
is still Borel measurable. This is a general result described in the following lemma which we will use shortly.
Lemma 11.3 Let h be a continuous map from U to Rn and let g : Rn (X, ) be Borel measurable. Then
g h is also Borel measurable.
Proof: Let V . Then
1

(g h)

(V ) = h1 g 1 (V ) = h1 (Borel set) .

The conclusion of the lemma will follow if we show that h1 (Borel set) = Borel set. Let


S E Borel sets : h1 (E) is Borel .
Then S is a algebra which contains the open sets and so it coincides with the Borel sets. This proves the
lemma.
Let Eq consist of all finite disjoint unions of sets of {Dp : p q}. By Lemma 5.6 and Corollary 5.7, Eq
is an algebra. Let E
q=1 Eq . Then E is also an algebra.
Lemma 11.4 Let E (E) . Then
Z
Z
Xh(EU1 ) (y) dy XEU1 (x) |det Dh (x)| dx

(11.4)

and the integrand on the left of the inequality is measurable.

Proof: Let
M {E (E) : (11.4) holds} .
Then from the monotone and dominated convergence theorems, M is a monotone class. By Lemma 11.2,
M contains E, it follows M equals (E) by the theorem on monotone classes. This proves the lemma.
Note that by Lemma 7.3 (E) contains the open sets and so it contains the Borel sets also. Now let
F be any Borel set and let V1 = h (U1 ). By the inverse function theorem,

V1 is open. Thus for F a Borel
set, F V1 is also a Borel set. Therefore, since F V1 = h h1 (F ) U1 , we can apply (11.4) in the first
inequality below and write
Z
Z
Z
XF (y) dy XF V1 (y) dy Xh1 (F )U1 (x) |det Dh (x)| dx
V1

Xh1 (F ) (x) |det Dh (x)| dx =

U1

XF (h (x)) |det Dh (x)| dx

(11.5)

U1

It follows that (11.5) holds for XF replaced with an arbitrary nonnegative Borel measurable simple function.
Now if g is a nonnegative Borel measurable function, we may obtain g as the pointwise limit of an increasing
sequence of nonnegative simple functions. Therefore, the monotone convergence theorem applies and we
may write the following inequality for all g a nonnegative Borel measurable function.
Z
Z
g (y) dy
g (h (x)) |det Dh (x)| dx.
(11.6)
V1

U1

With this preparation, we are ready for the main result in this section, the change of variables formula.

11.1. GENERALIZATIONS

203

Theorem 11.5 Let U be an open set and let h be one to one, C 1 , and det Dh (x) 6= 0 on U. For V = h (U )
and g 0 and Borel measurable,
Z
Z
g (y) dy =
g (h (x)) |det Dh (x)| dx.
(11.7)
V

Proof: By the inverse function theorem, h1 is also a C 1 function. Therefore, using (11.6) on x = h1 (y) ,
along with the chain rule and the property of determinants which states that the determinant of the product
equals the product of the determinants,
Z
Z
g (y) dy
g (h (x)) |det Dh (x)| dx
V1

Z
V1

Z
V1

U1




 
g (y) det Dh h1 (y)  det Dh1 (y) dy





g (y) det D h h1 (y) dy =

g (y) dy

V1

which shows the formula (11.7) holds for U1 replacing U. To verify the theorem, let Uk be an increasing
sequence of open sets whose union is U and whose closures are compact as in Theorem 3.32. Then from the
above, (11.7) holds for U replaced with Uk and V replaced with Vk . Now (11.7) follows from the monotone
convergence theorem. This proves the theorem.

11.1

Generalizations

In this section we give some generalizations of the theorem of the last section. The first generalization will
be to remove the assumption that det Dh (x) 6= 0. This is accomplished through the use of the following
fundamental lemma known as Sards lemma. Actually the following lemma is a special case of Sards lemma.
Lemma 11.6 (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.

Proof: Let {Uk }k=1 be the increasing sequence of open sets whose closures are compact and whose union
equals U which exists by Theorem 3.32 and let Zk Z Uk . We will show that h (Zk ) has measure zero.
Let W be an open set contained in Uk+1 which contains Zk and satisfies
mn (Zk ) +  > mn (W )
where we will always assume  < 1. Let
C
r = dist Uk , Uk+1

and let r1 > 0 be the constant of Lemma 11.1 such that whenever x Uk and 0 < ||v|| r1 ,
||h (x + v) h (x) Dh (x) v|| <  ||v|| .
Now let
f
W =
i=1 Ri

(11.8)

CHANGE OF VARIABLES FOR C 1 MAPS

204

n o
fi are disjoint half open cubes from Dq where q is chosen so large that for each i,
where the R
 
n
o
fi sup ||x y|| : x, y R
fi < r1 .
diam R
Denote by {Ri } those cubes which have nonempty intersection with Zk , let di be the diameter of Ri , and let
zi be a point in Ri Zk . Since zi Zk , it follows Dh (zi ) B (0,di ) = Di where Di is contained in a subspace
of Rn having dimension p n 1 and the diameter of Di is no larger than Ck di where
Ck max {||Dh (x)|| : x Zk }
Then by (11.8), if z Ri ,
h (z) h (zi ) Di + B (0, di )


Di + B0 0, 1 di .
(Recall B0 refers to the ball taken with respect to the usual norm.) Therefore, by Theorem 2.24, there exists
an orthogonal linear transformation, Q, which preserves distances measured with respect to the usual norm
and QDi Rn1 . Therefore, for z Ri ,


Qh (z) Qh (zi ) QDi + B0 0,  1 di .
Refering to (11.1), it follows that
mn (h (Ri )) = |det (Q)| mn (h (Ri )) = mn (Qh (Ri ))
mn QDi + B0 0,  1 di
Ck di + 2 1 di

n1

mn QDi + B 0,  1 di

2 1 di Ck,n mn (Ri ) .

Therefore,
mn (h (Zk ))

m (h (Ri )) Ck,n 

i=1

mn (Ri ) Ck,n mn (W )

i=1

Ck,n  (mn (Zk ) + ) .

Since  > 0 is arbitrary, this shows mn (h (Zk )) = 0. Now this implies
mn (h (Z)) = lim mn (h (Zk )) = 0
k

and this proves the lemma.

With Sards lemma it is easy to remove the assumption that det Dh (x) 6= 0.
Theorem 11.7 Let U be an open set in Rn and let h be a one to one mapping from U to Rn which is C 1 .
Then if g 0 is a Borel measurable function,
Z
Z
g (y) dy =
g (h (x)) |det Dh (x)| dx
h(U )

and every function which needs to be measurable, is. In particular the integrand of the second integral is
Lebesgue measurable.

11.1. GENERALIZATIONS

205

Proof: We observe first that h (U ) is measurable because, thanks to Theorem 3.32,

U =
i=1 Ui


where Ui is compact. Therefore, h (U ) =
i=1 h Ui , a Borel set. The function, x g (h (x)) is also
measurable from Lemma 11.3. Letting Z be described above, it follows by continuity of the derivative, Z
is a closed set. Thus the inverse function theorem applies and we may say that h (U \ Z) is an open set.
Therefore, if g 0 is a Borel measurable function,
Z
Z
g (y) dy =
g (y) dy =
h(U \Z)

h(U )

g (h (x)) |det Dh (x)| dx =

U \Z

g (h (x)) |det Dh (x)| dx,

the middle equality holding by Theorem 11.5 and the first equality holding by Sards lemma. This proves
the theorem.
It is possible to extend this theorem to the case when g is only assumed to be Lebesgue measurable. We
do this next using the following lemma.
Lemma 11.8 Suppose 0 f g, g is measurable with respect to the algebra of Lebesgue measurable sets,
S, and g = 0 a.e. Then f is also Lebesgue measurable.
Proof: Let a 0. Then
f 1 ((a, ]) g 1 ((a, ]) {x : g (x) > 0} ,
a set of measure zero. Therefore, by completeness of Lebesgue measure, it follows f 1 ((a, ]) is also
Lebesgue measurable. This proves the lemma.
To extend Theorem 11.7 to the case where g is only Lebesgue measurable, first suppose F is a Lebesgue
measurable set which has measure zero. Then by regularity of Lebesgue measure, there exists a Borel set,
G F such that mn (G) = 0 also. By Theorem 11.7,
Z
Z
0=
XG (y) dy =
XG (h (x)) |det Dh (x)| dx
h(U )

and the integrand of the second integral is Lebesgue measurable. Therefore, this measurable function,
x XG (h (x)) |det Dh (x)| ,
is equal to zero a.e. Thus,
0 XF (h (x)) |det Dh (x)| XG (h (x)) |det Dh (x)|
and Lemma 11.8 implies the function x XF (h (x)) |det Dh (x)| is also Lebesgue measurable and
Z
XF (h (x)) |det Dh (x)| dx = 0.
U

Now let F be an arbitrary bounded Lebesgue measurable set and use the regularity of the measure to
obtain G F E where G and E are Borel sets with the property that mn (G \ E) = 0. Then from what
was just shown,
x XF \E (h (x)) |det Dh (x)|

CHANGE OF VARIABLES FOR C 1 MAPS

206
is Lebesgue measurable and
Z
U

XF \E (h (x)) |det Dh (x)| dx = 0.

It follows since
XF (h (x)) |det Dh (x)| = XF \E (h (x)) |det Dh (x)| + XE (h (x)) |det Dh (x)|
and x XE (h (x)) |det Dh (x)| is measurable, that
x XF (h (x)) |det Dh (x)|
is measurable and so
Z

XF (h (x)) |det Dh (x)| dx =

XE (h (x)) |det Dh (x)| dx

XE (y) dy =

h(U )

XF (y) dy.

(11.9)

h(U )

To obtain the result in the case where F is an arbitrary Lebesgue measurable set, let
Fk F B (0, k)
and apply (11.9) to Fk and then let k and use the monotone convergence theorem to obtain (11.9) for
a general Lebesgue measurable set, F. It follows from this formula that we may replace XF with an arbitrary
nonnegative simple function. Now if g is an arbitrary nonnegative Lebesgue measurable function, we obtain
g as a pointwise limit of an increasing sequence of nonnegative simple functions and use the monotone
convergence theorem again to obtain (11.9) with XF replaced with g. This proves the following corollary.
Corollary 11.9 The conclusion of Theorem 11.7 hold if g is only assumed to be Lebesgue measurable.
Corollary 11.10 If g L1 (h (U ) , S, mn ) , then the conclusion of Theorem 11.7 holds.
Proof: Apply Corollary 11.9 to the positive and negative parts of the real and complex parts of g.
There are many other ways to prove the above results. To see some alternative presentations, see [24],
[19], or [11].
Next we give a version of this theorem which considers the case where h is only C 1 , not necessarily 1-1.
For
U+ {x U : |det Dh (x)| > 0}
and Z the set where |det Dh (x)| = 0, Lemma 11.6 implies m(h(Z)) = 0. For x U+ , the inverse function
theorem implies there exists an open set Bx such that
x Bx U+ , h is 1 1 on Bx .
Let {Bi } be a countable subset of {Bx }xU+ 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 1 1 on Ei , Ei Ej = ,
and each Ei is a Borel set contained in the open set Bi . Now we define
n(y) =

X
i=1

Thus n(y) 0 and is Borel measurable.

Xh(Ei ) (y) + Xh(Z) (y).

11.1. GENERALIZATIONS

207

Lemma 11.11 Let F h(U ) be measurable. Then

Z
Z
n(y)XF (y)dy =
XF (h(x))| det Dh(x)|dx.
h(U )

Proof: Using Lemma 11.6 and the Monotone Convergence Theorem or Fubinis Theorem,
!
Z
Z

X
n(y)XF (y)dy =
Xh(Ei ) (y) + Xh(Z) (y) XF (y)dy
h(U )

h(U )

=
=
=
=

Xh(Ei ) (y)XF (y)dy

i=1 h(U )
Z
X

Xh(Ei ) (y)XF (y)dy

i=1 h(Bi )
Z
X

XEi (x)XF (h(x))| det Dh(x)|dx

i=1 Bi
Z
X

XEi (x)XF (h(x))| det Dh(x)|dx

i=1

i=1

Z
X

Z X

XEi (x)XF (h(x))| det Dh(x)|dx

U i=1

XF (h(x))| det Dh(x)|dx =

U+

XF (h(x))| det Dh(x)|dx.

This proves the lemma.

Definition 11.12 For y h(U ),
#(y) |h1 (y)|.
Thus #(y) number of elements in h1 (y).
We observe that
#(y) = n(y) a.e.

(11.10)

And thus # is a measurable function. This follows because n(y) = #(y) if y

/ h(Z), a set of measure 0.
Theorem 11.13 Let g 0, g measurable, and let h be C 1 (U ). Then
Z
Z
#(y)g(y)dy =
g(h(x))| det Dh(x)|dx.
h(U )

(11.11)

Proof: From (11.10) and Lemma 11.11, (11.11) holds for all g, a nonnegative simple function. Approximating an arbitrary g 0 with an increasing pointwise convergent sequence of simple functions yields
(11.11) for g 0, g measurable. This proves the theorem.

CHANGE OF VARIABLES FOR C 1 MAPS

208

11.2

Exercises

1. The gamma function is defined by

()

et t1

for > 0. Show this is a well defined function of these values of and verify that ( + 1) = () .
What is (n + 1) for n a nonnegative integer?

R s2
R
2
2
2. Show that e 2 ds = 2. Hint: Denote this integral by I and observe that I 2 = R2 e(x +y )/2 dxdy.
Then change variables to polar coordinates, x = r cos () , y = r sin .
3. Now that you know what the gamma function is, consider in the formula for ( + 1) the following
change of variables. t = + 1/2 s. Then in terms of the new variable, s, the formula for ( + 1) is


Z
Z
h 

i

1
s
ln 1+ s s
+ 12
s

e + 2
ds
=
e

e
1
+
e
ds

Show the integrand converges to e

s2
2

. Show that then

Z

s2
( + 1)
lim +(1/2) =
e 2 ds = 2.
e

Hint: You will need to obtain a dominating function for theintegral

so that you can use the dominated
convergence theorem. You might try considering s ( , ) first and consider something like
2

e1(s /4) on this interval. Then look for another function for s > . This formula is known as
Stirlings formula.

The Lp Spaces
12.1

Basic inequalities and properties

The Lebesgue integral makes it possible to define and prove theorems about the space of functions described
below. These Lp spaces are very useful in applications of real analysis and this chapter is about these spaces.
In what follows (, S, ) will be a measure space.
Definition 12.1 Let 1 p < . We define
p

L () {f : f is measurable and

|f ()|p d < }

and
||f ||Lp

 p1
|f |p d
||f ||p .

Z

In fact || ||p is a norm if things are interpreted correctly. First we need to obtain Holders inequality. We
will always use the following convention for each p > 1.
1 1
+ = 1.
p q
Often one uses p0 instead of q in this context.

Theorem 12.2 (Holders inequality) If f and g are measurable functions, then if p > 1,
Z

|f | |g| d

Z

 p1 Z
 q1
|f |p d
|g|q d .

Proof: To begin with, we prove Youngs inequality.

Lemma 12.3 If 0 a, b then ab

ap
p

bq
q .

209

(12.1)

THE LP SPACES

210
Proof: Consider the following picture:

x
b
x = tp1
t = xq1
t
a

ab

p1

dt +

0
p

xq1 dx =

ap
bq
+ .
p
q

Note equalityR occurs when

R a =b .
If either |f |p d or |g|p d equals or 0, the inequality (12.1) is obviously valid. Therefore assume
both of these are less than and not equal to 0. By the lemma,
Z
Z
Z
|f |
|g|
1
|f |p
1
|g|q
d
d = 1.
p d +
||f ||p ||g||q
p
||f ||p
q
||g||qq
Hence,
Z

|f | |g| d ||f ||p ||g||q .

This proves Holders inequality.

Corollary 12.4 (Minkowski inequality) Let 1 p < . Then
||f + g||p ||f ||p + ||g||p .

(12.2)

Proof: If p = 1, this is obvious. Let p > 1. We can assume ||f ||p and ||g||p < and ||f + g||p 6= 0 or
there is nothing to prove. Therefore,
Z

Z
p
p1
p
p
|f + g| d 2
|f | + |g| d < .
Now
Z
Z

p1

|f + g|p d
Z

|f |d + |f + g|p1 |g|d
Z
Z
p
p
=
|f + g| q |f |d + |f + g| q |g|d
Z
Z
Z
Z
1
1
1
1
p
p
p
q
p
q
( |f + g| d) ( |f | d) + ( |f + g| d) ( |g|p d) p.
|f + g|

R
1
Dividing both sides by ( |f + g|p d) q yields (12.2). This proves the corollary.

12.1. BASIC INEQUALITIES AND PROPERTIES

211

This shows that if f, g Lp , then f + g Lp . Also, it is clear that if a is a constant and f Lp , then
af Lp . Hence Lp is a vector space. Also we have the following from the Minkowski inequality and the
definition of || ||p .
a.) ||f ||p 0, ||f ||p = 0 if and only if f = 0 a.e.
b.) ||af ||p = |a| ||f ||p if a is a scalar.
c.) ||f + g||p ||f ||p + ||g||p .
We see that || ||p would be a norm if ||f ||p = 0 implied f = 0. If we agree to identify all functions in Lp
that differ only on a set of measure zero, then || ||p is a norm and Lp is a normed vector space. We will do
so from now on.
Definition 12.5 A complete normed linear space is called a Banach space.
Next we show that Lp is a Banach space.
Theorem 12.6 The following hold for Lp ()
a.) Lp () is complete.
b.) If {fn } is a Cauchy sequence in Lp (), then there exists f Lp () and a subsequence which converges
a.e. to f Lp (), and ||fn f ||p 0.
Proof: Let {fn } be a Cauchy sequence in Lp (). This means that for every > 0 there exists N
such that if n, m N , then ||fn fm ||p < . Now we will select a subsequence. Let n1 be such that
||fn fm ||p < 21 whenever n, m n1 . Let n2 be such that n2 > n1 and ||fn fm ||p < 22 whenever
n, m n2 . If n1 , , nk have been chosen, let nk+1 > nk and whenever n, m nk+1 , ||fn fm ||p < 2(k+1) .
The subsequence will be {fnk }. Thus, ||fnk fnk+1 ||p < 2k . Let
gk+1 = fnk+1 fnk .
Then by the Minkowski inequality,

>

||gk+1 ||p

k=1

k=1

for all m. It follows that

Z

m
X

m
X

!p

|gk+1 |



m
 X



||gk+1 ||p 
|gk+1 |



k=1

k=1

||gk+1 ||p

!p

<

(12.3)

k=1

for all m and so the monotone convergence theorem implies that the sum up to m in (12.3) can be replaced
by a sum up to . Thus,

|gk+1 (x)| < a.e. x.

k=1

Therefore,

k=1 gk+1 (x)

converges for a.e. x and for such x, let

f (x) = fn1 (x) +

gk+1 (x).

k=1

Pm
Note that k=1 gk+1 (x) = fnm+1 (x)fn1 (x). Therefore limk fnk (x) = f (x) for all x
/ E where (E) = 0.
If we redefine fnk to equal 0 on E and let f (x) = 0 for x E, it then follows that limk fnk (x) = f (x) for
all x. By Fatous lemma,
||f fnk ||p lim inf ||fnm fnk ||p
m

X


fni+1 fni  2(k1).
p
i=k

THE LP SPACES

212
Therefore, f Lp () because
||f ||p ||f fnk ||p + ||fnk ||p < ,

and limk ||fnk f ||p = 0. This proves b.). To see that the original Cauchy sequence converges to f in
Lp (), we write
||f fn ||p ||f fnk ||p + ||fnk fn ||p .
If > 0 is given, let 2(k1) < 2 . Then if n > nk,

+ = .
2 2

||f fn || < 2(k1) + 2k <

This proves part a.) and completes the proof of the theorem.
In working with the Lp spaces, the following inequality also known as Minkowskis inequality is very
useful.
Theorem 12.7 Let (X, S, ) and (Y, F, ) be -finite measure spaces and let f be product measurable. Then
the following inequality is valid for p 1.
Z Z
X

|f (x, y)| d

 p1

 p1
Z
p
( |f (x, y)| d) d .

Z

Proof: Let Xn X, Yn Y, (Yn ) < , (Xn ) < , and let


f (x, y) if |f (x, y)| m,
fm (x, y) =
m
if |f (x, y)| > m.
Thus
Z

Z
(

Yn

|fm (x, y)|d)p d

 p1

< .

Xk

Let
J(y) =

|fm (x, y)|d.

Xk

Then
Z
Yn

p
|fm (x, y)|d d

Z
Xk

Z
J(y)p1
|fm (x, y)|d d
Yn
Xk
Z Z
=
J(y)p1 |fm (x, y)|d d
=

Xk

Yn

by Fubinis theorem. Now we apply Holders inequality and recall p 1 = pq . This yields
Z

Z
Yn

Z

Xk

Xk

|fm (x, y)|d

p

Z
Yn

J(y) d

 q1 Z

Yn

|fm (x, y)| d

 p1

 p1

Yn
p

J(y) d

 q1 Z
Xk

Z
Yn

|fm (x, y)| d

(12.4)

12.2. DENSITY OF SIMPLE FUNCTIONS

Z

Z
(

Yn

213

|fm (x, y)|d)p d

 q1 Z

Xk

Z

Xk

|fm (x, y)|p d

 p1

d.

Yn

Therefore,
Z
Yn

Z

|fm (x, y)|d

p

 p1

Xk

Z
Xk

Z

|fm (x, y)|p d

 p1

d.

(12.5)

Yn

To obtain (12.4) let m and use the Monotone Convergence theorem to replace fm by f in (12.5). Next
let k and use the same theorem to replace Xk with X. Finally let n and use the Monotone
Convergence theorem again to replace Yn with Y . This yields (12.4).
Next, we develop some properties of the Lp spaces.

12.2

Density of simple functions

Theorem 12.8 Let p 1 and let (, S, ) be a measure space. Then the simple functions are dense in
Lp ().
Proof: By breaking an arbitrary function into real and imaginary parts and then considering the positive
and negative parts of these, we see that there is no loss of generality in assuming f has values in [0, ]. By
Theorem 5.31, there is an increasing sequence of simple functions, {sn }, converging pointwise to f (x)p . Let
1
tn (x) = sn (x) p . Thus, tn (x) f (x) . Now
|f (x) tn (x)| |f (x)|.
By the Dominated Convergence theorem, we may conclude
Z
0 = lim
|f (x) tn (x)|p d.
n

Thus simple functions are dense in L .

Recall that for a topological space, Cc () is the space of continuous functions with compact support
in . Also recall the following definition.
Definition 12.9 Let (, S, ) be a measure space and suppose (, ) is also a topological space. Then
(, S, ) is called a regular measure space if the -algebra of Borel sets is contained in S and for all E S,
(E) = inf{(V ) : V E and V open}
and
(E) = sup{(K) : K E and K is compact }.
For example Lebesgue measure is an example of such a measure.
Lemma 12.10 Let be a locally compact metric space and let K be a compact subset of V , an open set.
Then there exists a continuous function f : [0, 1] such that f (x) = 1 for all x K and spt(f ) is a
compact subset of V .
Proof: Let K W W V and W is compact. Define f by
f (x) =

dist(x, W C )
.
dist(x, K) + dist(x, W C )

It is not necessary to be in a metric space to do this. You can accomplish the same thing using Urysohns
lemma.

THE LP SPACES

214

Theorem 12.11 Let (, S, ) be a regular measure space as in Definition 12.9 where the conclusion of
Lemma 12.10 holds. Then Cc () is dense in Lp ().
Proof: Let f Lp () and pick a simple function, s, such that ||s f ||p <

where > 0 is arbitrary.

Let
m
X

s(x) =

ci XEi (x)

i=1

where c1 , , cm are the distinct nonzero values of s. Thus the Ei are disjoint and (Ei ) < for each i.
Therefore there exist compact sets, Ki and open sets, Vi , such that Ki Ei Vi and
m
X

|ci |(Vi \ Ki ) p <

i=1

.
2

Let hi Cc () satisfy
hi (x) = 1 for x Ki
spt(hi ) Vi .
Let
g=

m
X

ci hi .

i=1

Then by Minkowskis inequality,

||g s||p

! p1
m
X
p
(
|ci | |hi (x) XEi (x)|) d

i=1

m Z
X
i=1
m
X

|ci | |hi (x) XEi (x)| d

 p1

|ci |(Vi \ Ki ) p <

i=1

.
2

Therefore,
||f g||p ||f s||p + ||s g||p <

+ = .
2 2

This proves the theorem.

12.3

Continuity of translation

Definition 12.12 Let f be a function defined on U Rn and let w Rn . Then fw will be the function
defined on w + U by
fw (x) = f (x w).
Theorem 12.13 (Continuity of translation in Lp ) Let f Lp (Rn ) with = m, Lebesgue measure. Then
lim ||fw f ||p = 0.

||w||0

12.4. SEPARABILITY

215

Proof: Let > 0 be given and let g Cc (Rn ) with ||g f ||p < 3 . Since Lebesgue measure is translation
invariant (m(w + E) = m(E)), ||gw fw ||p = ||g f ||p < 3 . Therefore
||f fw ||p

<

||f g||p + ||g gw ||p + ||gw fw ||

2
+ ||g gw ||p .
3

But lim|w|0 gw (x) = g(x) uniformly in x because g is uniformly continuous. Therefore, whenever |w| is
small enough, ||g gw ||p < 3 . Thus ||f fw ||p < whenever |w| is sufficiently small. This proves the
theorem.

12.4

Separability

Theorem 12.14 For p 1, Lp (Rn , m) is separable. This means there exists a countable set, D, such that
if f Lp (Rn ) and > 0, there exists g D such that ||f g||p < .
Proof: Let Q be all functions of the form cX[a,b) where
[a, b) [a1 , b1 ) [a2 , b2 ) [an , bn ),
and both ai , bi are rational, while c has rational real and imaginary parts. Let D be the set of all finite
sums of functions in Q. Thus, D is countable. We now show that D is dense in Lp (Rn , m). To see this, we
need to show that for every f Lp (Rn ), there exists an element of D, s such that ||s f ||p < . By Theorem
12.11 we can assume without loss of generality that f Cc (Rn ). Let Pm consist of all sets of the form
[a, b) where ai = j2m and bi = (j + 1)2m for j an integer. Thus Pm consists of a tiling of Rn into half open
1
rectangles having diameters 2m n 2 . There are countably many of these rectangles; so, let Pm = {[ai , bi )}
m
and Rn =
i=1 [ai , bi ). Let ci be complex numbers with rational real and imaginary parts satisfying
m
|f (ai ) cm
,
i |<5

|cm
i | |f (ai )|.

(12.6)

P m
Let sm (x) =
i=1 ci X[ai ,bi ) . Since f (ai ) = 0 except for finitely many values of i, (12.6) implies sm
D. It is also clear that, since f is uniformly continuous, limm sm (x) = f (x) uniformly in x. Hence
limx0 ||sm f ||p = 0.
Corollary 12.15 Let be any Lebesgue measurable subset of Rn . Then Lp () is separable. Here the
algebra of measurable sets will consist of all intersections of Lebesgue measurable sets with and the measure
will be mn restricted to these sets.
e be the restrictions of D to . If f Lp (), let F be the zero extension of f to all of Rn .
Proof: Let D
Let > 0 be given. By Theorem 12.14 there exists s D such that ||F s||p < . Thus
||s f ||Lp () ||s F ||Lp (Rn ) <
e is dense in Lp ().
and so the countable set D

12.5

Mollifiers and density of smooth functions

Definition 12.16 Let U be an open subset of Rn . Cc (U ) is the vector space of all infinitely differentiable
functions which equal zero for all x outside of some compact set contained in U .

THE LP SPACES

216
Example 12.17 Let U = {x Rn : |x| < 2}


1 

2
exp |x| 1
if |x| < 1,
(x) =

0 if |x| 1.
Then a little work shows Cc (U ). The following also is easily obtained.
Lemma 12.18 Let U be any open set. Then Cc (U ) 6= .

Definition 12.19 Let U = {x Rn : |x| < 1}. A sequence { m } Cc (U ) is called a mollifier (sometimes
an approximate identity) if
m (x) 0, m (x) = 0, if |x|
and

1
,
m

m (x) = 1.
R
As before, f (x, y)d(y) will mean x is fixed and the function y f (x, y) is being integrated. We may
also write dx for dm(x) in the case of Lebesgue measure.
Example 12.20 Let
Cc (B(0, 1)) (B(0, 1) = {x : |x| < 1})
R
R
with (x) 0 and dm = 1. Let m (x) = cm (mx) where cm is chosen in such a way that m dm = 1.
By the change of variables theorem we see that cm = mn .
Definition 12.21 A function, f , is said to be in L1loc (Rn , ) if f is measurable and if |f |XK L1 (Rn , ) for
every compact set, K. Here is a Radon measure on Rn . Usually = m, Lebesgue measure. When this is
so, we write L1loc (Rn ) or Lp (Rn ), etc. If f L1loc (Rn ), and g Cc (Rn ),
Z
Z
f g(x) f (x y)g(y)dm = f (y)g(x y)dm.

The following lemma will be useful in what follows.

Lemma 12.22 Let f L1loc (Rn ), and g Cc (Rn ). Then f g is an infinitely differentiable function.
Proof: We look at the difference quotient for calculating a partial derivative of f g.
Z
f g (x + tej ) f g (x)
1
g(x + tej y) g (x y)
=
f (y)
dm (y) .
t
t
t
g(x+te y)g(xy)

j
Using the fact that g Cc (Rn ) , we can argue that the quotient,
is uniformly bounded.
t
Therefore, there exists a dominating function for the integrand of the above integral which is of the form
C |f (y)| XK where K is a compact set containing the support of g. It follows we can take the limit of the
difference quotient above inside the integral and write
Z

(f g) (x) = f (y)
g (x y) dm (y) .
xj
xj

Now letting x
g play the role of g in the above argument, we can continue taking partial derivatives of all
j
orders. This proves the lemma.

12.5. MOLLIFIERS AND DENSITY OF SMOOTH FUNCTIONS

217

Theorem 12.23 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.
Proof: Let r > 0 be small enough that K + B(0, 3r) U. Let Kr = K + B(0, r).

Kr U

1
Consider XKr m where m is a mollifier. Let m be so large that m
< r. Then using Lemma 12.22 it is
straightforward to verify that h = XKr m satisfies the desired conclusions.
Although we will not use the following corollary till later, it follows as an easy consequence of the above
theorem and is useful. Therefore, we state it here.

Corollary 12.24 Let K be a compact set in Rn and let {Ui }i=1 be an open cover of K. Then there exists
functions, k Cc (Ui ) such that i Ui and

i (x) = 1.

i=1

If K1 is a compact subset of U1 we may also take 1 such that 1 (x) = 1 for all x K1 .
Proof: This follows from a repeat of the proof of Theorem 6.11, replacing the lemma used in that proof
with Theorem 12.23.
Theorem 12.25 For each p 1, Cc (Rn ) is dense in Lp (Rn ).

gm

Proof: Let f Lp (Rn ) and let > 0 be given. Choose g Cc (Rn ) such that ||f g||p < 2 . Now let
= g m where { m } is a mollifier.
[gm (x + hei ) gm (x)]h1

= h1

g(y)[ m (x + hei y) m (x y)]dm.

The integrand is dominated by C|g(y)|h for some constant C depending on

max {| m (x)/xj | : x Rn , j {1, 2, , n}}.
By the Dominated Convergence theorem, the limit as h 0 exists and yields
gm (x)
=
xi

g(y)

m (x y)
dy.
xi

THE LP SPACES

218

Similarly, all other partial derivatives exist and are continuous as are all higher order derivatives. Conse1
).
quently, gm Cc (Rn ). It vanishes if x
/ spt(g) + B(0, m
||g gm ||p

Z

|g(x)

 p1
g(x y) m (y)dm(y)| dm(x)
p

Z Z
 p1
p

( |g(x) g(x y)| m (y)dm(y)) dm(x)

Z Z

 p1
|g(x) g(x y)|p dm(x)
m (y)dm(y)

1
B(0, m
)

<

||g gy ||p m (y)dm(y)

whenever m is large enough. This follows from Theorem 12.13. Theorem 12.7 was used to obtain the third
inequality. There is no measurability problem because the function
(x, y) |g(x) g(x y)| m (y)
is continuous so it is surely Borel measurable, hence product measurable. Thus when m is large enough,
||f gm ||p ||f g||p + ||g gm ||p <

+ = .
2 2

This proves the theorem.

12.6

Exercises

1. 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.
Note f is continuous at 0 and f (0) > 0. Remember continuity of translation in Lp .
2. Give an example of a sequence of functions in Lp (R) which converges to zero in Lp but does not
converge pointwise to 0. Does this contradict the proof of the theorem that Lp is complete?
R
3. Let m Cc (Rn ), m (x) 0,and Rn m (y)dy = 1 with limm sup {|x| : x sup (m )} = 0. Show
if f Lp (Rn ), limm f m = f in Lp (Rn ).

4. Let : R R be convex. This means

(x + (1 )y) (x) + (1 )(y)
whenever [0, 1]. Show that if is convex, then is continuous. Also verify that if x < y < z, then
(y)(x)
(z)(y)
and that (z)(x)
(z)(y)
.
yx
zy
zx
zy

12.6. EXERCISES

219

1
5. Prove
Jensens
R
R inequality. If : R
R R is convex, () = 1, and f : R is in L (), then
( f du) (f )d. Hint: Let s = f d and show there exists such that (s) (t)+(st)
for all t.
0

6. Let p1 + p10 = 1, p > 1, let f Lp (R), g Lp (R). Show f g is uniformly continuous on R and
|(f g)(x)| ||f ||Lp ||g||Lp0 .
R1
R
7. B(p, q) = 0 xp1 (1 x)q1 dx, (p) = 0 et tp1 dt for p, q > 0. The first of these is called the beta
function, while the second is the gamma function. Show a.) (p + 1) = p(p); b.) (p)(q) =
B(p, q)(p + q).
Rx
8. Let f Cc (0, ) and define F (x) = x1 0 f (t)dt. Show
||F ||Lp (0,)
Hint: Use xF 0 = f F and integrate

p
||f ||Lp (0,) whenever p > 1.
p1

R
0

|F (x)|p dx by parts.

Rx
9. Now suppose f Lp (0, ), p > 1, and f not necessarily in Cc (0, ). Note that F (x) = x1 0 f (t)dt
still makes sense for each x > 0. Is the inequality of Problem 8 still valid? Why? This inequality is
called Hardys inequality.
10. When does equality hold in Holders inequality? Hint: First suppose f, g 0. This isolates the most
interesting aspect of the question.
11. Consider Hardys inequality of Problems 8 and 9. Show equality holds only
If
R if f = 0 a.e.
R Hint:
p
equality holds, we can assume f 0. Why? You might then establish (p 1) 0 F p dx = p 0 F q f dx
and use Problem 10.
12. In Hardys inequality, show the constant p(p 1)1 cannot be improved. Also show that if f > 0
1
and f L1 , then F
/ L1 so it is important that p > 1. Hint: Try f (x) = x p X[A1 ,A] .
13. AR set of functions, L1 , is uniformly integrable if for all > 0 there exists a > 0 such that
 f du < whenever (E) < . Prove Vitalis Convergence theorem: Let {fn } be uniformly
E
1
integrable, ()
R < , fn (x) f (x) a.e. where f is measurable and |f (x)| < a.e. Then f L
and limn |fn f |d = 0. Hint: Use Egoroffs theorem to show {fn } is a Cauchy sequence in
L1 () . This yields a different and easier proof than what was done earlier.
14. Show the Vitali Convergence theorem implies the Dominated Convergence theorem for finite measure
spaces.
15. Suppose () < , {fn } L1 (), and
Z

h (|fn |) d < C

for all n where h is a continuous, nonnegative function satisfying

lim

h (t)
= .
t

Show {fn } is uniformly integrable.

16. Give an example of a situation in which the Vitali Convergence theorem applies, but the Dominated
Convergence theorem does not.

THE LP SPACES

220

17. Sometimes, especially in books on probability, a different definition of uniform integrability is used
than that presented here. A set of functions, S, defined on a finite measure space, (, S, ) is said to
be uniformly integrable if for all  > 0 there exists > 0 such that for all f S,
Z
|f | d .
[|f |]

Show that this definition is equivalent to the definition of uniform integrability given earlier with the
addition of the condition that there is a constant, C < such that
Z
|f | d C
for all f S. If this definition of uniform integrability is used, show that if fn () f () a.e., then
it is automatically the case that |f ()| < a.e. so it is not necessary to check this condition in the
hypotheses for the Vitali convergence theorem.
18. We say f L (, ) if there exists a set of measure zero, E, and a constant C < such that
|f (x)| C for all x
/ E.
||f || inf{C : |f (x)| C a.e.}.
Show || || is a norm on L (, ) provided we identify f and g if f (x) = g(x) a.e. Show L (, ) is
complete.
19. Suppose f L L1 . Show limp ||f ||Lp = ||f || .
R1
R1
20. Suppose : R R and ( 0 f (x)dx) 0 (f (x))dx for every real bounded measurable f . Can it be
concluded that is convex?
21. Suppose () < . Show that if 1 p < q, then Lq () Lp ().
22. Show L1 (R) * L2 (R) and L2 (R) * L1 (R) if Lebesgue measure is used.
23. Show that if x [0, 1] and p 2, then
(

1x p
1
1+x p
) +(
) (1 + xp ).
2
2
2

Note this is obvious if p = 2. Use this to conclude the following inequality valid for all z, w C and
p 2.




p
p
 z + w p  z w  p

 +
 |z| + |w| .
 2 
 2 
2
2
Hint: For the first part, divide both sides by xp , let y =
for all y 1. If |z| |w| > 0, this takes the form

1
x

and show the resulting inequality is valid

1
1
1
| (1 + rei )|p + | (1 rei )|p (1 + rp )
2
2
2
whenever 0 < 2 and r [0, 1]. Show the expression on the left is maximized when = 0 and use
the first part.
24. If p 2, establish Clarksons inequality. Whenever f, g Lp ,

  p 
p
 1



 (f + g) +  1 (f g) 1 ||f ||p + 1 ||g||p .
p
 2

 2

2
2
p
p

For more on Clarkson inequalities (there are others), see Hewitt and Stromberg [15] or Ray [22].

12.6. EXERCISES

221

25. Show that for p 2, Lp is uniformly convex. This means that if {fn }, {gn } Lp , ||fn ||p , ||gn ||p 1,
and ||fn + gn ||p 2, then ||fn gn ||p 0.
26. Suppose that [0, 1] and r, s, q > 0 with
1
1
= +
.
q
r
s
show that
Z
Z
Z
( |f |q d)1/q (( |f |r d)1/r ) (( |f |s d)1/s )1.
If q, r, s 1 this says that
||f ||q ||f ||r ||f ||1
.
s
Hint:
Z
Now note that 1 =

q
r

q(1)
s

|f | d =

|f |q |f |q(1) d.

and use Holders inequality.

27. Generalize Theorem 12.7 as follows. Let 0 p1 p2 < . Then

Z Z
Y

p1

|f (x, y)|

p2 /p1 !1/p2

d
d

Z Z
X

p2

|f (x, y)|

p1 /p2

!1/p1

222

THE LP SPACES

Fourier Transforms
13.1

The Schwartz class

The Fourier transform of a function in L1 (Rn ) is given by

Z
F f (t) (2)n/2
eitx f (x)dx.
Rn

However, we want to take the Fourier transform of many other kinds of functions. In particular we want to
take the Fourier transform of functions in L2 (Rn ) which is not a subset of L1 (Rn ). Thus the above integral
may not make sense. In defining what is meant by the Fourier Transform of more general functions, it is
convenient to use a special class of functions known as the Schwartz class which is a subset of Lp (Rn ) for all
p 1. The procedure is to define the Fourier transform of functions in the Schwartz class and then use this
to define the Fourier transform of a more general function in terms of what happens when it is multiplied
by the Fourier transform of functions in the Schwartz class.
The functions in the Schwartz class are infinitely differentiable and they vanish very rapidly as |x|
along with all their partial derivatives. To describe precisely what we mean by this, we need to present some
notation.
Definition 13.1 = (1 , , n ) for 1 n positive integers is called a multi-index. For a multi-index,
|| 1 + + n and if x Rn ,
x = (x1 , , xn ),
and f a function, we define
n

1 2
x x
1 x2 xn , D f (x)

|| f (x)
.
n
x
n

2
1
x
1 x2

Definition 13.2 We say f S, the Schwartz class, if f C (Rn ) and for all positive integers N,
N (f ) <
where
2

N (f ) = sup{(1 + |x| )N |D f (x)| : x Rn , || N }.

Thus f S if and only if f C (Rn ) and
sup{|x D f (x)| : x Rn } <
for all multi indices and .
223

(13.1)

224

FOURIER TRANSFORMS

Also note that if f S, then p(f ) S for any polynomial, p with p(0) = 0 and that
S Lp (Rn ) L (Rn )
for any p 1.
Definition 13.3 (Fourier transform on S ) For f S,
Z
F f (t) (2)n/2
eitx f (x)dx,
Rn

F 1 f (t) (2)n/2

eitx f (x)dx.

Rn

Here x t =

Pn

i=1

xi ti .

It will follow from the development given below that (F F 1 )(f ) = f and (F 1 F )(f ) = f whenever
f S, thus justifying the above notation.
Theorem 13.4 If f S, then F f and F 1 f are also in S.
Proof: To begin with, let = ej = (0, 0, , 1, 0, , 0), the 1 in the jth slot.
Z
F 1 f (t + hej ) F 1 f (t)
eihxj 1
n/2
= (2)
)dx.
eitx f (x)(
h
h
Rn
Consider the integrand in (13.2).


ihxj
 itx
1 
e f (x)( e
)

h

(13.2)

 i(h/2)x

j
 e
ei(h/2)xj 

= |f (x)| (
)
h


 i sin ((h/2) xj ) 

= |f (x)| 

(h/2)
|f (x)| |xj |

and this is a function in L1 (Rn ) because f S. Therefore by the Dominated Convergence Theorem,
Z
F 1 f (t)
n/2
= (2)
eitx ixj f (x)dx
tj
Rn
Z
n/2
= i(2)
eitx xej f (x)dx.
Rn

Now xej f (x) S and so we may continue in this way and take derivatives indefinitely. Thus F 1 f C (Rn )
and from the above argument,
Z

D F 1 f (t) =(2)n/2
eitx (ix) f (x)dx.
Rn

To complete showing F 1 f S,
t D F 1 f (t) =(2)n/2

eitx t (ix) f (x)dx.

Rn

Integrate this integral by parts to get

t D F

n/2

f (t) =(2)

Rn

i|| eitx D ((ix) f (x))dx.

(13.3)

13.1. THE SCHWARTZ CLASS

225

Here is how this is done.

Z

eitj xj tj j (ix) f (x)dxj

eitj xj j
t (ix) f (x) |
+
itj j
Z
1
i eitj xj tj j Dej ((ix) f (x))dxj

where the boundary term vanishes because f S. Returning to (13.3), we use (13.1), and the fact that
|eia | = 1 to conclude
Z
a
1
|t D F f (t)| C
|D ((ix) f (x))|dx < .
Rn

It follows F 1 f S. Similarly F f S whenever f S.

Theorem 13.5 F F 1 (f ) = f and F 1 F (f ) = f whenever f S.
Before proving this theorem, we need a lemma.
Lemma 13.6
(2)n/2

e(1/2)uu du = 1,

(13.4)

e(1/2)(uia)(uia) du = 1.

(13.5)

Rn

n/2

(2)

Rn

Proof:
Z
2
( ex /2 dx)2

Z Z

R R
Z Z 2

e(x

+y 2 )/2

rer

/2

dxdy

ddr = 2.

Therefore
Z

e(1/2)uu du =

Rn

n Z
Y

i=1

exj /2 dxj = (2)n/2.

This proves (13.4). To prove (13.5) it suffices to show that

Z
2
e(1/2)(xia) dx = (2)1/2.

(13.6)

Define h(a) to be the left side of (13.6). Thus

Z
2
2
h(a) = ( e(1/2)x (cos(ax) + i sin ax) dx)ea /2
ZR
2
2
= ( e(1/2)x cos(ax)dx)ea /2
R

because sin is an odd function. We know h (0) = (2)1/2 .

Z
2
0
a2 /2 d
( ex /2 cos(ax)dx).
h (a) = ah(a) + e
da R

(13.7)

226

FOURIER TRANSFORMS

Forming difference quotients and using the Dominated Convergence Theorem, we can take
integral in (13.7) to obtain
Z
2
xe(1/2)x sin(ax)dx.

d
da

inside the

Integrating this by parts yields

Z
Z
2
2
d
( ex /2 cos(ax)dx) = a( ex /2 cos(ax)dx).
da R
R
Therefore
h0 (a)

= ah(a) aea

/2

ex

/2

cos(ax)dx

= ah(a) ah(a) = 0.
This proves the lemma since h(0) = (2)1/2 .
Proof of Theorem 13.5 Let
2

g (x) = e(

/2)xx

Thus 0 g (x) 1 and lim0+ g (x) = 1. By the Dominated Convergence Theorem,

Z
n
(F F 1 )f (x) = lim (2) 2
F 1 f (t)g (t)eitx dt.
0

Rn

Therefore,
(F F 1 )f (x) =

=
=
=

lim (2)n

Rn

lim (2)n

ei(yx)t f (y)g (t)dtdy

Z
n
2
f (y)[(2)
ei(yx)t g (t)dt]dy.

Rn

n
2

lim (2)

eiyt f (y)g (t)eixt dydt

Rn

Rn

Rn

(13.8)

Rn

Consider [ ] in (13.8). This equals

n
2 n

(2)

1 xy 2
|

e 2 (uia)(uia) e 2 |

du,

Rn

where a = 1 (y x), and |z| = (z z) 2 . Applying Lemma 13.6

n

1 xy 2

(2) 2 [ ] = (2) 2 n e 2 | |
m (y x) = m (x y)
and by Lemma 13.6,
Z

Rn

m (y)dy = 1.

(13.9)

13.1. THE SCHWARTZ CLASS

227

Thus from (13.8),

(F F 1 )f (x)

lim

f (y)m (x y)dy

(13.10)

Rn

lim f m (x).

n
2

m (y)dy = (2)

|y|

Z
(

1 y 2

e 2 | | dy)n.

|y|

Using polar coordinates,

= (2)n/2

2
2
e(1/(2 )) n1 ddn

S n1

Z
= (2)n/2 (

e(1/2) n1 d)Cn .

This clearly converges to 0 as 0+ because of the Dominated Convergence Theorem and the fact that
2
n1 e /2 is in L1 (R). Hence
lim

m (y)dy = 0.

|y|

Let be small enough that |f (x) f (x y)| < whenever |y| . Therefore, from Formulas (13.9) and
(13.10),
|f (x) (F F 1 )f (x)| = lim |f (x) f m (x)|
0

lim sup

lim sup

|f (x) f (x y)|m (y)dy

Rn

|f (x) f (x y)|m (y)dy+

|y|>

|f (x) f (x y)|m (y)dy

|y|

Z
lim sup ((
0

|y|>

m (y)dy)2||f || + ) = .

Since > 0 is arbitrary, f (x) = (F F 1 )f (x) whenever f S. This proves Theorem 13.5 and justifies the
notation in Definition 13.3.

228

13.2

FOURIER TRANSFORMS

Fourier transforms of functions in L2 (Rn )

With this preparation, we are ready to begin the consideration of F f and F 1 f for f L2 (Rn ). First note
that the formula given for F f and F 1 f when f S will not work for f L2 (Rn ) unless f is also in L1 (Rn ).
The following theorem will make possible the definition of F f and F 1 f for arbitrary f L2 (Rn ).
Theorem 13.7 For S, ||F ||2 = ||F 1 ||2 = ||||2 .
Proof: First note that for S,
= F 1 () , F 1 ()
= F ().
F ()
This follows from the definition. Let , S.
Z
Z Z
(F (t))(t)dt = (2)n/2
(t)(x)eitx dxdt
n
n
n
R
R
R
Z
=
(x)(F (x))dx.

(13.11)

(13.12)

Rn

Similarly,
Z

(x)(F 1 (x))dx =

Rn

(F 1 (t))(t)dt.

(13.13)

Rn

Now, (13.11) - (13.13) imply

Z

|(x)| dx =

Rn

(x)F 1 (F (x))dx

(x)F (F (x))dx

Rn

Rn

ZR

F (x)(F (x))dx
n

|F |2 dx.

Rn

Similarly
||||2 = ||F 1 ||2 .
This proves the theorem.
With this theorem we are now able to define the Fourier transform of a function in L2 (Rn ) .
Definition 13.8 Let f L2 (Rn ) and let {k } be a sequence of functions in S converging to f in L2 (Rn ) .
We know such a sequence exists because S is dense in L2 (Rn ) . (Recall that even Cc (Rn ) is dense in
L2 (Rn ) .) Then F f limk F k where the limit is taken in L2 (Rn ) . A similar definition holds for F 1 f.
Lemma 13.9 The above definition is well defined.
Proof: We need to verify two things. First we need to show that limk F (k ) exists in L2 (Rn ) and
next we need to verify that this limit is independent of the sequence of functions in S which is converging
to f.
To verify the first claim, note that sincelimk ||f k ||2 = 0, it follows that {k } is a Cauchy sequence.
Therefore, by Theorem 13.7 {F k } and F 1 k are also Cauchy sequences in L2 (Rn ) . Therefore, they
both converge in L2 (Rn ) to a unique element of L2 (Rn ). This verifies the first part of the lemma.

13.2. FOURIER TRANSFORMS OF FUNCTIONS IN L2 RN

229

Now suppose {k } and { k } are two sequences of functions in S which converge to f L2 (Rn ) .
Do {F k } and {F k } converge to the same element of L2 (Rn )? We know that for k large, ||k k ||2
||k f ||2 +||f k ||2 < 2 + 2 = . Therefore, for large enough k, we have ||F k F k ||2 = ||k k ||2 <
and so, since > 0 is arbitrary, it follows that {F k } and {F k } converge to the same element of L2 (Rn ) .
We leave as an easy exercise the following identity for , S.
Z
Z
(x)F (x)dx =
F (x)(x)dx
Rn

Rn

and
Z

(x)F

(x)dx =

Rn

F 1 (x)(x)dx

Rn

Theorem 13.10 If f L2 (Rn ), F f and F 1 f are the unique elements of L2 (Rn ) such that for all S,
Z
Z
F f (x)(x)dx =
f (x)F (x)dx,
(13.14)
Rn

Rn

F 1 f (x)(x)dx =

Rn

f (x)F 1 (x)dx.

(13.15)

Rn

Proof: Let {k } be a sequence in S converging to f in L2 (Rn ) so F k converges to F f in L2 (Rn ) .

Therefore, by Holders inequality or the Cauchy Schwartz inequality,
Z
Z
F f (x)(x)dx = lim
F k (x)(x)dx
k Rn
Rn
Z
k (x)F (x)dx
= lim
k Rn
Z
=
f (x)F (x)dx.
Rn

A similar formula holds for F 1 . It only remains to verify uniqueness. Suppose then that for some G
L2 (Rn ) ,
Z
Z
G (x) (x) dx =
F f (x)(x)dx
(13.16)
Rn

Rn

for all S. Does it follow that G = F f in L2 (Rn )? Let {k } be a sequence of elements of S converging
in L2 (Rn ) to G F f . Then from (13.16),
Z
Z
2
0 = lim
(G (x) F f (x)) k (x) dx =
|G (x) F f (x)| dx.
k

Rn

Rn

Thus G = F f and this proves uniqueness. A similar argument applies for F 1 .

Theorem 13.11 (Plancherel) For f L2 (Rn ).
(F 1 F )(f ) = f = (F F 1 )(f ),

(13.17)

||f ||2 = ||F f ||2 = ||F 1 f ||2 .

(13.18)

230

FOURIER TRANSFORMS

Proof: Let S.
Z

(F

F )(f )dx =

Rn

Ff F

dx =

Rn

f F (F 1 ())dx

Rn

f dx.

Thus (F 1 F )(f ) = f because S is dense in L2 (Rn ). Similarly (F F 1 )(f ) = f . This proves (13.17). To
show (13.18), we use the density of S to obtain a sequence, {k } converging to f in L2 (Rn ) . Then
||F f ||2 = lim ||F k ||2 = lim ||k ||2 = ||f ||2 .
k

Similarly,
||f ||2 = ||F 1 f ||2.
This proves the theorem.
The following corollary is a simple generalization of this. To prove this corollary, we use the following
simple lemma which comes as a consequence of the Cauchy Schwartz inequality.
Lemma 13.12 Suppose fk f in L2 (Rn ) and gk g in L2 (Rn ) . Then
Z
Z
lim
fk gk dx =
f gdx
k

Rn

Rn

Proof:
Z




fk gk dx

Rn

Rn

 Z
 
f gdx 

fk gk dx

Rn

fk gdx

Rn

Rn

Rn



fk gdx +



f gdx

||fk ||2 ||g gk ||2 + ||g||2 ||fk f ||2 .

Now ||fk ||2 is a Cauchy sequence and so it is bounded independent of k. Therefore, the above expression is
smaller than whenever k is large enough. This proves the lemma.
Corollary 13.13 For f, g L2 (Rn ),
Z
Z
f gdx =
Rn

F f F gdx =

Rn

F 1 f F 1 gdx.

Rn

Proof: First note the above formula is obvious if f, g S. To see this, note
Z
Z
Z
1
F f F gdx =
F f (x)
eixt g (t) dtdx
n/2
Rn
Rn
Rn
(2)
Z
Z
1
eixt F f (x) dxg (t)dt
=
n/2
Rn
Rn (2)
Z


=
F 1 F f (t) g (t)dt
n
ZR
=
f (t) g (t)dt.
Rn

13.2. FOURIER TRANSFORMS OF FUNCTIONS IN L2 RN

231

The formula with F 1 is exactly similar.

Now to verify the corollary, let k f in L2 (Rn ) and let k g in L2 (Rn ) . Then
Z
Z
F f F gdx = lim
F k F k dx
k Rn
Rn
Z
= lim
k k dx
k Rn
Z
=
f gdx
Rn

A similar argument holds for F 1 .This proves the corollary.

How do we compute F f and F 1 f ?
Theorem 13.14 For f L2 (Rn ), let fr = f XEr where Er is a bounded measurable set with Er Rn . Then
the following limits hold in L2 (Rn ) .
F f = lim F fr , F 1 f = lim F 1 fr .
r

Proof: ||f fr ||2 0 and so ||F f F fr ||2 0 and ||F 1 f F 1 fr ||2 0 by Plancherels Theorem.
This proves the theorem.
What are F fr and F 1 fr ? Let S
Z
Z
F fr dx =
fr F dx
Rn
Rn
Z Z
n
2
= (2)
fr (x)eixy (y)dydx
R n Rn
Z
Z
n
=
[(2) 2
fr (x)eixy dx](y)dy.
Rn

Rn

Since this holds for all S, a dense subset of L2 (Rn ), it follows that
Z
n
F fr (y) = (2) 2
fr (x)eixy dx.
Rn

Similarly
Z

F 1 fr (y) = (2) 2

fr (x)eixy dx.

Rn

This shows that to take

transform of a function in L2 (Rn ) , it suffices to take the limit as r
R the Fourier
2
n
n
ixy
in L (R ) of (2) 2 Rn fr (x)e
dx. A similar procedure works for the inverse Fourier transform.
Definition 13.15 For f L1 (Rn ) , define
n/2

F f (x) (2)

eixy f (y) dy,

Rn

n/2

F 1 f (x) (2)

eixy f (y) dy.

Rn

Thus, for f L1 (Rn ), F f and F 1 f are uniformly bounded.

232

FOURIER TRANSFORMS

Theorem 13.16 Let h L2 (Rn ) and let f L1 (Rn ). Then h f L2 (Rn ),

n/2

F 1 (h f ) = (2)

F 1 hF 1 f,

n/2

F (h f ) = (2)

F hF f,

and
||h f ||2 ||h||2 ||f ||1 .

(13.19)

Proof: Without loss of generality, we may assume h and f are both Borel measurable. Then an application of Minkowskis inequality yields
Z

Rn

Hence

Z

|h (x y)| |f (y)| dy

2

!1/2

dx

Rn

||f ||1 ||h||2 .

(13.20)

|h (x y)| |f (y)| dy < a.e. x and

x

h (x y) f (y) dy

is in L2 (Rn ). Let Er Rn , m (Er ) < . Thus,

hr XEr h L2 (Rn ) L1 (Rn ),
and letting S,
Z

F (hr f ) () dx

(hr f ) (F ) dx
Z Z Z
n/2
= (2)
hr (x y) f (y) eixt (t) dtdydx

Z Z Z
n/2
= (2)
hr (x y) ei(xy)t dx f (y) eiyt dy (t) dt
Z
n/2
=
(2)
F hr (t) F f (t) (t) dt.
Since is arbitrary and S is dense in L2 (Rn ) ,
n/2

F (hr f ) = (2)

F hr F f.

Now by Minkowskis Inequality, hr f h f in L2 (Rn ) and also it is clear that hr h in L2 (Rn ) ; so, by
Plancherels theorem, we may take the limit in the above and conclude
n/2

F (h f ) = (2)

F hF f.

The assertion for F 1 is similar and (13.19) follows from (13.20).

13.3. TEMPERED DISTRIBUTIONS

13.3

233

Tempered distributions

In this section we give an introduction to the general theory of Fourier transforms. Recall that S is the set
of all C (Rn ) such that for N = 1, 2, ,
N ()

sup
||N,xRn

1 + |x|

N

|D (x)| < .

The set, S is a vector space and we can make it into a topological space by letting sets of the form be a
basis for the topology.
B N (, r) { S such that N ( ) < r}.
Note the functions, N are increasing in N. Then S0 , the space of continuous linear functions defined on S
mapping S to C, are called tempered distributions. Thus,
Definition 13.17 f S0 means f : S C, f is linear, and f is continuous.
How can we verify f S0 ? The following lemma is about this question along with the similar question
of when a linear map from S to S is continuous.
Lemma 13.18 Let f : S C be linear and let L : S S be linear. Then f is continuous if and only if
|f ()| CN ()

(a.)

for some N . Also, L is continuous if and only if for each N, there exists M such that
N (L) CM ()

(b.)

for some C independent of .

Proof: It suffices to verify continuity at 0 because f and L are linear. We verify (b.). Let 0 U where
N
U is an open set. Then

0 B (0, r) U for some r > 0 and N . Then if M and C are as described in (b.),
M
1
and B
0, C r , we have
N (L) CM () < r;
so, this shows




B M 0, C 1 r L1 B N (0, r) L1 (U ),
which shows that L is continuous at 0. The argument for f and the only if part of the proof is left for the
reader.
The key to extending the Fourier transform to S0 is the following theorem which states that F and F 1
are continuous. This is analogous to the procedure in defining the Fourier transform for functions in L2 (Rn ).
Recall we proved these mappings preserved the L2 norms of functions in S.
Theorem 13.19 For F and F 1 the Fourier and inverse Fourier transforms,
Z
n/2
F (x) (2)
eixy (y) dy,
Rn

n/2

F 1 (x) (2)

Rn

F and F 1 are continuous linear maps from S to S.

eixy (y) dy,

234

FOURIER TRANSFORMS

Proof: Let || N where N > 0, and let x 6= 0. Then


N 

N Z

2
2
1



1 + |x|
D F (x) Cn 1 + |x|




y (y) dy .

ixy

Rn

(13.21)

Suppose x2j 1 for j {i1 , , ir } and x2j < 1 if j

/ {i1 , , ir }. Then after integrating by parts in the
integral of (13.21), we obtain the following for the right side of (13.21):


N Z
Y

2
ixy



Cn 1 + |x|
e
D
(y

(y))
dy
x2N
,
(13.22)
j
 n

R

j{i1 ,,ir }

where
2N

r
X

ei k ,

k=1

the vector with 0 in the jth slot if j

/ {i1 , , ir } and a 2N in the jth slot for j {i1 , , ir }. Now letting
C (n, N ) denote a generic constant depending only on n and N, the product rule and a little estimating
yields
nN


D (y (y)) 1 + |y|2
2N n () C (n, N ).

nN
2
is integrable. Therefore, the expression in (13.22) is no larger than
Also the function y 1 + |y|

N
2
C (n, N ) 1 + |x|

x2N
2N n ()
j

j{i1 ,,ir }

C (n, N ) 1 + n +

j{i1 ,,ir }

C (n, N )

2
|xj |

2N

|xj |

j{i1 ,,ir }

x2N
2N n ()
j

j{i1 ,,ir }

2N

|xj |

2N n ()

j{i1 ,,ir }

C (n, N ) 2N n ().
Therefore, if x2j 1 for some j,


1 + |x|

N 

D F 1 (x) C (n, N ) 2N n ().

If x2j < 1 for all j, we can use (13.21) to obtain



Z
N 

N 
1


1 + |x|
D F (x) Cn (1 + n) 
2



y (y) dy 

ixy

Rn

C (n, N ) R ()
for some R depending on N and n. Let M max (R, 2nN ) and we see


N F 1 CM ()
where M and C depend only on N and n. By Lemma 13.18, F 1 is continuous. Similarly, F is continuous.
This proves the theorem.

13.3. TEMPERED DISTRIBUTIONS

235

Definition 13.20 For f S0 , we define F f and F 1 f in S0 by

F f () f (F ), F 1 f () f F 1 .
To see this is a good definition, consider the following.
|F f ()| |f (F )| CN (F ) CM ().
Also note that F and F 1 are both one to one and onto. This follows from the fact that these mappings
map S one to one onto S.
What are some examples of things in S0 ? In answering this question, we will use the following lemma.
R
Lemma 13.21 If f L1loc (Rn ) and Rn f dx = 0 for all Cc (Rn ) , then f = 0 a.e.
Proof: It is enough to verify this for f 0. Let
E {x :f (x) r}, ER E B (0,R).
Let Kn be an increasing sequence of compact sets and let Vn be a decreasing sequence of open sets satisfying
Kn ER Vn , m (Vn \ Kn ) 2n , V1 is bounded.
Let
n Cc (Vn ) , Kn n Vn .
Then n (x) XER (x) a.e. because the set where n (x) fails to converge is contained in the set of all
x which are in infinitely many of the sets Vn \ Kn . This set has measure zero and so, by the dominated
convergence theorem,
Z
Z
Z
f n dx =
f dx rm (ER ).
f n dx = lim
0 = lim
n

Rn

V1

ER

Thus, m (ER ) = 0 and therefore m (E) = 0. Since r > 0 is arbitrary, it follows

m ([x :f (x) > 0]) = 0.
This proves the lemma.
Theorem 13.22 Let f be a measurable function with polynomial growth,

N
2
|f (x)| C 1 + |x|
for some N,
or let f Lp (Rn ) for some p [1, ]. Then f S0 if we define
Z
f () f dx.
Proof: Let f have polynomial growth first. Then
Z
Z 
nN
2
|f | || dx C
1 + |x|
|| dx
Z 
nN 
2nN
2
2
1 + |x|
1 + |x|
C
dx2nN ()
C (N, n) 2nN () < .

236

FOURIER TRANSFORMS

Therefore we can define

f ()

f dx

and it follows that

|f ()| C (N, n) 2nN ().
By Lemma 13.18, f S0 . Next suppose f Lp (Rn ).
Z
Z

M
2
|f | || dx |f | 1 + |x|
dxM ()
M

0
2
Lp . Then by Holders Inequality,
where we choose M large enough that 1 + |x|
|f ()| ||f ||p Cn M ().
By Lemma 13.18, f S0 . This proves the theorem.
Definition 13.23 If f S0 and S, then f S0 if we define
f () f ().
We need to verify that with this definition, f S0 . It is clearly linear. There exist constants C and N
such that
|f ()|

|f ()| CN ()

N
2
= C
sup
1 + |x|
|D ()|
xRn , ||N

C (, n, N ) N ().
Thus by Lemma 13.18, f S0 .
The next topic is that of convolution. This was discussed in Corollary 13.16 but we review it here. This
corollary implies that if f L2 (Rn ) S0 and S, then
f L2 (Rn ), ||f ||2 ||f ||2 ||||1 ,
and also
n/2

F (f ) (x) = F (x) F f (x) (2)

(13.23)
n/2

F 1 (f ) (x) = F 1 (x) F 1 f (x) (2)

By Definition 13.23,
n/2

F (f ) = (2)

F F f

whenever f L2 (Rn ) and S.

Now it is easy to see the proper way to define f when f is only in S0 and S.
Definition 13.24 Let f S0 and let S. Then we define
n/2

f (2)

F 1 (F F f ) .

13.3. TEMPERED DISTRIBUTIONS

237

Theorem 13.25 Let f S0 and let S.

n/2

F (f ) = (2)

n/2

F 1 (f ) = (2)

F F f,

(13.24)

F 1 F 1 f.

Proof: Note that (13.24) follows from Definition 13.24 and both assertions hold for f S. Next we
write for S,


F 1 F 1 (x)
=

Z Z Z

( F F ) (x) .

iyy1 iy1 z

(x y) e

(z) dzdy1 dy (2)

Z Z Z

n
iy
y1 i
y1 z
=
(x y) e
e
(z) dzd
y1 dy (2)

Now for S,
n/2

n/2
F F 1 F 1 f () (2)
F 1 F 1 f (F )

(2)

n/2

(2)

n/2
F 1 f F 1 F (2)
f F 1 F 1 F =

n/2 1
f (2)
F


F F 1 F 1 (F )

f F 1 F 1 = f ( F F )
n/2

(2)

n/2

F 1 (F F f ) () (2)

n/2

(2)

(F F f ) F 1

n/2
F f F F 1 (2)
f F F F 1 =

by (13.23),
 


n/2
= f F (2)
F F 1



= f F F 1 (F F ) = f ( F F ) .
Comparing the above shows
n/2

(2)

n/2 1
F F 1 F 1 f = (2)
F (F F f ) f

and ; so,
n/2

(2)
which proves the theorem.

F 1 F 1 f = F 1 (f )

238

FOURIER TRANSFORMS

13.4

Exercises

1. Suppose f L1 (Rn ) and ||gk f ||1 0. Show F gk and F 1 gk converge uniformly to F f and F 1 f
respectively.
2. Let f L1 (Rn ),
F f (t) (2)n/2

eitx f (x)dx,

Rn

F 1 f (t) (2)n/2

eitx f (x)dx.

Rn

Show that F 1 f and F f are both continuous and bounded. Show also that
lim F 1 f (x) = lim F f (x) = 0.

|x|

|x|

Are the Fourier and inverse Fourier transforms of a function in L1 (Rn ) uniformly continuous?
3. Suppose F 1 f L1 (Rn ). Observe that just as in Theorem 13.5,
(F F 1 )f (x) = lim f m (x).
0

Use this to argue that if f and F 1 f L1 (Rn ), then

F F 1 f (x) = f (x) a.e. x.
Similarly
(F 1 F )f (x) = f (x) a.e.
if f and F f L1 (Rn ). Hint: Show f m f in L1 (Rn ). Thus there is a subsequence k 0 such
that f mk (x) f (x) a.e.
4. Show that if F 1 f L1 or F f L1 , then f equals a continuous bounded function a.e.
n/2

5. Let f, g L1 (Rn ). Show f g L1 and F (f g) = (2)

F f F g.

6. Suppose f, g L1 (R) and F f = F g. Show f = g a.e.

7. Suppose f f = f or f f = 0 and f L1 (R). Show f = 0.
R
Rr
8. 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
(a)
(b)

sin(u)

u du = 2
R
limr sin(ru)
du
u
1
0

= 0 whenever > 0.
R
(c) If f L (R) , then limr R sin (ru) f (u) du = 0.

R
RR
R
Hint: For the first two, use u1 = 0 eut dt and apply Fubinis theorem to 0 sin u R eut 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 sometimes called the Riemann Lebesgue lemma.

13.4. EXERCISES

239

9. Suppose that g L1 (R) and that at some x > 0 we have that g is locally Holder continuous from the
right and from the left. By this we mean
lim g (x + r) g (x+)

r0+

exists,
lim g (x r) g (x)

r0+

exists and there exist constants K, > 0 and r (0, 1] such that for |x y| < ,
r

|g (x+) g (y)| < K |x y|

for y > x and

|g (x) g (y)| < K |x y|

for y < x. Show that under these conditions,



Z
2 sin (ur) g (x u) + g (x + u)
lim
du
r 0
u
2
=

g (x+) + g (x)
.
2

10. 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
eity g (y) dydt =
.
R 2 R
2

This is very interesting. If g L2 (R), this shows F 1 (F g) (x) = g(x+)+g(x)

, the midpoint of the
2
jump in g at the point, x. In particular, if g S, F 1 (F g) = g. Hint: Show the left side of the above
equation reduces to


Z
2 sin (ur) g (x u) + g (x + u)
du
0
u
2
and then use Problem 9 to obtain the result.
11. We say a measurable function g defined on (0, ) has exponential growth if |g (t)| Cet for some
. For Re (s) > , we can define the Laplace Transform by
Z
Lg (s)
esu g (u) du.
0

Assume that g has exponential growth as above and is Holder continuous from the right and from the
left at t. Pick > . Show that
Z R
1
g (t+) + g (t)
lim
et eiyt Lg ( + iy) dy =
.
R 2 R
2
This formula is sometimes written in the form
Z +i
1
est Lg (s) ds
2i i

240

FOURIER TRANSFORMS
and is called the complex inversion integral for Laplace transforms. It can be used to find inverse
Laplace transforms. Hint:
Z R
1
et eiyt Lg ( + iy) dy =
2 R
1
2

et eiyt

e(+iy)u g (u) dudy.

Now use Fubinis theorem and do the integral from R to R to get this equal to
Z
sin (R (t u))
et u
e
g (u)
du

tu
where g is the zero extension of g off [0, ). Then this equals
Z
et (tu)
sin (Ru)
e
g (t u)
du

u
which equals
2et

Z
0

g (t u) e(tu) + g (t + u) e(t+u) sin (Ru)

du
2
u

and then apply the result of Problem 9.

12. Several times in the above chapter we used an argument like the following. Suppose
0 for all S. Therefore, f = 0 in L2 (Rn ) . Prove the validity of this argument.

Rn

f (x) (x) dx =

13. Suppose f S, fxj L1 (Rn ). Show F (fxj )(t) = itj F f (t).

14. Let f S and let k be a positive integer.
||f ||k,2 (||f ||22 +

||D f ||22 )1/2.

||k

One could also define

Z
|||f |||k,2 (

|F f (x)|2 (1 + |x|2 )k dx)1/2.

Rn

Show both || ||k,2 and ||| |||k,2 are norms on S and that they are equivalent. These are Sobolev space
norms. What are some possible advantages of the second norm over the first? Hint: For which values
of k do these make sense?
15. Define H k (Rn ) by f L2 (Rn ) such that
Z
1
( |F f (x)|2 (1 + |x|2 )k dx) 2 < ,

|||f |||k,2

Z
1
( |F f (x)|2 (1 + |x|2 )k dx) 2.

Show H k (Rn ) is a Banach space, and that if k is a positive integer, H k (Rn ) ={ f L2 (Rn ) : there
exists {uj } S with ||uj f ||2 0 and {uj } is a Cauchy sequence in || ||k,2 of Problem 14}.
This is one way to define Sobolev Spaces. Hint: One way to do the second part of this is to let
gs F f in L2 ((1 + |x|2 )k dx) where gs Cc (Rn ). We can do this because (1 + |x|2 )k dx is a Radon
measure. By convolving with a mollifier, we can, without loss of generality, assume gs Cc (Rn ).
Thus gs = F fs , fs S. Then by Problem 14, fs is Cauchy in the norm || ||k,2 .

13.4. EXERCISES

241

16. If 2k > n, show that if f H k (Rn ), then f equals a bounded continuous function a.e. Hint: Show
F f L1 (Rn ), and then use Problem 4. To do this, write
k

|F f (x)| = |F f (x)|(1 + |x|2 ) 2 (1 + |x|2 )

k
2

So
Z

|F f (x)|dx =

|F f (x)|(1 + |x|2 ) 2 (1 + |x|2 )

k
2

dx.

Use Holders Inequality. This is an example of a Sobolev Embedding Theorem.

17. Let u S. Then we know F u S and so, in particular, it makes sense to form the integral,
Z
F u (x0 , xn ) dxn
R
0

where (x , xn ) = x R . For u S, define u (x0 ) u (x0 , 0) . Find a constant such that F (u) (x0 )
equals this constant times the above integral. Hint: By the dominated convergence theorem
Z
Z
2
0
F u (x , xn ) dxn = lim
e(xn ) F u (x0 , xn ) dxn .
n

Now use the definition of the Fourier transform and Fubinis theorem as required in order to obtain
the desired relationship.
18. Recall from the chapter on Fourier series that the Fourier series of a function in L2 (, ) con2
verges to the function in the mean square
sense. Prove
n
o a similar theorem with L (, ) replaced
(1/2) inx
2
by L (m, m) and the functions (2)
e
used in the Fourier series replaced with
nZ
n
o
(1/2) i n x
. Now suppose f is a function in L2 (R) satisfying F f (t) = 0 if |t| > m.
(2m)
e m
nZ

Show that if this is so, then



1X
n sin ( (mx + n))
f (x) =
f
.

m
mx +
nZ

Here m is a positive integer. This is sometimes called the Shannon sampling theorem.

242

FOURIER TRANSFORMS

Banach Spaces
14.1

Baire category theorem

Functional analysis is the study of various types of vector spaces which are also topological spaces and the
linear operators defined on these spaces. As such, it is really a generalization of linear algebra and calculus.
The vector spaces which are of interest in this subject include the usual spaces Rn and Cn but also many
which are infinite dimensional such as the space C (X; Rn ) discussed in Chapter 4 in which we think of a
function as a point or a vector. When the topology comes from a norm, the vector space is called a normed
linear space and this is the case of interest here. A normed linear space is called real if the field of scalars is
R and complex if the field of scalars is C. We will assume a linear space is complex unless stated otherwise.
A normed linear space may be considered as a metric space if we define d (x, y) ||x y||. As usual, if every
Cauchy sequence converges, the metric space is called complete.
Definition 14.1 A complete normed linear space is called a Banach space.
The purpose of this chapter is to prove some of the most important theorems about Banach spaces. The
next theorem is called the Baire category theorem and it will be used in the proofs of many of the other
theorems.
Theorem 14.2 Let (X, d) be a complete metric space and let {Un }
n=1 be a sequence of open subsets of X
satisfying Un = X (Un is dense). Then D
n=1 Un is a dense subset of X.
Proof: Let p X and let r0 > 0. We need to show D B(p, r0 ) 6= . Since U1 is dense, there exists
p1 U1 B(p, r0 ), an open set. Let p1 B(p1 , r1 ) B(p1 , r1 ) U1 B(p, r0 ) and r1 < 21 . We are using
Theorem 3.14.

 r0 p
p 1
There exists p2 U2 B(p1 , r1 ) because U2 is dense. Let
p2 B(p2 , r2 ) B(p2 , r2 ) U2 B(p1 , r1 ) U1 U2 B(p, r0 ).
and let r2 < 22 . Continue in this way. Thus
rn < 2n,
B(pn , rn ) U1 U2 ... Un B(p, r0 ),
243

244

BANACH SPACES
B(pn , rn ) B(pn1 , rn1 ).

Consider the Cauchy sequence, {pn }. Since X is complete, let

lim pn = p .

Since all but finitely many terms of {pn } are in B(pm , rm ), it follows that p B(pm , rm ). Since this holds
for every m,

p
m=1 B(pm , rm ) i=1 Ui B(p, r0 ).

This proves the theorem.

Corollary 14.3 Let X be a complete metric space and suppose X =
i=1 Fi where each Fi is a closed set.
Then for some i, interior Fi 6= .
The set D of Theorem 14.2 is called a G set because it is the countable intersection of open sets. Thus
D is a dense G set.
Recall that a norm satisfies:
a.) ||x|| 0, ||x|| = 0 if and only if x = 0.
b.) ||x + y|| ||x|| + ||y||.
c.) ||cx|| = |c| ||x|| if c is a scalar and x X.
We also recall the following lemma which gives a simple way to tell if a function mapping a metric space
to a metric space is continuous.
Lemma 14.4 If (X, d), (Y, p) are metric spaces, f is continuous at x if and only if
lim xn = x

implies
lim f (xn ) = f (x).

The proof is left to the reader and follows quickly from the definition of continuity. See Problem 5 of
Chapter 3. For the sake of simplicity, we will write xn x sometimes instead of limn xn = x.
Theorem 14.5 Let X and Y be two normed linear spaces and let L : X Y be linear (L(ax + by) =
aL(x) + bL(y) for a, b scalars and x, y X). The following are equivalent
a.) L is continuous at 0
b.) L is continuous
c.) There exists K > 0 such that ||Lx||Y K ||x||X for all x X (L is bounded).
Proof: a.)b.) Let xn x. Then (xn x) 0. It follows Lxn Lx 0 so Lxn Lx. b.)c.)
Since L is continuous, L is continuous at 0. Hence ||Lx||Y < 1 whenever ||x||X for some . Therefore,
suppressing the subscript on the || ||,


x
||L
|| 1.
||x||
Hence
||Lx||
c.)a.) is obvious.

1
||x||.

14.1. BAIRE CATEGORY THEOREM

245

Definition 14.6 Let L : X Y be linear and continuous where X and Y are normed linear spaces. We
denote the set of all such continuous linear maps by L(X, Y ) and define
||L|| = sup{||Lx|| : ||x|| 1}.

(14.1)

The proof of the next lemma is left to the reader.

Lemma 14.7 With ||L|| defined in (14.1), L(X, Y ) is a normed linear space. Also ||Lx|| ||L|| ||x||.
For example, we could consider the space of linear transformations defined on Rn having values in Rm ,
and the above gives a way to measure the distance between two linear transformations. In this case, the
n
linear transformations are all continuous because if L is such a linear transformation, and {ek }k=1 and
m
n
m
{ei }i=1 are the standard basis vectors in R and R respectively, there are scalars lik such that
L (ek ) =

m
X

lik ei .

i=1

Thus, letting a =

Pn

k=1

ak ek ,
L (a) = L

n
X

ak ek

k=1

n
X

ak L (ek ) =

k=1

n X
m
X

ak lik ei .

k=1 i=1

Consequently, letting K |lik | for all i, k,

||La||


2 1/2
m X
n

X



ak lik 
Km1/2 n (max {|ak | , k = 1, , n})



i=1 k=1
1/2

Km

n
X

a2k

!1/2

= Km1/2 n ||a||.

k=1

This type of thing occurs whenever one is dealing with a linear transformation between finite dimensional
normed linear spaces. Thus, in finite dimensions the algebraic condition that an operator is linear is sufficient
to imply the topological condition that the operator is continuous. The situation is not so simple in infinite
dimensional spaces such as C (X; Rn ). This is why we impose the topological condition of continuity as a
criterion for membership in L (X, Y ) in addition to the algebraic condition of linearity.
Theorem 14.8 If Y is a Banach space, then L(X, Y ) is also a Banach space.
Proof: Let {Ln } be a Cauchy sequence in L(X, Y ) and let x X.
||Ln x Lm x|| ||x|| ||Ln Lm ||.
Thus {Ln x} is a Cauchy sequence. Let
Lx = lim Ln x.
n

Then, clearly, L is linear. Also L is continuous. To see this, note that {||Ln ||} is a Cauchy sequence of real
numbers because |||Ln || ||Lm ||| ||Ln Lm ||. Hence there exists K > sup{||Ln || : n N}. Thus, if x X,
||Lx|| = lim ||Ln x|| K||x||.
n

This proves the theorem.

246

14.2

BANACH SPACES

Uniform boundedness closed graph and open mapping theorems

The next big result is sometimes called the Uniform Boundedness theorem, or the Banach-Steinhaus theorem.
This is a very surprising theorem which implies that for a collection of bounded linear operators, if they are
bounded pointwise, then they are also bounded uniformly. As an example of a situation in which pointwise
bounded does not imply uniformly bounded, consider the functions f (x) X(,1) (x) x1 for (0, 1) and
X(,1) (x) equals zero if x
/ (, 1) and one if x (, 1) . Clearly each function is bounded and the collection
of functions is bounded at each point of (0, 1), but there is no bound for all the functions taken together.
Theorem 14.9 Let X be a Banach space and let Y be a normed linear space. Let {L } be a collection
of elements of L(X, Y ). Then one of the following happens.
a.) sup{||L || : } <
b.) There exists a dense G set, D, such that for all x D,
sup{||L x|| } = .
Proof: For each n N, define
Un = {x X : sup{||L x|| : } > n}.
Then Un is an open set. Case b.) is obtained from Theorem 14.2 if each Un is dense. The other case is that for
some n, Un is not dense. If this occurs, there exists x0 and r > 0 such that for all x B(x0 , r), ||L x|| n
for all . Now if y B(0, r), x0 + y B(x0 , r). Consequently, for all such y, ||L (x0 + y)|| n. This
implies that for such y and all ,
||L y|| n + ||L (x0 )|| 2n.
Hence ||L || 2n
r for all , and we obtain case a.).
The next theorem is called the Open Mapping theorem. Unlike Theorem 14.9 it requires both X and Y
to be Banach spaces.
Theorem 14.10 Let X and Y be Banach spaces, let L L(X, Y ), and suppose L is onto. Then L maps
open sets onto open sets.
To aid in the proof of this important theorem, we give a lemma.
Lemma 14.11 Let a and b be positive constants and suppose
B(0, a) L(B(0, b)).
Then
L(B(0, b)) L(B(0, 2b)).
Proof of Lemma 14.11: Let y L(B(0, b)). Pick x1 B(0, b) such that ||y Lx1 || < a2 . Now
2y 2Lx1 B(0, a) L(B(0, b)).
Then choose x2 B(0, b) such that ||2y 2Lx1 Lx2 || < a/2. Thus ||y Lx1 L
in this way, we pick x3 , x4 , ... in B(0, b) such that
||y

n
X
i=1

2(i1) L(xi )|| = ||y L

n
X
i=1

2(i1) xi || < a2n.

x2
2

|| < a/22 . Continuing

(14.2)

14.2. UNIFORM BOUNDEDNESS CLOSED GRAPH AND OPEN MAPPING THEOREMS

Let x =

i=1

247

2(i1) xi . The series converges because X is complete and

||

n
X

2(i1) xi || b

2(i1) = b 2m+2 .

i=m

i=m

Thus the sequence of partial sums is Cauchy. Letting n in (14.2) yields ||y Lx|| = 0. Now
||x|| = lim ||
n

lim

n
X
i=1

n
X

2(i1) xi ||

i=1

2(i1) ||xi || < lim

n
X

2(i1) b = 2b.

i=1

This proves the lemma.

Proof of Theorem 14.10: Y =
n=1 L(B(0, n)). By Corollary 14.3, the set, L(B(0, n0 )) has nonempty
interior for some n0 . Thus B(y, r) L(B(0, n0 )) for some y and some r > 0. Since L is linear B(y, r)
L(B(0, n0 )) also (why?). Therefore
B(0, r) B(y, r) + B(y, r)
{x + z : x B (y, r) and z B (y, r)}
L(B(0, 2n0 ))
By Lemma 14.11, L(B(0, 2n0 )) L(B(0, 4n0 )). Letting a = r(4n0 )1 , it follows, since L is linear, that
B(0, a) L(B(0, 1)).
Now let U be open in X and let x + B(0, r) = B(x, r) U . Then
L(U ) L(x + B(0, r))
= Lx + L(B(0, r)) Lx + B(0, ar) = B(Lx, ar)
(L(B(0, r)) B(0, ar) because L(B(0, 1)) B(0, a) and L is linear). Hence
Lx B(Lx, ar) L(U ).
This shows that every point, Lx LU , is an interior point of LU and so LU is open. This proves the
theorem.
This theorem is surprising because it implies that if || and |||| are two norms with respect to which a
vector space X is a Banach space such that || K ||||, then there exists a constant k, such that |||| k || .
This can be useful because sometimes it is not clear how to compute k when all that is needed is its existence.
To see the open mapping theorem implies this, consider the identity map ix = x. Then i : (X, ||||) (X, ||)
is continuous and onto. Hence i is an open map which implies i1 is continuous. This gives the existence of
the constant k. Of course there are many other situations where this theorem will be of use.
Definition 14.12 Let f : D E. The set of all ordered pairs of the form {(x, f (x)) : x D} is called the
graph of f .
Definition 14.13 If X and Y are normed linear spaces, we make X Y into a normed linear space by
using the norm ||(x, y)|| = ||x|| + ||y|| along with component-wise addition and scalar multiplication. Thus
a(x, y) + b(z, w) (ax + bz, ay + bw).
There are other ways to give a norm for X Y . See Problem 5 for some alternatives.

248

BANACH SPACES

Lemma 14.14 The norm defined in Definition 14.13 on X Y along with the definition of addition and
scalar multiplication given there make X Y into a normed linear space. Furthermore, the topology induced
by this norm is identical to the product topology defined in Chapter 3.
Lemma 14.15 If X and Y are Banach spaces, then X Y with the norm and vector space operations
defined in Definition 14.13 is also a Banach space.
Lemma 14.16 Every closed subspace of a Banach space is a Banach space.
Definition 14.17 Let X and Y be Banach spaces and let D X be a subspace. A linear map L : D Y is
said to be closed if its graph is a closed subspace of X Y . Equivalently, L is closed if xn x and Lxn y
implies x D and y = Lx.
Note the distinction between closed and continuous. If the operator is closed the assertion that y = Lx
only follows if it is known that the sequence {Lxn } converges. In the case of a continuous operator, the
convergence of {Lxn } follows from the assumption that xn x. It is not always the case that a mapping
which is closed is necessarily continuous. Consider the function f (x) = tan (x) if x is not an odd multiple of

2 and f (x) 0 at every odd multiple of 2 . Then the graph is closed and the function is defined on R but
it clearly fails to be continuous. The next theorem, the closed graph theorem, gives conditions under which
closed implies continuous.
Theorem 14.18 Let X and Y be Banach spaces and suppose L : X Y is closed and linear. Then L is
continuous.
Proof: Let G be the graph of L. G = {(x, Lx) : x X}. Define P : G X by P (x, Lx) = x. P
maps the Banach space G onto the Banach space X and is continuous and linear. By the open mapping
theorem, P maps open sets onto open sets. Since P is also 1-1, this says that P 1 is continuous. Thus
||P 1 x|| K||x||. Hence
||x|| + ||Lx|| K||x||
and so ||Lx|| (K 1)||x||. This shows L is continuous and proves the theorem.

14.3

Hahn Banach theorem

The closed graph, open mapping, and uniform boundedness theorems are the three major topological theorems in functional analysis. The other major theorem is the Hahn-Banach theorem which has nothing to do
with topology. Before presenting this theorem, we need some preliminaries.
Definition 14.19 Let F be a nonempty set. F is called a partially ordered set if there is a relation, denoted
here by , such that
x x for all x F.
If x y and y z then x z.
C F is said to be a chain if every two elements of C are related. By this we mean that if x, y C, then
either x y or y x. Sometimes we call a chain a totally ordered set. C is said to be a maximal chain if
whenever D is a chain containing C, D = C.
The most common example of a partially ordered set is the power set of a given set with being the
relation. The following theorem is equivalent to the axiom of choice. For a discussion of this, see the appendix
on the subject.

14.3. HAHN BANACH THEOREM

249

Theorem 14.20 (Hausdorff Maximal Principle) Let F be a nonempty partially ordered set. Then there
exists a maximal chain.
Definition 14.21 Let X be a real vector space : X R is called a gauge function if
(x + y) (x) + (y),
(ax) = a(x) if a 0.

(14.3)

Suppose M is a subspace of X and z

/ M . Suppose also that f is a linear real-valued function having the
property that f (x) (x) for all x M . We want to consider the problem of extending f to M Rz such
that if F is the extended function, F (y) (y) for all y M Rz and F is linear. Since F is to be linear,
we see that we only need to determine how to define F (z). Letting a > 0, we need to have the following
hold for all x, y M .
F (x + az) (x + az), F (y az) (y az).
Multiplying by a1 using the fact that M is a subspace, and (14.3), we see this is the same as
f (x) + F (z) (x + z), f (y) (y z) F (z)
for all x, y M . Hence we need to have F (z) such that for all x, y M
f (y) (y z) F (z) (x + z) f (x).

(14.4)

Is there any such number between f (y) (y z) and (x + z) f (x) for every pair x, y M ? This is where
we use that f (x) (x) on M . For x, y M ,
(x + z) f (x) [f (y) (y z)]
= (x + z) + (y z) (f (x) + f (y))
(x + y) f (x + y) 0.
Therefore there exists a number between
sup {f (y) (y z) : y M }
and
inf {(x + z) f (x) : x M }
We choose F (z) to satisfy (14.4). With this preparation, we state a simple lemma which will be used to
prove the Hahn Banach theorem.
Lemma 14.22 Let M be a subspace of X, a real linear space, and let be a gauge function on X. Suppose
f : M R is linear and z
/ M, and f (x) (x) for all x M . Then f can be extended to M Rz such
that, if F is the extended function, F is linear and F (x) (x) for all x M Rz.
Proof: Let f (y) (y z) F (z) (x + z) f (x) for all x, y M and let F (x + az) = f (x) + aF (z)
whenever x M, a R. If a > 0
F (x + az)

= f (x) + aF (z)

 x i
h x
+z f
f (x) + a
a
a
= (x + az).

250

BANACH SPACES

If a < 0,
F (x + az)

= f (x) + aF (z)
 



x
x
f (x) + a f

z
a
a
= f (x) f (x) + (x + az) = (x + az).

This proves the lemma.

Theorem 14.23 (Hahn Banach theorem) Let X be a real vector space, let M be a subspace of X, let
f : M R be linear, let be a gauge function on X, and suppose f (x) (x) for all x M . Then there
exists a linear function, F : X R, such that
a.) F (x) = f (x) for all x M
b.) F (x) (x) for all x X.
Proof: Let F = {(V, g) : V M, V is a subspace of X, g : V R is linear, g(x) = f (x) for all x M ,
and g(x) (x)}. Then (M, f ) F so F =
6 . Define a partial order by the following rule.
(V, g) (W, h)
means
V W and h(x) = g(x) if x V.
Let C be a maximal chain in F (Hausdorff Maximal theorem). Let Y = {V : (V, g) C}. Let h : Y R
be defined by h(x) = g(x) where x V and (V, g) C. This is well defined since C is a chain. Also h is
clearly linear and h(x) (x) for all x Y . We want to argue that Y = X. If not, there exists z X \ Y
and we can extend

h to Y Rz using Lemma 14.22. But this will contradict the maximality of C. Indeed,
C { Y Rz, h } would be a longer chain where h is the extended h. This proves the Hahn Banach theorem.
This is the original version of the theorem. There is also a version of this theorem for complex vector
spaces which is based on a trick.
Corollary 14.24 (Hahn Banach) Let M be a subspace of a complex normed linear space, X, and suppose
f : M C is linear and satisfies |f (x)| K||x|| for all x M . Then there exists a linear function, F ,
defined on all of X such that F (x) = f (x) for all x M and |F (x)| K||x|| for all x.
Proof: First note f (x) = Re f (x) + i Im f (x) and so
Re f (ix) + i Im f (ix) = f (ix) = i f (x) = i Re f (x) Im f (x).
Therefore, Im f (x) = Re f (ix), and we may write
f (x) = Re f (x) i Re f (ix).
If c is a real scalar
Re f (cx) i Re f (icx) = cf (x) = c Re f (x) i c Re f (ix).
Thus Re f (cx) = c Re f (x). It is also clear that Re f (x + y) = Re f (x) + Re f (y). Consider X as a real vector
space and let (x) = K||x||. Then for all x M ,
| Re f (x)| K||x|| = (x).

14.3. HAHN BANACH THEOREM

251

From Theorem 14.23, Re f may be extended to a function, h which satisfies

h(ax + by) = ah(x) + bh(y) if a, b R
|h(x)| K||x|| for all x X.
Let
F (x) h(x) i h(ix).
It is routine to show F is linear. Now wF (x) = |F (x)| for some |w| = 1. Therefore
|F (x)| = wF (x) = h(wx) i h(iwx) = h(wx)
= |h(wx)| K||x||.
This proves the corollary.
Definition 14.25 Let X be a Banach space. We denote by X 0 the space L(X, C). By Theorem 14.8, X 0 is
a Banach space. Remember
||f || = sup{|f (x)| : ||x|| 1}
for f X 0 . We call X 0 the dual space.
Definition 14.26 Let X and Y be Banach spaces and suppose L L(X, Y ). Then we define the adjoint
map in L(Y 0 , X 0 ), denoted by L , by
L y (x) y (Lx)
for all y Y 0 .
X

Y0
Y

In terms of linear algebra, this adjoint map is algebraically very similar to, and is in fact a generalization
of, the transpose of a matrix considered as a map on Rn . Recall that if A is such a matrix, AT satisfies
AT x y = xAy. In the case of Cn the adjoint is similar to the conjugate transpose of the matrix and it
behaves the same with respect to the complex inner product on Cn . What is being done here is to generalize
this algebraic concept to arbitrary Banach spaces.
Theorem 14.27 Let L L(X, Y ) where X and Y are Banach spaces. Then
a.) L L(Y 0 , X 0 ) as claimed and ||L || ||L||.
b.) If L is 1-1 onto a closed subspace of Y , then L is onto.
c.) If L is onto a dense subset of Y , then L is 1-1.
Proof: Clearly L y is linear and L is also a linear map.
||L || =
=

sup ||L y || = sup

||y ||1

sup
||y ||1

Hence, ||L || ||L|| and this shows part a.).

sup |L y (x)|

||y ||1 ||x||1

sup |y (Lx)| sup |(Lx)| = ||L||

||x||1

||x||1

252

BANACH SPACES

If L is 1-1 and onto a closed subset of Y , then we can apply the Open Mapping theorem to conclude that
L1 : L(X) X is continuous. Hence
||x|| = ||L1 Lx|| K||Lx||
for some K. Now let x X 0 be given. Define f L(L(X), C) by f (Lx) = x (x). Since L is 1-1, it follows
that f is linear and well defined. Also
|f (Lx)| = |x (x)| ||x || ||x|| K||x || ||Lx||.
By the Hahn Banach theorem, we can extend f to an element y Y 0 such that ||y || K||x ||. Then
L y (x) = y (Lx) = f (Lx) = x (x)
so L y = x and we have shown L is onto. This shows b.).
Now suppose LX is dense in Y . If L y = 0, then y (Lx) = 0 for all x. Since LX is dense, this can only
happen if y = 0. Hence L is 1-1.
Corollary 14.28 Suppose X and Y are Banach spaces, L L(X, Y ), and L is 1-1 and onto. Then L is
also 1-1 and onto.
There exists a natural mapping from a normed linear space, X, to the dual of the dual space.
Definition 14.29 Define J : X X 00 by J(x)(x ) = x (x). This map is called the James map.
Theorem 14.30 The map, J, has the following properties.
a.) J is 1-1 and linear.
b.) ||Jx|| = ||x|| and ||J|| = 1.
c.) J(X) is a closed subspace of X 00 if X is complete.
Also if x X 0 ,
||x || = sup {|x (x )| : ||x || 1, x X 00 } .
Proof: To prove this, we will use a simple but useful lemma which depends on the Hahn Banach theorem.
Lemma 14.31 Let X be a normed linear space and let x X. Then there exists x X 0 such that ||x || = 1
and x (x) = ||x||.
Proof: Let f : Cx C be defined by f (x) = ||x||. Then for y Cx, |f (y)| kyk. By the Hahn
Banach theorem, there exists x X 0 such that x (x) = f (x) and ||x || 1. Since x (x) = ||x|| it follows
that ||x || = 1. This proves the lemma.
Now we prove the theorem. It is obvious that J is linear. If Jx = 0, then let x (x) = ||x|| with ||x || = 1.
0 = J(x)(x ) = x (x) = ||x||.
This shows a.). To show b.), let x X and x (x) = ||x|| with ||x || = 1. Then
||x|| sup{|y (x)| : ||y || 1} = sup{|J(x)(y )| : ||y || 1} = ||Jx||
|J(x)(x )| = |x (x)| = ||x||
||J|| = sup{||Jx|| : ||x|| 1} = sup{||x|| : ||x|| 1} = 1.
This shows b.). To verify c.), use b.). If Jxn y X 00 then by b.), xn is a Cauchy sequence converging
to some x X. Then Jx = limn Jxn = y .

14.4. EXERCISES

253

Finally, to show the assertion about the norm of x , use what was just shown applied to the James map
from X 0 to X 000 . More specifically,
||x || = sup {|x (x)| : ||x|| 1} = sup {|J (x) (x )| : ||Jx|| 1}
sup {|x (x )| : ||x || 1} = sup {|J (x ) (x )| : ||x || 1}
||Jx || = ||x ||.
This proves the theorem.
Definition 14.32 When J maps X onto X 00 , we say that X is Reflexive.
Later on we will give examples of reflexive spaces. In particular, it will be shown that the space of square
integrable and pth power integrable functions for p > 1 are reflexive.

14.4

Exercises

1. Show that no countable dense subset of R is a G set. In particular, the rational numbers are not a
G set.
2. Let f : R C be a function. Let r f (x) = sup{|f (z) f (y)| : y, z B(x, r)}. Let f (x) =
limr0 r f (x). Show f is continuous at x if and only if f (x) = 0. Then show the set of points where
f is continuous is a G set (try Un = {x : f (x) < n1 }). Does there exist a function continuous at only
the rational numbers? Does there exist a function continuous at every irrational and discontinuous

elsewhere? Hint: Suppose D is any countable set, D = {di }i=1 , and define the function,
P fn (x) to
equal zero for every x
/ {d1 , , dn } and 2n for x in this finite set. Then consider g (x) n=1 fn (x) .
Show that this series converges uniformly.
3. Let f C([0, 1]) and suppose f 0 (x) exists. Show there exists a constant, K, such that |f (x) f (y)|
K|x y| for all y [0, 1]. Let Un = {f C([0, 1]) such that for each x [0, 1] there exists y [0, 1]
such that |f (x)f (y)| > n|xy|}. Show that Un is open and dense in C([0, 1]) where for f C ([0, 1]),
||f || sup {|f (x)| : x [0, 1]} .
Show that if f C([0, 1]), there exists g C([0, 1]) such that ||g f || < but g 0 (x) does not exist for
any x [0, 1].
4. Let X be a normed linear space and suppose A X is weakly bounded. This means that for each
x X 0 , sup{|x (x)| : x A} < . Show A is bounded. That is, show sup{||x|| : x A} < .
5. Let X and Y be two Banach spaces. Define the norm
|||(x, y)||| max (||x||X , ||y||Y ).
Show this is a norm on X Y which is equivalent to the norm given in the chapter for X Y . Can
you do the same for the norm defined by
1/2

2
2
?
|(x, y)| ||x||X + ||y||Y
6. Prove Lemmas 14.14 - 14.16.

254

BANACH SPACES

7. Let f : R C be continuous and periodic with

P period 2. That is f (x + 2) = f (x) for all x. Then it
is not unreasonable to try to write f (x) = n= an einx . Find what an should be. Hint: Multiply
R
R P PR
both sides by eimx and do . Pretend there is no problem writing
=
. Recall the series
which results is called a Fourier series.
8. If you did 7 correctly, you found
an =

Z


f (x)einx dx (2)1.

th

The n partial sum will be denoted by Sn f and defined by Sn f (x) =

R
f (y)Dn (x y)dy where

Dn (t) =

Pn

k=n

ak eikx . Show Sn f (x) =

sin((n + 12 )t)
.
2 sin( 2t )

This is called the Dirichlet kernel. If you have trouble, review the chapter on Fourier series.
9. Let Y = {f such that f is continuous, defined on R, and 2 periodic}. Define ||f ||Y = sup{|f (x)| :
x [, ]}. Show that (Y, || ||Y ) is a Banach space. Let x R and define Ln (f ) = Sn f (x). Show
Ln Y 0 but limn ||Ln || = . Hint: Let f (y) approximate sign(Dn (x y)).
10. Show there exists a dense G subset of Y such that for f in this set, |Sn f (x)| is unbounded. Show
there is a dense G subset of Y having the property that |Sn f (x)| is unbounded on a dense G subset
of R. This shows Fourier series can fail to converge pointwise to continuous periodic functions in a
fairly spectacular way. Hint: First show there is a dense G subset of Y, G, such that for all f G,
we have sup {|Sn f (x)| : n N} = for all x Q. Of course Q is not a G set but this is still pretty
impressive. Next consider Hk {(x, f ) R Y : supn |Sn f (x)| > k} and argue that Hk is open and
dense. Next let Hk1 {x R : for some f Y, (x, f ) Hk }and define Hk2 similarly. Argue that Hki
i
is open and dense and then consider Pi
k=1 Hk .




R
11. Let n f = 0 sin n + 12 y f (y) dy for f L1 (0, ) . Show that sup {||n || : n = 1, 2, } < using
the Riemann Lebesgue lemma.
12. Let X be a normed linear space and let M be a convex open set containing 0. Define
(x) = inf{t > 0 :

x
M }.
t

Show is a gauge function defined on X. This particular example is called a Minkowski functional.
Recall a set, M , is convex if x + (1 )y M whenever [0, 1] and x, y M .
13. This problem explores the use of the Hahn Banach theorem in establishing separation theorems.
Let M be an open convex set containing 0. Let x
/ M . Show there exists x X 0 such that

Re x (x) 1 > Re x (y) for all y M . Hint: If y M, (y) < 1. Show this. If x
/ M, (x) 1. Try
f (x) = (x) for R. Then extend f to F , show F is continuous, then fix it so F is the real part
of x X 0 .
14. A Banach space is said to be strictly convex if whenever ||x|| = ||y|| and x 6= y, then


 x + y  


 2  < ||x||.

F : X X 0 is said to be a duality map if it satisfies the following: a.) ||F (x)|| = ||x||. b.)
F (x)(x) = ||x||2 . Show that if X 0 is strictly convex, then such a duality map exists. Hint: Let
f (x) = ||x||2 and use Hahn Banach theorem, then strict convexity.

14.4. EXERCISES

255

15. Suppose D X, a Banach space, and L : D Y is a closed operator. D might not be a Banach space
with respect to the norm on X. Define a new norm on D by ||x||D = ||x||X + ||Lx||Y . Show (D, || ||D )
is a Banach space.
16. Prove the following theorem which is an improved version of the open mapping theorem, [8]. Let X
and Y be Banach spaces and let A L (X, Y ). Then the following are equivalent.
AX = Y,
A is an open map.
There exists a constant M such that for every y Y , there exists x X with y = Ax and
||x|| M ||y||.
17. Here is an example of a closed unbounded operator. Let X = Y = C([0, 1]) and let
D = {f C 1 ([0, 1]) : f (0) = 0}.
L : D C([0, 1]) is defined by Lf = f 0 . Show L is closed.
18. Suppose D X and D is dense in X. Suppose L : D Y is linear and ||Lx|| K||x|| for all x D.
e defined on all of X with ||Lx||
e K||x|| and L
e is linear.
Show there is a unique extension of L, L,

19. A Banach space is uniformly convex if whenever ||xn ||, ||yn || 1 and ||xn + yn || 2, it follows that
||xn yn || 0. Show uniform convexity implies strict convexity.

20. We say that xn converges weakly to x if for every x X 0 , x (xn ) x (x). We write xn * x to
denote weak convergence. Show that if ||xn x|| 0, then xn * x.
21. Show that if X is uniformly convex, then xn * x and ||xn || ||x|| implies ||xn x|| 0. Hint:
Use Lemma 14.31 to obtain f X 0 with ||f || = 1 and f (x) = ||x||. See Problem 19 for the definition
of uniform convexity.

22. Suppose L L (X, Y ) and M L (Y, Z). Show M L L (X, Z) and that (M L) = L M .
23. In Theorem 14.27, it was shown that ||L || ||L||. Are these actually equal? Hint: You might show
that supB supA a (, ) = supA supB a (, ) and then use this in the string of inequalities
used to prove ||L || ||L|| along with the fact that ||Jx|| = ||x|| which was established in Theorem
14.30.

256

BANACH SPACES

Hilbert Spaces
15.1

Basic theory

Let X be a vector space. An inner product is a mapping from X X to C if X is complex and from X X
to R if X is real, denoted by (x, y) which satisfies the following.
(x, x) 0, (x, x) = 0 if and only if x = 0,

(15.1)

(x, y) = (y, x).

(15.2)

(ax + by, z) = a(x, z) + b(y, z).

(15.3)

For a, b C and x, y, z X,

Note that (15.2) and (15.3) imply (x, ay + bz) = a(x, y) + b(x, z).
1/2
We will show that if (, ) is an inner product, then (x, x)
defines a norm and we say that a normed
linear space is an inner product space if ||x|| = (x, x)1/2.
Definition 15.1 A normed linear space in which the norm comes from an inner product as just described
is called an inner product space. A Hilbert space is a complete inner product space.
Thus a Hilbert space is a Banach space whose norm comes from an inner product as just described. The
difference between what we are doing here and the earlier references to Hilbert space is that here we will be
making no assumption that the Hilbert space is finite dimensional. Thus we include the finite dimensional
material as a special case of that which is presented here.
Pn
Example 15.2 Let X = Cn with the inner product given by (x, y) i=1 xi y i . This is a complex Hilbert
space.
Example 15.3 Let X = Rn , (x, y) = x y. This is a real Hilbert space.
Theorem 15.4 (Cauchy Schwarz) In any inner product space
|(x, y)| ||x|| ||y||.
Proof: Let C, || = 1, and (x, y) = |(x, y)| = Re(x, y). Let
F (t) = (x + ty, x + ty).
If y = 0 there is nothing to prove because
(x, 0) = (x, 0 + 0) = (x, 0) + (x, 0)
257

258

HILBERT SPACES

and so (x, 0) = 0. Thus, we may assume y 6= 0. Then from the axioms of the inner product, (15.1),
F (t) = ||x||2 + 2t Re(x, y) + t2 ||y||2 0.
This yields
||x||2 + 2t|(x, y)| + t2 ||y||2 0.
Since this inequality holds for all t R, it follows from the quadratic formula that
4|(x, y)|2 4||x||2 ||y||2 0.
This yields the conclusion and proves the theorem.
Earlier it was claimed that the inner product defines a norm. In this next proposition this claim is proved.
1/2

Proposition 15.5 For an inner product space, ||x|| (x, x)

does specify a norm.

Proof: All the axioms are obvious except the triangle inequality. To verify this,
2

||x + y||

(x + y, x + y) ||x|| + ||y|| + 2 Re (x, y)

2

||x|| + ||y|| + 2 |(x, y)|

||x|| + ||y|| + 2 ||x|| ||y|| = (||x|| + ||y||) .

The following lemma is called the parallelogram identity.

Lemma 15.6 In an inner product space,
||x + y||2 + ||x y||2 = 2||x||2 + 2||y||2.
The proof, a straightforward application of the inner product axioms, is left to the reader. See Problem
7. Also note that
||x|| = sup |(x, y)|

(15.4)

||y||1

because by the Cauchy Schwarz inequality, if x 6= 0,

||x|| sup |(x, y)|
||y||1

x,

x
||x||

= ||x|| .

It is obvious that (15.4) holds in the case that x = 0.

One of the best things about Hilbert space is the theorem about projection onto a closed convex set.
Recall that a set, K, is convex if whenever [0, 1] and x, y K, x + (1 )y K.
Theorem 15.7 Let K be a closed convex nonempty subset of a Hilbert space, H, and let x H. Then there
exists a unique point P x K such that ||P x x|| ||y x|| for all y K.
Proof: First we show uniqueness. Suppose ||zi x|| ||yx|| for all y K. Then using the parallelogram
identity and
||z1 x|| ||y x||
for all y K,
||z1 x||2

z1 + z2
z1 x z2 x 2
x||2 = ||
+
||
2
2
2
z1 x 2
z2 x 2
z1 z 2 2
2(||
|| + ||
|| ) ||
||
2
2
2
z1 z2 2
||z1 x||2 ||
|| ,
2
||

15.1. BASIC THEORY

259

where the last inequality holds because

||z2 x|| ||z1 x||.
Hence z1 = z2 and this shows uniqueness.
Now let = inf{||x y|| : y K} and let yn be a minimizing sequence. Thus limn ||x yn || = , yn
K. Then by the parallelogram identity,
yn + ym
x||2 )
2
2(||yn x||2 + ||ym x||2 ) 42.

||yn ym ||2

2(||yn x||2 + ||ym x||2 ) 4(||

Since ||x yn || , this shows {yn } is a Cauchy sequence. Since H is complete, yn y for some y H
which must be in K because K is closed. Therefore ||x y|| = and we let P x = y.
Corollary 15.8 Let K be a closed, convex, nonempty subset of a Hilbert space, H, and let x
/ K. Then
for z K, z = P x if and only if
Re(x z, y z) 0

(15.5)

for all y K.
Before proving this, consider what it says in the case where the Hilbert space is Rn.

yX
yX
X
z

-x

Condition (15.5) says the angle, , shown in the diagram is always obtuse. Remember, the sign of x y
is the same as the sign of the cosine of their included angle.
The inequality (15.5) is an example of a variational inequality and this corollary characterizes the projection of x onto K as the solution of this variational inequality.
Proof of Corollary: Let z K. Since K is convex, every point of K is in the form z + t(y z) where
t [0, 1] and y K. Therefore, z = P x if and only if for all y K and t [0, 1],
||x (z + t(y z))||2 = ||(x z) t(y z)||2 ||x z||2
for all t [0, 1] if and only if for all t [0, 1] and y K
2

||x z|| + t2 ||y z|| 2t Re (x z, y z) ||x z||

which is equivalent to (15.5). This proves the corollary.

Definition 15.9 Let H be a vector space and let U and V be subspaces. We write U V = H if every
element of H can be written as a sum of an element of U and an element of V in a unique way.
The case where the closed convex set is a closed subspace is of special importance and in this case the
above corollary implies the following.
Corollary 15.10 Let K be a closed subspace of a Hilbert space, H, and let x
/ K. Then for z K, z = P x
if and only if
(x z, y) = 0
for all y K. Furthermore, H = K K where
K {x H : (x, k) = 0 for all k K}

260

HILBERT SPACES

Proof: Since K is a subspace, the condition (15.5) implies

Re(x z, y) 0
for all y K. But this implies this inequality holds with replaced with =. To see this, replace y with y.
Now let || = 1 and
(x z, y) = |(x z, y)|.
Since y K for all y K,
0 = Re(x z, y) = (x z, y) = (x z, y) = |(x z, y)|.
Now let x H. Then x = x P x + P x and from what was just shown, x P x K and P x K. This
shows that K + K = H. We need to verify that K K = {0} because this implies that there is at most
one way to write an element of H as a sum of one from K and one from K . Suppose then that z K K .
Then from what was just shown, (z, z) = 0 and so z = 0. This proves the corollary.
The following theorem is called the Riesz representation theorem for the dual of a Hilbert space. If z H
then we may define an element f H 0 by the rule
(x, z) f (x).
It follows from the Cauchy Schwartz inequality and the properties of the inner product that f H 0 . The
Riesz representation theorem says that all elements of H 0 are of this form.
Theorem 15.11 Let H be a Hilbert space and let f H 0 . Then there exists a unique z H such that
f (x) = (x, z) for all x H.
Proof: If f = 0, there is nothing to prove so assume without loss of generality that f 6= 0. Let
M = {x H : f (x) = 0}. Thus M is a closed proper subspace of H. Let y
/ M . Then y P y w has the
property that (x, w) = 0 for all x M by Corollary 15.10. Let x H be arbitrary. Then
xf (w) f (x)w M
so
0 = (f (w)x f (x)w, w) = f (w)(x, w) f (x)||w||2.
Thus
f (x) = (x,

f (w)w
)
||w||2

and so we let
z=

f (w)w
.
||w||2

This proves the existence of z. If f (x) = (x, zi ) i = 1, 2, for all x H, then for all x H,
(x, z1 z2 ) = 0.
Let x = z1 z2 to conclude uniqueness. This proves the theorem.

15.2. ORTHONORMAL SETS

15.2

261

Orthonormal sets

The concept of an orthonormal set of vectors is a generalization of the notion of the standard basis vectors
of Rn or Cn .
Definition 15.12 Let H be a Hilbert space. S H is called an orthonormal set if ||x|| = 1 for all x S
and (x, y) = 0 if x, y S and x 6= y. For any set, D, we define D {x H : (x, d) = 0 for all d D} . If
S is a set, we denote by span (S) the set of all finite linear combinations of vectors from S.
We leave it as an exercise to verify that D is always a closed subspace of H.
Theorem 15.13 In any separable Hilbert space, H, there exists a countable orthonormal set, S = {xi } such
that the span of these vectors is dense in H. Furthermore, if x H, then
x=

(x, xi ) xi lim

i=1

n
X

(x, xi ) xi .

(15.6)

i=1

Proof: Let F denote the collection of all orthonormal subsets of H. We note F is nonempty because
{x} F where ||x|| = 1. The set, F is a partially ordered set if we let the order be given by set inclusion.
By the Hausdorff maximal theorem, there exists a maximal chain, C in F. Then we let S C. It follows
S must be a maximal orthonormal set of vectors. It remains to verify that S is countable span (S) is dense,
and the condition, (15.6) holds. To see S is countable note that if x, y S, then
2

||x y|| = ||x|| + ||y|| 2 Re (x, y) = ||x|| + ||y|| = 2.

Therefore, the open sets, B x, 21 for x S are disjoint and cover S. Since H is assumed to be separable,
there exists a point from a countable dense set in each of these disjoint balls showing there can only be
countably many of the balls and that consequently, S is countable as claimed.
It remains to verify (15.6) and that span (S) is dense. If span (S) is not dense, then span (S) is a closed

proper subspace of H and letting y

/ span (S) we see that z y P y span (S) . But then S {z} would
be a larger orthonormal set of vectors contradicting the maximality of S.

It remains to verify (15.6). Let S = {xi }i=1 and consider the problem of choosing the constants, ck in
such a way as to minimize the expression

 2
n


X


ck xk  =
 x


k=1

||x|| +

n
X

|ck |

k=1

n
X

ck (x, xk )

n
X

ck (x, xk ).

k=1

k=1

We see this equals

2

||x|| +

n
X

|ck (x, xk )|

n
X

|(x, xk )|

k=1

k=1

and therefore, this minimum is achieved when ck = (x, xk ) . Now since span (S) is dense, there exists n large
enough that for some choice of constants, ck ,

2
n


X


ck xk  < .
x


k=1

262

HILBERT SPACES

However, from what was just shown,


  2 
 2
n
n




X
X




(x, xi ) xi  x
ck xk  <
x




i=1

k=1

Pn

showing that limn i=1 (x, xi ) xi = x as claimed. This proves the theorem.
In the proof of this theorem, we established the following corollary.
Corollary 15.14 Let S be any orthonormal set of vectors and let {x1 , , xn } S. Then if x H

2 
2
n
n




X
X




ck xk  x
(x, xi ) xi 
x




i=1

k=1

for all choices of constants, ck . In addition to this, we have Bessels inequality

2

||x||

n
X

|(x, xk )| .

k=1

If S is countable and span (S) is dense, then letting {xi }i=1 = S, we obtain (15.6).
Definition 15.15 Let A L (H, H) where H is a Hilbert space. Then |(Ax, y)| ||A|| ||x|| ||y|| and so
the map, x (Ax, y) is continuous and linear. By the Riesz representation theorem, there exists a unique
element of H, denoted by A y such that
(Ax, y) = (x, A y) .
It is clear y A y is linear and continuous. We call A the adjoint of A. We say A is a self adjoint operator
if A = A . Thus for all x, y H, (Ax, y) = (x, Ay) . We say A is a compact operator if whenever {xk } is a
bounded sequence, there exists a convergent subsequence of {Axk } .
The big result in this subject is sometimes called the Hilbert Schmidt theorem.
Theorem 15.16 Let A be a compact self adjoint operator defined on a Hilbert space, H. Then there exists
a countable set of eigenvalues, {i } and an orthonormal set of eigenvectors, ui satisfying
i is real, |n | |n+1 | , Aui = i ui ,

(15.7)

lim n = 0,

(15.8)

span (u1 , , un ) = H.

(15.9)

and either
n

or for some n,

In any case,

span ({ui }i=1 ) is dense in A (H) .

(15.10)

and for all x H,

Ax =

k=1

k (x, uk ) uk .

(15.11)

15.2. ORTHONORMAL SETS

263

This sequence of eigenvectors and eigenvalues also satisfies

|n | = ||An || ,

(15.12)

An : Hn Hn .

(15.13)

and

where H H1 and Hn {u1 , , un1 }

and An is the restriction of A to Hn .

Proof: If A = 0 then we may pick u H with ||u|| = 1 and let 1 = 0. Since A (H) = 0 it follows the
span of u is dense in A (H) and we have proved the theorem in this case.
2
Thus, we may assume A 6= 0. Let 1 be real and 21 ||A|| . We know from the definition of ||A||
there exists xn , ||xn || = 1, and ||Axn || ||A|| = |1 | . Now it is clear that A2 is also a compact self adjoint
operator. We consider




2
21 A2 xn , xn = 21 ||Axn || 0.
Since A is compact, we may replace {xn } by a subsequence, still denoted by {xn } such that Axn converges
to some element of H. Thus since 21 A2 satisfies




21 A2 y, y 0




in addition to being self adjoint, it follows x, y 21 A2 x, y satisfies all the axioms for an inner product
except for the one which says that (z, z) = 0 only if z = 0. Therefore, the Cauchy Schwartz inequality (see
Problem 6) may be used to write





 21 A2 xn , y 

1/2

21 A2 y, y
en ||y|| .

1/2
21 A2 xn , xn

where en 0 as n . Therefore, taking the sup over all ||y|| 1, we see




lim  21 A2 xn  = 0.
n

Since A2 xn converges, it follows since 1 6= 0 that {xn } is a Cauchy sequence converging to x with ||x|| = 1.
Therefore, A2 xn A2 x and so
 2


 1 A2 x = 0.
Now

(1 I A) (1 I + A) x = (1 I + A) (1 I A) x = 0.
y
x
. If (1 I A) x = y 6= 0, we let u1 ||y||
.
If (1 I A) x = 0, we let u1 ||x||
Suppose we have found {u1 , , un } such that Auk = k uk and |k | |k+1 | , |k | = ||Ak || and
Ak : Hk Hk for k n. If

span (u1 , , un ) = H
we have obtained the conclusion of the theorem and we are in the situation of (15.9). Therefore, we assume
the span of these vectors is always a proper subspace of H. We show that An+1 : Hn+1 Hn+1 . Let

y Hn+1 {u1 , , un }

264

HILBERT SPACES

Then for k n
(Ay, uk ) = (y, Auk ) = k (y, uk ) = 0,
showing An+1 : Hn+1 Hn+1 as claimed. We have two cases to consider. Either n = 0 or it is not. In
the case where n = 0 we see An = 0. Then every element of H is the sum of one in span (u1 , , un ) and

one in span (u1 , , un ) . (note span (u1 , , un ) is a closed subspace. See Problem 11.) Thus, if x H,

we can
Pn write x = y + z where y span (u1 , , un ) and z span (u1 , , un ) and Az = 0. Therefore,
y = j=1 cj uj and so
Ax = Ay =

n
X

cj Auj

j=1

n
X

cj j uj span (u1 , , un ) .

j=1

It follows that we have the conclusion of the theorem in this case because the above equation holds if we let
ci = (x, ui ) .
Now consider the case where n 6= 0. In this case we repeat the above argument used to find u1 and 1

for the operator, An+1 . This yields un+1 Hn+1 {u1 , , un } such that
||un+1 || = 1, ||Aun+1 || = |n+1 | = ||An+1 || ||An || = |n |
and if it is ever the case that n = 0, it follows from the above argument that the conclusion of the theorem
is obtained.
Now we claim limn n = 0. If this were not so, we would have 0 < = limn |n | but then
2

||Aun Aum ||

= ||n un m um ||
2

= |n | + |m | 22

and so there would not exist a convergent subsequence of {Auk }k=1 contrary to the assumption that A is
compact. Thus we have verified the claim that limn n = 0. It remains to verify that span ({ui }) is dense

in A (H) . If w span ({ui }) then w Hn for all n and so for all n, we have
||Aw|| ||An || ||w|| |n | ||w|| .
Therefore, Aw = 0. Now every vector from H can be written as a sum of one from

span ({ui }) = span ({ui })

and one from span ({ui }). Therefore, if x H, we can write x = y + w where y span ({ui }) and

w span ({ui }) . From what we just showed, we see Aw = 0. Also, since y span ({ui }), there exist
constants, ck and n such that


n


X


ck uk  < .
y


k=1

Therefore, from Corollary 15.14,


 

n
n

 

X
X

 

(y, uk ) uk  = y
(x, uk ) uk  < .
y

 

k=1

k=1

15.2. ORTHONORMAL SETS

265

Therefore,
||A||


!
n


X


> A y
(x, uk ) uk 


k=1


n


X


(x, uk ) k uk  .
= Ax


k=1

Since is arbitrary, this shows span ({ui }) is dense in A (H) and also implies (15.11). This proves the
theorem.
Note that if we define v u L (H, H) by
v u (x) = (x, u) v,
then we can write (15.11) in the form
A=

k uk uk

k=1

We give the following useful corollary.

Corollary 15.17 Let A be a compact self adjoint operator defined on a separable Hilbert space, H. Then
there exists a countable set of eigenvalues, {i } and an orthonormal set of eigenvectors, vi satisfying
Avi = i vi , ||vi || = 1,

(15.14)

span ({vi }i=1 ) is dense in H.

(15.15)

Furthermore, if i 6= 0, the space, Vi {x H : Ax = i x} is finite dimensional.

Proof: In the proof of the above theorem, let W span ({ui }) . By Theorem 15.13, there is an

orthonormal set of vectors, {wi }i=1 whose span is dense in W. As shown in the proof of the above theorem,

Aw = 0 for all w W. Let {vi }i=1 = {ui }i=1 {wi }i=1 .

It remains to verify the space, Vi , is finite dimensional. First we observe that A : Vi Vi . Since A is
continuous, it follows that A : Vi Vi . Thus A is a compact self adjoint operator on Vi and by Theorem
15.16, we must be in the situation of (15.9) because the only eigenvalue is i . This proves the corollary.
Note the last claim of this corollary holds independent of the separability of H.
Suppose
/ {n } and 6= 0. Then we can use the above formula for A, (15.11), to give a formula for
1
2
(A I) . We note first that since limn n = 0, it follows that 2n / (n ) must be bounded, say by
a positive constant, M.

Corollary 15.18 Let A be a compact self adjoint operator and let

/ {n }n=1 and 6= 0 where the n are
the eigenvalues of A. Then
1

(A I)

1
1 X k
x= x+
(x, uk ) uk .

(15.16)

k=1

Proof: Let m < n. Then since the {uk } form an orthonormal set,
 n

!1/2
2
n 
X

X
k


k
2
(x, uk ) uk  =
|(x, uk )|



k
k
k=m
k=m
!1/2
n
X
2
M
|(x, uk )|
.
k=m

(15.17)

266

HILBERT SPACES

But we have from Bessels inequality,

|(x, uk )| ||x||

k=1

and so for m large enough, the first term in (15.17) is smaller than . This shows the infinite series in (15.16)
converges. It is now routine to verify that the formula in (15.16) is the inverse.

15.3

The Fredholm alternative

Recall that if A is an n n matrix and if the only solution to the system, Ax = 0 is x = 0 then for any
y Rn it follows that there exists a unique solution to the system Ax = y. This holds because the first
condition implies A is one to one and therefore, A1 exists. Of course things are much harder in a Hilbert
space. Here is a simple example.
Example 15.19 Let L2 (N; ) = H where is counting measure. Thus an element of H is a sequence,

a = {ai }i=1 having the property that

||a||H

|ak |

!1/2

< .

k=1

We define A : H H by
Aa b {0, a1 , a2 , } .
Thus A slides the sequence to the right and puts a zero in the first slot. Clearly A is one to one and linear
but it cannot be onto because it fails to yield e1 {1, 0, 0, } .
Notwithstanding the above example, there are theorems which are like the linear algebra theorem mentioned above which hold in an arbitrary Hilbert space in the case where some operator is compact. To begin
with we give a simple lemma which is interesting for its own sake.
Lemma 15.20 Suppose A is a compact operator defined on a Hilbert space, H. Then (I A) (H) is a closed
subspace of H.
Proof: Suppose (I A) xn y. Let
n dist (xn , ker (I A))
and let zn ker (I A) be such that
n ||xn zn ||

1
1+
n

n .

Thus (I A) (xn zn ) y.
Case 1: {xn zn } has a bounded subsequence.
If this is so, the compactness of A implies there exists a subsequence, still denoted by n such that

{A (xn zn )}n=1 is a Cauchy sequence. Since (I A) (xn zn ) y, this implies {(xn zn )} is also a
Cauchy sequence converging to a point, x H. Then, taking the limit as n , we see (I A) x = y and
so y (I A) (H) .
Case 2: limn ||xn zn || = . We will show this case cannot occur.
n
In this case, we let wn ||xxnn z
zn || . Thus (I A) wn 0 and wn is bounded. Therefore, we can take
a subsequence, still denoted by n such that {Awn } is a Cauchy sequence. This implies {wn } is a Cauchy

15.3. THE FREDHOLM ALTERNATIVE

267

sequence which must converge to some w H. Therefore, (I A) w = 0 and so w ker (I A) .

However, this is impossible because of the following argument. If z ker (I A) ,
||wn z|| =

1
||xn zn ||xn zn || z||
||xn zn ||
1
n
n


n
=
.
||xn zn ||
n+1
1 + n1 n

Taking the limit, we see ||w z|| 1. Since z ker (I A) is arbitrary, this shows dist (w , ker (I A))
1.
Since Case 2 does not occur, this proves the lemma.
Theorem 15.21 Let A be a compact operator defined on a Hilbert space, H and let f H. Then there
exists a solution, x, to
x Ax = f

(15.18)

(f, y) = 0

(15.19)

y A y = 0.

(15.20)

if and only if

for all y solving

Proof: Suppose x is a solution to (15.18) and let y be a solution to (15.20). Then

(f, y)

= (x Ax, y) = (x, y) (Ax, y)

= (x, y) (x, A y) = (x, y A y) = 0.

Next suppose (f, y) = 0 for all y solving (15.20). We want to show there exists x solving (15.18). By
Lemma 15.20, (I A) (H) is a closed subspace of H. Therefore, there exists a point, (I A) x, in this
subspace which is the closest point to f. By Corollary 15.10, we must have
(f (I A) x, (I A) y (I A) x) = 0
for all y H. Therefore,
((I A ) [f (I A) x] , y x) = 0
for all y H. This implies x is a solution to
(I A ) (I A) x = (I A ) f
and so
(I A ) [(I A) x f ] = 0.
Therefore (f, f (I A) x) = 0. Since (I A) x (I A) (H) , we also have
((I A) x, f (I A) x) = 0
and so
(f (I A) x, f (I A) x) = 0,
showing that f = (I A) x. This proves the theorem.
The following corollary is called the Fredholm alternative.

268

HILBERT SPACES

Corollary 15.22 Let A be a compact operator defined on a Hilbert space, H. Then there exists a solution
to the equation
x Ax = f

(15.21)

for all f H if and only if (I A ) is one to one.

Proof: Suppose (I A) is one to one first. Then if y A y = 0 it follows y = 0 and so for any f H,
(f, y) = (f, 0) = 0. Therefore, by Theorem 15.21 there exists a solution to (I A) x = f.
Now suppose there exists a solution, x, to (I A) x = f for every f H. If (I A ) y = 0, we can let
(I A) x = y and so
2

||y|| = (y, y) = ((I A) x, y) = (x, (I A ) y) = 0.

Therefore, (I A ) is one to one.

15.4

Sturm Liouville problems

A Sturm Liouville problem involves the differential equation,

0

(p (x) y 0 ) + (q (x) + r (x)) y = 0, x [a, b]

(15.22)

where we assume that q (t) 6= 0 for any t and some boundary conditions,
boundary condition at a
boundary condition at b

(15.23)

For example we typically have boundary conditions of the form

C1 y (a) + C2 y 0 (a)
C3 y (b) + C4 y 0 (b)

=
=

0
0

(15.24)

where
C12 + C22 > 0, and C32 + C42 > 0.

(15.25)

We will assume all the functions involved are continuous although this could certainly be weakened. Also
we assume here that a and b are finite numbers. In the example the constants, Ci are given and is called
the eigenvalue while a solution of the differential equation and given boundary conditions corresponding to
is called an eigenfunction.
There is a simple but important identity related to solutions of the above differential equation. Suppose
i and yi for i = 1, 2 are two solutions of (15.22). Thus from the equation, we obtain the following two
equations.
0

(p (x) y10 ) y2 + (1 q (x) + r (x)) y1 y2 = 0,

0

(p (x) y20 ) y1 + (2 q (x) + r (x)) y1 y2 = 0.

Subtracting the second from the first yields
0

(p (x) y10 ) y2 (p (x) y20 ) y1 + (1 2 ) q (x) y1 y2 = 0.

(15.26)

15.4. STURM LIOUVILLE PROBLEMS

269

Now we note that

0

(p (x) y10 ) y2 (p (x) y20 ) y1 =

d
((p (x) y10 ) y2 (p (x) y20 ) y1 )
dx

and so integrating (15.26) from a to b, we obtain

((p (x) y10 ) y2 (p (x) y20 ) y1 ) |ba + (1 2 )

q (x) y1 (x) y2 (x) dx = 0

(15.27)

We have been purposely vague about the nature of the boundary conditions because of a desire to not lose
generality. However, we will always assume the boundary conditions are such that whenever y1 and y2 are
two eigenfunctions, it follows that
((p (x) y10 ) y2 (p (x) y20 ) y1 ) |ba = 0

(15.28)

In the case where the boundary conditions are given by (15.24), and (15.25), we obtain (15.28). To see why
this is so, consider the top limit. This yields
p (b) [y10 (b) y2 (b) y20 (b) y1 (b)]
However we know from the boundary conditions that
C3 y1 (b) + C4 y10 (b) =
C3 y2 (b) + C4 y20 (b) =

0
0

and that from (15.25) that not both C3 and C4 equal zero. Therefore the determinant of the matrix of
coefficients must equal zero. But this is implies [y10 (b) y2 (b) y20 (b) y1 (b)] = 0 which yields the top limit is
equal to zero. A similar argument holds for the lower limit.
With the identity (15.27) we can give a result on orthogonality of the eigenfunctions.
Proposition 15.23 Suppose yi solves the problem (15.22) , (15.23), and (15.28) for = i where 1 6= 2 .
Then we have the orthogonality relation
Z b
q (x) y1 (x) y2 (x) dx = 0.
(15.29)
a

In addition to this, if u, v are two solutions to the differential equation corresponding to , (15.22), not
necessarily the boundary conditions, then there exists a constant, C such that
W (u, v) (x) p (x) = C
for all x [a, b] . In this formula, W (u, v) denotes the Wronskian given by


u (x) v (x)
det
.
u0 (x) v 0 (x)

(15.30)

(15.31)

Proof: The orthogonality relation, (15.29) follows from the fundamental assumption, (15.28) and (15.27).
It remains to verify (15.30). We have
0

(p (x) u0 ) v (p (x) v 0 ) u + ( ) q (x) uv = 0.

Now the first term equals
d
d
(p (x) u0 v p (x) v 0 u) =
(p (x) W (v, u) (x))
dx
dx
and so p (x) W (u, v) (x) = p (x) W (v, u) (x) = C as claimed.
Now consider the differential equation,
0

(p (x) y 0 ) + r (x) y = 0.
This is obtained from the one of interest by letting = 0.

(15.32)

270

HILBERT SPACES

Criterion 15.24 Suppose we are able to find functions, u and v such that they solve the differential equation,
(15.32) and u solves the boundary condition at x = a while v solves the boundary condition at x = b. Suppose
also that W (u, v) 6= 0.
If p (x) > 0 on [a, b] it is clear from the fundamental existence and uniqueness theorems for ordinary
differential equations that such functions, u and v exist. (See any good differential equations book or
Problem 10 of Chapter 10.) The following lemma gives an easy to check sufficient condition for Criterion
15.24 to occur in the case where the boundary conditions are given in (15.24), (15.25).
Lemma 15.25 Suppose p (x) 6= 0 for all x [a, b] . Then for C1 and C2 given in (15.24) and u a nonzero
solution of (15.32), if
C3 u (b) + C4 u0 (b) 6= 0,
Then if v is any nonzero solution of the equation (15.32) which satisfies the boundary condition at x = b, it
follows W (u, v) 6= 0.
Proof: Thanks to Proposition 15.23 W (u, v) (x) is either equal to zero for all x [a, b] or it is never
equal to zero on this interval. If the conclusion of the lemma is not so, then uv equals a constant. This
is easy to see from using the quotient rule in which the Wronskian occurs in the numerator. Therefore,
v (x) = u (x) c for some nonzero constant, c But then
C3 v (b) + C4 v 0 (b) = c (C3 u (b) + C4 u0 (b)) 6= 0,
contrary to the assumption that v satisfies the boundary condition at x = b. This proves the lemma.
Lemma 15.26 Assume Criterion 15.24. In the case where the boundary conditions are given by (15.24) and
(15.25), a function, y is a solution to the boundary conditions, (15.24) and (15.25) along with the equation,
0

(p (x) y 0 ) + r (x) y = g

(15.33)

if and only if
y (x) =

G (t, x) g (t) dt

(15.34)

where
G1 (t, x) =

c1 (p (t) v (x) u (t)) if t < x

.
c1 (p (t) v (t) u (x)) if t > x

(15.35)

where c is the constant of Proposition 15.23 which satisfies p (x) W (u, v) (x) = c.
Proof: We can verify that if
Z
Z
1 x
1 b
yp (x) =
g (t) p (t) u (t) v (x) dt +
g (t) p (t) v (t) u (x) dt,
c a
c x
then yp is a particular solution of the equation (15.33) which satisfies the boundary conditions, (15.24) and
(15.25). Therefore, every solution of the equation must be of the form
y (x) = u (x) + v (x) + yp (x) .
Substituting in to the given boundary conditions, (15.24), we obtain
(C1 v (a) + C2 v 0 (a)) = 0.

15.4. STURM LIOUVILLE PROBLEMS

271

If 6= 0, then we have
C1 v (a) + C2 v 0 (a)
C1 u (a) + C2 u0 (a)

=
=

0
0.

Since C12 + C22 6= 0, we must have

(v (a) u0 (a) u (a) v 0 (a)) = W (v, u) (a) = 0
which contradicts the assumption in Criterion 15.24 about the Wronskian. Thus = 0. Similar reasoning
applies to show = 0 also. This proves the lemma.
Now in the case of Criterion 15.24, y is a solution to the Sturm Liouville eigenvalue problem, (15.22),
(15.24), and (15.25) if and only if y solves the boundary conditions, (15.24) and the equation,
0

(p (x) y 0 ) + r (x) y (x) = q (x) y (x) .

This happens if and only if
Z
Z
x
b
y (x) =
q (t) y (t) p (t) u (t) v (x) dt +
q (t) y (t) p (t) v (t) u (x) dt,
c a
c x

(15.36)

Letting = 1 , this if of the form

y (x) =

G (t, x) y (t) dt

(15.37)

c1 (q (t) p (t) v (x) u (t)) if t < x

.
c1 (q (t) p (t) v (t) u (x)) if t > x

(15.38)

where
G (t, x) =

Could = 0? If this happened, then from Lemma 15.26, we would have that 0 is the solution of (15.33)
where the right side is q (t) y (t) which would imply that q (t) y (t) = 0 for all t which implies y (t) = 0 for
all t. It follows from (15.38) that G : [a, b] [a, b] R is continuous and symmetric, G (t, x) = G (x, t) .
Also we see that for f C ([a, b]) , and
w (x)

G (t, x) f (t) dt,

Lemma 15.26 implies w is the solution to the boundary conditions (15.24), (15.25) and the equation
0

(p (x) y 0 ) + r (x) y = q (x) f (x)

(15.39)

Theorem 15.27 Suppose u, v are given in Criterion 15.24. Then there exists a sequence of functions,

{yn }n=1 and real numbers, n such that

0

(p (x) yn0 ) + (n q (x) + r (x)) yn = 0, x [a, b] ,

C1 yn (a) + C2 yn0 (a)
C3 yn (b) + C4 yn0 (b)

= 0,
= 0.

(15.40)

(15.41)

and
lim |n | =

(15.42)

272

HILBERT SPACES

such that for all f C ([a, b]) , whenever w satisfies (15.39) and the boundary conditions, (15.24),
w (x) =

X
1
(f, yn ) yn .

n=1 n

(15.43)

Also the functions, {yn } form a dense set in L2 (a, b) which satisfy the orthogonality condition, (15.29).
Rb
Proof: Let Ay (x) a G (t, x) y (t) dt where G is defined above in (15.35). Then A : L2 (a, b)
C ([a, b]) L2 (a, b). Also, for y L2 (a, b) we may use Fubinis theorem and obtain
(Ay, z)L2 (a,b)

G (t, x) y (t) z (x) dtdx

a
b

G (t, x) y (t) z (x) dxdt

= (y, Az)L2 (a,b)

showing that A is self adjoint.
Now suppose D L2 (a, b) is a bounded set and pick y D. Then
Z

 b



|Ay (x)| 
G (t, x) y (t) dt
 a

Z b

|G (t, x)| |y (t)| dt

a

CG ||y||L2 (a,b) C
where C is a constant which depends on G and the uniform bound on functions from D. Therefore, the
functions, {Ay : y D} are uniformly bounded. Now for y D, we use the uniform continuity of G on
[a, b] [a, b] to conclude that when |x z| is sufficiently small, |G (t, x) G (t, z)| < and that therefore,
Z

 b



(G (t, x) G (t, z)) y (t) dt
|Ay (x) Ay (z)| = 
 a

Z b

|y (t)| b a ||y||L2 (a,b)

a

Thus {Ay : y D} is uniformly equicontinuous and so by the Ascoli Arzela theorem, Theorem 4.4, this set of
functions is precompact in C ([a, b]) which means there exists a uniformly convergent subsequence, {Ayn } .
However this sequence must then be uniformly Cauchy in the norm of the space, L2 (a, b) and so A is a
compact self adjoint operator defined on the Hilbert space, L2 (a, b). Therefore, by Theorem 15.16, there
exist functions yn and real constants, n such that ||yn ||L2 = 1 and Ayn = n yn and


|n | n+1  , Aui = i ui ,
(15.44)
and either
lim n = 0,

(15.45)

span (y1 , , yn ) = H L2 (a, b) .

(15.46)

or for some n,

15.4. STURM LIOUVILLE PROBLEMS

273

Since (15.46) does not occur, we must have (15.45). Also from Theorem 15.16,

span ({yi }i=1 ) is dense in A (H) .

(15.47)

and so for all f C ([a, b]) ,

Af =

k (f, yk ) yk .

(15.48)

k=1

Thus for w a solution of (15.39) and the boundary conditions (15.24),

X
1
(f, yk ) yk .
k

w = Af =

k=1

The last claim follows from Corollary 15.17 and the observation above that is never equal to zero. This
proves the theorem.
Note that if since q (x) 6= 0 we can say that for a given g C ([a, b]) we can define f by g (x) = q (x) f (x)
and so if w is a solution to the boundary conditions, (15.24) and the equation
0

(p (x) w0 (x)) + r (x) w (x) = g (x) = q (x) f (x) ,

we may write the formula
w (x)

X
1
(f, yk ) yk
k
k=1



X
1 g
=
, y k yk .
k
q

k=1

Theorem 15.28 Suppose f L2 (a, b) . Then

f=

ak yk

(15.49)

f (x) yk (x) q (x) dx

Rb 2
y (x) q (x) dx
a k

(15.50)

k=1

where
ak =

Rb
a

and the convergence of the partial sums takes place in L2 (a, b) .

Proof: By Theorem 15.27 there exist bk and n such that


n


X


bk yk  < .
f


k=1

L2

Now we can define an equivalent norm on L2 (a, b) by

|||f |||L2 (a,b)

Z
a

!1/2

|f (x)| |q (x)| dx

274

HILBERT SPACES

Thus there exist constants and independent of g such that

||g|| |||g||| ||g||
Therefore,


n


X


bk yk   
f


k=1

< .
L2

Rb
This new norm also comes from the inner product ((f, g)) a f (x) g (x) |q (x)| dx. Then as in Theorem
9.21 a completing the square argument shows if we let bk be given as in (15.50) then the distance between
f and the linear combination of the first n of the yk measured in the norm, |||||| , is minimized Thus letting
ak be given by (15.50), we see that



 


n
n
n



 


X
X
X



 


bk yk  <
ak yk  f
ak yk  f
f




 

k=1

L2

k=1

L2

k=1

L2

Since is arbitrary, this proves the theorem.

More can be said about convergence of these series based on the eigenfunctions of a Sturm Liouville
problem. In particular, it can be shown that for reasonable functions the pointwise convergence properties
are like those of Fourier series and that the series converges to the midpoint of the jump. For more on these
topics see the old book by Ince, written in Egypt in the 1920s, [17].

15.5

Exercises

R1
1. Prove Examples 2.14 and 2.15 are Hilbert spaces. For f, g C ([0, 1]) let (f, g) = 0 f (x) g (x)dx.
Show this is an inner product. What does the Cauchy Schwarz inequality say in this context?
2. Generalize the Fredholm theory presented above to the case where A : X Y for X, Y Banach
spaces. In this context, A : Y 0 X 0 is given by A y (x) y (Ax) . We say A is compact if
A (bounded set) = precompact set, exactly as in the Hilbert space case.
3. Let S denote the unit sphere in a Banach space, X, S {x X : ||x|| = 1} . Show that if Y is a
Banach space, then A L (X, Y ) is compact if and only if A (S) is precompact.
4. Show that A L (X, Y ) is compact if and only if A is compact. Hint: Use the result of 3 and
the Ascoli Arzela theorem to argue that for S the unit ball in X 0 , there is a subsequence, {yn } S
such that yn converges uniformly on the compact set, A (S). Thus {A yn } is a Cauchy sequence in X 0 .
To get the other implication, apply the result just obtained for the operators A and A . Then use
results about the embedding of a Banach space into its double dual space.
5. Prove a version of Problem 4 for Hilbert spaces.
6. Suppose, in the definition of inner product, Condition (15.1) is weakened to read only (x, x) 0. Thus
the requirement that (x, x) = 0 if and only if x = 0 has been dropped. Show that then |(x, y)|
1/2
1/2
|(x, x)| |(y, y)| . This generalizes the usual Cauchy Schwarz inequality.
7. Prove the parallelogram identity. Next suppose (X, || ||) is a real normed linear space. Show that if
the parallelogram identity holds, then (X, || ||) is actually an inner product space. That is, there exists
an inner product (, ) such that ||x|| = (x, x)1/2 .
8. Let K be a closed convex subset of a Hilbert space, H, and let P be the projection map of the chapter.
Thus, ||P x x|| ||y x|| for all y K. Show that ||P x P y|| ||x y|| .

15.5. EXERCISES

275

9. Show that every inner product space is uniformly convex. This means that if xn , yn are vectors whose
norms are no larger than 1 and if ||xn + yn || 2, then ||xn yn || 0.
10. Let H be separable and let S be an orthonormal set. Show S is countable.
11. Suppose {x1 , , xm } is a linearly independent set of vectors in a normed linear space. Show span (x1 , , xm )
is a closed subspace. Also show that if {x1 , , xm } is an orthonormal set then span (x1 , , xm ) is a
closed subspace.
12. Show every Hilbert space, separable or not, has a maximal orthonormal set of vectors.
13. Prove BesselsPinequality, which says that if {xn }
n=1 is an orthonormal set in H,
Pn then for all

x H, ||x||2 k=1 |(x, xk )|2 . Hint: Show that if M = span(x1 , , xn ), then P x = k=1 xk (x, xk ).
Then observe ||x||2 = ||x P x||2 + ||P x||2 .
14. Show S is a maximal orthonormal set if and only if
span(S) {all finite linear combinations of elements of S}
is dense in H.
15. Suppose {xn }
n=1 is a maximal orthonormal set. Show that
x=

(x, xn )xn lim

n=1

and ||x||2 =

i=1

|(x, xi )|2 . Also show (x, y) =

N
X

(x, xn )xn

n=1

n=1 (x, xn )(y, xn ).

16. Let S = {einx (2) 2 }

n= . Show S is an orthonormal set if the inner product is given by (f, g) =
R
f (x) g (x)dx.

17. Show that if Bessels equation,

2

||y|| =

|(y, n )| ,

n=1

holds for all y H where

and

{n }n=1

is an orthonormal set, then {n }n=1 is a maximal orthonormal set



N


X


lim y
(y, n ) n  = 0.
N 

n=1

18. Suppose X is an infinite dimensional Banach space and suppose

{x1 xn }
are linearly independent with ||xi || = 1. Show span (x1 xn ) Xn is a closed linear subspace of X.
Now let z
/ Xn and pick y Xn such that ||z y|| 2 dist (z, Xn ) and let
xn+1 =

zy
.
||z y||

Show the sequence {xk } satisfies ||xn xk || 1/2 whenever k < n. Hint:

 

 z y
  z y xk ||z y|| 





.
 ||z y|| xk  = 

||z y||

Now show the unit ball {x X : ||x|| 1} is compact if and only if X is finite dimensional.

276

HILBERT SPACES

19. Show that if A is a self adjoint operator and Ay = y for a complex number and y 6= 0, then must
be real. Also verify that if A is self adjoint and Ax = x while Ay = y, then if 6= , it must be the
case that (x, y) = 0.
20. If Y is a closed subspace of a reflexive Banach space X, show Y is reflexive.
21. Show H k (Rn ) is a Hilbert space. See Problem 15 of Chapter 13 for a definition of this space.

Brouwer Degree
This chapter is on the Brouwer degree, a very useful concept with numerous and important 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 invariance of domain theorem. It also is used in bifurcation theory
and many other areas in which it is an essential tool. Our emphasis in this chapter will be on the Brouwer
degree 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 [7] and much of the presentation
here follows this reference.

16.1

Preliminary results

Definition 16.1 For a bounded open set, we denote by C the set of functions which are continuous




on and by C m , m the space of restrictions of functions in Cc (Rn ) to . The norm in C is
defined as follows.


||f || = ||f ||C () sup |f (x)| : x .




If the functions take values in Rn we will write C m ; Rn or C ; Rn if there is no differentiability


assumed. The norm on C ; Rn is defined in the same way as above,


||f || = ||f ||C (;Rn ) sup |f (x)| : x .
Also, we will denote by C (; Rn ) functions which are continuous on that have values in Rn and by
C m (; Rn ) functions which have m continuous derivatives defined on .




Theorem 16.2 Let be a bounded open set in Rn and let f C . Then there exists g C with
||g f ||C () .
Proof: This follows immediately from the Stone Weierstrass theorem. Let i (x) xi . Then the functions
i and the constant function, 0 (x) 1 separate the points of and annihilate
no point. Therefore, the


algebra generated by these functions, the polynomials, is dense in C . Thus we may take g to be a
polynomial.
Applying this result to the components of a vector valued function yields the following corollary.
Corollary 16.3 If f C ; Rn


C ; Rn such that

for a bounded subset of Rn , then for all > 0, there exists g

||g f || < .
277

278

BROUWER DEGREE

We make essential use of the following lemma a little later in establishing the definition of the degree,
given below, is well defined.
Lemma 16.4 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

where here (Dg)ij gi,j

Proof: det (Dg) =

gi
xj .

Pn

i=1 gi,j cof(Dg)ij

and so

det (Dg)
= cof (Dg)ij .
gi,j

(16.1)

Also
kj det (Dg) =

gi,k (cof (Dg))ij .

(16.2)

The reason for this is that if k 6= j this is just the expansion of a determinant of a matrix in which the k th
and j th columns are equal. Differentiate (16.2) with respect to xj and sum on j. This yields
X

kj

r,s,j

X
X
(det Dg)
gr,sj =
gi,kj (cof (Dg))ij +
gi,k cof (Dg)ij,j .
gr,s
ij
ij

Hence, using kj = 0 if j 6= k and (16.1),

X

(cof (Dg))rs gr,sk =

rs

gr,ks (cof (Dg))rs +

rs

gi,k cof (Dg)ij,j .

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

If det (gi,k ) 6= 0 so that (gi,k ) is invertible, this shows

(cof (Dg))ij,j = 0. If det (Dg) = 0, let

gk = g + k I
where k 0 and det (Dg + k I) det (Dgk ) 6= 0. Then
X
j

(cof (Dg))ij,j = lim

(cof (Dgk ))ij,j = 0

and this proves the lemma.

16.2

Definitions and elementary properties

In this section, f : Rn will be a continuous map. We make the following definitions.

16.2. DEFINITIONS AND ELEMENTARY PROPERTIES

279


Definition 16.5 Uy f C ; Rn : y
/ f () . For two functions,
f , g Uy ,
we will say f g if there exists a continuous function,
h : [0, 1] Rn
such that h (x, 1) = g (x) and h (x, 0) = f (x) and x h (x,t) Uy . This function, h, is called a homotopy
and we say that f and g are homotopic.
Definition 16.6 For W an open set in Rn and g C 1 (W ; Rn ) we say y is a regular value of g if whenever
x g1 (y) , det (Dg (x)) 6= 0. Note that if g1 (y) = , it follows that y is a regular value from this
definition.
Lemma 16.7 The relation, , is an equivalence relation and, denoting by [f ] the equivalence class deter

mined by f , it follows that [f ] is an open subset of Uy . Furthermore, Uy is an open set in C ; Rn . In


addition to this, if f Uy , there exists g [f ] C 2 ; Rn for which y is a regular value.
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 we define a homotopy of f and k as follows.



h1 (x,2t) if t 0, 21
 .
h (x,t)
h2 (x,2t 1) if t 12 , 1


It is obvious that Uy is an open subset of C ; Rn . We need to argue that [f ] is also an open set. However,


if f Uy , There exists > 0 such that B (y, 2) f () = . Let f1 C ; Rn with ||f1 f || < . Then
if t [0, 1] , and x
|f (x) + t (f1 (x) f (x)) y| |f (x) y| t ||f f 1 || > 2 t > 0.
Therefore, B (f ,) [f ] because if f1 B (f , ) , 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, there exists g [f ]


C 2 ; Rn . If y is a regular value of g, leave g unchanged. Otherwise, let


S x : det Dg (x) = 0
and pick > 0 small enough that B (y, 2) g () = . By Sards lemma, g (S) is a set of measure zero
e B (y, ) \ g (S) . Thus y
e is a regular value of g. Now define g1 (x) g (x) + ye
and so there exists y
y. It
e and so, since Dg (x) = Dg1 (x) , y is a regular value of g1 .
follows that g1 (x) = y if and only if g (x) = y
Then for t [0, 1] and x ,
|g (x) + t (g1 (x) g (x)) y| |g (x) y| t |ye
y| > 2 t > 0.
It follows g1 g and so g1 f . This proves the lemma since y is a regular value of g1 .


Definition 16.8 Let g Uy C 2 ; Rn and let y be a regular value of g. Then
X

d (g, , y)
sign (det Dg (x)) : x g1 (y) ,
d (g, , y) = 0 if g1 (y) = ,

(16.3)
(16.4)

280

BROUWER DEGREE

and if f Uy ,
d (f , , y) = d (g, , y)

(16.5)

n

where g [f ] C 2 ; R , and y is a regular value of g. This number, denoted by d (f , , y) is called the

degree or Brouwer degree.
We need to verify this is well defined. We begin with the definition, (16.3). We need to show that the
sum is finite.
Lemma 16.9 When y is a regular value, the sum in (16.3) is finite.
Proof: This follows from the inverse function theorem because g1 (y) is a closed, hence compact subset
of due to the assumption that y
/ g () . Since y is a regular value, it follows that det (Dg (x)) 6= 0 for
each x g1 (y) . By the inverse function theorem, there is an open set, Ux , containing x such that g is
one to one on this open set. Since g1 (y) is compact, this means there are finitely many sets, Ux which
cover g1 (y) , each containing only one point of g1 (y) . Therefore, this set is finite and so the sum is well
defined.
A much more difficult problem
is to show there are no contradictions in the two ways d (f , , y) is defined
in the case when f C 2 ; Rn and y is a regular value of f . We need to verify that if g0 g1 for


gi C 2 ; Rn and y a regular value of gi , it follows that d (g1 , , y) = d (g2 , , y) under the conditions of
(16.3) and (16.4). To aid in this, we give the following lemma.
Lemma 16.10 Suppose k l. Then there exists a sequence of functions of Uy ,
m

{gi }i=1 ,


such that gi C 2 ; Rn , y is a regular value for gi , and defining g0 k and gm+1 l, there exists > 0
such that for i = 1, , m + 1,
B (y, ) (tgi + (1 t) gi1 ) () = , for all t [0, 1] .

(16.6)

Proof: Let h : [0, 1] Rn be a function which shows k and l are equivalent. Now let 0 = t0 < t1 <
< tm = 1 be such that
||h (, ti ) h (, ti1 )|| <

(16.7)

where > 0 is small enough that

B (y, 8) h ( [0, 1]) = .


Now for i {1, , m} , let gi Uy C 2 ; Rn be such that

(16.8)

||gi h (, ti )|| < .

(16.9)


This is possible because C 2 ; R is dense in C ; Rn from Corollary 16.3. If y is a regular value for gi ,
e be a regular value of gi close enough to y that
leave gi unchanged. Otherwise, using Sards lemma, let y
e also satisfies (16.9). Then gei (x) = y if and only if gi (x) = y
e . Thus y is a
the function, gei gi + y y
regular value for gei and we may replace gi with gei in (16.9). Therefore, we can assume that y is a regular
value for gi in (16.9). Now from this construction,


n

||gi gi1 || ||gi h (, ti )||

+ ||h (, ti ) h (, ti1 )|| + ||gi1 h (, ti1 )|| < 3.

16.2. DEFINITIONS AND ELEMENTARY PROPERTIES

281

Now we verify (16.6). We just showed that for each x ,

gi (x) B (gi1 (x) , 3)
and we also know from (16.8) and (16.9) that for any j,
|gj (x) y|

|gj (x) h (x, tj )| + |h (x, tj ) y|

8 = 7.

Therefore, for x ,
|tgi (x) + (1 t) gi1 (x) y| = |gi1 (x) + t (gi (x) gi1 (x)) y|
7 t |gi (x) gi1 (x)| > 7 t3 4 > .
This proves the lemma.
We make the following definition of a set of functions.
Definition 16.11 For each > 0, let

Cc

(B (0, )) , 0,

dx = 1.

Lemma 16.12 Let g Uy C 2 ; Rn and let y be a regular value of g. Then according to the definition
of degree given in (16.3) and (16.4),
Z
d (g, , y) =
(g (x) y) det Dg (x) dx
(16.10)

whenever is small enough. Also y + v is a regular value of g whenever |v| is small enough.
m

Proof: Let the points in g1 (y) be {xi }i=1 . By the inverse function theorem, there exist disjoint open
sets, Ui , 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 be small enough that
B (y, ) m
i=1 g (Ui )
and let Vi g1 (B (y, )) Ui . Therefore, for any this small,
Z

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

m Z
X
i=1

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

Vi

The reason for this is as follows. The integrand is nonzero only if g (x) y B (0, ) which occurs only
if g (x) B (y, ) which is the same as x g1 (B (y, )). Therefore, the integrand is nonzero only if x is
contained in exactly one of the sets, Vi . Now using the change of variables theorem,
=

m Z
X
i=1

g(Vi )y




(z) det Dg g1 (y + z) det Dg1 (y + z) dz

By the chain rule, I = Dg g1 (y + z) Dg1 (y + z) and so






det Dg g1 (y + z) det Dg1 (y + z) = sign det Dg g1 (y + z)

282

BROUWER DEGREE
= sign (det Dg (x)) = sign (det Dg (xi )) .

Therefore, this reduces to

m
X

sign (det Dg (xi ))

(z) dz =

g(Vi )y

i=1

m
X

sign (det Dg (xi ))

(z) dz =

B(0,)

i=1

m
X

sign (det Dg (xi )) .

i=1

In case g1 (y) = , there exists > 0 such that g B (y, ) = and so for this small,
Z
(g (x) y) det Dg (x) dx = 0.

The last assertion of the lemma follows from the observation that g (S) is a compact set and so its
complement is an open set. This proves the lemma.
Now we are ready to prove a lemma which will complete the task of showing the above definition of the
degree is well defined. In the following lemma, and elsewhere, a comma followed by an index indicates the
f
partial derivative with respect to the indicated variable. Thus, f,j will mean x
.
j


Lemma 16.13 Suppose f , g are two functions in C 2 ; Rn and
B (y, ) ((1 t) f + tg) () =
for all t [0, 1] . Then
Z

(f (x) y) det (Df (x)) dx =

(g (x) y) det (Dg (x)) dx.

(16.11)

(16.12)

Proof: Define for t [0, 1] ,

H (t)

(f y + t (g f )) det (D (f + t (g f ))) dx.

Then if t (0, 1) ,
H 0 (t) =

Z X

, (f (x) y + t (g (x) f (x)))

(g (x) f (x)) det D (f + t (g f )) dx

+

(f y + t (g f ))

det D (f + t (g f )),j (g f ),j dx A + B.

,j

In this formula, the function det is considered as a function of the n2 entries in the n n matrix. Now as in
the proof of Lemma 16.4,
det D (f + t (g f )),j = (cof D (f +t (g f )))j

16.2. DEFINITIONS AND ELEMENTARY PROPERTIES

283

and so
B=

Z XX

(f y + t (g f ))

(cof D (f +t (g f )))j (g f ),j dx.

By hypothesis
x (f (x) y + t (g (x) f (x))) (cof D (f (x) +t (g (x) f (x))))j
is in Cc1 () because if x , it follows by (16.11) that
f (x) y + t (g (x) f (x))
/ B (0, ) .
Therefore, we may integrate by parts and write
Z XX

( (f y + t (g f )))
B=
x
j

j
(cof D (f +t (g f )))j (g f ) dx +

Z XX

(f y + t (g f )) (cof D (f +t (g f )))j,j (g f ) dx.

The second term equals zero by Lemma 16.4. Simplifying the first term yields
Z XXX
B=
, (f y + t (g f ))

(f,j + t (g,j f,j )) (cof D (f +t (g f )))j (g f ) dx

=

Z XX

Z X

, (f y + t (g f )) det (D (f +t (g f ))) (g f ) dx

, (f y + t (g f )) det (D (f +t (g f ))) (g f ) dx = A.

Therefore, H 0 (t) = 0 and so H is a constant. This proves the lemma.

Now we are ready to prove that the Brouwer degree is well defined.
Proposition 16.14 Definition 16.8 is well defined.


Proof: We only need to verify that for k, l Uy k l, k, l C 2 ; Rn , and y a regular value of k, l,
d (k, , y) = d (l, , y) as given in the first part of Definition 16.8. Let the functions, gi , i = 1, , m be as
described in Lemma 16.10. By Lemma 16.8 we may take > 0 small enough that equation (16.10) holds for
g = k, l. Then by Lemma 16.13 and letting g0 k, and gm+1 = l,
Z
Z
(gi (x) y) det Dgi (x) dx =
(gi1 (x) y) det Dgi1 (x) dx

for i = 1, , m + 1. In particular d (k, , y) = d (l, , y) proving the proposition.

The degree has many wonderful properties. We begin with a simple technical lemma which will allow us
to establish them.

284

BROUWER DEGREE

Lemma 16.15 Let y1

/ f () . Then d (f , , y1 ) = d (f , , y) whenever y is close enough to y1 . Also,
d (f , , y) equals an integer.
Proof: It follows immediately from the definition of the degree that
d (f , , y)


is an integer. Let g C 2 ; Rn for which y1 is a regular value and let h : [0, 1] Rn be a continuous
function for which h (x, 0) = f (x) and h (x, 1) = g (x) such that h (, t) does not contain y1 for any
t [0, 1] . Then let 1 be small enough that
B (y1 , 1 ) h ( [0, 1]) = .
From Lemma 16.12, we may take < 1 small enough that whenever |v| < , y1 + v is a regular value of g
and
X

d (g, , y1 ) =
sign Dg (x) : x g1 (y1 )
=

X

sign Dg (x) : x g1 (y1 + v) = d (g, , y1 + v) .

The second equal sign above needs some justification. We know g1 (y1 ) = {x1 , , xm } and by the inverse
function theorem, there are open sets, Ui such that xi Ui and g is one to one on Ui having an inverse
on the open set g (Ui ) which is also C 2 . We want to say that for |v| small enough, g1 (y1 + v) m
j=1 Ui .
If not, there exists vk 0 and
zk g1 (y1 + vk ) \ m
j=1 Ui .
But then, taking a subsequence, still denoted by zk we could have zk z
/ m
j=1 Ui and so
g (z) = lim g (zk ) = lim (y1 + vk ) = y1 ,
k

contradicting the fact that g1 (y1 ) m

j=1 Ui . This justifies the second equal sign.
For the above homotopy of f and g, if x ,
|h (x, t) (y1 + v)| |h (x, t) y1 | |v| > 1 > 0.
Therefore, by the definition of the degree,
d (f , , y) = d (g, , y) = d (g, , y1 +v) = d (f , , y1 +v) .
This proves the lemma.
Here are some important properties of the degree.
Theorem 16.16 The degree satisfies the following properties. In what follows, id (x) = x.
1. d (id, , y) = 1 if y .


2. If i , i open, and 1 2 = and if y
/ f \ (1 2 ) , then d (f , 1 , y) + d (f , 2 , y) =
d (f , , y) .


3. If y
/ f \ 1 and 1 is an open subset of , then
d (f , , y) = d (f , 1 , y) .

16.2. DEFINITIONS AND ELEMENTARY PROPERTIES

285

4. d (f , , y) 6= 0 implies f 1 (y) 6= .
5. If f , g are homotopic with a homotopy, h : [0, 1] for which h (, t) does not contain y, then
d (g,, y) = d (f , , y) .
6. d (, , y) is defined and constant on




g C ; Rn : ||g f || < r
where r = dist (y, f ()) .
7. d (f , , ) is constant on every connected component of Rn \ f () .
8. d (g,, y) = d (f , , y) if g| = f | .
Proof: The first property follows immediately from the definition of the degree.
To obtain the second property, let be small enough that


B (y, 3) f \ (1 2 ) = .


Next pick g C 2 ; Rn such that ||f g|| < . Letting y1 be a regular value of g with |y1 y| < ,
and defining g1 (x) g (x) + y y1 , we see that ||g g1 || < and y is a regular value of g1 . Then if
x \ (1 2 ) , it follows that for t [0, 1] ,
|f (x) + t (g1 (x) f (x)) y| |f (x) y| t ||g1 f || > 3 2 > 0


and so g1 \ (1 2 ) does not contain y. Hence g11 (y) 1 2 and g1 f . Therefore, from the
definition of the degree,
d (f , , y) d (g, , y) = d (g, 1 , y) + d (g, 2 , y)
d (f , 1 , y) + d (f , 2 , y)
The third property follows from the second if we let 2 = . In the above formula, d (g, 2 , y) = 0 in this
case.


Now consider the fourth property. If f 1 (y) = , then for > 0 small enough, B (y, 3) f = . Let
g be in C 2 and ||f g|| < with y a regular point of g. Then d (f , , y) = d (g, , y) and g1 (y) = so
d (g, , y) = 0.


From the definition of degree, there exists k C 2 ; Rn for which y is a regular point which is
homotopic to g and d (g,, y) = d (k,, y). But the property of being homotopic is an equivalence relation
and so from the definition of degree again, d (k,, y) = d (f ,, y) . This verifies the fifth property.
The sixth property follows from the fifth. Let the homotopy be
h (x, t) f (x) + t (g (x) f (x))
for t [0, 1] . Then for x ,
|h (x, t) y| |f (x) y| t ||g f || > r r = 0.
The seventh property follows from Lemma 16.15. The connected components are open connected sets.
This lemma implies d (f , , ) is continuous. However, this function is also integer valued. If it is not constant
on K, a connected component, there exists a number, r such that the values of this function are contained
in (, r) (r, ) with the function having values in both of these disjoint open sets. But then we could
consider the open sets A {z K : d (f , , z) > r} and B {z K : d (f , , z) < r}. Now K = A B and
we see that K is not connected.
The last property results from the homotopy
h (x, t) = f (x) + t (g (x) f (x)) .
Since g = f on , it follows that h (, t) does not contain y and so the conclusion follows from property
5.

286

BROUWER DEGREE

Definition 16.17 We say that a bounded open set, is symmetric if = . We say a continuous
function, f : Rn is odd if f (x) = f (x) .


Suppose is symmetric and g C 2 ; Rn is an odd map for which 0 is a regular point. Then the
chain rule implies Dg (x) = Dg (x) and so d (g, , 0) must equal an odd integer because if x g1 (0) , it
follows that x g1 (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. We also have 0 g1 (0) and so we have that the degree equals an
even integer added to sign (det Dg (0)) , an odd integer. It seems reasonable to expect that this would hold
for an arbitrary continuous odd function defined on symmetric . In fact this is the case and we will show
this next. The following lemma is the key result used.




Lemma 16.18 Let g C 2 ; Rn be an odd map. Then for every > 0, there exists h C 2 ; Rn
such that h is also an odd map, ||h g|| < , and 0 is a regular point of h.
Proof: Let h0 (x) = g (x) + x where is chosen such that det Dh0 (0) 6= 0 and < 2 . Now let
i {x : xi 6= 0} . Define h1 (x) h0 (x) y1 x31 where |y1 | < and y1 is a regular value of the
function,
x
for x 1 . Thus h1 (x) = 0 if and only if y1 =
det

det

h0 (x)
x31

h0 (x)
.
x31

Since y1 is a regular value,

!

x31 h0i (x)

h0i,j (x) x31 x
j
=
x61

xj
x61

h0i,j (x) x31

!
x31 y1i x31

6= 0

implying that



det h0i,j (x)

x31 y1i = det (Dh1 (x)) 6= 0.
xj
We have shown that 0 is a regular value of h1 on the set 1 . Now we define h2 (x) h1 (x) y2 x32 where
|y2 | < and y2 is a regular value of
x

h1 (x)
x32

for x 2 . Thus, as in the step going from h0 to h1 , for such x h1

2 (0) ,



det h1i,j (x)

x3 y2i = det (Dh2 (x)) 6= 0.
xj 2
1
Actually, det (Dh2 (x)) 6= 0 for x (1 2 ) h1
2 (0) because if x (1 \ 2 ) h2 (0) , then x2 = 0.
From the above formula for det (Dh2 (x)) , we see that in this case,

det (Dh2 (x)) = det (Dh1 (x)) = det (h1i,j (x)) .

We continue in this way, finding a sequence of odd functions, hi where
hi+1 (x) = hi (x) yi x3i
for |yi | < and 0 a regular value of hi on ij=1 j . Let hn h. Then 0 is a regular value of h for x nj=1 j .
The point of which is not in nj=1 j is 0. If x = 0, then from the construction, Dh (0) = Dh0 (0) and so
0 is a regular value of h for x . By choosing small enough, we see that ||h g|| < . This proves the
lemma.

16.3. APPLICATIONS

287

Theorem 16.19 (Borsuk) Let f C

d (f , , 0) equals an odd integer.

; Rn

be odd and let be symmetric with 0

/ f (). Then

Proof: Let be small enough that B (0,3) f () = . Let

g1 C 2 ; Rn
be such that ||f g1 || < and let g denote the odd part of g1 . Thus
g (x)

1
(g1 (x) g1 (x)) .
2

Then since f is odd, it follows that ||f g|| < also. By Lemma 16.18 there exists odd h C 2 ; Rn
for which 0 is a regular value and ||h g|| < . Therefore, ||f h|| < 2 and from Theorem 16.16
m
1
d (f , , 0) = d (h, , 0) . However, since 0 is a regular
= {xi , xi , 0}i=1 , and since h
Pm point of h, h (0) P
m
is odd, Dh (xi ) = Dh (xi ) and so d (h, , 0) i=1 sign det (Dh (xi )) + i=1 sign det (Dh (xi )) + sign
det (Dh (0)) , an odd integer.

16.3

Applications

With these theorems it is possible to give easy proofs of some very important and difficult theorems.
Definition 16.20 If f : U Rn Rn , we say f is locally one to one if for every x U, there exists > 0
such that f is one to one on B (x, ) .
To begin with we consider the Invariance of domain theorem.
Theorem 16.21 (Invariance of domain)Let be any open subset of Rn and let f : Rn be continuous
and locally one to one. Then f maps open subsets of to open sets in Rn .
Proof: Suppose not. Then there exists an open set, U with U open but f (U ) is not open . This
means that there exists y0 , not an interior point of f (U ) where y0 = f (x0 ) for x0 U. Let B (x0 , r) U be
such that f is one to one on B (x0 , r).
Let e
f (x) f (x + x0 ) y0 . Then e
f : B (0,r) Rn , e
f (0) = 0. If x B (0,r) and t [0, 1] , then




x
tx
e
f
e
f
6= 0
(16.13)
1+t
1+t
because if this quantity were to equal zero, then since e
f is one to one on B (0,r), it would follow that
x
tx
=
1+t
1+t





x
which is not so unless x = 0
/ B (0,r) . Let h (x, t) e
f 1+t
e
f tx
1+t . Then we just showed that
0
/ h (, t) for all t [0, 1] . By Borsuks theorem, d (h (1, ) , B (0,r) , 0) equals an odd integer. Also by
part 5 of Theorem 16.16, the homotopy invariance assertion,


d (h (1, ) , B (0,r) , 0) = d (h (0, ) , B (0,r) , 0) = d e
f , B (0,r) , 0 .
Now from the case where t = 0 in (16.13), there exists > 0 such that B (0, ) e
f (B (0,r)) = . Therefore,
B (0, ) is a subset of a component of Rn \ e
f (B (0, r)) and so




d e
f , B (0,r) , z = d e
f , B (0,r) , 0 6= 0

288

BROUWER DEGREE

for all z B (0,) . It follows that for all z B (0, ) , e

f 1 (z) B (0,r) 6= . In terms of f , this says that for
all |z| < , there exists x B (0,r) such that
f (x + x0 ) y0 = z.
In other words, f (U ) f (B (x0 , r)) B (y0 , ) showing that y0 is an interior point of f (U ) after all. This
proves the theorem..
Corollary 16.22 If n > m there does not exist a continuous one to one map from Rn to Rm .
T

Proof: Suppose not and let f be such a continuous map, f (x) (f1 (x) , , fm (x)) . Then let g (x)
T
(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 16.23 If f is locally one to one, 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. 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 , we see that 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 16.24 (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. Therefore, by the homotopy invariance,
n

0 = d (f id, B, 0) = d (id, B, 0) = (1) ,

a contradiction.
Corollary 16.25 (Brouwer fixed point) Let K be any convex compact set in Rn and let f : K K be
continuous. Then f has a fixed point.
Proof: Let B B (0, R) where R is large enough that B K, and let P denote the projection map
onto K. Let g : B B be defined as g (x) f (P (x)) . Then g is continuous and so by Theorem 16.24 it
has a fixed point, x. But x K and so x = f (P (x)) = f (x) .


Definition 16.26 We say 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 16.27 There does not exist a retraction of B (0, r)

onto B (0, r) .

Proof: Suppose f were such a retraction. Then for all x , f (x) = x and so from the properties of
the degree,
1 = d (id, , 0) = d (f , , 0)

16.4. THE PRODUCT FORMULA AND JORDAN SEPARATION THEOREM

289

which is clearly impossible because f 1 (0) = which implies d (f , , 0) = 0.

In the next two theorems we make use of the Tietze extension theorem which states that in a metric
space (more generally a normal topological space) every continuous function defined on a closed subset of
the space having values in [a, b] may be extended to a continuous function defined on the whole space having
values in [a, b]. For a discussion of this important theorem and an outline of its proof see Problems 9 - 11 of
Chapter 4.
Theorem 16.28 Let be a symmetric open set in Rn such that 0 and let f : V be continuous
where V is an m dimensional subspace of Rn , m < n. Then f (x) = f (x) for some x .


Proof: Suppose not. Using the Tietze extension theorem, extend f to all of , f V. Let g (x) =
f (x) f (x) . Then 0
/ g () and so for some r > 0, B (0,r) Rn \ g () . For z B (0,r) ,
d (g, , z) = d (g, , 0) 6= 0
because B (0,r) is contained in a component of Rn \ g () and Borsuks theorem implies that d (g, , 0) 6= 0
since g is odd. Hence
V g () B (0,r)
and this is a contradiction because V is m dimensional. This proves the theorem.
This theorem is called the Borsuk Ulam theorem. Note that it implies that there exist two points on
opposite sides of the surface of the earth that have the same atmospheric pressure and temperature. The
next theorem is an amusing result which is like combing hair. It gives the existence of a cowlick.
Theorem 16.29 Let n be odd and let be an open bounded set in Rn with 0 . Suppose f : Rn \{0}
is continuous. Then for some x and 6= 0, f (x) = x.
Proof: Using the Tietze extension theorem, extend f to all of . Suppose for all x , f (x) 6= x for
all R. Then
0
/ tf (x) + (1 t) x, (x, t) [0, 1]
0
/ tf (x) (1 t) x, (x, t) [0, 1] .
Then by the homotopy invariance of degree,
d (f , , 0) = d (id, , 0) , d (f , , 0) = d (id, , 0) .
n

But this is impossible because d (id, , 0) = 1 but d (id, , 0) = (1) . This proves the theorem.

16.4

The Product formula and Jordan separation theorem

In this section we present the product formula for the degree and use it to prove a very important theorem
in topology. To begin with we give the following lemma.






Lemma 16.30 Let y1 , , yr be points not in f () . Then there exists e
f C 2 ; Rn such that e
f f 
and yi is a regular value for e
f for each i.

<

290

BROUWER DEGREE

e1 such that y
e1
Proof: Let f0 C 2 ; Rn , ||f0 f || < 2 . For S0 the singular set of f0 , pick y
/

e1 . Thus y1 is a regular
f0 (S0 ) .(e
y1 is a regular value of f0 ) and |e
y1 y1 | < 3r . Let f1 (x) f0 (x) + y1 y
value of f1 and
||f f 1 || ||f f 0 || + ||f0 f1 || <
e2 such that |e
Letting S1 be the singular set of f1 , choose y
y2 y2 | <

+ .
3r 2

3r

and

e2
y
/ f1 (S1 ) (f1 (S1 ) + y2 y1 ) .

e2 . Thus if f2 (x) = y1 , then

Let f2 (x) f2 (x) + y2 y

e2
f1 (x) + y2 y1 = y

and so x
/ S1 . Thus y1 is a regular value of f2 . If f2 (x) = y2 , then
e2
y2 = f1 (x) + y2 y

e2 implying x
and so f1 (x) = y
/ S1 showing det Df2 (x) = det Df1 (x) 6= 0. Thus y1 and y2 are both regular
values of f2 and
||f2 f || ||f2 f1 || + ||f1 f ||

+
+ .
3r 3r 2




We continue in this way. Let e
f fr . Then e
f f  < 2 + 3 < and each yi is a regular value of e
f.
<

Definition 16.31 Let the connected components of Rn \ f () be denoted by Ki . We know from the properties of the degree that d (f , , ) is constant on each of these components. We will denote by d (f , , Ki )
the constant value on the component, Ki .


Lemma 16.32 Let f C ; Rn , g C 2 (Rn , Rn ) , and y
/ g (f ()) . Suppose also that y is a regular
value of g. Then we have the following product formula where Ki are the bounded components of Rn \ f () .
d (g f , , y) =

d (f , , Ki ) d (g, Ki , y) .

i=1

Proof: First note that if Ki is unbounded,

d (f , , Ki ) = 0 because there exists a point, z Ki such


that f 1 (z) = due to the fact that f is compact and is consequently bounded. Thus it makes no
difference in the above formula whether we let Ki be general components or insist on their being bounded.
 ki
Let xij j=1 denote the point of g1 (y) which are contained in Ki , the ith component of Rn \ f () . Note


also that g1 (y) f is a compact set covered by the components of Rn \ f () and so it is covered by
finitely many of these components. For the other components, d (f , , Ki ) = 0 and so this is actually a finite
sum. There are no convergence problems. Now let > 0 be small enough that
B (y, 3) g (f ()) = ,
and for each xij g1 (y)


B xij , 3 f () = .

16.4. THE PRODUCT FORMULA AND JORDAN SEPARATION THEOREM

291

Next choose > 0 small enough that < and if z1 and z2 are any two points of f with |z1 z2 | < ,
it follows that |g (z1 ) g (z2 )| < .






Now choose e
f C 2 ; Rn such that e
f f  < and each point, xij is a regular value of e
f . From

i
the properties of the degree we know that d (f , , Ki ) = d f , , xj for each j = 1, , mi . For x , and
t [0, 1] ,






f (x) f (x) xij  > 3 t > 0
f (x) + t e

and so





d e
f , , xij = d f , , xij = d (f , , Ki )

(16.14)

independent of j. Also for x , and t [0, 1] ,



 




f (x) g (f (x)) y 3 t > 0
g (f (x)) + t g e

and so



d (g f , , y) = d ge
f , , y .
Now let

(16.15)

n okij




uij
be the points of e
f 1 xij . Therefore, kij < because e
f 1 xij , a bounded open
l
l=1

set. It follows from (16.15) that



d (g f , , y) = d ge
f , , y

kij
mi X
X
X
i=1 j=1 l=1

mi
X
X

det Dg

  
 
sgn det Dg e
f uij
sgn det De
f uij
l
l

xij

i=1 j=1

 X




i
e
d f , , xj =
d (g, Ki , y) d e
f , , xij
i=1

d (g, Ki , y) d (f , , Ki ) .

i=1

With this lemma we are ready to prove the product formula.

Theorem 16.33 (product formula) Let {Ki }i=1 be the bounded components of Rn \f () for f C ; Rn ,
let g C (Rn , Rn ) , and suppose that y
/ g (f ()). Then
d (g f , , y) =

d (g, Ki , y) d (f , , Ki ) .

i=1

e C 2 (Rn , Rn ) be such that

Proof: Let B (y,3) g (f ()) = and let g



sup |e
g (z) g (z)| : z f <

(16.16)

292

BROUWER DEGREE

e. Then from the above inequality, if x and t [0, 1] ,

And also y is a regular value of g
|g (f (x)) + t (e
g (f (x)) g (f (x))) y| 3 t > 0.

It follows that
d (g f , , y) = d (e
g f , , y) .

(16.17)

Now also, Ki f () and so if z Ki , then g (z) g (f ()) . Consequently, for such z,

|g (z) + t (e
g (z) g (z)) y| |g (z) y| t > 3 t > 0
which shows that
d (g, Ki , y) = d (e
g, Ki , y) .

(16.18)

Therefore, by Lemma 16.32,

d (g f , , y) = d (e
g f , , y) =

d (e
g, Ki , y) d (f , , Ki )

i=1

d (g, Ki , y) d (f , , Ki ) .

i=1

This proves the product formula. Note there are no convergence

problems because these sums are actually
finite sums because, as in the previous lemma, g1 (y) f is a compact set covered by the components
of Rn \ f () and so it is covered by finitely many of these components. For the other components,
d (f , , Ki ) = 0.
With the product formula is possible to give a fairly easy proof of the Jordan separation theorem, a very
profound result in the topology of Rn .
Theorem 16.34 (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 let f be an extension of f to all of Rn and let f 1 denote an extension of f 1 to all of Rn .
Pick K K and take y K. Let H denote the set of bounded components of Rn \ f (K) (note K C).
Since f 1 f equals the identity, id, on K, it follows that


1 = d (id, K, y) = d f 1 f , K, y .
By the product formula,



X


1 = d f 1 f , K, y =
d f , K, H d f 1 , H, y ,
HH

the sum being a finite sum. 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. Letting GH denote those sets of L which are contained in H, we note that
H \ GH f (C) .

16.5. INTEGRATION AND THE DEGREE

293

This is because if z
/ GH , then z cannot be contained in any set of L which has nonempty intersection
with H since then, that whole set of L would be contained in H due to the fact that the sets of H are
disjoint open set and the sets of L are connected. It follows that z is either an element of f (C) which would
correspond to being contained in none of the sets of L, or else, z is contained in some set of L which has
empty intersection with H. But the sets of L are open and so this point, z cannot, in this latter case, be
contained in H. Therefore, the
above inclusion is verified.
Claim: y
/ f 1 H \ GH .
Proof of the claim: If not, then f 1 (z) = y where z H \ GH f (C) and so f 1 (z) = y C. But
y
/ C and this contradiction proves the claim.
Now every set of L is contained in some
H. What about those sets of H which contain no set of L?
 set of 


1
1
From the claim, y
/ f
H and so d f , H, y = 0. Therefore, letting H1 denote those sets of H which
contain some set of L, properties of the degree imply


X
X



X 
1=
d f , K, H d f 1 , H, y =
d f , K, H
d f 1 , L, y
HH1

HH1

X X

HH1 LGH

LL

LGH

 X





d f , K, L d f 1 , L, y =
d f , K, L d f 1 , L, y
LL

 X





d f , K, L d f 1 , L, y =
d f , K, L d f 1 , L, K
LL

and each sum is finite. Letting |K| denote the number of elements in K,

X X


|K| =
d f , K, L d f 1 , L, K .
KK LL

By symmetry, we may use the above argument to write

|L| =

X X

LL KK




d f , K, L d f 1 , L, K .

It follows |K| = |L| and this proves the theorem because if n > 1 there is exactly one unbounded component
and if n = 1 there are exactly two.

16.5

Integration and the degree

There is a very interesting application of the degree to integration. Recall Lemma 16.12. We will use
Theorem 16.16 to generalize this lemma next. In this proposition, we let be the mollifier of Definition
16.11.


Proposition 16.35 Let g Uy C 1 ; Rn then whenever > 0 is small enough,
Z
d (g, , y) =
(g (x) y) det Dg (x) dx.

Proof: Let 0 > 0 be small enough that

B (y, 30 ) g () = .

294

BROUWER DEGREE

Now let m be a mollifier and let

gm g m .


Thus gm C ; Rn and
||gm g|| , ||Dgm Dg|| 0

(16.19)

as m . Choose M such that for m M,

||gm g|| < 0 .

(16.20)

Thus gm Uy C 2 ; Rn
Letting z B (y, ) for < 0 , and x ,
|(1 t) gm (x) + gk (x) t z| (1 t) |gm (x) z| + t |gk (x) z|
> (1 t) |g (x) z| + t |g (x) z| 0
= |g (x) z| 0
|g (x) y| |y z| 0
> 30 0 0 = 0 > 0
showing that B (y, ) ((1 t) gm + tgk ) () = . By Lemma 16.13
Z
(gm (x) y) det (Dgm (x)) dx =

(gk (x) y) det (Dgk (x)) dx

(16.21)

for all k, m M.
em
We may assume that y is a regular value of gm since if it is not, we will simply replace gm with g
defined by
em (x) (ye
gm (x) g
y)

e is a regular value of g chosen close enough to y such that (16.20) holds for g
em in place of gm . Thus
where y
em (x) = y if and only if gm (x) = y
e and so y is a regular value of g
em . Therefore,
g
Z
d (y,, gm ) =
(gm (x) y) det (Dgm (x)) dx
(16.22)

for all small enough by Lemma 16.12. For x , and t [0, 1] ,

|(1 t) g (x) + tgm (x) y| (1 t) |g (x) y| + t |gm (x) y|
(1 t) |g (x) y| + t |g (x) y| 0
> 30 0 > 0
and so by Theorem 16.16, and (16.22), we can write
d (y,, g) = d (y,, gm ) =

16.5. INTEGRATION AND THE DEGREE

Z

295

(gm (x) y) det (Dgm (x)) dx

whenever is small enough. Fix such an < 0 and use (16.21) to conclude the right side of the above
equations is independent of m > M. Then let m and use (16.19) to take a limit as m and
conclude
Z
d (y,, g) = lim
(gm (x) y) det (Dgm (x)) dx
m
Z
=
(g (x) y) det (Dg (x)) dx.

This proves the proposition.

With this proposition, we are ready to present the interesting change of variables

theorem. Let U be a
bounded open set with the property that U has measure zero and let f C 1 U ; Rn . Then Theorem 11.13
implies that f (U ) also has measure zero. From Proposition 16.35 we see that for y
/ f (U ) ,
Z
d (y, U , f ) = lim
(f (x) y) det Df (x) dx,
0

showing that y d (y, U , f ) is a measurable function. Also,

Z
y
(f (x) y) det Df (x) dx

(16.23)

is a function bounded independent of because det Df (x) is bounded and the integral of equals one.
Letting h Cc (Rn ), we can therefore, apply the dominated convergence theorem and the above observation
that f (U ) has measure zero to write
Z
Z
Z
h (y) d (y, U , f ) dy = lim h (y)
(f (x) y) det Df (x) dxdy.
0

Now we will assume for convenience that has the following simple form.
(x)

1 x

R
where the support of is contained in B (0,1), (x) 0, and (x) dx = 1. Therefore, interchanging the
order of integration in the above,
Z
Z
Z
h (y) d (y, U , f ) dy = lim
det Df (x) h (y) (f (x) y) dydx
0 U
Z
Z
= lim
det Df (x)
h (f (x) u) (u) dudx.
0

B(0,1)

Now
Z

Z
Z




h (f (x) u) (u) dudx
det Df (x) h (f (x)) dx =
 det Df (x)
 U

B(0,1)
U
Z
Z


h (f (x) u) (u) dudx
 det Df (x)
 U
B(0,1)

296

BROUWER DEGREE



det Df (x)
h (f (x)) (u) dudx

U
B(0,1)


Z
Z




det Df (x) |h (f (x) u) h (f (x))| dx (u) du

 B(0,1) U


By the uniform continuity of h we see this converges to zero as 0. Consequently,

Z
Z
Z
h (y) d (y, U , f ) dy = lim
det Df (x) h (y) (f (x) y) dydx
0 U
Z
=
det Df (x) h (f (x)) dx
U

which proves the following lemma.

Lemma 16.36 Let h Cc (Rn ) and let f C 1 U ; Rn where U has measure zero for U a bounded open
set. Then everything is measurable which needs to be and we have the following formula.
Z
Z
h (y) d (y, U , f ) dy =
det Df (x) h (f (x)) dx.
U

Next we give a simple corollary which replaces Cc (Rn ) with L1 (Rn ) .

Corollary 16.37 Let h L (Rn ) and let f C 1 U ; Rn where U has measure zero for U a bounded
open set. Then everything is measurable which needs to be and we have the following formula.
Z
Z
h (y) d (y, U , f ) dy =
det Df (x) h (f (x)) dx.
U

Proof: For all y

/ f (U ) a set of measure zero, d (y, U , f ) is bounded by some constant which is
independent of y
/ U due to boundedness of the formula (16.23). The integrand of the integral on the
left equals zero off some bounded set because if y
/ f (U ) , d (y, U , f ) = 0. Therefore, we can modify h
off a bounded set and assume without loss of generality that h L (Rn ) L1 (Rn ) . Letting hk be a
sequence of functions of Cc (Rn ) which converges pointwise a.e. to h and in L1 (Rn ) , in such a way that
|hk (y)| ||h|| + 1 for all x,
|det Df (x) hk (f (x))| |det Df (x)| (||h|| + 1)
and
Z
U

|det Df (x)| (||h|| + 1) dx <

because U is bounded and h L (Rn ) . Therefore, we may apply the dominated convergence theorem to
the equation
Z
Z
hk (y) d (y, U , f ) dy =
det Df (x) hk (f (x)) dx
U

and obtain the desired result.

Differential forms
17.1

Manifolds

Manifolds are sets which resemble Rn locally. To make this more precise, we need some definitions.
Definition 17.1 Let X Y where (Y, ) is a topological space. Then we define the relative topology on X
to be the set of all intersections with X of sets from . We say these relatively open sets are open in X.
Similarly, we say a subset of X is closed in X if it is closed with respect to the relative topology on X.
We leave as an easy exercise the following lemma.
Lemma 17.2 Let X and Y be defined as in Definition 17.1. Then the relative topology defined there is a
topology for X. Furthermore, the sets closed in X consist of the intersections of closed sets from Y with X.
With the above lemma and definition, we are ready to define manifolds.
Definition 17.3 A closed and bounded subset of Rm , , will be called an n dimensional manifold with boundn
ary if there are finitely many sets, Ui , open in and continuous one
functions,
 to one
Ri : Ui p Ri Ui R
1
n
n
1
such that Ri and Ri both are continuous, Ri Ui is open in R u R : u 0 , and = i=1 Ui . These
mappings, Ri ,together with their domains, Ui , are called charts and the totality of all the charts, (Ui , Ri ) just
described
an atlas
i,Ri x Rn< } where

 is called
for the manifold. We also define int () {x n : for some
n
1
n
n
R< u R : u < 0 . We define {x : for some i, Ri x R0 } where R0 u Rn : u1 = 0
and we refer to as the boundary of .
This definition is a little too restrictive. In general we do not require the collection of sets, Ui to be
finite. However, in the case where is closed and bounded, we can always reduce to this because of the
compactness of and since this is the case of most interest to us here, the assumption that the collection of
sets, Ui , is finite is made.
Lemma 17.4 Let and int () be as defined above. Then int () is open in and is closed in .
Furthermore, int () = , = int () , and is an n 1 dimensional manifold for which
() = . In addition to this, the property of being in int () or does not depend on the choice of atlas.
Proof: It is clear that = int () . We show that int () = . Suppose this does not happen.
Then there would exist x int () . 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 R1
i , a continuous one to one map
from Rn to Rn having a continuous inverse. Choosing r > 0 small enough, we may obtain that
R1
i B (Ri x,r) Ui Uj .
n
n
Therefore, Rj R1
i (B (Ri x,r)) R and contains a point on R0 . However, this cannot occur because it
contradicts the theorem on invariance of domain, Theorem 16.21, which requires that Rj R1
i (B (Ri x,r))

297

298

DIFFERENTIAL FORMS

must be an open subset of Rn . Therefore, we have shown that int () = as claimed. This same
argument shows that the property of being in int () or does not depend on the choice of the atlas. To
verify that () = , let P1 : Rn Rn1 be defined by P1 (u1 , , un ) = (u2 , , un ) and consider the
n1
maps P1 Ri ke1 where k is large enough that the images of these maps are in Rn1
.
< . Here e1 refers to R
We now show that int () is open in . If x int () , then for some i, Ri x Rn< and so whenever, r > 0
is small enough, B (Ri x,r) Rn< and R1
i (B (Ri x,r)) is a set open in and contained in Ui . Therefore,
all the points of R1
(B
(R
x,r))
are
in
int () which shows that int () is open in as claimed. Now it
i
i
follows that is closed because = \ int () .


Definition 17.5 Let V Rn . We denote by C k V ; Rm the set of functions which are restrictions to V of
some function defined on Rn which has k continuous derivatives and compact support.
Definition 17.6 We will say an n
with boundary, is a C k manifold with boundary
 dimensional manifold



1
k
n
for some k 1 if Rj Ri C Ri (Ui Uj ); R
and R1
C k Ri Ui ; Rm . We say is orientable if
i
in addition to this there exists an atlas, (Ur , Rr ) , such that whenever Ui Uj 6= ,



det D Rj R1
(u) > 0
(17.1)
i
whenever u Ri (Ui Uj ) . The mappings, Ri R1
are called the overlap maps. We will refer to an atlas
j
satisfying (17.1) as an oriented atlas. We will also assume that if an oriented n manifold has nonempty
boundary, then n 2. Thus we are not defining the concept of an oriented one manifold with boundary.
The following lemma is immediate from the definitions.
Lemma 17.7 If is a C k oriented manifold with boundary, then is also a C k oriented manifold with
empty boundary.
Proof: We simply let an atlas consist of (Vr , Sr ) where Vr is the intersection of Ur with and Sr is of
the form
Sr (x) P1 Rr (x) kr e1
= (u2 kr , , un )
where P1 is defined above in Lemma 17.4, e1 refers to Rn1 and kr is large enough that for all x Vr , Sr (x)
Rn1
< .
When we refer to an oriented manifold, , we will always regard as an oriented manifold according
to the construction of Lemma 17.7.
The study of manifolds is really a generalization of something with which everyone 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. Remember, there are lots of different
coordinate systems which describe a point. There are spherical coordinates, cylindrical coordinates and
rectangular coordinates to name the three most popular coordinate systems. These coordinates are like
the vector u. The point, x is like the geometric point although we are always assuming x has rectangular
coordinates in Rm for some m. Under fairly general conditions, it has been shown there is no loss of generality
in making such an assumption and so we are doing so.

17.2

The integration of differential forms on manifolds

In this section we consider the integration of differential forms on manifolds. This topic is a generalization of
what you did in calculus when you found 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 generalization of this idea and it turns
out they are what it makes sense to integrate on manifolds. The following lemma, used in establishing the
definition of the degree and proved in that chapter is also the fundamental result in discussing the integration
of differential forms.

17.2. THE INTEGRATION OF DIFFERENTIAL FORMS ON MANIFOLDS

299

Lemma 17.8 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

where here (Dg)ij gi,j

gi
xj .

We will also need the following fundamental lemma on partitions of unity which is also discussed earlier,
Corollary 12.24.

Lemma 17.9 Let K be a compact set in Rn and let {Ui }i=1 be an open cover of K. Then there exists
functions, k Cc (Ui ) such that i Ui and

i (x) = 1.

i=1

The following lemma will also be used.

Lemma 17.10 Let {i1 , , in } {1, , m} and let R C 1 V ; Rm . Letting x = Ru, we define


xi1 xin
(u1 un )
to be the following determinant.

det

xi1
u1

..
.

xin
u1

xi1
un

..
.

xin
un

Then letting R1 C 1 W ; Rm and x = R1 v = Ru, we have the following formula.






xi1 xin
xi1 xin u1 un
=
.
(v 1 v n )
(u1 un ) (v 1 v n )
Proof: We define for I {i1 , , in } , the mapping PI : Rm span (ei1 , , ein ) by
i
x1
..
PI x .
xin

Thus


xi1 xin
= det (D (PI R) (u)) .
(u1 un )
since R1 (v) = R (u) = x,
PI R1 (v) = PI R (u)
and so the chain rule implies


D (PI R1 ) (v) = D (PI R) (u) D R1 R1 (v)

300

DIFFERENTIAL FORMS

and so


xi1 xin
= det (D (PI R1 ) (v)) =
(v 1 v n )




det (D (PI R) (u)) det D R1 R1 (v) =




xi1 xin u1 un
(u1 un ) (v 1 v n )
as claimed.
With these three lemmas, we first define what a differential form is and then describe how to integrate
one.
Definition 17.11 We will let I denote an ordered list of n indices from the set, {1, , m} . Thus I =
(i1 , , in ) . We say it is an ordered list because the order matters. Thus if we switch two indices, I would
be changed. A differential form of order n in Rm is a formal expression,
X
=
aI (x) dxI
I

where aI is at least Borel measurable or continuous if you wish 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 an orientable n
dimensional manifold with boundary, we define
Z

(17.2)

according to the following procedure. We let (Ui , Ri ) be an oriented atlas for . Each Ui is the intersection
of an open set in Rm , with and so there exists a C partition of unity subordinate
to the open cover, {Oi }
P
which sums to 1 on . Thus i Cc (Oi ) , has values in [0, 1] and satisfies i i (x) = 1 for all x . We
call this a partition of unity subordinate to {Ui } in this context. Then we define (17.2) by


Z
p XZ
X



xi1 xin
1
1

i Ri (u) aI Ri (u)
du
(17.3)
(u1 un )

Ri Ui
i=1
I

Of course there are all sorts of questions related to whether this definition is well defined. The formula
(17.2) makes no mention of partitions of unity or a particular atlas.
R What if we picked a different atlas and a
different partition of unity? Would we get the same number for ? In general, the answer is no. However,
there is a sense in which (17.2) is well defined. This involves the concept of orientation.
Definition 17.12 Suppose is an n dimensional C k orientable manifold with boundary and let (Ui , Ri )
and (Vi , Si ) be two oriented atlass of . We say they have the same orientation if whenever Ui Vj 6= ,




det D Ri S1
(v) > 0
(17.4)
j
for all v Sj (Ui Vj ) .




Note that by the chain rule, (17.4) is equivalent to saying det D Sj R1
(u) > 0 for all u
i
Ri (Ui Vj ) .
Now we are ready to discuss the manner in which (17.2) is well defined.

17.2. THE INTEGRATION OF DIFFERENTIAL FORMS ON MANIFOLDS

301

Theorem 17.13 Suppose is an n dimensional C k orientable manifold with boundary and let (U
R i , Ri )
and (Vi , Si ) be two oriented atlass of . Suppose the two atlass have the same orientation. Then if is
computed with respect to the two atlass the same number is obtained.
Proof: In Definition 17.11 let { i } be a partition of unity as described there which is associated with
the atlas (Ui , Ri ) and let { i } be a partition of unity associated
in the same manner with the atlas (Vi , Si ) .
Then since the orientations are the same, letting u = Ri S1
v,
j
S1
j

det D Ri

u1 un
(v)
>0
(v 1 v n )

and so using the change of variables formula,

xi1 xin
1
i R1
(u)
a
R
(u)
du =
I
i
i
(u1 un )
Ri Ui

XZ
I

q XZ
X
j=1

Ri (Ui Vj )

q XZ
X
j=1

S1
j

Sj (Ui Vj )

R1
i

(17.5)

xi1 xin
1
1
du =
(u) i Ri (u) aI Ri (u)
(u1 un )

xi1 xin u1 un
1
1
dv
(v) i Sj (v) aI Sj (v)
(u1 un ) (v 1 v n )

which by Lemma 17.10 equals

=

q XZ
X
j=1

xi1 xin
1
1
dv.
j S1
(v)

S
(v)
a
S
(v)
I
i
j
j
j
(v 1 v n )
Sj (Ui Vj )

We sum over i in (17.6) and (17.5) to obtain

the definition of

p XZ
X
i=1

i R1
i

Ri Ui

using (Ui , Ri )

xi1 xin
1
du =
(u) aI Ri (u)
(u1 un )

p X
q XZ
X
i=1 j=1

xi1 xin
1
1
j S1
(v)

S
(v)
a
S
(v)
dv
I
i
j
j
j
(v 1 v n )
Sj (Ui Vj )

q XZ
X
j=1

Sj (Ui Vj )

S1
j

xi1 xin
1
dv =
(v) aI Sj (v)
(v 1 v n )

the definition of
This proves the theorem.

using (Vi , Si ) .

(17.6)

302

17.3

DIFFERENTIAL FORMS

Some examples of orientable manifolds

We show in this section that there are lots of orientable manifolds. The following simple proposition will
give abundant examples.
Proposition 17.14 Suppose is a bounded open subset of Rn with n 2 having the property that for all
e , containing p, an open interval, (a, b) , an open set, B Rn1 ,
p \ , there exists
an open set, U
k
and a function, g C B; R such that for some k {1, , n}
xk = g (c
xk )
e , and U
e equals either
whenever x U


ck B and xk (a, g (c
x Rn : x
xk ))

(17.7)

or

ck B and xk (g (c
x Rn : x
xk ) , b) .

(17.8)

Then is an orientable C k manifold with boundary. Here

ck x1 , , xk1 xk+1 , , xn
x

T

e and g be as described above. In the case of (17.7) define

Proof: Let U
R (x)

xk g (c
xk ) x2

x1

xn

T

where the x1 is in the k th slot. Then it is a simple exercise to verify that det (DR (x)) = 1. Now in case
(17.8) holds, we let
R (x)

g (c
xk ) xk

x2

x1

xn

T

We see that in this case we also have det (DR (x)) = 1.

Also, in either case we see that R is one to one and k times continuously differentiable mapping into Rn .
In case (17.7) the inverse is given by
R1 (u) =

u1

u2

uk + g (c
uk )

un

T

and in case (17.8), there is also a simple formula for R1 . Now modifying these functions outside of suitable
compact sets, we may assume they are all of the sort needed in the definition of a C k manifold.
fj covering along with functions Rj as just
The set, is compact and so there are p of these sets, U
described. Let U0 satisfy
\ pi=1 Ui U0 U0


and let R0 (x) x1 k x2 xn where k is chosen large enough that R0 maps U0 into Rn< .
Modifying this function off some compact set containing U0 , to equal zero off this set, we see (Ur , Rr ) is an
fr . The chain rule shows the derivatives of the overlap maps have
oriented atlas for if we define Ur U
positive determinants.
For example, a ball of radius r > 0 is an oriented n manifold with boundary because it satisfies the
conditions of the above proposition. This proposition gives examples of n manifolds in Rn but we want to
have examples of n manifolds in Rm for m > n. The following lemma will help.

17.3. SOME EXAMPLES OF ORIENTABLE MANIFOLDS

303

Lemma 17.15 Suppose O is a bounded open subset of Rn and let F : O Rm be a function in C k O; Rm
where m n with the property that for all x O, DF (x) has rank n. Then if y0 = F (x0 ) , there exists a
bounded open set in Rm , W, which contains

y0 , a bounded open set, U O which contains x0 and a function
G : W U such that G is in C k W ; Rn and for all x U,
G (F (x)) = x.
Furthermore, G = G1 P on W where P is a map of the form


P (y) = y i1 , , y in
for some list of indices, i1 < < in .
Proof: Consider the system
F (x) = y.
Since DF (x0 ) has rank n, the inverse function theorem can be applied to some system of the form
F ij (x) = y ij , j = 1, , n
to obtain open sets, U1 and V1 , subsets of Rn and a function, G1 C k (V1 ; U1 ) such that G1 is one to one and
onto and is the inverse of the function FI where FI (x) is the vector valued function whose j th component
is F ij (x) . If we restrict the open sets, U1 and V1 , calling the
restricted open sets, U and V respectively, we
can modify G1 off a compact set to obtain G1 C k V ; Rn and G1 is the inverse of FI . Now let P and
G be as defined above and let W = V Z where Z is a bounded open set in Rmn such that W contains
F (U ) . (If n = m, we let W = V.) We can modify
P off a compact set which contains W so the resulting


function, still denoted by P is in C k W ; Rn Then for x U


G (F (x)) = G1 (P (F (x))) = G1 FI (x) = x.
This proves the lemma.
With this lemma we can give a theorem which will provide many other examples.
Theorem 17.16 Let be an n
manifold with boundary in Rn and suppose O, an open bounded subset
of Rn . Suppose F C k O; Rm is one to one on O and DF (x) has rank n for all x O. Then F () is
also a manifold with boundary and F () = F () . If is a C l manifold for l k, then so is F () . If
is orientable, then so is F () .
Proof: Let (Ur , Rr ) be an atlas for and suppose Ur = Or where Or is an open subset of O. Let
g
x0 Ur . By Lemma 17.15 there exists
an open set, Wx0 in Rm containing F (x0 ), an open set in Rn , U
x0
k
n
containing x0 , and Gx0 C Wx0 ; R such that
Gx0 (F (x)) = x
g
g
for all x U
x0 . Let Ux0 Ur Ux0 .
Claim: F (Ux0 ) is open in F () .
Proof: Let x Ux0 . If F (x1 ) is close enough to F (x) where x1 , then F (x1 ) Wx0 and so
|x x1 | = |Gx0 (F (x)) Gx0 (F (x1 ))|
K |(F (x)) F (x1 )|
where K is some constant which depends only on
max {||DGx0 (y)|| : y Rm } .

304

DIFFERENTIAL FORMS

Therefore, if F (x1 ) is close enough to F (x) , it follows we can conclude |x x1 | is very small. Since Ux0
is open in it follows that whenever F (x1 ) is sufficiently close to F (x) , we have x1 Ux0 . Consequently
F (x1 ) F (Ux0 ) . This shows F (Ux0 ) is open in F () and proves the claim.
With this claim it follows that (F (Ux0 ) , Rr Gx0 ) is a chart. The inverse map of Rr Gx0 being F R1
r .
Since is compact there are finitely many of these sets, F (Uxi ) covering . This yields an atlas for F ()


of the form (F (Uxi ) , Rr Gxi ) where xi Ur and proves the first part. If the R1
are in C l Rr Ur ; Rn ,
r
then the overlap map for two of these charts is of the form,




Rs Gxj F R1
= Rs R1
r
r
while the inverse of one of the maps in the chart is of the form
F R1
r
showing that if is a C l manifold, then F () is also. This also verifies the claim that if (Ur , Rr ) is an
oriented atlas for , then F () also has an oriented atlas since the overlap maps described above are all of
the form Rs R1
r .
It remains to verify the assertion about boundaries. y F () if and only if for some xi Ur ,
Rr Gxi (y) Rn0
if and only if
Gxi (y)
if and only if
Gxi (F (x)) = x
where F (x) = y if and only if y F () . This proves the theorem.
A function F satisfying the condifions listed in Theorem17.16 is called a regular mapping.

17.4

Stokes theorem

One of the most important theorems in this subject is Stokes theorem which relates an integral of a differential
form on an oriented manifold with boundary to another integral of a differential form on the boundary.
P
Lemma 17.17 Let (Ui , Ri ) be an oriented atlas for , an oriented manifold. Also let =
aI dxI be a
differential form for which aI has compact support contained in Ur for each I. Then


Z
XZ
xi1 xin
1
du
=
.
(17.9)
aI Rr (u)
(u1 un )

Rr Ur
I

Proof: Let K Ur be a compact set for which aI = 0 off K for all I. Then consider the atlas, (Ui0 , Ri )
where Ui0 Ui K C for all i 6= r, Ur Ur0 . Thus (Ui0 , Ri ) is also an oriented atlas. Now let { i }be a partition
of unity subordinate to the sets {Ui0 } . Then if i 6= r
i (x) aI (x) = 0
for all I. Therefore,
Z

xi1 xin

( i aI )
(u)
du
(u1 un )
Ri Ui
i
I


XZ
xi1 xin
1
=
aI Rr (u)
du
(u1 un )
Rr Ur
XXZ

R1
i

and this proves the lemma.

Before proving Stokes theorem we need a definition. (This subject has a plethora of definitions.)

17.4. STOKES THEOREM

305

P
Definition 17.18 Let = I aI (x) dxi1 dxin1 be a differential form of order n 1 where aI is in
Cc1 (Rn ) . Then we define d, a differential form of order n by replacing aI (x) with
daI (x)

m
X
aI (x)

dxk

(17.10)

dxk dxi1 dxin1 .

(17.11)

k=1

xk

and putting a wedge after the dxk . Therefore,

d

m
XX
aI (x)
I

xk

k=1

Having wallowed in definitions, we are finally ready to prove Stokes theorem. The proof is very elementary, amounting to a computation which comes from the definitions.
P
Theorem 17.19 (Stokes theorem) Let be a C 2 orientable manifold with boundary and let I aI (x) dxi1
dxin1 be a differential form of order n 1 for which aI is C 1 . Then
Z
Z
=
d.
(17.12)

Proof: We let (Ur , Rr ) , r {1, , p} be an oriented atlas for and let { r } be a partition of unity
subordinate to {Ur } . Let B {r : Rr Ur Rn0 6= } .




Z
p Z
m XX
X
xj xi1 xin 1
aI
1
(u)
du
d
r j Rr
x
(u1 un )

r=1 Rr Ur
j=1
I

Now
r

aI
( r aI ) r
=

aI
xj
xj
xj

Therefore,
Z

d =

p Z
m XX
X
j=1

Rr Ur

r=1

p Z
m XX
X
j=1

r=1

Rr Ur


xj xi1 xin 1
( r aI )
1
Rr
(u)
du
xj
(u1 un )


xj xi1 xin 1
r
1
aI Rr
(u)
du
xj
(u1 un )

(17.13)

r
Consider the second line in (17.13). The expression,
xj aI has compact support in Ur and so by Lemma
17.17, this equals
p Z X
m X
X
r

a dxj dxi1 dxin1 =

j I
x

r=1
j=1

Z X
p
m XX

j=1

Z X
m X

j
x
j=1
I

r=1

xj

p
X
r=1

aI dxj dxi1 dxin1 =

aI dxj dxi1 dxin1 = 0

306
because

DIFFERENTIAL FORMS
P

r = 1 on . Thus we are left to consider the first line in (17.13).

n
X
xj xi1 xin 1
xj 1k
=
A
(u1 un )
uk
k=1

Thus, letting x = R1
r (u) and using the chain rule,




p Z
m XX
X
xj xi1 xin 1
( r aI )
1
Rr
(u)
du
xj
(u1 un )
j=1 I r=1 Rr Ur

X
p Z
m XX
n
X
( r aI )
xj 1k
(x)
A du
j
x
uk
j=1 I r=1 Rr Ur
k=1

!
p X
m XX
n Z
X
r aI R1
r
A1k du
k
u
j=1 I r=1 k=1 Rr Ur

!
p Z
n
m XX
X
X
r aI R1
r

A1k du
k
n
u
R

r=1
j=1

where A1k denotes the cofactor of

Z

x
.
uk

k=1

There are two cases here on the kth term in the above sum over k. The first case is where k 6= 1. In this
case we integrate by parts and obtain the kth term is
!
Z
p
m XX
1k
X
1 A

r aI Rr
du
uk
Rn

r=1
j=1
I

In the case where k = 1, we integrate by parts and obtain the kth term equals
p Z
m XX
X
j=1

r=1

Rn1

11 0
r aI R1
r A | du2 dun

Rn

r aI R1
r

A11
du
u1

Adding all these terms up and using Lemma 17.8, we finally obtain
Z

d =

m XXZ
X

n1
j=1 I rB R
m XXZ
X

j=1

rB

p Z
m XX
X
I

r=1

r aI

R1
r

j=1

Rr Ur Rn
0

r aI

Rn1

11 0
r aI R1
r A | =

xi1 xin1
(u2 un )

R1
r

0, u2 , , un du2 dun

xi1 xin1
(u2 un )

0, u2 , , un du2 dun

This proves Stokes theorem.

17.5

A generalization

We used that R1
is C 2 in the proof of Stokes theorem but the end result is a formula which involves only
i
the first derivatives of R1
i . This suggests that it is not necessary to make this assumption. This is in fact

17.5. A GENERALIZATION

307

the case. We give an argument which shows that Stokes theorem holds for oriented C 1 manifolds. For a still
more general theorem see Section 20.7. We do not present this more general result here because it depends
on too many hard theorems which are not proved until later.
Now suppose is only a C 1 orientable manifold. Then in the proof of Stokes theorem, we can say there
exists some subsequence, n such that
Z

d lim

p Z
m XX
X
j=1

r=1

Rr Ur




xj xi1 xin 1
aI
1
du
r j Rr N (u)
x
(u1 un )

where N is a mollifier and

xj xi1 xin 1
(u1 un )

is obtained from
x = R1
r N (u) .
The reason we can assert this limit is that from the dominated convergence theorem, it is routine to show




1
R1
N
r N ,i = Rr
,i




1
and by results presented in Chapter 12 using Minkowskis inequality, we see limN R1
r N ,i = Rr
,i
in Lp (Rr Ur ) for every p. Taking an appropriate subsequence, we can obtain, in addition to this,
almost
P
everywhere convergence for every partial derivative and every Rr .We may also arrange to have
r = 1
near . We may do this as follows. If Ur = Or where Or is open in Rm , we see that the compact set, is
covered by the open sets, Or . Consider the compact set, + B (0, ) K where < dist (, Rm \ pi=1 Oi ) .
Then take a partition of unity subordinate to the open P
cover {Oi } which sums to 1 on K. Then for N large
enough, R1

(R
U
)
will
lie
in
this
set,
K,
where
r = 1.
r i
N
r
Then we do the computations as in the proof of Stokes theorem. Using the same computations, with
1
R1
r N in place of Rr , along with the dominated convergence theorem,
Z
d =

lim

m XXZ
X
j=1

rB

m XXZ
X
j=1

rB

Rr Ur Rn
0

Rr Ur Rn
0

r aI

r aI

R1
r

R1
r

xi1 xin1
N
(u2 un )

xi1 xin1
(u2 un )

0, u2 , , un du2 dun

0, u2 , , un du2 dun

This yields the following generalization of Stokes theorem to the case of C 1 manifolds.
Theorem 17.20 (Stokes theorem) Let be a C 1 oriented manifold with boundary and let
dxin1 be a differential form of order n 1 for which aI is C 1 . Then
Z
Z
=
d.

aI (x) dxi1

(17.14)

308

17.6

DIFFERENTIAL FORMS

Surface measures

Let be a C 1 manifold in Rm , oriented or not. Let f be a continuous function defined on , and let (Ui , Ri )
be an
{ i } be a C partition of unity subordinate to the sets, Ui as described earlier. If
Patlas and let
I
= I aI (x) dx is a differential form, we may always assume
dxI = dxi1 dxin
where i1 < i2 < < in . The reason for this is that in taking an integral used to integrate the differential
(xi1 xin )
form, a switch in two of the dxj results in switching two rows in the determinant, (u1 un ) , implying that
any two of these differ only by a multiple of 1. Therefore, there is no loss of generality in assuming from
now on that in the sum for , I is always a list of indices which are strictly increasing. The case where
some term of has a repeat, dxir = dxis can be ignored because such terms deliver zero in integrating the
differential form because they involve a determinant having two equal rows. We emphasize again that from
now on I will refer to an increasing list of indices.
Let

!2 1/2
X xi1 xin

Ji (u)
(u1 un )
I

1
where here the sum

is taken over all possible increasing lists of n indices, I, from {1, , m} and x = Ri u.
m
Thus there are n terms in the sum. Note that if m = n we obtain only one term in the sum, the absolute
value of the determinant of Dx (u) . We define a positive linear functional, on C () as follows:

p Z
X
i=1

Ri Ui

f i R1
i (u) Ji (u) du.

(17.15)

We will now show this is well defined.

Lemma 17.21 The functional defined in (17.15) does not depend on the choice of atlas or the partition of
unity.
Proof: In (17.15), let { i } be a partition of unity as described there which is associated with the atlas
(Ui , Ri ) and let { i } be a partition of unity associated
in the same manner with the atlas (Vi , Si ) . Using


the change of variables formula with u = Ri S1
v
j
p Z
X
i=1

Ri Ui

p X
q Z
X
i=1 j=1

p X
q Z
X
i=1 j=1

Sj (Ui Vj )

p X
q Z
X
i=1 j=1

i f R1
i (u) Ji (u) du =

Ri Ui

(17.16)

j i f R1
i (u) Ji (u) du =



 u1 un









1
1
S1
 dv
j (v) i Sj (v) f Sj (v) Ji (u) 
 (v 1 v n ) 

Sj (Ui Vj )

1
1
j S1
j (v) i Sj (v) f Sj (v) Jj (v) dv.

(17.17)

17.6. SURFACE MEASURES

309

This yields
the definition of f using (Ui , Ri )
p Z
X

Ri Ui

i=1

p X
q Z
X

Sj (Ui Vj )

i=1 j=1

1
1
j S1
j (v) i Sj (v) f Sj (v) Jj (v) dv

q Z
X
j=1

i f R1
i (u) Ji (u) du =

Sj (Vj )

1
j S1
j (v) f Sj (v) Jj (v) dv

the definition of f using (Vi , Si ) .

This proves the lemma.
This lemma implies the following theorem.
Theorem 17.22 Let be a C k manifold with boundary. Then there exists a unique Radon measure, ,
defined on such that whenever f is a continuous function defined on and (Ui , Ri ) denotes an atlas and
{ i } a partition of unity subordinate to this atlas, we have
f =

f d =

p Z
X
i=1

Ri Ui

i f R1
i (u) Ji (u) du.

(17.18)

Furthermore, for any f L1 (, ) ,

Z

p Z
X

f d =

i=1

Ri Ui

i f R1
i (u) Ji (u) du

(17.19)

and a subset, A, of is measurable if and only if for all r, Rr (Ur A) is Jr (u) du measurable.
Proof:We begin by proving the following claim.
Claim :A set, S Ui , has measure zero in Ui , if and only if Ri S has measure zero in Ri Ui with
respect to the measure, Ji (u) du.
Proof of the claim:Let > 0 be given. By outer regularity, there exists a set, V Ui , open in
such that (V ) < and S V Ui . Then Ri V is open in Rn and contains Ri S. Letting h O, where
O Rn = Ri V and mn (O) < mn (Ri V ) + , and letting h1 (x) h (Ri (x)) for x Ui , we see h1 V. By
Corollary 12.24,
that
spt (h1 ) {x Rm : i (x) = 1} . Thus

we can also choose our partition of unity so
1
1
j h1 Rj (u) = 0 unless j = i when this reduces to h1 Ri (u) . Thus

(V )

Z
V

Z
Ri Ui

Z
O

h1 d =

h1 d =

h1 R1
i (u) Ji (u) du =

h (u) Ji (u) du Ki ,

p Z
X
j=1

Rj Uj

Z
Ri Ui

j h1 R1
j (u) Jj (u) du

h (u) Ji (u) du =

Z
Ri V

h (u) Ji (u) du

310

DIFFERENTIAL FORMS

where Ki ||Ji || . Now this holds for all h O and so

Z
Z
Z
Ji (u) du
Ji (u) du
Ji (u) du (1 + Ki ) .
Ri S

Ri V

Since is arbitrary, this shows Ri S has mesure zero with respect to the measure, Ji (u) du as claimed.
Now we prove the converse. Suppose Ri S has Jr (u) du measure zero. Then there exists an open set, O
such that O Ri S and
Z
Ji (u) du < .
O
1
Thus R1
i (O Ri Ui ) is open in and contains S. Let h Ri (O Ri Ui ) be such that
Z


hd + > R1
i (O Ri Ui ) (S)

As in the first part, we can choose our partition of unity such that h (x) = 0 off the set,
{x Rm : i (x) = 1}
and so as in this part of the argument,
Z

hd

p Z
X

ZRi Ui

j=1

Rj Uj

j h R1
j (u) Jj (u) du

h R1
i (u) Ji (u) du


h R1
i (u) Ji (u) du

ZORi Ui

Ji (u) du <
O

and so (S) 2. Since is arbitrary, this proves the claim.

Now let A Ur be measurable. By the regularity of the measure, there exists an F set, F and a G
set, G such that Ur G A F and (G \ F ) = 0.(Recall a G set is a countable intersection of open sets
and an F set is a countable union of closed sets.) Then since is compact, it follows each of the closed sets
whose union equals F is a compact set. Thus if F =
k=1 Fk we know Rr (Fk ) is also a compact set and so
Rr (F ) =
R
(F
)
is
a
Borel
set.
Similarly,
R
(G)
is also a Borel set. Now by the claim,
k
r
k=1 r
Z
Jr (u) du = 0.
Rr (G\F )

We also see that since Rr is one to one,

Rr G \ Rr F = Rr (G \ F )
and so
Rr (F ) Rr (A) Rr (G)
where Rr (G) \ Rr (F ) has measure zero. By completeness of the measure, Ji (u) du, we see Rr (A) is
measurable. It follows that if A is measurable, then Rr (Ur A) is Jr (u) du measurable for all r. The
converse is entirely similar.

17.7.

DIVERGENCE THEOREM

311

Letting f L1 (, ) , we use the fact that is a Radon mesure to obtain a sequence

 of continuous


functions, {fk } which converge to f in L1 (, ) and for a.e. x. Therefore, the sequence fk R1
() is
i




a Cauchy sequence in L1 Ri Ui ; i R1
i (u) Ji (u) du . It follows there exists




g L1 Ri Ui ; i R1
i (u) Ji (u) du






1
Ri Ui ; i R1
such that fk R1
i () g in L
i (u) Ji (u) du . By the pointwise convergence, g (u) =


1
1
f R1
for a.e. u Ri Ui and so
i (u) for a.e. Ri (u) Ui . By the above claim, g = f Ri
f R1
L1 (Ri Ui ; Ji (u) du)
i
and we can write
Z

f d

=
=

lim

fk d = lim

p Z
X

i=1

Ri Ui

i fk R1
i (u) Ji (u) du

i R1
i (u) g (u) Ji (u) du

i=1 Ri Ui
p Z
X
i=1

p Z
X

i f R1
i (u) Ji (u) du.

Ri Ui

This proves the theorem.

Corollary 17.23 Let f L1 (; ) and suppose f (x) = 0 for all x
/ Ur where (Ur , Rr ) is a chart in a C k
atlas for . Then
Z
Z
Z


f R1
(17.20)
f d =
f d =
r (u) Jr (u) du.

Ur

Rr Ur

Proof: Using regularity of the measures, we can pick a compact subset, K, of Ur such that

Z
Z



f d
f d < .

K

Ur

Now by Corollary 12.24, we can choose the partition of unity such that K {x Rm : r (x) = 1} . Then
Z

f d

p Z
X

Zi=1

Ri Ui

Rr Ur

i f XK R1
i (u) Ji (u) du

f XK R1
r (u) Jr (u) du.

Therefore, letting Kl Rr Ur we can take a limit and conclude

Z

Z




1

 .
f
d

f
R
(u)
J
(u)
du
r
r


Ur

Rr Ur

Since is arbitrary, this proves the corollary.

17.7

Divergence theorem

What about writing the integral of a differential form in terms of this measure? This is a useful idea because
it allows us to obtain various important formulas such as the divergence theorem which are traditionally

312

DIFFERENTIAL FORMS

written not in terms of differential forms but in terms of measure on the surface and outer normals. Let
be a differential form,
X
(x) =
aI (x) dxI
I

where aI is continuous and the sum is taken over all increasing lists from {1, , m} . We assume is a C k
manifold which is orientable and that (Ur , Rr ) is an oriented atlas for while, { r } is a C partition of
unity subordinate to the Ur .
Lemma 17.24 Consider the set,


S x : for some r, x = R1
.
r (u) where x Ur and Jr (u) = 0
Then (S) = 0.
p
Proof: Let Sr denote those points, x, of Ur for which x = R1
r (u) and Jr (u) = 0. Thus S = r=1 Sr .
From Corollary 17.23
Z
Z


XSr d =
XSr R1
r (u) Jr (u) du = 0

Rr Ur

and so
(S)

p
X

(Sk ) = 0.

k=1

This proves the lemma.

With respect to the above atlas, we define a function of x in the following way. For I = (i1 , , in ) an
increasing list of indices,


(xi1 xin )
/Jr (u) , if x Ur \ S
1
n
I
(u u )
o (x)

0 if x S

Now it follows from Lemma 17.10 that if we had used a different atlas having the same orientation, then
m
oI (x) would be unchanged. We define a vector in R( n ) by letting the I th component of o (x) be defined by
oI (x) . Also note that since (S) = 0,
X
2
oI (x) = 1 a.e.
I

Define
(x) o (x)

aI (x) oI (x) ,

From the definition of what we mean by the integral of a differential form, Definition 17.11, it follows that


Z
p XZ
X



xi1 xin
1
1
r Rr (u) aI Rr (u)

du
(u1 un )
Rr Ur

r=1 I
p Z
X






1
1
=
r R1
r (u) Rr (u) o Rr (u) Jr (u) du
Rr Ur

r=1
Z

od

Note that o is bounded and measurable so is in L1 .

(17.21)

17.7.

DIVERGENCE THEOREM

313

Lemma 17.25 Let be a C k oriented manifold in Rn with an oriented atlas, (Ur , Rr ). Letting x = R1
r u
and letting 2 j n, we have


n
x1 xbi xn
X
xi
i+1
(1)
=0
(17.22)
j
u
(u2 un )
i=1
for each r. Here, xbi means this is deleted. If for each r,


x1 xn
0,
(u1 un )
then for each r


n
x1 xbi xn
X
xi
i+1
(1)
0 a.e.
1
u
(u2 un )
i=1

(17.23)

Proof: (17.23) follows from the observation that



1
bi xn
n

x
i
X
x x
x
i+1
=
(1)
1
(u1 un )
u
(u2 un )
i=1
1

by expanding the determinant,

x1 xn
,
(u1 un )
along the first column. Formula (17.22) follows from the observation that the sum in (17.22) is just the
determinant of a matrix which has two equal columns. This proves the lemma.
With this lemma, it is easy to verify a general form of the divergence theorem from Stokes theorem.
First we recall the definition of the divergence of a vector field.
Pn
Definition 17.26 Let O be an open subset of Rn and let F (x) k=1 F k (x) ek be a vector field for which
F k C 1 (O) . Then
div (F)

n
X
Fk

k=1

xk

Theorem 17.27 Let be an orientable C k n manifold with boundary in Rn having an oriented atlas,
(Ur , Rr ) which satisfies


x1 xn
0
(17.24)
(u1 un )
for all r. Then letting n (x) be the vector field whose ith component taken with respect to the usual basis of
Rn is given by


(
1
n
bi
i+1 x x x
(1)
ni (x)
(17.25)
(u2 un ) /Jr (u) if Jr (u) 6= 0
0 if Jr (u) = 0
for x Ur , it follows n (x) is independent of the choice of atlas provided
the orientation remains


unchanged. Also n (x) is a unit vector for a.e. x . Let F C 1 ; Rn . Then we have the following
formula which is called the divergence theorem.
Z
Z
div (F) dx =
F nd,
(17.26)

where is the surface measure on defined above.

314

DIFFERENTIAL FORMS

Proof: Recall that on

 2 1/2

1
bi xn
n

x
X

Jr (u) =

2 un )

(u
i=1
From Lemma 17.10 and the definition of two atlass having the same orientation, we see that aside from sets
of measure zero, the assertion about the independence of choice of atlas for the normal, n (x) is verified.
Also, by Lemma 17.24, we know Jr (u) 6= 0 off some set of measure zero for each atlas and so n (x) is a unit
vector for a.e. x.
Now we define the differential form,

n
X

i+1

(1)

i=1

ci dxn .
Fi (x) dx1 dx

Then from the definition of d,

d = div (F) dx1 dxn .
Now let { r } be a partition of unity subordinate to the Ur . Then using (17.24) and the change of variables
formula,


Z
p Z
X

x1 xn
1
d
( r div (F)) Rr (u)
du
(u1 un )

r=1 Rr Ur
Z
p Z
X
=
( r div (F)) (x) dx =
div (F) dx.
r=1

Ur

Now
Z

p Z
X

i=1

r=1



x1 xbi xn


du2 dun
r Fi R1
r (u)
2 un )
n

(u
R
U
R
r r
0
r=1
!
n
X


i
2
n
r
Fi n
R1
r (u) Jr (u) du du

p Z
X
i+1
(1)

n
X

Rr Ur Rn
0

i=1

F nd.

By Stokes theorem,
Z

div (F) dx =

d =

F nd

and this proves the theorem.

Definition 17.28 In the context of the divergence theorem, the vector, n is called the unit outer normal.
(x1 xn )
Since we did not assume (u1 un ) 6= 0 for all x = R1
r (u) , this is about all we can say about the
geometric significance of n. However, it is usually the case that we are in a situation where this determinant
is non zero. This is the case in the context of Proposition 17.14 for example, when this determinant was
equal to one. The next proposition shows that n does what we would hope it would do if it really does
(x1 xn )
deserve to be called a unit outer normal when (u1 un ) 6= 0.

17.7.

DIVERGENCE THEOREM

315

Proposition 17.29 Let n be given by (17.25) at a point of the boundary,

n
x0 = R1
r (u0 ) , u0 R0 ,

where (Ur , Rr ) is a chart for the manifold, , of Theorem 17.27 and

x1 xn
(u0 ) > 0
(u1 un )

(17.27)

at this point, then |n| = 1 and for all t > 0 small enough,
x0 + tn
/

(17.28)

and
n

x
=0
uj

(17.29)

for all j = 2, , n.
Proof: First note that (17.27) implies Jr (u0 ) > 0 and that we have already verified that |n| = 1 and
(17.29). Suppose the proposition is not true. Then there exists a sequence, {tj } such that tj > 0 and tj 0
for which
R1
r (u0 ) + tj n .
Since Ur is open in , it follows that for all j large enough,
R1
r (u0 ) + tj n Ur .
Therefore, there exists uj Rn such that
1
R1
r (uj ) = Rr (u0 ) + tj n.

Now by the inverse function theorem, this implies

uj

= Rr R1
r (u0 ) + tj n


= DRr R1
r (u0 ) ntj + u0 + o (tj )

= DR1
r (u0 ) ntj + u0 + o (tj ) .

At this point we take the first component of both sides and utilize the fact that the first component of u0
equals zero. Then


0 u1j = tj DR1
(17.30)
r (u0 ) n e1 + o (tj ) .


We consider the quantity, DR1
r (u0 ) n e1 . From the formula for the inverse in terms of the transpose of
the cofactor matrix, we obtain


DR1
r (u0 ) n e1 =

1

(1)1+j Mj1 nj
det DR1
(u
)
r
0

where
M j1 =



x1 xbj xn
(u2 un )

316

DIFFERENTIAL FORMS

is the determinant of the matrix obtained from deleting the j th row and the first column of the matrix,
x1

x1
u1 (u0 )
un (u0 )

..
..

.
.
xn
xn
u1 (u0 )
un (u0 )
which is just the matrix of the linear transformation, DR1
r (u0 ) taken

with respect to the usual basis
vectors. Now using the given formula for nj we see that DR1
(u
)
n
e1 > 0. Therefore, we can divide
0
r
by the positive number tj in (17.30) and choose j large enough that




1
 o (tj ) 
DR
(u
)
n
e1
0
r


 tj  <
2

to obtain a contradiction. This proves the proposition and gives a geometric justification of the term unit
outer normal applied to n.

17.8

Exercises

1. In the section on surface measure we used

!2 1/2
X xi1 xin

Ji (u)
(u1 un )
I

What if we had used instead

"



p #1/p
X  xi1 xin 
Ji (u)?


 (u1 un ) 
I

Would everything have worked out the same? Why is there a preference for the exponent, 2?
(x1 xn )
2. Suppose is an oriented C 1 n manifold in Rn and that for one of the charts, (u1 un ) > 0. Can it be
concluded that this condition holds for all the charts? What if we also assume is connected?

3. We defined manifolds with boundary in terms of the maps, Ri mapping into the special half space,
{u Rn : u1 0} . We retained this special half space in the discussion of oriented manifolds. However,
we could have used any half space in our definition. Show that if n 2, there was no loss of generality
in restricting our attention to this special half space. Is there a problem in defining oriented manifolds
in this manner using this special half space in the case where n = 1?


4. Let be an oriented Lipschitz or C k n manifold with boundary in Rn and let f C 1 ; Rk . Show
that
Z
Z
f
dx =
f nj d
j
x

where is the surface measure on discussed above. This says essentially that we can exchange
differentiation with respect to xj on with multiplication by the j th component of the exterior normal
on . Compare to the divergence theorem.
m
5. Recall the vector valued function, o (x) , for a C 1 oriented manifold which has values in R( n ) . Show
m
that for an orientable manifold this function is continuous as well as having its length in R( n ) equal
to one where the length is measured in the usual way.

17.8. EXERCISES

317

6. In the proof of Lemma 17.4 we used a very hard result, the invariance of domain theorem. Assume
the manifold in question is a C 1 manifold and give a much easier proof based on the inverse function
theorem.
7. Suppose is an oriented, C 1 2 manifold in R3 . And consider the function
!
!
!






x2 x3
x1 x3
x1 x2
/Ji (u) e1
/Ji (u) e2 +
/Ji (u) e3 .
n (x)
(u1 u2 )
(u1 u2 )
(u1 u2 )

(17.31)

Show this function has unit length in R3 , is independent of the choice of atlas having the same orientation, and is a continuous function of x . Also show this function is perpendicular to at every
point by verifying its dot product with x/ui equals zero. To do this last thing, observe the following
determinant.


 x1 x1 x3 
 ui u1 u2 
 x2 x2 x3 
 ui u1 u2 
 x3 x3 x3 


i
1
2
u

8. Take a long rectangular piece of paper, put one twist in it and then glue or tape the ends together.
This is called a Moebus band. Take a pencil and draw a line down the center. If you keep drawing,
you will be able to return to your starting point without ever taking the pencil off the paper. In other
words, the shape you have obtained has only one side. Now if we consider the pencil as a normal vector
to the plane, can you explain why the Moebus band is not orientable? For more fun with scissors and
paper, cut the Moebus band down the center line and see what happens. You might be surprised.
9. Let be a C k 2 manifold with boundary in R3 and let = a1 (x) dx1 + a2 (x) dx2 + a3 (x) dx3 be a
one form where the ai are C 1 functions. Show that


a1
a2
2 dx1 dx2 +
d =
x1
x



a3
a1
a3
a2
1
3

dx

dx
+

dx2 dx3 .
x1
x3
x2
x3
R
R
Stokes theorm would say that = d. This is the classical form of Stokes theorem.


10. In the context of 9, Stokes theorem is usually written in terms of vector notation rather than differential form notation. This involves the curl of a vector field and a normal to the given 2 manifold. Let n
be given as in (17.31) and let a C 1 vector field be given by a (x) a1 (x) e1 + a2 (x) e2 + a3 (x) e3 where
the ej are the standard unit vectors. Recall from elementary calculus courses that


 e1
e2
e3 



curl (a) =  x
1

x2
x3
 a1 (x) a2 (x) a3 (x) 


a3
a2
=
3 e1 +
x2
x




a1
a3
a2
a1
2
1 e +
2 e3 .
x3
x
x1
x
Letting be the surface measure on and 1 the surface measure on defined above, show that
Z
Z
d =
curl (a) nd

318

DIFFERENTIAL FORMS
and so Stokes formula takes the form
Z
Z
curl (a) nd =
a1 (x) dx1 + a2 (x) dx2 + a3 (x) dx3

Z
=
a Td1

where T is a unit tangent vector to given by


 

 x 
x

/  2  .
T (x)
u2
u
 x 
Assume  u2  6= 0. This means you have a well defined unit tangent vector to .

x
11. It is nice to understand the geometric relationship between n and T. Show that u
1 points into


2
x
x
x
while u2 points along and that n u2 u1 = Ji (u) > 0. Using the geometric description of
the cross product from elementary calculus, show n is the direction of a person walking arround
with on his left hand. The following picture is illustrative of the situation.

x
u
1


x

 u2

 P
I
@
PP
n@

PP
@ 
@

Representation Theorems
18.1

Radon Nikodym Theorem

This chapter is on various representation theorems. The first theorem, the Radon Nikodym Theorem, is a
representation theorem for one measure in terms of another. The approach given here is due to Von Neumann
and depends on the Riesz representation theorem for Hilbert space.
Definition 18.1 Let and be two measures defined on a -algebra, S, of subsets of a set, . We say that
is absolutely continuous with respect to and write << if (E) = 0 whenever (E) = 0.
Theorem 18.2 (Radon Nikodym) Let and be finite measures defined on a -algebra, S, of subsets of
. Suppose << . Then there exists f L1 (, ) such that f (x) 0 and
Z
(E) =
f d.
E

Proof: Let : L2 (, + ) C be defined by

g =

g d.

By Holders inequality,
|g|

Z

1 d

1/2 Z

1/2
1/2
|g| d ( + )
= () ||g||2
2

and so (L2 (, + ))0 . By the Riesz representation theorem in Hilbert space, Theorem 15.11, there
exists h L2 (, + ) with
Z
Z
g =
g d =
hgd( + ).
(18.1)

Letting E ={x : Im h(x) > 0}, and letting g = XE , (18.1) implies

Z
(E) =
(Re h + i Im h)d( + ).
E

Since the left side of (18.2) is real, this shows ( + )(E) = 0. Similarly, if
E = {x : Im h (x) < 0},
319

(18.2)

320

REPRESENTATION THEOREMS

then ( + )(E) = 0. Thus we may assume h is real-valued. Now let E = {x : h(x) < 0} and let g = XE .
Then from (18.2)
Z
(E) =
h d( + ).
E

Since h(x) < 0 on E, it follows ( + )(E) = 0 or else the right side of this equation would be negative.
Thus we can take h 0. Now let E = {x : h(x) 1} and let g = XE . Then
Z
(E) =
h d( + ) (E) + (E).
E

Therefore (E) = 0. Since << , it follows that (E) = 0 also. Thus we can assume
0 h(x) < 1
for all x. From (18.1), whenever g L2 (, + ),
Z
Z
g(1 h)d =
hgd.

Let g(x) =

Pn

i=0

hi (x)XE (x) in (18.3). This yields

Z

(1 hn+1 (x))d =

Let f (x) =

i=1

(18.3)

Z n+1
X

hi (x)d.

(18.4)

E i=1

hi (x) and use the Monotone Convergence theorem in (18.4) to let n and conclude
Z
(E) =
f d.
E

We know f L (, ) because is finite. This proves the theorem.

Note that the function, f is unique a.e. because, if g is another function which also serves to represent
, we could consider the set,
E {x : f (x) g (x) > > 0}
and conclude that
0=

f (x) g (x) d (E) .

Since this holds for every > 0, it must be the case that the set where f is larger than g has measure zero.
Similarly, the set where g is larger than f has measure zero. The f in the theorem is sometimes denoted by
d
.
d
The next corollary is a generalization to finite measure spaces.
Corollary 18.3 Suppose << and there exist sets Sn S with
Sn Sm = ,
n=1 Sn = ,
and (Sn ), (Sn ) < . Then there exists f 0, where f is measurable, and
Z
(E) =
f d
E

for all E S. The function f is + a.e. unique.

18.2. VECTOR MEASURES

321

Proof: Let Sn = {E Sn : E S}. Clearly Sn is a algebra of subsets of Sn , , are both finite

measures on Sn , and << . Thus, by Theorem 18.2, there exists an Sn measurable function fn , fn (x) 0,
with
Z
(E) =
fn d
E

for all E Sn . Define f (x) = fn (x) for x Sn . Then f is measurable because

1
f 1 ((a, ]) =
n=1 fn ((a, ]) S.

Also, for E S,
(E)

=
=

(E Sn ) =

n=1
Z
X

Z
X

XESn (x)fn (x)d

n=1

XESn (x)f (x)d

n=1

f d.

To see f is unique, suppose f1 and f2 both work and consider

E {x : f1 (x) f2 (x) > 0}.
Then
0 = (E Sn ) (E Sn ) =

f1 (x) f2 (x)d.

ESn

Hence (E Sn ) = 0 so (E) = 0. Hence (E) = 0 also. Similarly

( + )({x : f2 (x) f1 (x) > 0}) = 0.
This version of the Radon Nikodym theorem will suffice for most applications, but more general versions
are available. To see one of these, one can read the treatment in Hewitt and Stromberg. This involves the
notion of decomposable measure spaces, a generalization of finite.

18.2

Vector measures

The next topic will use the Radon Nikodym theorem. It is the topic of vector and complex measures. Here we
are mainly concerned with complex measures although a vector measure can have values in any topological
vector space.
Definition 18.4 Let (V, ||||) be a normed linear space and let (, S) be a measure space. We call a function
: S V a vector measure if is countably additive. That is, if {Ei }
i=1 is a sequence of disjoint sets of S,
(
i=1 Ei )

(Ei ).

i=1

Definition 18.5 Let (, S) be a measure space and let be a vector measure defined on S. A subset, (E),
of S is called a partition of E if (E) consists of finitely many disjoint sets of S and (E) = E. Let
X
||(E) = sup{
||(F )|| : (E) is a partition of E}.
F (E)

|| is called the total variation of .

322

REPRESENTATION THEOREMS

The next theorem may seem a little surprising. It states that, if finite, the total variation is a nonnegative
measure.
Theorem 18.6 If ||() < , then || is a measure on S.
Proof: Let E1 E2 = and let {Ai1 Aini } = (Ei ) with
||(Ei ) <

ni
X

||(Aij )|| i = 1, 2.

j=1

Let (E1 E2 ) = (E1 ) (E2 ). Then

X

||(E1 E2 )

||(F )|| > ||(E1 ) + ||(E2 ) 2.

F (E1 E2 )

Since > 0 was arbitrary, it follows that

||(E1 E2 ) ||(E1 ) + ||(E2 ).

(18.5)

Let {Ej }
j=1 be a sequence of disjoint sets of S. Let E = j=1 Ej and let

{A1 , , An } = (E )
be such that
||(E ) <

n
X

||(Ai )||.

i=1

But ||(Ai )||

j=1

||(Ai Ej )||. Therefore,

||(E )

<

n X

i=1 j=1
X
n
X

||(Ai Ej )||
||(Ai Ej )||

j=1 i=1

||(Ej ).

j=1

The interchange in order of integration follows from Fubinis theorem or else Theorem 5.44 on the equality
of double sums, and the last inequality follows because A1 Ej , , An Ej is a partition of Ej .
Since > 0 is arbitrary, this shows
||(
j=1 Ej )

||(Ej ).

j=1

By induction, (18.5) implies that whenever the Ei are disjoint,

||(nj=1 Ej )

n
X
j=1

||(Ej ).

18.2. VECTOR MEASURES

323

Therefore,

n
||(Ej ) ||(
j=1 Ej ) ||(j=1 Ej )

n
X

||(Ej ).

j=1

j=1

Since n is arbitrary, this implies

||(
j=1 Ej )

||(Ej )

j=1

which proves the theorem.

In the case where V = C, it is automatically the case that ||() < . This is proved in Rudin [24]. We
will not need to use this fact, so it is left for the interested reader to look up.
Theorem 18.7 Let (, S) be a measure space and let : S C be a complex vector measure with ||() <
. Let : S [0, ()] be a finite measure such that << . Then there exists a unique f L1 () such
that for all E S,
Z
f d = (E).
E

Proof: It is clear that Re and Im are real-valued vector measures on S. Since ||() < , it follows
easily that | Re |() and | Im |() < . Therefore, each of
| Re | + Re | Re | Re() | Im | + Im
| Im | Im()
,
,
, and
2
2
2
2
are finite measures on S. It is also clear that each of these finite measures are absolutely continuous with
respect to . Thus there exist unique nonnegative functions in L1 (), f1, f2 , g1 , g2 such that for all E S,
Z
1
(| Re | + Re )(E) =
f1 d,
2
ZE
1
(| Re | Re )(E) =
f2 d,
2
ZE
1
(| Im | + Im )(E) =
g1 d,
2
E
Z
1
(| Im | Im )(E) =
g2 d.
2
E
Now let f = f1 f2 + i(g1 g2 ).
The following corollary is about representing a vector measure in terms of its total variation.
1
Corollary 18.8 Let
R be a complex vector measure with ||() < . Then there exists a unique f L ()
such that (E) = E f d||. Furthermore, |f | = 1 || a.e. This is called the polar decomposition of .

Proof: First we note that << || and so such an L1 function exists and is unique. We have to show
|f | = 1 a.e.
Lemma 18.9 Suppose (, S, ) is a measure space and f is a function in L1 (, ) with the property that
Z
|
f d| (E)
E

for all E S. Then |f | 1 a.e.

324

REPRESENTATION THEOREMS

Proof of the lemma: Consider the following picture.

B(p, r)
(0, 0)


p.

where B(p, r) B(0, 1) = . Let E = f 1 (B(p, r)). If (E) 6= 0 then





Z
Z
 1

 1




f d p = 
(f p)d
 (E)
(E) E
E
Z
1

|f p|d < r.
(E) E
Hence
|

1
(E)

f d| > 1,

contradicting the assumption of the lemma. It follows (E) = 0. Since {z C : |z| > 1} can be covered by
countably many such balls, it follows that
(f 1 ({z C : |z| > 1} )) = 0.
Thus |f (x)| 1 a.e. as claimed. This proves the lemma.
To finish the proof of Corollary 18.8, if ||(E) 6= 0,

 

Z
 (E)   1


=
 1.
f
d||
 ||(E)   ||(E)

E
Therefore |f | 1, || a.e. Now let

En = {x : |f (x)| 1

1
}.
n

Let {F1 , , Fm } be a partition of En such that

m
X

|(Fi )| ||(En ) .

i=1

Then
||(En )

m
X
i=1




m Z
X
|(Fi )|
|
i=1

1
1
n

X
m

1
1
n

f d || |

Fi

|| (Fi )

i=1

|| (En )

and so
1
|| (En ) .
n
Since was arbitrary, this shows that || (En ) = 0. But {x : |f (x)| < 1} =
n=1 En . So ||({x :
|f (x)| < 1}) = 0. This proves Corollary 18.8.

18.3. REPRESENTATION THEOREMS FOR THE DUAL SPACE OF LP

18.3

325

Representation theorems for the dual space of Lp

In Chapter 14 we discussed the definition of a Banach space and the dual space of a Banach space. We also
saw in Chapter 12 that the Lp spaces are Banach spaces. The next topic deals with the dual space of Lp for
p 1 in the case where the measure space is finite or finite.
Theorem 18.10 (Riesz representation theorem) Let p > 1 and let (, S, ) be a finite measure space. If
(Lp ())0 , then there exists a unique h Lq () ( p1 + 1q = 1) such that
f =

hf d.

Proof: (Uniqueness) If h1 and h2 both represent , consider

f = |h1 h2 |q2 (h1 h2 ),
where h denotes complex conjugation. By Holders inequality, it is easy to see that f Lp (). Thus
0 = f f =
Z

h1 |h1 h2 |q2 (h1 h2 ) h2 |h1 h2 |q2 (h1 h2 )d

|h1 h2 |q d.

Therefore h1 = h2 and this proves uniqueness.

Now let (E) = (XE ). Let A1 , , An be a partition of .
|XAi | = wi (XAi ) = (wi XAi )
for some wi C, |wi | = 1. Thus
n
X

|(Ai )| =

i=1

n
X
i=1

n
X
|(XAi )| = (
wi XAi )
i=1

Z
Z X
n
1
1
1
||||( |
wi XAi |p d) p = ||||( d) p = ||||() p.

i=1

Therefore ||() < . Also, if {Ei }

i=1 is a sequence of disjoint sets of S, let
Fn = ni=1 Ei , F =
i=1 Ei .
Then by the Dominated Convergence theorem,
||XFn XF ||p 0.
Therefore,
(F ) = (XF ) = lim (XFn ) = lim
n

n
X

k=1

(XEk ) =

k=1

(Ek ).

326

REPRESENTATION THEOREMS

This shows is a complex measure with || finite. It is also clear that << . Therefore, by the Radon
Nikodym theorem, there exists h L1 () with
Z
(E) =
hd = (XE ).
E

Now let s =

Pm

i=1 ci XEi

be a simple function. We have

(s) =

m
X

ci (XEi ) =

i=1

m
X

ci

i=1

hd =

hsd.

(18.6)

Ei

Claim: If f is uniformly bounded and measurable, then

Z
(f ) = hf d.
Proof of claim: Since f is bounded and measurable, there exists a sequence of simple functions, {sn }
which converges to f pointwise and in Lp () . Then
Z
Z
(f ) = lim (sn ) = lim
hsn d = hf d,
n

the first equality holding because of continuity of and the second equality holding by the dominated
convergence theorem.
Let En = {x : |h(x)| n}. Thus |hXEn | n. Then
|hXEn |q2 (hXEn ) Lp ().
By the claim, it follows that
||hXEn ||qq =

h|hXEn |q2 (hXEn )d = (|hXEn |q2 (hXEn ))

q

|||| || |hXEn |q2 hXEn ||p = |||| ||hXEn ||qp.

Therefore, since q

q
p

= 1, it follows that
||hXEn ||q ||||.

Letting n , the Monotone Convergence theorem implies

||h||q ||||.

(18.7)

Now that h has been shown to be in Lq (), it follows from (18.6) and the density of the simple functions,
Theorem 12.8, that
Z
f = hf d
for all f Lp (). This proves Theorem 18.10.
Corollary 18.11 If h is the function of Theorem 18.10 representing , then ||h||q = ||||.
R
Proof: |||| = sup{ hf : ||f ||p 1} ||h||q |||| by (18.7), and Holders inequality.
To represent elements of the dual space of L1 (), we need another Banach space.

18.3. REPRESENTATION THEOREMS FOR THE DUAL SPACE OF LP

327

Definition 18.12 Let (, S, ) be a measure space. L () is the vector space of measurable functions such
that for some M > 0, |f (x)| M for all x outside of some set of measure zero (|f (x)| M a.e.). We define
f = g when f (x) = g(x) a.e. and ||f || inf{M : |f (x)| M a.e.}.
Theorem 18.13 L () is a Banach space.
Proof: It is clear that L () is a vector space and it is routine to verify that || || is a norm.
To verify completeness, let {fn } be a Cauchy sequence in L (). Let
|fn (x) fm (x)| ||fn fm ||
for all x
/ Enm , a set of measure 0. Let E = n,m Enm . Thus (E) = 0 and for each x
/ E, {fn (x)}
n=1 is
a Cauchy sequence in C. Let

0 if x E
f (x) =
= lim XE C (x)fn (x).
limn fn (x) if x
/E
n
Then f is clearly measurable because it is the limit of measurable functions. If
Fn = {x : |fn (x)| > ||fn || }
and F =
/ F E,
n=1 Fn , it follows (F ) = 0 and that for x
|f (x)| lim inf |fn (x)| lim inf ||fn || <
n

because {fn } is a Cauchy sequence. Thus f L (). Let n be large enough that whenever m > n,
||fm fn || < .
Thus, if x
/ E,
|f (x) fn (x)| =

lim |fm (x) fn (x)|

lim inf ||fm fn || < .

Hence ||f fn || < for all n large enough. This proves the theorem.

0
The next theorem is the Riesz representation theorem for L1 () .
Theorem 18.14 (Riesz representation theorem) Let (, S, ) be a finite measure space. If (L1 ())0 ,
then there exists a unique h L () such that
Z
(f ) =
hf d

for all f L1 ().

Proof: Just as in the proof of Theorem 18.10, there exists a unique h L1 () such that for all simple
functions, s,
Z
(s) = hs d.
(18.8)
To show h L (), let > 0 be given and let
E = {x : |h(x)| |||| + }.

328

REPRESENTATION THEOREMS

Let |k| = 1 and hk = |h|. Since the measure space is finite, k L1 (). Let {sn } be a sequence of simple
functions converging to k in L1 (), and pointwise. Also let |sn | 1. Therefore
Z
Z
(kXE ) = lim (sn XE ) = lim
hsn d =
hkd
n

where the last equality holds by the Dominated Convergence theorem. Therefore,
Z
Z
||||(E) |(kXE )| = |
hkXE d| =
|h|d

(|||| + )(E).
It follows that (E) = 0. Since > 0 was arbitrary, |||| ||h|| . Now we have seen that h L (), the
density of the simple functions and (18.8) imply
Z
f =
hf d , |||| ||h|| .
(18.9)

This proves the existence part of the theorem. To verify uniqueness, suppose h1 and h2 both represent
and let f L1 () be such that |f | 1 and f (h1 h2 ) = |h1 h2 |. Then
Z
Z
0 = f f = (h1 h2 )f d = |h1 h2 |d.
Thus h1 = h2 .
Corollary 18.15 If h is the function in L () representing (L1 ())0 , then ||h|| = ||||.
R
Proof: |||| = sup{| hf d| : ||f ||1 1} ||h|| |||| by (18.9).
Next we extend these results to the finite case.
Lemma 18.16 Let (, S, ) be a measure space and suppose there exists
R a measurable function, r such that
r (x) > 0 for all x, there exists M such that |r (x)| < M for all x, and rd < . Then for
(Lp (, ))0 , p 1,
there exists a unique h Lq (, ), L (, ) if p = 1 such that
Z
f = hf d.
Also ||h|| = ||||. (||h|| = ||h||q if p > 1, ||h|| if p = 1). Here
1 1
+ = 1.
p q
Proof: We present the proof in the case where p > 1 and leave the case p = 1 to the reader. It involves
routine modifications of the case presented. Define a new measure
e, according to the rule
Z

e (E)
rd.
(18.10)
E

Thus
e is a finite measure on S. Now define a mapping, : Lp (,
e) Lp (, ) by
1

f = r p f.

18.3. REPRESENTATION THEOREMS FOR THE DUAL SPACE OF LP

329

Then is one to one and onto. Also it is routine to show

||f ||Lp () = ||f ||Lp (e) .

(18.11)

Consider the diagram below which is descriptive of the situation.

q

L (e
)

L (e
)

Lp (e
)

Lp () ,
Lp ()

By the Riesz representation theorem for finite measures, there exists a unique h Lq (,
e) such that for all
f Lp (e
) ,
Z
(f ) = (f )
f hde
,
(18.12)

||h||Lq (e)

= || || = |||| ,

the last equation holding because of (18.11). But from (18.10),

Z
Z 
 1
1
f hde
=
r p f hr q d

Z
 1
=
(f ) hr q d

and so we see that

(f ) =

 1
(f ) hr q d
1

for all f Lp (e
) . Since is onto, this shows hr q represents as claimed. It only remains to verify
 q1 
|||| = hr  . However, this equation comes immediately form (18.10) and (18.12). This proves the
q

lemma.
A situation in which the conditions of the lemma are satisfied is the case where the measure space is
finite. This allows us to state the following theorem.
Theorem 18.17 (Riesz representation theorem) Let (, S, ) be finite and let
(Lp (, ))0 , p 1.
Then there exists a unique h Lq (, ), L (, ) if p = 1 such that
Z
f = hf d.
Also ||h|| = ||||. (||h|| = ||h||q if p > 1, ||h|| if p = 1). Here
1 1
+ = 1.
p q
Proof: Let {n } be a sequence of disjoint elements of S having the property that
0 < (n ) < ,
n=1 n = .

330

REPRESENTATION THEOREMS

Define
r(x) =

X
1
1
X
(x)
(
)
,

e
(E)
=
rd.

n
n2 n
E
n=1

Thus
Z

rd =
e() =

X
1
<
n2
n=1

so
e is a finite measure. By the above lemma we obtain the existence part of the conclusion of the theorem.
Uniqueness is done as before.
With the Riesz representation theorem, it is easy to show that
Lp (), p > 1
is a reflexive Banach space.
Theorem 18.18 For (, S, ) a finite measure space and p > 1, Lp () is reflexive.
0

Proof: Let r : (Lr ())0 Lr () be defined for

Z

1
r

1
r0

= 1 by

( r )g d = g

for all g Lr (). From Theorem 18.17 r is 1-1, onto, continuous and linear. By the Open Map theorem,
1
is also 1-1, onto, and continuous ( r equals the representor of ). Thus r is also 1-1, onto, and
r

q 0

p 0
continuous by Corollary 14.28. Now observe that J = p 1
q . To see this, let z (L ) , y (L ) ,

p 1
q ( q z )(y )

J( q z )(y )

= ( p z )(y )
= z ( p y )
Z
=
( q z )( p y )d,

= y ( q z )
Z
=
( p y )( p z )d.

q 0
p
Therefore p 1
q = J on q (L ) = L . But the two maps are onto and so J is also onto.

18.4

Riesz Representation theorem for non finite measure spaces

It is not necessary to assume is either finite or finite to establish the Riesz representation theorem for
1 < p < . This involves the notion of uniform convexity. First we recall Clarksons inequality for p 2.
This was Problem 24 in Chapter 12.
Lemma 18.19 (Clarkson inequality p 2) For p 2,
||

f +g p
f g p
1
1
||p + ||
||p ||f ||pp + ||g||pp .
2
2
2
2

18.4. RIESZ REPRESENTATION THEOREM FOR NON FINITE MEASURE SPACES

331

m
Definition 18.20 A Banach space, X, is said to be uniformly convex if whenever ||xn || 1 and || xn +x
||
2
1 as n, m , then {xn } is a Cauchy sequence and xn x where ||x|| = 1.

Observe that Clarksons inequality implies Lp is uniformly convex for all p 2. Uniformly convex spaces
have a very nice property which is described in the following lemma. Roughly, this property is that any
element of the dual space achieves its norm at some point of the closed unit ball.
Lemma 18.21 Let X be uniformly convex and let L X 0 . Then there exists x X such that
||x|| = 1, Lx = ||L||.
Proof: Let ||e
xn || 1 and |Le
xn | ||L||. Let xn = wn x
en where |wn | = 1 and
wn Le
xn = |Le
xn |.

Thus Lxn = |Lxn | = |Le

xn | ||L||.
Lxn ||L||, ||xn || 1.
We can assume, without loss of generality, that
Lxn = |Lxn |

||L||
2

and L 6= 0.
m
Claim || xn +x
|| 1 as n, m .
2
Proof of Claim: Let n, m be large enough that Lxn , Lxm ||L|| 2 where 0 < . Then ||xn +xm || 6= 0
because if it equals 0, then xn = xm so Lxn = Lxm but both Lxn and Lxm are positive. Therefore we
m
can consider ||xxnn +x
+xm || a vector of norm 1. Thus,
||L|| |L

(xn + xm )
2||L||
|
.
||xn + xm ||
||xn + xm ||

Hence
||xn + xm || ||L|| 2||L|| .
Since > 0 is arbitrary, limn,m ||xn + xm || = 2. This proves the claim.
By uniform convexity, {xn } is Cauchy and xn x, ||x|| = 1. Thus Lx = limn Lxn = ||L||. This
proves Lemma 18.21.
The proof of the Riesz representation theorem will be based on the following lemma which says that if
you can show certain things, then you can represent a linear functional.
Lemma 18.22 (McShane) Let X be a complex normed linear space and let L X 0 . Suppose there exists
x X, ||x|| = 1 with Lx = ||L|| 6= 0. Let y X and let y (t) = ||x + ty|| for t R. Suppose 0y (0) exists for
each y X. Then for all y X,
0y (0) + i 0iy (0) = ||L||1 Ly.
Proof: Suppose first that ||L|| = 1. Then
L(x + t(y L(y)x)) = Lx = 1 = ||L||.
Therefore, ||x + t(y L(y)x)|| 1. Also for small t, |L(y)t| < 1, and so
1 ||x + t(y L(y)x)|| = ||(1 L(y)t)x + ty||

332

REPRESENTATION THEOREMS




t


|1 L (y) t| x +
y .
1 L (y) t

This implies

1
t
||x +
y||.
|1 tL(y)|
1 L(y)t

(18.13)

Using the formula for the sum of a geometric series,

1
= 1 + tLy + o (t)
1 tLy
where limt0 o (t) (t1 ) = 0. Using this in (18.13), we obtain
|1 + tL(y) + o (t) | ||x + ty + o(t)||
Now if t > 0, since ||x|| = 1, we have
( y (t) y (0))t1

= (||x + ty|| ||x||)t1

(|1 + t L(y)| 1)t1 +
Re L(y) +

o(t)
t

o(t)
.
t

If t < 0,
( y (t) y (0))t1 Re L(y) +

o(t)
.
t

Since 0y (0) is assumed to exist, this shows

0y (0) = Re L(y).
Now
Ly = Re L(y) + i Im L(y)
so
L(iy) = i(Ly) = i Re L(y) + Im L(y)
and
L(iy) = Re L (iy) + i Im L (iy).
Hence
Re L(iy) = Im L(y).
Consequently, by (18.14)
Ly = Re L(y) + i Im L(y) = Re L (y) + i Re L (iy)
= 0y (0) + i 0iy (0).

(18.14)

18.4. RIESZ REPRESENTATION THEOREM FOR NON FINITE MEASURE SPACES

333

This proves the lemma when ||L|| = 1. For arbitrary L 6= 0, let Lx = ||L||, ||x|| = 1. Then from above, if
1
L1 y ||L|| L (y) , ||L1 || = 1 and so from what was just shown,
L(y)
= 0y (0) + i iy (0)
||L||

L1 (y) =

and this proves McShanes lemma.

We will use the uniform convexity and this lemma to prove the Riesz representation theorem next. Let
p 2 and let : Lq (Lp )0 be defined by
Z
(g)(f ) =
gf d.
(18.15)

Theorem 18.23 (Riesz representation theorem p 2) The map is 1-1, onto, continuous, and
||g|| = ||g||, |||| = 1.
R
Proof: Obviously is linear. Suppose g = 0. Then 0 = gf d for all f Lp . Let
f = |g|q2 g.
Then f Lp and so 0 =
inequality. In fact,

|g|q d. Hence g = 0 and is 1-1. That g (Lp )0 is obvious from the Holder

|(g)(f )| ||g||q ||f ||p ,

and so ||(g)|| ||g||q. To see that equality holds, let
f = |g|q2 g ||g||1q
.
q
Then ||f ||p = 1 and
(g)(f ) =

|g|q d||g||1q
= ||g||q .
q

Thus |||| = 1. It remains to show is onto. Let L (Lp )0 . We need show L = g for some g Lq . Without
loss of generality, we may assume L 6= 0. Let
Lg = ||L||, g Lp , ||g|| = 1.
We can assert the existence of such a g by Lemma 18.21. For f Lp ,
1

f (t) ||g + tf ||p f (t) p .

We show 0f (0) exists. Let [g = 0] denote the set {x : g (x) = 0}.
f (t) f (0)
=
t
1
t

Z
[g6=0]

(|g + tf |p |g|p )d =

1
t

|t| |f |p d

[g=0]

p|g(x) + s(x)f (x)|p2 Re[(g(x) + s(x)f (x))f(x)]d

(18.16)

334

REPRESENTATION THEOREMS

where the Mean Value theorem is being used on the function t |g (x) + tf (x) |p and s(x) is between 0 and
t, the integrand in the second integral of (18.16) equaling
1
(|g(x) + tf (x)|p |g(x)|p ).
t
Now if |t| < 1, the integrand in the last integral of (18.16) is bounded by


|f (x)|p
(|g(x)| + |f (x)|)p
p
+
q
p
which is a function in L1 since f, g are in Lp (we used the inequality ab
apply the Dominated Convergence theorem and obtain
Z
0f (0) = p |g(x)|p2 Re(g(x)f(x))d.

aq
q

bp
p ).

Because of this, we can

Hence
0f (0) = ||g||
Note

1
p

p
q

|g(x)|p2 Re(g(x)f(x))d.

|g(x)|p2 Re(ig(x)f(x))d.

1 = 1q . Therefore,
0if (0) = ||g||

p
q

But Re(ig f) = Im(g f) and so by the McShane lemma,

Z
p
Lf = ||L|| ||g|| q
|g(x)|p2 [Re(g(x)f(x)) + i Re(ig(x)f(x))]d
Z
p
q
= ||L|| ||g||
|g(x)|p2 [Re(g(x)f(x)) + i Im(g(x)f(x))]d
Z
p
= ||L|| ||g|| q
|g(x)|p2 g(x)f (x)d.
This shows that
L = (||L|| ||g||

p
q

|g|p2 g)

and verifies is onto. This proves the theorem.

To prove the Riesz representation theorem for 1 < p < 2, one can verify that Lp is uniformly convex and
then repeat the above argument. Note that no reference to p 2 was used in the proof. Unfortunately,
this requires Clarksons Inequalities for p (1, 2) which are more technical than the case where p 2. To
see this done see Hewitt & Stromberg [15] or Ray [22]. Here we take a different approach using the Milman
theorem which states that uniform convexity implies the space is Reflexive.
Theorem 18.24 (Riesz representation theorem) Let 1 < p < and let : Lq (Lp )0 be given by (18.15).
Then is 1-1, onto, and ||g|| = ||g||.
Proof: Everything is the same as the proof of Theorem 18.23 except for the assertion that is onto.
Suppose 1 < p < 2. (The case p 2 was done in Theorem 18.23.) Then q > 2 and so we know from Theorem
18.23 that : Lp (Lq )0 defined as
Z
f (g)
f gd

18.5. THE DUAL SPACE OF C (X)

335

is onto and ||f || = ||f ||. Then : (Lq )00 (Lp )0 is also 1-1, onto, and || L|| = ||L||. By Milmans
theorem, J is onto from Lq (Lq )00 . This occurs because of the uniform convexity of Lq which follows from
Clarksons inequality. Thus both maps in the following diagram are 1-1 and onto.
J

Lq (Lq )00 (Lp )0.

Now if g Lq , f Lp , then
J(g)(f ) = Jg(f ) = (f )(g) =

f g d.

Thus if : Lq (Lp )0 is the mapping of (18.15), this shows = J. Also

||g|| = || Jg|| = ||Jg|| = ||g||.
This proves the theorem.
In the case where p = 1, it is also possible to give the Riesz representation theorem in a more general
context than finite spaces. To see this done, see Hewitt and Stromberg [15]. The dual space of L has
also been studied. See Dunford and Schwartz [9].

18.5

The dual space of C (X)

Next we represent the dual space of C(X) where X is a compact Hausdorff space. It will turn out to be a
space of measures. The theorems we will present hold for X a compact or locally compact Hausdorff space
but we will only give the proof in the case where X is also a metric space. This is because the proof we use
depends on the Riesz representation theorem for positive linear functionals and we only gave such a proof
in the special case where X was a metric space. With the more general theorem in hand, the arguments
give here will apply unchanged to the more general setting. Thus X will be a compact metric space in what
follows.
Let L C(X)0 . Also denote by C + (X) the set of nonnegative continuous functions defined on X. Define
for f C + (X)
(f ) = sup{|Lg| : |g| f }.
Note that (f ) < because |Lg| ||L|| ||g|| ||L|| ||f || for |g| f .
Lemma 18.25 If c 0, (cf ) = c(f ), f1 f2 implies f1 f2 , and
(f1 + f2 ) = (f1 ) + (f2 ).
Proof: The first two assertions are easy to see so we consider the third. Let |gj | fj and let gej = eij gj
where j is chosen such that eij Lgj = |Lgj |. Thus Le
gj = |Lgj |. Then

Hence

|e
g1 + ge2 | f1 + f2.
|Lg1 | + |Lg2 | = Le
g1 + Le
g2 =
L(e
g1 + ge2 ) = |L(e
g1 + ge2 )| (f1 + f2 ).

Choose g1 and g2 such that |Lgi | + > (fi ). Then (18.17) shows

(f1 ) + (f2 ) 2 (f1 + f2 ).

(18.17)

336

REPRESENTATION THEOREMS

Since > 0 is arbitrary, it follows that

(f1 ) + (f2 ) (f1 + f2 ).

(18.18)

Now let |g| f1 + f2 , |Lg| (f1 + f2 ) . Let

(
hi (x) =

fi (x)g(x)
f1 (x)+f2 (x)

if f1 (x) + f2 (x) > 0,

0 if f1 (x) + f2 (x) = 0.

Then hi is continuous and h1 (x) + h2 (x) = g(x), |hi | fi . Therefore,

+ (f1 + f2 ) |Lg| |Lh1 + Lh2 | |Lh1 | + |Lh2 |
(f1 ) + (f2 ).
Since > 0 is arbitrary, this shows with (18.18) that
(f1 + f2 ) (f1 ) + (f2 ) (f1 + f2 )
which proves the lemma.
Let C(X; R) be the real-valued functions in C(X) and define
R (f ) = f + f
for f C(X; R). Using Lemma 18.25 and the identity
(f1 + f2 )+ + f1 + f2 = f1+ + f2+ + (f1 + f2 )
to write
(f1 + f2 )+ (f1 + f2 ) = f1+ f1 + f2+ f2 ,
we see that R (f1 +f2 ) = R (f1 )+R (f2 ). To show that R is linear, we need to verify that R (cf ) = cR (f )
for all c R. But
(cf ) = cf ,
if c 0 while
(cf )+ = c(f ),
if c < 0 and
(cf ) = (c)f +,
if c < 0. Thus, if c < 0,




R (cf ) = (cf )+ (cf ) = (c) f (c)f +
= c(f ) + c(f + ) = c((f + ) (f )).
(If this looks familiar it may be because we used this approach earlier in defining the integral of a real-valued
function.) Now let
f = R (Re f ) + iR (Im f )
for arbitrary f C(X). It is easy to see that is a positive linear functional on C(X) (= Cc (X) since X is
compact). By the Riesz representation theorem for positive linear functionals, there exists a unique Radon
measure such that
Z
f =
f d
X

for all f C(X). Thus (1) = (X). Now we present the Riesz representation theorem for C(X)0 .

18.5. THE DUAL SPACE OF C (X)

337

Theorem 18.26 Let L (C(X))0 . Then there exists a Radon measure and a function L (X, ) such
that
Z
L(f ) =
f d.
X

Proof: Let f C(X). Then there exists a unique Radon measure such that
Z
|Lf | (|f |) =
|f |d = ||f ||1 .
X

Since is a Radon measure, we know C(X) is dense in L1 (X, ). Therefore L extends uniquely to an element
of (L1 (X, ))0 . By the Riesz representation theorem for L1 , there exists a unique L (X, ) such that
Z
Lf =
f d
X

for all f C(X).

It is possible to give a simple generalization of the above theorem to locally compact Hausdorff spaces.
We will do this in the special case where X = Rn . Define
e
X

n
Y

[, ]

i=1

With the product topology where a subbasis for a topology on [, ] will consist of sets of the form
e into a metric space by using the metric,
[, a), (a, b) or (a, ]. We can also make X
(x, y)

n
X

|arctan xi arctan yi |

i=1

We also define by C0 (X) the space of continuous functions, f , defined on X such that
lim

f (x) = 0.

e
dist(x,X\X
)0

For this space of functions, ||f ||0 sup {|f (x)| : x X} is a norm which makes this into a Banach space.
Then the generalization is the following corollary.
0

Corollary 18.27 Let L (C0 (X)) where X = Rn . Then there exists L (X, ) for a Radon measure
such that for all f C0 (X),
Z
L (f ) =
f d.
X

Proof: Let
n
 
o
e f C X
e : f (z) = 0 if z X
e \X .
D
 
e is a closed subspace of the Banach space C X
e . Let : C0 (X) D
e be defined by
Thus D
f (x) =

f (x) if x X,
e \X
0 if x X

338

REPRESENTATION THEOREMS

e (||u|| = ||u|| .)It follows we have the following diagram.

Then is an isometry of C0 (X) and D.
C0 (X)

 0
e
D

 0
e
C X

C0 (X)

e
D

 
e
C X

 0
e such that i L1 = L. Now we apply Theorem
By the Hahn Banach theorem, there exists L1 C X


e and a function L X,
e 1 , such that
18.26 to get the existence of a Radon measure, 1 , on X
L1 g =

gd1 .

e
X

Letting the algebra of 1 measurable sets be denoted by S1 , we define

n
n
o
o
e \ X : E S1
S E\ X
and let be the restriction of 1 to S. If f C0 (X),

Lf = i L1 f L1 if = L1 f =

f d1 =

e
X

f d.

This proves the corollary.

18.6

Weak convergence

A very important sort of convergence in applications of functional analysis is the concept of weak or weak
convergence. It is important because it allows us to extract a convergent subsequence of a given bounded
sequence. The only problem is the convergence is very weak so it does not tell us as much as we would like.
Nevertheless, it is a very useful concept. The big theorems in the subject are the Eberlein Smulian theorem
and the Banach Alaoglu theorem about the weak or weak compactness of the closed unit balls in either a
Banach space or its dual space. These theorems are proved in Yosida [29]. Here we will consider a special
case which turns out to be by far the most important in applications and it is not hard to get from the
results of this chapter. First we define what we mean by weak and weak convergence.
Definition 18.28 Let X 0 be the dual of a Banach space X and let {xn } be a sequence of elements of X 0 .
Then we say xn converges weak to x if and only if for all x X,
lim xn (x) = x (x) .

We say a sequence in X, {xn } converges weakly to x X if and only if for all x X 0

lim x (xn ) = x (x) .

The main result is contained in the following lemma.

Lemma 18.29 Let X 0 be the dual of a Banach space, X and suppose X is separable. Then if {xn } is a
bounded sequence in X 0 , there exists a weak convergent subsequence.
Proof: Let D be a dense countable set in X. Then the sequence, {xn (x)} is bounded for all x and in
particular for all x D. Use the Cantor diagonal process to obtain a subsequence, still denoted by n such

18.7. EXERCISES

339

that xn (d) converges for each d D. Now let x X be completely arbitrary. We will show {xn (x)} is a
Cauchy sequence. Let > 0 be given and pick d D such that for all n
|xn (x) xn (d)| <

.
3

We can do this because D is dense. By the first part of the proof, there exists N such that for all m, n > N ,
|xn (d) xm (d)| <

.
3

Then for such m, n,

|xn (x) xm (x)| |xn (x) xn (d)| + |xn (d) xm (d)|

+ + = .
+ |xm (d) xm (x)| <
3 3 3
Since is arbitrary, this shows {xn (x)} is a Cauchy sequence for all x X.
Now we define f (x) limn xn (x) . Since each xn is linear, it follows f is also linear. In addition to
this,
|f (x)| = lim |xn (x)| K ||x||
n

where K is some constant which is larger than all the norms of the xn . We know such a constant exists
because we assumed that the sequence, {xn } was bounded. This proves the lemma.
The lemma implies the following important theorem.
Theorem 18.30 Let be a measurable subset of Rn and let {fk } be a bounded sequence in Lp () where
1 < p . Then there exists a weak convergent subsequence.
0

Proof: We know from Corollary 12.15 that Lp () is separable. From the Riesz representation theorem,
we obtain the desired result.
Note that from the Riesz representation theorem, it follows that if p < , then we also have the
susequence converges weakly.

18.7

Exercises

1. Suppose (E) =

f d where and are two measures and f L1 (). Show << .

2. Show that << for and two finite measures, if and only if for every > 0 there exists > 0
such that whenever (E) < , it follows (E) < .
3. Suppose , are two finite measures defined on a algebra S. Show = S + A where S and A are
finite measures satisfying
S (E) = S (E S) , (S) = 0 for some S S,
A << .
This is called the Lebesgue decomposition. Hint: This is just a generalization of the Radon Nikodym
theorem. In the proof of this theorem, let
S = {x : h(x) = 1}, S (E) = (E S),
A (E) = (E S c ).
We write S and A << in this situation.

340

REPRESENTATION THEOREMS

4. Generalize the result of Problem 3 to the case where is finite and is finite.
5. Let F be a nondecreasing right continuous, bounded function,
lim F (x) = 0.

R
Let Lf = f dF where f Cc (R) and this is just the Riemann Stieltjes integral. Let be the Radon
measure representing L. Show
((a, b]) = F (b) F (a) , ((a, b)) = F (b) F (a).
6. Using Problems 3,R4, and 5, show there is a bounded nondecreasing function G(x) such that G(x)
x
F (x) and G(x) = `(t)dt for some ` 0 , ` L1 (m). Also, if F (x) G(x) = S(x), then
R
S(x) is non decreasing and if S is the measure representing f dS, then S m. Hint: Consider
G(x) = A ((, x]).
7. Let and be two measures defined on S, a algebra of subsets of . Suppose is finite and g
0 with g measurable. Show that
Z
d
g=
, ((E) =
gd)
d
E
if and only if for all A S ,and , , 0,
(A {x : g(x) }) (A {x : g(x) }),
(A {x : g(x) < }) (A {x : g(x) < }).
Hint: To show g =

d
d

from the two conditions, use the conditions to argue that for (A) < ,
(A {x : g(x) [, )}) (A {x : g(x) [, )})
(A {x : g(x) [, )}).

8. Let r, p, q (1, ) satisfy

1
1 1
= + 1
r
p q
and let f Lp (Rn ) , g Lq (Rn ) , f, g 0. Youngs inequality says that
||f g||r ||g||q ||f ||p .
Prove Youngs inequality by justifying or modifying the steps in the following argument using Problem
27 of Chapter 12. Let


1
1
r0
n
h L (R ) .
+ 0 =1
r
r
Z

f g (x) |h (x)| dx =

Z Z

f (y) g (x y) |h (x)| dxdy.

18.7. EXERCISES

341

Let r = p so (0, 1), p0 (1 ) = q, p0 r0 . Then the above

Z Z

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

|h (x)| dydx

Z Z 

Z 

"Z Z


"Z

r0

|h (x)|

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

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

|g (x y)|

|f (y)|

q/r

= ||g||q

Z

q/p0

r

|g (x y)|

|g (x y)|

dx

1/r

dy

p/r #1/p
dx
dy

#1/r0
p0 r0 /p0
dx

|h (x)| dy

(1)p0

||g||q

r

#1/p0
r0 p0 /r0
|h (x)| dx
dy

Z

r0 1/r0
|h (x)| dx

|g (x y)|

"Z Z


"Z

|g (x y)|

"Z Z


dy

p/r #1/p
dx
dy

r0 /p0

#1/r0

dx

q/r

||g||q

||f ||p

||f ||p ||h||r0 = ||g||q ||f ||p ||h||r0 .

Therefore ||f g||r ||g||q ||f ||p . Does this inequality continue to hold if r, p, q are only assumed to be
in [1, ]? Explain.
9. Let X = [0, ] with a subbasis for the topology given by sets of the form [0, b) and (a, ]. Show that
X is a compact set. Consider all functions of the form
n
X

ak exkt

k=0

where t > 0. Show that this collection of functions is an algebra of functions in C (X) if we define
et 0,
and that it separates the points and annihilates no point. Conclude that this algebra is dense in C (X).

342

REPRESENTATION THEOREMS

10. Suppose f C (X) and for all t large enough, say t ,

Z
etx f (x) dx = 0.
0

Show that f (x) = 0 for all x (0, ). Hint: Show the measure given by d = ex dx is a regular
measure and so C (X) is dense in L2 (X, ). Now use Problem 9 to conclude that f (x) = 0 a.e.
11. Suppose f is continuous on (0, ), and for some > 0,


f (x) ex  C

for all x (0, ), and suppose also that for all t large enough,
Z
f (x) etx dx = 0.
0

Show f (x) = 0 for all x (0, ). A common procedure in elementary differential equations classes is
to obtain the Laplace transform of an unknown function and then, using a table of Laplace transforms,
find what the unknown function is. Can the result of this problem be used to justify this procedure?
12. The following theorem is called the mean ergodic theorem.
Theorem 18.31 Let (, S, ) be a finite measure space and let : satisfy 1 (E) S for all
E S. Also suppose for all positive integers, n, that


n (E) K (E) .
For f Lp () , and p 1, let
T f f .

(18.19)

||T n || K 1/p .

(18.20)

Then T L (Lp () , Lp ()) with

Defining An L (Lp () , Lp ()) by

An

n1
1X k
T ,
n
k=0

there exists A L (Lp () , Lp ()) such that for all f Lp () ,

An f Af weakly

(18.21)

and A is a projection, A2 = A, onto the space of all f Lp () such that T f = f . The norm of A
satisfies
||A|| K 1/p .

(18.22)

To prove this theorem first show (18.20) by verifying it for simple functions and then using the density
of simple functions in Lp () . Next let
n
o
M g Lp () : ||An g||p 0.

18.7. EXERCISES

343

Show M is a closed subspace of Lp () containing (I T ) (Lp ()) . Next show that if

Ank f h weakly,
R
0
then T h = h. If Lp () is such that gd = 0 for all g M, then using the definition of An , it
follows that for all k Lp () ,
Z
Z
Z
n
n
kd = (T k + (I T ) k) d = T n kd
and so
Z

kd =

An kd.

Using (18.23) show that if Ank f g weakly and Amk f h weakly, then
Z
Z
Z
Z
Z
gd = lim
Ank f d = f d = lim
Amk f d = hd.
k

(18.23)

(18.24)

Now argue that

T n g = weak lim T n Ank f = weak lim Ank T n f = g.
k

Conclude that An (g h) = g h so if g 6= h, then g h

/ M because
An (g h) g h 6= 0.
R
R
0
Now show there exists Lp () such that (g h) d 6= 0 but kd = 0 for all k M,
contradicting (18.24). Use this to conclude that if p > 1 then An f converges weakly for each f Lp () .
Let Af denote this weak limit and verify that the conclusions of the theorem hold in the case where
p > 1. To get the case where p = 1 use the theorem just proved for p > 1 on L2 () L1 () a dense
subset of L1 () along with the observation that L () L2 () due to the assumption that we are
in a finite measure space.
13. Generalize Corollary 18.27 to X = Cn .
14. Show that in a reflexive Banach space, weak and weak convergence are the same.

344

REPRESENTATION THEOREMS

Weak Derivatives
19.1

Test functions and weak derivatives

In elementary courses in mathematics, functions are often thought of as things which have a formula associated with them and it is the formula which receives the most attention. For example, in beginning calculus
courses the derivative of a function is defined as the limit of a difference quotient. We start with one function
which we tend to identify with a formula and, by taking a limit, we get another formula for the derivative.
A jump in abstraction occurs as soon as we encounter the derivative of a function of n variables where the
derivative is defined as a certain linear transformation which is determined not by a formula but by what it
does to vectors. When this is understood, we see that it reduces to the usual idea in one dimension. The
idea of weak partial derivatives goes further in the direction of defining something in terms of what it does
rather than by a formula, and extra generality is obtained when it is used. In particular, it is possible to
differentiate almost anything if we use a weak enough notion of what we mean by the derivative. This has
the advantage of letting us talk about a weak partial derivative of a function without having to agonize over
the important question of existence but it has the disadvantage of not allowing us to say very much about
this weak partial derivative. Nevertheless, it is the idea of weak partial derivatives which makes it possible
to use functional analytic techniques in the study of partial differential equations and we will show in this
chapter that the concept of weak derivative is useful for unifying the discussion of some very important
theorems. We will also show that certain things we wish were true, such as the equality of mixed partial
derivatives, are true within the context of weak derivatives.
Let Rn . A distribution on is defined to be a linear functional on Cc (), called the space of test
functions. The space of all such linear functionals will be denoted by D () . Actually, more is sometimes
done here. One imposes a topology on Cc () making it into something called a topological vector space,
and when this has been done, D0 () is defined as the dual space of this topological vector space. To see
this, consult the book by Yosida [29] or the book by Rudin [25].
Example: The space L1loc () may be considered as a subset of D () as follows.
Z
f ()
f (x) (x) dx

Cc

L1loc

for all
(). Recall that f
() if f XK L1 () whenever K is compact.

Example: x D () where x () (x).

It will be observed from the above two examples and a little thought that D () is truly enormous. We
shall define the derivative of a distribution in such a way that it agrees with the usual notion of a derivative
on those distributions which are also continuously differentiable functions. With this in mind, let f be the
restriction to of a smooth function defined on Rn . Then Dxi f makes sense and for Cc ()
Z
Z
Dxi f ()
Dxi f (x) (x) dx =
f Dxi dx = f (Dxi ).

Motivated by this we make the following definition.

345

346

WEAK DERIVATIVES

Definition 19.1 For T D ()

Dxi T () T (Dxi ).
Of course one can continue taking derivatives indefinitely. Thus,


Dxi xj T Dxi Dxj T
and it is clear that all mixed partial derivatives are equal because this holds for the functions in Cc ().
Thus we can differentiate virtually anything, even functions that may be discontinuous everywhere. However
the notion of derivative is very weak, hence the name, weak derivatives.
Example: Let = R and let

1 if x 0,
H (x)
0 if x < 0.
Then
DH () =

H (x) 0 (x) dx = (0) = 0 ().

Note that in this example, DH is not a function.

What happens when Df is a function?
Theorem 19.2 Let = (a, b) and suppose that f and Df are both in L1 (a, b). Then f is equal to a
continuous function a.e., still denoted by f and
Z x
f (x) = f (a) +
Df (t) dt.
a

In proving Theorem 19.2 we shall use the following lemma.

Lemma 19.3 Let T D (a, b) and suppose DT = 0. Then there exists a constant C such that
Z b
T () =
Cdx.
a

Proof: T (D) = 0 for all

Cc

(a, b) from the definition of DT = 0. Let

Z b
0 Cc (a, b) ,
0 (x) dx = 1,
a

and let
(x) =

[ (t)

(y) dy 0 (t)]dt

for Cc (a, b). Thus Cc (a, b) and

D =

(y) dy 0 .

Therefore,
= D +

Z
a

(y) dy 0

19.1. TEST FUNCTIONS AND WEAK DERIVATIVES

347

and so
T () = T (D ) +

(y) dy T (0 ) =

T (0 ) (y) dy.

Let C = T 0 . This proves the lemma.

Proof of Theorem 19.2 Since f and Df are both in L1 (a, b),
Df ()

Df (x) (x) dx = 0.

Consider
f ()

()

Df (t) dt

and let Cc (a, b).

D f ()

()

Df (t) dt ()

f (x) 0 (x) dx +

= Df () +

= Df ()


Df (t) dt 0 (x) dx

Z

Df (t) 0 (x) dxdt

t
b

Df (t) (t) dt = 0.

By Lemma 19.3, there exists a constant, C, such that

f ()

()

Df (t) dt () =

C (x) dx

for all Cc (a, b). Thus

Z


Z
{ f (x)


Df (t) dt C} (x) dx = 0

for all Cc (a, b). It follows from Lemma 19.6 in the next section that
Z x
f (x)
Df (t) dt C = 0 a.e. x.
a

Thus we let f (a) = C and write

f (x) = f (a) +

Z
a

This proves Theorem 19.2.

Df (t) dt.

348

WEAK DERIVATIVES

Theorem 19.2 says that

f (x) = f (a) +

Df (t) dt

Rx
whenever it makes sense to write a Df (t) dt, if Df is interpreted as a weak derivative. Somehow, this is
the way it ought to be. It follows from the fundamental theorem of calculus in Chapter 20 that f 0 (x) exists
for a.e. x where the derivative is taken in the sense of a limit of difference quotients and f 0 (x) = Df (x).
This raises an interesting question. Suppose f is continuous on [a, b] and f 0 (x) exists in the classical sense
for a.e. x. Does it follow that
Z x
f (x) = f (a) +
f 0 (t) dt?
a

The answer is no. To see an example, consider Problem 3 of Chapter 7 which gives an example of a function
which is continuous on [0, 1], has a zero derivative for a.e. x but climbs from 0 to 1 on [0, 1]. Thus this
function is not recovered from integrating its classical derivative.
In summary, if the notion of weak derivative is used, one can at least give meaning to the derivative of
almost anything, the mixed partial derivatives are always equal, and, in one dimension, one can recover the
function from integrating its derivative. None of these claims are true for the classical derivative. Thus weak
derivatives are convenient and rule out pathologies.

19.2

Weak derivatives in Lploc

Definition 19.4 We say f Lploc (Rn ) if f XK Lp whenever K is compact.

Definition 19.5 For = (k1 , , kn ) where the ki are nonnegative integers, we define
||

n
X

|kxi |, D f (x)

i=1

|| f (x)
.
xk11 xk22 xknn

We want to consider the case where u and D u for || = 1 are each in Lploc (Rn ). The next lemma is the
one alluded to in the proof of Theorem 19.2.
Lemma 19.6 Suppose f L1loc (Rn ) and suppose
Z

f dx = 0

for all Cc (Rn ). Then f ( x) = 0 a.e. x.

Proof: Without loss of generality f is real-valued. Let
E { x : f (x) > }
and let
Em E B(0, m).
We show that m (Em ) = 0. If not, there exists an open set, V , and a compact set K satisfying
K Em V B (0, m) , m (V \ K) < 41 m (Em ) ,

19.2. WEAK DERIVATIVES IN LP

LOC

349

|f | dx < 41 m (Em ) .

V \K

Let H and W be open sets satisfying

KHHW W V
and let
HgW
where the symbol, , has the same meaning as it does in Chapter 6. Then let be a mollifier and let
h g for small enough that
K h V.
Thus
0

f hdx =

f dx +

f hdx

V \K

K
1

m (K) 4 m (Em )


 m (Em ) 41 m (Em ) 41 m (Em )
21 m(Em ).
Therefore, m (Em ) = 0, a contradiction. Thus
m (E)

m (Em ) = 0

m=1

and so, since  > 0 is arbitrary,

m ({ x : f ( x) > 0}) = 0.
Similarly m ({ x : f ( x) < 0}) = 0. This proves the lemma.
This lemma allows the following definition.
Definition 19.7 We say for u L1loc (Rn ) that D u L1loc (Rn ) if there exists a function g L1loc (Rn ),
necessarily unique by Lemma 19.6, such that for all Cc (Rn ),
Z
Z
||

gdx = D u () = (1) u (D ) dx.

We call g D u when this occurs.
Lemma 19.8 Let u L1loc and suppose u,i L1loc , where the subscript on the u following the comma denotes
the ith weak partial derivative. Then if  is a mollifier and u u  , we can conclude that u,i u,i  .
Proof: If Cc (Rn ), then
Z
Z
u ( x y) ,i ( x) dx =
u (z) ,i (z + y) dz
Z
= u,i (z) (z + y) dz
Z
= u,i (x y) ( x) dx.

350

WEAK DERIVATIVES

Therefore,
u,i ()

Z Z
u ,i =
u ( x y)  ( y) ,i ( x) d ydx
Z Z
=
u ( x y) ,i ( x)  ( y) dxdy
Z Z
=
u,i ( x y) ( x)  ( y) dxdy
Z
=
u,i  ( x) ( x) dx.
=

The technical questions about product measurability in the use of Fubinis theorem may be resolved by
picking a Borel measurable representative for u. This proves the lemma.
Next we discuss a form of the product rule.
Lemma 19.9 Let C (Rn ) and suppose u, u,i Lploc (Rn ). Then (u),i and u are in Lploc and
(u),i = u,i + u ,i .
Proof: Let Cc (Rn ) then
(u),i ()

u,i dx

= u[(),i ,i ]dx
Z


=
u,i + u ,i dx.
This proves the lemma. We recall the notation for the gradient of a function.
T

u (x) (u,1 (x) u,n (x))

thus
Du (x) v =u (x) v.

19.3

Morreys inequality

The following inequality will be called Morreys inequality. It relates an expression which is given pointwise
to an integral of the pth power of the derivative.
Lemma 19.10 Let u C 1 (Rn ) and p > n. Then there exists a constant, C, depending only on n such that
for any x, y Rn ,
|u (x) u (y)|

Z
B(x,2|xy|)

|u (z) | dz

!1/p


| x y|(1n/p) .

19.3. MORREYS INEQUALITY

351

Proof: In the argument C will be a generic constant which depends on n.

Z
Z rZ
|u (x) u (y)| dy =
|u (x + ) u (x)| n1 dd
B(x,r)

S n1

0
r

S n1
Z r

S n1

= Cr

|u (x + t) |n1 ddtd

S n1

= Crn

Z
Z

S n1

|u (x + t) |n1 dtdd

Crn

|u (x + t) |dtd

Z
0

|u (x + t) | n1
t
dtd
tn1
|u (z)|

Z
B(x,r)

n1 dz.

| z x|

Thus if we define
Z
f dx =
E

1
m (E)

f dx,

then
Z

|u (y) u (x)| dy C

B(x,r)

1n

|u (z) ||z x|

dz.

(19.1)

B(x,r)

Now let r = |x y| and

U = B (x, r), V = B (y, r), W = U V.
Thus W equals the intersection of two balls of radius r with the center of one on the boundary of the other.
It is clear there exists a constant, C, depending only on n such that
m (W )
m (W )
=
= C.
m (U )
m (V )
Then from (19.1),
Z
|u (x) u (y)| =

|u (x) u (y)| dz

Z
|u (x) u (z)| dz +

C
m (U )

|u (z) u (y)| dz

Z

|u (x) u (z)| dz +

|u (z) u (y)| dz

Z

Z
C |u (x) u (z)| dz + |u (y) u (z)| dz
U

352

WEAK DERIVATIVES
C

Z

1n

|u (z) ||z x|

dz +

1n

|u (z) ||z y|

|u (z) | dz

=C

dz

!(p1)/p

B(x,r)

|u (z) | dz

!1/p Z

B(x,r)

=C

1n p/(p1)

(|z x|

B(x,r)

(19.2)

Consider the first of these two integrals. This is no smaller than

!1/p Z
Z
C


dz .

|u (z) | dz

p(1n)/(p1) n1

dd

(p1)/p

S n1

!1/p Z

B(x,r)

(1n)/(p1) d

(p1)/p

=C

|u (z) | dz

!1/p

r(1n/p)

(19.3)

B(x,r)

|u (z) | dz

!1/p

r(1n/p).

(19.4)

B(x,2|xy|)

The second integral in (19.2) is dominated by the same expression found in (19.3) except the ball over which
the integral is taken is centered at y not x. Thus this integral is also dominated by the expression in (19.4)
and so,
!1/p
Z
|u (z) |p dz

|u (x) u (y)| C

(1n/p)

|x y|

(19.5)

B(x,2|xy|)

which proves the lemma.

19.4

Rademachers theorem

Next we extend this inequality to the case where we only have u and u,i in Lploc for p > n. This leads to an
elegant proof of the differentiability a.e. of a Lipschitz continuous function. Let k Cc (Rn ) , k 0, and
k (z) = 1 for all z B (0, k). Then
u k , (u k ),i Lp (Rn ).
Let  be a mollifier and consider
(u k ) u k  .
By Lemma 19.8,
(u k ),i = (u k ),i  .
Therefore
(u k ),i (u k ),i in Lp (Rn )

(19.6)

19.4. RADEMACHERS THEOREM

353

and
(u k ) u k in Lp (Rn )

(19.7)

as  0. By (19.7), there exists a subsequence  0 such that

(u k ) (z) u k (z) a.e.

(19.8)

(u k ) (z) u (z)

(19.9)

Since k (z) = 1 for |z| < k, this shows

and for a.e. z with |z| <k. Denoting the exceptional set of (19.9) by Ek , let

x, y
/ k=1 Ek E.
Also let k be so large that
B (0,k) B (x,2|x y|).
Then by (19.5),
|(u k ) (x) (u k ) (y)|

|(u k ) |p dz

!1/p

(1n/p)

|x y|

B(x,2|yx|)

where C depends only on n. Now by (19.8), there exists a subsequence,  0, such that (19.9), holds for
z = x, y. Thus, from (19.6),
|u (x) u (y)| C

|u| dz

!1/p

(1n/p)

|x y|

(19.10)

B(x,2|yx|)

Redefining u on E, in the case where p > n, we can obtain (19.10) for all x, y. This has proved the following
theorem.
Theorem 19.11 Suppose u, u,i Lploc (Rn ) for i = 1, , n and p > n. Then u has a representative, still
denoted by u, such that for all x, y Rn ,
|u (x) u (y)| C

|u|p dz

!1/p

(1n/p)

|x y|

B(x,2|yx|)

The next corollary is a very remarkable result. It says that not only is u continuous by virtue of having
weak partial derivatives in Lp for large p, but also it is differentiable a.e.
Corollary 19.12 Let u, u,i Lploc (Rn ) for i = 1, , n and p > n. Then the representative of u described
in Theorem 19.11 is differentiable a.e.
Proof: Consider
|u (y) u (x) u (x) (y x)|

354

WEAK DERIVATIVES

where u,i is a representative of u,i , an element of Lp . Define

g (z) u (z) + u (x) (y z).
Then the above expression is of the form
|g (y) g (x) |
and
g (z) = u (z) u (x).
Therefore g Lploc (Rn ) and g,i Lploc (Rn ). It follows from Theorem 19.11 that
|g (y) g (x) | = |u (y) u (x) u (x) (y x)|

|g (z) | dz

!1/p

(1n/p)

|x y|

B(x,2|yx|)

=C

|u (z) u (x) |p dz

!1/p

(1n/p)

|x y|

B(x,2|yx|)

=C

|u (z) u (x) | dz

!1/p

|x y|.

B(x,2|yx|)

This last expression is o (|y x|) at every Lebesgue point, x, of u. This proves the corollary and shows
u is the gradient a.e.
Now suppose u is Lipschitz on Rn ,
|u (x) u (y)| K |x y|
for some constant K. We define Lip (u) as the smallest value of K that works in this inequality. The following
corollary is known as Rademachers theorem. It states that every Lipschitz function is differentiable a.e.
Corollary 19.13 If u is Lipschitz continuous then u is differentiable a.e. and ||u,i || Lip (u).
Proof: We do this by showing that Lipschitz continuous functions have weak derivatives in L (Rn )
and then using the previous results. Let
Dehi u (x) h1 [u (x + hei ) u (x)].
Then Dehi u is bounded in L (Rn ) and
||Dehi u|| Lip (u).
It follows that Dehi u is contained in a ball in L (Rn ), the dual space of L1 (Rn ). By Theorem 18.30 there
is a subsequence h 0 such that
Dehi u * w, ||w|| Lip (u)

(19.11)

19.5. EXERCISES

355

where the convergence takes place in the weak topology of L (Rn ). Let Cc (Rn ). Then
Z
Z
wdx = lim Dehi udx
h0

= lim

h0

u (x)
Z

( (x hei ) (x))
dx
h
u (x) ,i (x) dx.

Thus w = u,i and we see that u,i L (Rn ) for each i. Hence u, u,i Lploc (Rn ) for all p > n and so u is
differentiable a.e. and u is given in terms of the weak derivatives of u by Corollary 19.12. This proves the
corollary.

19.5

Exercises

1. Let K be a bounded subset of Lp (Rn ) and suppose that for all  > 0, there exist a > 0 and G such
that G is compact such that if |h| < , then
Z
p
|u (x + h) u (x)| dx < p
for all u K and
Z

|u (x)| dx < p

Rn \G

for all u K. Show that K is precompact in Lp (Rn ). Hint: Let k be a mollifier and consider
Kk {u k : u K} .
Verify the conditions of the Ascoli Arzela theorem for these functions defined on G and show there is
an  net for each  > 0. Can you modify this to let an arbitrary open set take the place of Rn ?
2. In (19.11), why is ||w|| Lip (u)?
3. Suppose Dehi u is bounded in Lp (Rn ) for p > n. Can we conclude as in Corollary 19.13 that u is
differentiable a.e.?
4. Show that a closed subspace of a reflexive Banach space is reflexive. Hint: The proof of this is an
exercise in the use of the Hahn Banach theorem. Let Y be the closed subspace of the reflexive space
X and let y Y 00 . Then i y X 00 and so i y = Jx for some x X because X is reflexive.
Now argue that x Y as follows. If x
/ Y , then there exists x such that x (Y ) = 0 but x (x) 6= 0.

Thus, i x = 0. Use this to get a contradiction. When you know that x = y Y , the Hahn Banach
theorem implies i is onto Y 0 and for all x X 0 ,
y (i x ) = i y (x ) = Jx (x ) = x (x) = x (iy) = i x (y).
5. For an arbitrary open set U Rn , we define X 1p (U ) as the set of all functions in Lp (U ) whose weak
partial derivatives are also in Lp (U ). Here we say a function in Lp (U ) , g equals u,i if and only if
Z
Z
gdx =
u,i dx
U

356

WEAK DERIVATIVES
for all Cc (U ). The norm in this space is given by
Z
1/p
p
p
||u||1p
|u| + |u| dx
.
U

Then we define the Sobolev space W (U ) to be the closure of C U in X 1p (U ) where C U is
defined to be restrictions of all functions in Cc (Rn ) to U . Show that this definition of weak derivative
is well defined and that X 1p (U ) is a reflexive Banach space. Hint: To do this, show the operator
n+1
u u,i is a closed operator and that X 1p (U ) can be considered as a closed subspace of Lp (U )
.
Show that in general the product of reflexive spaces is reflexive and then use Problem 4 above which
states that a closed subspace of a reflexive Banach space is reflexive. Thus, conclude that W 1p (U ) is
also a reflexive Banach space.
1p

6. Theorem 19.11 shows that if the weak derivatives of a function u Lp (Rn ) are in Lp (Rn ) , for p > n,
then the function has a continuous representative. (In fact, one can conclude more than continuity
from this theorem.) It is also important to consider the case when p < n. To aid in the study of this
case which will be carried out in the next few problems, show the following inequality for n 2.
1/n1
Z Y
n
n Z
Y
n1
|wj (x)| dmn
|wj (x)|
dmn1
Rn j=1

Rn1

i=1

where wj does not depend on the jth component of x, xj . Hint: First show it is true for n = 2 and
then use Holders inequality and induction. You might benefit from first trying the case n = 3 to get
the idea.
7. Show that if Cc (Rn ), then

n 
1 X  
||||n/(n1)
.
n
n j=1  xj 1

Hint: First show that if ai 0, then

n
Y

1 X

ai .
n
n j=1

1/n

ai

i=1

Then observe that

| (x)|

so
n/(n1)

||||n/(n1) =

Z Y
n Z
j=1



,j (x) dxj
n/(n1)

||



,j (x) dxj

dmn

1/(n1)

dmn

1/(n1)
n Z
Y


,j (x) dmn
.

j=1

Hence
||||n/(n1)

1/n
n Z
Y


,j (x) dmn
.

j=1

19.5. EXERCISES

357

8. Show that if Cc (Rn ), then if

1
q

1
p

n1 , where p < n, then

n

1 (n 1) p X  
||||q

.
n
n n p j=1 ,i p
Also show that if u W 1p (Rn ), then u Lq (Rn ) and the inclusion map is continuous. This is part of
the Sobolev embedding theorem. For more on Sobolev spaces see Adams [1]. Hint: Let r > 1. Then
r
|| Cc (Rn ) and



 r
r1 
,i  .
||,i  = r ||
Now apply the result of Problem 7 to write
Z

||

rn
n1

r X

n
n i=1

Z

dmn

(n1)/n

 p
,i 

r X

n
n i=1

1/p Z 

r1

||

r1

||

p/(p1)

 
,i  dmn

dmn

(p1)/p

Now choose r such that

(r 1) p
rn
=
p1
n1
so that the last term on the right can be cancelled with the first term on the left and simplify.

358

WEAK DERIVATIVES

Fundamental Theorem of Calculus

One of the most remarkable theorems in Lebesgue integration is the Lebesgue fundamental theorem of
calculus which says that if f is a function in L1 , then the indefinite integral,
Z x
x
f (t) dt,
a

can be differentiated for a.e. x and gives f (x) a.e. This is a very significant generalization of the usual
fundamental theorem of calculus found in calculus. To prove this theorem, we use a covering theorem due to
Vitali and theorems on maximal functions. This approach leads to very general results without very many
painful technicalities. We will be in the context of (Rn , S, m) where m is n-dimensional Lebesgue measure.
When this important theorem is established, it will be used to prove the very useful theorem about change
of variables in multiple integrals.
By Lemma 6.6 of Chapter 6 and the completeness of m, we know that the Lebesgue measurable sets are
exactly those measurable in the sense of Caratheodory. Also, we can regard m as an outer measure defined
on all of P(Rn ). We will use the following notation.
B(p, r) = {x : |x p| < r}.

(20.1)

If
= B(p, 5r).
B = B(p, r), then B

20.1

The Vitali covering theorem

Lemma 20.1 Let F be a collection of balls as in (20.1). Suppose

> M sup{r : B(p, r) F} > 0.
Then there exists G F such that
if B(p, r) G then r >

M
,
2

if B1 , B2 G then B1 B2 = ,

(20.2)
(20.3)

G is maximal with respect to Formulas (20.2) and (20.3).

Proof: Let H = {B F such that (20.2) and (20.3) hold}. Obviously H =
6 because there exists
.
Partially
order
H
by
set
inclusion
and
use
the
Hausdorff
maximal theorem (see
B(p, r) F with r > M
2
the appendix on set theory) to let C be a maximal chain in H. Clearly C satisfies (20.2) and (20.3). If C
is not maximal with respect to these two properties, then C was not a maximal chain. Let G = C.
359

360

FUNDAMENTAL THEOREM OF CALCULUS

Theorem 20.2 (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: Let G1 F satisfy

B(p, r) G1 implies r >

M
,
2

B1 , B2 G1 implies B1 B2 = ,

(20.4)
(20.5)

G1 is maximal with respect to Formulas (20.4), and (20.5).

Suppose G1 , , Gm1 have been chosen, m 2. Let
Fm = {B F : B Rn \ {G1 Gm1 }}.
Let Gm Fm satisfy the following.
B(p, r) Gm implies r >

M
,
2m

B1 , B2 Gm implies B1 B2 = ,

(20.6)
(20.7)

Gm is a maximal subset of Fm with respect to Formulas (20.6) and (20.7).

If Fm = , Gm = . Define
G
k=1 Gk .
b : B G} covers A. Let x A. Then
Thus G is a collection of disjoint balls in F. We need to show {B
x B(p, r) F. Pick m such that
M
M
< r m1 .
m
2
2
b for some B G1 Gm . To see this, note that B(p, r) must intersect some set of
We claim x B
G1 Gm because if it didnt, then
0
Gm {B(p, r)} = Gm
0
would satisfy Formulas (20.6) and (20.7), and Gm
% Gm contradicting the maximality of Gm . Let the set
intersected be B(p0 , r0 ). Thus r0 > M 2m .

r0
p0

.x
p
?

Then if x B(p, r),

|x p0 |

|x p| + |p p0 | < r + r0 + r
2M

+ r0 < 4r0 + r0 = 5r0

2m1
since r0 > M/2m . Hence B(p, r) B(p0 , 5r0 ) and this proves the theorem.

20.2. DIFFERENTIATION WITH RESPECT TO LEBESGUE MEASURE

20.2

361

Differentiation with respect to Lebesgue measure

The covering theorem just presented will now be used to establish the fundamental theorem of calculus. In
discussing this, we introduce the space of functions which is locally integrable in the following definition.
This space of functions is the most general one for which the maximal function defined below makes sense.
Definition 20.3 f L1loc (Rn ) means f XB(0,R) L1 (Rn ) for all R > 0. For f L1loc (Rn ), the Hardy
Littlewood Maximal Function, M f , is defined by
Z
1
M f (x) sup
|f (y)|dy.
r>0 m(B(x, r)) B(x,r)
Theorem 20.4 If f L1 (Rn ), then for > 0,
m([M f > ])

5n
||f ||1 .

(Here and elsewhere, [M f > ] {x Rn : M f (x) > } with other occurrences of [ ] being defined
similarly.)
Proof: Let S [M f > ]. For x S, choose rx > 0 with
Z
1
|f | dm > .
m(B(x, rx )) B(x,rx )
The rx are all bounded because
1
m(B(x, rx )) <

|f | dm <

B(x,rx )

1
||f ||1 .

By the Vitali covering theorem, there are disjoint balls B(xi , ri ) such that
S xS B(x, rx )
i=1 B(xi , 5ri )
and
1
m(B(xi , ri ))

|f | dm > .

B(xi ,ri )

Therefore
m(S)

m(B(xi , 5ri )) = 5n

i=1

m(B(xi , ri ))

i=1

Z
5n X
|f | dm
i=1 B(xi ,ri )
Z
5n
|f | dm,
Rn

the last inequality being valid because the balls B(xi , ri ) are disjoint. This proves the theorem.
Lemma 20.5 Let f 0, and f L1 , then
1
r0 m(B(x, r))
lim

Z
B(x,r)

f (y)dy = f (x) a.e. x.

362

FUNDAMENTAL THEOREM OF CALCULUS

Proof: Let > 0 and let



#
Z


1


f (y)dy f (x) > .
B = lim sup 

r0  m(B(x, r)) B(x,r)
"

Then for any g Cc (Rn ), B equals



"
#
Z


1


lim sup 
f (y) g(y)dy (f (x) g(x)) >

r0  m(B(x, r)) B(x,r)

because for any g Cc (Rn ),

1
lim
r0 m(B(x, r))

g(y)dy = g(x).

B(x,r)

Thus
B [M (|f g|) + |f g| > ],
and so
h
i h
i
B M (|f g|) >
|f g| >
.
2
2
Now
h
i
m |f g| >
=
2
2
2

[|f g|>
2]

dx

[|f g|>
2]

|f g|dx ||f g||1 .

Therefore by Theorem 20.4,

m(B )

2(5n )
2
+

||f g||1 .

Since Cc (Rn ) is dense in L1 (Rn ) and g is arbitrary, this estimate shows m(B ) = 0. It follows by Lemma
6.6, since B is measurable in the sense of Caratheodory, that B is Lebesgue measurable and m(B ) = 0.


"
#
Z


1


1
lim sup 
f (y)dy f (x) > 0
(20.8)
m=1 B m

r0  m(B(x, r)) B(x,r)
and each set B1/m has measure 0 so the set on the left in (20.8) is also Lebesgue measurable and has measure
0. Thus, if x is not in this set,


Z


1


0 = lim sup 
f (y)dy f (x)

r0  m(B(x, r)) B(x,r)


Z


1


lim inf 
f (y)dy f (x) 0.
r0  m(B(x, r)) B(x,r)


This proves the lemma.

20.2. DIFFERENTIATION WITH RESPECT TO LEBESGUE MEASURE

Corollary 20.6 If f 0 and f L1loc (Rn ), then
Z
1
lim
f (y)dy = f (x) a.e. x.
r0 m(B(x, r)) B(x,r)

363

(20.9)

Proof: Apply Lemma 20.5 to f XB(0,R) for R = 1, 2, 3, . Thus (20.9) holds for a.e. x B(0, R) for
each R = 1, 2, .
Theorem 20.7 (Fundamental Theorem of Calculus) Let f L1loc (Rn ). Then there exists a set of measure
0, B, such that if x
/ B, then
Z
1
lim
|f (y) f (x)|dy = 0.
r0 m(B(x, r)) B(x,r)
Proof: Let {di }
i=1 be a countable dense subset of C. By Corollary 20.6, there exists a set of measure 0,
Bi , such that if x
/ Bi
Z
1
lim
|f (y) di |dy = |f (x) di |.
(20.10)
r0 m(B(x, r)) B(x,r)
Let B =
/ B. Pick di such that |f (x) di | < 2 . Then
i=1 Bi and let x
Z
Z
1
1
|f (y) f (x)|dy
|f (y) di |dy
m(B(x, r)) B(x,r)
m(B(x, r)) B(x,r)
1
+
m(B(x, r))

|f (x) di |dy

B(x,r)

1
+
2 m(B(x, r))

|f (y) di |dy.

B(x,r)

By (20.10)
1
m(B(x, r))

|f (y) f (x)|dy

B(x,r)

whenever r is small enough. This proves the theorem.

Definition 20.8 For B the set of Theorem 20.7, B C is called the Lebesgue set or the set of Lebesgue points.
Let f L1loc (Rn ). Then
1
lim
r0 m(B(x, r))

f (y)dy = f (x) a.e. x.

B(x,r)

The next corollary is a one dimensional version of what was just presented.
Corollary 20.9 Let f L1 (R) and let
F (x) =

Then for a.e. x, F 0 (x) = f (x).

f (t)dt.

364

FUNDAMENTAL THEOREM OF CALCULUS

Proof: For h > 0

1
h

x+h

|f (y) f (x)|dy 2(

1
)
2h

x+h

|f (y) f (x)|dy

xh

By Theorem 20.7, this converges to 0 a.e. Similarly

Z
1 x
|f (y) f (x)|dy
h xh
converges to 0 a.e. x.


Z
 F (x + h) F (x)
 1 x+h



f
(x)
|f (y) f (x)|dy

 h
h
x

and



Z
 1 x
 F (x) F (x h)



f
(x)
|f (y) f (x)|dy.

 h
h
xh

Therefore,

lim

h0

F (x + h) F (x)
= f (x) a.e. x
h

This proves the corollary.

20.3

The change of variables formula for Lipschitz maps

This section is on a generalization of the change of variables formula for multiple integrals presented in
Chapter 11. In this section, will be a Lebesgue measurable set in Rn and h : Rn will be Lipschitz.
We recall Rademachers theorem a proof of which was given in Chapter 19.
Theorem 20.10 Let f :Rn Rm be Lipschitz. Then Df (x) exists a.e.and ||fi,j || Lip (f ) .
It turns out that a Lipschitz function defined on some subset of Rn always has a Lipschitz extension to
all of Rn . The next theorem gives a proof of this. For more on this sort of theorem we refer to [12].
Theorem 20.11 If h : Rm is Lipschitz, then there exists h : Rn Rm which extends h and is also
Lipschitz.
Proof: It suffices to assume m = 1 because if this is shown, it may be applied to the components of h
to get the desired result. Suppose
|h (x) h (y)| K |x y|.

(20.11)

(x) inf{h (w) + K |x w| : w }.

h

(20.12)

Define

If x , then for all w ,

h (w) + K |x w| h (x)
(x). But also we can take w = x in (20.12) which yields h
(x) h (x).
by (20.11). This shows h (x) h

Therefore h (x) = h (x) if x .

20.3. THE CHANGE OF VARIABLES FORMULA FOR LIPSCHITZ MAPS

365



(y). Without loss of generality we may assume h
(x)
Now suppose x, y Rn and consider 
h (x) h
(y) . (If not, repeat the following argument with x and y interchanged.) Pick w such that
h
(y).
h (w) + K |y w|  < h
Then


(y) = h
(x) h
(y) h (w) + K |x w|

h (x) h
[h (w) + K |y w| ] K |x y| + .
Since  is arbitrary,


(y) K |x y|

h (x) h
and this proves the theorem.
We will use h to denote a Lipschitz extension of the Lipschitz function h. From now on h will denote a
Lipschitz map from a measurable set in Rn to Rn . The next lemma is an application of the Vitali covering
theorem. It states that every open set can be filled with disjoint balls except for a set of measure zero.
Lemma 20.12 Let V be an open set in Rr , mr (V ) < . Then there exists a sequence of disjoint open balls
{Bi } having radii less than and a set of measure 0, T , such that
V = (
i=1 Bi ) T.
Proof: This is left as a problem. See Problem 8 in this chapter.
We wish to show that h maps Lebesgue measurable sets to Lebesgue measurable sets. In showing this
the key result is the next lemma which states that h maps sets of measure zero to sets of measure zero.


Lemma 20.13 If mn (T ) = 0 then mn h (T ) = 0.
Proof: Let V be an open set containing T whose measure is less than . Now using the Vitali covering
theorem, there exists a sequence
n ofodisjoint balls {Bi }, Bi = B (xi , ri ), which are contained in V such that
bi , having the same center but 5 times the radius, covers T . Then
the sequence of enlarged balls, B
 



b
mn h (T ) mn h
i=1 Bi

X
i=1

X
i=1

  
bi
mn h B

n n n


n X
(n) Lip h
5 ri = 5n Lip h
mn (Bi )
i=1

n n


n n
Lip h
5 mn (V )  Lip h
5.

Since  is arbitrary, this proves the lemma.

Actually, the argument in this lemma holds in other contexts which do not imply h is Lipschitz continuous.
For one such example, see Problem 23.
With the conclusion of this lemma, the next lemma is fairly easy to obtain.

366

FUNDAMENTAL THEOREM OF CALCULUS

Lemma 20.14 If A is Lebesgue measurable, then h (A ) is Lebesgue measurable. Furthermore,

(A) Lip h n mn (A).
mn h

(20.13)

Proof: Let Ak = A B (0, k) , k N. We establish (20.13) for Ak in place of A and then let k
to obtain (20.13). Let V Ak and let mn (V ) < . By Lemma 20.12, there is a sequence of disjoint balls
{Bi }, and a set of measure 0, T , such that
V =
i=1 Bi T, Bi = B(xi , ri ).
Then by Lemma 20.13,




(Ak ) mn h
(V )
mn h






(

mn h
i=1 Bi ) + mn h (T ) = mn h (i=1 Bi )

X
i=1

X
i=1

X




(Bi )
(xi ) , Lip h
ri
mn h
mn B h
i=1

X


ri n = Lip h
n
n mn (V ).
(n) Lip h
mn (Bi ) = Lip h
i=1

Since V is an arbitrary open set containing Ak , it follows from regularity of Lebesgue measure that




(Ak ) Lip h
n mn (Ak ).
mn h
(20.14)
(A) is Lebesgue measurable.
Now let k to obtain (20.13). This proves the formula. It remains to show h
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 = Ak .
(F ) h
(Ak ) h
(F ) h
(T ). By continuity of h,
h
(F ) is a countable union of compact sets and so
Then h
it is Borel. By (20.14) with T in place of Ak ,


(T ) = 0
mn h
(T ) is Lebesgue measurable. Therefore, h
(Ak ) is Lebesgue measurable because mn is a complete
and so h
(Ak ) between two Lebesgue measurable sets whose difference has measure
measure and we have exhibited h
0. Now
(A) =

h
k=1 h (Ak )
(A) is also Lebesgue measurable and this proves the lemma.
so h
The following lemma, found in Rudin [25], is interesting for its own sake and will serve as the basis for
many of the theorems and lemmas which follow. Its proof is based on the Brouwer fixed point theorem, a
short proof of which is given in the chapter on the Brouwer degree. The idea is that if a continuous function
mapping a ball in Rk to Rk doesnt move any point very much, then the image of the ball must contain a
slightly smaller ball.

20.3. THE CHANGE OF VARIABLES FORMULA FOR LIPSCHITZ MAPS

367

Lemma 20.15 Let B = B (0, r), a ball in Rk and let F : B Rk be continuous and suppose for some
 < 1,
|F (v) v| < r
for all v B. Then


F B B (0, r (1 )).


Proof: Suppose a B (0, r (1 )) \ F B and let
G (v)

r (a F (v))
.
|a F (v)|

If |v| = r,
v (a F (v)) = v a v F (v)
= v a v (F (v) v) r2
< r2 (1 ) + r2 r2 = 0.
Then for |v| = r, G (v) 6= v because we just showed that v G (v) < 0 but v v =r2 > 0. If |v| < r, it
follows that G (v) 6= v because |G (v)| = r but |v| < r. This lack of a fixed point contradicts the Brouwer
fixed point theorem and this proves the lemma.
We are interested in generalizing the change of variables formula. Since h is only Lipschitz, Dh (x) may
(x) exists a.e. x.
not exist for all x but from the theorem of Rademacher Dh
In the arguments below, we will define a measure and use the Radon Nikodym theorem to obtain a
function which is of interest to us. Then we will identify this function. In order to do this, we need some
technical lemmas.
(x)1 and Dh
(x) exists. Then if  (0, 1) the following
Lemma 20.16 Let x be a point where Dh
hold for all r small enough.
 





(x) B (0, r (1 )) ,
mn h B (x,r) = mn h (B (x,r)) mn Dh
(20.15)
(B (x, r)) h
(x) + Dh
(x) B (0, r (1 + )),
h

(20.16)

(B (x,r)) mn Dh
(x) B (0, r (1 + ))
mn h

(20.17)

If x is also a point of density of , then

(B (x, r) )
mn h

= 1.
lim
(B (x, r))
r0
mn h

(20.18)

(x) exists,
Proof: Since Dh
(x + v)
h

(x) + Dh
(x) v+o (|v|)
= h


(x) + Dh
(x) v+Dh
(x)1 o (|v|)
= h

(20.19)
(20.20)

368

FUNDAMENTAL THEOREM OF CALCULUS

Consequently, when r is small enough, (20.16) holds. Therefore, (20.17) holds. From (20.20),
(x + v) = h
(x) + Dh
(x) (v+o (|v|)).
h
1

(x)
Thus, from the assumption that Dh
1

(x)
Dh

exists,

(x + v) Dh
(x)1 h
(x) v =o(|v|).
h

(20.21)

Letting
(x + v) Dh
(x),
(x)1 h
(x)1 h
F (v) = Dh
we can apply Lemma 20.15 in (20.21) to conclude that for r small enough,
1

(x)
Dh

(x + v) Dh
(x) B (0, (1 ) r).
(x)1 h
h

Therefore,


(x) + Dh
(x) B (0, (1 ) r)
B (x,r) h
h
which implies
 



(x) B (0, r (1 ))
mn h B (x,r) mn Dh
which shows (20.15).
Now suppose that x is also a point of density of . Then whenever r is small enough,
mn (B (x,r) \ ) <  (n) rn.
Then for such r we write
(B (x, r) )
mn h


1
(B (x, r))
mn h

(B (x, r)) mn h
(B (x,r) \ )
mn h


.
(B (x, r))
mn h

From Lemma 20.14, and (20.15), this is no larger than

n  (n) rn
Lip h


1
(x) B (0, r (1 )) .
mn D h
By the theorem on the change of variables for a linear map, this expression equals


n  (n) rn
Lip h


1 
(x)  rn (n) (1 )n 1 g ()
det Dh
where lim0 g () = 0. Then for all r small enough,

(B (x, r) )
mn h

1 g ()
1
(B (x, r))
mn h
which proves the lemma since  is arbitrary.



(x)  .
For simplicity in notation, we write J (x) for the expression det Dh

(20.22)

20.3. THE CHANGE OF VARIABLES FORMULA FOR LIPSCHITZ MAPS

369

(x) does not exist }. Then N has measure zero and if x N

Theorem 20.17 Let N {x : Dh
/ then


mn h (B (x, r))
J (x) = lim
.
(20.23)
r0
mn (B (x,r))
1

(x)
Proof: Suppose first that Dh
for linear maps,
n

J (x) (1 )

exists. Using (20.15), (20.17) and the change of variables formula

(x) B (0,r (1 ))
(x, r))
mn Dh
mn h(B

mn (B (x, r))
mn (B (x, r))

mn Dh (x) B (0,r (1 + ))

n
= J (x) (1 + )
mn (B (x, r))

whenever r is small enough. It follows that since  > 0 is arbitrary, (20.23) holds.
(x)1 does not exist. Then from the definition of the derivative,
Now suppose Dh
h (x + v) = h (x) + Dh (x) v + o (v) ,
and so for all r small enough, h (B (x, r)) lies in a cylinder having height r and diameter no more than


Dh (x) 2r (1 + ) .
Therefore, for such r,







Dh (x) r (1 + ) n1 r
mn h (B (x, r))
C

mn (B (x,r))
rn
Since is arbitrary,


mn h (B (x, r))
J (x) = 0 = lim
.
r0
mn (B (x,r))
This proves the theorem.
We define the following two sets for future reference
1

(x) exists but Dh

(x)
S {x : Dh

does not exist}

(x) does not exist},

N {x : Dh

(20.24)
(20.25)

and we assume for now that h : Rn is one to one and Lipschitz. Since h is one to one, Lemma 20.14
implies we can define a measure, , on the algebra of Lebesgue measurable sets as follows.
(E) mn (h (E )).
By Lemma 20.14, we see this is a measure and << mn . Therefore by the corollary to the Radon Nikodym
theorem, Corollary 18.3, there exists f L1loc (Rn ) , f 0, f (x) = 0 if x
/ , and
Z
Z
(E) =
f dm =
f dm.
E

We want to identify f . Define

Q {x : x is not a point of density of } N

370

FUNDAMENTAL THEOREM OF CALCULUS

{x : x is not a Lebesgue point of f }.

Then E is a set of measure zero and if x ( \ Q) S C , Lemma 20.16 and Theorem 20.17 imply


Z
mn h (B (x,r) )
1
f (x) = lim
f (y) dm = lim
r0
r0 mn (B (x,r)) B(x,r)
mn (B (x,r))




mn h (B (x,r) ) mn h (B (x,r))


= lim
= J (x).
r0
mn (B (x,r))
mn h (B (x,r))
On the other hand, if x ( \ Q) S, then by Theorem 20.17,
mn h (B (x,r) )
f (x)
f (y) dm = lim
r0
mn (B (x,r))
B(x,r)


mn h (B (x,r))
lim
= J (x) = 0.
r0
mn (B (x,r))
1
= lim
r0 mn (B (x,r))

Therefore, f (x) = J (x) a.e., whenever x \ Q.

Now let F be a Borel measurable set in Rn . Recall this implies F is Lebesgue measurable. Then
Z
Z



XF (y) dmn = XF h() (y) dmn = mn h h1 (F )
h()

= h

(F ) =

Xh1 (F ) (x) J (x) dmn =

XF (h (x)) J (x) dmn .

(20.26)

Can we write a similar formula for F only mn measurable? Note that there are no measurability questions
in the above formula because h1 (F ) is a Borel set due to the continuity of h but it is not clear h1 (F ) is
measurable for F only Lebesgue measurable.
First consider the case where E is only Lebesgue measurable but
mn (E h ()) = 0.
By regularity of Lebesgue measure, there exists a Borel set F E h () such that
mn (F ) = mn (E h ()) = 0.
Then from (20.26),
Xh1 (F ) (x) J (x) = 0 a.e.
But
0 Xh1 (E) (x) J (x) Xh1 (F ) (x) J (x)
which shows the two functions in (20.27) are equal a.e. Therefore
Xh1 (E) (x) J (x)
is Lebesgue measurable and so from (20.26),
Z
Z
0 = XEh() (y) dmn = XF h() (y) dmn

(20.27)

20.3. THE CHANGE OF VARIABLES FORMULA FOR LIPSCHITZ MAPS

=

Xh1 (F ) (x) J (x) dm =

Xh1 (E) (x) J (x) dm,

371
(20.28)

which shows (20.26) holds in this case where

mn (E h ()) = 0.
Now let R B (0,R) where R is large enough that R 6= and let E be mn measurable. By
regularity of Lebesgue measure, there exists F E h (R ) such that F is Borel and
mn (F \ (E h (R ))) = 0.

(20.29)

Now
(E h (R )) (F \ (E h (R )) h (R )) = F h (R )
and so
XR h1 (F ) J = XR h1 (E) J + XR h1 (F \(Eh(R ))) J
where from (20.29) and (20.28), the second function on the right of the equal sign is Lebesgue measurable
and equals zero a.e. Therefore, by completeness of Lebesgue measure, the first function on the right of the
equal sign is also Lebesgue measurable and equals the function on the left a.e. Thus,
Z
Z
XEh(R ) (y) dmn = XF h(R ) (y) dmn

XR h1 (F ) (x) J (x) dmn =

XR h1 (E) (x) J (x) dmn .

(20.30)

Letting R we obtain (20.30) with replacing R and the function

x Xh1 (E) (x) J (x)
is Lebesgue measurable. Writing this in a more familiar form yields
Z
Z
XE (y) dmn =
XE (h (x)) J (x) dm.
h()

(20.31)

From this, it follows that if s is a nonnegative mn measurable simple function, (20.31) continues to be valid
with s in place of XE . Then approximating an arbitrary nonnegative mn measurable function, g, by an
increasing sequence of simple functions, it follows that (20.31) holds with g in place of XE and there are
no measurability problems because x g (h (x)) J (x) is Lebesgue measurable. This proves the following
change of variables theorem.
Theorem 20.18 Let g : h () [0, ] be Lebesgue measurable where h is one to one and Lipschitz on ,
and is a Lebesgue measurable set. Then if J (x) is defined to equal 0 for x N ,
x (g h) (x) J (x)
is Lebesgue measurable and
Z
h()

g (y) dmn =

g (h (x)) J (x) dmn .

For another version of this theorem based on the same arguments given here, in which the function h is
not assumed to be Lipschitz, see Problems 23 - 26.

372

20.4

FUNDAMENTAL THEOREM OF CALCULUS

Mappings that are not one to one

In this section, h :Rn will only be Lipschitz. We drop the requirement that h be one to one. Let S and
N be given in (20.24) and (20.25). The following lemma is a version of Sards theorem.
Lemma 20.19 For S defined above, mn (h (S)) = 0.
Proof: From Theorem 20.17, whenever x S and r is small enough,


mn h (B (x,r))
< .
mn (B (x,r))
Therefore, whenever x S and r small enough,


mn h (B (x,r))  (n) rn.

(20.32)

Let Sk = S B (0,k) and for each x Sk , let rx be such that (20.32) holds with r replaced by 5rx and
B (x,rx ) B (0,k).
By the Vitali coveringntheorem,
there is a disjoint subsequence of these balls, {B (xi , ri )}, with the property
o
bi covers Sk . Then by the way these balls were defined, with (20.32) holding for
that {B (xi , 5ri )} B
r = 5ri ,

  
X

X
bi
mn h (Sk )
mn h B
5n 
(n) rin
i=1

= 5n 

i=1

mn (B (xi , ri )) 5n mn (B (0, k)).

i=1

Since  is arbitrary, this shows mn h (Sk ) = 0. Now letting k , this shows mn h (S) = 0 which
proves the lemma.
Thus mn (N ) = 0 and mn (h (S)) = 0 and so by Lemma 20.14
mn (h (S N )) mn (h (S)) + mn (h (N )) = 0.

(20.33)

Let B \ (S N ).
A similar lemma to the following was proved in the section on the change of variables formula for a C 1
map. There the proof was based on the inverse function theorem. However, this is no longer possible; so, a
slightly more technical argument is required.
Lemma 20.20 There exists a sequence of disjoint measurable sets, {Fi }, such that

i=1 Fi = B
and h is one to one on Fi .
Proof: L (Rn , Rn ) is a finite dimensional normed linear space. In fact,
{ei ej : i, j {1, , n}}
is easily seen to be a basis. Let I be the elements of L (Rn , Rn ) which are invertible and let F be a countable
dense subset of I. Also let C be a countable dense subset of B. For c C and T F,


E (c, T, i) {b B c, i1 B such that (a.) and (b.) hold}

20.4. MAPPINGS THAT ARE NOT ONE TO ONE

373

where the conditions (a.) and (b.) are as follows.

1
|T v| |Dh (b) v| for all v
1+

(a.)

|h (a) h (b) Dh (b) (a b) | |T (a b) |

(b.)

for all a B b, 2i1 . Here 0 <  < 1/2.
Obviously, there are countably many E (c, T, i). Now suppose a, b E (c, T, i) and h (a) = h (b). Then
|a b| |a c| + |c b| <

2
.
i

Therefore, from (a.) and (b.),

1
|T (a b)|
1+

|Dh (b) (a b)|

= |h (a) h (b) Dh (b) (a b)|  |T (a b)|.

Since T is one to one, this shows that a = b. Thus h is one to one on E (c, T, i).
Now let b B. Choose T F such that

1

1 
||Dh (b) T || <  Dh (b)  .
Then for all v Rn ,

and so


1

1 
|T v Dh (b) v|  Dh (b)  |v|  |Dh (b) v|
|T v| (1 + ) |Dh (b) v|

which yields (a.). Now choose i large enough that for |a b| < 2i1 ,

|a b|
||T 1 ||
 |T (a b)|

|h (a) h (b) Dh (b) (a b)| <

and pick c CB b,i1 . Then b E (c, T, i) and this shows that B equals the union of these sets.
Let {Ei } be an enumeration of these sets and define F1 E1 , and if F1 , , Fn have been chosen,
Fn+1 En+1 \ ni=1 Fi . Then {Fi } satisfies the conditions of the lemma and this proves the lemma.
The following corollary is also of interest.
Corollary 20.21 For each Ei in Lemma 20.20, h1 is Lipschitz on h (Ei ).
Proof: Pick a, b Ei . Then by condition a. and b.,
|h (a) h (b) | |Dh (b) (a b)|  |T (a b)|


1
 |T (a b) | r |a b|
1+

374

FUNDAMENTAL THEOREM OF CALCULUS

for some r > 0 by the equivalence of all norms on a finite dimensional space. Therefore,
 1

h (h (a)) h1 (h (b)) 1 |h (a) h (b)|
r
and this proves the corollary.
Now let g : h () [0, ] be mn measurable. By Theorem 20.18,
Z
Z
Xh(Fi ) (y) g (y) dmn =
g (h (x)) J (x) dm.
h()

(20.34)

Fi

Now define
n (y) =

Xh(Fi ) (y).

i=1

By Lemma 20.14, h (Fi ) is mn measurable and so n is a mn measurable function. For each y B, n (y)
gives the number of elements in h1 (y) B. From (20.34),
Z
Z
n (y) g (y) dmn =
g (h (x)) J (x) dm.
(20.35)
h()

Now define
# (y) number of elements in h1 (y).
Theorem 20.22 The function y # (y) is mn measurable and if
g : h () [0, ]
is mn measurable, then
Z

g (y) # (y) dmn =

h()

g (h (x)) J (x) dm.

Proof: If y
/ h (S N ), then n (y) = # (y). By (20.33)
mn (h (S N )) = 0
and so n (y) = # (y) a.e. Since mn is a complete measure, # () is mn measurable. Letting
G h () \ h (S N ),
(20.35) implies
Z

g (y) # (y) dmn

h()

g (y) n (y) dmn

g (h (x)) J (x) dm

ZB

g (h (x)) J (x) dm.

This proves the theorem.

This theorem is a special case of the area formula proved in [19] and [11].

20.5. DIFFERENTIAL FORMS ON LIPSCHITZ MANIFOLDS

20.5

375

Differential forms on Lipschitz manifolds

With the change of variables formula for Lipschitz maps, we can generalize much of what was done for
C k manifolds to Lipschitz manifolds. To begin with we define what these are and then we will discuss the
integration of differential forms on Lipschitz manifolds.
In order to show the integration of differential forms is well defined, we need a version of the chain rule
valid in the context of Lipschitz maps.
Theorem 20.23 Let f ,g and h be Lipschitz mappings from Rn to Rn with g (f (x)) = h (x) on A, a
measurable set. Also suppose that f 1 is Lipschitz on f (A) . Then for a.e. x A, Dg (f (x)), Df (x),
Dh (x) ,and D (f g) (x) all exist and
Dh (x) = D (g f ) (x) = Dg (f (x)) Df (x).
The proof of this theorem is based on the following lemma.
Lemma 20.24 Let k : Rn Rn be Lipschitz, then if k (x) = 0 for all x A, then det (Dk (x)) = 0
a.e.x A.
R
R
Proof: By the change of variables formula, Theorem 20.22, 0 = {0} # (y) dy = A |det (Dk (x))| dx and
so det (Dk (x)) = 0 a.e.
Proof of the theorem: On A, g (f (x)) h (x) = 0 and so by the lemma, there exists a set of measure
zero, N1 such that if x
/ N1 , D (g f ) (x) Dh (x) = 0. Let M be the set of points in f (A) where g fails
to be differentiable and let N2 f 1 (M ) A, also a set of measure zero because by Lemma 20.13, Lipschitz
functions map sets of measure zero to sets of measure zero. Finally let N3 be the set of points where f
fails to be differentiable. Then if x A\ (N1 N2 N3 ) , the chain rule implies Dh (x) = D (g f ) (x) =
Dg (f (x)) Df (x). This proves the theorem.
Definition 20.25 We say is a Lipschitz manifold with boundary if it is a manifold with boundary for
1
which the maps, R

i and

Ri are Lipschitz. It will be called orientable if there is an atlas, (Ui , Ri ) for which
1
det D Ri Rj (u) > 0 for a.e. u Rj (Ui Uj ) . We will call such an atlas a Lipschitz atlas.
Note we had to say the determinant of the derivative of the overlap map is greater than zero a.e. because
all we know is that this map is Lipschitz and so it only has a derivative a.e.
Lemma 20.26 Suppose (V, S) and (U, R) are two charts for a Lipschitz n manifold, Rm , that det (Du (v))
exists for a.e. v S (V U ) , and det (Dv (u)) exists for a.e. u R (U V ) . Here we are writing u (v)
for R S1 (v) with a similar definition for v (u) . Then letting I = (i1 , , in ) be a sequence of indices, we
have the following formula for a.e. u R (U V ) .






xi1 xin
xi1 xin v 1 v n
=
.
(20.36)
(u1 un )
(v 1 v n ) (u1 un )


Proof: Let xI Rn denote xi1 , , xin . Then using Theorem 20.23, we can say that there is a set of
Lebesgue measure zero, N R (U V ) , such that if u
/ N, then
DxI (u) = DxI (v) Dv (u) .
Taking determinants, we obtain (20.36) for u
/ N. This proves the lemma.
Similarly we have the following formula which holds for a.e. v S (U V ) .






xi1 xin u1 un
xi1 xin
=
.
(v 1 v n )
(u1 un ) (v 1 v n )

376

20.6

FUNDAMENTAL THEOREM OF CALCULUS

Some examples of orientable Lipschitz manifolds

The following simple proposition will give many examples of Lipschitz manifolds.
Proposition 20.27 Suppose is a bounded open subset of Rn with n 2 having the property that for all
p \ , there exists an open set, U1 , containing p, and an orthogonal transformation, P : Rn Rn
whose determinant equals 1, having the following properties. The set, P (U1 ) is of the form
B (a, b)
where B is a bounded open subset of Rn1 . Letting
e ,
U U
it follows that

while



b B and u1 (a, g (b
P (U ) = u Rn : u
u))

(20.37)



b B and u1 = g (b
P ( U1 ) = u Rn : u
u)

(20.38)

where g is a Lipschitz map and

b u2 , , un
u

T

where u = u1 , u2 , un

T

Then is an oriented Lipschitz manifold.

The following picture is descriptive of the situation described in the proposition.

U1
@
b
@
H
H
@
H
P
P (U1 )
U
@
P (U )
x
@
@
a
Proof: Let U1 , U, and g be as described above. For x U define
R (x) P (x)
where
u1 g (b
u) u2

(u)

un

T

The pair, (U, R) is a chart in an atlas for . It is clear that R is Lipschitz continuous. Furthermore, it is
clear that the inverse map, R1 is given by the formula
R1 (u) = P 1 (u)
where
1 (u) =

u1 + g (b
u) u2

un

T

20.6. SOME EXAMPLES OF ORIENTABLE LIPSCHITZ MANIFOLDS

377

and that R , R1 , and 1 are all are Lipschitz continuous. Then the version of the chain rule found in
Theorem 20.23 implies DR (x) , and DR1 (x) exist and have postive determinant a.e. Thus if (U, R) and
(V, S) are two of these charts, we may apply Theorem 20.23 to find that for a.e. v,








det D R S1 (v) = det DR S1 (v) DS1 (v)





= det DR S1 (v) det DS1 (v) > 0.
The set, is compact and so there are p of these sets, U1j covering along with functions Rj as just
described. Let U0 satisfy
\ pj=1 U1j U0 U0


and let R0 (x) x1 k x2 xn where k is chosen large enough that R0 maps U0 into Rn< . Then
(Ur , Rr ) is an oriented atlas for if we define Ur U1r . As above, the chain rule shows the derivatives
of the overlap maps have positive determinants.
For example, a ball of radius r > 0 is an oriented Lipschitz
Qn n manifold with boundary because it satisfies
the conditions of the above proposition. So is a box, i=1 (ai , bi ) . This proposition gives examples of
Lipschitz n manifolds in Rn but we want to have examples of n manifolds in Rm for m > n. We recall the
following lemma from Section 17.3.


Lemma 20.28 Suppose O is a bounded open subset of Rn and let F : O Rm be a function in C k O; Rm
where m n with the property that for all x O, DF (x) has rank n. Then if y0 = F (x0 ) , there exists a
bounded open set in Rm , W, which contains

y0 , a bounded open set, U O which contains x0 and a function
G : W U such that G is in C k W ; Rn and for all x U,
G (F (x)) = x.
Furthermore, G = G1 P on W where P is a map of the form


P (y) = y i1 , , y in
for some list of indices, i1 < < in .
With this lemma we can give a theorem which will provide many other examples. We note that G, and
F are Lipschitz continuous in the above because of the requirement that they are restrictions of C k functions
defined on Rn which have compact support.
Theorem 20.29 Let be a Lipschitz
n manifold with boundary in Rn and suppose O, an open bounded
subset of Rn . Suppose F C 1 O; Rm is one to one on O and DF (x) has rank n for all x O. Then F ()
is also a Lipschitz manifold with boundary and F () = F () .
Proof: Let (Ur , Rr ) be an atlas for and suppose Ur = Or where Or is an open subset of O. Let
g
x0 Ur . By Lemma 20.28 there exists
an open set, Wx0 in Rm containing F (x0 ), an open set in Rn , U
x0
containing x0 , and Gx0 C k Wx0 ; Rn such that
Gx0 (F (x)) = x
g
g
for all x U
x0 . Let Ux0 Ur Ux0 .
Claim: F (Ux0 ) is open in F () .
Proof: Let x Ux0 . If F (x1 ) is close enough to F (x) where x1 , then F (x1 ) Wx0 and so
|x x1 | = |Gx0 (F (x)) Gx0 (F (x1 ))|
K |(F (x)) F (x1 )|

378

FUNDAMENTAL THEOREM OF CALCULUS

where K is some constant which depends only on

max {||DGx0 (y)|| : y Rm } .
Therefore, if F (x1 ) is close enough to F (x) , it follows we can conclude |x x1 | is very small. Since Ux0
is open in it follows that whenever F (x1 ) is sufficiently close to F (x) , we have x1 Ux0 . Consequently
F (x1 ) F (Ux0 ) . This shows F (Ux0 ) is open in F () and proves the claim.
With this claim it follows that (F (Ux0 ) , Rr Gx0 ) is a chart and since Rr is given to be Lipschitz
continuous, we know the map, Rr Gx0 is Lipschitz continuous. The inverse map of Rr Gx0 is also seen to
equal F R1
r , also Lipschitz continuous. Since is compact there are finitely many of these sets, F (Uxi )
covering F () . This yields an atlas for F () of the form (F (Uxi ) , Rr Gxi ) where xi Ur and proves
F () is a Lipschitz manifold.
Since the R1
r are Lipschitz, the overlap map for two of these charts is of the form,




Rs Gxj F R1
= Rs R1
r ,
r
showing that if (Ur , Rr ) is an oriented atlas for , then F () also has an oriented atlas since the overlap
maps described above are all of the form Rs R1
r .
It remains to verify the assertion about boundaries. y F () if and only if for some xi Ur ,
Rr Gxi (y) Rn0
if and only if
Gxi (y)
if and only if
Gxi (F (x)) = x
where F (x) = y if and only if y F () . This proves the theorem.

20.7

Stokes theorem on Lipschitz manifolds

In the proof of Stokes theorem, we used that R1

is C 2 but the end result is a formula which involves only
i
1
the first derivatives of Ri . This suggests that it is not necessary to make this assumption. Here we will
show that one can do with assuming the manifold in question is only a Lipschitz manifold.
Using the Theorem 20.22 and the version of the chain rule for Lipschitz maps given above, we can verify
that the integral of a differential form makes sense for an orientable Lipschitz manifold using the same
arguments given earlier in the chapter on differential forms. Now suppose is only a Lipschitz orientable n
manifold. Then in the proof of Stokes theorem, we can say that for some sequence of n ,




Z
p Z
m XX
X


xj xi1 xin 1
aI
1
d lim
r j Rr N (u)
du
(20.39)
n
x
(u1 un )

r=1 Rr Ur
j=1
I

where N is a mollifier and

xj xi1 xin 1
(u1 un )

is obtained from
x = R1
r N (u) .

20.8. SURFACE MEASURES ON LIPSCHITZ MANIFOLDS

379

The reason we can write the above limit is that from Rademachers theorem and the dominated convergence
theorem we may write




1
R1
N .
r N ,i = Rr
,i




1
As in Chapter 12 we see R1
in Lp (Rr Ur ) for every p. Taking an appropriate subser N ,i Rr
,i
quence, we can obtain, in addition to this, almost everywhere convergence for every partial derivative and
every Rr yielding (20.39) for thatP
subsequence.
We may also
arrange
to
have
r = 1 near . Then for N large enough, R1
r N (Rr Ui ) will lie in
P
this set where
r = 1. Then we do the computations as in the proof of Stokes theorem. Using the same
1
computations, with R1
r N in place of Rr ,
Z
d =

lim

m XXZ
X
j=1

rB

m XXZ
X
j=1

rB

Rr Ur Rn
0

Rr Ur Rn
0

r aI

r aI

R1
r

R1
r

xi1 xin1
N
(u2 un )

xi1 xin1
(u2 un )

0, u2 , , un du2 dun

0, u , , u

du2 dun

This yields the following significant generalization of Stokes theorem.

Theorem 20.30 (Stokes theorem) Let be a Lipschitz orientable manifold with boundary and let
P
i1
in1
be a differential form of order n 1 for which aI is C 1 . Then
I aI (x) dx dx
Z
Z
=
d.
(20.40)

The theorem can be generalized further but this version seems particularly natural so we stop with this
one. You need a change of variables formula and you need to be able to take the derivative of R1
a.e. These
i
ingredients are available for more general classes of functions than Lipschitz continuous functions. See [19]
in the exercises on the area formula. You could also relax the assumptions on aI slightly.

20.8

Surface measures on Lipschitz manifolds

Let be a Lipschitz manifold in Rm , oriented or not. Let f be a continuous function defined on , and let
(Ui , RiP
) be an atlas and let { i } be a C partition of unity subordinate to the sets, Ui as described earlier.
If = I aI (x) dxI is a differential form, we may always assume
dxI = dxi1 dxin
where i1 < i2 < < in . The reason for this is that in taking an integral used to integrate the differential
(xi1 xin )
form, a switch in two of the dxj results in switching two rows in the determinant, (u1 un ) , implying that
any two of these differ only by a multiple of 1. Therefore, there is no loss of generality in assuming from
now on that in the sum for , I is always a list of indices which are strictly increasing. The case where
some term of has a repeat, dxir = dxis can be ignored because such terms deliver zero in integrating the
differential form because they involve a determinant having two equal rows. We emphasize again that from
now on I will refer to an increasing list of indices.

380

FUNDAMENTAL THEOREM OF CALCULUS

Let

X
Ji (u)
I

!2 1/2
xi1 xin

(u1 un )

1
where here the sum

is taken over all possible increasing lists of indices, I, from {1, , m} and x = Ri u.
m
Thus there are n terms in the sum. We define a positive linear functional, on C () as follows:
p Z
X


f
f i R1
(20.41)
i (u) Ji (u) du.
Ri Ui

i=1

We will now show this is well defined.

Lemma 20.31 The functional defined in (20.41) does not depend on the choice of atlas or the partition of
unity.
Proof: In (20.41), let { i } be a partition of unity as described there which is associated with the atlas
(Ui , Ri ) and let { i } be a partition of unity associated in the same
manner with the atlas (Vi , Si ) . Using
the change of variables formula, Theorem 20.22 with u = Ri S1
v,
j
p Z
X

Ri Ui

i=1

i f R1
i (u) Ji (u) du =

p X
q Z
X

Ri Ui

i=1 j=1

p X
q Z
X

Sj (Ui Vj )

i=1 j=1

Sj (Ui Vj )

i=1 j=1

j i f R1
i (u) Ji (u) du =



 u1 un









1
1
(v) i Sj (v) f Sj (v) Ji (u) 
 dv
 (v 1 v n ) 

S1
j

p X
q Z
X

1
1
j S1
j (v) i Sj (v) f Sj (v) Jj (v) dv.

This yields
the definition of f using (Ui , Ri )
p Z
X
i=1

p X
q Z
X
i=1 j=1

Sj (Ui Vj )

q Z
X
j=1

(20.42)

Ri Ui

i f R1
i (u) Ji (u) du =

1
1
j S1
j (v) i Sj (v) f Sj (v) Jj (v) dv

Sj (Vj )

1
j S1
j (v) f Sj (v) Jj (v) dv

the definition of f using (Vi , Si ) .

This proves the lemma.
This lemma implies the following theorem.

(20.43)

20.8. SURFACE MEASURES ON LIPSCHITZ MANIFOLDS

381

Theorem 20.32 Let be a Lipschitz manifold with boundary. Then there exists a unique Radon measure,
, defined on such that whenever f is a continuous function defined on and (Ui , Ri ) denotes an atlas
and { i } a partition of unity subordinate to this atlas, we have
f =

f d =

p Z
X
i=1

Ri Ui

i f R1
i (u) Ji (u) du.

(20.44)

Furthermore, for any f L1 (, ) ,

Z

p Z
X

f d =

i=1

Ri Ui

i f R1
i (u) Ji (u) du

(20.45)

and a subset, A, of is measurable if and only if for all r, Rr (Ur A) is Jr (u) du measurable.
Proof:We begin by proving the following claim.
Claim :A set, S Ui , has measure zero in Ui , if and only if Ri S has measure zero in Ri Ui with
respect to the measure, Ji (u) du.
Proof of the claim:Let > 0 be given. By outer regularity, there exists a set, V Ui , open in
such that (V ) < and S V Ui . Then Ri V is open in Rn and contains Ri S. Letting h O, where
O Rn = Ri V and mn (O) < mn (Ri V ) + , and letting h1 (x) h (Ri (x)) for x Ui , we see h1 V. By
Corollary 12.24,
that
spt (h1 ) {x Rm : i (x) = 1} . Thus

we can also choose our partition of unity so
1
1
j h1 Rj (u) = 0 unless j = i when this reduces to h1 Ri (u) . Thus

(V )

h1 d =

h1 d =

h1 R1
i

p Z
X
j=1

(u) Ji (u) du =

ZRi Ui

h (u) Ji (u) du Ki ,

Rj Uj

j h1 R1
j (u) Jj (u) du

h (u) Ji (u) du =

Ri Ui

h (u) Ji (u) du

Ri V

where Ki ||Ji || . Now this holds for all h O and so

Z
Z
Z
Ji (u) du
Ji (u) du
Ji (u) du (1 + Ki ) .
Ri S

Ri V

Since is arbitrary, this shows Ri S has mesure zero with respect to the measure, Ji (u) du as claimed.
Now we prove the converse. Suppose Ri S has Jr (u) du measure zero. Then there exists an open set, O
such that O Ri S and
Z
Ji (u) du < .
O
1
Thus R1
i (O Ri Ui ) is open in and contains S. Let h Ri (O Ri Ui ) be such that
Z


hd + > R1
i (O Ri Ui ) (S)

As in the first part, we can choose our partition of unity such that h (x) = 0 off the set,
{x Rm : i (x) = 1}

382

FUNDAMENTAL THEOREM OF CALCULUS

and so as in this part of the argument,

Z

hd

p Z
X

ZRi Ui

j=1

Rj Uj

j h R1
j (u) Jj (u) du

h R1
i (u) Ji (u) du


h R1
i (u) Ji (u) du

ZORi Ui

Ji (u) du <
O

and so (S) 2. Since is arbitrary, this proves the claim.

Now let A Ur be measurable. By the regularity of the measure, there exists an F set, F and a G
set, G such that Ur G A F and (G \ F ) = 0.(Recall a G set is a countable intersection of open sets
and an F set is a countable union of closed sets.) Then since is compact, it follows each of the closed sets
whose union equals F is a compact set. Thus if F =
k=1 Fk we know Rr (Fk ) is also a compact set and so
Rr (F ) =
R
(F
)
is
a
Borel
set.
Similarly,
R
(G)
is also a Borel set. Now by the claim,
r
k
r
k=1
Z
Jr (u) du = 0.
Rr (G\F )

We also see that since Rr is one to one,

Rr G \ Rr F = Rr (G \ F )
and so
Rr (F ) Rr (A) Rr (G)
where Rr (G) \ Rr (F ) has measure zero. By completeness of the measure, Ji (u) du, we see Rr (A) is
measurable. It follows that if A is measurable, then Rr (Ur A) is Jr (u) du measurable for all r. The
converse is entirely similar.
Letting f L1 (, ) , we use the fact that is a Radon mesure to obtain a sequence
 of continuous


functions, {fk } which converge to f in L1 (, ) and for a.e. x. Therefore, the sequence fk R1
is
i ()




a Cauchy sequence in L1 Ri Ui ; i R1
(u)
J
(u)
du
.
It
follows
there
exists
i
i




g L1 Ri Ui ; i R1
i (u) Ji (u) du






such that fk R1
() g in L1 Ri Ui ; i R1
(u) Ji (u) du . By the pointwise convergence, g (u) =
i
i


1
1
f R1
for a.e. u Ri Ui and so
i (u) for a.e. Ri (u) Ui . By the above claim, g = f Ri
f R1
L1 (Ri Ui ; Ji (u) du)
i
and we can write
Z

f d

=
=

lim

fk d = lim

p Z
X

i=1 Ri Ui
p Z
X
i=1

Ri Ui

p Z
X
i=1

Ri Ui

i fk R1
i (u) Ji (u) du

i R1
i (u) g (u) Ji (u) du


i f R1
i (u) Ji (u) du.

20.9. THE DIVERGENCE THEOREM FOR LIPSCHITZ MANIFOLDS

383

This proves the theorem.

In the case of a Lipschitz manifold, note that by Rademachers theorem the set in Rr Ur on which R1
r
has no derivative has Lebesgue measure zero and so contributes nothing to the definition of and can be
ignored. We will do so from now on. Other sets of measure zero in the sets Rr Ur can also be ignored and
we do so whenever convenient.
Corollary 20.33 Let f L1 (; ) and suppose f (x) = 0 for all x
/ Ur where (Ur , Rr ) is a chart in a
Lipschitz atlas for . Then
Z
Z
Z


f d =
f d =
f R1
(20.46)
r (u) Jr (u) du.

Ur

Rr Ur

Proof: Using regularity of the measures, we can pick a compact subset, K, of Ur such that
Z

Z



 < .
f
d

f
d


Ur

Now by Corollary 12.24, we can choose the partition of unity such that K {x Rm : r (x) = 1} . Then
Z

f d

p Z
X

Zi=1

Ri Ui

Rr Ur

i f XK R1
i (u) Ji (u) du

f XK R1
r (u) Jr (u) du.

Therefore, letting Kl Rr Ur we can take a limit and conclude


Z
Z




1

f d
f Rr (u) Jr (u) du .

Ur

Rr Ur

Since is arbitrary, this proves the corollary.

20.9

The divergence theorem for Lipschitz manifolds

What about writing the integral of a differential form in terms of this measure? Let be a differential form,
X
(x) =
aI (x) dxI
I

where aI is continuous or Borel measurable if you desire, and the sum is taken over all increasing lists from
{1, , m} . We assume is a Lipschitz or C k manifold which is orientable and that (Ur , Rr ) is an oriented
atlas for while, { r } is a C partition of unity subordinate to the Ur .
Lemma 20.34 Consider the set,


S x : for some r, x = R1
.
r (u) where x Ur and Jr (u) = 0
Then (S) = 0.
p
Proof: Let Sr denote those points, x, of Ur for which x = R1
r (u) and Jr (u) = 0. Thus S = r=1 Sr .
From Corollary 20.33
Z
Z


XSr d =
XSr R1
r (u) Jr (u) du = 0

Rr Ur

384

FUNDAMENTAL THEOREM OF CALCULUS

and so
(S)

p
X

(Sk ) = 0.

k=1

This proves the lemma.

With respect to the above atlas, we define a function of x in the following way. For I = (i1 , , in ) an
increasing list of indices,


(xi1 xin )
/Jr (u) , if x Ur \ S
1
n
(u u )
oI (x)

0 if x S

Then (20.36) implies this is well defined aside from a possible set of measure zero, N. Now it follows from
Corollary 20.33 that R1 (N ) is a set of measure zero on and that oI is given by the top line off a set of
measure zero. It also shows that if we had used a different atlas having the same orientation, then o (x)
defined using the new atlas would change only on a set of measure zero. Also note
X
2
oI (x) = 1 a.e.
I

Thus we may consider o (x) to be well defined if we regard two functions as equal provided they differ only
m
on a set of measure zero. Now we define a vector valued function, o having values in R( n ) by letting the I th
component of o (x) equal oI (x) . Also define
X
(x) o (x)
aI (x) oI (x) ,
I

From the definition of what we mean by the integral of a differential form, Definition 17.11, it follows that


Z
p XZ
X



xi1 xin
1
1
du

r Rr (u) aI Rr (u)
(u1 un )

Rr Ur
r=1 I
p Z
X






1
1
=
r R1
r (u) Rr (u) o Rr (u) Jr (u) du
Rr Ur

r=1
Z

od

(20.47)

Note that o is bounded and measurable so is in L1 . For convenience, we state the following lemma whose
proof is essentially the same as the proof of Lemma 17.25.
Lemma 20.35 Let be a Lipschitz oriented manifold in Rn with an oriented atlas, (Ur , Rr ). Letting
x = R1
r u and letting 2 j n, we have


1
bi xn
n

x
i
X
x
i+1
(1)
= 0 a.e.
(20.48)
j
u
(u2 un )
i=1
for each r. Here, xbi means this is deleted. If for each r,


x1 xn
0 a.e.,
(u1 un )
then for each r


1
bi xn
n

x
i
X
x
i+1
(1)
0 a.e.
1
u
(u2 un )
i=1

(20.49)

20.9. THE DIVERGENCE THEOREM FOR LIPSCHITZ MANIFOLDS

385

Proof: (20.49) follows from the observation that





1
bi xn
n

x
i
X
x1 xn
x
i+1
=
(1)
1
(u1 un )
u
(u2 un )
i=1
by expanding the determinant,


x1 xn
,
(u1 un )
along the first column. Formula (20.48) follows from the observation that the sum in (20.48) is just the
determinant of a matrix which has two equal columns. This proves the lemma.
With this lemma, it is easy to verify a general form of the divergence theorem from Stokes theorem.
First we recall the definition of the divergence of a vector field.
Pn
Definition 20.36 Let O be an open subset of Rn and let F (x) k=1 F k (x) ek be a vector field for which
F k C 1 (O) . Then
div (F)

n
X
Fk

k=1

xk

Theorem 20.37 Let be an orientable Lipschitz n manifold with boundary in Rn having an oriented atlas,
(Ur , Rr ) which satisfies


x1 xn
0
(20.50)
(u1 un )
for all r. Then letting n (x) be the vector field whose ith component taken with respect to the usual basis of
Rn is given by


(
1
n
bi
i+1 x x x
i
(1)
n (x)
(20.51)
(u2 un ) /Jr (u) if Jr (u) 6= 0
0 if Jr (u) = 0
for x Ur , it follows n (x) is independent of the choice of atlas provided
the orientation remains


unchanged. Also n (x) is a unit vector for a.e. x . Let F C 1 ; Rn . Then we have the following
formula which is called the divergence theorem.
Z
Z
div (F) dx =
F nd,
(20.52)

where is the surface measure on defined above.

Proof: Recall that on

 2 1/2

1
b
i xn
n

x
X

Jr (u) =

2 un )

(u
i=1
From Lemma 20.26 and the definition of two atlass having the same orientation, we see that aside from sets
of measure zero, the assertion about the independence of choice of atlas for the normal, n (x) is verified.
Also, by Lemma 20.34, we know Jr (u) 6= 0 off some set of measure zero for each atlas and so n (x) is a unit
vector for a.e. x.

386

FUNDAMENTAL THEOREM OF CALCULUS

Now we define the differential form,

n
X

i+1

(1)

i=1

ci dxn .
Fi (x) dx1 dx

Then from the definition of d,

d = div (F) dx1 dxn .
Now let { r } be a partition of unity subordinate to the Ur . Then using (20.50) and the change of variables
formula,


Z
p Z
X

x1 xn
1
( r div (F)) Rr (u)
d
du
(u1 un )

r=1 Rr Ur
Z
p Z
X
=
( r div (F)) (x) dx =
div (F) dx.
Ur

r=1

Now
Z

p Z
X

i=1



x1 xbi xn


r Fi R1
du2 dun
r (u)
2 un )
n

(u
R
U
R
r r
0
r=1
!
n
X


2
n
r
Fi ni R1
r (u) Jr (u) du du

p Z
X
i+1
(1)

n
X

Rr Ur Rn
0

r=1

i=1

F nd.

By Stokes theorem,
Z

div (F) dx =

d =

F nd

and this proves the theorem.

Definition 20.38 In the context of the divergence theorem, the vector, n is called the unit outer normal.

20.10

Exercises

1. Let E be a Lebesgue measurable set. We say x E is a point of density if

m(E B(x, r))
= 1.
r0
m(B(x, r))
lim

Show that a.e. point of E is a point of density.

2. Let (, S, ) be any finite measure space, f 0, f real-valued, and measurable. Let be an
increasing C 1 function with (0) = 0. Show
Z
Z
f d =
0 (t)([f (x) > t])dt.

Hint:
Z

(f (x))d =

Z Z

f (x)

0 (t)dtd =

Z Z

X[0,f (x)) (t)0 (t)dtd.

Argue 0 (t)X[0,f (x)) (t) is product measurable and use Fubinis theorem. The function t ([f (x) > t])
is called the distribution function.

20.10.

EXERCISES

387

3. Let f be in L1loc (Rn ). Show M f is Borel measurable.

4. If f Lp , 1 < p < , show M f Lp and
||M f ||p A(p, n)||f ||p .
Hint: Let
f1 (x)

f (x) if |f (x)| > /2,

0 if |f (x)| /2.

Argue [M f (x) > ] [M f1 (x) > /2]. Then by Problem 2,

Z
Z
(M f )p dx =
pp1 m([M f > ])d
0

pp1 m([M f1 > /2])d.

Now use Theorem 20.4 and Fubinis Theorem as needed.

5. Show |f (x)| M f (x) at every Lebesgue point of f whenever f L1 (Rn ).
6. In the proof of the Vitali covering theorem there is nothing sacred about the constant 12 . Go through
the proof replacing this constant with where (0, 1) . Show that it follows that for every > 0,
b
the conclusion of the Vitali covering theorem follows with 5 replaced by (3 + ) in the definition of B.
7. Suppose A is covered by a finite collection of Balls, F. Show that then there exists a disjoint collection
p
bi where 5 can be replaced with 3 in the definition of B.
b
of these balls, {Bi }i=1 , such that A pi=1 B
Hint: Since the collection of balls is finite, we can arrange them in order of decreasing radius.
8. The result of this Problem is sometimes called the Vitali Covering Theorem. It is very important in
some applications. It has to do with covering sets in except for a set of measure zero with disjoint
balls. Let E Rn be Lebesgue measurable, m(E) < , and let F be a collection of balls that cover
E in the sense of Vitali. This means that if x E and > 0, then there exists B F, diameter
of B < and x B. Show there exists a countable sequence of disjoint balls of F, {Bj }, such that
m(E \
j=1 Bj ) = 0. Hint: Let E U , U open and
m(E) > (1 10n )m(U ).
Let {Bj } be disjoint,

E
j=1 Bj , Bj U.
Thus
m(E) 5n m(
j=1 Bj ).
Then
m(E) > (1 10n )m(U )

(1 10n )[m(E \
j=1 Bj ) + m(j=1 Bj )]

388

FUNDAMENTAL THEOREM OF CALCULUS

n
(1 10n )[m(E \
m(E)].
j=1 Bj ) + 5

Hence
n 1
m(E \
) (1 (1 10n )5n )m(E).
j=1 Bj ) (1 10

Let (1 10n )1 (1 (1 10n )5n ) < < 1 and pick N1 large enough that
1
m(E) m(E \ N
j=1 Bj ).

N1
1
Let F1 = {B F : Bj B = , j = 1, , N1 }. If E \ N
j=1 Bj 6= , then F1 6= and covers E \ j=1 Bj
1
in the sense of Vitali. Repeat the same argument, letting E \ N
j=1 Bj play the role of E.

9. Suppose E is a Lebesgue measurable set which has positive measure and let B be an arbitrary open
ball and let D be a set dense in Rn . Establish the result of Smital, [4]which says that under these
conditions, mn ((E + D) B) = mn (B) where here mn denotes the outer measure determined by mn .
Is this also true for X, an arbitrary possibly non measurable set replacing E in which mn (X) > 0?
Hint: Let x be a point of density of E and let D0 denote those elements of D, d, such that d + x B.
Thus D0 is dense in B. Now use translation invariance of Lebesgue measure to verify that there exists,
R > 0 such that if r < R, we have the following holding for d D0 and rd < R.
mn ((E + D) B (x + d, rd ))
mn ((E + d) B (x + d, rd )) (1 ) mn (B (x + d, rd )) .
Argue the balls, mn (B (x + d, rd )) , form a Vitali cover of B.
10. Suppose is a Radon measure on Rn , and (S) < where mn (S) = 0 and (E) = (E S). (If
(E) = (E S) where mn (S) = 0 we say mn .) Show that for mn a.e. x the following holds. If
i)
Bi {x}, then limi m(B
= 0. Hint: You might try this. Set , r > 0, and let
n (Bi )
E = {x S C : there exists {Bix }, Bix {x} with

(Bix )
}.
mn (Bix )

Let K be compact, (S \ K) < r. Let F consist of those balls just described that do not intersect K
and which have radius < 1. This is a Vitali cover of E . Let B1 , , Bk be disjoint balls from F and
mn (E \ ki=1 Bi ) < r.
Then
mn (E ) < r +

k
X

mn (Bi ) < r + 1

i=1

r + 1

k
X

k
X

(Bi ) =

i=1

(Bi S) r + 1 (S \ K) < 2r.

i=1

Since r is arbitrary, mn (E ) = 0. Consider E =

/ S E.
k=1 Ek1 and let x
11. Is it necessary to assume (S) < in Problem 10? Explain.

20.10.

EXERCISES

389

12. Let S be an increasing function on R which is right continuous,

lim S(x) = 0,

and S is bounded. Let be the measure representing

Show S 0 (x) = 0 m a.e. Hint:

f dS. Thus ((, x]) = S(x). Suppose m.

0 h1 (S(x + h) S(x))

((x, x + h])
((x h, x + 2h))
3
.
m((x, x + h])
m((x h, x + 2h))

Now apply Problem 10. Similarly h1 (S(x) S(x h)) 0.

13. Let f be increasing, bounded above and below, and right continuous. Show f 0 (x) exists a.e. Hint:
See Problem 6 of Chapter 18.
14. Suppose |f (x) f (y)| K|x y|. Show there exists g L (R), ||g|| K, and
Z y
f (y) f (x) =
g(t)dt.
x

R
Hint: Let F (x) = Kx + f (x) and let be the measure representing f dF . Show << m. What
does this imply about the differentiability of a Lipschitz continuous function?
15. Let f be increasing. Show f 0 (x) exists a.e.
R1
16. Let f (x) = x2 . Thus 1 f (x)dx = 2/3. Lets change variables. u = x2 , du = 2xdx = 2u1/2 dx. Thus
2/3 =

x2 dx =

u/2u1/2 du = 0.

Can this be done correctly using a change of variables theorem?

17. Consider the construction employed to obtain the Cantor set, but instead of removing the middle third
interval, remove only enough that the sum of the lengths of all the open intervals which are removed
is less than one. That which remains is called a fat Cantor set. Show it is a compact set which has
measure greater than zero which contains no interval and has the property that every point is a limit
point of the set. Let P be such a fat Cantor set and consider
Z x
f (x) =
XP C (t) dt.
0

Show that f is a strictly increasing function which has the property that its derivative equals zero on
a set of positive measure.
18. Let f be a function defined on an interval, (a, b) . The Dini derivates are defined as
D+ f (x) lim inf

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

D f (x) lim inf

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

h0+

h0+

390

FUNDAMENTAL THEOREM OF CALCULUS

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) .

19. 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) + x, consider the situation where D+ fe(x) >0 for all
but countably many x. If in this situation, fe(c) > fe(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.
20. Consider in the formula for ( + 1) the following change of variables. t = + 1/2 s. Then in terms
of the new variable, s, the formula for ( + 1) is


Z
Z

i
h 

s
ln 1+ s s
+ 12
s
+ 12

e
e
1+
ds = e
ds
e

Show the integrand converges to e

s2
2

. Show that then

Z

s2
( + 1)
lim +(1/2) =
e 2 ds = 2.
e

You will need to obtain a dominating function for the integral so that you can use the dominated
convergence theorem. This formula is known as Stirlings formula.
21. Let be an oriented Lipschitz n manifold in Rn for which


x1 xn
0 a.e.
(u1 un )
for all the charts, x = Rr (u) , and suppose F : Rn Rm for m n is a Lipschitz continuous function
such that there exists a Lipschitz continuous function, G : Rm Rn such that G P
F (x) = x for all
x . Show that F () is an oriented Lipschitz n manifold. Now suppose I aI (y) dyI is a
differential form on F () . Show


Z
Z X
y i1 y in
=
aI (F (x))
dx.
(x1 xn )
F()

In this case, we say that F () is a parametrically defined manifold. Note this shows how to compute
the integral of a differential form on such a manifold without dragging in a partition of unity. Also
note that could be a box or some collection of boxes pased togenter along edges. Can you get a
similar result in the case where F satisfies the conditions of Theorem 20.29?
22. Let h :Rn Rn and h is Lipschitz. Let
A = {x : h (x) = c}
where c is a constant vector. Show J (x) = 0 a.e. on A. Hint: Use Theorem 20.22.

20.10.

EXERCISES

391

23. Let U be an open subset of Rn and let h :U Rn be differentiable on A U for some A a Lebesgue
measurable set. Show that if T A and mn (T ) = 0, then mn (h (T )) = 0. Hint: Let
Tk {x T : ||Dh (x)|| < k}
and let  > 0 be given. Now let V be an open set containing Tk which is contained in U such that
mn (V ) < kn5n and let > 0 be given. Using differentiability of h, for each x Tk there exists rx <
such that B (x,5rx ) V and
h (B (x,rx )) B (h (x) , 5krx ).
Use the same argument found in Lemma 20.13 to conclude
mn (h (Tk )) = 0.
Now
mn (h (T )) = lim mn (h (Tk )) = 0.
k

24. In the context of 23 show that if S is a Lebesgue measurable subset of A, then h (S) is mn measurable.
Hint: Use the same argument found in Lemma 20.14.
25. Suppose also that h is differentiable on U . Show the following holds. Let x A be a point where
1
Dh (x) exists. Then if  (0, 1) the following hold for all r small enough.

 
(20.53)
mn h B (x,r) = mn (h (B (x,r))) mn (Dh (x) B (0, r (1 ))),
h (B (x, r)) h (x) + Dh (x) B (0, r (1 + )),

(20.54)

mn (h (B (x,r))) mn (Dh (x) B (0, r (1 + ))) .

(20.55)

If U \ A has measure 0, then for x A,

lim

r0

mn (h (B (x, r) A))
= 1.
mn (h (B (x, r)))

(20.56)

mn (h (B (x, r)))
,
mn (B (x,r))

(20.57)

Also show that for x A,

J (x) = lim

r0

where J (x) det (Dh (x)).

26. Assuming the context of the above problems, let h be one to one on A and establish that for F Borel
measurable in Rn
Z
Z
XF (y) dmn =
XF (h (x)) J (x) dm.
h(A)

This is like (20.26). Next show, using the arguments of (20.27) - (20.31), that a change of variables
formula of the form
Z
Z
g (y) dmn =
g (h (x)) J (x) dm
h(A)

holds whenever g : h (A) [0, ] is mn measurable.

27. Extend the theorem about integration and the Brouwer degree to more general classes of mappings
than C 1 mappings.

392

FUNDAMENTAL THEOREM OF CALCULUS

The complex numbers

In this chapter we consider the complex numbers, C and a few basic topics such as the roots of a complex
number. Just as a real number should be considered as a point on the line, a complex number is considered
a point in the plane. We can identify a point in the plane in the usual way using the Cartesian coordinates
of the point. Thus (a, b) identifies a point whose x coordinate is a and whose y coordinate is b. In dealing
with complex numbers, we write such a point as a + ib and multiplication and addition are defined in the
most obvious way subject to the convention that i2 = 1. Thus,
(a + ib) + (c + id) = (a + c) + i (b + d)
and
(a + ib) (c + id) = (ac bd) + i (bc + ad) .
We can also verify that every non zero complex number, a + ib, with a2 + b2 6= 0, has a unique multiplicative
inverse.
1
a ib
a
b
= 2
= 2
i 2
.
a + ib
a + b2
a + b2
a + b2
Theorem 21.1 The complex numbers with multiplication and addition defined as above form a field.
The field of complex numbers is denoted as C. An important construction regarding complex numbers
is the complex conjugate denoted by a horizontal line above the number. It is defined as follows.
a + ib = a ib.
What it does is reflect a given complex number across the x axis. Algebraically, the following formula is
easy to obtain.


a + ib (a + ib) = a2 + b2 .
The length of a complex number, refered to as the modulus of z and denoted by |z| is given by
|z| x2 + y 2

1/2

1/2

= (zz)

and we make C into a metric space by defining the distance between two complex numbers, z and w as
d (z, w) |z w| .
We see therefore, that this metric on C is the same as the usual metric of R2 . A sequence, zn z if and
n
only if xn x in R and yn y in R where z = x + iy and zn = xn + iyn . For example if zn = n+1
+ i n1 ,
then zn 1 + 0i = 1.
393

394

THE COMPLEX NUMBERS

Definition 21.2 A sequence of complex numbers, {zn } is a Cauchy sequence if for every > 0 there exists
N such that n, m > N implies |zn zm | < .
This is the usual definition of Cauchy sequence. There are no new ideas here.
Proposition 21.3 The complex numbers with the norm just mentioned forms a complete normed linear
space.
Proof: Let {zn } be a Cauchy sequence of complex numbers with zn = xn + iyn . Then {xn } and {yn }
are Cauchy sequences of real numbers and so they converge to real numbers, x and y respectively. Thus
zn = xn + iyn x + iy. By Theorem 21.1 C is a linear space with the field of scalars equal to C. It only
remains to verify that | | satisfies the axioms of a norm which are:
|z + w| |z| + |w|
|z| 0 for all z
|z| = 0 if and only if z = 0
|z| = || |z| .
We leave this as an exercise.
Definition 21.4 An infinite sum of complex numbers is defined as the limit of the sequence of partial sums.
Thus,

k=1

ak lim

n
X

ak .

k=1

Just as in the case of sums of real numbers, we see that an infinite sum converges if and only if the
sequence of partial sums is a Cauchy sequence.
Definition 21.5 We say a sequence of functions of a complex variable, {fn } converges uniformly to a
function, g for z S if for every > 0 there exists N such that if n > N , then
|fn (z) g (z)| <
for all z S. The infinite sum
S.

k=1

fn converges uniformly on S if the partial sums converge uniformly on

Proposition 21.6 A sequence of functions, {fn } defined on a set S, converges uniformly to some function,
g if and only if for all > 0 there exists N such that whenever m, n > N ,
||fn fm || < .
Here ||f || sup {|f (z)| : z S} .
Just as in the case of functions of a real variable, we have the Weierstrass M test.
Proposition 21.7 Let {fn } be a sequence
defined on S C. Suppose there
P of complex valued functions
P
exists Mn such that ||fn || < Mn and
Mn converges. Then
fn converges uniformly on S.

395
Since every complex number can be
a point in R2 , we define the polar form of a complex
 considered

y
x
number as follows. If z = x + iy then |z| , |z| is a point on the unit circle because


x
|z|

2

y
|z|

2

= 1.

Therefore, there is an angle such that



x y
,
|z| |z|

= (cos , sin ) .

It follows that
z = x + iy = |z| (cos + i sin ) .
This is the polar form of the complex number, z = x + iy.
One of the most important features of the complex numbers is that you can always obtain n nth roots
of any complex number. To begin with we need a fundamental result known as De Moivres theorem.
Theorem 21.8 Let r > 0 be given. Then if n is a positive integer,
n

[r (cos t + i sin t)] = rn (cos nt + i sin nt) .

Proof: It is clear the formula holds if n = 1. Suppose it is true for n.
n+1

[r (cos t + i sin t)]

= [r (cos t + i sin t)] [r (cos t + i sin t)]

which by induction equals

= rn+1 (cos nt + i sin nt) (cos t + i sin t)
= rn+1 ((cos nt cos t sin nt sin t) + i (sin nt cos t + cos nt sin t))
= rn+1 (cos (n + 1) t + i sin (n + 1) t)
by standard trig. identities.
Corollary 21.9 Let z be a non zero complex number. Then there are always exactly k kth roots of z in C.
Proof: Let z = x + iy. Then
z = |z|

y
x
+i
|z|
|z|

and from the definition of |z| ,


Thus

y
x
|z| , |z|

x
|z|

2

y
|z|

2

= 1.

is a point on the unit circle and so

y
x
= sin t,
= cos t
|z|
|z|

for a unique t [0, 2). By De Moivres theorem, a number is a kth root of z if and only if it is of the form





t + 2l
t + 2l
1/k
|z|
cos
+ i sin
k
k
for l an integer. By the fact that the cos and sin are 2 periodic, if l = k in the above formula the same
complex number is obtained as if l = 0. Thus there are exactly k of these numbers.
If S C and f : S C, we say f is continuous if whenever zn z S, it follows that f (zn ) f (z) .
Thus f is continuous if it takes converging sequences to converging sequences.

396

THE COMPLEX NUMBERS

21.1

Exercises

1. Let z = 3 + 4i. Find the polar form of z and obtain all cube roots of z.
2. Prove Propositions 21.6 and 21.7.
3. Verify the complex numbers form a field.
Qn
Qn
4. Prove that k=1 zk = k=1 z k . In words, show the conjugate of a product is equal to the product of
the conjugates.
Pn
Pn
5. Prove that k=1 zk = k=1 z k . In words, show the conjugate of a sum equals the sum of the conjugates.
6. Let P (z) be a polynomial having real coefficients. Show the zeros of P (z) occur in conjugate pairs.
7. If A is a real n n matrix and Ax = x, show that Ax = x.
8. Tell what is wrong with the following proof that 1 = 1.
q

2
2
1 = i = 1 1 = (1) = 1 = 1.
9. If z = |z| (cos + i sin ) and w = |w| (cos + i sin ) , show
zw = |z| |w| (cos ( + ) + i sin ( + )) .
10. Since each complex number, z = x + iy can be considered a vector in R2 , we can also consider it a
vector in R3 and consider the cross product of two complex numbers. Recall from calculus that for
x (a, b, c) and y (d, e, f ) , two vectors in R3 ,

i j k
x y det a b c
d e f
and that geometrically |x y| = |x| |y| sin , the area of the parallelogram spanned by the two vectors,
x, y and the triple, x, y, x y forms a right handed system. Show
z1 z2 = Im (z 1 z2 ) k.
Thus the area of the parallelogram spanned by z1 and z2 equals |Im (z 1 z2 )| .
11. Prove that f : S C C is continuous at z S if and only if for all > 0 there exists a > 0 such
that whenever w S and |w z| < , it follows that |f (w) f (z)| < .
12. Verify that every polynomial p (z) is continuous on C.
13. Show that if {fn } is a sequence of functions converging uniformly to a function, f on S C and if fn
is continuous on S, then so is f.
P
1
14. Show that if |z| < 1, then k=0 z k = 1z
.
P
15. Show that whenever
an converges
P it follows that limn an = 0. Give an example in which
limn an = 0, an an+1 and yet
an fails to converge to a number.
1/n

16. Prove the root test for series of complex numbers. If ak C and r lim supn |an |

converges absolutely if r < 1

X
diverges if r > 1
ak

test fails if r = 1.
k=0

then

21.2. THE EXTENDED COMPLEX PLANE

17. Does limn n

397

2+i n
3

exist? Tell why and find the limit if it does exist.

Pn
18. Let A0 = 0 and let An k=1 ak if n > 0. Prove the partial summation formula,
q
X

ak bk = Aq bq Ap1 bp +

k=p

q1
X

Ak (bk bk+1 ) .

k=p

Now using this formula, suppose {bn } is a sequence of real numbers

P which converges to 0 and is
decreasing. Determine those values of such that || = 1 and k=1 bk k converges. Hint: From
Problem 15 you have an example of a sequence {bn } which shows that = 1 is not one of those values
of .
19. Let f : U C C be given by f (x + iy) = u (x, y) + iv (x, y) . Show f is continuous on U if and only
if u : U R and v : U R are both continuous.

21.2

The extended complex plane

The set of complex numbers has already been considered along with the topology of C which is nothing but
the topology of R2 . Thus, for zn = xn + iyn we say zn z x + iy if and only if xn x and yn y. The
norm in C is given by

1/2
1/2
|x + iy| ((x + iy) (x iy))
= x2 + y 2
which is just the usual norm in R2 identifying (x, y) with x + iy. Therefore, C is a complete metric space
and we have the Heine Borel theorem that compact sets are those which are closed and bounded. Thus, as
far as topology is concerned, there is nothing new about C.
We need to consider another general topological space which is related to C. It is called the extended
b and consisting of the complex plane, C along with another point not in C known
complex plane, denoted by C
as . For example, could be any point in R3 . We say a sequence of complex numbers, zn , converges to
if, whenever K is a compact set in C, there exists a number, N such that for all n > N, zn
/ K. Since
compact sets in C are closed and bounded, this is equivalent to saying that for all R > 0, there exists N
such that if n > N, then zn
/ B (0, R) which is the same as saying limn |zn | = where this last symbol
has the same meaning as it does in calculus.
A geometric way of understanding this in terms of more familiar objects involves a concept known as the
Riemann sphere.
2
Consider the unit sphere, S 2 given by (z 1) + y 2 + x2 = 1. We define a map from the unit sphere with
the point, (0, 0, 2) left out which is one to one onto R2 as follows.

@
@
@ p
@
@

@
@(p)

We extend a line from the north pole of the sphere, the point (0, 0, 2) , through the point on the sphere,
p, until it intersects a unique point on R2 . This mapping, known as stereographic projection, which we will
denote for now by , is clearly continuous because it takes converging sequences, to converging sequences.
b we see a
Furthermore, it is clear that 1 is also continuous. In terms of the extended complex plane, C,
1
sequence, zn converges to if and only if zn converges to (0, 0, 2) and a sequence, zn converges to z C
if and only if 1 (zn ) 1 (z) .

398

21.3

THE COMPLEX NUMBERS

Exercises

1. Try to find an explicit formula for and 1 .

2. What does the mapping 1 do to lines and circles?
3. Show that S 2 is compact but C is not. Thus C 6= S 2 . Show that a set, K is compact (connected) in C
if and only if 1 (K) is compact (connected) in S 2 \ {(0, 0, 2)} .
4. Let K be a compact set in C. Show that C \ K has exactly one unbounded component and that this
component is the one which is a subset of the component of S 2 \ K which contains . If you need to
rewrite using the mapping, to make sense of this, it is fine to do so.
b into a topological space as follows. We define a basis for a topology on C
b to be all open sets
5. Make C
and all complements of compact sets, the latter type being those which are said to contain the point
b into a compact Hausdorff space. Also verify that
. Show this is a basis for a topology which makes C
b
C with this topology is homeomorphic to the sphere, S 2 .

Riemann Stieltjes integrals

In the theory of functions of a complex variable, the most important results are those involving contour
integration. Before we define what we mean by contour integration, it is necessary to define the notion of
a Riemann Steiltjes integral, a generalization of the usual Riemann integral and the notion of a function of
bounded variation.
Definition 22.1 Let : [a, b] C be a function. We say is of bounded variation if
( n
)
X
sup
| (ti ) (ti1 )| : a = t0 < < tn = b V (, [a, b]) <
i=1

where the sums are taken over all possible lists, {a = t0 < < tn = b} .
The idea is that it makes sense to talk of the length of the curve ([a, b]) , defined as V (, [a, b]) . For this
reason, in the case that is continuous, such an image of a bounded variation function is called a rectifiable
curve.
Definition 22.2 Let : [a, b] C be of bounded variation and let f : [a, b] C. Letting P {t0 , , tn }
where a = t0 < t1 < < tn = b, we define
||P|| max {|tj tj1 | : j = 1, , n}
and the Riemann Steiltjes sum by
S (P)

n
X

f ( j ) ( (tj ) (tj1 ))

j=1

where j [tj1 , tj ] . (Note this notation is a little sloppy because

R it does not identify the specific point, j
used. It is understood that this point is arbitrary.) We define f (t) d (t) as the unique number which
satisfies the following condition. For all > 0 there exists a > 0 such that if ||P|| , then

Z


 f (t) d (t) S (P) < .



Sometimes this is written as

f (t) d (t) lim S (P) .

||P||0

The function, ([a, b]) is a set of points in C and as t moves from a to b, (t) moves from (a) to (b) .
Thus ([a, b]) has a first point and a last point. If : [c, d] [a, b] is a continuous nondecreasing function,
then : [c, d] C is also of bounded variation and yields the same set of points in C with the same first
and last points. In the case where the values of the function, f, which are of interest are those on ([a, b]) ,
we have the following important theorem on change of parameters.
399

400

RIEMANN STIELTJES INTEGRALS

Theorem 22.3 Let and be as just described. Then assuming that

Z
f ( (t)) d (t)

exists, so does
Z

f ( ( (s))) d ( ) (s)

and
Z

f ( (t)) d (t) =

f ( ( (s))) d ( ) (s) .

(22.1)

Proof: There exists > 0 such that if P is a partition of [a, b] such that ||P|| < , then
Z



 f ( (t)) d (t) S (P) < .



By continuity of , there exists > 0 such that if Q is a partition of [c, d] with ||Q|| < , Q = {s0 , , sn } ,
then | (sj ) (sj1 )| < . Thus letting P denote the points in [a, b] given by (sj ) for sj Q, it follows
that ||P|| < and so


Z

n
X


 f ( (t)) d (t)
f ( ( ( j ))) ( ( (sj )) ( (sj1 ))) <



j=1
where j [sj1 , sj ] . Therefore, from the definition we see that (22.1) holds and that
Z
f ( ( (s))) d ( ) (s)

exists.
R
This theorem shows that f ( (t)) d (t) is independent of the particular used in its computation to
the extent that if is any nondecreasing function from another interval, [c, d] , mapping to [a, b] , then the
same value is obtained by replacing with .
The fundamental result in this subject is the following theorem.
Theorem
22.4 Let f : [a, b] C be continuous and let : [a, b] C be of bounded variation. Then
R
1
, then
f
(t)
d
(t)
exists. Also if m > 0 is such that |t s| < m implies |f (t) f (s)| < m

Z



 f (t) d (t) S (P) 2V (, [a, b])


m

whenever ||P|| < m .

Proof: The function, f , is uniformly continuous because it is defined on a compact set. Therefore, there
exists a decreasing sequence of positive numbers, { m } such that if |s t| < m , then
|f (t) f (s)| <

1
.
m

Let
Fm {S (P) : ||P|| < m }.

401
Thus Fm is a closed set. (When we write S (P) in the above definition, we mean to include all sums
corresponding to P for any choice of j .) We wish to show that
diam (Fm )

2V (, [a, b])
m

(22.2)

R
because then there will exist a unique point, I
m=1 Fm . It will then follow that I = f (t) d (t) . To
verify (22.2), it suffices to verify that whenever P and Q are partitions satisfying ||P|| < m and ||Q|| < m ,
|S (P) S (Q)|

2
V (, [a, b]) .
m

(22.3)

Suppose ||P|| < m and Q P. Then also ||Q|| < m . To begin with, suppose that P {t0 , , tp , , tn }
and Q {t0 , , tp1 , t , tp , , tn } . Thus Q contains only one more point than P. Letting S (Q) and S (P)
be Riemann Steiltjes sums,
S (Q)

p1
X

f ( j ) ( (tj ) (tj1 )) + f ( ) ( (t ) (tp1 ))

j=1

+f ( ) ( (tp ) (t )) +

n
X

f ( j ) ( (tj ) (tj1 )) ,

j=p+1

S (P)

p1
X

f ( j ) ( (tj ) (tj1 )) + f ( p ) ( (t ) (tp1 ))

j=1

+f ( p ) ( (tp ) (t )) +

n
X

f ( j ) ( (tj ) (tj1 )) .

j=p+1

Therefore,
|S (P) S (Q)|

p1
X
1
1
| (tj ) (tj1 )| +
| (t ) (tp1 )| +
m
m
j=1

n
X
1
1
1
| (tp ) (t )| +
| (tj ) (tj1 )| V (, [a, b]) .
m
m
m
j=p+1

(22.4)

Clearly the extreme inequalities would be valid in (22.4) if Q had more than one extra point. Let S (P) and
S (Q) be Riemann Steiltjes sums for which ||P|| and ||Q|| are less than m and let R P Q. Then
|S (P) S (Q)| |S (P) S (R)| + |S (R) S (Q)|

2
V (, [a, b]) .
m

and this shows (22.3) which proves

R (22.2). Therefore, there exists a unique complex number, I m=1 Fm
which satisfies the definition of f (t) d (t) . This proves the theorem.
The following theorem follows easily from the above definitions and theorem.

402

RIEMANN STIELTJES INTEGRALS

Theorem 22.5 Let f C ([a, b]) and let : [a, b] C be of bounded variation. Let
M max {|f (t)| : t [a, b]} .

(22.5)

Then
Z



 f (t) d (t) M V (, [a, b]) .



(22.6)

Also if {fn } is a sequence of functions of C ([a, b]) which is converging uniformly to the function, f, then
Z
Z
lim
fn (t) d (t) =
f (t) d (t) .
(22.7)
n

Proof: Let (22.5) hold. From the proof of the above theorem we know that when ||P|| < m ,
Z



 f (t) d (t) S (P) 2 V (, [a, b])

 m

and so

Z



 f (t) d (t) |S (P)| + 2 V (, [a, b])


m

n
X

M | (tj ) (tj1 )| +

j=1

M V (, [a, b]) +

2
V (, [a, b])
m

2
V (, [a, b]) .
m

This proves (22.6) since m is arbitrary. To verify (22.7) we use the above inequality to write
Z
 Z

Z

 

 f (t) d (t) fn (t) d (t) =  (f (t) fn (t)) d (t)

 


max {|f (t) fn (t)| : t [a, b]} V (, [a, b]) .

Since the convergence is assumed to be uniform, this proves (22.7).
It turns out that we will be mainly interested in the case where is also continuous in addition to being
of bounded variation. Also, it turns out to be much easier to evaluate such integrals in the case where is
also C 1 ([a, b]) . The following theorem about approximation will be very useful.
Theorem 22.6 Let : [a, b] C be continuous and of bounded variation, let f : [a, b] K C be
continuous for K a compact set in C, and let > 0 be given. Then there exists : [a, b] C such that
(a) = (a) , (b) = (b) , C 1 ([a, b]) , and
|| || < ,
Z

Z


 f (t, z) d (t) f (t, z) d (t) < ,



(22.8)

(22.9)

V (, [a, b]) V (, [a, b]) ,

where || || max {| (t) (t)| : t [a, b]} .

(22.10)

403
Proof: We extend to be defined on all R according to (t) = (a) if t < a and (t) = (b) if t > b.
Now we define
2h
Z t+ (ba)
(ta)
1
h (t)
(s) ds.
2h
2h 2h+t+ (ba)
(ta)
where the integral is defined in the obvious way. That is,
Z b
Z b
Z
(t) + i (t) dt
(t) dt + i
a

(t) dt.

Therefore,
h (b) =

h (a) =

b+2h

1
2h

1
2h

(s) ds = (b) ,

b
a

(s) ds = (a) .

a2h

Also, because of continuity of and the fundamental theorem of calculus,

 


1
2h
2h
t+
(t a)
1+

0h (t) =
2h
ba
ba



2h
2h
2h + t +
(t a)
1+
ba
ba
and so h C 1 ([a, b]) . The following lemma is significant.
Lemma 22.7 V ( h , [a, b]) V (, [a, b]) .
Proof: Let a = t0 < t1 < < tn = b. Then using the definition of h and changing the variables to
make all integrals over [0, 2h] ,
n
X

| h (tj ) h (tj1 )| =

j=1

 Z

n 
2h  
X
2h
 1

2h
+
t
+
(t

a)


j
j
 2h 0
ba
j=1



2h
s 2h + tj1 +
(tj1 a) 
ba
1

2h

Z
0

n 
2h X




 s 2h + tj + 2h (tj a)

ba
j=1



2h
s 2h + tj1 +
(tj1 a)  ds.
ba

404

RIEMANN STIELTJES INTEGRALS

2h
For a given s [0, 2h] , the points, s 2h + tj + ba
(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 (, [a 2h, b + 2h]) = V (, [a, b]) .
It follows
n
X

| h (tj ) h (tj1 )| V (, [a, b])

j=1

which proves the lemma.

With this lemma the proof of the theorem can be completed without too much trouble. First of all, if
> 0 is given, there exists 1 such that if h < 1 , then for all t,
| (t) h (t)|

1
2h

<

1
2h

2h
t+ (ba)
(ta)

2h
2h+t+ (ba)
(ta)

| (s) (t)| ds

2h
t+ (ba)
(ta)

ds =

(22.11)

2h
(ta)
2h+t+ (ba)

due to the uniform continuity of . This proves (22.8). From (22.2) there exists 2 such that if ||P|| < 2 ,
then for all z K,
Z

Z



 

 f (t, z) d (t) S (P) < , 
f (t, z) d h (t) Sh (P) <

 3 
 3

for all h. Here S (P) is a Riemann Steiltjes sum of the form

n
X

f ( i , z) ( (ti ) (ti1 ))

i=1

and Sh (P) is a similar Riemann Steiltjes sum taken with respect to h instead of . Therefore, fix the
partition, P, and choose h small enough that in addition to this, we have the following inequality valid for
all z K.

|S (P) Sh (P)| <

3
We can do this thanks to (22.11) and the uniform continuity of f on [a, b] K. It follows
Z

Z




f (t, z) d h (t)
 f (t, z) d (t)


h
Z



 f (t, z) d (t) S (P) + |S (P) Sh (P)|





Z




+ Sh (P)
f (t, z) d h (t) < .


h

Formula (22.10) follows from the lemma. This proves the theorem.
Of course the same result is obtained without the explicit dependence of f on z. R
This is a very useful theorem because if is C 1 ([a, b]) , it is easy to calculate f (t) d (t) . We will
typically reduce to the case where is C 1 by using the above theorem. The next theorem shows how easy
it is to compute these integrals in the case where is C 1 .

405
Theorem 22.8 If f : [a, b] C and : [a, b] C is in C 1 ([a, b]) , then
Z

f (t) d (t) =

f (t) 0 (t) dt.

(22.12)

Proof: Let P be a partition of [a, b], P = {t0 , , tn } and ||P|| is small enough that whenever |t s| <
||P|| ,
|f (t) f (s)| <

(22.13)

and


Z

n
X


 f (t) d (t)

f
(
)
(
(t
)

(t
))
j
j
j1  < .



j=1

Now
n
X

f ( j ) ( (tj ) (tj1 )) =

n
bX

f ( j ) X(tj1 ,tj ] (s) 0 (s) ds

a j=1

j=1

and thanks to (22.13),



Z b n

Z b
X


0
0

f ( j ) X(tj1 ,tj ] (s) (s) ds
f (s) (s) ds

a
 a j=1

<

| 0 (s)| ds.

It follows that

Z
Z b
Z b




0
f (s) (s) ds <
| 0 (s)| ds + .
 f (t) d (t)


a
a

Since is arbitrary, this verifies (22.12).

Definition 22.9 Let : [a, b] U be a continuous function with bounded variation and let f : U C be a
continuous function. Then we define,
Z
Z
f (z) dz
f ( (t)) d (t) .

The expression, f (z) dz, is called a contour integral and is referred to as the contour. We also say that
a function f : U C for U an open set in C has a primitive if there exists a function, F, the primitive, such
that F 0 (z) = f (z) . Thus F is just an antiderivative. Also if k : [ak , bk ] C is continuous and of bounded
variation, for k = 1, , m and k (bk ) = k+1 (ak ) , we define
Z

Pm

k=1

f (z) dz
k

m Z
X

k=1

f (z) dz.

(22.14)

In addition to this, for : [a, b] C, we define : [a, b] C by (t) (b + a t) . Thus simply

traces out the points of ([a, b]) in the opposite order.

406

RIEMANN STIELTJES INTEGRALS

The following lemma is useful and follows quickly from Theorem 22.3.
Lemma 22.10 In the above definition, there exists a continuous bounded variation function, defined on
some closed interval, [c, d] , such that ([c, d]) = m
k=1 k ([ak , bk ]) and (c) = 1 (a1 ) while (d) = m (bm ) .
Furthermore,
Z
m Z
X
f (z) dz =
f (z) dz.

k=1

If : [a, b] C is of bounded variation and continuous, then

Z
Z
f (z) dz =
f (z) dz.

Theorem 22.11 Let K be a compact set in C and let f : U K C be continuous for U an open set in C.
Also let : [a, b] U be continuous with bounded variation. Then if r > 0 is given, there exists : [a, b] U
such that (a) = (a) , (b) = (b) , is C 1 ([a, b]) , and
Z

Z


 f (z, w) dz f (z, w) dz  < r, || || < r.



Proof: Let > 0 be given and let H be an open set containing ([a, b]) such that H is compact. Then
f is uniformly continuous on H K and so there exists a > 0 such that if zj H, j = 1, 2 and wj K for
j = 1, 2 such that if
|z1 z2 | + |w1 w2 | < ,
then
|f (z1 , w1 ) f (z2 , w2 )| < .
By Theorem 22.6, let : [a, b] C be such that ([a, b]) H, (x) = (x) for x = a, b, C 1 ([a, b]) ,
|| || < min (, r) , V (, [a, b]) < V (, [a, b]) , and
Z

Z


 f ( (t) , w) d (t) f ( (t) , w) d (t) <



for all w K. Then, since |f ( (t) , w) f ( (t) , w)| < for all t [a, b] ,

Z
Z


 f ( (t) , w) d (t) f ( (t) , w) d (t) < V (, [a, b]) V (, [a, b]) .



Therefore,

Z

Z


 f (z, w) dz f (z, w) dz  =




Z
Z


 f ( (t) , w) d (t) f ( (t) , w) d (t) < + V (, [a, b]) .



Since > 0 is arbitrary, this proves the theorem.

We will be very interested in the functions which have primitives. It turns out, it is not enough for f to
be continuous in order to possess a primitive. This is in stark contrast to the situation for functions of a real
variable in which the fundamental theorem of calculus will deliver a primitive for any continuous function.
The reason for our interest in such functions is the following theorem and its corollary.

22.1. EXERCISES

407

Theorem 22.12 Let : [a, b] C be continuous and of bounded variation. Also suppose F 0 (z) = f (z) for
all z U, an open set containing ([a, b]) and f is continuous on U. Then
Z
f (z) dz = F ( (b)) F ( (a)) .

Proof: By Theorem 22.11 there exists C 1 ([a, b]) such that (a) = (a) , and (b) = (b) such that
Z

Z


 f (z) dz f (z) dz  < .



Then since is in C 1 ([a, b]) , we may write

Z
Z b
Z
0
f (z) dz =
f ( (t)) (t) dt =

dF ( (t))
dt
dt

= F ( (b)) F ( (a)) = F ( (b)) F ( (a)) .

Therefore,


Z


(F ( (b)) F ( (a))) f (z) dz  <



and since > 0 is arbitrary, this proves the theorem.

Corollary 22.13 If : [a, b] C is continuous, has bounded variation, is a closed curve, (a) = (b) , and
([a, b]) U where U is an open set on which F 0 (z) = f (z) , then
Z
f (z) dz = 0.

22.1

Exercises

1. Let : [a, b] R be increasing. Show V (, [a, b]) = (b) (a) .

2. Suppose : [a, b] C satisfies a Lipschitz condition, | (t) (s)| K |s t| . Show is of bounded
variation and that V (, [a, b]) K |b a| .
3. We say : [c0 , cm ] C is piecewise smooth if there exist numbers, ck , k = 1, , m such that
c0 < c1 < < cm1 < cm such that is continuous and : [ck , ck+1 ] C is C 1 . Show that such
piecewise smooth functions are of bounded variation and give an estimate for V (, [c0 , cm ]) .
R
4. Let : [0, 2] C be given by (t) = r (cos mt + i sin mt) for m an integer. Find dz
z .
5. Show that if : [a, b] C then there exists an increasing function h : [0, 1] [a, b] such that
h ([0, 1]) = ([a, b]) .
6. Let : [a, b] C be an arbitrary continuous curve having bounded variation and let f, g have continuous derivatives on some open set containing ([a, b]) . Prove the usual integration by parts formula.
Z
Z
0
f g dz = f ( (b)) g ( (b)) f ( (a)) g ( (a)) f 0 gdz.

(1/2)

7. RLet f (z) |z|

ei 2 where z = |z| ei . This function is called the principle branch of z (1/2) . Find
f (z) dz where is the semicircle in the upper half plane which goes from (1, 0) to (1, 0) in the

counter clockwise direction. Next do the integral in which goes in the clockwise direction along the
semicircle in the lower half plane.

408

RIEMANN STIELTJES INTEGRALS

8. Prove an open set, U is connected if and only if for every two points in U, there exists a C 1 curve
having values in U which joins them.
9. Let P, Q be two partitions of [a, b] with P Q. Each of these partitions can be used to form an
approximation to V (, [a, b]) as described above. Recall the total variation was the supremum of sums
of a certain form determined by a partition. How is the sum associated with P related to the sum
associated with Q? Explain.
10. Consider the curve,
(t) =

t + it2 sin
0 if t = 0

1
t

if t (0, 1]

Is a continuous curve having bounded variation? What if the t2 is replaced with t? Is the resulting
curve continuous? Is it a bounded variation curve?
R
11. Suppose : [a, b] R is given by (t) = t. What is f (t) d? Explain.

Analytic functions
In this chapter we define what we mean by an analytic function and give a few important examples of
functions which are analytic.
Definition 23.1 Let U be an open set in C and let f : U C. We say f is analytic on U if for every
z U,
f (z + h) f (z)
f 0 (z)
h0
h
lim

exists and is a continuous function of z U. Here h C.

Note that if f is analytic, it must be the case that f is continuous. It is more common to not include the
requirement that f 0 is continuous but we will show later that the continuity of f 0 follows.
What are some examples of analytic functions? The simplest example is any polynomial. Thus
p (z)

n
X

ak z k

k=0

is an analytic function and

p0 (z) =

n
X

ak kz k1 .

k=1

We leave the verification of this as an exercise. More generally, power series are analytic. We will show
this later. For now, we consider the very important Cauchy Riemann equations which give conditions under
which complex valued functions of a complex variable are analytic.
Theorem 23.2 Let U be an open subset of C and let f : U C be a function, such that for z = x + iy U,
f (z) = u (x, y) + iv (x, y) .
Then f is analytic if and only if u, v are C 1 (U ) and
u
v u
v
=
,
= .
x
y y
x
Furthermore, we have the formula,
f 0 (z) =

u
v
(x, y) + i
(x, y) .
x
x
409

410

ANALYTIC FUNCTIONS

Proof: Suppose f is analytic first. Then letting t R,

f 0 (z) = lim

t0

lim

t0

f (z + t) f (z)
=
t

u (x + t, y) + iv (x + t, y) u (x, y) + iv (x, y)

t
t

v (x, y)
u (x, y)
+i
.
x
x

But also
f (z + it) f (z)
=
t0
it

f 0 (z) = lim

lim

t0

u (x, y + t) + iv (x, y + t) u (x, y) + iv (x, y)

it
it


u (x, y)
v (x, y)
+i
y
y

v (x, y)
u (x, y)
i
.
y
y

1
i

This verifies the Cauchy Riemann equations. We are assuming that z f 0 (z) is continuous. Therefore, the
partial derivatives of u and v are also continuous. To see this, note that from the formulas for f 0 (z) given
above, and letting z1 = x1 + iy1


 v (x, y) v (x1 , y1 ) 

 |f 0 (z) f 0 (z1 )| ,

 y

y
showing that (x, y) v(x,y)
is continuous since (x1 , y1 ) (x, y) if and only if z1 z. The other cases are
y
similar.
Now suppose the Cauchy Riemann equations hold and the functions, u and v are C 1 (U ) . Then letting
h = h1 + ih2 ,
f (z + h) f (z) = u (x + h1 , y + h2 )
+iv (x + h1 , y + h2 ) (u (x, y) + iv (x, y))
We know u and v are both differentiable and so
f (z + h) f (z) =

u
u
(x, y) h1 +
(x, y) h2 +
x
y

v
v
(x, y) h1 +
(x, y) h2
x
y

+ o (h) .

23.1. EXERCISES

411

Dividing by h and using the Cauchy Riemann equations,

f (z + h) f (z)
=
h

u
x

v
i x
(x, y) h1 +

v
(x, y) h1 + i y
(x, y) h2

u
y

(x, y) h2

o (h)
h

u
h1 + ih2
v
h1 + ih2
o (h)
(x, y)
+i
(x, y)
+
x
h
x
h
h

Taking the limit as h 0, we obtain

f 0 (z) =

u
v
(x, y) + i
(x, y) .
x
x

It follows from this formula and the assumption that u, v are C 1 (U ) that f 0 is continuous.
It is routine to verify that all the usual rules of derivatives hold for analytic functions. In particular, we
have the product rule, the chain rule, and quotient rule.

23.1

Exercises

1. Verify all the usual rules of differentiation including the product and chain rules.
2. Suppose f and f 0 : U C are analytic and f (z) = u (x, y) + iv (x, y) . Verify uxx + uyy = 0
and vxx + vyy = 0. This partial differential equation satisfied by the real and imaginary parts of
an analytic function is called Laplaces equation. We say these functions satisfying Laplaces equation
R y are harmonic
R x functions. If u is a harmonic function defined on B (0, r) show that v (x, y)
u
(x,
t)
dt

u (t, 0) dt is such that u + iv is analytic.

0 x
0 y
3. Define a function f (z) z x iy where z = x + iy. Is f analytic?
4. If f (z) = u (x, y) + iv (x, y) and f is analytic, verify that


ux uy
2
det
= |f 0 (z)| .
vx vy
5. Show that if u (x, y) + iv (x, y) = f (z) is analytic, then u v = 0. Recall
u (x, y) = hux (x, y) , uy (x, y)i.
6. Show that every polynomial is analytic.
7. If (t) = x (t) + iy (t) is a C 1 curve having values in U, an open set of C, and if f : U C is analytic,
0
we can consider f , another C 1 curve having values in C. Also, 0 (t) and (f ) (t) are complex
2
numbers so these can be considered as vectors in R as follows. The complex number, x+iy corresponds
to the vector, hx, yi. Suppose that and are two such C 1 curves having values in U and that
0
(t0 ) = (s0 ) = z and suppose that f : U C is analytic. Show that the angle between (f ) (t0 )
0
0
0
0
and (f ) (s0 ) is the same as the angle between (t0 ) and (s0 ) assuming that f (z) 6= 0. Thus
analytic mappings preserve angles at points where the derivative is nonzero. Such mappings are called
isogonal. . Hint: To make this easy to show, first observe that hx, yi ha, bi = 12 (zw + zw) where
z = x + iy and w = a + ib.

412

ANALYTIC FUNCTIONS

8. Analytic functions are even better than what is described in Problem 7. In addition to preserving
angles, they also preserve orientation. To verify this show that if z = x + iy and w = a + ib are
two complex numbers, then hx, y, 0i and ha, b, 0i are two vectors in R3 . Recall that the cross product,
hx, y, 0i ha, b, 0i, yields a vector normal to the two given vectors such that the triple, hx, y, 0i, ha, b, 0i,
and hx, y, 0i ha, b, 0i satisfies the right hand rule and has magnitude equal to the product of the
sine of the included angle times the product of the two norms of the vectors. In this case, the
cross product either points in the direction of the positive z axis or in the direction of the negative z axis. Thus, either the vectors hx, y, 0i, ha, b, 0i, k form a right handed system or the vectors
ha, b, 0i, hx, y, 0i, k form a right handed system. These are the two possible orientations. Show that
in the situation of Problem 7 the orientation of 0 (t0 ) , 0 (s0 ) , k is the same as the orientation of the
0
0
vectors (f ) (t0 ) , (f ) (s0 ) , k. Such mappings are called conformal. Hint: You can do this by
0
0
verifying that (f ) (t0 ) (f ) (s0 ) = 0 (t0 ) 0 (s0 ). To make the verification easier, you might
first establish the following simple formula for the cross product where here x + iy = z and a + ib = w.
hx, y, 0i ha, b, 0i = Re (ziw) k.
9. Write the Cauchy Riemann equations in terms of polar coordinates. Recall the polar coordinates are
given by
x = r cos , y = r sin .

23.2

Examples of analytic functions

A very important example of an analytic function is ez ex (cos y + i sin y) exp (z) . We can verify
this is an analytic function by considering the Cauchy Riemann equations. Here u (x, y) = ex cos y and
v (x, y) = ex sin y. The Cauchy Riemann equations hold and the two functions u and v are C 1 (C) . Therefore,
z ez is an analytic function on all of C. Also from the formula for f 0 (z) given above for an analytic function,
d z
e = ex (cos y + i sin y) = ez .
dz
We also see that ez = 1 if and only if z = 2k for k an integer. Other properties of ez follow from the
formula for it. For example, let zj = xj + iyj where j = 1, 2.
ez1 ez2

ex1 (cos y1 + i sin y1 ) ex2 (cos y2 + i sin y2 )

= ex1 +x2 (cos y1 cos y2 sin y1 sin y2 ) +
iex1 +x2 (sin y1 cos y2 + sin y2 cos y1 )

= ex1 +x2 (cos (y1 + y2 ) + i sin (y1 + y2 )) = ez1 +z2 .

Another example of an analytic function is any polynomial. We can also define the functions cos z and
sin z by the usual formulas.
sin z

eiz + eiz
eiz eiz
, cos z
.
2i
2

By the rules of differentiation, it is clear these are analytic functions which agree with the usual functions
in the case where z is real. Also the usual differentiation formulas hold. However,
cos ix =

ex + ex
= cosh x
2

and so cos z is not bounded. Similarly sin z is not bounded.

23.3. EXERCISES

413

A more interesting example is the log function. We cannot define the log for all values of z but if we
leave out the ray, (, 0], then it turns out we can do so. On R + i (, ) it is easy to see that ez is one
to one, mapping onto C \ (, 0]. Therefore, we can define the log on C \ (, 0] in the usual way,
elog z z = eln|z| ei arg(z) ,
where arg (z) is the unique angle in (, ) for which the equal sign in the above holds. Thus we need
log z = ln |z| + i arg (z) .

(23.1)

There are many other ways to define a logarithm. In fact, we could take any ray from 0 and define a logarithm
on what is left. It turns out that all these logarithm functions are analytic. This will be clear from the open
mapping theorem presented later but for now you may verify by brute force that the usual definition of the
logarithm, given in (23.1) and referred to as the principle branch of the logarithm is analytic. This can be
done by verifying the Cauchy Riemann equations in the following.
!!


x
1/2
if y < 0,
log z = ln x2 + y 2
+ i arccos p
x2 + y 2
2

log z = ln x + y

2 1/2

+ i arccos

log z = ln x2 + y 2

1/2

x
p
x2 + y 2

!!

if y > 0,


 y 
+ i arctan
if x > 0.
x

With the principle branch of the logarithm defined, we may define the principle branch of z for any C.
We define
z e log(z) .

23.3

Exercises

1. Verify the principle branch of the logarithm is an analytic function.

2. Find ii corresponding to the principle branch of the logarithm.
3. Show that sin (z + w) = sin z cos w + cos z sin w.
4. If f is analytic on U, an open set in C, when can it be concluded that |f | is analytic? When can it be
concluded that |f | is continuous? Prove your assertions.
5. Let f (z) = z where z x iy for z = x + iy. Describe geometrically what f does and discuss whether
f is analytic.
6. A fractional linear transformation is a function of the form
f (z) =

az + b
cz + d

where ad bc 6= 0. Note that if c = 0, this reduces to a linear transformation (a/d) z + (b/d) . Special
cases of these are given defined as follows.
dilations: z z, 6= 0, inversions: z

1
,
z

414

ANALYTIC FUNCTIONS
translations: z z + .
In the case where c 6= 0, let S1 (z) = z + dc , S2 (z) = z1 , S3 (z) = (bcad)
z and S4 (z) = z + ac . Verify
c2
that f (z) = S4 S3 S2 S1 . Now show that in the case where c = 0, f is still a finite composition of
dilations, inversions, and translations.

7. Show that for a fractional linear transformation described in Problem 6 circles and lines are mapped
to circles or lines. Hint: This is obvious for dilations, and translations. It only remains to verify this
for inversions. Note that all circles and lines may be put in the form




x2 + y 2 2ax 2by = r2 a2 + b2
where = 1 gives a circle centered at (a, b) with radius r and = 0 gives a line. In terms of complex
variables we may consider all possible circles and lines in the form
zz + z + z + = 0,
Verify every circle or line is of this form and that conversely, every expression of this form yields either
a circle or a line. Then verify that inversions do what is claimed.
8. It is desired to find an analytic function, L (z) defined for all z C \ {0} such that eL(z) = z. Is this
possible? Explain why or why not.
9. If f is analytic, show that z f (z) is also analytic.
10. Find the real and imaginary parts of the principle branch of z 1/2 .

Cauchys formula for a disk

In this chapter we prove the Cauchy formula for a disk. Later we will generalize this formula to much more
general situations but the version given here will suffice to prove many interesting theorems needed in the
later development of the theory. First we give a few preliminary results from advanced calculus.
Lemma 24.1 Let f : [a, b] C. Then f 0 (t) exists if and only if Re f 0 (t) and Im f 0 (t) exist. Furthermore,
f 0 (t) = Re f 0 (t) + i Im f 0 (t) .
Proof: The if part of the equivalence is obvious.
Now suppose f 0 (t) exists. Let both t and t + h be contained in [a, b]
 


 Re f (t + h) Re f (t)
  f (t + h) f (t)

0
0



Re (f (t)) 
f (t)

h
h

and this converges to zero as h 0. Therefore, Re f 0 (t) = Re (f 0 (t)) . Similarly, Im f 0 (t) = Im (f 0 (t)) .
Lemma 24.2 If g : [a, b] C and g is continuous on [a, b] and differentiable on (a, b) with g 0 (t) = 0, then
g (t) is a constant.
Proof: From the above lemma, we can apply the mean value theorem to the real and imaginary parts
of g.
Lemma 24.3 Let : [a, b] [c, d] R be continuous and let
Z b
g (t)
(s, t) ds.

(24.1)

Then g is continuous. If

exists and is continuous on [a, b] [c, d] , then

Z b
(s, t)
g 0 (t) =
ds.
t
a

(24.2)

Proof: The first claim follows from the uniform continuity of on [a, b][c, d] , which uniform continuity
results from the set being compact. To establish (24.2), let t and t + h be contained in [c, d] and form, using
the mean value theorem,
Z
g (t + h) g (t)
1 b
=
[ (s, t + h) (s, t)] ds
h
h a
Z
1 b (s, t + h)
=
hds
h a
t
Z b
(s, t + h)
=
ds,
t
a
415

416

CAUCHYS FORMULA FOR A DISK

where may depend on s but is some number between 0 and 1. Then by the uniform continuity of
follows that (24.2) holds.

t ,

it

Corollary 24.4 Let : [a, b] [c, d] C be continuous and let

g (t)

(s, t) ds.

(24.3)

Then g is continuous. If

exists and is continuous on [a, b] [c, d] , then

g 0 (t) =

Z
a

(s, t)
ds.
t

(24.4)

Proof: Apply Lemma 24.3 to the real and imaginary parts of .

With this preparation we are ready to prove Cauchys formula for a disk.
Theorem 24.5 Let f : U C be analytic on the open set, U and let
B (z0 , r) U.
Let (t) z0 + reit for t [0, 2] . Then if z B (z0 , r) ,
1
f (z) =
2i

f (w)
dw.
wz

(24.5)

Proof: Consider for [0, 1] ,

g ()

Z
0

f z + z0 + reit z
reit + z0 z

rieit dt.

If equals one, this reduces to the integral in (24.5). We will show g is a constant and that g (0) = f (z) 2i.
First we consider the claim about g (0) .
2


reit
g (0) =
dt if (z)
reit + z0 z
0

Z 2
1
= if (z)
dt
0
1 zz
0
reit
Z 2 X

n
= if (z)
rn eint (z z0 ) dt
Z

n=0



0
because  zz
< 1. Since this sum converges uniformly we may interchange the sum and the integral to
reit
obtain
g (0)

= if (z)

n=0

=
because

R 2
0

eint dt = 0 if n > 0.

2if (z)

(z z0 )

Z
0

eint dt

417
Next we show that g is constant. By Corollary 24.4, for (0, 1) ,





Z 2 0
f z + z0 + reit z
reit + z0 z
0
g () =
rieit dt
reit + z0 z
0
Z 2



=
f 0 z + z0 + reit z rieit dt
0


Z 2


1
d
=
f z + z0 + reit z
dt
dt

0


1


1
= f z + z0 + rei2 z
f z + z0 + re0 z
= 0.

Now g is continuous on [0, 1] and g 0 (t) = 0 on (0, 1) so by Lemma 24.2, g equals a constant. This constant
can only be g (0) = 2if (z) . Thus,
Z
f (w)
g (1) =
dw = g (0) = 2if (z) .
wz
This proves the theorem.
This is a very significant theorem. We give a few applications next.
Theorem 24.6 Let f : U C be analytic where U is an open set in C. Then f has infinitely many
derivatives on U . Furthermore, for all z B (z0 , r) ,
Z
n!
f (w)
(n)
f (z) =
dw
(24.6)
2i (w z)n+1
where (t) z0 + reit , t [0, 2] for r small enough that B (z0 , r) U.
Proof: Let z B (z0 , r) U and let B (z0 , r) U. Then, letting (t) z0 + reit , t [0, 2] , and h
small enough,
Z
Z
1
f (w)
1
f (w)
f (z) =
dw, f (z + h) =
dw
2i w z
2i w z h
Now
1
1
h

=
wzh wz
(w + z + h) (w + z)
and so
f (z + h) f (z)
h

=
=

Z
1
hf (w)
dw
2hi (w + z + h) (w + z)
Z
1
f (w)
dw.
2i (w + z + h) (w + z)

Now for all h sufficiently small, there exists a constant C independent of such h such that




1
1



 (w + z + h) (w + z) (w + z) (w + z) 




h


= 
 C |h|
 (w z h) (w z)2 

418

CAUCHYS FORMULA FOR A DISK

and so, the integrand converges uniformly as h 0 to

=

f (w)
2

(w z)

Therefore, we may take the limit as h 0 inside the integral to obtain

Z
1
f (w)
0
dw.
f (z) =
2i (w z)2
Continuing in this way, we obtain (24.6).
This is a very remarkable result. We just showed that the existence of one continuous derivative implies
the existence of all derivatives, in contrast to the theory of functions of a real variable. Actually, we just
showed a little more than what the theorem states. The above proof establishes the following corollary.
Corollary 24.7 Suppose f is continuous on B (z0 , r) and suppose that for all z B (z0 , r) ,
Z
f (w)
1
f (z) =
dw,
2i w z
where (t) z + reit , t [0, 2] . Then f is analytic on B (z0 , r) and in fact has infinitely many derivatives
on B (z0 , r) .
We also have the following simple lemma as an application of the above.
Lemma 24.8 Let (t) = z0 + reit , for t [0, 2], suppose fn f uniformly on B (z0 , r), and suppose
Z
1
fn (w)
fn (z) =
dw
(24.7)
2i w z
for z B (z0 , r) . Then
1
f (z) =
2i

f (w)
dw,
wz

(24.8)

implying that f is analytic on B (z0 , r) .

Proof: From (24.7) and the uniform convergence of fn to f on ([0, 2]) , we have that the integrals in
(24.7) converge to
Z
1
f (w)
dw.
2i w z
Therefore, the formula (24.8) follows.
Proposition 24.9 Let {an } denote a sequence of complex numbers. Then there exists R [0, ] such that

ak (z z0 )

k=0

converges absolutely if |z z0 | < R, diverges if |z z0 | > R and converges uniformly on B (z0 , r) for all
r < R. Furthermore, if R > 0, the function,
f (z)

k=0

is analytic on B (z0 , R) .

ak (z z0 )

419
Proof: The assertions about absolute convergence are routine from the root test if we define

1
1/n
R lim sup |an |
n

with R = if the quantity in parenthesis equals zero.

PThe assertion about uniform convergence follows
from the Weierstrass M test if we use Mn |an | rn . ( n=0 |an | rn < by the root test). It only remains
to verify the assertion about f (z) being analytic in the case where R > 0. Let 0 < r < R and define
Pn
k
fn (z) k=0 ak (z z0 ) . Then fn is a polynomial and so it is analytic. Thus, by the Cauchy integral
formula above,
Z
1
fn (w)
fn (z) =
dw
2i w z
where (t) = z0 + reit , for t [0, 2] . By Lemma 24.8 and the first part of this proposition involving uniform
convergence, we obtain
Z
f (w)
1
f (z) =
dw.
2i w z
Therefore, f is analytic on B (z0 , r) by Corollary 24.7. Since r < R is arbitrary, this shows f is analytic on
B (z0 , R) .
This proposition shows that all functions which are given as power series are analytic on their circle
of convergence, the set of complex numbers, z, such that |z z0 | < R. Next we show that every analytic
function can be realized as a power series.
Theorem 24.10 If f : U C is analytic and if B (z0 , r) U, then
f (z) =

an (z z0 )

(24.9)

n=0

for all |z z0 | < r. Furthermore,

an =

f (n) (z0 )
.
n!

Proof: Consider |z z0 | < r and let (t) = z0 + reit , t [0, 2] . Then for w ([0, 2]) ,


 z z0 


 w z0  < 1

and so, by the Cauchy integral formula, we may write

Z
1
f (w)
f (z) =
dw
2i w z
Z
1
f (w)

 dw
=
2i (w z ) 1 zz0
0
wz0
n
Z

X
1
f (w)
z z0
=
dw.
2i (w z0 ) n=0 w z0

Since the series converges uniformly, we may interchange the integral and the sum to obtain
!
Z

X
1
f (w)
n
f (z) =
(z z0 )
n+1
2i
(w

z
)

0
n=0

n=0

an (z z0 )

(24.10)

420

CAUCHYS FORMULA FOR A DISK

By Theorem 24.6 we see that (24.10) holds.

The following theorem pertains to functions which are analytic on all of C, entire functions.
Theorem 24.11 (Liouvilles theorem) If f is a bounded entire function then f is a constant.
Proof: Since f is entire, we can pick any z C and write
1
f (z) =
2i
0

Z
R

f (w)

2 dw

(w z)

where R (t) = z + Reit for t [0, 2] . Therefore,

|f 0 (z)| C

1
R

where C is some constant depending on the assumed bound on f. Since R is arbitrary, we can take R to
obtain f 0 (z) = 0 for any z C. It follows from this that f is constant for if zj j = 1, 2 are two complex numbers, we can consider h (t) = f (z1 + t (z2 z1 )) for t [0, 1] . Then h0 (t) = f 0 (z1 + t (z2 z1 )) (z2 z1 ) = 0.
By Lemma 24.2 h is a constant on [0, 1] which implies f (z1 ) = f (z2 ) .
With Liouvilles theorem it becomes possible to give an easy proof of the fundamental theorem of algebra.
It is ironic that all the best proofs of this theorem in algebra come from the subjects of analysis or topology.
Out of all the proofs that have been given of this very important theorem, the following one based on
Liouvilles theorem is the easiest.
Theorem 24.12 (Fundamental theorem of Algebra) Let
p (z) = z n + an1 z n1 + + a1 z + a0
be a polynomial where n 1 and each coefficient is a complex number. Then there exists z0 C such that
p (z0 ) = 0.
1

Proof: Suppose not. Then p (z)

is an entire function. Also



n
n1
|p (z)| |z| |an1 | |z|
+ + |a1 | |z| + |a0 |



1 
1
and so lim|z| |p (z)| = which implies lim|z| p (z)  = 0. It follows that, since p (z) is bounded
1

for z in any bounded set, we must have that p (z) is a bounded entire function. But then it must be
1
1
constant. However since p (z) 0 as |z| , this constant can only be 0. However, p(z)
is never equal
to zero. This proves the theorem.

24.1

Exercises

P


1. Show that if |ek | , then  k=m ek rk rk+1  < if 0 r < 1. Hint: Let || = 1 and verify that

ek rk r

k=m

where < Re (ek ) < .

k+1

X
 X






=
ek rk rk+1  =
Re (ek ) rk rk+1


k=m

k=m

24.1. EXERCISES

421

P
P
n
2. Abels theoremPsays that if n=0 an (z a) P
has radius of P
convergence equal to
1 and if A = n=0 an ,

then limr1 n=0Pan rn = A. Hint: Show k=0 ak rk = k=0 Ak rk rk+1 where Ak denotes the
kth partial sum of
aj . Thus

k=0

ak rk =

k=m+1

m

X


Ak rk rk+1 +
Ak rk rk+1 ,
k=0

where |Ak A| < for all k m. In the first sum, write Ak = A + ek and use Problem 1. Use this
P
k
1
theorem to verify that arctan (1) = k=0 (1) 2k+1
.
3. Find the integrals using the Cauchy integral formula.
R
z
it
(a) sin
zi dz where (t) = 2e : t [0, 2] .
R 1
(b) za dz where (t) = a + reit : t [0, 2]
R
z
it
(c) cos
z 2 dz where (t) = e : t [0, 2]
R
1 it
(d) log(z)
z n dz where (t) = 1 + 2 e : t [0, 2] and n = 0, 1, 2.
R z2 +4
4. Let (t) = 4eit : t [0, 2] and find z(z
2 +1) dz.
P
5. Suppose f (z) = n=0 an z n for all |z| < R. Show that then
1
2

Z
0

X



2
f rei 2 d =
|an | r2n
n=0

Pn
for all r [0, R). Hint: Let fn (z) k=0 ak z k , show
then take limits as n using uniform convergence.

1
2


R 2 
fn rei 2 d = Pn |ak |2 r2k and
k=0
0

6. The Cauchy integral formula, marvelous as it is, can actually be improved upon. The Cauchy integral
formula involves representing f by the values of f on the boundary of the disk, B (a, r) . It is possible
to represent f by using only the values of Re f on the boundary. This leads to the Schwarz formula .
Supply the details in the following outline.
Suppose f is analytic on |z| < R and
f (z) =

an z n

(24.11)

n=0

with the series converging uniformly on |z| = R. Then letting |w| = R,

2u (w) = f (w) + f (w)
and so
2u (w) =

ak wk +

k=0

Now letting (t) = Reit , t [0, 2]

Z

2u (w)
dw
w

ak (w) .

k=0

= (a0 + a0 )

1
dw
w

2i (a0 + a0 ) .

(24.12)

422

CAUCHYS FORMULA FOR A DISK

Thus, multiplying (24.12) by w1 ,
1
i

u (w)
dw = a0 + a0 .
w

Now multiply (24.12) by w(n+1) and integrate again to obtain

Z
1
u (w)
an =
dw.
i wn+1
Using these formulas for an in (24.11), we can interchange the sum and the integral (Why can we do
this?) to write the following for |z| < R.
Z

1 X  z k+1
1
f (z) =
u (w) dw a0
i z
w
k=0
Z
u (w)
1
=
dw a0 ,
i w z
R u(w)
1
which is the Schwarz formula. Now Re a0 = 2i
dw and a0 = Re a0 i Im a0 . Therefore, we can
w
also write the Schwarz formula as
Z
1
u (w) (w + z)
f (z) =
dw + i Im a0 .
(24.13)
2i (w z) w
7. Take the real parts of the second form of the Schwarz formula to derive the Poisson formula for a disk,




Z 2


u Rei R2 r2
1
i
u re
=
d.
(24.14)
2 0 R2 + r2 2Rr cos ( )
8. Suppose that u (w) is a given real continuous function defined on B (0, R) and define f (z) for |z| < R
by (24.13). Show that f, so defined is analytic. Explain why u given in (24.14) is harmonic. Show that




lim u rei = u Rei .
rR

Thus u is a harmonic function which approaches a given function on the boundary and is therefore, a
solution to the Dirichlet problem.
P
P
k
k1
9. Suppose f (z) = k=0 ak (z z0 ) for all |z z0 | < R. Show that f 0 (z) = k=0 ak k (z z0 )
for
0
all |z z0 | < R. Hint: Let fn (z) be a partial sum of f. Show that fn converges uniformly to some
function, g on |z z0 | r for any r < R. Now use the Cauchy integral formula for a function and its
derivative to identify g with f 0 .

k
P
10. Use Problem 9 to find the exact value of k=0 k 2 13 .
11. Prove the binomial formula,

(1 + z) =

 
X

n=0

zn

where
 

( n + 1)

.
n
n!
n

Can this be used to give a proof of the binomial formula, (a + b) =

Pn

k=0

n
k

ank bk ? Explain.

The general Cauchy integral formula

25.1

The Cauchy Goursat theorem

In this section we prove a fundamental theorem which is essential to the development which follows and is
closely related to the question of when a function has a primitive. First of all, if we are given two points in
C, z1 and z2 , we may consider (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 we 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 zm1 to zm . We denote this piecewise linear curve as (z1 , , zm ) . Now let T be a
triangle with vertices z1 , z2 and z3 encountered in the counter clockwise direction as shown.
z3
@
@
@
@
@ z2
z1
R
R
Then we will denote by T f (z) dz, the expression, (z1 ,z2 ,z3 ,z1 ) f (z) dz. Consider the following picture.
z3
@

@
I T
@
T11
@
@
R
@
T21
@
@

@
@
I
@
T31
T41
I
@
@ z
@
z1
2
By Lemma 22.10 we may conclude that
Z

f (z) dz =

4 Z
X

k=1

f (z) dz.

(25.1)

Tk1

On the inside lines the integrals cancel as claimed in Lemma 22.10 because there are two integrals going
in opposite directions for each of these inside lines. Now we are ready to prove the Cauchy Goursat theorem.
Theorem 25.1 (Cauchy Goursat) Let f : U C have the property that f 0 (z) exists for all z U and let
T be a triangle contained in U. Then
Z
f (w) dw = 0.
T

423

424

THE GENERAL CAUCHY INTEGRAL FORMULA

Proof: Suppose not. Then

Z




From (25.1) it follows



f (w) dw = 6= 0.



4 Z

X



f (w) dw

 T 1

k=1

and so for at least one of these Tk1 , denoted from now on as T1 , we must have
Z




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




 .
f
(w)
dw

 42
T2

Continue in this way, obtaining a sequence of triangles,

Tk Tk+1 , diam (Tk ) diam (T ) 2k ,

and
Z




f (w) dw k .
4
Tk

0
Then let z
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)| < |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 Corollary 22.13.
Z
Tk

f (w) dw =

g (w) dw.

Tk

From the definition, of the integral, we see

Z






g (w) dw diam (Tk ) (length of Tk )
k
4
Tk
2k (length of T ) diam (T ) 2k ,

and so
(length of T ) diam (T ) .
R
Since is arbitrary, this shows = 0, a contradiction. Thus T f (w) dw = 0 as claimed.
This fundamental result yields the following important theorem.

25.1. THE CAUCHY GOURSAT THEOREM

425

Theorem 25.2 (Morera) Let U be an open set and let f 0 (z) exist for all z U . Let D B (z0 , r) U.
Then there exists > 0 such that f has a primitive on B (z0 , r + ).
Proof: Choose > 0 small enough that B (z0 , r + ) U. Then for w B (z0 , r + ) , define
Z
F (w)
f (u) du.
(z0 ,w)

Then by the Cauchy Goursat theorem, and w B (z0 , r + ) , it follows that for |h| small enough,
Z
F (w + h) F (w)
1
=
f (u) du
h
h (w,w+h)
1
=
h

f (w + th) hdt =

f (w + th) dt

which converges to f (w) due to the continuity of f at w. This proves the theorem.
We can also give the following corollary whose proof is similar to the proof of the above theorem.
Corollary 25.3 Let U be an open set and suppose that whenever
(z1 , z2 , z3 , z1 )
is a closed curve bounding a triangle T, which is contained in U, and f is a continuous function defined on
U, it follows that
Z
f (z) dz = 0,
(z1 ,z2 ,z3 ,z1 )

then f is analytic on U.
Proof: As in the proof of Moreras theorem, let B (z0 , r) U and use the given condition to construct
a primitive, F for f on B (z0 , r) . Then F is analytic and so by Theorem 24.6, it follows 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 U.
Theorem 25.4 Let U be an open set in C and suppose f : U C has the property that f 0 (z) exists for
each z U. Then f is analytic on U.
Proof: Let z0 U and let B (z0 , r) U. By Moreras theorem f has a primitive, F on B (z0 , r) . It follows
that F is analytic because it has a derivative, f, and this derivative is continuous. Therefore, by Theorem
24.6 F has infinitely many derivatives on B (z0 , r) implying that f also has infinitely many derivatives on
B (z0 , r) . Thus f is analytic as claimed.
It follows that we can say a function is analytic on an open set, U if and only if f 0 (z) exists for z U.
We just proved the derivative, if it exists, is automatically continuous.
The same proof used to prove Theorem 25.2 implies the following corollary.
Corollary 25.5 Let U be a convex open set and suppose that f 0 (z) exists for all z U. Then f has a
primitive on U.
Note that this implies that if U is a convex open set on which f 0 (z) exists and if : [a, b] U is a closed,
continuous curve having bounded variation, then letting F be a primitive of f Theorem 22.12 implies
Z
f (z) dz = F ( (b)) F ( (a)) = 0.

426

THE GENERAL CAUCHY INTEGRAL FORMULA

Notice how different this is from the situation of a function of a real variable. It is possible for a function
of a real variable to have a derivative everywhere and yet the derivative can be discontinuous. A simple
example is the following.


 2
x sin x1 if x 6= 0
f (x)
.
0 if x = 0
Then f 0 (x) exists for all x R. Indeed, if x 6= 0, the derivative equals 2x sin x1 cos x1 which has no limit
as x 0. However, from the definition of the derivative of a function of one variable, we see easily that
f 0 (0) = 0.

25.2

The Cauchy integral formula

Here we develop the general version of the Cauchy integral formula valid for arbitrary closed rectifiable
curves. The key idea in this development is the notion of the winding number. This is the number defined
in the following theorem, also called the index. We make use of this winding number along with the earlier
results, especially Liouvilles theorem, to give an extremely general Cauchy integral formula.
Theorem 25.6 Let : [a, b] C be continuous and have bounded variation with (a) = (b) . Also suppose
that z
/ ([a, b]) . We define
Z
1
dw
n (, z)
.
(25.2)
2i w z
Then n (, ) is continuous and integer valued. Furthermore, there exists a sequence, k : [a, b] C such
that k is C 1 ([a, b]) ,
1
, (a) = k (b) = (a) = (b) ,
k k

|| k || <

and n ( k , z) = n (, z) for all k large enough. Also n (, ) is constant on every component of C \ ([a, b])
and equals zero on the unbounded component of C \ ([a, b]) .
Proof: First we verify the assertion about continuity.
Z 

|n (, z) n (, z1 )| C 

1
1

wz
w z1

e (Length of ) |z1 z|
C



dw

whenever z1 is close enough to z. This proves the continuity assertion.

Next we need to show the winding number equals an integer. To do so, use Theorem 22.11 to obtain k ,
a function in C 1 ([a, b]) such that z
/ k ([a, b]) for all k large enough, k (x) = (x) for x = a, b, and


Z
 1 Z dw
1
dw  1
1



 < , || k || < .
 2i w z

2i k w z
k
k
We will show each of

1
2i

dw
k wz

is an integer. To simplify the notation, we write instead of k .

Z

dw
=
wz

Z
a

0 (s) ds
.
(s) z

25.2. THE CAUCHY INTEGRAL FORMULA

427

We define
g (t)

Z
a

0 (s) ds
.
(s) z

(25.3)

Then


0
eg(t) ( (t) z)

= eg(t) 0 (t) eg(t) g 0 (t) ( (t) z)

= eg(t) 0 (t) eg(t) 0 (t) = 0.

It follows that eg(t) ( (t) z) equals a constant. In particular, using the fact that (a) = (b) ,
eg(b) ( (b) z) = eg(a) ( (a) z) = ( (a) z) = ( (b) z)
and so eg(b) = 1. This happens if and only if g (b) = 2mi for some integer m. Therefore, (25.3) implies
Z
Z b 0
dw
(s) ds
=
.
2mi =

(s)

z
w
z

a
R
R dw
1
dw
1
Therefore, 2i
is a sequence of integers converging to 2i
n (, z) and so n (, z) must also
k wz
wz
be an integer and n ( k , z) = n (, z) for all k large enough.
Since n (, ) is continuous and integer valued, it follows that it must be constant on every connected
component of C \ ([a, b]) . It is clear that n (, z) equals zero on the unbounded component because from
the formula,


1
lim |n (, z)| lim V (, [a, b])
z
z
|z| c
where c max {|w| : w ([a, b])} .This proves the theorem.
It is a good idea to consider a simple case to get an idea of what the winding number is measuring. To
do so, consider : [a, b] C such that is continuous, closed and bounded variation. Suppose also that is
one to one on (a, b) . Such a curve is called a simple closed curve. It can be shown that such a simple closed
curve divides the plane into exactly two components, an inside bounded component and an outside
unbounded component. This is called the Jordan Curve theorem or the Jordan separation theorem. For a
proof of this difficult result, see the chapter on degree theory. For now, it suffices to simply assume that
is such that this result holds. This will usually be obvious anyway. We also suppose
that it is possible to


change the parameter to be in [0, 2] , in such a way that (t) + z + reit (t) z 6= 0 for all t [0, 2]
and [0, 1] . (As t goes from 0 to 2 the point (t) traces the curve ([0, 2]) in the counter clockwise
direction.) Suppose z D, the inside of the simple closed curve and consider the curve (t) = z + reit for
t [0, 2] where r is chosen small enough that B (z, r) D. Then we claim that n (, z) = n (, z) .
Proposition 25.7 Under the above conditions, n (, z) = n (, z) and n (, z) = 1.
Proof: By changing the parameter, we may assume that [a, b] =
[0, 2] . From Theorem 25.6 it suffices
to assume also that is C 1 . Define h (t) (t) + z + reit (t) for [0, 1] . (This function is called
a homotopy of the curves and .) Note that for each [0, 1] , t h (t) is a closed C 1 curve. Also,


Z
Z 2
0 (t) + rieit 0 (t)
1
1
1
dw =
dt.
2i h w z
2i 0 (t) + (z + reit (t)) z
We know this number is an integer and it is routine to verify that it is a continuous function of . When
= 0 it equals n (, z) and when = 1 it equals n (, z). Therefore, n (, z) = n (, z) . It only remains to
compute n (, z) .
Z 2
1
rieit
n (, z) =
dt = 1.
2i 0
reit

428

THE GENERAL CAUCHY INTEGRAL FORMULA

This proves the proposition.

Now if was not one to one but caused the point, (t) to travel around ([a, b]) twice, we could modify
the above argument to have the parameter interval, [0, 4] and still find n (, z) = n (, z) only this time,
n (, z) = 2. Thus the winding number is just what its name suggests. It measures the number of times the
curve winds around the point. One might ask why bother with the winding number if this is all it does. The
reason is that the notion of counting the number of times a curve winds around a point is rather vague. The
winding number is precise. It is also the natural thing to consider in the general Cauchy integral formula
presented below. We have in mind a situation typified by the following picture in which U is the open set
between the dotted curves and j are closed rectifiable curves in U.

3

2

The following theorem is the general Cauchy integral formula.

Theorem 25.8 Let U be an open subset of the plane and let f : U C be analytic. If k : [ak , bk ]
U, k = 1, , m are continuous closed curves having bounded variation such that for all z
/ U,
m
X

n ( k , z) = 0,

k=1

then for all z U \ m

k=1 k ([ak , bk ]) ,
f (z)

m
X

n ( k , z) =

k=1

Z
m
X
f (w)
1
dw.
2i k w z

k=1

Proof: Let be defined on U U by

(z, w)

f (w)f (z)
wz
0

if w 6= z
.
f (z) if w = z

Then is analytic as a function of both z and w and is continuous in U U. The claim that this function
is analytic as a function of both z and w is obvious at points where z 6= w, and is most easily seen using
Theorem 24.10 at points, where z = w. Indeed, if (z, z) is such a point, we need to verify that w (z, w)
is analytic even at w = z. But by Theorem 24.10, for all h small enough,


(z, z + h) (z, z)
1 f (z + h) f (z)
=
f 0 (z)
h
h
h
"
#
1 1 X f (k) (z) k
0
=
h f (z)
h h
k!
k=1
"
#
X f (k) (z)
f 00 (z)
k2
=
h

.
k!
2!
k=2

25.2. THE CAUCHY INTEGRAL FORMULA

429

Similarly, z (z, w) is analytic even if z = w.

We define
m

1 X
2i

h (z)

(z, w) dw.

k=1

We wish to show that h is analytic on U. To do so, we verify

Z
h (z) dz = 0
T

for every triangle, T, contained in U and apply Corollary 25.3. To do this we use Theorem 22.11 to obtain
for each k, a sequence of functions, kn C 1 ([ak , bk ]) such that
kn (x) = k (x) for x {ak , bk }
and
kn ([ak , bk ]) U, || kn k || <

1
,
n

Z

Z


1


(z, w) dw
(z, w) dw < ,

 kn
 n
k

for all z T. Then applying Fubinis theorem, we can write

Z Z
Z
Z
(z, w) dwdz =
T

kn

kn

because is given to be analytic. By (25.4),

Z Z
Z
(z, w) dwdz = lim
T

and so h is analytic on U as claimed.

Now let H denote the set,
(
H

z C\

m
k=1

(25.4)

(z, w) dzdw = 0

k ([ak , bk ]) :

(z, w) dwdz = 0

kn

m
X

n ( k , z) = 0 .

k=1

Pm
We know that H is an open set because z k=1 n ( k , z) is integer valued and continuous. Define
(
h (z) if z U
Pm R f (w)
.
(25.5)
g (z)
1
k=1 wz dw if z H
2i
k

We need to verify that g (z) is well defined. For z U H, we know z

/ m
k=1 k ([ak , bk ]) and so
m

g (z)

=
=
=

1 X
2i
1
2i
1
2i

k=1 k
m Z
X
k=1 k
m Z
X

k=1

f (w) f (z)
dw
wz
m

f (w)
1 X
dw
wz
2i

k=1

f (w)
dw
wz

Z
k

f (z)
dw
wz

430

THE GENERAL CAUCHY INTEGRAL FORMULA

because z H. This shows g (z) is well defined. Also, g is analytic on U because it equals h there. It is
routine to verify that g is analytic on H also. By assumption, U C H and so U H = C showing that g is
an entire function.P
m
Now note that k=1 n ( k , z) = 0 for all z contained in the unbounded component of C\ m
k=1 k ([ak , bk ])
C
which component contains B (0, r) for r large enough. It follows that for |z| > r, it must be the case that
z H and so for such z, the bottom description of g (z) found in (25.5) is valid. Therefore, it follows
lim |g (z)| = 0

|z|

and so g is bounded and entire. By Liouvilles theorem, g is a constant. Hence, from the above equation,
the constant can only equal zero.
For z U \ m
k=1 k ([ak , bk ]) ,
m

0=

1 X
2i

k=1

1 X
2i

k=1

f (w) f (z)
dw =
wz
m

Z
k

X
f (w)
dw f (z)
n ( k , z) .
wz
k=1

This proves the theorem.

Corollary 25.9 Let U be an open set and let k : [ak , bk ] U , k = 1, , m, be closed, continuous and of
bounded variation. Suppose also that
m
X

n ( k , z) = 0

k=1

for all z
/ U. Then if f : U C is analytic, we have
m Z
X

k=1

f (w) dw = 0.

Proof: This follows from Theorem 25.8 as follows. Let

g (w) = f (w) (w z)
where z U \ m
k=1 k ([ak , bk ]) . Then by this theorem,
0=0

m
X

k=1

n ( k , z) = g (z)

m
X

n ( k , z) =

k=1

Z
m
m Z
X
1
g (w)
1 X
dw =
f (w) dw.
2i k w z
2i
k

k=1

k=1

Another simple corollary to the above theorem is Cauchys theorem for a simply connected region.
Definition 25.10 We say an open set, U C is a region if it is open and connected. We say U is simply
b \U is connected.
connected if C

25.2. THE CAUCHY INTEGRAL FORMULA

431

Corollary 25.11 Let : [a, b] U be a continuous closed curve of bounded variation where U is a simply
connected region in C and let f : U C be analytic. Then
Z
f (w) dw = 0.

b ([a, b]). Thus C\

b ([a, b]) .Then the
Proof:
Let D denote the unbounded component of C\
b \ U is contained in D since every point of C
b \ U must be in some component of C\
b ([a, b])
connected set, C
b
b
and is contained in both C\U and D. Thus D must be the component that contains C \ U. It follows that
b \ U, its value being its value on D. However, for z D,
n (, ) must be constant on C
Z
1
1
n (, z) =
dw
2i w z
and so lim|z| n (, z) = 0 showing n (, z) = 0 on D. Therefore we have verified the hypothesis of Theorem
25.8. Let z U D and define
g (w) f (w) (w z) .
Thus g is analytic on U and by Theorem 25.8,
0 = n (z, ) g (z) =

1
2i

g (w)
1
dw =
wz
2i

f (w) dw.

This proves the corollary.

The following is a very significant result which will be used later.
Corollary 25.12 Suppose U is a simply connected open set and f : U C is analytic. Then f has a
primitive, F, on U. Recall this means there exists F such that F 0 (z) = f (z) for all z U.
Proof: Pick a point, z0 U and let V denote those points, z of U for which there exists a curve,
: [a, b] U such that is continuous, of bounded variation, (a) = z0 , and (b) = z. Then it is easy to
verify that V is both open and closed in U and therefore, V = U because U is connected. Denote by z0 ,z
such a curve from z0 to z and define
Z
F (z)
f (w) dw.
z0 ,z

Then F is well defined because if j , j = 1, 2 are two such curves, it follows from Corollary 25.11 that
Z
Z
f (w) dw +
f (w) dw = 0,
2

implying that
Z

f (w) dw =

f (w) dw.

Now this function, F is a primitive because, thanks to Corollary 25.11

Z
1
1
(F (z + h) F (z)) h
=
f (w) dw
h z,z+h
Z
1 1
=
f (z + th) hdt
h 0
and so, taking the limit as h 0, we see F 0 (z) = f (z) .

432

25.3

THE GENERAL CAUCHY INTEGRAL FORMULA

Exercises

1. If U is simply connected, f is analytic on U and f has no zeros in U, show there exists an analytic
function, F, defined on U such that eF = f.
P
k
2. Let U be an open set and let f be analytic on U. Show that if a U, then f (z) = k=0 bk (z a)
whenever |z a| < R where R is the distance between a and the nearest point where f fails to have
a derivative. The number R, is called the radius of convergence and the power series is said to be
expanded about a.
1
3. Find the radius of convergence of the function 1+z
2 expanded about a = 2. Note there is nothing wrong
1
with the function, 1+x2 when considered as a function of a real variable, x for any value of x. However,
if we insist on using power series, we find that there is a limitation on the values of x for which the
power series converges due to the presence in the complex plane of a point, i, where the function fails
to have a derivative.

4. What if we defined an open set, U to be simply connected if C \ U is connected. Would it amount to

the same thing? Hint: Consider the outside of B (0, 1) .
R
5. Let (t) = eit : t [0, 2] . Find z1n dz for n = 1, 2, .

2n 1

R 2
R
2n
it
6. Show i 0 (2 cos ) d = z + z1
: t [0, 2] . Then evaluate this
z dz where (t) = e
integral using the binomial theorem and the previous problem.
7. Let f : U C be analytic and f (z) = u (x, y) + iv (x, y) . Show u, v and uv are all harmonic although
it can happen that u2 is not. Recall that a function, w is harmonic if wxx + wyy = 0.
8. Suppose that for some constants a, b 6= 0, a, b R, f (z + ib) = f (z) for all z C and f (z + a) = f (z)
for all z C. If f is analytic, show that f must be constant. Can you generalize this? Hint: This
uses Liouvilles theorem.

The open mapping theorem

In this chapter we present the open mapping theorem for analytic functions. This important result states
that analytic functions map connected open sets to connected open sets or else to single points. It is very
different than the situation for a function of a real variable.

26.1

Zeros of an analytic function

In this section we give a very surprising property of analytic functions which is in stark contrast to what
takes place for functions of a real variable. It turns out the zeros of an analytic function which is not constant
on some region cannot have a limit point.
Theorem 26.1 Let U be a connected open set (region) and let f : U C be analytic. Then the following
are equivalent.
1. f (z) = 0 for all z U
2. There exists z0 U such that f (n) (z0 ) = 0 for all n.
3. There exists z0 U which is a limit point of the set,
Z {z U : f (z) = 0} .
Proof: It is clear the first condition implies the second two. Suppose the third holds. Then for z near
z0 we have
f (z) =

X
f (n) (z0 )
n
(z z0 )
n!

n=k

where k 1 since z0 is a zero of f. Suppose k < . Then,

k

f (z) = (z z0 ) g (z)
where g (z0 ) 6= 0. Letting zn z0 where zn Z, zn 6= z0 , it follows
k

0 = (zn z0 ) g (zn )
which implies g (zn ) = 0. Then by continuity of g, we see that g (z0 ) = 0 also, contrary to the choice of k.
Therefore, k cannot be less than and so z0 is a point satisfying the second condition.
Now suppose the second condition and let
n
o
S z U : f (n) (z) = 0 for all n .
433

434

THE OPEN MAPPING THEOREM

It is clear that S is a closed set which by assumption is nonempty. However, this set is also open. To see
this, let z S. Then for all w close enough to z,
f (w) =

X
f (k) (z)

k=0

k!

(w z) = 0.

Thus f is identically equal to zero near z S. Therefore, all points near z are contained in S also, showing
that S is an open set. Now U = S (U \ S) , the union of two disjoint open sets, S being nonempty. It
follows the other open set, U \ S, must be empty because U is connected. Therefore, the first condition is
verified. This proves the theorem. (See the following diagram.)
1.)
.%
2.)

&

3.)

Note how radically different this from the theory of functions of a real variable. Consider, for example
the function


 2
x sin x1 if x 6= 0
f (x)
0 if x = 0
which has a derivative for all x R and for which 0 is a limit point of the set, Z, even though f is not
identically equal to zero.

26.2

The open mapping theorem

With this preparation we are ready to prove the open mapping theorem, an even more surprising result than
the theorem about the zeros of an analytic function.
Theorem 26.2 (Open mapping theorem) Let U be a region in C and suppose f : U C is analytic. Then
f (U ) is either a point or a region. In the case where f (U ) is a region, it follows that for each z0 U, there
exists an open set, V containing z0 such that for all z V,
m

f (z) = f (z0 ) + (z)

(26.1)

where : V B (0, ) is one to one, analytic and onto, (z0 ) = 0, 0 (z) 6= 0 on V and 1 analytic on
B (0, ) . If f is one to one, then m = 1 for each z0 and f 1 : f (U ) U is analytic.
Proof: Suppose f (U ) is not a point. Then if z0 U it follows there exists r > 0 such that f (z) 6= f (z0 )
for all z B (z0 , r) \ {z0 } . Otherwise, z0 would be a limit point of the set,
{z U : f (z) f (z0 ) = 0}
which would imply from Theorem 26.1 that f (z) = f (z0 ) for all z U. Therefore, making r smaller if
necessary, we may write, using the power series of f,
m

f (z) = f (z0 ) + (z z0 ) g (z)

0

for all z B (z0 , r) , where g (z) 6= 0 on B (z0 , r) . Then gg is an analytic function on B (z0 , r) and so
by Corollary 25.5 it has a primitive on B (z0 , r) , h. Therefore, using the product rule and the chain rule,

0
geh = 0 and so there exists a constant, C = ea+ib such that on B (z0 , r) ,
geh = ea+ib .

26.2. THE OPEN MAPPING THEOREM

435

Therefore,
g (z) = eh(z)+a+ib
and so, modifying h by adding in the constant, a + ib, we see g (z) = eh(z) where h0 (z) =
Letting
(z) = (z z0 ) e

g 0 (z)
g(z)

on B (z0 , r) .

h(z)
m

we obtain the formula (26.1) valid on B (z0 , r) . Now

0 (z0 ) = e

h(z0 )
m

6= 0

and so, restricting r we may assume that 0 (z) 6= 0 for all z B (z0 , r). We need to verify that there is an
open set, V contained in B (z0 , r) such that maps V onto B (0, ) for some > 0.
Let (z) = u (x, y) + iv (x, y) where z = x + iy. Then

 

u (x0 , y0 )
0
=
v (x0 , y0 )
0
because for z0 = x0 + iy0 , (z0 ) = 0. In addition to this, the functions u and v are in C 1 (B (0, r)) because
is analytic. By the Cauchy Riemann equations,

 

 ux (x0 , y0 ) uy (x0 , y0 )   ux (x0 , y0 ) vx (x0 , y0 ) 

=

 vx (x0 , y0 ) vy (x0 , y0 )   vx (x0 , y0 ) ux (x0 , y0 ) 

2
= u2x (x0 , y0 ) + vx2 (x0 , y0 ) = 0 (z0 ) 6= 0.

Therefore, by the inverse function theorem there exists an open set, V, containing z0 and > 0 such that
T
(u, v) maps V one to one onto B (0, ) . Thus is one to one onto B (0, ) as claimed. It follows that m
maps V onto B (0, m ) . Therefore, the formula (26.1) implies that f maps the open set, V, containing z0 to
an open set. This shows f (U ) is an open set. It is connected because f is continuous and U is connected.
Thus f (U ) is a region. It only remains to verify that 1 is analytic on B (0, ) . We show this by verifying
the Cauchy Riemann equations.
Let

 

u (x, y)
u
=
(26.2)
v (x, y)
v
T

for (u, v) B (0, ) . Then, letting w = u + iv, it follows that 1 (w) = x (u, v) + iy (u, v) . We need to
verify that
xu = yv , xv = yu .

(26.3)

The inverse function theorem has already given us the continuity of these partial derivatives. From the
equations (26.2), we have the following systems of equations.
ux xu + uy yu = 1
ux xv + uy yv = 0
,
.
vx xu + vy yu = 0
vx xv + vy yv = 1
Solving these for xu , yv , xv , and yu , and using the Cauchy Riemann equations for u and v, yields (26.3).
2i
It only remains to verify the assertion about the case where f is one to one. If m > 1, then e m 6= 1 and
so for z1 V,
e

2i
m

(z1 ) 6= (z1 ) .

436
But e
then

THE OPEN MAPPING THEOREM

2i
m

(z1 ) B (0, ) and so there exists z2 6= z1 (since is one to one) such that (z2 ) = e
m

(z2 )

2i
m

(z1 ) . But

 2i
m
m
= (z1 )
= e m (z1 )

implying f (z2 ) = f (z1 ) contradicting the assumption that f is one to one. Thus m = 1 and f 0 (z) = 0 (z) 6=
0 on V. Since f maps open sets to open sets, it follows that f 1 is continuous and so we may write

0
f 1 (f (z))

f 1 (f (z1 )) f 1 (f (z))
f (z1 ) f (z)
f (z1 )f (z)
z1 z
1
= lim
= 0
.
z1 z f (z1 ) f (z)
f (z)
=

lim

This proves the theorem.

One does not have to look very far to find that this sort of thing does not hold for functions mapping R
to R. Take for example, the function f (x) = x2 . Then f (R) is neither a point nor a region. In fact f (R)
fails to be open.

26.3

Applications of the open mapping theorem

Definition 26.3 We will denote by a ray starting at 0. Thus is a straight line of infinite length extending
in one direction with its initial point at 0.
As a simple application of the open mapping theorem, we give the following theorem about branches of
the logarithm.
Theorem 26.4 Let be a ray starting at 0. Then there exists an analytic function, L (z) defined on C \
such that
eL(z) = z.
We call L a branch of the logarithm.
Proof: Let be an angle of the ray, . The function, ez is a one to one and onto mapping from
R + i (, + 2) to C \ and so we may define L (z) for z C \ such that eL(z) = z and we see that L
defined in this way is analytic on C \ because of the open mapping theorem. Note we could just as well
have considered R + i ( 2, ) . This would have given another branch of the logarithm valid on C \ .
Also, there are infinitely many choices for , each of which leads to a branch of the logarithm by the process
just described.
Here is another very significant theorem known as the maximum modulus theorem which follows immediately from the open mapping theorem.
Theorem 26.5 (maximum modulus theorem) Let U be a bounded region and let f : U C be analytic and
f : U C continuous. Then if z U,
|f (z)| max {|f (w)| : w U } .

(26.4)

If equality is achieved for any z U, then f is a constant.

Proof: Suppose f is not a constant. Then f (U ) is a region and so if z U, there exists
 r > 0 such that

B (f (z) , r) f (U ) . It follows there exists z1 U with |f (z1 )| > |f (z)| . Hence max |f (w)| : w U is
not achieved at any interior point of U. Therefore, the point at which the maximum is achieved must lie on
the boundary of U and so


max {|f (w)| : w U } = max |f (w)| : w U > |f (z)|
for all z U or else f is a constant. This proves the theorem.

26.4. COUNTING ZEROS

26.4

437

Counting zeros

The above proof of the open mapping theorem relies on the very important inverse function theorem from
real analysis. The proof features this and the Cauchy Riemann equations to indicate how the assumption
f is analytic is used. There are other approaches to this important theorem which do not rely on the
big theorems from real analysis and are more oriented toward the use of the Cauchy integral formula and
specialized techniques from complex analysis. We give one of these approaches next which involves the notion
of counting zeros. The next theorem is the one about counting zeros. We will use the theorem later in the
proof of the Riemann mapping theorem.
Theorem 26.6 Let U be a region and let : [a, b] U be closed, continuous, bounded variation, and
n (, z) = 0 for all z
/ U. Suppose also that f is analytic on U having zeros a1 , , am where the zeros are
repeated according to multiplicity, and suppose that none of these zeros are on ([a, b]) . Then
1
2i
Proof: We are given f (z) =

Qm

j=1

X
f 0 (z)
dz =
n (, ak ) .
f (z)
k=1

(z aj ) g (z) where g (z) 6= 0 on U. Hence

m

f 0 (z) X 1
g 0 (z)
=
+
f (z)
z aj
g (z)
j=1
and so
1
2i

X
f 0 (z)
1
dz =
n (, aj ) +
f (z)
2i
j=1

g 0 (z)
dz.
g (z)

(z)
But the function, z gg(z)
is analytic and so by Corollary 25.9, the last integral in the above expression
equals 0. Therefore, this proves the theorem.

Theorem 26.7 Let U be a region, let : [a, b] U be continuous, closed and bounded variation such
that n (, z) = 0 for all z
/ U. Also suppose f : U C be analytic and that
/ f ( ([a, b])) . Then
f : [a, b] C is continuous, closed, and bounded variation. Also suppose {a1 , , am } = f 1 () where
these points are counted according to their multiplicities as zeros of the function f Then
n (f , ) =

m
X

n (, ak ) .

k=1

Proof: It is clear that f is closed and continuous. It only remains to verify that it is of bounded
variation. Suppose first that ([a, b]) B B U where B is a ball. Then
|f ( (t)) f ( (s))| =
Z






f ( (s) + ( (t) (s))) ( (t) (s)) d
0

C | (t) (s)|


where C max |f 0 (z)| : z B . Hence, in this case,

V (f , [a, b]) CV (, [a, b]) .

438

THE OPEN MAPPING THEOREM

Now let denote the distance between ([a, b]) and C \ U. Since ([a, b]) is compact, > 0. By uniform

continuity there exists = ba

p for p a positive integer such that if |s t| < , then | (s) (t)| < 2 . Then


([t, t + ]) B (t) ,
U.
2
n

o
Let C max |f 0 (z)| : z pj=1 B (tj ) , 2
where tj pj (b a) + a. Then from what was just shown,
V (f , [a, b])

p1
X

V (f , [tj , tj+1 ])

j=0

p1
X

V (, [tj , tj+1 ]) <

j=0

showing that f is bounded variation as claimed. Now from Theorem 25.6 there exists C 1 ([a, b]) such
that
(a) = (a) = (b) = (b) , ([a, b]) U,
and
n (, ak ) = n (, ak ) , n (f , ) = n (f , )

(26.5)

for k = 1, , m. Then
n (f , ) = n (f , )
=
=
=
=

1
2i

f
b

dw
w

1
f 0 ( (t)) 0
(t) dt
2i a f ( (t))
Z
1
f 0 (z)
dz
2i f (z)
m
X
n (, ak )
Z

k=1

Pm

By Theorem 26.6. By (26.5), this equals k=1 n (, ak ) which proves the theorem.
The next theorem is very interesting for its own sake.
Theorem 26.8 Let f : B (a, R) C be analytic and let
m

f (z) = (z a) g (z) , > m 1

where g (z) 6= 0 in B (a, R) . (f (z) has a zero of order m at z = a.) Then there exist , > 0 with the
property that for each z satisfying 0 < |z | < , there exist points,
{a1 , , am } B (a, ) ,
such that
f 1 (z) B (a, ) = {a1 , , am }
and each ak is a zero of order 1 for the function f () z.

26.4. COUNTING ZEROS

439

Proof: By Theorem 26.1 f is not constant on B (a, R) because it has a zero of order m. Therefore, using
this theorem again, there exists > 0 such that B (a, 2) B (a, R) and there are no solutions to the equation
f (z) = 0 for z B (a, 2) except a. Also we may assume is small enough that for 0 < |z a| 2,
f 0 (z) 6= 0. Otherwise, a would be a limit point of a sequence of points, zn , having f 0 (zn ) = 0 which would
imply, by Theorem 26.1 that f 0 = 0 on B (0, R) , contradicting the assumption that f has a zero of order m
and is therefore not constant.
Now pick (t) = a + eit , t [0, 2] . Then
/ f ( ([0, 2])) so there exists > 0 with
B (, ) f ( ([0, 2])) = .

(26.6)

Therefore, B (, ) is contained on one component of C \ f ( ([0, 2])) . Therefore, n (f , ) = n (f , z)

for all z B (, ) . Now consider f restricted to B (a, 2) . For z B (, ) , f 1 (z) must consist of a finite
set of points because f 0 (w) 6= 0 for all w in B (a, 2) \ {a} implying that the zeros of f () z in B (a, 2)
are isolated. Since B (a, 2) is compact, this means there are only finitely many. By Theorem 26.7,
n (f , z) =

p
X

n (, ak )

(26.7)

k=1

where {a1 , , ap } = f 1 (z) . Each point, ak of f 1 (z) is either inside the circle traced out by , yielding
n (, ak ) = 1, or it is outside this circle yielding n (, ak ) = 0 because of (26.6). It follows the sum in (26.7)
reduces to the number of points of f 1 (z) which are contained in B (a, ) . Thus, letting those points in
f 1 (z) which are contained in B (a, ) be denoted by {a1 , , ar }
n (f , ) = n (f , z) = r.
We need to verify that r = m. We do this by computing n (f , ) . However, this is easy to compute by
Theorem 26.6 which states
n (f , ) =

m
X

n (, a) = m.

k=1

Therefore, r = m. Each of these ak is a zero of order 1 of the function f () z because f 0 (ak ) 6= 0. This
proves the theorem.
This is a very fascinating result partly because it implies that for values of f near a value, , at which
f () has a root of order m for m > 1, the inverse image of these values includes at least m points, not
just one. Thus the topological properties of the inverse image changes radically. This theorem also shows
that f (B (a, )) B (, ) .
Theorem 26.9 (open mapping theorem) Let U be a region and f : U C be analytic. Then f (U ) is either
a point of a region. If f is one to one, then f 1 : f (U ) U is analytic.
Proof: If f is not constant, then for every f (U ) , it follows from Theorem 26.1 that f ()
has a zero of order m < and so from Theorem 26.8 for each a U there exist , > 0 such that
f (B (a, )) B (, ) which clearly implies that f maps open sets to open sets. Therefore, f (U ) is open,
connected because f is continuous. If f is one to one, Theorem 26.8 implies that for every f (U ) the zero
of f () is of order 1. Otherwise, that theorem implies that for z near , there are m points which f maps
to z contradicting the assumption that f is one to one. Therefore, f 0 (z) 6= 0 and since f 1 is continuous,
due to f being an open map, it follows we may write

0
f 1 (f (z))

This proves the theorem.

f 1 (f (z1 )) f 1 (f (z))
f (z1 ) f (z)
f (z1 )f (z)
z1 z
1
= lim
= 0
.
z1 z f (z1 ) f (z)
f (z)
=

lim

440

THE OPEN MAPPING THEOREM

26.5

Exercises

1. Use Theorem 26.6 to give an alternate proof of the fundamental theorem of algebra. Hint: Take a
R
0
(z)
contour of the form r = reit where t [0, 2] . Consider pp(z)
dz and consider the limit as r .
r

2. Prove the following version of the maximum modulus theorem. Let f : U C be analytic where U is
a region. Suppose there exists a U such that |f (a)| |f (z)| for all z U. Then f is a constant.
3. Let M be an n n matrix. Recall that the eigenvalues of M are given by the zeros of the polynomial,
pM (z) = det (M zI) where I is the n n identity. Formulate a theorem which describes how the
eigenvalues depend on small changes in M. Hint: You could define a norm on the space of n n
1/2
matrices as ||M || tr (M M )
where M is the conjugate transpose of M. Thus
1/2

X
2
||M || =
|Mjk | .
j,k

Argue that small changes will produce small changes in pM (z) . Then apply Theorem 26.6 using k a
very small circle surrounding zk , the kth eigenvalue.
4. Suppose that two analytic functions defined on a region are equal on some set, S which contains a
limit point. (Recall p is a limit point of S if every open set which contains p, also contains infinitely
many points of S. ) Show the two functions coincide. We defined ez ex (cos y + i sin y) earlier and
we showed that ez , defined this way was analytic on C. Is there any other way to define ez on all of C
such that the function coincides with ex on the real axis?
5. We know various identities for real valued functions. For example cosh2 x sinh2 x = 1. If we define
z
z
z
z
cosh z e +e
and sinh z e e
, does it follow that
2
2
cosh2 z sinh2 z = 1
for all z C? What about
sin (z + w) = sin z cos w + cos z sin w?
Can you verify these sorts of identities just from your knowledge about what happens for real arguments?
6. Was it necessary that U be a region in Theorem 26.1? Would the same conclusion hold if U were only
assumed to be an open set? Why? What about the open mapping theorem? Would it hold if U were
not a region?
7. Let f : U C be analytic and one to one. Show that f 0 (z) 6= 0 for all z U. Does this hold for a
function of a real variable?
8. We say a real valued function, u is subharmonic if uxx + uyy 0. Show that if u is subharmonic on a
bounded region, (open connected set) U, and continuous on U and u m on U, then u m on U.
Hint: If not, u achieves its maximum at (x0 , y0 ) U. Let u (x0 , y0 ) > m + where > 0. Now consider
u (x, y) = x2 + u (x, y) where is small enough that 0 < x2 < for all (x, y) U. Show that u also
achieves its maximum at some point of U and that therefore, uxx + uyy 0 at that point implying
that uxx + uyy , a contradiction.
9. If u is harmonic on some region, U, show that u coincides locally with the real part of an analytic
function and that therefore, u has infinitely many derivatives on U. Hint: Consider the case where
0 U. You can always reduce to this case by a suitable translation. Now let B (0, r) U and use the
Schwarz formula to obtain an analytic function whose real part coincides with u on B (0, r) . Then
use Problem 8.

26.5. EXERCISES

441

10. Show the solution to the Dirichlet problem of Problem 8 in the section on the Cauchy integral formula
for a disk is unique. You need to formulate this precisely and then prove uniqueness.

442

THE OPEN MAPPING THEOREM

Singularities
27.1

The Laurent series

In this chapter we consider the functions which are analytic in some open set except at isolated points. The
fundamental formula in this subject which is used to classify isolated singularities is the Laurent series.
Definition 27.1 We define ann (a, R1 , R2 ) {z : R1 < |z a| < R2 } .
Thus ann (a, 0, R) would denote the punctured ball, B (a, R) \ {0} . We now consider an important lemma
which will be used in what follows.
it
Lemma
R 27.2 Let g be analytic on ann (a, R1 , R2 ) . Then if r (t) a+re for t [0, 2] and r (R1 , R2 ) ,
then g (z) dz is independent of r.
r

i(2t)
Proof: Let R1 < r1 < r2 < R2 and denote by r (t) the curve, r (t) a

+ re

for t [0, 2] .
25.7
to
conclude
n

,
z
+
n

,
z
= 0. Also if
Then if z B (a, R1 ), we can apply Proposition
r1
r2



z
/ B (a, R2 ) , then by Corollary 25.11 we have n rj , z = 0 for j = 1, 2. Therefore, we can apply Theorem
25.8 and conclude that for all z ann (a, R1 , R2 ) \ 2j=1 rj ([0, 2]) ,





0 n r2 , z + n r1 , z =
1
2i

Z
r2

1
g (w) (w z)
dw
wz
2i

Z
r1

g (w) (w z)
dw
wz

which proves the desired result.

With this preparation we are ready to discuss the Laurent series.
Theorem 27.3 Let f be analytic on ann (a, R1 , R2 ) . Then there exist numbers, an C such that for all
z ann (a, R1 , R2 ) ,
f (z) =

an (z a) ,

(27.1)

n=

where the series converges absolutely and uniformly on ann (a, r1 , r2 ) whenever R1 < r1 < r2 < R2 . Also
Z
1
f (w)
dw
(27.2)
an =
2i (w a)n+1
where (t) = a + reit , t [0, 2] for any r (R1 , R2 ) . Furthermore the series is unique in the sense that if
(27.1) holds for z ann (a, R1 , R2 ) , then we obtain (27.2).
443

444

SINGULARITIES

Proof: Let R1 < r1 < r2 < R2 and define 1 (t) a + (r1 ) eit and 2 (t) a + (r2 + ) eit for
t [0, 2] and chosen small enough that R1 < r1 < r2 + < R2 .

2
1
A

A

Then by Proposition 25.7 and Corollary 25.11, we see that

n ( 1 , z) + n ( 2 , z) = 0
off ann (a, R1 , R2 ) and that on ann (a, r1 , r2 ) ,
n ( 1 , z) + n ( 2 , z) = 1.
Therefore, by Theorem 25.8,
f (z)

"Z

1
2i

Z
1
f (w)
h
2i
1 (z a) 1

f (w)
dw +
wz

1
=
2i
1
2i

Z
2

Z
1

Z
2

wa
za

f (w)
dw
wz
i dw +

Z
2

f (w)
h
(w a) 1

za
wa

i dw

n

f (w) X z a
dw +
w a n=0 w a
n

f (w) X w a
dw.
(z a) n=0 z a

(27.3)

From the formula (27.3), it follows

n that for z ann (a, r1 , r2 ), the terms in the first sum are bounded

nby an
r2
r1
expression of the form C r2 +
while those in the second are bounded by one of the form C r1
and
so by the Weierstrass M test, the convergence is uniform and so we may interchange the integrals and the
sums in the above formula and rename the variable of summation to obtain
!
Z

X
1
f (w)
n
f (z) =
n+1 dw (z a) +
2i
(w

a)

2
n=0
1
X

n=

1
2i

f (w)
n+1

(w a)

(z a) .

(27.4)

By Lemma 27.2, we may write this as

f (z) =

n=

1
2i

Z
r

f (w)

n+1 dw

(w a)

(z a) .

27.1. THE LAURENT SERIES

445

where r (R1P
, R2 ) is arbitrary.

n
If f (z) = n= an (z a) on ann (a, R1 , R2 ) let
fn (z)

n
X

ak (z a)

(27.5)

k=n

and verify from a repeat of the above argument that

!
Z

X
1
fn (w)
k
dw (z a) .
fn (z) =
2i r (w a)k+1
k=

(27.6)

Therefore, using (27.5) directly, we see

1
2i

fn (w)

k+1

(w a)

dw = ak

for each k [n, n] . However,

1
2i

Z
r

fn (w)

1
dw =
k+1
2i
(w a)

because if l > n or l < n, then it is easy to verify that

Z
l
al (w a)

k+1

(w a)

f (w)

k+1

(w a)

dw

dw = 0

for all k [n, n] . Therefore,

ak =

1
2i

f (w)
k+1

(w a)

dw

and so this establishes uniqueness. This proves the theorem.

Definition 27.4 We say f has an isolated singularity at a C if there exists R > 0 such that f is analytic
on ann (a, 0, R) . Such an isolated singularity is said to be a pole of order m if am 6= 0 but ak = 0 for all
k < m. The singularity is said to be removable if an = 0 for all n < 0, and it is said to be essential if am 6= 0
for infinitely many m < 0.
Note that thanks to the Laurent series, the possibilities enumerated in the above definition are the only
ones possible. Also observe that a is removable if and only if f (z) = g (z) for some g analytic near a. How
can we recognize a removable singularity or a pole without computing the Laurent series? This is the content
of the next theorem.
Theorem 27.5 Let a be an isolated singularity of f . Then a is removable if and only if
lim (z a) f (z) = 0

(27.7)

lim |f (z)| = .

(27.8)

za

and a is a pole if and only if

za

The pole is of order m if

m+1

lim (z a)

za

f (z) = 0

but
m

lim (z a) f (z) 6= 0.

za

446

SINGULARITIES

Proof: First suppose a is a removable singularity. Then it is clear that (27.7) holds since am = 0 for all
m < 0. Now suppose that (27.7) holds and f is analytic on ann (a, 0, R). Then define

(z a) f (z) if z 6= a
h (z)
0 if z = a
We verify that h is analytic near a by using Moreras Rtheorem. Let T be a triangle in B (a, R) . If T does
not contain the point, a, then Corollary 25.11 implies T h (z) dz = 0. Therefore, we may assume a T. If
a is a vertex, then, denoting by b and c the other two vertices, we pick p and q, points on the sides, ab and
ac respectively which are close to a. Then by Corollary 25.11,
Z
h (z) dz = 0.
(q,c,b,p,q)

But
by continuity of h, it follows that Ras p and q are moved closer to a the above integral converges to
R
h
(z) dz, showing that in this case, T h (z) dz = 0 also. It only remains to consider the case where a
T
is not a vertex but is in T. In this case we subdivide the triangle T into either 3 or 2 subtriangles having a
as one vertex, depending on whether a is in the interior or on an edge. Then, applying the above result to
these triangles and noting that the integrals
R over the interior edges cancel out due to the integration being
taken in opposite directions, we see that T h (z) dz = 0 in this case also.
Now we know h is analytic. Since h equals zero at a, we can conclude that
h (z) = (z a) g (z)
where g (z) is analytic in B (a, R) . Therefore, for all z 6= a,
(z a) g (z) = (z a) f (z)
showing that f (z) = g (z) for all z 6= a and g is analytic on B (0, R) . This proves the converse.
It is clear that if f has a pole at a, then (27.8) holds. Suppose conversely that (27.8) holds. Then we
know from the first part of this theorem that 1/f (z) has a removable singularity at a. Also, if g (z) = 1/f (z)
for z near a, then g (a) = 0. Therefore, for z 6= a,
m

1/f (z) = (z a) h (z)

for some analytic function, h (z) for which h (a) 6= 0. It follows that 1/h r is analytic near a with r (a) 6= 0.
Therefore, for z near a,
m

f (z) = (z a)

ak (z a) , a0 6= 0,

k=0

showing that f has a pole of order m. This proves the theorem.

Note that this is very different than
 whatoccurs for functions of a real variable. Consider for example,
1/2
1/2
1/2
the function, f (x) = x
. We see x |x|
0 but clearly |x|
cannot equal a differentiable function
near 0.
What about rational functions, those which are a quotient of two polynomials? It seems reasonable to
suppose, since every finite partial sum of the Laurent series is a rational function just as every finite sum of
a power series is a polynomial, it might be the case that something interesting can be said about rational
functions in the context of Laurent series. In fact we will show the existence of the partial fraction expansion
for rational functions. First we need the following simple lemma.
Lemma 27.6 If f is a rational function which has no poles in C then f is a polynomial.

27.1. THE LAURENT SERIES

447

Proof: We can write

l

f (z) =

p0 (z b1 ) 1 (z bn ) n
r
r ,
(z a1 ) 1 (z am ) m

where we can assume the fraction has been reduced to lowest terms. Thus none of the bj equal any of the
ak . But then, by Theorem 27.5 we would have poles at each ak . Therefore, the denominator must reduce to
1 and so f is a polynomial.
Theorem 27.7 Let f (z) be a rational function,
l

f (z) =

p0 (z b1 ) 1 (z bn ) n
r
r ,
(z a1 ) 1 (z am ) m

(27.9)

where the expression is in lowest terms. Then there exist numbers, bkj and a polynomial, p (z) , such that
f (z) =

rl
m X
X
l=1 j=1

blj
j

(z al )

+ p (z) .

(27.10)

Proof: We see that f has a pole at a1 and it is clear this pole must be of order r1 since otherwise
we could not achieve equality between (27.9) and the Laurent series for f near a1 due to different rates of
growth. Therefore, for z ann (a1 , 0, R1 )
f (z) =

r1
X

b1j

+ p1 (z)

(z a1 )

j=1

where p1 is analytic in B (a1 , R1 ) . Then define

f1 (z) f (z)

r1
X
j=1

b1j
j

(z a1 )

so that f1 is a rational function coinciding with p1 near a1 which has no pole at a1 . We see that f1 has a pole
at a2 or order r2 by the same reasoning. Therefore, we may subtract off the principle part of the Laurent
series for f1 near a2 like we just did for f. This yields
f (z) =

r1
X
j=1

b1j
j

(z a1 )

r2
X
j=1

b2j
j

(z a2 )

+ p2 (z) .

Letting

r1
X
f (z)
j=1

b1j
j

(z a1 )

r2
X
j=1

b2j
j

(z a2 )

= f2 (z) ,

and continuing in this way we finally obtain

f (z)

rl
m X
X
l=1 j=1

blj
j

(z al )

= fm (z)

where fm is a rational function which has no poles. Therefore, it must be a polynomial. This proves the
theorem.

448

SINGULARITIES

How does this relate to the usual partial fractions routine of calculus? Recall in that case we had to
consider irreducible quadratics and all the constants were real. In the case from calculus, since the coefficients
of the polynomials were real, the roots of the denominator occurred in conjugate pairs. Thus we would have
paired terms like
b

(z a)

(z a)

occurring in the sum. We leave it to the reader to verify this version of partial fractions does reduce to the
version from calculus.
We have considered the case of a removable singularity or a pole and proved theorems about this case.
What about the case where the singularity is essential? We give an interesting theorem about this case next.
Theorem 27.8 (Casorati Weierstrass) If f has an essential singularity at a then for all r > 0,
f (ann (a, 0, r)) = C
Proof: If not there exists c C and r > 0 such that c
/ f (ann (a, 0, r)). Therefore,there exists > 0
such that B (c, ) f (ann (a, 0, r)) = . It follows that
1

lim |z a|

za
1

and so by Theorem 27.5 z (z a)

|f (z) c| =

(f (z) c) has a pole at a. It follows that for m the order of the pole,
1

(z a)

m
X

(f (z) c) =

ak
k

+ g (z)

k=1

(z a)

k1

+ g (z) (z a) ,

where g is analytic near a. Therefore,

f (z) c =

m
X

ak

k=1 (z a)

showing that f has a pole at a rather than an essential singularity. This proves the theorem.
This theorem is much weaker than the best result known, the Picard theorem which we state next. A
proof of this famous theorem may be found in Conway [6].
Theorem 27.9 If f is an analytic function having an essential singularity at z, then in every open set
containing z the function f, assumes each complex number, with one possible exception, an infinite number
of times.

27.2

Exercises

1. Classify the singular points of the following functions according to whether they are poles or essential
singularities. If poles, determine the order of the pole.
(a)

cos z
z2

(b)

z 3 +1
z(z1)

(c) cos

1
z

2. Suppose f is defined on an open set, U, and it is known that f is analytic on U \ {z0 } but continuous
at z0 . Show that f is actually analytic on U.

27.2. EXERCISES

449

3. A function defined on C has finitely many poles and lim|z| f (z) exists. Show f is a rational function.
Hint: First show that if h has only one pole at 0 and if lim|z| h (z) exists, then h is a rational
function. Now consider
Qm
r
(z z ) k
Qm r k f (z)
h (z) k=1
k
k=1 z
where zk is a pole of order rk .

450

SINGULARITIES

Residues and evaluation of integrals

It turns out that the theory presented above about singularities and the Laurent series is very useful in
computing the exact value of many hard integrals. First we define what we mean by a residue.

Definition 28.1 Let a be an isolated singularity of f. Thus

f (z) =

an (z a)

n=

for all z near a. Then we define the residue of f at a by

Res (f, a) = a1 .
Now suppose that U is an open set and f : U \ {a1 , , am } C is analytic where the ak are isolated
singularities of f.
'



H
H
1

a1

2
a2

&

Let be a simple closed continuous, and bounded variation curve enclosing these isolated singularities such that ([a, b]) U and {a1 , , am } D U, where D is the bounded component (inside) of
C \ ([a, b]) . Also assume n (, z) = 1 for all z D. As explained earlier, this would occur if (t) traces out
the curve in the counter clockwise direction. Choose r small enough that B (aj , r) B (ak , r) = whenever
j 6= k, B (ak , r) U for all k, and define
k (t) ak + re(2t)i , t [0, 2] .
Thus n ( k , ai ) = 1 and if z is in the unbounded component of C\ ([a, b]) , n (, z) = 0 and n ( k , z) = 0.
If z
/ U \ {a1 , , am } , then z either equals one of the ak or else z is in the unbounded component just
451

452

RESIDUES AND EVALUATION OF INTEGRALS

described. Either way,

Pm

k=1

n ( k , z) + n (, z) = 0. Therefore, by Theorem 25.8, if z

/ D,

Z
Z
m
X
1
(w z)
1
(w z)
f (w)
dw +
f (w)
dw =
2i
(w

z)
2i
(w z)
j

j=1
Z
Z
m
X
1
1
f (w) dw +
f (w) dw =
2i j
2i
j=1
!
m
X
n ( k , z) + n (, z) f (z) (z z) = 0.
k=1

and so, taking r small enough,

1
2i

f (w) dw

Z
m
X
1
=
f (w) dw
2i j
j=1
Z
m

1 X X k
l
=
al
(w ak ) dw
2i

k
k=1 l=
Z
m
X
1
1
=
ak1
(w ak ) dw
2i
k
k=1

m
X

k=1

ak1 =

m
X

Res (f, ak ) .

k=1

Now we give some examples of hard integrals which can be evaluated by using this idea. This will be
done by integrating over various closed curves having bounded variation.
Example 28.2 The first example we consider is the following integral.
Z
1
dx
1
+
x4

One could imagine evaluating this integral by the method of partial fractions and it should work out by
that method. However, we will consider the evaluation of this integral by the method of residues instead.
To do so, consider the following picture.
y

Let r (t) = reit , t [0, ] and let r (t) = t : t [r, r] . Thus r parameterizes the top curve and r
parameterizes the straight line from r to r along the x axis. Denoting by r the closed curve traced out

453
by these two, we see from simple estimates that
Z
lim
r

1
dz = 0.
1 + z4

This follows from the following estimate.

Z



1
1


dz
r.


 r 1 + z 4  r4 1

Therefore,

1
dx = lim
r
1 + x4

1
We compute r 1+z
4 dz using the method of residues.
points, z where 1 + z 4 = 0. These points are

Z
r

1
dz.
1 + z4

The only residues of the integrand are located at

1
1
1
2 i 2, z =
2
2
2
2
1
1
1
2 + i 2, z =
2+
2
2
2

1
i 2,
2
1
i 2
2

and it is only the last two which are found in the inside of r . Therefore, we need to calculate the residues
at these points. Clearly this function has a pole of order one at each of these points and so we may calculate
the residue at in this list by evaluating
lim (z )

1
1 + z4

Thus

1
1
Res f,
2+ i 2
2
2



1
1
1
lim z
2+ i 2

1
1
2
2
1
+
z4
z 2 2+ 2 i 2


Similarly we may find the other residue in the same way



1
1
Res f,
2+ i 2
2
2




1
1
1
lim
z
2+ i 2

1
1
2
2
1
+
z4
z 2 2+ 2 i 2

=
=

1
1
2 i 2
8
8

=
1
1
= i 2+
2.
8
8

Therefore,



1
1
1
1
1
1
dz
=
2i

i
2
+
2
+

i
2
=
2.
4
8
8
8
8
2
r 1 + z

R 1
Thus, taking the limit we obtain 12 2 = 1+x
4 dx.
Obviously many different variations of this are possible. The main idea being that the integral over the
semicircle converges to zero as r . Sometimes one must be fairly creative to determine the sort of curve
to integrate over as well as the sort of function in the integrand and even the interpretation of the integral
which results.
Z

454

RESIDUES AND EVALUATION OF INTEGRALS

Example 28.3 This example illustrates the comment about the integral.
Z
sin x
dx
x
0
Rr
By this integral we mean limr 0 sinx x dx. The function is not absolutely integrable so the meaning of
the integral is in terms of the limit just described. To do this integral, we note the integrand is even and so
it suffices to find
Z R ix
e
lim
dx
R R x
called the Cauchy principle value, take the imaginary part to get
lim

sin x
dx
x

and then divide by two. In order to do so, we let R > r and consider the curve which goes along the x
axis from (R, 0) to (r, 0), from (r, 0) to (r, 0) along the semicircle in the upper half plane, from (r, 0)
to (R, 0) along the x axis, and finally from (R, 0) to (R, 0) along the semicircle in the upper half plane as
shown in the following picture.
y

iz

On the inside of this curve, the function, ez has no singularities and so it has no residues. Pick R large
and let r 0 + . The integral along the small semicircle is
Z

it

ere rieit
dt =
reit

it
e(re ) dt.

and this clearly converges to as r 0. Now we consider the top integral. For z = Reit ,
it

eiRe = eR sin t cos (R cos t) + ieR sin t sin (R cos t)

and so


 iReit 
e
 eR sin t .

Therefore, along the top semicircle we get the absolute value of the integral along the top is,
Z
 Z


iReit 

e
dt
eR sin t dt

0

455

eR sin dt +

R sin

eR sin t dt +

eR sin t dt

whenever is small enough. Letting be this small, it follows that

Z



it
lim 
eiRe dt
R
0

and since is arbitrary, this shows the integral over the top semicircle converges to 0. Therefore, for some
function e (r) which converges to zero as r 0,
Z R ix
Z r ix
Z
e
e
eiz
dz +
dx +
dx
e (r) =
z
x
r
R x
top semicircle
Letting r 0, we see
=

Z
top semicircle

eiz
dz +
z

eix
dx
x

and so, taking R ,

= lim

eix
dx = 2 lim
R
x

Z
0

sin x
,
x

sin x
x dx

showing that = 0
with the above interpretation of the integral.
Sometimes we dont blow up the curves and take limits. Sometimes the problem of interest reduces
directly to a complex integral over a closed curve. Here is an example of this.
Example 28.4 The integral is
Z
0

cos
d
2 + cos

This integrand is even and so we may write it as

Z
1 cos
d.
2 2 + cos


For z on the unit circle, z = ei , z = z1 and therefore, cos = 12 z + z1 . Thus dz = iei d and so d = dz
iz .
Note that we are proceeding formally in order to get a complex integral which reduces to the one of interest.
It follows that a complex integral which reduces to the one we want is


Z
Z
1
1
1
1
z2 + 1
2 z + z
dz
=
dz
2i 2 + 12 z + z1 z
2i z (4z + z 2 + 1)


where is the unit circle. Now the integrand has poles of order 1 at those points where z 4z + z 2 + 1 = 0.
These points are

0, 2 + 3, 2 3.
Only the first two are inside the unit circle. It is also clear the function has simple poles at these points.
Therefore,


z2 + 1
Res (f, 0) = lim z
= 1.
z0
z (4z + z 2 + 1)

456

RESIDUES AND EVALUATION OF INTEGRALS



Res f, 2 + 3 =

lim

z2+ 3



z 2 + 3

z2 + 1
2
=

3.
z (4z + z 2 + 1)
3

It follows
Z
0

z2 + 1
dz
2
z (4z + z + 1)


1
2
=
2i 1
3
2i
3



2
= 1
3 .
3

cos
d
2 + cos

1
2i

Other rational functions of the trig functions will work out by this method also.
Sometimes we have to be clever about which version of an analytic function that reduces to a real function
we should use. The following is such an example.
Example 28.5 The integral here is
Z
0

ln x
dx.
1 + x4

We would like to use the same curve we used in the integral involving sinx x but this will create problems
with the log since the usual version of the log is not defined on the negative real axis. This does not need
to concern us however. We simply use another branch of the logarithm. We leave out the ray from 0 along
the negative y axis and use Theorem 26.4 to define L (z) on this set. Thus L (z) = ln |z| + i arg1 (z) where
i
arg1 (z) will be the angle, , between 2 and 3
2 such that z = |z| e . Now the only singularities contained
in this curve are
1
1
1
1
2 + i 2,
2+ i 2
2
2
2
2
and the integrand, f has simple poles at these points. Thus using the same procedure as in the other
examples,


1
1
Res f,
2+ i 2 =
2
2
1
1
2 i 2
32
32
and
1
1
Res f,
2+ i 2
2
2


3
3
2 + i 2.
32
32
We need to consider the integral along the small semicircle of radius r. This reduces to
Z 0


ln |r| + it
rieit dt
4
it
1 + (re )

457
which clearly converges to zero as r 0 because r ln r 0. Therefore, taking the limit as r 0,
Z

L (z)
dz + lim
r0+
1 + z4

large semicircle

lim

r0+

Observing that

Z
r

ln t
dt = 2i
1 + t4

L(z)
dz
large semicircle 1+z 4

e (R) + 2 lim

r0+

ln (t) + i
dt +
1 + t4


3
3
1
1
2 + i 2 +
2 i 2 .
32
32
32
32

0 as R , we may write
R

Z
r

ln t
dt + i
1 + t4

1
dt =
1 + t4

1 1
+ i 2 2
8 4

where e (R) 0 as R . From an earlier example this becomes

e (R) + 2 lim

r0+

ln t
dt + i
1 + t4

1 1
+ i 2 2.
8 4

Now letting r 0+ and R , we see

Z
0

1 1
=
+ i 2 2 i
8 4
1 2
=
2 ,
8


ln t
dt
1 + t4

and so

Z
0

ln t
1 2
dt =
2 ,
4
1+t
16

which is probably not the first thing you would thing of. You might try to imagine how this could be obtained
using elementary techniques.

Example 28.6 The Fresnel integrals are

Z
0

cos x dx,

sin x2 dx.

To evaluate these integrals we will consider f (z) =eiz on the curve which goes from the origin to the

along a circle of radius r, and from there back

point r on the x axis and from this point to the point r 1+i
2
to the origin as illustrated in the following picture.

458

RESIDUES AND EVALUATION OF INTEGRALS

Thus the curve we integrate over is shaped like a slice of pie. Denote by r the curved part. Since f is
analytic,

Z
Z r
Z r   1+i 2 
2
2
1+i
i t 2

dt
0 =
eiz dz +
eix dx
e
2
r
0
0


Z
Z r
Z r
1+i
iz 2
ix2
t2

dt
=
e dz +
e dx
e
2
r
0
0


Z
Z r
2
2
1+i

=
eiz dz +
eix dx
+ e (r)
2
2
r
0

R
2
where e (r) 0 as r . Here we used the fact that 0 et dt = 2 . Now we need to examine the first
of these integrals.

Z

Z


 4

2



i(reit )
iz 2
it 
e dz  = 
e
rie dt

 r

 0

Z 4
2
r
er sin 2t dt
0

r (3/2)

r
2

1
r

du +
2
2
1u

er u

du
1 u2

Z
0

1 u2

e(r

1/2

which converges to zero as r . Therefore, taking the limit as r ,

 Z

2
1+i

=
eix dx
2
2
0
and so we can now find the Fresnel integrals

Z
Z

2
sin x dx = =
cos x2 dx.
2 2
0
0
The next example illustrates the technique of integrating around a branch point.
Example 28.7

R
0

xp1
1+x dx,

p (0, 1) .

459
Since the exponent of x in the numerator is larger than 1. The integral does converge. However, the
techniques of real analysis dont tell us what it converges to. The contour we will use is as follows: From
(, 0) to (r, 0) along the x axis and then from (r, 0) to (r, 0) counter clockwise along the circle of radius r,
then from (r, 0) to (, 0) along the x axis and from (, 0) to (, 0) , clockwise along the circle of radius . You
should draw a picture of this contour. The interesting thing about this is that we cannot define z p1 all the
way around 0. Therefore, we use a branch of z p1 corresponding to the branch of the logarithm obtained by
deleting the positive x axis. Thus
z p1 = e(ln|z|+iA(z))(p1)
where z = |z| eiA(z) and A (z) (0, 2) . Along the integral which goes in the positive direction on the x
axis, we will let A (z) = 0 while on the one which goes in the negative direction, we take A (z) = 2. This is
the appropriate choice obtained by replacing the line from (, 0) to (r, 0) with two lines having a small gap
and then taking a limit as the gap closes. We leave it as an exercise to verify that the two integrals taken
along the circles of radius and r converge to 0 as 0 and as r . Therefore, taking the limit,
Z 0 p1 
Z p1

x
x
dx +
e2i(p1) dx = 2i Res (f, 1) .
1+x
1+x
0
Calculating the residue of the integrand at 1, and simplifying the above expression, we obtain

 Z xp1
1 e2i(p1)
dx = 2ie(p1)i .
1+x
0
Upon simplification we see that
Z

xp1

dx =
.
1+x
sin p

The following example is one of the most interesting. By an auspicious choice of the contour it is possible
to obtain a very interesting formula for cot z known as the Mittag Leffler expansion of cot z.
Example 28.8 We let N be the contour which goes from N 12 N i horizontally to N + 12 N i and from
there, vertically to N + 12 + N i and then horizontally to N 12 + N i and finally vertically to N 12 N i.
Thus the contour is a large rectangle and the direction of integration is in the counter clockwise direction.
We will look at the following integral.
Z
cos z
IN
dz
sin
z
(2 z 2 )
N
where R is not an integer. This will be used to verify the formula of Mittag Leffler,

X
1
2
cot
+
=
.
2
2
2

n=1

(28.1)

We leave it as an exercise to verify that cot z is bounded on this contour and that therefore, IN 0 as
N . Now we compute the residues of the integrand at and at n where |n| < N + 12 for n an integer.
These are the only singularities of the integrand in this contour and therefore, we can evaluate IN by using
these. We leave it as an exercise to calculate these residues and find that the residue at is
cos
2 sin
while the residue at n is
2

1
.
n2

460

RESIDUES AND EVALUATION OF INTEGRALS

Therefore,
0 = lim IN = lim 2i
N

"

N
X

n=N

cot
1

2 n2

which establishes the following formula of Mittag Leffler.

lim

N
X

n=N

1
cot
=
.
2
n

Writing this in a slightly nicer form, we obtain (28.1).

28.1

The argument principle and Rouches theorem

This technique of evaluating integrals by computing the residues also leads to the proof of a theorem referred
to as the argument principle.
Definition 28.9 We say a function defined on U, an open set, is meromorphic if its only singularities are
poles, isolated singularities, a, for which
lim |f (z)| = .

za

Theorem 28.10 (argument principle) Let f be meromorphic in U and let its poles be {p1 , , pm } and its
zeros be {z1 , , zn } . Let zk be a zero of order rk and let pk be a pole of order lk . Let : [a, b] U be
a continuous simple closed curve having bounded variation for which the inside of ([a, b]) contains all the
poles and zeros of f and is contained in U. Also let n (, z) = 1 for all z contained in the inside of ([a, b]) .
Then
Z 0
n
m
X
X
1
f (z)
dz =
rk
lk
2i f (z)
k=1

k=1

Proof: This theorem follows from computing the residues of f 0 /f. It has residues at poles and zeros. See
Problem 4.
With the argument
Pmprinciple, we can prove Rouches
Pntheorem . In the argument principle, we will denote
by Zf the quantity k=1 rk and by Pf the quantity k=1 lk . Thus Zf is the number of zeros of f counted
according to the order of the zero with a similar definition holding for Pf .
Z 0
1
f (z)
dz = Zf Pf
2i f (z)
Theorem 28.11 (Rouches theorem) Let f, g be meromorphic in U and let Zf and Pf denote respectively
the numbers of zeros and poles of f counted according to order. Let Zg and Pg be defined similarly. Let
: [a, b] U be a simple closed continuous curve having bounded variation such that all poles and zeros
of both f and g are inside ([a, b]) . Also let n (, z) = 1 for every z inside ([a, b]) . Also suppose that for
z ([a, b])
|f (z) + g (z)| < |f (z)| + |g (z)| .
Then
Zf Pf = Zg Pg .

28.2. EXERCISES

461

Proof: We see from the hypotheses that









1 + f (z)  < 1 +  f (z) 



g (z)
g (z) 

which shows that for all z ([a, b]) ,

f (z)
C \ [0, ).
g (z)
Letting l denote a branch of the logarithm defined on C \ [0, ), it follows that l
the function,

(f /g)
(f /g)

f (z)
g(z)

is a primitive for

. Therefore, by the argument principle,

0

1
2i

(f /g)
1
dz =
(f /g)
2i

Z 

f0
g0

f
g

dz

= Zf Pf (Zg Pg ) .
This proves the theorem.

28.2

Exercises

1. In Example 28.2 we found the integral of a rational function of a certain sort. The technique used in
(x)
this example typically works for rational functions of the form fg(x)
where deg (g (x)) deg f (x) + 2
provided the rational function has no poles on the real axis. State and prove a theorem based on these
observations.
2. Fill in the missing details of Example 28.8 about IN 0. Note how important it was that the contour
was chosen just right for this to happen. Also verify the claims about the residues.
3. Suppose f has a pole of order m at z = a. Define g (z) by
m

g (z) = (z a) f (z) .
Show
Res (f, a) =

1
g (m1) (a) .
(m 1)!

Hint: Use the Laurent series.

4. Give a proof of Theorem 28.10. Hint: Let p be a pole. Show that near p, a pole of order m,
P
k
m + k=1 bk (z p)
f 0 (z)
=
P
k
f (z)
(z p) + k=2 ck (z p)
Show that Res (f, p) = m. Carry out a similar procedure for the zeros.
5. Use Rouches theorem to prove the fundamental theorem of algebra which says that if p (z) = z n +
an1 z n1 +a1 z + a0 , then p has n zeros in C. Hint: Let q (z) = z n and let be a large circle,
(t) = reit for r sufficiently large.
6. Consider the two polynomials z 5 + 3z 2 1 and z 5 + 3z 2 . Show that on |z| = 1, we have the conditions
for Rouches theorem holding. Now use Rouches theorem to verify that z 5 + 3z 2 1 must have two
zeros in |z| < 1.

462

RESIDUES AND EVALUATION OF INTEGRALS

7. Consider the polynomial, z 11 + 7z 5 + 3z 2 17. Use Rouches theorem to find a bound on the zeros of
this polynomial. In other words, find r such that if z is a zero of the polynomial, |z| < r. Try to make
r fairly small if possible.
8. Verify that

R
0

et dt =

2 .

Hint: Use polar coordinates.

9. Use the contour described in Example 28.2 to compute the exact values of the following improper
integrals.
(a)

(b)

(c)

x
dx
(x2 +4x+13)2
0

x2
dx
(x2 +a2 )2

dx
, a, b
(x2 +a2 )(x2 +b2 )

>0

10. Evaluate the following improper integrals.

(a)

R
0

cos ax
dx
(x2 +b2 )2

(b)

x sin x
dx
(x2 +a2 )2

11. Find the Cauchy principle value of the integral

Z

(x2

sin x
dx
+ 1) (x 1)

defined as
lim

0+

Z

12. Find a formula for the integral

sin x
dx +
(x2 + 1) (x 1)

dx
(1+x2 )n+1

13. Using the contour of Example 28.3 find

1+


sin x
dx .
(x2 + 1) (x 1)

where n is a nonnegative integer.

sin2 x
x2 dx.

14. If m < n for m and n integers, show

Z
0

15. Find

16. Find

28.3

x2m

dx =
2n
1+x
n sin

1
2m+1
2n

.

1
dx.
(1+x4 )2

ln(x)
1+x2 dx

=0

The Poisson formulas and the Hilbert transform

In this section we consider various applications of the above ideas by focussing on the contour, R shown
below, which represents a semicircle of radius R in the right half plane the direction of integration indicated

28.3. THE POISSON FORMULAS AND THE HILBERT TRANSFORM

463

by the arrows.
y

z
x

AA


AA

We will suppose that f is analytic in a region containing the right half plane and use the Cauchy integral
formula to write
Z
Z
f (w)
1
f (w)
1
dw, 0 =
dw,
f (z) =
2i R w z
2i R w + z
the second integral equaling zero because the integrand is analytic as indicated in the picture. Therefore,
multiplying the second integral by and subtracting from the first we obtain


Z
1
w + z w + z
f (z) =
f (w)
dw.
(28.2)
2i R
(w z) (w + z)
We would like to have the integrals over the semicircular part of the contour converge to zero as R .
This requires some sort of growth condition on f. Let
n
h io


M (R) = max f Reit  : t ,
.
2 2
We leave it as an exercise to verify that when
M (R)
= 0 for = 1
R

(28.3)

lim M (R) = 0 for 6= 1,

(28.4)

lim

and
R

then this condition that the integrals over the curved part of R converge to zero is satisfied. We assume
this takes place in what follows. Taking the limit as R


Z
1
i + z i + z
f (z) =
f (i)
d
(28.5)
2
(i z) (i + z)
the negative sign occurring because the direction of integration along the y axis is negative. If = 1 and
z = x + iy, this reduces to
!
Z
1
x
d,
(28.6)
f (z) =
f (i)
2

|z i|
which is called the Poisson formula for a half plane.. If we assume M (R) 0, and take = 1, (28.5)
reduces to
!
Z
y
i
d.
(28.7)
f (i)
2

|z i|

464

RESIDUES AND EVALUATION OF INTEGRALS

Of course we can consider real and imaginary parts of f in these formulas. Let
f (i) = u () + iv () .
From (28.6) we obtain upon taking the real part,
1
u (x, y) =

u ()

x
2

d.

(28.8)

d,

(28.9)

d.

(28.10)

|z i|

Taking real and imaginary parts in (28.7) gives the following.

1
u (x, y) =

|z i|

1
v (x, y) =

v ()

u ()

y
2

|z i|

These are called the conjugate Poisson formulas because knowledge of the imaginary part on the y axis leads
to knowledge of the real part for Re z > 0 while knowledge of the real part on the imaginary axis leads to
knowledge of the real part on Re z > 0.
We obtain the Hilbert transform by formally letting z = iy in the conjugate Poisson formulas and picking
x = 0. Letting u (0, y) = u (y) and v (0, y) = v (y) , we obtain, at least formally


Z
1
1
u (y) =
v ()
d,

y


Z
1
1
v (y) =
u ()
d.

y
Of course there are major problems in writing these integrals due to the integrand possessing a nonintegrable
singularity at y. There is a large theory connected with the meaning of such integrals as these known as the
theory of singular integrals. Here we evaluate these integrals by taking a contour which goes around the
singularity and then taking a limit to obtain a principle value integral.
The case when = 0 in (28.5) yields
Z
1
f (i)
d.
(28.11)
f (z) =
2 (z i)
We will use this formula in considering the problem of finding the inverse Laplace transform.
We say a function, f, defined on (0, ) is of exponential type if
|f (t)| < Aeat
for some constants A and a. For such a function we can define the Laplace transform as follows.
Z
f (t) est dt Lf.
F (s)

(28.12)

(28.13)

We leave it as an exercise to show that this integral makes sense for all Re s > a and that the function so
defined is analytic on Re z > a. Using the estimate, (28.12), we obtain that for Re s > a,


 A 

.
|F (s)| 
(28.14)
s a

28.3. THE POISSON FORMULAS AND THE HILBERT TRANSFORM

465

We will show that if f (t) is given by the formula,

Z
1
e(a+)t f (t)
eit F (i + a + ) d,
2
then Lf = F for all s large enough.
Z
Z


1
L e(a+)t f (t) =
est
eit F (i + a + ) ddt
2 0

Now if
Z

|F (i + a + )| d < ,

(28.15)

we can use Fubinis theorem to interchange the order of integration. Unfortunately, we do not know this.
The best we have is the estimate (28.14). However, this is a very crude estimate and often (28.15) will hold.
Therefore, we shall assume whatever we need in order to continue with the symbol pushing and interchange
the order of integration to obtain with the aid of (28.11) the following:

Z Z


1
L e(a+)t f (t)
=
e(si)t dt F (i + a + ) d
2
0
Z
F (i + a + )
1
d
=
2
s i
= F (s + a + )
for all s > 0. (The reason for fussing with + a + rather than just is so the function, F ( + a + )
will be analytic on Re > , a region containing the right half plane allowing us to use (28.11).) Now with
this information, we may verify that L (f ) (s) = F (s) for all s > a. We just showed
Z
ewt e(a+)t f (t) dt = F (w + a + )
0

whenever Re w > 0. Let s = w + a + . Then L (f ) (s) = F (s) whenever Re s > a + . Since is arbitrary,
this verifies L (f ) (s) = F (s) for all s > a. It follows that if we are given F (s) which is analytic for Re s > a
and we want to find f such that L (f ) = F, we should pick c > a and define
Z
1
ct
e f (t)
eit F (i + c) d.
2
Changing the variable, to let s = i + c, we may write this as
Z c+i
1
f (t) =
est F (s) ds,
2i ci

(28.16)

and we know from the above argument that we can expect this procedure to work if things are not too
pathological. This integral is called the Bromwich integral for the inversion of the Laplace transform. The
function f (t) is the inverse Laplace transform.
s
We illustrate this procedure with a simple example. Suppose F (s) = (s2 +1)
2 . In this case, F is analytic
for Re s > 0. Let c = 1 and integrate over a contour which goes from c iR vertically to c + iR and then
follows a semicircle in the counter clockwise direction back to c iR. Clearly the integrals over the curved
portion of the contour converge to 0 as R . There are two residues of this function, one at i and one at
i. At both of these points the poles are of order two and so we find the residue at i by
!
2
d ets s (s i)
Res (f, i) = lim
2
si ds
(s2 + 1)
=

iteit
4

466

RESIDUES AND EVALUATION OF INTEGRALS

and the residue at i is

2

ets s (s + i)

d
Res (f, i) = lim
si ds

(s2 + 1)

iteit
4

Now evaluating the contour integral and taking R , we find that the integral in (28.16) equals
 it

ite
iteit
2i
+
= it sin t
4
4
and therefore,
f (t) =

1
t sin t.
2

You should verify that this actually works giving L (f ) =

28.4

s
.
(s2 +1)2

Exercises

1. Verify that the integrals over the curved part of R in (28.2) converge to zero when (28.3) and (28.4)
are satisfied.
2. Obtain similar formulas to (28.8) for the imaginary part in the case where = 1 and formulas (28.9)
- (28.10) in the case where = 1. Observe that these formulas give an explicit formula for f (z) if
either the real or the imaginary parts of f are known along the line x = 0.
3. Verify that the formula for the Laplace transform, (28.13) makes sense for all s > a and that F is
analytic for Re z > a.
4. Find inverse Laplace transforms for the functions,
a
a
1
s
s2 +a2 , s2 (s2 +a2 ) , s7 , (s2 +a2 )2 .

5. Consider the analytic function ez . Show it satisfies the necessary conditions in order to apply formula
(28.6). Use this to verify the formulas,
Z
1
x cos
ex cos y =
d,
x2 + (y )2
Z
1
x sin
x
e sin y =
d.
x2 + (y )2
6. The Poisson formula gives
1
u (x, y) =

u (0, )

x
2

x2 + (y )

whenever u is the real part of a function analytic in the right half plane which has a suitable growth
condition. Show that this implies
!
Z
1
x
1=
d.
x2 + (y )2

28.5. INFINITE PRODUCTS

467

7. Now consider an arbitrary continuous function, u () and define

1
u (x, y)

u ()

x
2

x2 + (y )

d.

Verify that for u (x, y) given by this formula,

lim |u (x, y) u (y)| = 0,

x0+

and that u is a harmonic function, uxx + uyy = 0, on x > 0. Therefore, this integral yields a solution
to the Dirichlet problem on the half plane which is to find a harmonic function which assumes given
boundary values.
8. To what extent can we relax the assumption that u () is continuous?

28.5

Infinite products

In this section we give an introduction to the topic of infinite products and apply the theory to the Gamma
function. To begin with we give a definition of what is meant by an infinite product.
Q
Qn
Definition 28.12 n=1 (1 + un ) limn k=1 (1 + uk ) whenever this limit exists. IfQun = un (z) for
n
z H, we say the infinite product converges uniformly on H if the partial products, k=1 (1 + uk (z))
converge uniformly on H.
QN
QN
Lemma 28.13 Let PN k=1 (1 + uk ) and let QN k=1 (1 + |uk |) . Then
!
N
X
QN exp
|uk | , |PN 1| QN 1
k=1

Proof: To verify the first inequality,

QN =

N
Y

(1 + |uk |)

k=1

N
Y

|uk |

= exp

k=1

N
X

|uk | .

k=1

The second claim is obvious if N = 1. Consider N = 2.

|(1 + u1 ) (1 + u2 ) 1| = |u2 + u1 + u1 u2 |
1 + |u1 | + |u2 | + |u1 | |u2 | 1
= (1 + |u1 |) (1 + |u2 |) 1
Continuing this way the desired inequality follows.
The main theorem is the following.
P
Theorem 28.14 Let H C and suppose that n=1 |un (z)| converges uniformly on H. Then
P (z)

(1 + un (z))

n=1

converges uniformly on H. If (n1 , n2 , ) is any permutation of (1, 2, ) , then for all z H,

P (z) =

(1 + unk (z))

k=1

and P has a zero at z0 if and only if un (z0 ) = 1 for some n.

468

RESIDUES AND EVALUATION OF INTEGRALS

Proof: We use Lemma 28.13 to write for m < n, and all z H,


 n
m

Y
Y


(1 + uk (z))
(1 + uk (z))



k=1

k=1

m
 n

Y
 Y





(1 + |uk (z)|) 
(1 + uk (z)) 1



k=1
k=m+1


!

n
 Y

X


exp
|uk (z)| 
(1 + |uk (z)|) 1


k=1
k=m+1
!
!

X
C exp
|uk (z)| 1
k=m+1

C (e 1)
whenever m is large enough. This shows the partial products form a uniformly Cauchy sequence and hence
converge uniformly on H. This verifies the first part of the theorem.
Next we need to verify the part about taking the product in different orders. Suppose then that
(n1 , n2 , ) is a permutation of the list, (1, 2, ) and choose M large enough that for all z H,



M
Y

Y


(1 + uk (z))
(1 + uk (z)) < .



k=1

k=1

Then for all N sufficiently large, {n1 , n2 , , nN } {1, 2, , M } . Then for N this large, we use Lemma
28.13 to obtain

M
N

Y
Y


(1 + uk (z))
(1 + unk (z))



k=1

k=1



 

M
Y

Y



(1 + uk (z)) 1
(1 + unk (z))




k=1
kN,nk >M


 

M
Y

Y



(1 + uk (z)) 
(1 + |unk (z)|) 1




k=1
kN,nk >M



M

Y
Y




(1 + uk (z)) 
(1 + |ul (z)|) 1




k=1
l=M

M
!
!

Y

X


|ul (z)| 1
(1 + uk (z)) exp



l=M
k=1
M

Y



(1 + uk (z)) (exp 1)



k=1


Y



(1 + |uk (z)|) (exp 1)



k=1

whenever M is large enough. Therefore, this shows, using (28.18) that



N

Y

Y


(1 + unk (z))
(1 + uk (z))



k=1

k=1

(28.17)

(28.18)

28.5. INFINITE PRODUCTS

469


N
M
Y

Y


(1 + uk (z)) +
(1 + unk (z))



k=1

k=1

M

Y
Y


(1 + uk (z))
(1 + uk (z))



k=1

k=1



!
Y



(1 + |uk (z)|) + (exp 1)




k=1

which verifies the claim about convergence of the permuted products.

It remains to verify the assertion about the points, z0 , where P (z0 ) = 0. Obviously, if un (z0 ) = 1,
then P (z0 ) = 0. Suppose then that P (z0 ) = 0. Letting nk = k and using (28.17), we may take the limit as
N to obtain


M
Y



(1 + uk (z0 )) =



k=1

Y

Y


(1 + uk (z0 ))
(1 + uk (z0 ))



k=1
k=1

M

Y


(1 + uk (z0 )) (exp 1) .



k=1

QM
If is chosen small enough in this inequality, we see this implies k=1 (1 + uk (z)) = 0 and therefore,
uk (z0 ) = 1 for some k M. This proves the theorem.
Now we present the Weierstrass product formula. This formula tells how to factor analytic functions into
an infinite product. It is a very interesting and useful theorem. First we need to give a definition of the
elementary factors.
Definition 28.15 Let E0 (z) 1 z and for p 1,
z2
zp
Ep (z) (1 z) exp z +
++
2
p


The fundamental factors satisfy an important estimate which is stated next.

Lemma 28.16 For all |z| 1 and p = 0, 1, 2, ,
p+1

|1 Ep (z)| |z|

Proof: If p = 0 this is obvious. Suppose therefore, that p 1.



z2
zp
Ep0 (z) = exp z +
++
+
2
p
z2
zp
(1 z) exp z +
++
2
p


1 + z + + z p1

470

RESIDUES AND EVALUATION OF INTEGRALS

and so, since (1 z) 1 + z + + z p1 = 1 z p ,
Ep0

z2
zp
(z) = z exp z +
++
2
p
p

which shows that Ep0 has a zero of order p at 0. Thus, from the equation just derived,
Ep0 (z) = z p

ak z k

k=0

where each ak 0 and a0 = 1. This last assertion

about the sign of the ak follows easily from differentiating

2
p
the function f (z) = exp z + z2 + + zp and evaluating the derivatives at z = 0. A primitive for Ep0 (z)
P
k+1+p
is of the form k=0 ak zk+p+1 and so integrating from 0 to z along (0, z) we see that
Ep (z) Ep (0) =

Ep (z) 1 =

ak

k=0

= z p+1

z k+p+1
k+p+1

ak

k=0

zk
k+p+1

which shows that (Ep (z) 1) /z p+1 has a removable singularity at z = 0.

Now from the formula for Ep (z) ,


zp
z2
++
1
Ep (z) 1 = (1 z) exp z +
2
p
and so
Ep (1) 1 = 1 =

ak

k=0

1
k+p+1

Since each ak 0, we see that for |z| = 1,

|1 Ep (z)| X
1

ak
= 1.
p+1
|z
|
k+p+1
k=1

Now by the maximum modulus theorem,

p+1

|1 Ep (z)| |z|
for all |z| 1. This proves the lemma.

Theorem 28.17 Let zn be a sequence of nonzero complex numbers which have no limit point in C and
suppose there exist, pn , nonnegative integers such that
1+pn

X
r
<
|zn |
n=1

(28.19)

28.5. INFINITE PRODUCTS

471

for all r R. Then

P (z)

Epn

n=1

z
zn

is analytic on C and has a zero at each point, zn and at no others. If w occurs m times in {zn } , then P has
a zero of order m at w.
Proof: The series


X
 z 1+pn
 
 zn 

n=1

converges uniformly on any compact set because if |z| r, then

 

 z 1+pn   r 1+pn




 zn

|zn |
and so we may apply the Weierstrass M test to obtain the uniform convergence of
Also,

 
 
pn +1


Epn z 1 |z|


zn
|zn |

P 
n=1

z
zn

1+pn

on |z| < r.

by Lemma 28.16 whenever n is large enough because the hypothesis that {zn } has no limit point requires
that limn |zn | = . Therefore, by Theorem 28.14,
 

Y
z
P (z)
Epn
zn
n=1

converges uniformly on compact subsets of C. Letting Pn (z) denote the nth partial product for P (z) , we
have for |z| < r
Z
1
Pn (w)
Pn (z) =
dw
2i r w z
where r (t) reit , t [0, 2] . By the uniform convergence of Pn to P on compact sets, it follows the same
formula holds for P in place of Pn showing that P is analytic in B (0, r) . Since r is arbitrary, we see that P
is analytic on all of C.
Now we ask where the zeros of P are. By Theorem 28.14, the zeros occur at exactly those points, z,
where
 
z
Epn
1 = 1.
zn
 
 
In that theorem Epn zzn 1 plays the role of un (z) . Thus we need Epn zzn = 0 for some n. However,
this occurs exactly when zzn = 1 so the zeros of P are the points {zn } .
If w occurs m times in the sequence, {zn } , we let n1 , , nm be those indices at which w occurs. Then
we choose a permutation of (1, 2, ) which starts with the list (n1 , , nm ) . By Theorem 28.14,

 

Y
z
z m
P (z) =
Epnk
= 1
g (z)
zn k
w
k=1

472

RESIDUES AND EVALUATION OF INTEGRALS

where g is an analytic function which is not equal to zero at w. It follows from this that P has a zero of
order m at w. This proves the theorem.
The next theorem is the Weierstrass factorization theorem which can be used to factor a given function,
f, rather than only deciding convergence questions.
Theorem 28.18 Let f be analytic on C, f (0) 6= 0, and let the zeros of f be {zk } , listed according to order.
(Thus if z is a zero of order m, it will be listed m times in the list, {zk } .) Then there exists an entire
function, g and a sequence of nonnegative integers, pn such that
g(z)

f (z) = e

Epn

n=1

z
zn

(28.20)

Note that eg(z) 6= 0 for any z and this is the interesting thing about this function.
Proof: We know {zn } cannot have a limit point because if there were a limit point of this sequence, it
would follow from Theorem 26.1 that f (z) = 0 for all z, contradicting the hypothesis that f (0) 6= 0. Hence
limn |zn | = and so
1+n1 X
n


X
r
r
=
<
|zn |
|zn |
n=1
n=1
by the root test. Therefore, by Theorem 28.17 we may write
P (z) =

Epn

n=1

z
zn

a function analytic on C by picking pn = n 1 or perhaps some other choice. (We know pn = n 1 works
but we do not know this is the only choice that might work.) Then f /P has only removable singularities in
C and no zeros thanks to Theorem 28.17. Thus, letting h (z) = f (z) /P (z) , we know from Corollary 25.12
that h0 /h has a primitive, ge. Then


heeg

0

=0

and so
h (z) = ea+ib ege(z)
for some constants, a, b. Therefore, letting g (z) = ge (z) + a + ib, we see that h (z) = eg(z) and thus (28.20)
holds. This proves the theorem.
Corollary 28.19 Let f be analytic on C, f has a zero of order m at 0, and let the other zeros of f be {zk } ,
listed according to order. (Thus if z is a zero of order l, it will be listed l times in the list, {zk } .) Then there
exists an entire function, g and a sequence of nonnegative integers, pn such that
m g(z)

f (z) = z e

n=1

Epn

z
zn

Proof: Since f has a zero of order m at 0, it follows from Theorem 26.1 that {zk } cannot have a limit
point in C and so we may apply Theorem 28.18 to the function, f (z) /z m which has a removable singularity
at 0. This proves the corollary.

28.6. EXERCISES

28.6

473

Exercises

1. Show
series.

n=2

1
n2

= 12 . Hint: Take the ln of the partial product and then exploit the telescoping

Q
2. Suppose P (z) = k=1 fk (z) 6= 0 for all z U, an open set, that convergence is uniform on compact
subsets of U, and fk is analytic on U. Show
P 0 (z) =

fk0 (z)

k=1

fn (z) .

n6=k

Hint: Use a branch of the logarithm, defined near P (z) and logarithmic differentiation.
3. Show that sinzz has a removable singularity at z = 0 and so there exists an analytic function, q, defined
on C such that sinzz = q (z) and q (0) = 1. Using the Weierstrass product formula, show that
Y 
z z
q (z) = eg(z)
1
ek
k
kZ,k6=0


Y
z2
= eg(z)
1 2
k
k=1

for some analytic function, g (z) and that we may take g (0) = 0.
4. Use Problem 2 along with Problem 3 to show that


Y
cos z
sin z
z2
g(z) 0

=
e
g
(z)
1

z
z 2
k2
k=1

2zeg(z)




X
1 Y
z2
1

.
n2
k2
n=1
k6=n

Now divide this by q (z) on both sides to show

cot z

X
1
1
= g 0 (z) + 2z
.
2
z
z n2
n=1

Use the Mittag Leffler expansion for the cot z to conclude from this that g 0 (z) = 0 and hence, g (z) = 0
so that


Y
sin z
z2
=
1 2 .
z
k
k=1

5. In the formula for the product expansion of sinzz , let z = 12 to obtain a formula for
formula. Is this formula you have come up with a good way to calculate ?

called Walliss

6. This and the next collection of problems are dealing with the gamma function. Show that
 C (z)

z  z


e n 1
 1+
n
n2
and therefore,


X
z  z


e n 1 <
 1+
n
n=1
with the convergence uniform on compact sets.

474

RESIDUES AND EVALUATION OF INTEGRALS

z
Q
7. Show n=1 1 + nz e n converges to an analytic function on C which has zeros only at the negative
integers and that therefore,

Y

1+

n=1

z 1 z
en
n

is a meromorphic function (Analytic except for poles) having simple poles at the negative integers.
8. Show there exists such that if

ez Y 
z 1 z
(z)
1+
en ,
z n=1
n

then (1) = 1. Hint:

n=1

(1 + n) e1/n = c = e .

9. Now show that

= lim

"

n
X
1
ln n
k

k=1

 P 

P 
Hint: Show = n=1 ln 1 + n1 n1 = n=1 ln (1 + n) ln n n1 .
10. Justify the following argument leading to Gausss formula
 ! z
n 
Y
z
k
e
ek
(z) = lim
n
k+z
z
k=1

 z
P
1
n!
e
z( n
)
k=1
k
= lim
e
n (1 + z) (2 + z) (n + z)
z
Pn
Pn
1
n!
= lim
ez( k=1 k ) ez[ k=1
n (1 + z) (2 + z) (n + z)
n!nz
= lim
.
n (1 + z) (2 + z) (n + z)

1
k ln n

11. Verify from the Gauss formula above that (z + 1) = (z) z and that for n a nonnegative integer,
(n + 1) = n!.
12. The usual definition of the gamma function for positive x is
Z
1 (x)
et tx1 dt.
0

Show 1

t n

et for t [0, n] . Then show

n
Z n
t
n!nx
1
tx1 dt =
.
n
x (x + 1) (x + n)
0

Use the first part and the dominated convergence theorem or heuristics if you have not studied this
theorem to conclude that
n!nx
= (x) .
n x (x + 1) (x + n)

1 (x) = lim
Hint: To show 1
u [0, 1].

t n
n

et for t [0, n] , verify this is equivalent to showing (1 u) enu for

28.6. EXERCISES

475

R
13. Show (z) = 0 et tz1 dt. whenever Re z > 0. Hint: You have already shown that this is true for
positive real numbers. Verify this formula for Re z yields an analytic function.




14. Show 12 = . Then find 52 .

476

RESIDUES AND EVALUATION OF INTEGRALS

The Riemann mapping theorem

We know from the open mapping theorem that analytic functions map regions to other regions or else to
single points. In this chapter we prove the remarkable Riemann mapping theorem which states that for every
simply connected region, U there exists an analytic function, f such that f (U ) = B (0, 1) and in addition to
this, f is one to one. The proof involves several ideas which have been developed up to now. We also need
the following important theorem, a case of Montels theorem.
Theorem 29.1 Let U be an open set in C and let F denote a set of analytic functions mapping U to

B (0, M ) . Then there exists a sequence of functions from F, {fn }n=1 and an analytic function, f such that
(k)
fn converges uniformly to f (k) on every compact subset of U.
Proof: First we note there exists a sequence of compact sets, Kn such that Kn int Kn+1 U for
all n where here int K denotes the interior of the set K, the union
in K and
 of all open sets

contained

C

n1 works for Kn .
n=1 Kn = U. We leave it as an exercise to verify that B (0, n) z U : dist z, U
Then there exist positive numbers, n such that if z Kn , then B (z, n ) int Kn+1 . Now denote by Fn the
set of restrictions of functions of F to Kn . Then let z Kn and let (t) z + n eit , t [0, 2] . It follows
that for z1 B (z, n ) , and f F,


 
Z
 1

1
1

|f (z) f (z1 )| = 
f (w)

dw
2i
wz
w z1

Z

z z1
1 

f
(w)
dw


2
(w z) (w z1 ) 
Letting |z1 z| <

n
2 ,

we can estimate this and write

|f (z) f (z1 )|

M
|z z1 |
2 n 2
2
n /2

2M

|z z1 |
.
n

It follows that Fn is equicontinuous and uniformly bounded so by the Arzela Ascoli theorem there exists a

sequence, {fnk }k=1 F which converges uniformly on Kn . Let {f1k }k=1 converge uniformly on K1 . Then

use the Arzela Ascoli theorem applied to this sequence to get a subsequence, denoted by {f2k }k=1 which

also converges uniformly on K2 . Continue in this way to obtain {fnk }k=1 which converges uniformly on

K1 , , Kn . Now the sequence {fnn }n=m is a subsequence of {fmk }

k=1 and so it converges uniformly on
Km for all m. Denoting fnn by fn for short, this is the sequence of functions promised by the theorem. It is

clear {fn }n=1 converges uniformly on every compact subset of U because every such set is contained in Km
for all m large enough. Let f (z) be the point to which fn (z) converges. Then f is a continuous function
defined on U. We need to verify f is analytic. But, letting T U,
Z
Z
f (z) dz = lim
fn (z) dz = 0.
T

477

478

THE RIEMANN MAPPING THEOREM

Therefore, by Moreras theorem we see that f is analytic. As for the uniform convergence of the derivatives
of f, this follows from the Cauchy integral formula. For z Kn , and (t) z + n eit , t [0, 2] ,
Z

1  fk (w) f (w) 
0
0
|f (z) fk (z)|
dw

2

2 
(w z)
1
1
2 n 2
2
n
1
= ||fk f || ,
n
||fk f ||

where here ||fk f || sup {|fk (z) f (z)| : z Kn } . Thus we get uniform convergence of the derivatives.
The consideration of the other derivatives is similar.
Since the family, F satisfies the conclusion of Theorem 29.1 it is known as a normal family of functions.
The following result is about a certain class of so called fractional linear transformations,
Lemma 29.2 For B (0, 1) , let
(z)

z
.
1 z

Then maps B (0, 1) one to one and onto B (0, 1), 1

= , and
0 () =

2.

1 ||

Proof: First we show (z) B (0, 1) whenever z B (0, 1) . If this is not so, there exists z B (0, 1)
such that
2

|z | |1 z| .
However, this requires
2

|z| + || > 1 + || |z|

and so
2

|z|

1 ||

> 1 ||

contradicting |z| < 1.

1
It remains
it is easy to verify

to verify is one to one and onto with the given formula for . But
(w) = w. Therefore, is onto and one to one. To verify the formula for 0 , just differentiate the
function. Thus,
2

0 (z) = (z ) (1) (1 z)

() + (1 z)

and the formula for the derivative follows.

The next lemma, known as Schwarzs lemma is interesting for its own sake but will be an important part
of the proof of the Riemann mapping theorem.
Lemma 29.3 Suppose F : B (0, 1) B (0, 1) , F is analytic, and F (0) = 0. Then for all z B (0, 1) ,
|F (z)| |z| ,

(29.1)

|F 0 (0)| 1.

(29.2)

and

If equality holds in (29.2) then there exists C with || = 1 and

F (z) = z.

(29.3)

479
Proof: We know F (z) = zG (z) where G is analytic. Then letting |z| < r < 1, the maximum modulus
theorem implies



F reit 
1
|G (z)| sup
.
r
r
Therefore, letting r 1 we get
|G (z)| 1

(29.4)

It follows that (29.1) holds. Since F 0 (0) = G (0) , (29.4) implies (29.2). If equality holds in (29.2), then from
the maximum modulus theorem, we see that G achieves its maximum at an interior point and is consequently
equal to a constant, , || = 1. Thus F (z) = z which shows (29.3). This proves the lemma.
Definition 29.4 We say a region, U has the square root property if whenever f, f1 : U C are both analytic,
it follows there exists : U C such that is analytic and f (z) = 2 (z) .
The next theorem will turn out to be equivalent to the Riemann mapping theorem.
Theorem 29.5 Let U 6= C for U a region and suppose U has the square root property. Then there exists
h : U B (0, 1) such that h is one to one, onto, and analytic.
Proof: We define F to be the set of functions, f such that f : U B (0, 1) is one to one and analytic.
We will show
 F isnonempty. Then we will show there is a function in F, h, such that for some fixed z0 U,
|h0 (z0 )|  0 (z0 ) for all F. When we have done this, we show h is actually onto. This will prove the
theorem.
Now we begin by showing F is nonempty. Since U 6= C it follows there exists
/ U. Then letting
f (z) z , it follows f and f1 are both analytic on U . Since U has the square root property, there exists
: U C such that 2 (z) = f (z) for all z U. By the open mapping theorem, there exists a such that for
some r < |a| ,
B (a, r) (U ) .
It follows that if z U, then (z)
/ B (a, r) because if this were to occur for some z1 U, then (z1 )
B (a, r) and so there exists z2 B (a, r) such that
(z1 ) = (z2 ) .
Squaring both sides, it follows that z1 = z2 and so z1 = z2 . Therefore, we would have (z2 ) = 0 and
so 0 B (a, r) contrary to the construction in which r < |a| . Now let
(z)

r
.
(z) + a

is well defined because we just verified the denominator is nonzero. It also follows that | (z)| 1 because
if not,
r > | (z) + a|
for some z U, contradicting what was just shown about (U ) B (a, r) = . Therefore, we have shown
that F =
6 .
For z0 U fixed, let



sup  0 (z0 ) : F .

480

THE RIEMANN MAPPING THEOREM

Thus > 0 because 0 (z0 ) 6= 0 for defined above. By Theorem 29.1, there exists a sequence, { n }, of
functions in F and an analytic function, h, such that
 0

 n (z0 ) (z0 )
(29.5)

and

n h, 0n h0 ,

(29.6)

uniformly on all compact sets of U. It follows



|h0 (z0 )| = lim  0n (z0 ) =
n

and for all z U,

|h (z)| = lim | n (z)| 1.
n

We need to verify that h is one to one. Suppose h (z1 ) = and z2 U. We must verify that h (z2 ) 6= . We
choose r > 0 such that h has no zeros on B (z2 , r), B (z2 , r) U, and
B (z2 , r) B (z1 , r) = .
We can do this because, the zeros of h are isolated since h is not constant due to the fact that h0 (z0 ) =
6= 0. Let n (z1 ) = n . Thus n n has a zero at z1 and since n is one to one, it has no zeros in B (z2 , r).
Thus by Theorem 26.6, the theorem on counting zeros, for (t) z2 + reit , t [0, 2] ,
Z
1
0n (w)
0 = lim
dw
n 2i n (w) n
Z
1
h0 (w)
=
dw,
2i h (w)
which shows that h has no zeros in B (z2 , r) . This shows that h is one to one since z2 6= z1 was arbitrary.
Therefore, h F. This completes the second step of the proof. It only remains to verify that h is onto.
To show h is onto, we use the fractional linear transformation of Lemma 29.2. Suppose h is not onto.
Then there exists B (0, 1) \ h (U ) . Then 0
/ h because
/ h (U ) . Therefore, since U has the square
root property, there exists g, an analytic function defined on U such that
g 2 = h.
The function g is one to one because if g (z1 ) = g (z2 ) , then we could square both sides and conclude that
h (z1 ) = h (z2 )
and since and h are one to one, this shows z1 = z2 . It follows that g F also. Now let g(z0 ) g. Thus
(z0 ) = 0. If we define s (w) w2 , then using Lemma 29.2, in particular, the description of 1
= , we
obtain
g = g(z0 )
and therefore,
h (z)

= g 2 (z)


=
s g(z0 ) (z)
=

(F ) (z)

29.1. EXERCISES

481





Now F (0) = 1
2
(0)
= 1
g 2 (z0 ) = h (z0 ) .

g(z0 )
There are two cases to consider. First suppose that h (z0 ) 6= 0. Then define
G h(z0 ) F.
Then G : B (0, 1) B (0, 1) and G (0) = 0. Therefore by the Schwarz lemma, Lemma 29.3,


!


1


0
F
(0)
|G0 (0)| = 
1
 1 |h (z0 )|2


which implies |F 0 (0)| < 1. In the case where h (z0 ) = 0, we note that because of the function, s, in the
definition of F, F is not one to one and so we cannot have F (z) = z for some || = 1. Therefore, by the
Schwarz lemma applied to F, we see |F 0 (0)| < 1. Therefore,


= |h0 (z0 )| = |F 0 ( (z0 ))|  0 (z0 )




= |F 0 (0)|  0 (z0 ) <  0 (z0 ) ,

contradicting the definition of . Therefore, h must be onto and this proves the theorem.
We now give a simple lemma which will yield the usual form of the Riemann mapping theorem.

Lemma 29.6 Let U be a simply connected region with U 6= C. Then U has the square root property.
0

both be analytic on U. Then ff is analytic on U so by Corollary 25.12, there exists


0
0
e
e
e
Fe, analytic on U such that Fe0 = ff on U. Then f eF = 0 and so f (z) = CeF = ea+ib eF . Now let
1
F = Fe + a + ib. Then F is still a primitive of f 0 /f and we have f (z) = eF (z) . Now let (z) e 2 F (z) . Then
is the desired square root and so U has the square root property.
Proof: Let f and

1
f

Corollary 29.7 (Riemann mapping theorem) Let U be a simply connected region with U 6= C and let a U .
Then there exists a function, f : U B (0, 1) such that f is one to one, analytic, and onto with f (a) = 0.
Furthermore, f 1 is also analytic.
Proof: From Theorem 29.5 and Lemma 29.6 there exists a function, g : U B (0, 1) which is one to
one, onto, and analytic. We need to show that there exists a function, f, which does what g does but in
addition, f (a) = 0. We can do so by letting f = g(a) g if g (a) 6= 0. The assertion that f 1 is analytic
follows from the open mapping theorem.

29.1

Exercises

1. Prove that in Theorem 29.1 it suffices to assume F is uniformly bounded on each compact subset of U.
2. Verify the conclusion of Theorem 29.1 involving the higher order derivatives.
3. What if U = C? Does there exist an analytic function, f mapping U one to one and onto B (0, 1)?
Explain why or why not. Was U 6= C used in the proof of the Riemann mapping theorem?
4. Verify that | (z)| = 1 if |z| = 1. Apply the maximum modulus theorem to conclude that | (z)| 1
for all |z| < 1.
5. Suppose that |f (z)| 1 for |z| = 1 and f () = 0 for || < 1. Show that |f (z)| | (z)| for all
which has a removable singularity at . Show the modulus of
z B (0, 1) . Hint: Consider f (z)(1z)
z
this function is bounded by 1 on |z| = 1. Then apply the maximum modulus theorem.

482

THE RIEMANN MAPPING THEOREM

Approximation of analytic functions

Consider the function, z1 = f (z) for z defined on U B (0, 1) \ {0} . Clearly f is analytic on U. Suppose we


could approximate f uniformly by polynomials
ann 0, 12 , 34 , a compact subset of U. Then, there would
 on
 1 R

1
exist a suitable polynomial p (z) , such that  2i
f (z) p (z) dz  < 10
where here is a circle of radius

R
R
2
1
1
3 . However, this is impossible because 2i f (z) dz = 1 while 2i p (z) dz = 0. This shows we cannot
expect to be able to uniformly approximate analytic functions on compact sets using polynomials. It turns
out we will be able to approximate by rational functions. The following lemma is the one of the key results
which will allow us to verify a theorem on approximation. We will use the notation
||f g||K, sup {|f (z) g (z)| : z K}
which describes the manner in which the approximation is measured.
Lemma 30.1 Let R be a rational function which has a pole only at a V, a component of C \ K where K
is a compact set. Suppose b V . Then for > 0 given, there exists a rational function, Q, having a pole
only at b such that
||R Q||K, < .

(30.1)

If it happens that V is unbounded, then there exists a polynomial, P such that

||R P ||K, < .

(30.2)

Proof: We say b V satisfies P if for all > 0 there exists a rational function, Qb , having a pole only
at b such that
||R Qb ||K, .
Now we define a set,
S {b V : b satisfies P } .
We observe that S 6= because a S.
We now show that S is open. Suppose b1 S. Then there exists a > 0 such that


 b1 b  1


 zb < 2

(30.3)



 b1 bn  1
for all z K whenever b B (b1 , ) . If not, there would exist a sequence bn b for which  dist(b
 2.
n ,K)
Then taking the limit and using the fact that dist (bn , K) dist (b, K) > 0, (why?) we obtain a contradiction.
483

484

APPROXIMATION OF ANALYTIC FUNCTIONS

Since b1 satisfies P, there exists a rational function Qb1 with the desired properties. We will show we can
approximate Qb1 with Qb thus yielding an approximation to R by the use of the triangle inequality,
||R Qb1 ||K, + ||Qb1 Qb ||K, ||R Qb ||K, .
Since Qb1 has poles only at b1 , it follows it is a sum of functions of the form
assume Qb1 is of the special form
Qb1 (z) =

n
(zb1 )n .

Therefore, it suffices to

1
n.
(z b1 )

However,
1
n
(z b1 )

(z b)
=

b1 b
zb

n


k

X
1
b1 b
Ak
.
n
zb
(z b)

(30.4)

k=0

We leave it as an exercise to find Ak and to verify using the Weierstrass M test that this series converges
absolutely and uniformly on K because of the estimate (30.3) which holds for all z K. Therefore, a
suitable partial sum can be made as close as desired to (zb1 1 )n . This shows that b satisfies P whenever b is
close enough to b1 verifying that S is open.
Next we show that S is closed in V. Let bn S and suppose bn b V. Then for all n large enough,
1
dist (b, K) |bn b|
2
and so we obtain the following for all n large enough.


 b bn  1


 z bn  < 2 ,

for all z K. Now a repeat of the above argument in (30.4) with bn playing the role of b1 shows that b S.
Since S is both open and closed in V it follows that, since S 6= , we must have S = V . Otherwise V would
fail to be connected.
Now let b V. Then a repeat of the argument that was just given to verify that S is closed shows that
b satisfies P and proves (30.1).
It remains to consider the case where V is unbounded. Since S = V, pick b V = S large enough that
z  1
 
(30.5)
 <
b
2
for all z K. As before, it suffices to assume that Qb is of the form
Qb (z) =

1
n
(z b)

Then we leave it as an exercise to verify that, thanks to (30.5),

n
 z k
1
(1) X
A
k
n =
bn
b
(z b)

(30.6)

k=0

with the convergence uniform on K. Therefore, we may approximate R uniformly by a polynomial consisting
of a partial sum of the above infinite sum.
The next theorem is interesting for its own sake. It gives the existence, under certain conditions, of a
contour for which the Cauchy integral formula holds.

485
Theorem 30.2 Let K U where K is compact and U is open. Then there exist linear mappings, k :
[0, 1] U \ K such that for all z K,
p

f (z) =

1 X
2i

k=1

f (w)
dw.
wz

(30.7)

Proof: Tile R2 = C with little squares having diameters less than where 0 < dist K, U C (see
m
Problem 3). Now let {Rj }j=1 denote those squares that have nonempty intersection with K. For example,
see the following picture.

@
@
@

@ U

@
@
@
@
@

@
@
@
@K
 4
Let vjk k=1 denote the four vertices of Rj where vj1 is the lower left, vj2 the lower right, vj3 the upper
right and vj4 the upper left. Let kj : [0, 1] U be defined as


kj (t) vjk + t vjk+1 vjk if k < 4,


4j (t) vj4 + t vj1 vj4 if k = 4.
Define
Z

g (w) dw

Rj

4 Z
X

k=1

g (w) dw.

k
j

 p
Thus we integrate over the boundary of the square in the counter clockwise direction. Let j j=1 denote

the curves, kj which

have the property that
kj ([0, 1]) K = .
Pp R
Pm R
Claim: j=1 Rj g (w) dw = j=1 g (w) dw.
j

Proof of the claim: If kj ([0, 1]) K 6= , then for some r 6= j,

lr ([0, 1]) = kj ([0, 1])

but lr = kj (The directions are opposite.). Hence, in the sum on the left, the only possibly nonzero
contributions come from those curves, kj for which kj ([0, 1]) K = and this proves the claim.
Now let z K and suppose z is in the interior of Rs , one of these squares which intersect K. Then by
the Cauchy integral formula,
Z
1
f (w)
f (z) =
dw,
2i Rs w z

486

APPROXIMATION OF ANALYTIC FUNCTIONS

and if j 6= s,
0=

1
2i

Z
Rj

f (w)
dw.
wz

Therefore,
m

f (z) =

1 X
2i j=1
1
2i

Rj

p Z
X

j=1

f (w)
dw
wz

f (w)
dw.
wz

This proves (30.7) in the case where z is in the interior of some Rs . The general case follows from using the
continuity of the functions, f (z) and
p

1 X
z
2i j=1

Z
j

f (w)
dw.
wz

This proves the theorem.

30.1

Runges theorem

With the above preparation we are ready to prove the very remarkable Runge theorem which says that we
can approximate analytic functions on arbitrary compact sets with rational functions which have a certain
nice form. Actually, the theorem we will present first is a variant of Runges theorem because it focuses on
a single compact set.
Theorem 30.3 Let K be a compact subset of an open set, U and let {bj } be a set which consists of one
point from the closure of each bounded component of C \ K. Let f be analytic on U. Then for each > 0,
there exists a rational function, Q whose poles are all contained in the set, {bj } such that
||Q f ||K, < .

(30.8)

Proof: By Theorem 30.2 there are curves, k described there such that for all z K,
p

f (z) =

1 X
2i

k=1

f (w)
dw.
wz

(30.9)

(w)
Defining g (w, z) fwz
for (w, z) pk=1 k ([0, 1]) K, we see that g is uniformly continuous and so there
exists a > 0 such that if ||P|| < , then for all z K,




p X
n
X


1
f
(
(
))
(
(t
)

(t
))
j
i
i1
k
k
k
f (z)
< .

 2
2i

(
)

z
j
k


j=1
k=1

The complicated expression is obtained by replacing each integral in (30.9) with a Riemann sum. Simplifying
the appearance of this, it follows there exists a rational function of the form
R (z) =

M
X

k=1

Ak
wk z

30.1. RUNGES THEOREM

487

where the wk are elements of components of C \ K and Ak are complex numbers such that
||R f ||K, <

.
2

k
Consider the rational function, Rk (z) wA
where wk Vj , one of the components of C \ K, the given
k z
point of Vj being bj or else Vj is unbounded. By Lemma 30.1, there exists a function, Qk which is either a
rational function having its only pole at bj or a polynomial, depending on whether Vj is bounded, such that

||Rk Qk ||K, <

Letting Q (z)

PM

k=1

.
2M

Qk (z) ,
||R Q||K, <

.
2

It follows
||f Q||K, ||f R||K, + ||R Q||K, < .
This proves the theorem.
Runges theorem concerns the case where the given points are contained in C \ U for U an open set
rather than a compact set. Note that here there could be uncountably many components of C \ U because
the components are no longer open sets. An easy example of this phenomenon in one dimension is where
U = [0, 1] \ P for P the Cantor set. Then you can show that R \ U has uncountably many components.
Nevertheless, Runges theorem will follow from Theorem 30.3 with the aid of the following interesting lemma.
Lemma 30.4 Let U be an open set in C. Then there exists a sequence of compact sets, {Kn } such that
U =
k=1 Kn , , Kn int Kn+1 ,

(30.10)

K Kn ,

(30.11)

and for any K U,

b \ Kn contains a component of C
b \ U.
for all n sufficiently large, and every component of C
Proof: Let
Vn {z : |z| > n}

z U
/

1
B z,
n

Thus {z : |z| > n} contains the point, . Now let

b \ Vn = C \ Vn U.
Kn C
We leave it as an exercise to verify that (30.10) and (30.11) hold. It remains to show that every component
b \ Kn contains a component of C
b \ U. Let D be a component of C
b \ Kn Vn .
of C
If
/ D, then D contains no point of {z : |z| > n} because this set is connected and D is a component.


S
(If it did contain a point of this set, it would have to contain the whole set..) Therefore, D
B z, n1
z U
/


and so D contains some point of B z, n1 for some z
/ U. Therefore, since this ball is connected, it follows
D must contain the whole ball and consequently D contains some point of U C . (The point z at the center

488

APPROXIMATION OF ANALYTIC FUNCTIONS

of the ball will do.) Since D contains z

/ U, it must contain the component, Hz , determined by this point.
The reason for this is that
b \U C
b \ Kn
Hz C
b \ Kn . Since it has a point in
and Hz is connected. Therefore, Hz can only have points in one component of C
D, it must therefore, be totally contained in D. This verifies the desired condition in the case where
/ D.
Now suppose that D. We know that
/ U because U is given to be a set in C. Letting H denote
b \ U determined by , it follows from similar reasoning to the above that H D and
the component of C
this proves the lemma.
Theorem 30.5 (Runge) Let U be an open set, and let A be a set which has one point in each bounded
b \ U and let f be analytic on U. Then there exists a sequence of rational functions, {Rn }
component of C
having poles only in A such that Rn converges uniformly to f on compact subsets of U.
b
Proof: Let Kn be the compact sets of Lemma 30.4 where each component of C\K
n contains a component
b
b
of C \ U. It follows each bounded component of C \ Kn contains a point of A. Therefore, by Theorem 30.3
there exists Rn a rational function with poles only in A such that
||Rn f ||Kn , <

1
.
n

It follows, since a given compact set, K is a subset of Kn for all n large enough, that Rn f uniformly on
K. This proves the theorem.
Corollary 30.6 Let U be simply connected and f is analytic on U. Then there exists a sequence of polynomials, {pn } such that pn f uniformly on compact sets of U.
b \ U is connected and so there are no
Proof: By definition of what is meant by simply connected, C
b \ U. Therefore, A = and it follows that Rn in the above theorem must be a
bounded components of C
polynomial since it is rational and has no poles.

30.2

Exercises

1. Let K be any nonempty set in C and define

dist (z, K) inf {|z w| : w K} .
Show that z dist (z, K) is a continuous function.
2. Verify the series in (30.4) converges
absolutely on K and find Ak . Also do the same for (30.6). Hint:
P
1
You know that for |z| < 1, 1z
= k=0 z k . Differentiate both sides as many times as needed to obtain
a formula for Ak . Then apply the Weierstrass M test and the ratio test.
3. In Theorem 30.2 we had a compact set, K contained in an open set U and we used the fact that




dist K, U C inf |z w| : w U C , z K > 0.
Prove this.
4. For U = [0, 1] \ P for P the Cantor set, show that R \ U has uncountably many components. Hint:
Show that the component of R \ U determined by p P, is just the single point, p and then show P is
uncountable.
5. In the proof of Lemma 30.4, verify that (30.10) and (30.11) are satisfied for the given choice of Kn .

The Hausdorff Maximal theorem

The purpose of this appendix is to prove the equivalence between the axiom of choice, the Hausdorff maximal
theorem, and the well-ordering principle. The Hausdorff maximal theorem and the well-ordering principle
are very useful but a little hard to believe; so, it may be surprising that they are equivalent to the axiom of
choice. First we give a proof that the axiom of choice implies the Hausdorff maximal theorem, a remarkable
theorem about partially ordered sets.
We say that a nonempty set is partially ordered if there exists a partial order, , satisfying
xx
and
if x y and y z then x z.
An example of a partially ordered set is the set of all subsets of a given set and . Note that we can not
conclude that any two elements in a partially ordered set are related. In other words, just because x, y are
in the partially ordered set, it does not follow that either x y or y x. We call a subset of a partially
ordered set, C, a chain if x, y C implies that either x y or y x. If either x y or y x we say that x
and y are comparable. A chain is also called a totally ordered set. We say C is a maximal chain if whenever
Ce is a chain containing C, it follows the two chains are equal. In other words C is a maximal chain if there
is no strictly larger chain.
Lemma A.1 Let F be a nonempty partially ordered set with partial order . Then assuming the axiom of
choice, there exists a maximal chain in F.
Proof: Let X be the set of all chains from F. For C X , let
SC = {x F such that C{x} is a chain strictly larger than C}.
If SC = for any C, then C is maximal and we are done. Thus, assume SC 6= for all C X . Let f (C) SC .
(This is where the axiom of choice is being used.) Let
g(C) = C {f (C)}.
Thus g(C) ) C and g(C) \ C ={f (C)} = {a single element of F}. We call a subset T of X a tower if
T,
C T implies g(C) T ,
and if S T is totally ordered with respect to set inclusion, then
S T .
489

490

THE HAUSDORFF MAXIMAL THEOREM

Here S is a chain with respect to set inclusion whose elements are chains.
Note that X is a tower. Let T0 be the intersection of all towers. Thus, T0 is a tower, the smallest tower.
We wish to show that any two sets in T0 are comparable in the sense of set inclusion so that T0 is actually a
chain. Let C0 be a set of T0 which is comparable to every set of T0 . Such sets exist, being an example. Let
B {D T0 : D ) C0 and f (C0 )
/ D} .
The picture represents sets of B. As illustrated in the picture, D is a set of B when D is larger than C0
but fails to be comparable to g (C0 ) . Thus there would be more than one chain ascending from C0 if B =
6 ,
rather like a tree growing upward in more than one direction from a fork in the trunk. We will show this
cant take place for any such C0 by showing B = .

C0

f (C0 )
D

This will be accomplished by showing Te0 T0 \ B is a tower. Since T0 is the smallest tower, this will
require that Te0 = T0 and so B = .
Claim: Te0 is a tower and so B = .
Proof of the claim: It is clear that Te0 because for to be contained in B it would be required to
properly contain C0 which is not possible. Suppose D Te0 . We need to verify g (D) Te0 .
Case 1: f (D) C0 . If D C0 , then since both D and {f (D)} are contained in C0 , it follows g (D) C0
and so g (D)
/ B. On the other hand, if D ) C0 , then since D Te0 , we know f (C0 ) D and so g (D) also
contains f (C0 ) implying g (D)
/ B. These are the only two cases to consider because we are given that C0
is comparable to every set of T0 .
Case 2: f (D)
/ C0 . If D ( C0 then we cant have f (D)
/ C0 because if this were so, g (D ) would not
compare to C0 .

f (C0 )
C0 f (D)

Hence if f (D)
/ C0 , then D C0 . If D = C 0 , then f (D) = f (C0 ) g (D) so g (D)
/ B. Therefore, assume
D ) C0 . Then, since D is in Te0 , f (C0 ) D and so f (C0 ) g (D) . Therefore, g (D) Te0 .
Now suppose S is a totally ordered subset of Te0 with respect to set inclusion. Then if every element of
S is contained in C0 , so is S and so S Te0 . If, on the other hand, some chain from S, C, contains C0
properly, then since C
/ B, f (C0 ) C S showing that S
/ B also. This has proved Te0 is a tower and
e
since T0 is the smallest tower, it follows T0 = T0 . We have shown roughly that no splitting into more than
one ascending chain can occur at any C0 which is comparable to every set of T0 . Next we will show that every
element of T0 has the property that it is comparable to all other elements of T0 . We will do so by showing
that these elements which possess this property form a tower.
Define T1 to be the set of all elements of T0 which are comparable to every element of T0 . (Recall the
elements of T0 are chains from the original partial order.)
Claim: T1 is a tower.
Proof of the claim: It is clear that T1 because is a subset of every set. Suppose C0 T1 . We need
to verify that g (C0 ) T1 . Let D T0 (Thus D C0 or else D ) C0 .)and consider g (C0 ) C0 {f (C0 )} . If
D C0 , then D g (C0 ) so g (C0 ) is comparable to D. If D ) C0 , then D g (C0 ) by what was just shown
(B = ). Hence g (C0 ) is comparable to D. Since D was arbitrary, it follows g (C0 ) is comparable to every
set of T0 . Now suppose S is a chain of elements of T1 and let D be an element of T0 . If every element in

491
the chain, S is contained in D, then S is also contained in D. On the other hand, if some set, C, from S
contains D properly, then S also contains D. Thus S T 1 since it is comparable to every D T0 .
This shows T1 is a tower and proves therefore, that T0 = T1 . Thus every set of T0 compares with every
other set of T0 showing T0 is a chain in addition to being a tower.
Now T0 , g (T0 ) T0 . Hence, because g (T0 ) is an element of T0 , and T0 is a chain of these, it follows
g (T0 ) T0 . Thus
T0 g (T0 ) ) T0 ,
a contradiction. Hence there must exist a maximal chain after all. This proves the lemma.
If X is a nonempty set, we say is an order on X if
x x,
and if x, y X, then
either x y or y x
and
if x y and y z then x z.
We say that is a well order and say that (X, ) is a well-ordered set if every nonempty subset of X has a
smallest element. More precisely, if S 6= and S X then there exists an x S such that x y for all y
S. A familiar example of a well-ordered set is the natural numbers.
Lemma A.2 The Hausdorff maximal principle implies every nonempty set can be well-ordered.
Proof: Let X be a nonempty set and let a X. Then {a} is a well-ordered subset of X. Let
F = {S X : there exists a well order for S}.
Thus F =
6 . We will say that for S1 , S2 F, S1 S2 if S1 S2 and there exists a well order for S2 ,
2 such that
(S2 , 2 ) is well-ordered
and if
y S2 \ S1 then x 2 y for all x S1 ,
and if 1 is the well order of S1 then the two orders are consistent on S1 . Then we observe that is a partial
order on F. By the Hausdorff maximal principle, we let C be a maximal chain in F and let
X C.
We also define an order, , on X as follows. If x, y are elements of X , we pick S C such that x, y are
both in S. Then if S is the order on S, we let x y if and only if x S y. This definition is well defined
because of the definition of the order, . Now let U be any nonempty subset of X . Then S U 6= for
some S C. Because of the definition of , if y S2 \ S1 , Si C, then x y for all x S1 . Thus, if
y X \ S then x y for all x S and so the smallest element of S U exists and is the smallest element
in U . Therefore X is well-ordered. Now suppose there exists z X \ X . Define the following order, 1 ,
on X {z}.
x 1 y if and only if x y whenever x, y X

492

THE HAUSDORFF MAXIMAL THEOREM

x 1 z whenever x X .

Then let
Ce = {S C or X {z}}.
Then Ce is a strictly larger chain than C contradicting maximality of C. Thus X \ X = and this shows X
is well-ordered by . This proves the lemma.
With these two lemmas we can now state the main result.
Theorem A.3 The following are equivalent.
The axiom of choice
The Hausdorff maximal principle
The well-ordering principle.
Proof: It only remains to prove that the well-ordering principle implies the axiom of choice. Let I be a
nonempty set and let Xi be a nonempty set for each i I. Let X = {Xi : i I} and well order X. Let
f (i) be the smallest element of Xi . Then
Y
f
Xi .
iI

A.1

Exercises

1. Zorns lemma states that in a nonempty partially ordered set, if every chain has an upper bound, there
exists a maximal element, x in the partially ordered set. When we say x is maximal, we mean that if
x y, it follows y = x. Show Zorns lemma is equivalent to the Hausdorff maximal theorem.
2. Let X be a vector space. We say Y X is a Hamel basis if every element of X can be written in a
unique way as a finite linear combination of elements in Y . Show that every vector space has a Hamel
basis and that if Y, Y1 are two Hamel bases of X, then there exists a one to one and onto map from
Y to Y1 .
3. Using the Baire category theorem of Chapter 14 show that any Hamel basis of a Banach space is
either finite or uncountable.
4. Consider the vector space of all polynomials defined on [0, 1] . Does there exist a norm, |||| defined on
these polynomials such that with this norm, the vector space of polynomials becomes a Banach space
(complete normed vector space)?

Bibliography
[1] Adams R. Sobolev Spaces, Academic Press, New York, San Francisco, London, 1975.
[2] Apostol T. Calculus Volume II Second editioin, Wiley 1969.
[3] Apostol, T. M. Mathematical Analysis. Addison Wesley Publishing Co., 1974.
[4] Bruckner A. , Bruckner J., and Thomson B., Real Analysis Prentice Hall 1997.
[5] Birkoff G. and MacLane S. A Survey of Modern Algebra, Macmillan 1977.
[6] Conway J. B. Functions of one Complex variable Second edition, Springer Verlag 1978.
[7] Deimling K. Nonlinear Functional Analysis, Springer-Verlag, 1985.
[8] Dontchev A.L. The Graves theorem Revisited, Journal of Convex Analysis, Vol. 3, 1996, No.1, 45-53.
[9] Dunford N. and Schwartz J.T. Linear Operators, Interscience Publishers, a division of John Wiley
and Sons, New York, part 1 1958, part 2 1963, part 3 1971.
[10] Edwards C. Advanced Calculus of Several Variables, Dover 1994.
[11] Evans L.C. and Gariepy, Measure Theory and Fine Properties of Functions, CRC Press, 1992.
[12] Federer H., Geometric Measure Theory, Springer-Verlag, New York, 1969.
[13] Gaskill H. and Narayanaswami P. Elements of Real Analysis, Prentice Hall 1998.
[14] Hardy G. A Course Of Pure Mathematics, Tenth edition Cambridge University Press 1992.
[15] Hewitt E. and Stromberg K. Real and Abstract Analysis, Springer-Verlag, New York, 1965.
[16] Hoffman K. and Kunze R. Linear Algebra Prentice Hall 1971.
[17] Ince, E.L., Ordinary Differential Equations, Dover 1956
[18] Jones F., Lebesgue Integration on Euclidean Space, Jones and Bartlett 1993.
[19] Kuttler K.L. Modern Analysis CRC Press 1998.
[20] Levinson, N. and Redheffer, R. Complex Variables, Holden Day, Inc. 1970
[21] Nobel B. and Daniel J. Applied Linear Algebra, Prentice Hall 1977.
[22] Ray W.O. Real Analysis, Prentice-Hall, 1988.
[23] Rudin W. Principles of Mathematical Analysis, McGraw Hill, 1976.
[24] Rudin W. Real and Complex Analysis, third edition, McGraw-Hill, 1987.
493

494
[25] Rudin W. Functional Analysis, second edition, McGraw-Hill, 1991.
[26] Silverman R. Complex Analysis with Applications, Dover, 1974.
[27] Spiegel, M.R. Complex Variables Schaums outline series, McGraw Hill, 1998.
[28] Wade W.R. An Introduction to Analysis, Prentice Hall, 1995.
[29] Yosida K. Functional Analysis, Springer-Verlag, New York, 1978.

BIBLIOGRAPHY

Index
C 1 functions, 178
F sets, 71
G , 244
G sets, 71
Lp
compactness, 355
L , 220
Lploc , 348

properties, 284
well defined, 283
Brouwer fixed point theorem, 288
Cantor diagonal process, 70
Cantor function, 127
Cantor set, 127
Caratheodory, 97
Caratheodory extension theorem, 136
Caratheodorys criterion, 103
Caratheodorys procedure, 98
Casorati Weierstrass theorem, 448
Cauchy
general Cauchy integral formula, 428
integral formula for disk, 416
Cauchy Riemann equations, 409
Cauchy Schwartz inequality, 23, 169
Cauchy Schwarz inequality, 257
Cauchy sequence, 46
Cayley Hamilton theorem, 39
chain rule, 176
change of variables, 120, 202
Lipschitz maps, 371
change of variables general case, 207
characteristic function, 80
characteristic polynomial, 38
chart, 297
Clarksons inequality, 220
closed graph theorem, 248
closed set, 44
closure of a set, 45
cofactor, 35
compact set, 47
complete measure space, 98
complete metric space, 46
completely separable, 46
complex numbers, 393
conformal maps, 412
conjugate Poisson formulas, 464
connected, 52
connected components, 53
continuous function, 46

Abels theorem, 421

adjugate, 37
Alexander subbasis theorem, 55
algebra of sets, 71
Analytic functions, 409
area formula, 374
argument principle, 460
Ascoli Arzela theorem, 62
at most countable, 13
atlas, 297
axiom of choice, 9, 12
axiom of extension, 9
axiom of specification, 9
axiom of unions, 9
Banach space, 211
Banach Steinhaus theorem, 246
basic open set, 43
basis, 16, 43
Bernstein polynomials, 64
Bessel inequality, 162
Bessels inequality, 262, 275
Borel Cantelli lemma, 79
Borel sets, 71
Borsuk, 286
Borsuk Ulam theorem, 289
bounded continuous linear functions, 244
bounded linear transformations, 172
bounded variation, 155, 399
branches of the logarithm, 436
Bromwich, 465
Brouwer degree, 279
Borsuk theorem, 286
495

496
convergence in measure, 79
convex
set, 258
convex functions, 218
convolution, 216, 236
cosine series
pointwise convergence, 157
countable, 13
counting zeros, 437
Cramers rule, 38
De Moivres theorem, 395
derivatives, 175
determinant, 32
product, 34
transpose, 34
differentiability
monotone functions, 389
differential equations
Peano existence theorem, 69
differential form, 300
derivative, 304
integral, 300
differential forms, 298
dimension of vector space, 17
Dini derivates, 389
Dirichlet Kernel, 149
Dirichlet kernel, 254
distribution, 345
distribution function, 146, 386
divergence, 313
divergence theorem, 313, 385
dominated convergence theorem, 86
dual space, 251
duality maps, 254
Egoroff theorem, 77
eigenvalue, 41
eigenvalues, 440
eigenvector, 41
elementary sets, 131
entire, 420
epsilon net, 49
equality of mixed partial derivatives, 184
equicontinuous, 62
equivalence of norms, 172
ergodic, 94
individual ergodic theorem, 91
exponential growth, 239
extended complex plane, 397
Fatous lemma, 86

INDEX
Fejer kernel, 158
Fejers theorem, 159
finite intersection property, 47
finite measure space, 74
Fourier series, 147, 254
convergence of Cesaro means, 159
mean square convergence, 164
pointwise convergence, 152
Fourier transform, 234
Fourier transform L1 , 237
Fourier transforms, 223
fractional linear transformations, 478
Frechet derivative, 175
Fredholm alternative, 267
Fubinis theorem, 119, 133
function, 11
fundamental theorem of algebra, 420
fundamental theorem of calculus
Lebesgue measure, 363
Gamma function, 219
gauge function, 249
Gausss formula, 474
Gramm Schmidt process, 24
Hahn Banach theorem, 250
Hardy Littlewood maximal function, 361
Hardys inequality, 219
harmonic functions, 411
Hausdorff maximal principle, 55, 249
Hausdorff maximal theorem, 489
Hausdorff metric, 58, 70
Hausdorff space, 44
Heine Borel theorem, 51, 58
higher order derivatives, 182
Hilbert Schmidt theorem, 262
Hilbert space, 22, 257
Hilbert transform, 464
Holders inequality, 209
implicit function theorem, 187
infinite products, 467
inner product, 22
inner product space, 22, 257
inner regular, 106
integral, 81, 84
Invariance of domain, 287
Inverse function theorem, 189
inverse function theorem, 196
inverses and determinants, 37
isogonal, 411

INDEX
James map, 252
Jensens inequality, 219
Jordan Separation theorem, 292
Kolmogorov, 138
Lagrange multipliers, 192
Laplace expansion, 35
Laplace transform, 129, 239, 464
Laplace transforms, 342
Laurent series, 443
Lebesgue
measure, 113
decomposition, 339
set, 363
Lebesgue measure
translation invariance, 115
lim inf, 76
lim sup, 76
limit point, 44
linear transformation, 18
linearly independent, 16
Liouville theorem, 420
Lipschitz maps
extension, 364
locally compact , 47
manifold
Lipschitz, 375
orientable, 375
parametrically defined, 390
manifolds, 297
boundary, 297
interior, 297
orientable, 298
radon measure, 309, 380
smooth, 298
surface mesure, 309, 380
Matrix
lower triangular, 38
upper triangular, 38
matrix
Hermitian, 30
non defective, 28
normal, 28
matrix of linear transformation, 18
maximal function
measurability, 387
maximal function strong estimates, 387
maximum modulus theorem, 436
McShanes lemma, 331
mean ergodic theorem, 342

497
mean square approximation, 162
measurable, 97
measurable function, 75
pointwise limits, 75
measurable functions
Borel, 79
combinations, 77
measurable rectangle, 131
measurable sets, 74, 97
measure space, 74
metric space, 45
Minkowski functional, 254
Minkowskis inequality, 212
minor, 35
Mittag Leffler, 459
monotone class, 72
monotone convergence theorem, 81
Montels theorem, 477
Morreys inequality, 350
multi-index, 223
norm, 22
normal family of functions, 478
normal matrix, 42
normal topological space, 44
nowhere differentiable functions, 253
odd map, 285
open cover, 47
Open mapping theorem, 246
open mapping theorem, 434
open sets, 43
operator norm, 172
orientable manifold, 298
orientation, 300
orthogonal matrix, 31
orthonormal basis, 26
orthonormal set, 261
outer measure, 79, 97
outer regular, 106
parallelogram identity, 31, 274
partial fractions, 447
partial order, 55, 248
partially ordered set, 489
partition of unity, 103
periodic function, 147
permutation
sign of a permutation, 32
Picard theorem, 448
Plancherel theorem, 229
point of density, 386

498
Poisson formula, 463
polar coordinates, 124
polar decomposition, 323
positive linear functional, 101
power series
analytic functions, 419
power set, 43
precompact, 47, 68
principal minor, 38
principal submatrix, 38
principle branch, 413
Product formula, 291
product measure, 133
product measure on infinite product, 139
product measures and regularity, 134
product rule, 350
product topology, 46
projection in Hilbert space, 259
Rademachers theorem, 352, 354
Radon Nikodym Theorem
finite measures, 320
finite measures, 319
random variables, 138
identically distributed, 144
independent, 144
rank of a matrix, 40
reflexive Banach Space, 253
reflexive space
closed subspace of, 276
regular, 106
regular mapping, 304
regular topological space, 44
relative topology, 297
Riemann Lebesgue lemma, 151, 254
Riemann sphere, 397
Riesz representation theorem, 24
C0 (X), 337
Hilbert space, 260
Riesz Representation theorem
C (X), 336
Riesz representation theorem Lp
finite measures, 325
Riesz representation theorem Lp
finite case, 329
non finite measure space, 334
Riesz representation theorem for L1
finite measures, 327
right polar decomposition, 28
Rouches theorem, 460
Runges theorem, 486

INDEX
Russells paradox, 10
Sards lemma, 203
Schroder Bernstein theorem, 12
Schur form of a matrix, 42
Schwartz class, 223
Schwarz formula, 421
Schwarzs lemma, 478
second derivative test, 194
second mean value theorem, 154
self adjoint, 26
separable space, 46
separated, 52
separation theorem, 254
sets, 9
Shannon sampling theorem, 241
sigma algebra, 71
sigma finite, 131
similar matrices, 20
similarity transformation, 20
simple function, 80
singularities, 445
Smtal, 388
Sobolev Space
embedding theorem, 240, 357
equivalent norms, 240
reflexive space, 356
stereographic projection, 397
Stirlings formula, 208, 390
Stokes theorem, 305
Lipschitz manifolds, 307, 379
stokes theorem, 317
Stone Weierstrass, 66
strict convexity, 255
strong ergodic theorem, 143
strong law of large numbers, 144
subbasis, 48
subspace, 15
Taylors formula, 193
tempered distributions, 232
Tietze extention theorem, 70
topological space, 43
total variation, 321
totally bounded set, 49
totally ordered set, 489
triangle inequality, 86
Tychonoff theorem, 56
uniform boundedness theorem, 246
uniform contractions, 186
uniform convexity, 255

INDEX
uniformly convex, 221
uniformly integrable, 88, 220
variational inequality, 259
vector measures, 321
vector space, 15
Vitali convergence theorem, 89, 219
Vitali covering theorem, 360
Vitali coverings, 388
Walliss formula, 473
weak * convergence, 338
weak convergence, 255
weak derivative, 346
Weierstrass, 65
Weierstrass M test, 394
well ordered sets, 491
winding number, 426
Wronskian, 269
Youngs inequality, 340

499