Analisis II - Note's Class

Bruce K.
Driver
Analysis Tools with Examples

August 6, 2004 File:anal.tex
Springer
Berlin Heidelberg NewYork
Hong Kong London
Milan Paris Tokyo
Contents
Part I Background Material
1 Introduction / User Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Set Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3 A Brief Review of Real and Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.1 The Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.1.1 The Decimal Representation of a Real Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 The Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4 Limits and Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.1 Limsups, Liminfs and Extended Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Sums of positive functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Sums of complex functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.4 Iterated sums and the Fubini and Tonelli Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.5.1 Limit Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.5.2 Dominated Convergence Theorem Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 `p – spaces, Minkowski and Holder Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Part II Metric, Banach, and Hilbert Space Basics

4 Contents
6 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.1 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.2 Completeness in Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.3 Supplementary Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.3.1 Word of Caution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.3.2 Riemannian Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7 Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.2 Bounded Linear Operators Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.3 General Sums in Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.4 Inverting Elements in L(X) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
8 Hilbert Space Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

8.1 Hilbert Space Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2 Some Spectral Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.3 Compact Operators on a Hilbert Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.3.1 The Spectral Theorem for Self Adjoint Compact Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
8.4 Supplement 1: Converse of the Parallelogram Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
8.5 Supplement 2. Non-complete inner product spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
9 Hölder Spaces as Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

9.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Part III Calculus and Ordinary Differential Equations in Banach Spaces
10 The Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

10.1 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
10.2 Integral Operators as Examples of Bounded Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
10.3 Linear Ordinary Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
10.4 Classical Weierstrass Approximation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
10.5 Iterated Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
10.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
11 Ordinary Differential Equations in a Banach Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

11.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
11.2 Uniqueness Theorem and Continuous Dependence on Initial Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
11.3 Local Existence (Non-Linear ODE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
11.4 Global Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Page: 4 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

Contents 5
11.5 Semi-Group Properties of time independent flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

11.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
12 Banach Space Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

12.1 The Differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
12.2 Product and Chain Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
12.3 Partial Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
12.4 Higher Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
12.5 Inverse and Implicit Function Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
12.6 Smooth Dependence of ODE’s on Initial Conditions* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
12.7 Existence of Periodic Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
12.8 Contraction Mapping Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.9.1 Alternate construction of g. To be made into an exercise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Part IV Topological Spaces
13 Topological Space Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

13.1 Constructing Topologies and Checking Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
13.2 Product Spaces I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
13.3 Closure operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
13.4 Countability Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
13.5 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
13.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
13.6.1 General Topological Space Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
13.6.2 Connectedness Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
13.6.3 Metric Spaces as Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
14 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
14.1 Metric Space Compactness Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
14.2 Compact Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
14.3 Local and σ – Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
14.4 Function Space Compactness Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
14.5 Tychonoff’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
14.6 Banach – Alaoglu’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
14.6.1 Weak and Strong Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
14.7 Weak Convergence in Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
14.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
14.8.1 Ascoli-Arzela Theorem Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
14.8.2 Tychonoff’s Theorem Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

6 Contents
15 Locally Compact Hausdorff Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

15.1 Locally compact form of Urysohn’s Metrization Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
15.2 Partitions of Unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
15.3 C0 (X) and the Alexanderov Compactification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
15.4 Stone-Weierstrass Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
15.5 *More on Separation Axioms: Normal Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
15.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
16 Baire Category Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

16.1 Metric Space Baire Category Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
16.2 Locally Compact Hausdorff Space Baire Category Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
16.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
Part V Lebesgue Integration Theory
17 Introduction: What are measures and why “measurable” sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

17.1 The problem with Lebesgue “measure” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
18 Measurability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
18.1 Algebras and σ – Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
18.2 Measurable Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
18.2.1 More general pointwise limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
18.3 σ – Function Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
18.4 Product σ – Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
18.4.1 Factoring of Measurable Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
18.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
19 Measures and Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

19.1 Example of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
19.1.1 ADD: Examples of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
19.2 Integrals of Simple functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
19.3 Integrals of positive functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
19.4 Integrals of Complex Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
19.5 Measurability on Complete Measure Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
19.6 Comparison of the Lebesgue and the Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
19.7 Determining Classes of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
19.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
20 Multiple Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

20.1 Fubini-Tonelli’s Theorem and Product Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
20.2 Lebesgue Measure on Rd and the Change of Variables Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
20.3 The Polar Decomposition of Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365

Contents 7
20.4 More proofs of the classical Weierstrass approximation Theorem 10.35 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
20.5 More Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
20.6 Sard’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
20.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
21 Lp -spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
21.1 Jensen’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
21.2 Modes of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
21.3 Completeness of Lp – spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
21.3.1 Summary: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
21.4 Converse of Hölder’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
21.5 Uniform Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
21.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
22 Approximation Theorems and Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415

22.1 Density Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
22.2 Convolution and Young’s Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
22.2.1 Smooth Partitions of Unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
22.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Part VI Further Hilbert and Banach Space Techniques
23 L2 - Hilbert Spaces Techniques and Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439

23.1 L2 -Orthonormal Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
23.2 Hilbert Schmidt Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
23.3 Fourier Series Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
23.3.1 Dirichlet, Fejér and Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
23.3.2 The Dirichlet Problems on D and the Poisson Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
23.4 Weak L2 -Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
23.5 *Conditional Expectation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
23.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
23.7 Fourier Series Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
23.8 Conditional Expectation Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
24 Complex Measures, Radon-Nikodym Theorem and the Dual of Lp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469

24.1 The Radon-Nikodym Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
24.2 The Structure of Signed Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
24.2.1 Hahn Decomposition Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
24.2.2 Jordan Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
24.3 Complex Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
24.4 Absolute Continuity on an Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
24.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489

8 Contents
25 Three Fundamental Principles of Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

25.1 The Hahn-Banach Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491
25.1.1 Hahn – Banach Theorem Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
25.1.2 *Quotient spaces, adjoints, and more reflexivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
25.2 The Open Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
25.3 Uniform Boundedness Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
25.3.1 Applications to Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
25.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
25.4.1 More Examples of Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
25.4.2 Hahn-Banach Theorem Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514
25.4.3 Open Mapping and Closed Operator Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
25.4.4 Weak Topology and Convergence Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
26 Weak and Strong Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519

26.1 Basic Definitions and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
26.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
27 Bochner Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
Part VII Construction and Differentiation of Measures
28 Examples of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541

28.1 Extending Premeasures to Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
28.1.1 Regularity and Density Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
28.2 The Riesz-Markov Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
28.2.1 Regularity Results For Radon Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
28.2.2 The dual of C0 (X) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
28.3 Classifying Radon Measures on R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557
28.3.1 Classifying Radon Measures on R using Theorem 28.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
28.3.2 Classifying Radon Measures on R using the Riesz-Markov Theorem 28.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
28.3.3 The Lebesgue-Stieljtes Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
28.4 Kolmogorov’s Existence of Measure on Products Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
29 Probability Measures on Lusin Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569

29.1 Weak Convergence Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
29.2 Haar Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
29.3 Hausdorff Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
29.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573
29.4.1 The Laws of Large Number Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575

Contents 9
30 Lebesgue Differentiation and the Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577

30.1 A Covering Lemma and Averaging Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 578
30.2 Maximal Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
30.3 Lebesque Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
30.4 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584
30.4.1 Increasing Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
30.4.2 Functions of Bounded Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
30.4.3 Alternative method to Proving Theorem 30.29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
30.5 The connection of Weak and pointwise derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
30.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 606
31 Constructing Measures Via Carathéodory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607

31.1 Construction of Premeasures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 608
31.1.1 Extending Premeasures to Aσ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
31.2 Outer Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 611
31.3 *The σ – Finite Extension Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612
31.4 General Extension and Construction Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 616
31.4.1 Extensions of General Premeasures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
31.5 Proof of the Riesz-Markov Theorem 28.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621
31.6 More Motivation of Carathéodory’s Construction Theorem 31.17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625
32 The Daniell – Stone Construction of Integration and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627

32.0.1 Examples of Daniell Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
32.1 Extending a Daniell Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
32.2 The Structure of L1 (I) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 640
32.3 Relationship to Measure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
32.4 Extensions of premeasures to measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
32.4.1 A Useful Version: BRUCE: delete this if incorporated above. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
32.5 Riesz Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 651
32.6 The General Riesz Representation by Daniell Integrals (Move Later?) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 656
32.7 Regularity Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658
32.8 Metric space regularity results resisted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 664
32.9 General Product Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
32.10Daniel Integral approach to dual spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 668
33 Class Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671

33.1 Monotone Class and π – λ Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
33.1.1 Some other proofs of previously proved theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674
33.2 Regularity of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 675
33.2.1 Another proof of Theorem 28.22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 679
33.2.2 Second Proof of Theorem 22.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 680

10 Contents
Part VIII The Fourier Transform and Generalized Functions
34 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683

34.1 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
34.2 Schwartz Test Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
34.3 Fourier Inversion Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
34.4 Summary of Basic Properties of F and F −1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
34.5 Fourier Transforms of Measures and Bochner’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
34.6 Supplement: Heisenberg Uncertainty Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697
34.6.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699
34.6.2 More Proofs of the Fourier Inversion Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700
35 Constant Coefficient partial differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703

35.1 Elliptic examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704
35.2 Poisson Semi-Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 706
35.3 Heat Equation on Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
35.4 Wave Equation on Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 712
35.5 Elliptic Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 718
35.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723
36 Elementary Generalized Functions / Distribution Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725

36.1 Distributions on U ⊂o Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725
36.2 Examples of distributions and related computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726
36.3 Other classes of test functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734
36.4 Compactly supported distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 739
36.5 Tempered Distributions and the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 742
36.6 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 749
36.7 Appendix: Topology on Cc∞ (U ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754
37 Convolutions involving distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 759

37.1 Tensor Product of Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 759
37.2 Elliptic Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 769
Part IX Appendices
A Multinomial Theorems and Calculus Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

A.1 Multinomial Theorems and Product Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
A.2 Taylor’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
B Zorn’s Lemma and the Hausdorff Maximal Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Contents 11
C Nets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Part I
Background Material
1
Introduction / User Guide
Not written as of yet. Topics to mention.

1. A better and more general integral.
a) Convergence Theorems
b) Integration over diverse collection of sets. (See probability theory.)
c) Integration relative to different weights or densities including singular
weights.
d) Characterization of dual spaces.
e) Completeness.
2. Infinite dimensional Linear algebra.
3. ODE and PDE.
4. Harmonic and Fourier Analysis.
5. Probability Theory
2
Set Operations
Let N denote the positive integers, N0 := N∪ {0} be the non-negative in- A4B := (B \ A) ∪ (A \ B) .
tegers and Z = N0 ∪ (−N) – the positive and negative integers including 0, Q
the rational numbers, R the real numbers (see Chapter 3 below), and C the As usual if {Aα }α∈I is an indexed collection of subsets of X we define the union
complex numbers. We will also use F to stand for either of the fields R or C. and the intersection of this collection by
Notation 2.1 Given two sets X and Y, let Y X denote the collection of all ∪α∈I Aα := {x ∈ X : ∃ α ∈ I 3 x ∈ Aα } and
functions f : X → Y. If X = N, we will say that f ∈ Y N is a sequence ∩α∈I Aα := {x ∈ X : x ∈ Aα ∀ α ∈ I }.
∞
with values in Y and often write fn for f (n) and express f as {fn }n=1 . If
X = {1, 2, . . . , N }, we will write Y N in place of Y {1,2,...,N } and denote f ∈ Y N
`
Notation 2.4 We will also write α∈I Aα for ∪α∈I Aα in the case that
by f = (f1 , f2 , . . . , fN ) where fn = f (n). {Aα }α∈I are pairwise disjoint, i.e. Aα ∩ Aβ = ∅ if α 6= β.
Notation Q 2.2 More generally if {Xα : α ∈ A} is a collection of non-empty sets, Notice that ∪ is closely related to ∃ and ∩ is closely related to ∀. For example
let XA = Xα and πα : XA → Xα be the canonical projection map defined ∞
let {An }n=1 be a sequence of subsets from X and define
α∈A Q
by πα (x) = xα . If If Xα = X for some fixed space X, then we will write Xα
α∈A
{An i.o.} := {x ∈ X : # {n : x ∈ An } = ∞} and
A
as X rather than XA . {An a.a.} := {x ∈ X : x ∈ An for all n sufficiently large}.
Recall that an element x ∈ XA is a “choice function,” i.e. an assignment (One should read {An i.o.} as An infinitely often and {An a.a.} as An almost
xα := x(α) ∈ Xα for each α ∈ A. The axiom of choice (see Appendix B.) always.) Then x ∈ {An i.o.} iff
states that XA 6= ∅ provided that Xα 6= ∅ for each α ∈ A.
∀N ∈ N ∃ n ≥ N 3 x ∈ An
Notation 2.3 Given a set X, let 2X denote the power set of X – the collection
of all subsets of X including the empty set. and this may be expressed as
The reason for writing the power set of X as 2X is that if we think of 2
X {An i.o.} = ∩∞
N =1 ∪n≥N An .
meaning {0, 1} , then an element of a ∈ 2X = {0, 1} is completely determined
by the set Similarly, x ∈ {An a.a.} iff
A := {x ∈ X : a(x) = 1} ⊂ X.
X
In this way elements in {0, 1} are in one to one correspondence with subsets ∃ N ∈ N 3 ∀ n ≥ N, x ∈ An
of X.
which may be written as
For A ∈ 2X let
Ac := X \ A = {x ∈ X : x ∈/ A} {An a.a.} = ∪∞
N =1 ∩n≥N An .
and more generally if A, B ⊂ X let
Definition 2.5. A set X is said to be countable if is empty or there is an
c injective function f : X → N, otherwise X is said to be uncountable.
B \ A := {x ∈ B : x ∈
/ A} = A ∩ B .
We also define the symmetric difference of A and B by Lemma 2.6 (Basic Properties of Countable Sets).
6 2 Set Operations
N
1. If A ⊂ X is a subset of a countable set X then A is countable. 6. Let us begin by showing 2N = {0, 1} is uncountable. For sake of
2. Any infinite subset Λ ⊂ N is in one to one correspondence with N. N
contradiction suppose f : N → {0, 1} is a surjection and write f (n) as
3. A non-empty set X is countable iff there exists a surjective map, g : N → X. N
(f1 (n) , f2 (n) , f3 (n) , . . . ) . Now define a ∈ {0, 1} by an := 1 − fn (n). By
4. If X and Y are countable then X × Y is countable. construction fn (n) 6= an for all n and so a ∈ / f (N) . This contradicts the as-
5. Suppose for each m ∈ N that Am is a countable subset of a set X, then sumption that f is surjective and shows 2N is uncountable. For the general
A = ∪∞ m=1 Am is countable. In short, the countable union of countable sets case, since Y0X ⊂ Y X for any subset Y0 ⊂ Y, if Y0X is uncountable then so
is still countable. is Y X . In this way we may assume Y0 is a two point set which may as well
6. If X is an infinite set and Y is a set with at least two elements, then Y X be Y0 = {0, 1} . Moreover, since X is an infinite set we may find an injective
is uncountable. In particular 2X is uncountable for any infinite set X. map x : N → X and use this to set up an injection, i : 2N → 2X by setting
i (A) := {xn : n ∈ N} ⊂ X for all A ⊂ N. If 2X were countable we could find
Proof. 1. If f : X → N is an injective map then so is the restriction, f |A ,
a surjective map f : 2X → N in which case f ◦ i : 2N → N would be surjec-
of f to the subset A. 2. Let f (1) = min Λ and define f inductively by
tive as well. However this is impossible since we have already seed that 2N is
f (n + 1) = min (Λ \ {f (1), . . . , f (n)}) . uncountable.
We end this section with some notation which will be used frequently in the
Since Λ is infinite the process continues indefinitely. The function f : N → Λ sequel.
defined this way is a bijection.
3. If g : N → X is a surjective map, let Notation 2.7 If f : X → Y is a function and E ⊂ 2Y let
f (x) = min g −1 ({x}) = min {n ∈ N : f (n) = x} . f −1 E := f −1 (E) := {f −1 (E)|E ∈ E}.
Then f : X → N is injective which combined with item If G ⊂ 2X , let

2. (taking Λ = f (X)) shows X is countable. Conversely if f : X → N is f∗ G := {A ∈ 2Y |f −1 (A) ∈ G}.
injective let x0 ∈ X be a fixed point and define g : N → X by g(n) = f −1 (n)
Definition 2.8. Let E ⊂ 2X be a collection of sets, A ⊂ X, iA : A → X be the
for n ∈ f (X) and g(n) = x0 otherwise.
inclusion map (iA (x) = x for all x ∈ A) and
4. Let us first construct a bijection, h, from N to N × N. To do this put the
elements of N × N into an array of the form EA = i−1
A (E) = {A ∩ E : E ∈ E} .
 
(1, 1) (1, 2) (1, 3) . . .
 (2, 1) (2, 2) (2, 3) . . . 

 (3, 1) (3, 2) (3, 3) . . . 
 2.1 Exercises
.. .. .. . .
 
. . . . Let f : X → Y be a function and {Ai }i∈I be an indexed family of subsets of Y,
verify the following assertions.
and then “count” these elements by counting the sets {(i, j) : i + j = k} one
at a time. For example let h (1) = (1, 1) , h(2) = (2, 1), h (3) = (1, 2), h(4) = Exercise 2.1. (∩i∈I Ai )c = ∪i∈I Aci .
(3, 1), h(5) = (2, 2), h(6) = (1, 3) and so on. If f : N →X and g : N →Y are
surjective functions, then the function (f × g) ◦ h : N →X × Y is surjective Exercise 2.2. Suppose that B ⊂ Y, show that B \ (∪i∈I Ai ) = ∩i∈I (B \ Ai ).
where (f × g) (m, n) := (f (m), g(n)) for all (m, n) ∈ N × N.
Exercise 2.3. f −1 (∪i∈I Ai ) = ∪i∈I f −1 (Ai ).
5. If A = ∅ then A is countable by definition so we may assume A 6= ∅.
With out loss of generality we may assume A1 6= ∅ and by replacing Am by Exercise 2.4. f −1 (∩i∈I Ai ) = ∩i∈I f −1 (Ai ).
A1 if necessary we may also assume Am 6= ∅ for all m. For each m ∈ N let
am : N →Am be a surjective function and then define f : N × N → ∪∞ m=1 Am by Exercise 2.5. Find a counterexample which shows that f (C ∩ D) = f (C) ∩
f (m, n) := am (n). The function f is surjective and hence so is the composition, f (D) need not hold.
f ◦ h : N → ∪∞m=1 Am , where h : N → N × N is the bijection defined above.

3
A Brief Review of Real and Complex Numbers
Although it is assumed that the reader of this book is familiar with the 3. If we further assume qn ≤ rn for all n, show limn→∞ qn ≤ limn→∞ rn . (It
properties of the real numbers, R, nevertheless I feel it is instructive to define suffices to consider the case where qn = 0 for all n.)
them here and sketch the development of their basic properties. It will most
certainly be assumed that the reader is familiar with basic algebraic properties The rational numbers Q suffer from the defect that they are not complete,
of the natural numbers N and the ordered field of rational numbers, i.e. not all Cauchy sequences are convergent. In fact, according to Corollary
nm o 3.14 below, “most” Cauchy sequences of rational numbers do not converge to a
Q= : m, n ∈ Z : n 6= 0 . rational number.
n
As usual, for q ∈ Q, we define Exercise 3.3. Use the following outline to construct a Cauchy sequence
∞
{qn }n=1 ⊂ Q which is not convergent in Q.
q if q ≥ 0
|q| =
−q if q ≤ 0. 1. Recall that there is no element q ∈ Q such that q 2 = 2.1 To each n ∈ N let
mn ∈ N be chosen so that
Notice that if q ∈ Q and |q| ≤ n−1 := n1 for all n, then q = 0. Since if q 6= 0,
then |q| = m 1
n for some m, n ∈ N and hence |q| ≥ n . A similar argument shows m2n (mn + 1)
2
1 <2< (3.1)
q ≥ 0 iff q ≥ − n for all n ∈ N. These trivial remarks will be used in the future
n2 n2
without further reference.
∞ and let qn := mnn .
Definition 3.1. A sequence {qn }n=1 ⊂ Q converges to q ∈ Q if |q − qn | → 0 ∞
∞ 2. Verify that qn2 → 2 as n → ∞ and that {qn }n=1 is a Cauchy sequence in Q.
as n → ∞, i.e. if for all N ∈ N, |q − qn | ≤ N1 for a.a. n. As usual if {qn }n=1 ∞
3. Show {qn }n=1 does not have a limit in Q.
converges to q we will write qn → q as n → ∞ or q = limn→∞ qn .
∞
Definition 3.2. A sequence {qn }n=1 ⊂ Q is Cauchy if |qn − qm | → 0 as
m, n → ∞. More precisely we require for each N ∈ N that |qm − qn | ≤ N1 3.1 The Real Numbers
for a.a. pairs (m, n) .
∞
∞
Exercise 3.1. Show that all convergent sequences {qn }n=1 ⊂ Q are Cauchy Let C denote the collection of Cauchy sequences a = {an }n=1 ⊂ Q and say
∞
and that all Cauchy sequences {qn }n=1 are bounded – i.e. there exists M ∈ N a, b ∈ C are equivalent (write a ∼ b) iff limn→∞ |an − bn | = 0. (The reader
such that should check that “ ∼ ” is an equivalence relation.)
|qn | ≤ M for all n ∈ N.
Definition 3.3. A real number is an equivalence class, ā := {b ∈ C : b ∼ a}
∞ ∞
Exercise 3.2. Suppose {qn }n=1 and {rn }n=1 are Cauchy sequences in Q. associated to some element a ∈ C. The collection of real numbers will be denoted
∞ ∞ by R. For q ∈ Q, let i (q) = ā where a is the constant sequence an = q for all
1. Show {qn + rn }n=1 and {qn · rn }n=1 are Cauchy.
∞ ∞ n ∈ N. We will simply write 0 for i (0) and 1 for i (1) .
Now assume that {qn }n=1 and {rn }n=1 are convergent sequences in Q.
∞ ∞
2. Show {qn + rn }n=1 {qn · rn }n=1 are convergent in Q and 1
This fact also shows that the intermediate value theorem, (see Theorem 13.50 be-
low.) fails when working with continuous functions defined over Q.
lim (qn + rn ) = lim qn + lim rn and
n→∞ n→∞ n→∞
lim (qn rn ) = lim qn · lim rn .
n→∞ n→∞ n→∞
8 3 A Brief Review of Real and Complex Numbers
Exercise 3.4. Given ā, b̄ ∈ R show that the definitions By exercise 3.6,
|ā| = ā − b̄ + b̄ ≤ ā − b̄ + b̄
−ā = (−a), ā + b̄ := (a + b) and ā · b̄ := a · b
and hence
∞
are well defined. Here −a, a + b and a · b denote the sequences {−an }n=1 , |ā| − b̄ ≤ ā − b̄
∞ ∞
{an + bn }n=1 and {an · bn }n=1 respectively. Further verify that with these op-
and by reversing the roles of ā and b̄ we also have
erations, R becomes a field and the map i : Q → R is injective homomorphism
of fields. Hint: if ā 6= 0 show that ā may be represented by a sequence a ∈ C
− |ā| − b̄ = b̄ − |ā| ≤ b̄ − ā = ā − b̄ .
with |an | ≥ N1 for all n and some N ∈ N. For this representative show the
∞
sequence a−1 := a−1

n n=1
∈ C. The multiplicative inverse to ā may now be Therefore,
1 −1
−1 ∞ |ā| − b̄ ≤ ā − b̄
constructed as: ā = ā := an n=1 .
∞
and consequently if {ān }n=1 ⊂ R converges to ā ∈ R then
Definition 3.4. Let ā, b̄ ∈ R. Then
1. ā > 0 if there exists an N ∈ N such that an > N1 for a.a. n. ||ān | − |ā|| ≤ |ān − ā| → 0 as n → ∞.
2. ā ≥ 0 iff either ā > 0 or ā = 0. Equivalently (as the reader should verify), Remark 3.6. The field i (Q) is dense in R in the sense that if ā ∈ R there exists
ā ≥ 0 iff for all N ∈ N, an ≥ − N1 for a.a. n. ∞
{qn }n=1 ⊂ Q such that i (qn ) → ā as n → ∞. Indeed, simply let qn = an where
3. Write ā > b̄ or b̄ < ā if ā − b̄ > 0 a represents ā. Since a is a Cauchy sequence, to any N ∈ N there exits M ∈ N
4. Write ā ≥ b̄ or b̄ ≤ ā if ā − b̄ ≥ 0. such that
1 1
Exercise 3.5. Show “ ≥ ” make R into a linearly ordered field and the map − ≤ am − an ≤ for all m, n ≥ M
N N
i : Q → R preserves order. Namely if ā, b̄ ∈ R then
and therefore
1. exactly one of the following relations hold: ā < b̄ or ā > b̄ or ā = b̄.
1 1
2. If ā ≥ 0 and b̄ ≥ 0 then ā + b̄ ≥ 0 and ā · b̄ ≥ 0. −i ≤ i (am ) − ā ≤ i for all m ≥ M.
3. If q, r ∈ Q then q ≤ r iff i (q) ≤ i (r) . N N
The absolute value of a real number ā is defined analogously to that of a This shows
rational number by
1

|i (qm ) − ā| = |i (am ) − ā| ≤ i for all m ≥ M

ā if ā ≥ 0
|ā| = . N
−ā if ā < 0
Observe this definition is consistent with our previous definition of the absolute and since N is arbitrary it follows that i (qm ) → ā as m → ∞.
value on Q, namely i (|q|) = |i (q)| . Also notice that ā = 0 (i.e. a ∼ 0 where 0 ∞
1 Definition 3.7. A sequence {ān }n=1 ⊂ R is Cauchy if |ān − ām | → 0 as
denotes the constant sequence of all zeros)
1
iff for all N ∈ N, |an | ≤ N for a.a. m, n → ∞. More precisely we require for each N ∈ N that |ām − ān | ≤ i N1
n. This is equivalent to saying |ā| ≤ i N for all N ∈ N iff ā = 0.
for a.a. pairs (m, n) .
∞
Definition 3.5. A sequence {ān }n=1 ⊂ R converges to ā ∈ R if |ā − ān | → 0
as n → ∞, i.e. if for all N ∈ N, |ā − ān | ≤ i N1 for a.a. n. As before (for Exercise 3.7. The analogues of the results in Exercises 3.1 and 3.2 hold with
∞
rational numbers) if {ān }n=1 converges to ā we will write ān → ā as n → ∞ Q replaced by R. (We now say a subset Λ ⊂ R is bounded if there exists M ∈ N
or ā = limn→∞ ān . such that |λ| ≤ i (M ) for all λ ∈ Λ.)
Exercise 3.6. Given ā, b̄ ∈ R show For the purposes of real analysis the most important property of R is that
it is “complete.”
āb̄ = |ā| b̄ and ā + b̄ ≤ |ā| + b̄ .
Theorem 3.8. The ordered field R is complete, i.e. all Cauchy sequences in
The latter inequality being referred to as the triangle inequality. R are convergent.

3.1 The Real Numbers 9
∞ ∞
Proof. Suppose that {ā (m)}m=1
is a Cauchy sequence in R. By Remark Proposition 3.11. If {an }n=1
⊂ R is an increasing (decreasing) sequence
∞
3.6, we may choose qm ∈ Q such that which is bounded from above (below), then {an }n=1 is convergent and
|ā (m) − i (qm )| ≤ i m−1 for all m ∈ N.

lim an = sup {an : n ∈ N} lim an = inf {an : n ∈ N} .
n→∞ n→∞
Given N ∈ N, choose M ∈ N such that |ā (m) − ā (n)| ≤ i N −1 for all m, n ≥

If Λ ⊂ R is a set bounded from above then there exists {λn } ⊂ Λ such that
M. Then λn ↑ M := sup Λ, as n → ∞, i.e. {λn } is increasing and limn→∞ λn = M.
|i (qm ) − i (qn )| ≤ |i (qm ) − ā (m)| + |ā (m) − ā (n)| + |ā (n) − i (qn )| Proof. Let M := sup {an : n ∈ N} , then for each N ∈ N there must exist
≤ i m−1 + i n−1 + i N −1 m ∈ N such that M − i N −1 < am ≤ M. Since an is increasing, it follows that

M − i N −1 < an ≤ M for all n ≥ m.

and therefore
|qm − qn | ≤ m−1 + n−1 + N −1 for all m, n ≥ M. From this we conclude that lim an exists and lim an = M. If M = sup Λ, for
each n ∈ N we may choose λn ∈ Λ such that
∞
It now follows that q = {qm }m=1 ∈ C and therefore q represents a point q̄ ∈ R.
M − i n−1 < λn ≤ M.

Using Remark 3.6 and the triangle inequality, (3.2)
By replacing λn by max {λ1 , . . . , λn }2 if necessary we may assume that λn is
|ā (m) − q̄| ≤ |ā (m) − i (qm )| + |i (qm ) − q̄|
increasing in n. It now follows easily from Eq. (3.2) that limn→∞ λn = M.
≤ i m−1 + |i (qm ) − q̄| → 0 as m → ∞

and therefore limm→∞ ā (m) = q̄. 3.1.1 The Decimal Representation of a Real Number
Pm Pm
Definition 3.9. A number M ∈ R is an upper bound for a set Λ ⊂ R if Let α ∈ R or α ∈ Q, m, n ∈ Z and S := k=n αk . If α = 1 then k=n αk =
λ ≤ M for all λ ∈ Λ and a number m ∈ R is an lower bound for a set Λ ⊂ R m − n + 1 while for α 6= 1,
if λ ≥ m for all λ ∈ Λ. Upper and lower bounds need not exist. If Λ has an αS − S = αm+1 − αn
upper (lower) bound, Λ is said to be bounded from above (below).
and solving for S gives the important geometric summation formula,
Theorem 3.10. To each non-empty set Λ ⊂ R which is bounded from above
m
(below) there is a unique least upper bound denoted by sup Λ ∈ R (respectively X αm+1 − αn
greatest lower bound denoted by inf Λ ∈ R). αk = if α 6= 1. (3.3)
α−1
k=n
Proof. Suppose Λ is bounded from above and for each n ∈ N, let mn ∈ Z
Taking α = 10−1 in Eq. (3.3) implies
be the smallest integer such that i mn
2n is an upper bound for Λ. The sequence
mn m
qn := 2n is Cauchy because qm ∈ [qn − 2−n , qn ] ∩ Q for all m ≥ n, i.e. X 10−(m+1) − 10−n 1 1 − 10−(m−n+1)
10−k = =
10−1 − 1 10n−1 9
|qm − qn | ≤ 2− min(m,n) → 0 as m, n → ∞. k=n
and in particular, for all M ≥ n,

Passing to the limit, n → ∞, in the inequality i (qn ) ≥ λ, which is valid for all
λ ∈ Λ implies m
X 1
M
X
q̄ = lim i (qn ) ≥ λ for all λ ∈ Λ. lim 10−k = ≥ 10−k .
n→∞ m→∞ 9 · 10n−1
k=n k=n
Thus q̄ is an upper bound for Λ. If there were another upper bound M ∈ R for Z
Λ such that M < q̄, it would follow that M ≤ i (qn ) < q̄ for some n. But this is Let D denote those sequences α ∈ {0, 1, 2, . . . , 9} with the following prop-
∞ erties:
a contradiction because {qn }n=1 is a decreasing sequence, i (qn ) ≥ i (qm ) for all
m ≥ n and therefore i (qn ) ≥ q̄ for all n. Therefore q̄ is the unique least upper 2
The notation, max Λ, denotes sup Λ along with the assertion that sup Λ ∈ Λ. Sim-
bound for Λ. The existence of lower bounds is proved analogously. ilarly, min Λ = inf Λ along with the assertion that inf Λ ∈ Λ.

10 3 A Brief Review of Real and Complex Numbers
1. there exists N ∈ N such that α−n = 0 for all n ≥ N and Theorem 3.12 (Decimal Representation). The map
2. αn 6= 0 for some n ∈ Z.
D : {±1} × D0 → R\ {0}
Associated to each α ∈ D is the sequence a = a (α) defined by
n is a bijection.
X
an := αk 10−k .
Proof. Suppose D (ε, α) = D (δ, β) for some (ε, α) and (δ, β) in {±1} × D.
k=−∞
Since D (ε, α) > 0 if ε = 1 and D (ε, α) < 0 if ε = −1 it follows that ε = δ. Let
Since for m > n, a = a (α) and b = a (β) be the sequences associated to α and β respectively.
m m Suppose that α 6= β and let j ∈ Z be the position where α and β first disagree,
X
−k
X 1 1 i.e. αn = βn for all n < j while αj 6= βj . For sake of definiteness suppose
|am − an | = αk 10 ≤ 9 10−k ≤ 9 = n,

9 · 10n 10 βj > αj . Then for n > j we have
k=n+1 k=n+1
it follows that n
X
bn − an = (βj − αj ) 10−j + (βk − αk ) 10−k
1
|am − an | ≤ → 0 as m, n → ∞. k=j+1
10min(m,n) n
X 1
Therefore a = a (α) ∈ C and we may define a map D : {±1} × D → R defined ≥ 10−j − 9 10−k ≥ 10−j − 9 = 0.
9 · 10j
by D (ε, α) = εa (α). As is customary we will denote D (ε, α) = εa (α) as k=j+1
ε · αm . . . α0 .α1 α2 . . . αn . . . (3.4) Therefore bn − an ≥ 0 for all n and lim (bn − an ) = 0 iff βj = αj + 1 and βk = 9
and αk = 0 for all k > j. In summary, D (ε, α) = D (δ, β) with α 6= β implies
where m is the largest integer in Z such that αk = 0 for all k < m. If m > 0 either α or β has an infinite tail of nines which shows that D is injective when
the expression in Eq. (3.4) should be interpreted as restricted to {±1} × D0 . To see that D is surjective it suffices to show any b̄ ∈ R
ε · 0.0 . . . 0αm αm+1 . . . . with 0 < b̄ < 1 is in the range of D. For each n ∈ N, let an = .α1 . . . αn with
αi ∈ {0, 1, 2, . . . , 9} such that
An element α ∈ D has a tail of all 9’s starting at N ∈ N if αn = 9 and for all
i (an ) < b̄ ≤ i (an ) + i 10−n .

n ≥ N and αN −1 6= 9. If α has a tail of 9’s starting at N ∈ N, then for n > N, (3.6)
N −1 n
X X Since an+1 = an + αn+1 10−(n+1) for some αn+1 ∈ {0, 1, 2, . . . , 9} , we see that
an (α) = αk 10−k + 9 10−k an+1 = .α1 . . . αn αn+1 , i.e. the first n digits in the decimal expansion of an+1
k=−∞ k=N
are the same as in the decimal expansion of an . Hence this defines αn uniquely
N −1
X 9 1 − 10−(n−N ) for all n ≥ 1. By setting αn = 0 when n ≤ 0, we have constructed from b̄ an
= αk 10−k + · element α ∈ D. Because of Eq. (3.6), D (1, α) = b̄.
10N −1 9
k=−∞
N
X −1 Notation 3.13 From now on we will identify Q with i (Q) ⊂ R and elements
→ αk 10−k + 10−(N −1) as n → ∞. in R with their decimal expansions.
k=−∞
To summarize, we have constructed a complete ordered field R “containing”
PN −1
If α0 is the digits in the decimal expansion of k=−∞ αk 10
−k
+ 10−(N −1) , then Q as a dense subset. Moreover every element in R (modulo those of the form
m10−n for some m ∈ Z and n ∈ N) has a unique decimal expansion.
α0 ∈ D0 := {α ∈ D : α does not have a tail of all 9’s} .
Corollary 3.14. The set (0, 1) := {a ∈ R : 0 < a < 1} is uncountable while
and we have just shown that D (ε, α) = D (ε, α0 ) . In particular this implies Q ∩ (0, 1) is countable.
D ({±1} × D0 ) = D ({±1} × D) . (3.5)

3.2 The Complex Numbers 11
N
Proof. By Theorem 3.12, the set {0, 1, 2 . . . , 8} can be mapped injectively Notice that
into (0, 1) and therefore it follows from Lemma 2.6 that (0, 1) is uncountable. For 1 1
each m ∈ N, let Am := m n
: n ∈ N with n < m . Since Q ∩ (0, 1) = ∪∞ m=1 Am Re z = (z + z̄) and Im z = (z − z̄) . (3.12)
and # (Am ) < ∞ for all m, another application of Lemma 2.6 shows Q ∩ (0, 1) 2 2i
is countable. Proposition 3.18. Complex conjugation and the modulus operators satisfy the
following properties.
3.2 The Complex Numbers 1. z̄ = z,

2. zw = z̄ w̄ and z̄ + w̄ = z + w.
Definition 3.15 (Complex Numbers). Let C = R2 equipped with multipli- 3. |z̄| = |z|
n
cation rule 4. |zw| = |z| |w| and in particular |z n | = |z| for all n ∈ N.
(a, b)(c, d) := (ac − bd, bc + ad) (3.7) 5. |Re z| ≤ |z| and |Im z| ≤ |z|
6. |z + w| ≤ |z| + |w| .
and the usual rule for vector addition. As is standard we will write 0 = (0, 0) ,
7. z = 0 iff |z| = 0.
1 = (1, 0) and i = (0, 1) so that every element z of C may be written as z =
8. If z 6= 0 then z −1 := |z|z̄ 2 (also written as z1 ) is the inverse of z.
(x, y) = x1 + yi which in the future will be written simply as z = x + iy. If −1 n
z = x + iy, let Re z = x and Im z = y. 9. z −1 = |z| and more generally |z n | = |z| for all n ∈ Z.
Writing z = a + ib and w = c + id, the multiplication rule in Eq. (3.7) Proof. All of these properties are direct computations except for possibly
becomes the triangle inequality in item 6 which is verified by the following computation;
(a + ib)(c + id) := (ac − bd) + i(bc + ad) (3.8) 2 2 2
|z + w| = (z + w) (z + w) = |z| + |w| + wz̄ + w̄z
and in particular 12 = 1 and i2 = −1. 2 2
= |z| + |w| + wz̄ + wz̄
Proposition 3.16. The complex numbers C with the above multiplication rule 2 2 2 2
= |z| + |w| + 2 Re (wz̄) ≤ |z| + |w| + 2 |z| |w|
satisfies the usual definitions of a field. For example wz = zw and z (w1 + w2 ) =
2
zw1 + zw2 , etc. Moreover if z 6= 0, z has a multiplicative inverse given by = (|z| + |w|) .
a b
z −1 = −i 2 . (3.9)
a2 + b2 a + b2
∞
Definition 3.19. A sequence {zn }n=1 ⊂ C is Cauchy if |zn − zm | → 0 as
Proof. The proof is a straightforward verification. Only the last assertion
m, n → ∞ and is convergent to z ∈ C if |z − zn | → 0 as n → ∞. As usual if
will be verified here. Suppose z = a + ib 6= 0, we wish to find w = c + id such ∞
{zn }n=1 converges to z we will write zn → z as n → ∞ or z = limn→∞ zn .
that zw = 1 and this happens by Eq. (3.8) iff
ac − bd = 1 and (3.10) Theorem 3.20. The complex numbers are complete,i.e. all Cauchy sequences
are convergent.
bc + ad = 0. (3.11)
a b
Proof. This follows from the completeness of real numbers and the easily
Solving these equations for c and d gives c = a2 +b2 and d = − a2 +b 2 as claimed. proved observations that if zn = an + ibn ∈ C, then
∞ ∞ ∞
1. {zn }n=1 ⊂ C is Cauchy iff {an }n=1 ⊂ R and {bn }n=1 ⊂ R are Cauchy and
Notation 3.17 (Conjugation and Modulus) If z = a + ib with a, b ∈ R let
2. zn → z = a + ib as n → ∞ iff an → a and bn → b as n → ∞.
z̄ = a − ib and
√ p q
2 2
|z| := z z̄ = a2 + b2 = |Re z| + |Im z| .
See Exercise 3.8 for the existence of the square root as a positive real number.

3.3 Exercises
Exercise 3.8. Show to every a ∈ R with a ≥ 0 there exists a unique

√ number
b ∈ R such√that b ≥ 0 and b2 = a. Of course we will call b = a. Also show
√
that a → a is an increasing function on [0, ∞). Hint: To construct b = a
for a > 0, to each n ∈ N let mn ∈ N0 be chosen so that
!
2 2
m2n
2
(mn + 1) mn (mn + 1)
<a≤ i.e. i <a≤i
n2 n2 n2 n2
mn ∞
and let qn := n . Then show b = {qn }n=1 ∈ R satisfies b > 0 and b2 = a.
4
Limits and Sums
4.1 Limsups, Liminfs and Extended Limits and bn = n−α with α > 0 shows the necessity for assuming right hand side of
Eq. (4.2) is not of the form ∞ · 0.
Notation 4.1 The extended real numbers is the set R̄ := R∪ {±∞} , i.e. it Proof. The proofs of items 1. and 2. are left to the reader.
is R with two new points called ∞ and −∞. We use the following conventions, Proof of Eq. (4.1). Let a := limn→∞ an and b = limn→∞ bn . Case 1.,
±∞ · 0 = 0, ±∞ · a = ±∞ if a ∈ R with a > 0, ±∞ · a = ∓∞ if a ∈ R with suppose b = ∞ in which case we must assume a > −∞. In this case, for every
a < 0, ±∞ + a = ±∞ for any a ∈ R, ∞ + ∞ = ∞ and −∞ − ∞ = −∞ while M > 0, there exists N such that bn ≥ M and an ≥ a − 1 for all n ≥ N and this
∞ − ∞ is not defined. A sequence an ∈ R̄ is said to converge to ∞ (−∞) if for implies
all M ∈ R there exists m ∈ N such that an ≥ M (an ≤ M ) for all n ≥ m. an + bn ≥ M + a − 1 for all n ≥ N.
∞ ∞ Since M is arbitrary it follows that an + bn → ∞ as n → ∞. The cases where
Lemma 4.2. Suppose {an }n=1 and {bn }n=1 are convergent sequences in R̄,
b = −∞ or a = ±∞ are handled similarly. Case 2. If a, b ∈ R, then for every
then:
ε > 0 there exists N ∈ N such that
1. If an ≤ bn for a.a. n then limn→∞ an ≤ limn→∞ bn . |a − an | ≤ ε and |b − bn | ≤ ε for all n ≥ N.
2. If c ∈ R, limn→∞ (can ) = c limn→∞ an .
∞
3. If {an + bn }n=1 is convergent and Therefore,
lim (an + bn ) = lim an + lim bn (4.1) |a + b − (an + bn )| = |a − an + b − bn | ≤ |a − an | + |b − bn | ≤ 2ε

n→∞ n→∞ n→∞
for all n ≥ N. Since n is arbitrary, it follows that limn→∞ (an + bn ) = a + b.
provided the right side is not of the form ∞ − ∞. Proof of Eq. (4.2). It will be left to the reader to prove the case where lim an
∞
4. {an bn }n=1 is convergent and and lim bn exist in R. I will only consider the case where a = limn→∞ an 6= 0
and limn→∞ bn = ∞ here. Let us also suppose that a > 0 (the case a < 0 is
lim (an bn ) = lim an · lim bn (4.2) handled similarly) and let α := min a2 , 1 . Given any M < ∞, there exists

n→∞ n→∞ n→∞
N ∈ N such that an ≥ α and bn ≥ M for all n ≥ N and for this choice of N,
provided the right hand side is not of the for ±∞ · 0 of 0 · (±∞) . an bn ≥ M α for all n ≥ N. Since α > 0 is fixed and M is arbitrary it follows
that limn→∞ (an bn ) = ∞ as desired.
Before going to the proof consider the simple example where an = n and For any subset Λ ⊂ R̄, let sup Λ and inf Λ denote the least upper bound and
bn = −αn with α > 0. Then greatest lower bound of Λ respectively. The convention being that sup Λ = ∞
 if ∞ ∈ Λ or Λ is not bounded from above and inf Λ = −∞ if −∞ ∈ Λ or Λ is
 ∞ if α < 1
lim (an + bn ) = 0 if α = 1 not bounded from below. We will also use the conventions that sup ∅ = −∞

−∞ if α > 1 and inf ∅ = +∞.
∞
Notation 4.3 Suppose that {xn }n=1 ⊂ R̄ is a sequence of numbers. Then
while
lim an + lim bn “ = ”∞ − ∞. lim inf xn = lim inf{xk : k ≥ n} and (4.3)
n→∞ n→∞ n→∞ n→∞
This shows that the requirement that the right side of Eq. (4.1) is not of form lim sup xn = lim sup{xk : k ≥ n}. (4.4)
n→∞ n→∞
∞−∞ is necessary in Lemma 4.2. Similarly by considering the examples an = n
We will also write lim for lim inf and lim for lim sup .
14 4 Limits and Sums
Remark 4.4. Notice that if ak := inf{xk : k ≥ n} and bk := sup{xk : k ≥ Hence by the definition of the limit, limk→∞ ak = a. If lim inf n→∞ an = ∞,
n}, then {ak } is an increasing sequence while {bk } is a decreasing sequence. then we know for all M ∈ (0, ∞) there is an integer N such that
Therefore the limits in Eq. (4.3) and Eq. (4.4) always exist in R̄ and
M ≤ inf{ak : k ≥ N }
lim inf xn = sup inf{xk : k ≥ n} and
n→∞ n and hence limn→∞ an = ∞. The case where lim supn→∞ an = −∞ is handled
lim sup xn = inf sup{xk : k ≥ n}. similarly.
n→∞ n
Conversely, suppose that limn→∞ an = A ∈ R̄ exists. If A ∈ R, then for
The following proposition contains some basic properties of liminfs and lim- every ε > 0 there exists N (ε) ∈ N such that |A − an | ≤ ε for all n ≥ N (ε), i.e.
sups.
A − ε ≤ an ≤ A + ε for all n ≥ N (ε).
Proposition 4.5. Let {an }∞
n=1 and {bn }∞
n=1 be two sequences of real numbers.
From this we learn that
Then
1. lim inf n→∞ an ≤ lim supn→∞ an and limn→∞ an exists in R̄ iff A − ε ≤ lim inf an ≤ lim sup an ≤ A + ε.
n→∞ n→∞
lim inf an = lim sup an ∈ R̄. Since ε > 0 is arbitrary, it follows that
n→∞ n→∞
A ≤ lim inf an ≤ lim sup an ≤ A,
2. There is a subsequence {ank }∞ ∞
k=1 of {an }n=1 such that limk→∞ ank = n→∞ n→∞
lim supn→∞ an . Similarly, there is a subsequence {ank }∞ ∞
k=1 of {an }n=1 such
that limk→∞ ank = lim inf n→∞ an . i.e. that A = lim inf n→∞ an = lim supn→∞ an . If A = ∞, then for all M > 0
3. there exists N = N (M ) such that an ≥ M for all n ≥ N. This show that
lim sup (an + bn ) ≤ lim sup an + lim sup bn (4.5) lim inf n→∞ an ≥ M and since M is arbitrary it follows that
n→∞ n→∞ n→∞
whenever the right side of this equation is not of the form ∞ − ∞. ∞ ≤ lim inf an ≤ lim sup an .
n→∞ n→∞
4. If an ≥ 0 and bn ≥ 0 for all n ∈ N, then
The proof for the case A = −∞ is analogous to the A = ∞ case.
lim sup (an bn ) ≤ lim sup an · lim sup bn , (4.6)
n→∞ n→∞ n→∞
provided the right hand side of (4.6) is not of the form 0 · ∞ or ∞ · 0. 4.2 Sums of positive functions
Proof. Item 1. will be proved here leaving the remaining items as an exercise In this and the next few sections, let X and Y be two sets. We will write α ⊂⊂ X
to the reader. Since to denote that α is a finite subset of X and write 2X f for those α ⊂⊂ X.
inf{ak : k ≥ n} ≤ sup{ak : k ≥ n} ∀ n, Definition 4.6. Suppose that a : X → [0, ∞] is a function and F ⊂ X is a

subset, then
lim inf an ≤ lim sup an .
n→∞ n→∞
( )
X X X
Now suppose that lim inf n→∞ an = lim supn→∞ an = a ∈ R. Then for all ε > 0, a= a(x) := sup a(x) : α ⊂⊂ F .
there is an integer N such that F x∈F x∈α
Remark 4.7. Suppose that X = N = {1, 2, 3, . . . } and a : X → [0, ∞], then

a − ε ≤ inf{ak : k ≥ N } ≤ sup{ak : k ≥ N } ≤ a + ε,
∞ N
i.e.
X X X
a= a(n) := lim a(n).
a − ε ≤ ak ≤ a + ε for all k ≥ N. N n=1
N →∞
n=1

4.2 Sums of positive functions 15
PN P
Indeed for all N, n=1 a(n) ≤ N a, and thus passing to the limit we learn Similarly, if α, β ⊂⊂ X, then
that X X X X X X
∞
X X a+ b≤ a+ b= (a + b) ≤ (a + b).
a(n) ≤ a. α β α∪β α∪β α∪β X
n=1 N
Taking sups over α and β then shows that
Conversely, if α ⊂⊂ N, then for all N large enough so that α ⊂ {1, 2, . . . , N }, X X X
P PN
we have α a ≤ n=1 a(n) which upon passing to the limit implies that a+ b≤ (a + b).
X X X
X ∞
X
a≤ a(n). Lemma 4.10. Let X and Y be sets, R ⊂ X × Y and suppose that a : R → R̄
α n=1 is a function. Let x R := {y ∈ Y : (x, y) ∈ R} and Ry := {x ∈ X : (x, y) ∈ R} .
Then
Taking the supremum over α in the previous equation shows
sup a(x, y) = sup sup a(x, y) = sup sup a(x, y) and
X ∞
X (x,y)∈R x∈X y∈x R y∈Y x∈Ry
a≤ a(n).
inf a(x, y) = inf inf a(x, y) = inf inf a(x, y).
N n=1 (x,y)∈R x∈X y∈x R y∈Y x∈Ry
(Recall the conventions: sup ∅ = −∞ and inf ∅ = +∞.)

P
Remark 4.8. Suppose a : X → [0, ∞] and X a < ∞, then {x ∈ X : a(x) > 0}
notice that for any ε > 0, the set {x :
is at most countable. To see this first P
Proof. Let M = sup(x,y)∈R a(x, y), Nx := supy∈x R a(x, y). Then a(x, y) ≤
a(x) ≥ ε} must be finite for otherwise X a = ∞. Thus
M for all (x, y) ∈ R implies Nx = supy∈x R a(x, y) ≤ M and therefore that
[
∞
{x ∈ X : a(x) > 0} = k=1 {x : a(x) ≥ 1/k} sup sup a(x, y) = sup Nx ≤ M. (4.7)
x∈X y∈x R x∈X
which shows that {x ∈ X : a(x) > 0} is a countable union of finite sets and thus Similarly for any (x, y) ∈ R,
countable by Lemma 2.6.
a(x, y) ≤ Nx ≤ sup Nx = sup sup a(x, y)
x∈X x∈X y∈x R
Lemma 4.9. Suppose that a, b : X → [0, ∞] are two functions, then
X X X and therefore
(a + b) = a+ b and M = sup a(x, y) ≤ sup sup a(x, y) (4.8)
X X X (x,y)∈R x∈X y∈x R
X X
λa = λ a Equations (4.7) and (4.8) show that
X X
sup a(x, y) = sup sup a(x, y).
(x,y)∈R x∈X y∈x R
for all λ ≥ 0.
The assertions involving infimums are proved analogously or follow from what
I will only prove the first assertion, the second being easy. Let α ⊂⊂ X be we have just proved applied to the function −a.
a finite set, then
X X X X X Theorem 4.11 (Monotone Convergence Theorem for Sums). Suppose
(a + b) = a+ b≤ a+ b that fn : X → [0, ∞] is an increasing sequence of functions and
α α α X X
f (x) := lim fn (x) = sup fn (x).
n→∞ n
which after taking sups over α shows that
X X X Then X X
(a + b) ≤ a+ b. lim fn = f
n→∞
X X X X X

and then taking the sup over all α ⊂⊂ X gives

X X
f ≤ A = lim fn (4.10)
n→∞
X X
which combined with Eq. (4.9) proves the theorem.
Lemma 4.12 (Fatou’s Lemma for Sums). Suppose that fn : X → [0, ∞] is

a sequence of functions, then
X X
lim inf fn ≤ lim inf fn .
n→∞ n→∞
X X
Fig. 4.1. The x and y – slices of a set R ⊂ X × Y. Proof. Define gk := inf fn so that gk ↑ lim inf n→∞ fn as k → ∞. Since
n≥k
gk ≤ fn for all n ≥ k,
X X
Proof. We will give two proofs. gk ≤ fn for all n ≥ k
First proof. Let X X
2X
f := {A ⊂ X : A ⊂⊂ X}. and therefore X X
Then gk ≤ lim inf fn for all k.
n→∞
X X X X X X
lim fn = sup fn = sup sup fn = sup sup fn
n→∞ n n α∈2X
α α∈2X n
α
We may now use the monotone convergence theorem to let k → ∞ to find
X X f f
X X
= sup lim fn = sup lim fn MCT
X X X X
n→∞ n→∞ lim inf fn = lim gk = lim gk ≤ lim inf fn .
α∈2X
f α α∈2X
f α n→∞ k→∞ k→∞ n→∞
X X X X
X X
= sup f= f.
α∈2X
f α X
P
P P Remark 4.13. If A = a < ∞, then for all ε > 0 there exists αε ⊂⊂ X such
X f. Since fn ≤ fm ≤ f for all
Second Proof. Let Sn = X
X fn and S = that
n ≤ m, it follows that X
A≥ a≥A−ε
Sn ≤ Sm ≤ S
α
which shows that limn→∞ Sn exists and is less that S, i.e. for all α ⊂⊂ X containing αε or equivalently,
X X
A := lim fn ≤ f. (4.9)
n→∞
X
X X A − a ≤ ε (4.11)

P P α

Noting that α fn ≤ X fn = Sn ≤ A for all α ⊂⊂ X and in particular,
P
X for all α ⊂⊂ X containing αε . Indeed, choose αε so that αε a ≥ A − ε.
fn ≤ A for all n and α ⊂⊂ X.
α
Letting n tend to infinity in this equation shows that

X
f ≤ A for all α ⊂⊂ X
α

4.3 Sums of complex functions 17
4.3 Sums of complex functions Thus it suffices to consider the case where a : X → R is a real function. Write
a = a+ − a− where
Definition 4.14. Suppose that a : X → C is a function, we say that
a+ (x) = max(a(x), 0) and a− (x) = max(−a(x), 0). (4.12)
X X
a= a(x)
Then |a| = a+ + a− and
X x∈X
X X X
exists and is equal to A ∈ C, if for all ε > 0 there is a finite subset αε ⊂ X ∞= |a| = a+ + a−
such that for all α ⊂⊂ X containing αε we have X X X
+ −
P P
P + X a = ∞ or
which shows that either X a = ∞. Suppose, with out loss

X 0
A −

a ≤ ε. of generality, that X a = ∞. Let X := {x ∈ X : a(x) ≥ 0}, then we know
that PX 0 a = ∞ which means there are finite subsets αn ⊂ X 0 ⊂ X such
P
α
that αP n
a ≥ n for all n. Thus if α ⊂⊂P X is any finite set, it follows that
The following lemma is left as an exercise to the reader. limn→∞ αn ∪α a = ∞, and therefore X a can not exist as a number in R.
Finally if a is summable, write X a = ρeiθ with ρ ≥ 0 and θ ∈ R, then
P P
Lemma
P 4.15. Suppose Pthat a, b : X → C are two functions such that X a
and X b exist, then X (a + λb) exists for all λ ∈ C and
X

X X
a = ρ = e−iθ a= e−iθ a

X X X
(a + λb) = a+λ b.
X

X X
X X X
X −iθ X +
Re e−iθ a

= Re e a ≤
Definition 4.16 (Summable). We call a function a : X → C summable if X
X X −iθ X
X
X ≤ Re e−iθ a ≤ e a ≤ |a| .
|a| < ∞. X X X
X P
Alternatively, this may be proved by approximating X a by a finite sum and
then using the triangle inequality of |·| .
P P
Proposition 4.17. Let a : X → C be a function, then X a exists iff X |a| <
∞, i.e. iff a is summable. Moreover if a is summable, then
Remark 4.18. Suppose that X = N and a : N → C is a sequence, then it is not
necessarily true that

X X
∞
a ≤ |a| .
X X

X X
a(n) = a(n). (4.13)
n=1 n∈N
P P± P ±
Proof. If X |a| < ∞, then X (Re a) < ∞ and X (Im a) < ∞ This is because
∞ N
and hence by Remark 4.13Pthese sums exists in the sense of Definition 4.14. X X
Therefore by Lemma 4.15, X a exists and a(n) = lim a(n)
N →∞
n=1 n=1
! P
X X +
X −
X +
X − depends on the ordering of the sequence a where as n∈N Pa(n) does not. For
a= (Re a) − (Re a) + i (Im a) − (Im a) . example, take a(n) = (−1)n
/n then
P
|a(n)| = ∞ i.e.
P∞ n∈N n∈N a(n) does not
X X X X X
exist while n=1 a(n) does exist. On the other hand, if
P
Conversely, if X |a| = ∞ then, because |a| ≤ |Re a| + |Im a| , we must have X ∞
X
X X |a(n)| = |a(n)| < ∞
|Re a| = ∞ or |Im a| = ∞. n∈N n=1
X X
then Eq. (4.13) is valid.


Theorem 4.19 (Dominated Convergence Theorem for Sums). Sup-
X

X X

pose that fn : X → C is a sequence of functions on X such that f (x) =
f − fn
=
(f − fn )
limn→∞ fn (x) ∈ C exists for all x ∈ X. Further assume there is a dominating β β β
function g : X → [0, ∞) such that ≤
X
|f − fn | =
X
|f − fn | +
X
|f − fn |
β α β\α
|fn (x)| ≤ g(x) for all x ∈ X and n ∈ N (4.14) X X
≤ |f − fn | + 2 g
and that g is summable. Then α β\α
X
≤ |f − fn | + 2ε.
X X
lim fn (x) = f (x). (4.15)
n→∞ α
x∈X x∈X
and hence that
Proof. Notice that |f | = lim |fn | ≤ g so that f is summable. By considering

X X X

the real and imaginary parts of f separately, it suffices to prove the theorem in f− fn ≤ |f − fn | + 2ε.

the case where f is real. By Fatou’s Lemma, β β α
X X X Since this last equation is true for all such β ⊂⊂ X, we learn that
(g ± f ) = lim inf (g ± fn ) ≤ lim inf (g ± fn )
n→∞ n→∞
X X X X X X
f− fn ≤ |f − fn | + 2ε
!
X X
= g + lim inf ± fn .
α
X X
n→∞
X X
which then implies that
Since lim inf n→∞ (−an ) = − lim supn→∞ an , we have shown,
X X

X
lim sup f− fn ≤ lim sup |f − fn | + 2ε

P
X X X lim inf n→∞ XPfn n→∞
X

X
n→∞
α
g± f≤ g+
− lim supn→∞ X fn = 2ε.
X X X
and therefore X X X Because ε > 0 is arbitrary we conclude that

lim sup fn ≤ f ≤ lim inf fn .
n→∞ n→∞ X X
X X X lim sup f− fn = 0.

P P n→∞
This shows that lim fn exists and is equal to f.

X X
n→∞ X X
Proof. (Second Proof.) Passing to the limit in Eq. (4.14) shows that |f | ≤ g which is the same as Eq. (4.15).
and in particular that f is summable. Given ε > 0, let α ⊂⊂ X such that
Remark 4.20. Theorem 4.19 may easily be generalized as follows. Suppose
X
g ≤ ε. fn , gn , g are summable
P functions
P on X such that fn → f and gn → g pointwise,
|fn | ≤ gn and X gn → X g as n → ∞. Then f is summable and Eq. (4.15)
X\α
still holds. For the proof we use Fatou’s Lemma to again conclude
Then for β ⊂⊂ X such that α ⊂ β, X X X
(g ± f ) = lim inf (gn ± fn ) ≤ lim inf (gn ± fn )
n→∞ n→∞
X X X
!
X X
= g + lim inf ± fn
n→∞
X X
and then proceed exactly as in the first proof of Theorem 4.19.

4.5 Exercises 19
4.4 Iterated sums and the Fubini and Tonelli Theorems Since α is arbitrary, it follows that
XX XX X
Let X and Y be two sets. The proof of the following lemma is left to the reader. a(x, y) = sup a(x, y) ≤ a
α⊂⊂X x∈α
x∈X y∈Y y∈Y X×Y
Lemma 4.21. Suppose that a : X →PC is function and
P F ⊂ X is a subset such
that a(x) = 0 for all x ∈
/ F. Then F a exists iff X a exists and when the which completes the proof.
sums exists,
Theorem 4.23 (Fubini’s Theorem for Sums). Now suppose that a : X ×
X X
a= a.
X F
Y → C is a summable function, i.e. by Theorem 4.22 any one of the following
equivalent conditions hold:
Theorem 4.22 (Tonelli’s Theorem for Sums). Suppose that a : X × Y → P
[0, ∞], then 1. PX×Y P |a| < ∞,
2. PX P Y |a| < ∞ or
X XX XX
a= a= a.
X×Y X Y Y X
3. Y X |a| < ∞.
Then X XX XX
Proof. It suffices to show, by symmetry, that a= a= a.
X XX X×Y X Y Y X
a= a
X×Y X Y Proof. If a : X → R is real valued the theorem follows by applying Theorem
4.22 to a± – the positive and negative parts of a. The general result holds for
Let Λ ⊂⊂ X × Y. Then for any α ⊂⊂ X and β ⊂⊂ Y such that Λ ⊂ α × β, we complex valued functions a by applying the real version just proved to the real
have X X XX XX XX and imaginary parts of a.
a≤ a= a≤ a≤ a,
Λ α×β α β α Y X Y
P P P
i.e. Λ a≤ X Y a. Taking the sup over Λ in this last equation shows 4.5 Exercises
X XX
a≤ a. Exercise 4.1. Now suppose for each n ∈ N := {1, 2, . . .} that fn : X → R is a
X×Y X Y function. Let
D := {x ∈ X : lim fn (x) = +∞}
For the reverse inequality, for each x ∈ X choose βnx ⊂⊂ Y such that βnx ↑ Y as n→∞
n ↑ ∞ and X X show that
a(x, y) = lim a(x, y).
n→∞
D = ∩∞ ∞
M =1 ∪N =1 ∩n≥N {x ∈ X : fn (x) ≥ M }. (4.16)
y∈Y x
y∈βn
Exercise 4.2. Let fn : X → R be as in the last problem. Let
If α ⊂⊂ X is a given finite subset of X, then
X X C := {x ∈ X : lim fn (x) exists in R}.
a(x, y) = lim a(x, y) for all x ∈ α n→∞
n→∞
y∈Y y∈βn
Find an expression for C similar to the expression for D in (4.16). (Hint: use
where βn := ∪x∈α βnx ⊂⊂ Y. Hence the Cauchy criteria for convergence.)
XX X X X X
a(x, y) = lim a(x, y) = lim a(x, y)
n→∞ n→∞
x∈α y∈Y x∈α y∈βn x∈α y∈βn
X X
= lim a(x, y) ≤ a.
n→∞
(x,y)∈α×βn X×Y

4.5.1 Limit Problems 2. There is a summable function g : Y → [0, ∞) such that
Exercise 4.3. Show lim inf n→∞ (−an ) = − lim supn→∞ an . |f (u, y)| ≤ g(y) for all y ∈ Y and u ∈ U.
Exercise 4.4. Suppose that lim supn→∞ an = M ∈ R̄, show that there is a Show that X
subsequence {ank }∞ ∞ F (u) := f (u, y) (4.19)
k=1 of {an }n=1 such that limk→∞ ank = M.
y∈Y
Exercise 4.5. Show that is a continuous function for u ∈ U.
lim sup(an + bn ) ≤ lim sup an + lim sup bn (4.17) Exercise 4.11. Suppose that Y is a set, J = (a, b) ⊂ R is an interval, and
n→∞ n→∞ n→∞
f : J × Y → C is a function satisfying:
provided that the right side of Eq. (4.17) is well defined, i.e. no ∞ − ∞ or
1. For each y ∈ Y, the function u → f (u, y) is differentiable on J,
−∞ + ∞ type expressions. (It is OK to have ∞ + ∞ = ∞ or −∞ − ∞ = −∞,
2. There is a summable function g : Y → [0, ∞) such that
etc.)
∂
Exercise 4.6. Suppose that an ≥ 0 and bn ≥ 0 for all n ∈ N. Show f (u, y) ≤ g(y) for all y ∈ Y and u ∈ J.
∂u
lim sup(an bn ) ≤ lim sup an · lim sup bn , (4.18) P
3. There is a u0 ∈ J such that y∈Y |f (u0 , y)| < ∞.
n→∞ n→∞ n→∞
provided the right hand side of (4.18) is not of the form 0 · ∞ or ∞ · 0. Show:
P
Exercise 4.7. Prove Lemma 4.15. a) for all u ∈ J P
that y∈Y |f (u, y)| < ∞.
b) Let F (u) := y∈Y f (u, y), show F is differentiable on J and that
Exercise 4.8. Prove Lemma 4.21. X ∂
Ḟ (u) = f (u, y).
∂u
4.5.2 Dominated Convergence Theorem Problems y∈Y
(Hint: Use the mean value theorem.)

Notation 4.24 For u0 ∈ Rn and δ > 0, let Bu0 (δ) := {x ∈ Rn : |x − u0 | < δ}
be the ball in Rn centered at u0 with radius δ. Exercise 4.12 (Differentiation of Power Series). P Suppose R > 0 and
∞ ∞ n
n {an }n=0 is a sequence of complex numbers such thatP∞ n=0n|an | r < ∞ for
Exercise 4.9. Suppose U ⊂ R is a set and u0 ∈ U is a point such that
all r ∈ (0, R). Show, using Exercise 4.11, f (x) := n=0 an x is continuously
U ∩ (Bu0 (δ) \ {u0 }) 6= ∅ for all δ > 0. Let G : U \ {u0 } → C be a function
differentiable for x ∈ (−R, R) and
on U \ {u0 }. Show that limu→u0 G(u) exists and is equal to λ ∈ C,1 iff for all
∞ ∞ ∞
sequences {un }n=1 ⊂ U \ {u0 } which converge to u0 (i.e. limn→∞ un = u0 ) we X X
have limn→∞ G(un ) = λ. f 0 (x) = nan xn−1 = nan xn−1 .
n=0 n=1
Exercise 4.10. Suppose that Y is a set, U ⊂ Rn is a set, and f : U × Y → C Exercise 4.13. Show the functions
is a function satisfying: ∞
X xn
1. For each y ∈ Y, the function u ∈ U → f (u, y) is continuous on U. 2 ex := , (4.20)
n=0
n!
1
More explicitly, limu→u0 G(u) = λ means for every every > 0 there exists a δ > 0 ∞
X n x2n+1
such that sin x := (−1) and (4.21)
|G(u) − λ| < whenever u ∈ U ∩ (Bu0 (δ) \ {u0 }) . n=0
(2n + 1)!
∞
X n x2n
2
To say g := f (·, y) is continuous on U means that g : U → C is continuous relative cos x = (−1) (4.22)
(2n)!
to the metric on Rn restricted to U. n=0

are infinitely differentiable and they satisfy
d x
e = ex with e0 = 1
dx
d
sin x = cos x with sin (0) = 0
dx
d
cos x = − sin x with cos (0) = 1.
dx
Exercise 4.14. Continue the notation of Exercise 4.13.
1. Use the product and the chain rule to show,
d h −x (x+y) i
e e =0
dx
and conclude from this, that e−x e(x+y) = ey for all x, y ∈ R. In particular
taking y = 0 this implies that e−x = 1/ex and hence that e(x+y) = ex ey .
x
PeN ↑ ∞
Use this result to show as x ↑ ∞ and ex ↓ 0 as x ↓ −∞.
xn n
Remark: since e ≥ n=0 n! when x ≥ 0, it follows that limx→∞ xex = 0
x
for any n ∈ N, i.e. ex grows at a rate faster than any polynomial in x as

x → ∞.
2. Use the product rule to show
d
cos2 x + sin2 x = 0

dx
and use this to conclude that cos2 x + sin2 x = 1 for all x ∈ R.
∞
P∞ 4.15. Let {an }n=−∞ be a summable sequence of complex numbers,
Exercise
i.e. n=−∞ |an | < ∞. For t ≥ 0 and x ∈ R, define
∞
X 2
F (t, x) = an e−tn einx ,
n=−∞
where as usual eix = cos(x) + i sin(x), this is motivated by replacing x in Eq.

(4.20) by ix and comparing the result to Eqs. (4.21) and (4.22).
1. F (t, x) is continuous for (t, x) ∈ [0, ∞) × R. Hint: Let Y = Z and u = (t, x)
and use Exercise 4.10.
2. ∂F (t, x)/∂t, ∂F (t, x)/∂x and ∂ 2 F (t, x)/∂x2 exist for t > 0 and x ∈ R.
Hint: Let Y = Z and u = t for computing ∂F (t, x)/∂t and u = x for
computing ∂F (t, x)/∂x and ∂ 2 F (t, x)/∂x2 via Exercise 4.11. In computing
the t derivative, you should let ε > 0 and apply Exercise 4.11 with t = u > ε
and then afterwards let ε ↓ 0.
3. F satisfies the heat equation, namely
∂F (t, x)/∂t = ∂ 2 F (t, x)/∂x2 for t > 0 and x ∈ R.
5
`p – spaces, Minkowski and Holder Inequalities
Z s Z t
In this chapter, let µ : X → (0, ∞) be a given function. Let F denote either
F (s) = f (s0 )ds0 and G(t) = g(t0 )dt0 .
R or C. For p ∈ (0, ∞) and f : X → F, let 0 0
X
!1/p Then for all s, t ≥ 0,
p
kf kp := |f (x)| µ(x) st ≤ F (s) + G(t)
x∈X
and equality holds iff t = f (s).
and for p = ∞ let
Proof. Let
kf k∞ = sup {|f (x)| : x ∈ X} .
Also, for p > 0, let As := {(σ, τ ) : 0 ≤ τ ≤ f (σ) for 0 ≤ σ ≤ s} and
Bt := {(σ, τ ) : 0 ≤ σ ≤ g(τ ) for 0 ≤ τ ≤ t}
`p (µ) = {f : X → F : kf kp < ∞}.
then as one sees from Figure 5.1, [0, s] × [0, t] ⊂ As ∪ Bt . (In the figure: s = 3,
In the case where µ(x) = 1 for all x ∈ X we will simply write `p (X) for `p (µ). t = 1, A3 is the region under t = f (s) for 0 ≤ s ≤ 3 and B1 is the region to the
left of the curve s = g(t) for 0 ≤ t ≤ 1.) Hence if m denotes the area of a region
Definition 5.1. A norm on a vector space Z is a function k·k : Z → [0, ∞)
in the plane, then
such that
1. (Homogeneity) kλf k = |λ| kf k for all λ ∈ F and f ∈ Z. st = m ([0, s] × [0, t]) ≤ m(As ) + m(Bt ) = F (s) + G(t).
2. (Triangle inequality) kf + gk ≤ kf k + kgk for all f, g ∈ Z.
As it stands, this proof is a bit on the intuitive side. However, it will become
3. (Positive definite) kf k = 0 implies f = 0.
rigorous if one takes m to be “Lebesgue measure” on the plane which will be
A function p : Z → [0, ∞) satisfying properties 1. and 2. but not necessarily introduced later.
3. above will be called a semi-norm on Z. We can also give a calculus proof of this theorem under the additional as-
A pair (Z, k·k) where Z is a vector space and k·k is a norm on Z is called a sumption that f is C 1 . (This restricted version of the theorem is all we need in
normed vector space. this section.) To do this fix t ≥ 0 and let
Z s
The rest of this section is devoted to the proof of the following theorem. h(s) = st − F (s) = (t − f (σ))dσ.
0
Theorem 5.2. For p ∈ [1, ∞], (`p (µ), k · kp ) is a normed vector space.
If σ > g(t) = f −1 (t), then t − f (σ) < 0 and hence if s > g(t), we have
Proof. The only difficulty is the proof of the triangle inequality which is
the content of Minkowski’s Inequality proved in Theorem 5.8 below. Z s Z g(t) Z s
h(s) = (t − f (σ))dσ = (t − f (σ))dσ + (t − f (σ))dσ
Proposition 5.3. Let f : [0, ∞) → [0, ∞) be a continuous strictly increasing 0 0 g(t)
function such that f (0) = 0 (for simplicity) and lim f (s) = ∞. Let g = f −1 g(t)
Z
s→∞ ≤ (t − f (σ))dσ = h(g(t)).
and for s, t ≥ 0 let 0
24 5 `p – spaces, Minkowski and Holder Inequalities
Combining this with h(0) = 0 we see that h(s) takes its maximum at some sp tq
st ≤ + for all s, t ≥ 0
point s ∈ (0, g(t)] and hence at a point where 0 = h0 (s) = t − f (s). The only p q
solution to this equation is s = g(t) and we have thus shown with equality if and only if tq = sp .
g(t) p 1
Proof. Let F (s) = sp for p > 1. Then f (s) = sp−1 = t and g(t) = t p−1 =
Z
st − F (s) = h(s) ≤ (t − f (σ))dσ = h(g(t)) q−1
0 t , wherein we have used q − 1 = p/ (p − 1) − 1 = 1/ (p − 1) . Therefore
R g(t) G(t) = tq /q and hence by Proposition 5.3,
with equality when s = g(t). To finish the proof we must show 0 (t −
f (σ))dσ = G(t). This is verified by making the change of variables σ = g(τ ) sp tq
st ≤ +
and then integrating by parts as follows: p q
g(t) t t with equality iff t = sp−1 , i.e. tq = sq(p−1) = sp . For those who do not want to
Z Z Z
(t − f (σ))dσ = (t − f (g(τ )))g 0 (τ )dτ = (t − τ )g 0 (τ )dτ use Proposition 5.3, here is a direct calculus proof. Fix t > 0 and let
0 0 0
Z t
sp
= g(τ )dτ = G(t). h (s) := st − .
0 p
Then h (0) = 0, lims→∞ h (s) = −∞ and h0 (s) = t − sp−1 which equals zero iff
1
s = t p−1 . Since
p p
tq

1 1 t p−1 p t p−1 q 1
h t p−1 =t p−1 t− =t p−1 − =t 1− = ,
p p p q
it follows from the first derivative test that

q
n 1 o t tq
max h = max h (0) , h t p−1 = max 0, = .
q q
So we have shown
sp tq
st − ≤ with equality iff t = sp−1 .
Fig. 5.1. A picture proof of Proposition 5.3. p q
Theorem 5.6 (Hölder’s inequality). Let p, q ∈ [1, ∞] be conjugate expo-

nents. For all f, g : X → F,
p
Definition 5.4. The conjugate exponent q ∈ [1, ∞] to p ∈ [1, ∞] is q := p−1
with the conventions that q = ∞ if p = 1 and q = 1 if p = ∞. Notice that q is kf gk1 ≤ kf kp · kgkq . (5.2)
characterized by any of the following identities:
If p ∈ (1, ∞) and f and g are not identically zero, then equality holds in Eq.
1 1 q p (5.2) iff
+ = 1, 1 + = q, p − = 1 and q(p − 1) = p. (5.1) p q
p q p q |f | |g|
= . (5.3)
p kf kp kgkq
Lemma 5.5. Let p ∈ (1, ∞) and q := p−1 ∈ (1, ∞) be the conjugate exponent.
Then

5 `p – spaces, Minkowski and Holder Inequalities 25
Proof. The proof of Eq. (5.2) for p ∈ {1, ∞} is easy and will be left to kf + gk∞ = sup |f + g| ≤ sup (|f | + |g|)
X X
the reader. The cases where kf kp = 0 or ∞ or kgkq = 0 or ∞ are easily dealt
with and are also left to the reader. So we will assume that p ∈ (1, ∞) and ≤ sup |f | + sup |g| = kf k∞ + kgk∞ .
X X
0 < kf kp , kgkq < ∞. Letting s = |f (x)| /kf kp and t = |g|/kgkq in Lemma 5.5
implies Now assume that p ∈ (1, ∞). Since
p q
|f (x) g (x)| 1 |f (x)| 1 |g (x)| p p p p p p
≤ p + q |f + g| ≤ (2 max (|f | , |g|)) = 2p max (|f | , |g| ) ≤ 2p (|f | + |g| )
kf kp kgkq p kf kp q kgkq
with equality iff it follows that
p q
kf + gkpp ≤ 2p kf kpp + kgkpp < ∞.

|f (x)| |g (x)|
= sp = tq = . (5.4)
kf kpp kgkqq Eq. (5.5) is easily verified if kf + gkp = 0, so we may assume kf + gkp > 0.
Multiplying this equation by µ (x) and then summing on x gives Multiplying the inequality,
p
kf gk1 1 1 |f + g| = |f + g||f + g|p−1 ≤ |f | |f + g|p−1 + |g||f + g|p−1 (5.6)
≤ + =1
kf kp kgkq p q by µ, then summing on x and applying Holder’s inequality on each term gives
with equality iff Eq. (5.4) holds for all x ∈ X, i.e. iff Eq. (5.3) holds.
X X X
|f + g|p µ ≤ |f | |f + g|p−1 µ + |g| |f + g|p−1 µ
X X X
Definition 5.7. For a complex number λ ∈ C, let
p−1
≤ (kf kp + kgkp ) |f + g| . (5.7)

λ q
if λ 6= 0
sgn(λ) = |λ|
0 if λ = 0. Since q(p − 1) = p, as in Eq. (5.1),
X X
For λ, µ ∈ C we will write sgn(λ) $ sgn(µ) if sgn(λ) = sgn(µ) or λµ = 0. k|f + g|p−1 kqq = (|f + g|p−1 )q µ = |f + g|p µ = kf + gkpp . (5.8)
X X
Theorem 5.8 (Minkowski’s Inequality). If 1 ≤ p ≤ ∞ and f, g ∈ `p (µ)
then Combining Eqs. (5.7) and (5.8) shows
kf + gkp ≤ kf kp + kgkp . (5.5)
kf + gkpp ≤ (kf kp + kgkp ) kf + gkp/q
p (5.9)
Moreover, assuming f and g are not identically zero, equality holds in Eq. (5.5)
iff and solving this equation for kf +gkp (making use of Eq. (5.1)) implies Eq. (5.5).
Now suppose that f and g are not identically zero and p ∈ (1, ∞) . Equality
sgn(f ) $ sgn(g) when p = 1 and holds in Eq. (5.5) iff equality holds in Eq. (5.9) iff equality holds in Eq. (5.7)
and Eq. (5.6). The latter happens iff
f = cg for some c > 0 when p ∈ (1, ∞).
sgn(f ) $ sgn(g) and
Proof. For p = 1, p p
|f + g|p

|f | |g|
= = . (5.10)
kf + gkpp
X X X X
kf + gk1 = |f + g|µ ≤ (|f | µ + |g|µ) = |f | µ + |g|µ kf kp kgkp
X X X X
wherein we have used
with equality iff
|f + g|p−1
q
|f + g|p
= .
|f | + |g| = |f + g| ⇐⇒ sgn(f ) $ sgn(g). k|f + g|p−1 kq kf + gkpp
Finally Eq. (5.10) is equivalent to |f | = c |g| with c = (kf kp /kgkp ) > 0 and this
For p = ∞,
equality along with sgn(f ) $ sgn(g) implies f = cg.

5.1 Exercises
Exercise 5.1. Generalize Proposition 5.3 as follows. Let a ∈ [−∞, 0] and f :

R ∩ [a, ∞) → [0, ∞) be a continuous strictly increasing function such that
lim f (s) = ∞, f (a) = 0 if a > −∞ or lims→−∞ f (s) = 0 if a = −∞. Also let
s→∞
g = f −1 , b = f (0) ≥ 0,
Z s Z t
0 0
F (s) = f (s )ds and G(t) = g(t0 )dt0 .
0 0
Then for all s, t ≥ 0,
st ≤ F (s) + G(t ∨ b) ≤ F (s) + G(t)

Fig. 5.3. When t ≤ b, notice that g(t) ≤ 0 but G(t) ≥ 0. Also notice that G(t) is no
and equality holds iff t = f (s). In particular, taking f (s) = es , prove Young’s longer needed to estimate st.
inequality stating
st ≤ es + (t ∨ 1) ln (t ∨ 1) − (t ∨ 1) ≤ es + t ln t − t,
where s ∨ t := min (s, t) . Hint: Refer to Figures 5.2 and 5.3..
Fig. 5.2. Comparing areas when t ≥ b goes the same way as in the text.
Part II
Metric, Banach, and Hilbert Space Basics

6
Metric Spaces
Definition 6.1. A function d : X × X → [0, ∞) is called a metric if Exercise 6.2. Let F be a collection of closed subsets of X, show ∩F := ∩F ∈F F
n
is closed. Also show that finite unions of closed sets are closed, i.e. if {Fk }k=1
1. (Symmetry) d(x, y) = d(y, x) for all x, y ∈ X are closed sets then ∪nk=1 Fk is closed. (By taking complements, this shows that
2. (Non-degenerate) d(x, y) = 0 if and only if x = y ∈ X the collection of open sets, τd , is closed under finite intersections and arbitrary
3. (Triangle inequality) d(x, z) ≤ d(x, y) + d(y, z) for all x, y, z ∈ X. unions.)
As primary examples, any normed space (X, k·k) (see Definition 5.1) is a The following “continuity” facts of the metric d will be used frequently in
metric space with d(x, y) := kx − yk . Thus the space `p (µ) (as in Theorem the remainder of this book.
5.2) is a metric space for all p ∈ [1, ∞]. Also any subset of a metric space is a
metric space. For example a surface Σ in R3 is a metric space with the distance Lemma 6.6. For any non empty subset A ⊂ X, let dA (x) := inf{d(x, a)|a ∈
between two points on Σ being the usual distance in R3 . A}, then
|dA (x) − dA (y)| ≤ d(x, y) ∀ x, y ∈ X (6.1)
Definition 6.2. Let (X, d) be a metric space. The open ball B(x, δ) ⊂ X and in particular if xn → x in X then dA (xn ) → dA (x) as n → ∞. Moreover
centered at x ∈ X with radius δ > 0 is the set the set Fε := {x ∈ X|dA (x) ≥ ε} is closed in X.
B(x, δ) := {y ∈ X : d(x, y) < δ}. Proof. Let a ∈ A and x, y ∈ X, then
We will often also write B(x, δ) as Bx (δ). We also define the closed ball cen- dA (x) ≤ d(x, a) ≤ d(x, y) + d(y, a).
tered at x ∈ X with radius δ > 0 as the set Cx (δ) := {y ∈ X : d(x, y) ≤ δ}. Take the infimum over a in the above equation shows that
∞
Definition 6.3. A sequence {xn }n=1
in a metric space (X, d) is said to be dA (x) ≤ d(x, y) + dA (y) ∀x, y ∈ X.
convergent if there exists a point x ∈ X such that limn→∞ d(x, xn ) = 0. In
this case we write limn→∞ xn = x or xn → x as n → ∞. Therefore, dA (x) − dA (y) ≤ d(x, y) and by interchanging x and y we also have
that dA (y) − dA (x) ≤ d(x, y) which implies Eq. (6.1). If xn → x ∈ X, then by
Exercise 6.1. Show that x in Definition 6.3 is necessarily unique. Eq. (6.1),
|dA (x) − dA (xn )| ≤ d(x, xn ) → 0 as n → ∞
Definition 6.4. A set E ⊂ X is bounded if E ⊂ B (x, R) for some x ∈ X and ∞
∞ so that limn→∞ dA (xn ) = dA (x) . Now suppose that {xn }n=1 ⊂ Fε and xn → x
R < ∞. A set F ⊂ X is closed iff every convergent sequence {xn }n=1 which is
in X, then
contained in F has its limit back in F. A set V ⊂ X is open iff V c is closed.
dA (x) = lim dA (xn ) ≥ ε
We will write F @ X to indicate F is a closed subset of X and V ⊂o X to n→∞
indicate the V is an open subset of X. We also let τd denote the collection of since dA (xn ) ≥ ε for all n. This shows that x ∈ Fε and hence Fε is closed.
open subsets of X relative to the metric d.
Corollary 6.7. The function d satisfies,
Definition 6.5. A subset A ⊂ X is a neighborhood of x if there exists an
|d(x, y) − d(x0 , y 0 )| ≤ d(y, y 0 ) + d(x, x0 ).
open set V ⊂o X such that x ∈ V ⊂ A. We will say that A ⊂ X is an open
neighborhood of x if A is open and x ∈ A. In particular d : X × X → [0, ∞) is “continuous” in the sense that d(x, y) is
close to d(x0 , y 0 ) if x is close to x0 and y is close to y 0 . (The notion of continuity
will be developed shortly.)
30 6 Metric Spaces
Proof. By Lemma 6.6 for single point sets and the triangle inequality for Definition 6.11. A function f : X → Y is continuous at x ∈ X if for all
the absolute value of real numbers, ε > 0 there is a δ > 0 such that
|d(x, y) − d(x0 , y 0 )| ≤ |d(x, y) − d(x, y 0 )| + |d(x, y 0 ) − d(x0 , y 0 )| d(f (x), f (x0 )) < ε provided that ρ(x, x0 ) < δ. (6.3)
≤ d(y, y 0 ) + d(x, x0 ).
The function f is said to be continuous if f is continuous at all points x ∈ X.
The following lemma gives two other characterizations of continuity of a

c
Example 6.8. Let x ∈ X and δ > 0, then Cx (δ) and Bx (δ) are closed subsets function at a point.
∞
of X. For example if {yn }n=1 ⊂ Cx (δ) and yn → y ∈ X, then d (yn , x) ≤ δ for
all n and using Corollary 6.7 it follows d (y, x) ≤ δ, i.e. y ∈ Cx (δ) . A similar Lemma 6.12 (Local Continuity Lemma). Suppose that (X, ρ) and (Y, d)
c
proof shows Bx (δ) is open, see Exercise 6.3. are two metric spaces and f : X → Y is a function defined in a neighborhood
of a point x ∈ X. Then the following are equivalent:
Exercise 6.3. Show that V ⊂ X is open iff for every x ∈ V there is a δ > 0
such that Bx (δ) ⊂ V. In particular show Bx (δ) is open for all x ∈ X and δ > 0. 1. f is continuous at x ∈ X.
Hint: by definition V is not open iff V c is not closed. 2. For all neighborhoods A ⊂ Y of f (x), f −1 (A) is a neighborhood of x ∈ X.
∞
3. For all sequences {xn }n=1 ⊂ X such that x = limn→∞ xn , {f (xn )} is con-
Lemma 6.9 (Approximating open sets from the inside by closed sets). vergent in Y and
Let A be a closed subset of X and Fε := {x ∈ X|dA (x) ≥ ε} @ X be as in lim f (xn ) = f lim xn .
Lemma 6.6. Then Fε ↑ Ac as ε ↓ 0. n→∞ n→∞
Proof. It is clear that dA (x) = 0 for x ∈ A so that Fε ⊂ Ac for each ε > 0 Proof. 1 =⇒ 2. If A ⊂ Y is a neighborhood of f (x) , there exists ε > 0
and hence ∪ε>0 Fε ⊂ Ac . Now suppose that x ∈ Ac ⊂o X. By Exercise 6.3 there such that Bf (x) (ε) ⊂ A and because f is continuous there exists a δ > 0 such
exists an ε > 0 such that Bx (ε) ⊂ Ac , i.e. d(x, y) ≥ ε for all y ∈ A. Hence that Eq. (6.3) holds. Therefore
x ∈ Fε and we have shown that Ac ⊂ ∪ε>0 Fε . Finally it is clear that Fε ⊂ Fε0
Bx (δ) ⊂ f −1 Bf (x) (ε) ⊂ f −1 (A)

whenever ε0 ≤ ε.
showing f −1 (A) is a neighborhood of x.
Definition 6.10. Given a set A contained in a metric space X, let Ā ⊂ X be ∞
2 =⇒ 3. Suppose that {xn }n=1 ⊂ X and x = limn→∞ xn . Then for any ε >
the closure of A defined by
0, Bf (x) (ε) is a neighborhood of f (x) and so f −1 Bf (x) (ε) is a neighborhood
Ā := {x ∈ X : ∃ {xn } ⊂ A 3 x = lim xn }.
n→∞
of x which must contain Bx (δ) for some δ > 0. Because xn → x, it follows that
xn ∈ Bx (δ) ⊂ f −1 Bf (x) (ε) for a.a. n and this implies f (xn ) ∈ Bf (x) (ε) for
That is to say Ā contains all limit points of A. We say A is dense in X if a.a. n, i.e. d(f (x), f (xn )) < ε for a.a. n. Since ε > 0 is arbitrary it follows that
Ā = X, i.e. every element x ∈ X is a limit of a sequence of elements from A. limn→∞ f (xn ) = f (x) .
3. =⇒ 1. We will show not 1. =⇒ not 3. If f is not continuous at x,
Exercise 6.4. Given A ⊂ X, show Ā is a closed set and in fact there exists an ε > 0 such that for all n ∈ N there exists a point xn ∈ X with
ρ (xn , x) < n1 yet d (f (xn ) , f (x)) ≥ ε. Hence xn → x as n → ∞ yet f (xn ) does
Ā = ∩{F : A ⊂ F ⊂ X with F closed}. (6.2) not converge to f (x) .
That is to say Ā is the smallest closed set containing A. Here is a global version of the previous lemma.
Lemma 6.13 (Global Continuity Lemma). Suppose that (X, ρ) and (Y, d)
are two metric spaces and f : X → Y is a function defined on all of X. Then
6.1 Continuity the following are equivalent:
Suppose that (X, ρ) and (Y, d) are two metric spaces and f : X → Y is a 1. f is continuous.
function. 2. f −1 (V ) ∈ τρ for all V ∈ τd , i.e. f −1 (V ) is open in X if V is open in Y.

6.2 Completeness in Metric Spaces 31
3. f −1 (C) is closed in X if C is closed in Y. 6.2 Completeness in Metric Spaces

4. For all convergent sequences {xn } ⊂ X, {f (xn )} is convergent in Y and
∞
Definition 6.16 (Cauchy sequences). A sequence {xn }n=1 in a metric space
lim f (xn ) = f lim xn . (X, d) is Cauchy provided that
n→∞ n→∞
c lim d(xn , xm ) = 0.
Proof. Since f −1 (Ac ) = f −1 (A) , it is easily seen that 2. and 3. are

m,n→∞
equivalent. So because of Lemma 6.12 it only remains to show 1. and 2. are
equivalent. If f is continuous and V ⊂ Y is open, then for every x ∈ f −1 (V ) , V Exercise 6.6. Show that convergent sequences are always Cauchy sequences.
is a neighborhood of f (x) and so f −1 (V ) is a neighborhood of x. Hence f −1 (V ) The converse is not always true. For example, let X = Q be the set of rational
∞
is a neighborhood of all of its points and from this and Exercise 6.3 it follows that numbers
√ and d(x, y) = |x−y|. Choose a sequence {xn }n=1 ⊂ Q which converges
∞
f −1 (V ) is open. Conversely, if x ∈ X and A ⊂ Y is a neighborhood of f (x) then to 2 ∈ R, then {xn }n=1 is (Q, d) – Cauchy but not (Q, d) – convergent. The
there exists V ⊂o X such that f (x) ∈ V ⊂ A. Hence x ∈ f −1 (V ) ⊂ f −1 (A) sequence does converge in R however.
and by assumption f −1 (V ) is open showing f −1 (A) is a neighborhood of x.
Definition 6.17. A metric space (X, d) is complete if all Cauchy sequences
Therefore f is continuous at x and since x ∈ X was arbitrary, f is continuous.
are convergent sequences.
Exercise 6.7. Let (X, d) be a complete metric space. Let A ⊂ X be a subset

Example 6.14. The function dA defined in Lemma 6.6 is continuous for each
of X viewed as a metric space using d|A×A . Show that (A, d|A×A ) is complete
A ⊂ X. In particular, if A = {x} , it follows that y ∈ X → d(y, x) is continuous
iff A is a closed subset of X.
for each x ∈ X.
Example 6.18. Examples 2. – 4. of complete metric spaces will be verified in
Exercise 6.5. Use Example 6.14 and Lemma 6.13 to recover the results of
Chapter 7 below.
Example 6.8.
1. X = R and d(x, y) = |x − y|, see Theorem 3.8 above.
The next result shows that there are lots of continuous functions on a metric Pn 2 1/2

space (X, d) . 2. X = Rn and d(x, y) = kx − yk2 = i=1 (xi − yi ) .
3. X = `p (µ) for p ∈ [1, ∞] and any weight function µ : X → (0, ∞).
Lemma 6.15 (Urysohn’s Lemma for Metric Spaces). Let (X, d) be a met- 4. X = C([0, 1], R) – the space of continuous functions from [0, 1] to R and
ric space and suppose that A and B are two disjoint closed subsets of X. Then
d(f, g) := max |f (t) − g(t)|.
t∈[0,1]
dB (x)
f (x) = for x ∈ X (6.4)
dA (x) + dB (x) This is a special case of Lemma 7.3 below.
5. Let X = C([0, 1], R) and
defines a continuous function, f : X → [0, 1], such that f (x) = 1 for x ∈ A and
f (x) = 0 if x ∈ B. Z 1
d(f, g) := |f (t) − g(t)| dt.
Proof. By Lemma 6.6, dA and dB are continuous functions on X. Since 0
A and B are closed, dA (x) > 0 if x ∈ / A and dB (x) > 0 if x ∈ / B. Since
−1 You are asked in Exercise 7.11 to verify that (X, d) is a metric space which
A ∩ B = ∅, dA (x) + dB (x) > 0 for all x and (dA + dB ) is continuous as well. is not complete.
The remaining assertions about f are all easy to verify.
Sometimes Urysohn’s lemma will be use in the following form. Suppose Exercise 6.8 (Completions of Metric Spaces). Suppose that (X, d) is a
F ⊂ V ⊂ X with F being closed and V being open, then there exists f ∈ (not necessarily complete) metric space.
Using the following outline show there
C (X, [0, 1])) such that f = 1 on F while f = 0 on V c . This of course follows exists a complete metric space X̄, d¯ and an isometric map i : X → X̄ such
from Lemma 6.15 by taking A = F and B = V c . that i (X) is dense in X̄, see Definition 6.10.

32 6 Metric Spaces
∞
1. Let C denote the collection of Cauchy sequences a = {an }n=1 ⊂ X. Given Lemma 6.20. Suppose that (X, |·|) is a normed vector space and d is the metric
two element a, b ∈ C show dC (a, b) := limn→∞ d (an , bn ) exists, dC (a, b) ≥ 0 on X defined by d(x, y) = |x − y| . Then
for all a, b ∈ C and dC satisfies the triangle inequality,
Bx (ε) = Cx (ε) and
dC (a, c) ≤ dC (a, b) + dC (b, c) for all a, b, c ∈ C. bd(Bx (ε)) = {y ∈ X : d(x, y) = ε}.
Thus (C, dC ) would be a metric space if it were true that dC (a, b) = 0 iff where the boundary operation, bd(·) is defined in Definition 13.29 (BRUCE:
a = b. This however is false, for example if an = bn for all n ≥ 100, then Forward Reference.) below.
dC (a, b) = 0 while a need not equal b.
2. Define two elements a, b ∈ C to be equivalent (write a ∼ b) when- Proof. We must show that C := Cx (ε) ⊂ Bx (ε) =: B̄. For y ∈ C, let
ever dC (a, b) = 0. Show “ ∼ ” is an equivalence relation on C and that v = y − x, then
dC (a0 , b0 ) = dC (a, b) if a ∼ a0 and b ∼ b0 . (Hint: see Corollary 6.7.) |v| = |y − x| = d(x, y) ≤ ε.
3. Given a ∈ C let ā := {b ∈ C : b ∼ a} denote the equivalence class containing Let αn = 1 − 1/n so that αn ↑ 1 as n → ∞. Let yn = x + αn v, then d(x, yn ) =
a and let X̄ := {ā : a ∈ C} denote the collection of such equivalence classes. αn d(x, y) < ε, so that yn ∈ Bx (ε) and d(y, yn ) = (1 − αn ) |v| → 0 as n → ∞.
Show that d¯ ā, b̄ := dC (a, b) is well defined on X̄ × X̄ and verify X̄, d¯ is

This shows that yn → y as n → ∞ and hence that y ∈ B̄.
a metric space.
4. For x ∈ X let i (x) = ā where a is the constant sequence, an = x for all n.
6.3.2 Riemannian Metrics
Verify that i: X → X̄ is an isometric map and that i (X) is dense in X̄.
∞
5. Verify X̄, d¯ is complete. Hint: if {ā(m)}m=1 is a Cauchy sequence in X̄ This subsection is not completely self contained and may safely be skipped.
choose bm ∈ X such that d¯(i (bm ) , ā(m)) ≤ 1/m. Then show ā(m) → b̄
∞
where b = {bm }m=1 . Lemma 6.21. Suppose that X is a Riemannian (or sub-Riemannian) manifold
and d is the metric on X defined by
d(x, y) = inf {`(σ) : σ(0) = x and σ(1) = y}

6.3 Supplementary Remarks
where `(σ) is the length of the curve σ. We define `(σ) = ∞ if σ is not piecewise
6.3.1 Word of Caution smooth.
Then
Example 6.19. Let (X, d) be a metric space. It is always true that Bx (ε) ⊂ Cx (ε)
since Cx (ε) is a closed set containing Bx (ε). However, it is not always true that Bx (ε) = Cx (ε) and
Bx (ε) = Cx (ε). For example let X = {1, 2} and d(1, 2) = 1, then B1 (1) = {1} , bd(Bx (ε)) = {y ∈ X : d(x, y) = ε}
B1 (1) = {1} while C1 (1) = X. For another counterexample, take
where the boundary operation, bd(·) is defined in Definition 13.29 below.
X = (x, y) ∈ R2 : x = 0 or x = 1

Proof. Let C := Cx (ε) ⊂ Bx (ε) =: B̄. We will show that C ⊂ B̄ by showing
with the usually Euclidean metric coming from the plane. Then B̄ c ⊂ C c . Suppose that y ∈ B̄ c and choose δ > 0 such that By (δ) ∩ B̄ = ∅. In
particular this implies that
B(0,0) (1) = (0, y) ∈ R2 : |y| < 1 ,

B(0,0) (1) = (0, y) ∈ R2 : |y| ≤ 1 , while

By (δ) ∩ Bx (ε) = ∅.
C(0,0) (1) = B(0,0) (1) ∪ {(0, 1)} . We will finish the proof by showing that d(x, y) ≥ ε + δ > ε and hence that
y ∈ C c . This will be accomplished by showing: if d(x, y) < ε + δ then By (δ) ∩
In spite of the above examples, Lemmas 6.20 and 6.21 below shows that for Bx (ε) 6= ∅. If d(x, y) < max(ε, δ) then either x ∈ By (δ) or y ∈ Bx (ε). In either
certain metric spaces of interest it is true that Bx (ε) = Cx (ε). case By (δ) ∩ Bx (ε) 6= ∅. Hence we may assume that max(ε, δ) ≤ d(x, y) < ε + δ.
Let α > 0 be a number such that

6.4 Exercises 33
6.4 Exercises
Exercise 6.9. Let (X, d) be a metric space. Suppose that {xn }∞n=1 ⊂ X is a
sequence and set εn := d(xn , xn+1 ). Show that for m > n that
m−1
X ∞
X
d(xn , xm ) ≤ εk ≤ εk .
k=n k=n
Fig. 6.1. An almost length minimizing curve joining x to y. Conclude from this that if
∞
X ∞
X
εk = d(xn , xn+1 ) < ∞
k=1 n=1
max(ε, δ) ≤ d(x, y) < α < ε + δ
then {xn }∞ ∞
n=1 is Cauchy. Moreover, show that if {xn }n=1 is a convergent se-
and choose a curve σ from x to y such that `(σ) < α. Also choose 0 < δ 0 < δ such quence and x = limn→∞ xn then
that 0 < α − δ 0 < ε which can be done since α − δ < ε. Let k(t) = d(y, σ(t)) a ∞
continuous function on [0, 1] and therefore k([0, 1]) ⊂ R is a connected set which d(x, xn ) ≤
X
εk .
contains 0 and d(x, y). Therefore there exists t0 ∈ [0, 1] such that d(y, σ(t0 )) = k=n
k(t0 ) = δ 0 . Let z = σ(t0 ) ∈ By (δ) then
Exercise 6.10. Show thatP∞ (X, d) is a complete metric space iff every sequence
d(x, z) ≤ `(σ|[0,t0 ] ) = `(σ) − `(σ|[t0 ,1] ) < α − d(z, y) = α − δ 0 < ε {xn }∞
n=1 ⊂ X such that n=1 d(xn , xn+1 ) < ∞ is a convergent sequence in X.
You may find it useful to prove the following statements in the course of the
and therefore z ∈ Bx (ε) ∩ Bx (δ) 6= ∅. proof.
Remark 6.22. Suppose again that X is a Riemannian (or sub-Riemannian) man- If {xn } is Cauchy sequence, then there is a subsequence yj := xnj such that
1. P
∞
ifold and j=1 d(yj+1 , yj ) < ∞.
d(x, y) = inf {`(σ) : σ(0) = x and σ(1) = y} . 2. If {xn }∞
n=1 is Cauchy and there exists a subsequence yj := xnj of {xn } such
that x = limj→∞ yj exists, then limn→∞ xn also exists and is equal to x.
Let σ be a curve from x to y and let ε = `(σ) − d(x, y). Then for all 0 ≤ u <
v ≤ 1, Exercise 6.11. Suppose that f : [0, ∞) → [0, ∞) is a C 2 – function such
that f (0) = 0, f 0 > 0 and f 00 ≤ 0 and (X, ρ) is a metric space. Show that
d(x, y) + ε = `(σ) = `(σ|[0,u] ) + `(σ|[u,v] ) + `(σ|[v,1] ) d(x, y) = f (ρ(x, y)) is a metric on X. In particular show that
≥ d(x, σ(u)) + `(σ|[u,v] ) + d(σ(v), y)
ρ(x, y)
d(x, y) :=
and therefore, using the triangle inequality, 1 + ρ(x, y)
`(σ|[u,v] ) ≤ d(x, y) + ε − d(x, σ(u)) − d(σ(v), y) is a metric on X. (Hint: use calculus to verify that f (a + b) ≤ f (a) + f (b) for
all a, b ∈ [0, ∞).)
≤ d(σ(u), σ(v)) + ε.
∞
Exercise 6.12. Let {(Xn , dn )}n=1 be a sequence of metric spaces, X :=
This leads to the following conclusions. If σ is within ε of a length minimizing
Q ∞ ∞ ∞
n=1 Xn , and for x = (x(n))n=1 and y = (y(n))n=1 in X let
curve from x to y then σ|[u,v] is within ε of a length minimizing curve from σ(u)
∞
to σ(v). In particular if σ is a length minimizing curve from x to y then σ|[u,v] X dn (x(n), y(n))
d(x, y) = 2−n .
is a length minimizing curve from σ(u) to σ(v). 1 + dn (x(n), y(n))
n=1
Show:

1. (X, d) is a metric space,
∞
2. a sequence {xk }k=1 ⊂ X converges to x ∈ X iff xk (n) → x(n) ∈ Xn as
k → ∞ for each n ∈ N and
3. X is complete if Xn is complete for all n.
Exercise 6.13. Suppose (X, ρ) and (Y, d) are metric spaces and A is a dense
subset of X.
1. Show that if F : X → Y and G : X → Y are two continuous functions
such that F = G on A then F = G on X. Hint: consider the set C :=
{x ∈ X : F (x) = G (x)} .
2. Suppose f : A → Y is a function which is uniformly continuous, i.e. for
every ε > 0 there exists a δ > 0 such that
d (f (a) , f (b)) < ε for all a, b ∈ A with ρ (a, b) < δ.
Show there is a unique continuous function F : X → Y such that F = f on

A. Hint: each point x ∈ X is a limit of a sequence consisting of elements
from A.
3. Let X = R = Y and A = Q ⊂ X, find a function f : Q → R which is
continuous on Q but does not extend to a continuous function on R.
7
Banach Spaces
∞
Let (X, k·k) be a normed vector space and d (x, y) := kx − yk be the asso- Proof. Let {fn }n=1 ⊂ `∞ (X) be a Cauchy sequence. Since for any x ∈ X,
∞
ciated metric on X. We say {xn }n=1 ⊂ X converges to x ∈ X (and write we have
limn→∞ xn = x or xn → x) if |fn (x) − fm (x)| ≤ kfn − fm k∞ (7.1)
∞
0 = lim d (x, xn ) = lim kx − xn k . which shows that {fn (x)}n=1 ⊂ F is a Cauchy sequence of numbers. Because F
n→∞ n→∞ (F = R or C) is complete, f (x) := limn→∞ fn (x) exists for all x ∈ X. Passing
∞ to the limit n → ∞ in Eq. (7.1) implies
Similarly {xn }n=1 ⊂ X is said to be a Cauchy sequence if
|f (x) − fm (x)| ≤ lim inf kfn − fm k∞
0= lim d (xm , xn ) = lim kxm − xn k . n→∞
m,n→∞ m,n→∞
and taking the supremum over x ∈ X of this inequality implies
Definition 7.1 (Banach space). A normed vector space (X, k·k) is a Banach
space if the associated metric space (X, d) is complete, i.e. all Cauchy sequences kf − fm k∞ ≤ lim inf kfn − fm k∞ → 0 as m → ∞
n→∞
are convergent.
∞
Remark 7.2. Since kxk = d (x, 0) , it follows from Lemma 6.6 that k·k is a con- showing fm → f in `∞ (X). For the second assertion, suppose that {fn }n=1 ⊂
tinuous function on X and that BC(X) ⊂ `∞ (X) and fn → f ∈ `∞ (X). We must show that f ∈ BC(X), i.e.
that f is continuous. To this end let x, y ∈ X, then
|kxk − kyk| ≤ kx − yk for all x, y ∈ X.
|f (x) − f (y)| ≤ |f (x) − fn (x)| + |fn (x) − fn (y)| + |fn (y) − f (y)|
It is also easily seen that the vector addition and scalar multiplication are ≤ 2 kf − fn k∞ + |fn (x) − fn (y)| .
continuous on any normed space as the reader is asked to verify in Exercise 7.5.
These facts will often be used in the sequel without further mention. Thus if ε > 0, we may choose n large so that 2 kf − fn k∞ < ε/2 and then for this
n there exists an open neighborhood Vx of x ∈ X such that |fn (x) − fn (y)| <
ε/2 for y ∈ Vx . Thus |f (x) − f (y)| < ε for y ∈ Vx showing the limiting function
7.1 Examples f is continuous.
Here is an application of this theorem.
Lemma 7.3. Suppose that X is a set then the bounded functions, `∞ (X), on
Theorem 7.4 (Metric Space Tietze Extension Theorem). Let (X, d) be a
X is a Banach space with the norm
metric space, D be a closed subset of X, −∞ < a < b < ∞ and f ∈ C(D, [a, b]).
kf k = kf k∞ = sup |f (x)| . (Here we are viewing D as a metric space with metric dD := d|D×D .) Then
x∈X there exists F ∈ C(X, [a, b]) such that F |D = f.
Moreover if X is a metric space (more generally a topological space, see Chapter Proof.
13) the set BC(X) ⊂ `∞ (X) = B(X) is closed subspace of `∞ (X) and hence is −1
1. By scaling and translation (i.e. by replacing f by (b − a) (f − a)), it suf-
also a Banach space.
fices to prove Theorem 7.4 with a = 0 and b = 1.
36 7 Banach Spaces
2. Suppose α ∈ (0, 1] and f : D → [0, α] is continuous function. Let A := Theorem 7.5 (Completeness of `p (µ)). Let X be a set and µ : X → (0, ∞)
f −1 ([0, 13 α]) and B := f −1 ([ 23 α, α]). By Lemma 6.15 there exists a function be a given function. Then for any p ∈ [1, ∞], (`p (µ), k·kp ) is a Banach space.
g̃ ∈ C(X, [0, α3 ]) such that g̃ = 0 on A and g̃ = 1 on B. Letting g := α3 g̃, we
have g ∈ C(X, [0, α3 ]) such that g = 0 on A and g = α3 on B. Further notice Proof. We have already proved this for p = ∞ in Lemma 7.3 so we now
∞
that assume that p ∈ [1, ∞). Let {fn }n=1 ⊂ `p (µ) be a Cauchy sequence. Since for
2 any x ∈ X,
0 ≤ f (x) − g(x) ≤ α for all x ∈ D.
3 1
3. Now suppose f : D → [0, 1] is a continuous function as in step 1. Let |fn (x) − fm (x)| ≤ kfn − fm kp → 0 as m, n → ∞
µ(x)
g1 ∈ C(X, [0, 1/3]) be as in step 2, see Figure 7.1. with α = 1 and let
∞
f1 := f − g1 |D ∈ C(D, [0, 2/3]). Apply step 2. with α = 2/3 and f = f1 to it follows that {fn (x)}n=1 is a Cauchy sequence of numbers and f (x) :=
2
find g2 ∈ C(X, [0, 31 23 ]) such that f2 := f − (g1 + g2 ) |D ∈ C(D, [0, 32 ]). limn→∞ fn (x) exists for all x ∈ X. By Fatou’s Lemma,
n−1
Continue this way inductively to find gn ∈ C(X, [0, 31 23 ]) such that X X
kfn − f kpp = µ · lim inf |fn − fm |p ≤ lim inf µ · |fn − fm |p
m→∞ m→∞
N
" #!
N X X
2
fm kpp
X
f− gn |D =: fN ∈ C D, 0, . (7.2) = lim inf kfn − → 0 as n → ∞.
3 m→∞
n=1
P∞ This then shows that f = (f − fn ) + fn ∈ `p (µ) (being the sum of two `p –
4. Define F := n=1 gn . Since `p
functions) and that fn −→ f.
∞ ∞ n−1
X X 1 2 1 1 Remark 7.6. Let X be a set, Y be a Banach space and `∞ (X, Y ) denote the
kgn k∞ ≤ = 2 = 1,
n=1 n=1
3 3 31− 3 bounded functions f : X → Y equipped with the norm
the series defining F is uniformly convergent so F ∈ C(X, [0, 1]). Passing to kf k = kf k∞ = sup kf (x)kY .
x∈X
the limit in Eq. (7.2) shows f = F |D .
If X is a metric space (or a general topological space, see Chapter 13), let
BC(X, Y ) denote those f ∈ `∞ (X, Y ) which are continuous. The same proof
used in Lemma 7.3 shows that `∞ (X, Y ) is a Banach space and that BC(X, Y )
is a closed subspace of `∞ (X, Y ). Similarly, if 1 ≤ p < ∞ we may define
 !1/p 
 X 
p
`p (X, Y ) = f : X → Y : kf kp = kf (x)kY <∞ .
 
x∈X

The same proof as in Theorem 7.5 would then show that `p (X, Y ) , k·kp is a
Banach space.
7.2 Bounded Linear Operators Basics

Fig. 7.1. Reducing f by subtracting off a globally defined function g1 ∈ C X, [0, 13 ] .

Definition 7.7. Let X and Y be normed spaces and T : X → Y be a linear

map. Then T is said to be bounded provided there exists C < ∞ such that
kT (x)kY ≤ CkxkX for all x ∈ X. We denote the best constant by kT kop =
kT kL(X,Y ) , i.e.

7.2 Bounded Linear Operators Basics 37
kT (x)kY Notation 7.9 Let L(X, Y ) denote the bounded linear operators from X to Y
kT kL(X,Y ) = sup = sup {kT (x)kY : kxkX = 1} .
x6=0 kxkX x6=0 and L (X) = L (X, X) . If Y = F we write X ∗ for L(X, F) and call X ∗ the
(continuous) dual space to X.
The number kT kL(X,Y ) is called the operator norm of T.
Lemma 7.10. Let X, Y be normed spaces, then the operator norm k·k on
In the future, we will usually drop the garnishing on the norms and sim- L(X, Y ) is a norm. Moreover if Z is another normed space and T : X → Y
ply write kxkX as kxk, kT kL(X,Y ) as kT k , etc. The reader should be able to and S : Y → Z are linear maps, then kST k ≤ kSkkT k, where ST := S ◦ T.
determine the norm that is to be used by context.
Proof. As usual, the main point in checking the operator norm is a norm
Proposition 7.8. Suppose that X and Y are normed spaces and T : X → Y is to verify the triangle inequality, the other axioms being easy to check. If
is a linear map. The the following are equivalent: A, B ∈ L(X, Y ) then the triangle inequality is verified as follows:
1. T is continuous. kAx + Bxk kAxk + kBxk
2. T is continuous at 0. kA + Bk = sup ≤ sup
x6=0 kxk x6=0 kxk
3. T is bounded.
kAxk kBxk
≤ sup + sup = kAk + kBk .
Proof. 1. ⇒ 2. trivial. 2. ⇒ 3. If T continuous at 0 then there exist δ > 0 such x6=0 kxk x6=0 kxk
that kT (x)k ≤ 1 if kxk ≤ δ. Therefore for any nonzero x ∈ X, kT (δx/kxk)k ≤ 1
which implies that kT (x)k ≤ 1δ kxk and hence kT k ≤ 1δ < ∞. 3. ⇒ 1. Let x ∈ X For the second assertion, we have for x ∈ X, that
and ε > 0 be given. Then
kST xk ≤ kSkkT xk ≤ kSkkT kkxk.
kT y − T xk = kT (y − x)k ≤ kT k ky − xk < ε From this inequality and the definition of kST k, it follows that kST k ≤ kSkkT k.
provided ky − xk < ε/kT k := δ.
The reader is asked to prove the following continuity lemma in Exercise 7.9.
For the next three exercises, let X = Rn and Y = Rm and T : X → Y be a
linear transformation so that T is given by matrix multiplication by an m × n Lemma 7.11. Let X, Y and Z be normed spaces. Then the maps
matrix. Let us identify the linear transformation T with this matrix.
(S, x) ∈ L(X, Y ) × X −→ Sx ∈ Y
Assume the norms on X and Y are the `1 – norms, i.e. for
Exercise 7.1. P
n n
x ∈ R , kxk = j=1 |xj | . Then the operator norm of T is given by and
(S, T ) ∈ L(X, Y ) × L(Y, Z) −→ ST ∈ L(X, Z)
m
X
kT k = max |Tij | . are continuous relative to the norms
1≤j≤n
i=1
k(S, x)kL(X,Y )×X := kSkL(X,Y ) + kxkX and
Exercise 7.2. Assume the norms on X and Y are the `∞ – norms, i.e. for
k(S, T )kL(X,Y )×L(Y,Z) := kSkL(X,Y ) + kT kL(Y,Z)
x ∈ Rn , kxk = max1≤j≤n |xj | . Then the operator norm of T is given by
n
X on L(X, Y ) × X and L(X, Y ) × L(Y, Z) respectively.
kT k = max |Tij | .
1≤i≤m
j=1
Proposition 7.12. Suppose that X is a normed vector space and Y is a Banach
space. Then (L(X, Y ), k · kop ) is a Banach space. In particular the dual space
Exercise 7.3. Assume norms on X and Y are the `2 – norms, i.e. for X ∗ is always a Banach space.
n 2 Pn the2 2
x ∈ R , kxk = j=1 xj . Show kT k is the largest eigenvalue of the matrix ∞
Proof. Let {Tn }n=1 be a Cauchy sequence in L(X, Y ). Then for each x ∈ X,
T tr T : Rn → Rn . Hint: Use the spectral theorem for orthogonal matrices.
kTn x − Tm xk ≤ kTn − Tm k kxk → 0 as m, n → ∞

38 7 Banach Spaces
∞ N ∞
showing {Tn x}n=1 is Cauchy in Y. Using the completeness of Y, there exists an X X
element T x ∈ Y such that yN +1 − y1 = (yn+1 − yn ) → s = (yn+1 − yn ) ∈ X as N → ∞.
n=1 n=1
lim kTn x − T xk = 0. ∞
n→∞ This shows that limN →∞ yN exists and is equal to x := y1 + s. Since {xn }n=1
The map T : X → Y is linear map, since for x, x0 ∈ X and λ ∈ F we have is Cauchy,
T (x + λx0 ) = lim Tn (x + λx0 ) = lim [Tn x + λTn x0 ] = T x + λT x0 , kx − xn k ≤ kx − yk k + kyk − xn k → 0 as k, n → ∞

n→∞ n→∞
wherein we have used the continuity of the vector space operations in the last showing that limn→∞ xn exists and is equal to x.
equality. Moreover,
Example 7.14. Here is another proof of Proposition 7.12 which makes use of
kT x − Tn xk ≤ kT x − Tm xk + kTm x − Tn xk ≤ kT x − Tm xk + kTm − Tn k kxk Theorem 7.13. Suppose that Tn ∈ L(X, Y ) is a sequence of operators such that
P∞
kTn k < ∞. Then
and therefore n=1
kT x − Tn xk ≤ lim inf (kT x − Tm xk + kTm − Tn k kxk) ∞ ∞

m→∞
X X
kTn xk ≤ kTn k kxk < ∞
= kxk · lim inf kTm − Tn k . n=1 n=1
m→∞
Hence ∞
P
and therefore by the completeness of Y, Sx := Tn x = limN →∞ SN x exists
kT − Tn k ≤ lim inf kTm − Tn k → 0 as n → ∞. n=1
m→∞
N
P
Thus we have shown that Tn → T in L(X, Y ) as desired. in Y, where SN := Tn . The reader should check that S : X → Y so defined
The following characterization of a Banach space will sometimes be useful n=1
is linear. Since,
in the sequel.
N ∞
Theorem 7.13. A normed space (X, k · k) is a Banach space iff every sequence X X
∞ N kSxk = lim kSN xk ≤ lim kTn xk ≤ kTn k kxk ,
∞ P P N →∞ N →∞
{xn }n=1 ⊂ X such that kxn k < ∞ implies limN →∞ xn = s exists in X n=1 n=1
n=1 n=1
(that is to say every absolutely convergent series is a convergent series in X.) S is bounded and
∞ ∞
X
P
As usual we will denote s by xn . kSk ≤ kTn k. (7.3)
n=1 n=1
Proof. (This is very similar to Exercise 6.10.) (⇒) If X is complete and Similarly,
∞
P N
P
kxn k < ∞ then sequence sN := xn for N ∈ N is Cauchy because (for kSx − SM xk = lim kSN x − SM xk
n=1 n=1 N →∞
N > M) N ∞
N X X
X ≤ lim kTn k kxk = kTn k kxk
ksN − sM k ≤ kxn k → 0 as M, N → ∞. N →∞
n=M +1 n=M +1
n=M +1
∞
P N
P and therefore,
Therefore s = xn := limN →∞ xn exists in X. (⇐=) Suppose that ∞
X
∞
n=1 n=1 kS − SM k ≤ kTn k → 0 as M → ∞.
{xn }n=1 is a Cauchy sequence and let {yk = xnk }∞
k=1 be a subsequence of n=M
∞
∞ P
{xn }n=1 such that kyn+1 − yn k < ∞. By assumption
n=1

7.2 Bounded Linear Operators Basics 39
For the remainder of this section let X be an infinite set, µ : X → (0, ∞) 1. By Holder’s inequality,
be a given function and p, q ∈ [1, ∞] such that q = p/ (p − 1) . It will also be X
convenient to define δx : X → R for x ∈ X by |f (x)| |g (x)| µ (x) ≤ kf kq kgkp
x∈X
1 if y = x
δx (y) = which shows that φf is well defined. The φf : `p (µ) → F is linear by the
0 if y 6= x.
linearity of sums and since
Notation 7.15 Let c0 (X) denote those functions f ∈ `∞ (X) which “vanish at
X

X
∞,” i.e. for every ε > 0 there exists a finite subset Λε ⊂ X such that |f (x)| < ε |φf (g)| =

f (x) g (x) µ (x) ≤

|f (x)| |g (x)| µ (x) ≤ kf kq kgkp ,
whenever x ∈ / Λε . Also let cf (X) denote those functions f : X → F with finite
x∈X

x∈X
support, i.e.
we learn that
cf (X) := {f ∈ `∞ (X) : # ({x ∈ X : f (x) 6= 0}) < ∞} . kφf k`p (µ)∗ ≤ kf kq . (7.5)
∗
Exercise 7.4. Therefore φf ∈ `p (µ) .
Show cf (X) is a dense subspace of the Banach spaces
2. The map φ in Eq. (7.4) is linear in f by the linearity properties of infinite

p
` (µ) , k·kp for 1 ≤ p < ∞, while the closure of cf (X) inside the Banach q−1
sums. For p ∈ (1, ∞) , define g (x) = sgn(f (x)) |f (x)| where
space, (`∞ (X) , k·k∞ ) is c0 (X) . Note from this it follows that c0 (X) is a closed
subspace of `∞ (X) . (See Proposition 15.23 below where this last assertion is
z
if z 6= 0
proved in a more general context.) sgn(z) := |z|
0 if z = 0.
Theorem 7.16. Let X be any set, µ : X → (0, ∞) be a function, p ∈ [1, ∞], Then
q := p/ (p − 1) be the conjugate exponent and for f ∈ `q (µ) define φf : `p (µ) → p
|f (x)|( p−1 ) µ (x)
−1 p
p
X (q−1)p
X
F by X kgkp = |f (x)| µ (x) =
φf (g) := f (x) g (x) µ (x) . x∈X x∈X
X q q
x∈X = |f (x)| µ (x) = kf kq
Then x∈X
∗ and
1. φf (g) is well defined and φf ∈ `p (µ) .
2. The map X q−1
X q−1
φ ∗ φf (g) = f (x) sgn(f (x)) |f (x)| µ (x) = |f (x)| |f (x)| µ (x)
f ∈ `q (µ) → φf ∈ `p (µ) (7.4) x∈X x∈X
q
= kf kq( q )
1 1
is an isometric linear map of Banach spaces. q +p
= kf kq kf kqp = kf kq kgkp .
∗
3. If p ∈ [1, ∞), then the map in Eq. (7.4) is also surjective and hence, `p (µ)
q
is isometrically isomorphic to ` (µ) . Hence kφf k`p (µ)∗ ≥ kf kq which combined with Eq. (7.5) shows kφf k`p (µ)∗ =
4. When p = ∞, the map kf kq . For p = ∞, let g (x) = sgn(f (x)), then kgk∞ = 1 and
1 ∗
f ∈ ` (µ) → φf ∈ c0 (X) X
|φf (g)| = f (x) sgn(f (x))µ (x)
is an isometric and surjective, i.e. `1 (µ) is isometrically isomorphic to x∈X
∗ X
c0 (X) . = |f (x)| µ (x) = kf k1 kgk∞
x∈X
(See Theorem 25.13 below for a continuation of this theorem.)
which shows kφf k`∞ (µ)∗ ≥ kf k`1 (µ) . Combining this with Eq. (7.5) shows
Proof.
kφf k`∞ (µ)∗ = kf k`1 (µ) . For p = 1,

40 7 Banach Spaces
q
|φf (δx )| = µ (x) |f (x)| = |f (x)| kδx k1 and solving this inequality for kf k`q (Λ,µ) (using q − q/p = 1) implies
kf k`q (Λ,µ) ≤ kλk`p (µ)∗ Taking the supremum of this inequality on Λ ⊂⊂ X
and therefore kφf k`1 (µ)∗ ≥ |f (x)| for all x ∈ X. Hence kφf k`1 (µ)∗ ≥ kf k∞
shows kf k`q (µ) ≤ kλk`p (µ)∗ , i.e. f ∈ `q (µ) . Since λ = φf agree on cf (X)
which combined with Eq. (7.5) shows kφf k`1 (µ)∗ = kf k∞ .
∗ ∗ and cf (X) is a dense subspace of `p (µ) for p < ∞ and cf (X) is dense
3. and 4. Suppose that p ∈ [1, ∞) and λ ∈ `p (µ) or p = ∞ and λ ∈ c0 (X) . subspace of c0 (X) when p = ∞, it follows that λ = φf .
We wish to find f ∈ `q (µ) such that λ = φf . If such an f exists, then
λ (δx ) = f (x) µ (x) and so we must define f (x) := λ (δx ) /µ (x) . As a
preliminary estimate,
|λ (δx )| kλk`p (µ)∗ kδx k`p (µ) 7.3 General Sums in Banach Spaces
|f (x)| = ≤
µ (x) µ (x)
1
Definition 7.17. Suppose X is a normed space.
kλk`p (µ)∗ [µ (x)] p − q1 ∞ P∞
= = kλk`p (µ)∗ [µ (x)] . 1. Suppose that {xn }n=1 is a sequence in X, then we say n=1 xn converges
µ (x) P∞
in X and n=1 xn = s if
When p = 1 and q = ∞, this implies kf k∞ ≤ kλk`1 (µ)∗ < ∞. If p ∈ (1, ∞] N
X
and Λ ⊂⊂ X, then lim xn = s in X.
N →∞
X X n=1
q q q−1
kf k`q (Λ,µ) := |f (x)| µ (x) = f (x) sgn(f (x)) |f (x)| µ (x)
x∈Λ x∈Λ P that {xα : α ∈ A} is a given collection of P
2. Suppose vectors in X. We say the
sum x
α∈A α converges in X and write s = x ∈ X if for all
X λ (δx ) q−1 P α
α∈A
= sgn(f (x)) |f (x)| µ (x) ε > 0 there exists a finite set Γε ⊂ A such that s − α∈Λ xα < ε for any
µ (x)
x∈Λ Λ ⊂⊂ A such that Γε ⊂ Λ.
X q−1
= λ (δx ) sgn(f (x)) |f (x)|
P
P Warning: As usual if X is a Banach space and α∈A kxα k < ∞ then
α∈A xα exists in X, see Exercise 7.13. However, unlike the case of real valued
x∈Λ
! P P
X q−1
sums the existence of α∈A xα does not imply α∈Λ kxα k < ∞. See Propo-
=λ sgn(f (x)) |f (x)| δx sition 8.19 below, from which one may manufacture counter-examples to this
x∈Λ false premise.

X
≤ kλk`p (µ)∗

sgn(f (x)) |f (x)|
q−1
δx .
Lemma 7.18. Suppose that {xα ∈ X : α ∈ A} is a given collection of vectors

x∈Λ
in a normed space, X.
p
P
1. If s = α∈A xα ∈ X Pexists and T : X → Y is a bounded linear map between
Since
normed spaces, then α∈A T xα exists in Y and
!1/p
X X X X
q−1 (q−1)p

sgn(f (x)) |f (x)| δx =

|f (x)| µ (x) Ts = T xα = T xα .
α∈A α∈A

x∈Λ p x∈Λ
P
!1/p 2. If s = α∈A xα exists in X then for every ε > 0 there exists Γε ⊂⊂ A such
q/p
X q P
= |f (x)| µ (x) = kf k`q (Λ,µ) that Pα∈Λ xα < ε for all Λ ⊂⊂ A \ Γε .

x∈Λ 3. If s = α∈A xα exists in X, the set Γ := {α ∈ A : xa 6= 0} is at most
∞
1/∞
countable. Moreover if Γ is infinite and {αn }n=1 is an enumeration of Γ,
which is also valid for p = ∞ provided kf k`1 (Λ,µ) := 1. Combining the last then
two displayed equations shows X∞ XN
s= xαn := lim xαn . (7.6)
q q/p N →∞
kf k`q (Λ,µ) ≤ kλk`p (µ)∗ kf k`q (Λ,µ) n=1 n=1

7.4 Inverting Elements in L(X) 41
∞
4. If we further assume that X is a P
Banach space
and suppose for all ε > 0 Therefore {sn }n=1is Cauchy and hence convergent in X, because X is a
exists Γε ⊂⊂ A such that α∈Λ xα < ε whenever Λ ⊂⊂ A \ Γε ,
there P Banach space. Let s := limn→∞ sn . Then for Λ ⊂⊂ A such that γn ⊂ Λ, we
then α∈A xα exists in X. have

Proof. X X 1
s − xα ≤ ks − sn k + ≤ ks − sn k + n .
xα

1. Let Γε be as in Definition 7.17 and Λ ⊂⊂ A such that Γε ⊂ Λ. Then

α∈Λ α∈Λ\γn
Since the right side of this equation goes to zero as n → ∞, it follows that

X X
T s − T xα ≤ kT k s − xα < kT k ε
P
α∈A xα exists and is equal to s.

α∈Λ α∈Λ
P
which shows that P α∈Λ T xα exists and is equal to T s.
2. Suppose that s = α∈A xα exists and ε > 0. Let Γε ⊂⊂ A be as in
Definition 7.17. Then for Λ ⊂⊂ A \ Γε , 7.4 Inverting Elements in L(X)
Definition 7.19. A linear map T : X → Y is an isometry if kT xkY = kxkX

X X X
xα = xα − xα for all x ∈ X. T is said to be invertible if T is a bijection and T −1 is bounded.

α∈Λ α∈Γε ∪Λ α∈Γε

X

X

Notation 7.20 We will write GL(X, Y ) for those T ∈ L(X, Y ) which are
≤

xα − s +

xα − s < 2ε.
invertible. If X = Y we simply write L(X) and GL(X) for L(X, X) and

α∈Γε ∪Λ

α∈Γε
GL(X, X) respectively.
P Proposition 7.21. Suppose X is a Banach space and Λ ∈ L(X) := L(X, X)
3. If s = P in X, for each n ∈ N there exists a finite subset
α∈A xα exists ∞
kΛn k < ∞. Then I − Λ is invertible and
P
Γn ⊂ A such that α∈Λ xα < n1 for all Λ ⊂⊂ A \ Γn . Without loss of satisfies
n=0
generality we may assume xα 6= 0 for all α ∈ Γn . Let Γ∞ := ∪∞
n=1 Γn – a
∞ ∞
countable subset of A. Then for any β ∈/ Γ∞ , we have {β} ∩ Γn = ∅ and 1 X X
(I − Λ)−1 = “ Λn and (I − Λ)−1 ≤ kΛn k.

therefore ”=

X

I −Λ n=0 n=0
1
kxβ k = xα ≤
→ 0 as n → ∞. In particular if kΛk < 1 then the above formula holds and

α∈{β} n
∞ (I − Λ)−1 ≤
1
Let {αn }n=1 be an enumeration of Γ and define γN := {αn : 1 ≤ n ≤ N } . .
1 − kΛk
Since for any M ∈ N, γN will eventually contain ΓM for N sufficiently large,
∞
we have Proof. Since L(X) is a Banach space and
P
kΛn k < ∞, it follows from
N
X 1 n=0
lim sup s − xαn ≤ → 0 as M → ∞.

M Theorem 7.13 that
N →∞
n=1
N
X
S := lim SN := lim Λn
Therefore Eq. (7.6) holds. N →∞ N →∞
P n=0
4. For n ∈ N, let Γn ⊂⊂ A such that P α∈Λ xα < n1 for all Λ ⊂⊂ A \ Γn .
Define γn := ∪nk=1 Γk ⊂ A and sn := α∈γn xα . Then for m > n, exists in L(X). Moreover, by Lemma 7.11,
(I − Λ) S = (I − Λ) lim SN = lim (I − Λ) SN
N →∞ N →∞
X
ksm − sn k = N
xα ≤ 1/n → 0 as m, n → ∞.
X
= lim (I − Λ) Λn = lim (I − ΛN +1 ) = I

α∈γm \γn
N →∞ N →∞
n=0

42 7 Banach Spaces
and similarly S (I − Λ) = I. This shows that (I − Λ)−1 exists and is equal to Therefore I − Λ is invertible and hence so is B (being the product of invertible
S. Moreover, (I − Λ)−1 is bounded because elements) with
∞ −1 −1
B −1 = (IX − Λ)−1 A−1 = IX − A−1 (A − B))

(I − Λ)−1 = kSk ≤
X A .
kΛn k.

n=0
Taking norms of the previous equation gives
n n
If we further assume kΛk < 1, then kΛ k ≤ kΛk and −1 1
B ≤ (IX − Λ)−1 kA−1 k ≤ kA−1 k

∞
X ∞
X 1 1 − kΛk
n
kΛn k ≤ kΛk = < ∞. kA−1 k
n=0 n=0
1 − kΛk ≤
1 − kA−1 k kA − Bk
which is the bound in Eq. (7.9). The bound in Eq. (7.10) holds because
Corollary 7.22. Let X and Y be Banach spaces. Then GL(X, Y ) is an open −1
B − A−1 = B −1 (A − B) A−1 ≤ B −1 A−1 kA − Bk

(possibly empty) subset of L(X, Y ). More specifically, if A ∈ GL(X, Y ) and
B ∈ L(X, Y ) satisfies kA−1 k2 kA − Bk
kB − Ak < kA−1 k−1 (7.7) ≤
1 − kA−1 k kA − Bk
.
then B ∈ GL(X, Y )
∞
X n For an application of these results to linear ordinary differential equations,
B −1 = IX − A−1 B A−1 ∈ L(Y, X),

(7.8) see Section 10.3.
n=0
B ≤ kA−1 k
−1 1
(7.9) 7.5 Exercises
1− kA−1 k kA − Bk
and Exercise 7.5. Let (X, k·k) be a normed space over F (R or C). Show the map
−1
B − A−1 ≤
kA−1 k2 kA − Bk
. (7.10)
1 − kA−1 k kA − Bk (λ, x, y) ∈ F × X × X → x + λy ∈ X
In particular the map
is continuous relative to the norm on F × X × X defined by
−1
A ∈ GL(X, Y ) → A ∈ GL(Y, X) (7.11)
k(λ, x, y)kF×X×X := |λ| + kxk + kyk .
is continuous.
(See Exercise 13.25 for more on the metric associated to this norm.) Also show
Proof. Let A and B be as above, then that k·k : X → [0, ∞) is continuous.
B = A − (A − B) = A IX − A−1 (A − B)) = A(IX − Λ)

Exercise 7.6. Let X = N and for p, q ∈ [1, ∞) let k·kp denote the `p (N) –
norm. Show k·kp and k·kq are inequivalent norms for p 6= q by showing
where Λ : X → X is given by
kf kp
Λ := A−1 (A − B) = IX − A−1 B. sup = ∞ if p < q.
f 6=0 kf kq
Now
Exercise 7.7. Suppose that (X, k·k) is a normed space and S ⊂ X is a linear
kΛk = A−1 (A − B)) ≤ kA−1 k kA − Bk < kA−1 kkA−1 k−1 = 1.

subspace.

7.5 Exercises 43
1. Show the closure S̄ of S is also a linear subspace. Exercise 7.13. SupposeP{xα ∈ X : α ∈ A} is a given collection of vectors in a
2. Now suppose that X is a Banach space. Show that S with the inherited Banach space X. Show α∈A xα exists in X and
norm from X is a Banach space iff S is closed.
X X
Exercise 7.8. Folland Problem 5.9. Showing C k ([0, 1]) is a Banach space. xα ≤ kxα k

α∈A α∈A
Exercise 7.9. Suppose that X, Y and Z are Banach spaces and Q : X ×Y → Z P
is a bilinear form, i.e. we are assuming x ∈ X → Q (x, y) ∈ Z is linear for each if α∈A kxα k < ∞. That is to say “absolute convergence” implies conver-
y ∈ Y and y ∈ Y → Q (x, y) ∈ Z is linear for each x ∈ X. Show Q is continuous gence in a Banach space.
relative to the product norm, k(x, y)kX×Y := kxkX + kykY , on X × Y iff there
is a constant M < ∞ such that Exercise 7.14. Suppose X is a Banach space and {fn : n ∈ N} is a sequence
PN
in X such that limn→∞ fn = f ∈ X. Show sN := N1 n=1 fn for N ∈ N is still
kQ (x, y)kZ ≤ M kxkX · kykY for all (x, y) ∈ X × Y. (7.12) a convergent sequence and
N
Then apply this result to prove Lemma 7.11. 1 X
lim fn = lim sN = f.
N →∞ N N →∞
Exercise 7.10. Let d : C(R) × C(R) → [0, ∞) be defined by n=1
∞ Exercise 7.15 (Dominated Convergence Theorem Again). Let X be a

X kf − gkn
d(f, g) = 2−n , Banach space, A be a set and suppose fn : A → X is a sequence of functions
n=1
1 + kf − gkn such that f (α) := limn→∞ fn (α) exists for all α ∈ A. Further assume there
exists a summable function g : A → [0, ∞) such that kfn (α)k ≤ g (α) for all
where kf kn := sup{|f (x)| : |x| ≤ n} = max{|f (x)| : |x| ≤ n}. P
α ∈ A. Show α∈A f (α) exists in X and
1. Show that d is a metric on C(R). X X
2. Show that a sequence {fn }∞ n=1 ⊂ C(R) converges to f ∈ C(R) as n → ∞
lim fn (α) = f (α) .
n→∞
iff fn converges to f uniformly on bounded subsets of R. α∈A α∈A
3. Show that (C(R), d) is a complete metric space.
Exercise 7.11. Let X = C([0, 1], R) and for f ∈ X, let

Z 1
kf k1 := |f (t)| dt.
0
Show that (X, k·k1 ) is normed space and show by example that this space is not
∞
complete. Hint: For the last assertion find a sequence of {fn }n=1 ⊂ X which is
“trying” to converge to the function f = 1[ 12 ,1] ∈
/ X.
Exercise 7.12. Let (X, k·k1 ) be the normed space in Exercise 7.11. Compute
the closure of A when
1. A = {f
n ∈ X : f (1/2) = 0} . o
2. A = f ∈ X : supt∈[0,1] f (t) ≤ 5 .
n R 1/2 o
3. A = f ∈ X : 0 f (t) dt = 0 .

8
Hilbert Space Basics
Definition 8.1. Let H be a complex vector space. An inner product on H is a

function, h·|·i : H × H → C, such that
1. hax + by|zi = ahx|zi + bhy|zi i.e. x → hx|zi is linear.
2. hx|yi = hy|xi.
3. kxk2 := hx|xi ≥ 0 with equality kxk2 = 0 iff x = 0.
Notice that combining properties (1) and (2) that x → hz|xi is conjugate
linear for fixed z ∈ H, i.e.
hz|ax + byi = āhz|xi + b̄hz|yi.
The following identity will be used frequently in the sequel without further Fig. 8.1. The picture behind the proof of the Schwarz inequality.
mention,
kx + yk2 = hx + y|x + yi = kxk2 + kyk2 + hx|yi + hy|xi

p
Corollary 8.3. Let (H, h·|·i) be an inner product space and kxk := hx|xi.
= kxk2 + kyk2 + 2Rehx|yi. (8.1) Then the Hilbertian norm, k · k, is a norm on H. Moreover h·|·i is continuous
on H × H, where H is viewed as the normed space (H, k·k).
Theorem 8.2 (Schwarz Inequality). Let (H, h·|·i) be an inner product space,
then for all x, y ∈ H Proof. If x, y ∈ H, then, using Schwarz’s inequality,
|hx|yi| ≤ kxkkyk
kx + yk2 = kxk2 + kyk2 + 2Rehx|yi
and equality holds iff x and y are linearly dependent.
≤ kxk2 + kyk2 + 2kxkkyk = (kxk + kyk)2 .
Proof. If y = 0, the result holds trivially. So assume that y 6= 0 and observe;
2
if x = αy for some α ∈ C, then hx|yi = α kyk and hence Taking the square root of this inequality shows k·k satisfies the triangle inequal-
ity.
2 Checking that k·k satisfies the remaining axioms of a norm is now routine
|hx|yi| = |α| kyk = kxkkyk.
and will be left to the reader. If x, x0 , y, y 0 ∈ H, then
Now suppose that x ∈ H is arbitrary, let z := x − kyk−2 hx|yiy. (So z is the
“orthogonal projection” of x onto y, see Figure 8.1.) Then |hx|yi − hx0 |y 0 i| = |hx − x0 |yi + hx0 |y − y 0 i|
2 2
≤ kykkx − x0 k + kx0 kky − y 0 k
hx|yi = kxk2 + |hx|yi| kyk2 − 2Rehx| hx|yi yi
0 ≤ kzk2 = ≤ kykkx − x0 k + (kxk + kx − x0 k) ky − y 0 k

x − 2
y 4
kyk kyk kyk2
= kykkx − x0 k + kxkky − y 0 k + kx − x0 kky − y 0 k
|hx|yi|2
= kxk2 −
kyk2 from which it follows that h·|·i is continuous.
from which it follows that 0 ≤ kyk2 kxk2 − |hx|yi|2 with equality iff z = 0 or Definition 8.4. Let (H, h·|·i) be an inner product space, we say x, y ∈ H are
equivalently iff x = kyk−2 hx|yiy. orthogonal and write x ⊥ y iff hx|yi = 0. More generally if A ⊂ H is a set,
46 8 Hilbert Space Basics
x ∈ H is orthogonal to A (write x ⊥ A) iff hx|yi = 0 for all y ∈ A. Let Example 8.8. Suppose X is a set and µ : X → (0, ∞) , then H := `2 (µ) is a
A⊥ = {x ∈ H : x ⊥ A} be the set of vectors orthogonal to A. A subset S ⊂ H Hilbert space when equipped with the inner product,
is an orthogonal set if x ⊥ y for all distinct elements x, y ∈ S. If S further X
satisfies, kxk = 1 for all x ∈ S, then S is said to be an orthonormal set. hf |gi := f (x) ḡ (x) µ (x) .
x∈X
Proposition 8.5. Let (H, h·|·i) be an inner product space then
In Exercise 8.7 you will show every Hilbert space H is “equivalent” to a Hilbert
1. (Parallelogram Law) space of this form with µ ≡ 1.
kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 (8.2) More examples of Hilbert spaces will be given later after we develop the
Lebesgue integral, see Example 23.1 below.
for all x, y ∈ H.
2. (Pythagorean Theorem) If S ⊂⊂ H is a finite orthogonal set, then Definition 8.9. A subset C of a vector space X is said to be convex if for all
x, y ∈ C the line segment [x, y] := {tx + (1 − t)y : 0 ≤ t ≤ 1} joining x to y is
X 2 X contained in C as well. (Notice that any vector subspace of X is convex.)

x = kxk2 . (8.3)

x∈S

x∈S Theorem 8.10 (Best Approximation Theorem). Suppose that H is a
Hilbert space and M ⊂ H is a closed convex subset of H. Then for any x ∈ H
3. If A ⊂ H is a set, then A⊥ is a closed linear subspace of H. there exists a unique y ∈ M such that
Remark 8.6. See Proposition 8.48 for the “converse” of the parallelogram law. kx − yk = d(x, M ) = inf kx − zk.
z∈M
Proof. I will assume that H is a complex Hilbert space, the real case being Moreover, if M is a vector subspace of H, then the point y may also be charac-
easier. Items 1. and 2. are proved by the following elementary computations; terized as the unique point in M such that (x − y) ⊥ M.
kx + yk2 + kx − yk2 Proof. Uniqueness. By replacing M by M − x := {m − x : m ∈ M } we
= kxk2 + kyk2 + 2Rehx|yi + kxk2 + kyk2 − 2Rehx|yi may assume x = 0. Let δ := d(0, M ) = inf m∈M kmk and y, z ∈ M, see Figure
8.2.
= 2kxk2 + 2kyk2 ,
and
X 2

X X X
x = h x| yi = hx|yi

x∈S x∈S y∈S x,y∈S
X X
= hx|xi = kxk2 .
x∈S x∈S
Item 3. is a consequence of the continuity of h·|·i and the fact that
A⊥ = ∩x∈A Nul(h·|xi)
where Nul(h·|xi) = {y ∈ H : hy|xi = 0} – a closed subspace of H. Fig. 8.2. The geometry of convex sets.
Definition 8.7. A Hilbert space is an inner product space (H, h·|·i) such that
the induced Hilbertian norm is complete.
By the parallelogram law and the convexity of M,

8 Hilbert Space Basics 47
2kyk2 + 2kzk2 = ky + zk2 + ky − zk2 Theorem 8.13 (Projection Theorem). Let H be a Hilbert space and M ⊂ H
y + z 2
be a closed subspace. The orthogonal projection PM satisfies:
= 4
+ ky − zk2 ≥ 4δ 2 + ky − zk2 . (8.4)
2 1. PM is linear and hence we will write PM x rather than PM (x).
2
2. PM = PM (PM is a projection).
Hence if kyk = kzk = δ, then 2δ 2 + 2δ 2 ≥ 4δ 2 + ky − zk2 , so that ky − zk2 = 0. ∗
3. PM = PM (PM is self-adjoint).
Therefore, if a minimizer for d(0, ·)|M exists, it is unique. 4. Ran(PM ) = M and Nul(PM ) = M ⊥ .
Existence. Let yn ∈ M be chosen such that kyn k = δn → δ ≡ d(0, M ).
Taking y = ym and z = yn in Eq. (8.4) shows Proof.
2
2δm + 2δn2 ≥ 4δ 2 + kyn − ym k2 . 1. Let x1 , x2 ∈ H and α ∈ F, then PM x1 + αPM x2 ∈ M and
Passing to the limit m, n → ∞ in this equation implies, PM x1 + αPM x2 − (x1 + αx2 ) = [PM x1 − x1 + α(PM x2 − x2 )] ∈ M ⊥
2δ 2 + 2δ 2 ≥ 4δ 2 + lim sup kyn − ym k2 , showing PM x1 + αPM x2 = PM (x1 + αx2 ), i.e. PM is linear.

m,n→∞ 2
2. Obviously Ran(PM ) = M and PM x = x for all x ∈ M . Therefore PM = PM .
⊥
i.e. lim supm,n→∞ kyn − ym k2 = 0. Therefore, by completeness of H, {yn }n=1
∞ 3. Let x, y ∈ H, then since (x − PM x) and (y − PM y) are in M ,
is convergent. Because M is closed, y := lim yn ∈ M and because the norm is
n→∞ hPM x|yi = hPM x|PM y + y − PM yi = hPM x|PM yi
continuous,
kyk = lim kyn k = δ = d(0, M ). = hPM x + (x − PM x)|PM yi = hx|PM yi.
n→∞
So y is the desired point in M which is closest to 0. 4. We have already seen, Ran(PM ) = M and PM x = 0 iff x = x − 0 ∈ M ⊥ ,
Now suppose M is a closed subspace of H and x ∈ H. Let y ∈ M be the i.e. Nul(PM ) = M ⊥ .
closest point in M to x. Then for w ∈ M, the function
g(t) := kx − (y + tw)k2 = kx − yk2 − 2tRehx − y|wi + t2 kwk2 Corollary 8.14. If M ⊂ H is a proper closed subspace of a Hilbert space H,
0
has a minimum at t = 0 and therefore 0 = g (0) = −2Rehx − y|wi. Since w ∈ M then H = M ⊕ M ⊥ .
is arbitrary, this implies that (x − y) ⊥ M. Proof. Given x ∈ H, let y = PM x so that x−y ∈ M ⊥ . Then x = y+(x−y) ∈
Finally suppose y ∈ M is any point such that (x − y) ⊥ M. Then for z ∈ M, 2
M +M ⊥ . If x ∈ M ∩M ⊥ , then x ⊥ x, i.e. kxk = hx|xi = 0. So M ∩M ⊥ = {0} .
by Pythagorean’s theorem,
kx − zk2 = kx − y + y − zk2 = kx − yk2 + ky − zk2 ≥ kx − yk2

Exercise 8.1. Suppose M is a subset of H, then M ⊥⊥ = span(M ).
2 2
which shows d(x, M ) ≥ kx − yk . That is to say y is the point in M closest to
Theorem 8.15 (Riesz Theorem). Let H ∗ be the dual space of H (Notation
x.
7.9). The map
j
Definition 8.11. Suppose that A : H → H is a bounded operator. The adjoint z ∈ H −→ h·|zi ∈ H ∗ (8.5)
of A, denoted A∗ , is the unique operator A∗ : H → H such that hAx|yi = 1
hx|A∗ yi. (The proof that A∗ exists and is unique will be given in Proposition is a conjugate linear isometric isomorphism.
8.16 below.) A bounded operator A : H → H is self - adjoint or Hermitian 1
Recall that j is conjugate linear if
if A = A∗ .
j (z1 + αz2 ) = jz1 + ᾱjz2
Definition 8.12. Let H be a Hilbert space and M ⊂ H be a closed subspace.
The orthogonal projection of H onto M is the function PM : H → H such that for all z1 , z2 ∈ H and α ∈ C.
for x ∈ H, PM (x) is the unique element in M such that (x − PM (x)) ⊥ M.

Proof. The map j is conjugate linear by the axioms of the inner products. hAx|y1 + λy2 iK = hAx|y1 iK + λ̄hAx|y2 iK
Moreover, for x, z ∈ H, = hx|A∗ (y1 )iK + λ̄hx|A∗ (y2 )iH
|hx|zi| ≤ kxk kzk for all x ∈ H = hx|A∗ (y1 ) + λA∗ (y2 )iH
with equality when x = z. This implies that kjzkH ∗ = kh·|zikH ∗ = kzk . There- and by the uniqueness of A∗ (y1 + λy2 ) we find
fore j is isometric and this implies j is injective. To finish the proof we must A∗ (y1 + λy2 ) = A∗ (y1 ) + λA∗ (y2 ).
show that j is surjective. So let f ∈ H ∗ which we assume, with out loss of gen-
erality, is non-zero. Then M =Nul(f ) – a closed proper subspace of H. Since, This shows A∗ is linear and so we will now write A∗ y instead of A∗ (y).
by Corollary 8.14, H = M ⊕ M ⊥ , f : H/M ∼ = M ⊥ → F is a linear isomorphism. Since
This shows that dim(M ⊥ ) = 1 and hence H = M ⊕ Fx0 where x0 ∈ M ⊥ \ {0} .2 hA∗ y|xiH = hx|A∗ yiH = hAx|yiK = hy|AxiK
2
Choose z = λx0 ∈ M ⊥ such that f (x0 ) = hx0 |zi, i.e. λ = f¯(x0 )/ kx0 k . Then ∗
it follows that A∗∗ = A. The assertion that (A + λB) = A∗ + λ̄B ∗ is Exercise
for x = m + λx0 with m ∈ M and λ ∈ F,
8.2.
f (x) = λf (x0 ) = λhx0 |zi = hλx0 |zi = hm + λx0 |zi = hx|zi Items 3. and 4. Making use of Schwarz’s inequality (Theorem 8.2), we
have
which shows that f = jz.
kA∗ k = sup kA∗ kk
k∈K:kkk=1
Proposition 8.16 (Adjoints). Let H and K be Hilbert spaces and A : H → K
be a bounded operator. Then there exists a unique bounded operator A∗ : K → H = sup sup |hA∗ k|hi|
k∈K:kkk=1 h∈H:khk=1
such that
hAx|yiK = hx|A∗ yiH for all x ∈ H and y ∈ K. (8.6) = sup sup |hk|Ahi| = sup kAhk = kAk
h∈H:khk=1 k∈K:kkk=1 h∈H:khk=1
Moreover, for all A, B ∈ L(H, K) and λ ∈ C,
so that kA∗ k = kAk . Since
∗
1. (A + λB) = A∗ + λ̄B ∗ ,
2
2. A∗∗ := (A∗ )∗ = A, kA∗ Ak ≤ kA∗ k kAk = kAk
3. kA∗ k = kAk and
4. kA∗ Ak = kAk .
2 and
∗
5. If K = H, then (AB) = B ∗ A∗ . In particular A ∈ L (H)
∗ has a bounded 2 2
−1 kAk = sup kAhk = sup |hAh|Ahi|
inverse iff A∗ has a bounded inverse and (A∗ ) = A−1 . h∈H:khk=1 h∈H:khk=1
Proof. For each y ∈ K, the map x → hAx|yiK is in H ∗ and therefore there = sup |hh|A∗ Ahi| ≤ sup kA∗ Ahk = kA∗ Ak (8.7)
h∈H:khk=1 h∈H:khk=1
exists, by Theorem 8.15, a unique vector z ∈ H (we will denote this z by A∗ (y))
such that 2 2
we also have kA∗ Ak ≤ kAk ≤ kA∗ Ak which shows kAk = kA∗ Ak .
hAx|yiK = hx|ziH for all x ∈ H. Alternatively, from Eq. (8.7),
This shows there is a unique map A∗ : K → H such that hAx|yiK = hx|A∗ (y)iH 2
kAk ≤ kA∗ Ak ≤ kAk kA∗ k (8.8)
for all x ∈ H and y ∈ K.
To see A∗ is linear, let y1 , y2 ∈ K and λ ∈ C, then for any x ∈ H, which then implies kAk ≤ kA∗ k . Replacing A by A∗ in this last inequality
2
Alternatively, choose x0 ∈ M ⊥ \ {0} such that f (x0 ) = 1. For x ∈ M ⊥ we have shows kA∗ k ≤ kAk and hence that kA∗ k = kAk . Using this identity back in
2
f (x − λx0 ) = 0 provided that λ := f (x). Therefore x − λx0 ∈ M ∩ M ⊥ = {0} , i.e. Eq. (8.8) proves kAk = kA∗ Ak .
x = λx0 . This again shows that M ⊥ is spanned by x0 . Now suppose that K = H. Then
hABh|ki = hBh|A∗ ki = hh|B ∗ A∗ ki

8.1 Hilbert Space Basis 49
∗
which shows (AB) = B ∗ A∗ . If A−1 exists then Proof. Let Γ ⊂⊂ T be any finite set. Then
∗ ∗ X X X
A−1 A∗ = AA−1 = I ∗ = I and 0 ≤ kx − hx|vivk2 = kxk2 − 2Re hx|vi hv|xi + |hx|vi|2
∗ ∗ v∈Γ v∈Γ v∈Γ
A∗ A−1 = A−1 A = I ∗ = I. X
2 2
∗ = kxk − |hx|vi|
−1
This shows that A∗ is invertible and (A∗ ) = A−1 . Similarly if A∗ is v∈Γ
invertible then so is A = A∗∗ .
|hx|vi|2 ≤ kxk2 . Taking the supremum of this inequality over
P
showing that
v∈Γ
Exercise 8.2. Let H, K, M be Hilbert spaces, A, B ∈ L(H, K), C ∈ L(K, M ) Γ ⊂⊂ T then proves Eq. (8.9).
∗ ∗
and λ ∈ C. Show (A + λB) = A∗ + λ̄B ∗ and (CA) = A∗ C ∗ ∈ L(M, H).
P
n m
Proposition 8.19. Suppose T ⊂ HPis an orthogonal set. Then s = v∈T v
Exercise 8.3. Let H = C and K = C equipped with the usual inner prod- 2
v∈T kvk < ∞. (In particular T must be
exists in H (see Definition 7.17) iff P
ucts, i.e. hz|wiH = z · w̄ for z, w ∈ H. Let A be an m × n matrix thought of as at most a countable set.) Moreover, if v∈T kvk2 < ∞, then
a linear operator from H to K. Show the matrix associated to A∗ : K → H is
2 2
the conjugate transpose of A.
P
1. ksk = Pv∈T kvk and
2. hs|xi = v∈T hv|xi for all x ∈ H.
Lemma 8.17. Suppose A : H → K is a bounded operator, then:
∞
Similarly if {vn }∞
P
1. Nul(A∗ ) = Ran(A)⊥ . n=1 is an orthogonal set, then s = vn exists in H iff
n=1
2. Ran(A) = Nul(A∗ )⊥ . ∞ ∞
kvn k2 < ∞. In particular if
P P
3. if K = H and V ⊂ H is an A – invariant subspace (i.e. A(V ) ⊂ V ), then vn exists, then it is independent of rear-
n=1 n=1
V ⊥ is A∗ – invariant. rangements of {vn }∞
n=1 .
Proof. An element y ∈ K is in Nul(A∗ ) iff 0 = hA∗ y|xi = hy|Axi for all Proof. Suppose s =
P
v exists. Then there exists Γ ⊂⊂ T such that
v∈T
x ∈ H which happens iff y ∈ Ran(A)⊥ . Because, by Exercise 8.1, Ran(A) =
Ran(A)⊥⊥ , and so by the first item, Ran(A) = Nul(A∗ )⊥ . Now suppose A(V ) ⊂ X 2

X 2
V and y ∈ V ⊥ , then kvk = v ≤ 1

v∈Λ v∈Λ
hA∗ y|xi = hy|Axi = 0 for all x ∈ V
for all Λ ⊂⊂ T \ Γ, wherein the first inequality we have used Pythagorean’s
2
which shows A∗ y ∈ V ⊥ .
P
theorem. Taking the supremum over such Λ shows that v∈T \Γ kvk ≤ 1 and
therefore X X
2 2
kvk ≤ 1 + kvk < ∞.
8.1 Hilbert Space Basis v∈T v∈Γ
2
P
Conversely, suppose that v∈T kvk < ∞. Then for all ε > 0 there exists
Proposition 8.18 (Bessel’s Inequality). Let T be an orthonormal set, then Γε ⊂⊂ T such that if Λ ⊂⊂ T \ Γε ,
for any x ∈ H,
X 2 X
X
|hx|vi|2 ≤ kxk2 for all x ∈ H. (8.9)
v = kvk2 < ε2 . (8.10)

v∈T

v∈Λ v∈Λ
In particular the set Tx := {v ∈ T : hx|vi =
6 0} is at most countable for all x ∈ P
H. Hence by Lemma 7.18, v∈T v exists. P
For item 1, let Γε be as above and set sε := v∈Γε v. Then
|ksk − ksε k| ≤ ks − sε k < ε

X X
and by Eq. (8.10), hP x|yi = hhx|uiu|yi = hx|uihu|yi. (8.14)
X X u∈β u∈β
2
0≤ kvk2 − ksε k = kvk2 ≤ ε2 .
v∈T v ∈Γ
/ ε Taking y ∈ β in Eq. (8.14) gives hP x|yi = hx|yi, i.e. that hx − P x|yi = 0 for
all y ∈ β. So (x − P x) ⊥ span β and by continuity we also have (x − P x) ⊥
Letting ε ↓ 0 we deduce from the previous two equations that ksε k → ksk and M = span β. Since P x is also in M, it follows from the definition of PM that
2 2
ksε k → v∈T kvk2 as ε ↓ 0 and therefore ksk = v∈T kvk2 .
P P
P x = PM x proving Eq. (8.11). Equations (8.12) and (8.13) now follow from
Item 2. is a special case of Lemma 7.18. For the final assertion, let sN := (8.14), Proposition 8.19 and the fact that hPM x|yi = hPM 2
x|yi = hPM x|PM yi
N
P ∞
vn and suppose that limN →∞ sN = s exists in H and in particular {sN }N =1 for all x, y ∈ H.
n=1
is Cauchy. So for N > M. Exercise 8.4. Let (H, h·|·i) be a Hilbert space and suppose that {Pn }∞n=1 is a
sequence of orthogonal projection operators on H such that Pn (H) ⊂ Pn+1 (H)
N
X for all n. Let M := ∪∞
n=1 Pn (H) (a subspace of H) and let P denote orthonormal
kvn k2 = ksN − sM k2 → 0 as M, N → ∞
projection onto M̄ . Show limn→∞ Pn x = P x for all x ∈ H. Hint: first prove
n=M +1
the result for x ∈ M ⊥ , then for x ∈ M and then for x ∈ M̄ .
∞ ∞
kvn k2 is convergent, i.e. kvn k2 < ∞.
P P
which shows that Definition 8.21 (Basis). Let H be a Hilbert space. A basis β of H is a max-
n=1 n=1
Alternative imal orthonormal subset β ⊂ H.
P proof of item 1. We could use the last result to prove Item
1. Indeed, if v∈T kvk2 < ∞, then T is countable and so we may write T = Proposition 8.22. Every Hilbert space has an orthonormal basis.
∞
{vn }n=1 . Then s = limN →∞ sN with sN as above. Since the norm, k·k , is
continuous on H, Proof. Let F be the collection of all orthonormal subsets of H ordered by
2 inclusion. If Φ ⊂ F is linearly ordered then ∪Φ is an upper bound. By Zorn’s
N N
Lemma (see Theorem B.7) there exists a maximal element β ∈ F.
X X
2 2 2
ksk = lim ksN k = lim vn = lim kvn k

N →∞ N →∞ N →∞ An orthonormal set β ⊂ H is said to be complete if β ⊥ = {0} . That is to
n=1 n=1
∞ say if hx|ui = 0 for all u ∈ β then x = 0.
X 2
X
= kvn k = kvk2 .
Lemma 8.23. Let β be an orthonormal subset of H then the following are
n=1 v∈T
equivalent:
1. β is a basis,
Corollary 8.20. Suppose H is a Hilbert space, β ⊂ H is an orthonormal set 2. β is complete and
and M = span β. Then 3. span β = H.
X
PM x = hx|uiu, (8.11) Proof. (1. ⇐⇒ 2.) If β is not complete, then there exists a unit vector
u∈β x ∈ β ⊥ \ {0} . The set β ∪ {x} is an orthonormal set properly containing β, so
β is not maximal. Conversely, if β is not maximal, there exists an orthonormal
X 2 2
|hx|ui| = kPM xk and (8.12)
u∈β
set β1 ⊂ H such that β & β1 . Then if x ∈ β1 \ β, we have hx|ui = 0 for all u ∈ β
X showing β is not complete.
hx|uihu|yi = hPM x|yi (8.13) (2. ⇐⇒ 3.) If β is not complete and x ∈ β ⊥ \ {0} , then span β ⊂ x⊥ which
u∈β is a proper subspace of H. Conversely if span β is a proper subspace of H, β ⊥ =
⊥
for all x, y ∈ H. span β is a non-trivial subspace by Corollary 8.14 and β is not complete.
2 2
Theorem 8.24. Let β ⊂ H be an orthonormal set. Then the following are
P
Proof. By Bessel’s inequality,
P u∈β |hx|ui| ≤ kxk for all x ∈ H and hence
by Proposition 8.18, P x := u∈β hx|uiu exists in H and for all x, y ∈ H, equivalent:

8.1 Hilbert Space Basis 51
1. β is complete, i.e. β is an orthonormal basis for H. Proof. Let D ⊂ H be a countable dense set D = {un }∞ n=1 . By Gram-Schmidt
process there exists β = {vn }∞
P
2. x = hx|uiu for all x ∈ H. n=1 an orthonormal set such that span{vn : n =
u∈β P 1, 2 . . . , N } ⊇ span{un : n = 1, 2 . . . , N }. So if hx|vn i = 0 for all n then hx|un i =
3. hx|yi = hx|ui hu|yi for all x, y ∈ H. 0 for all n. Since D ⊂ H is dense we may choose {wk } ⊂ D such that x =
u∈β
4. kxk2 =
P
|hx|ui|2 for all x ∈ H. limk→∞ wk and therefore hx|xi = limk→∞ hx|wk i = 0. That is to say x = 0 and
u∈β β is complete. Conversely if β ⊂ H is a countable orthonormal basis, then the
countable set
Proof. Let M = span β and PP= PM .  
(1) ⇒ (2) By Corollary 8.20, hx|uiu = PM x. Therefore X 
u∈β D= au u : au ∈ Q + iQ : #{u : au 6= 0} < ∞
 
X u∈β
x− hx|uiu = x − PM x ∈ M ⊥ = β ⊥ = {0} .
u∈β is dense in H. Finally let β = {un }∞
n=1 be an orthonormal basis and β1 ⊂ H be
(2) ⇒ (3) is a consequence of Proposition 8.19. another orthonormal basis. Then the sets
(3) ⇒ (4) is obvious, just take y = x.
Bn = {v ∈ β1 : hv|un i =
6 0}
(4) ⇒ (1) If x ∈ β ⊥ , then by 4), kxk = 0, i.e. x = 0. This shows that β is
complete. ∞
S
∞
Suppose Γ := {un }n=1 is a collection of vectors in an inner product space are countable for each n ∈ N and hence B := Bn is a countable subset of β1 .
n=1
(H, h·|·i) . The standard Gram-Schmidt process produces from Γ an orthonor- Suppose there exists v ∈ β1 \B, then hv|un i = 0 for all n and since β = {un }∞
n=1
mal subset, β = {vn }∞ n=1 , such that every element un ∈ Γ is a finite linear com- is an orthonormal basis, this implies v = 0 which is impossible since kvk = 1.
bination of elements from β. Recall the procedure is to define vn inductively by Therefore β1 \ B = ∅ and hence β1 = B is countable.
setting
Xn Notation 8.28 If f : X → C and g : Y → C are two functions, let f ⊗ g :
ṽn+1 := vn+1 − hun+1 |vj ivj = vn+1 − Pn vn+1 X × Y → C be defined by f ⊗ g (x, y) := f (x) g (y) .
j=1
n Proposition 8.29. Suppose X and Y are sets and µ : X → (0, ∞) and
where Pn is orthogonal projection onto Mn := span({vk }k=1 ). If vn+1 := 0, let
−1 ν : Y → (0, ∞) are given weight functions. If β ⊂ `2 (µ) and γ ⊂ `2 (ν) are
ṽn+1 = 0, otherwise set vn+1 := kṽn+1 k ṽn+1 . Finally re-index the resulting orthonormal bases, then
sequence so as to throw out those vn with vn = 0. The result is an orthonormal
subset, β ⊂ H, with the desired properties. β ⊗ γ := {f ⊗ g : f ∈ β and g ∈ γ}
Definition 8.25. A subset, Γ, of a normed space X is said to be total if is an orthonormal basis for `2 (µ ⊗ ν) .
span(Γ ) is dense in X.
Proof. Let f, f 0 ∈ `2 (µ) and g, g 0 ∈ `2 (ν) , then by the Tonelli’s Theorem
Remark 8.26. Suppose that {un }∞ ∞
n=1 is a total subset of H. Let {vn }n=1 be
∞ 4.22 for sums and Hölder’s inequality,
the vectors found by performing Gram-Schmidt on the set {un }n=1 . Then β :=
{vn }∞n=1 is an orthonormal basis for H. Indeed, if h ∈ H is orthogonal to β then
X
f ⊗ g · f 0 ⊗ g 0 µ ⊗ ν =
X X
f f 0 µ · gg 0 ν
∞
h is orthogonal to {un }∞n=1 and hence also span {un }n=1 = H. In particular h X×Y X Y
is orthogonal to itself and so h = 0. This generalizes the corresponding results
≤ kf k`2 (µ) kf k`2 (µ) kgk`2 (ν) kg 0 k`2 (ν) = 1 < ∞.
0
for finite dimensional inner product spaces.
Proposition 8.27. A Hilbert space H is separable (BRUCE: has separable been So by Fubini’s Theorem 4.23 for sums,
defined yet?) iff H has a countable orthonormal basis β ⊂ H. Moreover, if H X X
is separable, all orthonormal bases of H are countable. (See Proposition 4.14 in hf ⊗ g|f 0 ⊗ g 0 i`2 (µ⊗ν) = f f¯0 µ · gḡ 0 ν
X Y
Conway’s, “A Course in Functional Analysis,” for a more general version of
this proposition.) = hf |f 0 i`2 (µ) hg|g 0 i`2 (ν) = δf,f 0 δg,g0 .

Therefore, β ⊗ γ is an orthonormal subset of `2 (µ ⊗ ν). So it only remains to as desired.

show β ⊗ γ is complete. We will give two proofs of this fact. Let F ∈ `2 (µ ⊗ ν). Second Proof. Suppose, for all f ∈ β and g ∈ γ that hF |f ⊗ gi = 0, i.e.
In the first proof we will verify item 4. of Theorem 8.24 while in the second we X X
will verify item 1 of Theorem 8.24. 0 = hF |f ⊗ gi`2 (µ⊗ν) = µ (x) ν (y) F (x, y)f¯(x)ḡ(y)
First Proof. By Tonelli’s Theorem, x∈X y∈Y
X
X
µ (x)
X 2 2
ν (y) |F (x, y)| = kF k`2 (µ⊗ν) < ∞ = µ (x) hF (x, ·)|gi`2 (ν) f¯(x). (8.16)
x∈X
x∈X y∈Y
and since µ > 0, it follows that Since

hF (x, ·)|gi`2 (ν) 2 µ (x) ≤
X X X 2
|F (x, y)| ν (y) < ∞,
X 2
|F (x, y)| ν (y) < ∞ for all x ∈ X, µ (x) (8.17)
y∈Y x∈X x∈X y∈Y
i.e. F (x, ·) ∈ `2 (ν) for all x ∈ X. By the completeness of γ, it follows from Eq. (8.16) and the completeness of β that hF (x, ·)|gi`2 (ν) = 0
for all x ∈ X. By the completeness of γ we conclude that F (x, y) = 0 for all
hF (x, ·) |gi`2 (ν) 2
X 2
X
|F (x, y)| ν (y) = hF (x, ·) |F (x, ·)i`2 (ν) = (x, y) ∈ X × Y.
y∈Y g∈γ
Definition 8.30. A linear map U : H → K is an isometry if kU xkK = kxkH
and therefore, for all x ∈ H and U is unitary if U is also surjective.
2
X X 2
kF k`2 (µ⊗ν) = µ (x) ν (y) |F (x, y)| Exercise 8.5. Let U : H → K be a linear map, show the following are equiva-
x∈X y∈Y lent:
hF (x, ·) |gi`2 (ν) 2 µ (x) .
X X
= (8.15)
1. U : H → K is an isometry,
x∈X g∈γ
2. hU x|U x0 iK = hx|x0 iH for all x, x0 ∈ H, (see Eq. (8.33) below)
and in particular, x → hF (x, ·) |gi`2 (ν) is in `2 (µ) . So by the completeness of 3. U ∗ U = idH .
β and the Fubini and Tonelli theorems, we find
2 Exercise 8.6. Let U : H → K be a linear map, show the following are equiva-
X 2 X X lent:
hF (x, ·) |gi`2 (ν) µ (x) = hF (x, ·) |gi`2 (ν) f¯ (x) µ (x)

X

f ∈β x∈X
1. U : H → K is unitary
  2 2. U ∗ U = idH and U U ∗ = idK .
X X X 3. U is invertible and U −1 = U ∗ .

 f¯ (x) µ (x)

=

 F (x, y) ḡ (y) ν (y)
f ∈β x∈X y∈Y Exercise 8.7. Let H be a Hilbert space. Use Theorem 8.24 to show there exists
2 a set X and a unitary map U : H → `2 (X). Moreover, if H is separable and
X X
= F (x, y) f ⊗ g (x, y) µ ⊗ ν (x, y)
dim(H) = ∞, then X can be taken to be N so that H is unitarily equivalent to

f ∈β (x,y)∈X×Y `2 = `2 (N).
X 2
= hF |f ⊗ gi`2 (µ⊗ν) .
f ∈β
8.2 Some Spectral Theory
Combining this result with Eq. (8.15) shows
For this section let H and K be two Hilbert spaces over C.
hF |f ⊗ gi`2 (µ⊗ν) 2
2
X
kF k`2 (µ⊗ν) =
f ∈β, g∈γ
Exercise 8.8. Suppose A : H → H is a bounded self-adjoint operator. Show:

8.2 Some Spectral Theory 53
1. If λ is an eigenvalue of A, i.e. Ax = λx for some x ∈ H \ {0} , then λ ∈ R. Lemma 8.33. Suppose that A ∈ L(H) is a normal operator, i.e. 0 = [A, A∗ ] :=
2. If λ and µ are two distinct eigenvalues of A with eigenvectors x and y AA∗ − A∗ A. Then λ ∈ σ(A) iff
respectively, then x ⊥ y.
inf k(A − λ1)ψk = 0. (8.18)
kψk=1
Unlike in finite dimensions, it is possible that an operator on a complex
Hilbert space may have no eigenvalues, see Example 8.36 and Lemma 8.37 In other words, λ ∈ σ (A) iff there is an “approximate sequence of eigenvectors”
below for a couple of examples. For this reason it is useful to generalize the for (A, λ) , i.e. there exists ψn ∈ H such that kψn k = 1 and Aψn − λψn → 0 as
notion of an eigenvalue as follows. n → ∞.
Definition 8.31. Suppose X is a Banach space over F (F = R or C) and A ∈ Proof. By replacing A by A − λI we may assume that λ = 0. If 0 ∈
/ σ(A),
L (X) . We say λ ∈ F is in the spectrum of A if A − λI does not have a then
bounded3 inverse. The spectrum will be denoted by σ (A) ⊂ F. The resolvent kAψk kψk
= 1/ A−1 > 0.

inf kAψk = inf = inf
set for A is ρ (A) := F\σ (A) . kψk=1 kψk kA−1 ψk
Remark 8.32. If λ is an eigenvalue of A, then A − λI is not injective and hence Now suppose that inf kψk=1 kAψk = ε > 0 or equivalently we have
not invertible. Therefore any eigenvalue of A is in the spectrum of A. If H is a
Hilbert space and A ∈ L (H) , it follows from item 5. of Proposition 8.16 that kAψk ≥ ε kψk
λ ∈ σ (A) iff λ̄ ∈ σ (A∗ ) , i.e. for all ψ ∈ H. Because A is normal,
σ (A∗ ) = λ̄ : λ ∈ σ (A) .

2 2
kAψk = hAψ|Aψi = hA∗ Aψ|ψi = hAA∗ ψ|ψi = hA∗ ψ|A∗ ψi = kA∗ ψk .
Exercise 8.9. Suppose X is a complex Banach space and A ∈ GL (X) . Show Therefore we also have
−1
σ A−1 = σ (A) := λ−1 : λ ∈ σ (A) .

kA∗ ψk = kAψk ≥ ε kψk ∀ ψ ∈ H. (8.19)
If we further assume A is both invertible and isometric, i.e. kAxk = kxk for all This shows in particular that A and A∗ are injective, Ran(A) is closed and
x ∈ X, then show hence by Lemma 8.17
σ (A) ⊂ S 1 := {z ∈ C : |z| = 1} .
Ran(A) = Ran(A) = Nul(A∗ )⊥ = {0}⊥ = H.
Hint: working formally,
Therefore A is algebraically invertible and the inverse is bounded by Eq. (8.19).
−1 1 1 Aλ
A−1 − λ−1 = 1 1 = λ−A
=
A − λ Aλ
λ−A
Lemma 8.34. Suppose that A ∈ L(H) is self-adjoint (i.e. A = A∗ ) then
−1 −1 −1 −1

from which you might expect that A −λ = −λA (A − λ) if λ ∈ h i
ρ (A) . σ(A) ⊂ − kAkop , kAkop ⊂ R.
Exercise 8.10. Suppose X is a Banachnspace and A ∈ L (X) . Use Corollary

o Proof. Writing λ = α + iβ with α, β ∈ R, then
7.22 to show σ (A) is a closed subset of λ ∈ F : |λ| ≤ kAk := kAkL(X) . 2 2 2 2
k(A + α + iβ) ψk = k(A + α)ψk + |β| kψk + 2 Re((A + α) ψ, iβψ)
3 2 2 2
It will follow by the open mapping Theorem 25.19 or the closed graph Theorem = k(A + α)ψk + |β| kψk (8.20)
25.22 that the word bounded may be omitted from this definition.
wherein we have used
Re [iβ((A + α) ψ, ψ)] = β Im((A + α) ψ, ψ) = 0

since 1. A is a bounded self-adjoint operator.

((A + α) ψ, ψ) = (ψ, (A + α) ψ) = ((A + α) ψ, ψ). 2. A has no eigenvalues.
3. σ (A) = [−2, 2] .
Eq. (8.20) along with Lemma 8.33 shows hthat λ ∈
/ σ(A) if iβ 6= 0, i.e. σ(A) ⊂ R.
The fact that σ (A) is now contained in − kAkop , kAkop is a consequence of Proof. For another (simpler) proof of this lemma, see Exercise 23.8 below.
Exercise 8.9. 1. Since
kAf k2 ≤ kf (· + 1)k2 + kf (· − 1)k2 = 2 kf k2 ,
∗
Remark It is not true that σ(A) ⊂ R implies A = A . For example let
8.35. kAkop ≤ 2 < ∞. Moreover, for f, g ∈ `2 (Z) ,
01
A= on H = C2 , then σ(A) = {0} yet A 6= A∗ .
00 X
hAf |gi = i (f (k + 1) − f (k − 1)) ḡ (k)
Example 8.36. Let S ∈ L(H) be a (not necessarily) normal operator. The proof k
of Lemma 8.33 gives λ ∈ σ(S) if Eq. (8.18) holds. However the converse is not X X
= if (k) ḡ (k − 1) − if (k) ḡ (k + 1)
always valid unless S is normal. For example, let S : `2 → `2 be the shift,
k k
S(ω1 , ω2 , . . . ) = (0, ω1 , ω2 , . . . ). Then for any λ ∈ D := {z ∈ C : |z| < 1} , X
= f (k) Ag (k) = hf |Agi,
k(S − λ) ψk = kSψ − λψk ≥ |kSψk − |λ| kψk| = (1 − |λ|) kψk k
and so there does not exists an approximate sequence of eigenvectors for (S, λ) . which shows A = A∗ .
However, as we will now show, σ (S) = D̄. 2. From Lemma 8.34, we know that σ (A) ⊂ [−2, 2] . If λ ∈ [−2, 2] and f ∈ H
To prove this it suffices to show by Remark 8.32 and Exercise 8.9 that satisfies Af = λf, then
D ⊂ σ (S ∗ ) . For if this is the case then D̄ ⊂ σ (S ∗ ) ⊂ D̄ and hence σ (S) = D̄
since D̄ is invariant under complex conjugation. f (k + 1) = −iλf (k) + f (k − 1) for all k ∈ Z. (8.21)
A simple computation shows,
This is a second order difference equations which can be solved analogously to
∗
S (ω1 , ω2 , . . . ) = (ω2 , ω3 , . . . ) second order ordinary differential equations. The idea is to start by looking for a
solution of the form f (k) = αk . Then Eq. (8.21) becomes, αk+1 = −iλαk +αk−1
and ω = (ω1 , ω2 , . . . ) is an eigenvector for S ∗ with eigenvalue λ ∈ C iff or equivalently that
α2 + iλα − 1 = 0.
0 = (S ∗ − λI) (ω1 , ω2 , . . . ) = (ω2 − λω1 , ω3 − λω2 , . . . ).
So we will have a solution if α ∈ {α± } where
Solving these equation shows √
−iλ ± 4 − λ2
ω2 = λω1 , ω3 = λω2 = λ2 ω1 , . . . , ωn = λn−1 ω1 , . . . . α± = .
2
Hence if λ ∈ D, we may let ω1 = 1 above to find For |λ| =
6 2, there are two distinct roots and the general solution to Eq. (8.21)
is of the form
S ∗ (1, λ, λ2 , . . . ) = λ(1, λ, λ2 , . . . ) f (k) = c+ α+k k
+ c− α− (8.22)
where (1, λ, λ2 , . . . ) ∈ `2 . Thus we have shown λ is an eigenvalue for S ∗ for all for some constants c± ∈ C and |λ| = 2, the general solution has the form
λ ∈ D and hence D ⊂ σ(S ∗ ).
k k
f (k) = cα+ + dkα+ (8.23)
Lemma 8.37. Let H = `2 (Z) and let A : H → H be defined by
Since in all cases, |α± | = 14 λ2 + 4 − λ2 = 1, it follows that neither of these

Af (k) = i (f (k + 1) − f (k − 1)) for all k ∈ Z. functions, f, will be in `2 (Z) unless they are identically zero. This shows that
A has no eigenvalues.
Then:

8.3 Compact Operators on a Hilbert Space 55
3. The above argument suggests a method for constructing approximate Therefore, if kf k = kgk = 1, it follows that
eigenfunctions. Namely, let λ ∈ [−2, 2] and define fn (k) := 1|k|≤n αk where
α = α+ . Then a simple computation shows M M
kf + gk2 + kf − gk2 = 2kf k2 + 2kgk2 = M.

|Rehf |T gi| ≤
4 4
k(A − λI) fn k2
lim =0 (8.24) By replacing f be eiθ f where θ is chosen so that eiθ hf |T gi is real, we find
n→∞ kfn k2
and therefore λ ∈ σ (A) . |hf |T gi| ≤ M for all kf k = kgk = 1.
Exercise 8.11. Verify Eq. (8.24). Also show by explicit computations that Hence
kT k = sup |hf |T gi| ≤ M.
kf k=kgk=1
k(A − λI) fn k2
lim 6= 0
n→∞ kfn k2 If f ∈ H \ {0} and kT k = |hT f |f i|/kf k2 then, using Schwarz’s inequality,
if λ ∈
/ [−2, 2] . |hT f |f i| kT f k
kT k = ≤ ≤ kT k. (8.25)
kf k2 kf k
The next couple of results will be needed for the next section.
This implies |hT f |f i| = kT f kkf k and forces equality in Schwarz’s inequality. So
Theorem 8.38 (Rayleigh quotient). Suppose T ∈ L(H) := L(H, H) is a by Theorem 8.2, T f and f are linearly dependent, i.e. T f = λf for some λ ∈ C.
bounded self-adjoint operator, then Substituting this into (8.25) shows that |λ| = kT k. Since T is self-adjoint,
|hf |T f i| 2
kT k = sup 2
. λkf k2 = hλf |f i = hT f |f i = hf |T f i = hf |λf i = λ̄hf |f i = λ̄ kf k ,
f 6=0 kf k
which implies that λ ∈ R and therefore, λ ∈ {±kT k}.
Moreover if there exists a non-zero element f ∈ H such that
|hT f |f i|
= kT k, 8.3 Compact Operators on a Hilbert Space
kf k2
then f is an eigenvector of T with T f = λf and λ ∈ {±kT k}. In this section let H and B be Hilbert spaces and U := {x ∈ H : kxk < 1} be the
unit ball in H. Recall from Definition 14.16 (BRUCE: forward reference. Think
Proof. Let about correct placement of this section.) that a bounded operator, K : H → B,
|hf |T f i|
M := sup 2
. is compact iff K(U ) is compact in B. Equivalently, for all bounded sequences
f 6=0 kf k
{xn }∞ ∞
n=1 ⊂ H, the sequence {Kxn }n=1 has a convergent subsequence in B.
We wish to show M = kT k. Since Because of Theorem 14.15, if dim(H) = ∞ and K : H → B is invertible, then
K is not compact.
|hf |T f i| ≤ kf kkT f k ≤ kT kkf k2 ,
Definition 8.39. K : H → B is said to have finite rank if Ran(K) ⊂ B is
we see M ≤ kT k. Conversely let f, g ∈ H and compute finite dimensional.
hf + g|T (f + g)i − hf − g|T (f − g)i The following result is a simple consequence of Corollaries 14.13 and 14.14.
= hf |T gi + hg|T f i + hf |T gi + hg|T f i Corollary 8.40. If K : H → B is a finite rank operator, then K is compact.
= 2[hf |T gi + hT g|f i] = 2[hf |T gi + hf |T gi] In particular if either dim(H) < ∞ or dim(B) < ∞ then any bounded operator
= 4Rehf |T gi. K : H → B is finite rank and hence compact.

Lemma 8.41. Let K := K(H, B) denote the compact operators from H to B. lim sup kK − Kn k = ε > 0,
n→∞
Then K(H, B) is a norm closed subspace of L(H, B).
in which case there exists xnk ∈ U such that k(K − Knk )xnk k ≥ ε for all nk .
Proof. The fact that K is a vector subspace of L(H, B) will be left to the
Since K is compact, by passing to a subsequence if necessary, we may assume
reader. To finish the proof, we must show that K ∈ L(H, B) is compact if there ∞
{Kxnk }nk =1 is convergent in B. Letting y := limk→∞ Kxnk ,
exists Kn ∈ K(H, B) such that limn→∞ kKn − Kkop = 0.
First Proof. Given ε > 0, choose N = N (ε) such that kKN − Kk < ε. k(K − Knk )xnk k = k(1 − Pnk )Kxnk k
Using the fact that KN U is precompact, choose a finite subset Λ ⊂ U such that ≤ k(1 − Pnk )(Kxnk − y)k + k(1 − Pnk )yk
minx∈Λ ky − KN xk < ε for all y ∈ KN (U ) . Then for z = Kx0 ∈ K(U ) and
≤ kKxnk − yk + k(1 − Pnk )yk → 0 as k → ∞.
x ∈ Λ,
But this contradicts the assumption that ε is positive and hence we must have
kz − Kxk = k(K − KN )x0 + KN (x0 − x) + (KN − K)xk
limn→∞ kK −Kn k = 0, i.e. K is an operator norm limit of finite rank operators.
≤ 2ε + kKN x0 − KN xk. The converse direction follows from Corollary 8.40 and Lemma 8.41.
Therefore minx∈Λ kz − Kxk < 3ε, which shows K(U ) is 3ε bounded for all Corollary 8.43. If K is compact then so is K ∗ .
ε > 0, so K(U ) is totally bounded and hence precompact.
∞ Proof. First Proof. Let Kn = Pn K be as in the proof of Proposition 8.42,
Second Proof. Suppose {xn}n=1 ∞is a bounded sequence in H. By
∞com-
∞ then Kn∗ = K ∗ Pn is still finite rank. Furthermore, using Proposition 8.16,
pactness, there is a subsequence x1n n=1 of {xn }n=1 such that K1 x1n n=1 is
convergent in B. Working inductively, we may construct subsequences kK ∗ − Kn∗ k = kK − Kn k → 0 as n → ∞
∞ ∞
∞
{xn }n=1 ⊃ x1n n=1 ⊃ x2n n=1 · · · ⊃ {xm
∞
n }n=1 ⊃ . . .
showing K ∗ is a limit of finite rank operators and hence compact.
∞
Second Proof. Let {xn }n=1 be a bounded sequence in B, then
∞
such that {Km xmn }n=1 is convergent in B for each m. By the usual Cantor’s di- 2
∞ ∞
agonalization procedure, let yn := xnn , then {yn }n=1 is a subsequence of {xn }n=1 kK ∗ xn − K ∗ xm k = hxn − xm |KK ∗ (xn − xm )i ≤ 2C kKK ∗ (xn − xm )k
∞ (8.26)
such that {Km yn }n=1 is convergent for all m. Since ∞
where C is a bound on the norms of the xn . Since {K ∗ xn }n=1 is also a bounded
kKyn − Kyl k ≤ k(K − Km ) yn k + kKm (yn − yl )k + k(Km − K) yl )k sequence, by the compactness of K there is a subsequence {x0n } of the {xn } such
≤ 2 kK − Km k + kKm (yn − yl )k , that KK ∗ x0n is convergent and hence by Eq. (8.26), so is the sequence {K ∗ x0n } .
lim sup kKyn − Kyl k ≤ 2 kK − Km k → 0 as m → ∞,

n,l→∞
∞ 8.3.1 The Spectral Theorem for Self Adjoint Compact Operators
which shows {Kyn }n=1 is Cauchy and hence convergent.
For the rest of this section, K ∈ K(H) := K(H, H) will be a self-adjoint compact
Proposition 8.42. A bounded operator K : H → B is compact iff there exists operator or S.A.C.O. for short. Because of Proposition 8.42, we might expect
finite rank operators, Kn : H → B, such that kK − Kn k → 0 as n → ∞. compact operators to behave very much like finite dimensional matrices. This
Proof. Since K(U ) is compact it contains a countable dense subset and is typically the case as we will see below.
from this it follows that K (H) is a separable subspace of B. Let {φn } be an Example 8.44 (Model S.A.C.O.). Let H = `2 and K be the diagonal matrix
orthonormal basis for K (H) ⊂ B and  
λ1 0 0 · · ·
N
X  0 λ2 0 · · · 
PN y = hy|φn iφn K =  0 0 λ3 · · ·  ,
 
n=1
.. .. . . . .
 
. . . .
be the orthogonal projection of y onto span{φn }N
n=1 . Then limN →∞ kPN y−yk =
0 for all y ∈ K(H). Define Kn := Pn K – a finite rank operator on H. For sake where limn→∞ |λn | = 0 and λn ∈ R. Then K is a self-adjoint compact operator.
of contradiction suppose that This assertion was proved in Example 14.17.

8.3 Compact Operators on a Hilbert Space 57
The main theorem (Theorem 8.46) of this subsection states that up to uni- 5. Using the {φn }N
n=1 above,
tary equivalence, Example 8.44 is essentially the most general example of an
N
S.A.C.O. X
Kf = λn hf |φn iφn for all f ∈ H. (8.29)
Proposition 8.45. Let K be a S.A.C.O., then either λ = kKk or λ = − kKk n=1
is an eigenvalue of K.
6. The spectrum of K is σ(K) = {0} ∪ {λn : n < N + 1} if dim H = ∞,
Proof. Without loss of generality we may assume that K is non-zero since otherwise σ(K) = {λn : n ≤ N } with N ≤ dim H.
otherwise the result is trivial. By Theorem 8.38, there exists un ∈ H such that
Proof. We will find λn ’s and φn ’s recursively. Let λ1 ∈ {±kKk} and φ1 ∈ H
kun k = 1 and
such that Kφ1 = λ1 φ1 as in Proposition 8.45.
|hun |Kun i| Take M1 = span(φ1 ) so K(M1 ) ⊂ M1 . By Lemma 8.17, KM1⊥ ⊂ M1⊥ . Define
= |hun |Kun i| −→ kKk as n → ∞. (8.27) K1 : M1⊥ → M1⊥ via K1 = K|M1⊥ . Then K1 is again a compact operator. If
kun k2
K1 = 0, we are done. If K1 6= 0, by Proposition 8.45 there exists λ2 ∈ {±kK1 k}
By passing to a subsequence if necessary, we may assume that λ := and φ2 ∈ M1⊥ such that kφ2 k = 1 and K1 φ2 = Kφ2 = λ2 φ2 . Let M2 :=
limn→∞ hun |Kun i exists and λ ∈ {±kKk}. By passing to a further subsequence span(φ1 , φ2 ).
if necessary, we may assume, using the compactness of K, that Kun is conver- Again K (M2 ) ⊂ M2 and hence K2 := K|M2⊥ : M2⊥ → M2⊥ is compact and
gent as well. We now compute: if K2 = 0 we are done. When K2 6= 0, we apply Proposition 8.45 again to find
λ3 ∈ {±kKk2 } and φ3 ∈ M2⊥ such that kφ3 k = 1 and K2 φ3 = Kφ3 = λ3 φ3 .
0 ≤ kKun − λun k2 = kKun k2 − 2λhKun |un i + λ2 Continuing this way indefinitely or until we reach a point where Kn = 0,
≤ λ2 − 2λhKun |un i + λ2 we construct a sequence {λn }N n=1 of eigenvalues and orthonormal eigenvectors
{φn }N
n=1 such that |λn | ≥ |λn+1 | with the further property that
→ λ2 − 2λ2 + λ2 = 0 as n → ∞.
kKφk
Hence |λn | = sup . (8.30)
φ⊥{φ1 ,φ2 ,...φn−1 } kφk
Kun − λun → 0 as n → ∞ (8.28)
and therefore When N < ∞, the remaining results in the theorem are easily verified. So from
1 now on let us assume that N = ∞.
u := lim un = lim Kun ∞
n→∞ λ n→∞ If ε := limn→∞ |λn | > 0, then λ−1n φn n=1 is a bounded sequence in H.
exists. By the continuity of the inner product, kuk = 1 6= 0. By passing to the Hence, by the compactness of K, there exists a subsequence {nk : k ∈ N} of
∞ ∞
limit in Eq. (8.28) we find that Ku = λu. N such that φnk = λ−1 nk Kφnk k=1 is a convergent. However, since {φnk }k=1
is an orthonormal set, this is impossible and hence we must conclude that
Theorem 8.46 (Compact Operator Spectral Theorem). Suppose that ε := limn→∞ |λn | = 0.
K : H → H is a non-zero S.A.C.O., then Let M := span{φn }∞ n=1 . Then K(M ) ⊂ M and hence, by Lemma 8.17,
K(M ⊥ ) ⊂ M ⊥ . Using Eq. (8.30),
1. there exists at least one eigenvalue λ ∈ {±kKk}.
2. There are at most countably many non-zero eigenvalues, {λn }N n=1 , where

kK|M ⊥ k ≤ K|Mn⊥ = |λn | −→ 0 as n → ∞
N = ∞ is allowed. (Unless K is finite rank (i.e. dim Ran (K) < ∞), N will
be infinite.) showing K|M ⊥ ≡ 0. Define P0 to be orthogonal projection onto M ⊥ . Then for
3. The λn ’s (including multiplicities) may be arranged so that |λn | ≥ |λn+1 | f ∈ H,
for all n. If N = ∞ then limn→∞ |λn | = 0. (In particular any eigenspace ∞
X
for K with non-zero eigenvalue is finite dimensional.) f = P0 f + (1 − P0 )f = P0 f + hf |φn iφn
4. The eigenvectors {φn }N n=1 can be chosen to be an O.N. set such that H =
n=1
span{φn } ⊕ Nul(K). and

∞
X ∞
X N p
X
Kf = KP0 f + K hf |φn iφn = λn hf |φn iφn Aψ := λn hψ|φn iφn for all ψ ∈ H.
n=1 n=1 n=1
which proves Eq. (8.29).
A simple computation shows, A2 = K ∗ K, and therefore,
Since {λn }∞
n=1 ⊂ σ(K) and σ(K) is closed, it follows that 0 ∈ σ(K) and
hence {λn }∞
n=1 ∪ {0} ⊂ σ(K). Suppose that z ∈ / {λn }∞n=1 ∪ {0} and let d 2
kAψk = hAψ|Aψi = ψ|A2 ψ

∞
be the distance between z and {λn }n=1 ∪ {0}. Notice that d > 0 because 2
limn→∞ λn = 0. = hψ|K ∗ Kψi = hKψ|Kψi = kKψk
A few simple computations show that:
for all ψ ∈ H. Hence we may define a unitary operator, u : Ran(A) → Ran(K)
∞
X by the formula
(K − zI)f = hf |φn i(λn − z)φn − zP0 f, uAψ = Kψ for all ψ ∈ H.
n=1
We then have
(K − z)−1 exists, N p
X
Kψ = uAψ = λn hψ|φn iuφn (8.32)
∞
X n=1
(K − zI)−1 f = hf |φn i(λn − z)−1 φn − z −1 P0 f, √
n=1 which proves the result with ψn := uφn and αn = λn .
It is instructive to find ψn explicitly and to verify Eq. (8.32) by brute force.
and −1/2
Since φn = λn Aφn ,
∞
X 1 1
k(K − zI)−1 f k2 = |hf |φn i|2 2
+ 2 kP0 f k2 ψn = λ−1/2
n uAφn = λ−1/2
n Kφn
n=1
|λn − z| |z|
∞ and
2 !
1 X 1
≤ |hf |φn i|2 + kP0 f k2 = kf k2 . hKφn |Kφm i = hφn |K ∗ Kφm i = λn δmn .
d n=1
d2
N
This verifies that {ψn }n=1 is an orthonormal set. Moreover,
−1 −1 −1
We have thus shown that (K − zI) exists, k(K − zI) k ≤ d < ∞ and
N p N p
hence z ∈
/ σ(K). X X
λn hψ|φn iψn = λn hψ|φn iλ−1/2
n Kφn
Theorem 8.47 (Structure of Compact Operators). Let K : H → B n=1 n=1
be a compact operator. Then there exists N ∈ N∪ {∞} , orthonormal subsets N
X
N N N =K hψ|φn iφn = Kψ
{φn }n=1 ⊂ H and {ψn }n=1 ⊂ B and a sequence {αn }n=1 ⊂ R+ such that
α1 ≥ α2 ≥ . . . (with limn→∞ αn = 0 if N = ∞), kψn k ≤ 1 for all n and n=1
PN
N
X since n=1 hψ|φn iφn = P ψ where P is orthogonal projection onto Nul(K)⊥ .
Kf = αn hf |φn iψn for all f ∈ H. (8.31) Second Proof. Let K = u |K| be the polar decomposition of K. Then |K|
n=1 is self-adjoint and compact, by Corollary ?? below, and hence by Theorem 8.46
N
there exists an orthonormal basis {φn }n=1 for Nul(|K|)⊥ = Nul(K)⊥ such that
Proof. Since K ∗ K is a self-adjoint compact operator, Theorem 8.46 implies
N |K| φn = λn φn , λ1 ≥ λ2 ≥ . . . and limn→∞ λn = 0 if N = ∞. For f ∈ H,
there exists an orthonormal set {φn }N n=1 ⊂ H and positive numbers {λn }n=1
such that N N N
N X X X
K ∗ Kψ =
X
λn hψ|φn iφn for all ψ ∈ H. Kf = u |K| hf |φn iφn = hf |φn iu |K| φn = λn hf |φn iuφn
n=1 n=1 n=1
n=1
Let A be the positive square root of K ∗ K defined by which is Eq. (8.31) with ψn := uφn .

8.4 Supplement 1: Converse of the Parallelogram Law 59
8.4 Supplement 1: Converse of the Parallelogram Law it suffices to show x → hx|yi is linear for all y ∈ H. (The rest of this proof may
safely be skipped by the reader.) For this we will need to derive an identity
Proposition 8.48 (Parallelogram Law Converse). If (X, k·k) is a normed from Eq. (8.2). To do this we make use of Eq. (8.2) three times to find
space such that Eq. (8.2) holds forpall x, y ∈ X, then there exists a unique inner
product on h·|·i such that kxk := hx|xi for all x ∈ X. In this case we say that kx + y + zk2 = −kx + y − zk2 + 2kx + yk2 + 2kzk2
k·k is a Hilbertian norm. = kx − y − zk2 − 2kx − zk2 − 2kyk2 + 2kx + yk2 + 2kzk2
Proof. If k·k is going to come from an inner product h·|·i, it follows from = ky + z − xk2 − 2kx − zk2 − 2kyk2 + 2kx + yk2 + 2kzk2
Eq. (8.1) that
= −ky + z + xk2 + 2ky + zk2 + 2kxk2
2Rehx|yi = kx + yk2 − kxk2 − kyk2
− 2kx − zk2 − 2kyk2 + 2kx + yk2 + 2kzk2 .
and
−2Rehx|yi = kx − yk2 − kxk2 − kyk2 . Solving this equation for kx + y + zk2 gives
Subtracting these two equations gives the “polarization identity,”
kx + y + zk2 = ky + zk2 + kx + yk2 − kx − zk2 + kxk2 + kzk2 − kyk2 . (8.34)
2 2
4Rehx|yi = kx + yk − kx − yk .
Using Eq. (8.34), for x, y, z ∈ H,
Replacing y by iy in this equation then implies that
2 2
4 Rehx + z|yi = kx + z + yk2 − kx + z − yk2
4Imhx|yi = kx + iyk − kx − iyk
= ky + zk2 + kx + yk2 − kx − zk2 + kxk2 + kzk2 − kyk2
from which we find − kz − yk2 + kx − yk2 − kx − zk2 + kxk2 + kzk2 − kyk2

1X
hx|yi = εkx + εyk2 (8.33)
4 = kz + yk2 − kz − yk2 + kx + yk2 − kx − yk2
ε∈G
= 4 Rehx|yi + 4 Rehz|yi. (8.35)
where G = {±1, ±i} – a cyclic subgroup of S 1 ⊂ C. Hence, if h·|·i is going to
exist we must define it by Eq. (8.33) and the uniqueness has been proved. Now suppose that δ ∈ G, then since |δ| = 1,
For existence, define hx|yi by Eq. (8.33) in which case,
1X 1X
1X 1 4hδx|yi = εkδx + εyk2 = εkx + δ −1 εyk2
εkx + εxk2 = k2xk2 + ikx + ixk2 − ikx − ixk2

hx|xi = 4 4
4 4 ε∈G ε∈G
ε∈G
1X
i i 2 = εδkx + δεyk2 = 4δhx|yi (8.36)
= kxk2 + 1 + i|2 kxk2 − 1 − i|2 kxk2 = kxk .

4
4 4 ε∈G
So to finish the proof, it only remains to show that hx|yi defined by Eq. (8.33) where in the third inequality, the substitution ε → εδ was made in the sum. So
is an inner product. Eq. (8.36) says h±ix|yi = ±ihx|yi and h−x|yi = −hx|yi. Therefore
Since
X X Imhx|yi = Re (−ihx|yi) = Reh−ix|yi
4hy|xi = εky + εxk2 = εkε (y + εx) k2
ε∈G ε∈G which combined with Eq. (8.35) shows
X
2 2
= εkεy + ε xk
Imhx + z|yi = Reh−ix − iz|yi = Reh−ix|yi + Reh−iz|yi
ε∈G
= Imhx|yi + Imhz|yi
= ky + xk2 − k − y + xk2 + ikiy − xk2 − ik − iy − xk2
= kx + yk2 − kx − yk2 + ikx − iyk2 − ikx + iyk2 and therefore (again in combination with Eq. (8.35)),
= 4hx|yi hx + z|yi = hx|yi + hz|yi for all x, y ∈ H.

2
Because of this equation and Eq. (8.36) to finish the proof that x → hx|yi is X
2
X 2
x − hx|uiu = kxk − |hx|ui| < ε

linear, it suffices to show hλx|yi = λhx|yi for all λ > 0. Now if λ = m ∈ N, then
u∈Λ u∈Λ
hmx|yi = hx + (m − 1)x|yi = hx|yi + h(m − 1)x|yi
hx|uiu. Suppose x = (x1 , x2 , . . . , xn , . . . ) ∈ β ⊥ . If 2) is
P
showing that x =
u∈β
so that by induction hmx|yi = mhx|yi. Replacing x by x/m then shows that valid then kxk2 = 0, i.e. x = 0. So β is maximal. Let us now construct a
hx|yi = mhm−1 x|yi so that hm−1 x|yi = m−1 hx|yi and so if m, n ∈ N, we find counterexample to prove the last assertion. Take H = Span{ei }∞ 2
i=1 ⊂ ` and
∞
n 1 n let ũn = e1 − (n + 1)en+1 for n = 1, 2 . . . . Applying Gram-Schmidt to {ũn }n=1
h x|yi = nh x|yi = hx|yi we construct an orthonormal set β = {un }∞ n=1 ⊂ H. I now claim that β ⊂ H is
m m m
maximal. Indeed if x = (x1 , x2 , . . . , xn , . . . ) ∈ β ⊥ then x ⊥ un for all n, i.e.
so that hλx|yi = λhx|yi for all λ > 0 and λ ∈ Q. By continuity, it now follows
that hλx|yi = λhx|yi for all λ > 0. 0 = hx|ũn i = x1 − (n + 1)xn+1 .
−1
Therefore xn+1 = (n + 1) x1 for all n. Since x ∈ Span{ei }∞ i=1 , xN = 0 for
some N sufficiently large and therefore x1 = 0 which in turn implies that xn = 0
8.5 Supplement 2. Non-complete inner product spaces for all n. So x = 0 and hence β is maximal in H. On the other hand, β is not
maximal in `2 . In fact the above argument shows that β ⊥ in `2 is given by the
Part of Theorem 8.24 goes through when H is a not necessarily complete inner span of v = (1, 12 , 13 , 14 , 51 , . . . ). Let P be the orthogonal projection of `2 onto
product space. We have the following proposition. the Span(β) = v ⊥ . Then
∞
Proposition 8.49. Let (H, h·|·i) be a not necessarily complete inner product X hx|vi
space and β ⊂ H be an orthonormal set. Then the following two conditions are hx|un iun = P x = x − 2 v,
n=1 kvk
equivalent:
∞
hx|un iun = x iff x ∈ Span(β) = v ⊥ ⊂ `2 . For example if x =
P
so that
P
1. x = hx|uiu for all x ∈ H.
u∈β n=1
2. kxk2 =
P
|hx|ui|2 for all x ∈ H. / v ⊥ and hence
(1, 0, 0, . . . ) ∈ H (or more generally for x = ei for any i), x ∈
∞
u∈β P
hx|un iun 6= x.
i=1
Moreover, either of these two conditions implies that β ⊂ H is a maxi-
mal orthonormal set. However β ⊂ H being a maximal orthonormal set is not
sufficient (without completeness of H) to show that items 1. and 2. hold! 8.6 Exercises
Proof. As in the proof of Theorem 8.24, 1) implies 2). For 2) implies 1) let Exercise 8.12. Prove Theorem 14.43. Hint: Let H0 := span {xn : n ∈ N} – a
Λ ⊂⊂ β and consider ∞
separable Hilbert subspace of H. Let {λm }m=1 ⊂ H0 be an orthonormal basis
2 and use Cantor’s diagonalization argument to find a subsequence yk := xnk such
that cm := limk→∞ hyk |λm i exists for all m ∈ N. Finish the proof by appealing
X X X
2 2 2
x − hx|uiu = kxk − 2 |hx|ui| + |hx|ui|

u∈Λ

u∈Λ u∈Λ
to Proposition 14.42. (BRUCE: forward reference.)
2
X 2 ∞
= kxk − |hx|ui| . Definition 8.50. We say a sequence {xn }n=1 of a Hilbert space, H, converges
w
u∈Λ weakly to x ∈ H (and denote this by writing xn → x ∈ H as n → ∞) iff
limn→∞ hxn , yi = hx, yi for all y ∈ H.
Since kxk2 = |hx|ui|2 , it follows that for every ε > 0 there exists Λε ⊂⊂ β
P
∞ w
u∈β Exercise 8.13. Suppose that {xn }n=1 ⊂ H and xn → x ∈ H as n → ∞.
such that for all Λ ⊂⊂ β such that Λε ⊂ Λ, Show xn → x as n → ∞ (i.e. limn→∞ kx − xn k = 0) iff limn→∞ kxn k = kxk .
(BRUCE: weak convergence has not been defined yet.)

∞ w
Exercise 8.14 (Banach-Saks). Suppose that {xn }n=1 ⊂ H, xn → x ∈ H as
n → ∞, and c := supn kxn k < ∞.4 Show there exists a subsequence, yk = xnk
such that
N
1 X
lim x − yk = 0,

N →∞ N
k=1
PN
i.e. N1 k=1 yk → x as N → ∞. Hints: 1. show it suffices to assume x = 0 and
∞
then choose {yk }k=1 so that |hyk |yl i| ≤ l−1 (or even smaller if you like) for all
k ≤ l.
Exercise 8.15 (The Mean Ergodic Theorem). Let U : H → H be a uni-

tary operator on a Hilbert space H, M = Nul(U − I), P = PM be orthogonal
Pn−1
projection onto M, and Sn = n1 k=0 U k . Show Sn → PM strongly by which
we mean limn→∞ Sn x = PM x for all x ∈ H.
Hints: 1. Show H is the orthogonal direct sum of M and Ran(U − I) by
first showing Nul(U ∗ − I) = Nul(U − I) and then using Lemma 8.17. 2. Verify
the result for x ∈ Nul(U − I) and x ∈ Ran(U − I). 3. Use a limiting argument
to verify the result for x ∈ Ran(U − I).
See Definition 14.36 and the exercises in Section 25.4 for more on the notion
of weak and strong convergence.
4
The assumption that c < ∞ is superfluous because of the “uniform boundedness
principle,” see Theorem 25.27 below.
9
Hölder Spaces as Banach Spaces
In this section, we will assume that reader has basic knowledge of the Rie- Lemma 9.3. Suppose u ∈ C 1 (Ω)∩BC(Ω) and ∂i u ∈ BC(Ω) for i = 1, 2, . . . , d,
mann integral and differentiability properties of functions. The results use here then u ∈ C 0,1 (Ω), i.e. [u]1 < ∞.
may be found in Part III below. (BRUCE: there are forward references in this
section.) The proof of this lemma is left to the reader as Exercise 9.1.
Notation 9.1 Let Ω be an open subset of Rd , BC(Ω) and BC(Ω̄) be the Theorem 9.4. Let Ω be an open subset of Rd . Then
bounded continuous functions on Ω and Ω̄ respectively. By identifying f ∈
BC(Ω̄) with f |Ω ∈ BC(Ω), we will consider BC(Ω̄) as a subset of BC(Ω). 1. Under the identification of u ∈ BC Ω̄ with u|Ω ∈ BC (Ω) , BC(Ω̄) is a
For u ∈ BC(Ω) and 0 < β ≤ 1 let closed subspace of BC(Ω).
2. Every element u ∈ C 0,β (Ω) has a unique extension to a continuous func-
tion (still denoted by u) on Ω̄. Therefore we may identify C 0,β (Ω) with

|u(x) − u(y)|
kuku := sup |u(x)| and [u]β := sup .
x∈Ω x,y∈Ω |x − y|β C 0,β (Ω̄) ⊂ BC(Ω̄). (In particular we may consider C 0,β (Ω) and C 0,β (Ω̄)
x6=y
to be the same when β > 0.)
If [u]β < ∞, then u is Hölder continuous with holder exponent1 β. The 3. The function u ∈ C 0,β (Ω) → kukC 0,β (Ω) ∈ [0, ∞) is a norm on C 0,β (Ω)
collection of β – Hölder continuous function on Ω will be denoted by which make C 0,β (Ω) into a Banach space.
C 0,β (Ω) := {u ∈ BC(Ω) : [u]β < ∞} Proof. 1. The first item is trivial since for u ∈ BC(Ω̄), the sup-norm of u
on Ω̄ agrees with the sup-norm on Ω and BC(Ω̄) is complete in this norm.
and for u ∈ C 0,β (Ω) let ∞
2. Suppose that [u]β < ∞ and x0 ∈ bd(Ω). Let {xn }n=1 ⊂ Ω be a sequence
such that x0 = limn→∞ xn . Then
kukC 0,β (Ω) := kuku + [u]β . (9.1)
β
Remark 9.2. If u : Ω → C and [u]β < ∞ for some β > 1, then u is constant on |u(xn ) − u(xm )| ≤ [u]β |xn − xm | → 0 as m, n → ∞
each connected component of Ω. Indeed, if x ∈ Ω and h ∈ Rd then ∞
showing {u(xn )}n=1 is Cauchy so that ū(x0 ) := limn→∞ u(xn ) exists. If
∞

u(x + th) − u(x) {yn }n=1 ⊂ Ω is another sequence converging to x0 , then
≤ [u]β tβ /t → 0 as t → 0
t
β
|u(xn ) − u(yn )| ≤ [u]β |xn − yn | → 0 as n → ∞,
which shows ∂h u(x) = 0 for all x ∈ Ω. If y ∈ Ω is in the same connected
component as x, then by Exercise 22.8 below there exists a smooth curve σ : showing ū(x0 ) is well defined. In this way we define ū(x) for all x ∈ bd(Ω) and
[0, 1] → Ω such that σ(0) = x and σ(1) = y. So by the fundamental theorem of let ū(x) = u(x) for x ∈ Ω. Since a similar limiting argument shows
calculus and the chain rule,
β
Z 1 Z 1 |ū(x) − ū(y)| ≤ [u]β |x − y| for all x, y ∈ Ω̄
d
u(y) − u(x) = u(σ(t))dt = 0 dt = 0.
0 dt 0 it follows that ū is still continuous and [ū]β = [u]β . In the sequel we will abuse
This is why we do not talk about Hölder spaces with Hölder exponents larger notation and simply denote ū by u.
than 1. 3. For u, v ∈ C 0,β (Ω),
1
If β = 1, u is is said to be Lipschitz continuous.
64 9 Hölder Spaces as Banach Spaces

|v(y) + u(y) − v(x) − u(x)| assumption, the set Kε := {x ∈ Ω : |u(x)| ≥ ε} is a compact subset of Ω. Since
[v + u]β = sup
x,y∈Ω |x − y|β bd(Ω) = Ω̄ \ Ω, bd(Ω) ∩ Kε = ∅ and therefore the distance, δ := d(Kε , bd(Ω)),
x6=y
between Kε and bd(Ω) is positive. So if x ∈ bd(Ω) and y ∈ Ω̄ and |y − x| < δ,
|v(y) − v(x)| + |u(y) − u(x)| then |ū(x) − ū(y)| = |u(y)| < ε which shows ū : Ω̄ → C is continuous. This also
≤ sup ≤ [v]β + [u]β
x,y∈Ω |x − y|β shows {|ū| ≥ ε} = {|u| ≥ ε} = Kε is compact in Ω and hence also in Ω̄. Since
x6=y
ε > 0 was arbitrary, this shows ū ∈C0 (Ω̄). Conversely if u ∈ C0 (Ω̄) such that
and for λ ∈ C it is easily seen that [λu]β = |λ| [u]β . This shows [·]β is a semi- u|bd(Ω) = 0 and ε > 0, then Kε := x ∈ Ω̄ : |u(x)| ≥ ε is a compact subset of
norm (see Definition 5.1) on C 0,β (Ω) and therefore k · kC 0,β (Ω) defined in Eq. Ω̄ which is contained in Ω since bd(Ω) ∩ Kε = ∅. Therefore Kε is a compact
∞ subset of Ω showing u|Ω ∈ C0 (Ω̄).
(9.1) is a norm. To see that C 0,β (Ω) is complete, let {un }n=1 be a C 0,β (Ω)–
Cauchy sequence. Since BC(Ω̄) is complete, there exists u ∈ BC(Ω̄) such that Definition 9.7. Let Ω be an open subset of Rd , k ∈ N∪ {0} and β ∈ (0, 1].
ku − un k∞ → 0 as n → ∞. For x, y ∈ Ω with x 6= y, Let BC k (Ω) (BC k (Ω̄)) denote the set of k – times continuously differentiable
functions u on Ω such that ∂ α u ∈ BC(Ω) (∂ α u ∈ BC(Ω̄))2 for all |α| ≤ k.
|u(x) − u(y)| |un (x) − un (y)|
β
= lim β
Similarly, let BC k,β (Ω) denote those u ∈ BC k (Ω) such that [∂ α u]β < ∞ for
|x − y| n→∞ |x − y| all |α| = k. For u ∈ BC k (Ω) let
≤ lim sup[un ]β ≤ lim kun kC 0,β (Ω) < ∞,
n→∞
X
n→∞ kukC k (Ω) = k∂ α uku and
|α|≤k
and so we see that u ∈ C 0,β (Ω). Similarly, X X
kukC k,β (Ω) = k∂ α uku + [∂ α u]β .
|u(x) − un (x) − (u(y) − un (y))| |(um − un )(x) − (um − un )(y)| |α|≤k |α|=k
β
= lim
|x − y| m→∞ |x − y|β
≤ lim sup[um − un ]β → 0 as n → ∞, Theorem 9.8. The spaces BC k (Ω) and BC k,β (Ω) equipped with k · kC k (Ω) and
m→∞ k · kC k,β (Ω) respectively are Banach spaces and BC k (Ω̄) is a closed subspace of
showing [u − un ]β → 0 as n → ∞ and therefore limn→∞ ku − un kC 0,β (Ω) = 0. BC k (Ω) and BC k,β (Ω) ⊂ BC k (Ω̄). Also
Notation 9.5 Since Ω and Ω̄ are locally compact Hausdorff spaces, we may C0k,β (Ω) = C0k,β (Ω̄) = {u ∈ BC k,β (Ω) : ∂ α u ∈ C0 (Ω) ∀ |α| ≤ k}
define C0 (Ω) and C0 (Ω̄) as in Definition 15.22. We will also let
is a closed subspace of BC k,β (Ω).
C00,β (Ω) := C 0,β
(Ω) ∩ C0 (Ω) and C00,β (Ω̄) := C 0,β
(Ω) ∩ C0 (Ω̄). ∞
Proof. Suppose that {un }n=1 ⊂ BC k (Ω) is a Cauchy sequence, then
α ∞
It has already been shown in Proposition 15.23 that C0 (Ω) and C0 (Ω̄) are {∂ un }n=1 is a Cauchy sequence in BC(Ω) for |α| ≤ k. Since BC(Ω) is com-
closed subspaces of BC(Ω) and BC(Ω̄) respectively. The next proposition de- plete, there exists gα ∈ BC(Ω) such that limn→∞ k∂ α un − gα k∞ = 0 for all
scribes the relation between C0 (Ω) and C0 (Ω̄). |α| ≤ k. Letting u := g0 , we must show u ∈ C k (Ω) and ∂ α u = gα for all
|α| ≤ k. This will be done by induction on |α| . If |α| = 0 there is nothing to
Proposition 9.6. Each u ∈ C0 (Ω) has a unique extension to a continuous prove. Suppose that we have verified u ∈ C l (Ω) and ∂ α u = gα for all |α| ≤ l
function on Ω̄ given by ū = u on Ω and ū = 0 on bd(Ω) and the extension ū for some l < k. Then for x ∈ Ω, i ∈ {1, 2, . . . , d} and t ∈ R sufficiently small,
is in C0 (Ω̄). Conversely if u ∈ C0 (Ω̄) and u|bd(Ω) = 0, then u|Ω ∈ C0 (Ω). In Z t
this way we may identify C0 (Ω) with those u ∈ C0 (Ω̄) such that u|bd(Ω) = 0. a a
∂ un (x + tei ) = ∂ un (x) + ∂i ∂ a un (x + τ ei )dτ.
0
Proof. Any extension u ∈ C0 (Ω) to an element ū ∈ C(Ω̄) is necessarily
unique, since Ω is dense inside Ω̄. So define ū = u on Ω and ū = 0 on bd(Ω). Letting n → ∞ in this equation gives
We must show ū is continuous on Ω̄ and ū ∈ C0 (Ω̄). For the continuity assertion 2
To say ∂ α u ∈ BC(Ω̄) means that ∂ α u ∈ BC(Ω) and ∂ α u extends to a continuous
it is enough to show ū is continuous at all points in bd(Ω). For any ε > 0, by function on Ω̄.

Z t
∂ a u(x + tei ) = ∂ a u(x) + gα+ei (x + τ ei )dτ
0
from which it follows that ∂i ∂ α u(x) exists for all x ∈ Ω and ∂i ∂ α u = gα+ei . This
completes the induction argument and also the proof that BC k (Ω) is complete.
It is easy to check that BC k (Ω̄) is a closed subspace of BC k (Ω) and by using
Exercise 9.1 and Theorem 9.4 that that BC k,β (Ω) is a subspace of BC k (Ω̄).
The fact that C0k,β (Ω) is a closed subspace of BC k,β (Ω) is a consequence of
∞
Proposition 15.23. To prove BC k,β (Ω) is complete, let {un }n=1 ⊂ BC k,β (Ω) be
a k · kC k,β (Ω) – Cauchy sequence. By the completeness of BC k (Ω) just proved,
there exists u ∈ BC k (Ω) such that limn→∞ ku − un kC k (Ω) = 0. An application
of Theorem 9.4 then shows limn→∞ k∂ α un − ∂ α ukC 0,β (Ω) = 0 for |α| = k and
therefore limn→∞ ku − un kC k,β (Ω) = 0.
The reader is asked to supply the proof of the following lemma.
Lemma 9.9. The following inclusions hold. For any β ∈ [0, 1]
BC k+1,0 (Ω) ⊂ BC k,1 (Ω) ⊂ BC k,β (Ω)

BC k+1,0 (Ω̄) ⊂ BC k,1 (Ω̄) ⊂ BC k,β (Ω).
9.1 Exercises
Exercise 9.1. Prove Lemma 9.3.

Part III
Calculus and Ordinary Differential Equations in Banach Spaces

10
The Riemann Integral
In this Chapter, the Riemann integral for Banach space valued functions is kf (un ) − f (vn )k ≥ ε while lim |un − vn | = 0.
n→∞
defined and developed. Our exposition will be brief, since the Lebesgue integral
and the Bochner Lebesgue integral will subsume the content of this chapter. By passing to subsequences if necessary we may assume that limn→∞ un
In Definition 14.1 below, we will give a general notion of a compact subset of and limn→∞ vn exists. Since limn→∞ |un − vn | = 0, we must have
a “topological” space. However, by Corollary 14.9 below, when we are working
with subsets of Rd this definition is equivalent to the following definition. lim un = u = lim vn
n→∞ n→∞
d
Definition 10.1. A subset A ⊂ R is said to be compact if A is closed and for some u ∈ K. Since f is continuous, vector addition is continuous and
bounded. the norm is continuous, we may now conclude that
Theorem 10.2. Suppose that K ⊂ Rd is a compact set and f ∈ C (K, X) . ε ≤ lim kf (un ) − f (vn )k = kf (u) − f (u)k = 0
n→∞
Then
∞ which is a contradiction.
1. Every sequence {un }n=1 ⊂ K has a convergent subsequence.
2. The function f is uniformly continuous on K, namely for every ε > 0 there
exists a δ > 0 only depending on ε such that kf (u) − f (v)k < ε whenever For the remainder of the chapter, let [a, b] be a fixed compact interval and
u, v ∈ K and |u − v| < δ where |·| is the standard Euclidean norm on Rd . X be a Banach space. The collection S = S([a, b], X) of step functions, f :
[a, b] → X, consists of those functions f which may be written in the form
Proof.
n−1
1. (This is a special case of Theorem 14.7 and Corollary 14.9 below.) Since X
d f (t) = x0 1[a,t1 ] (t) + xi 1(ti ,ti+1 ] (t), (10.1)
K is bounded, K ⊂ [−R, R] for some sufficiently large d. Let tn be the i=1
first component of un so that tn ∈ [−R, R] for all n. Let J1 = [0, R] if
tn ∈ J1 for infinitely many n otherwise let J1 = [−R, 0]. Similarly split J1 where π := {a = t0 < t1 < · · · < tn = b} is a partition of [a, b] and xi ∈ X. For
in half and let J2 ⊂ J1 be one of the halves such that tn ∈ J2 for infinitely f as in Eq. (10.1), let
many n. Continue this way inductively to find a nested sequence of intervals n−1
J1 ⊃ J2 ⊃ J3 ⊃ J4 ⊃ . . . such that the length of Jk is 2−(k−1) R and for each
X
∞ I(f ) := (ti+1 − ti )xi ∈ X. (10.2)
k, tn ∈ Jk for infinitely many n. We may now choose a subsequence, {nk }k=1 i=0
∞ ∞
of {n}n=1 such that τk := tnk ∈ Jk for all k. The sequence {τk }k=1 is
∞ Exercise 10.1. Show that I(f ) is well defined, independent of how f is repre-
Cauchy and hence convergent. Thus by replacing {un }n=1 by a subsequence
∞ sented as a step function. (Hint: show that adding a point to a partition π of
if necessary we may assume the first component of {un }n=1 is convergent.
Repeating this argument for the second, then the third and all the way [a, b] does not change the right side of Eq. (10.2).) Also verify that I : S → X
∞ is a linear operator.
through the dth – components of {un }n=1 , we may, by passing to further
subsequences, assume all of the components of un are convergent. But this
Notation 10.3 Let S̄ denote the closure of S inside the Banach space,
implies lim un = u exists and since K is closed, u ∈ K.
`∞ ([a, b], X) as defined in Remark 7.6.
2. (This is a special case of Exercise 14.6 below.) If f were not uniformly
∞
continuous on K, there would exists an ε > 0 and sequences {un }n=1 and The following simple “Bounded Linear Transformation” theorem will often
∞
{vn }n=1 in K such that be used in the sequel to define linear transformations.
70 10 The Riemann Integral
n−1 n−1
Theorem 10.4 (B. L. T. Theorem). Suppose that Z is a normed space, X X X
is a Banach space, and S ⊂ Z is a dense linear subspace of Z. If T : S → X is a kI(f )k ≤ (ti+1 − ti )kxi k ≤ (ti+1 − ti )kf k∞ ≤ (b − a)kf k∞ . (10.5)
i=0 i=0
bounded linear transformation (i.e. there exists C < ∞ such that kT zk ≤ C kzk
for all z ∈ S), then T has a unique extension to an element T̄ ∈ L(Z, X) and The existence of I¯ satisfying Eq. (10.3) is a consequence of Theorem 10.4. Given
this extension still satisfies f ∈ C([a, b], X), π := {a = t0 < t1 < · · · < tn = b} a partition of [a, b], and

T̄ z ≤ C kzk for all z ∈ S̄. cπi ∈ [ti , ti+1 ] for i = 0, 1, 2 . . . , n − 1, let fπ ∈ S be defined by
n−1
Exercise 10.2. Prove Theorem 10.4.
X
fπ (t) := f (c0 )0 1[t0 ,t1 ] (t) + f (cπi )1(ti ,ti+1 ] (t).
i=1
Proposition 10.5 (Riemann Integral). The linear function I : S → X ex-
tends uniquely to a continuous linear operator I¯ from S̄ to X and this operator Then by the uniform continuity of f on [a, b] (Theorem 10.2), lim|π|→0 kf −
satisfies, fπ k∞ = 0 and therefore f ∈ S̄. Moreover,
¯ )k ≤ (b − a) kf k∞ for all f ∈ S̄.
kI(f (10.3)
n−1
¯ ) may be com-
Furthermore, C([a, b], X) ⊂ S̄ ⊂ `∞ ([a, b], X) and for f ∈, I(f
X
I (f ) = lim I(fπ ) = lim f (cπi )(ti+1 − ti )
puted as |π|→0 |π|→0
i=0
n−1
X
¯ ) = lim
I(f f (cπi )(ti+1 − ti ) (10.4) which proves Eq. (10.4).
|π|→0
i=0 If fn ∈ S and f ∈ S̄ such that limn→∞
kf − fn k∞ = 0, then for a ≤ α <
where π := {a = t0 < t1 < · · · < tn = b} denotes a partition of [a, b], |π| = β ≤ b, then 1(α,β] fn ∈ S and limn→∞ 1(α,β] f − 1(α,β] fn ∞ = 0. This shows
max {|ti+1 − ti | : i = 0, . . . , n − 1} is the mesh size of π and cπi may be chosen 1(α,β] f ∈ S̄ whenever f ∈ S̄.
arbitrarily inside [ti , ti+1 ]. See Figure 10.1.
Notation 10.6 For f ∈ S̄ and a ≤ α ≤ β ≤ b we will write denote I(1 ¯ (α,β] f )
Rβ R
by α f (t) dt or (α,β] f (t)dt. Also following the usual convention, if a ≤ β ≤
α ≤ b, we will let
Z β Z α
¯ (β,α] f ) = −
f (t) dt = −I(1 f (t) dt.
α β
The next Lemma, whose proof is left to the reader contains some of the
many familiar properties of the Riemann integral.
Lemma 10.7. For f ∈ S̄([a, b], X) and α, β, γ ∈ [a, b], the Riemann integral
satisfies:
R
β
1. α f (t) dt ≤ (β − α) sup {kf (t)k : α ≤ t ≤ β} .

Rγ XRβ Rγ
2. α f (t) dt = α f (t) dt + β f (t) dt.
Rt
Fig. 10.1. The usual picture associated to the Riemann integral. 3. The function G(t) := a f (τ )dτ is continuous on [a, b].
4. If Y is another Banach space and T ∈ L(X, Y ), then T f ∈ S̄([a, b], Y ) and
! Z
Z β β
Proof. Taking the norm of Eq. (10.2) and using the triangle inequality T f (t) dt = T f (t) dt.
shows, α α

10.1 The Fundamental Theorem of Calculus 71
d
5. The function t → kf (t)kX is in S̄([a, b], R) and As for the case of real valued functions, the derivative operator dt is easily
Z seen to be linear. The next two results have proofs very similar to their real
b Z b
valued function analogues.

f (t) dt ≤ kf (t)kX dt.

a a
X Lemma 10.10 (Product Rules). Suppose that t → U (t) ∈ L (X) , t →
6. If f, g ∈ S̄([a, b], R) and f ≤ g, then V (t) ∈ L (X) and t → x (t) ∈ X are differentiable at t = t0 , then
d
1. dt |t0 [U (t) x (t)] ∈ X exists and
Z b Z b
f (t) dt ≤ g(t) dt.
a a d h i
|t0 [U (t) x (t)] = U̇ (t0 ) x (t0 ) + U (t0 ) ẋ (t0 )
Exercise 10.3. Prove Lemma 10.7. dt
and
Remark 10.8 (BRUCE: todo?). Perhaps the Riemann Stieljtes integral, Lemma d
28.38, should be done here. Maybe this should be done in the more genral
2. dt |t0 [U (t) V (t)] ∈ L (X) exists and
context of Banach valued functions in preparation of T. Lyon’s rough path d h i
analysis. The point would be to let Xt take values in a Banach |t0 [U (t) V (t)] = U̇ (t0 ) V (t0 ) + U (t0 ) V̇ (t0 ) .
P space and assume
dt
that Xt had finite variation. Then define µX (t) := supπ l Xt∧tl − Xt∧tl−1 .
−1
Then we could define 3. If U (t0 ) is invertible, then t → U (t) is differentiable at t = t0 and
Z T X d −1 −1 −1
Zt dXt := lim Ztl−1 Xt∧tl − Xt∧tl−1 |t U (t) = −U (t0 ) U̇ (t0 ) U (t0 ) . (10.6)
0 |π|→0 dt 0
for continuous operator valued paths, Zt . This integral would then satisfy the Proof. The reader is asked to supply the proof of the first two items in
−1
estimates, Exercise 10.9. Before proving item 3., let us assume that U (t) is differentiable,
Z Z then using the product rule we would learn
T T
Zt dXt ≤ kZt k dµX (t) ≤ sup kZt k µX (T ) .
i d
d d h −1 −1 −1
0 0 0≤t≤T 0 = |t0 I = |t0 U (t) U (t) = |t0 U (t) U (t0 ) + U (t0 ) U̇ (t0 ) .
dt dt dt
d −1
Solving this equation for dt |t0 U (t) gives the formula in Eq. (10.6). The prob-
10.1 The Fundamental Theorem of Calculus −1
lem with this argument is that we have not yet shown t → U (t) is invertible
Our next goal is to show that our Riemann integral interacts well with differ- at t0 . Here is the formal proof. Since U (t) is differentiable at t0 , U (t) → U (t0 )
entiation, namely the fundamental theorem of calculus holds. Before doing this as t → t0 and by Corollary 7.22, U (t0 + h) is invertible for h near 0 and
we will need a couple of basic definitions and results of differential calculus, −1 −1
U (t0 + h) → U (t0 ) as h → 0.
more details and the next few results below will be done in greater detail in
Chapter 12. Therefore, using Lemma 7.11, we may let h → 0 in the identity,
Definition 10.9. Let (a, b) ⊂ R. A function f : (a, b) → X is differentiable at −1 −1
U (t0 + h) − U (t0 ) −1 U (t0 ) − U (t0 + h) −1
t ∈ (a, b) iff = U (t0 + h) U (t0 ) ,
h h
f (t + h) − f (t)
L := lim h−1 [f (t + h) − f (t)] = lim “

” to learn
h→0 h→0 h
−1 −1
U (t0 + h) − U (t0 )
exists in X. The limit L, if it exists, will be denoted by f˙(t) or df
dt (t). We also lim = −U (t0 )
−1 −1
U̇ (t0 ) U (t0 ) .
say that f ∈ C 1 ((a, b) → X) if f is differentiable at all points t ∈ (a, b) and h→0 h
f˙ ∈ C((a, b) → X).

Proposition 10.11 (Chain Rule). Suppose s → x (s) ∈ X is differentiable Theorem 10.14 (Fundamental Theorem of Calculus). Suppose that f ∈
at s = s0 and t → T (t) ∈ R is differentiable at t = t0 and T (t0 ) = s0 , then C([a, b], X), Then
t → x (T (t)) is differentiable at t0 and d
Rt
1. dt a
f (τ ) dτ = f (t) for all t ∈ (a, b).
d 2. Now assume that F ∈ C([a, b], X), F is continuously differentiable on (a, b)
|t x (T (t)) = x0 (T (t0 )) T 0 (t0 ) .
dt 0 (i.e. Ḟ (t) exists and is continuous for t ∈ (a, b)) and Ḟ extends to a con-
tinuous function on [a, b] which is still denoted by Ḟ . Then
The proof of the chain rule is essentially the same as the real valued function
Z b
case, see Exercise 10.10.
Ḟ (t) dt = F (b) − F (a).
Proposition 10.12. Suppose that f : [a, b] → X is a continuous function such a
that f˙(t) exists and is equal to zero for t ∈ (a, b). Then f is constant. Proof. Let h > 0 be a small number and consider
Z Z
t+h Z t t+h
Proof. Let ε > 0 and α ∈ (a, b) be given. (We will later let ε ↓ 0.) By the

f (τ )dτ − f (τ )dτ − f (t)h = (f (τ ) − f (t)) dτ

definition of the derivative, for all τ ∈ (a, b) there exists δτ > 0 such that

a a t
Z t+h
kf (t) − f (τ )k = f (t) − f (τ ) − f˙(τ )(t − τ ) ≤ ε |t − τ | if |t − τ | < δτ .

≤ k(f (τ ) − f (t))k dτ ≤ hε(h),
t
(10.7)
Let where ε(h) := maxτ ∈[t,t+h] k(f (τ ) − f (t))k. Combining this with a similar com-
A = {t ∈ [α, b] : kf (t) − f (α)k ≤ ε(t − α)} (10.8) putation when h < 0 shows, for all h ∈ R sufficiently small, that
Z
and t0 be the least upper bound for A. We will now use a standard argument t+h Z t
f (τ )dτ − f (τ )dτ − f (t)h ≤ |h|ε(h),

which is referred to as continuous induction to show t0 = b. Eq. (10.7) with
a a
τ = α shows t0 > α and a simple continuity argument shows t0 ∈ A, i.e.
where now ε(h) := maxτ ∈[t−|h|,t+|h|] k(f (τ ) − f (t))k. By continuity of f at t,
kf (t0 ) − f (α)k ≤ ε(t0 − α). (10.9) d
Rt
ε(h) → 0 and hence dt a
f (τ ) dτ exists and is equal to f (t). For the second
Rt
For the sake of contradiction, suppose that t0 < b. By Eqs. (10.7) and (10.9), item, set G(t) := a Ḟ (τ ) dτ − F (t). Then G is continuous by Lemma 10.7 and
Ġ(t) = 0 for all t ∈ (a, b) by item 1. An application of Proposition 10.12 shows
kf (t) − f (α)k ≤ kf (t) − f (t0 )k + kf (t0 ) − f (α)k Rb
G is a constant and in particular G(b) = G(a), i.e. a Ḟ (τ ) dτ − F (b) = −F (a).
≤ ε(t0 − α) + ε(t − t0 ) = ε(t − α)
for 0 ≤ t − t0 < δt0 which violates the definition of t0 being an upper bound. Corollary 10.15 (Mean Value Inequality). Suppose that f : [a, b] → X is
Thus we have shown b ∈ A and hence a continuous function such that f˙(t) exists for t ∈ (a, b) and f˙ extends to a
continuous function on [a, b]. Then
kf (b) − f (α)k ≤ ε(b − α). Z b
kf (b) − f (a)k ≤ kf˙(t)kdt ≤ (b − a) · f˙ . (10.10)

Since ε > 0 was arbitrary we may let ε ↓ 0 in the last equation to conclude ∞
a
f (b) = f (α) . Since α ∈ (a, b) was arbitrary it follows that f (b) = f (α) for all Rb
α ∈ (a, b] and then by continuity for all α ∈ [a, b], i.e. f is constant. Proof. By the fundamental theorem of calculus, f (b) − f (a) = a
f˙(t)dt
and then by Lemma 10.7,
Remark 10.13. The usual real variable proof of Proposition 10.12 makes use Z Z
b b
Rolle’s theorem which in turn uses the extreme value theorem. This latter the-

kf (b) − f (a)k = f˙(t)dt ≤ kf˙(t)kdt

orem is not available to vector valued functions. However with the aid of the a a
Hahn Banach Theorem 25.4 below and Lemma 10.7, it is possible to reduce Z b
the proof of Proposition 10.12 and the proof of the Fundamental Theorem of ≤
˙
f dt = (b − a) · f˙ .

Calculus 10.14 to the real valued case, see Exercise 25.4. a ∞ ∞

10.2 Integral Operators as Examples of Bounded Operators 73
and the latter expression tends to 0 as x → z by uniform continuity of K.

Therefore T f ∈ C([0, 1]) and by the linearity of the Riemann integral, T :
Corollary 10.16 (Change of Variable Formula). Suppose that f ∈
C([0, 1]) → C([0, 1]) is a linear map. Moreover,
C([a, b], X) and T : [c, d] → (a, b) is a continuous function such that T (s)
Z 1 Z 1
is continuously differentiable for s ∈ (c, d) and T 0 (s) extends to a continuous
|T f (x)| ≤ |K(x, y)| |f (y)| dy ≤ |K(x, y)| dy · kf k∞ ≤ A kf k∞
function on [c, d]. Then 0 0
Z d Z T (d) where
f (T (s)) T 0 (s) ds = f (t) dt. 1
Z
c T (c) A := sup |K(x, y)| dy < ∞. (10.12)
x∈[0,1] 0
Rt
Proof. For s ∈ (a, b) define F (t) := T (c) f (τ ) dτ. Then F ∈ C 1 ((a, b) , X) This shows kT k ≤ A < ∞ and therefore T is bounded. We may in fact show
and by the fundamental theorem of calculus and the chain rule, kT k = A. To do this let x0 ∈ [0, 1] be such that
d Z 1 Z 1
F (T (s)) = F 0 (T (s)) T 0 (s) = f (T (s)) T 0 (s) . sup |K(x, y)| dy = |K(x0 , y)| dy.
ds x∈[0,1] 0 0
Integrating this equation on s ∈ [c, d] and using the chain rule again gives
Such an x0 can be found since, using a similar argument to that in Eq. (10.11),
Z d Z T (d) R1
x → 0 |K(x, y)| dy is continuous. Given ε > 0, let
f (T (s)) T 0 (s) ds = F (T (d)) − F (T (c)) = f (t) dt.
c T (c) K(x0 , y)
fε (y) := q
2
ε + |K(x0 , y)|
and notice that limε↓0 kfε k∞ = 1 and
10.2 Integral Operators as Examples of Bounded 2
1
|K(x0 , y)|
Z
Operators kT fε k∞ ≥ |T fε (x0 )| = T fε (x0 ) = q dy.
0 2
ε + |K(x0 , y)|
In the examples to follow all integrals are the standard Riemann integrals and
we will make use of the following notation. Therefore,
d 1 2
Notation 10.17 Given an open set U ⊂ R , let Cc (U ) denote the collection |K(x0 , y)|
Z
1
of real valued continuous functions f on U such that kT k ≥ lim q dy
ε↓0 kfε k∞ 0 2
ε + |K(x0 , y)|
supp(f ) := {x ∈ U : f (x) 6= 0} 1 2
|K(x0 , y)|
Z
= lim dy = A
is a compact subset of U.
q
ε↓0 0 2
ε + |K(x0 , y)|
Example 10.18. Suppose that K : [0, 1] × [0, 1] → C is a continuous function.
For f ∈ C([0, 1]), let since
Z 1 2
|K(x0 , y)|
T f (x) = K(x, y)f (y)dy. 0 ≤ |K(x0 , y)| − q
0 2
ε + |K(x0 , y)|
Since q
|K(x0 , y)| 2
Z 1 =q ε + |K(x0 , y)| − |K(x0 , y)|
|T f (x) − T f (z)| ≤ |K(x, y) − K(z, y)| |f (y)| dy ε + |K(x0 , y)|
2
0 q
≤ kf k∞ max |K(x, y) − K(z, y)| (10.11) 2
≤ ε + |K(x0 , y)| − |K(x0 , y)|
y

and the latter expression tends to zero uniformly in y as ε ↓ 0. ẏ(t) = A(t)y(t) + h (t) where y(0) = x ∈ X, (10.13)
We may also consider other norms on C([0, 1]). Let (for now) L1 ([0, 1]) 1
where y is an unknown function in C (J → X). This equation may be written in
denote C([0, 1]) with the norm
its equivalent (as the reader should verify) integral form, namely we are looking
Z 1
for y ∈ C(J, X) such that
kf k1 = |f (x)| dx, Z t Z t
0
y(t) = x + h (τ ) dτ + A(τ )y(τ )dτ. (10.14)
then T : L1 ([0, 1], dm) → C([0, 1]) is bounded as well. Indeed, let M = 0 0
sup {|K(x, y)| : x, y ∈ [0, 1]} , then In what follows, we will abuse notation and use k·k to denote the operator norm
Z 1 on L (X) associated to then norm, k·k , on X and let kφk∞ := maxt∈J kφ(t)k
|(T f )(x)| ≤ |K(x, y)f (y)| dy ≤ M kf k1 for φ ∈ BC(J, X) or BC(J, L (X)).
0
which shows kT f k∞ ≤ M kf k1 and hence, Notation 10.19 For t ∈ R and n ∈ N, let

{(τ1 , . . . , τn ) ∈ Rn : 0 ≤ τ1 ≤ · · · ≤ τn ≤ t} if t ≥ 0

kT kL1 →C ≤ max {|K(x, y)| : x, y ∈ [0, 1]} < ∞. ∆n (t) =
{(τ1 , . . . , τn ) ∈ Rn : t ≤ τn ≤ · · · ≤ τ1 ≤ 0} if t ≤ 0
We can in fact show that kT k = M as follows. Let (x0 , y0 ) ∈ [0, 1]2 satisfying
and also write dτ = dτ1 . . . dτn and
|K(x0 , y0 )| = M. Then given ε > 0, there exists a neighborhood U = I × Z t
Z Z τn Z τ2
J of (x0 , y0 ) such that |K(x, y) − K(x0 , y0 )| < ε for all (x, y) ∈ U. Let f ∈ n·1t<0
R1 f (τ1 , . . . τn )dτ : = (−1) dτn dτn−1 . . . dτ1 f (τ1 , . . . τn ).
Cc (I, [0, ∞)) such that 0 f (x)dx = 1. Choose α ∈ C such that |α| = 1 and ∆n (t) 0 0 0
αK(x0 , y0 ) = M, then
Lemma 10.20. Suppose that ψ ∈ C (R, R) , then
Z 1 Z
Z Z t n
|(T αf )(x0 )| = K(x0 , y)αf (y)dy = K(x0 , y)αf (y)dy n·1t<0 1
0 I (−1) ψ(τ1 ) . . . ψ(τn )dτ = ψ(τ )dτ . (10.15)
Z ∆n (t) n! 0
≥ Re αK(x0 , y)f (y)dy Rt
Proof. Let Ψ (t) := 0 ψ(τ )dτ. The proof will go by induction on n. The
Z I
case n = 1 is easily verified since
≥ (M − ε) f (y)dy = (M − ε) kαf kL1 Z t
I
Z
(−1)1·1t<0 ψ(τ1 )dτ1 = ψ(τ )dτ = Ψ (t).
and hence ∆1 (t) 0
kT αf kC ≥ (M − ε) kαf kL1
Now assume the truth of Eq. (10.15) for n − 1 for some n ≥ 2, then
showing that kT k ≥ M − ε. Since ε > 0 is arbitrary, we learn that kT k ≥ M Z
and hence kT k = M. (−1)
n·1t<0
ψ(τ1 ) . . . ψ(τn )dτ
One may also view T as a map from T : C([0, 1]) → L1 ([0, 1]) in which case ∆n (t)
one may show Z t Z τn Z τ2
Z 1 = dτn dτn−1 . . .
dτ1 ψ(τ1 ) . . . ψ(τn )
kT kL1 →C ≤ max |K(x, y)| dx < ∞. 0 0 0
0 y Z t Z t
Ψ n−1 (τn ) Ψ n−1 (τn )
= dτn ψ(τn ) = dτn Ψ̇ (τn )
0 (n − 1)! 0 (n − 1)!
10.3 Linear Ordinary Differential Equations Z Ψ (t)
un−1 Ψ n (t)
= du = ,
0 (n − 1)! n!
Let X be a Banach space, J = (a, b) ⊂ R be an open interval with 0 ∈ J,
h ∈ C(J → X) and A ∈ C(J → L(X)). In this section we are going to consider wherein we made the change of variables, u = Ψ (τn ), in the second to last
the ordinary differential equation, equality.

10.3 Linear Ordinary Differential Equations 75
Remark 10.21. Eq. (10.15) is equivalent to Proof. Define Λ : BC(J, X) → BC(J, X) by

!n Z t
Z Z
1 (Λy)(t) = A(τ )y(τ )dτ.
ψ(τ1 ) . . . ψ(τn )dτ = ψ(τ )dτ 0
∆n (t) n! ∆1 (t)
Then y solves Eq. (10.14) iff y = φ + Λy or equivalently iff (I − Λ)y = φ. An
R
and another way to understand this equality is to view ∆n (t) ψ(τ1 ) . . . ψ(τn )dτ induction argument shows
as a multiple integral (see Chapter 20 below) rather than an iterated integral. Z t
Indeed, taking t > 0 for simplicity and letting Sn be the permutation group on (Λn φ)(t) = dτn A(τn )(Λn−1 φ)(τn )
0
{1, 2, . . . , n} we have Z t Z τn
= dτn dτn−1 A(τn )A(τn−1 )(Λn−2 φ)(τn−1 )
[0, t]n = ∪σ∈Sn {(τ1 , . . . , τn ) ∈ Rn : 0 ≤ τσ1 ≤ · · · ≤ τσn ≤ t} 0 0
..
with the union being “essentially” disjoint. Therefore, making a change of vari- .
ables and using the fact that ψ(τ1 ) . . . ψ(τn ) is invariant under permutations, Z t Z τn Z τ2
we find = dτn dτn−1 . . . dτ1 A(τn ) . . . A(τ1 )φ(τ1 )
0 0 0
Z t n Z Z
n·1t<0
ψ(τ )dτ = ψ(τ1 ) . . . ψ(τn )dτ = (−1) A(τn ) . . . A(τ1 )φ(τ1 )dτ.
0 [0,t]n ∆n (t)
X Z Taking norms of this equation and using the triangle inequality along with
= ψ(τ1 ) . . . ψ(τn )dτ
{(τ1 ,...,τn )∈Rn :0≤τσ1 ≤···≤τσn ≤t} Lemma 10.20 gives,
σ∈Sn
Z
X Z
= ψ(sσ−1 1 ) . . . ψ(sσ−1 n )ds k(Λn φ)(t)k ≤ kφk∞ · kA(τn )k . . . kA(τ1 )kdτ
{(s1 ,...,sn )∈Rn :0≤s1 ≤···≤sn ≤t} ∆n (t)
σ∈Sn !n
X Z Z
1
= ψ(s1 ) . . . ψ(sn )ds ≤kφk∞ · kA(τ )kdτ
{(s1 ,...,sn )∈Rn :0≤s1 ≤···≤sn ≤t} n! ∆1 (t)
σ∈Sn
Z Z n
1
= n! ψ(τ1 ) . . . ψ(τn )dτ. ≤kφk∞ · kA(τ )kdτ .
∆n (t) n! J
Therefore,
Theorem 10.22. Let φ ∈ BC(J, X), then the integral equation Z n
1
kΛn kop ≤ kA(τ )kdτ (10.18)
Z t n! J
y(t) = φ(t) + A(τ )y(τ )dτ (10.16)
0
and
∞
X R
kA(τ )kdτ
has a unique solution given by kΛn kop ≤ e J <∞
n=0
∞
X n·1t<0
Z where k·kop denotes the operator norm on L (BC(J, X)) . An application of
y(t) = φ(t) + (−1) A(τn ) . . . A(τ1 )φ(τ1 )dτ (10.17) ∞
Proposition 7.21 now shows (I − Λ)−1 = Λn exists and
P
n=1 ∆n (t)
n=0
and this solution satisfies the bound R
(I − Λ)−1 ≤ e J kA(τ )kdτ .

R op
kA(τ )kdτ
kyk∞ ≤ kφk∞ e J . It is now only a matter of working through the notation to see that these
assertions prove the theorem.

Corollary 10.23. Suppose h ∈ C(J → X) and x ∈ X, then there exits a unique 1. (Semi-group property.) T0 = IdX and Tt Ts = Tt+s for all s, t ≥ 0.
solution, y ∈ C 1 (J, X) , to the linear ordinary differential Eq. (10.13). 2. (Norm Continuity) t → Tt is continuous at 0, i.e. kTt − IkL(X) → 0 as
t ↓ 0.
Proof. Let Z t
Then there exists A ∈ L(X) such that Tt = etA where etA is defined in Eq.
φ (t) = x + h (τ ) dτ.
0 (10.19).
By applying Theorem 10.22 with and J replaced by any open interval J0 such
that 0 ∈ J0 and J¯0 is a compact subinterval1 of J, there exists a unique solution
yJ0 to Eq. (10.13) which is valid for t ∈ J0 . By uniqueness of solutions, if J1 is 10.4 Classical Weierstrass Approximation Theorem
a subinterval of J such that J0 ⊂ J1 and J¯1 is a compact subinterval of J, we
have yJ1 = yJ0 on J0 . Because of this observation, we may construct a solution Definition 10.26 (Support). Let f : X → Z be a function from a metric
y to Eq. (10.13) which is defined on the full interval J by setting y (t) = yJ0 (t) space (X, ρ) to a vector space Z. The support of f is the closed subset, supp(f ),
for any J0 as above which also contains t ∈ J. of X defined by
supp(f ) := {x ∈ X : f (x) 6= 0}.
Corollary 10.24. Suppose that A ∈ L(X) is independent of time, then the
Example 10.27. For example if f : R → R is defined by f (x) = sin(x)1[0,4π] (x) ∈
solution to
R, then
ẏ(t) = Ay(t) with y(0) = x
{f 6= 0} = (0, 4π) \ {π, 2π, 3π}
is given by y(t) = etA x where
and therefore supp(f ) = [0, 4π].
∞ n
X t n For the remainder of this section, Z will be used to denote a Banach space.
etA = A . (10.19)
n=0
n!
Definition 10.28 (Convolution). For f, g ∈ C (R) with either f or g having
Moreover, compact support, we define the convolution of f and g by
e(t+s)A = etA esA for all s, t ∈ R. (10.20) Z Z
f ∗ g(x) = f (x − y)g(y)dy = f (y)g(x − y)dy.
Proof. The first assertion is a simple consequence of Eq. 10.17 and Lemma R R
10.20 with ψ = 1. The assertion in Eq. (10.20) may be proved by explicit We will also use this definition when one of the functions, either f or g, takes
computation but the following proof is more instructive. Given x ∈ X, let values in a Banach space Z.
y (t) := e(t+s)A x. By the chain rule,
∞
Lemma 10.29 (Approximate δ – sequences). Suppose that {qn }n=1 is a
d d sequence non-negative continuous real valued functions on R with compact sup-
y (t) = |τ =t+s eτ A x = Aeτ A x|τ =t+s
dt dτ port that satisfy
= Ae(t+s)A x = Ay (t) with y (0) = esA x. Z
qn (x) dx = 1 and (10.21)
The unique solution to this equation is given by
ZR
y (t) = etA x (0) = etA esA x. lim
n→∞
qn (x)dx = 0 for all ε > 0. (10.22)
|x|≥ε
(t+s)A
This completes the proof since, by definition, y (t) = e x.
We also have the following converse to this corollary whose proof is outlined If f ∈ BC(R, Z), then
in Exercise 10.20 below. Z
qn ∗ f (x) := qn (y)f (x − y)dy
Theorem 10.25. Suppose that Tt ∈ L(X) for t ≥ 0 satisfies R
1
We do this so that φ|J0 will be bounded. converges to f uniformly on compact subsets of R.

10.4 Classical Weierstrass Approximation Theorem 77
Proof. Let x ∈ R, then because of Eq. (10.21),

Z

kqn ∗ f (x) − f (x)k = qn (y) (f (x − y) − f (x)) dy

Z R
≤ qn (y) kf (x − y) − f (x)k dy.
R
Let M = sup {kf (x)k : x ∈ R} . Then for any ε > 0, using Eq. (10.21),
Z
kqn ∗ f (x) − f (x)k ≤ qn (y) kf (x − y) − f (x)k dy
|y|≤ε
Z
+ qn (y) kf (x − y) − f (x)k dy
|y|>ε
Z
≤ sup kf (x + w) − f (x)k + 2M qn (y)dy.
|w|≤ε |y|>ε Fig. 10.2. A plot of q1 , q50 , and q100 . The most peaked curve is q100 and the least is
q1 . The total area under each of these curves is one.
So if K is a compact subset of R (for example a large interval) we have
sup kqn ∗ f (x) − f (x)k Z

2
R1
(1 − x2 )n dx
(x)∈K ε
qn (x)dx = Rε R1
2 0 (1 − x2 )n dx + 2 ε (1 − x2 )n dx
Z
|x|≥ε
≤ sup kf (x + w) − f (x)k + 2M qn (y)dy R1 x
|w|≤ε, x∈K kyk>ε (1 − x2 )n dx (1 − x2 )n+1 |1ε
≤ Rεε xε =
0 ε
(1 − x2 )n dx (1 − x2 )n+1 |ε0
and hence by Eq. (10.22),
(1 − ε2 )n+1
= → 0 as n → ∞,
lim sup sup kqn ∗ f (x) − f (x)k 1 − (1 − ε2 )n+1
n→∞ x∈K
≤ sup kf (x + w) − f (x)k . the proof is complete.

|w|≤ε, x∈K
Notation 10.31 Let Z+ := N ∪ {0} and for x ∈ Rd and α ∈ Zd+ let xα =
Qd αi Pd d
This finishes the proof since the right member of this equation tends to 0 as i=1 xi and |α| = i=1 αi . A polynomial on R with values in Z is a function
ε ↓ 0 by uniform continuity of f on compact subsets of R. p : Rd → Z of the form
Let qn : R →[0, ∞) be defined by X
p(x) = pα xα with pα ∈ Z and N ∈ Z+ .
Z 1
1 α:|α|≤N
qn (x) := (1 − x2 )n 1|x|≤1 where cn := (1 − x2 )n dx. (10.23)
cn −1 If pα 6= 0 for some α such that |α| = N, then we define deg(p) := N to be
Figure 10.2 displays the key features of the functions qn . the degree of p. If Z is a complex Banach space, the function pQhas a natural
d
extension to z ∈ Cd , namely p(z) = α:|α|≤N pα z α where z α = i=1 ziαi .
P
∞
Lemma 10.30. The sequence {qn }n=1 is an approximate δ – sequence, i.e. they
satisfy Eqs. (10.21) and (10.22). Given a compact subset K ⊂ Rd and f ∈ C (K, C)2 , we are going to show, in
the Weierstrass approximation Theorem 10.35 below, that f may be uniformly
Proof. By construction, qn ∈ Cc (R, [0, ∞)) for each n and Eq. 10.21 holds. 2
Note that f is automatically bounded because if not there would exist un ∈ K
Since such that limn→∞ |f (un )| = ∞. Using Theorem 10.2 we may, by passing to a

approximated by polynomial functions on K. The next theorem addresses this Lemma 10.33. Suppose J = [a, b] is a compact subinterval of R and K is
question when K is a compact subinterval of R. a compact subset of Rd−1 , then the linear mapping R : C (J × K, Z) →
C (J, C (K, Z)) defined by (Rf ) (t) = f (t, ·) ∈ C (K, Z) for t ∈ J is an iso-
Theorem 10.32 (Weierstrass Approximation Theorem). Suppose −∞ < metric isomorphism of Banach spaces.
a < b < ∞, J = [a, b] and f ∈ C(J, Z). Then there exists polynomials pn on R
such that pn → f uniformly on J. Proof. By uniform continuity of f on J × K (see Theorem 10.2),
Proof. By replacing f by F where k(Rf ) (t) − (Rf ) (s)kC(K,Z) = max kf (t, y) − f (s, y)kZ → 0 as s → t
y∈K
F (t) := f (a + t (b − a)) − [f (a) + t (f (b) − f (a))] for t ∈ [0, 1] , which shows that Rf is indeed in C (J → C (K, Z)) . Moreover,
it suffices to assume a = 0, b = 1 and f (0) = f (1) = 0. Furthermore we may kRf kC(J→C(K,Z)) = max k(Rf ) (t)kC(K,Z)
now extend f to a continuous function on all R by setting f ≡ 0 on R \ [0, 1] . t∈J
With qn defined as in Eq. (10.23), let fn (x) := (qn ∗ f )(x) and recall from = max max kf (t, y)kZ = kf kC(J×K,Z) ,
t∈J y∈K
Lemma 10.29 that fn (x) → f (x) as n → ∞ with the convergence being uniform
in x ∈ [0, 1]. This completes the proof since fn is equal to a polynomial function showing R is isometric and therefore injective.
on [0, 1] . Indeed, there are polynomials, ak (y) , such that To see that R is surjective, let F ∈ C (J → C (K, Z)) and define f (t, y) :=
F (t) (y) . Since
2n
X
(1 − (x − y)2 )n = ak (y) xk , kf (t, y) − f (s, y 0 )kZ ≤ kf (t, y) − f (s, y)kZ + kf (s, y) − f (s, y 0 )kZ
k=0
≤ kF (t) − F (s)kC(K,Z) + kF (s) (y) − F (s) (y 0 )kZ ,
and therefore, for x ∈ [0, 1] ,
it follows by the continuity of t → F (t) and y → F (s) (y) that
Z
fn (x) = qn (x − y)f (y)dy kf (t, y) − f (s, y 0 )kZ → 0 as (t, y) → (s, y 0 ) .
R
Z
1 This shows f ∈ C (J × K, Z) and thus completes the proof because Rf = F by
f (y) (1 − (x − y)2 )n 1|x−y|≤1 dy

=
cn [0,1] construction.
Z
1
= f (y)(1 − (x − y)2 )n dy Corollary 10.34 (Weierstrass Approximation Theorem). Let d ∈ N,
cn [0,1]
Ji = [ai , bi ] be compact subintervals of R for i = 1, 2, . . . , d, J := J1 × · · · × Jd
2n 2n
and f ∈ C(J, Z). Then there exists polynomials pn on Rd such that pn → f
Z
1 X X
= f (y) ak (y) xk dy = Ak xk uniformly on J.
cn [0,1]
k=0 k=0
Proof. The proof will be by induction on d with the case d = 1 being the
where Z content of Theorem 10.32. Now suppose that d > 1 and the theorem holds with
Ak = f (y) ak (y) dy. d replaced by d−1. Let K := J2 ×· · ·×Jd , Z0 = C (K, Z) , R : C (J1 × K, Z) →
[0,1]
C (J1 , Z0 ) be as in Lemma 10.33 and F := Rf. By Theorem 10.32, for any ε > 0
there exists a polynomial function
subsequence if necessary, assume un → u ∈ K as n → ∞ . Now the continuity of f n
would then imply
X
p (t) = ck tk
∞ = lim |f (un )| = |f (u)|
n→∞ k=0
which is absurd since f takes values in C.
with ck ∈ Z0 = C (K, Z) such that kF − pkC(J1 ,Z0 ) < ε. By the induction
hypothesis, there exists polynomial functions qk : K → Z such that

10.5 Iterated Integrals 79
ε
kck − qk kZ0 < k
. Corollary 10.37 (Complex Weierstrass Approximation Theorem).
n (|a| + |b|) Suppose that K ⊂ Cd is a compact set and f ∈ C(K, C). Then there exists
It is now easily verified (you check) that the polynomial function, polynomials pn (z, z̄) for z ∈ Cd such that supz∈K |pn (z, z̄) − f (z)| → 0 as
n → ∞.
n
Proof. This is an immediate consequence of Theorem 10.35 and Remark
X
ρ (x) := xk1 qk (x2 , . . . , xd ) for x ∈ J
k=0 10.36.
satisfies kf − ρkC(J,Z) < 2ε and this completes the induction argument and Example 10.38. Let K = S 1 = {z ∈ C : |z| = 1} and A be the set of polynomials
hence the proof. in (z, z̄) restricted to S 1 . Then A is dense in C(S 1 ).3 Since z̄ = z −1 on S 1 ,
The reader is referred to Chapter 20 for a two more alternative proofs of we have shown polynomials in z and z −1 are dense in C(S 1 ). This example
d
this corollary. generalizes in an obvious way to K = S 1 ⊂ Cd .
Theorem 10.35 (Weierstrass Approximation Theorem). Suppose that Exercise 10.4. Suppose −∞ < a < b < ∞ and f ∈ C ([a, b], C) satisfies
K ⊂ Rd is a compact subset and f ∈ C(K, C). Then there exists polynomi- Z b
als pn on Rd such that pn → f uniformly on K. f (t) tn dt = 0 for n = 0, 1, 2 . . . .
a
Proof. Choose λ > 0 and b ∈ Rd such that
Show f ≡ 0.
K0 := λK − b := {λx − b : x ∈ K} ⊂ Bd Exercise 10.5. Suppose f ∈ C (R, C) is a 2π – periodic function (i.e.
d f (x + 2π) = f (x) for all x ∈ R) and
where Bd := (0, 1) . The function F (y) := f λ−1 (y + b) for y ∈ K0 is in

Z 2π
C (K0 , C) and if p̂n (y) are polynomials on Rd such that p̂n → F uniformly
on K0 then pn (x) := p̂n (λx − b) are polynomials on Rd such that pn → f f (x) einx dx = 0 for all n ∈ Z,
0
uniformly on K. Hence we may now assume that K is a compact subset of Bd .
show again that f ≡ 0. Hint: Use Example 10.38 to show that any 2π – periodic
Let g ∈ C (K ∪ Bdc ) be defined by
continuous function g on R is the uniform limit of trigonometric polynomials

f (x) if x ∈ K of the form
g (x) = n
0 if x ∈ Bdc
X
p (x) = pk eikx with pk ∈ C for all k.
k=−n
and then use the Tietze extension Theorem 7.4 (applied to the real and imag-
inary parts of F ) to find a continuous function F ∈ C(Rd , C) such that
F = g|K∪Bdc . If pn are polynomials on Rd such that pn → F uniformly on 10.5 Iterated Integrals
d
[0, 1] then pn also converges to f uniformly on K. Hence, by replacing f by F,
d Theorem 10.39 (Baby Fubini Theorem). Let ai , bi ∈ R with ai 6= bi for i =
we may now assume that f ∈ C(Rd , C), K = B̄d = [0, 1] , and f ≡ 0 on Bdc .
The result now follows by an application of Corollary 10.34 with Z = C. 1, 2 . . . , n, f (t1 , t2 , . . . , tn ) ∈ Z be a continuous function of (t1 , t2 , . . . , tn ) where
ti between ai and bi for each i and for any given permutation, σ, of {1, 2 . . . , n}
Remark 10.36. The mapping (x, y) ∈ Rd × Rd → z = x + iy ∈ Cd is an iso- let
morphism of vector spaces. Letting z̄ = x − iy as usual, we have x = z+z̄ 2 and
b σ1 b σn
Z Z
y = z−z̄ . Therefore under this identification any polynomial p(x, y) on R d
× Rd Iσ (f ) := dtσ1 . . . dtσn f (t1 , t2 , . . . , tn ). (10.24)
2i aσ1 aσn
may be written as a polynomial q in (z, z̄), namely
Then Iσ (f ) is well defined and independent of σ, i.e. the order of iterated inte-
z + z̄ z − z̄ grals is irrelevant under these hypothesis.
q(z, z̄) = p( , ).
2 2i 3
Note that it is easy to extend f ∈ C(S 1 ) to a function F ∈ C(C) by setting
Conversely a polynomial q in (z, z̄) may be thought of as a polynomial p in z
F (z) = zf ( |z| ) for z 6= 0 and F (0) = 0. So this special case does not require the
(x, y), namely p(x, y) = q(x + iy, x − iy). Tietze extension theorem.

Proof. Let Ji := [min (ai , bi ) , max (ai , bi )] , J := J1 × · · · × Jn and |Ji | := and then by Fubini’s Theorem 10.39 we learn
max (ai , bi ) − min (ai , bi ) . Using the uniform continuity of f (Theorem 10.2) Z s Z t Z s
∂ ∂ ∂
and the continuity of the Riemann integral, it is easy to prove (compare with f (s, t) = f (s0 , t) + f (σ, t0 )dσ + dτ dσ f (σ, τ ).
the proof of Lemma 10.33) that the map s0 ∂σ t0 s0 ∂τ ∂σ
Z b σn
Differentiating this equation in t and then in s (again using two more applica-
(t1 , . . . t̂σn , . . . , tn ) ∈ (J1 × · · · × Jˆσn × · · · × Jn ) → dtσn f (t1 , t2 , . . . , tn ) tions of Theorem 10.14) shows Eq. (10.26) holds.
aσn
is continuous, where the hat is used to denote a missing element from a list. From 10.6 Exercises
this remark, it follows that each of the integrals in Eq. (10.24) are well defined
and hence so is Iσ (f ) . Moreover by an induction argument using Lemma 10.33 Throughout these problems, (X, k·k) is a Banach space.
and the boundedness of the Riemann integral, we have the estimate, Exercise 10.6. Show f = (f1 , . . . , fn ) ∈ S̄([a, b], Rn ) iff fi ∈ S̄([a, b], R) for
Yn
! i = 1, 2, . . . , n and
kIσ (f )kZ ≤ |Ji | kf kC(J,Z) . (10.25) Z b Z b Z b !
i=1 f (t)dt = f1 (t)dt, . . . , fn (t)dt .
a a a
Now suppose τ is another permutation. Because of Eq. (10.25), Iσ and Iτ
are bounded operators on C (J, Z) and so to shows Iσ = Iτ is suffices to shows Here Rn is to be equipped with the usual Euclidean norm. Hint: Use Lemma
there are equal on the dense set of polynomial functions (see Corollary 10.34) 10.7 to prove the forward implication.
in C (J, Z) . Moreover by linearity, it suffices to show Iσ (f ) = Iτ (f ) when f Exercise 10.7. Give another proof of Proposition 10.40 which does not use
has the form Fubini’s Theorem 10.39 as follows.
f (t1 , t2 , . . . , tn ) = tk11 . . . tknn z
1. By a simple translation argument we may assume (0, 0) ∈ Q and we are
for some ki ∈ N0 and z ∈ Z. However for this function, explicit computations trying to prove Eq. (10.26) holds at (s, t) = (0, 0).
show ∂ ∂
n
! 2. Let h(s, t) := ∂t ∂s f (s, t) and
Y bki i +1 − aki i +1
Iσ (f ) = Iτ (f ) = · z. Z s Z t
i=1
ki + 1 G(s, t) := dσ dτ h(σ, τ )
0 0
so that Eq. (10.27) states

Proposition 10.40 (Equality of Mixed Partial Derivatives). Let Q = Z s
∂
(a, b) × (c, d) be an open rectangle in R2 and f ∈ C(Q, Z). Assume that f (s, t) = f (0, t) + f (σ, t0 )dσ + G(s, t)
∂ ∂ ∂ ∂ 0 ∂σ
∂t f (s, t), ∂s f (s, t) and ∂t ∂s f (s, t) exists and are continuous for (s, t) ∈ Q,
∂ ∂
then ∂s ∂t f (s, t) exists for (s, t) ∈ Q and and differentiating this equation at t = 0 shows
∂ ∂ ∂
∂ ∂ ∂ ∂ f (s, 0) = f (0, 0) + G(s, 0). (10.28)
f (s, t) = f (s, t) for (s, t) ∈ Q. (10.26) ∂t ∂t ∂t
∂s ∂t ∂t ∂s
Now show using the definition of the derivative that
Proof. Fix (s0 , t0 ) ∈ Q. By two applications of Theorem 10.14, Z s
∂
Z s G(s, 0) = dσh(σ, 0). (10.29)
∂ ∂t 0
f (s, t) = f (st0 , t) + f (σ, t)dσ
s0 ∂σ Hint: Consider
Z s Z s Z t s s t
∂ ∂ ∂
Z Z Z
= f (s0 , t) + f (σ, t0 )dσ + dσ dτ f (σ, τ ) (10.27) G(s, t) − t dσh(σ, 0) = dσ dτ [h(σ, τ ) − h(σ, 0)] .
s0 ∂σ s0 t0 ∂τ ∂σ 0 0 0

10.6 Exercises 81
3. Now differentiate Eq. (10.28) in s using Theorem 10.14 to finish the proof. Prove that V (t) is invertible and that V −1 (t) = U (t)4 , where by abuse of
−1
notation I am writing V −1 (t) for [V (t)] . Hints: 1) show dt d
[U (t)V (t)] = 0
Exercise 10.8. Give another proof of Eq. (10.24) in Theorem 10.39 based on (which is sufficient if dim(X) < ∞) and 2) show compute y(t) := V (t)U (t)
Proposition 10.40. To do this let t0 ∈ (c, d) and s0 ∈ (a, b) and define solves a linear differential ordinary differential equation that has y ≡ Id as an
Z t Z s obvious solution. (The results of Exercise 10.11 may be useful here.) Then use
G(s, t) := dτ dσf (σ, τ ) the uniqueness of solutions to linear ODEs.
t0 s0
Exercise 10.13. Suppose that (X, k·k) is a Banach space, J = (a, b) with
Show G satisfies the hypothesis of Proposition 10.40 which combined with two −∞ ≤ a < b ≤ ∞ and fn : R → X are continuously differentiable functions
applications of the fundamental theorem of calculus implies ∞
such that there exists a summable sequence {an }n=1 satisfying
∂ ∂ ∂ ∂
G(s, t) = G(s, t) = f (s, t). kfn (t)k + f˙n (t) ≤ an for all t ∈ J and n ∈ N.

∂t ∂s ∂s ∂t
Use two more applications of the fundamental theorem of calculus along with Show:
the observation that G = 0 if t = t0 or s = s0 to conclude n o
1. sup fn (t+h)−f n (t)
: (t, h) ∈ J × 3 t + h ∈ J and h 6
= 0 ≤ an .

Z s Z t Z s Z t h R
∂ ∂ ∂ 2. The function F : R → X defined by
G(s, t) = dσ dτ G(σ, τ ) = dσ dτ f (σ, τ ). (10.30)
s0 t0 ∂τ ∂σ s0 t0 ∂τ
∞
X
Finally let s = b and t = d in Eq. (10.30) and then let s0 ↓ a and t0 ↓ c to prove F (t) := fn (t) for all t ∈ J
Eq. (10.24). n=1
is differentiable and for t ∈ J,

Exercise 10.9 (Product Rule). Prove items 1. and 2. of Lemma 10.10. This
can be modeled on the standard proof for real valued functions. ∞
X
Ḟ (t) = f˙n (t) .
Exercise 10.10 (Chain Rule). Prove the chain rule in Proposition 10.11. n=1
Again this may be modeled on the on the standard proof for real valued func-
tions. Exercise 10.14. Suppose that A ∈ L(X). Show directly that:
Exercise 10.11. To each A ∈ L (X) , we may define LA , RA : L (X) → L (X) 1. etA define in Eq. (10.19) is convergent in L(X) when equipped with the
by operator norm.
d tA
LA B = AB and RA B = BA for all B ∈ L (X) . 2. etA is differentiable in t and that dt e = AetA .
Show LA , RA ∈ L (L (X)) and that Exercise 10.15. Suppose that A ∈ L(X) and v ∈ X is an eigenvector of A
with eigenvalue λ, i.e. that Av = λv. Show etA v = etλ v. Also show that if
kLA kL(L(X)) = kAkL(X) = kRA kL(L(X)) . X = Rn and A is a diagonalizable n × n matrix with
Exercise 10.12. Suppose that A : R → L(X) is a continuous function and A = SDS −1 with D = diag(λ1 , . . . , λn )
U, V : R → L(X) are the unique solution to the linear differential equations
then etA = SetD S −1 where etD = diag(etλ1 , . . . , etλn ). Here diag(λ1 , . . . , λn )
V̇ (t) = A(t)V (t) with V (0) = I (10.31) denotes the diagonal matrix Λ such that Λii = λi for i = 1, 2, . . . , n.
and Exercise 10.16. Suppose that A, B ∈ L(X) and [A, B] := AB −BA = 0. Show
U̇ (t) = −U (t)A(t) with U (0) = I. (10.32) that e(A+B) = eA eB .
4
The fact that U (t) must be defined as in Eq. (10.32) follows from Lemma 10.10.

d
Exercise 10.17. Suppose A ∈ C(R, L(X)) satisfies [A(t), A(s)] = 0 for all 6. Now show, using the semigroup property and step 4., that dt Tt = ATt for
s, t ∈ R. Show Rt
all t > 0.
d −tA
y(t) := e( 0 A(τ )dτ ) x 7. Using step 5, show dt e Tt = 0 for all t > 0 and therefore e−tA Tt =
−0A
e T0 = I.
is the unique solution to ẏ(t) = A(t)y(t) with y(0) = x.
Exercise 10.21 (Duhamel’ s Principle I). Suppose that A : R → L(X) is
Exercise 10.18. Compute etA when a continuous function and V : R → L(X) is the unique solution to the linear

0 1
differential equation in Eq. (10.31). Let x ∈ X and h ∈ C(R, X) be given. Show
A= that the unique solution to the differential equation:
−1 0
and use the result to prove the formula ẏ(t) = A(t)y(t) + h(t) with y(0) = x (10.33)
cos(s + t) = cos s cos t − sin s sin t. is given by Z t

tA sA (t+s)A y(t) = V (t)x + V (t) V (τ )−1 h(τ ) dτ. (10.34)
Hint: Sum the series and use e e =e . 0
d −1
Exercise 10.19. Compute etA when Hint: compute dt [V (t)y(t)] (see Exercise 10.12) when y solves Eq. (10.33).
Exercise 10.22 (Duhamel’ s Principle II). Suppose that A : R → L(X) is

 
0ab
A = 0 0 c a continuous function and V : R → L(X) is the unique solution to the linear
000 differential equation in Eq. (10.31). Let W0 ∈ L(X) and H ∈ C(R, L(X)) be
given. Show that the unique solution to the differential equation:
with a, b, c ∈ R. Use your result to compute et(λI+A) where λ ∈ R and I is the
3 × 3 identity matrix. Hint: Sum the series. Ẇ (t) = A(t)W (t) + H(t) with W (0) = W0 (10.35)
Exercise 10.20. Prove Theorem 10.25 using the following outline. is given by Z t
1. Using the right continuity at 0 and the semi-group property for Tt , show W (t) = V (t)W0 + V (t) V (τ )−1 H(τ ) dτ. (10.36)
0
there are constants M and C such that kTt kL(X) ≤ M C t for all t > 0.
2. Show t ∈ [0, ∞) → Tt ∈ R εL(X) is continuous.
3. For ε > 0, let Sε := 1ε 0 Tτ dτ ∈ L(X). Show Sε → I as ε ↓ 0 and conclude
from this that Sε is invertible when ε > 0 is sufficiently small. For the
remainder of the proof fix such a small ε > 0.
4. Show
1 t+ε
Z
T t Sε = Tτ dτ
ε t
and conclude from this that

Tt − I 1
lim Sε = (Tε − IdX ) .
t↓0 t ε
5. Using the fact that Sε is invertible, conclude A = limt↓0 t−1 (Tt − I) exists
in L(X) and that
1
A = (Tε − I) Sε−1 .
ε

11
Ordinary Differential Equations in a Banach Space
( x
Let X be a Banach space, U ⊂o X, J = (a, b) 3 0 and Z ∈ C (J × U, X) . √
n−1
if n > 1
y(t) = y(t, x) = 1−(n−1)txn−1 (11.4)
The function Z is to be interpreted as a time dependent vector-field on U ⊂ X. t
ex if n = 1.
In this section we will consider the ordinary differential equation (ODE for
short) The reader should verify by direct calculation that y(t, x) defined above does
ẏ(t) = Z(t, y(t)) with y(0) = x ∈ U. (11.1) indeed solve Eq. (11.3). The above argument shows that these are the only
The reader should check that any solution y ∈ C 1 (J, U ) to Eq. (11.1) gives a possible solutions to the Equations in (11.3).
solution y ∈ C(J, U ) to the integral equation: Notice that when n = 1, the solution exists for all time while for n > 1, we
must require
Z t
1 − (n − 1)txn−1 > 0
y(t) = x + Z(τ, y(τ ))dτ (11.2)
0 or equivalently that
1
and conversely if y ∈ C(J, U ) solves Eq. (11.2) then y ∈ C (J, U ) and y solves 1
Eq. (11.1). t< if xn−1 > 0 and
(1 − n)xn−1
Remark 11.1. For notational simplicity we have assumed that the initial condi- 1 n−1
t>− n−1 if x < 0.
tion for the ODE in Eq. (11.1) is taken at t = 0. There is no loss in generality (1 − n) |x|
in doing this since if ỹ solves
−1
Moreover for n > 1, y(t, x) blows up as t approaches (n − 1) x1−n . The reader
dỹ
(t) = Z̃(t, ỹ(t)) with ỹ(t0 ) = x ∈ U should also observe that, at least for s and t close to 0,
dt
iff y(t) := ỹ(t + t0 ) solves Eq. (11.1) with Z(t, x) = Z̃(t + t0 , x). y(t, y(s, x)) = y(t + s, x) (11.5)
for each of the solutions above. Indeed, if n = 1 Eq. (11.5) is equivalent to the
well know identity, et es = et+s and for n > 1,
11.1 Examples
Let X = R, Z(x) = xn with n ∈ N and consider the ordinary differential
equation
ẏ(t) = Z(y(t)) = y n (t) with y(0) = x ∈ R. (11.3)
If y solves Eq. (11.3) with x 6= 0, then y(t) is not zero for t near 0. Therefore
up to the first time y possibly hits 0, we must have

Z t
ẏ(τ )
Z y(t)  [y(t)]1−n −x1−n if n > 1
−n 1−n
t= dτ = u du =
0 y(τ )
n
y(0)  ln y(t)
x

if n = 1
and solving these equations for y(t) implies

84 11 Ordinary Differential Equations in a Banach Space
1
y(s, x)

y(t, y(s, x)) = p y(t) := ((1 − α)t) 1−α if t > 0 ,
n−1
1 − (n − 1)ty(s, x)n−1 0 if t ≤ 0
n−1
√ x
1−(n−1)sxn−1
= s (for example if α = 1/2 then y(t) = 14 t2 1t≥0 ) then the reader may easily check
n−1 y also solve Eq. (11.6). Furthermore, ya (t) := y(t − a) also solves Eq. (11.6) for
n−1
√ x
1 − (n − 1)t n−1 all a ≥ 0, see Figure 11.1 below.
1−(n−1)sxn−1
√ x
n−1
1−(n−1)sxn−1
= q
xn−1
1 − (n − 1)t 1−(n−1)sx
n−1
n−1
x
= n−1
p
1 − (n − 1)sxn−1 − (n − 1)txn−1
x
= n−1
p = y(t + s, x).
1 − (n − 1)(s + t)xn−1
α
Now suppose Z(x) = |x| with 0 < α < 1 and we now consider the ordinary
differential equation
α
ẏ(t) = Z(y(t)) = |y(t)| with y(0) = x ∈ R. (11.6)
Working as above we find, if x 6= 0 that

Fig. 11.1. Three different solutions to the ODE ẏ(t) = |y(t)|1/2 with y(0) = 0.
t y(t) 1−α
− x1−α
Z Z
ẏ(τ ) −α [y(t)]
t= α dτ = |u| du = ,
0 |y(t)| y(0) 1−α
With these examples in mind, let us now go to the general theory. The case
1−α
where u1−α := |u| sgn(u). Since sgn(y(t)) = sgn(x) the previous equation of linear ODE’s has already been studied in Section 10.3 above.
implies
h i
1−α 1−α
sgn(x)(1 − α)t = sgn(x) sgn(y(t)) |y(t)| − sgn(x) |x| 11.2 Uniqueness Theorem and Continuous Dependence
= |y(t)|
1−α
− |x|
1−α on Initial Data
and therefore, Lemma 11.2. Gronwall’s Lemma. Suppose that f, ε, and k are non-negative
1
functions of a real variable t such that
1−α
1−α
y(t, x) = sgn(x) |x| + sgn(x)(1 − α)t (11.7) Z t

f (t) ≤ ε(t) +
k(τ )f (τ )dτ . (11.8)
1−α 0
is uniquely determined by this formula until the first time t where |x| +
sgn(x)(1 − α)t = 0. Then Z t
As before y(t) = 0 is a solution to Eq. (11.6) when x = 0, however it is far |
Rt
k(s)ds|
f (t) ≤ ε(t) +
k(τ )ε(τ )e τ dτ , (11.9)
from being the unique solution. For example letting x ↓ 0 in Eq. (11.7) gives a 0
function 1 and in particular if ε and k are constants we find that
y(t, 0+) = ((1 − α)t) 1−α
which solves Eq. (11.6) for t > 0. Moreover if we define f (t) ≤ εek|t| . (11.10)

11.3 Local Existence (Non-Linear ODE) 85
Proof. I will only prove the case t ≥ 0. The case t ≤ 0 can be derived 11.3 Local Existence (Non-Linear ODE)
by applying the t ≥ 0 to f˜(t) = f (−t), k̃(t) = k(−t) and ε(t) = ε(−t). Set
Rt
F (t) = 0 k(τ )f (τ )dτ . Then by (11.8), We now show that Eq. (11.1) has a unique solution when Z satisfies the Lips-
chitz condition in Eq. (11.14). See Exercise 14.21 below for another existence
Ḟ = kf ≤ kε + kF. theorem.
Hence, Theorem 11.4 (Local Existence). Let T > 0, J = (−T, T ), x0 ∈ X, r > 0
d − R t k(s)ds Rt Rt
(e 0 F ) = e− 0 k(s)ds (Ḟ − kF ) ≤ kεe− 0 k(s)ds . and
dt C(x0 , r) := {x ∈ X : kx − x0 k ≤ r}
Integrating this last inequality from 0 to t and then solving for F yields:
Z t Z t be the closed r – ball centered at x0 ∈ X. Assume
Rt Rτ Rt
F (t) ≤ e 0 k(s)ds · dτ k(τ )ε(τ )e− 0 k(s)ds = dτ k(τ )ε(τ )e τ k(s)ds .
0 0 M = sup {kZ(t, x)k : (t, x) ∈ J × C(x0 , r)} < ∞ (11.13)
But by the definition of F we have that and there exists K < ∞ such that
f ≤ ε + F,
kZ(t, x) − Z(t, y)k ≤ K kx − yk for all x, y ∈ C(x0 , r) and t ∈ J. (11.14)
and hence the last two displayed equations imply (11.9). Equation (11.10) fol-
lows from (11.9) by a simple integration. Let T0 < min {r/M, T } and J0 := (−T0 , T0 ), then for each x ∈ B(x0 , r − M T0 )
there exists a unique solution y(t) = y(t, x) to Eq. (11.2) in C (J0 , C(x0 , r)) .
Corollary 11.3 (Continuous Dependence on Initial Data). Let U ⊂o X, Moreover y(t, x) is jointly continuous in (t, x), y(t, x) is differentiable in t,
0 ∈ (a, b) and Z : (a, b) × U → X be a continuous function which is K–Lipschitz ẏ(t, x) is jointly continuous for all (t, x) ∈ J0 × B(x0 , r − M T0 ) and satisfies
function on U, i.e. kZ(t, x)−Z(t, x0 )k ≤ Kkx−x0 k for all x and x0 in U. Suppose Eq. (11.1).
y1 , y2 : (a, b) → U solve
Proof. The uniqueness assertion has already been proved in Corollary 11.3.
dyi (t)
= Z(t, yi (t)) with yi (0) = xi for i = 1, 2. (11.11) To prove existence, let Cr := C(x0 , r), Y := C (J0 , C(x0 , r)) and
dt
Z t
Then
Sx (y)(t) := x + Z(τ, y(τ ))dτ. (11.15)
ky2 (t) − y1 (t)k ≤ kx2 − x1 keK|t| for t ∈ (a, b) (11.12) 0
and in particular, there is at most one solution to Eq. (11.1) under the above
With this notation, Eq. (11.2) becomes y = Sx (y), i.e. we are looking for a fixed
Lipschitz assumption on Z.
point of Sx . If y ∈ Y, then
Proof. Let f (t) := ky2 (t) − y1 (t)k. Then by the fundamental theorem of Z t
calculus,
kSx (y)(t) − x0 k ≤ kx − x0 k +

kZ(τ, y(τ ))k dτ ≤ kx − x0 k + M |t|
Z t 0
f (t) = ky2 (0) − y1 (0) + (ẏ2 (τ ) − ẏ1 (τ )) dτ k ≤ kx − x0 k + M T0 ≤ r − M T0 + M T0 = r,
0
Z t
showing Sx (Y ) ⊂ Y for all x ∈ B(x0 , r − M T0 ). Moreover if y, z ∈ Y,
≤ f (0) +
kZ(τ, y2 (τ )) − Z(τ, y1 (τ ))k dτ
0 Z t
Z t
kSx (y)(t) − Sx (z)(t)k = −

[Z(τ, y(τ )) Z(τ, z(τ ))] dτ
= kx2 − x1 k + K f (τ ) dτ .
0

0 Z t

Therefore by Gronwall’s inequality we have, ≤ kZ(τ, y(τ )) − Z(τ, z(τ ))k dτ
0
ky2 (t) − y1 (t)k = f (t) ≤ kx2 − x1 keK|t| .
Z t

≤K ky(τ ) − z(τ )k dτ . (11.16)
0

Let y0 (t, x) = x and yn (·, x) ∈ Y defined inductively by kyn+1 (·, x) − yn (·, x)k∞,J0 := sup {kyn+1 (t, x) − yn (t, x)k : t ∈ J0 } .
Z t
So y(t, x) := limn→∞ yn (t, x) exists uniformly for t ∈ J and using Eq. (11.14)
yn (·, x) := Sx (yn−1 (·, x)) = x + Z(τ, yn−1 (τ, x))dτ. (11.17)
0
we also have
Using the estimate in Eq. (11.16) repeatedly we find sup{ kZ(t, y(t)) − Z(t, yn−1 (t))k : t ∈ J0 }
≤ K ky(·, x) − yn−1 (·, x)k∞,J0 → 0 as n → ∞.
|| yn+1 (t) − yn (t) ||
Z t
Now passing to the limit in Eq. (11.17) shows y solves Eq. (11.2). From this
≤K kyn (τ ) − yn−1 (τ )k dτ equation it follows that y(t, x) is differentiable in t and y satisfies Eq. (11.1).
0
Z t Z t1 The continuity of y(t, x) follows from Corollary 11.3 and mean value inequality
2

≤K dt1
dt2 kyn−1 (t2 ) − yn−2 (t2 )k (Corollary 10.15):
0 0
.. ky(t, x) − y(t0 , x0 )k ≤ ky(t, x) − y(t, x0 )k + ky(t, x0 ) − y(t0 , x0 )k
. Z t
Z t Z Z tn−1
t1 0
0

n
= ky(t, x) − y(t, x )k + Z(τ, y(τ, x ))dτ
≤K
dt1 dt2 . . .
dtn ky1 (tn ) − y0 (tn )k . . .

t0

0 0 0 Z t
≤ ky(t, x) − y(t, x0 )k + kZ(τ, y(τ, x0 ))k dτ
Z
n
≤ K ky1 (·, x) − y0 (·, x)k∞ dτ t0
∆n (t) Z t
n
K n |t| ≤ kx − x0 keKT + kZ(τ, y(τ, x0 ))k dτ

(11.19)
= ky1 (·, x) − y0 (·, x)k∞ (11.18) 0
t
n!
≤ kx − x0 keKT + M |t − t0 | .
wherein we have also made use of Lemma 10.20. Combining this estimate with
Z t

Z t

The continuity of ẏ(t, x) is now a consequence Eq. (11.1) and the continuity of
ky1 (t, x) − y0 (t, x)k =
Z(τ, x)dτ ≤
kZ(τ, x)k dτ ≤ M0 , y and Z.
0 0
Corollary 11.5. Let J = (a, b) 3 0 and suppose Z ∈ C(J × X, X) satisfies
where
(Z
T0 Z 0
) kZ(t, x) − Z(t, y)k ≤ K kx − yk for all x, y ∈ X and t ∈ J. (11.20)
M0 = max kZ(τ, x)k dτ, kZ(τ, x)k dτ ≤ M T0 ,
0 −T0 Then for all x ∈ X, there is a unique solution y(t, x) (for t ∈ J) to Eq. (11.1).
Moreover y(t, x) and ẏ(t, x) are jointly continuous in (t, x).
shows n
K n |t| K n T0n Proof. Let J0 = (a0 , b0 ) 3 0 be a precompact subinterval of J and Y :=
kyn+1 (t, x) − yn (t, x)k ≤ M0 ≤ M0
n! n! BC (J0 , X) . By compactness, M := supt∈J¯0 kZ(t, 0)k < ∞ which combined
and this implies with Eq. (11.20) implies
∞
X sup kZ(t, x)k ≤ M + K kxk for all x ∈ X.
sup{ kyn+1 (·, x) − yn (·, x)k∞,J0 : t ∈ J0 } t∈J¯0
n=0
∞
X K n T0n Using this estimate and Lemma 10.7 one easily shows Sx (Y ) ⊂ Y for all x ∈ X.
≤ M0 = M0 eKT0 < ∞ The proof of Theorem 11.4 now goes through without any further change.
n=0
n!
where

11.4 Global Properties 87
11.4 Global Properties Thus if we let K := max {2M/ε, K0 } , we have shown
Definition 11.6 (Local Lipschitz Functions). Let U ⊂o X, J be an open kZ(t, y) − Z(t, z)k ≤ Kky − zk for all y, z ∈ Eε and t ∈ I.
interval and Z ∈ C(J × U, X). The function Z is said to be locally Lipschitz in
x if for all x ∈ U and all compact intervals I ⊂ J there exists K = K(x, I) < ∞
and ε = ε(x, I) > 0 such that B(x, ε(x, I)) ⊂ U and Proposition 11.8 (Maximal Solutions). Let Z ∈ C(J × U, X) be a locally
Lipschitz function in x and let x ∈ U be fixed. Then there is an interval Jx =
kZ(t, x1 ) − Z(t, x0 )k ≤ K(x, I)kx1 − x0 k ∀ x0 , x1 ∈ B(x, ε(x, I)) & t ∈ I.
(a(x), b(x)) with a ∈ [−∞, 0) and b ∈ (0, ∞] and a C 1 –function y : J → U with
(11.21)
the following properties:
For the rest of this section, we will assume J is an open interval containing
1. y solves ODE in Eq. (11.1).
0, U is an open subset of X and Z ∈ C(J × U, X) is a locally Lipschitz function.
2. If ỹ : J˜ = (ã, b̃) → U is another solution of Eq. (11.1) (we assume that
Lemma 11.7. Let Z ∈ C(J × U, X) be a locally Lipschitz function in X and 0 ∈ J)˜ then J˜ ⊂ J and ỹ = y| ˜.
J
E be a compact subset of U and I be a compact subset of J. Then there exists
The function y : J → U is called the maximal solution to Eq. (11.1).
ε > 0 such that Z(t, x) is bounded for (t, x) ∈ I × Eε and and Z(t, x) is K –
Lipschitz on Eε for all t ∈ I, where Proof. Suppose that yi : Ji = (ai , bi ) → U, i = 1, 2, are two solutions to
Eq. (11.1). We will start by showing that y1 = y2 on J1 ∩ J2 . To do this1 let
Eε := {x ∈ U : dist(x, E) < ε} .
J0 = (a0 , b0 ) be chosen so that 0 ∈ J¯0 ⊂ J1 ∩ J2 , and let E := y1 (J¯0 ) ∪ y2 (J¯0 ) –
Proof. Let ε(x, I) and K(x, I) be as in Definition 11.6 above. Since E is com- a compact subset of X. Choose ε > 0 as in Lemma 11.7 so that Z is Lipschitz
pact, there exists a finite subset Λ ⊂ E such that E ⊂ V := ∪x∈Λ B(x, ε(x, I)/2). on Eε . Then y1 |J0 , y2 |J0 : J0 → Eε both solve Eq. (11.1) and therefore are equal
If y ∈ V, there exists x ∈ Λ such that ky − xk < ε(x, I)/2 and therefore by Corollary 11.3. Since J0 = (a0 , b0 ) was chosen arbitrarily so that [a0 , b0 ] ⊂
J1 ∩ J2 , we may conclude that y1 = y2 on J1 ∩ J2 . Let (yα , Jα = (aα , bα ))α∈A
kZ(t, y)k ≤ kZ(t, x)k + K(x, I) ky − xk ≤ kZ(t, x)k + K(x, I)ε(x, I)/2 denote the possible solutions to (11.1) such that 0 ∈ Jα . Define Jx = ∪Jα and
≤ sup {kZ(t, x)k + K(x, I)ε(x, I)/2} =: M < ∞. set y = yα on Jα . We have just checked that y is well defined and the reader
x∈Λ,t∈I may easily check that this function y : Jx → U satisfies all the conclusions of
the theorem.
This shows Z is bounded on I × V. Let
1 Notation 11.9 For each x ∈ U, let Jx = (a(x), b(x)) be the maximal interval
ε := d(E, V c ) ≤ min ε(x, I) on which Eq. (11.1) may be solved, see Proposition 11.8. Set D(Z) := ∪x∈U (Jx ×
2 x∈Λ
{x}) ⊂ J × U and let φ : D(Z) → U be defined by φ(t, x) = y(t) where y is
and notice that ε > 0 since E is compact, V c is closed and E ∩ V c = ∅. the maximal solution to Eq. (11.1). (So for each x ∈ U, φ(·, x) is the maximal
If y, z ∈ Eε and ky − zk < ε, then as before there exists x ∈ Λ such that solution to Eq. (11.1).)
ky − xk < ε(x, I)/2. Therefore 1
Here is an alternate proof of the uniqueness. Let
kz − xk ≤ kz − yk + ky − xk < ε + ε(x, I)/2 ≤ ε(x, I)
T ≡ sup{t ∈ [0, min{b1 , b2 }) : y1 = y2 on [0, t]}.
and since y, z ∈ B(x, ε(x, I)), it follows that (T is the first positive time after which y1 and y2 disagree.
Suppose, for sake of contradiction, that T < min{b1 , b2 }. Notice that y1 (T ) =
kZ(t, y) − Z(t, z)k ≤ K(x, I)ky − zk ≤ K0 ky − zk
y2 (T ) =: x0 . Applying the local uniqueness theorem to y1 (· − T ) and y2 (· − T )
where K0 := maxx∈Λ K(x, I) < ∞. On the other hand if y, z ∈ Eε and thought as function from (−δ, δ) → B(x0 , (x0 )) for some δ sufficiently small, we
learn that y1 (· − T ) = y2 (· − T ) on (−δ, δ). But this shows that y1 = y2 on [0, T + δ)
ky − zk ≥ ε, then
which contradicts the definition of T. Hence we must have the T = min{b1 , b2 }, i.e.
2M y1 = y2 on J1 ∩J2 ∩[0, ∞). A similar argument shows that y1 = y2 on J1 ∩J2 ∩(−∞, 0]
kZ(t, y) − Z(t, z)k ≤ 2M ≤ ky − zk . as well.
ε

Proposition 11.10. Let Z ∈ C(J × U, X) be a locally Lipschitz function in x Example 11.12. (Not worked out completely.) Let X = U = `2 , ψ ∈ C ∞ (R2 )
and y : Jx = (a(x), b(x)) → U be the maximal solution to Eq. (11.1). If b(x) < b, be a smooth function such that ψ = 1 in a neighborhood of the line segment
then either lim supt↑b(x) kZ(t, y(t))k = ∞ or y(b(x)−) := limt↑b(x) y(t) exists joining (1, 0) to (0, 1) and being supported within the 1/10 – neighborhood of
and y(b(x)−) ∈ / U. Similarly, if a > a(x), then either lim supt↓a(x) ky(t)k = ∞ this segment. Choose an ↑ ∞ and bn ↑ ∞ and define
or y(a(x)+) := limt↓a(x) y(t) exists and y(a(x)+) ∈/ U. ∞
X
Proof. Suppose that b < b(x) and M := lim supt↑b(x) kZ(t, y(t))k < ∞. Z(x) = an ψ(bn (xn , xn+1 ))(en+1 − en ). (11.22)
Then there is a b0 ∈ (0, b(x)) such that kZ(t, y(t))k ≤ 2M for all t ∈ (b0 , b(x)). n=1
Thus, by the usual fundamental theorem of calculus argument,
Z 0 For any x ∈ `2 , only a finite number of terms are non-zero in the above sum
t in a neighborhood of x. Therefor Z : `2 → `2 is a smooth and hence locally
ky(t) − y(t0 )k ≤ kZ(t, y(τ ))k dτ ≤ 2M |t − t0 |

t Lipschitz vector field. Let (y(t), J = (a, b)) denote the maximal solution to
for all t, t0 ∈ (b0 , b(x)). From this it is easy to conclude that y(b(x)−) = ẏ(t) = Z(y(t)) with y(0) = e1 .
limt↑b(x) y(t) exists. If y(b(x)−) ∈ U, by the local existence Theorem 11.4, there
exists δ > 0 and w ∈ C 1 ((b(x) − δ, b(x) + δ), U ) such that Then if the an and bn are chosen appropriately, then b < ∞ and there will
exist tn ↑ b such that y(tn ) is approximately en for all n. So again y(tn ) does
ẇ(t) = Z(t, w(t)) and w(b(x)) = y(b(x)−). not have a limit yet supt∈[0,b) ky(t)k < ∞. The idea is that Z is constructed to
Now define ỹ : (a, b(x) + δ) → U by “blow” the particle from e1 to e2 to e3 to e4 etc. etc. with the time it takes to
travel from en to en+1 being on order 1/2n . The vector field in Eq. (11.22) is
y(t) if t ∈ Jx a first approximation at such a vector field, it may have to be adjusted a little
ỹ(t) = .
w(t) if t ∈ [b(x), b(x) + δ) more to provide an honest example. In this example, we are having problems
because y(t) is “going off in dimensions.”
The reader may now easily show ỹ solves the integral Eq. (11.2) and hence also
solves Eq. 11.1 for t ∈ (a(x), b(x) + δ).2 But this violates the maximality of y Here is another version of Proposition 11.10 which is more useful when
and hence we must have that y(b(x)−) ∈ / U. The assertions for t near a(x) are dim(X) < ∞.
proved similarly.
Proposition 11.13. Let Z ∈ C(J × U, X) be a locally Lipschitz function in x
Example 11.11. Let X = R2 , J = R, U = (x, y) ∈ R2 : 0 < r < 1 where
and y : Jx = (a(x), b(x)) → U be the maximal solution to Eq. (11.1).
r2 = x2 + y 2 and
1 1
Z(x, y) = (x, y) + (−y, x). 1. If b(x) < b, then for every compact subset K ⊂ U there exists TK < b(x)
r 1 − r2 such that y(t) ∈
/ K for all t ∈ [TK , b(x)).
Then the unique solution (x(t), y(t)) to 2. When dim(X) < ∞, we may write this condition as: if b(x) < b, then either
d 1
(x(t), y(t)) = Z(x(t), y(t)) with (x(0), y(0)) = ( , 0) lim sup ky(t)k = ∞ or lim inf dist(y(t), U c ) = 0.
dt 2 t↑b(x) t↑b(x)
is given by
Proof. 1) Suppose that b(x) < b and, for sake of contradiction, there exists
1 1 1 a compact set K ⊂ U and tn ↑ b(x) such that y(tn ) ∈ K for all n. Since K
(x(t), y(t)) = t+ cos , sin
2 1/2 − t 1/2 − t is compact, by passing to a subsequence if necessary, we may assume y∞ :=
for t ∈ J(1/2,0) = (−1/2, 1/2) . Notice that kZ(x(t), y(t))k → ∞ as t ↑ 1/2 and limn→∞ y(tn ) exists in K ⊂ U. By the local existence Theorem 11.4, there exists
dist((x(t), y(t)), U c ) → 0 as t ↑ 1/2. T0 > 0 and δ > 0 such that for each x0 ∈ B (y∞ , δ) there exists a unique solution
w(·, x0 ) ∈ C 1 ((−T0 , T0 ), U ) solving
2
See the argument in Proposition 11.13 for a slightly different method of extending
y which avoids the use of the integral equation (11.2). w(t, x0 ) = Z(t, w(t, x0 )) and w(0, x0 ) = x0 .

11.4 Global Properties 89
Now choose n sufficiently large so that tn ∈ (b(x) − T0 /2, b(x)) and y(tn ) ∈
B (y∞ , δ) . Define ỹ : (a(x), b(x) + T0 /2) → U by

y(t) if t ∈ Jx
ỹ(t) =
w(t − tn , y(tn )) if t ∈ (tn − T0 , b(x) + T0 /2).
wherein we have used (tn − T0 , b(x) + T0 /2) ⊂ (tn − T0 , tn + T0 ). By uniqueness

of solutions to ODE’s ỹ is well defined, ỹ ∈ C 1 ((a(x), b(x) + T0 /2) , X) and ỹ
solves the ODE in Eq. 11.1. But this violates the maximality of y.
2) For each n ∈ N let
Kn := {x ∈ U : kxk ≤ n and dist(x, U c ) ≥ 1/n} .
Then Kn ↑ U and each Kn is a closed bounded set and hence compact if

dim(X) < ∞. Therefore if b(x) < b, by item 1., there exists Tn ∈ [0, b(x))
such that y(t) ∈ / Kn for all t ∈ [Tn , b(x)) or equivalently ky(t)k > n or
dist(y(t), U c ) < 1/n for all t ∈ [Tn , b(x)).
Remark 11.14 (This remark is still rather rough.). In general it is not true that
Fig. 11.2. Manufacturing vector fields where b(x) is discontinuous.
the functions a and b are continuous. For example, let U be the region in R2
described in polar coordinates by r > 0 and 0 < θ < 3π/2 and Z(x, y) = (0, −1)
as in Figure 11.2 below. Then b(x, y) = y for all x ≥ 0 and y > 0 while
b(x, y) = ∞ for all x < 0 and y ∈ R which shows b is discontinuous. On the Proof. Let |J0 | = b0 − a0 , I = J¯0 and E := y(J¯0 ) – a compact subset of U
other hand notice that and let ε > 0 and K < ∞ be given as in Lemma 11.7, i.e. K is the Lipschitz
constant for Z on Eε . Also recall the notation: ∆1 (t) = [0, t] if t > 0 and
{b > t} = {x < 0} ∪ {(x, y) : x ≥ 0, y > t} ∆1 (t) = [t, 0] if t < 0. Suppose that x ∈ Eε , then by Corollary 11.3,
is an open set for all t > 0. An example of a vector field for which b(x) is kφ(t, x) − φ(t, x0 )k ≤ kx − x0 keK|t| ≤ kx − x0 keK|J0 | (11.24)
discontinuous is given in the top left hand corner of Figure 11.2. The map
ψ (r (cos θ, sin θ)) := ln r, tan 23 θ − π2 , would allow the reader to find an for all t ∈ J0 ∩Jx such that such that φ (∆1 (t), x) ⊂ Eε . Letting δ := εe−K|J0 | /2,
example on R2 if so desired. Some calculations shows that Z transferred to R2 and assuming x ∈ B(x0 , δ), the previous equation implies
by the map ψ is given by the new vector
kφ(t, x) − φ(t, x0 )k ≤ ε/2 < ε ∀ t ∈ J0 ∩ Jx 3 φ (∆1 (t), x) ⊂ Eε .
3π 3 3π 3
Z̃(x, y) = −e−x sin + tan−1 (y) , cos + tan−1 (y) . This estimate further shows that φ(t, x) remains bounded and strictly away
8 4 8 4
from the boundary of U for all such t. Therefore, it follows from Proposition
11.8 and “continuous induction3 ” that J0 ⊂ Jx and Eq. (11.24) is valid for all
Theorem 11.15 (Global Continuity). Let Z ∈ C(J × U, X) be a locally t ∈ J0 . This proves Eq. (11.23) with C := eK|J0 | . Suppose that (t0 , x0 ) ∈ D(Z)
Lipschitz function in x. Then D(Z) is an open subset of J ×U and the functions and let 0 ∈ J0 @@ Jx0 such that t0 ∈ J0 and δ be as above. Then we have just
φ : D(Z) → U and φ̇ : D(Z) → U are continuous. More precisely, for all x0 ∈ U shown J0 × B(x0 , δ) ⊂ D(Z) which proves D(Z) is open. Furthermore, since the
and all open intervals J0 such that 0 ∈ J0 @@ Jx0 there exists δ = δ(x0 , J0 , Z) > evaluation map
0 and C = C(x0 , J0 , Z) < ∞ such that for all x ∈ B(x0 , δ), J0 ⊂ Jx and e
(t0 , y) ∈ J0 × BC(J0 , U ) → y(t0 ) ∈ X
3
kφ(·, x) − φ(·, x0 )kBC(J0 ,U ) ≤ C kx − x0 k . (11.23) See the argument in the proof of Proposition 10.12.

is continuous (as the reader should check) it follows that φ = e ◦ (x → φ(·, x)) : 3. For arbitrary t, s ∈ R, etZ ◦ esZ ⊂ e(t+s)Z .
J0 × B(x0 , δ) → U is also continuous; being the composition of continuous
maps. The continuity of φ̇(t0 , x) is a consequence of the continuity of φ and the Proof. Item 1. For simplicity assume that t, s ≥ 0. The case t, s ≤ 0 is left
differential equation 11.1 Alternatively using Eq. (11.2), to the reader. Suppose that x ∈ D(etZ ◦ esZ ). Then by assumption x ∈ D(esZ )
and esZ (x) ∈ D(etZ ). Define the path y(τ ) via:
kφ(t0 , x) − φ(t, x0 )k ≤ kφ(t0 , x) − φ(t0 , x0 )k + kφ(t0 , x0 ) − φ(t, x0 )k τZ
Z t0 e (x) if 0 ≤ τ ≤ s
y(τ ) = .
e(τ −s)Z (x) if s ≤ τ ≤ t + s

≤ C kx − x0 k + kZ(τ, φ(τ, x0 ))k dτ
t
≤ C kx − x0 k + M |t0 − t| It is easy to check that y solves ẏ(τ ) = Z(y(τ )) with y(0) = x. But since,
eτ Z (x) is the maximal solution we must have that x ∈ D(e(t+s)Z ) and y(t +
where C is the constant in Eq. (11.23) and M = supτ ∈J0 kZ(τ, φ(τ, x0 ))k < ∞. s) = e(t+s)Z (x). That is e(t+s)Z (x) = etZ ◦ esZ (x). Hence we have shown that
This clearly shows φ is continuous. etZ ◦ esZ ⊂ e(t+s)Z . To finish the proof of item 1. it suffices to show that
D(e(t+s)Z ) ⊂ D(etZ ◦ esZ ). Take x ∈ D(e(t+s)Z ), then clearly x ∈ D(esZ ). Set
y(τ ) = e(τ +s)Z (x) defined for 0 ≤ τ ≤ t. Then y solves
11.5 Semi-Group Properties of time independent flows
ẏ(τ ) = Z(y(τ )) with y(0) = esZ (x).
To end this chapter we investigate the semi-group property of the flow associated
to the vector-field Z. It will be convenient to introduce the following suggestive But since τ → eτ Z (esZ (x)) is the maximal solution to the above initial valued
notation. For (t, x) ∈ D(Z), set etZ (x) = φ(t, x). So the path t → etZ (x) is the problem we must have that y(τ ) = eτ Z (esZ (x)), and in particular at τ = t,
maximal solution to e(t+s)Z (x) = etZ (esZ (x)). This shows that x ∈ D(etZ ◦esZ ) and in fact e(t+s)Z ⊂
etZ ◦ esZ .
d tZ Item 2. Let x ∈ D(e−tZ ) – again assume for simplicity that t ≥ 0. Set
e (x) = Z(etZ (x)) with e0Z (x) = x.
dt y(τ ) = e(τ −t)Z (x) defined for 0 ≤ τ ≤ t. Notice that y(0) = e−tZ (x) and
This exponential notation will be justified shortly. It is convenient to have the ẏ(τ ) = Z(y(τ )). This shows that y(τ ) = eτ Z (e−tZ (x)) and in particular that
following conventions. x ∈ D(etZ ◦ e−tZ ) and etZ ◦ e−tZ (x) = x. This proves item 2.
Item 3. I will only consider the case that s < 0 and t + s ≥ 0, the other
Notation 11.16 We write f : X → X to mean a function defined on some cases are handled similarly. Write u for t + s, so that t = −s + u. We know that
open subset D(f ) ⊂ X. The open set D(f ) will be called the domain of f. Given etZ = euZ ◦ e−sZ by item 1. Therefore
two functions f : X → X and g : X → X with domains D(f ) and D(g)
respectively, we define the composite function f ◦ g : X → X to be the function etZ ◦ esZ = (euZ ◦ e−sZ ) ◦ esZ .
with domain
Notice in general, one has (f ◦ g) ◦ h = f ◦ (g ◦ h) (you prove). Hence, the above
D(f ◦ g) = {x ∈ X : x ∈ D(g) and g(x) ∈ D(f )} = g −1 (D(f )) displayed equation and item 2. imply that
given by the rule f ◦ g(x) = f (g(x)) for all x ∈ D(f ◦ g). We now write f = g etZ ◦ esZ = euZ ◦ (e−sZ ◦ esZ ) = e(t+s)Z ◦ ID(esZ ) ⊂ e(t+s)Z .
iff D(f ) = D(g) and f (x) = g(x) for all x ∈ D(f ) = D(g). We will also write
f ⊂ g iff D(f ) ⊂ D(g) and g|D(f ) = f.
The following result is trivial but conceptually illuminating partial converse
Theorem 11.17. For fixed t ∈ R we consider etZ as a function from X to X to Theorem 11.17.
with domain D(etZ ) = {x ∈ U : (t, x) ∈ D(Z)}, where D(φ) = D(Z) ⊂ R × U,
D(Z) and φ are defined in Notation 11.9. Conclusions: Proposition 11.18 (Flows and Complete Vector Fields). Suppose U ⊂o
X, φ ∈ C(R × U, U ) and φt (x) = φ(t, x). Suppose φ satisfies:
1. If t, s ∈ R and t · s ≥ 0, then etZ ◦ esZ = e(t+s)Z .
2. If t ∈ R, then etZ ◦ e−tZ = IdD(e−tZ ) . 1. φ0 = IU ,

11.6 Exercises 91
2. φt ◦ φs = φt+s for all t, s ∈ R, and where (y00 , . . . , y0n−1 ) is given in U, has a unique solution for small t ∈ J. Hint:
3. Z(x) := φ̇(0, x) exists for all x ∈ U and Z ∈ C(U, X) is locally Lipschitz. let y(t) = y(t), ẏ(t), . . . , y (n−1) (t) and rewrite Eq. (11.25) as a first order ODE
of the form
Then φt = etZ . ẏ(t) = Z(t, y(t)) with y(0) = (y00 , . . . , y0n−1 ).
Proof. Let x ∈ U and y(t) := φt (x). Then using Item 2., Exercise 11.6. Use the results of Exercises 10.19 and 11.5 to solve
d d d ÿ(t) − 2ẏ(t) + y(t) = 0 with y(0) = a and ẏ(0) = b.
ẏ(t) = |0 y(t + s) = |0 φ(t+s) (x) = |0 φs ◦ φt (x) = Z(y(t)).
ds ds ds
Hint: The 2 × 2 matrix associated to this system, A, has only one eigenvalue
Since y(0) = x by Item 1. and Z is locally Lipschitz by Item 3., we know by
1 and may be written as A = I + B where B 2 = 0.
uniqueness of solutions to ODE’s (Corollary 11.3) that φt (x) = y(t) = etZ (x).
Exercise 11.7 (Non-Homogeneous ODE). Suppose that U ⊂o X is open
and Z : R × U → X is a continuous function. Let J = (a, b) be an interval
and t0 ∈ J. Suppose that y ∈ C 1 (J, U ) is a solution to the “non-homogeneous”
11.6 Exercises differential equation:
Exercise 11.1. Find a vector field Z such that e(t+s)Z is not contained in ẏ(t) = Z(t, y(t)) with y(to ) = x ∈ U. (11.26)
etZ ◦ esZ .
Define Y ∈ C 1 (J − t0 , R × U ) by Y (t) := (t + t0 , y(t + t0 )). Show that Y solves
Definition 11.19. A locally Lipschitz function Z : U ⊂o X → X is said to be the “homogeneous” differential equation
a complete vector field if D(Z) = R × U. That is for any x ∈ U, t → etZ (x) is
Ẏ (t) = Z̃(Y (t)) with Y (0) = (t0 , y0 ), (11.27)
defined for all t ∈ R.
where Z̃(t, x) := (1, Z(x)). Conversely, suppose that Y ∈ C 1 (J − t0 , R × U ) is a
Exercise 11.2. Suppose that Z : X → X is a locally Lipschitz function. As-
solution to Eq. (11.27). Show that Y (t) = (t+t0 , y(t+t0 )) for some y ∈ C 1 (J, U )
sume there is a constant C > 0 such that
satisfying Eq. (11.26). (In this way the theory of non-homogeneous O.D.E.’s may
kZ(x)k ≤ C(1 + kxk) for all x ∈ X. be reduced to the theory of homogeneous O.D.E.’s.)
Exercise 11.8 (Differential Equations with Parameters). Let W be an-
Then Z is complete. Hint: use Gronwall’s Lemma 11.2 and Proposition 11.10.
other Banach space, U × V ⊂o X × W and Z ∈ C(U × V, X) be a locally
Exercise 11.3. Suppose y is a solution to ẏ(t) = |y(t)|
1/2
with y(0) = 0. Show Lipschitz function on U × V. For each (x, w) ∈ U × V, let t ∈ Jx,w → φ(t, x, w)
there exists a, b ∈ [0, ∞] such that denote the maximal solution to the ODE
 1 ẏ(t) = Z(y(t), w) with y(0) = x. (11.28)
 4 (t − b)2 if t≥b
y(t) = 0 if −a < t < b Prove
 1
− 4 (t + a)2 if t ≤ −a. D := {(t, x, w) ∈ R × U × V : t ∈ Jx,w } (11.29)
Exercise 11.4. Using the fact that the solutions to Eq. (11.3) are never 0 if is open in R × U × V and φ and φ̇ are continuous functions on D.
x 6= 0, show that y(t) = 0 is the only solution to Eq. (11.3) with y(0) = 0. Hint: If y(t) solves the differential equation in (11.28), then v(t) := (y(t), w)
solves the differential equation,
Exercise 11.5 (Higher Order ODE). Let X be a Banach space, , U ⊂o X n
and f ∈ C (J × U, X) be a Locally Lipschitz function in x = (x1 , . . . , xn ). Show v̇(t) = Z̃(v(t)) with v(0) = (x, w), (11.30)
the nth ordinary differential equation,
where Z̃(x, w) := (Z(x, w), 0) ∈ X × W and let ψ(t, (x, w)) := v(t). Now apply
y (n) (t) = f (t, y(t), ẏ(t), . . . , y (n−1) (t)) with y (k) (0) = y0k for k < n (11.25) the Theorem 11.15 to the differential equation (11.30).

Exercise 11.9 (Abstract Wave Equation). For A ∈ L(X) and t ∈ R, let
∞
X (−1)n 2n 2n
cos(tA) := t A and
n=0
(2n)!
∞
sin(tA) X (−1)n 2n+1 2n
:= t A .
A n=0
(2n + 1)!
Show that the unique solution y ∈ C 2 (R, X) to
ÿ(t) + A2 y(t) = 0 with y(0) = y0 and ẏ(0) = ẏ0 ∈ X (11.31)
is given by
sin(tA)
y(t) = cos(tA)y0 + ẏ0 .
A
Remark 11.20. Exercise 11.9 can be done by direct verification. Alternatively
and more instructively, rewrite Eq. (11.31) as a first order ODE using Exercise
11.5. In doing so you will be lead to compute etB where B ∈ L(X × X) is given
by
0 I
B= ,
−A2 0

x1
where we are writing elements of X × X as column vectors, . You should
x2
then show
cos(tA) sin(tA)

etB = A
−A sin(tA) cos(tA)
where
∞
X (−1)n 2n+1 2(n+1)
A sin(tA) := t A .
n=0
(2n + 1)!
Exercise 11.10 (Duhamel’s Principle for the Abstract Wave Equa-

tion). Continue the notation in Exercise 11.9, but now consider the ODE,
ÿ(t) + A2 y(t) = f (t) with y(0) = y0 and ẏ(0) = ẏ0 ∈ X (11.32)
where f ∈ C(R, X). Show the unique solution to Eq. (11.32) is given by
Z t
sin(tA) sin((t − τ ) A)
y(t) = cos(tA)y0 + ẏ0 + f (τ )dτ (11.33)
A 0 A
Hint: Again this could be proved by direct calculation. However it is more
instructive to deduce Eq. (11.33) from Exercise 10.21 and the comments in
Remark 11.20.
12
Banach Space Calculus
In this section, X and Y will be Banach space and U will be an open subset Remark 12.4. The linear transformation Λ in Definition 12.3 is necessarily
of X. unique. Indeed if Λ1 is another linear transformation such that Eq. (12.1) holds
with Λ replaced by Λ1 , then
Notation 12.1 (ε, O, and o notation) Let 0 ∈ U ⊂o X, and f : U → Y be
a function. We will write: (Λ − Λ1 )h = o(h),
1. f (x) = ε(x) if limx→0 kf (x)k = 0. i.e.
2. f (x) = O(x) if there are constants C < ∞ and r > 0 such that k(Λ − Λ1 )hk
kf (x)k ≤ Ckxk for all x ∈ B(0, r). This is equivalent to the condition lim sup = 0.
h→0 khk
that lim supx→0 kxk−1 kf (x)k < ∞, where
On the other hand, by definition of the operator norm,
kf (x)k
lim sup := lim sup{kf (x)k : 0 < kxk ≤ r}. k(Λ − Λ1 )hk
x→0 kxk r↓0 lim sup = kΛ − Λ1 k.
h→0 khk
3. f (x) = o(x) if f (x) = ε(x)O(x), i.e. limx→0 kf (x)k/kxk = 0. The last two equations show that Λ = Λ1 .
Example 12.2. Here are some examples of properties of these symbols. Exercise 12.1. Show that a function f : (a, b) → X is a differentiable at
1. A function f : U ⊂o X → Y is continuous at x0 ∈ U if f (x0 + h) = t ∈ (a, b) in the sense of Definition 10.9 iff it is differentiable in the sense of
f (x0 ) + ε(h). Definition 12.3. Also show Df (t)v = v f˙(t) for all v ∈ R.
2. If f (x) = ε(x) and g(x) = ε(x) then f (x) + g(x) = ε(x).
Example 12.5. If T ∈ L (X, Y ) and x, h ∈ X, then
Now let g : Y → Z be another function where Z is another Banach space.
3. If f (x) = O(x) and g(y) = o(y) then g ◦ f (x) = o(x). T (x + h) − T (x) − T h = 0
4. If f (x) = ε(x) and g(y) = ε(y) then g ◦ f (x) = ε(x).
which shows T 0 (x) = T for all x ∈ X.
12.1 The Differential Example 12.6. Assume that GL(X, Y ) is non-empty. Then by Corollary 7.22,
GL(X, Y ) is an open subset of L(X, Y ) and the inverse map f : GL(X, Y ) →
Definition 12.3. A function f : U ⊂o X → Y is differentiable at x0 ∈ U if GL(Y, X), defined by f (A) := A−1 , is continuous. We will now show that f is
there exists a linear transformation Λ ∈ L(X, Y ) such that differentiable and
f (x0 + h) − f (x0 ) − Λh = o(h). (12.1) f 0 (A)B = −A−1 BA−1 for all B ∈ L(X, Y ).
We denote Λ by f 0 (x0 ) or Df (x0 ) if it exists. As with continuity, f is differ- This is a consequence of the identity,
entiable on U if f is differentiable at all points in U.
f (A + H) − f (A) = (A + H)−1 (A − (A + H)) A−1 = −(A + H)−1 HA−1
which may be used to find the estimate,

94 12 Banach Space Calculus
f (A + H) − f (A) + A−1 HA−1 = A−1 − (A + H)−1 HA−1

A(x0 + h)f (x0 + h)
≤ A−1 − (A + H)−1 kHk A−1 = (A(x0 ) + DA(x0 )h + o(h))(f (x0 ) + f 0 (x0 )h + o(h))

kA−1 k3 kHk
2 = A(x0 )f (x0 ) + A(x0 )f 0 (x0 )h + [DA(x0 )h]f (x0 ) + o(h),
2
≤ = O kHk
1 − kA−1 k kHk verifies the product rule holds. This may also be considered as a special case
of Proposition 12.9. Chain Rule. Using f (x0 + h) − f (x0 ) = O(h) (see Eq.
wherein we have used the bound in Eq. (7.10) of Corollary 7.22 for the last
(12.1)) and o(O(h)) = o(h),
inequality.
(g◦f )(x0 + h)
= g(f (x0 )) + g 0 (f (x0 ))(f (x0 + h) − f (x0 )) + o(f (x0 + h) − f (x0 ))
12.2 Product and Chain Rules
= g(f (x0 )) + g 0 (f (x0 ))(Df (x0 )x0 + o(h)) + o(f (x0 + h) − f (x0 )
The following theorem summarizes some basic properties of the differential. = g(f (x0 )) + g 0 (f (x0 ))Df (x0 )h + o(h).
Theorem 12.7. The differential D has the following properties: Converse Chain Rule. Since g is differentiable at y0 = f (x0 ) and g 0 (y0 ) is
invertible,
1. Linearity: D is linear, i.e. D(f + λg) = Df + λDg.
2. Product Rule: If f : U ⊂o X → Y and A : U ⊂o X → L(X, Z) are g(f (x0 + h)) − g(f (x0 ))
differentiable at x0 then so is x → (Af )(x) := A(x)f (x) and = g 0 (f (x0 ))(f (x0 + h) − f (x0 )) + o(f (x0 + h) − f (x0 ))
= g 0 (f (x0 )) [f (x0 + h) − f (x0 ) + o(f (x0 + h) − f (x0 ))] .
D(Af )(x0 )h = (DA(x0 )h)f (x0 ) + A(x0 )Df (x0 )h.
And since g ◦ f is differentiable at x0 ,
3. Chain Rule: If f : U ⊂o X → V ⊂o Y is differentiable at x0 ∈ U, and
g : V ⊂o Y → Z is differentiable at y0 := f (x0 ), then g ◦ f is differentiable (g ◦ f )(x0 + h) − g(f (x0 )) = (g ◦ f )0 (x0 )h + o(h).
at x0 and (g ◦ f )0 (x0 ) = g 0 (y0 )f 0 (x0 ).
4. Converse Chain Rule: Suppose that f : U ⊂o X → V ⊂o Y is contin- Comparing these two equations shows that
uous at x0 ∈ U, g : V ⊂o Y → Z is differentiable y0 := f (ho ), g 0 (y0 ) is
f (x0 + h) − f (x0 ) + o(f (x0 + h) − f (x0 ))
invertible, and g ◦ f is differentiable at x0 , then f is differentiable at x0 and
= g 0 (f (x0 ))−1 [(g ◦ f )0 (x0 )h + o(h)]
0 0 −1 0
f (x0 ) := [g (x0 )] (g ◦ f ) (x0 ). (12.2)
which is equivalent to
Proof. Linearity. Let f, g : U ⊂o X → Y be two functions which are
differentiable at x0 ∈ U and λ ∈ R, then f (x0 + h) − f (x0 ) + o(f (x0 + h) − f (x0 ))
= g 0 (f (x0 ))−1 [(g ◦ f )0 (x0 )h + o(h)]
(f + λg)(x0 + h)
= g 0 (f (x0 ))−1 {(g ◦ f )0 (x0 )h + o(h) − o(f (x0 + h) − f (x0 ))}
= f (x0 ) + Df (x0 )h + o(h) + λ(g(x0 ) + Dg(x0 )h + o(h)
= g 0 (f (x0 ))−1 (g ◦ f )0 (x0 )h + o(h) + o(f (x0 + h) − f (x0 )). (12.3)
= (f + λg)(x0 ) + (Df (x0 ) + λDg(x0 ))h + o(h),
Using the continuity of f, f (x0 + h) − f (x0 ) is close to 0 if h is close to zero,
which implies that (f + λg) is differentiable at x0 and that and hence
1
ko(f (x0 + h) − f (x0 ))k ≤ kf (x0 + h) − f (x0 )k (12.4)
D(f + λg)(x0 ) = Df (x0 ) + λDg(x0 ). 2
for all h sufficiently close to 0. (We may replace 21 by any number α > 0 above.)
Product Rule. The computation,
Taking the norm of both sides of Eq. (12.3) and making use of Eq. (12.4) shows,
for h close to 0, that

12.2 Product and Chain Rules 95
kf (x0 + h) − f (x0 )k Proof. Let us equip X with the norm

1
≤ kg 0 (f (x0 ))−1 (g ◦ f )0 (x0 )kkhk + o(khk) + kf (x0 + h) − f (x0 )k.

kxkX := max kxi kXi .
2
Solving for kf (x0 + h) − f (x0 )k in this last equation shows that If T is continuous then T is continuous at 0. If T is continuous at 0, using
that kT (x)kY ≤ 1 whenever
T (0) = 0, there exists a δ > 0 such kxkX ≤ δ. Now
f (x0 + h) − f (x0 ) = O(h). (12.5) −1 −1
if x ∈ X is arbitrary, let x0 := δ kx1 kX1 x1 , . . . , kxn kXn xn . Then kx0 kX ≤ δ
(This is an improvement, since the continuity of f only guaranteed that f (x0 + and hence
h) − f (x0 ) = ε(h).) Because of Eq. (12.5), we now know that o(f (x0 + h) −
n
!

nY −1
f (x0 )) = o(h), which combined with Eq. (12.3) shows that δ kxi kXi T (x1 , . . . , xn ) = kT (x0 )k ≤ 1

i=1 Y
f (x0 + h) − f (x0 ) = g 0 (f (x0 ))−1 (g ◦ f )0 (x0 )h + o(h),
from which Eq. (12.6) follows with C = δ −n .
i.e. f is differentiable at x0 and f 0 (x0 ) = g 0 (f (x0 ))−1 (g ◦ f )0 (x0 ). n
PnNow suppose that Eq. (12.6) holds. For x, h ∈ X and ε ∈ {0, 1} let |ε| =
Corollary 12.8 (Chain Rule). Suppose that σ : (a, b) → U ⊂o X is differ- i=1 εi and
entiable at t ∈ (a, b) and f : U ⊂o X → Y is differentiable at σ(t) ∈ U. Then xε (h) := ((1 − ε1 ) x1 + ε1 h1 , . . . , (1 − εn ) xn + εn hn ) ∈ X.

f ◦ σ is differentiable at t and
By the multi-linearity of T,
d(f ◦ σ)(t)/dt = f 0 (σ(t))σ̇(t).
X
T (x + h) = T (x1 + h1 , . . . , xn + hn ) = T (xε (h))
Proposition 12.9 (Product Rule II). Suppose that X := X1 × · · · × Xn with n
ε∈{0,1}
each Xi being a Banach space and T : X1 × · · · × Xn → Y is a multilinear map,
n
i.e. X
= T (x) + T (x1 , . . . , xi−1 , hi , xi+1 , . . . , xn )
xi ∈ Xi → T (x1 , . . . , xi−1 , xi , xi+1 , . . . , xn ) ∈ Y
i=1
is linear when x1 , . . . , xi−1 , xi+1 , . . . , xn are held fixed. Then the following are
X
+ T (xε (h)) . (12.8)
equivalent: n
ε∈{0,1} :|ε|≥2
1. T is continuous.
From Eq. (12.6),
2. T is continuous at 0 ∈ X.
3. There exists a constant C < ∞ such that

X
ε = O khk2 ,

n
Y
T (x (h))
kT (x)kY ≤ C kxi kXi (12.6) ε∈{0,1}n :|ε|≥2
i=1
and so it follows from Eq. (12.8) that T 0 (x) exists and is given by Eq. (12.7).
for all x = (x1 , . . . , xn ) ∈ X. This completes the proof since it is trivial to check that T being differentiable
4. T is differentiable at all x ∈ X1 × · · · × Xn . at x ∈ X implies continuity of T at x ∈ X.
Moreover if T the differential of T is given by Exercise 12.2. Let det : L (Rn ) → R be the determinant function on n × n
n
X matrices and for A ∈ L (6 Rn ) we will let Ai denote the ith – column of A and
T 0 (x) h = T (x1 , . . . , xi−1 , hi , xi+1 , . . . , xn ) (12.7) write A = (A1 |A2 | . . . |An ) .
i=1
1. Show det0 (A) exists for all A ∈ L (6 Rn ) and
where h = (h1 , . . . , hn ) ∈ X.

n
X More explicitly
det 0 (A) H = det (A1 | . . . |Ai−1 |Hi |Ai+1 | . . . |An ) (12.9)
i=1
[f 00 (A)B] C = A−1 BA−1 CA−1 + A−1 CA−1 BA−1 . (12.10)
n
for all H ∈ L (R ) . Hint: recall that det (A) is a multilinear function of its
Working inductively one shows f : GL(X) → GL(X) defined by f (A) := A−1
columns.
is C ∞ .
2. Use Eq. (12.9) along with basic properties of the determinant to show
det0 (I) H = tr(H).
3. Suppose now that A ∈ GL (Rn ) , show
12.3 Partial Derivatives
det 0 (A) H = det (A) tr(A−1 H).
Definition 12.13 (Partial or Directional Derivative). Let f : U ⊂o X →
Hint: Notice that det (A + H) = det (A) det I + A−1 H .

Y be a function, x0 ∈ U, and v ∈ X. We say that f is differentiable at x0 in
4. If A ∈ L (Rn ) , show det eA = etr(A) . Hint: use the previous item and d
the direction v iff dt |0 (f (x0 + tv)) =: (∂v f )(x0 ) exists. We call (∂v f )(x0 ) the
Corollary 12.8 to show directional or partial derivative of f at x0 in the direction v.
d Notice that if f is differentiable at x0 , then ∂v f (x0 ) exists and is equal to
det etA = det etA tr(A).

dt f 0 (x0 )v, see Corollary 12.8.
Definition 12.10. Let X and Y be Banach spaces and let L1 (X, Y ) := Proposition 12.14. Let f : U ⊂o X → Y be a continuous function and D ⊂ X
L(X, Y ) and for k ≥ 2 let Lk (X, Y ) be defined inductively by Lk+1 (X, Y ) = be a dense subspace of X. Assume ∂v f (x) exists for all x ∈ U and v ∈ D, and
L(X, Lk (X, Y )). For example L2 (X, Y ) = L(X, L(X, Y )) and L3 (X, Y ) = there exists a continuous function A : U → L(X, Y ) such that ∂v f (x) = A(x)v
L (X, L(X, L(X, Y ))) . for all v ∈ D and x ∈ U ∩ D. Then f ∈ C 1 (U, Y ) and Df = A.
Suppose f : U ⊂o X → Y is a function. If f is differentiable on U, then it Proof. Let x0 ∈ U, ε > 0 such that B(x0 , 2ε) ⊂ U and M := sup{kA(x)k :
makes sense to ask if f 0 = Df : U → L(X, Y ) = L1 (X, Y ) is differentiable. If x ∈ B(x0 , 2ε)} < ∞1 . For x ∈ B(x0 , ε) ∩ D and v ∈ D ∩ B(0, ε), by the
Df is differentiable on U then f 00 = D2 f := DDf : U → L2 (X, Y ). Similarly fundamental theorem of calculus,
we define f (n) = Dn f : U → Ln (X, Y ) inductively.
Z 1
df (x + tv)
Definition 12.11. Given k ∈ N, let C k (U, Y ) denote those functions f : U → f (x + v) − f (x) = dt
Y such that f (j) := Dj f : U → Lj (X, Y ) exists and is continuous for j = 0 dt
Z 1 Z 1
1, 2, . . . , k.
= (∂v f )(x + tv) dt = A(x + tv) v dt. (12.11)
0 0
Example 12.12. Let us continue on with Example 12.6 but now let X = Y to
simplify the notation. So f : GL(X) → GL(X) is the map f (A) = A−1 and For general x ∈ B(x0 , ε) and v ∈ B(0, ε), choose xn ∈ B(x0 , ε) ∩ D and vn ∈
D ∩ B(0, ε) such that xn → x and vn → v. Then
f 0 (A) = −LA−1 RA−1 , i.e. f 0 = −Lf Rf .
1
It should be noted well, unlike in finite dimensions closed and bounded sets need not
where LA B = AB and RA B = AB for all A, B ∈ L(X). As the reader may be compact, so it is not sufficient to choose sufficiently small so that B(x0 , 2) ⊂
∞
easily check, the maps U. Here is a counter example. Let P∞X ≡ H be a Hilbert space, {en }n=1 be an
orthonormal set. Define f (x) ≡ n=1 nφ(kx − en k), where φ is any continuous
A ∈ L(X) → LA , RA ∈ L(L(X)) function on R such that φ(0) = 1 and φ is supported
√ in (−1, 1). Notice that ken −
em k2 = 2 for all m 6= n, so that ken − em k = 2. Using this fact it is rather easy
are linear and bounded. So by the chain and the product rule we find f 00 (A) to check that for any x0 ∈ H, there is an > 0 such that for all x ∈ B(x0 , ),
exists for all A ∈ L(X) and only one term in the sum defining f is non-zero. Hence, f is continuous. However,
f (en ) = n → ∞ as n → ∞.
f 00 (A)B = −Lf 0 (A)B Rf − Lf Rf 0 (A)B .

12.4 Higher Order Derivatives 97
Z 1 d h i
f (xn + vn ) − f (xn ) =
X
A(xn + tvn ) vn dt (12.12) f (x + tv) − f (x) = f x + tv (i) − f x + tv (i−1)
0
i=1
holds for all n. The left side of this last equation tends to f (x + v) − f (x) by d Z
X 1
d
the continuity of f. For the right side of Eq. (12.12) we have = f x + tv (i−1) + stvi ei ds
i=1 0
ds
Z 1 Z 1 d Z 1
k A(x + tv) v dt −

A(xn + tvn ) vn dtk
X
0 0
= tvi ∂i f x + tv (i−1) + stvi ei ds
i=1 0
Z 1
≤ kA(x + tv) − A(xn + tvn ) kkvk dt + M kv − vn k. d
X Z 1
0 = tvi A x + tv (i−1) + stvi ei ei ds.
i=1 0
It now follows by the continuity of A, the fact that kA(x + tv) − A(xn + tvn ) k ≤
M, and the dominated convergence theorem that right side of Eq. (12.12) con- Using the continuity of A, it now follows that
R1
verges to 0 A(x + tv) v dt. Hence Eq. (12.11) is valid for all x ∈ B(x0 , ε) and
d Z 1
v ∈ B(0, ε). We also see that f (x + tv) − f (x) X
lim = vi lim A x + tv (i−1) + stvi ei ei ds
t→0 t i=1
t→0 0
f (x + v) − f (x) − A(x)v = ε(v)v, (12.13)
Xd Z 1
R1
where ε(v) := [A(x + tv) − A(x)] dt. Now = vi A (x) ei ds = A (x) v
0 0
i=1
Z 1
kε(v)k ≤ kA(x + tv) − A(x)k dt which shows ∂v f (x) exists and ∂v f (x) = A (x) v. The result now follows from
0 an application of Proposition 12.14.
≤ max kA(x + tv) − A(x)k → 0 as v → 0,
t∈[0,1]
by the continuity of A. Thus, we have shown that f is differentiable and that 12.4 Higher Order Derivatives
Df (x) = A(x).
It is somewhat inconvenient to work with the Banach spaces Lk (X, Y ) in Def-
d
Corollary 12.15. Suppose now that X = R , f : U ⊂o X → Y be a continuous inition 12.10. For this reason we will introduce an isomorphic Banach space,
function such that ∂i f (x) := ∂ei f (x) exists and is continuous on U for i = Mk (X, Y ).
d
1, 2, . . . , d, where {ei }i=1 is the standard basis for Rd . Then f ∈ C 1 (U, Y ) and
Df (x) ei = ∂i f (x) for all i. Definition 12.16. For k ∈ {1, 2, 3, . . .}, let Mk (X, Y ) denote the set of func-
tions f : X k → Y such that
Proof. For x ∈ U, let A (x) : Rd → Y be the unique linear map such that
A (x) ei = ∂i f (x) for i = 1, 2, . . . , d. Then A : U → L(Rd , Y ) is a continuous 1. For i ∈ {1, 2, . . . , k}, v ∈ X → f hv1 , v2 , . . . , vi−1 , v, vi+1 , . . . , vk i ∈ Y is
map. Now let v ∈ Rd and v (i) := (v1 , v2 , . . . , vi , 0, . . . , 0) for i = 1, 2, . . . , d and linear 2 for all {vi }ni=1 ⊂ X.
v (0) := 0. Then for t ∈ R near 0, using the fundamental theorem of calculus 2. The norm kf kMk (X,Y ) should be finite, where
and the definition of ∂i f (x) ,
kf hv1 , v2 , . . . , vk ikY
kf kMk (X,Y ) := sup{ : {vi }ki=1 ⊂ X \ {0}}.
kv1 kkv2 k · · · kvk k
2
I will routinely write f hv1 , v2 , . . . , vk i rather than f (v1 , v2 , . . . , vk ) when the func-
tion f depends on each of variables linearly, i.e. f is a multi-linear function.

Lemma 12.17. There are linear operators jk : Lk (X, Y ) → Mk (X, Y ) defined Proof. We will prove the theorem by induction on k. We have already proved
inductively as follows: j1 = IdL(X,Y ) (notice that M1 (X, Y ) = L1 (X, Y ) = the theorem when k = 1, see Proposition 12.14. Now suppose that k > 1 and
L(X, Y )) and that the statement of the theorem holds when k is replaced by k − 1. Hence we
know that Dl f (x) = Al (x) for all x ∈ U and l = 1, 2, . . . , k − 1. We are also
(jk+1 A)hv0 , v1 , . . . , vk i = (jk (Av0 ))hv1 , v2 , . . . , vk i ∀vi ∈ X. given that
(Notice that Av0 ∈ Lk (X, Y ).) Moreover, the maps jk are isometric isomor- (∂v1 ∂v2 · · · ∂vk f )(x) = Ak (x)hv1 , v2 , . . . , vk i ∀x ∈ U ∩ D, {vi } ⊂ D. (12.14)
phisms.
Now we may write (∂v2 · · · ∂vk f )(x) as (Dk−1 f )(x)hv2 , v3 , . . . , vk i so that Eq.
Proof. To get a feeling for what jk is let us write out j2 and j3 explicitly. (12.14) may be written as
If A ∈ L2 (X, Y ) = L(X, L(X, Y )), then (j2 A)hv1 , v2 i = (Av1 )v2 and if A ∈
L3 (X, Y ) = L(X, L(X, L(X, Y ))), (j3 A)hv1 , v2 , v3 i = ((Av1 )v2 )v3 for all vi ∈ X. ∂v1 (Dk−1 f )(x)hv2 , v3 , . . . , vk i)
It is easily checked that jk is linear for all k. We will now show by induction
= Ak (x)hv1 , v2 , . . . , vk i ∀x ∈ U ∩ D, {vi } ⊂ D. (12.15)
that jk is an isometry and in particular that jk is injective. Clearly this is true
if k = 1 since j1 is the identity map. For A ∈ Lk+1 (X, Y ), So by the fundamental theorem of calculus, we have that
kjk+1 AkMk+1 (X,Y ) ((Dk−1 f )(x + v1 ) − (Dk−1 f )(x))hv2 , v3 , . . . , vk i
k(jk (Av0 ))hv1 , v2 , . . . , vk ikY Z 1
:= sup{ : {vi }ki=0 ⊂ X \ {0}}
kv0 kkv1 kkv2 k · · · kvk k = Ak (x + tv1 )hv1 , v2 , . . . , vk i dt (12.16)
0
k(jk (Av0 ))kMk (X,Y )
= sup{ : v0 ∈ X \ {0}} for all x ∈ U ∩D and {vi } ⊂ D with v1 sufficiently small. By the same argument
kv0 k
kAv0 kLk (X,Y ) given in the proof of Proposition 12.14, Eq. (12.16) remains valid for all x ∈
= sup{ : v0 ∈ X \ {0}} U and {vi } ⊂ X with v1 sufficiently small. We may write this last equation
kv0 k
alternatively as,
= kAkL(X,Lk (X,Y )) := kAkLk+1 (X,Y ) ,
Z 1
wherein the second to last inequality we have used the induction hypothesis. (Dk−1 f )(x + v1 ) − (Dk−1 f )(x) = Ak (x + tv1 )hv1 , · · · i dt. (12.17)
0
This shows that jk+1 is an isometry provided jk is an isometry. To finish the
proof it suffices to show that jk is surjective for all k. Again this is true for Hence
k = 1. Suppose that jk is invertible for some k ≥ 1. Given f ∈ Mk+1 (X, Y ) we
must produce A ∈ Lk+1 (X, Y ) = L(X, Lk (X, Y )) such that jk+1 A = f. If such (Dk−1 f )(x + v1 ) − (Dk−1 f )(x) − Ak (x)hv1 , · · · i
an equation is to hold, then for v0 ∈ X, we would have jk (Av0 ) = f hv0 , · · · i. Z 1
That is Av0 = jk−1 (f hv0 , · · · i). It is easily checked that A so defined is linear, = [Ak (x + tv1 ) − Ak (x)]hv1 , · · · i dt
0
bounded, and jk+1 A = f.
From now on we will identify Lk with Mk without further mention. In from which we get the estimate,
particular, we will view Dk f as function on U with values in Mk (X, Y ).
k(Dk−1 f )(x + v1 ) − (Dk−1 f )(x) − Ak (x)hv1 , · · · ik ≤ ε(v1 )kv1 k (12.18)
Theorem 12.18 (Differentiability). Suppose k ∈ {1, 2, . . .} and D is a dense
R1
subspace of X, f : U ⊂o X → Y is a function such that (∂v1 ∂v2 · · · ∂vl f )(x) where ε(v1 ) := 0 kAk (x + tv1 ) − Ak (x)k dt. Notice by the continuity of Ak
exists for all x ∈ D ∩ U, {vi }li=1 ⊂ D, and l = 1, 2, . . . k. Further assume that ε(v1 ) → 0 as v1 → 0. Thus it follow from Eq. (12.18) that Dk−1 f is
there exists continuous functions Al : U ⊂o X → Ml (X, Y ) such that such differentiable and that (Dk f )(x) = Ak (x).
that (∂v1 ∂v2 · · · ∂vl f )(x) = Al (x)hv1 , v2 , . . . , vl i for all x ∈ D ∩ U, {vi }li=1 ⊂ D,
and l = 1, 2, . . . k. Then Dl f (x) exists and is equal to Al (x) for all x ∈ U and Example 12.19. Let f : GL(X, Y ) → GL(Y, X) be defined by f (A) := A−1 . We
l = 1, 2, . . . , k. assume that GL(X, Y ) is not empty. Then f is infinitely differentiable and

12.5 Inverse and Implicit Function Theorems 99
1
(Dk f )(A)hV1 , V2 , . . . , Vk i
Z
d d
(∂v F )(x) := |0 F (x + sv) = |0 f (x(t) + sv(t)) dt
X ds ds
= (−1)k {B −1 Vσ(1) B −1 Vσ(2) B −1 · · · B −1 Vσ(k) B −1 }, (12.19) Z 1
0
Z 1
σ d
= |0 f (x(t) + sv(t)) dt = f 0 (x(t))v(t) dt.
0 ds 0
where sum is over all permutations of σ of {1, 2, . . . , k}.
Similar computations show that
Let me check Eq. (12.19) in the case that k = 2. Notice that we have already Z 1
shown that (∂V1 f )(B) = Df (B)V1 = −B −1 V1 B −1 . Using the product rule we (∂v1 ∂v2 · · · ∂vk f )(x) = f (k) (x(t))v1 (t) · · · vk (t) dt =: Ak (x)hv1 , v2 , . . . , vk i.
find that 0
Now for x, y ∈ X,
(∂V2 ∂V1 f )(B) = B −1 V2 B −1 V1 B −1 + B −1 V1 B −1 V2 B −1 =: A2 (B)hV1 , V2 i.
|Ak (x)hv1 , v2 , . . . , vk i − Ak (y)hv1 , v2 , . . . , vk i|
Notice that kA2 (B)hV1 , V2 ik ≤ 2kB −1 k3 kV1 k · kV2 k, so that kA2 (B)k ≤ Z 1
2kB −1 k3 < ∞. Hence A2 : GL(X, Y ) → M2 (L(X, Y ), L(Y, X)). Also ≤ |f (k) (x(t)) − f (k) (y(t))| · |v1 (t) · · · vk (t) |dt
0
k
k(A2 (B) − A2 (C))hV1 , V2 ik ≤ 2kB −1 V2 B −1 V1 B −1 − C −1 V2 C −1 V1 C −1 k 1
Y Z
≤ kvi k |f (k) (x(t)) − f (k) (y(t))|dt,
≤ 2kB −1 V2 B −1 V1 B −1 − B −1 V2 B −1 V1 C −1 k i=1 0
−1 −1 −1 −1 −1 −1
+ 2kB V2 B V1 C −B V2 C V1 C k which shows that
−1 −1 −1 −1 −1 −1
+ 2kB V2 C V1 C −C V2 C V1 C k 1
Z
kAk (x) − Ak (y)k ≤ |f (k) (x(t)) − f (k) (y(t))|dt.
≤ 2kB −1 k2 kV2 kkV1 kkB −1 − C −1 k 0
+ 2kB −1 kkC −1 kkV2 kkV1 kkB −1 − C −1 k This last expression is easily seen to go to zero as y → x in X. Hence Ak is
+ 2kC −1 2
k kV2 kkV1 kkB −1
−C −1
k. continuous. Thus we may apply Theorem 12.18 to conclude that Eq. (12.20) is
valid.
This shows that
kA2 (B) − A2 (C)k ≤ 2kB −1 − C −1 k{kB −1 k2 + kB −1 kkC −1 k + kC −1 k2 }. 12.5 Inverse and Implicit Function Theorems
Since B → B −1 is differentiable and hence continuous, it follows that A2 (B) is In this section, let X be a Banach space, R > 0, U = B = B(0, R) ⊂ X and
also continuous in B. Hence by Theorem 12.18 D2 f (A) exists and is given as ε : U → X be a continuous function such that ε (0) = 0. Our immediate goal is
in Eq. (12.19) to give a sufficient condition on ε so that F (x) := x + ε(x) is a homeomorphism
from U to F (U ) with F (U ) being an open subset of X. Let’s start by looking
Example 12.20. Suppose that f : R → R is a C ∞ – function and F (x) := at the one dimensional case first. So for the moment assume that X = R,
R1
0
f (x(t)) dt for x ∈ X := C([0, 1], R) equipped with the norm kxk := U = (−1, 1), and ε : U → R is C 1 . Then F will be injective iff F is either
maxt∈[0,1] |x(t)|. Then F : X → R is also infinitely differentiable and strictly increasing or decreasing. Since we are thinking that F is a “small”
perturbation of the identity function we will assume that F is strictly increasing,
1
i.e. F 0 = 1 + ε0 > 0. This positivity condition is not so easily interpreted for
Z
(Dk F )(x)hv1 , v2 , . . . , vk i = f (k) (x(t))v1 (t) · · · vk (t) dt, (12.20) operators on a Banach space. However the condition that |ε0 | ≤ α < 1 is easily
0
interpreted in the Banach space setting and it implies 1 + ε0 > 0.
for all x ∈ X and {vi } ⊂ X.
Lemma 12.21. Suppose that U = B = B(0, R) (R > 0) is a ball in X and
To verify this example, notice that ε : B → X is a C 1 function such that kDεk ≤ α < ∞ on U. Then
kε(x) − ε(y)k ≤ αkx − yk for all x, y ∈ U. (12.21)

Proof. By the fundamental theorem of calculus and the chain rule:

Z 1
d
ε(y) − ε(x) = ε(x + t(y − x))dt
0 dt
Z 1
= [Dε(x + t(y − x))](y − x)dt.
0
Therefore, by the triangle inequality and the assumption that kDε(x)k ≤ α on

B,
Z 1 Fig. 12.1. Nesting of F (B(x0 , r)) between B(z0 , (1 − α)r) and B(z0 , (1 + α)r).
kε(y) − ε(x)k ≤ kDε(x + t(y − x))kdt · k(y − x)k ≤ αk(y − x)k.
0
for all x, y ∈ U. This implies

1
Remark 12.22. It is easily checked that if ε : U = B(0, R) → X is C and kx − yk ≤ (1 − α)−1 kF (x) − F (y)k (12.25)
satisfies (12.21) then kDεk ≤ α on U.
which shows F is injective on U and hence shows the inverse function G =
Using the above remark and the analogy to the one dimensional example,
F −1 : V := F (U ) → U is well defined. Moreover, replacing x, y in Eq.
one is lead to the following proposition.
(12.25) by G (x) and G (y) respectively with x, y ∈ V shows
Proposition 12.23. Suppose α ∈ (0, 1), R > 0, U = B(0, R) ⊂o X and ε :
kG (x) − G (y) k ≤ (1 − α)−1 kx − yk for all x, y ∈ V. (12.26)
U → X is a continuous function such that ε (0) = 0 and
Hence G is Lipschitz on V and hence continuous.
kε(x) − ε(y)k ≤ αkx − yk ∀ x, y ∈ U. (12.22)
2. Let x0 ∈ U, r > 0 and z0 = F (x0 ) = x0 + ε(x0 ) be as in item 2. The second
Then F : U → X defined by F (x) := x + ε(x) for x ∈ U satisfies: inclusion in Eq. (12.23) follows from the simple computation:
1. F is an injective map and G = F −1 : V := F (U ) → U is continuous. kF (x0 + h) − z0 k = kh + ε (x0 + h) − ε (x0 )k

2. If x0 ∈ U, z0 = F (x0 ) and r > 0 such the B(x0 , r) ⊂ U, then ≤ khk + kε (x0 + h) − ε (x0 )k
≤ (1 + α) khk < (1 + α) r
B(z0 , (1 − α)r) ⊂ F (B(x0 , r)) ⊂ B(z0 , (1 + α)r). (12.23)
for all h ∈ B (0, r) . To prove the first inclusion in Eq. (12.23) we must find,
In particular, for all r ≤ R,
for every z ∈ B(z0 , (1 − α)r), an h ∈ B (0, r) such that z = F (x0 + h) or
B(0, (1 − α) r) ⊂ F (B(0, r)) ⊂ B(0, (1 + α) r), (12.24) equivalently an h ∈ B (0, r) solving
see Figure 12.1 below. z − z0 = F (x0 + h) − F (x0 ) = h + ε(x0 + h) − ε(x0 ).

3. V := F (U ) is open subset of X and F : U → V is a homeomorphism. Let k := z − z0 and for h ∈ B (0, r) , let δ (h) := ε(x0 + h) − ε(x0 ). With
Proof. this notation it suffices to show for each k ∈ B(z0 , (1 − α)r) there exists
h ∈ B (0, r) such that k = h + δ (h) . Notice that δ (0) = 0 and
1. Using the definition of F and the estimate in Eq. (12.22),
kδ (h1 ) − δ (h2 )k = kε(x0 + h1 ) − ε(x0 + h2 )k ≤ α kh1 − h2 k (12.27)
kx − yk = k(F (x) − F (y)) − (ε(x) − ε(y))k
for all h1 , h2 ∈ B (0, r) . We are now going to solve the equation k = h+δ (h)
≤ kF (x) − F (y)k + kε(x) − ε(y)k
for h by the method of successive approximations starting with h0 = 0 and
≤ kF (x) − F (y)k + αk(x − y)k then defining hn inductively by

12.5 Inverse and Implicit Function Theorems 101
hn+1 = k − δ (hn ) . (12.28) Then ε(h) = o(h) as h → 0 and
A simple induction argument using Eq. (12.27) shows that kε(h0 ) − ε(h)k ≤ α (R) kh0 − hk for all h, h0 ∈ B X (0, R). (12.33)
khn+1 − hn k ≤ αn kkk for all n ∈ N0 If α (R) < 1 (which may be achieved by shrinking R if necessary), then T 0 (x)
is invertible for all x ∈ B X (x0 , R) and
and in particular that
0 −1 1 0
T (x0 )−1

N −1 N −1 sup T (x)
L(Y,X)
≤ L(Y,X)
. (12.34)
1 − α (R)
X X
x∈B X (x0 ,R)
khN k = (hn+1 − hn ) ≤ khn+1 − hn k

n=0 n=0 −1
N −1 Proof. By definition of T 0 (x0 ) and using T 0 (x0 ) exists,
N
X
n 1−α
≤ α kkk = kkk . (12.29)
1−α T (x0 + h) − T (x0 ) = T 0 (x0 )h + o(h)
n=0
Since kkk < (1 − α) r, this implies that khN k < r for all N showing the from which it follows that ε(h) = o(h). In fact by the fundamental theorem of
approximation procedure is well defined. Let calculus, Z 1
T 0 (x0 )−1 T 0 (x0 + th) − I hdt

∞
X ε(h) =
0
h := lim hn = (hn+1 − hn ) ∈ X
N →∞
n=0 but we will not use this here. Let h, h0 ∈ B X (0, R) and apply the fundamental
theorem of calculus to t → T (x0 + t(h0 − h)) to conclude
which exists since the sum in the previous equation is absolutely convergent.
Passing to the limit in Eqs. (12.29) and (12.28) shows that khk ≤ (1 − ε(h0 ) − ε(h) = T 0 (x0 )−1 [T (x0 + h0 ) − T (x0 + h)] − (h0 − h)
α)−1 kkk < r and h = k − δ (h) , i.e. h ∈ B (0, r) solves k = h + δ (h) as Z 1
desired. T 0 (x0 )−1 T 0 (x0 + t(h0 − h)) − I dt (h0 − h).

=
3. Given x0 ∈ U, the first inclusion in Eq. (12.23) shows that z0 = F (x0 ) 0
is in the interior of F (U ) . Since z0 ∈ F (U ) was arbitrary, it follows that
Taking norms of this equation gives
V = F (U ) is open. The continuity of the inverse function has already been
proved in item 1. Z 1
kε(h0 ) − ε(h)k ≤
0
T (x0 )−1 T 0 (x0 + t(h0 − h)) − I dt kh0 − hk

0
For the remainder of this section let X and Y be two Banach spaces, U ⊂o X, ≤ α (R) kh0 − hk
k ≥ 1, and f ∈ C k (U, Y ).
It only remains to prove Eq. (12.34), so suppose now that α (R) < 1. Then by
Lemma 12.24. Suppose x0 ∈ U, R > 0 is such that B X (x0 , R) ⊂ U and Proposition 7.21, T 0 (x0 )−1 T 0 (x) = I − I − T 0 (x0 )−1 T 0 (x) is invertible and
T : B X (x0 , R) → Y is a C 1 – function such that T 0 (x0 ) is invertible. Let

0 −1 1
T (x0 )−1 T 0 (x) ≤ for all x ∈ B X (x0 , R).
0
T (x0 )−1 T 0 (x) − I

α (R) := sup L(X)
(12.30) 1 − α (R)
x∈B X (x0 ,R)
Since T 0 (x) = T 0 (x0 ) T 0 (x0 )−1 T 0 (x) this implies T 0 (x) is invertible and

and ε ∈ C 1 B X (0, R), X be defined by
0 −1 −1 0 1
T 0 (x0 )−1 T 0 (x) T (x0 )−1 ≤
0
T (x0 )−1

0 −1 T (x) =
ε (h) = T (x0 ) [T (x0 + h) − T (x0 )] − h (12.31)
1 − α (R)
so that
for all x ∈ B X (x0 , R).
T (x0 + h) = T (x0 ) + T 0 (x0 ) (h + ε(h)) . (12.32)

Theorem 12.25 (Inverse Function Theorem). Suppose U ⊂o X, k ≥ 1 1. Using Eqs. (12.32) and (12.24),
and T ∈ C k (U, Y ) such that T 0 (x) is invertible for all x ∈ U. Further assume
T B X (x0 , r) = T (x0 ) + T 0 (x0 ) (I + ε) B X (0, r)

x0 ∈ U and R > 0 such that B X (x0 , R) ⊂ U. (12.37)
0 X
⊂ T (x0 ) + T (x0 ) B (0, (1 + α (r)) r)
1. For all r ≤ R,
T (B X (x0 , r)) ⊂ T (x0 ) + T 0 (x0 ) B X (0, (1 + α (r))r) . (12.35)
2. Now assume α (R) < 1, then by Eqs. (12.37) and (12.24),
2. If we further assume that
T (x0 ) + T 0 (x0 ) B X (0, (1 − α (r)) r)
0
T (x0 )−1 T 0 (x) − I < 1,

α (R) := sup ⊂ T (x0 ) + T 0 (x0 ) (I + ε) B X (0, r) = T B X (x0 , r)

x∈B X (x0 ,R)

which may always be achieved by taking R sufficiently small, then
3. Notice that h ∈ X → T (x0 ) + T 0 (x0 ) h ∈ Y is a homeomorphism. The fact
T (x0 ) + T 0 (x0 ) B X (0, (1 − α (r))r) ⊂ T (B X (x0 , r)) (12.36) that T is an open map follows easily from Eq. (12.36) which shows that
T (x0 ) is interior of T (W ) for any W ⊂o X with x0 ∈ W.
for all r ≤ R, see Figure 12.2. 4. The fact that T |B X (x0 ,R) : B X (x0 , R) → T (B X (x0 , R)) is invertible with
3. T : U → Y is an open mapping, in particular V := T (U ) ⊂o Y. a continuous inverse follows from Eq. (12.32) and Proposition 12.23. It
4. Again if R is sufficiently small so that α (R) < 1, then now follows from the converse to the chain rule, Theorem 12.7, that g :=
T |−1 X
X
T |B X (x0 ,R) : B X (x0 , R) → T (B X (x0 , R)) is invertible and X
B (x0 ,R)
: T B (x0 , R) → B (x0 , R) is differentiable and
T |−1

B X (x0 ,R)
: T B X (x0 , R) → B X (x0 , R) is a C k – map.
−1
g 0 (y) = [T 0 (g (y))] for all y ∈ T B X (x0 , R) .

5. If T is injective, then T −1 : V → U is also a C k – map and
0
T −1 (y) = T 0 (T −1 (y))
−1
for all y ∈ V. This equation shows g is C 1 . Now suppose that k ≥ 2. Since T 0 ∈
C k−1 (B, L(X)) and i(A) := A−1 is a smooth map by Example 12.19,
g 0 = i ◦ T 0 ◦ g is C 1 , i.e. g is C 2 . If k ≥ 2, we may use the same argu-
ment to now show g is C 3 . Continuing this way inductively, we learn g is
Ck.
5. Since differentiability and smoothness is local, the assertion in item 5. fol-
lows directly from what has already been proved.
Theorem 12.26 (Implicit Function Theorem). Suppose that X, Y, and

W are three Banach spaces, k ≥ 1, A ⊂ X × Y is an open set, (x0 , y0 ) is
a point in A, and f : A → W is a C k – map such f (x0 , y0 ) = 0. Assume
that D2 f (x0 , y0 ) := D(f (x0 , ·))(y0 ) : Y → W is a bounded invertible linear
transformation. Then there is an open neighborhood U0 of x0 in X such that
for all connected open neighborhoods U of x0 contained in U0 , there is a unique
Fig. 12.2. The nesting of T (B X (x0 , r)) between T (x0 ) + T 0 (x0 ) B X (0, (1 − α (r))r) continuous function u : U → Y such that u(x0 ) = yo , (x, u(x)) ∈ A and
andT (x0 ) + T 0 (x0 ) B X (0, (1 + α (r))r) . f (x, u(x)) = 0 for all x ∈ U. Moreover u is necessarily C k and
Du(x) = −D2 f (x, u(x))−1 D1 f (x, u(x)) for all x ∈ U. (12.38)

Proof. Let ε ∈ C 1 B X (0, R), X be as defined in Eq. (12.31).

12.6 Smooth Dependence of ODE’s on Initial Conditions* 103
Proof. By replacing f by (x, y) → D2 f (x0 , y0 )−1 f (x, y) if necessary, we From Eq. (12.40) and the continuity of u0 and u1 , limx̃→x R(x̃) = D2 f (x, u0 (x))
may assume with out loss of generality that W = Y and D2 f (x0 , y0 ) = IY . which is invertible.3 Thus R(x̃) is invertible for all x̃ sufficiently close to x which
Define F : A → X × Y by F (x, y) := (x, f (x, y)) for all (x, y) ∈ A. Notice that combined with Eq. (12.39) implies that u1 (x̃) = u0 (x̃) for all x̃ sufficiently close
to x. Since x ∈ O was arbitrary, we have shown that O is open.
I D1 f (x, y)
DF (x, y) =
0 D2 f (x, y)
12.6 Smooth Dependence of ODE’s on Initial Conditions*
which is invertible iff D2 f (x, y) is invertible and if D2 f (x, y) is invertible then
In this subsection, let X be a Banach space, U ⊂o X and J be an open interval
I −D1 f (x, y)D2 f (x, y)−1

DF (x, y)−1 = . with 0 ∈ J.
0 D2 f (x, y)−1
Lemma 12.27. If Z ∈ C(J × U, X) such that Dx Z(t, x) exists for all (t, x) ∈
Since D2 f (x0 , y0 ) = I is invertible, the inverse function theorem guarantees J × U and Dx Z(t, x) ∈ C(J × U, X) then Z is locally Lipschitz in x, see Defi-
that there exists a neighborhood U0 of x0 and V0 of y0 such that U0 × V0 ⊂ A, nition 11.6.
F (U0 × V0 ) is open in X × Y, F |(U0 ×V0 ) has a C k –inverse which we call F −1 .
Let π2 (x, y) := y for all (x, y) ∈ X × Y and define C k – function u0 on U0 by Proof. Suppose I @@ J and x ∈ U. By the continuity of DZ, for every t ∈ I
u0 (x) := π2 ◦ F −1 (x, 0). Since F −1 (x, 0) = (x̃, u0 (x)) iff there an open neighborhood Nt of t ∈ I and εt > 0 such that B(x, εt ) ⊂ U and
(x, 0) = F (x̃, u0 (x)) = (x̃, f (x̃, u0 (x))), sup {kDx Z(t0 , x0 )k : (t0 , x0 ) ∈ Nt × B(x, εt )} < ∞.
it follows that x = x̃ and f (x, u0 (x)) = 0. Thus By the compactness of I, there exists a finite subset Λ ⊂ I such that I ⊂ ∪t∈I Nt .
Let ε(x, I) := min {εt : t ∈ Λ} and
(x, u0 (x)) = F −1 (x, 0) ∈ U0 × V0 ⊂ A
K(x, I) := sup {kDZ(t, x0 )k(t, x0 ) ∈ I × B(x, ε(x, I))} < ∞.
and f (x, u0 (x)) = 0 for all x ∈ U0 . Moreover, u0 is C k being the composition
of the C k – functions, x → (x, 0), F −1 , and π2 . So if U ⊂ U0 is a connected set Then by the fundamental theorem of calculus and the triangle inequality,
containing x0 , we may define u := u0 |U to show the existence of the functions Z 1
u as described in the statement of the theorem. The only statement left to kZ(t, x1 ) − Z(t, x0 )k ≤ kDx Z(t, x0 + s(x1 − x0 )k ds kx1 − x0 k
prove is the uniqueness of such a function u. Suppose that u1 : U → Y is 0
another continuous function such that u1 (x0 ) = y0 , and (x, u1 (x)) ∈ A and ≤ K(x, I)kx1 − x0 k
f (x, u1 (x)) = 0 for all x ∈ U. Let
for all x0 , x1 ∈ B(x, ε(x, I)) and t ∈ I.
O := {x ∈ U |u(x) = u1 (x)} = {x ∈ U |u0 (x) = u1 (x)}. Theorem 12.28 (Smooth Dependence of ODE’s on Initial Condi-
tions). Let X be a Banach space, U ⊂o X, Z ∈ C(R × U, X) such that
Clearly O is a (relatively) closed subset of U which is not empty since x0 ∈ O.
Dx Z ∈ C(R × U, X) and φ : D(Z) ⊂ R × X → X denote the maximal so-
Because U is connected, if we show that O is also an open set we will have
lution operator to the ordinary differential equation
shown that O = U or equivalently that u1 = u0 on U. So suppose that x ∈ O,
i.e. u0 (x) = u1 (x). For x̃ near x ∈ U, ẏ(t) = Z(t, y(t)) with y(0) = x ∈ U, (12.41)
0 = 0 − 0 = f (x̃, u0 (x̃)) − f (x̃, u1 (x̃)) = R(x̃)(u1 (x̃) − u0 (x̃)) (12.39) see Notation 11.9 and Theorem 11.15. Then φ ∈ C 1 (D(Z), U ), ∂t Dx φ(t, x)
exists and is continuous for (t, x) ∈ D(Z) and Dx φ(t, x) satisfies the linear
where differential equation,
Z 1
R(x̃) := D2 f ((x̃, u0 (x̃) + t(u1 (x̃) − u0 (x̃)))dt. (12.40) 3
0 Notice that DF (x, u0 (x)) is invertible for all x ∈ U0 since F |U0 ×V0 has a C 1 inverse.
Therefore D2 f (x, u0 (x)) is also invertible for all x ∈ U0 .

d Proof. Let h ∈ Y be sufficiently small and τ ∈ J, then by fundamental
Dx φ(t, x) = [(Dx Z) (t, φ(t, x))]Dx φ(t, x) with Dx φ(0, x) = IX (12.42)
dt theorem of calculus,
for t ∈ Jx .
Z(τ,y(τ ) + h(τ )) − Z(τ, y(τ ))
Proof. Let x0 ∈ U and J be an open interval such that 0 ∈ J ⊂ J¯ @@ Jx0 ,
Z 1
y0 := y(·, x0 )|J and = [Dx Z(τ, y(τ ) + rh(τ )) − Dx Z(τ, y(τ ))]dr
0
Oε := {y ∈ BC(J, U ) : ky − y0 k∞ < ε} ⊂o BC(J, X). and therefore,
By Lemma 12.27, Z is locally Lipschitz and therefore Theorem 11.15 is appli- f (y + h) − f (y) − Λy h(t)
cable. By Eq. (11.23) of Theorem 11.15, there exists ε > 0 and δ > 0 such that Z t
G : B(x0 , δ) → Oε defined by G(x) := φ(·, x)|J is continuous. By Lemma 12.29 = [Z(τ, y(τ ) + h(τ )) − Z(τ, y(τ )) − Dx Z(τ, y(τ ))h(τ ) ] dτ
0
below, for ε > 0 sufficiently small the function F : Oε → BC(J, X) defined by Z t Z 1
Z · = dτ dr[Dx Z(τ, y(τ ) + rh(τ )) − Dx Z(τ, y(τ ))]h(τ ).
F (y) := y − Z(t, y(t))dt. (12.43) 0 0
0
Therefore,
1
is C and k(f (y + h) − f (y) − Λy h)k∞ ≤ khk∞ δ(h) (12.45)
Z ·
DF (y)v = v − Dy Z(t, y(t))v(t)dt. (12.44) where
0
Z Z 1
By the existence and uniqueness Theorem 10.22 for linear ordinary differential δ(h) := dτ dr kDx Z(τ, y(τ ) + rh(τ )) − Dx Z(τ, y(τ ))k .
equations, DF (y) is invertible for any y ∈ BC(J, U ). By the definition of φ, J 0
F (G(x)) = h(x) for all x ∈ B(x0 , δ) where h : X → BC(J, X) is defined by
With the aide of Lemmas 12.27 and Lemma 11.7,
h(x)(t) = x for all t ∈ J, i.e. h(x) is the constant path at x. Since h is a bounded
linear map, h is smooth and Dh(x) = h for all x ∈ X. We may now apply the (r, τ, h) ∈ [0, 1] × J × Y → kDx Z(τ, y(τ ) + rh(τ ))k
converse to the chain rule in Theorem 12.7 to conclude G ∈ C 1 (B(x0 , δ), O) and
DG(x) = [DF (G(x))]−1 Dh(x) or equivalently, DF (G(x))DG(x) = h which in is bounded for small h provided ε > 0 is sufficiently small. Thus it follows from
turn is equivalent to the dominated convergence theorem that δ(h) → 0 as h → 0 and hence Eq.
Z t (12.45) implies f 0 (y) exists and is given by Λy . Similarly,
Dx φ(t, x) − [DZ(φ(τ, x)]Dx φ(τ, x) dτ = IX .
0 ||f 0 (y + h) − f 0 (y)||op
Z
As usual this equation implies Dx φ(t, x) is differentiable in t, Dx φ(t, x) is con- ≤ kDx Z(τ, y(τ ) + h(τ )) − Dx Z(τ, y(τ ))k dτ → 0 as h → 0
tinuous in (t, x) and Dx φ(t, x) satisfies Eq. (12.42). J
0
Lemma 12.29. Continuing the notation used in the proof of Theorem 12.28 showing f is continuous.
and further let Z · Remark 12.30. If Z ∈ C k (U, X), then an inductive argument shows that
f (y) := Z(τ, y(τ )) dτ for y ∈ Oε . φ ∈ C k (D(Z), X). For example if Z ∈ C 2 (U, X) then (y(t), u(t)) :=
0 (φ(t, x), Dx φ(t, x)) solves the ODE,
1
Then f ∈ C (Oε , Y ) and for all y ∈ Oε ,
d
Z · (y(t), u(t)) = Z̃ ((y(t), u(t))) with (y(0), u(0)) = (x, IdX )
0 dt
f (y)h = Dx Z(τ, y(τ ))h(τ ) dτ =: Λy h.
0 where Z̃ is the C 1 – vector field defined by

12.7 Existence of Periodic Solutions 105
Z 2π
Z̃ (x, u) = (Z(x), Dx Z(x)u) .
Rt
h0 1 − e−c(f ) = e− τ
f (s)ds
k(τ )dτ
0
Therefore Theorem 12.28 may be applied to this equation to deduce: Dx2 φ(t, x)
and Dx2 φ̇(t, x) exist and are continuous. We may now differentiate Eq. (12.42) where Z 2π Z 2π
to find Dx2 φ(t, x) satisfies the ODE, c(f ) = f (τ )dτ = p0 (y(τ ))dτ > 0.
0 0
d 2 1
The unique solution h ∈ C2π (R, R) to P 0 (y)h = k is given by
D φ(t, x) = [ ∂Dx φ(t,x) Dx Z (t, φ(t, x))]Dx φ(t, x)
dt x Z 2π R Z t R
−1 R t
+ [(Dx Z) (t, φ(t, x))]Dx2 φ(t, x)

−c(f ) − 0 f (τ )dτ − τt f (s)ds t
h(t) = 1 − e e e k(τ )dτ + e− τ f (s)ds k(τ )dτ
0 0
with Dx2 φ(0, x) = 0. −1 Rt
Z 2π Rt
Z t Rt
= 1 − e−c(f ) e− 0
f (s)ds
e− τ
f (s)ds
k(τ )dτ + e− τ
f (s)ds
k(τ )dτ.
0 0
12.7 Existence of Periodic Solutions Therefore P 0 (y) is invertible for all y. Hence by the inverse function Theorem
12.25, P : Ũ → Ṽ is an open mapping which is locally invertible.
A detailed discussion of the inverse function theorem on Banach and Frechét Step 2. Let us now prove P : Ũ → Ṽ is injective. For this suppose y1 , y2 ∈ Ũ
spaces may be found in Richard Hamilton’s, “The Inverse Function Theorem of such that P (y1 ) = g = P (y2 ) and let z = y2 − y1 . Since
Nash and Moser.” The applications in this section are taken from this paper.
In what follows we say f ∈ C2π k k
(R, (c, d)) if f ∈ C2π (R, (c, d)) and f is 2π – ż(t) + p(y2 (t)) − p(y1 (t)) = g(t) − g(t) = 0,
periodic, i.e. f (x + 2π) = f (x) for all x ∈ R. if tm ∈ R is point where z(tm ) takes on its maximum, then ż(tm ) = 0 and hence
Theorem 12.31 (Taken from Hamilton, p. 110.). Let p : U := (a, b) → p(y2 (tm )) − p(y1 (tm )) = 0.
V := (c, d) be a smooth function with p0 > 0 on (a, b). For every g ∈
∞ ∞ Since p is increasing this implies y2 (tm ) = y1 (tm ) and hence z(tm ) = 0. This
C2π (R, (c, d)) there exists a unique function y ∈ C2π (R, (a, b)) such that
shows z(t) ≤ 0 for all t and a similar argument using a minimizer of z shows
ẏ(t) + p(y(t)) = g(t). z(t) ≥ 0 for all t. So we conclude y1 = y2 .
0 0 1
Step 3. Let W := P (Ũ ), we wish to show W = Ṽ . By step 1., we know
Proof. Let Ṽ := C2π (R, (c, d)) ⊂o C2π (R, R) and Ũ ⊂o C2π (R, (a, b)) be W is an open subset of Ṽ and since Ṽ is connected, to finish the proof it
given by suffices to show W is relatively closed in Ṽ . So suppose yj ∈ Ũ such that
1
Ũ := y ∈ C2π

(R, R) : a < y(t) < b & c < ẏ(t) + p(y(t)) < d ∀ t . gj := P (yj ) → g ∈ Ṽ . We must now show g ∈ W, i.e. g = P (y) for some y ∈ W.
If tm is a maximizer of yj , then ẏj (tm ) = 0 and hence gj (tm ) = p(yj (tm )) < d
The proof will be completed by showing P : Ũ → Ṽ defined by and therefore yj (tm ) < b because p is increasing. A similar argument works for
the minimizers then allows us to conclude Ran (p ◦ yj ) ⊂ Ran (gj ) @@ (c, d) for
P (y)(t) = ẏ(t) + p(y(t)) for y ∈ Ũ and t ∈ R all j. Since gj is converging uniformly to g, there exists c < γ < δ < d such that
Ran(p ◦ yj ) ⊂ Ran(gj ) ⊂ [γ, δ] for all j. Again since p0 > 0,
is bijective. Note that if P (y) is smooth then so is y.
Step 1. The differential of P is given by P 0 (y)h = ḣ + p0 (y)h, see Exercise Ran(yj ) ⊂ p−1 ([γ, δ]) = [α, β] @@ (a, b) for all j.
12.8. We will now show that the linear mapping P 0 (y) is invertible. Indeed let In particular sup {|ẏj (t)| : t ∈ R and j} < ∞ since
f = p0 (y) > 0, then the general solution to the Eq. ḣ + f h = k is given by
Z t R ẏj (t) = gj (t) − p(yj (t)) ⊂ [γ, δ] − [γ, δ] (12.46)
Rt t
h(t) = e− 0 f (τ )dτ h0 + e− τ f (s)ds k(τ )dτ which is a compact subset of R. The Ascoli-Arzela Theorem (see Theoerem
0
14.29 below) now allows us to assume, by passing to a subsequence if necessary,
0
where h0 is a constant. We wish to choose h0 so that h(2π) = h0 , i.e. so that that yj is converging uniformly to y ∈ C2π (R, [α, β]). It now follows that

m−1 ∞
ẏj (t) = gj (t) − p(yj (t)) → g − p(y) X
k n
X αn
ρ(xm , xn ) ≤ α ρ(x1 , x0 ) ≤ ρ(x1 , x0 )α αl = ρ(x1 , x0 ) .
1 0
1−α
uniformly in t. Hence we concluded that y ∈ C2π (R, R) ∩ C2π (R, [α, β]), ẏj → y k=n l=0
and P (y) = g. This has proved that g ∈ W and hence that W is relatively This last equation shows that ρ(xm , xn ) → 0 as m, n → ∞, i.e. {xn }∞
n=0 is a
closed in Ṽ . Cauchy sequence.
Corollary 12.33 (Contraction Mapping Principle II). Suppose that
(X, ρ) is a complete metric space and S : X → X is a continuous map such
12.8 Contraction Mapping Principle that S (n) is a contraction for some n ∈ N. Here
n times
Some of the arguments uses in this chapter and in Chapter 11 may be abstracted z }| {
(n)
to a general principle of finding fixed points on a complete metric space. This S := S ◦ S ◦ . . . ◦ S
is the content of this section. and we are assuming there exists α ∈ (0, 1) such that ρ(S (n) (x), S (n) (y)) ≤
Theorem 12.32 (Contraction Mapping Principle). Suppose that (X, ρ) αρ(x, y) for all x, y ∈ X. Then S has a unique fixed point in X.
is a complete metric space and S : X → X is a contraction, i.e. there exists Proof. Let T := S (n) , then T : X → X is a contraction and hence T has a
α ∈ (0, 1) such that ρ(S(x), S(y)) ≤ αρ(x, y) for all x, y ∈ X. Then S has unique fixed point x ∈ X. Since any fixed point of S is also a fixed point of T,
a unique fixed point in X, i.e. there exists a unique point x ∈ X such that we see if S has a fixed point then it must be x. Now
S(x) = x.
T (S(x)) = S (n) (S(x)) = S(S (n) (x)) = S(T (x)) = S(x),
Proof. For uniqueness suppose that x and x0 are two fixed points of S, then which shows that S(x) is also a fixed point of T. Since T has only one fixed
0 0 0 point, we must have that S(x) = x. So we have shown that x is a fixed point
ρ(x, x ) = ρ(S(x), S(x )) ≤ αρ(x, x ).
of S and this fixed point is unique.
Therefore (1 − α)ρ(x, x0 ) ≤ 0 which implies that ρ(x, x0 ) = 0 since 1 − α > 0. Lemma 12.34. Suppose that (X, ρ) is a complete metric space, n ∈ N, Z is
Thus x = x0 . For existence, let x0 ∈ X be any point in X and define xn ∈ X a topological space, and α ∈ (0, 1). Suppose for each z ∈ Z there is a map
inductively by xn+1 = S(xn ) for n ≥ 0. We will show that x := limn→∞ xn Sz : X → X with the following properties:
exists in X and because S is continuous this will imply, (n) (n)
Contraction property ρ(Sz (x), Sz (y)) ≤ αρ(x, y) for all x, y ∈ X and z ∈ Z.
x = lim xn+1 = lim S(xn ) = S( lim xn ) = S(x), Continuity in z For each x ∈ X the map z ∈ Z → Sz (x) ∈ X is continuous.
n→∞ n→∞ n→∞
By Corollary 12.33 above, for each z ∈ Z there is a unique fixed point
showing x is a fixed point of S. So to finish the proof, because X is complete, G(z) ∈ X of Sz .
it suffices to show {xn }∞
n=1 is a Cauchy sequence in X. An easy inductive com- Conclusion: The map G : Z → X is continuous.
putation shows, for n ≥ 0, that (n)
Proof. Let Tz := Sz . If z, w ∈ Z, then
ρ(xn+1 , xn ) = ρ(S(xn ), S(xn−1 )) ≤ αρ(xn , xn−1 ) ≤ · · · ≤ αn ρ(x1 , x0 ). ρ(G(z), G(w)) = ρ(Tz (G(z)), Tw (G(w)))
Another inductive argument using the triangle inequality shows, for m > n, ≤ ρ(Tz (G(z)), Tw (G(z))) + ρ(Tw (G(z)), Tw (G(w)))
that, ≤ ρ(Tz (G(z)), Tw (G(z))) + αρ(G(z), G(w)).
m−1
X Solving this inequality for ρ(G(z), G(w)) gives
ρ(xm , xn ) ≤ ρ(xm , xm−1 ) + ρ(xm−1 , xn ) ≤ · · · ≤ ρ(xk+1 , xk ). 1
k=n ρ(G(z), G(w)) ≤ ρ(Tz (G(z)), Tw (G(z))).
1−α
Combining the last two inequalities gives (using again that α ∈ (0, 1)), Since w → Tw (G(z)) is continuous it follows from the above equation that
G(w) → G(z) as w → z, i.e. G is continuous.

12.9 Exercises 107
12.9 Exercises Use Exercise 12.4 to conclude that w is C 1 and that w0 (t, 0) := dw(t, s)/ds|s=0
satisfies the differential equation,
Exercise 12.3. Suppose that A : R → L(X) is a continuous function and
d 0
V : R → L(X) is the unique solution to the linear differential equation w (t, 0) = Aw0 (t, 0) + BetA with w(0, 0) = 0 ∈ L(X). (12.52)
dt
V̇ (t) = A(t)V (t) with V (0) = I. (12.47) Solve this equation by Duhamel’s principle (Exercise 10.22) and then apply
Proposition 12.14 to conclude that f is differentiable with differential given by
Assuming that V (t) is invertible for all t ∈ R, show that V −1 (t) := [V (t)]−1
Eq. (12.50).
must solve the differential equation
Exercise 12.6 (Local ODE Existence). Let Sx be defined as in Eq. (11.15)
d −1
V (t) = −V −1 (t)A(t) with V −1 (0) = I. (12.48) from the proof of Theorem 11.4. Verify that Sx satisfies the hypothesis of Corol-
dt lary 12.33. In particular we could have used Corollary 12.33 to prove Theorem
See Exercise 10.12 as well. 11.4.
Exercise 12.4 (Differential Equations with Parameters). Let W be an- Exercise 12.7 (Local ODE Existence Again). Let J = (−1, 1) , Z ∈
other Banach space, U × V ⊂o X × W and Z ∈ C 1 (U × V, X). For each C 1 (X, X), Y := BC(J, X) and for y ∈ Y and s ∈ J let ys ∈ Y be defined
(x, w) ∈ U × V, let t ∈ Jx,w → φ(t, x, w) denote the maximal solution to the by ys (t) := y(st). Use the following outline to prove the ODE
ODE
ẏ(t) = Z(y(t)) with y(0) = x (12.53)
ẏ(t) = Z(y(t), w) with y(0) = x (12.49)
and has a unique solution for small t and this solution is C 1 in x.
D := {(t, x, w) ∈ R × U × V : t ∈ Jx,w } 1. If y solves Eq. (12.53) then ys solves
as in Exercise 11.8. ẏs (t) = sZ(ys (t)) with ys (0) = x
1
1. Prove that φ is C and that Dw φ(t, x, w) solves the differential equation: or equivalently Z t
d ys (t) = x + s Z(ys (τ ))dτ. (12.54)
Dw φ(t, x, w) = (Dx Z)(φ(t, x, w), w)Dw φ(t, x, w) + (Dw Z)(φ(t, x, w), w)
dt 0
Notice that when s = 0, the unique solution to this equation is y0 (t) = x.
with Dw φ(0, x, w) = 0 ∈ L(W, X). Hint: See the hint for Exercise 11.8 with
2. Let F : J × Y → J × Y be defined by
the reference to Theorem 11.15 being replace by Theorem 12.28.
Z t
2. Also show with the aid of Duhamel’s principle (Exercise 10.22) and Theorem
12.28 that F (s, y) := (s, y(t) − s Z(y(τ ))dτ ).
0
Z t
Dw φ(t, x, w) = Dx φ(t, x, w) Dx φ(τ, x, w)−1 (Dw Z)(φ(τ, x, w), w)dτ Show the differential of F is given by
0 Z t Z ·
F 0 (s, y)(a, v) = a, t → v(t) − s Z 0 (y(τ ))v(τ )dτ − a Z(y(τ ))dτ .
Exercise 12.5. (Differential of eA ) Let f : L(X) → GL(X) be the exponen- 0 0
tial function f (A) = eA . Prove that f is differentiable and that
3. Verify F 0 (0, y) : R × Y → R × Y is invertible for all y ∈ Y and notice that
Z 1 F (0, y) = (0, y).
Df (A)B = e(1−t)A BetA dt. (12.50) 4. For x ∈ X, let Cx ∈ Y be the constant path at x, i.e. Cx (t) = x for all t ∈ J.
0
Use the inverse function Theorem 12.25 to conclude there exists ε > 0 and
Hint: Let B ∈ L(X) and define w(t, s) = et(A+sB) for all t, s ∈ R. Notice that a C 1 map φ : (−ε, ε) × B(x0 , ε) → Y such that
dw(t, s)/dt = (A + sB)w(t, s) with w(0, s) = I ∈ L(X). (12.51) F (s, φ(s, x)) = (s, Cx ) for all (s, x) ∈ (−ε, ε) × B(x0 , ε).

5. Show, for s ≤ ε that ys (t) := φ(s, x)(t) satisfies Eq. (12.54). Now define
y(t, x) = φ(ε/2, x)(2t/ε) and show y(t, x) solve Eq. (12.53) for |t| < ε/2 and
x ∈ B(x0 , ε).
Exercise 12.8. Show P defined in Theorem 12.31 is continuously differentiable

and P 0 (y)h = ḣ + p0 (y)h.
Exercise 12.9. Embedded sub-manifold problems.
Exercise 12.10. Lagrange Multiplier problems.
12.9.1 Alternate construction of g. To be made into an exercise.
Suppose U ⊂o X and f : U → Y is a C 2 – function. Then we are looking for a

function g(y) such that f (g(y)) = y. Fix an x0 ∈ U and y0 = f (x0 ) ∈ Y. Suppose
such a g exists and let x(t) = g(y0 + th) for some h ∈ Y. Then differentiating
f (x(t)) = y0 + th implies
d
f (x(t)) = f 0 (x(t))ẋ(t) = h
dt
or equivalently that
−1
ẋ(t) = [f 0 (x(t))] h = Z(h, x(t)) with x(0) = x0 (12.55)
−1
where Z(h, x) = [f 0 (x(t))] h. Conversely if x solves Eq. (12.55) we have
d
dt f (x(t)) = h and hence that
f (x(1)) = y0 + h.
Thus if we define
g(y0 + h) := eZ(h,·) (x0 ),
then f (g(y0 + h)) = y0 + h for all h sufficiently small. This shows f is an open
mapping.
Part IV
Topological Spaces
13
Topological Space Basics
Using the metric space results above as motivation we will axiomatize the 3. Let X = {1, 2, 3}, then τ = {∅, X, {2, 3}} is a topology on X which does
notion of being an open set to more general settings. not come from a metric.
4. Again let X = {1, 2, 3}. Then τ = {{1}, {2, 3}, ∅, X}. is a topology, and the
Definition 13.1. A collection of subsets τ of X is a topology if sets X, {1}, {2, 3}, ∅ are open and closed. The sets {1, 2} and {1, 3} are
neither open nor closed.
1. ∅, X ∈ τ S
2. τ is closed under arbitrary unions, i.e. if Vα ∈ τ, for α ∈ I then Vα ∈ τ .
α∈I
3. τ is closed under finite intersections, i.e. if V1 , . . . , Vn ∈ τ then V1 ∩· · ·∩Vn ∈
τ.
A pair (X, τ ) where τ is a topology on X will be called a topological space.
Notation 13.2 Let (X, τ ) be a topological space.

1. The elements, V ∈ τ, are called open sets. We will often write V ⊂o X to
indicate V is an open subset of X.
2. A subset F ⊂ X is closed if F c is open and we will write F @ X if F is a
closed subset of X.
3. An open neighborhood of a point x ∈ X is an open set V ⊂ X such that
x ∈ V. Let τx = {V ∈ τ : x ∈ V } denote the collection of open neighborhoods
of x.
Fig. 13.1. A topology.
4. A subset W ⊂ X is a neighborhood of x if there exists V ∈ τx such that
V ⊂ W.
5. A collection η ⊂ τx is called a neighborhood base at x ∈ X if for all
V ∈ τx there exists W ∈ η such that W ⊂ V .
Definition 13.4. Let (X, τX ) and (Y, τY ) be topological spaces. A function f :
The notation τx should not be confused with X → Y is continuous if
f −1 (τY ) := f −1 (V ) : V ∈ τY ⊂ τX .

τ{x} := i−1
{x} (τ ) = {{x} ∩ V : V ∈ τ } = {∅, {x}} .
Example 13.3. 1. Let (X, d) be a metric space, we write τd for the collection of We will also say that f is τX /τY –continuous or (τX , τY ) – continuous. Let
d – open sets in X. We have already seen that τd is a topology, see Exercise C(X, Y ) denote the set of continuous functions from X to Y.
6.2. The collection of sets η = {Bx (ε) : ε ∈ D} where D is any dense subset Exercise 13.1. Show f : X → Y is continuous iff f −1 (C) is closed in X for all
of (0, 1] is a neighborhood base at x. closed subsets C of Y.
2. Let X be any set, then τ = 2X is the discrete topology on X. In this topology
all subsets of X are both open and closed. At the opposite extreme we have Definition 13.5. A map f : X → Y between topological spaces is called a
the trivial topology, τ = {∅, X} . In this topology only the empty set and homeomorphism provided that f is bijective, f is continuous and f −1 : Y →
X are open (closed). X is continuous. If there exists f : X → Y which is a homeomorphism, we say
112 13 Topological Space Basics
that X and Y are homeomorphic. (As topological spaces X and Y are essentially
the same.)
13.1 Constructing Topologies and Checking Continuity

Proposition 13.6. Let E be any collection of subsets of X. Then there exists
a unique smallest topology τ (E) which contains E.
Proof. Since 2X is a topology and E ⊂ 2X , E is always a subset of a topology.
It is now easily seen that
\
τ (E) := {τ : τ is a topology and E ⊂ τ }
Fig. 13.2. Fitting balls in the intersection.
is a topology which is clearly the smallest possible topology containing E.

The following proposition gives an explicit descriptions of τ (E).
Exercise 13.2. Suppose that S is a sub-base for a topology τ on a set X.
Proposition 13.7. Let X be a set and E ⊂ 2X . For simplicity of notation,
1. Show V := Sf (Sf is the collection of finite intersections of elements from
assume that X, ∅ ∈ E. (If this is not the case simply replace E by E ∪ {X, ∅} .)
S) is a base for τ.
Then
2. Show S is itself a base for τ iff
τ (E) := {arbitrary unions of finite intersections of elements from E}. (13.1)
V1 ∩ V2 = ∪{S ∈ S : S ⊂ V1 ∩ V2 }.
Proof. Let τ be given as in the right side of Eq. (13.1). From the definition of for every pair of sets V1 , V2 ∈ S.
a topology any topology containing E must contain τ and hence E ⊂ τ ⊂ τ (E).
The proof will be completed by showing τ is a topology. The validation of τ Remark 13.9. Let (X, d) be a metric space, then E = {Bx (δ) : x ∈ X and δ > 0}
being a topology is routine except for showing that τ is closed under taking is a base for τd – the topology associated to the metric d. This is the content
finite intersections. Let V, W ∈ τ which by definition may be expressed as of Exercise 6.3.
Let us check directly that E is a base for a topology. Suppose that x, y ∈ X
V = ∪α∈A Vα and W = ∪β∈B Wβ , and ε, δ > 0. If z ∈ B(x, δ) ∩ B(y, ε), then
where Vα and Wβ are sets which are finite intersection of elements from E. Then B(z, α) ⊂ B(x, δ) ∩ B(y, ε) (13.2)
[
V ∩ W = (∪α∈A Vα ) ∩ (∪β∈B Wβ ) = Vα ∩ Wβ . where α = min{δ − d(x, z), ε − d(y, z)}, see Figure 13.2. This is a for-
(α,β)∈A×B mal consequence of the triangle inequality. For example let us show that
B(z, α) ⊂ B(x, δ). By the definition of α, we have that α ≤ δ − d(x, z) or
Since for each (α, β) ∈ A × B, Vα ∩ Wβ is still a finite intersection of elements that d(x, z) ≤ δ − α. Hence if w ∈ B(z, α), then
from E, V ∩ W ∈ τ showing τ is closed under taking finite intersections.
d(x, w) ≤ d(x, z) + d(z, w) ≤ δ − α + d(z, w) < δ − α + α = δ
Definition 13.8. Let (X, τ ) be a topological space. We say that S ⊂ τ is a sub-
base for the topology τ iff τ = τ (S) and X = ∪S := ∪V ∈S V. We say V ⊂ τ is a which shows that w ∈ B(x, δ). Similarly we show that w ∈ B(y, ε) as well.
base for the topology τ iff V is a sub-base with the property that every element Owing to Exercise 13.2, this shows E is a base for a topology. We do not
V ∈ τ may be written as need to use Exercise 13.2 here since in fact Equation (13.2) may be generalized
to finite intersection of balls. Namely if xi ∈ X, δi > 0 and z ∈ ∩ni=1 B(xi , δi ),
V = ∪{B ∈ V : B ⊂ V }. then

13.1 Constructing Topologies and Checking Continuity 113
B(z, α) ⊂ ∩ni=1 B(xi , δi ) (13.3) Lemma 13.14. Suppose that f : X → Y is a function, E ⊂ 2Y and A ⊂ Y,
then
where now α := min {δi − d(xi , z) : i = 1, 2, . . . , n} . By Eq. (13.3) it follows
that any finite intersection of open balls may be written as a union of open τ f −1 (E) = f −1 (τ (E)) and

(13.4)
balls.
τ (EA ) = (τ (E))A . (13.5)
Exercise 13.3. Suppose f : X → Y is a function and τX and τY are topologies
on X and Y respectively. Show Moreover, if τY = τ (E) and τX is a topology on X, then f is (τX , τY ) – con-
tinuous iff f −1 (E) ⊂ τX .
f −1 τY := f −1 (V ) ⊂ X : V ∈ τY and f∗ τX := V ⊂ Y : f −1 (V ) ∈ τX

Proof. We will give two proof of Eq. (13.4). The first proof is more con-
(as in Notation 2.7) are also topologies on X and Y respectively. structive than the second, but the second proof will work in the context of σ –
algebras to be developed later.
Remark 13.10. Let f : X → Y be a function. Given a topology τY ⊂ 2Y , the
First Proof. There is no harm (as the reader should verify) in replacing E
topology τX := f −1 (τY ) is the smallest topology on X such that f is (τX , τY )
by E ∪ {∅, Y } if necessary so that we may assume that ∅, Y ∈ E. By Proposition
- continuous. Similarly, if τX is a topology on X then τY = f∗ τX is the largest
13.7, the general element V of τ (E) is an arbitrary unions of finite intersections
topology on Y such that f is (τX , τY ) - continuous.
of elements from E. Since f −1 preserves all of the set operations, it follows
Definition 13.11. Let (X, τ ) be a topological space and A subset of X. The that f −1 τ (E) consists of sets which are arbitrary unions of finite intersections
of elements from f −1 E, which is precisely τ f −1 (E) by another application of

relative topology or induced topology on A is the collection of sets
Proposition 13.7.
τA = i−1
A (τ ) = {A ∩ V : V ∈ τ } , Second Proof. By Exercise 13.3, f −1 (τ (E)) is a topology and since E ⊂
τ (E) , f −1 (E) ⊂ f −1 (τ (E)). It now follows that τ (f −1 (E)) ⊂ f −1 (τ (E)). For
where iA : A → X be the inclusion map as in Definition 2.8. the reverse inclusion notice that
Lemma 13.12. The relative topology, τA , is a topology on A. Moreover a subset
f∗ τ f −1 (E) = B ⊂ Y : f −1 (B) ∈ τ f −1 (E)

B ⊂ A is τA – closed iff there is a τ – closed subset, C, of X such that B = C∩A.
is a topology which contains E and thus τ (E) ⊂ f∗ τ f −1 (E) . Hence

Proof. The first assertion is a consequence of Exercise 13.3. For the second, if B ∈ τ (E)
we know that f −1 (B) ∈ τ f −1 (E) , i.e. f −1 (τ (E)) ⊂ τ f −1 (E) and Eq. (13.4)

B ⊂ A is τA – closed iff A \ B = A ∩ V for some V ∈ τ which is equivalent to
B = A \ (A ∩ V ) = A ∩ V c for some V ∈ τ. has been proved. Applying Eq. (13.4) with X = A and f = iA being the
inclusion map implies
Exercise 13.4. Show if (X, d) is a metric space and τ = τd is the topology
coming from d, then (τd )A is the topology induced by making A into a metric (τ (E))A = i−1 −1
A (τ (E)) = τ (iA (E)) = τ (EA ).
space using the metric d|A×A .
Lastly if f −1 E ⊂ τX , then f −1 τ (E) = τ f −1 E ⊂ τX which shows f is (τX , τY )

Lemma 13.13. Suppose that (X, τX ), (Y, τY ) and (Z, τZ ) are topological – continuous.
spaces. If f : (X, τX ) → (Y, τY ) and g : (Y, τY ) → (Z, τZ ) are continuous
functions then g ◦ f : (X, τX ) → (Z, τZ ) is continuous as well. Corollary 13.15. If (X, τ ) is a topological space and f : X → R is a function
then the following are equivalent:
Proof. This is easy since by assumption g −1 (τZ ) ⊂ τY and f −1 (τY ) ⊂ τX
so that 1. f is (τ, τR ) - continuous,
−1
(g ◦ f ) (τZ ) = f −1 g −1 (τZ ) ⊂ f −1 (τY ) ⊂ τX .

2. f −1 ((a, b)) ∈ τ for all −∞ < a < b < ∞,
3. f −1 ((a, ∞)) ∈ τ and f −1 ((−∞, b)) ∈ τ for all a, b ∈ Q.
The following elementary lemma turns out to be extremely useful because (We are using τR to denote the standard topology on R induced by the metric
it may be used to greatly simplify the verification that a given function is d(x, y) = |x − y|.)
continuous.

Proof. Apply Lemma 13.14 with appropriate choices of E. Since each A ∈ U is open, τA ⊂ τ and by assumption, f |−1 A (V ) ∈ τA ⊂ τ. Hence
Eq. (13.6) shows f −1 (V ) is a union of τ – open sets and hence is also τ – open.
Definition 13.16. Let (X, τX ) and (Y, τY ) be topological spaces. A function
f : X → Y is continuous at a point x ∈ X if for every open neighborhood V
of f (x) there is an open neighborhood U of x such that U ⊂ f −1 (V ). See Figure Exercise 13.6 (A Baby Extension Theorem). Suppose V ∈ τ and f : V →
13.3. C is a continuous function. Further assume there is a closed subset C such that
{x ∈ V : f (x) 6= 0} ⊂ C ⊂ V, then F : X → C defined by

f (x) if x ∈ V
F (x) =
0 if x ∈ /V
is continuous.
Exercise 13.7 (Building Continuous Functions). Prove the following vari-

ant of item 2. of Proposition 13.19. Namely, suppose there exists a finite col-
lection F of closed subsets of X such that X = ∪A∈F A and f |A is continuous
for all A ∈ F, then f is continuous. Given an example showing that the as-
sumption that F is finite can not be eliminated. Hint: consider f −1 (C) where
C is a closed subset of Y.
Fig. 13.3. Checking that a function is continuous at x ∈ X.
13.2 Product Spaces I
Exercise 13.5. Show f : X → Y is continuous (Definition 13.16) iff f is con- Definition 13.20. Let X be a set and suppose there is a collection of topological
tinuous at all points x ∈ X. spaces {(Yα , τα ) : α ∈ A} and functions fα : X → Yα for all α ∈ A. Let τ (fα :
α ∈ A) denote the smallest topology on X such that each fα is continuous, i.e.
Definition 13.17. Given topological spaces (X, τ ) and (Y, τ 0 ) and a subset A ⊂
X. We say a function f : A → Y is continuous iff f is τA /τ 0 – continuous. τ (fα : α ∈ A) = τ (∪α fα−1 (τα )).
Proposition 13.21 (Topologies Generated by Functions). Assuming the

Definition 13.18. Let (X, τ ) be a topological S S A ⊂ X. A collection of
space and
subsets U ⊂ τ is an open cover of A if A ⊂ U := U ∈U U. notation in Definition 13.20 and additionally let (Z, τZ ) be a topological space
and g : Z → X be a function. Then g is (τZ , τ (fα : α ∈ A)) – continuous iff
Proposition 13.19 (Localizing Continuity). Let (X, τ ) and (Y, τ 0 ) be topo- fα ◦ g is (τZ , τα )–continuous for all α ∈ A.
logical spaces and f : X → Y be a function.
Proof. (⇒) If g is (τZ , τ (fα : α ∈ A)) – continuous, then the composition
1. If f is continuous and A ⊂ X then f |A : A → Y is continuous. fα ◦ g is (τZ , τα ) – continuous by Lemma 13.13. (⇐) Let
2. Suppose there exist an open cover, U ⊂ τ, of X such that f |A is continuous
τX = τ (fα : α ∈ A) = τ ∪α∈A fα−1 (τα ) .

for all A ∈ U, then f is continuous.
Proof. 1. If f : X → Y is a continuous, f −1 (V ) ∈ τ for all V ∈ τ 0 and If fα ◦ g is (τZ , τα ) – continuous for all α, then
therefore
f |−1 −1
(V ) ∈ τA for all V ∈ τ 0 . g −1 fα−1 (τα ) ⊂ τZ ∀ α ∈ A
A (V ) = A ∩ f
2. Let V ∈ τ 0 , then and therefore
f −1 (V ) = ∪A∈U f −1 (V ) ∩ A = ∪A∈U f |−1 g −1 ∪α∈A fα−1 (τα ) = ∪α∈A g −1 fα−1 (τα ) ⊂ τZ

A (V ). (13.6)

13.2 Product Spaces I 115
Hence Proof. 1. Since
g −1 (τX ) = g −1 τ ∪α∈A fα−1 (τα ) = τ (g −1 ∪α∈A fα−1 (τα ) ⊂ τZ ∪α πα−1 Eα ⊂ ∪α πα−1 τα = ∪α πα−1 (τ (Eα ))

= ∪α τ (πα−1 Eα ) ⊂ τ ∪α πα−1 Eα ,

which shows that g is (τZ , τX ) – continuous. Q
Let {(Xα , τα )}α∈A be a collection of topological spaces, X = XA = Xα
α∈A it follows that
τ ∪α πα−1 Eα ⊂ ⊗α τα ⊂ τ ∪α πα−1 Eα .

and πα : XA → Xα be the canonical projection map as in Notation 2.2.
2. Now let U = ∪α πα−1 τα f denote the collection of sets consisting of finite

Definition 13.22. The product topology τ = ⊗α∈A τα is the smallest topology
on XA such that each projection πα is continuous. Explicitly, τ is the topology intersections of elements from ∪α πα−1 τα . Notice that U may be described as
generated by the collection of sets, those sets in Eq. (13.9) where Vα ∈ τα for all α ∈ Λ. By Exercise 13.2, U is a
base for the product topology, ⊗α∈A τα . Hence for W ∈ ⊗α∈A τα and x ∈ W,
E = {πα−1 (Vα ) : α ∈ A, Vα ∈ τα } = ∪α∈A π −1 τα . (13.7) there exists a V ∈ U of the form in Eq. (13.9) such that x ∈ V ⊂ W. Since Bα
is a base for τα , there exists Uα ∈ Bα such that xα ∈ Uα ⊂ Vα for each α ∈ Λ.
Applying Proposition 13.21 in this setting implies the following proposition.
With this notation, the set UΛ × XA\Λ ∈ V and x ∈ UΛ × XA\Λ ⊂ V ⊂ W. This
Proposition 13.23. Suppose Y is a topological space and f : Y → XA is a shows that every open set in X may be written as a union of elements from V,
map. Then f is continuous iff πα ◦ f : Y → Xα is continuous for all α ∈ A. In i.e. V is a base for the product topology.
particular if A = {1, 2, . . . , n} so that XA = X1 × X2 × · · · × Xn and f (y) =
Notation 13.26 Let Ei ⊂ 2Xi be a collection of subsets of a set Xi for each
(f1 (y), f2 (y), . . . , fn (y)) ∈ X1 × X2 × · · · × Xn , then f : Y → XA is continuous
i = 1, 2, . . . , n. We will write, by abuse of notation, E1 × E2 × · · · × En for the
iff fi : Y → Xi is continuous for all i.
collection of subsets of X1 ×· · ·×Xn of the form A1 ×A2 ×· · ·×An with Ai ∈ Ei
Proposition 13.24. Suppose that (X, τ ) is a topological space and {fn } ⊂ X A for all i. That is we are identifying (A1 , A2 , . . . , An ) with A1 × A2 × · · · × An .
(see Notation 2.2) is a sequence. Then fn → f in the product topology of X A
Corollary 13.27. Suppose A = {1, 2, . . . , n} so X = X1 × X2 × · · · × Xn .
iff fn (α) → f (α) for all α ∈ A.
1. If Ei ⊂ 2Xi , τi = τ (Ei ) and Xi ∈ Ei for each i, then
Proof. Since πα is continuous, if fn → f then fn (α) = πα (fn ) → πα (f ) =
f (α) for all α ∈ A. Conversely, fn (α) → f (α) for all α ∈ A iff πα (fn ) → πα (f ) τ1 ⊗ τ2 ⊗ · · · ⊗ τn = τ (E1 × E2 × · · · × En ) (13.10)
for all α ∈ A. Therefore if V = πα−1 (Vα ) ∈ E (with E as in Eq. (13.7)) and
f ∈ V, then πα (f ) ∈ Vα and πα (fn ) ∈ Vα for a.a. n and hence fn ∈ V for a.a. and in particular
n. This shows that fn → f as n → ∞.
τ1 ⊗ τ2 ⊗ · · · ⊗ τn = τ (τ1 × · · · × τn ). (13.11)
Proposition 13.25. Suppose that (Xα , τα )α∈A is aQ collection of topological
spaces and ⊗α∈A τα is the product topology on X := α∈A Xα . 2. Furthermore if Bi ⊂ τi is a base for the topology τi for each i, then B1 ×
· · · × Bn is a base for the product topology, τ1 ⊗ τ2 ⊗ · · · ⊗ τn .
1. If Eα ⊂ τα generates τα for each α ∈ A, then
Proof. (The proof is a minor variation on the proof of Proposition 13.25.) 1.
⊗α∈A τα = τ ∪α∈A πα−1 (Eα )

(13.8)
Let ∪i∈A πi−1 (Ei ) f denotes the collection of sets which are finite intersections

2. If Bα ⊂ τα is a base for τα for each α, then the collection of sets, V, of the from ∪i∈A πi−1 (Ei ), then, using Xi ∈ Ei for all i,
form
∪i∈A πi−1 (Ei ) ⊂ E1 × E2 × · · · × En ⊂ ∪i∈A πi−1 (Ei ) f .

Y Y
V = ∩α∈Λ πα−1 Vα = Vα × Xα =: VΛ × XA\Λ , (13.9)
α∈Λ α∈Λ
/ Therefore

τ = τ ∪i∈A πi−1 (Ei ) ⊂ τ (E1 × E2 × · · · × En ) ⊂ τ ∪i∈A πi−1 (Ei ) f = τ.

where Λ ⊂⊂ A and Vα ∈ Bα for all α ∈ Λ is base for ⊗α∈A τα .

2. Observe that τ1 × · · · × τn is closed under finite intersections and generates 3. The accumulation points of A is the set
τ1 ⊗ τ2 ⊗ · · · ⊗ τn , therefore τ1 × · · · × τn is a base for the product topology. The
proof that B1 × · · · × Bn is also a base for τ1 ⊗ τ2 ⊗ · · · ⊗ τn follows the same acc(A) = {x ∈ X : V ∩ A \ {x} =
6 ∅ for all V ∈ τx }.
method used to prove item 2. in Proposition 13.25.
4. The boundary of A is the set bd(A) := Ā \ Ao .
Lemma 13.28. Let (Xi , di ) for i = 1, . . . , n be metric spaces, X := X1 × · · · ×
Xn and for x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ) in X let Remark 13.30. The relationships between the interior and the closure of a set
are:
n
X \ \
d(x, y) = di (xi , yi ). (13.12) (Ao )c = {V c : V ∈ τ and V ⊂ A} = {C : C is closed C ⊃ Ac } = Ac
i=1
and similarly, (Ā)c = (Ac )o . Hence the boundary of A may be written as
Then the topology, τd , associated to the metric d is the product topology on X,
i.e. bd(A) := Ā \ Ao = Ā ∩ (Ao )c = Ā ∩ Ac , (13.13)
τd = τd1 ⊗ τd2 ⊗ · · · ⊗ τdn .
which is to say bd(A) consists of the points in both the closure of A and Ac .
Proof. Let ρ(x, y) = max{di (xi , yi ) : i = 1, 2, . . . , n}. Then ρ is equivalent
to d and hence τρ = τd . Moreover if ε > 0 and x = (x1 , x2 , . . . , xn ) ∈ X, then Proposition 13.31. Let A ⊂ X and x ∈ X.
Bxρ (ε) = Bxd11 (ε) × · · · × Bxdnn (ε). 1. If V ⊂o X and A ∩ V = ∅ then Ā ∩ V = ∅.
2. x ∈ Ā iff V ∩ A 6= ∅ for all V ∈ τx .
By Remark 13.9, 3. x ∈ bd(A) iff V ∩ A 6= ∅ and V ∩ Ac 6= ∅ for all V ∈ τx .
E := {Bxρ (ε) : x ∈ X and ε > 0} 4. Ā = A ∪ acc(A).
is a base for τρ and by Proposition 13.25 E is also a base for τd1 ⊗ τd2 ⊗ · · · ⊗ τdn .
Proof. 1. Since A ∩ V = ∅, A ⊂ V c and since V c is closed, Ā ⊂ V c . That
Therefore, c
is to say Ā ∩ V = ∅. 2. By Remark 13.301 , Ā = ((Ac )o ) so x ∈ Ā iff x ∈
/ (Ac )o
τd1 ⊗ τd2 ⊗ · · · ⊗ τdn = τ (E) = τρ = τd . c
which happens iff V * A for all V ∈ τx , i.e. iff V ∩ A 6= ∅ for all V ∈ τx . 3.
This assertion easily follows from the Item 2. and Eq. (13.13). 4. Item 4. is an
easy consequence of the definition of acc(A) and item 2.
13.3 Closure operations Lemma 13.32. Let A ⊂ Y ⊂ X, ĀY denote the closure of A in Y with its
relative topology and Ā = ĀX be the closure of A in X, then ĀY = ĀX ∩ Y.
Definition 13.29. Let (X, τ ) be a topological space and A be a subset of X. Proof. Using Lemma 13.12,
1. The closure of A is the smallest closed set Ā containing A, i.e.
ĀY = ∩ {B @ Y : A ⊂ B} = ∩ {C ∩ Y : A ⊂ C @ X}
Ā := ∩ {F : A ⊂ F @ X} . = Y ∩ (∩ {C : A ⊂ C @ X}) = Y ∩ ĀX .
(Because of Exercise 6.4 this is consistent with Definition 6.10 for the clo- Alternative proof. Let x ∈ Y then x ∈ ĀY iff V ∩ A 6= ∅ for all V ∈ τY
sure of a set in a metric space.) such that x ∈ V. This happens iff for all U ∈ τx , U ∩ Y ∩ A = U ∩ A 6= ∅ which
2. The interior of A is the largest open set Ao contained in A, i.e. happens iff x ∈ ĀX . That is to say ĀY = ĀX ∩ Y.
The support of a function may now be defined as in Definition 10.26 above.
Ao = ∪ {V ∈ τ : V ⊂ A} .
1
c showing x ∈
Here is another direct proof of item 2. which goes by / Ā iff there exists
(With this notation the definition of a neighborhood of x ∈ X may be stated V ∈ τx such that V ∩ A = ∅. If x ∈ / Ā then V = Ā ∈ τx and V ∩ A ⊂ V ∩ Ā = ∅.
as: A ⊂ X is a neighborhood of a point x ∈ X if x ∈ Ao .) Conversely if there exists V ∈ τx such that A ∩ V = ∅ then by Item 1. Ā ∩ V = ∅.

13.4 Countability Axioms 117
Definition 13.33 (Support). Let f : X → Y be a function from a topological Example 13.36. Let X = {1, 2, 3} and τ = {X, ∅, {1, 2}, {2, 3}, {2}} and xn = 2
space (X, τX ) to a vector space Y. Then we define the support of f by for all n. Then xn → x for every x ∈ X. So limits need not be unique!
supp(f ) := {x ∈ X : f (x) 6= 0}, Definition 13.37 (First Countable). A topological space, (X, τ ), is first
countable iff every point x ∈ X has a countable neighborhood base as defined
a closed subset of X. in Notation 13.2
The next result is included for completeness but will not be used in the Example
13.38. spaces, (X, d) , are first countable. Indeed, if x ∈ X
All metric
sequel so may be omitted. then B x, n1 : n ∈ N is a countable neighborhood base at x ∈ X.
Lemma 13.34. Suppose that f : X → Y is a map between topological spaces. Exercise 13.8. Suppose X is an uncountable set and let V ∈ τ iff V c is finite or
Then the following are equivalent: countable of V = ∅. Show τ is a topology on X which is closed under countable
1. f is continuous. intersections and that (X, τ ) is not first countable.
2. f (Ā) ⊂ f (A) for all A ⊂ X
Exercise 13.9. Let {0, 1} be equipped with the discrete topology and X =
3. f −1 (B) ⊂ f −1 (B̄) for all B ⊂ Y. R
{0, 1} be equipped with the product topology, τ. Show (X, τ ) is not first
countable.

Proof. If f is continuous, then f −1 f (A) is closed and since A ⊂

f −1 (f (A)) ⊂ f −1 f (A) it follows that Ā ⊂ f −1 f (A) . From this equa- The spaces described in Exercises 13.8 and 13.9 are examples of topological
spaces which are not metrizable, i.e. the topology is not induced by any metric
tion we learn that f (Ā) ⊂ f (A) so that (1) implies (2) Now assume (2), then on X. Like for metric spaces, when τ is first countable, we may formulate many
for B ⊂ Y (taking A = f −1 (B̄)) we have topological notions in terms of sequences.
f (f −1 (B)) ⊂ f (f −1 (B̄)) ⊂ f (f −1 (B̄)) ⊂ B̄ Proposition 13.39. If f : X → Y is continuous at x ∈ X and limn→∞ xn =
x ∈ X, then limn→∞ f (xn ) = f (x) ∈ Y. Moreover, if there exists a countable
and therefore neighborhood base η of x ∈ X, then f is continuous at x iff lim f (xn ) = f (x)
f −1 (B) ⊂ f −1 (B̄). (13.14) ∞
n→∞
for all sequences {xn }n=1 ⊂ X such that xn → x as n → ∞.
This shows that (2) implies (3) Finally if Eq. (13.14) holds for all B, then when
B is closed this shows that Proof. If f : X → Y is continuous and W ∈ τY is a neighborhood of f (x) ∈
Y, then there exists a neighborhood V of x ∈ X such that f (V ) ⊂ W. Since
f −1 (B) ⊂ f −1 (B̄) = f −1 (B) ⊂ f −1 (B) xn → x, xn ∈ V a.a. and therefore f (xn ) ∈ f (V ) ⊂ W a.a., i.e. f (xn ) → f (x)
as n → ∞. Conversely suppose that η := {Wn }∞ n=1 is a countable neighborhood
which shows that ∞
base at x and lim f (xn ) = f (x) for all sequences {xn }n=1 ⊂ X such that
f −1 (B) = f −1 (B). n→∞
xn → x. By replacing Wn by W1 ∩ · · · ∩ Wn if necessary, we may assume that
Therefore f −1 (B) is closed whenever B is closed which implies that f is con- ∞
{Wn }n=1 is a decreasing sequence of sets. If f were not continuous at x then
tinuous. o
/ f −1 (V ) . Therefore, Wn is not a subset
there exists V ∈ τf (x) such that x ∈
of f (V ) for all n. Hence for each n, we may choose xn ∈ Wn \ f −1 (V ). This
−1
sequence then has the property that xn → x as n → ∞ while f (xn ) ∈ / V for all
13.4 Countability Axioms n and hence limn→∞ f (xn ) 6= f (x).
∞ ∞
Definition 13.35. Let (X, τ ) be a topological space. A sequence {xn }n=1 ⊂ Lemma 13.40. Suppose there exists {xn }n=1 ⊂ A such that xn → x, then
X converges to a point x ∈ X if for all V ∈ τx , xn ∈ V almost always x ∈ Ā. Conversely if (X, τ ) is a first countable space (like a metric space) then
∞
(abbreviated a.a.), i.e. # ({n : xn ∈
/ V }) < ∞. We will write xn → x as n → ∞ if x ∈ Ā there exists {xn }n=1 ⊂ A such that xn → x.
or limn→∞ xn = x when xn converges to x.

∞
Proof. Suppose {xn }n=1 ⊂ A and xn → x ∈ X. Since Āc is an open Exercise 13.12. Every separable metric space, (X, d) is second countable.
set, if x ∈ Āc then xn ∈ Āc ⊂ Ac a.a. contradicting the assumption that
∞
{xn }n=1 ⊂ A. Hence x ∈ Ā. For the converse we now assume that (X, τ ) is first Exercise 13.13. Suppose E ⊂ 2X is a countable collection of subsets of X,
∞ then τ = τ (E) is a second countable topology on X.
countable and that {Vn }n=1 is a countable neighborhood base at x such that
V1 ⊃ V2 ⊃ V3 ⊃ . . . . By Proposition 13.31, x ∈ Ā iff V ∩ A 6= ∅ for all V ∈ τx .
Hence x ∈ Ā implies there exists xn ∈ Vn ∩ A for all n. It is now easily seen
that xn → x as n → ∞. 13.5 Connectedness
Definition 13.41. A topological space, (X, τ ), is second countable if there Definition 13.44. (X, τ ) is disconnected if there exist non-empty open sets
exists a countable base V for τ, i.e. V ⊂ τ is a countable set such that for every U and V of X such that U ∩ V = ∅ and X = U ∪ V . We say {U, V } is a
W ∈ τ, disconnection of X. The topological space (X, τ ) is called connected if it
W = ∪{V : V ∈ V such that V ⊂ W }. is not disconnected, i.e. if there is no disconnection of X. If A ⊂ X we say
A is connected iff (A, τA ) is connected where τA is the relative topology on
Definition 13.42. A subset D of a topological space X is dense if D̄ = X. A
A. Explicitly, A is disconnected in (X, τ ) iff there exists U, V ∈ τ such that
topological space is said to be separable if it contains a countable dense subset,
U ∩ A 6= ∅, U ∩ A 6= ∅, A ∩ U ∩ V = ∅ and A ⊂ U ∪ V.
D.
The reader should check that the following statement is an equivalent defi-
Example 13.43. The following are examples of countable dense sets.
nition of connectivity. A topological space (X, τ ) is connected iff the only sets
1. The rational numbers, Q, are dense in R equipped with the usual topology. A ⊂ X which are both open and closed are the sets X and ∅. This version of
2. More generally, Qd is a countable dense subset of Rd for any d ∈ N. the definition is often used in practice.
3. Even more generally, for any function µ : N → (0, ∞), `p (µ) is separable for
all 1 ≤ p < ∞. For example, let Γ ⊂ F be a countable dense set, then Remark 13.45. Let A ⊂ Y ⊂ X. Then A is connected in X iff A is connected in
Y.
D := {x ∈ `p (µ) : xi ∈ Γ for all i and #{j : xj 6= 0} < ∞}.
Proof. Since
The set Γ can be taken to be Q if F = R or Q + iQ if F = C.
4. If (X, d) is a metric space which is separable then every subset Y ⊂ X is τA := {V ∩ A : V ⊂ X} = {V ∩ A ∩ Y : V ⊂ X} = {U ∩ A : U ⊂o Y },
also separable in the induced topology.
the relative topology on A inherited from X is the same as the relative topology
To prove 4. above, let A = {xn }∞
n=1 ⊂ X be a countable dense subset of X. on A inherited from Y . Since connectivity is a statement about the relative
Let dY (x) = inf{d(x, y) : y ∈ Y } be the distance topologies on A, A is connected in X iff A is connected in Y.
from x to Y and recall that
dY : X → [0, ∞) is continuous. Let εn = max dY (xn ), n1 ≥ 0 and for each n The following elementary but important lemma is left as an exercise to the
let yn ∈ Bxn (2εn ). Then if y ∈ Y and ε > 0 we may choose n ∈ N such that reader.
d(y, xn ) ≤ εn < ε/3. Then d(yn , xn ) ≤ 2εn < 2ε/3 and therefore
Lemma 13.46. Suppose that f : X → Y is a continuous map between topolog-
d(y, yn ) ≤ d(y, xn ) + d(xn , yn ) < ε. ical spaces. Then f (X) ⊂ Y is connected if X is connected.
This shows that B := {yn }∞ Here is a typical way these connectedness ideas are used.
n=1 is a countable dense subset of Y.
Exercise 13.10. Show `∞ (N) is not separable. Example 13.47. Suppose that f : X → Y is a continuous map between two
topological spaces, the space X is connected and the space Y is “T1 ,” i.e. {y}
Exercise 13.11. Show every second countable topological space (X, τ ) is sep- is a closed set for all y ∈ Y as in Definition 15.35 below. Further assume f is
arable. Show the converse is not true by showing X := R with τ = {∅} ∪ locally constant, i.e. for all x ∈ X there exists an open neighborhood V of x
{V ⊂ R : 0 ∈ V } is a separable, first countable but not a second countable topo- in X such that f |V is constant. Then f is constant, i.e. f (X) = {y0 } for some
logical space. y0 ∈ Y. To prove this, let y0 ∈ f (X) and let W := f −1 ({y0 }). Since {y0 } ⊂ Y

13.5 Connectedness 119
S
is a closed set and since f is continuous W ⊂ X is also closed. Since f is locally 3. Let Y := α∈A Eα . By Remark 13.45, we know that Eα is connected in
constant, W is open as well and since X is connected it follows that W = X, Y for each α ∈ A. If {U, V } were a disconnection of Y, by item (1), either
i.e. f (X) = {y0 } . Eα ⊂ U or Eα ⊂ V for all α. Let Λ = {α ∈ A : Eα ⊂ U } then U = ∪α∈Λ Eα
As a concrete application of this result, suppose that X is a connected open and V = ∪α∈A\Λ Eα . (Notice that neither Λ or A \ Λ can be empty since U
subset of Rd and f : X → R is a C 1 – function such that ∇f ≡ 0. If x ∈ X and and V are not empty.) Since
ε > 0 such that B (x, ε) ⊂ X, we have, for any |v| < ε and t ∈ [−1, 1] , that [ \
d ∅=U ∩V = α∈Λ,β∈Λc (Eα ∩ Eβ ) ⊃ Eα 6= ∅.
f (x + tv) = ∇f (x + tv) · v = 0. α∈A
dt
Therefore f (x + v) = f (x) for all |v| < ε and this shows f is locally constant. we have reached a contradiction and hence no such disconnection exists.
Hence, by what we have just proved, f is constant on X. 4. (A good example to keep in mind here is X = R, A = (0, 1) and B = [1, 2).)
For sake of contradiction suppose that {U, V } were a disconnection of Y =
Theorem 13.48 (Properties of Connected Sets). Let (X, τ ) be a topolog- A ∪ B. By item (1) either A ⊂ U or A ⊂ V, say A ⊂ U in which case B ⊂ V.
ical space. Since Y = A ∪ B we must have A = U and B = V and so we may conclude:
1. If B ⊂ X is a connected set and X is the disjoint union of two open sets U A and B are disjoint subsets of Y which are both open and closed. This
and V, then either B ⊂ U or B ⊂ V. implies
A = ĀY = Ā ∩ Y = Ā ∩ (A ∪ B) = A ∪ Ā ∩ B

2. If A ⊂ X is connected,
a) then Ā is connected. and therefore
b) More generally, if A is connected and B ⊂ acc(A), then A ∪ B is con-

nected as well. (Recall that acc(A) – the set of accumulation points of ∅ = A ∩ B = A ∪ Ā ∩ B ∩ B = Ā ∩ B 6= ∅
A was defined in Definition 13.29 above.)
which gives us the desired contradiction.
T
3. If {EαS}α∈A is a collection of connected sets such that α∈A Eα 6= ∅, then
Y := α∈A Eα is connected as well. 5. Let C denote the collection of connected subsets C ⊂ X such that x ∈ C.
4. Suppose A, B ⊂ X are non-empty connected subsets of X such that Ā∩B 6= Then by item 3., the set Cx := ∪C is also a connected subset of X which
∅, then A ∪ B is connected in X. contains x and clearly this is the unique maximal connected set containing
5. Every point x ∈ X is contained in a unique maximal connected subset Cx of x. Since C̄x is also connected by item (2) and Cx is maximal, Cx = C̄x , i.e.
X and this subset is closed. The set Cx is called the connected component Cx is closed.
of x.
Proof.
Theorem 13.49 (The Connected Subsets of R). The connected subsets of
1. Since B is the disjoint union of the relatively open sets B ∩ U and B ∩ V, we R are intervals.
must have B ∩ U = B or B ∩ V = B for otherwise {B ∩ U, B ∩ V } would
be a disconnection of B. Proof. Suppose that A ⊂ R is a connected subset and that a, b ∈ A with
2. a. Let Y = Ā be equipped with the relative topology from X. Suppose that a < b. If there exists c ∈ (a, b) such that c ∈ / A, then U := (−∞, c) ∩ A
U, V ⊂o Y form a disconnection of Y = Ā. Then by 1. either A ⊂ U or and V := (c, ∞) ∩ A would form a disconnection of A. Hence (a, b) ⊂ A. Let
A ⊂ V. Say that A ⊂ U. Since U is both open an closed in Y, it follows that α := inf(A) and β := sup(A) and choose αn , βn ∈ A such that αn < βn and
Y = Ā ⊂ U. Therefore V = ∅ and we have a contradiction to the assumption αn ↓ α and βn ↑ β as n → ∞. By what we have just shown, (αn , βn ) ⊂ A for all
that {U, V } is a disconnection of Y = Ā. Hence we must conclude that n and hence (α, β) = ∪∞ n=1 (αn , βn ) ⊂ A. From this it follows that A = (α, β),
Y = Ā is connected as well. [α, β), (α, β] or [α, β], i.e. A is an interval.
b. Now let Y = A ∪ B with B ⊂ acc(A), then Conversely suppose that A is an interval, and for sake of contradiction,
suppose that {U, V } is a disconnection of A with a ∈ U, b ∈ V. After relabelling
ĀY = Ā ∩ Y = (A ∪ acc(A)) ∩ Y = A ∪ B.
U and V if necessary we may assume that a < b. Since A is an interval [a, b] ⊂ A.
Because A is connected in Y, by (2a) Y = A ∪ B = ĀY is also connected. Let p = sup ([a, b] ∩ U ) , then because U and V are open, a < p < b. Now p

can not be in U for otherwise sup ([a, b] ∩ U ) > p and p can not be in V for
otherwise p < sup ([a, b] ∩ U ) . From this it follows that p ∈
/ U ∪ V and hence
A 6= U ∪ V contradicting the assumption that {U, V } is a disconnection.
Theorem 13.50 (Intermediate Value Theorem). Suppose that (X, τ ) is
a connected topological space and f : X → R is a continuous map. Then f
satisfies the intermediate value property. Namely, for every pair x, y ∈ X such
that f (x) < f (y) and c ∈ (f (x) , f (y)), there exits z ∈ X such that f (z) = c.
Proof. By Lemma 13.46, f (X) is connected subset of R. So by Theorem
13.49, f (X) is a subinterval of R and this completes the proof.
Definition 13.51. A topological space X is path connected if to every pair of
points {x0 , x1 } ⊂ X there exists a continuous path, σ ∈ C([0, 1], X), such that
σ(0) = x0 and σ(1) = x1 . The space X is said to be locally path connected
if for each x ∈ X, there is an open neighborhood V ⊂ X of x which is path
connected.
Proposition 13.52. Let X be a topological space.
1. If X is path connected then X is connected. Fig. 13.4. This picture illustrates why the connected component of p in X × Y must
contain all points of X × Y.
2. If X is connected and locally path connected, then X is path connected.
3. If X is any connected open subset of Rn , then X is path connected.
Proof. The reader is asked to prove this proposition in Exercises 13.20 – if Cp is a proper subset of XA , then XA \ Cp is a non-empty open set. By the
13.22 below. definition of the product topology, this would imply that XA \ Cp contains an
Proposition 13.53 (Stability of Connectedness Under Products). Let open set of the form
(Xα , τα ) be connected topological spaces. Then the product space XA =
Q V := ∩α∈Λ πα−1 (Vα ) = VΛ × XA\Λ
α∈A Xα equipped with the product topology is connected.
Proof. Let us begin with the case of two factors, namely assume that X and where Λ ⊂⊂ A and Vα ∈ τα for all α ∈ Λ. We will now show that no such V
Y are connected topological spaces, then we will show that X × Y is connected can exist and hence XA = Cp , i.e. XA is connected.
as well. Given x ∈ X, let fx : Y → X × Y be the map fx (y) = (x, y) and notice Define φ : XΛ → XA by φ(y) = x where
that fx is continuous since πX ◦ fx (y) = x and πY ◦ fx (y) = y are continuous
yα if α ∈ Λ
maps. From this we conclude that {x} × Y = fx (Y ) is connected by Lemma xα =
pα if α ∈
/ Λ.
13.46. A similar argument shows X × {y} is connected for all y ∈ Y.
Let p = (x0 , y0 ) ∈ X × Y and Cp denote the connected component of p. If α ∈ Λ, πα ◦ φ(y) = yα = πα (y) and if α ∈ A \ Λ then πα ◦ φ(y) = pα so that in
Since {x0 } × Y is connected and p ∈ {x0 } × Y it follows that {x0 } × Y ⊂ Cp every case πα ◦ φ : XΛ → Xα is continuous and therefore φ is continuous. Since
and hence Cp is also the connected component (x0 , y) for all y ∈ Y. Similarly, XΛ is a product of a finite number of connected spaces and so is connected
X × {y} ⊂ C(x0 ,y) = Cp is connected, and therefore X × {y} ⊂ Cp . So we have and thus so is the continuous image, φ(XΛ ) = XΛ × {pα }α∈A\Λ ⊂ XΛ . Now
shown (x, y) ∈ Cp for all x ∈ X and y ∈ Y, see Figure 13.4. By induction the
p ∈ φ(XΛ ) and φ(XΛ ) is connected implies that φ(XΛ ) ⊂ C. On the other hand
theorem holds whenever A is a finite set, i.e. for products of a finite number of
one easily sees that
connected spaces.
∅=6 V ∩ φ(XΛ ) ⊂ V ∩ C
For the general case, again choose a point p ∈ XA = X A and again let
C = Cp be the connected component of p. Recall that Cp is closed and therefore contradicting the assumption that V ⊂ C c .

13.6 Exercises 121
13.6 Exercises Exercise 13.22. Prove item 3. of Proposition 13.52.
13.6.1 General Topological Space Problems Exercise 13.23. Let
X := (x, y) ∈ R2 : y = sin(x−1 ) ∪ {(0, 0)}

Exercise 13.14. Let V be an open subset of R. Show V may be written as
a disjoint union of open intervals Jn = (an , bn ), where an , bn ∈ R∪ {±∞} for
n = 1, 2, · · · < N with N = ∞ possible. equipped with the relative topology induced from the standard topology on R2 .
Show X is connected but not path connected.
Exercise 13.15. Let (X, τ ) and (Y, τ 0 ) be a topological spaces, f : X → Y be
n
a function, U be an open cover of X and {Fj }j=1 be a finite cover of X by
closed sets. 13.6.3 Metric Spaces as Topological Spaces
1. If A ⊂ X is any set and f : X → Y is (τ, τ 0 ) – continuous then f |A : A → Y Definition 13.54. Two metrics d and ρ on a set X are said to be equivalent
is (τA , τ 0 ) – continuous. if there exists a constant c ∈ (0, ∞) such that c−1 ρ ≤ d ≤ cρ.
2. Show f : X → Y is (τ, τ 0 ) – continuous iff f |U : U → Y is (τU , τ 0 ) –
continuous for all U ∈ U. Exercise 13.24. Suppose that d and ρ are two metrics on X.
3. Show f : X → Y is (τ, τ 0 ) – continuous iff f |Fj : Fj → Y is (τFj , τ 0 ) – 1. Show τd = τρ if d and ρ are equivalent.
continuous for all j = 1, 2, . . . , n. 2. Show by example that it is possible for τd = τρ even thought d and ρ are
Exercise 13.16. Suppose that X is a set, {(Yα , τα ) : α ∈ A} is a family of inequivalent.
topological spaces and fα : X → Yα is a given function for all α ∈ A. Assuming Exercise 13.25. Let (Xi , di ) for i = 1, . . . , n be a finite collection of metric
that Sα ⊂ τα is a sub-base for the topology τα for each α ∈ A, show S := spacesQand for 1 ≤ p ≤ ∞ and x = (x1 , x2 , . . . , xn ) and y = (y1 , . . . , yn ) in
∪α∈A fα−1 (Sα ) is a sub-base for the topology τ := τ (fα : α ∈ A). n
X := i=1 Xi , let
p 1/p
Pn
13.6.2 Connectedness Problems ( i=1 [di (xi , yi )] ) if p 6= ∞ .
ρp (x, y) =
maxi di (xi , yi ) if p = ∞
Exercise 13.17. Show any non-trivial interval in Q is disconnected.
Exercise 13.18. Suppose a < b and f : (a, b) → R is a non-decreasing function. 1. Show (X, ρp ) is a metric space for p ∈ [1, ∞]. Hint: Minkowski’s inequality.
Show if f satisfies the intermediate value property (see Theorem 13.50), then 2. Show for any p, q ∈ [1, ∞], the metrics ρp and ρq are equivalent. Hint: This
f is continuous. can be done with explicit estimates or you could use Theorem 14.12 below.
∞
Exercise 13.19. Suppose −∞ < a < b ≤ ∞ and f : [a, b) → R is a strictly Notation 13.55 Let X be a set and p := {pn }n=0 be a family of semi-metrics
increasing continuous function. By Lemma 13.46, f ([a, b)) is an interval and on X, i.e. pn : X × X → [0, ∞) are functions satisfying the assumptions of
since f is strictly increasing it must of the form [c, d) for some c ∈ R and d ∈ R̄ metric except for the assertion that pn (x, y) = 0 implies x = y. Further assume
with c < d. Show the inverse function f −1 : [c, d) → [a, b) is continuous and is that pn (x, y) ≤ pn+1 (x, y) for all n and if pn (x, y) = 0 for all n ∈ N then x = y.
strictly increasing. In particular if n ∈ N, apply this result to f (x) = xn for Given n ∈ N and x ∈ X let
x ∈ [0, ∞) to construct the positive nth – root of a real number. Compare with
Exercise 3.8 Bn (x, ε) := {y ∈ X : pn (x, y) < ε} .
Exercise 13.20. Prove item 1. of Proposition 13.52. Hint: show X is not con- We will write τ (p) form the smallest topology on X such that pn (x, ·) : X →
nected implies X is not path connected. [0, ∞) is continuous for all n ∈ N and x ∈ X, i.e. τ (p) := τ (pn (x·) : n ∈ N and
x ∈ X).
Exercise 13.21. Prove item 2. of Proposition 13.52. Hint: fix x0 ∈ X and let
W denote the set of x ∈ X such that there exists σ ∈ C([0, 1], X) satisfying
σ(0) = x0 and σ(1) = x. Then show W is both open and closed.

Exercise 13.26. Using Notation 13.55, show that collection of balls,
B := {Bn (x, ε) : n ∈ N, x ∈ X and ε > 0} ,
forms a base for the topology τ (p). Hint: Use Exercise 13.16 to show B is a
sub-base for the topology τ (p) and then use Exercise 13.2 to show B is in fact
a base for the topology τ (p).
Exercise 13.27 (A minor variant of Exercise 6.12). Let pn be as in No-

tation 13.55 and
∞
X pn (x, y)
d(x, y) := 2−n .
n=0
1 + pn (x, y)
∞
Show d is a metric on X and τd = τ (p). Conclude that a sequence {xk }k=1 ⊂ X
converges to x ∈ X iff
lim pn (xk , x) = 0 for all n ∈ N.

k→∞
∞
Exercise
Q∞ 13.28. Let {(Xn , dn )}n=1 be a sequence of metric spaces, X :=
∞ ∞
n=1 X n , and for x = (x(n))n=1 and y = (y(n))n=1 in X let
∞
X dn (x(n), y(n))
d(x, y) = 2−n .
n=1
1 + dn (x(n), y(n))
(See Exercise 6.12.) Moreover, let πn : X → Xn be the projection maps, show
τd = ⊗∞
n=1 τdn := τ ({πn : n ∈ N}).
That is show the d – metric topology is the same as the product topology on
X. Suggestions: 1) show πn is τd continuous for each n and 2) show for each
x ∈ X that d (x, ·) is ⊗∞
n=1 τdn – continuous.
For the second assertion notice
P∞ −n dn (x(n),·)
that d (x, ·) = n=1 fn where fn = 2 1+dn (x(n),·) ◦ πn .
14
Compactness
Definition 14.1. The subset A of a topological space (X τ ) is said to be com- Proof. (⇒)TSuppose that X is compact and F ⊂ 2X is a collection of closed
pact if every open cover (Definition 13.18) of A has finite a sub-cover, i.e. if U sets such that F = ∅. Let
is an open cover of A there exists U0 ⊂⊂ U such that U0 is a cover of A. (We
will write A @@ X to denote that A ⊂ X and A is compact.) A subset A ⊂ X U = F c := {C c : C ∈ F} ⊂ τ,
is precompact if Ā is compact.
then U is a cover of X and hence has a finite subcover, U0 . Let F0 = U0c ⊂⊂ F,
Proposition 14.2. Suppose that K ⊂ X is a compact set and F ⊂ K is a then ∩F0 = ∅ so that F does not have the finite intersection property. (⇐) If
n
closed subset. Then F is compact. If {Ki }i=1 is a finite collections of compact X is not compact, there exists an open cover U of X with no finite subcover.
subsets of X then K = ∪ni=1 Ki is also a compact subset of X. Let
F = U c := {U c : U ∈ U } ,
Proof. Let U ⊂ τ be an open cover of F, then U∪ {F c } is an open cover of
K. The cover U∪ {F c } of K has a finite subcover which we denote by U0 ∪ {F c } T F is a collection of closed sets with the finite intersection property while
then
where U0 ⊂⊂ U. Since F ∩ F c = ∅, it follows that U0 is the desired subcover F = ∅.
of F. For the second assertion suppose U ⊂ τ is an open cover of K. Then U
covers each compact set Ki and therefore there exists a finite subset Ui ⊂⊂ U Exercise 14.4. Let (X, τ ) be a topological space. Show that A ⊂ X is compact
for each i such that Ki ⊂ ∪Ui . Then U0 := ∪ni=1 Ui is a finite cover of K. iff (A, τA ) is a compact topological space.
Exercise 14.1 (Suggested by Michael Gurvich). Show by example that

the intersection of two compact sets need not be compact. (This pathology 14.1 Metric Space Compactness Criteria
disappears if one assumes the topology is Hausdorff, see Definition 15.2 below.)
Exercise 14.2. Suppose f : X → Y is continuous and K ⊂ X is compact, then Let (X, d) be a metric space and for x ∈ X and ε > 0 let
f (K) is a compact subset of Y. Give an example of continuous map, f : X → Y,
Bx0 (ε) := Bx (ε) \ {x}
and a compact subset K of Y such that f −1 (K) is not compact.
Exercise 14.3 (Dini’s Theorem). Let X be a compact topological space be the ball centered at x of radius ε > 0 with x deleted. Recall from Definition
and fn : X → [0, ∞) be a sequence of continuous functions such that fn (x) ↓ 0 13.29 that a point x ∈ X is an accumulation point of a subset E ⊂ X if
as n → ∞ for each x ∈ X. Show that in fact fn ↓ 0 uniformly in x, i.e. ∅=6 E ∩ V \ {x} for all open neighborhoods, V, of x. The proof of the following
supx∈X fn (x) ↓ 0 as n → ∞. Hint: Given ε > 0, consider the open sets Vn := elementary lemma is left to the reader.
{x ∈ X : fn (x) < ε}.
Lemma 14.5. Let E ⊂ X be a subset of a metric space (X, d) . Then the fol-
Definition 14.3. A collection F of closed subsets of a topological space (X, τ ) lowing are equivalent:
has the finite intersection property if ∩F0 6= ∅ for all F0 ⊂⊂ F.
1. x ∈ X is an accumulation point of E.
The notion of compactness may be expressed in terms of closed sets as 2. Bx0 (ε) ∩ E 6= ∅ for all ε > 0.
follows. 3. Bx (ε) ∩ E is an infinite set for all ε > 0.
∞
4. There exists {xn }n=1 ⊂ E \ {x} with limn→∞ xn = x.
Proposition 14.4. A topological space X is compact iff every family of closed
sets F ⊂ 2X having the finite intersection property satisfies F =
T
6 ∅.
124 14 Compactness
Definition 14.6. A metric space (X, d) is ε – bounded (ε > 0) if there exists so that d (xn , xm ) ≥ ε > 0 for every m 6= n and hence no subsequence of
∞
a finite cover of X by balls of radius ε and it is totally bounded if it is ε – {xn }n=1 can be Cauchy.
bounded for all ε > 0.
Theorem 14.7. Let (X, d) be a metric space. The following are equivalent.
(a) X is compact.
(b) Every infinite subset of X has an accumulation point.
∞
(c) Every sequence {xn }n=1 ⊂ X has a convergent subsequence.
(d) X is totally bounded and complete.
Proof. The proof will consist of showing that a ⇒ b ⇒ c ⇒ d ⇒ a.

(a ⇒ b) We will show that not b ⇒ not a. Suppose there exists an infinite
subset E ⊂ X which has no accumulation points. Then for all x ∈ X there
exists δx > 0 such that Vx := Bx (δx ) satisfies (Vx \ {x}) ∩ E = ∅. Clearly
V = {Vx }x∈X is a cover of X, yet V has no finite sub cover. Indeed, for each
x ∈ X, Vx ∩ E ⊂ {x} and hence if Λ ⊂⊂ X, ∪x∈Λ Vx can only contain a finite Fig. 14.2. Constructing a set with out an accumulation point.
number of points from E (namely Λ ∩ E). Thus for any Λ ⊂⊂ X, E " ∪x∈Λ Vx
and in particular X 6= ∪x∈Λ Vx . (See Figure 14.1.)
(d ⇒ a) For sake of contradiction, assume there exists an open cover V =
{Vα }α∈A of X with no finite subcover. Since X is totally bounded for each
n ∈ N there exists Λn ⊂⊂ X such that
[ [
X= Bx (1/n) ⊂ Cx (1/n).
x∈Λn x∈Λn
Fig. 14.1. The construction of an open cover with no finite sub-cover. Choose x1 ∈ Λ1 such that no finite subset of V covers K1 := Cx1 (1). Since
K1 = ∪x∈Λ2 K1 ∩ Cx (1/2), there exists x2 ∈ Λ2 such that K2 := K1 ∩ Cx2 (1/2)
can not be covered by a finite subset of V, see Figure 14.3. Continuing this way
∞
(b ⇒ c) Let {xn }n=1 ⊂ X be a sequence and E := {xn : n ∈ N} . If inductively, we construct sets Kn = Kn−1 ∩Cxn (1/n) with xn ∈ Λn such that no
∞ ∞
#(E) < ∞, then {xn }n=1 has a subsequence {xnk }k=1 which is constant and Kn can be covered by a finite subset of V. Now choose yn ∈ Kn for each n. Since
∞
hence convergent. On the other hand if #(E) = ∞ then by assumption E has {Kn }n=1 is a decreasing sequence of closed sets such that diam(Kn ) ≤ 2/n, it
∞
an accumulation point and hence by Lemma 14.5, {xn }n=1 has a convergent follows that {yn } is a Cauchy and hence convergent with
subsequence.
∞ y = lim yn ∈ ∩∞
m=1 Km .
(c ⇒ d) Suppose {xn }n=1 ⊂ X is a Cauchy sequence. By assumption there n→∞
∞
exists a subsequence {xnk }k=1 which is convergent to some point x ∈ X. Since
∞
{xn }n=1 is Cauchy it follows that xn → x as n → ∞ showing X is complete. We Since V is a cover of X, there exists V ∈ V such that y ∈ V. Since Kn ↓ {y}
now show that X is totally bounded. Let ε > 0 be given and choose an arbitrary and diam(Kn ) → 0, it now follows that Kn ⊂ V for some n large. But this
point x1 ∈ X. If possible choose x2 ∈ X such that d(x2 , x1 ) ≥ ε, then if possible violates the assertion that Kn can not be covered by a finite subset of V.
choose x3 ∈ X such that d{x1 ,x2 } (x3 ) ≥ ε and continue inductively choosing
n
points {xj }j=1 ⊂ X such that d{x1 ,...,xn−1 } (xn ) ≥ ε. (See Figure 14.2.) This Corollary 14.8. Any compact metric space (X, d) is second countable and
∞
process must terminate, for otherwise we would produce a sequence {xn }n=1 ⊂ hence also separable by Exercise 13.11. (See Example 15.25 below for an ex-
X which can have no convergent subsequences. Indeed, the xn have been chosen ample of a compact topological space which is not separable.)

14.1 Metric Space Compactness Criteria 125
n
X 2
d2 (x, y) = (yi − xi ) ≤ nδ 2
i=1
√ √
which shows that d(x, y) ≤ nδ. Hence if choose δ < ε/ n we have shows that
d(x, y) < ε, i.e. Eq. (14.1) holds.
Example 14.10. Let X = `p (N) with p ∈ [1, ∞) and µ ∈ `p (N) such that µ(k) ≥
0 for all k ∈ N. The set
K := {x ∈ X : |x(k)| ≤ µ(k) for all k ∈ N}

∞
is compact. To prove this, let {xn }n=1 ⊂ K be a sequence. By compactness of
∞
closed bounded sets in C, for each k ∈ N there is a subsequence of {xn (k)}n=1 ⊂
Fig. 14.3. Nested Sequence of cubes.
C which is convergent. By Cantor’s diagonalization trick, we may choose a
∞ ∞
subsequence {yn }n=1 of {xn }n=1 such that y(k) := limn→∞ yn (k) exists for all
1
k ∈ N. Since |yn (k)| ≤ µ(k) for all n it follows that |y(k)| ≤ µ(k), i.e. y ∈ K.
Proof. To each integer n, there exists Λn ⊂⊂ X such that X = Finally
∪x∈Λn B(x, 1/n). The collection of open balls,
∞
X ∞
X
p p p
V := ∪n∈N ∪x∈Λn {B(x, 1/n)} lim ky − yn kp = lim |y(k) − yn (k)| = lim |y(k) − yn (k)| = 0
n→∞ n→∞ n→∞
k=1 k=1
forms a countable basis for the metric topology on X. To check this, suppose
wherein we have used the Dominated convergence theorem. (Note
that x0 ∈ X and ε > 0 are given and choose n ∈ N such that 1/n < ε/2
and x ∈ Λn such that d (x0 , x) < 1/n. Then B(x, 1/n) ⊂ B (x0 , ε) because for p
|y(k) − yn (k)| ≤ 2p µp (k)
y ∈ B(x, 1/n),
d (y, x0 ) ≤ d (y, x) + d (x, x0 ) < 2/n < ε. and µp is summable.) Therefore yn → y and we are done.
Alternatively, we can prove K is compact by showing that K is closed and
∞
totally bounded. It is simple to show K is closed, for if {xn }n=1 ⊂ K is a
Corollary 14.9. The compact subsets of Rn are the closed and bounded sets. convergent sequence in X, x := limn→∞ xn , then
Proof. This is a consequence of Theorem 10.2 and Theorem 14.7. Here is |x(k)| ≤ lim |xn (k)| ≤ µ(k) ∀ k ∈ N.
n→∞
another proof. If K is closed and bounded then K is complete (being the closed
subset of a complete space) and K is contained in [−M, M ]n for some positive This shows that x ∈ K and hence K is closed. To see that K is totally
integer M. For δ > 0, let P∞ p 1/p
bounded, let ε > 0 and choose N such that k=N +1 |µ(k)| < ε. Since
Λδ = δZn ∩ [−M, M ]n := {δx : x ∈ Zn and δ|xi | ≤ M for i = 1, 2, . . . , n}. 1
The argument is as follows. Let {n1j }∞ ∞
j=1 be a subsequence of N = {n}n=1 such that
limj→∞ xn1 (1) exists. Now choose a subsequence {n2j }∞ j=1 of {n 1 ∞
}
j j=1 such that
j
We will show, by choosing δ > 0 sufficiently small, that
limj→∞ xn2 (2) exists and similarly {n3j }∞ 2 ∞
j=1 of {nj }j=1 such that limj→∞ xn3 (3)
j j
K ⊂ [−M, M ]n ⊂ ∪x∈Λδ B(x, ε) (14.1) exists. Continue on this way inductively to get
{n}∞ 1 ∞ 2 ∞ 3 ∞
n=1 ⊃ {nj }j=1 ⊃ {nj }j=1 ⊃ {nj }j=1 ⊃ . . .
which shows that K is totally bounded. Hence by Theorem 14.7, K is compact.
Suppose that y ∈ [−M, M ]n , then there exists x ∈ Λδ such that |yi − xi | ≤ δ such that limj→∞ xnk (k) exists for all k ∈ N. Let mj := njj so that eventually
j
for i = 1, 2, . . . , n. Hence {mj }∞ k ∞
j=1 is a subsequence of {nj }j=1 for all k. Therefore, we may take yj := xmj .

126 14 Compactness
QN n
k=1Cµ(k) (0) ⊂ CN is closed and bounded, it is compact. Therefore there Proof. Let {fi }i=1 be a basis for L and define a new norm on L by
QN
exists a finite subset Λ ⊂ k=1 Cµ(k) (0) such that v
Xn u n
uX 2
N ai fi := t |ai | for ai ∈ F.

Y
Cµ(k) (0) ⊂ ∪z∈Λ BzN (ε)
i=1

2 i=1
k=1
By the triangle inequality for the norm |·| , we find
where BzN (ε) is the open ball centered at z ∈ CN relative to the
`p ({1, 2, 3, . . . , N }) – norm. For each z ∈ Λ, let z̃ ∈ X be defined by v v
Xn X n u n
X
u n Xn
uX
2t 2
z̃(k) = z(k) if k ≤ N and z̃(k) = 0 for k ≥ N + 1. I now claim that
u
ai fi ≤ |ai | |fi | ≤ |fi | |ai | ≤ M ai fi
t

i=1 i=1 i=1 i=1 i=1 2
K ⊂ ∪z∈Λ Bz̃ (2ε) (14.2)
qP
n 2
which, when verified, shows K is totally bounded. To verify Eq. (14.2), let where M = i=1 |fi | . Thus we have
x ∈ K and write x = u + v where u(k) = x(k) for k ≤ N and u(k) = 0 for
k < N. Then by construction u ∈ Bz̃ (ε) for some z̃ ∈ Λ and |f | ≤ M kf k2
!1/p
∞
X for all f ∈ L and this inequality shows that |·| is continuous relative to
p
kvkp ≤ |µ(k)| < ε. k·k2 . Since the normed space (L, k·k2 ) is homeomorphic and isomorphic
k=N +1 to Fn with the standard euclidean norm, the closed bounded set, S :=
So we have {f ∈ L : kf k2 = 1} ⊂ L, is a compact subset of L relative to k·k2 . Therefore by
Exercise 14.5 there exists f0 ∈ S such that
kx − z̃kp = ku + v − z̃kp ≤ ku − z̃kp + kvkp < 2ε.
m = inf {|f | : f ∈ S} = |f0 | > 0.
Exercise 14.5 (Extreme value theorem). Let (X, τ ) be a compact topo-
f
logical space and f : X → R be a continuous function. Show −∞ < inf f ≤ Hence given 0 6= f ∈ L, then kf k2 ∈ S so that
sup f < ∞ and there exists a, b ∈ X such that f (a) = inf f and f (b) = sup f 2 .
f
Hint: use Exercise 14.2 and Corollary 14.9. = |f | 1

m ≤
kf k 2
kf k 2
Exercise 14.6 (Uniform Continuity). Let (X, d) be a compact metric space,
(Y, ρ) be a metric space and f : X → Y be a continuous function. Show that f is or equivalently
uniformly continuous, i.e. if ε > 0 there exists δ > 0 such that ρ(f (y), f (x)) < ε 1
|f | .
kf k2 ≤
if x, y ∈ X with d(x, y) < δ. Hint: you could follow the argument in the proof m
of Theorem 10.2. This shows that |·| and k·k2 are equivalent norms. Similarly one shows that k·k
Definition 14.11. Let L be a vector space. We say that two norms, |·| and k·k , and k·k2 are equivalent and hence so are |·| and k·k .
on L are equivalent if there exists constants α, β ∈ (0, ∞) such that Corollary 14.13. If (L, k·k) is a finite dimensional normed space, then A ⊂ L
kf k ≤ α |f | and |f | ≤ β kf k for all f ∈ L. is compact iff A is closed and bounded relative to the given norm, k·k .
Theorem 14.12. Let L be a finite dimensional vector space. Then any two Corollary 14.14. Every finite dimensional normed vector space (L, k·k) is
norms |·| and k·k on L are equivalent. (This is typically not true for norms on complete. In particular any finite dimensional subspace of a normed vector space
infinite dimensional spaces, see for example Exercise 7.6.) is automatically closed.
∞ ∞
2
Here is a proof if X is a metric space. Let {xn }∞
n=1 ⊂ X be a sequence such that
Proof. If {fn }n=1 ⊂ L is a Cauchy sequence, then {fn }n=1 is bounded and
f (xn ) ↑ sup f. By compactness of X we may assume, by passing to a subsequence hence has a convergent subsequence, gk = fnk , by Corollary 14.13. It is now
if necessary that xn → b ∈ X as n → ∞. By continuity of f, f (b) = sup f. routine to show limn→∞ fn = f := limk→∞ gk .

14.2 Compact Operators 127
Theorem 14.15. Suppose that (X, k·k) is a normed vector in which the unit 14.2 Compact Operators
ball, V := B0 (1) , is precompact. Then dim X < ∞.
Definition 14.16. Let A : X → Y be a bounded operator between two Banach
Proof. Since V̄ is compact, we may choose Λ ⊂⊂ X such that spaces. Then A is compact if A [BX (0, 1)] is precompact in Y or equivalently
for any {xn }∞
n=1 ⊂ X such that kxn k ≤ 1 for all n the sequence yn := Axn ∈ Y
1
V̄ ⊂ ∪x∈Λ x + V (14.3) has a convergent subsequence.
2
Example 14.17. Let X = `2 = Y and λn ∈ C such that limn→∞ λn = 0, then
where, for any δ > 0,
A : X → Y defined by (Ax)(n) = λn x(n) is compact.
δV := {δx : x ∈ V } = B0 (δ) . 2
Proof. Suppose {xj }∞ 2 2
P
j=1 ⊂ ` such that kxj k = |xj (n)| ≤ 1 for all j.
Let Y := span(Λ), then Eq. (14.3) implies, By Cantor’s Diagonalization argument, there exists {jk } ⊂ {j} such that, for
each n, x̃k (n) = xjk (n) converges to some x̃(n) ∈ C as k → ∞. By Fatou’s
1 Lemma 4.12,
V ⊂ V̄ ⊂ Y + V.
2 ∞ ∞ ∞
X X 2
X 2
2
Multiplying this equation by 1
then shows |x̃(n)| = lim inf |x̃k (n)| ≤ lim inf |x̃k (n)| ≤ 1,
2 k→∞ k→∞
n=1 n=1 n=1
1 1 1 1
V ⊂ Y + V =Y + V which shows x̃ ∈ `2 .
2 2 4 4 Let λ∗M = max |λn |. Then
n≥M
and hence
1 1 1 ∞
V ⊂ Y + V ⊂ Y + Y + V = Y + V. X 2
2 4 4 kAx̃k − Ax̃k2 = |λn |2 |x̃k (n) − x̃(n)|
Continuing this way inductively then shows that n=1
M ∞
1
X X
V ⊂ Y + n V for all n ∈ N. (14.4) ≤ |λn |2 |x̃k (n) − x̃(n)|2 + |λ∗M |2 |x̃k (n) − x̃(n)|2
2 n=1 M +1
M
Indeed, if Eq. (14.4) holds, then X 2
≤ |λn |2 |x̃k (n) − x̃(n)|2 + |λ∗M |2 kx̃k − x̃k

1 1 1 1 n=1
V ⊂Y + V ⊂Y + Y + V =Y + V. M
2 2 2n 2n+1 X
≤ |λn |2 |x̃k (n) − x̃(n)|2 + 4|λ∗M |2 .
−n
Hence if x ∈ V, there exists yn ∈ Y and zn ∈ B0 (2 ) such that yn + zn → x. n=1
Since limn→∞ zn = 0, it follows that x = limn→∞ yn ∈ Ȳ . Since dim Y ≤ Passing to the limit in this inequality then implies
# (Λ) < ∞, Corollary 14.14 implies Y = Ȳ and so we have shown that V ⊂ Y.
1
Since for any x ∈ X, 2kxk x ∈ V ⊂ Y, we have x ∈ Y for all x ∈ X, i.e. X = Y. lim sup kAx̃k − Ax̃k2 ≤ 4|λ∗M |2 → 0 as M → ∞
k→∞
Exercise 14.7. Suppose (Y, k·kY ) is a normed space and (X, k·kX ) is a finite and this completes the proof the A is a compact operator.
dimensional normed space. Show every linear transformation T : X → Y is A B
necessarily bounded. Lemma 14.18. If X −→ Y −→ Z are bounded operators such the either A or
B is compact then the composition BA : X → Z is also compact.

128 14 Compactness
Proof. Let BX (0, 1) be the open unit ball in X. If A is compact and B Exercise 14.8. If (X, τ ) is locally compact and second countable, then there is
is bounded, then BA(BX (0, 1)) ⊂ B(ABX (0, 1)) which is compact since the a countable basis B0 for the topology consisting of precompact open sets. Use
image of compact sets under continuous maps are compact. Hence we con- this to show (X, τ ) is σ - compact.
clude that BA(BX (0, 1)) is compact, being the closed subset of the compact
Exercise 14.9. Every separable locally compact metric space is σ – compact.
set B(ABX (0, 1)). If A is continuous and B is compact, then A(BX (0, 1)) is a
bounded set and so by the compactness of B, BA(BX (0, 1)) is a precompact Exercise 14.10. Every σ – compact metric space is second countable (or equiv-
subset of Z, i.e. BA is compact. alently separable), see Corollary 14.8.
Exercise 14.11. Suppose that (X, d) is a metric space and U ⊂ X is an open
subset.
14.3 Local and σ – Compactness
1. If X is locally compact then (U, d) is locally compact.
Notation 14.19 If X is a topological space and Y is a normed space, let 2. If X is σ – compact then (U, d) is σ – compact. Hint: Mimic Example 14.22,
replacing {x ∈ Rn : |x| ≤ k} by compact sets Xk @@ X such that Xk ↑ X.
BC(X, Y ) := {f ∈ C(X, Y ) : sup kf (x)kY < ∞}
x∈X Lemma 14.23. Let (X, τ ) be locally and σ – compact. Then there exists com-
o
pact sets Kn ↑ X such that Kn ⊂ Kn+1 ⊂ Kn+1 for all n.
and
Cc (X, Y ) := {f ∈ C(X, Y ) : supp(f ) is compact}. Proof. Suppose that C ⊂ X is a compact set. For each x ∈ C let Vx ⊂o X
be an open neighborhood of x such that V̄x is compact. Then C ⊂ ∪x∈C Vx so
If Y = R or C we will simply write C(X), BC(X) and Cc (X) for C(X, Y ),
there exists Λ ⊂⊂ C such that
BC(X, Y ) and Cc (X, Y ) respectively.
C ⊂ ∪x∈Λ Vx ⊂ ∪x∈Λ V̄x =: K.
Remark 14.20. Let X be a topological space and Y be a Banach space. By com-
bining Exercise 14.2 and Theorem 14.7 it follows that Cc (X, Y ) ⊂ BC(X, Y ). Then K is a compact set, being a finite union of compact subsets of X, and
C ⊂ ∪x∈Λ Vx ⊂ K o . Now let Cn ⊂ X be compact sets such that Cn ↑ X as
Definition 14.21 (Local and σ – compactness). Let (X, τ ) be a topological n → ∞. Let K1 = C1 and then choose a compact set K2 such that C2 ⊂ K2o .
space. Similarly, choose a compact set K3 such that K2 ∪ C3 ⊂ K3o and continue
o
1. (X, τ ) is locally compact if for all x ∈ X there exists an open neighborhood inductively to find compact sets Kn such that Kn ∪ Cn+1 ⊂ Kn+1 for all n.
∞
V ⊂ X of x such that V̄ is compact. (Alternatively, in light of Definition Then {Kn }n=1 is the desired sequence.
13.29 (also see Definition 6.5), this is equivalent to requiring that to each Remark 14.24. Lemma 14.23 may also be stated as saying there exists precom-
x ∈ X there exists a compact neighborhood Nx of x.) ∞
pact open sets {Gn }n=1 such that Gn ⊂ Ḡn ⊂ Gn+1 for all n and Gn ↑ X as
2. (X, τ ) is σ – compact if there exists compact sets Kn ⊂ X such that X = ∞ ∞
n → ∞. Indeed if {Gn }n=1 are as above, let Kn := Ḡn and if {Kn }n=1 are as
∪∞n=1 Kn . (Notice that we may assume, by replacing Kn by K1 ∪K2 ∪· · ·∪Kn in Lemma 14.23, let Gn := Kno .
if necessary, that Kn ↑ X.)
Proposition 14.25. Suppose X is a locally compact metric space and U ⊂o X
Example 14.22. Any open subset of U ⊂ Rn is a locally compact and σ – com- and K @@ U. Then there exists V ⊂o X such that K ⊂ V ⊂ V ⊂ U ⊂ X and
pact metric space. The proof of local compactness is easy and is left to the V̄ is compact.
reader. To see that U is σ – compact, for k ∈ N, let
Proof. (This is done more generally in Proposition 15.7 below.) By local
Kk := {x ∈ U : |x| ≤ k and dU c (x) ≥ 1/k} . compactness of X, for each x ∈ K there exists εx > 0 such that Bx (εx ) is
compact and by shrinking εx if necessary we may assume,
Then Kk is a closed and bounded subset of Rn and hence compact. Moreover
Kko ↑ U as k → ∞ since3 Bx (εx ) ⊂ Cx (εx ) ⊂ Bx (2εx ) ⊂ U
for each x ∈ K. By compactness of K, there exists Λ ⊂⊂ K such that K ⊂
Kko ⊃ {x ∈ U : |x| < k and dU c (x) > 1/k} ↑ U as k → ∞.
∪x∈Λ Bx (εx ) =: V. Notice that V̄ ⊂ ∪x∈Λ Bx (εx ) ⊂ U and V̄ is a closed subset
3
In fact this is an equality, but we will not need this here. of the compact set ∪x∈Λ Bx (εx ) and hence compact as well.

14.4 Function Space Compactness Criteria 129
S Λ

Definition 14.26. Let U be an open subset of a topological space (X, τ ). We From this it follows that F = Fφ : φ ∈ D where, for φ ∈ DΛ ,
will write f ≺ U to mean a function f ∈ Cc (X, [0, 1]) such that supp(f ) :=
{f 6= 0} ⊂ U. Fφ := {f ∈ F : |f (y) − φ(x)| < ε for y ∈ Vx and x ∈ Λ}.

Lemma 14.27 (Urysohn’s Lemma for Metric Spaces). Let X be a locally Let Γ := φ ∈ DΛ : Fφ 6= ∅ and for each φ ∈ Γ choose fφ ∈ Fφ ∩ F. For
compact metric space and K @@ U ⊂o X. Then there exists f ≺ U such that f ∈ Fφ , x ∈ Λ and y ∈ Vx we have
f = 1 on K. In particular, if K is compact and C is closed in X such that
K ∩ C = ∅, there exists f ∈ Cc (X, [0, 1]) such that f = 1 on K and f = 0 on C. |f (y) − fφ (y)| ≤ |f (y) − φ(x))| + |φ(x) − fφ (y)| < 2ε.
Proof. Let V be as in Proposition 14.25 and then use Lemma 6.15 to find So kf − fφ k∞ < 2ε for all f ∈ Fφ showing that Fφ ⊂ Bfφ (2ε). Therefore,
a function f ∈ C(X, [0, 1]) such that f = 1 on K and f = 0 on V c . Then
supp(f ) ⊂ V̄ ⊂ U and hence f ≺ U. F = ∪φ∈Γ Fφ ⊂ ∪φ∈Γ Bfφ (2ε)
and because ε > 0 was arbitrary we have shown that F is totally bounded.
(⇒) (*The rest of this proof may safely be skipped.) Since k·k∞ : C(X) →
14.4 Function Space Compactness Criteria [0, ∞) is a continuous function on C(X) it is bounded on any compact subset
F ⊂ C(X). This shows that sup {kf k∞ : f ∈ F} < ∞ which clearly implies that
In this section, let (X, τ ) be a topological space. F is pointwise bounded.4 Suppose F were not equicontinuous at some point
x ∈ X that is to say there exists ε > 0 such that for all V ∈ τx , sup sup |f (y) −
Definition 14.28. Let F ⊂ C(X). y∈V f ∈F
f (x)| > ε.5 Equivalently said, to each V ∈ τx we may choose
1. F is equicontinuous at x ∈ X iff for all ε > 0 there exists U ∈ τx such
that |f (y) − f (x)| < ε for all y ∈ U and f ∈ F. fV ∈ F and xV ∈ V 3 |fV (x) − fV (xV )| ≥ ε. (14.5)
2. F is equicontinuous if F is equicontinuous at all points x ∈ X.
3. F is pointwise bounded if sup{|f (x)| : f ∈ F} < ∞ for all x ∈ X. k·k∞
Set CV = {fW : W ∈ τx and W ⊂ V } ⊂ F and notice for any V ⊂⊂ τx that
Theorem 14.29 (Ascoli-Arzela Theorem). Let (X, τ ) be a compact topolog-
∩V ∈V CV ⊇ C∩V 6= ∅,
ical space and F ⊂ C(X). Then F is precompact in C(X) iff F is equicontinuous
and point-wise bounded. 4
One could also prove that F is pointwise bounded by considering the continuous
evaluation maps ex : C(X) → R given by ex (f ) = f (x) for all x ∈ X.
Proof. (⇐) Since C(X) ⊂ `∞ (X) is a complete metric space, we must show 5
If X is first countable we could finish the proof with the following argument. Let
F is totally bounded. Let ε > 0 be given. By equicontinuity, for all x ∈ X, there {Vn }∞
n=1 be a neighborhood base at x such that V1 ⊃ V2 ⊃ V3 ⊃ . . . . By the
exists Vx ∈ τx such that |f (y) − f (x)| < ε/2 if y ∈ Vx and f ∈ F. Since X assumption that F is not equicontinuous at x, there exist fn ∈ F and xn ∈ Vn such
is compact we may choose Λ ⊂⊂ X such that X = ∪x∈Λ Vx . We have now that |fn (x) − fn (xn )| ≥ ∀ n. Since F is a compact metric space by passing to
decomposed X into “blocks” {Vx }x∈Λ such that each f ∈ F is constant to a subsequence if necessary we may assume that fn converges uniformly to some
within ε on Vx . Since sup {|f (x)| : x ∈ Λ and f ∈ F} < ∞, it is now evident f ∈ F. Because xn → x as n → ∞ we learn that
that ≤ |fn (x) − fn (xn )| ≤ |fn (x) − f (x)| + |f (x) − f (xn )| + |f (xn ) − fn (xn )|
M = sup {|f (x)| : x ∈ X and f ∈ F} ≤ 2kfn − f k + |f (x) − f (xn )| → 0 as n → ∞
≤ sup {|f (x)| : x ∈ Λ and f ∈ F} + ε < ∞. which is a contradiction.
Let D := {kε/2 : k ∈ Z} ∩ [−M, M ]. If f ∈ F and φ ∈ DΛ (i.e. φ : Λ → D is a

function) is chosen so that |φ(x) − f (x)| ≤ ε/2 for all x ∈ Λ, then
|f (y) − φ(x)| ≤ |f (y) − f (x)| + |f (x) − φ(x)| < ε ∀ x ∈ Λ and y ∈ Vx .

130 14 Compactness
∞
so that {CV }V ∈ τx ⊂ F has the finite intersection property.6 Since F is Proof. Let {Kn }n=1 be the compact subsets of X constructed in Lemma
compact, it follows that there exists some 14.23. We may now apply Theorem 14.29 repeatedly to find a nested family of
\ subsequences
1 2 3
f∈ CV 6= ∅. {fm } ⊃ {gm } ⊃ {gm } ⊃ {gm } ⊃ ...
V ∈τx n ∞
such that the sequence {gm }m=1 ⊂ C(X) is uniformly convergent on Kn . Using
Since f is continuous, there exists V ∈ τx such that |f (x) − f (y)| < ε/3 for all Cantor’s trick, define the subsequence {hn } of {fm } by hn := gnn . Then {hn }
y ∈ V. Because f ∈ CV , there exists W ⊂ V such that kf − fW k < ε/3. We is uniformly convergent on Kl for each l ∈ N. Now if K ⊂ X is an arbitrary
now arrive at a contradiction; compact set, there exists l < ∞ such that K ⊂ Klo ⊂ Kl and therefore {hn } is
uniformly convergent on K as well.
ε ≤ |fW (x) − fW (xW )|
Proposition 14.31. Let Ω ⊂o Rd such that Ω̄ is compact and 0 ≤ α < β ≤ 1.
≤ |fW (x) − f (x)| + |f (x) − f (xW )| + |f (xW ) − fW (xW )|
Then the inclusion map i : C β (Ω) ,→ C α (Ω) is a compact operator. See Chapter
< ε/3 + ε/3 + ε/3 = ε. 9 and Lemma 9.9 for the notation being used here.
Let {un }∞ β
n=1 ⊂ C (Ω) such that kun kC β ≤ 1, i.e. kun k∞ ≤ 1 and
Exercise 14.12. Give an alternative proof of the implication, (⇐) , in Theorem |un (x) − un (y)| ≤ |x − y|β for all x, y ∈ Ω.
14.29 by showing every subsequence {fn : n ∈ N} ⊂ F has a convergence sub-
sequence. By the Arzela-Ascoli Theorem 14.29, there exists a subsequence of {ũn }∞
n=1 of
{un }∞
n=1 and u ∈ C o
(Ω̄) such that ũ n → u in C 0
. Since
2
Exercise 14.13. Suppose k ∈ C [0, 1] , R and for f ∈ C ([0, 1] , R) , let
|u(x) − u(y)| = lim |ũn (x) − ũn (y)| ≤ |x − y|β ,
n→∞
Z 1
Kf (x) := k (x, y) f (y) dy for all x ∈ [0, 1] .
0
u ∈ C β as well. Define gn := u − ũn ∈ C β , then
Show K is a compact operator on (C ([0, 1] , R) , k·k∞ ) . [gn ]β + kgn kC 0 = kgn kC β ≤ 2
The following result is a corollary of Lemma 14.23 and Theorem 14.29. and gn → 0 in C 0 . To finish the proof we must show that gn → 0 in C α . Given
δ > 0,
Corollary 14.30 (Locally Compact Ascoli-Arzela Theorem). Let (X, τ ) |gn (x) − gn (y)|
be a locally compact and σ – compact topological space and {fm } ⊂ C(X) be a [gn ]α = sup ≤ An + Bn
x6=y |x − y|α
pointwise bounded sequence of functions such that {fm |K } is equicontinuous for
any compact subset K ⊂ X. Then there exists a subsequence {mn } ⊂ {m} such where
∞
that {gn := fmn }n=1 ⊂ C(X) is a sequence which is uniformly convergent on
|gn (x) − gn (y)|

compact subsets of X. An = sup : x 6
= y and |x − y| ≤ δ
|x − y|α
6
If we are willing to use Net’s described in Appendix C below we could finish the |gn (x) − gn (y)| β−α
proof as follows. Since F is compact, the net {fV }V ∈τx ⊂ F has a cluster point = sup · |x − y| : x 6 = y and |x − y| ≤ δ
|x − y|β
f ∈ F ⊂ C(X). Choose a subnet {gα }α∈A of {fV }V ∈τX such that gα → f uniformly.
Then, since xV → x implies xVα → x, we may conclude from Eq. (14.5) that ≤ δ β−α · [gn ]β ≤ 2δ β−α
≤ |gα (x) − gα (xVα )| → |g(x) − g(x)| = 0 and

which is a contradiction. |gn (x) − gn (y)|
Bn = sup : |x − y| > δ ≤ 2δ −α kgn kC 0 → 0 as n → ∞.
|x − y|α

14.5 Tychonoff’s Theorem 131
Therefore,
lim sup [gn ]α ≤ lim sup An + lim sup Bn ≤ 2δ β−α + 0 → 0 as δ ↓ 0.

n→∞ n→∞ n→∞
This proposition generalizes to the following theorem which the reader is asked
to prove in Exercise 14.22 below.
Theorem 14.32. Let Ω be a precompact open subset of Rd , α, β ∈ [0, 1] and

k, j ∈ N0 . If j + β > k + α, then C j,β Ω̄ is compactly contained in C k,α Ω̄ .
14.5 Tychonoff ’s Theorem

The goal of this section is to show that arbitrary products of compact spaces is Fig. 14.4. Constructing the open set Ux .
still compact. Before going to the general case of an arbitrary number of factors
let us start with only two factors.
∪V0 = ∪x∈Λ ∪α∈Γx (Vα × Wα ) ⊃ ∪x∈Λ (Ux × Y ) = X × Y.
Proposition 14.33. Suppose that X and Y are non-empty compact topological
spaces, then X × Y is compact in the product topology.
The results of Exercises 14.23 and 13.28 prove Tychonoff’s Theorem for a
Proof. Let U be an open cover of X × Y. Then for each (x, y) ∈ X × Y countable product of compact metric spaces. We now state the general version
there exist U ∈ U such that (x, y) ∈ U. By definition of the product topology, of the theorem.
there also exist Vx ∈ τxX and Wy ∈ τyY such that Vx × Wy ⊂ U. Therefore
V := {Vx × Wy : (x, y) ∈ X × Y } is also an open cover of X × Y. We will now Theorem 14.34 (Tychonoff ’s Theorem). Q Let {Xα }α∈A be a collection of
show that V has a finite sub-cover, say V0 ⊂⊂ V. Assuming this is proved for non-empty compact spaces. Then X := XA = Xα is compact in the product
α∈A
the moment, this implies that U also has a finite subcover because each V ∈ V0 space topology. (Compare with Exercise 14.23 which covers the special case of
is contained in some UV ∈ U. So to complete the proof it suffices to show every a countable product of compact metric spaces.)
cover V of the form V = {Vα × Wα : α ∈ A} where Vα ⊂o X and Wα ⊂o Y has
a finite subcover. Given x ∈ X, let fx : Y → X × Y be the map fx (y) = (x, y) Proof. (The proof is taken from Loomis [14] which followed Bourbaki. Re-
and notice that fx is continuous since πX ◦ fx (y) = x and πY ◦ fx (y) = y are mark 14.35 below should help the reader understand the strategy of the proof
continuous maps. From this we conclude that {x} × Y = fx (Y ) is compact. to follow.) The proof requires a form of “induction” known as Zorn’s lemma
Similarly, it follows that X × {y} is compact for all yS∈ Y. Since V is a cover of which is equivalent to the axiom of choice, see Theorem B.7 of Appendix B
{x} × Y, there exist Γx ⊂⊂ A such that {x} × Y ⊂ (Vα × Wα ) without loss below.
α∈Γx For α ∈ A let πα denote the projection map from X to Xα . Suppose that F
of generalityTwe may assume that Γx is chosen so that x ∈ Vα for all α ∈ Γx . is a family of closed subsets of X which has the finite intersection property, see
Let Ux := Vα ⊂o X, and notice that Definition 14.3. By Proposition 14.4 the proof will be complete if we can show
α∈Γx
∩F 6= ∅.
The first step is to apply Zorn’s lemma to construct a maximal collection,
[ [
(Vα × Wα ) ⊃ (Ux × Wα ) = Ux × Y, (14.6)
α∈Γx α∈Γx
F0 , of (not necessarily closed) subsets of X
with the finite intersection
property
such that F ⊂ F0 . To do this, let Γ := G ⊂ 2X : F ⊂ G equipped with the
see Figure 14.4 below. Since {Ux }x∈X is now an open cover of X and X is partial order, G1 < G2 if G1 ⊂ G2 . If Φ is a linearly ordered subset of Γ, then
compact, there exists Λ ⊂⊂ X such that X = ∪x∈Λ Ux . The finite subcollection, G:= ∪Φ is an upper bound for Γ which still has the finite intersection property
V0 := {Vα × Wα : α ∈ ∪x∈Λ Γx }, of V is the desired finite subcover. Indeed using as the reader should check. So by Zorn’s lemma, Γ has a maximal element F0 .
Eq. (14.6), The maximal F0 has the following properties.

132 14 Compactness
1. F0 is closed under finite intersections. Indeed, if we let (F0 )f denote the

collection of all finite intersections of elements from F0 , then (F0 )f has
the finite intersection property and contains F0 . Since F0 is maximal, this
implies (F0 )f = F0 .
2. If B ⊂ X and B ∩ F 6= ∅ for all F ∈ F0 then B ∈ F0 . For if not F0 ∪
{B} would still satisfy the finite intersection property and would properly
contain F0 and this would violate the maximallity of F0 .
3. For each α ∈ A,
πa (F0 ) := {πα (F ) ⊂ Xα : F ∈ F0 }
n
has the finite intersection property. Indeed, if {Fi }i=1 ⊂ F0 , then
∩ni=1 πα (Fi ) ⊃ πα (∩ni=1 Fi ) 6= ∅.
Fig. 14.5. Here F = {F1 , F2 } where F1 and F2 are the two parabolic arcs and
Since Xα is compact, property 3. above along with Proposition 14.4 implies F1 ∩ F2 = {p, q}.
∩F ∈F0 πα (F ) 6= ∅. Since this true for each α ∈ A, using the axiom of choice,
there exists p ∈ X such that pα = πα (p) ∈ ∩F ∈F0 πα (F ) for all α ∈ A. The
proof will be completed
by showing
∩F 6= ∅ by showing p ∈ ∩F. 14.6 Banach – Alaoglu’s Theorem
Since C := ∩ F̄ : F ∈ F0 ⊂ ∩F, it suffices to show p ∈ C. Let U be an
open neighborhood of p in X. By the definition of the product topology (or item 14.6.1 Weak and Strong Topologies
2. of Proposition 13.25), there exists Λ ⊂⊂ A and open sets Uα ⊂ Xα for all
α ∈ Λ such that p ∈ ∩α∈Λ πα−1 (Uα ) ⊂ U. Since pα ∈ ∩F ∈F0 πα (F ) and pα ∈ Uα Definition 14.36. Let X and Y be be a normed vector spaces and L(X, Y ) the
for all α ∈ Λ, it follows that Uα ∩ πα (F ) 6= ∅ for all F ∈ F0 and all α ∈ Λ. This normed space of bounded linear transformations from X to Y.
then implies πα−1 (Uα ) ∩ F 6= ∅ for all F ∈ F0 and all α ∈ Λ. By property 2.7
above we concluded that πα−1 (Uα ) ∈ F0 for all α ∈ Λ and then by property 1. 1. The weak topology on X is the topology generated by X ∗ , i.e. the smallest
that ∩α∈Λ πα−1 (Uα ) ∈ F0 . In particular topology on X such that every element f ∈ X ∗ is continuous.
2. The weak-∗ topology on X ∗ is the topology generated by X, i.e. the smallest
6 F ∩ ∩α∈Λ πα−1 (Uα ) ⊂ F ∩ U for all F ∈ F0 topology on X ∗ such that the maps f ∈ X ∗ → f (x) ∈ C are continuous for

∅=
all x ∈ X.
which shows p ∈ F̄ for each F ∈ F0 , i.e. p ∈ C. 3. The strong operator topology on L(X, Y ) is the smallest topology such
that T ∈ L(X, Y ) −→ T x ∈ Y is continuous for all x ∈ X.
Remark 14.35. Consider the following simple example where X = [−1, 1] × 4. The weak operator topology on L(X, Y ) is the smallest topology such
[−1, 1] and F = {F1 , F2 } as in Figure 14.5. Notice that πi (F1 )∩πi (F2 ) = [−1, 1] that T ∈ L(X, Y ) −→ f (T x) ∈ C is continuous for all x ∈ X and f ∈ Y ∗ .
for each i and so gives no help in trying to find the ith – coordinate of one of the
two points in F1 ∩F2 . This is why it is necessary to introduce the collection F0 in Let us be a little more precise about the topologies described in the above
the proof of Theorem 14.34. In this case one might take F0 to be the collection definitions.
of all subsets F ⊂ X such that p ∈ F. We then have ∩F ∈F0 πi (F ) = {pi } ,
so the ith – coordinate of p may now be determined by observing the sets, 1. The weak topology has a neighborhood base at x0 ∈ X consisting of sets
{πi (F ) : F ∈ F0 } . of the form
N = ∩ni=1 {x ∈ X : |fi (x) − fi (x0 )| < ε}
7
Here is where we use that F0 is maximal among the collection of all, not just closed, where fi ∈ X ∗ and ε > 0.
sets having the finite intersection property and containing F . 2. The weak-∗ topology on X ∗ has a neighborhood base at f ∈ X ∗ consisting
of sets of the form

14.7 Weak Convergence in Hilbert Spaces 133
N := ∩ni=1 {g ∈ X ∗ : |f (xi ) − g(xi )| < ε} Proof. The routine check that ρ is a metric is left to the reader. Let τρ
be the topology on C ∗ induced by ρ. For any g ∈ X and n ∈ N, the map
where xi ∈ X and ε > 0. f ∈ X ∗ → (f (xn ) − g(xn )) ∈C is τw∗ continuous and since the sum in Eq.
3. The strong operator topology on L(X, Y ) has a neighborhood base at (14.7) is uniformly convergent for f ∈ C ∗ , it follows that f → ρ(f, g) is τC ∗ –
T ∈ X ∗ consisting of sets of the form continuous. This implies the open balls relative to ρ are contained in τC ∗ and
N := ∩ni=1 {S ∈ L (X, Y ) : kSxi − T xi k < ε} therefore τρ ⊂ τC ∗ . We now wish to prove τC ∗ ⊂ τρ . Since τC ∗ is the topology
generated by {x̂|C ∗ : x ∈ C} , it suffices to show x̂ is τρ – continuous for all
where xi ∈ X and ε > 0. ∞
x ∈ C. But given x ∈ C there exists a subsequence yk := xnk of {xn }n=1 such
4. The weak operator topology on L(X, Y ) has a neighborhood base at that such that x = limk→∞ yk . Since
T ∈ X ∗ consisting of sets of the form
sup |x̂(f ) − ŷk (f )| = sup |f (x − yk )| ≤ kx − yk k → 0 as k → ∞,
N := ∩ni=1 {S ∈ L (X, Y ) : |fi (Sxi − T xi )| < ε} f ∈C ∗ f ∈C ∗
where xi ∈ X, fi ∈ X ∗ and ε > 0. ŷk → x̂ uniformly on C ∗ and using ŷk is τρ – continuous for all k (as is easily
Theorem 14.37 (Alaoglu’s Theorem). If X is a normed space the unit ball checked) we learn x̂ is also τρ continuous. Hence τC ∗ = τ (x̂|C ∗ : x ∈ X) ⊂ τρ .
in X ∗ is weak - ∗ compact. (Also see Theorem 14.44 and Proposition 26.16.) The compactness assertion follows from Theorem 14.37. The compactness asser-
tion may also be verified directly using: 1) sequential compactness is equivalent
Proof. For all x ∈ X let Dx = {z ∈ C : |z| ≤ kxk}.
Q Then Dx ⊂ C is a to compactness for metric spaces and 2) a Cantor’s diagonalization argument
compact set and so by Tychonoff’s Theorem Ω := Dx is compact in the as in the proof of Theorem 14.44. (See Proposition 26.16 below.)
x∈X
product topology. If f ∈ C ∗ := {f ∈ X ∗ : kf k ≤ 1}, |f (x)| ≤ kf k kxk ≤ kxk
which implies that f (x) ∈ Dx for all x ∈ X, i.e. C ∗ ⊂ Ω. The topology on C ∗
inherited from the weak–∗ topology on X ∗ is the same as that relative topology 14.7 Weak Convergence in Hilbert Spaces
coming from the product topology on Ω. So to finish the proof it suffices to show ∞
C ∗ is a closed subset of the compact space Ω. To prove this let πx (f ) = f (x) Suppose H is an infinite dimensional Hilbert space and {xn }n=1 is an orthonor-
be the projection maps. Then mal subset of H. Then, by Eq. (8.1), kxn − xm k2 = 2 for all m 6= n and in
∞
particular, {xn }n=1 has no convergent subsequences. From this we conclude
C ∗ = {f ∈ Ω : f is linear} that C := {x ∈ H : kxk ≤ 1} , the closed unit ball in H, is not compact. To
= {f ∈ Ω : f (x + cy) − f (x) − cf (y) = 0 for all x, y ∈ X and c ∈ C} overcome this problems it is sometimes useful to introduce a weaker topology
\ \ on X having the property that C is compact.
= {f ∈ Ω : f (x + cy) − f (x) − cf (y) = 0}
x,y∈X c∈C Definition 14.39. Let (X, k·k) be a Banach space and X ∗ be its continuous
∞
−1 dual. The weak topology, τw , on X is the topology generated by X ∗ . If {xn }n=1 ⊂
\ \
= (πx+cy − πx − cπy ) ({0})
w
x,y∈X c∈C X is a sequence we will write xn → x as n → ∞ to mean that xn → x in the
weak topology.
which is closed because (πx+cy − πx − cπy ) : Ω → C is continuous.
Theorem 14.38 (Alaoglu’s Theorem for separable spaces). Suppose that Because τw = τ (X ∗ ) ⊂ τk·k := τ ({kx − ·k : x ∈ X}), it is harder for a
X is a separable Banach space, C ∗ := {f ∈ X ∗ : kf k ≤ 1} is the closed unit ball function f : X → F to be continuous in the τw – topology than in the norm
∞
in X ∗ and {xn }n=1 is an countable dense subset of C := {x ∈ X : kxk ≤ 1} . topology, τk·k . In particular if φ : X → F is a linear functional which is τw –
Then continuous, then φ is τk·k – continuous and hence φ ∈ X ∗ .
∞
X 1
ρ(f, g) := n
|f (xn ) − g(xn )| (14.7) Exercise 14.14. Show the vector space operations of X are continuous in the
n=1
2 weak topology, i.e. show:
defines a metric on C ∗ which is compatible with the weak topology on C ∗ , τC ∗ :=
1. (x, y) ∈ X × X → x + y ∈ X is (τw ⊗ τw , τw ) – continuous and
(τw∗ )C ∗ = {V ∩ C : V ∈ τw∗ } . Moreover (C ∗ , ρ) is a compact metric space.

134 14 Compactness
2. (λ, x) ∈ F × X → λx ∈ X is (τF ⊗ τw , τw ) – continuous. Proof. 1. =⇒ 2. This is a consequence of Theorem 8.15 and Proposition
w
14.40. 2. =⇒ 3. is trivial. 3. =⇒ 1. Let M := supn kxn k and H0 denote the
∞
Proposition 14.40. Let {xn }n=1 ⊂ X be a sequence, then xn → x ∈ X as algebraic span of β. Then for y ∈ H and z ∈ H0 ,
n → ∞ iff φ(x) = limn→∞ φ(xn ) for all φ ∈ X ∗ .
w
|hx − xn |yi| ≤ |hx − xn |zi| + |hx − xn |y − zi| ≤ |hx − xn |zi| + 2M ky − zk .
Proof. By definition of τw , we have xn → x ∈ X iff for all Γ ⊂⊂ X ∗ and
ε > 0 there exists an N ∈ N such that |φ(x) − φ(xn )| < ε for all n ≥ N and φ ∈ Passing to the limit in this equation implies lim supn→∞ |hx − xn |yi| ≤
Γ. This later condition is easily seen to be equivalent to φ(x) = limn→∞ φ(xn ) 2M ky − zk which shows lim supn→∞ |hx − xn |yi| = 0 since H0 is dense in H. To
for all φ ∈ X ∗ . prove the last assertion, let Γ ⊂⊂ β. Then by Bessel’s inequality (Proposition
The topological space (X, τw ) is still Hausdorff as follows from the Hahn 8.18), X X
2 2 2
Banach Theorem, see Theorem 25.6 below. For the moment we will concentrate |cy | = lim |hxn |yi| ≤ lim inf kxn k ≤ M 2 .
n→∞ n→∞
on the special case where X = H is a Hilbert space in which case H ∗ = y∈Γ y∈Γ
{φz := h·|zi : z ∈ H} , see Theorem 8.15. If x, y ∈ H and z := y − x 6= 0, then P 2
Since Γ ⊂⊂ β was arbitrary,
P we conclude that y∈β |cy | ≤ M < ∞ and hence
2 we may define x := y∈β cy y. By construction we have
0 < ε := kzk = φz (z) = φz (y) − φz (x).
Thus hx|yi = cy = lim hxn |yi for all y ∈ β

n→∞
Vx := {w ∈ H : |φz (x) − φz (w)| < ε/2} and w

and hence xn → x ∈ H as n → ∞ by what we have just proved.
Vy := {w ∈ H : |φz (y) − φz (w)| < ε/2} ∞
Theorem 14.43. Suppose {xn }n=1 is a bounded sequence in a Hilbert space,
∞
are disjoint sets from τw which contain x and y respectively. This shows that H. Then there exists a subsequence yk := xnk of {xn }n=1 and x ∈ X such that
w
(H, τw ) is a Hausdorff space. In particular, this shows that weak limits are yk → x as k → ∞.
unique if they exist.
Proof. This is a consequence of Proposition 14.42 and a Cantor’s diagonal-
∞
Remark 14.41. Suppose that H is an infinite dimensional Hilbert space {xn }n=1 ization argument which is left to the reader, see Exercise 8.12.
is an orthonormal subset of H. Then Bessel’s inequality (Proposition 8.18) im-
w w Theorem 14.44 (Alaoglu’s Theorem for Hilbert Spaces). Suppose that
plies xn → 0 ∈ H as n → ∞. This points out the fact that if xn → x ∈ H as
n → ∞, it is no longer necessarily true that kxk = limn→∞ kxn k . However we H is a separable Hilbert space, C := {x ∈ H : kxk ≤ 1} is the closed unit ball
∞
do always have kxk ≤ lim inf n→∞ kxn k because, in H and {en }n=1 is an orthonormal basis for H. Then
2 ∞
kxk = lim hxn |xi ≤ lim inf [kxn k kxk] = kxk lim inf kxn k .
X 1
n→∞ n→∞ n→∞ ρ(x, y) := n
|hx − y|en i| (14.8)
n=1
2
Proposition 14.42. Let H be a Hilbert space, β ⊂ H be an orthonormal ba-
∞ defines a metric on C which is compatible with the weak topology on C,
sis for H and {xn }n=1 ⊂ H be a bounded sequence, then the following are
equivalent: τC := (τw )C = {V ∩ C : V ∈ τw } . Moreover (C, ρ) is a compact metric space.
w
(This theorem will be extended to Banach spaces, see Theorems 14.37 and 14.38
1. xn → x ∈ H as n → ∞. below.)
2. hx|yi = limn→∞ hxn |yi for all y ∈ H.
3. hx|yi = limn→∞ hxn |yi for all y ∈ β. Proof. The routine check that ρ is a metric is left to the reader. Let τρ
be the topology on C induced by ρ. For any y ∈ H and n ∈ N, the map
P 2
Moreover, if cy := limn→∞ hxn |yi exists for all y ∈ β, then y∈β |cy | < ∞ x ∈ H → hx − y|en i = hx|en i − hy|en i is τw continuous and since the sum in
w
Eq. (14.8) is uniformly convergent for x, y ∈ C, it follows that x → ρ(x, y) is
P
and xn → x := y∈β cy y ∈ H as n → ∞.
τC – continuous. This implies the open balls relative to ρ are contained in τC

14.8 Exercises 135
and therefore τρ ⊂ τC . For the converse inclusion, let z ∈ H, x → φz (x) = hx|zi Now suppose, for the sake of contradiction, that fn does not converge uni-
PN
be an element of H ∗ , and for N ∈ N let zN := n=1 hz|en ien . Then φzN = formly to f. In this case there exists ε > 0 and a subsequence, gk := fnk of
PN F, such that kf − gk k∞ ≥ ε for all k. However, another application of the
n=1 hz|en iφen is ρ continuous, being a finite linear combination of the φen ∞
which are easily seen to be ρ – continuous. Because zN → z as N → ∞ it Arzela-Ascoli Theorem 14.29 shows there is a further subsquence, {gkl }l=1 of
∞
follows that {gk }k=1 such that liml→∞ kf − gkl k∞ = 0 which leads to the contradiction that
0 = ε > 0.
sup |φz (x) − φzN (x)| = kz − zN k → 0 as N → ∞.
x∈C Exercise 14.20. Let T ∈ (0, ∞) and F ⊂ C([0, T ]) be a family of functions
such that:
Therefore φz |C is ρ – continuous as well and hence τC = τ (φz |C : z ∈ H) ⊂
τρ . The last assertion follows directly from Theorem 14.43 and the fact that 1. f˙(t) exists for all t ∈ (0, T ) and f ∈ F.
sequential compactness is equivalent to compactness for metric spaces. 2. supf ∈F |f (0)| < ∞ and
˙
3. M := sup f ∈Fsup f (t) < ∞.
t∈(0,T )
14.8 Exercises Show F is precompact in the Banach space C([0, T ]) equipped with the
norm kf k∞ = supt∈[0,T ] |f (t)| .
Exercise 14.15. Prove Lemma 14.5.
Exercise 14.21 (Peano’s Existence Theorem). Suppose Z : R × Rd → Rd
Exercise 14.16. Let C be a closed proper subset of Rn and x ∈ Rn \ C. Show is a bounded continuous function. Then for each T < ∞8 there exists a solution
there exists a y ∈ C such that d(x, y) = dC (x). to the differential equation
Exercise 14.17. Let F = R in this problem and A ⊂ `2 (N) be defined by ẋ(t) = Z(t, x(t)) for − T < t < T with x(0) = x0 . (14.9)
A = {x ∈ `2 (N) : x(n) ≥ 1 + 1/n for some n ∈ N} Do this by filling in the following outline for the proof.
= ∪∞ 2
n=1 {x ∈ ` (N) : x(n) ≥ 1 + 1/n}.
1. Given ε > 0, show there exists a unique function xε ∈ C([−ε, ∞) → Rd )
such that xε (t) := x0 for −ε ≤ t ≤ 0 and
Show A is a closed subset of `2 (N) with the property that dA (0) = 1 while there Z t
is no y ∈ A such that d(0, y) = 1. (Remember that in general an infinite union xε (t) = x0 + Z(τ, xε (τ − ε))dτ for all t ≥ 0. (14.10)
of closed sets need not be closed.) 0
Exercise 14.18. Let p ∈ [1, ∞] and X be an infinite set. Show directly, without Here
using Theorem 14.15, the closed unit ball in `p (X) is not compact. Z t Z t Z t
Z(τ, xε (τ −ε))dτ = Z1 (τ, xε (τ − ε))dτ, . . . , Zd (τ, xε (τ − ε))dτ
0 0 0
14.8.1 Ascoli-Arzela Theorem Problems
where Z = (Z1 , . . . , Zd ) and the integrals are either the Lebesgue or the
Exercise 14.19. Let (X, τ ) be a compact topological space and F := Riemann integral since they are equal on continuous functions. Hint: For
∞
{fn }n=1 ⊂ C (X) is a sequence of functions which are equicontinuous and t ∈ [0, ε], it follows from Eq. (14.10) that
pointwise convergent. Show f (x) := limn→∞ fn (x) is continuous and that Z t
limn→∞ kf − fn k∞ = 0, i.e. fn → f uniformly as n → ∞. xε (t) = x0 + Z(τ, x0 )dτ.
0
Solution to Exercise (14.19). By the Arzela-Ascoli Theorem 14.29, there
Now that xε (t) is known for t ∈ [−ε, ε] it can be found by integration for
exists a subsequence, gk := fnk of F which is uniformly convergent to a function
t ∈ [−ε, 2ε]. The process can be repeated.
g ∈ C (X) . Since also gk → f pointwise by assumption, it follows that f = g ∈
8
C (X) . Using Corollary 14.30, we may in fact allow T = ∞.

∞
2. Then use Exercise 14.20 to show there exists {εk }k=1 ⊂ (0, ∞) such that
limk→∞ εk = 0 and xεk converges to some x ∈ C([0, T ]) with respect to the
sup-norm: kxk∞ = supt∈[0,T ] |x(t)|). Also show for this sequence that
lim sup |xεk (τ − εk ) − x (τ )| = 0.

k→∞ εk ≤τ ≤T
3. Pass to the limit (with justification) in Eq. (14.10) with ε replaced by εk

to show x satisfies
Z t
x(t) = x0 + Z(τ, x(τ ))dτ ∀ t ∈ [0, T ].
0
4. Conclude from this that ẋ(t) exists for t ∈ (0, T ) and that x solves Eq.
(14.9).
5. Apply what you have just proved to the ODE,
ẏ(t) = −Z(−t, y(t)) for 0 ≤ t < T with y(0) = x0 .
Then extend x(t) above to (−T, T ) by setting x(t) = y(−t) if t ∈ (−T, 0].
Show x so defined solves Eq. (14.9) for t ∈ (−T, T ).

Exercise 14.22. Prove Theorem 14.32. Hint: First prove C j,β Ω̄ @@
C j,α Ω̄ is compact if 0 ≤ α < β ≤ 1. Then use Lemma 14.18 repeatedly
to handle all of the other cases.
14.8.2 Tychonoff ’s Theorem Problem
Exercise 14.23 (Tychonoff ’s Theorem for Compact Metric Spaces).

Let us continue the Notation used in Exercise 6.12. Further assume that the
spaces Xn are compact for all n. Show, without using Theorem 14.34, (X, d) is
∞
compact. Hint: Either use Cantor’s method to show every sequence {xm }m=1 ⊂
X has a convergent subsequence or alternatively show (X, d) is complete and
totally bounded. (Compare with Example 14.10.)
15
Locally Compact Hausdorff Spaces
In this section X will always be a topological space with topology τ. We Proposition 15.5. Suppose that (X, τ ) is a Hausdorff space, K @@ X and
are now interested in restrictions on τ in order to insure there are “plenty” of x ∈ K c . Then there exists U, V ∈ τ such that U ∩ V = ∅, x ∈ U and K ⊂ V.
continuous functions. One such restriction is to assume τ = τd – is the topol- In particular K is closed. (So compact subsets of Hausdorff topological spaces
ogy induced from a metric on X. For example the results in Lemma 6.15 and are closed.) More generally if K and F are two disjoint compact subsets of X,
Theorem 7.4 above shows that metric spaces have lots of continuous functions. there exist disjoint open sets U, V ∈ τ such that K ⊂ V and F ⊂ U.
The main thrust of this section is to study locally compact (and σ – com-
pact) “Hausdorff” spaces as defined in Definitions 15.2 and 14.21. We will see Proof. Because X is Hausdorff, for all y ∈ K there exists Vy ∈ τy and
again that this class of topological spaces have an ample supply of continuous Uy ∈ τx such that Vy ∩ Uy = ∅. The cover {Vy }y∈K of K has a finite subcover,
functions. We will start out with the notion of a Hausdorff topology. The follow- {Vy }y∈Λ for some Λ ⊂⊂ K. Let V = ∪y∈Λ Vy and U = ∩y∈Λ Uy , then U, V ∈ τ
ing example shows a pathology which occurs when there are not enough open satisfy x ∈ U, K ⊂ V and U ∩ V = ∅. This shows that K c is open and hence
sets in a topology. that K is closed. Suppose that K and F are two disjoint compact subsets of X.
For each x ∈ F there exists disjoint open sets Ux and Vx such that K ⊂ Vx and
Example 15.1. As in Example 13.36, let x ∈ Ux . Since {Ux }x∈F is an open cover of F, there exists a finite subset Λ of F
such that F ⊂ U := ∪x∈Λ Ux . The proof is completed by defining V := ∩x∈Λ Vx .
X := {1, 2, 3} with τ := {X, ∅, {1, 2}, {2, 3}, {2}}.
Example 13.36 shows limits need not be unique in this space and moreover it Exercise 15.1. Show any finite set X admits exactly one Hausdorff topology
is easy to verify that the only continuous functions, f : Y → R, are necessarily τ.
constant.
Exercise 15.2. Let (X, τ ) and (Y, τY ) be topological spaces.
Definition 15.2 (Hausdorff Topology). A topological space, (X, τ ), is
1. Show τ is Hausdorff iff ∆ := {(x, x) : x ∈ X} is a closed set in X × X
Hausdorff if for each pair of distinct points, x, y ∈ X, there exists disjoint
equipped with the product topology τ ⊗ τ.
open neighborhoods, U and V of x and y respectively. (Metric spaces are typical
2. Suppose τ is Hausdorff and f, g : Y → X are continuous maps. If
examples of Hausdorff spaces.) Y
{f = g} = Y then f = g. Hint: make use of the map f × g : Y → X × X
Remark 15.3. When τ is Hausdorff the “pathologies” appearing in Example 15.1 defined by (f × g) (y) = (f (y), g(y)).
do not occur. Indeed if xn → x ∈ X and y ∈ X \ {x} we may choose V ∈ τx
Exercise 15.3. Give an example of a topological space which has a non-closed
and W ∈ τy such that V ∩ W = ∅. Then xn ∈ V a.a. implies xn ∈ / W for all
compact subset.
but a finite number of n and hence xn 9 y, so limits are unique.
Proposition 15.6. Suppose that X is a compact topological space, Y is a Haus-
Proposition 15.4.
Q Let (Xα , τα ) be Hausdorff topological spaces. Then the prod- dorff topological space, and f : X → Y is a continuous bijection then f is a
uct space XA = α∈A Xα equipped with the product topology is Hausdorff.
homeomorphism, i.e. f −1 : Y → X is continuous as well.
Proof. Let x, y ∈ XA be distinct points. Then there exists α ∈ A such that Proof. Since closed subsets of compact sets are compact, continuous images
πα (x) = xα 6= yα = πα (y). Since Xα is Hausdorff, there exists disjoint open of compact subsets are compact and compact subsets of Hausdorff spaces are
sets U, V ⊂ Xα such πα (x) ∈ U and πα (y) ∈ V. Then πα−1 (U ) and πα−1 (V ) are −1
closed, it follows that f −1 (C) = f (C) is closed in X for all closed subsets
disjoint open sets in XA containing x and y respectively.
C of X. Thus f −1 is continuous.
138 15 Locally Compact Hausdorff Spaces
The next two results show that locally compact Hausdorff spaces have plenty Definition 14.26) such that f = 1 on K. In particular, if K is compact and C is
of open sets and plenty of continuous functions. closed in X such that K ∩ C = ∅, there exists f ∈ Cc (X, [0, 1]) such that f = 1
on K and f = 0 on C.
Proposition 15.7. Suppose X is a locally compact Hausdorff space and U ⊂o
X and K @@ U. Then there exists V ⊂o X such that K ⊂ V ⊂ V ⊂ U ⊂ X Proof. For notational ease later it is more convenient to construct g := 1−f
and V̄ is compact. (Compare with Proposition 14.25 above.) rather than f. To motivate the proof, suppose g ∈ C(X, [0, 1]) such that g = 0
on K and g = 1 on U c . For r > 0, let Ur = {g < r} . Then for 0 < r < s ≤ 1,
Proof. By local compactness, for all x ∈ K, there exists Ux ∈ τx such that Ur ⊂ {g ≤ r} ⊂ Us and since {g ≤ r} is closed this implies
Ūx is compact. Since K is compact, there exists Λ ⊂⊂ K such that {Ux }x∈Λ is a
cover of K. The set O = U ∩ (∪x∈Λ Ux ) is an open set such that K ⊂ O ⊂ U and K ⊂ Ur ⊂ Ūr ⊂ {g ≤ r} ⊂ Us ⊂ U.
O is precompact since Ō is a closed subset of the compact set ∪x∈Λ Ūx . (∪x∈Λ Ūx .
is compact because it is a finite union of compact sets.) So by replacing U by Therefore associated to the function g is the collection open sets {Ur }r>0 ⊂ τ
O if necessary, we may assume that Ū is compact. Since Ū is compact and with the property that K ⊂ Ur ⊂ Ūr ⊂ Us ⊂ U for all 0 < r < s ≤ 1 and
bd(U ) = Ū ∩U c is a closed subset of Ū , bd(U ) is compact. Because bd(U ) ⊂ U c , Ur = X if r > 1. Finally let us notice that we may recover the function g from
it follows that bd(U ) ∩ K = ∅, so by Proposition 15.5, there exists disjoint open the sequence {Ur }r>0 by the formula
sets V and W such that K ⊂ V and bd(U ) ⊂ W. By replacing V by V ∩ U if
g(x) = inf{r > 0 : x ∈ Ur }. (15.1)
necessary we may further assume that K ⊂ V ⊂ U, see Figure 15.1. Because
The idea of the proof to follow is to turn these remarks around and define g by
Eq. (15.1).
Step 1. (Construction of the Ur .) Let
D := k2−n : k = 1, 2, . . . , 2−n , n = 1, 2, . . .

be the dyadic rationals in (0, 1]. Use Proposition 15.7 to find a precompact open
set U1 such that K ⊂ U1 ⊂ Ū1 ⊂ U. Apply Proposition 15.7 again to construct
an open set U1/2 such that
K ⊂ U1/2 ⊂ Ū1/2 ⊂ U1
and similarly use Proposition 15.7 to find open sets U1/2 , U3/4 ⊂o X such that
K ⊂ U1/4 ⊂ Ū1/4 ⊂ U1/2 ⊂ Ū1/2 ⊂ U3/4 ⊂ Ū3/4 ⊂ U1 .

Fig. 15.1. The construction of V.
Likewise there exists open set U1/8 , U3/8 , U5/8 , U7/8 such that
K ⊂ U1/8 ⊂ Ū1/8 ⊂ U1/4 ⊂ Ū1/4 ⊂ U3/8 ⊂ Ū3/8 ⊂ U1/2

Ū ∩ W c is a closed set containing V and bd(U ) ∩ W c = ∅, ⊂ Ū1/2 ⊂ U5/8 ⊂ Ū5/8 ⊂ U3/4 ⊂ Ū3/4 ⊂ U7/8 ⊂ Ū7/8 ⊂ U1 .
V̄ ⊂ Ū ∩ W c = (U ∪ bd(U )) ∩ W c = U ∩ W c ⊂ U ⊂ Ū . Continuing this way inductively, one shows there exists precompact open sets
{Ur }r∈D ⊂ τ such that
Since Ū is compact it follows that V̄ is compact and the proof is complete.
The following Lemma is analogous to Lemma 14.27. K ⊂ Ur ⊂ U r ⊂ Us ⊂ U1 ⊂ Ū1 ⊂ U
Lemma 15.8 (Urysohn’s Lemma for LCH Spaces). Let X be a locally for all r, s ∈ D with 0 < r < s ≤ 1.
compact Hausdorff space and K @@ U ⊂o X. Then there exists f ≺ U (see Step 2. Let Ur := X if r > 1 and define

15.1 Locally compact form of Urysohn’s Metrization Theorem 139
F |K = f. Moreover given c ∈ [a, b], F can be chosen so that supp(F − c) =

{F 6= c} ⊂ U.
The proof of this theorem is similar to Theorem 7.4 and will be left to the
reader, see Exercise 15.6.
15.1 Locally compact form of Urysohn’s Metrization

Theorem
Definition 15.10 (Polish spaces). A Polish space is a separable topological
space (X, τ ) which admits a complete metric, ρ, such that τ = τρ .
Notation 15.11 Let Q := [0, 1]N denote the (infinite dimensional) unit cube
in RN . For a, b ∈ Q let
Fig. 15.2. Determining g from {Ur } .
∞
X 1
d(a, b) := n
|an − bn | . (15.2)
n=1
2
g(x) = inf{r ∈ D ∪ (1, 2) : x ∈ Ur },
The metric introduced in Exercise 14.23 would be defined, in this context,
see Figure 15.2. Then g(x) ∈ [0, 1] for all x ∈ X, g(x) = 0 for x ∈ K since ˜ b) := P∞ 1n |an −bn | . Since 1 ≤ 1 + |an − bn | ≤ 2, it follows that d˜ ≤
as d(a, n=1 2 1+|an −bn |
x ∈ K ⊂ Ur for all r ∈ D. If x ∈ U1c , then x ∈ / Ur for all r ∈ D and hence d ≤ 2d. So the metrics d and d˜ are equivalent and in particular the topologies
g(x) = 1. Therefore f := 1 − g is a function such that f = 1 on K and induced by d and d˜ are the same. By Exercises 13.28, the d – topology on Q
{f 6= 0} = {g 6= 1} ⊂ U1 ⊂ Ū1 ⊂ U so that supp(f ) = {f 6= 0} ⊂ Ū1 ⊂ U is is the same as the product topology and by Tychonoff’s Theorem 14.34 or by
a compact subset of U. Thus it only remains to show f, or equivalently g, is Exercise 14.23, (Q, d) is a compact metric space.
continuous.
Since E = {(α, ∞), (−∞, α) : α ∈ R} generates the standard topology on R, Theorem 15.12. To every separable metric space (X, ρ), there exists a contin-
to prove g is continuous it suffices to show {g < α} and {g > α} are open sets uous injective map G : X → Q such that G : X → G(X) ⊂ Q is a homeomor-
for all α ∈ R. But g(x) < α iff there exists r ∈ D ∪ (1, ∞) with r < α such that phism. Moreover if the metric, ρ, is also complete, then G (X) is a Gδ –set, i.e.
x ∈ Ur . Therefore the G (X) is the countable intersection of open subsets of (Q, d) . In short, any
[ separable metrizable space X is homeomorphic to a subset of (Q, d) and if X is
{g < α} = {Ur : r ∈ D ∪ (1, ∞) 3 r < α} a Polish space then X is homeomorphic to a Gδ – subset of (Q, d).
which is open in X. If α ≥ 1, {g > α} = ∅ and if α < 0, {g > α} = X. If Proof. (See Rogers and Williams [19], Theorem 82.5 on p. 106.) By replacing
ρ ∞
α ∈ (0, 1), then g(x) > α iff there exists r ∈ D such that r > α and x ∈ / Ur . ρ by 1+ρ if necessary, we may assume that 0 ≤ ρ < 1. Let D = {an }n=1 be a
Now if r > α and x ∈ / Ur then for s ∈ D ∩ (α, r), x ∈
/ Ūs ⊂ Ur . Thus we have countable dense subset of X and define
shown that [n c o G (x) = (ρ (x, a1 ) , ρ (x, a2 ) , ρ (x, a3 ) , . . . ) ∈ Q
{g > α} = Us : s ∈ D 3 s > α
and
which is again an open subset of X. ∞
X 1
γ (x, y) = d (G (x) , G (y)) = n
|ρ (x, an ) − ρ (y, an )|
Theorem 15.9 (Locally Compact Tietz Extension Theorem). Let (X, τ ) n=1
2
be a locally compact Hausdorff space, K @@ U ⊂o X, f ∈ C(K, R), a = for x, y ∈ X. To prove the first assertion, we must show G is injective and γ is
min f (K) and b = max f (K). Then there exists F ∈ C(X, [a, b]) such that a metric on X which is compatible with the topology determined by ρ.

If G (x) = G (y) , then ρ (x, a) = ρ (y, a) for all a ∈ D. Since D is a dense is a metric ρ on X such that τ = τρ . Moreover, ρ may be chosen so that X is
subset of X, we may choose αk ∈ D such that isometric to a subset Q0 ⊂ Q equipped with the metric d in Eq. (15.2). In this
metric X is totally bounded and hence the completion of X (which is isometric
0 = lim ρ (x, αk ) = lim ρ (y, αk ) = ρ (y, x) to Q̄0 ⊂ Q) is compact. (Also see Theorem 15.43.)
k→∞ k→∞
and therefore x = y. A simple argument using the dominated convergence Proof. Let B be a countable base for τ and set
theorem shows y → γ (x, y) is ρ – continuous, i.e. γ (x, y) is small if ρ (x, y) is
Γ := {(U, V ) ∈ B × B | Ū ⊂ V and Ū is compact}.
small. Conversely,
To each O ∈ τ and x ∈ O there exist (U, V ) ∈ Γ such that x ∈ U ⊂ V ⊂ O.
ρ (x, y) ≤ ρ (x, an ) + ρ (y, an ) = 2ρ (x, an ) + ρ (y, an ) − ρ (x, an )
Indeed, since B is a base for τ, there exists V ∈ B such that x ∈ V ⊂ O.
≤ 2ρ (x, an ) + |ρ (x, an ) − ρ (y, an )| ≤ 2ρ (x, an ) + 2n γ (x, y) . Now apply Proposition 15.7 to find U 0 ⊂o X such that x ∈ U 0 ⊂ Ū 0 ⊂ V
with Ū 0 being compact. Since B is a base for τ, there exists U ∈ B such that
Hence if ε > 0 is given, we may choose n so that 2ρ (x, an ) < ε/2 and so if
x ∈ U ⊂ U 0 and since Ū ⊂ Ū 0 , Ū is compact so (U, V ) ∈ Γ. In particular this
γ (x, y) < 2−(n+1) ε, it will follow that ρ (x, y) < ε. This shows τγ = τρ . Since
shows that B 0 := {U ∈ B : (U, V ) ∈ Γ for some V ∈ B} is still a base for τ. If Γ
G : (X, γ) → (Q, d) is isometric, G is a homeomorphism.
is a finite, then B 0 is finite and τ only has a finite number of elements as well.
Now suppose that (X, ρ) is a complete metric space. Let S := G (X) and σ N
Since (X, τ ) is Hausdorff, it follows that X is a finite set. Letting {xn }n=1 be
be the metric on S defined by σ (G (x) , G (y)) = ρ (x, y) for all x, y ∈ X. Then
an enumeration of X, define T : X → Q by T (xn ) = en for n = 1, 2, . . . , N
(S, σ) is a complete metric (being the isometric image of a complete metric
where en = (0, 0, . . . , 0, 1, 0, . . . ), with the 1 occurring in the nth spot. Then
space) and by what we have just prove, τσ = τdS .Consequently, if u ∈ S and ε >
ρ(x, y) := d(T (x), T (y)) for x, y ∈ X is the desired metric.
0 is given, we may find δ 0 (ε) such that Bσ (u, δ 0 (ε)) ⊂ Bd (u, ε) . Taking δ (ε) = ∞
So we may now assume that Γ is an infinite set and let {(Un , Vn )}n=1 be an
min (δ 0 (ε) , ε) , we have diamd (Bd (u, δ (ε))) < ε and diamσ (Bd (u, δ (ε))) < ε
enumeration of Γ. By Urysohn’s Lemma 15.8 there exists fU,V ∈ C(X, [0, 1])
where
such that fU,V = 0 on Ū and fU,V = 1 on V c . Let F := {fU,V | (U, V ) ∈ Γ }
diamσ (A) := {sup σ (u, v) : u, v ∈ A} and and set fn := fUn ,Vn – an enumeration of F. We will now show that
diamd (A) := {sup d (u, v) : u, v ∈ A} . X∞
1
ρ(x, y) := n
|fn (x) − fn (y)|
Let S̄ denote the closure of S inside of (Q, d) and for each n ∈ N let n=1
2
Nn := {N ∈ τd : diamd (N ) ∨ diamσ (N ∩ S) < 1/n} is the desired metric on X. The proof will involve a number of steps.
and let Un := ∪Nn ∈ τd . From the previous paragraph, it follows that S ⊂ Un 1. (ρ is a metric on X.) It is routine to show ρ satisfies the triangle inequal-
and therefore S ⊂ S̄ ∩ (∩∞ ity and ρ is symmetric. If x, y ∈ X are distinct points then there exists
n=1 Un ) . c
Conversely if u ∈ S̄ ∩ (∩∞ (Un0 , Vn0 ) ∈ Γ such that x ∈ Un0 and Vn0 ⊂ O := {y} . Since fn0 (x) = 0
n=1 Un ) and n ∈ N, there exists Nn ∈ Nn such −n0
that u ∈ Nn . Moreover, since N1 ∩ · · · ∩ Nn is an open neigborhood of u ∈ S̄, and fn0 (y) = 1, it follows that ρ(x, y) ≥ 2 > 0.
there exists un ∈ N1 ∩ · · · ∩ Nn ∩ S for each n ∈ N. From the definition of 2. (Let τ0 = τ (fn : n ∈ N) , then τ = τ0 = τρ .) As usual we have τ0 ⊂ τ.
Nn , we have limn→∞ d (u, un ) = 0 and σ (un , um ) ≤ max n−1 , m−1 → 0 as
Since, for each x ∈ X, y → ρ(x, y) is τ0 – continuous (being the uni-
∞
m, n → ∞. Since (S, σ) is complete, it follows that {un }n=1 is convergent in formly convergent sum of continuous functions), it follows that Bx (ε) :=
(S, σ) to some element u0 ∈ S. Since (S, dS ) has the same topology as (S, σ) {y ∈ X : ρ(x, y) < ε} ∈ τ0 for all x ∈ X and ε > 0. Thus τρ ⊂ τ0 ⊂ τ.
it follows that d (un , u0 ) → 0 as well and thus that u = u0 ∈ S. We have Suppose that O ∈ τ and x ∈ O. Let (Un0 , Vn0 ) ∈ Γ be such that x ∈ Un0
now shown,TS = S̄ ∩(∩∞ and Vn0 ⊂ O. Then fn0 (x) = 0 and fn0 = 1 on Oc . Therefore if y ∈ X and
n=1 Un ) . This completes the proof because we may
write S̄ =
∞
S where S := u ∈ Q : d u, S̄ < 1/n

and therefore, fn0 (y) < 1, then y ∈ O so x ∈ {fn0 < 1} ⊂ O. This shows that O may be
n=1 T 1/n 1/n
T∞
S = ( n=1 Un ) ∩
∞ written as a union of elements from τ0 and therefore O ∈ τ0 . So τ ⊂ τ0 and
n=1 S1/n is a Gδ set.
hence τ = τ0 . Moreover, if y ∈ Bx (2−n0 ) then 2−n0 > ρ(x, y) ≥ 2−n0 fn0 (y)
Theorem 15.13 (Urysohn Metrization Theorem for LCH’s). Every sec- and therefore x ∈ Bx (2−n0 ) ⊂ {fn0 < 1} ⊂ O. This shows O is ρ – open
ond countable locally compact Hausdorff space, (X, τ ) , is metrizable, i.e. there and hence τρ ⊂ τ0 ⊂ τ ⊂ τρ .

15.2 Partitions of Unity 141
3. (X is isometric to some Q0 ⊂ Q.) Let T : X → Q be defined by T (x) = Then Fj is compact, Fj ⊂ Uj for all j, and K ⊂ ∪nj=1 Fj . By Urysohn’s Lemma
(f1 (x), f2 (x), . . . , fn (x), . . . ). Then T is an isometry by the very definitions 15.8 there exists fj ≺ Uj such that fj = 1 on Fj for j = 1, 2, . . . , n and by
of d and ρ and therefore X is isometric to Q0 := T (X). Since Q0 is a subset convention let fn+1 ≡ 1. We will now give two methods to finish the proof.
of the compact metric space (Q, d), Q0 is totally bounded and therefore X Method 1. Let h1 = f1 , h2 = f2 (1 − h1 ) = f2 (1 − f1 ),
is totally bounded.
h3 = f3 (1 − h1 − h2 ) = f3 (1 − f1 − (1 − f1 )f2 ) = f3 (1 − f1 )(1 − f2 )
BRUCE: Add Stone Chech Compactification results. and continue on inductively to define
k−1
Y
hk = (1 − h1 − · · · − hk−1 )fk = fk · (1 − fj ) ∀ k = 2, 3, . . . , n (15.3)
15.2 Partitions of Unity j=1
Definition 15.14. Let (X, τ ) be a topological space and X0 ⊂ X be a set. A and to show
collection of sets {Bα }α∈A ⊂ 2X is locally finite on X0 if for all x ∈ X0 , there n
is an open neighborhood Nx ∈ τ of x such that #{α ∈ A : Bα ∩ Nx 6= ∅} < ∞.
Y
hn+1 = (1 − h1 − · · · − hn ) · 1 = 1 · (1 − fj ). (15.4)
j=1
Definition 15.15. Suppose that U is an open cover of X0 ⊂ X. A collection
{φα }α∈A ⊂ C(X, [0, 1]) (N = ∞ is allowed here) is a partition of unity on From these equations it clearly follows that hj Q ∈ Cc (X, [0, 1]) and that
X0 subordinate to the cover U if: n
supp(hj ) ⊂ supp(fj ) ⊂ Uj , i.e. hj ≺ Uj . Since j=1 (1 − fj ) = 0 on K,
Pn n
1. for all α there is a U ∈ U such that supp(φα ) ⊂ U, j=1 hj = 1 on K and {hj }j=1 is the desired partition of unity.
n
the collection of sets, {supp(φα )}α∈A , is locally finite on X, and
2. P Method 2. Let g :=
P
fj ∈ Cc (X). Then g ≥ 1 on K and hence K ⊂
3. α∈A φα = 1 on X0 . j=1
{g > 21 }. Choose φ ∈ Cc (X, [0, 1]) such that φ = 1 on K and supp(φ) ⊂ {g > 21 }
Notice by item 2. that, for each x ∈ X, there is a neighborhood Nx such that and define f0 := 1 − φ. Then f0 = 0 on K, f0 = 1 if g ≤ 12 and therefore,
Λ := {α ∈ A : supp(φα ) ∩ Nx 6= ∅} f0 + f1 + · · · + fn = f0 + g > 0
P P
is
P a finite set. Therefore, α∈A φα |Nx = α∈Λ φα |Nx which shows the sum on X. The desired partition of unity may be constructed as
α∈A φα is well defined and defines a continuous function on Nx and there-
fore on X since continuity is P a local property. We will summarize these last fj (x)
hj (x) = .
comments by saying the sum, α∈A φα , is locally finite. f0 (x) + · · · + fn (x)
Proposition 15.16 (Partitions of Unity: The Compact Case). Suppose Indeed supp (hj ) = supp (fj ) ⊂ Uj , hj ∈ Cc (X, [0, 1]) and on K,
that X is a locally compact Hausdorff space, K ⊂ X is a compact set and
n
U = {Uj }j=1 is an open cover of K. Then there exists a partition of unity f1 + · · · + fn f1 + · · · + fn
h1 + · · · + hn = = = 1.
n
{hj }j=1 of K such that hj ≺ Uj for all j = 1, 2, . . . , n. f0 + f1 + · · · + fn f1 + · · · + fn
Proof. For all x ∈ K choose a precompact open neighborhood, Vx , of x

such that V x ⊂ S
Uj . Since K is compact, there exists a finite subset, Λ, of K Proposition 15.17. Let (X, τ ) be a locally compact and σ – compact Hausdorff
such that K ⊂ Vx . Let space. Suppose that U ⊂ τ is an open cover of X. Then we may construct two
x∈Λ
locally finite open covers V = {Vi }N N
i=1 and W = {Wi }i=1 of X (N = ∞ is
allowed here) such that:
Fj = ∪ V̄x : x ∈ Λ and V x ⊂ Uj .
1. Wi ⊂ W̄i ⊂ Vi ⊂ V̄i and V̄i is compact for all i.

2. For each i there exist U ∈ U such that V̄i ⊂ U. Proof. Let V = {Vi }N N
i=1 and W = {Wi }i=1 be open covers of X with the
properties described in Proposition 15.17. By Urysohn’s Lemma 15.8, there
Proof. By Remark 14.24, there exists an open cover of G = {Gn }∞ n=1 of X exists fi ≺ Vi such that fi = 1 on W̄i for each i. As in the proof of Proposition
such that Gn ⊂ Ḡn ⊂ Gn+1 . Then X = ∪∞ k=1 (Ḡ k \ Ḡ k−1 ), where by convention 15.16 there are two methods to finish the proof.
G−1 = G0 = ∅. For the moment fix k ≥ 1. For each x ∈ Ḡk \Gk−1 , let Ux ∈ U be Method 1. Define h1 = f1 , hj by Eq. (15.3) for all other j. Then as in Eq.
chosen so that x ∈ Ux and by Proposition 15.7 choose an open neighborhood (15.4), for all n < N + 1,
Nx of x such that N̄x ⊂ Ux ∩ (Gk+1 \ Ḡk−2 ), see Figure 15.3 below. Since  
X∞ Yn
1− hj = lim fn (1 − fj ) = 0
n→∞
j=1 j=1
since for x ∈ X, fj (x) = 1 for some j. As in the proof of Proposition 15.16, it

N
is easily checked that {hi }i=1 is the desired partition of unity.
PN
Method 2. Let f := i=1 fi , a locally finite sum, so that f ∈ C(X). Since
∞
{Wi }i=1 is a cover of X, f ≥ 1 on X so that 1/f ∈ C (X)) as well. The functions
hi := fi /f for i = 1, 2, . . . , N give the desired partition of unity.
Lemma 15.19. Let (X, τ ) be a locally compact Hausdorff space.

1. A subset E ⊂ X is closed iff E ∩ K is closed for all K @@ X.
2. Let {Cα }α∈A be a locally finite collection of closed subsets of X, then C =
∪α∈A Cα is closed in X. (Recall that in general closed sets are only closed
under finite unions.)
Fig. 15.3. Constructing the {Wi }N Proof. 1. Since compact subsets of Hausdorff spaces are closed, E ∩ K is
i=1 .
closed if E is closed and K is compact. Now suppose that E ∩ K is closed for
all compact subsets K ⊂ X and let x ∈ E c . Since X is locally compact, there
1
{Nx }x∈Ḡk \Gk−1 is an open cover of the compact set Ḡk \ Gk−1 , there exist a c open neighborhood, V, of x. By assumption E ∩ V̄ is closed
exists a precompact
so x ∈ E ∩ V̄ – an open subset of X. By Proposition 15.7 there exists an
finite subset Γk ⊂ {Nx }x∈Ḡk \Gk−1 which also covers Ḡk \ Gk−1 . c
open set U such that x ∈ U ⊂ Ū ⊂ E ∩ V̄ , see Figure 15.4. Let W := U ∩ V.
By construction, for each W ∈ Γk , there is a U ∈ U such that W̄ ⊂ U ∩ Since
(Gk+1 \ Ḡk−2 ) and by another application of Proposition 15.7, there exists an W ∩ E = U ∩ V ∩ E ⊂ U ∩ V̄ ∩ E = ∅,
open set VW such that W̄ ⊂ VW ⊂ V̄W ⊂ U ∩ (Gk+1 \ Ḡk−2 ). We now choose
and enumeration {Wi }N ∞ and W is an open neighborhood of x and x ∈ E c was arbitrary, we have shown
i=1 of the countable open cover, ∪k=1 Γk , of X and define
N N
Vi = VWi . Then the collection {Wi }i=1 and {Vi }i=1 are easily checked to satisfy E c is open hence E is closed.
all the conclusions of the proposition. In particular notice that for each k; 2. Let K be a compact subset of X and for each x ∈ K let Nx be an
Vi ∩ Gk 6= ∅ for only a finite number of i’s. open neighborhood of x such that #{α ∈ A : Cα ∩ Nx 6= ∅} < ∞. Since K is
compact, there exists a finite subset Λ ⊂ K such that K ⊂ ∪x∈Λ Nx . Letting
Theorem 15.18 (Partitions of Unity for σ – Compact LCH Spaces). Λ0 := {α ∈ A : Cα ∩ K 6= ∅}, then
Let (X, τ ) be locally compact, σ – compact and Hausdorff and let U ⊂ τ be an 1
If X were a metric space we could finish the proof as follows. If there does not
open cover of X. Then there exists a partition of unity of {hi }N
i=1 (N = ∞ is exist an open neighborhood of x which is disjoint from E, then there would exists
allowed here) subordinate to the cover U such that supp(hi ) is compact for all xn ∈ E such that xn → x. Since E ∩ V̄ is closed and xn ∈ E ∩ V̄ for all large n,
i. it follows (see Exercise 6.4) that x ∈ E ∩ V̄ and in particular that x ∈ E. But we
chose x ∈ E c .

15.3 C0 (X) and the Alexanderov Compactification 143
By Item 2. of Lemma 15.19, ∪j∈Γk supp(gj ) is closed and therefore,
supp(hk ) = {hk 6= 0} = ∪j∈Γk {gj 6= 0} ⊂ ∪j∈Γk supp(gj ) ⊂ Uk

PN
and hence hk ≺ Uk and the sum k=1 hk is still locally finite. (Why?) The
desired partition of unity is now formed by letting hαk := hk for k < N + 1 and
/ {αi }N
hα ≡ 0 if α ∈ i=1 .
Corollary 15.21. Let (X, τ ) be a locally compact and σ – compact Hausdorff

space and A, B be disjoint closed subsets of X. Then there exists f ∈ C(X, [0, 1])
such that f = 1 on A and f = 0 on B. In fact f can be chosen so that
supp(f ) ⊂ B c .
Fig. 15.4. Showing E c is open.
Proof. Let U1 = Ac and U2 = B c , then {U1 , U2 } is an open cover of X.
By Corollary 15.20 there exists h1 , h2 ∈ C(X, [0, 1]) such that supp(hi ) ⊂ Ui
X for i = 1, 2 and h1 + h2 = 1 on X. The function f = h2 satisfies the desired
# (Λ0 ) ≤ #{α ∈ A : Cα ∩ Nx 6= ∅} < ∞
properties.
x∈Λ
and hence K ∩ (∪α∈A Cα ) = K ∩ (∪α∈Λ0 Cα ) . The set (∪α∈Λ0 Cα ) is a finite

union of closed sets and hence closed. Therefore, K ∩ (∪α∈A Cα ) is closed and 15.3 C0 (X) and the Alexanderov Compactification
by item 1. it follows that ∪α∈A Cα is closed as well.
Definition 15.22. Let (X, τ ) be a topological space. A continuous function f :
Corollary 15.20. Let (X, τ ) be a locally compact and σ – compact Hausdorff X → C is said to vanish at infinity if {|f | ≥ ε} is compact in X for all
space and U = {Uα }α∈A ⊂ τ be an open cover of X. Then there exists a partition ε > 0. The functions, f ∈ C(X), vanishing at infinity will be denoted by C0 (X).
of unity of {hα }α∈A subordinate to the cover U such that supp(hα ) ⊂ Uα for (Notice that C0 (X) = C (X) whenever X is compact.)
all α ∈ A. (Notice that we do not assert that hα has compact support. However
if Ūα is compact then supp(hα ) will be compact.) Proposition 15.23. Let X be a topological space, BC(X) be the space of
bounded continuous functions on X with the supremum norm topology. Then
Proof. By the σ – compactness of X, we may choose a countable sub-
N 1. C0 (X) is a closed subspace of BC(X).
set, {αi }N
i=1 (N = ∞ allowed here), of A such that {Ui := Uαi }i=1 is still
∞ 2 2. If we further assume that X is a locally compact Hausdorff space, then
an open cover of X. Let {gj }j=1 be a partition of unity subordinate to the
cover {Ui }N C0 (X) = Cc (X).
as inTheorem 15.18. Define Γ̃k := {j : supp(gj ) ⊂ Uk } and
i=1
Γk := Γ̃k \ ∪k−1
j=1 Γ̃k , where by convention Γ̃0 = ∅. Then Proof.
∞ ∞ 1. If f ∈ C0 (X), K1 := {|f | ≥ 1} is a compact subset of X and therefore f (K1 )
is a compact and hence bounded subset of C and so M := supx∈K1 |f (x)| <
[ a
N= Γ̃k = Γk .
k=1 k=1 ∞. Therefore kf k∞ ≤ M ∨ 1 < ∞ showing f ∈ BC(X). Now suppose
P fn ∈ C0 (X) and fn → f in BC(X). Let ε > 0 be given and choose n
If Γk = ∅ let hk := 0 otherwise let hk := j∈Γk gj , a locally finite sum. Then sufficiently large so that kf − fn k∞ ≤ ε/2. Since
N ∞
X X |f | ≤ |fn | + |f − fn | ≤ |fn | + kf − fn k∞ ≤ |fn | + ε/2,
hk = gj = 1.
k=1 j=1 {|f | ≥ ε} ⊂ {|fn | + ε/2 ≥ ε} = {|fn | ≥ ε/2} .
2
So as to simplify the indexing we assume there countable number of gj ’s. This can Because {|f | ≥ ε} is a closed subset of the compact set {|fn | ≥ ε/2} ,
always be arranged by taking gk ≡ 0 for large k if necessary. {|f | ≥ ε} is compact and we have shown f ∈ C0 (X).

2. Since C0 (X) is a closed subspace of BC(X) and Cc (X) ⊂ C0 (X), we always on X. Moreover, for all ε > 0 there exists an open neighborhood V ∈ τ ∗ of ∞
have Cc (X) ⊂ C0 (X). Now suppose that f ∈ C0 (X) and let Kn := {|f | ≥ such that
1
n } @@ X. By Lemma 15.8 we may choose φn ∈ Cc (X, [0, 1]) such that |g(x)| = |f (x) − f (∞)| < ε for all x ∈ V.
φn ≡ 1 on Kn . Define fn := φn f ∈ Cc (X). Then Since V is an open neighborhood of ∞, there exists a compact subset,
1 K ⊂ X, such that V = X ∗ \ K. By the previous equation we see that
kf − fn ku = k(1 − φn )f k∞ ≤ → 0 as n → ∞. {x ∈ X : |g(x)| ≥ ε} ⊂ K, so {|g| ≥ ε} is compact and we have shown g vanishes
n
at ∞.
This shows that f ∈ Cc (X). Conversely if g ∈ C0 (X), extend g to X ∗ by setting g(∞) = 0. Given
ε > 0, the set K = {|g| ≥ ε} is compact, hence X ∗ \ K is open in X ∗ . Since
g(X ∗ \ K) ⊂ (−ε, ε) we have shown that g is continuous at ∞. Since g is also
Proposition 15.24 (Alexanderov Compactification). Suppose that (X, τ ) continuous at all points in X it follows that g is continuous on X ∗ . Now it
is a non-compact locally compact Hausdorff space. Let X ∗ = X ∪ {∞} , where f = g + c with c ∈ C and g ∈ C0 (X), it follows by what we just proved that
{∞} is a new symbol not in X. The collection of sets, defining f (∞) = c extends f to a continuous function on X ∗ .
∗ Example 15.25. Let X be an uncountable set and τ be the discrete topology
τ ∗ = τ ∪ {X ∗ \ K : K @@ X} ⊂ 2X , on X. Let (X ∗ = X ∪ {∞} , τ ∗ ) be the one point compactification of X. The
smallest dense subset of X ∗ is the uncountable set X. Hence X ∗ is a compact
is a topology on X ∗ and (X ∗ , τ ∗ ) is a compact Hausdorff space. Moreover f ∈
but non-separable and hence non-metrizable space.
C(X) extends continuously to X ∗ iff f = g + c with g ∈ C0 (X) and c ∈ C in
R
which case the extension is given by f (∞) = c. Exercise 15.4. Let X := {0, 1} and τ be the product topology on X where
{0, 1} is equipped with the discrete topology. Show (X, τ ) is separable. (Com-
Proof. 1. (τ ∗ is a topology.) Let F := {F ⊂ X ∗ : X ∗ \ F ∈ τ ∗ }, i.e. F ∈ F bining this with Exercise 13.9 and Tychonoff’s Theorem 14.34, we see that
iff F is a compact subset of X or F = F0 ∪ {∞} with F0 being a closed subset of (X, τ ) is compact and separable but not first countable.)
X. Since the finite union of compact (closed) subsets is compact (closed), it is
easily seen that F is closed under finite unions. Because arbitrary intersections Solution to Exercise (15.4). We begin by observing that a basic open neigh-
of closed subsets of X are closed and closed subsets of compact subsets of X are borhood of g ∈ X is of the form
compact, it is also easily checked that F is closed under arbitrary intersections.
VΛ := {f ∈ X : f = g on Λ}
Therefore F satisfies the axioms of the closed subsets associated to a topology
and hence τ ∗ is a topology. where Λ ⊂⊂ R. Therefore to see that X is separable, we must find a countable
2. ((X ∗ , τ ∗ ) is a Hausdorff space.) It suffices to show any point x ∈ X can be set D ⊂ X such that for any g ∈ X (g : R → {0, 1}) and any Λ ⊂⊂ R, there
separated from ∞. To do this use Proposition 15.7 to find an open precompact exists f ∈ D such that f = g on Λ.
neighborhood, U, of x. Then U and V := X ∗ \ Ū are disjoint open subsets of Kevin Costello’s construction. Let
X ∗ such that x ∈ U and ∞ ∈ V.
3. ((X ∗ , τ ∗ ) is compact.) Suppose that U ⊂ τ ∗ is an open cover of X ∗ . Mm,k := 1[k/m,(k+1)/m)
Since U covers ∞, there exists a compact set K ⊂ X such that X ∗ \ K ∈ U. be the characteristic function of the interval [k/m, (k + 1)/m) and let D ⊂
Clearly X is covered by U0 := {V \ {∞} : V ∈ U} and by the definition of τ ∗ R
{0, 1} be the set of all finite sums of Mm,k which still have range in {0, 1}, i.e.
(or using (X ∗ , τ ∗ ) is Hausdorff), U0 is an open cover of X. In particular U0 is
the set of sums over disjoint intervals.
an open cover of K and since K is compact there exists Λ ⊂⊂ U such that R
Now suppose g ∈ {0, 1} and Λ ⊂⊂ R. Let
K ⊂ ∪ {V \ {∞} : V ∈ Λ} . It is now easily checked that Λ ∪ {X ∗ \ K} ⊂ U is
a finite subcover of X ∗ . S := {x ∈ Λ : g (x) = 0} and T = {x ∈ Λ : g (x) = 1} .
4. (Continuous functions on C(X ∗ ) statements.) Let i : X → X ∗ be the `
inclusion map. Then i is continuous and open, i.e. i(V ) is open in X ∗ for all V Then Λ = S T and we may take intervals Jt := [k/m, (k + 1)/m) 3 t for each
open in X. If f ∈ C(X ∗ ), then g = f |X − f (∞) = f ◦ i − f (∞) is continuous t ∈ T which
P are small enough to be disjoint and not contain any points in S.
Then f = t∈T 1Jt ∈ D and f = g on Λ showing f ∈ VΛ .

15.4 Stone-Weierstrass Theorem 145
The next proposition gathers a number of results involving countability lim sup ||t| − pn (t)| = 0.
n→∞ |t|≤M
assumptions which have appeared in the exercises.
Proposition 15.26 (Summary). Let (X, τ ) be a topological space. By replacing pn by pn − pn (0) if necessary we may assume that pn (0) = 0. Since
A is an algebra, it follows that fn = pn (f ) ∈ A and |f | ∈ A, because |f | is the
1. If (X, τ ) is second countable, then (X, τ ) is separable; see Exercise 13.11. uniform limit of the fn ’s. Since
2. If (X, τ ) is separable and metrizable then (X, τ ) is second countable; see
Exercise 13.12. 1
f ∨g = (f + g + |f − g|) and
3. If (X, τ ) is locally compact and metrizable then (X, τ ) is σ – compact iff 2
(X, τ ) is separable; see Exercises 14.10 and 14.11. 1
f ∧ g = (f + g − |f − g|),
4. If (X, τ ) is locally compact and second countable, then (X, τ ) is σ - compact, 2
see Exercise 14.8. we have shown A is a lattice.
5. If (X, τ ) is locally compact and metrizable, then (X, τ ) is σ – compact iff
(X, τ ) is separable, see Exercises 14.9 and 14.10. Lemma 15.30. Let A ⊂ C(X, R) be an algebra which separates points and
6. There exists spaces, (X, τ ) , which are both compact and separable but not suppose x and y are distinct points of X. If there exits such that f, g ∈ A such
first countable and in particular not metrizable, see Exercise 15.4. that
f (x) 6= 0 and g(y) 6= 0, (15.5)
then
15.4 Stone-Weierstrass Theorem V := {(f (x), f (y)) : f ∈ A}= R2 . (15.6)
We now wish to generalize Theorem 10.35 to more general topological spaces. Proof. It is clear that V is a non-zero subspace of R2. If dim(V ) = 1, then
We will first need some definitions. V = span(a, b) for some (a, b) ∈ R2 which, necessarily by Eq. (15.5), satisfy
a 6= 0 6= b. Since (a, b) = (f (x), f (y)) for some f ∈ A and f 2 ∈ A, it follows
Definition 15.27. Let X be a topological space and A ⊂ C(X) = C(X, R) or
that (a2 , b2 ) = (f 2 (x), f 2 (y)) ∈ V as well. Since dim V = 1, (a, b) and (a2 , b2 )
C(X, C) be a collection of functions. Then
are linearly dependent and therefore
1. A is said to separate points if for all distinct points x, y ∈ X there exists
a b
f ∈ A such that f (x) 6= f (y). 0 = det = ab2 − a2 b = ab(b − a)
a2 b2
2. A is an algebra if A is a vector subspace of C(X) which is closed under
pointwise multiplication. (Note well: we do not assume 1 ∈ A.) which implies that a = b. But this the implies that f (x) = f (y) for all f ∈ A,
3. A ⊂ C(X, R) is called a lattice if f ∨g := max(f, g) and f ∧g = min(f, g) ∈ violating the assumption that A separates points. Therefore we conclude that
A for all f, g ∈ A. dim(V ) = 2, i.e. V = R2 .
4. A ⊂ C(X, C) is closed under conjugation if f¯ ∈ A whenever f ∈ A.
Theorem 15.31 (Stone-Weierstrass Theorem). Suppose X is a locally
Remark 15.28. If X is a topological space such that C(X, R) separates points compact Hausdorff space and A ⊂ C0 (X, R) is a closed subalgebra which sepa-
then X is Hausdorff. Indeed if x, y ∈ X and f ∈ C(X, R) such that f (x) 6= f (y), rates points. For x ∈ X let
then f −1 (J) and f −1 (I) are disjoint open sets containing x and y respectively
when I and J are disjoint intervals containing f (x) and f (y) respectively. Ax := {f (x) : f ∈ A} and
Lemma 15.29. If A is a closed sub-algebra of BC(X, R) then |f | ∈ A for all Ix = {f ∈ C0 (X, R) : f (x) = 0}.
f ∈ A and A is a lattice.
Then either one of the following two cases hold.
Proof. Let f ∈ A and let M = sup |f (x)| . Using Theorem 10.35 or Exercise
x∈X 1. A = C0 (X, R) or
15.12, there are polynomials pn (t) such that 2. there exists a unique point x0 ∈ X such that A = Ix0 .

Moreover, case 1. holds iff Ax = R for all x ∈ X and case 2. holds iff there
exists a point x0 ∈ X such that Ax0 = {0} .
Proof. If there exists x0 such that Ax0 = {0} (x0 is unique since A separates
points) then A ⊂ Ix0 . If such an x0 exists let C = Ix0 and if Ax = R for all
x, set C = C0 (X, R). Let f ∈ C be given. By Lemma 15.30, for all x, y ∈ X
such that x 6= y, there exists gxy ∈ A such that f = gxy on {x, y}.3 When X is
compact the basic idea of the proof is contained in the following identity,
f (z) = inf sup gxy (z) for all z ∈ X. (15.7)
x∈X y∈X
Fig. 15.5. Constructing the “dominating approximates,” gx for each x ∈ X.

To prove this identity, let gx := supy∈X gxy and notice that gx ≥ f since
gxy (y) = f (y) for all y ∈ X. Moreover, gx (x) = f (x) for all x ∈ X since
gxy (x) = f (x) for all x. Therefore,
To complete the proof of the theorem, let g∞ be a fixed element of A such
inf sup gxy = inf gx = f. that f < g∞ + ε on X; for example let g∞ = gx0 ∈ A for some fixed x0 ∈ X.
x∈X y∈X x∈X
For each x ∈ X, let Ux be a neighborhood of x such that |f − gx | < ε on Ux .
The rest of the proof is devoted to replacing the inf and the sup above by Choose n εo n εo
min and max over finite sets at the expense of Eq. (15.7) becoming only an Γ ⊂⊂ F := |f | ≥ ∪ |g∞ | ≥
approximate identity. We also have to modify Eq. (15.7) slightly to take care 2 2
S
of the non-compact case. such that F ⊂ Ux (Γ exists since F is compact) and define
x∈Γ
Claim. Given ε > 0 and x ∈ X there exists gx ∈ A such that gx (x) = f (x) and
f < gx + ε on X. g = min{gx : x ∈ Γ ∪ {∞}} ∈ A.
To prove this, let Vy be an open neighborhood of y such that |f − gxy | < ε

S
Then, for x ∈ F, gx < f + ε on Ux and hence g < f + ε on Ux ⊃ F. Likewise,
on Vy ; in particular f < ε + gxy on Vy . Also let gx,∞ be any fixed element in A x∈Γ
such that gx,∞ (x) = f (x) and let
g ≤ g∞ < ε/2 < f + ε on F c .
n εo n εo
K = |f | ≥ ∪ |gx,∞ | ≥ . (15.8)
2 2 Therefore we have now shown,
S
Since K is compact, there exists Λ ⊂⊂ K such that K ⊂ Vy . Define
y∈Λ f < g + ε and g < f + ε on X,
gx (z) = max{gxy : y ∈ Λ ∪ {∞}}. i.e. |f − g| < ε on X. Since ε > 0 is arbitrary it follows that f ∈ Ā = A and so
A = C.
Since
f < ε + gxy < ε + gx on Vy , Corollary 15.32 (Complex Stone-Weierstrass Theorem). Let X be a lo-
for any y ∈ Λ, and cally compact Hausdorff space. Suppose A ⊂ C0 (X, C) is closed in the uniform
ε topology, separates points, and is closed under complex conjugation. Then either
f< < ε + gx,∞ ≤ gx + ε on K c , A = C0 (X, C) or
2
f < ε + gx on X and by construction f (x) = gx (x), see Figure 15.5. This A = IxC0 := {f ∈ C0 (X, C) : f (x0 ) = 0}
completes the proof of the claim.
3 for some x0 ∈ X.
If Ax0 = {0} and x = x0 or y = x0 , then gxy exists merely by the fact that A
separates points.

15.5 *More on Separation Axioms: Normal Spaces 147
Proof. Since 15.5 *More on Separation Axioms: Normal Spaces
f + f¯ f − f¯
Re f = and Im f = ,
2 2i (This section may safely be omitted on the first reading.)
Re f and Im f are both in A. Therefore
Definition 15.35 (T0 – T2 Separation Axioms). Let (X, τ ) be a topological
AR = {Re f, Im f : f ∈ A} space. The topology τ is said to be:
is a real sub-algebra of C0 (X, R) which separates points. Therefore either AR = 1. T0 if for x 6= y in X there exists V ∈ τ such that x ∈ V and y ∈ / V or V
C0 (X, R) or AR = Ix0 ∩ C0 (X, R) for some x0 and hence A = C0 (X, C) or IxC0 such that y ∈ V but x ∈ / V.
respectively. 2. T1 if for every x, y ∈ X with x 6= y there exists V ∈ τ such that x ∈ V and
As an easy application, Theorem 15.31 and Corollary 15.32 imply Theorem y∈/ V. Equivalently, τ is T1 iff all one point subsets of X are closed.4
10.35 and Corollary 10.37 respectively. Here are a few more applications. 3. T2 if it is Hausdorff.
Example 15.33. Let f ∈ C([a, b]) be a positive function which is injective. Then Note T2 implies T1 which implies T0 . The topology in Example 15.1 is T0
PN
functions of the form k=1 ak f k with ak ∈ C and N ∈ N are dense in C([a, b]). but not T1 . If X is a finite set and τ is a T1 – topology on X then τ = 2X . To
For example if a = 1 and b = 2, then one may take f (x) = xα for any α 6= 0, prove this let x ∈ X be fixed. Then for every y 6= x in X there exists Vy ∈ τ
or f (x) = ex , etc. such that x ∈ Vy while y ∈ / Vy . Thus {x} = ∩y6=x Vy ∈ τ showing τ contains all
one point subsets of X and therefore all subsets of X. So we have to look to
Exercise 15.5. Let (X, d) be a separable compact metric space. Show that infinite sets for an example of T1 topology which is not T2 .
C(X) is also separable. Hint: Let E ⊂ X be a countable dense set and then
consider the algebra, A ⊂ C(X), generated by {d(x, ·)}x∈E . Example 15.36. Let X be any infinite set and let τ = {A ⊂ X : #(Ac ) < ∞} ∪
{∅} – the so called cofinite topology. This topology is T1 because if x 6= y in
Example 15.34. Let X = [0, ∞), λ > 0 be fixed, A be the real algebra generated X, then V = {x}c ∈ τ with x ∈ / V while y ∈ V. This topology however is not
by t → e−λt . So the general element f ∈ A is of the form f (t) = p(e−λt ), where T2 . Indeed if U, V ∈ τ are open sets such that x ∈ U, y ∈ V and U ∩ V = ∅
p(x) is a polynomial function in x with real coefficients. Since A ⊂ C0 (X, R) then U ⊂ V c . But this implies #(U ) < ∞ which is impossible unless U = ∅
separates points and e−λt ∈ A is pointwise positive, Ā = C0 (X, R). which is impossible since x ∈ U.
As an application of Example 15.34, suppose that g ∈ Cc (X, R) satisfies, The uniqueness of limits of sequences which occurs for Hausdorff topologies
Z ∞ (see Remark 15.3) need not occur for T1 – spaces. For example, let X = N and
g (t) e−λt dt = 0 for all λ > 0. (15.9) τ be the cofinite topology on X as in Example 15.36. Then xn = n is a sequence
0 in X such that xn → x as n → ∞ for all x ∈ N. For the most part we will
avoid these pathologies in the future by only considering Hausdorff topologies.
(Note well that the integral in Eq. (15.9) is really over a finite interval since g
is compactly supported.) Equation (15.9) along with linearity of the Riemann Definition 15.37 (Normal Spaces: T4 – Separation Axiom). A topological
integral implies space (X, τ ) is said to be normal or T4 if:
Z ∞
g (t) f (t) dt = 0 for all f ∈ A.
0
1. X is Hausdorff and
2. if for any two closed disjoint subsets A, B ⊂ X there exists disjoint open
We may now choose fn ∈ A such that fn → g uniformly and therefore, using the
sets V, W ⊂ X such that A ⊂ V and B ⊂ W.
continuity of the Riemann integral under uniform convergence (see Proposition
10.5), 4
If one point subsets are closed and x 6= y in X then V := {x}c is an open set
Z ∞ Z ∞
containing y but not x. Conversely if τ is T1 and x ∈ X there exists Vy ∈ τ such
0 = lim g (t) fn (t) dt = g 2 (t) dt.
n→∞ 0 0
/ Vy for all y 6= x. Therefore, {x}c = ∪y6=x Vy ∈ τ.
that y ∈ Vy and x ∈
From this last equation it is easily deduced, using the continuity of g, that
g ≡ 0. See Theorem 22.12 below, where this is done in greater generality.

Example 15.38. By Lemma 6.15 and Corollary 15.21 it follows that metric Theorem 15.43 (Urysohn Metrization Theorem for Normal Spaces).
spaces and topological spaces which are locally compact, σ – compact and Every second countable normal space, (X, τ ) , is metrizable, i.e. there is a metric
Hausdorff (in particular compact Hausdorff spaces) are normal. Indeed, in each ρ on X such that τ = τρ . Moreover, ρ may be chosen so that X is isometric
case if A, B are disjoint closed subsets of X, there f ∈ C(X, [0, 1]) such
exists to a subset Q0 ⊂ Q (Q is as in Notation 15.11) equipped with the metric d in
that f = 1 on A and f = 0 on B. Now let U = f > 21 and V = {f < 12 }. Eq. (15.2). In this metric X is totally bounded and hence the completion of X
(which is isometric to Q̄0 ⊂ Q) is compact.
Remark 15.39. A topological space, (X, τ ), is normal iff for any C ⊂ W ⊂ X
with C being closed and W being open there exists an open set U ⊂o X such Proof. (The proof here will be very similar to the proof of Theorem 15.13.)
that Let B be a countable base for τ and set
C ⊂ U ⊂ Ū ⊂ W.
To prove this first suppose X is normal. Since W c is closed and C ∩ W c = ∅, Γ := {(U, V ) ∈ B × B | Ū ⊂ V }.
there exists disjoint open sets U and V such that C ⊂ U and W c ⊂ V. Therefore
To each O ∈ τ and x ∈ O there exist (U, V ) ∈ Γ such that x ∈ U ⊂ V ⊂ O.
C ⊂ U ⊂ V c ⊂ W and since V c is closed, C ⊂ U ⊂ Ū ⊂ V c ⊂ W.
Indeed, since B is a base for τ, there exists V ∈ B such that x ∈ V ⊂ O. Because
For the converse direction suppose A and B are disjoint closed subsets of
{x}∩V c = ∅, there exists disjoint open sets U e and W such that x ∈ Ue, V c ⊂ W
X. Then A ⊂ B c and B c is open, and so by assumption there exists U ⊂o X
and U ∩ W = ∅. Choose U ∈ B such that x ∈ U ⊂ U . Since U ⊂ U e ⊂ W c,
such that A ⊂ U ⊂ Ū ⊂ B c and by the same token there exists W ⊂o X such
e e
c
that Ū ⊂ W ⊂ W̄ ⊂ B c . Taking complements of the last expression implies U ⊂ W ⊂ V and hence (U, V ) ∈ Γ. See Figure 15.6 below. In particular this
B ⊂ W̄ c ⊂ W c ⊂ Ū c .
Let V = W̄ c . Then A ⊂ U ⊂o X, B ⊂ V ⊂o X and U ∩ V ⊂ U ∩ W c = ∅.
Theorem 15.40 (Urysohn’s Lemma for Normal Spaces). Let X be a
normal space. Assume A, B are disjoint closed subsets of X. Then there exists
f ∈ C(X, [0, 1]) such that f = 0 on A and f = 1 on B.
Proof. To make the notation match Lemma 15.8, let U = Ac and K = B.
Then K ⊂ U and it suffices to produce a function f ∈ C(X, [0, 1]) such that
f = 1 on K and supp(f ) ⊂ U. The proof is now identical to that for Lemma
15.8 except we now use Remark 15.39 in place of Proposition 15.7.
Theorem 15.41 (Tietze Extension Theorem). Let (X, τ ) be a normal
space, D be a closed subset of X, −∞ < a < b < ∞ and f ∈ C(D, [a, b]). Fig. 15.6. Constructing (U, V ) ∈ Γ.
Then there exists F ∈ C(X, [a, b]) such that F |D = f.
Proof. The proof is identical to that of Theorem 7.4 except we now use
Theorem 15.40 in place of Lemma 6.15. shows that
B0 := {U ∈ B : (U, V ) ∈ Γ for some V ∈ B}
Corollary 15.42. Suppose that X is a normal topological space, D ⊂ X is
closed, F ∈ C(D, R). Then there exists F ∈ C(X) such that F |D = f. is still a base for τ.
If Γ is a finite set, the previous comment shows that τ only has a finite
Proof. Let g = arctan(f ) ∈ C(D, (− π2 , π2 )). Then by the Tietze ex- number of elements as well. Since (X, τ ) is Hausdorff, it follows that X is a
tension theorem, there exists G ∈ C(X, [− π2 , π2 ]) such that G|D = g. Let N
finite set. Letting {xn }n=1 be an enumeration of X, define T : X → Q by
B := G−1 ({− π2 , π2 }) @ X, then B ∩ D = ∅. By Urysohn’s lemma (Theo- T (xn ) = en for n = 1, 2, . . . , N where en = (0, 0, . . . , 0, 1, 0, . . . ), with the 1
rem 15.40) there exists h ∈ C(X, [0, 1]) such that h ≡ 1 on D and h = 0 occurring in the nth spot. Then ρ(x, y) := d(T (x), T (y)) for x, y ∈ X is the
on B and in particular hG ∈ C(D, (− π2 , π2 )) and (hG) |D = g. The function desired metric.
F := tan(hG) ∈ C(X) is an extension of f.

15.6 Exercises 149
∞
So we may now assume that Γ is an infinite set and let {(Un , Vn )}n=1
be
an enumeration of Γ. By Urysohn’s Lemma for normal spaces (Theorem 15.40)
there exists fU,V ∈ C(X, [0, 1]) such that fU,V = 0 on Ū and fU,V = 1 on V c .
Let F := {fU,V | (U, V ) ∈ Γ } and set fn := fUn ,Vn – an enumeration of F. The
proof that
∞
X 1
ρ(x, y) := n
|fn (x) − fn (y)|
n=1
2
is the desired metric on X now follows exactly as the corresponding argument
in the proof of Theorem 15.13.
15.6 Exercises
Fig. 15.7. Sterographic projection and the one point compactification of Rn .
Exercise 15.6. Prove Theorem 15.9. Hints:
1. By Proposition 15.7, there exists a precompact open set V such that K ⊂
V ⊂ V̄ ⊂ U. Now suppose that f : K → [0, α] is continuous with α ∈ (0, 1] Exercise 15.11. In this problem, suppose Theorem 15.31 has only been proved
and let A := f −1 ([0, 13 α]) and B := f −1 ([ 23 α, 1]). Appeal to Lemma 15.8 to when X is compact. Show that it is possible to prove Theorem 15.31 by using
find a function g ∈ C(X, [0, α/3]) such that g = α/3 on B and supp(g) ⊂ Proposition 15.24 to reduce the non-compact case to the compact case.
V \ A. Hints:
2. Now follow the argument in the proof of Theorem 7.4 to construct F ∈
1. If Ax = R for all x ∈ X let X ∗ = X ∪{∞} be the one point compactification
C(X, [a, b]) such that F |K = f.
of X.
3. For c ∈ [a, b], choose φ ≺ U such that φ = 1 on K and replace F by
2. If Ax0 = {0} for some x0 ∈ X, let Y := X \ {x0 } and Y ∗ = Y ∪ {∞} be
Fc := φF + (1 − φ)c.
the one point compactification of Y.
Exercise 15.7 (Sterographic Projection). Let X = Rn , X ∗ := X ∪ {∞} 3. For f ∈ A define f (∞) = 0. In this way A may be considered to be a
be the one point compactification of X, S n := {y ∈ Rn+1 : |y| = 1} be the sub-algebra of C(X ∗ , R) in case 1. or a sub-algebra of C(Y ∗ , R) in case 2.
unit sphere in Rn+1 and N = (0, . . . , 0, 1) ∈ Rn+1 . Define f : S n → X ∗ by
Exercise 15.12. Let M < ∞, show there are polynomials pn (t) such that
f (N ) = ∞, and for y ∈ S n \ {N } let f (y) = b ∈ Rn be the unique point such
that (b, 0) is on the line containing N and y, see Figure 15.7 below. Find a lim sup ||t| − pn (t)| = 0
formula for f and show f : S n → X ∗ is a homeomorphism. (So the one point n→∞ |t|≤M
compactification of Rn is homeomorphic to the n sphere.)

using the following outline.
Exercise 15.8. Let (X, τ ) be a locally compact Hausdorff space. Show (X, τ ) √
is separable iff (X ∗ , τ ∗ ) is separable. 1. Let f (x) = 1 − x for |x| ≤ 1 and use Taylor’s theorem with integral
remainder (see Eq. A.15 of Appendix A), or analytic function theory if you
Exercise 15.9. Show by example that there exists a locally compact metric know it, to show there are constants5 cn > 0 for n ∈ N such that
space (X, d) such that the one point compactification, (X ∗ := X ∪ {∞} , τ ∗ ) , ∞
is not metrizable. Hint: use exercise 15.8. √ X
1−x=1− cn xn for all |x| < 1. (15.10)
n=1
Exercise 15.10. Suppose (X, d) is a locally compact and σ – compact metric
space. Show the one point compactification, (X ∗ := X ∪ {∞} , τ ∗ ) , is metriz- 5
In fact cn := (2n−3)!!
, but this is not needed.
2n n!
able.

Pm n
P∞
2. Let qm (x) := 1− n=1 cn x . Use (15.10) to show n=1 cn = 1 and conclude
from this that √
lim sup | 1 − x − qm (x)| = 0. (15.11)
m→∞ |x|≤1
3. Let 1 − x = t2 /M 2 , i.e. x = 1 − t2 /M 2 , then

|t|
− qm (1 − t2 /M 2 ) = 0

lim sup
m→∞ |t|≤M M
so that pm (t) := M qm (1 − t2 /M 2 ) are the desired polynomials.
Exercise 15.13. Given a continuous function f : R → C which is 2π -

periodic and ε > 0. Show there exists a trigonometric polynomial, p(θ) =
n
αn einθ , such that |f (θ) − P (θ)| < ε for all θ ∈ R. Hint: show that there
P
n=−N
exists a unique function F ∈ C(S 1 ) such that f (θ) = F (eiθ ) for all θ ∈ R.
Remark 15.44. Exercise 15.13 generalizes to 2π – periodic functions on Rd , i.e.

d
functions such that f (θ + 2πei ) = f (θ) for all i = 1, 2, . . . , d where {ei }i=1 is the
standard basis for Rd . A trigonometric polynomial p(θ) is a function of θ ∈ Rd
of the form X
p(θ) = αn ein·θ
n∈Γ
d
where Γ is a finite subset of Z . The assertion is again that these trigonometric
polynomials are dense in the 2π – periodic functions relative to the supremum
norm.
16
Baire Category Theorem
o τ ) be a topological space. A set E ⊂ X is said to be

Definition 16.1. Let (X, all n and hence x ∈ G. Moreover, x ∈ B(x1 , ε1 ) ⊂ W ∩ V1 implies x ∈ W and
nowhere dense if Ē = ∅ i.e. Ē has empty interior. hence x ∈ W ∩ G showing W ∩ G 6= ∅.
2. The second assertion is equivalently to showing
Notice that E is nowhere dense is equivalent to
∞
!c ∞ ∞
o c c [ \ c \ o
X= Ē = Ē
o
= (E c ) . ∅=6 Ēn = Ēn = (Enc ) .
n=1 n=1 n=1
That is to say E is nowhere dense iff E c has dense interior. o
nowhere dense is equivalent to (Enc ) being a dense
As we have observed, En is T
∞ o
open set, hence by part 1), n=1 (Enc ) is dense in X and hence not empty.
16.1 Metric Space Baire Category Theorem Example 16.3. Suppose that X is a countable set and ρ is a metric on X for
which no single point set is open. Then (X, ρ) is not complete. Indeed we may
Theorem 16.2 (Baire Category Theorem). Let (X, ρ) be a complete metric assume X = N and let En := {n} ⊂ N for all n ∈ N. Then En is closed and by
space. assumption it has empty interior. Since X = ∪n∈N En , it follows from the Baire
∞
∞
T Category Theorem 16.2 that (X, ρ) can not be complete.
1. If {Vn }n=1 is a sequence of dense open sets, then G := Vn is dense in
n=1
X.
∞ S∞
If {En }n=1 is a sequence of nowhere
2. S dense sets, then n=1 En ⊂ 16.2 Locally Compact Hausdorff Space Baire Category
∞ S ∞
n=1 Ē n & X and in particular X 6
= n=1 E n . Theorem
Proof. 1. We must show that Ḡ = X which is equivalent to showing that
Here is another version of the Baire Category theorem when X is a locally
W ∩G 6= ∅ for all non-empty open sets W ⊂ X. Since V1 is dense, W ∩V1 6= ∅ and
compact Hausdorff space.
hence there exists x1 ∈ X and ε1 > 0 such that
Theorem 16.4. Let X be a locally compact Hausdorff space.
B(x1 , ε1 ) ⊂ W ∩ V1 .
∞
Proposition 16.5. 1. If {Vn }n=1 is a sequence of dense open sets, then G :=
Since V2 is dense, B(x1 , ε1 ) ∩ V2 6= ∅ and hence there exists x2 ∈ X and ε2 > 0 ∞
T
such that Vn is dense in X.
n=1
B(x2 , ε2 ) ⊂ B(x1 , ε1 ) ∩ V2 . ∞ S∞
2. If {En }n=1 is a sequence of nowhere dense sets, then X 6= n=1 En .
∞
Continuing this way inductively, we may choose {xn ∈ X and εn > 0}n=1 such
that Proof. As in the previous proof, the second assertion is a consequence of
B(xn , εn ) ⊂ B(xn−1 , εn−1 ) ∩ Vn ∀n. the first. To finish the proof, if suffices to show G ∩ W 6= ∅ for all open sets
W ⊂ X. Since V1 is dense, there exists x1 ∈ V1 ∩ W and by Proposition 15.7
Furthermore we can clearly do this construction in such a way that εn ↓ 0 as there exists U1 ⊂o X such that x1 ∈ U1 ⊂ Ū1 ⊂ V1 ∩ W with Ū1 being compact.
n ↑ ∞. Hence {xn }∞
n=1 is Cauchy sequence and x = lim xn exists in X since Similarly, there exists a non-empty open set U2 such that U2 ⊂ Ū2 ⊂ U1 ∩ V2 .
n→∞ ∞
X is complete. Since B(xn , εn ) is closed, x ∈ B(xn , εn ) ⊂ Vn so that x ∈ Vn for Working inductively, we may find non-empty open sets {Uk }k=1 such that Uk ⊂
152 16 Baire Category Theorem
Ūk ⊂ Uk−1 ∩ Vk . Since ∩nk=1 Ūk = Ūn 6= ∅ for all n, the finite intersection Theorem 16.9. Let N ⊂ C([0, 1], R) be the set of nowhere differentiable
characterization of Ū1 being compact implies that functions. (Here a function f is said to be differentiable at 0 if f 0 (0) :=
limt↓0 f (t)−f (0)
exists and at 1 if f 0 (1) := limt↑0 f (1)−f (t)
exists.) Then N is
6 ∩∞
∅= k=1 Ūk ⊂ G ∩ W.
t 1−t
a residual set so the “generic” continuous functions is nowhere differentiable.
Proof. If f ∈ / N , then f 0 (x0 ) exists for some x0 ∈ [0, 1] and by the defini-
Definition 16.6. A subset E ⊂ X is meager or of the first category if tion of the derivative and compactness of [0, 1], there exists n ∈ N such that
∞
E =
S
En where each En is nowhere dense. And a set R ⊂ X is called |f (x) − f (x0 )| ≤ n|x − x0 | ∀ x ∈ [0, 1]. Thus if we define
n=1
residual if Rc is meager. En := {f ∈ C([0, 1]) : ∃ x0 ∈ [0, 1] 3 |f (x) − f (x0 )| ≤ n|x − x0 | ∀ x ∈ [0, 1]} ,
Remarks 16.7 For those readers that already know some measure theory may then we have just shown N c ⊂ E := ∪∞ n=1 En . So to finish the proof it suffices
want to think of meager as being the topological analogue of sets of measure to show (for each n) En is a closed subset of C([0, 1], R) with empty interior.
0 and residual as being the topological analogue of sets of full measure. (This ∞
1. To prove En is closed, let {fm }m=1 ⊂ En be a sequence of functions such
analogy should not be taken too seriously, see Exercise 19.19.) that there exists f ∈ C([0, 1], R) such that kf − fm k∞ → 0 as m → ∞. Since
1. R is residual iff R contains a countable intersection of dense open sets. fm ∈ En , there exists xm ∈ [0, 1] such that
Indeed if R is a residual set, then there exists nowhere dense sets {En } |fm (x) − fm (xm )| ≤ n|x − xm | ∀ x ∈ [0, 1]. (16.1)
such that
Rc = ∪∞ ∞
n=1 En ⊂ ∪n=1 Ēn . Since [0, 1] is a compact metric space, by passing to a subsequence if neces-
Taking complements of this equation shows that sary, we may assume x0 = limm→∞ xm ∈ [0, 1] exists. Passing to the limit
in Eq. (16.1), making use of the uniform convergence of fn → f to show
∩∞ c
n=1 Ēn ⊂ R, limm→∞ fm (xm ) = f (x0 ), implies
i.e. R contains a set of the form ∩∞ c
n=1 Vn with each Vn (= Ēn ) being an |f (x) − f (x0 )| ≤ n|x − x0 | ∀ x ∈ [0, 1]
open dense subset of X.
Conversely, if ∩∞
n=1 Vn ⊂ R with each Vn being an open dense subset of X, and therefore that f ∈ En . This shows En is a closed subset of C([0, 1], R).
then Rc ⊂ ∪∞ c c ∞ c
n=1 Vn and hence R = ∪n=1 En where each En = R ∩ Vn , is
c
2. To finish the proof, we will show En0 = ∅ by showing for each f ∈ En
a nowhere dense subset of X. and ε > 0 given, there exists g ∈ C([0, 1], R) \ En such that kf − gk∞ < ε.
2. A countable union of meager sets is meager and any subset of a meager set We now construct g. Since [0, 1] is compact and f is continuous there exists
is meager. N ∈ N such that |f (x) − f (y)| < ε/2 whenever |y − x| < 1/N. Let k denote the
3. A countable intersection of residual sets is residual. piecewise linear function on [0, 1] such that k( m m
N ) = f ( N ) for m = 0, 1, . . . , N
and k 00 (x) = 0 for x ∈
/ πN := {m/N : m = 0, 1, . . . , N } . Then it is easily seen
Remarks 16.8 The Baire Category Theorems may now be stated as follows.
that kf − kku < ε/2 and for x ∈ ( m m+1
N , N ) that
If X is a complete metric space or X is a locally compact Hausdorff space, then
1. all residual sets are dense in X and |f ( m+1 m
N ) − f ( N )|
|k 0 (x)| = 1 < N ε/2.
2. X is not meager. N
It should also be remarked that incomplete metric spaces may be meager. We now make k “rougher” by adding a small wiggly function h which we define
For example, let X ⊂ C([0, 1]) be the subspace of polynomial functions on [0, 1] as follows. Let M ∈ N be chosen so that 4εM > 2n and define h uniquely
equipped with the supremum norm. Then X = ∪∞ n=1 En where En ⊂ X denotes by h( Mm
) = (−1)m ε/2 for m = 0, 1, . . . , M and h00 (x) = 0 for x ∈/ πM . Then
the subspace of polynomials of degree less than or equal to n. You are asked khk∞ < ε and |h0 (x)| = 4εM > 2n for x ∈ / πM . See Figure 16.1 below. Finally
to show in Exercise 16.1 below that En is nowhere dense for all n. Hence X is define g := k + h. Then
meager and the empty set is residual in X.
Here is an application of Theorem 16.2. kf − gk∞ ≤ kf − kk∞ + khk∞ < ε/2 + ε/2 = ε

16.2 Locally Compact Hausdorff Space Baire Category Theorem 153
which is a closed subset of R since f is continuous. Moreover, if x ∈ W and

M = |f 0 (x)| , then
|f (y) − f (x)| = |f 0 (x) (y − x) + o (y − x)|

≤ (M + 1) |y − x|
for y close to x. (Here o(y − x) denotes a function such that limy→x o(y −
x)/(y − x) = 0.) In particular, this shows that x ∈ Ek for all k sufficiently
large. Therefore W = ∪∞
k=1 Ek and since W is not meager by the Baire category
Theorem 16.4, some Ek has non-empty interior. That is there exists x0 ∈ Ek ⊂
W and ε > 0 such that
J := (x0 − ε, x0 + ε) ⊂ Ek ⊂ W.
Fig. 16.1. Constgructing a rough approximation, g, to a continuous function f.
For x ∈ J, we have |f (x + z) − f (x)| ≤ k |z| provided that |z| ≤ k −1 and
therefore that |f 0 (x)| ≤ k for x ∈ J. Therefore x0 ∈ U ∩ W showing U is dense.
and
|g 0 (x)| ≥ |h0 (x)| − |k 0 (x)| > 2n − n = n ∀x ∈
/ πM ∪ πN .
Remark 16.11. This proposition generalizes to functions f : Rn → Rm in an
It now follows from this last equation and the mean value theorem that for any obvious way.
x0 ∈ [0, 1],
For our next application of Theorem 16.2, let X := BC ∞ ((−1, 1)) denote

g(x) − g(x0 )
x − x0 > n

the set of smooth functions f on (−1, 1) such that f and all of its derivatives
are bounded. In the metric
for all x ∈ [0, 1] sufficiently close to x0 . This shows g ∈
/ En and so the proof is
∞
(k)
complete. X f − g (k)
−k ∞ for f, g ∈ X,
Here is an application of the Baire Category Theorem 16.4. ρ(f, g) := 2
k=0
1 + f (k) − g (k) ∞
Proposition 16.10. Suppose that f : R → R is a function such that f 0 (x) ex-
ists for all x ∈ R. Let X becomes a complete metric space.
∞
[
( ) Theorem 16.12. Given an increasing sequence of positive numbers {Mn }n=1 ,
0 the set
U := x ∈ R : sup |f (x + y)| < ∞ . (n)
|y|<ε f (0)
ε>0 F := f ∈ X : lim sup ≥1
n→∞ Mn
Then U is a dense open set. (It is not true that U = R in general, see Example
is dense in X. In particular, there is a dense set of f ∈ X such that the power
30.27 below.)
series expansion of f at 0 has zero radius of convergence.
Proof. It is easily seen from the definition of U that U is open. Let W ⊂o R
Proof. Step 1. Let n ∈ N. Choose g ∈ Cc∞ ((−1, 1)) such that kgk∞ < 2−n
be an open subset of R. For k ∈ N, let
while g 0 (0) = 2Mn and define

1 x tn−1 t2
Ek := x ∈ W : |f (y) − f (x)| ≤ k |y − x| when |y − x| ≤
Z Z Z
k fn (x) := dtn−1 dtn−2 . . . dt1 g(t1 ).
\ 0 0 0
= {x ∈ W : |f (x + z) − f (x)| ≤ k |z|} ,
z:|z|≤k−1 Then for k < n,

154 16 Baire Category Theorem
x tn−k−1 t2
Z Z Z
fn (0) n f (0)
fn(k) (x) = dtn−k−1 dtn−k−2 . . . dt1 g(t1 ), lim sup x = lim sup
n n
n!x

n! 2
0 0 0 n→∞ n→∞ (n!)

(n) (k)
f (n) (x) = g 0 (x), fn (0) = 2Mn and fn
f (0)
satisfies n n
= lim sup 2 · lim n! |x | = ∞
n→∞ (n!) n→∞
2−n
≤ 2−n for k < n.
(k)
fn ≤
∞ (n − 1 − k)!
where we have used limn→∞ n! |xn | = ∞ and lim supn→∞ f(n!)
n (0)
2 ≥ 1.
Consequently,
∞
Remark 16.13. Given a sequence of real number {an }n=0 there always exists
∞
(k) f ∈ X such that f (n) (0) = an . To construct such a function f, let φ ∈ Cc∞ (−1, 1)
X fn
ρ(fn , 0) = 2−k ∞
(k) be a function such that φ = 1 inPa neighborhood of 0 and εn ∈ (0, 1) be chosen
1 + ∞
k=0 fn
∞
so that εn ↓ 0 as n → ∞ and n=0 |an | εnn < ∞. The desired function f can
n−1 ∞ then be defined by
X X
2−k 2−n + 2−k · 1 ≤ 2 2−n + 2−n = 4 · 2−n .

≤ ∞ ∞
k=0 k=n
X an n X
f (x) = x φ(x/εn ) =: gn (x). (16.2)
n=0
n! n=0
Thus we have constructed fn ∈ X such that limn→∞ ρ(fn , 0) = 0 while
(n)
fn (0) = 2Mn for all n. The fact that f is well defined and continuous follows from the estimate:
Step 2. The set a kφk
n ∞
|gn (x)| = xn φ(x/εn ) ≤ |an | εnn
n o
Gn := ∪m≥n f ∈ X : f (m) (0) > Mm n! n!

P∞
and the assumption that n=0 |an | εnn < ∞. The estimate
is a dense open subset of X. The fact that Gn is open is clear. To see that Gn is
dense, let g ∈ X be given and define gm := g +εm fm where εm := sgn(g (m) (0)).
an an n 0

0 n−1

Then |gn (x)| = x φ(x/εn ) + x φ (x/εn )

(m) (m) (m)
(n − 1)! n!εn
gm (0) = g (0) + fm (0) ≥ 2Mm > Mm for all m. 0
kφk∞ kφ k ∞
≤ |an | εn−1
n + |an | εnn
Therefore, gm ∈ Gn for all m ≥ n and since (n − 1)! n!
≤ (kφk∞ + kφ0 k∞ ) |an | εnn
ρ(gm , g) = ρ(fm , 0) → 0 as m → ∞
P∞ n 1
it follows that g ∈ Ḡn . and the P assumption that n=0 |an | εn < ∞ shows f ∈ C (−1, 1) and
0 ∞ 0
Step 3. By the Baire Category theorem, ∩Gn is a dense subset of X. This f (x) = g (x). Similar arguments show f ∈ Cc (−1, 1) and f (k) (x) =
k
P∞ (k) n=0 n
completes the proof of the first assertion since n=0 gn (x) for all x and k ∈ N. This completes the proof since, using
(k)
(n) φ(x/εn ) = 1 for x in a neighborhood of 0, gn (0) = δk,n ak and hence
f (0)
F = f ∈ X : lim sup ≥1 ∞
n→∞ Mn X
(n) f (k) (0) = gn(k) (0) = ak .
f (0)
= ∩∞n=1 f ∈ X :
Mn
≥ 1 for some n ≥ m ⊃ ∩∞
n=1 Gn .
n=0
2
Step 4. Take Mn = (n!) and recall that the power series expansion for f 16.3 Exercises
P∞
near 0 is given by n=0 fnn!(0) xn . This series can not converge for any f ∈ F
and any x 6= 0 because Exercise 16.1. Let (X, k·k) be a normed space and E ⊂ X be a subspace.

1. If E is closed and proper subspace of X then E is nowhere dense.
2. If E is a proper finite dimensional subspace of X then E is nowhere dense.
Exercise 16.2. Now suppose that (X, k·k) is an infinite dimensional Banach
space. Show that X can not have a countable algebraic basis. More explicitly,
there is no countable subset S ⊂ X such that every element x ∈ X may be
written as a finite linear combination of elements from S. Hint: make use of
Exercise 16.1 and the Baire category theorem.
Part V
Lebesgue Integration Theory

17
Introduction: What are measures and why “measurable” sets
Definition 17.1 (Preliminary). A measure µ “on” a set X is a function Example 17.4. Suppose that X is a set λ : X → [0, ∞] is a function. Then
µ : 2X → [0, ∞] such that X
µ := λ(x)δx
1. µ(∅) = 0 x∈X
N
2. If {Ai }i=1 is a finite (N < ∞) or countable (N = ∞) collection of subsets
of X which are pair-wise disjoint (i.e. Ai ∩ Aj = ∅ if i 6= j) then is a measure, explicitly X
µ(A) = λ(x)
N
X x∈A
µ(∪N
i=1 Ai ) = µ(Ai ).
i=1
for all A ⊂ X.
Example 17.2. Suppose that X is any set and x ∈ X is a point. For A ⊂ X, let
17.1 The problem with Lebesgue “measure”
1 if x ∈ A
δx (A) =
0 if x ∈
/ A. So far all of the examples of measures given above are “counting” type measures,
Then µ = δx is a measure on X called the Dirac delta measure at x. i.e. a weighted count of the number of points in a set. We certainly are going to
want other types of measures too. In particular, it will be of great interest to
Example 17.3. Suppose that µ is a measure on X and λ > 0, then λ · µ have a measure on R (called Lebesgue measure) which measures the “length”
Pis also a
measure on X. Moreover, if {µα }α∈J are all measures on X, then µ = α∈J µα , of a subset of R. Unfortunately as the next theorem shows, there is no such
i.e. X reasonable measure of length if we insist on measuring all subsets of R.
µ(A) = µα (A) for all A ⊂ X
α∈J
Theorem 17.5. There is no measure µ : 2R →[0, ∞] such that
is a measure on X. (See Section 2 for the meaning of this sum.) To prove this 1. µ([a, b)) = (b − a) for all a < b and
∞
we must show that µ is countably additive. Suppose that {Ai }i=1 is a collection 2. is translation invariant, i.e. µ(A + x) = µ(A) for all x ∈ R and A ∈ 2R ,
of pair-wise disjoint subsets of X, then where
A + x := {y + x : y ∈ A} ⊂ R.
∞
X ∞ X
X
µ(∪∞
i=1 Ai ) = µ(Ai ) = µα (Ai ) In fact the theorem is still true even if (1) is replaced by the weaker condition
i=1 i=1 α∈J that 0 < µ((0, 1]) < ∞.
∞
XX X
= µα (Ai ) = µα (∪∞
i=1 Ai )
The counting measure µ (A) = # (A) is translation invariant. However
α∈J i=1 α∈J µ((0, 1]) = ∞ in this case and so µ does not satisfy condition 1.
= µ(∪∞ Proof. First proof. Let us identify [0, 1) with the unit circle S 1 := {z ∈
i=1 Ai )
C : |z| = 1} by the map
wherein the third equality we used Theorem 4.22 and in the fourth we used
that fact that µα is a measure. φ(t) = ei2πt = (cos 2πt + i sin 2πt) ∈ S 1
160 17 Introduction: What are measures and why “measurable” sets
`
for t ∈ [0, 1). Using this identification we may use µ to define a function ν on where α Aα is used to denote the union of pair-wise disjoint sets {Aα } . By
1
2S by ν(φ(A)) = µ(A) for all A ⊂ [0, 1). This new function is a measure on S 1 Eqs. (17.2) and (17.3),
with the property that 0 < ν((0, 1]) < ∞. For z ∈ S 1 and N ⊂ S 1 let X X
ν(S 1 ) = ν(rN ) = ν(N ).
zN := {zn ∈ S 1 : n ∈ N }, (17.1) r∈R r∈R
that is to say eiθ N is N rotated counter clockwise by angle θ. We now claim The right member from this equation is either 0 or ∞, 0 if ν(N ) = 0 and ∞ if
that ν is invariant under these rotations, i.e. ν(N ) > 0. In either case it is not equal ν(S 1 ) ∈ (0, 1). Thus we have reached
the desired contradiction.
ν(zN ) = ν(N ) (17.2) Proof. Second proof of Theorem 17.5. For N ⊂ [0, 1) and α ∈ [0, 1), let
for all z ∈ S 1 and N ⊂ S 1 . To verify this, write N = φ(A) and z = φ(t) for N α = N + α mod 1
some t ∈ [0, 1) and A ⊂ [0, 1). Then = {a + α mod 1 ∈ [0, 1) : a ∈ N }
φ(t)φ(A) = φ(t + A mod 1) = (α + N ∩ {a < 1 − α}) ∪ ((α − 1) + N ∩ {a ≥ 1 − α}) .
where for A ⊂ [0, 1) and α ∈ [0, 1), Then
t + A mod 1 := {a + t mod 1 ∈ [0, 1) : a ∈ N } µ (N α ) = µ (α + N ∩ {a < 1 − α}) + µ ((α − 1) + N ∩ {a ≥ 1 − α})

= (a + A ∩ {a < 1 − t}) ∪ ((t − 1) + A ∩ {a ≥ 1 − t}) . = µ (N ∩ {a < 1 − α}) + µ (N ∩ {a ≥ 1 − α})
= µ (N ∩ {a < 1 − α} ∪ (N ∩ {a ≥ 1 − α}))
Thus
= µ(N ). (17.4)
ν(φ(t)φ(A)) = µ(t + A mod 1) We will now construct a bad set N which coupled with Eq. (17.4) will lead to
= µ ((a + A ∩ {a < 1 − t}) ∪ ((t − 1) + A ∩ {a ≥ 1 − t})) a contradiction. Set
= µ ((a + A ∩ {a < 1 − t})) + µ (((t − 1) + A ∩ {a ≥ 1 − t}))
Qx := {x + r ∈ R : r∈ Q} =x + Q.
= µ (A ∩ {a < 1 − t}) + µ (A ∩ {a ≥ 1 − t})
= µ ((A ∩ {a < 1 − t}) ∪ (A ∩ {a ≥ 1 − t})) Notice that Qx ∩ Qy 6= ∅ implies that Qx = Qy . Let O = {Qx : x ∈ R} – the
= µ(A) = ν(φ(A)). orbit space of the Q action. For all A ∈ O choose f (A) ∈ [0, 1/3) ∩ A1 and
define N = f (O). Then observe:
Therefore it suffices to prove that no finite non-trivial measure ν on S 1 such that
1. f (A) = f (B) implies that A ∩ B 6= ∅ which implies that A = B so that f is
Eq. (17.2) holds. To do this we will “construct” a non-measurable set N = φ(A)
injective.
for some A ⊂ [0, 1). Let
2. O = {Qn : n ∈ N }.
R := {z = ei2πt : t ∈ Q} = {z = ei2πt : t ∈ [0, 1) ∩ Q} Let R be the countable set,
– a countable subgroup of S 1 . As above R acts on S 1 by rotations and divides R := Q ∩ [0, 1).
S 1 up into equivalence classes, where z, w ∈ S 1 are equivalent if z = rw for some
r ∈ R. Choose (using the axiom of choice) one representative point n from each We now claim that
of these equivalence classes and let N ⊂ S 1 be the set of these representative
points. Then every point z ∈ S 1 may be uniquely written as z = nr with n ∈ N N r ∩ N s = ∅ if r 6= s and (17.5)
and r ∈ R. That is to say a [0, 1) = ∪r∈R N r . (17.6)
S1 = (rN ) (17.3) 1 Q
r∈R We have used the Axiom of choice here, i.e. A∈F (A ∩ [0, 1/3]) 6= ∅

Indeed, if x ∈ N r ∩ N s 6= ∅ then x = r + n mod 1 and x = s + n0 mod 1, then
n − n0 ∈ Q, i.e. Qn = Qn0 . That is to say, n = f (Qn ) = f (Qn0 ) = n0 and hence
that s = r mod 1, but s, r ∈ [0, 1) implies that s = r. Furthermore, if x ∈ [0, 1)
and n := f (Qx ), then x − n = r ∈ Q and x ∈ N r mod 1 . Now that we have
constructed N, we are ready for the contradiction. By Equations (17.4–17.6)
we find
X X
1 = µ([0, 1)) = µ(N r ) = µ(N )
r∈R r∈R

∞ if µ(N ) > 0
= .
0 if µ(N ) = 0
which is certainly inconsistent. Incidentally we have just produced an example

of so called “non – measurable” set.
Because of Theorem 17.5, it is necessary to modify Definition 17.1. Theorem
17.5 points out that we will have to give up the idea of trying to measure all
subsets of R but only measure some sub-collections of “measurable” sets. This
leads us to the notion of σ – algebra discussed in the next chapter. Our revised
notion of a measure will appear in Definition 19.1 of Chapter 19 below.
18
Measurability
\
18.1 Algebras and σ – Algebras A(E) := {A : A is an algebra such that E ⊂ A}
Definition 18.1. A collection of subsets A of a set X is an algebra if and \

σ(E) := {M : M is a σ – algebra such that E ⊂ M}.
1. ∅, X ∈ A
2. A ∈ A implies that Ac ∈ A
3. A is closed under finite unions, i.e. if A1 , . . . , An ∈ A then A1 ∪· · ·∪An ∈ A.
In view of conditions 1. and 2., 3. is equivalent to Example 18.5. Suppose X = {1, 2, 3} and E = {∅, X, {1, 2}, {1, 3}}, see Figure
30 . A is closed under finite intersections. 18.1.
Definition 18.2. A collection of subsets M of X is a σ – algebra (or some-
times called a σ – field) if M is an algebra which also closed under countable
∞
unions, i.e. if {Ai }i=1 ⊂ M, then ∪∞i=1 Ai ∈ M. (Notice that since M is also
closed under taking complements, M is also closed under taking countable in-
tersections.) A pair (X, M), where X is a set and M is a σ – algebra on X, is
called a measurable space.
The reader should compare these definitions with that of a topology in

Definition 13.1. Recall that the elements of a topology are called open sets.
Analogously, elements of and algebra A or a σ – algebra M will be called
measurable sets.
Fig. 18.1. A collection of subsets.
Example 18.3. Here are some examples of algebras.
1. M = 2X , then M is a topology, an algebra and a σ – algebra.
2. Let X = {1, 2, 3}, then τ = {∅, X, {2, 3}} is a topology on X which is not Then
an algebra.
3. τ = A = {{1}, {2, 3}, ∅, X} is a topology, an algebra, and a σ – algebra on τ (E) = {∅, X, {1}, {1, 2}, {1, 3}}
X. The sets X, {1}, {2, 3}, ∅ are open and closed. The sets {1, 2} and {1, 3} A(E) = σ(E) = 2X .
are neither open nor closed and are not measurable.
The next proposition is the analogue to Proposition 13.7 for topologies and
The reader should compare this example with Example 13.3. enables us to give and explicit descriptions of A(E). On the other hand it should
be noted that σ(E) typically does not admit a simple concrete description.
Proposition 18.4. Let E be any collection of subsets of X. Then there exists
a unique smallest algebra A(E) and σ – algebra σ(E) which contains E. Proposition 18.6. Let X be a set and E ⊂ 2X . Let E c := {Ac : A ∈ E} and
Ec := E ∪ {X, ∅} ∪ E c Then
Proof. The proof is the same as the analogous Proposition 13.6 for topolo-
gies, i.e. A(E) := {finite unions of finite intersections of elements from Ec }. (18.1)
164 18 Measurability
Proof. Let A denote the right member of Eq. (18.1). From the definition of • if E ∈ E, then E c is a finite disjoint union of sets from E. (In particular
an algebra, it is clear that E ⊂ A ⊂ A(E). Hence to finish that proof it suffices X = ∅c is a finite disjoint union of elements from E.)
to show A is an algebra. The proof of these assertions are routine except for
Example 18.9. Let X = R, then
possibly showing that A is closed under complementation. To check A is closed
under complementation, let Z ∈ A be expressed as

E := (a, b] ∩ R : a, b ∈ R̄
N \
[ K = {(a, b] : a ∈ [−∞, ∞) and a < b < ∞} ∪ {∅, R}
Z= Aij
is an elementary family.
i=1 j=1
Exercise 18.2. Let A ⊂ 2X and B ⊂ 2Y be elementary families. Show the
where Aij ∈ Ec . Therefore, writing Bij = Acij ∈ Ec , we find that
collection
K
N [ K E = A × B = {A × B : A ∈ A and B ∈ B}
\ [
Zc = Bij = (B1j1 ∩ B2j2 ∩ · · · ∩ BN jN ) ∈ A is also an elementary family.
i=1 j=1 j1 ,...,jN =1
Proposition 18.10. Suppose E ⊂ 2X is an elementary family, then A = A(E)
wherein we have used the fact that B1j1 ∩B2j2 ∩· · ·∩BN jN is a finite intersection consists of sets which may be written as finite disjoint unions of sets from E.
of sets from Ec .
Proof. This could be proved making use of Proposition 18.6. However it
Remark 18.7. One might think that in general σ(E) may be described as the is easier to give a direct proof. Let A denote the collection of sets which may
countable unions of countable intersections of sets in E c . However this is in be written as finite disjoint unions of sets from E. Clearly E ⊂ A ⊂ A(E) so it
general false, since if suffices to show A is an algebra since A(E) is the smallest algebra containing
[∞ \ ∞ E. By the properties of E, we know that ∅, X ∈ A. Now suppose that Ai =
Z= Aij `
F ∈Λi F ∈ A where, for i = 1, 2, . . . , n, Λi is a finite collection of disjoint sets
i=1 j=1 from E. Then
with Aij ∈ Ec , then n
\ \n a
!
[
∞ ∞
! Ai = F = (F1 ∩ F2 ∩ · · · ∩ Fn )
i=1 i=1 F ∈Λi (F1 ,,...,Fn )∈Λ1 ×···×Λn
[ \
c
Z = Ac`,j`
and this is a disjoint (you check) union of elements ` from E. Therefore A is
j1 =1,j2 =1,...jN =1,... `=1
closed under finite intersections. Similarly, if A =
T F ∈Λ F with Λ being a finite
which is now an uncountable union. Thus the above description is not correct.
collection of disjoint sets from E, then Ac = F ∈Λ F c . Since by assumption
In general it is complicated to explicitly describe σ(E), see Proposition 1.23 on
F c ∈ A for F ∈ Λ ⊂ E and A is closed under finite intersections, it follows that
page 39 of Folland for details. Also see Proposition 18.13 below.
Ac ∈ A.
Exercise 18.1. Let τ be a topology on a set X and A = A(τ ) be the algebra Definition 18.11. Let X be a set. We say that a family of sets F ⊂ 2X is a
generated by τ. Show A is the collection of subsets of X which may be written partition of X if distinct members of F are disjoint and if X is the union of
as finite union of sets of the form F ∩ V where F is closed and V is open. the sets in F.
The following notion will be useful in the sequel and plays an analogous role Example 18.12. Let X be a set and E = {A1 , . . . , An } where A1 , . . . , An is a
for algebras as a base (Definition 13.8) does for a topology. partition of X. In this case
Definition 18.8. A set E ⊂ 2X is said to be an elementary family or ele- A(E) = σ(E) = τ (E) = {∪i∈Λ Ai : Λ ⊂ {1, 2, . . . , n}}
mentary class provided that
where ∪i∈Λ Ai := ∅ when Λ = ∅. Notice that
• ∅∈E
• E is closed under finite intersections # (A(E)) = #(2{1,2,...,n} ) = 2n .

18.2 Measurable Functions 165
∞
Proposition 18.13. Suppose that M ⊂ 2X is a σ – algebra and M is at most (a) (a, b) =
S
(a, b − n1 ] ∈ σ(E).
a countable set. Then there exists a unique finite partition F of X such that n=1
F ⊂ M and every element B ∈ M is of the form (b) All of the
T standard open subsets of R are in σ(E).
(c) {x} = x − n1 , x ∈ σ(E)
B = ∪ {A ∈ F : A ⊂ B} . (18.2) n
(d) [a, b] = {a} ∪ (a, b] ∈ σ(E)
In particular M is actually a finite set and # (M) = 2n for some n ∈ N. (e) Any countable subset of R is in σ(E).
Proof. For each x ∈ X let Remark 18.15. In the above example, one may replace E by E = {(a, ∞) : a ∈
Q} ∪ {R, ∅}, in which case A(E) may be described as being those sets which are
Ax = ∩ {A ∈ M : x ∈ A} ∈ M, finite disjoint unions of sets from the following list
wherein we have used M is a countable σ – algebra to insure Ax ∈ M. Hence {(a, ∞), (−∞, a], (a, b] : a, b ∈ Q} ∪ {∅, R} .
Ax is the smallest set in M which contains x. Let C = Ax ∩ Ay . If x ∈
/ C then
Ax \ C ⊂ Ax is an element of M which contains x and since Ax is the smallest This shows that A(E) is a countable set – a useful fact which will be needed
member of M containing x, we must have that C = ∅. Similarly if y ∈ / C then later.
C = ∅. Therefore if C 6= ∅, then x, y ∈ Ax ∩ Ay ∈ M and Ax ∩ Ay ⊂ Ax and
Ax ∩ Ay ⊂ Ay from which it follows that Ax = Ax ∩ Ay = Ay . This shows that Notation 18.16 For a general topological space (X, τ ), the Borel σ – algebra
F = {Ax : x ∈ X} ⊂ M is a (necessarily countable) partition of X for which is the σ – algebra BX := σ(τ ) on X. In particular if X = Rn , BRn will be used
Eq. (18.2) holds for all B ∈ M. Enumerate the elements of F as F = {Pn }N to denote the Borel σ – algebra on Rn when Rn is equipped with its standard
n=1
where N ∈ N or N = ∞. If N = ∞, then the correspondence Euclidean topology.
N
a ∈ {0, 1} → Aa = ∪{Pn : an = 1} ∈ M Exercise 18.3. Verify the σ – algebra, BR , is generated by any of the following
collection of sets:
is bijective and therefore, by Lemma 2.6, M is uncountable. Thus any countable
σ – algebra is necessarily finite. This finishes the proof modulo the uniqueness 1. {(a, ∞) : a ∈ R} , 2. {(a, ∞) : a ∈ Q} or 3. {[a, ∞) : a ∈ Q} .
assertion which is left as an exercise to the reader.
Proposition 18.17. If τ is a second countable topology on X and E is a count-
Example 18.14. Let X = R and able collection of subsets of X such that τ = τ (E), then BX := σ(τ ) = σ(E), i.e.
σ(τ (E)) = σ(E).
E = {(a, ∞) : a ∈ R} ∪ {R, ∅} = {(a, ∞) ∩ R : a ∈ R̄} ⊂ 2R .
Proof. Let Ef denote the collection of subsets of X which are finite inter-
Notice that Ef = E and that E is closed under unions, which shows that τ (E) = section of elements from E along with X and ∅. Notice that Ef is still countable
E, i.e. E is already a topology. Since (a, ∞)c = (−∞, a] we find that Ec = (you prove). A setS Z is in τ (E) iff Z is an arbitrary union of sets from Ef .
{(a, ∞), (−∞, a], −∞ ≤ a < ∞} ∪ {R, ∅}. Noting that Therefore Z = A for some subset F ⊂ Ef which is necessarily count-
A∈F
(a, ∞) ∩ (−∞, b] = (a, b] able. Since Ef ⊂ σ(E) and σ(E) is closed under countable unions it follows
that Z ∈ σ(E) and hence that τ (E) ⊂ σ(E). Lastly, since E ⊂ τ (E) ⊂ σ(E),
it follows that A(E) = A(Ẽ) where σ(E) ⊂ σ(τ (E)) ⊂ σ(E).

Ẽ := (a, b] ∩ R : a, b ∈ R̄ .
18.2 Measurable Functions
Since Ẽ is an elementary family of subsets of R, Proposition 18.10 implies A(E)
may be described as being those sets which are finite disjoint unions of sets
Our notion of a “measurable” function will be analogous to that for a continuous
from Ẽ. The σ – algebra, σ(E), generated by E is very complicated. Here are
function. For motivational purposes, suppose (X, M, µ) is a measure space and
some sets in σ(E) – most of which are not in A(E).

fR : X → R+ . Roughly speaking, in the next Chapter we are going to define Lemma 18.21. Suppose that (X, M), (Y, F) and (Z, G) are measurable spaces.
f dµ as a certain limit of sums of the form, If f : (X, M) → (Y, F) and g : (Y, F) → (Z, G) are measurable functions then
X g ◦ f : (X, M) → (Z, G) is measurable as well.
∞
X Proof. By assumption g −1 (G) ⊂ F and f −1 (F) ⊂ M so that
ai µ(f −1 (ai , ai+1 ]).
−1
(g ◦ f ) (G) = f −1 g −1 (G) ⊂ f −1 (F) ⊂ M.
0<a1 <a2 <a3 <...

For this to make sense we will need to require f −1 ((a, b]) ∈ M for all a < b.
Because of Lemma 18.22 below, this last condition is equivalent to the condition
f −1 (BR ) ⊂ M. Lemma 18.22. Suppose that f : X → Y is a function and E ⊂ 2Y and A ⊂ Y
then
Definition 18.18. Let (X, M) and (Y, F) be measurable spaces. A function
σ f −1 (E) = f −1 (σ(E)) and

f : X → Y is measurable if f −1 (F) ⊂ M. We will also say that f is M/F – (18.3)
measurable or (M, F) – measurable. (σ(E))A = σ(EA ). (18.4)
Example 18.19 (Characteristic Functions). Let (X, M) be a measurable space (Similar assertion hold with σ (·) being replaced by A (·) .) Moreover, if F = σ(E)
and A ⊂ X. We define the characteristic function 1A : X → R by and M is a σ – algebra on X, then f is (M, F) – measurable iff f −1 (E) ⊂ M.
Proof. By Exercise 18.4, f −1 (σ(E)) is a σ – algebra and since E ⊂ F,

1 if x ∈ A
1A (x) =
0 if x ∈
/ A. f (E) ⊂ f −1 (σ(E)). It now follows that σ (f −1 (E)) ⊂ f −1 (σ (E)). For the
−1
reverse inclusion, notice that

If A ∈ M, then 1A is (M, 2R ) – measurable because 1−1
A (W ) is either ∅, X,
f∗ σ f −1 (E) = B ⊂ Y : f −1 (B) ∈ σ f −1 (E)

A or Ac for any W ⊂ R. Conversely, if F is any σ – algebra on R containing
a set W ⊂ R such that 1 ∈ W and 0 ∈ W c , then A ∈ M if 1A is (M, F) –
⊂ f∗ σ f −1 (E) . Hence if

is a σ – algebra which contains E and thus σ(E)
measurable. This is because A = 1−1
A (W ) ∈ M.
B ∈ σ(E) we know that f −1 (B) ∈ σ f −1 (E) , i.e. f −1 (σ(E)) ⊂ σ f −1 (E)

Exercise 18.4. Suppose f : X → Y is a function, F ⊂ 2Y and M ⊂ 2X . Show and Eq. (18.3) has been proved. Applying Eq. (18.3) with X = A and f = iA
f −1 F and f∗ M (see Notation 2.7) are algebras (σ – algebras) provided F and being the inclusion map implies
M are algebras (σ – algebras).
(σ(E))A = i−1 −1
A (σ(E)) = σ(iA (E)) = σ(EA ).
Remark 18.20. Let f : X → Y be a function. Given a σ – algebra F ⊂ 2Y ,
Lastly if f −1 E ⊂ M, then f −1 σ (E) = σ f −1 E ⊂ M which shows f is (M, F)

the σ – algebra M := f −1 (F) is the smallest σ – algebra on X such that f is
– measurable.
(M, F) - measurable . Similarly, if M is a σ - algebra on X then F = f∗ M is
the largest σ – algebra on Y such that f is (M, F) - measurable . Corollary 18.23. Suppose that (X, M) is a measurable space. Then the fol-
lowing conditions on a function f : X → R are equivalent:
Recall from Definition 2.8 that for E ⊂ 2X and A ⊂ X that
1. f is (M, BR ) – measurable,
EA = i−1
A (E) = {A ∩ E : E ∈ E} 2. f −1 ((a, ∞)) ∈ M for all a ∈ R,
3. f −1 ((a, ∞)) ∈ M for all a ∈ Q,
where iA : A → X is the inclusion map. Because of Exercise 13.3, when E = M 4. f −1 ((−∞, a]) ∈ M for all a ∈ R.
is an algebra (σ – algebra), MA is an algebra (σ – algebra) on A and we call
MA the relative or induced algebra (σ – algebra) on A. Proof. An exercise in using Lemma 18.22 and is the content of Exercise
The next two Lemmas are direct analogues of their topological counter parts 18.8.
in Lemmas 13.13 and 13.14. For completeness, the proofs will be given even Here is yet another way to generate σ – algebras. (Compare with the anal-
though they are same as those for Lemmas 13.13 and 13.14. ogous topological Definition 13.20.)

18.2 Measurable Functions 167
Definition 18.24 (σ – Algebras Generated by Functions). Let X be a set Proposition 18.29 (Localizing Measurability). Let (X, M) and (Y, F) be
and suppose there is a collection of measurable spaces {(Yα , Fα ) : α ∈ A} and measurable spaces and f : X → Y be a function.
functions fα : X → Yα for all α ∈ A. Let σ(fα : α ∈ A) denote the smallest σ
– algebra on X such that each fα is measurable, i.e. 1. If f is measurable and A ⊂ X then f |A : A → Y is measurable.
2. Suppose there exist An ∈ M such that X = ∪∞ n=1 An and f |An is MAn
σ(fα : α ∈ A) = σ(∪α fα−1 (Fα )). measurable for all n, then f is M – measurable.
Proposition 18.25. Assuming the notation in Definition 18.24 and addition- Proof. As the reader will notice, the proof given below is essentially iden-
ally let (Z, M) be a measurable space and g : Z → X be a function. Then g is tical to the proof of Proposition 13.19 which is the topological analogue of this
(M, σ(fα : α ∈ A)) – measurable iff fα ◦g is (M, Fα )–measurable for all α ∈ A. proposition. 1. If f : X → Y is measurable, f −1 (B) ∈ M for all B ∈ F and
therefore
Proof. This proof is essentially the same as the proof of the topological f |−1
A (B) = A ∩ f
−1
(B) ∈ MA for all B ∈ F.
analogue in Proposition 13.21. (⇒) If g is (M, σ(fα : α ∈ A)) – measurable, 2. If B ∈ F, then
then the composition fα ◦ g is (M, Fα ) – measurable by Lemma 18.21. (⇐) Let
−1
f −1 (B) = ∪∞ −1
(B) ∩ An = ∪∞

n=1 f n=1 f |An (B).
G = σ(fα : α ∈ A) = σ ∪α∈A fα−1 (Fα ) .

Since each An ∈ M, MAn ⊂ M and so the previous displayed equation shows

If fα ◦ g is (M, Fα ) – measurable for all α, then f −1 (B) ∈ M.
g −1 fα−1 (Fα ) ⊂ M ∀ α ∈ A Proposition 18.30. If (X, M) is a measurable space, then
and therefore f = (f1 , f2 , . . . , fn ) : X → Rn
g −1 ∪α∈A fα−1 (Fα ) = ∪α∈A g −1 fα−1 (Fα ) ⊂ M. is (M, BRn ) – measurable iff fi : X → R is (M, BR ) – measurable for each i.

In particular, a function f : X → C is (M, BC ) – measurable iff Re f and Im f
Hence are (M, BR ) – measurable.
g −1 (G) = g −1 σ ∪α∈A fα−1 (Fα ) = σ(g −1 ∪α∈A fα−1 (Fα ) ⊂ M

Proof. This is formally a consequence of Corollary 18.65 and Proposition
18.60 below. Nevertheless it is instructive to give a direct proof now. Let τ = τRn
which shows that g is (M, G) – measurable. denote the usual topology on Rn and πi : Rn → R be projection onto the ith
– factor. Since πi is continuous, πi is BRn /BR – measurable and therefore if
Definition 18.26. A function f : X → Y between two topological spaces is f : X → Rn is measurable then so is fi = πi ◦ f. Now suppose fi : X → R is
Borel measurable if f −1 (BY ) ⊂ BX . measurable for all i = 1, 2, . . . , n. Let
Proposition 18.27. Let X and Y be two topological spaces and f : X → Y be E := {(a, b) : a, b ∈ Qn 3 a < b} ,
a continuous function. Then f is Borel measurable.
where, for a, b ∈ Rn , we write a < b iff ai < bi for i = 1, 2, . . . , n and let
Proof. Using Lemma 18.22 and BY = σ(τY ),
(a, b) = (a1 , b1 ) × · · · × (an , bn ) .
f −1 (BY ) = f −1 (σ(τY )) = σ(f −1 (τY )) ⊂ σ(τX ) = BX .
Since E ⊂ τ and every element V ∈ τ may be written as a (necessarily) count-
able union of elements from E, we have σ (E) ⊂ BRn = σ (τ ) ⊂ σ (E) , i.e.
σ (E) = BRn . (This part of the proof is essentially a direct proof of Corollary
Definition 18.28. Given measurable spaces (X, M) and (Y, F) and a subset
18.65 below.) Because
A ⊂ X. We say a function f : A → Y is measurable iff f is MA /F – measur-
able. f −1 ((a, b)) = f1−1 ((a1 , b1 )) ∩ f2−1 ((a2 , b2 )) ∩ · · · ∩ fn−1 ((an , bn )) ∈ M

for all a, b ∈ Q with a < b, it follows that f −1 E ⊂ M and therefore Proof. Let us first observe that
f −1 BRn = f −1 σ (E) = σ f −1 E ⊂ M.

{−∞} = ∩∞ ∞ c
n=1 [−∞, −n) = ∩n=1 [−n, ∞] ∈ BR̄ ,
∞
{∞} = ∩n=1 [n, ∞] ∈ BR̄ and R = R̄\ {±∞} ∈ BR̄ .
Corollary 18.31. Let (X, M) be a measurable space and f, g : X → C be Letting i : R → R̄ be the inclusion map,
(M, BC ) – measurable functions. Then f ± g and f · g are also (M, BC ) –
i−1 (BR̄ ) = σ i−1 [a, ∞] : a ∈ R̄ = σ i−1 ([a, ∞]) : a ∈ R̄

measurable.
= σ [a, ∞] ∩ R : a ∈ R̄ = σ ({[a, ∞) : a ∈ R}) = BR .
Proof. Define F : X → C × C, A± : C × C → C and M : C × C −→ C by
F (x) = (f (x), g(x)), A± (w, z) = w ± z and M (w, z) = wz. Then A± and M Thus we have shown
are continuous and hence (BC2 , BC ) – measurable. Also F is (M, BC ⊗ BC ) =
(M, BC2 ) – measurable since π1 ◦ F = f and π2 ◦ F = g are (M, BC ) – measur- BR = i−1 (BR̄ ) = {A ∩ R : A ∈ BR̄ }.
able. Therefore A± ◦ F = f ± g and M ◦ F = f · g, being the composition of
This implies:
measurable functions, are also measurable.
Lemma 18.32. Let α ∈ C, (X, M) be a measurable space and f : X → C be a 1. A ∈ BR̄ =⇒ A ∩ R ∈BR and
(M, BC ) – measurable function. Then 2. if A ⊂ R̄ is such that A∩R ∈BR there exists B ∈ BR̄ such that A∩R = B∩R.
1 Because A∆B ⊂ {±∞} and {∞} , {−∞} ∈ BR̄ we may conclude that
if f (x) 6= 0 A ∈ BR̄ as well.
F (x) := f (x)
α if f (x) = 0
This proves Eq. (18.6).
is measurable. The proofs of the next two corollaries are left to the reader, see Exercises
18.5 and 18.6.
Proof. Define i : C → C by
1 Corollary 18.34. Let (X, M) be a measurable space and f : X → R̄ be a
z if z 6= 0 function. Then the following are equivalent
i(z) =
0 if z = 0.
1. f is (M, BR̄ ) - measurable,
For any open set V ⊂ C we have 2. f −1 ((a, ∞]) ∈ M for all a ∈ R,
i−1 (V ) = i−1 (V \ {0}) ∪ i−1 (V ∩ {0}) 3. f −1 ((−∞, a]) ∈ M for all a ∈ R,
4. f −1 ({−∞}) ∈ M, f −1 ({∞}) ∈ M and f 0 : X → R defined by
Because i is continuous except at z = 0, i−1 (V \ {0}) is an open set and hence
in BC . Moreover, i−1 (V ∩ {0}) ∈ BC since i−1 (V ∩ {0}) is either the empty 0 f (x) if f (x) ∈ R
f (x) := 1R (f (x)) =
set or the one point set {0} . Therefore i−1 (τC ) ⊂ BC and hence i−1 (BC ) = 0 if f (x) ∈ {±∞}
i−1 (σ(τC )) = σ(i−1 (τC )) ⊂ BC which shows that i is Borel measurable. Since
is measurable.
F = i ◦ f is the composition of measurable functions, F is also measurable.
We will often deal with functions f : X → R̄ = R∪ {±∞} . When talking Corollary 18.35. Let (X, M) be a measurable space, f, g : X → R̄ be functions
about measurability in this context we will refer to the σ – algebra on R̄ defined and define f · g : X → R̄ and (f + g) : X → R̄ using the conventions, 0 · ∞ = 0
by and (f + g) (x) = 0 if f (x) = ∞ and g (x) = −∞ or f (x) = −∞ and g (x) =
BR̄ := σ ({[a, ∞] : a ∈ R}) . (18.5) ∞. Then f · g and f + g are measurable functions on X if both f and g are
Proposition 18.33 (The Structure of BR̄ ). Let BR and BR̄ be as above, then measurable.
BR̄ = {A ⊂ R̄ : A ∩ R ∈BR }. (18.6) Exercise 18.5. Prove Corollary 18.34 noting that the equivalence of items 1. –
3. is a direct analogue of Corollary 18.23. Use Proposition 18.33 to handle item
In particular {∞} , {−∞} ∈ BR̄ and BR ⊂ BR̄ . 4.

18.3 σ – Function Algebras 169
Exercise 18.6. Prove Corollary 18.35. Proof. Let V ∈ τd and Wm := {y ∈ Y : dV c (y) > 1/m} for m = 1, 2, . . . .
Then Wm ∈ τd ,
Proposition 18.36 (Closure under sups, infs and limits). Suppose that
(X, M) is a measurable space and fj : (X, M) → R for j ∈ N is a sequence of Wm ⊂ W̄m ⊂ {y ∈ Y : dV c (y) ≥ 1/m} ⊂ V
M/BR – measurable functions. Then
for all m and Wm ↑ V as m → ∞. The proof will be completed by verifying the
supj fj , inf j fj , lim sup fj and lim inf fj identity,
j→∞ j→∞
f −1 (V ) = ∪∞ ∞ −1
m=1 ∪N =1 ∩n≥N fn (Wm ) ∈ M.
are all M/BR – measurable functions. (Note that this result is in generally false If x ∈ f −1 (V ) then f (x) ∈ V and hence f (x) ∈ Wm for some m. Since fn (x) →
when (X, M) is a topological space and measurable is replaced by continuous in f (x), fn (x) ∈ Wm for almost all n. That is x ∈ ∪∞ ∞ −1
m=1 ∪N =1 ∩n≥N fn (Wm ).
the statement.) ∞ ∞ −1
Conversely when x ∈ ∪m=1 ∪N =1 ∩n≥N fn (Wm ) there exists an m such that
fn (x) ∈ Wm ⊂ W̄m for almost all n. Since fn (x) → f (x) ∈ W̄m ⊂ V, it follows
Proof. Define g+ (x) := sup j fj (x), then that x ∈ f −1 (V ).
{x : g+ (x) ≤ a} = {x : fj (x) ≤ a ∀ j} Remark 18.40. In the previous Lemma 18.39 it is possible to let (Y, τ ) be any
= ∩j {x : fj (x) ≤ a} ∈ M topological space which has the “regularity” property that if V ∈ τ there exists
Wm ∈ τ such that Wm ⊂ W̄m ⊂ V and V = ∪∞ m=1 Wm . Moreover, some extra
so that g+ is measurable. Similarly if g− (x) = inf j fj (x) then condition is necessary on the topology τ in order for Lemma 18.39 to be correct.
For example if Y = {1, 2, 3} and τ = {Y, ∅, {1, 2}, {2, 3}, {2}} as in Example
{x : g− (x) ≥ a} = ∩j {x : fj (x) ≥ a} ∈ M. 13.36 and X = {a, b} with the trivial σ – algebra. Let fj (a) = fj (b) = 2 for
all j, then fj is constant and hence measurable. Let f (a) = 1 and f (b) = 2,
Since
then fj → f as j → ∞ with f being non-measurable. Notice that the Borel σ
lim sup fj = inf sup {fj : j ≥ n} and – algebra on Y is 2Y .
j→∞ n
lim inf fj = sup inf {fj : j ≥ n}

j→∞ n
18.3 σ – Function Algebras
we are done by what we have already proved.
In this subsection, we are going to relate σ – algebras of subsets of a set X to
Definition 18.37. Given a function f : X → R̄ let f+ (x) := max {f (x), 0} and certain algebras of functions on X. We will begin this endeavor after proving
f− (x) := max (−f (x), 0) = − min (f (x), 0) . Notice that f = f+ − f− . the simple but very useful approximation Theorem 18.42 below.
Corollary 18.38. Suppose (X, M) is a measurable space and f : X → R̄ is a Definition 18.41. Let (X, M) be a measurable space. A function φ : X → F
function. Then f is measurable iff f± are measurable. (F denotes either R, C or [0, ∞] ⊂ R̄) is a simple function if φ is M – BF
measurable and φ(X) contains only finitely many elements.
Proof. If f is measurable, then Proposition 18.36 implies f± are measurable.
Conversely if f± are measurable then so is f = f+ − f− . Any such simple functions can be written as
n
X
18.2.1 More general pointwise limits φ= λi 1Ai with Ai ∈ M and λi ∈ F. (18.7)
i=1
Lemma 18.39. Suppose that (X, M) is a measurable space, (Y, d) is a metric
space and fj : X → Y is (M, BY ) – measurable for all j. Also assume that for Indeed, take λ1 , λ2 , . . . , λn to be an enumeration of the range of φ and Ai =
each x ∈ X, f (x) = limn→∞ fn (x) exists. Then f : X → Y is also (M, BY ) – φ−1 ({λi }). Note that this argument shows that any simple function may be
measurable. written intrinsically as

X
φ= y1φ−1 ({y}) . (18.8) (2n , ∞] = (2n , 2n+1 ] ∪ (2n+1 , ∞],
y∈F
and so for x ∈ f −1 ((2n+1 , ∞]), φn (x) = 2n < 2n+1 = φn+1 (x) and for x ∈
The next theorem shows that simple functions are “pointwise dense” in the
f −1 ((2n , 2n+1 ]), φn+1 (x) ≥ 2n = φn (x). Therefore φn ≤ φn+1 for all n. It is
space of measurable functions.
clear by construction that φn (x) ≤ f (x) for all x and that 0 ≤ f (x) − φn (x) ≤
Theorem 18.42 (Approximation Theorem). Let f : X → [0, ∞] be mea- 2−n if x ∈ X2n . Hence we have shown that φn (x) ↑ f (x) for all x ∈ X and
surable and define, see Figure 18.2, φn ↑ f uniformly on bounded sets. For the second assertion, first assume that
f : X → R is a measurable function and choose φ± n to be simple functions such
22n −1
X k that φ± ↑ f± as n → ∞ and define φn = φ+
− φ−
n . Then
φn (x) := 1 −1 k k+1 (x) + 2n 1f −1 ((2n ,∞]) (x) n n
2n f (( 2n , 2n ]) − + −
k=0
|φn | = φ+
n + φn ≤ φn+1 + φn+1 = |φn+1 |
22n −1
X k n − −
= 1 k k+1 (x) + 2 1{f >2n } (x) and clearly |φn | = φ+ +
n + φn ↑ f+ + f− = |f | and φn = φn − φn → f+ − f− = f
2n { 2n <f ≤ 2n }
k=0 as n → ∞. Now suppose that f : X → C is measurable. We may now choose
then φn ≤ f for all n, φn (x) ↑ f (x) for all x ∈ X and φn ↑ f uniformly simple function un and vn such that |un | ↑ |Re f | , |vn | ↑ |Im f | , un → Re f and
on the sets XM := {x ∈ X : f (x) ≤ M } with M < ∞. Moreover, if f : X → vn → Im f as n → ∞. Let φn = un + ivn , then
C is a measurable function, then there exists simple functions φn such that 2 2 2 2
|φn | = u2n + vn2 ↑ |Re f | + |Im f | = |f |
limn→∞ φn (x) = f (x) for all x and |φn | ↑ |f | as n → ∞.
and φn = un + ivn → Re f + i Im f = f as n → ∞.
For the rest of this section let X be a given set.
Definition 18.43 (Bounded Convergence). We say that a sequence of

functions fn from X to R or C converges boundedly to a function f if
limn→∞ fn (x) = f (x) for all x ∈ X and
sup{|fn (x)| : x ∈ X and n = 1, 2, . . .} < ∞.
Definition 18.44. A function algebra H on X is a linear subspace of

`∞ (X, R) which contains 1 and is closed under pointwise multiplication, i.e.
H is a subalgebra of `∞ (X, R) which contains 1. If H is further closed under
bounded convergence then H is said to be a σ – function algebra.
Example 18.45. Suppose M is a σ – algebra on X, then
`∞ (M, R) := {f ∈ `∞ (X, R) : f is M/BR – measurable} (18.9)

Fig. 18.2. Constructing simple functions approximating a function, f : X → [0, ∞].
is a σ – function algebra. The next theorem will show that these are the only
example of σ – function algebras. (See Exercise 18.7 below for examples of
Proof. Since function algebras on X.)
k k+1 2k 2k + 1 2k + 1 2k + 2 Notation 18.46 If H ⊂ `∞ (X, R) be a function algebra, let
( , ] = ( n+1 , n+1 ] ∪ ( n+1 , n+1 ],
2n 2n 2 2 2 2
M (H) := {A ⊂ X : 1A ∈ H} . (18.10)
if x ∈ f −1 ( 2n+1 2k
, 2k+1 ] then φn (x) = φn+1 (x) = 2n+1 2k

and if x ∈
−1 2k+1 2k+2
2n+1 2k 2k+1
f ( 2n+1 , 2n+1 ] then φn (x) = 2n+1 < 2n+1 = φn+1 (x). Similarly Theorem 18.47. Let H be a σ – function algebra on a set X. Then

18.3 σ – Function Algebras 171
1. M (H) is a σ – algebra on X. M = {A ⊂ X : 1A ∈ `∞ (M, R)}

2. H = `∞ (M (H) , R) . = {A ⊂ X : 1A ∈ `∞ (F, R)} = F.
3. The map
M ∈ {σ – algebras on X} → `∞ (M, R) ∈ {σ – function algebras on X} Notation 18.48 Suppose M is a subset of `∞ (X, R) .

(18.11)
is bijective with inverse given by H → M (H) . 1. Let H (M ) denote the smallest subspace of `∞ (X, R) which contains M and
the constant functions and is closed under bounded convergence.
Proof. Let M := M (H) . 2. Let Hσ (M ) denote the smallest σ – function algebra containing M.
1. Since 0, 1 ∈ H, ∅, X ∈ M. If A ∈ M then, since H is a linear subspace of Theorem 18.49. Suppose M is a subset of `∞ (X, R) , then Hσ (M ) =
∞
`∞ (X, R) , 1Ac = 1 − 1A ∈ H which shows Ac ∈ M. If {An }n=1 ⊂ M, then `∞ (σ (M ) , R) or in other words the following diagram commutes:
since H is an algebra, M −→ σ (M )
t t
N
Y M ∈ {Multiplicative Subsets} −→ {σ – algebras} 3M
1∩N
n=1 An
= 1An =: fN ∈ H ↓ ↓ ↓ ↓
n=1
Hσ (M ) ∈ {σ– function algebras} = {σ– function algebras} 3 `∞ (M, R) .
for all N ∈ N. Because H is closed under bounded convergence it follows Proof. Since `∞ (σ (M ) , R) is σ – function algebra which contains M it
that follows that
1∩∞
n=1 An
= lim fN ∈ H Hσ (M ) ⊂ `∞ (σ (M ) , R) .
N →∞
and this implies ∩∞

n=1 An ∈ M. Hence we have shown M is a σ – algebra. For the opposite inclusion, let
2. Since H is an algebra, p (f ) ∈ H for any f ∈ H and any polynomial p on R. M = M (Hσ (M )) := {A ⊂ X : 1A ∈ Hσ (M )} .
The Weierstrass approximation Theorem 10.35, asserts that polynomials on
R are uniformly dense in the space of continuous functions on any compact By Theorem 18.47, M ⊂ Hσ (M ) = `∞ (M, R) which implies that every f ∈ M
subinterval of R. Hence if f ∈ H and φ ∈ C (R) , there exists polynomials is M – measurable. This then implies σ (M ) ⊂ M and therefore
pn on R such that pn ◦ f (x) converges to φ ◦ f (x) uniformly (and hence `∞ (σ (M ) , R) ⊂ `∞ (M, R) = Hσ (M ) .
boundedly) in x ∈ X as n → ∞. Therefore φ ◦ f ∈ H for all f ∈ H and
φ ∈ C (R) and in particular |f | ∈ H and f± := |f |±f 2 ∈ H if f ∈ H.
1/n
Fix an α ∈ R and for n ∈ N let φn (t) := (t − α)+ , where (t − α)+ := Definition 18.50 (Multiplicative System). A collection of bounded real or
max {t − α, 0} . Then φn ∈ C (R) and φn (t) → 1t>α as n → ∞ and the complex valued functions, M, on a set X is called a multiplicative system if
convergence is bounded when t is restricted to any compact subset of R. f · g ∈ M whenever f and g are in M.
Hence if f ∈ H it follows that 1f >α = limn→∞ φn (f ) ∈ H for all α ∈ R, i.e. Theorem 18.51 (Dynkin’s Multiplicative System Theorem). Suppose
{f > α} ∈ M for all α ∈ R. Therefore if f ∈ H then f ∈ `∞ (M, R) and M ⊂ `∞ (X, R) is a multiplicative system, then
we have shown H ⊂ `∞ (M, R) . Conversely if f ∈ `∞ (M, R) , then for any
α < β, {α < f ≤ β} ∈ M = M (H) and so by assumption 1{α<f ≤β} ∈ H. H (M ) = Hσ (M ) = `∞ (σ (M ) , R) . (18.12)
Combining this remark with the approximation Theorem 18.42 and the fact This can also be stated as follows.
that H is closed under bounded convergence shows that f ∈ H. Hence we Suppose H is a linear subspace of `∞ (X, R) such that: 1 ∈ H, H is closed
have shown `∞ (M, R) ⊂ H which combined with H ⊂ `∞ (M, R) already under bounded convergence, and M ⊂ H. Then H contains all bounded real
proved shows H = `∞ (M (H) , R) . valued σ(M )-measurable functions, i.e. `∞ (σ (M ) , R) ⊂ H.
3. Items 1. and 2. shows the map in Eq. (18.11) is surjective. To see the (In words, the smallest subspace of bounded real valued functions on X which
map is injective suppose M and F are two σ – algebras on X such that contains M that is closed under bounded convergence is the same as the space
`∞ (M, R) = `∞ (F, R) , then of bounded real valued σ (M ) – measurable functions on X.)

Proof. The assertion that Hσ (M ) = `∞ (σ (M ) , R) has already been proved Definition 18.53. A collection of subsets, C, of X is a multiplicative
in Theorem 18.49. Since any σ – function algebra containing M is also a class(or a π – class) if C is closed under finite intersections.
subspace of `∞ (X, R) which contains the constant functions and is closed
under bounded convergence (compare with Exercise 18.13), it follows that Corollary 18.54. Suppose H is a subspace of `∞ (X, R) which is closed under
H (M ) ⊂ Hσ (M ) . To complete the proof it suffices to show the inclusion, bounded convergence and 1 ∈ H. If C ⊂ 2X is a multiplicative class such that
H (M ) ⊂ Hσ (M ) , is an equality. We will accomplish this below by showing 1A ∈ H for all A ∈ C, then H contains all bounded σ(C) – measurable functions.
H (M ) is also a σ – function algebra. Proof. Let M = {1}∪{1A : A ∈ C} . Then M ⊂ H is a multiplicative system
For any f ∈ H := H (M ) let and the proof is completed with an application of Theorem 18.51.
Hf := {g ∈ H : f g ∈ H} ⊂ H
Corollary 18.55. Suppose that (X, d) is a metric space and BX = σ(τd ) is the
∞
and notice that Hf is a linear subspace of ` (X, R) which is closed under Borel σ – algebra on X and H is a subspace of `∞ (X, R) such that BC(X, R) ⊂
bounded convergence. Moreover if f ∈ M, M ⊂ Hf since M is multiplicative. H and H is closed under bounded convergence1 . Then H contains all bounded
Therefore Hf = H and we have shown that f g ∈ H whenever f ∈ M and BX – measurable real valued functions on X. (This may be stated as follows:
g ∈ H. Given this it now follows that M ⊂ Hf for any f ∈ H and by the the smallest vector space of bounded functions which is closed under bounded
same reasoning just used, Hf = H. Since f ∈ H is arbitrary, we have shown convergence and contains BC(X, R) is the space of bounded BX – measurable
f g ∈ H for all f, g ∈ H, i.e. H is an algebra, which by the definition of H (M ) real valued functions on X.)
in Notation 18.48 contains the constant functions, i.e. H (M ) is a σ – function
Proof. Let V ∈ τd be an open subset of X and for n ∈ N let
algebra.
Theorem 18.52 (Complex Multiplicative System Theorem). Suppose H fn (x) := min(n · dV c (x), 1) for all x ∈ X.
is a complex linear subspace of `∞ (X, C) such that: 1 ∈ H, H is closed under
complex conjugation, and H is closed under bounded convergence. If M ⊂ H Notice that fn = φn ◦ dV c where φn (t) = min(nt, 1) (see Figure 18.3) which
is multiplicative system which is closed under conjugation, then H contains all is continuous and hence fn ∈ BC(X, R) for all n. Furthermore, fn converges
bounded complex valued σ(M )-measurable functions, i.e. `∞ (σ (M ) , C) ⊂ H. boundedly to 1dV c >0 = 1V as n → ∞ and therefore 1V ∈ H for all V ∈ τ. Since
τ is a π – class, the result now follows by an application of Corollary 18.54.
Proof. Let M0 = spanC (M ∪ {1}) be the complex span of M. As the
reader should verify, M0 is an algebra, M0 ⊂ H, M0 is closed under com-
plex conjugation and that σ (M0 ) = σ (M ) . Let HR := H ∩ `∞ (X, R) and
M0R = M ∩ `∞ (X, R). Then (you verify) M0R is a multiplicative system,
M0R ⊂ HR and HR is a linear space containing 1 which isclosed
under bounded
convergence. Therefore by Theorem 18.51, `∞ σ M0R , R ⊂ HR . Since H
and M0 are complex linear spaces closed under complex conjugation, for any
f ∈ H or f ∈ M0 , the functions Re f = 21 f + f¯ and Im f = 2i 1
f − f¯ are

in H (M 0 ) or M0 respectively. Therefore H = H + iH , M0 = M0 + iM0 ,

R R R R
R
σ M0 = σ (M0 ) = σ (M ) and
`∞ (σ (M ) , C) = `∞ σ M0R , R + i`∞ σ M0R , R

⊂ HR + iHR = H. Fig. 18.3. Plots of φ1 , φ2 and φ3 .
Exercise 18.7 (Algebra analogue of Theorem 18.47). Call a function al-

gebra H ⊂ `∞ (X, R) a simple function algebra if the range of each function
Here are some more variants of Corollary 18.55.
f ∈ H is a finite subset of R. Prove there is a one to one correspondence between
1
algebras A on a set X and simple function algebras H on X. Recall that BC(X, R) are the bounded continuous functions on X.

18.4 Product σ – Algebras 173
∞
Proposition 18.56. Let (X, d) be a metric space, BX = σ(τd ) be the Borel σ 2. Let {Kn }n=1 be compact subsets of F c such that Kn ↑ F c as n → ∞ and
– algebra and assume there exists compact sets Kk ⊂ X such that Kko ↑ X. set Vn := Knc = X \ Kn . Then Vn ↓ F and by Proposition 15.5, Vn is open
Suppose that H is a subspace of `∞ (X, R) such that Cc (X, R) ⊂ H (Cc (X, R) is for each n.
∞
the space of continuous functions with compact support) and H is closed under 3. Let U ⊂ X be an open set and {Kn }n=1 be compact subsets of U such that
bounded convergence. Then H contains all bounded BX – measurable real valued Kn ↑ U. By Urysohn’s Lemma 15.8, there exist fn ≺ U such that fn = 1
functions on X. on Kn . These functions satisfy, 1U = limn→∞ fn .
4. By item 3., 1U is σ(Cc (X, R)) – measurable for all U ∈ τ and hence τ ⊂
Proof. Let k and n be positive integers and set ψn,k (x) = min(1, n ·
σ(Cc (X, R)). Therefore BX = σ(τ ) ⊂ σ(Cc (X, R)). The converse inclusion
d(K o )c (x)). Then ψn,k ∈ Cc (X, R) and {ψn,k 6= 0} ⊂ Kko . Let Hn,k denote
k holds because continuous functions are always Borel measurable.
those bounded BX – measurable functions, f : X → R, such that ψn,k f ∈ H.
It is easily seen that Hn,k is closed under bounded convergence and that Hn,k
contains BC(X, R) and therefore by Corollary 18.55, ψn,k f ∈ H for all bounded Here is a variant of Corollary 18.55.
measurable functions f : X → R. Since ψn,k f → 1Kko f boundedly as n → ∞,
1Kko f ∈ H for all k and similarly 1Kko f → f boundedly as k → ∞ and therefore Corollary 18.58. Suppose that (X, τ ) is a second countable locally compact
f ∈ H. Hausdorff space and BX = σ(τ ) is the Borel σ – algebra on X. If H is a
subspace of `∞ (X, R) which is closed under bounded convergence and contains
Lemma 18.57. Suppose that (X, τ ) is a locally compact second countable Haus- Cc (X, R), then H contains all bounded BX – measurable real valued functions
dorff space.2 Then: on X.
1. every open subset U ⊂ X is σ – compact. In fact U is still a locally compact Proof. By Item 3. of Lemma 18.57, for every U ∈ τ the characteristic
second countable Hausdorff space. function, 1U , may be written as a bounded pointwise limit of functions from
2. If F ⊂ X is a closed set, there exist open sets Vn ⊂ X such that Vn ↓ F as Cc (X, R) . Therefore 1U ∈ H for all U ∈ τ. Since τ is a π – class, the proof is
n → ∞. finished with an application of Corollary 18.54
3. To each open set U ⊂ X there exists fn ≺ U (i.e. fn ∈ Cc (U, [0, 1])) such
that limn→∞ fn = 1U .
4. BX = σ(Cc (X, R)), i.e. the σ – algebra generated by Cc (X) is the Borel σ
– algebra on X.
18.4 Product σ – Algebras
Q
Let {(Xα , Mα )}α∈A be a collection of measurable spaces X = XA = Xα
Proof. α∈A
and πα : XA → Xα be the canonical projection map as in Notation 2.2.
1. Let U be an open subset of X, V be a countable base for τ and
Definition 18.59 (Product σ – Algebra). The product σ – algebra,
V U := {W ∈ V : W̄ ⊂ U and W̄ is compact}.
⊗α∈A Mα , is the smallest σ – algebra on X such that each πα for α ∈ A is
For each x ∈ U, by Proposition 15.7, there exists an open neighborhood V measurable, i.e.
of x such that V̄ ⊂ U and V̄ is compact. Since V is a base for the topology
⊗α∈A Mα := σ(πα : α ∈ A) = σ ∪α πα−1 (Mα ) .

τ, there exists W ∈ V such that x ∈ W ⊂ V. Because W̄ ⊂ V̄ , it follows
that W̄ is compact and hence W ∈ V U . As x ∈ U was arbitrary, U = ∪V U .
This shows V U is a countable basis for the topology on U and that U is still Applying Proposition 18.25 in this setting implies the following proposition.
locally compact. Proposition 18.60. Suppose Y is a measurable space and f : Y → X = XA
∞
Let {Wn }n=1 be an enumeration of V U and set Kn := ∪nk=1 W̄k . Then is a map. Then f is measurable iff πα ◦ f : Y → Xα is measurable for all
Kn ↑ U as n → ∞ and Kn is compact for each n. This shows U is σ – α ∈ A. In particular if A = {1, 2, . . . , n} so that X = X1 × X2 × · · · × Xn and
compact. (See Exercise 14.7.) f (y) = (f1 (y), f2 (y), . . . , fn (y)) ∈ X1 × X2 × · · · × Xn , then f : Y → XA is
2
For example any separable locally compact metric space and in particular any open measurable iff fi : Y → Xi is measurable for all i.
subset of Rn .

Proposition 18.61. Suppose that (Xα , Mα )α∈A is a collection of measurable Remark 18.62. One can not relax the assumption that Xα ∈ Eα in the second
spaces and Eα ⊂ Mα generates Mα for each α ∈ A, then part of Proposition 18.61. For example, if X1 = X2 = {1, 2} and E1 = E2 =
{{1}} , then σ(E1 ×E2 ) = {∅, X1 ×X2 , {(1, 1)}} while σ(σ(E1 )×σ(E2 )) = 2X1 ×X2 .
⊗α∈A Mα = σ ∪α∈A πα−1 (Eα )

(18.13)
Theorem 18.63. Let {Xα }α∈A be a sequence of sets where A is at most count-
Moreover, suppose that A is either finite or countably infinite, Xα ∈ Eα for each able. Suppose for each α ∈ A we are given a countable set Eα ⊂ 2Xα . Let
α ∈ A, and Mα = σ(Eα ) for each α ∈ A. Then the product σ – algebra satisfies τQα = τ (Eα ) be the topology on Xα generated by Eα and X be the product space
α∈A Xα equipped with the product topology τ := ⊗α∈A τ (Eα ). Then the Borel
( )!
Y
⊗α∈A Mα = σ Eα : Eα ∈ Eα for all α ∈ A . (18.14) σ – algebra BX = σ(τ ) is the same as the product σ – algebra:
α∈A
BX = ⊗α∈A BXα ,
In particular if A = {1, 2, . . . , n} , then X = X1 × X2 × · · · × Xn and
where BXα = σ(τ (Eα )) = σ(Eα ) for all α ∈ A.
M1 ⊗ M2 ⊗ · · · ⊗ Mn = σ(M1 × M2 × · · · × Mn ), In particular if A = {1, 2, . . . , n} and each (Xi , τi ) is a second countable
where M1 × M2 × · · · × Mn is as defined in Notation 13.26. topological space, then
Proof. Since ∪α πα−1 (Eα ) ⊂ ∪α πα−1 (Mα ), it follows that BX := σ(τ1 ⊗ τ2 ⊗ · · · ⊗ τn ) = σ(BX1 × · · · × BXn ) =: BX1 ⊗ · · · ⊗ BXn .
F := σ ∪α πα−1 (Eα ) ⊂ σ ∪α πα−1 (Mα ) = ⊗α∈A Mα . Proof. By Proposition 13.25, the topology τ may be described as the small-

est topology containing E = ∪α∈A πα−1 (Eα ). Now E is the countable union of
Conversely, countable sets so is still countable. Therefore by Proposition 18.17 and Propo-
F ⊃ σ(πα−1 (Eα )) = πα−1 (σ(Eα )) = πα−1 (Mα ) sition 18.61,
holds for all α implies that
BX = σ(τ ) = σ(τ (E)) = σ(E) = ⊗α∈A σ(Eα )
∪α πα−1 (Mα ) ⊂ F = ⊗α∈A σ(τα ) = ⊗α∈A BXα .
and hence that ⊗α∈A Mα ⊂ F. We now prove Eq. (18.14). Since we are assum-
ing that Xα ∈ Eα for each α ∈ A, we see that
( ) Corollary 18.64. If (Xi , di ) are separable metric spaces for i = 1, . . . , n, then
Y
∪α πα−1 (Eα ) ⊂ Eα : Eα ∈ Eα for all α ∈ A BX1 ⊗ · · · ⊗ BXn = B(X1 ×···×Xn )
α∈A
where BXi is the Borel σ – algebra on Xi and B(X1 ×···×Xn ) is the Borel σ – alge-
and therefore by Eq. (18.13) × · · · × Xn equipped with the metric topology associated to the metric
bra on X1P
n
(
Y
)! d(x, y) = i=1 di (xi , yi ) where x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ).
⊗α∈A Mα = σ ∪α πα−1 (Eα ) ⊂ σ

Eα : Eα ∈ Eα for all α ∈ A . Proof. This is a combination of the results in Lemma 13.28, Exercise 13.12
α∈A
and Theorem 18.63.
This last statement is true independent as to whether A is countable or not. Because all norms on finite dimensional spaces are equivalent, the usual
For the reverse inclusion it suffices to notice that since A is countable, Euclidean norm on Rm × Rn is equivalent to the “product” norm defined by
Y
Eα = ∩α∈A πα−1 (Eα ) ∈ ⊗α∈A Mα k(x, y)kRm ×Rn = kxkRm + kykRn .
α∈A
Hence by Lemma 13.28, the Euclidean topology on Rm+n is the same as the
and hence product topology on Rm+n ∼
= Rm × Rn . Here we are identifying Rm × Rn with
m+n
by the map
( )!
Y R
σ Eα : Eα ∈ Eα for all α ∈ A ⊂ ⊗α∈A Mα .
α∈A
(x, y) ∈ Rm × Rn → (x1 , . . . , xm , y1 , . . . , yn ) ∈ Rm+n .
These comments along with Corollary 18.64 proves the following result.

18.5 Exercises 175
Corollary 18.65. After identifying Rm × Rn with Rm+n as above and letting Exercise 18.11. Show that every monotone function f : R → R is (BR , BR ) –
BRn denote the Borel σ –algebra on Rn , we have measurable.
n–times Exercise 18.12. Show by example that the supremum of an uncountable fam-
z }| {
BRm+n = BRn ⊗ BRm and BRn = BR ⊗ · · · ⊗ BR . ily of measurable functions need not be measurable. (Folland problem 2.6 on p.
48.)
18.4.1 Factoring of Measurable Maps Exercise 18.13. Let X = {1, 2, 3, 4} , A = {1, 2} , B = {2, 3} and M :=
{1A , 1B } . Show Hσ (M ) 6= H (M ) in this case.
Lemma 18.66. Suppose that (Y, F) is a measurable space and F : X → Y is
a map. Then to every (σ(F ), BR̄ ) – measurable function, H : X → R̄, there is
a (F, BR̄ ) – measurable function h : Y → R̄ such that H = h ◦ F.
Proof. First suppose that H = 1A where A ∈ σ(F ) = F −1 (F). Let B ∈ F

such that A = F −1 (B) then 1A = 1F −1 (B) = 1B ◦ F andPhence the Lemma
is valid in this case with h = 1B . More generally if H = ai 1Ai is a simple
function, then
P there exists B i ∈ F such that 1 Ai = 1Bi ◦ F and hence H = h ◦ F
with h := ai 1Bi – a simple function on R̄. For general (σ(F ), F) – measurable
function, H, from X → R̄, choose simple functions Hn converging to H. Let hn
be simple functions on R̄ such that Hn = hn ◦ F. Then it follows that
H = lim Hn = lim sup Hn = lim sup hn ◦ F = h ◦ F

n→∞ n→∞ n→∞
where h := lim supn→∞ hn – a measurable function from Y to R̄.

The following is an immediate corollary of Proposition 18.25 and Lemma
18.66.
Corollary 18.67. Let X and A be sets, and suppose for α ∈ A we are Q give a
measurable space (Yα , Fα ) and a function fα : X → Yα . Let Y := α∈A Yα ,
F := ⊗α∈A Fα be the product σ – algebra on Y and M := σ(fα : α ∈ A) be the
smallest σ – algebra on X such that each fα is measurable. Then the function
F : X → Y defined by [F (x)]α := fα (x) for each α ∈ A is (M, F) – measurable
and a function H : X → R̄ is (M, BR̄ ) – measurable iff there exists a (F, BR̄ ) –
measurable function h from Y to R̄ such that H = h ◦ F.
18.5 Exercises
Exercise 18.8. Prove Corollary 18.23. Hint: See Exercise 18.3.
Exercise 18.9. If M is the σ – algebra generated by E ⊂ 2X , then M is the

union of the σ – algebras generated by countable subsets F ⊂ E.
Exercise 18.10. Let (X, M) be a measure space and fn : X → F be a sequence

of measurable functions on X. Show that {x : limn→∞ fn (x) exists in F} ∈ M.

19
Measures and Integration
Definition 19.1. A measure µ on a measurable space (X, M) is a function Proposition 19.3 (Basic properties of measures). Suppose that (X, M, µ)
∞
µ : M → [0, ∞] such that is a measure space and E, F ∈ M and {Ej }j=1 ⊂ M, then :
1. µ(∅) = 0 and 1. µ(E) ≤ µ(F
n P) if E ⊂ F.
2. (Finite Additivity) If {Ai }i=1 ⊂ M are pairwise disjoint, i.e. Ai ∩ Aj = ∅ 2. µ(∪Ej ) ≤ µ(Ej ).
when i 6= j, then 3. If µ(E1 ) < ∞ and Ej ↓ E, i.e. E1 ⊃ E2 ⊃ E3 ⊃ . . . and E = ∩j Ej , then
n n
[ X µ(Ej ) ↓ µ(E) as j → ∞.
µ( Ai ) = µ(Ai ).
i=1 i=1 Proof.
3. (Continuity) If An ∈ M and An ↑ A, then µ(An ) ↑ µ(A).
1. Since F = E ∪ (F \ E),
We call a triple (X, M, µ), where (X, M) is a measurable space and µ :
M → [0, ∞] is a measure, a measure space. µ(F ) = µ(E) + µ(F \ E) ≥ µ(E).
Remark 19.2. Properties 2) and 3) in Definition 19.1 are equivalent to the fol- ej = Ej \ (E1 ∪ · · · ∪ Ej−1 ) so that the Ẽj ’s are pair-wise disjoint and
2. Let E
∞
lowing condition. If {Ai }i=1 ⊂ M are pairwise disjoint then E = ∪Eej . Since Ẽj ⊂ Ej it follows from Remark 19.2 and part (1), that
∞
[ ∞
X X X
µ( Ai ) = µ(Ai ). (19.1) µ(E) = ej ) ≤
µ(E µ(Ej ).
i=1 i=1
3. Define Di := E1 \ Ei then Di ↑ E1 \ E which implies that
To prove this assume that Properties 2) and 3) in Definition 19.1 hold and
n ∞
∞ S S
µ(E1 ) − µ(E) = lim µ(Di ) = µ(E1 ) − lim µ(Ei )
{Ai }i=1 ⊂ M are pairwise disjoint. Letting Bn := Ai ↑ B := Ai , we i→∞ i→∞
i=1 i=1
have
which shows that limi→∞ µ(Ei ) = µ(E).
∞ n ∞
[ (3) (2) X X
µ( Ai ) = µ(B) = lim µ(Bn ) = lim µ(Ai ) = µ(Ai ).
n→∞ n→∞
i=1 i=1 i=1
Definition 19.4. A set E ⊂ X is a null set if E ∈ M and µ(E) = 0. If P is
Conversely, if Eq. (19.1) holds we may take Aj = ∅ for all j > n to see that some “property” which is either true or false for each x ∈ X, we will use the
Property 2) of Definition 19.1 holds. Also if An ↑ A, let Bn := An \ An−1 with terminology P a.e. (to be read P almost everywhere) to mean
∞
A0 := ∅. Then {Bn }n=1 are pairwise disjoint, An = ∪nj=1 Bj and A = ∪∞ j=1 Bj .
So if Eq. (19.1) holds we have E := {x ∈ X : P is false for x}
∞
X is a null set. For example if f and g are two measurable functions on (X, M, µ),
µ(A) = µ ∪∞
j=1 Bj = µ(Bj ) f = g a.e. means that µ(f 6= g) = 0.
j=1
n
X Definition 19.5. A measure space (X, M, µ) is complete if every subset of a
= lim µ(Bj ) = lim µ(∪nj=1 Bj ) = lim µ(An ). null set is in M, i.e. for all F ⊂ X such that F ⊂ E ∈ M with µ(E) = 0
n→∞ n→∞ n→∞
j=1 implies that F ∈ M.
178 19 Measures and Integration
Proposition 19.6 (Completion of a Measure). Let (X, M, µ) be a measure Many theorems in the sequel will require some control on the size of a
space. Set measure µ. The relevant notion for our purposes (and most purposes) is that
of a σ – finite measure defined next.
N = N µ := {N ⊂ X : ∃ F ∈ M such that N ⊂ F and µ(F ) = 0} ,
M̄ = M̄µ := {A ∪ N : A ∈ M and N ∈ N } and Definition 19.7. Suppose X is a set, E ⊂ M ⊂ 2X and µ : M → [0, ∞] is
a function. The function µ is σ – finite on E if there exists En ∈ E such that
µ̄(A ∪ N ) := µ(A) for A ∈ M and N ∈ N ,
µ(En ) < ∞ and X = ∪∞ n=1 En . If M is a σ – algebra and µ is a measure on M
see Fig. 19.1. Then M̄ is a σ – algebra, µ̄ is a well defined measure on M̄, µ̄ which is σ – finite on M we will say (X, M, µ) is a σ – finite measure space.
is the unique measure on M̄ which extends µ on M, and (X, M̄, µ̄) is complete The reader should check that if µ is a finitely additive measure on an algebra,
measure space. The σ-algebra, M̄, is called the completion of M relative to µ M, then µ is σ – finite on M iff there exists Xn ∈ M such that Xn ↑ X and
and µ̄, is called the completion of µ. µ(Xn ) < ∞.
Proof. Clearly X, ∅ ∈ M̄. Let A ∈ M and N ∈ N and choose F ∈ M such
19.1 Example of Measures
Most σ – algebras and σ -additive measures are somewhat difficult to describe

and define. However, one special case is fairly easy to understand. Namely
suppose that F ⊂ 2X is a countable or finite partition of X and M ⊂ 2X is the
σ – algebra which consists of the collection of sets A ⊂ X such that
A = ∪ {α ∈ F : α ⊂ A} . (19.2)
It is easily seen that M is a σ – algebra.

Any measure µ : M → [0, ∞] is determined uniquely by its values on F.
Fig. 19.1. Completing a σ – algebra.
Conversely, if we are given any function λ : F → [0, ∞] we may define, for
A ∈ M, X X
µ(A) = λ(α) = λ(α)1α⊂A
that N ⊂ F and µ(F ) = 0. Since N c = (F \ N ) ∪ F c , α∈F 3α⊂A α∈F
c c c c
(A ∪ N ) = A ∩ N = A ∩ (F \ N ∪ F ) c where 1α⊂A is one if α ⊂ A and `∞zero otherwise. We may check that µ is a
measure on M. Indeed, if A = i=1 Ai and α ∈ F, then α ⊂ A iff α ⊂ Ai for
= [Ac ∩ (F \ N )] ∪ [Ac ∩ F c ] P∞
one and hence exactly one Ai . Therefore 1α⊂A = i=1 1α⊂Ai and hence
where [Ac ∩ (F \ N )] ∈ N and [Ac ∩ F c ] ∈ M. Thus M̄ is closed under X X ∞
X
complements. If Ai ∈ M and Ni ⊂ Fi ∈ M such that µ(Fi ) = 0 then µ(A) = λ(α)1α⊂A = λ(α) 1α⊂Ai
∪(Ai ∪ NiP) = (∪Ai ) ∪ (∪Ni ) ∈ M̄ since ∪Ai ∈ M and ∪Ni ⊂ ∪Fi and α∈F α∈F i=1
µ(∪Fi ) ≤ µ(Fi ) = 0. Therefore, M̄ is a σ – algebra. Suppose A∪N1 = B ∪N2 X∞ X ∞
X
with A, B ∈ M and N1 , N2 , ∈ N . Then A ⊂ A ∪ N1 ⊂ A ∪ N1 ∪ F2 = B ∪ F2 = λ(α)1α⊂Ai = µ(Ai )
which shows that i=1 α∈F i=1
µ(A) ≤ µ(B) + µ(F2 ) = µ(B).
as desired. Thus we have shown that there is a one to one correspondence
Similarly, we show that µ(B) ≤ µ(A) so that µ(A) = µ(B) and hence µ̄(A ∪ between measures µ on M and functions λ : F → [0, ∞].
N ) := µ(A) is well defined. It is left as an exercise to show µ̄ is a measure, i.e. The construction of measures will be covered in Chapters 31 – 32 below.
that it is countable additive. However, let us record here the existence of an interesting class of measures.

19.2 Integrals of Simple functions 179
Theorem 19.8. To every right continuous non-decreasing function F : R → R That is to say

there exists a unique measure µF on BR such that 1
m((0, ]) = 1/n.
n
µF ((a, b]) = F (b) − F (a) ∀ − ∞ < a ≤ b < ∞ (19.3)
Similarly, m((0, nl ]) = l/n for all l, n ∈ N and therefore by the translation
Moreover, if A ∈ BR then invariance of m,
(∞ )
X m((a, b]) = b − a for all a, b ∈ Q with a < b.
µF (A) = inf (F (bi ) − F (ai )) : A ⊂ ∪∞
i=1 (ai , bi ] (19.4)
i=1
(∞ Finally for a, b ∈ R such that a < b, choose an , bn ∈ Q such that bn ↓ b and
∞
)
X a an ↑ a, then (an , bn ] ↓ (a, b] and thus
= inf (F (bi ) − F (ai )) : A ⊂ (ai , bi ] . (19.5)
i=1 i=1
m((a, b]) = lim m((an , bn ]) = lim (bn − an ) = b − a,
n→∞ n→∞
In fact the map F → µF is a one to one correspondence between right continuous
functions F with F (0) = 0 on one hand and measures µ on BR such that i.e. m is Lebesgue measure. To prove Eq. (19.7) we may assume that λ 6= 0
µ(J) < ∞ on any bounded set J ∈ BR on the other. −1
since this case is trivial to prove. Now let mλ (B) := |λ| m(λB). It is easily
Proof. See Section 28.3 below or Theorem 28.40 below. checked that mλ is again a measure on BR which satisfies
Example 19.9. The most important special case of Theorem 19.8 is when F (x) = mλ ((a, b]) = λ−1 m ((λa, λb]) = λ−1 (λb − λa) = b − a
x, in which case we write m for µF . The measure m is called Lebesgue measure.
if λ > 0 and
Theorem 19.10. Lebesgue measure m is invariant under translations, i.e. for
−1 −1
B ∈ BR and x ∈ R, mλ ((a, b]) = |λ| m ([λb, λa)) = − |λ| (λb − λa) = b − a
m(x + B) = m(B). (19.6)
if λ < 0. Hence mλ = m.
Moreover, m is the unique measure on BR such that m((0, 1]) = 1 and Eq.
We are now going to develop integration theory relative to a measure. The
(19.6) holds for B ∈ BR and x ∈ R. Moreover, m has the scaling property
integral defined in the case for Lebesgue measure, m, will be an extension of
m(λB) = |λ| m(B) (19.7) the standard Riemann integral on R.
where λ ∈ R, B ∈ BR and λB := {λx : x ∈ B}.

19.1.1 ADD: Examples of Measures
Proof. Let mx (B) := m(x + B), then one easily shows that mx is a measure
on BR such that mx ((a, b]) = b − a for all a < b. Therefore, mx = m by the BRUCE: ADD details.
uniqueness assertion in Theorem 19.8. For the converse, suppose that m is
translation invariant and m((0, 1]) = 1. Given n ∈ N, we have 1. Product measure for the flipping of a coin.
2. Haar Measure
k−1 k k−1 1 3. Measure on embedded submanifolds, i.e. Hausdorff measure.
(0, 1] = ∪nk=1 ( , ] = ∪nk=1 + (0, ] .
n n n n 4. Wiener measure.
5. Gibbs states.
Therefore, 6. Measure associated to self-adjoint operators and classifying them.
n
X k−1 1
1 = m((0, 1]) = m + (0, ]
n n
k=1
n
19.2 Integrals of Simple functions
X 1 1
= m((0, ]) = n · m((0, ]).
n n Let (X, M, µ) be a fixed measure space in this section.
k=1

X
Definition 19.11. Let F = C or [0, ∞) and suppose that φ : X → F is a simple Iµ (φ + ψ) = z µ(φ + ψ = z)
function as in Definition 18.41. If F = C assume further that µ(φ−1 ({y})) < ∞ z∈F
for all y 6= 0 in C. For such functions φ, define Iµ (φ) by =
X
z µ (∪w∈F {φ = w, ψ = z − w})
X z∈F
Iµ (φ) = yµ(φ−1 ({y})). X X
y∈F
= z µ(φ = w, ψ = z − w)
z∈F w∈F
Proposition 19.12. Let λ ∈ F and φ and ψ be two simple functions, then Iµ
X
= (z + w)µ(φ = w, ψ = z)
satisfies: z,w∈F
X X
1. = z µ(ψ = z) + w µ(φ = w)
Iµ (λφ) = λIµ (φ). (19.8) z∈F w∈F
2. = Iµ (ψ) + Iµ (φ).
Iµ (φ + ψ) = Iµ (ψ) + Iµ (φ). which proves 2. For 3. if φ and ψ are non-negative simple functions such that
3. If φ and ψ are non-negative simple functions such that φ ≤ ψ then φ≤ψ
X X
Iµ (φ) ≤ Iµ (ψ). Iµ (φ) = aµ(φ = a) = aµ(φ = a, ψ = b)
a≥0 a,b≥0
−1
Proof. Let us write {φ = y} for the set φ ({y}) ⊂ X and µ(φ = y) for
X X
≤ bµ(φ = a, ψ = b) = bµ(ψ = b) = Iµ (ψ),
µ({φ = y}) = µ(φ−1 ({y})) so that a,b≥0 b≥0
wherein the third inequality we have used {φ = a, ψ = b} = ∅ if a > b.

X
Iµ (φ) = yµ(φ = y).
y∈F
We will also write {φ = a, ψ = b} for φ−1 ({a})∩ψ −1 ({b}). This notation is more 19.3 Integrals of positive functions
intuitive for the purposes of this proof. Suppose that λ ∈ F then
X X Definition 19.13. Let L+ = L+ (M) = {f : X → [0, ∞] : f is measurable}.
Iµ (λφ) = y µ(λφ = y) = y µ(φ = y/λ) Define
y∈F y∈F
Z Z
X f (x) dµ (x) = f dµ := sup {Iµ (φ) : φ is simple and φ ≤ f } .
= λz µ(φ = z) = λIµ (φ) X X
z∈F
We say the f ∈ L+ is integrable if X f dµ < ∞. If A ∈ M, let
R
provided that λ 6= 0. The case λ = 0 is clear, so we have proved 1. Suppose that Z Z Z

φ and ψ are two simple functions, then f (x) dµ (x) = f dµ := 1A f dµ.
A A X
Remark 19.14. Because

R of item 3. of Proposition
R 19.12, if φ is a non-negative
simple function, X φdµ = Iµ (φ) so that X is an extension of Iµ . This extension
still has the monotonicity property of Iµ ; namely if 0 ≤ f ≤ g then
Z
f dµ = sup {Iµ (φ) : φ is simple and φ ≤ f }
X
Z
≤ sup {Iµ (φ) : φ is simple and φ ≤ g} ≤ gdµ.
X

19.3 Integrals of positive functions 181
Z
Similarly if c > 0,
X
Z Z fρ ≤ f dµ.
X
cf dµ = c f dµ. Λ
X X
Since Λ is arbitrary, this implies
Also notice that if f is integrable, then µ ({f = ∞}) = 0.
X Z
Lemma 19.15 (Sums as Integrals). Let X be a set and ρ : X → [0, ∞] be a fρ ≤ f dµ.
X
function, let µ = x∈X ρ(x)δx on M = 2X , i.e.
P
X
X
µ(A) = ρ(x).
x∈A Theorem 19.16 (Monotone Convergence Theorem). Suppose fn ∈ L+ is
a sequence of functions such that fn ↑ f (f is necessarily in L+ ) then
If f : X → [0, ∞] is a function (which is necessarily measurable), then Z Z
Z X fn ↑ f as n → ∞.
f dµ = f ρ.
X X Proof. Since fn ≤ fm ≤ f, for all n ≤ m < ∞,
P Proof. Suppose that φ : X → [0, ∞) is a simple function, then φ =

Z Z Z
fn ≤ fm ≤ f
z∈[0,∞) z1{φ=z} and
X X X X X R
φρ = ρ(x) z1{φ=z} (x) = z ρ(x)1{φ=z} (x) from which if follows fn is increasing in n and
X x∈X z∈[0,∞) z∈[0,∞) x∈X Z Z
lim fn ≤ f. (19.9)
X Z
= zµ({φ = z}) = φdµ. n→∞
z∈[0,∞) X
For the opposite inequality, let φ : X → [0, ∞) be a simple function such
So if φ : X → [0, ∞) is a simple function such that φ ≤ f, then that 0 ≤ φ ≤ f, α ∈ (0, 1) and XnR := {fn ≥ αφ} . Notice that Xn ↑ X and
Z fn ≥ α1Xn φ and so by definition of fn ,
X X
φdµ = φρ ≤ f ρ.
Z Z Z
X X X fn ≥ α1Xn φ = α 1Xn φ. (19.10)
Taking the sup over φ in this last equation then shows that Then using the continuity property of µ,
Z X Z Z
f dµ ≤
X
f ρ. lim 1Xn φ = lim 1Xn y1{φ=y}
X X n→∞ n→∞
y>0
X X
For the reverse inequality, let Λ ⊂⊂ X be a finite set and N ∈ (0, ∞). Set = lim yµ(Xn ∩ {φ = y}) = y lim µ(Xn ∩ {φ = y})
n→∞ n→∞
f N (x) = min {N, f (x)} and let φN,Λ be the simple function given by φN,Λ (x) := y>0 y>0
1Λ (x)f N (x). Because φN,Λ (x) ≤ f (x), X Z
= y lim µ({φ = y}) = φ.
Z Z n→∞
X X y>0
fNρ = φN,Λ ρ = φN,Λ dµ ≤ f dµ.
Λ X X X This identity allows us to let n → ∞ in Eq. (19.10) to conclude
Since f N ↑ f as N → ∞, we may let N → ∞ in this last equation to concluded

Z Z
1
φ≤ lim fn .
X α n→∞

Z Z Z Z
Since this is true for all non-negative simple functions φ with φ ≤ f ; (f1 + f2 ) = lim (φn + ψn ) = lim φn + ψn
Z Z Z n→∞ n→∞
1 Z Z Z Z
f = sup φ : φ is simple and φ ≤ f ≤ lim fn . = lim φn + lim ψn = f1 + f2 .
X α n→∞ n→∞ n→∞
R R
Because α ∈ (0, 1) was arbitrary, it follows that f ≤ lim fn which combined N
P ∞
P
n→∞ Now to the general case. Let gN := fn and g = fn , then gN ↑ g and so
with Eq. (19.9) proves the theorem. n=1 1
The following simple lemma will be used often in the sequel. again by monotone convergence theorem and the additivity just proved,
∞ Z N Z N
Lemma 19.17 (Chebyshev’s Inequality). Suppose that f ≥ 0 is a measur- X X Z X
able function, then for any ε > 0, fn := lim fn = lim fn
N →∞ N →∞
n=1 n=1 n=1
∞
Z
1 Z Z Z
µ(f ≥ ε) ≤
X
f dµ. (19.11) = lim gN = g =: fn .
ε X N →∞
n=1
R
In particular if X f dµ < ∞ then µ(f = ∞) = 0 (i.e. f < ∞ a.e.) and the set
{f > 0} is σ – finite.
Remark 19.19. It is in the proof of this corollary (i.e. the linearity of the integral)
Proof. Since 1{f ≥ε} ≤ 1{f ≥ε} 1ε f ≤ 1ε f, that we really make use of the
R assumption that all of our functions are measur-
Z Z Z able. In fact the definition f dµ makes sense for all functions f : X → [0, ∞]
1 1 not just measurable functions. Moreover the monotone convergence theorem
µ(f ≥ ε) = 1{f ≥ε} dµ ≤ 1{f ≥ε} f dµ ≤ f dµ.
X X ε ε X holds in this generality with no change in the proof. However, in the proof of
R Corollary 19.18, we use the approximation Theorem 18.42 which relies heavily
If M := X f dµ < ∞, then on the measurability of the functions to be approximated.
M The following Lemma and the next Corollary are simple applications of
µ(f = ∞) ≤ µ(f ≥ n) ≤ → 0 as n → ∞
n Corollary 19.18.
and {f ≥ 1/n} ↑ {f > 0} with µ(f ≥ 1/n) ≤ nM < ∞ for all n. Lemma 19.20 (The First Borell – Carntelli Lemma). Let (X, M, µ) be
a measure space, An ∈ M, and set
Corollary 19.18. If fn ∈ L+ is a sequence of functions then
∞ [
\
∞
Z X ∞ Z
X {An i.o.} = {x ∈ X : x ∈ An for infinitely many n’s} = An .
fn = fn . N =1 n≥N
n=1 n=1 P∞
P∞ R P∞ If n=1 µ(An ) < ∞ then µ({An i.o.}) = 0.
In particular, if n=1 fn < ∞ then n=1 fn < ∞ a.e.
Proof. (First Proof.) Let us first observe that
Proof. First off we show that
∞
( )
Z Z Z X
{An i.o.} = x ∈ X : 1An (x) = ∞ .
(f1 + f2 ) = f1 + f2
n=1
P∞
by choosing non-negative simple function φn and ψn such that φn ↑ f1 and Hence if n=1 µ(An ) < ∞ then
ψn ↑ f2 . Then (φn + ψn ) is simple as well and (φn + ψn ) ↑ (f1 + f2 ) so by the ∞ ∞ Z ∞
Z X
monotone convergence theorem,
X X
∞> µ(An ) = 1An dµ = 1An dµ
n=1 n=1 X X n=1

19.3 Integrals of positive functions 183
∞
implies that
P
1An (x) < ∞ for µ - a.e. x. That is to say µ({An i.o.}) = 0. Example 19.23. Suppose −∞ < a < b < ∞, f ∈ C([a, b], [0, ∞)) and m be
n=1 Lebesgue measure on R. Also let πk = {a = ak0 < ak1 < · · · < aknk = b} be a
(Second Proof.) Of course we may give a strictly measure theoretic proof of this sequence of refining partitions (i.e. πk ⊂ πk+1 for all k) such that
fact:
mesh(πk ) := max{akj − ak+1

j−1 : j = 1, . . . , nk } → 0 as k → ∞.
 
[
µ(An i.o.) = lim µ  An  For each k, let
N →∞
n≥N
k −1
nX
X
min f (x) : akl ≤ x ≤ akl+1 1(akl ,akl+1 ] (x)

≤ lim µ(An ) fk (x) = f (a)1{a} +
N →∞
n≥N l=0
and the last limit is zero since

P∞
µ(An ) < ∞. then fk ↑ f as k → ∞ and so by the monotone convergence theorem,
n=1
Z b Z Z b
∞
Corollary 19.21. Suppose that (X, M, µ) is a measure space and {An }n=1 ⊂ f dm := f dm = lim fk dm
M is a collection of sets such that µ(Ai ∩ Aj ) = 0 for all i 6= j, then a [a,b] k→∞ a
k −1
nX
∞
min f (x) : akl ≤ x ≤ akl+1 m (akl , akl+1 ]

X = lim
µ (∪∞
n=1 An ) = µ(An ). k→∞
l=0
n=1 Z b
Proof. Since = f (x)dx.

a
Z
µ (∪∞ The latter integral being the Riemann integral.
n=1 An ) = 1∪∞
n=1 An
dµ and
X
∞
X ∞
Z X We can use the above result to integrate some non-Riemann integrable func-
µ(An ) = 1An dµ tions:
n=1 X n=1
Example 19.24. For all λ > 0,
it suffices to show Z ∞ Z
∞ 1
X
1An = 1∪∞ µ – a.e. (19.12) e−λx dm(x) = λ−1 and dm(x) = π.
n=1 An 0 R 1 + x2
n=1
P∞ P∞ The proof of these identities are similar. By the monotone convergence theorem,
Now n=1 1An ≥ 1∪∞ n=1 An
and n=1 1An (x) 6= 1∪∞ n=1 An
(x) iff x ∈ Ai ∩ Aj for Example 19.23 and the fundamental theorem of calculus for Riemann integrals
some i 6= j, that is (or see Theorem 10.14 above or Theorem 19.40 below),
∞
( )
X Z ∞ Z N Z N
x: 1An (x) 6= 1∪∞
n=1 An
(x) = ∪i<j Ai ∩ Aj e−λx
dm(x) = lim e −λx
dm(x) = lim e−λx dx
n=1 0 N →∞ 0 N →∞ 0
1
and the latter set has measure 0 being the countable union of sets of measure = − lim e−λx |N
0 =λ
−1
N →∞ λ
zero. This proves Eq. (19.12) and hence the corollary.
and
Notation 19.22 If m is Lebesgue measure on BR , f is a non-negative Borel Z Z N Z N
Rb
measurable function and a < b with a, b ∈ R̄, we will often write a f (x) dx or 1 1 1
dm(x) = lim dm(x) = lim dx
Rb R R 1 + x2 N →∞ −N 1 + x2 N →∞ −N 1 + x2
a
f dm for (a,b]∩R f dm.
= lim tan−1 (N ) − tan−1 (−N ) = π.

N →∞

Let us also consider the functions x−p , RProposition 19.26. Suppose that f ≥ 0 is a measurable function. Then
Z
1
Z 1
1 X R Also if f, g ≥ 0 are measurable functions
f dµ = 0 iff fR = 0 a.e. R such
R that
dm(x) = lim 1( n1 ,1] (x) p dm(x) f ≤ g a.e. then f dµ ≤ gdµ. In particular if f = g a.e. then f dµ = gdµ.
xp n→∞ x
(0,1] 0
Z 1 1 Proof. If f = 0 a.e. and φ ≤ f is a simple function R then φ = 0 a.e. This
1 x−p+1 implies that µ(φ−1
({y})) = 0 for all y > 0 and hence φdµ = 0 and therefore
= lim dx = lim X
n→∞ 1 xp n→∞ 1 − p
R R
n 1/n X
f dµ = 0. Conversely, if f dµ = 0, then by (Lemma 19.17),
1
if p < 1
= 1−p
Z
∞ if p > 1 µ(f ≥ 1/n) ≤ n f dµ = 0 for all n.
If p = 1 we find P∞
Therefore, µ(f > 0) ≤ n=1 µ(f ≥ 1/n) = 0, i.e. f = 0 a.e. For the second
1
assertion let E be the exceptional set where f > g, i.e. E := {x ∈ X : f (x) >
Z Z
1 1
dm(x) = lim dx = lim ln(x)|11/n = ∞. g(x)}. By assumption E is a null set and 1E c f ≤ 1E c g everywhere. Because
(0,1] xp n→∞ 1
n
x n→∞
g = 1E c g + 1E g and 1E g = 0 a.e.,
Example 19.25. Let {rn }∞
n=1 be an enumeration of the points in Q ∩ [0, 1] and Z Z Z Z
define gdµ = 1E c gdµ + 1E gdµ = 1E c gdµ
∞
X 1
f (x) = 2−n p
n=1 |x − rn | and similarly
R R
f dµ = 1E c f dµ. Since 1E c f ≤ 1E c g everywhere,
with the convention that Z Z Z Z
1 f dµ = 1E c f dµ ≤ 1E c gdµ = gdµ.
p = 5 if x = rn .
|x − rn |
Since, By Theorem 19.40,
Z 1 Z 1 Z rn Corollary 19.27. Suppose that {fn } is a sequence of non-negative measurable
1 1 1 functions and f is a measurable function such that fn ↑ f off a null set, then
p dx = √ dx + √ dx
0 |x − rn | rn x − rn 0 rn − x Z Z
√ √ √ √ fn ↑ f as n → ∞.
= 2 x − rn |1rn − 2 rn − x|r0n = 2 1 − rn − rn
≤ 4,
Proof. Let E ⊂ X be a null set such that fn 1E c ↑ f 1E c as n → ∞. Then
we find by the monotone convergence theorem and Proposition 19.26,
Z ∞ Z ∞ Z Z Z Z
X 1 X
f (x)dm(x) = 2−n p dx ≤ 2−n 4 = 4 < ∞. fn = fn 1E c ↑ f 1E c = f as n → ∞.
[0,1] n=1 [0,1] |x − rn | n=1
In particular, m(f = ∞) = 0, i.e. that f < ∞ for almost every x ∈ [0, 1] and
this implies that Lemma 19.28 (Fatou’s Lemma). If fn : X → [0, ∞] is a sequence of mea-
∞ surable functions then
X 1
2−n p < ∞ for a.e. x ∈ [0, 1]. Z Z
n=1 |x − rn |
lim inf fn ≤ lim inf fn
n→∞ n→∞
This result is somewhat surprising since the singularities of the summands form
a dense subset of [0, 1].

19.4 Integrals of Complex Valued Functions 185
Proof. Define gk := inf fn so that gk ↑ lim inf n→∞ fn as k → ∞. Since Proposition 19.32. The map
n≥k
gk ≤ fn for all k ≤ n,
Z
1
Z Z f ∈ L (µ; R) → f dµ ∈ R
gk ≤ fn for all n ≥ k X
R R
and therefore is linear and has the monotonicity property: f dµ ≤ gdµ for all f, g ∈
L1 (µ; R) such that f ≤ g a.e.
Z Z
gk ≤ lim inf fn for all k.
n→∞
Proof. Let f, g ∈ L1 (µ; R) and a, b ∈ R. By modifying f and g on a null set,
We may now use the monotone convergence theorem to let k → ∞ to find we may assume that f, g are real valued functions. We have af + bg ∈ L1 (µ; R)
Z Z
MCT
Z Z because
lim inf fn = lim gk = lim gk ≤ lim inf fn . |af + bg| ≤ |a| |f | + |b| |g| ∈ L1 (µ; R) .
n→∞ k→∞ k→∞ n→∞
If a < 0, then
(af )+ = −af− and (af )− = −af+
so that Z Z Z Z Z Z
19.4 Integrals of Complex Valued Functions af = −a f− + a f+ = a( f+ − f− ) = a f.
Definition 19.29. A measurable function f : X → R̄ is integrable if f+ := A similar calculation works for a > 0 and the case a = 0 is trivial so we have
f 1{f ≥0} and f− = −f 1{f ≤0} are integrable. We write L1 (µ; R) for the space shown that
of real valued integrable functions. For f ∈ L1 (µ; R) , let
Z Z
Z Z Z af = a f.
f dµ = f+ dµ − f− dµ Now set h = f + g. Since h = h+ − h− ,
Convention: If f, g : X → R̄ are two measurable functions, let f +g denote h+ − h− = f+ − f− + g+ − g−

the collection of measurable functions h : X → R̄ such that h(x) = f (x) + g(x) or
whenever f (x) + g(x) is well defined, i.e. is not of the form ∞ − ∞ or −∞ + ∞. h+ + f− + g− = h− + f+ + g+ .
We use a similar convention for f − g. Notice that if f, g ∈ L1 (µ; R) and
h1 , h2 ∈ f + g, then h1 = h2 a.e. because |f | < ∞ and |g| < ∞ a.e. Therefore, Z Z Z Z Z Z
Notation 19.30 (Abuse of notation) We will sometimes denote the inte- h+ + f− + g− = h− + f+ + g+
R
gral X f dµ by µ (f ) . With this notation we have µ (A) = µ (1A ) for all A ∈ M. and hence
Remark 19.31. Since
Z Z Z Z Z Z Z Z Z
f± ≤ |f | ≤ f+ + f− , h = h+ − h− = f+ + g+ − f− − g− = f + g.
R
a measurable function f is integrable iff |f | dµ < ∞. Hence Finally if f+ − f− = f ≤ g = g+ − g− then f+ + g− ≤ g+ + f− which implies
Z that Z Z Z Z
1
L (µ; R) := f : X → R̄ : f is measurable and |f | dµ < ∞ . f+ + g− ≤ g+ + f−
X
1
If f, g ∈ L (µ; R) and Rf = g a.e.
R then f± = g± a.e. and so it follows from or equivalently that
Proposition 19.26 that f dµ = gdµ. In particular if f, g ∈ L1 (µ; R) we may Z Z Z Z Z Z
define Z Z f = f+ − f− ≤ g+ − g− = g.
(f + g) dµ = hdµ
X X The monotonicity property is also a consequence of the linearity of the integral,
where h is any element of f + g. the fact that f ≤ g a.e. implies 0 ≤ g − f a.e. and Proposition 19.26.

Definition
R 19.33. A measurable function f : X → C is integrable if 2. The following are equivalent
|f | dµ < ∞. Analogously to the real case, let
R R
X a) RE f = E g for all E ∈ M
Z b) |f − g| = 0
1 X
L (µ; C) := f : X → C : f is measurable and |f | dµ < ∞ . c) f = g a.e.
X
denote√the complex valued integrable functions. Because, max (|Re f | , |Im f |) ≤ Proof. 1. By Chebyshev’s inequality, Lemma 19.17,
R
|f | ≤ 2 max (|Re f | , |Im f |) , |f | dµ < ∞ iff Z
1
Z Z µ(|f | ≥ ) ≤ n |f | dµ < ∞
n X
|Re f | dµ + |Im f | dµ < ∞.
for all n. 2. (a) =⇒ (c) Notice that
For f ∈ L1 (µ; C) define Z Z Z
Z Z Z f= g⇔ (f − g) = 0
f dµ = Re f dµ + i Im f dµ. E E E
for all E ∈ M. Taking E = {Re(f − g) > 0} and using 1E Re(f − g) ≥ 0, we

It is routine to show the integral is still linear on L1 (µ; C) (prove!). In the learn that
remainder of this section, let L1 (µ) be either L1 (µ; C) or L1 (µ; R) . If A ∈ M Z Z
and f ∈ L1 (µ; C) or f : X → [0, ∞] is a measurable function, let 0 = Re (f − g)dµ = 1E Re(f − g) =⇒ 1E Re(f − g) = 0 a.e.
Z Z E
f dµ := 1A f dµ. This implies that 1E = 0 a.e. which happens iff

A X
Proposition 19.34. Suppose that f ∈ L1 (µ; C) , then µ ({Re(f − g) > 0}) = µ(E) = 0.
Similar µ(Re(f − g) < 0) = 0 so that Re(f − g) = 0 a.e. Similarly, Im(f − g) = 0

Z Z

f dµ ≤ |f | dµ. (19.13)
a.e and hence f − g = 0 a.e., i.e. f = g a.e. (c) =⇒ (b) is clear and so is (b)
X X
=⇒ (a) since Z Z Z
by writing X f dµ = Reiθ with R ≥ 0. We may assume that
R
Proof.
R Start

f−

g ≤ |f − g| = 0.
R = f dµ > 0 since otherwise there is nothing to prove. Since
X

E E
Z Z Z Z
R = e−iθ e−iθ f dµ = Re e−iθ f dµ + i Im e−iθ f dµ,

f dµ =
X X X X Definition 19.36. Let (X, M, µ) be a measure space and L1 (µ) = L1 (X, M, µ)
denote the set of L1 (µ) functions modulo the equivalence relation; f ∼ g iff
it must be that X Im e−iθ f dµ = 0. Using the monotonicity in Proposition
R
f = g a.e. We make this into a normed space using the norm
19.26, Z
kf − gkL1 = |f − g| dµ
Z Z Z Z
−iθ −iθ

f dµ =
Re e f dµ ≤ Re e f dµ ≤
|f | dµ.
X X X X
and into a metric space using ρ1 (f, g) = kf − gkL1 .
Warning: in the future we will often not make much of a distinction between
Proposition 19.35. Let f, g ∈ L1 (µ) , then
L1 (µ) and L1 (µ) . On occasion this can be dangerous and this danger will be
1 1 pointed out when necessary.
1. The set {f 6= 0} is σ – finite, in fact {|f | ≥ n} ↑ {f 6= 0} and µ(|f | ≥ n) <
∞ for all n.

19.4 Integrals of Complex Valued Functions 187
Remark 19.37. More generally we may define Lp (µ) = Lp (X, M, µ) for p ∈ Exercise 19.1. Give another proof of Proposition 19.34 by first proving Eq.
[1, ∞) as the set of measurable functions f such that (19.13) with f being a cylinder function in which case the triangle inequality for
Z complex numbers will do the trick. Then use the approximation Theorem 18.42
p
|f | dµ < ∞ along with the dominated convergence Theorem 19.38 to handle the general
X case.
modulo the equivalence relation; f ∼ g iff f = g a.e. Corollary 19.39. Let {fnP
∞
}n=1 ⊂ L1 (µ) be a sequence such that
P∞ ∞
We will see in Chapter 21 that n=1 kfn kL1 (µ) < ∞, then n=1 fn is convergent a.e. and
∞ ∞ Z
Z !
Z 1/p X X
kf kLp =
p
|f | dµ for f ∈ L (µ)p fn dµ = fn dµ.
X n=1 n=1 X
P∞ P∞
is a norm and (Lp (µ), k·kLp ) is a Banach space in this norm. Proof. The condition n=1 kfn kL1 (µ) < ∞ is equivalent to n=1 |fn | ∈
P∞ PN
Theorem 19.38 (Dominated Convergence Theorem).RSuppose fn ,Rgn , g ∈ L1 (µ) . Hence n=1 fn is almost everywhere convergent and if SN := n=1 fn ,
L1 (µ) , fn → f a.e., |fn | ≤ gn ∈ L1 (µ) , gn → g a.e. and X gn dµ → X gdµ. then
N ∞
Then f ∈ L1 (µ) and
X X
Z Z |SN | ≤ |fn | ≤ |fn | ∈ L1 (µ) .
f dµ = lim fn dµ. n=1 n=1
X h→∞ X So by the dominated convergence theorem,
(In most typical applications of this theorem gn = g ∈ L1 (µ) for all n.)
∞
Z ! Z Z
X
Proof. Notice that |f | = limn→∞ |fn | ≤ limn→∞ |gn | ≤ g a.e. so that fn dµ = lim SN dµ = lim SN dµ
X X N →∞ N →∞ X
f ∈ L1 (µ) . By considering the real and imaginary parts of f separately, it n=1
N Z ∞ Z
suffices to prove the theorem in the case where f is real. By Fatou’s Lemma, X X
= lim fn dµ = fn dµ.
Z Z Z N →∞ X X
n=1 n=1
(g ± f )dµ = lim inf (gn ± fn ) dµ ≤ lim inf (gn ± fn ) dµ
X X n→∞ n→∞ X
Z Z
= lim gn dµ + lim inf ± fn dµ Theorem 19.40 (The Fundamental Theorem of Calculus). Suppose
n→∞ X n→∞ X Rx
Z Z −∞ < a < b < ∞, f ∈ C((a, b), R)∩L1 ((a, b), m) and F (x) := a f (y)dm(y).
= gdµ + lim inf ± fn dµ Then
X n→∞ X
1. F ∈ C([a, b], R) ∩ C 1 ((a, b), R).
Since lim inf n→∞ (−an ) = − lim supn→∞ an , we have shown, 2. F 0 (x) = f (x) for all x ∈ (a, b).
Z Z Z R 3. If G ∈ C([a, b], R) ∩ C 1 ((a, b), R) is an anti-derivative of f on (a, b) (i.e.
lim inf n→∞ X Rfn dµ
gdµ ± f dµ ≤ gdµ + f = G0 |(a,b) ) then
X X X − lim supn→∞ X fn dµ Z b
f (x)dm(x) = G(b) − G(a).
and therefore a
Z Z Z R
lim sup fn dµ ≤ f dµ ≤ lim inf fn dµ. Proof. Since F (x) := R 1(a,x) (y)f (y)dm(y), limx→z 1(a,x) (y) = 1(a,z) (y) for
n→∞ X X n→∞ X m – a.e. y and 1(a,x) (y)f (y) ≤ 1(a,b) (y) |f (y)| is an L1 – function, it follows
R R from the dominated convergence Theorem 19.38 that F is continuous on [a, b].
This shows that lim fn dµ exists and is equal to f dµ. Simple manipulations show,
n→∞ X X

 R ∞
x+h
[f (y) − f (x)] dm(y) if h > 0 P∞ Suppose R > 0 and {an }n=0 is a
Example 19.42 (Integration of Power Series).

F (x + h) − F (x)
= 1

Rx

− f (x) x
sequence of complex numbers such that n=0 |an | rn < ∞ for all r ∈ (0, R).
h |h| 

[f (y) − f (x)] dm(y) if h < 0

x+h Then
(R
x+h
1
Rxx |f (y) − f (x)| dm(y) if h > 0
∞ ∞ ∞
Z β X ! Z β
≤ n
X X β n+1 − αn+1
|h| |f (y) − f (x)| dm(y) if h < 0 an x dm(x) = an xn dm(x) = an
x+h
α n=0 n=0 α n=0
n+1
≤ sup {|f (y) − f (x)| : y ∈ [x − |h| , x + |h|]}
for all −R < α < β < R. Indeed this follows from Corollary 19.39 since
and the latter expression, by the continuity of f, goes to zero as h → 0 . This
shows F 0 = f on (a, b). For the converse direction, we have by assumption that ∞ Z ∞
!
X β X Z |β| Z |α|
n n n
G0 (x) = F 0 (x) for x ∈ (a, b). Therefore by the mean value theorem, F − G = C |an | |x| dm(x) ≤ |an | |x| dm(x) + |an | |x| dm(x)
α 0 0
for some constant C. Hence n=0 n=0
∞ n+1 n+1 ∞
Z b X |β| + |α| X
f (x)dm(x) = F (b) = F (b) − F (a) ≤ |an | ≤ 2r |an | rn < ∞
n=0
n+1 n=0
a
= (G(b) + C) − (G(a) + C) = G(b) − G(a).
where r = max(|β| , |α|).
Corollary 19.43 (Differentiation Under the Integral). Suppose that J ⊂

Example 19.41. The following limit holds, R is an open interval and f : J × X → C is a function such that
Z n
x
lim (1 − )n dm(x) = 1. 1. x → f (t, x) is measurable for each t ∈ J.
n→∞ 0 n
2. f (t0 , ·) ∈ L1 (µ) for some t0 ∈ J.
Let fn (x) = (1 − x n
n ) 1[0,n] (x) and notice that limn→∞ fn (x) = e−x . We will 3. ∂f
∂t (t, x) exists for all (t, x).
now show 4. There is a function g ∈ L1 (µ) such that ∂f

(t, ·) ≤ g for each t ∈ J.

∂t
0 ≤ fn (x) ≤ e−x for all x ≥ 0.
Then f (t, ·) ∈ L1 (µ) for all t ∈ J (i.e. X |f (t, x)| dµ(x) < ∞), t →
R
It suffices to consider x ∈ [0, n]. Let g(x) = ex fn (x), then for x ∈ (0, n), R
f (t, x)dµ(x) is a differentiable function on J and
d 1 1 1 X
ln g(x) = 1 + n x (− ) = 1 − ≤0
dx (1 − n ) n (1 − nx ) d
Z Z
∂f
f (t, x)dµ(x) = (t, x)dµ(x).
which shows that ln g(x) and hence g(x) is decreasing on [0, n]. Therefore g(x) ≤ dt X X ∂t
g(0) = 1, i.e.
Proof. (The proof is essentially the same as for sums.) By considering the
0 ≤ fn (x) ≤ e−x .
real and imaginary parts of f separately, we may assume that f is real. Also
From Example 19.24, we know notice that
Z ∞ ∂f
(t, x) = lim n(f (t + n−1 , x) − f (t, x))
e−x dm(x) = 1 < ∞, ∂t n→∞
0
and therefore, for x → ∂f
∂t (t, x) is a sequential limit of measurable functions and
so that e−x is an integrable function on [0, ∞). Hence by the dominated con-
hence is measurable for all t ∈ J. By the mean value theorem,
vergence theorem,
Z n Z ∞
x n |f (t, x) − f (t0 , x)| ≤ g(x) |t − t0 | for all t ∈ J (19.14)
lim (1 − ) dm(x) = lim fn (x)dm(x)
n→∞ 0 n n→∞
Z ∞ 0 Z ∞ and hence
= lim fn (x)dm(x) = e−x dm(x) = 1.
0 n→∞ 0 |f (t, x)| ≤ |f (t, x) − f (t0 , x)| + |f (t0 , x)| ≤ g(x) |t − t0 | + |f (t0 , x)| .

19.5 Measurability on Complete Measure Spaces 189
This shows f (t, ·) ∈ L1 (µ) for all t ∈ J. Let G(t) := X f (t, x)dµ(x), then n
xn e−λx dm(x). Recall that
R R
That is n! = λ [0,∞)
G(t) − G(t0 ) f (t, x) − f (t0 , x)

Z Z
= dµ(x). Γ (t) := xt−1 e−x dx for t > 0.
t − t0 X t − t0 [0,∞)
By assumption,
(The reader should check that Γ (t) < ∞ for all t > 0.) We have just shown
f (t, x) − f (t0 , x) ∂f that Γ (n + 1) = n! for all n ∈ N.
lim = (t, x) for all x ∈ X
t→t0 t − t0 ∂t
Remark 19.45. Corollary 19.43 may be generalized by allowing the hypothesis
and by Eq. (19.14), to hold for x ∈ X \ E where E ∈ M is a fixed null set, i.e. E must be

f (t, x) − f (t0 , x) independent
R ∞of t. Consider what happens if we formally apply Corollary 19.43
≤ g(x) for all t ∈ J and x ∈ X. to g(t) := 0 1x≤t dm(x),
t − t0
d ∞
Z Z ∞
Therefore, we may apply the dominated convergence theorem to conclude ? ∂
ġ(t) = 1x≤t dm(x) = 1x≤t dm(x).
dt 0 0 ∂t
G(tn ) − G(t0 ) f (tn , x) − f (t0 , x)
Z
lim = lim dµ(x) ∂
n→∞ tn − t0 n→∞ X tn − t0 The last integral is zero since ∂t 1x≤t = 0 unless t = x in which case it is not
Z
f (tn , x) − f (t0 , x) defined. On the other hand g(t) = t so that ġ(t) = 1. (The reader should decide
= lim dµ(x) which hypothesis of Corollary 19.43 has been violated in this example.)
n→∞ tn − t0
ZX
∂f
= (t0 , x)dµ(x)
X ∂t 19.5 Measurability on Complete Measure Spaces
for all sequences tn ∈ J \ {t0 } such that tn → t0 . Therefore, Ġ(t0 ) =
limt→t0 G(t)−G(t0)
exists and In this subsection we will discuss a couple of measurability results concerning
t−t0
completions of measure spaces.
Z
∂f
Ġ(t0 ) = (t0 , x)dµ(x). Proposition 19.46. Suppose that (X, M, µ) is a complete measure space1 and
X ∂t
f : X → R is measurable.
1. If g : X → R is a function such that f (x) = g(x) for µ – a.e. x, then g is
Example 19.44. Recall from Example 19.24 that measurable.
Z 2. If fn : X → R are measurable and f : X → R is a function such that
λ−1 = e−λx dm(x) for all λ > 0. limn→∞ fn = f, µ - a.e., then f is measurable as well.
[0,∞)
Let ε > 0. For λ ≥ 2ε > 0 and n ∈ N there exists Cn (ε) < ∞ such that Proof. 1. Let E = {x : f (x) 6= g(x)} which is assumed to be in M and
n µ(E) = 0. Then g = 1E c f + 1E g since f = g on E c . Now 1E c f is measurable
d so g will be measurable if we show 1E g is measurable. For this consider,
0≤ − e−λx = xn e−λx ≤ C(ε)e−εx .
dλ
E ∪ (1E g)−1 (A \ {0}) if 0 ∈ A
c
−1
(1E g) (A) = (19.15)
Using this fact, Corollary 19.43 and induction gives (1E g)−1 (A) if 0 ∈
/A
n Z n
d d Since (1E g)−1 (B) ⊂ E if 0 ∈
n!λ−n−1 = − λ−1 = − e−λx dm(x) / B and µ(E) = 0, it follow by completeness of
dλ [0,∞) dλ M that (1E g)−1 (B) ∈ M if 0 ∈/ B. Therefore Eq. (19.15) shows that 1E g is
Z
= xn e−λx dm(x). 1
Recall this means that if N ⊂ X is a set such that N ⊂ A ∈ M and µ(A) = 0,
[0,∞) then N ∈ M as well.

measurable. 2. Let E = {x : lim fn (x) 6= f (x)} by assumption E ∈ M and 19.6 Comparison of the Lebesgue and the Riemann
n→∞
µ(E) = 0. Since g := 1E f = limn→∞ 1E c fn , g is measurable. Because f = g Integral
on E c and µ(E) = 0, f = g a.e. so by part 1. f is also measurable.
The above results are in general false if (X, M, µ) is not complete. For For the rest of this chapter, let −∞ < a < b < ∞ and f : [a, b] → R be a
example, let X = {0, 1, 2}, M = {{0}, {1, 2}, X, φ} and µ = δ0 . Take g(0) = bounded function. A partition of [a, b] is a finite subset π ⊂ [a, b] containing
0, g(1) = 1, g(2) = 2, then g = 0 a.e. yet g is not measurable. {a, b}. To each partition
Lemma 19.47. Suppose that (X, M, µ) is a measure space and M̄ is the com- π = {a = t0 < t1 < · · · < tn = b} (19.17)
pletion of M relative to µ and µ̄ is the extension of µ to M̄. Then a function
of [a, b] let
f : X → R is (M̄, B = B R ) – measurable iff there exists a function g : X → R
mesh(π) := max{|tj − tj−1 | : j = 1, . . . , n},
that is (M, B) – measurable such E = {x : f (x) 6= g(x)} ∈ M̄ and µ̄ (E) = 0,
i.e. f (x) = g(x) for µ̄ – a.e. x. Moreover for such a pair f and g, f ∈ L1 (µ̄) iff
g ∈ L1 (µ) and in which case Mj = sup{f (x) : tj ≤ x ≤ tj−1 }, mj = inf{f (x) : tj ≤ x ≤ tj−1 }
X n n
X
Z Z
Gπ = f (a)1{a} + Mj 1(tj−1 ,tj ] , gπ = f (a)1{a} + mj 1(tj−1 ,tj ] and
f dµ̄ = gdµ.
X X 1 1
X X
Proof. Suppose first that such a function g exists so that µ̄(E) = 0. Since g Sπ f = Mj (tj − tj−1 ) and sπ f = mj (tj − tj−1 ).
is also (M̄, B) – measurable, we see from Proposition 19.46 that f is (M̄, B) –
measurable. Conversely if f is (M̄, B) – measurable, by considering f± we may Notice that Z b Z b
assume that f ≥ 0. Choose (M̄, B) – measurable simple function φn ≥ 0 such Sπ f = Gπ dm and sπ f = gπ dm.
that φn ↑ f as n → ∞. Writing a a
X The upper and lower Riemann integrals are defined respectively by
φn = ak 1Ak
Z b Z a
with Ak ∈ M̄, we may choose Bk ∈ M such that Bk ⊂ Ak and µ̄(Ak \ Bk ) = 0. f (x)dx = inf Sπ f and f (x)dx = sup sπ f.
a π π
b
Letting X
φ̃n := ak 1Bk Rb Rb
Definition 19.48. The function f is Riemann integrable iff a f = f ∈ R
Rb a
we have produced a (M, B) – measurable simple function φ̃P n ≥ 0 such that and which case the Riemann integral a f is defined to be the common value:
En := {φn 6= φ̃n } has zero µ̄ – measure. Since µ̄ (∪n En ) ≤ n µ̄ (En ) , there
b b b
exists F ∈ M such that ∪n En ⊂ F and µ(F ) = 0. It now follows that
Z Z Z
f (x)dx = f (x)dx = f (x)dx.
a a a
1F φ̃n = 1F φn ↑ g := 1F f as n → ∞.
The proof of the following Lemma is left to the reader as Exercise 19.20.
This shows that g = 1F f is (M, B) – Rmeasurable R and that {f 6= g} ⊂ F has µ̄
– measure zero. Since f = g, µ̄ – a.e., X f dµ̄ = X gdµ̄ so to prove Eq. (19.16) Lemma 19.49. If π 0 and π are two partitions of [a, b] and π ⊂ π 0 then
it suffices to prove
Gπ ≥ Gπ0 ≥ f ≥ gπ0 ≥ gπ and
Z Z
gdµ̄ = gdµ. (19.16)
X X Sπ f ≥ Sπ0 f ≥ sπ0 f ≥ sπ f.
Because µ̄ = µ on M, Eq. (19.16) is easily verified for non-negative M – ∞
There exists an increasing sequence of partitions {πk }k=1 such that mesh(πk ) ↓
measurable simple functions. Then by the monotone convergence theorem and
0 and
the approximation Theorem 18.42 it holds for all M – measurable functions Z b Z b
g : X → [0, ∞]. The rest of the assertions follow in the standard way by S πk f ↓ f and sπk f ↑ f as k → ∞.
considering (Re g)± and (Im g)± . a a

19.6 Comparison of the Lebesgue and the Riemann Integral 191
If we let for all k large enough, since eventually Gk (x) is the supremum of f (y) over
G := lim Gπk and g := lim gπk (19.18) some interval contained in [x − ε, x + ε]. Again letting k → ∞ implies
k→∞ k→∞
sup f (y) ≥ G(x) and therefore, that
then by the dominated convergence theorem, |y−x|≤ε
Z Z Z b H(x) = lim sup f (y) ≥ G(x)

gdm = lim gπk = lim sπk f = f (x)dx (19.19) y→x
[a,b] k→∞ [a,b] k→∞ a
and for all x ∈

/ π. Combining this equation with Eq. (19.22) then implies H(x) =
G(x) if x ∈ / π. A similar argument shows that h(x) = g(x) if x ∈ / π and
Z Z Z b
hence Eq. (19.21) is proved.
Gdm = lim Gπk = lim Sπk f = f (x)dx. (19.20)
[a,b] k→∞ [a,b] k→∞ a 3. The functions G and g are limits of measurable functions and hence mea-
surable. Since H = G and h = g except possibly on the countable set π,
Notation 19.50 For x ∈ [a, b], let both H and h are also Borel measurable. (You justify this statement.)
H(x) = lim sup f (y) := lim sup{f (y) : |y − x| ≤ ε, y ∈ [a, b]} and
y→x ε↓0
h(x) = lim inf f (y) := lim inf {f (y) : |y − x| ≤ ε, y ∈ [a, b]}. Theorem 19.52. Let f : [a, b] → R be a bounded function. Then
y→x ε↓0
Z b Z Z b Z
Lemma 19.51. The functions H, h : [a, b] → R satisfy: f= Hdm and f= hdm (19.23)
a [a,b] a [a,b]
1. h(x) ≤ f (x) ≤ H(x) for all x ∈ [a, b] and h(x) = H(x) iff f is continuous
and the following statements are equivalent:
at x.
∞
2. If {πk }k=1 is any increasing sequence of partitions such that mesh(πk ) ↓ 0 1. H(x) = h(x) for m -a.e. x,
and G and g are defined as in Eq. (19.18), then 2. the set
E := {x ∈ [a, b] : f is discontinuous at x}
G(x) = H(x) ≥ f (x) ≥ h(x) = g(x) / π := ∪∞
∀x∈ k=1 πk . (19.21)
is an m̄ – null set.
(Note π is a countable set.) 3. f is Riemann integrable.
3. H and h are Borel measurable.
If f is Riemann integrable then f is Lebesgue measurable2 , i.e. f is L/B –
Proof. Let Gk := Gπk ↓ G and gk := gπk ↑ g. measurable where L is the Lebesgue σ – algebra and B is the Borel σ – algebra
on [a, b]. Moreover if we let m̄ denote the completion of m, then
1. It is clear that h(x) ≤ f (x) ≤ H(x) for all x and H(x) = h(x) iff lim f (y)
y→x Z Z b Z Z
exists and is equal to f (x). That is H(x) = h(x) iff f is continuous at x. Hdm = f (x)dx = f dm̄ = hdm. (19.24)
2. For x ∈/ π, [a,b] a [a,b] [a,b]
Gk (x) ≥ H(x) ≥ f (x) ≥ h(x) ≥ gk (x) ∀ k ∞
Proof. Let {πk }k=1 be an increasing sequence of partitions of [a, b] as de-
and letting k → ∞ in this equation implies scribed in Lemma 19.49 and let G and g be defined as in Lemma 19.51. Since
m(π) = 0, H = G a.e., Eq. (19.23) is a consequence of Eqs. (19.19) and (19.20).
G(x) ≥ H(x) ≥ f (x) ≥ h(x) ≥ g(x) ∀ x ∈
/ π. (19.22) From Eq. (19.23), f is Riemann integrable iff
Moreover, given ε > 0 and x ∈
Z Z
/ π,
Hdm = hdm
[a,b] [a,b]
sup{f (y) : |y − x| ≤ ε, y ∈ [a, b]} ≥ Gk (x)
2
f need not be Borel measurable.

and because h ≤ f ≤ H this happens iff h(x) = H(x) for m - a.e. x. Since Corollary 19.56. Suppose that (X, τ ) is a topological space, BX = σ(τ ) is the
E = {x : H(x) 6= h(x)}, this last condition is equivalent to E being a m – null Borel σ – algebra on X and µ and ν are two measures on BX which are σ –
set. In light of these results and Eq. (19.21), the remaining assertions including finite on τ. If µ = ν on τ then µ = ν on BX , i.e. µ ≡ ν.
Eq. (19.24) are now consequences of Lemma 19.47.
Corollary 19.57. Suppose that µ and ν are two measures on BRn which are
Rb
Notation 19.53 In view of this theorem we will often write a f (x)dx for finite on bounded sets and such that µ(A) = ν(A) for all sets A of the form
Rb
a
f dm. A = (a, b] = (a1 , b1 ] × · · · × (an , bn ]
with a, b ∈ Rn and a < b, i.e. ai < bi for all i. Then µ = ν on BRn .

19.7 Determining Classes of Measures
Proposition 19.58. Suppose that (X, d) is a metric space, µ and ν are two
X
Definition 19.54 (σ – finite). Let X be a set and E ⊂ F ⊂ 2 . We say that measures on BX := σ(τd ) which are finite on bounded measurable subsets of X
a function µ : F → [0, ∞] is σ – finite on E if there exist Xn ∈ E such that and Z Z
Xn ↑ X and µ(Xn ) < ∞ for all n. f dµ = f dν (19.25)
X X
Theorem 19.55 (Uniqueness). Suppose that C ⊂ 2X is a π – class (see Def- for all f ∈ BCb (X, R) where
inition 18.53), M = σ(C) and µ and ν are two measure on M. If µ and ν are
σ – finite on C and µ = ν on C, then µ = ν on M. BCb (X, R) = {f ∈ BC(X, R) : supp(f ) is bounded}. (19.26)
Proof. We begin first with the special case where µ(X) < ∞ and therefore Then µ ≡ ν.
also
Proof. To prove this fix a o ∈ X and let
ν(X) = lim ν (Xn ) = lim µ (Xn ) = µ(X) < ∞.
n→∞ n→∞
ψR (x) = ([R + 1 − d(x, o)] ∧ 1) ∨ 0 (19.27)
Let
H := {f ∈ `∞ (M, R) : µ (f ) = ν (f )} . so that ψR ∈ BCb (X, [0, 1]), supp(ψR ) ⊂ B(o, R + 2) and ψR ↑ 1 as R → ∞.
Then H is a linear subspace which is closed under bounded convergence (by the Let HR denote the space of bounded real valued BX – measurable functions f
dominated convergence theorem), contains 1 and contains the multiplicative such that Z Z
system, M := {1C : C ∈ C} . Therefore, by Theorem 18.51 or Corollary 18.54, ψR f dµ = ψR f dν. (19.28)
H = `∞ (M, R) and hence µ = ν. For the general case, let Xn1 , Xn2 ∈ C be X X
chosen so that Xn1 ↑ X and Xn2 ↑ X as n → ∞ and µ Xn1 + ν Xn2 < ∞ for Then HR is closed under bounded convergence and because of Eq. (19.25)
all n. Then Xn := Xn1 ∩ Xn2 ∈ C increases to X and ν (Xn ) = µ (Xn ) < ∞ for contains BC(X, R). Therefore by Corollary 18.55, HR contains all bounded
all n. For each n ∈ N, define two measures µn and νn on M by measurable functions on X. Take f = 1A in Eq. (19.28) with A ∈ BX , and then
use the monotone convergence theorem to let R → ∞. The result is µ(A) = ν(A)
µn (A) := µ(A ∩ Xn ) and νn (A) = ν(A ∩ Xn ). for all A ∈ BX .
Here is another version of Proposition 19.58.
Then, as the reader should verify, µn and νn are finite measure on M such
that µn = νn on C. Therefore, by the special case just proved, µn = νn on M. Proposition 19.59. Suppose that (X, d) is a metric space, µ and ν are two
Finally, using the continuity properties of measures, measures on BX = σ(τd ) which are both finite on compact sets. Further assume
there exist compact sets Kk ⊂ X such that Kko ↑ X. If
µ(A) = lim µ(A ∩ Xn ) = lim ν(A ∩ Xn ) = ν(A) Z Z
n→∞ n→∞
f dµ = f dν (19.29)
for all A ∈ M. X X
As an immediate consequence we have the following corollaries. for all f ∈ Cc (X, R) then µ ≡ ν.

19.8 Exercises 193
Proof. Let ψn,k be defined as in the proof of Proposition 18.56 and let Hn,k Exercise 19.4. Suppose that µn : M → [0, ∞] are measures on M for n ∈ N.
denote those bounded BX – measurable functions, f : X → R such that Also suppose that µn (A) is increasing in n for all A ∈ M. Prove that µ : M →
Z Z [0, ∞] defined by µ(A) := limn→∞ µn (A) is also a measure.
f ψn,k dµ = f ψn,k dν.
X X Exercise 19.5. Now suppose that Λ is some index set and for eachPλ ∈ Λ, µλ :
M → [0, ∞] is a measure on M. Define µ : M → [0, ∞] by µ(A) = λ∈Λ µλ (A)
By assumption BC(X, R) ⊂ Hn,k and one easily checks that Hn,k is closed for each A ∈ M. Show that µ is also a measure.
under bounded convergence. Therefore, by Corollary 18.55, Hn,k contains all
bounded measurable function. In particular for A ∈ BX , Exercise 19.6. Let (X, M, µ) be a measure space
R and ρ : X → [0, ∞] be a
Z Z measurable function. For A ∈ M, set ν(A) := A ρdµ.
1A ψn,k dµ = 1A ψn,k dν.
X X
1. Show ν : M → [0, ∞] is a measure.
2. Let f : X → [0, ∞] be a measurable function, show
Letting n → ∞ in this equation, using the dominated convergence theorem, one Z Z
shows Z Z f dν = f ρdµ. (19.30)
1A 1Kko dµ = 1A 1Kko dν X X
X X
Hint: first prove the relationship for characteristic functions, then for sim-
holds for k. Finally using the monotone convergence theorem we may let k → ∞
ple functions, and then for general positive measurable functions.
to conclude
3. Show that a measurable function f : X → C is in L1 (ν) iff |f | ρ ∈ L1 (µ)
Z Z
µ(A) = 1A dµ = 1A dν = ν(A) and if f ∈ L1 (ν) then Eq. (19.30) still holds.
X X
for all A ∈ BX . Notation 19.60 It is customary to informally describe ν defined in Exercise
19.6 by writing dν = ρdµ.
19.8 Exercises Exercise 19.7. Let (X, M, µ) be a measure space, (Y, F) be a measurable
space and f : X → Y be a measurable map. Define a function ν : F → [0, ∞]
Exercise 19.2. Let µ be a measure on an algebra A ⊂ 2X , then µ(A)+µ(B) = by ν(A) := µ(f −1 (A)) for all A ∈ F.
µ(A ∪ B) + µ(A ∩ B) for all A, B ∈ A.
1. Show ν is a measure. (We will write ν = f∗ µ or ν = µ ◦ f −1 .)
Exercise 19.3 (From problem 12 on p. 27 of Folland.). Let (X, M, µ) 2. Show Z Z
be a finite measure space and for A, B ∈ M let ρ(A, B) = µ(A∆B) where gdν = (g ◦ f ) dµ (19.31)
Y X
A∆B = (A \ B) ∪ (B \ A) . It is clear that ρ (A, B) = ρ (B, A) . Show:
for all measurable functions g : Y → [0, ∞]. Hint: see the hint from Exercise
1. ρ satisfies the triangle inequality: 19.6.
3. Show a measurable function g : Y → C is in L1 (ν) iff g ◦ f ∈ L1 (µ) and
ρ (A, C) ≤ ρ (A, B) + ρ (B, C) for all A, B, C ∈ M. that Eq. (19.31) holds for all g ∈ L1 (ν).
2. Define A ∼ B iff µ(A∆B) = 0 and notice that ρ (A, B) = 0 iff A ∼ B. Show Exercise 19.8. Let F : R → R be a C 1 -function such that F 0 (x) > 0 for all
“∼ ” is an equivalence relation. x ∈ R and limx→±∞ F (x) = ±∞. (Notice that F is strictly increasing so that
3. Let M/ ∼ denote M modulo the equivalence relation, ∼, and let [A] := F −1 : R → R exists and moreover, by the inverse function theorem that F −1 is
{B ∈ M : B ∼ A} . Show that ρ̄ ([A] , [B]) := ρ (A, B) is gives a well defined a C 1 – function.) Let m be Lebesgue measure on BR and
metric on M/ ∼ .
4. Similarly show µ̃ ([A]) = µ (A) is a well defined function on M/ ∼ and show −1
ν(A) = m(F (A)) = m( F −1 (A)) = F∗−1 m (A)

µ̃ : (M/ ∼) → R+ is ρ̄ – continuous.

for all A ∈ BR . Show dν = F 0 dm. Use this result to prove the change of variable Exercise 19.17. Folland 2.28 on p. 60.
formula,
Exercise 19.18. Folland 2.31b and 2.31e on p. 60. (The answer in 2.13b is
Z Z
0
h ◦ F · F dm = hdm (19.32) wrong by a factor of −1 and the sum is on k = 1 to ∞. In part e, s should be
R R
taken to be a. You may also freely use the Taylor series expansion
which is valid for all Borel measurable functions h : R → [0, ∞].
Hint: Start by showing dν = F 0 dm on sets of the form A = (a, b] with X∞ ∞
(2n − 1)!! n X (2n)! n
a, b ∈ R and a < b. Then use the uniqueness assertions in Theorem 19.8 (or (1 − z)−1/2 = z = 2 z for |z| < 1.
2n n! n
n=0 4 (n!)
see Corollary 19.57) to conclude dν = F 0 dm on all of BR . To prove Eq. (19.32) n=0
apply Exercise 19.7 with g = h ◦ F and f = F −1 .
Exercise 19.19. There exists a meager (see Definition 16.6 and Theorem 16.4)
∞ subsets of R which have full Lebesgue measure, i.e. whose complement is a
Exercise 19.9. Let (X, M, µ) be a measure space and {An }n=1 ⊂ M, show
Lebesgue null set. (This is Folland 5.27. Hint: Consider the generalized Cantor
µ({An a.a.}) ≤ lim inf µ (An ) sets discussed on p. 39 of Folland.)
n→∞
and if µ (∪m≥n Am ) < ∞ for some n, then Exercise 19.20. Prove Lemma 19.49.
µ({An i.o.}) ≥ lim sup µ (An ) .

n→∞
Exercise 19.10. BRUCE: Delete this exercise which is contained in Lemma

19.17. Suppose (X, M, µ) be
R a measure space and f : X → [0 ∞] be a mea-
surable function such that X f dµ < ∞. Show µ ({f = ∞}) = 0 and the set
{f > 0} is σ – finite.
Exercise 19.11. Folland 2.13 on p. 52. Hint: “Fatou times two.”
Exercise 19.12. Folland 2.14 on p. 52. BRUCE: delete this exercise
Exercise 19.13. Give examples Rof measurable functions {fn } on R such that
fn decreases to 0 uniformly yet fn dm = ∞ for all n. Also give an example
R a sequence of measurable functions {gn } on [0, 1] such that gn → 0 while
of
gn dm = 1 for all n.
Exercise 19.14. Folland 2.19 on p. 59. (This problem is essentially covered in

the previous exercise.)
∞
Exercise 19.15. Suppose {an }n=−∞ ⊂ C is a summable sequence (i.e.
P∞ P∞ inθ
n=−∞ |an | < ∞), then f (θ) := n=−∞ an e is a continuous function for
θ ∈ R and Z π
1
an = f (θ)e−inθ dθ.
2π −π
Exercise
R 19.16. For any function f ∈ L1 (m) , show x ∈
R → (−∞,x] f (t) dm (t) is continuous in x. Also find a finite measure, µ,
R
on BR such that x → (−∞,x] f (t) dµ (t) is not continuous.

20
Multiple Integrals
In this chapter we will introduce iterated integrals and product measures. For the general (M ⊗ N , BR ) – measurable functions f : X × Y → F and
We are particularly interested in when it is permissible to interchange the order M ∈ N, let fM := 1|f |≤M f ∈ `∞ (M ⊗ N , F) . Then Eqs. (20.1) and (20.2) hold
of integration in multiple integrals. with f replaced by fM and hence for f as well by letting M → ∞.
Example 20.1. As an example let X = [1, ∞) and Y = [0, 1] equipped with Notation 20.3 (Iterated Integrals) If (X, M, µ) and (Y, N , ν) are two
their Borel σ - algebras and let µ = ν = m, where m is Lebesgue measure. The measure spaces and f : X × Y → C is a M ⊗ N – measurable function, the
iterated integrals of the function f (x, y) := e−xy − 2e−2xy satisfy, iterated integrals of f (when they make sense) are:
Z 1 Z ∞ Z 1
1 − e−y
Z Z Z Z
−xy −2xy −y
(e − 2e )dx dy = e dy ∈ (0, ∞) dµ(x) dν(y)f (x, y) := f (x, y)dν(y) dµ(x)
0 1 0 y X Y X Y
and and Z Z Z Z
Z ∞ Z 1 Z ∞
1 − e−x
dν(y) dµ(x)f (x, y) := f (x, y)dµ(x) dν(y).
(e−xy − 2e−2xy )dy dx = − e−x dx ∈ (−∞, 0) Y X Y X
1 0 1 x
Notation 20.4 Suppose that f : X → C and g : Y → C are functions, let f ⊗g
and therefore are not equal. Hence it is not always true that order of integration denote the function on X × Y given by
is irrelevant.
f ⊗ g(x, y) = f (x)g(y).
Lemma 20.2. Let F be either [0, ∞), R or C. Suppose (X, M) and (Y, N ) are
two measurable spaces and f : X × Y → F is a (M ⊗ N , BF ) – measurable Notice that if f, g are measurable, then f ⊗ g is (M ⊗ N , BC ) – measurable.
function, then for each y ∈ Y, To prove this let F (x, y) = f (x) and G(x, y) = g(y) so that f ⊗ g = F · G will
be measurable provided that F and G are measurable. Now F = f ◦ π1 where
x → f (x, y) is (M, BF ) measurable, (20.1) π1 : X × Y → X is the projection map. This shows that F is the composition
for each x ∈ X, of measurable functions and hence measurable. Similarly one shows that G is
y → f (x, y) is (N , BF ) measurable. (20.2) measurable.
Proof. Suppose that E = A × B ∈ E := M × N and f = 1E . Then

20.1 Fubini-Tonelli’s Theorem and Product Measure
f (x, y) = 1A×B (x, y) = 1A (x)1B (y)
Theorem 20.5. Suppose (X, M, µ) and (Y, N , ν) are σ-finite measure spaces
from which it follows that Eqs. (20.1) and (20.2) hold for this function. Let H
and f is a nonnegative (M ⊗ N , BR ) – measurable function, then for each y ∈ Y,
be the collection of all bounded (M ⊗ N , BF ) – measurable functions on X × Y
such that Eqs. (20.1) and (20.2) hold, here we assume F = R or C. Because x → f (x, y) is M – B[0,∞] measurable, (20.3)
measurable functions are closed under taking linear combinations and pointwise
limits, H is linear subspace of `∞ (M ⊗ N , F) which is closed under bounded for each x ∈ X,
convergence and contain 1E ∈ H for all E in the π – class, E. Therefore by by y → f (x, y) is N – B[0,∞] measurable, (20.4)
Corollary 18.54, that H = `∞ (M ⊗ N , F) .
196 20 Multiple Integrals
Z
x→ f (x, y)dν(y) is M – B[0,∞] measurable, (20.5) µm (A) = µ(Xm ∩ A) and νn (B) = ν(Yn ∩ B) for all A ∈ M and B ∈ N or
Y
Z equivalently dµm = 1Xm dµ and dνn = 1Yn dν. By what we have just proved
Eqs. (20.3) – (20.7) with µ replaced by µm and ν by νn for all (M ⊗ N , BR̄ ) –
y→ f (x, y)dµ(x) is N – B[0,∞] measurable, (20.6)
X measurable functions, f : X × Y → [0, ∞]. The validity of Eqs. (20.3) – (20.7)
then follows by passing to the limits m → ∞ and then n → ∞ making use of
and Z Z Z Z the monotone convergence theorem in the form,
dµ(x) dν(y)f (x, y) = dν(y) dµ(x)f (x, y). (20.7) Z Z Z
X Y Y X
udµm = u1Xm dµ ↑ udµ as m → ∞
Proof. Suppose that E = A × B ∈ E := M × N and f = 1E . Then X X X
and
f (x, y) = 1A×B (x, y) = 1A (x)1B (y)
Z Z Z
vdµn = v1Yn dµ ↑ vdµ as n → ∞
Y Y Y
and one sees that Eqs. (20.3) and (20.4) hold. Moreover
Z Z for all u ∈ L+ (X, M) and v ∈ L+ (Y, N ).
f (x, y)dν(y) = 1A (x)1B (y)dν(y) = 1A (x)ν(B), Corollary 20.6. Suppose (X, M, µ) and (Y, N , ν) are σ – finite measure
Y Y
spaces. Then there exists a unique measure π on M ⊗ N such that π(A × B) =
so that Eq. (20.5) holds and we have µ(A)ν(B) for all A ∈ M and B ∈ N . Moreover π is given by
Z Z Z Z Z Z
dµ(x) dν(y)f (x, y) = ν(B)µ(A). (20.8) π(E) = dµ(x) dν(y)1E (x, y) = dν(y) dµ(x)1E (x, y) (20.9)
X Y X Y Y X
Similarly, for all E ∈ M ⊗ N and π is σ – finite.

Z
Proof. Notice that any measure π such that π(A × B) = µ(A)ν(B) for
f (x, y)dµ(x) = µ(A)1B (y) and
all A ∈ M and B ∈ N is necessarily σ – finite. Indeed, let Xn ∈ M and
ZX
Yn ∈ N be chosen so that µ(Xn ) < ∞, ν(Yn ) < ∞, Xn ↑ X and Yn ↑ Y,
Z
dν(y) dµ(x)f (x, y) = ν(B)µ(A) then Xn × Yn ∈ M ⊗ N , Xn × Yn ↑ X × Y and π(Xn × Yn ) < ∞ for all n.
Y X
The uniqueness assertion is a consequence of Theorem 19.55 or see Theorem
from which it follows that Eqs. (20.6) and (20.7) hold in this case as well. For 33.6 below with E = M × N . For the existence, it suffices to observe, using the
the moment let us further assume that µ(X) < ∞ and ν(Y ) < ∞ and let H be monotone convergence theorem, that π defined in Eq. (20.9) is a measure on
the collection of all bounded (M ⊗ N , BR ) – measurable functions on X × Y M ⊗ N . Moreover this measure satisfies π(A × B) = µ(A)ν(B) for all A ∈ M
such that Eqs. (20.3) – (20.7) hold. Using the fact that measurable functions and B ∈ N from Eq. (20.8).
are closed under pointwise limits and the dominated convergence theorem (the
dominating function always being a constant), one easily shows that H closed Notation 20.7 The measure π is called the product measure of µ and ν and
under bounded convergence. Since we have just verified that 1E ∈ H for all E in will be denoted by µ ⊗ ν.
the π – class, E, it follows by Corollary 18.54 that H is the space of all bounded
Theorem 20.8 (Tonelli’s Theorem). Suppose (X, M, µ) and (Y, N , ν) are
(M ⊗ N , BR ) – measurable functions on X × Y. Finally if f : X × Y → [0, ∞]
σ – finite measure spaces and π = µ ⊗ ν is the product measure on M ⊗ N .
is a (M ⊗ N , BR̄ ) – measurable function, let fM = M ∧ f so that fM ↑ f as
If f ∈ L+ (X × Y, M ⊗ N ), then f (·, y) ∈ L+ (X, M) for all y ∈ Y, f (x, ·) ∈
M → ∞ and Eqs. (20.3) – (20.7) hold with f replaced by fM for all M ∈ N.
L+ (Y, N ) for all x ∈ X,
Repeated use of the monotone convergence theorem allows us to pass to the
limit M → ∞ in these equations to deduce the theorem in the case µ and ν Z Z
are finite measures. For the σ – finite case, choose Xn ∈ M, Yn ∈ N such that f (·, y)dν(y) ∈ L+ (X, M), f (x, ·)dµ(x) ∈ L+ (Y, N )
Xn ↑ X, Yn ↑ Y, µ(Xn ) < ∞ and ν(Yn ) < ∞ for all m, n ∈ N. Then define Y X

20.1 Fubini-Tonelli’s Theorem and Product Measure 197
and which implies that µ (E) = 0. Let f± be the positive and negative parts of f,
Z Z Z then using the above convention we have
f dπ = dµ(x) dν(y)f (x, y) (20.10) Z Z
X×Y
ZX ZY f (x, y) dν (y) = 1E c (x) f (x, y) dν (y)
Y
= dν(y) dµ(x)f (x, y). (20.11) ZY
Y X = 1E c (x) [f+ (x, y) − f− (x, y)] dν (y)
Y
Proof. By Theorem 20.5 and Corollary 20.6, the theorem holds when Z Z
f = 1E with E ∈ M ⊗ N . Using the linearity of all of the statements, the = 1E c (x) f+ (x, y) dν (y) − 1E c (x) f− (x, y) dν (y) .
theorem is also true for non-negative simple functions. Then using the mono- Y Y
tone convergence theorem repeatedly along with the approximation Theorem (20.16)
18.42, one deduces the theorem for general f ∈ L+ (X × Y, M ⊗ N ). Noting that 1E c (x) f± (x, y) = (1E c ⊗ 1Y · f± ) (x, y) is a positive M ⊗ N –
The following convention will be in force for the rest of this chapter. measurable Rfunction, it follows from another application of Tonelli’s theorem
Convention: If (X, M, µ) is a measure R space and f : X → C is a mea- that x → Y f (x, y) dν (y) is M – measurable, being the difference of two
surableR but non-integrable function, i.e. X |f | dµ = ∞, by convention we will measurable functions. Moreover
R function (i.e. f : X → [0, ∞])
define X f dµ := 0. However if f is a non-negative Z Z Z Z
is a non-integrable function we will still write X f dµ = ∞.
f (x, y) dν (y) dµ (x) ≤ |f (x, y)| dν (y) dµ (x) < ∞,

X Y X Y
Theorem 20.9 (Fubini’s Theorem). Suppose (X, M, µ) and (Y, N , ν) are
which shows Y f (·, y)dv(y) ∈ L1 (µ). Integrating Eq. (20.16) on x and using
R
σ – finite measure spaces, π = µ ⊗ ν is the product measure on M ⊗ N and
f : X × Y → C is a M ⊗ N – measurable function. Then the following three Tonelli’s theorem repeatedly implies,
conditions are equivalent: Z Z
Z f (x, y) dν (y) dµ (x)
|f | dπ < ∞, i.e. f ∈ L1 (π), (20.12) X Y
Z Z Z Z
X×Y
Z Z = dµ (x) dν (y) 1E c (x) f+ (x, y) − dµ (x) dν (y) 1E c (x) f− (x, y)
|f (x, y)| dν(y) dµ(x) < ∞ and (20.13) ZX ZY ZX ZY
X Y
Z Z = dν (y) dµ (x) 1E c (x) f+ (x, y) − dν (y) dµ (x) 1E c (x) f− (x, y)
Y X Y X
|f (x, y)| dµ(x) dν(y) < ∞. (20.14) Z Z Z Z
Y X = dν (y) dµ (x) f+ (x, y) − dν (y) dµ (x) f− (x, y)
If any one (and hence all) of these fR(x, ·) ∈ L1 (ν) for µ-a.e. ZY X
Z Y X
R condition hold, then
Z Z
x, f (·, y) ∈ L (µ) for ν-a.e. y, Y f (·, y)dv(y) ∈ L (µ), X f (x, ·)dµ(x) ∈ L1 (ν)
1 1 = f+ dπ − f− dπ = (f+ − f− ) dπ = f dπ (20.17)
X×Y X×Y X×Y X×Y
and Eqs. (20.10) and (20.11) are still valid.
which proves Eq. (20.10) holds.
Proof. The equivalence of Eqs. (20.12) – (20.14) is a direct consequence of
Now suppose that f = u + iv is complex valued and again let E be as in Eq.
Tonelli’s Theorem 20.8. Now suppose f ∈ L1 (π) is a real valued function and
(20.15). Just as above we still have E ∈ M and µ (E) = 0. By our convention,
let Z Z Z Z
E := x ∈ X : |f (x, y)| dν (y) = ∞ . (20.15) f (x, y) dν (y) = 1E c (x) f (x, y) dν (y) = 1E c (x) [u (x, y) + iv (x, y)] dν (y)
Y Y
R ZY ZY
Then by Tonelli’s theorem, x → Y |f (x, y)| dν (y) is measurable and hence
= 1E c (x) u (x, y) dν (y) + i 1E c (x) v (x, y) dν (y)
E ∈ M. Moreover Tonelli’s theorem implies Y Y
Z Z Z
which is measurable in x by what we have just proved. Similarly one shows
|f (x, y)| dν (y) dµ (x) = |f | dπ < ∞ R
f (·, y) dν (y) ∈ L1 (µ) and Eq. (20.10) still holds by a computation similar
X Y X×Y Y

R
to that done in Eq. (20.17). The assertions pertaining to Eq. (20.11) may be exists and equals 0 for µ a.e. x and similarly that X h(x, y)dµ(x) exists and
proved in the same way. equals 0 for ν a.e. y. Therefore
Notation 20.10 Given E ⊂ X × Y and x ∈ X, let
Z Z Z Z Z
0= hdλ = hdµ dν = hdν dµ.
X×Y Y X X Y
xE := {y ∈ Y : (x, y) ∈ E}.
For general f ∈ L1 (λ), we may choose g ∈ L1 (M⊗N , µ⊗ν) such that f (x, y) =
Similarly if y ∈ Y is given let
g(x, y) for λ− a.e. (x, y). Define h := f − g. Then h = 0, λ− a.e. Hence by what
Ey := {x ∈ X : (x, y) ∈ E}. we have just proved and Theorem 20.8 f = g + h has the following properties:
1. For µ a.e. x, y → f (x, y) = g(x, y) + h(x, y) is in L1 (ν) and
If f : X × Y → C is a function let fx = f (x, ·) and f y := f (·, y) so that
fx : Y → C and f y : X → C. Z Z
f (x, y)dν(y) = g(x, y)dν(y).
Theorem 20.11. Suppose (X, M, µ) and (Y, N , ν) are complete σ – finite mea- Y Y
sure spaces. Let (X × Y, L, λ) be the completion of (X × Y, M ⊗ N , µ ⊗ ν). If f
2. For ν a.e. y, x → f (x, y) = g(x, y) + h(x, y) is in L1 (µ) and
is L – measurable and (a) f ≥ 0 or (b) f ∈ L1 (λ) then fx is N – measurable
for µ a.e. x and f y is M – measurable for ν a.e. y and in case (b) fx ∈ L1 (ν) Z Z
and f y ∈ L1 (µ) for µ a.e. x and ν a.e. y respectively. Moreover, f (x, y)dµ(x) = g(x, y)dµ(x).
X X
Z Z
x→ 1
fx dν ∈ L (µ) and y → f dµ ∈ L1 (ν)
y From these assertions and Theorem 20.8, it follows that
Y X Z Z Z Z
and dµ(x) dν(y)f (x, y) = dµ(x) dν(y)g(x, y)
X Y
ZX ZY
Z Z Z Z Z
f dλ = dν dµ f = dµ dν f.
X×Y Y X X Y = dν(y) dν(x)g(x, y)
ZY Y
Proof. If E ∈ M ⊗ N is a µ ⊗ ν null set (i.e. (µ ⊗ ν)(E) = 0), then
= g(x, y)d(µ ⊗ ν)(x, y)
ZX×Y
Z Z
0 = (µ ⊗ ν)(E) = ν(x E)dµ(x) = µ(Ey )dν(y).
= f (x, y)dλ(x, y).
X X X×Y
This shows that Similarly it is shown that

µ({x : ν(x E) 6= 0}) = 0 and ν({y : µ(Ey ) 6= 0}) = 0,
Z Z Z
dν(y) dµ(x)f (x, y) = f (x, y)dλ(x, y).
Y X X×Y
i.e. ν(x E) = 0 for µ a.e. x and µ(Ey ) = 0 for ν a.e. y. If h is L measurable and
h = 0 for λ – a.e., then there exists E ∈ M ⊗ N such that {(x, y) : h(x, y) 6=
0} ⊂ E and (µ ⊗ ν)(E) = 0. Therefore |h(x, y)| ≤ 1E (x, y) and (µ ⊗ ν)(E) = 0. The previous theorems have obvious generalizations to products of any finite
Since number of σ – finite measure spaces. For example the following theorem holds.
n
{hx 6= 0} = {y ∈ Y : h(x, y) 6= 0} ⊂ x E and Theorem 20.12. Suppose {(Xi , Mi , µi )}i=1 are σ – finite measure spaces
{hy 6= 0} = {x ∈ X : h(x, y) 6= 0} ⊂ Ey and X := X1 × · · · × Xn . Then there exists a unique measure, π, on
(X, M1 ⊗ · · · ⊗ Mn ) such that
we learn that for µ a.e. x and ν a.e. y that {hx 6= 0} ∈ M, {h
R y 6= 0} ∈ N ,
ν({hx 6= 0}) = 0 and a.e. and µ({hy 6= 0}) = 0. This implies Y h(x, y)dν(y) π(A1 × · · · × An ) = µ1 (A1 ) . . . µn (An ) for all Ai ∈ Mi .

20.1 Fubini-Tonelli’s Theorem and Product Measure 199
Z
(This measure and its completion will be denoted by µ1 ⊗ · · · ⊗ µn .) If f : X → M sin x 1 e−M Λ
−Λx
e dx − π + arctan Λ ≤ C (20.21)

[0, ∞] is a M1 ⊗ · · · ⊗ Mn – measurable function then
0 x 2 M
Z Z Z
f dπ = dµσ(1) (xσ(1) ) . . . dµσ(n) (xσ(n) ) f (x1 , . . . , xn ) (20.18) where C = maxx≥0 1+x
= √1 ∼
= 1.2. In particular Eq. (20.19) is valid.
1+x2 2 2−2
X Xσ(1) Xσ(n)
To verify these assertions, first notice that by the fundamental theorem of
where σ is any permutation of {1, 2, . . . , n}. This equation also holds for any calculus,
f ∈ L1 (π) and moreover, f ∈ L1 (π) iff
Z x Z x Z x

|sin x| = cos ydy ≤ |cos y| dy ≤ 1dy = |x|
Z Z
dµσ(1) (xσ(1) ) . . . dµσ(n) (xσ(n) ) |f (x1 , . . . , xn )| < ∞ 0 0 0
Xσ(1) Xσ(n)

so sinx x ≤ 1 for all x 6= 0. Making use of the identity
for some (and hence all) permutations, σ.
Z ∞
This theorem can be proved by the same methods as in the two factor case, e−tx dt = 1/x
see Exercise 20.5. Alternatively, one can use the theorems already proved and 0
induction on n, see Exercise 20.6 in this regard. and Fubini’s theorem,

Example 20.13. In this example we will show Z M
sin x −Λx
Z M Z ∞
e dx = dx sin x e−Λx e−tx dt
Z M
sin x 0 x 0 0
lim dx = π/2. (20.19) Z ∞ Z M
M →∞ 0 x = dt dx sin x e−(Λ+t)x
R∞ 0 0
To see this write 1
= e−tx dt and use Fubini-Tonelli to conclude that ∞
1 − (cos M + (Λ + t) sin M ) e−M (Λ+t)
Z
x 0
= 2 dt
Z M
sin x
Z M Z ∞ 0 (Λ + t) + 1
−tx
dx = e sin x dt dx Z ∞
1
Z ∞
cos M + (Λ + t) sin M −M (Λ+t)
0 x 0 0 = dt − e dt
2 2
(Λ + t) + 1 (Λ + t) + 1
"Z #
Z ∞ M 0 0
= e−tx sin x dx dt 1
0 0 = π − arctan Λ − ε(M, Λ) (20.22)
Z ∞
2
1
1 − te−M t sin M − e−M t cos M dt

= 2 where
1 + t Z ∞
cos M + (Λ + t) sin M
Z0 ∞ ε(M, Λ) = e−M (Λ+t) dt.
1 π 2
→ dt = as M → ∞, 0 (Λ + t) + 1
1 + t 2 2
0
Since
cos M + (Λ + t) sin M 1 + (Λ + t)
wherein we have used the dominated convergence theorem to pass to the limit.
≤ ≤ C,

2 2
(Λ + t) + 1 (Λ + t) + 1
The next example is a refinement of this result.
Z ∞
e−M Λ
Example 20.14. We have |ε(M, Λ)| ≤ e−M (Λ+t) dt = C .
0 M
Z ∞
sin x −Λx 1 This estimate along with Eq. (20.22) proves Eq. (20.21) from which Eq. (20.19)
e dx = π − arctan Λ for all Λ > 0 (20.20)
0 x 2 follows by taking Λ → ∞ and Eq. (20.20) follows (using the dominated conver-
gence theorem again) by letting M → ∞.
and forΛ, M ∈ [0, ∞),

20.2 Lebesgue Measure on Rd and the Change of Theorem 20.19 (Change of Variables Theorem). Let Ω ⊂o Rd be an open
Variables Theorem set and T : Ω → T (Ω) ⊂o Rd be a C 1 – diffeomorphism,1 see Figure 20.1. Then
for any Borel measurable function, f : T (Ω) → [0, ∞],
Notation 20.15 Let Z Z
d times
f (T (x)) | det T 0 (x) |dx = f (y) dy, (20.23)
d times
Ω
z }| { z }| { T (Ω)
d
m := m ⊗ · · · ⊗ m on BRd = BR ⊗ · · · ⊗ BR
where T 0 (x) is the linear transformation on Rd defined by T 0 (x)v := dt
d
|0 T (x +
be the d – fold product of Lebesgue measure m on BR . We will also use md d 0
tv). More explicitly, viewing vectors in R as columns, T (x) may be represented
to denote its completion and let Ld be the completion of BRd relative to md .
by the matrix
A subset A ∈ Ld is called a Lebesgue measurable set and md is called d –  
∂1 T1 (x) . . . ∂d T1 (x)
dimensional Lebesgue measure, or just Lebesgue measure for short. .. ..
T 0 (x) = 
 .. ,

(20.24)
. . .
Definition 20.16. A function f : Rd → R is Lebesgue measurable if ∂1 Td (x) . . . ∂d Td (x)
f −1 (BR ) ⊂ Ld .
i.e. the i - j – matrix entry of T 0 (x) is given by T 0 (x)ij = ∂i Tj (x) where
d
Notation 20.17 I will often be sloppy in the sequel and write m for m and T (x) = (T1 (x), . . . , Td (x))tr and ∂i = ∂/∂xi .
dx for dm(x) = dmd (x), i.e.
Z Z Z
f (x) dx = f dm = f dmd .
Rd Rd Rd
Hopefully the reader will understand the meaning from the context.
Theorem 20.18. Lebesgue measure md is translation invariant. Moreover md

is the unique translation invariant measure on BRd such that md ((0, 1]d ) = 1.
Proof. Let A = J1 × · · · × Jd with Ji ∈ BR and x ∈ Rd . Then
x + A = (x1 + J1 ) × (x2 + J2 ) × · · · × (xd + Jd )
and therefore by translation invariance of m on BR we find that
md (x + A) = m(x1 + J1 ) . . . m(xd + Jd ) = m(J1 ) . . . m(Jd ) = md (A)
and hence md (x + A) = md (A) for all A ∈ BRd by Corollary 19.57. From this
fact we see that the measure md (x + ·) and md (·) have the same null sets. Using
this it is easily seen that m(x + A) = m(A) for all A ∈ Ld . The proof of the
Fig. 20.1. The geometric setup of Theorem 20.19.
second assertion is Exercise 20.7.
Exercise 20.1. In this problem you are asked to show there is no reasonable
notion of Lebesgue measure on an infinite dimensional Hilbert space. To be
Remark 20.20. Theorem 20.19 is best remembered as the statement: if we make
more precise, suppose H is an infinite dimensional Hilbert space and m is a
the change of variables y = T (x) , then dy = | det T 0 (x) |dx. As usual, you must
countably additive measure on BH which is invariant under translations and
satisfies, m(B0 (ε)) > 0 for all ε > 0. Show m(V ) = ∞ for all non-empty open 1
That is T : Ω → T (Ω) ⊂o Rd is a continuously differentiable bijection and the
subsets V ⊂ H. inverse map T −1 : T (Ω) → Ω is also continuously differentiable.

20.2 Lebesgue Measure on Rd and the Change of Variables Theorem 201
Z Z
also change the limits of integration appropriately, i.e. if x ranges through Ω 1(an ,bn ) · (f ◦ T ) · |T 0 |dm = 1T ((an ,bn )) · f ◦ T · |T 0 |dm

then y must range through T (Ω) .
Ω Ω
Z
Proof. The proof will be by induction on d. The case d = 1 was essentially = lim 1T ([αk ,βk ]) · f ◦ T · |T 0 | dm

done in Exercise 19.8. Nevertheless, for the sake of completeness let us give k→∞
Ω
a proof here. Suppose d = 1, a < α < β < b such that [a, b] is a compact Z
subinterval of Ω. Then | det T 0 | = |T 0 | and = lim 1T ([αk ,βk ]) · f dm
k→∞
Z Z Z β T (Ω)
1T ((α,β]) (T (x)) |T 0 (x)| dx = 1(α,β] (x) |T 0 (x)| dx = |T 0 (x)| dx.
Z
[a,b] [a,b] α = 1T ((an ,bn )) · f dm.
T (Ω)
If T 0 (x) > 0 on [a, b] , then
Z β Z β Summing this equality on n, then shows Eq. (20.23) holds.
0 0 To carry out the induction step, we now suppose d > 1 and suppose the
|T (x)| dx = T (x) dx = T (β) − T (α)
α α
Z theorem is valid with d being replaced by d − 1. For notational compactness, let
= m (T ((α, β])) = 1T ((α,β]) (y) dy us write vectors in Rd as row vectors rather than column vectors. Nevertheless,
T ([a,b]) the matrix associated to the differential, T 0 (x) , will always be taken to be given
as in Eq. (20.24).
while if T 0 (x) < 0 on [a, b] , then Case 1. Suppose T (x) has the form
Z β Z β
T (x) = (xi , T2 (x) , . . . , Td (x)) (20.26)
|T 0 (x)| dx = − T 0 (x) dx = T (α) − T (β)
α α
Z or
= m (T ((α, β])) = 1T ((α,β]) (y) dy. T (x) = (T1 (x) , . . . , Td−1 (x) , xi ) (20.27)
T ([a,b])
for some i ∈ {1, . . . , d} . For definiteness we will assume T is as in Eq. (20.26),
Combining the previous three equations shows the case of T in Eq. (20.27) may be handled similarly. For t ∈ R, let it : Rd−1 →
Z Z Rd be the inclusion map defined by
0
f (T (x)) |T (x)| dx = f (y) dy (20.25)
[a,b] T ([a,b]) it (w) := wt := (w1 , . . . , wi−1 , t, wi+1 , . . . , wd−1 ) ,
whenever f is of the form f = 1T ((α,β]) with a < α < β < b. An application Ωt be the (possibly empty) open subset of Rd−1 defined by
of Dynkin’s multiplicative system Theorem 18.51 then implies that Eq. (20.25)
holds for every bounded measurable function f : T ([a, b]) → R. (Observe that Ωt := w ∈ Rd−1 : (w1 , . . . , wi−1 , t, wi+1 , . . . , wd−1 ) ∈ Ω

|T 0 (x)| is continuous and hence bounded for x in the compact interval, [a, b] .)
and Tt : Ωt → Rd−1 be defined by
`N
From Exercise 13.14, Ω = n=1 (an , bn ) where an , bn ∈ R∪ {±∞} for n =
1, 2, · · · < N with N = ∞ possible. Hence if f : T (Ω) → R + is a Borel
measurable function and an < αk < βk < bn with αk ↓ an and βk ↑ bn , then by Tt (w) = (T2 (wt ) , . . . , Td (wt )) ,
what we have already proved and the monotone convergence theorem
see Figure 20.2. Expanding det T 0 (wt ) along the first row of the matrix T 0 (wt )
shows
|det T 0 (wt )| = |det Tt0 (w)| .
Now by the Fubini-Tonelli Theorem and the induction hypothesis,

Case 2. (Eq. (20.23) is true locally.) Suppose that T : Ω → Rd is a general

map as in the statement of the theorem and x0 ∈ Ω is an arbitrary point. We
will now show there exists an open neighborhood W ⊂ Ω of x0 such that
Z Z
f ◦ T | det T 0 |dm = f dm
T (W )
W
holds for all Borel measurable function, f : T (W ) → [0, ∞]. Let Mi be the 1-i
minor of T 0 (x0 ) , i.e. the determinant of T 0 (x0 ) with the first row and ith –
column removed. Since
d
X i+1
0 6= det T 0 (x0 ) = (−1) ∂i Tj (x0 ) · Mi ,
i=1
Fig. 20.2. In this picture d = i = 3 and Ω is an egg-shaped region with an egg-shaped
hole. The picture indicates the geometry associated with the map T and slicing the there must be some i such that Mi 6= 0. Fix an i such that Mi 6= 0 and let,
set Ω along planes where x3 = t.
S (x) := (xi , T2 (x) , . . . , Td (x)) . (20.28)
Z Z Observe that |det S 0 (x0 )| = |Mi | =
6 0. Hence by the inverse function Theorem
f ◦ T | det T 0 |dm = 1Ω · f ◦ T | det T 0 |dm 12.25, there exist an open neighborhood W of x0 such that W ⊂o Ω and
Ω Rd S (W ) ⊂o Rd and S : W → S (W ) is a C 1 – diffeomorphism. Let R : S (W ) →
Z T (W ) ⊂o Rd to be the C 1 – diffeomorphism defined by
= 1Ω (wt ) (f ◦ T ) (wt ) | det T 0 (wt ) |dwdt
Rd R (z) := T ◦ S −1 (z) for all z ∈ S (W ) .
 
Z Z
 (f ◦ T ) (wt ) | det T 0 (wt ) |dw dt Because
=
R
Ωt (T1 (x) , . . . , Td (x)) = T (x) = R (S (x)) = R ((xi , T2 (x) , . . . , Td (x)))
 
Z Z
=  f (t, Tt (w)) | det Tt0 (w) |dw dt for all x ∈ W, if
R
Ωt (z1 , z2 , . . . , zd ) = S (x) = (xi , T2 (x) , . . . , Td (x))
   
Z Z Z Z
=

f (t, z) dz  dt =
  1T (Ω) (t, z) f (t, z) dz  dt then
R (z) = T1 S −1 (z) , z2 , . . . , zd .

R R (20.29)
Tt (Ωt ) Rd−1
Z Observe that S is a map of the form in Eq. (20.26), R is a map of the form in Eq.
= f (y) dy (20.27), T 0 (x) = R0 (S (x)) S 0 (x) (by the chain rule) and (by the multiplicative
T (Ω) property of the determinant)
wherein the last two equalities we have used Fubini-Tonelli along with the iden- |det T 0 (x)| = | det R0 (S (x)) | |det S 0 (x)| ∀ x ∈ W.
tity;
So if f : T (W ) → [0, ∞] is a Borel measurable function, two applications of the
a a
T (Ω) = T (it (Ω)) = {(t, z) : z ∈ Tt (Ωt )} .
t∈R t∈R
results in Case 1. shows,

20.2 Lebesgue Measure on Rd and the Change of Variables Theorem 203
Z Z
f ◦ T · | det T 0 |dm = (f ◦ R · | det R0 |) ◦ S · |det S 0 | dm Eq. (20.30) is implies Eq. (20.23) with Ω = (a, b) . On the other hand if T 0 < 0
on (a, b) then T ((a, b)) = (T (b) , T (a)) and Eq. (20.30) is equivalent to
W W
Z Z
0
Z Z T (a) Z
= f ◦ R · | det R |dm = f dm f (T (x)) (− |T 0 (x)|) dx = − f (y) dy = − f (y) dy
S(W ) R(S(W )) (a,b) T (b) T ((a,b))
Z
= f dm which is again implies Eq. (20.23). On the other hand Eq. Eq. (20.30) is more
T (W ) general than Eq. (20.23) since it does not require T to be injective. The standard
proof of Eq. (20.30) is as follows. For z ∈ T ([a, b]) , let
and Case 2. is proved.
Case 3. (General Case.) Let f : Ω → [0, ∞] be a general non-negative Borel Z z
measurable function and let F (z) := f (y) dy.
T (a)
c
Kn := {x ∈ Ω : dist(x, Ω ) ≥ 1/n and |x| ≤ n} . Then by the chain rule and the fundamental theorem of calculus,
Then each Kn is a compact subset of Ω and Kn ↑ Ω as n → ∞. Using the Z b Z b Z b
d
0 0 0
compactness of Kn and case 2, for each n ∈ N, there is a finite open cover Wn f (T (x)) T (x) dx = F (T (x)) T (x) dx = [F (T (x))] dx
a a a dx
of Kn such that W ⊂ Ω and Eq. (20.23) holds with Ω replaced by W for each T (b)
∞
Z
W ∈ Wn . Let {Wi }i=1 be an enumeration of ∪∞ n=1 W`
n and set W̃1 = W1 and = F (T (x)) |ba = f (y) dy.
∞
W̃i := Wi \ (W1 ∪ · · · ∪ Wi−1 ) for all i ≥ 2. Then Ω = i=1 W̃i and by repeated T (a)
use of case 2.,
An application of Dynkin’s multiplicative systems theorem (in the form of
Z ∞ Z
X Corollary 18.55) now shows that Eq. (20.30) holds for all bounded measur-
f ◦ T | det T 0 |dm = 1W̃i · (f ◦ T ) · | det T 0 |dm able functions f on (a, b) . Then by the usual truncation argument, it also holds
Ω i=1 Ω for all positive measurable functions on (a, b) .
∞ Z h
X i
= 1T (W̃i ) f ◦ T · | det T 0 |dm Example 20.22. Continuing the setup in Theorem 20.19, if A ∈ BΩ , then
i=1W Z Z
i
∞
X Z n
X Z m (T (A)) = 1T (A) (y) dy = 1T (A) (T x) |det T 0 (x)| dx
= 1T (W̃i ) · f dm = 1T (W̃i ) · f dm Rd Rd
Z
i=1
T (Wi )
i=1
T (Ω) = 1A (x) |det T 0 (x)| dx
Z Rd
= f dm.
wherein the second equality we have made the change of variables, y = T (x) .
T (Ω) Hence we have shown
d (m ◦ T ) = |det T 0 (·)| dm.

Remark 20.21. When d = 1, one often learns the change of variables formula as
In particular if T ∈ GL(d, R) = GL(Rd ) – the space of d × d invertible matrices,
Z b Z T (b) then m ◦ T = |det T | m, i.e.
f (T (x)) T 0 (x) dx = f (y) dy (20.30)
a T (a) m (T (A)) = |det T | m (A) for allA ∈ BRd . (20.31)
where f : [a, b] → R is a continuous function and T is C 1 – function defined in This equation also shows that m ◦ T and m have the same null sets and hence
a neighborhood of [a, b] . If T 0 > 0 on (a, b) then T ((a, b)) = (T (a) , T (b)) and the equality in Eq. (20.31) is valid for any A ∈ Ld .

2
Ux Vy − Uy Vx = Ux2 + Vx2 = |f 0 | .

Exercise 20.2. Show that f ∈ L1 T (Ω) , md iff
Z
Therefore
|f ◦ T | | det T 0 |dm < ∞ 2
dudv = |f 0 (x + iy)| dxdy.
Ω
Example 20.25. In this example we will evaluate the integral
and if f ∈ L1 T (Ω) , md , then Eq. (20.23) holds.
ZZ
x4 − y 4 dxdy

Example 20.23 (Polar Coordinates). Suppose T : (0, ∞) × (0, 2π) → R2 is de- I :=
Ω
fined by
x = T (r, θ) = (r cos θ, r sin θ) , where
Ω = (x, y) : 1 < x2 − y 2 < 2, 0 < xy < 1 ,

i.e. we are making the change of variable,
see Figure 20.3. We are going to do this by making the change of variables,
x1 = r cos θ and x2 = r sin θ for 0 < r < ∞ and 0 < θ < 2π.
In this case
cos θ − r sin θ
T 0 (r, θ) =
sin θ r cos θ
and therefore
dx = |det T 0 (r, θ)| drdθ = rdrdθ.
Observing that
R2 \ T ((0, ∞) × (0, 2π)) = ` := {(x, 0) : x ≥ 0}
has m2 – measure zero, it follows from the change of variables Theorem 20.19
that Z Z 2π Z ∞
f (x)dx = dθ dr r · f (r (cos θ, sin θ)) (20.32)
R2 0 0
Fig. 20.3. The region Ω consists of the two curved rectangular regions shown.
for any Borel measurable function f : R2 → [0, ∞].
Example 20.24 (Holomorphic Change of Variables). Suppose that f : Ω ⊂o C ∼ =
R2 → C is an injective holomorphic function such that f 0 (z) 6= 0 for all z ∈ Ω.
(u, v) := T (x, y) = x2 − y 2 , xy ,

We may express f as
f (x + iy) = U (x, y) + iV (x, y) in which case

for all z = x + iy ∈ Ω. Hence if we make the change of variables, 2x −2y
dxdy = 2 x2 + y 2 dxdy

dudv = det
y x
w = u + iv = f (x + iy) = U (x, y) + iV (x, y)
Notice that
then
Ux Uy 1
x4 − y 4 = x2 − y 2 x2 + y 2 = u x2 + y 2 = ududv.

dudv = det dxdy = |Ux Vy − Uy Vx | dxdy.
Vx Vy 2
Recalling that U and V satisfy the Cauchy Riemann equations, Ux = Vy and The function T is not injective on Ω but it is injective on each of its connected
Uy = −Vx with f 0 = Ux + iVx , we learn components. Let D be the connected component in the first quadrant so that

20.3 The Polar Decomposition of Lebesgue Measure 205
Ω = −D ∪ D and T (±D) = (1, 2) × (0, 1) . The change of variables theorem Proof. By Tonelli’s theorem and induction,
then implies Z
2 2
Id (a) = e−a|y| e−at md−1 (dy) dt
1 u2 2
ZZ ZZ
4 4
1 3 Rd−1 ×R
I± := x − y dxdy = ududv = |1 · 1 =
±D 2 (1,2)×(0,1) 2 2 4 = Id−1 (a)I1 (a) = I1d (a). (20.33)
and therefore I = I+ + I− = 2 · (3/4) = 3/2. So it suffices to compute:
Exercise 20.3 (Spherical Coordinates). Let T : (0, ∞) × (0, π) × (0, 2π) →
Z Z
2 2 2
R3 be defined by I2 (a) = e−a|x| dm(x) = e−a(x1 +x2 ) dx1 dx2 .

R2 R2 \{0}
T (r, φ, θ) = (r sin φ cos θ, r sin φ sin θ, r cos φ)
Using polar coordinates, see Eq. (20.32), we find,
= r (sin φ cos θ, sin φ sin θ, cos φ) ,
Z ∞ Z 2π Z ∞
2 2
see Figure 20.4. By making the change of variables x = T (r, φ, θ) , show I2 (a) = dr r dθ e−ar = 2π re−ar dr
0 0 0
M 2 M
e−ar
Z Z
−ar 2 2π
= 2π lim re dr = 2π lim = = π/a.
M →∞ 0 M →∞ −2a 0 2a
This shows that I2 (a) = π/a and the result now follows from Eq. (20.33).
20.3 The Polar Decomposition of Lebesgue Measure

Let
d
X
2
S d−1 = {x ∈ Rd : |x| := x2i = 1}
i=1
Fig. 20.4. The relation of x to (r, φ, θ) in spherical coordinates.
be the unit sphere in Rd equipped with its Borel σ – algebra, BS d−1 and Φ :
−1
Rd \ {0} → (0, ∞) × S d−1 be defined by Φ(x) := (|x| , |x| x). The inverse map,
Φ−1 : (0, ∞) × S d−1 → Rd \ {0} , is given by Φ−1 (r, ω) = rω. Since Φ and Φ−1
Z Z π Z 2π Z ∞ are continuous, they are both Borel measurable. For E ∈ BS d−1 and a > 0, let
f (x)dx = dφ dθ dr r2 sin φ · f (T (r, φ, θ))
R3 0 0 0
Ea := {rω : r ∈ (0, a] and ω ∈ E} = Φ−1 ((0, a] × E) ∈ BRd .
for any Borel measurable function, f : R3 → [0, ∞].
Definition 20.27. For E ∈ BS d−1 , let σ(E) := d · m(E1 ). We call σ the surface
Lemma 20.26. Let a > 0 and measure on S d−1 .
It is easy to check that σ is a measure. Indeed if E ∈ BS d−1 , then `∞E1 =

Z
2
Id (a) := e−a|x| dm(x). Φ−1 ((0, 1] × E) ∈ B so that m(E ) is well defined. Moreover if E =
`∞ R d 1 i=1 Ei ,
Rd then E1 = i=1 (Ei )1 and
Then Id (a) = (π/a)d/2 . ∞
X ∞
X
σ(E) = d · m(E1 ) = m ((Ei )1 ) = σ(Ei ).
i=1 i=1

The intuition behind this definition is as follows. If E ⊂ S d−1 is a set and ε > 0 Proof. By Exercise 19.7,
is a small number, then the volume of Z Z Z
f ◦ Φ−1 ◦ Φ dm = f ◦ Φ−1

f dm = d (Φ∗ m) (20.36)
(1, 1 + ε] · E = {rω : r ∈ (1, 1 + ε] and ω ∈ E}
Rd Rd \{0} (0,∞)×S d−1
should be approximately given by m ((1, 1 + ε] · E) ∼

= σ(E)ε, see Figure 20.5 and therefore to prove Eq. (20.34) we must work out the measure Φ∗ m on
below. On the other hand
B(0,∞) ⊗ BS d−1 defined by
Φ∗ m(A) := m Φ−1 (A) ∀ A ∈ B(0,∞) ⊗ BS d−1 .

(20.37)
If A = (a, b] × E with 0 < a < b and E ∈ BS d−1 , then
Φ−1 (A) = {rω : r ∈ (a, b] and ω ∈ E} = bE1 \ aE1
wherein we have used Ea = aE1 in the last equality. Therefore by the basic
scaling properties of m and the fundamental theorem of calculus,
(Φ∗ m) ((a, b] × E) = m (bE1 \ aE1 ) = m(bE1 ) − m(aE1 )

Z b
Fig. 20.5. Motivating the definition of surface measure for a sphere. = bd m(E1 ) − ad m(E1 ) = d · m(E1 ) rd−1 dr. (20.38)
a
Letting dρ(r) = rd−1 dr, i.e.

Z
m ((1, 1 + ε]E) = m (E1+ε \ E1 ) = (1 + ε)d − 1 m(E1 ).

ρ(J) = rd−1 dr ∀ J ∈ B(0,∞) , (20.39)
J
Therefore we expect the area of E should be given by
Eq. (20.38) may be written as

(1 + ε)d − 1 m(E1 )
σ(E) = lim = d · m(E1 ). (Φ∗ m) ((a, b] × E) = ρ((a, b]) · σ(E) = (ρ ⊗ σ) ((a, b] × E) . (20.40)
ε↓0 ε
Since
The following theorem is motivated by Example 20.23 and Exercise 20.3. E = {(a, b] × E : 0 < a < b and E ∈ BS d−1 } ,
d
Theorem 20.28 (Polar Coordinates). If f : R → [0, ∞] is a (BRd , B)– is a π class (in fact it is an elementary class) such that σ(E) = B(0,∞) ⊗ BS d−1 ,
measurable function then it follows from Theorem 19.55 and Eq. (20.40) that Φ∗ m = ρ ⊗ σ. Using this
Z Z result in Eq. (20.36) gives
f (x)dm(x) = f (rω)rd−1 drdσ(ω). (20.34) Z Z
f ◦ Φ−1 d (ρ ⊗ σ)

Rd (0,∞)×S d−1 f dm =
Rd (0,∞)×S d−1
In particular if f : R+ → R+ is measurable then
Z Z ∞ which combined with Tonelli’s Theorem 20.8 proves Eq. (20.36).
f (|x|)dx = f (r)dV (r) (20.35) Corollary 20.29. The surface area σ(S d−1 ) of the unit sphere S d−1 ⊂ Rd is
0
Rd
2π d/2
d
where V (r) = m (B(0, r)) = r m (B(0, 1)) = d −1
σ S d−1
d
r . σ(S d−1 ) = (20.41)
Γ (d/2)

20.4 More proofs of the classical Weierstrass approximation Theorem 10.35 207
where Γ is the gamma function given by 20.4 More proofs of the classical Weierstrass
Z ∞ approximation Theorem 10.35
Γ (x) := ux−1 e−u dr (20.42)
0 In each of these proofs we will use the reduction explained the previous proof
√ of Theorem 10.35 to reduce to the case where f ∈ C([0, 1]d ). The first proof we
Moreover, Γ (1/2) = π, Γ (1) = 1 and Γ (x + 1) = xΓ (x) for x > 0.
will give here is based on the “weak law” of large numbers. The second will be
Proof. Using Theorem 20.28 we find another approximate δ – function argument.
Z ∞ Proof. of Theorem 10.35. Let 0 : = (0, 0, . . . , 0), 1 : = (1, 1, . . . , 1) and
Z Z ∞
2 2 [0, 1] := [0, 1]d . By considering the real and imaginary parts of f separately,
Id (1) = dr rd−1 e−r dσ = σ(S d−1 ) rd−1 e−r dr.
0 0 it suffices to assume f ∈ C([0, 1], R). For x ∈ [0, 1], let νx be the measure on
S d−1 {0, 1} such that νx ({0}) = 1 − x and νx ({1}) = x. Then
We simplify this last integral by making the change of variables u = r2 so that Z
r = u1/2 and dr = 21 u−1/2 du. The result is ydνx (y) = 0 · (1 − x) + 1 · x = x and (20.44)
{0,1}
Z ∞ Z ∞ Z
2 d−1 1 (y − x)2 dνx (y) = x2 (1 − x) + (1 − x)2 · x = x(1 − x).
rd−1 e−r dr = u 2 e−u u−1/2 du (20.45)
0 0 2 {0,1}
1 ∞ d −1 −u
Z
1
= u 2 e du = Γ (d/2). (20.43) For x ∈ [0, 1] let µx = νx1 ⊗ · · · ⊗ νxd be the product of νx1 , . . . , νxd on Ω :=
2 0 2 d
{0, 1} . Alternatively the measure µx may be described by
Combing the the last two equations with Lemma 20.26 which states that Id (1) =
d
π d/2 , we conclude that Y 1−εi
µx ({ε}) = (1 − xi ) xεi i (20.46)
1 i=1
π d/2 = Id (1) = σ(S d−1 )Γ (d/2)
2 for ε ∈ Ω. Notice that µx ({ε}) is a degree d polynomial in x for each ε ∈ Ω.
which proves Eq. (20.41). Example 19.24 implies Γ (1) = 1 and from Eq. (20.43), For n ∈ N and x ∈ [0, 1], let µnx denote the n – fold product of µx with itself
Z ∞ Z ∞ on Ω n , Xi (ω) = ωi ∈ Ω ⊂ Rd for ω ∈ Ω n and let
−r 2 2
Γ (1/2) = 2 e dr = e−r dr Sn = (Sn1 , . . . , Snd ) := (X1 + X2 + · · · + Xn )/n,
0 −∞
√
= I1 (1) = π. so Sn : Ω n → Rd . The reader is asked to verify (Exercise 20.4) that
The relation, Γ (x + 1) = xΓ (x) is the consequence of the following integration Z Z Z
by parts argument: Sn dµnx := Sn1 dµnx , . . . , Snd dµnx = (x1 , . . . , xd ) = x (20.47)
Ωn Ωn Ωn
Z ∞ Z ∞
−u x+1 du x d −u and
Γ (x + 1) = e u = u − e du d
0 u 0 du Z
2 1X d
Z ∞ |Sn − x| dµnx = xi (1 − xi ) ≤ . (20.48)
=x ux−1 e−u du = x Γ (x). Ωn n i=1 n
0
From these equations it follows that Sn is concentrating near x as n → ∞, a
manifestation of the law of large numbers. Therefore it is reasonable to expect
BRUCE: add Morrey’s Inequality ?? here. Z
pn (x) := f (Sn )dµnx (20.49)
Ωn

Z
should approach f (x) as n → ∞. Let ε > 0 be given, M = Qn (x) dx = 1 and (20.51)
sup {|f (x)| : x ∈ [0, 1]} and ZRd
δε = sup {|f (y) − f (x)| : x, y ∈ [0, 1] and |y − x| ≤ ε} . lim Qn (x)dx = 0 for all ε > 0. (20.52)
n→∞
|x|≥ε
By uniform continuity of f on [0, 1], limε↓0 δε = 0. Using these definitions and
the fact that µnx (Ω n ) = 1, For f ∈ BC(Rd ), Qn ∗ f converges to f uniformly on compact subsets of Rd .
Z Z Proof. The proof is exactly the same as the proof of Lemma 10.29, it is
n
|f (x) − f (Sn )| dµnx only necessary to replace R by Rd everywhere in the proof.

|f (x) − pn (x)| =
(f (x) − f (Sn )) dµx ≤
ZΩ
n Ω n Define
Qn : Rn → [0, ∞) by Qn (x) = qn (x1 ) . . . qn (xd ).
Z
n (20.53)
≤ |f (x) − f (Sn )| dµx + |f (x) − f (Sn )| dµnx
{|Sn −x|>ε} {|Sn −x|≤ε} where qn is defined in Eq. (10.23).
≤ 2M µnx (|Sn − x| > ε) + δε . (20.50) ∞
Lemma 20.31. The sequence {Qn }n=1 is an approximate δ – sequence, i.e.
they satisfy Eqs. (20.51) and (20.52).
By Chebyshev’s inequality,
Z Proof. The fact that Qn integrates to one is an easy consequence of Tonelli’s
1 d theorem and the fact that qn integrates to one. Since all norms on Rd are
µnx (|Sn − x| > ε) ≤ 2 (Sn − x) 2
dµnx = 2,
ε Ωn nε equivalent, we may assume that |x| = max {|xi | : i = 1, 2, . . . , d} when proving
Eq. (20.52). With this norm
and therefore, Eq. (20.50) yields the estimate
x ∈ Rd : |x| ≥ ε = ∪di=1 x ∈ Rd : |xi | ≥ ε

2dM
kf − pn k∞ ≤ + δε and therefore by Tonelli’s theorem,
nε2
Z d Z Z
and hence
X
Qn (x)dx ≤ Qn (x)dx = d qn (t)dt
lim sup kf − pn k∞ ≤ δε → 0 as ε ↓ 0. i=1
n→∞ {|x|≥ε} {|xi |≥ε} {x∈R|x|≥ε}
This completes the proof since, using Eq. (20.46), which tends to zero as n → ∞ by Lemma 10.30.
Proof. Proof of Theorem 10.35. Again we assume f ∈ C Rd , C and f ≡ 0
n
d
on Qcd where Qd := (0, 1) . Let Qn (x) be defined as in Eq. (20.53). Then by
X X Y
pn (x) = f (Sn (ω))µnx ({ω}) = f (Sn (ω)) µx ({ωi }),
ω∈Ω n ω∈Ω n i=1 Lemma 20.31 and 20.30, pn (x) := (Qn ∗ F )(x) → F (x) uniformly for x ∈ [0, 1]
as n → ∞. So to finish the proof it only remains to show pn (x) is a polynomial
is an nd – degree polynomial in x ∈ Rd ). when x ∈ [0, 1]. For x ∈ [0, 1],
Z
Exercise 20.4. Verify Eqs. (20.47) and (20.48). This is most easily done using pn (x) = Qn (x − y)f (y)dy
Eqs. (20.44) and (20.45) and Fubini’s theorem repeatedly. (Of course Fubini’s Rd
theorem here is over kill since these are only finite sums after all. Nevertheless Z d
1 Y −1
cn (1 − (xi − yi )2 )n 1|xi −yi |≤1 dy

it is convenient to use this formulation.) = f (y)
cn [0,1] i=1
The second proof requires the next two lemmas. 1
Z d
Y −1
cn (1 − (xi − yi )2 )n dy.

∞
= f (y)
Lemma 20.30 (Approximate δ – sequences). Suppose that {Qn }n=1 is a cn [0,1] i=1
sequence of positive functions on Rd such that
Since the product in the above integrand is a polynomial if (x, y) ∈ Rd × Rd , it
follows easily that pn (x) is polynomial in x.

20.5 More Spherical Coordinates 209
20.5 More Spherical Coordinates and more generally,
In this section we will define spherical coordinates in all dimensions. Along x1 = r sin φn−2 . . . sin φ2 sin φ1 cos θ
the way we will develop an explicit method for computing surface integrals on x2 = r sin φn−2 . . . sin φ2 sin φ1 sin θ
spheres. As usual when n = 2 define spherical coordinates (r, θ) ∈ (0, ∞) ×
x3 = r sin φn−2 . . . sin φ2 cos φ1
[0, 2π) so that
x1 r cos θ
..
= = T2 (θ, r). .
x2 r sin θ
xn−2 = r sin φn−2 sin φn−3 cos φn−4
For n = 3 we let x3 = r cos φ1 and then
xn−1 = r sin φn−2 cos φn−3
x1
= T2 (θ, r sin φ1 ), xn = r cos φn−2 . (20.54)
x2
as can be seen from Figure 20.6, so that By the change of variables formula,
Z
f (x)dm(x)
Rn
Z ∞ Z
= dr dφ1 . . . dφn−2 dθ∆n (θ, φ1 , . . . , φn−2 , r)f (Tn (θ, φ1 , . . . , φn−2 , r))
0 0≤φi ≤π,0≤θ≤2π
(20.55)
where
∆n (θ, φ1 , . . . , φn−2 , r) := |det Tn0 (θ, φ1 , . . . , φn−2 , r)| .
Proposition 20.32. The Jacobian, ∆n is given by
Fig. 20.6. Setting up polar coordinates in two and three dimensions.
∆n (θ, φ1 , . . . , φn−2 , r) = rn−1 sinn−2 φn−2 . . . sin2 φ2 sin φ1 . (20.56)
If f is a function on rS n−1 – the sphere of radius r centered at 0 inside of Rn ,

    then
x1 r sin φ1 cos θ
 x2  = T2 (θ, r sin φ1 ) =  r sin φ1 sin θ  =: T3 (θ, φ1 , r, ).
Z Z
n−1
r cos φ1 f (x)dσ(x) = r f (rω)dσ(ω)
x3 r cos φ1 rS n−1 S n−1
Z
We continue to work inductively this way to define = f (Tn (θ, φ1 , . . . , φn−2 , r))∆n (θ, φ1 , . . . , φn−2 , r)dφ1 . . . dφn−2 dθ
  0≤φi ≤π,0≤θ≤2π
x1
 ..  (20.57)
 .  Tn (θ, φ1 , . . . , φn−2 , r sin φn−1 , )
= = Tn+1 (θ, φ1 , . . . , φn−2 , φn−1 , r). Proof. We are going to compute ∆n inductively. Letting ρ := r sin φn−1
r cos φn−1

 xn 
xn+1 and writing ∂T ∂Tn
∂ξ for ∂ξ (θ, φ1 , . . . , φn−2 , ρ) we have
n
So for example, ∆n+1 (θ,φ1 , . . . , φn−2 , φn−1 , r)

∂Tn ∂Tn
∂Tn ∂Tn ∂Tn

x1 = r sin φ2 sin φ1 cos θ . . . ∂φ ∂ρ r cos φn−1 ∂ρ sin φn−1

= ∂θ ∂φ1 n−2
x2 = r sin φ2 sin φ1 sin θ 0 0 ... 0 −r sin φn−1 cos φn−1

= r cos2 φn−1 + sin2 φn−1 ∆n (, θ, φ1 , . . . , φn−2 , ρ)

x3 = r sin φ2 cos φ1
x4 = r cos φ2 = r∆n (θ, φ1 , . . . , φn−2 , r sin φn−1 ),

i.e. Indeed,
2k + 2 2k + 2 (2k)!! [2(k + 1)]!!
∆n+1 (θ, φ1 , . . . , φn−2 , φn−1 , r) = r∆n (θ, φ1 , . . . , φn−2 , r sin φn−1 ). (20.58) γ2(k+1)+1 = γ2k+1 = 2 =2
2k + 3 2k + 3 (2k + 1)!! (2(k + 1) + 1)!!
To arrive at this result we have expanded the determinant along the bottom and
row. Staring with ∆2 (θ, r) = r already derived in Example 20.23, Eq. (20.58) 2k + 1 2k + 1 (2k − 1)!! (2k + 1)!!
γ2(k+1) = γ2k = π =π .
implies, 2k + 1 2k + 2 (2k)!! (2k + 2)!!
The recursion relation in Eq. (20.59) may be written as
∆3 (θ, φ1 , r) = r∆2 (θ, r sin φ1 ) = r2 sin φ1
σ(S n ) = σ S n−1 γn−1

(20.60)
∆4 (θ, φ1 , φ2 , r) = r∆3 (θ, φ1 , r sin φ2 ) = r3 sin2 φ2 sin φ1
.. which combined with σ S 1 = 2π implies
.
σ S 1 = 2π,

∆n (θ, φ1 , . . . , φn−2 , r) = rn−1 sinn−2 φn−2 . . . sin2 φ2 sin φ1
σ(S 2 ) = 2π · γ1 = 2π · 2,
which proves Eq. (20.56). Equation (20.57) now follows from Eqs. (20.34), 1 22 π 2
(20.55) and (20.56). σ(S 3 ) = 2π · 2 · γ2 = 2π · 2 · π = ,
2 2!!
As a simple application, Eq. (20.57) implies 22 π 2 22 π 2 2 23 π 2
σ(S 4 ) = · γ3 = ·2 =
Z 2!! 2!! 3 3!!
σ(S n−1 ) = sinn−2 φn−2 . . . sin2 φ2 sin φ1 dφ1 . . . dφn−2 dθ 1 2 31 23 π 3
5
0≤φi ≤π,0≤θ≤2π σ(S ) = 2π · 2 · π · 2 · π= ,
n−2
2 3 42 4!!
1 2 31 42 24 π 3
Y
= 2π γk = σ(S n−2 )γn−2 (20.59) σ(S 6 ) = 2π · 2 · π · 2 · π· 2=
k=1 2 3 42 53 5!!
Rπ and more generally that
where γk := 0
sink φdφ. If k ≥ 1, we have by integration by parts that, n n+1
2 (2π) (2π)
π π π σ(S 2n ) = and σ(S 2n+1 ) = (20.61)
(2n − 1)!!
Z Z Z
(2n)!!
γk = sink φdφ = − sink−1 φ d cos φ = 2δk,1 + (k − 1) sink−2 φ cos2 φdφ
0
Z π0 0 which is verified inductively using Eq. (20.60). Indeed,
k−2 2 n n+1

= 2δk,1 + (k − 1) sin φ 1 − sin φ dφ = 2δk,1 + (k − 1) [γk−2 − γk ] 2 (2π) (2n − 1)!! (2π)
0 σ(S 2n+1 ) = σ(S 2n )γ2n = π =
(2n − 1)!! (2n)!! (2n)!!
and hence γk satisfies γ0 = π, γ1 = 2 and the recursion relation and
n+1 n+1
k−1 (2π) (2n)!! 2 (2π)
γk = γk−2 for k ≥ 2. σ(S (n+1) ) = σ(S 2n+2 ) = σ(S 2n+1 )γ2n+1 = 2 = .
k (2n)!! (2n + 1)!! (2n + 1)!!
Hence we may conclude Using
(2n)!! = 2n (2(n − 1)) . . . (2 · 1) = 2n n!
1 2 31 42 531 n+1
γ0 = π, γ1 = 2, γ2 = π, γ3 = 2, γ4 = π, γ5 = 2, γ6 = π we may write σ(S 2n+1 ) = 2πn! which shows that Eqs. (20.34) and (20.61 are
2 3 42 53 642
in agreement. We may also write the formula in Eq. (20.61) as
and more generally by induction that  n/2
 2(2π)
n (n−1)!! for n even
(2k − 1)!! (2k)!! σ(S ) = n+1
γ2k = π and γ2k+1 = 2 .  (2π) 2 for n odd.
(2k)!! (2k + 1)!! (n−1)!!

20.6 Sard’s Theorem 211
20.6 Sard’s Theorem m (f (Qj )) = m (f (pj ) + (f 0 (pj ) + Rj ) (Qj − pj ))

= m ((f 0 (pj ) + Rj ) (Qj − pj ))
See p. 538 of Taylor and references. Also see Milnor’s topology book. Add in the
= m (Oj (f 0 (pj ) + Rj ) (Qj − pj )) (20.62)
Brower’s Fixed point theorem here as well. Also Spivak’s calculus on manifolds.
Theorem 20.33. Let U ⊂o Rm , f ∈ C ∞ (U, Rd ) and C := where Oj ∈ SO(n) is chosen so that Oj f 0 (pj )Rn ⊂ Rm−1 × {0} . Now
0
{x ∈ U : rank(f (x)) < n} be the set of critical points of f. Then the critical Oj f 0 (pj )(Qj − pj ) is contained in Γ × {0} where Γ ⊂ Rm−1 is a cube cen-
values, f (C), is a Borel measurable subset of Rd of Lebesgue measure 0. tered at 0 ∈ Rm−1 with side length at most 2 |f 0 (pj )| /N ≤ 2M/N where
M = maxp∈K |f 0 (p)| . It now follows that Oj (f 0 (pj ) + Rj ) (Qj − pj ) is con-
Remark 20.34. This result clearly extends to manifolds. tained the set of all points within ε(N )/N of Γ × {0} and in particular
For simplicity in the proof given below it will be convenient to use the norm, Oj (f 0 (pj ) + Rj ) (Qj − pj ) ⊂ (1 + ε(N )/N ) Γ × [ε(N )/N, ε(N )/N ].
|x| := maxi |xi | . Recall that if f ∈ C 1 (U, Rd ) and p ∈ U, then
Z 1 Z 1 From this inclusion and Eq. (20.62) it follows that
f (p + x) = f (p) + f 0 (p + tx)xdt = f (p) + f 0 (p)x + [f 0 (p + tx) − f 0 (p)] xdt m−1
0 0
M
m (f (Qj )) ≤ 2 (1 + ε(N )/N ) 2ε(N )/N
N
so that if
m−1 1
Z 1 = 2m M m−1 [(1 + ε(N )/N )] ε(N )
Nm
R(p, x) := f (p + x) − f (p) − f 0 (p)x = [f 0 (p + tx) − f 0 (p)] xdt
0 and therefore,
we have 1
1
X m−1
m (f (Qj )) ≤ N n 2m M m−1 [(1 + ε(N )/N )]
Z
0 0 m (f (C ∩ K)) ≤ ε(N )
|R(p, x)| ≤ |x| |f (p + tx) − f (p)| dt = |x| ε(p, x). Nm
0 j
By uniform continuity, it follows for any compact subset K ⊂ U that n−1 1

= 2n M n−1 [(1 + ε(N )/N )] ε(N ) → 0 as N → ∞
N m−n
sup {|ε(p, x)| : p ∈ K and |x| ≤ δ} → 0 as δ ↓ 0.
since m ≥ n. This proves the easy case since we may write U as a countable
0 m
Proof. Notice that if x ∈ U \C, then f (x) : R → R is surjective, which is n union of cubes K as above.
an open condition, so that U \C is an open subset of U. This shows C is relatively Remark. The case (m < n) also follows from the case m = n as follows.
closed in U, i.e. there exists C̃ @ Rm such that C = C̃ ∩ U. Let Kn ⊂ U be When m < n, C = U and we must show m(f (U )) = 0. Letting F : U ×
compact subsets of U such that Kn ↑ U, then Kn ∩ C ↑ C and Kn ∩ C = Kn ∩ C̃ Rn−m → Rn be the map F (x, y) = f (x). Then F 0 (x, y)(v, w) = f 0 (x)v, and
is compact for each n. Therefore, f (Kn ∩ C) ↑ f (C) i.e. f (C) = ∪n f (Kn ∩ C) is hence CF := U × Rn−m . So if the assertion holds for m = n we have
a countable union of compact sets and therefore is Borel measurable. Moreover,
m(f (U )) = m(F (U × Rn−m )) = 0.
since m(f (C)) = limn→∞ m(f (Kn ∩ C)), it suffices to show m(f (K)) = 0 for
all compact subsets K ⊂ C. Case 1. (n ≤ m) Let K = [a, a + γ] be a cube Case 2. (m > n) This is the hard case and the case we will need in the co-area
contained in U and by scaling the domain we may assume γ = (1, 1, 1, . . . , 1). formula to be proved later. Here I will follow the proof in Milnor. Let
n
For N ∈ N and j ∈ SN := {0, 1, . . . , N − 1} let Kj = j/N + [a, a + γ/N ] so
that K = ∪j∈SN Kj with Kjo ∩ Kjo0 = ∅ if j 6= j 0 . Let {Qj : j = 1 . . . , M } be the Ci := {x ∈ U : ∂ α f (x) = 0 when |α| ≤ i}
collection of those {Kj : j ∈ SN } which intersect C. For each j, let pj ∈ Qj ∩ C
and for x ∈ Qj − pj we have so that C ⊃ C1 ⊃ C2 ⊃ C3 ⊃ . . . . The proof is by induction on n and goes by
the following steps:
f (pj + x) = f (pj ) + f 0 (pj )x + Rj (x)
1. m(f (C \ C1 )) = 0.
where |Rj (x)| ≤ εj (N )/N and ε(N ) := maxj εj (N ) → 0 as N → ∞. Now 2. m(f (Ci \ Ci+1 )) = 0 for all i ≥ 1.

3. m(f (Ci )) = 0 for all i sufficiently large. f (C ∩ V ) = g(C 0 ) = ∪t [{t} × ut (Ct0 )] .

Step 1. If m = 1, there is nothing to prove since C = C1 so we may assume By the induction hypothesis, mm−1 (ut (Ct0 )) = 0 for all t, and therefore by
m ≥ 2. Suppose that x ∈ C \ C1 , then f 0 (p) 6= 0 and so by reordering the Fubini’s theorem,
components of x and f (p) if necessary we may assume that ∂1 f1 (p) 6= 0 where Z
we are writing ∂f (p)/∂xi as ∂i f (p) . The map h(x) := (f1 (x), x2 , . . . , xn ) has m(f (C ∩ V )) = mm−1 (ut (Ct0 ))1V 0 6=∅ dt = 0.
t
R
differential  
∂1 f1 (p) ∂2 f1 (p) . . . ∂n f1 (p) Since C \C1 may be covered by a countable collection of open sets V as above, it
 0 1 0 0  follows that m(f (C \C1 )) = 0. Step 2. Suppose that p ∈ Ck \Ck+1 , then there is
h0 (p) = 
 
.. .. . .. .
..  an α such that |α| = k+1 such that ∂ α f (p) = 0 while ∂ β f (p) = 0 for all |β| ≤ k.
 . . 
0 0 0 1 Again by permuting coordinates we may assume that α1 6= 0 and ∂ α f1 (p) 6= 0.
Let w(x) := ∂ α−e1 f1 (x), then w(p) = 0 while ∂1 w(p) 6= 0. So again the implicit
which is not singular. So by the implicit function theorem, there exists there function theorem there exists V ∈ τp such that h(x) := (w (x) , x2 , . . . , xn ) maps
exists V ∈ τp such that h : V → h(V ) ∈ τh(p) is a diffeomorphism and in V → V 0 := h(V ) ∈ τh(p) in a diffeomorphic way and in particular ∂1 w(x) 6= 0
particular ∂f1 (x)/∂x1 6= 0 for x ∈ V and hence V ⊂ U \ C1 . Consider the map on V so that V ⊂ U \ Ck+1 . As before, let g := f ◦ h−1 and notice that
g := f ◦ h−1 : V 0 := h(V ) → Rm , which satisfies Ck0 := h(Ck ∩ V ) ⊂ {0} × Rn−1 and
(f1 (x), f2 (x), . . . , fm (x)) = f (x) = g(h(x)) = g((f1 (x), x2 , . . . , xn )) f (Ck ∩ V ) = g(Ck0 ) = ḡ (Ck0 )
which implies g(t, y) = (t, u(t, y)) for (t, y) ∈ V 0 := h(V ) ∈ τh(p) , see Figure where ḡ := g|({0}×Rn−1 )∩V 0 . Clearly Ck0 is contained in the critical points of ḡ,
20.7 below where p = x̄ and m = p. Since and therefore, by induction
0 = m(ḡ(Ck0 )) = m(f (Ck ∩ V )).
Since Ck \ Ck+1 is covered by a countable collection of such open sets, it follows
that
m(f (Ck \ Ck+1 )) = 0 for all k ≥ 1.
Step 3. Suppose that Q is a closed cube with edge length δ contained in U
and k > n/m − 1. We will show m(f (Q ∩ Ck )) = 0 and since Q is arbitrary it
will follows that m(f (Ck )) = 0 as desired. By Taylor’s theorem with (integral)
remainder, it follows for x ∈ Q ∩ Ck and h such that x + h ∈ Q that
f (x + h) = f (x) + R(x, h)
where
k+1
|R(x, h)| ≤ c khk
Fig. 20.7. Making a change of variable so as to apply induction. where c = c(Q, k). Now subdivide Q into rn cubes of edge size δ/r and let Q0 be
one of the cubes in this subdivision such that Q0 ∩ Ck 6= ∅ and let x ∈ Q0 ∩ Ck .
It then follows that f (Q0 ) is contained in a cube centered at f (x) ∈ Rm with
k+1 m m(k+1)
side length at most 2c (δ/r) and hence volume at most (2c) (δ/r) .
1 0 n
Therefore, f (Q ∩ Ck ) is contained in the union of at most r cubes of volume
g 0 (t, y) =
∂t u(t, y) ∂y u(t, y) m m(k+1)
(2c) (δ/r) and hence meach
it follows that (t, y) is a critical point of g iff y ∈ Ct0 – the set of critical points m m(k+1) n m
m (f (Q ∩ Ck )) ≤ (2c) (δ/r) r = (2c) δ m(k+1) rn−m(k+1) → 0 as r ↑ ∞
of y → u(t, y). Since h is a diffeomorphism we have C 0 := h(C ∩ V ) are the
critical points of g in V 0 and provided that n − m(k + 1) < 0, i.e. provided k > n/m − 1.

20.7 Exercises 213
20.7 Exercises Exercise 20.9. Folland Problem 2.48 on p. 69. (Counter example related to
Fubini Theorem involving counting measures.)
Exercise 20.5. Prove Theorem 20.12. Suggestion, to get started define
Z Z Exercise 20.10. Folland Problem 2.50 on p. 69 pertaining to area under a
π (A) := dµ (x1 ) . . . dµ (xn ) 1A (x1 , . . . , xn ) curve. (Note the M × BR should be M ⊗ BR̄ in this problem.)
X1 Xn
and then show Eq. (20.18) holds. Use the case of two factors as the model of Exercise 20.11. Folland Problem 2.55 on p. 77. (Explicit integrations.)
your proof. Exercise 20.12. Folland Problem 2.56 on p. 77. Let f ∈ L1 ((0, a), dm), g(x) =
R a f (t) 1
Exercise 20.6. Let (Xj , Mj , µj ) for j = 1, 2, 3 be σ – finite measure spaces. x t dt for x ∈ (0, a), show g ∈ L ((0, a), dm) and
Let F : (X1 × X2 ) × X3 → X1 × X2 × X3 be defined by Z a Z a
F ((x1 , x2 ), x3 ) = (x1 , x2 , x3 ). g(x)dx = f (t)dt.
0 0
1. Show F is ((M1 ⊗ M2 ) ⊗ M3 , M1 ⊗ M2 ⊗ M3 ) – measurable and F −1 is R∞
(M1 ⊗ M2 ⊗ M3 , (M1 ⊗ M2 ) ⊗ M3 ) – measurable. That is Exercise 20.13. Show 0 sinx x dm(x) = ∞. So sinx x ∈
/ L1 ([0, ∞), m) and
R ∞ sin x
0 x dm(x) is not defined as a Lebesgue integral.
F : ((X1 × X2 )×X3 , (M1 ⊗ M2 )⊗M3 ) → (X1 ×X2 ×X3 , M1 ⊗M2 ⊗M3 )
Exercise 20.14. Folland Problem 2.57 on p. 77.
is a “measure theoretic isomorphism.”
2. Let π := F∗ [(µ1 ⊗ µ2 ) ⊗ µ3 ] , i.e. π(A) = [(µ1 ⊗ µ2 ) ⊗ µ3 ] (F −1 (A)) for all Exercise 20.15. Folland Problem 2.58 on p. 77.
A ∈ M1 ⊗ M2 ⊗ M3 . Then π is the unique measure on M1 ⊗ M2 ⊗ M3
such that Exercise 20.16. Folland Problem 2.60 on p. 77. Properties of the Γ – function.
π(A1 × A2 × A3 ) = µ1 (A1 )µ2 (A2 )µ3 (A3 )
Exercise 20.17. Folland Problem 2.61 on p. 77. Fractional integration.
for all Ai ∈ Mi . We will write π := µ1 ⊗ µ2 ⊗ µ3 .
3. Let f : X1 × X2 × X3 → [0, ∞] be a (M1 ⊗ M2 ⊗ M3 , BR̄ ) – measurable Exercise 20.18. Folland Problem 2.62 on p. 80. Rotation invariance of surface
function. Verify the identity, measure on S n−1 .
Z Z Z Z
f dπ = dµ3 (x3 ) dµ2 (x2 ) dµ1 (x1 )f (x1 , x2 , x3 ), Exercise 20.19. Folland Problem 2.64 on p. 80. On the integrability of
a b
X1 ×X2 ×X3 X3 X2 X1 |x| |log |x|| for x near 0 and x near ∞ in Rn .
makes sense and is correct.
4. (Optional.) Also show the above identity holds for any one of the six possible Exercise 20.20. Show, using Problem 20.18 that
orderings of the iterated integrals. Z
1
ωi ωj dσ (ω) = δij σ S d−1 .

Exercise 20.7. Prove the second assertion of Theorem 20.18. That is show md S d−1 d
is the unique translation invariant measure on BRd such that md ((0, 1]d ) = 1.
Hint: show S d−1 ωi2 dσ (ω) is independent of i and therefore
R
Hint: Look at the proof of Theorem 19.10.
Exercise 20.8. (Part of Folland Problem 2.46 on p. 69.) Let X = [0, 1], M = Z
1X
d Z
B[0,1] be the Borel σ – field on X, m be Lebesgue measure on [0, 1] and ν be ωi2 dσ (ω) = ωj2 dσ (ω) .
S d−1 d j=1 S d−1
counting measure, ν(A) = #(A). Finally let D = {(x, x) ∈ X 2 : x ∈ X} be the
diagonal in X 2 . Show
Z Z Z Z
1D (x, y)dν(y) dm(x) 6= 1D (x, y)dm(x) dν(y)
X X X X
by explicitly computing both sides of this equation.

21
Lp -spaces
Let (X, M, µ) be a measure space and for 0 < p < ∞ and a measurable Proof. The cases where kf kq = 0 or ∞ or kgkp = 0 or ∞ are easy to deal
function f : X → C let with and are left to the reader. So we will now assume that 0 < kf kq , kgkp < ∞.
Z 1/p Let s = |f | /kf kp and t = |g|/kgkq then Lemma 5.5 implies
p p
kf kp := |f | dµ . (21.1) |f g| 1 |f | 1 |g|q
X ≤ + (21.5)
kf kp kgkq p kf kp q kgkq
When p = ∞, let
p−1 (p−1) p/q p/q
with equality iff |g/kgkq | = |f | /kf kp = |f | /kf kp , i.e. |g|q kf kpp =
p
kf k∞ = inf {a ≥ 0 : µ(|f | > a) = 0} (21.2) kgkqq |f | . Integrating Eq. (21.5) implies
For 0 < p ≤ ∞, let kf gk1 1 1

≤ + =1
kf kp kgkq p q
Lp (X, M, µ) = {f : X → C : f is measurable and kf kp < ∞}/ ∼
with equality iff Eq. (21.4) holds. The proof is finished since it is easily checked
p q q p
where f ∼ g iff f = g a.e. Notice that kf − gkp = 0 iff f ∼ g and if f ∼ g then that equality holds in Eq. (21.3) when |f | = c |g| of |g| = c |f | for some
kf kp = kgkp . In general we will (by abuse of notation) use f to denote both constant c.
the function f and the equivalence class containing f. The following corollary is an easy extension of Hölder’s inequality.
Remark 21.1. Suppose that kf k∞ ≤ M, then for all a > M, µ(|f | > a) = 0 and Corollary 21.3. Suppose that fi : X → C are measurable functions Pn for i =
therefore µ(|f | > M ) = limn→∞ µ(|f | > M + 1/n) = 0, i.e. |f (x)| ≤ M for µ - 1, . . . , n and p1 , . . . , pn and r are positive numbers such that i=1 p−1
i = r−1 ,
a.e. x. Conversely, if |f | ≤ M a.e. and a > M then µ(|f | > a) = 0 and hence then n n n
kf k∞ ≤ M. This leads to the identity:
Y Y X
f ≤ kf k where p−1 = r−1 .

i i pi i

i=1 r i=1 i=1
kf k∞ = inf {a ≥ 0 : |f (x)| ≤ a for µ – a.e. x} .
Proof. To prove this inequality, start with n = 2, then for any p ∈ [1, ∞],
The next theorem is a generalization Theorem 5.6 to general integrals and Z
the proof is essentially identical to the proof of Theorem 5.6. r
kf gkr =
r r
|f | |g| dµ ≤ kf r kp kg r kp∗
X
Theorem 21.2 (Hölder’s inequality). Suppose that 1 ≤ p ≤ ∞ and q := p
p −1
+ q −1 = 1. If f and g are measurable functions then where p∗ = p−1 is the conjugate exponent. Let p1 = pr and p2 = p∗ r so that
p−1 , or equivalently p
p−1 −1
1 + p2 = r
−1
as desired. Then the previous equation states that
kf gk1 ≤ kf kp · kgkq . (21.3)
kf gkr ≤ kf kp1 kgkp2
p
Assuming p ∈ (1, ∞) and kf kp · kgkq < ∞, equality holds in Eq. (21.3) iff |f |
q as desired. The general case is now proved by induction. Indeed,
and |g| are linearly dependent as elements of L1 which happens iff
n+1 n n
p
Y Y Y
|g|q kf kpp = kgkqq |f | a.e. (21.4)
fi = fi · fn+1 ≤ fi kfn+1 kpn+1

i=1 r i=1 r i=1 q
216 21 Lp -spaces
Pn
where q −1 + p−1
n+1 = r
−1
. Since i=1 p−1
i = q −1 , we may now use the induction sgn(f ) $ sgn(g) and
hypothesis to conclude n

|f |
p
|f + g|p

|g|
p
n = = a.e. (21.8)
kf + gkpp
Y
kf kp kgkp
Y
fi ≤ kfi kpi ,

i=1 q i=1
Therefore
which combined with the previous displayed equation proves the generalized Z Z
form of Holder’s inequality. k|f + g|p−1 kqq = (|f + g|p−1 )q dµ = |f + g|p dµ. (21.9)
X X
Theorem 21.4 (Minkowski’s Inequality). If 1 ≤ p ≤ ∞ and f, g ∈ Lp then
Combining Eqs. (21.7) and (21.9) implies
kf + gkp ≤ kf kp + kgkp . (21.6)
kf + gkpp ≤ kf kp kf + gkp/q
p + kgkp kf + gkp/q
p (21.10)
Moreover, assuming f and g are not identically zero, equality holds in Eq. (21.6)
iff sgn(f ) $ sgn(g) a.e. (see the notation in Definition 5.7) when p = 1 and with equality iff Eq. (21.8) holds which happens iff f = cg a.e. with c > 0.
f = cg a.e. for some c > 0 for p ∈ (1, ∞). Solving for kf + gkp in Eq. (21.10) gives Eq. (21.6).
The next theorem gives another example of using Hölder’s inequality
Proof. When p = ∞, |f | ≤ kf k∞ a.e. and |g| ≤ kgk∞ a.e. so that |f + g| ≤
Theorem 21.5. Suppose that (X, M, µ) and (Y, N , ν) be σ – finite measure
|f | + |g| ≤ kf k∞ + kgk∞ a.e. and therefore
spaces, p ∈ [1, ∞], q = p/(p − 1) and k : X × Y → C be a M ⊗ N – measurable
kf + gk∞ ≤ kf k∞ + kgk∞ . function. Assume there exist finite constants C1 and C2 such that
Z
When p < ∞, |k(x, y)| dµ(x) ≤ C1 for ν a.e. y and
p p p p p p ZX
|f + g| ≤ (2 max (|f | , |g|)) = 2p max (|f | , |g| ) ≤ 2p (|f | + |g| ) ,
|k(x, y)| dν(y) ≤ C2 for µ a.e. x.
Y
kf + gkpp ≤ 2p kf kpp + kgkpp < ∞.

If f ∈ Lp (ν), then
In case p = 1,
Z
Z Z Z
|k(x, y)f (y)| dν(y) < ∞ for µ – a.e. x,
kf + gk1 = |f + g|dµ ≤ |f | dµ + |g|dµ Y
X X X
k(x, y)f (y)dν(y) ∈ Lp (µ) and
R
x → Kf (x) :=
with equality iff |f | + |g| = |f + g| a.e. which happens iff sgn(f ) $ sgn(g) a.e. In Y
case p ∈ (1, ∞), we may assume kf + gkp , kf kp and kgkp are all positive since 1/p 1/q
otherwise the theorem is easily verified. Now kKf kLp (µ) ≤ C1 C2 kf kLp (ν) (21.11)
|f + g|p = |f + g||f + g|p−1 ≤ (|f | + |g|)|f + g|p−1 Proof. Suppose p ∈ (1, ∞) to begin with and let q = p/(p − 1), then by
Hölder’s inequality,
with equality iff sgn(f ) $ sgn(g). Integrating this equation and applying Z Z
1/q 1/p
Holder’s inequality with q = p/(p − 1) gives |k(x, y)f (y)| dν(y) = |k(x, y)| |k(x, y)| |f (y)| dν(y)
Y Y
Z Z Z Z 1/q Z 1/p
p p−1
|f + g| dµ ≤ |f | |f + g| dµ + |g| |f + g|p−1 dµ ≤ |k(x, y)| dν(y)
p
|k(x, y)| |f (y)| dν(y)
X X X
Y Y
p−1
≤ (kf kp + kgkp ) k |f + g| kq (21.7) Z 1/p
1/q p
≤ C2 |k(x, y)| |f (y)| dν(y) .
with equality iff Y

21.1 Jensen’s Inequality 217
Therefore,
Z p Z Z p

|k(·, y)f (y)| dν(y) = dµ(x) |k(x, y)f (y)| dν(y)

Y Lp (µ) X Y
Z Z
p/q p
≤ C2 dµ(x) dν(y) |k(x, y)| |f (y)|
X Y
Z Z
p/q p
= C2 dν(y) |f (y)| dµ(x) |k(x, y)|
Z Y X
p/q p p/q p
≤ C2 C1 dν(y) |f (y)| = C2 C1 kf kLp (ν) ,
Y
wherein
R we used Tonelli’s theorem in third line. From this it follows that
Y
|k(x, y)f (y)| dν(y) < ∞ for µ - a.e. x, Fig. 21.1. A convex function along with two cords corresponding to x0 = −2 and
Z x1 = 4 and x0 = −5 and x1 = −2.
x → Kf (x) := k(x, y)f (y)dν(y) ∈ Lp (µ)
Y
and that Eq. (21.11) holds. Proposition 21.8. Suppose ϕ : (a, b) → R is a convex function, then
Similarly if p = ∞,
Z Z 1. For all u, v, w, z ∈ (a, b) such that u < z, w ∈ [u, z) and v ∈ (u, z],
|k(x, y)f (y)| dν(y) ≤ kf kL∞ (ν) · |k(x, y)| dν(y) ≤ C2 kf kL∞ (ν) for µ – a.e. x. ϕ(v) − ϕ(u) ϕ(z) − ϕ(w)
Y Y ≤ . (21.12)
v−u z−w
so that kKf kL∞ (µ) ≤ C2 kf kL∞ (ν) . If p = 1, then
Z Z Z Z 2. For each c ∈ (a, b), the right and left sided derivatives ϕ0± (c) exists in R and
dµ(x) dν(y) |k(x, y)f (y)| = dν(y) |f (y)| dµ(x) |k(x, y)| if a < u < v < b, then
X Y Y X
Z ϕ0+ (u) ≤ ϕ0− (v) ≤ ϕ0+ (v). (21.13)
≤ C1 dν(y) |f (y)|
Y 3. The function ϕ is continuous and differentiable except on an at most count-
which shows kKf kL1 (µ) ≤ C1 kf kL1 (ν) . able subset of (a, b) .
4. For all t ∈ (a, b) and β ∈ [ϕ0− (t), ϕ0+ (t)], ϕ(x) ≥ ϕ(t) + β(x − t) for all
x ∈ (a, b). In particular,
21.1 Jensen’s Inequality
ϕ(x) ≥ ϕ(t) + ϕ0− (t)(x − t) for all x, t ∈ (a, b).
Definition 21.6. A function φ : (a, b) → R is convex if for all a < x0 < x1 < b
Proof. 1a) Suppose first that u < v = w < z, in which case Eq. (21.12) is
and t ∈ [0, 1] φ(xt ) ≤ tφ(x1 ) + (1 − t)φ(x0 ) where xt = tx1 + (1 − t)x0 .
equivalent to
Example 21.7. The functions exp(x) and − log(x) are convex and xp is convex iff
p ≥ 1 as follows from Corollary 21.9 below which in part states that any φ ∈ (ϕ(v) − ϕ(u)) (z − v) ≤ (ϕ(z) − ϕ(v)) (v − u)
C 2 ((a, b) , R) such that φ00 ≥ 0 is convex.
which after solving for ϕ(v) is equivalent to the following equations holding:
The following Proposition is clearly motivated by Figure 21.1.
v−u z−v
BRUCE: See the Appendix (page 500) of Revuz and Yor for facts and better ϕ(v) ≤ ϕ(z) + ϕ(u) .
proofs of facts about convex functions. z−u z−u

218 21 Lp -spaces
But this last equation states that ϕ(v) ≤ ϕ(z)t + ϕ(u) (1 − t) where t = v−u
z−u ϕ0+ (u) ≤ ϕ0− (z) for all a < u < z < b.
and v = tz + (1 − t)u and hence is valid by the definition of ϕ being convex.
1b) Now assume u = w < v < z, in which case Eq. (21.12) is equivalent to The inequality, ϕ0− (z) ≤ ϕ0+ (z), is also an easy consequence of Eq. (21.12).
3) Since ϕ(x) has both left and right finite derivatives, it follows that ϕ
(ϕ(v) − ϕ(u)) (z − u) ≤ (ϕ(z) − ϕ(u)) (v − u) is continuous. (For an alternative proof, see Rudin.) Since z → ϕ0− (z) is an
increasing function, it has at most a countable set of discontinuities. If ϕ0− is
which after solving for ϕ(v) is equivalent to continuous at u, then by Eq. (21.13),
ϕ(v) (z − u) ≤ ϕ(z) (v − u) + ϕ(u) (z − v) ϕ0+ (u) ≤ lim ϕ0− (v) = ϕ0− (u) ≤ ϕ0+ (u)
v↓u
which is equivalent to from which it follows that ϕ0− (u) = ϕ0+ (u) and ϕ is differentiable at u.
4) Given t, let β ∈ [ϕ0− (t), ϕ0+ (t)], then by Eqs. (21.14) and (21.15),
v−u z−v
ϕ(v) ≤ ϕ(z) + ϕ(u) . ϕ(t) − ϕ(u) ϕ(z) − ϕ(t)
z−u z−u ≤ ϕ0− (t) ≤ β ≤ ϕ0+ (t) ≤
t−u z−t
Again this equation is valid by the convexity of ϕ.
1c) u < w < v = z, in which case Eq. (21.12) is equivalent to for all a < u < t < z < b. Item 4. now follows.
Corollary 21.9. Suppose ϕ : (a, b) → R is differentiable then ϕ is convex iff ϕ0
(ϕ(z) − ϕ(u)) (z − w) ≤ (ϕ(z) − ϕ(w)) (z − u) is non decreasing. In particular if ϕ ∈ C 2 (a, b) then ϕ is convex iff ϕ00 ≥ 0.
and this is equivalent to the inequality, Proof. By Proposition 21.8, if ϕ is convex then ϕ0 is non-decreasing. Con-
versely if ϕ0 is increasing then by the mean value theorem,
w−u z−w
ϕ(w) ≤ ϕ(z) + ϕ(u) ϕ(x1 ) − ϕ(c)
z−u z−u = ϕ0 (ξ1 ) for some ξ1 ∈ (c, x1 )
x1 − c
which again is true by the convexity of ϕ.
1) General case. If u < w < v < z, then by 1a-1c) and
ϕ(c) − ϕ(x0 )
= ϕ0 (ξ2 ) for some ξ2 ∈ (x0 , c).
ϕ(z) − ϕ(w) ϕ(v) − ϕ(w) ϕ(v) − ϕ(u) c − x0
≥ ≥ Hence
z−w v−w v−u
ϕ(x1 ) − ϕ(c) ϕ(c) − ϕ(x0 )
≥
and if u < v < w < z x1 − c c − x0
ϕ(z) − ϕ(w) ϕ(w) − ϕ(v) ϕ(w) − ϕ(u) for all x0 < c < x1 . Solving this inequality for ϕ(c) gives
≥ ≥ .
z−w w−v w−u c − x0 x1 − c
ϕ(c) ≤ ϕ(x1 ) + ϕ(x0 )
x1 − x0 x1 − x0
We have now taken care of all possible cases.
2) On the set a < w < z < b, Eq. (21.12) shows that (ϕ(z) − ϕ(w)) / (z − w) showing ϕ is convex.
is a decreasing function in w and an increasing function in z and therefore ϕ0± (x)
Theorem 21.10 (Jensen’s Inequality). Suppose that (X, M, µ) is a prob-
exists for all x ∈ (a, b). Also from Eq. (21.12) we learn that
ability space, i.e. µ is a positive measure and µ(X) = 1. Also suppose that
ϕ(z) − ϕ(w) f ∈ L1 (µ), f : X → (a, b), and ϕ : (a, b) → R is a convex function. Then
ϕ0+ (u) ≤ for all a < u < w < z < b, (21.14)
z−w
Z Z
ϕ(v) − ϕ(u) ϕ f dµ ≤ ϕ(f )dµ
≤ ϕ0− (z) for all a < u < v < z < b, (21.15) X X
v−u 1
Rwhere if ϕ ◦ f ∈
/ L (µ), then ϕ ◦ f is integrable in the extended sense and
and letting w ↑ z in the first equation also implies that X
ϕ(f )dµ = ∞.

21.2 Modes of Convergence 219
3. fn → f in Lp iff f ∈ Lp and fn ∈ Lp for all n, and limn→∞ kfn − f kp = 0.

R
Proof. Let t = X f dµ ∈ (a, b) and let β ∈ R be such that ϕ(s) − ϕ(t) ≥
β(s−t) for all s ∈ (a, b). Then integrating the inequality, ϕ(f )−ϕ(t) ≥ β(f −t),
implies that Definition 21.13. 1. {fn } is a.e. Cauchy if there is a set E ∈ M such that
Z Z Z µ(E) = 0 and{1E c fn } is a pointwise Cauchy sequences.
0≤ ϕ(f )dµ − ϕ(t) = ϕ(f )dµ − ϕ( f dµ). 2. {fn } is Cauchy in µ – measure (or L0 – Cauchy) if limm,n→∞ µ(|fn −fm | >
X X X ε) = 0 for all ε > 0.
3. {fn } is Cauchy in Lp if limm,n→∞ kfn − fm kp = 0.
Moreover, if ϕ(f ) is not integrable, then ϕ(f ) ≥ ϕ(t)
R + β(f − t) which shows
that negative part of ϕ(f ) is integrable. Therefore, X ϕ(f )dµ = ∞ in this case. Lemma 21.14 (Chebyshev’s inequality again). Let p ∈ [1, ∞) and f ∈ Lp ,
then
1
Example 21.11. The convex functions in Example 21.7 lead to the following µ (|f | ≥ ε) ≤ p kf kpp for all ε > 0.
ε
inequalities,
In particular if {fn } ⊂ Lp is Lp – convergent (Cauchy) then {fn } is also con-
vergent (Cauchy) in measure.
Z Z
exp f dµ ≤ ef dµ, (21.16)
X X
Z Z Proof. By Chebyshev’s inequality (19.11),
log(|f |)dµ ≤ log |f | dµ Z
p p 1 p 1
X X µ (|f | ≥ ε) = µ (|f | ≥ ε ) ≤ p |f | dµ = p kf kpp
ε X ε
and for p ≥ 1, Z p Z p Z

f dµ ≤ |f | dµ ≤
p
|f | dµ. and therefore if {fn } is Lp – Cauchy, then

X X X
1
The last equation may also easily be derived using Hölder’s inequality. As a µ (|fn − fm | ≥ ε) ≤ kfn − fm kpp → 0 as m, n → ∞
εp
special case of the first equation, we get another proof of PLemma 5.5. Indeed,
n
more generally, suppose pi , si > 0 for i = 1, 2, . . . , n and i=1 p1i = 1, then showing {fn } is L0 – Cauchy. A similar argument holds for the Lp – convergent
case.
n n
Pn
ln si
Pn 1 p
ln si i
X 1 ln spi i X spi i
s1 . . . sn = e i=1 =e i=1 pi
≤ e = (21.17)
p
i=1 i i=1
pi
Pn 1the inequality followspi from Eq. (21.16) with X = {1, 2, . . . , n} , µ =

where
i=1 pi δi and f (i) := ln si . Of course Eq. (21.17) may be proved directly
using the convexity of the exponential function.
m
Here is a sequence of functions where fn → 0 a.e., fn 9 0 in L1 , fn → 0.
21.2 Modes of Convergence
As usual let (X, M, µ) be a fixed measure space, assume 1 ≤ p ≤ ∞ and let
∞
{fn }n=1 ∪ {f } be a collection of complex valued measurable functions on X.
We have the following notions of convergence and Cauchy sequences.
Definition 21.12. 1. fn → f a.e. if there is a set E ∈ M such that µ(E) = 0
and limn→∞ 1E c fn = 1E c f.
2. fn → f in µ – measure if limn→∞ µ(|fn − f | > ε) = 0 for all ε > 0. We Above is a sequence of functions where fn → 0 a.e., yet fn 9 0 in L1 . or in
µ
will abbreviate this by saying fn → f in L0 or by fn → f. measure.

220 21 Lp -spaces
µ({|gj+1 − gj | > εj }) ≤ εj .
Let Ej = {|gj+1 − gj | > εj } ,
∞
[ ∞
[
FN = Ej = {|gj+1 − gj | > εj }
m
Here is a sequence of functions where fn → 0 a.e., fn → 0 but fn 9 0 in L1 . j=N j=N
and
∞
\ ∞ [
\ ∞
E := FN = Ej = {|gj+1 − gj | > εj i.o.}.
N =1 N =1 j=N
Then µ(E) = 0 by Lemma 19.20 or the computation

∞
X ∞
X
µ(E) ≤ µ(Ej ) ≤ εj = δN → 0 as N → ∞.
j=N j=N
If x ∈/ FN , i.e. |gj+1 (x) − gj (x)| ≤ εj for all j ≥ N, then by Lemma 21.15,

f (x) = lim gj (x) exists and |f (x) − gj (x)| ≤ δj for all j ≥ N. Therefore,
j→∞
∞
c
FNc , lim gj (x) = f (x) exists for all x ∈
S
since E = / E. Moreover, {x :
N =1 j→∞
|f (x) − gj (x)| > δj } ⊂ Fj for all j ≥ N and hence
Above is a sequence of functions where fn → 0 in L1 , fn 9 0 a.e., and
m
fn → 0. µ(|f − gj | > δj ) ≤ µ(Fj ) ≤ δj → 0 as j → ∞.
∞
P µ
Lemma 21.15. Suppose an ∈ C and |an+1 − an | ≤ εn and εn < ∞. Then Therefore gj → f as j → ∞. Since
n=1
∞
lim an = a ∈ C exists and |a − an | ≤ δn :=
P
εk . {|fn − f | > ε} = {|f − gj + gj − fn | > ε}
n→∞ k=n ⊂ {|f − gj | > ε/2} ∪ {|gj − fn | > ε/2},
Proof. (This is a special case of Exercise 6.9.) Let m > n then
µ({|fn − f | > ε}) ≤ µ({|f − gj | > ε/2}) + µ(|gj − fn | > ε/2)
m−1 m−1 ∞
P
|am − an | =

(ak+1 − ak ) ≤
P
|ak+1 − ak | ≤
P
εk := δn . (21.18) and
k=n k=n k=n
µ({|fn − f | > ε}) ≤ lim sup µ(|gj − fn | > ε/2) → 0 as n → ∞.
j→∞
So |am − an | ≤ δmin(m,n) → 0 as , m, n → ∞, i.e. {an } is Cauchy. Let m → ∞
in (21.18) to find |a − an | ≤ δn . µ
If there is another function g such that fn → g as n → ∞, then arguing as
Theorem 21.16. Suppose {fn } is L0 -Cauchy. Then there exists a subsequence above
µ
gj = fnj of {fn } such that lim gj := f exists a.e. and fn → f as n → ∞. µ(|f − g| > ε) ≤ µ({|f − fn | > ε/2}) + µ(|g − fn | > ε/2) → 0 as n → ∞.
µ
Moreover if g is a measurable function such that fn → g as n → ∞, then f = g
a.e. Hence
∞ ∞
1 X 1
εn < ∞ (εn = 2−n would do) and set µ(|f − g| > 0) = µ(∪∞
P
Proof. Let εn > 0 such that n=1 {|f − g| > }) ≤ µ(|f − g| > ) = 0,
n=1 n n=1
n
∞
P
δn = εk . Choose gj = fnj such that {nj } is a subsequence of N and
k=n
i.e. f = g a.e.

21.3 Completeness of Lp – spaces 221
Corollary 21.17 (Dominated Convergence Theorem). Suppose {fn } , 21.3 Completeness of Lp – spaces
{gn } , and g are in L1 and f ∈ L0 are functions such that
Z Z Theorem 21.19. Let k·k∞ be as defined in Eq. (21.2), then
µ µ ∞
|fn | ≤ gn a.e., fn −→ f, gn −→ g, and gn → g as n → ∞. (L∞ (X, M, µ), k·k∞ ) is a Banach space. A sequence {fn }n=1 ⊂ L∞ con-
∞
verges to f ∈ L iff there exists E ∈ M such that µ(E) = 0 and fn → f
Then f R∈ L1 and 1
R limn→∞ kf − fn k1 = 0, i.e. fn → f in L . In particular uniformly on E c . Moreover, bounded simple functions are dense in L∞ .
limn→∞ fn = f.
Proof. By Minkowski’s Theorem 21.4, k·k∞ satisfies the triangle inequality.
Proof. First notice that |f | ≤ g a.e. and hence f ∈ L1 since g ∈ L1 . To see The reader may easily check the remaining conditions that ensure k·k∞ is a
that |f | ≤ g, use Theorem 21.16 to find subsequences {fnk } and {gnk } of {fn } ∞
norm. Suppose that {fn }n=1 ⊂ L∞ is a sequence such fn → f ∈ L∞ , i.e.
and {gn } respectively which are almost everywhere convergent. Then kf − fn k∞ → 0 as n → ∞. Then for all k ∈ N, there exists Nk < ∞ such that
|f | = lim |fnk | ≤ lim gnk = g a.e. µ |f − fn | > k −1 = 0 for all n ≥ Nk .

k→∞ k→∞
If (for sake of contradiction) limn→∞ kf − fn k1 6= 0 there exists ε > 0 and a Let

subsequence {fnk } of {fn } such that E = ∪∞
−1

k=1 ∪n≥Nk |f − fn | > k .
Z
|f − fnk | ≥ ε for all k. (21.19) Then µ(E) = 0 and for x ∈ E c , |f (x) − fn (x)| ≤ k −1 for all n ≥ Nk . This
shows that fn → f uniformly on E c . Conversely, if there exists E ∈ M such
Using Theorem 21.16 again, we may assume (by passing to a further subse- that µ(E) = 0 and fn → f uniformly on E c , then for any ε > 0,
quences if necessary) that fnk → fR and gnk → gR almost everywhere. Not-
µ (|f − fn | ≥ ε) = µ ({|f − fn | ≥ ε} ∩ E c ) = 0
ing, |f − fnk | ≤ g + gnk → 2g and (g + gnk ) → 2g, Ran application of the
dominated convergence Theorem 19.38 implies limk→∞ |f − fnk | = 0 which for all n sufficiently large. That is to say lim supn→∞ kf − fn k∞ ≤ ε for all
contradicts Eq. (21.19). ε > 0. The density of simple functions follows from the approximation Theorem
Lemma). If fn ≥ 0 and fn → f in measure, then 18.42. So the last item to prove is the completeness of L∞ for which we will use
RExercise 21.1 (Fatou’s
R
f ≤ lim inf n→∞ fn . Theorem 7.13.
∞ P∞
Suppose that {fn }n=1 ⊂ L∞ is a sequence such that n=1 kfn k∞ < ∞. Let
Theorem 21.18 (Egoroff ’s Theorem). Suppose µ(X) < ∞ and fn → f
Mn := kfn k∞ , En := {|fn | > Mn } , and E := ∪∞ n=1 En so that µ(E) = 0. Then
a.e. Then for all ε > 0 there exists E ∈ M such that µ(E) < ε and fn → f
µ ∞ ∞
uniformly on E c . In particular fn −→ f as n → ∞. X X
sup |fn (x)| ≤ Mn < ∞
Proof. Let fn → f a.e. Then µ({|fn − f | > k1 i.o. n}) = 0 for all k > 0, i.e. n=1 x∈E
c
n=1
   
∞ [ PN
[ 1 \ 1 which
P∞ shows thatc SN (x) = n=1 fn (x) converges uniformly to S(x) :=
lim µ  {|fn − f | > } = µ  {|fn − f | > } = 0. f (x) on E , i.e. lim kS − Sn k∞ = 0.
N →∞ k k n=1 n n→∞
n≥N N =1 n≥N
Alternatively, suppose εm,n := kfm − fn k∞ → 0 as m, n → ∞. Let
∞ Em,n = {|fn − fm | > εm,n } and E := ∪Em,n , then µ(E) = 0 and
{|fn − f | > k1 } and choose an increasing sequence {Nk }k=1 such
S
Let Ek :=
n≥Nk
that µ(Ek ) < ε2−k for all k. Setting E := ∪Ek , µ(E) < k ε2−k = ε and if sup |fm (x) − fn (x)| ≤ εm,n → 0 as m, n → ∞.
P
x∈E c
x∈/ E, then |fn − f | ≤ k1 for all n ≥ Nk and all k. That is fn → f uniformly
on E c . Therefore, f := limn→∞ fn exists on E c and the limit is uniform on E c . Letting
Exercise 21.2. Show that Egoroff’s Theorem remains valid when the assump- f = limn→∞ 1E c fn , it then follows that limn→∞ kfn − f k∞ = 0.
tion µ(X) < ∞ is replaced by the assumption that |fn | ≤ g ∈ L1 for all n. Theorem 21.20 (Completeness of Lp (µ)). For 1 ≤ p ≤ ∞, Lp (µ) equipped
Hint: make use of Theorem 21.18 applied to fn |Xk where Xk := |g| ≥ k −1 . with the Lp – norm, k·kp (see Eq. (21.1)), is a Banach space.

222 21 Lp -spaces
Proof. By Minkowski’s Theorem 21.4, k·kp satisfies the triangle inequality. The reader may easily check this final formula is correct even when q = ∞
As above the reader may easily check the remaining conditions that ensure k·kp provided we interpret 1/p − 1/∞ to be 1/p.
is a norm. So we are left to prove the completeness of Lp (µ) for 1 ≤ p < ∞, the
case p = ∞ being done in Theorem 21.19. Proposition 21.22. Suppose that 0 < p0 < p1 ≤ ∞, λ ∈ (0, 1) and pλ ∈
∞ (p0 , p1 ) be defined by
Let {fn }n=1 ⊂ Lp (µ) be a Cauchy sequence. By Chebyshev’s inequality
(Lemma 21.14), {fn } is L0 -Cauchy (i.e. Cauchy in measure) and by Theorem 1 1−λ λ
= + (21.20)
21.16 there exists a subsequence {gj } of {fn } such that gj → f a.e. By Fatou’s pλ p0 p1
Lemma, with the interpretation that λ/p1 = 0 if p1 = ∞.1 Then Lpλ ⊂ Lp0 + Lp1 , i.e.
Z Z every function f ∈ Lpλ may be written as f = g + h with g ∈ Lp0 and h ∈ Lp1 .
p
kgj − f kp = lim inf |gj − gk | dµ ≤ lim inf |gj − gk |p dµ
p
For 1 ≤ p0 < p1 ≤ ∞ and f ∈ Lp0 + Lp1 let
k→∞ k→∞
= lim inf kgj − gk kpp → 0 as j → ∞.

n o
k→∞ kf k := inf kgkp0 + khkp1 : f = g + h .
Lp
In particular, kf kp ≤ kgj − f kp + kgj kp < ∞ so the f ∈ Lp and gj −→ f . The Then (Lp0 + Lp1 , k·k) is a Banach space and the inclusion map from Lpλ to
proof is finished because, Lp0 + Lp1 is bounded; in fact kf k ≤ 2 kf kpλ for all f ∈ Lpλ .
kfn − f kp ≤ kfn − gj kp + kgj − f kp → 0 as j, n → ∞. Proof. Let M > 0, then the local singularities of f are contained in the
set E := {|f | > M } and the behavior of f at “infinity” is solely determined by
f on E c . Hence let g = f 1E and h = f 1E c so that f = g + h. By our earlier
The Lp (µ) – norm controls two types of behaviors of f, namely the “behavior discussion we expect that g ∈ Lp0 and h ∈ Lp1 and this is the case since,
at infinity” and the behavior of “local singularities.” So in particular, if f blows
up at a point x0 ∈ X, then locally near x0 it is harder for f to be in Lp (µ) p
Z
p
Z p0
f
as p increases. On the other hand a function f ∈ Lp (µ) is allowed to decay kgkp00 = |f | 0 1|f |>M = M p0 1|f |>M
M
at “infinity” slower and slower as p increases. With these insights in mind, Z pλ
f p
we should not in general expect Lp (µ) ⊂ Lq (µ) or Lq (µ) ⊂ Lp (µ). However, ≤ M p0 1|f |>M ≤ M p0 −pλ kf kpλλ < ∞

there are two notable exceptions. (1) If µ(X) < ∞, then there is no behavior at M
infinity to worry about and Lq (µ) ⊂ Lp (µ) for all q ≥ p as is shown in Corollary and
21.21 below. (2) If µ is counting measure, i.e. µ(A) = #(A), then all functions
Z p1
in Lp (µ) for any p can not blow up on a set of positive measure, so there are no
Z
p p1 p1 p1
f
khkp11 = f 1|f |≤M p = |f | 1|f |≤M = M 1|f |≤M
local singularities. In this case Lp (µ) ⊂ Lq (µ) for all q ≥ p, see Corollary 21.25

1 M
below. Z pλ
f p
≤ M p1 1|f |≤M ≤ M p1 −pλ kf kpλλ < ∞.
Corollary 21.21. If µ(X) < ∞ and 0 < p < q ≤ ∞, then Lq (µ) ⊂ Lp (µ), the M
inclusion map is bounded and in fact
Moreover this shows
kf kp ≤ [µ(X)]( p q ) kf kq .
1
−1
p /p0 p /p1
kf k ≤ M 1−pλ /p0 kf kpλλ + M 1−pλ /p1 kf kpλλ .
Proof. Take a ∈ [1, ∞] such that
Taking M = λ kf kpλ then gives
1 1 1 pq
= + , i.e. a = . 1
A little algebra shows that λ may be computed in terms of p0 , pλ and p1 by
p a q q−p
p0 p1 − pλ
Then by Corollary 21.3, λ= · .
pλ p1 − p0
1 1
kf kp = kf · 1kp ≤ kf kq · k1ka = µ(X)1/a kf kq = µ(X)( p − q ) kf kq .

21.4 Converse of Hölder’s Inequality 223

kf k ≤ λ1−pλ /p0 + λ1−pλ /p1 kf kpλ Proof. Suppose that 0 < p < q = ∞, then
p p
X p p
and then taking λ = 1 shows kf k ≤ 2 kf kpλ . The proof that (Lp0 + Lp1 , k·k) is kf k∞ = sup {|f (x)| : x ∈ X} ≤ |f (x)| = kf kp ,
a Banach space is left as Exercise 21.7 to the reader. x∈X
Corollary 21.23 (Interpolation of Lp – norms). Suppose that 0 < p0 < i.e. kf k∞ ≤ kf kp for all 0 < p < ∞. For 0 < p ≤ q ≤ ∞, apply Corollary 21.23
p1 ≤ ∞, λ ∈ (0, 1) and pλ ∈ (p0 , p1 ) be defined as in Eq. (21.20), then Lp0 ∩ with p0 = p and p1 = ∞ to find
Lp1 ⊂ Lpλ and p/q 1−p/q p/q 1−p/q
λ 1−λ
kf kpλ ≤ kf kp0 kf kp1 . (21.21) kf kq ≤ kf kp kf k∞ ≤ kf kp kf kp = kf kp .
Further assume 1 ≤ p0 < pλ < p1 ≤ ∞, and for f ∈ Lp0 ∩ Lp1 let
kf k := kf kp0 + kf kp1 .
21.3.1 Summary:
p0 p1 p0 p1
Then (L ∩ L , k·k) is a Banach space and the inclusion map of L ∩ L into p
1. Since µ(|f | > ε) ≤ ε−p kf kp , Lp – convergence implies L0 – convergence.
Lpλ is bounded, in fact
2. L0 – convergence implies almost everywhere convergence for some subse-
quence.

kf kpλ ≤ max λ−1 , (1 − λ)−1 kf kp0 + kf kp1 . (21.22)
3. If µ(X) < ∞ then almost everywhere convergence implies uniform con-
vergence off certain sets of small measure and in particular we have L0 –
The heuristic explanation of this corollary is that if f ∈ Lp0 ∩ Lp1 , then f
convergence.
has local singularities no worse than an Lp1 function and behavior at infinity
4. If µ(X) < ∞, then Lq ⊂ Lp for all p ≤ q and Lq – convergence implies Lp
no worse than an Lp0 function. Hence f ∈ Lpλ for any pλ between p0 and p1 .
– convergence.
Proof. Let λ be determined as above, a = p0 /λ and b = p1 /(1 − λ), then
5. Lp0 ∩ Lp1 ⊂ Lq ⊂ Lp0 + Lp1 for any q ∈ (p0 , p1 ).
by Corollary 21.3,
6. If p ≤ q, then `p ⊂ `q andkf kq ≤ kf kp .

λ 1−λ λ 1−λ λ 1−λ
kf kpλ = |f | |f | ≤ |f | |f | = kf kp0 kf kp1 .
pλ a b
21.4 Converse of Hölder’s Inequality
It is easily checked that k·k is a norm on Lp0 ∩ Lp1 . To show this space is
complete, suppose that {fn } ⊂ Lp0 ∩ Lp1 is a k·k – Cauchy sequence. Then
Throughout this section we assume (X, M, µ) is a σ – finite measure space,
{fn } is both Lp0 and Lp1 – Cauchy. Hence there exist f ∈ Lp0 and g ∈ Lp1 such
q ∈ [1, ∞] and p ∈ [1, ∞] are conjugate exponents, i.e. p−1 + q −1 = 1. For
that limn→∞ kf − fn kp0 = 0 and limn→∞ kg − fn kpλ = 0. By Chebyshev’s
g ∈ Lq , let φg ∈ (Lp )∗ be given by
inequality (Lemma 21.14) fn → f and fn → g in measure and therefore by
Theorem 21.16, f = g a.e. It now is clear that limn→∞ kf − fn k = 0. The Z
estimate in Eq. (21.22) is left as Exercise 21.6 to the reader. φg (f ) = gf dµ =: hg, f i. (21.23)
Remark 21.24. Combining Proposition 21.22 and Corollary 21.23 gives By Hölder’s inequality
Lp0 ∩ Lp1 ⊂ Lpλ ⊂ Lp0 + Lp1
Z
|φg (f )| ≤ |gf |dµ ≤ kgkq kf kp (21.24)
for 0 < p0 < p1 ≤ ∞, λ ∈ (0, 1) and pλ ∈ (p0 , p1 ) as in Eq. (21.20).
which implies that
Corollary 21.25. Suppose now that µ is counting measure on X. Then Lp (µ) ⊂
Lq (µ) for all 0 < p < q ≤ ∞ and kf kq ≤ kf kp . kφg k(Lp )∗ := sup{|φg (f )| : kf kp = 1} ≤ kgkq . (21.25)

224 21 Lp -spaces
Proposition 21.26 (Converse of Hölder’s Inequality). Let (X, M, µ) be and

a σ – finite measure space and 1 ≤ p ≤ ∞ as above. For all g ∈ Lq , Z Z
n o |φg (f )| = sgn(g)gdµ = |g|dµ
kgkq = kφg k(Lp )∗ := sup |φg (f )| : kf kp = 1 (21.26) Xn ∩{|g|≥M −ε} Xn ∩{|g|≥M −ε}
≥ (M − ε)µ(Xn ∩ {|g| ≥ M − ε}) = (M − ε)kf k1 .

and for any measurable function g : X → C,
Z Since ε > 0 is arbitrary, it follows from this equation that kφg k(L1 )∗ ≥ M =
kgkq = sup |g| f dµ : kf kp = 1 and f ≥ 0 . (21.27) kgk∞ .
X Now for the proof of Eq. (21.27). The key new point is that we no longer are
assuming that g ∈ Lq . Let M (g) denote the right member in Eq. (21.27) and
Proof. We begin by proving Eq. (21.26). Assume first that q < ∞ so p > 1. set gn := 1Xn ∩{|g|≤n} g. Then |gn | ↑ |g| as n → ∞ and it is clear that M (gn )
Then Z Z is increasing in n. Therefore using Lemma 4.10 and the monotone convergence

|φg (f )| = gf dµ ≤ |gf | dµ ≤ kgkq kf kp
theorem,
Z
and equality occurs in the first inequality when sgn(gf ) is constant a.e. while lim M (gn ) = sup M (gn ) = sup sup |gn | f dµ : kf kp = 1 and f ≥ 0
p
equality in the second occurs, by Theorem 21.2, when |f | = c|g|q for some n→∞ n n X
constant c > 0. So let f := sgn(g)|g|q/p which for p = ∞ is to be interpreted as
Z
q/∞ = sup sup |gn | f dµ : kf kp = 1 and f ≥ 0
f = sgn(g), i.e. |g| ≡ 1. When p = ∞, n X
Z Z
|φg (f )| = g sgn(g)dµ = kgkL1 (µ) = kgk1 kf k∞ = sup lim |gn | f dµ : kf kp = 1 and f ≥ 0
n→∞ X
X Z
which shows that kφg k(L∞ )∗ ≥ kgk1 . If p < ∞, then = sup |g| f dµ : kf kp = 1 and f ≥ 0 = M (g).
X
Z Z
p
kf kpp = |f | = |g|q = kgkqq Since gn ∈ Lq for all n and M (gn ) = kφgn k(Lp )∗ (as you should verify), it
follows from Eq. (21.26) that M (gn ) = kgn kq . When q < ∞ (by the monotone
while Z Z Z convergence theorem) and when q = ∞ (directly from the definitions) one learns
that limn→∞ kgn kq = kgkq . Combining this fact with limn→∞ M (gn ) = M (g)
φg (f ) = gf dµ = |g||g|q/p dµ = |g|q dµ = kgkqq .
just proved shows M (g) = kgkq .
Hence As an application we can derive a sweeping generalization of Minkowski’s
|φg (f )| kgkqq 1
q(1− p ) inequality. (See Reed and Simon, Vol II. Appendix IX.4 for a more thorough
= q/p
= kgkq = kgkq . discussion of complex interpolation theory.)
kf kp kgkq
This shows that ||φg k ≥ kgkq which combined with Eq. (21.25) implies Eq. Theorem 21.27 (Minkowski’s Inequality for Integrals). Let (X, M, µ)
(21.26). and (Y, N , ν) be σ – finite measure spaces and 1 ≤ p ≤ ∞. If f is a M ⊗ N
The last case to consider is p = 1 and q = ∞. Let M := kgk∞ and choose measurable function, then y → kf (·, y)kLp (µ) is measurable and
Xn ∈ M such that Xn ↑ X as n → ∞ and µ(Xn ) < ∞ for all n. For any
ε > 0, µ(|g| ≥ M − ε) > 0 and Xn ∩ {|g| ≥ M − ε} ↑ {|g| ≥ M − ε}. Therefore, 1. if f is a positive M ⊗ N measurable function, then
µ(Xn ∩ {|g| ≥ M − ε}) > 0 for n sufficiently large. Let Z

Z
f (·, y)dν(y)
≤ kf (·, y)kLp (µ) dν(y). (21.28)
f = sgn(g)1Xn ∩{|g|≥M −ε} , Y Lp (µ) Y
then 2. RIf f : X × Y → C is a M ⊗ N measurable function and

kf k1 = µ(Xn ∩ {|g| ≥ M − ε}) ∈ (0, ∞) Y
kf (·, y)kLp (µ) dν(y) < ∞ then

21.4 Converse of Hölder’s Inequality 225
Z Z Z
a) for µ – a.e. x, f (x, ·) ∈ L1 (ν),
f (·, y)dν(y)

≤ |f (·, y)| dν(y)
≤ kf (·, y)kLp (µ) dν(y).
b) the µ –a.e. defined function, x → Y f (x, y)dν(y), is in Lp (µ) and
R
Y Lp (µ) Y Lp (µ) Y
c) the bound in Eq. (21.28) holds.
Hence the function, x ∈ X → Y f (x, y)dν(y), is in Lp (µ) and the bound in Eq.
R
Proof. For p ∈ [1, ∞], let Fp (y) := kf (·, y)kLp (µ) . If p ∈ [1, ∞) (21.28) holds.
Z 1/p Here is an application of Minkowski’s inequality for integrals. In this theorem
Fp (y) = kf (·, y)kLp (µ) =
p
|f (x, y)| dµ(x) we will be using the convention that x−1/∞ := 1.
X
Theorem 21.28 (Theorem 6.20 in Folland). Suppose that k : (0, ∞) ×
is a measurable function on Y by Fubini’s theorem. To see that F∞ is measur- (0, ∞) → C is a measurable function such that k is homogenous of degree −1,
able, let Xn ∈ M such that Xn ↑ X and µ(Xn ) < ∞ for all n. Then by Exercise i.e. k(λx, λy) = λ−1 k(x, y) for all λ > 0. If, for some p ∈ [1, ∞],
21.5, Z ∞
F∞ (y) = lim lim kf (·, y)1Xn kLp (µ) Cp := |k(x, 1)| x−1/p dx < ∞
n→∞ p→∞
0
which shows that F∞ is (Y, N ) – measurable as well. This shows that integral
on the right side of Eq. (21.28) is well defined. then for f ∈ L ((0, ∞), m), k(x, ·)f (·) ∈ L1 ((0, ∞), m) for m – a.e. x. Moreover,
p
Now suppose that f ≥ 0, q = p/(p − 1)and g ∈ Lq (µ) such that g ≥ 0 and the m – a.e. defined function
kgkLq (µ) = 1. Then by Tonelli’s theorem and Hölder’s inequality, Z ∞
(Kf )(x) = k(x, y)f (y)dy (21.29)
Z Z Z Z 0
f (x, y)dν(y) g(x)dµ(x) = dν(y) dµ(x)f (x, y)g(x) p
X Y Y X is in L ((0, ∞), m) and
Z
≤ kgkLq (µ) kf (·, y)kLp (µ) dν(y) kKf kLp ((0,∞),m) ≤ Cp kf kLp ((0,∞),m) .
Y
Z
= kf (·, y)kLp (µ) dν(y). Proof. By the homogeneity of k, k(x, y) = x−1 k(1, xy ). Using this relation
Y and making the change of variables, y = zx, gives
Z ∞ Z ∞
Therefore by the converse to Hölder’s inequality (Proposition 21.26), y
|k(x, y)f (y)| dy = x−1 k(1, )f (y) dy

Z
0 x

f (·, y)dν(y)
Z0 ∞ Z ∞
= x−1 |k(1, z)f (xz)| xdz = |k(1, z)f (xz)| dz.

Y Lp (µ)
Z Z 0 0
= sup f (x, y)dν(y) g(x)dµ(x) : kgkLq (µ) = 1 and g ≥ 0 Since
X Y Z ∞ Z ∞
p dx
kf (· z)kpLp ((0,∞),m) = |f (yz)|p dy = |f (x)| ,
Z
≤ kf (·, y)kLp (µ) dν(y) 0 0 z
Y
kf (· z)kLp ((0,∞),m) = z −1/p kf kLp ((0,∞),m) .
proving Eq. (21.28) in this case.
Now let f : X × Y → C be as in item 2) of the theorem. Applying the first Using Minkowski’s inequality for integrals then shows
part of the theorem to |f | shows Z ∞

Z ∞
Z
|k(·, y)f (y)| dy
≤ |k(1, z)| kf (·z)kLp ((0,∞),m) dz
0 Lp ((0,∞),m) 0
|f (x, y)| dν(y) < ∞ for µ– a.e. x, Z ∞
Y
R R = kf kLp ((0,∞),m) |k(1, z)| z −1/p dz
1 0
i.e. f (x, ·) ∈ L (ν) for the µ –a.e. x. Since Y f (x, y)dν(y) ≤ Y |f (x, y)| dν(y)
it follows by item 1) that = Cp kf kLp ((0,∞),m) < ∞.

226 21 Lp -spaces
Z
This shows that Kf in Eq. (21.29) is well defined from m – a.e. x. The proof is

p
Mq (g) = sup φgdµ : φ ∈ L (µ) with kφkp = 1 = kgkq

finished by observing X
Z ∞
where the last equality follows by Proposition 21.26. So it remains to show that

kKf kLp ((0,∞),m) ≤
|k(·, y)f (y)| dy
p ≤ Cp kf kLp ((0,∞),m) if φg ∈ L1 for all φ ∈ Sf and Mq (g) < ∞ then g ∈ Lq (µ). For n ∈ N, let
0 L ((0,∞),m)
gn := 1Xn 1|g|≤n g. Then gn ∈ Lq (µ), in fact kgn kq ≤ nµ(Xn )1/q < ∞. So by the
for all f ∈ Lp ((0, ∞), m). previous paragraph, kgn kq = Mq (gn ) and hence
The following theorem is a strengthening of Proposition 21.26. It may be Z
skipped on the first reading. p

kgn kq = sup φ1Xn 1|g|≤n gdµ : φ ∈ L (µ) with kφkp = 1

Theorem 21.29 (Converse of Hölder’s Inequality II). Assume that X

≤ Mq (g) φ1Xn 1|g|≤n p ≤ Mq (g) · 1 = Mq (g)
(X, M, µ) is a σ – finite measure space, q, p ∈ [1, ∞] are conjugate exponents
and let Sf denote the set of simple functions φ on X such that µ (φ 6= 0) < ∞. wherein the second to last inequality we have made use of the definition of
Let g : X → C be a measurable function such that φg ∈ L1 (µ) for all φ ∈ Sf ,2 Mq (g) and the fact that φ1Xn 1|g|≤n ∈ Sf . If q ∈ (1, ∞), an application of the
and define monotone convergence theorem (or Fatou’s Lemma) along with the continuity
Z

of the norm, k·kp , implies
Mq (g) := sup φgdµ : φ ∈ Sf with kφkp = 1 .
(21.30)
X kgkq = lim kgn kq ≤ Mq (g) < ∞.
n→∞
If Mq (g) < ∞ then g ∈ Lq (µ) and Mq (g) = kgkq .
If q = ∞, then kgn k∞ ≤ Mq (g) < ∞ for all n implies |gn | ≤ Mq (g) a.e. which
Proof. Let Xn ∈ M be sets such that µ(Xn ) < ∞ and Xn ↑ X as n ↑ ∞. then implies that |g| ≤ Mq (g) a.e. since |g| = limn→∞ |gn | . That is g ∈ L∞ (µ)
Suppose that q = 1 and hence p = ∞. Choose simple functions φn on X such and kgk∞ ≤ M∞ (g).
that |φn | ≤ 1 and sgn(g) = limn→∞ φn in the pointwise sense. Then 1Xm φn ∈ Sf
and therefore Z

1X φn gdµ ≤ Mq (g) 21.5 Uniform Integrability
m
X
This section will address the question as to what extra conditions are needed
for all m, n. By assumption 1Xm g ∈ L1 (µ) and therefore by the dominated
in order that an L0 – convergent sequence is Lp – convergent.
convergence theorem we may let n → ∞ in this equation to find
Z Notation 21.30 For f ∈ L1 (µ) and E ∈ M, let
1Xm |g| dµ ≤ Mq (g) Z
X
µ(f : E) := f dµ.
E
for all m. The monotone convergence theorem then implies that
Z Z and more generally if A, B ∈ M let
|g| dµ = lim 1Xm |g| dµ ≤ Mq (g) Z
X m→∞ X µ(f : A, B) := f dµ.
A∩B
1
showing g ∈ L (µ) and kgk1 ≤ Mq (g). Since Holder’s inequality implies that
Mq (g) ≤ kgk1 , we have proved the theorem in case q = 1. For q > 1, we will Lemma 21.31. Suppose g ∈ L1 (µ), then for any ε > 0 there exist a δ > 0 such
begin by assuming that g ∈ Lq (µ). Since p ∈ [1, ∞) we know that Sf is a dense that µ(|g| : E) < ε whenever µ(E) < δ.
subspace of Lp (µ) and therefore, using φg is continuous on Lp (µ),
Proof. If the Lemma is false, there would exist ε > 0 and sets En such that
2
This is equivalent to requiring 1A g ∈ L1 (µ) for all A ∈ M such that µ(A) < ∞. µ(En ) → 0 while µ(|g| : En ) ≥ ε for all n. Since |1En g| ≤ |g| ∈ L1 and for any
δ ∈ (0, 1), µ(1En |g| > δ) ≤ µ(En ) → 0 as n → ∞, the dominated convergence

21.5 Uniform Integrability 227
theorem of Corollary 21.17 implies limn→∞ µ(|g| : En ) = 0. This contradicts µ(|fn |) = µ(|fn | : |fn | ≥ M ) + µ(|fn | : |fn | < M )
µ(|g| : En ) ≥ ε for all n and the proof is complete. ≤ sup µ(|fn | : |fn | ≥ M ) + M µ(X) < ∞.
∞
Suppose that {fn }n=1 is a sequence of measurable functions which converge n
in L1 (µ) to a function f. Then for E ∈ M and n ∈ N,
Moreover,
|µ(fn : E)| ≤ |µ(f − fn : E)| + |µ(f : E)| ≤ kf − fn k1 + |µ(f : E)| .
µ(|fn | : E) = µ(|fn | : |fn | ≥ M, E) + µ(|fn | : |fn | < M, E)
Let εN := supn>N kf − fn k1 , then εN ↓ 0 as N ↑ ∞ and ≤ sup µ(|fn | : |fn | ≥ M ) + M µ(E).
n
sup |µ(fn : E)| ≤ sup |µ(fn : E)| ∨ (εN + |µ(f : E)|) ≤ εN + µ (gN : E) ,
n n≤N So given ε > 0 choose M so large that supn µ(|fn | : |fn | ≥ M ) < ε/2 and then
(21.31) take δ = ε/ (2M ) .
PN 1
where gN = |f | + n=1 |fn | ∈ L . From Lemma 21.31 and Eq. (21.31) one
Remark 21.35. It is not in general true that if {fn } ⊂ L1 (µ) is uniformly inte-
easily concludes,
grable then supn µ(|fn |) < ∞. For example take X = {∗} and µ({∗}) = 1. Let
∀ ε > 0 ∃ δ > 0 3 sup |µ(fn : E)| < ε when µ(E) < δ. (21.32) fn (∗) = n. Since for δ < 1 a set E ⊂ X such that µ(E) < δ is in fact the empty
n set, we see that Eq. (21.33) holds in this example. However, for finite measure
∞ spaces with out “atoms”, for every δ > 0 we may find a finite partition of X by
Definition 21.32. Functions {fn }n=1 ⊂ L1 (µ) satisfying Eq. (21.32) are said k
sets {E` }`=1 with µ(E` ) < δ. Then if Eq. (21.33) holds with 2ε = 1, then
to be uniformly integrable.
k
Remark 21.33. Let {fn } be real functions satisfying Eq. (21.32), E be a set X
µ(|fn |) = µ(|fn | : E` ) ≤ k
where µ(E) < δ and En = E ∩ {fn ≥ 0} . Then µ(En ) < δ so that µ(fn+ : E) =
`=1
µ(fn : En ) < ε and similarly µ(fn− : E) < ε. Therefore if Eq. (21.32) holds then
showing that µ(|fn |) ≤ k for all n.
sup µ(|fn | : E) < 2ε when µ(E) < δ. (21.33)
n
The following Lemmas gives a concrete necessary and sufficient conditions
Similar arguments work for the complex case by looking at the real and imagi- for verifying a sequence of functions is uniformly bounded and uniformly inte-
∞ grable.
nary parts of fn . Therefore {fn }n=1 ⊂ L1 (µ) is uniformly integrable iff
∀ ε > 0 ∃ δ > 0 3 sup µ(|fn | : E) < ε when µ(E) < δ. (21.34) Lemma 21.36. Suppose that µ(X) < ∞, and Λ ⊂ L0 (X) is a collection of
n functions.
Lemma 21.34. Assume that µ(X) < ∞, then {fn } is uniformly bounded in 1. If there exists a non decreasing function φ : R+ → R+ such that
L1 (µ) (i.e. K = supn kfn k1 < ∞) and {fn } is uniformly integrable iff limx→∞ φ(x)/x = ∞ and
lim sup µ(|fn | : |fn | ≥ M ) = 0. (21.35) K := sup µ(φ(|f |)) < ∞ (21.36)
M →∞ n
f ∈Λ
Proof. Since {fn } is uniformly bounded in L1 (µ), µ(|fn | ≥ M ) ≤ K/M. So

then
if (21.34) holds and ε > 0 is given, we may choose M sufficiently large so that
lim sup µ |f | 1|f |≥M = 0. (21.37)
µ(|fn | ≥ M ) < δ(ε) for all n and therefore, M →∞ f ∈Λ
sup µ(|fn | : |fn | ≥ M ) ≤ ε. 2. Conversely if Eq. (21.37) holds, there exists a non-decreasing continuous
n function φ : R+ → R+ such that φ(0) = 0, limx→∞ φ(x)/x = ∞ and Eq.
(21.36) is valid.
Since ε is arbitrary, we concluded that Eq. (21.35) must hold. Conversely, sup-
pose that Eq. (21.35) holds, then automatically K = supn µ(|fn |) < ∞ because

228 21 Lp -spaces
x
Proof. 1. Let φ be as in item 1. above and set εM := supx≥M φ(x) → 0 as and hence
∞
M → ∞ by assumption. Then for f ∈ Λ
X
sup µ (φ(|f |)) ≤ (n + 1) εMn < ∞.
f ∈Λ
|f | n=0
µ(|f | : |f | ≥ M ) = µ( φ (|f |) : |f | ≥ M ) ≤ εM µ(φ (|f |) : |f | ≥ M )
φ (|f |)
≤ εM µ(φ (|f |)) ≤ KεM Theorem 21.37 (Vitali Convergence Theorem). (Folland 6.15) Suppose
that 1 ≤ p < ∞. A sequence {fn } ⊂ Lp is Cauchy iff
and hence
lim sup µ |f | 1|f |≥M ≤ lim KεM = 0. 1. {fn } is L0 – Cauchy,
M →∞ f ∈Λ M →∞ p
2. {|fn | } – is uniformly integrable.
2. By assumption, εM := supf ∈Λ µ |f | 1|f |≥M → 0 as M → ∞. Therefore we 3. RFor all ε > 0, there exists a set E ∈ M such that µ(E) < ∞ and
p
may choose Mn ↑ ∞ such that |f | dµ < ε for all n. (This condition is vacuous when µ(X) < ∞.)
Ec n
∞
X Proof. (=⇒) Suppose {fn } ⊂ Lp is Cauchy. Then (1) {fn } is L0 –
(n + 1) εMn < ∞ Cauchy by Lemma 21.14. (2) By completeness of Lp , there exists f ∈ Lp such
n=0
that kfn − f kp → 0 as n → ∞. By the mean value theorem,
where by convention M0 := 0. Now define φ so that φ(0) = 0 and
p
∞
||f | − |fn |p | ≤ p(max(|f | , |fn |))p−1 ||f | − |fn || ≤ p(|f | + |fn |)p−1 ||f | − |fn ||
X
φ0 (x) = (n + 1) 1(Mn ,Mn+1 ] (x), and therefore by Hölder’s inequality,
n=0
Z Z Z
p p p−1
i.e.
∞
||f | − |fn | | dµ ≤ p (|f | + |fn |) ||f | − |fn || dµ ≤ p (|f | + |fn |)p−1 |f − fn |dµ
Z x X
φ(x) = φ0 (y)dy = (n + 1) (x ∧ Mn+1 − x ∧ Mn ) .
0
≤ pkf − fn kp k(|f | + |fn |)p−1 kq = pk |f | + |fn |kp/q
p kf − fn kp
n=0
By construction φ is continuous, φ(0) = 0, φ0 (x) is increasing (so φ is convex) ≤ p(kf kp + kfn kp )p/q kf − fn kp
and φ0 (x) ≥ (n + 1) for x ≥ Mn . In particular p
where q := p/(p − 1). This shows that ||f | − |fn |p | dµ → 0 as n → ∞.3 By
R
p
φ(x) φ(Mn ) + (n + 1)x the remarks prior to Definition 21.32, {|fn | } is uniformly integrable. To verify
≥ ≥ n + 1 for x ≥ Mn (3), for M > 0 and n ∈ N let EM = {|f | ≥ M } and EM (n) = {|fn | ≥ M }.
x x
Then µ(EM ) ≤ M1p kf ||pp < ∞ and by the dominated convergence theorem,
from which we conclude limx→∞ φ(x)/x = ∞. We also have φ0 (x) ≤ (n + 1) on
[0, Mn+1 ] and therefore
Z Z
p p
|f | dµ = |f | 1|f |<M dµ → 0 as M → 0.
φ(x) ≤ (n + 1)x for x ≤ Mn+1 . c
EM
So for f ∈ Λ, Moreover,
∞
X fn 1E c ≤ f 1E c + (fn − f )1E c ≤ f 1E c + kfn − f k . (21.38)
µ (φ(|f |)) = µ φ(|f |)1(Mn ,Mn+1 ] (|f |) M p M p M p M p p
n=0
X∞ R all pn ≥ N, kf −
So given ε > 0, choose N sufficiently large such that for
fn kpp < ε. Then choose M sufficiently small such that E c |f | dµ < ε and

≤ (n + 1) µ |f | 1(Mn ,Mn+1 ] (|f |) M
n=0
3
X∞ ∞
X Here is an alternative proof. Let hn ≡ ||fn | − |f | | ≤ |fn | + |f |p =: gn ∈ L1 and
p p p
µ µ
≤ (n + 1) µ |f | 1|f |≥Mn ≤ g ≡ 2|f |p . Then gn → g, hn → 0 and gn →
R R
(n + 1) εMn R g. Therefore by the dominated
n=0 n=0 convergence theorem in Corollary 21.17, lim hn dµ = 0.
n→∞

21.5 Uniform Integrability 229
R p
c (n) |f |
EM
dµ < ε for all n = 1, 2, . . . , N − 1. Letting E := EM ∪ EM (1) ∪ · · · ∪ By property (2), δ(ε) → 0 as ε → 0. Therefore
EM (N − 1), we have
lim sup kfn − fm kp ≤ 2ε1/p + 0 + 2δ(ε) → 0 as ε ↓ 0
Z m,n→∞
µ(E) < ∞, |fn |p dµ < ε for n ≤ N − 1
Ec and therefore {fn } is Lp -Cauchy.
Here is another version of Vitali’s Convergence Theorem.
and by Eq. (21.38)
Z Theorem 21.38 (Vitali Convergence Theorem). (This is problem 9 on p.
p
|fn | dµ < (ε 1/p
+ε 1/p p p
) ≤ 2 ε for n ≥ N. 133 in Rudin.) Assume that µ(X) < ∞, {fn } is uniformly integrable, fn → f
Ec a.e. and |f | < ∞ a.e., then f ∈ L1 (µ) and fn → f in L1 (µ).
Therefore we have found E ∈ M such that µ(E) < ∞ and Proof. Let ε > 0 be given and choose δ > 0 as in the Eq. (21.33). Now use
Z Egoroff’s Theorem 21.18 to choose a set E c where {fn } converges uniformly on
sup
p
|fn | dµ ≤ 2p ε E c and µ(E) < δ. By uniform convergence on E c , there is an integer N < ∞
n Ec such that |fn − fm | ≤ 1 on E c for all m, n ≥ N. Letting m → ∞, we learn that
which verifies (3) since ε > 0 was arbitrary. (⇐=) Now suppose{fn } ⊂ Lp |fN − f | ≤ 1 on E c .
satisfies conditions (1) - (3). Let ε > 0, E be as in (3) and
Therefore |f | ≤ |fN | + 1 on E c and hence
Amn := {x ∈ E|fm (x) − fn (x)| ≥ ε}.
µ(|f |) = µ(|f | : E c ) + µ(|f | : E)
Then
≤ µ(|fN |) + µ(X) + µ(|f | : E).
k(fn − fm ) 1E c kp ≤ kfn 1E c kp + kfm 1E c kp < 2ε1/p
and Now by Fatou’s lemma,
kfn − fm kp = k(fn − fm )1E c kp + k(fn − fm )1E\Amn kp µ(|f | : E) ≤ lim inf µ(|fn | : E) ≤ 2ε < ∞
n→∞
+ k(fn − fm )1Amn kp
by Eq. (21.33). This shows that f ∈ L1 . Finally
1/p
≤ k(fn − fm )1E\Amn kp + k(fn − fm )1Amn kp + 2ε . (21.39)
µ(|f − fn |) = µ(|f − fn | : E c ) + µ(|f − fn | : E)
Using properties (1) and (3) and 1E∩{|fm −fn |<ε} |fm − fn |p ≤ εp 1E ∈ L1 , the ≤ µ(|f − fn | : E c ) + µ(|f | + |fn | : E)
dominated convergence theorem in Corollary 21.17 implies
≤ µ(|f − fn | : E c ) + 4ε
Z
k(fn − fm ) 1E\Amn kpp = 1E∩{|fm −fn |<ε} |fm − fn |p −→ 0. and so by the Dominated convergence theorem we learn that
m,n→∞
lim sup µ(|f − fn |) ≤ 4ε.

which combined with Eq. (21.39) implies n→∞
lim sup kfn − fm kp ≤ lim sup k(fn − fm )1Amn kp + 2ε1/p . Since ε > 0 was arbitrary this completes the proof.
m,n→∞ m,n→∞
Theorem 21.39 (Vitali again). Suppose that fn → f in µ measure and Eq.
Finally (21.35) holds, then fn → f in L1 .
k(fn − fm )1Amn kp ≤ kfn 1Amn kp + kfm 1Amn kp ≤ 2δ(ε)
Proof. This could of course be proved using 21.38 after passing to subse-
where quences to get {fn } to converge a.s. However I wish to give another proof. First
δ(ε) := sup sup{ kfn 1E kp : E ∈ M 3 µ(E) ≤ ε} off, by Fatou’s lemma, f ∈ L1 (µ). Now let
n

230 21 Lp -spaces
φK (x) = x1|x|≤K + K1|x|>K . Exercise 21.5. Let f ∈ Lp ∩ L∞ for some p < ∞. Show kf k∞ = limq→∞ kf kq .
If we further assume µ(X) < ∞, show kf k∞ = limq→∞ kf kq for all mea-
µ
then φK (fn ) → φK (f ) because |φK (f ) − φK (fn )| ≤ |f − fn | and since surable functions f : X → C. In particular, f ∈ L∞ iff limq→∞ kf kq < ∞.
Hints: Use Corollary 21.23 to show lim supq→∞ kf kq ≤ kf k∞ and to show
|f − fn | ≤ |f − φK (f )| + |φK (f ) − φK (fn )| + |φK (fn ) − fn | lim inf q→∞ kf kq ≥ kf k∞ , let M < kf k∞ and make use of Chebyshev’s in-
equality.
we have that
Exercise 21.6. Prove Eq. (21.22) in Corollary 21.23. (Part of Folland 6.3 on
µ|f − fn | ≤ µ |f − φK (f )| + µ|φK (f ) − φK (fn )| + µ |φK (fn ) − fn | p. 186.) Hint: Use the inequality, with a, b ≥ 1 with a−1 + b−1 = 1 chosen
= µ(|f | : |f | ≥ K) + µ|φK (f ) − φK (fn )| + µ(|fn | : |fn | ≥ K). appropriately,
sa tb
Therefore by the dominated convergence theorem st ≤ + ,
a b
(see Lemma 5.5 for Eq. (21.17)) applied to the right side of Eq. (21.21).
lim sup µ|f − fn | ≤ µ(|f | : |f | ≥ K) + lim sup µ(|fn | : |fn | ≥ K).
n→∞ n→∞ Exercise 21.7. Complete the proof of Proposition 21.22 by showing (Lp +
This last expression goes to zero as K → ∞ by uniform integrability. Lr , k·k) is a Banach space. Hint: you may find using Theorem 7.13 is helpful
here.
Exercise 21.8. Folland 6.5 on p. 186.
21.6 Exercises Exercise 21.9. By making the change of variables, u = ln x, prove the following
facts:
Definition 21.40. The essential range of f, essran(f ), consists of those λ ∈ Z 1/2
C such that µ(|f − λ| < ε) > 0 for all ε > 0. b
x−a |ln x| dx < ∞ ⇐⇒ a < 1 or a = 1 and b < −1
Definition 21.41. Let (X, τ ) be a topological space and ν be a measure on Z0 ∞
b
BX = σ(τ ). The support of ν, supp(ν), consists of those x ∈ X such that x−a |ln x| dx < ∞ ⇐⇒ a > 1 or a = 1 and b < −1
ν(V ) > 0 for all open neighborhoods, V, of x. 2
Z 1
b
Exercise 21.3. Let (X, τ ) be a second countable topological space and ν be a x−a |ln x| dx < ∞ ⇐⇒ a < 1 and b > −1
0
measure on BX – the Borel σ – algebra on X. Show Z ∞
b
x−a |ln x| dx < ∞ ⇐⇒ a > 1 and b > −1.
1. supp(ν) is a closed set. (This is actually true on all topological spaces.) 1
2. ν(X \ supp(ν)) = 0 and use this to conclude that W := X \ supp(ν) is the Suppose 0 < p0 < p1 ≤ ∞ and m is Lebesgue measure on (0, ∞) . Use the above
largest open set in X such that ν(W ) = 0. Hint: let U ⊂ τ be a countable results to manufacture a function f on (0, ∞) such that f ∈ Lp ((0, ∞) , m) iff
base for the topology τ. Show that W may be written as a union of elements (a) p ∈ (p0 , p1 ) , (b) p ∈ [p0 , p1 ] and (c) p = p0 .
from V ∈ V with the property that µ(V ) = 0.
Exercise 21.4. Prove the following facts about essran(f ).
Exercise 21.11. Folland 6.10 on p. 186. Use the strong form of Theorem 19.38.
1. Let ν = f∗ µ := µ ◦ f −1 – a Borel measure on C. Show essran(f ) = supp(ν). Exercise 21.12. Let (X, M, µ) and (Y, N , ν) be σ – finite measure spaces,
2. essran(f ) is a closed set and f (x) ∈ essran(f ) for almost every x, i.e. µ(f ∈
/ f ∈ L2 (ν) and k ∈ L2 (µ ⊗ ν). Show
essran(f )) = 0. Z
3. If F ⊂ C is a closed set such that f (x) ∈ F for almost every x then |k(x, y)f (y)| dν(y) < ∞ for µ – a.e. x.
essran(f ) ⊂ F. So essran(f ) is the smallest closed set F such that f (x) ∈ F
for almost every x.
R
Let Kf (x) := Y k(x, y)f (y)dν(y) when the integral is defined. Show Kf ∈
4. kf k∞ = sup {|λ| : λ ∈ essran(f )} . L2 (µ) and K : L2 (ν) → L2 (µ) is a bounded operator with kKkop ≤ kkkL2 (µ⊗ν) .

Exercise 21.13. Folland 6.27 on p. 196. Hint: Theorem 21.28.

22
Approximation Theorems and Convolutions
22.1 Density Theorems 1. M ⊂ Lp (µ, F) and σ (M ) = M.

2. There exists ψk ∈ M such that ψk → 1 boundedly.
In this section, (X, M, µ) will be a measure space A will be a subalgebra of M. 3. If F = C we further assume that M is closed under complex conjugation.
Notation 22.1 Suppose (X, M, µ) is a measure space and A ⊂ M is a sub- Then to every function f ∈ Lp (µ, F) , there exist φn ∈ M such that
algebra of M. Let S(A) denote those simple functions φ : X → C such that limn→∞ kf − φn kLp (µ) = 0, i.e. M is dense in Lp (µ, F) .
φ−1 ({λ}) ∈ A for all λ ∈ C and let Sf (A, µ) denote those φ ∈ S(A) such that
µ(φ 6= 0) < ∞. Proof. Fix k ∈ N for the moment and let H denote those bounded M –
∞
p
measurable functions, f : X → F, for which there exists {φn }n=1 ⊂ M such
p
P
Remark 22.2. For φ ∈ Sf (A, µ) and p ∈ [1, ∞), |φ| = z6=0 |z| 1{φ=z} and that limn→∞ kψk f − φn kLp (µ) = 0. A routine check shows H is a subspace of
hence Z `∞ (M, F) such that 1 ∈ H, M ⊂ H and H is closed under complex conjugation
p
X
|φ| dµ = |z|p µ(φ = z) < ∞ (22.1) if F = C. Moreover, H is closed under bounded convergence. To see this suppose
z6=0 fn ∈ H and fn → f boundedly. Then, by the dominated convergence theorem,
so that Sf (A, µ) ⊂ L (µ). Conversely if φ ∈ S(A) ∩ Lp (µ), then from Eq. (22.1)
p limn→∞ kψk (f − fn )kLp (µ) = 0.1 (Take the dominating function to be g =
p ∞
it follows that µ (φ = z) < ∞ for all z 6= 0 and therefore µ (φ 6= 0) < ∞. Hence [2C |ψk |] where C is a constant bounding all of the {|fn |}n=1 .) We may now
we have shown, for any 1 ≤ p < ∞, choose φn ∈ M such that kφn − ψk fn kLp (µ) ≤ n1 then
Sf (A, µ) = S(A) ∩ Lp (µ). lim sup kψk f − φn kLp (µ) ≤ lim sup kψk (f − fn )kLp (µ)
n→∞ n→∞
Lemma 22.3 (Simple Functions are Dense). The simple functions, + lim sup kψk fn − φn kLp (µ) = 0 (22.2)
Sf (M, µ), form a dense subspace of Lp (µ) for all 1 ≤ p < ∞. n→∞
∞
Proof. Let {φn }n=1 be the simple functions in the approximation Theorem which implies f ∈ H. An application of Dynkin’s Multiplicative System The-
18.42. Since |φn | ≤ |f | for all n, φn ∈ Sf (M, µ) and orem 18.51 if F = R or Theorem 18.52 if F = C now shows H contains all
p p
bounded measurable functions on X.
|f − φn |p ≤ (|f | + |φn |) ≤ 2p |f | ∈ L1 (µ) . Let f ∈ Lp (µ) be given. The dominated convergence theorem implies

limk→∞ ψk 1{|f |≤k} f − f Lp (µ) = 0. (Take the dominating function to be
Therefore, by the dominated convergence theorem, p
Z Z g = [2C |f |] where C is a bound on all of the |ψk | .) Using this and what
lim |f − φn |p dµ = lim |f − φn |p dµ = 0. we have just proved, there exists φk ∈ M such that
n→∞ n→∞
ψk 1{|f |≤k} f − φk p ≤ 1 .

L (µ) k
The goal of this section is to find a number of other dense subspaces of
Lp (µ) for p ∈ [1, ∞). The next theorem is the key result of this section. The same line of reasoning used in Eq. (22.2) now implies
limk→∞ kf − φk kLp (µ) = 0.
Theorem 22.4 (Density Theorem). Let p ∈ [1, ∞), (X, M, µ) be a measure
space and M be an algebra of bounded F – valued (F = R or F = C) measurable 1
It is at this point that the proof would break down if p = ∞.
functions such that
234 22 Approximation Theorems and Convolutions
Definition 22.5. Let (X, τ ) be a topological space and µ be a measure on BX = Proof. Consider X to be a metric space with usual metric induced from Rn .
σ (τ ) . A locally integrable function is a Borel measurable function f : X → C Then X is a locally compact separable metric space and therefore Cc (X, C) =
such that K |f | dµ < ∞ for all compact subsets K ⊂ X. We will write L1loc (µ) C(X, C) is dense in Lp (µ) for all p ∈ [1, ∞). Since, by the dominated con-
R
for the space of locally integrable functions. More generally we say f ∈ Lploc (µ) vergence theorem, uniform convergence implies Lp (µ) – convergence, it follows
iff k1K f kLp (µ) < ∞ for all compact subsets K ⊂ X. from the Weierstrass approximation theorem (see Theorem 10.35 and Corollary
10.37 or Theorem 15.31 and Corollary 15.32) that polynomials are also dense
Definition 22.6. Let (X, τ ) be a topological space. A K-finite measure on in Lp (µ).
X is Borel measure µ such that µ (K) < ∞ for all compact subsets K ⊂ X.
Lemma 22.11. Let (X, τ ) be a second countable locally compact Hausdorff
Lebesgue measure on R is an example of a K-finite measure while counting space and µ : BX → [0, ∞] be a K-finite measure on X. If h ∈ L1loc (µ) is a
measure on R is not a K-finite measure. function such that Z
Example 22.7. Suppose that µ is a K-finite measure on BRd . An application of f hdµ = 0 for all f ∈ Cc (X) (22.3)
p d X
Theorem 22.4 shows Cc (R,
d
C) is dense in L (R , BRd , µ; C). To applyd Theorem
then h(x) = 0 for µ – a.e. x. (See also Corollary 28.26 below.)
22.4, let M := Cc R , C and ψk (x) := ψ (x/k) where ψ ∈ Cc R , C with
ψ (x) = 1 in
a neighborhood of 0. The proof is completed by showing σ (M ) = Proof. Let dν(x) = |h(x)| dx, then ν is a K-finite measure on X and hence
σ Cc Rd , C = BRd , which follows directly from Lemma 18.57. Cc (X) is dense in L1 (ν) by Theorem 22.8. Notice that
We may also give a more down to earth proof as follows. Let x0 ∈ Rd , R > 0, Z Z
c 1/n
A := B (x0 , R) and fn (x) := dA (x) . Then fn ∈ M and fn → 1B(x0 ,R) as f · sgn(h)dν = f hdµ = 0 for all f ∈ Cc (X). (22.4)
n → ∞ which shows 1B(x0 ,R) is σ (M )-measurable, i.e. B (x0 , R) ∈ σ (M ) . Since X X
x0 ∈ Rd and R > 0 were arbitrary, σ (M ) = BRd . ∞
Let {Kk }k=1 be a sequence of compact sets such that Kk ↑ X as in Lemma
More generally we have the following result. 14.23. Then 1Kk sgn(h) ∈ L1 (ν) and therefore there exists fm ∈ Cc (X) such
that fm → 1Kk sgn(h) in L1 (ν). So by Eq. (22.4),
Theorem 22.8. Let (X, τ ) be a second countable locally compact Hausdorff Z Z
space and µ : BX → [0, ∞] be a K-finite measure. Then Cc (X) (the space of ν(Kk ) = 1Kk dν = lim fm sgn(h)dν = 0.
continuous functions with compact support) is dense in Lp (µ) for all p ∈ [1, ∞). X m→∞ X
(See also Proposition 28.23 below.) R
Since Kk ↑ X as k → ∞, 0 = ν(X) = X |h| dµ, i.e. h(x) = 0 for µ – a.e. x.
Proof. Let M := Cc (X) and use Item 3. of Lemma 18.57 to find functions As an application of Lemma 22.11 and Example 15.34, we will show that
ψk ∈ M such that ψk → 1 to boundedly as k → ∞. The result now follows the Laplace transform is injective.
from an application of Theorem 22.4 along with the aid of item 4. of Lemma
Theorem 22.12 (Injectivity of the Laplace Transform). For f ∈
18.57.
L1 ([0, ∞), dx), the Laplace transform of f is defined by
Exercise 22.1. Show that BC (R, C) is not dense in L∞ (R, BR , m; C). Hence Z ∞
the hypothesis that p < ∞ in Theorem 22.4 can not be removed. Lf (λ) := e−λx f (x)dx for all λ > 0.
0
Corollary 22.9. Suppose X ⊂ Rn is an open set, BX is the Borel σ – algebra If Lf (λ) := 0 then f (x) = 0 for m -a.e. x.
on X and µ be a K-finite measure on (X, BX ) . Then Cc (X) is dense in Lp (µ)
for all p ∈ [1, ∞). Proof. Suppose that f ∈ L1 ([0, ∞), dx) such that Lf (λ) ≡ 0. Let g ∈
C0 ([0, ∞), R) and ε > 0 be given. By Example 15.34 we may choose {aλ }λ>0
Corollary 22.10. Suppose that X is a compact subset of Rn and µ is a finite such that # ({λ > 0 : aλ 6= 0}) < ∞ and
measure on (X, BX ), then polynomials are dense in Lp (X, µ) for all 1 ≤ p < ∞. X
|g(x) − aλ e−λx | < ε for all x ≥ 0.
λ>0

22.1 Density Theorems 235
Then Exercise 22.2. (BRUCE: Should drop this exercise.) Suppose that (X, d) is
Z ∞ Z ∞ X
!
a metric space, µ is a measure on BX := σ(τd ) which is finite on bounded
−λx

g(x)f (x)dx =
g(x) − aλ e f (x)dx
measurable subsets of X. Show BCb (X, R), defined in Eq. (19.26), is dense in
Lp (µ) . Hints: let ψk be as defined in Eq. (19.27) which incidentally may be

0 0
λ>0

Z ∞ X

used to show σ (BCb (X, R)) = σ (BC(X, R)) . Then use the argument in the
≤ g(x) − aλ e−λx |f (x)| dx ≤ εkf k1 . proof of Corollary 18.55 to show σ (BC(X, R)) = BX .

0
λ>0

Theorem 22.14. Suppose p ∈ [1, ∞), A ⊂ M is an algebra such that σ(A) =
R∞
Since ε > 0 is arbitrary, it follows that 0 g(x)f (x)dx = 0 for all g ∈ M and µ is σ – finite on A. Then Sf (A, µ) is dense in Lp (µ). (See also Remark
C0 ([0, ∞), R). The proof is finished by an application of Lemma 22.11. 28.7 below.)
Here is another variant of Theorem 22.8. Proof. Let M := Sf (A, µ). By assumption there exits Xk ∈ A such that
Theorem 22.13. Let (X, d) be a metric space, τd be the topology on X gener- µ(Xk ) < ∞ and Xk ↑ X as k → ∞. If A ∈ A, then Xk ∩A ∈ A and µ (Xk ∩ A) <
ated by d and BX = σ(τd ) be the Borel σ – algebra. Suppose µ : BX → [0, ∞] ∞ so that 1Xk ∩A ∈ M. Therefore 1A = limk→∞ 1Xk ∩A is σ (M ) – measurable
is a measure which is σ – finite on τd and let BCf (X) denote the bounded for every A ∈ A. So we have shown that A ⊂ σ (M ) ⊂ M and therefore
continuous functions on X such that µ(f 6= 0) < ∞. Then BCf (X) is a dense M = σ (A) ⊂ σ (M ) ⊂ M, i.e. σ (M ) = M. The theorem now follows from
subspace of Lp (µ) for any p ∈ [1, ∞). Theorem 22.4 after observing ψk := 1Xk ∈ M and ψk → 1 boundedly.
Proof. Let Xk ∈ τd be open sets such that Xk ↑ X and µ(Xk ) < ∞ and let Theorem 22.15 (Separability of Lp – Spaces). Suppose, p ∈ [1, ∞), A ⊂
M is a countable algebra such that σ(A) = M and µ is σ – finite on A. Then
ψk (x) = min(1, k · dXkc (x)) = φk (dXkc (x)), Lp (µ) is separable and
see Figure 22.1 below. It is easily verified that M := BCf (X) is an algebra,
X
D={ aj 1Aj : aj ∈ Q + iQ, Aj ∈ A with µ(Aj ) < ∞}
is a countable dense subset.

Proof. It is left to reader to check D is dense in Sf (A, µ) relative to the
Lp (µ) – norm. The proof is then complete since Sf (A, µ) is a dense subspace of
Lp (µ) by Theorem 22.14.
Pn
Example 22.16. The collection of functions of the form φ = k=1 ck 1(ak ,bk ]
with ak , bk ∈ Q and ak < bk are dense in Lp (R, BR , m; C) and Lp (R, BR , m; C)
is separable for any p ∈ [1, ∞). To prove this simply apply Theorem 22.14 with
A being the algebra on R generated by the half open intervals `n(a, b] ∩ R with
Fig. 22.1. The plot of φn for n = 1, 2, and 4. Notice that φn → 1(0,∞) . a < b and a, b ∈ Q∪ {±∞} , i.e. A consists of sets of the form k=1 (ak , bk ] ∩ R,
where ak , bk ∈ Q∪ {±∞} .
Exercise 22.3. Show L∞ ([0, 1] , BR , m; C) is not separable. Hint: Suppose Γ is
ψk ∈ M for all k and ψk → 1 boundedly as k → ∞. Given V ∈ τ and k, n ∈ N,let a dense subset of L∞ ([0, 1] , BR , m; C) and for λ ∈ (0, 1) , let fλ (x) := 1[0,λ] (x) .
For each λ ∈ (0, 1) , choose gλ ∈ Γ such that kfλ − gλ k∞ < 1/2 and then show
fk,n (x) := min(1, n · d(V ∩Xk )c (x)).
the map λ ∈ (0, 1) → gλ ∈ Γ is injective. Use this to conclude that Γ must be
Then {fk,n 6= 0} = V ∩ Xk so fk,n ∈ BCf (X). Moreover uncountable.
lim lim fk,n = lim 1V ∩Xk = 1V Corollary 22.17 (Riemann Lebesgue Lemma). Suppose that
k→∞ n→∞ k→∞ f ∈ L1 (R, m), then Z
which shows V ∈ σ (M ) and hence σ (M ) = BX . The proof is now completed lim f (x)eiλx dm(x) = 0.
by an application of Theorem 22.4. λ→±∞ R

Pn Z n Z
Proof. By Example 22.16, given ε > 0 there exists φ = k=1 ck 1(ak ,bk ] with X
|1B − f | dµ = |1B − f | dµ
ak , bk ∈ R such that X
Z i=0 Ai
|f − φ|dm < ε. Xn
"Z Z #
R = µ (A0 ∩ B) + |1B − λi | dµ + |1B − λi | dµ
Notice that i=1 Ai ∩B Ai \B
n
Z n
Z X X
φ(x)eiλx dm(x) = ck 1(ak ,bk ] (x)eiλx dm(x) = µ (A0 ∩ B) + [|1 − λi | µ (B ∩ Ai ) + |λi | µ (Ai \ B)] (22.5)
R R k=1 i=1
n
n Z bk n X
X X ≥ µ (A0 ∩ B) + min {µ (B ∩ Ai ) , µ (Ai \ B)} (22.6)
= ck eiλx dm(x) = ck λ−1 eiλx |bakk
ak i=1
k=1 k=1
n
X where the last equality is a consequence of the fact that 1 ≤ |λi | + |1 − λi | . Let
= λ−1 ck eiλbk − eiλak → 0 as |λ| → ∞.

k=1 0 if µ (B ∩ Ai ) < µ (Ai \ B)
αi =
Combining these two equations with 1 if µ (B ∩ Ai ) ≥ µ (Ai \ B)
Pn
and g = αi 1Ai = 1D where
Z Z Z
i=1
f (x)eiλx dm(x) ≤ (f (x) − φ(x)) eiλx dm(x) + φ(x)eiλx dm(x)

R
Z
R
Z R
D := ∪ {Ai : i > 0 & αi = 1} ∈ A.
|f − φ|dm + φ(x)eiλx dm(x)

≤
R R Equation (22.5) with λi replaced by αi and f by g implies
Z
iλx
n
≤ ε + φ(x)e dm(x)
Z X
R |1B − 1D | dµ = µ (A0 ∩ B) + min {µ (B ∩ Ai ) , µ (Ai \ B)} .
X i=1
we learn that
R
Z Z The latter expression, by Eq. (22.6), is bounded by X |1B − f | dµ < ε and
iλx iλx

lim sup f (x)e dm(x) ≤ ε + lim sup φ(x)e dm(x) = ε.
therefore, Z
|λ|→∞ R |λ|→∞ R
µ(B4D) = |1B − 1D | dµ < ε.
X
Since ε > 0 is arbitrary, this completes the proof of the Riemann Lebesgue
lemma.
Corollary 22.18. Suppose A ⊂ M is an algebra such that σ(A) = M and µ Remark 22.19. We have to assume that µ(B) < ∞ as the following example
is σ – finite on A. Then for every B ∈ M such that µ(B) < ∞ and ε > 0 there shows. Let X = R, M = B, µ = m, A be the algebra generated by half open
exists D ∈ A such that µ(B4D) < ε. (See also Remark 28.7 below.) intervals of the form (a, b], and B = ∪∞
n=1 (2n, 2n + 1]. It is easily checked that
for every D ∈ A, that m(B∆D) = ∞.
n
Proof. By Theorem 22.14, there exists a collection,
R {Ai }i=1 , of pairwise
disjoint
Pn subsets of A and λ i ∈ R such that |1
X B
− f | dµ < ε where f =
n
λ 1
i=1 i Ai . Let A 0 := X \ ∪ A
i=1 i ∈ A then 22.2 Convolution and Young’s Inequalities
Throughout this section we will be solely concerned with d – dimensional

Lebesgue measure, m, and we will simply write Lp for Lp Rd , m .

22.2 Convolution and Young’s Inequalities 237
Definition 22.20 (Convolution). Let f, g : Rd → C be measurable functions. inverse operation to differentiation.) Since τz L = Lτz for all z ∈ Rd , (this is
We define Z another way to characterize constant coefficient differential operators) and
f ∗ g(x) = f (x − y)g(y)dy (22.7) L−1 = K we should have τz K = Kτz . Writing out this equation then says
Rd
Z
whenever the integral is defined, i.e. either f (x − ·) g (·) ∈ L1 (Rd , m) or k(x − z, y)g(y)dy = (Kg) (x − z) = τz Kg(x) = (Kτz g) (x)
Rd
f (x − ·) g (·) ≥ 0. Notice that the condition that f (x − ·) g (·) ∈ L1 (Rd , m) Z Z
is equivalent to writing |f | ∗ |g| (x) < ∞. By convention, if the integral in Eq. = k(x, y)g(y − z)dy = k(x, y + z)g(y)dy.
Rd Rd
(22.7) is not defined, let f ∗ g(x) := 0.
Since g is arbitrary we conclude that k(x − z, y) = k(x, y + z). Taking y = 0
Notation 22.21 Given a multi-index α ∈ Zd+ , let |α| = α1 + · · · + αd , then gives
d α d αj k(x, z) = k(x − z, 0) =: ρ(x − z).
Y α ∂ Y ∂
xα := xj j , and ∂xα = := . We thus find that Kg = ρ ∗ g. Hence we expect the convolution operation
j=1
∂x j=1
∂xj to appear naturally when solving constant coefficient partial differential
equations. More about this point later.
For z ∈ Rd and f : Rd → C, let τz f : Rd → C be defined by τz f (x) = f (x − z).
Proposition 22.23. Suppose p ∈ [1, ∞], f ∈ L1 and g ∈ Lp , then f ∗ g(x)
Remark 22.22 (The Significance of Convolution). exists for almost every x, f ∗ g ∈ Lp and
1. Suppose that f, g ∈ L1 (m) are positive functions and let µ be the measure kf ∗ gkp ≤ kf k1 kgkp .
2
on Rd defined by
Proof. This follows directly from Minkowski’s inequality for integrals, The-
dµ (x, y) := f (x) g (y) dm (x) dm (y) . orem 21.27.
Proposition 22.24. Suppose that p ∈ [1, ∞), then τz : Lp → Lp is an isometric
Then if h : R → [0, ∞] is a measurable function we have
isomorphism and for f ∈ Lp , z ∈ Rd → τz f ∈ Lp is continuous.
Z Z
h (x + y) dµ (x, y) = h (x + y) f (x) g (y) dm (x) dm (y) Proof. The assertion that τz : Lp → Lp is an isometric isomorphism follows
(Rd )2 (Rd )2 from translation invariance of Lebesgue measure and the fact that τ−z ◦ τz = id.
Z
For the continuity assertion, observe that
= h (x) f (x − y) g (y) dm (x) dm (y)
(Rd )2 kτz f − τy f kp = kτ−y (τz f − τy f )kp = kτz−y f − f kp
Z
= h (x) f ∗ g (x) dm (x) . from which it follows that it is enough to show τz f → f in Lp as z → 0 ∈ Rd .
Rd
When f ∈ Cc (Rd ), τz f → f uniformly and since the K := ∪|z|≤1 supp(τz f ) is
In other words, this shows the measure (f ∗ g) m is the same as S∗ µ where compact, it follows by the dominated convergence theorem that τz f → f in Lp
S (x, y) := x + y. In probability lingo, the distribution of a sum of two “in- as z → 0 ∈ Rd . For general g ∈ Lp and f ∈ Cc (Rd ),
dependent” (i.e. product measure) random variables is the the convolution
of the individual distributions. kτz g − gkp ≤ kτz g − τz f kp + kτz f − f kp + kf − gkp
2. Suppose that L = |α|≤k aα ∂ α is a constant coefficient differential operator
P
= kτz f − f kp + 2 kf − gkp
and suppose that we can solve (uniquely) the equation Lu = g in the form
Z and thus
u(x) = Kg(x) := k(x, y)g(y)dy lim sup kτz g − gkp ≤ lim sup kτz f − f kp + 2 kf − gkp = 2 kf − gkp .
Rd z→0 z→0
d p
where k(x, y) is an “integral kernel.” (This is a natural sort of assumption Because Cc (R ) is dense in L , the term kf − gkp may be made as small as we
since, in view of the fundamental theorem of calculus, integration is the please.

Exercise 22.4. Let p ∈ [1, ∞] and kτz − IkL(Lp (m)) be the operator norm τz −I. Lemma 22.27. Suppose f, g, h : Rd → C are measurable functions and assume
Show kτz − IkL(Lp (m)) = 2 for all z ∈ Rd \ {0} and conclude from this that that x is a point in Rd such that |f | ∗ |g| (x) < ∞ and |f | ∗ (|g| ∗ |h|) (x) < ∞,
z ∈ Rd → τz ∈ L (Lp (m)) is not continuous. then

Hints: 1) Show kτz − IkL(Lp (m)) = τ|z|e1 − I L(Lp (m)) . 2) Let z = te1 1. f ∗ g(x) = g ∗ f (x)
with t > 0 and look for f ∈ Lp (m) such that τz f is approximately equal to −f. 2. f ∗ (g ∗ h)(x) = (f ∗ g) ∗ h(x)
(In fact, if p = ∞, you can find f ∈ L∞ (m) such that τz f = −f.) (BRUCE: 3. If z ∈ Rd and τz (|f | ∗ |g|)(x) = |f | ∗ |g| (x − z) < ∞, then
add on a problem somewhere showing that σ (τz ) = S 1 ⊂ C. This is very simple
to prove if p = 2 by using the Fourier transform.) τz (f ∗ g)(x) = τz f ∗ g(x) = f ∗ τz g(x)
Definition 22.25. Suppose that (X, τ ) is a topological space and µ is a measure 4. If x ∈

/ suppm (f ) + suppm (g) then f ∗ g(x) = 0 and in particular,
on BX = σ(τ ). For a measurable function f : X → C we define the essential
support of f by suppm (f ∗ g) ⊂ suppm (f ) + suppm (g)
suppµ (f ) = {x ∈ X : µ({y ∈ V : f (y) 6= 0}}) > 0 ∀ neighborhoods V of x}. where in defining suppm (f ∗g) we will use the convention that “f ∗g(x) 6= 0”
(22.8) when |f | ∗ |g| (x) = ∞.
Equivalently, x ∈/ suppµ (f ) iff there exists an open neighborhood V of x such
that 1V f = 0 a.e. Proof. For item 1.,
Z Z
It is not hard to show that if supp(µ) = X (see Definition 21.41) and |f | ∗ |g| (x) = |f | (x − y) |g| (y)dy = |f | (y) |g| (y − x)dy = |g| ∗ |f | (x)
f ∈ C(X) then suppµ (f ) = supp(f ) := {f 6= 0} , see Exercise 22.7. Rd Rd
Lemma 22.26. Suppose (X, τ ) is second countable and f : X → C is a mea- where in the second equality we made use of the fact that Lebesgue measure
surable function and µ is a measure on BX . Then X := U \ suppµ (f ) may be invariant under the transformation y → x−y. Similar computations prove all of
described as the largest open set W such that f 1W (x) = 0 for µ – a.e. x. Equiv- the remaining assertions of the first three items of the lemma. Item 4. Since f ∗
alently put, C := suppµ (f ) is the smallest closed subset of X such that f = f 1C g(x) = f˜∗ g̃(x) if f = f˜ and g = g̃ a.e. we may, by replacing f by f 1suppm (f ) and
a.e. g by g1suppm (g) if necessary, assume that {f 6= 0} ⊂ suppm (f ) and {g 6= 0} ⊂
suppm (g). So if x ∈ / (suppm (f ) + suppm (g)) then x ∈ / ({f 6= 0} + {g 6= 0}) and
Proof. To verify that the two descriptions of suppµ (f ) are equivalent, sup- for all y ∈ Rd , either x − y ∈ / {f 6= 0} or y ∈/ {g 6= 0} . That is to say either
pose suppµ (f ) is defined as in Eq. (22.8) and W := X \ suppµ (f ). Then x−y ∈ {f = 0} or y ∈ {g = 0} and hence f (x−y)g(y) = 0 for all y and therefore

d
W = {x ∈ X : ∃ τ 3 V 3 x such that µ({y ∈ V : f (y) 6= 0}}) = 0} f ∗ g(x) = 0. This shows that f ∗ g = 0 on R \ suppm (f ) + suppm (g) and
= ∪ {V ⊂o X : µ (f 1V 6= 0) = 0} therefore
Rd \ suppm (f ) + suppm (g) ⊂ Rd \ suppm (f ∗ g),
= ∪ {V ⊂o X : f 1V = 0 for µ – a.e.} .
i.e. suppm (f ∗ g) ⊂ suppm (f ) + suppm (g).
So to finish the argument it suffices to show µ (f 1W 6= 0) = 0. To to this let U
be a countable base for τ and set Remark 22.28. Let A, B be closed sets of Rd , it is not necessarily true that A+B
is still closed. For example, take
Uf := {V ∈ U : f 1V = 0 a.e.}.
A = {(x, y) : x > 0 and y ≥ 1/x} and B = {(x, y) : x < 0 and y ≥ 1/ |x|} ,
Then it is easily seen that W = ∪Uf and since Uf is countable
X then every point of A + B has a positive y - component and hence is not zero.
µ (f 1W 6= 0) ≤ µ (f 1V 6= 0) = 0. On the other hand, for x > 0 we have (x, 1/x) + (−x, 1/x) = (0, 2/x) ∈ A + B
V ∈Uf
for all x and hence 0 ∈ A + B showing A+B is not closed. Nevertheless if one of
the sets A or B is compact, then A + B is closed again. Indeed, if A is compact

and xn = an + bn ∈ A + B and xn → x ∈ Rd , then by passing to a subsequence kfλ kp = λ−d/p kf k and (f ∗ g)λ = λd fλ ∗ gλ .

if necessary we may assume limn→∞ an = a ∈ A exists. In this case
Therefore if Eq. (22.10) holds for some p, q, r ∈ [1, ∞], we would also have
lim bn = lim (xn − an ) = x − a ∈ B kf ∗ gkr = λd/r k(f ∗ g)λ kr ≤ λd/r λd kfλ kp kgλ kq = λ(d+d/r−d/p−d/q) kf kp kgkq
n→∞ n→∞
exists as well, showing x = a + b ∈ A + B. for all λ > 0. This is only possible if Eq. (22.9) holds.
Proposition 22.29. Suppose that p, q ∈ [1, ∞] and p and q are conjugate ex- Proof. By the usual sorts of arguments, we may assume f and g are positive
ponents, f ∈ Lp and g ∈ Lq , then f ∗ g ∈ BC(Rd ), kf ∗ gk∞ ≤ kf kp kgkq and functions. Let α, β ∈ [0, 1] and p1 , p2 ∈ (0, ∞] satisfy p−1 −1
1 + p2 + r
−1
= 1. Then
by Hölder’s inequality, Corollary 21.3,
if p, q ∈ (1, ∞) then f ∗ g ∈ C0 (Rd ). Z h i
Proof. The existence of f ∗ g(x) and the estimate |f ∗ g| (x) ≤ kf kp kgkq f ∗ g(x) = f (x − y)(1−α) g(y)(1−β) f (x − y)α g(y)β dy
Rd
for all x ∈ Rd is a simple consequence of Holders inequality and the translation Z 1/r Z 1/p1
invariance of Lebesgue measure. In particular this shows kf ∗ gk∞ ≤ kf kp kgkq . (1−α)r (1−β)r αp1
≤ f (x − y) g(y) dy f (x − y) dy ×
By relabeling p and q if necessary we may assume that p ∈ [1, ∞). Since Rd Rd
Z 1/p2
βp2
kτz (f ∗ g) − f ∗ gku = kτz f ∗ g − f ∗ gku × g(y) dy
Rd
≤ kτz f − f kp kgkq → 0 as z → 0 Z 1/r
α β
= f (x − y)(1−α)r g(y)(1−β)r dy kf kαp1 kgkβp2 .
it follows that f ∗ g is uniformly continuous. Finally if p, q ∈ (1, ∞), we learn Rd
from Lemma 22.27 and what we have just proved that fm ∗ gm ∈ Cc (Rd ) where
Taking the rth power of this equation and integrating on x gives
fm = f 1|f |≤m and gm = g1|g|≤m . Moreover,
Z Z
r (1−α)r (1−β)r α β
kf ∗ g − fm ∗ gm k∞ ≤ kf ∗ g − fm ∗ gk∞ + kfm ∗ g − fm ∗ gm k∞ kf ∗ gkr ≤ f (x − y) g(y) dy dx · kf kαp1 kgkβp2
Rd Rd
≤ kf − fm kp kgkq + kfm kp kg − gm kq (1−α)r (1−β)r αr βr
= kf k(1−α)r kgk(1−β)r kf kαp1 kgkβp2 . (22.11)
≤ kf − fm kp kgkq + kf kp kg − gm kq → 0 as m → ∞
Let us now suppose, (1 − α)r = αp1 and (1 − β)r = βp2 , in which case Eq.
d
showing, with the aid of Proposition 15.23, f ∗ g ∈ C0 (R ). (22.11) becomes,
r r r
kf ∗ gkr ≤ kf kαp1 kgkβp2
Theorem 22.30 (Young’s Inequality). Let p, q, r ∈ [1, ∞] satisfy
which is Eq. (22.10) with
1 1 1 p := (1 − α)r = αp1 and q := (1 − β)r = βp2 . (22.12)
+ =1+ . (22.9)
p q r
So to finish the proof, it suffices to show p and q are arbitrary indices in
If f ∈ Lp and g ∈ Lq then |f | ∗ |g| (x) < ∞ for m – a.e. x and [1, ∞] satisfying p−1 + q −1 = 1 + r−1 . If α, β, p1 , p2 satisfy the relations above,
then
kf ∗ gkr ≤ kf kp kgkq . (22.10) r r
α= and β =
r + p1 r + p2
In particular L1 is closed under convolution. (The space (L1 , ∗) is an example and
of a “Banach algebra” without unit.) 1 1 1 1 1 r + p1 1 r + p2
+ = + = +
Remark 22.31. Before going to the formal proof, let us first understand Eq. p q αp1 αp2 p1 r p2 r
(22.9) by the following scaling argument. For λ > 0, let fλ (x) := f (λx), then 1 1 2 1
= + + =1+ .
after a few simple change of variables we find p1 p2 r r

Z
Conversely, if p, q, r satisfy Eq. (22.9), then let α and β satisfy p = (1 − α)r and kφt ∗ f − af kp ≤ kτtz f − f kp |φ(z)| dz → 0 as t ↓ 0.
q = (1 − β)r, i.e. Rd
r−p p r−q q Item 2. is proved similarly. Indeed, form Eq. (22.13)

α := = 1 − ≤ 1 and β = = 1 − ≤ 1.
r r r r Z
kφt ∗ f − af k∞ ≤ kτtz f − f k∞ |φ(z)| dz
Using Eq. (22.9) we may also express α and β as Rd
1 1 which again tends to zero by the dominated convergence theorem because

α = p(1 − ) ≥ 0 and β = q(1 − ) ≥ 0 limt↓0 kτtz f − f k∞ = 0 uniformly in z by the uniform continuity of f.
q p
Item 3. Let BR = B(0, R) be a large ball in Rd and K @@ U, then
and in particular we have shown α, β ∈ [0, 1]. If we now define p1 := p/α ∈
sup |φt ∗ f (x) − af (x)|
(0, ∞] and p2 := q/β ∈ (0, ∞], then x∈K
Z Z
1 1 1 1 1 1
≤ [f (x − tz) − f (x)] φ(z)dz + [f (x − tz) − f (x)] φ(z)dz

+ + =β +α +
p1 p2 r q p r BR BRc
1 1 1 Z Z
= (1 − ) + (1 − ) + ≤ |φ(z)| dz · sup |f (x − tz) − f (x)| + 2 kf k∞ |φ(z)| dz
q p r BR x∈K,z∈BR c
BR

1 1 Z
=2− 1+ + =1 ≤ kφk1 · sup |f (x − tz) − f (x)| + 2 kf k∞ |φ(z)| dz
r r
x∈K,z∈BR |z|>R
as desired. so that using the uniform continuity of f on compact subsets of U,
Theorem 22.32 (Approximate δ – functions). Let p ∈ [1, ∞], φ ∈ L1 (Rd ),
Z
a := Rd φ(x)dx, and for t > 0 let φt (x) = t−d φ(x/t). Then
R lim sup sup |φt ∗ f (x) − af (x)| ≤ 2 kf k∞ |φ(z)| dz → 0 as R → ∞.
t↓0 x∈K |z|>R
p p
1. If f ∈ L with p < ∞ then φt ∗ f → af in L as t ↓ 0.
2. If f ∈ BC(Rd ) and f is uniformly continuous then kφt ∗ f − af k∞ → 0 as See Theorem 8.15 of Folland for a statement about almost everywhere con-
t ↓ 0. vergence.
3. If f ∈ L∞ and f is continuous on U ⊂o Rd then φt ∗ f → af uniformly on
Exercise 22.5. Let
compact subsets of U as t ↓ 0.
e−1/t if t > 0
f (t) =
Proof. Making the change of variables y = tz implies 0 if t ≤ 0.
Z Z Show f ∈ C ∞ (R, [0, 1]).
φt ∗ f (x) = f (x − y)φt (y)dy = f (x − tz)φ(z)dz Lemma 22.33. There exists φ ∈ Cc∞ (Rd , [0, ∞)) such that φ(0) > 0, supp(φ) ⊂
Rd Rd R
B̄(0, 1) and Rd φ(x)dx = 1.
so that
Z R Then2 h ∈
Proof. Define h(t) = f (1−t)f (t+1) where f is as in Exercise 22.5.
φt ∗ f (x) − af (x) = [f (x − tz) − f (x)] φ(z)dz Cc∞ (R, [0, 1]), −2
supp(h) ⊂ [−1, 1] and h(0) = e > 0. Define c = Rd h(|x| )dx.
2
Rd
Z Then φ(x) = c−1 h(|x| ) is the desired function.
= [τtz f (x) − f (x)] φ(z)dz. (22.13) The reader asked to prove the following proposition in Exercise 22.9 below.
Rd
Proposition 22.34. Suppose that f ∈ L1loc (Rd , m) and φ ∈ Cc1 (Rd ), then f ∗
Hence by Minkowski’s inequality for integrals (Theorem 21.27), Proposition φ ∈ C 1 (Rd ) and ∂i (f ∗ φ) = f ∗ ∂i φ. Moreover if φ ∈ Cc∞ (Rd ) then f ∗ φ ∈
22.24 and the dominated convergence theorem, C ∞ (Rd ).

Z Z Z
Corollary 22.35 (C ∞ – Uryshon’s Lemma). Given K @@ U ⊂o Rd , there ∆k f (x)e−iξ·x dm(x) =
2k
f (x)∆kx e−iξ·x dm(x) = − |ξ| f (x)e−iξ·x dm(x)
exists f ∈ Cc∞ (Rd , [0, 1]) such that supp(f ) ⊂ U and f = 1 on K. Rd Rd Rd
2k
Proof. Let φ be as in Lemma 22.33, φt (x) = t−d φ(x/t) be as in Theorem = −(2π)d/2 |ξ| fˆ(ξ)
d
22.32, d be the standard metric on
R and ε = d(K, U c ). Since K is compact
c d for any k ∈ N. Hence
and U is closed, ε > 0. Let Vδ = x ∈ R : d(x, K) < δ and f = φε/3 ∗ 1Vε/3 ,
then
−2k
(2π)d/2 fˆ(ξ) ≤ |ξ| ∆k f → 0

supp(f ) ⊂ supp(φε/3 ) + Vε/3 ⊂ V̄2ε/3 ⊂ U. 1
Since V̄2ε/3 is closed and bounded, f ∈ Cc∞ (U ) and for x ∈ K,

as |ξ| → ∞ and fˆ ∈ C0 (Rd ). Suppose that f ∈ L1 (m) and
fk ∈
∞ d
Cc (R ) is a
Z Z ˆ ˆ
sequence such that limk→∞ kf − fk k1 = 0, then limk→∞ f − fk = 0. Hence
f (x) = 1d(y,K)<ε/3 · φε/3 (x − y)dy = φε/3 (x − y)dy = 1. ∞
Rd Rd fˆ ∈ C0 (Rd ) by an application of Proposition 15.23.
The proof will be finished after the reader (easily) verifies 0 ≤ f ≤ 1.
Here is an application of this corollary whose proof is left to the reader, Corollary 22.38. Let X ⊂ Rd be an open set and µ be a K-finite measure on
Exercise 22.10. BX .
Lemma 22.36 (Integration by Parts). Suppose f and g are measur- 1. Then Cc∞ (X) is dense in Lp (µ) for all 1 ≤ p < ∞.
able functions on Rd such that t → f (x1 , . . . , xi−1 , t, xi+1 , . . . , xd ) and t → 2. If h ∈ L1loc (µ) satisfies
g(x1 , . . . , xi−1 , t, xi+1 , . . . , xd ) are continuously differentiable functions on R for Z
∂f ∂g
each fixed x = (x1 , . . . , xd ) ∈ Rd . Moreover assume f · g, ∂x i
· g and f · ∂x i
are f hdµ = 0 for all f ∈ Cc∞ (X) (22.14)
1 d X
in L (R , m). Then
Z
∂f
Z
∂g then h(x) = 0 for µ – a.e. x.
· gdm = − f· dm.
Rd ∂xi Rd ∂xi Proof. Let f ∈ Cc (X), φ be as in Lemma 22.33, φt be as in Theorem 22.32
With this result we may give another proof of the Riemann Lebesgue and set ψt := φt ∗ (f 1X ) . Then by Proposition 22.34 ψt ∈ C ∞ (X) and by
Lemma. Lemma 22.27 there exists a compact set K ⊂ X such that supp(ψt ) ⊂ K for
all t sufficiently small. By Theorem 22.32, ψt → f uniformly on X as t ↓ 0
Lemma 22.37 (Riemann Lebesgue Lemma). For f ∈ L1 (Rd , m) let
Z 1. The dominated convergence theorem (with dominating function being
ˆ
f (ξ) := (2π)−d/2
f (x)e−iξ·x dm(x) kf k∞ 1K ), shows ψt → f in Lp (µ) as t ↓ 0. This proves Item 1., since
Rd Theorem 22.8 guarantees that Cc (X) is dense in Lp (µ).
2. Keeping the same notation as above, the dominated convergence theorem

be the Fourier transform of f. Then fˆ ∈ C0 (Rd ) and fˆ ≤ (2π)−d/2 kf k1 .

∞ (with dominating function being kf k∞ |h| 1K ) implies
(The choice of the normalization factor, (2π)−d/2 , in fˆ is for later convenience.) Z Z Z
Proof. The fact that fˆ is continuous is a simple application of the dominated 0 = lim ψt hdµ = lim ψt hdµ = f hdµ.
t↓0 X X t↓0 X
convergence theorem. Moreover,
Z The proof is now finished by an application of Lemma 22.11.
ˆ
f (ξ) ≤ |f (x)| dm(x) ≤ (2π)−d/2 kf k1
Rd
so it only remains to see that fˆ(ξ) → 0 as |ξ| → ∞. First suppose that f ∈

∂2
Pd
Cc∞ (Rd ) and let ∆ = j=1 ∂x d ∂ −iξ·x
2 be the Laplacian on R . Notice that ∂x e
j
=
j
2
−iξj e−iξ·x and ∆e−iξ·x = − |ξ| e−iξ·x . Using Lemma 22.36 repeatedly,

22.2.1 Smooth Partitions of Unity differentiable. Also assume f · g, ∂x f · g and f · ∂x g are integrable relative to
Lebesgue measure on R × Rd−1 , where ∂x f (x, y) := dt d
f (x + t, y)|t=0 . Show
We have the following smooth variants of Proposition 15.16, Theorem 15.18 Z Z
and Corollary 15.20. The proofs of these results are the same as their continu- ∂x f (x, y) · g(x, y)dxdy = − f (x, y) · ∂x g(x, y)dxdy. (22.15)
ous counterparts. One simply uses the smooth version of Urysohn’s Lemma of R×Rd−1 R×Rd−1
Corollary 22.35 in place of Lemma 15.8. (Note: this result and Fubini’s theorem proves Lemma 22.36.)
Hints: Let ψ ∈ Cc∞ (R) be a function which is 1 in a neighborhood of
Proposition 22.39 (Smooth Partitions of Unity for Compacts). Sup-
n 0 ∈ R and set ψε (x) = ψ(εx). First verify Eq. (22.15) with f (x, y) replaced by
pose that X is an open subset of Rd , K ⊂ X is a compact set and U = {Uj }j=1
ψε (x)f (x, y) by doing the x – integral first. Then use the dominated convergence
is an open cover of K. Then there exists a smooth (i.e. hj ∈ C ∞ (X, [0, 1])) theorem to prove Eq. (22.15) by passing to the limit, ε ↓ 0.
n
partition of unity {hj }j=1 of K such that hj ≺ Uj for all j = 1, 2, . . . , n.
Exercise 22.11. Let µ be a finite measure on BRd , then D := span{eiλ·x : λ ∈
Theorem 22.40 (Locally Compact Partitions of Unity). Suppose that X Rd } is a dense subspace of Lp (µ) for all 1 ≤ p < ∞. Hints: By Theorem 22.8,
is an open subset of Rd and U is an open cover of X. Then there exists a smooth Cc (Rd ) is a dense subspace of Lp (µ). For f ∈ Cc (Rd ) and N ∈ N, let
partition of unity of {hi }N
i=1 (N = ∞ is allowed here) subordinate to the cover
X
U such that supp(hi ) is compact for all i. fN (x) := f (x + 2πN n).
n∈Zd
d
Corollary 22.41. Suppose that X is an open subset of R and U = {Uα }α∈A ⊂ Show fN ∈ BC(Rd ) and x → fN (N x) is 2π – periodic, so by Exercise 15.13,
τ is an open cover of X. Then there exists a smooth partition of unity of x → fN (N x) can be approximated uniformly by trigonometric polynomials.
{hα }α∈A subordinate to the cover U such that supp(hα ) ⊂ Uα for all α ∈ A. p
Use this fact to conclude that fN ∈ D̄L (µ) . After this show fN → f in Lp (µ).
Moreover if Ūα is compact for each α ∈ A we may choose hα so that hα ≺ Uα .
Exercise 22.12. Suppose that µ and ν are two finite measures on Rd such that
Z Z
eiλ·x dµ(x) = eiλ·x dν(x) (22.16)
22.3 Exercises Rd Rd
d
for all λ ∈ R . Show µ = ν.
Exercise 22.6. Let (X, τ ) be a topological space, µ a measure on BX = σ(τ )
Hint: Perhaps the easiest way to do this is to use Exercise 22.11 with the
and f : X → C be a measurable function. Letting ν be the measure, dν = |f | dµ,
measure µ being replaced by µ + ν. Alternatively, use the Rmethod of proof
show supp(ν) = suppµ (f ), where supp(ν) is defined in Definition 21.41). R
of Exercise 22.11 to show Eq. (22.16) implies Rd f dµ(x) = Rd f dν(x) for all
Exercise 22.7. Let (X, τ ) be a topological space, µ a measure on BX = σ(τ ) f ∈ Cc (Rd ) and then apply Corollary 18.58.
such that supp(µ) = X (see Definition 21.41). Show suppµ (f ) = supp(f ) = Exercise R 22.13. Again let µ be a finite measure on BRd . Further assume that
{f 6= 0} for all f ∈ C(X). CM := Rd eM |x| dµ(x) < ∞ for all M ∈ (0, ∞). Let P(Rd ) be the space of
p
polynomials, ρ(x) = |α|≤N ρα xα with ρα ∈ C, on Rd . (Notice that |ρ(x)| ≤
P
Exercise 22.8. Prove the following strong version of item 3. of Proposition
13.52, namely to every pair of points, x0 , x1 , in a connected open subset V of CeM |x| for some constant C = C(ρ, p, M ), so that P(Rd ) ⊂ Lp (µ) for all 1 ≤
Rd there exists σ ∈ C ∞ (R, V ) such that σ(0) = x0 and σ(1) = x1 . Hint: First p < ∞.) Show P(Rd ) is dense in Lp (µ) for all 1 ≤ p < ∞. Here is a possible
choose a continuous path γ : [0, 1] → V such that γ (t) = x0 for t near 0 and outline.
n
γ (t) = x1 for t near 1 and then use a convolution argument to smooth γ. Outline: Fix a λ ∈ Rd and let fn (x) = (λ · x) /n! for all n ∈ N.
α
1. Use calculus to verify supt≥0 tα e−M t = (α/M ) e−α for all α ≥ 0 where
Exercise 22.9. Prove Proposition 22.34 by appealing to Corollary 19.43. 0
(0/M ) := 1. Use this estimate along with the identity
Exercise 22.10 (Integration by Parts). Suppose that (x, y) ∈ R × Rd−1 → pn pn pn

pn

pn
|λ · x| ≤ |λ| |x| = |x| e−M |x| |λ| eM |x|
f (x, y) ∈ C and (x, y) ∈ R × Rd−1 → g(x, y) ∈ C are measurable functions
such that for each fixed y ∈ Rd , x → f (x, y) and x → g(x, y) are continuously to find an estimate on kfn kp .

22.3 Exercises 243
P∞
2. Use your estimate on kfn kp to show n=0 kfn kp < ∞ and conclude This violates the definition of T and therefore T = ∞ and in particular we
may take T = 1 to learn
N
iλ·(·) X n Z
lim e − i fn = 0. h(x)eiλ·x dµ(x) = 0 for all λ ∈ Rd .
N →∞
n=0 p Rd
3. Now finish by appealing to Exercise 22.11. 5. Use Exercise 22.11 to conclude that
Z
Exercise 22.14. Again let µ be R a finite measure on BRd but now assume there h(x)g(x)dµ(x) = 0
exists an ε > 0 such that RC := Rd eε|x| dµ(x) < ∞. Also let q > 1 and h ∈ Lq (µ) Rd
be a function such that Rd h(x)xα dµ(x) = 0 for all α ∈ Nd0 . (As mentioned in
for all g ∈ Lp (µ). Now choose g judiciously to finish the proof.
Exercise 22.14, P(Rd ) ⊂ Lp (µ) for all 1 ≤ p < ∞, so x → h(x)xα is in L1 (µ).)
Show h(x) = 0 for µ– a.e. x using the following outline.
n
Outline: Fix a λ ∈ Rd , let fn (x) = (λ · x) /n! for all n ∈ N, and let
p = q/(q − 1) be the conjugate exponent to q.
α
1. Use calculus to verify supt≥0 tα e−εt = (α/ε) e−α for all α ≥ 0 where
0
(0/ε) := 1. Use this estimate along with the identity

pn pn pn pn pn
|λ · x| ≤ |λ| |x| = |x| e−ε|x| |λ| eε|x|
to find an estimate on kfn kp .

2. Use your estimate on kfn kp to show there exists δ > 0 such that
P∞ iλ·x
n=0 kfn kp < ∞ when |λ| ≤ δ and conclude for |λ| ≤ δ that e = Lp (µ)-
P∞ n
n=0 i fn (x). Conclude from this that
Z
h(x)eiλ·x dµ(x) = 0 when |λ| ≤ δ.
Rd
3. Let λ ∈ Rd (|λ| not necessarily small) and set g(t) := Rd eitλ·x h(x)dµ(x)
R
for t ∈ R. Show g ∈ C ∞ (R) and

Z
g (n) (t) = (iλ · x)n eitλ·x h(x)dµ(x) for all n ∈ N.
Rd
4. Let T = sup{τ ≥ 0 : g|[0,τ ] ≡ 0}. By Step 2., T ≥ δ. If T < ∞, then

Z
0 = g (n) (T ) = (iλ · x)n eiT λ·x h(x)dµ(x) for all n ∈ N.
Rd
Use Step 3. with h replaced by eiT λ·x h(x) to conclude

Z
g(T + t) = ei(T +t)λ·x h(x)dµ(x) = 0 for all t ≤ δ/ |λ| .
Rd

Part VI
Further Hilbert and Banach Space Techniques

23
L2 - Hilbert Spaces Techniques and Fourier Series
This section is concerned with Hilbert spaces presented as in the following Remark 23.3 (An Interesting Phenomena). Let H = L2 ([−1, 1], dm) and B :=
example. {1, x3 , x6 , x9 , . . . }. Then again A is total in H by the same argument as in
item 2. Example 23.2. This is true even though B is a proper subset of A.
Example 23.1. Let (X, M, µ) be a measure space. Then H := L2 (X, M, µ) with Notice that A is an algebraic basis for the polynomials on [−1, 1] while B is
inner product Z not! The following computations may help relieve some of the reader’s anxiety.
hf |gi = f · ḡdµ Let f ∈ L2 ([−1, 1], dm), then, making the change of variables x = y 1/3 , shows
X that
is a Hilbert space. Z 1 Z 1 2 1 Z 1 2
2
|f (x)| dx = f (y 1/3 ) y −2/3 dy = f (y 1/3 ) dµ(y) (23.2)

It will be convenient to define −1 −1 3 −1
Z
hf, gi := f (x) g (x) dµ (x) (23.1) where dµ(y) = 31 y −2/3 dy. Since µ([−1, 1]) = m([−1, 1]) = 2, µ is a finite mea-
X sure on [−1, 1] and hence by Exercise 22.13 A := {1, x, x2 , x3 . . . } is total (see
Definition 8.25) in L2 ([−1, 1], dµ). In particular for any ε > 0 there exists a
for all measurable functions f, g on X such that f g ∈ L1 (µ) . So with this polynomial p(y) such that
notation we have hf |gi = hf, ḡi for all f, g ∈ H.
Z 1 2
Exercise 23.1. Let K : L2 (ν) → L2 (µ) be the operator defined in Exercise f (y 1/3 ) − p(y) dµ(y) < ε2 .

21.12. Show K ∗ : L2 (µ) → L2 (ν) is the operator given by −1
Z However, by Eq. (23.2) we have

K ∗ g(y) = k̄(x, y)g(x)dµ(x). Z 1 2 Z 1
X
ε2 > f (y 1/3 ) − p(y) dµ(y) = f (x) − p(x3 )2 dx.

−1 −1
23.1 L2 -Orthonormal Basis Alternatively, if f ∈ C([−1, 1]), then g(y) = f (y 1/3 ) is back in C([−1, 1]).
Therefore for any ε > 0, there exists a polynomial p(y) such that
Example 23.2. 1. Let H = L2 ([−1, 1], dm), A := {1, x, x2 , x3 . . . } and β ⊂ H
be the result of doing the Gram-Schmidt procedure on A. By the Stone- ε > kg − pk∞ = sup {|g(y) − p(y)| : y ∈ [−1, 1]}
Weierstrass theorem or by Exercise 22.13 directly, A is total in H. Hence = sup g(x3 ) − p(x3 ) : x ∈ [−1, 1]

by Remark 8.26, β is an orthonormal basis for H. The basis, β, consists
= sup f (x) − p(x3 ) : x ∈ [−1, 1] .

of polynomials which up to normalization are the so called “Legendre
polynomials.” This gives another proof the polynomials in x3 are dense in C([−1, 1]) and hence
1 2
2. Let H = L2 (R, e− 2 x dx) and A := {1, x, x2 , x3 . . . }. Again by Exercise in L2 ([−1, 1]).
22.13, A is total in H and hence the Gram-Schmidt procedure applied to A
produces an orthonormal basis, β, of polynomial functions for H. This basis Exercise 23.2. Suppose (X, M, µ) and (Y, N , ν) are σ-finite measure
∞
consists, up to normalizations, of the so called “Hermite polynomials” spaces such that L2 (µ) and L2 (ν) are separable. If {fn }n=1 and
∞ 2 2
on R. {gm }m=1 are orthonormal bases for L (µ) and L (ν) respectively, then
248 23 L2 - Hilbert Spaces Techniques and Fourier Series
β := {fn ⊗ gm : m, n ∈ N} is an orthonormal basis for L2 (µ ⊗ ν) . (Recall that 2. For all K ∈ L(H, B), kKkHS = kK ∗ kHS and
f ⊗ g (x, y) := f (x) g (y) , see Notation 20.4.) Hint: model your proof on the
proof of Proposition 8.29. kKkHS ≥ kKkop := sup {kKhk : h ∈ H such that khk = 1} .
Definition 23.4 (External direct sum of Hilbert spaces). Suppose that 3. The set HS(H, B) is a subspace of L (H, B) (the bounded operators from
∞
{Hn }n=1 is aQsequence of Hilbert spaces. Let ⊕∞
n=1 Hn denote the space of se- H → B), k·kHS is a norm on HS(H, B) for which (HS(H, B), k·kHS ) is a
∞
quences, f ∈ n=1 Hn such that Hilbert space, and the corresponding inner product is given by
v
u∞ ∞
X
uX
kf k = t
2
kf (n)kHn < ∞. hK1 |K2 iHS = hK1 en |K2 en i . (23.3)
n=1
n=1
4. If K : H → B is a bounded
PN finite rank operator, then K is Hilbert Schmidt.
It is easily seen that (⊕∞ n=1 Hn , k·k) is a Hilbert space with inner product defined,
for all f, g ∈ ⊕∞ H , by
5. Let PN x := n=1 hx|en i en be orthogonal projection onto
n=1 n
span{en : n ≤ N } ⊂ H and for K ∈ HS(H, B), let KN := KPN .
∞
X Then
2 2
hf |gi⊕∞
n=1 Hn
= hf (n) |g (n)iHn . kK − KN kop ≤ kK − KN kHS → 0 as N → ∞,
n=1
which shows that finite rank operators are dense in (HS(H, B), k·kHS ) . In
Exercise 23.3. Suppose H is a Hilbert space and {Hn : n ∈ N} are closed sub- particular of HS(H, B) ⊂ K(H, B) – the space of compact operators from
spaces of H such that Hn ⊥ Hm for all m 6= n and if fP ∈ H with f ⊥ Hn for all H → B.
∞
n ∈ N, then f = 0. For f ∈ ⊕∞ n=1 Hn , show the sum n=1 f (n) is convergent 6. If Y is another Hilbert space and A : Y → H and C : B → Y are bounded
∞
in H and the map U : ⊕∞
P
n=1 Hn → H defined by U f := n=1 f (n) is unitary. operators, then
`∞
Exercise 23.4. Suppose (X, M, µ) is a measure space and X = n=1 Xn with kKAkHS ≤ kKkHS kAkop and kCKkHS ≤ kKkHS kCkop ,
Xn ∈ M and µ (Xn ) > 0 for all n. Then U : L2 (X, µ) → ⊕∞ 2
n=1 L (Xn , µ)
defined by (U f ) (n) := f 1Xn is unitary. in particular HS(H, H) is an ideal in L (H) .
Proof. Items 1. and 2. By Parseval’s equality and Fubini’s theorem for

23.2 Hilbert Schmidt Operators sums,
∞ ∞ X
∞
In this section H and B will be Hilbert spaces.
X 2
X 2
kKen k = |hKen |um i|
n=1 n=1 m=1
Proposition 23.5. Let H and B be a separable Hilbert spaces, K : H → B be ∞ X ∞ ∞
∞
a bounded linear operator, {en }∞ n=1 and {um }m=1 be orthonormal basis for H
X 2
X 2
= |hen |K ∗ um i| = kK ∗ um k .
and B respectively. Then: m=1 n=1 m=1
P∞ 2 P∞ 2
1. n=1 kKen k = m=1 kK ∗ um k allowing for the possibility that the sums This proves kKkHS is well defined independent of basis and that kKkHS =
are infinite. In particular the Hilbert Schmidt norm of K, kK ∗ kHS . For x ∈ H \ {0} , x/ kxk may be taken to be the first element in an
∞
X orthonormal basis for H and hence
2 2
kKkHS := kKen k ,
K x ≤ kKk .

n=1
kxk HS
is well defined independent of the choice of orthonormal basis {en }∞
n=1 . We
say K : H → B is a Hilbert Schmidt operator if kKkHS < ∞ and let Multiplying this inequality by kxk shows kKxk ≤ kKkHS kxk and hence
HS(H, B) denote the space of Hilbert Schmidt operators from H to B. kKkop ≤ kKkHS .

23.2 Hilbert Schmidt Operators 249
Item 3. For K1 , K2 ∈ L(H, B), Item 4. Since Nul(K ∗ )⊥ = Ran (K) = Ran (K) ,
v N
u∞ 2 2
X 2
uX 2 kKkHS = kK ∗ kHS = kK ∗ vn kH < ∞
kK1 + K2 kHS = t kK1 en + K2 en k
n=1
n=1
N
v
u∞ where N := dim Ran (K) and {vn }n=1 is an orthonormal basis for Ran (K) =
uX 2 K (H) .
≤t [kK1 en k + kK2 en k]
Item 5. Simply observe,
n=1
2 2
X 2
∞
= k{kK1 en k + kK2 en k}n=1 k`2 kK − KN kop ≤ kK − KN kHS = kKen k → 0 as N → ∞.
∞ ∞ n>N
≤ k{kK1 en k}n=1 k`2 + k{kK2 en k}n=1 k`2
Item 6. For C ∈ L(B, Y ) and K ∈ L(H, B) then
= kK1 kHS + kK2 kHS .
∞
X ∞
X
2 2 2 2 2 2
From this triangle inequality and the homogeneity properties of k·kHS , we now kCKkHS = kCKen k ≤ kCkop kKen k = kCkop kKkHS
easily see that HS(H, B) is a subspace of L(H, B) and k·kHS is a norm on n=1 n=1
HS(H, B). Since and for A ∈ L (Y, H) ,

∞
X ∞
X kKAkHS = kA∗ K ∗ kHS ≤ kA∗ kop kK ∗ kHS = kAkop kKkHS .
|hK1 en |K2 en i| ≤ kK1 en k kK2 en k
n=1 n=1
v v
u∞ u∞ Remark 23.6. The separability assumptions made in Proposition 23.5 are un-
uX uX
2t 2
≤ t kK1 en k kK2 en k = kK1 kHS kK2 kHS , necessary. In general, we define
n=1 n=1 2
X 2
kKkHS = kKek
e∈β
the sum in Eq. (23.3) is well defined and is easily checked to define an inner
2
product on HS(H, B) such that kKkHS = hK|KiHS . where β ⊂ H is an orthonormal basis. The same proof of Item 1. of Proposition
2
The proof that
2
HS(H, B), k·kHS is complete is very similar to the proof 23.5 shows kKkHS is well defined and kKkHS = kK ∗ kHS . If kKkHS < ∞,
∞ then there exists a countable subset β0 ⊂ β such that Ke = 0 if e ∈ β \ β0 . Let
of Theorem 7.5. Indeed, suppose {Km }m=1 is a k·kHS – Cauchy sequence in H0 := span(β0 ) and B0 := K(H0 ). Then K (H) ⊂ B0 , K|H0⊥ = 0 and hence by
HS(H, B). Because L(H, B) is complete, there exists K ∈ L(H, B) such that
applying the results of Proposition 23.5 to K|H0 : H0 → B0 one easily sees that
kK − Km kop → 0 as m → ∞. Thus, making use of Fatou’s Lemma 4.12,
the separability of H and B are unnecessary in Proposition 23.5.
∞
2
X 2
Example 23.7. Let (X, µ) be a measure space, H = L2 (X, µ) and
kK − Km kHS = k(K − Km ) en k n
n=1
X
∞ k(x, y) := fi (x)gi (y)
X 2 i=1
= lim inf k(Kl − Km ) en k
l→∞
n=1 where
∞
X fi , gi ∈ L2 (X, µ) for i = 1, . . . , n.
2
≤ lim inf k(Kl − Km ) en k
l→∞ Define Z
n=1
2 (Kf )(x) = k(x, y)f (y)dµ(y),
= lim inf kKl − Km kHS → 0 as m → ∞. X
l→∞
then K : L2 (X, µ) → L2 (X, µ) is a finite rank operator and hence Hilbert
2
Hence K ∈ HS(H, B) and limm→∞ kK − Km kHS = 0. Schmidt.

Exercise 23.5. Suppose that (X, µ) is a σ–finite measure space such that H = by the dominated convergence theorem. For any ε > 0,
L2 (X, µ) is separable and k : X × X → R is a measurable function, such that Z 2
1
Z α dxdy < ∞,
2
kkkL2 (X×X,µ⊗µ) := |k(x, y)|2 dµ(x)dµ(y) < ∞. Ω×Ω |x − y| + ε
X×X
and hence Kε is Hilbert Schmidt and hence compact. By Eq. (23.4), Kε → K
Define, for f ∈ H, Z as ε ↓ 0 and hence it follows that K is compact as well.
Kf (x) = k(x, y)f (y)dµ(y), Exercise 23.6. Let H := L2 ([0, 1] , m) , k (x, y) := min (x, y) for x, y ∈ [0, 1]
X
and define K : H → H by
when the integral makes sense. Show:
Z 1
1. Kf (x) is defined for µ–a.e. x in X. Kf (x) = k (x, y) f (y) dy.
2. The resulting function Kf is in H and K : H → H is linear. 0
3. kKkHS = kkkL2 (X×X,µ⊗µ) < ∞. (This implies K ∈ HS(H, H).) By Exercise 23.5, K is a Hilbert Schmidt operator and it is easily seen that K
is self-adjoint. Show:
Example 23.8. Suppose that Ω ⊂ Rn is a bounded set, α < n, then the operator
K : L2 (Ω, m) → L2 (Ω, m) defined by 1. If g ∈ C 2 ([0, 1]) with g (0) = 0 = g 0 (1) , then Kg 00 = −g. Use this to
Z conclude hKf |g 00 i = −hf |gi for all g ∈ Cc∞ ((0, 1)) and consequently that
1 Nul(K) = {0} .
Kf (x) := α f (y)dy
Ω |x − y| 2. Now suppose that f ∈ H is an eigenvector of K with eigenvalue λ 6= 0.
Show that there is a version1 of f which is in C ([0, 1]) ∩ C 2 ((0, 1)) and this
is compact.
version, still denoted by f, solves
Proof. For ε ≥ 0, let
λf 00 = −f with f (0) = f 0 (1) = 0. (23.5)
Z
1
Kε f (x) := α f (y)dy = [gε ∗ (1Ω f )] (x) where f 0 (1) := limx↑1 f 0 (x) .
Ω |x − y| + ε
3. Use Eq. (23.5) to find all the eigenvalues and eigenfunctions of K.
where gε (x) = |x|α1+ε 1C (x) with C ⊂ Rn a sufficiently large ball such that 4. Use the results above along with the spectral Theorem 8.46, to show
Ω − Ω ⊂ C. Since α < n, it follows that √

1
2 sin n+ πx : n ∈ N0
−α 2
gε ≤ g0 = |·| 1C ∈ L1 (Rn , m).
is an orthonormal basis for L2 ([0, 1] , m) .
Hence it follows by Proposition 22.23 that
Exercise 23.7. Let (X, M, µ) be a σ – finite measure space, a ∈ L∞ (µ) and
k(K − Kε ) f kL2 (Ω) ≤ k(g0 − gε ) ∗ (1Ω f )kL2 (Rn ) let A be the bounded operator on H := L2 (µ) defined by Af (x) = a (x) f (x)
≤ k(g0 − gε )kL1 (Rn ) k1Ω f kL2 (Rn ) for all f ∈ H. (We will denote A by Ma in the future.) Show:
= k(g0 − gε )kL1 (Rn ) kf kL2 (Ω) 1. kAkop = kakL∞ (µ) .

2. A∗ = Mā .
which implies 3. σ (A) = essran(a) where σ (A) is the spectrum of A and essran(a) is the
essential range of a, see Definitions 8.31 and 21.40 respectively.
kK − Kε kB(L2 (Ω)) ≤ kg0 − gε kL1 (Rn ) 4. Show λ is an eigenvalue for A = Ma iff µ ({a = λ}) > 0, i.e. iff a has a “flat
Z spot of height λ.”

1 1
= α
− α dx → 0 as ε ↓ 0 (23.4)
C |x| + ε |x| 1
A measurable function g is called a version of f iff g = f a.e..

23.3 Fourier Series Considerations 251
23.3 Fourier Series Considerations Exercise 23.8. Let A be the operator defined in Lemma 8.37 and for g ∈
L2 (T) , let U g (k) := g̃ (k) so that U : L2 (T) → `2 (Z) is unitary. Show
Throughout this section we will let dθ, dx, dα, etc. denote Lebesgue measure U −1 AU = Ma where a ∈ Cper ∞
(R) is a function to be found. Use this rep-
on Rd normalized so that the cube, Q := (−π, π]d , has measure one, i.e. dθ = resentation and the results in Exercise 23.7 to give a simple proof of the results
(2π)−d dm(θ) where m is standard Lebesgue measure on Rd . As usual, for α ∈ in Lemma 8.37.
Nd0 , let
|α|
1 ∂ |α| 23.3.1 Dirichlet, Fejér and Kernels
Dθα = .
i ∂θ1 . . . ∂θdαd
α1
k Although the sum in Eq. (23.7) is guaranteed to converge relative to the Hilber-
Notation 23.9 Let Cper (Rd ) denote the 2π – periodic functions in C k (Rd ),
tian norm on H it certainly need not converge pointwise even if f ∈ Cper Rd
that is f ∈ Cper (R ) iff f ∈ C k (Rd ) and f (θ + 2πei ) = f (θ) for all θ ∈ Rd and
k d
as will be proved in Section 25.3.1 below. Nevertheless, if f is sufficiently regu-
i = 1, 2, . . . , d. Further let h·|·i denote the inner product on the Hilbert space, lar, then the sum in Eq. (23.7) will converge pointwise as we will now show. In
H := L2 ([−π, π]d ), given by the process we will give a direct and constructive proof of the result in Exercise
Z d Z 15.13, see Theorem 23.12 below.
1
hf |gi := f (θ)ḡ(θ)dθ = f (θ)ḡ(θ)dm (θ) Let us restrict our attention to d = 1 here. Consider
Q 2π Q " #
X X 1 Z
and define φk (θ) := eik·θ for all k ∈ Zd . For f ∈ L1 (Q), we will write f˜(k) for fn (θ) = f˜(k)φk (θ) = f (x)e−ik·x
dx φk (θ)
2π [−π,π]
the Fourier coefficient, |k|≤n |k|≤n
Z
Z 1 X
f˜(k) := hf |φk i = f (θ)e−ik·θ dθ. (23.6) = f (x) eik·(θ−x) dx
2π [−π,π]
Q |k|≤n
Z
1
Since any 2π – periodic functions on Rd may be identified with function on = f (x)Dn (θ − x)dx (23.8)
d 2π [−π,π]
the d - dimensional torus, Td ∼
= Rd / (2πZ) ∼
d
= S 1 , I may also write C k (Td )
k
for Cper (Rd ) and Lp Td for Lp (Q) where elements in f ∈ Lp (Q) are to be where
n
thought of as there extensions to 2π – periodic functions on Rd . X
Dn (θ) := eikθ
Theorem 23.10 (Fourier Series). The functions β := φk : k ∈ Zd form k=−n
an orthonormal basis for H, i.e. if f ∈ H then
is called the Dirichlet kernel. Letting α = eiθ/2 , we have
X X
f= hf |φk iφk = f˜(k)φk (23.7) n
X α2(n+1) − α−2n α2n+1 − α−(2n+1)
k∈Zd k∈Zd Dn (θ) = α2k = =
α2 − 1 α − α−1
k=−n
where the convergence takes place in L2 ([−π, π]d ).
2i sin(n + 12 )θ sin(n + 12 )θ
= = .

Proof. Simple computations show β := φk : k ∈ Zd is an orthonormal 2i sin 12 θ sin 12 θ
set. We now claim that β is an orthonormal basis. To see this recall that
Cc ((−π, π)d ) is dense in L2 ((−π, π)d , dm). Any f ∈ Cc ((−π, π)) may be ex- and therefore
n
tended to be a continuous 2π – periodic function on R and hence by Exercise X sin(n + 21 )θ
15.13 and Remark 15.44, f may uniformly (and hence in L2 ) be approximated Dn (θ) := eikθ = , (23.9)
k=−n
sin 12 θ
by a trigonometric polynomial. Therefore β is a total orthonormal set, i.e. β is
an orthonormal basis. see Figure 23.3.1.
This may also be proved by first proving the case d = 1 as above and then
using Exercise 23.2 inductively to get the result for any d.

N N Z
1 X 1 X 1 X
FN (θ) = fn (θ) = f (x) eik·(θ−x) dx
N + 1 n=0 N + 1 n=0 2π [−π,π]
|k|≤n
Z
1
= KN (θ − x)f (x)dx
2π [−π,π]
where
N
1 X X ik·θ
KN (θ) := e (23.11)
N + 1 n=0
|k|≤n
is the Fejér kernel.
Theorem 23.12. The Fejér kernel KN in Eq. (23.11) satisfies:

This is a plot D1 and D10 .
1.
with the understanding that the right side of this equation is 2n + 1 whenever N
θ ∈ 2πZ. X |n|
KN (θ) = 1− einθ (23.12)
N +1
1 n=−N
Theorem 23.11. Suppose f ∈ L ([−π, π] , dm) and f is differentiable at some
1 sin2 N2+1 θ

θ ∈ [−π, π] , then limn→∞ fn (θ) = f (θ) where fn is as in Eq. (23.8).
= . (23.13)
N + 1 sin2 θ2
Proof. Observe that
1
Z
1
Z X 2. KNR(θ) ≥ 0.
Dn (θ − x)dx = eik·(θ−x) dx = 1 1
3. 2π
π
KN (θ)dθ = 1
2π [−π,π] 2π [−π,π] −π
|k|≤n 4. supε≤|θ|≤π KN (θ) → 0 as N → ∞ for all ε > 0, see Figure 23.1.
5. For any continuous 2π – periodic function f on R, KN ∗ f (θ) → f (θ)
and therefore, uniformly in θ as N → ∞, where
Z
1 Z π
fn (θ) − f (θ) = [f (x) − f (θ)] Dn (θ − x)dx 1
2π [−π,π] KN ∗ f (θ) = KN (θ − α)f (α)dα
Z 2π −π
1 N
= [f (x) − f (θ − x)] Dn (x)dx

X |n|
2π [−π,π] = 1− f˜ (n) einθ . (23.14)
N +1
f (θ − x) − f (θ)
Z
1 1 n=−N
= 1 sin(n + )x dx. (23.10)
2π [−π,π] sin 2 x 2
If f is differentiable at θ, the last expression in Eq. (23.10) tends to 0 as n → ∞ Proof. 1. Equation (23.12) is a consequence of the identity,
by the Riemann Lebesgue Lemma (Corollary 22.17 or Lemma 22.37) and the N X
fact that 1[−π,π] (x) f (θ−x)−f (θ)
∈ L1 (dx) .
X X X
sin 21 x eik·θ = eik·θ = (N + 1 − |k|) eik·θ .
Despite the Dirichlet kernel not being positive, it still satisfies the approx- n=0 |k|≤n |k|≤n≤N |k|≤N
1
imate δ – sequence property, 2π Dn → δ0 as n → ∞, when acting on C 1 –
periodic functions in θ. In order to improve the convergence properties it is rea- Moreover, letting α = eiθ/2 and using Eq. (3.3) shows
sonable to try to replace {fn : n ∈ N0 } by the sequence of averages (see Exercise
7.14),

23.3 Fourier Series Considerations 253
1 π
Z

|KN ∗ f (θ) − f (θ)| = KN (θ − α) [f (α) − f (θ)] dα
2π −π
Z π
1
≤ KN (α) |f (θ − α) − f (θ)| dα
2π −π
Z
1 1 1 1
≤ kf k∞ + KN (α) |f (θ − α) − f (θ)| dα
π N + 1 sin2 2ε

2π |α|≤ε
1 1 1
≤ kf k∞ + sup |f (θ − α) − f (θ)| .
π N + 1 sin2 2ε |α|≤ε
Fig. 23.1. Plots of KN (θ) for N = 2, 7 and 13. Therefore,
lim sup kKN ∗ f − f k∞ ≤ sup sup |f (θ − α) − f (θ)| → 0 as ε ↓ 0.

N →∞ θ |α|≤ε
N X N
1 X 1 X α2n+2 − α−2n
KN (θ) = α2k =
N + 1 n=0 N + 1 n=0 α2 − 1
|k|≤n
1 XN 23.3.2 The Dirichlet Problems on D and the Poisson Kernel

− α−2n−1
2n+1
= −1
α
(N + 1) (α − α ) n=0 Let D := {z ∈ C : |z| < 1} be the open unit disk in C ∼
= R2 , write z ∈ C as
N iθ ∂2 ∂2
1 X z = x + iy or z = re , and let ∆ = ∂x2 + ∂y2 be the Laplacian acting on
αα − α−1 α−2n
2n
= C 2 (D) .
(N + 1) (α − α−1 ) n=0
−2N −2
2N +2
1 α −1 −1 α −1 Theorem 23.13 (Dirichlet problem for D). To every continuous function
= α −α g ∈ C (bd(D)) there exists a unique function u ∈ C(D̄) ∩ C 2 (D) solving
(N + 1) (α − α−1 ) α2 − 1 α−2 − 1
1 h
2(N +1) −2(N +1)
i
= 2 α − 1 + α − 1 ∆u(z) = 0 for z ∈ D and u|∂D = g. (23.15)
(N + 1) (α − α−1 )
1 h i2 Moreover for r < 1, u is given by,
= 2 α(N +1) − α−(N +1)
(N + 1) (α − α−1 ) 1
Z π
1 sin2 ((N + 1) θ/2) u(reiθ ) = Pr (θ − α)u(eiα )dα =: Pr ∗ u(eiθ ) (23.16)
= . 2π −π
N +1 sin2 (θ/2) 1
Z π
1 + rei(θ−α)
= Re i(θ−α)
u(eiα )dα (23.17)
Items 2. and 3. follow easily from Eqs. (23.13) and (23.12) respectively. Item 2π −π 1 − re
4. is a consequence of the elementary estimate;
where Pr is the Poisson kernel defined by
1 1
sup KN (θ) ≤ 1 − r2
N + 1 sin2 2ε

ε≤|θ|≤π Pr (δ) := .
1 − 2r cos δ + r2
and is clearly indicated in Figure 23.1. Item 5. now follows by the standard (The problem posed in Eq. (23.15) is called the Dirichlet problem for D.)
approximate δ – function arguments, namely,
1
Proof. In this proof, we are going to be identifying S = bd(D) :=
z ∈ D̄ : |z| = 1 with [−π, π]/ (π ∼ −π) by the map θ ∈ [−π, π] → eiθ ∈ S 1 .
Also recall that the Laplacian ∆ may be expressed in polar coordinates as,

1
Z π
1
∆u = r−1 ∂r r−1 ∂r u + 2 ∂θ2 u,

Ak = u(eiθ )e−ikθ dθ = g̃(k).
r 2π −π
where Hence if u is a solution to Eq. (23.15) then u must be given by
∂ ∂
(∂r u) reiθ = u reiθ and (∂θ u) reiθ = u reiθ .

X
∂r ∂θ u(reiθ ) = g̃(k)r|k| eikθ for r < 1. (23.22)
Uniqueness. Suppose u is a solution to Eq. (23.15) and let k∈Z
1
Z π or equivalently,
g̃(k) := g(eikθ )e−ikθ dθ
X X
2π −π u(z) = g̃(k)z k + g̃(−k)z̄ k .
k∈N0 k∈N
and Z π Notice that the theory of the Fourier series implies Eq. (23.22) is valid in the
1
ũ(r, k) := u(reiθ )e−ikθ dθ (23.18) L2 (dθ) - sense. However more is true, since for r < 1, the series in Eq. (23.22) is
2π −π
absolutely convergent and in fact defines a C ∞ – function (see Exercise 4.11 or
be the Fourier coefficients of g (θ) and θ → u reiθ respectively. Then for Corollary 19.43) which must agree with the continuous function, θ → u reiθ ,
r ∈ (0, 1) , for almost every θ and hence for all θ. This completes the proof of uniqueness.
Z π Existence. Given g ∈ C (bd(D)) , let u be defined as in Eq. (23.22). Then,
1
r−1 ∂r (r∂r ũ(r, k)) = r−1 ∂r r−1 ∂r u (reiθ )e−ikθ dθ again by Exercise 4.11 or Corollary 19.43, u ∈ C ∞ (D) . So to finish the proof

2π −π it suffices to show limx→y u (x) = g (y) for all y ∈ bd(D). Inserting the formula
Z π
1 1 2 for g̃(k) into Eq. (23.22) gives
=− ∂ u(reiθ )e−ikθ dθ
2π −π r2 θ Z π
iθ 1
1 1
Z π u(re ) = Pr (θ − α) u(eiα )dα for all r < 1
=− 2 u(reiθ )∂θ2 e−ikθ dθ 2π −π
r 2π −π
where
1
= 2 k 2 ũ(r, k) ∞ ∞
r X X X
Pr (δ) = r|k| eikδ = rk eikδ + rk e−ikδ − 1 =
or equivalently k∈Z k=0 k=0
r∂r (r∂r ũ(r, k)) = k 2 ũ(r, k). (23.19)
1

1 + reiδ

= Re 2 − 1 = Re
Recall the general solution to 1 − reiδ 1 − reiδ
" #
1 + reiδ 1 − re−iδ

1 − r2 + 2ir sin δ

r∂r (r∂r y(r)) = k 2 y(r) (23.20) = Re = Re (23.23)
2 1 − 2r cos δ + r2
α
|1 − reiδ |
may be found by trying solutions of the form y(r) = r which then implies
α2 = k 2 or α = ±k. From this one sees that ũ(r, k) solving Eq. (23.19) may 1 − r2
= .
be written as ũ(r, k) = Ak r|k| + Bk r−|k| for some constants Ak and Bk when 1 − 2r cos δ + r2
k 6= 0. If k = 0, the solution to Eq. (23.20) is gotten by simple integration and The Poisson kernel again solves the usual approximate δ – function proper-
the result is ũ(r, 0) = A0 + B0 ln r. Since ũ(r, k) is bounded near the origin for ties (see Figure 2), namely:
each k it must be that Bk = 0 for all k ∈ Z. Hence we have shown there exists
Ak ∈ C such that, for all r ∈ (0, 1), 1. Pr (δ) > 0 and
Z π Z π X
Z π 1 1
1 Pr (θ − α) dα = r|k| eik(θ−α) dα
Ak r|k| = ũ(r, k) = u(reiθ )e−ikθ dθ. (23.21) 2π −π 2π −π
2π −π k∈Z
1 X |k| π ik(θ−α)
Z
Since all terms of this equation are continuous for r ∈ [0, 1], Eq. (23.21) remains = r e dα = 1
2π −π
valid for all r ∈ [0, 1] and in particular we have, at r = 1, that k∈Z

23.4 Weak L2 -Derivatives 255
and 23.4 Weak L2 -Derivatives
2.
1 − r2 Theorem 23.15 (Weak and Strong Differentiability). Suppose that f ∈
sup Pr (θ) ≤ → 0 as r ↑ 1.
ε≤|θ|≤π 1 − 2r cos ε + r2 L2 (Rn ) and v ∈ Rn \ {0} . Then the following are equivalent:
∞
1. There exists {tn }n=1 ⊂ R\ {0} such that limn→∞ tn = 0 and

f (· + tn v) − f (·)
sup
< ∞.
n tn
2
2. There exists g ∈ L2 (Rn ) such that hf, ∂v φi = −hg, φi for all φ ∈ Cc∞ (Rn ).
L2 L2
3. There exists g ∈ L2 (Rn ) and fn ∈ Cc∞ (Rn ) such that fn → f and ∂v fn → g
as n → ∞.
A plot of Pr (δ) for r = 0.2, 0.5 and 0.7. 4. There exists g ∈ L2 such that
Therefore by the same argument used in the proof of Theorem 23.12, f (· + tv) − f (·) L2
→ g as t → 0.
t
lim sup u reiθ − g eiθ = lim sup (Pr ∗ g) eiθ − g eiθ = 0

r↑1 θ r↑1 θ (See Theorem 26.18 for the Lp generalization of this theorem.)
which certainly implies limx→y u (x) = g (y) for all y ∈ bd(D). Proof. 1. =⇒ 2. We may assume, using Theorem 14.43 and passing to a
(·) w
subsequence if necessary, that f (·+tntv)−f → g for some g ∈ L2 (Rn ). Now for
Remark 23.14 (Harmonic Conjugate). Writing z = reiθ , Eq. (23.17) may be ∞ n
φ ∈ Cc (R ),
n
rewritten as Z π
1 1 + ze−iα
f (· + tn v) − f (·)

φ(· − tn v) − φ(·)

u(z) = Re −iα
u(eiα )dα hg|φi = lim , φ = lim f,
2π −π 1 − ze n→∞ tn n→∞ tn
which shows u = Re F where

φ(· − tn v) − φ(·)
= f, lim = −hf, ∂v φi,
1
Z π
1 + ze−iα
n→∞ tn
F (z) := u(eiα )dα.
2π −π 1 − ze−iα wherein we have used the translation invariance of Lebesgue measure and
the dominated convergence theorem. 2. =⇒ 3. Let φ ∈ Cc∞ (Rn , R) such
Moreover it follows from Eq. (23.23) that
that Rn φ(x)dx = 1 and let φm (x) = mn φ(mx), then by Proposition 22.34,
R
1
Z π
r sin(θ − α) hm := φm ∗ f ∈ C ∞ (Rn ) for all m and
Im F (reiθ ) = Im 2
g(eiα )dα
π −π 1 − 2r cos(θ − α) + r
Z
∂v hm (x) = ∂v φm ∗ f (x) = ∂v φm (x − y)f (y)dy = hf, −∂v [φm (x − ·)]i
=: (Qr ∗ u) (eiθ ) Rn
= hg, φm (x − ·)i = φm ∗ g(x).
where
r sin(δ)
Qr (δ) := . By Theorem 22.32, hm → f ∈ L2 (Rn ) and ∂v hm = φm ∗ g → g in L2 (Rn ) as
1 − 2r cos(δ) + r2
m → ∞. This shows 3. holds except for the fact that hm need not have compact
From these remarks it follows that v =: (Qr ∗ g) (eiθ ) is the harmonic conjugate support. To fix this let ψ ∈ Cc∞ (Rn , [0, 1]) such that ψ = 1 in a neighborhood
of u and P̃r = Qr . For more on this point see Section ?? below. of 0 and let ψε (x) = ψ(εx) and (∂v ψ)ε (x) := (∂v ψ) (εx). Then
∂v (ψε hm ) = ∂v ψε hm + ψε ∂v hm = ε (∂v ψ)ε hm + ψε ∂v hm

so that ψε hm → hm in L2 and ∂v (ψε hm ) → ∂v hm in L2 as ε ↓ 0. Let fm = 23.5 *Conditional Expectation
ψεm hm where εm is chosen to be greater than zero but small enough so that
In this section let (Ω, F, P ) be a probability space, i.e. (Ω, F, P ) is a measure
kψεm hm − hm k2 + k∂v (ψεm hm ) → ∂v hm k2 < 1/m. space and P (Ω) = 1. Let G ⊂ F be a sub – sigma algebra of F and write f ∈ Gb
if f : Ω → C is bounded and f is (G, BC ) – measurable. In this section we will
Then fm ∈ Cc∞ (Rn ), fm → f and ∂v fm → g in L2 as m → ∞. 3. =⇒ 4. By the write Z
fundamental theorem of calculus
Ef := f dP.
τ−tv fm (x) − fm (x) fm (x + tv) − fm (x) Ω
=
t t Definition 23.16 (Conditional Expectation). Let EG : L2 (Ω, F, P ) →
1 1 d
Z 1
L2 (Ω, G, P ) denote orthogonal projection of L2 (Ω, F, P ) onto the closed sub-
Z
= fm (x + stv)ds = (∂v fm ) (x + stv)ds.
t 0 ds 0 space L2 (Ω, G, P ). For f ∈ L2 (Ω, G, P ), we say that EG f ∈ L2 (Ω, F, P ) is the
(23.24) conditional expectation of f.
Let Theorem 23.17. Let (Ω, F, P ) and G ⊂ F be as above and f, g ∈ L2 (Ω, F, P ).

Z 1 Z 1
Gt (x) := τ−stv g(x)ds = g(x + stv)ds 1. If f ≥ 0, P – a.e. then EG f ≥ 0, P – a.e.
0 0
2. If f ≥ g, P – a.e. there EG f ≥ EG g, P – a.e.
which is defined for almost every x and is in L2 (Rn ) by Minkowski’s inequality 3. |EG f | ≤ EG |f | , P – a.e.
for integrals, Theorem 21.27. Therefore 4. kEG f kL1 ≤ kf kL1 for all f ∈ L2 . So by the B.L.T. Theorem 10.4, EG
Z 1 extends uniquely to a bounded linear map from L1 (Ω, F, P ) to L1 (Ω, G, P )
τ−tv fm (x) − fm (x) which we will still denote by EG .
− Gt (x) = [(∂v fm ) (x + stv) − g(x + stv)] ds
t 0 5. If f ∈ L1 (Ω, F, P ) then F = EG f ∈ L1 (Ω, G, P ) iff
and hence again by Minkowski’s inequality for integrals, E(F h) = E(f h) for all h ∈ Gb .
Z 1
τ−tv fm − fm
6. If g ∈ Gb and f ∈ L1 (Ω, F, P ), then EG (gf ) = g · EG f, P – a.e.

− Gt ≤
kτ−stv (∂v fm ) − τ−stv gk2 ds
t 2 0
Z 1 Proof. By the definition of orthogonal projection for h ∈ Gb ,
= k∂v fm − gk2 ds.
0 E(f h) = E(f · EG h) = E(EG f · h).
Letting m → ∞ in this equation implies (τ−tv f − f ) /t = Gt a.e. Finally one So if f, h ≥ 0 then 0 ≤ E(f h) ≤ E(EG f · h) and since this holds for all h ≥ 0
more application of Minkowski’s inequality for integrals implies, in Gb , EG f ≥ 0, P – a.e. This proves (1). Item (2) follows by applying item

τ−tv f − f

Z 1

(1). to f − g. If f is real, ±f ≤ |f | and so by Item (2), ±EG f ≤ EG |f | , i.e.
− g = kGt − gk2 =
(τ−stv g − g) ds |EG f | ≤ EG |f | , P – a.e. For complex f, let h ≥ 0 be a bounded and G –
t 0

2
Z 1
2 measurable function. Then
≤ kτ−stv g − gk2 ds.
h i h i
0
E [|EG f | h] = E EG f · sgn (EG f )h = E f · sgn (EG f )h
By the dominated convergence theorem and Proposition 22.24, the latter term ≤ E [|f | h] = E [EG |f | · h] .
tends to 0 as t → 0 and this proves 4. The proof is now complete since 4. =⇒
Since h is arbitrary, it follows that |EG f | ≤ EG |f | , P – a.e. Integrating this
1. is trivial.
inequality implies
kEG f kL1 ≤ E |EG f | ≤ E [EG |f | · 1] = E [|f |] = kf kL1 .

23.5 *Conditional Expectation 257
Item (5). Suppose f ∈ L1 (Ω, F, P ) and h ∈ Gb . Let fn ∈ L2 (Ω, F, P ) be a is a µ – null set. Since
sequence of functions such that fn → f in L1 (Ω, F, P ). Then Z Z Z
E [|F ◦ πX |] = dµ(x) dν(y) |F (x)| ρ(x, y) = dµ(x) |F (x)| ρ̄(x)
E(EG f · h) = E( lim EG fn · h) = lim E(EG fn · h)
n→∞ n→∞ ZX ZY X

= lim E(fn · h) = E(f · h). (23.25) = dµ(x) ν(dy)f (x, y)ρ(x, y)
n→∞
ZX ZY
This equation uniquely determines EG , for if F ∈ L1 (Ω, G, P ) also satisfies ≤ dµ(x) ν(dy) |f (x, y)| ρ(x, y) < ∞,
E(F · h) = E(f · h) for all h ∈ Gb , then taking h = sgn (F − EG f ) in Eq. X Y
(23.25) gives 1
F ◦ πX ∈ L (Ω, G, P ). Let h = H ◦ πX be a bounded G – measurable function,
0 = E((F − EG f ) h) = E(|F − EG f |).
then
This shows F = EG f, P – a.e. Item (6) is now an easy consequence of this Z Z
characterization, since if h ∈ Gb , E [F ◦ πX · h] = dµ(x) dν(y)F (x)H(x)ρ(x, y)
ZX Y
E [(gEG f ) h] = E [EG f · hg] = E [f · hg] = E [gf · h] = E [EG (gf ) · h] . = dµ(x)F (x)H(x)ρ̄(x)
Thus EG (gf ) = g · EG f, P – a.e. ZX Z
= dµ(x)H(x) ν(dy)f (x, y)ρ(x, y)
Proposition 23.18. If G0 ⊂ G1 ⊂ F. Then X Y
= E [hf ]
EG0 EG1 = EG1 EG0 = EG0 . (23.26)
and hence EG f = F ◦ πX as claimed.
Proof. Equation (23.26) holds on L2 (Ω, F, P ) by the basic properties of or-
thogonal projections. It then hold on L1 (Ω, F, P ) by continuity and the density This example shows that conditional expectation is a generalization of the
of L2 (Ω, F, P ) in L1 (Ω, F, P ). notion of performing integration over a partial subset of the variables in the
integrand. Whereas to compute the expectation, one should integrate over all
Example 23.19. Suppose that (X, M, µ) and (Y, N , ν) are two σ – finite measure of the variables. See also Exercise 23.25 to gain more intuition about conditional
spaces. Let Ω = X × Y, F = M ⊗ N and P (dx, dy) =R ρ(x, y)µ(dx)ν(dy) where expectations.
ρ ∈ L1 (Ω, F, µ ⊗ ν) is a positive function such that X×Y ρd (µ ⊗ ν) = 1. Let
πX : Ω → X be the projection map, πX (x, y) = x, and Theorem 23.20 (Jensen’s inequality). Let (Ω, F, P ) be a probability space
and ϕ : R → R be a convex function. Assume f ∈ L1 (Ω, F, P ; R) is a function
−1
G := σ(πX ) = πX (M) = {A × Y : A ∈ M} . such that (for simplicity) ϕ(f ) ∈ L1 (Ω, F, P ; R), then ϕ(EG f ) ≤ EG [ϕ(f )] , P
– a.e.
Then f : Ω → R is G – measurable iff f = F ◦ πX for some function F : X → R
which is N – measurable, see Lemma 18.66. For f ∈ L1 (Ω, F, P ), we will now Proof. Let us first assume that φ is C 1 and f is bounded. In this case
show EG f = F ◦ πX where
ϕ(x) − ϕ(x0 ) ≥ ϕ0 (x0 )(x − x0 ) for all x0 , x ∈ R. (23.27)
Z
1
F (x) = 1(0,∞) (ρ̄(x)) · f (x, y)ρ(x, y)ν(dy), Taking x0 = EG f and x = f in this inequality implies
ρ̄(x) Y
ρ̄(x)
R R
:= Y ρ(x, y)ν(dy). (By convention, Y f (x, y)ρ(x, y)ν(dy) := 0 if ϕ(f ) − ϕ(EG f ) ≥ ϕ0 (EG f )(f − EG f )
R
Y
|f (x, y)| ρ(x, y)ν(dy) = ∞.) and then applying EG to this inequality gives
By Tonelli’s theorem, the set
Z EG [ϕ(f )] − ϕ(EG f ) = EG [ϕ(f ) − ϕ(EG f )]
E := {x ∈ X : ρ̄(x) = ∞} ∪ x ∈ X : |f (x, y)| ρ(x, y)ν(dy) = ∞ ≥ ϕ0 (EG f )(EG f − EG EG f ) = 0
Y

The same proof works for general φ, one need only use Proposition 21.8 to
replace Eq. (23.27) by
ϕ(x) − ϕ(x0 ) ≥ ϕ0− (x0 )(x − x0 ) for all x0 , x ∈ R
where ϕ0− (x0 ) is the left hand derivative of φ at x0 . If f is not bounded, apply
what we have just proved to f M = f 1|f |≤M , to find
Plot of ψ1 0. Plot of ψ1 1.
EG ϕ(f M ) ≥ ϕ(EG f M ).

(23.28)
Since EG : L1 (Ω, F, P ; R) → L1 (Ω, F, P ; R) is a bounded operator and f M → f

∞
and ϕ(f M ) → φ(f ) in L1 (Ω, F, P ; R) as M → ∞, there exists {Mk }k=1 such
Mk Mk
that Mk ↑ ∞ and f → f and ϕ(f ) → φ(f ), P – a.e. So passing to the
limit in Eq. (23.28) shows EG [ϕ(f )] ≥ ϕ(EG f ), P – a.e.
23.6 Exercises Plot of ψ2 0. Plot of ψ2 1.
Exercise 23.9. Let (X, M, µ) be a measure space and H := L2 (X, M, µ).

Given f ∈ L∞ (µ) let Mf : H → H be the multiplication operator defined by
Mf g = f g. Show Mf2 = Mf iff there exists A ∈ M such that f = 1A a.e.
Exercise 23.10 (Haar Basis). In this problem, let L2 denote L2 ([0, 1], m)
with the standard inner product,
ψ(x) = 1[0,1/2) (x) − 1[1/2,1) (x) Plot of ψ2 2. Plot of ψ2 3.

and for k, j ∈ N0 := N∪{0} with 0 ≤ j < 2k let
1. Let M0 = span({1}) and for n ∈ N let
ψkj (x) = 2k/2 ψ(2k x − j)
Mn := span {1} ∪ ψkj : 0 ≤ k < n and 0 ≤ j < 2k ,

= 2k/2 12−k [j,j+1/2) (x) − 12−k [j+1/2,j+1) (x) .

where 1 denotes the constant function 1. Show
The following pictures shows the graphs of ψ0,0 , ψ1,0 , ψ1,1 , ψ2,1 , ψ2,2 and ψ2,3
Mn = span {1[j2−n ,(j+1)2−n ) : and 0 ≤ j < 2n .

respectively.

2. Show β := {1}∪ ψkj : 0 ≤ k and 0 ≤ j < 2k is anorthonormal set. Hint:
show ψk+1,j ∈ Mk⊥ for all 0 ≤ j < 2k+1 and show ψkj : 0 ≤ j < 2k is an
orthonormal set for fixed k.
3. Show ∪∞ 2
n=1 Mn is a dense subspace of L and therefore β is an orthonormal
2
basis for L . Hint: see Theorem 22.15.
4. For f ∈ L2 , let
Plot of ψ0 , 0. n−1
X 2X
k
−1
Hn f := hf |1i1 + hf |ψkj iψkj .
k=0 j=0

23.7 Fourier Series Exercises 259
Show (compare with Exercise 23.25) Exercise 23.18. Suppose f ∈ L1 ([−π, π]d ) is a function such that f˜ ∈ `1 (Zd )
n
and set X
−1 (j+1)2−n f˜(k)eik·x (pointwise).
2X
!
g(x) :=
Z
Hn f = 2n f (x)dx 1[j2−n ,(j+1)2−n ) k∈Zd
j=0 j2−n
1. Show g ∈ Cper (Rd ).
and use this to show kf − Hn f k∞ → 0 as n → ∞ for all f ∈ C([0, 1]). 2. Show g(x) = f (x) for m – a.e. x in [−π, π]d . Hint: Show g̃(k) = f˜(k) and
Hint: Compute orthogonal projection onto Mn using a judiciously chosen then use approximation arguments to show
basis for Mn . Z Z
f (x)h(x)dx = g(x)h(x)dx ∀ h ∈ C([−π, π]d )
Exercise 23.11. Let O(n) be the orthogonal groups consisting of n × n real [−π,π]d [−π,π]d
orthogonal matrices O, i.e. Otr O = I. For O ∈ O(n) and f ∈ L2 (Rn ) let and then refer to Lemma 22.11.
UO f (x) = f (O−1 x). Show 3. Conclude that f ∈ L1 ([−π, π]d ) ∩ L∞ ([−π, π]d ) and in particular f ∈
1. UO f is well defined, namely if f = g a.e. then UO f = UO g a.e. Lp ([−π, π]d ) for all p ∈ [1, ∞].
2. UO : L2 (Rn ) → L2 (Rn ) is unitary and satisfies UO1 UO2 = UO1 O2 for all Exercise 23.19. Suppose m ∈ N0 , α is a multi-index such that |α| ≤ 2m and
O1 , O2 ∈ O(n). That is to say the map O ∈ O(n) → U(L2 (Rn )) – the 2m
f ∈ Cper (Rd )2 .
unitary operators on L2 (Rn ) is a group homomorphism, i.e. a “unitary
representation” of O(n). 1. Using integration by parts, show (using Notation 22.21) that
3. For each f ∈ L2 (Rn ), the map O ∈ O(n) → UO f ∈ L2 (Rn ) is continuous. (ik)α f˜(k) = h∂ α f |ek i for all k ∈ Zd .
Take the topology on O(n) to be that inherited from the Euclidean topology
on the vector space of all n × n matrices. Hint: see the proof of Proposition Note: This equality implies
22.24.

˜
1 1
f (k) ≤ α k∂ α f kH ≤ α k∂ α f k∞ .
k k
Exercise 23.12. Euclidean group representation and its infinitesimal genera- Pd
tors including momentum and angular momentum operators. 2. Now let ∆f = i=1 ∂ 2 f /∂x2i , Working as in part 1) show
2
Exercise 23.13. Spherical Harmonics. h(1 − ∆)m f |ek i = (1 + |k| )m f˜(k). (23.29)
Remark 23.21. Suppose that m is an even integer, α is a multi-index and f ∈
Exercise 23.14. The gradient and the Laplacian in spherical coordinates. m+|α|
Cper (Rd ), then
Exercise 23.15. Legendre polynomials.  2  2
X X 2 2
|k α | f˜(k) =  |h∂ α f |ek i| (1 + |k| )m/2 (1 + |k| )−m/2 

k∈Zd k∈Zd
23.7 Fourier Series Exercises  2
X
2
P∞ m/2 α −m/2
Exercise 23.16. Show k=1 k −2 = π 2 /6, by taking f (x) = x on [−π, π] and = h(1 − ∆) ∂ f |ek i (1 + |k| )

2
computing kf k2 directly and then in terms of the Fourier Coefficients f˜ of f. k∈Zd
X 2 X
2
≤ h(1 − ∆)m/2 ∂ α f |ek i · (1 + |k| )−m

Exercise 23.17 (Riemann Lebesgue Lemma for Fourier Series). Show
for f ∈ L1 ([−π, π]d ) that f˜ ∈ c0 (Zd ), i.e. f˜ : Zd → C and limk→∞ f˜(k) = k∈Zd
2
k∈Zd
0. Hint: If f ∈ H, this follows form Bessel’s inequality. Now use a density = Cm (1 − ∆)m/2 ∂ α f

argument. H
2 2
We view Cper (R) as a subspace of H = L ([−π, π]) by identifying f ∈ Cper (R)
with f |[−π,π] ∈ H.

2 −d/2
X X
where Cm := k∈Zd (1 + |k| )−m < ∞ iff m > d/2. So the smoother f is the F̂ (k)eik·x for m – a.e. x
P
F (x + 2πk) = (2π) (23.30)
faster f˜ decays at infinity. The next problem is the converse of this assertion k∈Zd k∈Zd
and hence smoothness of f corresponds to decay of f˜ at infinity and visa-versa.
and form this conclude that f ∈ L1 ([−π, π]d ) ∩ L∞ ([−π, π]d ).
Exercise 23.20 (A Sobolev Imbedding Theorem). Suppose s ∈ R and Hint: see the hint for item 2. of Exercise 23.18.
4. Suppose we now assume that F ∈ C(Rd ) and F satisfies |F (x)| ≤ C(1 +

ck ∈ C : k ∈ Zd are coefficients such that
|x|)−s for some s > d and C < ∞. Under these added assumptions on F,
show Eq. (23.30) holds for all x ∈ Rd and in particular
X 2 2
|ck | (1 + |k| )s < ∞.
k∈Zd
−d/2
X X
F (2πk) = (2π) F̂ (k).
d
Show if s > 2 + m, the function f defined by k∈Zd k∈Zd
For notational simplicity, in the remaining problems we will assume that

X
f (x) = ck eik·x
d = 1.
k∈Zd
m Exercise 23.22 (Heat Equation 1.). Let (t, x) ∈ [0, ∞) × R → u(t, x) be a

is in Cper (Rd ). Hint: Work as in the above remark to show
continuous function such that u(t, ·) ∈ Cper (R) for all t ≥ 0, u̇ := ut , ux , and
X uxx exists and are continuous when t > 0. Further assume that u satisfies the
|ck | |k α | < ∞ for all |α| ≤ m.
heat equation u̇ = 12 uxx . Let ũ(t, k) := hu(t, ·)|ek i for k ∈ Z. Show for t > 0 and
k∈Zd d
k ∈ Z that ũ(t, k) is differentiable in t and dt ũ(t, k) = −k 2 ũ(t, k)/2. Use this
Exercise 23.21 (Poisson Summation Formula). Let F ∈ L1 (Rd ), result to show X t 2
  u(t, x) = e− 2 k f˜(k)eikx (23.31)
 X  k∈Z
E := x ∈ Rd : |F (x + 2πk)| = ∞ where f (x) := u(0, x) and as above
 d

k∈Z
Z π Z π
1
and set f˜(k) = hf |ek i = f (y)e−iky dy = f (y)e−iky dm (y) .
Z
−π 2π −π
−d/2
F̂ (k) := (2π) F (x)e−ik·x dx.
Rd Notice from Eq. (23.31) that (t, x) → u(t, x) is C ∞ for t > 0.
1 d
Further assume F̂ ∈ ` (Z ). 1
P − 2t k2 ikx
Exercise 23.23 (Heat Equation 2.). Let qt (x) := 2π k∈Z e e . Show
d
1. RShow m(E)P = 0 and E + 2πk = E for all k ∈ Z . Hint: Compute that Eq. (23.31) may be rewritten as
[−π,π]d k∈Zd |F (x + 2πk)| dx. Z π
2. Let P u(t, x) = qt (x − y)f (y)dy
f (x) := k∈Zd F (x + 2πk) for x ∈
/E −π
0 if x ∈ E.
and
−d/2
X
Show f ∈ L1 ([−π, π]d ) and f˜(k) = (2π) F̂ (k). qt (x) = pt (x + k2π)
3. Using item 2) and the assumptions on F, show k∈Z
1 2
X X −d/2 where pt (x) := √ 1 e− 2t x . Also show u(t, x) may be written as
f (x) = f˜(k)eik·x = (2π) F̂ (k)eik·x for m – a.e. x, 2πt
k∈Zd k∈Zd Z
u(t, x) = pt ∗ f (x) := pt (x − y)f (y)dy.
i.e. Rd

P
Hint: To show qt (x) = k∈Z pt (x + k2π), use the Poisson summation formula
and the Gaussian integration identity,
Z
1 1 t 2
p̂t (ω) = √ pt (x)eiωx dx = √ e− 2 ω . (23.32)
2π R 2π
Equation (23.32) will be discussed in Example 34.4 below.
Exercise 23.24 (Wave Equation). Let u ∈ C 2 (R × R) be such that u(t, ·) ∈
Cper (R) for all t ∈ R. Further assume that u solves the wave equation, utt = uxx .
Let f (x) := u(0, x) and g(x) = u̇(0, x). Show ũ(t, k) := hu(t, ·), ek i for k ∈ Z
d2 2
is twice continuously differentiable in t and dt 2 ũ(t, k) = −k ũ(t, k). Use this
result to show
X
sin kt ikx
u(t, x) = ˜
f (k) cos(kt) + g̃(k) e (23.33)
k
k∈Z
with the sum converging absolutely. Also show that u(t, x) may be written as
1 t
Z
1
u(t, x) = [f (x + t) + f (x − t)] + g(x + τ )dτ. (23.34)
2 2 −t
Hint: To show Eq. (23.33) implies (23.34) use
eikt + e−ikt
cos kt = ,
2
ikt −ikt
e −e
sin kt = , and
2i
t
eik(x+t) − eik(x−t)
Z
= eik(x+τ ) dτ.
ik −t
23.8 Conditional Expectation Exercises

∞
Exercise 23.25. Suppose (Ω, F, P ) is a probability
`∞ space and A := {Ai }i=1 ⊂
F is a partition of Ω. (Recall this means Ω = i=1 Ai .) Let G be the σ –
algebra generated by A. Show:
1. B ∈ G iff B = ∪i∈Λ Ai for some Λ ⊂ P N.
∞
2. g : Ω → R is G – measurable iff g = i=1 λi 1Ai for some λi ∈ R.
1
3. For f ∈ L (Ω, F, P ), let E(f |Ai ) := E [1Ai f ] /P (Ai ) if P (Ai ) 6= 0 and
E(f |Ai ) = 0 otherwise. Show
∞
X
EG f = E(f |Ai )1Ai .
i=1
24
Complex Measures, Radon-Nikodym Theorem and the Dual of Lp
Definition 24.1. A signed measure ν on a measurable space (X, M) is a Example 24.4. Suppose that µ is a positive measure on (X, M) and g∈ L1 (µ),
function ν : M → R such that then ν given as in Eq. (24.1) is a complex measure on (X, M). Also if µr± , µi±
is any collection of four positive finite measures on (X, M), then
1. Either
ν (M) := {ν (A) : A ∈ M} ⊂ (−∞, ∞] ν := µr+ − µr− + i µi+ − µi−

(24.2)
or ν(M) ⊂ [−∞, ∞). `∞ is a complex measure.
2. ν is countably additive, this is to say if E = j=1 Ej with Ej ∈ M, then
∞
If ν is given as in Eq. 24.1, then ν may be written as in Eq. (24.2) with
X dµr± = (Re g)± dµ and dµi± = (Im g)± dµ.
ν(E) = ν(Ej ).
j=1
∞
If ν(E) ∈ R then the series
P
ν(Ej ) is absolutely convergent since it is
24.1 The Radon-Nikodym Theorem
j=1
independent of rearrangements. Definition 24.5. Let ν be a complex or signed measure on (X, M). A set E ∈
3. ν(∅) = 0. M is a null set or precisely a ν – null set if ν(A) = 0 for all A ∈ M such that
A ⊂ E, i.e. ν|ME = 0. Recall that ME := {A ∩ E : A ∈ M} = i−1 E (M) is the
If there exists Xn ∈ M such that |ν(Xn )| < ∞ and X = ∪∞ n=1 Xn , then ν “trace of M on E.
is said to be σ – finite and if ν(M) ⊂ R then ν is said to be a finite signed
measure. Similarly, a countably additive set function ν : M → C such that We will eventually show that every complex and σ – finite signed measure
ν(∅) = 0 is called a complex measure. ν may be described as in Eq. (24.1). The next theorem is the first result in this
direction.
Example 24.2. Suppose that µ+ and µ− are two positive measures on M such
that either µ+ (X) < ∞ or µ− (X) < ∞, then ν = µ+ − µ− is a signed measure. Theorem 24.6 (A Baby Radon-Nikodym Theorem). Suppose (X, M) is
If both µ+ (X) and µ− (X) are finite then ν is a finite signed measure and may a measurable space, µ is a positive finite measure on M and ν is a complex
also be considered to be a complex measure. measure on M such that |ν(A)| ≤ µ(A) for all A ∈ M. Then dν = ρdµ where
|ρ| ≤ 1. Moreover if ν is a positive measure, then 0 ≤ ρ ≤ 1.
24.3. Suppose that g : X → R is measurable and either E g + dµ or
R
Example
R − P
E
g dµ < ∞, then Proof. For a simple function, f ∈ S(X, M), let ν(f ) := a∈C aν(f = a).
Z
Then
ν(A) = gdµ ∀ A ∈ M (24.1) X X Z
A |ν(f )| ≤ |a| |ν(f = a)| ≤ |a| µ(f = a) = |f | dµ.
a∈C a∈C X
defines a signed Rmeasure. This is actually a special case of the last example
with µ± (A) := A g ± dµ. Notice that the measure µ± in this example have So, by the B.L.T. Theorem 10.4, ν extends to a continuous linear functional on
the property that they are concentrated on disjoint sets, namely µ+ “lives” on L1 (µ) satisfying the bounds
{g > 0} and µ− “lives” on the set {g < 0} . Z p
|ν(f )| ≤ |f | dµ ≤ µ(X) kf kL2 (µ) for all f ∈ L1 (µ).
X
264 24 Complex Measures, Radon-Nikodym Theorem and the Dual of Lp
The Riesz representation Theorem 8.15 then implies there exists a unique ρ ∈ Definition 24.9 (Lebesgue Decomposition). Suppose that ν is a signed
L2 (µ) such that Z (complex) measure and µ is a positive measure on (X, M). Two signed (com-
plex) measures νa and νs form a Lebesgue decomposition of ν relative to µ
ν(f ) = f ρdµ for all f ∈ L2 (µ).
X if
Taking A ∈ M and f = sgn(ρ)1A in this equation shows 1. If ν (A) = ∞ (ν (A) = −∞) for some A ∈ M then νa (A) 6= −∞
(νa (A) 6= +∞) and νs (A) 6= −∞ (νs (A) 6= +∞) .
Z Z
|ρ| dµ = ν(sgn(ρ)1A ) ≤ µ(A) = 1dµ 2. ν = νa + νs which is well defined by assumption 1.
A A
3. νa µ and νs ⊥ µ.
from which it followsR that |ρ| ≤ 1, µ – a.e. If ν is a positive measure,R then for real
2
f, 0 = Im [ν(f )] = X Im ρf dµ and taking f = Im ρ shows 0 = X [Im ρ] dµ, Lemma 24.10. Let ν is a signed (complex) measure and µ is a positive measure
i.e. Im(ρ(x)) = 0 for µ – a.e. x and we have shown ρ is real a.e. Similarly, on (X, M). If there exists a Lebesgue decomposition, ν = νs +νa , of the measure
Z ν relative to µ then it is unique. Moreover:
0 ≤ ν(Re ρ < 0) = ρdµ ≤ 0,
{Re ρ<0} 1. if ν is positive then νs and νa are positive.
2. If ν is a σ – finite measure then so are νs and νa .
shows ρ ≥ 0 a.e.
Definition 24.7. Let µ and ν be two signed or complex measures on (X, M). Proof. Since νs ⊥ µ, there exists A ∈ M such that µ(A) = 0 and Ac is νs –
Then: null and because νa µ, A is also a null set for νa . So for C ∈ M, νa (C ∩A) = 0
and νs (C ∩ Ac ) = 0 from which it follows that
1. µ and ν are mutually singular (written as µ ⊥ ν) if there exists A ∈ M
such that A is a ν – null set and Ac is a µ – null set. ν(C) = ν(C ∩ A) + ν(C ∩ Ac ) = νs (C ∩ A) + νa (C ∩ Ac )
2. The measure ν is absolutely continuous relative to µ (written as ν
µ) provided ν(A) = 0 whenever A is a µ – null set, i.e. all µ – null sets are and hence,
ν – null sets as well.
νs (C) = νs (C ∩ A) = ν(C ∩ A) and
As an example, suppose that µ is a positive measure and ρ ∈ L1 (µ) . Then
the measure, ν := ρµ is absolutely continuous relative to µ. Indeed, if µ (A) = 0 νa (C) = νa (C ∩ Ac ) = ν(C ∩ Ac ). (24.3)
then Z Item 1. is now obvious from Eq. (24.3).
ρ (A) = ρdµ = 0 For Item 2., if ν is a σ – finite measure then there exists Xn ∈ M such that
A
X = ∪∞ n=1 Xn and |ν(Xn )| < ∞ for all n. Since ν(Xn ) = νa (Xn ) + νs (Xn ), we
as well.
must have νa (Xn ) ∈ R and νs (Xn ) ∈ R showing νa and νs are σ – finite as well.
Lemma 24.8. If µ1 , µ2 and ν are signed measures on (X, M) such that µ1 ⊥ ν For the uniqueness assertion, if we have another decomposition ν = ν̃a + ν̃s
∞
and µ2 ⊥ ν and µ1 + µ2 is well defined, then (µ1 + µ2 ) ⊥ ν. If P{µi }i=1 is a with ν̃s ⊥ µ and ν̃a µ we may choose Ã ∈ M such that µ(Ã) = 0 and Ãc is
∞
sequence of positive measures such that µi ⊥ ν for all i then µ = i=1 µi ⊥ ν ν̃s – null. Then B = A ∪ Ã is still a µ - null set and B c = Ac ∩ Ãc is a null set
as well. for both νs and ν̃s . Therefore by the same arguments which proved Eq. (24.3),
Proof. In both cases, choose Ai ∈ M such that Ai is ν – null and Aci is νs (C) = ν(C ∩ B) = ν̃s (C) and
µi -null for all i. Then by Lemma 24.16, A := ∪i Ai is still a ν –null set. Since
νa (C) = ν(C ∩ B c ) = ν̃a (C) for all C ∈ M.
Ac = ∩i Aci ⊂ Acm for all m
P∞
we see that Ac is a µi - null set for all i and is therefore a null set for µ = i=1 µi .
This shows that µ ⊥ ν. Lemma 24.11. Suppose µ is a positive measure on (X, M) and f, g : X → R̄
Throughout the remainder of this section µ will be always be a positive are extended integrable functions such that
measure on (X, M) .

24.1 The Radon-Nikodym Theorem 265
Z Z
f dµ = gdµ for all A ∈ M, (24.4) and then take f = g1{h<1} (1 − h)−1 with g ≥ 0 in Eq. (24.6) to learn
A A
R R ν(g1{h<1} ) = µ(g1{h<1} (1 − h)−1 h) = µ(ρg).
f dµ < ∞, X g− dµ < ∞, and the measures |f | dµ and |g| dµ are σ –
X −
finite. Then f (x) = g(x) for µ – a.e. x. Hence if we define
νa := 1{h<1} ν and νs := 1{h=1} ν,
Proof. By assumption there exists Xn ∈ M such that Xn ↑ X and
we then have νs ⊥ µ (since νs “lives” on {h = 1} while µ (h = 1) = 0) and
R R
Xn
|f | dµ < ∞ and Xn
|g| dµ < ∞ for all n. Replacing A by A ∩ Xn in Eq.
(24.4) implies νa = ρµ and in particular νa µ. Hence ν = νa + νs is the desired Lebesgue
Z Z Z Z decomposition of ν.1
If we further assume that ν µ, then µ (h = 1) = 0 implies ν (h = 1) = 0
1Xn f dµ = f dµ = gdµ = 1Xn gdµ
A A∩Xn A∩Xn A and hence that νs = 0 and we conclude ` that ν = νa = ρµ.
∞
For the σ – finite case, write X = n=1 Xn where Xn ∈ M are chosen so
for all A ∈ M. Since 1Xn f and 1Xn g are in L1 (µ) for all n, this equation implies that µ(Xn ) < ∞ and ν(Xn ) < ∞ for all n. Let dµn = 1Xn dµ and dνn = 1Xn dν.
1Xn f = 1Xn g, µ – a.e. Letting n → ∞ then shows that f = g, µ – a.e. Then by what we have just proved there exists ρn ∈ L1 (X, µn ) ⊂ L1 (X, µ) and
measure νns such that dνn = ρn dµn + dνns with νns ⊥ µn . Since µn and νns “live”
Remark 24.12. Suppose that f and g are two positive measurable functions on on Xn (see Eq. (24.3) there exists An ∈ MXn such that µ (An ) = µn (An ) = 0
(X, M, µ) such that Eq. (28.32) holds. It is not in general true that f = g, µ – and
a.e. A trivial counterexample is to take M = 2X , µ(A) = ∞ for all non-empty νns (X \ An ) = νns (Xn \ An ) = 0.
A ∈ M, f = 1X and g = 2 · 1X . Then Eq. (24.4) holds yet f 6= g. P∞
This shows that νns ⊥ µ for all n and so by Lemma 24.8, νs := n=1 νns is
Theorem 24.13 (Radon Nikodym Theorem for Positive Measures). singular relative to µ. Since
Suppose that µ and ν are σ – finite positive measures on (X, M). Then ν has
∞ ∞ ∞
a unique Lebesgue decomposition ν = νa + νs relative to µ and there exists X X X
ν= νn = (ρn µn + νns ) = (ρn 1Xn µ + νns ) = ρµ + νs ,
a unique (modulo sets of µ – measure 0) function ρ : X → [0, ∞) such that
n=1 n=1 n=1
dνa = ρdµ. Moreover, νs = 0 iff ν µ.
P∞
where ρ := n=1 1Xn ρn , it follows that ν = νa + νs with νa = ρµ is the
Proof. The uniqueness assertions follow directly from Lemmas 24.10 and
Lebesgue decomposition of ν relative to µ.
24.11.
1
Existence. (Von-Neumann’s Proof.) First suppose that µ and ν are finite Here is the motivation for this construction.
` Suppose that dν = dνs + ρdµ is the
measures and let λ = µ + ν. By Theorem 24.6, dν = hdλ with 0 ≤ h ≤ 1 and Radon-Nikodym decompostion and X = A B such that νs (B) = 0 and µ(A) = 0.
this implies, for all non-negative measurable functions f, that Then we find
νs (f ) + µ(ρf ) = ν(f ) = λ(f g) = ν(f g) + µ(f g).

ν(f ) = λ(f h) = µ(f h) + ν(f h) (24.5)
Letting f → 1A f then implies that
or equivalently
ν(f (1 − h)) = µ(f h). (24.6) νs (1A f ) = ν(1A f g)
Taking f = 1{h=1} in Eq. (24.6) shows that which show that g = 1 ν –a.e. on A. Also letting f → 1B f implies that
µ ({h = 1}) = ν(1{h=1} (1 − h)) = 0, µ(ρ1B f (1 − g)) = ν(1B f (1 − g)) = µ(1B f g) = µ(f g)
i.e. 0 ≤ h (x) < 1 for µ - a.e. x. Let which shows that

ρ(1 − g) = ρ1B (1 − g) = g µ − a.e..
h This shows that ρ = g
µ – a.e.
ρ := 1{h<1} 1−g
1−h

Theorem 24.14 (Dual of Lp – spaces). Let (X, M, µ) be a σ – finite measure for all simple functions f on X. Replacing f by 1{|g|≤M } f in Eq. (24.10) shows
space and suppose that p, q ∈ [1, ∞] are conjugate exponents. Then for p ∈ Z
[1, ∞), the map g ∈ Lq → φg ∈ (Lp )∗ (where φg = h·, giµ was defined in Eq. φ(f 1{|g|≤M } ) = 1{|g|≤M } gf dµ
21.23) is an isometric isomorphism of Banach spaces. We summarize this by X
writing (Lp )∗ = Lq for all 1 ≤ p < ∞. (The result is in general false for p = 1
as can be seen from Theorem 25.13 and Lemma 25.14 below.) holds for all simple functions f and then by continuity for all f ∈ Lp (µ). By
the converse to Holder’s inequality, (Proposition 21.26) we learn that
Proof. The only results of this theorem which are not covered in Proposition
21.26 is the surjectivity of the map g ∈ Lq → φg ∈ (Lp )∗ . When p = 2, this 1{|g|≤M } g = sup φ(f 1{|g|≤M } )
q
kf kp =1
surjectivity is a direct consequence of the Riesz Theorem 8.15.
Case 1. We will begin the proof under the extra assumption that µ(X) < ∞ ≤ sup kφk(Lp )∗ f 1{|g|≤M } p ≤ kφk(Lp )∗ .
∗ kf kp =1
in which cased bounded functions are in Lp (µ) for all p. So let φ ∈ (Lp ) . We
need to find g ∈ Lq (µ) such that φ = φg . When p ∈ [1, 2], L2 (µ) ⊂ Lp (µ) so that
Using the monotone convergence theorem we may let M → ∞ in the previous
we may restrict φ to L2 (µ) and again the result follows fairly easily from the
equation to learn kgkq ≤ kφk(Lp )∗ .With this result, Eq. (24.10) extends by
Riesz Theorem, see Exercise 24.3 below. To handle general p ∈ [1, ∞), define
ν(A) := φ(1A ). If A = ∞
` continuity to hold for all f ∈ Lp (µ) and hence we have shown that φ = φg .
n=1 n with An ∈ M, then
A
Case 2. Now suppose that µ is σ – finite and Xn ∈ M are sets such that
N
X p1 µ(Xn ) < ∞ and Xn ↑ X as n → ∞. We will identify f ∈ Lp (Xn , µ) with
k p = µ(∪∞

k1A − 1An kLp = k1∪∞
n=N +1 An L n=N +1 An ) → 0 as N → ∞. f 1Xn ∈ Lp (X, µ) and this way we may consider Lp (Xn , µ) as a subspace of
n=1 Lp (X, µ) for all n and p ∈ [1, ∞]. By Case 1. there exists gn ∈ Lq (Xn , µ) such
Therefore that Z
∞ ∞
X X φ(f ) = gn f dµ for all f ∈ Lp (Xn , µ)
ν(A) = φ(1A ) = φ(1An ) = ν(An ) Xn
1 1
and
showing ν is a complex measure.2 For A ∈ M, let |ν| (A) be the “total variation”
kgn kq = sup |φ(f )| : f ∈ Lp (Xn , µ) and kf kLp (Xn ,µ) = 1 ≤ kφk[Lp (µ)]∗ .

of A defined by
|ν| (A) := sup {|φ(f 1A )| : |f | ≤ 1} (24.7)
It is easy to see that gn = gm a.e. on Xn ∩Xm for all m, n so that g := limn→∞ gn
and notice that exists µ – a.e. By the Rabove inequality and Fatou’s lemma, kgkq ≤ kφk[Lp (µ)]∗ <
|ν(A)| ≤ |ν| (A) ≤ kφk(Lp )∗ µ(A)1/p for all A ∈ M. (24.8) ∞ and since φ(f ) = Xn gf dµ for all f ∈ Lp (Xn , µ) and n and ∪∞ p
n=1 L (Xn , µ)
p
R
is dense in L (X, µ) it follows by continuity that φ(f ) = X gf dµ for all f ∈
You are asked to show in Exercise 24.4 that |ν| is a measure on (X, M). (This Lp (X, µ),i.e. φ = φg .
can also be deduced from Lemma 24.29 and Proposition 24.33 below.) By Eq.
(24.8) |ν| µ, by Theorem 24.6 dν = hd |ν| for some |h| ≤ 1 and by Theorem
24.13 d |ν| = ρdµ for some ρ ∈ L1 (µ). Hence, letting g = ρh ∈ L1 (µ), dν = gdµ 24.2 The Structure of Signed Measures
or equivalently Z
φ(1A ) = g1A dµ ∀ A ∈ M. (24.9) Definition 24.15. Let ν be a signed measure on (X, M) and E ∈ M, then
X
1. E is positive if for all A ∈ M such that A ⊂ E, ν(A) ≥ 0, i.e. ν|ME ≥ 0.
By linearity this equation implies
2. E is negative if for all A ∈ M such that A ⊂ E, ν(A) ≤ 0, i.e. ν|ME ≤ 0.
Z
φ(f ) = gf dµ (24.10) Lemma 24.16. Suppose that ν is a signed measure on (X, M). Then
X
2
It is at this point that the proof breaks down when p = ∞. 1. Any subset of a positive set is positive.

24.2 The Structure of Signed Measures 267
2. The countable union of positive (negative or null) sets is still positive (neg- 24.2.1 Hahn Decomposition Theorem
ative or null).
3. Let us now further assume that ν(M) ⊂ [−∞, ∞) and E ∈ M is a set Definition 24.17. Suppose that ν is a signed measure on (X, M). A Hahn
such that ν (E) ∈ (0, ∞). Then there exists a positive set P ⊂ E such that decomposition for ν is a partition {P, N = P c } of X such that P is positive
ν(P ) ≥ ν(E). and N is negative.
Proof. The first assertion is obvious. If Pj ∈ M are positive sets, let P = Theorem 24.18 (Hahn Decomposition Theorem). Every signed measure
∞ n−1
! space (X, M, ν) has a Hahn decomposition, {P, N }. Moreover, if {P̃ , Ñ } is
S S
Pn . By replacing Pn by the positive set Pn \ Pj we may assume another Hahn decomposition, then P ∆P̃ = N ∆Ñ is a null set, so the decom-
n=1 j=1 position is unique modulo null sets.
∞
∞ `
that the {Pn }n=1 are pairwise disjoint so that P = Pn . Now if E ⊂ P and Proof. With out loss of generality we may assume that ν(M) ⊂ [−∞, ∞).
n=1
∞
` P∞ If not just consider −ν instead.
E ∈ M, E = (E ∩ Pn ) so ν(E) = n=1 ν(E ∩ Pn ) ≥ 0.which shows that P Uniqueness. For any A ∈ M, we have
n=1
is positive. The proof for the negative and the null case is analogous.
The idea for proving the third assertion is to keep removing “big” sets of ν(A) = ν(A ∩ P ) + ν(A ∩ N ) ≤ ν(A ∩ P ) ≤ ν(P ).
negative measure from E. The set remaining from this procedure will be P. We
In particular, taking A = P ∪ P̃ , we learn
now proceed to the formal proof. For all A ∈ M let
ν(P ) ≤ ν(P ∪ P̃ ) ≤ ν(P )
n(A) = 1 ∧ sup{−ν(B) : B ⊂ A}.

Since ν(∅) = 0, n(A) ≥ 0 and n(A) = 0 iff A is positive. Choose A0 ⊂ E or equivalently that ν (P ) = ν P ∪ P̃ . Of course by symmetry we also have
such that −ν(A0 ) ≥ 12 n(E) and set E1 = E \ A0 , then choose A1 ⊂ E1 such
that −ν(A1 ) ≥ 12 n(E1 ) and set E2 = E \ (A0 ∪ A1 ) . Continue this procedure ν (P ) = ν P ∪ P̃ = ν P̃ =: s.
k−1
`
inductively, namely if A0 , . . . , Ak−1 have been chosen let Ek = E \ Ai and Since also,
i=0
∞ ∞
choose Ak ⊂ Ek such that −ν(Ak ) ≥ 21 n(Ek ). Let P := E \
` T
Ak = Ek , s = ν(P ∪ P̃ ) = ν(P ) + ν(P̃ ) − ν(P ∩ P̃ ) = 2s − ν(P ∩ P̃ ),
k=0 k=0
∞
` h i`
then E = P ∪ Ak and hence we also have ν(P ∩ P̃ ) = s. Finally using P ∪ P̃ = P ∩ P̃ P̃ ∆P , we
k=0
conclude that
∞
X ∞
X
(0, ∞) 3 ν(E) = ν(P ) + ν(Ak ) = ν(P ) − −ν(Ak ) ≤ ν(P ). (24.11) s = ν(P ∪ P̃ ) = ν(P ∩ P̃ ) + ν(P̃ ∆P ) = s + ν(P̃ ∆P )
k=0 k=0
P∞ which shows ν(P̃ ∆P ) = 0. Thus N ∆Ñ = P̃ ∆P is a positive set with zero mea-
From Eq. (24.11) we learn that k=0 −ν(Ak ) < ∞ and in particular that sure, i.e. N ∆Ñ = P̃ ∆P is a null set and this proves the uniqueness assertion.
limk→∞ (−ν(Ak )) = 0. Since 0 ≤ 12 n(Ek ) ≤ −ν(Ak ), this also implies Existence. Let
limk→∞ n(Ek ) = 0. If A ∈ M with A ⊂ P, then A ⊂ Ek for all k and so, s := sup{ν(A) : A ∈ M}
for k large so that n(Ek ) < 1, we find −ν(A) ≤ n(Ek ). Letting k → ∞ in
which is non-negative since ν(∅) = 0. If s = 0, we are done since P = ∅ and
this estimate shows −ν(A) ≤ 0 or equivalently ν(A) ≥ 0. Since A ⊂ P was
N = X is the desired decomposition. So assume s > 0 and choose An ∈ M
arbitrary, we conclude that P is a positive set such that ν(P ) ≥ ν(E).
such that ν(An ) > 0 and limn→∞ ν(An ) = s. By Lemma 24.16 there exists
positive sets Pn ⊂ An such that ν(Pn ) ≥ ν(An ). Then s ≥ ν(Pn ) ≥ ν(An ) → s
as n → ∞ implies that s = limn→∞ ν(Pn ). The set P := ∪∞ n=1 Pn is a positive
set being the union of positive sets and since Pn ⊂ P for all n,

ν(P ) ≥ ν(Pn ) → s as n → ∞. Therefore ν(A) ∈ R iff ν(A ∩ P ) ∈ R and ν(A ∩ N ) ∈ R iff |ν| (A) < ∞. Finally,
This shows that ν(P ) ≥ s and hence by the definition of s, s = ν(P ) < ∞. ν(A) = ν(A ∩ P ) + ν(A ∩ N )
I now claim that N = P c is a negative set and therefore, {P, N } is the desired = |ν| (A ∩ P ) − |ν| (A ∩ N )
Hahn decomposition. If N were not negative, we could find E ⊂ N = P c such Z
that ν(E) > 0. We then would have = (1P − 1N )d |ν|
A
ν(P ∪ E) = ν(P ) + ν(E) = s + ν(E) > s
which shows that dν = gd |ν| .
which contradicts the definition of s. Lemma 24.22. Suppose that µ is a positive measure on (X, M) and g : X → R
is an extended µ-integrable function. If ν is the signed measure dν = gdµ, then
24.2.2 Jordan Decomposition dν± = g± dµ and d |ν| = |g| dµ. We also have
Z
Theorem 24.19 (Jordan Decomposition). If ν is a signed measure on |ν| (A) = sup{ f dν : |f | ≤ 1} for all A ∈ M. (24.13)
(X, M) , there exist unique positive measure ν± on (X, M) such that ν+ ⊥ ν− A
and ν = ν+ − ν− . This decomposition is called the Jordan decomposition of
ν. Proof. The pair, P = {g > 0} and N = {g ≤ 0} = P c is a Hahn decompo-
sition for ν. Therefore
Proof. Let {P, N } be a Hahn decomposition for ν and define Z Z Z
ν+ (A) = ν(A ∩ P ) = gdµ = 1{g>0} gdµ = g+ dµ,
ν+ (E) := ν(P ∩ E) and ν− (E) := −ν(N ∩ E) ∀ E ∈ M. A∩P A A
Z Z Z
Then it is easily verified that ν = ν+ − ν− is a Jordan decomposition of ν. The ν− (A) = −ν(A ∩ N ) = − gdµ = − 1{g≤0} gdµ = − g− dµ.
reader is asked to prove the uniqueness of this decomposition in Exercise 24.10. A∩N A A
and
Z Z
Definition 24.20. The measure, |ν| := ν+ + ν− is called the total variation |ν| (A) = ν+ (A) + ν− (A) = g+ dµ − g− dµ
of ν. A signed measure is called σ – finite provided that ν± (or equivalently
ZA A
|ν| := ν+ + ν− ) are σ -finite measures.
Z
= (g+ − g− ) dµ = |g| dµ.
A A
Lemma 24.21. Let ν be a signed measure on (X, M) and A ∈ M. If ν(A) ∈
R then ν(B) ∈ R for all B ⊂ A. Moreover, ν(A) ∈ R iff |ν| (A) < ∞. In If A ∈ M and |f | ≤ 1, then
particular, ν is σ finite iff |ν| is σ – finite. Furthermore if P, N ∈ M is a Hahn Z Z Z Z Z
decomposition for ν and g = 1P − 1N , then dν = gd |ν| , i.e.
f dν = f dν+ −

f dν− ≤ f dν+ + f dν−

A
ZA ZA ZA A
Z
ν(A) = gd |ν| for all A ∈ M.
A ≤ |f | dν+ + |f | dν− = |f | d |ν| ≤ |ν| (A).
A A A
Proof. Suppose that B ⊂ A and |ν(B)| = ∞ then since ν(A) = ν(B) +
For the reverse inequality, let f := 1P − 1N then
ν(A \ B) we must have |ν(A)| = ∞. Let P, N ∈ M be a Hahn decomposition
for ν, then
Z
f dν = ν(A ∩ P ) − ν(A ∩ N ) = ν+ (A) + ν− (A) = |ν| (A).
A
ν(A) = ν(A ∩ P ) + ν(A ∩ N ) = |ν(A ∩ P )| − |ν(A ∩ N )| and
|ν| (A) = ν(A ∩ P ) − ν(A ∩ N ) = |ν(A ∩ P )| + |ν(A ∩ N )| . (24.12)

24.2 The Structure of Signed Measures 269
Definition 24.23. Let ν be a signed measure on (X, M), let |ν| (C) = ν(C ∩ P ) − ν(C ∩ N )
= νs (C ∩ P ) − νs (C ∩ N ) + νa (C ∩ P ) − νa (C ∩ N )
L1 (ν) := L1 (ν+ ) ∩ L1 (ν− ) = L1 (|ν|)
= |νs | (C) + |νa | (C).
and for f ∈ L1 (ν) we define
Z Z Z
f dν := f dν+ − f dν− . Lemma 24.26.
X X X
1. Let ν be a signed measure and µ be a positive measure on (X, M) such that
Lemma 24.24. Let µ be a positive measure on (X, M), g be an extended inte- ν µ and ν ⊥ µ, P then ν ≡ 0.
grable function on (X, M, µ) and dν = gdµ. Then L1 (ν) = L1 (|g| dµ) and for ∞
2. Suppose that ν = i=1 νi where νi are positive measures on (X, M) such
f ∈ L1 (ν), Z Z that νi µ, then ν µ.
f dν = f gdµ. 3. Also if ν1 and ν2 are two signed measure such that νi µ for i = 1, 2 and
X X ν = ν1 + ν2 is well defined, then ν µ.
Proof. By Lemma 24.22, dν+ = g+ dµ, dν− = g− dµ, and d |ν| = |g| dµ so
Proof. 1. Because ν ⊥ µ, there exists A ∈ M such that A is a ν – null set
that L1 (ν) = L1 (|ν|) = L1 (|g| dµ) and for f ∈ L1 (ν),
and B = Ac is a µ - null set. Since B is µ – null and ν µ, B is also ν – null.
Z Z Z Z Z This shows by Lemma 24.16 that X = A ∪ B is also ν – null, i.e. ν is the zero
f dν = f dν+ − f dν− = f g+ dµ − f g− dµ measure. The proof of items 2. and 3. are easy and will be left to the reader.
X
ZX X
Z X X
= f (g+ − g− ) dµ = f gdµ. Theorem 24.27 (Radon Nikodym Theorem for Signed Measures). Let
X X ν be a σ – finite signed measure and µ be a σ – finite positive measure on
(X, M). Then ν has a unique Lebesgue decomposition ν = νa + νs relative to
µ and there exists a unique (modulo sets of µ – measure 0) extended integrable
Lemma 24.25. Suppose ν is a signed measure, µ is a positive measure and function ρ : X → R such that dνa = ρdµ. Moreover, νs = 0 iff ν µ, i.e.
ν = νa + νs is a Lebesgue decomposition (see Definition 24.9) of ν relative to dν = ρdµ iff ν µ.
µ, then |ν| = |νa | + |νs | .
Proof. Uniqueness. Is a direct consequence of Lemmas 24.10 and 24.11.
Proof. Let A ∈ ` M be chosen so that A is a null set for νa and Ac is a null Existence. Let ν = ν+ −ν− be the Jordan decomposition of ν. Assume, without
`set
for νs . Let A = P 0
N 0 be a Hahn decomposition of νs |MA and Ac = P̃ Ñ loss of generality, that ν+ (X) < ∞, i.e. ν(A) < ∞ for all A ∈ M. By the Radon
be a Hahn decomposition of νa |MAc . Let P = P 0 ∪ P̃ and N = N 0 ∪ Ñ . Since Nikodym Theorem 24.13 for positive measures there exist functions f± : X →
for C ∈ M, [0, ∞) and measures λ± such that ν± = µf± + λ± with λ± ⊥ µ. Since
ν(C ∩ P ) = ν(C ∩ P 0 ) + ν(C ∩ P̃ ) ∞ > ν+ (X) = µf+ (X) + λ+ (X),

= νs (C ∩ P 0 ) + νa (C ∩ P̃ ) ≥ 0 f+ ∈ L1 (µ) and λ+ (X) < ∞ so that f = f+ − f− is an extended integrable
function, dνa := f dµ and νs = λ+ − λ− are signed measures. This finishes the
and existence proof since
ν(C ∩ N ) = ν(C ∩ N 0 ) + ν(C ∩ Ñ )
ν = ν+ − ν− = µf+ + λ+ − µf− + λ− = νa + νs
= νs (C ∩ N 0 ) + νa (C ∩ Ñ ) ≤ 0
and νs = (λ+ − λ− ) ⊥ µ by Lemma 24.8. For the final statement, if νs = 0, then
we see that {P, N } is a Hahn decomposition for ν. It also easy to see that {P, N } dν = ρdµ and hence ν µ. Conversely if ν µ, then dνs = dν − ρdµ µ, so
is a Hahn decomposition for both νs and νa as well. Therefore, by Lemma 24.16, νs = 0. Alternatively just use the uniqueness of the Lebesgue
decomposition to conclude νa = ν and νs = 0. Or more directly, choose B ∈ M

such that µ(B c ) = 0 and B is a νs – null set. Since ν µ, B c is also a ν – null Definition 24.30. Given a complex measure ν, let νr = Re ν and νi = Im ν so
set so that, for A ∈ M, that νr and νi are finite signed measures such that
ν(A) = ν(A ∩ B) = νa (A ∩ B) + νs (A ∩ B) = νa (A ∩ B). ν(A) = νr (A) + iνi (A) for all A ∈ M.
Let L1 (ν) := L1 (νr ) ∩ L1 (νi ) and for f ∈ L1 (ν) define

Notation 24.28 The function f is called the Radon-Nikodym derivative of ν Z Z Z
dν
relative to µ and we will denote this function by dµ . f dν := f dνr + i f dνi .
X X X
ExampleR 24.31. Suppose that µ is a positive measure on (X, M), g ∈ L1 (µ) and
24.3 Complex Measures ν(A) = A gdµ as in Example 24.4, then L1 (ν) = L1 (|g| dµ) and for f ∈ L1 (ν)
Suppose that ν is a complex measure on (X, M), let νr := Re ν, νi := Im ν and Z Z
µ := |νr | + |νi |. Then µ is a finite positive measure on M such that νr µ f dν = f gdµ. (24.14)
X X
and νi µ. By the Radon-Nikodym Theorem 24.27, there exists real functions
h, k ∈ L1 (µ) such that dνr = h dµ and dνi = k dµ. So letting g := h+ik ∈ L1 (µ), To check Eq. (24.14), notice that dνr = Re g dµ and dνi = Im g dµ so that
(using Lemma 24.24)
dν = (h + ik)dµ = gdµ
showing every complex measure may be written as in Eq. (24.1). L1 (ν) = L1 (Re gdµ) ∩ L1 (Im gdµ) = L1 (|Re g| dµ) ∩ L1 (|Im g| dµ) = L1 (|g| dµ).
Lemma 24.29. Suppose that ν is a complex measure on (X, M), and for i = If f ∈ L1 (ν), then
1, 2 let µi be a finite positive measure on (X, M) such that dν = gi dµi with Z Z Z Z
gi ∈ L1 (µi ). Then f dν := f Re gdµ + i f Im gdµ = f gdµ.
Z Z X X X X
|g1 | dµ1 = |g2 | dµ2 for all A ∈ M. Remark 24.32. Suppose that ν is a complex measure on (X, M) such that dν =
A A
gdµ and as above d |ν| = |g| dµ. Letting
In particular, we may define a positive measure |ν| on (X, M) by
g
Z if |g| =
6 0
ρ = sgn(ρ) := |g|
|ν| (A) = |g1 | dµ1 for all A ∈ M. 1 if |g| = 0
A
The finite positive measure |ν| is called the total variation measure of ν. we see that
dν = gdµ = ρ |g| dµ = ρd |ν|
Proof. Let λ = µ1 + µ2 so that µi λ. Let ρi = dµi /dλ ≥ 0 and hi = ρi gi .
Since and |ρ| = 1 and ρ is uniquely defined modulo |ν| – null sets. We will denote ρ
by dν/d |ν| . With this notation, it follows from Example 24.31 that L1 (ν) :=
Z Z Z
ν(A) = gi dµi = gi ρi dλ = hi dλ for all A ∈ M,
A A A L1 (|ν|) and for f ∈ L1 (ν),
h1 = h2 , λ –a.e. Therefore Z Z
dν
Z Z Z f dν = f d |ν| .
X X d |ν|
|g1 | dµ1 = |g1 | ρ1 dλ = |h1 | dλ
A
ZA Z A Z We now give a number of methods for computing the total variation, |ν| , of a
= |h2 | dλ = |g2 | ρ2 dλ = |g2 | dµ2 . complex or signed measure ν.
A A A

24.3 Complex Measures 271
Proposition 24.33 (Total Variation). Suppose A ⊂ 2X is an algebra, M = This is evident from Figure 24.1 and formally follows from the fact that
σ(A), ν is a complex (or a signed measure which is σ – finite on A) on (X, M) 2
d −1
h
−1
i
and for E ∈ M let ρ(x) − t |zk | zk = 2 t − Re(|zk | zk ρ(x)) ≥ 0

( n ) dt
X when t ≥ 1. Therefore if we define
µ0 (E) = sup |ν(Ej )| : Ej ∈ AE 3 Ei ∩ Ej = δij Ei , n = 1, 2, . . .
1
( n )
X
µ1 (E) = sup |ν(Ej )| : Ej ∈ ME 3 Ei ∩ Ej = δij Ei , n = 1, 2, . . .
1
(∞ )
X
µ2 (E) = sup |ν(Ej )| : Ej ∈ ME 3 Ei ∩ Ej = δij Ei
1
Z

µ3 (E) = sup f dν : f is measurable with |f | ≤ 1

ZE

µ4 (E) = sup f dν : f ∈ Sf (A, |ν|) with |f | ≤ 1 .

E
Fig. 24.1. Sliding points to the unit circle.
then µ0 = µ1 = µ2 = µ3 = µ4 = |ν| .
Proof. Let ρ = dν/d |ν| and recall that |ρ| = 1, |ν| – a.e.
Step 1. µ4 ≤ |ν| = µ3 . If f is measurable with |f | ≤ 1 then −1
|zk | zk if |zk | > 1
wk :=
Z

Z

Z

Z zk if |zk | ≤ 1
f dν = f ρd |ν| ≤ |f | d |ν| ≤ 1d |ν| = |ν| (E)

E E E E N
P
and ρ̃n = wk 1Ak then
from which we conclude that µ4 ≤ µ3 ≤ |ν| . Taking f = ρ̄ above shows k=1
Z

Z

Z |ρ(x) − ρn (x)| ≥ |ρ(x) − ρ̃n (x)|
f dν = ρ̄ ρ d |ν| = 1 d |ν| = |ν| (E)
and therefore ρ̃n → ρ1Xm in L1 (|ν|). So we now assume that ρn is as in Eq.

E E E
(24.15) with |zk | ≤ 1. Now
which shows that |ν| ≤ µ3 and hence |ν| = µ3 .
Step 2. µ4 ≥ |ν| . Let Xm ∈ A be chosen so that |ν| (Xm ) < ∞ and
Z Z Z

ρ̄n dν − ρ̄1Xm dν ≤ (ρ̄n dν − ρ̄1Xm ) ρd |ν|
Xm ↑ X as m → ∞. By Theorem 22.15 (or Remark 28.7 or Corollary 32.42

E E
below), there exists ρn ∈ Sf (A, µ) such that ρn → ρ1Xm in L1 (|ν|) and each ρn ZE
may be written in the form ≤ |ρ̄n − ρ̄1Xm | d |ν| → 0 as n → ∞
E
N
ρn =
X
z k 1 Ak (24.15) and hence Z

k=1 µ4 (E) ≥ ρ̄1Xm dν = |ν| (E ∩ Xm ) for all m.
E
where zk ∈ C and Ak ∈ A and Ak ∩ Aj = ∅ if k 6= j. I claim that we may Letting m ↑ ∞ in this equation shows µ4 ≥ |ν| which combined with step 1.
assume that |zk | ≤ 1 in Eq. (24.15) for if |zk | > 1 and x ∈ Ak , shows µ3 = µ4 = |ν| .
∞
Step 3. µ0 = µ1 = µ2 = |ν| . Clearly µ0 ≤ µ1 ≤ µ2 . Suppose {Ej }j=1 ⊂
−1
|ρ(x) − zk | ≥ ρ(x) − |zk | zk . ME be a collection of pairwise disjoint sets, then


∞ ∞ Z ∞
X X X Proof. In all cases we have |ν(A)| ≤ |ν| (A) for all A ∈ M which clearly
|ν(Ej )| = ρd |ν| ≤ |ν| (Ej ) = |ν| (∪Ej ) ≤ |ν| (E)
Ej
shows that |ν| (A) = 0 implies A is a ν – null set. Conversely if A is a ν – null
j=1 j=1 j=1
set, then, by definition, ν|MA ≡ 0 so by Proposition 24.33
which shows that µ2 ≤ |ν| = µ4 . So it suffices to show µ4 ≤ µ0 . But if f ∈ (∞ )
PN
Sf (A, |ν|) with |f | ≤ 1, then f may be expressed as f = k=1 zk 1Ak with |ν| (A) = sup
X
|ν(Ej )| : Ej ∈ MA 3 Ei ∩ Ej = δij Ei = 0.
|zk | ≤ 1 and Ak ∩ Aj = δij Ak. Therefore, 1
Z N N
since Ej ⊂ A implies µ(Ej ) = 0 and hence ν(Ej ) = 0.
X X
f dν = z ν(A ∩ E) ≤ |zk | |ν(Ak ∩ E)|

k k
Alternate Proofs that A is ν – null implies |ν| (A) = 0.

E k=1

k=1
1) Suppose ν is a signed measure and {P, N = P c } ⊂ M is a Hahn decom-
N
X position for ν. Then
≤ |ν(Ak ∩ E)| ≤ µ0 (A).
k=1
|ν| (A) = ν(A ∩ P ) − ν(A ∩ N ) = 0.
Since this equation holds for all f ∈ Sf (A, |ν|) with |f | ≤ 1, µ4 ≤ µ0 as claimed.
Now suppose that ν is a complex measure. Then A is a null set for both νr :=
Re ν and νi := Im ν. Therefore |ν| (A) ≤ |νr | (A) + |νi | (A) = 0.
Theorem 24.34 (Radon Nikodym Theorem for Complex Measures). dν
2) Here is another proof in the complex case. Let ρ = d|ν| , then by assump-
Let ν be a complex measure and µ be a σ – finite positive measure on (X, M). tion of A being ν – null,
Then ν has a unique Lebesgue decomposition ν = νa + νs relative to µ and there Z
exists a unique element ρ ∈ L1 (µ) such that such that dνa = ρdµ. Moreover,
0 = ν(B) = ρd |ν| for all B ∈ MA .
νs = 0 iff ν µ, i.e. dν = ρdµ iff ν µ. B
Proof. Uniqueness. Is a direct consequence of Lemmas 24.10 and 24.11. This shows that ρ1A = 0, |ν| – a.e. and hence
Existence. Let g : X → S 1 ⊂ C be a function such that dν = gd |ν| . By Z Z
Theorem 24.13, there exists h ∈ L1 (µ) and a positive measure |ν|s such that
|ν| (A) = |ρ| d |ν| = 1A |ρ| d |ν| = 0.
|ν|s ⊥ µ and d |ν| = hdµ + d |ν|s . Hence we have dν = ρdµ + dνs with ρ := A X
gh ∈ L1 (µ) and dνs := gd |ν|s . This finishes the proof since, as is easily verified,
νs ⊥ µ.
Theorem 24.36 ( – δ Definition of Absolute Continuity). Let ν be a
complex measure and µ be a positive measure on (X, M). Then ν µ iff for
24.4 Absolute Continuity on an Algebra all ε > 0 there exists a δ > 0 such that |ν(A)| < ε whenever A ∈ M and
µ(A) < δ.
The following results will be needed in Section 30.4 below.
Proof. (⇐=) If µ(A) = 0 then |ν(A)| < ε for all ε > 0 which shows that
Exercise 24.1. Let ν = νr + i
iν is a complex measure on a measurable space, ν(A) = 0, i.e. ν µ.
(X, M) , then |ν r | ≤ |ν| , ν i ≤ |ν| and |ν| ≤ |ν r | + ν i .
(=⇒) Since ν µ iff |ν| µ and |ν(A)| ≤ |ν| (A) for all A ∈ M, it suffices
Exercise 24.2. Let ν be a signed measure on a measurable space, (X, M) . If to assume ν ≥ 0 with ν(X) < ∞. Suppose for the sake of contradiction there
A ∈ M is set such that there exists M < ∞ such that |ν (B)| ≤ M for all exists ε > 0 and An ∈ M such that ν(An ) ≥ ε > 0 while µ(An ) ≤ 21n . Let
B ∈ MA = {C ∩ A : C ∈ M} , then |ν| (A) ≤ 2M. If ν is complex measure
∞ [
with A ∈ M and M < ∞ as above, then |ν| (A) ≤ 4M. \
A = {An i.o.} = An
Lemma 24.35. Let ν be a complex or a signed measure on (X, M). Then A ∈ N =1 n≥N
M is a ν – null set iff |ν| (A) = 0. In particular if µ is a positive measure on
(X, M), ν µ iff |ν| µ. so that

24.5 Exercises 273
∞
X 24.5 Exercises
µ(A) = lim µ (∪n≥N An ) ≤ lim µ(An ) ≤ lim 2−(N −1) = 0.
N →∞ N →∞ N →∞
n=N
Exercise 24.3. Prove Theorem 24.14 for p ∈ [1, 2] by directly applying the
On the other hand, Riesz theorem to φ|L2 (µ) .
ν(A) = lim ν (∪n≥N An ) ≥ lim inf ν(An ) ≥ ε > 0 Exercise 24.4. Show |ν| be defined as in Eq. (24.7) is a positive measure. Here
N →∞ n→∞
is an outline.
showing that ν is not absolutely continuous relative to µ.
1 1. Show
Corollary 24.37. Let µ be a positive measure M) and f ∈ L (dµ).
on (X,
R |ν| (A) + |ν| (B) ≤ |ν| (A ∪ B). (24.16)
Then for all ε > 0 there exists δ > 0 such that f dµ < ε for all A ∈ M such
A when A,
`B are disjoint sets in M.
that µ(A) < δ. ∞
2. If A = n=1 An with An ∈ M then
R
Proof. Apply theorem 24.36 to the signed measure ν(A) = f dµ for all ∞
A
X
A ∈ M. |ν| (A) ≤ |ν| (An ). (24.17)
n=1
Theorem 24.38 (Absolute Continuity on an Algebra). Let ν be a complex
measure and µ be a positive measure on (X, M). Suppose that A ⊂ M is an 3. From Eqs. (24.16) and (24.17) it follows that |ν| is finitely additive, and
algebra such that σ(A) = M and that µ is σ – finite on A. Then ν µ iff for hence
N N
all ε > 0 there exists a δ > 0 such that |ν(A)| < ε for all A ∈ A which satisfy X X
|ν| (A) = |ν| (An ) + |ν| (∪n>N An ) ≥ |ν| (An ).
µ(A) < δ. n=1 n=1
Proof. (=⇒) This implication is a consequence of Theorem 24.36. P∞
Letting N → ∞ in this inequality shows |ν| (A) ≥ n=1 |ν| (An ) which
(⇐=) If |ν(A)| < ε for all A ∈ A with µ(A) < δ, then by Exercise 24.2, combined with Eq. (24.17) shows |ν| is countably additive.
|ν| (A) ≤ 4ε for all A ∈ A with µ(A) < δ. Because of this argument, we may
now replace ν by |ν| and hence we may assume that ν is a positive finite measure. Exercise 24.5. Suppose X is a set, A ⊂ 2X is an algebra, and ν : A → C is a
Let ε > 0 and δ > 0 be such that ν(A) < ε for all A ∈ A with µ(A) < δ. finitely additive measure. For any A ∈ A, let
Suppose that B ∈ M with µ(B) < δ and α ∈ (0, δ − µ (B)) . By Corollary ( n )
n
22.18, there exists A ∈ A such that X a
|ν| (A) := sup |ν (Ai )| : A = Ai with Ai ∈ A and n ∈ N .
µ (A∆B) + ν (A∆B) = (µ + ν) (A∆B) < α. i=1 i=1
n m
In particular it follows that µ (A) ≤ µ (B) + µ (A∆B) < δ and hence by as- 1. Suppose P := {Ai }i=1 ⊂ A is a partition of A ∈ A and {Bj }j=1 ⊂ A is
sumption ν (A) < ε. Therefore, partition of A which refines P (i.e. for each j there exists an i such that
ν (B) ≤ ν (A) + ν (A∆B) < ε + α Bj ⊂ Ai ), then
Xn m
X
and letting α ↓ 0 in this inequality shows ν (B) ≤ ε. |ν (Ai )| ≤ |ν (Bj )| . (24.18)
Alternative Proof. Let ε > 0 and δ > 0 be such that ν(A) < ε for all i=1 j=1
A ∈ A with µ(A) < δ. Suppose that B ∈ M with µ(B) < δ. Use the regularity 2. The total variation,|ν| : A → [0, ∞] , of ν is a finitely additive measure
Theorem 28.6 below (or see Theorem 33.9 or Corollary 32.42) to find A ∈ Aσ on A.
such that B ⊂ A and µ(B) ≤ µ(A) < δ. Write A = ∪n An with An ∈ A. By
∞
replacing An by ∪nj=1 Aj if necessary we may assume that An is increasing in n. Exercise 24.6. Suppose that {νn }n=1 are complex measures on a measurable
Then µ(An ) ≤ µ(A) < δ for each n and hence by assumption ν(An ) < ε. Since space, (X, M) .
B ⊂ A = ∪n An it follows that ν(B) ≤ ν(A) = limn→∞ ν(An ) ≤ ε. Thus we P∞ P∞
have shown that ν(B) ≤ ε for all B ∈ M such that µ(B) < δ. 1. If n=1 |νn | (X) < ∞, then ν := n=1 νn is a complex measure.

2. If there is a finite positive measure, µ : M → [0, ∞) such that |νn (A)| ≤
µ (A) for all A ∈ M and ν (A) := limn→∞ νn (A) exists for all A ∈ M, then
ν is also a complex measure.
Exercise 24.7. Suppose µi , νi are σ – finite positive measures on measurable
spaces, (Xi , Mi ), for i = 1, 2. If νi µi for i = 1, 2 then ν1 ⊗ ν2 µ1 ⊗ µ2 and
in fact
d(ν1 ⊗ ν2 )
(x1 , x2 ) = ρ1 ⊗ ρ2 (x1 , x2 ) := ρ1 (x1 )ρ2 (x2 )
d(µ1 ⊗ µ2 )
where ρi := dνi /dµi for i = 1, 2.
Exercise 24.8. Let X = [0, 1] , M := B[0,1] , m be Lebesgue measure and µ be
counting measure on X. Show
1. m µ yet there is not function ρ such that dm = ρdµ.
2. Counting measure µ has no Lebesgue decomposition relative to m.
Exercise 24.9. Suppose that ν is a signed or complex measure on (X, M) and
An ∈ M such that either An ↑ A or An ↓ A and ν(A1 ) ∈ R, then show
ν(A) = limn→∞ ν(An ).
Exercise 24.10. Let (X, M) be a measurable space, ν : M → [−∞, ∞) be a
signed measure, and ν = ν+ − ν− be a Jordan decomposition of ν. If ν := α − β
with α and β being positive measures and α (X) < ∞, show ν+ ≤ α and
ν− ≤ β. Us this result to prove the uniqueness of Jordan decompositions stated
in Theorem 24.19.
Exercise 24.11. Let ν1 and ν2 be two signed measures on (X, M) which are
assumed to be valued in [−∞, ∞). Show, |ν1 + ν2 | ≤ |ν1 | + |ν2 | . Hint: use
Exercise 24.10 along with the observation that ν1 + ν2 = (ν1+ + ν2+ ) − (ν1− + ν2− ),
where νi± := (νi )± .
Exercise 24.12. Folland Exercise 3.7a on p. 88.
Exercise 24.13. Show Theorem 24.36 may fail if ν is not finite. (For a hint,
see problem 3.10 on p. 92 of Folland.)
Exercise 24.16. If ν is a complex measure on (X, M) such that |ν| (X) =
ν (X) , then ν = |ν| .
Exercise 24.17. Suppose ν is a complex or a signed measure on a measurable
space, (X, M) . Show A ∈ M is a ν - null set iff |ν| (A) = 0. Use this to conclude
that if µ is a positive measure, then ν ⊥ µ iff |ν| ⊥ µ.
25
Three Fundamental Principles of Banach Spaces
25.1 The Hahn-Banach Theorem |f (x)| = λf (x) = f (λx) = u(λx) ≤ p(λx) = p(x)
Our goal here is to show that continuous dual, X ∗ , of a Banach space, X, is holds for all x ∈ X.
always large. This will be the content of the Hahn-Banach Theorem 25.4 below.
Definition 25.2 (Minkowski functional). A function p : X → R is a
Proposition 25.1. Let X be a complex vector space over C and let XR denote Minkowski functional if
X thought of as a real vector space. If f ∈ X ∗ and u = Ref ∈ XR∗ then
1. p(x + y) ≤ p(x) + p(y) for all x, y ∈ X and
f (x) = u(x) − iu(ix). (25.1) 2. p(cx) = cp(x) for all c ≥ 0 and x ∈ X.
Conversely if u ∈ XR∗ and f is defined by Eq. (25.1), then f ∈ X ∗ and kukXR∗ = Example 25.3. Suppose that X = R and
kf kX ∗ . More generally if p is a semi-norm (see Definition 5.1) on X, then
p(x) = inf {λ ≥ 0 : x ∈ λ[−1, 2] = [−λ, 2λ]} .
|f | ≤ p iff u ≤ p.
Notice that if x ≥ 0, then p(x) = x/2 and if x ≤ 0 then p(x) = −x, i.e.
Proof. Let v(x) = Im f (x), then

v(ix) = Im f (ix) = Im(if (x)) = Ref (x) = u(x). x/2 if x ≥ 0
p(x) =
|x| if x ≤ 0.
Therefore
From this formula it is clear that p(cx) = cp(x) for all c ≥ 0 but not for c < 0.
f (x) = u(x) + iv(x) = u(x) + iu(−ix) = u(x) − iu(ix). Moreover, p satisfies the triangle inequality, indeed if p(x) = λ and p(y) = µ,
then x ∈ λ[−1, 2] and y ∈ µ[−1, 2] so that
Conversely for u ∈ XR∗ let f (x) = u(x) − iu(ix). Then
x + y ∈ λ[−1, 2] + µ[−1, 2] ⊂ (λ + µ) [−1, 2]
f ((a + ib)x) = u(ax + ibx) − iu(iax − bx)
= au(x) + bu(ix) − i(au(ix) − bu(x)) kf k2 = sup |f (x)|2 = sup (|u(x)|2 + |u(ix)|2 ).
kxk=1 kxk=1
while
Suppose that M = sup |u(x)| and this supremum is attained at x0 ∈ X with
(a + ib)f (x) = au(x) + bu(ix) + i(bu(x) − au(ix)). kxk=1
kx0 k = 1. Replacing x0 by −x0 if necessary, we may assume that u(x0 ) = M.
So f is complex linear. Because |u(x)| = |Ref (x)| ≤ |f (x)|, it follows that
Since u has a maximum at x0 ,
kuk ≤ kf k. For x ∈ X choose λ ∈ S 1 ⊂ C such that |f (x)| = λf (x) so
|f (x)| = f (λx) = u(λx) ≤ kuk kλxk = kukkxk.

d x0 + itx0
0= u
dt 0 kx0 + itx0 k
Since x ∈ X is arbitrary, this shows that kf k ≤ kuk so kf k = kuk.1 For the last
d 1
assertion, it is clear that |f | ≤ p implies that u ≤ |u| ≤ |f | ≤ p. Conversely if = (u(x0 ) + tu(ix0 )) = u(ix0 )
dt |1 + it|
u ≤ p and x ∈ X, choose λ ∈ S 1 ⊂ C such that |f (x)| = λf (x). Then 0
d d
√
1
Proof. To understand better why kf k = kuk, notice that since | |1
dt 0
+ it| = |
dt 0
1 + t2 = 0.This explains why kf k = kuk.
276 25 Three Fundamental Principles of Banach Spaces
which shows that p(x + y) ≤ λ + µ = p(x) + p(y). To check the last set inclusion wherein we use f ≤ p on M and the triangle inequality for p. In conclusion, if
let a, b ∈ [−1, 2], then α := supz∈M [f (z) − p(z − x)] and F (y + λx) := f (y) + λα, then by following
the above logic backwards, we have F |M = f and F ≤ p on M ⊕ Rx showing
λ µ F is the desired extension.
λa + µb = (λ + µ) a+ b ∈ (λ + µ) [−1, 2]
λ+µ λ+µ Step 2. Let us now write F : X → R to mean F is defined on a linear
µ
subspace D(F ) ⊂ X and F : D(F ) → R is linear. For F, G : X → R we will say
λ
since [−1, 2] is a convex set and λ+µ + λ+µ = 1. F ≺ G if D(F ) ⊂ D(G) and F = G|D(F ) , that is G is an extension of F. Let
BRUCE: Add in the relationship to convex sets and separation theorems, F = {F : X → R : f ≺ F and F ≤ p on D(F )}.
see Reed and Simon Vol. 1. for example.
Then (F, ≺) is a partially ordered set. If Φ ⊂ F is a chain (i.e. a linearlySordered
Theorem 25.4 (Hahn-Banach). Let X be a real vector space, p : X → R be a subset of F) then Φ has an upper bound G ∈ F defined by D(G) = D(F )
Minikowski functional, M ⊂ X be a subspace f : M → R be a linear functional F ∈Φ
such that f ≤ p on M. Then there exists a linear functional F : X → R such and G(x) = F (x) for x ∈ D(F ). Then it is easily checked that D(G) is a
that F |M = f and F ≤ p on X. linear subspace, G ∈ F, and F ≺ G for all F ∈ Φ. We may now apply Zorn’s
Lemma2 (see Theorem B.7) to conclude there exists a maximal element F ∈ F.
Proof. Step 1. We show for all x ∈ X \ M there exists and extension F Necessarily, D(F ) = X for otherwise we could extend F by step (1), violating
to M ⊕ Rx with the desired properties. If F exists and α = F (x), then for all the maximality of F. Thus F is the desired extension of f.
y ∈ M and λ ∈ R we must have
Corollary 25.5. Suppose that X is a complex vector space, p : X → [0, ∞) is
f (y) + λα = F (y + λx) ≤ p(y + λx). (25.2) a semi-norm, M ⊂ X is a linear subspace, and f : M → C is linear functional
such that |f (x)| ≤ p(x) for all x ∈ M. Then there exists F ∈ X 0 (X 0 is the
Dividing this equation by |λ| allows us to conclude that Eq. (25.2) is valid for algebraic dual of X) such that F |M = f and |F | ≤ p.
all y ∈ M and λ ∈ R iff
Proof. Let u = Ref then u ≤ p on M and hence by Theorem 25.4, there
f (y) + εα ≤ p(y + εx) for all y ∈ M and ε ∈ {±1} . exists U ∈ XR0 such that U |M = u and U ≤ p on M . Define F (x) = U (x) −
iU (ix) then as in Proposition 25.1, F = f on M and |F | ≤ p.
Equivalently put we must have, for all y, z ∈ M, that
Theorem 25.6. Let X be a normed space M ⊂ X be a closed subspace and
α ≤ p(y + x) − f (y) and x ∈ X \ M . Then there exists f ∈ X ∗ such that kf k = 1, f (x) = δ = d(x, M )
and f = 0 on M .
f (z) − p(z − x) ≤ α.
Proof. Define h : M ⊕ Cx → C by h(m + λx) := λδ for all m ∈ M and
Hence it is possible to find an α ∈ R such that Eq. (25.2) holds iff λ ∈ C. Then
f (z) − p(z − x) ≤ p(y + x) − f (y) for all y, z ∈ M. (25.3) |λ| δ δ δ
khk := sup = sup = =1
m∈M and λ6=0 km + λxk m∈M and λ6=0 kx + m/λk δ
(If Eq. (25.3) holds, then supz∈M [f (z) − p(z − x)] ≤ inf y∈M [p(y + x) − f (y)]
and so we may choose α = supz∈M [f (z) − p(z − x)] for example.) Now Equa- and by the Hahn – Banach theorem there exists f ∈ X ∗ such that f |M ⊕Cx = h
tion (25.3) is equivalent to having and kf k ≤ 1. Since 1 = khk ≤ kf k ≤ 1, it follows that kf k = 1.
2
The use of Zorn’s lemma in this step may be avoided in the case that p (x) is a
f (z) + f (y) = f (z + y) ≤ p(y + x) + p(z − x) for all y, z ∈ M
norm and X may be written as M ⊕ span(β) where β := {xn }∞ n=1 is a countable
subset of X. In this case, by step (1) and induction, f : M → R may be extended to
and this last equation is valid because
a linear functional F : M ⊕ span(β) → R with F (x) ≤ p (x) for x ∈ M ⊕ span(β).
This function F then extends by continuity to X and gives the desired extension
f (z + y) ≤ p (z + y) = p(y + x + z − x) ≤ p(y + x) + p(z − x),
of f.

25.1 The Hahn-Banach Theorem 277
Corollary 25.7. To each x ∈ X, let x̂ ∈ X ∗∗ be defined by x̂(f ) = f (x) for Exercise 25.2. Prove Item 3. of Lemma 25.11. Also show that it is possible
0
all f ∈ X ∗ . Then the map x ∈ X → x̂ ∈ X ∗∗ is a linear isometry of Banach that N̄ 6= N ⊥ . Hint: let Y be a non-reflexive Banach space (see Theorem
spaces. 7.16 and Theorem 25.13 below) and take N = X̂ ⊂ X ∗∗ = Y ∗ .
Proof. Since Solution to Exercise. 25.2If f ∈ N, then f |N ⊥ = 0 by definition of N ⊥ . This
0 0 0
|x̂(f )| = |f (x)| ≤ kf kX ∗ kxkX for all f ∈ X ∗ , shows f ∈ N ⊥ and hence N ⊂ N ⊥ . Since N ⊥ is a closed subspace it
0 0
now follows that N̄ ⊂ N ⊥ . If there exists f ∈ N ⊥ \ N̄ , then by the Hahn
it follows that kx̂kX ∗∗ ≤ kxkX . Now applying Theorem 25.6 with M = {0} , Banach theorem there exists λ ∈ X ∗∗ such that λ|N = 0 while λ (f ) 6= 0. If X
there exists f ∈ X ∗ such that kf k = 1 and |x̂(f )| = f (x) = kxk , which shows is reflexive, λ = x̂ for some x ∈ X and we have x̂ (N ) = 0 which is equivalent
that kx̂kX ∗∗ ≥ kxkX . This shows that x ∈ X → x̂ ∈ X ∗∗ is an isometry. Since 0
to x ∈ N ⊥ . Since f ∈ N ⊥ , we arrive at the contradiction,
isometries are necessarily injective, we are done.
0 = f (x) = x̂ (f ) = λ (f ) 6= 0,
Definition 25.8. A Banach space X is reflexive if the map x ∈ X → x̂ ∈ X ∗∗ 0
is surjective. and hence it follows that N̄ = N ⊥ .
Now let X = Y ∗ where Y is a non-reflexive Banach space and let N = X̂ ⊂
Example 25.9. Every Hilbert space H is reflexive. This is a consequence of the ⊥
X = Y ∗ . Notice that N is a closed subspace of X ∗∗ and that f ∈ N ⊂ Y =
∗∗
Riesz Theorem 8.15.
X ∗ iff 0 = x̂ (f ) = f (x) for all x ∈ X. This clearly shows that N ⊥ = {0} and
0
Exercise 25.1. Show all finite dimensional Banach spaces are reflexive. therefore N ⊥ = Y ∗ = X ∗∗ which properly contains N = X̂ when X is not
reflexive.
Definition 25.10. For subsets, M ⊂ X and N ⊂ X ∗ , let 0
Proposition 25.12. Suppose X is a Banach space, then X ∗∗∗ = (X [ ∗ ) ⊕ X̂
M 0 := {f ∈ X ∗ : f |M = 0} and
where
N ⊥ := {x ∈ X : f (x) = 0 for all f ∈ N }.
0
X̂ = {λ ∈ X ∗∗∗ : λ (x̂) = 0 for all x ∈ X} .
We call M 0 the annihilator of M and N ⊥ the backwards annihilator of N. In particular X is reflexive iff X ∗ is reflexive.
Lemma 25.11. Let M ⊂ X and N ⊂ X ∗ , then Proof. Let ψ ∈ X ∗∗∗ and define fψ ∈ X ∗ by fψ (x) := ψ(x̂) for all x ∈ X
and set ψ 0 := ψ − fˆψ . For x ∈ X (so x̂ ∈ X ∗∗ ) we have
1. M 0 and N ⊥ are always closed subspace of X ∗ and X respectively.
⊥
2. If M is a subspace of X, then M 0 = M̄ . ψ 0 (x̂) = ψ(x̂) − fˆψ (x̂) = fψ (x) − x̂(fψ ) = fψ (x) − fψ (x) = 0.
0
3. If N is a subspace, then N̄ ⊂ N ⊥ with equality if X is reflexive. See This shows ψ 0 ∈ X̂ 0 and we have shown X ∗∗∗ = X c∗ + X̂ 0 . If ψ ∈ X
c∗ ∩ X̂ 0 , then
Proposition 25.16 below. ψ = fˆ for some f ∈ X and 0 = fˆ(x̂) = x̂(f ) = f (x) for all x ∈ X, i.e. f = 0 so
∗
Proof. Since ψ = 0. Therefore X ∗∗∗ = X c∗ ⊕ X̂ 0 as claimed.

∗∗
If X is reflexive, then X̂ = X ∗∗ and so X̂ 0 = {0} showing (X ∗ ) = X ∗∗∗ =
M 0 = ∩x∈M Nul(x̂) and N ⊥ = ∩f ∈M Nul(f ), [ ∗ ), i.e. X ∗ is reflexive. Conversely if X ∗ is reflexive we conclude that X̂
0
(X =
M 0 and N ⊥ are both formed as an intersection of closed subspaces and hence {0} and therefore
are themselves closed subspaces. ⊥
⊥
⊥ X ∗∗ = {0} = X̂ 0 = X̂,
If x ∈ M, then f (x) = 0 for all f ∈ M 0 so that x ∈ M 0 and hence
0 ⊥ ∗

M̄ ⊂ M . If x ∈
/ M̄ , then there exists (by Theorem 25.6) f ∈ X such that which shows X̂ is reflexive. Here we have used
⊥ ⊥
f |M = 0 while f (x) 6= 0, i.e. f ∈ M 0 yet f (x) 6= 0. This shows x ∈ / M0 X̂ 0 = X̂ = X̂
⊥
and we have shown M 0 ⊂ M̄ . The proof of Item 3. is left to the reader in
Exercise 25.2. since X̂ is a closed subspace of X ∗∗ .

Theorem 25.13 (Continuation of Theorem 7.16). Let X be an infinite Exercise 25.3. Suppose p ∈ (1, ∞) and µ is a σ – finite measure on a measur-
set, µ : X → (0, ∞) be a function, p ∈ [1, ∞], q := p/ (p − 1) be the conjugate able space (X, M), then Lp (X, M, µ) is reflexive. Hint: model your proof on
exponent and for f ∈ `q (µ) define φf : `p (µ) → F by the proof of item 1. of Theorem 25.13 making use of Theorem 24.14.
X Lemma 25.14. Suppose that (X, o) is a pointed Hausdorff topological space
φf (g) := f (x) g (x) µ (x) . (25.4)
(i.e. o ∈ X is a fixed point) and ν is a finite measure on BX such that
x∈X
1. supp(ν) = X while ν ({o}) = 0 and
1. `p (µ) is reflexive for p ∈ (1, ∞) . 2. there exists fn ∈ C (X) such that fn → 1{o} boundedly as n → ∞.
∗
2. The map φ : `1 (µ) → `∞ (X) is not surjective.
3. `1 (µ) and `∞ (X) are not reflexive. (For example suppose X = [0, 1], o = 0, and µ = m.)
Then the map
∗
See Lemma 25.14 and Exercise 28.3 below for more examples of non-reflexive g ∈ L1 (ν) → φg ∈ L∞ (ν)
spaces.
is not surjective and the Banach space L1 (ν) is not reflexive. (In other words,
Proof. Theorem 24.14 may fail when p = ∞ and L1 - spaces need not be reflexive.)
Proof. Since supp(ν) = X, if f ∈ C (X) we have
1. This basically follows from two applications of item 3 of Theorem 7.16.
∗∗ ∗
More precisely if λ ∈ `p (µ) , let λ̃ ∈ `q (µ) be defined by λ̃ (g) = λ (φg ) kf kL∞ (ν) = sup {|f (x)| x ∈ X}
for g ∈ `q (µ) . Then by item 3., there exists f ∈ `p (µ) such that, for all
g ∈ `q (µ) , and we may view C (X) as a closed subspace of L∞ (ν) . For f ∈ C (X) , let
λ (φg ) = λ̃ (g) = φf (g) = φg (f ) = fˆ (φg ) . λ (f ) = f (o) . Then kλkC(X)∗ = 1, and therefore by Corollary 25.5 of the Hahn-
Banach Theorem, there exists an extension Λ ∈ (L∞ (ν))∗ such that λ = Λ|C(X)
∗
Since `p (µ) = {φg : g ∈ `q (µ)} , this implies that λ = fˆ and so `p (µ) is and kΛk = 1.
reflexive. If Λ = φg for some g ∈ L1 (ν) then we would have
2. Recall c0 (X) as defined in Notation 7.15 and is a closed subspace of `∞ (X) , Z
see Exercise 7.4. Let 1 ∈ `∞ (X) denote the constant function 1 on X. Notice f (o) = λ (f ) = Λ(f ) = φg (f ) = f gdν for all f ∈ C (X) .
that k1 − f k∞ ≥ 1 for all f ∈ c0 (X) and therefore, by the Hahn - Banach X
∗
Theorem, there exists λ ∈ `∞ (X) such that λ (1) = 0 while λ|c0 (X) ≡ 0. Applying this equality to the
∞
{fn }n=1 in item 2. of the statement of the lemma
1
Now if λ = φf for some f ∈ ` (µ) , then µ (x) f (x) = λ (δx ) = 0 for all x and then passing to the limit using the dominated convergence theorem, we
and f would have to be zero. This is absurd. arrive at the following contradiction;
∗ ∗ ∗
3. As we have seen `1 (µ) ∼ = `∞ (X) while `∞ (X) ∼ = c0 (X) 6= `1 (µ) . Let Z Z
∗
λ ∈ `∞ (X) be the linear functional as described above. We view this as 1 = lim fn (o) = lim fn gdν = 1{o} gdν = 0.
∗∗
an element of `1 (µ) by using n→∞ n→∞ X X
Hence we must conclude that Λ 6= φg for any g ∈ L1 (ν) .

λ̃ (φg ) := λ (g) for all g ∈ `∞ (X) . ∗
Since, by Theorem 24.14, the map f ∈ L∞ (ν) → φf ∈ L1 (ν) is an isomet-
∗∗
ric isomorphism of Banach spaces we may define L ∈ L1 (ν) by
Suppose that λ̃ = fˆ for some f ∈ `1 (µ) , then
L (φf ) := Λ (f ) for all f ∈ L∞ (ν) .
λ (g) = λ̃ (φg ) = fˆ (φg ) = φg (f ) = φf (g) .
If L were to equal ĝ for some g ∈ L1 (ν) , then
1
But λ was constructed in such a way that λ 6= φf for any f ∈ ` (µ) . It now Z
∗
follows from Proposition 25.12 that `1 (µ) ∼= `∞ (X) is also not reflexive. Λ (f ) = L (φf ) = ĝ (φf ) = φf (g) = f gdν
X
for all f ∈ C (X) ⊂ L∞ (ν) . But we have just seen this is impossible and
6 ĝ for any g ∈ L1 (ν) and thus L1 (ν) is not reflexive.
therefore L =

25.1.1 Hahn – Banach Theorem Problems 25.1.2 *Quotient spaces, adjoints, and more reflexivity
Exercise 25.4. Give another proof Corollary 10.15 based on Remark 10.13. Definition 25.15. Let X and Y be Banach spaces and A : X → Y be a linear
Hint: the Hahn Banach theorem implies operator. The transpose of A is the linear operator A† : Y ∗ → X ∗ defined by
A f (x) = f (Ax) for f ∈ Y ∗ and x ∈ X. The null space of A is the subspace
†

|λ(f (b)) − λ(f (a))|
kf (b) − f (a)k = sup . Nul(A) := {x ∈ X : Ax = 0} ⊂ X. For M ⊂ X and N ⊂ X ∗ let
λ∈X ∗ , λ6=0 kλk
Exercise 25.5. Prove Theorem 10.39 using the following strategy. M 0 := {f ∈ X ∗ : f |M = 0} and
1. Use the results from the proof in the text of Theorem 10.39 that N ⊥ := {x ∈ X : f (x) = 0 for all f ∈ N }.
Z d Z b
Proposition 25.16 (Basic properties of transposes and annihilators).
s→ f (s, t)dt and t → f (s, t)ds
c a
1. kAk = A† and A†† x̂ = Ax c for all x ∈ X.
are continuous maps. 2. M 0 and N ⊥ are always closed subspaces of X ∗ and X respectively.
2. For the moment take X = R and prove Eq. (10.24) holds by first proving it ⊥
3. M 0 = M̄ .
holds when f (s, t) = sm tn with m, n ∈ N0 . Then use this result along with 0
Theorem 10.35 to show Eq. (10.24) holds for all f ∈ C ([a, b] × [c, d], R) . 4. N̄ ⊂ N ⊥ with equality when X is reflexive.
3. For the general case, use the special case proved in item 2. along with Hahn 5. Nul(A) = Ran(A† )⊥ and Nul(A† ) = Ran(A)0 . Moreover, Ran(A) =
Banach Theorem 25.4. Nul(A† )⊥ and if X is reflexive, then Ran(A† ) = Nul(A)0 .
Exercise 25.6 (Liouville’s Theorem). (This exercise requires knowledge 6. X is reflexive iff X ∗ is reflexive. More generally X ∗∗∗ = X
c∗ ⊕ X̂ 0 where
of complex variables.) Let X be a Banach space and f : C →X be a
X̂ 0 = {λ ∈ X ∗∗∗ : λ (x̂) = 0 for all x ∈ X} .
function which is complex differentiable at all points z ∈ C, i.e. f 0 (z) :=
limh→0 (f (z + h) − f (z) /h exists for all z ∈ C. If we further suppose that Proof.
M := sup kf (z)k < ∞,
z∈C 1.
then f is constant. Hint: use the Hahn Banach Theorem 25.4 and the fact the kAk = sup kAxk = sup sup |f (Ax)|
result holds if X = C. kxk=1 kxk=1 kf k=1
= sup sup A† f (x) = sup A† f = A† .

Exercise 25.7. Let M be a finite dimensional subspace of a normed space,
kf k=1 kxk=1 kf k=1
X. Show there exists a closed subspace, N, such that X = M ⊕ N. Hint: let
β = {x1 , . . . , xn } ⊂ M be a basis for M and construct N making use of λi ∈ X ∗ 2. This is an easy consequence of the assumed continuity off all linear func-
which you should construct to satisfy, tionals involved. ⊥

1 if i = j 3. If x ∈ M, then f (x) = 0 for all f ∈ M 0 so that x ∈ M 0 . Therefore
λi (xj ) = δij = ⊥
0 if i 6= j. M̄ ⊂ M 0 . If x ∈ / M̄ , then there exists f ∈ X ∗ such that f |M = 0 while
⊥
Exercise 25.8. Folland 5.21, p. 160. f (x) 6= 0, i.e. f ∈ M 0 yet f (x) 6= 0. This shows x ∈ / M0 and we have
0 ⊥

shown M ⊂ M̄ .
Exercise 25.9. Let X be a Banach space such that X ∗ is separable. Show 0 0
X is separable as well. (The converse is not true as can be seen by taking 4. It is again simple to show N ⊂ N ⊥ and therefore N̄ ⊂ N ⊥ . Moreover,
X = `1 (N) .) Hint: use the greedy algorithm, i.e. suppose D ⊂ X ∗ \ {0} is a as above if f ∈ / N̄ there exists ψ ∈ X ∗∗ such that ψ|N̄ = 0 while ψ(f ) 6= 0.
countable dense subset of X ∗ , for ` ∈ D choose x` ∈ X such that kx` k = 1 and If X is reflexive, ψ = x̂ for some x ∈ X and since g(x) = ψ(g) = 0 for
|`(x` )| ≥ 21 k`k. all g ∈ N̄ , we have x ∈ N ⊥ . On the other hand, f (x) = ψ(f ) 6= 0 so
0 0
f∈ / N ⊥ . Thus again N ⊥ ⊂ N̄ .
Exercise 25.10. Folland 5.26.

5. 1. k·kX/M is a norm on X/M.

∗ 2. The projection map π : X → X/M has norm 1, kπk = 1.
Nul(A) = {x ∈ X : Ax = 0} = {x ∈ X : f (Ax) = 0 ∀ f ∈ X }
3. (X/M, k·kX/M ) is a Banach space.
= x ∈ X : A† f (x) = 0 ∀ f ∈ X ∗

4. If Y is another normed space and T : X → Y is a bounded linear transfor-
= x ∈ X : g(x) = 0 ∀ g ∈ Ran(A† ) = Ran(A† )⊥ . mation such that M ⊂ Nul(T ), then there exists a unique linear transfor-

mation S : X/M → Y such that T = S ◦ π and moreover kT k = kSk .
Similarly,
Proof. 1) Clearly kx + M k ≥ 0 and if kx + M k = 0, then there exists
Nul(A† ) = f ∈ Y ∗ : A† f = 0 = f ∈ Y ∗ : (A† f )(x) = 0 ∀ x ∈ X

mn ∈ M such that kx + mn k → 0 as n → ∞, i.e. x = lim mn ∈ M̄ = M. Since
n→∞
= {f ∈ Y ∗ : f (Ax) = 0 ∀ x ∈ X} x ∈ M, x + M = 0 ∈ X/M. If c ∈ C\ {0} , x ∈ X, then
= f ∈ Y ∗ : f |Ran(A) = 0 = Ran(A)0 .

kcx + M k = inf kcx + mk = |c| inf kx + m/ck = |c| kx + M k
6. Let ψ ∈ X ∗∗∗ and define fψ ∈ X ∗ by fψ (x) = ψ(x̂) for all x ∈ X and set m∈M m∈M
ψ 0 := ψ − fˆψ . For x ∈ X (so x̂ ∈ X ∗∗ ) we have because m/c runs through M as m runs through M. Let x1 , x2 ∈ X and
ψ (x̂) = ψ(x̂) − fˆψ (x̂) = fψ (x) − x̂(fψ ) = fψ (x) − fψ (x) = 0.
0 m1 , m2 ∈ M then
This shows ψ 0 ∈ X̂ 0 and we have shown X ∗∗∗ = X c∗ + X̂ 0 . If ψ ∈ Xc∗ ∩ X̂ 0 , kx1 + x2 + M k ≤ kx1 + x2 + m1 + m2 k ≤ kx1 + m1 k + kx2 + m2 k.
ˆ ∗ ˆ
then ψ = f for some f ∈ X and 0 = f (x̂) = x̂(f ) = f (x) for all x ∈ X, i.e.
Taking infimums over m1 , m2 ∈ M then implies
f = 0 so ψ = 0. Therefore X ∗∗∗ = X c∗ ⊕ X̂ 0 as claimed. If X is reflexive,
then X̂ = X ∗∗ and so X̂ 0 = {0} showing X ∗∗∗ = X c∗ , i.e. X ∗ is reflexive.
kx1 + x2 + M k ≤ kx1 + M k + kx2 + M k.
Conversely if X ∗ is reflexive we conclude that X̂ 0 = {0} and therefore
⊥
⊥
X ∗∗ = {0} = X̂ 0 = X̂, so that X is reflexive. and we have completed the proof the (X/M, k · k) is a normed space. 2) Since
kπ(x)k = inf m∈M kx + mk ≤ kxk for all x ∈ X, kπk ≤ 1. To see kπk = 1, let
Alternative proof. Notice that fψ = J † ψ, where J : X → X ∗∗ is given x ∈ X \ M so that π(x) 6= 0. Given α ∈ (0, 1), there exists m ∈ M such that
by Jx = x̂, and the composition
ˆ J † kx + mk ≤ α−1 kπ(x)k .
f ∈ X ∗ → fˆ ∈ X ∗∗∗ → J † fˆ ∈ X ∗
Therefore,
is the identity map since J † fˆ (x) = fˆ(Jx) = fˆ(x̂) = x̂(f ) = f (x) for all kπ(x + m)k kπ(x)k α kx + mk
= ≥ =α
∗ ˆ kx + mk kx + mk kx + mk
x ∈ X. Thus it follows that X → X ∗∗∗ is invertible iff J † is its inverse which
0
can happen iff Nul(J † ) = {0} . But as above Nul(J † ) = Ran (J) which will which shows kπk ≥ α. Since α ∈ (0, 1) is arbitrary
P we conclude that kπ(x)k = 1.
be zero iff Ran(J) = X ∗∗ and since J is an isometry this is equivalent to 3) Let π(xn ) ∈ X/M be a sequence such that kπ(xn )k < ∞.PAs above there
saying Ran (J) = X ∗∗ . So we have again shown X ∗ is reflexive iff X is exists mn ∈ M such that kπ(xn )k ≥ 21 kxn + mn k and hence kxn + mn k ≤
reflexive. P ∞
P
2 kπ(xn )k < ∞. Since X is complete, x := (xn + mn ) exists in X and
n=1
therefore by the continuity of π,
Theorem 25.17. Let X be a Banach space, M ⊂ X be a proper closed sub-
∞ ∞
space, X/M the quotient space, π : X → X/M the projection map π(x) = x+M X X
for x ∈ X and define the quotient norm on X/M by π(x) = π(xn + mn ) = π(xn )
n=1 n=1
kπ(x)kX/M = kx + M kX/M = inf kx + mkX .
m∈M showing X/M is complete. 4) The existence of S is guaranteed by the “factor
Then: theorem” from linear algebra. Moreover kSk = kT k because

n
kT k = kS ◦ πk ≤ kSk kπk = kSk some of the fk ’s if necessary, assume that {fk }k=1 is a linearly indepen-
n
dent set. Since the {fk }k=1 are linearly independent, the map x ∈ X →
and (f1 (x), . . . , fn (x)) ∈ Cn is surjective (why) and hence there exists x ∈ X
kS(π(x))k kT xk such that
kSk = sup = sup (f1 (x), . . . , fn (x)) = T x = (λ (f1 ) , . . . , λ(fn )) (25.7)
x∈M
/ kπ(x)k x∈M
/ kπ(x)k
kT xk kT xk as desired.
≥ sup = sup = kT k . 3. Let λ ∈ C ∗∗ ⊂ X ∗∗ and N be the weak - ∗ open neighborhood of λ as in Eq.
x∈M
/ kxk x6=0 kxk
(25.5). Working as before, given ε > 0, we need to find x ∈ C such that Eq.
(25.6). It will be left to the reader to verify that it suffices again to assume
n
{fk }k=1 is a linearly independent set. (Hint: Suppose that {f1 , . . . , fm } were
Theorem 25.18. Let X be a Banach space. Then n
a maximal linearly dependent subset of {fk }k=1 , then each fk with k > m
may be written as a linear combination {f1 , . . . , fm } .) As in the proof of
1. Identifying X with X̂ ⊂ X ∗∗ , the weak – ∗ topology on X ∗∗ induces the weak
item 2., there exists x ∈ X such that Eq. (25.7) holds. The problem is
topology on X. More explicitly, the map x ∈ X → x̂ ∈ X̂ is a homeomor-
that x may not be in C. To remedy this, let N := ∩nk=1 Nul(fk ) = Nul(T ),
phism when X is equipped with its weak topology and X̂ with the relative ∗
π : X → X/N ∼ = Cn be the projection map and f¯k ∈ (X/N ) be chosen so
topology coming from the weak-∗ topology on X ∗∗ . ¯
that fk = fk ◦ π for k = 1, 2, . . . , n. Then we have produced x ∈ X such
2. X̂ ⊂ X ∗∗ is dense in the weak-∗ topology on X ∗∗ .
that
3. Letting C and C ∗∗ be the closed unit balls in X and X ∗∗ respectively, then
Ĉ := {x̂ ∈ C ∗∗ : x ∈ C} is dense in C ∗∗ in the weak – ∗ topology on X ∗∗. . (λ (f1 ) , . . . , λ(fn )) = (f1 (x), . . . , fn (x)) = (f¯1 (π(x)), . . . , f¯n (π(x))).
4. X is reflexive iff C is weakly compact.
∗
Since f¯1 , . . . , f¯n is a basis for (X/N ) we find

(See Definition 14.36 for the topologies being used here.)
Pn
¯ Pn

i=1 αi fi (π(x))
| i=1 αi λ(fi )|
Proof. kπ(x)k = sup = sup Pn
n αi f¯i

α∈Cn \{0} k i=1 αi fi k
P
α∈Cn \{0} i=1
1. The weak – ∗ topology on X ∗∗ is generated by Pn
|λ( i=1 αi fi )|
= sup Pn
α∈C \{0} k i=1 αi fi k
n o
fˆ : f ∈ X ∗ = {ψ ∈ X ∗∗ → ψ(f ) : f ∈ X ∗ } . n
Pn
k i=1 αi fi k
≤ kλk sup Pn = 1.
So the induced topology on X is generated by α∈Cn \{0} k i=1 αi fi k
{x ∈ X → x̂ ∈ X ∗∗ → x̂(f ) = f (x) : f ∈ X ∗ } = X ∗ Hence we have shown kπ(x)k ≤ 1 and therefore for any α > 1 there
exists y = x + n ∈ X such that kyk < α and (λ (f1 ) , . . . , λ(fn )) =
and so the induced topology on X is precisely the weak topology. (f1 (y), . . . , fn (y)). Hence
2. A basic weak - ∗ neighborhood of a point λ ∈ X ∗∗ is of the form
|λ(fi ) − fi (y/α)| ≤ fi (y) − α−1 fi (y) ≤ (1 − α−1 ) |fi (y)|

N := ∩nk=1 {ψ ∈ X ∗∗ : |ψ(fk ) − λ(fk )| < ε} (25.5)
n
which can be arbitrarily small (i.e. less than ε) by choosing α sufficiently
for some {fk }k=1 ⊂ X ∗ and ε > 0. be given. We must now find x ∈ X such close to 1.
that x̂ ∈ N , or equivalently so that 4. Let Ĉ := {x̂ : x ∈ C} ⊂ C ∗∗ ⊂ X ∗∗ . If X is reflexive, Ĉ = C ∗∗ is weak - ∗
compact and hence by item 1., C is weakly compact in X. Conversely if C
|x̂(fk ) − λ(fk )| = |fk (x) − λ(fk )| < ε for k = 1, 2, . . . , n. (25.6)
is weakly compact, then Ĉ ⊂ C ∗∗ is weak – ∗ compact being the continuous
In fact we will show there exists x ∈ X such that λ(fk ) = fk (x) for image of a continuous map. Since the weak – ∗ topology on X ∗∗ is Hausdorff,
weak–∗
k = 1, 2, . . . , n. To prove this stronger assertion we may, by discarding it follows that Ĉ is weak – ∗ closed and so by item 3, C ∗∗ = Ĉ = Ĉ. So

n
if λ ∈ X ∗∗ , λ/ kλk ∈ C ∗∗ = Ĉ, i.e. there exists x ∈ C such that x̂ = λ/ kλk . X
ˆ ky − T xk k < αn+1 δ for all n ∈ N. (25.8)
This shows λ = (kλk x) and therefore X̂ = X ∗∗ .
k=1
∞ ∞ ∞
xn exists and kxk < 1, i.e. x ∈ B1X .
P P P
Since kxn k < αn < 1, x :=
n=1 n=1 n=1
Passing to the limit in Eq. (25.8) shows, ky − T xk = 0 and hence y ∈ T (B1X ) =
25.2 The Open Mapping Theorem E1 . Therefore we have shown BδX ⊂ E1 . The same scaling argument as above
X
then shows Bαδ ⊂ Eα for all α > 0.
Theorem 25.19 (Open Mapping Theorem). Let X, Y be Banach spaces,
3. If x ∈ V ⊂o X and y = T x ∈ T V we must show that T V contains a
T ∈ L(X, Y ). If T is surjective then T is an open mapping, i.e. T (V ) is open
ball B Y (y, ε) = T x + BεY for some ε > 0. Now B Y (y, ε) = T x + BεY ⊂ T V iff
in Y for all open subsets V ⊂ X.
BεY ⊂ T V − T x = T (V − x). Since V − x is a neighborhood of 0 ∈ X, there
Proof. For all α > 0 let BαX = {x ∈ X : kxkX < α} ⊂ X, BαY = exists α > 0 such that BαX ⊂ (V − x) and hence by assertion 2.,
{y ∈ Y : kykY < α} ⊂ Y and Eα = T (BαX ) ⊂ Y. The proof will be carried Y
out by proving the following three assertions. Bαδ ⊂ T BαX ⊂ T (V − x) = T (V ) − y
1. There exists δ > 0 such that BδαY

⊂ Eα for all α > 0. and therefore B Y (y, ε) ⊂ T V with ε := αδ.
Y
2. For the same δ > 0, Bδα ⊂ Eα , i.e. we may remove the closure in assertion
1. Corollary 25.20. If X, Y are Banach spaces and T ∈ L(X, Y ) is invertible
3. The last assertion implies T is an open mapping. (i.e. a bijective linear transformation) then the inverse map, T −1 , is bounded,
i.e. T −1 ∈ L(Y, X). (Note that T −1 is automatically linear.)
∞
S
1. Since Y = En , the Baire category Theorem 16.2 implies there exists Definition 25.21. Let X and Y be normed spaces and T : X → Y be linear
n−1
0 (not necessarily continuous) map.
n such that 6= ∅, i.e. there exists y ∈ E n and ε > 0 such that
En BY
(y, ε) ⊂ E n .
Suppose ky 0 k < ε then y and y + y 0 are in B Y (y, ε) ⊂ E n hence there exists 1. Let Γ : X → X × Y be the linear map defined by Γ (x) := (x, T (x)) for all
x̃, x ∈ BnX such that kT x̃ − (y + y 0 )k and kT x − yk may be made as small as x ∈ X and let
we please, which we abbreviate as follows Γ (T ) = {(x, T (x)) : x ∈ X}
kT x̃ − (y + y 0 )k ≈ 0 and kT x − yk ≈ 0. be the graph of T.
2. The operator T is said to be closed if Γ (T ) is closed subset of X × Y.
Hence by the triangle inequality,
Exercise 25.11. Let T : X → Y be a linear map between normed vector
kT (x̃ − x) − y 0 k = kT x̃ − (y + y 0 ) − (T x − y)k ∞
spaces, show T is closed iff for all convergent sequences {xn }n=1 ⊂ X such that
∞
≤ kT x̃ − (y + y 0 )k + kT x − yk ≈ 0 {T xn }n=1 ⊂ Y is also convergent, we have limn→∞ T xn = T (limn→∞ xn ) .
(Compare this with the statement that T is continuous iff for every convergent
∞ ∞
with x̃−x ∈ B2nX
. This shows that y 0 ∈ E2n which implies B Y (0, ε) ⊂ E2n . Since sequences {xn }n=1 ⊂ X we have {T xn }n=1 ⊂ Y is necessarily convergent and
α
the map φα : Y → Y given by φα (y) = 2n y is a homeomorphism, φα (E2n ) = Eα limn→∞ T xn = T (limn→∞ xn ) .)
Y Y αε Y ε
and φα (B (0, ε)) = B (0, 2n ), it follows that Bδα ⊂ Eα where δ := 2n > 0.
Y
2. Let δ be as in assertion 1., y ∈ Bδ and α1 ∈ (kyk /δ, 1). Choose Theorem 25.22 (Closed Graph Theorem). Let X and Y be Banach spaces
∞ P∞ and T : X → Y be linear map. Then T is continuous iff T is closed.
{αn }n=2 ⊂ (0, ∞) such that n=1 αn < 1. Since y ∈ BαY1 δ ⊂ Eα1 = T BαX1
by assertion 1. there exists x1 ∈ BαX1 such that ky − T x1 k < α2 δ. (Notice that Proof. If T is continuous and (xn , T xn ) → (x, y) ∈ X × Y as n → ∞ then
ky − T x1 k can be made as small as we please.) Similarly, since y−T x1 ∈ BαY2 δ ⊂ T xn → T x = y which implies (x, y) = (x, T x) ∈ Γ (T ). Conversely suppose T is

Ēα2 = T BαX2 there exists x2 ∈ BαX2 such that ky − T x1 − T x2 k < α3 δ. Con- closed, i.e. Γ (T ) is a closed subspace of X ×Y and is therefore a Banach space in
tinuing this way inductively, there exists xn ∈ BαXn such that its own right. The map π2 : X × Y → X is continuous and π1 |Γ (T ) : Γ (T ) → X

25.2 The Open Mapping Theorem 283
is continuous bijection which implies π1 |−1

Γ (T ) is bounded by the open mapping to a subsequence if necessary we know that g(x) = limn→∞ fn (x) = f (x) for m
Theorem 25.19. Therefore T = π2 ◦ Γ = π2 ◦ π1 |−1 - a.e. x. So f = g ∈ M.
Γ (T ) is bounded, being the
composition of bounded operators since the following diagram commutes i) Let i : (M, k · k∞ ) → (M, k · k2 ) be the identity map. Then i is bounded
and bijective. By the open mapping theorem, j = i−1 is bounded as well. Hence
Γ (T ) there exists A < ∞ such that kf k∞ = kj(f )k ≤ A kf k2 for all f ∈ M.
Γ = π1 |−1
Γ (T ) % & π2 ii) Let j : (M, k · k2 ) → (M, k · k∞ ) be the identity map. We will shows that
. j is a closed operator and hence bounded by the closed graph Theorem 25.22.
X −→ Y
T Suppose that fn ∈ M such that fn → f in L2 and fn = j(fn ) → g in C([0, 1]).
Then as in the first paragraph, we conclude that g = f = j(f ) a.e. showing j is
closed. Now finish as in last line of proof i).
As an application we have the following proposition. 2) For x ∈ [0, 1], let ex : M → C be the evaluation map ex (f ) = f (x). Then
Proposition 25.23. Let H be a Hilbert space. Suppose that T : H → H is a |ex (f )| ≤ |f (x)| ≤ kf k∞ ≤ Akf kL2
linear (not necessarily bounded) map such that there exists T ∗ : H → H such
that which shows that ex ∈ M∗ . Hence there exists a unique element gx ∈ M such
hT x|Y i = hx|T ∗ Y i ∀ x, y ∈ H. that
Then T is bounded. f (x) = ex (f ) = hf, gx i for all f ∈ M.
Moreover kgx kL2 = kex kM∗ ≤ A.
Proof. It suffices to show T is closed. To prove this suppose that xn ∈ H
3) Let {fj }nj=1 be an L2 – orthonormal subset of M. Then
such that (xn , T xn ) → (x, y) ∈ H × H. Then for any z ∈ H,
n n
hT xn |zi = hxn |T ∗ zi −→ hx|T ∗ zi = hT x|zi as n → ∞.
X X
A2 ≥ kex k2M∗ = kgx k2L2 ≥ |hfj , gx i|2 = |fj (x)|2
j=1 j=1
On the other hand limn→∞ hT xn |zi = hy|zi as well and therefore hT x|zi = hy|zi
for all z ∈ H. This shows that T x = y and proves that T is closed. and integrating this equation over x ∈ [0, 1] implies that
Here is another example.
n Z
X 1 n
X
Example 25.24. Suppose that M ⊂ L2 ([0, 1], m) is a closed subspace such that A2 ≥ |fj (x)|2 dx = 1=n
each element of M has a representative in C([0, 1]). We will abuse notation and j=1 0 j=1
simply write M ⊂ C([0, 1]). Then
which shows that n ≤ A2 . Hence dim(M) ≤ A2 .
1. There exists A ∈ (0, ∞) such that kf k∞ ≤ Akf kL2 for all f ∈ M.
2. For all x ∈ [0, 1] there exists gx ∈ M such that Remark 25.25. Keeping the notation in Example 25.24, G(x, y) = gx (y) for all
Z 1 x, y ∈ [0, 1]. Then
f (x) = hf |gx i := f (y) gx (y) dy for all f ∈ M. Z 1
0 f (x) = ex (f ) = f (y)G(x, y)dy for all f ∈ M.
0
Moreover we have kgx k ≤ A.
3. The subspace M is finite dimensional and dim(M) ≤ A2 . The function G is called the reproducing kernel for M.
Proof. 1) I will give a two proofs of part 1. Each proof requires that we first The above example generalizes as follows.
show that (M, k · k∞ ) is a complete space. To prove this it suffices to show M
is a closed subspace of C([0, 1]). So let {fn } ⊂ M and f ∈ C([0, 1]) such that Proposition 25.26. Suppose that (X, M, µ) is a finite measure space, p ∈
kfn − f k∞ → 0 as n → ∞. Then kfn − fm kL2 ≤ kfn − fm k∞ → 0 as m, n → [1, ∞) and W is a closed subspace of Lp (µ) such that W ⊂ Lp (µ) ∩ L∞ (µ).
∞, and since M is closed in L2 ([0, 1]), L2 − limn→∞ fn = g ∈ M. By passing Then dim(W ) < ∞.

Proof. With out loss of generality we may assume that µ(X) = 1. As in 25.3 Uniform Boundedness Principle
Example 25.24, we shows that W is a closed subspace of L∞ (µ) and hence by
the open mapping theorem, there exists a constant A < ∞ such that kf k∞ ≤ Theorem 25.27 (Uniform Boundedness Principle). Let X and Y be
A kf kp for all f ∈ W. Now if 1 ≤ p ≤ 2, then normed vector spaces, A ⊂ L(X, Y ) be a collection of bounded linear opera-
tors from X to Y,
kf k∞ ≤ A kf kp ≤ A kf k2
F = FA = {x ∈ X : sup kAxk < ∞} and
p 2 p−2 A∈A
and if p ∈ (2, ∞), then kf kp ≤ kf k2 kf k∞ or equivalently,
R = RA = F c = {x ∈ X : sup kAxk = ∞}. (25.9)
1−2/p A∈A
2/p 1−2/p 2/p
kf kp ≤ kf k2 kf k∞ ≤ kf k2 A kf kp
1. If sup kAk < ∞ then F = X.
1−2/p A∈A
from which we learn that kf kp ≤ A kf k2 and therefore that kf k∞ ≤ 2. If F is not meager, then sup kAk < ∞.
AA1−2/p kf k2 so that in any case there exists a constant B < ∞ such A∈A
N 3. If X is a Banach space, F is not meager iff sup kAk < ∞. In particular,
that kf k∞ ≤ B kf k2 . Let {fn }n=1 be an orthonormal subset of W and A∈A
PN
f = n=1 cn fn with cn ∈ C, then
if sup kAxk < ∞ for all x ∈ X then sup kAk < ∞.
N 2 A∈A A∈A
X N
X
2 2 2
cn fn ≤ B |cn | ≤ B 2 |c|

4. If X is a Banach space, then sup kAk = ∞ iff R is residual. In particular
n=1 ∞ n=1 A∈A
if sup kAk = ∞ then sup kAxk = ∞ for x in a dense subset of X.
2 PN 2
where |c| := n=1 |cn | . For each c ∈ CN , there is an exception set Ec such A∈A A∈A
that for x ∈
/ Ec , Proof. 1. If M := sup kAk < ∞, then sup kAxk ≤ M kxk < ∞ for all
N 2
X A∈A A∈A
2 x ∈ X showing F = X.
cn fn (x) ≤ B 2 |c| .

n=1
2. For each n ∈ N, let En ⊂ X be the closed sets given by
N
Let D := (Q + iQ) and E = ∩c∈D Ec . Then µ(E) = 0 and for x ∈ / E,
\
En = {x : sup kAxk ≤ n} = {x : kAxk ≤ n}.
PN 2 2
c f (x) ≤ B |c| for all c ∈ By continuity it then follows for x ∈
/E A∈A

n=1 n n D. A∈A
that
X N
2
Then F = ∪∞ n=1 En which is assumed to be non-meager and hence there exists

2
cn fn (x) ≤ B 2 |c| for all c ∈ CN . an n ∈ N such that En has non-empty interior. Let Bx (δ) be a ball such that

n=1
Bx (δ) ⊂ En . Then for y ∈ X with kyk = δ we know x − y ∈ Bx (δ) ⊂ En , so
that Ay = Ax − A(x − y) and hence for any A ∈ A,
Taking cn = fn (x) in this inequality implies that
N 2 N
kAyk ≤ kAxk + kA(x − y)k ≤ n + n = 2n.
X X
2 2
|fn (x)| ≤ B 2 |fn (x)| for all x ∈
/E

Hence it follows that kAk ≤ 2n/δ for all A ∈ A, i.e. sup kAk ≤ 2n/δ < ∞.
n=1 n=1 A∈A
3. If X is a Banach space, F = X is not meager by the Baire Category
and therefore that Theorem 16.2. So item 3. follows from items 1. and 2 and the fact that F = X
N
iff sup kAxk < ∞ for all x ∈ X.
X 2
|fn (x)| ≤ B 2 for all x ∈
/ E.
A∈A
n=1
4. Item 3. is equivalent to F is meager iff sup kAk = ∞. Since R = F c , R
2 2 A∈A
Integrating this equation over x then implies that N ≤ B , i.e. dim(W ) ≤ B .
is residual iff F is meager, so R is residual iff sup kAk = ∞.
A∈A

25.3 Uniform Boundedness Principle 285
∗ PN
Remarks 25.28 Let S ⊂ X be the unit sphere in X, fA (x) = Ax for x ∈ S Proof. Let fN ∈ `1 be given by fN (a) = n=1 an cn and set MN :=
and A ∈ A. max {|cn | : n = 1, . . . , N } . Then
1. The assertion sup kAxk < ∞ for all x ∈ X implies sup kAk < ∞ may |fN (a)| ≤ MN kak`1
A∈A A∈A
be interpreted as follows. If supA∈A kfA (x)k < ∞ for all x ∈ S, then
and by taking a = ek with k such MN = |ck | , we learn that kfN k = MN . Now
sup kfA k∞ < ∞ where kfA k∞ := supx∈S kfA (x)k = kAk .
A∈A by assumption, limN →∞ fN (a) exists for all a ∈ `1 and in particular,
2. If dim(X) < ∞ we may give a simple proof of this assertion. Indeed
N sup |fN (a)| < ∞ for all a ∈ `1 .
P{en }n=1 ⊂
if S is a basis for X there is a constant ε > 0 such that N
N PN
n=1 λn en ≥ ε n=1 |λn | and so the assumption supA∈A kfA (x)k < ∞

So by the uniform boundedness principle, Theorem 25.27,
implies
P
N
∞ > sup kfN k = sup MN = sup {|cn | : n = 1, 2, 3, . . . } .
n=1 λn Aen PN

|λn | kAen k N N
sup kAk = sup sup P
≤ sup sup n=1 PN
N ε n=1 |λn |
n=1 λn en
A∈A A∈A λ6=0 A∈A λ6=0
≤ ε−1 sup sup kAen k = ε−1 sup sup kAen k < ∞. 25.3.1 Applications to Fourier Series
A∈A n n A∈A
Notice that we have used the linearity of each A ∈ A in a crucial way. Let T = S 1 be the unit circle in S 1 , φn (z) := z n for all n ∈ Z, and m denote
3. If we drop the linearity assumption, so that fA ∈ C(S, Y ) for all A ∈ A the normalized arc length measure on T, i.e. if f : T → [0, ∞) is measurable,
– some index set, then it is no longer true that supA∈A kfA (x)k < ∞ then Z Z Z π
for all x ∈ S, then sup kfA k∞ < ∞. The reader is invited to construct a 1
A∈A f (w)dw := f dm := f (eiθ )dθ.
∞ T T 2π −π
counterexample when X = R2 and Y = R by finding a sequence {fn }n=1
of continuous functions on S 1 such that limn→∞ fn (x) = 0 for all x ∈ S 1 From Section 23.3, we know {φn }n∈Z is an orthonormal basis for L2 (T ). For
while limn→∞ kfn kC(S 1 ) = ∞. n ∈ N and z ∈ T, let
4. The assumption that X is a Banach space in item 3.of Theorem 25.27 can n Z
not be dropped. For example, let X ⊂ C([0, 1]) be the polynomial functions
X
sn (f, z) := hf |φn iφk (z) = f (w)dn (z w̄)dw
on [0, 1] equipped with the uniform norm k·k∞ and for t ∈ (0, 1], let ft (x) := k=−n T
d
(x(t) − x(0)) /t for all x ∈ X. Then limt→0 ft (x) = dt |0 x(t) and therefore
supt∈(0,1] |ft (x)| < ∞ for all x ∈ X. If the conclusion of Theorem 25.27 where
n
(item 3.) were true we would have M := supt∈(0,1] kft k < ∞. This would
X sin(n + 21 )θ
dn (eiθ ) := eikθ = ,
then imply k=−n
sin 12 θ

x(t) − x(0) see Eqs. (23.8) and (23.9). By Theorem 23.10, for all f ∈ L2 (T ) we know
≤ M kxk for all x ∈ X and t ∈ (0, 1].
∞
t
f = L2 (T ) − lim sn (f, ·).
Letting t ↓ 0 in this equation gives, |ẋ(0)| ≤ M kxk∞ for all x ∈ X. But n→∞
taking x(t) = tn in this inequality shows M = ∞. On the other hand the next proposition shows; if we fix z ∈ T, then
∞
Example 25.29. Suppose that {cn }n=1 ⊂ C is a sequence of numbers such that limn→∞ sn (f, z) does not even exist for the “typical” f ∈ C(T ) ⊂ L2 (T ).
N
X Proposition 25.30 (Lack of pointwise convergence). For each z ∈ T,
lim an cn exists in C for all a ∈ `1 . there exists a residual set Rz ⊂ C(T ) such that supn |sn (f, z)| = ∞ for all
N →∞
n=1 f ∈ Rz . Recall that C(T ) is a complete metric space, hence Rz is a dense
∞
Then c ∈ ` . subset of C(T ).

Proof. By symmetry considerations, it suffices to assume z = 1 ∈ T. Let Lemma 25.31. For f ∈ L1 (T ), let
Λn : C(T ) → C be given by Z
Z f˜(n) := hf, φn i = f (w)w̄n dw.
T
Λn f := sn (f, 1) = f (w)dn (w̄)dw.
T Then f˜ ∈ c0 := C0 (Z) (i.e limn→∞ f˜(n) = 0) and the map f ∈ L1 (T ) → f˜ ∈ c0
An application of Corollary 32.68 below shows, is a one to one bounded linear transformation into but not onto c0 .
2
Proof. By Bessel’s inequality, n∈Z f˜(n) < ∞ for all f ∈ L2 (T ) and in
Z P
kΛn k = kdn k1 = |dn (w̄)| dw
T particular lim|n|→∞ f˜(n) = 0. Given f ∈ L1 (T ) and g ∈ L2 (T ) we have

sin(n + 12 )θ
Z π Z π
1 −iθ
1
= dn (e ) dθ = dθ. (25.10) Z

2π −π 2π −π sin 21 θ ˜ n

f (n) − ĝ(n) = [f (w) − g(w)] w̄ dw ≤ kf − gk1

T
Of course we may prove this directly as follows. Since and hence
Z Z Z
lim sup f˜(n) = lim sup f˜(n) − ĝ(n) ≤ kf − gk1

|Λn f | = f (w)dn (w̄)dw ≤
|f (w)dn (w̄)| dw ≤ kf k∞ |dn (w̄)| dw, n→∞ n→∞
T T T
R for all g ∈ L (T ). Since L (T ) is dense in L1 (T ), it follows
2 2
that
we learn kΛn k ≤ |dn (w̄)| dw. For all ε > 0, let
T
lim supn→∞ f˜(n) = 0 for all f ∈ L1 , i.e. f˜ ∈ c0 . Since f˜(n) ≤ kf k1 ,

dn (z̄)

fε (z) := p . we have f˜ ≤ kf k1 showing that Λf := f˜ is a bounded linear transfor-

d2n (z̄) + ε c0
1 ˜
R from L (T ) to c0 . To see that Λ is injective, suppose f = Λf ≡ 0,
mation
Then kfε kC(T ) ≤ 1 and hence then T f (w)p(w, w̄)dw = 0 for all polynomials p in w and w̄. By the Stone -
Wierestrass and the dominated convergence theorem, this implies that
d2 (z̄)
Z Z
kΛn k ≥ lim |Λn fε | = lim p n |dn (z̄)| dw
Z
dw =
ε↓0 ε↓0 T d2n (z̄) + ε T f (w)g(w)dw = 0
T
and the verification of Eq. (25.10) is complete. for all g ∈ C(T ). Lemma 22.11 now implies f = 0 a.e. If Λ were surjective,
Using Z x Z x the open mapping theorem would imply that Λ−1 : c0 → L1 (T ) is bounded. In

|sin x| =

cos ydy ≤

|cos y| dy ≤ |x| particular this implies there exists C < ∞ such that

0 0
kf kL1 ≤ C f˜ for all f ∈ L1 (T ). (25.12)

in Eq. (25.10) implies that c0

Z π
sin(n + 12 )θ
Z π Taking f = dn , we find (because dñ (k) = 1|k|≤n ) that dñ = 1 while (by Eq.

1 dθ = 2 sin(n + 1 )θ dθ

kΛn k ≥
1
c0
2π −π
2θ
π 0 2 θ (25.11)) limn→∞ kdn kL1 = ∞ contradicting Eq. (25.12). Therefore Ran(Λ) 6=
Z (n+ 12 )π c0 .
2 π
Z
1 dθ dy
= sin(n + )θ = |sin y| → ∞ as n → ∞ (25.11)
π 0 2 θ 0 y
and hence supn kΛn k = ∞. So by Theorem 25.27, 25.4 Exercises
R1 = {f ∈ C(T ) : sup |Λn f | = ∞} 25.4.1 More Examples of Banach Spaces

n
Exercise 25.12. Let (X, M) be a measurable space and M (X) denote the
is a residual set. space of complex measures on (X, M) and for µ ∈ M (X) let kµk := |µ| (X).
See Rudin Chapter 5 for more details. Show (M (X), k·k) is a Banach space. (Move to Section 30.)

25.4 Exercises 287
Exercise 25.13. Folland 5.9, p. 155. (Drop this problem, or move to Chapter Exercise 25.20. Let X = C ([0, 1]) and Y = C 1 ([0, 1]) ⊂ X with both X and
9.) Y being equipped with the uniform norm. Let T : Y → X be the linear map,
T f = f 0 . Here C 1 ([0, 1]) denotes those functions, f ∈ C 1 ((0, 1)) ∩ C ([0, 1])
Exercise 25.14. Folland 5.10, p. 155. (Drop this problem, or move later where such that
it can be done.) f 0 (1) := lim f 0 (x) and f 0 (0) := lim f 0 (x)
x↑1 x↓0
Exercise 25.15. Folland 5.11, p. 155. (Drop this problem, or move to Chapter exist.
9.)
1. Y is a proper dense subspace of X and in particular Y is not complete.
25.4.2 Hahn-Banach Theorem Problems 2. T : Y → X is a closed operator which is not bounded.
Exercise 25.21. Folland 5.31, p. 164.

Exercise 25.16. Let X be a normed vector space. Show a linear functional,
f : X → C, is bounded iff M := f −1 ({0}) is closed. Hint: if M is closed yet Exercise 25.22. Let X be a vector space equipped with two norms, k·k1 and
f is not continuous, consider yn := x0 − xn /f (xn ) where x0 ∈ X such that k·k2 such that k·k1 ≤ k·k2 and X is complete relative to both norms. Show
f (x0 ) = 1 and xn ∈ X such that kxn k = 1 and limn→∞ |f (xn )| = ∞. there is a constant C < ∞ such that k·k2 ≤ C k·k1 .
Exercise 25.17. Let M be a closed subspace of a normed space, X, and x ∈ Exercise 25.23. Show that it is impossible to find a sequence, {an }n∈N ⊂
X \ M. Show M ⊕ Cx is closed. Hint: make use of a λ ∈ X ∗ which you should (0, ∞) , with the following property: if {λn }n∈N is a sequence in C, then
P∞
construct so that λ (M ) = 0 while λ (x) 6= 0. −1
n=1 |λn | < ∞ iff sup an |λn | < ∞. (Poetically speaking, there is no “slowest
rate” of decay for the summands of absolutely convergent series.)
Exercise 25.18. (Uses quotient spaces.) Let X be an infinite dimensional
Outline: For sake of contradiction suppose such a “magic” sequence
normed vector space. Show:
{an }n∈N ⊂ (0, ∞) were to exists.
∞
1. There exists a sequence {xn }n=1 ⊂ X such that kxn k = 1 for all n and 1. For f ∈ `∞ (N) , let (T f ) (n) := an f (n) for n ∈ N. Verify that T f ∈ `1 (N)
kxm − xn k ≥ 21 for all m 6= n. and T : `∞ (N) → `1 (N) is a bounded linear operator.
2. Show X is not locally compact. 2. Show T : `∞ (N) → `1 (N) must be an invertible operator and that
T −1 : `1 (N) → `∞ (N) is necessarily bounded, i.e. T : `∞ (N) → `1 (N)
25.4.3 Open Mapping and Closed Operator Problems is a homeomorphism.
3. Arrive at a contradiction by showing either that T −1 is not bounded or by
Exercise 25.19. Let X = `1 (N) , using the fact that, D, the set of finitely supported sequences, is dense in
( ∞
) `1 (N) but not in `∞ (N) .
X
Y = f ∈X: n |f (n)| < ∞ Exercise 25.24. Folland 5.34, p. 164. (Not a very good problem, delete.)
n=1
Exercise 25.25. Folland 5.35, p. 164. (A quotient space exercise.)
with Y being equipped with the `1 (N) - norm, and T : Y → X be defined by
(T f ) (n) = nf (n) . Show: Exercise 25.26. Folland 5.36, p. 164. (A quotient space exercise.)
1. Y is a proper dense subspace of X and in particular Y is not complete Exercise 25.27. Suppose T : X → Y is a linear map between two Banach
2. T : Y → X is a closed operator which is not bounded. spaces such that f ◦ T ∈ X ∗ for all f ∈ Y ∗ . Show T is bounded.
3. T : Y → X is algebraically invertible, S := T −1 : X → Y is bounded and
surjective but not open. Exercise 25.28. Suppose Tn : X → Y for n ∈ N is a sequence of bounded
linear operators between two Banach spaces such limn→∞ Tn x exists for all
x ∈ X. Show T x := limn→∞ Tn x defines a bounded linear operator from X to
Y.

Exercise 25.29. Let X, Y and Z be Banach spaces and B : X × Y → Z be a Exercise 25.36. Let X be a Banach space. Show every weakly compact subset
bilinear map such that B (x, ·) ∈ L (Y, Z) and B (·, y) ∈ L (X, Z) for all x ∈ X of X is norm bounded and every weak–∗ compact subset of X ∗ is norm bounded.
and y ∈ Y. Show there is a constant M < ∞ such that
Exercise 25.37. A vector subspace of a normed space X is norm-closed iff it
kB(x, y)k ≤ M kxk kyk for all (x, y) ∈ X × Y is weakly closed. (If X is not reflexive, it is not necessarily true that a normed
closed subspace of X ∗ need be weak* closed, see Exercise 25.39.) (Hint: this
and conclude from this that B : X × Y → Z is continuous problem only uses the Hahn-Banach Theorem.)
∞ ∞
Exercise 25.30. Folland 5.40, p. 165. (Condensation of singularities). Exercise 25.38. Let X be a Banach space, {Tn }n=1 and {Sn }n=1 be two se-
quences of bounded operators on X such that Tn → T and Sn → S strongly,
Exercise 25.31. Folland 5.41, p. 165. (Drop this exercise, it is 16.2.) ∞
and suppose {xn }n=1 ⊂ X such that limn→∞ kxn − xk = 0. Show:
25.4.4 Weak Topology and Convergence Problems 1. limn→∞ kTn xn − T xk = 0 and that
2. Tn Sn → T S strongly as n → ∞.
∞
Definition 25.32. A sequence {xn }n=1 ⊂ X is weakly Cauchy if for all
V ∈ τw such that 0 ∈ V, xn − xm ∈ V for all m, n sufficiently large. Similarly a
∞
sequence {fn }n=1 ⊂ X ∗ is weak–∗ Cauchy if for all V ∈ τw∗ such that 0 ∈ V,
fn − fm ∈ V for all m, n sufficiently large.
∞
Remark 25.33. These conditions are equivalent to {f (xn )}n=1 being Cauchy for
∞
all f ∈ X ∗ and {fn (x)}n=1 being Cauchy for all x ∈ X respectively.
Exercise 25.32. Let X and Y be Banach spaces. Show:

1. Every weakly Cauchy sequence in X is bounded.
2. Every weak-* Cauchy sequence in X ∗ is bounded.
∞
3. If {Tn }n=1 ⊂ L (X, Y ) converges weakly (or strongly) then
supn kTn kL(X,Y ) < ∞.
Exercise 25.33. Let X be a Banach space, C := {x ∈ X : kxk ≤ 1} and C ∗ :=

{λ ∈ X ∗ : kλkX ∗ ≤ 1} be the closed unit balls in X and X ∗ respectively.
1. Show C is weakly closed and C ∗ is weak-* closed in X and X ∗ respectively.
2. If E ⊂ X is a norm-bounded set, then the weak closure, Ē w ⊂ X, is also
norm bounded.
3. If F ⊂ X ∗ is a norm-bounded set, then the weak-* closure, Ē w−∗ ⊂ X ∗ , is
also norm bounded.
4. Every weak-* Cauchy sequence {fn } ⊂ X ∗ is weak-* convergent to some
f ∈ X ∗.
Exercise 25.35. If X is a separable normed linear space, the weak-* topology

on the closed unit ball in X ∗ is second countable and hence metrizable. (See
Theorem 14.38.)

26
Weak and Strong Derivatives
For this section, let Ω be an open subset of Rd , p, q, r ∈ [1, ∞], Lp (Ω) = hu, ∂ α φi = (−1)|α| hg, φi for all φ ∈ Cc∞ (Ω).
L (Ω, BΩ , m) and Lploc (Ω) = Lploc (Ω, BΩ , m), where m is Lebesgue measure on
p
More generally if p(ξ) = |α|≤N aα ξ α is a polynomial in ξ ∈ Rn , then p(∂)u
P
BRd and BΩ is the Borel σ – algebra on Ω. If Ω = Rd , we will simply write Lp
and Lploc for Lp (Rd ) and Lploc (Rd ) respectively. Also let exists weakly in Lp (Ω) (Lploc (Ω)) iff there exists g ∈ Lp (Ω) (Lploc (Ω)) such
Z that
hf, gi := f gdm hu, p(−∂)φi = hg, φi for all φ ∈ Cc∞ (Ω) (26.2)
Ω
and we denote g by w−p(∂)u.
1
for any pair of measurable functions f, g : Ω → C such that f g ∈ L (Ω). For
By Corollary 22.38, there is at most one g ∈ L1loc (Ω) such that Eq. (26.2)
example, by Hölder’s inequality, if hf, gi is defined for f ∈ Lp (Ω) and g ∈ Lq (Ω)
p holds, so w−p(∂)u is well defined.
when q = p−1 .
∞ Lemma 26.4. Let p(ξ) be a polynomial on Rd , k = deg (p) ∈ N, and u ∈
Definition 26.1. A sequence {un }n=1 ⊂ Lploc (Ω) is said to converge to u ∈ L1loc (Ω) such that p(∂)u exists weakly in L1loc (Ω). Then
Lploc (Ω) if limn→∞ ku − un kLq (K) = 0 for all compact subsets K ⊂ Ω.
1. suppm (w−p(∂)u) ⊂ suppm (u), where suppm (u) is the essential support of
The following simple but useful remark will be used (typically without fur- u relative to Lebesgue measure, see Definition 22.25.
ther comment) in the sequel. 2. If deg p = k and u|U ∈ C k (U, C) for some open set U ⊂ Ω, then w−p(∂)u =
p (∂) u a.e. on U.
Remark 26.2. Suppose r, p, q ∈ [1, ∞] are such that r−1 = p−1 +q −1 and ft → f
in Lp (Ω) and gt → g in Lq (Ω) as t → 0, then ft gt → f g in Lr (Ω). Indeed, Proof.
kft gt − f gkr = k(ft − f ) gt + f (gt − g)kr 1. Since

≤ kft − f kp kgt kq + kf kp kgt − gkq → 0 as t → 0
hw−p(∂)u, φi = −hu, p(−∂)φi = 0 for all φ ∈ Cc∞ (Ω \ suppm (u)),
an application of Corollary 22.38 shows w−p(∂)u = 0 a.e. on Ω \

26.1 Basic Definitions and Properties suppm (u). So by Lemma 22.26, Ω \ suppm (u) ⊂ Ω \ suppm (w−p(∂)u),
i.e. suppm (w−p(∂)u) ⊂ suppm (u).
Definition 26.3 (Weak Differentiability). Let v ∈ Rd and u ∈ Lp (Ω) 2. Suppose that u|U is C k and let ψ ∈ Cc∞ (U ). (We view ψ as a function
(u ∈ Lploc (Ω)) then ∂v u is said to exist weakly in Lp (Ω) (Lploc (Ω)) if there in Cc∞ (Rd ) by setting ψ ≡ 0 on Rd \ U.) By Corollary 22.35, there exists
exists a function g ∈ Lp (Ω) (g ∈ Lploc (Ω)) such that γ ∈ Cc∞ (Ω) such that 0 ≤ γ ≤ 1 and γ = 1 in a neighborhood of supp(ψ).
Then by setting γu = 0 on Rd \ supp(γ) we may view γu ∈ Cck (Rd ) and
hu, ∂v φi = −hg, φi for all φ ∈ Cc∞ (Ω). (26.1) so by standard integration by parts (see Lemma 22.36) and the ordinary
(w)
product rule,
The function g if it exists will be denoted by ∂v u. Similarly if α ∈ Nd0 and
∂ α is as in Notation 22.21, we say ∂ α u exists weakly in Lp (Ω) (Lploc (Ω)) iff hw−p(∂)u, ψi = hu, p(−∂)ψi = −hγu, p(−∂)ψi
there exists g ∈ Lp (Ω) (Lploc (Ω)) such that = hp(∂) (γu) , ψi = hp(∂)u, ψi (26.3)
290 26 Weak and Strong Derivatives
Proof. Let φ ∈ Cck (Ω) and choose η ∈ Cc∞ (B (0, 1)) such that Rd η(x)dx =
R
wherein the last equality we have γ is constant on supp(ψ). Since Eq.
(26.3) is true for all ψ ∈ Cc∞ (U ), an application of Corollary 22.38 with 1 and let ηε (x) := ε−d η(x/ε). Then ηε ∗ φ ∈ Cc∞ (Ω) for ε sufficiently small and
h = w−p(∂)u − p (∂) u and µ = m shows w−p(∂)u = p (∂) u a.e. on U. p (−∂) [ηε ∗ φ] = ηε ∗ p (−∂) φ → p (−∂) φ and ηε ∗ φ → φ uniformly on compact
sets as ε ↓ 0. Therefore by the dominated convergence theorem,
Notation 26.5 In light of Lemma 26.4 there is no danger in simply writing hp (∂) u, φi = limhp (∂) u, ηε ∗ φi = limhu, p (−∂) (ηε ∗ φ)i = hu, p (−∂) φi.
ε↓0 ε↓0
p (∂) u for w−p(∂)u. So in the sequel we will always interpret p(∂)u in the weak
or “distributional” sense.
Lemma 26.10 (Product Rule). Let u ∈ L1loc (Ω), v ∈ Rd and φ ∈ C 1 (Ω). If
Example 26.6. Suppose u(x) = |x| for x ∈ R, then ∂u(x) = sgn(x) in L1loc (R) (w) (w)
while ∂ 2 u(x) = 2δ(x) so ∂ 2 u(x) does not exist weakly in L1loc (R) . ∂v u exists in L1loc (Ω), then ∂v (φu) exists in L1loc (Ω) and
∂v(w) (φu) = ∂v φ · u + φ∂v(w) u a.e.

Example 26.7. Suppose d = 2 and u(x, y) = 1y>x . Then u ∈ L1loc R2 , while
∂x 1y>x = −δ (y − x) and ∂y 1y>x = δ (y − x) and so that neither ∂x u or ∂y u Moreover if φ ∈ Cc1 (Ω) and F := φu ∈ L1 (here we define F on Rd by setting
exists weakly. On the other hand (∂x + ∂y ) u = 0 weakly. To prove these as- (w)
F = 0 on Rd \ Ω ), then ∂ (w) F = ∂v φ · u + φ∂v u exists weakly in L1 (Rd ).
sertions, notice u ∈ C ∞ R2 \ ∆ where ∆ = (x, x) : x ∈ R2 . So by Lemma
26.4, for any polynomial p (ξ) without constant term, if p (∂) u exists weakly Proof. Let ψ ∈ Cc∞ (Ω), then using Lemma 26.9,
then p (∂) u = 0. However,
Z Z −hφu, ∂v ψi = −hu, φ∂v ψi = −hu, ∂v (φψ) − ∂v φ · ψi
hu, −∂x φi = − φx (x, y)dxdy = − φ(y, y)dy, = h∂v(w) u, φψi + h∂v φ · u, ψi
y>x R
Z Z = hφ∂v(w) u, ψi + h∂v φ · u, ψi.
hu, −∂y φi = − φy (x, y)dxdy = φ(x, x)dx and
y>x R This proves the first assertion. To prove the second assertion let γ ∈ Cc∞ (Ω) such
hu, −(∂x + ∂y )φi = 0 that 0 ≤ γ ≤ 1 and γ = 1 on a neighborhood of supp(φ). So for ψ ∈ Cc∞ (Rd ),
using ∂v γ = 0 on supp(φ) and γψ ∈ Cc∞ (Ω), we find
from which it follows that ∂x u and ∂y u can not be zero while (∂x + ∂y )u = 0.
On the other hand if p(ξ) and q (ξ) are two polynomials and u ∈ L1loc (Ω) hF, ∂v ψi = hγF, ∂v ψi = hF, γ∂v ψi = h(φu) , ∂v (γψ) − ∂v γ · ψi
is a function such that p(∂)u exists weakly in L1loc (Ω) and q (∂) [p (∂) u] exists = h(φu) , ∂v (γψ)i = −h∂v(w) (φu) , (γψ)i
weakly in L1loc (Ω) then (qp) (∂) u exists weakly in L1loc (Ω) . This is because
= −h∂v φ · u + φ∂v(w) u, γψi = −h∂v φ · u + φ∂v(w) u, ψi.
hu, (qp) (−∂) φi = hu, p (−∂) q(−∂)φi (w) (w)
This show ∂v F = ∂v φ · u + φ∂v u as desired.
= hp (∂) u, q(−∂)φi = hq(∂)p (∂) u, φi for all φ ∈ Cc∞ (Ω) .
Lemma 26.11. Suppose q ∈ [1, ∞), p(ξ) is a polynomial in ξ ∈ Rd and u ∈
Example 26.8. Let u(x, y) = 1x>0 + 1y>0 in L1loc R2 . Then ∂x u(x, y) = δ(x) ∞
Lqloc (Ω). If there exists {um }m=1 ⊂ Lqloc (Ω) such that p (∂) um exists in Lqloc (Ω)
and ∂y u(x,
y) = δ(y) so ∂x u(x, y) and ∂y u(x, y) do not exist weakly in for all m and there exists g ∈ Lqloc (Ω) such that for all φ ∈ Cc∞ (Ω),
L1loc R2 . However ∂y ∂x u does exists weakly and is the zero function. This
shows ∂y ∂x u may exists lim hum , φi = hu, φi and lim hp (∂) um , φi = hg, φi
weakly despite the fact both ∂x u and ∂y u do not m→∞ m→∞
exists weakly in L1loc R2 .
then p (∂) u exists in Lqloc (Ω) and p (∂) u = g.
Lemma 26.9. Suppose u ∈ L1loc (Ω) and p(ξ) is a polynomial of degree k such
that p (∂) u exists weakly in L1loc (Ω) then Proof. Since
hp (∂) u, φi = hu, p (−∂) φi for all φ ∈ Cck (Ω) . (26.4) hu, p (∂) φi = lim hum , p (∂) φi = − lim hp (∂) um , φi = hg, φi
m→∞ m→∞
Note: The point here is that Eq. (26.4) holds for all φ ∈ Cck (Ω) not just for all φ ∈Cc∞ (Ω),p (∂) u exists and is equal to g ∈ Lqloc (Ω).
φ ∈ Cc∞ (Ω) . Conversely we have the following proposition.

26.1 Basic Definitions and Properties 291
Proposition 26.12 (Mollification). Suppose q ∈ [1, ∞), p1 (ξ), . . . , pN (ξ) is lim kf ∗ ηε − f kLp (K) = 0. (26.5)
ε↓0
a collection of polynomials in ξ ∈ Rd and u ∈ Lqloc (Ω) such that pl (∂)u exists
weakly in Lqloc (Ω) for l = 1, 2, . . . , N. Then there exists un ∈ Cc∞ (Ω) such that Now let p(ξ) be a polynomial on Rd , u ∈ Lqloc (Ω) such that p (∂) u ∈ Lqloc (Ω)
un → u in Lqloc (Ω) and pl (∂) un → pl (∂) u in Lqloc (Ω) for l = 1, 2, . . . , N. and vε := ηε ∗ u ∈ C ∞ (Ωε ) as above. Then for x ∈ K and ε < ε0 ,
Proof. Let η ∈ Cc∞ (B(0, 1)) such that Rd ηdm = 1 and ηε (x) := ε−d η(x/ε)
R
Z Z
be as in the proof of Lemma 26.9. For any function f ∈ L1loc (Ω) , ε > 0 and p(∂)vε (x) = u(y)p(∂x )ηε (x − y)dy = u(y)p(−∂y )ηε (x − y)dy
x ∈ Ωε := {y ∈ Ω : dist(y, Ω c ) > ε} , let ZΩ Ω
Z = u(y)p(−∂y )ηε (x − y)dy = hu, p(∂)ηε (x − ·)i

fε (x) := f ∗ ηε (x) := 1Ω f ∗ ηε (x) = f (y)ηε (x − y)dy. Ω
Ω = hp(∂)u, ηε (x − ·)i = (p(∂)u)ε (x). (26.6)
∞
Notice that fε ∈ C (Ωε ) and Ωε ↑ Ω as ε ↓ 0. Given a compact set K ⊂ Ω
From Eq. (26.6) we may now apply Eq. (26.5) with f = u and f = pl (∂)u for
let Kε := {x ∈ Ω : dist(x, K) ≤ ε} . Then Kε ↓ K as ε ↓ 0, there exists ε0 > 0
1 ≤ l ≤ N to find
such that K0 := Kε0 is a compact subset of Ω0 := Ωε0 ⊂ Ω (see Figure 26.1)
and for x ∈ K, N
X
Z Z kvε − ukLp (K) + kpl (∂)vε − pl (∂)ukLp (K) → 0 as ε ↓ 0.
f ∗ ηε (x) := f (y)ηε (x − y)dy = f (y)ηε (x − y)dy. l=1
Ω Kε
For n ∈ N, let
Therefore, using Theorem 22.32,
Kn := {x ∈ Ω : |x| ≤ n and d(x, Ω c ) ≥ 1/n}
o
(so Kn ⊂ Kn+1 ⊂ Kn+1 for all n and Kn ↑ Ω as n → ∞ or see Lemma 14.23)
and choose ψn ∈ Cc∞ (Kn+1
o
, [0, 1]), using Corollary 22.35, so that ψn = 1 on a
neighborhood of Kn . Choose εn ↓ 0 such that Kn+1 ⊂ Ωεn and
N
X
kvεn − ukLp (Kn ) + kpl (∂)vεn − pl (∂)ukLp (Kn ) ≤ 1/n.
l=1
Then un := ψn · vεn ∈ Cc∞ (Ω) and since un = vεn on Kn we still have

N
X
kun − ukLp (Kn ) + kpl (∂)un − pl (∂)ukLp (Kn ) ≤ 1/n. (26.7)
l=1
Since any compact set K ⊂ Ω is contained in Kno for all n sufficiently large, Eq.
Fig. 26.1. The geomentry of K ⊂ K0 ⊂ Ω0 ⊂ Ω. (26.7) implies
" N
#
X
lim kun − ukLp (K) + kpl (∂)un − pl (∂)ukLp (K) = 0.
n→∞
kf ∗ ηε − f kLp (K) = k(1K0 f ) ∗ ηε − 1K0 f kLp (K) l=1
≤ k(1K0 f ) ∗ ηε − 1K0 f kLp (Rd ) → 0 as ε ↓ 0.

The following proposition is another variant of Proposition 26.12 which the
Hence, for all f ∈ Lqloc (Ω), f ∗ ηε ∈ C ∞ (Ωε ) and reader is asked to prove in Exercise 26.2 below.

Proposition 26.13. Suppose q ∈ [1, ∞), p1 (ξ),. . . , pN (ξ) is a collection of Proposition 26.16 (Weak-∗ Compactness: Banach - Alaoglu’s Theo-
polynomials in ξ ∈ Rd and u ∈ Lq = Lq Rd such that pl (∂)u ∈ Lq for rem). Let X be a separable Banach space and {fn } ⊂ X ∗ be a bounded sequence,
l = 1, 2, . . . , N. Then there exists un ∈ Cc∞ Rd such that then there exist a subsequence {fñ } ⊂ {fn } such that lim fn (x) = f (x) for all
n→∞
" # x ∈ X with f ∈ X ∗ .
N
X
lim kun − ukLq + kpl (∂)un − pl (∂)ukLq = 0. Proof. Let D ⊂ X be a countable linearly independent subset of X such
n→∞
l=1 that span(D) = X. Using Cantor’s diagonal trick, choose {fñ } ⊂ {fn } such
that λx := lim fñ (x) exist for all x ∈ D. Define f : span(D) → R by the
Notation 26.14 (Difference quotients) For v ∈ Rd and h ∈ R \ {0} and a n→∞
function u : Ω → C, let formula X X
f( ax x) = ax λx
u(x + hv) − u(x) x∈D x∈D
∂vh u(x) :=
h where by assumption # ({x ∈ D : ax 6= 0}) < ∞. Then f : span(D) → R is
for those x ∈ Ω such that x + hv ∈ Ω. When v is one of the standard basis linear and moreover fñ (y) → f (y) for all y ∈ span(D). Now
elements, ei for 1 ≤ i ≤ d, we will write ∂ih u(x) rather than ∂ehi u(x). Also let
|f (y)| = lim fñ (y) ≤ lim sup kfñ k kyk ≤ Ckyk for all y ∈ span(D).

n→∞
∇h u(x) := ∂1h u(x), . . . , ∂nh u(x) n→∞

Hence by the B.L.T. Theorem 10.4, f extends uniquely to a bounded linear

be the difference quotient approximation to the gradient.
functional on X. We still denote the extension of f by f ∈ X ∗ . Finally, if x ∈ X
Definition 26.15 (Strong Differentiability). Let v ∈ Rd and u ∈ Lp , then and y ∈ span(D)
∂v u is said to exist strongly in Lp if the limh→0 ∂vh u exists in Lp . We will
(s) |f (x) − fñ (x)| ≤ |f (x) − f (y)| + |f (y) − fñ (y)| + |fñ (y) − fñ (x)|
denote the limit by ∂v u.
≤ kf k kx − yk + kfñ k kx − yk + |f (y) − fñ (y)k
(s) (w)
It is easily verified that if u ∈ Lp , v ∈ Rd and ∂v u ∈ Lp exists then ∂v u
(w) (s) ≤ 2Ckx − yk + |f (y) − fñ (y)| → 2Ckx − yk as n → ∞.
exists and ∂v u = ∂v u. The key to checking this assertion is the identity,
Therefore
u(x + hv) − u(x)
Z
h
h∂v u, φi = φ(x)dx
h lim sup f (x) − fñ (x) ≤ 2Ckx − yk → 0 as y → x.
d

ZR n→∞
φ(x − hv) − φ(x) h
= u(x) dx = hu, ∂−v φi. (26.8)
R d h
p
(s)
Hence if ∂v u = limh→0 ∂vh u exists in Lp and φ ∈ Cc∞ (Rd ), then Corollary 26.17. Let p ∈ (1, ∞] and q = p−1 . Then to every bounded sequence
∞ ∞
{un }n=1 ⊂ L (Ω) there is a subsequence {ũn }n=1 and an element u ∈ Lp (Ω)
p
d such that
h∂v(s) u, φi = lim h∂vh u, φi = lim hu, ∂−v
h
φi = |0 hu, φ (· − hv)i = −hu, ∂v φi
h→0 h→0 dh lim hũn , gi = hu, gi for all g ∈ Lq (Ω) .
n→∞
wherein Corollary 19.43 has been used in the last equality to bring the deriva-
(w) (s) Proof. By Theorem 24.14, the map
tive past the integral. This shows ∂v u exists and is equal to ∂v u. What is
(w) ∗
somewhat more surprising is that the converse assertion that if ∂v u exists v ∈ Lp (Ω) → hv, ·i ∈ (Lq (Ω))
(s)
then so does ∂v u. Theorem 26.18 is a generalization of Theorem 23.15 from
2 p
L to L . For the reader’s convenience, let us give a self-contained proof of the is an isometric isomorphism of Banach spaces. By Theorem 22.15, Lq (Ω) is
version of the Banach - Alaoglu’s Theorem which will be used in the proof of separable for all q ∈ [1, ∞) and hence the result now follows from Proposition
Theorem 26.18. (This is the same as Theorem 14.38 above.) 26.16.

Theorem 26.18 (Weak and Strong Differentiability). Suppose p ∈ [1, ∞), Then um ∈ Cc∞ (Rd ), um → u and ∂v um → g in Lp as m → ∞. 3. =⇒ 4. By
u ∈ Lp (Rd ) and v ∈ Rd \ {0} . Then the following are equivalent: the fundamental theorem of calculus
∞
1. There exists g ∈ Lp (Rd ) and {hn }n=1 ⊂ R\ {0} such that limn→∞ hn = 0 um (x + hv) − um (x)
∂vh um (x) =
and h
lim h∂vhn u, φi = hg, φi for all φ ∈ Cc∞ (Rd ). 1 1 d
Z Z 1
n→∞ = um (x + shv)ds = (∂v um ) (x + shv)ds. (26.9)
(w)
h 0 ds 0
2. ∂v u exists and is equal to g ∈ Lp (Rd ), i.e. hu, ∂v φi = −hg, φi for all
φ∈ Cc∞ (Rd ). and therefore,
Lp Lp
3. There exists g ∈ Lp (Rd ) and un ∈ Cc∞ (Rd ) such that un → u and ∂v un → g Z 1
as n → ∞. ∂vh um (x) − ∂v um (x) = [(∂v um ) (x + shv) − ∂v um (x)] ds.
(s)
4. ∂v u exists and is is equal to g ∈ Lp (Rd ), i.e. ∂vh u → g in Lp as h → 0. 0
So by Minkowski’s inequality for integrals, Theorem 21.27,

Moreover if p ∈ (1, ∞) any one of the equivalent conditions 1. – 4. above
are implied by the following condition. h
Z 1
∂v um (x) − ∂v um ≤ k(∂v um ) (· + shv) − ∂v um kp ds
∞ p
10 . There exists
{hn }n=1 ⊂ R\ {0} such that limn→∞ hn = 0 and 0
supn ∂vhn u p < ∞.
and letting m → ∞ in this equation then implies
Proof. 4. =⇒ 1. is simply the assertion that strong convergence implies Z 1
weak convergence. 1. =⇒ 2. For φ ∈ Cc∞ (Rd ), Eq. (26.8) and the dominated
h
∂v u − g ≤ kg(· + shv) − gkp ds.
p
convergence theorem implies 0
hn By the dominated convergence theorem and Proposition 22.24, the right mem-
hg, φi = lim h∂vhn u, φi = lim hu, ∂−v φi = −hu, ∂v φi.
n→∞ n→∞ ber of this equation tends to zero as h → 0 and this shows item 4. holds. (10 . =⇒
1. when p > 1) This is a consequence of Corollary 26.17 (or see Theorem 14.38
2. =⇒ 3. Let η ∈ Cc∞ (Rd , R) such that Rd η(x)dx = 1 and let ηm (x) =
R
w
above) which asserts, by passing to a subsequence if necessary, that ∂vhn u → g
md η(mx), then by Proposition 22.34, hm := ηm ∗ u ∈ C ∞ (Rd ) for all m and p d
for some g ∈ L (R ).
Z
∂v hm (x) = ∂v ηm ∗ u(x) = ∂v ηm (x − y)u(y)dy Example 26.19. The fact that (10 ) does not imply the equivalent conditions 1 –
Rd 4 in Theorem 26.18 when p = 1 is demonstrated by the following example. Let
= hu, −∂v [ηm (x − ·)]i = hg, ηm (x − ·)i = ηm ∗ g(x). u := 1[0,1] , then
Z
By Theorem 22.32, hm → u ∈ Lp (Rd ) and ∂v hm = ηm ∗ g → g in Lp (Rd ) as u(x + h) − u(x)
Z
dx = 1

1[−h,1−h] (x) − 1[0,1] (x) dx = 2
m → ∞. This shows 3. holds except for the fact that hm need not have compact R
h |h| R
support. To fix this let ψ ∈ Cc∞ (Rd , [0, 1]) such that ψ = 1 in a neighborhood
of 0 and let ψε (x) = ψ(εx) and (∂v ψ)ε (x) := (∂v ψ) (εx). Then for |h| < 1. On the other hand the distributional derivative of u is ∂u(x) =
δ(x) − δ(x − 1) which is not in L1 .
∂v (ψε hm ) = ∂v ψε hm + ψε ∂v hm = ε (∂v ψ)ε hm + ψε ∂v hm Alternatively, if there exists g ∈ L1 (R, dm) such that
so that ψε hm → hm in Lp and ∂v (ψε hm ) → ∂v hm in Lp as ε ↓ 0. Let um = u(x + hn ) − u(x)

lim = g(x) in L1
ψεm hm where εm is chosen to be greater than zero but small enough so that n→∞ hn
∞
kψεm hm − hm kp + k∂v (ψεm hm ) → ∂v hm kp < 1/m. for some sequence {hn }n=1 as above. Then for φ ∈ Cc∞ (R) we would have on
one hand,

u(x + hn ) − u(x) φ(x − hn ) − φ(x)
Z Z
φ(x)dx = u(x)dx and hence by item 3. of Theorem 26.18, ∂i u ∈ L2 for each i. Since
R hn R hn Z
Z 1 2 2
k∇ukL2 = lim |∇un | dm = (−∆un , un )L2 → −(f, u) as n → ∞
→− φ0 (x)dx = (φ(0) − φ(1)) as n → ∞, n→∞ Rd
0
we also learn that
while on the other hand,
2
Z
u(x + hn ) − u(x)
Z k∇ukL2 = −(f, u) ≤ kf kL2 · kukL2 . (26.11)
φ(x)dx → g(x)φ(x)dx.
R hn R Let us now consider
These two equations imply d Z
X d Z
X
2
Z |∂i ∂j un | dm = − ∂j un ∂i2 ∂j un dm
g(x)φ(x)dx = φ(0) − φ(1) for all φ ∈ Cc∞ (R) (26.10) i,j=1 Rd i,j=1 Rd
R d Z d Z
X X
=− ∂j un ∂j ∆un dm = ∂j2 un ∆un dm
R
and in particular that R g(x)φ(x)dx = 0 for all φ ∈ Cc (R\ {0, 1}). By Corollary
j=1 Rd j=1 Rd
22.38, g(x) = 0 for m – a.e. x ∈ R\ {0, 1} and hence g(x) = 0 for m – a.e. x ∈ R. Z
But this clearly contradicts Eq. (26.10). This example also shows that the unit =
2
|∆un | dm = k∆un kL2 .
2
ball in L1 (R, dm) is not weakly sequentially compact. Compare with Lemma Rd
25.14 below.
Replacing un by un − um in this calculation shows
Corollary 26.20. If 1 ≤ p < ∞, u ∈ Lp such that ∂v u ∈ Lp , then ∂vh u Lp ≤
d Z
k∂v ukLp for all h 6= 0 and v ∈ Rd . X 2
|∂i ∂j (un − um )| dm = k∆(un − um )kL2 → 0 as m, n → ∞
2
i,j=1 Rd
Proof. By Minkowski’s inequality for integrals, Theorem 21.27, we may let
m → ∞ in Eq. (26.9) to find
and therefore by Lemma 26.4 (also see Exercise 26.4), ∂i ∂j u ∈ L2 Rd for all
Z 1 i, j and
∂vh u(x) = (∂v u) (x + shv)ds for a.e. x ∈ Rd d Z
X 2 2 2
0 |∂i ∂j u| dm = k∆ukL2 = kf kL2 . (26.12)
i,j=1 Rd
and Z 1
Combining Eqs. (26.11) and (26.12) gives the estimate
h
∂v u p ≤ k(∂v u) (· + shv)kLp ds = k∂v ukLp .
L
0 X 2 2 2
k∂ α ukL2 ≤ kukL2 + kf kL2 · kukL2 + kf kL2
|α|≤2
Proposition 26.21 (A weak form of Weyls Lemma). If u ∈ L2 (Rd ) such 2 2
2 d α 2 d

that f := 4u ∈ L (R ) then ∂ u ∈ L R for |α| ≤ 2. Furthermore if k ∈ N0 = kukL2 + k∆ukL2 · kukL2 + k∆ukL2 . (26.13)
and ∂ β f ∈ L2 Rd for all |β| ≤ k, then ∂ α u ∈ L2 Rd for |α| ≤ k + 2.
Let us now further assume ∂i f = ∂i ∆u ∈ L2 Rd . Then for h ∈ R \ {0} ,
Proof. By Proposition 26.13, there exists un ∈ Cc∞ Rd such that un → u
∂ih u ∈ L2 (Rd ) and ∆∂ih u = ∂ih ∆u = ∂ih f ∈ L2 (Rd ) and hence by Eq. (26.13)

and ∆un → ∆u = f in L2 Rd . By integration by parts we find and what we have just proved, ∂ α ∂ih u = ∂ih ∂ α u ∈ L2 and
∂ih ∂ α u 2 2 d ≤ ∂ih u 2 2 + ∂ih f 2 · ∂ih u 2 + ∂ih f 2 2

Z X
2 L (R ) L L L L
|∇(un − um )| dm = (−∆(un − um ), (un − um ))L2
|α|≤2
Rd
2 2
→ − (f − f, u − u) = 0 as m, n → ∞ ≤ k∂i ukL2 + k∂i f kL2 · k∂i ukL2 + k∂i f kL2

where the last inequality follows from Corollary 26.20. Therefore applying The- 2. If k∂ih ukLp (V ) ≤ CV for all h sufficiently small then by Corollary 26.17 there
orem 26.18 again we learn that ∂i ∂ α u ∈ L2 (Rd ) for all |α| ≤ 2 and w
exists hn → 0 such that ∂ihn u → v ∈ Lp (V ). Hence if ϕ ∈ Cc∞ (V ),
X 2 2 2
k∂i ∂ α ukL2 (Rd ) ≤ k∂i ukL2 + k∂i f kL2 · k∂i ukL2 + k∂i f kL2
Z Z Z
vϕdm = lim hn
∂i uϕdm = lim u∂i−hn ϕdm
|α|≤2 n→∞ n→∞ Ω
2 2 V Ω
≤ k∇ukL2 + k∂i f kL2 · k∇ukL2 + k∂i f kL2 Z Z
≤ kf kL2 · kukL2 =− u∂i ϕ dm = − u∂i ϕ dm.
q Ω V
2
+ k∂i f kL2 · kf kL2 · kukL2 + k∂i f kL2 .
Therefore ∂i u = v ∈ Lp (V ) and k∂i ukLp (V ) ≤ kvkLp (V ) ≤ CV .1 Finally if
The remainder of the proof, which is now an induction argument using the C := supV ⊂⊂Ω CV < ∞, then by the dominated convergence theorem,
above ideas, is left as an exercise to the reader.
k∂i ukLp (Ω) = lim k∂i ukLp (V ) ≤ C.
V ↑Ω
Theorem 26.22. Suppose that Ω is an open subset of Rd and V is an open
precompact subset of Ω.
We will now give a couple of applications of Theorem 26.18.
1. If 1 ≤ p < ∞, u ∈ Lp (Ω) and ∂i u ∈ Lp (Ω), then k∂ih ukLp (V ) ≤ k∂i ukLp (Ω)
for all 0 < |h| < 12 dist(V, Ω c ). Lemma 26.23. Let v ∈ Rd .
2. Suppose that 1 < p ≤ ∞, u ∈ Lp (Ω) and assume there exists a constants
1. If h ∈ L1 and ∂v h exists in L1 , then Rd ∂v h(x)dx = 0.
R
CV < ∞ and εV ∈ (0, 12 dist(V, Ω c )) such that
2. If p, q, r ∈ [1, ∞) satisfy r−1 = p−1 + q −1 , f ∈ Lp and g ∈ Lq are functions
k∂ih ukLp (V ) ≤ CV for all 0 < |h| < εV . such that ∂v f and ∂v g exists in Lp and Lq respectively, then ∂v (f g) exists in
Lr and ∂v (f g) = ∂v f · g + f · ∂v g. Moreover if r = 1 we have the integration
Then ∂i u ∈ Lp (V ) and k∂i ukLp (V ) ≤ CV . Moreover if C := supV ⊂⊂Ω CV < by parts formula,
∞ then in fact ∂i u ∈ Lp (Ω) and k∂i ukLp (Ω) ≤ C. h∂v f, gi = −hf, ∂v gi. (26.15)
Proof. 1. Let U ⊂o Ω such that V̄ ⊂ U and Ū is a compact subset of Ω. 3. If p = 1, ∂v f exists in L1 and g ∈ BC 1 (Rd ) (i.e. g ∈ C 1 (Rd ) with g and
For u ∈ C 1 (Ω) ∩ Lp (Ω), x ∈ B and 0 < |h| < 12 dist(V, U c ), its first derivatives being bounded) then ∂v (gf ) exists in L1 and ∂v (f g) =
∂v f · g + f · ∂v g and again Eq. (26.15) holds.
1
u(x + hei ) − u(x)
Z
∂ih u(x) = = ∂i u(x + thei ) dt Proof. 1) By item 3. of Theorem 26.18 there exists hn ∈ Cc∞ (Rd ) such that
h 0 hn → h and ∂v hn → ∂v h in L1 . Then
and in particular, Z
d
Z
d
Z
Z 1 ∂v hn (x)dx = |0 hn (x + hv)dx = |0 hn (x)dx = 0
|∂ih u(x)| ≤ |∂u(x + thei )|dt. Rd dt Rd dt Rd
0
1
Here we have used the result that if f ∈ Lp and fn ∈ Lp such that hfn , φi → hf, φi
Therefore by Minikowski’s inequality for integrals,
for all φ ∈ Cc∞ (V ) , then kf kLp (V ) ≤ lim inf n→∞ kfn kLp (V ) . To prove this, we have
p
Z 1 with q = p−1 that
k∂ih ukLp (V ) ≤ k∂u(· + thei )kLp (V ) dt ≤ k∂i ukLp (U ) . (26.14)
0 |hf, φi| = lim |hfn , φi| ≤ lim inf kfn kLp (V ) · kφkLq (V )
n→∞ n→∞
For general u ∈ Lp (Ω) with ∂i u ∈ Lp (Ω), by Proposition 26.12, there exists and therefore,
un ∈ Cc∞ (Ω) such that un → u and ∂i un → ∂i u in Lploc (Ω). Therefore we may
replace u by un in Eq. (26.14) and then pass to the limit to find |hf, φi|
kf kLp (V ) = sup ≤ lim inf kfn kLp (V ) .
φ6=0 kφkLq (V ) n→∞
k∂ih ukLp (V ) ≤ k∂i ukLp (U ) ≤ k∂i ukLp (Ω) .

and letting n → ∞ proves the first assertion. 2) Similarly there exists fn , gn ∈ which tends to zero as h → 0. The second item is proved analogously, or just
Cc∞ (Rd ) such that fn → f and ∂v fn → ∂v f in Lp and gn → g and ∂v gn → ∂v g make use of the fact that f ∗ g = g ∗ f and apply Item 1. Using the fact that
in Lq as n → ∞. So by the standard product rule and Remark 26.2, fn gn → g(x − ·) ∈ Cc∞ (Rd ) and the definition of the weak derivative,
f g ∈ Lr as n → ∞ and Z Z
f ∗ ∂v g(x) = f (y) (∂v g) (x − y)dy = − f (y) (∂v g(x − ·)) (y)dy
∂v (fn gn ) = ∂v fn · gn + fn · ∂v gn → ∂v f · g + f · ∂v g in Lr as n → ∞. d Rd
ZR
It now follows from another application of Theorem 26.18 that ∂v (f g) exists in = ∂v f (y)g(x − y)dy = ∂v f ∗ g(x).
Lr and ∂v (f g) = ∂v f · g + f · ∂v g. Eq. (26.15) follows from this product rule and Rd
item 1. when r = 1. 3) Let fn ∈ Cc∞ (Rd ) such that fn → f and ∂v fn → ∂v f in Item 3. is a consequence of this equality and items 1. and 2.
L1 as n → ∞. Then as above, gfn → gf in L1 and ∂v (gfn ) → ∂v g · f + g∂v f in
L1 as n → ∞. In particular if φ ∈ Cc∞ (Rd ), then Proposition 26.25. Let Ω = (α, β) ⊂ R be an open interval and f ∈ L1loc (Ω)
such that ∂ (w) f = 0 in L1loc (Ω). Then there exists c ∈ C such that f = c
hgf, ∂v φi = lim hgfn , ∂v φi = − lim h∂v (gfn ) , φi a.e. More generally, suppose F : Cc∞ (Ω) → C is a linear functional such that
n→∞ n→∞
F (φ0 ) = 0 for all φ ∈ Cc∞ (Ω), where φ0 (x) = dx
d
φ(x), then there exists c ∈ C
= − lim h∂v g · fn + g∂v fn , φi = −h∂v g · f + g∂v f, φi.
n→∞ such that Z
This shows ∂v (f g) exists (weakly) and ∂v (f g) = ∂v f · g + f · ∂v g. Again Eq. F (φ) = hc, φi = cφ(x)dx for all φ ∈ Cc∞ (Ω). (26.16)
Ω
(26.15) holds in this case by item 1. already proved.
Proof. Before giving a proof
R of the second assertion, let us show it includes
Lemma 26.24. Let p, q, r ∈ [1, ∞] satisfy p−1 + q −1 = 1 + r−1 , f ∈ Lp , g ∈ Lq the first. Indeed, if F (φ) := Ω φf dm and ∂ (w) f = 0, then F (φ0 ) = 0 for all
and v ∈ Rd . φ ∈ Cc∞ (Ω) and therefore there exists c ∈ C such that
1. If ∂v f exists strongly in Lr , then ∂v (f ∗ g) exists strongly in Lp and Z Z
φf dm = F (φ) = chφ, 1i = c φf dm.
∂v (f ∗ g) = (∂v f ) ∗ g. Ω Ω
2. If ∂v g exists strongly in Lq , then ∂v (f ∗ g) exists strongly in Lr and But this implies f = c a.e.
R So it only remains to prove the second assertion. Let
η ∈ Cc∞ (Ω) such that Ω ηdm = 1. Given φ ∈ Cc∞ (Ω) ⊂ Cc∞ (R) , let ψ(x) =
∂v (f ∗ g) = f ∗ ∂v g. Rx
(φ(y) − η(y)hφ, 1i) dy. Then ψ 0 (x) = φ(x) − η(x)hφ, 1i and ψ ∈ Cc∞ (Ω) as
−∞
3. If ∂v f exists weakly in Lp and g ∈ Cc∞ (Rd ), then f ∗ g ∈ C ∞ (Rd ), ∂v (f ∗ g) the reader should check. Therefore,
exists strongly in Lr and 0 = F (ψ) = F (φ − hφ, ηiη) = F (φ) − hφ, 1iF (η)
∂v (f ∗ g) = f ∗ ∂v g = (∂v f ) ∗ g.
which shows Eq. (26.16) holds with c = F (η). This concludes the proof, however
Proof. Items 1 and 2. By Young’s inequality (Theorem 22.30) and simple it will be instructive to give another proof of the first assertion.
computations: Alternative proof of first assertion. Suppose f ∈ L1loc (Ω) and ∂ (w) f = 0
0
and fm := f ∗ ηm as is in the proof of Lemma 26.9. Then fm = ∂ (w) f ∗ ηm = 0,
τ−hv (f ∗ g) − f ∗ g
so fm = cm for some constant cm ∈ C. By Theorem 22.32, fm → f in L1loc (Ω)

− (∂v f ) ∗ g
h
and therefore if J = [a, b] is a compact subinterval of Ω,
r
τ−hv f ∗ g − f ∗ g Z
= − (∂v f ) ∗ g 1
h
r
|cm − ck | = |fm − fk | dm → 0 as m, k → ∞.
b−a J
τ−hv f − f
= − (∂v f ) ∗ g ∞
So {cm }m=1 is a Cauchy sequence and therefore c := limm→∞ cm exists and
h
r

τ−hv f − f

f = limm→∞ fm = c a.e.
≤ − (∂v f )
kgkq We will say more about the connection of weak derivatives to pointwise
h p derivatives in Section 30.5 below.

26.2 Exercises 297
n
26.2 Exercises Exercise 26.8. Suppose that f ∈ L1loc (Ω) and v ∈ Rd and {ej }j=1is the
standard basis for Rd . If ∂j f := ∂ej f exists weakly inPL1loc (Ω) for all j =
Exercise 26.1. Give another proof of Lemma 26.10 base on Proposition 26.12. n
1, 2, . . . , n then ∂v f exists weakly in L1loc (Ω) and ∂v f = j=1 vj ∂j f.
Exercise 26.2. Prove Proposition 26.13. Hints: 1. Use uε as defined in the
Exercise 26.9. Suppose, f ∈ L1loc (Rd ) and ∂v f exists weakly and ∂v f = 0 in
proof of Proposition 26.12 to show it suffices
to consider the case where L1loc (Rd ) for all v ∈ Rd . Then there exists λ ∈ C such that f (x) = λ for m – a.e.
u ∈ C ∞ Rd ∩ Lq Rd with ∂ α u ∈ Lq Rd for all α ∈ Nd0 . 2. Then let

x ∈ Rd . Hint: See steps 1. and 2. in the outline given in Exercise 26.10 below.
ψ ∈ Cc∞ (B(0, 1), [0, 1]) such that ψ = 1 on a neighborhood of 0 and let
un (x) := u(x)ψ(x/n). Exercise 26.10 (A generalization of Exercise 26.9). Suppose Ω is a con-
d
Exercise 26.3. Suppose p(ξ) is a polynomial in ξ ∈ R , p ∈ (1, ∞), q := p nected open subset of Rd and f ∈ L1loc (Ω). If ∂ α f = 0 weakly for α ∈ Zn+ with
p−1 ,
|α| = N + 1, then f (x) = p(x) for m – a.e. x where p(x) is a polynomial of
u ∈ Lp such that p(∂)u ∈ Lp and v ∈ Lq such that p (−∂) v ∈ Lq . Show
degree at most N. Here is an outline.
hp (∂) u, vi = hu, p (−∂) vi.
1. Suppose x0 ∈ Ω and ε > 0 such that C := Cx0 (ε) ⊂ Ω and let ηn be a
Exercise 26.4. Let p ∈ [1, ∞), α be a multi index (if α = 0 let ∂ 0 be the
sequence of approximate δ – functions such supp(ηn ) ⊂ B0 (1/n) for all n.
identity operator on Lp ),
Then for n large enough, ∂ α (f ∗ ηn ) = (∂ α f ) ∗ ηn on C for |α| = N + 1. Now
D(∂ α ) := {f ∈ Lp (Rn ) : ∂ α f exists weakly in Lp (Rn )} use Taylor’s theorem to conclude there exists a polynomial pn of degree at
most N such that fn = pn on C.
and for f ∈ D(∂ α ) (the domain of ∂ α ) let ∂ α f denote the α – weak derivative 2. Show p := limn→∞ pn exists on C and then let n → ∞ in step 1. to show
of f. (See Definition 26.3.) there exists a polynomial p of degree at most N such that f = p a.e. on C.
3. Use Taylor’s theorem to show if p and q are two polynomials on Rd which
1. Show ∂ α is a densely defined operator on Lp , i.e. D(∂ α ) is a dense linear
agree on an open set then p = q.
subspace of Lp and ∂ α : D(∂ α ) → Lp is a linear transformation.
4. Finish the proof with a connectedness argument using the results of steps
2. Show ∂ α : D(∂ α ) → Lp is a closed operator, i.e. the graph,
2. and 3. above.
Γ (∂ α ) := {(f, ∂ α f ) ∈ Lp × Lp : f ∈ D(∂ α )} ,
Exercise 26.11. Suppose Ω ⊂o Rd and v, w ∈ Rd . Assume f ∈ L1loc (Ω)
is a closed subspace of Lp × Lp . and that ∂v ∂w f exists weakly in L1loc (Ω), show ∂w ∂v f also exists weakly and
3. Show ∂ α : D(∂ α ) ⊂ Lp → Lp is not bounded unless α = 0. (The norm on ∂w ∂v f = ∂v ∂w f.
D(∂ α ) is taken to be the Lp – norm.)
Exercise 26.12. Let d = 2 and f (x, y) = 1x≥0 . Show ∂ (1,1) f = 0 weakly in
Exercise 26.5. Let p ∈ [1, ∞), f ∈ Lp and α be a multi index. Show ∂ α f exists L1loc despite the fact that ∂1 f does not exist weakly in L1loc !
weakly (see Definition 26.3) in Lp iff there exists fn ∈ Cc∞ (Rn ) and g ∈ Lp such
that fn → f and ∂ α fn → g in Lp as n → ∞. Hints: See exercises 26.2 and
26.4.
Exercise 26.6. 8.8 on p. 246.
Exercise 26.7. Assume n = 1 and let ∂ = ∂e1 where e1 = (1) ∈ R1 = R.
1. Let f (x) = |x| , show ∂f exists weakly in L1loc (R) and ∂f (x) = sgn(x) for
m – a.e. x.
2. Show ∂(∂f ) does not exists weakly in L1loc (R).
3. Generalize item 1. as follows. Suppose f ∈ C(R, R) and there exists a finite
set Λ := {t1 < t2 < · · · < tN } ⊂ R such that f ∈ C 1 (R \ Λ, R). Assuming
∂f ∈ L1loc (R) , show ∂f exists weakly and ∂ (w) f (x) = ∂f (x) for m – a.e. x.

27
Bochner Integral
Part VII
Construction and Differentiation of Measures

28
Examples of Measures
In this chapter we are going to state a couple of construction theorems for 2. If µ is also countably sub-additive on E, then µ is a premeasure on A.
measures. The proofs of these theorems will be deferred until the next chapter, 3. If µ is a premeasure on A then
also see Chapter 32. Our goal in this chapter is to apply these construction ∞ ∞
theorems to produce a fairly broad class of examples of measures.
X a
µ̄(A) = inf{ µ(En ) : A ⊂ En with En ∈ E} (28.2)
n=1 n=1
X∞ [∞
28.1 Extending Premeasures to Measures = inf{ µ(En ) : A ⊂ En with En ∈ E} (28.3)
n=1 n=1
Throughout this chapter, X will be a given set which will often be taken to be
a locally compact Hausdorff space. extends µ to a measure µ̄ on σ (A) = σ(E).
4. If we further assume µ is σ – finite on E, then µ̄ is the unique measure on
Definition 28.1. Suppose that E ⊂ 2X is a collection of subsets of X and σ(E) such that µ̄|E = µ.
µ : E → [0, ∞] is a function. Then
Proof. Item 1. is Proposition 31.3, item 2. is Proposition 31.5, item 3. is
1. µ is additive or finitely additive on E if contained in Theorem 31.18 (or see Theorems 31.15 or 32.41 for the σ – finite
n
X case) and item 4. is a consequence of Theorem 19.55. The equivalence of Eqs.
µ(E) = µ(Ei ) (28.1) (28.2) and (28.3) requires a little comment. S∞
i=1 Suppose µ̄ is defined by Eq. (28.2) and A ⊂ n=1 En with En ∈ E and let
`∞
`n Ẽn := En \ (E1 ∪ · · · ∪ En−1 ) ∈ A (E) , where E0 := ∅. Then A ⊂ n=1 Ẽn
whenever E = i=1 Ei ∈ E with Ei ∈ E for i = 1, 2, . . . , n < ∞. If in `Nn
addition E = A is an algebra and µ (∅) = 0, then µ is a finitely additive and by Proposition 18.10 Ẽn = j=1 En,j for some En,j ∈ E. Therefore, A ⊂
`∞ `Nn
measure. n=1 j=1 En,j and hence
2. µ is σ – additive (or countable additive) on E if item 1. holds even
∞ X
Nn ∞ ∞
when n = ∞. If in addition E = A is an algebra and µ (∅) = 0, then µ is X X X
called a premeasure on A. µ̄(A) ≤ µ (En,j ) = µ Ẽn ≤ µ (En ) ,
n=1 j=1 n=1 n=1
3. µ is sub-additive (finitely sub-additive) on E if
n
X which easily implies the equality in Eq. (28.3).
µ(E) ≤ µ(Ei )
i=1
Example 28.3. The uniqueness assertion in item 4. of Theorem 28.2 may fail if
Sn the σ – finiteness assumption is dropped. For example, let X = R and A denote
whenever E = i=1 Ei ∈ E with n ∈ N∪ {∞} (n ∈ N). the algebra generated by
Theorem 28.2. Suppose that E ⊂ 2X is an elementary family (Definition E := {(a, b] ∩ R : −∞ ≤ a ≤ b ≤ ∞} .
18.8), A = A(E) is the algebra generated by E (see Proposition 18.10) and
µ : E → [0, ∞] is a function such that µ (∅) = 0. Then each of the following three distinct measures on BR restrict to the same
premeasure on A;
1. If µ is additive on E, then µ has a unique extension to a finitely additive
measure on A which will still be denoted by µ. 1. µ1 = ∞ except on the empty set,
304 28 Examples of Measures
2. µ2 is counting measure, and Proof. 1. The first item is an easy consequence of the third item in Theorem
3. µ3 (A) = µ2 (A ∩ D) where D is any dense subset of R. 28.2 with A = E.
2. Let Xm ∈ A such that µ̄ (Xm ) < ∞ and Xm ↑ X as n → ∞ and let
The next exercise is a minor variant of Remark 19.2 and Proposition 19.3. Bm := Xm ∩ B. Then by item 1., there exists Cm ∈ Aσ such that Bm ⊂ Cm
∞
and µ̄ (Cm \ Bm ) < ε2−(m+1) . So, letting C =
S
Exercise 28.1. Suppose µ : A → [0, ∞] is a finitely additive measure. Show Cm , C ∈ Aσ and
m=1
∞
1. µ is a premeasure on A iff µ (An ) ↑ µ(A) for all {An }n=1 ⊂ A such that ∞ ∞
An ↑ A ∈ A.
X X ε
µ̄ (C \ B) ≤ µ̄ (Cm \ B) ≤ µ̄ (Cm \ Bm ) < .
2. Further assume µ is finite (i.e. µ (X) < ∞). Then µ is a premeasure on A m=1 m=1
2
∞
iff µ(An ) ↓ 0 for all {An }n=1 ⊂ A such that An ↓ ∅.
Applying this result to B c implies there exists D ∈ Aσ such that B c ⊂ D and
28.1.1 Regularity and Density Results ε
µ̄ (B \ Dc ) = µ̄ (D \ B c ) < .
2
Definition 28.4. Given a collection of subsets, E, of X, let Eσ denote the col-
lection of subsets of X which are finite or countable unions of sets from E. Therefore if we let A := Dc ∈ Aδ, then A ⊂ B and µ̄ (B \ A) < ε/2 and
Similarly let Eδ denote the collection of subsets of X which are finite or count- therefore
able intersections of sets from E. We also write Eσδ = (Eσ )δ and Eδσ = (Eδ )σ , µ̄ (C \ A) = µ̄ (B \ A) + µ̄ (C \ B) < ε.
etc. 3. By item 2 there exist Am ⊂ B ⊂ Cm with Cm ∈ Aσ , Am ∈ Aδ and
∞
S ∞
T
Lemma 28.5. Suppose that A ⊂ 2X is an algebra. Then: µ̄ (Cm \ Am ) < 1/m for all m. Letting A := Am ∈ Aδσ and C := Cm ∈
m=1 m=1
1. Aσ is closed under taking countable unions and finite intersections. Aσδ , we have
2. Aδ is closed under taking countable intersections and finite unions. µ̄ (C \ A) ≤ µ̄ (Cm \ Am ) → 0 as m → ∞.
3. {Ac : A ∈ Aσ } = Aδ and {Ac : A ∈ Aδ } = Aσ .
Proof. By construction Aσ is closed under countable unions. Moreover if Remark 28.7. Using this result we may recover Corollary 22.18 and Theorem
A = ∪∞ ∞
i=1 Ai and B = ∪j=1 Bj with Ai , Bj ∈ A, then 22.14 which state, under the assumptions of Theorem 28.6;
A ∩ B = ∪∞
i,j=1 Ai ∩ Bj ∈ Aσ , 1. for every ε > 0 and B ∈ σ (A) such that µ̄(B) < ∞, there exists D ∈ A
such that µ̄(B4D) < ε.
which shows that Aσ is also closed under finite intersections. Item 3. is straight 2. Sf (A, µ) is dense in Lp (µ) for all 1 ≤ p < ∞.
forward and item 2. follows from items 1. and 3.
Indeed by Theorem 28.6 (also see Corollary 33.10), there exists C ∈ Aσ
Theorem 28.6 (Regularity Theorem). Suppose that µ is a σ – finite pre- such B ⊂ C and µ̄(C \ B) < ε. Now write C = ∪∞ n=1 Cn with Cn ∈ A for
measure on an algebra A, µ̄ is the extension described in Theorem 28.2 and each n. By replacing Cn by ∪nk=1 Ck ∈ A if necessary, we may assume that
B ∈ σ (A) . Then: Cn ↑ C as n → ∞. Since Cn \ B ↑ C \ B, B \ Cn ↓ B \ C = ∅ as n → ∞, and
µ̄(B \ C1 ) ≤ µ̄(B) < ∞, we know that
1.
µ̄ (B) := inf {µ̄ (C) : B ⊂ C ∈ Aσ } . lim µ̄(Cn \ B) = µ̄(C \ B) < ε and lim µ̄(B \ Cn ) = µ̄(B \ C) = 0
n→∞ n→∞
2. For any ε > 0 there exists A ⊂ B ⊂ C such that A ∈ Aδ , C ∈ Aσ and
µ̄(C \ A) < ε. Hence for n sufficiently large,
3. There exists A ⊂ B ⊂ C such that A ∈ Aδσ , C ∈ Aσδ and µ̄(C \ A) = 0.
µ̄(B4Cn ) = µ̄(Cn \ B) + µ̄(B \ Cn ) < ε.

28.2 The Riesz-Markov Theorem 305
Hence we are done with the first item by taking D = Cn ∈ A for an n sufficiently Example 28.11. Let X = R and τ = τd = 2X be the discrete topology on X.
large. Now let µ(A) = 0 if A is countable and µ(A) = ∞ otherwise. Since K ⊂ X is
For the second item, notice that compact iff # (K) < ∞, µ is a K-finite measure on X and
Z Z
|1B − 1D |p dµ = µ̄(B4D) < ε (28.4) Iµ (f ) = f dµ = 0 for all f ∈ Cc (X).
X X
from which it easily follows that any simple function in Sf (M, µ) may be ap- This shows that the correspondence µ → Iµ from K-finite measures to positive
proximated arbitrary well by an element from Sf (A, µ). This completes the linear functionals on Cc (X) is not injective without further restriction.
proof of item 2. since Sf (M, µ) is dense in Lp (µ) by Lemma 22.3.
Definition 28.12. Suppose that µ is a Borel measure on X and B ∈ BX . We
say µ is inner regular on B if
28.2 The Riesz-Markov Theorem µ(B) = sup{µ(K) : K @@ B} (28.5)
Now suppose that X is a locally compact Hausdorff space and B = BX is the and µ is outer regular on B if
Borel σ – algebra on X. Open subsets of Rd and locally compact separable
metric spaces are examples of such spaces, see Section 14.3. µ(B) = inf{µ(U ) : B ⊂ U ⊂o X}. (28.6)
Definition 28.8. A linear functional I on Cc (X) is positive if I(f ) ≥ 0 for The measure µ is said to be a regular Borel measure on X, if it is both inner
all f ∈ Cc (X, [0, ∞)). and outer regular on all Borel measurable subsets of X.
Proposition 28.9. If I is a positive linear functional on Cc (X) and K is a Definition 28.13. A measure µ : BX → [0, ∞] is a Radon measure on X
compact subset of X, then there exists CK < ∞ such that |I(f )| ≤ CK kf k∞ for if µ is a K-finite measure which is inner regular on all open subsets of X and
all f ∈ Cc (X) with supp(f ) ⊂ K. outer regular on all Borel subsets of X.
Proof. By Urysohn’s Lemma 15.8, there exits φ ∈ Cc (X, [0, 1]) such that The measure in Example 28.11 is an example of a K-finite measure on X
φ = 1 on K. Then for all f ∈ Cc (X, R) such that supp(f ) ⊂ K, |f | ≤ kf k∞ φ or which is not a Radon measure on X. BRUCE: Add exercise stating the sum of
equivalently kf k∞ φ ± f ≥ 0. Hence kf k∞ I(φ) ± I(f ) ≥ 0 or equivalently which two radon measures is still a radon measure. It is not true for countable sums
is to say |I(f )| ≤ kf k∞ I(φ). Letting CK := I(φ), we have shown that |I(f )| ≤ since this does not even preserve the K - finite condition.
CK kf k∞ for all f ∈ Cc (X, R) with supp(f ) ⊂ K. For general f ∈ Cc (X, C)
with supp(f ) ⊂ K, choose |α| = 1 such that αI(f ) ≥ 0. Then Example 28.14. If the topology on a set, X, is the discrete topology, then a
measure µ on BX is a Radon measure iff µ is of the form
|I(f )| = αI(f ) = I(α f ) = I(Re(αf )) ≤ CK k Re (αf ) k∞ ≤ CK kf k∞ . X
µ= µx δx (28.7)
x∈X
Example 28.10. If µ is a K-finite measure on X, then where µx ∈ [0, ∞) for all x ∈ X. To verify this first notice that BX = τX = 2X
Z and hence every measure on BX is necessarily outer regular on all subsets of X.
Iµ (f ) = f dµ ∀f ∈ Cc (X) The measure µ is K-finite iff µx := µ ({x}) < ∞ for all x ∈ X. If µ is a Radon
X measure, then for A ⊂ X we have, by inner regularity,
defines a positive linear functional on Cc (X). In the future, we will often simply (
X
)
X
write µ(f ) for Iµ (f ). µ(A) = sup {µ(Λ) : Λ ⊂⊂ A} = sup µx : Λ ⊂⊂ A = µx .
x∈Λ x∈A
The Riesz-Markov Theorem 28.16 below asserts that every positive linear
functional on Cc (X) comes from a K-finite measure µ. On the other hand if µ is given by Eq. (28.7) and A ⊂ X, then

( )
X X µ(U ) = µ0 (U ) = sup{Iµ (f ) : f ≺ U }. (28.13)
µ(A) = µx = sup µ(Λ) = µx : Λ ⊂⊂ A
x∈A x∈Λ If µ and ν are two Radon measures such that Iµ = Iν . Then by Eq. (28.13)
it follows that µ = ν on all open sets. Then by outer regularity, µ = ν on BX
showing µ is inner regular on all (open) subsets of X.
and this shows the map µ → Iµ is injective.
Recall from Definition 14.26 that if U is an open subset of X, we write f ≺ U Item 2. Let K ⊂ X be a compact set, then by monotonicity of the integral,
to mean that f ∈ Cc (X, [0, 1]) with supp(f ) := {f 6= 0} ⊂ U.
µ(K) ≤ inf{Iµ (f ) : f ∈ Cc (X) with f ≥ 1K }. (28.14)
Notation 28.15 Given a positive linear functional, I, on Cc (X) define µ = µI
on BX by To prove the reverse inequality, choose, by outer regularity, U ⊂o X such that
µ(U ) = sup{I(f ) : f ≺ U } (28.8) K ⊂ U and µ(U \ K) < ε. By Urysohn’s Lemma 15.8 there exists f ≺ U such
that f = 1 on K and hence,
for all U ⊂o X and then define Z Z
µ(B) = inf{µ(U ) : B ⊂ U and U is open}. (28.9) Iµ (f ) = f dµ = µ(K) + f dµ ≤ µ(K) + µ(U \ K) < µ(K) + ε.
X
U \K
Theorem 28.16 (Riesz-Markov Theorem). The map µ → Iµ taking Radon
measures on X to positive linear functionals on Cc (X) is bijective. Moreover Consequently,
if I is a positive linear functional on Cc (X), the function µ := µI defined in
Notation 28.15 has the following properties. inf{Iµ (f ) : f ∈ Cc (X) with f ≥ 1K } < µ(K) + ε
1. µ is a Radon measure on X and the map I → µI is the inverse to the map and because ε > 0 was arbitrary, the reverse inequality in Eq. (28.14) holds and
µ → Iµ . Eq. (28.10) is verified.
2. For all compact subsets K ⊂ X, Item 3. If f ∈ Cc (X), then
Z Z
µ(K) = inf{I(f ) : 1K ≤ f ≺ X}. (28.10) |Iµ (f )| ≤ |f | dµ = |f | dµ ≤ kf k∞ µ(supp(f )) ≤ kf k∞ µ(X)
X supp(f )
3. If kIµ k denotes the dual norm of I = Iµ on Cc (X, R)∗ , then kIk = µ(X). (28.15)
In particular, the linear functional, Iµ , is bounded iff µ(X) < ∞. and thus kIµ k ≤ µ(X). For the reverse inequality let K be a compact subset
of X and use Urysohn’s Lemma 15.8 again to find a function f ≺ X such that
Proof. The proof of the surjectivity of the map µ → Iµ and the assertion
f = 1 on K. By Eq. (28.12) we have
in item 1. is the content of Theorem 31.21 below.
Injectivity of µ → Iµ . Suppose that µ is a is a Radon measure on X. To
Z
each open subset U ⊂ X let µ(K) ≤ f dµ = Iµ (f ) ≤ kIµ k kf k∞ = kIµ k ,
X
µ0 (U ) := sup{Iµ (f ) : f ≺ U }. (28.11) which by the inner regularity of µ on open sets implies
It is evident that µ0 (U ) ≤ µ(U ) because f ≺ U implies f ≤ 1U . Given a µ(X) = sup{µ(K) : K @@ X} ≤ kIµ k .

compact subset K ⊂ U, Urysohn’s Lemma 15.8 implies there exists f ≺ U such
that f = 1 on K. Therefore,
Z
Example 28.17 (Discrete Version of Theorem 28.16). Suppose X is a set, τ =
µ(K) ≤ f dµ ≤ µ0 (U ) ≤ µ(U ) (28.12)
X 2X is the discrete topology on X and for x ∈ X, let ex ∈ Cc (X) be defined
by ex (y) = 1{x} (y). Let I be positive linear functional on Cc (X) and define a
By assumption µ is inner regular on open sets, and therefore taking the supre- Radon measure, µ, on X by
mum of Eq. (28.12) over compact subsets, K, of U shows

X
µ(A) := I(ex ) for all A ⊂ X. µ(K) = µ(F ) − µ(F ∩ V ) ≈ µ(U ) ≈ µ(A),
x∈A
or more formally,
Then for f ∈ Cc (X) (so f is a complex valued function on X supported on a
finite set), µ(K) = µ(F ) − µ(F ∩ V ) ≥ µ(U ) − ε − µ(F ∩ V )
! ≥ µ(U ) − 2ε ≥ µ(A) − 3ε,
Z X X
f dµ = f (x)I(ex ) = I f (x)ex = I(f ), we see that µ(A \ K) ≤ 3ε. This proves the proposition when µ(A) < ∞.
X x∈X x∈X If µ(A) = ∞ and there exists An ↑ A as n → ∞ with µ(An ) < ∞. Then by
the first part, there exist compact set Kn such that Kn ⊂ An and µ(An \ Kn ) <
so that I = Iµ . It is easy to see in this example that µ defined above is the 1/n or equivalently µ(Kn ) > µ(An ) − 1/n → ∞ as n → ∞.
unique regular radon measure on X such that I = Iµ while example Example
28.11 shows the uniqueness is lost if the regularity assumption is dropped. Corollary 28.19. Every σ-finite Radon measure, µ, is a regular Borel measure,
i.e. µ is both outer and inner regular on all Borel subsets.
28.2.1 Regularity Results For Radon Measures Notation 28.20 If (X, τ ) is a topological space, let Fσ denote the collection of
sets formed by taking countable unions of closed sets and Gδ = τδ denote the
Proposition 28.18. If µ is a Radon measure on X then µ is inner regular on collection of sets formed by taking countable intersections of open sets.
all σ-finite Borel sets.
Proposition 28.21. Suppose that µ is a σ -finite Radon measure and B ∈ B.
Proof. Suppose A ∈ BX and µ(A) < ∞ and ε > 0 is given. By outer Then
regularity of µ, there exist an open set U ⊂o X such that A ⊂ U and µ(U \A) <
ε. By inner regularity on open sets, there exists a compact set F @@ U such 1. For all ε > 0 there exists sets F ⊂ B ⊂ U with F closed, U open and
that µ(U \ F ) < ε. Again by outer regularity of µ, there exist V ⊂o X such µ(U \ F ) < ε.
that (U \ A) ⊂ V and µ(V ) < ε. Then K := F \ V is compact set and 2. There exists A ∈ Fσ and C ∈ Gδ such that A ⊂ B ⊂ C such that and
µ(C \ A) = 0.
K ⊂ F \ (U \ A) = F ∩ (U ∩ Ac )c = F ∩ (U c ∪ A) = F ∩ A,
Proof. 1. Let Xn ∈ B such that Xn ↑ X and µ (Xn ) < ∞ and choose
see Figure 28.1. Since, set Un such that B ∩ Xn ⊂ Un and µ (Un \ (B ∩ Xn )) < ε2−(n+1) . Then
open S
∞
U := n=1 Un is an open set such that
∞ ∞
X X ε
µ (U \ B) ≤ µ (Un \ B) ≤ µ (Un \ (B ∩ Xn )) < .
n=1 n=1
2
c
Applying this same result to B allows us to find a closed set F such that
B c ⊂ F c and
ε
µ (B \ F ) = µ (F c \ B c ) < .
2
Thus F ⊂ B ⊂ U and µ (U \ F ) < ε as desired.
2. This a simple consequence of item 1.
Theorem 28.22. Let X be a locally compact Hausdorff space such that every
open set V ⊂o X is σ – compact, i.e. there exists Kn @@ V such that V =
∪n Kn . Then any K-finite measure ν on X is a Radon measure and in fact is a
Fig. 28.1. Constructing the compact set K.
regular Borel measure. (The reader should check that if X is second countable,
then open sets are σ compact, see Exercise 14.8. In particular this condition
holds for Rn with the standard topology.)

Proof. By the Riesz-Markov Theorem 28.16, the positive linear functional, If B ∈ B = BX and ε > 0, by Proposition 28.21, there exists F ⊂ B ⊂ U
Z such that F is closed, U is open and µ(U \ F ) < ε. Since U \ F is open,
I (f ) := f dν for all f ∈ Cc (X), ν(U \ F ) = µ(U \ F ) < ε and therefore
X
ν (U ) − ε ≤ ν (B) ≤ ν (U ) and
may
R be represented by a Radon measure µ on (X, B) , i.e. such that I (f ) = µ (U ) − ε ≤ µ (B) ≤ µ (U ) .
f dµ for all f ∈ Cc (X). By Corollary 28.19, µ is also a regular Borel measure
X Since ν (U ) = µ (U ) , ν (B) = ∞ iff µ (B) = ∞ and if ν (B) < ∞ then
on (X, B) . So to finish the proof it suffices to show ν = µ. We will give two |ν (B) − µ (B)| < ε. Because ε > 0 is arbitrary, we may conclude that
proofs of this statement. ν (B) = µ (B) for all B ∈ B.
First Proof. The same arguments used in the proof of Lemma 18.57 shows
σ (Cc (X)) = BX . Let K be a compact subset of X and use Urysohn’s Lemma Proposition 28.23 (Density of Cc (X) in Lp (µ)). If µ is a Radon measure
15.8 to find ϕ ≺ X such that ϕ ≥ 1K . By a simple application of the multi- on X, then Cc (X) is dense in Lp (µ) for all 1 ≤ p < ∞.
plicative system Theorem 18.51 one shows
Proof. Let ε > 0 and B ∈ BX with µ(B) < ∞. By Proposition 28.18, there
exists K @@ B ⊂ U ⊂o X such that µ(U \ K) < εp and by Urysohn’s Lemma
Z Z
ϕf dν = ϕf dµ
15.8, there exists f ≺ U such that f = 1 on K. This function f satisfies
X X
Z Z
p p p
for all bounded BX = σ (Cc (X)) – measurable functions on X. Taking f = 1K kf − 1B kp = |f − 1B | dµ ≤ |f − 1B | dµ ≤ µ(U \ K) < εp .
X U \K
then shows that ν (K) = µ (K) with K @@ X. An application of Theorem 19.55
implies µ = ν on σ – algebra generated by the compact sets. This completes From this it easy to conclude that Cc (X) is dense in Sf (B, µ) – the simple
the proof, since, by assumption, this σ – algebra contains all of the open sets functions on X which are in L1 (µ) . Combining this with Lemma 22.3 which
and hence is the Borel σ – algebra. asserts that Sf (B, µ) is dense in Lp (µ) completes the proof of the theorem.
Second Proof. Since µ is a Radon measure on X, it follows from Eq.
(28.13), that Theorem 28.24 (Lusin’s Theorem). Suppose (X, τ ) is a locally compact
    Hausdorff space, BX is the Borel σ – algebra on X, and µ is a Radon mea-
Z  Z  sure on (X, BX ) . Also let ε > 0 be given. If f : X → C is a measurable function
µ(U ) = sup f dµ : f ≺ U = sup f dν : f ≺ U ≤ ν (U ) (28.16) such that µ(f 6= 0) < ∞, there exists a compact set K ⊂ {f 6= 0} such that f |K
   
X X is continuous and µ({f 6= 0} \ K) < ε. Moreover there exists φ ∈ Cc (X) such
that µ(f 6= φ) < ε and if f is bounded the function φ may be chosen so that
for all open subsets U of X. For each compact subset K ⊂ U, there exists, by
Uryshon’s Lemma 15.8, a function f ≺ U such that f ≥ 1K . Thus kφku ≤ kf ku := sup |f (x)| .
Z Z x∈X
ν(K) ≤ f dν = f dµ ≤ µ (U ) . (28.17) Proof. Suppose first that f is bounded, in which case

X
X Z
Combining Eqs. (28.16) and (28.17) implies ν(K) ≤ µ(U ) ≤ ν(U ). By assump- |f | dµ ≤ kf ku µ(f 6= 0) < ∞.
X
tion there exists compact sets, Kn ⊂ U, such that Kn ↑ U as n → ∞ and
therefore by continuity of ν, By Proposition 28.23, there exists fn ∈ Cc (X) such that fn → f in L1 (µ) as
n → ∞. By passing to a subsequence if necessary, we may assume kf − fn k1 <
ν(U ) = lim ν(Kn ) ≤ µ(U ) ≤ ν(U ). εn−1 2−n and hence by Chebyshev’s inequality (Lemma 19.17),
n→∞
µ |f − fn | > n−1 < ε2−n for all n.

Hence we have shown, ν (U ) = µ (U ) for all U ∈ τ.

Let E := ∪∞ −1

n=1 |f − fn | > n , so that µ(E) < ε. On E c , |f − fn | ≤ 1/n, i.e. The following result is a slight generalization of Lemma 22.11.
c
fn → f uniformly on E and hence f |E c is continuous. By Proposition 28.18,
there exists a compact set K and open set V such that Corollary 28.26. Let (X, τ ) be a locally compact Hausdorff space, µ : BX →
[0, ∞] be a Radon measure on X and h ∈ L1loc (µ). If
K ⊂ {f 6= 0} \ E ⊂ V Z
f hdµ = 0 for all f ∈ Cc (X) (28.18)
such that µ(V \ K) < ε. Notice that X
µ({f 6= 0} \ K) = µ (({f 6= 0} \ K) \ E) + µ (({f 6= 0} \ K) ∩ E) then 1K h = 0 for µ – a.e. for every compact subset K ⊂ X. (BRUCE: either
≤ µ(V \ K) + µ(E) < 2ε. show h = 0 a.e. or give a counterexample. Also, either prove or give a coun-
terexample to the question to the statement the dν = ρdµ is a Radon measure
By the Tietze extension Theorem 15.9, there exists F ∈ C(X) such that if ρ ≥ 0 and in L1loc (µ) .)
f = F |K . By Urysohn’s Lemma 15.8 there exists ψ ≺ V such that ψ = 1 on
K. So letting φ = ψF ∈ Cc (X), we have φ = f on K, kφk∞ ≤ kf k∞ and Proof. By considering the real and imaginary parts of h we may assume
since {φ 6= f } ⊂ E ∪ (V \ K), µ(φ 6= f ) < 3ε. This proves the assertions in the with out loss of generality that h is real valued. Let K be a compact subset of
theorem when f is bounded. X. Then 1K sgn(h̄) ∈ L1 (µ) and by Proposition 28.23, there exists fn ∈ Cc (X)
Suppose that f : X → C is (possibly) unbounded and ε > 0 is given. Then such that limn→∞ kfn − 1K sgn(h)kL1 (µ) = 0. Let φ ∈ Cc (X, [0, 1]) such that
BN := {0 < |f | ≤ N } ↑ {f 6= 0} as N → ∞ and therefore for all N sufficiently φ = 1 on K and gn = φ min (−1, max (1, fn )) . Since
large,
µ ({f 6= 0} \ BN ) < ε/3. |gn − 1K sgn(h)| ≤ |fn − 1K sgn(h)|
Since µ is a Radon measure, Proposition 28.18, guarantee’s there is a compact we have found gn ∈ Cc (X, R) such that |gn | ≤ 1supp(φ) and gn → 1K sgn(h) in
set C ⊂ {f 6= 0} such that µ ({f 6= 0} \ C) < ε/3. Therefore, L1 (µ) . By passing to a sub-sequence if necessary we may assume the conver-
gence happens µ – almost everywhere. Using Eq. (28.18) and the dominated
µ ({f 6= 0} \ (BN ∩ C)) < 2ε/3.
convergence theorem (the dominating function is |h| 1supp(φ) ) we conclude that
We may now apply the bounded case already proved to the function 1BN ∩C f Z Z Z
to find a compact subset K and an open set V such that K ⊂ V, 0 = lim gn hdµ = 1K sgn(h̄)hdµ = |h| dµ
n→∞ X X K
K ⊂ {1BN ∩C f 6= 0} = BN ∩ C ∩ {f 6= 0}
which shows h (x) = 0 for µ - a.e. x ∈ K.
such thatµ ((BN ∩ C ∩ {f 6= 0}) \ K) < ε/3 and φ ∈ Cc (X) such that φ =
1BN ∩C f = f on K. This completes the proof, since 28.2.2 The dual of C0 (X)
µ ({f 6= 0} \ K) ≤ µ ((BN ∩ C ∩ {f 6= 0}) \ K) + µ ({f 6= 0} \ (BN ∩ C)) < ε
Definition 28.27. Let (X, τ ) be a locally compact Hausdorff space and B =
which implies µ (f 6= φ) < ε. σ (τ ) be the Borel σ – algebra. A signed Radon measure is a signed measure
µ on B such that the measures, µ± , in the Jordan decomposition of µ are both
Example 28.25. To illustrate Theorem 28.24, suppose that X = (0, 1), µ = m is Radon measures. A complex Radon measure is a complex measure µ on B
Lebesgue measure and f = 1(0,1)∩Q . Then Lusin’s theorem asserts for any ε > 0 such that Re µ and Im µ are signed radon measures.
there exists a compact set K ⊂ (0, 1) such that m((0, 1) \ K) < ε and f |K is
∞
continuous. To see this directly, let {rn }n=1 be an enumeration of the rationals Example 28.28. Every complex measure µ on BRd is a Radon measure. BRUCE:
in (0, 1), add some more examples and perhaps some exercises here.
Jn = (rn − ε2−n , rn + ε2−n ) ∩ (0, 1) and W = ∪∞
n=1 Jn . BRUCE: Compare and combine with results from Section 32.10.
Then W is an open subset of X and µ(W ) < ε. Therefore Kn := [1/n, 1 − 1/n] \ Proposition 28.29. Suppose (X, τ ) is a topological space and I ∈ C0 (X, R)∗ .
W is a compact subset of X and m(X \ Kn ) ≤ n2 + µ(W ). Taking n sufficiently Then we may write I = I+ − I− where I± ∈ C0 (X, R)∗ are positive linear
large we have m(X \ Kn ) < ε and f |Kn ≡ 0 which is of course continuous. functionals.

Proof. For f ∈ C0 (X, [0, ∞)), let Remark 28.30. The above proof works for functionals on linear spaces of
I+ (f ) := sup {I(g) : g ∈ C0 (X, [0, ∞)) and g ≤ f }

bounded functionsR 1 which are closed under taking f ∧ g and f ∨ g. As an exam-
ple, let λ(f ) = 0 f (x)dx for all bounded measurable functions f : [0, 1] → R.
By the Hahn Banach theorem, we may extend λ to a linear functional Λ on all
and notice that |I+ (f )| ≤ kIk kf k . If c > 0, then I+ (cf ) = cI+ (f ). Suppose
bounded functions on [0, 1] in such a way that kΛk = 1. Let Λ+ be as above, then
that f1 , f2 ∈ C0 (X, [0, ∞)) and gi ∈ C0 (X, [0, ∞)) such that gi ≤ fi , then
Λ+ = λ on bounded measurable functions and kΛ+ k = 1. Define µ(A) := Λ(1A )
g1 + g2 ≤ f1 + f2 so that
for all A ⊂ [0, 1] and notice that if A is measurable, the µ(A) = m(A). So µ is
I(g1 ) + I(g2 ) = I(g1 + g2 ) ≤ I+ (f1 + f2 ) a finitely additive extension of m to all subsets of [0, 1].
and therefore Exercise 28.2. Suppose that µ is a signed Radon measure and I = Iµ . Let
I+ (f1 ) + I+ (f2 ) ≤ I+ (f1 + f2 ). (28.19) µ+ and µ− be the Radon measures associated to I± with I± being constructed
as in the proof of Proposition 28.29. Show that µ = µ+ − µ− is the Jordan
Moreover, if g ∈ C0 (X, [0, ∞)) and g ≤ f1 + f2 , let g1 = min(f1 , g), so that decomposition of µ.
0 ≤ g2 := g − g1 ≤ f1 − g1 + f2 ≤ f2 . Solution to Exercise (28.2). Let X = P ∪ P c where P is a positive set for
µ and P c is a negative set. Then for A ∈ BX ,
Hence I(g) = I(g1 ) + I(g2 ) ≤ I+ (f1 ) + I+ (f2 ) for all such g and therefore,
µ(P ∩ A) = µ+ (P ∩ A) − µ− (P ∩ A) ≤ µ+ (P ∩ A) ≤ µ+ (A). (28.21)
I+ (f1 + f2 ) ≤ I+ (f1 ) + I+ (f2 ). (28.20)
To finish the proof we need only prove the reverse inequality. To this end let
Combining Eqs. (28.19) and (28.20) shows that I+ (f1 + f2 ) = I+ (f1 ) + I+ (f2 ).
ε > 0 and choose K @@ P ∩ A ⊂ U ⊂o X such that |µ| (U \ K) < ε. Let
For general f ∈ C0 (X, R), let I+ (f ) = I+ (f+ ) − I+ (f− ) where f+ = max(f, 0)
f, g ∈ Cc (U, [0, 1]) with f ≤ g, then
and f− = − min(f, 0). (Notice that f = f+ − f− .) If f = h − g with h, g ∈
C0 (X, R), then g + f+ = h + f− and therefore, I(f ) = µ(f ) = µ(f : K) + µ(f : U \ K) ≤ µ(g : K) + O (ε)
I+ (g) + I+ (f+ ) = I+ (h) + I+ (f− ) ≤ µ(K) + O (ε) ≤ µ(P ∩ A) + O (ε) .
and hence I+ (f ) = I+ (h) − I+ (g). In particular, Taking the supremum over all such f ≤ g, we learn that I+ (g) ≤ µ(P ∩A)+O (ε)
and then taking the supremum over all such g shows that
I+ (−f ) = I+ (f− − f+ ) = I+ (f− ) − I+ (f+ ) = −I+ (f )
µ+ (U ) ≤ µ(P ∩ A) + O (ε) .
so that I+ (cf ) = cI+ (f ) for all c ∈ R. Also,
Taking the infimum over all U ⊂o X such that P ∩ A ⊂ U shows that
I+ (f + g) = I+ (f+ + g+ − (f− + g− )) = I+ (f+ + g+ ) − I+ (f− + g− )
= I+ (f+ ) + I+ (g+ ) − I+ (f− ) − I+ (g− ) µ+ (P ∩ A) ≤ µ(P ∩ A) + O (ε) (28.22)
= I+ (f ) + I+ (g).
From Eqs. (28.21) and (28.22) it follows that µ(P ∩ A) = µ+ (P ∩ A). Since
Therefore I+ is linear. Moreover,
I− (f ) = sup I(g) − I(f ) = sup I(g − f ) = sup −I(f − g) = sup −I(h)
0≤g≤f 0≤g≤f 0≤g≤f 0≤h≤f
|I+ (f )| ≤ max (|I+ (f+ )| , |I+ (f− )|) ≤ kIk max (kf+ k , kf− k) = kIk kf k
the same argument applied to −I shows that
which shows that kI+ k ≤ kIk . Let I− = I+ − I ∈ C0 (X, R)∗ , then for f ≥ 0,
−µ(P c ∩ A) = µ− (P c ∩ A).
I− (f ) = I+ (f ) − I(f ) ≥ 0
Since
by definition of I+ , so I− ≥ 0 as well.

28.3 Classifying Radon Measures on R 311
µ(A) = µ(P ∩ A) + µ(P c ∩ A) = µ+ (P ∩ A) − µ− (P c ∩ A) and Exercise 28.3. Let (X, τ ) be a compact Hausdorff P space which supports a
µ(A) = µ+ (A) − µ− (A) positive measure ν on B = σ (τ ) such that ν (X) 6= x∈X ν ({x}) , i.e. ν is a
not a counting type measure. (Example X = [0, 1] .) Then C (X) is not reflexive.
it follows that ∗
Hint: recall that C (X) is isomorphic to the space of complex Radon mea-
µ+ (A \ P ) = µ− (A \ P c ) = µ− (A ∩ P ). ∗∗
sures on (X, B) . Using this isomorphism, define λ ∈ C (X) by
Taking A = P then shows that µ− (P ) = 0 and taking A = P c shows that X
µ+ (P c ) = 0 and hence λ (µ) = µ ({x})
x∈X
µ(P ∩ A) = µ+ (P ∩ A) = µ+ (A) and
−µ(P c ∩ A) = µ− (P c ∩ A) = µ− (A) and then show λ 6= fˆ for any f ∈ C (X) .
as was to be proved. Solution to Exercise (28.3). Suppose there exists f ∈ C (X) such that
λ (µ) = fˆ (µ) = µ (f ) = X f dµ for all complex Radon measures, µ. Taking
R
Theorem 28.31. Let X be a locally compact Hausdorff space, M (X) be the
space of complex Radon measures on X and for µ ∈ M (X) let kµk = |µ|(X). µ = δx with x ∈ X then implies
Then the map X
f (x) = µ (δx ) = λ (δx ) = δx ({y}) = 1.
µ ∈ M (X) → Iµ ∈ C0 (X)∗
R y∈X
is an isometric isomorphism. Here again Iµ (f ) := X f dµ.
This shows f ≡ 1. However, this f can not work since
Proof. To show that the map M (X) → C0 (X)∗ is surjective, let I ∈ C0 (X)∗
and then write I = I re + iI im be the decomposition into real and imaginary
X
fˆ (ν) = ν (X) 6= ν ({x}) = λ (ν) .
parts. Then further decompose these into there plus and minus parts so x∈X
re re im im

I = I+ − I− + i I+ − I−
and let µre im
± and µ± be the corresponding positive Radon measures associated 28.3 Classifying Radon Measures on R
re im
to I± and I± . Then I = Iµ where
Throughout this section, let X = R, E be the elementary class
µ = µre re im im

+ − µ− + i µ+ − µ− .
To finish the proof it suffices to show kIµ kC0 (X)∗ = kµk = |µ|(X). We have E = {(a, b] ∩ R : −∞ ≤ a ≤ b ≤ ∞}, (28.23)
Z

and A = A(E) be the algebra formed by taking finite disjoint unions of elements
kIµ kC0 (X)∗ = sup f dµ : f ∈ C0 (X) 3 kf k∞ ≤ 1 from E, see Proposition 18.10. The aim of this section is to prove Theorem 19.8
ZX which we restate here for convenience.

≤ sup f dµ : f measurable and kf k∞ ≤ 1 = kµk .
X Theorem 28.32. The collection of K-finite measure on (R, BR ) are in one
to one correspondence with a right continuous non-decreasing functions, F :
To prove the opposite inequality, write dµ = gd |µ| with g a complex measurable
R → R, with F (0) = 0. The correspondence is as follows. If F is a right contin-
function such that |g| = 1. By Proposition 28.23, there exist fn ∈ Cc (X) such
uous non-decreasing function F : R → R, then there exists a unique measure,
that fn → g in L1 (|µ|) as n → ∞. Let gn = φ(fn ) where φ : C → C is the
µF , on BR such that
continuous function defined by φ(z) = z if |z| ≤ 1 and φ(z) = z/ |z| if |z| ≥ 1.
Then |gn | ≤ 1 and gn → g in L1 (µ). Thus µF ((a, b]) = F (b) − F (a) ∀ − ∞ < a ≤ b < ∞
Z Z Z
kµk = |µ| (X) = d |µ| = ḡdµ = lim ḡn dµ ≤ kIµ kC0 (X)∗ . and this measure may be defined by
X X n→∞ X

(∞ )
X Moreover, one easily checks using the additivity of µ that Eq. (28.26) holds
µF (A) = inf (F (bi ) − F (ai )) : A ⊂ ∪∞
i=1 (ai , bi ] for this F. Conversely, suppose F : R̄ → R̄ is an increasing function such that
i=1
(∞ F −1 ({−∞}) ⊂ {−∞}, F −1 ({∞}) ⊂ {∞}. Define µ on E using the formula in
∞
)
X a Eq. (28.26). The argument will be completed by showing µ is additive on E
= inf (F (bi ) − F (ai )) : A ⊂ (ai , bi ] (28.24)
and hence, by Proposition 31.3, has a unique extension to a finitely additive
i=1 i=1
measure on A. Suppose that
for allA ∈ BR . Conversely if µ is K-finite measure on (R, BR ) , then
n
a
(a, b] = (ai , bi ].

−µ((x, 0]) if x ≤ 0
F (x) := (28.25) i=1
µ((0, x]) if x ≥ 0
is a right continuous non-decreasing function and this map is the inverse to the By reordering (ai , bi ] if necessary, we may assume that
map, F → µF .
a = a1 < b1 = a2 < b2 = a3 < · · · < bn−1 = an < bn = b.
There are three aspects to this theorem; namely the existence of the map
F → µF , the surjectivity of the map and the injectivity of this map. Assuming Therefore, by the telescoping series argument,
the map F → µF exists, the surjectivity follows from Eq. (28.25) and the n
X n
X
injectivity is an easy consequence of Theorem 19.55. The rest of this section is µ((a, b]) = F (b) − F (a) = [F (bi ) − F (ai )] = µ((ai , bi ]).
devoted to giving two proofs for the existence of the map F → µF . i=1 i=1
Exercise 28.4. Show by direct means any measure µ = µF satisfying Eq.

(28.24) is outer regular on all Borel sets. Hint: it suffices to show if B := Now let F : R → R be an increasing function, F (±∞) := limx→±∞ F (x)
`∞
i=1 i , bi ], then there exists V ⊂o R such that µ (V \ B) is as small as you
(a and µ = µF be the finitely additive measure on (R, A) described in Proposition
please. 28.33. If µ happens to be a premeasure on A, then, letting An = (a, bn ] with
bn ↓ b as n → ∞, implies
28.3.1 Classifying Radon Measures on R using Theorem 28.2
F (bn ) − F (a) = µ((a, bn ]) ↓ µ((a, b]) = F (b) − F (a).
Proposition 28.33. To each finitely additive measures µ : A → [0, ∞] which ∞
Since {bn }n=1 was an arbitrary sequence such that bn ↓ b, we have shown
is finite on bounded sets there is a unique increasing function F : R̄ → R̄ such
limy↓b F (y) = F (b), i.e. F is right continuous. The next proposition shows the
that F (0) = 0, F −1 ({−∞}) ⊂ {−∞} , F −1 ({∞}) ⊂ {∞} and
converse is true as well. Hence premeasures on A which are finite on bounded
µ((a, b] ∩ R) = F (b) − F (a) ∀ a ≤ b in R̄. (28.26) sets are in one to one correspondences with right continuous increasing functions
which vanish at 0.
Conversely, given an increasing function F : R̄ → R̄ such that F −1 ({−∞}) ⊂
{−∞} and F −1 ({∞}) ⊂ {∞} , there is a unique finitely additive measure µ = Proposition 28.34. To each right continuous increasing function F : R → R
µF on A such that the relation in Eq. (28.26) holds. there exists a unique premeasure µ = µF on A such that
Proof. If F is going to exist, then µF ((a, b]) = F (b) − F (a) ∀ − ∞ < a < b < ∞.
µ((0, b] ∩ R) = F (b) − F (0) = F (b) if b ∈ [0, ∞], Proof. As above, let F (±∞) := limx→±∞ F (x) and µ = µF be as in Propo-
µ((a, 0]) = F (0) − F (a) = −F (a) if a ∈ [−∞, 0] sition 28.33. Because of Proposition 31.5, to finish the proof it suffices to show
µ is sub-additive on E.
from which we learn First suppose that −∞ < a < b < ∞, J = (a, b], Jn = (an , bn ] such that
∞
−µ((x, 0]) if x ≤ 0
`
J= Jn . We wish to show
F (x) = n=1
µ((0, x] ∩ R) if x ≥ 0.

28.3 Classifying Radon Measures on R 313
∞
X ∞
a
µ(J) ≤ µ(Jn ). (28.27) IM := (a, M ] = J ∩ IM = Jn ∩ IM
n=1 n=1
To do this choose numbers ã > a, b̃n > bn in which case I := (ã, b] ⊂ J, and so by what we have already proved,
∞ ∞
Jñ := (an , b̃n ] ⊃ Jño := (an , b̃n ) ⊃ Jn .
X X
F (M ) − F (a) = µ(IM ) ≤ µ(Jn ∩ IM ) ≤ µ(Jn ).
∞ n=1 n=1
Since I¯ = [a, b] is compact and I¯ ⊂ J ⊂ Jño there exists N < ∞ such that
S
n=1
Now let M → ∞ in this last inequality to find that
∞
N N X
[ [ µ((a, ∞)) = F (∞) − F (a) ≤ µ(Jn ).
I ⊂ I¯ ⊂ Jño ⊂ Jñ .
n=1
n=1 n=1
The other cases where a = −∞ and b ∈ R and a = −∞ and b = ∞ are handled
Hence by finite sub-additivity of µ,
similarly.
N ∞
X X Corollary 28.35. The map F → µF in Theorem 28.32 exists.
F (b) − F (ã) = µ(I) ≤ µ(Jñ ) ≤ µ(Jñ ).
n=1 n=1 Proof. This is simply a combination of Proposition 28.34 and Theorem 28.2.
Using the right continuity of F and letting ã ↓ a in the above inequality,
∞
X 28.3.2 Classifying Radon Measures on R using the Riesz-Markov
µ (J) = µ((a, b]) = F (b) − F (a) ≤ µ Jñ
Theorem 28.16
n=1
∞ ∞
X X For the moment let X be an arbitrary set. We are going to start by introducing
= µ (Jn ) + µ(Jñ \ Jn ). (28.28) a simple integral associated to an additive measure, µ, on an algebra A ⊂ 2X .
n=1 n=1
Definition 28.36. Let µ be a finitely additive measure on an algebra A ⊂ 2X ,
Given ε > 0, we may use the right continuity of F to choose b̃n so that S = Sf (A, µ) be the collection of simple functions defined in Notation 22.1 and
for f ∈ S defined the integral I(f ) = Iµ (f ) by
µ(Jñ \ Jn ) = F (b̃n ) − F (bn ) ≤ ε2−n ∀ n ∈ N.
X
Using this in Eq. (28.28) shows Iµ (f ) = yµ(f = y). (28.29)
y∈R
∞
The same proof used for Proposition 19.12 shows Iµ : S → R is linear and
X
µ(J) = µ((a, b]) ≤ µ (Jn ) + ε
n=1
positive, i.e. I(f ) ≥ 0 if f ≥ 0. Taking absolute values of Eq. (28.29) gives
X
which verifies Eq. (28.27) since ε > 0 was arbitrary. |I(f )| ≤ |y| µ(f = y) ≤ kf k∞ µ(f 6= 0) (28.30)
The hard work is now done but we still have to check the cases where y∈R
a = −∞ or b = ∞. For example, suppose that b = ∞ so that
where kf k∞ = supx∈X |f (x)| . For A ∈ A, let SA := {f ∈ S : {f 6= 0} ⊂ A}.
∞
a The estimate in Eq. (28.30) implies
J = (a, ∞) = Jn
n=1 |I(f )| ≤ µ(A) kf k∞ for all f ∈ SA . (28.31)
with Jn = (an , bn ] ∩ R. Then Let S̄A denote the closure of SA inside `∞ (X, R).

∞ k −1
nX
Proposition 28.37. Let (A, µ, I = Iµ ) be as in Definition 28.36, then we may
Z
min f (x) : akl ≤ x ≤ akl+1 F (akl+1 ) − F (akl ) .

extend I to f dF = lim
[ −∞ k→∞
l=0
S̃ := {S̄A : A ∈ A with µ(A) < ∞}
(28.33)
by defining I(f ) = IA (f ) when f ∈ S̄A with µ(A) < ∞. Moreover this extension
is still positive. The following Exercise is an abstraction of Lemma 28.38.
Proof. Because of Eq. (28.31) and the B.L.T. Theorem 10.4, I has a unique Exercise 28.5. Continue the notation of Definition 28.36 and Proposition
extension IA to S̄A ⊂ `∞ (X, R) for any A ∈ A such that µ(A) < ∞. The 28.37. Further assume that X is a metric space, there exists open sets Xn ⊂o X
extension IA is still positive. Indeed, let f ∈ S̄A with f ≥ 0 and let fn ∈ SA be such that Xn ↑ X and for each n ∈ N and δ > 0 there exists a finite collection
k
a sequence such that kf − fn k∞ → 0 as n → ∞. Then fn ∨ 0 ∈ SA and of sets {Ai }i=1 ⊂ A such that diam(Ai ) < δ, µ(Ai ) < ∞ and Xn ⊂ ∪ki=1 Ai .
Then Cc (X, R) ⊂ S̃ and so I is well defined on Cc (X, R).
kf − fn ∨ 0k∞ ≤ kf − fn k∞ → 0 as n → ∞.
Therefore, IA (f ) = limn→∞ IA (fn ∨ 0) ≥ 0. 28.3.3 The Lebesgue-Stieljtes Integral

Now suppose that A, B ∈ A are sets such that µ(A) + µ(B) < ∞. Then
Notation 28.39 Given an increasing function F : R → R, let F (x−) =
SA ∪ SB ⊂ SA∪B and so S̄A ∪ S̄B ⊂ S̄A∪B . Therefore IA (f ) = IA∪B (f ) = IB (f )
limy↑x F (y), F (x+) = limy↓x F (y) and F (±∞) = limx→±∞ F (x) ∈ R̄. Since
for all f ∈ S̄A ∩ S̄B . Therefore I(f ) := IA (f ) for f ∈ S̄A is well defined.
F is increasing all of theses limits exists.
We now specialize the previous results to the case where X = R, A = A(E)
with E as in Eq. (28.23), and F and µ are R ∞as in Proposition
R∞ 28.33. In this Theorem 28.40. If F : R → R is an increasing function (not necessarily right
setting, for f ∈ S̃, we will write Iµ (f ) as −∞ f dF or −∞ f (x)dF (x) and to continuous), there exists a unique measure µ = µF on BR such that
this integral as the Riemann Stieljtes integral of f relative to F. Z ∞ Z
R∞ f dF = f dµ for all f ∈ Cc (R, R), (28.34)
Lemma 28.38. Using the notation above, the map f ∈ S̃ → −∞ f dF is linear,
−∞ R
positive and satisfies the estimate
R∞
Z ∞

where −∞ f dF is as in Lemma 28.38 above. This measure may also be char-

f dF ≤ (F (b) − F (a)) kf k∞ (28.32) acterized as the unique measure on BR such that
−∞
µ ((a, b]) = F (b+) − F (a+) for all − ∞ < a < b < ∞. (28.35)
if supp(f ) ⊂ (a, b). Moreover Cc (R, R) ⊂ S̃.
Moreover, if A ∈ BR then
Proof. The only new point of the lemma is to prove Cc (R, R) ⊂ S̃, the
(∞ )
remaining assertions follow directly from Proposition 28.37. The fact that X
∞
Cc (R, R) ⊂ S̃ has essentially already been done in Example 19.23. In more µF (A) = inf (F (bi +) − F (ai +)) : A ⊂ ∪i=1 (ai , bi ]
detail, let f ∈ Cc (R, R) and choose a < b such that supp(f ) ⊂ (a, b). Then i=1
(∞ ∞
)
define fk ∈ S as in Example 19.23, i.e. X a
= inf (F (bi +) − F (ai +)) : A ⊂ (ai , bi ] . (28.36)
k −1
nX i=1 i=1
min f (x) : akl ≤ x ≤ akl+1 1(akl ,akl+1 ] (x)

fk (x) =
l=0
Proof. An application of the Riesz-Markov Theorem 28.16 implies there
exists a unique measure µ on BR such Eq. (28.34) is valid. Let −∞ < a < b < ∞,
where πk = {a = ak0 < ak1 < · · · < aknk = b}, for k = 1, 2, 3, . . . , is a sequence of ε > 0 be small and ϕε (x) be the function defined in Figure 28.2, i.e. ϕε is one on
refining partitions such that mesh(πk ) → 0 as k → ∞. Since supp(f ) is compact [a+2ε, b+ε], linearly interpolates to zero on [b+ε, b+2ε] and on [a+ε, a+2ε] and
and f is continuous, f is uniformly continuous on R. Therefore kf − fk k∞ → 0 is zero on (a, b + 2ε)c . Since ϕε → 1(a,b] it follows by the dominated convergence
as k → ∞, showing f ∈ S̃. Incidentally, for f ∈ Cc (R, R), it follows that theorem that

28.4 Kolmogorov’s Existence of Measure on Products Spaces 315
µ0 (A) ≤ inf {µ (V ) : A ⊂ V ⊂o R} = µ (A) .
Combining this with Eq. (28.39) shows µ0 (A) = µ (A) which is precisely Eq.
(28.36).
Corollary 28.41. The map F → µF is a one to one correspondence between

right continuous non-decreasing functions F such that F (0) = 0 and Radon
(same as K - finite) measures on (R, BR ) .
Fig. 28.2. .
28.4 Kolmogorov’s Existence of Measure on Products
Spaces
Z Z
µ((a, b]) = lim ϕε dµ = lim ϕε dF. (28.37) Throughout this section, let {(XQ α , τα )}α∈A be second countable locally compact
ε↓0 R ε↓0 R
Hausdorff spaces and let X := Xα be equipped with the product topology,
On the other hand we have α∈A Q
τ := ⊗α∈A τα . More generally for Λ ⊂ A, let XΛ := α∈Λ Xα and τΛ := ⊗α∈Λ τα
1(a+2ε,b+ε] ≤ ϕε ≤ 1(a+ε,b+2ε] , (28.38) and Λ ⊂ Γ ⊂ A, let πΛ,Γ : XΓ → XΛ be the projection map; πΛ,Γ (x) = x|Λ for
and therefore applying IF to both sides of Eq. (28.38) shows; x ∈ XΓ . We will simply write πΛ for πΛ,A : X → XΛ . (Notice that if Λ is a finite
Z subset of A then (XΛ , τΛ ) is still second countable as the reader should verify.)
Let M = ⊗α∈A Bα be the product σ – algebra on X = XA and BΛ = σ (τΛ ) be
F (b + ε) − F (a + 2ε) = 1(a+2ε,b+ε] dF
the Borel σ – algebra on XΛ .
ZR
≤ ϕε dF Theorem 28.42 (Kolmogorov’s Existence Theorem). Suppose
ZR {µΛ : Λ ⊂⊂ A} are probability measures on (XΛ , BΛ ) satisfying the following
≤ 1(a+ε,b+2ε] dF = F (b + 2ε) − F (a + ε). compatibility condition:
R
• (πΛ,Γ )∗ µΓ = µΛ whenever Λ ⊂ Γ ⊂⊂ A.
Letting ε ↓ 0 in this equation and using Eq. (28.37) shows
Then there exists a unique probability measure, µ, on (X, M) such that
F (b+) − F (a+) ≤ µ((a, b]) ≤ F (b+) − F (a+).
(πΛ )∗ µ = µΛ whenever Λ ⊂⊂ A. Recall, see Exercise 19.8, that the condition
For the last assertion let (πΛ )∗ µ = µΛ is equivalent to the statement;
(∞ ∞
) Z Z
X a
0
µ (A) = inf (F (bi ) − F (ai )) : A ⊂ (ai , bi ] F (πΛ (x))dµ(x) = F (y)dµΛ (y) (28.40)
i=1 i=1 X
XΛ
= inf {µ (B) : A ⊂ B ∈ Aσ } ,
for all Λ ⊂⊂ A and F : XΛ → R bounded a measurable.
where A is the algebra generated by the half open intervals on R. By mono-
tonicity of µ, it follows that We will first prove the theorem in the following special case. The full proof
will be given after Exercise 28.6 below.
µ0 (A) ≥ µ (A) for all A ∈ B. (28.39)
Theorem 28.43. Theorem 28.42 holds under the additional assumption that
For the `reverse inequality, let A ⊂ V ⊂o R and notice by Exercise 13.14 that each of the spaces, {(Xα , τα )}α∈A , are compact second countable and Hausdorff
∞
V = i=1 (ai , bi ) for some collection of disjoint open intervals in R. Since and A is countable.
(ai , bi ) ∈ Aσ (as the reader should verify!), it follows that V ∈ Aσ and therefore,

Proof. Recall from Theorem 18.63 that the Borel σ – algebra, BΛ = σ (τΛ ) , Exercise 28.6. Let (Y, τ ) be a locally compact Hausdorff space and (Y ∗ =
and the product σ – algebra, ⊗α∈Λ Bα , are the same for any Λ ⊂ A. By Ty- Y ∪ {∞} , τ ∗ ) be the one point compactification of Y. Then
chonoff’s Theorem 14.34 and Proposition 15.4, X and XΛ for any Λ ⊂ A are
still compact Hausdorff spaces which are second countable if Λ is finite. By the BY ∗ := σ(τ ∗ ) = {A ⊂ Y ∗ : A ∩ Y ∈ BY = σ(τ )}
Stone Weierstrass Theorem 15.31,
or equivalently put
D := {f ∈ C(X) : f = F ◦ πΛ with F ∈ C(XΛ ) and Λ ⊂⊂ A}
BY ∗ = BY ∪ {A ∪ {∞} : A ∈ BY } .
is a dense subspace of C(X). For f = F ◦ πΛ ∈ D, let
Z Also shows that (Y ∗ = Y ∪ {∞} , τ ∗ ) is second countable if (Y, τ ) was second
I(f ) = F ◦ πΛ (x)dµΛ (x). (28.41) countable.
XΛ
Proof. Proof of Theorem 28.42.
Let us verify that I is well defined. Suppose that f may also be expressed as
Case 1; A is a countable. Let (Xα∗ = Xα ∪ {∞α ∗
Q} , τα∗) be the one point
f = F 0 ◦ πΛ0 with Λ0 ⊂⊂ A and F 0 ∈ C(XΛ0 ). Let Γ := Λ ∪ Λ0 and define ∗
compactification of (Xα , τα ) . For Λ ⊂ A, let XΛ := Xα equipped with the
G ∈ C (XΓ ) by G := F ◦ πΛ,Γ . Hence, using Exercise 19.8, α∈Λ
∗
Z Z Z

Z product topology and Borel σ – algebra, BΛ . Since Λ is at most countable, the
GdµΓ = F ◦ πΛ,Γ dµΓ = F d (πΛ,Γ )∗ µΓ = F dµΛ set, \
XΓ XΓ XΛ XΛ
XΛ := {πα = ∞α } ,
wherein we have used the compatibility condition in the last equality. Similarly, α∈A
using G = F 0 ◦ πΛ0 ,Γ (as the reader should verify), one shows is a measurable subset of XΛ∗ . Therefore for each Λ ⊂⊂ A, we may extend µΛ
to a measure, µ̄Λ , on (XΛ∗ , BΛ
∗
) using the formula,
Z Z
GdµΓ = F 0 dµΛ0 .
XΓ XΛ0
µ̄Λ (B) = µΛ (B ∩ XΛ ) for all B ∈ XΛ∗ .
Therefore Z Z Z
An application of Theorem 28.43 shows there exists a unique probability mea-
F 0 dµΛ0 = GdµΓ = F dµΛ ,
XΛ0 XΓ XΛ sure, µ̄, on X ∗ := XA
∗
such that (πΛ )∗ µ̄ = µ̄Λ for all Λ ⊂⊂ A. Since
which shows I in Eq. (28.41) is well defined. [
X∗ \ X = {πα = ∞α }
Since |I(f )| ≤ kf k∞ , the B.L.T. Theorem 10.4 allows us to extend I from
α∈A
the dense subspace, D, to a continuous linear functional, I, ¯ on C(X). Because I
¯
was positive on D, it is easy to check that I is still positive on C(X). So by the and µ̄ ({πα = ∞}) = µ̄{α} ({∞α }) = 0, it follows that µ̄ (X ∗ \ X) = 0. Hence
Riesz-Markov R Theorem 28.16, there exists a Radon measure on B = M such µ := µ̄|BX is a probability measure on (X, BX ) . Finally if B ∈ BX ⊂ BX ∗ ,
¯ ) = f dµ for all f ∈ C(X). By the definition of I¯ in now follows that
that I(f −1

X µΛ (B) = µ̄Λ (B) = (πΛ )∗ µ̄ (B) = µ̄ πΛ (B)
−1
(B) ∩ X = µ πΛ |−1
Z Z Z
¯ ◦ πΛ ) = = µ̄ πΛ X (B)
F d (πΛ )∗ µ = F ◦ πΛ dµ = I(F F dµΛ
XΛ XΛ XΛ which shows µ is the required probability measure on BX .
Case 2; A is uncountable. By case 1. for each countable or finite subset
for all F ∈ C(X Λ ) and Λ ⊂⊂ A. Since XΛ is a second countable locally compact
Γ ⊂ A there is a measure µΓ on (XΓ , BΓ ) such that (πΛ,Γ )∗ µΓ = µΛ for all
Hausdorff space, this identity implies, see Theorem 22.81 , that (πΛ )∗ µ = µΛ .
Λ ⊂⊂ Γ. By Exercise 18.9,
The uniqueness assertion of the theorem follows from the fact that the measure
−1
µ is determined uniquely by its values on the algebra A := ∪Λ⊂⊂A πΛ (BXΛ )
[
πΓ−1 (BΓ ) : Γ is a countable subset of A ,

M=
which generates B = M, see Theorem 19.55.
1
Alternatively, use Theorems 28.22 and the uniquness assertion in Markov-Riesz i.e. every B ∈ M may be written in the form B = πΓ−1 (C) for some countable
Theorem 28.16 to conclude (πΛ )∗ µ = µΛ . subset, Γ ⊂ A, and C ∈ BΓ . For such a B we define µ (B) := µΓ (C) . It is left

to the reader to check that µ is well defined and that µ is a measure on M.
(Keep in mind the countable union of countable sets is countable.) If Λ ⊂⊂ A
and C ∈ BΛ , then
−1

[(πΛ )∗ µ] (C) = µ πΛ (C) := µΛ (C) ,
i.e. (πΛ )∗ µ = µΛ as desired.
Corollary 28.44. Suppose that {µα }α∈A are probability measure on (Xα , Bα )
for all α ∈ A and if Λ ⊂⊂ A let µΛ := ⊗α∈Λ µα be the product measure on
(XΛ , BΛ = ⊗α∈Λ Bα ) . Then there exists a unique probability measure, µ, on
(X, M) such that (πΛ )∗ µ = µΛ for all Λ ⊂⊂ A. (It is possible remove the
topology from this corollary, see Theorem 32.65 below.)
Exercise 28.7. Prove Corollary 28.44 by showing the measures µΛ := ⊗α∈Λ µα

satisfy the compatibility condition in Theorem 28.42.
29
Probability Measures on Lusin Spaces
Definition 29.1 (Lusin spaces). A Lusin space is a topological space (X, τ ) Proposition 29.6. The following are equivalent:
which is homeomorphic to a Borel subset of a compact metric space. w
1. Pn → P as n → ∞
Example 29.2. By Theorem 15.12, every Polish (i.e. complete separable metric 2. Pn (f ) → P (f ) for every f ∈ BC(X) which is uniformly continuous.
space) is a Lusin space. Moreover, any Borel subset of Lusin space is again a 3. lim supn→∞ Pn (F ) ≤ P (F ) for all F @ X.
Lusin space. 4. lim inf n→∞ Pn (G) ≥ P (G) for all G ⊂o X.
5. limn→∞ Pn (A) = P (A) for all A ∈ B such that P (bd(A)) = 0.
Definition 29.3. Two measurable spaces, (X, M) and (Y, N ) are said to be
isomorphic if there exists a bijective map, f : X → Y such that f (M) = N Proof. 1. =⇒ 2. is obvious. For 2. =⇒ 3.,
and f −1 (N ) = M, i.e. both f and f −1 are measurable. 
 1 if t ≤ 0
φ(t) := 1 − t if 0 ≤ t ≤ 1 (29.1)
0 if t ≥ 1

29.1 Weak Convergence Results
and let fn (x) := φ(nd(x, F )). Then fn ∈ BC(X, [0, 1]) is uniformly continuous,
The following is an application of theorem 14.7 characterizing compact sets in 0 ≤ 1F ≤ fn for all n and fn ↓ 1F as n → ∞. Passing to the limit n → ∞ in
metric spaces. (BRUCE: add Helly’s selection principle here.) the equation
Proposition 29.4. Suppose that (X, ρ) is a complete separable metric space 0 ≤ Pn (F ) ≤ Pn (fm )
and µ is a probability measure on B = σ(τρ ). Then for all ε > 0, there exists gives
Kε @@ X such that µ(Kε ) ≥ 1 − ε. 0 ≤ lim sup Pn (F ) ≤ P (fm )
n→∞
∞
Proof. Let {xk }k=1 be a countable dense subset of X. Then X = and then letting m → ∞ in this inequality implies item 3. 3. ⇐⇒ 4. Assuming
∪k Cxk (1/n) for all n ∈ N. Hence by continuity of µ, there exists, for all item 3., let F = Gc , then
n ∈ N, Nn < ∞ such that µ(Fn ) ≥ 1 − ε2−n where Fn := ∪N k=1 Cxk (1/n).
n
∞
Let K := ∩n=1 Fn then 1 − lim inf Pn (G) = lim sup (1 − Pn (G)) = lim sup Pn (Gc )
n→∞ n→∞ n→∞
µ(X \ K) = µ(∪∞ c
n=1 Fn )
c
≤ P (G ) = 1 − P (G)
∞ ∞ ∞
which implies 4. Similarly 4. =⇒ 3. 3. ⇐⇒ 5. Recall that bd(A) = Ā \ Ao ,
X X X
≤ µ(Fnc ) = (1 − µ(Fn )) ≤ ε2−n = ε
n=1 n=1 n=1 so if P (bd(A)) = 0 and 3. (and hence also 4. holds) we have
so that µ(K) ≥ 1 − ε. Moreover K is compact since K is closed and totally lim sup Pn (A) ≤ lim sup Pn (Ā) ≤ P (Ā) = P (A) and
bounded; K ⊂ Fn for all n and each Fn is 1/n – bounded. n→∞ n→∞
lim inf Pn (A) ≥ lim inf Pn (Ao ) ≥ P (Ao ) = P (A)
∞ n→∞ n→∞
Definition 29.5. A sequence of probability measures {Pn }n=1 is said to con-
verge to a probability P if for every f ∈ BC(X), Pn (f ) → P (f ). This is actually from which it follows that limn→∞ Pn (A) = P (A). Conversely, let F @ X and
weak-* convergence when viewing Pn ∈ BC(X)∗ . set Fδ := {x ∈ X : ρ(x, F ) ≤ δ} . Then
320 29 Probability Measures on Lusin Spaces
bd(Fδ ) ⊂ Fδ \ {x ∈ X : ρ(x, F ) < δ} = Aδ so that Eq. (29.3) becomes,

where Aδ := {x ∈ X : ρ(x, F ) = δ} . Since {Aδ }δ>0 are all disjoint, we must k−1 k−1
1X 1X
have P (Fi ) ≤ P (f ) ≤ P (Fi ) + 1/k.
X
P (Aδ ) ≤ P (X) ≤ 1 k i=1 k i=1
δ>0
Using this equation with P = Pn and then with P = P we find
and in particular the set Λ := {δ > 0 : P (Aδ ) > 0} is at most countable. Let
δn ∈
/ Λ be chosen so that δn ↓ 0 as n → ∞, then
" k−1 #
1X
lim sup Pn (f ) ≤ lim sup Pn (Fi ) + 1/k
P (Fδm ) = lim Pn (Fδn ) ≥ lim sup Pn (F ). n→∞ n→∞ k
n→∞ i=1
n→∞
k−1
Let m → ∞ this equation to conclude P (F ) ≥ lim supn→∞ Pn (F ) as desired. 1X
≤ P (Fi ) + 1/k ≤ P (f ) + 1/k.
To finish the proof we will now show 3. =⇒ 1. By an affine change of variables k i=1
it suffices to consider f ∈ C(X, (0, 1)) in which case we have ≤
k k
(i − 1)
X i X
Since k is arbitrary,
1{ (i−1) ≤f < i } ≤ f ≤ 1{ (i−1) ≤f < i } . (29.2)
i=1
k k k
i=1
k k k lim sup Pn (f ) ≤ P (f ).
n→∞
Let Fi := ki ≤ f and notice that Fk = ∅, then we for any probability P that

This inequality also hold for 1 − f and this implies lim inf n→∞ Pn (f ) ≥ P (f )
k k
and hence limn→∞ Pn (f ) = P (f ) as claimed.
X (i − 1) X i
[P (Fi−1 ) − P (Fi )] ≤ P (f ) ≤ [P (Fi−1 ) − P (Fi )] . (29.3) Definition 29.7. Let X be a topological space. A collection of probability mea-
i=1
k i=1
k
sures Λ on (X, BX ) is said to be tight if for every ε > 0 there exists a compact
Now set Kε ∈ BX such that P (Kε ) ≥ 1 − ε for all P ∈ Λ.
k ∞
X (i − 1) Theorem 29.8. Suppose X is a separable metrizable space and Λ = {Pn }n=1 is
[P (Fi−1 ) − P (Fi )]
i=1
k a tight sequence of probability measures on BX . Then there exists a subsequence
∞
k k
{Pnk }k=1 which is weakly convergent to a probability measure P on BX .
X (i − 1) X (i − 1)
= P (Fi−1 ) − P (Fi ) Proof. First suppose that X is compact. In this case C(X) is a Banach
i=1
k i=1
k
space which is separable by the Stone – Weirstrass theorem, see Exercise 15.5.
k−1 k k−1
X i X i−1 1X By the Riesz theorem, Corollary 32.68, we know that C(X)∗ is in one to one cor-
= P (Fi ) − P (Fi ) = P (Fi )
k k k i=1 respondence with complex measure on (X, BX ). We have also seen that C(X)∗
i=1 i=1
is metrizable and the unit ball in C(X)∗ is weak - * compact, see Theorem
∞
and 14.38. Hence there exists a subsequence {Pnk }k=1 which is weak -* convergent
k to a probability measure P on X. Alternatively, use the cantor’s diagonaliza-
X i ∞
[P (Fi−1 ) − P (Fi )] tion procedure on a countable dense set Γ ⊂ C(X) so find {Pnk }k=1 such that
i=1
k Λ(f ) := limk→∞ Pnk (f ) exists for all f ∈ Γ. Then for g ∈ C(X) and f ∈ Γ, we
k k have
X i−1 X 1
= [P (Fi−1 ) − P (Fi )] + [P (Fi−1 ) − P (Fi )]
i=1
k i=1
k |Pnk (g) − Pnl (g)| ≤ |Pnk (g) − Pnk (f )| + |Pnk (f ) − Pnl (f )|
k−1
X 1 + |Pnl (f ) − Pnl (g)|
= P (Fi ) + ≤ 2 kg − f k∞ + |Pnk (f ) − Pnl (f )|
i=1
k

29.4 Exercises 321
which shows 29.4 Exercises

lim sup |Pnk (g) − Pnl (g)| ≤ 2 kg − f k∞ .
k,l→∞ Exercise 29.1. Let E ∈ BR with m(E) > 0. Then for any α ∈ (0, 1) there
Letting f ∈ Λ tend to g in C(X) shows lim supk,l→∞ |Pnk (g) − Pnl (g)| = 0 and exists a bounded open interval J ⊂ R such that m(E ∩ J) ≥ αm(J).1 Hints: 1.
hence Λ(g) := limk→∞ Pnk (g) for all g ∈ C(X). It is now clear that Λ(g) ≥ 0 Reduce to the case where m(E) ∈ (0, ∞). 2) Approximate E from the outside
for all g ≥ 0 so that Λ is a positive linear functional on X and thus there is a by an open set V ⊂ R. 3. Make use of Exercise 13.14, which states that V may
probability measure P such that Λ(g) = P (g). be written as a disjoint union of open intervals.
General case. By Theorem 15.12 we may assume that X is a subset of
Exercise 29.2. Let F : R → R be a right continuous increasing function and
a compact metric space which we will denote by X̄. We now extend Pn to X̄
µ = µF be as in Theorem 28.32. For a < b, find the values of µ ({a}) , µ ([a, b)) ,
∩ X̄) for all
by setting P̄n (A) := P̄n (A
∞
A ∈ BX̄ . By what we have just proved,
µ ([a, b]) and µ ((a, b)) in terms of the function F.
there is a subsequence P̄k0 := P̄nk k=1 such that P̄k0 converges weakly to a
probability measure P̄ on X̄. The main thing we now have to prove is that Exercise 29.3. Suppose that F ∈ C 1 (R) is an increasing function and µF is
“P̄ (X) = 1,” this is where the tightness assumption is going to be used. Given the unique Borel measure on R such that µF ((a, b]) = F (b) − F (a) for all a ≤ b.
ε > 0, let Kε ⊂ X be a compact set such that P̄n (Kε ) ≥ 1 − ε for all n. Since Show that dµF = ρdm for some function ρ ≥ 0. Find ρ explicitly in terms of F.
Kε is compact in X it is compact in X̄ as well and in particular a closed subset
of X̄. Therefore by Proposition 29.6 Exercise 29.4. Suppose that F (x) = e1x≥3 +π1x≥7 and µF is the is the unique
0
Borel measure on R such that µF ((a, b]) = F (b) − F (a) for all a ≤ b. Give an
P̄ (Kε ) ≥ lim sup P̄k (Kε ) = 1 − ε. explicit description of the measure µF .
k→∞
Exercise 29.5. Let (X, A, µ) be as in Definition 28.36 and Proposition 28.37,
Since ε > 0 is arbitrary, this shows with X0 := ∪∞ n=1 K1/n satisfies P̄ (X0 ) = 1. Y be a Banach space and S(Y ) := Sf (X, A, µ; Y ) be the collection of functions
Because X0 ∈ BX ∩ BX̄ , we may view P̄ as a measure on BX by letting P (A) := f : X → Y such that #(f (X)) < ∞, f −1 ({y}) ∈ A for all y ∈ Y and µ(f 6=
P̄ (A ∩ X0 ) for all A ∈ BX . Given a closed subset F ⊂ X, choose F̃ @ X̄ such 0) < ∞. We may define a linear functional I : S(Y ) → Y by
that F = F̃ ∩ X. Then
X
I(f ) = yµ(f = y).
lim sup Pk0 (F ) = lim sup P̄k0 (F̃ ) ≤ P̄ (F̃ ) = P̄ (F̃ ∩ X0 ) = P (F ),
k→∞ k→∞ y∈Y
w
which shows Pk0 → P. Verify the following statements.
1. Let kf k∞ = supx∈X kf (x)kY be the sup norm on `∞ (X, Y ), then for f ∈
S(Y ),
29.2 Haar Measures kI(f )kY ≤ kf k∞ µ(f 6= 0).
Hence if µ(X) < ∞, I extends to a bounded linear transformation from
To be written.
S̄(Y ) ⊂ `∞ (X, Y ) to Y.
2. Assuming (X, A, µ) satisfies the hypothesis in Exercise 28.5, then
C(X, Y ) ⊂ S̄(Y ).
29.3 Hausdorff Measure 3. Now assume the notation in Section 28.3.3, i.e. X = [−M, M ] for some
M ∈ R and µ is determined by an increasing function F. Let π := {−M =
To be written. t0 < t1 < · · · < tn = M } denote a partition of J := [−M, M ] along with a
choice ci ∈ [ti , ti+1 ] for i = 0, 1, 2 . . . , n − 1. For f ∈ C([−M, M ], Y ), set
1
See also the Lebesgue differentiation Theorem 30.13 from which one may prove the
much stronger form of this theorem, namely for m -a.e. x ∈ E there exits rα (x) > 0
such that m(E ∩ (x − r, x + r)) ≥ αm((x − r, x + r)) for all r ≤ rα (x).

322 29 Probability Measures on Lusin Spaces
n−1
X 29.4.1 The Laws of Large Number Exercises
fπ := f (c0 )1[t0 ,t1 ] + f (ci )1(ti ,ti+1 ] .
i=1 For the rest of the problems of this section, let ν be a probability measure on
BR such that
Show that fπ ∈ S and Z
|x| dν(x) < ∞.
kf − fπ kF → 0 as |π| := max{(ti+1 − ti ) : i = 0, 1, 2 . . . , n − 1} → 0. R
By Corollary 28.44, there exists a unique measure µ on (X := RN , B := BRN =

Conclude from this that ⊗∞
n=1 BR ) such that
n−1
X Z Z
I(f ) = lim f (ci )(F (ti+1 ) − F (ti )). f (x1 , x2 , . . . , xN )dµ(x) = f (x1 , x2 , . . . , xN )dν(x1 ) . . . dν(xN ) (29.4)
|π|→0 X
i=0
RN
RM RM
As usual we will write this integral as −M
f dF and as −M
f (t)dt if F (t) = for all N ∈ N and bounded measurable functions f : RN → R, i.e. µ = ⊗∞
n=1 µn
t. with µn = ν for every n. We will also use the following notation:
Exercise 29.6. Let (X, τ ) be a locally compact Hausdorff space and I : n
1X
C0 (X, R) → R be a positive linear functional. Show I is necessarily bounded, Sn (x) := xk for x ∈ X,
n
i.e. there exists a C < ∞ such that |I(f )| ≤ C kf k∞ for all f ∈ C0 (X, R). k=1
Z
Outline. (BRUCE: see Nate’s proof below and then rewrite this outline to m := xdν(x)
make the problem much easier and to handle more general circumstances.) R
Z Z
1. By the Riesz-Markov Theorem 28.16, R there exists a unique Radon measure σ 2 := (x − m)2 dν(x) = x2 dν(x) − m2 , and
µ on (X, BX ) such that µ (f ) := X f dµ = I (f ) for all f ∈ Cc (X, R) . Show ZR R
µ (f ) ≤ I (f ) for all f ∈ C0 (X, [0, ∞)). γ := (x − m)4 dν(x).

2. Show that if µ (X) = ∞, then there exists a function f ∈ C0 (X, [0, ∞)) R
such that ∞ = µ(f ) ≤ I (f ) contradicting the assumption that I is real As is customary, m is said to be the mean or average of ν and σ 2 is the variance
valued. of ν.
3. By item 2., µ (X) < ∞ and therefore C0 (X, R) ⊂ L1 (µ) and µ : C0 (X, R) →
R is a well defined positive linear functional. Let J (f ) := I (f )−µ (f ) , which Exercise
R 29.8 (Weak Law of Large Numbers). Assume σ 2 < ∞. Show
by item 1. is a positive linear functional such that J|Cc (X,R) ≡ 0. Show that S dµ = m.
X n
σ2
Z
any positive linear functional, J, on C0 (X, R) satisfying these properties 2 2
kSn − mk2 = (Sn − m) dµ = ,
must necessarily be zero. Thus I (f ) = µ (f ) and kIk = µ (X) < ∞ as X n
claimed. σ2
and µ(|Sn − m| > ε) ≤ nε2 for all ε > 0 and n ∈ N.
Exercise 29.7. BRUCE (Drop this exercise or move somewhere else, it is
Exercise 29.9 (A simple
R form of the Strong Law of Large Numbers).
already proved in the notes in more general terms.) Suppose that I :
Suppose now that γ := R (x − m)4 dν(x) < ∞. Show for all ε > 0 and n ∈ N
Cc∞ (R, R) → R is a positive linear functional. Show
that
1. For each compact subset K @@ R there exists a constant CK < ∞ such 4
Z
4 1
(Sn − m) dµ = 4 nγ + 3n(n − 1)σ 4

that kSn − mk4 =
X n
|I(f )| ≤ CK kf k∞ 1 −1
= 2 n γ + 3 1 − n−1 σ 4 and

whenever supp(f ) ⊂ K. n
R Radon measure µ on BR (the Borel σ – algebra n−1 γ + 3 1 − n−1 σ 4

2. Show there exists a unique
on R) such that I(f ) = R f dµ for all f ∈ Cc∞ (R, R). µ(|Sn − m| > ε) ≤ .
ε 4 n2

Conclude from the last estimate and the first Borel Cantelli Lemma 19.20 that
limn→∞ Sn (x) = m for µ – a.e. x ∈ X.
Exercise 29.10. Suppose γ := R (x − m)4 dν(x) < ∞ and m = R (x −

R R
m)dν(x) 6= 0. For λ > 0 let Tλ : RN → RN be defined by Tλ (x) =

(λx1 , λx2 , . . . , λxn , . . . ), µλ = µ ◦ Tλ−1 and
 
n
 1 X 
Xλ := x ∈ RN : lim xj = λ .
 n→∞ n 
j=1
Show
1 if λ = λ0

µλ (Xλ0 ) = δλ,λ0 =
0 if λ 6= λ0
and use this to show if λ 6= 1, then dµλ 6= ρdµ for any measurable function
ρ : RN → [0, ∞].
30
Lebesgue Differentiation and the Fundamental Theorem of Calculus
BRUCE: replace Rn by Rd in this section? Proof. Choose a compact set K ⊂ U such that m(K) > c and then let E1 ⊂
E be a finite subcover of K. Choose B1 ∈ E1 to be a ball with largest diameter
Notation 30.1 In this chapter, let B = BRn denote the Borel σ – algebra
in E1 . Let E2 = {A ∈ E1 : A ∩ B1 = ∅}. If E2 is not empty, choose B2 ∈ E2 to be
on Rn and m be Lebesgue measure on B. If V is an open subset of Rn , let
a ball with largest diameter in E2 . Similarly let E3 = {A ∈ E2 : A ∩ B2 = ∅} and
L1loc (V ) := L1loc (V, m) and simply write L1loc for L1loc (Rn ). We will also write
if E3 is not empty, choose B3 ∈ E3 to be a ball with largest diameter in E3 .
|A| for m(A) when A ∈ B.
Continue choosing Bi ∈ E for i = 1, 2, . . . , k this way until Ek+1 is empty, see
Definition 30.2. A collection of measurable sets {E}r>0 ⊂ B is said to shrink Figure 30.1 below. If B = B(x0 , r) ⊂ Rn , let B ∗ = B(x0 , 3r) ⊂ Rn , that is B ∗
nicely to x ∈ Rn if (i) Er ⊂ B(x, r) for all r > 0 and (ii) there exists α > 0
such that m(Er ) ≥ αm(B(x, r)). We will abbreviate this by writing Er ↓ {x}
nicely. (Notice that it is not required that x ∈ Er for any r > 0.
The main result of this chapter is the following theorem.
Theorem 30.3. Suppose that ν is a complex measure on (Rn , B) , then there
exists g ∈ L1 (Rn , m) and a complex measure νs such that νs ⊥ m, dν = gdm +
dνs , and for m - a.e. x
ν(Er )
g(x) = lim (30.1)
r↓0 m(Er )
for any collection of {Er }r>0 ⊂ B which shrink nicely to {x} . (Eq. (30.1) holds
for all x ∈ L (g) – the Lebesgue set of g, see Definition 30.11 and Theorem
30.12 below.)
Fig. 30.1. Picking out the large disjoint balls via the “greedy algorithm.”
Proof. The existence of g and νs such that νs ⊥ m and dν = gdm + dνs is
a consequence of the Radon-Nikodym Theorem 24.34. Since
Z
ν(Er ) 1 νs (Er ) is the ball concentric with B which has three times the radius of B. We will now
= g(x)dm(x) +
m(Er ) m(Er ) Er m(Er ) show K ⊂ ∪ki=1 Bi∗ . For each A ∈ E1 there exists a first i such that Bi ∩ A 6= ∅.
Eq. (30.1) is a consequence of Theorem 30.13 and Corollary 30.15 below. In this case diam(A) ≤ diam(Bi ) and A ⊂ Bi∗ . Therefore A ⊂ ∪ki=1 Bi∗ and
The rest of this chapter will be devoted to filling in the details of the proof hence K ⊂ ∪{A : A ∈ E1 } ⊂ ∪ki=1 Bi∗ . Hence by sub-additivity,
of this theorem. k
X k
X
c < m(K) ≤ m(Bi∗ ) ≤ 3n m(Bi ).
i=1 i=1
30.1 A Covering Lemma and Averaging Operators
Lemma 30.4 (Covering Lemma). Let E be a collection of open balls in Rn Definition 30.5. For f ∈ L1loc , x ∈ Rn and r > 0 let
and U = ∪B∈E B. If c < m(U ), then there exists disjoint balls B1 , . . . , Bk ∈ E 1
Z
k (Ar f )(x) = f dm (30.2)
such that c < 3n
P
m(Bj ). |B(x, r)|
j=1 B(x,r)
326 30 Lebesgue Differentiation and the Fundamental Theorem of Calculus
where B(x, r) = B(x, r) ⊂ Rn , and |A| := m(A). This should be compared with Chebyshev’s inequality which states that
Lemma 30.6. Let f ∈ L1loc , then for each x ∈ Rn , (0, ∞) 3 r → (Ar f )(x) ∈ C kf kL1
m (|f | > α) ≤ .
is continuous and for each r > 0, Rn 3 x → (Ar f ) (x) ∈ C is measurable. α
Proof. Recall that |B(x, r)| = m(E1 )rn which is continuous in r. Also Proof. Let Eα := {Hf > α}. For all x ∈ Eα there exists rx such that
limr→r0 1B(x,r) (y) = 1B(x,r0 ) (y) if |y| =
6 r0 and since m ({y : |y| =
6 r0 }) = 0 (you Arx |f | (x) > α, i.e. Z
prove!), limr→r0 1B(x,r) (y) = 1B(x,r0 ) (y) for m -a.e. y. So by the dominated 1
|Bx (rx )| < f dm.
convergence theorem, α Bx (rx )
Z Z Since Eα ⊂ ∪x∈Eα Bx (rx ), if c < m(Eα ) ≤ m(∪x∈Eα Bx (rx )) then, using Lemma
lim f dm = f dm 30.4, there exists x1 , . . . , xk ∈ Eα and disjoint balls Bi = Bxi (rxi ) for i =
r→r0
B(x,r) B(x,r0 ) 1, 2, . . . , k such that
k
and therefore Z X
n
X 3n Z 3n
Z
3n
1 c< 3 |Bi | < |f | dm ≤ |f | dm = kf kL1 .
(Ar f )(x) = f dm α Bi α Rn α
m(E1 )rn i=1
B(x,r)
This shows that c < 3n α−1 kf kL1 for all c < m(Eα ) which proves m(Eα ) ≤
is continuous in r. Let gr (x, y) := 1B(x,r) (y) = 1|x−y|<r . Then gr is B ⊗ B – 3n α−1 kf k.
measurable (for example write it as a limit of continuous functions or just notice
that F : Rn × Rn → R defined by F (x, y) := |x − y| is continuous) and so that Theorem 30.9. If f ∈ L1loc then lim(Ar f )(x) = f (x) for m – a.e. x ∈ Rn .
r↓0
by Fubini’s theorem
Z Z Proof. With out loss of generality we may assume f ∈ L1 (m). We now
x→ f dm = gr (x, y)f (y)dm(y) begin with the special case where f = g ∈ L1 (m) is also continuous. In this
B(x,r) B(x,r) case we find:
Z
1
is B – measurable and hence so is x → (Ar f ) (x). |(Ar g)(x) − g(x)| ≤ |g(y) − g(x)|dm(y)
|B(x, r)| B(x,r)
≤ sup |g(y) − g(x)| → 0 as r → 0.
y∈B(x,r)
30.2 Maximal Functions
In fact we have shown that (Ar g)(x) → g(x) as r → 0 uniformly for x in
Definition 30.7. For f ∈ L1 (m), the Hardy - Littlewood maximal function Hf compact subsets of Rn . For general f ∈ L1 (m),
is defined by
(Hf )(x) = sup Ar |f | (x). |Ar f (x) − f (x)| ≤ |Ar f (x) − Ar g(x)| + |Ar g(x) − g(x)| + |g(x) − f (x)|
r>0
= |Ar (f − g)(x)| + |Ar g(x) − g(x)| + |g(x) − f (x)|
Lemma 30.6 allows us to write ≤ H(f − g)(x) + |Ar g(x) − g(x)| + |g(x) − f (x)|
(Hf )(x) = sup Ar |f | (x) and therefore,
r∈Q, r>0
from which it follows that Hf is measurable. lim|Ar f (x) − f (x)| ≤ H(f − g)(x) + |g(x) − f (x)|.
r↓0
1
Theorem 30.8 (Maximal Inequality). If f ∈ L (m) and α > 0, then So if α > 0, then
3n n
αo n αo
m (Hf > α) ≤ kf kL1 . Eα := lim|Ar f (x) − f (x)| > α ⊂ H(f − g) > ∪ |g − f | >
α r↓0 2 2

30.3 Lebesque Set 327
and thus Proof. For w ∈ C define gw (x) = |f (x) − w|p and

α α
m(Eα ) ≤ m H(f − g) > + m |g − f | >
2 2 Ew := x : lim (Ar gw ) (x) 6= gw (x) .
r↓0
3n 1
≤ kf − gkL1 + kf − gkL1
α/2 α/2 Then by Theorem 30.9 m(Ew ) = 0 for all w ∈ C and therefore m(E) = 0 where
≤ 2(3n + 1)α−1 kf − gkL1 , [
E= Ew .
where in the second inequality we have used the Maximal inequality (The- w∈Q+iQ
orem 30.8) and Chebyshev’s inequality. Since this is true for all continuous
g ∈ C(Rn ) ∩ L1 (m) and this set is dense in L1 (m), we may make kf − gkL1 as By definition of E, if x ∈
/ E then.
small as we please. This shows that
lim(Ar |f (·) − w|p )(x) = |f (x) − w|p
∞ r↓0
X
m x : lim|Ar f (x) − f (x)| > 0 = m(∪∞
n=1 E1/n ) ≤ m(E1/n ) = 0. p
r↓0 for all w ∈ Q + iQ. Letting q := p−1 (so that p/q = p − 1) we have
n=1
p
|f (·) − f (x)|p ≤ (|f (·) − w| + |w − f (x)|)
p p
Corollary 30.10. If dµ = gdm with g ∈ L1loc then ≤ 2p/q (|f (·) − w| + |w − f (x)|p ) = 2p−1 (|f (·) − w| + |w − f (x)|p ) ,
p
µ(B(x, r)) (Ar |f (·) − f (x)|p )(x) ≤ 2p−1 (Ar |f (·) − w| ) (x) + 2p−1 |w − f (x)|
= Ar g(x) → g(x) for m – a.e. x.
|B(x, r)| and hence for x ∈
/ E,
lim(Ar |f (·) − f (x)|p )(x) ≤ 2p−1 |f (x) − w|p + 2p−1 |w − f (x)|p = 2p |f (x) − w|p .
30.3 Lebesque Set r↓0
Since this is true for all w ∈ Q + iQ, we see that

Definition 30.11. For f ∈ L1loc (m), the Lebesgue set of f is
  lim(Ar |f (·) − f (x)|p )(x) = 0 for all x ∈
/ E,
 Z  r↓0

n 1 
L (f ) := x ∈ R : lim |f (y) − f (x)|dy = 0 c

 r↓0 |B(x, r)| 
 i.e. E c ⊂ Lp (f ) or equivalently (Lp (f )) ⊂ E. So m (Rn \ Lp (f )) ≤ m(E) = 0.
B(x,r)

= x ∈ Rn : lim (Ar |f (·) − f (x)|) (x) = 0 . Theorem 30.13 (Lebesque Differentiation Theorem). If f ∈ Lploc and
r↓0
x ∈ Lp (f ) (so in particular for m – a.e. x), then
More generally, if p ∈ [1, ∞) and f ∈ Lploc (m) , let Z
1
lim |f (y) − f (x)|p dy = 0
r↓0 m(Er ) E
 
 Z  r
 1 
Lp (f ) := x ∈ Rn : lim |f (y) − f (x)|p dy = 0 and
r↓0 |B(x, r)| Z


B(x,r)

 1
lim f (y)dy = f (x)
r↓0 m(Er ) Er
Theorem 30.12. Suppose 1 ≤ p < ∞ and f ∈ Lploc (m), then m Rd \ Lp (f ) =

when Er ↓ {x} nicely, see Definition 30.2.
0.

Proof. For x ∈ Lp (f ) , by Hölder’s inequality (Theorem 21.2) or Jensen’s

inequality (Theorem 21.10), we have
Z p Z p
1 1

m(Er ) f (y)dy − f (x) =
(f (y) − f (x)) dy
Er m(E r )
Z Er
1
≤ |f (y) − f (x)|p dy
m(Er ) Er
P∞ −2
δ1/n and A = R2 \
Z
1 Fig. 30.2. In this picture we imagine that λ = n=1 n
≤ |f (y) − f (x)|p dy {(−1/n, 0) : n ∈ N} . We may approximatePA by the open sets, VN := R2 \
αm(B(x, r)) ∞ −2
B(x,r) {(−1/n, 0) : 1 ≤ n ≤ N } , since λ (VN ) = n=N +1 n → 0 as N → ∞. (Of
course we could simplify matters in this setting by choosing A = V := R2 \
which tends to zero as r ↓ 0 by Theorem 30.12. In the second inequality we ({(−1/n, 0) : 1 ≤ n ≤ N } ∪ {0}) , but this would not be very enlightening.)
have used the fact that m(B(x, r) \ B(x, r)) = 0.
Lemma 30.14. Suppose λ is positive K – finite measure on B := BRn such

N N
that λ ⊥ m. Then for m – a.e. x, X X
c < 3n m(Bi ) < k3n λ(Bi )
λ(B(x, r)) i=1 i=1
lim = 0. ≤ k3 λ(U ) ≤ k3n λ(Vε ) ≤ k3n ε.
n
r↓0 m(B(x, r))
Proof. Let A ∈ B such that λ(A) = 0 and m(Ac ) = 0. By the regularity Since c < m(U ) is arbitrary we learn that m(U ) ≤ k3n ε. This argument shows
theorem (see Theorem 28.22, Corollary 32.42 or Exercise 33.4), for all ε > 0 open sets Uε such that Fk ⊂ Uε and m(Uε ) ≤ k3n ε for all ε > 0. Therefore
there exists an open set Vε ⊂ Rn such that A ⊂ Vε and λ(Vε ) < ε. For the Fk ⊂ G := ∩∞l=1 U1/l ∈ B with m (G) = 0 which shows Fk ∈ L and m(Fk ) = 0.
rest of this argument, we will assume m has been extended to the Lebesgue Since
λ(B(x, r))
measurable sets, L := B̄ m . Let F∞ := x ∈ A : lim > 0 = ∪∞ k=1 Fk ∈ L,
r↓0 m(B(x, r))

λ(B(x, r)) 1 it also follows that F∞ ∈ L and m (F∞ ) = 0. Since
Fk := x ∈ A : lim >
r↓0 m(B(x, r)) k
n λ(B(x, r))
the for x ∈ Fk choose rx > 0 such that Bx (rx ) ⊂ Vε (see Figure 30.2) and x ∈ R : lim > 0 ⊂ F∞ ∪ Ac
r↓0 m(B(x, r))
λ(B(x,rx )) 1
m(B(x,rx )) > k , i.e.
m(B(x, rx )) < kλ(B(x, rx )). and m(Ac ) = 0, we have shown
S
Let E = {B(x, rx )}x∈Fk and U := B(x, rx ) ⊂ Vε . Heuristically if all the

n λ(B(x, r))
x∈Fk m x ∈ R : lim >0 = 0.
r↓0 m(B(x, r))
balls in E were disjoint and E were countable, then
X X
m(Fk ) ≤ m(B(x, rx )) < k λ(B(x, rx ))
x∈Fk x∈Fk Corollary 30.15. Let λ be a complex or a K – finite signed measure (i.e.
= kλ(U ) ≤ k λ(Vε ) ≤ kε. ν(K) ∈ R for all K @@ Rn ) such that λ ⊥ m. Then for m – a.e. x,
Since ε > 0 is arbitrary this would imply that Fk ∈ L and m(Fk ) = 0. To fix λ(Er )
lim =0
the above argument, suppose that c < m(U ) and use the covering lemma to r↓0 m(Er )
find disjoint balls B1 , . . . , BN ∈ E such that
whenever Er ↓ {x} nicely.

Proof. By Exercise 24.17, λ ⊥ m implies |λ| ⊥ m. Hence the result follows 4. There exists a unique measure, νG on B = BR such that
from Lemma 30.14 and the inequalities,
νG ((a, b]) = G(b) − G(a) for all a < b.
|λ(Er )| |λ|(Er ) |λ|(B(x, r)) |λ|(B(x, 2r))
≤ ≤ ≤ . 5. For m – a.e. x, F 0 (x) and G0 (x) exists and F 0 (x) = G0 (x). (Notice that
m(Er ) αm(B(x, r)) αm(B(x, r)) α2−n m(B(x, 2r))
F 0 (x) and G0 (x) are non-negative when they exist.)
6. The function F 0 is in L1loc (m) and there exists a unique positive measure νs
on (R, BR ) such that
Proposition 30.16. TODO Add in almost everywhere convergence result of
convolutions by approximate δ – functions.
Z b
F (b+) − F (a+) = F 0 dm + νs ((a, b]) for all − ∞ < a < b < ∞.
a
30.4 The Fundamental Theorem of Calculus Moreover the measure νs is singular relative to m and F 0 ∈ L1 (R, m) if F
is bounded.
In this section we will restrict the results above to the one dimensional setting.
Proof. Item 1. is a consequence of Eq. (28.35) of Theorem 28.40. Neverthe-
The following notation will be in force for the rest of this chapter. (BRUCE:
less we will still give a direct proof here as well.
make sure this notation agrees with the notation in Notation 30.21.)
1. The following observation shows G is increasing: if x < y then
Notation 30.17 Let
1. m be one dimensional Lebesgue measure on B := BR , F (x−) ≤ F (x) ≤ F (x+) = G(x) ≤ F (y−) ≤ F (y) ≤ F (y+) = G(y).
2. α, β be numbers in R̄ such that −∞ ≤ α < β ≤ ∞, (30.4)
3. A = A[α,β] be the algebra generated by sets of the form (a, b] ∩ [α, β] with Since G is increasing, G(x) ≤ G(x+). If y > x then G(x+) ≤ F (y) and
−∞ ≤ a < b ≤ ∞, hence G(x+) ≤ F (x+) = G(x), i.e. G(x+) = G(x) which is to say G is
4. Ab denote those sets in A which are bounded, right continuous.
5. and B[α,β] be the Borel σ – algebra on [α, β] ∩ R. 2. Since G(x) ≤ F (y−) ≤ F (y) for all y > x, it follows that
Notation 30.18 Given a function F : R → R̄ or F : R → C, let F (x−) = G(x) ≤ lim F (y−) ≤ lim F (y) = G(x)
y↓x y↓x
limy↑x F (y), F (x+) = limy↓x F (y) and F (±∞) = limx→±∞ F (x) whenever the
limits exist. Notice that if F is a monotone functions then F (±∞) and F (x±) showing G(x) = limy↓x F (y−).
exist for all x. 3. By Eq. (30.4), if x 6= y then
(F (x−), F (x+)] ∩ (F (y−), F (y+)] = ∅.

30.4.1 Increasing Functions
Therefore, {(F (x−), F (x+)]}x∈R are disjoint possible empty intervals in R.
Theorem 30.19 (Monotone functions). Let F : R → R be increasing and Let N ∈ N and α ⊂⊂ (−N, N ) be a finite set, then
define G(x) = F (x+). Then: a
(F (x−), F (x+)] ⊂ (F (−N ), F (N )]
1. The function G is increasing and right continuous.
x∈α
2. For x ∈ R, G(x) = limy↓x F (y−).
3. The set {x ∈ R : F (x+) > F (x−)} is countable and for each N > 0, and and therefore,
moreover, X
X [F (x+) − F (x−)] ≤ F (N ) − F (−N ) < ∞.
[F (x+) − F (x−)] ≤ F (N ) − F (−N ) < ∞. (30.3) x∈α
x∈(−N,N ]
Since this is true for all α ⊂⊂ (−N, N ], Eq. (30.3) holds. Eq. (30.3) shows


ΓN := {x ∈ (−N, N )|F (x+) − F (x−) > 0} H(x + r) − H(x) |H(x + r)| + |H(x)|
≤
r |r|
is countable and hence so is
H(x + |r|) + H(x − |r|) + H(x)
≤
Γ := {x ∈ R|F (x+) − F (x−) > 0} = ∪∞
N =1 ΓN .
|r|
λ([x − |r| , x + |r|])
4. Item 4. is a direct consequence of Theorem 28.32. Notice that νG is a finite ≤2
2 |r|
measure when F and hence G is bounded.
5. By Theorem 30.3, and the last term goes to zero as r → 0 because {[x − r, x + r]}r>0 shrinks
dνG = gdm + dνs , nicely to {x} as r ↓ 0 and m([x − |r| , x + |r|]) = 2 |r| . Hence we conclude
where νs ⊥ m, g ∈ L1loc (R, m) with g ∈ L1 (R, m) if F is bounded, and for for m – a.e. x that H 0 (x) = 0.
m - a.e. x; for all sequences {Er }r>0 which shrink nicely to {x} , 6. From Theorem 30.3, item 5. and Eqs. (30.5) and (30.6), F 0 = G0 ∈ L1loc (m)
and dνG = F 0 dm + dνs where νs is a positive measure such that νs ⊥ m.
g (x) = lim(νG (Er )/m(Er )) Applying this equation to an interval of the form (a, b] gives
r↓0
Z b
with the limit being independent of the choice of the sequence {Er }r>0 . F (b+) − F (a+) = νG ((a, b]) = F 0 dm + νs ((a, b]). (30.7)
Since (x, x + r] ↓ {x} and (x − r, x] ↓ {x} nicely, a
νG (x, x + r]) G(x + r) − G(x) d The uniqueness of νs such that this equation holds is a consequence of
g (x) = lim = lim = + G(x) (30.5) Theorem 19.55. As we have already mentioned, when F is bounded then
r↓0 m((x, x + r]) r↓0 r dx
F 0 ∈ L1 (R, m) . This can also be seen directly by letting a → −∞ and
and b → +∞ in Eq. (30.7).
νG ((x − r, x]) G(x) − G(x − r)
g (x) = lim = lim
r↓0m((x − r, x]) r↓0 r
G(x − r) − G(x) d Example 30.20. Let C ⊂ [0, 1] denote the cantor set constructed as follows. Let
= lim = − G(x) (30.6) C1 = [0, 1] \ (1/3, 2/3), C2 := C1 \ [(1/9, 2/9) ∪ (7/9, 8/9)] , etc., so that we keep
r↓0 −r dx
removing the middle thirds at each stage in the construction. Then
exist and are equal for m - a.e. x, i.e. G0 (x) = g (x) exists for m -a.e. x.  
For x ∈ R, let  X ∞ 
C := ∩∞ C
n=1 n = x = a j 3−j
: aj ∈ {0, 2}
H(x) := G(x) − F (x) = F (x+) − F (x) ≥ 0. 
j=0

Since F (x) = G(x) − H(x), the proof of 5. will be complete once we show and
H 0 (x) = 0 for m – a.e. x. From Item 3.,
1 2 22

Λ := {x ∈ R : F (x+) > F (x)} ⊂ {x ∈ R : F (x+) > F (x−)} m(C) = 1 − + + + ...
3 9 33
∞ n
is a countable set and 1X 2 1 1
=1− =1− = 0.
X X X 3 n=0 3 3 1 − 2/3
H(x) = (F (x+)−F (x)) ≤ (F (x+)−F (x−)) < ∞
x∈(−N,N ) x∈(−N,N ) x∈(−N,N ) Associated to this set is the so called cantor function F (x) := limn→∞ fn (x)
∞
for all N < ∞. Therefore λ :=
P
H(x)δx (i.e. λ(A) :=
P
H(x) for all where the {fn }n=1 are continuous non-decreasing functions such that fn (0) = 0,
x∈A
x∈R fn (1) = 1 with the fn pictured in Figure 30.3 below. From the pictures one sees
A ∈ BR ) defines a Radon measure on BR . Since λ(Λc ) = 0 and m(Λ) = 0, that {fn } are uniformly Cauchy, hence there exists F ∈ C([0, 1]) such that
the measure λ ⊥ m. By Corollary 30.15 for m - a.e. x, F (x) := limn→∞ fn (x). The function F has the following properties,

note the closure of X in R and X̄∞ denote the closure of X in R̄ := [−∞, ∞] .1

We further let A denote the algebra of half open intervals in X, i.e. the alge-
bra generated by the sets, {(a, b] ∩ X : −∞ ≤ a ≤ b ≤ ∞} . Also let Ab be those
A ∈ A which are bounded.
Definition 30.22. For −∞ ≤ a < b ≤ ∞, a partition P of [a, b] ∩ X̄ is a finite
subset of [a, b] ∩ X̄ such that {a, b} ∩ X̄ ⊂ P. For x ∈ [min P, max P], let
x+ = xP+ = min {y ∈ P : y > x} ∧ max P
where min ∅ := ∞.
For example, if X = (α, ∞) , then a partition of X̄ = [α, ∞) is a finite
subset, P, of [α, ∞) such that α ∈ P and if α ≤ a < b < ∞, then a partition of
[a, b] is a finite subset, P, of [a, b] such that a, b ∈ P, see Figure 30.4.
Fig. 30.3. Constructing the Cantor function.
1. F is continuous and non-decreasing.

2. F 0 (x) = 0 for m – a.e. x ∈ [0, 1] because F is flat on all of the middle third
open intervals used to construct the cantor set C and the total measure of
these intervals is 1 as proved above. Fig. 30.4. In this figure, X = (α, ∞) and partitions of X and [a, b] with [a, b] ⊂ X̄
3. The measure on B[0,1] associated to F, namely ν([0, b]) = F (b) is singular have been shown. The meaing of x+ is also depicted.
relative to Lebesgue measure and ν ({x}) = 0 for all x ∈ [0, 1]. Notice that
ν ([0, 1]) = 1. In particular, the function F certainly does not satisfy the
fundamental theorem of calculus despite the fact that F 0 (x) = 0 for a.e. x.
The following proposition will help motivate a number of concepts which
will need to introduce.
30.4.2 Functions of Bounded Variation
Proposition 30.23. Suppose ν is a complex measure on (X, B) and F : X̄ → C
Our next goal is to prove an analogue of Theorem 30.19 for complex valued is a function
F. Let α, β ∈ R with α < β be fixed. The following notation will be used ν((a, b]) = F (b) − F (a)
throughout this section. for all a, b ∈ X̄ with a < b.(For example one may let F (x) := ν((−∞, x] ∩ X)).)
Then
Notation 30.21 Let (X, B) denote one of the
following four measure spaces: 1

(R, BR ) , (−∞, β], B(−∞,β) , (α, ∞) , B(α,∞) or (α, β], B(α,β] and let X̄ de- So X̄ is either R, (−∞, β], [α, ∞), or [α, β] respectively and X̄∞ is either [−∞, ∞] ,
[−∞, β], [α, ∞], or [α, β] respectively.

1. F : X̄ → C is a right continuous function, Definition 30.24 (Total variation of a function). The total variation of
2. For all a, b ∈ X̄ with a < b, a function F : X̄→ C on (a, b] ∩ X ⊂ X̄∞ (b = ∞ is allowed here) is defined by
X X X
|ν| (a, b] = sup |ν(x, x+ ]| = sup |F (x+ ) − F (x)| (30.8) TF ((a, b] ∩ X) = sup |F (x+ ) − F (x)|
P P P
x∈P x∈P x∈P
where supremum is over all partitions P of [a, b]. where supremum is over all partitions P of [a, b] ∩ X. Also let
3. If inf X = −∞ then Eq. (30.8) remains valid for a = −∞ and moreover,
TF (b) := TF ((inf X, b]) for all b ∈ X.
|ν| ((−∞, b]) = lim |ν| (a, b]. (30.9)
a→−∞
The function F is said to have bounded variation on (a, b] ∩ X if TF ((a, b] ∩
Similar statements hold in case sup X = +∞ in which case we may take X) < ∞ and F is said to be of bounded variation, and we write F ∈ BV (X) ,
b = ∞ above. In particular if X = R, then if TF (X) < ∞.
( )
X Definition 30.25 (Absolute continuity). A function F : X̄→ C is abso-
|ν| (R) = sup |F (x+ ) − F (x)| : P is a partition of R lutely continuous if for all ε > 0 there exists δ > 0 such that
x∈P
( ) n
X
|F (bi ) − F (ai )| < ε (30.11)
X
= lim sup |F (x+ ) − F (x)| : P is a partition of [a, b] .
a→−∞ i=1
b→∞ x∈P
n
n
4. ν m on X iff for all ε > 0 there exists δ > 0 such that
P
whenever {(ai , bi ]}i=1 are disjoint subintervals of X such that (bi − ai ) < δ.
i=1
n
X n
X
|ν ((ai , bi ])| = |F (bi ) − F (ai )| < ε (30.10) Exercise 30.2. Let F, G : X̄→ C be and λ ∈ C be given. Show
i=1 i=1
1. TF +G ≤ TF + TG and TλF = |λ| TF . Conclude from this that BV (X) is a
n
n P vector space.
whenever {(ai , bi ]}i=1 are disjoint subintervals of X such that (bi − ai ) <
i=1 2. TRe F ≤ TF , TIm F ≤ TF , and TF ≤ TRe F +TIm F . In particular F ∈ BV (X)
δ. iff Re F and Im F are in BV (X) .
3. If F : X̄→ C is absolutely continuous then F : X̄→ C is continuous and in
Proof. 1. The right continuity of F is a consequence of the continuity of ν fact is uniformly continuous.
under decreasing limits of sets.
2 and 3. When a, b ∈ X̄, Eq. (30.8) follows from Proposition 24.33 and the Lemma 30.26 (Examples). Let F : X̄ → F be given, where F is either R of
fact that B = σ(A). The verification of item 3. is left for Exercise 30.1. C.
4. Equation (30.10) is a consequence of Theorem 24.38 and the following
remarks: 1. If F : X̄ → R is a monotone function, then TF ((a, b]) = |F (b) − F (a)| for
all a, b ∈ X̄ with a < b. So F ∈ BV (X) iff F is bounded (which will be the
n n
a) {(ai , bi ) ∩ X}i=1 are disjoint intervals iff {(ai , bi ] ∩ X}i=1 are disjoint inter- case if X = [α, β]).
vals, 2. If F : [α, β] → C is absolutely continuous then F ∈ BV ((α, β]).
n
b) m (X ∩ (∪ni=1 (ai , bi ])) ≤
P
(bi − ai ), and 3. If F ∈ C ([α, β] → R) , F 0 (x) is differentiable for all x ∈ (α, β) , and
i=1 `n supx∈(α,β) |F 0 (x)| = M < ∞, then F is absolutely continuous2 and
c) the general element A ∈ Ab is of the form A = X ∩ ( i=1 (ai , bi ]) .
TF ((a, b]) ≤ M (b − a) ∀ α ≤ a < b ≤ β.
2
It is proved in Natanson or in Rudin that this is also true if F ∈ C([α, β]) such
Exercise 30.1. Prove Item 3. of Proposition 30.23. that F 0 (x) exists for all x ∈ (α, β) and F 0 ∈ L1 ([α, β] , m) .

n
n Z

4. Let f ∈ L1 (X, m) and set X X
|F (bi ) − F (ai )| = f dm

Z
i=1 i=1
(ai ,bi ]
F (x) = f dm for all x ∈ X̄. (30.12) Xn Z
(−∞,x]∩X̄
≤ |f | dm
i=1 (ai ,bi ]
Then F : X̄→ C is absolutely continuous. Z
= |f | dm = ν(∪ni=1 (ai , bi ]). (30.13)
Proof. ∪n
i=1 (ai ,bi ]
1. If F is monotone increasing and P is a partition of (a, b] then Since ν is absolutely continuous relative to m, by Theorem 24.38, for all
X X ε > 0 there exist δ > 0 such that ν(A) < ε if m(A) < δ. Applying this
|F (x+ ) − F (x)| = (F (x+ ) − F (x)) = F (b) − F (a) result with A = ∪ni=1 (ai , bi ], it follows from Eq. (30.13) that F satisfies
x∈P x∈P the definition of being absolutely continuous. Furthermore, Eq. (30.13) also
may be used to show
so that TF ((a, b]) = F (b) − F (a). Similarly, one shows Z
TF ((a, b]) ≤ |f | dm.
TF ((a, b]) = F (a) − F (b) = |F (b) − F (a)| (a,b]
if F is monotone decreasing. Also note that F ∈ BV (R) iff

|F (∞) − F (−∞)| < ∞, where F (±∞) = limx→±∞ F (x) . Example 30.27 (See I. P. Natanson,“Theory of functions of a real variable,”
2. Since F is absolutely continuous, there exists δ > 0 such that whenever p.269.). In each of the two examples below, f ∈ C([−1, 1]).
a, b ∈ X̄ with a < b and b − a < δ, then
3/2
X 1. Let f (x) = |x| sin x1 with f (0) = 0, then f is everywhere differentiable
|F (x+ ) − F (x)| ≤ 1 but f 0 is not bounded near zero. However, f 0 is in L1 ([−1, 1]).
x∈P 2. Let f (x) = x2 cos xπ2 with f (0) = 0, then f is everywhere differentiable but
f0 ∈
/ L1 (−ε, ε) for any ε ∈ (0, 1) . Indeed, if 0 ∈/ (α, β) then
for all partitions, P, of [a, b] . This shows that TF ((a, b]) ≤ 1 for all a < b
Z β
with b − a < δ. Thus using Eq. (30.14), it follows that TF ((a, b]) ≤ N < ∞ π π
provided N ∈ N is chosen so that b − a < N δ. f 0 (x)dx = f (β) − f (α) = β 2 cos 2 − α2 cos 2 .
n α β α
3. Suppose that {(ai , bi ]}i=1 are disjoint subintervals of (a, b], then by the q
2
√
mean value theorem, Now take αn := 4n+1 and βn = 1/ 2n. Then
n n Z βn
2 π(4n + 1) 1 1
X X
|F (bi ) − F (ai )| ≤ |F 0 (ci )| (bi − ai ) ≤ M · m (∪ni=1 (ai , bi )) f 0 (x)dx = cos − cos 2nπ =
i=1 i=1 αn 4n + 1 2 2n 2n
n Rε 0
∞
and noting that {(αn , βn )}n=1 are all disjoint, we find 0 |f (x)| dx = ∞.
X
≤M (bi − ai ) ≤ M (b − a)
i=1
Theorem 30.28. Let F : R → C be any function.
form which it easily follows that F is absolutely continuous. Moreover we 1. For a < b < c,
may conclude that TF ((a, b]) ≤ M (b − a). TF ((a, c]) = TF ((a, b]) + TF ((b, c]). (30.14)
n
4. Let ν be the positive measure dν = |f | dm on (a, b]. Again let {(ai , bi ]}i=1
be disjoint subintervals of (a, b], then Letting a = α in this expression implies
TF (c) = TF (b) + TF ((b, c]) (30.15)
and in particular TF is monotone increasing.

2. Now suppose F : R → R and F ∈ BV (R) . Then the functions F± := |F (b)| ≤ |F (b) − F (0)| + |F (0)| ≤ TF (0, b] + |F (0)|
(TF ± F ) /2 are bounded and increasing functions. ≤ TF (0, ∞) + |F (0)|
3. (Optional) A function F : R → R is in BV iff F = F+ − F− where F± are
bounded increasing functions. and similarly
|F (a)| ≤ |F (0)| + TF (−∞, 0)
Proof.
which shows that F is bounded by |F (0)| + TF (R). Therefore the functions,
1. (Item 1. is a special case of Exercise 24.5. Nevertheless we will give a proof F+ and F− are bounded as well.
here.) By the triangle inequality, if P and P0 are partition of [a, c] such that 3. By Exercise 30.2 if F = F+ − F− , then
P ⊂ P0 , then
X X TF ((a, b]) ≤ TF+ ((a, b]) + TF− ((a, b])
|F (x+ ) − F (x)| ≤ |F (x+ ) − F (x)| . = |F+ (b) − F+ (a)| + |F− (b) − F− (a)|
x∈P x∈P0
which is bounded showing that F ∈ BV. Conversely if F is bounded varia-
So if P is a partition of [a, c], then P ⊂ P0 := P∪ {b} implies tion, then F = F+ − F− where F± are defined as in Item 1.
X X
|F (x+ ) − F (x)| ≤ |F (x+ ) − F (x)|
x∈P x∈P0
X X Theorem 30.29 (Bounded variation functions). Suppose F : X̄ → C is in
= |F (x+ ) − F (x)| + |F (x+ ) − F (x)| BV (X) , then
x∈P0 ∩(a,b] x∈P0 ∩[b,c]
≤ TF ((a, b]) + TF ((b, c]). 1. F (x+) := limy↓x F (y) and F (x−) := limy↑x F (y) exist for all x ∈ X̄.
By convention, if X ⊂ (α, ∞] then F (α−) = F (α) and if X ⊂ (−∞, β]
Thus we see that then F (β+) := F (β) . Let G(x) := F (x+) and G (±∞) = F (±∞) where
appropriate.
TF ((a, c]) ≤ TF ((a, b]) + TF ((b, c]). 2. If inf X = −∞, then F (−∞) := limx→−∞ F (x) exists and if sup X = +∞
then F (∞) := limx→∞ F (x) exists.
Similarly if P1 is a partition of [a, b] and P2 is a partition of [b, c], then 3. The set of points of discontinuity, {x ∈ X : limy→x F (y) 6= F (x)}, of F is
P = P1 ∪ P2 is a partition of [a, c] and at most countable and in particular G(x) = F (x+) for all but a countable
X X X number x ∈ X.
|F (x+ ) − F (x)|+ |F (x+ ) − F (x)| = |F (x+ ) − F (x)| ≤ TF ((a, c]).
4. For m – a.e. x, F 0 (x) and G0 (x) exist and F 0 (x) = G0 (x).
x∈P1 x∈P2 x∈P
5. The function G is right continuous on X. Moreover, there exists a unique
From this we conclude complex measure, ν = νF , on (X, B) such that, for all a, b ∈ X̄ with a < b,
TF ((a, b]) + TF ((b, c]) ≤ TF ((a, c]) ν ((a, b]) = G (b) − G (a) = F (b+) − F (a+) . (30.17)
which finishes the proof of Eqs. (30.14) and (30.15). 6. F 0 ∈ L1 (X, m) and the Lebesgue decomposition of ν may be written as
2. By Item 1., for all a < b,
dνF = F 0 dm + dνs (30.18)
TF (b) − TF (a) = TF ((a, b]) ≥ |F (b) − F (a)| (30.16)
where νs is a measure singular to m. In particular,
and therefore Z b
TF (b) ± F (b) ≥ TF (a) ± F (a) F (b+) − F (a+) = F 0 dm + νs ((a, b]) (30.19)
a
which shows that F± are increasing. Moreover from Eq. (30.16), for b ≥ 0
and a ≤ 0, whenever a, b ∈ X̄ with a < b.

7. νs = 0 iff F is absolutely continuous on X̄. where the supremum is over all partition of [a, b] . This shows that |ν| = |ν̃| on
A ⊂ B and so by the measure uniqueness Theorem 19.55, |ν| = |ν̃| on B. It now
Proof. If X 6= R, extend F to all of R by requiring F be constant on each follows that |ν| + ν and |ν̃| + ν̃ are finite positive measure on B such that, for
of the connected components of R \ X o . For example if X = [α, β] , extend F all a < b,
to R by setting F (x) := F (α) for x ≤ α and F (x) = F (β) for x ≥ β. With
this extension it is easily seen that TF (R) = TF (X) and TF (x) is constant (|ν| + ν) ((a, b]) = |ν| ((a, b]) + (G(b) − G(a))
on the connected components of R \ X o . Thus we may now assume X = R = |ν̃| ((a, b]) + (G(b) − G(a))
and TF (R) < ∞. Moreover, by considering the real and imaginary parts of F = (|ν̃| + ν̃) ((a, b]) .
separately we may assume F is real valued. So we now assume X = R and
F : R → R is in BV := BV (R) . Hence another application of Theorem 19.55 shows
By Theorem 30.37, the functions F± := (TF ± F ) /2 are bounded and in-
|ν| + ν = |ν̃| + ν̃ = |ν| + ν̃ on B,
creasing functions. Since F = F+ − F− , items 1. – 4. are now easy consequences
of Theorem 30.19 applies to F+ and F− . and hence ν = ν̃ on B.
Let G± (x) := F± (x+) and G± (∞) = F± (∞) and G± (−∞) = F± (−∞) , Alternative proofs of uniqueness of ν. The uniqueness may be proved
then by any number of other means. For example one may directly apply the mul-
G (x) = F (x+) = G+ (x) − G− (x) tiplicative system Theorem 18.51 with RH being the collection of bounded mea-
R
and as in Theorem 30.19 (see Theorem 28.32), there exists unique positive finite surable functions such that R f dν = R f dν̃ and M being the multiplicative
measures, ν± , such that system,
M := 1(a,b] : a, b ∈ R with a < b .
ν± ((a, b]) = G± (b) − G± (a) for all a < b. Alternatively one could apply the monotone class Theorem (Lemma 33.3) with
C := {A ∈ B : ν (A) = ν̃ (A)} and A the algebra of half open intervals. Or one
Then ν := ν+ − ν− is a finite signed measure with the property that could use the π – λ Theorem 33.5, with D = {A ∈ B : ν (A) = ν̃ (A)} and
ν ((a, b]) = G (b) − G (a) = F (b+) − F (a+) for all a < b. C := {(a, b] : a, b ∈ R with a < b} .
Corollary 30.30. If F ∈ BV (X) then νF ⊥ m iff F 0 = 0 m a.e.
Since ν± have Lebesgue decompositions given by
Proof. This is a consequence of Eq. (30.18) and the uniqueness of the
dν± = F±0 dm + d (ν± )s Lebesgue decomposition. In more detail, if F 0 (x) = 0 for m a.e. x, then by Eq.
(30.18), νF = νs ⊥ m. If νF ⊥ m, then by Eq. (30.18), F 0 dm = dνF − dνs ⊥ dm
with F±0 ∈ L1 (m) and (ν± )s ⊥ m, it follows that and by Lemma 24.8 F 0 dm = 0, i.e. F 0 = 0 m -a.e.
dν = F+0 − F−0 dm + dνs = F 0 dm + dνs Corollary 30.31. Let F : X̄ → C be a right continuous function in BV (X) ,

νF be the associated complex measure and
with F 0 = F+0 − F−0 (m -a.e.), F 0 ∈ L1 (R, m) and νs ⊥ m, where
dνF = F 0 dm + dνs (30.20)
νs := (ν+ )s − (ν− )s .
be the its Lebesgue decomposition. Then the following are equivalent,
Thus we have proven everything but the uniqueness of the measure ν satisfying 1. F is absolutely continuous,
Eq. (30.17). 2. νF m,
Uniqueness of ν. So it only remains to prove that ν is unique. Suppose 3. νs = 0, and
that ν̃ is another such measure such that Eq. (30.17) holds with ν replaced by 4. for all a, b ∈ X with a < b,
ν̃. Then for (a, b] ⊂ R, Z
X F (b) − F (a) = F 0 (x)dm(x). (30.21)
|ν| (a, b] = sup |G(x+ ) − G(x)| = |ν̃| (a, b]
P (a,b]
x∈P

n
Proof. The equivalence of 1. and 2. was established in Proposition 30.23 Proof. Suppose that {(ai , bi ]}i=1 is a sequence of disjoint intervals in [α, β],
and the equivalence of 2. and 3. is trivial. (If νF m, then dνs = dνF − then
F 0 dm dm which implies, by Lemma 24.26, that νs = 0.) If νF m and n
X
G (x) := F (x+) , then the identity, |F (bi )G(bi ) − F (ai )G(ai )|
i=1
Z b
n n
F (b) − F (a) = F (b+) − F (a−) = F 0 (x) dm (x) , X X
a ≤ |F (bi )| |G(bi ) − G(ai )| + |F (bi ) − F (ai )| |G(ai )|
i=1 i=1
implies F is continuous. n
X n
X
(The equivalence of 4. and 1., 2., and 3.) If F is absolutely continuous, ≤ kF k∞ |G(bi ) − G(ai )| + kGk∞ |F (bi ) − F (ai )| .
then νs = 0 and Eq. (30.21) follows from Eq. (30.20). Conversely let i=1 i=1
Z From this inequality, one easily deduces the absolutely continuity of the product
ρ(A) := F 0 (x)dm(x) for all A ∈ B. F G from the absolutely continuity of F and G. Therefore,
A Z β Z β
0
Recall by the Radon - Nikodym theorem that R |F 0 (x)| dm(x) < ∞ so that ρ
R β
F G|α = (F G) dm = (F 0 G + F G0 )dm.
α α
is a complex measure on B. So if Eq. (30.21) holds, then ρ = νF on the algebra
generated by half open intervals. Therefore ρ = νF as in the uniqueness part of
the proof of Theorem 30.29. Therefore dνF = F 0 dm dm.
30.4.3 Alternative method to Proving Theorem 30.29
Theorem 30.32 (The fundamental theorem of calculus). Suppose that
F : [α, β] → C is a measurable function. Then the following are equivalent: For simplicity assume that α = −∞, β = ∞, F ∈ BV,
1. F is absolutely continuous on [α, β]. Ab := {A ∈ A : A is bounded} ,

2. There exists f ∈ L1 ([α, β]), dm) such that Pn
and Sc (A) denote simple functions of the form f = i=1 λi 1Ai with Ai ∈ Ab .
Let ν 0 = νF0 be the finitely additive set function on such that ν 0 ((a, b]) =
Z x
F (x) − F (α) = f dm ∀x ∈ [α, β] (30.22) F (b) − F (a) for all −∞ < a < b < ∞. As in the case of an increasing function
α
F (see Lemma 28.38 and the text preceding it) we may define a linear functional,
3. F 0 exists a.e., F 0 ∈ L1 ([α, β], dm) and IF : Sc (A) → C, by X
Z x IF (f ) = λν 0 (f = λ).
F (x) − F (α) = F 0 dm ∀x ∈ [α, β]. (30.23) λ∈C
α PN N
If we write f = i=1 λi 1(ai ,bi ] with {(ai , bi ]}i=1 pairwise disjoint subsets of Ab
Proof. 1. =⇒ 3. If F is absolutely continuous then F ∈ BV ((α, β]) and F is inside (a, b] we learn
continuous on [α, β]. Hence Eq. (30.23) holds by Corollary 30.31. The assertion
N N
3. =⇒ 2. is trivial and we have already seen in Lemma 30.26 that 2. implies 1. X X
|IF (f )| = λi (F (bi ) − F (ai ) ≤ |λi | |F (bi ) − F (ai )| ≤ kf k∞ TF ((a, b]).

i=1 i=1
Corollary 30.33 (Integration by parts). Suppose −∞ < α < β < ∞ and (30.24)
F, G : [α, β] → C are two absolutely continuous functions. Then In the usual way this estimate allows us to extend IF to the those compactly
supported functions, Sc (A), in the closure of Sc (A). As usual we will still denote
Z β Z β the extension of IF to Sc (A) by IF and recall that Sc (A) contains Cc (R, C).
F 0 Gdm = − F G0 dm + F G|βα . The estimate in Eq. (30.24) still holds for this extension and in particular we
α α
have

30.5 The connection of Weak and pointwise derivatives 337
|I(f )| ≤ TF (∞) · kf k∞ for all f ∈ Cc (R, C).

∗
Therefore I extends uniquely by continuity to an element of C0 (R, C) . So by
appealing to the complex Riesz Theorem (Corollary 32.68) there exists a unique
complex measure ν = νF such that
Z
IF (f ) = f dν for all f ∈ Cc (R). (30.25)
R
This leads to the following theorem.

Theorem 30.34. To each function F ∈ BV there exists a unique measure
ν = νF on (R, BR ) such that Eq. (30.25) holds. Moreover, F (x+) = limy↓x F (y)
exists for all x ∈ R and the measure ν satisfies
ν((a, b]) = F (b+) − F (a+) for all − ∞ < a < b < ∞. (30.26)
Remark 30.35. By applying Theorem 30.34 to the function x → F (−x) one

shows every F ∈ BV has left hand limits as well, i.e F (x−) = limy↑x F (y)
exists for all x ∈ R.
Proof. We must still prove F (x+) exists for all x ∈ R and Eq. (30.26) Fig. 30.5. A couple of functions in Sc (A).
holds. To prove let ψb and φε be the functions shown in Figure 30.5 below. The
reader should check that ψb ∈ Sc (A). Notice that
Similarly this equation holds with b replaced by a, i.e.
IF (ψb+ε ) = IF (ψα + 1(α,b+ε] ) = IF (ψα ) + F (b + ε) − F (α)
Z
and since kφε − ψb+ε k∞ = 1, ψα dν + ν((α, a]) = IF (ψα ) + F (a+) − F (α).
R
|I(φε ) − IF (ψb+ε )| = |IF (φε − ψb+ε )|
Subtracting the last two equations proves Eq. (30.26).
≤ TF ([b + ε, b + 2ε]) = TF (b + 2ε) − TF (b + ε),
Remark 30.36. Given Theorem 30.34 we may now prove Theorem 30.29 in the
which implies O (ε) := I(φε ) − IF (ψb+ε ) → 0 as ε ↓ 0 because TF is monotonic. same we proved Theorem 30.19.
Therefore,
I(φε ) = IF (ψb+ε ) + I(φε ) − IF (ψb+ε )

= IF (ψα ) + F (b + ε) − F (α) + O (ε) . (30.27)
30.5 The connection of Weak and pointwise derivatives
Because φε converges boundedly to ψb as ε ↓ 0, the dominated convergence Theorem 30.37. Suppose Let Ω ⊂ R be an open interval and f ∈ L1loc (Ω).
theorem implies Then there exists a complex measure µ on BΩ such that
Z Z Z Z
0
lim I(φε ) = lim φε dν = ψb dν = ψα dν + ν((α, b]). − hf, φ i = µ(φ) := φdµ for all φ ∈ Cc∞ (Ω) (30.28)
ε↓0 ε↓0 R R R Ω
So we may let ε ↓ 0 in Eq. (30.27) to learn F (b+) exists and iff there exists a right continuous function F of bounded variation such that
Z F = f a.e. In this case µ = µF , i.e. µ((a, b]) = F (b) − F (a) for all −∞ < a <
ψα dν + ν((α, b]) = IF (ψα ) + F (b+) − F (α). b < ∞.
R

Proof. Suppose f = F a.e. where F is as above and let µ = µF be the dY (u(x), u(x0 )) ≤ CdX (x, x0 ) for all x, x0 ∈ X
associated measure on BΩ . Let G(t) = F (t) − F (−∞) = µ((−∞, t]), then using
Fubini’s theorem and the fundamental theorem of calculus, and said to be locally Lipschitz if for all compact subsets K ⊂ X there exists
Z Z CK < ∞ such that
−hf, φ0 i = −hF, φ0 i = −hG, φ0 i = − φ0 (t) 1(−∞,t] (s)dµ(s) dt dY (u(x), u(x0 )) ≤ CK dX (x, x0 ) for all x, x0 ∈ K.
Ω
Z Z ZΩ
=− 0
φ (t)1(−∞,t] (s)dtdµ(s) = φ(s)dµ(s) = µ(φ). Proposition 30.40. Let u ∈ L1loc (Ω). Then there exists a locally Lipschitz
Ω Ω Ω function ũ : Ω → C such that ũ = u a.e. iff (weak − ∂i u) ∈ L1loc (Ω) exists
and is locally (essentially) bounded for i = 1, 2, . . . , d.
Conversely if Eq. (30.28) holds for some measure µ, let F (t) := µ((−∞, t]) then
working backwards from above, Proof. Suppose u = ũ a.e. and ũ is Lipschitz and let p ∈ (1, ∞) and V be a
Z Z Z precompact open set such that V̄ ⊂ W and let Vε := x ∈ Ω : dist(x, V̄ ) ≤ ε .
−hf, φ0 i = µ(φ) = φ(s)dµ(s) = − φ0 (t)1(−∞,t] (s)dtdµ(s) Then for ε < dist(V̄ , Ω c ), Vε ⊂ Ω and therefore there is constant C(V, ε) < ∞
Z Ω Ω Ω such that |ũ(y) − ũ(x)| ≤ C(V, ε) |y − x| for all x, y ∈ Vε . So for 0 < |h| ≤ 1
and v ∈ Rd with |v| = 1,
=− φ0 (t)F (t)dt.
Ω
u(x + hv) − u(x) p ũ(x + hv) − ũ(x) p
Z Z
dx = dx ≤ C(V, ε) |v|p .
This shows ∂ (w) (f − F ) = 0 and therefore by Proposition 26.25, f = F + c a.e. V
h
V
h
for some constant c ∈ C. Since F +c is right continuous with bounded variation,
the proof is complete. Therefore Theorem 26.18 may be applied to conclude ∂v u exists in Lp and
moreover,
Proposition 30.38. Let Ω ⊂ R be an open interval and f ∈ L1loc (Ω). Then
ũ(x + hv) − ũ(x)
∂ w f exists in L1loc (Ω) iff f has a continuous version f˜ which is absolutely con- lim = ∂v u(x) for m – a.e. x ∈ V.
h→0 h
tinuous on all compact subintervals of Ω. Moreover, ∂ w f = f˜0 a.e., where f˜0 (x)
∞
is the usual pointwise derivative. Since there exists {hn }n=1 ⊂ R\ {0} such that limn→∞ hn = 0 and

Proof. If f is locally absolutely continuous and φ ∈ Cc∞ (Ω) with supp(φ) ⊂ ũ(x + hn v) − ũ(x)
|∂v u(x)| = lim ≤ C(V ) for a.e. x ∈ V,
[a, b] ⊂ Ω, then by integration by parts, Corollary 30.33, n→∞ hn
Z
0
Z b
0
Z b
0
Z it follows that k∂v uk∞ ≤ C(V ) where C(V ) := limε↓0 C(V, ε).
f φdm = f φdm = − f φ dm + f φ|ba =− f φ0 dm. Conversely, let Ωε := {x ∈ Ω : dist(x, Ω c ) > ε} and η ∈ Cc∞ (B(0, 1), [0, ∞))
Ω a a Ω
such that Rn η(x)dx = 1, ηm (x) = mn η(mx) and um := u∗ηm as in the proof of
R
This shows ∂ w f exists and ∂ w f = f 0 ∈ L1loc (Ω). Now Rsuppose that ∂ w f exists Theorem 26.18. Suppose V ⊂o Ω with V̄ ⊂ Ω and ε is sufficiently small. Then
x
in L1loc (Ω) and a ∈ Ω. Define F ∈ C (Ω) by F (x) := a ∂ w f (y)dy. Then F is um ∈ C ∞ (Ωε ), ∂v um = ∂v u ∗ ηm , |∂v um (x)| ≤ k∂v ukL∞ (V −1 ) =: C(V, m) < ∞
m
absolutely continuous on compacts and therefore by fundamental theorem of and therefore for x, y ∈ V̄ with |y − x| ≤ ε,
calculus for absolutely continuous functions (Theorem 30.32), F 0 (x) exists and Z 1
is equal to ∂ w f (x) for a.e. x ∈ Ω. Moreover, by the first part of the argument, d
|um (y) − um (x)| =
um (x + t(y − x))dt
∂ w F exists and ∂ w F = ∂ w f, and so by Proposition 26.25 there is a constant c 0 dt
such that Z 1

f˜(x) := F (x) + c = f (x) for a.e. x ∈ Ω. = (y − x) · ∇um (x + t(y − x))dt
0
Z 1
≤ |y − x| · |∇um (x + t(y − x))| dt ≤ C(V, m) |y − x| .
Definition 30.39. Let X and Y be metric spaces. A function u : X → Y is 0
said to be Lipschitz if there exists C < ∞ such that (30.29)

30.5 The connection of Weak and pointwise derivatives 339
By passing to a subsequence if necessary, we may assume that limm→∞ um (x) = Theorem 30.41. Let p ≥ 1 and Ω be an open subset of Rd , x ∈ Rd be written
u(x) for m – a.e. x ∈ V̄ and then letting m → ∞ in Eq. (30.29) implies as x = (y, t) ∈ Rd−1 × R,
Y := y ∈ Rd−1 : ({y} × R) ∩ Ω 6= ∅

|u(y) − u(x)| ≤ C(V ) |y − x| for all x, y ∈ V̄ \ E and |y − x| ≤ ε (30.30)
where E ⊂ V̄ is a m – null set. Define ũV : V̄ → C by ũV = u on V̄ \ E and and u ∈ Lp (Ω). Then ∂t u exists weakly in Lp (Ω) iff there is a version ũ of
ũV (x) = limy→x u(y) if x ∈ E. Then clearly ũV = u a.e. on V̄ and it is easy to u such that for a.e. y ∈ Y the function t → ũ(y, t) is absolutely continuous,
∂t u(y, t) = ∂ ũ(y,t)
∂ ũ
a.e., and ∂t L (Ω) < ∞.
y ∈E
/ p
∂t
show ũV is well defined and ũV : V̄ → C is continuous and still satisfies
Proof. For the proof of Theorem 30.41, it suffices to consider the case where
|ũV (y) − ũV (x)| ≤ CV |y − x| for x, y ∈ V̄ with |y − x| ≤ ε. Ω = (0, 1)d . Write x ∈ Ω as x = (y, t) ∈ Y × (0, 1) = (0, 1)d−1 × (0, 1) and ∂t u
for the weak derivative ∂ed u. By assumption
Since ũV is continuous on V̄ there exists MV < ∞ such that |ũV | ≤ MV on V̄ .
Hence if x, y ∈ V̄ with |x − y| ≥ ε, we find
Z
|∂t u(y, t)| dydt = k∂t uk1 ≤ k∂t ukp < ∞
Ω
|ũV (y) − ũV (x)| 2M
≤ and so by Fubini’s theorem there exists a set of full measure, Y0 ⊂ Y, such that
|y − x| ε
Z 1
and hence |∂t u(y, t)| dt < ∞ for y ∈ Y0 .
0
2MV
|ũV (y) − ũV (x)| ≤ max CV , |y − x| for x, y ∈ V̄ Rt
ε So for y ∈ Y0 , the function v(y, t) := 0 ∂t u(y, τ )dτ is well defined and absolutely
continuous in t with ∂t ∂
v(y, t) = ∂t u(y, t) for a.e. t ∈ (0, 1). Let ξ ∈ Cc∞ (Y ) and
showing ũV is Lipschitz on V̄ . To complete the proof, choose precompact open ∞
η ∈ Cc ((0, 1)) , then integration by parts for absolutely functions implies
sets Vn such that Vn ⊂ V̄n ⊂ Vn+1 ⊂ Ω for all n and for x ∈ Vn let ũ(x) :=
Z 1 Z 1
ũVn (x). ∂
Alternative way to construct the function ũV . For x ∈ V \ E, v(y, t)η̇(t)dt = − v(y, t)η(t)dt for all y ∈ Y0 .
0 0 ∂t
Z Z
Multiplying both sides of this equation by ξ(y) and integrating in y shows
|um (x) − u(x)| = u(x − y)η(my)mn dy − u(x) = [u(x − y/m) − u(x)] η(y)dy

ZV ZV
Z Z
∂
C v(x)η̇(t)ξ(y)dydt = − v(y, t)η(t)ξ(y)dydt
≤ |u(x − y/m) − u(x)| η(y)dy ≤ |y| η(y)dy Ω Ω ∂t
V m V Z
=− ∂t u(y, t)η(t)ξ(y)dydt.
wherein the last equality we have Rused Eq. (30.30) with V replaced by Vε for Ω
some small ε > 0. Letting K := C V |y| η(y)dy < ∞ we have shown
Using the definition of the weak derivative, this equation may be written as
kum − uk∞ ≤ K/m → 0 as m → ∞ Z Z
u(x)η̇(t)ξ(y)dydt = − ∂t u(x)η(t)ξ(y)dydt
Ω Ω
and consequently
and comparing the last two equations shows
kum − un k∞ = kum − un k∞ ≤ 2K/m → 0 as m → ∞. Z
[v(x) − u(x)] η̇(t)ξ(y)dydt = 0.
Therefore, un converges uniformly to a continuous function ũV . Ω
The next theorem is from Chapter 1. of Maz’ja [15].
Since ξ ∈ Cc∞ (Y
) is arbitrary, this implies there exists a set Y1 ⊂ Y0 of full
measure such that

Z
[v(y, t) − u(y, t)] η̇(t)dt = 0 for all y ∈ Y1 Exercise 30.6. Folland 3.27 on p. 107.
Ω
from which we conclude, using Proposition 26.25, that u(y, t) = v(y, t) + C(y)
for t ∈ Jy where md−1 (Jy ) = 1, here mk denotes k – dimensional Lebesgue Exercise 30.8. Folland 3.30 on p. 107.
measure. In conclusion we have shown that
Z t Exercise 30.9. Folland 3.33 on p. 108.
u(y, t) = ũ(y, t) := ∂t u(y, τ )dτ + C(y) for all y ∈ Y1 and t ∈ Jy . (30.31) Exercise 30.10. Folland 3.35 on p. 108.
0
We can be more precise about the formula for ũ(y, t) by integrating both sides Exercise 30.11. Folland 3.37 on p. 108.
of Eq. (30.31) on t we learn
Z 1 Z t Z 1
C(y) = dt ∂τ u(y, τ )dτ − u(y, t)dt Exercise 30.13. Folland 3.40 on p. 108.
0 0 0
Z 1 Z 1 Exercise 30.14. Folland 8.4 on p. 239.
= (1 − τ ) ∂τ u(y, τ )dτ − u(y, t)dt
0 0
Z 1
= [(1 − t) ∂t u(y, t) − u(y, t)] dt
0
and hence
Z t Z 1
ũ(y, t) := ∂τ u(y, τ )dτ + [(1 − τ ) ∂τ u(y, τ ) − u(y, τ )] dτ
0 0
which is well defined for y ∈ Y0 . For the converse suppose that such a ũ exists,
then for φ ∈ Cc∞ (Ω) ,
Z Z
u(y, t)∂t φ(y, t)dydt = ũ(y, t)∂t φ(y, t)dtdy
Ω Ω
Z
∂ ũ(y, t)
=− φ(y, t)dtdy
Ω ∂t
wherein we have used integration by parts for absolutely continuous functions.
From this equation we learn the weak derivative ∂t u(y, t) exists and is given
by ∂ ũ(y,t)
∂t a.e.
30.6 Exercises

31
Constructing Measures Via Carathéodory
The main goals of this chapter is to prove the two measure construction Proof.
` Since (by Proposition 18.10) every element A ∈ A is of the form
Theorems 28.2 and 28.16. Throughout this chapter, X will be a given set. The A = i Ei for a finite collection of Ei ∈ E, it is clear that if µ extends to a
following definition is a continuation of the terminology introduced in Definition measure then the extension is unique and must be given by
28.1. X
µ(A) = µ(Ei ). (31.3)
Definition 31.1. Suppose that E ⊂ 2X is a collection of subsets of X and i
µ : E → [0, ∞] is a function. Then To prove existence, the main point
` is to show that µ(A) in Eq. (31.3) is well
1. µ is super-additive (finitely super-additive) on E if defined; i.e. if we also have A = j Fj with Fj ∈ E, then we must show
X X
n
X µ(Ei ) = µ(Fj ). (31.4)
µ(E) ≥ µ(Ei ) (31.1) i j
i=1 `
`n P Ei = j (Ei ∩ Fj ) and the property that µ is additive on E implies µ(Ei ) =
But
whenever E = i=1 Ei ∈ E with n ∈ N∪ {∞} (n ∈ N). j µ(Ei ∩ Fj ) and hence
2. µ is monotonic if µ (A) ≤ µ (B) for all A, B ∈ E with A ⊂ B. X XX X
µ(Ei ) = µ(Ei ∩ Fj ) = µ(Ei ∩ Fj ).
Remark 31.2. If E = A is an algebra and µ is finitely additive on A, then µ is i i j i,j
sub-additive on A iff
Similarly, or by symmetry,
∞
X ∞
a X X
µ(A) ≤ µ(Ai ) for A = Ai (31.2) µ(Fj ) = µ(Ei ∩ Fj )
i=1 i=1 j i,j
∞ S∞
which combined with the previous equation shows that Eq. (31.4) holds. It is
where A ∈ A and {Ai }i=1 ⊂ A are`pairwise disjoint sets. Indeed if A = i=1 Bi
∞ now easy to verify that µ extended to A as in Eq. (31.3) is an additive measure
with A ∈ A and Bi ∈ A, then A = i=1 Ai where Ai := Bi \(B1 ∪ . . . Bi−1 ) ∈ A
and B0 = ∅. Therefore using the monotonicity of µ and Eq. (31.2) on A.
∞
X ∞
X Proposition 31.4. Suppose that A ⊂ 2X is an algebra and µ : A → [0, ∞] is a
µ(A) ≤ µ(Ai ) ≤ µ(Bi ). finitely additive measure on A. Then µ is automatically super-additive on A.
i=1 i=1
Proof. Since ! !
N
a N
[
A= Ai ∪ A\ Ai ,
31.1 Construction of Premeasures i=1 i=1
N N
! N
X [ X
Proposition 31.3 (Construction of Finitely Additive Measures). Sup- µ(A) = µ(Ai ) + µ A \ Ai ≥ µ(Ai ).
pose E ⊂ 2X is an elementary family (see Definition 18.8) and A = A(E) is i=1 i=1 i=1
the algebra generated by E. Then every additive function µ : E → [0, ∞] extends ∞
P
uniquely to an additive measure (which we still denote by µ) on A. Letting N → ∞ in this last expression shows that µ(A) ≥ µ(Ai ).
i=1
342 31 Constructing Measures Via Carathéodory
∞
Proposition 31.5. Suppose that E ⊂ 2X is an elementary family, A = A(E) Proof. Suppose {An }n=1 ⊂ A, A0 := ∅, and A = ∪∞ n=1 An ∈ Aσ . By
and µ : A → [0, ∞] is a finitely additive measure. Then µ is a premeasure on replacing each An by An \ (A1 ∪ · · · ∪ An−1 ) if necessary we may assume that
∞
A iff µ is sub-additive on E. collection of sets {An }n=1 are pairwise disjoint. Hence every element A ∈ Aσ
∞
`
Proof. Clearly if µ is a premeasure on A then µ is σ-additive and hence may be expressed as a disjoint union, A = An with An ∈ A. With A
n=1
sub-additive on E. Because of Proposition 31.4, to prove the converse it suffices expressed this way we must define
to show that the sub-additivity of µ on E implies the sub-additivity of µ on A.
∞
` ∞
So suppose A = An with A ∈ A and each An ∈ A which we express as
X
n=1
µ (A) := µ (An ) .
`k `Nn n=1
A = j=1 Ej with Ej ∈ E and An = i=1 En,i with En,i ∈ E. Then
The proof that µ (A) is well defined follows the same argument used in the proof
∞
a ∞ a
a Nn ∞
`
Ej = A ∩ Ej = An ∩ Ej = En,i ∩ Ej of Proposition 31.3. Explicitly, suppose also that A = Bk with Bk ∈ A, then
k=1
n=1 n=1 i=1 `∞
for each n, An = k=1 (An ∩ Bk ) and therefore because µ is a premeasure,
which is a countable union and hence by assumption,
∞
X
∞ X
X Nn µ(An ) = µ(An ∩ Bk ).
µ(Ej ) ≤ µ (En,i ∩ Ej ) . k=1
n=1 i=1
Summing this equation on n shows,
Summing this equation on j and using the finite additivity of µ shows
∞
X ∞ X
X ∞ ∞ X
X ∞
k
X ∞ X
k X
X Nn µ(An ) = µ(An ∩ Bk ) = µ(An ∩ Bk )
µ(A) = µ(Ej ) ≤ µ (En,i ∩ Ej ) n=1 n=1 k=1 k=1 n=1
j=1 j=1 n=1 i=1
∞ X
Nn X
k ∞ X
Nn ∞
wherein the last equality we have used Tonelli’s theorem for sums. By symmetry
=
X
µ (En,i ∩ Ej ) =
X
µ (En,i ) =
X
µ (An ) , we also have
X∞ X∞ X∞
n=1 i=1 j=1 n=1 i=1 n=1 µ(Bk ) = µ(An ∩ Bk )
k=1 k=1 n=1
which proves (using Remark 31.2) the sub-additivity of µ on A. P∞ P∞
and comparing the last two equations gives n=1 µ(An ) = k=1 µ(Bk ) which
shows the extension of µ to Aσ is well defined.
31.1.1 Extending Premeasures to Aσ ∞
Countable additive of µ on Aσ . If {An }n=1 is a collectionòf pairwise
∞
Proposition 31.6. Let µ be a premeasure on an algebra A, then µ has a unique disjoint subsets of Aσ , then there exists Ani ∈ A such that An = i=1 Ani for
extension (still called µ) to a countably additive function on Aσ . Moreover the all n, and therefore,
extended function µ satisfies the following properties.1 
∞

∞
a X
1. (Continuity) If An ∈ A and An ↑ A ∈ Aσ , then µ (An ) ↑ µ (A) as n → ∞. µ (∪∞ A
n=1 n ) = µ  Ani
 := µ (Ani )
2. (Strong Additivity) If A, B ∈ Aσ , then i,n=1 i,n=1
∞ X
X ∞ ∞
X
µ (A ∪ B) + µ (A ∩ B) = µ (A) + µ (B) . (31.5) = µ (Ani ) = µ (An ) .
n=1 i=1 n=1
3. (Sub-Additivity on Aσ ) The function µ is sub-additive on Aσ .
1
Again there are no problems in manipulating the above sums since all summands
The remaining results in this proposition may be skipped in which case the reader are non-negative.
should also skip Section 31.3.

31.2 Outer Measures 343
ǹ
(∞ ∞
)
Continuity of µ. Suppose An ∈ A and An ↑ A ∈ Aσ . Then An = Bi
X [
i=1
ρ∗ (A) = inf ρ(Ei ) : A ⊂ Ei with Ei ∈ E . (31.7)
∞
` i=1 i=1
and A = Bi where Bn := An \ (A1 ∪ · · · ∪ An−1 ) ∈ A. So by definition of
i=1 Then ρ∗ is an outer measure.
µ (A) ,
∞ n Proof. It is clear that ρ∗ is monotonic and ρ∗ (∅) = 0. Suppose for i ∈ N,
Ai ∈ 2X and ρ∗ (Ai ) < ∞; otherwise there will be nothing to prove. Let ε > 0
X X
µ (A) = µ (Bi ) = lim µ (Bi ) = lim µ (An )
n→∞ n→∞ ∞ ∞
i=1 i=1 Eij and ρ∗ (Ai ) ≥ ρ(Eij ) − 2−i ε.
S P
and choose Eij ∈ E such that Ai ⊂
which proves the continuity assertion. j=1 j=1
∞ ∞
Strong additivity of µ. Let A and B be in Aσ and choose An , Bn ∈ A Since
S
Ai ⊂
S
Eij ,
such that An ↑ A and Bn ↑ B as n → ∞ then i=1 i,j=1
∞ ∞
∞ X ∞ ∞
!
µ (An ∪ Bn ) + µ (An ∩ Bn ) = µ (An ) + µ (Bn ) . (31.6) ∗
[ X X X
ρ Ai ≤ ρ(Eij ) ≤ (ρ∗ (Ai ) + 2−i ε) = ρ∗ (Ai ) + ε.
Indeed if µ (An ) + µ (Bn ) = ∞ the identity is true because ∞ = ∞ and if i=1 i=1 j=1 i=1 i=1
µ (An ) + µ (Bn ) < ∞ the identity follows from the finite additivity of µ on A Since ε > 0 is arbitrary in this inequality, we have shown ρ∗ is sub-additive.
and the set identity, The following lemma is an easy consequence of Proposition 31.6 and the
remarks in the proof of Theorem 28.2.
An ∪ Bn = [An ∩ Bn ] ∪ [An \ (An ∩ Bn )] ∪ [Bn \ (An ∩ Bn )] .
Lemma 31.9. Suppose that µ is a premeasure on an algebra A and µ∗ is the
Since An ∪ Bn ↑ A ∪ B and An ∩ Bn ↑ A ∩ B, Eq. (31.5) follows by passing to outer measure associated to µ as in Proposition 31.8. Then
the limit as n → ∞ in Eq. (31.6) while making use of the continuity property
of µ. µ∗ (B) = inf {µ (C) : B ⊂ C ∈ Aσ } ∀ B ⊂ X
Sub-Additivity on Aσ` . Suppose An ∈ Aσ and A = ∪∞ n=1 An . Choose and µ∗ = µ on A, where µ has been extended to Aσ as described in Proposition
∞ ∞
An,j ∈ A such that An := j=1 An,j , let {Bk }k=1 be an enumeration of the 31.6.
collection of sets, {An,j : n, j ∈ N} , and define Ck := `
Bk \(B1 ∪ · · · ∪ Bk−1 ) ∈ A
∞ Lemma 31.10. Suppose (X, τ ) is a locally compact Hausdorff space, I is a
with the usual convention that B0 = ∅. Then A = k=1 Ck and therefore by
the definition of µ on Aσ and the monotonicity of µ on A, positive linear functionals on Cc (X), and let µ : τ → [0, ∞] be defined in Eq.
(28.8). Then µ is sub-additive on τ and the associate outer measure, µ∗ : 2X →
∞ ∞ ∞ X
∞ ∞
X X X X [0, ∞] associated to µ as in Proposition 31.8 may be described by
µ (A) = µ (Ck ) ≤ µ (Bk ) = µ (An,j ) = µ (An ) .
k=1 k=1 n=1 j=1 n=1 µ∗ (E) = inf {µ (U ) : E ⊂ U ⊂o X} . (31.8)
In particular µ∗ = µ on τ.
In future we will tacitly assume that any premeasure, µ, on an algebra, A, ∞
Proof. Let {Uj }j=1 ⊂ τ, U := ∪∞ j=1 Uj , f ≺ U and K = supp(f ). Since K is
has been extended to Aσ as described in Proposition 31.6.
compact, K ⊂ ∪nj=1 Uj for some n ∈ N sufficiently large. By Proposition 15.16
Pn
(partitions of unity proposition) we may choose hj ≺ Uj such that j=1 hj = 1
Pn
31.2 Outer Measures on K. Since f = j=1 hj f and hj f ≺ Uj ,
n n ∞
Definition 31.7. A function ν : 2X → [0, ∞] is an outer measure if ν(∅) = 0,
X X X
I (f ) = I (hj f ) ≤ µ (Uj ) ≤ µ (Uj ) .
ν is monotonic and sub-additive. j=1 j=1 j=1
P∞
Proposition 31.8 (Example of an outer measure.). Let E ⊂ 2X be arbi- Since this is true for all f ≺ U we conclude µ (U ) ≤ j=1 µ (Uj ) proving
trary collection of subsets of X such that ∅, X ∈ E. Let ρ : E → [0, ∞] be a the countable sub-additivity of µ on τ. The remaining assertions are a direct
function such that ρ(∅) = 0. For any A ⊂ X, define consequence of this sub-additivity.

31.3 *The σ – Finite Extension Theorem µ∗ (B) = sup {µ (A) : Aδ 3 A ⊂ B}

= sup {µ (X) − µ (Ac ) : Aδ 3 A ⊂ B}
This section may be skipped (at the loss of some motivation), since the results
= sup {µ (X) − µ (C) : Aδ 3 C c ⊂ B}
here will be subsumed by those in Section 31.4 below.
= µ (X) − inf {µ (C) : B c ⊂ C ∈ Aσ }
Notation 31.11 (Inner Measure) If µ is a finite (i.e. µ (X) < ∞) premea- = µ (X) − µ∗ (B c ) .
sure on an algebra A, we extend µ to Aδ by defining
Thus the condition that µ∗ (B) = µ∗ (B) is equivalent to requiring that
µ (A) := µ (X) − µ (Ac ) . (31.9)
µ (X) = µ∗ (X) = µ∗ (B c ) + µ∗ (B) . (31.13)
(Note: µ (Ac ) is defined since Ac ∈ Aσ .) Also let
3. By definition B ⊂ X iff µ∗ (B) = µ∗ (B) which happens iff for each ε > 0
µ∗ (B) := sup {µ (A) : Aδ 3 A ⊂ B} ∀ B ⊂ X there exists A ∈ Aδ and C ∈ Aσ such that A ⊂ B ⊂ C and µ (C) − µ (A) < ε;
i.e. by item 1, µ (C \ A) < ε. The containment, A ⊂ M, follows from what we
and define have just proved or is a direct consequence of µ being additive on A and the
M = M (µ) := {B ⊂ X : µ∗ (B) = µ∗ (B)} (31.10) fact that µ∗ = µ∗ = µ on A.
∗
and µ̄ := µ |M . In words, B is in M iff B may be well approximated from both 4. Suppose A, B ∈ Aδ are disjoint sets, then by the strong additivity of µ
inside and out by sets µ can measure. on Aσ (use Eq. (31.5) with A and B being replaced by Ac and B c respectively)
gives
Remark 31.12. If A ∈ Aσ ∩ Aδ , then A, Ac ∈ Aσ and so by the strong additivity
c
of µ, µ (A) + µ (Ac ) = µ (X) from which it follows that the extension of µ to 2µ (X) − µ (A ∪ B) = µ (X) + µ ([A ∪ B] ) = µ (Ac ∪ B c ) + µ (Ac ∩ B c )
Aδ is consistent with the extension of µ to Aσ . = µ (Ac ) + µ (B c ) = 2µ (X) − µ (A) − µ (B) ,
Lemma 31.13. Let µ be a finite premeasure on an algebra A ⊂ 2X and con- i.e. µ (A ∪ B) = µ (A) + µ (B) .
tinue the setup in Notation 31.11.
Theorem 31.14 (Finite Premeasure Extension Theorem). If µ is a finite
1. If A ∈ Aδ and C ∈ Aσ with A ⊂ C, then premeasure on an algebra A, then M = M (µ) (as in Eq. (31.10)) is a σ –
algebra, A ⊂ M and µ̄ = µ∗ |M is a countably additive measure such that µ̄ = µ
µ (C \ A) = µ (C) − µ (A) . (31.11) on A.
2. For all B ⊂ X, µ∗ (B) = µ (X) − µ∗ (B c ) , and Proof. By Lemma 31.13, φ, X ∈ A ⊂ M and from Eq. (31.12) it follows that
M is closed under complementation. Now suppose N ∈ {2, 3, . . . } ∪ {∞} and
M := {B ⊂ X : µ (X) = µ∗ (B) + µ∗ (B c )} . (31.12) Bi ∈ M for i < N. Given ε > 0, by Lemma 31.13 there exists Ai ⊂ Bi ⊂ Ci
with Ai ∈ Aδ and Ci ∈ Aσ such that µ (Ci \ Ai ) < ε2−i for all i < N. Let
3. As subset B ⊂ X is in M iff for all ε > 0 there exists A ∈ Aδ and C ∈ Aσ
B = ∪i<N Bi , C := ∪i<N Ci and A := ∪i<N Ai so that A ⊂ B ⊂ C ∈ Aσ .
such that A ⊂ B ⊂ C and µ (C \ A) < ε. In particular A ⊂ M.
For the moment assume N < ∞, then A ∈ Aδ , C \ A = C ∩ Ac ∈ Aσ ,
4. µ is additive on Aδ .
C \ A = ∪i<N (Ci \ A) ⊂ ∪i<N (Ci \ Ai ) ∈ Aσ
Proof. 1. The strong additivity Eq. (31.5) with B = C ∈ Aσ and A being
replaced by Ac ∈ Aσ implies and so by the sub-additivity of µ on Aσ (Proposition 31.6),
µ (Ac ∪ C) + µ (C \ A) = µ (Ac ) + µ (C) .
X X
µ (C \ A) ≤ µ (Ci \ Ai ) < ε2−i < ε.
i<N i<N
Since X = Ac ∪ C and µ (Ac ) = µ (X) − µ (A) , the previous equality implies
Eq. (31.11). Since ε > 0 was arbitrary, it follows again by Lemma 31.13 that B ∈ M and
2. For the second assertion we have we have shown M is an algebra.

31.3 *The σ – Finite Extension Theorem 345
Now suppose that N = ∞. Because M is an algebra, to show M is a σ – Exercise 31.1. Keeping the same hypothesis and notation as in Theorem 31.14
algebra it suffices to show B = ∪∞ i=1 Bi ∈ M under the additional assumption and suppose B ∈ M. Show there exists A ⊂ B ⊂ C such that A ∈ Aδσ ,
∞
that the collection of sets, {Bi }i=1 , are also pairwise disjoint in which case the C ∈ Aσδ and µ̄(C \ A) = 0. (Hint: see the proof of Theorem 28.6 where the
∞
sets, {Ai }i=1 , are pairwise disjoint. Since µ is additive on Aδ (Lemma 31.13), same statement is proved with M replaced by σ (A) .) Conclude from this that
for any n ∈ N, µ̄ is the completion of µ̄|σ(A) . (See Lemma 19.47 for more about completion of
n n
measures.)
X X
µ (Ai ) + ε2−i ≤ µ (∪ni=1 Ai ) + ε.

µ (Ci ) ≤ Exercise 31.2. Keeping the same hypothesis and notation as in Theorem
i=1 i=1 31.14, show M = M0 where M0 consists of those subset B ⊂ X such that
This implies, using µ∗ (E) = µ∗ (B ∩ E) + µ∗ (B c ∩ E) ∀ E ⊂ X. (31.15)
c
µ (∪ni=1 Ai ) = µ (X) − µ ([∪ni=1 Ai ] ) ≤ µ (X) , Hint: To verify Eq. (31.15) holds for B ∈ M, “approximate” E ⊂ X from the
outside by a set C ∈ Aσ and then make use the sub-additivity, the monotonicity
that
∞
X n
X of µ∗ and the fact that µ∗ is a measure on M.
µ (Ci ) = lim µ (Ci ) ≤ µ (X) + ε < ∞. (31.14)
i=1
n→∞
i=1 Theorem 31.15. Suppose that µ is a σ – finite premeasure on an algebra A.
`n Then
Let n ∈ N and An := i=1 Ai ∈ Aδ . Then Aδ 3 An ⊂ B ⊂ C ∈ Aσ , µ̄ (B) := inf {µ (C) : B ⊂ C ∈ Aσ } ∀ B ∈ σ (A) (31.16)
C \ An ∈ Aσ and
defined a measure on σ (A) and this measure is the unique measure on σ (A)
C \ An = ∪∞ n n
∞
i=1 (Ci \ A ) ⊂ [∪i=1 (Ci \ Ai )] ∪ ∪i=n+1 Ci ∈ Aσ . which extends µ.
Therefore, using the sub-additivity of µ on Aσ and the estimate (31.14), Proof. The uniqueness of the extension µ̄ was already proved in Theorem
∞
19.55. For existence, let {Xn }n=1 ⊂ A be chosen so that µ (Xn ) < ∞ for all n
n ∞
X X and Xn ↑ X as n → ∞ and let
µ (C \ An ) ≤ µ (Ci \ Ai ) + µ (Ci )
i=1 i=n+1 µn (A) := µn (A ∩ Xn ) for all A ∈ A.
∞
X
≤ε+ µ (Ci ) → ε as n → ∞. Each µn is a premeasure (as is easily verified) on A and hence by Theorem
i=n+1 31.14 each µn has an extension, µ̄n , to a measure on σ (A) . Since the measure
µ̄n are increasing, µ̄ := limn→∞ µ̄n is a measure which extends µ, see Exercise
Since ε > 0 was arbitrary it now follows from Lemma 31.9 that B ∈ M. 19.4.
Moreover, since The proof will be completed by verifying that Eq. (31.16) holds by repeating
µ∗ (Bi ) ≤ µ (Ci ) ≤ µ (Ai ) + 2−i ε, an argument already used in the proof of Theorem 28.6. Let B ∈ σ (A) , Bm =
n
X n Xm ∩ B and ε > 0 be given. By Theorem 31.14, there exists Cm ∈ Aσ such
X
µ∗ (Bi ) − 2−i ε ≤ µ (Ai ) = µ (An ) ≤ µ∗ (B) . that Bm ⊂ Cm ⊂ Xm and µ̄(Cm \ Bm ) = µ̄m (Cm \ Bm ) < ε2−n . Then C :=
i=1 i=1 ∪∞m=1 Cm ∈ Aσ and, as usual,
Letting n → ∞ in this equation implies
∞ ∞ ∞
!
[ X X
∞
X ∞
X µ̄(C \ B) ≤ µ̄ (Cm \ B) ≤ µ̄(Cm \ B) ≤ µ̄(Cm \ Bm ) < ε.
µ∗ (Bi ) − ε ≤ µ∗ (B) ≤ µ∗ (Bi ) . m=1 m=1 m=1
i=1 i=1
Thus
P∞
Because ε > 0 was arbitrary, it follows that ∗ ∗
i=1 µ (Bi ) = µ (B) and we have
µ̄ (B) ≤ µ̄ (C) = µ̄ (B) + µ̄(C \ B) ≤ µ̄ (B) + ε
also shown µ̄ = µ∗ |M is a measure on M. which proves the first item since ε > 0 was arbitrary.

31.4 General Extension and Construction Theorem µ∗ (A ∪ B) = µ∗ (A) + µ∗ (B)
Exercise 31.2 motivates the following definition. and µ = µ∗ |M is finitely additive on M.

We now must show that M is a σ – algebra and the µ is σ – additive.
Sn Let
Definition 31.16. Let µ∗ : 2X → [0, ∞] be an outer measure. Define the µ∗ - Ai ∈ M (without loss of generality assume Ai ∩ Aj = ∅ if i 6= j) Bn = i=1 Ai ,
∞
measurable sets to be and B =
S
Aj , then for E ⊂ X we have
j=1
M(µ∗ ) := {B ⊂ X : µ∗ (E) ≥ µ∗ (E ∩ B) + µ∗ (E ∩ B c ) ∀ E ⊂ X}.
µ∗ (E ∩ Bn ) = µ∗ (E ∩ Bn ∩ An ) + µ∗ (E ∩ Bn ∩ Acn )
∗ ∗
Because of the sub-additivity of µ , we may equivalently define M(µ ) by = µ∗ (E ∩ An ) + µ∗ (E ∩ Bn−1 ).
M(µ∗ ) = {B ⊂ X : µ∗ (E) = µ∗ (E ∩ B) + µ∗ (E ∩ B c ) ∀ E ⊂ X}. (31.17) and so by induction,
n
Theorem 31.17 (Carathéodory’s Construction Theorem). Let µ∗ be an X
µ∗ (E ∩ Bn ) = µ∗ (E ∩ Ak ). (31.21)
outer measure on X and M := M(µ∗ ). Then M is a σ-algebra and µ := µ∗ |M
k=1
is a complete measure.
Therefore we find that
Proof. Clearly ∅, X ∈ M and if A ∈ M then Ac ∈ M. So to show that M is
an algebra we must show that M is closed under finite unions, i.e. if A, B ∈ M µ∗ (E) = µ∗ (E ∩ Bn ) + µ∗ (E ∩ Bnc )
and E ∈ 2X then Xn
= µ∗ (E ∩ Ak ) + µ∗ (E ∩ Bnc )
µ∗ (E) ≥ µ∗ (E ∩ (A ∪ B)) + µ(E \ (A ∪ B)). k=1
Xn
Using the definition of M three times, we have ≥ µ∗ (E ∩ Ak ) + µ∗ (E ∩ B c )
k=1
µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E \ A) (31.18)
where the last inequality is a consequence of the monotonicity of µ∗ and the
= µ∗ (E ∩ A ∩ B) + µ∗ ((E ∩ A) \ B) fact that B c ⊂ Bnc . Letting n → ∞ in this equation shows that
+ µ∗ ((E \ A) ∩ B) + µ∗ ((E \ A) \ B). (31.19) ∞
X
µ∗ (E) ≥ µ∗ (E ∩ Ak ) + µ∗ (E ∩ B c )
By the sub-additivity of µ∗ and the set identity,
k=1
E ∩ (A ∪ B) = (E ∩ A) ∪ (E ∩ B) ≥ µ (∪k (E ∩ Ak )) + µ∗ (E \ B)
∗
= [((E ∩ A) \ B) ∪ (E ∩ A ∩ B)] ∪ [((E ∩ B) \ A) ∪ (E ∩ A ∩ B)] = µ∗ (E ∩ B) + µ∗ (E \ B) ≥ µ∗ (E),

= [E ∩ A ∩ B] ∪ [(E ∩ A) \ B] ∪ [(E \ A) ∩ B] , wherein we have used the sub-additivity µ∗ twice. Hence B ∈ M and we have
shown M is a σ – algebra. Since µ∗ (E) ≥ µ∗ (E ∩ Bn ) we may let n → ∞ in
we have Eq. (31.21) to find
∞
µ∗ (E ∩ A ∩ B) + µ∗ ((E ∩ A) \ B) + µ∗ ((E \ A) ∩ B) ≥ µ∗ (E ∩ (A ∪ B)) .
X
µ∗ (E) ≥ µ∗ (E ∩ Ak ).
k=1
Using this inequality in Eq. (31.19) shows ∞
Letting E = B = ∪Ak in this inequality then implies µ∗ (B) ≥ µ∗ (Ak ) and
P
∗ ∗ ∗
µ (E) ≥ µ (E ∩ (A ∪ B)) + µ (E \ (A ∪ B)) (31.20) k=1
∞
hence, by the sub-additivity of µ∗ , µ∗ (B) = µ∗ (Ak ). Therefore, µ = µ∗ |M
P
which implies A ∪ B ∈ M. So M is an algebra. Now suppose A, B ∈ M are k=1
disjoint, then taking E = A ∪ B in Eq. (31.18) implies is countably additive on M.

31.4 General Extension and Construction Theorem 347
Finally we show µ is complete. If N ⊂ F ∈ M and µ(F ) = 0 = µ∗ (F ), then 2. This is a`direct consequence of item 1. and Theorem 31.17.
∗ ∞ ∞
µ (N ) = 0 and 3. If A := j=1 Aj with {Aj }j=1 ⊂ A being a collection of pairwise disjoint
sets, then
µ∗ (E) ≤ µ∗ (E ∩ N ) + µ∗ (E ∩ N c ) = µ∗ (E ∩ N c ) ≤ µ∗ (E). ∞
X X∞
ν (A) = ν(Aj ) = µ(Aj ) = µ (A) .
which shows that N ∈ M. j=1 j=1
This shows ν = µ = µ̄ on Aσ . Consequently, if B ∈ M, then

31.4.1 Extensions of General Premeasures
ν(B) ≤ inf {ν (A) : B ⊂ A ∈ Aσ }
In this subsection let X be a set, A be a subalgebra of 2X and µ0 : A → [0, ∞] = inf {µ (A) : B ⊂ A ∈ Aσ } = µ∗ (B) = µ̄(B). (31.22)
be a premeasure on A.
If µ̄(B) < ∞ and ε > 0 is given, there exists A ∈ Aσ such that B ⊂ A and
Theorem 31.18. Let A ⊂ 2X be an algebra, µ be a premeasure on A and
µ̄ (A) = µ (A) ≤ µ̄ (B) + ε. From Eq. (31.22), this implies
µ∗ be the associated outer measure as defined in Eq. (31.7) with ρ = µ. Let
M := M(µ∗ ) ⊃ σ(A), then: ν (A \ B) ≤ µ̄(A \ B) ≤ ε.
∗ ∗
1. A ⊂ M(µ ) and µ |A = µ.
Therefore,
2. µ̄ = µ∗ |M is a measure on M which extends µ.
3. If ν : M → [0, ∞] is another measure such that ν = µ on A and B ∈ M, ν (B) ≤ µ̄ (B) ≤ µ̄ (A) = ν (A) = ν (B) + ν (A \ B) ≤ ν (B) + ε
then ν(B) ≤ µ̄(B) and ν(B) = µ̄(B) whenever µ̄(B) < ∞.
4. If µ is σ-finite on A then the extension, µ̄, of µ to M is unique and moreover which shows µ̄(B) = ν(B) because ε > 0 was arbitrary.
µ̄|σ(A)
M = σ (A) . 4. For the σ – finite case, choose Xj ∈ M such that Xj ↑ X and µ̄(Xj ) < ∞
then
Proof. Recall from Proposition 31.6 and Lemma 31.9 that µ extends to a µ̄(B) = lim µ̄(B ∩ Xj ) = lim ν(B ∩ Xj ) = ν(B).
countably additive function on Aσ and µ∗ = µ on A. j→∞ j→∞
1. Let A ∈ A and E ⊂ X such that µ∗ (E) < ` ∞. Given ε > 0 choose pairwise
∞
disjoint sets, Bj ∈ A, such that E ⊂ B := j=1 Bj and Theorem 31.19 (Regularity Theorem). Suppose that µ is a σ – finite pre-
∞
measure on an algebra A, µ̄ is the extension described in Theorem 31.18 and
µ∗ (E) + ε ≥ µ (B) =
X
µ(Bj ). B ∈ M := M (µ∗ ) . Then:
j=1 1.
`∞ c c
`∞ c µ̄ (B) := inf {µ̄ (C) : B ⊂ C ∈ Aσ } .
Since A ∩ E ⊂ j=1 (Bj ∩ A ) and E ∩ A ⊂ j=1 (Bj ∩ A ), using the
sub-additivity of µ∗ and the additivity of µ on A we have, 2. For any ε > 0 there exists A ⊂ B ⊂ C such that A ∈ Aδ , C ∈ Aσ and
∞ ∞
µ̄(C \ A) < ε.
3. There exists A ⊂ B ⊂ C such that A ∈ Aδσ , C ∈ Aσδ and µ̄(C \ A) = 0.
X X
µ∗ (E) + ε ≥ µ(Bj ) = [µ(Bj ∩ A) + µ(Bj ∩ Ac )]
j=1 j=1
4. The σ-algebra, M, is the completion of σ (A) with respect to µ̄|σ(A) .
≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). Proof. The proofs of items 1. – 3. are the same as the proofs of the corre-
sponding results in Theorem 28.6 and so will be omitted. Moreover, item 4. is
Since ε > 0 is arbitrary this shows that a simple consequence of item 3. and Proposition 19.6.
µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) The following proposition shows that measures may be “restricted” to non-
measurable sets.
and therefore that A ∈ M(µ∗ ).

Proposition 31.20. Suppose that (X, M, µ) is a probability space and Ω ⊂ X 31.5 Proof of the Riesz-Markov Theorem 28.16
is any set. Let MΩ := {A ∩ Ω : A ∈ M} and set P (A ∩ Ω) := µ∗ (A ∩ Ω). Then
P is a measure on the σ - algebra MΩ . Moreover, if P ∗ is the outer measure This section is devoted to completing the proof of the Riesz-Markov Theorem
generated by P, then P ∗ (A) = µ∗ (A) for all A ⊂ Ω. 28.16.
Proof. Let A, B ∈ M such that A ∩ B = ∅. Then since A ∈ M ⊂ M(µ∗ ) it Theorem 31.21. Suppose (X, τ ) is a locally compact Hausdorff space, I is a
follows from Eq. (31.15) with E := (A ∪ B) ∩ Ω that positive linear functional on Cc (X) and µ := µI be as in Notation 28.15. Then
µ∗ ((A ∪ B) ∩ Ω) = µ∗ ((A ∪ B) ∩ Ω ∩ A) + µ∗ ((A ∪ B) ∩ Ω ∩ Ac ) µ is a Radon measure on X such that I = Iµ , i.e.
= µ∗ (Ω ∩ A) + µ∗ (B ∩ Ω) Z
`∞ I (f ) = f dµ for all f ∈ Cc (X) .
which shows that P is finitely additive. Now suppose A = j=1 Aj with Aj ∈ M X
`∞
and let Bn := j=n+1 Aj ∈ M. By what we have just proved,
Proof. Let µ : τ → [0, ∞] be as in Eq. (28.8) and µ∗ : 2X → [0, ∞] be
n
X n
X the associate outer measure as in Proposition 31.8. As we have seen in Lemma
µ∗ (A ∩ Ω) = µ∗ (Aj ∩ Ω) + µ∗ (Bn ∩ Ω) ≥ µ∗ (Aj ∩ Ω). 31.10, µ is sub-additive on τ and
j=1 j=1
µ∗ (E) = inf {µ (U ) : E ⊂ U ⊂o X} .
Passing to the limit as n → ∞ in this last expression and using the sub-
additivity of µ∗ we find By Theorem 31.17, M := M (µ∗ ) is a σ-algebra and µ∗ |M is a measure on M.
∞
X ∞
X To show BX ⊂ M it suffices to show U ∈ M for all U ∈ τ, i.e. we must
µ∗ (Aj ∩ Ω) ≥ µ∗ (A ∩ Ω) ≥ µ∗ (Aj ∩ Ω). show;
j=1 j=1 µ∗ (E) ≥ µ∗ (E ∩ U ) + µ∗ (E \ U ) (31.24)
Thus for every E ⊂ X such that µ∗ (E) < ∞. First suppose E is open, in which case
∞
X
µ∗ (A ∩ Ω) = µ∗ (Aj ∩ Ω) E ∩ U is open as well. Let f ≺ E ∩ U and K := supp(f ). Then E \ U ⊂ E \ K
j=1 and if g ≺ E \ K ∈ τ then f + g ≺ E (see Figure 31.1) and hence
and we have shown that P = µ |MΩ is a measure. Now let P ∗ be the outer
∗
µ∗ (E) ≥ I (f + g) = I (f ) + I (g) .
measure generated by P. For A ⊂ Ω, we have
P ∗ (A) = inf {P (B) : A ⊂ B ∈ MΩ } Taking the supremum of this inequality over g ≺ E \ K shows
= inf {P (B ∩ Ω) : A ⊂ B ∈ M} µ∗ (E) ≥ I (f ) + µ∗ (E \ K) ≥ I (f ) + µ∗ (E \ U ) .
= inf {µ∗ (B ∩ Ω) : A ⊂ B ∈ M} (31.23)
Taking the supremum of this inequality over f ≺ U shows Eq. (31.24) is valid
∗ ∗
and since µ (B ∩ Ω) ≤ µ (B), for E ∈ τ.
P ∗ (A) ≤ inf {µ∗ (B) : A ⊂ B ∈ M} For general E ⊂ X, let V ∈ τ with E ⊂ V, then
= inf {µ(B) : A ⊂ B ∈ M} = µ∗ (A). µ∗ (V ) ≥ µ∗ (V ∩ U ) + µ∗ (V \ U ) ≥ µ∗ (E ∩ U ) + µ∗ (E \ U )
On the other hand, for A ⊂ B ∈ M, we have µ∗ (A) ≤ µ∗ (B ∩ Ω) and therefore
and taking the infimum of this inequality over such V shows Eq. (31.24) is valid
by Eq. (31.23)
for general E ⊂ X. Thus U ∈ M for all U ∈ τ and therefore BX ⊂ M.
µ∗ (A) ≤ inf {µ∗ (B ∩ Ω) : A ⊂ B ∈ M} = P ∗ (A). Up to this point it has been shown that µ = µ∗ |BX is a measure which, by
very construction, is outer regular. We now verify that µ satisfies Eq. (28.10),
and we have shown
namely that µ (K) = ν (K) for all compact sets K ⊂ X where
µ∗ (A) ≤ P ∗ (A) ≤ µ∗ (A).
ν (K) := inf {I (f ) : f ∈ Cc (X, [0, 1]) 3 f ≥ 1K } .

31.5 Proof of the Riesz-Markov Theorem 28.16 349
Fig. 31.1. Constructing a function g which approximates 1E\U .
To do this let f ∈ Cc (X, [0, 1]) with f ≥ 1K and ε > 0 be given. Let
−1 Fig. 31.2. This sequence of figures shows how the function fn is constructed. The
Uε := {f > 1 − ε} ∈ τ and g ≺ Uε , then g ≤ (1 − ε) f and hence
−1 idea is to think of f as describing a “cake” set on a “table,” X. We then slice the cake
I (g) ≤ (1 − ε) I (f ) . Taking the supremum of this inequality over all g ≺ Uε
into slabs, each of which is placed back on the table. Each of these slabs is described
then gives,
−1 by one of the functions, fn , as in Eq. (31.27).
µ (K) ≤ µ (Uε ) ≤ (1 − ε) I (f ) .
Since ε > 0 was arbitrary, we learn µ (K) ≤ I (f ) for all 1K ≤ f ≺ X and
therefore, µ (K) ≤ ν (K) . Now suppose that U ∈ τ and K ⊂ U. By Urysohn’s P∞
see Figure 31.2. It should be clear from Figure 31.2 that f = n=1 fn with
Lemma 15.8 (also see Lemma 14.27), there exists f ≺ U such that f ≥ 1K and
therefore n
fn ≡ 0 for all n sufficiently large.
the sum actually being a finitesum since
Let K0 := supp(f ) and Kn := f ≥ N . Then (again see Figure 31.2) for all
µ (K) ≤ ν (K) ≤ I (f ) ≤ µ (U ) . n ∈ N,
By the outer regularity of µ, we have 1Kn ≤ N fn ≤ 1Kn−1
µ (K) ≤ ν (K) ≤ inf {µ (U ) : K ⊂ U ⊂o X} = µ (K) , which upon integrating on µ gives
i.e. µ (Kn ) ≤ N Iµ (fn ) ≤ µ (Kn−1 ) . (31.28)
µ (K) = ν (K) = inf {I (f ) : f ∈ Cc (X, [0, 1]) 3 f ≥ 1K } . (31.25) Moreover, if U is any open set containing Kn−1 , then N fn ≺ U and so by Eq.
(31.25) and the definition of µ, we have
This inequality clearly establishes that µ is K-finite and therefore
Cc (X, [0, ∞)) ⊂ L1 (µ) . µ (Kn ) ≤ N I (fn ) ≤ µ (U ) . (31.29)
Next we will establish,
Z From the outer regularity of µ, it follows from Eq. (31.29) that
I (f ) = Iµ (f ) := f dµ (31.26)
X µ (Kn ) ≤ N I (fn ) ≤ µ (Kn−1 ) . (31.30)
for all f ∈ Cc (X) . By the linearity, it suffices to verify Eq. (31.26) holds for
As a consequence of Eqs. (31.28) and (31.30), we have
f ∈ Cc (X, [0, ∞)) . To do this we will use the “layer cake method” to slice f
into thin pieces. Explicitly, fix an N ∈ N and for n ∈ N let N |Iµ (fn ) − I (fn )| ≤ µ (Kn−1 ) − µ (Kn ) = µ (Kn−1 \ Kn ) .

n−1 1
fn := min max f − ,0 , , (31.27) Therefore
N N

∞ ∞
X X
X X µ∗ (A) ≤ ei ) = µ(B) ≤
µ(E µ(Ei ).
|Iµ (f ) − I (f )| = Iµ (fn ) − I (fn ) ≤ |Iµ (fn ) − I (fn )|

Therefore, µ∗ (A) = inf{µ(B) : B ∈ M and A ⊂ B}.

n=1 n=1
∞
1 X 1 Item 2. If µ∗ (E) = ∞ Eq. (31.31) holds trivially. So assume that µ∗ (E) < ∞.
≤ µ (Kn−1 \ Kn ) = µ (K0 ) → 0 as N → ∞ Let ε > 0 be given and choose, by Item 1., B ∈ M such that E ⊂ B and
N N
n=1
µ(B) ≤ µ∗ (E) + ε. Then
which establishes Eq. (31.26).
µ∗ (E) + ε ≥ µ(B) = µ(B ∩ A) + µ(B ∩ Ac )
It now only remains to show µ is inner regular on open sets to complete the
proof. If U ∈ τ and µ (U ) < ∞, then for any ε > 0 there exists f ≺ U such that ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ).
Z Since ε > 0 is arbitrary we are done.
µ (U ) ≤ I (f ) + ε = f dµ + ε ≤ µ (supp(f )) + ε. Item 3. Let us begin by assuming the µ(X) < ∞. We have already seen that
X
M ⊂ M(µ∗ ). Suppose that A ∈ 2X satisfies,
Hence if K = supp(f ), we have K ⊂ U and µ (U \ K) < ε and this shows µ is
inner regular on open sets with finite measure. Finally if U ∈ τ and µ (U ) = ∞, µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ∀ E ∈ 2X . (31.32)
there exists fn ≺ U such that I (fn ) ↑ ∞ as n → ∞. Then, letting Kn =
By Item 1., there exists Bn ∈ M such that A ⊂ Bn and µ∗ (Bn ) ≤ µ∗ (A) + n1
supp(fn ), we have Kn ⊂ U and µ (Kn ) ≥ I (fn ) and therefore µ (Kn ) ↑ µ (U ) =
for all n ∈ N. Therefore B = ∩Bn ⊃ A and µ(B) ≤ µ∗ (A) + n1 for all n which
∞.
implies that µ(B) ≤ µ∗ (A) which implies that µ(B) = µ∗ (A). Similarly there
exists C ∈ M such that Ac ⊂ C and µ∗ (Ac ) = µ(C). Taking E = X in Eq.
(31.32) shows
31.6 More Motivation of Carathéodory’s Construction µ(X) = µ∗ (A) + µ∗ (Ac ) = µ(B) + µ(C)
Theorem 31.17
so
The next Proposition helps to motivate this definition and the Carathéodory’s µ(C c ) = µ(X) − µ (C) = µ(B).
construction Theorem 31.17. Thus letting D = C c , we have
Proposition 31.22. Suppose E = M is a σ – algebra, ρ = µ : M → [0, ∞] is D ⊂ A ⊂ B and µ(D) = µ∗ (A) = µ(B)

a measure and µ∗ is defined as in Eq. (31.7). Then
so µ(B \ D) = 0 and hence
1. For A ⊂ X
µ∗ (A) = inf{µ(B) : B ∈ M and A ⊂ B}. A = D ∪ [(B\D) ∩ A]
∗
In particular, µ = µ on M. where D ∈ M and (B\D) ∩ A ∈ N showing that A ∈ M̄ and µ∗ (A) = µ̄(A).
2. Then M ⊂ M(µ∗ ), i.e. if A ∈ M and E ⊂ X then Now if µ is σ – finite, choose Xn ∈ M such that µ(Xn ) < ∞ and Xn ↑ X.
Given A ∈ M(µ∗ ) set An = Xn ∩ A. Therefore
µ∗ (E) ≥ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ). (31.31)
µ∗ (E) = µ∗ (E ∩ A) + µ∗ (E ∩ Ac ) ∀ E ∈ 2X .
3. Assume further that µ is σ – finite on M, then M(µ∗ ) = M̄ = M̄µ and
µ∗ |M(µ∗ ) = µ̄ where (M̄ = M̄µ , µ̄) is the completion of (M, µ) . Replace E by Xn to learn,
Proof. Item 1. If Ei ∈ M such that A ⊂ ∪Ei = B and E

ei = Ei \ (E1 ∪ · · · ∪ µ∗ (Xn ) = µ∗ (An ) + µ∗ (Xn \ A) = µ∗ (An ) + µ∗ (Xn \ An ).
Ei−1 ) then X X The same argument as above produces sets Dn ⊂ An ⊂ Bn such that µ(Dn ) =
µ(Ei ) ≥ µ(Eei ) = µ(B)
µ∗ (An ) = µ(Bn ). Hence An = Dn ∪ Nn and Nn := (Bn \ Dn ) ∩ An ∈ N . So we
so learn that

A = D ∪ N := (∪Dn ) ∪ (∪Nn ) ∈ M ∪ N = M̄.
We also see that µ∗ (A) = µ(D) since D ⊂ A ⊂ D ∪ F where F ∈ M such that
N ⊂ F and
µ(D) = µ∗ (D) ≤ µ∗ (A) ≤ µ(D ∪ F ) = µ(D).
32
The Daniell – Stone Construction of Integration and Measures
Now that we have developed integration theory relative to a measure on a σ Definition 32.3. A set, L, of extended real valued functions on X is an ex-
– algebra, it is time to show how to construct the measures that we have been tended vector space (or a vector space for short) if L is closed under scalar
using. This is a bit technical because there tends to be no “explicit” description multiplication and addition in the following sense: if f, g ∈ L and λ ∈ R then
of the general element of the typical σ – algebras. On the other hand, we do (f + λg) ⊂ L. A vector space L is said to be an extended lattice (or a lattice
know how to explicitly describe algebras which are generated by some class for short) if it is also closed under the lattice operations;
of sets E ⊂ 2X . Therefore, we might try to define measures on σ(E) by there
f ∨ g = max(f, g) and f ∧ g = min(f, g).
restrictions to A(E). Theorem 19.55 or Theorem 33.6 shows this is a plausible
method. A linear functional I on L is a function I : L → R such that
So the strategy of this section is as follows: 1) construct finitely additive
measure on an algebra, 2) construct “integrals” associated to such finitely ad- I(f + λg) = I(f ) + λI(g) for all f, g ∈ L and λ ∈ R. (32.2)
ditive measures, 3) extend these integrals (Daniell’s method) when possible to A linear functional I is positive if I(f ) ≥ 0 when f ∈ L+ .
a larger class of functions, 4) construct a measure from the extended integral
Equation (32.2) is to be interpreted as I(h) = I(f ) + λI(g) for all h ∈
(Daniell – Stone construction theorem).
(f + λg), and in particular I is required to take the same value on all members
In this chapter, X will be a given set and we will be dealing with certain
of (f + λg).
spaces of extended real valued functions f : X → R̄ on X.
Remark 32.4. Notice that an extended lattice L is closed under the absolute
Notation 32.1 Given functions f, g : X → R̄, let f + g denote the collection of value operation since |f | = f ∨ 0 − f ∧ 0 = f ∨ (−f ). Also if I is positive on L
functions h : X → R̄ such that h(x) = f (x)+g(x) for all x for which f (x)+g(x) then I(f ) ≤ I(g) when f, g ∈ L and f ≤ g. Indeed, f ≤ g implies (g − f )0 ≥ 0,
is well defined, i.e. not of the form ∞ − ∞. so
For example, if X = {1, 2, 3} and f (1) = ∞, f (2) = 2 and f (3) = 5 0 = I(0) ≤ I((g − f )0 ) = I(g) − I(f )
and g(1) = g(2) = −∞ and g(3) = 4, then h ∈ f + g iff h(2) = −∞ and and hence I(f ) ≤ I(g). If L is a vector space of real-valued functions on X,
h(3) = 7. The value h(1) may be chosen freely. More generally if a, b ∈ R and then L is a lattice iff f + = f ∨ 0 ∈ L for all f ∈ L. This is because
f, g : X → R̄ we will write af + bg for the collection of functions h : X → R̄
|f | = f + + (−f )+ ,
such that h(x) = af (x) + bg(x) for those x ∈ X where af (x) + bg(x) is well
defined with the values of h(x) at the remaining points being arbitrary. It will 1
f ∨ g = (f + g + |f − g|) and
also be useful to have some explicit representatives for af + bg which we define, 2
for α ∈ R̄, by 1
f ∧ g = (f + g − |f − g|) .
2
af (x) + bg(x) when defined In the remainder of this chapter we fix a sub-lattice, S ⊂ `∞ (X, R) and a
(af + bg)α (x) = (32.1)
α otherwise. positive linear functional I : S → R.
We will make use of this definition with α = 0 and α = ∞ below. Definition 32.5 (Property (D)). A non-negative linear functional I on S is
∞
said to be continuous under monotone limits if I(fn ) ↓ 0 for all {fn }n=1 ⊂
Notation 32.2 Given a collection of extended real valued functions C on X, +
S satisfying (pointwise) fn ↓ 0. A positive linear functional on S satisfying
let C + := {f ∈ C : f ≥ 0} – denote the subset of positive functions f ∈ C. property (D) is called a Daniell integral on S. We will also write S as D(I)
– the domain of I.
354 32 The Daniell – Stone Construction of Integration and Measures
Lemma 32.6. Let I be a non-negative linear functional on a lattice S. Then 0 ≤ I(fn ) ≤ CK kfn k∞ ↓ 0 as n → ∞.
property (D) is equivalent to either of the following two properties:
D1 If φ, φn ∈ S satisfy; φn ≤ φn+1 for all n and φ ≤ limn→∞ φn , then I(φ) ≤ For example if X = R and F is an increasing function on R, then I (f ) :=
limn→∞ I(φn ).
R
P∞ P∞ R
f dF is a Daniell integral on Cc (R, R), see Lemma 28.38. However it is not
D2 If uj ∈ S+ and φ ∈ S is such that φ ≤ j=1 uj then I(φ) ≤ j=1 I(uj ). generally true in this case that I(fn ) ↓ 0 for all fn ∈ S (S is the collection of
compactly supported step functions on R) such that fn ↓ 0. The next example
Proof. (D) =⇒ (D1 ) Let φ, φn ∈ S be as in D1 . Then φ ∧ φn ↑ φ and
and proposition addresses this question.
φ − (φ ∧ φn ) ↓ 0 which implies
Example 32.8. Suppose F : R → R is an increasing function which is not right
I(φ) − I(φ ∧ φn ) = I(φ − (φ ∧ φn )) ↓ 0.
continuous at x0 ∈ R. Then, letting fn = 1(x0 ,x0 +n−1 ] ∈ S, we have fn ↓ 0 as
Hence n → ∞ but
I(φ) = lim I(φ ∧ φn ) ≤ lim I(φn ). Z
fn dF = F x0 + n−1 − F (x0 ) → F (x0 +) − F (x0 ) 6= 0.

n→∞ n→∞
Pn
(D1 ) =⇒ (D2 ) Apply (D1 ) with φn = j=1 uj . (D2 ) =⇒ (D) Suppose φn ∈ S R
PN
with φn ↓ 0 and let un = φn − φn+1 . Then n=1 un = φ1 − φN +1 ↑ φ1 and Proposition 32.9. Let (A, µ, S = Sf (A, µ), I = Iµ ) be as in Definition 28.36.
hence If µ is a premeasure (Definition 31.1) on A, then
∞ N
X X ∀ fn ∈ S with fn ↓ 0 =⇒ I(fn ) ↓ 0 as n → ∞. (32.3)
I(φ1 ) ≤ I(un ) = lim I(un )
N →∞
n=1 n=1
Hence I is a Daniell integral on S.
= lim I(φ1 − φN +1 ) = I(φ1 ) − lim I(φN +1 )
N →∞ N →∞
Proof. Let ε > 0 be given. Then
from which it follows that limN →∞ I(φN +1 ) ≤ 0. Since I(φN +1 ) ≥ 0 for all N
we conclude that limN →∞ I(φN +1 ) = 0. fn = fn 1fn >εf1 + fn 1fn ≤εf1 ≤ f1 1fn >εf1 + εf1 ,
X
I(fn ) ≤ I (f1 1fn >εf1 ) + εI(f1 ) = aµ (f1 = a, fn > εa) + εI(f1 ),
32.0.1 Examples of Daniell Integrals a>0
Proposition 32.7. Suppose that (X, τ ) is locally compact Hausdorff space and and hence
I is a positive linear functional on S := Cc (X, R). Then for each compact subset X
K ⊂ X there is a constant CK < ∞ such that |I(f )| ≤ CK kf k∞ for all lim sup I(fn ) ≤ a lim sup µ (f1 = a, fn > εa) + εI(f1 ). (32.4)
n→∞ n→∞
a>0
f ∈ Cc (X, R) with supp(f ) ⊂ K. Moreover, if fn ∈ Cc (X, [0, ∞)) and fn ↓ 0
(pointwise) as n → ∞, then I(fn ) ↓ 0 as n → ∞ and in particular I is Because, for a > 0,
necessarily a Daniell integral on S.
A 3 {f1 = a, fn > εa} ↓ ∅ as n → ∞
Proof. Let f ∈ Cc (X, R) with supp(f ) ⊂ K. By Lemma 15.8 there exists
ψK ≺ X such that ψK = 1 on K. Since kf k∞ ψK ± f ≥ 0, and µ (f1 = a) < ∞, lim supn→∞ µ (f1 = a, fn > εa) = 0. Combining this
with Eq. (32.4) and making use of the fact that ε > 0 is arbitrary we learn
0 ≤ I(kf k∞ ψK ± f ) = kf k∞ I(ψK ) ± I(f )
lim supn→∞ I(fn ) = 0.
from which it follows that |I(f )| ≤ I(ψK ) kf k∞ . So the first assertion holds
with CK = I(ψK ) < ∞. Now suppose that fn ∈ Cc (X, [0, ∞)) and fn ↓ 0 as
n → ∞. Let K = supp(f1 ) and notice that supp(fn ) ⊂ K for all n. By Dini’s
Theorem (see Exercise 14.3), kfn k∞ ↓ 0 as n → ∞ and hence

32.1 Extending a Daniell Integral 355
32.1 Extending a Daniell Integral I↓ (g) = lim I (gn ) if S 3 gn ↓ g.

n→∞
In the remainder of this chapter we fix a lattice, S, of bounded functions, f : If f ∈ S↑ ∩ S↓ , then there exists fn , gn ∈ S such that fn ↑ f and gn ↓ f. Hence
X → R, and a positive linear functional I : S → R satisfying Property (D) of S 3 (gn − fn ) ↓ 0 and hence by the continuity property (D),
Definition 32.5.
I↓ (f ) − I↑ (f ) = lim [I (gn ) − I (fn )] = lim I (gn − fn ) = 0.
n→∞ n→∞
Lemma 32.10. Suppose that {fn } , {gn } ⊂ S.
Therefore I↓ = I↑ on S↑ ∩ S↓ .
1. If fn ↑ f and gn ↑ g with f, g : X → (−∞, ∞] such that f ≤ g, then
Notation 32.12 Using the above comments we may now simply write I (f ) for
lim I(fn ) ≤ lim I(gn ). (32.5) I↑ (f ) or I↓ (f ) when f ∈ S↑ or f ∈ S↓ . Henceforth we will now view I as a
n→∞ n→∞
function on S↑ ∩ S↑ .
2. If fn ↓ f and gn ↓ g with f, g : X → [−∞, ∞) such that f ≤ g, then Eq.
Again because of Lemma 32.10, let I↑ := I|S↑ or I↓ := I|S↓ are positive
(32.5) still holds.
functionals; i.e. if f ≤ g then I(f ) ≤ I(g).
In particular, in either case if f = g, then
Exercise 32.1. Show S↓ = −S↑ and for f ∈ S↓ ∪ S↑ that I(−f ) = −I(f ) ∈ R̄.
lim I(fn ) = lim I(gn ).
n→∞ n→∞ Proposition 32.13. The set S↑ and the extension of I to S↑ in Definition 32.11
satisfies:
Proof.
1. (Monotonicity) I(f ) ≤ I(g) if f, g ∈ S↑ with f ≤ g.
1. Fix n ∈ N, then gk ∧ fn ↑ fn as k → ∞ and gk ∧ fn ≤ gk and hence
2. S↑ is closed under the lattice operations, i.e. if f, g ∈ S↑ then f ∧ g ∈ S↑
I(fn ) = lim I(gk ∧ fn ) ≤ lim I(gk ). and f ∨ g ∈ S↑ . Moreover, if I(f ) < ∞ and I(g) < ∞, then I(f ∨ g) < ∞
k→∞ k→∞ and I(f ∧ g) < ∞.
Passing to the limit n → ∞ in this equation proves Eq. (32.5). 3. (Positive Linearity) I (f + λg) = I(f ) + λI(g) Pfor all f, g ∈ S↑ and λ ≥ 0.
∞
2. Since −fn ↑ (−f ) and −gn ↑ (−g) and −g ≤ (−f ), what we just proved 4. f ∈ S+ iff there exists φn ∈ S+
such that f = n=1 φn . Moreover, I(f ) =
P∞ ↑
shows m=1 I(φ m ). P∞ P∞
5. If fn ∈ S+
↑ , then
+
n=1 fn =: f ∈ S↑ and I(f ) = n=1 I(fn ).
− lim I(gn ) = lim I(−gn ) ≤ lim I(−fn ) = − lim I(fn )
n→∞ n→∞ n→∞ n→∞ Remark 32.14. Similar results hold for the extension of I to S↓ in Definition
which is equivalent to Eq. (32.5). 32.11.
Proof.
Definition 32.11. Let 1. Monotonicity follows directly from Lemma 32.10.

2. If fn , gn ∈ S are chosen so that fn ↑ f and gn ↑ g, then fn ∧ gn ↑ f ∧ g and
S↑ = {f : X → (−∞, ∞] : ∃ fn ∈ S such that fn ↑ f } fn ∨ gn ↑ f ∨ g. If we further assume that I(g) < ∞, then f ∧ g ≤ g and
hence I(f ∧ g) ≤ I(g) < ∞. In particular it follows that I(f ∧ 0) ∈ (−∞, 0]
and for all f ∈ S↑ . Combining this with the identity,
S↓ = {f : X → [−∞, ∞) : ∃ fn ∈ S such that fn ↓ f } .
I(f ) = I (f ∧ 0 + f ∨ 0) = I (f ∧ 0) + I(f ∨ 0) ,
Because of Lemma 32.10, for f ∈ S↑ and g ∈ S↓ we may define
shows I(f ) < ∞ iff I(f ∨ 0) < ∞. Since f ∨ g ≤ f ∨ 0 + g ∨ 0, if both
I↑ (f ) = lim I (fn ) if S 3 fn ↑ f I(f ) < ∞ and I(g) < ∞ then
n→∞
and I(f ∨ g) ≤ I (f ∨ 0) + I (g ∨ 0) < ∞.

3. Let fn , gn ∈ S be chosen so that fn ↑ f and gn ↑ g, then (fn + λgn ) ↑ Remark 32.17. A function g : X → R̄ is integrable iff for any ε > 0 there exists
(f + λg) and therefore f ∈ S↓ ∩ L1 (I) and h ∈ S↑ ∩ L1 (I)1 such that f ≤ g ≤ h and I(h − f ) < ε.
Indeed if g is integrable, then I∗ (g) = I ∗ (g) and there exists f ∈ S↓ ∩ L1 (I)
I (f + λg) = lim I (fn + λgn ) = lim I(fn ) + λ lim I(gn ) and h ∈ S↑ ∩ L1 (I) such that f ≤ g ≤ h and 0 ≤ I∗ (g) − I(f ) < ε/2 and
n→∞ n→∞ n→∞
0 ≤ I(h) − I ∗ (g) < ε/2. Adding these two inequalities implies 0 ≤ I(h) − I(f ) =
= I(f ) + λI(g).
I(h − f ) < ε. Conversely, if there exists f ∈ S↓ ∩ L1 (I) and h ∈ S↑ ∩ L1 (I) such
4. Let f ∈ S+ that f ≤ g ≤ h and I(h − f ) < ε, then
↑ and fn ∈ S be chosen so that fn ↑ f. By replacing fn by fn ∨ 0
if necessary we may assume that fn ∈ S+ . Now set φn = fn −Pfn−1 ∈ S for I(f ) = I∗ (f ) ≤ I∗ (g) ≤ I∗ (h) = I(h) and
∞
n = 1, 2, 3, . . . with the convention that f0 = 0 ∈ S. Then n=1 φn = f
and I(f ) = I ∗ (f ) ≤ I ∗ (g) ≤ I ∗ (h) = I(h)
n
X n
X ∞
X and therefore
I(f ) = lim I(fn ) = lim I( φm ) = lim I(φm ) = I(φm ).
n→∞ n→∞ n→∞
m=1 m=1 m=1 0 ≤ I ∗ (g) − I∗ (g) ≤ I(h) − I(f ) = I(h − f ) < ε.
P∞ Pn
Conversely, if f = m=1 φm with φm ∈ S+ , then fn := m=1 φm ↑ f as Since ε > 0 is arbitrary, this shows I ∗ (g) = I∗ (g).
n → ∞ and fn ∈ S+ .P
∞
5. Using Item 4., fn = m=1 φn,m with φn,m ∈ S+ . Thus Proposition 32.18. Given functions f, g : X → R̄, then:
∞ X
X ∞ X 1. I ∗ (λf ) = λI ∗ (f ) for all λ ≥ 0.
f= φn,m = lim φn,m ∈ S↑ 2. (Chebyshev’s Inequality.) Suppose f : X → [0, ∞] is a function and α ∈
N →∞
n=1 m=1 m,n≤N (0, ∞), then I ∗ (1{f ≥α} ) ≤ α1 I ∗ (f ) and if I ∗ (f ) < ∞ then I ∗ (1{f =∞} ) = 0.
3. I ∗ is sub-additive, i.e. if I ∗ (f )+I ∗ (g) is not of the form ∞−∞ or −∞+∞,
and then
X X I ∗ (f + g) ≤ I ∗ (f ) + I ∗ (g). (32.6)
I(f ) = lim I( φn,m ) = lim I(φn,m )
N →∞ N →∞
m,n≤N m,n≤N This inequality is to be interpreted to mean,
∞ X
∞ ∞
I ∗ (h) ≤ I ∗ (f ) + I ∗ (g) for all h ∈ (f + g).
X X
= I(φn,m ) = I(fn ).
n=1 m=1 n=1
4. I∗ (−g) = −I ∗ (g).
5. I∗ (g) ≤ I ∗ (g).
Definition 32.15. Given an arbitrary function g : X → R̄, let 6. If f ≤ g then I ∗ (f ) ≤ I ∗ (g) and I∗ (f ) ≤ I∗ (g).
7. If g ∈ S↑ and I(g) < ∞ or g ∈ S↓ and I(g) > −∞ then I∗ (g) = I ∗ (g) =
I ∗ (g) = inf {I(f ) : g ≤ f ∈ S↑ } ∈ R̄ and I(g).
I∗ (g) = sup {I(f ) : S↓ 3 f ≤ g} ∈ R̄. Proof.

with the convention that sup ∅ = −∞ and inf ∅ = +∞. 1. Suppose that λ > 0 (the λ = 0 case being trivial), then
I ∗ (λf ) = inf {I(h) : λf ≤ h ∈ S↑ } = inf I(h) : f ≤ λ−1 h ∈ S↑

Definition 32.16. A function g : X → R̄ is integrable if I∗ (g) = I ∗ (g) ∈ R.
Let = inf {I(λg) : f ≤ g ∈ S↑ } = λ inf {I(g) : f ≤ g ∈ S↑ } = λI ∗ (f ).
L1 (I) := g : X → R̄ : I∗ (g) = I ∗ (g) ∈ R

1
Equivalently, f ∈ S↓ with I(f ) > −∞ and h ∈ S↑ with I(h) < ∞.
¯
and for g ∈ L1 (I), let I(g) denote the common value I∗ (g) = I ∗ (g).

32.1 Extending a Daniell Integral 357
2. For α ∈ (0, ∞), α1{f ≥α} ≤ f and therefore, 6. If f ≤ g then
αI ∗ (1{f ≥α} ) = I ∗ (α1{f ≥α} ) ≤ I ∗ (f ). I ∗ (f ) = inf {I(h) : f ≤ h ∈ S↑ } ≤ inf {I(h) : g ≤ h ∈ S↑ } = I ∗ (g)
Since N 1{f =∞} ≤ f for all N ∈ (0, ∞), and

N I ∗ (1{f =∞} ) = I ∗ (N 1{f =∞} ) ≤ I ∗ (f ). I∗ (f ) = sup {I(h) : S↓ 3 h ≤ f } ≤ sup {I(h) : S↓ 3 h ≤ g} = I∗ (g).
So if I ∗ (f ) < ∞, this inequality implies I ∗ (1{f =∞} ) = 0 because N is 7. Let g ∈ S↑ with I(g) < ∞ and choose gn ∈ S such that gn ↑ g. Then
arbitrary.
3. If I ∗ (f ) + I ∗ (g) = ∞ the inequality is trivial so we may assume that I ∗ (g) ≥ I∗ (g) ≥ I(gn ) → I(g) as n → ∞.
I ∗ (f ), I ∗ (g) ∈ [−∞, ∞). If I ∗ (f ) + I ∗ (g) = −∞ then we may assume,
by interchanging f and g if necessary, that I ∗ (f ) = −∞ and I ∗ (g) < ∞. Combining this with
By definition of I ∗ , there exists fn ∈ S↑ and gn ∈ S↑ such that f ≤ fn and
g ≤ gn and I(fn ) ↓ −∞ and I(gn ) ↓ I ∗ (g). Since f + g ≤ fn + gn ∈ S↑ , (i.e. I ∗ (g) = inf {I(f ) : g ≤ f ∈ S↑ } = I(g)
h ≤ fn + gn for all h ∈ (f + g) which holds because fn , gn > −∞) and
shows
∗
I(fn + gn ) = I(fn ) + I(gn ) ↓ −∞ + I (g) = −∞, I ∗ (g) ≥ I∗ (g) ≥ I(g) = I ∗ (g)
and hence I∗ (g) = I(g) = I ∗ (g). If g ∈ S↓ and I(g) > −∞, then by what
it follows that I ∗ (f +g) = −∞, i.e. I ∗ (h) = −∞ for all h ∈ f +g. Henceforth
we have just proved,
we may assume I ∗ (f ), I ∗ (g) ∈ R. Let k ∈ (f + g) and f ≤ h1 ∈ S↑ and
g ≤ h2 ∈ S↑ . Then k ≤ h1 + h2 ∈ S↑ because if (for example) f (x) = ∞ I∗ (−g) = I(−g) = I ∗ (−g).
and g(x) = −∞, then h1 (x) = ∞ and h2 (x) > −∞ since h2 ∈ S↑ . Thus
h1 (x) + h2 (x) = ∞ ≥ k(x) no matter the value of k(x). It now follows This finishes the proof since I∗ (−g) = −I ∗ (g) and I(−g) = −I(g).
from the definitions that I ∗ (k) ≤ I(h1 ) + I(h2 ) for all f ≤ h1 ∈ S↑ and
g ≤ h2 ∈ S↑ . Therefore,
Lemma 32.19 (Countable Sub-additivity of I ∗ ). Let fn : X → [0, ∞] be
I ∗ (k) ≤ inf {I(h1 ) + I(h2 ) : f ≤ h1 ∈ S↑ and g ≤ h2 ∈ S↑ } P∞
a sequence of functions and F := n=1 fn . Then
= I ∗ (f ) + I ∗ (g)
∞
X ∞
X
and since k ∈ (f + g) is arbitrary we have proven Eq. (32.6). I ∗ (F ) = I ∗ ( fn ) ≤ I ∗ (fn ). (32.7)
4. From the definitions and Exercise 32.1, n=1 n=1
P∞
I∗ (−g) = sup {I(f ) : f ≤ −g ∈ S↓ } = sup {I(f ) : g ≤ −f ∈ S↑ } Proof. Suppose n=1 I ∗ (fn ) < ∞, for otherwise the result is trivial. Let
= sup {I(−h) : g ≤ h ∈ S↑ } = − inf {I(h) : g ≤ h ∈ S↑ } ε > 0 be given and choose gn ∈ S+ I(gn ) = I ∗ (fn ) + εn
↑ such that fn ≤ gn and P
P∞ ∞
= −I ∗ (g). where n=1 εn ≤ ε. (For example take εn ≤ 2 ε.) Then n=1 gn =: G ∈ S+
−n
↑,
F ≤ G and so
5. The assertion is trivially true if I ∗ (g) = I∗ (g) = ∞ or I ∗ (g) = I∗ (g) = −∞. ∞ ∞ ∞
So we now assume that I ∗ (g) and I∗ (g) are not both ∞ or −∞. Since
X X X
I ∗ (F ) ≤ I ∗ (G) = I(G) = I(gn ) = (I ∗ (fn ) + εn ) ≤ I ∗ (fn ) + ε.
0 ∈ (g − g) and I ∗ (g − g) ≤ I ∗ (g) + I ∗ (−g) (by Item 1), n=1 n=1 n=1
0 = I ∗ (0) ≤ I ∗ (g) + I ∗ (−g) = I ∗ (g) − I∗ (g) Since ε > 0 is arbitrary, the proof is complete.
provided the right side is well defined which it is by assumption. So again Proposition 32.20. The space L1 (I) is an extended lattice and I¯ : L1 (I) → R
we deduce that I∗ (g) ≤ I ∗ (g). is linear in the sense of Definition 32.3.

Proof. Let us begin by showing that L1 (I) is a vector space. Suppose that 0 ≤ h1 ∨h2 − f1 ∨f2 ≤ h1 − f1 + h2 − f2 ,
g1 , g2 ∈ L1 (I), and g ∈ (g1 + g2 ). Given ε > 0 there exists fi ∈ S↓ ∩ L1 (I) and
hi ∈ S↑ ∩ L1 (I) such that fi ≤ gi ≤ hi and I(hi − fi ) < ε/2. Let us now show because, for example, if h1 ∨h2 = h1 and f1 ∨f2 = f2 then
h1 ∨h2 − f1 ∨f2 = h1 − f2 ≤ h1 − f1 .
f1 (x) + f2 (x) ≤ g(x) ≤ h1 (x) + h2 (x) ∀x ∈ X. (32.8)
Therefore,
This is clear at points x ∈ X where g1 (x) + g2 (x) is well defined. The other
I (h1 ∨h2 − f1 ∨f2 ) ≤ I (h1 − f1 + h2 − f2 ) < ε
case to consider is where g1 (x) = ∞ = −g2 (x) in which case h1 (x) = ∞ and
f2 (x) = −∞ while , h2 (x) > −∞ and f1 (x) < ∞ because h2 ∈ S↑ and f1 ∈ S↓ . and hence by Remark 32.17, g1 ∨g2 ∈ L1 (I).
Therefore, f1 (x)+f2 (x) = −∞ and h1 (x)+h2 (x) = ∞ so that Eq. (32.8) is valid
Theorem 32.21 (Monotone convergence theorem). If fn ∈ L1 (I) and
no matter how g(x) is chosen. Since f1 + f2 ∈ S↓ ∩ L1 (I), h1 + h2 ∈ S↑ ∩ L1 (I) ¯ n ) = supn I(f
¯ n ) < ∞ in which case
fn ↑ f, then f ∈ L1 (I) iff limn→∞ I(f
and ¯ ¯
¯ i ) ≤ I(fi ) + ε/2 and − ε/2 + I(hi ) ≤ I(g
¯ i ), I(f ) = limn→∞ I(fn ).
I(g
Proof. If f ∈ L1 (I), then by monotonicity I(f ¯ n ) ≤ I(f
¯ ) for all n and
we find ¯ ¯ ¯ n) <
therefore limn→∞P∞I(fn ) ≤ I(f ) < ∞. Conversely, suppose ` := limn→∞ I(f
¯ 1 ) + I(g
I(g ¯ 2 ) − ε ≤ I(f1 ) + I(f2 ) = I(f1 + f2 ) ≤ I∗ (g) ≤ I ∗ (g) ∞ and let g := n=1 (fn+1 −fn )0 . The reader should check that f ≤ (f1 +g)∞ ∈
¯ 1 ) + I(g
¯ 2 ) + ε. (f1 + g) . So by Lemma 32.19,
≤ I(h1 + h2 ) = I(h1 ) + I(h2 ) ≤ I(g
¯ 1 ) + I(g
¯ 2) = I ∗ (f ) ≤ I ∗ ((f1 + g)∞ ) ≤ I ∗ (f1 ) + I ∗ (g)
Because ε > 0 is arbitrary, we have shown that g ∈ L1 (I) and I(g ∞ ∞
¯
I(g), ¯ 1 + g2 ) = I(g
i.e. I(g ¯ 1 ) + I(g
¯ 2 ). It is a simple matter to show λg ∈ L1 (I) X X
≤ I ∗ (f1 ) + ¯ 1) +
I ∗ ((fn+1 − fn )0 ) = I(f I¯ (fn+1 − fn )
¯
and I(λg) ¯
= λI(g) for all g ∈ L1 (I) and λ ∈ R. For example if λ = −1 (the n=1 n=1
most interesting case), choose f ∈ S↓ ∩ L1 (I) and h ∈ S↑ ∩ L1 (I) such that ∞
X
f ≤ g ≤ h and I(h − f ) < ε. Therefore, ¯ 1) + ¯ n+1 ) − I(f
¯ n ) = I(f
¯ 1 ) + ` − I(f
¯ 1 ) = `.

= I(f I(f (32.9)
n=1
S↓ ∩ L1 (I) 3 −h ≤ −g ≤ −f ∈ S↑ ∩ L1 (I)
Because fn ≤ f, it follows that I(f ¯ n ) = I∗ (fn ) ≤ I∗ (f ) which upon passing
with I(−f − (−h)) = I(h − f ) < ε and this shows that −g ∈ L1 (I) and to limit implies ` ≤ I∗ (f ). This inequality and the one in Eq. (32.9) shows
¯
I(−g) ¯
= −I(g). We have now shown that L1 (I) is a vector space of extended ¯ ) = ` = limn→∞ I(f
I ∗ (f ) ≤ ` ≤ I∗ (f ) and therefore, f ∈ L1 (I) and I(f ¯ n ).
real valued functions and I¯ : L1 (I) → R is linear. To show L1 (I) is a lattice, let +
g1 , g2 ∈ L1 (I) and fi ∈ S↓ ∩ L1 (I) and hi ∈ S↑ ∩ L1 (I) such that fi ≤ gi ≤ hi Lemma 32.22 (Fatou’s Lemma). Suppose {fn } ⊂ L1 (I) , then inf fn ∈
and I(hi − fi ) < ε/2 as above. Then using Proposition 32.13 and Remark 32.14, ¯ n ) < ∞, then lim inf n→∞ fn ∈ L1 (I) and in this case
L1 (I). If lim inf n→∞ I(f
¯
I(lim ¯ n ).
inf fn ) ≤ lim inf I(f
S↓ ∩ L1 (I) 3 f1 ∧ f2 ≤ g1 ∧ g2 ≤ h1 ∧ h2 ∈ S↑ ∩ L1 (I). n→∞ n→∞
Moreover, Proof. Let gk := f1 ∧· · ·∧fk ∈ L1 (I), then gk ↓ g := inf n fn . Since −gk ↑ −g,
0 ≤ h1 ∧ h2 − f1 ∧ f2 ≤ h1 − f1 + h2 − f2 , ¯
−gk ∈ L1 (I) for all k and I(−g ¯
k ) ≤ I(0) = 0, it follow from Theorem 32.21
that −g ∈ L (I) and hence so is inf n fn = g ∈ L1 (I). By what we have just
1
because, for example, if h1 ∧ h2 = h1 and f1 ∧ f2 = f2 then
proved, uk := inf n≥k fn ∈ L1 (I) for all k. Notice that uk ↑ lim inf n→∞ fn , and
¯ k ) ≤ I(f
by monotonicity that I(u ¯ k ) for all k. Therefore,
h1 ∧ h2 − f1 ∧ f2 = h1 − f2 ≤ h2 − f2 .
¯ k ) = lim inf I(u
lim I(u ¯ k ) ≤ lim inf I(f
¯ n) < ∞
Therefore, k→∞ k→∞ k→∞
I (h1 ∧ h2 − f1 ∧ f2 ) ≤ I (h1 − f1 + h2 − f2 ) < ε
and by the monotone convergence Theorem 32.21, lim inf n→∞ fn =
and hence by Remark 32.17, g1 ∧ g2 ∈ L1 (I). Similarly limk→∞ uk ∈ L1 (I) and

32.2 The Structure of L1 (I) 359
¯
I(lim ¯ k ) ≤ lim inf I(f
inf fn ) = lim I(u ¯ n ). 6. By Proposition 32.20, |f | ∈ L1 (I) and so by Chebyshev’s inequality (Item
n→∞ k→∞ n→∞
2 of Proposition 32.18), {|f | = ∞} is a null set.
Before stating the dominated convergence theorem, it is helpful to remove

some of the annoyances of dealing with extended real valued functions. As we Theorem 32.25 (Dominated Convergence Theorem). Suppose that
have done when studying integrals associated to a measure, we can do this by {fn : n ∈ N} ⊂ L1 (I) such that f := lim fn exists pointwise and there exists
modifying integrable functions by a “null” function. g ∈ L1 (I) such that |fn | ≤ g for all n. Then f ∈ L1 (I) and
Definition 32.23. A function n : X → R̄ is a null function if I ∗ (|n|) = 0. ¯ n ) = I(
lim I(f ¯ lim fn ) = I(f
¯ ).
n→∞ n→∞
A subset E ⊂ X is said to be a null set if 1E is a null function. Given two
functions f, g : X → R̄ we will write f = g a.e. if {f 6= g} is a null set. Proof. By Proposition 32.24, the set E := {g = ∞} is a null set and
¯ E c fn ) = I(f
I(1 ¯ n ) and I(1
¯ E c g) = I(g).
¯ Since
Here are some basic properties of null functions and null sets.
¯ E c (g ± fn )) ≤ 2I(1
I(1 ¯ E c g) = 2I(g)
¯ < ∞,
Proposition 32.24. Suppose that n : X → R̄ is a null function and f : X → R̄
is an arbitrary function. Then we may apply Fatou’s Lemma 32.22 to find 1E c (g ± f ) ∈ L1 (I) and
¯
1. n ∈ L1 (I) and I(n) = 0. ¯ E c (g ± f )) ≤ lim inf I(1
I(1 ¯ E c (g ± fn ))
n→∞
2. The function n · f is a null function.
¯ E c g) ± I(1
¯ E c fn ) = lim inf I(g)

¯ ± I(f
¯ n) .

= lim inf I(1
3. The set {x ∈ X : n(x) 6= 0} is a null set. n→∞ n→∞
¯ ) = I(1
4. If E is a null set and f ∈ L1 (I), then 1E c f ∈ L1 (I) and I(f ¯ E c f ).
1 1 ¯ ¯ Since f = 1E c f a.e. and 1E c f = 12 1E c (g + f − (g + f )) ∈ L1 (I), Proposition
5. If g ∈ L (I) and f = g a.e. then f ∈ L (I) and I(f ) = I(g).
32.24 implies f ∈ L1 (I). So the previous inequality may be written as
6. If f ∈ L1 (I), then E := {|f | = ∞} is a null set.
¯ ± I(f
I(g) ¯ ) = I(1
¯ E c g) ± I(1
¯ Ec f )
Proof.
¯ n)

¯ E c (g ± f )) ≤ I(g)
¯ + lim inf n→∞ I(f
∗ ∗ ∗
1. If n is null, using ±n ≤ |n| we find I (±n) ≤ I (|n|) = 0, i.e. I (n) ≤ 0 and = I(1 ¯ n ),
− lim supn→∞ I(f
−I∗ (n) = I ∗ (−n) ≤ 0. Thus it follows that I ∗ (n) ≤ 0 ≤ I∗ (n) and therefore
n ∈ L1 (I) with I¯ (n) = 0. wherein we have used lim inf n→∞ (−an ) = − lim sup an . These two inequal-
2. Since |n · f | ≤ ∞ · |n| , I ∗ (|n · f |) ≤ I ∗ (∞ · |n|) . For k ∈ N, k |n| ∈ L1 (I) ities imply lim supn→∞ I(f¯ n ) ≤ I(f
¯ ) ≤ lim inf n→∞ I(f
¯ n ) which shows that
¯ |n|) = kI (|n|) = 0, so k |n| is a null function. By the monotone
and I(k ¯ ¯
lim I(fn ) exists and is equal to I(f ).
n→∞
convergence Theorem 32.21 and the fact k |n| ↑ ∞ · |n| ∈ L1 (I) as k ↑ ∞,
I¯ (∞ · |n|) = limk→∞ I¯ (k |n|) = 0. Therefore ∞ · |n| is a null function and
hence so is n · f. 32.2 The Structure of L1 (I)
3. Since 1{n6=0} ≤ ∞ · 1{n6=0} = ∞ · |n| , I ∗ 1{n6=0} ≤ I ∗ (∞ · |n|) = 0 showing

{n 6= 0} is a null set. Let S↑↓ (I) denote the collections of functions f : X → R̄ for which there exists
4. Since 1E f ∈ L1 (I) and I¯ (1E f ) = 0, fn ∈ S↑ ∩ L1 (I) such that fn ↓ f as n → ∞ and limn→∞ I(f¯ n ) > −∞. Applying
the monotone convergence theorem to f1 − fn , it follows that f1 − f ∈ L1 (I)
f 1E c = (f − 1E f )0 ∈ (f − 1E f ) ⊂ L1 (I) and hence −f ∈ L1 (I) so that S↑↓ (I) ⊂ L1 (I).
¯ 1E c ) = I(f
and I(f ¯ ) − I(1
¯ E f ) = I(f
¯ ). Lemma 32.26. Let f : X → R̄ be a function. If I ∗ (f ) ∈ R, then there exists
5. Letting E be the null set {f 6= g} , then 1E c f = 1E c g ∈ L1 (I) and 1E f is a ¯
g ∈ S↑↓ (I) such that f ≤ g and I ∗ (f ) = I(g). (Consequently, n : X → [0, , ∞)
null function and therefore, f = 1E f + 1E c f ∈ L1 (I) and is a positive null function iff there exists g ∈ S↑↓ (I) such that g ≥ n and
¯ = 0.) Moreover, f ∈ L1 (I) iff there exists g ∈ S↑↓ (I) such that g ≥ f and
I(g)
¯ ) = I(1
I(f ¯ E f ) + I(f
¯ 1E c ) = I(1
¯ E c f ) = I(1
¯ E c g) = I(g).
¯
f = g a.e.

Proof. By definition of I ∗ (f ) we may choose a sequence of functions gk ∈ (f ∧ g) ∧ h = f ∧ (g ∧ h) and (f ∨ g) ∧ h = (f ∧ h) ∨ (g ∧ h)

¯ k ) ↓ I ∗ (f ). By replacing gk by g1 ∧ · · · ∧ gk if
S↑ ∩ L1 (I) such that gk ≥ f and I(g
1
necessary (g1 ∧· · ·∧gk ∈ S↑ ∩L (I) by Proposition 32.13), we may assume that gk and the fact that L1 (I) is a lattice. If fn : X → [0, ∞] is a sequence of measur-
is a decreasing sequence. Then limk→∞ gk =: g ≥ f and, since limk→∞ I(g ¯ k) = able functions such that f = limn→∞ fn exists pointwise, then for h ∈ L1 (I),
∗
I (f ) > −∞, g ∈ S↑↓ (I) . By the monotone convergence theorem applied to we have h ∧ fn → h ∧ f . By the dominated convergence theorem (using
g1 − gk , |h ∧ fn | ≤ |h|) it follows that h ∧ f ∈ L1 (I). Since h ∈ L1 (I) is arbitrary
¯ 1 − g) = lim I(g
I(g ¯ 1 − gk ) = I(g ¯ 1 ) − I ∗ (f ), we conclude that f is measurable as well.
k→∞
¯ = I ∗ (f ). Now suppose that f ∈ L1 (I), then (g − f )0 ≥ 0 and
so I(g) Lemma 32.30. A non-negative function f on X is measurable iff φ∧f ∈ L1 (I)
for all φ ∈ S.
I¯ ((g − f )0 ) = I¯ (g) − I(f
¯ ) = I(g)
¯ − I ∗ (f ) = 0.
Proof. Suppose f : X → [0, ∞] is a function such that φ ∧ f ∈ L1 (I)
Therefore (g − f )0 is a null functions and hence so is ∞ · (g − f )0 . Because for all φ ∈ S and let g ∈ S↑ ∩ L1 (I). Choose φn ∈ S such that φn ↑ g as
n → ∞, then φn ∧ f ∈ L1 (I) and by the monotone convergence Theorem 32.21,
1{f 6=g} = 1{f <g} ≤ ∞ · (g − f )0 , φn ∧ f ↑ g ∧ f ∈ L1 (I). Similarly, using the dominated convergence Theorem
32.25, it follows that g ∧ f ∈ L1 (I) for all g ∈ S↑↓ (I) . Finally for any h ∈ L1 (I),
{f 6= g} is a null set so if f ∈ L1 (I) there exists g ∈ S↑↓ (I) such that f = g a.e. there exists g ∈ S↑↓ (I) such that h = g a.e. and hence h ∧ f = g ∧ f a.e. and
The converse statement has already been proved in Proposition 32.24. therefore by Proposition 32.24, h ∧ f ∈ L1 (I). This completes the proof since
the converse direction is trivial.
Proposition 32.27. Suppose that I and S are as above and J is another
Daniell integral on a vector lattice T such that S ⊂ T and I = J|S . (We abbre- Definition 32.31. A set A ⊂ X is measurable if 1A is measurable and A
viate this by writing I ⊂ J.) Then L1 (I) ⊂ L1 (J) and I¯ = J¯ on L1 (I), or in integrable if 1A ∈ L1 (I). Let R denote the collection of measurable subsets of
abbreviated form: if I ⊂ J then I¯ ⊂ J.
¯ X.
Proof. From the construction of the extensions, it follows that S↑ ⊂ T↑ Remark 32.32. Suppose that f ≥ 0, then f ∈ L1 (I) iff f is measurable and
and the I = J on S↑ . Similarly, it follows that S↑↓ (I) ⊂ T↑↓ (J) and I¯ = J¯ on I ∗ (f ) < ∞. Indeed, if f is measurable and I ∗ (f ) < ∞, there exists g ∈ S↑ ∩L1 (I)
S↑↓ (I) . From Lemma 32.26 we learn, if n ≥ 0 is an I – null function then there such that f ≤ g. Since f is measurable, f = f ∧g ∈ L1 (I). In particular if A ∈ R,
exists g ∈ S↑↓ (I) ⊂ T↑↓ (J) such that n ≤ g and 0 = I(g) = J(g). This shows then A is integrable iff I ∗ (1A ) < ∞.
that n is also a J – null function and in particular every I – null set is a J –
null set. Again by Lemma 32.26, if f ∈ L1 (I) there exists g ∈ S↑↓ (I) ⊂ T↑↓ (J) Lemma 32.33. The set R is a ring which is a σ – algebra if 1 is measurable.
such that {f 6= g} is an I – null set and hence a J – null set. So by Proposition (Notice that 1 is measurable iff 1 ∧ φ ∈ L1 (I) for all φ ∈ S. This condition is
32.24, f ∈ L1 (J) and I(f ) = I(g) = J(g) = J(f ). clearly implied by assuming 1 ∧ φ ∈ S for all φ ∈ S. This will be the typical case
in applications.)
Proof. Suppose that A, B ∈ R, then A ∩ B and A ∪ B are in R by Lemma

32.3 Relationship to Measure Theory 32.29 because
1A∩B = 1A ∧ 1B and 1A∪B = 1A ∨ 1B .
Definition 32.28. A function f : X → [0, ∞] is said to I-measurable (or
just measurable) if f ∧ g ∈ L1 (I) for all g ∈ L1 (I). If Ak ∈ R, then the identities,
Lemma 32.29. The set of non-negative measurable functions is closed under 1∪∞
k=1 Ak
= lim 1∪nk=1 Ak and 1∩∞
k=1 Ak
= lim 1∩nk=1 Ak
n→∞ n→∞
pairwise minimums and maximums and pointwise limits.
along with Lemma 32.29 shows that ∪∞ ∞
k=1 Ak and ∩k=1 Ak are in R as well. Also
Proof. Suppose that f, g : X → [0, ∞] are measurable functions. The fact if A, B ∈ R and g ∈ S, then
that f ∧ g and f ∨ g are measurable (i.e. (f ∧ g) ∧ h and (f ∨ g) ∨ h are in L1 (I)
for all h ∈ L1 (I)) follows from the identities g ∧ 1A\B = g ∧ 1A − g ∧ 1A∩B + g ∧ 0 ∈ L1 (I) (32.10)

32.3 Relationship to Measure Theory 361
showing the A \ B ∈ R as well.2 Thus we have shown that R is a ring. If 1 = 1X Notice by Remark 32.32 that
is measurable it follows that X ∈ R and R becomes a σ – algebra.
¯ A ) if
I(1 A is integrable
µ+ (A) =
Lemma 32.34 (Chebyshev’s Inequality). Suppose that 1 is measurable. ∞ if A ∈ R but A is not integrable.
+ Proof. Since 1∅ = 0, µ± (∅) = I(0) ¯ = 0 and if A, B ∈ R, A ⊂ B then
1. If f ∈ L1 (I) then, for all α ∈ R, the set {f > α} is measurable. More-
¯ {f >α} ) ≤ α−1 I(f
over, if α > 0 then {f > α} is integrable and I(1 ¯ ). µ+ (A) = I ∗ (1A ) ≤ I ∗ (1B ) = µ+ (B) and similarly, µ− (A) = I∗ (1A ) ≤ I∗ (1B ) =
2. σ(S) ⊂ R. µ− (B). Hence µ± are monotonic. By Remark 32.32 if µ+ (A) < ∞ then A is
integrable so
Proof.
¯ A ) = I ∗ (1A ) = µ+ (A).
µ− (A) = I∗ (1A ) = I(1
1. If α < 0, {f > α} = X ∈ R since 1 is measurable. So now assume that ∞
α ≥ 0. If α = 0 let g = f ∈ L1 (I) and if α > 0 let g = α−1 f − α−1 f ∧ 1. Now suppose that {Ej }j=1 ⊂ R is a sequence of pairwise disjoint sets and let
(Notice that g is a difference of two L1 (I) – functions and hence in L1 (I).) E := ∪∞ j=1 Ej ∈ R. If µ+ (Ei ) = ∞ for some i then by monotonicity µ+ (E) = ∞
+ Pn +
The function g ∈ L1 (I) has been manufactured so that {g > 0} = {f >

as well. If µ+ (Ej ) < ∞ for all j then fn := j=1 1Ej ∈ L1 (I) with fn ↑ 1E .
+
Therefore, by the monotone convergence theorem, 1E is integrable iff

α}. Now let φn := (ng) ∧ 1 ∈ L1 (I) then φn ↑ 1{f >α} as n → ∞ showing
1{f >α} is measurable and hence that {f > α} is measurable. Finally if ∞
X
α > 0, ¯ n) =
lim I(f µ+ (Ej ) < ∞
n→∞
1{f >α} = 1{f >α} ∧ α−1 f ∈ L1 (I)

j=1
showing the {f > α} is integrable and ¯ n ) = I(1

in which case 1E ∈ L1 (I) and limn→∞ I(f ¯ E ) = µ+ (E). Thus we have
shown that µ+ is a measure and µ− (E) = µ+ (E) whenever µ+ (E) < ∞. The
¯ {f >α} ) = I(1
¯ {f >α} ∧ α−1 f ) ≤ I(α
¯ −1 f ) = α−1 I(f
¯ ).

I(1 fact the µ− is a measure will be shown in the course of the proof of Theorem
32.38.
2. Since f ∈ S+ is R measurable by (1) and S = S+ − S+ , it follows that any
f ∈ S is R measurable, σ(S) ⊂ R. Example 32.36. Suppose X is a set, S = {0} is the trivial vector space and
I(0) = 0. Then clearly I is a Daniel integral,

∗ ∞ if g(x) > 0 for some x
Lemma 32.35. Let 1 be measurable. Define µ± : R → [0, ∞] by I (g) =
0 if g≤0
µ+ (A) = I ∗ (1A ) and µ− (A) = I∗ (1A ) and similarly,
−∞ if g(x) < 0 for some x
Then µ± are measures on R such that µ− ≤ µ+ and µ− (A) = µ+ (A) whenever I∗ (g) =
0 if g ≥ 0.
µ+ (A) < ∞.
Therefore, L1 (I) = {0} and for any A ⊂ X we have 1A ∧ 0 = 0 ∈ S so that
2
Indeed, for x ∈ A ∩ B, x ∈ A \ B and x ∈ Ac , Eq. (32.10) evaluated at x states, R = 2X . Since 1A ∈/ L1 (I) = {0} unless A = ∅ set, the measure µ+ in Lemma
respectively, that 32.35 is given by µ+ (A) = ∞ if A 6= ∅ and µ+ (∅) = 0, i.e. µ+ (A) = I ∗ (1A )
g ∧ 0 = g ∧ 1 − g ∧ 1 + g ∧ 0,
while µ− ≡ 0.
g ∧ 1 = g ∧ 1 − g ∧ 0 + g ∧ 0 and Lemma 32.37. For A ∈ R, let
g ∧ 0 = g ∧ 0 − g ∧ 0 + g ∧ 0,
α(A) := sup{µ+ (B) : B ∈ R, B ⊂ A and µ+ (B) < ∞},
all of which are true.
then α is a measure on R such that α(A) = µ+ (A) whenever µ+ (A) < ∞. If ν
is any measure on R such that ν(B) = µ+ (B) when µ+ (B) < ∞, then α ≤ ν.
Moreover, α ≤ µ− .

Proof. Clearly α(A) = µ+ (A) whenever µ+ (A) < ∞. Now let A = ∪∞ n=1 An 4. Conversely if ν is any measure on σ(S) such that ν(A) = µ+ (A) when
∞
with{An }n=1 ⊂ R being a collection of pairwise disjoint subsets. Let Bn ⊂ An A ∈ σ(S) and µ+ (A) < ∞, then Eq. (32.12) is valid.
with µ+ (Bn ) < ∞, then B N := ∪N N
n=1 Bn ⊂ A and µ+ (B ) < ∞ and hence
Proof.
N +
X 1. Suppose that f ∈ L1 (I) , then Lemma 32.34 implies that f is R mea-
α(A) ≥ µ+ (B N ) = µ+ (Bn )
n=1
surable. Given n ∈ N, let
2n 2n
2 2
and since Bn ⊂ An with µ+ (Bn ) < ∞ is arbitrary it follows that α(A) ≥ X k −n
X
PN P∞ φn := 1 k k+1 = 2 1{ kn <f } . (32.13)
n=1 α(An ) and hence letting N → ∞ implies α(A) ≥ n=1 α(An ). Con- 2n { 2n <f ≤ 2n } 2
k=1 k=1
versely, if B ⊂ A with µ+ (B) < ∞, then B ∩ An ⊂ An and µ+ (B ∩ An ) < ∞.
Therefore, n
Then we know { 2kn < f } ∈ R and that 1{ kn <f } = 1{ kn <f } ∧ 2k f ∈ L1 (I),

∞ ∞
X X 2 + 2
i.e. µ+ 2kn < f < ∞. Therefore φn ∈ L1 (I) and φn ↑ f. Suppose that

µ+ (B) = µ+ (B ∩ An ) ≤ α(An )
n=1 n=1 ν is any measure such that ν(A) = µ+ (A) when µ+ (A) < ∞, then by the
P∞
for all such B and hence α(A) ≤ n=1 α(An ). Using the definition of α and monotone convergence theorems for I¯ and the Lebesgue integral,
the assumption that ν(B) = µ+ (B) when µ+ (B) < ∞, 2n 2n
2 2
¯ ) = lim I(φ
¯ n ) = lim 2−n
X
¯ −n
X k
α(A) = sup{ν(B) : B ∈ R, B ⊂ A and µ+ (B) < ∞} ≤ ν(A), I(f I(1{ k
<f } ) = lim 2 µ+ <f
n→∞ n→∞ 2n n→∞ 2n
k=1 k=1
showing α ≤ ν. Similarly, 22n Z Z

−n
X k
= lim 2 ν < f = lim φn dν = f dν. (32.14)
¯ B ) : B ∈ R, B ⊂ A and µ+ (B) < ∞}
α(A) = sup{I(1
n→∞ 2n n→∞ X X
k=1
= sup{I∗ (1B ) : B ∈ R, B ⊂ A and µ+ (B) < ∞} ≤ I∗ (1A ) = µ− (A). +

¯ ) =

This shows that f ∈ L1 (ν)
R
and that I(f f dν. Since every
X
− ±
+
f ∈ L (I) is of the form f = f − f with f ∈ L1 (I) , it fol-
1 +
lows that L1 (I) ⊂ L1 (µ+ ) ⊂ L1 (ν) ⊂ L1 (α) and Eq. (32.12) holds for
Theorem 32.38 (Stone). Suppose that 1 is measurable and µ+ and µ− are as +
all f ∈ L1 (I). Conversely suppose that f ∈ L1 (µ+ ) . Define φn as in
defined in Lemma 32.35, then:
Eq. (32.13). Chebyshev’s inequality implies that µ+ ( 2kn < f ) < ∞ and
1. L1 (I) = L1 (X, R, µ+ ) = L1 (µ+ ) and for integrable f ∈ L1 (µ+ ), hence { 2kn < f } is I – integrable. Again by the monotone convergence for
Z Lebesgue integrals and the computations in Eq. (32.14),
¯
I(f ) = f dµ+ . (32.11) Z
X ∞> ¯ n)
f dµ+ = lim I(φ
X n→∞
2. If ν is any measure on R such that S ⊂ L1 (ν) and
Z ¯ f ∈ L1 (I) and
and therefore by the monotone convergence theorem for I,
¯
I(f ) = f dν for all f ∈ S (32.12) Z
X ¯ n ) = I(f
f dµ+ = lim I(φ ¯ ).
X n→∞
then µ− (A) ≤ ν(A) ≤ µ+ (A) for all A ∈ R with µ− (A) = ν(A) = µ+ (A)
whenever µ+ (A) < ∞. 2. Suppose that ν is any measure such that Eq. (32.12) holds. Then by the
3. Letting α be as defined in Lemma 32.37, µ− = α and hence µ− is a measure. monotone convergence theorem,
(So µ+ is the maximal and µ− is the minimal measure for which Eq. (32.12) Z
holds.) I(f ) = f dν for all f ∈ S↑ ∪ S↓ .
X

32.4 Extensions of premeasures to measures 363
Let A ∈ R and assume that µ+ (A) < ∞, i.e. 1A ∈ L1 (I). Then there exists µ0 (Xn ) < ∞ and Xn ↑ X as n → ∞. Then µ0 has a unique extension to a
f ∈ S↑ ∩ L1 (I) such that 1A ≤ f and integrating this inequality relative to measure, µ, on M := σ(A). Moreover, if A ∈ M and ε > 0 is given, there
ν implies Z Z exists B ∈ Aσ such that A ⊂ B and µ(B \ A) < ε. In particular,
ν(A) = 1A dν ≤ ¯ ).
f dν = I(f
X X
µ(A) = inf{µ0 (B) : A ⊂ B ∈ Aσ } (32.15)
∞ ∞
Taking the infimum of this equation over those f ∈ S↑ such that 1A ≤ f X a
implies ν(A) ≤ I ∗ (1A ) = µ+ (A). If µ+ (A) = ∞ in this inequality holds = inf{ µ0 (An ) : A ⊂ An with An ∈ A}. (32.16)
n=1 n=1
trivially. Similarly, if A ∈ R and f ∈ S↓ such that 0 ≤ f ≤ 1A , then
Z Z Proof. Let (A, µ0 , I = Iµ0 ) be as in Definition 28.36. By Proposition 32.9, I
ν(A) = 1A dν ≥ ¯ ).
f dν = I(f
X X
P∞ S = Sf (A, µ0 ). It is clear that 1 ∧ φ ∈ S for all
is a Daniell integral on the lattice
φ ∈ S. Since 1Xn ∈ S+ and n=1 1Xn > 0 on X, by Remark 32.45 there exists
Taking the supremum of this equation over those f ∈ S↓ such that 0 ≤ f ≤ χ ∈ S↑ such that I(χ) < ∞ and χ > 0. So the hypothesis of TheoremR 32.44
1A then implies ν(A) ≥ µ− (A). So we have shown that µ− ≤ ν ≤ µ+ . hold and hence there exists a unique measure µ on M such that I(f ) = X f dµ
3. By Lemma 32.37, ν = α is a measure as in (2) satisfying α ≤ µ− and for all f ∈ S. Taking f = 1A with A ∈ A and µ0 (A) < ∞ shows µ(A) = µ0 (A).
therefore µ− ≤ α and hence we have shown that α = µ− . This also shows For general A ∈ A, we have
that µ− is a measure.
4. This can be done by the same type of argument used in the proof of (1). µ(A) = lim µ(A ∩ Xn ) = lim µ0 (A ∩ Xn ) = µ0 (A).
n→∞ n→∞
The fact that µ is the only extension of µ0 to M follows from Theorem 33.6
Proposition 32.39 (Uniqueness). Suppose that 1 is measurable and there or Theorem 19.55. It is also can be proved using Theorem 32.44. Indeed, if ν is
exists a function χ ∈ L1 (I) such that χ(x) > 0 for all x. Then there is only one another measure on M such that ν = µ on A, then Iν = I on S. Therefore by
measure µ on σ(S) such that the uniqueness assertion in Theorem 32.44, µ = ν on M. By Eq. (32.20), for
Z A ∈ M,
¯ )=
I(f f dµ for all f ∈ S.
X µ(A) = I ∗ (1A ) = inf {I(f ) : f ∈ S↑ with 1A ≤ f }
Z
Remark 32.40. The existence of a function χ ∈ L1 (I) such that χ(x) > 0 for = inf f dµ : f ∈ S↑ with 1A ≤ f .
¯
all x is equivalent to the existence of a function χ ∈ S↑ such that I(χ) <∞ X
1
and χ(x) > 0 for all x ∈ X. Indeed by Lemma 32.26, if χ ∈ L (I) there exists
For the moment suppose µ(A) < ∞ and ε > 0 is given. Choose f ∈ S↑ such
χ̃ ∈ S↑ ∩ L1 (I) such χ̃ ≥ χ.
that 1A ≤ f and Z
Proof. As in Remark 32.40, we may assume χ ∈ S↑ ∩ L1 (I). The sets f dµ = I(f ) < µ(A) + ε. (32.17)
Xn := {χ > 1/n} ∈ σ(S) ⊂ R satisfy µ(Xn ) ≤ nI(χ) ¯ < ∞. The proof is X
completed using Theorem 32.38 to conclude, for any A ∈ σ(S), that Let fn ∈ S be a sequence such that fn ↑ f as n → ∞ and for α ∈ (0, 1) set
µ+ (A) = lim µ+ (A ∩ Xn ) = lim µ− (A ∩ Xn ) = µ− (A).
n→∞ n→∞ Bα := {f > α} = ∪∞
n=1 {fn > α} ∈ Aσ .
Since µ− ≤ µ ≤ µ+ = µ−, we see that µ = µ+ = µ− . Then A ⊂ {f ≥ 1} ⊂ Bα and by Chebyshev’s inequality,

Z
µ(Bα ) ≤ α−1 f dµ = α−1 I(f )
32.4 Extensions of premeasures to measures X
Theorem 32.41. Let X be a set, A be a subalgebra of 2X and µ0 be a pre- which combined with Eq. (32.17) implies µ(Bα ) < µ(A) + ε for all α sufficiently
measure on A which is σ – finite on A, i.e. there exists Xn ∈ A such that close to 1. For such α we then have A ⊂ Bα ∈ Aσ and µ(Bα \ A) = µ(Bα ) −

Z Z
µ(A) < ε. For general A ∈ A, choose Xn ↑ X with Xn ∈ A. Then there exists p
kh − gkp =
p
|h − g| dµ = |h − g|
p−1
|h − g| dµ
Bn ∈ Aσ such that µ(Bn \ (An ∩ Xn )) < ε2−n . Define B := ∪∞
n=1 Bn ∈ Aσ . X
Z X
Then p−1 p−1
≤ (2n) |h − g| dµ < (2n) ε.
∞
X X
µ(B \ A) = µ (∪∞
n=1 (Bn \ A)) ≤ µ ((Bn \ A))
Since ε > 0 was arbitrary, this shows S is dense in Lp (µ; R).
n=1
∞
X Remark 32.43. If we drop the σ – finiteness assumption on µ0 we may loose
≤ µ ((Bn \ (A ∩ Xn )) < ε.
uniqueness assertion in Theorem 32.41. For example, let X = R, BR and A be
n=1
the algebra generated by E := {(a, b]∩R : −∞ ≤ a < b ≤ ∞}. Recall BR = σ(E).
Eq. (32.15) is an easy consequence of this result and the fact that µ(B) = µ0 (B). Let D ⊂ R be a countable dense set and define µD (A) := #(D ∩ A). Then
µD (A) = ∞ for all A ∈ A such that A 6= ∅. So if D0 ⊂ R is another countable
dense subset of R, µD0 = µD on A while µD 6= µD0 on BR . Also notice that µD
Corollary 32.42 (Regularity of µ). Let A ⊂ 2X be an algebra of sets, M = is σ – finite on BR but not on A.
σ(A) and µ : M → [0, ∞] be a measure on M which is σ – finite on A. Then
It is now possible to use Theorem 32.41 to give a proof of Theorem 19.8, see
1. For all A ∈ M, subsection 31.4.1 below. However rather than do this now let us give another
µ(A) = inf {µ(B) : A ⊂ B ∈ Aσ } . (32.18) application of Theorem 32.41 based on Proposition 32.9 and use the result to
2. If A ∈ M and ε > 0 are given, there exists B ∈ Aσ such that A ⊂ B and prove Theorem 19.8.
µ(B \ A) < ε.
3. For all A ∈ M and ε > 0 there exists B ∈ Aδ such that B ⊂ A and 32.4.1 A Useful Version: BRUCE: delete this if incorporated above.
µ(A \ B) < ε.
4. For any B ∈ M there exists A ∈ Aδσ and C ∈ Aσδ such that A ⊂ B ⊂ C We are now in a position to state the main construction theorem. The theorem
and µ(C \ A) = 0. we state here is not as general as possible but it will suffice for our present
5. The linear space S := Sf (A, µ) is dense in Lp (µ) for all p ∈ [1, ∞), briefly purposes.
Lp (µ)
put, Sf (A, µ) = Lp (µ). Theorem 32.44 (Daniell-Stone). Let S be a lattice of bounded functions on
a set X such that 1 ∧ φ ∈ S for all φ ∈ S and let I be a Daniel integral on S.
Proof. Items 1. and 2. follow by applying Theorem 32.41 to µ0 = µ|A . Items
Further assume there exists χ ∈ S↑ such that I(χ) < ∞ and χ(x) > 0 for all
3. and 4. follow from Items 1. and 2. as in the proof of Corollary 33.10 above.
x ∈ X. Then there exists a unique measure µ on M := σ(S) such that
Item 5. This has already been proved in Theorem 22.15 but we will give yet
another proof here. When p = 1 and g ∈RL1 (µ; R), there exists, by Eq. (32.20), Z
∞ I(f ) = f dµ for all f ∈ S. (32.19)
h ∈ S↑ such that g ≤ h and kh − gk1 = X (h − g)dµ < ε. Let {hn }n=1 ⊂ S be
X
chosen so that hn ↑ h as n → ∞. Then by the dominated convergence theorem,
khn − gk1 → kh − gk1 < ε as n → ∞. Therefore for n large we have hn ∈ S Moreover, for all g ∈ L1 (X, M, µ),
L1 (µ)
with khn − gk1 < ε. Since ε > 0 is arbitrary this shows, Sf (A, µ) = L1 (µ). Z
p sup {I(f ) : S↓ 3 f ≤ g} = gdµ = inf {I(h) : g ≤ h ∈ S↑ } . (32.20)
Now suppose p > 1, g ∈ L (µ; R) and Xn ∈ A are sets such that Xn ↑ X and
X
µ(Xn ) < ∞. By the dominated convergence theorem, 1Xn ·[(g ∧ n) ∨ (−n)] → g
in Lp (µ) as n → ∞, so it suffices to consider g ∈ Lp (µ; R) with {g 6= 0} ⊂ Xn Proof. Only a sketch of the proof will be given here. Full details may be
and |g| ≤ n for some large n ∈ N. By Hölder’s inequality, such a g is in L1 (µ). found in Section 32 below. Existence. For g : X → R̄, define
So if ε > 0, by the p = 1 case, we may find h ∈ S such that kh − gk1 < ε. By
replacing h by (h ∧ n) ∨ (−n) ∈ S, we may assume h is bounded by n as well I ∗ (g) := inf{I(h) : g ≤ h ∈ S↑ },
and hence
I∗ (g) := sup{I(f ) : S↓ 3 f ≤ g}

32.5 Riesz Representation Theorem 365
Z Z
and set µ(Xn ) = 1Xn dµ ≤ nχdµ = nI(χ) < ∞.
L1 (I) := {g : X → R̄ : I ∗ (g) = I∗ (g) ∈ R}. X X
¯
For g ∈ L1 (I), let I(g) = I ∗ (g) = I∗ (g). Then, as shown in Proposition 32.20, Since χ > 0 on X, Xn ↑ X and therefore by continuity of ν and µ,
L (I) is a “extended” vector space and I¯ : L1 (I) → R is linear as defined in
1
Definition 32.3 below. By Proposition 32.18, if f ∈ S↑ with I(f ) < ∞ then f ∈ ν(A) = lim ν(A ∩ Xn ) = lim µ(A ∩ Xn ) = µ(A)
n→∞ n→∞
L1 (I). Moreover, I¯ obeys the monotone convergence theorem, Fatou’s lemma,
and the dominated convergence theorem, see Theorem 32.21, Lemma 32.22 and for all A ∈ σ(S).
Theorem 32.25 respectively. Let
Remark 32.45. To check the hypothesis in Theorem 32.44 that there exists χ ∈
+
R := A ⊂ X : 1A ∧ f ∈ L1 (I) for all f ∈ S P∞I(χ) < ∞ and χ(x) > 0 for all x ∈ X, it suffices to find φn ∈ S
S↑ such that

such that P n=1 φn > 0 on X. To see this let Mn := max (kφn k∞ , I(φn ) , 1) and
and for A ∈ R set µ(A) := I ∗ (1A ). It can then be shown: 1) R is a σ algebra ∞
define χ := n=1 Mn12n φn , then χ ∈ S↑ , 0 < χ ≤ 1 and I(χ) ≤ 1 < ∞.
(Lemma 32.33) containing σ(S) (Lemma 32.34), µ is a measure on R (Lemma
32.35), and that Eq. (32.19) holds. In fact it is shown in Theorem 32.38 and
Proposition 32.39 below that L1 (X, M, µ) ⊂ L1 (I) and 32.5 Riesz Representation Theorem
Z
¯ =
I(g) gdµ for all g ∈ L1 (X, M, µ). Definition 32.46. Given a second countable locally compact Hausdorff space
X
(X, τ ), let M+ denote the collection of positive measures, µ, on BX := σ(τ ) with
The assertion in Eq. (32.20) is a consequence of the definition of L1 (I) and the property that µ(K) < ∞ for all compact subsets K ⊂ X. Such a measure
I¯ and this last equation. Uniqueness. Suppose that ν is another measure on µ will be Rcalled a Radon measure on X. For µ ∈ M+ and f ∈ Cc (X, R) let
σ(S) such that Z Iµ (f ) := X f dµ.
I(f ) = f dν for all f ∈ S. BRUCE: Consolidate the next theorem and Theorem 32.63.
X
By the monotone convergence theorem and the definition of I on S↑ , Theorem 32.47 (Riesz Representation Theorem). Let (X, τ ) be a second
Z countable3 locally compact Hausdorff space. Then the map µ → Iµ taking M+
I(f ) = f dν for all f ∈ S↑ . to positive linear functionals on Cc (X, R) is bijective. Moreover every measure
X
µ ∈ M+ has the following properties:
Therefore if A ∈ σ(S) ⊂ R,
1. For all ε > 0 and B ∈ BX , there exists F ⊂ B ⊂ U such that U is open
µ(A) = I ∗ (1A ) = inf{I(h) : 1A ≤ h ∈ S↑ } and F is closed and µ(U \ F ) < ε. If µ(B) < ∞, F may be taken to be a
Z Z compact subset of X.
= inf{ hdν : 1A ≤ h ∈ S↑ } ≥ 1A dν = ν(A) 2. For all B ∈ BX there exists A ∈ Fσ and C ∈ τδ (τδ is more conventionally
X X
written as Gδ ) such that A ⊂ B ⊂ C and µ(C \ A) = 0.
which shows ν ≤ µ. If A ∈ σ(S) ⊂ R with µ(A) < ∞, then, by Remark 32.32 3. For all B ∈ BX ,
below, 1A ∈ L1 (I) and therefore
µ(B) = inf{µ(U ) : B ⊂ U and U is open} (32.21)
¯ A ) = I∗ (1A ) = sup{I(f ) : S↓ 3 f ≤ 1A }
µ(A) = I ∗ (1A ) = I(1
Z = sup{µ(K) : K ⊂ B and K is compact}. (32.22)
= sup{ f dν : S↓ 3 f ≤ 1A } ≤ ν(A). 3
The second countability is assumed here in order to avoid certain technical issues.
X
Recall from Lemma 18.57 that under these assumptions, σ(S) = BX . Also recall
Hence µ(A) ≤ ν(A) for all A ∈ σ(S) and ν(A) = µ(A) when µ(A) < ∞. from Uryshon’s metrizatoin theorem that X is metrizable. We will later remove the
To prove ν(A) = µ(A) for all A ∈ σ(S), let Xn := {χ ≥ 1/n} ∈ σ(S). Since second countability assumption.
1Xn ≤ nχ,

4. For all open subsets, U ⊂ X, Applying this result to B c shows there exists a closed set F @ X such that
Z B c ⊂ F c and
µ(U ) = sup{ f dµ : f ≺ X} = sup{Iµ (f ) : f ≺ X}. (32.23) µ(B \ F ) = µ(F c \ B c ) < ε.
X
So we have produced F ⊂ B ⊂ U such that µ(U \F ) = µ(U \B)+µ(B\F ) <
5. For all compact subsets K ⊂ X, 2ε. If µ(B) < ∞, using B \ (Kn ∩ F ) ↑ B \ F as n → ∞, we may choose n
sufficiently large so that µ(B \ (Kn ∩ F )) < ε. Hence we may replace F by
µ(K) = inf{Iµ (f ) : 1K ≤ f ≺ X}. (32.24) the compact set F ∩ Kn if necessary.
2. Choose Fn ⊂ B ⊂ Un such Fn is closed, Un is open and µ(Un \ Fn ) < 1/n.
6. If kIµ k denotes the dual norm on Cc (X, R)∗ , then kIµ k = µ(X). In partic- Let B = ∪n Fn ∈ Fσ and C := ∩Un ∈ τδ . Then A ⊂ B ⊂ C and
ular Iµ is bounded iff µ(X) < ∞.
7. Cc (X, R) is dense in Lp (µ; R) for all 1 ≤ p < ∞. 1
µ(C \ A) ≤ µ(Fn \ Un ) < → 0 as n → ∞.
n
Proof. First notice that Iµ is a positive linear functional on S := Cc (X, R)
for all µ ∈ M+ and S is a lattice such that 1 ∧ f ∈ S for all f ∈ S. Proposition 3. From Item 1, one easily concludes that
32.7shows that any positive linear functional, I, on S := Cc (X, R) is a Daniell
integral on S. By Lemma 14.23, there exists compact sets Kn ⊂ X such that µ(B) = inf {µ(U ) : B ⊂ U ⊂o X}
Kn ↑ X. By Urysohn’s
P∞ lemma, there exists φn ≺ X such that φn = 1 on Kn . for all B ∈ BX and
Since φn ∈ S+ and n=1 φn > 0 on X it follows from Remark 32.45 that there
exists χ ∈ S↑ such that χ > 0 on X and I(χ) < ∞. So the hypothesis of the µ(B) = sup {µ(K) : K @@ B}
Daniell – Stone Theorem 32.44 hold and hence there exists a unique measure µ
on σ(S) =BX (Lemma 18.57) such that I = Iµ . Hence the map µ → Iµ taking for all B ∈ BX with µ(B) < ∞. So now suppose B ∈ BX and µ(B) = ∞.
M+ to positive linear functionals on Cc (X, R) is bijective. We will now prove Using the notation at the end of the proof of Item 1., we have µ(F ) = ∞ and
the remaining seven assertions of the theorem. µ(F ∩ Kn ) ↑ ∞ as n → ∞. This shows sup {µ(K) : K @@ B} = ∞ = µ(B)
as desired.
1. Suppose ε > 0 and B ∈ BX satisfies µ(B) < ∞. Then 1B ∈ L1 (µ) so there
4. For U ⊂o X, let
exists functions fn ∈ Cc (X, R) such that fn ↑ f, 1B ≤ f, and
ν(U ) := sup{Iµ (f ) : f ≺ U }.
Z
f dµ = I(f ) < µ(B) + ε. (32.25) It is evident that ν(U ) ≤ µ(U ) because f ≺ U implies f ≤ 1U . Let K be a
X compact subset of U. By Urysohn’s Lemma 15.8, there exists f ≺ U such
that f = 1 on K. Therefore,
Let α ∈ (0, 1) and Ua := {f > α} ∪∞ n=1 {fn > α} ∈ τ. Since R1B ≤ f,
B ⊂ {f ≥ 1} ⊂ Uα and by Chebyshev’s inequality, µ(Uα ) ≤ α−1 X f dµ =
Z
α−1 I(f ). Combining this estimate with Eq. (32.25) shows µ(Uα \ B) = µ(K) ≤ f dµ ≤ ν(U ) (32.26)
X
µ(Uα )−µ(B) < ε for α sufficiently closet to 1. For general B ∈ BX , by what
we have just proved, there exists open sets Un ⊂ X such that B ∩ Kn ⊂ Un and we have
and µ(Un \ (B ∩ Kn )) < ε2−n for all n. Let U = ∪∞ n=1 Un , then B ⊂ U ∈ τ
and µ(K) ≤ ν(U ) ≤ µ(U ) for all U ⊂o X and K @@ U. (32.27)
∞
X By Item 3.,
µ(U \ B) = µ(∪∞
n=1 (Un \ B)) ≤ µ(Un \ B)
n=1 µ(U ) = sup{µ(K) : K @@ U } ≤ ν(U ) ≤ µ(U )
∞
X ∞
X
≤ µ(Un \ (B ∩ Kn )) ≤ ε2−n = ε. which shows that µ(U ) = ν(U ), i.e. Eq. (32.23) holds.
n=1 n=1

32.5 Riesz Representation Theorem 367
5. Now suppose K is a compact subset of X. From Eq. (32.26),
µ(K) ≤ inf{Iµ (f ) : 1K ≤ f ≺ X} ≤ µ(U ) Remark 32.48. We may give a direct proof of the fact that µ → Iµ is injective.
Indeed, suppose µ, ν ∈ M+ satisfy Iµ (f ) = Iν (f ) for all f ∈ Cc (X, R). By
for any open subset U such that K ⊂ U. Consequently by Eq. (32.21), Theorem 22.8, if A ∈ BX is a set such that µ(A) + ν(A) < ∞, there exists
fn ∈ Cc (X, R) such that fn → 1A in L1 (µ + ν). Since fn → 1A in L1 (µ) and
µ(K) ≤ inf{Iµ (f ) : 1K ≤ f ≺ X} ≤ inf{µ(U ) : K ⊂ U ⊂o X} = µ(K) L1 (ν),
µ(A) = lim Iµ (fn ) = lim Iν (fn ) = ν(A).
which proves Eq. (32.24). n→∞ n→∞
6. For f ∈ Cc (X, R), For general A ∈ BX , choose compact subsets Kn ⊂ X such that Kn ↑ X. Then
Z
|Iµ (f )| ≤ |f | dµ ≤ kf k∞ µ(supp(f )) ≤ kf k∞ µ(X) (32.28) µ(A) = lim µ(A ∩ Kn ) = lim ν(A ∩ Kn ) = ν(A)
n→∞ n→∞
X
which shows kIµ k ≤ µ(X). Let K @@ X and f ≺ X such that f = 1 on K. showing µ = ν. Therefore the map µ → Iµ is injective.
By Eq. (32.26), Theorem 32.49 (Lusin’s Theorem). Suppose (X, τ ) is a locally compact and
Z second countable Hausdorff space, BX is the Borel σ – algebra on X, and µ is a
µ(K) ≤ f dµ = Iµ (f ) ≤ kIµ k kf k∞ = kIµ k measure on (X, BX ) which is finite on compact sets of X. Also let ε > 0 be given.
X
If f : X → C is a measurable function such that µ(f 6= 0) < ∞, there exists a
and therefore, compact set K ⊂ {f 6= 0} such that f |K is continuous and µ({f 6= 0} \ K) < ε.
Moreover there exists φ ∈ Cc (X) such that µ(f 6= φ) < ε and if f is bounded
µ(X) = sup{µ(K) : K @@ X} ≤ kIµ k . the function φ may be chosen so that kφk∞ ≤ kf k∞ := supx∈X |f (x)| .
7. This has already been proved by two methods in Theorem 22.8 but we will Proof. Suppose first that f is bounded, in which case
give yet another proof here. When p = 1 and g ∈ L1 (µ; R), there exists,
R by Z
Eq. (32.20), h ∈ S↑ = Cc (X, R)↑ such that g ≤ h and kh − gk1 = X (h − |f | dµ ≤ kf kµ µ(f 6= 0) < ∞.
∞
g)dµ < ε. Let {hn }n=1 ⊂ S = Cc (X, R) be chosen so that hn ↑ h as n → ∞. X
Then by the dominated convergence theorem (notice that |hn | ≤ |h1 | + |h|),
By Theorem 22.8 or Item 7. of Theorem 32.47, there exists fn ∈ Cc (X) such
khn − gk1 → kh − gk1 < ε as n → ∞. Therefore for n large we have
that fn → f in L1 (µ) as n → ∞. By passing to a subsequence if necessary,
hn ∈ Cc (X, R) with khn − gk1 < ε. Since ε > 0 is arbitrary this shows,
we may assume kf − fn k1 < εn−1 −n −1

L1 (µ)
2 for all n and thus µ |f − fn | > n <
∞
Sf (A, µ) = L1 (µ). Now suppose p > 1, g ∈ Lp (µ; R) and {Kn }n=1 are −n ∞ −1

ε2 for all n. Let E := ∪n=1 |f − fn | > n , so that µ(E) < ε. On E c ,
as above. By the dominated convergence theorem, 1Kn (g ∧ n)∨(−n) → g in c
|f − fn | ≤ 1/n, i.e. fn → f uniformly on E and hence f |E c is continuous. Let
Lp (µ) as n → ∞, so it suffices to consider g ∈ Lp (µ; R) with supp(g) ⊂ Kn A := {f 6= 0}\E. By Theorem 32.47 (or see Exercises 33.4 and 33.5) there exists
and |g| ≤ n for some large n ∈ N. By Hölder’s inequality, such a g is a compact set K and open set V such that K ⊂ A ⊂ V such that µ(V \ K) < ε.
in L1 (µ). So if ε > 0, by the p = 1 case, there exists h ∈ S such that Notice that
kh − gk1 < ε. By replacing h by (h ∧ n) ∨ (−n) ∈ S, we may assume h is µ({f 6= 0} \ K) ≤ µ(A \ K) + µ(E) < 2ε.
bounded by n in which case
Z Z By the Tietze extension Theorem 15.9, there exists F ∈ C(X) such that f =
p p p−1 F |K . By Urysohn’s Lemma 15.8 there exists ψ ≺ V such that ψ = 1 on K.
kh − gkp = |h − g| dµ = |h − g| |h − g| dµ
X X So letting φ = ψF ∈ Cc (X), we have φ = f on K, kφk∞ ≤ kf k∞ and since
p−1
Z
p−1
{φ 6= f } ⊂ E ∪ (V \ K), µ(φ 6= f ) < 3ε. This proves the assertions in the
≤ (2n) |h − g| dµ < (2n) ε. theorem when f is bounded. Suppose that f : X → C is (possibly) unbounded.
X ∞
By Lemmas 18.57 and 14.23, there exists compact sets {KN }N =1 of X such that
Since ε > 0 was arbitrary, this shows S is dense in Lp (µ; R). KN ↑ X. Hence BN := KN ∩ {0 < |f | ≤ N } ↑ {f 6= 0} as N → ∞. Therefore if

ε > 0 is given there exists an N such that µ({f 6= 0} \ BN ) < ε. We now apply Solution to Exercise (32.2). Let Vn ⊂o X be sets such that Vn ↓ K as
what we have just proved to 1BN f to find a compact set K ⊂ {1BN f 6= 0} , n → ∞ and use P
Uryshon’s Lemma to find fn ∈ Cc (Vn , [0, 1]) such that fn = 1
∞
and open set V ⊂ X and φ ∈ Cc (V ) ⊂ Cc (X) such that µ(V \ K) < ε, on K. Let f = 2−n fn . Hence if x ∈ K c , then x ∈
P∞ n=1−n / Vn for some n and
µ({1BN f 6= 0} \ K) < ε and φ = f on K. The proof is now complete since hence f (x) < n=1 2 = 1.
{φ 6= f } ⊂ ({f 6= 0} \ BN ) ∪ ({1BN f 6= 0} \ K) ∪ (V \ K) This exercise shows that σ(compact Gδ0 s) ⊂ σ(Cc (X)). Indeed, if K is a
compact Gδ then by Exercise 32.2, there exist f ≺ X such that f = 1 on K
so that µ(φ 6= f ) < 3ε. and f < 1 on K c . Therefore 1K = limn→∞ f n is BX
0
– measurable. Therefore
0 0
To illustrate Theorem 32.49, suppose that X = (0, 1), µ = m is Lebesgue we have proved BX = σ(compact Gδ s).
measure and f = 1(0,1)∩Q . Then Lusin’s theorem asserts for any ε > 0 there
exists a compact set K ⊂ (0, 1) such that m((0, 1) \ K) < ε and f |K is con- Definition 32.51. Let (X, τ ) be a local compact topological space. We say that
∞
tinuous. To see this directly, let {rn }n=1 be an enumeration of the rationals in E ⊂ X is bounded if E ⊂ K for some compact set K and E is σ – bounded if
∞
(0, 1), E ⊂ ∪Kn for some sequence of compact sets {Kn }n=1 .
Jn = (rn − ε2−n , rn + ε2−n ) ∩ (0, 1) and W = ∪∞
n=1 Jn . 0
Lemma 32.52. If A ∈ BX , then either A or Ac is σ – bounded.
Then W is an open subset of X and µ(W ) < ε. Therefore Kn := [1/n, 1 − 1/n] \
W is a compact subset of X and m(X \ Kn ) ≤ n2 + µ(W ). Taking n sufficiently Proof. Let
large we have m(X \ Kn ) < ε and f |Kn ≡ 0 is continuous.
F := {A ⊂ X : either A or Ac is σ – bounded}.
Clearly X ∈ F and F is closed under complementation. Moreover if Ai ∈ F

32.6 The General Riesz Representation by Daniell then A = ∪i Ai ∈ F. Indeed, if each Ai is σ – bounded then A is σ – bounded
Integrals (Move Later?) and if some Acj is σ – bounded then
This section is rather a mess and is certainly not complete. Here is the upshot Ac = ∩i Aci ⊂ Acj
of what I understand at this point.
is σ – bounded. Therefore, F is a σ – algebra containing the compact Gδ0 s and
When using the Daniell integral to construct measures on locally compact 0
therefore BX ⊂ F.
Hausdorff spaces the natural answer is in terms of measures on the Baire σ – 0
Now the σ – algebra BX is called and may not necessarily be as large as the
algebra. To get the Rudin or Folland version of the theorem one has to extend
Borel σ – algebra. However if every open subset of X is σ – compact, then the
this measure to the Borel σ – algebra. Checking all of the details here seems to
Borel BX and and the Baire σ – algebras are the same. Indeed, if U ⊂o X and
be rather painful. Just as painful and giving the full proof in Rudin!! Argh.
Kn ↑ U with Kn being compact. There exists fn ≺ U such that fn = 1 on Kn .
0
Definition 32.50. Let X be a locally compact Hausdorff space. The Baire σ – Now f := limn→∞ fn = 1U showing U ∈ σ(Cc (X)) = BX .
0
algebra on X is BX := σ(Cc (X)). Lemma 32.53. In Halmos on p.221 it is shown that a compact Baire set is
+ − ±
Notice that if f ∈ Cc (X, R) then f = f − f with f ∈ Cc (X, R+ ). necessarily a compact Gδ .
0
Therefore BX is generated by sets of the form K := {f ≥ α} ⊂ supp(f ) with Proof. Let K be a compact Baire set and let KGδ denote the space of
α > 0. Notice that K is compact and K = ∩∞ n=1 {f > α − 1/n} showing K is a compact Gδ ’s. Recall in general that if D is some collection of subsets of a
compact Gδ . Thus we have shown BX 0
⊂ σ(compact Gδ0 s). For the converse we space X, then
will need the following exercise.
σ(D) = ∪ {σ(E) : E is a countable subset of D} .
Exercise 32.2. Suppose that X is a locally compact Hausdorff space and K ⊂
X is a compact Gδ then there exists f ∈ Cc (X, [0, 1]) such that f = 1 on K and This is because the right member of this equation is a σ – algebra. Therefore,
f < 1 on K c . ∞ ∞
there exist {Cn }n=1 ⊂ KGδ such that K ∈ σ ({Cn }n=1 ) . Let fn ∈ C(X, [0, 1])
such that Cn = {fn = 0}, see Exercise 32.2 above. Now define

32.7 Regularity Results 369
∞
X is compact, we may replace U by V if necessary and assume that U is bounded.
d(x, y) := 2−n |fn (x) − fn (y)| .
Let f ∈ C(X, [0, 1]) such that f = 0 on K and f = 1 on U c . Take U0 = {f <
n=1
1/2} and K0 = {f ≤ 1/2}. Then K @@ U0 ⊂ K0 ⊂ U, K0 is compact Gδ and
Then d would be a metric on X except for the fact that d(x, y) may be zero U0 is a σ – compact open set since U0 = ∪∞
n=3 {f ≤ 1/2 + 1/n}. 5. Let K @@ X,
even though x 6= y. Let X ∼ y iff d(x, y) = 0 iff fn (x) = fn (y) for all n It is and D be a countable dense subset of X. For all x ∈
/ K there exist disjoint open
easily seen that ∼ is an equivalence relation and Z := X/ ∼ with the induced sets Vx and Ux such that x ∈ Ux and K ⊂ Vx . (I don’t see how to finish this off
metric d¯ is a metric space. Also let π : X → Z be the canonical projection map. at the moment.)
Notice that if x ∈ Cn then y ∈ Cn for all x ∼ y, and therefore π −1 (π(Cn )) = Cn
for all n. In particular this shows that
∞
32.7 Regularity Results
K ∈ σ ({Cn }n=1 ) ⊂ π −1 (P(Z)) ,
Proposition 32.55. Let X be a compact Hausdorff space and µ be a Baire
i.e. K = π −1 (π(K)). Now π is continuous, since if x ∈ X and y ∈ measure on BX 0
. Then for each A ∈ BX 0
and ε > 0 there exists K ⊂ A ⊂ V
∩Nk=1 {|fk (y) − fk (x)| < ε} ⊂o X then where K is a compact Gδ and V is an open, Baire and σ -compact, such that
¯ µ(V \ K) < ε.
d(π(x), π(y)) = d(x, y) < ε + 2−N +1
R
Proof. Let I(f ) = X f dµ for f ∈ S := C(X), so that I is a Daniell integral
which can be made as small as we please. Hence π(K) is compact and hence
on C(X). Since 1 ∈ S, the measure µ from the Daniell – Stone construction
closed in Z. Let Wn := {z ∈ Z : d¯π(K) (z) < 1/n}, then Wn is open in Z and
theorem is the same as the measure µ. Hence for g ∈ L1 (µ), we have
Wn ↓ π(K) as n → ∞. Let Vn := π −1 (Wn ), open in X since π is continuous,
then Vn ↓ K as n → ∞.
Z
The following facts are taken from Halmos, section 50 starting on p. 216. sup {I(f ) : f ∈ S↓ with f ≤ h} = gdµ
X
Z
Theorem 32.54. 1. It K @@ X and K ⊂ U ∪ V with U, V ∈ τ, then K = = inf hdµ : h ∈ S↑ with g ≤ h .
K1 ∪ K2 with K1 @@ U and K2 @@ V.
2. If K @@ X and F @ X are disjoint, then there exists f ∈ C(X, [0, 1]) such 0
that f = 0 on K and f = 1 on F. Suppose ε > 0 and B ∈ BX are given. There exists hn ∈ S such that hn ↑ h,
3. If f is a real valued continuous function, then for all c ∈ R the sets {f ≥ c} , 1B ≤ h, and µ(h) < µ(B) + ε. The condition 1B ≤ h, implies 1B ≤ 1{h≥1} ≤ h
{f ≤ c} and {f = c} are closed Gδ . and hence
4. If K @@ U ⊂o X then there exists K @@ U0 ⊂ K0 ⊂ U such that K0 is a µ(B) ≤ µ(h ≥ 1) ≤ µ(h) < µ(B) + ε. (32.29)
compact Gδ and U0 is a σ – compact open set. Moreover, letting
5. If X is separable, then every compact subset of X is a Gδ . (I think the proof
of this point is wrong in Halmos!) Vm := ∪∞ ∞ ∞
n=1 {hn > 1 − 1/m} = ∪n=1 ∪k=1 {hn ≥ 1 − 1/m + 1/k}
Proof. 1. K \ U and K \ V are disjoint compact sets and hence there exists (a σ – compact, open Baire set) we have Vm ↓ {h ≥ 1} ⊃ B hence µ(Vm ) ↓
two disjoint open sets U 0 and V 0 such that µ(h ≥ 1) ≥ µ(B) as m → ∞. Combining this observation with Eq. (32.29), we
may choose m sufficiently large so that B ⊂ Vm and
K \ U ⊂ V 0 and K \ V ⊂ U 0 .
µ(Vm \ B) = µ(Vm ) − µ(B) < ε.
Let K1 := K \ V 0 ⊂ U and K2 = K \ U 0 ⊂ V. 2. Tietze extension theorem with
elementary proof in Halmos. 3. {f ≤ c} = ∩∞ n=1 {f < c + 1/n} with similar Hence there exists V ∈ τ such that B ⊂ V and µ(V \ B) < ε. Similarly, there
formula for the other cases. The converse has already been mentioned. 4. For exists f ∈ S↓ such that f ≤ 1B and µ(B) < µ(f ) + ε. We clearly may assume
each x ∈ K, let Vx be an open neighborhood of K such that V̄x @@ U, and set that f ≥ 0. Let fn ∈ S be chosen so that fn ↓ f as n → ∞. Since 0 ≤ f ≤ 1B
V = ∪x∈Λ Vx where Λ ⊂⊂ K is a finite set such that K ⊂ V. Since V̄ = ∪x∈Λ V̄x we have

0 ≤ f ≤ 1{f >0} ≤ 1B Proof. Let A be a σ – bounded set and Kn be compact Gδ ’s (which exist by
so that {f > 0} ⊂ B and µ(B) < µ(f > 0) + ε. For each m ∈ N, let Theorem 32.54) such that A ⊂ ∪Kn . By replacing Kn by ∪nk=1 Kk if necessary,
we may assume that Kn is increasing in n. By Proposition 32.55, there exists
Km := ∩∞ ∞ ∞
n=1 {fn ≥ 1/m} = ∩k=1 ∩n=1 {fn > 1/m − 1/k} , compact Gδ0 s, Cn , such that Cn ⊂ A∩Kn and µ(A∩Kn \Cn ) < ε2−n for all n. Let
a compact Gδ , then Km ↑ {f > 0} as m → ∞. Therefore for large m we will C N := ∪Nn=1 Cn , then C
N
is a compact Gδ , C N ⊂ A and µ(A ∩ KN \ CN ) < ε for
have µ(B) < µ(Km ) + ε, i.e. Km ⊂ B and µ(B \ Km ) < ε. all N. From this equation it follows that µ(A \ CN ) < ε for large N if µ(A) < ∞
and µ(CN ) → ∞ if µ(A) = ∞. In either case we conclude that ` ν(A) = µ(A).
Remark 32.56. The above proof does not in general work when X is a locally 0 ∞
Now let us show that ν is a measure on BX . Suppose A = n=1 An and
0
compact Hausdorff space and µ is a finite Baire measure on BX since it may Kn ⊂ An for each n with Kn being a compact Gδ . Then K N := ∪N n=1 Kn is also
happen that µ 6= µ+ , i.e. µ+ (X) might be infinite, see Example 32.57 below. a compact Gδ and since K N ⊂ A, it follows that
However, if µ+ (X) < ∞, then the above proof works in this context as well.
N
Example 32.57. Let X be an uncountable and τ = 2X be the discrete topology
X
ν(A) ≥ µ(K N ) = µ(Kn ).
on X. In this case K ⊂ X is compact iff K is a finite set. Since every set is n=1
0
open, K is necessarily a Gδ and hence a Baire set. So BX is the σ – algebra
0 0 PN
generated by the finite subsets of X. We may describe BX by A ∈ BX iff A is Since Kn ⊂ An are arbitrary, we learn that ν(A) ≥ n=1 ν(An ) for all N and
c 0
countable or A is countable. For A ∈ BX , let hence letting N → ∞ shows

0 if A is countable ∞
µ(A) = X
1 if Ac is uncountable ν(A) ≥ ν(An ).
n=1
0
To see that µ is a measure suppose that A is the disjoint union ofP{An } ⊂ BX .
∞
If An is countable for all n, then A is countable and µ(A) = 0 = n=1 µ(An ). We now wish to prove the converse inequality.
P∞ Owing to the above inequality,
If Acm is countable for some m, then Ai ⊂ Acm is countable for all i 6= m. it suffices not to consider the case where n=1 ν(An ) < ∞. Let K ⊂ A be a
P∞
Therefore, n=1 µ(An ) = 1, now Ac = ∩Acn ⊂ Acm is countable as well, so compact Gδ . Then
µ(A) = 1. Therefore µ is a measure. ∞ ∞ ∞
0
The measureR µ is clearly a finite Baire measure on BX which is non-regular.
X X X
µ(K) = µ(K ∩ An ) = ν(K ∩ An ) ≤ ν(An )
Letting I(f ) = X f dµ for all f ∈ S = Cc (X) – the functions with finite support, n=1 n=1 n=1
then I(f ) = 0 for all f. If B ⊂ X is a set such that B c is countable, there are P∞
no functions f ∈ (Cc (X))↑ such that 1B ≤ f. Therefore µ+ (B) = I ∗ (1B ) = ∞. and since K is arbitrary, it follows that ν(A) ≤ n=1 ν(An ). So ν is a measure.
0
That is Finally if A ∈ BX , then
0 if A is countable
µ+ (A) = sup {ν(K) : K ⊂ A and K is a compact Gδ }
∞ otherwise.
0
On the other hand, one easily sees that µ− (A) = 0 for all A ∈ BX . The measure = sup {µ(K) : K ⊂ A and K is a compact Gδ } = ν(A)
µ− represents I as well.
showing ν is regular.
Definition 32.58. A Baire measure µ on a locally compact Hausdorff space is
0
regular if for each A ∈ BX 0
, (BX – being the Baire σ – algebra) Corollary 32.60. Suppose that µ is a finite Baire measure on X such
µ(A) = sup {µ(K) : K ⊂ A and K is a compact Gδ } . µ(X) := sup {µ(K) : K ⊂ X and K is a compact Gδ } ,
Proposition 32.59. Let µ be a Baire measure on X and set then µ = ν, in particular µ is regular.
ν(A) := sup {µ(K) : K ⊂ A and K is a compact Gδ } .
Proof. The assumption asserts that µ(X) = ν(X). Since µ = ν on the π
Then ν(A) = µ(A) for any σ – bounded sets A and ν is a regular Baire measure – class consisting of the compact Gδ0 s, we may apply Theorem 19.55 to learn
on X. µ = ν.

32.7 Regularity Results 371
Proposition 32.61. Suppose that µ is a Baire measure on X, then for all A ∈ Theorem 32.63 (Riesz Representation Theorem). Let X be a locally com-
0 0
BX which is σ – bounded and ε > 0 there exists V ∈ τ ∩ BX such that A ⊂ V pact Hausdorff space. The map ν → Iν taking Radon measures on X to positive
and µ(V \ A) < ε. Moreover if µ is regular then linear functionals on Cc (X) is bijective. Moreover if I is a positive linear func-
0
tionals on Cc (X), then I = Iν where ν is the unique Radon measure ν such that
µ(A) = inf µ(V ) : A ⊂ V ∈ τ ∩ BX . (32.30) ν(U ) = sup{I(f ) : f ≺ U } for all U ⊂o X.
0
holds for all A ∈ BX . Proof. Given a positive linear functional on Cc (X), the Daniell - Stone
0
Proof. Suppose A is σ – bounded Baire set. Let Kn be compact Gδ ’s (which integral construction theorem gives the existence of a measure µ on BX :=
exist by Theorem 32.54) such that A ⊂ ∪Kn . By replacing Kn by ∪nk=1 Kk if σ(Cc (X)) (the Baire σ – algebra) such that
necessary, we may assume that Kn is increasing in n. By Proposition 32.55 Z
(applied to dµn := 1Un0 dµ with Un0 an open Baire set such that Kn ⊂ Un0 and f dµ = I(f ) for all f ∈ Cc (X)
Un0 ⊂ Cn where Cn is a compact Baire set, see Theorem 32.54), there exists open X
Baire sets Vn of X such that A∩Kn ⊂ Vn and µ(Vn \A∩Kn ) < ε2−n for all n. Let 1
and for g ∈ L (µ),
V = ∪∞ 0
n=1 Vn ∈ τ ∩ BX , A ⊂ V and µ(V \ A)< ε. Now suppose that µ is regular
0 0
and A ∈ BX . If µ(A) = ∞ then clearly inf µ(V ) : A ⊂ V ∈ τ ∩ BX = ∞. So
Z
we will now assume that µ(A) < ∞. By inner regularity, there exists compact sup {I(f ) : S↓ 3 f ≤ g} = gdµ = inf {I(h) : g ≤ h ∈ S↑ }
X
Gδ0 s, Kn , such that Kn ↑, Kn ⊂ A for all n and µ(A \ Kn ) ↓ 0 as n → ∞.
Letting B = ∪Kn ⊂ A, then B is a σ – bounded set, µ(A \ B) = 0. Since B is with S := Cc (X, R). Suppose that K is a compact subset of X and E ⊂ K is a
σ – bounded there exists an open Baire V such that B ⊂ V and µ(V \ B) is a Baire set. Let f ≺ X be a function such that f = 1 on K, then 1E ≤ f implies
small as we please. These remarks reduce the problem to considering the truth µ(E) = I(1E ) ≤ I(f ) < ∞. Therefore any bounded (i.e. subset of a compact
of the proposition for the null set A \ B. So we now assume that µ(A) = 0. If set) Baire set E has finite measure. Suppose that K is a compact Baire set, i.e.
A is σ – bounded we are done by the first part of the proposition, so we will a compact Gδ , and f is as in Exercise 32.2, then
now assume that A is not σ – bounded. By Lemma 32.52, it follows that Ac is Z
σ – bounded. (I am a little stuck here, so assume for now that µ(X) < ∞. in µ(K) ≤ f n dµ = I(f n ) < ∞
which case we do not use the fact that Ac can be assumed to be σ – bounded.)
If µ(X) < ∞ and ε > 0 is given, by inner regularity there exists a compact showing µ is finite on compact Baire sets and by the dominated convergence
Baire subset K ⊂ Ac such that theorem that
ε > µ(Ac \ K) = µ(K c \ A) µ(K) = lim I(f n )
n→∞
0
and since A ⊂ K c is an open, Baire set the proof is finished when µ is a finite showing µ is uniquely determined on compact Baire sets. Suppose that A ∈ BX
∗
measure. and µ(A) = I (1A ) < ∞. Given ε > 0, there exists f ∈ S↑ such that 1A ≤ f
and µ(f ) < µ(A) + ε. Let fn ∈ Cc (X) such that fn ↑ f, then 1A ≤ 1{f ≥1} ≤ f
Example 32.62. 1) Suppose that X = R with the standard topology and µ is which shows
counting measure on X. Then clearly µ is not finite on all compact sets, so µ is Z
not K-finite measure. 2) Let X = R and τ = τd = 2X be the discrete topology µ(A) ≤ µ(f ≥ 1) ≤ f dµ = I(f ) < µ(A) + ε.
on X. Now let µ(A) = 0 if A is countable and µ(A) = ∞ otherwise. Then
µ(K) = 0 < ∞ if K is τd – compact yet µ is not inner regular on open sets, i.e.
Let Vm := ∪∞n=1 {fn > 1 − 1/m} , then Vm is open and Vm ↓ {f ≥ 1} as m → ∞.
all sets. So again µ is not Radon. Moreover, the functional
Z Notice that
Iµ (f ) = f dµ = 0 for all f ∈ Cc (X). µ(Vm ) = lim µ (fn > 1 − 1/m) ≤ µ (f > 1 − 1/m)
X n→∞
This shows that with out the restriction that µ is Radon in Example 28.17, the 1 1
≤ µ(f ) < (µ(A) + ε)
correspondence µ → Iµ is not injective. 1 − 1/m 1 − 1/m

0
showing µ(Vm ) < µ(A) + ε for all m large enough. Therefore if A ∈ BX and 1. For all ε > 0 and B ∈ M there exists an open set V ∈ τ and a closed set
µ(A) < ∞, there exists a Baire open set, V, such A ⊂ V and µ(V \A) is as small F such that F ⊂ B ⊂ V and µ(V \ F ) ≤ ε.
0
as we please. Suppose that A ∈ BX is a σ – bounded Baire set, then using Item 2. For all B ∈ M, there exists A ∈ Fσ and C ∈ Gδ such that A ⊂ B ⊂ C and
4. of Theorem 32.54 there exists compact Gδ , Kn , such that A ⊂ ∪Kn . Hence µ(C \ A) = 0. Here Fσ denotes the collection of subsets of X which may be
there exists Vn open Baire sets such that Kn ∩A ⊂ Vn and µ(Vn \Kn ∩A) < ε2−n written as a countable union of closed sets and Gδ = τδ is the collection of
for all n. Now let V := ∪Vn , an open Baire set, then A ⊂ V and µ(V \ A) < ε. subsets of X which may be written as a countable intersection of open sets.
Hence we have shown if A is σ – bounded then 3. The space BCf (X) of bounded continuous functions on X such that µ(f 6=
0) < ∞ is dense in Lp (µ).
µ(A) = inf {µ(V ) : A ⊂ V ⊂o X and V is Baire.} R
Proof. Let S := BCf (X), I(f ) := X f dµ for f ∈ S and Xn ∈ τ be chosen
Again let Aand Kn be as above. Replacing Kn by ∪nk=1 Kk we may also assume
so that µ(Xn ) < ∞ and Xn ↑ X as n → ∞. Then 1 ∧ f ∈ S for all f ∈ S
that Kn ↑ as n ↑ . Then Kn ∩ A is a bounded Baire set. Let Fn be a compact
and if φn = 1 ∧ ndXnc ∈ S+ , then φn ↑ 1 as n → ∞ and so by Remark 32.45
Gδ such that Kn ∩ A ⊂ Fn and choose σ – compact open set Vn such that
there exists χ ∈ S↑ such that χ > 0 on X and I(χ) < ∞. Similarly if V ∈ τ,
Fn \ Kn ∩ A ⊂ Vn and µ(Vn \ (Fn \ Kn ∩ A)) < ε2−n . ....... In the end the
the function gn := 1 ∧ nd(Xn ∩V )c ∈ S and gn → 1V as n → ∞ showing
desired measure ν should be defined by
σ(S) =BX . If fn ∈ S+ and fn ↓ 0 as n → ∞, it follows by the dominated
ν(U ) = sup{I(f ) : f ≺ U } for all U ⊂o X convergence theorem that I(fn ) ↓ 0 as n → ∞. So the hypothesis of the Daniell
– Stone Theorem 32.44 hold and hence µ is the unique measure on BX such
and for general A ∈ BX we set that I = Iµ and for B ∈ BX and
ν(A) := inf {ν(U ) : A ⊂ U ⊂o X} . µ(B) = I ∗ (1B ) = inf {I(f ) : f ∈ S↑ with 1B ≤ f }
Z
Let us note that if f ≺ U and K = supp(f ), then there exists K ⊂ U0 ⊂ K0 ⊂ U
= inf f dµ : f ∈ S↑ with 1B ≤ f .
as in Theorem 32.54. Therefore, f ≤ 1K0 and hence I(f ) ≤ µ(K0 ) which shows X
that
ν(U ) ≤ sup {µ(K0 ) : K0 ⊂ U and K0 is a compact Gδ } . Suppose ε > 0 and B ∈ BX are given. There exists fn ∈ BCf (X) such that
fn ↑ f, 1B ≤ f, and µ(f ) < µ(B) + ε. The condition 1B ≤ f, implies 1B ≤
The converse inequality is easily proved by letting g ≺ U such that g = 1 on 1{f ≥1} ≤ f and hence that
K0 . Then µ(K0 ) ≤ I(g) ≤ ν(U ) and hence
µ(B) ≤ µ(f ≥ 1) ≤ µ(f ) < µ(B) + ε. (32.31)
ν(U ) = sup {µ(K0 ) : K0 ⊂ U and K0 is a compact Gδ } .
Let us note that ν is sub-additive on open sets ,see p. 314 of Royden. Let Moreover, letting Vm := ∪∞
n=1 {fn ≥ 1 − 1/m} ∈ τd , we have Vm ↓ {f ≥ 1} ⊃ B
hence µ(Vm ) ↓ µ(f ≥ 1) ≥ µ(B) as m → ∞. Combining this observation with
ν ∗ (A) := inf {ν(U ) : A ⊂ U ⊂o X} Eq. (32.31), we may choose m sufficiently large so that B ⊂ Vm and
Then ν ∗ is an outer measure as well I think and N := {A ⊂ X : ν ∗ (A) = 0} is µ(Vm \ B) = µ(Vm ) − µ(B) < ε.
closed under countable unions. Moreover if E is Baire measurable and E ∈ N ,
then there exists O open ν(O) < ε and E ⊂ O. Hence for all compact Gδ , K ⊂ O, Hence there exists V ∈ τ such that B ⊂ V and µ(V \ B) < ε. Applying this
µ(K) < ε. Royden uses assumed regularity here to show that ν(E) = 0. I don’t result to B c shows there exists F @ X such that B c ⊂ F c and
see how to get this assume regularity at this point.
µ(B \ F ) = µ(F c \ B c ) < ε.
32.8 Metric space regularity results resisted So we have produced F ⊂ B ⊂ V such that µ(V \F ) = µ(V \B)+µ(B \F ) < 2ε.
The second assertion is an easy consequence of the first and the third follows
Proposition 32.64. Let (X, d) be a metric space and µ be a measure on M = in similar manner to any of the proofs of Item 7. in Theorem 32.47.
BX which is σ – finite on τ := τd .

32.9 General Product Measures 373
32.9 General Product Measures where x × y ∈ XΛn is defined by (x × y) (α) = x(α) if α ∈ Λk and (x × y) (α) =
y(α) for α ∈ Λn \ Λk . By convention we set Fnn = Fn . Since fn is decreasing it
k
In this section we drop the topological assumptions used in the last section. follows that Fn+1 ≤ Fnk for all k and n ≥ k and therefore F k := limn→∞ Fnk
exists. By Fubini’s theorem,
Theorem 32.65. Let {(Xα , Mα , µα )}α∈A be Q a collection of probability spaces, Z
that is µα (Xa ) = 1 for all α ∈ A. Let X := Xα , M = σ(πα : α ∈ A) and
α∈A Fnk (x) = Fnk+1 (x × y)dµΛk+1 \Λk (y) when k + 1 ≤ n
Q XΛn \Λk
for Λ ⊂⊂ A let XΛ := Qα∈Λ Xα and πΛ : X → XΛ be the projection map
πΛ (x) = x|Λ and µΛ := α∈Λ µα be product measure on MΛ := ⊗α∈Λ Mα . and hence letting n → ∞ in this equation shows
Then there exists a unique measure µ on M such that (πΛ )∗ µ = µΛ for all
Λ ⊂⊂ A, i.e. if f : XΛ → R is a bounded measurable function then
Z
F k (x) = F k+1 (x × y)dµΛk+1 \Λk (y) (32.33)
XΛn \Λk
Z Z
f (πΛ (x))dµ(x) = f (y)dµΛ (y). (32.32)
X
XΛ
for all k. Now
Z Z
Proof. Let S denote the collection of functions f : X → R such that there F 1 (x)dµΛ1 (x) = lim Fn1 (x)dµΛ1 (x) = lim I(fn ) = ε > 0
n→∞ n→∞
exists Λ ⊂⊂ A and a bounded measurable function R F : XΛ → R such that XΛ1 XΛ1
f = F ◦ πΛ . For f = F ◦ πΛ ∈ S, let I(f ) = XΛ F dµΛ . Let us verify that

so there exists
I is well defined. Suppose that f may also be expressed as f = G ◦ πΓ with
x1 ∈ XΛ1 such that F 1 (x1 ) ≥ ε.
Γ ⊂⊂ A and G : XΓ → R bounded and measurable. By replacing Γ by Γ ∪ Λ
if necessary, we may assume that Λ ⊂ Γ. Making use of Fubini’s theorem we From Eq. (32.33) with k = 1 and x = x1 it follows that
learn Z
Z Z ε≤ F 2 (x1 × y)dµΛ2 \Λ1 (y)
G(z) dµΓ (z) = F ◦ πΛ (x) dµΛ (x)dµΓ \Λ (y) XΛ2 \Λ1
XΓ XΛ ×XΓ \Λ
Z Z and hence there exists
= F ◦ πΛ (x) dµΛ (x) · dµΓ \Λ (y)
XΛ
Z
XΓ \Λ
x2 ∈ XΛ2 \Λ1 such that F 2 (x1 × x2 ) ≥ ε.

= µΓ \Λ XΓ \Λ · F ◦ πΛ (x) dµΛ (x)
XΛ Working this way inductively using Eq. (32.33) implies there exists
Z
= F ◦ πΛ (x) dµΛ (x), xi ∈ XΛi \Λi−1 such that F n (x1 × x2 × · · · × xn ) ≥ ε
XΛ
for all n. Now Fkn ≥ F n for all k ≤ n and in particular for k = n, thus
wherein we have used the fact that µΛ (XΛ ) = 1 for all Λ ⊂⊂ A since µα (Xα ) =
1 for all α ∈ A. It is now easy to check that I is a positive linear functional Fn (x1 × x2 × · · · × xn ) = Fnn (x1 × x2 × · · · × xn )
on the lattice S. We will now show that I is a Daniel integral. Suppose that
≥ F n (x1 × x2 × · · · × xn ) ≥ ε (32.34)
fn ∈ S+ is a decreasing sequence such that inf n I(fn ) = ε > 0. We need to show
f := limn→∞ fn is not identically zero. As in the proof that I is well defined, for all n. Let x ∈ X be any point such that
there exists Λn ⊂⊂ A and bounded measurable functions Fn : XΛn → [0, ∞)
such that Λn is increasing in n and fn = Fn ◦ πΛn for each n. For k ≤ n, let πΛn (x) = x1 × x2 × · · · × xn
Fnk : XΛk → [0, ∞) be the bounded measurable function
Z for all n. From Eq. (32.34) it follows that
Fnk (x) = Fn (x × y)dµΛn \Λk (y)
XΛn \Λk fn (x) = Fn ◦ πΛn (x) = Fn (x1 × x2 × · · · × xn ) ≥ ε

for all n and therefore f (x) := limn→∞ fn (x) ≥ ε showing f is not zero. There- I(g1 ) + I(g2 ) = I(g1 + g2 ) ≤ I+ (f1 + f2 ).
fore, I is a Daniel integral and there exists by Theorem 32.47 a unique measure
µ on (X, σ(S) = M) such that Therefore,
I+ (f1 ) + I+ (f2 ) = sup{I(g1 ) + I(g2 ) : S+ 3 gi ≤ fi } ≤ I+ (f1 + f2 ). (32.35)

Z
I(f ) = f dµ for all f ∈ S.
X
For the opposite inequality, suppose g ∈ S+ and g ≤ f1 + f2 . Let g1 = f1 ∧ g,
Taking f = 1A ◦ πΛ in this equation implies then

−1
µΛ (A) = I(f ) = µ ◦ πΛ (A) 0 if g ≤ f1
0 ≤ g2 := g − g1 = g − f1 ∧ g =
g − f1 if g ≥ f1
and the result is proved.

0 if g ≤ f1
≤ ≤ f2 .
f1 + f2 − f1 if g ≥ f1
Remark 32.66. (Notion of kernel needs more explanation here.) The above the-
orem may be Jazzed up as follows. Let {(Xα , Mα )}α∈A be a collection of mea- Since g = g1 + g2 with S+ 3 gi ≤ fi ,
surable spaces. Suppose for each pair Λ ⊂ Γ ⊂⊂ A there is a kernel µΛ,Γ (x, dy)
for x ∈ XΛ and y ∈ XΓ \Λ such that if Λ ⊂ Γ ⊂ K ⊂⊂ A then I(g) = I(g1 ) + I(g2 ) ≤ I+ (f1 ) + I+ (f2 )
µΛ,K (x, dy × dz) = µΛ,Γ (x, dy)µΓ,K (x × y, dz). and since S+ 3 g ≤ f1 + f2 was arbitrary, we may conclude
Then there exists a unique measure µ on M such that I+ (f1 + f2 ) ≤ I+ (f1 ) + I+ (f2 ). (32.36)
Z Z
Combining Eqs. (32.35) and (32.36) shows that
f (πΛ (x))dµ(x) = f (y)dµ∅,Λ (y)
X
XΛ I+ (f1 + f2 ) = I+ (f1 ) + I+ (f2 ) for all fi ∈ S+ . (32.37)
for all Λ ⊂⊂ A and f : XΛ → R bounded and measurable. To prove this We now extend I+ to S by defining, for f ∈ S,
assertion, just use the proof of Theorem 32.65 replacing µΓ \Λ (dy) by µΛ,Γ (x, dy)
everywhere in the proof. I+ (f ) = I+ (f+ ) − I+ (f− )
where f+ = f ∨0 and f− = − (f ∧ 0) = (−f )∨0. (Notice that f = f+ −f− .) We

32.10 Daniel Integral approach to dual spaces will now shows that I+ is linear. If c ≥ 0, we may use (cf )± = cf± to conclude
that
BRUCE: compare and consolidate with Section 28.2.2. I+ (cf ) = I+ (cf+ ) − I+ (cf− ) = cI+ (f+ ) − cI+ (f− ) = cI+ (f ).
Similarly, using (−f )± = f∓ it follows that I+ (−f ) = I+ (f− ) − I+ (f+ ) =
Proposition 32.67. Let S be a vector lattice of bounded real functions on a
−I+ (f ). Therefore we have shown
set X. We equip S with the sup-norm topology and suppose I ∈ S∗ . Then there
exists I± ∈ S∗ which are positive such that then I = I+ − I− . I+ (cf ) = cI+ (f ) for all c ∈ R and f ∈ S.
Proof. For f ∈ S+ , let If f = u − v with u, v ∈ S+ then
I+ (f ) := sup I(g) : g ∈ S+ and g ≤ f .

v + f+ = u + f− ∈ S+
One easily sees that |I+ (f )| ≤ kIk kf k for all f ∈ S+ and I+ (cf ) = cI+ (f ) for and so by Eq. (32.37), I+ (v) + I+ (f+ ) = I+ (u) + I+ (f− ) or equivalently
all f ∈ S+ and c > 0. Let f1 , f2 ∈ S+ . Then for any gi ∈ S+ such that gi ≤ fi ,
we have S+ 3 g1 + g2 ≤ f1 + f2 and hence I+ (f ) = I+ (f+ ) − I+ (f− ) = I+ (u) − I+ (v). (32.38)

32.10 Daniel Integral approach to dual spaces 375
Now if f, g ∈ S, then To finish the proof we need only prove the reverse inequality. To this end let
ε > 0 and choose K @@ P ∩ A ⊂ U ⊂o X such that |µ| (U \ K) < ε. Let
I+ (f + g) = I+ (f+ + g+ − (f− + g− )) f, g ∈ Cc (U, [0, 1]) with f ≤ g, then
= I+ (f+ + g+ ) − I+ (f− + g− )
I(f ) = µ(f ) = µ(f : K) + µ(f : U \ K) ≤ µ(g : K) + O (ε)
= I+ (f+ ) + I+ (g+ ) − I+ (f− ) − I+ (g− )
≤ µ(K) + O (ε) ≤ µ(P ∩ A) + O (ε) .
= I+ (f ) + I+ (g),
wherein the second equality we used Eq. (32.38). The last two paragraphs show Taking the supremum over all such f ≤ g, we learn that I+ (g) ≤ µ(P ∩A)+O (ε)
I+ : S → R is linear. Moreover, and then taking the supremum over all such g shows that
|I+ (f )| = |I+ (f+ ) − I+ (f− )| ≤ max (|I+ (f+ )| , |I+ (f− )|) µ+ (U ) ≤ µ(P ∩ A) + O (ε) .
≤ kIk max (kf+ k , kf− k) = kIk kf k Taking the infimum over all U ⊂o X such that P ∩ A ⊂ U shows that
which shows that kI+ k ≤ kIk . That is I+ is a bounded positive linear functional µ+ (P ∩ A) ≤ µ(P ∩ A) + O (ε) (32.40)
on S. Let I− = I+ −I ∈ S∗ . Then by definition of I+ (f ), I− (f ) = I+ (f )−I(f ) ≥
0 for all S 3 f ≥ 0. Therefore I = I+ − I− with I± being positive linear From Eqs. (32.39) and (32.40) it follows that µ(P ∩ A) = µ+ (P ∩ A). Since
functionals on S.
I− (f ) = sup I(g) − I(f ) = sup I(g − f ) = sup −I(f − g) = sup −I(h)
Corollary 32.68. Suppose X is a second countable locally compact Hausdorff 0≤g≤f 0≤g≤f 0≤g≤f 0≤h≤f
∗
space and I ∈ C0 (X, R) , then there exists
R µ = µ+ − µ− where µ is a finite
signed measure on BR such that I(f ) = R f dµ for all f ∈ C0 (X, R).R Similarly the same argument applied to −I shows that
∗
if I ∈ C0 (X, C) there exists a complex measure µ such that I(f ) = R f dµ for
all f ∈ C0 (X, C). TODO Add in the isometry statement here. −µ(P c ∩ A) = µ− (P c ∩ A).
Proof. Let I = I+ −I− be the decomposition given as above. Then we know Since
there exists finite measure µ± such that
µ(A) = µ(P ∩ A) + µ(P c ∩ A) = µ+ (P ∩ A) − µ− (P c ∩ A) and
Z
I± (f ) = f dµ± for all f ∈ C0 (X, R). µ(A) = µ+ (A) − µ− (A)
X
R it follows that
and therefore I(f ) = X f dµ for all f ∈ C0R(X, R) where µ = µ+ −µ− . Moreover µ+ (A \ P ) = µ− (A \ P c ) = µ− (A ∩ P ).
the measure µ is unique. Indeed if I(f ) = X f dµ
R for some finite signed measure
µ, then the next result shows that I± (f ) = X f dµ± where µ± is the Hahn Taking A = P then shows that µ− (P ) = 0 and taking A = P c shows that
decomposition of µ. Now the measures µ± are uniquely determined by I± . The µ+ (P c ) = 0 and hence
complex case is a consequence of applying the real case just proved to Re I and
µ(P ∩ A) = µ+ (P ∩ A) = µ+ (A) and
Im I.
−µ(P c ∩ A) = µ− (P c ∩ A) = µ− (A)
Proposition 32.69. Suppose that µ is a signed Radon measure and I = Iµ .
Let µ+ and µ− be the Radon measures associated to I± , then µ = µ+ − µ− is as was to be proved.
the Jordan decomposition of µ.
Proof. Let X = P ∪ P c where P is a positive set for µ and P c is a negative

set. Then for A ∈ BX ,
µ(P ∩ A) = µ+ (P ∩ A) − µ− (P ∩ A) ≤ µ+ (P ∩ A) ≤ µ+ (A). (32.39)

33
Class Arguments
33.1 Monotone Class and π – λ Theorems argument shows that C(C) is closed under decreasing limits. Thus we have
shown that C(C) is a monotone class for all C ∈ C. If A ∈ A ⊂ C, then
Definition 33.1. Let C ⊂ 2X be a collection of sets. A ∩ B, A ∩ B c , B ∩ Ac ∈ A ⊂ C for all B ∈ A and hence it follows that
A ⊂ C(A) ⊂ C. Since C is the smallest monotone class containing A and C(A) is
1. C is a monotone class if it is closed under countable increasing unions
a monotone class containing A, we conclude that C(A) = C for any A ∈ A. Let
and countable decreasing intersections,
B ∈ C and notice that A ∈ C(B) happens iff B ∈ C(A). This observation and
2. C is a π – class if it is closed under finite intersections and
the fact that C(A) = C for all A ∈ A implies A ⊂ C(B) ⊂ C for all B ∈ C. Again
3. C is a λ–class if C satisfies the following properties:
since C is the smallest monotone class containing A and C(B) is a monotone
a) X ∈ C class we conclude that C(B) = C for all B ∈ C. That is to say, if A, B ∈ C then
b) If A, B ∈ C and A ∩ B = ∅, then A ∪ B ∈ C. (Closed under disjoint A ∈ C = C(B) and hence A ∩ B, A ∩ B c , Ac ∩ B ∈ C. So C is closed under
unions.) complements (since X ∈ A ⊂ C) and finite intersections and increasing unions
c) If A, B ∈ C and A ⊃ B, then A \ B ∈ C. (Closed under proper differ- from which it easily follows that C is a σ – algebra.
ences.) Let E ⊂ 2X×Y be given by
d) If An ∈ C and An ↑ A, then A ∈ C. (Closed under countable increasing
unions.) E = M × N = {A × B : A ∈ M, B ∈ N }
4. C is a λ0 – class if C satisfies conditions a) – c) but not necessarily d).
and recall from Exercise 18.2 that E is an elementary family. Hence the algebra
Remark 33.2. Notice that every λ – class is also a monotone class. A = A(E) generated by E consists of sets which may be written as disjoint
unions of sets from E.
(The reader wishing to shortcut this section may jump to Theorem 33.5
where he/she should then only read the second proof.) Lemma 33.4. If D is a λ0 – class which contains a π – class, C, then D contains
A (C) – the algebra generated by C.
Lemma 33.3 (Monotone Class Theorem). Suppose A ⊂ 2X is an algebra
Proof. We will give two proofs of this lemma. The first proof is “construc-
and C is the smallest monotone class containing A. Then C = σ(A).
tive” and makes use of Proposition 18.6 which tells how to construct A(C) from
Proof. For C ∈ C let C. The key to the first proof is the following claim which will be proved by
induction.
C(C) = {B ∈ C : C ∩ B, C ∩ B c , B ∩ C c ∈ C}, Claim. Let C˜0 = C and Cñ denote the collection of subsets of X of the form
then C(C) is a monotone class. Indeed, if Bn ∈ C(C) and Bn ↑ B, then Bnc ↓ B c Ac1 ∩ · · · ∩ Acn ∩ B = B \ A1 \ A2 \ · · · \ An . (33.1)
and so
with Ai ∈ C and B ∈ C ∪ {X} . Then Cñ ⊂ D for all n, i.e. C˜ := ∪∞ ˜
n=0 Cn ⊂ D.
C 3 C ∩ Bn ↑ C ∩ B By assumption C˜0 ⊂ D and when n = 1,
C 3 C ∩ Bnc ↓ C ∩ B c and B \ A1 = B \ (A1 ∩ B) ∈ D
C 3 Bn ∩ C c ↑ B ∩ C c .
when A1 , B ∈ C ⊂ D since A1 ∩ B ∈ C ⊂ D. Therefore, C˜1 ⊂ D. For the
c c
Since C is a monotone class, it follows that C ∩ B, C ∩ B , B ∩ C ∈ C, i.e. induction step, let B ∈ C ∪ {X} and Ai ∈ C ∪ {X} and let En denote the set in
B ∈ C(C). This shows that C(C) is closed under increasing limits and a similar Eq. (33.1) We now assume Cñ ⊂ D and wish to show En+1 ∈ D, where
378 33 Class Arguments
En+1 = En \ An+1 = En \ (An+1 ∩ En ). Then C ⊂ D1 and D1 is also a λ–class because as we now check.
` a) X ∈ D1 . b)
If A, B ∈ D1 with A ∩ B = ∅, then (A ∪ B) ∩ C = (A ∩ C) (B ∩ C) ∈ D for
Because all C ∈ C. c) If A, B ∈ D1 with B ⊂ A, then (A \ B) ∩ C = A ∩ C \ (B ∩ C) ∈ D
An+1 ∩ En = Ac1 ∩ · · · ∩ Acn ∩ (B ∩ An+1 ) ∈ Cñ ⊂ D for all C ∈ C. d) If An ∈ D1 and An ↑ A as n → ∞, then An ∩ C ∈ D for all
and (An+1 ∩ En ) ⊂ En ∈ Cñ ⊂ D, we have En+1 ∈ D as well. This finishes the C ∈ D and hence An ∩ C ↑ A ∩ C ∈ D. Since C ⊂ D1 ⊂ D and D is the smallest
proof of the claim. λ – class containing C it follows that D1 = D. From this we conclude that if
Notice that C˜ is still a multiplicative class and from Proposition 18.6 (using A ∈ D and B ∈ C then A ∩ B ∈ D.
the fact that C is a multiplicative class), A(C) consists of finite unions of elements Let
˜ By applying the claim to C,
from C. ˜ Ac ∩ · · · ∩ Acn ∈ D for all Ai ∈ C˜ and hence D2 := {A ∈ D : A ∩ D ∈ D ∀ D ∈ D}.
1
c Then D2 is a λ–class (as you should check) which, by the above paragraph,
A1 ∪ · · · ∪ An = (Ac1 ∩ · · · ∩ Acn ) ∈ D.
contains C. As above this implies that D = D2 , i.e. we have shown that D is
Thus we have shown A(C) ⊂ D which completes the proof. closed under finite intersections. Since λ – classes are closed under complemen-
Second Proof. With out loss of generality, we may assume that D is the tation, D is an algebra which is closed under increasing unions and hence is
smallest λ0 – class containing C for if not just replace D by the intersection of closed under arbitrary countable unions, i.e. D is a σ – algebra. Since C ⊂ D
all λ0 – classes containing C. Let we must have σ(C) ⊂ D and in fact σ(C) = D.
D1 := {A ∈ D : A ∩ C ∈ D ∀ C ∈ C}. 33.1.1 Some other proofs of previously proved theorems

Then C ⊂ D1 and D1 is also a λ0 –class as we now check.` a) X ∈ D1 . b) If Proof. Other Proof of Corollary 18.54. Let D := {A ⊂ X : 1A ∈ H}. Then by
A, B ∈ D1 with A ∩ B = ∅, then (A ∪ B) ∩ C = (A ∩ C) (B ∩ C) ∈ D for all
assumption C ⊂ D and since 1 ∈ H we know X ∈ D. If A, B ∈ D are disjoint
C ∈ C. c) If A, B ∈ D1 with B ⊂ A, then (A \ B) ∩ C = A ∩ C \ (B ∩ C) ∈ D
then 1A∪B = 1A + 1B ∈ H so that A ∪ B ∈ D and if A, B ∈ D and A ⊂ B, then
for all C ∈ C. Since C ⊂ D1 ⊂ D and D is the smallest λ0 – class containing C
1B\A = 1B − 1A ∈ H. Finally if An ∈ D and An ↑ A as n → ∞ then 1An → 1A
it follows that D1 = D. From this we conclude that if A ∈ D and B ∈ C then
boundedly so 1A ∈ H and hence A ∈ D. So D is λ – class containing C and
A ∩ B ∈ D. Let
hence D contains σ(C). From this it follows that H contains 1A for all A ∈ σ(C)
D2 := {A ∈ D : A ∩ D ∈ D ∀ D ∈ D}.
and hence all σ(C) – measurable simple functions by linearity. The proof is now
Then D2 is a λ0 –class (as you should check) which, by the above paragraph, con- complete with an application of the approximation Theorem 18.42 along with
tains C. As above this implies that D = D2 , i.e. we have shown that D is closed the assumption that H is closed under bounded convergence.
under finite intersections. Since λ0 – classes are closed under complementation, Proof. Other Proof of Theorems 18.51 and 18.52. Let F be R or C. Let C
D is an algebra and hence A (C) ⊂ D. In fact D = A(C). be the family of all sets of the form:
This Lemma along with the monotone class theorem immediately implies
Dynkin’s very useful “π – λ theorem.” B := {x ∈ X : f1 (x) ∈ R1 , . . . , fm (x) ∈ Rm } (33.2)
Theorem 33.5 (π – λ Theorem). If D is a λ class which contains a contains where m = 1, 2, . . . , and for k = 1, 2, . . . , m, fk ∈ M and Rk is an open interval
a π – class, C, then σ(C) ⊂ D. if F = R or Rk is an open rectangle in C if F = C. The family C is easily seen
to be a π – system such that σ(M ) = σ(C). So By Corollary 18.54, to finish
Proof. First Proof. Since D is a λ0 – class, Lemma 33.4 implies that the proof it suffices to show 1B ∈ H for all B ∈ C. It is easy toconstruct, for
A(C) ⊂ D and so by Remark 33.2 and Lemma 33.3, σ(C) ⊂ D. Let us pause to ∞
each k, a uniformly bounded sequence of continuous functions φkn n=1 on F
give a second, stand-alone, proof of this Theorem. converging to the characteristic function 1Rk . By Weierstrass’ theorem, there
Second Proof. With out loss of generality, we may assume that D is the

exists polynomials pkm (x) such that pkn (x) − φkn (x) ≤ 1/n for |x| ≤ kφk k∞ in
smallest λ – class containing C for if not just replace D by the intersection of the real case and polynomials pkm (z, z̄) in z and z̄ such that pkn (z, z̄) − φkn (z) ≤
all λ – classes containing C. Let 1/n for |z| ≤ kφk k∞ in the complex case. The functions
D1 := {A ∈ D : A ∩ C ∈ D ∀ C ∈ C}.

33.2 Regularity of Measures 379
Fn :=p1n (f1 )p2n (f2 ) . . . pm

n (fm ) (real case) 33.2 Regularity of Measures
¯ ¯
Fn :=pn (f1 f1 )pn (f2 , f2 ) . . . pm
1 2 ¯
n (fm , fm ) (complex case) Definition 33.7. Suppose that E is a collection of subsets of X, let Eσ denote
on X are uniformly bounded, belong to H and converge pointwise to 1B as the collection of subsets of X which are finite or countable unions of sets from
n → ∞, where B is the set in Eq. (33.2). Thus 1B ∈ H and the proof is E. Similarly let Eδ denote the collection of subsets of X which are finite or
complete. countable intersections of sets from E. We also write Eσδ for (Eσ )δ and Eδσ for
(Eδ )σ , etc.
Theorem 33.6 (Uniqueness). Suppose that E ⊂ 2X is an elementary class Remark 33.8. Notice that if A is an algebra and C = ∪Ci and D = ∪Dj with
and M = σ(E) (the σ – algebra generated by E). If µ and ν are two measures Ci , Dj ∈ Aσ , then
on M which are σ – finite on E and such that µ = ν on E then µ = ν on M. C ∩ D = ∪i,j (Ci ∩ Dj ) ∈ Aσ
Proof. Let A := A(E) be the algebra generated by E. Since every element so that Aσ is closed under finite intersections.
of A is a disjoint union of elements from E, it is clear that µ = ν on A.
The following theorem shows how recover a measure µ on σ(A) from its
Henceforth we may assume that E = A. We begin first with the special case
values on an algebra A.
where µ(X) < ∞ and hence ν(X) = µ(X) < ∞. Let
Theorem 33.9 (Regularity Theorem). Let A ⊂ 2X be an algebra of sets,
C = {A ∈ M : µ(A) = ν(A)} M = σ(A) and µ : M → [0, ∞] be a measure on M which is σ – finite on A.
Then for all A ∈ M,
The reader may easily check that C is a monotone class. Since A ⊂ C, the
monotone class lemma asserts that M = σ(A) ⊂ C ⊂ M showing that C = M µ(A) = inf {µ(B) : A ⊂ B ∈ Aσ } . (33.4)
and hence that µ = ν on M. For the σ – finite case, let Xn ∈ A be sets such
Moreover, if A ∈ M and ε > 0 are given, then there exists B ∈ Aσ such that
that µ(Xn ) = ν(Xn ) < ∞ and Xn ↑ X as n → ∞. For n ∈ N, let
A ⊂ B and µ(B \ A) ≤ ε.
µn (A) := µ(A ∩ Xn ) and νn (A) = ν(A ∩ Xn ) (33.3) Proof. For A ⊂ X, define
for all A ∈ M. Then one easily checks that µn and νn are finite measure on M µ∗ (A) = inf {µ(B) : A ⊂ B ∈ Aσ } .
such that µn = νn on A. Therefore, by what we have just proved, µn = νn on
We are trying to show µ∗ = µ on M. We will begin by first assuming that µ is
M. Hence or all A ∈ M, using the continuity of measures,
a finite measure, i.e. µ(X) < ∞. Let
µ(A) = lim µ(A ∩ Xn ) = lim ν(A ∩ Xn ) = ν(A). F = {B ∈ M : µ∗ (B) = µ(B)} = {B ∈ M : µ∗ (B) ≤ µ(B)}.
n→∞ n→∞
It is clear that A ⊂ F, so the finite case will be finished by showing F is

Using Dynkin’s π – λ Theorem 33.5 we may strengthen Theorem 33.6 to a monotone class. Suppose Bn ∈ F, Bn ↑ B as n → ∞ and let ε > 0 be
the following. given. Since µ∗ (Bn ) = µ(Bn ) there exists An ∈ Aσ such that Bn ⊂ An and
Proof. Second Proof of Theorem 19.55. As in the proof of Theorem µ(An ) ≤ µ(Bn ) + ε2−n i.e.
33.6, it suffices to consider the case where µ and ν are finite measure such that
µ(X) = ν(X) < ∞. In this case the reader may easily verify from the basic µ(An \ Bn ) ≤ ε2−n .
properties of measures that Let A = ∪n An ∈ Aσ , then B ⊂ A and
∞
D = {A ∈ M : µ(A) = ν(A)} X
µ(A \ B) = µ(∪n (An \ B)) ≤ µ((An \ B))
is a λ – class. By assumption C ⊂ D and hence by the π– λ theorem, D contains n=1
∞ ∞
M = σ(C). X X
≤ µ((An \ Bn )) ≤ ε2−n = ε.
n=1 n=1

380 33 Class Arguments
Therefore, µ(A \ B) < ε.

µ∗ (B) ≤ µ(A) ≤ µ(B) + ε
Furthermore, for any B ∈ M there exists A ∈ Aδσ and C ∈ Aσδ such that
and since ε > 0 was arbitrary it follows that B ∈ F. Now suppose that Bn ∈ F A ⊂ B ⊂ C and µ(C \ A) = 0.
and Bn ↓ B as n → ∞ so that
Proof. By Theorem 33.9, there exist C ∈ Aσ such that Ac ⊂ C and µ(C \
µ(Bn ) ↓ µ(B) as n → ∞. A ) ≤ ε. Let B = C c ⊂ A and notice that B ∈ Aδ and that C \ Ac = B c ∩ A =
c
As above choose An ∈ Aσ such that Bn ⊂ An and A \ B, so that

µ(A \ B) = µ(C \ Ac ) ≤ ε.
0 ≤ µ(An ) − µ(Bn ) = µ(An \ Bn ) ≤ 2−n . Finally, given B ∈ M, we may choose An ∈ Aδ and Cn ∈ Aσ such that
An ⊂ B ⊂ Cn and µ(Cn \ B) ≤ 1/n and µ(B \ An ) ≤ 1/n. By replacing AN by
Combining the previous two equations shows that limn→∞ µ(An ) = µ(B). Since
∪N N
n=1 An and CN by ∩n=1 Cn , we may assume that An ↑ and Cn ↓ as n increases.
µ∗ (B) ≤ µ(An ) for all n, we conclude that µ∗ (B) ≤ µ(B), i.e. that B ∈ F. Since
Let A = ∪An ∈ Aδσ and C = ∩Cn ∈ Aσδ , then A ⊂ B ⊂ C and
F is a monotone class containing the algebra A, the monotone class theorem
asserts that µ(C \ A) = µ(C \ B) + µ(B \ A) ≤ µ(Cn \ B) + µ(B \ An )
M = σ(A) ⊂ F ⊂ M
≤ 2/n → 0 as n → ∞.
showing the F = M and hence that µ∗ = µ on M. For the σ – finite case, let
Xn ∈ A be sets such that µ(Xn ) < ∞ and Xn ↑ X as n → ∞. Let µn be the
finite measure on M defined by µn (A) := µ(A ∩ Xn ) for all A ∈ M. Suppose For Exercises 33.1 – 33.3 let τ ⊂ 2X be a topology, M = σ(τ ) and µ
that ε > 0 and A ∈ M are given. By what we have just proved, for all A ∈ M, : M → [0, ∞) be a finite measure, i.e. µ(X) < ∞.
there exists Bn ∈ Aσ such that A ⊂ Bn and
Exercise 33.1. Let
µ ((Bn ∩ Xn ) \ (A ∩ Xn )) = µn (Bn \ A) ≤ ε2−n .
F := {A ∈ M : µ(A) = inf {µ(V ) : A ⊂ V ∈ τ }} . (33.5)
Notice that since Xn ∈ Aσ , Bn ∩ Xn ∈ Aσ and
1. Show F may be described as the collection of set A ∈ M such that for all
B := ∪∞
n=1 (Bn ∩ Xn ) ∈ Aσ . ε > 0 there exists V ∈ τ such that A ⊂ V and µ(V \ A) < ε.
2. Show F is a monotone class.
Moreover, A ⊂ B and
∞ ∞ Exercise 33.2. Give an example of a topology τ on X = {1, 2} and a measure
µ on M = σ(τ ) such that F defined in Eq. (33.5) is not M.
X X
µ(B \ A) ≤ µ((Bn ∩ Xn ) \ A) ≤ µ((Bn ∩ Xn ) \ (A ∩ Xn ))
n=1 n=1
∞ Exercise 33.3. Suppose now τ ⊂ 2X is a topology with the property that to
every closed set C ⊂ X, there exists Vn ∈ τ such that Vn ↓ C as n → ∞. Let
X
≤ ε2−n = ε.
n=1 A = A(τ ) be the algebra generated by τ.
Since this implies that 1. With the aid of Exercise 18.1, show that A ⊂ F. Therefore by exercise 33.1
and the monotone class theorem, F = M, i.e.
µ(A) ≤ µ(B) ≤ µ(A) + ε
µ(A) = inf {µ(V ) : A ⊂ V ∈ τ } .
and ε > 0 is arbitrary, this equation shows that Eq. (33.4) holds.
2. Show this result is equivalent to following statement: for every ε > 0 and
Corollary 33.10. Let A ⊂ 2X be an algebra of sets, M = σ(A) and µ : M → A ∈ M there exist a closed set C and an open set V such that C ⊂ A ⊂ V
[0, ∞] be a measure on M which is σ – finite on A. Then for all A ∈ M and and µ(V \ C) < ε. (Hint: Apply part 1. to both A and Ac .)
ε > 0 there exists B ∈ Aδ such that B ⊂ A and

33.2 Regularity of Measures 381
Exercise 33.4 (Generalization to the σ – finite case). Let τ ⊂ 2X be a Fε = {x ∈ X : dF (x) < ε} = ∪x∈F Bx (ε) ∈ τd .
topology with the property that to every closed set F ⊂ X, there exists Vn ∈ τ
Show that if F is closed, then Fε ↓ F as ε ↓ 0 and in particular Vn := F1/n ∈ τd
such that Vn ↓ F as n → ∞. Also let M = σ(τ ) and µ : M → [0, ∞] be a
are open sets decreasing to F. Therefore the results of Exercises 33.3 and 33.4
measure which is σ – finite on τ.
apply to measures on metric spaces with the Borel σ – algebra, B = σ(τd ).
1. Show that for all ε > 0 and A ∈ M there exists an open set V ∈ τ and a
Corollary 33.11. Let X ⊂ Rn be an open set and B = BX be the Borel σ –
closed set F such that F ⊂ A ⊂ V and µ(V \ F ) ≤ ε.
algebra on X equipped with the standard topology induced by open balls with
2. Let Fσ denote the collection of subsets of X which may be written as a
respect to the Euclidean distance. Suppose that µ : B → [0, ∞] is a measure
countable union of closed sets. Use item 1. to show for all B ∈ M, there
such that µ(K) < ∞ whenever K is a compact set.
exists C ∈ τδ (τδ is customarily written as Gδ ) and A ∈ Fσ such that
A ⊂ B ⊂ C and µ(C \ A) = 0. 1. Then for all A ∈ B and ε > 0 there exist a closed set F and an open set V
such that F ⊂ A ⊂ V and µ(V \ F ) < ε.
33.2.1 Another proof of Theorem 28.22 2. If µ(A) < ∞, the set F in item 1. may be chosen to be compact.
3. For all A ∈ B we may compute µ(A) using
Proof. The main part of this proof is an application of Exercise 33.4. So we
begin by checking the hypothesis of this exercise. Suppose that C @ X is a µ(A) = inf{µ(V ) : A ⊂ V and V is open} (33.8)
closed set, then by assumption there exists Kn @@ X such that C c = ∪∞ n=1 Kn . = sup{µ(K) : K ⊂ A and K is compact}. (33.9)
Letting VN := ∩N c
n=1 Kn ⊂o X, by taking complements of the last equality we
find that VN ↓ C as N → ∞. Also by assumption there exists Kn @@ X such Proof. For k ∈ N, let
that Kn ↑ X as n → ∞. For each x ∈ Kn , let Vx ⊂o X be a precompact
neighborhood of x. By compactness of Kn there is a finite set Λ ⊂⊂ Kn such Kk := {x ∈ X : |x| ≤ k and dX c (x) ≥ 1/k} . (33.10)
that Kn ⊂ Vn := ∪x∈Λ Vx . Since V̄ = ∪x∈Λ V̄x is a finite union of compact set, Then Kk is a closed and bounded subset of Rn and hence compact. Moreover
V̄n is compact and hence µ(Vn ) ≤ µ(V̄n ) < ∞. Since X = ∪Kn ⊂ ∪Vn we learn Kko ↑ X as k → ∞ since1
that µ is σ finite on open sets of X. By Exercise 33.4, we conclude that for all
ε > 0 and A ∈ BX there exists V ⊂o X and F @ X such that F ⊂ A ⊂ V and {x ∈ X : |x| < k and dX c (x) > 1/k} ⊂ Kko
µ(V \ F ) < ε. For this F and V we have
and {x ∈ X : |x| < k and dX c (x) > 1/k} ↑ X as k → ∞.This shows µ is σ –
µ(A) ≤ µ(V ) = µ(A) + µ(V \ A) ≤ µ(A) + µ(V \ F ) < µ(A) + ε (33.6) finite on τX and Item 1. follows from Exercises 33.4 and 33.5. If µ(A) < ∞
and F ⊂ A ⊂ V as in item 1. Then Kk ∩ F ↑ F as k → ∞ and therefore since
and
µ(V ) < ∞, µ(V \Kk ∩F ) ↓ µ(V \F ) as k → ∞. Hence by choosing k sufficiently
µ(F ) ≤ µ(A) = µ(F ) + µ(A \ F ) < µ(F ) + ε. (33.7)
large, µ(V \ Kk ∩ F ) < ε and we may replace F by the compact set F ∩ Kk and
From Eq. (33.6) we see that µ is outer regular on BX . To finish the proof of item 1. still holds. This proves item 2. Item 3. Item 1. easily implies that Eq.
inner regularity, let Kn @@ X such that Kn ↑ X. If µ(A) = ∞, it follows from (33.8) holds and item 2. implies Eq. (33.9) holds when µ(A) < ∞. So we need
Eq. (33.7) that µ(F ) = ∞. Since F ∩ Kn ↑ F, µ(F ∩ Kn ) ↑ ∞ = µ(A) which only check Eq. (33.9) when µ(A) = ∞. By Item 1. there is a closed set F ⊂ A
shows that µ is inner regular on A because F ∩ Kn is a compact subset of A for such that µ(A \ F ) < 1 and in particular µ(F ) = ∞. Since Kn ∩ F ↑ F, and
each n. If µ(A) < ∞, we again have F ∩ Kn ↑ F and hence by Eq. (33.7) for n Kn ∩ F is compact, it follows that the right side of Eq. (33.9) is infinite and
sufficiently large we still have hence equal to µ(A).
µ(F ∩ Kn ) ≤ µ(A) < µ(F ∩ Kn ) + ε
33.2.2 Second Proof of Theorem 22.13
from which it follows that µ is inner regular on A.
Proof. Second Proof of Theorem 22.13 Since Sf (M, µ) is dense in Lp (µ) it
Exercise 33.5 (Metric Space Examples). Suppose that (X, d) is a metric suffices to show any φ ∈ Sf (M, µ) may be well approximated by f ∈ BCf (X).
space and τd is the topology of d – open subsets of X. To each set F ⊂ X and
1
ε > 0 let In fact this is an equality, but we will not need this here.

Moreover, to prove this it suffices to show for A ∈ M with µ(A) < ∞ that 1A
may be well approximated by an f ∈ BCf (X). By Exercises 33.4 and 33.5, for
any ε > 0 there exists a closed set F and an open set V such that F ⊂ A ⊂ V
and µ(V \ F ) < ε. (Notice that µ(V ) < µ(A) + ε < ∞.) Let f be as in Eq. (6.4),
then f ∈ BCf (X) and since |1A − f | ≤ 1V \F ,
Z Z
p
|1A − f | dµ ≤ 1V \F dµ = µ(V \ F ) ≤ ε (33.11)
or equivalently
k1A − f k ≤ ε1/p .
Since ε > 0 is arbitrary, we have shown that 1A can be approximated in Lp (µ)
arbitrarily well by functions from BCf (X)).
Part VIII
The Fourier Transform and Generalized Functions

34
Fourier Transform
The underlying space in this section is Rn with Lebesgue measure. The The following notation will also be convenient; given a multi-index α ∈ Zn+ ,
Fourier inversion formula is going to state that let |α| = α1 + · · · + αn ,
n Z n α n αj
∂ ∂
Z
1 α
Y αj α
Y
f (x) = dξeiξ·x dyf (y)e−iy·ξ . (34.1) x := xj , ∂x = := and
2π Rn Rn j=1
∂x j=1
∂xj
|α| α α
If we let ξ = 2πη, this may be written as α 1 ∂ 1 ∂
Dx = = .
Z Z i ∂x i ∂x
f (x) = dηei2πη·x dyf (y)e−i2πy·η
Rn Rn Also let
2
hxi := (1 + |x| )1/2
1 n

and we have removed the multiplicative factor of in Eq. (34.1) at the
2π and for s ∈ R let
expense of placing factors of 2π in the arguments of the exponentials. Another νs (x) = (1 + |x|)s .
way to avoid writing the 2π’s altogether is to redefine dx and dξ and this is
what we will do here.
Notation 34.1 Let m be Lebesgue measure on Rn and define: 34.1 Fourier Transform

1
n
1
n Definition 34.2 (Fourier Transform). For f ∈ L1 , let
dx = √ dm(x) and dξ := √ dm(ξ).
2π 2π
Z
fˆ(ξ) = Ff (ξ) := e−ix·ξ f (x)dx (34.2)
n
To be consistent with this new normalization of Lebesgue measure we will rede- R
Z
fine kf kp and hf, gi as ∨ −1
g (x) = F g(x) = eix·ξ g(ξ)dξ = Fg(−x) (34.3)
Rn
1/p n/2 Z !1/p
The next theorem summarizes some more basic properties of the Fourier
Z
p 1 p
kf kp = |f (x)| dx = |f (x)| dm(x) transform.
Rn 2π Rn
Theorem 34.3. Suppose that f, g ∈ L1 . Then
and Z
f (x)g(x)dx when f g ∈ L1 . 1. fˆ ∈ C0 (Rn ) and fˆ ≤ kf k1 .

hf, gi := ∞
Rn
2. For y ∈ Rn , (τy f ) ˆ(ξ) = e−iy·ξ fˆ(ξ) where, as usual, τy f (x) := f (x − y).
Similarly we will define the convolution relative to these normalizations by ˆ
1 n/2
3. The Fourier transform takes convolution to products, i.e. (f Fg) = fˆĝ.
f Fg := 2π f ∗ g, i.e. 4. For f, g ∈ L1 , hfˆ, gi = hf, ĝi.
Z Z n/2 5. If T : Rn → Rn is an invertible linear transformation, then
1
f Fg(x) = f (x − y)g(y)dy = f (x − y)g(y) dm(y). ∧ −1 ∗
(f ◦ T ) (ξ) = |det T | fˆ( T −1 ξ) and
Rn Rn 2π
∨ −1 ∗
(f ◦ T ) (ξ) = |det T | f ∨ ( T −1 ξ)
386 34 Fourier Transform
Z Z
6. If (1 + |x|)k f (x) ∈ L1 , then fˆ ∈ C k and ∂ α fˆ ∈ C0 for all |α| ≤ k. Moreover, ˆ
(∂ α f ) (ξ) = ∂xα f (x)e−ix·ξ dx = (−1)|α| f (x)∂xα e−ix·ξ dx
R n R n
α
∂ξα fˆ(ξ) = F [(−ix) f (x)] (ξ) (34.4)
Z
= (−1) |α| α −ix·ξ
f (x)(−iξ) e dx = (iξ)α fˆ(ξ).
Rn
for all |α| ≤ k.
7. If f ∈ C k and ∂ α f ∈ L1 for all |α| ≤ k, then (1 + |ξ|)k fˆ(ξ) ∈ C0 and ˆ
Since ∂ α f ∈ L1 for all |α| ≤ k, it follows that (iξ)α fˆ(ξ) = (∂ α f ) (ξ) ∈ C0 for
all |α| ≤ k. Since
ˆ
(∂ α f ) (ξ) = (iξ)α fˆ(ξ) (34.5) !k
X n X
k
for all |α| ≤ k. (1 + |ξ|) ≤ 1 + |ξi | = cα |ξ α |
8. Suppose g ∈ L1 (Rk ) and h ∈ L1 (Rn−k ) and f = g ⊗ h, i.e. i=1 |α|≤k
f (x) = g(x1 , . . . , xk )h(xk+1 , . . . , xn ), where 0 < cα < ∞,

X
k
then fˆ = ĝ ⊗ ĥ. (1 + |ξ|) fˆ(ξ) ≤ cα ξ α fˆ(ξ) → 0 as ξ → ∞.

|α|≤k
Proof. Item 1. is the Riemann Lebesgue Lemma 22.37. Items 2. – 5. are
Item 8. is a simple application of Fubini’s theorem.
proved by the following straight forward computations:
2 2
Z Z Example 34.4. If f (x) = e−|x| /2
then fˆ(ξ) = e−|ξ| /2 , in short
(τy f ) ˆ(ξ) = e −ix·ξ
f (x − y)dx = e−i(x+y)·ξ f (x)dx = e−iy·ξ fˆ(ξ), 2 2 2 2
ZRn
Z Rn
Z Fe−|x| /2
= e−|ξ| /2
and F −1 e−|ξ| /2
= e−|x| /2
. (34.6)
hfˆ, gi = fˆ(ξ)g(ξ)dξ = dξg(ξ) dxe−ix·ξ f (x) More generally, for t > 0 let
Rn Rn Rn
2
Z Z 1
= dxdξe −ix·ξ
g(ξ)f (x) = dxĝ(x)f (x) = hf, ĝi, pt (x) := t−n/2 e− 2t |x| (34.7)
Rn ×Rn Rn ×Rn
Z Z Z then 2
ˆ t
(f Fg) (ξ) = e −ix·ξ
f Fg(x)dx = e−ix·ξ
f (x − y)g(y)dy dx pbt (ξ) = e− 2 |ξ| and (b
pt )∨ (x) = pt (x). (34.8)
n Rn Rn
ZR Z By Item 8. of Theorem 34.3, to prove Eq. (34.6) it suffices toconsider the 1
= dy dxe−ix·ξ f (x − y)g(y) 2 Qn 2 2
Rn Rn – dimensional case because e−|x| /2 = i=1 e−xi /2 . Let g(ξ) := Fe−x /2 (ξ) ,
Z Z
then by Eq. (34.4) and Eq. (34.5),
= dy dxe−i(x+y)·ξ f (x)g(y)
n n
ZR R
h 2
i d −x2 /2
g 0 (ξ) = F (−ix) e−x /2 (ξ) = iF
Z
e (ξ)
= dye −iy·ξ
g(y) dxe−ix·ξ f (x) = fˆ(ξ)ĝ(ξ) dx
Rn Rn h 2
i
= i(iξ)F e−x /2 (ξ) = −ξg(ξ). (34.9)
−1
and letting y = T x so that dx = |det T | dy
Z Z Lemma 20.26 implies
ˆ −1 −1
(f ◦ T ) (ξ) = e−ix·ξ f (T x)dx = e−iT y·ξ f (y) |det T | dy Z
2 1
Z
2
Rn Rn g(0) = e−x /2
dx = √ e−x /2
dm(x) = 1,
−1 ∗ 2π
= |det T | fˆ( T −1 ξ). R R
h 2
i 2
and so solving Eq. (34.9) with g(0) = 1 gives F e−x /2 (ξ) = g(ξ) = e−ξ /2 as
Item 6. is simply a matter of differentiating under the integral sign which is
2 2
easily justified because (1 + |x|)k f (x) ∈ L1 . Item 7. follows by using Lemma desired. The assertion that F −1 e−|ξ| /2
= e−|x| /2
follows similarly or by using
22.36 repeatedly (i.e. integration by parts) to find Eq. (34.3) to conclude,

34.2 Schwartz Test Functions 387
h 2
i h 2
i h 2
i 2
F −1 e−|ξ| /2
(x) = F e−|−ξ| /2
(x) = F e−|ξ| /2
(x) = e−|x| /2
. Remark 34.7. Since Cc∞ (Rn ) ⊂ S ⊂ L2 (Rn ) , it follows that S is dense in
L2 (Rn ).
The results in Eq. (34.8) now follow from Eq. √ (34.6) and item 5 of Theorem Exercise 34.1. Let
34.3. For example, since pt (x) = t−n/2 p1 (x/ t), X
L= aα (x)∂ α (34.10)
√ n √ t 2 |α|≤k
pt )(ξ) = t−n/2
(b t p̂1 ( tξ) = e− 2 |ξ| .
with aα ∈ P. Show L(S) ⊂ S and in particular ∂ α f and xα f are back in S for
pt )(ξ) = t−n/2 p 1t (ξ). Using this and the fact that
This may also be written as (b all multi-indices α.
pt is an even function,
Notation 34.8 Suppose that p(x, ξ) = Σ|α|≤N aα (x)ξ α where each function
pt )∨ (x) = F pbt (−x) = t−n/2 Fp 1t (−x) = t−n/2 tn/2 pt (−x) = pt (x).
(b aα (x) is a smooth function. We then set
p(x, Dx ) := Σ|α|≤N aα (x)Dxα

34.2 Schwartz Test Functions and if each aα (x) is also a polynomial in x we will let
Definition 34.5. A function f ∈ C(Rn , C) is said to have rapid decay or p(−Dξ , ξ) := Σ|α|≤N aα (−Dξ )Mξα
rapid decrease if
where Mξα is the operation of multiplication by ξ α .
N
sup (1 + |x|) |f (x)| < ∞ for N = 1, 2, . . . .
x∈Rn Proposition 34.9. Let p(x, ξ) be as above and assume each aα (x) is a polyno-
mial in x. Then for f ∈ S,
Equivalently, for each N ∈ N there exists constants CN < ∞ such that |f (x)| ≤
∧
CN (1 + |x|)−N for all x ∈ Rn . A function f ∈ C(Rn , C) is said to have (at (p(x, Dx )f ) (ξ) = p(−Dξ , ξ)fˆ (ξ) (34.11)
most) polynomial growth if there exists N < ∞ such
and
−N
sup (1 + |x|) |f (x)| < ∞, p(ξ, Dξ )fˆ(ξ) = [p(Dx , −x)f (x)]∧ (ξ). (34.12)
α −ix·ξ
i.e. there exists N ∈ N and C < ∞ such that |f (x)| ≤ C(1 + |x|)N for all Proof. The identities (−Dξ ) e = xα e−ix·ξ and Dxα eix·ξ = ξ α eix·ξ im-
n
x ∈ Rn . ply, for any polynomial function q on R ,
Definition 34.6 (Schwartz Test Functions). Let S denote the space of func- q(−Dξ )e−ix·ξ = q(x)e−ix·ξ and q(Dx )eix·ξ = q(ξ)eix·ξ . (34.13)
tions f ∈ C ∞ (Rn ) such that f and all of its partial derivatives have rapid decay
and let Therefore using Eq. (34.13) repeatedly,
kf kN,α = sup (1 + |x|)N ∂ α f (x)

Z
∧
X
x∈Rn (p(x, Dx )f ) (ξ) = aα (x)Dxα f (x) · e−ix·ξ dξ
so that Rn |α|≤N
n o
S = f ∈ C ∞ (Rn ) : kf kN,α < ∞ for all N and α . Z X
= Dxα f (x) · aα (−Dξ )e−ix·ξ dξ
∞ n Rn |α|≤N
Also let P denote those functions g ∈ C (R ) such that g and all of its deriva-
tives have at most polynomial growth, i.e. g ∈ C ∞ (Rn ) is in P iff for all multi- Z X α
aα (−Dξ )e−ix·ξ dξ

indices α, there exists Nα < ∞ such = f (x) (−Dx )
Rn |α|≤N
−Nα α
sup (1 + |x|) |∂ g(x)| < ∞.
Z X
aα (−Dξ ) ξ α e−ix·ξ dξ = p(−Dξ , ξ)fˆ (ξ)

= f (x)
Rn
(Notice that any polynomial function on Rn is in P.) |α|≤N

wherein the third inequality we have used Lemma 22.36 to do repeated integra- Proof. First notice that fˆ ∈ C0 (Rn ) ⊂ L∞ and fˆ ∈ L1 by assumption, so
1 2
tion by parts, the fact that mixed partial derivatives commute in the fourth, that fˆ ∈ L1 ∩ L∞ . Let pt (x) := t−n/2 e− 2t |x| be as in Example 34.4 so that
t 2
and in the last we have repeatedly used Corollary 19.43 to differentiate under pbt (ξ) = e− 2 |ξ| and pb∨ ˆ∨
t = pt . Define f0 := f ∈ C0 then
the integral. The proof of Eq. (34.12) is similar: Z Z
Z Z f0 (x) = (fˆ)∨ (x) = fˆ(ξ)eiξ·x dξ = lim fˆ(ξ)eiξ·x pbt (ξ)dξ
t↓0 Rn
p(ξ, Dξ )fˆ(ξ) = p(ξ, Dξ ) f (x)e−ix·ξ dx = f (x)p(ξ, −x)e−ix·ξ dx Z Z Rn
n n
X Z
R R
= lim f (y)eiξ·(x−y) pbt (ξ)dξ dy
α −ix·ξ t↓0 Rn Rn
= f (x)(−x) aα (ξ)e dx Z
Rn
|α|≤N = lim f (y)pt (x − y)dy = f (x) a.e.
X Z t↓0 Rn
= f (x)(−x)α aα (−Dx )e−ix·ξ dx wherein we have used Theorem 22.32
|α|≤N Rn √ R in the last equality along with the obser-
vations that pt (y) = p1 (y/ t) and Rn p1 (y)dy = 1 so that
X Z
= e−ix·ξ aα (Dx ) [(−x)α f (x)] dx 1
Z
|α|≤N Rn L – lim f (y)pt (x − y)dy = f (x) .
t↓0 Rn
= [p(Dx , −x)f (x)]∧ (ξ).
In particular this shows that f ∈ L1 ∩ L∞ . A similar argument shows that
F −1 F f = f0 as well. Let us now compute the L2 – norm of fˆ,
Z Z Z
Corollary 34.10. The Fourier transform preserves the space S, i.e. F(S) ⊂ S. kfˆk22 = fˆ(ξ)fˆ(ξ)dξ = dξ fˆ(ξ) dxf (x)eix·ξ
n n n
ZR Z R R
Proof. Let p(x, ξ) = Σ|α|≤N aα (x)ξ α with each aα (x) being a polynomial = dx f (x) ˆ
dξ f (ξ)eix·ξ
(by Fubini)
function in x. If f ∈ S then p(Dx , −x)f ∈ S ⊂ L1 and so by Eq. (34.12), n Rn
ZR
p(ξ, Dξ )fˆ(ξ) is bounded in ξ, i.e.
= dx f (x)f (x) = kf k22
Rn
sup |p(ξ, Dξ )fˆ(ξ)| ≤ C(p, f ) < ∞.
dξ fˆ(ξ)e ix·ξ
= F −1 fˆ(x) = f (x) a.e.
R
ξ∈Rn because Rn
2 Corollary 34.12. By the B.L.T. Theorem 10.4, the maps F|S and F −1 |S ex-
Taking p(x, ξ) = (1 + |x| )N ξ α with N ∈ Z+ in this estimate shows fˆ(ξ) and
tend to bounded linear maps F̄ and F̄ −1 from L2 → L2 . These maps satisfy the
all of its derivatives have rapid decay, i.e. fˆ is in S.
following properties:
1. F̄ and F̄ −1 are unitary and are inverses to one another as the notation
34.3 Fourier Inversion Formula suggests.
2. If f ∈ L2 , then F̄f is uniquely characterized as the function, G ∈ L2 such
Theorem 34.11 (Fourier Inversion Theorem). Suppose that f ∈ L1 and that
fˆ ∈ L1 (for example suppose f ∈ S), then hG, ψi = hf, ψ̂i for all ψ ∈ Cc∞ (Rn ).
3. If f ∈ L1 ∩ L2 , then F̄f = fˆ a.e.
1. there exists f0 ∈ C0 (Rn ) such that f = f0 a.e.,
4. For f ∈ L2 we may compute F̄ and F̄ −1 by
2. f0 = F −1 F f and f0 = FF −1 f,
3. f and fˆ are
Z
1 ∞
in L ∩ L and F̄f (ξ) = L2 – lim f (x)e−ix·ξ dx and (34.14)
ˆ R→∞ |x|≤R
4. kf k2 = f .
2 Z
In particular, F : S → S is a linear isomorphism of vector spaces. F̄ −1 f (ξ) = L2 – lim f (x)eix·ξ dx. (34.15)
R→∞ |x|≤R

34.3 Fourier Inversion Formula 389
5. We may further extend F̄ to a map from L1 + L2 → C0 + L2 (still denote holds with F = ĥ + F̄g ∈ C0 + L2 and ϕ ∈ S. Now suppose G ∈ L1loc and
by F̄) defined by F̄f = ĥ + F̄g where f = h + g ∈ L1 + L2 . For f ∈ L1 + L2 , hG, ϕi = hf, ϕ̂i for all ϕ ∈ Cc∞ (Rn ). Then by what we just proved, hG, ϕi =
F̄f may be characterized as the unique function F ∈ L1loc (Rn ) such that hF, ϕi for all ϕ ∈ Cc∞ (Rn ) and so another application of Corollary 22.38 shows
G = F ∈ C0 + L2 .
hF, ϕi = hf, ϕ̂i for all ϕ ∈ Cc∞ (Rn ). (34.16)
Notation 34.13 Given the results of Corollary 34.12, there is little danger in
Moreover if Eq. (34.16) holds then F ∈ C0 + L2 ⊂ L1loc (Rn ) and Eq.(34.16) writing fˆ or Ff for F̄f when f ∈ L1 + L2 .
is valid for all ϕ ∈ S.
Corollary 34.14. If f and g are L1 functions such that fˆ, ĝ ∈ L1 , then
2 2
Proof. 1. and 2. If f ∈ L and ϕn ∈ S such that ϕn → f in L , then
F̄f := limn→∞ ϕ̂n . Since ϕ̂n ∈ S ⊂ L1 , we may concluded that kϕ̂n k2 = kϕn k2 F(f g) = fˆFĝ and F −1 (f g) = f ∨ Fg ∨ .
for all n. Thus
Since S is closed under pointwise products and F : S → S is an isomorphism
it follows that S is closed under convolution as well.

F̄f = lim kϕ̂n k = lim kϕn k = kf k
2 n→∞ 2 n→∞ 2 2
Proof. By Theorem 34.11, f, g, fˆ, ĝ ∈ L1 ∩ L∞ and hence f · g ∈ L1 ∩ L∞

which shows that F̄ is an isometry from L2 to L2 and similarly F̄ −1 is an
and fˆFĝ ∈ L1 ∩ L∞ . Since
isometry. Since F̄ −1 F̄ = F −1 F = id on the dense set S, it follows by continuity
−1
that F̄ −1 F̄ = id on all of L2 . Hence F̄ F̄ = id, and thus F̄ −1 is the inverse

F −1 fˆFĝ = F −1 fˆ · F −1 (ĝ) = f · g ∈ L1
of F̄. This proves item 1. Moreover, if ψ ∈ Cc∞ (Rn ), then
we may conclude from Theorem 34.11 that
hF̄f, ψi = lim hϕ̂n , ψi = lim hϕn , ψ̂i = hf, ψi (34.17)
n→∞ n→∞
fˆFĝ = FF −1 fˆFĝ = F(f · g).
and this equation uniquely characterizes F̄f by Corollary 22.38. Notice that
Eq. (34.17) also holds for all ψ ∈ S.
Similarly one shows F −1 (f g) = f ∨ Fg ∨ .
3. If f ∈ L1 ∩ L2 , we have already seen that fˆ ∈ C0 (Rn ) ⊂ L1loc and
that hfˆ, ψi = hf, ψ̂i for all ψ ∈ Cc∞ (Rn ). Combining this with item 2. shows Corollary 34.15. Let p(x, ξ) and p(x, Dx ) be as in Notation 34.8 with each
hfˆ− F̄f, ψi = 0 or all ψ ∈ Cc∞ (Rn ) and so again by Corollary 22.38 we conclude function aα (x) being a smooth function of x ∈ Rn . Then for f ∈ S,
that fˆ − F̄f = 0 a.e. Z
4. Let f ∈ L2 and R < ∞ and set fR (x) := f (x)1|x|≤R . Then fR ∈ L1 ∩ p(x, Dx )f (x) = p(x, ξ)fˆ (ξ) eix·ξ dξ. (34.19)
L2 and therefore F̄fR = fˆR . Since F̄ is an isometry and (by the dominated Rn
convergence theorem) fR → f in L2 , it follows that Proof. For f ∈ S, we have

F̄f = L2 – lim F̄fR = L2 – lim fˆR . Z
R→∞ R→∞ p(x, Dx )f (x) = p(x, Dx ) F −1 fˆ (x) = p(x, Dx ) fˆ (ξ) eix·ξ dξ
Rn
5. If f = h + g ∈ L1 + L2 and ϕ ∈ S, then by Eq. (34.17) and item 4. of
Z Z
= ˆ ix·ξ
f (ξ) p(x, Dx )e dξ = fˆ (ξ) p(x, ξ)eix·ξ dξ.
Theorem 34.3, Rn Rn
hĥ + F̄g, ϕi = hh, ϕ̂i + hg, ϕ̂i = hh + g, ϕ̂i. (34.18) If p(x, ξ) is a more general function of (x, ξ) then that given in Notation 34.8,
the right member of Eq. (34.19) may still make sense, in which case we may
In particular if h + g = 0 a.e., then hĥ + F̄g, ϕi = 0 for all ϕ ∈ S and since use it as a definition of p(x, Dx ). A linear operator defined this way is called
ĥ + F̄g ∈ L1loc it follows from Corollary 22.38 that ĥ + F̄g = 0 a.e. This shows a pseudo differential operator and they turn out to be a useful class of
that F̄f is well defined independent of how f ∈ L1 + L2 is decomposed into operators to study when working with partial differential equations.
the sum of an L1 and an L2 function. Moreover Eq. (34.18) shows Eq. (34.16)

aα ξ α is a polynomial in ξ ∈ Rn and 34.5 Fourier Transforms of Measures and Bochner’s

P
Corollary 34.16. Suppose p(ξ) = |α|≤N
f ∈ L2 . Then p(∂)f exists in L2 (see Definition 26.3) iff ξ → p(iξ)fˆ(ξ) ∈ L2 Theorem
in which case
ˆ
(p(∂)f ) (ξ) = p(iξ)fˆ(ξ) for a.e. ξ. To motivate the next definition suppose that µ is a finite measure on Rn which
is absolutely continuous relative to Lebesgue measure, dµ(x) = ρ(x)dx. Then
In particular, if g ∈ L2 then f ∈ L2 solves the equation, p(∂)f = g iff
it is reasonable to require
p(iξ)fˆ(ξ) = ĝ(ξ) for a.e. ξ. Z Z
Proof. By definition p(∂)f = g in L2 iff µ̂(ξ) := ρ̂(ξ) = e−iξ·x ρ(x)dx = e−iξ·x dµ(x)
Rn Rn
hg, ϕi = hf, p(−∂)ϕi for all ϕ ∈ Cc∞ (Rn ). (34.20)
and
If follows from repeated use of Lemma 26.23 that the previous equation is Z Z
equivalent to (µFg) (x) := ρFg(x) = g(x − y)ρ(x)dx = g(x − y)dµ(y)
hg, ϕi = hf, p(−∂)ϕi for all ϕ ∈ S(Rn ). (34.21) Rn Rn
This may also be easily proved directly as well as follows. Choose ψ ∈ Cc∞ (Rn ) when g : Rn → C is a function such that the latter integral is defined, for exam-
such that ψ(x) = 1 for x ∈ B0 (1) and for ϕ ∈ S(Rn ) let ϕn (x) := ψ(x/n)ϕ(x). ple assume g is bounded. These considerations lead to the following definitions.
By the chain rule and the product rule (Eq. A.5 of Appendix A), Definition 34.17. The Fourier transform, µ̂, of a complex measure µ on BRn
X α is defined by
∂ α ϕn (x) = n−|β| ∂ β ψ (x/n) · ∂ α−β ϕ(x)
Z
β≤α
β µ̂(ξ) = e−iξ·x dµ(x) (34.22)
Rn
along with the dominated convergence theorem shows ϕn → ϕ and ∂ α ϕn → ∂ α ϕ and the convolution with a function g is defined by
in L2 as n → ∞. Therefore if Eq. (34.20) holds, we find Eq. (34.21) holds because Z
hg, ϕi = lim hg, ϕn i = lim hf, p(−∂)ϕn i = hf, p(−∂)ϕi. (µFg) (x) = g(x − y)dµ(y)
n→∞ n→∞ Rn
To complete the proof simply observe that hg, ϕi = hĝ, ϕ∨ i and when the integral is defined.
∨
hf, p(−∂)ϕi = hfˆ, [p(−∂)ϕ] i = hfˆ(ξ), p(iξ)ϕ∨ (ξ)i It follows from the dominated convergence theorem that µ̂ is continuous.
= hp(iξ)fˆ(ξ), ϕ∨ (ξ)i Also by a variant of Exercise 22.12, if µ and ν are two complex measure on BRn
such that µ̂ = ν̂, then µ = ν. The reader is asked to give another proof of this
for all ϕ ∈ S(Rn ). From these two observations and the fact that F is bijective fact in Exercise 34.4 below.
on S, one sees that Eq. (34.21) holds iff ξ → p(iξ)fˆ(ξ) ∈ L2 and ĝ(ξ) = p(iξ)fˆ(ξ)
for a.e. ξ. Example 34.18. Let σt be the surface measure on the sphere St of radius t
centered at zero in R3 . Then
sin t |ξ|
34.4 Summary of Basic Properties of F and F −1 σ̂t (ξ) = 4πt .
|ξ|
The following table summarizes some of the basic properties of the Fourier Indeed,
transform and its inverse. Z Z
f ←→ fˆ or f ∨ σ̂t (ξ) = e−ix·ξ dσ(x) = t2 e−itx·ξ dσ(x)
tS 2 S2
Smoothness ←→ Decay at infinity Z Z 2π Z π
α
∂α ←→ Multiplication by (±iξ) = t2 e−itx3 |ξ| dσ(x) = t2 dθ dϕ sin ϕe−it cos ϕ|ξ|
S ←→ S S2 0 0
1
L2 (Rn ) ←→ L2 (Rn ) 2 sin t |ξ|
Z
−itu|ξ| 1
= 2πt 2
e du = 2πt 2
e−itu|ξ| |u=1
u=−1 = 4πt .
Convolution ←→ Products. −1 −it |ξ| t |ξ|

34.5 Fourier Transforms of Measures and Bochner’s Theorem 391
m
Definition 34.19. A function χ : Rn → C is said to be positive (semi) From this it follows that for any m ∈ N and ⊂ Rn , the matrix A :=
{ξj }j=1
m m
definite iff the matrices A := {χ(ξk − ξj )}k,j=1 are positive definite for all {µ̂(ξk − ξj )}k,j=1 is self-adjoint. Moreover if λ ∈ C , m
m
m ∈ N and {ξj }j=1 ⊂ Rn .
m
X Z m
X
Lemma 34.20. If χ ∈ C(Rn , C) is a positive definite function, then µ̂(ξk − ξj )λk λ̄j = e−i(ξk −ξj )·x λk λ̄j dµ(x)
k,j=1 Rn k,j=1
1. χ(0) ≥ 0. Z m
2. χ(−ξ) = χ(ξ) for all ξ ∈ Rn .
X
= e−iξk ·x λk e−iξj ·x λj dµ(x)
3. |χ(ξ)| ≤ χ(0) for all ξ ∈ Rn . Rn k,j=1
4. For all f ∈ S(Rd ),
Z
Xm
2

−iξk ·x
= e λk dµ(x) ≥ 0
Z

χ(ξ − η)f (ξ)f (η)dξdη ≥ 0. (34.23) Rn k=1

Rn ×Rn
showing A is positive definite.
2
Proof. Taking m = 1 and ξ1 = 0 we learn χ(0) |λ| ≥ 0 for all λ ∈ C which
proves item 1. Taking m = 2, ξ1 = ξ and ξ2 = η, the matrix Theorem 34.22 (Bochner’s Theorem). Suppose χ ∈ C(Rn , C) is positive
definite function, then there exists a unique positive measure µ on BRn such
χ(0) χ(ξ − η) that χ = µ̂.
A :=
χ(η − ξ) χ(0)
Proof. If χ(ξ) = µ̂(ξ), then for f ∈ S we would have
∗
is positive definite from which we conclude χ(ξ − η) = χ(η − ξ) (since A = A Z Z Z
ˆ
by definition) and f dµ = (f ∨ ) dµ = f ∨ (ξ)µ̂(ξ)dξ.
Rn Rn Rn
χ(0) χ(ξ − η) 2 2
0 ≤ det = |χ(0)| − |χ(ξ − η)| . This suggests that we define
χ(η − ξ) χ(0)
Z
and hence |χ(ξ)| ≤ χ(0) for all ξ. This proves items 2. and 3. Item 4. follows by I(f ) := χ(ξ)f ∨ (ξ)dξ for all f ∈ S.
approximating the integral in Eq. (34.23) by Riemann sums, Rn
Z We will now show I is positive in the sense if f ∈ S and f ≥ 0 then I(f ) ≥ 0.

χ(ξ − η)f (ξ)f (η)dξdη For general f ∈ S we have
Rn ×Rn
X Z ∨ Z
= lim ε−2n

χ(ξ − η)f (ξ)f (η) ≥ 0. 2 2
χ(ξ) f ∨ Ff¯∨ (ξ)dξ

ε↓0
I(|f | ) = χ(ξ) |f | (ξ)dξ =
ξ,η∈(εZn )∩[−ε−1 ,ε−1 ]n ZR
n Rn
Z
= ∨ ¯∨
χ(ξ)f (ξ − η)f (η)dηdξ = χ(ξ)f ∨ (ξ − η)f ∨ (−η)dηdξ
The details are left to the reader.
Rn Rn
Z
Lemma 34.21. If µ is a finite positive measure on BRn , then χ := µ̂ ∈ = χ(ξ − η)f ∨ (ξ)f ∨ (η)dηdξ ≥ 0. (34.24)
C(Rn , C) is a positive definite function. Rn
2
Proof. As has already been observed after Definition 34.17, the dominated For t > 0 let pt (x) := t−n/2 e−|x| /2t
∈ S and define
convergence theorem implies µ̂ ∈ C(Rn , C). Since µ is a positive measure (and p 2
hence real), It (x) := IFpt (x) := I(pt (x − ·)) = I( pt (x − ·) )

Z Z
iξ·x
e−iξ·x dµ(x) = µ̂(−ξ).
p
µ̂(−ξ) = e dµ(x) = which is non-negative by Eq. (34.24) and the fact that pt (x − ·) ∈ S. Using
Rn Rn

ξ−η
Z Z
∨
pt (x − y)eiy·ξ dy = pt (y)ei(y+x)·ξ dy F −1 e−iηx fε (x) (ξ) = ε−n f ∨ (

[pt (x − ·)] (ξ) = )
Rn Rn ε
ix·ξ ∨ ix·ξ −t|ξ|2 /2
= e pt (ξ) =e e , and therefore, from Eq. (34.26),
ξ−η
Z Z
Z e−iηx fε (x)dµ(x) = χ(ξ)ε−n f ∨ ( )dξ. (34.27)
hIt , ψi = I(pt (x − ·))ψ(x)dx Rn Rn ε
n
ZR Z Because Rn f ∨ (ξ)dξ = Ff ∨ (0) = f (0) = 1, we may apply the approximate δ
R
∨
= χ(ξ) [pt (x − ·)] (ξ)ψ(x)dξ dx – function Theorem 22.32 to Eq. (34.27) to find
Rn Rn
Z Z Z
ix·ξ −t|ξ|2 /2
= χ(ξ)e e ψ(x)dξ dx e−iηx fε (x)dµ(x) → χ(η) as ε ↓ 0. (34.28)
n Rn
ZR Rn
2
= χ(ξ)ψ ∨ (ξ)e−t|ξ| /2 dξ On the the other hand, when η = 0, the monotone convergence theorem implies
Rn
µ(fε ) ↑ µ(1) = µ(Rn ) and therefore µ(Rn ) = µ(1) = χ(0) < ∞. Now knowing
which coupled with the dominated convergence theorem shows the µ is a finite measure we may use the dominated convergence theorem to
concluded
µ(e−iηx fε (x)) → µ(e−iηx ) = µ̂(η) as ε ↓ 0
Z
hIFpt , ψi → χ(ξ)ψ ∨ (ξ)dξ = I(ψ) as t ↓ 0.
Rn
for all η. Combining this equation with Eq. (34.28) shows µ̂(η) = χ(η) for all
Hence if ψ ≥ 0, then I(ψ) = limt↓0 hIt , ψi ≥ 0. η ∈ Rn .
Let K ⊂ R be a compact set and ψ ∈ Cc (R, [0, ∞)) be a function such that
ψ = 1 on K. If f ∈ Cc∞ (R, R) is a smooth function with supp(f ) ⊂ K, then
0 ≤ kf k∞ ψ − f ∈ S and hence 34.6 Supplement: Heisenberg Uncertainty Principle
0 ≤ hI, kf k∞ ψ − f i = kf k∞ hI, ψi − hI, f i Suppose that H is a Hilbert space and A, B are two densely defined symmetric
operators on H. More explicitly, A is a densely defined symmetric linear operator
and therefore hI, f i ≤ kf k∞ hI, ψi. Replacing f by −f implies, −hI, f i ≤ on H means there is a dense subspace DA ⊂ H and a linear map A : DA → H
kf k∞ hI, ψi and hence we have proved such that hAϕ|ψi = hϕ|Aψi for all ϕ, ψ ∈ DA . Let
|hI, f i| ≤ C(supp(f )) kf k∞ (34.25) DAB := {ϕ ∈ H : ϕ ∈ DB and Bϕ ∈ DA }
for all f ∈ DRn := Cc∞ (Rn , R) where C(K) is a finite constant for each compact and for ϕ ∈ DAB let (AB) ϕ = A(Bϕ) with a similar definition of DBA and
subset of Rn . Because of the estimate in Eq. (34.25), it follows that I|DRn has a BA. Moreover, let DC := DAB ∩ DBA and for ϕ ∈ DC , let
unique extension I to Cc (Rn , R) still satisfying the estimates in Eq. (34.25) and
moreover this extension is still positive. So by the Riesz – Markov Theorem 1 1
Cϕ = [A, B]ϕ = (AB − BA) ϕ.
32.47, there exists a unique Radon – measure µ on Rn such that such that i i
hI, f i = µ(f ) for all f ∈ Cc (Rn , R). Notice that for ϕ, ψ ∈ DC we have
To finish the proof we must show µ̂(η) = χ(η) for all η ∈ Rn given
1 1
Z hCϕ|ψi = {hABϕ|ψi − hBAϕ|ψi} = {hBϕ|Aψi − hAϕ|Bψi}
µ(f ) = χ(ξ)f ∨ (ξ)dξ for all f ∈ Cc∞ (Rn , R). (34.26) i i
Rn 1
= {hϕ|BAψi − hϕ|ABψi} = hϕ|Cψi,
i
Let f ∈ Cc∞ (Rn , R+ ) be a radial function such f (0) = 1 and f (x) is decreasing
as |x| increases. Let fε (x) := f (εx), then by Theorem 34.3, so that C is symmetric as well.

34.6 Supplement: Heisenberg Uncertainty Principle 393
Theorem 34.23 (Heisenberg Uncertainty Principle). Continue the above and C = −I on DC . Therefore for a unit vector ψ ∈ DC ,
notation and assumptions,
1 1 0
1
q q ≤ ψ − aψ · kxψ − bψk
2
2 2
|hψ|Cψi| ≤ kAψk − hψ|Aψi · kBψk − hψ|Bψi (34.29) 2 i
2
2
2
where a = i R ψ ψ̄ 0 dm 1 and b = R x |ψ(x)| dm(x). Thus we have
R R
for all ψ ∈ DC . Moreover if kψk = 1 and equality holds in Eq. (34.29), then
Z Z 2 Z
(A − hψ|AψiI) ψ = iλ (B − hψ|BψiI) ψ or 1 1 2

2 2 2
= |ψ| dm ≤ (k − a) ψ̂(k) dk · (x − b) |ψ(x)| dx. (34.32)

4 4 R
(B − hψ|BψiI) ψ = iλ (A − hψ|AψiI) ψ (34.30) R R
Equality occurs if there exists λ ∈ R such that

for some λ ∈ R.
1
Proof. By homogeneity (34.29) we may assume that kψk = 1. Let a := iλ (x − b) ψ(x) = ( ∂x − a)ψ(x) a.e.
i
hψ|Aψi, b = hψ|Bψi, Ã = A − aI, and B̃ = B − bI. Then we have still have
Working formally, this gives rise to the ordinary differential equation (in weak
[Ã, B̃] = [A − aI, B − bI] = iC. form),
ψx = [−λ(x − b) + ia] ψ (34.33)
Now
which has solutions (see Exercise 34.5 below)
ihψ|Cψi = −hψ|iCψi = −hψ|[Ã, B̃]ψi = −hψ|ÃB̃ψi + hψ|B̃ Ãψi Z
λ
= − hÃψ|B̃ψi − hB̃ψ|Ãψi = −2i ImhÃψ|B̃ψi ψ = C exp [−λ(x − b) + ia] dx = C exp − (x − b)2 + iax . (34.34)
R 2
from which we learn Let λ = 1
and choose C so that kψk2 = 1 to find
2t
|hψ|Cψi| = 2 ImhÃψ|B̃ψi ≤ 2 hÃψ|B̃ψi ≤ 2 Ãψ B̃ψ (34.31)

1/4
1 1
ψt,a,b (x) = exp − (x − b)2 + iax
with equality iff RehÃψ|B̃ψi = 0 and Ãψ and B̃ψ are linearly dependent, i.e. 2t 4t
iff Eq. (34.30) holds. Equation (34.29) now follows from the inequality in Eq. are the functions (called coherent states) which saturate the Heisenberg un-
(34.31) and the identities, certainty principle in Eq. (34.32).
2
2 2 2
Ãψ = kAψ − aψk = kAψk + a2 kψk − 2a RehAψ|ψi

34.6.1 Exercises
2 2 2 2
= kAψk + a − 2a = kAψk − hAψ|ψi
Exercise 34.2. Let f ∈ L2 (Rn ) and α be a multi-index. If ∂ α f exists in L2 (Rn )
α
and similarly then F(∂ α f ) = (iξ) fˆ(ξ) in L2 (Rn ) and conversely if ξ → ξ α fˆ(ξ) ∈ L2 (Rn )
2
B̃ψ = kBψk − hBψ|ψi. then ∂ α f exists.

1
The constant a may also be described as
2 1 √
Example 34.24. As an example, take H = L (R), A = i ∂x
and B =
Z Z
ˆ
0 0 a = i ψ ψ̄ 0 dm = 2πi ψ̂(ξ) ψ̄ 0 (ξ)dξ
M x with D A := {f ∈ H : f ∈ H} (f is the weak derivative) and DB :=
n R 2
o Z R 2
R
f ∈ H : R |xf (x)| dx < ∞ . In this case, = ξ ψ̂(ξ) dm(ξ).

R
DC = {f ∈ H : f 0 , xf and xf 0 are in H}

Exercise 34.3. Suppose p(ξ) is a polynomial in ξ ∈ Rd and u ∈ L2 such that 1. Further assume that f is continuously differentiable in a neighborhood of
p (∂) u ∈ L2 . Show 0. Show that Z
F (p (∂) u) (ξ) = p(iξ)û (ξ) ∈ L2 . Λ := fˆ(ξ)dξ = f (0).
R
Conversely if u ∈ L2 such that p(iξ)û (ξ) ∈ L2 , show p (∂) u ∈ L2 .
Hint: by the dominated convergence theorem, Λ := limM →∞ |ξ|≤M fˆ(ξ)dξ.
R
n
Exercise 34.4. Suppose µ is a complex measure on R and µ̂(ξ) is its Fourier
Now use the definition of fˆ(ξ), Fubini’s theorem and Exercise 34.6.
transform as defined in Definition 34.17. Show µ satisfies,
2. Apply part 1. of this exercise with f replace by τy f for some y ∈ R to prove
Z Z
Z
hµ̂, ϕi := µ̂(ξ)ϕ(ξ)dξ = µ(ϕ̂) := ϕ̂dµ for all ϕ ∈ S
Rn Rn f (y) = fˆ(ξ)eiy·ξ dξ (34.36)
R
and use this to show if µ is a complex measure such that µ̂ ≡ 0, then µ ≡ 0.
provided f is now continuously differentiable near y.
Exercise 34.5. Show that ψ described in Eq. (34.34) is the general solution to
The goal of the next exercises is to give yet another proof of the Fourier
Eq. (34.33). Hint: Suppose that ϕ is any solution to Eq. (34.33) and ψ is given
inversion formula.
as in Eq. (34.34) with C = 1. Consider the weak – differential equation solved
by ϕ/ψ. Notation 34.25 For L > 0, let CLk (R) denote the space of C k – 2πL periodic
functions:
34.6.2 More Proofs of the Fourier Inversion Theorem
CLk (R) := f ∈ C k (R) : f (x + 2πL) = f (x) for all x ∈ R .

Exercise 34.6. Suppose that f ∈ L1 (R) and assume that f continuously dif-
ferentiable in a neighborhood of 0, show Also let h·, ·iL denote the inner product on the Hilbert space HL :=
L2 ([−πL, πL]) given by
Z ∞
sin M x
lim f (x)dx = πf (0) (34.35)
Z
1
M →∞ −∞ x hf |giL := f (x)ḡ(x)dx.
2πL [−πL,πL]
using the following steps.
Exercise 34.8. Recall that χL k (x) := e
ikx/L
: k ∈ Z is an orthonormal basis
1. Use Example 20.14 to deduce, for HL and in particular for f ∈ HL ,
Z 1 Z M
sin M x sin x
X
lim dx = lim dx = π. f= hf |χLk iL χk
L
(34.37)
M →∞ −1 x M →∞ −M x k∈Z
2. Explain why where the convergence takes place in L2 ([−πL, πL]). Suppose now that f ∈
Z CL2 (R)2 . Show (by two integration by parts)
f (x)
0 = lim sin M x · dx and
M →∞ |x|≥1 x L2 00
hf |χL

k iL ≤ 2 kf k∞

f (x) − f (0) k
Z
0 = lim sin M x · dx.
M →∞ |x|≤1 x where kgk∞ denote the uniform norm of a function g. Use this to conclude that
the sum in Eq. (34.37) is uniformly convergent and from this conclude that Eq.
3. Add the previous two equations and use part (1) to prove Eq. (34.35).
(34.37) holds pointwise. BRUCE: it is enough to assume f ∈ CL (R) by making
Exercise 34.7 (Fourier Inversion Formula). Suppose that f ∈ L1 (R) such use of the identity,
that fˆ ∈ L1 (R). 2
We view CL2 (R) as a subspace of HL by identifying f ∈ CL2 (R) with f |[−πL,πL] ∈ HL .

hf |χL
L 0 L
k iL = hf |χk iL

|k|
along with the Cauchy Schwarz inequality to see
 2
X L 2

0 L 2
X X
hf |χL

k iL ≤ hf |χk iL · .
 
|k|
k6=0 k6=0 k6=0
Exercise 34.9 (Fourier Inversion Formula on S). Let f ∈ S(R), L > 0

and X
fL (x) := f (x + 2πkL). (34.38)
k∈Z
Show:
1. The sum defining fL is convergent and moreover that fL ∈ CL∞ (R).
2. Show hfL |χL √ 1 fˆ(k/L).
k iL = 2πL
3. Conclude from Exercise 34.8 that
1 Xˆ
fL (x) = √ f (k/L)eikx/L for all x ∈ R. (34.39)
2πL k∈Z
4. Show, by passing to the limit, L → ∞, in Eq. (34.39) that Eq. (34.36) holds
for all x ∈ R. Hint: Recall that fˆ ∈ S.
Exercise 34.11. Folland 8.14 on p. 254. (Wirtinger’s inequality.)
Exercise 34.12. Folland 8.15 on p. 255. (The sampling Theorem. Modify to

agree with notation in notes, see Solution ?? below.)
Exercise 34.15. .Folland 8.19 on p. 256. (The Fourier transform of a function

whose support has finite measure.)
Exercise 34.16. Folland 8.22 on p. 256. (Bessel functions.)
Exercise 34.17. Folland 8.23 on p. 256. (Hermite Polynomial problems and

Harmonic oscillators.)
Exercise 34.18. Folland 8.31 on p. 263. (Poisson Summation formula prob-

lem.)
35
Constant Coefficient partial differential equations
aα ξ α with aα ∈ C and 35.1 Elliptic examples

P
Suppose that p(ξ) = |α|≤k
α
1 As a specific example consider the equation
L = p(Dx ) := Σ|α|≤N aα Dxα = Σ|α|≤N aα ∂x . (35.1)
i
−∆ + m2 f = g

(35.2)
Then for f ∈ S Pn
c (ξ) = p(ξ)fˆ(ξ),
Lf where f, g : Rn → C and ∆ = i=1 ∂ 2 /∂x2i is the usual Laplacian on Rn . By
Corollary 34.16 (i.e. taking the Fourier transform of this equation), solving Eq.
that is to say the Fourier transform takes a constant coefficient partial differ- (35.2) with f, g ∈ L2 is equivalent to solving
ential operator to multiplication by a polynomial. This fact can often be used
to solve constant coefficient partial differential equation. For example suppose 2
|ξ| + m2 fˆ(ξ) = ĝ(ξ). (35.3)
g : Rn → C is a given function and we want to find a solution to the equation
Lf = g. Taking the Fourier transform of both sides of the equation Lf = g The unique solution to this latter equation is
would imply p(ξ)fˆ(ξ) = ĝ(ξ) and therefore fˆ(ξ) = ĝ(ξ)/p(ξ) provided p(ξ) is
never zero. (We will discuss what happens when p(ξ) has zeros a bit more later −1
2
fˆ(ξ) = |ξ| + m2 ĝ(ξ)
on.) So we should expect
and therefore,

−1 1 −1 1
f (x) = F ĝ(ξ) (x) = F Fg(x).
p(ξ) p(ξ) −1
−1
2
f (x) = F −1 |ξ| + m2 ĝ(ξ) (x) =: −∆ + m2 g(x).
Definition 35.1. Let L = p(Dx ) as in Eq. (35.1). Then we let
σ(L) :=Ran(p) ⊂ C and call σ(L) the spectrum of L. Given a measur-
We expect
able function G : σ(L) → C, we define (a possibly unbounded operator)
G(L) : L2 (Rn , m) → L2 (Rn , m) by
2
−1 Z
F −1 |ξ| + m2 ĝ(ξ) (x) = Gm Fg(x) = Gm (x − y)g(y)dy,
G(L)f := F −1 MG◦p F Rn
where
where MG◦p denotes the operation on L2 (Rn , m) of multiplication by G ◦ p, i.e. −1 Z
−1

2 2 1 iξ·x
Gm (x) := F |ξ| + m (x) = 2e dξ.
Rn m2 + |ξ|
MG◦p f = (G ◦ p) f
−1
2
2 2 At the moment F −1 |ξ| + m2 only makes sense when n = 1, 2, or 3 because
with domain given by those f ∈ L such that (G ◦ p) f ∈ L . −1
2
only then is |ξ| + m2 ∈ L2 (Rn ).
At a formal level we expect
For now we will restrict our attention to the one dimensional case, n = 1,
G(L)f = F −1 (G ◦ p) Fg. in which case Z
1 1
Gm (x) = √ eiξx dξ. (35.4)
2π R (ξ + mi) (ξ − mi)
398 35 Constant Coefficient partial differential equations
Z x Z ∞
The function Gm may be computed using standard complex variable contour 1 1
f (x) := − (x − y) g(y)dy − (y − x) g(y)dy
integration methods to find, for x ≥ 0, 2 −∞ 2 x
2
1 ei mx 1 √ so that
Gm (x) = √ 2πi = 2πe−mx
2π 2im 2m 1 x 1 ∞
Z Z
0
f (x) = − g(y)dy + g(y)dy and
and since Gm is an even function, 2 −∞ 2 x
1 1
−1 √ f 00 (x) = − g(x) − g(x).
−1

2 2 2π −m|x| 2 2
Gm (x) = F |ξ| + m (x) = e . (35.5)
2m
This result is easily verified to be correct, since 35.2 Poisson Semi-Group
"√ # √ Z
2π −m|x| 2π Let us now consider the problems of finding a function (x0 , x) ∈ [0, ∞) × Rn →
F e (ξ) = e−m|x| e−ix·ξ dx
2m 2m R u(x0 , x) ∈ C such that
Z ∞ Z 0
1 −mx −ix·ξ mx −ix·ξ
2
∂

= e e dx + e e dx + ∆ u = 0 with u(0, ·) = f ∈ L2 (Rn ). (35.6)
2m 0 −∞ ∂x20

1 1 1 1
= + = 2 . Let û(x0 , ξ) := Rn u(x0 , x)e−ix·ξ dx denote the Fourier transform of u in the
R
2m m + iξ m − iξ m + ξ2
x ∈ Rn variable. Then Eq. (35.6) becomes
Hence in conclusion we find that −∆ + m2 f = g has solution given by 2
∂ 2
√ − |ξ| û(x0 , ξ) = 0 with û(0, ξ) = fˆ(ξ) (35.7)
∂x20
Z Z
2π −m|x−y| 1
f (x) = Gm Fg(x) = e g(y)dy = e−m|x−y| g(y)dy.
2m R 2m R
and the general solution to this differential equation ignoring the initial condi-
Question. Why do we get a unique answer here given that f (x) = tion is of the form
A sinh(x) + B cosh(x) solves
û(x0 , ξ) = A(ξ)e−x0 |ξ| + B(ξ)ex0 |ξ| (35.8)
−∆ + m2 f = 0?

for some function A(ξ) and B(ξ). Let us now impose the extra condition that
The answer is that such an f is not in L2 unless f = 0! More generally it is u(x0 , ·) ∈ L2 (Rn ) or equivalently that û(x0 , ·) ∈ L2 (Rn ) for all x0 ≥ 0. The
worth noting that A sinh(x) + B cosh(x) is not in P unless A = B = 0. solution in Eq. (35.8) will not have this property unless B(ξ) decays very rapidly
What about when m = 0 in which case m2 + ξ 2 becomes ξ 2 which has a at ∞. The simplest way to achieve this is to assume B = 0 in which case we
zero at 0. Noting that constants are solutions to ∆f = 0, we might look at now get a unique solution to Eq. (35.7), namely
√ √
2π −m|x| 2π û(x0 , ξ) = fˆ(ξ)e−x0 |ξ| .
lim (Gm (x) − 1) = lim (e − 1) = − |x| .
m↓0 m↓0 2m 2
Applying the inverse Fourier transform gives
as a solution, i.e. we might conjecture that h i √
Z u(x0 , x) = F −1 fˆ(ξ)e−x0 |ξ| (x) =: e−x0 −∆ f (x)
1
f (x) := − |x − y| g(y)dy
2 R and moreover √
e−x0 −∆
f (x) = Px0 ∗ f (x)
solves the equation −f 00 = g. To verify this we have

35.3 Heat Equation on Rn 399
−n/2
F −1 e−x0 |ξ|

where Px0 (x) = (2π) (x). From Exercise 35.1, Theorem 35.3. Let
x0 −n/2 −|x−y|2 /2t
Px0 (x) = (2π)
−n/2
F −1 e−x0 |ξ| (x) = cn 2 ρ(t, x, y) := (2πt) e (35.11)
2
(x0 + |x| )(n+1)/2
be the heat kernel on Rn . Then
where
−n/2 Γ ((n + 1)/2) Γ ((n + 1)/2) 1
cn = (2π) √ n/2 = n (n+1)/2 . ∂t − ∆x ρ(t, x, y) = 0 and lim ρ(t, x, y) = δx (y), (35.12)
π2 2 π 2 t↓0
Hence we have proved the following proposition. where δx is the δ – function at x in Rn . More precisely, if f is a continuous
bounded (can be relaxed considerably) function on Rn , then
Proposition 35.2. For f ∈ L2 (Rn ),
Z
√
e−x0 −∆
f = Px0 ∗ f for all x0 ≥ 0 u(t, x) = ρ(t, x, y)f (y)dy
Rn
√
and the function u(x0 , x) := e−x0 −∆
f (x) is C ∞ for (x0 , x) ∈ (0, ∞) × Rn and is a solution to Eq. (35.9) where u(0, x) := limt↓0 u(t, x).
solves Eq. (35.6).
Proof. Direct computations show that ∂t − 21 ∆x ρ(t, x, y) = 0 and an

application Rof Theorem 22.32 shows limt↓0 ρ(t, x, y) = δx (y) or equivalently

35.3 Heat Equation on Rn that limt↓0 Rn ρ(t, x, y)f (y)dy = f (x) uniformly on compact subsets of Rn .
This shows that limt↓0 u(t, x) = f (x) uniformly on compact subsets of Rn .
The heat equation for a function u : R+ × Rn → C is the partial differential This notation suggests that we should be able to compute the solution to g
equation to (∆ − m2 )g = f using
1 Z ∞
∂t − ∆ u = 0 with u(0, x) = f (x), (35.9) −1 2

2 g(x) = m2 − ∆ f (x) = e−(m −∆)t f (x)dt
0
where f is a given function on Rn . By Fourier transforming Eq. (35.9) in the x Z ∞
−m2 t
– variables only, one finds that (35.9) implies that = e p2t Ff (x)dt,
0

1 2
∂t + |ξ| û(t, ξ) = 0 with û(0, ξ) = fˆ(ξ). (35.10) a fact which is easily verified using the Fourier transform. This gives us a method
2 to compute Gm (x) from the previous section, namely
2
and hence that û(t, ξ) = e−t|ξ| /2 fˆ(ξ). Inverting the Fourier transform then Z ∞
−m2 t
Z ∞
2 1 2
shows that Gm (x) = e p2t (x)dt = (2t)−n/2 e−m t− 4t |x| dt.

0 0
2
2

u(t, x) = F −1 e−t|ξ| /2 fˆ(ξ) (x) = F −1 e−t|ξ| /2 Ff (x) =: et∆/2 f (x). 2
We make the change of variables, λ = |x| /4t (t = |x| /4λ, dt = − |x|
2 2
4λ2 dλ) to
find
From Example 34.4,
2 −n/2
!
Z ∞ Z ∞ 2
|x| |x|
2
1 2
F −1 e−t|ξ| /2 (x) = pt (x) = t−n/2 e− 2t |x| −n/2 −m2 t− 4t |x|2
1 2 2
Gm (x) = (2t) e dt = e−m |x| /4λ−λ 2 dλ
0 0 2λ (2λ)
and therefore, 2(n/2−2) ∞ n/2−2 −λ −m2 |x|2 /4λ
Z
= λ e e dλ. (35.13)
Z
n−2
u(t, x) = pt (x − y)f (y)dy. |x| 0
Rn
This suggests the following theorem. In case n = 3, Eq. (35.13) becomes

√ Z ∞ √
π 1 −λ −m2 |x|2 /4λ π −m|x| fy0 (λ) = y 2 − λ−2 and fy00 (λ) = 2λ−3 and fy000 (λ) = −6λ−4

Gm (x) = √ √ e e dλ = √ e
2 |x| 0 πλ 2 |x|
so by Taylor’s theorem with remainder we learn
where the last equality follows from Exercise 35.1. Hence when n = 3 we have
found fy (λ) ∼
= 2y + y 3 (λ − y −1 )2 for all λ > 0,
Z √
−1 −3/2 π see Figure 35.3 below.
2
m −∆ f (x) = Gm Ff (x) = (2π) √ e−m|x−y| f (y)dy
R3 2 |x − y|
Z
1
= e−m|x−y| f (y)dy. (35.14)
R3 4π |x − y|
1
The function 4π|x| e−m|x| is called the Yukawa potential.
Let us work out Gm (x) for n odd. By differentiating Eq. (35.27) of Exercise
35.1 we find
Z ∞ Z ∞ k
1
k−1/2 − 4λ x2 −λm2 1 − 1 x2 d
dλλ e e = dλ √ e 4λ − e−λa |a=m2
0 0 λ da
k √
d π √
= − √ e− ax = pm,k (x)e−mx
da a
where pm,k (x) is a polynomial in x with deg pm = k with Plot of f4 and its second order Taylor approximation.
√
k
√ 13 So by the usual asymptotics arguments,
d 2k − 1 2k+1
pm,k (0) = π − a−1/2 |a=m2 = π( ... )m
da 22 2
Z
2 −1
√ ψ(y) ∼
= y n−2 λn/2−2 e−(λy +λ ) dλ
= m2k+1 π2−k (2k − 1)!!. (−ε+y −1 ,y −1 +ε)
Z
∼
= y n−2 λn/2−2 exp −2y − y 3 (λ − y −1 )2 dλ

Letting k − 1/2 = n/2 − 2 and m = 1 we find k = n−1 2 − 2 ∈ N for n = 3, 5, . . . .
(−ε+y −1 ,y −1 +ε)
and we find Z
∼ n−2 −2y
λn/2−2 exp −y 3 (λ − y −1 )2 dλ (let λ → λy −1 )
Z ∞
1 2 = y e
λn/2−2 e− 4λ x e−λ dλ = p1,k (x)e−x for all x > 0. R
Z
0
−2y n−2 −n/2+1
λn/2−2 exp −y(λ − 1)2 dλ

=e y y
Therefore, ZR
= e−2y y n−2 y −n/2+1 (λ + 1)n/2−2 exp −yλ2 dλ.

∞
2(n/2−2) 2(n/2−2)
Z
2
|x|2 /4λ
Gm (x) = n−2 λn/2−2 e−λ e−m dλ = n−2 p1,n/2−2 (m |x|)e−m|x| . R
|x| 0 |x|
The point is we are still going to get exponential decay at ∞.
Now for even m, I think we get Bessel functions in the answer. (BRUCE: When m = 0, Eq. (35.13) becomes
look this up.) Let us at least work out the asymptotics of Gm (x) for x → ∞.
∞
2(n/2−2) 2(n/2−2)
Z
To this end let dλ
G0 (x) = n−2 λn/2−1 e−λ = n−2 Γ (n/2 − 1)
Z ∞
−1 2
Z ∞
2 −1 |x| 0 λ |x|
ψ(y) := λn/2−2 e−(λ+λ y ) dλ = y n−2 λn/2−2 e−(λy +λ ) dλ
0 0 where Γ (x) in the gamma function defined in Eq. (20.42). Hence for “reason-
The function fy (λ) := (y λ + λ 2 −1
) satisfies, able” functions f (and n 6= 2) we expect that (see Proposition 35.4 below)

35.4 Wave Equation on Rn 401
Z
1 where M is a bound on f and supp(f ) ⊂ B (0, R) . Similarly, |x| ≤ r, we have
(−∆)−1 f (x) = G0 Ff (x) = 2(n/2−2) Γ (n/2 − 1)(2π)−n/2 n−2 f (y)dy
Rn |x − y|
1 1
sup |f (x − y)| ≤ M 1{|y|≤R+r} ∈ L1 (dy) ,
Z
1 1 n−2 n−2
= Γ (n/2 − 1) n−2 f (y)dy. |x|≤r |y| |y|
4π n/2 R |x − y|
n
from which it follows that G ∗ f is a continuous function. Similar arguments

The function show if f ∈ Cc2 (Rn ) , then G ∗ f ∈ C 2 (Rn ) and ∆ (G ∗ f ) = G ∗ ∆f. So to finish
1 1
G(x) := n/2
Γ (n/2 − 1) n−2 (35.15) the proof it suffices to show G ∗ ∆u = u.
4π |x| For this we now write, making use of Fubini-Tonelli, integration by parts,
is a “Green’s function” for −∆. Recall from Exercise 20.16 that, for n = 2k, the fact that ∂t ρt (y) = 21 ∆ρt (y) and the dominated convergence theorem,
Γ ( n2 − 1) = Γ (k − 1) = (k − 2)!, and for n = 2k + 1, Z Z ∞
1
G ∗ ∆u (x) = ∆u (x − y) ρt (y) dt dy
n √ 1 · 3 · 5 · · · · · (2k − 3) 2 Rn 0
Γ( − 1) = Γ (k − 1/2) = Γ (k − 1 + 1/2) = π Z ∞ Z
2 2k−1 1
√ (2k − 3)!! = dt ∆u (x − y) ρt (y) dy
= π where (−1)!! =: 1. 2 0 n
2k−1 Z ∞ ZR
1
= dt ∆y u (x − y) ρt (y) dy
Hence 2 0 Rn
1
1 1 πk
(k − 2)! if n = 2k 1 ∞
Z Z
G(x) = 1 (2k−3)!! = dt u (x − y) ∆y ρt (y) dy
4 |x|n−2 π k 2k−1
if n = 2k + 1 2 0 Rn
Z ∞ Z
and in particular when n = 3, d
= dt u (x − y) ρt (y) dy
1 1 0 n dt
G(x) = Z ∞R Z
4π |x| d
= lim dt u (x − y) ρt (y) dy
ε↓0 ε R n dt
which is consistent with Eq. (35.14) with m = 0. Z Z ∞
d

= lim u (x − y) ρt (y) dt dy
Proposition 35.4. Let n ≥ 3 and for x ∈ 6 Rn , let ρt (x) = ρ (t, x, 0) := ε↓0 Rn ε dt
Z
1 n/2 − 2t
1 2
e |x| (see Eq. (35.11))and G (x) be as in Eq. (35.15) so that

2πt = − lim u (x − y) ρε (y) dy = u (x) ,
ε↓0 Rn
1 ∞
Z
Cn
G (x) := n−2 = 2 ρt (x) dt for x 6= 0. where in the last equality we have used the fact that ρt is an approximate δ –
|x| 0 sequence.
Then
−∆ (G ∗ u) = −G ∗ ∆u = u
35.4 Wave Equation on Rn
for all u ∈ Cc2 n
(R ) .
Let us now consider the wave equation on Rn ,
Proof. For f ∈ Cc (Rn ) ,
0 = ∂t2 − ∆ u(t, x) with

Z
1
G ∗ f (x) = Cn f (x − y) n−2 dy u(0, x) = f (x) and ut (0, x) = g(x). (35.16)
Rn |y|
Taking the Fourier transform in the x variables gives the following equation
is well defined, since
2
Z Z 0 = ût t (t, ξ) + |ξ| û(t, ξ) with
1 1
|f (x − y)| n−2 dy ≤M n−2 dy <∞ û(0, ξ) = fˆ(ξ) and ût (0, ξ) = ĝ(ξ). (35.17)
Rn |y| |y|≤R+|x| |y|

The solution to these equations is We can arrive at this same solution by more elementary means as follows.
sin t |ξ| We first note in the one dimensional case that wave operator factors, namely
û(t, ξ) = fˆ(ξ) cos (t |ξ|) + ĝ(ξ)
|ξ| 0 = ∂t2 − ∂x2 u(t, x) = (∂t − ∂x ) (∂t + ∂x ) u(t, x).

and hence we should have

sin t |ξ|
Let U (t, x) := (∂t + ∂x ) u(t, x), then the wave equation states (∂t − ∂x ) U = 0
u(t, x) = F −1 fˆ(ξ) cos (t |ξ|) + ĝ(ξ) (x) and hence by the chain rule dt d
U (t, x − t) = 0. So
|ξ|
sin t |ξ| U (t, x − t) = U (0, x) = g(x) + f 0 (x)
= F −1 cos (t |ξ|) Ff (x) + F −1 Fg (x)
|ξ|

d −1 sin t |ξ|
and replacing x by x + t in this equation shows
−1 sin t |ξ|
= F Ff (x) + F Fg (x) . (35.18)
dt |ξ| |ξ|
(∂t + ∂x ) u(t, x) = U (t, x) = g(x + t) + f 0 (x + t).
The question now is how interpret this equation. In particular what are
the inverse Fourier transforms of F −1 cos (t |ξ|) and F −1 sin|ξ|t|ξ| . Since Working similarly, we learn that
−1 sin t|ξ|
d
dt F |ξ| hFf (x)i = F −1 cos (t |ξ|)Ff (x), it really suffices to under- d
u(t, x + t) = g(x + 2t) + f 0 (x + 2t)
stand F −1 sin|ξ|t|ξ| . The problem we immediately run into here is that dt
sin t|ξ| which upon integration implies
|ξ| ∈ L2 (Rn ) iff n
= 1 so that is the case we should start with.
Again by complex contour integration methods one can show Z t
π u(t, x + t) = u(0, x) + {g(x + 2τ ) + f 0 (x + 2τ )} dτ
F −1 ξ −1 sin tξ (x) = √

1x+t>0 − 1(x−t)>0 0
2π Z t
π π 1
= √ (1x>−t − 1x>t ) = √ 1[−t,t] (x) = f (x) + g(x + 2τ )dτ + f (x + 2τ )|t0
2π 2π 0 2
Z t
where in writing the last line we have assume that t ≥ 0. Again this easily seen 1
= (f (x) + f (x + 2t)) + g(x + 2τ )dτ.
to be correct because 2 0
Z
π 1 1 −iξ·x t Replacing x → x − t in this equation gives
F √ 1[−t,t] (x) (ξ) = 1[−t,t] (x)e−iξ·x dx = e |−t
2π 2 R −2iξ
Z t
1 iξt 1
e − e−iξt = ξ −1 sin tξ.

= u(t, x) = (f (x − t) + f (x + t)) + g(x − t + 2τ )dτ
2iξ 2 0
Therefore, and then letting y = x − t + 2τ in the last integral shows again that
1 t
Z
−1 −1

F ξ sin tξ Ff (x) = f (x − y)dy
2 −t 1 x+t
Z
1
u(t, x) = (f (x − t) + f (x + t)) + g(y)dy.
and the solution to the one dimensional wave equation is 2 2 x−t
d 1 t 1 t
Z Z h i
u(t, x) =
dt 2 −t
f (x − y)dy +
2 −t
g(x − y)dy When n > 3 it is necessary to treat F −1 sin|ξ|t|ξ| as a “distribution” or
“generalized function,” see Section 36 below. So for now let us take n = 3, in
1 t
Z
1
= (f (x − t) + f (x + t)) + g(x − y)dy which case from Example 34.18 it follows that
2 2 −t

1 x+t −1 sin t |ξ| t
Z
1 F = σt = tσ̄t (35.19)
= (f (x − t) + f (x + t)) + g(y)dy. |ξ| 4πt2
2 2 x−t

35.4 Wave Equation on Rn 403
1
where σ̄t is 4πt2 σt , the surface measure on St normalized to have total measure wherein we have made use of Example 34.18. This completes the proof since
one. Hence from Eq. (35.18) the solution to the three dimensional wave equation û(t, ξ) solves Eq. (35.17) as desired.
should be given by Proof 2. Differentiating
Z
d
u(t, x) = (tσ̄t Ff (x)) + tσ̄t Fg (x) . (35.20) S(t, x) := g(x + tω)dσ̄1 (ω)
dt S1
Using this definition in Eq. (35.20) gives in t gives

Z Z
d
Z
1
u(t, x) = t f (x − y)dσ̄t (y) + t g(x − y)dσ̄t (y) St (t, x) = ∇g(x + tω) · ωdσ(ω)
dt St St 4π S1
Z Z Z
d 1
= t f (x − tω)dσ̄1 (ω) + t g(x − tω)dσ̄1 (ω) = ∇ω · ∇g(x + tω)dm(ω)
dt S S 4π B(0,1)
Z 1 Z 1 Z
d t
= t f (x + tω)dσ̄1 (ω) + t g(x + tω)dσ̄1 (ω). (35.21) = ∆g(x + tω)dm(ω)
dt S1 S1 4π B(0,1)
Z
1
Proposition 35.5. Suppose f ∈ C 3 (R3 ) and g ∈ C 2 (R3 ), then u(t, x) defined = ∆g(x + y)dm(y)
by Eq. (35.21) is in C 2 R × R3 and is a classical solution of the wave equation 4πt2 B(0,t)
Z t Z
in Eq. (35.16). 1 2
= dr r ∆g(x + y)dσ(y)
4πt2 0 |y|=r
Proof. The fact that u ∈ C 2 R × R3 follows by the usual differentiation
under the integral arguments. Suppose we can prove the proposition R in the where we have used the divergence theorem, made the change of variables y = tω
special case that f ≡ 0. Then for f ∈ C 3 (R3 ), the function v(t, x) = +t S1 g(x+ and used the disintegration formula in Eq. (20.34),

tω)dσ̄1 (ω) solves the wave equation 0 = ∂t2 − ∆ v(t, x) with v(0, x) = 0 and Z Z Z ∞ Z
vt (0, x) = g(x). Differentiating the wave equation in t shows u = vt also solves f (x)dm(x) = f (r ω) dσ(ω)r n−1
dr = dr f (y)dσ(y).
the wave equation with u(0, x) = g(x) and ut (0, x) = vtt (0, x) = −∆x v(0, x) = 0 |y|=r
Rd [0,∞)×S n−1
0. These remarks reduced the problems to showing u in Eq. (35.21) with f ≡ 0
solves the wave equation. So let Since u(t, x) = tS(t, x) if follows that
Z
u(t, x) := t g(x + tω)dσ̄1 (ω). (35.22) ∂
utt (t, x) = [S(t, x) + tSt (t, x)]
S1 ∂t " #
Z t Z
We now give two proofs the u solves the wave equation. ∂ 1 2
= St (t, x) + dr r ∆g(x + y)dσ(y)
Proof 1. Since solving the wave equation is a local statement and u(t, x) ∂t 4πt 0 |y|=r
only depends on the values of g in B(x, t) we it suffices to consider the case Z t Z
1
where g ∈ Cc2 R3 . Taking the Fourier transform of Eq. (35.22) in the x variable = St (t, x) − dr ∆g(x + y)dσ(y)
4πt2 0 |y|=r
shows Z
Z Z 1
+ ∆g(x + y)dσ(y)
û(t, ξ) = t dσ̄1 (ω) g(x + tω)e−iξ·x dx 4πt |y|=t
S1 R3 Z
Z Z Z t
−iξ·x itω·ξ = St (t, x) − St (t, x) + ∆g(x + tω)dσ(ω)
=t dσ̄1 (ω) g(x)e e dx = ĝ(ξ)t eitω·ξ dσ̄1 (ω) 4πt2 |y|=1
S1 R3 S1
sin |tk| sin (t |ξ|) = t∆u(t, x)
= ĝ(ξ)t = ĝ(ξ)
|tk| |ξ| as required.

Fig. 35.1. The geometry of the solution to the wave equation in three dimensions.
The observer sees a flash at t = 0 and x = 0 only at time t = |x| . The wave progates
sharply with speed 1.
The solution in Eq. (35.21) exhibits a basic property of wave equations,

namely finite propagation speed. To exhibit the finite propagation speed, sup-
pose that f = 0 (for simplicity) and g has compact support near the origin, for Fig. 35.2. The geometry of the solution to the wave equation in two dimensions. A
example think of g = δ0 (x). Then x + tw = 0 for some w iff |x| = t. Hence the flash at 0 ∈ R2 looks like a line of flashes to the fictitious 3 – d observer and hence she
“wave front” propagates at unit speed and the wave front is sharp. See Figure sees the effect of the flash for t ≥ |x| . The wave still propagates with speed 1. However
35.1 below. there is no longer sharp propagation of the wave front, similar to water waves.
The solution of the two dimensional wave equation may be found using
“Hadamard’s method of decent” which we now describe. Suppose now that f
and g are functions on R2 which we may view as functions on R3 which happen Proof. As usual it suffices to consider the case where f ≡ 0. By symmetry
not to depend on the third coordinate. We now go ahead and solve the three u may be written as
dimensional wave equation using Eq. (35.21) and f and g as initial conditions. Z Z
It is easily seen that the solution u(t, x, y, z) is again independent of z and hence
u(t, x) = 2t g(x − y)dσ̄t (y) = 2t g(x + y)dσ̄t (y)
is a solution to the two dimensional wave equation. See figure 35.2 below. St+ St+
Notice that we still have finite speed of propagation but no longer sharp
propagation. The explicit formula for u is given in the next proposition. where St+ is the portion +
√ of St with z ≥ 0. The surface St may be 2parametrized
by R(u, v) = (u, v, t2 − u2 − v 2 ) with (u, v) ∈ Dt := (u, v) : u + v 2 ≤ t2 .
Proposition 35.6. Suppose f ∈ C 3 (R2 ) and g ∈ C 2 (R2 ), then In these coordinates we have
" ZZ # p p
∂ t f (x + tw) 4πt2 dσ̄t = −∂u t2 − u2 − v 2 , −∂v t2 − u2 − v 2 , 1 dudv

u(t, x) := p dm(w)
∂t 2π D1 1 − |w|2
u v
t
ZZ
g(x + tw) = √ ,√ , 1 dudv
2
t −u −v2 2 2 2
t −u −v 2
+ p dm(w)
2π D1 1 − |w|2 r
2 2
u +v |t|
= + 1dudv = √ dudv
t2 − u 2 − v 2 t − u2 − v 2
2

is in C 2 R × R2 and solves the wave equation in Eq. (35.16).

35.5 Elliptic Regularity 405
and therefore,
|t|
Z
2t p
u(t, x) = g(x + (u, v, t2 − u2 − v 2 )) √ dudv
4πt2 Dt t − u2 − v 2
2
Z
1 g(x + (u, v))
= sgn(t) √ dudv.
2π Dt t2 − u 2 − v 2
This may be written as
ZZ
1 g(x + w)
u(t, x) = sgn(t) p dm(w)
2π Dt t2 − |w|2
2 ZZ
1 t g(x + tw)
= sgn(t) p dm(w)
2π |t| D1 1 − |w|2 Fig. 35.3. The region M and the cutoff functions, θ and α.
ZZ
1 g(x + tw)
= t p dm(w)
2π D1 1 − |w|2
which makes sense for x near 0. Therefore we find
Z
u(x) = G ∗ (θv) (x) − β(y)∆y [G(x − y)θ(y)] · u(y)dy.
Rn
35.5 Elliptic Regularity
Clearly all of the above manipulations were correct if we know u were C 2 to
∞
The following theorem is a special case of the main theorem (Theorem 35.11) begin with. So for the general case, let un = u ∗ δn with {δn }n=1 – the usual
of this section. sort of δ – sequence approximation. Then ∆un = v ∗ δn =: vn away from ∂M
and
Theorem 35.7. Suppose that M ⊂o Rn , v ∈ C ∞ (M ) and u ∈ L1loc (M ) satisfies
∆u = v weakly, then u has a (necessarily unique) version ũ ∈ C ∞ (M ).
Z
un (x) = G ∗ (θvn ) (x) − β(y)∆y [G(x − y)θ(y)] · un (y)dy. (35.23)
Proof. We may always assume n ≥ 3, by embedding the n = 1 and n = 2 Rn
cases in the n = 3 cases. For notational simplicity, assume 0 ∈ M and we will
Since un → u in L1loc (O) where O is a sufficiently small neighborhood of 0, we
show u is smooth near 0. To this end let θ ∈ Cc∞ (M ) such that θ = 1 in a
may pass to the limit in Eq. (35.23) to find u(x) = ũ(x) for a.e. x ∈ O where
neighborhood of 0 and α ∈ Cc∞ (M ) such that supp(α) ⊂ {θ = 1} and α = 1
in a neighborhood of 0 as well, see Figure 35.3 Then formally, we have with
Z
β := 1 − α, ũ(x) := G ∗ (θv) (x) − β(y)∆y [G(x − y)θ(y)] · u(y)dy.
Rn
G ∗ (θv) = G ∗ (θ∆u) = G ∗ (θ∆(αu + βu)) This concluded the proof since ũ is smooth for x near 0.
= G ∗ (∆(αu) + θ∆(βu)) = αu + G ∗ (θ∆(βu))
DefinitionP 35.8. We say L = p(Dx ) as defined in Eq. (35.1) is elliptic if
so that pk (ξ) := |α|=k aα ξ α is zero iff ξ = 0. We will also say the polynomial p(ξ) :=
u(x) = G ∗ (θv) (x) − G ∗ (θ∆(βu))(x) P α
|α|≤k aα ξ is elliptic if this condition holds.
for x ∈ supp(α). The last term is formally given by
Remark 35.9. If p(ξ) := |α|≤k aα ξ α is an elliptic polynomial, then there exists
P
Z
G ∗ (θ∆(βu))(x) = G(x − y)θ(y)∆(β(y)u(y))dy A < ∞ such that inf |ξ|≥A |p(ξ)| > 0. Since pk (ξ) is everywhere non-zero for
ZR
n
ξ ∈ S n−1 and S n−1 ⊂ Rn is compact, ε := inf |ξ|=1 |pk (ξ)| > 0. By homogeneity
= β(y)∆y [G(x − y)θ(y)] · u(y)dy this implies
k
Rn |pk (ξ)| ≥ ε |ξ| for all ξ ∈ An .

Since 3. There exists G ∈ Cc∞ (Rn \ {0}) such that for all multi-indices α,

X

limM →∞ ∂ α GM (x) = ∂ α G(x) uniformly on compact subsets in Rn \ {0} .
X
aα ξ α ≥ |pk (ξ)| − aα ξ α

|p(ξ)| = pk (ξ) + Proof. We have already proved the first two items. For item 3., we notice
|α|<k |α|<k that
(1 − χ(ξ)) χ(ξ/M )ξ α
Z
k k−1 β β
≥ ε |ξ| − C 1 + |ξ| α
(−x) D GM (x) = (−D)ξ eix·ξ dξ
Rn p(ξ)
α

β (1 − χ(ξ)) ξ
Z
for some constant C < ∞ from which it is easily seen that for A sufficiently
= Dξ χ(ξ/M ) eix·ξ dξ
large, Rn p(ξ)
ε k
|p(ξ)| ≥ |ξ| for all |ξ| ≥ A. (1 − χ(ξ)) ξ α
Z
2 = Dξβ · χ(ξ/M )eix·ξ dξ + RM (x)
Rn p(ξ)
For the rest of this section, let L = p(Dx ) be an elliptic operator and M ⊂0
Rn . As mentioned at the beginning of this section, the formal solution to Lu = v where
for v ∈ L2 (Rn ) is given by X β (1 − χ(ξ)) ξ α
Z
M |γ|−|β| Dξγ · Dβ−γ χ (ξ/M )eix·ξ dξ.

RM (x) =
γ Rn p(ξ)
u = L−1 v = G ∗ v γ<β
Using
where γ ξα

≤ C |ξ||α|−m−|γ|
Z
1 ix·ξ D (1 − χ(ξ))
G(x) := e dξ. ξ p(ξ)
Rn p(ξ)
and the fact that
Of course this integral may not be convergent because of the possible zeros of
1
supp( Dβ−γ χ (ξ/M )) ⊂ {ξ ∈ Rn : A ≤ |ξ| /M ≤ 2A}

p and the fact p(ξ) may not decay fast enough at infinity. We we will introduce
a smooth cut off function χ(ξ) which is 1 on C0 (A) := {x ∈ Rn : |x| ≤ A} and = {ξ ∈ Rn : AM ≤ |ξ| ≤ 2AM }
supp(χ) ⊂ C0 (2A) where A is as in Remark 35.9. Then for M > 0 let
we easily estimate
(1 − χ(ξ)) χ(ξ/M ) ix·ξ
Z
GM (x) = e dξ, (35.24) X β Z
|α|−m−|γ|
R n p(ξ) |RM (x)| ≤ C M |γ|−|β| |ξ| dξ
Z γ {ξ∈Rn :AM ≤|ξ|≤2AM }
γ<β
δ(x) := χ∨ (x) = χ(ξ)eix·ξ dξ, and δM (x) = M n δ(M x). (35.25) X β
Rn ≤C M |γ|−|β| M |α|−m−|γ|+n = CM |α|−|β|−m+n .
γ
γ<β
R
Notice δ(x)dx = Fδ(0) = χ(0) = 1, δ ∈ S since χ ∈ S and
Rn
Z Z Therefore, RM → 0 uniformly in x as M → ∞ provided |β| > |α| − m + n. It
LGM (x) = (1 − χ(ξ)) χ(ξ/M )eix·ξ dξ = [χ(ξ/M ) − χ(ξ)] eix·ξ dξ follows easily now that GM → G in Cc∞ (Rn \ {0}) and furthermore that
Rn Rn
(1 − χ(ξ)) ξ α ix·ξ
Z
β
= δM (x) − δ(x) (−x) Dα G(x) = Dξβ · e dξ
Rn p(ξ)
provided M > 2. provided β is sufficiently large. In particular we have shown,
Proposition 35.10. Let p be an elliptic polynomial of degree m. The function α
k (1 − χ(ξ)) ξ
Z
1
GM defined in Eq. (35.24) satisfies the following properties, Dα G(x) = 2k
(−∆ξ ) · eix·ξ dξ
|x| Rn p(ξ)
1. GM ∈ S for all M > 0. provided m − |α| + 2k > n, i.e. k > (n − m + |α|) /2. We are now ready to use
2. LGM (x) = M n δ(M x) − δ(x). this result to prove elliptic regularity for the constant coefficient case.

35.6 Exercises 407
Theorem 35.11. Suppose L = p(Dξ ) is an elliptic differential operator on Rn , Therefore we have shown,
M ⊂o Rn , v ∈ C ∞ (M ) and u ∈ L1loc (M ) satisfies Lu = v weakly, then u has a Z
(necessarily unique) version ũ ∈ C ∞ (M ). u(x) = G ∗ (θv) (x) − L∗y {θ(y)G(x − y)} · (βu) (y)dy + δ ∗ (αu) (x)
Rn
Proof. For notational simplicity, assume 0 ∈ M and we will show u is
smooth near 0. To this end let θ ∈ Cc∞ (M ) such that θ = 1 in a neighborhood of for almost every x in a neighborhood of 0. (Again it suffices to prove this
0 and α ∈ Cc∞ (M ) such that supp(α) ⊂ {θ = 1} , and α = 1 in a neighborhood equation and in particular Eq. (35.26) assuming u ∈ C 2 (M ) because of the
of 0 as well. Then formally, we have with β := 1 − α, same convolution argument we have use above.) Since the right side of this
equation is the linear combination of smooth functions we have shown u has a
GM ∗ (θv) = GM ∗ (θLu) = GM ∗ (θL(αu + βu))
smooth version in a neighborhood of 0.
= GM ∗ (L(αu) + θL(βu))
Remarks 35.12 We could avoid introducing GM (x) if deg(p) > n, in which
= δM ∗ (αu) − δ ∗ (αu) + GM ∗ (θL(βu))
case (1−χ(ξ))
p(ξ) ∈ L1 and so
so that
(1 − χ(ξ)) ix·ξ
Z
δM ∗ (αu) (x) = GM ∗ (θv) (x) − GM ∗ (θL(βu))(x) + δ ∗ (αu) . (35.26) G(x) := e dξ
Rn p(ξ)
Since
(1 − χ(ξ)) χ(ξ/M ) is already well defined function with G ∈ C ∞ (Rn \{0})∩BC(Rn ). If deg(p) < n,
ˆ ˆ
F [GM ∗ (θv)] (ξ) = ĜM (ξ) (θv) (ξ) = (θv) (ξ) k
we may consider the operator Lk = [p(Dx )] = pk (Dx ) where k is chosen so
p(ξ)
that k · deg(p) > n. Since Lu = v implies Lk u = Lk−1 v weakly, we see to prove
(1 − χ(ξ)) ˆ
→ (θv) (ξ) as M → ∞ the hypoellipticity of L it suffices to prove the hypoellipticity of Lk .
p(ξ)
with the convergence taking place in L2 (actually in S), it follows that
Z
(1 − χ(ξ))
35.6 Exercises
ˆ
GM ∗ (θv) → “G ∗ (θv) ”(x) := (θv) (ξ)eix·ξ dξ
R n p(ξ) Exercise 35.1. Using

−1 (1 − χ(ξ)) ˆ
=F (θv) (ξ) (x) ∈ S. 1
Z ∞
2
p(ξ) +m2 )
2 = e−λ(|ξ| dλ,
|ξ| + m2 0
So passing the the limit, M → ∞, in Eq. (35.26) we learn for almost every
x ∈ Rn , the identity in Eq. (35.5) and Example 34.4, show for m > 0 and x ≥ 0 that
u(x) = G ∗ (θv) (x) − lim GM ∗ (θL(βu))(x) + δ ∗ (αu) (x) m
Z ∞
1 1 2 2
M →∞
e−mx = √ dλ √ e− 4λ x e−λm (let λ → λ/m2 ) (35.27)
for a.e. x ∈ supp(α). Using the support properties of θ and β we see for x near π λ
Z ∞ 0
0 that (θL(βu))(y) = 0 unless y ∈ supp(θ) and y ∈ / {α = 1} , i.e. unless y is in 1 m2 2
= dλ √ e−λ e− 4λ x . (35.28)
an annulus centered at 0. So taking x sufficiently close to 0, we find x − y stays 0 πλ
away from 0 as y varies through the above mentioned annulus, and therefore
Z Use this formula and Example 34.4 to show, in dimension n, that
GM ∗ (θL(βu))(x) = GM (x − y)(θL(βu))(y)dy h i Γ ((n + 1)/2) m
Rn F e−m|x| (ξ) = 2n/2 √ 2
Z π (m + |ξ| )(n+1)/2
2
= L∗y {θ(y)GM (x − y)} · (βu) (y)dy
Rn
Z where Γ (x) in the gamma function defined in Eq. (20.42). (I am not absolutely
→ L∗y {θ(y)G(x − y)} · (βu) (y)dy as M → ∞. positive I have got all the constants exactly right, but they should be close.)
Rn

36
Elementary Generalized Functions / Distribution Theory
This chapter has been highly influenced by Friedlander’s book [7]. Definition 36.4. Let Y be a topological space and Ty ∈ D0 (U ) for all y ∈ Y.
We say that Ty → T ∈ D0 (U ) as y → y0 iff
36.1 Distributions on U ⊂o Rn lim hTy , ϕi = hT, ϕi for all ϕ ∈ D(U ).

y→y0
Let U be an open subset of Rn and

Cc∞ (U ) = ∪K@@U C ∞ (K) (36.1) 36.2 Examples of distributions and related computations
denote the set of smooth functions on U with compact support in U.
Example 36.5. Let µ be aRpositive Radon measure on U and f ∈ L1loc (U ). Define
∞
Definition 36.1. A sequence {ϕk }k=1
⊂ D(U ) converges to ϕ ∈ D(U ), iff there T ∈ D0 (U ) by hTf , ϕi = U ϕf dµ for all ϕ ∈ D(U ). Notice that if ϕ ∈ C ∞ (K)
is a compact set K @@ U such that supp(ϕk ) ⊂ K for all k and ϕk → ϕ in then Z Z
C ∞ (K). |hTf , ϕi| ≤ |ϕf | dµ = |ϕf | dµ ≤ CK kϕk∞
U K
Definition 36.2 (Distributions on U ⊂o Rn ). A generalized function T on
|f | dµ < ∞. Hence Tf ∈ D0 (U ). Furthermore, the map
R
U ⊂o Rn is a continuous linear functional on D(U ), i.e. T : D(U ) → C is linear where CK := K
and limn→∞ hT, ϕk i = 0 for all {ϕk } ⊂ D(U ) such that ϕk → 0 in D(U ). We
denote the space of generalized functions by D0 (U ). f ∈ L1loc (U ) → Tf ∈ D0 (U )
Proposition 36.3. Let T : D(U ) → C be a linear functional. Then T ∈ D0 (U ) is injective. Indeed, Tf = 0 is equivalent to
iff for all K @@ U, there exist n ∈ N and C < ∞ such that Z
|T (ϕ)| ≤ Cpn (ϕ) for all ϕ ∈ C ∞ (K). (36.2) ϕf dµ = 0 for all ϕ ∈ D(U ). (36.3)
U
Proof. Suppose that {ϕk } ⊂ D(U ) such that ϕk → 0 in D(U ). Let K be
for all ϕ ∈ C ∞ (K). By the dominated convergence theorem and the usual
a compact set such that supp(ϕk ) ⊂ K for all k. Since limk→∞ pn (ϕk ) = 0, it
convolution argument, this is equivalent to
follows that if Eq. (36.2) holds that limn→∞ hT, ϕk i = 0. Conversely, suppose
that there is a compact set K @@ U such that for no choice of n ∈ N and Z
C < ∞, Eq. (36.2) holds. Then we may choose non-zero ϕn ∈ C ∞ (K) such ϕf dµ = 0 for all ϕ ∈ Cc (U ). (36.4)
U
that
|T (ϕn )| ≥ npn (ϕn ) for all n. Now fix a compact set K @@ U and ϕn ∈ Cc (U ) such that ϕn → sgn(f )1K in
Let ψn = 1
npn (ϕn ) ϕn
∞
∈ C (K), then pn (ψn ) = 1/n → 0 as n → ∞ which L1 (µ). By replacing ϕn by χ(ϕn ) if necessary, where
shows that ψn → 0 in D(U ). On the other hence |T (ψn )| ≥ 1 so that
limn→∞ hT, ψn i 6= 0. Alternate Proof:The definition of T being continuous z if |z| ≤ 1
χ(z) = z
is equivalent to T |C ∞ (K) being sequentially continuous for all K @@ U. Since |z| if |z| ≥ 1,
C ∞ (K) is a metric space, sequential continuity and continuity are the same
we may assume that |ϕn | ≤ 1. By passing to a further subsequence, we may
thing. Hence T is continuous iff T |C ∞ (K) is continuous for all K @@ U. Now
assume that ϕn → sgn(f )1K a.e.. Thus we have
T |C ∞ (K) is continuous iff a bound like Eq. (36.2) holds.
410 36 Elementary Generalized Functions / Distribution Theory
Z Z Z
0 = lim ϕn f dµ = sgn(f )1K f dµ = |f | dµ. Definition 36.7 (Multiplication by smooth functions). Suppose that g ∈
n→∞ U U K C ∞ (U ) and T ∈ D0 (U ) then we define gT ∈ D0 (U ) by
This shows that |f (x)| = 0 for µ -a.e. x ∈ K. Since K is arbitrary and U is hgT, ϕi = hT, gϕi for all ϕ ∈ D(U ).
the countable union of such compact sets K, it follows that f (x) = 0 for µ -a.e.
x ∈ U. It is easily checked that gT is continuous.
The injectivity may also be proved slightly more directly as follows. As Definition 36.8 (Differentiation). For T ∈ D0 (U ) and i ∈ {1, 2, . . . , n} let
before, it suffices to prove Eq. (36.4) implies that f (x) = 0 for µ – a.e. x. ∂i T ∈ D0 (U ) be the distribution defined by
We may further assume that f is real by considering real and imaginary parts
separately. Let K @@ U and ε > 0 be given. Set A = {f > 0} ∩ K, then h∂i T, ϕi = −hT, ∂i ϕi for all ϕ ∈ D(U ).
µ(A) < ∞ and hence since all σ finite measure on U are Radon, there exists
F ⊂ A ⊂ V with F compact and V ⊂o U such that µ(V \ F ) < δ. By Uryshon’s Again it is easy to check that ∂i T is a distribution.
lemma, there exists ϕ ∈ Cc (V ) such that 0 ≤ ϕ ≤ 1 and ϕ = 1 on F. Then by
More generally if L = |α|≤m aα ∂ α with aα ∈ C ∞ (U ) for all α, then LT is
P
Eq. (36.4)
the distribution defined by
Z Z Z Z Z
X
0= ϕf dµ = ϕf dµ + ϕf dµ = ϕf dµ + ϕf dµ hLT, ϕi = hT, (−1)|α| ∂ α (aα ϕ)i for all ϕ ∈ D(U ).
U F V \F F V \F
|α|≤m
so that
Hence we can talk about distributional solutions to differential equations of the
Z Z Z

f dµ = ϕf dµ ≤ |f | dµ < ε form LT = S.

F V \F V \F
Example 36.9. Suppose that f ∈ L1loc and g ∈ C ∞ (U ), then gTf = Tgf . If
provided that δ is chosen sufficiently small by the ε – δ definition of absolute
further f ∈ C 1 (U ), then ∂i Tf = T∂i f . If f ∈ C m (U ), then LTf = TLf .
continuity. Similarly, it follows that
Z Z Example 36.10. Suppose that a ∈ U, then
0≤ f dµ ≤ f dµ + ε ≤ 2ε.
A F h∂i δa , ϕi = −∂i ϕ(a)
R
Since ε > 0 is arbitrary, it follows that A f dµ = 0. Since K was arbitrary, we and more generally we have
learn that Z
X
f dµ = 0 hLδa , ϕi = (−1)|α| ∂ α (aα ϕ) (a).
{f >0}
|α|≤m
which shows that f ≤ 0 µ – a.e. Similarly, one shows that f ≥ 0 µ – a.e. and
hence f = 0 µ – a.e. Example 36.11. Consider the distribution T := T|x| for x ∈ R, i.e. take U = R.
Then
d d2
R
Example 36.6. Let us now assume that µ = m and write hTf , ϕi = U ϕf dm. T = Tsgn(x) and 2 T = 2δ0 .
For the moment let us also assume that U = R. Then we have dx d x
More generally, suppose that f is piecewise C 1 , the
1. limM →∞ Tsin M x = 0
2. limM →∞ TM −1 sin M x = πδ0 Rwhere δ0 is the point measure at 0. d X
3. If f ∈ L1 (Rn , dm) with Rn f dm = 1 and fε (x) = ε−n f (x/ε), then Tf = Tf 0 + (f (x+) − f (x−)) δx .
dx
limε↓0 Tfε = δ0 . As a special case,
consider limε↓0 π(x2ε+ε2 ) = δ0 .

36.2 Examples of distributions and related computations 411
Example 36.12. Consider T = Tln|x| on D(R). Then Now by deforming the contour we have
Z Z Z Z Z
1 1 1
hT 0 , ϕi = − ln |x| ϕ0 (x)dx = − lim ln |x| ϕ0 (x)dx dx = dx + dz
R ε↓0 |x|>ε |x|≤1 x + iy ε<|x|≤1 x + iy Cε z + iy
Z
= − lim ln |x| ϕ0 (x)dx where Cε : z = εeiθ with θ : π → 0. Therefore,
ε↓0 |x|>ε
Z
1
Z Z Z
1 1 1
= lim ϕ(x)dx − lim [ln ε(ϕ(ε) − ϕ(−ε))] lim dx = lim dx + lim dz
ε↓0 |x|>ε x ε↓0 y↓0 |x|≤1 x + iy y↓0 ε<|x|≤1 x + iy y↓0 C z + iy
ε
Z
1
Z Z
1 1
= lim ϕ(x)dx. = dx + dz = 0 − π.
ε↓0 |x|>ε x ε<|x|≤1 x Cε z
We will write T 0 = P V 1
x in the future. Here is another formula for T 0 , Hence we have shown that T+ = P V x1 − iπδ0 . Similarly, one shows that T− =
Z
1
Z
1 P V x1 + iπδ0 . Notice that it follows from these computations that T− − T+ =
0
hT , ϕi = lim ϕ(x)dx + ϕ(x)dx i2πδ0 . Notice that
ε↓0 1≥|x|>ε x |x|>1 x 1 1 2iy
Z Z − = 2
1 1 x − iy x + iy x + y2
= lim [ϕ(x) − ϕ(0)]dx + ϕ(x)dx
ε↓0 1≥|x|>ε x |x|>1 x y
Z Z and hence we conclude that limy↓0 x2 +y 2 = πδ0 – a result that we saw in
1 1 Example 36.6, item 3.
= [ϕ(x) − ϕ(0)]dx + ϕ(x)dx.
1≥|x| x |x|>1 x
Example 36.14. Suppose that µ is a complex measure on R and F (x) =
Please notice in the last example that x1 ∈
/ L1loc (R) so that T1/x is not well µ((−∞, x]), then TF0 = µ. Moreover, if f ∈ L1loc (R) and Tf0 = µ, then f = F + C
defined. This is an example of the so called division problem of distributions. a.e. for some constant C.
Here is another possible interpretation of x1 as a distribution.
Proof. Let ϕ ∈ D := D(R), then
1
Example 36.13. Here we try to define 1/x as limy↓0 x±iy , that is we want to Z Z Z
define a distribution T± by hTF0 , ϕi = −hTF , ϕ0 i = − F (x)ϕ0 (x)dx = − dx dµ(y)ϕ0 (x)1y≤x
Z R R R
1
hT± , ϕi := lim
Z Z Z
ϕ(x)dx.
y↓0 x ± iy = − dµ(y) dxϕ0 (x)1y≤x = dµ(y)ϕ(y) = hµ, ϕi
R R R
Let us compute T+ explicitly,
Z by Fubini’s theorem and the fundamental theorem of calculus. If Tf0 = µ, then
1 Tf0 −F = 0 and the result follows from Corollary 36.16 below.
lim ϕ(x)dx
y↓0 R x + iy
Z
1
Z
1 Lemma 36.15. Suppose that T ∈ D0 (Rn ) is a distribution such that ∂i T = 0 for
= lim ϕ(x)dx + lim ϕ(x)dx some i, then there exists a distribution S ∈ D0 (Rn−1 ) such that hT, ϕi = hS, ϕ̄i i
y↓0 |x|≤1 x + iy y↓0 |x|>1 x + iy
Z Z for all φ ∈ D(Rn ) where
1 1
= lim [ϕ(x) − ϕ(0)] dx + ϕ(0) lim dx Z
y↓0 |x|≤1 x + iy y↓0 |x|≤1 x + iy
Z ϕ̄i = τtei φdt ∈ D(Rn−1 ).
1 R
+ ϕ(x)dx
|x|>1 x n
Z Z R x ∈ R as x =
Proof. To simplify notation, assume that i = n and write
1 1 (y, z) with y ∈ R n−1 ∞
and z ∈ R. Let θ ∈ Cc (R) such that R θ(z)dz = 1 and
= PV ϕ(x)dx + ϕ(0) lim dx.
R x y↓0 |x|≤1 x + iy for ψ ∈ D(Rn−1 ), let ψ ⊗ θ(x) = ψ(y)θ(z). The mapping

ψ ∈ D(Rn−1 ) ∈ ψ ⊗ θ ∈ D(Rn ) for some distribution S ∈ D0 (Rn−1 ). If α is a multi-index such that αn = 0 and
|α| = m, then
is easily seen to be sequentially continuous and therefore hS, ψi := hT, ψ ⊗ θi
defined a distribution in D0 (Rn ). Now suppose that φ ∈ D(Rn ). If φ = ∂n f for 0 = h∂ α T − ∂ α Tpn , φi = hT − Tpn , ∂ α φi = hS, (∂ α φ)n i
n
R
some f ∈ D(R ) we would have to have φ(y, z)dz = 0. This is not generally = hS, ∂ α ϕ̄n i = (−1)|α| h∂ α S, ϕ̄n i.
true, however the function φ − ϕ̄ ⊗ θ does have this property. Define
Z z and in particular by taking φ = ψ ⊗ θ, we learn that h∂ α S, ψi = 0 for all
f (y, z) := [φ(y, z 0 ) − ϕ̄(y)θ(z 0 )] dz 0 , ψ ∈ D(Rn−1 ). Thus by the induction hypothesis, S = Tr for some polynomial
−∞ (r) of degree at most m on Rn−1 . Letting p(y, z) = pn (y, z)+r(y) – a polynomial
of degree at most m on Rn , it is easily checked that T = Tp .
then f ∈ D(Rn ) and ∂n f = φ − ϕ̄ ⊗ θ. Therefore,
Example 36.17. Consider the wave equation
0 = −h∂n T, f i = hT, ∂n f i = hT, φi − hT, ϕ̄ ⊗ θi = hT, φi − hS, ϕ̄i.
(∂t − ∂x ) (∂t + ∂x ) u(t, x) = ∂t2 − ∂x2 u(t, x) = 0.

From this equation one learns that u(t, x) = f (x + t) + g(x − t) solves the wave
Corollary 36.16. Suppose that T ∈ D0 (Rn ) is a distribution such that there equation for f, g ∈ C 2 . Suppose that f is a bounded Borel measurable function
exists m ≥ 0 such that on R and consider the function f (x + t) as a distribution on R. We compute
∂ α T = 0 for all |α| = m,
Z
h(∂t − ∂x ) f (x + t), φ(x, t)i = f (x + t) (∂x − ∂t ) φ(x, t)dxdt
R2
then T = Tp where p(x) is a polynomial on Rn of degree less than or equal to Z
m − 1, where by convention if deg(p) = −1 then p := 0. = f (x) [(∂x − ∂t ) φ] (x − t, t)dxdt
R2
Z
Proof. The proof will be by induction on n and m. The corollary is trivially d
=− f (x) [φ(x − t, t)] dxdt
true when m = 0 and n is arbitrary. Let n = 1 and assume the corollary holds 2 dt
ZR
for m = k − 1 with k ≥ 1. Let T ∈ D0 (R) such that 0 = ∂ k T = ∂ k−1 ∂T. By
= − f (x) [φ(x − t, t)] |t=∞t=−∞ dx = 0.
R x there exists a polynomial, q, of degree k − 2 such that
the induction hypothesis,
R
T 0 = Tq . Let p(x) = 0 q(z)dz, then p is a polynomial of degree at most k − 1
such that p0 = q and hence Tp0 = Tq = T 0 . So (T − Tp )0 = 0 and hence by This shows that (∂t − ∂x ) f (x + t) = 0 in the distributional sense. Similarly,
Lemma 36.15, T − Tp = TC where C = hT − Tp , θi and θ is as in the proof of (∂t + ∂x ) g(x − t) = 0 in the distributional sense. Hence u(t, x) = f (x + t) +
Lemma 36.15. This proves the he result for n = 1. For the general induction, g(x − t) solves the wave equation in the distributional sense whenever f and g
suppose there exists (m, n) ∈ N2 with m ≥ 0 and n ≥ 1 such that assertion are bounded Borel measurable functions on R.
in the corollary holds for pairs (m0 , n0 ) such that either n0 < n of n0 = n and
m0 ≤ m. Suppose that T ∈ D0 (Rn ) is a distribution such that Example 36.18. Consider f (x) = ln |x| for x ∈ R2 and let T = Tf . Then, point-
wise we have
∂ α T = 0 for all |α| = m + 1. x 2 x
∇ ln |x| = 2 and ∆ ln |x| = 2 − 2x · 4 = 0.
In particular this implies that ∂ α ∂n T = 0 for all |α| = m − 1 and hence by |x| |x| |x|
induction ∂n T =R Tqn where qn is a polynomial of degree at most m − 1 on Hence ∆f (x) = 0 for all x ∈ R2 except at x = 0 where it is not defined. Does
z
Rn . Let pn (x) = 0 qn (y, z 0 )dz 0 a polynomial of degree at most m on Rn . The this imply that ∆T = 0? No, in fact ∆T = 2πδ as we shall now prove. By
polynomial pn satisfies, 1) ∂ α pn = 0 if |α| = m and αn = 0 and 2) ∂n pn = qn . definition of ∆T and the dominated convergence theorem,
Hence ∂n (T − Tpn ) = 0 and so by Lemma 36.15, Z Z
hT − Tpn , φi = hS, ϕ̄n i h∆T, φi = hT, ∆φi = ln |x| ∆φ(x)dx = lim ln |x| ∆φ(x)dx.
R2 ε↓0 |x|>ε

36.3 Other classes of test functions 413
Using the divergence theorem, Proof. Pointwise we have ∂¯ z1 = 0 so we shall work as above. We then have
Z Z
¯ ¯ 1¯
ln |x| ∆φ(x)dx h∂T, φi = −hT, ∂φi = − ∂φ(z)dm(z)
|x|>ε R 2 z
Z Z Z
1¯
=− ∇ ln |x| · ∇φ(x)dx + ln |x| ∇φ(x) · n(x)dS(x) = − lim ∂φ(z)dm(z)
|x|>ε ∂{|x|>ε} ε↓0 |z|>ε z
Z Z Z
1
= ∆ ln |x| φ(x)dx − ∇ ln |x| · n(x)φ(x)dS(x) = lim ∂¯ φ(z)dm(z)
|x|>ε ∂{|x|>ε} ε↓0 |z|>ε z
Z Z
1 1
+ ln |x| (∇φ(x) · n(x)) dS(x) − lim φ(z) (n1 (z) + in2 (z)) dσ(z)
∂{|x|>ε} ε↓0 ∂{|z|>ε} z 2
Z
1 −z
Z
1
= ln |x| (∇φ(x) · n(x)) dS(x) = 0 − lim φ(z) dσ(z)
∂{|x|>ε} ε↓0 ∂{|z|>ε} z 2 |z|
Z Z
1 1
− ∇ ln |x| · n(x)φ(x)dS(x), = lim φ(z)dσ(z)
∂{|x|>ε} 2 ε↓0 ∂{|z|>ε} |z|
Z
1
where n(x) is the outward pointing normal, i.e. n(x) = −x̂ := x/ |x| . Now = π lim φ(z)dσ(z) = πφ(0).
ε↓0 2πε ∂{|z|>ε}
Z

ln |x| (∇φ(x) · n(x)) dS(x) ≤ C ln ε−1 2πε → 0 as ε ↓ 0

∂{|x|>ε}
where C is a bound on (∇φ(x) · n(x)) . While 36.3 Other classes of test functions
Z Z
x̂
∇ ln |x| · n(x)φ(x)dS(x) = · (−x̂)φ(x)dS(x) (For what follows, see Exercises 13.26 and 13.27 of Chapter 18.
∂{|x|>ε} ∂{|x|>ε} |x|
∞
1
Z Notation 36.20 Suppose that X is a vector space and {pn }n=0 is a family of
=− φ(x)dS(x) semi-norms on X such that pn ≤ pn+1 for all n and with the property that
ε ∂{|x|>ε}
pn (x) = 0 for all n implies that x = 0. (We allow for pn = p0 for all n in which
→ −2πφ(0) as ε ↓ 0. case X is a normed vector space.) Let τ be the smallest topology on X such
that pn (x − ·) : X → [0, ∞) is continuous for all n ∈ N and x ∈ X. For n ∈ N,
Combining these results shows
x ∈ X and ε > 0 let Bn (x, ε) := {y ∈ X : pn (x − y) < ε} .
h∆T, φi = 2πφ(0). Proposition 36.21. The balls B := {Bn (x, ε) : n ∈ N, x ∈ X and ε > 0} for a
basis for the topology τ. This topology is the same as the topology induced by the
Exercise 36.1. Carry out a similar computation to that in Example 36.18 to
metric d on X defined by
show
∆T1/|x| = −4πδ ∞
X pn (x − y)
d(x, y) = 2−n .
where now x ∈ R3 . 1 + pn (x − y)
n=0
Example 36.19. Let z = x + iy, and ∂¯ = 1

2 (∂x + i∂y ). Let T = T1/z , then Moreover, a sequence {xk } ⊂ X is convergent to x ∈ X iff limk→∞ d(x, xk ) = 0
iff limn→∞ pn (x, xk ) = 0 for all n ∈ N and {xk } ⊂ X is Cauchy in X iff
¯ = πδ0 or imprecisely ∂¯ 1 = πδ(z).
∂T limk,l→∞ d(xl , xk ) = 0 iff limk,l→∞ pn (xl , xk ) = 0 for all n ∈ N.
z

Proof. Suppose that z ∈ Bn (x, ε) ∩ Bm (y, δ) and assume with out loss of Lemma 36.22. Suppose that X and Y are vector spaces equipped with se-
generality that m ≥ n. Then if pm (w − z) < α, we have quences of norms {pn } and {qn } as in Notation 36.20. Then a linear map
T : X → Y is continuous if for all n ∈ N there exists Cn < ∞ and mn ∈ N
pm (w − y) ≤ pm (w − z) + pm (z − y) < α + pm (z − y) < δ such that qn (T x) ≤ Cn pmn (x) for all x ∈ X. In particular, f ∈ X ∗ iff
|f (x)| ≤ Cpm (x) for some C < ∞ and m ∈ N. (We may also characterize
provided that α ∈ (0, δ − pm (z − y)) and similarly
continuity by sequential convergence since both X and Y are metric spaces.)
pn (w − x) ≤ pm (w − x) ≤ pm (w − z) + pm (z − x) < α + pm (z − x) < ε Proof. Suppose that T is continuous, then {x : qn (T x) < 1} is an open
neighborhood of 0 in X. Therefore, there exists m ∈ N and ε > 0 such that
provided that α ∈ (0, ε − pm (z − x)). So choosing Bm (0, ε) ⊂ {x : qn (T x) < 1} . So for x ∈ X and α < 1, αεx/pm (x) ∈ Bm (0, ε)
1 and thus
δ= min (δ − pm (z − y), ε − pm (z − x)) , αε 1
2 qn T x < 1 =⇒ qn (T x) < pm (x)
pm (x) αε
we have shown that Bm (z, α) ⊂ Bn (x, ε) ∩ Bm (y, δ). This shows that B forms for all x. Letting α ↑ 1 shows that qn (T x) ≤ 1ε pm (x) for all x ∈ X. Conversely,
a basis for a topology. In detail, V ⊂o X iff for all x ∈ V there exists n ∈ N if T satisfies
and ε > 0 such that Bn (x, ε) := {y ∈ X : pn (x − y) < ε} ⊂ V. Let τ (B) be the qn (T x) ≤ Cn pmn (x) for all x ∈ X,
topology generated by B. Since|pn (x − y) − pn (x − z)| ≤ pn (y − z), we see that
pn (x−·) is continuous on relative to τ (B) for each x ∈ X and n ∈ N. This shows then
that τ ⊂ τ (B). On the other hand, since pn (x − ·) is τ – continuous, it follows qn (T x − T x0 ) = qn (T (x − x0 )) ≤ Cn pmn (x − x0 ) for all x, y ∈ X.
that Bn (x, ε) = {y ∈ X : pn (x − y) < ε} ∈ τ for all x ∈ X, ε > 0 and n ∈ N.
This shows that B ⊂ τ and therefore that τ (B) ⊂ τ. Thus τ = τ (B). Given This shows T x0 → T x as x0 → x, i.e. that T is continuous.
x ∈ X and ε > P0, let Bd (x, ε) = {y ∈ X : d(x, y) < ε} be a d – ball. Choose N Definition 36.23. A Frechét space is a vector space X equipped with a family
∞
large so that n=N +1 2−n < ε/2. Then y ∈ BN (x, ε/4) we have {pn } of semi-norms such that X is complete in the associated metric d.
N Example 36.24. Let K @@ Rn and C ∞ (K) := {f ∈ Cc∞ (Rn ) : supp(f ) ⊂ K} .
X
−n ε
d(x, y) = pN (x − y) 2 + ε/2 < 2 + ε/2 < ε For m ∈ N, let
4 X
n=0 pm (f ) := k∂ α f k∞ .
|α|≤m
which shows that BN (x, ε/4) ⊂ Bd (x, ε). Conversely, if d(x, y) < ε, then
∞
Then (C ∞ (K), {pm }m=1 ) is a Frechét space. Moreover the derivative operators
pn (x − y) {∂k } and multiplication by smooth functions are continuous linear
2−n <ε R maps from
1 + pn (x − y) C ∞ (K) to C ∞ (K). If µ is a finite measure on K, then T (f ) := K ∂ α f dµ is an
element of C ∞ (K)∗ for any multi index α.
which implies that
2−n ε Example 36.25. Let U ⊂o Rn and for m ∈ N, and a compact set K @@ U let
pn (x − y) < =: δ
1 − 2−n ε X X
pKm (f ) := k∂ α f k∞,K := max |∂ α f (x)| .
when 2−n ε < 1 which shows that Bn (x, δ) contains Bd (x, ε) with ε and δ as |α|≤m |α|≤m
x∈K
above. This shows that τ and the topology generated by d are the same. The o
moreover statements are now easily proved and are left to the reader. Choose a sequence Km @@ U such that Km ⊂ Km+1 ⊂ Km+1 @@ U for
Km ∞ ∞
all m and set qm (f ) = pm (f ). Then (C (K), {pm }m=1 ) is a Frechét space
Exercise 36.2. Keeping the same notation as Proposition 36.21 and further and the topology in independent of the choice of sequence of compact sets K
assume that {p0n }n∈N is another family of semi-norms as in Notation 36.20. exhausting U. Moreover the derivative operators {∂k } and multiplication by
Then the topology τ 0 determined by {p0n }n∈N is weaker then the topology τ smooth functions are continuous linear maps from C ∞ (U R ) to C ∞ (U ). If µ is a
determined by {pn }n∈N (i.e. τ 0 ⊂ τ ) iff for every n ∈ N there is an m ∈ N and α
finite measure with compact support in U, then T (f ) := K ∂ f dµ is an element
C < ∞ such that p0n ≤ Cpm . of C ∞ (U )∗ for any multi index α.

36.3 Other classes of test functions 415
Proposition 36.26. A linear functional T on C ∞ (U ) is continuous, i.e. T ∈ 1

R
where C = Rn (1+|x| 2 n dξ < ∞. Using the product rule repeatedly, it is not
)
C ∞ (U )∗ iff there exists a compact set K @@ U, m ∈ N and C < ∞ such that hard to show

∞
|hT, φi| ≤ CpK
m (φ) for all φ ∈ C (U ).
2 2
|g(·)| (1 + |·| )n = (1 + |·| )n (1 − ∆)m ((−ix)α f )

∞
2 X ∞
Notation 36.27 Let νs (x) := (1 + |x|)s (or change to νs (x) = (1 + |x| )s/2 = ≤k
2 n+|α|/2 β
(1 + |·| )

∂ f
hxis ?) for x ∈ Rn and s ∈ R. ∞
|β|≤2m
Example 36.28. Let S denote the space of functions f ∈ C ∞ (Rn ) such that f ≤ kp02m+n (f )
and all of its partial derivatives decay faster that (1 + |x|)−m for all m > 0 as
for some constant k < ∞. Combining the last two displayed equations implies
in Definition 34.6. Define
that p0m (fˆ) ≤ Ckp02m+n (f ) for all f ∈ S, and thus F is continuous.
X X
pm (f ) = k(1 + | · |)m ∂ α f (·)k∞ = k(µm ∂ α f (·)k∞ , Proposition 36.30. The subspace Cc∞ (Rn ) is dense in S(Rn ).
|α|≤m |α|≤m
Proof. Let θ ∈ Cc∞ (Rn ) such that θ = 1 in a neighborhood of 0 and set
then (S, {pm }) is a Frechét space. Again the derivative operators {∂k } and θm (x) = θ(x/m) for all m ∈ N. We will now show for all f ∈ S that θm f
multiplication by function f ∈ P are examples of continuous linear operators converges to f in S. The main point is by the product rule,
on S. For an example of an element T ∈ S ∗ , let µ be a measure on Rn such
that X α
α
∂ (θm f − f ) (x) = ∂ α−β θm (x)∂ β f (x) − f
Z
(1 + |x|)−N d|µ|(x) < ∞ β≤α
β
X α 1
∂ α f dµ defines and element of S ∗ .
R
for some N ∈ N. Then T (f ) := K = ∂ α−β θ(x/m)∂ β f (x).
β m|α−β|
β≤α:β6=α
Proposition 36.29. The Fourier transform F : S → S is a continuous linear

transformation.

Since max ∂ β θ ∞ : β ≤ α is bounded it then follows from the last equation
that kµt ∂ α (θm f − f )k∞ = O(1/m) for all t > 0 and α. That is to say θm f → f
Proof. For the purposes of this proof, it will be convenient to use the semi-
in S.
norms X
p0m (f ) = (1 + | · |2 )m ∂ α f (·) .

∞ Lemma 36.31 (Peetre’s Inequality). For all x, y ∈ Rn and s ∈ R,
|α|≤m n o
This is permissible, since by Exercise 36.2 they give rise to the same topology (1 + |x + y|)s ≤ min (1 + |y|)|s| (1 + |x|)s , (1 + |y|)s (1 + |x|)|s| (36.5)
on S. Let f ∈ S and m ∈ N, then
that is to say νs (x + y) ≤ ν|s| (x)νs (y) and νs (x + y) ≤ νs (x)ν|s| (y) for all s ∈ R,
2 2
(1 + |ξ| )m ∂ α fˆ(ξ) = (1 + |ξ| )m F ((−ix)α f ) (ξ) where νs (x) = (1 + |x|)s as in Notation 36.27. We also have the same results
= F [(1 − ∆)m ((−ix)α f )] (ξ) for hxi, namely
n o
and therefore if we let g = (1 − ∆)m ((−ix)α f ) ∈ S, hx + yis ≤ 2|s|/2 min hxi|s| hyis , hxis hyi|s| . (36.6)
Z
2
(1 + |ξ| )m ∂ α fˆ(ξ) ≤ kgk1 =

|g(x)| dx Proof. By elementary estimates,
Rn
Z
2 1 (1 + |x + y|) ≤ 1 + |x| + |y| ≤ (1 + |x|)(1 + |y|)
= |g(x)| (1 + |x| )n 2 dξ
R n (1 + |x| )n and so for Eq. (36.5) holds if s ≥ 0. Now suppose that s < 0, then

2
≤ C |g(·)| (1 + |·| )n

∞ (1 + |x + y|)s ≥ (1 + |x|)s (1 + |y|)s

and letting x → x − y and y → −y in this inequality implies 36.4 Compactly supported distributions
(1 + |x|)s ≥ (1 + |x + y|)s (1 + |y|)s . Definition 36.33. For a distribution T ∈ D0 (U ) and V ⊂o U, we say T |V = 0
if hT, φi = 0 for all φ ∈ D(V ).
This inequality is equivalent to
Proposition 36.34. Suppose that V := {Vα }α∈A is a collection of open subset
(1 + |x + y|)s ≤ (1 + |x|)s (1 + |y|)−s = (1 + |x|)s (1 + |y|)|s| .
of U such that T |Vα = 0 for all α, then T |W = 0 where W = ∪α∈A Vα .
By symmetry we also have Proof. Let {ψα }α∈A be a smooth partition of unity subordinate to V, i.e.
s |s|
(1 + |x + y|) ≤ (1 + |x|) (1 + |y|) . s supp(ψα ) ⊂ Vα for all α ∈ A, for each point x ∈ W there exists a neighborhood
P
Nx ⊂o W such that #{α ∈ A : supp(ψ P α ) ∩ Nx 6= ∅} < ∞ and 1W = α∈A ψα .
For the proof of Eq. (36.6 Then for φ ∈ D(W ), we have φ = α∈A φψα and there are only a finite number
of nonzero terms in the sum since supp(φ) is compact. Since φψα ∈ D(Vα ) for
2 2 2 2
hx + yi2 = 1 + |x + y| ≤ 1 + (|x| + |y|) = 1 + |x| + |y| + 2 |x| |y| all α, X X
2 2 2 2
≤ 1 + 2 |x| + 2 |y| ≤ 2(1 + |x| )(1 + |y| ) = 2hxi2 hyi2 . hT, φi = hT, φψα i = hT, φψα i = 0.
α∈A α∈A
From this it follows that hxi−2 ≤ 2hx + yi−2 hyi2 and hence
hx + yi−2 ≤ 2hxi−2 hyi2 . Definition 36.35. The support, supp(T ), of a distribution T ∈ D0 (U ) is the
relatively closed subset of U determined by
So if s ≥ 0, then
hx + yis ≤ 2s/2 hxis hyis U \ supp(T ) = ∪ {V ⊂o U : T |V = 0} .
and
By Proposition 36.26, supp(T ) may described as the smallest (relatively) closed
hx + yi−s ≤ 2s/2 hxi−s hyis .
set F such that T |U \F = 0.
Proposition 36.36. If f ∈ L1loc (U ), then supp(Tf ) = ess sup(f ), where

Proposition 36.32. Suppose that f, g ∈ S then f ∗ g ∈ S.
ess sup(f ) := {x ∈ U : m({y ∈ V : f (y) 6= 0}}) > 0 for all neighborhoods V of x}
Proof. First proof. Since F(f ∗g) = fˆĝ ∈ S it follows that f ∗g = F −1 (fˆĝ) ∈
S as well. For the second proof we will make use of Peetre’s inequality. We have as in Definition 22.25.
for any k, l ∈ N that
Z Proof. The key point is that Tf |V = 0 iff f = 0 a.e. on V and therefore
νt (x) |∂ (f ∗ g)(x)| = νt (x) |∂ f ∗ g(x)| ≤ νt (x) |∂ α f (x − y)| |g(y)| dy
α α
U \ supp(Tf ) = ∪ {V ⊂o U : f 1V = 0 a.e.} .
Z Z
≤ Cνt (x) ν−k (x − y)ν−l (y)dy ≤ Cνt (x) ν−k (x)νk (y)ν−l (y)dy On the other hand,
U \ ess sup(f ) = {x ∈ U : m({y ∈ V : f (y) 6= 0}}) = 0 for some neighborhood V of x}

Z
= Cνt−k (x) νk−l (y)dy.
= ∪{x ∈ U : f 1V = 0 a.e. for some neighborhood V of x}
Choosing k = t and l > t + n we learn that = ∪ {V ⊂o U : f 1V = 0 a.e.}
Z
νt (x) |∂ α (f ∗ g)(x)| ≤ C νk−l (y)dy < ∞
Definition 36.37. Let E 0 (U ) := {T ∈ D0 (U ) : supp(T ) ⊂ U is compact} – the
α
showing kνt ∂ (f ∗ g)k∞ < ∞ for all t ≥ 0 and α ∈ N . n
compactly supported distributions in D0 (U ).

36.5 Tempered Distributions and the Fourier Transform 417
X X
Lemma 36.38. Suppose that T ∈ D0 (U ) and f ∈ C ∞ (U ) is a function such k∂ α (θf )k∞ ∂ β f

|hT, f i| = |hT, θf i| = C ≤C ∞,F
that K := supp(T ) ∩ supp(f ) is a compact subset of U. Then we may define |α|≤m |β|≤m
hT, f i := hT, θf i, where θ ∈ D(U ) is any function such that θ = 1 on a neigh-
borhood of K. Moreover, if K @@ U is a given compact set and F @@ U is a and this estimate holds for all f ∈ C ∞ (U ) such that supp(T ) ∩ supp(f ) ⊂ K.
compact set such that K ⊂ F o , then there exists m ∈ N and C < ∞ such that Theorem 36.39. The restriction of T ∈ C ∞ (U )∗ to Cc∞ (U ) defines an element
X
∂ β f
in E 0 (U ). Moreover the map
|hT, f i| ≤ C ∞,F
(36.7)
|β|≤m i
T ∈ C ∞ (U )∗ → T |D(U ) ∈ E 0 (U )
for all f ∈ C ∞ (U ) such that supp(T ) ∩ supp(f ) ⊂ K. In particular if T ∈ E 0 (U )
is a linear isomorphism of vector spaces. The inverse map is defined as follows.
then T extends uniquely to a linear functional on C ∞ (U ) and there is a compact
Given S ∈ E 0 (U ) and θ ∈ Cc∞ (U ) such that θ = 1 on K = supp(S) then
subset F @@ U such that the estimate in Eq. (36.7) holds for all f ∈ C ∞ (U ).
i−1 (S) = θS, where θS ∈ C ∞ (U )∗ defined by
Proof. Suppose that θ̃ is another such cutoff function andlet V be an open
hθS, φi = hS, θφi for all φ ∈ C ∞ (U ).
neighborhood of K such that θ = θ̃ = 1 on V. Setting g := θ − θ̃ f ∈ D(U )
we see that Proof. Suppose that T ∈ C ∞ (U )∗ then there exists a compact set K @@ U,
m ∈ N and C < ∞ such that
supp(g) ⊂ supp(f ) \ V ⊂ supp(f ) \ K = supp(f ) \ supp(T ) ⊂ U \ supp(T ),
∞
|hT, φi| ≤ CpK
m (φ) for all φ ∈ C (U )
see Figure 36.1 below. Therefore,
where pKm is defined in Example 36.25. It is clear using the sequential notion of
0 = hT, gi = hT, θ − θ̃ f i = hT, θf i − hT, θ̃f i continuity that T |D(U ) is continuous on D(U ), i.e. T |D(U ) ∈ D0 (U ). Moreover,
if θ ∈ Cc∞ (U ) such that θ = 1 on a neighborhood of K then
which shows that hT, f i is well defined. Moreover, if F @@ U is a compact set
|hT, θφi − hT, φi| = |hT, (θ − 1) φi| ≤ CpK
m ((θ − 1) φ) = 0,
which shows θT = T. Hence supp(T ) = supp(θT ) ⊂ supp(θ) @@ U showing

that T |D(U ) ∈ E 0 (U ). Therefore the map i is well defined and is clearly linear.
I also claim that i is injective because if T ∈ C ∞ (U )∗ and i(T ) = T |D(U ) ≡ 0,
then hT, φi = hθT, φi = hT |D(U ) , θφi = 0 for all φ ∈ C ∞ (U ). To show i is
surjective suppose that S ∈ E 0 (U ). By Lemma 36.38 we know that S extends
uniquely to an element S̃ of C ∞ (U )∗ such that S̃|D(U ) = S, i.e. i(S̃) = S. and
K = supp(S).
Lemma 36.40. The space E 0 (U ) is a sequentially dense subset of D0 (U ).
o
Proof. Choose Kn @@ U such that Kn ⊂ Kn+1 ⊂ Kn+1 ↑ U as n → ∞.
Let θn ∈ Cc (Kn+1 ) such that θn = 1 on K. Then for T ∈ D0 (U ), θn T ∈ E 0 (U )
∞ 0
and θn T → T as n → ∞.
Fig. 36.1. Intersecting the supports. 36.5 Tempered Distributions and the Fourier Transform
The space of tempered distributions S 0 (Rn ) is the continuous dual to S =
o
such that K ⊂ F and θ ∈ Cc∞ (F 0 ) is a function which is 1 on a neighborhood S(Rn ). A linear functional T on S is continuous iff there exists k ∈ N and
of K, we have C < ∞ such that

X
kνk ∂ α φk∞
R
|hT, φi| ≤ Cpk (φ) := C (36.8) Proof. If µ̂ = 0, then µ = 0 as a distribution, i.e. φdµ = 0 for all φ ∈ S
|α|≤k and in particular for all φ ∈ D. By Example 36.5 this implies that µ is the zero
measure.
for all φ ∈ S. Since D = D (Rn ) is a dense subspace of S any element T ∈ S 0 More generally we have the following analogous theorem for compactly sup-
is determined by its restriction to D. Moreover, if T ∈ S 0 it is easy to see that ported distributions.
T |D ∈ D0 . Conversely and element T ∈ D0 satisfying an estimate of the form
in Eq. (36.8) for all φ ∈ D extend uniquely to an element of S 0 . In this way we Theorem 36.45. Let S ∈ E 0 (Rn ), then Ŝ is an analytic function and Ŝ(z) =
may view S 0 as a subspace of D0 . hS(x), e−ix·z i. Also if supp(S) @@ B(0, M ), then Ŝ(z) satisfies a bound of the
form
Example 36.41. Any compactly supported distribution is tempered, i.e. E 0 (U ) ⊂

Ŝ(z) ≤ C(1 + |z|)m eM |Im z|

S 0 (Rn ) for any U ⊂o Rn .
for some m ∈ N and C < ∞. If S ∈ D(Rn ), i.e. if S is assumed to be smooth,
One of the virtues of S 0 is that we may extend the Fourier transform to S 0 . then for all m ∈ N there exists Cm < ∞ such that
Recall that for L1 functions f and g we have the identity,

Ŝ(z) ≤ Cm (1 + |z|)−m eM |Im z| .

hfˆ, gi = hf, ĝi.
This suggests the following definition. Proof. The function h(z) = hS(ξ), e−iz·ξ i for z ∈ Cn is analytic since the
map z ∈ Cn → e−iz·ξ ∈ C ∞ (ξ ∈ Rn ) is analytic and S is complex linear.
Definition 36.42. The Fourier and inverse Fourier transform of a tempered Moreover, we have the bound
distribution T ∈ S 0 are the distributions T̂ = FT ∈ S 0 and T ∨ = F −1 T ∈
S 0 defined by
X
|h(z)| = hS(ξ), e−iz·ξ i ≤ C ∂ξα e−iz·ξ

∞,B(0,M )
|α|≤m
hT̂ , φi = hT, ϕ̂i and hT ∨ , φi = hT, φ∨ i for all φ ∈ S. X
z α e−iz·ξ

=C ∞,B(0,M )
Since F : S → S is a continuous isomorphism with inverse F −1 , one easily |α|≤m
checks that T̂ and T ∨ are well defined elements of S and that F −1 is the inverse X |α| −iz·ξ
≤C |z| e ≤ C(1 + |z|)m eM |Im z| .
of F on S 0 . ∞,B(0,M )
|α|≤m
n
Example 36.43. Suppose that µ is aR complex measure on R . Then we may view
If we now assume that S ∈ D(Rn ), then
µ as an element of S 0 via hµ, φi = φdµ for all φ ∈ S 0 . Then by Fubini-Tonelli,
Z Z
α α −iz·ξ α −iz·ξ
Z Z Z
z Ŝ(z) = S(ξ)z e dξ = S(ξ)(i∂ξ ) e dξ

−ix·ξ

hµ̂, φi = hµ, ϕ̂i = ϕ̂(x)dµ(x) = φ(ξ)e dξ dµ(x) n n
ZR R Z
(−i∂ξ )α S(ξ)e−iz·ξ dξ ≤ eM |Im z| |∂ξ α S(ξ)| dξ
Z Z
=
= φ(ξ)e−ix·ξ dµ(x) dξ Rn Rn
showing
whichR shows that µ̂R is the distribution associated to the continuous function
|z α | Ŝ(z) ≤ eM |Im z| k∂ α Sk1

ξ → e−ix·ξ dµ(x). e−ix·ξ dµ(x)We will somewhat abuse notation and identify
the distribution µ̂ with the function ξ → e−ix·ξ dµ(x). When dµ(x) = f (x)dx
R
and therefore
with f ∈ L1 , we have µ̂ = fˆ, so the definitions are all consistent. X
(1 + |z|)m Ŝ(z) ≤ CeM |Im z| k∂ α Sk1 ≤ CeM |Im z| .

Corollary 36.44. Suppose that µ is a complex measure such that µ̂ = 0, then |α|≤m
µ = 0. So complex measures on Rn are uniquely determined by their Fourier
transform.

36.5 Tempered Distributions and the Fourier Transform 419
So to finish the proof it suffices to show h = Ŝ in the sense of distributions1 . Remark 36.46. Notice that
For this let φ ∈ D, K @@ Rn be a compact set for ε > 0 let
X ∂ α Ŝ(z) = hS(x), ∂zα e−ix·z i = hS(x), (−ix)α e−ix·z i = h(−ix)α S(x), e−ix·z i
ϕ̂ε (ξ) = (2π)−n/2 εn φ(x)e−ix·ξ .
and (−ix)α S(x) ∈ E 0 (Rn ). Therefore, we find a bound of the form
x∈εZn
0
∂ Ŝ(z) ≤ C(1 + |z|)m eM |Im z|
α
This is a finite sum and

sup |∂ α (ϕ̂ε (ξ) − ϕ̂(ξ))|

ξ∈K where C and m0 depend on α. In particular, this shows that Ŝ ∈ P, i.e. S 0 is

Z

preserved under multiplication by Ŝ.
X α α
(−iy) φ(y)e−iy·ξ − (−ix) φ(x)e−ix·ξ dx

= sup The converse of this theorem holds as well. For the moment we only have
ξ∈K n
y∈εZn y+ε(0,1] the tools to prove the smooth converse. The general case will follow by using
X Z the notion of convolution to regularize a distribution to reduce the question to
sup y α φ(y)e−iy·ξ − xα φ(x)e−ix·ξ dx

≤ the smooth case.
y∈εZn y+ε(0,1]n ξ∈K
Theorem 36.47. Let S ∈ S(Rn ) and assume that Ŝ is an analytic function
By uniform continuity of xα φ(x)e−ix·ξ for (ξ, x) ∈ K × Rn (φ has compact and there exists an M < ∞ such that for all m ∈ N there exists Cm < ∞ such
support), that
Ŝ(z) ≤ Cm (1 + |z|)−m eM |Im z| .

y φ(y)e−iy·ξ − xα φ(x)e−ix·ξ → 0 as ε ↓ 0
α
δ(ε) = sup sup sup
ξ∈K y∈εZn x∈y+ε(0,1]n
Then supp(S) ⊂ B(0, M ).
which shows
sup |∂ α (ϕ̂ε (ξ) − ϕ̂(ξ))| ≤ Cδ(ε) Proof. By the Fourier inversion formula,
ξ∈K Z
where C is the volume of a cube in Rn which contains the support of φ. This S(x) = Ŝ(ξ)eiξ·x dξ
Rn
shows that ϕ̂ε → ϕ̂ in C ∞ (Rn ). Therefore,
X and by deforming the contour, we may express this integral as
hŜ, φi = hS, ϕ̂i = limhS, ϕ̂ε i = lim(2π)−n/2 εn φ(x)hS(ξ), e−ix·ξ i Z Z
ε↓0 ε↓0
x∈εZn S(x) = Ŝ(ξ)eiξ·x dξ = Ŝ(ξ + iη)ei(ξ+iη)·x dξ
Rn +iη Rn
X Z
= lim(2π)−n/2 εn φ(x)h(x) = φ(x)h(x)dx = hh, φi.
ε↓0
x∈εZn Rn for any η ∈ Rn . From this last equation it follows that
Z Z
−η·x −η·x M |η|
|S(x)| ≤ e Ŝ(ξ + iη) dξ ≤ Cm e e (1 + |ξ + iη|)−m dξ

1 Rn Rn
This is most easily done using Fubini’s Theorem 37.2 for distributions proved below. Z
This proof goes as follows. Let θ, η ∈ D(Rn ) such that θ = 1 on a neighborhood of
≤ Cm e−η·x eM |η| (1 + |ξ|)−m dξ ≤ C̃m e−η·x eM |η|
supp(S) and η = 1 on a neighborhood of supp(φ) then Rn
−ix·ξ −ix·ξ
hh, φi = hφ(x), hS(ξ), e ii = hη(x)φ(x), hS(ξ), θ(ξ)e ii where C̃m < ∞ if m > n. Letting η = λx with λ > 0 we learn
−ix·ξ
= hφ(x), hS(ξ), η(x)θ(ξ)e ii.
2

|S(x)| ≤ C̃m exp −λ |x| + M |x| = C̃m eλ|x|(M −|x|) . (36.9)
We may now apply Theorem 37.2 to conclude,
hh, φi = hS(ξ), hφ(x), η(x)θ(ξ)e−ix·ξ ii = hS(ξ), θ(ξ)hφ(x), e−ix·ξ ii = hS(ξ), hφ(x), e−ix·ξ ii Hence if |x| > M, we may let λ → ∞ in Eq. (36.9) to show S(x) = 0. That is
to say supp(S) ⊂ B(0, M ).
= hS(ξ), φ̂(ξ)i.
Let us now pause to work out some specific examples of Fourier transform
of measures.

Example 36.48 (Delta Functions). Let a ∈ Rn and δa be the point mass measure and
at a, then Z Z
δ̂a (ξ) = e−ia·ξ . p(i∂ξ )ϕ̂(ξ) = p(i∂ξ ) φ(x)e−ix·ξ
dx = p(x)φ(x)e−ix·ξ dx = (pφ) ˆ.
In particular it follows that
Thus
−1 −ia·ξ hFLT, φi = hT, p(i∂)ϕ̂i = hT, (pφ) î = hT̂ , pφi = hpT̂ , φi
F e = δa .
To see the content of this formula, let φ ∈ S. Then which proves the lemma.
Z Example 36.51. Let n = 1, −∞ < a < b < ∞, and dµ(x) = 1[a,b] (x)dx. Then
e−ia·ξ φ∨ (ξ)dξ = he−ia·ξ , F −1 φi = hF −1 e−ia·ξ , φi = hδa , φi = φ(a)
b
1 e−ix·ξ b 1 e−ib·ξ − e−ia·ξ
Z
µ̂(ξ) = e−ix·ξ dx = √ |a = √
which is precisely the Fourier inversion formula. a 2π −iξ 2π −iξ
−ia·ξ −ib·ξ
1 e −e
Example 36.49. Suppose that p(x) is a polynomial. Then =√ .
2π iξ
Z
hp̂, φi = hp, ϕ̂i = p(ξ)ϕ̂(ξ)dξ. So by the inversion formula we may conclude that
1 e−ia·ξ − e−ib·ξ

Now F −1
√ (x) = 1[a,b] (x) (36.10)
Z Z 2π iξ
p(ξ)ϕ̂(ξ) = φ(x)p(ξ)e−iξ·x dx = φ(x)p(i∂x )e−iξ·x dx
in the sense of distributions. This also true at the Level of L2 – functions. When
a = −b and b > 0 these formula reduce to
Z
= p(−i∂x )φ(x)e−iξ·x dx = F (p(−i∂)φ) (ξ)
1 eib·ξ − e−ib·ξ 2 sin bξ
F1[−b,b] = √ =√
which combined with the previous equation implies 2π iξ 2π ξ
Z
and
hp̂, φi = F (p(−i∂)φ) (ξ)dξ = F −1 F (p(−i∂)φ) (0) = p(−i∂)φ(0)

2 sin bξ
F −1 √ = 1[−b,b] .
= hδ0 , p(−i∂)φi = hp(i∂)δ0 , φi. 2π ξ
Let us pause to work out Eq. (36.10) by first principles. For M ∈ (0, ∞) let
Thus we have shown that p̂ = p(i∂)δ0 . νN be the complex measure on Rn defined by
Lemma 36.50. Let p(ξ) be a polynomial in ξ ∈ Rn , L = p(−i∂) (a constant
coefficient partial differential operator) and T ∈ S 0 , then dνM (ξ) = √ 1|ξ|≤M dξ,
2π iξ
Fp(−i∂)T = pT̂ .
then
In particular if T = δ0 , we have √ = lim νM in the S 0 topology.
2π iξ M →∞
Fp(−i∂)δ0 = p · δ̂0 = p. Hence


Proof. By definition, F −1 √ (x) = lim F −1 νM
2π iξ M →∞
hFLT, φi = hLT, ϕ̂i = hp(−i∂)T, ϕ̂i = hT, p(i∂)ϕ̂i and

36.6 Wave Equation 421
M −ia·ξ −ib·ξ
−e
Z
1 e Let us again pause to try to compute this inverse Fourier transform directly.
F −1 νM (ξ) = √ eiξx dξ.
−M 2π iξ To this end, let fM (ξ) := sint|ξ|
t|ξ|
1|ξ|≤M . By the dominated convergence theorem,
−ia·ξ −ib·ξ sin t|ξ|
Since is ξ → √12π e −e
iξ eiξx is a holomorphic function on C we may deform it follows that fM → t|ξ| in S 0 , i.e. pointwise on S. Therefore,
the contour to any contour in C starting at −M and ending at M. Let ΓM
denote the straight line path from −M to −1 along the real axis followed by sin t |ξ| sin t |ξ| −1
hF −1 , φi = h , F φi = lim hfM , F −1 φi = lim hF −1 fM , φi
the contour eiθ for θ going from π to 2π and then followed by the straight line t |ξ| t |ξ| M →∞ M →∞
path from 1 to M. Then

and
1 e−ia·ξ − e−ib·ξ iξx 1 e−ia·ξ − e−ib·ξ iξx
Z Z
√ e dξ = √ e dξ sin t |ξ|
Z
3/2 −1 3/2
|ξ|≤M 2π iξ ΓM 2π iξ (2π) F fM (x) = (2π) 1|ξ|≤M eiξ·x dξ
R3 t |ξ|
1 ei(x−a)·ξ − ei(x−b)·ξ
Z
M 2π π
√
Z Z Z
= dξ sin tr ir|x| cos φ 2
ΓM 2π iξ = e r sin φdrdφdθ
r=0 θ=0 φ=0 tr
ei(x−a)·ξ − ei(x−b)·ξ
Z
1 M 2π 1
= dm(ξ).
Z Z Z
sin tr ir|x|u 2
2πi ΓM iξ = e r drdudθ
r=0 θ=0 u=−1 tr
By the usual contour methods we find M
sin tr eir|x| − e−ir|x|
Z
1
Z
eiyξ

1 if y > 0 = 2π rdr
lim dm(ξ) = r=0 t ir |x|
M →∞ 2πi Γ ξ 0 if y < 0 Z M
M 4π
= sin tr sin r |x| dr
and therefore we have t |x| r=0
Z M
4π 1
F −1 √ (x) = lim F −1 νM (x) = 1x>a − 1x>b = 1[a,b] (x). = (cos(r(t + |x|) − cos(r(t − |x|)) dr
2π iξ M →∞ t |x| r=0 2
Example 36.52. Let σt be the surface measure on the sphere St of radius t 4π 1
= (sin(r(t + |x|) − sin(r(t − |x|)) |M
r=0
centered at zero in R3 . Then t |x| 2(t + |x|)

sin t |ξ| 4π 1 sin(M (t + |x|) sin(M (t − |x|)
σ̂t (ξ) = 4πt . = −
|ξ| t |x| 2 t + |x| t − |x|
Indeed, Now make use of the fact that sin M x
→ πδ(x) in one dimension to finish the
Z Z x
proof.
σ̂t (ξ) = e−ix·ξ dσ(x) = t2 e−itx·ξ dσ(x)
tS 2 S2
Z Z 2π Z π
−itx3 |ξ|
=t 2
e dσ(x) = t 2
dθ dφ sin φe−it cos φ|ξ| 36.6 Wave Equation
S2 0 0
1
sin t |ξ|
Z
1
= 2πt2 e−itu|ξ| du = 2πt2 e−itu|ξ| |u=−1
u=1
= 4πt2 . Given a distribution T and a test function φ, we wish to define T ∗ φ ∈ C ∞ by
−1 −it |ξ| t |ξ| the formula
By the inversion formula, it follows that Z
sin t |ξ| t T ∗ φ(x) = “ T (y)φ(x − y)dy” = hT, φ(x − ·)i.
F −1 = σt = tσ̄t
|ξ| 4πt2
As motivation for wanting to understand convolutions of distributions let us
1
where σ̄t is 4πt2 σt , the surface measure on St normalized to have total measure reconsider the wave equation in Rn ,
one.

0 = ∂t2 − ∆ u(t, x) with

We can arrive at this same solution by more elementary means as follows.
u(0, x) = f (x) and ut (0, x) = g(x). We first note in the one dimensional case that wave operator factors, namely
0 = ∂t2 − ∂x2 u(t, x) = (∂t − ∂x ) (∂t + ∂x ) u(t, x).

Taking the Fourier transform in the x variables gives the following equation
2
0 = ût t (t, ξ) + |ξ| û(t, ξ)with Let U (t, x) := (∂t + ∂x ) u(t, x), then the wave equation states (∂t − ∂x ) U = 0
d
û(0, ξ) = fˆ(ξ) and ût (0, ξ) = ĝ(ξ). and hence by the chain rule dt U (t, x − t) = 0. So
The solution to these equations is U (t, x − t) = U (0, x) = g(x) + f 0 (x)

sin t |ξ| and replacing x by x + t in this equation shows
û(t, ξ) = fˆ(ξ) cos (t |ξ|) + ĝ(ξ)
|ξ|
(∂t + ∂x ) u(t, x) = U (t, x) = g(x + t) + f 0 (x + t).
and hence we should have

sin t |ξ| Working similarly, we learn that
u(t, x) = F −1 fˆ(ξ) cos (t |ξ|) + ĝ(ξ) (x)
|ξ|
d
sin t |ξ| u(t, x + t) = g(x + 2t) + f 0 (x + 2t)
= F −1 cos (t |ξ|) ∗ f (x) + F −1 ∗ g (x) dt
|ξ|
d sin t |ξ| sin t |ξ| which upon integration implies
= F −1 ∗ f (x) + F −1 ∗ g (x) .
dt |ξ| |ξ| Z t
u(t, x + t) = u(0, x) + {g(x + 2τ ) + f 0 (x + 2τ )} dτ.
The question now is how interpret this equation. In particular what are the 0
inverse Fourier transforms of F −1 cos (t |ξ|) and F −1 sin|ξ|t|ξ| . Since dt
d
F −1 sin|ξ|t|ξ| ∗
Z t
1
= f (x) + g(x + 2τ )dτ + f (x + 2τ )|t0
f (x) = F −1 cos (t |ξ|) ∗ f (x), it really suffices to understand F −1 sin|ξ|t|ξ| . This 0 2
Z t
was worked out in Example 36.51 when n = 1 where we found 1
= (f (x) + f (x + 2t)) + g(x + 2τ )dτ.
π 2 0
F −1 ξ −1 sin tξ (x) = √

1x+t>0 − 1(x−t)>0
2π Replacing x → x − t in this equation then implies
π π
= √ (1x>−t − 1x>t ) = √ 1[−t,t] (x) Z t
2π 2π 1
u(t, x) = (f (x − t) + f (x + t)) + g(x − t + 2τ )dτ.
where in writing the last line we have assume that t ≥ 0. Therefore, 2 0
1 t Finally, letting y = x − t + 2τ in the last integral gives

Z
F −1 ξ −1 sin tξ ∗ f (x) =

f (x − y)dy
2 −t
1 x+t
Z
1
Therefore the solution to the one dimensional wave equation is u(t, x) = (f (x − t) + f (x + t)) + g(y)dy
2 2 x−t
d 1 t 1 t
Z Z
u(t, x) = f (x − y)dy + g(x − y)dy as derived using the Fourier transform.
dt 2 −t 2 −t For the three dimensional case we have
1 t
Z
1
= (f (x − t) + f (x + t)) + g(x − y)dy d −1 sin t |ξ| sin t |ξ|
2 2 −t u(t, x) = F ∗ f (x) + F −1 ∗ g (x)
dt |ξ| |ξ|
1 x+t
Z
1 d
= (f (x − t) + f (x + t)) + g(y)dy. = (tσ̄t ∗ f (x)) + tσ̄t ∗ g (x) .
2 2 x−t dt

36.6 Wave Equation 423
The question is what is µ ∗ g(x) where µ is a measure. To understand the

definition, suppose first that dµ(x) = ρ(x)dx, then we should have
Z Z
µ ∗ g(x) = ρ ∗ g(x) = g(x − y)ρ(x)dx = g(x − y)dµ(y).
Rn Rn
Thus we expect our solution to the wave equation should be given by

Z Z
d
u(t, x) = t f (x − y)dσ̄t (y) + t g(x − y)dσ̄t (y)
dt St St
Z Z
d
= t f (x − tω)dω + t g(x − tω)dω
dt S1 S1
Z Z
d
= t f (x + tω)dω + t g(x + tω)dω (36.11)
dt S1 S1
where dω := dσ̄1 (ω). Notice the sharp propagation of speed. To understand this Fig. 36.3. The geometry of the solution to the wave equation in two dimensions.
suppose that f = 0 for simplicity and g has compact support near the origin,
for example think of g = δ0 (x), the x + tw = 0 for some w iff |x| = t. Hence the
wave front propagates at unit speed in a sharp way. See figure below. that the solution u(t, x, y, z) is again independent of z and hence is a solution
to the two dimensional wave equation. See figure below.
Notice that we still have finite speed of propagation but no longer sharp
propagation. In fact we can work out the solution analytically as follows. Again
for simplicity assume that f ≡ 0. Then
Z 2π Z π
t
u(t, x, y) = dθ dφ sin φg((x, y) + t(sin φ cos θ, sin φ sin θ))
4π 0 0
Z 2π Z π/2
t
= dθ dφ sin φg((x, y) + t(sin φ cos θ, sin φ sin θ))
2π 0 0
√
and letting u = sin φ, so that du = cos φdφ = 1 − u2 dφ we find
Z 2π Z 1
t du
u(t, x, y) = dθ √ ug((x, y) + ut(cos θ, sin θ))
2π 0 0 1 − u2
Fig. 36.2. The geometry of the solution to the wave equation in three dimensions. and then letting r = ut we learn,
Z 2π Z t
1 dr r
u(t, x, y) = dθ p g((x, y) + r(cos θ, sin θ))
2π 0 0 1 − r /t t
2 2
We may also use this solution to solve the two dimensional wave equation Z 2π Z t
using Hadamard’s method of decent. Indeed, suppose now that f and g are 1 dr
= dθ √ rg((x, y) + r(cos θ, sin θ))
function on R2 which we may view as functions on R3 which do not depend on 2π 0 0 t2 − r 2
ZZ
the third coordinate say. We now go ahead and solve the three dimensional wave 1 g((x, y) + w))
= dm(w).
equation using Eq. (36.11) and f and g as initial conditions. It is easily seen
p
2π Dt t2 − |w|2

Here is a better alternative derivation of this result. We begin by using Definition 36.53 (Indcutive Limit Topology). Let X be a set, Xα ⊂ X
symmetry to find for α ∈ A (A is an index set) and assume that τα ⊂ 2Xα is a topology on Xα
Z Z for each α. Let iα : Xα → X denote the inclusion maps. The inductive limit
u(t, x) = 2t g(x − y)dσ̄t (y) = 2t g(x + y)dσ̄t (y) topology on X is the largest topology τ on X such that iα is continuous for all
St+ St+ α ∈ A. That is to say, τ = ∩α∈A iα∗ (τα ), i.e. a set U ⊂ X is open (U ∈ τ ) iff
i−1
α (A) = A ∩ Xα ∈ τα for all α ∈ A.
where St+ is the√ portion of St with z ≥ 0. This sphere is parametrized
by
R(u, v) = (u, v, t2 − u2 − v 2 ) with (u, v) ∈ Dt := (u, v) : u2 + v 2 ≤ t2 . In Notice that C ⊂ X is closed iff C ∩ Xα is closed in Xα for all α. Indeed,
these coordinates we have C ⊂ X is closed iff C c = X \ C ⊂ X is open, iff C c ∩ Xα = Xα \ C is open in
p p Xα iff Xα ∩ C = Xα \ (Xα \ C) is closed in Xα for all α ∈ A.
4πt2 dσ̄t = −∂u t2 − u2 − v 2 , −∂v t2 − u2 − v 2 , 1 dudv

Definition 36.54. Let D(U ) denote Cc∞ (U ) equipped with the inductive limit
u v
topology arising from writing Cc∞ (U ) as in Eq. (36.12) and using the Frechét
= √ , √ , 1 dudv
t2 − u 2 − v 2 t2 − u 2 − v 2 topologies on C ∞ (K) as defined in Example 36.24.
r
u2 + v 2 |t|
= 2 2 2
+ 1dudv = √ dudv For each K @@ U, C ∞ (K) is a closed subset of D(U ). Indeed if F is another
t −u −v t − u2 − v 2
2
compact subset of U, then C ∞ (K) ∩ C ∞ (F ) = C ∞ (K ∩ F ), which is a closed
and therefore, subset of C ∞ (F ). The set U ⊂ D(U ) defined by
 
|t|
Z
2t p  
u(t, x) = g(x + (u, v, t2 − u2 − v 2 )) √ dudv
X
4πt2 t2 − u2 − v 2 U = ψ ∈ D(U ) : k∂ α (ψ − φ)k∞ < ε (36.13)
St+  
Z |α|≤m
1 g(x + (u, v))
= sgn(t) √ dudv.
2π St+ t2 − u 2 − v 2 for some φ ∈ D(U ) and ε > 0 is an open subset of D(U ). Indeed, if K @@ U,
then  
This may be written as  X 
ZZ U ∩ C ∞ (K) = ψ ∈ C ∞ (K) : k∂ α (ψ − φ)k∞ < ε
1 g((x, y) + w)) 
|α|≤m

u(t, x) = sgn(t) p dm(w)
2π Dt t2 − |w|2
is easily seen to be open in C ∞ (K).
as before. (I should check on the sgn(t) term.)
Proposition 36.55. Let (X, τ ) be as described in Definition 36.53 and f : X →
Y be a function where Y is another topological space. Then f is continuous iff
f ◦ iα : Xα → Y is continuous for all α ∈ A.
36.7 Appendix: Topology on Cc∞ (U )
Proof. Since the composition of continuous maps is continuous, it follows
Let U be an open subset of Rn and that f ◦ iα : Xα → Y is continuous for all α ∈ A if f : X → Y is continuous.
Conversely, if f ◦ iα is continuous for all α ∈ A, then for all V ⊂o Y we have
Cc∞ (U ) = ∪K@@U C ∞ (K) (36.12)
−1
τα 3 (f ◦ iα ) (V ) = i−1
α (f
−1
(V )) = f −1 (V ) ∩ Xα for all α ∈ A
denote the set of smooth functions on U with compact support in U. Our goal
is to topologize Cc∞ (U ) in a way which is compatible with he topologies defined showing that f −1 (V ) ∈ τ.
in Example 36.24 above. This leads us to the inductive limit topology which we
now pause to introduce. Lemma 36.56. Let us continue the notation introduced in Definition 36.53.
Suppose further that there exists αk ∈ A such that Xk0 := Xαk ↑ X as k → ∞
and for each α ∈ A there exists an k ∈ N such that Xα ⊂ Xk0 and the inclusion

36.7 Appendix: Topology on Cc∞ (U ) 425
∞
map is continuous. Then τ = {A ⊂ X : A ∩ Xk0 ⊂o Xk0 for all k} and a function

X |φ(x)|
p(φ) = sup : x ∈ Kk+1 \ Kko .
f : X → Y is continuous iff f |Xk0 : Xk0 → Y is continuous for all k. In short |φk (xk )|
k=0
the inductive limit topology on X arising from the two collections of subsets
{Xα }α∈A and {Xk0 }k∈N are the same. One then checks that
N
!
Proof. Suppose that A ⊂ X, if A ∈ τ then A ∩ Xk0 = A ∩ Xαk ⊂o Xk0 by
X 1
p(φ) ≤ kφk∞
definition. Now suppose that A ∩ Xk0 ⊂o Xk0 for all k. For α ∈ A choose k such k=0
|φk (xk )|
that Xα ⊂ Xk0 , then A ∩ Xα = (A ∩ Xk0 ) ∩ Xα ⊂o Xα since A ∩ Xk0 is open in Xk0
and by assumption that Xα is continuously embedded in Xk0 , V ∩ Xα ⊂o Xα for for φ ∈ C ∞ (KN +1 ). This shows that p|C ∞ (KN +1 ) is continuous for all N and
all V ⊂o Xk0 . The characterization of continuous functions is prove similarly. hence p is continuous on D(U ). Since p is continuous on D(U ) and φk → 0
Let Kk @@ U for k ∈ N such that Kko ⊂ Kk ⊂ Kk+1 o
⊂ Kk+1 for all k and in D(U ), it follows that limk→∞ p(φk ) = p(limk→∞ φk ) = p(0) = 0. While on
Kk ↑ U as k → ∞. Then it follows for any K @@ U, there exists an k such that the other hand, p(φk ) ≥ 1 by construction and hence we have arrived at a
∞
K ⊂ Kko ⊂ Kk . One now checks that the map C ∞ (K) embeds continuously contradiction. Thus for any convergent sequence {φk }k=1 ⊂ D(U ) there is a
into C ∞ (Kk ) and moreover, C ∞ (K) is a closed subset of C ∞ (Kk+1 ). Therefore compact set K @@ U such that supp(φk ) ⊂ K for all k. We will now show that
∞
we may describe D(U ) as Cc∞ (U ) with the inductively limit topology coming {φk }k=1 is convergent to φ in C ∞ (K). To this end let U ⊂ D(U ) be the open
from ∪k∈N C ∞ (Kk ). set described in Eq. (36.13), then φk ∈ U for almost all k and in particular,
φk ∈ U ∩ C ∞ (K) for almost all k. (Letting ε > 0 tend to zero shows that
∞
Lemma 36.57. Suppose that {φk }k=1 ⊂ D(U ), then φk → φ ∈ D(U ) iff φk − supp(φ) ⊂ K, i.e. φ ∈ C ∞ (K).) Since sets of the form U ∩ C ∞ (K) with U as in
φ → 0 ∈ D(U ). Eq. (36.13) form a neighborhood base for the C ∞ (K) at φ, we concluded that
φk → φ in C ∞ (K).
Proof. Let φ ∈ D(U ) and U ⊂ D(U ) be a set. We will begin by showing that
U is open in D(U ) iff U − φ is open in D(U ). To this end let Kk be the compact Definition 36.59 (Distributions on U ⊂o Rn ). A generalized function on
sets described above and choose k0 sufficiently large so that φ ∈ C ∞ (Kk ) for all U ⊂o Rn is a continuous linear functional on D(U ). We denote the space of
k ≥ k0 . Now U − φ ⊂ D(U ) is open iff (U − φ) ∩ C ∞ (Kk ) is open in C ∞ (Kk ) for generalized functions by D0 (U ).
all k ≥ k0 . Because φ ∈ C ∞ (Kk ), we have (U − φ)∩C ∞ (Kk ) = U ∩C ∞ (Kk )−φ Proposition 36.60. Let f : D(U ) → C be a linear functional. Then the fol-
which is open in C ∞ (Kk ) iff U ∩ C ∞ (Kk ) is open C ∞ (Kk ). Since this is true lowing are equivalent.
for all k ≥ k0 we conclude that U − φ is an open subset of D(U ) iff U is open
in D(U ). Now φk → φ in D(U ) iff for all φ ∈ U ⊂o D(U ), φk ∈ U for almost all 1. f is continuous, i.e. f ∈ D0 (U ).
k which happens iff φk − φ ∈ U − φ for almost all k. Since U − φ ranges over all 2. For all K @@ U, there exist n ∈ N and C < ∞ such that
open neighborhoods of 0 when U ranges over the open neighborhoods of φ, the |f (φ)| ≤ Cpn (φ) for all φ ∈ C ∞ (K). (36.14)
result follows.
∞
3. For all sequences {φk } ⊂ D(U ) such that φk → 0 in D(U ), limk→∞ f (φk ) =
Lemma 36.58. A sequence {φk }k=1 ⊂ D(U ) converges to φ ∈ D(U ), iff there 0.
is a compact set K @@ U such that supp(φk ) ⊂ K for all k and φk → φ in
C ∞ (K). Proof. 1) ⇐⇒ 2). If f is continuous, then by definition of the inductive
limit topology f |C ∞ (K) is continuous. Hence an estimate of the type in Eq.
Proof. If φk → φ in C ∞ (K), then for any open set V ⊂ D(U ) with φ ∈ V we (36.14) must hold. Conversely if estimates of the type in Eq. (36.14) hold for all
have V ∩ C ∞ (K) is open in C ∞ (K) and hence φk ∈ V ∩ C ∞ (K) ⊂ V for almost compact sets K, then f |C ∞ (K) is continuous for all K @@ U and again by the
all k. This shows that φk → φ ∈ D(U ). For the converse, suppose that there definition of the inductive limit topologies, f is continuous on D0 (U ). 1) ⇐⇒ 3)
∞
exists {φk }k=1 ⊂ D(U ) which converges to φ ∈ D(U ) yet there is no compact By Lemma 36.58, the assertion in item 3. is equivalent to saying that f |C ∞ (K)
set K such that supp(φk ) ⊂ K for all k. Using Lemma36.57, we may replace is sequentially continuous for all K @@ U. Since the topology on C ∞ (K) is
φk by φk − φ if necessary so that we may assume φk → 0 in D(U ). By passing first countable (being a metric topology), sequential continuity and continuity
to a subsequences of {φk } and {Kk } if necessary, we may also assume there are the same think. Hence item 3. is equivalent to the assertion that f |C ∞ (K)
xk ∈ Kk+1 \ Kk such that φk (xk ) 6= 0 for all k. Let p denote the semi-norm on is continuous for all K @@ U which is equivalent to the assertion that f is
Cc∞ (U ) defined by continuous on D0 (U ).

Proposition 36.61. The maps (λ, φ) ∈ C × D(U ) → λφ ∈ D(U ) and (φ, ψ) ∈
D(U ) × D(U ) → φ + ψ ∈ D(U ) are continuous. (Actually, I will have to look
up how to decide to this.) What is obvious is that all of these operations are
sequentially continuous, which is enough for our purposes.
37
Convolutions involving distributions
37.1 Tensor Product of Distributions By standard arguments, we know that ∂ α fn → ∂ α φ uniformly on Rk as n → ∞.

Moreover for x ∈ Ω, it follows from Eq. (37.3) that
Let X ⊂o Rn and Y ⊂o Rm and S ∈ D0 (X) and T ∈ D0 (Y ). We wish to define
k
S ⊗ T ∈ D0 (X × Y ). Informally, we should have
Z Y
fn (x) := ckn φ(y) (1 − (xi − yi )2 )n dyi = pn (x)
Ω
Z
i=1
hS ⊗ T, φi = S(x)T (y)φ(x, y)dxdy
ZX×Y
Z Z Z where pn (x) is a polynomial in x. Notice that pn ∈ C ∞ ((0, 1))⊗· · ·⊗C ∞ ((0, 1))
= dxS(x) dyT (y)φ(x, y) = dyT (y) dxS(x)φ(x, y). so that we are almost there.1 We need only cutoff these functions so that they
X Y Y X have compact support. To this end, let θ ∈ Cc∞ ((0, 1)) be a function such that
θ = 1 on a neighborhood of [a, b] and define
Of course we should interpret this last equation as follows,
φn = (θ ⊗ · · · ⊗ θ) fn
hS ⊗ T, φi = hS(x), hT (y), φ(x, y)ii = hT (y), hS(x), φ(x, y)ii. (37.1)
= (θ ⊗ · · · ⊗ θ) pn ∈ Cc∞ ((0, 1)) ⊗ · · · ⊗ Cc∞ ((0, 1)).
This formula takes on particularly simple form when φ = u ⊗ v with u ∈ D(X)
and v ∈ D(Y ) in which case I claim now that φn → φ in D(Ω). Certainly by construction supp(φn ) ⊂
[a, b]k @@ Ω for all n. Also
hS ⊗ T, u ⊗ vi = hS, uihT, vi. (37.2)
∂ α (φ − φn ) = ∂ α (φ − (θ ⊗ · · · ⊗ θ) fn )
We begin with the following smooth version of the Weierstrass approximation = (θ ⊗ · · · ⊗ θ) (∂ α φ − ∂ α fn ) + Rn (37.4)
theorem which will be used to show Eq. (37.2) uniquely determines S ⊗ T.
where Rn is a sum of terms of the form ∂ β (θ ⊗ · · · ⊗ θ) · ∂ γ fn with β 6= 0.
Theorem 37.1 (Density Theorem). Suppose that X ⊂o Rn and Y ⊂o Rm , Since ∂ β (θ ⊗ · · · ⊗ θ) = 0 on [a, b]k and ∂ γ fn converges uniformly to zero on
then D(X) ⊗ D(Y ) is dense in D(X × Y ). Rk \[a, b]k , it follows that Rn → 0 uniformly as n → ∞. Combining this with Eq.
Proof. First let us consider the special case where X = (0, 1)n and Y = (37.4) and the fact that ∂ α fn → ∂ α φ uniformly on Rk as n → ∞, we see that
(0, 1)m so that X × Y = (0, 1)m+n . To simplify notation, let m + n = k and φn → φ in D(Ω). This finishes the proof in the case X = (0, 1)n and Y = (0, 1)m .
Ω = (0, 1)k and πi : Ω → (0, 1) be projection onto the ith factor of Ω. Suppose For the general case, let K = supp(φ) @@ X × Y and K1 = π1 (K) @@ X and
that φ ∈ Cc∞ (Ω) and K = supp(φ). We will view φ ∈ Cc∞ (Rk ) by setting φ = 0 1
One could also construct fn ∈ C ∞ (R)⊗k such that ∂ α fn → ∂ α f uniformlly as
outside of Ω. Since K is compact πi (K) ⊂ [ai , bi ] for some 0 < ai < bi < 1. n → ∞ using Fourier series. To this end, let φ̃ be the 1 – periodic extension of φ
Let a = min {ai : i = 1, . . . , k} and b = max {bi : i = 1, . . . , k} . Then supp(φ) = ∞
to Rk . Then φ̃ ∈ Cperiodic (Rk ) and hence it may be written as
K ⊂ [a, b]k ⊂ Ω. As in the proof of the Weierstrass approximation theorem, X
let qn (t) = cn (1 − t2 )n 1|t|≤1 where cn is chosen so that R qn (t)dt = 1. Also set cm ei2πm·x
R
φ̃(x) =
Qk
Qn = qn ⊗ · · · ⊗ qn , i.e. Qn (x) = i=1 qn (xi ) for x ∈ Rk . Let m∈Zk
where the cm : m ∈ Zk are the Fourier coefficients of φ̃ which decay faster that

Z k
(1 + |m|)−l for any l > 0. Thus fn (x) := m∈Zk :|m|≤n cm ei2πm·x ∈ C ∞ (R)⊗k and
Y P
fn (x) := Qn ∗ φ(x) = ckn φ(y) (1 − (xi − yi )2 )n 1|xi −yi |≤1 dyi . (37.3)
Rk i=1
∂ α fn → ∂ α φ unifromly on Ω as n → ∞.
428 37 Convolutions involving distributions
K2 = π2 (K) @@ Y where π1 and π2 are projections from X × Y to X and Y and hence

a b
respectively. Then K @ K1 × K2 @@ X × Y. Let {Vi }i=1 and {Uj }j=1 be finite
∂ β φ(x + te , ·) − ∂ β φ(x, ·)

y i y ∂ β
covers of K1 and K2 respectively by open sets Vi = (ai , bi ) and Uj = (cj , dj ) with − ∂y φ(x, ·)

t ∂xi

∞
ai , bi ∈ X and cj , dj ∈ Y. Also let αi ∈ CcP (Vi ) for i = 1, . . . , a and βj ∈ Cc∞ (Uj )
∞
a
for j = 1, . . . , b be functions such that i=1 αi = 1 on a neighborhood of K1 Z 1
∂ β ∂

β

Pb Pa Pb
and j=1 βj = 1 on a neighborhood of K2 . Then φ = i=1 j=1 (αi ⊗ βj ) φ ≤ ∂xi ∂y φ(x + τ tei , ·) − ∂xi ∂y φ(x, ·) dτ

0 ∞
and by what we have just proved (after scaling and translating) each term in
this sum, (αi ⊗ βj ) φ, may be written as a limit of elements in D(X) ⊗ D(Y ) in which tends to zero as t → 0. Thus we have checked that
the D(X × Y ) topology.
∂ φ(x + tei , ·) − φ(x, ·)
φ(x, ·) = D0 (Y )– lim
Theorem 37.2 (Distribution-Fubini-Theorem). Let S ∈ D0 (X), T ∈ ∂xi t→0 t
D0 (Y ), h(x) := hT (y), φ(x, y)i and g(y) := hS(x), φ(x, y)i. Then h =
hφ ∈ D(X), g = gφ ∈ D(Y ), ∂ α h(x) = hT (y), ∂xα φ(x, y)i and ∂ β g(y) = and therefore,
hS(x), ∂yβ φ(x, y)i for all multi-indices α and β. Moreover h(x + tei ) − h(x) φ(x + tei , ·) − φ(x, ·) ∂
= hT, i → hT, φ(x, ·)i
hS(x), hT (y), φ(x, y)ii = hS, hi = hT, gi = hT (y), hS(x), φ(x, y)ii. (37.5) t t ∂xi
We denote this common value by hS ⊗ T, φi and call S ⊗ T the tensor product of ∂
as t → 0 showing ∂i h(x) exists and is given by hT, ∂x i
φ(x, ·)i. By what
S and T. This distribution is uniquely determined by its values on D(X)⊗D(Y ) ∂
we have proved above, it follows that ∂i h(x) = hT, ∂xi φ(x, ·)i is continuous
and for u ∈ D(X) and v ∈ D(Y ) we have in x. By induction on |α| , it follows that ∂ α h(x) exists and is continuous
hS ⊗ T, u ⊗ vi = hS, uihT, vi. and ∂ α h(x) = hT (y), ∂xα φ(x, y)i for all α. Now if x ∈/ K1 , then φ(x, ·) ≡ 0
showing that {x ∈ X : h(x) 6= 0} ⊂ K1 and hence supp(h) ⊂ K1 @@ X.
Proof. Let K = supp(φ) @@ X × Y and K1 = π1 (K) and K2 = π2 (K). Thus h has compact support. This proves all of the assertions made about
Then K1 @@ X and K2 @@ Y and K ⊂ K1 ×K2 ⊂ X ×Y. If x ∈ X and y ∈ / K2 , h. The assertions pertaining to the function g are prove analogously. Let
then φ(x, y) = 0 and more generally ∂xα φ(x, y) = 0 so that {y : ∂xα φ(x, y) 6= 0} ⊂ hΓ, φi = hS(x), hT (y), φ(x, y)ii = hS, hφ i for φ ∈ D(X × Y ). Then Γ is clearly
K2 . Thus for all x ∈ X, supp(∂ α φ(x, ·)) ⊂ K2 ⊂ Y. By the fundamental theorem linear and we have
of calculus,
Z 1 |hΓ, φi| = |hS, hφ i|
∂yβ φ(x + v, y) − ∂yβ φ(x, y) = ∂vx ∂yβ φ(x + τ v, y)dτ (37.6)
X X
≤C k∂xα hφ k∞,K1 = C khT (y), ∂xα φ(·, y)ik∞,K1
0
|α|≤m |α|≤m
and therefore
β
Z 1 which combined with the estimate
∂y φ(x + v, ·) − ∂yβ φ(x, ·) ≤ |v| ∇x ∂yβ φ(x + τ v, ·) dτ

∞ ∞
X
|hT (y), ∂xα φ(x, y)i| ≤ C ∂yβ ∂xα φ(x, y)i

0 ∞,K 2
≤ |v| ∇x ∂yβ φ ∞ → 0 as ν → 0.

|β|≤p
This shows that x ∈ X → φ(x, ·) ∈ D(Y ) is continuous. Thus h is continuous shows X X

∂yβ ∂xα φ(x, y)i

being the composition of continuous functions. Letting v = tei in Eq. (37.6) we |hΓ, φi| ≤ C .
∞,K 1 ×K2
find |α|≤m |β|≤p
∂yβ φ(x + tei , y) − ∂yβ φ(x, y) ∂ β So Γ is continuous, i.e. Γ ∈ D0 (X × Y ), i.e.

− ∂ φ(x, y)
t ∂xi y
Z 1 φ ∈ D(X × Y ) → hS(x), hT (y), φ(x, y)ii
∂ β ∂ β
= ∂ φ(x + τ tei , y) − ∂ φ(x, y) dτ
0 ∂xi y ∂xi y defines a distribution. Similarly,

37.1 Tensor Product of Distributions 429
Z 1
φ ∈ D(X × Y ) → hT (y), hS(x), φ(x, y)ii
|∂ α φ(x − z) − ∂ α φ(y − z)| ≤ C |x − y| µ−M (z) µM (y + τ (x − y))dτ
0
also defines a distribution and since both of these distributions agree on the
≤ C(x, y) |x − y| µ−M (z) (37.9)
dense subspace D(X) ⊗ D(Y ), it follows they are equal.
where C(x, y) is a continuous function of (x, y). Putting all of this together we
Theorem 37.3. If (T, φ) is a distribution test function pair satisfying one of
see that
the following three conditions
|h(x) − h(y)| ≤ C̃(x, y) |x − y| → 0 as x → y,
1. T ∈ E 0 (Rn ) and φ ∈ C ∞ (Rn ) showing h is continuous. Let us now compute a partial derivative of h. Suppose
2. T ∈ D0 (Rn ) and φ ∈ D(Rn ) or that v ∈ Rn is a fixed vector, then by Eq. (37.8),
3. T ∈ S 0 (Rn ) and φ ∈ S(Rn ),
φ(x + tv − z) − φ(x − z)
let − ∂v φ(x − z)
t
Z
T ∗ φ(x) = “ T (y)φ(x − y)dy” = hT, φ(x − ·)i. (37.7) Z 1
= ∇φ(x + τ tv − z) · vdτ − ∂v φ(x − z)
Then T ∗ φ ∈ C ∞ (Rn ), ∂ α (T ∗ φ) = (∂ α T ∗ φ) = (T ∗ ∂ α φ) for all α and 0
Z 1
supp(T ∗ φ) ⊂ supp(T ) + supp(φ). Moreover if (3) holds then T ∗ φ ∈ P – the
= [∂v φ(x + τ tv − z) − ∂v φ(x − z)] dτ.
space of smooth functions with slow decrease. 0
Proof. I will supply the proof for case (3) since the other cases are similar This then implies
and easier. Let h(x) := T ∗ φ(x). Since T ∈ S 0 (Rn ), there exists m ∈ N and
α φ(x + tv − z) − φ(x − z)
C < ∞ such that |hT, φi| ≤ Cpm (φ) for all φ ∈ S, where pm is defined in

∂z − ∂v φ(x − z)
Example 36.28. Therefore,
t
Z 1
α

|h(x) − h(y)| = |hT, φ(x − ·) − φ(y − ·)i| ≤ Cpm (φ(x − ·) − φ(y − ·)) =
∂z [∂v φ(x + τ tv − z) − ∂v φ(x − z)] dτ
X 0
=C kµm (∂ α φ(x − ·) − ∂ α φ(y − ·))k∞ . Z 1
|α|≤m ≤ |∂zα [∂v φ(x + τ tv − z) − ∂v φ(x − z)]| dτ.

0
Let ψ := ∂ α φ, then But by the same argument as above, it follows that

Z 1
ψ(x − z) − ψ(y − z) = ∇ψ(y + τ (x − y) − z) · (x − y)dτ (37.8) |∂zα [∂v φ(x + τ tv − z) − ∂v φ(x − z)]| ≤ C(x + τ tv, x) |τ tv| µ−M (z)
0
and thus
and hence
α φ(x + tv − z) − φ(x − z)
Z 1 ∂z − ∂v φ(x − z)
|ψ(x − z) − ψ(y − z)| ≤ |x − y| · |∇ψ(y + τ (x − y) − z)| dτ
t
0 Z 1
Z 1 ≤ tµ−M (z) C(x + τ tv, x)τ dτ |v| µ−M (z).
≤ C |x − y| µ−M (y + τ (x − y) − z)dτ 0
0
Putting this all together shows
for any M < ∞. By Peetre’s inequality,
µM ∂zα φ(x + tv − z) − φ(x − z) − ∂v φ(x − z) = O(t)

µ−M (y + τ (x − y) − z) ≤ µ−M (z)µM (y + τ (x − y)) t
∞
so that → 0 as t → 0.

That is to say φ(x+tv−·)−φ(x−·)

t → ∂v φ(x − ·) in S as t → 0. Hence since T is Theorem 37.4. Suppose that if (T, φ) is a distribution test function pair satis-
continuous on S, we learn fying one the three condition in Theorem 37.3, then T ∗ φ as a distribution may
φ(x + tv − ·) − φ(x − ·) be characterized by
∂v (T ∗ φ) (x) = ∂v hT, φ(x − ·)i = lim hT, i hT ∗ φ, ψi = hT, φ̃ ∗ ψi (37.11)
t→0 t
= hT, ∂v φ(x − ·)i = T ∗ ∂v φ(x). for all ψ ∈ D(Rn ). Moreover, if T ∈ S 0 and φ ∈ S then Eq. (37.11) holds for
all ψ ∈ S.
By the first part of the proof, we know that ∂v (T ∗ φ) is continuous and hence
by induction it now follows that T ∗ φ is C ∞ and ∂ α T ∗ φ = T ∗ ∂ α φ. Since Proof. Let us first assume that T ∈ D0 and φ, ψ ∈ D and θ ∈ D be a
function such that θ = 1 on a neighborhood of the support of ψ. Then
T ∗ ∂ α φ(x) = hT (z), (∂ α φ) (x − z)i = (−1)α hT (z), ∂zα φ(x − z)i Z
= h∂zα T (z), φ(x − z)i = ∂ α T ∗ φ(x) hT ∗ φ, ψi = hT, φ(x − ·)iψ(x)dx = hψ(x), hT (y), φ(x − y)ii
Rn
the proof is complete except for showing T ∗ φ ∈ P. For the last statement, it = hθ(x)ψ(x), hT (y), φ(x − y)ii
suffices to prove |T ∗ φ(x)| ≤ CµM (x) for some C < ∞ and M < ∞. This goes = hψ(x), θ(x)hT (y), φ(x − y)ii
as follows
X = hψ(x), hT (y), θ(x)φ(x − y)ii.
|h(x)| = |hT, φ(x − ·)i| ≤ Cpm (φ(x − ·)) = C kµm (∂ α φ(x − ·)k∞
Now the function, θ(x)φ(x−y) ∈ D(Rn ×Rn ), so we may apply Fubini’s theorem
|α|≤m
for distributions to conclude that
and using Peetre’s inequality, |∂ α φ(x − z)| ≤ Cµ−m (x − z) ≤ Cµ−m (z)µm (x)
so that hT ∗ φ, ψi = hψ(x), hT (y), θ(x)φ(x − y)ii
kµm (∂ α φ(x − ·)k∞ ≤ Cµm (x). = hT (y), hψ(x), θ(x)φ(x − y)ii
Thus it follows that |T ∗ φ(x)| ≤ Cµm (x) for some C < ∞. If x ∈ Rn \ = hT (y), hθ(x)ψ(x), φ(x − y)ii
(supp(T ) + supp(φ)) and y ∈ supp(φ) then x − y ∈/ supp(T ) for otherwise = hT (y), hψ(x), φ(x − y)ii
x = x − y + y ∈ supp(T ) + supp(φ). Thus
= hT (y), ψ ∗ φ̃(y)i = hT, ψ ∗ φ̃i
supp(φ(x − ·)) = x − supp(φ) ⊂ Rn \ supp(T )
as claimed. If T ∈ E 0 , let α ∈ D(Rn ) be a function such that α = 1 on a
n neighborhood of supp(T ), then working as above,
and hence h(x) = hT, φ(x − ·)i = 0 for all x ∈ R \ (supp(T ) + supp(φ)) . This
implies that {h 6= 0} ⊂ supp(T ) + supp(φ) and hence
hT ∗ φ, ψi = hψ(x), hT (y), θ(x)φ(x − y)ii
supp(h) = {h 6= 0} ⊂ supp(T ) + supp(φ). = hψ(x), hT (y), α(y)θ(x)φ(x − y)ii
and since α(y)θ(x)φ(x − y) ∈ D(Rn × Rn ) we may apply Fubini’s theorem for
As we have seen in the previous theorem, T ∗ φ is a smooth function and distributions to conclude again that
hence may be used to define a distribution in D0 (Rn ) by
Z Z hT ∗ φ, ψi = hT (y), hψ(x), α(y)θ(x)φ(x − y)ii
hT ∗ φ, ψi = T ∗ φ(x)ψ(x)dx = hT, φ(x − ·)iψ(x)dx. = hα(y)T (y), hθ(x)ψ(x), φ(x − y)ii
= hT (y), hψ(x), φ(x − y)ii = hT, ψ ∗ φ̃i.
Using the linearity of T we might expect that
Z Z Now suppose that T ∈ S 0 and φ, ψ ∈ S. Let φn , ψn ∈ D be a sequences such
hT, φ(x − ·)iψ(x)dx = hT, φ(x − ·)ψ(x)dxi that φn → φ and ψn → ψ in S, then using arguments similar to those in the
proof of Theorem 37.3, one shows
or equivalently that
hT ∗ φ, ψi = hT, φ̃ ∗ ψi (37.10) hT ∗ φ, ψi = lim hT ∗ φn , ψn i = lim hT, ψn ∗ φ̃n i = hT, ψ ∗ φ̃i.
n→∞ n→∞
where φ̃(x) := φ(−x).

37.1 Tensor Product of Distributions 431
Theorem 37.5. Let U ⊂o Rn , then D(U ) is sequentially dense in E 0 (U ). When Definition 37.6 (Convolution of Distributions). Suppose that T ∈ D0 and
U = Rn we have E 0 (Rn ) is a dense subspace of S 0 (Rn ) ⊂ D0 (Rn ). Hence we S ∈ E 0 , then define T ∗ S ∈ D0 by
have the following inclusions,
hT ∗ S, φi = hT ⊗ S, φ+ i
D(U ) ⊂ E 0 (U ) ⊂ D0 (U ),
D(Rn ) ⊂ E 0 (Rn ) ⊂ S 0 (Rn ) ⊂ D0 (Rn ) and where φ+ (x, y) = φ(x + y) for all x, y ∈ Rn . More generally we may define T ∗ S
for any two distributions having the property that supp(T ⊗ S) ∩ supp(φ+ ) =
D(Rn ) ⊂ S(Rn ) ⊂ S 0 (Rn ) ⊂ D0 (Rn )
[supp(T ) × supp(S)] ∩ supp(φ+ ) is compact for all φ ∈ D.
with all inclusions being dense in the next space up.
Proposition 37.7. Suppose that T ∈ D0 and S ∈ E 0 then T ∗ S is well defined
Proof. The key point is to show D(U )Ris dense in E 0 (U ). Choose θ ∈ Cc∞ (Rn ) and
such that supp(θ) ⊂ B(0, 1), θ = θ and θ(x)dx = 1. Let θm (x) = m−n θ(mx)
so that supp(θm ) ⊂ B(0, 1/m). An element in T ∈ E 0 (U ) may be viewed as an hT ∗ S, φi = hT (x), hS(y), φ(x + y)ii = hS(y), hT (x), φ(x + y)ii. (37.12)
element in E 0 (Rn ) in a natural way. Namely if χ ∈ Cc∞ (U ) such that χ = 1
on a neighborhood of supp(T ), and φ ∈ C ∞ (Rn ), let hT, φi = hT, χφi. Define Moreover, if T ∈ S 0 then T ∗ S ∈ S 0 and F(T ∗ S) = Ŝ T̂ . Recall from Remark
Tm = T ∗ θm . It is easily seen that supp(Tn ) ⊂ supp(T ) + B(0, 1/m) ⊂ U for 36.46 that Ŝ ∈ P so that Ŝ T̂ ∈ S 0 .
all m sufficiently large. Hence Tm ∈ D(U ) for large enough m. Moreover, if
Proof. Let θ ∈ D be a function such that θ = 1 on a neighborhood of
ψ ∈ D(U ), then
supp(S), then by Fubini’s theorem for distributions,
hTm , ψi = hT ∗ θm , ψi = hT, θm ∗ ψi = hT, θm ∗ ψi → hT, ψi
hT ⊗ S, φ+ i = hT ⊗ S(x, y), θ(y)φ(x + y)i = hT (x)S(y), θ(y)φ(x + y)i
since θm ∗ ψ → ψ in D(U ) by standard arguments. If U = Rn , T ∈ E 0 (Rn ) ⊂ = hT (x), hS(y), θ(y)φ(x + y)ii = hT (x), hS(y), φ(x + y)ii
S 0 (Rn ) and ψ ∈ S, the same argument goes through to show hTm , ψi → hT, ψi
provided we show θm ∗ ψ → ψ in S(Rn ) as m → ∞. This latter is proved by and
showing for all α and t > 0, I
hT ⊗ S, φ+ i = hT (x)S(y), θ(y)φ(x + y)i = hS(y), hT (x), θ(y)φ(x + y)ii
kµt (∂ α θm ∗ ψ − ∂ α ψ)k∞ → 0 as m → ∞,
= hS(y), θ(y)hT (x), φ(x + y)ii = hS(y), hT (x), φ(x + y)ii
which is a consequence of the estimates:
proving Eq. (37.12). Suppose that T ∈ S 0 , then
|∂ α θm ∗ ψ(x) − ∂ α ψ(x)| = |θm ∗ ∂ α ψ(x) − ∂ α ψ(x)|
X
kµm ∂xα hS(y), φ(· + y)ik∞
Z
α α
|hT ∗ S, φi| = |hT (x), hS(y), φ(x + y)ii| ≤ C
= θm (y) [∂ ψ(x − y) − ∂ ψ(x)] dy

|α|≤m
sup |∂ α ψ(x − y) − ∂ α ψ(x)|

X
≤ α
=C kµm hS(y), ∂ φ(· + y)ik∞
|y|≤1/m
|α|≤m
1
≤ sup |∇∂ α ψ(x − y)|
m |y|≤1/m and
1 X
|hS(y), ∂ α φ(x + y)i| ≤ C sup ∂ β ∂ α φ(x + y)

≤ C sup µ−t (x − y)
m |y|≤1/m |β|≤p
y∈K
1 ≤ Cpm+p (φ) sup µ−m−p (x + y)

≤ Cµ−t (x − y) sup µt (y)
m |y|≤1/m y∈K
1 t ≤ Cpm+p (φ)µ−m−p (x) sup µm+p (y)

≤ C 1 + m−1 µ−t (x).

y∈K
m
= C̃µ−m−p (x)pm+p (φ).

Combining the last two displayed equations shows hF(T ∗ S), φi = hS(y), θ(y)hT̂ (ξ), φ(ξ)e−iξ·y ii
|hT ∗ S, φi| ≤ Cpm+p (φ) = hS(y), hT̂ (ξ), φ(ξ)θ(y)e−iξ·y ii

= hT̂ (ξ), hS(y), φ(ξ)θ(y)e−iξ·y ii
0
which shows that T ∗ S ∈ S . We still should check that
= hT̂ (ξ), φ(ξ)hS(y), e−iξ·y ii
hT ∗ S, φi = hT (x), hS(y), φ(x + y)ii = hS(y), hT (x), φ(x + y)ii = hT̂ (ξ), φ(ξ)Ŝ(ξ)i = hŜ(ξ)T̂ (ξ), φ(ξ)i. (37.16)
still holds for all φ ∈ S. This is a matter of showing that all of the expressions Since F(T ∗ S) ∈ S 0 and Ŝ T̂ ∈ S 0 , we conclude that Eq. (37.16) holds for all
are continuous in S when restricted to D. Explicitly, let φm ∈ D be a sequence φ ∈ S and hence F(T ∗ S) = Ŝ T̂ as was to be proved.
of functions such that φm → φ in S, then
hT ∗ S, φi = lim hT ∗ S, φn i = lim hT (x), hS(y), φn (x + y)ii (37.13)

n→∞ n→∞ 37.2 Elliptic Regularity
and P α
Theorem 37.8 (Hypoellipticity). Suppose that p(x) = |α|≤m aα ξ is a
n
hT ∗ S, φi = lim hT ∗ S, φn i = lim hS(y), hT (x), φn (x + y)ii. (37.14) polynomial on R and L is the constant coefficient differential operator
n→∞ n→∞
1 X 1 X
So it suffices to show the map φ ∈ S → hS(y), φ(· + y)i ∈ S is continuous and L = p( ∂) = aα ( ∂)α = aα (−i∂)α .
i i
φ ∈ S → hT (x), φ(x + ·)i ∈ C ∞ (Rn ) are continuous maps. These may verified |α|≤m |α|≤m
by methods similar to what we have been doing, so I will leave the details to
the reader. Given these continuity assertions, we may pass to the limits in Eq. Also assume there exists a distribution T ∈ D0 (Rn ) such that R := δ − LT ∈
(37.13d (37.14) to learn C ∞ (Rn ) and T |Rn \{0} ∈ C ∞ (Rn \ {0}). Then if v ∈ C ∞ (U ) and u ∈ D0 (U )
solves Lu = v then u ∈ C ∞ (U ). In particular, all solutions u to the equation
hT ∗ S, φi = hT (x), hS(y), φ(x + y)ii = hS(y), hT (x), φ(x + y)ii Lu = 0 are smooth.
still holds for all φ ∈ S. The last and most important point is to show F(T ∗S) = Proof. We must show for each x0 ∈ U that u is smooth on a neighborhood of
Ŝ T̂ . Using x0 . So let x0 ∈ U and θ ∈ D(U ) such that 0 ≤ θ ≤ 1 and θ = 1 on neighborhood
Z Z V of x0 . Also pick α ∈ D(V ) such that 0 ≤ α ≤ 1 and α = 1 on a neighborhood
−iξ·(x+y) of x0 . Then
ϕ̂(x + y) = φ(ξ)e dξ = φ(ξ)e−iξ·y e−iξ·x dξ
R n R n
θu = δ ∗ (θu) = (LT + R) ∗ (θu) = (LT ) ∗ (θu) + R ∗ (θu)
= F φ(ξ)e−iξ·y (x)

= T ∗ L (θu) + R ∗ (θu)
and the definition of F on S 0 we learn = T ∗ {αL (θu) + (1 − α)L (θu)} + R ∗ (θu)
= T ∗ {αLu + (1 − α)L (θu)} + R ∗ (θu)
hF(T ∗ S), φi = hT ∗ S, ϕ̂i = hS(y), hT (x), ϕ̂(x + y)ii
= T ∗ (αv) + R ∗ (θu) + T ∗ [(1 − α)L (θu)] .
= hS(y), hT (x), F φ(ξ)e−iξ·y (x)ii

= hS(y), hT̂ (ξ), φ(ξ)e−iξ·y ii. (37.15) Since αv ∈ D(U ) and T ∈ D0 (Rn ) it follows that R ∗ (θu) ∈ C ∞ (Rn ). Also
since R ∈ C ∞ (Rn ) and θu ∈ E 0 (U ), R ∗ (θu) ∈ C ∞ (Rn ). So to show θu, and
Let θ ∈ D be a function such that θ = 1 on a neighborhood of supp(S) and hence u, is smooth near x0 it suffices to show T ∗ g is smooth near x0 where
assume φ ∈ D for the moment. Then from Eq. (37.15) and Fubini’s theorem for g := (1 − α)L (θu) . Working formally for the moment,
distributions we find Z Z
T ∗ g(x) = T (x − y)g(y)dy = T (x − y)g(y)dy
Rn Rn \{α=1}

37.2 Elliptic Regularity 433
which should be smooth for x near x0 since in this case x−y 6= 0 when g(y) 6= 0. for all |ξ| ≥ M with M sufficiently large since limξ→∞ |r(ξ)|
|ξ|m = 0. Choose θ ∈
To make this precise, let δ > 0 be chosen
c
so that α = 1 on a neighborhood of n
D(R ) such that θ = 1 on a neighborhood of B(0, M ) and let
B(x0 , δ) so that supp(g) ⊂ B(x0 , δ) . For φ ∈ D(B(x0 , δ/2),
1 − θ(ξ) β(ξ)
hT ∗ g, φi = hT (x), hg(y), φ(x + y)ii = hT, hi h(ξ) = = ∈ C ∞ (Rn )
p(ξ) p(ξ)
where h(x) := hg(y), φ(x + y)i. If |x| ≤ δ/2 where β = 1 − θ. Since h(ξ) is bounded (in fact limξ→∞ h(ξ) = 0), h ∈ S 0 (Rn )
so there exists T := F −1 h ∈ S 0 (Rn ) is well defined. Moreover,
supp(φ(x + ·)) = supp(φ) − x ⊂ B(x0 , δ/2) − x ⊂ B(x0 , δ)
c F (δ − LT ) = 1 − p(ξ)h(ξ) = 1 − β(ξ) = θ(ξ) ∈ D(Rn )
so that h(x) = 0 and hence supp(h) ⊂ B(x0 , δ/2) . Hence if we let γ ∈
D(B(0, δ/2)) be a function such that γ = 1 near 0, we have γh ≡ 0, and which shows that
thus R := δ − LT ∈ S(Rn ) ⊂ C ∞ (Rn ).
hT ∗ g, φi = hT, hi = hT, h − γhi = h(1 − γ)T, hi = h[(1 − γ)T ] ∗ g, φi. So to finish the proof it suffices to show
Since this last equation is true for all φ ∈ D(B(x0 , δ/2)), T ∗ g = [(1 − γ)T ] ∗ g T |Rn \{0} ∈ C ∞ (Rn \ {0}).
on B(x0 , δ/2) and this finishes the proof since [(1 − γ)T ] ∗ g ∈ C ∞ (Rn ) because
To prove this recall that
(1 − γ)T ∈ C ∞ (Rn ).
Definition 37.9. Suppose that p(x) = |α|≤m aα ξ α is a polynomial on Rn and

P F (xα T ) = (i∂)α T̂ = (i∂)α h.
L is the constant coefficient differential operator
By the chain
c
rule and the fact that any derivative of β is has compact support
1 X 1 X in B(0, M ) and any derivative of p1 is non-zero on this set,
L = p( ∂) = aα ( ∂)α = aα (−i∂)α .
i i
|α|≤m |α|≤m 1
∂ α h = β∂ α + rα
P α p
Let σp (L)(ξ) := |α|=m aα ξ and call σp (L) the principle symbol of L. The
operator L is said to be elliptic provided that σp (L)(ξ) 6= 0 if ξ 6= 0. where rα ∈ D(Rn ). Moreover,
Theorem 37.10 (Existence of Parametrix). Suppose that L = p( 1i ∂) is an 1 ∂i p 1 ∂i p ∂j ∂i p ∂i p

∂i = − 2 and ∂j ∂i = −∂j 2 = − 2 + 2 3
elliptic constant coefficient differential operator, then there exists a distribution p p p p p p
T ∈ D0 (Rn ) such that R := δ − LT ∈ C ∞ (Rn ) and T |Rn \{0} ∈ C ∞ (Rn \ {0}).
from which it follows that
Proof. The idea is to try to find T such that LT = δ. Taking the Fourier
β(ξ)∂i 1 (ξ) ≤ C |ξ|−(m+1) and β(ξ)∂j ∂i 1 ≤ C |ξ|−(m+2) .

transform of this equation implies that p(ξ)T̂ (ξ) = 1 and hence we should try
to define T̂ (ξ) = 1/p(ξ). The main problem with this definition is that p(ξ)
p p
may have zeros. However, these zeros can not Poccur for large ξ by the ellipticity More generally, one shows by inductively that
assumption. Indeed, let q(ξ) := σp (L)(ξ) = |α|=m aα ξ α , r(ξ) = p(ξ) − q(ξ) =
α
β(ξ)∂ α 1 ≤ C |ξ|−(m+|α|) .
P
|α|<m aα ξ and let c = min {|q(ξ)| : |ξ| = 1} ≤ max {|q(ξ)| : |ξ| = 1} =: C.

(37.17)
Then because |q(·)| is a nowhere vanishing continuous function on the compact p
set S := {ξ ∈ Rn : |ξ| = 1|} , 0 < c ≤ C < ∞. For ξ ∈ Rn , let ξˆ = ξ/ |ξ| and
In particular, if k ∈ N is given and α is chosen so that |α| + m > n + k, then
notice k
|ξ| ∂ α h(ξ) ∈ L1 (ξ) and therefore
m m |r(ξ)|
|p(ξ)| = |q(ξ)| − |r(ξ)| ≥ c |ξ| − |r(ξ)| = |ξ| (c − m )>0 xα T = F −1 [(i∂)α h] ∈ C k (Rn ).
|ξ|

Hence we learn for any k ∈ N, we may choose p sufficiently large so that
2p
|x| T ∈ C k (Rn ).
This shows that T |Rn \{0} ∈ C ∞ (Rn \ {0}).

Here is the induction argument that proves Eq. (37.17). Let qα :=
p|α|+1 ∂ α p−1 with q0 = 1, then

∂i ∂ α p−1 = ∂i p−|α|−1 qα = (− |α| − 1) p−|α|−2 qα ∂i p + p−|α|−1 ∂i qα
so that
qα+ei = p|α|+2 ∂i ∂ α p−1 = (− |α| − 1) qα ∂i p + p∂i qα .
It follows by induction that qα is a polynomial in ξ and letting dα := deg(qα ),
we have dα+ei ≤ dα + m − 1 with d0 = 1. Again by induction this implies
dα ≤ |α| (m − 1). Therefore
qα d −m(|α|+1) |α|(m−1)−m(|α|+1) −(m+|α|)
∂ α p−1 = ∼ |ξ| α = |ξ| = |ξ|
p|α|+1
as claimed in Eq. (37.17).

Part IX
Appendices
A
Multinomial Theorems and Calculus Results
Given a multi-index α ∈ Zn+ , let |α| = α1 + · · · + αn , α! := α1 ! · · · αn !, Now suppose that a, b ∈ Rn and α is a multi-index, we have
n
∂
α n
∂
αj X α! X α!
α (a + b)α = aβ bα−β = aβ bδ (A.2)
Y Y
α
x := xj j and ∂xα = := . β!(α − β)! β!δ!
j=1
∂x j=1
∂xj β≤α β+δ=α
We also write Indeed, by the standard Binomial formula,

d
∂v f (x) := f (x + tv)|t=0 . X αi !
dt (ai + bi )αi = aβi bαi −βi
βi !(αi − βi )!
βi ≤αi
A.1 Multinomial Theorems and Product Rules from which Eq. (A.2) follows. Eq. (A.2) generalizes in the obvious way to
For a = (a1 , a2 , . . . , an ) ∈ Cn , m ∈ N and (i1 , . . . , im ) ∈ {1, 2, . . . , n}m let α

X α!
(a1 + · · · + ak ) = aβ1 . . . aβkk (A.3)
α̂j (i1 , . . . , im ) = # {k : ik = j} . Then β1 ! · · · βk ! 1
β1 +···+βk =α
n
!m n
X X X where a1 , a2 , . . . , ak ∈ Rn and α ∈ Zn+ .
ai = ai1 . . . aim = C(α)aα Now let us consider the product rule for derivatives. Let us begin with the
i=1 i1 ,...,im =1 |α|=m dn
one variable case (write dn f for f (n) = dx n f ) where we will show by induction
where that
n
n
X n k
d (f g) = d f · dn−k g. (A.4)
C(α) = # {(i1 , . . . , im ) : α̂j (i1 , . . . , im ) = αj for j = 1, 2, . . . , n} k
k=0
I claim that C(α) = m! Indeed assuming Eq. (A.4) we find

α! .
Indeed, one possibility for such a sequence (a1 , . . . , aim )
for a given α is gotten by choosing n n
X n X n
α1 α2 αn dn+1 (f g) = dk+1 f · dn−k g + dk f · dn−k+1 g
k k
z }| { z }| { z }| { k=0 k=0
(a1 , . . . , a1 , a2 , . . . , a2 , . . . , an , . . . , an ).
n+1 n
X n k n−k+1
X n k
Now there are m! permutations of this list. However, only those permutations = d f ·d g+ d f · dn−k+1 g
k−1 k
leading to a distinct list are to be counted. So for each of these m! permutations k=1 k=0
n+1
we must divide by the number of permutation which just rearrange the groups of X n n
ai ’s among themselves for each i. There are α! := α1 ! · · · αn ! such permutations. = + dk f · dn−k+1 g + dn+1 f · g + f · dn+1 g.
k−1 k
k=1
Therefore, C(α) = m!/α! as advertised. So we have proved
n
!m Since
X X m!
ai = aα . (A.1)
i=1
α!
|α|=m
6 A Multinomial Theorems and Calculus Results
t t

n n n! n!
Z Z
d
+ = + f (x(t)) = f (x(0)) + f (x(s))ds = f (x(0)) + ∂ẋ(s) f (x(s))ds (A.8)
k−1 k (n − k + 1)!(k − 1)! (n − k)!k! 0 ds 0

n! 1 1 and in particular,
= +
(k − 1)! (n − k)! (n − k + 1) k Z 1

n! n+1 n+1 f (x(1)) = f (x(0)) + ∂ẋ(s) f (x(s))ds.
= = 0
(k − 1)! (n − k)! (n − k + 1) k k
This proves Eq. (A.6) when N = 1. We will now complete the proof using
the result follows. induction on N. Applying Eq. (A.8) with f replaced by (N −1)! 1
∂ẋ(s) ∂vNs−1 f
Now consider the multi-variable case gives
n
! n
" α #
i
α 1 1
∂ẋ(s) ∂vNs−1 f (x(s)) = ∂ẋ(s) ∂vNs−1 f (x(0))

i
Y Y X
α k α −k
∂ α (f g) = ∂i i (f g) = ∂i i f · ∂i i i g (N − 1)! (N − 1)!
i=1 i=1 ki =0
ki Z s
1
∂ẋ(s) ∂vNs−1 ∂ẋ(t) f (x(t))dt

α1 αn Yn X α +
X X αi k (N − 1)! 0
= ··· ∂ f · ∂ α−k g = ∂ k f · ∂ α−k g
i=1
ki k 1

d N

1
Z s
d N

k1 =0 kn =0 k≤α
=− ∂ f (x(0)) − ∂ ∂ẋ(t) f (x(t))d
N ! ds vs N ! 0 ds vs
where k = (k1 , k2 , . . . , kn ) and
wherein we have used the fact that mixed partial derivatives commute to show
n d N N −1
ds ∂vs f = N ∂ẋ(s) ∂vs f. Integrating this equation on s ∈ [0, 1] shows, using the

α Y αi α!
:= = . fundamental theorem of calculus,
k i=1
ki k!(α − k)!
Z
1 N
1 d N
So we have proved RN = ∂ f (x(0)) − ∂ ∂ẋ(t) f (x(t))dsdt
X α N! v N ! 0≤t≤s≤1 ds vs
α
∂ β f · ∂ α−β g.
Z
∂ (f g) = (A.5) 1 1
∂ N f (x(0)) + ∂ N ∂ẋ(t) f (x(t))dt

β =
β≤α N! v (N + 1)! 0≤t≤1 wt
1
∂vN f (x(0)) + RN +1

=
N!
A.2 Taylor’s Theorem
which completes the inductive proof.
Theorem A.1. Suppose X ⊂ Rn is an open set, x : [0, 1] → X is a C 1 – path, Remark A.2. Using Eq. (A.1) with ai replaced by vi ∂i (although {vi ∂i }i=1 are
n
and f ∈ C N (X, C). Let vs := x(1) − x(s) and v = v1 = x(1) − x(0), then not complex numbers they are commuting symbols), we find
n
!m
N −1 X m!
1 m
X
vα ∂ α .
X
f (x(1)) = (∂vm f ) (x(0)) + RN (A.6) ∂v f = vi ∂i f=
m! i=1
α!
m=0 |α|=m
where Using this fact we may write Eqs. (A.6) and (A.7) as
1 Z 1 X 1
v α ∂ α f (x(0)) + RN
Z
1 1 d N f (x(1)) =
∂ẋ(s) ∂vNs−1 f (x(s))ds =

RN = − ∂ f (x(s))ds. α!
(N − 1)! 0 N! 0 ds vs |α|≤N −1
(A.7)
and
and 0! := 1. X 1 Z 1 d
RN = − vsα ∂ α f (x(s))ds.
Proof. By the fundamental theorem of calculus and the chain rule, α! 0 ds
|α|=N

A.2 Taylor’s Theorem 7
Corollary A.3. Suppose X ⊂ Rn is an open set which contains x(s) = (1 − Now for x ∈ (−1, 1) and N > β,
s)x0 + sx1 for 0 ≤ s ≤ 1 and f ∈ C N (X, C). Then
1 1 1
N (1 − s)N −1 N (1 − s)N −1
Z Z Z
N
N −1 1 0≤ ds ≤ ds = N (1 − s)β−1 ds =
(1 − sx)N −β (1 − s)N −β
Z
X 1 1 β
(∂vm f ) (x0 ) + ∂vN f (x(s))dνN (s) 0 0 0

f (x1 ) = (A.9)
m=0
m! N! 0
and therefore,
X 1 Z 1
X 1
= ∂ α f (x(0))(x1 − x0 )α + ∂ f (x(s))dνN (s) (x1 − x0 )α
α
|x|
N
α! α! 0 |RN (x)| ≤ |(β − 1) . . . (β − N + 1)| =: ρN .
|α|<N α:|α|=N
(N − 1)!
(A.10)
Since
where v := x1 − x0 and dνN is the probability measure on [0, 1] given by ρN +1 N −β
lim sup = |x| · lim sup = |x| < 1
N −1 N →∞ ρ N N →∞ N
dνN (s) := N (1 − s) ds. (A.11)
and so by the Ratio test, |RN (x)| ≤ ρN → 0 (exponentially fast) as N → ∞.
If we let x = x0 and y = x1 − x0 (so x + y = x1 ) Eq. (A.10) may be written as Therefore by passing to the limit in Eq. (A.13) we have proved
X ∂ α f (x) X 1 Z 1
∞
f (x + y) = x
yα + ∂xα f (x + sy)dνN (s) y α . (A.12) X (−1)m
α! α! 0 (1 − x)β = 1 + β(β − 1) . . . (β − m + 1)xm (A.14)
|α|<N α:|α|=N
m=1
m!
Proof. This is a special case of Theorem A.1. Notice that
which is valid for |x| < 1 and β ∈ R. An important
P∞ special cases is β = −1
1
vs = x(1) − x(s) = (1 − s)(x1 − x0 ) = (1 − s)v in which case, Eq. (A.14) becomes 1−x = m=0 xm , the standard geometric
series formula. Another another useful special case is β = 1/2 in which case Eq.
and hence (A.14) becomes
Z 1 Z 1
1 d 1 √ ∞
− (1 − s)N ∂vN f ∂vN f (x(s))N (1−s)N −1 ds. (−1)m 1 1

RN = (x(s))ds = X 1
N! 0 ds N! 0 1−x=1+ ( − 1) . . . ( − m + 1)xm
m=1
m! 2 2 2
∞
X (2m − 3)!! m
Example A.4. Let X = (−1, 1) ⊂ R, β ∈ R and f (x) = (1 − x) . The reader β =1− x for all |x| < 1. (A.15)
m=1
2m m!
should verify
f (m) (x) = (−1)m β(β − 1) . . . (β − m + 1)(1 − x)β−m
and therefore by Taylor’s theorem (Eq. (??) with x = 0 and y = x)

N −1
X 1
(1 − x)β = 1 + (−1)m β(β − 1) . . . (β − m + 1)xm + RN (x) (A.13)
m=1
m!
where
1
xN
Z
RN (x) = (−1)N β(β − 1) . . . (β − N + 1)(1 − sx)β−N dνN (s)
N! 0
1
xN N (1 − s)N −1
Z
= (−1)N β(β − 1) . . . (β − N + 1) ds.
N! 0 (1 − sx)N −β

B
Zorn’s Lemma and the Hausdorff Maximal Principle
Definition B.1. A partial order ≤ on X is a relation with following properties 2. The Hausdorff Maximal Principle: Every partially ordered set has a
maximal (relative to the inclusion order) linearly ordered subset.
(i) If x ≤ y and y ≤ z then x ≤ z. 3. Zorn’s Lemma: If X is partially ordered set such that every linearly or-
(ii)If x ≤ y and y ≤ x then x = y. dered subset of X has an upper bound, then X has a maximal element.1
(iii)x ≤ x for all x ∈ X.
Proof. (2 ⇒ 3) Let X be a partially ordered subset as in 3 and let F =
Example B.2. Let Y be a set and X = 2Y . There are two natural partial orders {E ⊂ X : E is linearly ordered} which we equip with the inclusion partial
on X. ordering. By 2. there exist a maximal element E ∈ F. By assumption, the
1. Ordered by inclusion, A ≤ B is A ⊂ B and linearly ordered set E has an upper bound x ∈ X. The element x is maximal,
2. Ordered by reverse inclusion, A ≤ B if B ⊂ A. for if y ∈ Y and y ≥ x, then E ∪ {y} is still an linearly ordered set containing
E. So by maximality of E, E = E ∪ {y} , i.e. y ∈ E and therefore y ≤ x showing
Definition B.3. Let (X, ≤) be a partially ordered set we say X is linearly or which combined with y ≥ x implies that y = x.2
totally ordered if for all x, y ∈ X either x ≤ y or y ≤ x. The real numbers R Q (3 ⇒ 1) Let {Xα }α∈A be a collection of non-empty sets, we must show
with the usual order ≤ is a typical example. `α∈A Xα is not empty. Let G denote the collection of functions g : D(g) →
α∈A Xα such that D(g) is a subset of A, and for all α ∈ D(g), g(α) ∈ Xα .
Definition B.4. Let (X, ≤) be a partial ordered set. We say x ∈ X is a max- Notice that G is not empty, for we may let α0 ∈ A and x0 ∈ Xα and then set
imal element if for all y ∈ X such that y ≥ x implies y = x, i.e. there is no D(g) = {α0 } and g(α0 ) = x0 to construct an element of G. We now put a partial
element larger than x. An upper bound for a subset E of X is an element order on G as follows. We say that f ≤ g for f, g ∈ G provided that D(f ) ⊂ D(g)
x ∈ X such that x ≥ y for all y ∈ E. and f = g|D(f ) . If Φ ⊂ G is a linearly ordered set, let D(h) = ∪g∈Φ D(g) and for
α ∈ D(g) let h(α) = g(α). Then h ∈ G is an upper bound for Φ. So by Zorn’s
Example B.5. Let

1
If X is a countable set we may prove Zorn’s Lemma by induction. Let {xn }∞ n=1
X= a = {1} b = {1, 2} c = {3} d = {2, 4} e = {2} be an enumeration of X, and define En ⊂ X inductively as follows. For n = 1 let
E1 = {x1 }, and if En have been chosen, let En+1 = En ∪ {xn+1 } if xn+1 is an upper
ordered by set inclusion. Then b and d are maximal elements despite that fact bound for En otherwise let En+1 = En . The set E = ∪∞ n=1 En is a linearly ordered
that b a and a b. We also have (you check) subset of X and hence by assumption E has an upper bound, x ∈ X. I
claim that his element is maximal, for if there exists y = xm ∈ X such that y ≥ x,
• If E = {a, e, c}, then E has no upper bound. then xm would be an upper bound for Em−1 and therefore y = xm ∈ Em ⊂ E.
That is to say if y ≥ x, then y ∈ E and hence y ≤ x, so y = x. (Hence we may view
Definition B.6. • If E = {a, e}, then b is an upper bound. Zorn’s lemma as a “ jazzed” up version of induction.)
• E = {e}, then b and d are upper bounds. 2
Similarly one may show that 3 ⇒ 2. Let F = {E ⊂ X : E is linearly ordered} S
and order F by inclusion. If M ⊂ F is linearly ordered, let E = ∪M = A. If
Theorem B.7. The following are equivalent. A∈M
x, y ∈ E then x ∈ A and y ∈ B for some A, B ⊂ M. Now M is linearly ordered
1. The axiom of choice: to each collection, ` {Xα }α∈A , of non-empty sets by set inclusion so A ⊂ B or B ⊂ A i.e. x, y ∈ A or x, y ∈ B. Since A and B
there exists a “choice function,” x : A → Xα such that x(α) ∈ Xα for are linearly order we must have either x ≤ y or y ≤ x, that is to say E is linearly
Q α∈A ordered. Hence by 3. there exists a maximal element E ∈ F which is the assertion
all α ∈ A, i.e. α∈A Xα 6= ∅. in 2.
10 B Zorn’s Lemma and the Hausdorff Maximal Principle
Lemma there exists a maximal element h ∈ G. To finish the proof we need only 1. P0 ∈ F1 .
show that D(h) = A. If this were not the case, then let α0 ∈ A \ D(h) and 2. If Φ ⊂ F1 is a totally ordered (by set inclusion) subset then ∪Φ ∈ F1 .
x0 ∈ Xα0 . We may now define D(h̃) = D(h) ∪ {α0 } and 3. If P ∈ F1 then g(P ) ∈ F1 .

h(α) if α ∈ D(h) Let us call a general subset F 0 ⊂ F satisfying these three conditions a tower
h̃(α) =
x0 if α = α0 . and let
F0 = ∩ {F 0 : F 0 is a tower} .
Then h ≤ h̃ while h 6= h̃ violating the fact that h was a maximal element.
(1 ⇒ 2) Let (X, ≤) be a partially ordered set. Let F be the collection of Standard arguments show that F0 is still a tower and clearly is the smallest
linearly ordered subsets of X which we order by set inclusion. Given x0 ∈ X, tower containing P0 . (Morally speaking F0 consists of all of the sets we were
{x0 } ∈ F is linearly ordered set so that F 6= ∅. Fix an element P0 ∈ F. If trying to constructed in the “idea section” of the proof.) We now claim that
P0 is not maximal there exists P1 ∈ F such that P0 P1 . In particular we F0 is a linearly ordered subset of F. To prove this let Γ ⊂ F0 be the linearly
may choose x ∈ / P0 such that P0 ∪ {x} ∈ F. The idea now is to keep repeating ordered set
this process of adding points x ∈ X until we construct a maximal element P
of F. We now have to take care of some details. We may assume with out Γ = {C ∈ F0 : for all A ∈ F0 either A ⊂ C or C ⊂ A} .
loss of generality that F̃ = {P ∈ F : P is not maximal} is a non-empty set.
Shortly we will show that Γ ⊂ F0 is a tower and hence that F0 = Γ. That is
For P ∈ F̃, let P ∗ = {x ∈ X : P ∪ {x} ∈ F} . As the above argument Q shows, to say F0 is linearly ordered. Assuming this for the moment let us finish the
P ∗ 6= ∅ for all P ∈ F̃. Using the axiom of choice, there exists f ∈ P ∈F̃ P ∗ .
proof.
We now define g : F → F by
Let P ≡ ∪F0 which is in F0 by property 2 and is clearly the largest element
in F0 . By 3. it now follows that P ⊂ g(P ) ∈ F0 and by maximality of P, we

P if P is maximal
g(P ) = (B.1) have g(P ) = P, the desired fixed point. So to finish the proof, we must show
P ∪ {f (x)} if P is not maximal.
that Γ is a tower. First off it is clear that P0 ∈ Γ so in particular Γ is not
The proof is completed by Lemma B.8 below which shows that g must have a empty. For each C ∈ Γ let
fixed point P ∈ F. This fixed point is maximal by construction of g.
ΦC := {A ∈ F0 : either A ⊂ C or g(C) ⊂ A} .
Lemma B.8. The function g : F → F defined in Eq. (B.1) has a fixed point.3
We will begin by showing that ΦC ⊂ F0 is a tower and therefore that ΦC = F0 .
Proof. The idea of the proof ∞is as follows. Let P0 ∈ F be chosen arbi-
trarily. Notice that Φ = g (n) (P0 ) n=0 ⊂ F is a linearly ordered set and it is 1. P0 ∈ ΦC since P0 ⊂ C for all C ∈ Γ ⊂ F0 . 2. If Φ ⊂ ΦC ⊂ F0 is totally
∞ ordered by set inclusion, then AΦ := ∪Φ ∈ F0 . We must show AΦ ∈ ΦC , that
g (n) (P0 ) ∈ F. Similarly we may repeat
S
therefore easily verified that P1 = is that AΦ ⊂ C or C ⊂ AΦ . Now if A ⊂ C for all A ∈ Φ, then AΦ ⊂ C and
n=0
∞
(n)
∞ hence AΦ ∈ ΦC . On the other hand if there is some A ∈ Φ such that g(C) ⊂ A
g (n) (P2 ) ∈ F, etc.
S S
the process to construct P2 = g (P1 ) ∈ F and P3 = then clearly g(C) ⊂ AΦ and again AΦ ∈ ΦC . 3. Given A ∈ ΦC we must show
n=0 n=0
etc. Then take P∞ = ∪∞
n=0 Pn and
start again with P0 replaced by P∞ . Then g(A) ∈ ΦC , i.e. that
keep going this way until eventually the sets stop increasing in size, in which g(A) ⊂ C or g(C) ⊂ g(A). (B.2)
case we have found our fixed point. The problem with this strategy is that we There are three cases to consider: either A C, A = C, or g(C) ⊂ A. In the
may never win. (This is very reminiscent of constructing measurable sets and case A = C, g(C) = g(A) ⊂ g(A) and if g(C) ⊂ A then g(C) ⊂ A ⊂ g(A) and
the way out is to use measure theoretic like arguments.) Eq. (B.2) holds in either of these cases. So assume that A C. Since C ∈ Γ,
Let us now start the formal proof. Again let P0 ∈ F and let F1 = {P ∈ either g(A) ⊂ C (in which case we are done) or C ⊂ g(A). Hence we may
F : P0 ⊂ P }. Notice that F1 has the following properties: assume that
3
Here is an easy proof if the elements of F happened to all be finite sets and there A C ⊂ g(A).
existed a set P ∈ F with a maximal number of elements. In this case the condition
Now if C were a proper subset of g(A) it would then follow that g(A) \ A would
that P ⊂ g(P ) would imply that P = g(P ), otherwise g(P ) would have more
consist of at least two points which contradicts the definition of g. Hence we
elements than P.

must have g(A) = C ⊂ C and again Eq. (B.2) holds, so ΦC is a tower. It is now
easy to show Γ is a tower. It is again clear that P0 ∈ Γ and Property 2. may be
checked for Γ in the same way as it was done for ΦC above. For Property 3., if
C ∈ Γ we may use ΦC = F0 to conclude for all A ∈ F0 , either A ⊂ C ⊂ g(C)
or g(C) ⊂ A, i.e. g(C) ∈ Γ. Thus Γ is a tower and we are done.
C
Nets
In this section (which may be skipped) we develop the notion of nets. Nets 2. x is a cluster point of {xα }α∈A if for all V ∈ τx , xα ∈ V frequently, i.e.
are generalization of sequences. Here is an example which shows that for general for all β ∈ A there exists α ≥ β such that xα ∈ V.
topological spaces, sequences are not always adequate.
Proposition C.5. Let X be a topological space and E ⊂ X. Then
Example C.1. Equip CR with the topology of pointwise convergence, i.e. the 1. x is an accumulation point of E (see Definition 13.29) iff there exists net
product topology and consider C(R, C) ⊂ CR . If {fn } ⊂ C(R, C) is a sequence {xα } ⊂ E \ {x} such that xα → x.
which converges such that fn → f ∈ CR pointwise then f is a Borel measurable 2. x ∈ Ē iff there exists {xα } ⊂ E such that xα → x.
function. Hence the sequential limits of elements in C(R, C) is necessarily con-
tained in the Borel measurable functions which is properly contained in CR . In Proof.
short the sequential closure of C(R, C) is a proper subset of CR . On the other 1. Suppose x is an accumulation point of E and let A = τx be ordered by
hand we have C(R, C) = CR . Indeed a typical open neighborhood of f ∈ CR is reverse set inclusion. To each α ∈ A = τx choose xα ∈ (α \ {x}) ∩ E which
of the form is possible sine x is an accumulation point of E. Then given V ∈ τx for
N = {g ∈ CR : |g(x) − f (x)| < ε for x ∈ Λ}, all α ≥ V (i.e. and α ⊂ V ), xα ∈ V and hence xα → x. Conversely if
where ε > 0 and Λ is a finite subset of R. Since N ∩ C(R, C) 6= ∅ it follows that {xα }α∈A ⊂ E \ {x} and xα → x then for all V ∈ τx there exists β ∈ A such
f ∈ C(R, C). that xα ∈ V for all α ≥ β. In particular xα ∈ (E \ {x}) ∩ V 6= ∅ and so
x ∈ acc(E) – the accumulation points of E.
Definition C.2. A directed set (A, ≤) is a set with a relation “≤” such that 2. If {xα } ⊂ E such that xα → x then for all V ∈ τx there exists β ∈ A such
that xα ∈ V ∩ E for all α ≥ β. In particular V ∩ E 6= ∅ for all V ∈ τx
1. α ≤ α for all α ∈ A. and this implies x ∈ Ē. For the converse recall Proposition 13.31 implies
2. If α ≤ β and β ≤ γ then α ≤ γ. E = E ∪ acc(E). If x ∈ acc(E) there exists a net {xα } ⊂ E such that
3. A is cofinite, i.e. α, β ∈ A there exists γ ∈ A such that α ≤ γ and β ≤ γ. xα → x by item 1. If x ∈ E we may simply take xn = x for all n ∈ A := N.
A net is function x : A → X where A is a directed set. We will often denote
a net x by {xα }α∈A .
Proposition C.6. Let X and Y be two topological spaces and f : X → Y be a
Example C.3 (Directed sets). function. Then f is continuous at x ∈ X iff f (xα ) → f (x) for all nets xα → x.
1. A = 2X ordered by inclusion, i.e. α ≤ β if α ⊂ β. If α ≤ β and β ≤ γ then Proof. If f is continuous at x and xα → x then for any V ∈ τf (x) there
α ⊂ β ⊂ γ and hence α ≤ γ. Similalry if α, β ∈ 2X then α, β ≤ α ∪ β =: γ. exists W ∈ τx such that f (W ) ⊂ V. Since xα ∈ W eventually, f (xα ) ∈ V
2. A = 2X ordered by reverse inclusion, i.e. α ≤ β if β ⊂ α. If α ≤ β and eventually and we have shown f (xα ) → f (x). Conversely, if f is not continuous
β ≤ γ then α ⊇ β ⊇ γ and so α ≤ γ and if α, β ∈ A then α, β ≤ α ∩ β. at x then there exists W ∈ τf (x) such that f (V ) * W for all V ∈ τx . Let
3. Let A = N equipped with the usual ordering on N. In this case nets are A = τx be ordered by reverse set inclusion and for V ∈ τx choose (axiom of
simply sequences. choice) xV ∈ V such that f (xV ) ∈/ W. Then xV → x since for any U ∈ τx ,
xV ∈ U if V ≥ U (i.e. V ⊂ U ). On the over hand f (xV ) ∈ / W for all V ∈ τx
Definition C.4. Let {xα }α∈A ⊂ X be a net then: showing f (xV ) 9 f (x).
1. xα converges to x ∈ X (written xα → x) iff for all V ∈ τx , xα ∈ V Definition C.7 ( Subnet). A net hyβ iβ∈B is a subnet of a net hxα iα∈A if
eventually, i.e. there exists β = βV ∈ A such that xα ∈ V for all α ≥ β. there exists a map β ∈ B → αβ ∈ A such that
14 C Nets
 
1. yβ = xαβ for all β ∈ B and [
2. for all α0 ∈ A there exists β0 ∈ B such that αβ ≥ α0 whenever β ≥ β0 , i.e. xα ∈ X \  Uj  6= ∅ for all α ∈ A.
αβ ≥ α0 eventually. j∈α
Proposition C.8. A point x ∈ X is a cluster point of a net hxα iα∈A iff there This net has no cluster point. To see this suppose x ∈ X and
S j ∈ J is chosen
exists a subnet hyβ iβ∈B such that yβ → x. so that x ∈ Uj . Then for all α ≥ {j} (i.e. j ∈ α), xα ∈
/ Uα ⊇ Uj and in
γ∈α
Proof. Suppose hyβ iβ∈B is a subnet of hxα iα∈A such that yβ = xαβ → x. particular xα ∈
/ Uj . This shows xα ∈
/ Uj frequently and hence x is not a cluster
Then for W ∈ τx and α0 ∈ A there exists β0 ∈ B such that yβ = xαβ ∈ W for point.
all β ≥ β0 . Choose β1 ∈ B such that αβ ≥ α0 for all β ≥ β1 then choose β3 ∈ B
such that β3 ≥ β1 and β3 ≥ β2 then αβ ≥ α0 and xαβ ∈ W for all β ≥ β3 which
implies xα ∈ W frequently. Conversely assume x is a cluster point of a net
hxα iα∈A . We mak B := τx × A into a directed set by defining (U, α) ≤ (U 0 , α0 )
iff α ≤ α0 and U ⊇ U 0 . For all (U, γ) ∈ B = τx × A, choose α(U,γ) ≥ γ in
A such that y(U,γ) = xα(U,γ) ∈ U. Then if α0 ∈ A for all (U 0 , γ 0 ) ≥ (U, α0 ),
i.e. γ 0 ≥ α0 and U 0 ⊂ U, α(U 0 ,γ 0 ) ≥ γ 0 ≥ α0 . Now if W ∈ τx is given, then
y(U,γ) ∈ U ⊂ W for all U ⊂ W . Hence fixing α ∈ A we see if (U, γ) ≥ (W, α)
then y(U,γ) = xα(U,γ) ∈ U ⊂ W showing that y(U,γ) → x.
Exercise C.1 (Folland #34, p. 121). Let hxα iα∈A be a net in a topological
space and forTeach α ∈ A let Eα ≡ {xβ : β ≥ α}. Then x is a cluster point of
hxα i iff x ∈ Eα.
α∈A
Solution to Exercise (C.1). If x is a cluster point, then given W ∈ τx we

know EαT ∩ W 6= ∅ for all α ∈ E since xβ ∈ W frequently thus x ∈ E α for all α,
i.e. x ∈ E α . Conversely if x is not a cluster point of hxα i then there exists
α∈A
W ∈ τx and α ∈ A such that xT
β ∈
/ W for all β ≥ α, i.e. W ∩ Eα = ∅. But this
/ E α and hence x ∈
shows x ∈ / Eα.
α∈A
Theorem C.9. A topological space X is compact iff every net has a cluster
point iff every net has a convergent subnet.
Proof. Suppose X is compact, hxα iα∈A ⊂ X is a net and let Fα :=

{xβ : β ≥ α}. Then Fα is closed for all α ∈ A, Fα ⊂ Fα0 if α ≥ α0 and
Fα1 ∩ · · · ∩ Fαn ⊇ Fγ whenever γ ≥ αi for i = 1, . . . , n. (Such a γ always
exists since A is a directed set.) Therefore Fα1 ∩ · · · ∩ Fαn 6= ∅ i.e. {Fα }α∈A has
the finite
T intersection property and since X is compact this implies there exists
x ∈ Fα By Exercise C.1, it follows that x is a cluster point of hxα iα∈A .
α∈a
Conversely, if X is not compact let {Uj }j∈J be an infinite cover with no finite
subcover. Let A be the directed set A = {α ⊂ J : # (α) < ∞} with α ≤ β iff
α ⊂ β. Define a net hxα iα∈A in X by choosing

References
1. Vladimir I. Bogachev, Gaussian measures, Mathematical Surveys and Mono- 11. Ronald B. Guenther and John W. Lee, Partial differential equations of mathemat-
graphs, vol. 62, American Mathematical Society, Providence, RI, 1998. MR ical physics and integral equations, Dover Publications Inc., Mineola, NY, 1996,
2000a:60004 Corrected reprint of the 1988 original. MR 97e:35001
2. Isaac Chavel, Riemannian geometry—a modern introduction, Cambridge Univer- 12. Hui Hsiung Kuo, Gaussian measures in Banach spaces, Springer-Verlag, Berlin,
sity Press, Cambridge, 1993. MR 95j:53001 1975, Lecture Notes in Mathematics, Vol. 463. MR 57 #1628
3. , The Laplacian on Riemannian manifolds, Spectral theory and geome- 13. Serge Lang, Real and functional analysis, third ed., Graduate Texts in Mathemat-
try (Edinburgh, 1998), Cambridge Univ. Press, Cambridge, 1999, pp. 30–75. MR ics, vol. 142, Springer-Verlag, New York, 1993. MR 94b:00005
2001c:58029 14. Lynn H. Loomis, An introduction to abstract harmonic analysis, D. Van Nostrand
4. Giuseppe Da Prato and Jerzy Zabczyk, Stochastic equations in infinite dimen- Company, Inc., Toronto-New York-London, 1953. MR 14,883c
sions, Cambridge University Press, Cambridge, 1992. MR 95g:60073 15. Vladimir G. Maz’ja, Sobolev spaces, Springer Series in Soviet Mathematics,
5. E. B. Davies, Heat kernels and spectral theory, Cambridge Tracts in Mathematics, Springer-Verlag, Berlin, 1985, Translated from the Russian by T. O. Shaposh-
vol. 92, Cambridge University Press, Cambridge, 1990. MR 92a:35035 nikova. MR 87g:46056
6. , Heat kernel bounds for higher order elliptic operators, Journées 16. A. Pazy, Semigroups of linear operators and applications to partial differential
“Équations aux Dérivées Partielles” (Saint-Jean-de-Monts, 1995), École Polytech., equations, Applied Mathematical Sciences, vol. 44, Springer-Verlag, New York,
Palaiseau, 1995, pp. Exp. No. III, 11. MR 96i:35020 1983. MR 85g:47061
7. F. G. Friedlander, Introduction to the theory of distributions, second ed., Cam- 17. Michael Reed and Barry Simon, Methods of modern mathematical physics. II.
bridge University Press, Cambridge, 1998, With additional material by M. Joshi. Fourier analysis, self-adjointness, Academic Press [Harcourt Brace Jovanovich
MR 2000g:46002: Call No. QA324 .F74 1998 Publishers], New York, 1975. MR 58 #12429b
8. Enrico Giusti, Minimal surfaces and functions of bounded variation, Birkhäuser 18. , Methods of modern mathematical physics. I, second ed., Academic Press
Verlag, Basel, 1984. MR 87a:58041 Inc. [Harcourt Brace Jovanovich Publishers], New York, 1980, Functional analysis.
9. Leonard Gross, Integration and nonlinear transformations in Hilbert space, Trans. MR 85e:46002
Amer. Math. Soc. 94 (1960), 404–440. MR 22 #2883 19. L. C. G. Rogers and David Williams, Diffusions, Markov processes, and mar-
10. , Abstract Wiener spaces, Proc. Fifth Berkeley Sympos. Math. Statist. tingales. Vol. 1, Cambridge Mathematical Library, Cambridge University Press,
and Probability (Berkeley, Calif., 1965/66), Vol. II: Contributions to Probability Cambridge, 2000, Foundations, Reprint of the second (1994) edition. MR
Theory, Part 1, Univ. California Press, Berkeley, Calif., 1967, pp. 31–42. MR 35 2001g:60188
#3027
20. Laurent Saloff-Coste, Aspects of Sobolev-type inequalities, London Mathematical
Society Lecture Note Series, vol. 289, Cambridge University Press, Cambridge,
2002. MR 2003c:46048
21. Robert Schatten, Norm ideals of completely continuous operators, Second printing.
Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 27, Springer-Verlag,
Berlin, 1970. MR 41 #2449
22. B. Simon, Convergence in trace ideals, Proc. Amer. Math. Soc. 83 (1981), no. 1,
39–43. MR 82h:47042
23. Barry Simon, Trace ideals and their applications, London Mathematical Society
Lecture Note Series, vol. 35, Cambridge University Press, Cambridge, 1979. MR
80k:47048
24. S. L. Sobolev and E. R. Dawson, Partial differential equations of mathematical
physics, Translated from the third Russian edition by E. R. Dawson; English
translation edited by T. A. A. Broadbent, Pergamon Press, Oxford, 1964. MR 31
#2478
25. Elias M. Stein, Singular integrals and differentiability properties of functions,
Princeton Mathematical Series, No. 30, Princeton University Press, Princeton,
N.J., 1970. MR 44 #7280
26. Frank W. Warner, Foundations of differentiable manifolds and Lie groups, Grad-
uate Texts in Mathematics, vol. 94, Springer-Verlag, New York, 1983, Corrected
reprint of the 1971 edition. MR 84k:58001
Absolute continuity, 601 Integration by parts, 609
Isomorphic measure spaces, 579
Banach space, 59
reflexive, 497 Lusin Space, 579
sums in, 69
Bounded variation, 601 Minikowski functional, 494
Cauchy, 11 Neighborhood, 197

Cauchy sequence base, 197
in a metric space, 53 open, 197
in a normed space, 59
Open, see Sets
Closed, see Sets
Open cover, 203
Coherent states, 715
Complete Polish Spaces, 250
Metric space, 54 Product topology, 204
Continuous function, 198, 202
Contraction Mapping Principle, 189 Radon Measure
Complex, 564
Function Signed, 564
continuous, 52, 198 Reflexive, see Banach space
continuous at a point, 52
Fundamental theorem of calculus, 608 Sets
closed, 197
Hahn-Banach Theorem, 495 open, 197
Helly’s selection principle, 579 Sub-base, see Topology
Homeomorphism, 198 Summable, 29
Index
Topological Space, 197 relative / induced, 201

Topology, 197 sub-base, 199
base, 199 trivial, 198
discrete, 198 Total variation, 601
generated by functions, 204
induced / relative, 201 version, 448

Analisis II - Note's Class

Uploaded by

Copyright:

Available Formats

Analisis II - Note's Class

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analisis II - Note's Class

Uploaded by

Copyright:

Available Formats

Bruce K.

Analysis Tools with Examples

Part I Background Material

1 Introduction / User Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 A Brief Review of Real and Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Limits and Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5 `p – spaces, Minkowski and Holder Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Part II Metric, Banach, and Hilbert Space Basics

8 Hilbert Space Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

9 Hölder Spaces as Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Part III Calculus and Ordinary Differential Equations in Banach Spaces

10 The Riemann Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

11 Ordinary Differential Equations in a Banach Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

Page: 4 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

11.5 Semi-Group Properties of time independent flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

12 Banach Space Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

Part IV Topological Spaces

13 Topological Space Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

Page: 5 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

15 Locally Compact Hausdorff Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

16 Baire Category Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

Part V Lebesgue Integration Theory

17 Introduction: What are measures and why “measurable” sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

19 Measures and Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

20 Multiple Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Page: 6 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

22 Approximation Theorems and Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415

Part VI Further Hilbert and Banach Space Techniques

23 L2 - Hilbert Spaces Techniques and Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439

24 Complex Measures, Radon-Nikodym Theorem and the Dual of Lp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469

Page: 7 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

25 Three Fundamental Principles of Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 491

26 Weak and Strong Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519

27 Bochner Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537

Part VII Construction and Differentiation of Measures

28 Examples of Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541

29 Probability Measures on Lusin Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569

Page: 8 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

30 Lebesgue Differentiation and the Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577

31 Constructing Measures Via Carathéodory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607

32 The Daniell – Stone Construction of Integration and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627

33 Class Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671

Page: 9 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

Part VIII The Fourier Transform and Generalized Functions

34 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683

35 Constant Coefficient partial differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703

36 Elementary Generalized Functions / Distribution Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725

37 Convolutions involving distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 759

A Multinomial Theorems and Calculus Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

B Zorn’s Lemma and the Hausdorff Maximal Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Page: 10 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

Page: 11 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

Not written as of yet. Topics to mention.

f (x) = min g −1 ({x}) = min {n ∈ N : f (n) = x} . f −1 E := f −1 (E) := {f −1 (E)|E ∈ E}.

Then f : X → N is injective which combined with item If G ⊂ 2X , let

Page: 6 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

Page: 8 job: anal macro: svmonob.cls date/time: 6-Aug-2004/17:08

M − i N −1 < an ≤ M for all n ≥ m.

and in particular, for all M ≥ n,